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 static size_t sizeof_imsm_dev(struct imsm_dev
*dev
)
295 size_t size
= sizeof(*dev
) - sizeof(struct imsm_map
) +
296 sizeof_imsm_map(get_imsm_map(dev
, 0));
298 /* migrating means an additional map */
299 if (dev
->vol
.migr_state
)
300 size
+= sizeof_imsm_map(get_imsm_map(dev
, 1));
305 static struct imsm_dev
*__get_imsm_dev(struct imsm_super
*mpb
, __u8 index
)
311 if (index
>= mpb
->num_raid_devs
)
314 /* devices start after all disks */
315 offset
= ((void *) &mpb
->disk
[mpb
->num_disks
]) - _mpb
;
317 for (i
= 0; i
<= index
; i
++)
319 return _mpb
+ offset
;
321 offset
+= sizeof_imsm_dev(_mpb
+ offset
);
326 static struct imsm_dev
*get_imsm_dev(struct intel_super
*super
, __u8 index
)
328 if (index
>= super
->anchor
->num_raid_devs
)
330 return super
->dev_tbl
[index
];
333 static __u32
get_imsm_disk_idx(struct imsm_map
*map
, int slot
)
335 __u32
*ord_tbl
= &map
->disk_ord_tbl
[slot
];
337 /* top byte is 'special' */
338 return __le32_to_cpu(*ord_tbl
& ~(0xff << 24));
341 static int get_imsm_raid_level(struct imsm_map
*map
)
343 if (map
->raid_level
== 1) {
344 if (map
->num_members
== 2)
350 return map
->raid_level
;
353 static int cmp_extent(const void *av
, const void *bv
)
355 const struct extent
*a
= av
;
356 const struct extent
*b
= bv
;
357 if (a
->start
< b
->start
)
359 if (a
->start
> b
->start
)
364 static struct extent
*get_extents(struct intel_super
*super
, struct dl
*dl
)
366 /* find a list of used extents on the given physical device */
367 struct extent
*rv
, *e
;
371 for (i
= 0; i
< super
->anchor
->num_raid_devs
; i
++) {
372 struct imsm_dev
*dev
= get_imsm_dev(super
, i
);
373 struct imsm_map
*map
= get_imsm_map(dev
, 0);
375 for (j
= 0; j
< map
->num_members
; j
++) {
376 __u32 index
= get_imsm_disk_idx(map
, j
);
378 if (index
== dl
->index
)
382 rv
= malloc(sizeof(struct extent
) * (memberships
+ 1));
387 for (i
= 0; i
< super
->anchor
->num_raid_devs
; i
++) {
388 struct imsm_dev
*dev
= get_imsm_dev(super
, i
);
389 struct imsm_map
*map
= get_imsm_map(dev
, 0);
391 for (j
= 0; j
< map
->num_members
; j
++) {
392 __u32 index
= get_imsm_disk_idx(map
, j
);
394 if (index
== dl
->index
) {
395 e
->start
= __le32_to_cpu(map
->pba_of_lba0
);
396 e
->size
= __le32_to_cpu(map
->blocks_per_member
);
401 qsort(rv
, memberships
, sizeof(*rv
), cmp_extent
);
403 e
->start
= __le32_to_cpu(dl
->disk
.total_blocks
) -
404 (MPB_SECTOR_CNT
+ IMSM_RESERVED_SECTORS
);
410 static void print_imsm_dev(struct imsm_dev
*dev
, int index
)
414 struct imsm_map
*map
= get_imsm_map(dev
, 0);
417 printf("[%s]:\n", dev
->volume
);
418 printf(" RAID Level : %d\n", get_imsm_raid_level(map
));
419 printf(" Members : %d\n", map
->num_members
);
420 for (slot
= 0; slot
< map
->num_members
; slot
++)
421 if (index
== get_imsm_disk_idx(map
, slot
))
423 if (slot
< map
->num_members
)
424 printf(" This Slot : %d\n", slot
);
426 printf(" This Slot : ?\n");
427 sz
= __le32_to_cpu(dev
->size_high
);
429 sz
+= __le32_to_cpu(dev
->size_low
);
430 printf(" Array Size : %llu%s\n", (unsigned long long)sz
,
431 human_size(sz
* 512));
432 sz
= __le32_to_cpu(map
->blocks_per_member
);
433 printf(" Per Dev Size : %llu%s\n", (unsigned long long)sz
,
434 human_size(sz
* 512));
435 printf(" Sector Offset : %u\n",
436 __le32_to_cpu(map
->pba_of_lba0
));
437 printf(" Num Stripes : %u\n",
438 __le32_to_cpu(map
->num_data_stripes
));
439 printf(" Chunk Size : %u KiB\n",
440 __le16_to_cpu(map
->blocks_per_strip
) / 2);
441 printf(" Reserved : %d\n", __le32_to_cpu(dev
->reserved_blocks
));
442 printf(" Migrate State : %s\n", dev
->vol
.migr_state
? "migrating" : "idle");
443 printf(" Dirty State : %s\n", dev
->vol
.dirty
? "dirty" : "clean");
444 printf(" Map State : %s\n", map_state_str
[map
->map_state
]);
447 static void print_imsm_disk(struct imsm_super
*mpb
, int index
)
449 struct imsm_disk
*disk
= __get_imsm_disk(mpb
, index
);
450 char str
[MAX_RAID_SERIAL_LEN
];
458 snprintf(str
, MAX_RAID_SERIAL_LEN
, "%s", disk
->serial
);
459 printf(" Disk%02d Serial : %s\n", index
, str
);
460 s
= __le32_to_cpu(disk
->status
);
461 printf(" State :%s%s%s%s\n", s
&SPARE_DISK
? " spare" : "",
462 s
&CONFIGURED_DISK
? " active" : "",
463 s
&FAILED_DISK
? " failed" : "",
464 s
&USABLE_DISK
? " usable" : "");
465 printf(" Id : %08x\n", __le32_to_cpu(disk
->scsi_id
));
466 sz
= __le32_to_cpu(disk
->total_blocks
) -
467 (MPB_SECTOR_CNT
+ IMSM_RESERVED_SECTORS
* mpb
->num_raid_devs
);
468 printf(" Usable Size : %llu%s\n", (unsigned long long)sz
,
469 human_size(sz
* 512));
472 static void examine_super_imsm(struct supertype
*st
, char *homehost
)
474 struct intel_super
*super
= st
->sb
;
475 struct imsm_super
*mpb
= super
->anchor
;
476 char str
[MAX_SIGNATURE_LENGTH
];
480 snprintf(str
, MPB_SIG_LEN
, "%s", mpb
->sig
);
481 printf(" Magic : %s\n", str
);
482 snprintf(str
, strlen(MPB_VERSION_RAID0
), "%s", get_imsm_version(mpb
));
483 printf(" Version : %s\n", get_imsm_version(mpb
));
484 printf(" Family : %08x\n", __le32_to_cpu(mpb
->family_num
));
485 printf(" Generation : %08x\n", __le32_to_cpu(mpb
->generation_num
));
486 sum
= __le32_to_cpu(mpb
->check_sum
);
487 printf(" Checksum : %08x %s\n", sum
,
488 __gen_imsm_checksum(mpb
) == sum
? "correct" : "incorrect");
489 printf(" MPB Sectors : %d\n", mpb_sectors(mpb
));
490 printf(" Disks : %d\n", mpb
->num_disks
);
491 printf(" RAID Devices : %d\n", mpb
->num_raid_devs
);
492 print_imsm_disk(mpb
, super
->disks
->index
);
493 if (super
->bbm_log
) {
494 struct bbm_log
*log
= super
->bbm_log
;
497 printf("Bad Block Management Log:\n");
498 printf(" Log Size : %d\n", __le32_to_cpu(mpb
->bbm_log_size
));
499 printf(" Signature : %x\n", __le32_to_cpu(log
->signature
));
500 printf(" Entry Count : %d\n", __le32_to_cpu(log
->entry_count
));
501 printf(" Spare Blocks : %d\n", __le32_to_cpu(log
->reserved_spare_block_count
));
502 printf(" First Spare : %llx\n", __le64_to_cpu(log
->first_spare_lba
));
504 for (i
= 0; i
< mpb
->num_raid_devs
; i
++)
505 print_imsm_dev(__get_imsm_dev(mpb
, i
), super
->disks
->index
);
506 for (i
= 0; i
< mpb
->num_disks
; i
++) {
507 if (i
== super
->disks
->index
)
509 print_imsm_disk(mpb
, i
);
513 static void brief_examine_super_imsm(struct supertype
*st
)
515 printf("ARRAY /dev/imsm metadata=imsm\n");
518 static void detail_super_imsm(struct supertype
*st
, char *homehost
)
520 printf("%s\n", __FUNCTION__
);
523 static void brief_detail_super_imsm(struct supertype
*st
)
525 printf("%s\n", __FUNCTION__
);
529 static int match_home_imsm(struct supertype
*st
, char *homehost
)
531 printf("%s\n", __FUNCTION__
);
536 static void uuid_from_super_imsm(struct supertype
*st
, int uuid
[4])
538 printf("%s\n", __FUNCTION__
);
543 get_imsm_numerical_version(struct imsm_super
*mpb
, int *m
, int *p
)
545 __u8
*v
= get_imsm_version(mpb
);
546 __u8
*end
= mpb
->sig
+ MAX_SIGNATURE_LENGTH
;
547 char major
[] = { 0, 0, 0 };
548 char minor
[] = { 0 ,0, 0 };
549 char patch
[] = { 0, 0, 0 };
550 char *ver_parse
[] = { major
, minor
, patch
};
554 while (*v
!= '\0' && v
< end
) {
555 if (*v
!= '.' && j
< 2)
556 ver_parse
[i
][j
++] = *v
;
564 *m
= strtol(minor
, NULL
, 0);
565 *p
= strtol(patch
, NULL
, 0);
569 static int imsm_level_to_layout(int level
)
577 return ALGORITHM_LEFT_ASYMMETRIC
;
579 return 0x102; //FIXME is this correct?
