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 reserved
[2]; /* 0x30 - 0x37 */
104 __u8 num_disks
; /* 0x38 Number of configured disks */
105 __u8 num_raid_devs
; /* 0x39 Number of configured volumes */
106 __u8 fill
[2]; /* 0x3A - 0x3B */
107 #define IMSM_FILLERS 39
108 __u32 filler
[IMSM_FILLERS
]; /* 0x3C - 0xD7 RAID_MPB_FILLERS */
109 struct imsm_disk disk
[1]; /* 0xD8 diskTbl[numDisks] */
110 /* here comes imsm_dev[num_raid_devs] */
111 } __attribute__ ((packed
));
114 static char *map_state_str
[] = { "normal", "uninitialized", "degraded", "failed" };
117 static unsigned int sector_count(__u32 bytes
)
119 return ((bytes
+ (512-1)) & (~(512-1))) / 512;
122 static unsigned int mpb_sectors(struct imsm_super
*mpb
)
124 return sector_count(__le32_to_cpu(mpb
->mpb_size
));
127 /* internal representation of IMSM metadata */
130 struct imsm_super
*mpb
;
133 int updates_pending
; /* count of pending updates for mdmon */
134 int creating_imsm
; /* flag to indicate container creation */
135 int current_vol
; /* index of raid device undergoing creation */
139 __u8 serial
[MAX_RAID_SERIAL_LEN
];
147 unsigned long long start
, size
;
150 /* definition of messages passed to imsm_process_update */
151 enum imsm_update_type
{
152 update_activate_spare
,
156 struct imsm_update_activate_spare
{
157 enum imsm_update_type type
;
161 struct imsm_update_activate_spare
*next
;
164 struct imsm_update_create_array
{
165 enum imsm_update_type type
;
170 static struct supertype
*match_metadata_desc_imsm(char *arg
)
172 struct supertype
*st
;
174 if (strcmp(arg
, "imsm") != 0 &&
175 strcmp(arg
, "default") != 0
179 st
= malloc(sizeof(*st
));
180 memset(st
, 0, sizeof(*st
));
181 st
->ss
= &super_imsm
;
182 st
->max_devs
= IMSM_MAX_DEVICES
;
183 st
->minor_version
= 0;
188 static __u8
*get_imsm_version(struct imsm_super
*mpb
)
190 return &mpb
->sig
[MPB_SIG_LEN
];
193 static struct imsm_disk
*get_imsm_disk(struct imsm_super
*mpb
, __u8 index
)
195 if (index
> mpb
->num_disks
- 1)
197 return &mpb
->disk
[index
];
200 static __u32
gen_imsm_checksum(struct imsm_super
*mpb
)
202 __u32 end
= mpb
->mpb_size
/ sizeof(end
);
203 __u32
*p
= (__u32
*) mpb
;
207 sum
+= __le32_to_cpu(*p
++);
209 return sum
- __le32_to_cpu(mpb
->check_sum
);
212 static size_t sizeof_imsm_dev(struct imsm_dev
*dev
)
214 size_t size
= sizeof(*dev
);
216 /* each map has disk_ord_tbl[num_members - 1] additional space */
217 size
+= sizeof(__u32
) * (dev
->vol
.map
[0].num_members
- 1);
219 /* migrating means an additional map */
220 if (dev
->vol
.migr_state
) {
221 size
+= sizeof(struct imsm_map
);
222 size
+= sizeof(__u32
) * (dev
->vol
.map
[1].num_members
- 1);
228 static struct imsm_dev
*get_imsm_dev(struct imsm_super
*mpb
, __u8 index
)
234 if (index
> mpb
->num_raid_devs
- 1)
237 /* devices start after all disks */
238 offset
= ((void *) &mpb
->disk
[mpb
->num_disks
]) - _mpb
;
240 for (i
= 0; i
<= index
; i
++)
242 return _mpb
+ offset
;
244 offset
+= sizeof_imsm_dev(_mpb
+ offset
);
249 static __u32
get_imsm_disk_idx(struct imsm_map
*map
, int slot
)
251 __u32
*ord_tbl
= &map
->disk_ord_tbl
[slot
];
253 /* top byte is 'special' */
254 return __le32_to_cpu(*ord_tbl
& ~(0xff << 24));
257 static int get_imsm_raid_level(struct imsm_map
*map
)
259 if (map
->raid_level
== 1) {
260 if (map
->num_members
== 2)
266 return map
->raid_level
;
269 static int cmp_extent(const void *av
, const void *bv
)
271 const struct extent
*a
= av
;
272 const struct extent
*b
= bv
;
273 if (a
->start
< b
->start
)
275 if (a
->start
> b
->start
)
280 static struct extent
*get_extents(struct intel_super
*super
, struct dl
*dl
)
282 /* find a list of used extents on the given physical device */
283 struct imsm_super
*mpb
= super
->mpb
;
284 struct imsm_disk
*disk
;
285 struct extent
*rv
, *e
;
289 disk
= get_imsm_disk(mpb
, dl
->index
);
293 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
294 struct imsm_dev
*dev
= get_imsm_dev(mpb
, i
);
295 struct imsm_map
*map
= dev
->vol
.map
;
297 for (j
= 0; j
< map
->num_members
; j
++) {
298 __u32 index
= get_imsm_disk_idx(map
, j
);
300 if (index
== dl
->index
)
304 rv
= malloc(sizeof(struct extent
) * (memberships
+ 1));
309 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
310 struct imsm_dev
*dev
= get_imsm_dev(mpb
, i
);
311 struct imsm_map
*map
= dev
->vol
.map
;
313 for (j
= 0; j
< map
->num_members
; j
++) {
314 __u32 index
= get_imsm_disk_idx(map
, j
);
316 if (index
== dl
->index
) {
317 e
->start
= __le32_to_cpu(map
->pba_of_lba0
);
318 e
->size
= __le32_to_cpu(map
->blocks_per_member
);
323 qsort(rv
, memberships
, sizeof(*rv
), cmp_extent
);
325 e
->start
= __le32_to_cpu(disk
->total_blocks
) -
326 (MPB_SECTOR_CNT
+ IMSM_RESERVED_SECTORS
);
332 static void print_imsm_dev(struct imsm_dev
*dev
, int index
)
336 struct imsm_map
*map
= dev
->vol
.map
;
339 printf("[%s]:\n", dev
->volume
);
340 printf(" RAID Level : %d\n", get_imsm_raid_level(map
));
341 printf(" Members : %d\n", map
->num_members
);
342 for (slot
= 0; slot
< map
->num_members
; slot
++)
343 if (index
== get_imsm_disk_idx(map
, slot
))
345 if (slot
< map
->num_members
)
346 printf(" This Slot : %d\n", slot
);
348 printf(" This Slot : ?\n");
349 sz
= __le32_to_cpu(dev
->size_high
);
351 sz
+= __le32_to_cpu(dev
->size_low
);
352 printf(" Array Size : %llu%s\n", (unsigned long long)sz
,
353 human_size(sz
* 512));
354 sz
= __le32_to_cpu(map
->blocks_per_member
);
355 printf(" Per Dev Size : %llu%s\n", (unsigned long long)sz
,
356 human_size(sz
* 512));
357 printf(" Sector Offset : %u\n",
358 __le32_to_cpu(map
->pba_of_lba0
));
359 printf(" Num Stripes : %u\n",
360 __le32_to_cpu(map
->num_data_stripes
));
361 printf(" Chunk Size : %u KiB\n",
362 __le16_to_cpu(map
->blocks_per_strip
) / 2);
363 printf(" Reserved : %d\n", __le32_to_cpu(dev
->reserved_blocks
));
364 printf(" Migrate State : %s\n", dev
->vol
.migr_state
? "migrating" : "idle");
365 printf(" Dirty State : %s\n", dev
->vol
.dirty
? "dirty" : "clean");
366 printf(" Map State : %s\n", map_state_str
[map
->map_state
]);
369 static void print_imsm_disk(struct imsm_super
*mpb
, int index
)
371 struct imsm_disk
*disk
= get_imsm_disk(mpb
, index
);
372 char str
[MAX_RAID_SERIAL_LEN
];
377 snprintf(str
, MAX_RAID_SERIAL_LEN
, "%s", disk
->serial
);
378 printf(" Disk%02d Serial : %s\n", index
, str
);
379 s
= __le32_to_cpu(disk
->status
);
