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
;
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
;
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 e
->start
= __le32_to_cpu(dl
->disk
.total_blocks
) -
444 (MPB_SECTOR_CNT
+ IMSM_RESERVED_SECTORS
);
449 static void print_imsm_dev(struct imsm_dev
*dev
, int index
)
453 struct imsm_map
*map
= get_imsm_map(dev
, 0);
457 printf("[%s]:\n", dev
->volume
);
458 printf(" RAID Level : %d\n", get_imsm_raid_level(map
));
459 printf(" Members : %d\n", map
->num_members
);
460 for (slot
= 0; slot
< map
->num_members
; slot
++)
461 if (index
== get_imsm_disk_idx(dev
, slot
))
463 if (slot
< map
->num_members
) {
464 ord
= get_imsm_ord_tbl_ent(dev
, slot
);
465 printf(" This Slot : %d%s\n", slot
,
466 ord
& IMSM_ORD_REBUILD
? " (out-of-sync)" : "");
468 printf(" This Slot : ?\n");
469 sz
= __le32_to_cpu(dev
->size_high
);
471 sz
+= __le32_to_cpu(dev
->size_low
);
472 printf(" Array Size : %llu%s\n", (unsigned long long)sz
,
473 human_size(sz
* 512));
474 sz
= __le32_to_cpu(map
->blocks_per_member
);
475 printf(" Per Dev Size : %llu%s\n", (unsigned long long)sz
,
476 human_size(sz
* 512));
477 printf(" Sector Offset : %u\n",
478 __le32_to_cpu(map
->pba_of_lba0
));
479 printf(" Num Stripes : %u\n",
480 __le32_to_cpu(map
->num_data_stripes
));
481 printf(" Chunk Size : %u KiB\n",
482 __le16_to_cpu(map
->blocks_per_strip
) / 2);
483 printf(" Reserved : %d\n", __le32_to_cpu(dev
->reserved_blocks
));
484 printf(" Migrate State : %s", dev
->vol
.migr_state
? "migrating" : "idle");
485 if (dev
->vol
.migr_state
)
486 printf(": %s", dev
->vol
.migr_type
? "rebuilding" : "initializing");
488 printf(" Map State : %s", map_state_str
[map
->map_state
]);
489 if (dev
->vol
.migr_state
) {
490 struct imsm_map
*map
= get_imsm_map(dev
, 1);
491 printf(" <-- %s", map_state_str
[map
->map_state
]);
494 printf(" Dirty State : %s\n", dev
->vol
.dirty
? "dirty" : "clean");
497 static void print_imsm_disk(struct imsm_super
*mpb
, int index
)
499 struct imsm_disk
*disk
= __get_imsm_disk(mpb
, index
);
500 char str
[MAX_RAID_SERIAL_LEN
+ 1];
508 snprintf(str
, MAX_RAID_SERIAL_LEN
+ 1, "%s", disk
->serial
);
509 printf(" Disk%02d Serial : %s\n", index
, str
);
510 s
= __le32_to_cpu(disk
->status
);
511 printf(" State :%s%s%s%s\n", s
&SPARE_DISK
? " spare" : "",
512 s
&CONFIGURED_DISK
? " active" : "",
513 s
&FAILED_DISK
? " failed" : "",
514 s
&USABLE_DISK
? " usable" : "");
515 printf(" Id : %08x\n", __le32_to_cpu(disk
->scsi_id
));
516 sz
= __le32_to_cpu(disk
->total_blocks
) -
517 (MPB_SECTOR_CNT
+ IMSM_RESERVED_SECTORS
* mpb
->num_raid_devs
);
518 printf(" Usable Size : %llu%s\n", (unsigned long long)sz
,
519 human_size(sz
* 512));
522 static void examine_super_imsm(struct supertype
*st
, char *homehost
)
524 struct intel_super
*super
= st
->sb
;
525 struct imsm_super
*mpb
= super
->anchor
;
526 char str
[MAX_SIGNATURE_LENGTH
];
530 snprintf(str
, MPB_SIG_LEN
, "%s", mpb
->sig
);
531 printf(" Magic : %s\n", str
);
532 snprintf(str
, strlen(MPB_VERSION_RAID0
), "%s", get_imsm_version(mpb
));
533 printf(" Version : %s\n", get_imsm_version(mpb
));
534 printf(" Family : %08x\n", __le32_to_cpu(mpb
->family_num
));
535 printf(" Generation : %08x\n", __le32_to_cpu(mpb
->generation_num
));
536 sum
= __le32_to_cpu(mpb
->check_sum
);
537 printf(" Checksum : %08x %s\n", sum
,
538 __gen_imsm_checksum(mpb
) == sum
? "correct" : "incorrect");
539 printf(" MPB Sectors : %d\n", mpb_sectors(mpb
));
540 printf(" Disks : %d\n", mpb
->num_disks
);
541 printf(" RAID Devices : %d\n", mpb
->num_raid_devs
);
542 print_imsm_disk(mpb
, super
->disks
->index
);
543 if (super
->bbm_log
) {
544 struct bbm_log
*log
= super
->bbm_log
;
547 printf("Bad Block Management Log:\n");
548 printf(" Log Size : %d\n", __le32_to_cpu(mpb
->bbm_log_size
));
549 printf(" Signature : %x\n", __le32_to_cpu(log
->signature
));
550 printf(" Entry Count : %d\n", __le32_to_cpu(log
->entry_count
));
551 printf(" Spare Blocks : %d\n", __le32_to_cpu(log
->reserved_spare_block_count
));
552 printf(" First Spare : %llx\n", __le64_to_cpu(log
->first_spare_lba
));
554 for (i
= 0; i
< mpb
->num_raid_devs
; i
++)
555 print_imsm_dev(__get_imsm_dev(mpb
, i
), super
->disks
->index
);
556 for (i
= 0; i
< mpb
->num_disks
; i
++) {
557 if (i
== super
->disks
->index
)
559 print_imsm_disk(mpb
, i
);
563 static void getinfo_super_imsm(struct supertype
*st
, struct mdinfo
*info
);
565 static void brief_examine_super_imsm(struct supertype
*st
)
567 /* We just write a generic DDF ARRAY entry
572 getinfo_super_imsm(st
, &info
);
573 fname_from_uuid(st
, &info
, nbuf
,'-');
574 printf("ARRAY /dev/imsm metadata=imsm UUID=%s\n", nbuf
+ 5);
577 static void detail_super_imsm(struct supertype
*st
, char *homehost
)
579 printf("%s\n", __FUNCTION__
);
582 static void brief_detail_super_imsm(struct supertype
*st
)
586 getinfo_super_imsm(st
, &info
);
587 fname_from_uuid(st
, &info
, nbuf
,'-');
588 printf(" UUID=%s", nbuf
+ 5);
592 static int match_home_imsm(struct supertype
*st
, char *homehost
)
594 printf("%s\n", __FUNCTION__
);
599 static void uuid_from_super_imsm(struct supertype
*st
, int uuid
[4])
601 /* The uuid returned here is used for:
602 * uuid to put into bitmap file (Create, Grow)
603 * uuid for backup header when saving critical section (Grow)
604 * comparing uuids when re-adding a device into an array
605 * In these cases the uuid required is that of the data-array,
606 * not the device-set.
607 * uuid to recognise same set when adding a missing device back
608 * to an array. This is a uuid for the device-set.
610 * For each of these we can make do with a truncated
611 * or hashed uuid rather than the original, as long as
613 * In each case the uuid required is that of the data-array,
614 * not the device-set.
616 /* imsm does not track uuid's so we synthesis one using sha1 on
617 * - The signature (Which is constant for all imsm array, but no matter)
618 * - the family_num of the container
619 * - the index number of the volume
620 * - the 'serial' number of the volume.
621 * Hopefully these are all constant.
623 struct intel_super
*super
= st
->sb
;
627 struct imsm_dev
*dev
= NULL
;
630 sha1_process_bytes(super
->anchor
->sig
, MAX_SIGNATURE_LENGTH
, &ctx
);
631 sha1_process_bytes(&super
->anchor
->family_num
, sizeof(__u32
), &ctx
);
632 if (super
->current_vol
>= 0)
633 dev
= get_imsm_dev(super
, super
->current_vol
);
635 __u32 vol
= super
->current_vol
;
636 sha1_process_bytes(&vol
, sizeof(vol
), &ctx
);
637 sha1_process_bytes(dev
->volume
, MAX_RAID_SERIAL_LEN
, &ctx
);
639 sha1_finish_ctx(&ctx
, buf
);
640 memcpy(uuid
, buf
, 4*4);
645 get_imsm_numerical_version(struct imsm_super
*mpb
, int *m
, int *p
)
647 __u8
*v
= get_imsm_version(mpb
);
648 __u8
*end
= mpb
->sig
+ MAX_SIGNATURE_LENGTH
;
649 char major
[] = { 0, 0, 0 };
650 char minor
[] = { 0 ,0, 0 };
651 char patch
[] = { 0, 0, 0 };
652 char *ver_parse
[] = { major
, minor
, patch
};
656 while (*v
!= '\0' && v
< end
) {
657 if (*v
!= '.' && j
< 2)
658 ver_parse
[i
][j
++] = *v
;
666 *m
= strtol(minor
, NULL
, 0);
667 *p
= strtol(patch
, NULL
, 0);
671 static int imsm_level_to_layout(int level
)
679 return ALGORITHM_LEFT_ASYMMETRIC
;
681 return 0x102; //FIXME is this correct?