584 static void getinfo_super_imsm_volume(struct supertype
*st
, struct mdinfo
*info
)
586 struct intel_super
*super
= st
->sb
;
587 struct imsm_dev
*dev
= get_imsm_dev(super
, super
->current_vol
);
588 struct imsm_map
*map
= get_imsm_map(dev
, 0);
590 info
->container_member
= super
->current_vol
;
591 info
->array
.raid_disks
= map
->num_members
;
592 info
->array
.level
= get_imsm_raid_level(map
);
593 info
->array
.layout
= imsm_level_to_layout(info
->array
.level
);
594 info
->array
.md_minor
= -1;
595 info
->array
.ctime
= 0;
596 info
->array
.utime
= 0;
597 info
->array
.chunk_size
= __le16_to_cpu(map
->blocks_per_strip
* 512);
599 info
->data_offset
= __le32_to_cpu(map
->pba_of_lba0
);
600 info
->component_size
= __le32_to_cpu(map
->blocks_per_member
);
602 info
->disk
.major
= 0;
603 info
->disk
.minor
= 0;
605 sprintf(info
->text_version
, "/%s/%d",
606 devnum2devname(st
->container_dev
),
607 info
->container_member
);
611 static void getinfo_super_imsm(struct supertype
*st
, struct mdinfo
*info
)
613 struct intel_super
*super
= st
->sb
;
614 struct imsm_disk
*disk
;
617 if (super
->current_vol
>= 0) {
618 getinfo_super_imsm_volume(st
, info
);
622 /* Set raid_disks to zero so that Assemble will always pull in valid
625 info
->array
.raid_disks
= 0;
626 info
->array
.level
= LEVEL_CONTAINER
;
627 info
->array
.layout
= 0;
628 info
->array
.md_minor
= -1;
629 info
->array
.ctime
= 0; /* N/A for imsm */
630 info
->array
.utime
= 0;
631 info
->array
.chunk_size
= 0;
633 info
->disk
.major
= 0;
634 info
->disk
.minor
= 0;
635 info
->disk
.raid_disk
= -1;
636 info
->reshape_active
= 0;
637 strcpy(info
->text_version
, "imsm");
638 info
->disk
.number
= -1;
639 info
->disk
.state
= 0;
642 disk
= &super
->disks
->disk
;
643 info
->disk
.number
= super
->disks
->index
;
644 info
->disk
.raid_disk
= super
->disks
->index
;
645 info
->data_offset
= __le32_to_cpu(disk
->total_blocks
) -
646 (MPB_SECTOR_CNT
+ IMSM_RESERVED_SECTORS
);
647 info
->component_size
= MPB_SECTOR_CNT
+ IMSM_RESERVED_SECTORS
;
648 s
= __le32_to_cpu(disk
->status
);
649 info
->disk
.state
= s
& CONFIGURED_DISK
? (1 << MD_DISK_ACTIVE
) : 0;
650 info
->disk
.state
|= s
& FAILED_DISK
? (1 << MD_DISK_FAULTY
) : 0;
651 info
->disk
.state
|= s
& USABLE_DISK
? (1 << MD_DISK_SYNC
) : 0;
655 static int update_super_imsm(struct supertype
*st
, struct mdinfo
*info
,
656 char *update
, char *devname
, int verbose
,
657 int uuid_set
, char *homehost
)
661 /* For 'assemble' and 'force' we need to return non-zero if any
662 * change was made. For others, the return value is ignored.
663 * Update options are:
664 * force-one : This device looks a bit old but needs to be included,
665 * update age info appropriately.
666 * assemble: clear any 'faulty' flag to allow this device to
668 * force-array: Array is degraded but being forced, mark it clean
669 * if that will be needed to assemble it.
671 * newdev: not used ????
672 * grow: Array has gained a new device - this is currently for
674 * resync: mark as dirty so a resync will happen.
675 * name: update the name - preserving the homehost
677 * Following are not relevant for this imsm:
678 * sparc2.2 : update from old dodgey metadata
679 * super-minor: change the preferred_minor number
680 * summaries: update redundant counters.
681 * uuid: Change the uuid of the array to match watch is given
682 * homehost: update the recorded homehost
683 * _reshape_progress: record new reshape_progress position.
686 //struct intel_super *super = st->sb;
687 //struct imsm_super *mpb = super->mpb;
689 if (strcmp(update
, "grow") == 0) {
691 if (strcmp(update
, "resync") == 0) {
692 /* dev->vol.dirty = 1; */
695 /* IMSM has no concept of UUID or homehost */
700 static size_t disks_to_mpb_size(int disks
)
704 size
= sizeof(struct imsm_super
);
705 size
+= (disks
- 1) * sizeof(struct imsm_disk
);
706 size
+= 2 * sizeof(struct imsm_dev
);
707 /* up to 2 maps per raid device (-2 for imsm_maps in imsm_dev */
708 size
+= (4 - 2) * sizeof(struct imsm_map
);
709 /* 4 possible disk_ord_tbl's */
710 size
+= 4 * (disks
- 1) * sizeof(__u32
);
715 static __u64
avail_size_imsm(struct supertype
*st
, __u64 devsize
)
717 if (devsize
< (MPB_SECTOR_CNT
+ IMSM_RESERVED_SECTORS
))
720 return devsize
- (MPB_SECTOR_CNT
+ IMSM_RESERVED_SECTORS
);
723 static int compare_super_imsm(struct supertype
*st
, struct supertype
*tst
)
727 * 0 same, or first was empty, and second was copied
728 * 1 second had wrong number
732 struct intel_super
*first
= st
->sb
;
733 struct intel_super
*sec
= tst
->sb
;
741 if (memcmp(first
->anchor
->sig
, sec
->anchor
->sig
, MAX_SIGNATURE_LENGTH
) != 0)
744 /* if an anchor does not have num_raid_devs set then it is a free
747 if (first
->anchor
->num_raid_devs
> 0 &&
748 sec
->anchor
->num_raid_devs
> 0) {
749 if (first
->anchor
->family_num
!= sec
->anchor
->family_num
)
751 if (first
->anchor
->mpb_size
!= sec
->anchor
->mpb_size
)
753 if (first
->anchor
->check_sum
!= sec
->anchor
->check_sum
)
760 static void fd2devname(int fd
, char *name
)
769 if (fstat(fd
, &st
) != 0)
771 sprintf(path
, "/sys/dev/block/%d:%d",
772 major(st
.st_rdev
), minor(st
.st_rdev
));
774 rv
= readlink(path
, dname
, sizeof(dname
));
779 nm
= strrchr(dname
, '/');
781 snprintf(name
, MAX_RAID_SERIAL_LEN
, "/dev/%s", nm
);
785 extern int scsi_get_serial(int fd
, void *buf
, size_t buf_len
);
787 static int imsm_read_serial(int fd
, char *devname
,
788 __u8 serial
[MAX_RAID_SERIAL_LEN
])
790 unsigned char scsi_serial
[255];
795 memset(scsi_serial
, 0, sizeof(scsi_serial
));
797 if (imsm_env_devname_as_serial()) {
798 char name
[MAX_RAID_SERIAL_LEN
];
800 fd2devname(fd
, name
);
801 strcpy((char *) serial
, name
);
805 rv
= scsi_get_serial(fd
, scsi_serial
, sizeof(scsi_serial
));
810 Name
": Failed to retrieve serial for %s\n",
815 rsp_len
= scsi_serial
[3];
816 for (i
= 0, cnt
= 0; i
< rsp_len
; i
++) {
817 if (!isspace(scsi_serial
[4 + i
]))
818 serial
[cnt
++] = scsi_serial
[4 + i
];
819 if (cnt
== MAX_RAID_SERIAL_LEN
)
823 serial
[MAX_RAID_SERIAL_LEN
- 1] = '\0';
829 load_imsm_disk(int fd
, struct intel_super
*super
, char *devname
, int keep_fd
)
836 __u8 serial
[MAX_RAID_SERIAL_LEN
];
838 rv
= imsm_read_serial(fd
, devname
, serial
);
843 /* check if this is a disk we have seen before. it may be a spare in
844 * super->disks while the current anchor believes it is a raid member,
845 * check if we need to update dl->index
847 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
848 if (memcmp(dl
->serial
, serial
, MAX_RAID_SERIAL_LEN
) == 0)
852 dl
= malloc(sizeof(*dl
));
859 Name
": failed to allocate disk buffer for %s\n",
866 dl
->major
= major(stb
.st_rdev
);
867 dl
->minor
= minor(stb
.st_rdev
);
868 dl
->next
= super
->disks
;
869 dl
->fd
= keep_fd
? fd
: -1;
870 dl
->devname
= devname
? strdup(devname
) : NULL
;
871 strncpy((char *) dl
->serial
, (char *) serial
, MAX_RAID_SERIAL_LEN
);
872 } else if (keep_fd
) {
877 /* look up this disk's index in the current anchor */
878 for (i
= 0; i
< super
->anchor
->num_disks
; i
++) {
879 struct imsm_disk
*disk_iter
;
881 disk_iter
= __get_imsm_disk(super
->anchor
, i
);
883 if (memcmp(disk_iter
->serial
, dl
->serial
,
884 MAX_RAID_SERIAL_LEN
) == 0) {
887 dl
->disk
= *disk_iter
;
888 status
= __le32_to_cpu(dl
->disk
.status
);
889 /* only set index on disks that are a member of a