380 printf(" State :%s%s%s%s\n", s
&SPARE_DISK
? " spare" : "",
381 s
&CONFIGURED_DISK
? " active" : "",
382 s
&FAILED_DISK
? " failed" : "",
383 s
&USABLE_DISK
? " usable" : "");
384 printf(" Id : %08x\n", __le32_to_cpu(disk
->scsi_id
));
385 sz
= __le32_to_cpu(disk
->total_blocks
) -
386 (MPB_SECTOR_CNT
+ IMSM_RESERVED_SECTORS
* mpb
->num_raid_devs
);
387 printf(" Usable Size : %llu%s\n", (unsigned long long)sz
,
388 human_size(sz
* 512));
391 static void examine_super_imsm(struct supertype
*st
, char *homehost
)
393 struct intel_super
*super
= st
->sb
;
394 struct imsm_super
*mpb
= super
->mpb
;
395 char str
[MAX_SIGNATURE_LENGTH
];
399 snprintf(str
, MPB_SIG_LEN
, "%s", mpb
->sig
);
400 printf(" Magic : %s\n", str
);
401 snprintf(str
, strlen(MPB_VERSION_RAID0
), "%s", get_imsm_version(mpb
));
402 printf(" Version : %s\n", get_imsm_version(mpb
));
403 printf(" Family : %08x\n", __le32_to_cpu(mpb
->family_num
));
404 printf(" Generation : %08x\n", __le32_to_cpu(mpb
->generation_num
));
405 sum
= __le32_to_cpu(mpb
->check_sum
);
406 printf(" Checksum : %08x %s\n", sum
,
407 gen_imsm_checksum(mpb
) == sum
? "correct" : "incorrect");
408 printf(" MPB Sectors : %d\n", mpb_sectors(mpb
));
409 printf(" Disks : %d\n", mpb
->num_disks
);
410 printf(" RAID Devices : %d\n", mpb
->num_raid_devs
);
411 print_imsm_disk(mpb
, super
->disks
->index
);
412 for (i
= 0; i
< mpb
->num_raid_devs
; i
++)
413 print_imsm_dev(get_imsm_dev(mpb
, i
), super
->disks
->index
);
414 for (i
= 0; i
< mpb
->num_disks
; i
++) {
415 if (i
== super
->disks
->index
)
417 print_imsm_disk(mpb
, i
);
421 static void brief_examine_super_imsm(struct supertype
*st
)
423 struct intel_super
*super
= st
->sb
;
424 struct imsm_super
*mpb
= super
->mpb
;
426 printf("ARRAY /dev/imsm family=%08x metadata=external:imsm\n",
427 __le32_to_cpu(mpb
->family_num
));
430 static void detail_super_imsm(struct supertype
*st
, char *homehost
)
432 printf("%s\n", __FUNCTION__
);
435 static void brief_detail_super_imsm(struct supertype
*st
)
437 printf("%s\n", __FUNCTION__
);
441 static int match_home_imsm(struct supertype
*st
, char *homehost
)
443 printf("%s\n", __FUNCTION__
);
448 static void uuid_from_super_imsm(struct supertype
*st
, int uuid
[4])
450 printf("%s\n", __FUNCTION__
);
455 get_imsm_numerical_version(struct imsm_super
*mpb
, int *m
, int *p
)
457 __u8
*v
= get_imsm_version(mpb
);
458 __u8
*end
= mpb
->sig
+ MAX_SIGNATURE_LENGTH
;
459 char major
[] = { 0, 0, 0 };
460 char minor
[] = { 0 ,0, 0 };
461 char patch
[] = { 0, 0, 0 };
462 char *ver_parse
[] = { major
, minor
, patch
};
466 while (*v
!= '\0' && v
< end
) {
467 if (*v
!= '.' && j
< 2)
468 ver_parse
[i
][j
++] = *v
;
476 *m
= strtol(minor
, NULL
, 0);
477 *p
= strtol(patch
, NULL
, 0);
481 static int imsm_level_to_layout(int level
)
489 return ALGORITHM_LEFT_SYMMETRIC
;
491 return 0x102; //FIXME is this correct?
496 static void getinfo_super_imsm_volume(struct supertype
*st
, struct mdinfo
*info
)
498 struct intel_super
*super
= st
->sb
;
499 struct imsm_super
*mpb
= super
->mpb
;
500 struct imsm_dev
*dev
= get_imsm_dev(mpb
, super
->current_vol
);
501 struct imsm_map
*map
= &dev
->vol
.map
[0];
503 info
->container_member
= super
->current_vol
;
504 info
->array
.raid_disks
= map
->num_members
;
505 info
->array
.level
= get_imsm_raid_level(map
);
506 info
->array
.layout
= imsm_level_to_layout(info
->array
.level
);
507 info
->array
.md_minor
= -1;
508 info
->array
.ctime
= 0;
509 info
->array
.utime
= 0;
510 info
->array
.chunk_size
= __le16_to_cpu(map
->blocks_per_strip
* 512);
512 info
->data_offset
= __le32_to_cpu(map
->pba_of_lba0
);
513 info
->component_size
= __le32_to_cpu(map
->blocks_per_member
);
515 info
->disk
.major
= 0;
516 info
->disk
.minor
= 0;
518 sprintf(info
->text_version
, "/%s/%d",
519 devnum2devname(st
->container_dev
),
520 info
->container_member
);
524 static void getinfo_super_imsm(struct supertype
*st
, struct mdinfo
*info
)
526 struct intel_super
*super
= st
->sb
;
527 struct imsm_super
*mpb
= super
->mpb
;
528 struct imsm_disk
*disk
;
531 if (super
->current_vol
>= 0) {
532 getinfo_super_imsm_volume(st
, info
);
535 info
->array
.raid_disks
= mpb
->num_disks
;
536 info
->array
.level
= LEVEL_CONTAINER
;
537 info
->array
.layout
= 0;
538 info
->array
.md_minor
= -1;
539 info
->array
.ctime
= 0; /* N/A for imsm */
540 info
->array
.utime
= 0;
541 info
->array
.chunk_size
= 0;
543 info
->disk
.major
= 0;
544 info
->disk
.minor
= 0;
545 info
->disk
.raid_disk
= -1;
546 info
->reshape_active
= 0;
547 strcpy(info
->text_version
, "imsm");
548 info
->disk
.number
= -1;
549 info
->disk
.state
= 0;
552 disk
= get_imsm_disk(mpb
, super
->disks
->index
);
553 info
->disk
.number
= super
->disks
->index
;
554 info
->disk
.raid_disk
= super
->disks
->index
;
555 info
->data_offset
= __le32_to_cpu(disk
->total_blocks
) -
556 (MPB_SECTOR_CNT
+ IMSM_RESERVED_SECTORS
);
557 info
->component_size
= MPB_SECTOR_CNT
+ IMSM_RESERVED_SECTORS
;
558 s
= __le32_to_cpu(disk
->status
);
559 info
->disk
.state
= s
& CONFIGURED_DISK
? (1 << MD_DISK_ACTIVE
) : 0;
560 info
->disk
.state
|= s
& FAILED_DISK
? (1 << MD_DISK_FAULTY
) : 0;
561 info
->disk
.state
|= s
& USABLE_DISK
? (1 << MD_DISK_SYNC
) : 0;
565 static int update_super_imsm(struct supertype
*st
, struct mdinfo
*info
,
566 char *update
, char *devname
, int verbose
,
567 int uuid_set
, char *homehost
)
571 /* For 'assemble' and 'force' we need to return non-zero if any
572 * change was made. For others, the return value is ignored.
573 * Update options are:
574 * force-one : This device looks a bit old but needs to be included,
575 * update age info appropriately.
576 * assemble: clear any 'faulty' flag to allow this device to
578 * force-array: Array is degraded but being forced, mark it clean
579 * if that will be needed to assemble it.
581 * newdev: not used ????
582 * grow: Array has gained a new device - this is currently for
584 * resync: mark as dirty so a resync will happen.
585 * name: update the name - preserving the homehost
587 * Following are not relevant for this imsm:
588 * sparc2.2 : update from old dodgey metadata
589 * super-minor: change the preferred_minor number
590 * summaries: update redundant counters.
591 * uuid: Change the uuid of the array to match watch is given
592 * homehost: update the recorded homehost
593 * _reshape_progress: record new reshape_progress position.
