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 __u32 disk_ord_tbl
[1]; /* disk_ord_tbl[num_members],
73 } __attribute__ ((packed
));
77 __u8 migr_state
; /* Normal or Migrating */
78 __u8 migr_type
; /* Initializing, Rebuilding, ... */
82 struct imsm_map map
[1];
83 /* here comes another one if migr_state */
84 } __attribute__ ((packed
));
87 __u8 volume
[MAX_RAID_SERIAL_LEN
];
90 __u32 status
; /* Persistent RaidDev status */
91 __u32 reserved_blocks
; /* Reserved blocks at beginning of volume */
92 #define IMSM_DEV_FILLERS 12
93 __u32 filler
[IMSM_DEV_FILLERS
];
95 } __attribute__ ((packed
));
98 __u8 sig
[MAX_SIGNATURE_LENGTH
]; /* 0x00 - 0x1F */
99 __u32 check_sum
; /* 0x20 - 0x23 MPB Checksum */
100 __u32 mpb_size
; /* 0x24 - 0x27 Size of MPB */
101 __u32 family_num
; /* 0x28 - 0x2B Checksum from first time this config was written */
102 __u32 generation_num
; /* 0x2C - 0x2F Incremented each time this array's MPB is written */
103 __u32 error_log_size
; /* 0x30 - 0x33 in bytes */
104 __u32 attributes
; /* 0x34 - 0x37 */
105 __u8 num_disks
; /* 0x38 Number of configured disks */
106 __u8 num_raid_devs
; /* 0x39 Number of configured volumes */
107 __u8 error_log_pos
; /* 0x3A */
108 __u8 fill
[1]; /* 0x3B */
109 __u32 cache_size
; /* 0x3c - 0x40 in mb */
110 __u32 orig_family_num
; /* 0x40 - 0x43 original family num */
111 __u32 pwr_cycle_count
; /* 0x44 - 0x47 simulated power cycle count for array */
112 __u32 bbm_log_size
; /* 0x48 - 0x4B - size of bad Block Mgmt Log in bytes */
113 #define IMSM_FILLERS 35
114 __u32 filler
[IMSM_FILLERS
]; /* 0x4C - 0xD7 RAID_MPB_FILLERS */
115 struct imsm_disk disk
[1]; /* 0xD8 diskTbl[numDisks] */
116 /* here comes imsm_dev[num_raid_devs] */
117 /* here comes BBM logs */
118 } __attribute__ ((packed
));
120 #define BBM_LOG_MAX_ENTRIES 254
122 struct bbm_log_entry
{
123 __u64 defective_block_start
;
124 #define UNREADABLE 0xFFFFFFFF
125 __u32 spare_block_offset
;
126 __u16 remapped_marked_count
;
128 } __attribute__ ((__packed__
));
131 __u32 signature
; /* 0xABADB10C */
133 __u32 reserved_spare_block_count
; /* 0 */
134 __u32 reserved
; /* 0xFFFF */
135 __u64 first_spare_lba
;
136 struct bbm_log_entry mapped_block_entries
[BBM_LOG_MAX_ENTRIES
];
137 } __attribute__ ((__packed__
));
141 static char *map_state_str
[] = { "normal", "uninitialized", "degraded", "failed" };
144 static unsigned int sector_count(__u32 bytes
)
146 return ((bytes
+ (512-1)) & (~(512-1))) / 512;
149 static unsigned int mpb_sectors(struct imsm_super
*mpb
)
151 return sector_count(__le32_to_cpu(mpb
->mpb_size
));
154 /* internal representation of IMSM metadata */
157 void *buf
; /* O_DIRECT buffer for reading/writing metadata */
158 struct imsm_super
*anchor
; /* immovable parameters */
160 size_t len
; /* size of the 'buf' allocation */
161 int updates_pending
; /* count of pending updates for mdmon */
162 int creating_imsm
; /* flag to indicate container creation */
163 int current_vol
; /* index of raid device undergoing creation */
164 #define IMSM_MAX_RAID_DEVS 2
165 struct imsm_dev
*dev_tbl
[IMSM_MAX_RAID_DEVS
];
169 __u8 serial
[MAX_RAID_SERIAL_LEN
];
172 struct imsm_disk disk
;
175 struct bbm_log
*bbm_log
;
179 unsigned long long start
, size
;
182 /* definition of messages passed to imsm_process_update */
183 enum imsm_update_type
{
184 update_activate_spare
,
188 struct imsm_update_activate_spare
{
189 enum imsm_update_type type
;
193 struct imsm_update_activate_spare
*next
;
196 struct imsm_update_create_array
{
197 enum imsm_update_type type
;
202 static int imsm_env_devname_as_serial(void)
204 char *val
= getenv("IMSM_DEVNAME_AS_SERIAL");
206 if (val
&& atoi(val
) == 1)
213 static struct supertype
*match_metadata_desc_imsm(char *arg
)
215 struct supertype
*st
;
217 if (strcmp(arg
, "imsm") != 0 &&
218 strcmp(arg
, "default") != 0
222 st
= malloc(sizeof(*st
));
223 memset(st
, 0, sizeof(*st
));
224 st
->ss
= &super_imsm
;
225 st
->max_devs
= IMSM_MAX_DEVICES
;
226 st
->minor_version
= 0;
231 static __u8
*get_imsm_version(struct imsm_super
*mpb
)
233 return &mpb
->sig
[MPB_SIG_LEN
];
236 /* retrieve a disk directly from the anchor when the anchor is known to be
237 * up-to-date, currently only at load time
239 static struct imsm_disk
*__get_imsm_disk(struct imsm_super
*mpb
, __u8 index
)
241 if (index
>= mpb
->num_disks
)
243 return &mpb
->disk
[index
];
246 /* retrieve a disk from the parsed metadata */
247 static struct imsm_disk
*get_imsm_disk(struct intel_super
*super
, __u8 index
)
251 for (d
= super
->disks
; d
; d
= d
->next
)
252 if (d
->index
== index
)
258 /* generate a checksum directly from the anchor when the anchor is known to be
259 * up-to-date, currently only at load or write_super after coalescing
261 static __u32
__gen_imsm_checksum(struct imsm_super
*mpb
)
263 __u32 end
= mpb
->mpb_size
/ sizeof(end
);
264 __u32
*p
= (__u32
*) mpb
;
268 sum
+= __le32_to_cpu(*p
++);
270 return sum
- __le32_to_cpu(mpb
->check_sum
);
273 static size_t sizeof_imsm_map(struct imsm_map
*map
)
275 return sizeof(struct imsm_map
) + sizeof(__u32
) * (map
->num_members
- 1);
278 struct imsm_map
*get_imsm_map(struct imsm_dev
*dev
, int second_map
)
280 struct imsm_map
*map
= &dev
->vol
.map
[0];
282 if (second_map
&& !dev
->vol
.migr_state
)
284 else if (second_map
) {
287 return ptr
+ sizeof_imsm_map(map
);
293 /* return the size of the device.
294 * migr_state increases the returned size if map[0] were to be duplicated
296 static size_t sizeof_imsm_dev(struct imsm_dev
*dev
, int migr_state
)
298 size_t size
= sizeof(*dev
) - sizeof(struct imsm_map
) +
299 sizeof_imsm_map(get_imsm_map(dev
, 0));
301 /* migrating means an additional map */
302 if (dev
->vol
.migr_state
)
303 size
+= sizeof_imsm_map(get_imsm_map(dev
, 1));
305 size
+= sizeof_imsm_map(get_imsm_map(dev
, 0));
310 static struct imsm_dev
*__get_imsm_dev(struct imsm_super
*mpb
, __u8 index
)
316 if (index
>= mpb
->num_raid_devs
)
319 /* devices start after all disks */
320 offset
= ((void *) &mpb
->disk
[mpb
->num_disks
]) - _mpb
;
322 for (i
= 0; i
<= index
; i
++)
324 return _mpb
+ offset
;
326 offset
+= sizeof_imsm_dev(_mpb
+ offset
, 0);
331 static struct imsm_dev
*get_imsm_dev(struct intel_super
*super
, __u8 index
)
333 if (index
>= super
->anchor
->num_raid_devs
)
335 return super
->dev_tbl
[index
];
338 static __u32
get_imsm_disk_idx(struct imsm_map
*map
, int slot
)
340 __u32
*ord_tbl
= &map
->disk_ord_tbl
[slot
];
342 /* top byte is 'special' */
343 return __le32_to_cpu(*ord_tbl
& ~(0xff << 24));
346 static int get_imsm_raid_level(struct imsm_map
*map
)
348 if (map
->raid_level
== 1) {
349 if (map
->num_members
== 2)
355 return map
->raid_level
;
358 static int cmp_extent(const void *av
, const void *bv
)
360 const struct extent
*a
= av
;
361 const struct extent
*b
= bv
;
362 if (a
->start
< b
->start
)
364 if (a
->start
> b
->start
)
369 static struct extent
*get_extents(struct intel_super
*super
, struct dl
*dl
)
371 /* find a list of used extents on the given physical device */
372 struct extent
*rv
, *e
;
376 for (i
= 0; i
< super
->anchor
->num_raid_devs
; i
++) {
377 struct imsm_dev
*dev
= get_imsm_dev(super
, i
);
378 struct imsm_map
*map
= get_imsm_map(dev
, 0);
380 for (j
= 0; j
< map
->num_members
; j
++) {
381 __u32 index
= get_imsm_disk_idx(map
, j
);
383 if (index
== dl
->index
)
387 rv
= malloc(sizeof(struct extent
) * (memberships
+ 1));
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
) {
400 e
->start
= __le32_to_cpu(map
->pba_of_lba0
);
401 e
->size
= __le32_to_cpu(map
->blocks_per_member
);
406 qsort(rv
, memberships
, sizeof(*rv
), cmp_extent
);
408 e
->start
= __le32_to_cpu(dl
->disk
.total_blocks
) -
409 (MPB_SECTOR_CNT
+ IMSM_RESERVED_SECTORS
);
415 static void print_imsm_dev(struct imsm_dev
*dev
, int index
