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
));
82 __u8 migr_state
; /* Normal or Migrating */
83 __u8 migr_type
; /* Initializing, Rebuilding, ... */
87 struct imsm_map map
[1];
88 /* here comes another one if migr_state */
89 } __attribute__ ((packed
));
92 __u8 volume
[MAX_RAID_SERIAL_LEN
];
95 __u32 status
; /* Persistent RaidDev status */
96 __u32 reserved_blocks
; /* Reserved blocks at beginning of volume */
97 #define IMSM_DEV_FILLERS 12
98 __u32 filler
[IMSM_DEV_FILLERS
];
100 } __attribute__ ((packed
));
103 __u8 sig
[MAX_SIGNATURE_LENGTH
]; /* 0x00 - 0x1F */
104 __u32 check_sum
; /* 0x20 - 0x23 MPB Checksum */
105 __u32 mpb_size
; /* 0x24 - 0x27 Size of MPB */
106 __u32 family_num
; /* 0x28 - 0x2B Checksum from first time this config was written */
107 __u32 generation_num
; /* 0x2C - 0x2F Incremented each time this array's MPB is written */
108 __u32 error_log_size
; /* 0x30 - 0x33 in bytes */
109 __u32 attributes
; /* 0x34 - 0x37 */
110 __u8 num_disks
; /* 0x38 Number of configured disks */
111 __u8 num_raid_devs
; /* 0x39 Number of configured volumes */
112 __u8 error_log_pos
; /* 0x3A */
113 __u8 fill
[1]; /* 0x3B */
114 __u32 cache_size
; /* 0x3c - 0x40 in mb */
115 __u32 orig_family_num
; /* 0x40 - 0x43 original family num */
116 __u32 pwr_cycle_count
; /* 0x44 - 0x47 simulated power cycle count for array */
117 __u32 bbm_log_size
; /* 0x48 - 0x4B - size of bad Block Mgmt Log in bytes */
118 #define IMSM_FILLERS 35
119 __u32 filler
[IMSM_FILLERS
]; /* 0x4C - 0xD7 RAID_MPB_FILLERS */
120 struct imsm_disk disk
[1]; /* 0xD8 diskTbl[numDisks] */
121 /* here comes imsm_dev[num_raid_devs] */
122 /* here comes BBM logs */
123 } __attribute__ ((packed
));
125 #define BBM_LOG_MAX_ENTRIES 254
127 struct bbm_log_entry
{
128 __u64 defective_block_start
;
129 #define UNREADABLE 0xFFFFFFFF
130 __u32 spare_block_offset
;
131 __u16 remapped_marked_count
;
133 } __attribute__ ((__packed__
));
136 __u32 signature
; /* 0xABADB10C */
138 __u32 reserved_spare_block_count
; /* 0 */
139 __u32 reserved
; /* 0xFFFF */
140 __u64 first_spare_lba
;
141 struct bbm_log_entry mapped_block_entries
[BBM_LOG_MAX_ENTRIES
];
142 } __attribute__ ((__packed__
));
146 static char *map_state_str
[] = { "normal", "uninitialized", "degraded", "failed" };
149 static unsigned int sector_count(__u32 bytes
)
151 return ((bytes
+ (512-1)) & (~(512-1))) / 512;
154 static unsigned int mpb_sectors(struct imsm_super
*mpb
)
156 return sector_count(__le32_to_cpu(mpb
->mpb_size
));
159 /* internal representation of IMSM metadata */
162 void *buf
; /* O_DIRECT buffer for reading/writing metadata */
163 struct imsm_super
*anchor
; /* immovable parameters */
165 size_t len
; /* size of the 'buf' allocation */
166 void *next_buf
; /* for realloc'ing buf from the manager */
168 int updates_pending
; /* count of pending updates for mdmon */
169 int creating_imsm
; /* flag to indicate container creation */
170 int current_vol
; /* index of raid device undergoing creation */
171 #define IMSM_MAX_RAID_DEVS 2
172 struct imsm_dev
*dev_tbl
[IMSM_MAX_RAID_DEVS
];
176 __u8 serial
[MAX_RAID_SERIAL_LEN
];
179 struct imsm_disk disk
;
182 struct dl
*add
; /* list of disks to add while mdmon active */
183 struct bbm_log
*bbm_log
;
187 unsigned long long start
, size
;
190 /* definition of messages passed to imsm_process_update */
191 enum imsm_update_type
{
192 update_activate_spare
,
197 struct imsm_update_activate_spare
{
198 enum imsm_update_type type
;
202 struct imsm_update_activate_spare
*next
;
205 struct imsm_update_create_array
{
206 enum imsm_update_type type
;
211 struct imsm_update_add_disk
{
212 enum imsm_update_type type
;
215 static int imsm_env_devname_as_serial(void)
217 char *val
= getenv("IMSM_DEVNAME_AS_SERIAL");
219 if (val
&& atoi(val
) == 1)
226 static struct supertype
*match_metadata_desc_imsm(char *arg
)
228 struct supertype
*st
;
230 if (strcmp(arg
, "imsm") != 0 &&
231 strcmp(arg
, "default") != 0
235 st
= malloc(sizeof(*st
));
236 memset(st
, 0, sizeof(*st
));
237 st
->ss
= &super_imsm
;
238 st
->max_devs
= IMSM_MAX_DEVICES
;
239 st
->minor_version
= 0;
245 static __u8
*get_imsm_version(struct imsm_super
*mpb
)
247 return &mpb
->sig
[MPB_SIG_LEN
];
251 /* retrieve a disk directly from the anchor when the anchor is known to be
252 * up-to-date, currently only at load time
254 static struct imsm_disk
*__get_imsm_disk(struct imsm_super
*mpb
, __u8 index
)
256 if (index
>= mpb
->num_disks
)
258 return &mpb
->disk
[index
];
262 /* retrieve a disk from the parsed metadata */
263 static struct imsm_disk
*get_imsm_disk(struct intel_super
*super
, __u8 index
)
267 for (d
= super
->disks
; d
; d
= d
->next
)
268 if (d
->index
== index
)
275 /* generate a checksum directly from the anchor when the anchor is known to be
276 * up-to-date, currently only at load or write_super after coalescing
278 static __u32
__gen_imsm_checksum(struct imsm_super
*mpb
)
280 __u32 end
= mpb
->mpb_size
/ sizeof(end
);
281 __u32
*p
= (__u32
*) mpb
;
285 sum
+= __le32_to_cpu(*p
++);
287 return sum
- __le32_to_cpu(mpb
->check_sum
);
290 static size_t sizeof_imsm_map(struct imsm_map
*map
)
292 return sizeof(struct imsm_map
) + sizeof(__u32
) * (map
->num_members
- 1);
295 struct imsm_map
*get_imsm_map(struct imsm_dev
*dev
, int second_map
)
297 struct imsm_map
*map
= &dev
->vol
.map
[0];
299 if (second_map
&& !dev
->vol
.migr_state
)
301 else if (second_map
) {
304 return ptr
+ sizeof_imsm_map(map
);
310 /* return the size of the device.
311 * migr_state increases the returned size if map[0] were to be duplicated
313 static size_t sizeof_imsm_dev(struct imsm_dev
*dev
, int migr_state
)
315 size_t size
= sizeof(*dev
) - sizeof(struct imsm_map
) +
316 sizeof_imsm_map(get_imsm_map(dev
, 0));
318 /* migrating means an additional map */
319 if (dev
->vol
.migr_state
)
320 size
+= sizeof_imsm_map(get_imsm_map(dev
, 1));
322 size
+= sizeof_imsm_map(get_imsm_map(dev
, 0));
327 static struct imsm_dev
*__get_imsm_dev(struct imsm_super
*mpb
, __u8 index
)
333 if (index
>= mpb
->num_raid_devs
)
336 /* devices start after all disks */
337 offset
= ((void *) &mpb
->disk
[mpb
->num_disks
]) - _mpb
;
339 for (i
= 0; i
<= index
; i
++)
341 return _mpb
+ offset
;
343 offset
+= sizeof_imsm_dev(_mpb
+ offset
, 0);
348 static struct imsm_dev
*get_imsm_dev(struct intel_super
*super
, __u8 index
)
350 if (index
>= super
->anchor
->num_raid_devs
)
352 return super
->dev_tbl
[index
];
355 static __u32
get_imsm_ord_tbl_ent(struct imsm_dev
*dev
, int slot
)
357 struct imsm_map
*map
;
359 if (dev
->vol
.migr_state
)
360 map
= get_imsm_map(dev
, 1);
362 map
= get_imsm_map(dev
, 0);
364 /* top byte identifies disk under rebuild */
365 return __le32_to_cpu(map
->disk_ord_tbl
[slot
]);
368 #define ord_to_idx(ord) (((ord) << 8) >> 8)
369 static __u32
get_imsm_disk_idx(struct imsm_dev
*dev
, int slot
)
371 __u32 ord
= get_imsm_ord_tbl_ent(dev
, slot
);
373 return ord_to_idx(ord
);
376 static void set_imsm_ord_tbl_ent(struct imsm_map
*map
, int slot
, __u32 ord
)
378 map
->disk_ord_tbl
[slot
] = __cpu_to_le32(ord
);
381 static int get_imsm_raid_level(struct imsm_map
*map
)
383 if (map
->raid_level
== 1) {
384 if (map
->num_members
== 2)
390 return map
->raid_level
;
393 static int cmp_extent(const void *av
, const void *bv
)
395 const struct extent
*a
= av
;
396 const struct extent
*b
= bv
;
397 if (a
->start
< b
->start
)
399 if (a
->start
> b
->start
)
404 static struct extent
*get_extents(struct intel_super
*super
, struct dl
*dl
)
406 /* find a list of used extents on the given physical device */
407 struct extent
*rv
, *e
;
410 __u32 reservation
= MPB_SECTOR_CNT
+ IMSM_RESERVED_SECTORS
;
412 for (i
= 0; i
< super
->anchor
->num_raid_devs
; i
++) {
413 struct imsm_dev
*dev
= get_imsm_dev(super
, i
);
414 struct imsm_map
*map
= get_imsm_map(dev
, 0);
416 for (j
= 0; j
< map
->num_members
; j
++) {
417 __u32 index
= get_imsm_disk_idx(dev
, j
);
419 if (index
== dl
->index
)
423 rv
= malloc(sizeof(struct extent
) * (memberships
+ 1));
428 for (i
= 0; i
< super
->anchor
->num_raid_devs
; i
++) {
429 struct imsm_dev
*dev
= get_imsm_dev(super
, i
);
430 struct imsm_map
*map
= get_imsm_map(dev
, 0);
432 for (j
= 0; j
< map
->num_members
; j
++) {
433 __u32 index
= get_imsm_disk_idx(dev
, j
);
435 if (index
== dl
->index
) {
436 e
->start
= __le32_to_cpu(map
->pba_of_lba0
);
437 e
->size
= __le32_to_cpu(map
->blocks_per_member
);
442 qsort(rv
, memberships
, sizeof(*rv
), cmp_extent
);
444 /* determine the start of the metadata
445 * when no raid devices are defined use the default
446 * ...otherwise allow the metadata to truncate the value
447 * as is the case with older versions of imsm
450 struct extent
*last
= &rv
[memberships
- 1];
453 remainder
= __le32_to_cpu(dl
->disk
.total_blocks
) -
454 (last
->start
+ last
->size
);
455 if (reservation
> remainder
)
456 reservation
= remainder
;
458 e
->start
= __le32_to_cpu(dl
->disk
.total_blocks
) - reservation
;
463 /* try to determine how much space is reserved for metadata from
464 * the last get_extents() entry, otherwise fallback to the
467 static __u32
imsm_reserved_sectors(struct intel_super
*super
, struct dl
*dl
)
473 /* for spares just return a minimal reservation which will grow
474 * once the spare is picked up by an array
477 return MPB_SECTOR_CNT
;
479 e
= get_extents(super
, dl
);
481 return MPB_SECTOR_CNT
+ IMSM_RESERVED_SECTORS
;
483 /* scroll to last entry */
484 for (i
= 0; e
[i
].size
; i
++)
487 rv
= __le32_to_cpu(dl
->disk
.total_blocks
) - e
[i
].start
;
495 static void print_imsm_dev(struct imsm_dev
*dev
, int index
)
499 struct imsm_map
*map
= get_imsm_map(dev
, 0);
503 printf("[%s]:\n", dev
->volume
);
504 printf(" RAID Level : %d\n", get_imsm_raid_level(map
));
505 printf(" Members : %d\n", map
->num_members
);
506 for (slot
= 0; slot
< map
->num_members
; slot
++)
507 if (index
== get_imsm_disk_idx(dev
, slot
))
509 if (slot
< map
->num_members
) {
510 ord
= get_imsm_ord_tbl_ent(dev
