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
];
360 /* retrieve a disk from the parsed metadata */
361 static struct imsm_disk
*get_imsm_disk(struct intel_super
*super
, __u8 index
)
365 for (d
= super
->disks
; d
; d
= d
->next
)
366 if (d
->index
== index
)
372 /* generate a checksum directly from the anchor when the anchor is known to be
373 * up-to-date, currently only at load or write_super after coalescing
375 static __u32
__gen_imsm_checksum(struct imsm_super
*mpb
)
377 __u32 end
= mpb
->mpb_size
/ sizeof(end
);
378 __u32
*p
= (__u32
*) mpb
;
382 sum
+= __le32_to_cpu(*p
);
386 return sum
- __le32_to_cpu(mpb
->check_sum
);
389 static size_t sizeof_imsm_map(struct imsm_map
*map
)
391 return sizeof(struct imsm_map
) + sizeof(__u32
) * (map
->num_members
- 1);
394 struct imsm_map
*get_imsm_map(struct imsm_dev
*dev
, int second_map
)
396 struct imsm_map
*map
= &dev
->vol
.map
[0];
398 if (second_map
&& !dev
->vol
.migr_state
)
400 else if (second_map
) {
403 return ptr
+ sizeof_imsm_map(map
);
409 /* return the size of the device.
410 * migr_state increases the returned size if map[0] were to be duplicated
412 static size_t sizeof_imsm_dev(struct imsm_dev
*dev
, int migr_state
)
414 size_t size
= sizeof(*dev
) - sizeof(struct imsm_map
) +
415 sizeof_imsm_map(get_imsm_map(dev
, 0));
417 /* migrating means an additional map */
418 if (dev
->vol
.migr_state
)
419 size
+= sizeof_imsm_map(get_imsm_map(dev
, 1));
421 size
+= sizeof_imsm_map(get_imsm_map(dev
, 0));
427 /* retrieve disk serial number list from a metadata update */
428 static struct disk_info
*get_disk_info(struct imsm_update_create_array
*update
)
431 struct disk_info
*inf
;
433 inf
= u
+ sizeof(*update
) - sizeof(struct imsm_dev
) +
434 sizeof_imsm_dev(&update
->dev
, 0);
440 static struct imsm_dev
*__get_imsm_dev(struct imsm_super
*mpb
, __u8 index
)
446 if (index
>= mpb
->num_raid_devs
)
449 /* devices start after all disks */
450 offset
= ((void *) &mpb
->disk
[mpb
->num_disks
]) - _mpb
;
452 for (i
= 0; i
<= index
; i
++)
454 return _mpb
+ offset
;
456 offset
+= sizeof_imsm_dev(_mpb
+ offset
, 0);
461 static struct imsm_dev
*get_imsm_dev(struct intel_super
*super
, __u8 index
)
463 struct intel_dev
*dv
;
465 if (index
>= super
->anchor
->num_raid_devs
)
467 for (dv
= super
->devlist
; dv
; dv
= dv
->next
)
468 if (dv
->index
== index
)
473 static __u32
get_imsm_ord_tbl_ent(struct imsm_dev
*dev
, int slot
)
475 struct imsm_map
*map
;
477 if (dev
->vol
.migr_state
)
478 map
= get_imsm_map(dev
, 1);
480 map
= get_imsm_map(dev
, 0);
482 /* top byte identifies disk under rebuild */
483 return __le32_to_cpu(map
->disk_ord_tbl
[slot
]);
486 #define ord_to_idx(ord) (((ord) << 8) >> 8)
487 static __u32
get_imsm_disk_idx(struct imsm_dev
*dev
, int slot
)
489 __u32 ord
= get_imsm_ord_tbl_ent(dev
, slot
);
491 return ord_to_idx(ord
);
494 static void set_imsm_ord_tbl_ent(struct imsm_map
*map
, int slot
, __u32 ord
)
496 map
->disk_ord_tbl
[slot
] = __cpu_to_le32(ord
);
499 static int get_imsm_disk_slot(struct imsm_map
*map
, unsigned idx
)
504 for (slot
= 0; slot
< map
->num_members
; slot
++) {
505 ord
= __le32_to_cpu(map
->disk_ord_tbl
[slot
]);
506 if (ord_to_idx(ord
) == idx
)
513 static int get_imsm_raid_level(struct imsm_map
*map
)
515 if (map
->raid_level
== 1) {
516 if (map
->num_members
== 2)
522 return map
->raid_level
;
525 static int cmp_extent(const void *av
, const void *bv
)
527 const struct extent
*a
= av
;
528 const struct extent
*b
= bv
;
529 if (a
->start
< b
->start
)
531 if (a
->start
> b
->start
)
536 static int count_memberships(struct dl
*dl
, struct intel_super
*super
)
541 for (i
= 0; i
< super
->anchor
->num_raid_devs
; i
++) {
542 struct imsm_dev
*dev
= get_imsm_dev(super
, i
);
543 struct imsm_map
*map
= get_imsm_map(dev
, 0);
545 if (get_imsm_disk_slot(map
, dl
->index
) >= 0)
552 static struct extent
*get_extents(struct intel_super
*super
, struct dl
*dl
)
554 /* find a list of used extents on the given physical device */
555 struct extent
*rv
, *e
;
557 int memberships
= count_memberships(dl
, super
);
558 __u32 reservation
= MPB_SECTOR_CNT
+ IMSM_RESERVED_SECTORS
;
560 rv
= malloc(sizeof(struct extent
) * (memberships
+ 1));
565 for (i
= 0; i
< super
->anchor
->num_raid_devs
; i
++) {
566 struct imsm_dev
*dev
= get_imsm_dev(super
, i
);
567 struct imsm_map
*map
= get_imsm_map(dev
, 0);
569 if (get_imsm_disk_slot(map
, dl
->index
) >= 0) {
570 e
->start
= __le32_to_cpu(map
->pba_of_lba0
);
571 e
->size
= __le32_to_cpu(map
->blocks_per_member
);
575 qsort(rv
, memberships
, sizeof(*rv
), cmp_extent
);
577 /* determine the start of the metadata
578 * when no raid devices are defined use the default
579 * ...otherwise allow the metadata to truncate the value
580 * as is the case with older versions of imsm
583 struct extent
*last
= &rv
[memberships
- 1];
586 remainder
= __le32_to_cpu(dl
->disk
.total_blocks
) -
587 (last
->start
+ last
->size
);
588 /* round down to 1k block to satisfy precision of the kernel
592 /* make sure remainder is still sane */
593 if (remainder
< (unsigned)ROUND_UP(super
->len
, 512) >> 9)
594 remainder
= ROUND_UP(super
->len
, 512) >> 9;
595 if (reservation
> remainder
)
596 reservation
= remainder
;
598 e
->start
= __le32_to_cpu(dl
->disk
.total_blocks
) - reservation
;
603 /* try to determine how much space is reserved for metadata from
604 * the last get_extents() entry, otherwise fallback to the
607 static __u32
imsm_reserved_sectors(struct intel_super
*super
, struct dl
*dl
)
613 /* for spares just return a minimal reservation which will grow
614 * once the spare is picked up by an array
617 return MPB_SECTOR_CNT
;
619 e
= get_extents(super
, dl
);
621 return MPB_SECTOR_CNT
+ IMSM_RESERVED_SECTORS
;
623 /* scroll to last entry */
624 for (i
= 0; e
[i
].size
; i
++)
627 rv
= __le32_to_cpu(dl
->disk
.total_blocks
) - e
[i
].start
;
634 static int is_spare(struct imsm_disk
*disk
)
636 return (disk
->status
& SPARE_DISK
) == SPARE_DISK
;
639 static int is_configured(struct imsm_disk
*disk
)
641 return (disk
->status
& CONFIGURED_DISK
) == CONFIGURED_DISK
;
644 static int is_failed(struct imsm_disk
*disk
)
646 return (disk
->status
& FAILED_DISK
) == FAILED_DISK
;
650 static __u64
blocks_per_migr_unit(struct imsm_dev
*dev
);
652 static void print_imsm_dev(struct imsm_dev
*dev
, char *uuid
, int disk_idx
)
656 struct imsm_map
*map
= get_imsm_map(dev
, 0);
660 printf("[%.16s]:\n", dev
->volume
);
661 printf(" UUID : %s\n", uuid
);
662 printf(" RAID Level : %d\n", get_imsm_raid_level(map
));
663 printf(" Members : %d\n", map
->num_members
);
664 printf(" Slots : [");
665 for (i
= 0; i
< map
->num_members
; i
++) {
666 ord
= get_imsm_ord_tbl_ent(dev
, i
);
667 printf("%s", ord
& IMSM_ORD_REBUILD
? "_" : "U");
670 slot
= get_imsm_disk_slot(map
, disk_idx
);
672 ord
= get_imsm_ord_tbl_ent(dev
, slot
);
673 printf(" This Slot : %d%s\n", slot
,
674 ord
& IMSM_ORD_REBUILD
? " (out-of-sync)" : "");
676 printf(" This Slot : ?\n");
677 sz
= __le32_to_cpu(dev
->size_high
);
679 sz
+= __le32_to_cpu(dev
->size_low
);
680 printf(" Array Size : %llu%s\n", (unsigned long long)sz
,
681 human_size(sz
* 512));
682 sz
= __le32_to_cpu(map
->blocks_per_member
);
683 printf(" Per Dev Size : %llu%s\n", (unsigned long long)sz
,
684 human_size(sz
* 512));
685 printf(" Sector Offset : %u\n",
686 __le32_to_cpu(map
->pba_of_lba0
));
687 printf(" Num Stripes : %u\n",
688 __le32_to_cpu(map
->num_data_stripes
));
689 printf(" Chunk Size : %u KiB\n",
690 __le16_to_cpu(map
->blocks_per_strip
) / 2);
691 printf(" Reserved : %d\n", __le32_to_cpu(dev
->reserved_blocks
));
692 printf(" Migrate State : ");
693 if (dev
->vol
.migr_state
) {
694 if (migr_type(dev
) == MIGR_INIT
)
695 printf("initialize\n");
696 else if (migr_type(dev
) == MIGR_REBUILD
)
698 else if (migr_type(dev
) == MIGR_VERIFY
)
700 else if (migr_type(dev
) == MIGR_GEN_MIGR
)
701 printf("general migration\n");
702 else if (migr_type(dev
) == MIGR_STATE_CHANGE
)
703 printf("state change\n");
704 else if (migr_type(dev
) == MIGR_REPAIR
)
707 printf("<unknown:%d>\n", migr_type(dev
));
710 printf(" Map State : %s", map_state_str
[map
->map_state
]);
711 if (dev
->vol
.migr_state
) {
712 struct imsm_map
*map
= get_imsm_map(dev
, 1);
714 printf(" <-- %s", map_state_str
[map
->map_state
]);
715 printf("\n Checkpoint : %u (%llu)",
716 __le32_to_cpu(dev
->vol
.curr_migr_unit
),
717 (unsigned long long)blocks_per_migr_unit(dev
));
720 printf(" Dirty State : %s\n", dev
->vol
.dirty
? "dirty" : "clean");
723 static void print_imsm_disk(struct imsm_super
*mpb
, int index
, __u32 reserved
)
725 struct imsm_disk
*disk
= __get_imsm_disk(mpb
, index
);
726 char str
[MAX_RAID_SERIAL_LEN
+ 1];
729 if (index
< 0 || !disk
)
733 snprintf(str
, MAX_RAID_SERIAL_LEN
+ 1, "%s", disk
->serial
);
734 printf(" Disk%02d Serial : %s\n", index
, str
);
735 printf(" State :%s%s%s\n", is_spare(disk
) ? " spare" : "",
736 is_configured(disk
) ? " active" : "",
737 is_failed(disk
) ? " failed" : "");
738 printf(" Id : %08x\n", __le32_to_cpu(disk
->scsi_id
));
739 sz
= __le32_to_cpu(disk
->total_blocks
) - reserved
;
740 printf(" Usable Size : %llu%s\n", (unsigned long long)sz
,
741 human_size(sz
* 512));
744 static void getinfo_super_imsm(struct supertype
*st
, struct mdinfo
*info
);
746 static void examine_super_imsm(struct supertype
*st
, char *homehost
)
748 struct intel_super
*super
= st
->sb
;
749 struct imsm_super
*mpb
= super
->anchor
;
750 char str
[MAX_SIGNATURE_LENGTH
];
755 __u32 reserved
= imsm_reserved_sectors(super
, super
->disks
);
758 snprintf(str
, MPB_SIG_LEN
, "%s", mpb
->sig
);
759 printf(" Magic : %s\n", str
);
760 snprintf(str
, strlen(MPB_VERSION_RAID0
), "%s", get_imsm_version(mpb
));
761 printf(" Version : %s\n", get_imsm_version(mpb
));
762 printf(" Orig Family : %08x\n", __le32_to_cpu(mpb
->orig_family_num
));
763 printf(" Family : %08x\n", __le32_to_cpu(mpb
->family_num
));
764 printf(" Generation : %08x\n", __le32_to_cpu(mpb
->generation_num
));
765 getinfo_super_imsm(st
, &info
);
766 fname_from_uuid(st
, &info
, nbuf
, ':');
767 printf(" UUID : %s\n", nbuf
+ 5);
768 sum
= __le32_to_cpu(mpb
->check_sum
);
769 printf(" Checksum : %08x %s\n", sum
,
770 __gen_imsm_checksum(mpb
) == sum
? "correct" : "incorrect");
771 printf(" MPB Sectors : %d\n", mpb_sectors(mpb
));
772 printf(" Disks : %d\n", mpb
->num_disks
);
773 printf(" RAID Devices : %d\n", mpb
->num_raid_devs
);
774 print_imsm_disk(mpb
, super
->disks
->index
, reserved
);
775 if (super
->bbm_log
) {
776 struct bbm_log
*log
= super
->bbm_log
;
779 printf("Bad Block Management Log:\n");
780 printf(" Log Size : %d\n", __le32_to_cpu(mpb
->bbm_log_size
));
781 printf(" Signature : %x\n", __le32_to_cpu(log
->signature
));
782 printf(" Entry Count : %d\n", __le32_to_cpu(log
->entry_count
));
783 printf(" Spare Blocks : %d\n", __le32_to_cpu(log
->reserved_spare_block_count
));
784 printf(" First Spare : %llx\n",
785 (unsigned long long) __le64_to_cpu(log
->first_spare_lba
));
787 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
789 struct imsm_dev
*dev
= __get_imsm_dev(mpb
, i
);
791 super
->current_vol
= i
;
792 getinfo_super_imsm(st
, &info
);
793 fname_from_uuid(st
, &info
, nbuf
, ':');
794 print_imsm_dev(dev
, nbuf
+ 5, super
->disks
->index
);
796 for (i
= 0; i
< mpb
->num_disks
; i
++) {
797 if (i
== super
->disks
->index
)
799 print_imsm_disk(mpb
, i
, reserved
);
803 static void brief_examine_super_imsm(struct supertype
*st
, int verbose
)
805 /* We just write a generic IMSM ARRAY entry */
808 struct intel_super
*super
= st
->sb
;
810 if (!super
->anchor
->num_raid_devs
) {
811 printf("ARRAY metadata=imsm\n");
815 getinfo_super_imsm(st
, &info
);
816 fname_from_uuid(st
, &info
, nbuf
, ':');
817 printf("ARRAY metadata=imsm UUID=%s\n", nbuf
+ 5);
820 static void brief_examine_subarrays_imsm(struct supertype
*st
, int verbose
)
822 /* We just write a generic IMSM ARRAY entry */
826 struct intel_super
*super
= st
->sb
;
829 if (!super
->anchor
->num_raid_devs
)
832 getinfo_super_imsm(st
, &info
);
833 fname_from_uuid(st
, &info
, nbuf
, ':');
834 for (i
= 0; i
< super
->anchor
->num_raid_devs
; i
++) {
835 struct imsm_dev
*dev
= get_imsm_dev(super
, i
);
837 super
->current_vol
= i
;
838 getinfo_super_imsm(st
, &info
);
839 fname_from_uuid(st
, &info
, nbuf1
, ':');
840 printf("ARRAY /dev/md/%.16s container=%s member=%d UUID=%s\n",
841 dev
->volume
, nbuf
+ 5, i
, nbuf1
+ 5);
845 static void export_examine_super_imsm(struct supertype
*st
)
847 struct intel_super
*super
= st
->sb
;
848 struct imsm_super
*mpb
= super
->anchor
;
852 getinfo_super_imsm(st
, &info
);
853 fname_from_uuid(st
, &info
, nbuf
, ':');
854 printf("MD_METADATA=imsm\n");
855 printf("MD_LEVEL=container\n");
856 printf("MD_UUID=%s\n", nbuf
+5);
857 printf("MD_DEVICES=%u\n", mpb
->num_disks
);
860 static void detail_super_imsm(struct supertype
*st
, char *homehost
)
865 getinfo_super_imsm(st
, &info
);
866 fname_from_uuid(st
, &info
, nbuf
, ':');
867 printf("\n UUID : %s\n", nbuf
+ 5);
870 static void brief_detail_super_imsm(struct supertype
*st
)
874 getinfo_super_imsm(st
, &info
);
875 fname_from_uuid(st
, &info
, nbuf
, ':');
876 printf(" UUID=%s", nbuf
+ 5);
879 static int imsm_read_serial(int fd
, char *devname
, __u8
*serial
);
880 static void fd2devname(int fd
, char *name
);
882 static int imsm_enumerate_ports(const char *hba_path
, int port_count
, int host_base
, int verbose
)
884 /* dump an unsorted list of devices attached to ahci, as well as
885 * non-connected ports
887 int hba_len
= strlen(hba_path
) + 1;
892 unsigned long port_mask
= (1 << port_count
) - 1;
894 if (port_count
> (int)sizeof(port_mask
) * 8) {
896 fprintf(stderr
, Name
": port_count %d out of range\n", port_count
);
900 /* scroll through /sys/dev/block looking for devices attached to
903 dir
= opendir("/sys/dev/block");
904 for (ent
= dir
? readdir(dir
) : NULL
; ent
; ent
= readdir(dir
)) {
915 if (sscanf(ent
->d_name
, "%d:%d", &major
, &minor
) != 2)
917 path
= devt_to_devpath(makedev(major
, minor
));
920 if (!path_attached_to_hba(path
, hba_path
)) {
926 /* retrieve the scsi device type */
927 if (asprintf(&device
, "/sys/dev/block/%d:%d/device/xxxxxxx", major
, minor
) < 0) {
929 fprintf(stderr
, Name
": failed to allocate 'device'\n");
933 sprintf(device
, "/sys/dev/block/%d:%d/device/type", major
, minor
);
934 if (load_sys(device
, buf
) != 0) {
936 fprintf(stderr
, Name
": failed to read device type for %s\n",
942 type
= strtoul(buf
, NULL
, 10);
944 /* if it's not a disk print the vendor and model */
945 if (!(type
== 0 || type
== 7 || type
== 14)) {
948 sprintf(device
, "/sys/dev/block/%d:%d/device/vendor", major
, minor
);
949 if (load_sys(device
, buf
) == 0) {
950 strncpy(vendor
, buf
, sizeof(vendor
));
951 vendor
[sizeof(vendor
) - 1] = '\0';
952 c
= (char *) &vendor
[sizeof(vendor
) - 1];
953 while (isspace(*c
) || *c
== '\0')
957 sprintf(device
, "/sys/dev/block/%d:%d/device/model", major
, minor
);
958 if (load_sys(device
, buf
) == 0) {
959 strncpy(model
, buf
, sizeof(model
));
960 model
[sizeof(model
) - 1] = '\0';
961 c
= (char *) &model
[sizeof(model
) - 1];
962 while (isspace(*c
) || *c
== '\0')
966 if (vendor
[0] && model
[0])
967 sprintf(buf
, "%.64s %.64s", vendor
, model
);
969 switch (type
) { /* numbers from hald/linux/device.c */
970 case 1: sprintf(buf
, "tape"); break;
971 case 2: sprintf(buf
, "printer"); break;
972 case 3: sprintf(buf
, "processor"); break;
974 case 5: sprintf(buf
, "cdrom"); break;
975 case 6: sprintf(buf
, "scanner"); break;
976 case 8: sprintf(buf
, "media_changer"); break;
977 case 9: sprintf(buf
, "comm"); break;
978 case 12: sprintf(buf
, "raid"); break;
979 default: sprintf(buf
, "unknown");
985 /* chop device path to 'host%d' and calculate the port number */
986 c
= strchr(&path
[hba_len
], '/');
989 fprintf(stderr
, Name
": %s - invalid path name\n", path
+ hba_len
);
994 if (sscanf(&path
[hba_len
], "host%d", &port
) == 1)
998 *c
= '/'; /* repair the full string */
999 fprintf(stderr
, Name
": failed to determine port number for %s\n",
1006 /* mark this port as used */
1007 port_mask
&= ~(1 << port
);
1009 /* print out the device information */
1011 printf(" Port%d : - non-disk device (%s) -\n", port
, buf
);
1015 fd
= dev_open(ent
->d_name
, O_RDONLY
);
1017 printf(" Port%d : - disk info unavailable -\n", port
);
1019 fd2devname(fd
, buf
);
1020 printf(" Port%d : %s", port
, buf
);
1021 if (imsm_read_serial(fd
, NULL
, (__u8
*) buf
) == 0)
1022 printf(" (%s)\n", buf
);
1037 for (i
= 0; i
< port_count
; i
++)
1038 if (port_mask
& (1 << i
))
1039 printf(" Port%d : - no device attached -\n", i
);
1045 static int detail_platform_imsm(int verbose
, int enumerate_only
)
1047 /* There are two components to imsm platform support, the ahci SATA
1048 * controller and the option-rom. To find the SATA controller we
1049 * simply look in /sys/bus/pci/drivers/ahci to see if an ahci
1050 * controller with the Intel vendor id is present. This approach
1051 * allows mdadm to leverage the kernel's ahci detection logic, with the
1052 * caveat that if ahci.ko is not loaded mdadm will not be able to
1053 * detect platform raid capabilities. The option-rom resides in a
1054 * platform "Adapter ROM". We scan for its signature to retrieve the
1055 * platform capabilities. If raid support is disabled in the BIOS the
1056 * option-rom capability structure will not be available.