890 * populated contianer, i.e. one with raid_devs
892 if (status
& SPARE_DISK
)
900 if (i
== super
->anchor
->num_disks
&& alloc
) {
903 Name
": failed to load disk with serial \'%s\' for %s\n",
904 dl
->serial
, devname
);
908 if (i
== super
->anchor
->num_disks
&& dl
->index
>= 0) {
911 Name
": confused... disk %d with serial \'%s\' "
912 "is not listed in the current anchor\n",
913 dl
->index
, dl
->serial
);
923 static void imsm_copy_dev(struct imsm_dev
*dest
, struct imsm_dev
*src
)
925 memcpy(dest
, src
, sizeof_imsm_dev(src
));
928 static int parse_raid_devices(struct intel_super
*super
)
931 struct imsm_dev
*dev_new
;
934 for (i
= 0; i
< super
->anchor
->num_raid_devs
; i
++) {
935 struct imsm_dev
*dev_iter
= __get_imsm_dev(super
->anchor
, i
);
937 len
= sizeof_imsm_dev(dev_iter
);
938 dev_new
= malloc(len
);
941 imsm_copy_dev(dev_new
, dev_iter
);
942 super
->dev_tbl
[i
] = dev_new
;
948 /* retrieve a pointer to the bbm log which starts after all raid devices */
949 struct bbm_log
*__get_imsm_bbm_log(struct imsm_super
*mpb
)
953 if (__le32_to_cpu(mpb
->bbm_log_size
)) {
955 ptr
+= mpb
->mpb_size
- __le32_to_cpu(mpb
->bbm_log_size
);
961 static void __free_imsm(struct intel_super
*super
, int free_disks
);
963 /* load_imsm_mpb - read matrix metadata
964 * allocates super->mpb to be freed by free_super
966 static int load_imsm_mpb(int fd
, struct intel_super
*super
, char *devname
)
968 unsigned long long dsize
;
969 unsigned long long sectors
;
971 struct imsm_super
*anchor
;
975 get_dev_size(fd
, NULL
, &dsize
);
977 if (lseek64(fd
, dsize
- (512 * 2), SEEK_SET
) < 0) {
980 Name
": Cannot seek to anchor block on %s: %s\n",
981 devname
, strerror(errno
));
985 if (posix_memalign((void**)&anchor
, 512, 512) != 0) {
988 Name
": Failed to allocate imsm anchor buffer"
989 " on %s\n", devname
);
992 if (read(fd
, anchor
, 512) != 512) {
995 Name
": Cannot read anchor block on %s: %s\n",
996 devname
, strerror(errno
));
1001 if (strncmp((char *) anchor
->sig
, MPB_SIGNATURE
, MPB_SIG_LEN
) != 0) {
1004 Name
": no IMSM anchor on %s\n", devname
);
1009 __free_imsm(super
, 0);
1010 super
->len
= __le32_to_cpu(anchor
->mpb_size
);
1011 super
->len
= ROUND_UP(anchor
->mpb_size
, 512);
1012 if (posix_memalign(&super
->buf
, 512, super
->len
) != 0) {
1015 Name
": unable to allocate %zu byte mpb buffer\n",
1020 memcpy(super
->buf
, anchor
, 512);
1022 sectors
= mpb_sectors(anchor
) - 1;
1025 rc
= load_imsm_disk(fd
, super
, devname
, 0);
1027 rc
= parse_raid_devices(super
);
1031 /* read the extended mpb */
1032 if (lseek64(fd
, dsize
- (512 * (2 + sectors
)), SEEK_SET
) < 0) {
1035 Name
": Cannot seek to extended mpb on %s: %s\n",
1036 devname
, strerror(errno
));
1040 if (read(fd
, super
->buf
+ 512, super
->len
- 512) != super
->len
- 512) {
1043 Name
": Cannot read extended mpb on %s: %s\n",
1044 devname
, strerror(errno
));
1048 check_sum
= __gen_imsm_checksum(super
->anchor
);
1049 if (check_sum
!= __le32_to_cpu(super
->anchor
->check_sum
)) {
1052 Name
": IMSM checksum %x != %x on %s\n",
1053 check_sum
, __le32_to_cpu(super
->anchor
->check_sum
),
1058 /* FIXME the BBM log is disk specific so we cannot use this global
1059 * buffer for all disks. Ok for now since we only look at the global
1060 * bbm_log_size parameter to gate assembly
1062 super
->bbm_log
= __get_imsm_bbm_log(super
->anchor
);
1064 rc
= load_imsm_disk(fd
, super
, devname
, 0);
1066 rc
= parse_raid_devices(super
);
1070 static void free_imsm_disks(struct intel_super
*super
)
1072 while (super
->disks
) {
1073 struct dl
*d
= super
->disks
;
1075 super
->disks
= d
->next
;
1084 /* free all the pieces hanging off of a super pointer */
1085 static void __free_imsm(struct intel_super
*super
, int free_disks
)
1094 free_imsm_disks(super
);
1095 for (i
= 0; i
< IMSM_MAX_RAID_DEVS
; i
++)
1096 if (super
->dev_tbl
[i
]) {
1097 free(super
->dev_tbl
[i
]);
1098 super
->dev_tbl
[i
] = NULL
;
1102 static void free_imsm(struct intel_super
*super
)
1104 __free_imsm(super
, 1);
1108 static void free_super_imsm(struct supertype
*st
)
1110 struct intel_super
*super
= st
->sb
;
1119 static struct intel_super
*alloc_super(int creating_imsm
)
1121 struct intel_super
*super
= malloc(sizeof(*super
));
1124 memset(super
, 0, sizeof(*super
));
1125 super
->creating_imsm
= creating_imsm
;
1126 super
->current_vol
= -1;
1133 static int load_super_imsm_all(struct supertype
*st
, int fd
, void **sbp
,
1134 char *devname
, int keep_fd
)
1137 struct intel_super
*super
;
1138 struct mdinfo
*sd
, *best
= NULL
;
1145 /* check if this disk is a member of an active array */
1146 sra
= sysfs_read(fd
, 0, GET_LEVEL
|GET_VERSION
|GET_DEVS
|GET_STATE
);
1150 if (sra
->array
.major_version
!= -1 ||
1151 sra
->array
.minor_version
!= -2 ||
1152 strcmp(sra
->text_version
, "imsm") != 0)
1155 super
= alloc_super(0);
1159 /* find the most up to date disk in this array, skipping spares */
1160 for (sd
= sra
->devs
; sd
; sd
= sd
->next
) {
1161 sprintf(nm
, "%d:%d", sd
->disk
.major
, sd
->disk
.minor
);
1162 dfd
= dev_open(nm
, keep_fd
? O_RDWR
: O_RDONLY
);
1167 rv
= load_imsm_mpb(dfd
, super
, NULL
);
1171 if (super
->anchor
->num_raid_devs
== 0)
1174 gen
= __le32_to_cpu(super
->anchor
->generation_num
);
1175 if (!best
|| gen
> bestgen
) {
1190 /* load the most up to date anchor */
1191 sprintf(nm
, "%d:%d", best
->disk
.major
, best
->disk
.minor
);
1192 dfd
= dev_open(nm
, O_RDONLY
);
1197 rv
= load_imsm_mpb(dfd
, super
, NULL
);
1204 /* re-parse the disk list with the current anchor */
1205 for (sd
= sra
->devs
; sd
; sd
= sd
->next
) {
1206 sprintf(nm
, "%d:%d", sd
->disk
.major
, sd
->disk
.minor
);
1207 dfd
= dev_open(nm
, keep_fd
? O_RDWR
: O_RDONLY
);
1212 load_imsm_disk(dfd
, super
, NULL
, keep_fd
);
1217 if (st
->subarray
[0]) {
1218 if (atoi(st
->subarray
) <= super
->anchor
->num_raid_devs
)
1219 super
->current_vol
= atoi(st
->subarray
);
1225 if (st
->ss
== NULL
) {
1226 st
->ss
= &super_imsm
;
1227 st
->minor_version
= 0;
1228 st
->max_devs
= IMSM_MAX_DEVICES
;
1229 st
->container_dev
= fd2devnum(fd
);
1236 static int load_super_imsm(struct supertype
*st
, int fd
, char *devname
)
1238 struct intel_super
*super
;
1242 if (load_super_imsm_all(st
, fd
, &st
->sb
, devname
, 1) == 0)
1245 if (st
->subarray
[0])
1246 return 1; /* FIXME */
1248 super
= alloc_super(0);
1251 Name
": malloc of %zu failed.\n",
1256 rv
= load_imsm_mpb(fd
, super
, devname
);
1261 Name
": Failed to load all information "
1262 "sections on %s\n", devname
);
1268 if (st
->ss
== NULL
) {
1269 st
->ss
= &super_imsm
;
1270 st
->minor_version
= 0;
1271 st
->max_devs
= IMSM_MAX_DEVICES
;
1277 static __u16
info_to_blocks_per_strip(mdu_array_info_t
*info
)
1279 if (info
->level
== 1)
1281 return info
->chunk_size
>> 9;
1284 static __u32
info_to_num_data_stripes(mdu_array_info_t
*info
)
1288 num_stripes
= (info
->size
* 2) / info_to_blocks_per_strip(info
);
1289 if (info
->level
== 1)
1295 static __u32
info_to_blocks_per_member(mdu_array_info_t
*info
)
1297 return (info
->size
* 2) & ~(info_to_blocks_per_strip(info
) - 1);
1300 static int init_super_imsm_volume(struct supertype
*st
, mdu_array_info_t
*info
,
1301 unsigned long long size
, char *name
,
1302 char *homehost
, int *uuid
)
1304 /* We are creating a volume inside a pre-existing container.