, slot
);
511 printf(" This Slot : %d%s\n", slot
,
512 ord
& IMSM_ORD_REBUILD
? " (out-of-sync)" : "");
514 printf(" This Slot : ?\n");
515 sz
= __le32_to_cpu(dev
->size_high
);
517 sz
+= __le32_to_cpu(dev
->size_low
);
518 printf(" Array Size : %llu%s\n", (unsigned long long)sz
,
519 human_size(sz
* 512));
520 sz
= __le32_to_cpu(map
->blocks_per_member
);
521 printf(" Per Dev Size : %llu%s\n", (unsigned long long)sz
,
522 human_size(sz
* 512));
523 printf(" Sector Offset : %u\n",
524 __le32_to_cpu(map
->pba_of_lba0
));
525 printf(" Num Stripes : %u\n",
526 __le32_to_cpu(map
->num_data_stripes
));
527 printf(" Chunk Size : %u KiB\n",
528 __le16_to_cpu(map
->blocks_per_strip
) / 2);
529 printf(" Reserved : %d\n", __le32_to_cpu(dev
->reserved_blocks
));
530 printf(" Migrate State : %s", dev
->vol
.migr_state
? "migrating" : "idle");
531 if (dev
->vol
.migr_state
)
532 printf(": %s", dev
->vol
.migr_type
? "rebuilding" : "initializing");
534 printf(" Map State : %s", map_state_str
[map
->map_state
]);
535 if (dev
->vol
.migr_state
) {
536 struct imsm_map
*map
= get_imsm_map(dev
, 1);
537 printf(" <-- %s", map_state_str
[map
->map_state
]);
540 printf(" Dirty State : %s\n", dev
->vol
.dirty
? "dirty" : "clean");
543 static void print_imsm_disk(struct imsm_super
*mpb
, int index
, __u32 reserved
)
545 struct imsm_disk
*disk
= __get_imsm_disk(mpb
, index
);
546 char str
[MAX_RAID_SERIAL_LEN
+ 1];
554 snprintf(str
, MAX_RAID_SERIAL_LEN
+ 1, "%s", disk
->serial
);
555 printf(" Disk%02d Serial : %s\n", index
, str
);
556 s
= __le32_to_cpu(disk
->status
);
557 printf(" State :%s%s%s%s\n", s
&SPARE_DISK
? " spare" : "",
558 s
&CONFIGURED_DISK
? " active" : "",
559 s
&FAILED_DISK
? " failed" : "",
560 s
&USABLE_DISK
? " usable" : "");
561 printf(" Id : %08x\n", __le32_to_cpu(disk
->scsi_id
));
562 sz
= __le32_to_cpu(disk
->total_blocks
) - reserved
;
563 printf(" Usable Size : %llu%s\n", (unsigned long long)sz
,
564 human_size(sz
* 512));
567 static void examine_super_imsm(struct supertype
*st
, char *homehost
)
569 struct intel_super
*super
= st
->sb
;
570 struct imsm_super
*mpb
= super
->anchor
;
571 char str
[MAX_SIGNATURE_LENGTH
];
574 __u32 reserved
= imsm_reserved_sectors(super
, super
->disks
);
576 snprintf(str
, MPB_SIG_LEN
, "%s", mpb
->sig
);
577 printf(" Magic : %s\n", str
);
578 snprintf(str
, strlen(MPB_VERSION_RAID0
), "%s", get_imsm_version(mpb
));
579 printf(" Version : %s\n", get_imsm_version(mpb
));
580 printf(" Family : %08x\n", __le32_to_cpu(mpb
->family_num
));
581 printf(" Generation : %08x\n", __le32_to_cpu(mpb
->generation_num
));
582 sum
= __le32_to_cpu(mpb
->check_sum
);
583 printf(" Checksum : %08x %s\n", sum
,
584 __gen_imsm_checksum(mpb
) == sum
? "correct" : "incorrect");
585 printf(" MPB Sectors : %d\n", mpb_sectors(mpb
));
586 printf(" Disks : %d\n", mpb
->num_disks
);
587 printf(" RAID Devices : %d\n", mpb
->num_raid_devs
);
588 print_imsm_disk(mpb
, super
->disks
->index
, reserved
);
589 if (super
->bbm_log
) {
590 struct bbm_log
*log
= super
->bbm_log
;
593 printf("Bad Block Management Log:\n");
594 printf(" Log Size : %d\n", __le32_to_cpu(mpb
->bbm_log_size
));
595 printf(" Signature : %x\n", __le32_to_cpu(log
->signature
));
596 printf(" Entry Count : %d\n", __le32_to_cpu(log
->entry_count
));
597 printf(" Spare Blocks : %d\n", __le32_to_cpu(log
->reserved_spare_block_count
));
598 printf(" First Spare : %llx\n", __le64_to_cpu(log
->first_spare_lba
));
600 for (i
= 0; i
< mpb
->num_raid_devs
; i
++)
601 print_imsm_dev(__get_imsm_dev(mpb
, i
), super
->disks
->index
);
602 for (i
= 0; i
< mpb
->num_disks
; i
++) {
603 if (i
== super
->disks
->index
)
605 print_imsm_disk(mpb
, i
, reserved
);
609 static void getinfo_super_imsm(struct supertype
*st
, struct mdinfo
*info
);
611 static void brief_examine_super_imsm(struct supertype
*st
)
613 /* We just write a generic DDF ARRAY entry
618 getinfo_super_imsm(st
, &info
);
619 fname_from_uuid(st
, &info
, nbuf
,'-');
620 printf("ARRAY /dev/imsm metadata=imsm UUID=%s\n", nbuf
+ 5);
623 static void detail_super_imsm(struct supertype
*st
, char *homehost
)
625 printf("%s\n", __FUNCTION__
);
628 static void brief_detail_super_imsm(struct supertype
*st
)
632 getinfo_super_imsm(st
, &info
);
633 fname_from_uuid(st
, &info
, nbuf
,'-');
634 printf(" UUID=%s", nbuf
+ 5);
638 static int match_home_imsm(struct supertype
*st
, char *homehost
)
640 printf("%s\n", __FUNCTION__
);
645 static void uuid_from_super_imsm(struct supertype
*st
, int uuid
[4])
647 /* The uuid returned here is used for:
648 * uuid to put into bitmap file (Create, Grow)
649 * uuid for backup header when saving critical section (Grow)
650 * comparing uuids when re-adding a device into an array
651 * In these cases the uuid required is that of the data-array,
652 * not the device-set.
653 * uuid to recognise same set when adding a missing device back
654 * to an array. This is a uuid for the device-set.
656 * For each of these we can make do with a truncated
657 * or hashed uuid rather than the original, as long as
659 * In each case the uuid required is that of the data-array,
660 * not the device-set.
662 /* imsm does not track uuid's so we synthesis one using sha1 on
663 * - The signature (Which is constant for all imsm array, but no matter)
664 * - the family_num of the container
665 * - the index number of the volume
666 * - the 'serial' number of the volume.
667 * Hopefully these are all constant.
669 struct intel_super
*super
= st
->sb
;
673 struct imsm_dev
*dev
= NULL
;
676 sha1_process_bytes(super
->anchor
->sig
, MAX_SIGNATURE_LENGTH
, &ctx
);
677 sha1_process_bytes(&super
->anchor
->family_num
, sizeof(__u32
), &ctx
);
678 if (super
->current_vol
>= 0)
679 dev
= get_imsm_dev(super
, super
->current_vol
);
681 __u32 vol
= super
->current_vol
;
682 sha1_process_bytes(&vol
, sizeof(vol
), &ctx
);
683 sha1_process_bytes(dev
->volume
, MAX_RAID_SERIAL_LEN
, &ctx
);
685 sha1_finish_ctx(&ctx
, buf
);
686 memcpy(uuid
, buf
, 4*4);
691 get_imsm_numerical_version(struct imsm_super
*mpb
, int *m
, int *p
)
693 __u8
*v
= get_imsm_version(mpb
);
694 __u8
*end
= mpb
->sig
+ MAX_SIGNATURE_LENGTH
;
695 char major
[] = { 0, 0, 0 };
696 char minor
[] = { 0 ,0, 0 };
697 char patch
[] = { 0, 0, 0 };
698 char *ver_parse
[] = { major
, minor
, patch
};
702 while (*v
!= '\0' && v
< end
) {
703 if (*v
!= '.' && j
< 2)
704 ver_parse
[i
][j
++] = *v
;
712 *m
= strtol(minor
, NULL
, 0);
713 *p
= strtol(patch
, NULL
, 0);
717 static int imsm_level_to_layout(int level
)
725 return ALGORITHM_LEFT_ASYMMETRIC
;
732 static void getinfo_super_imsm_volume(struct supertype
*st
, struct mdinfo
*info
)
734 struct intel_super
*super
= st
->sb
;
735 struct imsm_dev
*dev
= get_imsm_dev(super
, super
->current_vol
);
736 struct imsm_map
*map
= get_imsm_map(dev
, 0);
738 info
->container_member
= super
->current_vol
;
739 info
->array
.raid_disks
= map
->num_members
;
740 info
->array
.level
= get_imsm_raid_level(map
);
741 info
->array
.layout
= imsm_level_to_layout(info
->array
.level
);
742 info
->array
.md_minor
= -1;
743 info
->array
.ctime
= 0;
744 info
->array
.utime
= 0;
745 info
->array
.chunk_size
= __le16_to_cpu(map
->blocks_per_strip
) << 9;
746 info
->array
.state
= !dev
->vol
.dirty
;
748 info
->disk
.major
= 0;
749 info
->disk
.minor
= 0;
751 info
->data_offset
= __le32_to_cpu(map
->pba_of_lba0
);
752 info
->component_size
= __le32_to_cpu(map
->blocks_per_member
);
753 memset(info
->uuid
, 0, sizeof(info
->uuid
));
755 if (map
->map_state
== IMSM_T_STATE_UNINITIALIZED
|| dev
->vol
.dirty
)
756 info
->resync_start
= 0;
757 else if (dev
->vol
.migr_state
)
758 info
->resync_start
= __le32_to_cpu(dev
->vol
.curr_migr_unit
);
760 info
->resync_start
= ~0ULL;
762 strncpy(info
->name
, (char *) dev
->volume
, MAX_RAID_SERIAL_LEN
);
763 info
->name
[MAX_RAID_SERIAL_LEN
] = 0;
765 info
->array
.major_version
= -1;
766 info
->array
.minor_version
= -2;
767 sprintf(info
->text_version
, "/%s/%d",
768 devnum2devname(st
->container_dev
),
769 info
->container_member
);
770 info
->safe_mode_delay
= 4000; /* 4 secs like the Matrix driver */
771 uuid_from_super_imsm(st
, info
->uuid
);
775 static void getinfo_super_imsm(struct supertype
*st
, struct mdinfo
*info
)
777 struct intel_super
*super
= st
->sb
;
778 struct imsm_disk
*disk
;
781 if (super
->current_vol
>= 0) {
782 getinfo_super_imsm_volume(st
, info
);
786 /* Set raid_disks to zero so that Assemble will always pull in valid
789 info
->array
.raid_disks
= 0;
790 info
->array
.level
= LEVEL_CONTAINER
;
791 info
->array
.layout
= 0;
792 info
->array
.md_minor
= -1;
793 info
->array
.ctime
= 0; /* N/A for imsm */
794 info
->array
.utime
= 0;
795 info
->array
.chunk_size
= 0;
797 info
->disk
.major
= 0;
798 info
->disk
.minor
= 0;
799 info
->disk
.raid_disk
= -1;
800 info
->reshape_active
= 0;
801 info
->array
.major_version
= -1;
802 info
->array
.minor_version
= -2;
803 strcpy(info
->text_version
, "imsm");
804 info
->safe_mode_delay
= 0;
805 info
->disk
.number
= -1;
806 info
->disk
.state
= 0;
810 __u32 reserved
= imsm_reserved_sectors(super
, super
->disks
);
812 disk
= &super
->disks
->disk
;
813 info
->disk
.number
= super
->disks
->index
;
814 info
->disk
.raid_disk
= super
->disks
->index
;
815 info
->data_offset
= __le32_to_cpu(disk
->total_blocks
) - reserved
;
816 info
->component_size
= reserved
;
817 s
= __le32_to_cpu(disk
->status
);
818 info
->disk
.state
= s
& CONFIGURED_DISK
? (1 << MD_DISK_ACTIVE
) : 0;
819 info
->disk
.state
|= s
& FAILED_DISK
? (1 << MD_DISK_FAULTY
) : 0;
820 info
->disk
.state
|= s
& USABLE_DISK
? (1 << MD_DISK_SYNC
) : 0;
822 uuid_from_super_imsm(st
, info
->uuid
);
825 static int update_super_imsm(struct supertype
*st
, struct mdinfo
*info
,
826 char *update
, char *devname
, int verbose
,
827 int uuid_set
, char *homehost
)
831 /* For 'assemble' and 'force' we need to return non-zero if any
832 * change was made. For others, the return value is ignored.
