2 * mdadm - Intel(R) Matrix Storage Manager Support
4 * Copyright (C) 2002-2007 Intel Corporation
6 * This program is free software; you can redistribute it and/or modify it
7 * under the terms and conditions of the GNU General Public License,
8 * version 2, as published by the Free Software Foundation.
10 * This program is distributed in the hope it will be useful, but WITHOUT
11 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
12 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for
15 * You should have received a copy of the GNU General Public License along with
16 * this program; if not, write to the Free Software Foundation, Inc.,
17 * 51 Franklin St - Fifth Floor, Boston, MA 02110-1301 USA.
26 /* MPB == Metadata Parameter Block */
27 #define MPB_SIGNATURE "Intel Raid ISM Cfg Sig. "
28 #define MPB_SIG_LEN (strlen(MPB_SIGNATURE))
29 #define MPB_VERSION_RAID0 "1.0.00"
30 #define MPB_VERSION_RAID1 "1.1.00"
31 #define MPB_VERSION_RAID5 "1.2.02"
32 #define MAX_SIGNATURE_LENGTH 32
33 #define MAX_RAID_SERIAL_LEN 16
34 #define MPB_SECTOR_CNT 418
35 #define IMSM_RESERVED_SECTORS 4096
37 /* Disk configuration info. */
38 #define IMSM_MAX_DEVICES 255
40 __u8 serial
[MAX_RAID_SERIAL_LEN
];/* 0xD8 - 0xE7 ascii serial number */
41 __u32 total_blocks
; /* 0xE8 - 0xEB total blocks */
42 __u32 scsi_id
; /* 0xEC - 0xEF scsi ID */
43 __u32 status
; /* 0xF0 - 0xF3 */
44 #define SPARE_DISK 0x01 /* Spare */
45 #define CONFIGURED_DISK 0x02 /* Member of some RaidDev */
46 #define FAILED_DISK 0x04 /* Permanent failure */
47 #define USABLE_DISK 0x08 /* Fully usable unless FAILED_DISK is set */
49 #define IMSM_DISK_FILLERS 5
50 __u32 filler
[IMSM_DISK_FILLERS
]; /* 0xF4 - 0x107 MPB_DISK_FILLERS for future expansion */
53 /* RAID map configuration infos. */
55 __u32 pba_of_lba0
; /* start address of partition */
56 __u32 blocks_per_member
;/* blocks per member */
57 __u32 num_data_stripes
; /* number of data stripes */
58 __u16 blocks_per_strip
;
59 __u8 map_state
; /* Normal, Uninitialized, Degraded, Failed */
60 #define IMSM_T_STATE_NORMAL 0
61 #define IMSM_T_STATE_UNINITIALIZED 1
62 #define IMSM_T_STATE_DEGRADED 2 /* FIXME: is this correct? */
63 #define IMSM_T_STATE_FAILED 3 /* FIXME: is this correct? */
65 #define IMSM_T_RAID0 0
66 #define IMSM_T_RAID1 1
67 #define IMSM_T_RAID5 5 /* since metadata version 1.2.02 ? */
68 __u8 num_members
; /* number of member disks */
70 __u32 filler
[7]; /* expansion area */
71 #define IMSM_ORD_REBUILD (1 << 24)
72 __u32 disk_ord_tbl
[1]; /* disk_ord_tbl[num_members],
73 * top byte contains some flags
75 } __attribute__ ((packed
));
79 __u8 migr_state
; /* Normal or Migrating */
80 __u8 migr_type
; /* Initializing, Rebuilding, ... */
84 struct imsm_map map
[1];
85 /* here comes another one if migr_state */
86 } __attribute__ ((packed
));
89 __u8 volume
[MAX_RAID_SERIAL_LEN
];
92 __u32 status
; /* Persistent RaidDev status */
93 __u32 reserved_blocks
; /* Reserved blocks at beginning of volume */
94 #define IMSM_DEV_FILLERS 12
95 __u32 filler
[IMSM_DEV_FILLERS
];
97 } __attribute__ ((packed
));
100 __u8 sig
[MAX_SIGNATURE_LENGTH
]; /* 0x00 - 0x1F */
101 __u32 check_sum
; /* 0x20 - 0x23 MPB Checksum */
102 __u32 mpb_size
; /* 0x24 - 0x27 Size of MPB */
103 __u32 family_num
; /* 0x28 - 0x2B Checksum from first time this config was written */
104 __u32 generation_num
; /* 0x2C - 0x2F Incremented each time this array's MPB is written */
105 __u32 error_log_size
; /* 0x30 - 0x33 in bytes */
106 __u32 attributes
; /* 0x34 - 0x37 */
107 __u8 num_disks
; /* 0x38 Number of configured disks */
108 __u8 num_raid_devs
; /* 0x39 Number of configured volumes */
109 __u8 error_log_pos
; /* 0x3A */
110 __u8 fill
[1]; /* 0x3B */
111 __u32 cache_size
; /* 0x3c - 0x40 in mb */
112 __u32 orig_family_num
; /* 0x40 - 0x43 original family num */
113 __u32 pwr_cycle_count
; /* 0x44 - 0x47 simulated power cycle count for array */
114 __u32 bbm_log_size
; /* 0x48 - 0x4B - size of bad Block Mgmt Log in bytes */
115 #define IMSM_FILLERS 35
116 __u32 filler
[IMSM_FILLERS
]; /* 0x4C - 0xD7 RAID_MPB_FILLERS */
117 struct imsm_disk disk
[1]; /* 0xD8 diskTbl[numDisks] */
118 /* here comes imsm_dev[num_raid_devs] */
119 /* here comes BBM logs */
120 } __attribute__ ((packed
));
122 #define BBM_LOG_MAX_ENTRIES 254
124 struct bbm_log_entry
{
125 __u64 defective_block_start
;
126 #define UNREADABLE 0xFFFFFFFF
127 __u32 spare_block_offset
;
128 __u16 remapped_marked_count
;
130 } __attribute__ ((__packed__
));
133 __u32 signature
; /* 0xABADB10C */
135 __u32 reserved_spare_block_count
; /* 0 */
136 __u32 reserved
; /* 0xFFFF */
137 __u64 first_spare_lba
;
138 struct bbm_log_entry mapped_block_entries
[BBM_LOG_MAX_ENTRIES
];
139 } __attribute__ ((__packed__
));
143 static char *map_state_str
[] = { "normal", "uninitialized", "degraded", "failed" };
146 static unsigned int sector_count(__u32 bytes
)
148 return ((bytes
+ (512-1)) & (~(512-1))) / 512;
151 static unsigned int mpb_sectors(struct imsm_super
*mpb
)
153 return sector_count(__le32_to_cpu(mpb
->mpb_size
));
156 /* internal representation of IMSM metadata */
159 void *buf
; /* O_DIRECT buffer for reading/writing metadata */
160 struct imsm_super
*anchor
; /* immovable parameters */
162 size_t len
; /* size of the 'buf' allocation */
163 int updates_pending
; /* count of pending updates for mdmon */
164 int creating_imsm
; /* flag to indicate container creation */
165 int current_vol
; /* index of raid device undergoing creation */
166 #define IMSM_MAX_RAID_DEVS 2
167 struct imsm_dev
*dev_tbl
[IMSM_MAX_RAID_DEVS
];
171 __u8 serial
[MAX_RAID_SERIAL_LEN
];
174 struct imsm_disk disk
;
177 struct bbm_log
*bbm_log
;
181 unsigned long long start
, size
;
184 /* definition of messages passed to imsm_process_update */
185 enum imsm_update_type
{
186 update_activate_spare
,
190 struct imsm_update_activate_spare
{
191 enum imsm_update_type type
;
195 struct imsm_update_activate_spare
*next
;
198 struct imsm_update_create_array
{
199 enum imsm_update_type type
;
204 static int imsm_env_devname_as_serial(void)
206 char *val
= getenv("IMSM_DEVNAME_AS_SERIAL");
208 if (val
&& atoi(val
) == 1)
215 static struct supertype
*match_metadata_desc_imsm(char *arg
)
217 struct supertype
*st
;
219 if (strcmp(arg
, "imsm") != 0 &&
220 strcmp(arg
, "default") != 0
224 st
= malloc(sizeof(*st
));
225 memset(st
, 0, sizeof(*st
));
226 st
->ss
= &super_imsm
;
227 st
->max_devs
= IMSM_MAX_DEVICES
;
228 st
->minor_version
= 0;
233 static __u8
*get_imsm_version(struct imsm_super
*mpb
)
235 return &mpb
->sig
[MPB_SIG_LEN
];
238 /* retrieve a disk directly from the anchor when the anchor is known to be
239 * up-to-date, currently only at load time
241 static struct imsm_disk
*__get_imsm_disk(struct imsm_super
*mpb
, __u8 index
)
243 if (index
>= mpb
->num_disks
)
245 return &mpb
->disk
[index
];
248 /* retrieve a disk from the parsed metadata */
249 static struct imsm_disk
*get_imsm_disk(struct intel_super
*super
, __u8 index
)
253 for (d
= super
->disks
; d
; d
= d
->next
)
254 if (d
->index
== index
)
260 /* generate a checksum directly from the anchor when the anchor is known to be
261 * up-to-date, currently only at load or write_super after coalescing
263 static __u32
__gen_imsm_checksum(struct imsm_super
*mpb
)
265 __u32 end
= mpb
->mpb_size
/ sizeof(end
);
266 __u32
*p
= (__u32
*) mpb
;
270 sum
+= __le32_to_cpu(*p
++);
272 return sum
- __le32_to_cpu(mpb
->check_sum
);
275 static size_t sizeof_imsm_map(struct imsm_map
*map
)
277 return sizeof(struct imsm_map
) + sizeof(__u32
) * (map
->num_members
- 1);
280 struct imsm_map
*get_imsm_map(struct imsm_dev
*dev
, int second_map
)
282 struct imsm_map
*map
= &dev
->vol
.map
[0];
284 if (second_map
&& !dev
->vol
.migr_state
)
286 else if (second_map
) {
289 return ptr
+ sizeof_imsm_map(map
);
295 /* return the size of the device.
296 * migr_state increases the returned size if map[0] were to be duplicated
298 static size_t sizeof_imsm_dev(struct imsm_dev
*dev
, int migr_state
)
300 size_t size
= sizeof(*dev
) - sizeof(struct imsm_map
) +
301 sizeof_imsm_map(get_imsm_map(dev
, 0));
303 /* migrating means an additional map */
304 if (dev
->vol
.migr_state
)
305 size
+= sizeof_imsm_map(get_imsm_map(dev
, 1));
307 size
+= sizeof_imsm_map(get_imsm_map(dev
, 0));
312 static struct imsm_dev
*__get_imsm_dev(struct imsm_super
*mpb
, __u8 index
)
318 if (index
>= mpb
->num_raid_devs
)
321 /* devices start after all disks */
322 offset
= ((void *) &mpb
->disk
[mpb
->num_disks
]) - _mpb
;
324 for (i
= 0; i
<= index
; i
++)
326 return _mpb
+ offset
;
328 offset
+= sizeof_imsm_dev(_mpb
+ offset
, 0);
333 static struct imsm_dev
*get_imsm_dev(struct intel_super
*super
, __u8 index
)
335 if (index
>= super
->anchor
->num_raid_devs
)
337 return super
->dev_tbl
[index
];
340 static __u32
get_imsm_disk_idx(struct imsm_map
*map
, int slot
)
342 __u32
*ord_tbl
= &map
->disk_ord_tbl
[slot
];
344 /* top byte identifies disk under rebuild
345 * why not just use the USABLE bit... oh well.