1305 * so st->sb is already set.
1307 struct intel_super
*super
= st
->sb
;
1308 struct imsm_super
*mpb
= super
->anchor
;
1309 struct imsm_dev
*dev
;
1310 struct imsm_vol
*vol
;
1311 struct imsm_map
*map
;
1312 int idx
= mpb
->num_raid_devs
;
1314 unsigned long long array_blocks
;
1316 size_t size_old
, size_new
;
1318 if (mpb
->num_raid_devs
>= 2) {
1319 fprintf(stderr
, Name
": This imsm-container already has the "
1320 "maximum of 2 volumes\n");
1324 /* ensure the mpb is large enough for the new data */
1325 size_old
= __le32_to_cpu(mpb
->mpb_size
);
1326 size_new
= disks_to_mpb_size(info
->nr_disks
);
1327 if (size_new
> size_old
) {
1329 size_t size_round
= ROUND_UP(size_new
, 512);
1331 if (posix_memalign(&mpb_new
, 512, size_round
) != 0) {
1332 fprintf(stderr
, Name
": could not allocate new mpb\n");
1335 memcpy(mpb_new
, mpb
, size_old
);
1338 super
->anchor
= mpb_new
;
1339 mpb
->mpb_size
= __cpu_to_le32(size_new
);
1340 memset(mpb_new
+ size_old
, 0, size_round
- size_old
);
1342 super
->current_vol
= idx
;
1343 /* when creating the first raid device in this container set num_disks
1344 * to zero, i.e. delete this spare and add raid member devices in
1345 * add_to_super_imsm_volume()
1347 if (super
->current_vol
== 0)
1349 sprintf(st
->subarray
, "%d", idx
);
1350 dev
= malloc(sizeof(*dev
) + sizeof(__u32
) * (info
->raid_disks
- 1));
1352 fprintf(stderr
, Name
": could not allocate raid device\n");
1355 strncpy((char *) dev
->volume
, name
, MAX_RAID_SERIAL_LEN
);
1356 array_blocks
= calc_array_size(info
->level
, info
->raid_disks
,
1357 info
->layout
, info
->chunk_size
,
1359 dev
->size_low
= __cpu_to_le32((__u32
) array_blocks
);
1360 dev
->size_high
= __cpu_to_le32((__u32
) (array_blocks
>> 32));
1361 dev
->status
= __cpu_to_le32(0);
1362 dev
->reserved_blocks
= __cpu_to_le32(0);
1364 vol
->migr_state
= 0;
1367 for (i
= 0; i
< idx
; i
++) {
1368 struct imsm_dev
*prev
= get_imsm_dev(super
, i
);
1369 struct imsm_map
*pmap
= get_imsm_map(prev
, 0);
1371 offset
+= __le32_to_cpu(pmap
->blocks_per_member
);
1372 offset
+= IMSM_RESERVED_SECTORS
;
1374 map
= get_imsm_map(dev
, 0);
1375 map
->pba_of_lba0
= __cpu_to_le32(offset
);
1376 map
->blocks_per_member
= __cpu_to_le32(info_to_blocks_per_member(info
));
1377 map
->blocks_per_strip
= __cpu_to_le16(info_to_blocks_per_strip(info
));
1378 map
->num_data_stripes
= __cpu_to_le32(info_to_num_data_stripes(info
));
1379 map
->map_state
= info
->level
? IMSM_T_STATE_UNINITIALIZED
:
1380 IMSM_T_STATE_NORMAL
;
1382 if (info
->level
== 1 && info
->raid_disks
> 2) {
1383 fprintf(stderr
, Name
": imsm does not support more than 2 disks"
1384 "in a raid1 volume\n");
1387 if (info
->level
== 10)
1388 map
->raid_level
= 1;
1390 map
->raid_level
= info
->level
;
1392 map
->num_members
= info
->raid_disks
;
1393 for (i
= 0; i
< map
->num_members
; i
++) {
1394 /* initialized in add_to_super */
1395 map
->disk_ord_tbl
[i
] = __cpu_to_le32(0);
1397 mpb
->num_raid_devs
++;
1398 super
->dev_tbl
[super
->current_vol
] = dev
;
1403 static int init_super_imsm(struct supertype
*st
, mdu_array_info_t
*info
,
1404 unsigned long long size
, char *name
,
1405 char *homehost
, int *uuid
)
1407 /* This is primarily called by Create when creating a new array.
1408 * We will then get add_to_super called for each component, and then
1409 * write_init_super called to write it out to each device.
1410 * For IMSM, Create can create on fresh devices or on a pre-existing
1412 * To create on a pre-existing array a different method will be called.
1413 * This one is just for fresh drives.