833 * Update options are:
834 * force-one : This device looks a bit old but needs to be included,
835 * update age info appropriately.
836 * assemble: clear any 'faulty' flag to allow this device to
838 * force-array: Array is degraded but being forced, mark it clean
839 * if that will be needed to assemble it.
841 * newdev: not used ????
842 * grow: Array has gained a new device - this is currently for
844 * resync: mark as dirty so a resync will happen.
845 * name: update the name - preserving the homehost
847 * Following are not relevant for this imsm:
848 * sparc2.2 : update from old dodgey metadata
849 * super-minor: change the preferred_minor number
850 * summaries: update redundant counters.
851 * uuid: Change the uuid of the array to match watch is given
852 * homehost: update the recorded homehost
853 * _reshape_progress: record new reshape_progress position.
856 //struct intel_super *super = st->sb;
857 //struct imsm_super *mpb = super->mpb;
859 if (strcmp(update
, "grow") == 0) {
861 if (strcmp(update
, "resync") == 0) {
862 /* dev->vol.dirty = 1; */
865 /* IMSM has no concept of UUID or homehost */
870 static size_t disks_to_mpb_size(int disks
)
874 size
= sizeof(struct imsm_super
);
875 size
+= (disks
- 1) * sizeof(struct imsm_disk
);
876 size
+= 2 * sizeof(struct imsm_dev
);
877 /* up to 2 maps per raid device (-2 for imsm_maps in imsm_dev */
878 size
+= (4 - 2) * sizeof(struct imsm_map
);
879 /* 4 possible disk_ord_tbl's */
880 size
+= 4 * (disks
- 1) * sizeof(__u32
);
885 static __u64
avail_size_imsm(struct supertype
*st
, __u64 devsize
)
887 if (devsize
< (MPB_SECTOR_CNT
+ IMSM_RESERVED_SECTORS
))
890 return devsize
- (MPB_SECTOR_CNT
+ IMSM_RESERVED_SECTORS
);
893 static int compare_super_imsm(struct supertype
*st
, struct supertype
*tst
)
897 * 0 same, or first was empty, and second was copied
898 * 1 second had wrong number
902 struct intel_super
*first
= st
->sb
;
903 struct intel_super
*sec
= tst
->sb
;
911 if (memcmp(first
->anchor
->sig
, sec
->anchor
->sig
, MAX_SIGNATURE_LENGTH
) != 0)
914 /* if an anchor does not have num_raid_devs set then it is a free
917 if (first
->anchor
->num_raid_devs
> 0 &&
918 sec
->anchor
->num_raid_devs
> 0) {
919 if (first
->anchor
->family_num
!= sec
->anchor
->family_num
)
923 /* if 'first' is a spare promote it to a populated mpb with sec's
926 if (first
->anchor
->num_raid_devs
== 0 &&
927 sec
->anchor
->num_raid_devs
> 0) {
928 first
->anchor
->num_raid_devs
= sec
->anchor
->num_raid_devs
;
929 first
->anchor
->family_num
= sec
->anchor
->family_num
;
935 static void fd2devname(int fd
, char *name
)
944 if (fstat(fd
, &st
) != 0)
946 sprintf(path
, "/sys/dev/block/%d:%d",
947 major(st
.st_rdev
), minor(st
.st_rdev
));
949 rv
= readlink(path
, dname
, sizeof(dname
));
954 nm
= strrchr(dname
, '/');
956 snprintf(name
, MAX_RAID_SERIAL_LEN
, "/dev/%s", nm
);
960 extern int scsi_get_serial(int fd
, void *buf
, size_t buf_len
);
962 static int imsm_read_serial(int fd
, char *devname
,
963 __u8 serial
[MAX_RAID_SERIAL_LEN
])
965 unsigned char scsi_serial
[255];
971 memset(scsi_serial
, 0, sizeof(scsi_serial
));
973 rv
= scsi_get_serial(fd
, scsi_serial
, sizeof(scsi_serial
));
975 if (rv
&& imsm_env_devname_as_serial()) {
976 memset(serial
, 0, MAX_RAID_SERIAL_LEN
);
977 fd2devname(fd
, (char *) serial
);
984 Name
": Failed to retrieve serial for %s\n",
989 /* trim leading whitespace */
990 rsp_len
= scsi_serial
[3];
991 rsp_buf
= (char *) &scsi_serial
[4];
996 /* truncate len to the end of rsp_buf if necessary */
997 if (c
+ MAX_RAID_SERIAL_LEN
> rsp_buf
+ rsp_len
)
998 len
= rsp_len
- (c
- rsp_buf
);
1000 len
= MAX_RAID_SERIAL_LEN
;
1002 /* initialize the buffer and copy rsp_buf characters */
1003 memset(serial
, 0, MAX_RAID_SERIAL_LEN
);
1004 memcpy(serial
, c
, len
);
1006 /* trim trailing whitespace starting with the last character copied */
1007 c
= (char *) &serial
[len
- 1];
1008 while (isspace(*c
) || *c
== '\0')
1014 static int serialcmp(__u8
*s1
, __u8
*s2
)
1016 return strncmp((char *) s1
, (char *) s2
, MAX_RAID_SERIAL_LEN
);
1019 static void serialcpy(__u8
*dest
, __u8
*src
)
1021 strncpy((char *) dest
, (char *) src
, MAX_RAID_SERIAL_LEN
);
1025 load_imsm_disk(int fd
, struct intel_super
*super
, char *devname
, int keep_fd
)
1032 __u8 serial
[MAX_RAID_SERIAL_LEN
];
1034 rv
= imsm_read_serial(fd
, devname
, serial
);
1039 /* check if this is a disk we have seen before. it may be a spare in
1040 * super->disks while the current anchor believes it is a raid member,
1041 * check if we need to update dl->index
1043 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
1044 if (serialcmp(dl
->serial
, serial
) == 0)
1048 dl
= malloc(sizeof(*dl
));
1055 Name
": failed to allocate disk buffer for %s\n",
1062 dl
->major
= major(stb
.st_rdev
);
1063 dl
->minor
= minor(stb
.st_rdev
);
1064 dl
->next
= super
->disks
;
1065 dl
->fd
= keep_fd
? fd
: -1;
1066 dl
->devname
= devname
? strdup(devname
) : NULL
;
1067 serialcpy(dl
->serial
, serial
);
1069 } else if (keep_fd
) {
1074 /* look up this disk's index in the current anchor */
1075 for (i
= 0; i
< super
->anchor
->num_disks
; i
++) {
1076 struct imsm_disk
*disk_iter
;
1078 disk_iter
= __get_imsm_disk(super
->anchor
, i
);
1080 if (serialcmp(disk_iter
->serial
, dl
->serial
) == 0) {
1083 dl
->disk
= *disk_iter
;
1084 status
= __le32_to_cpu(dl
->disk
.status
);
1085 /* only set index on disks that are a member of a
1086 * populated contianer, i.e. one with raid_devs
1088 if (status
& FAILED_DISK
)
1090 else if (status
& SPARE_DISK
)
1105 static void imsm_copy_dev(struct imsm_dev
*dest
, struct imsm_dev
*src
)
1107 memcpy(dest
, src
, sizeof_imsm_dev(src
, 0));
1111 /* When migrating map0 contains the 'destination' state while map1
1112 * contains the current state. When not migrating map0 contains the
1113 * current state. This routine assumes that map[0].map_state is set to
1114 * the current array state before being called.