347 return __le32_to_cpu(*ord_tbl
& ~(0xff << 24));
350 static __u32
get_imsm_ord_tbl_ent(struct imsm_dev
*dev
, int slot
)
352 struct imsm_map
*map
;
354 if (dev
->vol
.migr_state
)
355 map
= get_imsm_map(dev
, 0);
357 map
= get_imsm_map(dev
, 1);
359 return map
->disk_ord_tbl
[slot
];
362 static int get_imsm_raid_level(struct imsm_map
*map
)
364 if (map
->raid_level
== 1) {
365 if (map
->num_members
== 2)
371 return map
->raid_level
;
374 static int cmp_extent(const void *av
, const void *bv
)
376 const struct extent
*a
= av
;
377 const struct extent
*b
= bv
;
378 if (a
->start
< b
->start
)
380 if (a
->start
> b
->start
)
385 static struct extent
*get_extents(struct intel_super
*super
, struct dl
*dl
)
387 /* find a list of used extents on the given physical device */
388 struct extent
*rv
, *e
;
392 for (i
= 0; i
< super
->anchor
->num_raid_devs
; i
++) {
393 struct imsm_dev
*dev
= get_imsm_dev(super
, i
);
394 struct imsm_map
*map
= get_imsm_map(dev
, 0);
396 for (j
= 0; j
< map
->num_members
; j
++) {
397 __u32 index
= get_imsm_disk_idx(map
, j
);
399 if (index
== dl
->index
)
403 rv
= malloc(sizeof(struct extent
) * (memberships
+ 1));
408 for (i
= 0; i
< super
->anchor
->num_raid_devs
; i
++) {
409 struct imsm_dev
*dev
= get_imsm_dev(super
, i
);
410 struct imsm_map
*map
= get_imsm_map(dev
, 0);
412 for (j
= 0; j
< map
->num_members
; j
++) {
413 __u32 index
= get_imsm_disk_idx(map
, j
);
415 if (index
== dl
->index
) {
416 e
->start
= __le32_to_cpu(map
->pba_of_lba0
);
417 e
->size
= __le32_to_cpu(map
->blocks_per_member
);
422 qsort(rv
, memberships
, sizeof(*rv
), cmp_extent
);
424 e
->start
= __le32_to_cpu(dl
->disk
.total_blocks
) -
425 (MPB_SECTOR_CNT
+ IMSM_RESERVED_SECTORS
);
431 static void print_imsm_dev(struct imsm_dev
*dev
, int index
)
435 struct imsm_map
*map
= get_imsm_map(dev
, 0);
438 printf("[%s]:\n", dev
->volume
);
439 printf(" RAID Level : %d\n", get_imsm_raid_level(map
));
440 printf(" Members : %d\n", map
->num_members
);
441 for (slot
= 0; slot
< map
->num_members
; slot
++)
442 if (index
== get_imsm_disk_idx(map
, slot
))
444 if (slot
< map
->num_members
)
445 printf(" This Slot : %d\n", slot
);
447 printf(" This Slot : ?\n");
448 sz
= __le32_to_cpu(dev
->size_high
);
450 sz
+= __le32_to_cpu(dev
->size_low
);
451 printf(" Array Size : %llu%s\n", (unsigned long long)sz
,
452 human_size(sz
* 512));
453 sz
= __le32_to_cpu(map
->blocks_per_member
);
454 printf(" Per Dev Size : %llu%s\n", (unsigned long long)sz
,
455 human_size(sz
* 512));
456 printf(" Sector Offset : %u\n",
457 __le32_to_cpu(map
->pba_of_lba0
));
458 printf(" Num Stripes : %u\n",
459 __le32_to_cpu(map
->num_data_stripes
));
460 printf(" Chunk Size : %u KiB\n",
461 __le16_to_cpu(map
->blocks_per_strip
) / 2);
462 printf(" Reserved : %d\n", __le32_to_cpu(dev
->reserved_blocks
));
463 printf(" Migrate State : %s", dev
->vol
.migr_state
? "migrating" : "idle");
464 if (dev
->vol
.migr_state
)
465 printf(": %s", dev
->vol
.migr_type
? "rebuilding" : "initializing");
467 printf(" Map State : %s", map_state_str
[map
->map_state
]);
468 if (dev
->vol
.migr_state
) {
469 struct imsm_map
*map
= get_imsm_map(dev
, 1);
470 printf(", %s", map_state_str
[map
->map_state
]);
473 printf(" Dirty State : %s\n", dev
->vol
.dirty
? "dirty" : "clean");
476 static void print_imsm_disk(struct imsm_super
*mpb
, int index
)
478 struct imsm_disk
*disk
= __get_imsm_disk(mpb
, index
);
479 char str
[MAX_RAID_SERIAL_LEN
];
487 snprintf(str
, MAX_RAID_SERIAL_LEN
, "%s", disk
->serial
);
488 printf(" Disk%02d Serial : %s\n", index
, str
);
489 s
= __le32_to_cpu(disk
->status
);
490 printf(" State :%s%s%s%s\n", s
&SPARE_DISK
? " spare" : "",
491 s
&CONFIGURED_DISK
? " active" : "",
492 s
&FAILED_DISK
? " failed" : "",
493 s
&USABLE_DISK
? " usable" : "");
494 printf(" Id : %08x\n", __le32_to_cpu(disk
->scsi_id
));
495 sz
= __le32_to_cpu(disk
->total_blocks
) -
496 (MPB_SECTOR_CNT
+ IMSM_RESERVED_SECTORS
* mpb
->num_raid_devs
);
497 printf(" Usable Size : %llu%s\n", (unsigned long long)sz
,
498 human_size(sz
* 512));
501 static void examine_super_imsm(struct supertype
*st
, char *homehost
)
503 struct intel_super
*super
= st
->sb
;
504 struct imsm_super
*mpb
= super
->anchor
;
505 char str
[MAX_SIGNATURE_LENGTH
];
509 snprintf(str
, MPB_SIG_LEN
, "%s", mpb
->sig
);
510 printf(" Magic : %s\n", str
);
511 snprintf(str
, strlen(MPB_VERSION_RAID0
), "%s", get_imsm_version(mpb
));
512 printf(" Version : %s\n", get_imsm_version(mpb
));
513 printf(" Family : %08x\n", __le32_to_cpu(mpb
->family_num
));
514 printf(" Generation : %08x\n", __le32_to_cpu(mpb
->generation_num
));
515 sum
= __le32_to_cpu(mpb
->check_sum
);
516 printf(" Checksum : %08x %s\n", sum
,
517 __gen_imsm_checksum(mpb
) == sum
? "correct" : "incorrect");
518 printf(" MPB Sectors : %d\n", mpb_sectors(mpb
));
519 printf(" Disks : %d\n", mpb
->num_disks
);
520 printf(" RAID Devices : %d\n", mpb
->num_raid_devs
);
521 print_imsm_disk(mpb
, super
->disks
->index
);
522 if (super
->bbm_log
) {
523 struct bbm_log
*log
= super
->bbm_log
;
526 printf("Bad Block Management Log:\n");
527 printf(" Log Size : %d\n", __le32_to_cpu(mpb
->bbm_log_size
));
528 printf(" Signature : %x\n", __le32_to_cpu(log
->signature
));
529 printf(" Entry Count : %d\n", __le32_to_cpu(log
->entry_count
));
530 printf(" Spare Blocks : %d\n", __le32_to_cpu(log
->reserved_spare_block_count
));
531 printf(" First Spare : %llx\n", __le64_to_cpu(log
->first_spare_lba
));
533 for (i
= 0; i
< mpb
->num_raid_devs
; i
++)
534 print_imsm_dev(__get_imsm_dev(mpb
, i
), super
->disks
->index
);
535 for (i
= 0; i
< mpb
->num_disks
; i
++) {
536 if (i
== super
->disks
->index
)
538 print_imsm_disk(mpb
, i
);
542 static void brief_examine_super_imsm(struct supertype
*st
)
544 printf("ARRAY /dev/imsm metadata=imsm\n");
547 static void detail_super_imsm(struct supertype
*st
, char *homehost
)
549 printf("%s\n", __FUNCTION__
);
552 static void brief_detail_super_imsm(struct supertype
*st
)
554 printf("%s\n", __FUNCTION__
);
558 static int match_home_imsm(struct supertype
*st
, char *homehost
)
560 printf("%s\n", __FUNCTION__
);
565 static void uuid_from_super_imsm(struct supertype
*st
, int uuid
[4])
567 printf("%s\n", __FUNCTION__
);
572 get_imsm_numerical_version(struct imsm_super
*mpb
, int *m
, int *p
)
574 __u8
*v
= get_imsm_version(mpb
);
575 __u8
*end
= mpb
->sig
+ MAX_SIGNATURE_LENGTH
;
576 char major
[] = { 0, 0, 0 };
577 char minor
[] = { 0 ,0, 0 };
578 char patch
[] = { 0, 0, 0 };
579 char *ver_parse
[] = { major
, minor
, patch
};
583 while (*v
!= '\0' && v
< end
) {
584 if (*v
!= '.' && j
< 2)
585 ver_parse
[i
][j
++] = *v
;
593 *m
= strtol(minor
, NULL
, 0);
594 *p
= strtol(patch
, NULL
, 0);
598 static int imsm_level_to_layout(int level
)
606 return ALGORITHM_LEFT_ASYMMETRIC
;
608 return 0x102; //FIXME is this correct?