686 static void getinfo_super_imsm_volume(struct supertype
*st
, struct mdinfo
*info
)
688 struct intel_super
*super
= st
->sb
;
689 struct imsm_dev
*dev
= get_imsm_dev(super
, super
->current_vol
);
690 struct imsm_map
*map
= get_imsm_map(dev
, 0);
692 info
->container_member
= super
->current_vol
;
693 info
->array
.raid_disks
= map
->num_members
;
694 info
->array
.level
= get_imsm_raid_level(map
);
695 info
->array
.layout
= imsm_level_to_layout(info
->array
.level
);
696 info
->array
.md_minor
= -1;
697 info
->array
.ctime
= 0;
698 info
->array
.utime
= 0;
699 info
->array
.chunk_size
= __le16_to_cpu(map
->blocks_per_strip
) << 9;
700 info
->array
.state
= !dev
->vol
.dirty
;
702 info
->disk
.major
= 0;
703 info
->disk
.minor
= 0;
705 info
->data_offset
= __le32_to_cpu(map
->pba_of_lba0
);
706 info
->component_size
= __le32_to_cpu(map
->blocks_per_member
);
707 memset(info
->uuid
, 0, sizeof(info
->uuid
));
709 if (map
->map_state
== IMSM_T_STATE_UNINITIALIZED
||
710 dev
->vol
.dirty
|| dev
->vol
.migr_state
)
711 info
->resync_start
= 0;
713 info
->resync_start
= ~0ULL;
715 strncpy(info
->name
, (char *) dev
->volume
, MAX_RAID_SERIAL_LEN
);
716 info
->name
[MAX_RAID_SERIAL_LEN
] = 0;
718 info
->array
.major_version
= -1;
719 info
->array
.minor_version
= -2;
720 sprintf(info
->text_version
, "/%s/%d",
721 devnum2devname(st
->container_dev
),
722 info
->container_member
);
723 info
->safe_mode_delay
= 4000; /* 4 secs like the Matrix driver */
724 uuid_from_super_imsm(st
, info
->uuid
);
728 static void getinfo_super_imsm(struct supertype
*st
, struct mdinfo
*info
)
730 struct intel_super
*super
= st
->sb
;
731 struct imsm_disk
*disk
;
734 if (super
->current_vol
>= 0) {
735 getinfo_super_imsm_volume(st
, info
);
739 /* Set raid_disks to zero so that Assemble will always pull in valid
742 info
->array
.raid_disks
= 0;
743 info
->array
.level
= LEVEL_CONTAINER
;
744 info
->array
.layout
= 0;
745 info
->array
.md_minor
= -1;
746 info
->array
.ctime
= 0; /* N/A for imsm */
747 info
->array
.utime
= 0;
748 info
->array
.chunk_size
= 0;
750 info
->disk
.major
= 0;
751 info
->disk
.minor
= 0;
752 info
->disk
.raid_disk
= -1;
753 info
->reshape_active
= 0;
754 info
->array
.major_version
= -1;
755 info
->array
.minor_version
= -2;
756 strcpy(info
->text_version
, "imsm");
757 info
->safe_mode_delay
= 0;
758 info
->disk
.number
= -1;
759 info
->disk
.state
= 0;
763 disk
= &super
->disks
->disk
;
764 info
->disk
.number
= super
->disks
->index
;
765 info
->disk
.raid_disk
= super
->disks
->index
;
766 info
->data_offset
= __le32_to_cpu(disk
->total_blocks
) -
767 (MPB_SECTOR_CNT
+ IMSM_RESERVED_SECTORS
);
768 info
->component_size
= MPB_SECTOR_CNT
+ IMSM_RESERVED_SECTORS
;
769 s
= __le32_to_cpu(disk
->status
);
770 info
->disk
.state
= s
& CONFIGURED_DISK
? (1 << MD_DISK_ACTIVE
) : 0;
771 info
->disk
.state
|= s
& FAILED_DISK
? (1 << MD_DISK_FAULTY
) : 0;
772 info
->disk
.state
|= s
& USABLE_DISK
? (1 << MD_DISK_SYNC
) : 0;
774 uuid_from_super_imsm(st
, info
->uuid
);
777 static int update_super_imsm(struct supertype
*st
, struct mdinfo
*info
,
778 char *update
, char *devname
, int verbose
,
779 int uuid_set
, char *homehost
)
783 /* For 'assemble' and 'force' we need to return non-zero if any
784 * change was made. For others, the return value is ignored.
785 * Update options are:
786 * force-one : This device looks a bit old but needs to be included,
787 * update age info appropriately.
788 * assemble: clear any 'faulty' flag to allow this device to
790 * force-array: Array is degraded but being forced, mark it clean
791 * if that will be needed to assemble it.
793 * newdev: not used ????
794 * grow: Array has gained a new device - this is currently for
796 * resync: mark as dirty so a resync will happen.
797 * name: update the name - preserving the homehost
799 * Following are not relevant for this imsm:
800 * sparc2.2 : update from old dodgey metadata
801 * super-minor: change the preferred_minor number
802 * summaries: update redundant counters.
803 * uuid: Change the uuid of the array to match watch is given
804 * homehost: update the recorded homehost
805 * _reshape_progress: record new reshape_progress position.
808 //struct intel_super *super = st->sb;
809 //struct imsm_super *mpb = super->mpb;
811 if (strcmp(update
, "grow") == 0) {
813 if (strcmp(update
, "resync") == 0) {
814 /* dev->vol.dirty = 1; */
817 /* IMSM has no concept of UUID or homehost */
822 static size_t disks_to_mpb_size(int disks
)
826 size
= sizeof(struct imsm_super
);
827 size
+= (disks
- 1) * sizeof(struct imsm_disk
);
828 size
+= 2 * sizeof(struct imsm_dev
);
829 /* up to 2 maps per raid device (-2 for imsm_maps in imsm_dev */
830 size
+= (4 - 2) * sizeof(struct imsm_map
);
831 /* 4 possible disk_ord_tbl's */
832 size
+= 4 * (disks
- 1) * sizeof(__u32
);
837 static __u64
avail_size_imsm(struct supertype
*st
, __u64 devsize
)
839 if (devsize
< (MPB_SECTOR_CNT
+ IMSM_RESERVED_SECTORS
))
842 return devsize
- (MPB_SECTOR_CNT
+ IMSM_RESERVED_SECTORS
);
845 static int compare_super_imsm(struct supertype
*st
, struct supertype
*tst
)
849 * 0 same, or first was empty, and second was copied
850 * 1 second had wrong number
854 struct intel_super
*first
= st
->sb
;
855 struct intel_super
*sec
= tst
->sb
;
863 if (memcmp(first
->anchor
->sig
, sec
->anchor
->sig
, MAX_SIGNATURE_LENGTH
) != 0)
866 /* if an anchor does not have num_raid_devs set then it is a free
869 if (first
->anchor
->num_raid_devs
> 0 &&
870 sec
->anchor
->num_raid_devs
> 0) {
871 if (first
->anchor
->family_num
!= sec
->anchor
->family_num
)
875 /* if 'first' is a spare promote it to a populated mpb with sec's
878 if (first
->anchor
->num_raid_devs
== 0 &&
879 sec
->anchor
->num_raid_devs
> 0) {
880 first
->anchor
->num_raid_devs
= sec
->anchor
->num_raid_devs
;
881 first
->anchor
->family_num
= sec
->anchor
->family_num
;
887 static void fd2devname(int fd
, char *name
)
896 if (fstat(fd
, &st
) != 0)
898 sprintf(path
, "/sys/dev/block/%d:%d",
899 major(st
.st_rdev
), minor(st
.st_rdev
));
901 rv
= readlink(path
, dname
, sizeof(dname
));
906 nm
= strrchr(dname
, '/');
908 snprintf(name
, MAX_RAID_SERIAL_LEN
, "/dev/%s", nm
);
912 extern int scsi_get_serial(int fd
, void *buf
, size_t buf_len
);
914 static int imsm_read_serial(int fd
, char *devname
,
915 __u8 serial
[MAX_RAID_SERIAL_LEN
])
917 unsigned char scsi_serial
[255];
923 memset(scsi_serial
, 0, sizeof(scsi_serial
));
925 rv
= scsi_get_serial(fd
, scsi_serial
, sizeof(scsi_serial
));
927 if (rv
&& imsm_env_devname_as_serial()) {
928 memset(serial
, 0, MAX_RAID_SERIAL_LEN
);
929 fd2devname(fd
, (char *) serial
);
936 Name
": Failed to retrieve serial for %s\n",
941 /* trim leading whitespace */
942 rsp_len
= scsi_serial
[3];
943 rsp_buf
= (char *) &scsi_serial
[4];
948 /* truncate len to the end of rsp_buf if necessary */
949 if (c
+ MAX_RAID_SERIAL_LEN
> rsp_buf
+ rsp_len
)
950 len
= rsp_len
- (c
- rsp_buf
);
952 len
= MAX_RAID_SERIAL_LEN
;
954 /* initialize the buffer and copy rsp_buf characters */
955 memset(serial
, 0, MAX_RAID_SERIAL_LEN
);
956 memcpy(serial
, c
, len
);
958 /* trim trailing whitespace starting with the last character copied */
959 c
= (char *) &serial
[len
- 1];
960 while (isspace(*c
) || *c
== '\0')
966 static int serialcmp(__u8
*s1
, __u8
*s2
)
968 return strncmp((char *) s1
, (char *) s2
, MAX_RAID_SERIAL_LEN
);
971 static void serialcpy(__u8
*dest
, __u8
*src
)
973 strncpy((char *) dest
, (char *) src
, MAX_RAID_SERIAL_LEN
);
977 load_imsm_disk(int fd
, struct intel_super
*super
, char *devname
, int keep_fd
)
984 __u8 serial
[MAX_RAID_SERIAL_LEN
];
986 rv
= imsm_read_serial(fd
, devname
, serial
);
991 /* check if this is a disk we have seen before. it may be a spare in
992 * super->disks while the current anchor believes it is a raid member,
993 * check if we need to update dl->index
995 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
996 if (serialcmp(dl
->serial
, serial
) == 0)
1000 dl
= malloc(sizeof(*dl
));
1007 Name
": failed to allocate disk buffer for %s\n",
1014 dl
->major
= major(stb
.st_rdev
);
1015 dl
->minor
= minor(stb
.st_rdev
);
1016 dl
->next
= super
->disks
;
1017 dl
->fd
= keep_fd
? fd
: -1;
1018 dl
->devname
= devname
? strdup(devname
) : NULL
;
1019 serialcpy(dl
->serial
, serial
);
1021 } else if (keep_fd
) {
1026 /* look up this disk's index in the current anchor */
1027 for (i
= 0; i
< super
->anchor
->num_disks
; i
++) {
1028 struct imsm_disk
*disk_iter
;
1030 disk_iter
= __get_imsm_disk(super
->anchor
, i
);
1032 if (serialcmp(disk_iter
->serial
, dl
->serial
) == 0) {
1035 dl
->disk
= *disk_iter
;
1036 status
= __le32_to_cpu(dl
->disk
.status
);
1037 /* only set index on disks that are a member of a
1038 * populated contianer, i.e. one with raid_devs
1040 if (status
& FAILED_DISK
)
1042 else if (status
& SPARE_DISK
)
1057 static void imsm_copy_dev(struct imsm_dev
*dest
, struct imsm_dev
*src
)
1059 memcpy(dest
, src
, sizeof_imsm_dev(src
, 0));
1063 /* When migrating map0 contains the 'destination' state while map1
1064 * contains the current state. When not migrating map0 contains the
1065 * current state. This routine assumes that map[0].map_state is set to
1066 * the current array state before being called.