)
419 struct imsm_map
*map
= get_imsm_map(dev
, 0);
422 printf("[%s]:\n", dev
->volume
);
423 printf(" RAID Level : %d\n", get_imsm_raid_level(map
));
424 printf(" Members : %d\n", map
->num_members
);
425 for (slot
= 0; slot
< map
->num_members
; slot
++)
426 if (index
== get_imsm_disk_idx(map
, slot
))
428 if (slot
< map
->num_members
)
429 printf(" This Slot : %d\n", slot
);
431 printf(" This Slot : ?\n");
432 sz
= __le32_to_cpu(dev
->size_high
);
434 sz
+= __le32_to_cpu(dev
->size_low
);
435 printf(" Array Size : %llu%s\n", (unsigned long long)sz
,
436 human_size(sz
* 512));
437 sz
= __le32_to_cpu(map
->blocks_per_member
);
438 printf(" Per Dev Size : %llu%s\n", (unsigned long long)sz
,
439 human_size(sz
* 512));
440 printf(" Sector Offset : %u\n",
441 __le32_to_cpu(map
->pba_of_lba0
));
442 printf(" Num Stripes : %u\n",
443 __le32_to_cpu(map
->num_data_stripes
));
444 printf(" Chunk Size : %u KiB\n",
445 __le16_to_cpu(map
->blocks_per_strip
) / 2);
446 printf(" Reserved : %d\n", __le32_to_cpu(dev
->reserved_blocks
));
447 printf(" Migrate State : %s", dev
->vol
.migr_state
? "migrating" : "idle");
448 if (dev
->vol
.migr_state
)
449 printf(": %s", dev
->vol
.migr_type
? "rebuilding" : "initializing");
451 printf(" Map State : %s", map_state_str
[map
->map_state
]);
452 if (dev
->vol
.migr_state
) {
453 struct imsm_map
*map
= get_imsm_map(dev
, 1);
454 printf(", %s", map_state_str
[map
->map_state
]);
457 printf(" Dirty State : %s\n", dev
->vol
.dirty
? "dirty" : "clean");
460 static void print_imsm_disk(struct imsm_super
*mpb
, int index
)
462 struct imsm_disk
*disk
= __get_imsm_disk(mpb
, index
);
463 char str
[MAX_RAID_SERIAL_LEN
];
471 snprintf(str
, MAX_RAID_SERIAL_LEN
, "%s", disk
->serial
);
472 printf(" Disk%02d Serial : %s\n", index
, str
);
473 s
= __le32_to_cpu(disk
->status
);
474 printf(" State :%s%s%s%s\n", s
&SPARE_DISK
? " spare" : "",
475 s
&CONFIGURED_DISK
? " active" : "",
476 s
&FAILED_DISK
? " failed" : "",
477 s
&USABLE_DISK
? " usable" : "");
478 printf(" Id : %08x\n", __le32_to_cpu(disk
->scsi_id
));
479 sz
= __le32_to_cpu(disk
->total_blocks
) -
480 (MPB_SECTOR_CNT
+ IMSM_RESERVED_SECTORS
* mpb
->num_raid_devs
);
481 printf(" Usable Size : %llu%s\n", (unsigned long long)sz
,
482 human_size(sz
* 512));
485 static void examine_super_imsm(struct supertype
*st
, char *homehost
)
487 struct intel_super
*super
= st
->sb
;
488 struct imsm_super
*mpb
= super
->anchor
;
489 char str
[MAX_SIGNATURE_LENGTH
];
493 snprintf(str
, MPB_SIG_LEN
, "%s", mpb
->sig
);
494 printf(" Magic : %s\n", str
);
495 snprintf(str
, strlen(MPB_VERSION_RAID0
), "%s", get_imsm_version(mpb
));
496 printf(" Version : %s\n", get_imsm_version(mpb
));
497 printf(" Family : %08x\n", __le32_to_cpu(mpb
->family_num
));
498 printf(" Generation : %08x\n", __le32_to_cpu(mpb
->generation_num
));
499 sum
= __le32_to_cpu(mpb
->check_sum
);
500 printf(" Checksum : %08x %s\n", sum
,
501 __gen_imsm_checksum(mpb
) == sum
? "correct" : "incorrect");
502 printf(" MPB Sectors : %d\n", mpb_sectors(mpb
));
503 printf(" Disks : %d\n", mpb
->num_disks
);
504 printf(" RAID Devices : %d\n", mpb
->num_raid_devs
);
505 print_imsm_disk(mpb
, super
->disks
->index
);
506 if (super
->bbm_log
) {
507 struct bbm_log
*log
= super
->bbm_log
;
510 printf("Bad Block Management Log:\n");
511 printf(" Log Size : %d\n", __le32_to_cpu(mpb
->bbm_log_size
));
512 printf(" Signature : %x\n", __le32_to_cpu(log
->signature
));
513 printf(" Entry Count : %d\n", __le32_to_cpu(log
->entry_count
));
514 printf(" Spare Blocks : %d\n", __le32_to_cpu(log
->reserved_spare_block_count
));
515 printf(" First Spare : %llx\n", __le64_to_cpu(log
->first_spare_lba
));
517 for (i
= 0; i
< mpb
->num_raid_devs
; i
++)
518 print_imsm_dev(__get_imsm_dev(mpb
, i
), super
->disks
->index
);
519 for (i
= 0; i
< mpb
->num_disks
; i
++) {
520 if (i
== super
->disks
->index
)
522 print_imsm_disk(mpb
, i
);
526 static void brief_examine_super_imsm(struct supertype
*st
)
528 printf("ARRAY /dev/imsm metadata=imsm\n");
531 static void detail_super_imsm(struct supertype
*st
, char *homehost
)
533 printf("%s\n", __FUNCTION__
);
536 static void brief_detail_super_imsm(struct supertype
*st
)
538 printf("%s\n", __FUNCTION__
);
542 static int match_home_imsm(struct supertype
*st
, char *homehost
)
544 printf("%s\n", __FUNCTION__
);
549 static void uuid_from_super_imsm(struct supertype
*st
, int uuid
[4])
551 printf("%s\n", __FUNCTION__
);
556 get_imsm_numerical_version(struct imsm_super
*mpb
, int *m
, int *p
)
558 __u8
*v
= get_imsm_version(mpb
);
559 __u8
*end
= mpb
->sig
+ MAX_SIGNATURE_LENGTH
;
560 char major
[] = { 0, 0, 0 };
561 char minor
[] = { 0 ,0, 0 };
562 char patch
[] = { 0, 0, 0 };
563 char *ver_parse
[] = { major
, minor
, patch
};
567 while (*v
!= '\0' && v
< end
) {
568 if (*v
!= '.' && j
< 2)
569 ver_parse
[i
][j
++] = *v
;
577 *m
= strtol(minor
, NULL
, 0);
578 *p
= strtol(patch
, NULL
, 0);
582 static int imsm_level_to_layout(int level
)
590 return ALGORITHM_LEFT_ASYMMETRIC
;
592 return 0x102; //FIXME is this correct?
597 static void getinfo_super_imsm_volume(struct supertype
*st
, struct mdinfo
*info
)
599 struct intel_super
*super
= st
->sb
;
600 struct imsm_dev
*dev
= get_imsm_dev(super
, super
->current_vol
);
601 struct imsm_map
*map
= get_imsm_map(dev
, 0);
603 info
->container_member
= super
->current_vol
;
604 info
->array
.raid_disks
= map
->num_members
;
605 info
->array
.level
= get_imsm_raid_level(map
);
606 info
->array
.layout
= imsm_level_to_layout(info
->array
.level
);
607 info
->array
.md_minor
= -1;
608 info
->array
.ctime
= 0;
609 info
->array
.utime
= 0;
610 info
->array
.chunk_size
= __le16_to_cpu(map
->blocks_per_strip
* 512);
612 info
->data_offset
= __le32_to_cpu(map
->pba_of_lba0
);
613 info
->component_size
= __le32_to_cpu(map
->blocks_per_member
);
615 info
->disk
.major
= 0;
616 info
->disk
.minor
= 0;
618 sprintf(info
->text_version
, "/%s/%d",
619 devnum2devname(st
->container_dev
),
620 info
->container_member
);
624 static void getinfo_super_imsm(struct supertype
*st
, struct mdinfo
*info
)
626 struct intel_super
*super
= st
->sb
;
627 struct imsm_disk
*disk
;
630 if (super
->current_vol
>= 0) {
631 getinfo_super_imsm_volume(st
, info
);
635 /* Set raid_disks to zero so that Assemble will always pull in valid
638 info
->array
.raid_disks
= 0;
639 info
->array
.level
= LEVEL_CONTAINER
;
640 info
->array
.layout
= 0;
641 info
->array
.md_minor
= -1;
642 info
->array
.ctime
= 0; /* N/A for imsm */
643 info
->array
.utime
= 0;
644 info
->array
.chunk_size
= 0;
646 info
->disk
.major
= 0;
647 info
->disk
.minor
= 0;
648 info
->disk
.raid_disk
= -1;
649 info
->reshape_active
= 0;
650 strcpy(info
->text_version
, "imsm");
651 info
->disk
.number
= -1;
652 info
->disk
.state
= 0;
655 disk
= &super
->disks
->disk
;
656 info
->disk
.number
= super
->disks
->index
;
657 info
->disk
.raid_disk
= super
->disks
->index
;
658 info
->data_offset
= __le32_to_cpu(disk
->total_blocks
) -
659 (MPB_SECTOR_CNT
+ IMSM_RESERVED_SECTORS
);
660 info
->component_size
= MPB_SECTOR_CNT
+ IMSM_RESERVED_SECTORS
;
661 s
= __le32_to_cpu(disk
->status
);
662 info
->disk
.state
= s
& CONFIGURED_DISK
? (1 << MD_DISK_ACTIVE
) : 0;
663 info
->disk
.state
|= s
& FAILED_DISK
? (1 << MD_DISK_FAULTY
) : 0;
664 info
->disk
.state
|= s
& USABLE_DISK
? (1 << MD_DISK_SYNC
) : 0;
668 static int update_super_imsm(struct supertype
*st
, struct mdinfo
*info
,
669 char *update
, char *devname
, int verbose
,
670 int uuid_set
, char *homehost
)
674 /* For 'assemble' and 'force' we need to return non-zero if any
675 * change was made. For others, the return value is ignored.
676 * Update options are:
677 * force-one : This device looks a bit old but needs to be included,
678 * update age info appropriately.