596 //struct intel_super *super = st->sb;
597 //struct imsm_super *mpb = super->mpb;
599 if (strcmp(update
, "grow") == 0) {
601 if (strcmp(update
, "resync") == 0) {
602 /* dev->vol.dirty = 1; */
605 /* IMSM has no concept of UUID or homehost */
610 static size_t disks_to_mpb_size(int disks
)
614 size
= sizeof(struct imsm_super
);
615 size
+= (disks
- 1) * sizeof(struct imsm_disk
);
616 size
+= 2 * sizeof(struct imsm_dev
);
617 /* up to 2 maps per raid device (-2 for imsm_maps in imsm_dev */
618 size
+= (4 - 2) * sizeof(struct imsm_map
);
619 /* 4 possible disk_ord_tbl's */
620 size
+= 4 * (disks
- 1) * sizeof(__u32
);
625 static __u64
avail_size_imsm(struct supertype
*st
, __u64 devsize
)
627 if (devsize
< (MPB_SECTOR_CNT
+ IMSM_RESERVED_SECTORS
))
630 return devsize
- (MPB_SECTOR_CNT
+ IMSM_RESERVED_SECTORS
);
633 static int compare_super_imsm(struct supertype
*st
, struct supertype
*tst
)
637 * 0 same, or first was empty, and second was copied
638 * 1 second had wrong number
642 struct intel_super
*first
= st
->sb
;
643 struct intel_super
*sec
= tst
->sb
;
651 if (memcmp(first
->mpb
->sig
, sec
->mpb
->sig
, MAX_SIGNATURE_LENGTH
) != 0)
653 if (first
->mpb
->family_num
!= sec
->mpb
->family_num
)
655 if (first
->mpb
->mpb_size
!= sec
->mpb
->mpb_size
)
657 if (first
->mpb
->check_sum
!= sec
->mpb
->check_sum
)
663 extern int scsi_get_serial(int fd
, void *buf
, size_t buf_len
);
665 static int imsm_read_serial(int fd
, char *devname
,
666 __u8 serial
[MAX_RAID_SERIAL_LEN
])
668 unsigned char scsi_serial
[255];
674 memset(scsi_serial
, 0, sizeof(scsi_serial
));
676 sg_fd
= sysfs_disk_to_sg(fd
);
680 Name
": Failed to open sg interface for %s: %s\n",
681 devname
, strerror(errno
));
685 rv
= scsi_get_serial(sg_fd
, scsi_serial
, sizeof(scsi_serial
));
691 Name
": Failed to retrieve serial for %s\n",
696 rsp_len
= scsi_serial
[3];
697 for (i
= 0, cnt
= 0; i
< rsp_len
; i
++) {
698 if (!isspace(scsi_serial
[4 + i
]))
699 serial
[cnt
++] = scsi_serial
[4 + i
];
700 if (cnt
== MAX_RAID_SERIAL_LEN
)
704 serial
[MAX_RAID_SERIAL_LEN
- 1] = '\0';
710 load_imsm_disk(int fd
, struct intel_super
*super
, char *devname
, int keep_fd
)
712 struct imsm_super
*mpb
= super
->mpb
;
715 struct imsm_disk
*disk
;
719 dl
= malloc(sizeof(*dl
));
723 Name
": failed to allocate disk buffer for %s\n",
727 memset(dl
, 0, sizeof(*dl
));
730 dl
->major
= major(stb
.st_rdev
);
731 dl
->minor
= minor(stb
.st_rdev
);
732 dl
->next
= super
->disks
;
733 dl
->fd
= keep_fd
? fd
: -1;
734 dl
->devname
= devname
? strdup(devname
) : NULL
;
737 rv
= imsm_read_serial(fd
, devname
, dl
->serial
);
742 /* look up this disk's index */
743 for (i
= 0; i
< mpb
->num_disks
; i
++) {
744 disk
= get_imsm_disk(mpb
, i
);
746 if (memcmp(disk
->serial
, dl
->serial
, MAX_RAID_SERIAL_LEN
) == 0)
750 if (i
> mpb
->num_disks
)
758 /* load_imsm_mpb - read matrix metadata
759 * allocates super->mpb to be freed by free_super
761 static int load_imsm_mpb(int fd
, struct intel_super
*super
, char *devname
)
763 unsigned long long dsize
;
764 size_t len
, mpb_size
;
765 unsigned long long sectors
;
767 struct imsm_super
*anchor
;
770 get_dev_size(fd
, NULL
, &dsize
);
772 if (lseek64(fd
, dsize
- (512 * 2), SEEK_SET
) < 0) {
775 Name
": Cannot seek to anchor block on %s: %s\n",
776 devname
, strerror(errno
));
781 if (posix_memalign((void**)&anchor
, 512, len
) != 0) {
784 Name
": Failed to allocate imsm anchor buffer"
785 " on %s\n", devname
);
788 if (read(fd
, anchor
, len
) != len
) {
791 Name
": Cannot read anchor block on %s: %s\n",
792 devname
, strerror(errno
));
797 if (strncmp((char *) anchor
->sig
, MPB_SIGNATURE
, MPB_SIG_LEN
) != 0) {
800 Name
": no IMSM anchor on %s\n", devname
);
805 mpb_size
= __le32_to_cpu(anchor
->mpb_size
);
806 mpb_size
= ROUND_UP(mpb_size
, 512);
807 if (posix_memalign(&super
->buf
, 512, mpb_size
) != 0) {
810 Name
": unable to allocate %zu byte mpb buffer\n",
815 memcpy(super
->buf
, anchor
, len
);
817 sectors
= mpb_sectors(anchor
) - 1;
820 return load_imsm_disk(fd
, super
, devname
, 0);
822 /* read the extended mpb */
823 if (lseek64(fd
, dsize
- (512 * (2 + sectors
)), SEEK_SET
) < 0) {
826 Name
": Cannot seek to extended mpb on %s: %s\n",
827 devname
, strerror(errno
));
831 len
= mpb_size
- 512;
832 if (read(fd
, super
->buf
+ 512, len
) != len
) {
835 Name
": Cannot read extended mpb on %s: %s\n",
836 devname
, strerror(errno
));
840 check_sum
= gen_imsm_checksum(super
->mpb
);
841 if (check_sum
!= __le32_to_cpu(super
->mpb
->check_sum
)) {
844 Name
": IMSM checksum %x != %x on %s\n",
845 check_sum
, __le32_to_cpu(super
->mpb
->check_sum
),
850 return load_imsm_disk(fd
, super
, devname
, 0);
853 static void free_imsm_disks(struct intel_super
*super
)
855 while (super
->disks
) {
856 struct dl
*d
= super
->disks
;
858 super
->disks
= d
->next
;
867 static void free_imsm(struct intel_super
*super
)
871 free_imsm_disks(super
);
876 static void free_super_imsm(struct supertype
*st
)
878 struct intel_super
*super
= st
->sb
;
887 static struct intel_super
*alloc_super(int creating_imsm
)
889 struct intel_super
*super
= malloc(sizeof(*super
));
892 memset(super
, 0, sizeof(*super
));
893 super
->creating_imsm
= creating_imsm
;
894 super
->current_vol
= -1;
901 static int load_super_imsm_all(struct supertype
*st
, int fd
, void **sbp
,
902 char *devname
, int keep_fd
)
905 struct intel_super
*super
;
906 struct mdinfo
*sd
, *best
= NULL
;
913 /* check if this disk is a member of an active array */
914 sra
= sysfs_read(fd
, 0, GET_LEVEL
|GET_VERSION
|GET_DEVS
|GET_STATE
);
918 if (sra
->array
.major_version
!= -1 ||
919 sra
->array
.minor_version
!= -2 ||
920 strcmp(sra
->text_version
, "imsm") != 0)
923 super
= alloc_super(0);
927 /* find the most up to date disk in this array */
928 for (sd
= sra
->devs
; sd
; sd
= sd
->next
) {
929 sprintf(nm
, "%d:%d", sd
->disk
.major
, sd
->disk
.minor
);
930 dfd
= dev_open(nm
, keep_fd
? O_RDWR
: O_RDONLY
);
935 rv
= load_imsm_mpb(dfd
, super
, NULL
);
939 gen
= __le32_to_cpu(super
->mpb
->generation_num
);
940 if (!best
|| gen
> bestgen
) {
955 /* load the most up to date anchor */
956 sprintf(nm
, "%d:%d", best
->disk
.major
, best
->disk
.minor
);
957 dfd
= dev_open(nm
, O_RDONLY
);
962 rv
= load_imsm_mpb(dfd
, super
, NULL
);
969 /* reset the disk list */
970 free_imsm_disks(super
);
972 /* populate disk list */
973 for (sd
= sra
->devs
; sd
; sd
= sd
->next
) {
974 sprintf(nm
, "%d:%d", sd
->disk
.major
, sd
->disk
.minor
);
975 dfd
= dev_open(nm
, keep_fd
? O_RDWR
: O_RDONLY
);
980 load_imsm_disk(dfd
, super
, NULL
, keep_fd
);
985 if (st
->subarray
[0]) {
986 if (atoi(st
->subarray
) <= super
->mpb
->num_raid_devs
)
987 super
->current_vol
= atoi(st
->subarray
);
993 if (st
->ss
== NULL
) {
994 st
->ss
= &super_imsm
;
995 st
->minor_version
= 0;
996 st
->max_devs
= IMSM_MAX_DEVICES
;
997 st
->container_dev
= fd2devnum(fd
);
1004 static int load_super_imsm(struct supertype
*st
, int fd
, char *devname
)
1006 struct intel_super
*super
;
1010 if (load_super_imsm_all(st
, fd
, &st
->sb
, devname
, 1) == 0)
1013 if (st
->subarray
[0])
1014 return 1; /* FIXME */
1016 super
= alloc_super(0);
1019 Name
": malloc of %zu failed.\n",
1024 rv
= load_imsm_mpb(fd
, super
, devname
);
1029 Name
": Failed to load all information "
1030 "sections on %s\n", devname
);
1036 if (st
->ss
== NULL
) {
1037 st
->ss
= &super_imsm
;
1038 st
->minor_version
= 0;
1039 st
->max_devs
= IMSM_MAX_DEVICES
;
1045 static __u16
info_to_blocks_per_strip(mdu_array_info_t
*info
)
1047 if (info
->level
== 1)
1049 return info
->chunk_size
>> 9;
1052 static __u32
info_to_num_data_stripes(mdu_array_info_t
*info
)
1056 num_stripes
= (info
->size
* 2) / info_to_blocks_per_strip(info
);
1057 if (info
->level
== 1)
1063 static int init_super_imsm_volume(struct supertype
*st
, mdu_array_info_t
*info
,
1064 unsigned long long size
, char *name
,
1065 char *homehost
, int *uuid
)
1067 /* We are creating a volume inside a pre-existing container.