1068 * Migration is indicated by one of the following states
1069 * 1/ Idle (migr_state=0 map0state=normal||unitialized||degraded||failed)
1070 * 2/ Initialize (migr_state=1 migr_type=0 map0state=normal
1071 * map1state=unitialized)
1072 * 3/ Verify (Resync) (migr_state=1 migr_type=1 map0state=normal
1074 * 4/ Rebuild (migr_state=1 migr_type=1 map0state=normal
1075 * map1state=degraded)
1077 static void migrate(struct imsm_dev
*dev
, __u8 to_state
, int rebuild_resync
)
1079 struct imsm_map
*dest
;
1080 struct imsm_map
*src
= get_imsm_map(dev
, 0);
1082 dev
->vol
.migr_state
= 1;
1083 dev
->vol
.migr_type
= rebuild_resync
;
1084 dest
= get_imsm_map(dev
, 1);
1086 memcpy(dest
, src
, sizeof_imsm_map(src
));
1087 src
->map_state
= to_state
;
1091 static int parse_raid_devices(struct intel_super
*super
)
1094 struct imsm_dev
*dev_new
;
1095 size_t len
, len_migr
;
1096 size_t space_needed
= 0;
1097 struct imsm_super
*mpb
= super
->anchor
;
1099 for (i
= 0; i
< super
->anchor
->num_raid_devs
; i
++) {
1100 struct imsm_dev
*dev_iter
= __get_imsm_dev(super
->anchor
, i
);
1102 len
= sizeof_imsm_dev(dev_iter
, 0);
1103 len_migr
= sizeof_imsm_dev(dev_iter
, 1);
1105 space_needed
+= len_migr
- len
;
1107 dev_new
= malloc(len_migr
);
1110 imsm_copy_dev(dev_new
, dev_iter
);
1111 super
->dev_tbl
[i
] = dev_new
;
1114 /* ensure that super->buf is large enough when all raid devices
1117 if (__le32_to_cpu(mpb
->mpb_size
) + space_needed
> super
->len
) {
1120 len
= ROUND_UP(__le32_to_cpu(mpb
->mpb_size
) + space_needed
, 512);
1121 if (posix_memalign(&buf
, 512, len
) != 0)
1124 memcpy(buf
, super
->buf
, len
);
1133 /* retrieve a pointer to the bbm log which starts after all raid devices */
1134 struct bbm_log
*__get_imsm_bbm_log(struct imsm_super
*mpb
)
1138 if (__le32_to_cpu(mpb
->bbm_log_size
)) {
1140 ptr
+= mpb
->mpb_size
- __le32_to_cpu(mpb
->bbm_log_size
);
1146 static void __free_imsm(struct intel_super
*super
, int free_disks
);
1148 /* load_imsm_mpb - read matrix metadata
1149 * allocates super->mpb to be freed by free_super
1151 static int load_imsm_mpb(int fd
, struct intel_super
*super
, char *devname
)
1153 unsigned long long dsize
;
1154 unsigned long long sectors
;
1156 struct imsm_super
*anchor
;
1160 get_dev_size(fd
, NULL
, &dsize
);
1162 if (lseek64(fd
, dsize
- (512 * 2), SEEK_SET
) < 0) {
1165 Name
": Cannot seek to anchor block on %s: %s\n",
1166 devname
, strerror(errno
));
1170 if (posix_memalign((void**)&anchor
, 512, 512) != 0) {
1173 Name
": Failed to allocate imsm anchor buffer"
1174 " on %s\n", devname
);
1177 if (read(fd
, anchor
, 512) != 512) {
1180 Name
": Cannot read anchor block on %s: %s\n",
1181 devname
, strerror(errno
));
1186 if (strncmp((char *) anchor
->sig
, MPB_SIGNATURE
, MPB_SIG_LEN
) != 0) {
1189 Name
": no IMSM anchor on %s\n", devname
);
1194 __free_imsm(super
, 0);
1195 super
->len
= ROUND_UP(anchor
->mpb_size
, 512);
1196 if (posix_memalign(&super
->buf
, 512, super
->len
) != 0) {
1199 Name
": unable to allocate %zu byte mpb buffer\n",
1204 memcpy(super
->buf
, anchor
, 512);
1206 sectors
= mpb_sectors(anchor
) - 1;
1209 rc
= load_imsm_disk(fd
, super
, devname
, 0);
1211 rc
= parse_raid_devices(super
);
1215 /* read the extended mpb */
1216 if (lseek64(fd
, dsize
- (512 * (2 + sectors
)), SEEK_SET
) < 0) {
1219 Name
": Cannot seek to extended mpb on %s: %s\n",
1220 devname
, strerror(errno
));
1224 if (read(fd
, super
->buf
+ 512, super
->len
- 512) != super
->len
- 512) {
1227 Name
": Cannot read extended mpb on %s: %s\n",
1228 devname
, strerror(errno
));
1232 check_sum
= __gen_imsm_checksum(super
->anchor
);
1233 if (check_sum
!= __le32_to_cpu(super
->anchor
->check_sum
)) {
1236 Name
": IMSM checksum %x != %x on %s\n",
1237 check_sum
, __le32_to_cpu(super
->anchor
->check_sum
),
1242 /* FIXME the BBM log is disk specific so we cannot use this global
1243 * buffer for all disks. Ok for now since we only look at the global
1244 * bbm_log_size parameter to gate assembly
1246 super
->bbm_log
= __get_imsm_bbm_log(super
->anchor
);
1248 rc
= load_imsm_disk(fd
, super
, devname
, 0);
1250 rc
= parse_raid_devices(super
);
1255 static void __free_imsm_disk(struct dl
*d
)
1264 static void free_imsm_disks(struct intel_super
*super
)
1266 while (super
->disks
) {
1267 struct dl
*d
= super
->disks
;
1269 super
->disks
= d
->next
;
1270 __free_imsm_disk(d
);
1274 /* free all the pieces hanging off of a super pointer */
1275 static void __free_imsm(struct intel_super
*super
, int free_disks
)
1284 free_imsm_disks(super
);
1285 for (i
= 0; i
< IMSM_MAX_RAID_DEVS
; i
++)
1286 if (super
->dev_tbl
[i
]) {
1287 free(super
->dev_tbl
[i
]);
1288 super
->dev_tbl
[i
] = NULL
;
1292 static void free_imsm(struct intel_super
*super
)
1294 __free_imsm(super
, 1);
1298 static void free_super_imsm(struct supertype
*st
)
1300 struct intel_super
*super
= st
->sb
;
1309 static struct intel_super
*alloc_super(int creating_imsm
)
1311 struct intel_super
*super
= malloc(sizeof(*super
));
1314 memset(super
, 0, sizeof(*super
));
1315 super
->creating_imsm
= creating_imsm
;
1316 super
->current_vol
= -1;
1323 static int load_super_imsm_all(struct supertype
*st
, int fd
, void **sbp
,
1324 char *devname
, int keep_fd
)
1327 struct intel_super
*super
;
1328 struct mdinfo
*sd
, *best
= NULL
;
1335 /* check if this disk is a member of an active array */
1336 sra
= sysfs_read(fd
, 0, GET_LEVEL
|GET_VERSION
|GET_DEVS
|GET_STATE
);
1340 if (sra
->array
.major_version
!= -1 ||
1341 sra
->array
.minor_version
!= -2 ||
1342 strcmp(sra
->text_version
, "imsm") != 0)
1345 super
= alloc_super(0);
1349 /* find the most up to date disk in this array, skipping spares */
1350 for (sd
= sra
->devs
; sd
; sd
= sd
->next
) {
1351 sprintf(nm
, "%d:%d", sd
->disk
.major
, sd
->disk
.minor
);
1352 dfd
= dev_open(nm
, keep_fd
? O_RDWR
: O_RDONLY
);
1357 rv
= load_imsm_mpb(dfd
, super
, NULL
);
1361 if (super
->anchor
->num_raid_devs
== 0)
1364 gen
= __le32_to_cpu(super
->anchor
->generation_num
);
1365 if (!best
|| gen
> bestgen
) {
1380 /* load the most up to date anchor */
1381 sprintf(nm
, "%d:%d", best
->disk
.major
, best
->disk
.minor
);
1382 dfd
= dev_open(nm
, O_RDONLY
);
1387 rv
= load_imsm_mpb(dfd
, super
, NULL
);
1394 /* re-parse the disk list with the current anchor */
1395 for (sd
= sra
->devs
; sd
; sd
= sd
->next
) {
1396 sprintf(nm
, "%d:%d", sd
->disk
.major
, sd
->disk
.minor
);
1397 dfd
= dev_open(nm
, keep_fd
? O_RDWR
: O_RDONLY
);
1402 load_imsm_disk(dfd
, super
, NULL
, keep_fd
);
1407 if (st
->subarray
[0]) {
1408 if (atoi(st
->subarray
) <= super
->anchor
->num_raid_devs
)
1409 super
->current_vol
= atoi(st
->subarray
);
1415 st
->container_dev
= fd2devnum(fd
);
1416 if (st
->ss
== NULL
) {
1417 st
->ss
= &super_imsm
;
1418 st
->minor_version
= 0;
1419 st
->max_devs
= IMSM_MAX_DEVICES
;
1421 st
->loaded_container
= 1;
1427 static int load_super_imsm(struct supertype
*st
, int fd
, char *devname
)
1429 struct intel_super
*super
;
1433 if (load_super_imsm_all(st
, fd
, &st
->sb
, devname
, 1) == 0)
1436 if (st
->subarray
[0])
1437 return 1; /* FIXME */
1439 super
= alloc_super(0);
1442 Name
": malloc of %zu failed.\n",
1447 rv
= load_imsm_mpb(fd
, super
, devname
);
1452 Name
": Failed to load all information "
1453 "sections on %s\n", devname
);
1459 if (st
->ss
== NULL
) {
1460 st
->ss
= &super_imsm
;
1461 st
->minor_version
= 0;
1462 st
->max_devs
= IMSM_MAX_DEVICES
;
1464 st
->loaded_container
= 0;
1469 static __u16
info_to_blocks_per_strip(mdu_array_info_t
*info
)
1471 if (info
->level
== 1)
1473 return info
->chunk_size
>> 9;
1476 static __u32
info_to_num_data_stripes(mdu_array_info_t
*info
)
1480 num_stripes
= (info
->size
* 2) / info_to_blocks_per_strip(info
);
1481 if (info
->level
== 1)
1487 static __u32
info_to_blocks_per_member(mdu_array_info_t
*info
)
1489 return (info
->size
* 2) & ~(info_to_blocks_per_strip(info
) - 1);
1492 static int init_super_imsm_volume(struct supertype
*st
, mdu_array_info_t
*info
,
1493 unsigned long long size
, char *name
,
1494 char *homehost
, int *uuid
)
1496 /* We are creating a volume inside a pre-existing container.