1116 * Migration is indicated by one of the following states
1117 * 1/ Idle (migr_state=0 map0state=normal||unitialized||degraded||failed)
1118 * 2/ Initialize (migr_state=1 migr_type=0 map0state=normal
1119 * map1state=unitialized)
1120 * 3/ Verify (Resync) (migr_state=1 migr_type=1 map0state=normal
1122 * 4/ Rebuild (migr_state=1 migr_type=1 map0state=normal
1123 * map1state=degraded)
1125 static void migrate(struct imsm_dev
*dev
, __u8 to_state
, int rebuild_resync
)
1127 struct imsm_map
*dest
;
1128 struct imsm_map
*src
= get_imsm_map(dev
, 0);
1130 dev
->vol
.migr_state
= 1;
1131 dev
->vol
.migr_type
= rebuild_resync
;
1132 dev
->vol
.curr_migr_unit
= 0;
1133 dest
= get_imsm_map(dev
, 1);
1135 memcpy(dest
, src
, sizeof_imsm_map(src
));
1136 src
->map_state
= to_state
;
1139 static void end_migration(struct imsm_dev
*dev
, __u8 map_state
)
1141 struct imsm_map
*map
= get_imsm_map(dev
, 0);
1143 dev
->vol
.migr_state
= 0;
1144 dev
->vol
.curr_migr_unit
= 0;
1145 map
->map_state
= map_state
;
1149 static int parse_raid_devices(struct intel_super
*super
)
1152 struct imsm_dev
*dev_new
;
1153 size_t len
, len_migr
;
1154 size_t space_needed
= 0;
1155 struct imsm_super
*mpb
= super
->anchor
;
1157 for (i
= 0; i
< super
->anchor
->num_raid_devs
; i
++) {
1158 struct imsm_dev
*dev_iter
= __get_imsm_dev(super
->anchor
, i
);
1160 len
= sizeof_imsm_dev(dev_iter
, 0);
1161 len_migr
= sizeof_imsm_dev(dev_iter
, 1);
1163 space_needed
+= len_migr
- len
;
1165 dev_new
= malloc(len_migr
);
1168 imsm_copy_dev(dev_new
, dev_iter
);
1169 super
->dev_tbl
[i
] = dev_new
;
1172 /* ensure that super->buf is large enough when all raid devices
1175 if (__le32_to_cpu(mpb
->mpb_size
) + space_needed
> super
->len
) {
1178 len
= ROUND_UP(__le32_to_cpu(mpb
->mpb_size
) + space_needed
, 512);
1179 if (posix_memalign(&buf
, 512, len
) != 0)
1182 memcpy(buf
, super
->buf
, len
);
1191 /* retrieve a pointer to the bbm log which starts after all raid devices */
1192 struct bbm_log
*__get_imsm_bbm_log(struct imsm_super
*mpb
)
1196 if (__le32_to_cpu(mpb
->bbm_log_size
)) {
1198 ptr
+= mpb
->mpb_size
- __le32_to_cpu(mpb
->bbm_log_size
);
1204 static void __free_imsm(struct intel_super
*super
, int free_disks
);
1206 /* load_imsm_mpb - read matrix metadata
1207 * allocates super->mpb to be freed by free_super
1209 static int load_imsm_mpb(int fd
, struct intel_super
*super
, char *devname
)
1211 unsigned long long dsize
;
1212 unsigned long long sectors
;
1214 struct imsm_super
*anchor
;
1218 get_dev_size(fd
, NULL
, &dsize
);
1220 if (lseek64(fd
, dsize
- (512 * 2), SEEK_SET
) < 0) {
1223 Name
": Cannot seek to anchor block on %s: %s\n",
1224 devname
, strerror(errno
));
1228 if (posix_memalign((void**)&anchor
, 512, 512) != 0) {
1231 Name
": Failed to allocate imsm anchor buffer"
1232 " on %s\n", devname
);
1235 if (read(fd
, anchor
, 512) != 512) {
1238 Name
": Cannot read anchor block on %s: %s\n",
1239 devname
, strerror(errno
));
1244 if (strncmp((char *) anchor
->sig
, MPB_SIGNATURE
, MPB_SIG_LEN
) != 0) {
1247 Name
": no IMSM anchor on %s\n", devname
);
1252 __free_imsm(super
, 0);
1253 super
->len
= ROUND_UP(anchor
->mpb_size
, 512);
1254 if (posix_memalign(&super
->buf
, 512, super
->len
) != 0) {
1257 Name
": unable to allocate %zu byte mpb buffer\n",
1262 memcpy(super
->buf
, anchor
, 512);
1264 sectors
= mpb_sectors(anchor
) - 1;
1267 rc
= load_imsm_disk(fd
, super
, devname
, 0);
1269 rc
= parse_raid_devices(super
);
1273 /* read the extended mpb */
1274 if (lseek64(fd
, dsize
- (512 * (2 + sectors
)), SEEK_SET
) < 0) {
1277 Name
": Cannot seek to extended mpb on %s: %s\n",
1278 devname
, strerror(errno
));
1282 if (read(fd
, super
->buf
+ 512, super
->len
- 512) != super
->len
- 512) {
1285 Name
": Cannot read extended mpb on %s: %s\n",
1286 devname
, strerror(errno
));
1290 check_sum
= __gen_imsm_checksum(super
->anchor
);
1291 if (check_sum
!= __le32_to_cpu(super
->anchor
->check_sum
)) {
1294 Name
": IMSM checksum %x != %x on %s\n",
1295 check_sum
, __le32_to_cpu(super
->anchor
->check_sum
),
1300 /* FIXME the BBM log is disk specific so we cannot use this global
1301 * buffer for all disks. Ok for now since we only look at the global
1302 * bbm_log_size parameter to gate assembly
1304 super
->bbm_log
= __get_imsm_bbm_log(super
->anchor
);
1306 rc
= load_imsm_disk(fd
, super
, devname
, 0);
1308 rc
= parse_raid_devices(super
);
1313 static void __free_imsm_disk(struct dl
*d
)
1322 static void free_imsm_disks(struct intel_super
*super
)
1324 while (super
->disks
) {
1325 struct dl
*d
= super
->disks
;
1327 super
->disks
= d
->next
;
1328 __free_imsm_disk(d
);
1332 /* free all the pieces hanging off of a super pointer */
1333 static void __free_imsm(struct intel_super
*super
, int free_disks
)
1342 free_imsm_disks(super
);
1343 for (i
= 0; i
< IMSM_MAX_RAID_DEVS
; i
++)
1344 if (super
->dev_tbl
[i
]) {
1345 free(super
->dev_tbl
[i
]);
1346 super
->dev_tbl
[i
] = NULL
;
1350 static void free_imsm(struct intel_super
*super
)
1352 __free_imsm(super
, 1);
1356 static void free_super_imsm(struct supertype
*st
)
1358 struct intel_super
*super
= st
->sb
;
1367 static struct intel_super
*alloc_super(int creating_imsm
)
1369 struct intel_super
*super
= malloc(sizeof(*super
));
1372 memset(super
, 0, sizeof(*super
));
1373 super
->creating_imsm
= creating_imsm
;
1374 super
->current_vol
= -1;
1381 static int load_super_imsm_all(struct supertype
*st
, int fd
, void **sbp
,
1382 char *devname
, int keep_fd
)
1385 struct intel_super
*super
;
1386 struct mdinfo
*sd
, *best
= NULL
;
1393 /* check if this disk is a member of an active array */
1394 sra
= sysfs_read(fd
, 0, GET_LEVEL
|GET_VERSION
|GET_DEVS
|GET_STATE
);
1398 if (sra
->array
.major_version
!= -1 ||
1399 sra
->array
.minor_version
!= -2 ||
1400 strcmp(sra
->text_version
, "imsm") != 0)
1403 super
= alloc_super(0);
1407 /* find the most up to date disk in this array, skipping spares */
1408 for (sd
= sra
->devs
; sd
; sd
= sd
->next
) {
1409 sprintf(nm
, "%d:%d", sd
->disk
.major
, sd
->disk
.minor
);
1410 dfd
= dev_open(nm
, keep_fd
? O_RDWR
: O_RDONLY
);
1415 rv
= load_imsm_mpb(dfd
, super
, NULL
);
1419 if (super
->anchor
->num_raid_devs
== 0)
1422 gen
= __le32_to_cpu(super
->anchor
->generation_num
);
1423 if (!best
|| gen
> bestgen
) {
1438 /* load the most up to date anchor */
1439 sprintf(nm
, "%d:%d", best
->disk
.major
, best
->disk
.minor
);
1440 dfd
= dev_open(nm
, O_RDONLY
);
1445 rv
= load_imsm_mpb(dfd
, super
, NULL
);
1452 /* re-parse the disk list with the current anchor */
1453 for (sd
= sra
->devs
; sd
; sd
= sd
->next
) {
1454 sprintf(nm
, "%d:%d", sd
->disk
.major
, sd
->disk
.minor
);
1455 dfd
= dev_open(nm
, keep_fd
? O_RDWR
: O_RDONLY
);
1460 load_imsm_disk(dfd
, super
, NULL
, keep_fd
);
1465 if (st
->subarray
[0]) {
1466 if (atoi(st
->subarray
) <= super
->anchor
->num_raid_devs
)
1467 super
->current_vol
= atoi(st
->subarray
);
1473 st
->container_dev
= fd2devnum(fd
);
1474 if (st
->ss
== NULL
) {
1475 st
->ss
= &super_imsm
;
1476 st
->minor_version
= 0;
1477 st
->max_devs
= IMSM_MAX_DEVICES
;
1479 st
->loaded_container
= 1;
1485 static int load_super_imsm(struct supertype
*st
, int fd
, char *devname
)
1487 struct intel_super
*super
;
1491 if (load_super_imsm_all(st
, fd
, &st
->sb
, devname
, 1) == 0)
1494 if (st
->subarray
[0])
1495 return 1; /* FIXME */
1497 super
= alloc_super(0);
1500 Name
": malloc of %zu failed.\n",
1505 rv
= load_imsm_mpb(fd
, super
, devname
);
1510 Name
": Failed to load all information "
1511 "sections on %s\n", devname
);
1517 if (st
->ss
== NULL
) {
1518 st
->ss
= &super_imsm
;
1519 st
->minor_version
= 0;
1520 st
->max_devs
= IMSM_MAX_DEVICES
;
1522 st
->loaded_container
= 0;
1527 static __u16
info_to_blocks_per_strip(mdu_array_info_t
*info
)
1529 if (info
->level
== 1)
1531 return info
->chunk_size
>> 9;
1534 static __u32
info_to_num_data_stripes(mdu_array_info_t
*info
)
1538 num_stripes
= (info
->size
* 2) / info_to_blocks_per_strip(info
);
1539 if (info
->level
== 1)
1545 static __u32
info_to_blocks_per_member(mdu_array_info_t
*info
)
1547 return (info
->size
* 2) & ~(info_to_blocks_per_strip(info
) - 1);
1550 static int init_super_imsm_volume(struct supertype
*st
, mdu_array_info_t
*info
,
1551 unsigned long long size
, char *name
,
1552 char *homehost
, int *uuid
)
1554 /* We are creating a volume inside a pre-existing container.
1555 * so st->sb is already set.
1557 struct intel_super
*super
= st
->sb
;
1558 struct imsm_super
*mpb
= super
->anchor
;
1559 struct imsm_dev
*dev
;
1560 struct imsm_vol
*vol
;
1561 struct imsm_map
*map
;
1562 int idx
= mpb
->num_raid_devs
;
1564 unsigned long long array_blocks
;
1566 size_t size_old
, size_new
;
1568 if (mpb
->num_raid_devs
>= 2) {
1569 fprintf(stderr
, Name
": This imsm-container already has the "
1570 "maximum of 2 volumes\n");
1574 /* ensure the mpb is large enough for the new data */
1575 size_old
= __le32_to_cpu(mpb
->mpb_size
);
1576 size_new
= disks_to_mpb_size(info
->nr_disks
);
1577 if (size_new
> size_old
) {
1579 size_t size_round
= ROUND_UP(size_new
, 512);
1581 if (posix_memalign(&mpb_new
, 512, size_round
) != 0) {
1582 fprintf(stderr
, Name
": could not allocate new mpb\n");
1585 memcpy(mpb_new
, mpb
, size_old
);
1588 super
->anchor
= mpb_new
;
1589 mpb
->mpb_size
= __cpu_to_le32(size_new
);
1590 memset(mpb_new
+ size_old
, 0, size_round
- size_old
);
1592 super
->current_vol
= idx
;
1593 /* when creating the first raid device in this container set num_disks
1594 * to zero, i.e. delete this spare and add raid member devices in
1595 * add_to_super_imsm_volume()
1597 if (super
->current_vol
== 0)
1599 sprintf(st
->subarray
, "%d", idx
);
1600 dev
= malloc(sizeof(*dev
) + sizeof(__u32
) * (info
->raid_disks
- 1));
1602 fprintf(stderr
, Name
": could not allocate raid device\n");
1605 strncpy((char *) dev
->volume
, name
, MAX_RAID_SERIAL_LEN
);
1606 array_blocks
= calc_array_size(info
->level
, info
->raid_disks
,
1607 info
->layout
, info
->chunk_size
,
1609 dev
->size_low
= __cpu_to_le32((__u32
) array_blocks
);
1610 dev
->size_high
= __cpu_to_le32((__u32
) (array_blocks
>> 32));
1611 dev
->status
= __cpu_to_le32(0);
1612 dev
->reserved_blocks
= __cpu_to_le32(0);
1614 vol
->migr_state
= 0;
1617 vol
->curr_migr_unit
= 0;
1618 for (i
= 0; i
< idx
; i
++) {
1619 struct imsm_dev
*prev
= get_imsm_dev(super
, i
);
1620 struct imsm_map
*pmap
= get_imsm_map(prev
, 0);
1622 offset
+= __le32_to_cpu(pmap
->blocks_per_member
);
1623 offset
+= IMSM_RESERVED_SECTORS
;
1625 map
= get_imsm_map(dev
, 0);
1626 map
->pba_of_lba0
= __cpu_to_le32(offset
);
1627 map
->blocks_per_member
= __cpu_to_le32(info_to_blocks_per_member(info
));
1628 map
->blocks_per_strip
= __cpu_to_le16(info_to_blocks_per_strip(info
));
1629 map
->num_data_stripes
= __cpu_to_le32(info_to_num_data_stripes(info
));
1630 map
->map_state
= info
->level
? IMSM_T_STATE_UNINITIALIZED
:
1631 IMSM_T_STATE_NORMAL
;
1633 if (info
->level
== 1 && info
->raid_disks
> 2) {
1634 fprintf(stderr
, Name
": imsm does not support more than 2 disks"
1635 "in a raid1 volume\n");
1638 if (info
->level
== 10)
1639 map
->raid_level
= 1;
1641 map
->raid_level
= info
->level
;
1643 map
->num_members
= info
->raid_disks
;
1644 for (i
= 0; i
< map
->num_members
; i
++) {
1645 /* initialized in add_to_super */
1646 set_imsm_ord_tbl_ent(map
, i
, 0);
1648 mpb
->num_raid_devs
++;
1649 super
->dev_tbl
[super
->current_vol
] = dev
;
1654 static int init_super_imsm(struct supertype
*st
, mdu_array_info_t
*info
,
1655 unsigned long long size
, char *name
,
1656 char *homehost
, int *uuid
)
1658 /* This is primarily called by Create when creating a new array.