679 * assemble: clear any 'faulty' flag to allow this device to
681 * force-array: Array is degraded but being forced, mark it clean
682 * if that will be needed to assemble it.
684 * newdev: not used ????
685 * grow: Array has gained a new device - this is currently for
687 * resync: mark as dirty so a resync will happen.
688 * name: update the name - preserving the homehost
690 * Following are not relevant for this imsm:
691 * sparc2.2 : update from old dodgey metadata
692 * super-minor: change the preferred_minor number
693 * summaries: update redundant counters.
694 * uuid: Change the uuid of the array to match watch is given
695 * homehost: update the recorded homehost
696 * _reshape_progress: record new reshape_progress position.
699 //struct intel_super *super = st->sb;
700 //struct imsm_super *mpb = super->mpb;
702 if (strcmp(update
, "grow") == 0) {
704 if (strcmp(update
, "resync") == 0) {
705 /* dev->vol.dirty = 1; */
708 /* IMSM has no concept of UUID or homehost */
713 static size_t disks_to_mpb_size(int disks
)
717 size
= sizeof(struct imsm_super
);
718 size
+= (disks
- 1) * sizeof(struct imsm_disk
);
719 size
+= 2 * sizeof(struct imsm_dev
);
720 /* up to 2 maps per raid device (-2 for imsm_maps in imsm_dev */
721 size
+= (4 - 2) * sizeof(struct imsm_map
);
722 /* 4 possible disk_ord_tbl's */
723 size
+= 4 * (disks
- 1) * sizeof(__u32
);
728 static __u64
avail_size_imsm(struct supertype
*st
, __u64 devsize
)
730 if (devsize
< (MPB_SECTOR_CNT
+ IMSM_RESERVED_SECTORS
))
733 return devsize
- (MPB_SECTOR_CNT
+ IMSM_RESERVED_SECTORS
);
736 static int compare_super_imsm(struct supertype
*st
, struct supertype
*tst
)
740 * 0 same, or first was empty, and second was copied
741 * 1 second had wrong number
745 struct intel_super
*first
= st
->sb
;
746 struct intel_super
*sec
= tst
->sb
;
754 if (memcmp(first
->anchor
->sig
, sec
->anchor
->sig
, MAX_SIGNATURE_LENGTH
) != 0)
757 /* if an anchor does not have num_raid_devs set then it is a free
760 if (first
->anchor
->num_raid_devs
> 0 &&
761 sec
->anchor
->num_raid_devs
> 0) {
762 if (first
->anchor
->family_num
!= sec
->anchor
->family_num
)
764 if (first
->anchor
->mpb_size
!= sec
->anchor
->mpb_size
)
766 if (first
->anchor
->check_sum
!= sec
->anchor
->check_sum
)
773 static void fd2devname(int fd
, char *name
)
782 if (fstat(fd
, &st
) != 0)
784 sprintf(path
, "/sys/dev/block/%d:%d",
785 major(st
.st_rdev
), minor(st
.st_rdev
));
787 rv
= readlink(path
, dname
, sizeof(dname
));
792 nm
= strrchr(dname
, '/');
794 snprintf(name
, MAX_RAID_SERIAL_LEN
, "/dev/%s", nm
);
798 extern int scsi_get_serial(int fd
, void *buf
, size_t buf_len
);
800 static int imsm_read_serial(int fd
, char *devname
,
801 __u8 serial
[MAX_RAID_SERIAL_LEN
])
803 unsigned char scsi_serial
[255];
808 memset(scsi_serial
, 0, sizeof(scsi_serial
));
810 if (imsm_env_devname_as_serial()) {
811 char name
[MAX_RAID_SERIAL_LEN
];
813 fd2devname(fd
, name
);
814 strcpy((char *) serial
, name
);
818 rv
= scsi_get_serial(fd
, scsi_serial
, sizeof(scsi_serial
));
823 Name
": Failed to retrieve serial for %s\n",
828 rsp_len
= scsi_serial
[3];
829 for (i
= 0, cnt
= 0; i
< rsp_len
; i
++) {
830 if (!isspace(scsi_serial
[4 + i
]))
831 serial
[cnt
++] = scsi_serial
[4 + i
];
832 if (cnt
== MAX_RAID_SERIAL_LEN
)
836 serial
[MAX_RAID_SERIAL_LEN
- 1] = '\0';
842 load_imsm_disk(int fd
, struct intel_super
*super
, char *devname
, int keep_fd
)
849 __u8 serial
[MAX_RAID_SERIAL_LEN
];
851 rv
= imsm_read_serial(fd
, devname
, serial
);
856 /* check if this is a disk we have seen before. it may be a spare in
857 * super->disks while the current anchor believes it is a raid member,
858 * check if we need to update dl->index
860 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
861 if (memcmp(dl
->serial
, serial
, MAX_RAID_SERIAL_LEN
) == 0)
865 dl
= malloc(sizeof(*dl
));
872 Name
": failed to allocate disk buffer for %s\n",
879 dl
->major
= major(stb
.st_rdev
);
880 dl
->minor
= minor(stb
.st_rdev
);
881 dl
->next
= super
->disks
;
882 dl
->fd
= keep_fd
? fd
: -1;
883 dl
->devname
= devname
? strdup(devname
) : NULL
;
884 strncpy((char *) dl
->serial
, (char *) serial
, MAX_RAID_SERIAL_LEN
);
885 } else if (keep_fd
) {
890 /* look up this disk's index in the current anchor */
891 for (i
= 0; i
< super
->anchor
->num_disks
; i
++) {
892 struct imsm_disk
*disk_iter
;
894 disk_iter
= __get_imsm_disk(super
->anchor
, i
);
896 if (memcmp(disk_iter
->serial
, dl
->serial
,
897 MAX_RAID_SERIAL_LEN
) == 0) {
900 dl
->disk
= *disk_iter
;
901 status
= __le32_to_cpu(dl
->disk
.status
);
902 /* only set index on disks that are a member of a
903 * populated contianer, i.e. one with raid_devs
905 if (status
& SPARE_DISK
)
913 if (i
== super
->anchor
->num_disks
&& alloc
) {
916 Name
": failed to load disk with serial \'%s\' for %s\n",
917 dl
->serial
, devname
);
921 if (i
== super
->anchor
->num_disks
&& dl
->index
>= 0) {
924 Name
": confused... disk %d with serial \'%s\' "
925 "is not listed in the current anchor\n",
926 dl
->index
, dl
->serial
);
936 static void imsm_copy_dev(struct imsm_dev
*dest
, struct imsm_dev
*src
)
938 memcpy(dest
, src
, sizeof_imsm_dev(src
, 0));
941 static void dup_map(struct imsm_dev
*dev
)
943 struct imsm_map
*dest
= get_imsm_map(dev
, 1);
944 struct imsm_map
*src
= get_imsm_map(dev
, 0);
946 memcpy(dest
, src
, sizeof_imsm_map(src
));
949 static int parse_raid_devices(struct intel_super
*super
)
952 struct imsm_dev
*dev_new
;
955 for (i
= 0; i
< super
->anchor
->num_raid_devs
; i
++) {
956 struct imsm_dev
*dev_iter
= __get_imsm_dev(super
->anchor
, i
);
958 len
= sizeof_imsm_dev(dev_iter
, 1);
959 dev_new
= malloc(len
);
962 imsm_copy_dev(dev_new
, dev_iter
);
963 super
->dev_tbl
[i
] = dev_new
;
969 /* retrieve a pointer to the bbm log which starts after all raid devices */
970 struct bbm_log
*__get_imsm_bbm_log(struct imsm_super
*mpb
)
974 if (__le32_to_cpu(mpb
->bbm_log_size
)) {
976 ptr
+= mpb
->mpb_size
- __le32_to_cpu(mpb
->bbm_log_size
);
982 static void __free_imsm(struct intel_super
*super
, int free_disks
);
984 /* load_imsm_mpb - read matrix metadata
985 * allocates super->mpb to be freed by free_super
987 static int load_imsm_mpb(int fd
, struct intel_super
*super
, char *devname
)
989 unsigned long long dsize
;
990 unsigned long long sectors
;
992 struct imsm_super
*anchor
;
996 get_dev_size(fd
, NULL
, &dsize
);
998 if (lseek64(fd
, dsize
- (512 * 2), SEEK_SET
) < 0) {
1001 Name
": Cannot seek to anchor block on %s: %s\n",
1002 devname
, strerror(errno
));
1006 if (posix_memalign((void**)&anchor
, 512, 512) != 0) {
1009 Name
": Failed to allocate imsm anchor buffer"
1010 " on %s\n", devname
);
1013 if (read(fd
, anchor
, 512) != 512) {
1016 Name
": Cannot read anchor block on %s: %s\n",
1017 devname
, strerror(errno
));
1022 if (strncmp((char *) anchor
->sig
, MPB_SIGNATURE
, MPB_SIG_LEN
) != 0) {
1025 Name
": no IMSM anchor on %s\n", devname
);
1030 __free_imsm(super
, 0);
1031 super
->len
= __le32_to_cpu(anchor
->mpb_size
);
1032 super
->len
= ROUND_UP(anchor
->mpb_size
, 512);
1033 if (posix_memalign(&super
->buf
, 512, super
->len
) != 0) {
1036 Name
": unable to allocate %zu byte mpb buffer\n",
1041 memcpy(super
->buf
, anchor
, 512);
1043 sectors
= mpb_sectors(anchor
) - 1;
1046 rc
= load_imsm_disk(fd
, super
, devname
, 0);
1048 rc
= parse_raid_devices(super
);
1052 /* read the extended mpb */
1053 if (lseek64(fd
, dsize
- (512 * (2 + sectors
)), SEEK_SET
) < 0) {
1056 Name
": Cannot seek to extended mpb on %s: %s\n",
1057 devname
, strerror(errno
));
1061 if (read(fd
, super
->buf
+ 512, super
->len
- 512) != super
->len
- 512) {
1064 Name
": Cannot read extended mpb on %s: %s\n",
1065 devname
, strerror(errno
));
1069 check_sum
= __gen_imsm_checksum(super
->anchor
);
1070 if (check_sum
!= __le32_to_cpu(super
->anchor
->check_sum
)) {
1073 Name
