2 * mdadm - Intel(R) Matrix Storage Manager Support
4 * Copyright (C) 2002-2008 Intel Corporation
6 * This program is free software; you can redistribute it and/or modify it
7 * under the terms and conditions of the GNU General Public License,
8 * version 2, as published by the Free Software Foundation.
10 * This program is distributed in the hope it will be useful, but WITHOUT
11 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
12 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for
15 * You should have received a copy of the GNU General Public License along with
16 * this program; if not, write to the Free Software Foundation, Inc.,
17 * 51 Franklin St - Fifth Floor, Boston, MA 02110-1301 USA.
20 #define HAVE_STDINT_H 1
24 #include "platform-intel.h"
30 /* MPB == Metadata Parameter Block */
31 #define MPB_SIGNATURE "Intel Raid ISM Cfg Sig. "
32 #define MPB_SIG_LEN (strlen(MPB_SIGNATURE))
33 #define MPB_VERSION_RAID0 "1.0.00"
34 #define MPB_VERSION_RAID1 "1.1.00"
35 #define MPB_VERSION_MANY_VOLUMES_PER_ARRAY "1.2.00"
36 #define MPB_VERSION_3OR4_DISK_ARRAY "1.2.01"
37 #define MPB_VERSION_RAID5 "1.2.02"
38 #define MPB_VERSION_5OR6_DISK_ARRAY "1.2.04"
39 #define MPB_VERSION_CNG "1.2.06"
40 #define MPB_VERSION_ATTRIBS "1.3.00"
41 #define MAX_SIGNATURE_LENGTH 32
42 #define MAX_RAID_SERIAL_LEN 16
44 #define MPB_ATTRIB_CHECKSUM_VERIFY __cpu_to_le32(0x80000000)
45 #define MPB_ATTRIB_PM __cpu_to_le32(0x40000000)
46 #define MPB_ATTRIB_2TB __cpu_to_le32(0x20000000)
47 #define MPB_ATTRIB_RAID0 __cpu_to_le32(0x00000001)
48 #define MPB_ATTRIB_RAID1 __cpu_to_le32(0x00000002)
49 #define MPB_ATTRIB_RAID10 __cpu_to_le32(0x00000004)
50 #define MPB_ATTRIB_RAID1E __cpu_to_le32(0x00000008)
51 #define MPB_ATTRIB_RAID5 __cpu_to_le32(0x00000010)
52 #define MPB_ATTRIB_RAIDCNG __cpu_to_le32(0x00000020)
54 #define MPB_SECTOR_CNT 418
55 #define IMSM_RESERVED_SECTORS 4096
56 #define SECT_PER_MB_SHIFT 11
58 /* Disk configuration info. */
59 #define IMSM_MAX_DEVICES 255
61 __u8 serial
[MAX_RAID_SERIAL_LEN
];/* 0xD8 - 0xE7 ascii serial number */
62 __u32 total_blocks
; /* 0xE8 - 0xEB total blocks */
63 __u32 scsi_id
; /* 0xEC - 0xEF scsi ID */
64 #define SPARE_DISK __cpu_to_le32(0x01) /* Spare */
65 #define CONFIGURED_DISK __cpu_to_le32(0x02) /* Member of some RaidDev */
66 #define FAILED_DISK __cpu_to_le32(0x04) /* Permanent failure */
67 __u32 status
; /* 0xF0 - 0xF3 */
68 __u32 owner_cfg_num
; /* which config 0,1,2... owns this disk */
69 #define IMSM_DISK_FILLERS 4
70 __u32 filler
[IMSM_DISK_FILLERS
]; /* 0xF4 - 0x107 MPB_DISK_FILLERS for future expansion */
73 /* RAID map configuration infos. */
75 __u32 pba_of_lba0
; /* start address of partition */
76 __u32 blocks_per_member
;/* blocks per member */
77 __u32 num_data_stripes
; /* number of data stripes */
78 __u16 blocks_per_strip
;
79 __u8 map_state
; /* Normal, Uninitialized, Degraded, Failed */
80 #define IMSM_T_STATE_NORMAL 0
81 #define IMSM_T_STATE_UNINITIALIZED 1
82 #define IMSM_T_STATE_DEGRADED 2
83 #define IMSM_T_STATE_FAILED 3
85 #define IMSM_T_RAID0 0
86 #define IMSM_T_RAID1 1
87 #define IMSM_T_RAID5 5 /* since metadata version 1.2.02 ? */
88 __u8 num_members
; /* number of member disks */
89 __u8 num_domains
; /* number of parity domains */
90 __u8 failed_disk_num
; /* valid only when state is degraded */
92 __u32 filler
[7]; /* expansion area */
93 #define IMSM_ORD_REBUILD (1 << 24)
94 __u32 disk_ord_tbl
[1]; /* disk_ord_tbl[num_members],
95 * top byte contains some flags
97 } __attribute__ ((packed
));
100 __u32 curr_migr_unit
;
101 __u32 checkpoint_id
; /* id to access curr_migr_unit */
102 __u8 migr_state
; /* Normal or Migrating */
104 #define MIGR_REBUILD 1
105 #define MIGR_VERIFY 2 /* analagous to echo check > sync_action */
106 #define MIGR_GEN_MIGR 3
107 #define MIGR_STATE_CHANGE 4
108 #define MIGR_REPAIR 5
109 __u8 migr_type
; /* Initializing, Rebuilding, ... */
111 __u8 fs_state
; /* fast-sync state for CnG (0xff == disabled) */
112 __u16 verify_errors
; /* number of mismatches */
113 __u16 bad_blocks
; /* number of bad blocks during verify */
115 struct imsm_map map
[1];
116 /* here comes another one if migr_state */
117 } __attribute__ ((packed
));
120 __u8 volume
[MAX_RAID_SERIAL_LEN
];
123 #define DEV_BOOTABLE __cpu_to_le32(0x01)
124 #define DEV_BOOT_DEVICE __cpu_to_le32(0x02)
125 #define DEV_READ_COALESCING __cpu_to_le32(0x04)
126 #define DEV_WRITE_COALESCING __cpu_to_le32(0x08)
127 #define DEV_LAST_SHUTDOWN_DIRTY __cpu_to_le32(0x10)
128 #define DEV_HIDDEN_AT_BOOT __cpu_to_le32(0x20)
129 #define DEV_CURRENTLY_HIDDEN __cpu_to_le32(0x40)
130 #define DEV_VERIFY_AND_FIX __cpu_to_le32(0x80)
131 #define DEV_MAP_STATE_UNINIT __cpu_to_le32(0x100)
132 #define DEV_NO_AUTO_RECOVERY __cpu_to_le32(0x200)
133 #define DEV_CLONE_N_GO __cpu_to_le32(0x400)
134 #define DEV_CLONE_MAN_SYNC __cpu_to_le32(0x800)
135 #define DEV_CNG_MASTER_DISK_NUM __cpu_to_le32(0x1000)
136 __u32 status
; /* Persistent RaidDev status */
137 __u32 reserved_blocks
; /* Reserved blocks at beginning of volume */
141 __u8 cng_master_disk
;
145 #define IMSM_DEV_FILLERS 10
146 __u32 filler
[IMSM_DEV_FILLERS
];
148 } __attribute__ ((packed
));
151 __u8 sig
[MAX_SIGNATURE_LENGTH
]; /* 0x00 - 0x1F */
152 __u32 check_sum
; /* 0x20 - 0x23 MPB Checksum */
153 __u32 mpb_size
; /* 0x24 - 0x27 Size of MPB */
154 __u32 family_num
; /* 0x28 - 0x2B Checksum from first time this config was written */
155 __u32 generation_num
; /* 0x2C - 0x2F Incremented each time this array's MPB is written */
156 __u32 error_log_size
; /* 0x30 - 0x33 in bytes */
157 __u32 attributes
; /* 0x34 - 0x37 */
158 __u8 num_disks
; /* 0x38 Number of configured disks */
159 __u8 num_raid_devs
; /* 0x39 Number of configured volumes */
160 __u8 error_log_pos
; /* 0x3A */
161 __u8 fill
[1]; /* 0x3B */
162 __u32 cache_size
; /* 0x3c - 0x40 in mb */
163 __u32 orig_family_num
; /* 0x40 - 0x43 original family num */
164 __u32 pwr_cycle_count
; /* 0x44 - 0x47 simulated power cycle count for array */
165 __u32 bbm_log_size
; /* 0x48 - 0x4B - size of bad Block Mgmt Log in bytes */
166 #define IMSM_FILLERS 35
167 __u32 filler
[IMSM_FILLERS
]; /* 0x4C - 0xD7 RAID_MPB_FILLERS */
168 struct imsm_disk disk
[1]; /* 0xD8 diskTbl[numDisks] */
169 /* here comes imsm_dev[num_raid_devs] */
170 /* here comes BBM logs */
171 } __attribute__ ((packed
));
173 #define BBM_LOG_MAX_ENTRIES 254
175 struct bbm_log_entry
{
176 __u64 defective_block_start
;
177 #define UNREADABLE 0xFFFFFFFF
178 __u32 spare_block_offset
;
179 __u16 remapped_marked_count
;
181 } __attribute__ ((__packed__
));
184 __u32 signature
; /* 0xABADB10C */
186 __u32 reserved_spare_block_count
; /* 0 */
187 __u32 reserved
; /* 0xFFFF */
188 __u64 first_spare_lba
;
189 struct bbm_log_entry mapped_block_entries
[BBM_LOG_MAX_ENTRIES
];
190 } __attribute__ ((__packed__
));
194 static char *map_state_str
[] = { "normal", "uninitialized", "degraded", "failed" };
197 static __u8
migr_type(struct imsm_dev
*dev
)
199 if (dev
->vol
.migr_type
== MIGR_VERIFY
&&
200 dev
->status
& DEV_VERIFY_AND_FIX
)
203 return dev
->vol
.migr_type
;
206 static void set_migr_type(struct imsm_dev
*dev
, __u8 migr_type
)
208 /* for compatibility with older oroms convert MIGR_REPAIR, into
209 * MIGR_VERIFY w/ DEV_VERIFY_AND_FIX status
211 if (migr_type
== MIGR_REPAIR
) {
212 dev
->vol
.migr_type
= MIGR_VERIFY
;
213 dev
->status
|= DEV_VERIFY_AND_FIX
;
215 dev
->vol
.migr_type
= migr_type
;
216 dev
->status
&= ~DEV_VERIFY_AND_FIX
;
220 static unsigned int sector_count(__u32 bytes
)
222 return ((bytes
+ (512-1)) & (~(512-1))) / 512;
225 static unsigned int mpb_sectors(struct imsm_super
*mpb
)
227 return sector_count(__le32_to_cpu(mpb
->mpb_size
));
231 struct imsm_dev
*dev
;
232 struct intel_dev
*next
;
236 /* internal representation of IMSM metadata */
239 void *buf
; /* O_DIRECT buffer for reading/writing metadata */
240 struct imsm_super
*anchor
; /* immovable parameters */
242 size_t len
; /* size of the 'buf' allocation */
243 void *next_buf
; /* for realloc'ing buf from the manager */
245 int updates_pending
; /* count of pending updates for mdmon */
246 int current_vol
; /* index of raid device undergoing creation */
247 __u32 create_offset
; /* common start for 'current_vol' */
248 __u32 random
; /* random data for seeding new family numbers */
249 struct intel_dev
*devlist
;
253 __u8 serial
[MAX_RAID_SERIAL_LEN
];
256 struct imsm_disk disk
;
259 struct extent
*e
; /* for determining freespace @ create */
260 int raiddisk
; /* slot to fill in autolayout */
262 struct dl
*add
; /* list of disks to add while mdmon active */
263 struct dl
*missing
; /* disks removed while we weren't looking */
264 struct bbm_log
*bbm_log
;
265 const char *hba
; /* device path of the raid controller for this metadata */
266 const struct imsm_orom
*orom
; /* platform firmware support */
267 struct intel_super
*next
; /* (temp) list for disambiguating family_num */
271 struct imsm_disk disk
;
272 #define IMSM_UNKNOWN_OWNER (-1)
274 struct intel_disk
*next
;
278 unsigned long long start
, size
;
281 /* definition of messages passed to imsm_process_update */
282 enum imsm_update_type
{
283 update_activate_spare
,
290 struct imsm_update_activate_spare
{
291 enum imsm_update_type type
;
295 struct imsm_update_activate_spare
*next
;
299 __u8 serial
[MAX_RAID_SERIAL_LEN
];
302 struct imsm_update_create_array
{
303 enum imsm_update_type type
;
308 struct imsm_update_kill_array
{
309 enum imsm_update_type type
;
313 struct imsm_update_rename_array
{
314 enum imsm_update_type type
;
315 __u8 name
[MAX_RAID_SERIAL_LEN
];
319 struct imsm_update_add_disk
{
320 enum imsm_update_type type
;
323 static struct supertype
*match_metadata_desc_imsm(char *arg
)
325 struct supertype
*st
;
327 if (strcmp(arg
, "imsm") != 0 &&
328 strcmp(arg
, "default") != 0
332 st
= malloc(sizeof(*st
));
335 memset(st
, 0, sizeof(*st
));
336 st
->ss
= &super_imsm
;
337 st
->max_devs
= IMSM_MAX_DEVICES
;
338 st
->minor_version
= 0;
344 static __u8
*get_imsm_version(struct imsm_super
*mpb
)
346 return &mpb
->sig
[MPB_SIG_LEN
];
350 /* retrieve a disk directly from the anchor when the anchor is known to be
351 * up-to-date, currently only at load time
353 static struct imsm_disk
*__get_imsm_disk(struct imsm_super
*mpb
, __u8 index
)
355 if (index
>= mpb
->num_disks
)
357 return &mpb
->disk
[index
];
361 /* retrieve a disk from the parsed metadata */
362 static struct imsm_disk
*get_imsm_disk(struct intel_super
*super
, __u8 index
)
366 for (d
= super
->disks
; d
; d
= d
->next
)
367 if (d
->index
== index
)
374 /* generate a checksum directly from the anchor when the anchor is known to be
375 * up-to-date, currently only at load or write_super after coalescing
377 static __u32
__gen_imsm_checksum(struct imsm_super
*mpb
)
379 __u32 end
= mpb
->mpb_size
/ sizeof(end
);
380 __u32
*p
= (__u32
*) mpb
;
384 sum
+= __le32_to_cpu(*p
);
388 return sum
- __le32_to_cpu(mpb
->check_sum
);
391 static size_t sizeof_imsm_map(struct imsm_map
*map
)
393 return sizeof(struct imsm_map
) + sizeof(__u32
) * (map
->num_members
- 1);
396 struct imsm_map
*get_imsm_map(struct imsm_dev
*dev
, int second_map
)
398 struct imsm_map
*map
= &dev
->vol
.map
[0];
400 if (second_map
&& !dev
->vol
.migr_state
)
402 else if (second_map
) {
405 return ptr
+ sizeof_imsm_map(map
);
411 /* return the size of the device.