1058 const struct imsm_orom
*orom
;
1059 struct sys_dev
*list
, *hba
;
1062 const char *hba_path
;
1066 if (enumerate_only
) {
1067 if (check_env("IMSM_NO_PLATFORM") || find_imsm_orom())
1072 list
= find_driver_devices("pci", "ahci");
1073 for (hba
= list
; hba
; hba
= hba
->next
)
1074 if (devpath_to_vendor(hba
->path
) == 0x8086)
1079 fprintf(stderr
, Name
": unable to find active ahci controller\n");
1080 free_sys_dev(&list
);
1083 fprintf(stderr
, Name
": found Intel SATA AHCI Controller\n");
1084 hba_path
= hba
->path
;
1086 free_sys_dev(&list
);
1088 orom
= find_imsm_orom();
1091 fprintf(stderr
, Name
": imsm option-rom not found\n");
1095 printf(" Platform : Intel(R) Matrix Storage Manager\n");
1096 printf(" Version : %d.%d.%d.%d\n", orom
->major_ver
, orom
->minor_ver
,
1097 orom
->hotfix_ver
, orom
->build
);
1098 printf(" RAID Levels :%s%s%s%s%s\n",
1099 imsm_orom_has_raid0(orom
) ? " raid0" : "",
1100 imsm_orom_has_raid1(orom
) ? " raid1" : "",
1101 imsm_orom_has_raid1e(orom
) ? " raid1e" : "",
1102 imsm_orom_has_raid10(orom
) ? " raid10" : "",
1103 imsm_orom_has_raid5(orom
) ? " raid5" : "");
1104 printf(" Chunk Sizes :%s%s%s%s%s%s%s%s%s%s%s%s%s%s%s%s\n",
1105 imsm_orom_has_chunk(orom
, 2) ? " 2k" : "",
1106 imsm_orom_has_chunk(orom
, 4) ? " 4k" : "",
1107 imsm_orom_has_chunk(orom
, 8) ? " 8k" : "",
1108 imsm_orom_has_chunk(orom
, 16) ? " 16k" : "",
1109 imsm_orom_has_chunk(orom
, 32) ? " 32k" : "",
1110 imsm_orom_has_chunk(orom
, 64) ? " 64k" : "",
1111 imsm_orom_has_chunk(orom
, 128) ? " 128k" : "",
1112 imsm_orom_has_chunk(orom
, 256) ? " 256k" : "",
1113 imsm_orom_has_chunk(orom
, 512) ? " 512k" : "",
1114 imsm_orom_has_chunk(orom
, 1024*1) ? " 1M" : "",
1115 imsm_orom_has_chunk(orom
, 1024*2) ? " 2M" : "",
1116 imsm_orom_has_chunk(orom
, 1024*4) ? " 4M" : "",
1117 imsm_orom_has_chunk(orom
, 1024*8) ? " 8M" : "",
1118 imsm_orom_has_chunk(orom
, 1024*16) ? " 16M" : "",
1119 imsm_orom_has_chunk(orom
, 1024*32) ? " 32M" : "",
1120 imsm_orom_has_chunk(orom
, 1024*64) ? " 64M" : "");
1121 printf(" Max Disks : %d\n", orom
->tds
);
1122 printf(" Max Volumes : %d\n", orom
->vpa
);
1123 printf(" I/O Controller : %s\n", hba_path
);
1125 /* find the smallest scsi host number to determine a port number base */
1126 dir
= opendir(hba_path
);
1127 for (ent
= dir
? readdir(dir
) : NULL
; ent
; ent
= readdir(dir
)) {
1130 if (sscanf(ent
->d_name
, "host%d", &host
) != 1)
1132 if (port_count
== 0)
1134 else if (host
< host_base
)
1137 if (host
+ 1 > port_count
+ host_base
)
1138 port_count
= host
+ 1 - host_base
;
1144 if (!port_count
|| imsm_enumerate_ports(hba_path
, port_count
,
1145 host_base
, verbose
) != 0) {
1147 fprintf(stderr
, Name
": failed to enumerate ports\n");
1155 static int match_home_imsm(struct supertype
*st
, char *homehost
)
1157 /* the imsm metadata format does not specify any host
1158 * identification information. We return -1 since we can never
1159 * confirm nor deny whether a given array is "meant" for this
1160 * host. We rely on compare_super and the 'family_num' fields to
1161 * exclude member disks that do not belong, and we rely on
1162 * mdadm.conf to specify the arrays that should be assembled.
1163 * Auto-assembly may still pick up "foreign" arrays.
1169 static void uuid_from_super_imsm(struct supertype
*st
, int uuid
[4])
1171 /* The uuid returned here is used for:
1172 * uuid to put into bitmap file (Create, Grow)
1173 * uuid for backup header when saving critical section (Grow)
1174 * comparing uuids when re-adding a device into an array
1175 * In these cases the uuid required is that of the data-array,
1176 * not the device-set.
1177 * uuid to recognise same set when adding a missing device back
1178 * to an array. This is a uuid for the device-set.
1180 * For each of these we can make do with a truncated
1181 * or hashed uuid rather than the original, as long as
1183 * In each case the uuid required is that of the data-array,
1184 * not the device-set.
1186 /* imsm does not track uuid's so we synthesis one using sha1 on
1187 * - The signature (Which is constant for all imsm array, but no matter)
1188 * - the orig_family_num of the container
1189 * - the index number of the volume
1190 * - the 'serial' number of the volume.
1191 * Hopefully these are all constant.
1193 struct intel_super
*super
= st
->sb
;
1196 struct sha1_ctx ctx
;
1197 struct imsm_dev
*dev
= NULL
;
1200 /* some mdadm versions failed to set ->orig_family_num, in which
1201 * case fall back to ->family_num. orig_family_num will be
1202 * fixed up with the first metadata update.
1204 family_num
= super
->anchor
->orig_family_num
;
1205 if (family_num
== 0)
1206 family_num
= super
->anchor
->family_num
;
1207 sha1_init_ctx(&ctx
);
1208 sha1_process_bytes(super
->anchor
->sig
, MPB_SIG_LEN
, &ctx
);
1209 sha1_process_bytes(&family_num
, sizeof(__u32
), &ctx
);
1210 if (super
->current_vol
>= 0)
1211 dev
= get_imsm_dev(super
, super
->current_vol
);
1213 __u32 vol
= super
->current_vol
;
1214 sha1_process_bytes(&vol
, sizeof(vol
), &ctx
);
1215 sha1_process_bytes(dev
->volume
, MAX_RAID_SERIAL_LEN
, &ctx
);
1217 sha1_finish_ctx(&ctx
, buf
);
1218 memcpy(uuid
, buf
, 4*4);
1223 get_imsm_numerical_version(struct imsm_super
*mpb
, int *m
, int *p
)
1225 __u8
*v
= get_imsm_version(mpb
);
1226 __u8
*end
= mpb
->sig
+ MAX_SIGNATURE_LENGTH
;
1227 char major
[] = { 0, 0, 0 };
1228 char minor
[] = { 0 ,0, 0 };
1229 char patch
[] = { 0, 0, 0 };
1230 char *ver_parse
[] = { major
, minor
, patch
};
1234 while (*v
!= '\0' && v
< end
) {
1235 if (*v
!= '.' && j
< 2)
1236 ver_parse
[i
][j
++] = *v
;
1244 *m
= strtol(minor
, NULL
, 0);
1245 *p
= strtol(patch
, NULL
, 0);
1249 static __u32
migr_strip_blocks_resync(struct imsm_dev
*dev
)
1251 /* migr_strip_size when repairing or initializing parity */
1252 struct imsm_map
*map
= get_imsm_map(dev
, 0);
1253 __u32 chunk
= __le32_to_cpu(map
->blocks_per_strip
);
1255 switch (get_imsm_raid_level(map
)) {
1260 return 128*1024 >> 9;
1264 static __u32
migr_strip_blocks_rebuild(struct imsm_dev
*dev
)
1266 /* migr_strip_size when rebuilding a degraded disk, no idea why
1267 * this is different than migr_strip_size_resync(), but it's good
1270 struct imsm_map
*map
= get_imsm_map(dev
, 1);
1271 __u32 chunk
= __le32_to_cpu(map
->blocks_per_strip
);
1273 switch (get_imsm_raid_level(map
)) {
1276 if (map
->num_members
% map
->num_domains
== 0)
1277 return 128*1024 >> 9;
1281 return max((__u32
) 64*1024 >> 9, chunk
);
1283 return 128*1024 >> 9;
1287 static __u32
num_stripes_per_unit_resync(struct imsm_dev
*dev
)
1289 struct imsm_map
*lo
= get_imsm_map(dev
, 0);
1290 struct imsm_map
*hi
= get_imsm_map(dev
, 1);
1291 __u32 lo_chunk
= __le32_to_cpu(lo
->blocks_per_strip
);
1292 __u32 hi_chunk
= __le32_to_cpu(hi
->blocks_per_strip
);
1294 return max((__u32
) 1, hi_chunk
/ lo_chunk
);
1297 static __u32
num_stripes_per_unit_rebuild(struct imsm_dev
*dev
)
1299 struct imsm_map
*lo
= get_imsm_map(dev
, 0);
1300 int level
= get_imsm_raid_level(lo
);
1302 if (level
== 1 || level
== 10) {
1303 struct imsm_map
*hi
= get_imsm_map(dev
, 1);
1305 return hi
->num_domains
;
1307 return num_stripes_per_unit_resync(dev
);
1310 static __u8
imsm_num_data_members(struct imsm_dev
*dev
)
1312 /* named 'imsm_' because raid0, raid1 and raid10
1313 * counter-intuitively have the same number of data disks
1315 struct imsm_map
*map
= get_imsm_map(dev
, 0);
1317 switch (get_imsm_raid_level(map
)) {
1321 return map
->num_members
;
1323 return map
->num_members
- 1;
1325 dprintf("%s: unsupported raid level\n", __func__
);
1330 static __u32
parity_segment_depth(struct imsm_dev
*dev
)
1332 struct imsm_map
*map
= get_imsm_map(dev
, 0);
1333 __u32 chunk
= __le32_to_cpu(map
->blocks_per_strip
);
1335 switch(get_imsm_raid_level(map
)) {
1338 return chunk
* map
->num_domains
;
1340 return chunk
* map
->num_members
;
1346 static __u32
map_migr_block(struct imsm_dev
*dev
, __u32 block
)
1348 struct imsm_map
*map
= get_imsm_map(dev
, 1);
1349 __u32 chunk
= __le32_to_cpu(map
->blocks_per_strip
);
1350 __u32 strip
= block
/ chunk
;
1352 switch (get_imsm_raid_level(map
)) {
1355 __u32 vol_strip
= (strip
* map
->num_domains
) + 1;
1356 __u32 vol_stripe
= vol_strip
/ map
->num_members
;
1358 return vol_stripe
* chunk
+ block
% chunk
;
1360 __u32 stripe
= strip
/ (map
->num_members
- 1);
1362 return stripe
* chunk
+ block
% chunk
;
1369 static __u64
blocks_per_migr_unit(struct imsm_dev
*dev
)
1371 /* calculate the conversion factor between per member 'blocks'
1372 * (md/{resync,rebuild}_start) and imsm migration units, return
1373 * 0 for the 'not migrating' and 'unsupported migration' cases
1375 if (!dev
->vol
.migr_state
)
1378 switch (migr_type(dev
)) {
1382 struct imsm_map
*map
= get_imsm_map(dev
, 0);
1383 __u32 stripes_per_unit
;
1384 __u32 blocks_per_unit
;
1393 /* yes, this is really the translation of migr_units to
1394 * per-member blocks in the 'resync' case
1396 stripes_per_unit
= num_stripes_per_unit_resync(dev
);
1397 migr_chunk
= migr_strip_blocks_resync(dev
);
1398 disks
= imsm_num_data_members(dev
);
1399 blocks_per_unit
= stripes_per_unit
* migr_chunk
* disks
;
1400 stripe
= __le32_to_cpu(map
->blocks_per_strip
) * disks
;
1401 segment
= blocks_per_unit
/ stripe
;
1402 block_rel
= blocks_per_unit
- segment
* stripe
;
1403 parity_depth
= parity_segment_depth(dev
);
1404 block_map
= map_migr_block(dev
, block_rel
);
1405 return block_map
+ parity_depth
* segment
;
1407 case MIGR_REBUILD
: {
1408 __u32 stripes_per_unit
;
1411 stripes_per_unit
= num_stripes_per_unit_rebuild(dev
);
1412 migr_chunk
= migr_strip_blocks_rebuild(dev
);
1413 return migr_chunk
* stripes_per_unit
;
1416 case MIGR_STATE_CHANGE
:
1422 static int imsm_level_to_layout(int level
)
1430 return ALGORITHM_LEFT_ASYMMETRIC
;
1437 static void getinfo_super_imsm_volume(struct supertype
*st
, struct mdinfo
*info
)
1439 struct intel_super
*super
= st
->sb
;
1440 struct imsm_dev
*dev
= get_imsm_dev(super
, super
->current_vol
);
1441 struct imsm_map
*map
= get_imsm_map(dev
, 0);
1445 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
1446 if (dl
->raiddisk
== info
->disk
.raid_disk
)
1448 info
->container_member
= super
->current_vol
;
1449 info
->array
.raid_disks
= map
->num_members
;
1450 info
->array
.level
= get_imsm_raid_level(map
);
1451 info
->array
.layout
= imsm_level_to_layout(info
->array
.level
);
1452 info
->array
.md_minor
= -1;
1453 info
->array
.ctime
= 0;
1454 info
->array
.utime
= 0;
1455 info
->array
.chunk_size
= __le16_to_cpu(map
->blocks_per_strip
) << 9;
1456 info
->array
.state
= !dev
->vol
.dirty
;
1457 info
->custom_array_size
= __le32_to_cpu(dev
->size_high
);
1458 info
->custom_array_size
<<= 32;
1459 info
->custom_array_size
|= __le32_to_cpu(dev
->size_low
);
1461 info
->disk
.major
= 0;
1462 info
->disk
.minor
= 0;
1464 info
->disk
.major
= dl
->major
;
1465 info
->disk
.minor
= dl
->minor
;
1468 info
->data_offset
= __le32_to_cpu(map
->pba_of_lba0
);
1469 info
->component_size
= __le32_to_cpu(map
->blocks_per_member
);
1470 memset(info
->uuid
, 0, sizeof(info
->uuid
));
1471 info
->recovery_start
= MaxSector
;
1472 info
->reshape_active
= 0;
1474 if (map
->map_state
== IMSM_T_STATE_UNINITIALIZED
|| dev
->vol
.dirty
) {
1475 info
->resync_start
= 0;
1476 } else if (dev
->vol
.migr_state
) {
1477 switch (migr_type(dev
)) {
1480 __u64 blocks_per_unit
= blocks_per_migr_unit(dev
);
1481 __u64 units
= __le32_to_cpu(dev
->vol
.curr_migr_unit
);
1483 info
->resync_start
= blocks_per_unit
* units
;
1487 /* we could emulate the checkpointing of
1488 * 'sync_action=check' migrations, but for now
1489 * we just immediately complete them
1492 /* this is handled by container_content_imsm() */
1494 case MIGR_STATE_CHANGE
:
1495 /* FIXME handle other migrations */
1497 /* we are not dirty, so... */
1498 info
->resync_start
= MaxSector
;
1501 info
->resync_start
= MaxSector
;
1503 strncpy(info
->name
, (char *) dev
->volume
, MAX_RAID_SERIAL_LEN
);
1504 info
->name
[MAX_RAID_SERIAL_LEN
] = 0;
1506 info
->array
.major_version
= -1;
1507 info
->array
.minor_version
= -2;
1508 devname
= devnum2devname(st
->container_dev
);
1509 *info
->text_version
= '\0';
1511 sprintf(info
->text_version
, "/%s/%d", devname
, info
->container_member
);
1513 info
->safe_mode_delay
= 4000; /* 4 secs like the Matrix driver */
1514 uuid_from_super_imsm(st
, info
->uuid
);
1517 /* check the config file to see if we can return a real uuid for this spare */
1518 static void fixup_container_spare_uuid(struct mdinfo
*inf
)
1520 struct mddev_ident_s
*array_list
;
1522 if (inf
->array
.level
!= LEVEL_CONTAINER
||
1523 memcmp(inf
->uuid
, uuid_match_any
, sizeof(int[4])) != 0)
1526 array_list
= conf_get_ident(NULL
);
1528 for (; array_list
; array_list
= array_list
->next
) {
1529 if (array_list
->uuid_set
) {
1530 struct supertype
*_sst
; /* spare supertype */
1531 struct supertype
*_cst
; /* container supertype */
1533 _cst
= array_list
->st
;
1535 _sst
= _cst
->ss
->match_metadata_desc(inf
->text_version
);
1540 memcpy(inf
->uuid
, array_list
->uuid
, sizeof(int[4]));
1549 static __u8
imsm_check_degraded(struct intel_super
*super
, struct imsm_dev
*dev
, int failed
);
1550 static int imsm_count_failed(struct intel_super
*super
, struct imsm_dev
*dev
);
1552 static struct imsm_disk
*get_imsm_missing(struct intel_super
*super
, __u8 index
)
1556 for (d
= super
->missing
; d
; d
= d
->next
)
1557 if (d
->index
== index
)
1562 static void getinfo_super_imsm(struct supertype
*st
, struct mdinfo
*info
)
1564 struct intel_super
*super
= st
->sb
;
1565 struct imsm_disk
*disk
;
1567 if (super
->current_vol
>= 0) {
1568 getinfo_super_imsm_volume(st
, info
);
1572 /* Set raid_disks to zero so that Assemble will always pull in valid
1575 info
->array
.raid_disks
= 0;
1576 info
->array
.level
= LEVEL_CONTAINER
;
1577 info
->array
.layout
= 0;
1578 info
->array
.md_minor
= -1;
1579 info
->array
.ctime
= 0; /* N/A for imsm */
1580 info
->array
.utime
= 0;
1581 info
->array
.chunk_size
= 0;
1583 info
->disk
.major
= 0;
1584 info
->disk
.minor
= 0;
1585 info
->disk
.raid_disk
= -1;
1586 info
->reshape_active
= 0;
1587 info
->array
.major_version
= -1;
1588 info
->array
.minor_version
= -2;
1589 strcpy(info
->text_version
, "imsm");
1590 info
->safe_mode_delay
= 0;
1591 info
->disk
.number
= -1;
1592 info
->disk
.state
= 0;
1594 info
->recovery_start
= MaxSector
;
1596 /* do we have the all the insync disks that we expect? */
1597 if (st
->loaded_container
) {
1598 struct imsm_super
*mpb
= super
->anchor
;
1599 int max_enough
= -1, i
;
1601 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
1602 struct imsm_dev
*dev
= get_imsm_dev(super
, i
);
1603 int failed
, enough
, j
, missing
= 0;
1604 struct imsm_map
*map
;
1607 failed
= imsm_count_failed(super
, dev
);
1608 state
= imsm_check_degraded(super
, dev
, failed
);
1609 map
= get_imsm_map(dev
, dev
->vol
.migr_state
);
1611 /* any newly missing disks?