": IMSM checksum %x != %x on %s\n",
1074 check_sum
, __le32_to_cpu(super
->anchor
->check_sum
),
1079 /* FIXME the BBM log is disk specific so we cannot use this global
1080 * buffer for all disks. Ok for now since we only look at the global
1081 * bbm_log_size parameter to gate assembly
1083 super
->bbm_log
= __get_imsm_bbm_log(super
->anchor
);
1085 rc
= load_imsm_disk(fd
, super
, devname
, 0);
1087 rc
= parse_raid_devices(super
);
1091 static void free_imsm_disks(struct intel_super
*super
)
1093 while (super
->disks
) {
1094 struct dl
*d
= super
->disks
;
1096 super
->disks
= d
->next
;
1105 /* free all the pieces hanging off of a super pointer */
1106 static void __free_imsm(struct intel_super
*super
, int free_disks
)
1115 free_imsm_disks(super
);
1116 for (i
= 0; i
< IMSM_MAX_RAID_DEVS
; i
++)
1117 if (super
->dev_tbl
[i
]) {
1118 free(super
->dev_tbl
[i
]);
1119 super
->dev_tbl
[i
] = NULL
;
1123 static void free_imsm(struct intel_super
*super
)
1125 __free_imsm(super
, 1);
1129 static void free_super_imsm(struct supertype
*st
)
1131 struct intel_super
*super
= st
->sb
;
1140 static struct intel_super
*alloc_super(int creating_imsm
)
1142 struct intel_super
*super
= malloc(sizeof(*super
));
1145 memset(super
, 0, sizeof(*super
));
1146 super
->creating_imsm
= creating_imsm
;
1147 super
->current_vol
= -1;
1154 static int load_super_imsm_all(struct supertype
*st
, int fd
, void **sbp
,
1155 char *devname
, int keep_fd
)
1158 struct intel_super
*super
;
1159 struct mdinfo
*sd
, *best
= NULL
;
1166 /* check if this disk is a member of an active array */
1167 sra
= sysfs_read(fd
, 0, GET_LEVEL
|GET_VERSION
|GET_DEVS
|GET_STATE
);
1171 if (sra
->array
.major_version
!= -1 ||
1172 sra
->array
.minor_version
!= -2 ||
1173 strcmp(sra
->text_version
, "imsm") != 0)
1176 super
= alloc_super(0);
1180 /* find the most up to date disk in this array, skipping spares */
1181 for (sd
= sra
->devs
; sd
; sd
= sd
->next
) {
1182 sprintf(nm
, "%d:%d", sd
->disk
.major
, sd
->disk
.minor
);
1183 dfd
= dev_open(nm
, keep_fd
? O_RDWR
: O_RDONLY
);
1188 rv
= load_imsm_mpb(dfd
, super
, NULL
);
1192 if (super
->anchor
->num_raid_devs
== 0)
1195 gen
= __le32_to_cpu(super
->anchor
->generation_num
);
1196 if (!best
|| gen
> bestgen
) {
1211 /* load the most up to date anchor */
1212 sprintf(nm
, "%d:%d", best
->disk
.major
, best
->disk
.minor
);
1213 dfd
= dev_open(nm
, O_RDONLY
);
1218 rv
= load_imsm_mpb(dfd
, super
, NULL
);
1225 /* re-parse the disk list with the current anchor */
1226 for (sd
= sra
->devs
; sd
; sd
= sd
->next
) {
1227 sprintf(nm
, "%d:%d", sd
->disk
.major
, sd
->disk
.minor
);
1228 dfd
= dev_open(nm
, keep_fd
? O_RDWR
: O_RDONLY
);
1233 load_imsm_disk(dfd
, super
, NULL
, keep_fd
);
1238 if (st
->subarray
[0]) {
1239 if (atoi(st
->subarray
) <= super
->anchor
->num_raid_devs
)
1240 super
->current_vol
= atoi(st
->subarray
);
1246 if (st
->ss
== NULL
) {
1247 st
->ss
= &super_imsm
;
1248 st
->minor_version
= 0;
1249 st
->max_devs
= IMSM_MAX_DEVICES
;
1250 st
->container_dev
= fd2devnum(fd
);
1257 static int load_super_imsm(struct supertype
*st
, int fd
, char *devname
)
1259 struct intel_super
*super
;
1263 if (load_super_imsm_all(st
, fd
, &st
->sb
, devname
, 1) == 0)
1266 if (st
->subarray
[0])
1267 return 1; /* FIXME */
1269 super
= alloc_super(0);
1272 Name
": malloc of %zu failed.\n",
1277 rv
= load_imsm_mpb(fd
, super
, devname
);
1282 Name
": Failed to load all information "
1283 "sections on %s\n", devname
);
1289 if (st
->ss
== NULL
) {
1290 st
->ss
= &super_imsm
;
1291 st
->minor_version
= 0;
1292 st
->max_devs
= IMSM_MAX_DEVICES
;
1298 static __u16
info_to_blocks_per_strip(mdu_array_info_t
*info
)
1300 if (info
->level
== 1)
1302 return info
->chunk_size
>> 9;
1305 static __u32
info_to_num_data_stripes(mdu_array_info_t
*info
)
1309 num_stripes
= (info
->size
* 2) / info_to_blocks_per_strip(info
);
1310 if (info
->level
== 1)
1316 static __u32
info_to_blocks_per_member(mdu_array_info_t
*info
)
1318 return (info
->size
* 2) & ~(info_to_blocks_per_strip(info
) - 1);
1321 static int init_super_imsm_volume(struct supertype
*st
, mdu_array_info_t
*info
,
1322 unsigned long long size
, char *name
,
1323 char *homehost
, int *uuid
)
1325 /* We are creating a volume inside a pre-existing container.
1326 * so st->sb is already set.
1328 struct intel_super
*super
= st
->sb
;
1329 struct imsm_super
*mpb
= super
->anchor
;
1330 struct imsm_dev
*dev
;
1331 struct imsm_vol
*vol
;
1332 struct imsm_map
*map
;
1333 int idx
= mpb
->num_raid_devs
;
1335 unsigned long long array_blocks
;
1337 size_t size_old
, size_new
;
1339 if (mpb
->num_raid_devs
>= 2) {
1340 fprintf(stderr
, Name
": This imsm-container already has the "
1341 "maximum of 2 volumes\n");
1345 /* ensure the mpb is large enough for the new data */
1346 size_old
= __le32_to_cpu(mpb
->mpb_size
);
1347 size_new
= disks_to_mpb_size(info
->nr_disks
);
1348 if (size_new
> size_old
) {
1350 size_t size_round
= ROUND_UP(size_new
, 512);
1352 if (posix_memalign(&mpb_new
, 512, size_round
) != 0) {
1353 fprintf(stderr
, Name
": could not allocate new mpb\n");
1356 memcpy(mpb_new
, mpb
, size_old
);
1359 super
->anchor
= mpb_new
;
1360 mpb
->mpb_size
= __cpu_to_le32(size_new
);
1361 memset(mpb_new
+ size_old
, 0, size_round
- size_old
);
1363 super
->current_vol
= idx
;
1364 /* when creating the first raid device in this container set num_disks
1365 * to zero, i.e. delete this spare and add raid member devices in
1366 * add_to_super_imsm_volume()
1368 if (super
->current_vol
== 0)
1370 sprintf(st
->subarray
, "%d", idx
);
1371 dev
= malloc(sizeof(*dev
) + sizeof(__u32
) * (info
->raid_disks
- 1));
1373 fprintf(stderr
, Name
": could not allocate raid device\n");
1376 strncpy((char *) dev
->volume
, name
, MAX_RAID_SERIAL_LEN
);
1377 array_blocks
= calc_array_size(info
->level
, info
->raid_disks
,
1378 info
->layout
, info
->chunk_size
,
1380 dev
->size_low
= __cpu_to_le32((__u32
) array_blocks
);
1381 dev
->size_high
= __cpu_to_le32((__u32
) (array_blocks
>> 32));
1382 dev
->status
= __cpu_to_le32(0);
1383 dev
->reserved_blocks
= __cpu_to_le32(0);
1385 vol
->migr_state
= 0;
1388 for (i
= 0; i
< idx
; i
++) {
1389 struct imsm_dev
*prev
= get_imsm_dev(super
, i
);
1390 struct imsm_map
*pmap
= get_imsm_map(prev
, 0);
1392 offset
+= __le32_to_cpu(pmap
->blocks_per_member
);
1393 offset
+= IMSM_RESERVED_SECTORS
;
1395 map
= get_imsm_map(dev
, 0);
1396 map
->pba_of_lba0
= __cpu_to_le32(offset
);
1397 map
->blocks_per_member
= __cpu_to_le32(info_to_blocks_per_member(info
));
1398 map
->blocks_per_strip
= __cpu_to_le16(info_to_blocks_per_strip(info
));
1399 map
->num_data_stripes
= __cpu_to_le32(info_to_num_data_stripes(info
));
1400 map
->map_state
= info
->level
? IMSM_T_STATE_UNINITIALIZED
:
1401 IMSM_T_STATE_NORMAL
;
1403 if (info
->level
== 1 && info
->raid_disks
> 2) {
1404 fprintf(stderr
, Name
": imsm does not support more than 2 disks"
1405 "in a raid1 volume\n");
1408 if (info
->level
== 10)
1409 map
->raid_level
= 1;
1411 map
->raid_level
= info
->level
;
1413 map
->num_members
= info
->raid_disks
;
1414 for (i
= 0; i
< map
->num_members
; i
++) {
1415 /* initialized in add_to_super */
1416 map
->disk_ord_tbl
[i
] = __cpu_to_le32(0);
1418 mpb
->num_raid_devs
++;
1419 super
->dev_tbl
[super
->current_vol
] = dev
;
1424 static int init_super_imsm(struct supertype
*st
, mdu_array_info_t
*info
,
1425 unsigned long long size
, char *name
,
1426 char *homehost
, int *uuid
)
1428 /* This is primarily called by Create when creating a new array.
1429 * We will then get add_to_super called for each component, and then
1430 * write_init_super called to write it out to each device.
1431 * For IMSM, Create can create on fresh devices or on a pre-existing
1433 * To create on a pre-existing array a different method will be called.
1434 * This one is just for fresh drives.