412 * migr_state increases the returned size if map[0] were to be duplicated
414 static size_t sizeof_imsm_dev(struct imsm_dev
*dev
, int migr_state
)
416 size_t size
= sizeof(*dev
) - sizeof(struct imsm_map
) +
417 sizeof_imsm_map(get_imsm_map(dev
, 0));
419 /* migrating means an additional map */
420 if (dev
->vol
.migr_state
)
421 size
+= sizeof_imsm_map(get_imsm_map(dev
, 1));
423 size
+= sizeof_imsm_map(get_imsm_map(dev
, 0));
429 /* retrieve disk serial number list from a metadata update */
430 static struct disk_info
*get_disk_info(struct imsm_update_create_array
*update
)
433 struct disk_info
*inf
;
435 inf
= u
+ sizeof(*update
) - sizeof(struct imsm_dev
) +
436 sizeof_imsm_dev(&update
->dev
, 0);
442 static struct imsm_dev
*__get_imsm_dev(struct imsm_super
*mpb
, __u8 index
)
448 if (index
>= mpb
->num_raid_devs
)
451 /* devices start after all disks */
452 offset
= ((void *) &mpb
->disk
[mpb
->num_disks
]) - _mpb
;
454 for (i
= 0; i
<= index
; i
++)
456 return _mpb
+ offset
;
458 offset
+= sizeof_imsm_dev(_mpb
+ offset
, 0);
463 static struct imsm_dev
*get_imsm_dev(struct intel_super
*super
, __u8 index
)
465 struct intel_dev
*dv
;
467 if (index
>= super
->anchor
->num_raid_devs
)
469 for (dv
= super
->devlist
; dv
; dv
= dv
->next
)
470 if (dv
->index
== index
)
475 static __u32
get_imsm_ord_tbl_ent(struct imsm_dev
*dev
, int slot
)
477 struct imsm_map
*map
;
479 if (dev
->vol
.migr_state
)
480 map
= get_imsm_map(dev
, 1);
482 map
= get_imsm_map(dev
, 0);
484 /* top byte identifies disk under rebuild */
485 return __le32_to_cpu(map
->disk_ord_tbl
[slot
]);
488 #define ord_to_idx(ord) (((ord) << 8) >> 8)
489 static __u32
get_imsm_disk_idx(struct imsm_dev
*dev
, int slot
)
491 __u32 ord
= get_imsm_ord_tbl_ent(dev
, slot
);
493 return ord_to_idx(ord
);
496 static void set_imsm_ord_tbl_ent(struct imsm_map
*map
, int slot
, __u32 ord
)
498 map
->disk_ord_tbl
[slot
] = __cpu_to_le32(ord
);
501 static int get_imsm_disk_slot(struct imsm_map
*map
, int idx
)
506 for (slot
= 0; slot
< map
->num_members
; slot
++) {
507 ord
= __le32_to_cpu(map
->disk_ord_tbl
[slot
]);
508 if (ord_to_idx(ord
) == idx
)
515 static int get_imsm_raid_level(struct imsm_map
*map
)
517 if (map
->raid_level
== 1) {
518 if (map
->num_members
== 2)
524 return map
->raid_level
;
527 static int cmp_extent(const void *av
, const void *bv
)
529 const struct extent
*a
= av
;
530 const struct extent
*b
= bv
;
531 if (a
->start
< b
->start
)
533 if (a
->start
> b
->start
)
538 static int count_memberships(struct dl
*dl
, struct intel_super
*super
)
543 for (i
= 0; i
< super
->anchor
->num_raid_devs
; i
++) {
544 struct imsm_dev
*dev
= get_imsm_dev(super
, i
);
545 struct imsm_map
*map
= get_imsm_map(dev
, 0);
547 if (get_imsm_disk_slot(map
, dl
->index
) >= 0)
554 static struct extent
*get_extents(struct intel_super
*super
, struct dl
*dl
)
556 /* find a list of used extents on the given physical device */
557 struct extent
*rv
, *e
;
559 int memberships
= count_memberships(dl
, super
);
560 __u32 reservation
= MPB_SECTOR_CNT
+ IMSM_RESERVED_SECTORS
;
562 rv
= malloc(sizeof(struct extent
) * (memberships
+ 1));
567 for (i
= 0; i
< super
->anchor
->num_raid_devs
; i
++) {
568 struct imsm_dev
*dev
= get_imsm_dev(super
, i
);
569 struct imsm_map
*map
= get_imsm_map(dev
, 0);
571 if (get_imsm_disk_slot(map
, dl
->index
) >= 0) {
572 e
->start
= __le32_to_cpu(map
->pba_of_lba0
);
573 e
->size
= __le32_to_cpu(map
->blocks_per_member
);
577 qsort(rv
, memberships
, sizeof(*rv
), cmp_extent
);
579 /* determine the start of the metadata
580 * when no raid devices are defined use the default
581 * ...otherwise allow the metadata to truncate the value
582 * as is the case with older versions of imsm
585 struct extent
*last
= &rv
[memberships
- 1];
588 remainder
= __le32_to_cpu(dl
->disk
.total_blocks
) -
589 (last
->start
+ last
->size
);
590 /* round down to 1k block to satisfy precision of the kernel
594 /* make sure remainder is still sane */
595 if (remainder
< ROUND_UP(super
->len
, 512) >> 9)
596 remainder
= ROUND_UP(super
->len
, 512) >> 9;
597 if (reservation
> remainder
)
598 reservation
= remainder
;
600 e
->start
= __le32_to_cpu(dl
->disk
.total_blocks
) - reservation
;
605 /* try to determine how much space is reserved for metadata from
606 * the last get_extents() entry, otherwise fallback to the
609 static __u32
imsm_reserved_sectors(struct intel_super
*super
, struct dl
*dl
)
615 /* for spares just return a minimal reservation which will grow
616 * once the spare is picked up by an array
619 return MPB_SECTOR_CNT
;
621 e
= get_extents(super
, dl
);
623 return MPB_SECTOR_CNT
+ IMSM_RESERVED_SECTORS
;
625 /* scroll to last entry */
626 for (i
= 0; e
[i
].size
; i
++)
629 rv
= __le32_to_cpu(dl
->disk
.total_blocks
) - e
[i
].start
;
636 static int is_spare(struct imsm_disk
*disk
)
638 return (disk
->status
& SPARE_DISK
) == SPARE_DISK
;
641 static int is_configured(struct imsm_disk
*disk
)
643 return (disk
->status
& CONFIGURED_DISK
) == CONFIGURED_DISK
;
646 static int is_failed(struct imsm_disk
*disk
)
648 return (disk
->status
& FAILED_DISK
) == FAILED_DISK
;
652 static __u64
blocks_per_migr_unit(struct imsm_dev
*dev
);
654 static void print_imsm_dev(struct imsm_dev
*dev
, char *uuid
, int disk_idx
)
658 struct imsm_map
*map
= get_imsm_map(dev
, 0);
662 printf("[%.16s]:\n", dev
->volume
);
663 printf(" UUID : %s\n", uuid
);
664 printf(" RAID Level : %d\n", get_imsm_raid_level(map
));
665 printf(" Members : %d\n", map
->num_members
);
666 slot
= get_imsm_disk_slot(map
, disk_idx
);
668 ord
= get_imsm_ord_tbl_ent(dev
, slot
);
669 printf(" This Slot : %d%s\n", slot
,
670 ord
& IMSM_ORD_REBUILD
? " (out-of-sync)" : "");
672 printf(" This Slot : ?\n");
673 sz
= __le32_to_cpu(dev
->size_high
);
675 sz
+= __le32_to_cpu(dev
->size_low
);
676 printf(" Array Size : %llu%s\n", (unsigned long long)sz
,
677 human_size(sz
* 512));
678 sz
= __le32_to_cpu(map
->blocks_per_member
);
679 printf(" Per Dev Size : %llu%s\n", (unsigned long long)sz
,
680 human_size(sz
* 512));
681 printf(" Sector Offset : %u\n",
682 __le32_to_cpu(map
->pba_of_lba0
));
683 printf(" Num Stripes : %u\n",
684 __le32_to_cpu(map
->num_data_stripes
));
685 printf(" Chunk Size : %u KiB\n",
686 __le16_to_cpu(map
->blocks_per_strip
) / 2);
687 printf(" Reserved : %d\n", __le32_to_cpu(dev
->reserved_blocks
));
688 printf(" Migrate State : ");
689 if (dev
->vol
.migr_state
) {
690 if (migr_type(dev
) == MIGR_INIT
)
691 printf("initialize\n");
692 else if (migr_type(dev
) == MIGR_REBUILD
)
694 else if (migr_type(dev
) == MIGR_VERIFY
)
696 else if (migr_type(dev
) == MIGR_GEN_MIGR
)
697 printf("general migration\n");
698 else if (migr_type(dev
) == MIGR_STATE_CHANGE
)
699 printf("state change\n");
700 else if (migr_type(dev
) == MIGR_REPAIR
)
703 printf("<unknown:%d>\n", migr_type(dev
));
706 printf(" Map State : %s", map_state_str
[map
->map_state
]);
707 if (dev
->vol
.migr_state
) {
708 struct imsm_map
*map
= get_imsm_map(dev
, 1);
710 printf(" <-- %s", map_state_str
[map
->map_state
]);
711 printf("\n Checkpoint : %u (%llu)",
712 __le32_to_cpu(dev
->vol
.curr_migr_unit
),
713 (unsigned long long)blocks_per_migr_unit(dev
));
716 printf(" Dirty State : %s\n", dev
->vol
.dirty
? "dirty" : "clean");
719 static void print_imsm_disk(struct imsm_super
*mpb
, int index
, __u32 reserved
)
721 struct imsm_disk
*disk
= __get_imsm_disk(mpb
, index
);
722 char str
[MAX_RAID_SERIAL_LEN
+ 1];
725 if (index
< 0 || !disk
)
729 snprintf(str
, MAX_RAID_SERIAL_LEN
+ 1, "%s", disk
->serial
);
730 printf(" Disk%02d Serial : %s\n", index
, str
);
731 printf(" State :%s%s%s\n", is_spare(disk
) ? " spare" : "",
732 is_configured(disk
) ? " active" : "",
733 is_failed(disk
) ? " failed" : "");
734 printf(" Id : %08x\n", __le32_to_cpu(disk
->scsi_id
));
735 sz
= __le32_to_cpu(disk
->total_blocks
) - reserved
;
736 printf(" Usable Size : %llu%s\n", (unsigned long long)sz
,
737 human_size(sz
* 512));
740 static void getinfo_super_imsm(struct supertype
*st
, struct mdinfo
*info
);
742 static void examine_super_imsm(struct supertype
*st
, char *homehost
)
744 struct intel_super
*super
= st
->sb
;
745 struct imsm_super
*mpb
= super
->anchor
;
746 char str
[MAX_SIGNATURE_LENGTH
];
751 __u32 reserved
= imsm_reserved_sectors(super
, super
->disks
);
754 snprintf(str
, MPB_SIG_LEN
, "%s", mpb
->sig
);
755 printf(" Magic : %s\n", str
);
756 snprintf(str
, strlen(MPB_VERSION_RAID0
), "%s", get_imsm_version(mpb
));
757 printf(" Version : %s\n", get_imsm_version(mpb
));
758 printf(" Orig Family : %08x\n", __le32_to_cpu(mpb
->orig_family_num
));
759 printf(" Family : %08x\n", __le32_to_cpu(mpb
->family_num
));
760 printf(" Generation : %08x\n", __le32_to_cpu(mpb
->generation_num
));
761 getinfo_super_imsm(st
, &info
);
762 fname_from_uuid(st
, &info
, nbuf
, ':');
763 printf(" UUID : %s\n", nbuf
+ 5);
764 sum
= __le32_to_cpu(mpb
->check_sum
);
765 printf(" Checksum : %08x %s\n", sum
,
766 __gen_imsm_checksum(mpb
) == sum
? "correct" : "incorrect");
767 printf(" MPB Sectors : %d\n", mpb_sectors(mpb
));
768 printf(" Disks : %d\n", mpb
->num_disks
);
769 printf(" RAID Devices : %d\n", mpb
->num_raid_devs
);
770 print_imsm_disk(mpb
, super
->disks
->index
, reserved
);
771 if (super
->bbm_log
) {
772 struct bbm_log
*log
= super
->bbm_log
;
775 printf("Bad Block Management Log:\n");
776 printf(" Log Size : %d\n", __le32_to_cpu(mpb
->bbm_log_size
));
777 printf(" Signature : %x\n", __le32_to_cpu(log
->signature
));
778 printf(" Entry Count : %d\n", __le32_to_cpu(log
->entry_count
));
779 printf(" Spare Blocks : %d\n", __le32_to_cpu(log
->reserved_spare_block_count
));
780 printf(" First Spare : %llx\n",
781 (unsigned long long) __le64_to_cpu(log
->first_spare_lba
));
783 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
785 struct imsm_dev
*dev
= __get_imsm_dev(mpb
, i
);
787 super
->current_vol
= i
;
788 getinfo_super_imsm(st
, &info
);
789 fname_from_uuid(st
, &info
, nbuf
, ':');
790 print_imsm_dev(dev
, nbuf
+ 5, super
->disks
->index
);
792 for (i
= 0; i
< mpb
->num_disks
; i
++) {
793 if (i
== super
->disks
->index
)
795 print_imsm_disk(mpb
, i
, reserved
);
799 static void brief_examine_super_imsm(struct supertype
*st
, int verbose
)
801 /* We just write a generic IMSM ARRAY entry */
804 struct intel_super
*super
= st
->sb
;
806 if (!super
->anchor
->num_raid_devs
) {
807 printf("ARRAY metadata=imsm\n");
811 getinfo_super_imsm(st
, &info
);
812 fname_from_uuid(st
, &info
, nbuf
, ':');
813 printf("ARRAY metadata=imsm UUID=%s\n", nbuf
+ 5);
816 static void brief_examine_subarrays_imsm(struct supertype
*st
, int verbose
)
818 /* We just write a generic IMSM ARRAY entry */
822 struct intel_super
*super
= st
->sb
;
825 if (!super
->anchor
->num_raid_devs
)
828 getinfo_super_imsm(st
, &info
);
829 fname_from_uuid(st
, &info
, nbuf
, ':');
830 for (i
= 0; i
< super
->anchor
->num_raid_devs
; i
++) {
831 struct imsm_dev
*dev
= get_imsm_dev(super
, i
);
833 super
->current_vol
= i
;
834 getinfo_super_imsm(st
, &info
);
835 fname_from_uuid(st
, &info
, nbuf1
, ':');
836 printf("ARRAY /dev/md/%.16s container=%s member=%d UUID=%s\n",
837 dev
->volume
, nbuf
+ 5, i
, nbuf1
+ 5);
841 static void export_examine_super_imsm(struct supertype
*st
)
843 struct intel_super
*super
= st
->sb
;
844 struct imsm_super
*mpb
= super
->anchor
;
848 getinfo_super_imsm(st
, &info
);
849 fname_from_uuid(st
, &info
, nbuf
, ':');
850 printf("MD_METADATA=imsm\n");
851 printf("MD_LEVEL=container\n");
852 printf("MD_UUID=%s\n", nbuf
+5);
853 printf("MD_DEVICES=%u\n", mpb
->num_disks
);
856 static void detail_super_imsm(struct supertype
*st
, char *homehost
)
861 getinfo_super_imsm(st
, &info
);
862 fname_from_uuid(st
, &info
, nbuf
, ':');
863 printf("\n UUID : %s\n", nbuf
+ 5);
866 static void brief_detail_super_imsm(struct supertype
*st
)
870 getinfo_super_imsm(st
, &info
);
871 fname_from_uuid(st
, &info
, nbuf
, ':');
872 printf(" UUID=%s", nbuf
+ 5);
875 static int imsm_read_serial(int fd
, char *devname
, __u8
*serial
);
876 static void fd2devname(int fd
, char *name
);
878 static int imsm_enumerate_ports(const char *hba_path
, int port_count
, int host_base
, int verbose
)
880 /* dump an unsorted list of devices attached to ahci, as well as
881 * non-connected ports
883 int hba_len
= strlen(hba_path
) + 1;
888 unsigned long port_mask
= (1 << port_count
) - 1;
890 if (port_count
> sizeof(port_mask
) * 8) {
892 fprintf(stderr
, Name
": port_count %d out of range\n", port_count
);
896 /* scroll through /sys/dev/block looking for devices attached to
899 dir
= opendir("/sys/dev/block");
900 for (ent
= dir
? readdir(dir
) : NULL
; ent
; ent
= readdir(dir
)) {
911 if (sscanf(ent
->d_name
, "%d:%d", &major
, &minor
) != 2)
913 path
= devt_to_devpath(makedev(major
, minor
));
916 if (!path_attached_to_hba(path
, hba_path
)) {
922 /* retrieve the scsi device type */
923 if (asprintf(&device
, "/sys/dev/block/%d:%d/device/xxxxxxx", major
, minor
) < 0) {
925 fprintf(stderr
, Name
": failed to allocate 'device'\n");
929 sprintf(device
, "/sys/dev/block/%d:%d/device/type", major
, minor
);
930 if (load_sys(device
, buf
) != 0) {
932 fprintf(stderr
, Name
": failed to read device type for %s\n",
938 type
= strtoul(buf
, NULL
, 10);
940 /* if it's not a disk print the vendor and model */
941 if (!(type
== 0 || type
== 7 || type
== 14)) {
944 sprintf(device
, "/sys/dev/block/%d:%d/device/vendor", major
, minor
);
945 if (load_sys(device
, buf
) == 0) {
946 strncpy(vendor
, buf
, sizeof(vendor
));
947 vendor
[sizeof(vendor
) - 1] = '\0';
948 c
= (char *) &vendor
[sizeof(vendor
) - 1];
949 while (isspace(*c
) || *c
== '\0')
953 sprintf(device
, "/sys/dev/block/%d:%d/device/model", major
, minor
);
954 if (load_sys(device
, buf
) == 0) {
955 strncpy(model
, buf
, sizeof(model
));
956 model
[sizeof(model
) - 1] = '\0';
957 c
= (char *) &model
[sizeof(model
) - 1];
958 while (isspace(*c
) || *c
== '\0')
962 if (vendor
[0] && model
[0])
963 sprintf(buf
, "%.64s %.64s", vendor
, model
);
965 switch (type
) { /* numbers from hald/linux/device.c */
966 case 1: sprintf(buf
, "tape"); break;
967 case 2: sprintf(buf
, "printer"); break;
968 case 3: sprintf(buf
, "processor"); break;
970 case 5: sprintf(buf
, "cdrom"); break;
971 case 6: sprintf(buf
, "scanner"); break;
972 case 8: sprintf(buf
, "media_changer"); break;
973 case 9: sprintf(buf
, "comm"); break;
974 case 12: sprintf(buf
, "raid"); break;
975 default: sprintf(buf
, "unknown");
981 /* chop device path to 'host%d' and calculate the port number */
982 c
= strchr(&path
[hba_len
], '/');
985 fprintf(stderr
, Name
": %s - invalid path name\n", path
+ hba_len
);
990 if (sscanf(&path
[hba_len
], "host%d", &port
) == 1)
994 *c
= '/'; /* repair the full string */
995 fprintf(stderr
, Name
": failed to determine port number for %s\n",
1002 /* mark this port as used */
1003 port_mask
&= ~(1 << port
);
1005 /* print out the device information */
1007 printf(" Port%d : - non-disk device (%s) -\n", port
, buf
);
1011 fd
= dev_open(ent
->d_name
, O_RDONLY
);
1013 printf(" Port%d : - disk info unavailable -\n", port
);
1015 fd2devname(fd
, buf
);
1016 printf(" Port%d : %s", port
, buf
);
1017 if (imsm_read_serial(fd
, NULL
, (__u8
*) buf
) == 0)
1018 printf(" (%s)\n", buf
);
1033 for (i
= 0; i
< port_count
; i
++)
1034 if (port_mask
& (1 << i
))
1035 printf(" Port%d : - no device attached -\n", i
);
1041 static int detail_platform_imsm(int verbose
, int enumerate_only
)
1043 /* There are two components to imsm platform support, the ahci SATA
1044 * controller and the option-rom. To find the SATA controller we
1045 * simply look in /sys/bus/pci/drivers/ahci to see if an ahci
1046 * controller with the Intel vendor id is present. This approach
1047 * allows mdadm to leverage the kernel's ahci detection logic, with the
1048 * caveat that if ahci.ko is not loaded mdadm will not be able to
1049 * detect platform raid capabilities. The option-rom resides in a
1050 * platform "Adapter ROM". We scan for its signature to retrieve the
1051 * platform capabilities. If raid support is disabled in the BIOS the
1052 * option-rom capability structure will not be available.
1054 const struct imsm_orom
*orom
;
1055 struct sys_dev
*list
, *hba
;
1058 const char *hba_path
;
1062 if (enumerate_only
) {
1063 if (check_env("IMSM_NO_PLATFORM") || find_imsm_orom())
1068 list
= find_driver_devices("pci", "ahci");
1069 for (hba
= list
; hba
; hba
= hba
->next
)
1070 if (devpath_to_vendor(hba
->path
) == 0x8086)
1075 fprintf(stderr
, Name
": unable to find active ahci controller\n");
1076 free_sys_dev(&list
);
1079 fprintf(stderr
, Name
": found Intel SATA AHCI Controller\n");
1080 hba_path
= hba
->path
;
1082 free_sys_dev(&list
);
1084 orom
= find_imsm_orom();
1087 fprintf(stderr
, Name
": imsm option-rom not found\n");
1091 printf(" Platform : Intel(R) Matrix Storage Manager\n");
1092 printf(" Version : %d.%d.%d.%d\n", orom
->major_ver
, orom
->minor_ver
,
1093 orom
->hotfix_ver
, orom
->build
);
1094 printf(" RAID Levels :%s%s%s%s%s\n",
1095 imsm_orom_has_raid0(orom
) ? " raid0" : "",
1096 imsm_orom_has_raid1(orom
) ? " raid1" : "",
1097 imsm_orom_has_raid1e(orom
) ? " raid1e" : "",
1098 imsm_orom_has_raid10(orom
) ? " raid10" : "",
1099 imsm_orom_has_raid5(orom
) ? " raid5" : "");
1100 printf(" Chunk Sizes :%s%s%s%s%s%s%s%s%s%s%s%s%s%s%s%s\n",
1101 imsm_orom_has_chunk(orom
, 2) ? " 2k" : "",
1102 imsm_orom_has_chunk(orom
, 4) ? " 4k" : "",
1103 imsm_orom_has_chunk(orom
, 8) ? " 8k" : "",
1104 imsm_orom_has_chunk(orom
, 16) ? " 16k" : "",
1105 imsm_orom_has_chunk(orom
, 32) ? " 32k" : "",
1106 imsm_orom_has_chunk(orom
, 64) ? " 64k" : "",
1107 imsm_orom_has_chunk(orom
, 128) ? " 128k" : "",
1108 imsm_orom_has_chunk(orom
, 256) ? " 256k" : "",
1109 imsm_orom_has_chunk(orom
, 512) ? " 512k" : "",
1110 imsm_orom_has_chunk(orom
, 1024*1) ? " 1M" : "",
1111 imsm_orom_has_chunk(orom
, 1024*2) ? " 2M" : "",
1112 imsm_orom_has_chunk(orom
, 1024*4) ? " 4M" : "",
1113 imsm_orom_has_chunk(orom
, 1024*8) ? " 8M" : "",
1114 imsm_orom_has_chunk(orom
, 1024*16) ? " 16M" : "",
1115 imsm_orom_has_chunk(orom
, 1024*32) ? " 32M" : "",
1116 imsm_orom_has_chunk(orom
, 1024*64) ? " 64M" : "");
1117 printf(" Max Disks : %d\n", orom
->tds
);
1118 printf(" Max Volumes : %d\n", orom
->vpa
);
1119 printf(" I/O Controller : %s\n", hba_path
);
1121 /* find the smallest scsi host number to determine a port number base */
1122 dir
= opendir(hba_path
);
1123 for (ent
= dir
? readdir(dir
) : NULL
; ent
; ent
= readdir(dir
)) {
1126 if (sscanf(ent
->d_name
, "host%d", &host
) != 1)
1128 if (port_count
== 0)
1130 else if (host
< host_base
)
1133 if (host
+ 1 > port_count
+ host_base
)
1134 port_count
= host
+ 1 - host_base
;
1140 if (!port_count
|| imsm_enumerate_ports(hba_path
, port_count
,
1141 host_base
, verbose
) != 0) {
1143 fprintf(stderr
, Name
": failed to enumerate ports\n");
1151 static int match_home_imsm(struct supertype
*st
, char *homehost
)
1153 /* the imsm metadata format does not specify any host
1154 * identification information. We return -1 since we can never
1155 * confirm nor deny whether a given array is "meant" for this
1156 * host. We rely on compare_super and the 'family_num' fields to
1157 * exclude member disks that do not belong, and we rely on
1158 * mdadm.conf to specify the arrays that should be assembled.
1159 * Auto-assembly may still pick up "foreign" arrays.
1165 static void uuid_from_super_imsm(struct supertype
*st
, int uuid
[4])
1167 /* The uuid returned here is used for:
1168 * uuid to put into bitmap file (Create, Grow)
1169 * uuid for backup header when saving critical section (Grow)
1170 * comparing uuids when re-adding a device into an array
1171 * In these cases the uuid required is that of the data-array,
1172 * not the device-set.
1173 * uuid to recognise same set when adding a missing device back
1174 * to an array. This is a uuid for the device-set.
1176 * For each of these we can make do with a truncated
1177 * or hashed uuid rather than the original, as long as
1179 * In each case the uuid required is that of the data-array,
1180 * not the device-set.
1182 /* imsm does not track uuid's so we synthesis one using sha1 on
1183 * - The signature (Which is constant for all imsm array, but no matter)
1184 * - the orig_family_num of the container
1185 * - the index number of the volume
1186 * - the 'serial' number of the volume.
1187 * Hopefully these are all constant.
1189 struct intel_super
*super
= st
->sb
;
1192 struct sha1_ctx ctx
;
1193 struct imsm_dev
*dev
= NULL
;
1196 /* some mdadm versions failed to set ->orig_family_num, in which
1197 * case fall back to ->family_num. orig_family_num will be
1198 * fixed up with the first metadata update.