613 static void getinfo_super_imsm_volume(struct supertype
*st
, struct mdinfo
*info
)
615 struct intel_super
*super
= st
->sb
;
616 struct imsm_dev
*dev
= get_imsm_dev(super
, super
->current_vol
);
617 struct imsm_map
*map
= get_imsm_map(dev
, 0);
619 info
->container_member
= super
->current_vol
;
620 info
->array
.raid_disks
= map
->num_members
;
621 info
->array
.level
= get_imsm_raid_level(map
);
622 info
->array
.layout
= imsm_level_to_layout(info
->array
.level
);
623 info
->array
.md_minor
= -1;
624 info
->array
.ctime
= 0;
625 info
->array
.utime
= 0;
626 info
->array
.chunk_size
= __le16_to_cpu(map
->blocks_per_strip
* 512);
628 info
->data_offset
= __le32_to_cpu(map
->pba_of_lba0
);
629 info
->component_size
= __le32_to_cpu(map
->blocks_per_member
);
631 info
->disk
.major
= 0;
632 info
->disk
.minor
= 0;
634 sprintf(info
->text_version
, "/%s/%d",
635 devnum2devname(st
->container_dev
),
636 info
->container_member
);
640 static void getinfo_super_imsm(struct supertype
*st
, struct mdinfo
*info
)
642 struct intel_super
*super
= st
->sb
;
643 struct imsm_disk
*disk
;
646 if (super
->current_vol
>= 0) {
647 getinfo_super_imsm_volume(st
, info
);
651 /* Set raid_disks to zero so that Assemble will always pull in valid
654 info
->array
.raid_disks
= 0;
655 info
->array
.level
= LEVEL_CONTAINER
;
656 info
->array
.layout
= 0;
657 info
->array
.md_minor
= -1;
658 info
->array
.ctime
= 0; /* N/A for imsm */
659 info
->array
.utime
= 0;
660 info
->array
.chunk_size
= 0;
662 info
->disk
.major
= 0;
663 info
->disk
.minor
= 0;
664 info
->disk
.raid_disk
= -1;
665 info
->reshape_active
= 0;
666 strcpy(info
->text_version
, "imsm");
667 info
->disk
.number
= -1;
668 info
->disk
.state
= 0;
671 disk
= &super
->disks
->disk
;
672 info
->disk
.number
= super
->disks
->index
;
673 info
->disk
.raid_disk
= super
->disks
->index
;
674 info
->data_offset
= __le32_to_cpu(disk
->total_blocks
) -
675 (MPB_SECTOR_CNT
+ IMSM_RESERVED_SECTORS
);
676 info
->component_size
= MPB_SECTOR_CNT
+ IMSM_RESERVED_SECTORS
;
677 s
= __le32_to_cpu(disk
->status
);
678 info
->disk
.state
= s
& CONFIGURED_DISK
? (1 << MD_DISK_ACTIVE
) : 0;
679 info
->disk
.state
|= s
& FAILED_DISK
? (1 << MD_DISK_FAULTY
) : 0;
680 info
->disk
.state
|= s
& USABLE_DISK
? (1 << MD_DISK_SYNC
) : 0;
684 static int update_super_imsm(struct supertype
*st
, struct mdinfo
*info
,
685 char *update
, char *devname
, int verbose
,
686 int uuid_set
, char *homehost
)
690 /* For 'assemble' and 'force' we need to return non-zero if any
691 * change was made. For others, the return value is ignored.
692 * Update options are:
693 * force-one : This device looks a bit old but needs to be included,
694 * update age info appropriately.
695 * assemble: clear any 'faulty' flag to allow this device to
697 * force-array: Array is degraded but being forced, mark it clean
698 * if that will be needed to assemble it.
700 * newdev: not used ????
701 * grow: Array has gained a new device - this is currently for
703 * resync: mark as dirty so a resync will happen.
704 * name: update the name - preserving the homehost
706 * Following are not relevant for this imsm:
707 * sparc2.2 : update from old dodgey metadata
708 * super-minor: change the preferred_minor number
709 * summaries: update redundant counters.
710 * uuid: Change the uuid of the array to match watch is given
711 * homehost: update the recorded homehost
712 * _reshape_progress: record new reshape_progress position.
715 //struct intel_super *super = st->sb;
716 //struct imsm_super *mpb = super->mpb;
718 if (strcmp(update
, "grow") == 0) {
720 if (strcmp(update
, "resync") == 0) {
721 /* dev->vol.dirty = 1; */
724 /* IMSM has no concept of UUID or homehost */
729 static size_t disks_to_mpb_size(int disks
)
733 size
= sizeof(struct imsm_super
);
734 size
+= (disks
- 1) * sizeof(struct imsm_disk
);
735 size
+= 2 * sizeof(struct imsm_dev
);
736 /* up to 2 maps per raid device (-2 for imsm_maps in imsm_dev */
737 size
+= (4 - 2) * sizeof(struct imsm_map
);
738 /* 4 possible disk_ord_tbl's */
739 size
+= 4 * (disks
- 1) * sizeof(__u32
);
744 static __u64
avail_size_imsm(struct supertype
*st
, __u64 devsize
)
746 if (devsize
< (MPB_SECTOR_CNT
+ IMSM_RESERVED_SECTORS
))
749 return devsize
- (MPB_SECTOR_CNT
+ IMSM_RESERVED_SECTORS
);
752 static int compare_super_imsm(struct supertype
*st
, struct supertype
*tst
)
756 * 0 same, or first was empty, and second was copied
757 * 1 second had wrong number
761 struct intel_super
*first
= st
->sb
;
762 struct intel_super
*sec
= tst
->sb
;
770 if (memcmp(first
->anchor
->sig
, sec
->anchor
->sig
, MAX_SIGNATURE_LENGTH
) != 0)
773 /* if an anchor does not have num_raid_devs set then it is a free
776 if (first
->anchor
->num_raid_devs
> 0 &&
777 sec
->anchor
->num_raid_devs
> 0) {
778 if (first
->anchor
->family_num
!= sec
->anchor
->family_num
)
780 if (first
->anchor
->mpb_size
!= sec
->anchor
->mpb_size
)
782 if (first
->anchor
->check_sum
!= sec
->anchor
->check_sum
)
789 static void fd2devname(int fd
, char *name
)
798 if (fstat(fd
, &st
) != 0)
800 sprintf(path
, "/sys/dev/block/%d:%d",
801 major(st
.st_rdev
), minor(st
.st_rdev
));
803 rv
= readlink(path
, dname
, sizeof(dname
));
808 nm
= strrchr(dname
, '/');
810 snprintf(name
, MAX_RAID_SERIAL_LEN
, "/dev/%s", nm
);
814 extern int scsi_get_serial(int fd
, void *buf
, size_t buf_len
);
816 static int imsm_read_serial(int fd
, char *devname
,
817 __u8 serial
[MAX_RAID_SERIAL_LEN
])
819 unsigned char scsi_serial
[255];
824 memset(scsi_serial
, 0, sizeof(scsi_serial
));
826 if (imsm_env_devname_as_serial()) {
827 char name
[MAX_RAID_SERIAL_LEN
];
829 fd2devname(fd
, name
);
830 strcpy((char *) serial
, name
);
834 rv
= scsi_get_serial(fd
, scsi_serial
, sizeof(scsi_serial
));
839 Name
": Failed to retrieve serial for %s\n",
844 rsp_len
= scsi_serial
[3];
845 for (i
= 0, cnt
= 0; i
< rsp_len
; i
++) {
846 if (!isspace(scsi_serial
[4 + i
]))
847 serial
[cnt
++] = scsi_serial
[4 + i
];
848 if (cnt
== MAX_RAID_SERIAL_LEN
)
852 serial
[MAX_RAID_SERIAL_LEN
- 1] = '\0';
858 load_imsm_disk(int fd
, struct intel_super
*super
, char *devname
, int keep_fd
)
865 __u8 serial
[MAX_RAID_SERIAL_LEN
];
867 rv
= imsm_read_serial(fd
, devname
, serial
);
872 /* check if this is a disk we have seen before. it may be a spare in
873 * super->disks while the current anchor believes it is a raid member,
874 * check if we need to update dl->index
876 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
877 if (memcmp(dl
->serial
, serial
, MAX_RAID_SERIAL_LEN
) == 0)
881 dl
= malloc(sizeof(*dl
));
888 Name
": failed to allocate disk buffer for %s\n",
895 dl
->major
= major(stb
.st_rdev
);
896 dl
->minor
= minor(stb
.st_rdev
);
897 dl
->next
= super
->disks
;
898 dl
->fd
= keep_fd
? fd
: -1;
899 dl
->devname
= devname
? strdup(devname
) : NULL
;
900 strncpy((char *) dl
->serial
, (char *) serial
, MAX_RAID_SERIAL_LEN
);
901 } else if (keep_fd
) {
906 /* look up this disk's index in the current anchor */
907 for (i
= 0; i
< super
->anchor
->num_disks
; i
++) {
908 struct imsm_disk
*disk_iter
;
910 disk_iter
= __get_imsm_disk(super
->anchor
, i
);
912 if (memcmp(disk_iter
->serial
, dl
->serial
,
913 MAX_RAID_SERIAL_LEN
) == 0) {
916 dl
->disk
= *disk_iter
;
917 status
= __le32_to_cpu(dl
->disk
.status
);
918 /* only set index on disks that are a member of a
919 * populated contianer, i.e. one with raid_devs
921 if (status
& SPARE_DISK
)
929 if (i
== super
->anchor
->num_disks
&& alloc
) {
932 Name
": failed to load disk with serial \'%s\' for %s\n",
933 dl
->serial
, devname
);
937 if (i
== super
->anchor
->num_disks
&& dl
->index
>= 0) {
940 Name
": confused... disk %d with serial \'%s\' "
941 "is not listed in the current anchor\n",
942 dl
->index
, dl
->serial
);
952 static void imsm_copy_dev(struct imsm_dev
*dest
, struct imsm_dev
*src
)
954 memcpy(dest
, src
, sizeof_imsm_dev(src
, 0));
957 static void dup_map(struct imsm_dev
*dev
)
959 struct imsm_map
*dest
= get_imsm_map(dev
, 1);
960 struct imsm_map
*src
= get_imsm_map(dev
, 0);
962 memcpy(dest
, src
, sizeof_imsm_map(src
));
965 static int parse_raid_devices(struct intel_super
*super
)
968 struct imsm_dev
*dev_new
;
971 for (i
= 0; i
< super
->anchor
->num_raid_devs
; i
++) {
972 struct imsm_dev
*dev_iter
= __get_imsm_dev(super
->anchor
, i
);
974 len
= sizeof_imsm_dev(dev_iter
, 1);
975 dev_new
= malloc(len
);
978 imsm_copy_dev(dev_new
, dev_iter
);
979 super
->dev_tbl
[i
] = dev_new
;
985 /* retrieve a pointer to the bbm log which starts after all raid devices */
986 struct bbm_log
*__get_imsm_bbm_log(struct imsm_super
*mpb
)
990 if (__le32_to_cpu(mpb
->bbm_log_size
)) {
992 ptr
+= mpb
->mpb_size
- __le32_to_cpu(mpb
->bbm_log_size
);
998 static void __free_imsm(struct intel_super
*super
, int free_disks
);
1000 /* load_imsm_mpb - read matrix metadata
1001 * allocates super->mpb to be freed by free_super
1003 static int load_imsm_mpb(int fd
, struct intel_super
*super
, char *devname
)
1005 unsigned long long dsize
;
1006 unsigned long long sectors
;
1008 struct imsm_super
*anchor
;
1012 get_dev_size(fd
, NULL