1612 * (catches single-degraded vs double-degraded)
1614 for (j
= 0; j
< map
->num_members
; j
++) {
1615 __u32 ord
= get_imsm_ord_tbl_ent(dev
, i
);
1616 __u32 idx
= ord_to_idx(ord
);
1618 if (!(ord
& IMSM_ORD_REBUILD
) &&
1619 get_imsm_missing(super
, idx
)) {
1625 if (state
== IMSM_T_STATE_FAILED
)
1627 else if (state
== IMSM_T_STATE_DEGRADED
&&
1628 (state
!= map
->map_state
|| missing
))
1630 else /* we're normal, or already degraded */
1633 /* in the missing/failed disk case check to see
1634 * if at least one array is runnable
1636 max_enough
= max(max_enough
, enough
);
1638 dprintf("%s: enough: %d\n", __func__
, max_enough
);
1639 info
->container_enough
= max_enough
;
1641 info
->container_enough
= -1;
1644 __u32 reserved
= imsm_reserved_sectors(super
, super
->disks
);
1646 disk
= &super
->disks
->disk
;
1647 info
->data_offset
= __le32_to_cpu(disk
->total_blocks
) - reserved
;
1648 info
->component_size
= reserved
;
1649 info
->disk
.state
= is_configured(disk
) ? (1 << MD_DISK_ACTIVE
) : 0;
1650 /* we don't change info->disk.raid_disk here because
1651 * this state will be finalized in mdmon after we have
1652 * found the 'most fresh' version of the metadata
1654 info
->disk
.state
|= is_failed(disk
) ? (1 << MD_DISK_FAULTY
) : 0;
1655 info
->disk
.state
|= is_spare(disk
) ? 0 : (1 << MD_DISK_SYNC
);
1658 /* only call uuid_from_super_imsm when this disk is part of a populated container,
1659 * ->compare_super may have updated the 'num_raid_devs' field for spares
1661 if (info
->disk
.state
& (1 << MD_DISK_SYNC
) || super
->anchor
->num_raid_devs
)
1662 uuid_from_super_imsm(st
, info
->uuid
);
1664 memcpy(info
->uuid
, uuid_match_any
, sizeof(int[4]));
1665 fixup_container_spare_uuid(info
);
1669 static int update_super_imsm(struct supertype
*st
, struct mdinfo
*info
,
1670 char *update
, char *devname
, int verbose
,
1671 int uuid_set
, char *homehost
)
1673 /* For 'assemble' and 'force' we need to return non-zero if any
1674 * change was made. For others, the return value is ignored.
1675 * Update options are:
1676 * force-one : This device looks a bit old but needs to be included,
1677 * update age info appropriately.
1678 * assemble: clear any 'faulty' flag to allow this device to
1680 * force-array: Array is degraded but being forced, mark it clean
1681 * if that will be needed to assemble it.
1683 * newdev: not used ????
1684 * grow: Array has gained a new device - this is currently for
1686 * resync: mark as dirty so a resync will happen.
1687 * name: update the name - preserving the homehost
1688 * uuid: Change the uuid of the array to match watch is given
1690 * Following are not relevant for this imsm:
1691 * sparc2.2 : update from old dodgey metadata
1692 * super-minor: change the preferred_minor number
1693 * summaries: update redundant counters.
1694 * homehost: update the recorded homehost
1695 * _reshape_progress: record new reshape_progress position.
1698 struct intel_super
*super
= st
->sb
;
1699 struct imsm_super
*mpb
;
1701 /* we can only update container info */
1702 if (!super
|| super
->current_vol
>= 0 || !super
->anchor
)
1705 mpb
= super
->anchor
;
1707 if (strcmp(update
, "uuid") == 0 && uuid_set
&& !info
->update_private
)
1709 Name
": '--uuid' not supported for imsm metadata\n");
1710 else if (strcmp(update
, "uuid") == 0 && uuid_set
&& info
->update_private
) {
1711 mpb
->orig_family_num
= *((__u32
*) info
->update_private
);
1713 } else if (strcmp(update
, "uuid") == 0) {
1714 __u32
*new_family
= malloc(sizeof(*new_family
));
1716 /* update orig_family_number with the incoming random
1717 * data, report the new effective uuid, and store the
1718 * new orig_family_num for future updates.
1721 memcpy(&mpb
->orig_family_num
, info
->uuid
, sizeof(__u32
));
1722 uuid_from_super_imsm(st
, info
->uuid
);
1723 *new_family
= mpb
->orig_family_num
;
1724 info
->update_private
= new_family
;
1727 } else if (strcmp(update
, "assemble") == 0)
1731 Name
": '--update=%s' not supported for imsm metadata\n",
1734 /* successful update? recompute checksum */
1736 mpb
->check_sum
= __le32_to_cpu(__gen_imsm_checksum(mpb
));
1741 static size_t disks_to_mpb_size(int disks
)
1745 size
= sizeof(struct imsm_super
);
1746 size
+= (disks
- 1) * sizeof(struct imsm_disk
);
1747 size
+= 2 * sizeof(struct imsm_dev
);
1748 /* up to 2 maps per raid device (-2 for imsm_maps in imsm_dev */
1749 size
+= (4 - 2) * sizeof(struct imsm_map
);
1750 /* 4 possible disk_ord_tbl's */
1751 size
+= 4 * (disks
- 1) * sizeof(__u32
);
1756 static __u64
avail_size_imsm(struct supertype
*st
, __u64 devsize
)
1758 if (devsize
< (MPB_SECTOR_CNT
+ IMSM_RESERVED_SECTORS
))
1761 return devsize
- (MPB_SECTOR_CNT
+ IMSM_RESERVED_SECTORS
);
1764 static void free_devlist(struct intel_super
*super
)
1766 struct intel_dev
*dv
;
1768 while (super
->devlist
) {
1769 dv
= super
->devlist
->next
;
1770 free(super
->devlist
->dev
);
1771 free(super
->devlist
);
1772 super
->devlist
= dv
;
1776 static void imsm_copy_dev(struct imsm_dev
*dest
, struct imsm_dev
*src
)
1778 memcpy(dest
, src
, sizeof_imsm_dev(src
, 0));
1781 static int compare_super_imsm(struct supertype
*st
, struct supertype
*tst
)
1785 * 0 same, or first was empty, and second was copied
1786 * 1 second had wrong number
1788 * 3 wrong other info
1790 struct intel_super
*first
= st
->sb
;
1791 struct intel_super
*sec
= tst
->sb
;
1799 /* if an anchor does not have num_raid_devs set then it is a free
1802 if (first
->anchor
->num_raid_devs
> 0 &&
1803 sec
->anchor
->num_raid_devs
> 0) {
1804 /* Determine if these disks might ever have been
1805 * related. Further disambiguation can only take place
1806 * in load_super_imsm_all
1808 __u32 first_family
= first
->anchor
->orig_family_num
;
1809 __u32 sec_family
= sec
->anchor
->orig_family_num
;
1811 if (memcmp(first
->anchor
->sig
, sec
->anchor
->sig
,
1812 MAX_SIGNATURE_LENGTH
) != 0)
1815 if (first_family
== 0)
1816 first_family
= first
->anchor
->family_num
;
1817 if (sec_family
== 0)
1818 sec_family
= sec
->anchor
->family_num
;
1820 if (first_family
!= sec_family
)
1826 /* if 'first' is a spare promote it to a populated mpb with sec's
1829 if (first
->anchor
->num_raid_devs
== 0 &&
1830 sec
->anchor
->num_raid_devs
> 0) {
1832 struct intel_dev
*dv
;
1833 struct imsm_dev
*dev
;
1835 /* we need to copy raid device info from sec if an allocation
1836 * fails here we don't associate the spare
1838 for (i
= 0; i
< sec
->anchor
->num_raid_devs
; i
++) {
1839 dv
= malloc(sizeof(*dv
));
1842 dev
= malloc(sizeof_imsm_dev(get_imsm_dev(sec
, i
), 1));
1849 dv
->next
= first
->devlist
;
1850 first
->devlist
= dv
;
1852 if (i
< sec
->anchor
->num_raid_devs
) {
1853 /* allocation failure */
1854 free_devlist(first
);
1855 fprintf(stderr
, "imsm: failed to associate spare\n");
1858 first
->anchor
->num_raid_devs
= sec
->anchor
->num_raid_devs
;
1859 first
->anchor
->orig_family_num
= sec
->anchor
->orig_family_num
;
1860 first
->anchor
->family_num
= sec
->anchor
->family_num
;
1861 memcpy(first
->anchor
->sig
, sec
->anchor
->sig
, MAX_SIGNATURE_LENGTH
);
1862 for (i
= 0; i
< sec
->anchor
->num_raid_devs
; i
++)
1863 imsm_copy_dev(get_imsm_dev(first
, i
), get_imsm_dev(sec
, i
));
1869 static void fd2devname(int fd
, char *name
)
1873 char dname
[PATH_MAX
];
1878 if (fstat(fd
, &st
) != 0)
1880 sprintf(path
, "/sys/dev/block/%d:%d",
1881 major(st
.st_rdev
), minor(st
.st_rdev
));
1883 rv
= readlink(path
, dname
, sizeof(dname
));
1888 nm
= strrchr(dname
, '/');
1890 snprintf(name
, MAX_RAID_SERIAL_LEN
, "/dev/%s", nm
);
1893 extern int scsi_get_serial(int fd
, void *buf
, size_t buf_len
);
1895 static int imsm_read_serial(int fd
, char *devname
,
1896 __u8 serial
[MAX_RAID_SERIAL_LEN
])
1898 unsigned char scsi_serial
[255];
1907 memset(scsi_serial
, 0, sizeof(scsi_serial
));
1909 rv
= scsi_get_serial(fd
, scsi_serial
, sizeof(scsi_serial
));
1911 if (rv
&& check_env("IMSM_DEVNAME_AS_SERIAL")) {
1912 memset(serial
, 0, MAX_RAID_SERIAL_LEN
);
1913 fd2devname(fd
, (char *) serial
);
1920 Name
": Failed to retrieve serial for %s\n",
1925 rsp_len
= scsi_serial
[3];
1929 Name
": Failed to retrieve serial for %s\n",
1933 rsp_buf
= (char *) &scsi_serial
[4];
1935 /* trim all whitespace and non-printable characters and convert
1938 for (i
= 0, dest
= rsp_buf
; i
< rsp_len
; i
++) {
1941 /* ':' is reserved for use in placeholder serial
1942 * numbers for missing disks
1950 len
= dest
- rsp_buf
;
1953 /* truncate leading characters */
1954 if (len
> MAX_RAID_SERIAL_LEN
) {
1955 dest
+= len
- MAX_RAID_SERIAL_LEN
;
1956 len
= MAX_RAID_SERIAL_LEN
;
1959 memset(serial
, 0, MAX_RAID_SERIAL_LEN
);
1960 memcpy(serial
, dest
, len
);
1965 static int serialcmp(__u8
*s1
, __u8
*s2
)
1967 return strncmp((char *) s1
, (char *) s2
, MAX_RAID_SERIAL_LEN
);
1970 static void serialcpy(__u8
*dest
, __u8
*src
)
1972 strncpy((char *) dest
, (char *) src
, MAX_RAID_SERIAL_LEN
);
1976 static struct dl
*serial_to_dl(__u8
*serial
, struct intel_super
*super
)
1980 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
1981 if (serialcmp(dl
->serial
, serial
) == 0)
1988 static struct imsm_disk
*
1989 __serial_to_disk(__u8
*serial
, struct imsm_super
*mpb
, int *idx
)
1993 for (i
= 0; i
< mpb
->num_disks
; i
++) {
1994 struct imsm_disk
*disk
= __get_imsm_disk(mpb
, i
);
1996 if (serialcmp(disk
->serial
, serial
) == 0) {
2007 load_imsm_disk(int fd
, struct intel_super
*super
, char *devname
, int keep_fd
)
2009 struct imsm_disk
*disk
;
2014 __u8 serial
[MAX_RAID_SERIAL_LEN
];
2016 rv
= imsm_read_serial(fd
, devname
, serial
);
2021 dl
= calloc(1, sizeof(*dl
));
2025 Name
": failed to allocate disk buffer for %s\n",
2031 dl
->major
= major(stb
.st_rdev
);
2032 dl
->minor
= minor(stb
.st_rdev
);
2033 dl
->next
= super
->disks
;
2034 dl
->fd
= keep_fd
? fd
: -1;
2035 assert(super
->disks
== NULL
);
2037 serialcpy(dl
->serial
, serial
);
2040 fd2devname(fd
, name
);
2042 dl
->devname
= strdup(devname
);
2044 dl
->devname
= strdup(name
);
2046 /* look up this disk's index in the current anchor */
2047 disk
= __serial_to_disk(dl
->serial
, super
->anchor
, &dl
->index
);
2050 /* only set index on disks that are a member of a
2051 * populated contianer, i.e. one with raid_devs
2053 if (is_failed(&dl
->disk
))
2055 else if (is_spare(&dl
->disk
))
2063 /* When migrating map0 contains the 'destination' state while map1
2064 * contains the current state. When not migrating map0 contains the
2065 * current state. This routine assumes that map[0].map_state is set to
2066 * the current array state before being called.
2068 * Migration is indicated by one of the following states
2069 * 1/ Idle (migr_state=0 map0state=normal||unitialized||degraded||failed)
2070 * 2/ Initialize (migr_state=1 migr_type=MIGR_INIT map0state=normal
2071 * map1state=unitialized)
2072 * 3/ Repair (Resync) (migr_state=1 migr_type=MIGR_REPAIR map0state=normal
2074 * 4/ Rebuild (migr_state=1 migr_type=MIGR_REBUILD map0state=normal
2075 * map1state=degraded)
2077 static void migrate(struct imsm_dev
*dev
, __u8 to_state
, int migr_type
)
2079 struct imsm_map
*dest
;
2080 struct imsm_map
*src
= get_imsm_map(dev
, 0);
2082 dev
->vol
.migr_state
= 1;
2083 set_migr_type(dev
, migr_type
);
2084 dev
->vol
.curr_migr_unit
= 0;
2085 dest
= get_imsm_map(dev
, 1);
2087 /* duplicate and then set the target end state in map[0] */
2088 memcpy(dest
, src
, sizeof_imsm_map(src
));
2089 if (migr_type
== MIGR_REBUILD
) {
2093 for (i
= 0; i
< src
->num_members
; i
++) {
2094 ord
= __le32_to_cpu(src
->disk_ord_tbl
[i
]);
2095 set_imsm_ord_tbl_ent(src
, i
, ord_to_idx(ord
));
2099 src
->map_state
= to_state
;
2102 static void end_migration(struct imsm_dev
*dev
, __u8 map_state
)
2104 struct imsm_map
*map
= get_imsm_map(dev
, 0);
2105 struct imsm_map
*prev
= get_imsm_map(dev
, dev
->vol
.migr_state
);
2108 /* merge any IMSM_ORD_REBUILD bits that were not successfully
2109 * completed in the last migration.