1415 struct intel_super
*super
;
1416 struct imsm_super
*mpb
;
1424 return init_super_imsm_volume(st
, info
, size
, name
, homehost
,
1427 super
= alloc_super(1);
1430 mpb_size
= disks_to_mpb_size(info
->nr_disks
);
1431 if (posix_memalign(&super
->buf
, 512, mpb_size
) != 0) {
1436 memset(mpb
, 0, mpb_size
);
1438 memcpy(mpb
->sig
, MPB_SIGNATURE
, strlen(MPB_SIGNATURE
));
1439 memcpy(mpb
->sig
+ strlen(MPB_SIGNATURE
), MPB_VERSION_RAID5
,
1440 strlen(MPB_VERSION_RAID5
));
1441 mpb
->mpb_size
= mpb_size
;
1447 static void add_to_super_imsm_volume(struct supertype
*st
, mdu_disk_info_t
*dk
,
1448 int fd
, char *devname
)
1450 struct intel_super
*super
= st
->sb
;
1451 struct imsm_super
*mpb
= super
->anchor
;
1453 struct imsm_dev
*dev
;
1454 struct imsm_map
*map
;
1457 dev
= get_imsm_dev(super
, super
->current_vol
);
1458 map
= get_imsm_map(dev
, 0);
1460 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
1461 if (dl
->major
== dk
->major
&&
1462 dl
->minor
== dk
->minor
)
1465 if (!dl
|| ! (dk
->state
& (1<<MD_DISK_SYNC
)))
1468 /* add a pristine spare to the metadata */
1469 if (dl
->index
< 0) {
1470 dl
->index
= super
->anchor
->num_disks
;
1471 super
->anchor
->num_disks
++;
1473 map
->disk_ord_tbl
[dk
->number
] = __cpu_to_le32(dl
->index
);
1474 status
= CONFIGURED_DISK
| USABLE_DISK
;
1475 dl
->disk
.status
= __cpu_to_le32(status
);
1477 /* if we are creating the first raid device update the family number */
1478 if (super
->current_vol
== 0) {
1480 struct imsm_dev
*_dev
= __get_imsm_dev(mpb
, 0);
1481 struct imsm_disk
*_disk
= __get_imsm_disk(mpb
, dl
->index
);
1485 sum
= __gen_imsm_checksum(mpb
);
1486 mpb
->family_num
= __cpu_to_le32(sum
);
1490 static void add_to_super_imsm(struct supertype
*st
, mdu_disk_info_t
*dk
,
1491 int fd
, char *devname
)
1493 struct intel_super
*super
= st
->sb
;
1495 unsigned long long size
;
1500 if (super
->current_vol
>= 0) {
1501 add_to_super_imsm_volume(st
, dk
, fd
, devname
);
1506 dd
= malloc(sizeof(*dd
));
1509 Name
": malloc failed %s:%d.\n", __func__
, __LINE__
);
1512 memset(dd
, 0, sizeof(*dd
));
1513 dd
->major
= major(stb
.st_rdev
);
1514 dd
->minor
= minor(stb
.st_rdev
);
1516 dd
->devname
= devname
? strdup(devname
) : NULL
;
1517 dd
->next
= super
->disks
;
1519 rv
= imsm_read_serial(fd
, devname
, dd
->serial
);
1522 Name
": failed to retrieve scsi serial, aborting\n");
1527 get_dev_size(fd
, NULL
, &size
);
1529 status
= USABLE_DISK
| SPARE_DISK
;
1530 strcpy((char *) dd
->disk
.serial
, (char *) dd
->serial
);
1531 dd
->disk
.total_blocks
= __cpu_to_le32(size
);
1532 dd
->disk
.status
= __cpu_to_le32(status
);
1533 if (sysfs_disk_to_scsi_id(fd
, &id
) == 0)
1534 dd
->disk
.scsi_id
= __cpu_to_le32(id
);
1536 dd
->disk
.scsi_id
= __cpu_to_le32(0);
1540 static int store_imsm_mpb(int fd
, struct intel_super
*super
);
1542 /* spare records have their own family number and do not have any defined raid
1545 static int write_super_imsm_spares(struct intel_super
*super
, int doclose
)
1547 struct imsm_super mpb_save
;
1548 struct imsm_super
*mpb
= super
->anchor
;
1553 mpb
->num_raid_devs
= 0;
1555 mpb
->mpb_size
= sizeof(struct imsm_super
);
1556 mpb
->generation_num
= __cpu_to_le32(1UL);
1558 for (d
= super
->disks
; d
; d
= d
->next
) {
1562 mpb
->disk
[0] = d
->disk
;
1563 sum
= __gen_imsm_checksum(mpb
);
1564 mpb
->family_num
= __cpu_to_le32(sum
);
1565 sum
= __gen_imsm_checksum(mpb
);
1566 mpb
->check_sum
= __cpu_to_le32(sum
);
1568 if (store_imsm_mpb(d
->fd
, super
)) {
1569 fprintf(stderr
, "%s: failed for device %d:%d %s\n",
1570 __func__
, d
->major
, d
->minor
, strerror(errno
));
1584 static int write_super_imsm(struct intel_super
*super
, int doclose
)
1586 struct imsm_super
*mpb
= super
->anchor
;
1594 /* 'generation' is incremented everytime the metadata is written */
1595 generation
= __le32_to_cpu(mpb
->generation_num
);
1597 mpb
->generation_num
= __cpu_to_le32(generation
);
1599 for (d
= super
->disks
; d
; d
= d
->next
) {
1604 mpb
->disk
[d
->index
] = d
->disk
;
1607 if (raid_disks
!= mpb
->num_disks
) {
1608 fprintf(stderr
, "%s: expected %d disks only found %d\n",
1609 __func__
, mpb
->num_disks
, raid_disks
);
1613 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
1614 struct imsm_dev
*dev
= __get_imsm_dev(mpb
, i
);
1616 imsm_copy_dev(dev
, super
->dev_tbl
[i
]);
1619 /* recalculate checksum */
1620 sum
= __gen_imsm_checksum(mpb
);
1621 mpb
->check_sum
= __cpu_to_le32(sum
);
1623 /* write the mpb for disks that compose raid devices */
1624 for (d
= super
->disks
; d
; d
= d
->next
) {
1627 if (store_imsm_mpb(d
->fd
, super
)) {
1628 fprintf(stderr
, "%s: failed for device %d:%d %s\n",
1629 __func__
, d
->major
, d
->minor
, strerror(errno
));
1639 return write_super_imsm_spares(super
, doclose
);
1644 static int write_init_super_imsm(struct supertype
*st
)
1646 if (st
->update_tail
) {
1647 /* queue the recently created array as a metadata update */
1649 struct imsm_update_create_array
*u
;
1650 struct intel_super
*super
= st
->sb
;
1651 struct imsm_dev
*dev
;
1654 if (super
->current_vol
< 0 ||
1655 !(dev
= get_imsm_dev(super
, super
->current_vol
))) {
1656 fprintf(stderr
, "%s: could not determine sub-array\n",
1662 len
= sizeof(*u
) - sizeof(*dev
) + sizeof_imsm_dev(dev
);
1665 fprintf(stderr
, "%s: failed to allocate update buffer\n",
1670 u
->type
= update_create_array
;
1671 u
->dev_idx
= super
->current_vol
;
1672 imsm_copy_dev(&u
->dev
, dev
);
1673 append_metadata_update(st
, u
, len
);
1675 for (d
= super
->disks
; d
; d
= d
->next
) {
1682 return write_super_imsm(st
->sb
, 1);
1685 static int store_zero_imsm(struct supertype
*st
, int fd
)
1687 unsigned long long dsize
;
1690 get_dev_size(fd
, NULL
, &dsize
);
1692 /* first block is stored on second to last sector of the disk */
1693 if (lseek64(fd
, dsize
- (512 * 2), SEEK_SET
) < 0)
1696 if (posix_memalign(&buf
, 512, 512) != 0)
1699 memset(buf
, 0, 512);
1700 if (write(fd
, buf
, 512) != 512)
1705 static int validate_geometry_imsm_container(struct supertype
*st
, int level
,
1706 int layout
, int raiddisks
, int chunk
,
1707 unsigned long long size
, char *dev
,
1708 unsigned long long *freesize
,
1712 unsigned long long ldsize
;
1714 if (level
!= LEVEL_CONTAINER
)
1719 fd
= open(dev
, O_RDONLY
|O_EXCL
, 0);
1722 fprintf(stderr
, Name
": imsm: Cannot open %s: %s\n",
1723 dev
, strerror(errno
));
1726 if (!get_dev_size(fd
, dev
, &ldsize
)) {
1732 *freesize
= avail_size_imsm(st
, ldsize
>> 9);
1737 /* validate_geometry_imsm_volume - lifted from validate_geometry_ddf_bvd
1738 * FIX ME add ahci details
1740 static int validate_geometry_imsm_volume(struct supertype
*st
, int level
,
1741 int layout
, int raiddisks
, int chunk
,
1742 unsigned long long size
, char *dev
,
1743 unsigned long long *freesize
,
1747 struct intel_super
*super
= st
->sb
;
1749 unsigned long long pos
= 0;
1750 unsigned long long maxsize
;
1754 if (level
== LEVEL_CONTAINER
)
1757 if (level
== 1 && raiddisks
> 2) {
1759 fprintf(stderr
, Name
": imsm does not support more "
1760 "than 2 in a raid1 configuration\n");
1764 /* We must have the container info already read in. */
1769 /* General test: make sure there is space for
1770 * 'raiddisks' device extents of size 'size' at a given
1773 unsigned long long minsize
= size
*2 /* convert to blocks */;
1774 unsigned long long start_offset
= ~0ULL;
1777 minsize
= MPB_SECTOR_CNT
+ IMSM_RESERVED_SECTORS
;
1778 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
1783 e
= get_extents(super
, dl
);
1786 unsigned long long esize
;
1787 esize
= e
[i
].start
- pos
;
1788 if (esize
>= minsize
)
1790 if (found
&& start_offset
== ~0ULL) {
1793 } else if (found
&& pos
!= start_offset
) {
1797 pos
= e
[i
].start
+ e
[i
].size
;
1799 } while (e
[i
-1].size
);
1804 if (dcnt
< raiddisks
) {
1806 fprintf(stderr
, Name
": imsm: Not enough "
1807 "devices with space for this array "
1814 /* This device must be a member of the set */
1815 if (stat(dev
, &stb
) < 0)
1817 if ((S_IFMT
& stb
.st_mode
) != S_IFBLK
)
1819 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
1820 if (dl
->major
== major(stb
.st_rdev
) &&
1821 dl
->minor
== minor(stb
.st_rdev
))
1826 fprintf(stderr
, Name
": %s is not in the "
1827 "same imsm set\n", dev
);
1830 e
= get_extents(super
, dl
);
1834 unsigned long long esize
;
1835 esize
= e
[i
].start
- pos
;
1836 if (esize
>= maxsize
)
1838 pos
= e
[i
].start
+ e
[i
].size
;
1840 } while (e
[i
-1].size
);
1841 *freesize
= maxsize
;
1846 int imsm_bbm_log_size(struct imsm_super
*mpb
)
1848 return __le32_to_cpu(mpb
->bbm_log_size
);
1851 static int validate_geometry_imsm(struct supertype
*st
, int level
, int layout
,
1852 int raiddisks
, int chunk
, unsigned long long size
,
1853 char *dev
, unsigned long long *freesize
,
1859 /* if given unused devices create a container
1860 * if given given devices in a container create a member volume
1862 if (level
== LEVEL_CONTAINER
) {
1863 /* Must be a fresh device to add to a container */
1864 return validate_geometry_imsm_container(st
, level
, layout
,
1865 raiddisks
, chunk
, size
,
1871 /* creating in a given container */
1872 return validate_geometry_imsm_volume(st
, level
, layout
,
1873 raiddisks
, chunk
, size
,
1874 dev
, freesize
, verbose
);
1877 /* limit creation to the following levels */
1889 /* This device needs to be a device in an 'imsm' container */
1890 fd
= open(dev
, O_RDONLY
|O_EXCL
, 0);
1894 Name
": Cannot create this array on device %s\n",
1899 if (errno
!= EBUSY
|| (fd
= open(dev
, O_RDONLY
, 0)) < 0) {
1901 fprintf(stderr
, Name
": Cannot open %s: %s\n",
1902 dev
, strerror(errno
));
1905 /* Well, it is in use by someone, maybe an 'imsm' container. */
1906 cfd
= open_container(fd
);
1910 fprintf(stderr
, Name
": Cannot use %s: It is busy\n",
1914 sra
= sysfs_read(cfd
, 0, GET_VERSION
);
1916 if (sra
&& sra
->array
.major_version
== -1 &&
1917 strcmp(sra
->text_version
, "imsm") == 0) {
1918 /* This is a member of a imsm container. Load the container
1919 * and try to create a volume
1921 struct intel_super
*super
;
1923 if (load_super_imsm_all(st
, cfd
, (void **) &super
, NULL
, 1) == 0) {
1925 st
->container_dev
= fd2devnum(cfd
);
1927 return validate_geometry_imsm_volume(st
, level
, layout
,
1933 } else /* may belong to another container */
1939 static struct mdinfo
*container_content_imsm(struct supertype
*st
)
1941 /* Given a container loaded by load_super_imsm_all,
1942 * extract information about all the arrays into
1945 * For each imsm_dev create an mdinfo, fill it in,
1946 * then look for matching devices in super->disks
1947 * and create appropriate device mdinfo.