1200 family_num
= super
->anchor
->orig_family_num
;
1201 if (family_num
== 0)
1202 family_num
= super
->anchor
->family_num
;
1203 sha1_init_ctx(&ctx
);
1204 sha1_process_bytes(super
->anchor
->sig
, MPB_SIG_LEN
, &ctx
);
1205 sha1_process_bytes(&family_num
, sizeof(__u32
), &ctx
);
1206 if (super
->current_vol
>= 0)
1207 dev
= get_imsm_dev(super
, super
->current_vol
);
1209 __u32 vol
= super
->current_vol
;
1210 sha1_process_bytes(&vol
, sizeof(vol
), &ctx
);
1211 sha1_process_bytes(dev
->volume
, MAX_RAID_SERIAL_LEN
, &ctx
);
1213 sha1_finish_ctx(&ctx
, buf
);
1214 memcpy(uuid
, buf
, 4*4);
1219 get_imsm_numerical_version(struct imsm_super
*mpb
, int *m
, int *p
)
1221 __u8
*v
= get_imsm_version(mpb
);
1222 __u8
*end
= mpb
->sig
+ MAX_SIGNATURE_LENGTH
;
1223 char major
[] = { 0, 0, 0 };
1224 char minor
[] = { 0 ,0, 0 };
1225 char patch
[] = { 0, 0, 0 };
1226 char *ver_parse
[] = { major
, minor
, patch
};
1230 while (*v
!= '\0' && v
< end
) {
1231 if (*v
!= '.' && j
< 2)
1232 ver_parse
[i
][j
++] = *v
;
1240 *m
= strtol(minor
, NULL
, 0);
1241 *p
= strtol(patch
, NULL
, 0);
1245 static __u32
migr_strip_blocks_resync(struct imsm_dev
*dev
)
1247 /* migr_strip_size when repairing or initializing parity */
1248 struct imsm_map
*map
= get_imsm_map(dev
, 0);
1249 __u32 chunk
= __le32_to_cpu(map
->blocks_per_strip
);
1251 switch (get_imsm_raid_level(map
)) {
1256 return 128*1024 >> 9;
1260 static __u32
migr_strip_blocks_rebuild(struct imsm_dev
*dev
)
1262 /* migr_strip_size when rebuilding a degraded disk, no idea why
1263 * this is different than migr_strip_size_resync(), but it's good
1266 struct imsm_map
*map
= get_imsm_map(dev
, 1);
1267 __u32 chunk
= __le32_to_cpu(map
->blocks_per_strip
);
1269 switch (get_imsm_raid_level(map
)) {
1272 if (map
->num_members
% map
->num_domains
== 0)
1273 return 128*1024 >> 9;
1277 return max((__u32
) 64*1024 >> 9, chunk
);
1279 return 128*1024 >> 9;
1283 static __u32
num_stripes_per_unit_resync(struct imsm_dev
*dev
)
1285 struct imsm_map
*lo
= get_imsm_map(dev
, 0);
1286 struct imsm_map
*hi
= get_imsm_map(dev
, 1);
1287 __u32 lo_chunk
= __le32_to_cpu(lo
->blocks_per_strip
);
1288 __u32 hi_chunk
= __le32_to_cpu(hi
->blocks_per_strip
);
1290 return max((__u32
) 1, hi_chunk
/ lo_chunk
);
1293 static __u32
num_stripes_per_unit_rebuild(struct imsm_dev
*dev
)
1295 struct imsm_map
*lo
= get_imsm_map(dev
, 0);
1296 int level
= get_imsm_raid_level(lo
);
1298 if (level
== 1 || level
== 10) {
1299 struct imsm_map
*hi
= get_imsm_map(dev
, 1);
1301 return hi
->num_domains
;
1303 return num_stripes_per_unit_resync(dev
);
1306 static __u8
imsm_num_data_members(struct imsm_dev
*dev
)
1308 /* named 'imsm_' because raid0, raid1 and raid10
1309 * counter-intuitively have the same number of data disks
1311 struct imsm_map
*map
= get_imsm_map(dev
, 0);
1313 switch (get_imsm_raid_level(map
)) {
1317 return map
->num_members
;
1319 return map
->num_members
- 1;
1321 dprintf("%s: unsupported raid level\n", __func__
);
1326 static __u32
parity_segment_depth(struct imsm_dev
*dev
)
1328 struct imsm_map
*map
= get_imsm_map(dev
, 0);
1329 __u32 chunk
= __le32_to_cpu(map
->blocks_per_strip
);
1331 switch(get_imsm_raid_level(map
)) {
1334 return chunk
* map
->num_domains
;
1336 return chunk
* map
->num_members
;
1342 static __u32
map_migr_block(struct imsm_dev
*dev
, __u32 block
)
1344 struct imsm_map
*map
= get_imsm_map(dev
, 1);
1345 __u32 chunk
= __le32_to_cpu(map
->blocks_per_strip
);
1346 __u32 strip
= block
/ chunk
;
1348 switch (get_imsm_raid_level(map
)) {
1351 __u32 vol_strip
= (strip
* map
->num_domains
) + 1;
1352 __u32 vol_stripe
= vol_strip
/ map
->num_members
;
1354 return vol_stripe
* chunk
+ block
% chunk
;
1356 __u32 stripe
= strip
/ (map
->num_members
- 1);
1358 return stripe
* chunk
+ block
% chunk
;
1365 static __u64
blocks_per_migr_unit(struct imsm_dev
*dev
)
1367 /* calculate the conversion factor between per member 'blocks'
1368 * (md/{resync,rebuild}_start) and imsm migration units, return
1369 * 0 for the 'not migrating' and 'unsupported migration' cases
1371 if (!dev
->vol
.migr_state
)
1374 switch (migr_type(dev
)) {
1378 struct imsm_map
*map
= get_imsm_map(dev
, 0);
1379 __u32 stripes_per_unit
;
1380 __u32 blocks_per_unit
;
1389 /* yes, this is really the translation of migr_units to
1390 * per-member blocks in the 'resync' case
1392 stripes_per_unit
= num_stripes_per_unit_resync(dev
);
1393 migr_chunk
= migr_strip_blocks_resync(dev
);
1394 disks
= imsm_num_data_members(dev
);
1395 blocks_per_unit
= stripes_per_unit
* migr_chunk
* disks
;
1396 stripe
= __le32_to_cpu(map
->blocks_per_strip
) * disks
;
1397 segment
= blocks_per_unit
/ stripe
;
1398 block_rel
= blocks_per_unit
- segment
* stripe
;
1399 parity_depth
= parity_segment_depth(dev
);
1400 block_map
= map_migr_block(dev
, block_rel
);
1401 return block_map
+ parity_depth
* segment
;
1403 case MIGR_REBUILD
: {
1404 __u32 stripes_per_unit
;
1407 stripes_per_unit
= num_stripes_per_unit_rebuild(dev
);
1408 migr_chunk
= migr_strip_blocks_rebuild(dev
);
1409 return migr_chunk
* stripes_per_unit
;
1412 case MIGR_STATE_CHANGE
:
1418 static int imsm_level_to_layout(int level
)
1426 return ALGORITHM_LEFT_ASYMMETRIC
;
1433 static void getinfo_super_imsm_volume(struct supertype
*st
, struct mdinfo
*info
)
1435 struct intel_super
*super
= st
->sb
;
1436 struct imsm_dev
*dev
= get_imsm_dev(super
, super
->current_vol
);
1437 struct imsm_map
*map
= get_imsm_map(dev
, 0);
1441 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
1442 if (dl
->raiddisk
== info
->disk
.raid_disk
)
1444 info
->container_member
= super
->current_vol
;
1445 info
->array
.raid_disks
= map
->num_members
;
1446 info
->array
.level
= get_imsm_raid_level(map
);
1447 info
->array
.layout
= imsm_level_to_layout(info
->array
.level
);
1448 info
->array
.md_minor
= -1;
1449 info
->array
.ctime
= 0;
1450 info
->array
.utime
= 0;
1451 info
->array
.chunk_size
= __le16_to_cpu(map
->blocks_per_strip
) << 9;
1452 info
->array
.state
= !dev
->vol
.dirty
;
1453 info
->custom_array_size
= __le32_to_cpu(dev
->size_high
);
1454 info
->custom_array_size
<<= 32;
1455 info
->custom_array_size
|= __le32_to_cpu(dev
->size_low
);
1457 info
->disk
.major
= 0;
1458 info
->disk
.minor
= 0;
1460 info
->disk
.major
= dl
->major
;
1461 info
->disk
.minor
= dl
->minor
;
1464 info
->data_offset
= __le32_to_cpu(map
->pba_of_lba0
);
1465 info
->component_size
= __le32_to_cpu(map
->blocks_per_member
);
1466 memset(info
->uuid
, 0, sizeof(info
->uuid
));
1467 info
->recovery_start
= MaxSector
;
1468 info
->reshape_active
= 0;
1470 if (map
->map_state
== IMSM_T_STATE_UNINITIALIZED
|| dev
->vol
.dirty
) {
1471 info
->resync_start
= 0;
1472 } else if (dev
->vol
.migr_state
) {
1473 switch (migr_type(dev
)) {
1476 __u64 blocks_per_unit
= blocks_per_migr_unit(dev
);
1477 __u64 units
= __le32_to_cpu(dev
->vol
.curr_migr_unit
);
1479 info
->resync_start
= blocks_per_unit
* units
;
1483 /* we could emulate the checkpointing of
1484 * 'sync_action=check' migrations, but for now
1485 * we just immediately complete them
1488 /* this is handled by container_content_imsm() */
1490 case MIGR_STATE_CHANGE
:
1491 /* FIXME handle other migrations */
1493 /* we are not dirty, so... */
1494 info
->resync_start
= MaxSector
;
1497 info
->resync_start
= MaxSector
;
1499 strncpy(info
->name
, (char *) dev
->volume
, MAX_RAID_SERIAL_LEN
);
1500 info
->name
[MAX_RAID_SERIAL_LEN
] = 0;
1502 info
->array
.major_version
= -1;
1503 info
->array
.minor_version
= -2;
1504 devname
= devnum2devname(st
->container_dev
);
1505 *info
->text_version
= '\0';
1507 sprintf(info
->text_version
, "/%s/%d", devname
, info
->container_member
);
1509 info
->safe_mode_delay
= 4000; /* 4 secs like the Matrix driver */
1510 uuid_from_super_imsm(st
, info
->uuid
);
1513 /* check the config file to see if we can return a real uuid for this spare */
1514 static void fixup_container_spare_uuid(struct mdinfo
*inf
)
1516 struct mddev_ident_s
*array_list
;
1518 if (inf
->array
.level
!= LEVEL_CONTAINER
||
1519 memcmp(inf
->uuid
, uuid_match_any
, sizeof(int[4])) != 0)
1522 array_list
= conf_get_ident(NULL
);
1524 for (; array_list
; array_list
= array_list
->next
) {
1525 if (array_list
->uuid_set
) {
1526 struct supertype
*_sst
; /* spare supertype */
1527 struct supertype
*_cst
; /* container supertype */
1529 _cst
= array_list
->st
;
1531 _sst
= _cst
->ss
->match_metadata_desc(inf
->text_version
);
1536 memcpy(inf
->uuid
, array_list
->uuid
, sizeof(int[4]));
1544 static void getinfo_super_imsm(struct supertype
*st
, struct mdinfo
*info
)
1546 struct intel_super
*super
= st
->sb
;
1547 struct imsm_disk
*disk
;
1549 if (super
->current_vol
>= 0) {
1550 getinfo_super_imsm_volume(st
, info
);
1554 /* Set raid_disks to zero so that Assemble will always pull in valid
1557 info
->array
.raid_disks
= 0;
1558 info
->array
.level
= LEVEL_CONTAINER
;
1559 info
->array
.layout
= 0;
1560 info
->array
.md_minor
= -1;
1561 info
->array
.ctime
= 0; /* N/A for imsm */
1562 info
->array
.utime
= 0;
1563 info
->array
.chunk_size
= 0;
1565 info
->disk
.major
= 0;
1566 info
->disk
.minor
= 0;
1567 info
->disk
.raid_disk
= -1;
1568 info
->reshape_active
= 0;
1569 info
->array
.major_version
= -1;
1570 info
->array
.minor_version
= -2;
1571 strcpy(info
->text_version
, "imsm");
1572 info
->safe_mode_delay
= 0;
1573 info
->disk
.number
= -1;
1574 info
->disk
.state
= 0;
1576 info
->recovery_start
= MaxSector
;
1579 __u32 reserved
= imsm_reserved_sectors(super
, super
->disks
);
1581 disk
= &super
->disks
->disk
;
1582 info
->data_offset
= __le32_to_cpu(disk
->total_blocks
) - reserved
;
1583 info
->component_size
= reserved
;
1584 info
->disk
.state
= is_configured(disk
) ? (1 << MD_DISK_ACTIVE
) : 0;
1585 /* we don't change info->disk.raid_disk here because
1586 * this state will be finalized in mdmon after we have
1587 * found the 'most fresh' version of the metadata
1589 info
->disk
.state
|= is_failed(disk
) ? (1 << MD_DISK_FAULTY
) : 0;
1590 info
->disk
.state
|= is_spare(disk
) ? 0 : (1 << MD_DISK_SYNC
);
1593 /* only call uuid_from_super_imsm when this disk is part of a populated container,
1594 * ->compare_super may have updated the 'num_raid_devs' field for spares
1596 if (info
->disk
.state
& (1 << MD_DISK_SYNC
) || super
->anchor
->num_raid_devs
)
1597 uuid_from_super_imsm(st
, info
->uuid
);
1599 memcpy(info
->uuid
, uuid_match_any
, sizeof(int[4]));
1600 fixup_container_spare_uuid(info
);
1604 static int update_super_imsm(struct supertype
*st
, struct mdinfo
*info
,
1605 char *update
, char *devname
, int verbose
,
1606 int uuid_set
, char *homehost
)
1608 /* For 'assemble' and 'force' we need to return non-zero if any
1609 * change was made. For others, the return value is ignored.
1610 * Update options are:
1611 * force-one : This device looks a bit old but needs to be included,
1612 * update age info appropriately.
1613 * assemble: clear any 'faulty' flag to allow this device to
1615 * force-array: Array is degraded but being forced, mark it clean
1616 * if that will be needed to assemble it.
1618 * newdev: not used ????
1619 * grow: Array has gained a new device - this is currently for
1621 * resync: mark as dirty so a resync will happen.
1622 * name: update the name - preserving the homehost
1623 * uuid: Change the uuid of the array to match watch is given
1625 * Following are not relevant for this imsm:
1626 * sparc2.2 : update from old dodgey metadata
1627 * super-minor: change the preferred_minor number
1628 * summaries: update redundant counters.
1629 * homehost: update the recorded homehost
1630 * _reshape_progress: record new reshape_progress position.
1633 struct intel_super
*super
= st
->sb
;
1634 struct imsm_super
*mpb
;
1636 /* we can only update container info */
1637 if (!super
|| super
->current_vol
>= 0 || !super
->anchor
)
1640 mpb
= super
->anchor
;
1642 if (strcmp(update
, "uuid") == 0 && uuid_set
&& !info
->update_private
)
1644 Name
": '--uuid' not supported for imsm metadata\n");
1645 else if (strcmp(update
, "uuid") == 0 && uuid_set
&& info
->update_private
) {
1646 mpb
->orig_family_num
= *((__u32
*) info
->update_private
);
1648 } else if (strcmp(update
, "uuid") == 0) {
1649 __u32
*new_family
= malloc(sizeof(*new_family
));
1651 /* update orig_family_number with the incoming random
1652 * data, report the new effective uuid, and store the
1653 * new orig_family_num for future updates.
1656 memcpy(&mpb
->orig_family_num
, info
->uuid
, sizeof(__u32
));
1657 uuid_from_super_imsm(st
, info
->uuid
);
1658 *new_family
= mpb
->orig_family_num
;
1659 info
->update_private
= new_family
;
1662 } else if (strcmp(update
, "assemble") == 0)
1666 Name
": '--update=%s' not supported for imsm metadata\n",
1669 /* successful update? recompute checksum */
1671 mpb
->check_sum
= __le32_to_cpu(__gen_imsm_checksum(mpb
));
1676 static size_t disks_to_mpb_size(int disks
)
1680 size
= sizeof(struct imsm_super
);
1681 size
+= (disks
- 1) * sizeof(struct imsm_disk
);
1682 size
+= 2 * sizeof(struct imsm_dev
);
1683 /* up to 2 maps per raid device (-2 for imsm_maps in imsm_dev */
1684 size
+= (4 - 2) * sizeof(struct imsm_map
);
1685 /* 4 possible disk_ord_tbl's */
1686 size
+= 4 * (disks
- 1) * sizeof(__u32
);
1691 static __u64
avail_size_imsm(struct supertype
*st
, __u64 devsize
)
1693 if (devsize
< (MPB_SECTOR_CNT
+ IMSM_RESERVED_SECTORS
))
1696 return devsize
- (MPB_SECTOR_CNT
+ IMSM_RESERVED_SECTORS
);
1699 static void free_devlist(struct intel_super
*super
)
1701 struct intel_dev
*dv
;
1703 while (super
->devlist
) {
1704 dv
= super
->devlist
->next
;
1705 free(super
->devlist
->dev
);
1706 free(super
->devlist
);
1707 super
->devlist
= dv
;
1711 static void imsm_copy_dev(struct imsm_dev
*dest
, struct imsm_dev
*src
)
1713 memcpy(dest
, src
, sizeof_imsm_dev(src
, 0));
1716 static int compare_super_imsm(struct supertype
*st
, struct supertype
*tst
)
1720 * 0 same, or first was empty, and second was copied
1721 * 1 second had wrong number
1723 * 3 wrong other info
1725 struct intel_super
*first
= st
->sb
;
1726 struct intel_super
*sec
= tst
->sb
;
1734 /* if an anchor does not have num_raid_devs set then it is a free
1737 if (first
->anchor
->num_raid_devs
> 0 &&
1738 sec
->anchor
->num_raid_devs
> 0) {
1739 /* Determine if these disks might ever have been
1740 * related. Further disambiguation can only take place
1741 * in load_super_imsm_all
1743 __u32 first_family
= first
->anchor
->orig_family_num
;
1744 __u32 sec_family
= sec
->anchor
->orig_family_num
;
1746 if (memcmp(first
->anchor
->sig
, sec
->anchor
->sig
,
1747 MAX_SIGNATURE_LENGTH
) != 0)
1750 if (first_family
== 0)
1751 first_family
= first
->anchor
->family_num
;
1752 if (sec_family
== 0)
1753 sec_family
= sec
->anchor
->family_num
;
1755 if (first_family
!= sec_family
)
1761 /* if 'first' is a spare promote it to a populated mpb with sec's
1764 if (first
->anchor
->num_raid_devs
== 0 &&
1765 sec
->anchor
->num_raid_devs
> 0) {
1767 struct intel_dev
*dv
;
1768 struct imsm_dev
*dev
;
1770 /* we need to copy raid device info from sec if an allocation
1771 * fails here we don't associate the spare
1773 for (i
= 0; i
< sec
->anchor
->num_raid_devs
; i
++) {
1774 dv
= malloc(sizeof(*dv
));
1777 dev
= malloc(sizeof_imsm_dev(get_imsm_dev(sec
, i
), 1));
1784 dv
->next
= first
->devlist
;
1785 first
->devlist
= dv
;
1787 if (i
< sec
->anchor
->num_raid_devs
) {
1788 /* allocation failure */
1789 free_devlist(first
);
1790 fprintf(stderr
, "imsm: failed to associate spare\n");
1793 first
->anchor
->num_raid_devs
= sec
->anchor
->num_raid_devs
;
1794 first
->anchor
->orig_family_num
= sec
->anchor
->orig_family_num
;
1795 first
->anchor
->family_num
= sec
->anchor
->family_num
;
1796 memcpy(first
->anchor
->sig
, sec
->anchor
->sig
, MAX_SIGNATURE_LENGTH
);
1797 for (i
= 0; i
< sec
->anchor
->num_raid_devs
; i
++)
1798 imsm_copy_dev(get_imsm_dev(first
, i
), get_imsm_dev(sec
, i
));
1804 static void fd2devname(int fd
, char *name
)
1808 char dname
[PATH_MAX
];
1813 if (fstat(fd
, &st
) != 0)
1815 sprintf(path
, "/sys/dev/block/%d:%d",
1816 major(st
.st_rdev
), minor(st
.st_rdev
));
1818 rv
= readlink(path
, dname
, sizeof(dname
));
1823 nm
= strrchr(dname
, '/');
1825 snprintf(name
, MAX_RAID_SERIAL_LEN
, "/dev/%s", nm
);
1828 extern int scsi_get_serial(int fd
, void *buf
, size_t buf_len
);
1830 static int imsm_read_serial(int fd
, char *devname
,
1831 __u8 serial
[MAX_RAID_SERIAL_LEN
])
1833 unsigned char scsi_serial
[255];
1842 memset(scsi_serial
, 0, sizeof(scsi_serial
));
1844 rv
= scsi_get_serial(fd
, scsi_serial
, sizeof(scsi_serial
));
1846 if (rv
&& check_env("IMSM_DEVNAME_AS_SERIAL")) {
1847 memset(serial
, 0, MAX_RAID_SERIAL_LEN
);
1848 fd2devname(fd
, (char *) serial
);
1855 Name
": Failed to retrieve serial for %s\n",
1860 rsp_len
= scsi_serial
[3];
1864 Name
": Failed to retrieve serial for %s\n",
1868 rsp_buf
= (char *) &scsi_serial
[4];
1870 /* trim all whitespace and non-printable characters and convert
1873 for (i
= 0, dest
= rsp_buf
; i
< rsp_len
; i
++) {
1876 /* ':' is reserved for use in placeholder serial
1877 * numbers for missing disks
1885 len
= dest
- rsp_buf
;
1888 /* truncate leading characters */
1889 if (len
> MAX_RAID_SERIAL_LEN
) {
1890 dest
+= len
- MAX_RAID_SERIAL_LEN
;
1891 len
= MAX_RAID_SERIAL_LEN
;
1894 memset(serial
, 0, MAX_RAID_SERIAL_LEN
);
1895 memcpy(serial
, dest
, len
);
1900 static int serialcmp(__u8
*s1
, __u8
*s2
)
1902 return strncmp((char *) s1
, (char *) s2
, MAX_RAID_SERIAL_LEN
);
1905 static void serialcpy(__u8
*dest
, __u8
*src
)
1907 strncpy((char *) dest
, (char *) src
, MAX_RAID_SERIAL_LEN
);
1911 static struct dl
*serial_to_dl(__u8
*serial
, struct intel_super
*super
)
1915 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
1916 if (serialcmp(dl
->serial
, serial
) == 0)
1923 static struct imsm_disk
*
1924 __serial_to_disk(__u8
*serial
, struct imsm_super
*mpb
, int *idx
)
1928 for (i
= 0; i
< mpb
->num_disks
; i
++) {
1929 struct imsm_disk
*disk
= __get_imsm_disk(mpb
, i
);
1931 if (serialcmp(disk
->serial
, serial
) == 0) {
1942 load_imsm_disk(int fd
, struct intel_super
*super
, char *devname
, int keep_fd
)
1944 struct imsm_disk
*disk
;
1949 __u8 serial
[MAX_RAID_SERIAL_LEN
];
1951 rv
= imsm_read_serial(fd
, devname
, serial
);
1956 dl
= calloc(1, sizeof(*dl
));
1960 Name
": failed to allocate disk buffer for %s\n",
1966 dl
->major
= major(stb
.st_rdev
);
1967 dl
->minor
= minor(stb
.st_rdev
);
1968 dl
->next
= super
->disks
;
1969 dl
->fd
= keep_fd
? fd
: -1;
1970 assert(super
->disks
== NULL
);
1972 serialcpy(dl
->serial
, serial
);
1975 fd2devname(fd
, name
);
1977 dl
->devname
= strdup(devname
);
1979 dl
->devname
= strdup(name
);
1981 /* look up this disk's index in the current anchor */
1982 disk
= __serial_to_disk(dl
->serial
, super
->anchor
, &dl
->index
);
1985 /* only set index on disks that are a member of a
1986 * populated contianer, i.e. one with raid_devs
1988 if (is_failed(&dl
->disk
))
1990 else if (is_spare(&dl
->disk
))
1998 /* When migrating map0 contains the 'destination' state while map1
1999 * contains the current state. When not migrating map0 contains the
2000 * current state. This routine assumes that map[0].map_state is set to
2001 * the current array state before being called.