2111 * FIXME add support for online capacity expansion and
2112 * raid-level-migration
2114 for (i
= 0; i
< prev
->num_members
; i
++)
2115 map
->disk_ord_tbl
[i
] |= prev
->disk_ord_tbl
[i
];
2117 dev
->vol
.migr_state
= 0;
2118 dev
->vol
.curr_migr_unit
= 0;
2119 map
->map_state
= map_state
;
2123 static int parse_raid_devices(struct intel_super
*super
)
2126 struct imsm_dev
*dev_new
;
2127 size_t len
, len_migr
;
2128 size_t space_needed
= 0;
2129 struct imsm_super
*mpb
= super
->anchor
;
2131 for (i
= 0; i
< super
->anchor
->num_raid_devs
; i
++) {
2132 struct imsm_dev
*dev_iter
= __get_imsm_dev(super
->anchor
, i
);
2133 struct intel_dev
*dv
;
2135 len
= sizeof_imsm_dev(dev_iter
, 0);
2136 len_migr
= sizeof_imsm_dev(dev_iter
, 1);
2138 space_needed
+= len_migr
- len
;
2140 dv
= malloc(sizeof(*dv
));
2143 dev_new
= malloc(len_migr
);
2148 imsm_copy_dev(dev_new
, dev_iter
);
2151 dv
->next
= super
->devlist
;
2152 super
->devlist
= dv
;
2155 /* ensure that super->buf is large enough when all raid devices
2158 if (__le32_to_cpu(mpb
->mpb_size
) + space_needed
> super
->len
) {
2161 len
= ROUND_UP(__le32_to_cpu(mpb
->mpb_size
) + space_needed
, 512);
2162 if (posix_memalign(&buf
, 512, len
) != 0)
2165 memcpy(buf
, super
->buf
, super
->len
);
2166 memset(buf
+ super
->len
, 0, len
- super
->len
);
2175 /* retrieve a pointer to the bbm log which starts after all raid devices */
2176 struct bbm_log
*__get_imsm_bbm_log(struct imsm_super
*mpb
)
2180 if (__le32_to_cpu(mpb
->bbm_log_size
)) {
2182 ptr
+= mpb
->mpb_size
- __le32_to_cpu(mpb
->bbm_log_size
);
2188 static void __free_imsm(struct intel_super
*super
, int free_disks
);
2190 /* load_imsm_mpb - read matrix metadata
2191 * allocates super->mpb to be freed by free_super
2193 static int load_imsm_mpb(int fd
, struct intel_super
*super
, char *devname
)
2195 unsigned long long dsize
;
2196 unsigned long long sectors
;
2198 struct imsm_super
*anchor
;
2201 get_dev_size(fd
, NULL
, &dsize
);
2205 Name
": %s: device to small for imsm\n",
2210 if (lseek64(fd
, dsize
- (512 * 2), SEEK_SET
) < 0) {
2213 Name
": Cannot seek to anchor block on %s: %s\n",
2214 devname
, strerror(errno
));
2218 if (posix_memalign((void**)&anchor
, 512, 512) != 0) {
2221 Name
": Failed to allocate imsm anchor buffer"
2222 " on %s\n", devname
);
2225 if (read(fd
, anchor
, 512) != 512) {
2228 Name
": Cannot read anchor block on %s: %s\n",
2229 devname
, strerror(errno
));
2234 if (strncmp((char *) anchor
->sig
, MPB_SIGNATURE
, MPB_SIG_LEN
) != 0) {
2237 Name
": no IMSM anchor on %s\n", devname
);
2242 __free_imsm(super
, 0);
2243 super
->len
= ROUND_UP(anchor
->mpb_size
, 512);
2244 if (posix_memalign(&super
->buf
, 512, super
->len
) != 0) {
2247 Name
": unable to allocate %zu byte mpb buffer\n",
2252 memcpy(super
->buf
, anchor
, 512);
2254 sectors
= mpb_sectors(anchor
) - 1;
2257 check_sum
= __gen_imsm_checksum(super
->anchor
);
2258 if (check_sum
!= __le32_to_cpu(super
->anchor
->check_sum
)) {
2261 Name
": IMSM checksum %x != %x on %s\n",
2263 __le32_to_cpu(super
->anchor
->check_sum
),
2271 /* read the extended mpb */
2272 if (lseek64(fd
, dsize
- (512 * (2 + sectors
)), SEEK_SET
) < 0) {
2275 Name
": Cannot seek to extended mpb on %s: %s\n",
2276 devname
, strerror(errno
));
2280 if ((unsigned)read(fd
, super
->buf
+ 512, super
->len
- 512) != super
->len
- 512) {
2283 Name
": Cannot read extended mpb on %s: %s\n",
2284 devname
, strerror(errno
));
2288 check_sum
= __gen_imsm_checksum(super
->anchor
);
2289 if (check_sum
!= __le32_to_cpu(super
->anchor
->check_sum
)) {
2292 Name
": IMSM checksum %x != %x on %s\n",
2293 check_sum
, __le32_to_cpu(super
->anchor
->check_sum
),
2298 /* FIXME the BBM log is disk specific so we cannot use this global
2299 * buffer for all disks. Ok for now since we only look at the global
2300 * bbm_log_size parameter to gate assembly
2302 super
->bbm_log
= __get_imsm_bbm_log(super
->anchor
);
2308 load_and_parse_mpb(int fd
, struct intel_super
*super
, char *devname
, int keep_fd
)
2312 err
= load_imsm_mpb(fd
, super
, devname
);
2315 err
= load_imsm_disk(fd
, super
, devname
, keep_fd
);
2318 err
= parse_raid_devices(super
);
2323 static void __free_imsm_disk(struct dl
*d
)
2334 static void free_imsm_disks(struct intel_super
*super
)
2338 while (super
->disks
) {
2340 super
->disks
= d
->next
;
2341 __free_imsm_disk(d
);
2343 while (super
->missing
) {
2345 super
->missing
= d
->next
;
2346 __free_imsm_disk(d
);
2351 /* free all the pieces hanging off of a super pointer */
2352 static void __free_imsm(struct intel_super
*super
, int free_disks
)
2359 free_imsm_disks(super
);
2360 free_devlist(super
);
2362 free((void *) super
->hba
);
2367 static void free_imsm(struct intel_super
*super
)
2369 __free_imsm(super
, 1);
2373 static void free_super_imsm(struct supertype
*st
)
2375 struct intel_super
*super
= st
->sb
;
2384 static struct intel_super
*alloc_super(void)
2386 struct intel_super
*super
= malloc(sizeof(*super
));
2389 memset(super
, 0, sizeof(*super
));
2390 super
->current_vol
= -1;
2391 super
->create_offset
= ~((__u32
) 0);
2392 if (!check_env("IMSM_NO_PLATFORM"))
2393 super
->orom
= find_imsm_orom();
2394 if (super
->orom
&& !check_env("IMSM_TEST_OROM")) {
2395 struct sys_dev
*list
, *ent
;
2397 /* find the first intel ahci controller */
2398 list
= find_driver_devices("pci", "ahci");
2399 for (ent
= list
; ent
; ent
= ent
->next
)
2400 if (devpath_to_vendor(ent
->path
) == 0x8086)
2403 super
->hba
= ent
->path
;
2406 free_sys_dev(&list
);
2414 /* find_missing - helper routine for load_super_imsm_all that identifies
2415 * disks that have disappeared from the system. This routine relies on
2416 * the mpb being uptodate, which it is at load time.
2418 static int find_missing(struct intel_super
*super
)
2421 struct imsm_super
*mpb
= super
->anchor
;
2423 struct imsm_disk
*disk
;
2425 for (i
= 0; i
< mpb
->num_disks
; i
++) {
2426 disk
= __get_imsm_disk(mpb
, i
);
2427 dl
= serial_to_dl(disk
->serial
, super
);
2431 dl
= malloc(sizeof(*dl
));
2437 dl
->devname
= strdup("missing");
2439 serialcpy(dl
->serial
, disk
->serial
);
2442 dl
->next
= super
->missing
;
2443 super
->missing
= dl
;
2449 static struct intel_disk
*disk_list_get(__u8
*serial
, struct intel_disk
*disk_list
)
2451 struct intel_disk
*idisk
= disk_list
;
2454 if (serialcmp(idisk
->disk
.serial
, serial
) == 0)
2456 idisk
= idisk
->next
;
2462 static int __prep_thunderdome(struct intel_super
**table
, int tbl_size
,
2463 struct intel_super
*super
,
2464 struct intel_disk
**disk_list
)
2466 struct imsm_disk
*d
= &super
->disks
->disk
;
2467 struct imsm_super
*mpb
= super
->anchor
;
2470 for (i
= 0; i
< tbl_size
; i
++) {
2471 struct imsm_super
*tbl_mpb
= table
[i
]->anchor
;
2472 struct imsm_disk
*tbl_d
= &table
[i
]->disks
->disk
;
2474 if (tbl_mpb
->family_num
== mpb
->family_num
) {
2475 if (tbl_mpb
->check_sum
== mpb
->check_sum
) {
2476 dprintf("%s: mpb from %d:%d matches %d:%d\n",
2477 __func__
, super
->disks
->major
,
2478 super
->disks
->minor
,
2479 table
[i
]->disks
->major
,
2480 table
[i
]->disks
->minor
);
2484 if (((is_configured(d
) && !is_configured(tbl_d
)) ||
2485 is_configured(d
) == is_configured(tbl_d
)) &&
2486 tbl_mpb
->generation_num
< mpb
->generation_num
) {
2487 /* current version of the mpb is a
2488 * better candidate than the one in
2489 * super_table, but copy over "cross
2490 * generational" status
2492 struct intel_disk
*idisk
;
2494 dprintf("%s: mpb from %d:%d replaces %d:%d\n",
2495 __func__
, super
->disks
->major
,
2496 super
->disks
->minor
,
2497 table
[i
]->disks
->major
,
2498 table
[i
]->disks
->minor
);
2500 idisk
= disk_list_get(tbl_d
->serial
, *disk_list
);
2501 if (idisk
&& is_failed(&idisk
->disk
))
2502 tbl_d
->status
|= FAILED_DISK
;
2505 struct intel_disk
*idisk
;
2506 struct imsm_disk
*disk
;
2508 /* tbl_mpb is more up to date, but copy
2509 * over cross generational status before
2512 disk
= __serial_to_disk(d
->serial
, mpb
, NULL
);
2513 if (disk
&& is_failed(disk
))
2514 d
->status
|= FAILED_DISK
;
2516 idisk
= disk_list_get(d
->serial
, *disk_list
);
2519 if (disk
&& is_configured(disk
))
2520 idisk
->disk
.status
|= CONFIGURED_DISK
;
2523 dprintf("%s: mpb from %d:%d prefer %d:%d\n",
2524 __func__
, super
->disks
->major
,
2525 super
->disks
->minor
,
2526 table
[i
]->disks
->major
,
2527 table
[i
]->disks
->minor
);
2535 table
[tbl_size
++] = super
;
2539 /* update/extend the merged list of imsm_disk records */
2540 for (j
= 0; j
< mpb
->num_disks
; j
++) {
2541 struct imsm_disk
*disk
= __get_imsm_disk(mpb
, j
);
2542 struct intel_disk
*idisk
;
2544 idisk
= disk_list_get(disk
->serial
, *disk_list
);
2546 idisk
->disk
.status
|= disk
->status
;
2547 if (is_configured(&idisk
->disk
) ||
2548 is_failed(&idisk
->disk
))
2549 idisk
->disk
.status
&= ~(SPARE_DISK
);
2551 idisk
= calloc(1, sizeof(*idisk
));
2554 idisk
->owner
= IMSM_UNKNOWN_OWNER
;
2555 idisk
->disk
= *disk
;
2556 idisk
->next
= *disk_list
;
2560 if (serialcmp(idisk
->disk
.serial
, d
->serial
) == 0)
2567 static struct intel_super
*
2568 validate_members(struct intel_super
*super
, struct intel_disk
*disk_list
,
2571 struct imsm_super
*mpb
= super
->anchor
;
2575 for (i
= 0; i
< mpb
->num_disks
; i
++) {
2576 struct imsm_disk
*disk
= __get_imsm_disk(mpb
, i
);
2577 struct intel_disk
*idisk
;
2579 idisk
= disk_list_get(disk
->serial
, disk_list
);
2581 if (idisk
->owner
== owner
||
2582 idisk
->owner
== IMSM_UNKNOWN_OWNER
)
2585 dprintf("%s: '%.16s' owner %d != %d\n",
2586 __func__
, disk
->serial
, idisk
->owner
,
2589 dprintf("%s: unknown disk %x [%d]: %.16s\n",
2590 __func__
, __le32_to_cpu(mpb
->family_num
), i
,
2596 if (ok_count
== mpb
->num_disks
)
2601 static void show_conflicts(__u32 family_num
, struct intel_super
*super_list
)
2603 struct intel_super
*s
;
2605 for (s
= super_list
; s
; s
= s
->next
) {
2606 if (family_num
!= s
->anchor
->family_num
)
2608 fprintf(stderr
, "Conflict, offlining family %#x on '%s'\n",
2609 __le32_to_cpu(family_num
), s
->disks
->devname
);
2613 static struct intel_super
*
2614 imsm_thunderdome(struct intel_super
**super_list
, int len
)
2616 struct intel_super
*super_table
[len
];
2617 struct intel_disk
*disk_list
= NULL
;
2618 struct intel_super
*champion
, *spare
;
2619 struct intel_super
*s
, **del
;
2624 memset(super_table
, 0, sizeof(super_table
));
2625 for (s
= *super_list
; s
; s
= s
->next
)
2626 tbl_size
= __prep_thunderdome(super_table
, tbl_size
, s
, &disk_list
);
2628 for (i
= 0; i
< tbl_size
; i
++) {
2629 struct imsm_disk
*d
;
2630 struct intel_disk
*idisk
;
2631 struct imsm_super
*mpb
= super_table
[i
]->anchor
;
2634 d
= &s
->disks
->disk
;
2636 /* 'd' must appear in merged disk list for its
2637 * configuration to be valid
2639 idisk
= disk_list_get(d
->serial
, disk_list
);
2640 if (idisk
&& idisk
->owner
== i
)
2641 s
= validate_members(s
, disk_list
, i
);
2646 dprintf("%s: marking family: %#x from %d:%d offline\n",
2647 __func__
, mpb
->family_num
,
2648 super_table
[i
]->disks
->major
,
2649 super_table
[i
]->disks
->minor
);
2653 /* This is where the mdadm implementation differs from the Windows
2654 * driver which has no strict concept of a container. We can only
2655 * assemble one family from a container, so when returning a prodigal
2656 * array member to this system the code will not be able to disambiguate
2657 * the container contents that should be assembled ("foreign" versus
2658 * "local"). It requires user intervention to set the orig_family_num
2659 * to a new value to establish a new container. The Windows driver in
2660 * this situation fixes up the volume name in place and manages the
2661 * foreign array as an independent entity.
2666 for (i
= 0; i
< tbl_size
; i
++) {
2667 struct intel_super
*tbl_ent
= super_table
[i
];
2673 if (tbl_ent
->anchor
->num_raid_devs
== 0) {
2678 if (s
&& !is_spare
) {
2679 show_conflicts(tbl_ent
->anchor
->family_num
, *super_list
);
2681 } else if (!s
&& !is_spare
)
2694 fprintf(stderr
, "Chose family %#x on '%s', "
2695 "assemble conflicts to new container with '--update=uuid'\n",
2696 __le32_to_cpu(s
->anchor
->family_num
), s
->disks
->devname
);
2698 /* collect all dl's onto 'champion', and update them to
2699 * champion's version of the status
2701 for (s
= *super_list
; s
; s
= s
->next
) {
2702 struct imsm_super
*mpb
= champion
->anchor
;
2703 struct dl
*dl
= s
->disks
;
2708 for (i
= 0; i
< mpb
->num_disks
; i
++) {
2709 struct imsm_disk
*disk
;
2711 disk
= __serial_to_disk(dl
->serial
, mpb
, &dl
->index
);
2714 /* only set index on disks that are a member of
2715 * a populated contianer, i.e. one with
2718 if (is_failed(&dl
->disk
))
2720 else if (is_spare(&dl
->disk
))
2726 if (i
>= mpb
->num_disks
) {
2727 struct intel_disk
*idisk
;
2729 idisk
= disk_list_get(dl
->serial
, disk_list
);
2730 if (idisk
&& is_spare(&idisk
->disk
) &&
2731 !is_failed(&idisk
->disk
) && !is_configured(&idisk
->disk
))
2739 dl
->next
= champion
->disks
;
2740 champion
->disks
= dl
;
2744 /* delete 'champion' from super_list */
2745 for (del
= super_list
; *del
; ) {
2746 if (*del
== champion
) {
2747 *del
= (*del
)->next
;
2750 del
= &(*del
)->next
;
2752 champion
->next
= NULL
;
2756 struct intel_disk
*idisk
= disk_list
;
2758 disk_list
= disk_list
->next
;
2765 static int load_super_imsm_all(struct supertype
*st
, int fd
, void **sbp
,
2766 char *devname
, int keep_fd
)
2769 struct intel_super
*super_list
= NULL
;
2770 struct intel_super
*super
= NULL
;
2771 int devnum
= fd2devnum(fd
);
2777 /* check if 'fd' an opened container */
2778 sra
= sysfs_read(fd
, 0, GET_LEVEL
|GET_VERSION
|GET_DEVS
|GET_STATE
);
2782 if (sra
->array
.major_version
!= -1 ||
2783 sra
->array
.minor_version
!= -2 ||
2784 strcmp(sra
->text_version
, "imsm") != 0) {
2789 for (sd
= sra
->devs
, i
= 0; sd
; sd
= sd
->next
, i
++) {
2790 struct intel_super
*s
= alloc_super();
2797 s
->next
= super_list
;
2801 sprintf(nm
, "%d:%d", sd
->disk
.major
, sd
->disk
.minor
);
2802 dfd
= dev_open(nm
, keep_fd
? O_RDWR
: O_RDONLY
);
2806 err
= load_and_parse_mpb(dfd
, s
, NULL
, keep_fd
);
2808 /* retry the load if we might have raced against mdmon */
2809 if (err
== 3 && mdmon_running(devnum
))
2810 for (retry
= 0; retry
< 3; retry
++) {
2812 err
= load_and_parse_mpb(dfd
, s
, NULL
, keep_fd
);
2822 /* all mpbs enter, maybe one leaves */
2823 super
= imsm_thunderdome(&super_list
, i
);
2829 if (find_missing(super
) != 0) {
2835 if (st
->subarray
[0]) {
2840 val
= strtoul(st
->subarray
, &ep
, 10);
2846 if (val
< super
->anchor
->num_raid_devs
)
2847 super
->current_vol
= val
;
2856 while (super_list
) {
2857 struct intel_super
*s
= super_list
;
2859 super_list
= super_list
->next
;
2868 st
->container_dev
= devnum
;
2869 if (err
== 0 && st
->ss
== NULL
) {
2870 st
->ss
= &super_imsm
;
2871 st
->minor_version
= 0;
2872 st
->max_devs
= IMSM_MAX_DEVICES
;
2874 st
->loaded_container
= 1;
2880 static int load_super_imsm(struct supertype
*st
, int fd
, char *devname
)
2882 struct intel_super
*super
;
2886 if (load_super_imsm_all(st
, fd
, &st
->sb
, devname
, 1) == 0)
2890 if (test_partition(fd
))
2891 /* IMSM not allowed on partitions */
2894 free_super_imsm(st
);
2896 super
= alloc_super();
2899 Name
": malloc of %zu failed.\n",
2904 rv
= load_and_parse_mpb(fd
, super
, devname
, 0);
2909 Name
": Failed to load all information "
2910 "sections on %s\n", devname
);
2915 if (st
->subarray
[0]) {
2919 val
= strtoul(st
->subarray
, &ep
, 10);
2925 if (val
< super
->anchor
->num_raid_devs
)
2926 super
->current_vol
= val
;
2934 if (st
->ss
== NULL
) {
2935 st
->ss
= &super_imsm
;
2936 st
->minor_version
= 0;
2937 st
->max_devs
= IMSM_MAX_DEVICES
;
2939 st
->loaded_container
= 0;
2944 static __u16
info_to_blocks_per_strip(mdu_array_info_t
*info
)
2946 if (info
->level
== 1)
2948 return info
->chunk_size
>> 9;
2951 static __u32
info_to_num_data_stripes(mdu_array_info_t
*info
, int num_domains
)
2955 num_stripes
= (info
->size
* 2) / info_to_blocks_per_strip(info
);
2956 num_stripes
/= num_domains
;
2961 static __u32
info_to_blocks_per_member(mdu_array_info_t
*info
)
2963 if (info
->level
== 1)
2964 return info
->size
* 2;
2966 return (info
->size
* 2) & ~(info_to_blocks_per_strip(info
) - 1);
2969 static void imsm_update_version_info(struct intel_super
*super
)
2971 /* update the version and attributes */
2972 struct imsm_super
*mpb
= super
->anchor
;
2974 struct imsm_dev
*dev
;
2975 struct imsm_map
*map
;
2978 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
2979 dev
= get_imsm_dev(super
, i
);
2980 map
= get_imsm_map(dev
, 0);
2981 if (__le32_to_cpu(dev
->size_high
) > 0)
2982 mpb
->attributes
|= MPB_ATTRIB_2TB
;
2984 /* FIXME detect when an array spans a port multiplier */
2986 mpb
->attributes
|= MPB_ATTRIB_PM
;
2989 if (mpb
->num_raid_devs
> 1 ||
2990 mpb
->attributes
!= MPB_ATTRIB_CHECKSUM_VERIFY
) {
2991 version
= MPB_VERSION_ATTRIBS
;
2992 switch (get_imsm_raid_level(map
)) {
2993 case 0: mpb
->attributes
|= MPB_ATTRIB_RAID0
; break;
2994 case 1: mpb
->attributes
|= MPB_ATTRIB_RAID1
; break;
2995 case 10: mpb
->attributes
|= MPB_ATTRIB_RAID10
; break;
2996 case 5: mpb
->attributes
|= MPB_ATTRIB_RAID5
; break;
2999 if (map
->num_members
>= 5)
3000 version
= MPB_VERSION_5OR6_DISK_ARRAY
;
3001 else if (dev
->status
== DEV_CLONE_N_GO
)
3002 version
= MPB_VERSION_CNG
;
3003 else if (get_imsm_raid_level(map
) == 5)
3004 version
= MPB_VERSION_RAID5
;
3005 else if (map
->num_members
>= 3)
3006 version
= MPB_VERSION_3OR4_DISK_ARRAY
;
3007 else if (get_imsm_raid_level(map
) == 1)
3008 version
= MPB_VERSION_RAID1
;
3010 version
= MPB_VERSION_RAID0
;
3012 strcpy(((char *) mpb
->sig
) + strlen(MPB_SIGNATURE
), version
);
3016 static int check_name(struct intel_super
*super
, char *name
, int quiet
)
3018 struct imsm_super
*mpb
= super
->anchor
;
3019 char *reason
= NULL
;
3022 if (strlen(name
) > MAX_RAID_SERIAL_LEN
)
3023 reason
= "must be 16 characters or less";
3025 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
3026 struct imsm_dev
*dev
= get_imsm_dev(super
, i
);
3028 if (strncmp((char *) dev
->volume
, name
, MAX_RAID_SERIAL_LEN
) == 0) {
3029 reason
= "already exists";
3034 if (reason
&& !quiet
)
3035 fprintf(stderr
, Name
": imsm volume name %s\n", reason
);
3040 static int init_super_imsm_volume(struct supertype
*st
, mdu_array_info_t
*info
,
3041 unsigned long long size
, char *name
,
3042 char *homehost
, int *uuid
)
3044 /* We are creating a volume inside a pre-existing container.