1497 * so st->sb is already set.
1499 struct intel_super
*super
= st
->sb
;
1500 struct imsm_super
*mpb
= super
->anchor
;
1501 struct imsm_dev
*dev
;
1502 struct imsm_vol
*vol
;
1503 struct imsm_map
*map
;
1504 int idx
= mpb
->num_raid_devs
;
1506 unsigned long long array_blocks
;
1508 size_t size_old
, size_new
;
1510 if (mpb
->num_raid_devs
>= 2) {
1511 fprintf(stderr
, Name
": This imsm-container already has the "
1512 "maximum of 2 volumes\n");
1516 /* ensure the mpb is large enough for the new data */
1517 size_old
= __le32_to_cpu(mpb
->mpb_size
);
1518 size_new
= disks_to_mpb_size(info
->nr_disks
);
1519 if (size_new
> size_old
) {
1521 size_t size_round
= ROUND_UP(size_new
, 512);
1523 if (posix_memalign(&mpb_new
, 512, size_round
) != 0) {
1524 fprintf(stderr
, Name
": could not allocate new mpb\n");
1527 memcpy(mpb_new
, mpb
, size_old
);
1530 super
->anchor
= mpb_new
;
1531 mpb
->mpb_size
= __cpu_to_le32(size_new
);
1532 memset(mpb_new
+ size_old
, 0, size_round
- size_old
);
1534 super
->current_vol
= idx
;
1535 /* when creating the first raid device in this container set num_disks
1536 * to zero, i.e. delete this spare and add raid member devices in
1537 * add_to_super_imsm_volume()
1539 if (super
->current_vol
== 0)
1541 sprintf(st
->subarray
, "%d", idx
);
1542 dev
= malloc(sizeof(*dev
) + sizeof(__u32
) * (info
->raid_disks
- 1));
1544 fprintf(stderr
, Name
": could not allocate raid device\n");
1547 strncpy((char *) dev
->volume
, name
, MAX_RAID_SERIAL_LEN
);
1548 array_blocks
= calc_array_size(info
->level
, info
->raid_disks
,
1549 info
->layout
, info
->chunk_size
,
1551 dev
->size_low
= __cpu_to_le32((__u32
) array_blocks
);
1552 dev
->size_high
= __cpu_to_le32((__u32
) (array_blocks
>> 32));
1553 dev
->status
= __cpu_to_le32(0);
1554 dev
->reserved_blocks
= __cpu_to_le32(0);
1556 vol
->migr_state
= 0;
1559 for (i
= 0; i
< idx
; i
++) {
1560 struct imsm_dev
*prev
= get_imsm_dev(super
, i
);
1561 struct imsm_map
*pmap
= get_imsm_map(prev
, 0);
1563 offset
+= __le32_to_cpu(pmap
->blocks_per_member
);
1564 offset
+= IMSM_RESERVED_SECTORS
;
1566 map
= get_imsm_map(dev
, 0);
1567 map
->pba_of_lba0
= __cpu_to_le32(offset
);
1568 map
->blocks_per_member
= __cpu_to_le32(info_to_blocks_per_member(info
));
1569 map
->blocks_per_strip
= __cpu_to_le16(info_to_blocks_per_strip(info
));
1570 map
->num_data_stripes
= __cpu_to_le32(info_to_num_data_stripes(info
));
1571 map
->map_state
= info
->level
? IMSM_T_STATE_UNINITIALIZED
:
1572 IMSM_T_STATE_NORMAL
;
1574 if (info
->level
== 1 && info
->raid_disks
> 2) {
1575 fprintf(stderr
, Name
": imsm does not support more than 2 disks"
1576 "in a raid1 volume\n");
1579 if (info
->level
== 10)
1580 map
->raid_level
= 1;
1582 map
->raid_level
= info
->level
;
1584 map
->num_members
= info
->raid_disks
;
1585 for (i
= 0; i
< map
->num_members
; i
++) {
1586 /* initialized in add_to_super */
1587 set_imsm_ord_tbl_ent(map
, i
, 0);
1589 mpb
->num_raid_devs
++;
1590 super
->dev_tbl
[super
->current_vol
] = dev
;
1595 static int init_super_imsm(struct supertype
*st
, mdu_array_info_t
*info
,
1596 unsigned long long size
, char *name
,
1597 char *homehost
, int *uuid
)
1599 /* This is primarily called by Create when creating a new array.
1600 * We will then get add_to_super called for each component, and then
1601 * write_init_super called to write it out to each device.
1602 * For IMSM, Create can create on fresh devices or on a pre-existing
1604 * To create on a pre-existing array a different method will be called.
1605 * This one is just for fresh drives.
1607 struct intel_super
*super
;
1608 struct imsm_super
*mpb
;
1616 return init_super_imsm_volume(st
, info
, size
, name
, homehost
,
1619 super
= alloc_super(1);
1622 mpb_size
= disks_to_mpb_size(info
->nr_disks
);
1623 if (posix_memalign(&super
->buf
, 512, mpb_size
) != 0) {
1628 memset(mpb
, 0, mpb_size
);
1630 memcpy(mpb
->sig
, MPB_SIGNATURE
, strlen(MPB_SIGNATURE
));
1631 memcpy(mpb
->sig
+ strlen(MPB_SIGNATURE
), MPB_VERSION_RAID5
,
1632 strlen(MPB_VERSION_RAID5
));
1633 mpb
->mpb_size
= mpb_size
;
1640 static void add_to_super_imsm_volume(struct supertype
*st
, mdu_disk_info_t
*dk
,
1641 int fd
, char *devname
)
1643 struct intel_super
*super
= st
->sb
;
1644 struct imsm_super
*mpb
= super
->anchor
;
1646 struct imsm_dev
*dev
;
1647 struct imsm_map
*map
;
1650 dev
= get_imsm_dev(super
, super
->current_vol
);
1651 map
= get_imsm_map(dev
, 0);
1653 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
1654 if (dl
->major
== dk
->major
&&
1655 dl
->minor
== dk
->minor
)
1658 if (!dl
|| ! (dk
->state
& (1<<MD_DISK_SYNC
)))
1661 /* add a pristine spare to the metadata */
1662 if (dl
->index
< 0) {
1663 dl
->index
= super
->anchor
->num_disks
;
1664 super
->anchor
->num_disks
++;
1666 set_imsm_ord_tbl_ent(map
, dk
->number
, dl
->index
);
1667 status
= CONFIGURED_DISK
| USABLE_DISK
;
1668 dl
->disk
.status
= __cpu_to_le32(status
);
1670 /* if we are creating the first raid device update the family number */
1671 if (super
->current_vol
== 0) {
1673 struct imsm_dev
*_dev
= __get_imsm_dev(mpb
, 0);
1674 struct imsm_disk
*_disk
= __get_imsm_disk(mpb
, dl
->index
);
1678 sum
= __gen_imsm_checksum(mpb
);
1679 mpb
->family_num
= __cpu_to_le32(sum
);
1683 static void add_to_super_imsm(struct supertype
*st
, mdu_disk_info_t
*dk
,
1684 int fd
, char *devname
)
1686 struct intel_super
*super
= st
->sb
;
1688 unsigned long long size
;
1693 if (super
->current_vol
>= 0) {
1694 add_to_super_imsm_volume(st
, dk
, fd
, devname
);
1699 dd
= malloc(sizeof(*dd
));
1702 Name
": malloc failed %s:%d.\n", __func__
, __LINE__
);
1705 memset(dd
, 0, sizeof(*dd
));
1706 dd
->major
= major(stb
.st_rdev
);
1707 dd
->minor
= minor(stb
.st_rdev
);
1709 dd
->devname
= devname
? strdup(devname
) : NULL
;
1711 rv
= imsm_read_serial(fd
, devname
, dd
->serial
);
1714 Name
": failed to retrieve scsi serial, aborting\n");
1719 get_dev_size(fd
, NULL
, &size
);
1721 status
= USABLE_DISK
| SPARE_DISK
;
1722 serialcpy(dd
->disk
.serial
, dd
->serial
);
1723 dd
->disk
.total_blocks
= __cpu_to_le32(size
);
1724 dd
->disk
.status
= __cpu_to_le32(status
);
1725 if (sysfs_disk_to_scsi_id(fd
, &id
) == 0)
1726 dd
->disk
.scsi_id
= __cpu_to_le32(id
);
1728 dd
->disk
.scsi_id
= __cpu_to_le32(0);
1730 if (st
->update_tail
) {
1731 dd
->next
= super
->add
;
1734 dd
->next
= super
->disks
;
1739 static int store_imsm_mpb(int fd
, struct intel_super
*super
);
1741 /* spare records have their own family number and do not have any defined raid
1744 static int write_super_imsm_spares(struct intel_super
*super
, int doclose
)
1746 struct imsm_super mpb_save
;
1747 struct imsm_super
*mpb
= super
->anchor
;
1752 mpb
->num_raid_devs
= 0;
1754 mpb
->mpb_size
= sizeof(struct imsm_super
);
1755 mpb
->generation_num
= __cpu_to_le32(1UL);
1757 for (d
= super
->disks
; d
; d
= d
->next
) {
1761 mpb
->disk
[0] = d
->disk
;
1762 sum
= __gen_imsm_checksum(mpb
);
1763 mpb
->family_num
= __cpu_to_le32(sum
);
1764 sum
= __gen_imsm_checksum(mpb
);
1765 mpb
->check_sum
= __cpu_to_le32(sum
);
1767 if (store_imsm_mpb(d
->fd
, super
)) {
1768 fprintf(stderr
, "%s: failed for device %d:%d %s\n",
1769 __func__
, d
->major
, d
->minor
, strerror(errno
));
1783 static int write_super_imsm(struct intel_super
*super
, int doclose
)
1785 struct imsm_super
*mpb
= super
->anchor
;
1791 __u32 mpb_size
= sizeof(struct imsm_super
) - sizeof(struct imsm_disk
);