1436 struct intel_super
*super
;
1437 struct imsm_super
*mpb
;
1445 return init_super_imsm_volume(st
, info
, size
, name
, homehost
,
1448 super
= alloc_super(1);
1451 mpb_size
= disks_to_mpb_size(info
->nr_disks
);
1452 if (posix_memalign(&super
->buf
, 512, mpb_size
) != 0) {
1457 memset(mpb
, 0, mpb_size
);
1459 memcpy(mpb
->sig
, MPB_SIGNATURE
, strlen(MPB_SIGNATURE
));
1460 memcpy(mpb
->sig
+ strlen(MPB_SIGNATURE
), MPB_VERSION_RAID5
,
1461 strlen(MPB_VERSION_RAID5
));
1462 mpb
->mpb_size
= mpb_size
;
1468 static void add_to_super_imsm_volume(struct supertype
*st
, mdu_disk_info_t
*dk
,
1469 int fd
, char *devname
)
1471 struct intel_super
*super
= st
->sb
;
1472 struct imsm_super
*mpb
= super
->anchor
;
1474 struct imsm_dev
*dev
;
1475 struct imsm_map
*map
;
1478 dev
= get_imsm_dev(super
, super
->current_vol
);
1479 map
= get_imsm_map(dev
, 0);
1481 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
1482 if (dl
->major
== dk
->major
&&
1483 dl
->minor
== dk
->minor
)
1486 if (!dl
|| ! (dk
->state
& (1<<MD_DISK_SYNC
)))
1489 /* add a pristine spare to the metadata */
1490 if (dl
->index
< 0) {
1491 dl
->index
= super
->anchor
->num_disks
;
1492 super
->anchor
->num_disks
++;
1494 map
->disk_ord_tbl
[dk
->number
] = __cpu_to_le32(dl
->index
);
1495 status
= CONFIGURED_DISK
| USABLE_DISK
;
1496 dl
->disk
.status
= __cpu_to_le32(status
);
1498 /* if we are creating the first raid device update the family number */
1499 if (super
->current_vol
== 0) {
1501 struct imsm_dev
*_dev
= __get_imsm_dev(mpb
, 0);
1502 struct imsm_disk
*_disk
= __get_imsm_disk(mpb
, dl
->index
);
1506 sum
= __gen_imsm_checksum(mpb
);
1507 mpb
->family_num
= __cpu_to_le32(sum
);
1511 static void add_to_super_imsm(struct supertype
*st
, mdu_disk_info_t
*dk
,
1512 int fd
, char *devname
)
1514 struct intel_super
*super
= st
->sb
;
1516 unsigned long long size
;
1521 if (super
->current_vol
>= 0) {
1522 add_to_super_imsm_volume(st
, dk
, fd
, devname
);
1527 dd
= malloc(sizeof(*dd
));
1530 Name
": malloc failed %s:%d.\n", __func__
, __LINE__
);
1533 memset(dd
, 0, sizeof(*dd
));
1534 dd
->major
= major(stb
.st_rdev
);
1535 dd
->minor
= minor(stb
.st_rdev
);
1537 dd
->devname
= devname
? strdup(devname
) : NULL
;
1538 dd
->next
= super
->disks
;
1540 rv
= imsm_read_serial(fd
, devname
, dd
->serial
);
1543 Name
": failed to retrieve scsi serial, aborting\n");
1548 get_dev_size(fd
, NULL
, &size
);
1550 status
= USABLE_DISK
| SPARE_DISK
;
1551 strcpy((char *) dd
->disk
.serial
, (char *) dd
->serial
);
1552 dd
->disk
.total_blocks
= __cpu_to_le32(size
);
1553 dd
->disk
.status
= __cpu_to_le32(status
);
1554 if (sysfs_disk_to_scsi_id(fd
, &id
) == 0)
1555 dd
->disk
.scsi_id
= __cpu_to_le32(id
);
1557 dd
->disk
.scsi_id
= __cpu_to_le32(0);
1561 static int store_imsm_mpb(int fd
, struct intel_super
*super
);
1563 /* spare records have their own family number and do not have any defined raid
1566 static int write_super_imsm_spares(struct intel_super
*super
, int doclose
)
1568 struct imsm_super mpb_save
;
1569 struct imsm_super
*mpb
= super
->anchor
;
1574 mpb
->num_raid_devs
= 0;
1576 mpb
->mpb_size
= sizeof(struct imsm_super
);
1577 mpb
->generation_num
= __cpu_to_le32(1UL);
1579 for (d
= super
->disks
; d
; d
= d
->next
) {
1583 mpb
->disk
[0] = d
->disk
;
1584 sum
= __gen_imsm_checksum(mpb
);
1585 mpb
->family_num
= __cpu_to_le32(sum
);
1586 sum
= __gen_imsm_checksum(mpb
);
1587 mpb
->check_sum
= __cpu_to_le32(sum
);
1589 if (store_imsm_mpb(d
->fd
, super
)) {
1590 fprintf(stderr
, "%s: failed for device %d:%d %s\n",
1591 __func__
, d
->major
, d
->minor
, strerror(errno
));
1605 static int write_super_imsm(struct intel_super
*super
, int doclose
)
1607 struct imsm_super
*mpb
= super
->anchor
;
1614 __u32 mpb_size
= sizeof(struct imsm_super
) - sizeof(struct imsm_disk
);
1616 /* 'generation' is incremented everytime the metadata is written */
1617 generation
= __le32_to_cpu(mpb
->generation_num
);
1619 mpb
->generation_num
= __cpu_to_le32(generation
);
1621 for (d
= super
->disks
; d
; d
= d
->next
) {
1626 mpb
->disk
[d
->index
] = d
->disk
;
1627 mpb_size
+= sizeof(struct imsm_disk
);
1630 if (raid_disks
!= mpb
->num_disks
) {
1631 fprintf(stderr
, "%s: expected %d disks only found %d\n",
1632 __func__
, mpb
->num_disks
, raid_disks
);
1636 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
1637 struct imsm_dev
*dev
= __get_imsm_dev(mpb
, i
);
1639 imsm_copy_dev(dev
, super
->dev_tbl
[i
]);
1640 mpb_size
+= sizeof_imsm_dev(dev
, 0);
1642 mpb_size
+= __le32_to_cpu(mpb
->bbm_log_size
);
1643 mpb
->mpb_size
= __cpu_to_le32(mpb_size
);
1645 /* recalculate checksum */
1646 sum
= __gen_imsm_checksum(mpb
);
1647 mpb
->check_sum
= __cpu_to_le32(sum
);
1649 /* write the mpb for disks that compose raid devices */
1650 for (d
= super
->disks
; d
; d
= d
->next
) {
1653 if (store_imsm_mpb(d
->fd
, super
)) {
1654 fprintf(stderr
, "%s: failed for device %d:%d %s\n",
1655 __func__
, d
->major
, d
->minor
, strerror(errno
));
1665 return write_super_imsm_spares(super
, doclose
);
1670 static int write_init_super_imsm(struct supertype
*st
)
1672 if (st
->update_tail
) {
1673 /* queue the recently created array as a metadata update */
1675 struct imsm_update_create_array
*u
;
1676 struct intel_super
*super
= st
->sb
;
1677 struct imsm_dev
*dev
;
1680 if (super
->current_vol
< 0 ||
1681 !(dev
= get_imsm_dev(super
, super
->current_vol
))) {
1682 fprintf(stderr
, "%s: could not determine sub-array\n",
1688 len
= sizeof(*u
) - sizeof(*dev
) + sizeof_imsm_dev(dev
, 0);
1691 fprintf(stderr
, "%s: failed to allocate update buffer\n",
1696 u
->type
= update_create_array
;
1697 u
->dev_idx
= super
->current_vol
;
1698 imsm_copy_dev(&u
->dev
, dev
);
1699 append_metadata_update(st
, u
, len
);
1701 for (d
= super
->disks
; d
; d
= d
->next
) {
1708 return write_super_imsm(st
->sb
, 1);
1711 static int store_zero_imsm(struct supertype
*st
, int fd
)
1713 unsigned long long dsize
;
1716 get_dev_size(fd
, NULL
, &dsize
);
1718 /* first block is stored on second to last sector of the disk */
1719 if (lseek64(fd
, dsize
- (512 * 2), SEEK_SET
) < 0)
1722 if (posix_memalign(&buf
, 512, 512) != 0)
1725 memset(buf
, 0, 512);
1726 if (write(fd
, buf
, 512) != 512)
1731 static int validate_geometry_imsm_container(struct supertype
*st
, int level
,
1732 int layout
, int raiddisks
, int chunk
,
1733 unsigned long long size
, char *dev
,
1734 unsigned long long *freesize
,
1738 unsigned long long ldsize
;
1740 if (level
!= LEVEL_CONTAINER
)
1745 fd
= open(dev
, O_RDONLY
|O_EXCL
, 0);
1748 fprintf(stderr
, Name
": imsm: Cannot open %s: %s\n",
1749 dev
, strerror(errno
));
1752 if (!get_dev_size(fd
, dev
, &ldsize
)) {
1758 *freesize
= avail_size_imsm(st
, ldsize
>> 9);
1763 /* validate_geometry_imsm_volume - lifted from validate_geometry_ddf_bvd
1764 * FIX ME add ahci details
1766 static int validate_geometry_imsm_volume(struct supertype
*st
, int level
,
1767 int layout
, int raiddisks
, int chunk
,
1768 unsigned long long size
, char *dev
,
1769 unsigned long long *freesize
,
1773 struct intel_super
*super
= st
->sb
;
1775 unsigned long long pos
= 0;
1776 unsigned long long maxsize
;
1780 if (level
== LEVEL_CONTAINER
)
1783 if (level
== 1 && raiddisks
> 2) {
1785 fprintf(stderr
, Name
": imsm does not support more "
1786 "than 2 in a raid1 configuration\n");
1790 /* We must have the container info already read in. */