2003 * Migration is indicated by one of the following states
2004 * 1/ Idle (migr_state=0 map0state=normal||unitialized||degraded||failed)
2005 * 2/ Initialize (migr_state=1 migr_type=MIGR_INIT map0state=normal
2006 * map1state=unitialized)
2007 * 3/ Repair (Resync) (migr_state=1 migr_type=MIGR_REPAIR map0state=normal
2009 * 4/ Rebuild (migr_state=1 migr_type=MIGR_REBUILD map0state=normal
2010 * map1state=degraded)
2012 static void migrate(struct imsm_dev
*dev
, __u8 to_state
, int migr_type
)
2014 struct imsm_map
*dest
;
2015 struct imsm_map
*src
= get_imsm_map(dev
, 0);
2017 dev
->vol
.migr_state
= 1;
2018 set_migr_type(dev
, migr_type
);
2019 dev
->vol
.curr_migr_unit
= 0;
2020 dest
= get_imsm_map(dev
, 1);
2022 /* duplicate and then set the target end state in map[0] */
2023 memcpy(dest
, src
, sizeof_imsm_map(src
));
2024 if (migr_type
== MIGR_REBUILD
) {
2028 for (i
= 0; i
< src
->num_members
; i
++) {
2029 ord
= __le32_to_cpu(src
->disk_ord_tbl
[i
]);
2030 set_imsm_ord_tbl_ent(src
, i
, ord_to_idx(ord
));
2034 src
->map_state
= to_state
;
2037 static void end_migration(struct imsm_dev
*dev
, __u8 map_state
)
2039 struct imsm_map
*map
= get_imsm_map(dev
, 0);
2040 struct imsm_map
*prev
= get_imsm_map(dev
, dev
->vol
.migr_state
);
2043 /* merge any IMSM_ORD_REBUILD bits that were not successfully
2044 * completed in the last migration.
2046 * FIXME add support for online capacity expansion and
2047 * raid-level-migration
2049 for (i
= 0; i
< prev
->num_members
; i
++)
2050 map
->disk_ord_tbl
[i
] |= prev
->disk_ord_tbl
[i
];
2052 dev
->vol
.migr_state
= 0;
2053 dev
->vol
.curr_migr_unit
= 0;
2054 map
->map_state
= map_state
;
2058 static int parse_raid_devices(struct intel_super
*super
)
2061 struct imsm_dev
*dev_new
;
2062 size_t len
, len_migr
;
2063 size_t space_needed
= 0;
2064 struct imsm_super
*mpb
= super
->anchor
;
2066 for (i
= 0; i
< super
->anchor
->num_raid_devs
; i
++) {
2067 struct imsm_dev
*dev_iter
= __get_imsm_dev(super
->anchor
, i
);
2068 struct intel_dev
*dv
;
2070 len
= sizeof_imsm_dev(dev_iter
, 0);
2071 len_migr
= sizeof_imsm_dev(dev_iter
, 1);
2073 space_needed
+= len_migr
- len
;
2075 dv
= malloc(sizeof(*dv
));
2078 dev_new
= malloc(len_migr
);
2083 imsm_copy_dev(dev_new
, dev_iter
);
2086 dv
->next
= super
->devlist
;
2087 super
->devlist
= dv
;
2090 /* ensure that super->buf is large enough when all raid devices
2093 if (__le32_to_cpu(mpb
->mpb_size
) + space_needed
> super
->len
) {
2096 len
= ROUND_UP(__le32_to_cpu(mpb
->mpb_size
) + space_needed
, 512);
2097 if (posix_memalign(&buf
, 512, len
) != 0)
2100 memcpy(buf
, super
->buf
, super
->len
);
2101 memset(buf
+ super
->len
, 0, len
- super
->len
);
2110 /* retrieve a pointer to the bbm log which starts after all raid devices */
2111 struct bbm_log
*__get_imsm_bbm_log(struct imsm_super
*mpb
)
2115 if (__le32_to_cpu(mpb
->bbm_log_size
)) {
2117 ptr
+= mpb
->mpb_size
- __le32_to_cpu(mpb
->bbm_log_size
);
2123 static void __free_imsm(struct intel_super
*super
, int free_disks
);
2125 /* load_imsm_mpb - read matrix metadata
2126 * allocates super->mpb to be freed by free_super
2128 static int load_imsm_mpb(int fd
, struct intel_super
*super
, char *devname
)
2130 unsigned long long dsize
;
2131 unsigned long long sectors
;
2133 struct imsm_super
*anchor
;
2136 get_dev_size(fd
, NULL
, &dsize
);
2138 if (lseek64(fd
, dsize
- (512 * 2), SEEK_SET
) < 0) {
2141 Name
": Cannot seek to anchor block on %s: %s\n",
2142 devname
, strerror(errno
));
2146 if (posix_memalign((void**)&anchor
, 512, 512) != 0) {
2149 Name
": Failed to allocate imsm anchor buffer"
2150 " on %s\n", devname
);
2153 if (read(fd
, anchor
, 512) != 512) {
2156 Name
": Cannot read anchor block on %s: %s\n",
2157 devname
, strerror(errno
));
2162 if (strncmp((char *) anchor
->sig
, MPB_SIGNATURE
, MPB_SIG_LEN
) != 0) {
2165 Name
": no IMSM anchor on %s\n", devname
);
2170 __free_imsm(super
, 0);
2171 super
->len
= ROUND_UP(anchor
->mpb_size
, 512);
2172 if (posix_memalign(&super
->buf
, 512, super
->len
) != 0) {
2175 Name
": unable to allocate %zu byte mpb buffer\n",
2180 memcpy(super
->buf
, anchor
, 512);
2182 sectors
= mpb_sectors(anchor
) - 1;
2185 check_sum
= __gen_imsm_checksum(super
->anchor
);
2186 if (check_sum
!= __le32_to_cpu(super
->anchor
->check_sum
)) {
2189 Name
": IMSM checksum %x != %x on %s\n",
2191 __le32_to_cpu(super
->anchor
->check_sum
),
2199 /* read the extended mpb */
2200 if (lseek64(fd
, dsize
- (512 * (2 + sectors
)), SEEK_SET
) < 0) {
2203 Name
": Cannot seek to extended mpb on %s: %s\n",
2204 devname
, strerror(errno
));
2208 if (read(fd
, super
->buf
+ 512, super
->len
- 512) != super
->len
- 512) {
2211 Name
": Cannot read extended mpb on %s: %s\n",
2212 devname
, strerror(errno
));
2216 check_sum
= __gen_imsm_checksum(super
->anchor
);
2217 if (check_sum
!= __le32_to_cpu(super
->anchor
->check_sum
)) {
2220 Name
": IMSM checksum %x != %x on %s\n",
2221 check_sum
, __le32_to_cpu(super
->anchor
->check_sum
),
2226 /* FIXME the BBM log is disk specific so we cannot use this global
2227 * buffer for all disks. Ok for now since we only look at the global
2228 * bbm_log_size parameter to gate assembly
2230 super
->bbm_log
= __get_imsm_bbm_log(super
->anchor
);
2236 load_and_parse_mpb(int fd
, struct intel_super
*super
, char *devname
, int keep_fd
)
2240 err
= load_imsm_mpb(fd
, super
, devname
);
2243 err
= load_imsm_disk(fd
, super
, devname
, keep_fd
);
2246 err
= parse_raid_devices(super
);
2251 static void __free_imsm_disk(struct dl
*d
)
2262 static void free_imsm_disks(struct intel_super
*super
)
2266 while (super
->disks
) {
2268 super
->disks
= d
->next
;
2269 __free_imsm_disk(d
);
2271 while (super
->missing
) {
2273 super
->missing
= d
->next
;
2274 __free_imsm_disk(d
);
2279 /* free all the pieces hanging off of a super pointer */
2280 static void __free_imsm(struct intel_super
*super
, int free_disks
)
2287 free_imsm_disks(super
);
2288 free_devlist(super
);
2290 free((void *) super
->hba
);
2295 static void free_imsm(struct intel_super
*super
)
2297 __free_imsm(super
, 1);
2301 static void free_super_imsm(struct supertype
*st
)
2303 struct intel_super
*super
= st
->sb
;
2312 static struct intel_super
*alloc_super(void)
2314 struct intel_super
*super
= malloc(sizeof(*super
));
2317 memset(super
, 0, sizeof(*super
));
2318 super
->current_vol
= -1;
2319 super
->create_offset
= ~((__u32
) 0);
2320 if (!check_env("IMSM_NO_PLATFORM"))
2321 super
->orom
= find_imsm_orom();
2322 if (super
->orom
&& !check_env("IMSM_TEST_OROM")) {
2323 struct sys_dev
*list
, *ent
;
2325 /* find the first intel ahci controller */
2326 list
= find_driver_devices("pci", "ahci");
2327 for (ent
= list
; ent
; ent
= ent
->next
)
2328 if (devpath_to_vendor(ent
->path
) == 0x8086)
2331 super
->hba
= ent
->path
;
2334 free_sys_dev(&list
);
2342 /* find_missing - helper routine for load_super_imsm_all that identifies
2343 * disks that have disappeared from the system. This routine relies on
2344 * the mpb being uptodate, which it is at load time.
2346 static int find_missing(struct intel_super
*super
)
2349 struct imsm_super
*mpb
= super
->anchor
;
2351 struct imsm_disk
*disk
;
2353 for (i
= 0; i
< mpb
->num_disks
; i
++) {
2354 disk
= __get_imsm_disk(mpb
, i
);
2355 dl
= serial_to_dl(disk
->serial
, super
);
2359 dl
= malloc(sizeof(*dl
));
2365 dl
->devname
= strdup("missing");
2367 serialcpy(dl
->serial
, disk
->serial
);
2370 dl
->next
= super
->missing
;
2371 super
->missing
= dl
;
2377 static struct intel_disk
*disk_list_get(__u8
*serial
, struct intel_disk
*disk_list
)
2379 struct intel_disk
*idisk
= disk_list
;
2382 if (serialcmp(idisk
->disk
.serial
, serial
) == 0)
2384 idisk
= idisk
->next
;
2390 static int __prep_thunderdome(struct intel_super
**table
, int tbl_size
,
2391 struct intel_super
*super
,
2392 struct intel_disk
**disk_list
)
2394 struct imsm_disk
*d
= &super
->disks
->disk
;
2395 struct imsm_super
*mpb
= super
->anchor
;
2398 for (i
= 0; i
< tbl_size
; i
++) {
2399 struct imsm_super
*tbl_mpb
= table
[i
]->anchor
;
2400 struct imsm_disk
*tbl_d
= &table
[i
]->disks
->disk
;
2402 if (tbl_mpb
->family_num
== mpb
->family_num
) {
2403 if (tbl_mpb
->check_sum
== mpb
->check_sum
) {
2404 dprintf("%s: mpb from %d:%d matches %d:%d\n",
2405 __func__
, super
->disks
->major
,
2406 super
->disks
->minor
,
2407 table
[i
]->disks
->major
,
2408 table
[i
]->disks
->minor
);
2412 if (((is_configured(d
) && !is_configured(tbl_d
)) ||
2413 is_configured(d
) == is_configured(tbl_d
)) &&
2414 tbl_mpb
->generation_num
< mpb
->generation_num
) {
2415 /* current version of the mpb is a
2416 * better candidate than the one in
2417 * super_table, but copy over "cross
2418 * generational" status
2420 struct intel_disk
*idisk
;
2422 dprintf("%s: mpb from %d:%d replaces %d:%d\n",
2423 __func__
, super
->disks
->major
,
2424 super
->disks
->minor
,
2425 table
[i
]->disks
->major
,
2426 table
[i
]->disks
->minor
);
2428 idisk
= disk_list_get(tbl_d
->serial
, *disk_list
);
2429 if (idisk
&& is_failed(&idisk
->disk
))
2430 tbl_d
->status
|= FAILED_DISK
;
2433 struct intel_disk
*idisk
;
2434 struct imsm_disk
*disk
;
2436 /* tbl_mpb is more up to date, but copy
2437 * over cross generational status before
2440 disk
= __serial_to_disk(d
->serial
, mpb
, NULL
);
2441 if (disk
&& is_failed(disk
))
2442 d
->status
|= FAILED_DISK
;
2444 idisk
= disk_list_get(d
->serial
, *disk_list
);
2447 if (disk
&& is_configured(disk
))
2448 idisk
->disk
.status
|= CONFIGURED_DISK
;
2451 dprintf("%s: mpb from %d:%d prefer %d:%d\n",
2452 __func__
, super
->disks
->major
,
2453 super
->disks
->minor
,
2454 table
[i
]->disks
->major
,
2455 table
[i
]->disks
->minor
);
2463 table
[tbl_size
++] = super
;
2467 /* update/extend the merged list of imsm_disk records */
2468 for (j
= 0; j
< mpb
->num_disks
; j
++) {
2469 struct imsm_disk
*disk
= __get_imsm_disk(mpb
, j
);
2470 struct intel_disk
*idisk
;
2472 idisk
= disk_list_get(disk
->serial
, *disk_list
);
2474 idisk
->disk
.status
|= disk
->status
;
2475 if (is_configured(&idisk
->disk
) ||
2476 is_failed(&idisk
->disk
))
2477 idisk
->disk
.status
&= ~(SPARE_DISK
);
2479 idisk
= calloc(1, sizeof(*idisk
));
2482 idisk
->owner
= IMSM_UNKNOWN_OWNER
;
2483 idisk
->disk
= *disk
;
2484 idisk
->next
= *disk_list
;
2488 if (serialcmp(idisk
->disk
.serial
, d
->serial
) == 0)
2495 static struct intel_super
*
2496 validate_members(struct intel_super
*super
, struct intel_disk
*disk_list
,
2499 struct imsm_super
*mpb
= super
->anchor
;
2503 for (i
= 0; i
< mpb
->num_disks
; i
++) {
2504 struct imsm_disk
*disk
= __get_imsm_disk(mpb
, i
);
2505 struct intel_disk
*idisk
;
2507 idisk
= disk_list_get(disk
->serial
, disk_list
);
2509 if (idisk
->owner
== owner
||
2510 idisk
->owner
== IMSM_UNKNOWN_OWNER
)
2513 dprintf("%s: '%.16s' owner %d != %d\n",
2514 __func__
, disk
->serial
, idisk
->owner
,
2517 dprintf("%s: unknown disk %x [%d]: %.16s\n",
2518 __func__
, __le32_to_cpu(mpb
->family_num
), i
,
2524 if (ok_count
== mpb
->num_disks
)
2529 static void show_conflicts(__u32 family_num
, struct intel_super
*super_list
)
2531 struct intel_super
*s
;
2533 for (s
= super_list
; s
; s
= s
->next
) {
2534 if (family_num
!= s
->anchor
->family_num
)
2536 fprintf(stderr
, "Conflict, offlining family %#x on '%s'\n",
2537 __le32_to_cpu(family_num
), s
->disks
->devname
);
2541 static struct intel_super
*
2542 imsm_thunderdome(struct intel_super
**super_list
, int len
)
2544 struct intel_super
*super_table
[len
];
2545 struct intel_disk
*disk_list
= NULL
;
2546 struct intel_super
*champion
, *spare
;
2547 struct intel_super
*s
, **del
;
2552 memset(super_table
, 0, sizeof(super_table
));
2553 for (s
= *super_list
; s
; s
= s
->next
)
2554 tbl_size
= __prep_thunderdome(super_table
, tbl_size
, s
, &disk_list
);
2556 for (i
= 0; i
< tbl_size
; i
++) {
2557 struct imsm_disk
*d
;
2558 struct intel_disk
*idisk
;
2559 struct imsm_super
*mpb
= super_table
[i
]->anchor
;
2562 d
= &s
->disks
->disk
;
2564 /* 'd' must appear in merged disk list for its
2565 * configuration to be valid
2567 idisk
= disk_list_get(d
->serial
, disk_list
);
2568 if (idisk
&& idisk
->owner
== i
)
2569 s
= validate_members(s
, disk_list
, i
);
2574 dprintf("%s: marking family: %#x from %d:%d offline\n",
2575 __func__
, mpb
->family_num
,
2576 super_table
[i
]->disks
->major
,
2577 super_table
[i
]->disks
->minor
);
2581 /* This is where the mdadm implementation differs from the Windows
2582 * driver which has no strict concept of a container. We can only
2583 * assemble one family from a container, so when returning a prodigal
2584 * array member to this system the code will not be able to disambiguate
2585 * the container contents that should be assembled ("foreign" versus
2586 * "local"). It requires user intervention to set the orig_family_num
2587 * to a new value to establish a new container. The Windows driver in
2588 * this situation fixes up the volume name in place and manages the
2589 * foreign array as an independent entity.