1068 * so st->sb is already set.
1070 struct intel_super
*super
= st
->sb
;
1071 struct imsm_super
*mpb
= super
->mpb
;
1072 struct imsm_dev
*dev
;
1073 struct imsm_vol
*vol
;
1074 struct imsm_map
*map
;
1075 int idx
= mpb
->num_raid_devs
;
1077 unsigned long long array_blocks
;
1079 size_t size_old
, size_new
;
1081 if (mpb
->num_raid_devs
>= 2) {
1082 fprintf(stderr
, Name
": This imsm-container already has the "
1083 "maximum of 2 volumes\n");
1087 /* ensure the mpb is large enough for the new data */
1088 size_old
= __le32_to_cpu(mpb
->mpb_size
);
1089 size_new
= disks_to_mpb_size(info
->nr_disks
);
1090 if (size_new
> size_old
) {
1092 size_t size_round
= ROUND_UP(size_new
, 512);
1094 if (posix_memalign(&mpb_new
, 512, size_round
) != 0) {
1095 fprintf(stderr
, Name
": could not allocate new mpb\n");
1098 memcpy(mpb_new
, mpb
, size_old
);
1101 super
->mpb
= mpb_new
;
1102 mpb
->mpb_size
= __cpu_to_le32(size_new
);
1103 memset(mpb_new
+ size_old
, 0, size_round
- size_old
);
1105 super
->current_vol
= idx
;
1106 sprintf(st
->subarray
, "%d", idx
);
1107 mpb
->num_raid_devs
++;
1108 dev
= get_imsm_dev(mpb
, idx
);
1109 strncpy((char *) dev
->volume
, name
, MAX_RAID_SERIAL_LEN
);
1110 array_blocks
= calc_array_size(info
->level
, info
->raid_disks
,
1111 info
->layout
, info
->chunk_size
,
1113 dev
->size_low
= __cpu_to_le32((__u32
) array_blocks
);
1114 dev
->size_high
= __cpu_to_le32((__u32
) (array_blocks
>> 32));
1115 dev
->status
= __cpu_to_le32(0);
1116 dev
->reserved_blocks
= __cpu_to_le32(0);
1118 vol
->migr_state
= 0;
1121 for (i
= 0; i
< idx
; i
++) {
1122 struct imsm_dev
*prev
= get_imsm_dev(mpb
, i
);
1123 struct imsm_map
*pmap
= &prev
->vol
.map
[0];
1125 offset
+= __le32_to_cpu(pmap
->blocks_per_member
);
1126 offset
+= IMSM_RESERVED_SECTORS
;
1129 map
->pba_of_lba0
= __cpu_to_le32(offset
);
1130 map
->blocks_per_member
= __cpu_to_le32(info
->size
* 2);
1131 map
->blocks_per_strip
= __cpu_to_le16(info_to_blocks_per_strip(info
));
1132 map
->num_data_stripes
= __cpu_to_le32(info_to_num_data_stripes(info
));
1133 map
->map_state
= info
->level
? IMSM_T_STATE_UNINITIALIZED
:
1134 IMSM_T_STATE_NORMAL
;
1136 if (info
->level
== 1 && info
->raid_disks
> 2) {
1137 fprintf(stderr
, Name
": imsm does not support more than 2 disks"
1138 "in a raid1 volume\n");
1141 if (info
->level
== 10)
1142 map
->raid_level
= 1;
1144 map
->raid_level
= info
->level
;
1146 map
->num_members
= info
->raid_disks
;
1147 for (i
= 0; i
< map
->num_members
; i
++) {
1148 /* initialized in add_to_super */
1149 map
->disk_ord_tbl
[i
] = __cpu_to_le32(0);
1155 static int init_super_imsm(struct supertype
*st
, mdu_array_info_t
*info
,
1156 unsigned long long size
, char *name
,
1157 char *homehost
, int *uuid
)
1159 /* This is primarily called by Create when creating a new array.
1160 * We will then get add_to_super called for each component, and then
1161 * write_init_super called to write it out to each device.
1162 * For IMSM, Create can create on fresh devices or on a pre-existing
1164 * To create on a pre-existing array a different method will be called.
1165 * This one is just for fresh drives.