1793 /* 'generation' is incremented everytime the metadata is written */
1794 generation
= __le32_to_cpu(mpb
->generation_num
);
1796 mpb
->generation_num
= __cpu_to_le32(generation
);
1798 for (d
= super
->disks
; d
; d
= d
->next
) {
1802 mpb
->disk
[d
->index
] = d
->disk
;
1803 mpb_size
+= sizeof(struct imsm_disk
);
1807 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
1808 struct imsm_dev
*dev
= __get_imsm_dev(mpb
, i
);
1810 imsm_copy_dev(dev
, super
->dev_tbl
[i
]);
1811 mpb_size
+= sizeof_imsm_dev(dev
, 0);
1813 mpb_size
+= __le32_to_cpu(mpb
->bbm_log_size
);
1814 mpb
->mpb_size
= __cpu_to_le32(mpb_size
);
1816 /* recalculate checksum */
1817 sum
= __gen_imsm_checksum(mpb
);
1818 mpb
->check_sum
= __cpu_to_le32(sum
);
1820 /* write the mpb for disks that compose raid devices */
1821 for (d
= super
->disks
; d
; d
= d
->next
) {
1824 if (store_imsm_mpb(d
->fd
, super
))
1825 fprintf(stderr
, "%s: failed for device %d:%d %s\n",
1826 __func__
, d
->major
, d
->minor
, strerror(errno
));
1834 return write_super_imsm_spares(super
, doclose
);
1840 static int create_array(struct supertype
*st
)
1843 struct imsm_update_create_array
*u
;
1844 struct intel_super
*super
= st
->sb
;
1845 struct imsm_dev
*dev
= get_imsm_dev(super
, super
->current_vol
);
1847 len
= sizeof(*u
) - sizeof(*dev
) + sizeof_imsm_dev(dev
, 0);
1850 fprintf(stderr
, "%s: failed to allocate update buffer\n",
1855 u
->type
= update_create_array
;
1856 u
->dev_idx
= super
->current_vol
;
1857 imsm_copy_dev(&u
->dev
, dev
);
1858 append_metadata_update(st
, u
, len
);
1863 static int _add_disk(struct supertype
*st
)
1865 struct intel_super
*super
= st
->sb
;
1867 struct imsm_update_add_disk
*u
;
1875 fprintf(stderr
, "%s: failed to allocate update buffer\n",
1880 u
->type
= update_add_disk
;
1881 append_metadata_update(st
, u
, len
);
1886 static int write_init_super_imsm(struct supertype
*st
)
1888 if (st
->update_tail
) {
1889 /* queue the recently created array / added disk
1890 * as a metadata update */
1891 struct intel_super
*super
= st
->sb
;
1895 /* determine if we are creating a volume or adding a disk */
1896 if (super
->current_vol
< 0) {
1897 /* in the add disk case we are running in mdmon
1898 * context, so don't close fd's
1900 return _add_disk(st
);
1902 rv
= create_array(st
);
1904 for (d
= super
->disks
; d
; d
= d
->next
) {
1911 return write_super_imsm(st
->sb
, 1);
1915 static int store_zero_imsm(struct supertype
*st
, int fd
)
1917 unsigned long long dsize
;
1920 get_dev_size(fd
, NULL
, &dsize
);
1922 /* first block is stored on second to last sector of the disk */
1923 if (lseek64(fd
, dsize
- (512 * 2), SEEK_SET
) < 0)
1926 if (posix_memalign(&buf
, 512, 512) != 0)
1929 memset(buf
, 0, 512);
1930 if (write(fd
, buf
, 512) != 512)
1935 static int imsm_bbm_log_size(struct imsm_super
*mpb
)
1937 return __le32_to_cpu(mpb
->bbm_log_size
);
1941 static int validate_geometry_imsm_container(struct supertype
*st
, int level
,
1942 int layout
, int raiddisks
, int chunk
,
1943 unsigned long long size
, char *dev
,
1944 unsigned long long *freesize
,
1948 unsigned long long ldsize
;
1950 if (level
!= LEVEL_CONTAINER
)
1955 fd
= open(dev
, O_RDONLY
|O_EXCL
, 0);
1958 fprintf(stderr
, Name
": imsm: Cannot open %s: %s\n",
1959 dev
, strerror(errno
));
1962 if (!get_dev_size(fd
, dev
, &ldsize
)) {
1968 *freesize
= avail_size_imsm(st
, ldsize
>> 9);
1973 /* validate_geometry_imsm_volume - lifted from validate_geometry_ddf_bvd
1974 * FIX ME add ahci details
1976 static int validate_geometry_imsm_volume(struct supertype
*st
, int level
,
1977 int layout
, int raiddisks
, int chunk
,
1978 unsigned long long size
, char *dev
,
1979 unsigned long long *freesize
,
1983 struct intel_super
*super
= st
->sb
;
1985 unsigned long long pos
= 0;
1986 unsigned long long maxsize
;
1990 if (level
== LEVEL_CONTAINER
)
1993 if (level
== 1 && raiddisks
> 2) {
1995 fprintf(stderr
, Name
": imsm does not support more "
1996 "than 2 in a raid1 configuration\n");
2000 /* We must have the container info already read in. */
2005 /* General test: make sure there is space for
2006 * 'raiddisks' device extents of size 'size' at a given
2009 unsigned long long minsize
= size
*2 /* convert to blocks */;
2010 unsigned long long start_offset
= ~0ULL;
2013 minsize
= MPB_SECTOR_CNT
+ IMSM_RESERVED_SECTORS
;
2014 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
2019 e
= get_extents(super
, dl
);
2022 unsigned long long esize
;
2023 esize
= e
[i
].start
- pos
;
2024 if (esize
>= minsize
)
2026 if (found
&& start_offset
== ~0ULL) {
2029 } else if (found
&& pos
!= start_offset
) {
2033 pos
= e
[i
].start
+ e
[i
].size
;
2035 } while (e
[i
-1].size
);
2040 if (dcnt
< raiddisks
) {
2042 fprintf(stderr
, Name
": imsm: Not enough "
2043 "devices with space for this array "
2050 /* This device must be a member of the set */
2051 if (stat(dev
, &stb
) < 0)
2053 if ((S_IFMT
& stb
.st_mode
) != S_IFBLK
)
2055 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
2056 if (dl
->major
== major(stb
.st_rdev
) &&
2057 dl
->minor
== minor(stb
.st_rdev
))
2062 fprintf(stderr
, Name
": %s is not in the "
2063 "same imsm set\n", dev
);
2066 e
= get_extents(super
, dl
);
2070 unsigned long long esize
;
2071 esize
= e
[i
].start
- pos
;
2072 if (esize
>= maxsize
)
2074 pos
= e
[i
].start
+ e
[i
].size
;
2076 } while (e
[i
-1].size
);
2077 *freesize
= maxsize
;
2082 static int validate_geometry_imsm(struct supertype
*st
, int level
, int layout
,
2083 int raiddisks
, int chunk
, unsigned long long size
,
2084 char *dev
, unsigned long long *freesize
,
2090 /* if given unused devices create a container
2091 * if given given devices in a container create a member volume
2093 if (level
== LEVEL_CONTAINER
) {
2094 /* Must be a fresh device to add to a container */
2095 return validate_geometry_imsm_container(st
, level
, layout
,
2096 raiddisks
, chunk
, size
,
2102 /* creating in a given container */
2103 return validate_geometry_imsm_volume(st
, level
, layout
,
2104 raiddisks
, chunk
, size
,
2105 dev
, freesize
, verbose
);
2108 /* limit creation to the following levels */
2120 /* This device needs to be a device in an 'imsm' container */
2121 fd
= open(dev
, O_RDONLY
|O_EXCL
, 0);
2125 Name
": Cannot create this array on device %s\n",
2130 if (errno
!= EBUSY
|| (fd
= open(dev
, O_RDONLY
, 0)) < 0) {
2132 fprintf(stderr
, Name
": Cannot open %s: %s\n",
2133 dev
, strerror(errno
));
2136 /* Well, it is in use by someone, maybe an 'imsm' container. */
2137 cfd
= open_container(fd
);
2141 fprintf(stderr
, Name
": Cannot use %s: It is busy\n",
2145 sra
= sysfs_read(cfd
, 0, GET_VERSION
);
2147 if (sra
&& sra
->array
.major_version
== -1 &&
2148 strcmp(sra
->text_version
, "imsm") == 0) {
2149 /* This is a member of a imsm container. Load the container
2150 * and try to create a volume
2152 struct intel_super
*super
;
2154 if (load_super_imsm_all(st
, cfd
, (void **) &super
, NULL
, 1) == 0) {
2156 st
->container_dev
= fd2devnum(cfd
);
2158 return validate_geometry_imsm_volume(st
, level
, layout
,
2164 } else /* may belong to another container */
2169 #endif /* MDASSEMBLE */
2171 static struct mdinfo
*container_content_imsm(struct supertype
*st
)
2173 /* Given a container loaded by load_super_imsm_all,
2174 * extract information about all the arrays into
2177 * For each imsm_dev create an mdinfo, fill it in,
2178 * then look for matching devices in super->disks
2179 * and create appropriate device mdinfo.