1795 /* General test: make sure there is space for
1796 * 'raiddisks' device extents of size 'size' at a given
1799 unsigned long long minsize
= size
*2 /* convert to blocks */;
1800 unsigned long long start_offset
= ~0ULL;
1803 minsize
= MPB_SECTOR_CNT
+ IMSM_RESERVED_SECTORS
;
1804 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
1809 e
= get_extents(super
, dl
);
1812 unsigned long long esize
;
1813 esize
= e
[i
].start
- pos
;
1814 if (esize
>= minsize
)
1816 if (found
&& start_offset
== ~0ULL) {
1819 } else if (found
&& pos
!= start_offset
) {
1823 pos
= e
[i
].start
+ e
[i
].size
;
1825 } while (e
[i
-1].size
);
1830 if (dcnt
< raiddisks
) {
1832 fprintf(stderr
, Name
": imsm: Not enough "
1833 "devices with space for this array "
1840 /* This device must be a member of the set */
1841 if (stat(dev
, &stb
) < 0)
1843 if ((S_IFMT
& stb
.st_mode
) != S_IFBLK
)
1845 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
1846 if (dl
->major
== major(stb
.st_rdev
) &&
1847 dl
->minor
== minor(stb
.st_rdev
))
1852 fprintf(stderr
, Name
": %s is not in the "
1853 "same imsm set\n", dev
);
1856 e
= get_extents(super
, dl
);
1860 unsigned long long esize
;
1861 esize
= e
[i
].start
- pos
;
1862 if (esize
>= maxsize
)
1864 pos
= e
[i
].start
+ e
[i
].size
;
1866 } while (e
[i
-1].size
);
1867 *freesize
= maxsize
;
1872 int imsm_bbm_log_size(struct imsm_super
*mpb
)
1874 return __le32_to_cpu(mpb
->bbm_log_size
);
1877 static int validate_geometry_imsm(struct supertype
*st
, int level
, int layout
,
1878 int raiddisks
, int chunk
, unsigned long long size
,
1879 char *dev
, unsigned long long *freesize
,
1885 /* if given unused devices create a container
1886 * if given given devices in a container create a member volume
1888 if (level
== LEVEL_CONTAINER
) {
1889 /* Must be a fresh device to add to a container */
1890 return validate_geometry_imsm_container(st
, level
, layout
,
1891 raiddisks
, chunk
, size
,
1897 /* creating in a given container */
1898 return validate_geometry_imsm_volume(st
, level
, layout
,
1899 raiddisks
, chunk
, size
,
1900 dev
, freesize
, verbose
);
1903 /* limit creation to the following levels */
1915 /* This device needs to be a device in an 'imsm' container */
1916 fd
= open(dev
, O_RDONLY
|O_EXCL
, 0);
1920 Name
": Cannot create this array on device %s\n",
1925 if (errno
!= EBUSY
|| (fd
= open(dev
, O_RDONLY
, 0)) < 0) {
1927 fprintf(stderr
, Name
": Cannot open %s: %s\n",
1928 dev
, strerror(errno
));
1931 /* Well, it is in use by someone, maybe an 'imsm' container. */
1932 cfd
= open_container(fd
);
1936 fprintf(stderr
, Name
": Cannot use %s: It is busy\n",
1940 sra
= sysfs_read(cfd
, 0, GET_VERSION
);
1942 if (sra
&& sra
->array
.major_version
== -1 &&
1943 strcmp(sra
->text_version
, "imsm") == 0) {
1944 /* This is a member of a imsm container. Load the container
1945 * and try to create a volume
1947 struct intel_super
*super
;
1949 if (load_super_imsm_all(st
, cfd
, (void **) &super
, NULL
, 1) == 0) {
1951 st
->container_dev
= fd2devnum(cfd
);
1953 return validate_geometry_imsm_volume(st
, level
, layout
,
1959 } else /* may belong to another container */
1965 static struct mdinfo
*container_content_imsm(struct supertype
*st
)
1967 /* Given a container loaded by load_super_imsm_all,
1968 * extract information about all the arrays into
1971 * For each imsm_dev create an mdinfo, fill it in,
1972 * then look for matching devices in super->disks
1973 * and create appropriate device mdinfo.
1975 struct intel_super
*super
= st
->sb
;
1976 struct imsm_super
*mpb
= super
->anchor
;
1977 struct mdinfo
*rest
= NULL
;
1980 /* do not assemble arrays that might have bad blocks */
1981 if (imsm_bbm_log_size(super
->anchor
)) {
1982 fprintf(stderr
, Name
": BBM log found in metadata. "
1983 "Cannot activate array(s).\n");
1987 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
1988 struct imsm_dev
*dev
= get_imsm_dev(super
, i
);
1989 struct imsm_vol
*vol
= &dev
->vol
;
1990 struct imsm_map
*map
= get_imsm_map(dev
, 0);
1991 struct mdinfo
*this;
1994 this = malloc(sizeof(*this));
1995 memset(this, 0, sizeof(*this));
1998 this->array
.level
= get_imsm_raid_level(map
);
1999 this->array
.raid_disks
= map
->num_members
;
2000 this->array
.layout
= imsm_level_to_layout(this->array
.level
);
2001 this->array
.md_minor
= -1;
2002 this->array
.ctime
= 0;
2003 this->array
.utime
= 0;
2004 this->array
.chunk_size
= __le16_to_cpu(map
->blocks_per_strip
) << 9;
2005 this->array
.state
= !vol
->dirty
;
2006 this->container_member
= i
;
2007 if (map
->map_state
== IMSM_T_STATE_UNINITIALIZED
||
2008 dev
->vol
.dirty
|| dev
->vol
.migr_state
)
2009 this->resync_start
= 0;
2011 this->resync_start
= ~0ULL;
2013 strncpy(this->name
, (char *) dev
->volume
, MAX_RAID_SERIAL_LEN
);
2014 this->name
[MAX_RAID_SERIAL_LEN
] = 0;
2016 sprintf(this->text_version
, "/%s/%d",
2017 devnum2devname(st
->container_dev
),
2018 this->container_member
);
2020 memset(this->uuid
, 0, sizeof(this->uuid
));
2022 this->component_size
= __le32_to_cpu(map
->blocks_per_member
);
2024 for (slot
= 0 ; slot
< map
->num_members
; slot
++) {
2025 struct mdinfo
*info_d
;
2032 idx
= get_imsm_disk_idx(map
, slot
);
2033 for (d
= super
->disks
; d
; d
= d
->next
)
2034 if (d
->index
== idx
)
2040 s
= d
? __le32_to_cpu(d
->disk
.status
) : 0;
2041 if (s
& FAILED_DISK
)
2043 if (!(s
& USABLE_DISK
))
2047 * if we skip some disks the array will be assmebled degraded;
2048 * reset resync start to avoid a dirty-degraded situation
2050 * FIXME handle dirty degraded
2052 if (skip
&& !dev
->vol
.dirty
)
2053 this->resync_start
= ~0ULL;
2057 info_d
= malloc(sizeof(*info_d
));
2059 fprintf(stderr
, Name
": failed to allocate disk"
2060 " for volume %s\n", (char *) dev
->volume
);
2065 memset(info_d
, 0, sizeof(*info_d
));
2066 info_d
->next
= this->devs
;
2067 this->devs
= info_d
;
2069 info_d
->disk
.number
= d
->index
;
2070 info_d
->disk
.major
= d
->major
;
2071 info_d
->disk
.minor
= d
->minor
;
2072 info_d
->disk
.raid_disk
= slot
;
2074 this->array
.working_disks
++;
2076 info_d
->events
= __le32_to_cpu(mpb
->generation_num
);
2077 info_d
->data_offset
= __le32_to_cpu(map
->pba_of_lba0
);
2078 info_d
->component_size
= __le32_to_cpu(map
->blocks_per_member
);
2080 strcpy(info_d
->name
, d
->devname
);
2089 static int imsm_open_new(struct supertype
*c
, struct active_array
*a
,
2092 struct intel_super
*super
= c
->sb
;
2093 struct imsm_super
*mpb
= super
->anchor
;
2095 if (atoi(inst
) >= mpb
->num_raid_devs
) {
2096 fprintf(stderr
, "%s: subarry index %d, out of range\n",
2097 __func__
, atoi(inst
));
2101 dprintf("imsm: open_new %s\n", inst
);
2102 a
->info
.container_member
= atoi(inst
);
2106 static __u8
imsm_check_degraded(struct intel_super
*super
, int n
, int failed
)
2108 struct imsm_dev
*dev
= get_imsm_dev(super
, n
);
2109 struct imsm_map
*map
= get_imsm_map(dev
, 0);
2112 return map
->map_state
== IMSM_T_STATE_UNINITIALIZED
?
2113 IMSM_T_STATE_UNINITIALIZED
: IMSM_T_STATE_NORMAL
;
2115 switch (get_imsm_raid_level(map
)) {
2117 return IMSM_T_STATE_FAILED
;
2120 if (failed
< map
->num_members
)
2121 return IMSM_T_STATE_DEGRADED
;
2123 return IMSM_T_STATE_FAILED
;
2128 * check to see if any mirrors have failed,
2129 * otherwise we are degraded
2131 int device_per_mirror
= 2; /* FIXME is this always the case?
2132 * and are they always adjacent?