1949 struct intel_super
*super
= st
->sb
;
1950 struct imsm_super
*mpb
= super
->anchor
;
1951 struct mdinfo
*rest
= NULL
;
1954 /* do not assemble arrays that might have bad blocks */
1955 if (imsm_bbm_log_size(super
->anchor
)) {
1956 fprintf(stderr
, Name
": BBM log found in metadata. "
1957 "Cannot activate array(s).\n");
1961 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
1962 struct imsm_dev
*dev
= get_imsm_dev(super
, i
);
1963 struct imsm_vol
*vol
= &dev
->vol
;
1964 struct imsm_map
*map
= get_imsm_map(dev
, 0);
1965 struct mdinfo
*this;
1968 this = malloc(sizeof(*this));
1969 memset(this, 0, sizeof(*this));
1973 this->array
.level
= get_imsm_raid_level(map
);
1974 this->array
.raid_disks
= map
->num_members
;
1975 this->array
.layout
= imsm_level_to_layout(this->array
.level
);
1976 this->array
.md_minor
= -1;
1977 this->array
.ctime
= 0;
1978 this->array
.utime
= 0;
1979 this->array
.chunk_size
= __le16_to_cpu(map
->blocks_per_strip
) << 9;
1980 this->array
.state
= !vol
->dirty
;
1981 this->container_member
= i
;
1982 if (map
->map_state
== IMSM_T_STATE_UNINITIALIZED
|| dev
->vol
.dirty
)
1983 this->resync_start
= 0;
1985 this->resync_start
= ~0ULL;
1987 strncpy(this->name
, (char *) dev
->volume
, MAX_RAID_SERIAL_LEN
);
1988 this->name
[MAX_RAID_SERIAL_LEN
] = 0;
1990 sprintf(this->text_version
, "/%s/%d",
1991 devnum2devname(st
->container_dev
),
1992 this->container_member
);
1994 memset(this->uuid
, 0, sizeof(this->uuid
));
1996 this->component_size
= __le32_to_cpu(map
->blocks_per_member
);
1998 for (slot
= 0 ; slot
< map
->num_members
; slot
++) {
1999 struct mdinfo
*info_d
;
2004 idx
= get_imsm_disk_idx(map
, slot
);
2005 for (d
= super
->disks
; d
; d
= d
->next
)
2006 if (d
->index
== idx
)
2010 break; /* shouldn't this be continue ?? */
2012 info_d
= malloc(sizeof(*info_d
));
2014 break; /* ditto ?? */
2015 memset(info_d
, 0, sizeof(*info_d
));
2016 info_d
->next
= this->devs
;
2017 this->devs
= info_d
;
2019 s
= __le32_to_cpu(d
->disk
.status
);
2021 info_d
->disk
.number
= d
->index
;
2022 info_d
->disk
.major
= d
->major
;
2023 info_d
->disk
.minor
= d
->minor
;
2024 info_d
->disk
.raid_disk
= slot
;
2025 info_d
->disk
.state
= s
& CONFIGURED_DISK
? (1 << MD_DISK_ACTIVE
) : 0;
2026 info_d
->disk
.state
|= s
& FAILED_DISK
? (1 << MD_DISK_FAULTY
) : 0;
2027 info_d
->disk
.state
|= s
& USABLE_DISK
? (1 << MD_DISK_SYNC
) : 0;
2029 this->array
.working_disks
++;
2031 info_d
->events
= __le32_to_cpu(mpb
->generation_num
);
2032 info_d
->data_offset
= __le32_to_cpu(map
->pba_of_lba0
);
2033 info_d
->component_size
= __le32_to_cpu(map
->blocks_per_member
);
2035 strcpy(info_d
->name
, d
->devname
);
2043 static int imsm_open_new(struct supertype
*c
, struct active_array
*a
,
2046 struct intel_super
*super
= c
->sb
;
2047 struct imsm_super
*mpb
= super
->anchor
;
2049 if (atoi(inst
) >= mpb
->num_raid_devs
) {
2050 fprintf(stderr
, "%s: subarry index %d, out of range\n",
2051 __func__
, atoi(inst
));
2055 dprintf("imsm: open_new %s\n", inst
);
2056 a
->info
.container_member
= atoi(inst
);
2060 static __u8
imsm_check_degraded(struct intel_super
*super
, int n
, int failed
)
2062 struct imsm_dev
*dev
= get_imsm_dev(super
, n
);
2063 struct imsm_map
*map
= get_imsm_map(dev
, 0);
2066 return map
->map_state
;
2068 switch (get_imsm_raid_level(map
)) {
2070 return IMSM_T_STATE_FAILED
;
2073 if (failed
< map
->num_members
)
2074 return IMSM_T_STATE_DEGRADED
;
2076 return IMSM_T_STATE_FAILED
;
2081 * check to see if any mirrors have failed,
2082 * otherwise we are degraded
2084 int device_per_mirror
= 2; /* FIXME is this always the case?
2085 * and are they always adjacent?