2594 for (i
= 0; i
< tbl_size
; i
++) {
2595 struct intel_super
*tbl_ent
= super_table
[i
];
2601 if (tbl_ent
->anchor
->num_raid_devs
== 0) {
2606 if (s
&& !is_spare
) {
2607 show_conflicts(tbl_ent
->anchor
->family_num
, *super_list
);
2609 } else if (!s
&& !is_spare
)
2622 fprintf(stderr
, "Chose family %#x on '%s', "
2623 "assemble conflicts to new container with '--update=uuid'\n",
2624 __le32_to_cpu(s
->anchor
->family_num
), s
->disks
->devname
);
2626 /* collect all dl's onto 'champion', and update them to
2627 * champion's version of the status
2629 for (s
= *super_list
; s
; s
= s
->next
) {
2630 struct imsm_super
*mpb
= champion
->anchor
;
2631 struct dl
*dl
= s
->disks
;
2636 for (i
= 0; i
< mpb
->num_disks
; i
++) {
2637 struct imsm_disk
*disk
;
2639 disk
= __serial_to_disk(dl
->serial
, mpb
, &dl
->index
);
2642 /* only set index on disks that are a member of
2643 * a populated contianer, i.e. one with
2646 if (is_failed(&dl
->disk
))
2648 else if (is_spare(&dl
->disk
))
2654 if (i
>= mpb
->num_disks
) {
2655 struct intel_disk
*idisk
;
2657 idisk
= disk_list_get(dl
->serial
, disk_list
);
2658 if (idisk
&& is_spare(&idisk
->disk
) &&
2659 !is_failed(&idisk
->disk
) && !is_configured(&idisk
->disk
))
2667 dl
->next
= champion
->disks
;
2668 champion
->disks
= dl
;
2672 /* delete 'champion' from super_list */
2673 for (del
= super_list
; *del
; ) {
2674 if (*del
== champion
) {
2675 *del
= (*del
)->next
;
2678 del
= &(*del
)->next
;
2680 champion
->next
= NULL
;
2684 struct intel_disk
*idisk
= disk_list
;
2686 disk_list
= disk_list
->next
;
2693 static int load_super_imsm_all(struct supertype
*st
, int fd
, void **sbp
,
2694 char *devname
, int keep_fd
)
2697 struct intel_super
*super_list
= NULL
;
2698 struct intel_super
*super
= NULL
;
2699 int devnum
= fd2devnum(fd
);
2704 enum sysfs_read_flags flags
;
2706 flags
= GET_LEVEL
|GET_VERSION
|GET_DEVS
|GET_STATE
;
2707 if (mdmon_running(devnum
))
2708 flags
|= SKIP_GONE_DEVS
;
2710 /* check if 'fd' an opened container */
2711 sra
= sysfs_read(fd
, 0, flags
);
2715 if (sra
->array
.major_version
!= -1 ||
2716 sra
->array
.minor_version
!= -2 ||
2717 strcmp(sra
->text_version
, "imsm") != 0) {
2722 for (sd
= sra
->devs
, i
= 0; sd
; sd
= sd
->next
, i
++) {
2723 struct intel_super
*s
= alloc_super();
2730 s
->next
= super_list
;
2734 sprintf(nm
, "%d:%d", sd
->disk
.major
, sd
->disk
.minor
);
2735 dfd
= dev_open(nm
, keep_fd
? O_RDWR
: O_RDONLY
);
2739 err
= load_and_parse_mpb(dfd
, s
, NULL
, keep_fd
);
2741 /* retry the load if we might have raced against mdmon */
2742 if (err
== 3 && mdmon_running(devnum
))
2743 for (retry
= 0; retry
< 3; retry
++) {
2745 err
= load_and_parse_mpb(dfd
, s
, NULL
, keep_fd
);
2755 /* all mpbs enter, maybe one leaves */
2756 super
= imsm_thunderdome(&super_list
, i
);
2762 if (find_missing(super
) != 0) {
2768 if (st
->subarray
[0]) {
2773 val
= strtoul(st
->subarray
, &ep
, 10);
2779 if (val
< super
->anchor
->num_raid_devs
)
2780 super
->current_vol
= val
;
2789 while (super_list
) {
2790 struct intel_super
*s
= super_list
;
2792 super_list
= super_list
->next
;
2801 st
->container_dev
= devnum
;
2802 if (err
== 0 && st
->ss
== NULL
) {
2803 st
->ss
= &super_imsm
;
2804 st
->minor_version
= 0;
2805 st
->max_devs
= IMSM_MAX_DEVICES
;
2807 st
->loaded_container
= 1;
2813 static int load_super_imsm(struct supertype
*st
, int fd
, char *devname
)
2815 struct intel_super
*super
;
2819 if (load_super_imsm_all(st
, fd
, &st
->sb
, devname
, 1) == 0)
2823 if (test_partition(fd
))
2824 /* IMSM not allowed on partitions */
2827 free_super_imsm(st
);
2829 super
= alloc_super();
2832 Name
": malloc of %zu failed.\n",
2837 rv
= load_and_parse_mpb(fd
, super
, devname
, 0);
2842 Name
": Failed to load all information "
2843 "sections on %s\n", devname
);
2848 if (st
->subarray
[0]) {
2852 val
= strtoul(st
->subarray
, &ep
, 10);
2858 if (val
< super
->anchor
->num_raid_devs
)
2859 super
->current_vol
= val
;
2867 if (st
->ss
== NULL
) {
2868 st
->ss
= &super_imsm
;
2869 st
->minor_version
= 0;
2870 st
->max_devs
= IMSM_MAX_DEVICES
;
2872 st
->loaded_container
= 0;
2877 static __u16
info_to_blocks_per_strip(mdu_array_info_t
*info
)
2879 if (info
->level
== 1)
2881 return info
->chunk_size
>> 9;
2884 static __u32
info_to_num_data_stripes(mdu_array_info_t
*info
, int num_domains
)
2888 num_stripes
= (info
->size
* 2) / info_to_blocks_per_strip(info
);
2889 num_stripes
/= num_domains
;
2894 static __u32
info_to_blocks_per_member(mdu_array_info_t
*info
)
2896 if (info
->level
== 1)
2897 return info
->size
* 2;
2899 return (info
->size
* 2) & ~(info_to_blocks_per_strip(info
) - 1);
2902 static void imsm_update_version_info(struct intel_super
*super
)
2904 /* update the version and attributes */
2905 struct imsm_super
*mpb
= super
->anchor
;
2907 struct imsm_dev
*dev
;
2908 struct imsm_map
*map
;
2911 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
2912 dev
= get_imsm_dev(super
, i
);
2913 map
= get_imsm_map(dev
, 0);
2914 if (__le32_to_cpu(dev
->size_high
) > 0)
2915 mpb
->attributes
|= MPB_ATTRIB_2TB
;
2917 /* FIXME detect when an array spans a port multiplier */
2919 mpb
->attributes
|= MPB_ATTRIB_PM
;
2922 if (mpb
->num_raid_devs
> 1 ||
2923 mpb
->attributes
!= MPB_ATTRIB_CHECKSUM_VERIFY
) {
2924 version
= MPB_VERSION_ATTRIBS
;
2925 switch (get_imsm_raid_level(map
)) {
2926 case 0: mpb
->attributes
|= MPB_ATTRIB_RAID0
; break;
2927 case 1: mpb
->attributes
|= MPB_ATTRIB_RAID1
; break;
2928 case 10: mpb
->attributes
|= MPB_ATTRIB_RAID10
; break;
2929 case 5: mpb
->attributes
|= MPB_ATTRIB_RAID5
; break;
2932 if (map
->num_members
>= 5)
2933 version
= MPB_VERSION_5OR6_DISK_ARRAY
;
2934 else if (dev
->status
== DEV_CLONE_N_GO
)
2935 version
= MPB_VERSION_CNG
;
2936 else if (get_imsm_raid_level(map
) == 5)
2937 version
= MPB_VERSION_RAID5
;
2938 else if (map
->num_members
>= 3)
2939 version
= MPB_VERSION_3OR4_DISK_ARRAY
;
2940 else if (get_imsm_raid_level(map
) == 1)
2941 version
= MPB_VERSION_RAID1
;
2943 version
= MPB_VERSION_RAID0
;
2945 strcpy(((char *) mpb
->sig
) + strlen(MPB_SIGNATURE
), version
);
2949 static int check_name(struct intel_super
*super
, char *name
, int quiet
)
2951 struct imsm_super
*mpb
= super
->anchor
;
2952 char *reason
= NULL
;
2955 if (strlen(name
) > MAX_RAID_SERIAL_LEN
)
2956 reason
= "must be 16 characters or less";
2958 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
2959 struct imsm_dev
*dev
= get_imsm_dev(super
, i
);
2961 if (strncmp((char *) dev
->volume
, name
, MAX_RAID_SERIAL_LEN
) == 0) {
2962 reason
= "already exists";
2967 if (reason
&& !quiet
)
2968 fprintf(stderr
, Name
": imsm volume name %s\n", reason
);
2973 static int init_super_imsm_volume(struct supertype
*st
, mdu_array_info_t
*info
,
2974 unsigned long long size
, char *name
,
2975 char *homehost
, int *uuid
)
2977 /* We are creating a volume inside a pre-existing container.
2978 * so st->sb is already set.
2980 struct intel_super
*super
= st
->sb
;
2981 struct imsm_super
*mpb
= super
->anchor
;
2982 struct intel_dev
*dv
;
2983 struct imsm_dev
*dev
;
2984 struct imsm_vol
*vol
;
2985 struct imsm_map
*map
;
2986 int idx
= mpb
->num_raid_devs
;
2988 unsigned long long array_blocks
;
2989 size_t size_old
, size_new
;
2990 __u32 num_data_stripes
;
2992 if (super
->orom
&& mpb
->num_raid_devs
>= super
->orom
->vpa
) {
2993 fprintf(stderr
, Name
": This imsm-container already has the "
2994 "maximum of %d volumes\n", super
->orom
->vpa
);
2998 /* ensure the mpb is large enough for the new data */
2999 size_old
= __le32_to_cpu(mpb
->mpb_size
);
3000 size_new
= disks_to_mpb_size(info
->nr_disks
);
3001 if (size_new
> size_old
) {
3003 size_t size_round
= ROUND_UP(size_new
, 512);
3005 if (posix_memalign(&mpb_new
, 512, size_round
) != 0) {
3006 fprintf(stderr
, Name
": could not allocate new mpb\n");
3009 memcpy(mpb_new
, mpb
, size_old
);
3012 super
->anchor
= mpb_new
;
3013 mpb
->mpb_size
= __cpu_to_le32(size_new
);
3014 memset(mpb_new
+ size_old
, 0, size_round
- size_old
);
3016 super
->current_vol
= idx
;
3017 /* when creating the first raid device in this container set num_disks
3018 * to zero, i.e. delete this spare and add raid member devices in
3019 * add_to_super_imsm_volume()
3021 if (super
->current_vol
== 0)
3024 if (!check_name(super
, name
, 0))
3026 sprintf(st
->subarray
, "%d", idx
);
3027 dv
= malloc(sizeof(*dv
));
3029 fprintf(stderr
, Name
": failed to allocate device list entry\n");
3032 dev
= malloc(sizeof(*dev
) + sizeof(__u32
) * (info
->raid_disks
- 1));
3035 fprintf(stderr
, Name
": could not allocate raid device\n");
3038 strncpy((char *) dev
->volume
, name
, MAX_RAID_SERIAL_LEN
);
3039 if (info
->level
== 1)
3040 array_blocks
= info_to_blocks_per_member(info
);
3042 array_blocks
= calc_array_size(info
->level
, info
->raid_disks
,
3043 info
->layout
, info
->chunk_size
,
3045 /* round array size down to closest MB */
3046 array_blocks
= (array_blocks
>> SECT_PER_MB_SHIFT
) << SECT_PER_MB_SHIFT
;
3048 dev
->size_low
= __cpu_to_le32((__u32
) array_blocks
);
3049 dev
->size_high
= __cpu_to_le32((__u32
) (array_blocks
>> 32));
3050 dev
->status
= __cpu_to_le32(0);
3051 dev
->reserved_blocks
= __cpu_to_le32(0);
3053 vol
->migr_state
= 0;
3054 set_migr_type(dev
, MIGR_INIT
);
3056 vol
->curr_migr_unit
= 0;
3057 map
= get_imsm_map(dev
, 0);
3058 map
->pba_of_lba0
= __cpu_to_le32(super
->create_offset
);
3059 map
->blocks_per_member
= __cpu_to_le32(info_to_blocks_per_member(info
));
3060 map
->blocks_per_strip
= __cpu_to_le16(info_to_blocks_per_strip(info
));
3061 map
->failed_disk_num
= ~0;
3062 map
->map_state
= info
->level
? IMSM_T_STATE_UNINITIALIZED
:
3063 IMSM_T_STATE_NORMAL
;
3066 if (info
->level
== 1 && info
->raid_disks
> 2) {
3069 fprintf(stderr
, Name
": imsm does not support more than 2 disks"
3070 "in a raid1 volume\n");
3074 map
->raid_level
= info
->level
;
3075 if (info
->level
== 10) {
3076 map
->raid_level
= 1;
3077 map
->num_domains
= info
->raid_disks
/ 2;
3078 } else if (info
->level
== 1)
3079 map
->num_domains
= info
->raid_disks
;
3081 map
->num_domains
= 1;
3083 num_data_stripes
= info_to_num_data_stripes(info
, map
->num_domains
);
3084 map
->num_data_stripes
= __cpu_to_le32(num_data_stripes
);
3086 map
->num_members
= info
->raid_disks
;
3087 for (i
= 0; i
< map
->num_members
; i
++) {
3088 /* initialized in add_to_super */
3089 set_imsm_ord_tbl_ent(map
, i
, IMSM_ORD_REBUILD
);
3091 mpb
->num_raid_devs
++;
3094 dv
->index
= super
->current_vol
;
3095 dv
->next
= super
->devlist
;
3096 super
->devlist
= dv
;
3098 imsm_update_version_info(super
);
3103 static int init_super_imsm(struct supertype
*st
, mdu_array_info_t
*info
,
3104 unsigned long long size
, char *name
,
3105 char *homehost
, int *uuid
)
3107 /* This is primarily called by Create when creating a new array.
3108 * We will then get add_to_super called for each component, and then
3109 * write_init_super called to write it out to each device.
3110 * For IMSM, Create can create on fresh devices or on a pre-existing
3112 * To create on a pre-existing array a different method will be called.
3113 * This one is just for fresh drives.
3115 struct intel_super
*super
;
3116 struct imsm_super
*mpb
;
3121 return init_super_imsm_volume(st
, info
, size
, name
, homehost
, uuid
);
3124 mpb_size
= disks_to_mpb_size(info
->nr_disks
);
3128 super
= alloc_super();
3129 if (super
&& posix_memalign(&super
->buf
, 512, mpb_size
) != 0) {
3134 fprintf(stderr
, Name
3135 ": %s could not allocate superblock\n", __func__
);
3138 memset(super
->buf
, 0, mpb_size
);
3140 mpb
->mpb_size
= __cpu_to_le32(mpb_size
);
3144 /* zeroing superblock */
3148 mpb
->attributes
= MPB_ATTRIB_CHECKSUM_VERIFY
;
3150 version
= (char *) mpb
->sig
;
3151 strcpy(version
, MPB_SIGNATURE
);
3152 version
+= strlen(MPB_SIGNATURE
);
3153 strcpy(version
, MPB_VERSION_RAID0
);
3159 static int add_to_super_imsm_volume(struct supertype
*st
, mdu_disk_info_t
*dk
,
3160 int fd
, char *devname
)
3162 struct intel_super
*super
= st
->sb
;
3163 struct imsm_super
*mpb
= super
->anchor
;
3165 struct imsm_dev
*dev
;
3166 struct imsm_map
*map
;
3169 dev
= get_imsm_dev(super
, super
->current_vol
);
3170 map
= get_imsm_map(dev
, 0);
3172 if (! (dk
->state
& (1<<MD_DISK_SYNC
))) {
3173 fprintf(stderr
, Name
": %s: Cannot add spare devices to IMSM volume\n",
3179 /* we're doing autolayout so grab the pre-marked (in
3180 * validate_geometry) raid_disk
3182 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
3183 if (dl
->raiddisk
== dk
->raid_disk
)
3186 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
3187 if (dl
->major
== dk
->major
&&
3188 dl
->minor
== dk
->minor
)
3193 fprintf(stderr
, Name
": %s is not a member of the same container\n", devname
);
3197 /* add a pristine spare to the metadata */
3198 if (dl
->index
< 0) {
3199 dl
->index
= super
->anchor
->num_disks
;
3200 super
->anchor
->num_disks
++;
3202 /* Check the device has not already been added */
3203 slot
= get_imsm_disk_slot(map
, dl
->index
);
3205 (get_imsm_ord_tbl_ent(dev
, slot
) & IMSM_ORD_REBUILD
) == 0) {
3206 fprintf(stderr
, Name
": %s has been included in this array twice\n",
3210 set_imsm_ord_tbl_ent(map
, dk
->number
, dl
->index
);
3211 dl
->disk
.status
= CONFIGURED_DISK
;
3213 /* if we are creating the first raid device update the family number */
3214 if (super
->current_vol
== 0) {
3216 struct imsm_dev
*_dev
= __get_imsm_dev(mpb
, 0);
3217 struct imsm_disk
*_disk
= __get_imsm_disk(mpb
, dl
->index
);
3219 if (!_dev
|| !_disk
) {
3220 fprintf(stderr
, Name
": BUG mpb setup error\n");
3226 sum
+= __gen_imsm_checksum(mpb
);
3227 mpb
->family_num
= __cpu_to_le32(sum
);
3228 mpb
->orig_family_num
= mpb
->family_num
;
3234 static int add_to_super_imsm(struct supertype
*st
, mdu_disk_info_t
*dk
,
3235 int fd
, char *devname
)
3237 struct intel_super
*super
= st
->sb
;
3239 unsigned long long size
;
3244 /* if we are on an RAID enabled platform check that the disk is
3245 * attached to the raid controller
3247 if (super
->hba
&& !disk_attached_to_hba(fd
, super
->hba
)) {
3249 Name
": %s is not attached to the raid controller: %s\n",
3250 devname
? : "disk", super
->hba
);
3254 if (super
->current_vol
>= 0)
3255 return add_to_super_imsm_volume(st
, dk
, fd
, devname
);
3258 dd
= malloc(sizeof(*dd
));
3261 Name
": malloc failed %s:%d.\n", __func__
, __LINE__
);
3264 memset(dd
, 0, sizeof(*dd
));
3265 dd
->major
= major(stb
.st_rdev
);
3266 dd
->minor
= minor(stb
.st_rdev
);
3268 dd
->devname
= devname
? strdup(devname
) : NULL
;
3271 rv
= imsm_read_serial(fd
, devname
, dd
->serial
);
3274 Name
": failed to retrieve scsi serial, aborting\n");
3279 get_dev_size(fd
, NULL
, &size
);
3281 serialcpy(dd
->disk
.serial
, dd
->serial
);
3282 dd
->disk
.total_blocks
= __cpu_to_le32(size
);
3283 dd
->disk
.status
= SPARE_DISK
;
3284 if (sysfs_disk_to_scsi_id(fd
, &id
) == 0)
3285 dd
->disk
.scsi_id
= __cpu_to_le32(id
);
3287 dd
->disk
.scsi_id
= __cpu_to_le32(0);
3289 if (st
->update_tail
) {
3290 dd
->next
= super
->add
;
3293 dd
->next
= super
->disks
;
3300 static int store_imsm_mpb(int fd
, struct imsm_super
*mpb
);
3304 struct imsm_super anchor
;
3305 } spare_record
__attribute__ ((aligned(512)));
3307 /* spare records have their own family number and do not have any defined raid
3310 static int write_super_imsm_spares(struct intel_super
*super
, int doclose
)
3312 struct imsm_super
*mpb
= super
->anchor
;
3313 struct imsm_super
*spare
= &spare_record
.anchor
;
3317 spare
->mpb_size
= __cpu_to_le32(sizeof(struct imsm_super
)),
3318 spare
->generation_num
= __cpu_to_le32(1UL),
3319 spare
->attributes
= MPB_ATTRIB_CHECKSUM_VERIFY
;
3320 spare
->num_disks
= 1,
3321 spare
->num_raid_devs
= 0,
3322 spare
->cache_size
= mpb
->cache_size
,
3323 spare
->pwr_cycle_count
= __cpu_to_le32(1),
3325 snprintf((char *) spare
->sig
, MAX_SIGNATURE_LENGTH
,
3326 MPB_SIGNATURE MPB_VERSION_RAID0
);
3328 for (d
= super
->disks
; d
; d
= d
->next
) {
3332 spare
->disk
[0] = d
->disk
;
3333 sum
= __gen_imsm_checksum(spare
);
3334 spare
->family_num
= __cpu_to_le32(sum
);
3335 spare
->orig_family_num
= 0;
3336 sum
= __gen_imsm_checksum(spare
);
3337 spare
->check_sum
= __cpu_to_le32(sum
);
3339 if (store_imsm_mpb(d
->fd
, spare
)) {
3340 fprintf(stderr
, "%s: failed for device %d:%d %s\n",
3341 __func__
, d
->major
, d
->minor
, strerror(errno
));
3353 static int write_super_imsm(struct intel_super
*super
, int doclose
)
3355 struct imsm_super
*mpb
= super
->anchor
;
3361 __u32 mpb_size
= sizeof(struct imsm_super
) - sizeof(struct imsm_disk
);
3363 /* 'generation' is incremented everytime the metadata is written */
3364 generation
= __le32_to_cpu(mpb
->generation_num
);
3366 mpb
->generation_num
= __cpu_to_le32(generation
);
3368 /* fix up cases where previous mdadm releases failed to set
3371 if (mpb
->orig_family_num
== 0)
3372 mpb
->orig_family_num
= mpb
->family_num
;
3374 mpb_size
+= sizeof(struct imsm_disk
) * mpb
->num_disks
;
3375 for (d
= super
->disks
; d
; d
= d
->next
) {
3379 mpb
->disk
[d
->index
] = d
->disk
;
3381 for (d
= super
->missing
; d
; d
= d
->next
)
3382 mpb
->disk
[d
->index
] = d
->disk
;
3384 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
3385 struct imsm_dev
*dev
= __get_imsm_dev(mpb
, i