, &dsize
);
1014 if (lseek64(fd
, dsize
- (512 * 2), SEEK_SET
) < 0) {
1017 Name
": Cannot seek to anchor block on %s: %s\n",
1018 devname
, strerror(errno
));
1022 if (posix_memalign((void**)&anchor
, 512, 512) != 0) {
1025 Name
": Failed to allocate imsm anchor buffer"
1026 " on %s\n", devname
);
1029 if (read(fd
, anchor
, 512) != 512) {
1032 Name
": Cannot read anchor block on %s: %s\n",
1033 devname
, strerror(errno
));
1038 if (strncmp((char *) anchor
->sig
, MPB_SIGNATURE
, MPB_SIG_LEN
) != 0) {
1041 Name
": no IMSM anchor on %s\n", devname
);
1046 __free_imsm(super
, 0);
1047 super
->len
= __le32_to_cpu(anchor
->mpb_size
);
1048 super
->len
= ROUND_UP(anchor
->mpb_size
, 512);
1049 if (posix_memalign(&super
->buf
, 512, super
->len
) != 0) {
1052 Name
": unable to allocate %zu byte mpb buffer\n",
1057 memcpy(super
->buf
, anchor
, 512);
1059 sectors
= mpb_sectors(anchor
) - 1;
1062 rc
= load_imsm_disk(fd
, super
, devname
, 0);
1064 rc
= parse_raid_devices(super
);
1068 /* read the extended mpb */
1069 if (lseek64(fd
, dsize
- (512 * (2 + sectors
)), SEEK_SET
) < 0) {
1072 Name
": Cannot seek to extended mpb on %s: %s\n",
1073 devname
, strerror(errno
));
1077 if (read(fd
, super
->buf
+ 512, super
->len
- 512) != super
->len
- 512) {
1080 Name
": Cannot read extended mpb on %s: %s\n",
1081 devname
, strerror(errno
));
1085 check_sum
= __gen_imsm_checksum(super
->anchor
);
1086 if (check_sum
!= __le32_to_cpu(super
->anchor
->check_sum
)) {
1089 Name
": IMSM checksum %x != %x on %s\n",
1090 check_sum
, __le32_to_cpu(super
->anchor
->check_sum
),
1095 /* FIXME the BBM log is disk specific so we cannot use this global
1096 * buffer for all disks. Ok for now since we only look at the global
1097 * bbm_log_size parameter to gate assembly
1099 super
->bbm_log
= __get_imsm_bbm_log(super
->anchor
);
1101 rc
= load_imsm_disk(fd
, super
, devname
, 0);
1103 rc
= parse_raid_devices(super
);
1107 static void __free_imsm_disk(struct dl
*d
)
1116 static void free_imsm_disks(struct intel_super
*super
)
1118 while (super
->disks
) {
1119 struct dl
*d
= super
->disks
;
1121 super
->disks
= d
->next
;
1122 __free_imsm_disk(d
);
1126 /* free all the pieces hanging off of a super pointer */
1127 static void __free_imsm(struct intel_super
*super
, int free_disks
)
1136 free_imsm_disks(super
);
1137 for (i
= 0; i
< IMSM_MAX_RAID_DEVS
; i
++)
1138 if (super
->dev_tbl
[i
]) {
1139 free(super
->dev_tbl
[i
]);
1140 super
->dev_tbl
[i
] = NULL
;
1144 static void free_imsm(struct intel_super
*super
)
1146 __free_imsm(super
, 1);
1150 static void free_super_imsm(struct supertype
*st
)
1152 struct intel_super
*super
= st
->sb
;
1161 static struct intel_super
*alloc_super(int creating_imsm
)
1163 struct intel_super
*super
= malloc(sizeof(*super
));
1166 memset(super
, 0, sizeof(*super
));
1167 super
->creating_imsm
= creating_imsm
;
1168 super
->current_vol
= -1;
1175 static int load_super_imsm_all(struct supertype
*st
, int fd
, void **sbp
,
1176 char *devname
, int keep_fd
)
1179 struct intel_super
*super
;
1180 struct mdinfo
*sd
, *best
= NULL
;
1187 /* check if this disk is a member of an active array */
1188 sra
= sysfs_read(fd
, 0, GET_LEVEL
|GET_VERSION
|GET_DEVS
|GET_STATE
);
1192 if (sra
->array
.major_version
!= -1 ||
1193 sra
->array
.minor_version
!= -2 ||
1194 strcmp(sra
->text_version
, "imsm") != 0)
1197 super
= alloc_super(0);
1201 /* find the most up to date disk in this array, skipping spares */
1202 for (sd
= sra
->devs
; sd
; sd
= sd
->next
) {
1203 sprintf(nm
, "%d:%d", sd
->disk
.major
, sd
->disk
.minor
);
1204 dfd
= dev_open(nm
, keep_fd
? O_RDWR
: O_RDONLY
);
1209 rv
= load_imsm_mpb(dfd
, super
, NULL
);
1213 if (super
->anchor
->num_raid_devs
== 0)
1216 gen
= __le32_to_cpu(super
->anchor
->generation_num
);
1217 if (!best
|| gen
> bestgen
) {
1232 /* load the most up to date anchor */
1233 sprintf(nm
, "%d:%d", best
->disk
.major
, best
->disk
.minor
);
1234 dfd
= dev_open(nm
, O_RDONLY
);
1239 rv
= load_imsm_mpb(dfd
, super
, NULL
);
1246 /* re-parse the disk list with the current anchor */
1247 for (sd
= sra
->devs
; sd
; sd
= sd
->next
) {
1248 sprintf(nm
, "%d:%d", sd
->disk
.major
, sd
->disk
.minor
);
1249 dfd
= dev_open(nm
, keep_fd
? O_RDWR
: O_RDONLY
);
1254 load_imsm_disk(dfd
, super
, NULL
, keep_fd
);
1259 if (st
->subarray
[0]) {
1260 if (atoi(st
->subarray
) <= super
->anchor
->num_raid_devs
)
1261 super
->current_vol
= atoi(st
->subarray
);
1267 if (st
->ss
== NULL
) {
1268 st
->ss
= &super_imsm
;
1269 st
->minor_version
= 0;
1270 st
->max_devs
= IMSM_MAX_DEVICES
;
1271 st
->container_dev
= fd2devnum(fd
);
1278 static int load_super_imsm(struct supertype
*st
, int fd
, char *devname
)
1280 struct intel_super
*super
;
1284 if (load_super_imsm_all(st
, fd
, &st
->sb
, devname
, 1) == 0)
1287 if (st
->subarray
[0])
1288 return 1; /* FIXME */
1290 super
= alloc_super(0);
1293 Name
": malloc of %zu failed.\n",
1298 rv
= load_imsm_mpb(fd
, super
, devname
);
1303 Name
": Failed to load all information "
1304 "sections on %s\n", devname
);
1310 if (st
->ss
== NULL
) {
1311 st
->ss
= &super_imsm
;
1312 st
->minor_version
= 0;
1313 st
->max_devs
= IMSM_MAX_DEVICES
;
1319 static __u16
info_to_blocks_per_strip(mdu_array_info_t
*info
)
1321 if (info
->level
== 1)
1323 return info
->chunk_size
>> 9;
1326 static __u32
info_to_num_data_stripes(mdu_array_info_t
*info
)
1330 num_stripes
= (info
->size
* 2) / info_to_blocks_per_strip(info
);
1331 if (info
->level
== 1)
1337 static __u32
info_to_blocks_per_member(mdu_array_info_t
*info
)
1339 return (info
->size
* 2) & ~(info_to_blocks_per_strip(info
) - 1);
1342 static int init_super_imsm_volume(struct supertype
*st
, mdu_array_info_t
*info
,
1343 unsigned long long size
, char *name
,
1344 char *homehost
, int *uuid
)
1346 /* We are creating a volume inside a pre-existing container.
1347 * so st->sb is already set.
1349 struct intel_super
*super
= st
->sb
;
1350 struct imsm_super
*mpb
= super
->anchor
;
1351 struct imsm_dev
*dev
;
1352 struct imsm_vol
*vol
;
1353 struct imsm_map
*map
;
1354 int idx
= mpb
->num_raid_devs
;
1356 unsigned long long array_blocks
;
1358 size_t size_old
, size_new
;
1360 if (mpb
->num_raid_devs
>= 2) {
1361 fprintf(stderr
, Name
": This imsm-container already has the "
1362 "maximum of 2 volumes\n");
1366 /* ensure the mpb is large enough for the new data */
1367 size_old
= __le32_to_cpu(mpb
->mpb_size
);
1368 size_new
= disks_to_mpb_size(info
->nr_disks
);
1369 if (size_new
> size_old
) {
1371 size_t size_round
= ROUND_UP(size_new
, 512);
1373 if (posix_memalign(&mpb_new
, 512, size_round
) != 0) {
1374 fprintf(stderr
, Name
": could not allocate new mpb\n");
1377 memcpy(mpb_new
, mpb
, size_old
);
1380 super
->anchor
= mpb_new
;
1381 mpb
->mpb_size
= __cpu_to_le32(size_new
);
1382 memset(mpb_new
+ size_old
, 0, size_round
- size_old
);
1384 super
->current_vol
= idx
;
1385 /* when creating the first raid device in this container set num_disks
1386 * to zero, i.e. delete this spare and add raid member devices in
1387 * add_to_super_imsm_volume()
1389 if (super
->current_vol
== 0)
1391 sprintf(st
->subarray
, "%d", idx
);
1392 dev
= malloc(sizeof(*dev
) + sizeof(__u32
) * (info
->raid_disks
- 1));
1394 fprintf(stderr
, Name
": could not allocate raid device\n");
1397 strncpy((char *) dev
->volume
, name
, MAX_RAID_SERIAL_LEN
);
1398 array_blocks
= calc_array_size(info
->level
, info
->raid_disks
,
1399 info
->layout
, info
->chunk_size
,
1401 dev
->size_low
= __cpu_to_le32((__u32
) array_blocks
);
1402 dev
->size_high
= __cpu_to_le32((__u32
) (array_blocks
>> 32));
1403 dev
->status
= __cpu_to_le32(0);
1404 dev
->reserved_blocks
= __cpu_to_le32(0);
1406 vol
->migr_state
= 0;
1409 for (i
= 0; i
< idx
; i
++) {
1410 struct imsm_dev
*prev
= get_imsm_dev(super
, i
);
1411 struct imsm_map
*pmap
= get_imsm_map(prev
, 0);
1413 offset
+= __le32_to_cpu(pmap
->blocks_per_member
);
1414 offset
+= IMSM_RESERVED_SECTORS
;
1416 map
= get_imsm_map(dev
, 0);
1417 map
->pba_of_lba0
= __cpu_to_le32(offset
);
1418 map
->blocks_per_member
= __cpu_to_le32(info_to_blocks_per_member(info
));
1419 map
->blocks_per_strip
= __cpu_to_le16(info_to_blocks_per_strip(info
));
1420 map
->num_data_stripes
= __cpu_to_le32(info_to_num_data_stripes(info
));
1421 map
->map_state
= info
->level
? IMSM_T_STATE_UNINITIALIZED
:
1422 IMSM_T_STATE_NORMAL
;
1424 if (info
->level
== 1 && info
->raid_disks
> 2) {
1425 fprintf(stderr
, Name
": imsm does not support more than 2 disks"
1426 "in a raid1 volume\n");
1429 if (info
->level
== 10)
1430 map
->raid_level
= 1;
1432 map
->raid_level
= info
->level
;
1434 map
->num_members
= info
->raid_disks
;
1435 for (i
= 0; i
< map
->num_members
; i
++) {
1436 /* initialized in add_to_super */
1437 map
->disk_ord_tbl
[i
] = __cpu_to_le32(0);
1439 mpb
->num_raid_devs
++;
1440 super
->dev_tbl
[super
->current_vol
] = dev
;
1445 static int init_super_imsm(struct supertype
*st
, mdu_array_info_t
*info
,
1446 unsigned long long size
, char *name
,
1447 char *homehost
, int *uuid
)
1449 /* This is primarily called by Create when creating a new array.