2090 for (i
= 0; i
< map
->num_members
; i
++) {
2091 int idx
= get_imsm_disk_idx(map
, i
);
2092 struct imsm_disk
*disk
= get_imsm_disk(super
, idx
);
2094 if (__le32_to_cpu(disk
->status
) & FAILED_DISK
)
2097 if (failed
>= device_per_mirror
)
2098 return IMSM_T_STATE_FAILED
;
2100 /* reset 'failed' for next mirror set */
2101 if (!((i
+ 1) % device_per_mirror
))
2105 return IMSM_T_STATE_DEGRADED
;
2109 return IMSM_T_STATE_DEGRADED
;
2111 return IMSM_T_STATE_FAILED
;
2117 return map
->map_state
;
2120 static int imsm_count_failed(struct intel_super
*super
, struct imsm_map
*map
)
2124 struct imsm_disk
*disk
;
2126 for (i
= 0; i
< map
->num_members
; i
++) {
2127 int idx
= get_imsm_disk_idx(map
, i
);
2129 disk
= get_imsm_disk(super
, idx
);
2130 if (__le32_to_cpu(disk
->status
) & FAILED_DISK
)
2137 static void imsm_set_array_state(struct active_array
*a
, int consistent
)
2139 int inst
= a
->info
.container_member
;
2140 struct intel_super
*super
= a
->container
->sb
;
2141 struct imsm_dev
*dev
= get_imsm_dev(super
, inst
);
2142 struct imsm_map
*map
= get_imsm_map(dev
, 0);
2143 int dirty
= !consistent
;
2147 if (a
->resync_start
== ~0ULL) {
2148 failed
= imsm_count_failed(super
, map
);
2149 map_state
= imsm_check_degraded(super
, inst
, failed
);
2150 /* complete recovery or initial resync */
2152 map_state
= IMSM_T_STATE_NORMAL
;
2153 if (map
->map_state
!= map_state
) {
2154 dprintf("imsm: map_state %d: %d\n",
2156 map
->map_state
= map_state
;
2157 super
->updates_pending
++;
2160 /* complete resync */
2161 if (!dirty
&& dev
->vol
.dirty
) {
2162 dprintf("imsm: mark 'clean'\n");
2164 super
->updates_pending
++;
2170 if (dirty
&& !dev
->vol
.dirty
) {
2171 dprintf("imsm: mark 'dirty' (%llu)\n", a
->resync_start
);
2173 super
->updates_pending
++;
2177 static void imsm_set_disk(struct active_array
*a
, int n
, int state
)
2179 int inst
= a
->info
.container_member
;
2180 struct intel_super
*super
= a
->container
->sb
;
2181 struct imsm_dev
*dev
= get_imsm_dev(super
, inst
);
2182 struct imsm_map
*map
= get_imsm_map(dev
, 0);
2183 struct imsm_disk
*disk
;
2186 int new_failure
= 0;
2188 if (n
> map
->num_members
)
2189 fprintf(stderr
, "imsm: set_disk %d out of range 0..%d\n",
2190 n
, map
->num_members
- 1);
2195 dprintf("imsm: set_disk %d:%x\n", n
, state
);
2197 disk
= get_imsm_disk(super
, get_imsm_disk_idx(map
, n
));
2199 /* check for new failures */
2200 status
= __le32_to_cpu(disk
->status
);
2201 if ((state
& DS_FAULTY
) && !(status
& FAILED_DISK
)) {
2202 status
|= FAILED_DISK
;
2203 disk
->status
= __cpu_to_le32(status
);
2205 super
->updates_pending
++;
2207 /* check if in_sync */
2208 if ((state
& DS_INSYNC
) && !(status
& USABLE_DISK
)) {
2209 status
|= USABLE_DISK
;
2210 disk
->status
= __cpu_to_le32(status
);
2211 super
->updates_pending
++;
2214 /* the number of failures have changed, count up 'failed' to determine
2215 * degraded / failed status
2217 if (new_failure
&& map
->map_state
!= IMSM_T_STATE_FAILED
)
2218 failed
= imsm_count_failed(super
, map
);
2220 /* determine map_state based on failed or in_sync count */
2222 map
->map_state
= imsm_check_degraded(super
, inst
, failed
);
2223 else if (map
->map_state
== IMSM_T_STATE_DEGRADED
) {
2227 for (d
= a
->info
.devs
; d
; d
= d
->next
)
2228 if (d
->curr_state
& DS_INSYNC
)
2231 if (working
== a
->info
.array
.raid_disks
) {
2232 map
->map_state
= IMSM_T_STATE_NORMAL
;
2233 super
->updates_pending
++;
2238 static int store_imsm_mpb(int fd
, struct intel_super
*super
)
2240 struct imsm_super
*mpb
= super
->anchor
;
2241 __u32 mpb_size
= __le32_to_cpu(mpb
->mpb_size
);
2242 unsigned long long dsize
;
2243 unsigned long long sectors
;
2245 get_dev_size(fd
, NULL
, &dsize
);
2247 if (mpb_size
> 512) {
2248 /* -1 to account for anchor */
2249 sectors
= mpb_sectors(mpb
) - 1;
2251 /* write the extended mpb to the sectors preceeding the anchor */
2252 if (lseek64(fd
, dsize
- (512 * (2 + sectors
)), SEEK_SET
) < 0)
2255 if (write(fd
, super
->buf
+ 512, 512 * sectors
) != 512 * sectors
)
2259 /* first block is stored on second to last sector of the disk */
2260 if (lseek64(fd
, dsize
- (512 * 2), SEEK_SET
) < 0)
2263 if (write(fd
, super
->buf
, 512) != 512)
2269 static void imsm_sync_metadata(struct supertype
*container
)
2271 struct intel_super
*super
= container
->sb
;
2273 if (!super
->updates_pending
)
2276 write_super_imsm(super
, 0);
2278 super
->updates_pending
= 0;
2281 static struct mdinfo
*imsm_activate_spare(struct active_array
*a
,
2282 struct metadata_update
**updates
)
2285 * Find a device with unused free space and use it to replace a
2286 * failed/vacant region in an array. We replace failed regions one a
2287 * array at a time. The result is that a new spare disk will be added
2288 * to the first failed array and after the monitor has finished
2289 * propagating failures the remainder will be consumed.
2291 * FIXME add a capability for mdmon to request spares from another
2295 struct intel_super
*super
= a
->container
->sb
;
2296 int inst
= a
->info
.container_member
;
2297 struct imsm_dev
*dev
= get_imsm_dev(super
, inst
);
2298 struct imsm_map
*map
= get_imsm_map(dev
, 0);
2299 int failed
= a
->info
.array
.raid_disks
;
2300 struct mdinfo
*rv
= NULL
;
2303 struct metadata_update
*mu
;
2305 struct imsm_update_activate_spare
*u
;
2309 for (d
= a
->info
.devs
; d
; d
= d
->next
) {
2310 if ((d
->curr_state
& DS_FAULTY
) &&
2312 /* wait for Removal to happen */
2314 if (d
->state_fd
>= 0)
2318 dprintf("imsm: activate spare: inst=%d failed=%d (%d) level=%d\n",
2319 inst
, failed
, a
->info
.array
.raid_disks
, a
->info
.array
.level
);
2320 if (imsm_check_degraded(super
, inst
, failed
) != IMSM_T_STATE_DEGRADED
)
2323 /* For each slot, if it is not working, find a spare */
2325 for (i
= 0; i
< a
->info
.array
.raid_disks
; i
++) {
2326 for (d
= a
->info
.devs
; d
; d
= d
->next
)
2327 if (d
->disk
.raid_disk
== i
)
2329 dprintf("found %d: %p %x\n", i
, d
, d
?d
->curr_state
:0);
2330 if (d
&& (d
->state_fd
>= 0))
2333 /* OK, this device needs recovery. Find a spare */
2334 for ( ; dl
; dl
= dl
->next
) {
2335 unsigned long long esize
;
2336 unsigned long long pos
;
2343 /* If in this array, skip */
2344 for (d2
= a
->info
.devs
; d2
; d2
= d2
->next
)
2345 if (d2
->disk
.major
== dl
->major
&&
2346 d2
->disk
.minor
== dl
->minor
) {
2347 dprintf("%x:%x already in array\n", dl
->major
, dl
->minor
);
2353 /* Does this unused device have the requisite free space?
2354 * We need a->info.component_size sectors
2356 ex
= get_extents(super
, dl
);
2358 dprintf("cannot get extents\n");
2364 array_start
= __le32_to_cpu(map
->pba_of_lba0
);
2367 /* check that we can start at pba_of_lba0 with
2368 * a->info.component_size of space
2370 esize
= ex
[j
].start
- pos
;
2371 if (array_start
>= pos
&&
2372 array_start
+ a
->info
.component_size
< ex
[j
].start
) {
2376 pos
= ex
[j
].start
+ ex
[j
].size
;
2379 } while (ex
[j
-1].size
);
2383 dprintf("%x:%x does not have %llu at %d\n",
2384 dl
->major
, dl
->minor
,
2385 a
->info
.component_size
,
2386 __le32_to_cpu(map
->pba_of_lba0
));
2391 /* found a usable disk with enough space */
2392 di
= malloc(sizeof(*di
));
2393 memset(di
, 0, sizeof(*di
));
2395 /* dl->index will be -1 in the case we are activating a
2396 * pristine spare. imsm_process_update() will create a
2397 * new index in this case. Once a disk is found to be
2398 * failed in all member arrays it is kicked from the
2401 di
->disk
.number
= dl
->index
;
2403 /* (ab)use di->devs to store a pointer to the device
2406 di
->devs
= (struct mdinfo
*) dl
;
2408 di
->disk
.raid_disk
= i
;
2409 di
->disk
.major
= dl
->major
;
2410 di
->disk
.minor
= dl
->minor
;
2412 di
->data_offset
= array_start
;
2413 di
->component_size
= a
->info
.component_size
;
2414 di
->container_member
= inst
;
2418 dprintf("%x:%x to be %d at %llu\n", dl
->major
, dl
->minor
,
2426 /* No spares found */
2428 /* Now 'rv' has a list of devices to return.