3045 * so st->sb is already set.
3047 struct intel_super
*super
= st
->sb
;
3048 struct imsm_super
*mpb
= super
->anchor
;
3049 struct intel_dev
*dv
;
3050 struct imsm_dev
*dev
;
3051 struct imsm_vol
*vol
;
3052 struct imsm_map
*map
;
3053 int idx
= mpb
->num_raid_devs
;
3055 unsigned long long array_blocks
;
3056 size_t size_old
, size_new
;
3057 __u32 num_data_stripes
;
3059 if (super
->orom
&& mpb
->num_raid_devs
>= super
->orom
->vpa
) {
3060 fprintf(stderr
, Name
": This imsm-container already has the "
3061 "maximum of %d volumes\n", super
->orom
->vpa
);
3065 /* ensure the mpb is large enough for the new data */
3066 size_old
= __le32_to_cpu(mpb
->mpb_size
);
3067 size_new
= disks_to_mpb_size(info
->nr_disks
);
3068 if (size_new
> size_old
) {
3070 size_t size_round
= ROUND_UP(size_new
, 512);
3072 if (posix_memalign(&mpb_new
, 512, size_round
) != 0) {
3073 fprintf(stderr
, Name
": could not allocate new mpb\n");
3076 memcpy(mpb_new
, mpb
, size_old
);
3079 super
->anchor
= mpb_new
;
3080 mpb
->mpb_size
= __cpu_to_le32(size_new
);
3081 memset(mpb_new
+ size_old
, 0, size_round
- size_old
);
3083 super
->current_vol
= idx
;
3084 /* when creating the first raid device in this container set num_disks
3085 * to zero, i.e. delete this spare and add raid member devices in
3086 * add_to_super_imsm_volume()
3088 if (super
->current_vol
== 0)
3091 if (!check_name(super
, name
, 0))
3093 sprintf(st
->subarray
, "%d", idx
);
3094 dv
= malloc(sizeof(*dv
));
3096 fprintf(stderr
, Name
": failed to allocate device list entry\n");
3099 dev
= malloc(sizeof(*dev
) + sizeof(__u32
) * (info
->raid_disks
- 1));
3102 fprintf(stderr
, Name
": could not allocate raid device\n");
3105 strncpy((char *) dev
->volume
, name
, MAX_RAID_SERIAL_LEN
);
3106 if (info
->level
== 1)
3107 array_blocks
= info_to_blocks_per_member(info
);
3109 array_blocks
= calc_array_size(info
->level
, info
->raid_disks
,
3110 info
->layout
, info
->chunk_size
,
3112 /* round array size down to closest MB */
3113 array_blocks
= (array_blocks
>> SECT_PER_MB_SHIFT
) << SECT_PER_MB_SHIFT
;
3115 dev
->size_low
= __cpu_to_le32((__u32
) array_blocks
);
3116 dev
->size_high
= __cpu_to_le32((__u32
) (array_blocks
>> 32));
3117 dev
->status
= __cpu_to_le32(0);
3118 dev
->reserved_blocks
= __cpu_to_le32(0);
3120 vol
->migr_state
= 0;
3121 set_migr_type(dev
, MIGR_INIT
);
3123 vol
->curr_migr_unit
= 0;
3124 map
= get_imsm_map(dev
, 0);
3125 map
->pba_of_lba0
= __cpu_to_le32(super
->create_offset
);
3126 map
->blocks_per_member
= __cpu_to_le32(info_to_blocks_per_member(info
));
3127 map
->blocks_per_strip
= __cpu_to_le16(info_to_blocks_per_strip(info
));
3128 map
->failed_disk_num
= ~0;
3129 map
->map_state
= info
->level
? IMSM_T_STATE_UNINITIALIZED
:
3130 IMSM_T_STATE_NORMAL
;
3133 if (info
->level
== 1 && info
->raid_disks
> 2) {
3136 fprintf(stderr
, Name
": imsm does not support more than 2 disks"
3137 "in a raid1 volume\n");
3141 map
->raid_level
= info
->level
;
3142 if (info
->level
== 10) {
3143 map
->raid_level
= 1;
3144 map
->num_domains
= info
->raid_disks
/ 2;
3145 } else if (info
->level
== 1)
3146 map
->num_domains
= info
->raid_disks
;
3148 map
->num_domains
= 1;
3150 num_data_stripes
= info_to_num_data_stripes(info
, map
->num_domains
);
3151 map
->num_data_stripes
= __cpu_to_le32(num_data_stripes
);
3153 map
->num_members
= info
->raid_disks
;
3154 for (i
= 0; i
< map
->num_members
; i
++) {
3155 /* initialized in add_to_super */
3156 set_imsm_ord_tbl_ent(map
, i
, IMSM_ORD_REBUILD
);
3158 mpb
->num_raid_devs
++;
3161 dv
->index
= super
->current_vol
;
3162 dv
->next
= super
->devlist
;
3163 super
->devlist
= dv
;
3165 imsm_update_version_info(super
);
3170 static int init_super_imsm(struct supertype
*st
, mdu_array_info_t
*info
,
3171 unsigned long long size
, char *name
,
3172 char *homehost
, int *uuid
)
3174 /* This is primarily called by Create when creating a new array.
3175 * We will then get add_to_super called for each component, and then
3176 * write_init_super called to write it out to each device.
3177 * For IMSM, Create can create on fresh devices or on a pre-existing
3179 * To create on a pre-existing array a different method will be called.
3180 * This one is just for fresh drives.
3182 struct intel_super
*super
;
3183 struct imsm_super
*mpb
;
3188 return init_super_imsm_volume(st
, info
, size
, name
, homehost
, uuid
);
3191 mpb_size
= disks_to_mpb_size(info
->nr_disks
);
3195 super
= alloc_super();
3196 if (super
&& posix_memalign(&super
->buf
, 512, mpb_size
) != 0) {
3201 fprintf(stderr
, Name
3202 ": %s could not allocate superblock\n", __func__
);
3205 memset(super
->buf
, 0, mpb_size
);
3207 mpb
->mpb_size
= __cpu_to_le32(mpb_size
);
3211 /* zeroing superblock */
3215 mpb
->attributes
= MPB_ATTRIB_CHECKSUM_VERIFY
;
3217 version
= (char *) mpb
->sig
;
3218 strcpy(version
, MPB_SIGNATURE
);
3219 version
+= strlen(MPB_SIGNATURE
);
3220 strcpy(version
, MPB_VERSION_RAID0
);
3226 static int add_to_super_imsm_volume(struct supertype
*st
, mdu_disk_info_t
*dk
,
3227 int fd
, char *devname
)
3229 struct intel_super
*super
= st
->sb
;
3230 struct imsm_super
*mpb
= super
->anchor
;
3232 struct imsm_dev
*dev
;
3233 struct imsm_map
*map
;
3236 dev
= get_imsm_dev(super
, super
->current_vol
);
3237 map
= get_imsm_map(dev
, 0);
3239 if (! (dk
->state
& (1<<MD_DISK_SYNC
))) {
3240 fprintf(stderr
, Name
": %s: Cannot add spare devices to IMSM volume\n",
3246 /* we're doing autolayout so grab the pre-marked (in
3247 * validate_geometry) raid_disk
3249 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
3250 if (dl
->raiddisk
== dk
->raid_disk
)
3253 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
3254 if (dl
->major
== dk
->major
&&
3255 dl
->minor
== dk
->minor
)
3260 fprintf(stderr
, Name
": %s is not a member of the same container\n", devname
);
3264 /* add a pristine spare to the metadata */
3265 if (dl
->index
< 0) {
3266 dl
->index
= super
->anchor
->num_disks
;
3267 super
->anchor
->num_disks
++;
3269 /* Check the device has not already been added */
3270 slot
= get_imsm_disk_slot(map
, dl
->index
);
3272 (get_imsm_ord_tbl_ent(dev
, slot
) & IMSM_ORD_REBUILD
) == 0) {
3273 fprintf(stderr
, Name
": %s has been included in this array twice\n",
3277 set_imsm_ord_tbl_ent(map
, dk
->number
, dl
->index
);
3278 dl
->disk
.status
= CONFIGURED_DISK
;
3280 /* if we are creating the first raid device update the family number */
3281 if (super
->current_vol
== 0) {
3283 struct imsm_dev
*_dev
= __get_imsm_dev(mpb
, 0);
3284 struct imsm_disk
*_disk
= __get_imsm_disk(mpb
, dl
->index
);
3286 if (!_dev
|| !_disk
) {
3287 fprintf(stderr
, Name
": BUG mpb setup error\n");
3293 sum
+= __gen_imsm_checksum(mpb
);
3294 mpb
->family_num
= __cpu_to_le32(sum
);
3295 mpb
->orig_family_num
= mpb
->family_num
;
3301 static int add_to_super_imsm(struct supertype
*st
, mdu_disk_info_t
*dk
,
3302 int fd
, char *devname
)
3304 struct intel_super
*super
= st
->sb
;
3306 unsigned long long size
;
3311 /* if we are on an RAID enabled platform check that the disk is
3312 * attached to the raid controller
3314 if (super
->hba
&& !disk_attached_to_hba(fd
, super
->hba
)) {
3316 Name
": %s is not attached to the raid controller: %s\n",
3317 devname
? : "disk", super
->hba
);
3321 if (super
->current_vol
>= 0)
3322 return add_to_super_imsm_volume(st
, dk
, fd
, devname
);
3325 dd
= malloc(sizeof(*dd
));
3328 Name
": malloc failed %s:%d.\n", __func__
, __LINE__
);
3331 memset(dd
, 0, sizeof(*dd
));
3332 dd
->major
= major(stb
.st_rdev
);
3333 dd
->minor
= minor(stb
.st_rdev
);
3335 dd
->devname
= devname
? strdup(devname
) : NULL
;
3338 rv
= imsm_read_serial(fd
, devname
, dd
->serial
);
3341 Name
": failed to retrieve scsi serial, aborting\n");
3346 get_dev_size(fd
, NULL
, &size
);
3348 serialcpy(dd
->disk
.serial
, dd
->serial
);
3349 dd
->disk
.total_blocks
= __cpu_to_le32(size
);
3350 dd
->disk
.status
= SPARE_DISK
;
3351 if (sysfs_disk_to_scsi_id(fd
, &id
) == 0)
3352 dd
->disk
.scsi_id
= __cpu_to_le32(id
);
3354 dd
->disk
.scsi_id
= __cpu_to_le32(0);
3356 if (st
->update_tail
) {
3357 dd
->next
= super
->add
;
3360 dd
->next
= super
->disks
;
3367 static int store_imsm_mpb(int fd
, struct imsm_super
*mpb
);
3371 struct imsm_super anchor
;
3372 } spare_record
__attribute__ ((aligned(512)));
3374 /* spare records have their own family number and do not have any defined raid
3377 static int write_super_imsm_spares(struct intel_super
*super
, int doclose
)
3379 struct imsm_super
*mpb
= super
->anchor
;
3380 struct imsm_super
*spare
= &spare_record
.anchor
;
3384 spare
->mpb_size
= __cpu_to_le32(sizeof(struct imsm_super
)),
3385 spare
->generation_num
= __cpu_to_le32(1UL),
3386 spare
->attributes
= MPB_ATTRIB_CHECKSUM_VERIFY
;
3387 spare
->num_disks
= 1,
3388 spare
->num_raid_devs
= 0,
3389 spare
->cache_size
= mpb
->cache_size
,
3390 spare
->pwr_cycle_count
= __cpu_to_le32(1),
3392 snprintf((char *) spare
->sig
, MAX_SIGNATURE_LENGTH
,
3393 MPB_SIGNATURE MPB_VERSION_RAID0
);
3395 for (d
= super
->disks
; d
; d
= d
->next
) {
3399 spare
->disk
[0] = d
->disk
;
3400 sum
= __gen_imsm_checksum(spare
);
3401 spare
->family_num
= __cpu_to_le32(sum
);
3402 spare
->orig_family_num
= 0;
3403 sum
= __gen_imsm_checksum(spare
);
3404 spare
->check_sum
= __cpu_to_le32(sum
);
3406 if (store_imsm_mpb(d
->fd
, spare
)) {
3407 fprintf(stderr
, "%s: failed for device %d:%d %s\n",
3408 __func__
, d
->major
, d
->minor
, strerror(errno
));
3420 static int write_super_imsm(struct intel_super
*super
, int doclose
)
3422 struct imsm_super
*mpb
= super
->anchor
;
3428 __u32 mpb_size
= sizeof(struct imsm_super
) - sizeof(struct imsm_disk
);
3430 /* 'generation' is incremented everytime the metadata is written */
3431 generation
= __le32_to_cpu(mpb
->generation_num
);
3433 mpb
->generation_num
= __cpu_to_le32(generation
);
3435 /* fix up cases where previous mdadm releases failed to set
3438 if (mpb
->orig_family_num
== 0)
3439 mpb
->orig_family_num
= mpb
->family_num
;
3441 mpb_size
+= sizeof(struct imsm_disk
) * mpb
->num_disks
;
3442 for (d
= super
->disks
; d
; d
= d
->next
) {
3446 mpb
->disk
[d
->index
] = d
->disk
;
3448 for (d
= super
->missing
; d
; d
= d
->next
)
3449 mpb
->disk
[d
->index
] = d
->disk
;
3451 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
3452 struct imsm_dev
*dev
= __get_imsm_dev(mpb
, i
);
3454 imsm_copy_dev(dev
, get_imsm_dev(super
, i
));
3455 mpb_size
+= sizeof_imsm_dev(dev
, 0);
3457 mpb_size
+= __le32_to_cpu(mpb
->bbm_log_size
);
3458 mpb
->mpb_size
= __cpu_to_le32(mpb_size
);
3460 /* recalculate checksum */
3461 sum
= __gen_imsm_checksum(mpb
);
3462 mpb
->check_sum
= __cpu_to_le32(sum
);
3464 /* write the mpb for disks that compose raid devices */
3465 for (d
= super
->disks
; d
; d
= d
->next
) {
3468 if (store_imsm_mpb(d
->fd
, mpb
))
3469 fprintf(stderr
, "%s: failed for device %d:%d %s\n",
3470 __func__
, d
->major
, d
->minor
, strerror(errno
));
3478 return write_super_imsm_spares(super
, doclose
);
3484 static int create_array(struct supertype
*st
, int dev_idx
)
3487 struct imsm_update_create_array
*u
;
3488 struct intel_super
*super
= st
->sb
;
3489 struct imsm_dev
*dev
= get_imsm_dev(super
, dev_idx
);
3490 struct imsm_map
*map
= get_imsm_map(dev
, 0);
3491 struct disk_info
*inf
;
3492 struct imsm_disk
*disk
;
3495 len
= sizeof(*u
) - sizeof(*dev
) + sizeof_imsm_dev(dev
, 0) +
3496 sizeof(*inf
) * map
->num_members
;
3499 fprintf(stderr
, "%s: failed to allocate update buffer\n",
3504 u
->type
= update_create_array
;
3505 u
->dev_idx
= dev_idx
;
3506 imsm_copy_dev(&u
->dev
, dev
);
3507 inf
= get_disk_info(u
);
3508 for (i
= 0; i
< map
->num_members
; i
++) {
3509 int idx
= get_imsm_disk_idx(dev
, i
);
3511 disk
= get_imsm_disk(super
, idx
);
3512 serialcpy(inf
[i
].serial
, disk
->serial
);
3514 append_metadata_update(st
, u
, len
);
3519 static int _add_disk(struct supertype
*st
)
3521 struct intel_super
*super
= st
->sb
;
3523 struct imsm_update_add_disk
*u
;
3531 fprintf(stderr
, "%s: failed to allocate update buffer\n",
3536 u
->type
= update_add_disk
;
3537 append_metadata_update(st
, u
, len
);
3542 static int write_init_super_imsm(struct supertype
*st
)
3544 struct intel_super
*super
= st
->sb
;
3545 int current_vol
= super
->current_vol
;
3547 /* we are done with current_vol reset it to point st at the container */
3548 super
->current_vol
= -1;
3550 if (st
->update_tail
) {
3551 /* queue the recently created array / added disk
3552 * as a metadata update */
3556 /* determine if we are creating a volume or adding a disk */
3557 if (current_vol
< 0) {
3558 /* in the add disk case we are running in mdmon
3559 * context, so don't close fd's
3561 return _add_disk(st
);
3563 rv
= create_array(st
, current_vol
);
3565 for (d
= super
->disks
; d
; d
= d
->next
) {
3573 for (d
= super
->disks
; d
; d
= d
->next
)
3574 Kill(d
->devname
, NULL
, 0, 1, 1);
3575 return write_super_imsm(st
->sb
, 1);
3580 static int store_super_imsm(struct supertype
*st
, int fd
)
3582 struct intel_super
*super
= st
->sb
;
3583 struct imsm_super
*mpb
= super
? super
->anchor
: NULL
;
3589 return store_imsm_mpb(fd
, mpb
);
3595 static int imsm_bbm_log_size(struct imsm_super
*mpb
)
3597 return __le32_to_cpu(mpb
->bbm_log_size
);
3601 static int validate_geometry_imsm_container(struct supertype
*st
, int level
,
3602 int layout
, int raiddisks
, int chunk
,
3603 unsigned long long size
, char *dev
,
3604 unsigned long long *freesize
,
3608 unsigned long long ldsize
;
3609 const struct imsm_orom
*orom
;
3611 if (level
!= LEVEL_CONTAINER
)
3616 if (check_env("IMSM_NO_PLATFORM"))
3619 orom
= find_imsm_orom();
3620 if (orom
&& raiddisks
> orom
->tds
) {
3622 fprintf(stderr
, Name
": %d exceeds maximum number of"
3623 " platform supported disks: %d\n",
3624 raiddisks
, orom
->tds
);
3628 fd
= open(dev
, O_RDONLY
|O_EXCL
, 0);
3631 fprintf(stderr
, Name
": imsm: Cannot open %s: %s\n",
3632 dev
, strerror(errno
));
3635 if (!get_dev_size(fd
, dev
, &ldsize
)) {