1167 struct intel_super
*super
;
1168 struct imsm_super
*mpb
;
1176 return init_super_imsm_volume(st
, info
, size
, name
, homehost
,
1179 super
= alloc_super(1);
1182 mpb_size
= disks_to_mpb_size(info
->nr_disks
);
1183 if (posix_memalign(&super
->buf
, 512, mpb_size
) != 0) {
1188 memset(mpb
, 0, mpb_size
);
1190 memcpy(mpb
->sig
, MPB_SIGNATURE
, strlen(MPB_SIGNATURE
));
1191 memcpy(mpb
->sig
+ strlen(MPB_SIGNATURE
), MPB_VERSION_RAID5
,
1192 strlen(MPB_VERSION_RAID5
));
1193 mpb
->mpb_size
= mpb_size
;
1199 static void add_to_super_imsm_volume(struct supertype
*st
, mdu_disk_info_t
*dk
,
1200 int fd
, char *devname
)
1202 struct intel_super
*super
= st
->sb
;
1203 struct imsm_super
*mpb
= super
->mpb
;
1205 struct imsm_dev
*dev
;
1206 struct imsm_map
*map
;
1207 struct imsm_disk
*disk
;
1210 dev
= get_imsm_dev(mpb
, super
->current_vol
);
1211 map
= &dev
->vol
.map
[0];
1213 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
1214 if (dl
->major
== dk
->major
&&
1215 dl
->minor
== dk
->minor
)
1217 if (!dl
|| ! (dk
->state
& (1<<MD_DISK_SYNC
)))
1220 map
->disk_ord_tbl
[dk
->number
] = __cpu_to_le32(dl
->index
);
1222 disk
= get_imsm_disk(mpb
, dl
->index
);
1223 status
= CONFIGURED_DISK
| USABLE_DISK
;
1224 disk
->status
= __cpu_to_le32(status
);
1227 static void add_to_super_imsm(struct supertype
*st
, mdu_disk_info_t
*dk
,
1228 int fd
, char *devname
)
1230 struct intel_super
*super
= st
->sb
;
1231 struct imsm_super
*mpb
= super
->mpb
;
1232 struct imsm_disk
*disk
;
1234 unsigned long long size
;
1239 if (super
->current_vol
>= 0) {
1240 add_to_super_imsm_volume(st
, dk
, fd
, devname
);
1245 dd
= malloc(sizeof(*dd
));
1248 Name
": malloc failed %s:%d.\n", __func__
, __LINE__
);
1251 memset(dd
, 0, sizeof(*dd
));
1252 dd
->major
= major(stb
.st_rdev
);
1253 dd
->minor
= minor(stb
.st_rdev
);
1254 dd
->index
= dk
->number
;
1255 dd
->devname
= devname
? strdup(devname
) : NULL
;
1256 dd
->next
= super
->disks
;
1258 rv
= imsm_read_serial(fd
, devname
, dd
->serial
);
1261 Name
": failed to retrieve scsi serial "
1262 "using \'%s\' instead\n", devname
);
1263 strcpy((char *) dd
->serial
, devname
);
1266 if (mpb
->num_disks
<= dk
->number
)
1267 mpb
->num_disks
= dk
->number
+ 1;
1269 disk
= get_imsm_disk(mpb
, dk
->number
);
1270 get_dev_size(fd
, NULL
, &size
);
1272 status
= USABLE_DISK
| SPARE_DISK
;
1273 strcpy((char *) disk
->serial
, (char *) dd
->serial
);
1274 disk
->total_blocks
= __cpu_to_le32(size
);
1275 disk
->status
= __cpu_to_le32(status
);
1276 if (sysfs_disk_to_scsi_id(fd
, &id
) == 0)
1277 disk
->scsi_id
= __cpu_to_le32(id
);
1279 disk
->scsi_id
= __cpu_to_le32(0);
1281 /* update the family number if we are creating a container */
1282 if (super
->creating_imsm
)
1283 mpb
->family_num
= __cpu_to_le32(gen_imsm_checksum(mpb
));
1288 static int store_imsm_mpb(int fd
, struct intel_super
*super
);
1290 static int write_super_imsm(struct intel_super
*super
, int doclose
)
1292 struct imsm_super
*mpb
= super
->mpb
;
1297 /* 'generation' is incremented everytime the metadata is written */
1298 generation
= __le32_to_cpu(mpb
->generation_num
);
1300 mpb
->generation_num
= __cpu_to_le32(generation
);
1302 /* recalculate checksum */
1303 sum
= gen_imsm_checksum(mpb
);
1304 mpb
->check_sum
= __cpu_to_le32(sum
);
1306 for (d
= super
->disks
; d
; d
= d
->next
) {
1307 if (store_imsm_mpb(d
->fd
, super
)) {
1308 fprintf(stderr
, "%s: failed for device %d:%d %s\n",
1309 __func__
, d
->major
, d
->minor
, strerror(errno
));
1321 static int write_init_super_imsm(struct supertype
*st
)
1323 if (st
->update_tail
) {
1324 /* queue the recently created array as a metadata update */
1326 struct imsm_update_create_array
*u
;
1327 struct intel_super
*super
= st
->sb
;
1328 struct imsm_super
*mpb
= super
->mpb
;
1329 struct imsm_dev
*dev
;
1330 struct imsm_map
*map
;
1333 if (super
->current_vol
< 0 ||
1334 !(dev
= get_imsm_dev(mpb
, super
->current_vol
))) {
1335 fprintf(stderr
, "%s: could not determine sub-array\n",
1341 map
= &dev
->vol
.map
[0];
1342 len
= sizeof(*u
) + sizeof(__u32
) * (map
->num_members
- 1);
1345 fprintf(stderr
, "%s: failed to allocate update buffer\n",
1350 u
->type
= update_create_array
;
1351 u
->dev_idx
= super
->current_vol
;
1352 memcpy(&u
->dev
, dev
, sizeof(*dev
));
1353 memcpy(u
->dev
.vol
.map
[0].disk_ord_tbl
, map
->disk_ord_tbl
,
1354 sizeof(__u32
) * map
->num_members
);
1355 append_metadata_update(st
, u
, len
);
1357 for (d
= super
->disks
; d
; d
= d
->next
) {
1364 return write_super_imsm(st
->sb
, 1);
1367 static int store_zero_imsm(struct supertype
*st
, int fd
)
1369 unsigned long long dsize
;
1372 get_dev_size(fd
, NULL
, &dsize
);
1374 /* first block is stored on second to last sector of the disk */
1375 if (lseek64(fd
, dsize
- (512 * 2), SEEK_SET
) < 0)
1378 if (posix_memalign(&buf
, 512, 512) != 0)
1381 memset(buf
, 0, sizeof(buf
));
1382 if (write(fd
, buf
, sizeof(buf
)) != sizeof(buf
))
1387 static int validate_geometry_imsm_container(struct supertype
*st
, int level
,
1388 int layout
, int raiddisks
, int chunk
,
1389 unsigned long long size
, char *dev
,
1390 unsigned long long *freesize
,
1394 unsigned long long ldsize
;
1396 if (level
!= LEVEL_CONTAINER
)
1401 fd
= open(dev
, O_RDONLY
|O_EXCL
, 0);
1404 fprintf(stderr
, Name
": imsm: Cannot open %s: %s\n",
1405 dev
, strerror(errno
));
1408 if (!get_dev_size(fd
, dev
, &ldsize
)) {
1414 *freesize
= avail_size_imsm(st
, ldsize
>> 9);
1419 /* validate_geometry_imsm_volume - lifted from validate_geometry_ddf_bvd
1420 * FIX ME add ahci details
1422 static int validate_geometry_imsm_volume(struct supertype
*st
, int level
,
1423 int layout
, int raiddisks
, int chunk
,
1424 unsigned long long size
, char *dev
,
1425 unsigned long long *freesize
,
1429 struct intel_super
*super
= st
->sb
;
1431 unsigned long long pos
= 0;
1432 unsigned long long maxsize
;
1436 if (level
== LEVEL_CONTAINER
)
1439 if (level
== 1 && raiddisks
> 2) {
1441 fprintf(stderr
, Name
": imsm does not support more "
1442 "than 2 in a raid1 configuration\n");
1446 /* We must have the container info already read in. */
1451 /* General test: make sure there is space for
1452 * 'raiddisks' device extents of size 'size' at a given
1455 unsigned long long minsize
= size
*2 /* convert to blocks */;
1456 unsigned long long start_offset
= ~0ULL;
1459 minsize
= MPB_SECTOR_CNT
+ IMSM_RESERVED_SECTORS
;
1460 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
1465 e
= get_extents(super
, dl
);
1468 unsigned long long esize
;
1469 esize
= e
[i
].start
- pos
;
1470 if (esize
>= minsize
)
1472 if (found
&& start_offset
== ~0ULL) {
1475 } else if (found
&& pos
!= start_offset
) {
1479 pos
= e
[i
].start
+ e
[i
].size
;
1481 } while (e
[i
-1].size
);
1486 if (dcnt
< raiddisks
) {
1488 fprintf(stderr
, Name
": imsm: Not enough "
1489 "devices with space for this array "
1496 /* This device must be a member of the set */
1497 if (stat(dev
, &stb
) < 0)
1499 if ((S_IFMT
& stb
.st_mode
) != S_IFBLK
)
1501 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
1502 if (dl
->major
== major(stb
.st_rdev
) &&
1503 dl
->minor
== minor(stb
.st_rdev
))
1508 fprintf(stderr
, Name
": %s is not in the "
1509 "same imsm set\n", dev
);
1512 e
= get_extents(super
, dl
);
1516 unsigned long long esize
;
1517 esize
= e
[i
].start
- pos
;
1518 if (esize
>= maxsize
)
1520 pos
= e
[i
].start
+ e
[i
].size
;
1522 } while (e
[i
-1].size
);
1523 *freesize
= maxsize
;
1528 static int validate_geometry_imsm(struct supertype
*st
, int level
, int layout
,
1529 int raiddisks
, int chunk
, unsigned long long size
,
1530 char *dev
, unsigned long long *freesize
,
1536 /* if given unused devices create a container
1537 * if given given devices in a container create a member volume
1539 if (level
== LEVEL_CONTAINER
) {
1540 /* Must be a fresh device to add to a container */
1541 return validate_geometry_imsm_container(st
, level
, layout
,
1542 raiddisks
, chunk
, size
,
1548 /* creating in a given container */
1549 return validate_geometry_imsm_volume(st
, level
, layout
,
1550 raiddisks
, chunk
, size
,
1551 dev
, freesize
, verbose
);
1554 /* limit creation to the following levels */
1566 /* This device needs to be a device in an 'imsm' container */
1567 fd
= open(dev
, O_RDONLY
|O_EXCL
, 0);
1571 Name
": Cannot create this array on device %s\n",
1576 if (errno
!= EBUSY
|| (fd
= open(dev
, O_RDONLY
, 0)) < 0) {
1578 fprintf(stderr
, Name
": Cannot open %s: %s\n",
1579 dev
, strerror(errno
));
1582 /* Well, it is in use by someone, maybe an 'imsm' container. */
1583 cfd
= open_container(fd
);
1587 fprintf(stderr
, Name
": Cannot use %s: It is busy\n",
1591 sra
= sysfs_read(cfd
, 0, GET_VERSION
);
1593 if (sra
&& sra
->array
.major_version
== -1 &&
1594 strcmp(sra
->text_version
, "imsm") == 0) {
1595 /* This is a member of a imsm container. Load the container
1596 * and try to create a volume
1598 struct intel_super
*super
;
1600 if (load_super_imsm_all(st
, cfd
, (void **) &super
, NULL
, 1) == 0) {
1602 st
->container_dev
= fd2devnum(cfd
);
1604 return validate_geometry_imsm_volume(st
, level
, layout
,
1610 } else /* may belong to another container */
1616 static struct mdinfo
*container_content_imsm(struct supertype
*st
)
1618 /* Given a container loaded by load_super_imsm_all,
1619 * extract information about all the arrays into
1622 * For each imsm_dev create an mdinfo, fill it in,
1623 * then look for matching devices in super->disks
1624 * and create appropriate device mdinfo.