2181 struct intel_super
*super
= st
->sb
;
2182 struct imsm_super
*mpb
= super
->anchor
;
2183 struct mdinfo
*rest
= NULL
;
2186 /* do not assemble arrays that might have bad blocks */
2187 if (imsm_bbm_log_size(super
->anchor
)) {
2188 fprintf(stderr
, Name
": BBM log found in metadata. "
2189 "Cannot activate array(s).\n");
2193 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
2194 struct imsm_dev
*dev
= get_imsm_dev(super
, i
);
2195 struct imsm_map
*map
= get_imsm_map(dev
, 0);
2196 struct mdinfo
*this;
2199 this = malloc(sizeof(*this));
2200 memset(this, 0, sizeof(*this));
2203 super
->current_vol
= i
;
2204 getinfo_super_imsm_volume(st
, this);
2205 for (slot
= 0 ; slot
< map
->num_members
; slot
++) {
2206 struct mdinfo
*info_d
;
2214 idx
= get_imsm_disk_idx(dev
, slot
);
2215 ord
= get_imsm_ord_tbl_ent(dev
, slot
);
2216 for (d
= super
->disks
; d
; d
= d
->next
)
2217 if (d
->index
== idx
)
2223 s
= d
? __le32_to_cpu(d
->disk
.status
) : 0;
2224 if (s
& FAILED_DISK
)
2226 if (!(s
& USABLE_DISK
))
2228 if (ord
& IMSM_ORD_REBUILD
)
2232 * if we skip some disks the array will be assmebled degraded;
2233 * reset resync start to avoid a dirty-degraded situation
2235 * FIXME handle dirty degraded
2237 if (skip
&& !dev
->vol
.dirty
)
2238 this->resync_start
= ~0ULL;
2242 info_d
= malloc(sizeof(*info_d
));
2244 fprintf(stderr
, Name
": failed to allocate disk"
2245 " for volume %s\n", (char *) dev
->volume
);
2250 memset(info_d
, 0, sizeof(*info_d
));
2251 info_d
->next
= this->devs
;
2252 this->devs
= info_d
;
2254 info_d
->disk
.number
= d
->index
;
2255 info_d
->disk
.major
= d
->major
;
2256 info_d
->disk
.minor
= d
->minor
;
2257 info_d
->disk
.raid_disk
= slot
;
2259 this->array
.working_disks
++;
2261 info_d
->events
= __le32_to_cpu(mpb
->generation_num
);
2262 info_d
->data_offset
= __le32_to_cpu(map
->pba_of_lba0
);
2263 info_d
->component_size
= __le32_to_cpu(map
->blocks_per_member
);
2265 strcpy(info_d
->name
, d
->devname
);
2275 static int imsm_open_new(struct supertype
*c
, struct active_array
*a
,
2278 struct intel_super
*super
= c
->sb
;
2279 struct imsm_super
*mpb
= super
->anchor
;
2281 if (atoi(inst
) >= mpb
->num_raid_devs
) {
2282 fprintf(stderr
, "%s: subarry index %d, out of range\n",
2283 __func__
, atoi(inst
));
2287 dprintf("imsm: open_new %s\n", inst
);
2288 a
->info
.container_member
= atoi(inst
);
2292 static __u8
imsm_check_degraded(struct intel_super
*super
, struct imsm_dev
*dev
, int failed
)
2294 struct imsm_map
*map
= get_imsm_map(dev
, 0);
2297 return map
->map_state
== IMSM_T_STATE_UNINITIALIZED
?
2298 IMSM_T_STATE_UNINITIALIZED
: IMSM_T_STATE_NORMAL
;
2300 switch (get_imsm_raid_level(map
)) {
2302 return IMSM_T_STATE_FAILED
;
2305 if (failed
< map
->num_members
)
2306 return IMSM_T_STATE_DEGRADED
;
2308 return IMSM_T_STATE_FAILED
;
2313 * check to see if any mirrors have failed,
2314 * otherwise we are degraded
2316 int device_per_mirror
= 2; /* FIXME is this always the case?
2317 * and are they always adjacent?
2322 for (i
= 0; i
< map
->num_members
; i
++) {
2323 int idx
= get_imsm_disk_idx(dev
, i
);
2324 struct imsm_disk
*disk
= get_imsm_disk(super
, idx
);
2328 else if (__le32_to_cpu(disk
->status
) & FAILED_DISK
)
2331 if (r10fail
>= device_per_mirror
)
2332 return IMSM_T_STATE_FAILED
;
2334 /* reset 'r10fail' for next mirror set */
2335 if (!((i
+ 1) % device_per_mirror
))
2339 return IMSM_T_STATE_DEGRADED
;
2343 return IMSM_T_STATE_DEGRADED
;
2345 return IMSM_T_STATE_FAILED
;
2351 return map
->map_state
;
2354 static int imsm_count_failed(struct intel_super
*super
, struct imsm_dev
*dev
)
2358 struct imsm_disk
*disk
;
2359 struct imsm_map
*map
= get_imsm_map(dev
, 0);
2361 for (i
= 0; i
< map
->num_members
; i
++) {
2362 __u32 ord
= get_imsm_ord_tbl_ent(dev
, i
);
2363 int idx
= ord_to_idx(ord
);
2365 disk
= get_imsm_disk(super
, idx
);
2367 __le32_to_cpu(disk
->status
) & FAILED_DISK
||
2368 ord
& IMSM_ORD_REBUILD
)
2375 static int is_resyncing(struct imsm_dev
*dev
)
2377 struct imsm_map
*migr_map
;
2379 if (!dev
->vol
.migr_state
)
2382 if (dev
->vol
.migr_type
== 0)
2385 migr_map
= get_imsm_map(dev
, 1);
2387 if (migr_map
->map_state
== IMSM_T_STATE_NORMAL
)
2393 static int is_rebuilding(struct imsm_dev
*dev
)
2395 struct imsm_map
*migr_map
;
2397 if (!dev
->vol
.migr_state
)
2400 if (dev
->vol
.migr_type
== 0)
2403 migr_map
= get_imsm_map(dev
, 1);
2405 if (migr_map
->map_state
== IMSM_T_STATE_DEGRADED
)
2411 /* Handle dirty -> clean transititions and resync. Degraded and rebuild
2412 * states are handled in imsm_set_disk() with one exception, when a
2413 * resync is stopped due to a new failure this routine will set the
2414 * 'degraded' state for the array.
2416 static int imsm_set_array_state(struct active_array
*a
, int consistent
)
2418 int inst
= a
->info
.container_member
;
2419 struct intel_super
*super
= a
->container
->sb
;
2420 struct imsm_dev
*dev
= get_imsm_dev(super
, inst
);
2421 struct imsm_map
*map
= get_imsm_map(dev
, 0);
2422 int failed
= imsm_count_failed(super
, dev
);
2423 __u8 map_state
= imsm_check_degraded(super
, dev
, failed
);
2425 if (consistent
== 2 &&
2426 (a
->resync_start
!= ~0ULL ||
2427 map_state
!= IMSM_T_STATE_NORMAL
||
2428 dev
->vol
.migr_state
))
2431 if (a
->resync_start
== ~0ULL) {
2432 /* complete intialization / resync,
2433 * recovery is completed in ->set_disk
2435 if (is_resyncing(dev
)) {
2436 dprintf("imsm: mark resync done\n");
2437 dev
->vol
.migr_state
= 0;
2438 map
->map_state
= map_state
;
2439 super
->updates_pending
++;
2441 } else if (!is_resyncing(dev
) && !failed
) {
2442 /* mark the start of the init process if nothing is failed */
2443 dprintf("imsm: mark resync start (%llu)\n", a
->resync_start
);
2444 map
->map_state
= map_state
;
2445 migrate(dev
, IMSM_T_STATE_NORMAL
,
2446 map
->map_state
== IMSM_T_STATE_NORMAL
);
2447 super
->updates_pending
++;
2450 /* mark dirty / clean */
2451 if (dev
->vol
.dirty
!= !consistent
) {
2452 dprintf("imsm: mark '%s' (%llu)\n",
2453 consistent
? "clean" : "dirty", a
->resync_start
);
2458 super
->updates_pending
++;
2463 static void imsm_set_disk(struct active_array
*a
, int n
, int state
)
2465 int inst
= a
->info
.container_member
;
2466 struct intel_super
*super
= a
->container
->sb
;
2467 struct imsm_dev
*dev
= get_imsm_dev(super
, inst
);
2468 struct imsm_map
*map
= get_imsm_map(dev
, 0);
2469 struct imsm_disk
*disk
;
2475 if (n
> map
->num_members
)
2476 fprintf(stderr
, "imsm: set_disk %d out of range 0..%d\n",
2477 n
, map
->num_members
- 1);
2482 dprintf("imsm: set_disk %d:%x\n", n
, state
);
2484 ord
= get_imsm_ord_tbl_ent(dev
, n
);
2485 disk
= get_imsm_disk(super
, ord_to_idx(ord
));
2487 /* check for new failures */
2488 status
= __le32_to_cpu(disk
->status
);
2489 if ((state
& DS_FAULTY
) && !(status
& FAILED_DISK
)) {
2490 status
|= FAILED_DISK
;
2491 disk
->status
= __cpu_to_le32(status
);
2492 disk
->scsi_id
= __cpu_to_le32(~(__u32
)0);
2493 memmove(&disk
->serial
[0], &disk
->serial
[1], MAX_RAID_SERIAL_LEN
- 1);
2494 super
->updates_pending
++;
2496 /* check if in_sync */
2497 if (state
& DS_INSYNC
&& ord
& IMSM_ORD_REBUILD
) {
2498 struct imsm_map
*migr_map
= get_imsm_map(dev
, 1);
2500 set_imsm_ord_tbl_ent(migr_map
, n
, ord_to_idx(ord
));
2501 super
->updates_pending
++;
2504 failed
= imsm_count_failed(super
, dev
);
2505 map_state
= imsm_check_degraded(super
, dev
, failed
);
2507 /* check if recovery complete, newly degraded, or failed */
2508 if (map_state
== IMSM_T_STATE_NORMAL
&& is_rebuilding(dev
)) {
2509 map
->map_state
= map_state
;
2510 dev
->vol
.migr_state
= 0;
2511 super
->updates_pending
++;
2512 } else if (map_state
== IMSM_T_STATE_DEGRADED
&&
2513 map
->map_state
!= map_state
&&
2514 !dev
->vol
.migr_state
) {
2515 dprintf("imsm: mark degraded\n");
2516 map
->map_state
= map_state
;
2517 super
->updates_pending
++;
2518 } else if (map_state
== IMSM_T_STATE_FAILED
&&
2519 map
->map_state
!= map_state
) {
2520 dprintf("imsm: mark failed\n");
2521 dev
->vol
.migr_state
= 0;
2522 map
->map_state
= map_state
;
2523 super
->updates_pending
++;
2527 static int store_imsm_mpb(int fd
, struct intel_super
*super
)
2529 struct imsm_super
*mpb
= super
->anchor
;
2530 __u32 mpb_size
= __le32_to_cpu(mpb
->mpb_size
);
2531 unsigned long long dsize
;
2532 unsigned long long sectors
;
2534 get_dev_size(fd
, NULL
, &dsize
);
2536 if (mpb_size
> 512) {
2537 /* -1 to account for anchor */
2538 sectors
= mpb_sectors(mpb
) - 1;
2540 /* write the extended mpb to the sectors preceeding the anchor */
2541 if (lseek64(fd
, dsize
- (512 * (2 + sectors
)), SEEK_SET
) < 0)
2544 if (write(fd
, super
->buf
+ 512, 512 * sectors
) != 512 * sectors
)
2548 /* first block is stored on second to last sector of the disk */
2549 if (lseek64(fd
, dsize
- (512 * 2), SEEK_SET
) < 0)
2552 if (write(fd
, super
->buf
, 512) != 512)
2558 static void imsm_sync_metadata(struct supertype
*container
)
2560 struct intel_super
*super
= container
->sb
;
2562 if (!super
->updates_pending
)
2565 write_super_imsm(super
, 0);
2567 super
->updates_pending