3641 *freesize
= avail_size_imsm(st
, ldsize
>> 9);
3646 static unsigned long long find_size(struct extent
*e
, int *idx
, int num_extents
)
3648 const unsigned long long base_start
= e
[*idx
].start
;
3649 unsigned long long end
= base_start
+ e
[*idx
].size
;
3652 if (base_start
== end
)
3656 for (i
= *idx
; i
< num_extents
; i
++) {
3657 /* extend overlapping extents */
3658 if (e
[i
].start
>= base_start
&&
3659 e
[i
].start
<= end
) {
3662 if (e
[i
].start
+ e
[i
].size
> end
)
3663 end
= e
[i
].start
+ e
[i
].size
;
3664 } else if (e
[i
].start
> end
) {
3670 return end
- base_start
;
3673 static unsigned long long merge_extents(struct intel_super
*super
, int sum_extents
)
3675 /* build a composite disk with all known extents and generate a new
3676 * 'maxsize' given the "all disks in an array must share a common start
3677 * offset" constraint
3679 struct extent
*e
= calloc(sum_extents
, sizeof(*e
));
3683 unsigned long long pos
;
3684 unsigned long long start
= 0;
3685 unsigned long long maxsize
;
3686 unsigned long reserve
;
3691 /* coalesce and sort all extents. also, check to see if we need to
3692 * reserve space between member arrays
3695 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
3698 for (i
= 0; i
< dl
->extent_cnt
; i
++)
3701 qsort(e
, sum_extents
, sizeof(*e
), cmp_extent
);
3706 while (i
< sum_extents
) {
3707 e
[j
].start
= e
[i
].start
;
3708 e
[j
].size
= find_size(e
, &i
, sum_extents
);
3710 if (e
[j
-1].size
== 0)
3719 unsigned long long esize
;
3721 esize
= e
[i
].start
- pos
;
3722 if (esize
>= maxsize
) {
3727 pos
= e
[i
].start
+ e
[i
].size
;
3729 } while (e
[i
-1].size
);
3735 /* FIXME assumes volume at offset 0 is the first volume in a
3738 if (start_extent
> 0)
3739 reserve
= IMSM_RESERVED_SECTORS
; /* gap between raid regions */
3743 if (maxsize
< reserve
)
3746 super
->create_offset
= ~((__u32
) 0);
3747 if (start
+ reserve
> super
->create_offset
)
3748 return 0; /* start overflows create_offset */
3749 super
->create_offset
= start
+ reserve
;
3751 return maxsize
- reserve
;
3754 static int is_raid_level_supported(const struct imsm_orom
*orom
, int level
, int raiddisks
)
3756 if (level
< 0 || level
== 6 || level
== 4)
3759 /* if we have an orom prevent invalid raid levels */
3762 case 0: return imsm_orom_has_raid0(orom
);
3765 return imsm_orom_has_raid1e(orom
);
3766 return imsm_orom_has_raid1(orom
) && raiddisks
== 2;
3767 case 10: return imsm_orom_has_raid10(orom
) && raiddisks
== 4;
3768 case 5: return imsm_orom_has_raid5(orom
) && raiddisks
> 2;
3771 return 1; /* not on an Intel RAID platform so anything goes */
3776 #define pr_vrb(fmt, arg...) (void) (verbose && fprintf(stderr, Name fmt, ##arg))
3778 validate_geometry_imsm_orom(struct intel_super
*super
, int level
, int layout
,
3779 int raiddisks
, int chunk
, int verbose
)
3781 if (!is_raid_level_supported(super
->orom
, level
, raiddisks
)) {
3782 pr_vrb(": platform does not support raid%d with %d disk%s\n",
3783 level
, raiddisks
, raiddisks
> 1 ? "s" : "");
3786 if (super
->orom
&& level
!= 1 &&
3787 !imsm_orom_has_chunk(super
->orom
, chunk
)) {
3788 pr_vrb(": platform does not support a chunk size of: %d\n", chunk
);
3791 if (layout
!= imsm_level_to_layout(level
)) {
3793 pr_vrb(": imsm raid 5 only supports the left-asymmetric layout\n");
3794 else if (level
== 10)
3795 pr_vrb(": imsm raid 10 only supports the n2 layout\n");
3797 pr_vrb(": imsm unknown layout %#x for this raid level %d\n",
3805 /* validate_geometry_imsm_volume - lifted from validate_geometry_ddf_bvd
3806 * FIX ME add ahci details
3808 static int validate_geometry_imsm_volume(struct supertype
*st
, int level
,
3809 int layout
, int raiddisks
, int chunk
,
3810 unsigned long long size
, char *dev
,
3811 unsigned long long *freesize
,
3815 struct intel_super
*super
= st
->sb
;
3816 struct imsm_super
*mpb
= super
->anchor
;
3818 unsigned long long pos
= 0;
3819 unsigned long long maxsize
;
3823 /* We must have the container info already read in. */
3827 if (!validate_geometry_imsm_orom(super
, level
, layout
, raiddisks
, chunk
, verbose
))
3831 /* General test: make sure there is space for
3832 * 'raiddisks' device extents of size 'size' at a given
3835 unsigned long long minsize
= size
;
3836 unsigned long long start_offset
= MaxSector
;
3839 minsize
= MPB_SECTOR_CNT
+ IMSM_RESERVED_SECTORS
;
3840 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
3845 e
= get_extents(super
, dl
);
3848 unsigned long long esize
;
3849 esize
= e
[i
].start
- pos
;
3850 if (esize
>= minsize
)
3852 if (found
&& start_offset
== MaxSector
) {
3855 } else if (found
&& pos
!= start_offset
) {
3859 pos
= e
[i
].start
+ e
[i
].size
;
3861 } while (e
[i
-1].size
);
3866 if (dcnt
< raiddisks
) {
3868 fprintf(stderr
, Name
": imsm: Not enough "
3869 "devices with space for this array "
3877 /* This device must be a member of the set */
3878 if (stat(dev
, &stb
) < 0)
3880 if ((S_IFMT
& stb
.st_mode
) != S_IFBLK
)
3882 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
3883 if (dl
->major
== (int)major(stb
.st_rdev
) &&
3884 dl
->minor
== (int)minor(stb
.st_rdev
))
3889 fprintf(stderr
, Name
": %s is not in the "
3890 "same imsm set\n", dev
);
3892 } else if (super
->orom
&& dl
->index
< 0 && mpb
->num_raid_devs
) {
3893 /* If a volume is present then the current creation attempt
3894 * cannot incorporate new spares because the orom may not
3895 * understand this configuration (all member disks must be
3896 * members of each array in the container).
3898 fprintf(stderr
, Name
": %s is a spare and a volume"
3899 " is already defined for this container\n", dev
);
3900 fprintf(stderr
, Name
": The option-rom requires all member"
3901 " disks to be a member of all volumes\n");
3905 /* retrieve the largest free space block */
3906 e
= get_extents(super
, dl
);
3911 unsigned long long esize
;
3913 esize
= e
[i
].start
- pos
;
3914 if (esize
>= maxsize
)
3916 pos
= e
[i
].start
+ e
[i
].size
;
3918 } while (e
[i
-1].size
);
3923 fprintf(stderr
, Name
": unable to determine free space for: %s\n",
3927 if (maxsize
< size
) {
3929 fprintf(stderr
, Name
": %s not enough space (%llu < %llu)\n",
3930 dev
, maxsize
, size
);
3934 /* count total number of extents for merge */
3936 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
3938 i
+= dl
->extent_cnt
;
3940 maxsize
= merge_extents(super
, i
);
3941 if (maxsize
< size
|| maxsize
== 0) {
3943 fprintf(stderr
, Name
": not enough space after merge (%llu < %llu)\n",
3948 *freesize
= maxsize
;
3953 static int reserve_space(struct supertype
*st
, int raiddisks
,
3954 unsigned long long size
, int chunk
,
3955 unsigned long long *freesize
)
3957 struct intel_super
*super
= st
->sb
;
3958 struct imsm_super
*mpb
= super
->anchor
;
3963 unsigned long long maxsize
;
3964 unsigned long long minsize
;
3968 /* find the largest common start free region of the possible disks */
3972 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
3978 /* don't activate new spares if we are orom constrained
3979 * and there is already a volume active in the container
3981 if (super
->orom
&& dl
->index
< 0 && mpb
->num_raid_devs
)
3984 e
= get_extents(super
, dl
);
3987 for (i
= 1; e
[i
-1].size
; i
++)
3995 maxsize
= merge_extents(super
, extent_cnt
);
4000 if (cnt
< raiddisks
||
4001 (super
->orom
&& used
&& used
!= raiddisks
) ||
4002 maxsize
< minsize
||
4004 fprintf(stderr
, Name
": not enough devices with space to create array.\n");
4005 return 0; /* No enough free spaces large enough */
4017 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
4019 dl
->raiddisk
= cnt
++;
4026 static int validate_geometry_imsm(struct supertype
*st
, int level
, int layout
,
4027 int raiddisks
, int chunk
, unsigned long long size
,
4028 char *dev
, unsigned long long *freesize
,
4035 /* if given unused devices create a container
4036 * if given given devices in a container create a member volume
4038 if (level
== LEVEL_CONTAINER
) {
4039 /* Must be a fresh device to add to a container */
4040 return validate_geometry_imsm_container(st
, level
, layout
,
4041 raiddisks
, chunk
, size
,
4047 if (st
->sb
&& freesize
) {
4048 /* we are being asked to automatically layout a
4049 * new volume based on the current contents of
4050 * the container. If the the parameters can be
4051 * satisfied reserve_space will record the disks,
4052 * start offset, and size of the volume to be
4053 * created. add_to_super and getinfo_super
4054 * detect when autolayout is in progress.
4056 if (!validate_geometry_imsm_orom(st
->sb
, level
, layout
,
4060 return reserve_space(st
, raiddisks
, size
, chunk
, freesize
);
4065 /* creating in a given container */
4066 return validate_geometry_imsm_volume(st
, level
, layout
,
4067 raiddisks
, chunk
, size
,
4068 dev
, freesize
, verbose
);
4071 /* This device needs to be a device in an 'imsm' container */
4072 fd
= open(dev
, O_RDONLY
|O_EXCL
, 0);
4076 Name
": Cannot create this array on device %s\n",
4081 if (errno
!= EBUSY
|| (fd
= open(dev
, O_RDONLY
, 0)) < 0) {
4083 fprintf(stderr
, Name
": Cannot open %s: %s\n",
4084 dev
, strerror(errno
));
4087 /* Well, it is in use by someone, maybe an 'imsm' container. */
4088 cfd
= open_container(fd
);
4092 fprintf(stderr
, Name
": Cannot use %s: It is busy\n",
4096 sra
= sysfs_read(cfd
, 0, GET_VERSION
);
4097 if (sra
&& sra
->array
.major_version
== -1 &&
4098 strcmp(sra
->text_version
, "imsm") == 0)
4102 /* This is a member of a imsm container. Load the container
4103 * and try to create a volume
4105 struct intel_super
*super
;
4107 if (load_super_imsm_all(st
, cfd
, (void **) &super
, NULL
, 1) == 0) {
4109 st
->container_dev
= fd2devnum(cfd
);
4111 return validate_geometry_imsm_volume(st
, level
, layout
,
4119 fprintf(stderr
, Name
": failed container membership check\n");
4125 static int default_chunk_imsm(struct supertype
*st
)
4127 struct intel_super
*super
= st
->sb
;
4132 return imsm_orom_default_chunk(super
->orom
);
4135 static void handle_missing(struct intel_super
*super
, struct imsm_dev
*dev
);
4137 static int kill_subarray_imsm(struct supertype
*st
)
4139 /* remove the subarray currently referenced by ->current_vol */
4141 struct intel_dev
**dp
;
4142 struct intel_super
*super
= st
->sb
;
4143 __u8 current_vol
= super
->current_vol
;
4144 struct imsm_super
*mpb
= super
->anchor
;
4146 if (super
->current_vol
< 0)
4148 super
->current_vol
= -1; /* invalidate subarray cursor */
4150 /* block deletions that would change the uuid of active subarrays
4152 * FIXME when immutable ids are available, but note that we'll
4153 * also need to fixup the invalidated/active subarray indexes in
4156 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
4159 if (i
< current_vol
)
4161 sprintf(subarray
, "%u", i
);
4162 if (is_subarray_active(subarray
, st
->devname
)) {
4164 Name
": deleting subarray-%d would change the UUID of active subarray-%d, aborting\n",
4171 if (st
->update_tail
) {
4172 struct imsm_update_kill_array
*u
= malloc(sizeof(*u
));
4176 u
->type
= update_kill_array
;
4177 u
->dev_idx
= current_vol
;
4178 append_metadata_update(st
, u
, sizeof(*u
));
4183 for (dp
= &super
->devlist
; *dp
;)
4184 if ((*dp
)->index
== current_vol
) {
4187 handle_missing(super
, (*dp
)->dev
);
4188 if ((*dp
)->index
> current_vol
)
4193 /* no more raid devices, all active components are now spares,
4194 * but of course failed are still failed
4196 if (--mpb
->num_raid_devs
== 0) {
4199 for (d
= super
->disks
; d
; d
= d
->next
)
4200 if (d
->index
> -2) {
4202 d
->disk
.status
= SPARE_DISK
;
4206 super
->updates_pending
++;
4211 static int update_subarray_imsm(struct supertype
*st
, char *update
, mddev_ident_t ident
)
4213 /* update the subarray currently referenced by ->current_vol */
4214 struct intel_super
*super
= st
->sb
;
4215 struct imsm_super
*mpb
= super
->anchor
;
4217 if (super
->current_vol
< 0)
4220 if (strcmp(update
, "name") == 0) {
4221 char *name
= ident
->name
;
4223 if (is_subarray_active(st
->subarray
, st
->devname
)) {
4225 Name
": Unable to update name of active subarray\n");
4229 if (!check_name(super
, name
, 0))
4232 if (st
->update_tail
) {
4233 struct imsm_update_rename_array
*u
= malloc(sizeof(*u
));
4237 u
->type
= update_rename_array
;
4238 u
->dev_idx
= super
->current_vol
;
4239 snprintf((char *) u
->name
, MAX_RAID_SERIAL_LEN
, "%s", name
);
4240 append_metadata_update(st
, u
, sizeof(*u
));
4242 struct imsm_dev
*dev
;
4245 dev
= get_imsm_dev(super
, super
->current_vol
);
4246 snprintf((char *) dev
->volume
, MAX_RAID_SERIAL_LEN
, "%s", name
);
4247 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
4248 dev
= get_imsm_dev(super
, i
);
4249 handle_missing(super
, dev
);
4251 super
->updates_pending
++;
4258 #endif /* MDASSEMBLE */
4260 static int is_rebuilding(struct imsm_dev
*dev
)
4262 struct imsm_map
*migr_map
;
4264 if (!dev
->vol
.migr_state
)
4267 if (migr_type(dev
) != MIGR_REBUILD
)
4270 migr_map
= get_imsm_map(dev
, 1);
4272 if (migr_map
->map_state
== IMSM_T_STATE_DEGRADED
)
4278 static void update_recovery_start(struct imsm_dev
*dev
, struct mdinfo
*array
)
4280 struct mdinfo
*rebuild
= NULL
;
4284 if (!is_rebuilding(dev
))
4287 /* Find the rebuild target, but punt on the dual rebuild case */
4288 for (d
= array
->devs
; d
; d
= d
->next
)
4289 if (d
->recovery_start
== 0) {
4296 /* (?) none of the disks are marked with
4297 * IMSM_ORD_REBUILD, so assume they are missing and the
4298 * disk_ord_tbl was not correctly updated
4300 dprintf("%s: failed to locate out-of-sync disk\n", __func__
);
4304 units
= __le32_to_cpu(dev
->vol
.curr_migr_unit
);
4305 rebuild
->recovery_start
= units
* blocks_per_migr_unit(dev
);
4309 static struct mdinfo
*container_content_imsm(struct supertype
*st
)
4311 /* Given a container loaded by load_super_imsm_all,
4312 * extract information about all the arrays into
4315 * For each imsm_dev create an mdinfo, fill it in,
4316 * then look for matching devices in super->disks
4317 * and create appropriate device mdinfo.