1450 * We will then get add_to_super called for each component, and then
1451 * write_init_super called to write it out to each device.
1452 * For IMSM, Create can create on fresh devices or on a pre-existing
1454 * To create on a pre-existing array a different method will be called.
1455 * This one is just for fresh drives.
1457 struct intel_super
*super
;
1458 struct imsm_super
*mpb
;
1466 return init_super_imsm_volume(st
, info
, size
, name
, homehost
,
1469 super
= alloc_super(1);
1472 mpb_size
= disks_to_mpb_size(info
->nr_disks
);
1473 if (posix_memalign(&super
->buf
, 512, mpb_size
) != 0) {
1478 memset(mpb
, 0, mpb_size
);
1480 memcpy(mpb
->sig
, MPB_SIGNATURE
, strlen(MPB_SIGNATURE
));
1481 memcpy(mpb
->sig
+ strlen(MPB_SIGNATURE
), MPB_VERSION_RAID5
,
1482 strlen(MPB_VERSION_RAID5
));
1483 mpb
->mpb_size
= mpb_size
;
1489 static void add_to_super_imsm_volume(struct supertype
*st
, mdu_disk_info_t
*dk
,
1490 int fd
, char *devname
)
1492 struct intel_super
*super
= st
->sb
;
1493 struct imsm_super
*mpb
= super
->anchor
;
1495 struct imsm_dev
*dev
;
1496 struct imsm_map
*map
;
1499 dev
= get_imsm_dev(super
, super
->current_vol
);
1500 map
= get_imsm_map(dev
, 0);
1502 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
1503 if (dl
->major
== dk
->major
&&
1504 dl
->minor
== dk
->minor
)
1507 if (!dl
|| ! (dk
->state
& (1<<MD_DISK_SYNC
)))
1510 /* add a pristine spare to the metadata */
1511 if (dl
->index
< 0) {
1512 dl
->index
= super
->anchor
->num_disks
;
1513 super
->anchor
->num_disks
++;
1515 map
->disk_ord_tbl
[dk
->number
] = __cpu_to_le32(dl
->index
);
1516 status
= CONFIGURED_DISK
| USABLE_DISK
;
1517 dl
->disk
.status
= __cpu_to_le32(status
);
1519 /* if we are creating the first raid device update the family number */
1520 if (super
->current_vol
== 0) {
1522 struct imsm_dev
*_dev
= __get_imsm_dev(mpb
, 0);
1523 struct imsm_disk
*_disk
= __get_imsm_disk(mpb
, dl
->index
);
1527 sum
= __gen_imsm_checksum(mpb
);
1528 mpb
->family_num
= __cpu_to_le32(sum
);
1532 static void add_to_super_imsm(struct supertype
*st
, mdu_disk_info_t
*dk
,
1533 int fd
, char *devname
)
1535 struct intel_super
*super
= st
->sb
;
1537 unsigned long long size
;
1542 if (super
->current_vol
>= 0) {
1543 add_to_super_imsm_volume(st
, dk
, fd
, devname
);
1548 dd
= malloc(sizeof(*dd
));
1551 Name
": malloc failed %s:%d.\n", __func__
, __LINE__
);
1554 memset(dd
, 0, sizeof(*dd
));
1555 dd
->major
= major(stb
.st_rdev
);
1556 dd
->minor
= minor(stb
.st_rdev
);
1558 dd
->devname
= devname
? strdup(devname
) : NULL
;
1559 dd
->next
= super
->disks
;
1561 rv
= imsm_read_serial(fd
, devname
, dd
->serial
);
1564 Name
": failed to retrieve scsi serial, aborting\n");
1569 get_dev_size(fd
, NULL
, &size
);
1571 status
= USABLE_DISK
| SPARE_DISK
;
1572 strcpy((char *) dd
->disk
.serial
, (char *) dd
->serial
);
1573 dd
->disk
.total_blocks
= __cpu_to_le32(size
);
1574 dd
->disk
.status
= __cpu_to_le32(status
);
1575 if (sysfs_disk_to_scsi_id(fd
, &id
) == 0)
1576 dd
->disk
.scsi_id
= __cpu_to_le32(id
);
1578 dd
->disk
.scsi_id
= __cpu_to_le32(0);
1582 static int store_imsm_mpb(int fd
, struct intel_super
*super
);
1584 /* spare records have their own family number and do not have any defined raid
1587 static int write_super_imsm_spares(struct intel_super
*super
, int doclose
)
1589 struct imsm_super mpb_save
;
1590 struct imsm_super
*mpb
= super
->anchor
;
1595 mpb
->num_raid_devs
= 0;
1597 mpb
->mpb_size
= sizeof(struct imsm_super
);
1598 mpb
->generation_num
= __cpu_to_le32(1UL);
1600 for (d
= super
->disks
; d
; d
= d
->next
) {
1604 mpb
->disk
[0] = d
->disk
;
1605 sum
= __gen_imsm_checksum(mpb
);
1606 mpb
->family_num
= __cpu_to_le32(sum
);
1607 sum
= __gen_imsm_checksum(mpb
);
1608 mpb
->check_sum
= __cpu_to_le32(sum
);
1610 if (store_imsm_mpb(d
->fd
, super
)) {
1611 fprintf(stderr
, "%s: failed for device %d:%d %s\n",
1612 __func__
, d
->major
, d
->minor
, strerror(errno
));
1626 static int write_super_imsm(struct intel_super
*super
, int doclose
)
1628 struct imsm_super
*mpb
= super
->anchor
;
1635 __u32 mpb_size
= sizeof(struct imsm_super
) - sizeof(struct imsm_disk
);
1637 /* 'generation' is incremented everytime the metadata is written */
1638 generation
= __le32_to_cpu(mpb
->generation_num
);
1640 mpb
->generation_num
= __cpu_to_le32(generation
);
1642 for (d
= super
->disks
; d
; d
= d
->next
) {
1647 mpb
->disk
[d
->index
] = d
->disk
;
1648 mpb_size
+= sizeof(struct imsm_disk
);
1651 if (raid_disks
!= mpb
->num_disks
) {
1652 fprintf(stderr
, "%s: expected %d disks only found %d\n",
1653 __func__
, mpb
->num_disks
, raid_disks
);
1657 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
1658 struct imsm_dev
*dev
= __get_imsm_dev(mpb
, i
);
1660 imsm_copy_dev(dev
, super
->dev_tbl
[i
]);
1661 mpb_size
+= sizeof_imsm_dev(dev
, 0);
1663 mpb_size
+= __le32_to_cpu(mpb
->bbm_log_size
);
1664 mpb
->mpb_size
= __cpu_to_le32(mpb_size
);
1666 /* recalculate checksum */
1667 sum
= __gen_imsm_checksum(mpb
);
1668 mpb
->check_sum
= __cpu_to_le32(sum
);
1670 /* write the mpb for disks that compose raid devices */
1671 for (d
= super
->disks
; d
; d
= d
->next
) {
1674 if (store_imsm_mpb(d
->fd
, super
)) {
1675 fprintf(stderr
, "%s: failed for device %d:%d %s\n",
1676 __func__
, d
->major
, d
->minor
, strerror(errno
));
1686 return write_super_imsm_spares(super
, doclose
);
1691 static int write_init_super_imsm(struct supertype
*st
)
1693 if (st
->update_tail
) {
1694 /* queue the recently created array as a metadata update */
1696 struct imsm_update_create_array
*u
;
1697 struct intel_super
*super
= st
->sb
;
1698 struct imsm_dev
*dev
;
1701 if (super
->current_vol
< 0 ||
1702 !(dev
= get_imsm_dev(super
, super
->current_vol
))) {
1703 fprintf(stderr
, "%s: could not determine sub-array\n",
1709 len
= sizeof(*u
) - sizeof(*dev
) + sizeof_imsm_dev(dev
, 0);
1712 fprintf(stderr
, "%s: failed to allocate update buffer\n",
1717 u
->type
= update_create_array
;
1718 u
->dev_idx
= super
->current_vol
;
1719 imsm_copy_dev(&u
->dev
, dev
);
1720 append_metadata_update(st
, u
, len
);
1722 for (d
= super
->disks
; d
; d
= d
->next
) {
1729 return write_super_imsm(st
->sb
, 1);
1732 static int store_zero_imsm(struct supertype
*st
, int fd
)
1734 unsigned long long dsize
;
1737 get_dev_size(fd
, NULL
, &dsize
);
1739 /* first block is stored on second to last sector of the disk */
1740 if (lseek64(fd
, dsize
- (512 * 2), SEEK_SET
) < 0)
1743 if (posix_memalign(&buf
, 512, 512) != 0)
1746 memset(buf
, 0, 512);
1747 if (write(fd
, buf
, 512) != 512)
1752 static int validate_geometry_imsm_container(struct supertype
*st
, int level
,
1753 int layout
, int raiddisks
, int chunk
,
1754 unsigned long long size
, char *dev
,
1755 unsigned long long *freesize
,
1759 unsigned long long ldsize
;
1761 if (level
!= LEVEL_CONTAINER
)
1766 fd
= open(dev
, O_RDONLY
|O_EXCL
, 0);
1769 fprintf(stderr
, Name
": imsm: Cannot open %s: %s\n",
1770 dev
, strerror(errno
));
1773 if (!get_dev_size(fd
, dev
, &ldsize
)) {
1779 *freesize
= avail_size_imsm(st
, ldsize
>> 9);
1784 /* validate_geometry_imsm_volume - lifted from validate_geometry_ddf_bvd
1785 * FIX ME add ahci details
1787 static int validate_geometry_imsm_volume(struct supertype
*st
, int level
,
1788 int layout
, int raiddisks
, int chunk
,
1789 unsigned long long size
, char *dev
,
1790 unsigned long long *freesize
,
1794 struct intel_super
*super
= st
->sb
;
1796 unsigned long long pos
= 0;
1797 unsigned long long maxsize
;
1801 if (level
== LEVEL_CONTAINER
)
1804 if (level
== 1 && raiddisks
> 2) {
1806 fprintf(stderr