1659 * We will then get add_to_super called for each component, and then
1660 * write_init_super called to write it out to each device.
1661 * For IMSM, Create can create on fresh devices or on a pre-existing
1663 * To create on a pre-existing array a different method will be called.
1664 * This one is just for fresh drives.
1666 struct intel_super
*super
;
1667 struct imsm_super
*mpb
;
1675 return init_super_imsm_volume(st
, info
, size
, name
, homehost
,
1678 super
= alloc_super(1);
1681 mpb_size
= disks_to_mpb_size(info
->nr_disks
);
1682 if (posix_memalign(&super
->buf
, 512, mpb_size
) != 0) {
1687 memset(mpb
, 0, mpb_size
);
1689 memcpy(mpb
->sig
, MPB_SIGNATURE
, strlen(MPB_SIGNATURE
));
1690 memcpy(mpb
->sig
+ strlen(MPB_SIGNATURE
), MPB_VERSION_RAID5
,
1691 strlen(MPB_VERSION_RAID5
));
1692 mpb
->mpb_size
= mpb_size
;
1699 static void add_to_super_imsm_volume(struct supertype
*st
, mdu_disk_info_t
*dk
,
1700 int fd
, char *devname
)
1702 struct intel_super
*super
= st
->sb
;
1703 struct imsm_super
*mpb
= super
->anchor
;
1705 struct imsm_dev
*dev
;
1706 struct imsm_map
*map
;
1709 dev
= get_imsm_dev(super
, super
->current_vol
);
1710 map
= get_imsm_map(dev
, 0);
1712 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
1713 if (dl
->major
== dk
->major
&&
1714 dl
->minor
== dk
->minor
)
1717 if (!dl
|| ! (dk
->state
& (1<<MD_DISK_SYNC
)))
1720 /* add a pristine spare to the metadata */
1721 if (dl
->index
< 0) {
1722 dl
->index
= super
->anchor
->num_disks
;
1723 super
->anchor
->num_disks
++;
1725 set_imsm_ord_tbl_ent(map
, dk
->number
, dl
->index
);
1726 status
= CONFIGURED_DISK
| USABLE_DISK
;
1727 dl
->disk
.status
= __cpu_to_le32(status
);
1729 /* if we are creating the first raid device update the family number */
1730 if (super
->current_vol
== 0) {
1732 struct imsm_dev
*_dev
= __get_imsm_dev(mpb
, 0);
1733 struct imsm_disk
*_disk
= __get_imsm_disk(mpb
, dl
->index
);
1737 sum
= __gen_imsm_checksum(mpb
);
1738 mpb
->family_num
= __cpu_to_le32(sum
);
1742 static void add_to_super_imsm(struct supertype
*st
, mdu_disk_info_t
*dk
,
1743 int fd
, char *devname
)
1745 struct intel_super
*super
= st
->sb
;
1747 unsigned long long size
;
1752 if (super
->current_vol
>= 0) {
1753 add_to_super_imsm_volume(st
, dk
, fd
, devname
);
1758 dd
= malloc(sizeof(*dd
));
1761 Name
": malloc failed %s:%d.\n", __func__
, __LINE__
);
1764 memset(dd
, 0, sizeof(*dd
));
1765 dd
->major
= major(stb
.st_rdev
);
1766 dd
->minor
= minor(stb
.st_rdev
);
1768 dd
->devname
= devname
? strdup(devname
) : NULL
;
1770 rv
= imsm_read_serial(fd
, devname
, dd
->serial
);
1773 Name
": failed to retrieve scsi serial, aborting\n");
1778 get_dev_size(fd
, NULL
, &size
);
1780 status
= USABLE_DISK
| SPARE_DISK
;
1781 serialcpy(dd
->disk
.serial
, dd
->serial
);
1782 dd
->disk
.total_blocks
= __cpu_to_le32(size
);
1783 dd
->disk
.status
= __cpu_to_le32(status
);
1784 if (sysfs_disk_to_scsi_id(fd
, &id
) == 0)
1785 dd
->disk
.scsi_id
= __cpu_to_le32(id
);
1787 dd
->disk
.scsi_id
= __cpu_to_le32(0);
1789 if (st
->update_tail
) {
1790 dd
->next
= super
->add
;
1793 dd
->next
= super
->disks
;
1798 static int store_imsm_mpb(int fd
, struct intel_super
*super
);
1800 /* spare records have their own family number and do not have any defined raid
1803 static int write_super_imsm_spares(struct intel_super
*super
, int doclose
)
1805 struct imsm_super mpb_save
;
1806 struct imsm_super
*mpb
= super
->anchor
;
1811 mpb
->num_raid_devs
= 0;
1813 mpb
->mpb_size
= sizeof(struct imsm_super
);
1814 mpb
->generation_num
= __cpu_to_le32(1UL);
1816 for (d
= super
->disks
; d
; d
= d
->next
) {
1820 mpb
->disk
[0] = d
->disk
;
1821 sum
= __gen_imsm_checksum(mpb
);
1822 mpb
->family_num
= __cpu_to_le32(sum
);
1823 sum
= __gen_imsm_checksum(mpb
);
1824 mpb
->check_sum
= __cpu_to_le32(sum
);
1826 if (store_imsm_mpb(d
->fd
, super
)) {
1827 fprintf(stderr
, "%s: failed for device %d:%d %s\n",
1828 __func__
, d
->major
, d
->minor
, strerror(errno
));
1842 static int write_super_imsm(struct intel_super
*super
, int doclose
)
1844 struct imsm_super
*mpb
= super
->anchor
;
1850 __u32 mpb_size
= sizeof(struct imsm_super
) - sizeof(struct imsm_disk
);
1852 /* 'generation' is incremented everytime the metadata is written */
1853 generation
= __le32_to_cpu(mpb
->generation_num
);
1855 mpb
->generation_num
= __cpu_to_le32(generation
);
1857 mpb_size
+= sizeof(struct imsm_disk
) * mpb
->num_disks
;
1858 for (d
= super
->disks
; d
; d
= d
->next
) {
1862 mpb
->disk
[d
->index
] = d
->disk
;
1865 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
1866 struct imsm_dev
*dev
= __get_imsm_dev(mpb
, i
);
1868 imsm_copy_dev(dev
, super
->dev_tbl
[i
]);
1869 mpb_size
+= sizeof_imsm_dev(dev
, 0);
1871 mpb_size
+= __le32_to_cpu(mpb
->bbm_log_size
);
1872 mpb
->mpb_size
= __cpu_to_le32(mpb_size
);
1874 /* recalculate checksum */
1875 sum
= __gen_imsm_checksum(mpb
);
1876 mpb
->check_sum
= __cpu_to_le32(sum
);
1878 /* write the mpb for disks that compose raid devices */
1879 for (d
= super
->disks
; d
; d
= d
->next
) {
1882 if (store_imsm_mpb(d
->fd
, super
))
1883 fprintf(stderr
, "%s: failed for device %d:%d %s\n",
1884 __func__
, d
->major
, d
->minor
, strerror(errno
));
1892 return write_super_imsm_spares(super
, doclose
);
1898 static int create_array(struct supertype
*st
)
1901 struct imsm_update_create_array
*u
;
1902 struct intel_super
*super
= st
->sb
;
1903 struct imsm_dev
*dev
= get_imsm_dev(super
, super
->current_vol
);
1905 len
= sizeof(*u
) - sizeof(*dev
) + sizeof_imsm_dev(dev
, 0);
1908 fprintf(stderr
, "%s: failed to allocate update buffer\n",
1913 u
->type
= update_create_array
;
1914 u
->dev_idx
= super
->current_vol
;
1915 imsm_copy_dev(&u
->dev
, dev
);
1916 append_metadata_update(st
, u
, len
);
1921 static int _add_disk(struct supertype
*st
)
1923 struct intel_super
*super
= st
->sb
;
1925 struct imsm_update_add_disk
*u
;
1933 fprintf(stderr
, "%s: failed to allocate update buffer\n",
1938 u
->type
= update_add_disk
;
1939 append_metadata_update(st
, u
, len
);
1944 static int write_init_super_imsm(struct supertype
*st
)
1946 if (st
->update_tail
) {
1947 /* queue the recently created array / added disk
1948 * as a metadata update */
1949 struct intel_super
*super
= st
->sb
;
1953 /* determine if we are creating a volume or adding a disk */
1954 if (super
->current_vol
< 0) {
1955 /* in the add disk case we are running in mdmon
1956 * context, so don't close fd's
1958 return _add_disk(st
);
1960 rv
= create_array(st
);
1962 for (d
= super
->disks
; d
; d
= d
->next
) {
1969 return write_super_imsm(st
->sb
, 1);
1973 static int store_zero_imsm(struct supertype
*st
, int fd
)
1975 unsigned long long dsize
;
1978 get_dev_size(fd
, NULL
, &dsize
);
1980 /* first block is stored on second to last sector of the disk */
1981 if (lseek64(fd
, dsize
- (512 * 2), SEEK_SET
) < 0)
1984 if (posix_memalign(&buf
, 512, 512) != 0)
1987 memset(buf
, 0, 512);
1988 if (write(fd
, buf
, 512) != 512)
1993 static int imsm_bbm_log_size(struct imsm_super
*mpb
)
1995 return __le32_to_cpu(mpb
->bbm_log_size
);
1999 static int validate_geometry_imsm_container(struct supertype
*st
, int level
,
2000 int layout
, int raiddisks
, int chunk
,
2001 unsigned long long size
, char *dev
,
2002 unsigned long long *freesize
,
2006 unsigned long long ldsize
;
2008 if (level
!= LEVEL_CONTAINER
)
2013 fd
= open(dev
, O_RDONLY
|O_EXCL
, 0);
2016 fprintf(stderr
, Name
": imsm: Cannot open %s: %s\n",
2017 dev
, strerror(errno
));
2020 if (!get_dev_size(fd
, dev
, &ldsize
)) {
2026 *freesize
= avail_size_imsm(st
, ldsize
>> 9);
2031 /* validate_geometry_imsm_volume - lifted from validate_geometry_ddf_bvd
2032 * FIX ME add ahci details
2034 static int validate_geometry_imsm_volume(struct supertype
*st
, int level
,
2035 int layout
, int raiddisks
, int chunk
,
2036 unsigned long long size
, char *dev
,
2037 unsigned long long *freesize
,
2041 struct intel_super
*super
= st
->sb
;
2043 unsigned long long pos
= 0;
2044 unsigned long long maxsize
;
2048 if (level
== LEVEL_CONTAINER
)
2051 if (level
== 1 && raiddisks
> 2) {
2053 fprintf(stderr
, Name
": imsm does not support more "
2054 "than 2 in a raid1 configuration\n");