1626 struct intel_super
*super
= st
->sb
;
1627 struct imsm_super
*mpb
= super
->mpb
;
1628 struct mdinfo
*rest
= NULL
;
1631 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
1632 struct imsm_dev
*dev
= get_imsm_dev(mpb
, i
);
1633 struct imsm_vol
*vol
= &dev
->vol
;
1634 struct imsm_map
*map
= vol
->map
;
1635 struct mdinfo
*this;
1639 this = malloc(sizeof(*this));
1640 memset(this, 0, sizeof(*this));
1644 this->array
.level
= get_imsm_raid_level(map
);
1645 this->array
.raid_disks
= map
->num_members
;
1646 this->array
.layout
= imsm_level_to_layout(this->array
.level
);
1647 this->array
.md_minor
= -1;
1648 this->array
.ctime
= 0;
1649 this->array
.utime
= 0;
1650 this->array
.chunk_size
= __le16_to_cpu(map
->blocks_per_strip
) << 9;
1651 this->array
.state
= !vol
->dirty
;
1652 this->container_member
= i
;
1653 if (map
->map_state
== IMSM_T_STATE_UNINITIALIZED
|| dev
->vol
.dirty
)
1654 this->resync_start
= 0;
1656 this->resync_start
= ~0ULL;
1658 strncpy(this->name
, (char *) dev
->volume
, MAX_RAID_SERIAL_LEN
);
1659 this->name
[MAX_RAID_SERIAL_LEN
] = 0;
1661 sprintf(this->text_version
, "/%s/%d",
1662 devnum2devname(st
->container_dev
),
1663 this->container_member
);
1665 memset(this->uuid
, 0, sizeof(this->uuid
));
1667 sz
= __le32_to_cpu(dev
->size_high
);
1669 sz
+= __le32_to_cpu(dev
->size_low
);
1670 this->component_size
= sz
;
1671 this->array
.size
= this->component_size
/ 2;
1673 for (slot
= 0 ; slot
< map
->num_members
; slot
++) {
1674 struct imsm_disk
*disk
;
1675 struct mdinfo
*info_d
;
1680 idx
= __le32_to_cpu(map
->disk_ord_tbl
[slot
] & ~(0xff << 24));
1681 for (d
= super
->disks
; d
; d
= d
->next
)
1682 if (d
->index
== idx
)
1686 break; /* shouldn't this be continue ?? */
1688 info_d
= malloc(sizeof(*info_d
));
1690 break; /* ditto ?? */
1691 memset(info_d
, 0, sizeof(*info_d
));
1692 info_d
->next
= this->devs
;
1693 this->devs
= info_d
;
1695 disk
= get_imsm_disk(mpb
, idx
);
1696 s
= __le32_to_cpu(disk
->status
);
1698 info_d
->disk
.number
= d
->index
;
1699 info_d
->disk
.major
= d
->major
;
1700 info_d
->disk
.minor
= d
->minor
;
1701 info_d
->disk
.raid_disk
= slot
;
1702 info_d
->disk
.state
= s
& CONFIGURED_DISK
? (1 << MD_DISK_ACTIVE
) : 0;
1703 info_d
->disk
.state
|= s
& FAILED_DISK
? (1 << MD_DISK_FAULTY
) : 0;
1704 info_d
->disk
.state
|= s
& USABLE_DISK
? (1 << MD_DISK_SYNC
) : 0;
1706 this->array
.working_disks
++;
1708 info_d
->events
= __le32_to_cpu(mpb
->generation_num
);
1709 info_d
->data_offset
= __le32_to_cpu(map
->pba_of_lba0
);
1710 info_d
->component_size
= __le32_to_cpu(map
->blocks_per_member
);
1712 strcpy(info_d
->name
, d
->devname
);
1720 static int imsm_open_new(struct supertype
*c
, struct active_array
*a
,
1723 struct intel_super
*super
= c
->sb
;
1724 struct imsm_super
*mpb
= super
->mpb
;
1726 if (atoi(inst
) + 1 > mpb
->num_raid_devs
) {
1727 fprintf(stderr
, "%s: subarry index %d, out of range\n",
1728 __func__
, atoi(inst
));
1732 dprintf("imsm: open_new %s\n", inst
);
1733 a
->info
.container_member
= atoi(inst
);
1737 static __u8
imsm_check_degraded(struct imsm_super
*mpb
, int n
, int failed
)
1739 struct imsm_dev
*dev
= get_imsm_dev(mpb
, n
);
1740 struct imsm_map
*map
= dev
->vol
.map
;
1743 return map
->map_state
;
1745 switch (get_imsm_raid_level(map
)) {
1747 return IMSM_T_STATE_FAILED
;
1750 if (failed
< map
->num_members
)
1751 return IMSM_T_STATE_DEGRADED
;
1753 return IMSM_T_STATE_FAILED
;
1758 * check to see if any mirrors have failed,
1759 * otherwise we are degraded
1761 int device_per_mirror
= 2; /* FIXME is this always the case?
1762 * and are they always adjacent?