2137 for (i
= 0; i
< map
->num_members
; i
++) {
2138 int idx
= get_imsm_disk_idx(map
, i
);
2139 struct imsm_disk
*disk
= get_imsm_disk(super
, idx
);
2141 if (__le32_to_cpu(disk
->status
) & FAILED_DISK
)
2144 if (failed
>= device_per_mirror
)
2145 return IMSM_T_STATE_FAILED
;
2147 /* reset 'failed' for next mirror set */
2148 if (!((i
+ 1) % device_per_mirror
))
2152 return IMSM_T_STATE_DEGRADED
;
2156 return IMSM_T_STATE_DEGRADED
;
2158 return IMSM_T_STATE_FAILED
;
2164 return map
->map_state
;
2167 static int imsm_count_failed(struct intel_super
*super
, struct imsm_map
*map
)
2171 struct imsm_disk
*disk
;
2173 for (i
= 0; i
< map
->num_members
; i
++) {
2174 int idx
= get_imsm_disk_idx(map
, i
);
2176 disk
= get_imsm_disk(super
, idx
);
2177 if (__le32_to_cpu(disk
->status
) & FAILED_DISK
)
2184 static void imsm_set_array_state(struct active_array
*a
, int consistent
)
2186 int inst
= a
->info
.container_member
;
2187 struct intel_super
*super
= a
->container
->sb
;
2188 struct imsm_dev
*dev
= get_imsm_dev(super
, inst
);
2189 struct imsm_map
*map
= get_imsm_map(dev
, 0);
2190 int dirty
= !consistent
;
2194 failed
= imsm_count_failed(super
, map
);
2195 map_state
= imsm_check_degraded(super
, inst
, failed
);
2197 if (consistent
&& !dev
->vol
.dirty
&&
2198 (dev
->vol
.migr_state
|| map_state
!= IMSM_T_STATE_NORMAL
))
2199 a
->resync_start
= 0ULL;
2201 if (a
->resync_start
== ~0ULL) {
2202 /* complete recovery or initial resync */
2203 if (map
->map_state
!= map_state
) {
2204 dprintf("imsm: map_state %d: %d\n",
2206 map
->map_state
= map_state
;
2207 super
->updates_pending
++;
2209 if (dev
->vol
.migr_state
) {
2210 dprintf("imsm: mark resync complete\n");
2211 dev
->vol
.migr_state
= 0;
2212 dev
->vol
.migr_type
= 0;
2213 super
->updates_pending
++;
2215 } else if (!dev
->vol
.migr_state
) {
2216 dprintf("imsm: mark '%s' (%llu)\n",
2217 failed
? "rebuild" : "initializing", a
->resync_start
);
2218 /* mark that we are rebuilding */
2219 map
->map_state
= failed
? map_state
: IMSM_T_STATE_NORMAL
;
2220 dev
->vol
.migr_state
= 1;
2221 dev
->vol
.migr_type
= failed
? 1 : 0;
2223 a
->check_degraded
= 1;
2224 super
->updates_pending
++;
2227 /* mark dirty / clean */
2228 if (dirty
!= dev
->vol
.dirty
) {
2229 dprintf("imsm: mark '%s' (%llu)\n",
2230 dirty
? "dirty" : "clean", a
->resync_start
);
2231 dev
->vol
.dirty
= dirty
;
2232 super
->updates_pending
++;
2236 static void imsm_set_disk(struct active_array
*a
, int n
, int state
)
2238 int inst
= a
->info
.container_member
;
2239 struct intel_super
*super
= a
->container
->sb
;
2240 struct imsm_dev
*dev
= get_imsm_dev(super
, inst
);
2241 struct imsm_map
*map
= get_imsm_map(dev
, 0);
2242 struct imsm_disk
*disk
;
2245 int new_failure
= 0;
2247 if (n
> map
->num_members
)
2248 fprintf(stderr
, "imsm: set_disk %d out of range 0..%d\n",
2249 n
, map
->num_members
- 1);
2254 dprintf("imsm: set_disk %d:%x\n", n
, state
);
2256 disk
= get_imsm_disk(super
, get_imsm_disk_idx(map
, n
));
2258 /* check for new failures */
2259 status
= __le32_to_cpu(disk
->status
);
2260 if ((state
& DS_FAULTY
) && !(status
& FAILED_DISK
)) {
2261 status
|= FAILED_DISK
;
2262 disk
->status
= __cpu_to_le32(status
);
2264 super
->updates_pending
++;
2266 /* check if in_sync */
2267 if ((state
& DS_INSYNC
) && !(status
& USABLE_DISK
)) {
2268 status
|= USABLE_DISK
;
2269 disk
->status
= __cpu_to_le32(status
);
2270 super
->updates_pending
++;
2273 /* the number of failures have changed, count up 'failed' to determine
2274 * degraded / failed status
2276 if (new_failure
&& map
->map_state
!= IMSM_T_STATE_FAILED
)
2277 failed
= imsm_count_failed(super
, map
);
2279 /* determine map_state based on failed or in_sync count */
2281 map
->map_state
= imsm_check_degraded(super
, inst
, failed
);
2282 else if (map
->map_state
== IMSM_T_STATE_DEGRADED
) {
2286 for (d
= a
->info
.devs
; d
; d
= d
->next
)
2287 if (d
->curr_state
& DS_INSYNC
)
2290 if (working
== a
->info
.array
.raid_disks
) {
2291 map
->map_state
= IMSM_T_STATE_NORMAL
;
2292 dev
->vol
.migr_state
= 0;
2293 dev
->vol
.migr_type
= 0;
2294 super
->updates_pending
++;
2299 static int store_imsm_mpb(int fd
, struct intel_super
*super
)
2301 struct imsm_super
*mpb
= super
->anchor
;
2302 __u32 mpb_size
= __le32_to_cpu(mpb
->mpb_size
);
2303 unsigned long long dsize
;
2304 unsigned long long sectors
;
2306 get_dev_size(fd
, NULL
, &dsize
);
2308 if (mpb_size
> 512) {
2309 /* -1 to account for anchor */
2310 sectors
= mpb_sectors(mpb
) - 1;
2312 /* write the extended mpb to the sectors preceeding the anchor */
2313 if (lseek64(fd
, dsize
- (512 * (2 + sectors
)), SEEK_SET
) < 0)
2316 if (write(fd
, super
->buf
+ 512, 512 * sectors
) != 512 * sectors
)
2320 /* first block is stored on second to last sector of the disk */
2321 if (lseek64(fd
, dsize
- (512 * 2), SEEK_SET
) < 0)
2324 if (write(fd
, super
->buf
, 512) != 512)
2330 static void imsm_sync_metadata(struct supertype
*container
)
2332 struct intel_super
*super
= container
->sb
;
2334 if (!super
->updates_pending
)
2337 write_super_imsm(super
, 0);
2339 super
->updates_pending
= 0;
2342 static struct dl
*imsm_readd(struct intel_super
*super
, int idx
, struct active_array
*a
)
2344 struct imsm_dev
*dev
= get_imsm_dev(super
, a
->info
.container_member
);
2345 struct imsm_map
*map
= get_imsm_map(dev
, 0);
2346 int i
= get_imsm_disk_idx(map
, idx
);
2349 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
2353 if (__le32_to_cpu(dl
->disk
.status
) & FAILED_DISK
)
2357 dprintf("%s: found %x:%x\n", __func__
, dl
->major
, dl
->minor
);
2362 static struct dl
*imsm_add_spare(struct intel_super
*super
, int idx
, struct active_array
*a
)
2364 struct imsm_dev
*dev
= get_imsm_dev(super
, a
->info
.container_member
);
2365 struct imsm_map
*map
= get_imsm_map(dev
, 0);
2366 unsigned long long esize
;
2367 unsigned long long pos
;
2376 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
2377 /* If in this array, skip */
2378 for (d
= a
->info
.devs
; d
; d
= d
->next
)
2379 if (d
->disk
.major
== dl
->major
&&
2380 d
->disk
.minor
== dl
->minor
) {
2381 dprintf("%x:%x already in array\n", dl
->major
, dl
->minor
);
2387 /* skip marked in use or failed drives */
2388 status
= __le32_to_cpu(dl
->disk
.status
);
2389 if (status
& FAILED_DISK
|| status
& CONFIGURED_DISK
) {
2390 dprintf("%x:%x status ( %s%s)\n",
2391 dl
->major
, dl
->minor
,
2392 status
& FAILED_DISK
? "failed " : "",
2393 status
& CONFIGURED_DISK
? "configured " : "");
2397 /* Does this unused device have the requisite free space?
2398 * We need a->info.component_size sectors
2400 ex
= get_extents(super
, dl
);
2402 dprintf("cannot get extents\n");
2408 array_start
= __le32_to_cpu(map
->pba_of_lba0
);
2411 /* check that we can start at pba_of_lba0 with
2412 * a->info.component_size of space
2414 esize
= ex
[j
].start
- pos
;
2415 if (array_start
>= pos
&&
2416 array_start
+ a
->info
.component_size
< ex
[j
].start
) {
2420 pos
= ex
[j
].start
+ ex
[j
].size
;
2423 } while (ex
[j
-1].size
);
2427 dprintf("%x:%x does not have %llu at %d\n",
2428 dl
->major
, dl
->minor
,
2429 a
->info
.component_size
,
2430 __le32_to_cpu(map
->pba_of_lba0
));
2440 static struct mdinfo
*imsm_activate_spare(struct active_array
*a
,
2441 struct metadata_update
**updates
)
2444 * Find a device with unused free space and use it to replace a
2445 * failed/vacant region in an array. We replace failed regions one a
2446 * array at a time. The result is that a new spare disk will be added
2447 * to the first failed array and after the monitor has finished
2448 * propagating failures the remainder will be consumed.
2450 * FIXME add a capability for mdmon to request spares from another
2454 struct intel_super
*super
= a
->container
->sb
;
2455 int inst
= a
->info
.container_member
;
2456 struct imsm_dev
*dev
= get_imsm_dev(super
, inst
);
2457 struct imsm_map
*map
= get_imsm_map(dev
, 0);
2458 int failed
= a
->info
.array
.raid_disks
;
2459 struct mdinfo
*rv
= NULL
;
2462 struct metadata_update
*mu
;
2464 struct imsm_update_activate_spare
*u
;
2468 for (d
= a
->info
.devs
; d
; d
= d
->next
) {
2469 if ((d
->curr_state
& DS_FAULTY
) &&
2471 /* wait for Removal to happen */
2473 if (d
->state_fd
>= 0)
2477 dprintf("imsm: activate spare: inst=%d failed=%d (%d) level=%d\n",
2478 inst
, failed
, a
->info
.array
.raid_disks
, a
->info
.array
.level
);
2479 if (imsm_check_degraded(super
, inst
, failed
) != IMSM_T_STATE_DEGRADED
)
2482 /* For each slot, if it is not working, find a spare */
2483 for (i
= 0; i
< a
->info
.array
.raid_disks
; i
++) {
2484 for (d
= a
->info
.devs
; d
; d
= d
->next
)
2485 if (d
->disk
.raid_disk
== i
)
2487 dprintf("found %d: %p %x\n", i
, d
, d
?d
->curr_state
:0);
2488 if (d
&& (d
->state_fd
>= 0))
2492 * OK, this device needs recovery. Try to re-add the previous
2493 * occupant of this slot, if this fails add a new spare
2495 dl
= imsm_readd(super
, i
, a
);
2497 dl
= imsm_add_spare(super
, i
, a
);
2501 /* found a usable disk with enough space */
2502 di
= malloc(sizeof(*di
));
2503 memset(di
, 0, sizeof(*di
));
2505 /* dl->index will be -1 in the case we are activating a
2506 * pristine spare. imsm_process_update() will create a
2507 * new index in this case. Once a disk is found to be
2508 * failed in all member arrays it is kicked from the
2511 di
->disk
.number
= dl
->index
;
2513 /* (ab)use di->devs to store a pointer to the device
2516 di
->devs
= (struct mdinfo
*) dl
;
2518 di
->disk
.raid_disk
= i
;
2519 di
->disk
.major
= dl
->major
;
2520 di
->disk
.minor
= dl
->minor
;
2522 di
->data_offset
= __le32_to_cpu(map
->pba_of_lba0
);
2523 di
->component_size
= a
->info
.component_size
;
2524 di
->container_member
= inst
;
2528 dprintf("%x:%x to be %d at %llu\n", dl
->major
, dl
->minor
,
2529 i
, di
->data_offset
);
2535 /* No spares found */
2537 /* Now 'rv' has a list of devices to return.