);
3387 imsm_copy_dev(dev
, get_imsm_dev(super
, i
));
3388 mpb_size
+= sizeof_imsm_dev(dev
, 0);
3390 mpb_size
+= __le32_to_cpu(mpb
->bbm_log_size
);
3391 mpb
->mpb_size
= __cpu_to_le32(mpb_size
);
3393 /* recalculate checksum */
3394 sum
= __gen_imsm_checksum(mpb
);
3395 mpb
->check_sum
= __cpu_to_le32(sum
);
3397 /* write the mpb for disks that compose raid devices */
3398 for (d
= super
->disks
; d
; d
= d
->next
) {
3401 if (store_imsm_mpb(d
->fd
, mpb
))
3402 fprintf(stderr
, "%s: failed for device %d:%d %s\n",
3403 __func__
, d
->major
, d
->minor
, strerror(errno
));
3411 return write_super_imsm_spares(super
, doclose
);
3417 static int create_array(struct supertype
*st
, int dev_idx
)
3420 struct imsm_update_create_array
*u
;
3421 struct intel_super
*super
= st
->sb
;
3422 struct imsm_dev
*dev
= get_imsm_dev(super
, dev_idx
);
3423 struct imsm_map
*map
= get_imsm_map(dev
, 0);
3424 struct disk_info
*inf
;
3425 struct imsm_disk
*disk
;
3428 len
= sizeof(*u
) - sizeof(*dev
) + sizeof_imsm_dev(dev
, 0) +
3429 sizeof(*inf
) * map
->num_members
;
3432 fprintf(stderr
, "%s: failed to allocate update buffer\n",
3437 u
->type
= update_create_array
;
3438 u
->dev_idx
= dev_idx
;
3439 imsm_copy_dev(&u
->dev
, dev
);
3440 inf
= get_disk_info(u
);
3441 for (i
= 0; i
< map
->num_members
; i
++) {
3442 int idx
= get_imsm_disk_idx(dev
, i
);
3444 disk
= get_imsm_disk(super
, idx
);
3445 serialcpy(inf
[i
].serial
, disk
->serial
);
3447 append_metadata_update(st
, u
, len
);
3452 static int _add_disk(struct supertype
*st
)
3454 struct intel_super
*super
= st
->sb
;
3456 struct imsm_update_add_disk
*u
;
3464 fprintf(stderr
, "%s: failed to allocate update buffer\n",
3469 u
->type
= update_add_disk
;
3470 append_metadata_update(st
, u
, len
);
3475 static int write_init_super_imsm(struct supertype
*st
)
3477 struct intel_super
*super
= st
->sb
;
3478 int current_vol
= super
->current_vol
;
3480 /* we are done with current_vol reset it to point st at the container */
3481 super
->current_vol
= -1;
3483 if (st
->update_tail
) {
3484 /* queue the recently created array / added disk
3485 * as a metadata update */
3489 /* determine if we are creating a volume or adding a disk */
3490 if (current_vol
< 0) {
3491 /* in the add disk case we are running in mdmon
3492 * context, so don't close fd's
3494 return _add_disk(st
);
3496 rv
= create_array(st
, current_vol
);
3498 for (d
= super
->disks
; d
; d
= d
->next
) {
3506 for (d
= super
->disks
; d
; d
= d
->next
)
3507 Kill(d
->devname
, NULL
, 0, 1, 1);
3508 return write_super_imsm(st
->sb
, 1);
3513 static int store_super_imsm(struct supertype
*st
, int fd
)
3515 struct intel_super
*super
= st
->sb
;
3516 struct imsm_super
*mpb
= super
? super
->anchor
: NULL
;
3522 return store_imsm_mpb(fd
, mpb
);
3528 static int imsm_bbm_log_size(struct imsm_super
*mpb
)
3530 return __le32_to_cpu(mpb
->bbm_log_size
);
3534 static int validate_geometry_imsm_container(struct supertype
*st
, int level
,
3535 int layout
, int raiddisks
, int chunk
,
3536 unsigned long long size
, char *dev
,
3537 unsigned long long *freesize
,
3541 unsigned long long ldsize
;
3542 const struct imsm_orom
*orom
;
3544 if (level
!= LEVEL_CONTAINER
)
3549 if (check_env("IMSM_NO_PLATFORM"))
3552 orom
= find_imsm_orom();
3553 if (orom
&& raiddisks
> orom
->tds
) {
3555 fprintf(stderr
, Name
": %d exceeds maximum number of"
3556 " platform supported disks: %d\n",
3557 raiddisks
, orom
->tds
);
3561 fd
= open(dev
, O_RDONLY
|O_EXCL
, 0);
3564 fprintf(stderr
, Name
": imsm: Cannot open %s: %s\n",
3565 dev
, strerror(errno
));
3568 if (!get_dev_size(fd
, dev
, &ldsize
)) {
3574 *freesize
= avail_size_imsm(st
, ldsize
>> 9);
3579 static unsigned long long find_size(struct extent
*e
, int *idx
, int num_extents
)
3581 const unsigned long long base_start
= e
[*idx
].start
;
3582 unsigned long long end
= base_start
+ e
[*idx
].size
;
3585 if (base_start
== end
)
3589 for (i
= *idx
; i
< num_extents
; i
++) {
3590 /* extend overlapping extents */
3591 if (e
[i
].start
>= base_start
&&
3592 e
[i
].start
<= end
) {
3595 if (e
[i
].start
+ e
[i
].size
> end
)
3596 end
= e
[i
].start
+ e
[i
].size
;
3597 } else if (e
[i
].start
> end
) {
3603 return end
- base_start
;
3606 static unsigned long long merge_extents(struct intel_super
*super
, int sum_extents
)
3608 /* build a composite disk with all known extents and generate a new
3609 * 'maxsize' given the "all disks in an array must share a common start
3610 * offset" constraint
3612 struct extent
*e
= calloc(sum_extents
, sizeof(*e
));
3616 unsigned long long pos
;
3617 unsigned long long start
= 0;
3618 unsigned long long maxsize
;
3619 unsigned long reserve
;
3624 /* coalesce and sort all extents. also, check to see if we need to
3625 * reserve space between member arrays
3628 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
3631 for (i
= 0; i
< dl
->extent_cnt
; i
++)
3634 qsort(e
, sum_extents
, sizeof(*e
), cmp_extent
);
3639 while (i
< sum_extents
) {
3640 e
[j
].start
= e
[i
].start
;
3641 e
[j
].size
= find_size(e
, &i
, sum_extents
);
3643 if (e
[j
-1].size
== 0)
3652 unsigned long long esize
;
3654 esize
= e
[i
].start
- pos
;
3655 if (esize
>= maxsize
) {
3660 pos
= e
[i
].start
+ e
[i
].size
;
3662 } while (e
[i
-1].size
);
3668 /* FIXME assumes volume at offset 0 is the first volume in a
3671 if (start_extent
> 0)
3672 reserve
= IMSM_RESERVED_SECTORS
; /* gap between raid regions */
3676 if (maxsize
< reserve
)
3679 super
->create_offset
= ~((__u32
) 0);
3680 if (start
+ reserve
> super
->create_offset
)
3681 return 0; /* start overflows create_offset */
3682 super
->create_offset
= start
+ reserve
;
3684 return maxsize
- reserve
;
3687 static int is_raid_level_supported(const struct imsm_orom
*orom
, int level
, int raiddisks
)
3689 if (level
< 0 || level
== 6 || level
== 4)
3692 /* if we have an orom prevent invalid raid levels */
3695 case 0: return imsm_orom_has_raid0(orom
);
3698 return imsm_orom_has_raid1e(orom
);
3699 return imsm_orom_has_raid1(orom
) && raiddisks
== 2;
3700 case 10: return imsm_orom_has_raid10(orom
) && raiddisks
== 4;
3701 case 5: return imsm_orom_has_raid5(orom
) && raiddisks
> 2;
3704 return 1; /* not on an Intel RAID platform so anything goes */
3709 #define pr_vrb(fmt, arg...) (void) (verbose && fprintf(stderr, Name fmt, ##arg))
3711 validate_geometry_imsm_orom(struct intel_super
*super
, int level
, int layout
,
3712 int raiddisks
, int chunk
, int verbose
)
3714 if (!is_raid_level_supported(super
->orom
, level
, raiddisks
)) {
3715 pr_vrb(": platform does not support raid%d with %d disk%s\n",
3716 level
, raiddisks
, raiddisks
> 1 ? "s" : "");
3719 if (super
->orom
&& level
!= 1 &&
3720 !imsm_orom_has_chunk(super
->orom
, chunk
)) {
3721 pr_vrb(": platform does not support a chunk size of: %d\n", chunk
);
3724 if (layout
!= imsm_level_to_layout(level
)) {
3726 pr_vrb(": imsm raid 5 only supports the left-asymmetric layout\n");
3727 else if (level
== 10)
3728 pr_vrb(": imsm raid 10 only supports the n2 layout\n");
3730 pr_vrb(": imsm unknown layout %#x for this raid level %d\n",
3738 /* validate_geometry_imsm_volume - lifted from validate_geometry_ddf_bvd
3739 * FIX ME add ahci details
3741 static int validate_geometry_imsm_volume(struct supertype
*st
, int level
,
3742 int layout
, int raiddisks
, int chunk
,
3743 unsigned long long size
, char *dev
,
3744 unsigned long long *freesize
,
3748 struct intel_super
*super
= st
->sb
;
3749 struct imsm_super
*mpb
= super
->anchor
;
3751 unsigned long long pos
= 0;
3752 unsigned long long maxsize
;
3756 /* We must have the container info already read in. */
3760 if (!validate_geometry_imsm_orom(super
, level
, layout
, raiddisks
, chunk
, verbose
))
3764 /* General test: make sure there is space for
3765 * 'raiddisks' device extents of size 'size' at a given
3768 unsigned long long minsize
= size
;
3769 unsigned long long start_offset
= MaxSector
;
3772 minsize
= MPB_SECTOR_CNT
+ IMSM_RESERVED_SECTORS
;
3773 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
3778 e
= get_extents(super
, dl
);
3781 unsigned long long esize
;
3782 esize
= e
[i
].start
- pos
;
3783 if (esize
>= minsize
)
3785 if (found
&& start_offset
== MaxSector
) {
3788 } else if (found
&& pos
!= start_offset
) {
3792 pos
= e
[i
].start
+ e
[i
].size
;
3794 } while (e
[i
-1].size
);
3799 if (dcnt
< raiddisks
) {
3801 fprintf(stderr
, Name
": imsm: Not enough "
3802 "devices with space for this array "
3810 /* This device must be a member of the set */
3811 if (stat(dev
, &stb
) < 0)
3813 if ((S_IFMT
& stb
.st_mode
) != S_IFBLK
)
3815 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
3816 if (dl
->major
== major(stb
.st_rdev
) &&
3817 dl
->minor
== minor(stb
.st_rdev
))
3822 fprintf(stderr
, Name
": %s is not in the "
3823 "same imsm set\n", dev
);
3825 } else if (super
->orom
&& dl
->index
< 0 && mpb
->num_raid_devs
) {
3826 /* If a volume is present then the current creation attempt
3827 * cannot incorporate new spares because the orom may not
3828 * understand this configuration (all member disks must be
3829 * members of each array in the container).
3831 fprintf(stderr
, Name
": %s is a spare and a volume"
3832 " is already defined for this container\n", dev
);
3833 fprintf(stderr
, Name
": The option-rom requires all member"
3834 " disks to be a member of all volumes\n");
3838 /* retrieve the largest free space block */
3839 e
= get_extents(super
, dl
);
3844 unsigned long long esize
;
3846 esize
= e
[i
].start
- pos
;
3847 if (esize
>= maxsize
)
3849 pos
= e
[i
].start
+ e
[i
].size
;
3851 } while (e
[i
-1].size
);
3856 fprintf(stderr
, Name
": unable to determine free space for: %s\n",
3860 if (maxsize
< size
) {
3862 fprintf(stderr
, Name
": %s not enough space (%llu < %llu)\n",
3863 dev
, maxsize
, size
);
3867 /* count total number of extents for merge */
3869 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
3871 i
+= dl
->extent_cnt
;
3873 maxsize
= merge_extents(super
, i
);
3874 if (maxsize
< size
|| maxsize
== 0) {
3876 fprintf(stderr
, Name
": not enough space after merge (%llu < %llu)\n",
3881 *freesize
= maxsize
;
3886 static int reserve_space(struct supertype
*st
, int raiddisks
,
3887 unsigned long long size
, int chunk
,
3888 unsigned long long *freesize
)
3890 struct intel_super
*super
= st
->sb
;
3891 struct imsm_super
*mpb
= super
->anchor
;
3896 unsigned long long maxsize
;
3897 unsigned long long minsize
;
3901 /* find the largest common start free region of the possible disks */
3905 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
3911 /* don't activate new spares if we are orom constrained
3912 * and there is already a volume active in the container
3914 if (super
->orom
&& dl
->index
< 0 && mpb
->num_raid_devs
)
3917 e
= get_extents(super
, dl
);
3920 for (i
= 1; e
[i
-1].size
; i
++)
3928 maxsize
= merge_extents(super
, extent_cnt
);
3933 if (cnt
< raiddisks
||
3934 (super
->orom
&& used
&& used
!= raiddisks
) ||
3935 maxsize
< minsize
||
3937 fprintf(stderr
, Name
": not enough devices with space to create array.\n");
3938 return 0; /* No enough free spaces large enough */
3950 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
3952 dl
->raiddisk
= cnt
++;
3959 static int validate_geometry_imsm(struct supertype
*st
, int level
, int layout
,
3960 int raiddisks
, int chunk
, unsigned long long size
,
3961 char *dev
, unsigned long long *freesize
,
3968 /* if given unused devices create a container
3969 * if given given devices in a container create a member volume
3971 if (level
== LEVEL_CONTAINER
) {
3972 /* Must be a fresh device to add to a container */
3973 return validate_geometry_imsm_container(st
, level
, layout
,
3974 raiddisks
, chunk
, size
,
3980 if (st
->sb
&& freesize
) {
3981 /* we are being asked to automatically layout a
3982 * new volume based on the current contents of
3983 * the container. If the the parameters can be
3984 * satisfied reserve_space will record the disks,
3985 * start offset, and size of the volume to be
3986 * created. add_to_super and getinfo_super
3987 * detect when autolayout is in progress.
3989 if (!validate_geometry_imsm_orom(st
->sb
, level
, layout
,
3993 return reserve_space(st
, raiddisks
, size
, chunk
, freesize
);
3998 /* creating in a given container */
3999 return validate_geometry_imsm_volume(st
, level
, layout
,
4000 raiddisks
, chunk
, size
,
4001 dev
, freesize
, verbose
);
4004 /* This device needs to be a device in an 'imsm' container */
4005 fd
= open(dev
, O_RDONLY
|O_EXCL
, 0);
4009 Name
": Cannot create this array on device %s\n",
4014 if (errno
!= EBUSY
|| (fd
= open(dev
, O_RDONLY
, 0)) < 0) {
4016 fprintf(stderr
, Name
": Cannot open %s: %s\n",
4017 dev
, strerror(errno
));
4020 /* Well, it is in use by someone, maybe an 'imsm' container. */
4021 cfd
= open_container(fd
);
4025 fprintf(stderr
, Name
": Cannot use %s: It is busy\n",
4029 sra
= sysfs_read(cfd
, 0, GET_VERSION
);
4030 if (sra
&& sra
->array
.major_version
== -1 &&
4031 strcmp(sra
->text_version
, "imsm") == 0)
4035 /* This is a member of a imsm container. Load the container
4036 * and try to create a volume
4038 struct intel_super
*super
;
4040 if (load_super_imsm_all(st
, cfd
, (void **) &super
, NULL
, 1) == 0) {
4042 st
->container_dev
= fd2devnum(cfd
);
4044 return validate_geometry_imsm_volume(st
, level
, layout
,
4052 fprintf(stderr
, Name
": failed container membership check\n");
4058 static void handle_missing(struct intel_super
*super
, struct imsm_dev
*dev
);
4060 static int kill_subarray_imsm(struct supertype
*st
)
4062 /* remove the subarray currently referenced by ->current_vol */
4064 struct intel_dev
**dp
;
4065 struct intel_super
*super
= st
->sb
;
4066 __u8 current_vol
= super
->current_vol
;
4067 struct imsm_super
*mpb
= super
->anchor
;
4069 if (super
->current_vol
< 0)
4071 super
->current_vol
= -1; /* invalidate subarray cursor */
4073 /* block deletions that would change the uuid of active subarrays
4075 * FIXME when immutable ids are available, but note that we'll
4076 * also need to fixup the invalidated/active subarray indexes in
4079 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
4082 if (i
< current_vol
)
4084 sprintf(subarray
, "%u", i
);
4085 if (is_subarray_active(subarray
, st
->devname
)) {
4087 Name
": deleting subarray-%d would change the UUID of active subarray-%d, aborting\n",
4094 if (st
->update_tail
) {
4095 struct imsm_update_kill_array
*u
= malloc(sizeof(*u
));
4099 u
->type
= update_kill_array
;
4100 u
->dev_idx
= current_vol
;
4101 append_metadata_update(st
, u
, sizeof(*u
));
4106 for (dp
= &super
->devlist
; *dp
;)
4107 if ((*dp
)->index
== current_vol
) {
4110 handle_missing(super
, (*dp
)->dev
);
4111 if ((*dp
)->index
> current_vol
)
4116 /* no more raid devices, all active components are now spares,
4117 * but of course failed are still failed
4119 if (--mpb
->num_raid_devs
== 0) {
4122 for (d
= super
->disks
; d
; d
= d
->next
)
4123 if (d
->index
> -2) {
4125 d
->disk
.status
= SPARE_DISK
;
4129 super
->updates_pending
++;
4134 static int update_subarray_imsm(struct supertype
*st
, char *update
, mddev_ident_t ident
)
4136 /* update the subarray currently referenced by ->current_vol */
4137 struct intel_super
*super
= st
->sb
;
4138 struct imsm_super
*mpb
= super
->anchor
;
4140 if (super
->current_vol
< 0)
4143 if (strcmp(update
, "name") == 0) {
4144 char *name
= ident
->name
;
4146 if (is_subarray_active(st
->subarray
, st
->devname
)) {
4148 Name
": Unable to update name of active subarray\n");
4152 if (!check_name(super
, name
, 0))
4155 if (st
->update_tail
) {
4156 struct imsm_update_rename_array
*u
= malloc(sizeof(*u
));
4160 u
->type
= update_rename_array
;
4161 u
->dev_idx
= super
->current_vol
;
4162 snprintf((char *) u
->name
, MAX_RAID_SERIAL_LEN
, "%s", name
);
4163 append_metadata_update(st
, u
, sizeof(*u
));
4165 struct imsm_dev
*dev
;
4168 dev
= get_imsm_dev(super
, super
->current_vol
);
4169 snprintf((char *) dev
->volume
, MAX_RAID_SERIAL_LEN
, "%s", name
);
4170 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
4171 dev
= get_imsm_dev(super
, i
);
4172 handle_missing(super
, dev
);
4174 super
->updates_pending
++;
4181 #endif /* MDASSEMBLE */
4183 static int is_rebuilding(struct imsm_dev
*dev
)
4185 struct imsm_map
*migr_map
;
4187 if (!dev
->vol
.migr_state
)
4190 if (migr_type(dev
) != MIGR_REBUILD
)
4193 migr_map
= get_imsm_map(dev
, 1);
4195 if (migr_map
->map_state
== IMSM_T_STATE_DEGRADED
)
4201 static void update_recovery_start(struct imsm_dev
*dev
, struct mdinfo
*array
)
4203 struct mdinfo
*rebuild
= NULL
;
4207 if (!is_rebuilding(dev
))
4210 /* Find the rebuild target, but punt on the dual rebuild case */
4211 for (d
= array
->devs
; d
; d
= d
->next
)
4212 if (d
->recovery_start
== 0) {
4218 units
= __le32_to_cpu(dev
->vol
.curr_migr_unit
);
4219 rebuild
->recovery_start
= units
* blocks_per_migr_unit(dev
);
4223 static struct mdinfo
*container_content_imsm(struct supertype
*st
)
4225 /* Given a container loaded by load_super_imsm_all,
4226 * extract information about all the arrays into
4229 * For each imsm_dev create an mdinfo, fill it in,
4230 * then look for matching devices in super->disks
4231 * and create appropriate device mdinfo.