1767 for (i
= 0; i
< map
->num_members
; i
++) {
1768 int idx
= get_imsm_disk_idx(map
, i
);
1769 struct imsm_disk
*disk
= get_imsm_disk(mpb
, idx
);
1771 if (__le32_to_cpu(disk
->status
) & FAILED_DISK
)
1774 if (failed
>= device_per_mirror
)
1775 return IMSM_T_STATE_FAILED
;
1777 /* reset 'failed' for next mirror set */
1778 if (!((i
+ 1) % device_per_mirror
))
1782 return IMSM_T_STATE_DEGRADED
;
1786 return IMSM_T_STATE_DEGRADED
;
1788 return IMSM_T_STATE_FAILED
;
1794 return map
->map_state
;
1797 static int imsm_count_failed(struct imsm_super
*mpb
, struct imsm_map
*map
)
1801 struct imsm_disk
*disk
;
1803 for (i
= 0; i
< map
->num_members
; i
++) {
1804 int idx
= get_imsm_disk_idx(map
, i
);
1806 disk
= get_imsm_disk(mpb
, idx
);
1807 if (__le32_to_cpu(disk
->status
) & FAILED_DISK
)
1814 static void imsm_set_array_state(struct active_array
*a
, int consistent
)
1816 int inst
= a
->info
.container_member
;
1817 struct intel_super
*super
= a
->container
->sb
;
1818 struct imsm_dev
*dev
= get_imsm_dev(super
->mpb
, inst
);
1819 struct imsm_map
*map
= &dev
->vol
.map
[0];
1820 int dirty
= !consistent
;
1824 if (a
->resync_start
== ~0ULL) {
1825 failed
= imsm_count_failed(super
->mpb
, map
);
1826 map_state
= imsm_check_degraded(super
->mpb
, inst
, failed
);
1828 map_state
= IMSM_T_STATE_NORMAL
;
1829 if (map
->map_state
!= map_state
) {
1830 dprintf("imsm: map_state %d: %d\n",
1832 map
->map_state
= map_state
;
1833 super
->updates_pending
++;
1837 if (dev
->vol
.dirty
!= dirty
) {
1838 dprintf("imsm: mark '%s' (%llu)\n",
1839 dirty
?"dirty":"clean", a
->resync_start
);
1841 dev
->vol
.dirty
= dirty
;
1842 super
->updates_pending
++;
1846 static void imsm_set_disk(struct active_array
*a
, int n
, int state
)
1848 int inst
= a
->info
.container_member
;
1849 struct intel_super
*super
= a
->container
->sb
;
1850 struct imsm_dev
*dev
= get_imsm_dev(super
->mpb
, inst
);
1851 struct imsm_map
*map
= dev
->vol
.map
;
1852 struct imsm_disk
*disk
;
1855 int new_failure
= 0;
1857 if (n
> map
->num_members
)
1858 fprintf(stderr
, "imsm: set_disk %d out of range 0..%d\n",
1859 n
, map
->num_members
- 1);
1864 dprintf("imsm: set_disk %d:%x\n", n
, state
);
1866 disk
= get_imsm_disk(super
->mpb
, get_imsm_disk_idx(map
, n
));
1868 /* check for new failures */
1869 status
= __le32_to_cpu(disk
->status
);
1870 if ((state
& DS_FAULTY
) && !(status
& FAILED_DISK
)) {
1871 status
|= FAILED_DISK
;
1872 disk
->status
= __cpu_to_le32(status
);
1874 super
->updates_pending
++;
1877 /* the number of failures have changed, count up 'failed' to determine
1878 * degraded / failed status
1880 if (new_failure
&& map
->map_state
!= IMSM_T_STATE_FAILED
)
1881 failed
= imsm_count_failed(super
->mpb
, map
);
1883 /* determine map_state based on failed or in_sync count */
1885 map
->map_state
= imsm_check_degraded(super
->mpb
, inst
, failed
);
1886 else if (map
->map_state
== IMSM_T_STATE_DEGRADED
) {
1890 for (d
= a
->info
.devs
; d
; d
= d
->next
)
1891 if (d
->curr_state
& DS_INSYNC
)
1894 if (working
== a
->info
.array
.raid_disks
) {
1895 map
->map_state
= IMSM_T_STATE_NORMAL
;
1896 super
->updates_pending
++;
1901 static int store_imsm_mpb(int fd
, struct intel_super
*super
)
1903 struct imsm_super
*mpb
= super
->mpb
;
1904 __u32 mpb_size
= __le32_to_cpu(mpb
->mpb_size
);
1905 unsigned long long dsize
;
1906 unsigned long long sectors
;
1908 get_dev_size(fd
, NULL
, &dsize
);
1910 if (mpb_size
> 512) {
1911 /* -1 to account for anchor */
1912 sectors
= mpb_sectors(mpb
) - 1;
1914 /* write the extended mpb to the sectors preceeding the anchor */
1915 if (lseek64(fd
, dsize
- (512 * (2 + sectors
)), SEEK_SET
) < 0)
1918 if (write(fd
, super
->buf
+ 512, 512 * sectors
) != 512 * sectors
)
1922 /* first block is stored on second to last sector of the disk */
1923 if (lseek64(fd
, dsize
- (512 * 2), SEEK_SET
) < 0)
1926 if (write(fd
, super
->buf
, 512) != 512)
1932 static void imsm_sync_metadata(struct supertype
*container
)
1934 struct intel_super
*super
= container
->sb
;
1936 if (!super
->updates_pending
)
1939 write_super_imsm(super
, 0);
1941 super
->updates_pending
= 0;
1944 static struct mdinfo
*imsm_activate_spare(struct active_array
*a
,
1945 struct metadata_update
**updates
)
1948 * Take a device that is marked spare in the metadata and use it to
1949 * replace a failed/vacant slot in an array. There may be a case where
1950 * a device is failed in one array but active in a second.
1951 * imsm_process_update catches this case and does not clear the SPARE_DISK
1952 * flag, allowing the second array to start using the device on failure.
1953 * SPARE_DISK is cleared when all arrays are using a device.
1955 * FIXME: is this a valid use of SPARE_DISK?
1958 struct intel_super
*super
= a
->container
->sb
;
1959 struct imsm_super
*mpb
= super
->mpb
;
1960 int inst
= a
->info
.container_member
;
1961 struct imsm_dev
*dev
= get_imsm_dev(mpb
, inst
);
1962 struct imsm_map
*map
= dev
->vol
.map
;
1963 int failed
= a
->info
.array
.raid_disks
;
1964 struct mdinfo
*rv
= NULL
;
1967 struct metadata_update
*mu
;
1969 struct imsm_update_activate_spare
*u
;
1973 for (d
= a
->info
.devs
; d
; d
= d
->next
) {
1974 if ((d
->curr_state
& DS_FAULTY
) &&
1976 /* wait for Removal to happen */
1978 if (d
->state_fd
>= 0)
1982 dprintf("imsm: activate spare: inst=%d failed=%d (%d) level=%d\n",
1983 inst
, failed
, a
->info
.array
.raid_disks
, a
->info
.array
.level
);
1984 if (imsm_check_degraded(mpb
, inst
, failed
) != IMSM_T_STATE_DEGRADED
)
1987 /* For each slot, if it is not working, find a spare */
1989 for (i
= 0; i
< a
->info
.array
.raid_disks
; i
++) {
1990 for (d
= a
->info
.devs
; d
; d
= d
->next
)
1991 if (d
->disk
.raid_disk
== i
)
1993 dprintf("found %d: %p %x\n", i
, d
, d
?d
->curr_state
:0);
1994 if (d
&& (d
->state_fd
>= 0))
1997 /* OK, this device needs recovery. Find a spare */
1998 for ( ; dl
; dl
= dl
->next
) {
1999 unsigned long long esize
;
2000 unsigned long long pos
;
2003 struct imsm_disk
*disk
;
2008 /* If in this array, skip */
2009 for (d2
= a
->info
.devs
; d2
; d2
= d2
->next
)
2010 if (d2
->disk
.major
== dl
->major
&&
2011 d2
->disk
.minor
== dl
->minor
) {
2012 dprintf("%x:%x already in array\n", dl
->major
, dl
->minor
);
2018 /* is this unused device marked as a spare? */
2019 disk
= get_imsm_disk(mpb
, dl
->index
);
2020 if (!(__le32_to_cpu(disk
->status
) & SPARE_DISK
))
2023 /* We are allowed to use this device - is there space?
2024 * We need a->info.component_size sectors */
2025 ex
= get_extents(super
, dl
);
2027 dprintf("cannot get extents\n");
2033 array_start
= __le32_to_cpu(map
->pba_of_lba0
);
2036 /* check that we can start at pba_of_lba0 with
2037 * a->info.component_size of space
2039 esize
= ex
[j
].start
- pos
;
2040 if (array_start
>= pos
&&
2041 array_start
+ a
->info
.component_size
< ex
[j
].start
) {
2045 pos
= ex
[j
].start
+ ex
[j
].size
;
2048 } while (ex
[j
-1].size
);
2052 dprintf("%x:%x does not have %llu at %d\n",
2053 dl
->major
, dl
->minor
,
2054 a
->info
.component_size
,
2055 __le32_to_cpu(map
->pba_of_lba0
));
2060 /* found a usable disk with enough space */
2061 di
= malloc(sizeof(*di
));
2062 memset(di
, 0, sizeof(*di
));
2063 di
->disk
.number
= dl
->index
;
2064 di
->disk
.raid_disk
= i
;
2065 di
->disk
.major
= dl
->major
;
2066 di
->disk
.minor
= dl
->minor
;
2068 di
->data_offset
= array_start
;
2069 di
->component_size
= a
->info
.component_size
;
2070 di
->container_member
= inst
;
2074 dprintf("%x:%x to be %d at %llu\n", dl
->major
, dl
->minor
,
2082 /* No spares found */
2084 /* Now 'rv' has a list of devices to return.