4319 struct intel_super
*super
= st
->sb
;
4320 struct imsm_super
*mpb
= super
->anchor
;
4321 struct mdinfo
*rest
= NULL
;
4324 /* do not assemble arrays that might have bad blocks */
4325 if (imsm_bbm_log_size(super
->anchor
)) {
4326 fprintf(stderr
, Name
": BBM log found in metadata. "
4327 "Cannot activate array(s).\n");
4331 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
4332 struct imsm_dev
*dev
= get_imsm_dev(super
, i
);
4333 struct imsm_map
*map
= get_imsm_map(dev
, 0);
4334 struct mdinfo
*this;
4337 /* do not publish arrays that are in the middle of an
4338 * unsupported migration
4340 if (dev
->vol
.migr_state
&&
4341 (migr_type(dev
) == MIGR_GEN_MIGR
||
4342 migr_type(dev
) == MIGR_STATE_CHANGE
)) {
4343 fprintf(stderr
, Name
": cannot assemble volume '%.16s':"
4344 " unsupported migration in progress\n",
4349 this = malloc(sizeof(*this));
4351 fprintf(stderr
, Name
": failed to allocate %zu bytes\n",
4355 memset(this, 0, sizeof(*this));
4358 super
->current_vol
= i
;
4359 getinfo_super_imsm_volume(st
, this);
4360 for (slot
= 0 ; slot
< map
->num_members
; slot
++) {
4361 unsigned long long recovery_start
;
4362 struct mdinfo
*info_d
;
4369 idx
= get_imsm_disk_idx(dev
, slot
);
4370 ord
= get_imsm_ord_tbl_ent(dev
, slot
);
4371 for (d
= super
->disks
; d
; d
= d
->next
)
4372 if (d
->index
== idx
)
4375 recovery_start
= MaxSector
;
4378 if (d
&& is_failed(&d
->disk
))
4380 if (ord
& IMSM_ORD_REBUILD
)
4384 * if we skip some disks the array will be assmebled degraded;
4385 * reset resync start to avoid a dirty-degraded
4386 * situation when performing the intial sync
4388 * FIXME handle dirty degraded
4390 if ((skip
|| recovery_start
== 0) && !dev
->vol
.dirty
)
4391 this->resync_start
= MaxSector
;
4395 info_d
= calloc(1, sizeof(*info_d
));
4397 fprintf(stderr
, Name
": failed to allocate disk"
4398 " for volume %.16s\n", dev
->volume
);
4399 info_d
= this->devs
;
4401 struct mdinfo
*d
= info_d
->next
;
4410 info_d
->next
= this->devs
;
4411 this->devs
= info_d
;
4413 info_d
->disk
.number
= d
->index
;
4414 info_d
->disk
.major
= d
->major
;
4415 info_d
->disk
.minor
= d
->minor
;
4416 info_d
->disk
.raid_disk
= slot
;
4417 info_d
->recovery_start
= recovery_start
;
4419 if (info_d
->recovery_start
== MaxSector
)
4420 this->array
.working_disks
++;
4422 info_d
->events
= __le32_to_cpu(mpb
->generation_num
);
4423 info_d
->data_offset
= __le32_to_cpu(map
->pba_of_lba0
);
4424 info_d
->component_size
= __le32_to_cpu(map
->blocks_per_member
);
4426 /* now that the disk list is up-to-date fixup recovery_start */
4427 update_recovery_start(dev
, this);
4435 static __u8
imsm_check_degraded(struct intel_super
*super
, struct imsm_dev
*dev
, int failed
)
4437 struct imsm_map
*map
= get_imsm_map(dev
, 0);
4440 return map
->map_state
== IMSM_T_STATE_UNINITIALIZED
?
4441 IMSM_T_STATE_UNINITIALIZED
: IMSM_T_STATE_NORMAL
;
4443 switch (get_imsm_raid_level(map
)) {
4445 return IMSM_T_STATE_FAILED
;
4448 if (failed
< map
->num_members
)
4449 return IMSM_T_STATE_DEGRADED
;
4451 return IMSM_T_STATE_FAILED
;
4456 * check to see if any mirrors have failed, otherwise we
4457 * are degraded. Even numbered slots are mirrored on
4461 /* gcc -Os complains that this is unused */
4462 int insync
= insync
;
4464 for (i
= 0; i
< map
->num_members
; i
++) {
4465 __u32 ord
= get_imsm_ord_tbl_ent(dev
, i
);
4466 int idx
= ord_to_idx(ord
);
4467 struct imsm_disk
*disk
;
4469 /* reset the potential in-sync count on even-numbered
4470 * slots. num_copies is always 2 for imsm raid10
4475 disk
= get_imsm_disk(super
, idx
);
4476 if (!disk
|| is_failed(disk
) || ord
& IMSM_ORD_REBUILD
)
4479 /* no in-sync disks left in this mirror the
4483 return IMSM_T_STATE_FAILED
;
4486 return IMSM_T_STATE_DEGRADED
;
4490 return IMSM_T_STATE_DEGRADED
;
4492 return IMSM_T_STATE_FAILED
;
4498 return map
->map_state
;
4501 static int imsm_count_failed(struct intel_super
*super
, struct imsm_dev
*dev
)
4505 struct imsm_disk
*disk
;
4506 struct imsm_map
*map
= get_imsm_map(dev
, 0);
4507 struct imsm_map
*prev
= get_imsm_map(dev
, dev
->vol
.migr_state
);
4511 /* at the beginning of migration we set IMSM_ORD_REBUILD on
4512 * disks that are being rebuilt. New failures are recorded to
4513 * map[0]. So we look through all the disks we started with and
4514 * see if any failures are still present, or if any new ones
4517 * FIXME add support for online capacity expansion and
4518 * raid-level-migration
4520 for (i
= 0; i
< prev
->num_members
; i
++) {
4521 ord
= __le32_to_cpu(prev
->disk_ord_tbl
[i
]);
4522 ord
|= __le32_to_cpu(map
->disk_ord_tbl
[i
]);
4523 idx
= ord_to_idx(ord
);
4525 disk
= get_imsm_disk(super
, idx
);
4526 if (!disk
|| is_failed(disk
) || ord
& IMSM_ORD_REBUILD
)
4534 static int imsm_open_new(struct supertype
*c
, struct active_array
*a
,
4537 struct intel_super
*super
= c
->sb
;
4538 struct imsm_super
*mpb
= super
->anchor
;
4540 if (atoi(inst
) >= mpb
->num_raid_devs
) {
4541 fprintf(stderr
, "%s: subarry index %d, out of range\n",
4542 __func__
, atoi(inst
));
4546 dprintf("imsm: open_new %s\n", inst
);
4547 a
->info
.container_member
= atoi(inst
);
4551 static int is_resyncing(struct imsm_dev
*dev
)
4553 struct imsm_map
*migr_map
;
4555 if (!dev
->vol
.migr_state
)
4558 if (migr_type(dev
) == MIGR_INIT
||
4559 migr_type(dev
) == MIGR_REPAIR
)
4562 migr_map
= get_imsm_map(dev
, 1);
4564 if (migr_map
->map_state
== IMSM_T_STATE_NORMAL
)
4570 /* return true if we recorded new information */
4571 static int mark_failure(struct imsm_dev
*dev
, struct imsm_disk
*disk
, int idx
)
4575 struct imsm_map
*map
;
4577 /* new failures are always set in map[0] */
4578 map
= get_imsm_map(dev
, 0);
4580 slot
= get_imsm_disk_slot(map
, idx
);
4584 ord
= __le32_to_cpu(map
->disk_ord_tbl
[slot
]);
4585 if (is_failed(disk
) && (ord
& IMSM_ORD_REBUILD
))
4588 disk
->status
|= FAILED_DISK
;
4589 disk
->status
&= ~CONFIGURED_DISK
;
4590 set_imsm_ord_tbl_ent(map
, slot
, idx
| IMSM_ORD_REBUILD
);
4591 if (map
->failed_disk_num
== 0xff)
4592 map
->failed_disk_num
= slot
;
4596 static void mark_missing(struct imsm_dev
*dev
, struct imsm_disk
*disk
, int idx
)
4598 mark_failure(dev
, disk
, idx
);
4600 if (disk
->scsi_id
== __cpu_to_le32(~(__u32
)0))
4603 disk
->scsi_id
= __cpu_to_le32(~(__u32
)0);
4604 memmove(&disk
->serial
[0], &disk
->serial
[1], MAX_RAID_SERIAL_LEN
- 1);
4607 static void handle_missing(struct intel_super
*super
, struct imsm_dev
*dev
)
4613 if (!super
->missing
)
4615 failed
= imsm_count_failed(super
, dev
);
4616 map_state
= imsm_check_degraded(super
, dev
, failed
);
4618 dprintf("imsm: mark missing\n");
4619 end_migration(dev
, map_state
);
4620 for (dl
= super
->missing
; dl
; dl
= dl
->next
)
4621 mark_missing(dev
, &dl
->disk
, dl
->index
);
4622 super
->updates_pending
++;
4625 /* Handle dirty -> clean transititions and resync. Degraded and rebuild
4626 * states are handled in imsm_set_disk() with one exception, when a
4627 * resync is stopped due to a new failure this routine will set the
4628 * 'degraded' state for the array.
4630 static int imsm_set_array_state(struct active_array
*a
, int consistent
)
4632 int inst
= a
->info
.container_member
;
4633 struct intel_super
*super
= a
->container
->sb
;
4634 struct imsm_dev
*dev
= get_imsm_dev(super
, inst
);
4635 struct imsm_map
*map
= get_imsm_map(dev
, 0);
4636 int failed
= imsm_count_failed(super
, dev
);
4637 __u8 map_state
= imsm_check_degraded(super
, dev
, failed
);
4638 __u32 blocks_per_unit
;
4640 /* before we activate this array handle any missing disks */
4641 if (consistent
== 2)
4642 handle_missing(super
, dev
);
4644 if (consistent
== 2 &&
4645 (!is_resync_complete(&a
->info
) ||
4646 map_state
!= IMSM_T_STATE_NORMAL
||
4647 dev
->vol
.migr_state
))
4650 if (is_resync_complete(&a
->info
)) {
4651 /* complete intialization / resync,
4652 * recovery and interrupted recovery is completed in
4655 if (is_resyncing(dev
)) {
4656 dprintf("imsm: mark resync done\n");
4657 end_migration(dev
, map_state
);
4658 super
->updates_pending
++;
4659 a
->last_checkpoint
= 0;
4661 } else if (!is_resyncing(dev
) && !failed
) {
4662 /* mark the start of the init process if nothing is failed */
4663 dprintf("imsm: mark resync start\n");
4664 if (map
->map_state
== IMSM_T_STATE_UNINITIALIZED
)
4665 migrate(dev
, IMSM_T_STATE_NORMAL
, MIGR_INIT
);
4667 migrate(dev
, IMSM_T_STATE_NORMAL
, MIGR_REPAIR
);
4668 super
->updates_pending
++;
4671 /* check if we can update curr_migr_unit from resync_start, recovery_start */
4672 blocks_per_unit
= blocks_per_migr_unit(dev
);
4673 if (blocks_per_unit
) {
4677 units
= a
->last_checkpoint
/ blocks_per_unit
;
4680 /* check that we did not overflow 32-bits, and that
4681 * curr_migr_unit needs updating
4683 if (units32
== units
&&
4684 __le32_to_cpu(dev
->vol
.curr_migr_unit
) != units32
) {
4685 dprintf("imsm: mark checkpoint (%u)\n", units32
);
4686 dev
->vol
.curr_migr_unit
= __cpu_to_le32(units32
);
4687 super
->updates_pending
++;
4691 /* mark dirty / clean */
4692 if (dev
->vol
.dirty
!= !consistent
) {
4693 dprintf("imsm: mark '%s'\n", consistent
? "clean" : "dirty");
4698 super
->updates_pending
++;
4703 static void imsm_set_disk(struct active_array
*a
, int n
, int state
)
4705 int inst
= a
->info
.container_member
;
4706 struct intel_super
*super
= a
->container
->sb
;
4707 struct imsm_dev
*dev
= get_imsm_dev(super
, inst
);
4708 struct imsm_map
*map
= get_imsm_map(dev
, 0);
4709 struct imsm_disk
*disk
;
4714 if (n
> map
->num_members
)
4715 fprintf(stderr
, "imsm: set_disk %d out of range 0..%d\n",
4716 n
, map
->num_members
- 1);
4721 dprintf("imsm: set_disk %d:%x\n", n
, state
);
4723 ord
= get_imsm_ord_tbl_ent(dev
, n
);
4724 disk
= get_imsm_disk(super
, ord_to_idx(ord
));
4726 /* check for new failures */
4727 if (state
& DS_FAULTY
) {
4728 if (mark_failure(dev
, disk
, ord_to_idx(ord
)))
4729 super
->updates_pending
++;
4732 /* check if in_sync */
4733 if (state
& DS_INSYNC
&& ord
& IMSM_ORD_REBUILD
&& is_rebuilding(dev
)) {
4734 struct imsm_map
*migr_map
= get_imsm_map(dev
, 1);
4736 set_imsm_ord_tbl_ent(migr_map
, n
, ord_to_idx(ord
));
4737 super
->updates_pending
++;
4740 failed
= imsm_count_failed(super
, dev
);
4741 map_state
= imsm_check_degraded(super
, dev
, failed
);
4743 /* check if recovery complete, newly degraded, or failed */
4744 if (map_state
== IMSM_T_STATE_NORMAL
&& is_rebuilding(dev
)) {
4745 end_migration(dev
, map_state
);
4746 map
= get_imsm_map(dev
, 0);
4747 map
->failed_disk_num
= ~0;
4748 super
->updates_pending
++;
4749 a
->last_checkpoint
= 0;
4750 } else if (map_state
== IMSM_T_STATE_DEGRADED
&&
4751 map
->map_state
!= map_state
&&
4752 !dev
->vol
.migr_state
) {
4753 dprintf("imsm: mark degraded\n");
4754 map
->map_state
= map_state
;
4755 super
->updates_pending
++;
4756 a
->last_checkpoint
= 0;
4757 } else if (map_state
== IMSM_T_STATE_FAILED
&&
4758 map
->map_state
!= map_state
) {
4759 dprintf("imsm: mark failed\n");
4760 end_migration(dev
, map_state
);
4761 super
->updates_pending
++;
4762 a
->last_checkpoint
= 0;
4766 static int store_imsm_mpb(int fd
, struct imsm_super
*mpb
)
4769 __u32 mpb_size
= __le32_to_cpu(mpb
->mpb_size
);
4770 unsigned long long dsize
;
4771 unsigned long long sectors
;
4773 get_dev_size(fd
, NULL
, &dsize
);
4775 if (mpb_size
> 512) {
4776 /* -1 to account for anchor */
4777 sectors
= mpb_sectors(mpb
) - 1;
4779 /* write the extended mpb to the sectors preceeding the anchor */
4780 if (lseek64(fd
, dsize
- (512 * (2 + sectors
)), SEEK_SET
) < 0)
4783 if ((unsigned long long)write(fd
, buf
+ 512, 512 * sectors
)
4788 /* first block is stored on second to last sector of the disk */
4789 if (lseek64(fd
, dsize
- (512 * 2), SEEK_SET
) < 0)
4792 if (write(fd
, buf
, 512) != 512)
4798 static void imsm_sync_metadata(struct supertype
*container
)
4800 struct intel_super
*super
= container
->sb
;
4802 if (!super
->updates_pending
)
4805 write_super_imsm(super
, 0);
4807 super
->updates_pending
= 0;
4810 static struct dl
*imsm_readd(struct intel_super
*super
, int idx
, struct active_array
*a
)
4812 struct imsm_dev
*dev
= get_imsm_dev(super
, a
->info
.container_member
);
4813 int i
= get_imsm_disk_idx(dev
, idx
);
4816 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
4820 if (dl
&& is_failed(&dl
->disk
))
4824 dprintf("%s: found %x:%x\n", __func__
, dl
->major
, dl
->minor
);
4829 static struct dl
*imsm_add_spare(struct intel_super
*super
, int slot
,
4830 struct active_array
*a
, int activate_new
)
4832 struct imsm_dev
*dev
= get_imsm_dev(super
, a
->info
.container_member
);
4833 int idx
= get_imsm_disk_idx(dev
, slot
);
4834 struct imsm_super
*mpb
= super
->anchor
;
4835 struct imsm_map
*map
;
4836 unsigned long long pos
;
4841 __u32 array_start
= 0;
4842 __u32 array_end
= 0;
4845 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
4846 /* If in this array, skip */
4847 for (d
= a
->info
.devs
; d
; d
= d
->next
)
4848 if (d
->state_fd
>= 0 &&
4849 d
->disk
.major
== dl
->major
&&
4850 d
->disk
.minor
== dl
->minor
) {
4851 dprintf("%x:%x already in array\n", dl
->major
, dl
->minor
);
4857 /* skip in use or failed drives */
4858 if (is_failed(&dl
->disk
) || idx
== dl
->index
||
4860 dprintf("%x:%x status (failed: %d index: %d)\n",
4861 dl
->major
, dl
->minor
, is_failed(&dl
->disk
), idx
);
4865 /* skip pure spares when we are looking for partially
4866 * assimilated drives
4868 if (dl
->index
== -1 && !activate_new
)
4871 /* Does this unused device have the requisite free space?
4872 * It needs to be able to cover all member volumes
4874 ex
= get_extents(super
, dl
);
4876 dprintf("cannot get extents\n");
4879 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
4880 dev
= get_imsm_dev(super
, i
);
4881 map
= get_imsm_map(dev
, 0);
4883 /* check if this disk is already a member of
4886 if (get_imsm_disk_slot(map
, dl
->index
) >= 0)
4892 array_start
= __le32_to_cpu(map
->pba_of_lba0
);
4893 array_end
= array_start
+
4894 __le32_to_cpu(map
->blocks_per_member
) - 1;
4897 /* check that we can start at pba_of_lba0 with
4898 * blocks_per_member of space
4900 if (array_start
>= pos
&& array_end
< ex
[j
].start
) {
4904 pos
= ex
[j
].start
+ ex
[j
].size
;
4906 } while (ex
[j
-1].size
);
4913 if (i
< mpb
->num_raid_devs
) {
4914 dprintf("%x:%x does not have %u to %u available\n",
4915 dl
->major
, dl
->minor
, array_start
, array_end
);
4925 static struct mdinfo
*imsm_activate_spare(struct active_array
*a
,
4926 struct metadata_update
**updates
)
4929 * Find a device with unused free space and use it to replace a
4930 * failed/vacant region in an array. We replace failed regions one a
4931 * array at a time. The result is that a new spare disk will be added
4932 * to the first failed array and after the monitor has finished
4933 * propagating failures the remainder will be consumed.
4935 * FIXME add a capability for mdmon to request spares from another
4939 struct intel_super
*super
= a
->container
->sb
;
4940 int inst
= a
->info
.container_member
;
4941 struct imsm_dev
*dev
= get_imsm_dev(super
, inst
);
4942 struct imsm_map
*map
= get_imsm_map(dev
, 0);
4943 int failed
= a
->info
.array
.raid_disks
;
4944 struct mdinfo
*rv
= NULL
;
4947 struct metadata_update
*mu
;
4949 struct imsm_update_activate_spare
*u
;
4953 for (d
= a
->info
.devs
; d
; d
= d
->next
) {
4954 if ((d
->curr_state
& DS_FAULTY
) &&
4956 /* wait for Removal to happen */
4958 if (d
->state_fd
>= 0)
4962 dprintf("imsm: activate spare: inst=%d failed=%d (%d) level=%d\n",
4963 inst
, failed
, a
->info
.array
.raid_disks
, a
->info
.array
.level
);
4964 if (imsm_check_degraded(super
, dev
, failed
) != IMSM_T_STATE_DEGRADED
)
4967 /* For each slot, if it is not working, find a spare */
4968 for (i
= 0; i
< a
->info
.array
.raid_disks
; i
++) {
4969 for (d
= a
->info
.devs
; d
; d
= d
->next
)
4970 if (d
->disk
.raid_disk
== i
)
4972 dprintf("found %d: %p %x\n", i
, d
, d
?d
->curr_state
:0);
4973 if (d
&& (d
->state_fd
>= 0))
4977 * OK, this device needs recovery. Try to re-add the
4978 * previous occupant of this slot, if this fails see if
4979 * we can continue the assimilation of a spare that was
4980 * partially assimilated, finally try to activate a new
4983 dl
= imsm_readd(super
, i
, a
);
4985 dl
= imsm_add_spare(super
, i
, a
, 0);
4987 dl
= imsm_add_spare(super
, i
, a
, 1);
4991 /* found a usable disk with enough space */
4992 di
= malloc(sizeof(*di
));
4995 memset(di
, 0, sizeof(*di
));
4997 /* dl->index will be -1 in the case we are activating a
4998 * pristine spare. imsm_process_update() will create a
4999 * new index in this case. Once a disk is found to be
5000 * failed in all member arrays it is kicked from the
5003 di
->disk
.number
= dl
->index
;
5005 /* (ab)use di->devs to store a pointer to the device
5008 di
->devs
= (struct mdinfo
*) dl
;
5010 di
->disk
.raid_disk
= i
;
5011 di
->disk
.major
= dl
->major
;
5012 di
->disk
.minor
= dl
->minor
;
5014 di
->recovery_start
= 0;
5015 di
->data_offset
= __le32_to_cpu(map
->pba_of_lba0
);
5016 di
->component_size
= a
->info
.component_size
;
5017 di
->container_member
= inst
;
5018 super
->random
= random32();
5022 dprintf("%x:%x to be %d at %llu\n", dl
->major
, dl
->minor
,
5023 i
, di
->data_offset
);
5029 /* No spares found */
5031 /* Now 'rv' has a list of devices to return.