4233 struct intel_super
*super
= st
->sb
;
4234 struct imsm_super
*mpb
= super
->anchor
;
4235 struct mdinfo
*rest
= NULL
;
4238 /* do not assemble arrays that might have bad blocks */
4239 if (imsm_bbm_log_size(super
->anchor
)) {
4240 fprintf(stderr
, Name
": BBM log found in metadata. "
4241 "Cannot activate array(s).\n");
4245 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
4246 struct imsm_dev
*dev
= get_imsm_dev(super
, i
);
4247 struct imsm_map
*map
= get_imsm_map(dev
, 0);
4248 struct mdinfo
*this;
4251 /* do not publish arrays that are in the middle of an
4252 * unsupported migration
4254 if (dev
->vol
.migr_state
&&
4255 (migr_type(dev
) == MIGR_GEN_MIGR
||
4256 migr_type(dev
) == MIGR_STATE_CHANGE
)) {
4257 fprintf(stderr
, Name
": cannot assemble volume '%.16s':"
4258 " unsupported migration in progress\n",
4263 this = malloc(sizeof(*this));
4265 fprintf(stderr
, Name
": failed to allocate %zu bytes\n",
4269 memset(this, 0, sizeof(*this));
4272 super
->current_vol
= i
;
4273 getinfo_super_imsm_volume(st
, this);
4274 for (slot
= 0 ; slot
< map
->num_members
; slot
++) {
4275 unsigned long long recovery_start
;
4276 struct mdinfo
*info_d
;
4283 idx
= get_imsm_disk_idx(dev
, slot
);
4284 ord
= get_imsm_ord_tbl_ent(dev
, slot
);
4285 for (d
= super
->disks
; d
; d
= d
->next
)
4286 if (d
->index
== idx
)
4289 recovery_start
= MaxSector
;
4292 if (d
&& is_failed(&d
->disk
))
4294 if (ord
& IMSM_ORD_REBUILD
)
4298 * if we skip some disks the array will be assmebled degraded;
4299 * reset resync start to avoid a dirty-degraded
4300 * situation when performing the intial sync
4302 * FIXME handle dirty degraded
4304 if ((skip
|| recovery_start
== 0) && !dev
->vol
.dirty
)
4305 this->resync_start
= MaxSector
;
4309 info_d
= calloc(1, sizeof(*info_d
));
4311 fprintf(stderr
, Name
": failed to allocate disk"
4312 " for volume %.16s\n", dev
->volume
);
4313 info_d
= this->devs
;
4315 struct mdinfo
*d
= info_d
->next
;
4324 info_d
->next
= this->devs
;
4325 this->devs
= info_d
;
4327 info_d
->disk
.number
= d
->index
;
4328 info_d
->disk
.major
= d
->major
;
4329 info_d
->disk
.minor
= d
->minor
;
4330 info_d
->disk
.raid_disk
= slot
;
4331 info_d
->recovery_start
= recovery_start
;
4333 if (info_d
->recovery_start
== MaxSector
)
4334 this->array
.working_disks
++;
4336 info_d
->events
= __le32_to_cpu(mpb
->generation_num
);
4337 info_d
->data_offset
= __le32_to_cpu(map
->pba_of_lba0
);
4338 info_d
->component_size
= __le32_to_cpu(map
->blocks_per_member
);
4340 /* now that the disk list is up-to-date fixup recovery_start */
4341 update_recovery_start(dev
, this);
4350 static int imsm_open_new(struct supertype
*c
, struct active_array
*a
,
4353 struct intel_super
*super
= c
->sb
;
4354 struct imsm_super
*mpb
= super
->anchor
;
4356 if (atoi(inst
) >= mpb
->num_raid_devs
) {
4357 fprintf(stderr
, "%s: subarry index %d, out of range\n",
4358 __func__
, atoi(inst
));
4362 dprintf("imsm: open_new %s\n", inst
);
4363 a
->info
.container_member
= atoi(inst
);
4367 static __u8
imsm_check_degraded(struct intel_super
*super
, struct imsm_dev
*dev
, int failed
)
4369 struct imsm_map
*map
= get_imsm_map(dev
, 0);
4372 return map
->map_state
== IMSM_T_STATE_UNINITIALIZED
?
4373 IMSM_T_STATE_UNINITIALIZED
: IMSM_T_STATE_NORMAL
;
4375 switch (get_imsm_raid_level(map
)) {
4377 return IMSM_T_STATE_FAILED
;
4380 if (failed
< map
->num_members
)
4381 return IMSM_T_STATE_DEGRADED
;
4383 return IMSM_T_STATE_FAILED
;
4388 * check to see if any mirrors have failed, otherwise we
4389 * are degraded. Even numbered slots are mirrored on
4393 /* gcc -Os complains that this is unused */
4394 int insync
= insync
;
4396 for (i
= 0; i
< map
->num_members
; i
++) {
4397 __u32 ord
= get_imsm_ord_tbl_ent(dev
, i
);
4398 int idx
= ord_to_idx(ord
);
4399 struct imsm_disk
*disk
;
4401 /* reset the potential in-sync count on even-numbered
4402 * slots. num_copies is always 2 for imsm raid10
4407 disk
= get_imsm_disk(super
, idx
);
4408 if (!disk
|| is_failed(disk
) || ord
& IMSM_ORD_REBUILD
)
4411 /* no in-sync disks left in this mirror the
4415 return IMSM_T_STATE_FAILED
;
4418 return IMSM_T_STATE_DEGRADED
;
4422 return IMSM_T_STATE_DEGRADED
;
4424 return IMSM_T_STATE_FAILED
;
4430 return map
->map_state
;
4433 static int imsm_count_failed(struct intel_super
*super
, struct imsm_dev
*dev
)
4437 struct imsm_disk
*disk
;
4438 struct imsm_map
*map
= get_imsm_map(dev
, 0);
4439 struct imsm_map
*prev
= get_imsm_map(dev
, dev
->vol
.migr_state
);
4443 /* at the beginning of migration we set IMSM_ORD_REBUILD on
4444 * disks that are being rebuilt. New failures are recorded to
4445 * map[0]. So we look through all the disks we started with and
4446 * see if any failures are still present, or if any new ones
4449 * FIXME add support for online capacity expansion and
4450 * raid-level-migration
4452 for (i
= 0; i
< prev
->num_members
; i
++) {
4453 ord
= __le32_to_cpu(prev
->disk_ord_tbl
[i
]);
4454 ord
|= __le32_to_cpu(map
->disk_ord_tbl
[i
]);
4455 idx
= ord_to_idx(ord
);
4457 disk
= get_imsm_disk(super
, idx
);
4458 if (!disk
|| is_failed(disk
) || ord
& IMSM_ORD_REBUILD
)
4465 static int is_resyncing(struct imsm_dev
*dev
)
4467 struct imsm_map
*migr_map
;
4469 if (!dev
->vol
.migr_state
)
4472 if (migr_type(dev
) == MIGR_INIT
||
4473 migr_type(dev
) == MIGR_REPAIR
)
4476 migr_map
= get_imsm_map(dev
, 1);
4478 if (migr_map
->map_state
== IMSM_T_STATE_NORMAL
)
4484 /* return true if we recorded new information */
4485 static int mark_failure(struct imsm_dev
*dev
, struct imsm_disk
*disk
, int idx
)
4489 struct imsm_map
*map
;
4491 /* new failures are always set in map[0] */
4492 map
= get_imsm_map(dev
, 0);
4494 slot
= get_imsm_disk_slot(map
, idx
);
4498 ord
= __le32_to_cpu(map
->disk_ord_tbl
[slot
]);
4499 if (is_failed(disk
) && (ord
& IMSM_ORD_REBUILD
))
4502 disk
->status
|= FAILED_DISK
;
4503 disk
->status
&= ~CONFIGURED_DISK
;
4504 set_imsm_ord_tbl_ent(map
, slot
, idx
| IMSM_ORD_REBUILD
);
4505 if (~map
->failed_disk_num
== 0)
4506 map
->failed_disk_num
= slot
;
4510 static void mark_missing(struct imsm_dev
*dev
, struct imsm_disk
*disk
, int idx
)
4512 mark_failure(dev
, disk
, idx
);
4514 if (disk
->scsi_id
== __cpu_to_le32(~(__u32
)0))
4517 disk
->scsi_id
= __cpu_to_le32(~(__u32
)0);
4518 memmove(&disk
->serial
[0], &disk
->serial
[1], MAX_RAID_SERIAL_LEN
- 1);
4521 static void handle_missing(struct intel_super
*super
, struct imsm_dev
*dev
)
4527 if (!super
->missing
)
4529 failed
= imsm_count_failed(super
, dev
);
4530 map_state
= imsm_check_degraded(super
, dev
, failed
);
4532 dprintf("imsm: mark missing\n");
4533 end_migration(dev
, map_state
);
4534 for (dl
= super
->missing
; dl
; dl
= dl
->next
)
4535 mark_missing(dev
, &dl
->disk
, dl
->index
);
4536 super
->updates_pending
++;
4539 /* Handle dirty -> clean transititions and resync. Degraded and rebuild
4540 * states are handled in imsm_set_disk() with one exception, when a
4541 * resync is stopped due to a new failure this routine will set the
4542 * 'degraded' state for the array.
4544 static int imsm_set_array_state(struct active_array
*a
, int consistent
)
4546 int inst
= a
->info
.container_member
;
4547 struct intel_super
*super
= a
->container
->sb
;
4548 struct imsm_dev
*dev
= get_imsm_dev(super
, inst
);
4549 struct imsm_map
*map
= get_imsm_map(dev
, 0);
4550 int failed
= imsm_count_failed(super
, dev
);
4551 __u8 map_state
= imsm_check_degraded(super
, dev
, failed
);
4552 __u32 blocks_per_unit
;
4554 /* before we activate this array handle any missing disks */
4555 if (consistent
== 2)
4556 handle_missing(super
, dev
);
4558 if (consistent
== 2 &&
4559 (!is_resync_complete(&a
->info
) ||
4560 map_state
!= IMSM_T_STATE_NORMAL
||
4561 dev
->vol
.migr_state
))
4564 if (is_resync_complete(&a
->info
)) {
4565 /* complete intialization / resync,
4566 * recovery and interrupted recovery is completed in
4569 if (is_resyncing(dev
)) {
4570 dprintf("imsm: mark resync done\n");
4571 end_migration(dev
, map_state
);
4572 super
->updates_pending
++;
4574 } else if (!is_resyncing(dev
) && !failed
) {
4575 /* mark the start of the init process if nothing is failed */
4576 dprintf("imsm: mark resync start\n");
4577 if (map
->map_state
== IMSM_T_STATE_UNINITIALIZED
)
4578 migrate(dev
, IMSM_T_STATE_NORMAL
, MIGR_INIT
);
4580 migrate(dev
, IMSM_T_STATE_NORMAL
, MIGR_REPAIR
);
4581 super
->updates_pending
++;
4584 /* check if we can update curr_migr_unit from resync_start, recovery_start */
4585 blocks_per_unit
= blocks_per_migr_unit(dev
);
4586 if (blocks_per_unit
&& failed
<= 1) {
4590 if (migr_type(dev
) == MIGR_REBUILD
)
4591 units
= min_recovery_start(&a
->info
) / blocks_per_unit
;
4593 units
= a
->info
.resync_start
/ blocks_per_unit
;
4596 /* check that we did not overflow 32-bits, and that
4597 * curr_migr_unit needs updating
4599 if (units32
== units
&&
4600 __le32_to_cpu(dev
->vol
.curr_migr_unit
) != units32
) {
4601 dprintf("imsm: mark checkpoint (%u)\n", units32
);
4602 dev
->vol
.curr_migr_unit
= __cpu_to_le32(units32
);
4603 super
->updates_pending
++;
4607 /* mark dirty / clean */
4608 if (dev
->vol
.dirty
!= !consistent
) {
4609 dprintf("imsm: mark '%s'\n", consistent
? "clean" : "dirty");
4614 super
->updates_pending
++;
4619 static void imsm_set_disk(struct active_array
*a
, int n
, int state
)
4621 int inst
= a
->info
.container_member
;
4622 struct intel_super
*super
= a
->container
->sb
;
4623 struct imsm_dev
*dev
= get_imsm_dev(super
, inst
);
4624 struct imsm_map
*map
= get_imsm_map(dev
, 0);
4625 struct imsm_disk
*disk
;
4630 if (n
> map
->num_members
)
4631 fprintf(stderr
, "imsm: set_disk %d out of range 0..%d\n",
4632 n
, map
->num_members
- 1);
4637 dprintf("imsm: set_disk %d:%x\n", n
, state
);
4639 ord
= get_imsm_ord_tbl_ent(dev
, n
);
4640 disk
= get_imsm_disk(super
, ord_to_idx(ord
));
4642 /* check for new failures */
4643 if (state
& DS_FAULTY
) {
4644 if (mark_failure(dev
, disk
, ord_to_idx(ord
)))
4645 super
->updates_pending
++;
4648 /* check if in_sync */
4649 if (state
& DS_INSYNC
&& ord
& IMSM_ORD_REBUILD
&& is_rebuilding(dev
)) {
4650 struct imsm_map
*migr_map
= get_imsm_map(dev
, 1);
4652 set_imsm_ord_tbl_ent(migr_map
, n
, ord_to_idx(ord
));
4653 super
->updates_pending
++;
4656 failed
= imsm_count_failed(super
, dev
);
4657 map_state
= imsm_check_degraded(super
, dev
, failed
);
4659 /* check if recovery complete, newly degraded, or failed */
4660 if (map_state
== IMSM_T_STATE_NORMAL
&& is_rebuilding(dev
)) {
4661 end_migration(dev
, map_state
);
4662 map
= get_imsm_map(dev
, 0);
4663 map
->failed_disk_num
= ~0;
4664 super
->updates_pending
++;
4665 } else if (map_state
== IMSM_T_STATE_DEGRADED
&&
4666 map
->map_state
!= map_state
&&
4667 !dev
->vol
.migr_state
) {
4668 dprintf("imsm: mark degraded\n");
4669 map
->map_state
= map_state
;
4670 super
->updates_pending
++;
4671 } else if (map_state
== IMSM_T_STATE_FAILED
&&
4672 map
->map_state
!= map_state
) {
4673 dprintf("imsm: mark failed\n");
4674 end_migration(dev
, map_state
);
4675 super
->updates_pending
++;
4679 static int store_imsm_mpb(int fd
, struct imsm_super
*mpb
)
4682 __u32 mpb_size
= __le32_to_cpu(mpb
->mpb_size
);
4683 unsigned long long dsize
;
4684 unsigned long long sectors
;
4686 get_dev_size(fd
, NULL
, &dsize
);
4688 if (mpb_size
> 512) {
4689 /* -1 to account for anchor */
4690 sectors
= mpb_sectors(mpb
) - 1;
4692 /* write the extended mpb to the sectors preceeding the anchor */
4693 if (lseek64(fd
, dsize
- (512 * (2 + sectors
)), SEEK_SET
) < 0)
4696 if (write(fd
, buf
+ 512, 512 * sectors
) != 512 * sectors
)
4700 /* first block is stored on second to last sector of the disk */
4701 if (lseek64(fd
, dsize
- (512 * 2), SEEK_SET
) < 0)
4704 if (write(fd
, buf
, 512) != 512)
4710 static void imsm_sync_metadata(struct supertype
*container
)
4712 struct intel_super
*super
= container
->sb
;
4714 if (!super
->updates_pending
)
4717 write_super_imsm(super
, 0);
4719 super
->updates_pending
= 0;
4722 static struct dl
*imsm_readd(struct intel_super
*super
, int idx
, struct active_array
*a
)
4724 struct imsm_dev
*dev
= get_imsm_dev(super
, a
->info
.container_member
);
4725 int i
= get_imsm_disk_idx(dev
, idx
);
4728 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
4732 if (dl
&& is_failed(&dl
->disk
))
4736 dprintf("%s: found %x:%x\n", __func__
, dl
->major
, dl
->minor
);
4741 static struct dl
*imsm_add_spare(struct intel_super
*super
, int slot
,
4742 struct active_array
*a
, int activate_new
)
4744 struct imsm_dev
*dev
= get_imsm_dev(super
, a
->info
.container_member
);
4745 int idx
= get_imsm_disk_idx(dev
, slot
);
4746 struct imsm_super
*mpb
= super
->anchor
;
4747 struct imsm_map
*map
;
4748 unsigned long long pos
;
4757 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
4758 /* If in this array, skip */
4759 for (d
= a
->info
.devs
; d
; d
= d
->next
)
4760 if (d
->state_fd
>= 0 &&
4761 d
->disk
.major
== dl
->major
&&
4762 d
->disk
.minor
== dl
->minor
) {
4763 dprintf("%x:%x already in array\n", dl
->major
, dl
->minor
);
4769 /* skip in use or failed drives */
4770 if (is_failed(&dl
->disk
) || idx
== dl
->index
||
4772 dprintf("%x:%x status (failed: %d index: %d)\n",
4773 dl
->major
, dl
->minor
, is_failed(&dl
->disk
), idx
);
4777 /* skip pure spares when we are looking for partially
4778 * assimilated drives
4780 if (dl
->index
== -1 && !activate_new
)
4783 /* Does this unused device have the requisite free space?
4784 * It needs to be able to cover all member volumes
4786 ex
= get_extents(super
, dl
);
4788 dprintf("cannot get extents\n");
4791 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
4792 dev
= get_imsm_dev(super
, i
);
4793 map
= get_imsm_map(dev
, 0);
4795 /* check if this disk is already a member of
4798 if (get_imsm_disk_slot(map
, dl
->index
) >= 0)
4804 array_start
= __le32_to_cpu(map
->pba_of_lba0
);
4805 array_end
= array_start
+
4806 __le32_to_cpu(map
->blocks_per_member
) - 1;
4809 /* check that we can start at pba_of_lba0 with
4810 * blocks_per_member of space
4812 if (array_start
>= pos
&& array_end
< ex
[j
].start
) {
4816 pos
= ex
[j
].start
+ ex
[j
].size
;
4818 } while (ex
[j
-1].size
);
4825 if (i
< mpb
->num_raid_devs
) {
4826 dprintf("%x:%x does not have %u to %u available\n",
4827 dl
->major
, dl
->minor
, array_start
, array_end
);
4837 static struct mdinfo
*imsm_activate_spare(struct active_array
*a
,
4838 struct metadata_update
**updates
)
4841 * Find a device with unused free space and use it to replace a
4842 * failed/vacant region in an array. We replace failed regions one a
4843 * array at a time. The result is that a new spare disk will be added
4844 * to the first failed array and after the monitor has finished
4845 * propagating failures the remainder will be consumed.