2538 * Create a metadata_update record to update the
2539 * disk_ord_tbl for the array
2541 mu
= malloc(sizeof(*mu
));
2542 mu
->buf
= malloc(sizeof(struct imsm_update_activate_spare
) * num_spares
);
2544 mu
->len
= sizeof(struct imsm_update_activate_spare
) * num_spares
;
2545 mu
->next
= *updates
;
2546 u
= (struct imsm_update_activate_spare
*) mu
->buf
;
2548 for (di
= rv
; di
; di
= di
->next
) {
2549 u
->type
= update_activate_spare
;
2550 u
->dl
= (struct dl
*) di
->devs
;
2552 u
->slot
= di
->disk
.raid_disk
;
2563 static int disks_overlap(struct imsm_map
*m1
, struct imsm_map
*m2
)
2569 for (i
= 0; i
< m1
->num_members
; i
++) {
2570 idx
= get_imsm_disk_idx(m1
, i
);
2571 for (j
= 0; j
< m2
->num_members
; j
++)
2572 if (idx
== get_imsm_disk_idx(m2
, j
))
2579 static void imsm_process_update(struct supertype
*st
,
2580 struct metadata_update
*update
)
2583 * crack open the metadata_update envelope to find the update record
2584 * update can be one of:
2585 * update_activate_spare - a spare device has replaced a failed
2586 * device in an array, update the disk_ord_tbl. If this disk is
2587 * present in all member arrays then also clear the SPARE_DISK
2590 struct intel_super
*super
= st
->sb
;
2591 struct imsm_super
*mpb
= super
->anchor
;
2592 enum imsm_update_type type
= *(enum imsm_update_type
*) update
->buf
;
2595 case update_activate_spare
: {
2596 struct imsm_update_activate_spare
*u
= (void *) update
->buf
;
2597 struct imsm_dev
*dev
= get_imsm_dev(super
, u
->array
);
2598 struct imsm_map
*map
= get_imsm_map(dev
, 0);
2599 struct active_array
*a
;
2600 struct imsm_disk
*disk
;
2607 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
2612 fprintf(stderr
, "error: imsm_activate_spare passed "
2613 "an unknown disk (index: %d serial: %s)\n",
2614 u
->dl
->index
, u
->dl
->serial
);
2618 super
->updates_pending
++;
2620 /* adding a pristine spare, assign a new index */
2621 if (dl
->index
< 0) {
2622 dl
->index
= super
->anchor
->num_disks
;
2623 super
->anchor
->num_disks
++;
2625 victim
= get_imsm_disk_idx(map
, u
->slot
);
2626 map
->disk_ord_tbl
[u
->slot
] = __cpu_to_le32(dl
->index
);
2628 status
= __le32_to_cpu(disk
->status
);
2629 status
|= CONFIGURED_DISK
;
2630 status
&= ~(SPARE_DISK
| USABLE_DISK
);
2631 disk
->status
= __cpu_to_le32(status
);
2633 /* count arrays using the victim in the metadata */
2635 for (a
= st
->arrays
; a
; a
= a
->next
) {
2636 dev
= get_imsm_dev(super
, a
->info
.container_member
);
2637 map
= get_imsm_map(dev
, 0);
2638 for (i
= 0; i
< map
->num_members
; i
++)
2639 if (victim
== get_imsm_disk_idx(map
, i
))
2643 /* clear some flags if the victim is no longer being
2647 disk
= get_imsm_disk(super
, victim
);
2648 status
= __le32_to_cpu(disk
->status
);
2649 status
&= ~(CONFIGURED_DISK
| USABLE_DISK
);
2650 disk
->status
= __cpu_to_le32(status
);
2651 /* at this point the disk can be removed from the
2652 * metadata, however we need to guarantee that we do
2653 * not race with any manager thread routine that relies
2654 * on dl->index or map->disk_ord_tbl
2659 case update_create_array
: {
2660 /* someone wants to create a new array, we need to be aware of
2661 * a few races/collisions:
2662 * 1/ 'Create' called by two separate instances of mdadm
2663 * 2/ 'Create' versus 'activate_spare': mdadm has chosen
2664 * devices that have since been assimilated via
2666 * In the event this update can not be carried out mdadm will
2667 * (FIX ME) notice that its update did not take hold.
2669 struct imsm_update_create_array
*u
= (void *) update
->buf
;
2670 struct imsm_dev
*dev
;
2671 struct imsm_map
*map
, *new_map
;
2672 unsigned long long start
, end
;
2673 unsigned long long new_start
, new_end
;
2677 /* handle racing creates: first come first serve */
2678 if (u
->dev_idx
< mpb
->num_raid_devs
) {
2679 dprintf("%s: subarray %d already defined\n",
2680 __func__
, u
->dev_idx
);
2684 /* check update is next in sequence */
2685 if (u
->dev_idx
!= mpb
->num_raid_devs
) {
2686 dprintf("%s: can not create array %d expected index %d\n",
2687 __func__
, u
->dev_idx
, mpb
->num_raid_devs
);
2691 new_map
= get_imsm_map(&u
->dev
, 0);
2692 new_start
= __le32_to_cpu(new_map
->pba_of_lba0
);
2693 new_end
= new_start
+ __le32_to_cpu(new_map
->blocks_per_member
);
2695 /* handle activate_spare versus create race:
2696 * check to make sure that overlapping arrays do not include
2699 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
2700 dev
= get_imsm_dev(super
, i
);
2701 map
= get_imsm_map(dev
, 0);
2702 start
= __le32_to_cpu(map
->pba_of_lba0
);
2703 end
= start
+ __le32_to_cpu(map
->blocks_per_member
);
2704 if ((new_start
>= start
&& new_start
<= end
) ||
2705 (start
>= new_start
&& start
<= new_end
))
2707 if (overlap
&& disks_overlap(map
, new_map
)) {
2708 dprintf("%s: arrays overlap\n", __func__
);
2712 /* check num_members sanity */
2713 if (new_map
->num_members
> mpb
->num_disks
) {
2714 dprintf("%s: num_disks out of range\n", __func__
);
2718 /* check that prepare update was successful */
2719 if (!update
->space
) {
2720 dprintf("%s: prepare update failed\n", __func__
);
2724 super
->updates_pending
++;
2725 dev
= update
->space
;
2726 update
->space
= NULL
;
2727 imsm_copy_dev(dev
, &u
->dev
);
2728 super
->dev_tbl
[u
->dev_idx
] = dev
;
2729 mpb
->num_raid_devs
++;
2731 /* fix up flags, if arrays overlap then the drives can not be
2734 for (i
= 0; i
< map
->num_members
; i
++) {
2735 struct imsm_disk
*disk
;
2738 disk
= get_imsm_disk(super
, get_imsm_disk_idx(map
, i
));
2739 status
= __le32_to_cpu(disk
->status
);
2740 status
|= CONFIGURED_DISK
;
2742 status
&= ~SPARE_DISK
;
2743 disk
->status
= __cpu_to_le32(status
);
2750 static void imsm_prepare_update(struct supertype
*st
,
2751 struct metadata_update
*update
)
2754 * Allocate space to hold new disk entries, raid-device entries or a
2755 * new mpb if necessary. We currently maintain an mpb large enough to
2756 * hold 2 subarrays for the given number of disks. This may not be
2757 * sufficient when reshaping.
2759 * FIX ME handle the reshape case.
2761 * The monitor will be able to safely change super->mpb by arranging
2762 * for it to be freed in check_update_queue(). I.e. the monitor thread
2763 * will start using the new pointer and the manager can continue to use
2764 * the old value until check_update_queue() runs.
2766 enum imsm_update_type type
= *(enum imsm_update_type
*) update
->buf
;
2769 case update_create_array
: {
2770 struct imsm_update_create_array
*u
= (void *) update
->buf
;
2771 size_t len
= sizeof_imsm_dev(&u
->dev
, 1);
2773 update
->space
= malloc(len
);
2783 struct superswitch super_imsm
= {
2785 .examine_super
= examine_super_imsm
,
2786 .brief_examine_super
= brief_examine_super_imsm
,
2787 .detail_super
= detail_super_imsm
,
2788 .brief_detail_super
= brief_detail_super_imsm
,
2789 .write_init_super
= write_init_super_imsm
,
2791 .match_home
= match_home_imsm
,
2792 .uuid_from_super
= uuid_from_super_imsm
,
2793 .getinfo_super
= getinfo_super_imsm
,
2794 .update_super
= update_super_imsm
,
2796 .avail_size
= avail_size_imsm
,
2798 .compare_super
= compare_super_imsm
,
2800 .load_super
= load_super_imsm
,
2801 .init_super
= init_super_imsm
,
2802 .add_to_super
= add_to_super_imsm
,
2803 .store_super
= store_zero_imsm
,
2804 .free_super
= free_super_imsm
,
2805 .match_metadata_desc
= match_metadata_desc_imsm
,
2806 .container_content
= container_content_imsm
,
2808 .validate_geometry
= validate_geometry_imsm
,
2812 .open_new
= imsm_open_new
,
2813 .load_super
= load_super_imsm
,
2814 .set_array_state
= imsm_set_array_state
,
2815 .set_disk
= imsm_set_disk
,
2816 .sync_metadata
= imsm_sync_metadata
,
2817 .activate_spare
= imsm_activate_spare
,
2818 .process_update
= imsm_process_update
,
2819 .prepare_update
= imsm_prepare_update
,