, Name
": imsm does not support more "
1807 "than 2 in a raid1 configuration\n");
1811 /* We must have the container info already read in. */
1816 /* General test: make sure there is space for
1817 * 'raiddisks' device extents of size 'size' at a given
1820 unsigned long long minsize
= size
*2 /* convert to blocks */;
1821 unsigned long long start_offset
= ~0ULL;
1824 minsize
= MPB_SECTOR_CNT
+ IMSM_RESERVED_SECTORS
;
1825 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
1830 e
= get_extents(super
, dl
);
1833 unsigned long long esize
;
1834 esize
= e
[i
].start
- pos
;
1835 if (esize
>= minsize
)
1837 if (found
&& start_offset
== ~0ULL) {
1840 } else if (found
&& pos
!= start_offset
) {
1844 pos
= e
[i
].start
+ e
[i
].size
;
1846 } while (e
[i
-1].size
);
1851 if (dcnt
< raiddisks
) {
1853 fprintf(stderr
, Name
": imsm: Not enough "
1854 "devices with space for this array "
1861 /* This device must be a member of the set */
1862 if (stat(dev
, &stb
) < 0)
1864 if ((S_IFMT
& stb
.st_mode
) != S_IFBLK
)
1866 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
1867 if (dl
->major
== major(stb
.st_rdev
) &&
1868 dl
->minor
== minor(stb
.st_rdev
))
1873 fprintf(stderr
, Name
": %s is not in the "
1874 "same imsm set\n", dev
);
1877 e
= get_extents(super
, dl
);
1881 unsigned long long esize
;
1882 esize
= e
[i
].start
- pos
;
1883 if (esize
>= maxsize
)
1885 pos
= e
[i
].start
+ e
[i
].size
;
1887 } while (e
[i
-1].size
);
1888 *freesize
= maxsize
;
1893 int imsm_bbm_log_size(struct imsm_super
*mpb
)
1895 return __le32_to_cpu(mpb
->bbm_log_size
);
1898 static int validate_geometry_imsm(struct supertype
*st
, int level
, int layout
,
1899 int raiddisks
, int chunk
, unsigned long long size
,
1900 char *dev
, unsigned long long *freesize
,
1906 /* if given unused devices create a container
1907 * if given given devices in a container create a member volume
1909 if (level
== LEVEL_CONTAINER
) {
1910 /* Must be a fresh device to add to a container */
1911 return validate_geometry_imsm_container(st
, level
, layout
,
1912 raiddisks
, chunk
, size
,
1918 /* creating in a given container */
1919 return validate_geometry_imsm_volume(st
, level
, layout
,
1920 raiddisks
, chunk
, size
,
1921 dev
, freesize
, verbose
);
1924 /* limit creation to the following levels */
1936 /* This device needs to be a device in an 'imsm' container */
1937 fd
= open(dev
, O_RDONLY
|O_EXCL
, 0);
1941 Name
": Cannot create this array on device %s\n",
1946 if (errno
!= EBUSY
|| (fd
= open(dev
, O_RDONLY
, 0)) < 0) {
1948 fprintf(stderr
, Name
": Cannot open %s: %s\n",
1949 dev
, strerror(errno
));
1952 /* Well, it is in use by someone, maybe an 'imsm' container. */
1953 cfd
= open_container(fd
);
1957 fprintf(stderr
, Name
": Cannot use %s: It is busy\n",
1961 sra
= sysfs_read(cfd
, 0, GET_VERSION
);
1963 if (sra
&& sra
->array
.major_version
== -1 &&
1964 strcmp(sra
->text_version
, "imsm") == 0) {
1965 /* This is a member of a imsm container. Load the container
1966 * and try to create a volume
1968 struct intel_super
*super
;
1970 if (load_super_imsm_all(st
, cfd
, (void **) &super
, NULL
, 1) == 0) {
1972 st
->container_dev
= fd2devnum(cfd
);
1974 return validate_geometry_imsm_volume(st
, level
, layout
,
1980 } else /* may belong to another container */
1986 static struct mdinfo
*container_content_imsm(struct supertype
*st
)
1988 /* Given a container loaded by load_super_imsm_all,
1989 * extract information about all the arrays into
1992 * For each imsm_dev create an mdinfo, fill it in,
1993 * then look for matching devices in super->disks
1994 * and create appropriate device mdinfo.
1996 struct intel_super
*super
= st
->sb
;
1997 struct imsm_super
*mpb
= super
->anchor
;
1998 struct mdinfo
*rest
= NULL
;
2001 /* do not assemble arrays that might have bad blocks */
2002 if (imsm_bbm_log_size(super
->anchor
)) {
2003 fprintf(stderr
, Name
": BBM log found in metadata. "
2004 "Cannot activate array(s).\n");
2008 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
2009 struct imsm_dev
*dev
= get_imsm_dev(super
, i
);
2010 struct imsm_vol
*vol
= &dev
->vol
;
2011 struct imsm_map
*map
= get_imsm_map(dev
, 0);
2012 struct mdinfo
*this;
2015 this = malloc(sizeof(*this));
2016 memset(this, 0, sizeof(*this));
2019 this->array
.level
= get_imsm_raid_level(map
);
2020 this->array
.raid_disks
= map
->num_members
;
2021 this->array
.layout
= imsm_level_to_layout(this->array
.level
);
2022 this->array
.md_minor
= -1;
2023 this->array
.ctime
= 0;
2024 this->array
.utime
= 0;
2025 this->array
.chunk_size
= __le16_to_cpu(map
->blocks_per_strip
) << 9;
2026 this->array
.state
= !vol
->dirty
;
2027 this->container_member
= i
;
2028 if (map
->map_state
== IMSM_T_STATE_UNINITIALIZED
||
2029 dev
->vol
.dirty
|| dev
->vol
.migr_state
)
2030 this->resync_start
= 0;
2032 this->resync_start
= ~0ULL;
2034 strncpy(this->name
, (char *) dev
->volume
, MAX_RAID_SERIAL_LEN
);
2035 this->name
[MAX_RAID_SERIAL_LEN
] = 0;
2037 sprintf(this->text_version
, "/%s/%d",
2038 devnum2devname(st
->container_dev
),
2039 this->container_member
);
2041 memset(this->uuid
, 0, sizeof(this->uuid
));
2043 this->component_size
= __le32_to_cpu(map
->blocks_per_member
);
2045 for (slot
= 0 ; slot
< map
->num_members
; slot
++) {
2046 struct mdinfo
*info_d
;
2054 idx
= get_imsm_disk_idx(map
, slot
);
2055 ord
= get_imsm_ord_tbl_ent(dev
, slot
);
2056 for (d
= super
->disks
; d
; d
= d
->next
)
2057 if (d
->index
== idx
)
2063 s
= d
? __le32_to_cpu(d
->disk
.status
) : 0;
2064 if (s
& FAILED_DISK
)
2066 if (!(s
& USABLE_DISK
))
2068 if (ord
& IMSM_ORD_REBUILD
)
2072 * if we skip some disks the array will be assmebled degraded;
2073 * reset resync start to avoid a dirty-degraded situation
2075 * FIXME handle dirty degraded
2077 if (skip
&& !dev
->vol
.dirty
)
2078 this->resync_start
= ~0ULL;
2082 info_d
= malloc(sizeof(*info_d
));
2084 fprintf(stderr
, Name
": failed to allocate disk"
2085 " for volume %s\n", (char *) dev
->volume
);
2090 memset(info_d
, 0, sizeof(*info_d
));
2091 info_d
->next
= this->devs
;
2092 this->devs
= info_d
;
2094 info_d
->disk
.number
= d
->index
;
2095 info_d
->disk
.major
= d
->major
;
2096 info_d
->disk
.minor
= d
->minor
;
2097 info_d
->disk
.raid_disk
= slot
;
2099 this->array
.working_disks
++;
2101 info_d
->events
= __le32_to_cpu(mpb
->generation_num
);
2102 info_d
->data_offset
= __le32_to_cpu(map
->pba_of_lba0
);
2103 info_d
->component_size
= __le32_to_cpu(map
->blocks_per_member
);
2105 strcpy(info_d
->name
, d
->devname
);
2114 static int imsm_open_new(struct supertype
*c
, struct active_array
*a
,
2117 struct intel_super
*super
= c
->sb
;
2118 struct imsm_super
*mpb
= super
->anchor
;
2120 if (atoi(inst
) >= mpb
->num_raid_devs
) {
2121 fprintf(stderr
, "%s: subarry index %d, out of range\n",
2122 __func__
, atoi(inst
));
2126 dprintf("imsm: open_new %s\n", inst
);
2127 a
->info
.container_member
= atoi(inst
);
2131 static __u8
imsm_check_degraded(struct intel_super
*super
, int n
, int failed
)
2133 struct imsm_dev
*dev
= get_imsm_dev(super
, n
);
2134 struct imsm_map
*map
= get_imsm_map(dev
, 0);
2137 return map
->map_state
== IMSM_T_STATE_UNINITIALIZED
?
2138 IMSM_T_STATE_UNINITIALIZED
: IMSM_T_STATE_NORMAL
;
2140 switch (get_imsm_raid_level(map
)) {
2142 return IMSM_T_STATE_FAILED
;
2145 if (failed
< map
->num_members
)
2146 return IMSM_T_STATE_DEGRADED
;
2148 return IMSM_T_STATE_FAILED
;
2153 * check to see if any mirrors have failed,
2154 * otherwise we are degraded
2156 int device_per_mirror
= 2; /* FIXME is this always the case?
2157 * and are they always adjacent?