4847 * FIXME add a capability for mdmon to request spares from another
4851 struct intel_super
*super
= a
->container
->sb
;
4852 int inst
= a
->info
.container_member
;
4853 struct imsm_dev
*dev
= get_imsm_dev(super
, inst
);
4854 struct imsm_map
*map
= get_imsm_map(dev
, 0);
4855 int failed
= a
->info
.array
.raid_disks
;
4856 struct mdinfo
*rv
= NULL
;
4859 struct metadata_update
*mu
;
4861 struct imsm_update_activate_spare
*u
;
4865 for (d
= a
->info
.devs
; d
; d
= d
->next
) {
4866 if ((d
->curr_state
& DS_FAULTY
) &&
4868 /* wait for Removal to happen */
4870 if (d
->state_fd
>= 0)
4874 dprintf("imsm: activate spare: inst=%d failed=%d (%d) level=%d\n",
4875 inst
, failed
, a
->info
.array
.raid_disks
, a
->info
.array
.level
);
4876 if (imsm_check_degraded(super
, dev
, failed
) != IMSM_T_STATE_DEGRADED
)
4879 /* For each slot, if it is not working, find a spare */
4880 for (i
= 0; i
< a
->info
.array
.raid_disks
; i
++) {
4881 for (d
= a
->info
.devs
; d
; d
= d
->next
)
4882 if (d
->disk
.raid_disk
== i
)
4884 dprintf("found %d: %p %x\n", i
, d
, d
?d
->curr_state
:0);
4885 if (d
&& (d
->state_fd
>= 0))
4889 * OK, this device needs recovery. Try to re-add the
4890 * previous occupant of this slot, if this fails see if
4891 * we can continue the assimilation of a spare that was
4892 * partially assimilated, finally try to activate a new
4895 dl
= imsm_readd(super
, i
, a
);
4897 dl
= imsm_add_spare(super
, i
, a
, 0);
4899 dl
= imsm_add_spare(super
, i
, a
, 1);
4903 /* found a usable disk with enough space */
4904 di
= malloc(sizeof(*di
));
4907 memset(di
, 0, sizeof(*di
));
4909 /* dl->index will be -1 in the case we are activating a
4910 * pristine spare. imsm_process_update() will create a
4911 * new index in this case. Once a disk is found to be
4912 * failed in all member arrays it is kicked from the
4915 di
->disk
.number
= dl
->index
;
4917 /* (ab)use di->devs to store a pointer to the device
4920 di
->devs
= (struct mdinfo
*) dl
;
4922 di
->disk
.raid_disk
= i
;
4923 di
->disk
.major
= dl
->major
;
4924 di
->disk
.minor
= dl
->minor
;
4926 di
->recovery_start
= 0;
4927 di
->data_offset
= __le32_to_cpu(map
->pba_of_lba0
);
4928 di
->component_size
= a
->info
.component_size
;
4929 di
->container_member
= inst
;
4930 super
->random
= random32();
4934 dprintf("%x:%x to be %d at %llu\n", dl
->major
, dl
->minor
,
4935 i
, di
->data_offset
);
4941 /* No spares found */
4943 /* Now 'rv' has a list of devices to return.
4944 * Create a metadata_update record to update the
4945 * disk_ord_tbl for the array
4947 mu
= malloc(sizeof(*mu
));
4949 mu
->buf
= malloc(sizeof(struct imsm_update_activate_spare
) * num_spares
);
4950 if (mu
->buf
== NULL
) {
4957 struct mdinfo
*n
= rv
->next
;
4966 mu
->len
= sizeof(struct imsm_update_activate_spare
) * num_spares
;
4967 mu
->next
= *updates
;
4968 u
= (struct imsm_update_activate_spare
*) mu
->buf
;
4970 for (di
= rv
; di
; di
= di
->next
) {
4971 u
->type
= update_activate_spare
;
4972 u
->dl
= (struct dl
*) di
->devs
;
4974 u
->slot
= di
->disk
.raid_disk
;
4985 static int disks_overlap(struct intel_super
*super
, int idx
, struct imsm_update_create_array
*u
)
4987 struct imsm_dev
*dev
= get_imsm_dev(super
, idx
);
4988 struct imsm_map
*map
= get_imsm_map(dev
, 0);
4989 struct imsm_map
*new_map
= get_imsm_map(&u
->dev
, 0);
4990 struct disk_info
*inf
= get_disk_info(u
);
4991 struct imsm_disk
*disk
;
4995 for (i
= 0; i
< map
->num_members
; i
++) {
4996 disk
= get_imsm_disk(super
, get_imsm_disk_idx(dev
, i
));
4997 for (j
= 0; j
< new_map
->num_members
; j
++)
4998 if (serialcmp(disk
->serial
, inf
[j
].serial
) == 0)
5005 static void imsm_delete(struct intel_super
*super
, struct dl
**dlp
, int index
);
5007 static void imsm_process_update(struct supertype
*st
,
5008 struct metadata_update
*update
)
5011 * crack open the metadata_update envelope to find the update record
5012 * update can be one of:
5013 * update_activate_spare - a spare device has replaced a failed
5014 * device in an array, update the disk_ord_tbl. If this disk is
5015 * present in all member arrays then also clear the SPARE_DISK
5018 struct intel_super
*super
= st
->sb
;
5019 struct imsm_super
*mpb
;
5020 enum imsm_update_type type
= *(enum imsm_update_type
*) update
->buf
;
5022 /* update requires a larger buf but the allocation failed */
5023 if (super
->next_len
&& !super
->next_buf
) {
5024 super
->next_len
= 0;
5028 if (super
->next_buf
) {
5029 memcpy(super
->next_buf
, super
->buf
, super
->len
);
5031 super
->len
= super
->next_len
;
5032 super
->buf
= super
->next_buf
;
5034 super
->next_len
= 0;
5035 super
->next_buf
= NULL
;
5038 mpb
= super
->anchor
;
5041 case update_activate_spare
: {
5042 struct imsm_update_activate_spare
*u
= (void *) update
->buf
;
5043 struct imsm_dev
*dev
= get_imsm_dev(super
, u
->array
);
5044 struct imsm_map
*map
= get_imsm_map(dev
, 0);
5045 struct imsm_map
*migr_map
;
5046 struct active_array
*a
;
5047 struct imsm_disk
*disk
;
5052 int victim
= get_imsm_disk_idx(dev
, u
->slot
);
5055 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
5060 fprintf(stderr
, "error: imsm_activate_spare passed "
5061 "an unknown disk (index: %d)\n",
5066 super
->updates_pending
++;
5068 /* count failures (excluding rebuilds and the victim)
5069 * to determine map[0] state
5072 for (i
= 0; i
< map
->num_members
; i
++) {
5075 disk
= get_imsm_disk(super
, get_imsm_disk_idx(dev
, i
));
5076 if (!disk
|| is_failed(disk
))
5080 /* adding a pristine spare, assign a new index */
5081 if (dl
->index
< 0) {
5082 dl
->index
= super
->anchor
->num_disks
;
5083 super
->anchor
->num_disks
++;
5086 disk
->status
|= CONFIGURED_DISK
;
5087 disk
->status
&= ~SPARE_DISK
;
5090 to_state
= imsm_check_degraded(super
, dev
, failed
);
5091 map
->map_state
= IMSM_T_STATE_DEGRADED
;
5092 migrate(dev
, to_state
, MIGR_REBUILD
);
5093 migr_map
= get_imsm_map(dev
, 1);
5094 set_imsm_ord_tbl_ent(map
, u
->slot
, dl
->index
);
5095 set_imsm_ord_tbl_ent(migr_map
, u
->slot
, dl
->index
| IMSM_ORD_REBUILD
);
5097 /* update the family_num to mark a new container
5098 * generation, being careful to record the existing
5099 * family_num in orig_family_num to clean up after
5100 * earlier mdadm versions that neglected to set it.
5102 if (mpb
->orig_family_num
== 0)
5103 mpb
->orig_family_num
= mpb
->family_num
;
5104 mpb
->family_num
+= super
->random
;
5106 /* count arrays using the victim in the metadata */
5108 for (a
= st
->arrays
; a
; a
= a
->next
) {
5109 dev
= get_imsm_dev(super
, a
->info
.container_member
);
5110 map
= get_imsm_map(dev
, 0);
5112 if (get_imsm_disk_slot(map
, victim
) >= 0)
5116 /* delete the victim if it is no longer being
5122 /* We know that 'manager' isn't touching anything,
5123 * so it is safe to delete
5125 for (dlp
= &super
->disks
; *dlp
; dlp
= &(*dlp
)->next
)
5126 if ((*dlp
)->index
== victim
)
5129 /* victim may be on the missing list */
5131 for (dlp
= &super
->missing
; *dlp
; dlp
= &(*dlp
)->next
)
5132 if ((*dlp
)->index
== victim
)
5134 imsm_delete(super
, dlp
, victim
);
5138 case update_create_array
: {
5139 /* someone wants to create a new array, we need to be aware of
5140 * a few races/collisions:
5141 * 1/ 'Create' called by two separate instances of mdadm
5142 * 2/ 'Create' versus 'activate_spare': mdadm has chosen
5143 * devices that have since been assimilated via
5145 * In the event this update can not be carried out mdadm will
5146 * (FIX ME) notice that its update did not take hold.
5148 struct imsm_update_create_array
*u
= (void *) update
->buf
;
5149 struct intel_dev
*dv
;
5150 struct imsm_dev
*dev
;
5151 struct imsm_map
*map
, *new_map
;
5152 unsigned long long start
, end
;
5153 unsigned long long new_start
, new_end
;
5155 struct disk_info
*inf
;
5158 /* handle racing creates: first come first serve */
5159 if (u
->dev_idx
< mpb
->num_raid_devs
) {
5160 dprintf("%s: subarray %d already defined\n",
5161 __func__
, u
->dev_idx
);
5165 /* check update is next in sequence */
5166 if (u
->dev_idx
!= mpb
->num_raid_devs
) {
5167 dprintf("%s: can not create array %d expected index %d\n",
5168 __func__
, u
->dev_idx
, mpb
->num_raid_devs
);
5172 new_map
= get_imsm_map(&u
->dev
, 0);
5173 new_start
= __le32_to_cpu(new_map
->pba_of_lba0
);
5174 new_end
= new_start
+ __le32_to_cpu(new_map
->blocks_per_member
);
5175 inf
= get_disk_info(u
);
5177 /* handle activate_spare versus create race:
5178 * check to make sure that overlapping arrays do not include
5181 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
5182 dev
= get_imsm_dev(super
, i
);
5183 map
= get_imsm_map(dev
, 0);
5184 start
= __le32_to_cpu(map
->pba_of_lba0
);
5185 end
= start
+ __le32_to_cpu(map
->blocks_per_member
);
5186 if ((new_start
>= start
&& new_start
<= end
) ||
5187 (start
>= new_start
&& start
<= new_end
))
5192 if (disks_overlap(super
, i
, u
)) {
5193 dprintf("%s: arrays overlap\n", __func__
);
5198 /* check that prepare update was successful */
5199 if (!update
->space
) {
5200 dprintf("%s: prepare update failed\n", __func__
);
5204 /* check that all disks are still active before committing
5205 * changes. FIXME: could we instead handle this by creating a
5206 * degraded array? That's probably not what the user expects,
5207 * so better to drop this update on the floor.
5209 for (i
= 0; i
< new_map
->num_members
; i
++) {
5210 dl
= serial_to_dl(inf
[i
].serial
, super
);
5212 dprintf("%s: disk disappeared\n", __func__
);
5217 super
->updates_pending
++;
5219 /* convert spares to members and fixup ord_tbl */
5220 for (i
= 0; i
< new_map
->num_members
; i
++) {
5221 dl
= serial_to_dl(inf
[i
].serial
, super
);
5222 if (dl
->index
== -1) {
5223 dl
->index
= mpb
->num_disks
;
5225 dl
->disk
.status
|= CONFIGURED_DISK
;
5226 dl
->disk
.status
&= ~SPARE_DISK
;
5228 set_imsm_ord_tbl_ent(new_map
, i
, dl
->index
);
5233 update
->space
= NULL
;
5234 imsm_copy_dev(dev
, &u
->dev
);
5235 dv
->index
= u
->dev_idx
;
5236 dv
->next
= super
->devlist
;
5237 super
->devlist
= dv
;
5238 mpb
->num_raid_devs
++;
5240 imsm_update_version_info(super
);
5243 /* mdmon knows how to release update->space, but not
5244 * ((struct intel_dev *) update->space)->dev
5246 if (update
->space
) {
5252 case update_kill_array
: {
5253 struct imsm_update_kill_array
*u
= (void *) update
->buf
;
5254 int victim
= u
->dev_idx
;
5255 struct active_array
*a
;
5256 struct intel_dev
**dp
;
5257 struct imsm_dev
*dev
;
5259 /* sanity check that we are not affecting the uuid of
5260 * active arrays, or deleting an active array
5262 * FIXME when immutable ids are available, but note that
5263 * we'll also need to fixup the invalidated/active
5264 * subarray indexes in mdstat
5266 for (a
= st
->arrays
; a
; a
= a
->next
)
5267 if (a
->info
.container_member
>= victim
)
5269 /* by definition if mdmon is running at least one array
5270 * is active in the container, so checking
5271 * mpb->num_raid_devs is just extra paranoia
5273 dev
= get_imsm_dev(super
, victim
);
5274 if (a
|| !dev
|| mpb
->num_raid_devs
== 1) {
5275 dprintf("failed to delete subarray-%d\n", victim
);
5279 for (dp
= &super
->devlist
; *dp
;)
5280 if ((*dp
)->index
== super
->current_vol
) {
5283 if ((*dp
)->index
> victim
)
5287 mpb
->num_raid_devs
--;
5288 super
->updates_pending
++;
5291 case update_rename_array
: {
5292 struct imsm_update_rename_array
*u
= (void *) update
->buf
;
5293 char name
[MAX_RAID_SERIAL_LEN
+1];
5294 int target
= u
->dev_idx
;
5295 struct active_array
*a
;
5296 struct imsm_dev
*dev
;
5298 /* sanity check that we are not affecting the uuid of
5301 snprintf(name
, MAX_RAID_SERIAL_LEN
, "%s", (char *) u
->name
);
5302 name
[MAX_RAID_SERIAL_LEN
] = '\0';
5303 for (a
= st
->arrays
; a
; a
= a
->next
)
5304 if (a
->info
.container_member
== target
)
5306 dev
= get_imsm_dev(super
, u
->dev_idx
);
5307 if (a
|| !dev
|| !check_name(super
, name
, 1)) {
5308 dprintf("failed to rename subarray-%d\n", target
);
5312 snprintf((char *) dev
->volume
, MAX_RAID_SERIAL_LEN
, name
);
5313 super
->updates_pending
++;
5316 case update_add_disk
:
5318 /* we may be able to repair some arrays if disks are
5321 struct active_array
*a
;
5323 super
->updates_pending
++;
5324 for (a
= st
->arrays
; a
; a
= a
->next
)
5325 a
->check_degraded
= 1;
5327 /* add some spares to the metadata */
5328 while (super
->add
) {
5332 super
->add
= al
->next
;
5333 al
->next
= super
->disks
;
5335 dprintf("%s: added %x:%x\n",
5336 __func__
, al
->major
, al
->minor
);
5343 static void imsm_prepare_update(struct supertype
*st
,
5344 struct metadata_update
*update
)
5347 * Allocate space to hold new disk entries, raid-device entries or a new
5348 * mpb if necessary. The manager synchronously waits for updates to
5349 * complete in the monitor, so new mpb buffers allocated here can be
5350 * integrated by the monitor thread without worrying about live pointers
5351 * in the manager thread.
5353 enum imsm_update_type type
= *(enum imsm_update_type
*) update
->buf
;
5354 struct intel_super
*super
= st
->sb
;
5355 struct imsm_super
*mpb
= super
->anchor
;
5360 case update_create_array
: {
5361 struct imsm_update_create_array
*u
= (void *) update
->buf
;
5362 struct intel_dev
*dv
;
5363 struct imsm_dev
*dev
= &u
->dev
;
5364 struct imsm_map
*map
= get_imsm_map(dev
, 0);
5366 struct disk_info
*inf
;
5370 inf
= get_disk_info(u
);
5371 len
= sizeof_imsm_dev(dev
, 1);
5372 /* allocate a new super->devlist entry */
5373 dv
= malloc(sizeof(*dv
));
5375 dv
->dev
= malloc(len
);
5380 update
->space
= NULL
;
5384 /* count how many spares will be converted to members */
5385 for (i
= 0; i
< map
->num_members
; i
++) {
5386 dl
= serial_to_dl(inf
[i
].serial
, super
);
5388 /* hmm maybe it failed?, nothing we can do about
5393 if (count_memberships(dl
, super
) == 0)
5396 len
+= activate
* sizeof(struct imsm_disk
);
5403 /* check if we need a larger metadata buffer */
5404 if (super
->next_buf
)
5405 buf_len
= super
->next_len
;
5407 buf_len
= super
->len
;
5409 if (__le32_to_cpu(mpb
->mpb_size
) + len
> buf_len
) {
5410 /* ok we need a larger buf than what is currently allocated
5411 * if this allocation fails process_update will notice that
5412 * ->next_len is set and ->next_buf is NULL
5414 buf_len
= ROUND_UP(__le32_to_cpu(mpb
->mpb_size
) + len
, 512);
5415 if (super
->next_buf
)
5416 free(super
->next_buf
);
5418 super
->next_len
= buf_len
;
5419 if (posix_memalign(&super
->next_buf
, 512, buf_len
) == 0)
5420 memset(super
->next_buf
, 0, buf_len
);
5422 super
->next_buf
= NULL
;
5426 /* must be called while manager is quiesced */
5427 static void imsm_delete(struct intel_super
*super
, struct dl
**dlp
, int index
)
5429 struct imsm_super
*mpb
= super
->anchor
;
5431 struct imsm_dev
*dev
;
5432 struct imsm_map
*map
;
5433 int i
, j
, num_members
;
5436 dprintf("%s: deleting device[%d] from imsm_super\n",
5439 /* shift all indexes down one */
5440 for (iter
= super
->disks
; iter
; iter
= iter
->next
)
5441 if (iter
->index
> index
)
5443 for (iter
= super
->missing
; iter
; iter
= iter
->next
)
5444 if (iter
->index
> index
)
5447 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
5448 dev
= get_imsm_dev(super
, i
);
5449 map
= get_imsm_map(dev
, 0);
5450 num_members
= map
->num_members
;
5451 for (j
= 0; j
< num_members
; j
++) {
5452 /* update ord entries being careful not to propagate
5453 * ord-flags to the first map
5455 ord
= get_imsm_ord_tbl_ent(dev
, j
);
5457 if (ord_to_idx(ord
) <= index
)
5460 map
= get_imsm_map(dev
, 0);
5461 set_imsm_ord_tbl_ent(map
, j
, ord_to_idx(ord
- 1));
5462 map
= get_imsm_map(dev
, 1);
5464 set_imsm_ord_tbl_ent(map
, j
, ord
- 1);
5469 super
->updates_pending
++;
5471 struct dl
*dl
= *dlp
;
5473 *dlp
= (*dlp
)->next
;
5474 __free_imsm_disk(dl
);
5477 #endif /* MDASSEMBLE */
5479 struct superswitch super_imsm
= {
5481 .examine_super
= examine_super_imsm
,
5482 .brief_examine_super
= brief_examine_super_imsm
,
5483 .brief_examine_subarrays
= brief_examine_subarrays_imsm
,
5484 .export_examine_super
= export_examine_super_imsm
,
5485 .detail_super
= detail_super_imsm
,
5486 .brief_detail_super
= brief_detail_super_imsm
,
5487 .write_init_super
= write_init_super_imsm
,
5488 .validate_geometry
= validate_geometry_imsm
,
5489 .add_to_super
= add_to_super_imsm
,
5490 .detail_platform
= detail_platform_imsm
,
5491 .kill_subarray
= kill_subarray_imsm
,
5492 .update_subarray
= update_subarray_imsm
,
5494 .match_home
= match_home_imsm
,
5495 .uuid_from_super
= uuid_from_super_imsm
,
5496 .getinfo_super
= getinfo_super_imsm
,
5497 .update_super
= update_super_imsm
,
5499 .avail_size
= avail_size_imsm
,
5501 .compare_super
= compare_super_imsm
,
5503 .load_super
= load_super_imsm
,
5504 .init_super
= init_super_imsm
,
5505 .store_super
= store_super_imsm
,
5506 .free_super
= free_super_imsm
,
5507 .match_metadata_desc
= match_metadata_desc_imsm
,
5508 .container_content
= container_content_imsm
,
5509 .default_layout
= imsm_level_to_layout
,
5516 .open_new
= imsm_open_new
,
5517 .load_super
= load_super_imsm
,
5518 .set_array_state
= imsm_set_array_state
,
5519 .set_disk
= imsm_set_disk
,
5520 .sync_metadata
= imsm_sync_metadata
,
5521 .activate_spare
= imsm_activate_spare
,
5522 .process_update
= imsm_process_update
,
5523 .prepare_update
= imsm_prepare_update
,
5524 #endif /* MDASSEMBLE */