2429 * Create a metadata_update record to update the
2430 * disk_ord_tbl for the array
2432 mu
= malloc(sizeof(*mu
));
2433 mu
->buf
= malloc(sizeof(struct imsm_update_activate_spare
) * num_spares
);
2435 mu
->len
= sizeof(struct imsm_update_activate_spare
) * num_spares
;
2436 mu
->next
= *updates
;
2437 u
= (struct imsm_update_activate_spare
*) mu
->buf
;
2439 for (di
= rv
; di
; di
= di
->next
) {
2440 u
->type
= update_activate_spare
;
2441 u
->dl
= (struct dl
*) di
->devs
;
2443 u
->slot
= di
->disk
.raid_disk
;
2454 static int disks_overlap(struct imsm_map
*m1
, struct imsm_map
*m2
)
2460 for (i
= 0; i
< m1
->num_members
; i
++) {
2461 idx
= get_imsm_disk_idx(m1
, i
);
2462 for (j
= 0; j
< m2
->num_members
; j
++)
2463 if (idx
== get_imsm_disk_idx(m2
, j
))
2470 static void imsm_process_update(struct supertype
*st
,
2471 struct metadata_update
*update
)
2474 * crack open the metadata_update envelope to find the update record
2475 * update can be one of:
2476 * update_activate_spare - a spare device has replaced a failed
2477 * device in an array, update the disk_ord_tbl. If this disk is
2478 * present in all member arrays then also clear the SPARE_DISK
2481 struct intel_super
*super
= st
->sb
;
2482 struct imsm_super
*mpb
= super
->anchor
;
2483 enum imsm_update_type type
= *(enum imsm_update_type
*) update
->buf
;
2486 case update_activate_spare
: {
2487 struct imsm_update_activate_spare
*u
= (void *) update
->buf
;
2488 struct imsm_dev
*dev
= get_imsm_dev(super
, u
->array
);
2489 struct imsm_map
*map
= get_imsm_map(dev
, 0);
2490 struct active_array
*a
;
2491 struct imsm_disk
*disk
;
2498 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
2503 fprintf(stderr
, "error: imsm_activate_spare passed "
2504 "an unknown disk (index: %d serial: %s)\n",
2505 u
->dl
->index
, u
->dl
->serial
);
2509 super
->updates_pending
++;
2511 /* adding a pristine spare, assign a new index */
2512 if (dl
->index
< 0) {
2513 dl
->index
= super
->anchor
->num_disks
;
2514 super
->anchor
->num_disks
++;
2516 victim
= get_imsm_disk_idx(map
, u
->slot
);
2517 map
->disk_ord_tbl
[u
->slot
] = __cpu_to_le32(dl
->index
);
2519 status
= __le32_to_cpu(disk
->status
);
2520 status
|= CONFIGURED_DISK
;
2521 status
&= ~(SPARE_DISK
| USABLE_DISK
);
2522 disk
->status
= __cpu_to_le32(status
);
2524 /* count arrays using the victim in the metadata */
2526 for (a
= st
->arrays
; a
; a
= a
->next
) {
2527 dev
= get_imsm_dev(super
, a
->info
.container_member
);
2528 map
= get_imsm_map(dev
, 0);
2529 for (i
= 0; i
< map
->num_members
; i
++)
2530 if (victim
== get_imsm_disk_idx(map
, i
))
2534 /* clear some flags if the victim is no longer being
2538 disk
= get_imsm_disk(super
, victim
);
2539 status
= __le32_to_cpu(disk
->status
);
2540 status
&= ~(CONFIGURED_DISK
| USABLE_DISK
);
2541 disk
->status
= __cpu_to_le32(status
);
2542 /* at this point the disk can be removed from the
2543 * metadata, however we need to guarantee that we do
2544 * not race with any manager thread routine that relies
2545 * on dl->index or map->disk_ord_tbl
2550 case update_create_array
: {
2551 /* someone wants to create a new array, we need to be aware of
2552 * a few races/collisions:
2553 * 1/ 'Create' called by two separate instances of mdadm
2554 * 2/ 'Create' versus 'activate_spare': mdadm has chosen
2555 * devices that have since been assimilated via
2557 * In the event this update can not be carried out mdadm will
2558 * (FIX ME) notice that its update did not take hold.
2560 struct imsm_update_create_array
*u
= (void *) update
->buf
;
2561 struct imsm_dev
*dev
;
2562 struct imsm_map
*map
, *new_map
;
2563 unsigned long long start
, end
;
2564 unsigned long long new_start
, new_end
;
2568 /* handle racing creates: first come first serve */
2569 if (u
->dev_idx
< mpb
->num_raid_devs
) {
2570 dprintf("%s: subarray %d already defined\n",
2571 __func__
, u
->dev_idx
);
2575 /* check update is next in sequence */
2576 if (u
->dev_idx
!= mpb
->num_raid_devs
) {
2577 dprintf("%s: can not create arrays out of sequence\n",
2582 new_map
= get_imsm_map(&u
->dev
, 0);
2583 new_start
= __le32_to_cpu(new_map
->pba_of_lba0
);
2584 new_end
= new_start
+ __le32_to_cpu(new_map
->blocks_per_member
);
2586 /* handle activate_spare versus create race:
2587 * check to make sure that overlapping arrays do not include
2590 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
2591 dev
= get_imsm_dev(super
, i
);
2592 map
= get_imsm_map(dev
, 0);
2593 start
= __le32_to_cpu(map
->pba_of_lba0
);
2594 end
= start
+ __le32_to_cpu(map
->blocks_per_member
);
2595 if ((new_start
>= start
&& new_start
<= end
) ||
2596 (start
>= new_start
&& start
<= new_end
))
2598 if (overlap
&& disks_overlap(map
, new_map
)) {
2599 dprintf("%s: arrays overlap\n", __func__
);
2603 /* check num_members sanity */
2604 if (new_map
->num_members
> mpb
->num_disks
) {
2605 dprintf("%s: num_disks out of range\n", __func__
);
2609 /* check that prepare update was successful */
2610 if (!update
->space
) {
2611 dprintf("%s: prepare update failed\n", __func__
);
2615 super
->updates_pending
++;
2616 dev
= update
->space
;
2617 update
->space
= NULL
;
2618 imsm_copy_dev(dev
, &u
->dev
);
2619 super
->dev_tbl
[u
->dev_idx
] = dev
;
2620 mpb
->num_raid_devs
++;
2622 /* fix up flags, if arrays overlap then the drives can not be
2625 for (i
= 0; i
< map
->num_members
; i
++) {
2626 struct imsm_disk
*disk
;
2629 disk
= get_imsm_disk(super
, get_imsm_disk_idx(map
, i
));
2630 status
= __le32_to_cpu(disk
->status
);
2631 status
|= CONFIGURED_DISK
;
2633 status
&= ~SPARE_DISK
;
2634 disk
->status
= __cpu_to_le32(status
);
2641 static void imsm_prepare_update(struct supertype
*st
,
2642 struct metadata_update
*update
)
2645 * Allocate space to hold new disk entries, raid-device entries or a
2646 * new mpb if necessary. We currently maintain an mpb large enough to
2647 * hold 2 subarrays for the given number of disks. This may not be
2648 * sufficient when reshaping.
2650 * FIX ME handle the reshape case.
2652 * The monitor will be able to safely change super->mpb by arranging
2653 * for it to be freed in check_update_queue(). I.e. the monitor thread
2654 * will start using the new pointer and the manager can continue to use
2655 * the old value until check_update_queue() runs.
2657 enum imsm_update_type type
= *(enum imsm_update_type
*) update
->buf
;
2660 case update_create_array
: {
2661 struct imsm_update_create_array
*u
= (void *) update
->buf
;
2662 size_t len
= sizeof_imsm_dev(&u
->dev
);
2664 update
->space
= malloc(len
);
2674 struct superswitch super_imsm
= {
2676 .examine_super
= examine_super_imsm
,
2677 .brief_examine_super
= brief_examine_super_imsm
,
2678 .detail_super
= detail_super_imsm
,
2679 .brief_detail_super
= brief_detail_super_imsm
,
2680 .write_init_super
= write_init_super_imsm
,
2682 .match_home
= match_home_imsm
,
2683 .uuid_from_super
= uuid_from_super_imsm
,
2684 .getinfo_super
= getinfo_super_imsm
,
2685 .update_super
= update_super_imsm
,
2687 .avail_size
= avail_size_imsm
,
2689 .compare_super
= compare_super_imsm
,
2691 .load_super
= load_super_imsm
,
2692 .init_super
= init_super_imsm
,
2693 .add_to_super
= add_to_super_imsm
,
2694 .store_super
= store_zero_imsm
,
2695 .free_super
= free_super_imsm
,
2696 .match_metadata_desc
= match_metadata_desc_imsm
,
2697 .container_content
= container_content_imsm
,
2699 .validate_geometry
= validate_geometry_imsm
,
2703 .open_new
= imsm_open_new
,
2704 .load_super
= load_super_imsm
,
2705 .set_array_state
= imsm_set_array_state
,
2706 .set_disk
= imsm_set_disk
,
2707 .sync_metadata
= imsm_sync_metadata
,
2708 .activate_spare
= imsm_activate_spare
,
2709 .process_update
= imsm_process_update
,
2710 .prepare_update
= imsm_prepare_update
,