2162 for (i
= 0; i
< map
->num_members
; i
++) {
2163 int idx
= get_imsm_disk_idx(map
, i
);
2164 struct imsm_disk
*disk
= get_imsm_disk(super
, idx
);
2166 if (__le32_to_cpu(disk
->status
) & FAILED_DISK
)
2169 if (failed
>= device_per_mirror
)
2170 return IMSM_T_STATE_FAILED
;
2172 /* reset 'failed' for next mirror set */
2173 if (!((i
+ 1) % device_per_mirror
))
2177 return IMSM_T_STATE_DEGRADED
;
2181 return IMSM_T_STATE_DEGRADED
;
2183 return IMSM_T_STATE_FAILED
;
2189 return map
->map_state
;
2192 static int imsm_count_failed(struct intel_super
*super
, struct imsm_map
*map
)
2196 struct imsm_disk
*disk
;
2198 for (i
= 0; i
< map
->num_members
; i
++) {
2199 int idx
= get_imsm_disk_idx(map
, i
);
2201 disk
= get_imsm_disk(super
, idx
);
2202 if (__le32_to_cpu(disk
->status
) & FAILED_DISK
)
2204 else if (!(__le32_to_cpu(disk
->status
) & USABLE_DISK
))
2211 static int imsm_set_array_state(struct active_array
*a
, int consistent
)
2213 int inst
= a
->info
.container_member
;
2214 struct intel_super
*super
= a
->container
->sb
;
2215 struct imsm_dev
*dev
= get_imsm_dev(super
, inst
);
2216 struct imsm_map
*map
= get_imsm_map(dev
, 0);
2217 int dirty
= !consistent
;
2221 failed
= imsm_count_failed(super
, map
);
2222 map_state
= imsm_check_degraded(super
, inst
, failed
);
2224 if (consistent
&& !dev
->vol
.dirty
&&
2225 (dev
->vol
.migr_state
|| map_state
!= IMSM_T_STATE_NORMAL
))
2226 a
->resync_start
= 0ULL;
2227 if (consistent
== 2 && a
->resync_start
!= ~0ULL)
2230 if (a
->resync_start
== ~0ULL) {
2231 /* complete recovery or initial resync */
2232 if (map
->map_state
!= map_state
) {
2233 dprintf("imsm: map_state %d: %d\n",
2235 map
->map_state
= map_state
;
2236 super
->updates_pending
++;
2238 if (dev
->vol
.migr_state
) {
2239 dprintf("imsm: mark resync complete\n");
2240 dev
->vol
.migr_state
= 0;
2241 dev
->vol
.migr_type
= 0;
2242 super
->updates_pending
++;
2244 } else if (!dev
->vol
.migr_state
) {
2245 dprintf("imsm: mark '%s' (%llu)\n",
2246 failed
? "rebuild" : "initializing", a
->resync_start
);
2247 /* mark that we are rebuilding */
2248 map
->map_state
= failed
? map_state
: IMSM_T_STATE_NORMAL
;
2249 dev
->vol
.migr_state
= 1;
2250 dev
->vol
.migr_type
= failed
? 1 : 0;
2252 a
->check_degraded
= 1;
2253 super
->updates_pending
++;
2256 /* mark dirty / clean */
2257 if (dirty
!= dev
->vol
.dirty
) {
2258 dprintf("imsm: mark '%s' (%llu)\n",
2259 dirty
? "dirty" : "clean", a
->resync_start
);
2260 dev
->vol
.dirty
= dirty
;
2261 super
->updates_pending
++;
2266 static void imsm_set_disk(struct active_array
*a
, int n
, int state
)
2268 int inst
= a
->info
.container_member
;
2269 struct intel_super
*super
= a
->container
->sb
;
2270 struct imsm_dev
*dev
= get_imsm_dev(super
, inst
);
2271 struct imsm_map
*map
= get_imsm_map(dev
, 0);
2272 struct imsm_disk
*disk
;
2275 int new_failure
= 0;
2277 if (n
> map
->num_members
)
2278 fprintf(stderr
, "imsm: set_disk %d out of range 0..%d\n",
2279 n
, map
->num_members
- 1);
2284 dprintf("imsm: set_disk %d:%x\n", n
, state
);
2286 disk
= get_imsm_disk(super
, get_imsm_disk_idx(map
, n
));
2288 /* check for new failures */
2289 status
= __le32_to_cpu(disk
->status
);
2290 if ((state
& DS_FAULTY
) && !(status
& FAILED_DISK
)) {
2291 status
|= FAILED_DISK
;
2292 disk
->status
= __cpu_to_le32(status
);
2294 super
->updates_pending
++;
2296 /* check if in_sync */
2297 if ((state
& DS_INSYNC
) && !(status
& USABLE_DISK
)) {
2298 status
|= USABLE_DISK
;
2299 disk
->status
= __cpu_to_le32(status
);
2300 super
->updates_pending
++;
2303 /* the number of failures have changed, count up 'failed' to determine
2304 * degraded / failed status
2306 if (new_failure
&& map
->map_state
!= IMSM_T_STATE_FAILED
)
2307 failed
= imsm_count_failed(super
, map
);
2309 /* determine map_state based on failed or in_sync count */
2311 map
->map_state
= imsm_check_degraded(super
, inst
, failed
);
2312 else if (map
->map_state
== IMSM_T_STATE_DEGRADED
) {
2316 for (d
= a
->info
.devs
; d
; d
= d
->next
)
2317 if (d
->curr_state
& DS_INSYNC
)
2320 if (working
== a
->info
.array
.raid_disks
) {
2321 map
->map_state
= IMSM_T_STATE_NORMAL
;
2322 dev
->vol
.migr_state
= 0;
2323 dev
->vol
.migr_type
= 0;
2324 super
->updates_pending
++;
2329 static int store_imsm_mpb(int fd
, struct intel_super
*super
)
2331 struct imsm_super
*mpb
= super
->anchor
;
2332 __u32 mpb_size
= __le32_to_cpu(mpb
->mpb_size
);
2333 unsigned long long dsize
;
2334 unsigned long long sectors
;
2336 get_dev_size(fd
, NULL
, &dsize
);
2338 if (mpb_size
> 512) {
2339 /* -1 to account for anchor */
2340 sectors
= mpb_sectors(mpb
) - 1;
2342 /* write the extended mpb to the sectors preceeding the anchor */
2343 if (lseek64(fd
, dsize
- (512 * (2 + sectors
)), SEEK_SET
) < 0)
2346 if (write(fd
, super
->buf
+ 512, 512 * sectors
) != 512 * sectors
)
2350 /* first block is stored on second to last sector of the disk */
2351 if (lseek64(fd
, dsize
- (512 * 2), SEEK_SET
) < 0)
2354 if (write(fd
, super
->buf
, 512) != 512)
2360 static void imsm_sync_metadata(struct supertype
*container
)
2362 struct intel_super
*super
= container
->sb
;
2364 if (!super
->updates_pending
)
2367 write_super_imsm(super
, 0);
2369 super
->updates_pending
= 0;
2372 static struct dl
*imsm_readd(struct intel_super
*super
, int idx
, struct active_array
*a
)
2374 struct imsm_dev
*dev
= get_imsm_dev(super
, a
->info
.container_member
);
2375 struct imsm_map
*map
= get_imsm_map(dev
, 0);
2376 int i
= get_imsm_disk_idx(map
, idx
);
2379 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
2383 if (__le32_to_cpu(dl
->disk
.status
) & FAILED_DISK
)
2387 dprintf("%s: found %x:%x\n", __func__
, dl
->major
, dl
->minor
);
2392 static struct dl
*imsm_add_spare(struct intel_super
*super
, int idx
, struct active_array
*a
)
2394 struct imsm_dev
*dev
= get_imsm_dev(super
, a
->info
.container_member
);
2395 struct imsm_map
*map
= get_imsm_map(dev
, 0);
2396 unsigned long long esize
;
2397 unsigned long long pos
;
2406 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
2407 /* If in this array, skip */
2408 for (d
= a
->info
.devs
; d
; d
= d
->next
)
2409 if (d
->disk
.major
== dl
->major
&&
2410 d
->disk
.minor
== dl
->minor
) {
2411 dprintf("%x:%x already in array\n", dl
->major
, dl
->minor
);
2417 /* skip marked in use or failed drives */
2418 status
= __le32_to_cpu(dl
->disk
.status
);
2419 if (status
& FAILED_DISK
|| status
& CONFIGURED_DISK
) {
2420 dprintf("%x:%x status ( %s%s)\n",
2421 dl
->major
, dl
->minor
,
2422 status
& FAILED_DISK
? "failed " : "",
2423 status
& CONFIGURED_DISK
? "configured " : "");
2427 /* Does this unused device have the requisite free space?
2428 * We need a->info.component_size sectors
2430 ex
= get_extents(super
, dl
);
2432 dprintf("cannot get extents\n");
2438 array_start
= __le32_to_cpu(map
->pba_of_lba0
);
2441 /* check that we can start at pba_of_lba0 with
2442 * a->info.component_size of space
2444 esize
= ex
[j
].start
- pos
;
2445 if (array_start
>= pos
&&
2446 array_start
+ a
->info
.component_size
< ex
[j
].start
) {
2450 pos
= ex
[j
].start
+ ex
[j
].size
;
2453 } while (ex
[j
-1].size
);
2457 dprintf("%x:%x does not have %llu at %d\n",
2458 dl
->major
, dl
->minor
,
2459 a
->info
.component_size
,
2460 __le32_to_cpu(map
->pba_of_lba0
));
2470 static struct mdinfo
*imsm_activate_spare(struct active_array
*a
,
2471 struct metadata_update
**updates
)
2474 * Find a device with unused free space and use it to replace a
2475 * failed/vacant region in an array. We replace failed regions one a
2476 * array at a time. The result is that a new spare disk will be added
2477 * to the first failed array and after the monitor has finished
2478 * propagating failures the remainder will be consumed.
2480 * FIXME add a capability for mdmon to request spares from another
2484 struct intel_super
*super
= a
->container
->sb
;
2485 int inst
= a
->info
.container_member
;
2486 struct imsm_dev
*dev
= get_imsm_dev(super
, inst
);
2487 struct imsm_map
*map
= get_imsm_map(dev
, 0);
2488 int failed
= a
->info
.array
.raid_disks
;
2489 struct mdinfo
*rv
= NULL
;
2492 struct metadata_update
*mu
;
2494 struct imsm_update_activate_spare
*u
;
2498 for (d
= a
->info
.devs
; d
; d
= d
->next
) {
2499 if ((d
->curr_state
& DS_FAULTY
) &&
2501 /* wait for Removal to happen */
2503 if (d
->state_fd
>= 0)
2507 dprintf("imsm: activate spare: inst=%d failed=%d (%d) level=%d\n",
2508 inst
, failed
, a
->info
.array
.raid_disks
, a
->info
.array
.level
);
2509 if (imsm_check_degraded(super
, inst
, failed
) != IMSM_T_STATE_DEGRADED
)
2512 /* For each slot, if it is not working, find a spare */
2513 for (i
= 0; i
< a
->info
.array
.raid_disks
; i
++) {
2514 for (d
= a
->info
.devs
; d
; d
= d
->next
)
2515 if (d
->disk
.raid_disk
== i
)
2517 dprintf("found %d: %p %x\n", i
, d
, d
?d
->curr_state
:0);
2518 if (d
&& (d
->state_fd
>= 0))
2522 * OK, this device needs recovery. Try to re-add the previous
2523 * occupant of this slot, if this fails add a new spare
2525 dl
= imsm_readd(super
, i
, a
);
2527 dl
= imsm_add_spare(super
, i
, a
);
2531 /* found a usable disk with enough space */
2532 di
= malloc(sizeof(*di
));
2533 memset(di
, 0, sizeof(*di
));
2535 /* dl->index will be -1 in the case we are activating a
2536 * pristine spare. imsm_process_update() will create a
2537 * new index in this case. Once a disk is found to be
2538 * failed in all member arrays it is kicked from the
2541 di
->disk
.number
= dl
->index
;
2543 /* (ab)use di->devs to store a pointer to the device
2546 di
->devs
= (struct mdinfo
*) dl
;
2548 di
->disk
.raid_disk
= i
;
2549 di
->disk
.major
= dl
->major
;
2550 di
->disk
.minor
= dl
->minor
;
2552 di
->data_offset
= __le32_to_cpu(map
->pba_of_lba0
);
2553 di
->component_size
= a
->info
.component_size
;
2554 di
->container_member
= inst
;
2558 dprintf("%x:%x to be %d at %llu\n", dl
->major
, dl
->minor
,
2559 i
, di
->data_offset
);
2565 /* No spares found */
2567 /* Now 'rv' has a list of devices to return.