5032 * Create a metadata_update record to update the
5033 * disk_ord_tbl for the array
5035 mu
= malloc(sizeof(*mu
));
5037 mu
->buf
= malloc(sizeof(struct imsm_update_activate_spare
) * num_spares
);
5038 if (mu
->buf
== NULL
) {
5045 struct mdinfo
*n
= rv
->next
;
5054 mu
->len
= sizeof(struct imsm_update_activate_spare
) * num_spares
;
5055 mu
->next
= *updates
;
5056 u
= (struct imsm_update_activate_spare
*) mu
->buf
;
5058 for (di
= rv
; di
; di
= di
->next
) {
5059 u
->type
= update_activate_spare
;
5060 u
->dl
= (struct dl
*) di
->devs
;
5062 u
->slot
= di
->disk
.raid_disk
;
5073 static int disks_overlap(struct intel_super
*super
, int idx
, struct imsm_update_create_array
*u
)
5075 struct imsm_dev
*dev
= get_imsm_dev(super
, idx
);
5076 struct imsm_map
*map
= get_imsm_map(dev
, 0);
5077 struct imsm_map
*new_map
= get_imsm_map(&u
->dev
, 0);
5078 struct disk_info
*inf
= get_disk_info(u
);
5079 struct imsm_disk
*disk
;
5083 for (i
= 0; i
< map
->num_members
; i
++) {
5084 disk
= get_imsm_disk(super
, get_imsm_disk_idx(dev
, i
));
5085 for (j
= 0; j
< new_map
->num_members
; j
++)
5086 if (serialcmp(disk
->serial
, inf
[j
].serial
) == 0)
5093 static void imsm_delete(struct intel_super
*super
, struct dl
**dlp
, unsigned index
);
5095 static void imsm_process_update(struct supertype
*st
,
5096 struct metadata_update
*update
)
5099 * crack open the metadata_update envelope to find the update record
5100 * update can be one of:
5101 * update_activate_spare - a spare device has replaced a failed
5102 * device in an array, update the disk_ord_tbl. If this disk is
5103 * present in all member arrays then also clear the SPARE_DISK
5106 struct intel_super
*super
= st
->sb
;
5107 struct imsm_super
*mpb
;
5108 enum imsm_update_type type
= *(enum imsm_update_type
*) update
->buf
;
5110 /* update requires a larger buf but the allocation failed */
5111 if (super
->next_len
&& !super
->next_buf
) {
5112 super
->next_len
= 0;
5116 if (super
->next_buf
) {
5117 memcpy(super
->next_buf
, super
->buf
, super
->len
);
5119 super
->len
= super
->next_len
;
5120 super
->buf
= super
->next_buf
;
5122 super
->next_len
= 0;
5123 super
->next_buf
= NULL
;
5126 mpb
= super
->anchor
;
5129 case update_activate_spare
: {
5130 struct imsm_update_activate_spare
*u
= (void *) update
->buf
;
5131 struct imsm_dev
*dev
= get_imsm_dev(super
, u
->array
);
5132 struct imsm_map
*map
= get_imsm_map(dev
, 0);
5133 struct imsm_map
*migr_map
;
5134 struct active_array
*a
;
5135 struct imsm_disk
*disk
;
5140 int victim
= get_imsm_disk_idx(dev
, u
->slot
);
5143 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
5148 fprintf(stderr
, "error: imsm_activate_spare passed "
5149 "an unknown disk (index: %d)\n",
5154 super
->updates_pending
++;
5156 /* count failures (excluding rebuilds and the victim)
5157 * to determine map[0] state
5160 for (i
= 0; i
< map
->num_members
; i
++) {
5163 disk
= get_imsm_disk(super
, get_imsm_disk_idx(dev
, i
));
5164 if (!disk
|| is_failed(disk
))
5168 /* adding a pristine spare, assign a new index */
5169 if (dl
->index
< 0) {
5170 dl
->index
= super
->anchor
->num_disks
;
5171 super
->anchor
->num_disks
++;
5174 disk
->status
|= CONFIGURED_DISK
;
5175 disk
->status
&= ~SPARE_DISK
;
5178 to_state
= imsm_check_degraded(super
, dev
, failed
);
5179 map
->map_state
= IMSM_T_STATE_DEGRADED
;
5180 migrate(dev
, to_state
, MIGR_REBUILD
);
5181 migr_map
= get_imsm_map(dev
, 1);
5182 set_imsm_ord_tbl_ent(map
, u
->slot
, dl
->index
);
5183 set_imsm_ord_tbl_ent(migr_map
, u
->slot
, dl
->index
| IMSM_ORD_REBUILD
);
5185 /* update the family_num to mark a new container
5186 * generation, being careful to record the existing
5187 * family_num in orig_family_num to clean up after
5188 * earlier mdadm versions that neglected to set it.
5190 if (mpb
->orig_family_num
== 0)
5191 mpb
->orig_family_num
= mpb
->family_num
;
5192 mpb
->family_num
+= super
->random
;
5194 /* count arrays using the victim in the metadata */
5196 for (a
= st
->arrays
; a
; a
= a
->next
) {
5197 dev
= get_imsm_dev(super
, a
->info
.container_member
);
5198 map
= get_imsm_map(dev
, 0);
5200 if (get_imsm_disk_slot(map
, victim
) >= 0)
5204 /* delete the victim if it is no longer being
5210 /* We know that 'manager' isn't touching anything,
5211 * so it is safe to delete
5213 for (dlp
= &super
->disks
; *dlp
; dlp
= &(*dlp
)->next
)
5214 if ((*dlp
)->index
== victim
)
5217 /* victim may be on the missing list */
5219 for (dlp
= &super
->missing
; *dlp
; dlp
= &(*dlp
)->next
)
5220 if ((*dlp
)->index
== victim
)
5222 imsm_delete(super
, dlp
, victim
);
5226 case update_create_array
: {
5227 /* someone wants to create a new array, we need to be aware of
5228 * a few races/collisions:
5229 * 1/ 'Create' called by two separate instances of mdadm
5230 * 2/ 'Create' versus 'activate_spare': mdadm has chosen
5231 * devices that have since been assimilated via
5233 * In the event this update can not be carried out mdadm will
5234 * (FIX ME) notice that its update did not take hold.
5236 struct imsm_update_create_array
*u
= (void *) update
->buf
;
5237 struct intel_dev
*dv
;
5238 struct imsm_dev
*dev
;
5239 struct imsm_map
*map
, *new_map
;
5240 unsigned long long start
, end
;
5241 unsigned long long new_start
, new_end
;
5243 struct disk_info
*inf
;
5246 /* handle racing creates: first come first serve */
5247 if (u
->dev_idx
< mpb
->num_raid_devs
) {
5248 dprintf("%s: subarray %d already defined\n",
5249 __func__
, u
->dev_idx
);
5253 /* check update is next in sequence */
5254 if (u
->dev_idx
!= mpb
->num_raid_devs
) {
5255 dprintf("%s: can not create array %d expected index %d\n",
5256 __func__
, u
->dev_idx
, mpb
->num_raid_devs
);
5260 new_map
= get_imsm_map(&u
->dev
, 0);
5261 new_start
= __le32_to_cpu(new_map
->pba_of_lba0
);
5262 new_end
= new_start
+ __le32_to_cpu(new_map
->blocks_per_member
);
5263 inf
= get_disk_info(u
);
5265 /* handle activate_spare versus create race:
5266 * check to make sure that overlapping arrays do not include
5269 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
5270 dev
= get_imsm_dev(super
, i
);
5271 map
= get_imsm_map(dev
, 0);
5272 start
= __le32_to_cpu(map
->pba_of_lba0
);
5273 end
= start
+ __le32_to_cpu(map
->blocks_per_member
);
5274 if ((new_start
>= start
&& new_start
<= end
) ||
5275 (start
>= new_start
&& start
<= new_end
))
5280 if (disks_overlap(super
, i
, u
)) {
5281 dprintf("%s: arrays overlap\n", __func__
);
5286 /* check that prepare update was successful */
5287 if (!update
->space
) {
5288 dprintf("%s: prepare update failed\n", __func__
);
5292 /* check that all disks are still active before committing
5293 * changes. FIXME: could we instead handle this by creating a
5294 * degraded array? That's probably not what the user expects,
5295 * so better to drop this update on the floor.
5297 for (i
= 0; i
< new_map
->num_members
; i
++) {
5298 dl
= serial_to_dl(inf
[i
].serial
, super
);
5300 dprintf("%s: disk disappeared\n", __func__
);
5305 super
->updates_pending
++;
5307 /* convert spares to members and fixup ord_tbl */
5308 for (i
= 0; i
< new_map
->num_members
; i
++) {
5309 dl
= serial_to_dl(inf
[i
].serial
, super
);
5310 if (dl
->index
== -1) {
5311 dl
->index
= mpb
->num_disks
;
5313 dl
->disk
.status
|= CONFIGURED_DISK
;
5314 dl
->disk
.status
&= ~SPARE_DISK
;
5316 set_imsm_ord_tbl_ent(new_map
, i
, dl
->index
);
5321 update
->space
= NULL
;
5322 imsm_copy_dev(dev
, &u
->dev
);
5323 dv
->index
= u
->dev_idx
;
5324 dv
->next
= super
->devlist
;
5325 super
->devlist
= dv
;
5326 mpb
->num_raid_devs
++;
5328 imsm_update_version_info(super
);
5331 /* mdmon knows how to release update->space, but not
5332 * ((struct intel_dev *) update->space)->dev
5334 if (update
->space
) {
5340 case update_kill_array
: {
5341 struct imsm_update_kill_array
*u
= (void *) update
->buf
;
5342 int victim
= u
->dev_idx
;
5343 struct active_array
*a
;
5344 struct intel_dev
**dp
;
5345 struct imsm_dev
*dev
;
5347 /* sanity check that we are not affecting the uuid of
5348 * active arrays, or deleting an active array
5350 * FIXME when immutable ids are available, but note that
5351 * we'll also need to fixup the invalidated/active
5352 * subarray indexes in mdstat
5354 for (a
= st
->arrays
; a
; a
= a
->next
)
5355 if (a
->info
.container_member
>= victim
)
5357 /* by definition if mdmon is running at least one array
5358 * is active in the container, so checking
5359 * mpb->num_raid_devs is just extra paranoia
5361 dev
= get_imsm_dev(super
, victim
);
5362 if (a
|| !dev
|| mpb
->num_raid_devs
== 1) {
5363 dprintf("failed to delete subarray-%d\n", victim
);
5367 for (dp
= &super
->devlist
; *dp
;)
5368 if ((*dp
)->index
== (unsigned)super
->current_vol
) {
5371 if ((*dp
)->index
> (unsigned)victim
)
5375 mpb
->num_raid_devs
--;
5376 super
->updates_pending
++;
5379 case update_rename_array
: {
5380 struct imsm_update_rename_array
*u
= (void *) update
->buf
;
5381 char name
[MAX_RAID_SERIAL_LEN
+1];
5382 int target
= u
->dev_idx
;
5383 struct active_array
*a
;
5384 struct imsm_dev
*dev
;
5386 /* sanity check that we are not affecting the uuid of
5389 snprintf(name
, MAX_RAID_SERIAL_LEN
, "%s", (char *) u
->name
);
5390 name
[MAX_RAID_SERIAL_LEN
] = '\0';
5391 for (a
= st
->arrays
; a
; a
= a
->next
)
5392 if (a
->info
.container_member
== target
)
5394 dev
= get_imsm_dev(super
, u
->dev_idx
);
5395 if (a
|| !dev
|| !check_name(super
, name
, 1)) {
5396 dprintf("failed to rename subarray-%d\n", target
);
5400 snprintf((char *) dev
->volume
, MAX_RAID_SERIAL_LEN
, "%s", name
);
5401 super
->updates_pending
++;
5404 case update_add_disk
:
5406 /* we may be able to repair some arrays if disks are
5409 struct active_array
*a
;
5411 super
->updates_pending
++;
5412 for (a
= st
->arrays
; a
; a
= a
->next
)
5413 a
->check_degraded
= 1;
5415 /* add some spares to the metadata */
5416 while (super
->add
) {
5420 super
->add
= al
->next
;
5421 al
->next
= super
->disks
;
5423 dprintf("%s: added %x:%x\n",
5424 __func__
, al
->major
, al
->minor
);
5431 static void imsm_prepare_update(struct supertype
*st
,
5432 struct metadata_update
*update
)
5435 * Allocate space to hold new disk entries, raid-device entries or a new
5436 * mpb if necessary. The manager synchronously waits for updates to
5437 * complete in the monitor, so new mpb buffers allocated here can be
5438 * integrated by the monitor thread without worrying about live pointers
5439 * in the manager thread.
5441 enum imsm_update_type type
= *(enum imsm_update_type
*) update
->buf
;
5442 struct intel_super
*super
= st
->sb
;
5443 struct imsm_super
*mpb
= super
->anchor
;
5448 case update_create_array
: {
5449 struct imsm_update_create_array
*u
= (void *) update
->buf
;
5450 struct intel_dev
*dv
;
5451 struct imsm_dev
*dev
= &u
->dev
;
5452 struct imsm_map
*map
= get_imsm_map(dev
, 0);
5454 struct disk_info
*inf
;
5458 inf
= get_disk_info(u
);
5459 len
= sizeof_imsm_dev(dev
, 1);
5460 /* allocate a new super->devlist entry */
5461 dv
= malloc(sizeof(*dv
));
5463 dv
->dev
= malloc(len
);
5468 update
->space
= NULL
;
5472 /* count how many spares will be converted to members */
5473 for (i
= 0; i
< map
->num_members
; i
++) {
5474 dl
= serial_to_dl(inf
[i
].serial
, super
);
5476 /* hmm maybe it failed?, nothing we can do about
5481 if (count_memberships(dl
, super
) == 0)
5484 len
+= activate
* sizeof(struct imsm_disk
);
5491 /* check if we need a larger metadata buffer */
5492 if (super
->next_buf
)
5493 buf_len
= super
->next_len
;
5495 buf_len
= super
->len
;
5497 if (__le32_to_cpu(mpb
->mpb_size
) + len
> buf_len
) {
5498 /* ok we need a larger buf than what is currently allocated
5499 * if this allocation fails process_update will notice that
5500 * ->next_len is set and ->next_buf is NULL
5502 buf_len
= ROUND_UP(__le32_to_cpu(mpb
->mpb_size
) + len
, 512);
5503 if (super
->next_buf
)
5504 free(super
->next_buf
);
5506 super
->next_len
= buf_len
;
5507 if (posix_memalign(&super
->next_buf
, 512, buf_len
) == 0)
5508 memset(super
->next_buf
, 0, buf_len
);
5510 super
->next_buf
= NULL
;
5514 /* must be called while manager is quiesced */
5515 static void imsm_delete(struct intel_super
*super
, struct dl
**dlp
, unsigned index
)
5517 struct imsm_super
*mpb
= super
->anchor
;
5519 struct imsm_dev
*dev
;
5520 struct imsm_map
*map
;
5521 int i
, j
, num_members
;
5524 dprintf("%s: deleting device[%d] from imsm_super\n",
5527 /* shift all indexes down one */
5528 for (iter
= super
->disks
; iter
; iter
= iter
->next
)
5529 if (iter
->index
> (int)index
)
5531 for (iter
= super
->missing
; iter
; iter
= iter
->next
)
5532 if (iter
->index
> (int)index
)
5535 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
5536 dev
= get_imsm_dev(super
, i
);
5537 map
= get_imsm_map(dev
, 0);
5538 num_members
= map
->num_members
;
5539 for (j
= 0; j
< num_members
; j
++) {
5540 /* update ord entries being careful not to propagate
5541 * ord-flags to the first map
5543 ord
= get_imsm_ord_tbl_ent(dev
, j
);
5545 if (ord_to_idx(ord
) <= index
)
5548 map
= get_imsm_map(dev
, 0);
5549 set_imsm_ord_tbl_ent(map
, j
, ord_to_idx(ord
- 1));
5550 map
= get_imsm_map(dev
, 1);
5552 set_imsm_ord_tbl_ent(map
, j
, ord
- 1);
5557 super
->updates_pending
++;
5559 struct dl
*dl
= *dlp
;
5561 *dlp
= (*dlp
)->next
;
5562 __free_imsm_disk(dl
);
5565 #endif /* MDASSEMBLE */
5567 struct superswitch super_imsm
= {
5569 .examine_super
= examine_super_imsm
,
5570 .brief_examine_super
= brief_examine_super_imsm
,
5571 .brief_examine_subarrays
= brief_examine_subarrays_imsm
,
5572 .export_examine_super
= export_examine_super_imsm
,
5573 .detail_super
= detail_super_imsm
,
5574 .brief_detail_super
= brief_detail_super_imsm
,
5575 .write_init_super
= write_init_super_imsm
,
5576 .validate_geometry
= validate_geometry_imsm
,
5577 .default_chunk
= default_chunk_imsm
,
5578 .add_to_super
= add_to_super_imsm
,
5579 .detail_platform
= detail_platform_imsm
,
5580 .kill_subarray
= kill_subarray_imsm
,
5581 .update_subarray
= update_subarray_imsm
,
5583 .match_home
= match_home_imsm
,
5584 .uuid_from_super
= uuid_from_super_imsm
,
5585 .getinfo_super
= getinfo_super_imsm
,
5586 .update_super
= update_super_imsm
,
5588 .avail_size
= avail_size_imsm
,
5590 .compare_super
= compare_super_imsm
,
5592 .load_super
= load_super_imsm
,
5593 .init_super
= init_super_imsm
,
5594 .store_super
= store_super_imsm
,
5595 .free_super
= free_super_imsm
,
5596 .match_metadata_desc
= match_metadata_desc_imsm
,
5597 .container_content
= container_content_imsm
,
5598 .default_layout
= imsm_level_to_layout
,
5605 .open_new
= imsm_open_new
,
5606 .set_array_state
= imsm_set_array_state
,
5607 .set_disk
= imsm_set_disk
,
5608 .sync_metadata
= imsm_sync_metadata
,
5609 .activate_spare
= imsm_activate_spare
,
5610 .process_update
= imsm_process_update
,
5611 .prepare_update
= imsm_prepare_update
,
5612 #endif /* MDASSEMBLE */