2085 * Create a metadata_update record to update the
2086 * disk_ord_tbl for the array
2088 mu
= malloc(sizeof(*mu
));
2089 mu
->buf
= malloc(sizeof(struct imsm_update_activate_spare
) * num_spares
);
2091 mu
->len
= sizeof(struct imsm_update_activate_spare
) * num_spares
;
2092 mu
->next
= *updates
;
2093 u
= (struct imsm_update_activate_spare
*) mu
->buf
;
2095 for (di
= rv
; di
; di
= di
->next
) {
2096 u
->type
= update_activate_spare
;
2097 u
->disk_idx
= di
->disk
.number
;
2098 u
->slot
= di
->disk
.raid_disk
;
2109 static int weight(unsigned int field
)
2113 for (weight
= 0; field
; weight
++)
2119 static int disks_overlap(struct imsm_map
*m1
, struct imsm_map
*m2
)
2125 for (i
= 0; i
< m1
->num_members
; i
++) {
2126 idx
= get_imsm_disk_idx(m1
, i
);
2127 for (j
= 0; j
< m2
->num_members
; j
++)
2128 if (idx
== get_imsm_disk_idx(m2
, j
))
2135 static void imsm_process_update(struct supertype
*st
,
2136 struct metadata_update
*update
)
2139 * crack open the metadata_update envelope to find the update record
2140 * update can be one of:
2141 * update_activate_spare - a spare device has replaced a failed
2142 * device in an array, update the disk_ord_tbl. If this disk is
2143 * present in all member arrays then also clear the SPARE_DISK
2146 struct intel_super
*super
= st
->sb
;
2147 struct imsm_super
*mpb
= super
->mpb
;
2148 enum imsm_update_type type
= *(enum imsm_update_type
*) update
->buf
;
2151 case update_activate_spare
: {
2152 struct imsm_update_activate_spare
*u
= (void *) update
->buf
;
2153 struct imsm_dev
*dev
= get_imsm_dev(mpb
, u
->array
);
2154 struct imsm_map
*map
= &dev
->vol
.map
[0];
2155 struct active_array
*a
;
2156 struct imsm_disk
*disk
;
2160 unsigned int members
;
2165 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
2166 if (dl
->index
== u
->disk_idx
)
2170 fprintf(stderr
, "error: imsm_activate_spare passed "
2171 "an unknown disk_idx: %d\n", u
->disk_idx
);
2175 super
->updates_pending
++;
2177 victim
= get_imsm_disk_idx(map
, u
->slot
);
2178 map
->disk_ord_tbl
[u
->slot
] = __cpu_to_le32(u
->disk_idx
);
2179 disk
= get_imsm_disk(mpb
, u
->disk_idx
);
2180 status
= __le32_to_cpu(disk
->status
);
2181 status
|= CONFIGURED_DISK
;
2182 disk
->status
= __cpu_to_le32(status
);
2184 /* map unique/live arrays using the spare */
2187 for (a
= st
->arrays
; a
; a
= a
->next
) {
2188 int inst
= a
->info
.container_member
;
2190 dev
= get_imsm_dev(mpb
, inst
);
2191 map
= &dev
->vol
.map
[0];
2192 if (map
->raid_level
> 0)
2193 members
|= 1 << inst
;
2194 for (d
= a
->info
.devs
; d
; d
= d
->next
)
2195 if (d
->disk
.major
== dl
->major
&&
2196 d
->disk
.minor
== dl
->minor
)
2200 /* until all arrays that can absorb this disk have absorbed
2201 * this disk it can still be considered a spare
2203 if (weight(found
) >= weight(members
)) {
2204 status
= __le32_to_cpu(disk
->status
);
2205 status
&= ~SPARE_DISK
;
2206 disk
->status
= __cpu_to_le32(status
);
2209 /* count arrays using the victim in the metadata */
2211 for (a
= st
->arrays
; a
; a
= a
->next
) {
2212 dev
= get_imsm_dev(mpb
, a
->info
.container_member
);
2213 map
= &dev
->vol
.map
[0];
2214 for (i
= 0; i
< map
->num_members
; i
++)
2215 if (victim
== get_imsm_disk_idx(map
, i
))
2219 /* clear some flags if the victim is no longer being
2222 disk
= get_imsm_disk(mpb
, victim
);
2224 status
= __le32_to_cpu(disk
->status
);
2225 status
&= ~(CONFIGURED_DISK
| USABLE_DISK
);
2226 disk
->status
= __cpu_to_le32(status
);
2230 case update_create_array
: {
2231 /* someone wants to create a new array, we need to be aware of
2232 * a few races/collisions:
2233 * 1/ 'Create' called by two separate instances of mdadm
2234 * 2/ 'Create' versus 'activate_spare': mdadm has chosen
2235 * devices that have since been assimilated via
2237 * In the event this update can not be carried out mdadm will
2238 * (FIX ME) notice that its update did not take hold.
2240 struct imsm_update_create_array
*u
= (void *) update
->buf
;
2241 struct imsm_dev
*dev
;
2242 struct imsm_map
*map
, *new_map
;
2243 unsigned long long start
, end
;
2244 unsigned long long new_start
, new_end
;
2248 /* handle racing creates: first come first serve */
2249 if (u
->dev_idx
< mpb
->num_raid_devs
) {
2250 dprintf("%s: subarray %d already defined\n",
2251 __func__
, u
->dev_idx
);
2255 /* check update is next in sequence */
2256 if (u
->dev_idx
!= mpb
->num_raid_devs
) {
2257 dprintf("%s: can not create arrays out of sequence\n",
2262 new_map
= &u
->dev
.vol
.map
[0];
2263 new_start
= __le32_to_cpu(new_map
->pba_of_lba0
);
2264 new_end
= new_start
+ __le32_to_cpu(new_map
->blocks_per_member
);
2266 /* handle activate_spare versus create race:
2267 * check to make sure that overlapping arrays do not include
2270 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
2271 dev
= get_imsm_dev(mpb
, i
);
2272 map
= &dev
->vol
.map
[0];
2273 start
= __le32_to_cpu(map
->pba_of_lba0
);
2274 end
= start
+ __le32_to_cpu(map
->blocks_per_member
);
2275 if ((new_start
>= start
&& new_start
<= end
) ||
2276 (start
>= new_start
&& start
<= new_end
))
2278 if (overlap
&& disks_overlap(map
, new_map
)) {
2279 dprintf("%s: arrays overlap\n", __func__
);
2283 /* check num_members sanity */
2284 if (new_map
->num_members
> mpb
->num_disks
) {
2285 dprintf("%s: num_disks out of range\n", __func__
);
2289 super
->updates_pending
++;
2290 mpb
->num_raid_devs
++;
2291 dev
= get_imsm_dev(mpb
, u
->dev_idx
);
2292 memcpy(dev
, &u
->dev
, sizeof(*dev
));
2293 map
= &dev
->vol
.map
[0];
2294 memcpy(map
->disk_ord_tbl
, new_map
->disk_ord_tbl
,
2295 sizeof(__u32
) * new_map
->num_members
);
2297 /* fix up flags, if arrays overlap then the drives can not be
2300 for (i
= 0; i
< map
->num_members
; i
++) {
2301 struct imsm_disk
*disk
;
2304 disk
= get_imsm_disk(mpb
, get_imsm_disk_idx(map
, i
));
2305 status
= __le32_to_cpu(disk
->status
);
2306 status
|= CONFIGURED_DISK
;
2308 status
&= ~SPARE_DISK
;
2309 disk
->status
= __cpu_to_le32(status
);
2316 static void imsm_prepare_update(struct supertype
*st
,
2317 struct metadata_update
*update
)
2319 /* Allocate space to hold a new mpb if necessary. We currently
2320 * allocate enough to hold 2 subarrays for the given number of disks.
2321 * This may not be sufficient iff reshaping.
2323 * FIX ME handle the reshape case.
2325 * The monitor will be able to safely change super->mpb by arranging
2326 * for it to be freed in check_update_queue(). I.e. the monitor thread
2327 * will start using the new pointer and the manager can continue to use
2328 * the old value until check_update_queue() runs.
2334 struct superswitch super_imsm
= {
2336 .examine_super
= examine_super_imsm
,
2337 .brief_examine_super
= brief_examine_super_imsm
,
2338 .detail_super
= detail_super_imsm
,
2339 .brief_detail_super
= brief_detail_super_imsm
,
2340 .write_init_super
= write_init_super_imsm
,
2342 .match_home
= match_home_imsm
,
2343 .uuid_from_super
= uuid_from_super_imsm
,
2344 .getinfo_super
= getinfo_super_imsm
,
2345 .update_super
= update_super_imsm
,
2347 .avail_size
= avail_size_imsm
,
2349 .compare_super
= compare_super_imsm
,
2351 .load_super
= load_super_imsm
,
2352 .init_super
= init_super_imsm
,
2353 .add_to_super
= add_to_super_imsm
,
2354 .store_super
= store_zero_imsm
,
2355 .free_super
= free_super_imsm
,
2356 .match_metadata_desc
= match_metadata_desc_imsm
,
2357 .container_content
= container_content_imsm
,
2359 .validate_geometry
= validate_geometry_imsm
,
2363 .open_new
= imsm_open_new
,
2364 .load_super
= load_super_imsm
,
2365 .set_array_state
= imsm_set_array_state
,
2366 .set_disk
= imsm_set_disk
,
2367 .sync_metadata
= imsm_sync_metadata
,
2368 .activate_spare
= imsm_activate_spare
,
2369 .process_update
= imsm_process_update
,
2370 .prepare_update
= imsm_prepare_update
,