2058 /* We must have the container info already read in. */
2063 /* General test: make sure there is space for
2064 * 'raiddisks' device extents of size 'size' at a given
2067 unsigned long long minsize
= size
*2 /* convert to blocks */;
2068 unsigned long long start_offset
= ~0ULL;
2071 minsize
= MPB_SECTOR_CNT
+ IMSM_RESERVED_SECTORS
;
2072 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
2077 e
= get_extents(super
, dl
);
2080 unsigned long long esize
;
2081 esize
= e
[i
].start
- pos
;
2082 if (esize
>= minsize
)
2084 if (found
&& start_offset
== ~0ULL) {
2087 } else if (found
&& pos
!= start_offset
) {
2091 pos
= e
[i
].start
+ e
[i
].size
;
2093 } while (e
[i
-1].size
);
2098 if (dcnt
< raiddisks
) {
2100 fprintf(stderr
, Name
": imsm: Not enough "
2101 "devices with space for this array "
2108 /* This device must be a member of the set */
2109 if (stat(dev
, &stb
) < 0)
2111 if ((S_IFMT
& stb
.st_mode
) != S_IFBLK
)
2113 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
2114 if (dl
->major
== major(stb
.st_rdev
) &&
2115 dl
->minor
== minor(stb
.st_rdev
))
2120 fprintf(stderr
, Name
": %s is not in the "
2121 "same imsm set\n", dev
);
2124 e
= get_extents(super
, dl
);
2128 unsigned long long esize
;
2129 esize
= e
[i
].start
- pos
;
2130 if (esize
>= maxsize
)
2132 pos
= e
[i
].start
+ e
[i
].size
;
2134 } while (e
[i
-1].size
);
2135 *freesize
= maxsize
;
2140 static int validate_geometry_imsm(struct supertype
*st
, int level
, int layout
,
2141 int raiddisks
, int chunk
, unsigned long long size
,
2142 char *dev
, unsigned long long *freesize
,
2148 /* if given unused devices create a container
2149 * if given given devices in a container create a member volume
2151 if (level
== LEVEL_CONTAINER
) {
2152 /* Must be a fresh device to add to a container */
2153 return validate_geometry_imsm_container(st
, level
, layout
,
2154 raiddisks
, chunk
, size
,
2160 /* creating in a given container */
2161 return validate_geometry_imsm_volume(st
, level
, layout
,
2162 raiddisks
, chunk
, size
,
2163 dev
, freesize
, verbose
);
2166 /* limit creation to the following levels */
2178 /* This device needs to be a device in an 'imsm' container */
2179 fd
= open(dev
, O_RDONLY
|O_EXCL
, 0);
2183 Name
": Cannot create this array on device %s\n",
2188 if (errno
!= EBUSY
|| (fd
= open(dev
, O_RDONLY
, 0)) < 0) {
2190 fprintf(stderr
, Name
": Cannot open %s: %s\n",
2191 dev
, strerror(errno
));
2194 /* Well, it is in use by someone, maybe an 'imsm' container. */
2195 cfd
= open_container(fd
);
2199 fprintf(stderr
, Name
": Cannot use %s: It is busy\n",
2203 sra
= sysfs_read(cfd
, 0, GET_VERSION
);
2205 if (sra
&& sra
->array
.major_version
== -1 &&
2206 strcmp(sra
->text_version
, "imsm") == 0) {
2207 /* This is a member of a imsm container. Load the container
2208 * and try to create a volume
2210 struct intel_super
*super
;
2212 if (load_super_imsm_all(st
, cfd
, (void **) &super
, NULL
, 1) == 0) {
2214 st
->container_dev
= fd2devnum(cfd
);
2216 return validate_geometry_imsm_volume(st
, level
, layout
,
2222 } else /* may belong to another container */
2227 #endif /* MDASSEMBLE */
2229 static struct mdinfo
*container_content_imsm(struct supertype
*st
)
2231 /* Given a container loaded by load_super_imsm_all,
2232 * extract information about all the arrays into
2235 * For each imsm_dev create an mdinfo, fill it in,
2236 * then look for matching devices in super->disks
2237 * and create appropriate device mdinfo.
2239 struct intel_super
*super
= st
->sb
;
2240 struct imsm_super
*mpb
= super
->anchor
;
2241 struct mdinfo
*rest
= NULL
;
2244 /* do not assemble arrays that might have bad blocks */
2245 if (imsm_bbm_log_size(super
->anchor
)) {
2246 fprintf(stderr
, Name
": BBM log found in metadata. "
2247 "Cannot activate array(s).\n");
2251 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
2252 struct imsm_dev
*dev
= get_imsm_dev(super
, i
);
2253 struct imsm_map
*map
= get_imsm_map(dev
, 0);
2254 struct mdinfo
*this;
2257 this = malloc(sizeof(*this));
2258 memset(this, 0, sizeof(*this));
2261 super
->current_vol
= i
;
2262 getinfo_super_imsm_volume(st
, this);
2263 for (slot
= 0 ; slot
< map
->num_members
; slot
++) {
2264 struct mdinfo
*info_d
;
2272 idx
= get_imsm_disk_idx(dev
, slot
);
2273 ord
= get_imsm_ord_tbl_ent(dev
, slot
);
2274 for (d
= super
->disks
; d
; d
= d
->next
)
2275 if (d
->index
== idx
)
2281 s
= d
? __le32_to_cpu(d
->disk
.status
) : 0;
2282 if (s
& FAILED_DISK
)
2284 if (!(s
& USABLE_DISK
))
2286 if (ord
& IMSM_ORD_REBUILD
)
2290 * if we skip some disks the array will be assmebled degraded;
2291 * reset resync start to avoid a dirty-degraded situation
2293 * FIXME handle dirty degraded
2295 if (skip
&& !dev
->vol
.dirty
)
2296 this->resync_start
= ~0ULL;
2300 info_d
= malloc(sizeof(*info_d
));
2302 fprintf(stderr
, Name
": failed to allocate disk"
2303 " for volume %s\n", (char *) dev
->volume
);
2308 memset(info_d
, 0, sizeof(*info_d
));
2309 info_d
->next
= this->devs
;
2310 this->devs
= info_d
;
2312 info_d
->disk
.number
= d
->index
;
2313 info_d
->disk
.major
= d
->major
;
2314 info_d
->disk
.minor
= d
->minor
;
2315 info_d
->disk
.raid_disk
= slot
;
2317 this->array
.working_disks
++;
2319 info_d
->events
= __le32_to_cpu(mpb
->generation_num
);
2320 info_d
->data_offset
= __le32_to_cpu(map
->pba_of_lba0
);
2321 info_d
->component_size
= __le32_to_cpu(map
->blocks_per_member
);
2323 strcpy(info_d
->name
, d
->devname
);
2333 static int imsm_open_new(struct supertype
*c
, struct active_array
*a
,
2336 struct intel_super
*super
= c
->sb
;
2337 struct imsm_super
*mpb
= super
->anchor
;
2339 if (atoi(inst
) >= mpb
->num_raid_devs
) {
2340 fprintf(stderr
, "%s: subarry index %d, out of range\n",
2341 __func__
, atoi(inst
));
2345 dprintf("imsm: open_new %s\n", inst
);
2346 a
->info
.container_member
= atoi(inst
);
2350 static __u8
imsm_check_degraded(struct intel_super
*super
, struct imsm_dev
*dev
, int failed
)
2352 struct imsm_map
*map
= get_imsm_map(dev
, 0);
2355 return map
->map_state
== IMSM_T_STATE_UNINITIALIZED
?
2356 IMSM_T_STATE_UNINITIALIZED
: IMSM_T_STATE_NORMAL
;
2358 switch (get_imsm_raid_level(map
)) {
2360 return IMSM_T_STATE_FAILED
;
2363 if (failed
< map
->num_members
)
2364 return IMSM_T_STATE_DEGRADED
;
2366 return IMSM_T_STATE_FAILED
;
2371 * check to see if any mirrors have failed, otherwise we
2372 * are degraded. Even numbered slots are mirrored on
2378 for (i
= 0; i
< map
->num_members
; i
++) {
2379 __u32 ord
= get_imsm_ord_tbl_ent(dev
, i
);
2380 int idx
= ord_to_idx(ord
);
2381 struct imsm_disk
*disk
;
2383 /* reset the potential in-sync count on even-numbered
2384 * slots. num_copies is always 2 for imsm raid10
2389 disk
= get_imsm_disk(super
, idx
);
2391 __le32_to_cpu(disk
->status
) & FAILED_DISK
||
2392 ord
& IMSM_ORD_REBUILD
)
2395 /* no in-sync disks left in this mirror the
2399 return IMSM_T_STATE_FAILED
;
2402 return IMSM_T_STATE_DEGRADED
;
2406 return IMSM_T_STATE_DEGRADED
;
2408 return IMSM_T_STATE_FAILED
;
2414 return map
->map_state
;
2417 static int imsm_count_failed(struct intel_super
*super
, struct imsm_dev
*dev
)
2421 struct imsm_disk
*disk
;
2422 struct imsm_map
*map
= get_imsm_map(dev
, 0);
2424 for (i
= 0; i
< map
->num_members
; i
++) {
2425 __u32 ord
= get_imsm_ord_tbl_ent(dev
, i
);
2426 int idx
= ord_to_idx(ord
);
2428 disk
= get_imsm_disk(super
, idx
);
2430 __le32_to_cpu(disk
->status
) & FAILED_DISK
||
2431 ord
& IMSM_ORD_REBUILD
)
2438 static int is_resyncing(struct imsm_dev
*dev
)
2440 struct imsm_map
*migr_map
;
2442 if (!dev
->vol
.migr_state
)
2445 if (dev
->vol
.migr_type
== 0)
2448 migr_map
= get_imsm_map(dev
, 1);
2450 if (migr_map
->map_state
== IMSM_T_STATE_NORMAL
)
2456 static int is_rebuilding(struct imsm_dev
*dev
)
2458 struct imsm_map
*migr_map
;
2460 if (!dev
->vol
.migr_state
)
2463 if (dev
->vol
.migr_type
== 0)
2466 migr_map
= get_imsm_map(dev
, 1);
2468 if (migr_map
->map_state
== IMSM_T_STATE_DEGRADED
)
2474 /* Handle dirty -> clean transititions and resync. Degraded and rebuild
2475 * states are handled in imsm_set_disk() with one exception, when a
2476 * resync is stopped due to a new failure this routine will set the
2477 * 'degraded' state for the array.