= 0;
2570 static struct dl
*imsm_readd(struct intel_super
*super
, int idx
, struct active_array
*a
)
2572 struct imsm_dev
*dev
= get_imsm_dev(super
, a
->info
.container_member
);
2573 int i
= get_imsm_disk_idx(dev
, idx
);
2576 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
2580 if (dl
&& __le32_to_cpu(dl
->disk
.status
) & FAILED_DISK
)
2584 dprintf("%s: found %x:%x\n", __func__
, dl
->major
, dl
->minor
);
2589 static struct dl
*imsm_add_spare(struct intel_super
*super
, int slot
, struct active_array
*a
)
2591 struct imsm_dev
*dev
= get_imsm_dev(super
, a
->info
.container_member
);
2592 int idx
= get_imsm_disk_idx(dev
, slot
);
2593 struct imsm_map
*map
= get_imsm_map(dev
, 0);
2594 unsigned long long esize
;
2595 unsigned long long pos
;
2604 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
2605 /* If in this array, skip */
2606 for (d
= a
->info
.devs
; d
; d
= d
->next
)
2607 if (d
->state_fd
>= 0 &&
2608 d
->disk
.major
== dl
->major
&&
2609 d
->disk
.minor
== dl
->minor
) {
2610 dprintf("%x:%x already in array\n", dl
->major
, dl
->minor
);
2616 /* skip in use or failed drives */
2617 status
= __le32_to_cpu(dl
->disk
.status
);
2618 if (status
& FAILED_DISK
|| idx
== dl
->index
) {
2619 dprintf("%x:%x status ( %s%s)\n",
2620 dl
->major
, dl
->minor
,
2621 status
& FAILED_DISK
? "failed " : "",
2622 idx
== dl
->index
? "in use " : "");
2626 /* Does this unused device have the requisite free space?
2627 * We need a->info.component_size sectors
2629 ex
= get_extents(super
, dl
);
2631 dprintf("cannot get extents\n");
2637 array_start
= __le32_to_cpu(map
->pba_of_lba0
);
2640 /* check that we can start at pba_of_lba0 with
2641 * a->info.component_size of space
2643 esize
= ex
[j
].start
- pos
;
2644 if (array_start
>= pos
&&
2645 array_start
+ a
->info
.component_size
< ex
[j
].start
) {
2649 pos
= ex
[j
].start
+ ex
[j
].size
;
2652 } while (ex
[j
-1].size
);
2656 dprintf("%x:%x does not have %llu at %d\n",
2657 dl
->major
, dl
->minor
,
2658 a
->info
.component_size
,
2659 __le32_to_cpu(map
->pba_of_lba0
));
2669 static struct mdinfo
*imsm_activate_spare(struct active_array
*a
,
2670 struct metadata_update
**updates
)
2673 * Find a device with unused free space and use it to replace a
2674 * failed/vacant region in an array. We replace failed regions one a
2675 * array at a time. The result is that a new spare disk will be added
2676 * to the first failed array and after the monitor has finished
2677 * propagating failures the remainder will be consumed.
2679 * FIXME add a capability for mdmon to request spares from another
2683 struct intel_super
*super
= a
->container
->sb
;
2684 int inst
= a
->info
.container_member
;
2685 struct imsm_dev
*dev
= get_imsm_dev(super
, inst
);
2686 struct imsm_map
*map
= get_imsm_map(dev
, 0);
2687 int failed
= a
->info
.array
.raid_disks
;
2688 struct mdinfo
*rv
= NULL
;
2691 struct metadata_update
*mu
;
2693 struct imsm_update_activate_spare
*u
;
2697 for (d
= a
->info
.devs
; d
; d
= d
->next
) {
2698 if ((d
->curr_state
& DS_FAULTY
) &&
2700 /* wait for Removal to happen */
2702 if (d
->state_fd
>= 0)
2706 dprintf("imsm: activate spare: inst=%d failed=%d (%d) level=%d\n",
2707 inst
, failed
, a
->info
.array
.raid_disks
, a
->info
.array
.level
);
2708 if (imsm_check_degraded(super
, dev
, failed
) != IMSM_T_STATE_DEGRADED
)
2711 /* For each slot, if it is not working, find a spare */
2712 for (i
= 0; i
< a
->info
.array
.raid_disks
; i
++) {
2713 for (d
= a
->info
.devs
; d
; d
= d
->next
)
2714 if (d
->disk
.raid_disk
== i
)
2716 dprintf("found %d: %p %x\n", i
, d
, d
?d
->curr_state
:0);
2717 if (d
&& (d
->state_fd
>= 0))
2721 * OK, this device needs recovery. Try to re-add the previous
2722 * occupant of this slot, if this fails add a new spare
2724 dl
= imsm_readd(super
, i
, a
);
2726 dl
= imsm_add_spare(super
, i
, a
);
2730 /* found a usable disk with enough space */
2731 di
= malloc(sizeof(*di
));
2732 memset(di
, 0, sizeof(*di
));
2734 /* dl->index will be -1 in the case we are activating a
2735 * pristine spare. imsm_process_update() will create a
2736 * new index in this case. Once a disk is found to be
2737 * failed in all member arrays it is kicked from the
2740 di
->disk
.number
= dl
->index
;
2742 /* (ab)use di->devs to store a pointer to the device
2745 di
->devs
= (struct mdinfo
*) dl
;
2747 di
->disk
.raid_disk
= i
;
2748 di
->disk
.major
= dl
->major
;
2749 di
->disk
.minor
= dl
->minor
;
2751 di
->data_offset
= __le32_to_cpu(map
->pba_of_lba0
);
2752 di
->component_size
= a
->info
.component_size
;
2753 di
->container_member
= inst
;
2757 dprintf("%x:%x to be %d at %llu\n", dl
->major
, dl
->minor
,
2758 i
, di
->data_offset
);
2764 /* No spares found */
2766 /* Now 'rv' has a list of devices to return.
2767 * Create a metadata_update record to update the
2768 * disk_ord_tbl for the array
2770 mu
= malloc(sizeof(*mu
));
2771 mu
->buf
= malloc(sizeof(struct imsm_update_activate_spare
) * num_spares
);
2773 mu
->len
= sizeof(struct imsm_update_activate_spare
) * num_spares
;
2774 mu
->next
= *updates
;
2775 u
= (struct imsm_update_activate_spare
*) mu
->buf
;
2777 for (di
= rv
; di
; di
= di
->next
) {
2778 u
->type
= update_activate_spare
;
2779 u
->dl
= (struct dl
*) di
->devs
;
2781 u
->slot
= di
->disk
.raid_disk
;
2792 static int disks_overlap(struct imsm_dev
*d1
, struct imsm_dev
*d2
)
2794 struct imsm_map
*m1
= get_imsm_map(d1
, 0);
2795 struct imsm_map
*m2
= get_imsm_map(d2
, 0);
2800 for (i
= 0; i
< m1
->num_members
; i
++) {
2801 idx
= get_imsm_disk_idx(d1
, i
);
2802 for (j
= 0; j
< m2
->num_members
; j
++)
2803 if (idx
== get_imsm_disk_idx(d2
, j
))
2810 static void imsm_delete(struct intel_super
*super
, struct dl
**dlp
, int index
);
2812 static void imsm_process_update(struct supertype
*st
,
2813 struct metadata_update
*update
)
2816 * crack open the metadata_update envelope to find the update record
2817 * update can be one of:
2818 * update_activate_spare - a spare device has replaced a failed
2819 * device in an array, update the disk_ord_tbl. If this disk is
2820 * present in all member arrays then also clear the SPARE_DISK
2823 struct intel_super
*super
= st
->sb
;
2824 struct imsm_super
*mpb
;
2825 enum imsm_update_type type
= *(enum imsm_update_type
*) update
->buf
;
2827 /* update requires a larger buf but the allocation failed */
2828 if (super
->next_len
&& !super
->next_buf
) {
2829 super
->next_len
= 0;
2833 if (super
->next_buf
) {
2834 memcpy(super
->next_buf
, super
->buf
, super
->len
);
2836 super
->len
= super
->next_len
;
2837 super
->buf
= super
->next_buf
;
2839 super
->next_len
= 0;
2840 super
->next_buf
= NULL
;
2843 mpb
= super
->anchor
;
2846 case update_activate_spare
: {
2847 struct imsm_update_activate_spare
*u
= (void *) update
->buf
;
2848 struct imsm_dev
*dev
= get_imsm_dev(super
, u
->array
);
2849 struct imsm_map
*map
= get_imsm_map(dev
, 0);
2850 struct imsm_map
*migr_map
;
2851 struct active_array
*a
;
2852 struct imsm_disk
*disk
;
2858 int victim
= get_imsm_disk_idx(dev
, u
->slot
);
2861 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
2866 fprintf(stderr
, "error: imsm_activate_spare passed "
2867 "an unknown disk (index: %d)\n",
2872 super
->updates_pending
++;
2874 /* count failures (excluding rebuilds and the victim)
2875 * to determine map[0] state
2878 for (i
= 0; i
< map
->num_members
; i
++) {
2881 disk
= get_imsm_disk(super
, get_imsm_disk_idx(dev
, i
));
2883 __le32_to_cpu(disk
->status
) & FAILED_DISK
)
2887 /* adding a pristine spare, assign a new index */
2888 if (dl
->index
< 0) {
2889 dl
->index
= super
->anchor
->num_disks
;
2890 super
->anchor
->num_disks
++;
2893 status
= __le32_to_cpu(disk
->status
);
2894 status
|= CONFIGURED_DISK
;
2895 status
&= ~SPARE_DISK
;
2896 disk
->status
= __cpu_to_le32(status
);
2899 to_state
= imsm_check_degraded(super
, dev
, failed
);
2900 map
->map_state
= IMSM_T_STATE_DEGRADED
;
2901 migrate(dev
, to_state
, 1);
2902 migr_map
= get_imsm_map(dev
, 1);
2903 set_imsm_ord_tbl_ent(map
, u
->slot
, dl
->index
);
2904 set_imsm_ord_tbl_ent(migr_map
, u
->slot
, dl
->index
| IMSM_ORD_REBUILD
);
2906 /* count arrays using the victim in the metadata */
2908 for (a
= st
->arrays
; a
; a
= a
->next
) {
2909 dev
= get_imsm_dev(super
, a
->info
.container_member
);
2910 for (i
= 0; i
< map
->num_members
; i
++)
2911 if (victim
== get_imsm_disk_idx(dev
, i
))
2915 /* delete the victim if it is no longer being
2921 for (dlp
= &super
->disks
; *dlp
; dlp
= &(*dlp
)->next
)
2922 if ((*dlp
)->index
== victim
)
2924 /* We know that 'manager' isn't touching anything,
2927 imsm_delete(super
, dlp
, victim
);
2931 case update_create_array
: {
2932 /* someone wants to create a new array, we need to be aware of
2933 * a few races/collisions:
2934 * 1/ 'Create' called by two separate instances of mdadm
2935 * 2/ 'Create' versus 'activate_spare': mdadm has chosen
2936 * devices that have since been assimilated via
2938 * In the event this update can not be carried out mdadm will
2939 * (FIX ME) notice that its update did not take hold.