2568 * Create a metadata_update record to update the
2569 * disk_ord_tbl for the array
2571 mu
= malloc(sizeof(*mu
));
2572 mu
->buf
= malloc(sizeof(struct imsm_update_activate_spare
) * num_spares
);
2574 mu
->len
= sizeof(struct imsm_update_activate_spare
) * num_spares
;
2575 mu
->next
= *updates
;
2576 u
= (struct imsm_update_activate_spare
*) mu
->buf
;
2578 for (di
= rv
; di
; di
= di
->next
) {
2579 u
->type
= update_activate_spare
;
2580 u
->dl
= (struct dl
*) di
->devs
;
2582 u
->slot
= di
->disk
.raid_disk
;
2593 static int disks_overlap(struct imsm_map
*m1
, struct imsm_map
*m2
)
2599 for (i
= 0; i
< m1
->num_members
; i
++) {
2600 idx
= get_imsm_disk_idx(m1
, i
);
2601 for (j
= 0; j
< m2
->num_members
; j
++)
2602 if (idx
== get_imsm_disk_idx(m2
, j
))
2609 static void imsm_delete(struct intel_super
*super
, struct dl
**dlp
);
2611 static void imsm_process_update(struct supertype
*st
,
2612 struct metadata_update
*update
)
2615 * crack open the metadata_update envelope to find the update record
2616 * update can be one of:
2617 * update_activate_spare - a spare device has replaced a failed
2618 * device in an array, update the disk_ord_tbl. If this disk is
2619 * present in all member arrays then also clear the SPARE_DISK
2622 struct intel_super
*super
= st
->sb
;
2623 struct imsm_super
*mpb
= super
->anchor
;
2624 enum imsm_update_type type
= *(enum imsm_update_type
*) update
->buf
;
2627 case update_activate_spare
: {
2628 struct imsm_update_activate_spare
*u
= (void *) update
->buf
;
2629 struct imsm_dev
*dev
= get_imsm_dev(super
, u
->array
);
2630 struct imsm_map
*map
= get_imsm_map(dev
, 0);
2631 struct active_array
*a
;
2632 struct imsm_disk
*disk
;
2639 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
2644 fprintf(stderr
, "error: imsm_activate_spare passed "
2645 "an unknown disk (index: %d serial: %s)\n",
2646 u
->dl
->index
, u
->dl
->serial
);
2650 super
->updates_pending
++;
2652 /* adding a pristine spare, assign a new index */
2653 if (dl
->index
< 0) {
2654 dl
->index
= super
->anchor
->num_disks
;
2655 super
->anchor
->num_disks
++;
2657 victim
= get_imsm_disk_idx(map
, u
->slot
);
2658 map
->disk_ord_tbl
[u
->slot
] = __cpu_to_le32(dl
->index
);
2660 status
= __le32_to_cpu(disk
->status
);
2661 status
|= CONFIGURED_DISK
;
2662 status
&= ~(SPARE_DISK
| USABLE_DISK
);
2663 disk
->status
= __cpu_to_le32(status
);
2665 /* count arrays using the victim in the metadata */
2667 for (a
= st
->arrays
; a
; a
= a
->next
) {
2668 dev
= get_imsm_dev(super
, a
->info
.container_member
);
2669 map
= get_imsm_map(dev
, 0);
2670 for (i
= 0; i
< map
->num_members
; i
++)
2671 if (victim
== get_imsm_disk_idx(map
, i
))
2675 /* clear some flags if the victim is no longer being
2680 for (dlp
= &super
->disks
; *dlp
; )
2681 if ((*dlp
)->index
== victim
)
2683 disk
= &(*dlp
)->disk
;
2684 status
= __le32_to_cpu(disk
->status
);
2685 status
&= ~(CONFIGURED_DISK
| USABLE_DISK
);
2686 disk
->status
= __cpu_to_le32(status
);
2687 /* We know that 'manager' isn't touching anything,
2690 imsm_delete(super
, dlp
);
2694 case update_create_array
: {
2695 /* someone wants to create a new array, we need to be aware of
2696 * a few races/collisions:
2697 * 1/ 'Create' called by two separate instances of mdadm
2698 * 2/ 'Create' versus 'activate_spare': mdadm has chosen
2699 * devices that have since been assimilated via
2701 * In the event this update can not be carried out mdadm will
2702 * (FIX ME) notice that its update did not take hold.
2704 struct imsm_update_create_array
*u
= (void *) update
->buf
;
2705 struct imsm_dev
*dev
;
2706 struct imsm_map
*map
, *new_map
;
2707 unsigned long long start
, end
;
2708 unsigned long long new_start
, new_end
;
2712 /* handle racing creates: first come first serve */
2713 if (u
->dev_idx
< mpb
->num_raid_devs
) {
2714 dprintf("%s: subarray %d already defined\n",
2715 __func__
, u
->dev_idx
);
2719 /* check update is next in sequence */
2720 if (u
->dev_idx
!= mpb
->num_raid_devs
) {
2721 dprintf("%s: can not create array %d expected index %d\n",
2722 __func__
, u
->dev_idx
, mpb
->num_raid_devs
);
2726 new_map
= get_imsm_map(&u
->dev
, 0);
2727 new_start
= __le32_to_cpu(new_map
->pba_of_lba0
);
2728 new_end
= new_start
+ __le32_to_cpu(new_map
->blocks_per_member
);
2730 /* handle activate_spare versus create race:
2731 * check to make sure that overlapping arrays do not include
2734 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
2735 dev
= get_imsm_dev(super
, i
);
2736 map
= get_imsm_map(dev
, 0);
2737 start
= __le32_to_cpu(map
->pba_of_lba0
);
2738 end
= start
+ __le32_to_cpu(map
->blocks_per_member
);
2739 if ((new_start
>= start
&& new_start
<= end
) ||
2740 (start
>= new_start
&& start
<= new_end
))
2742 if (overlap
&& disks_overlap(map
, new_map
)) {
2743 dprintf("%s: arrays overlap\n", __func__
);
2747 /* check num_members sanity */
2748 if (new_map
->num_members
> mpb
->num_disks
) {
2749 dprintf("%s: num_disks out of range\n", __func__
);
2753 /* check that prepare update was successful */
2754 if (!update
->space
) {
2755 dprintf("%s: prepare update failed\n", __func__
);
2759 super
->updates_pending
++;
2760 dev
= update
->space
;
2761 update
->space
= NULL
;
2762 imsm_copy_dev(dev
, &u
->dev
);
2763 super
->dev_tbl
[u
->dev_idx
] = dev
;
2764 mpb
->num_raid_devs
++;
2766 /* fix up flags, if arrays overlap then the drives can not be
2769 for (i
= 0; i
< map
->num_members
; i
++) {
2770 struct imsm_disk
*disk
;
2773 disk
= get_imsm_disk(super
, get_imsm_disk_idx(map
, i
));
2774 status
= __le32_to_cpu(disk
->status
);
2775 status
|= CONFIGURED_DISK
;
2777 status
&= ~SPARE_DISK
;
2778 disk
->status
= __cpu_to_le32(status
);
2785 static void imsm_prepare_update(struct supertype
*st
,
2786 struct metadata_update
*update
)
2789 * Allocate space to hold new disk entries, raid-device entries or a
2790 * new mpb if necessary. We currently maintain an mpb large enough to
2791 * hold 2 subarrays for the given number of disks. This may not be
2792 * sufficient when reshaping.
2794 * FIX ME handle the reshape case.
2796 * The monitor will be able to safely change super->mpb by arranging
2797 * for it to be freed in check_update_queue(). I.e. the monitor thread
2798 * will start using the new pointer and the manager can continue to use
2799 * the old value until check_update_queue() runs.
2801 enum imsm_update_type type
= *(enum imsm_update_type
*) update
->buf
;
2804 case update_create_array
: {
2805 struct imsm_update_create_array
*u
= (void *) update
->buf
;
2806 size_t len
= sizeof_imsm_dev(&u
->dev
, 1);
2808 update
->space
= malloc(len
);
2818 /* must be called while manager is quiesced */
2819 static void imsm_delete(struct intel_super
*super
, struct dl
**dlp
)
2821 struct imsm_super
*mpb
= super
->anchor
;
2822 struct dl
*dl
= *dlp
;
2824 struct imsm_dev
*dev
;
2825 struct imsm_map
*map
;
2828 dprintf("%s: deleting device %x:%x from imsm_super\n",
2829 __func__
, dl
->major
, dl
->minor
);
2831 /* shift all indexes down one */
2832 for (iter
= super
->disks
; iter
; iter
= iter
->next
)
2833 if (iter
->index
> dl
->index
)
2836 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
2837 dev
= get_imsm_dev(super
, i
);
2838 map
= get_imsm_map(dev
, 0);
2840 for (j
= 0; j
< map
->num_members
; j
++) {
2841 int idx
= get_imsm_disk_idx(map
, j
);
2843 if (idx
> dl
->index
)
2844 map
->disk_ord_tbl
[j
] = __cpu_to_le32(idx
- 1);
2849 super
->updates_pending
++;
2850 *dlp
= (*dlp
)->next
;
2851 __free_imsm_disk(dl
);
2854 struct superswitch super_imsm
= {
2856 .examine_super
= examine_super_imsm
,
2857 .brief_examine_super
= brief_examine_super_imsm
,
2858 .detail_super
= detail_super_imsm
,
2859 .brief_detail_super
= brief_detail_super_imsm
,
2860 .write_init_super
= write_init_super_imsm
,
2862 .match_home
= match_home_imsm
,
2863 .uuid_from_super
= uuid_from_super_imsm
,
2864 .getinfo_super
= getinfo_super_imsm
,
2865 .update_super
= update_super_imsm
,
2867 .avail_size
= avail_size_imsm
,
2869 .compare_super
= compare_super_imsm
,
2871 .load_super
= load_super_imsm
,
2872 .init_super
= init_super_imsm
,
2873 .add_to_super
= add_to_super_imsm
,
2874 .store_super
= store_zero_imsm
,
2875 .free_super
= free_super_imsm
,
2876 .match_metadata_desc
= match_metadata_desc_imsm
,
2877 .container_content
= container_content_imsm
,
2879 .validate_geometry
= validate_geometry_imsm
,
2883 .open_new
= imsm_open_new
,
2884 .load_super
= load_super_imsm
,
2885 .set_array_state
= imsm_set_array_state
,
2886 .set_disk
= imsm_set_disk
,
2887 .sync_metadata
= imsm_sync_metadata
,
2888 .activate_spare
= imsm_activate_spare
,
2889 .process_update
= imsm_process_update
,
2890 .prepare_update
= imsm_prepare_update
,