2479 static int imsm_set_array_state(struct active_array
*a
, int consistent
)
2481 int inst
= a
->info
.container_member
;
2482 struct intel_super
*super
= a
->container
->sb
;
2483 struct imsm_dev
*dev
= get_imsm_dev(super
, inst
);
2484 struct imsm_map
*map
= get_imsm_map(dev
, 0);
2485 int failed
= imsm_count_failed(super
, dev
);
2486 __u8 map_state
= imsm_check_degraded(super
, dev
, failed
);
2488 if (consistent
== 2 &&
2489 (!is_resync_complete(a
) ||
2490 map_state
!= IMSM_T_STATE_NORMAL
||
2491 dev
->vol
.migr_state
))
2494 if (is_resync_complete(a
)) {
2495 /* complete intialization / resync,
2496 * recovery is completed in ->set_disk
2498 if (is_resyncing(dev
)) {
2499 dprintf("imsm: mark resync done\n");
2500 end_migration(dev
, map_state
);
2501 super
->updates_pending
++;
2503 } else if (!is_resyncing(dev
) && !failed
) {
2504 /* mark the start of the init process if nothing is failed */
2505 dprintf("imsm: mark resync start (%llu)\n", a
->resync_start
);
2506 map
->map_state
= map_state
;
2507 migrate(dev
, IMSM_T_STATE_NORMAL
,
2508 map
->map_state
== IMSM_T_STATE_NORMAL
);
2509 super
->updates_pending
++;
2512 /* check if we can update the migration checkpoint */
2513 if (dev
->vol
.migr_state
&&
2514 __le32_to_cpu(dev
->vol
.curr_migr_unit
) != a
->resync_start
) {
2515 dprintf("imsm: checkpoint migration (%llu)\n", a
->resync_start
);
2516 dev
->vol
.curr_migr_unit
= __cpu_to_le32(a
->resync_start
);
2517 super
->updates_pending
++;
2520 /* mark dirty / clean */
2521 if (dev
->vol
.dirty
!= !consistent
) {
2522 dprintf("imsm: mark '%s' (%llu)\n",
2523 consistent
? "clean" : "dirty", a
->resync_start
);
2528 super
->updates_pending
++;
2533 static void imsm_set_disk(struct active_array
*a
, int n
, int state
)
2535 int inst
= a
->info
.container_member
;
2536 struct intel_super
*super
= a
->container
->sb
;
2537 struct imsm_dev
*dev
= get_imsm_dev(super
, inst
);
2538 struct imsm_map
*map
= get_imsm_map(dev
, 0);
2539 struct imsm_disk
*disk
;
2545 if (n
> map
->num_members
)
2546 fprintf(stderr
, "imsm: set_disk %d out of range 0..%d\n",
2547 n
, map
->num_members
- 1);
2552 dprintf("imsm: set_disk %d:%x\n", n
, state
);
2554 ord
= get_imsm_ord_tbl_ent(dev
, n
);
2555 disk
= get_imsm_disk(super
, ord_to_idx(ord
));
2557 /* check for new failures */
2558 status
= __le32_to_cpu(disk
->status
);
2559 if ((state
& DS_FAULTY
) && !(status
& FAILED_DISK
)) {
2560 status
|= FAILED_DISK
;
2561 disk
->status
= __cpu_to_le32(status
);
2562 disk
->scsi_id
= __cpu_to_le32(~(__u32
)0);
2563 memmove(&disk
->serial
[0], &disk
->serial
[1], MAX_RAID_SERIAL_LEN
- 1);
2564 super
->updates_pending
++;
2566 /* check if in_sync */
2567 if (state
& DS_INSYNC
&& ord
& IMSM_ORD_REBUILD
) {
2568 struct imsm_map
*migr_map
= get_imsm_map(dev
, 1);
2570 set_imsm_ord_tbl_ent(migr_map
, n
, ord_to_idx(ord
));
2571 super
->updates_pending
++;
2574 failed
= imsm_count_failed(super
, dev
);
2575 map_state
= imsm_check_degraded(super
, dev
, failed
);
2577 /* check if recovery complete, newly degraded, or failed */
2578 if (map_state
== IMSM_T_STATE_NORMAL
&& is_rebuilding(dev
)) {
2579 end_migration(dev
, map_state
);
2580 super
->updates_pending
++;
2581 } else if (map_state
== IMSM_T_STATE_DEGRADED
&&
2582 map
->map_state
!= map_state
&&
2583 !dev
->vol
.migr_state
) {
2584 dprintf("imsm: mark degraded\n");
2585 map
->map_state
= map_state
;
2586 super
->updates_pending
++;
2587 } else if (map_state
== IMSM_T_STATE_FAILED
&&
2588 map
->map_state
!= map_state
) {
2589 dprintf("imsm: mark failed\n");
2590 end_migration(dev
, map_state
);
2591 super
->updates_pending
++;
2595 static int store_imsm_mpb(int fd
, struct intel_super
*super
)
2597 struct imsm_super
*mpb
= super
->anchor
;
2598 __u32 mpb_size
= __le32_to_cpu(mpb
->mpb_size
);
2599 unsigned long long dsize
;
2600 unsigned long long sectors
;
2602 get_dev_size(fd
, NULL
, &dsize
);
2604 if (mpb_size
> 512) {
2605 /* -1 to account for anchor */
2606 sectors
= mpb_sectors(mpb
) - 1;
2608 /* write the extended mpb to the sectors preceeding the anchor */
2609 if (lseek64(fd
, dsize
- (512 * (2 + sectors
)), SEEK_SET
) < 0)
2612 if (write(fd
, super
->buf
+ 512, 512 * sectors
) != 512 * sectors
)
2616 /* first block is stored on second to last sector of the disk */
2617 if (lseek64(fd
, dsize
- (512 * 2), SEEK_SET
) < 0)
2620 if (write(fd
, super
->buf
, 512) != 512)
2626 static void imsm_sync_metadata(struct supertype
*container
)
2628 struct intel_super
*super
= container
->sb
;
2630 if (!super
->updates_pending
)
2633 write_super_imsm(super
, 0);
2635 super
->updates_pending
= 0;
2638 static struct dl
*imsm_readd(struct intel_super
*super
, int idx
, struct active_array
*a
)
2640 struct imsm_dev
*dev
= get_imsm_dev(super
, a
->info
.container_member
);
2641 int i
= get_imsm_disk_idx(dev
, idx
);
2644 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
2648 if (dl
&& __le32_to_cpu(dl
->disk
.status
) & FAILED_DISK
)
2652 dprintf("%s: found %x:%x\n", __func__
, dl
->major
, dl
->minor
);
2657 static struct dl
*imsm_add_spare(struct intel_super
*super
, int slot
, struct active_array
*a
)
2659 struct imsm_dev
*dev
= get_imsm_dev(super
, a
->info
.container_member
);
2660 int idx
= get_imsm_disk_idx(dev
, slot
);
2661 struct imsm_map
*map
= get_imsm_map(dev
, 0);
2662 unsigned long long esize
;
2663 unsigned long long pos
;
2672 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
2673 /* If in this array, skip */
2674 for (d
= a
->info
.devs
; d
; d
= d
->next
)
2675 if (d
->state_fd
>= 0 &&
2676 d
->disk
.major
== dl
->major
&&
2677 d
->disk
.minor
== dl
->minor
) {
2678 dprintf("%x:%x already in array\n", dl
->major
, dl
->minor
);
2684 /* skip in use or failed drives */
2685 status
= __le32_to_cpu(dl
->disk
.status
);
2686 if (status
& FAILED_DISK
|| idx
== dl
->index
) {
2687 dprintf("%x:%x status ( %s%s)\n",
2688 dl
->major
, dl
->minor
,
2689 status
& FAILED_DISK
? "failed " : "",
2690 idx
== dl
->index
? "in use " : "");
2694 /* Does this unused device have the requisite free space?
2695 * We need a->info.component_size sectors
2697 ex
= get_extents(super
, dl
);
2699 dprintf("cannot get extents\n");
2705 array_start
= __le32_to_cpu(map
->pba_of_lba0
);
2708 /* check that we can start at pba_of_lba0 with
2709 * a->info.component_size of space
2711 esize
= ex
[j
].start
- pos
;
2712 if (array_start
>= pos
&&
2713 array_start
+ a
->info
.component_size
< ex
[j
].start
) {
2717 pos
= ex
[j
].start
+ ex
[j
].size
;
2720 } while (ex
[j
-1].size
);
2724 dprintf("%x:%x does not have %llu at %d\n",
2725 dl
->major
, dl
->minor
,
2726 a
->info
.component_size
,
2727 __le32_to_cpu(map
->pba_of_lba0
));
2737 static struct mdinfo
*imsm_activate_spare(struct active_array
*a
,
2738 struct metadata_update
**updates
)
2741 * Find a device with unused free space and use it to replace a
2742 * failed/vacant region in an array. We replace failed regions one a
2743 * array at a time. The result is that a new spare disk will be added
2744 * to the first failed array and after the monitor has finished
2745 * propagating failures the remainder will be consumed.
2747 * FIXME add a capability for mdmon to request spares from another
2751 struct intel_super
*super
= a
->container
->sb
;
2752 int inst
= a
->info
.container_member
;
2753 struct imsm_dev
*dev
= get_imsm_dev(super
, inst
);
2754 struct imsm_map
*map
= get_imsm_map(dev
, 0);
2755 int failed
= a
->info
.array
.raid_disks
;
2756 struct mdinfo
*rv
= NULL
;
2759 struct metadata_update
*mu
;
2761 struct imsm_update_activate_spare
*u
;
2765 for (d
= a
->info
.devs
; d
; d
= d
->next
) {
2766 if ((d
->curr_state
& DS_FAULTY
) &&
2768 /* wait for Removal to happen */
2770 if (d
->state_fd
>= 0)
2774 dprintf("imsm: activate spare: inst=%d failed=%d (%d) level=%d\n",
2775 inst
, failed
, a
->info
.array
.raid_disks
, a
->info
.array
.level
);
2776 if (imsm_check_degraded(super
, dev
, failed
) != IMSM_T_STATE_DEGRADED
)
2779 /* For each slot, if it is not working, find a spare */
2780 for (i
= 0; i
< a
->info
.array
.raid_disks
; i
++) {
2781 for (d
= a
->info
.devs
; d
; d
= d
->next
)
2782 if (d
->disk
.raid_disk
== i
)
2784 dprintf("found %d: %p %x\n", i
, d
, d
?d
->curr_state
:0);
2785 if (d
&& (d
->state_fd
>= 0))
2789 * OK, this device needs recovery. Try to re-add the previous
2790 * occupant of this slot, if this fails add a new spare
2792 dl
= imsm_readd(super
, i
, a
);
2794 dl
= imsm_add_spare(super
, i
, a
);
2798 /* found a usable disk with enough space */
2799 di
= malloc(sizeof(*di
));
2800 memset(di
, 0, sizeof(*di
));
2802 /* dl->index will be -1 in the case we are activating a
2803 * pristine spare. imsm_process_update() will create a
2804 * new index in this case. Once a disk is found to be
2805 * failed in all member arrays it is kicked from the
2808 di
->disk
.number
= dl
->index
;
2810 /* (ab)use di->devs to store a pointer to the device
2813 di
->devs
= (struct mdinfo
*) dl
;
2815 di
->disk
.raid_disk
= i
;
2816 di
->disk
.major
= dl
->major
;
2817 di
->disk
.minor
= dl
->minor
;
2819 di
->data_offset
= __le32_to_cpu(map
->pba_of_lba0
);
2820 di
->component_size
= a
->info
.component_size
;
2821 di
->container_member
= inst
;
2825 dprintf("%x:%x to be %d at %llu\n", dl
->major
, dl
->minor
,
2826 i
, di
->data_offset
);
2832 /* No spares found */
2834 /* Now 'rv' has a list of devices to return.