2941 struct imsm_update_create_array
*u
= (void *) update
->buf
;
2942 struct imsm_dev
*dev
;
2943 struct imsm_map
*map
, *new_map
;
2944 unsigned long long start
, end
;
2945 unsigned long long new_start
, new_end
;
2949 /* handle racing creates: first come first serve */
2950 if (u
->dev_idx
< mpb
->num_raid_devs
) {
2951 dprintf("%s: subarray %d already defined\n",
2952 __func__
, u
->dev_idx
);
2956 /* check update is next in sequence */
2957 if (u
->dev_idx
!= mpb
->num_raid_devs
) {
2958 dprintf("%s: can not create array %d expected index %d\n",
2959 __func__
, u
->dev_idx
, mpb
->num_raid_devs
);
2963 new_map
= get_imsm_map(&u
->dev
, 0);
2964 new_start
= __le32_to_cpu(new_map
->pba_of_lba0
);
2965 new_end
= new_start
+ __le32_to_cpu(new_map
->blocks_per_member
);
2967 /* handle activate_spare versus create race:
2968 * check to make sure that overlapping arrays do not include
2971 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
2972 dev
= get_imsm_dev(super
, i
);
2973 map
= get_imsm_map(dev
, 0);
2974 start
= __le32_to_cpu(map
->pba_of_lba0
);
2975 end
= start
+ __le32_to_cpu(map
->blocks_per_member
);
2976 if ((new_start
>= start
&& new_start
<= end
) ||
2977 (start
>= new_start
&& start
<= new_end
))
2979 if (overlap
&& disks_overlap(dev
, &u
->dev
)) {
2980 dprintf("%s: arrays overlap\n", __func__
);
2984 /* check num_members sanity */
2985 if (new_map
->num_members
> mpb
->num_disks
) {
2986 dprintf("%s: num_disks out of range\n", __func__
);
2990 /* check that prepare update was successful */
2991 if (!update
->space
) {
2992 dprintf("%s: prepare update failed\n", __func__
);
2996 super
->updates_pending
++;
2997 dev
= update
->space
;
2998 map
= get_imsm_map(dev
, 0);
2999 update
->space
= NULL
;
3000 imsm_copy_dev(dev
, &u
->dev
);
3001 map
= get_imsm_map(dev
, 0);
3002 super
->dev_tbl
[u
->dev_idx
] = dev
;
3003 mpb
->num_raid_devs
++;
3006 for (i
= 0; i
< map
->num_members
; i
++) {
3007 struct imsm_disk
*disk
;
3010 disk
= get_imsm_disk(super
, get_imsm_disk_idx(dev
, i
));
3011 status
= __le32_to_cpu(disk
->status
);
3012 status
|= CONFIGURED_DISK
;
3013 status
&= ~SPARE_DISK
;
3014 disk
->status
= __cpu_to_le32(status
);
3018 case update_add_disk
:
3020 /* we may be able to repair some arrays if disks are
3023 struct active_array
*a
;
3024 for (a
= st
->arrays
; a
; a
= a
->next
)
3025 a
->check_degraded
= 1;
3027 /* add some spares to the metadata */
3028 while (super
->add
) {
3032 super
->add
= al
->next
;
3033 al
->next
= super
->disks
;
3035 dprintf("%s: added %x:%x\n",
3036 __func__
, al
->major
, al
->minor
);
3043 static void imsm_prepare_update(struct supertype
*st
,
3044 struct metadata_update
*update
)
3047 * Allocate space to hold new disk entries, raid-device entries or a new
3048 * mpb if necessary. The manager synchronously waits for updates to
3049 * complete in the monitor, so new mpb buffers allocated here can be
3050 * integrated by the monitor thread without worrying about live pointers
3051 * in the manager thread.
3053 enum imsm_update_type type
= *(enum imsm_update_type
*) update
->buf
;
3054 struct intel_super
*super
= st
->sb
;
3055 struct imsm_super
*mpb
= super
->anchor
;
3060 case update_create_array
: {
3061 struct imsm_update_create_array
*u
= (void *) update
->buf
;
3063 len
= sizeof_imsm_dev(&u
->dev
, 1);
3064 update
->space
= malloc(len
);
3071 /* check if we need a larger metadata buffer */
3072 if (super
->next_buf
)
3073 buf_len
= super
->next_len
;
3075 buf_len
= super
->len
;
3077 if (__le32_to_cpu(mpb
->mpb_size
) + len
> buf_len
) {
3078 /* ok we need a larger buf than what is currently allocated
3079 * if this allocation fails process_update will notice that
3080 * ->next_len is set and ->next_buf is NULL
3082 buf_len
= ROUND_UP(__le32_to_cpu(mpb
->mpb_size
) + len
, 512);
3083 if (super
->next_buf
)
3084 free(super
->next_buf
);
3086 super
->next_len
= buf_len
;
3087 if (posix_memalign(&super
->next_buf
, buf_len
, 512) != 0)
3088 super
->next_buf
= NULL
;
3092 /* must be called while manager is quiesced */
3093 static void imsm_delete(struct intel_super
*super
, struct dl
**dlp
, int index
)
3095 struct imsm_super
*mpb
= super
->anchor
;
3097 struct imsm_dev
*dev
;
3098 struct imsm_map
*map
;
3099 int i
, j
, num_members
;
3102 dprintf("%s: deleting device[%d] from imsm_super\n",
3105 /* shift all indexes down one */
3106 for (iter
= super
->disks
; iter
; iter
= iter
->next
)
3107 if (iter
->index
> index
)
3110 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
3111 dev
= get_imsm_dev(super
, i
);
3112 map
= get_imsm_map(dev
, 0);
3113 num_members
= map
->num_members
;
3114 for (j
= 0; j
< num_members
; j
++) {
3115 /* update ord entries being careful not to propagate
3116 * ord-flags to the first map
3118 ord
= get_imsm_ord_tbl_ent(dev
, j
);
3120 if (ord_to_idx(ord
) <= index
)
3123 map
= get_imsm_map(dev
, 0);
3124 set_imsm_ord_tbl_ent(map
, j
, ord_to_idx(ord
- 1));
3125 map
= get_imsm_map(dev
, 1);
3127 set_imsm_ord_tbl_ent(map
, j
, ord
- 1);
3132 super
->updates_pending
++;
3134 struct dl
*dl
= *dlp
;
3136 *dlp
= (*dlp
)->next
;
3137 __free_imsm_disk(dl
);
3140 #endif /* MDASSEMBLE */
3142 struct superswitch super_imsm
= {
3144 .examine_super
= examine_super_imsm
,
3145 .brief_examine_super
= brief_examine_super_imsm
,
3146 .detail_super
= detail_super_imsm
,
3147 .brief_detail_super
= brief_detail_super_imsm
,
3148 .write_init_super
= write_init_super_imsm
,
3149 .validate_geometry
= validate_geometry_imsm
,
3150 .add_to_super
= add_to_super_imsm
,
3152 .match_home
= match_home_imsm
,
3153 .uuid_from_super
= uuid_from_super_imsm
,
3154 .getinfo_super
= getinfo_super_imsm
,
3155 .update_super
= update_super_imsm
,
3157 .avail_size
= avail_size_imsm
,
3159 .compare_super
= compare_super_imsm
,
3161 .load_super
= load_super_imsm
,
3162 .init_super
= init_super_imsm
,
3163 .store_super
= store_zero_imsm
,
3164 .free_super
= free_super_imsm
,
3165 .match_metadata_desc
= match_metadata_desc_imsm
,
3166 .container_content
= container_content_imsm
,
3172 .open_new
= imsm_open_new
,
3173 .load_super
= load_super_imsm
,
3174 .set_array_state
= imsm_set_array_state
,
3175 .set_disk
= imsm_set_disk
,
3176 .sync_metadata
= imsm_sync_metadata
,
3177 .activate_spare
= imsm_activate_spare
,
3178 .process_update
= imsm_process_update
,
3179 .prepare_update
= imsm_prepare_update
,
3180 #endif /* MDASSEMBLE */