2835 * Create a metadata_update record to update the
2836 * disk_ord_tbl for the array
2838 mu
= malloc(sizeof(*mu
));
2839 mu
->buf
= malloc(sizeof(struct imsm_update_activate_spare
) * num_spares
);
2841 mu
->len
= sizeof(struct imsm_update_activate_spare
) * num_spares
;
2842 mu
->next
= *updates
;
2843 u
= (struct imsm_update_activate_spare
*) mu
->buf
;
2845 for (di
= rv
; di
; di
= di
->next
) {
2846 u
->type
= update_activate_spare
;
2847 u
->dl
= (struct dl
*) di
->devs
;
2849 u
->slot
= di
->disk
.raid_disk
;
2860 static int disks_overlap(struct imsm_dev
*d1
, struct imsm_dev
*d2
)
2862 struct imsm_map
*m1
= get_imsm_map(d1
, 0);
2863 struct imsm_map
*m2
= get_imsm_map(d2
, 0);
2868 for (i
= 0; i
< m1
->num_members
; i
++) {
2869 idx
= get_imsm_disk_idx(d1
, i
);
2870 for (j
= 0; j
< m2
->num_members
; j
++)
2871 if (idx
== get_imsm_disk_idx(d2
, j
))
2878 static void imsm_delete(struct intel_super
*super
, struct dl
**dlp
, int index
);
2880 static void imsm_process_update(struct supertype
*st
,
2881 struct metadata_update
*update
)
2884 * crack open the metadata_update envelope to find the update record
2885 * update can be one of:
2886 * update_activate_spare - a spare device has replaced a failed
2887 * device in an array, update the disk_ord_tbl. If this disk is
2888 * present in all member arrays then also clear the SPARE_DISK
2891 struct intel_super
*super
= st
->sb
;
2892 struct imsm_super
*mpb
;
2893 enum imsm_update_type type
= *(enum imsm_update_type
*) update
->buf
;
2895 /* update requires a larger buf but the allocation failed */
2896 if (super
->next_len
&& !super
->next_buf
) {
2897 super
->next_len
= 0;
2901 if (super
->next_buf
) {
2902 memcpy(super
->next_buf
, super
->buf
, super
->len
);
2904 super
->len
= super
->next_len
;
2905 super
->buf
= super
->next_buf
;
2907 super
->next_len
= 0;
2908 super
->next_buf
= NULL
;
2911 mpb
= super
->anchor
;
2914 case update_activate_spare
: {
2915 struct imsm_update_activate_spare
*u
= (void *) update
->buf
;
2916 struct imsm_dev
*dev
= get_imsm_dev(super
, u
->array
);
2917 struct imsm_map
*map
= get_imsm_map(dev
, 0);
2918 struct imsm_map
*migr_map
;
2919 struct active_array
*a
;
2920 struct imsm_disk
*disk
;
2926 int victim
= get_imsm_disk_idx(dev
, u
->slot
);
2929 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
2934 fprintf(stderr
, "error: imsm_activate_spare passed "
2935 "an unknown disk (index: %d)\n",
2940 super
->updates_pending
++;
2942 /* count failures (excluding rebuilds and the victim)
2943 * to determine map[0] state
2946 for (i
= 0; i
< map
->num_members
; i
++) {
2949 disk
= get_imsm_disk(super
, get_imsm_disk_idx(dev
, i
));
2951 __le32_to_cpu(disk
->status
) & FAILED_DISK
)
2955 /* adding a pristine spare, assign a new index */
2956 if (dl
->index
< 0) {
2957 dl
->index
= super
->anchor
->num_disks
;
2958 super
->anchor
->num_disks
++;
2961 status
= __le32_to_cpu(disk
->status
);
2962 status
|= CONFIGURED_DISK
;
2963 status
&= ~SPARE_DISK
;
2964 disk
->status
= __cpu_to_le32(status
);
2967 to_state
= imsm_check_degraded(super
, dev
, failed
);
2968 map
->map_state
= IMSM_T_STATE_DEGRADED
;
2969 migrate(dev
, to_state
, 1);
2970 migr_map
= get_imsm_map(dev
, 1);
2971 set_imsm_ord_tbl_ent(map
, u
->slot
, dl
->index
);
2972 set_imsm_ord_tbl_ent(migr_map
, u
->slot
, dl
->index
| IMSM_ORD_REBUILD
);
2974 /* count arrays using the victim in the metadata */
2976 for (a
= st
->arrays
; a
; a
= a
->next
) {
2977 dev
= get_imsm_dev(super
, a
->info
.container_member
);
2978 for (i
= 0; i
< map
->num_members
; i
++)
2979 if (victim
== get_imsm_disk_idx(dev
, i
))
2983 /* delete the victim if it is no longer being
2989 for (dlp
= &super
->disks
; *dlp
; dlp
= &(*dlp
)->next
)
2990 if ((*dlp
)->index
== victim
)
2992 /* We know that 'manager' isn't touching anything,
2995 imsm_delete(super
, dlp
, victim
);
2999 case update_create_array
: {
3000 /* someone wants to create a new array, we need to be aware of
3001 * a few races/collisions:
3002 * 1/ 'Create' called by two separate instances of mdadm
3003 * 2/ 'Create' versus 'activate_spare': mdadm has chosen
3004 * devices that have since been assimilated via
3006 * In the event this update can not be carried out mdadm will
3007 * (FIX ME) notice that its update did not take hold.
3009 struct imsm_update_create_array
*u
= (void *) update
->buf
;
3010 struct imsm_dev
*dev
;
3011 struct imsm_map
*map
, *new_map
;
3012 unsigned long long start
, end
;
3013 unsigned long long new_start
, new_end
;
3017 /* handle racing creates: first come first serve */
3018 if (u
->dev_idx
< mpb
->num_raid_devs
) {
3019 dprintf("%s: subarray %d already defined\n",
3020 __func__
, u
->dev_idx
);
3024 /* check update is next in sequence */
3025 if (u
->dev_idx
!= mpb
->num_raid_devs
) {
3026 dprintf("%s: can not create array %d expected index %d\n",
3027 __func__
, u
->dev_idx
, mpb
->num_raid_devs
);
3031 new_map
= get_imsm_map(&u
->dev
, 0);
3032 new_start
= __le32_to_cpu(new_map
->pba_of_lba0
);
3033 new_end
= new_start
+ __le32_to_cpu(new_map
->blocks_per_member
);
3035 /* handle activate_spare versus create race:
3036 * check to make sure that overlapping arrays do not include
3039 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
3040 dev
= get_imsm_dev(super
, i
);
3041 map
= get_imsm_map(dev
, 0);
3042 start
= __le32_to_cpu(map
->pba_of_lba0
);
3043 end
= start
+ __le32_to_cpu(map
->blocks_per_member
);
3044 if ((new_start
>= start
&& new_start
<= end
) ||
3045 (start
>= new_start
&& start
<= new_end
))
3047 if (overlap
&& disks_overlap(dev
, &u
->dev
)) {
3048 dprintf("%s: arrays overlap\n", __func__
);
3052 /* check num_members sanity */
3053 if (new_map
->num_members
> mpb
->num_disks
) {
3054 dprintf("%s: num_disks out of range\n", __func__
);
3058 /* check that prepare update was successful */
3059 if (!update
->space
) {
3060 dprintf("%s: prepare update failed\n", __func__
);
3064 super
->updates_pending
++;
3065 dev
= update
->space
;
3066 map
= get_imsm_map(dev
, 0);
3067 update
->space
= NULL
;
3068 imsm_copy_dev(dev
, &u
->dev
);
3069 map
= get_imsm_map(dev
, 0);
3070 super
->dev_tbl
[u
->dev_idx
] = dev
;
3071 mpb
->num_raid_devs
++;
3074 for (i
= 0; i
< map
->num_members
; i
++) {
3075 struct imsm_disk
*disk
;
3078 disk
= get_imsm_disk(super
, get_imsm_disk_idx(dev
, i
));
3079 status
= __le32_to_cpu(disk
->status
);
3080 status
|= CONFIGURED_DISK
;
3081 status
&= ~SPARE_DISK
;
3082 disk
->status
= __cpu_to_le32(status
);
3086 case update_add_disk
:
3088 /* we may be able to repair some arrays if disks are
3091 struct active_array
*a
;
3092 for (a
= st
->arrays
; a
; a
= a
->next
)
3093 a
->check_degraded
= 1;
3095 /* add some spares to the metadata */
3096 while (super
->add
) {
3100 super
->add
= al
->next
;
3101 al
->next
= super
->disks
;
3103 dprintf("%s: added %x:%x\n",
3104 __func__
, al
->major
, al
->minor
);
3111 static void imsm_prepare_update(struct supertype
*st
,
3112 struct metadata_update
*update
)
3115 * Allocate space to hold new disk entries, raid-device entries or a new
3116 * mpb if necessary. The manager synchronously waits for updates to
3117 * complete in the monitor, so new mpb buffers allocated here can be
3118 * integrated by the monitor thread without worrying about live pointers
3119 * in the manager thread.
3121 enum imsm_update_type type
= *(enum imsm_update_type
*) update
->buf
;
3122 struct intel_super
*super
= st
->sb
;
3123 struct imsm_super
*mpb
= super
->anchor
;
3128 case update_create_array
: {
3129 struct imsm_update_create_array
*u
= (void *) update
->buf
;
3131 len
= sizeof_imsm_dev(&u
->dev
, 1);
3132 update
->space
= malloc(len
);
3139 /* check if we need a larger metadata buffer */
3140 if (super
->next_buf
)
3141 buf_len
= super
->next_len
;
3143 buf_len
= super
->len
;
3145 if (__le32_to_cpu(mpb
->mpb_size
) + len
> buf_len
) {
3146 /* ok we need a larger buf than what is currently allocated
3147 * if this allocation fails process_update will notice that
3148 * ->next_len is set and ->next_buf is NULL
3150 buf_len
= ROUND_UP(__le32_to_cpu(mpb
->mpb_size
) + len
, 512);
3151 if (super
->next_buf
)
3152 free(super
->next_buf
);
3154 super
->next_len
= buf_len
;
3155 if (posix_memalign(&super
->next_buf
, buf_len
, 512) != 0)
3156 super
->next_buf
= NULL
;
3160 /* must be called while manager is quiesced */
3161 static void imsm_delete(struct intel_super
*super
, struct dl
**dlp
, int index
)
3163 struct imsm_super
*mpb
= super
->anchor
;
3165 struct imsm_dev
*dev
;
3166 struct imsm_map
*map
;
3167 int i
, j
, num_members
;
3170 dprintf("%s: deleting device[%d] from imsm_super\n",
3173 /* shift all indexes down one */
3174 for (iter
= super
->disks
; iter
; iter
= iter
->next
)
3175 if (iter
->index
> index
)
3178 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
3179 dev
= get_imsm_dev(super
, i
);
3180 map
= get_imsm_map(dev
, 0);
3181 num_members
= map
->num_members
;
3182 for (j
= 0; j
< num_members
; j
++) {
3183 /* update ord entries being careful not to propagate
3184 * ord-flags to the first map
3186 ord
= get_imsm_ord_tbl_ent(dev
, j
);
3188 if (ord_to_idx(ord
) <= index
)
3191 map
= get_imsm_map(dev
, 0);
3192 set_imsm_ord_tbl_ent(map
, j
, ord_to_idx(ord
- 1));
3193 map
= get_imsm_map(dev
, 1);
3195 set_imsm_ord_tbl_ent(map
, j
, ord
- 1);
3200 super
->updates_pending
++;
3202 struct dl
*dl
= *dlp
;
3204 *dlp
= (*dlp
)->next
;
3205 __free_imsm_disk(dl
);
3208 #endif /* MDASSEMBLE */
3210 struct superswitch super_imsm
= {
3212 .examine_super
= examine_super_imsm
,
3213 .brief_examine_super
= brief_examine_super_imsm
,
3214 .detail_super
= detail_super_imsm
,
3215 .brief_detail_super
= brief_detail_super_imsm
,
3216 .write_init_super
= write_init_super_imsm
,
3217 .validate_geometry
= validate_geometry_imsm
,
3218 .add_to_super
= add_to_super_imsm
,
3220 .match_home
= match_home_imsm
,
3221 .uuid_from_super
= uuid_from_super_imsm
,
3222 .getinfo_super
= getinfo_super_imsm
,
3223 .update_super
= update_super_imsm
,
3225 .avail_size
= avail_size_imsm
,
3227 .compare_super
= compare_super_imsm
,
3229 .load_super
= load_super_imsm
,
3230 .init_super
= init_super_imsm
,
3231 .store_super
= store_zero_imsm
,
3232 .free_super
= free_super_imsm
,
3233 .match_metadata_desc
= match_metadata_desc_imsm
,
3234 .container_content
= container_content_imsm
,
3240 .open_new
= imsm_open_new
,
3241 .load_super
= load_super_imsm
,
3242 .set_array_state
= imsm_set_array_state
,
3243 .set_disk
= imsm_set_disk
,
3244 .sync_metadata
= imsm_sync_metadata
,
3245 .activate_spare
= imsm_activate_spare
,
3246 .process_update
= imsm_process_update
,
3247 .prepare_update
= imsm_prepare_update
,
3248 #endif /* MDASSEMBLE */