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
->container_dev
= NoMdDev
;
337 st
->ss
= &super_imsm
;
338 st
->max_devs
= IMSM_MAX_DEVICES
;
339 st
->minor_version
= 0;
345 static __u8
*get_imsm_version(struct imsm_super
*mpb
)
347 return &mpb
->sig
[MPB_SIG_LEN
];
351 /* retrieve a disk directly from the anchor when the anchor is known to be
352 * up-to-date, currently only at load time
354 static struct imsm_disk
*__get_imsm_disk(struct imsm_super
*mpb
, __u8 index
)
356 if (index
>= mpb
->num_disks
)
358 return &mpb
->disk
[index
];
361 /* retrieve a disk from the parsed metadata */
362 static struct imsm_disk
*get_imsm_disk(struct intel_super
*super
, __u8 index
)
366 for (d
= super
->disks
; d
; d
= d
->next
)
367 if (d
->index
== index
)
373 /* generate a checksum directly from the anchor when the anchor is known to be
374 * up-to-date, currently only at load or write_super after coalescing
376 static __u32
__gen_imsm_checksum(struct imsm_super
*mpb
)
378 __u32 end
= mpb
->mpb_size
/ sizeof(end
);
379 __u32
*p
= (__u32
*) mpb
;
383 sum
+= __le32_to_cpu(*p
);
387 return sum
- __le32_to_cpu(mpb
->check_sum
);
390 static size_t sizeof_imsm_map(struct imsm_map
*map
)
392 return sizeof(struct imsm_map
) + sizeof(__u32
) * (map
->num_members
- 1);
395 struct imsm_map
*get_imsm_map(struct imsm_dev
*dev
, int second_map
)
397 struct imsm_map
*map
= &dev
->vol
.map
[0];
399 if (second_map
&& !dev
->vol
.migr_state
)
401 else if (second_map
) {
404 return ptr
+ sizeof_imsm_map(map
);
410 /* return the size of the device.
411 * migr_state increases the returned size if map[0] were to be duplicated
413 static size_t sizeof_imsm_dev(struct imsm_dev
*dev
, int migr_state
)
415 size_t size
= sizeof(*dev
) - sizeof(struct imsm_map
) +
416 sizeof_imsm_map(get_imsm_map(dev
, 0));
418 /* migrating means an additional map */
419 if (dev
->vol
.migr_state
)
420 size
+= sizeof_imsm_map(get_imsm_map(dev
, 1));
422 size
+= sizeof_imsm_map(get_imsm_map(dev
, 0));
428 /* retrieve disk serial number list from a metadata update */
429 static struct disk_info
*get_disk_info(struct imsm_update_create_array
*update
)
432 struct disk_info
*inf
;
434 inf
= u
+ sizeof(*update
) - sizeof(struct imsm_dev
) +
435 sizeof_imsm_dev(&update
->dev
, 0);
441 static struct imsm_dev
*__get_imsm_dev(struct imsm_super
*mpb
, __u8 index
)
447 if (index
>= mpb
->num_raid_devs
)
450 /* devices start after all disks */
451 offset
= ((void *) &mpb
->disk
[mpb
->num_disks
]) - _mpb
;
453 for (i
= 0; i
<= index
; i
++)
455 return _mpb
+ offset
;
457 offset
+= sizeof_imsm_dev(_mpb
+ offset
, 0);
462 static struct imsm_dev
*get_imsm_dev(struct intel_super
*super
, __u8 index
)
464 struct intel_dev
*dv
;
466 if (index
>= super
->anchor
->num_raid_devs
)
468 for (dv
= super
->devlist
; dv
; dv
= dv
->next
)
469 if (dv
->index
== index
)
474 static __u32
get_imsm_ord_tbl_ent(struct imsm_dev
*dev
, int slot
)
476 struct imsm_map
*map
;
478 if (dev
->vol
.migr_state
)
479 map
= get_imsm_map(dev
, 1);
481 map
= get_imsm_map(dev
, 0);
483 /* top byte identifies disk under rebuild */
484 return __le32_to_cpu(map
->disk_ord_tbl
[slot
]);
487 #define ord_to_idx(ord) (((ord) << 8) >> 8)
488 static __u32
get_imsm_disk_idx(struct imsm_dev
*dev
, int slot
)
490 __u32 ord
= get_imsm_ord_tbl_ent(dev
, slot
);
492 return ord_to_idx(ord
);
495 static void set_imsm_ord_tbl_ent(struct imsm_map
*map
, int slot
, __u32 ord
)
497 map
->disk_ord_tbl
[slot
] = __cpu_to_le32(ord
);
500 static int get_imsm_disk_slot(struct imsm_map
*map
, unsigned idx
)
505 for (slot
= 0; slot
< map
->num_members
; slot
++) {
506 ord
= __le32_to_cpu(map
->disk_ord_tbl
[slot
]);
507 if (ord_to_idx(ord
) == idx
)
514 static int get_imsm_raid_level(struct imsm_map
*map
)
516 if (map
->raid_level
== 1) {
517 if (map
->num_members
== 2)
523 return map
->raid_level
;
526 static int cmp_extent(const void *av
, const void *bv
)
528 const struct extent
*a
= av
;
529 const struct extent
*b
= bv
;
530 if (a
->start
< b
->start
)
532 if (a
->start
> b
->start
)
537 static int count_memberships(struct dl
*dl
, struct intel_super
*super
)
542 for (i
= 0; i
< super
->anchor
->num_raid_devs
; i
++) {
543 struct imsm_dev
*dev
= get_imsm_dev(super
, i
);
544 struct imsm_map
*map
= get_imsm_map(dev
, 0);
546 if (get_imsm_disk_slot(map
, dl
->index
) >= 0)
553 static struct extent
*get_extents(struct intel_super
*super
, struct dl
*dl
)
555 /* find a list of used extents on the given physical device */
556 struct extent
*rv
, *e
;
558 int memberships
= count_memberships(dl
, super
);
559 __u32 reservation
= MPB_SECTOR_CNT
+ IMSM_RESERVED_SECTORS
;
561 rv
= malloc(sizeof(struct extent
) * (memberships
+ 1));
566 for (i
= 0; i
< super
->anchor
->num_raid_devs
; i
++) {
567 struct imsm_dev
*dev
= get_imsm_dev(super
, i
);
568 struct imsm_map
*map
= get_imsm_map(dev
, 0);
570 if (get_imsm_disk_slot(map
, dl
->index
) >= 0) {
571 e
->start
= __le32_to_cpu(map
->pba_of_lba0
);
572 e
->size
= __le32_to_cpu(map
->blocks_per_member
);
576 qsort(rv
, memberships
, sizeof(*rv
), cmp_extent
);
578 /* determine the start of the metadata
579 * when no raid devices are defined use the default
580 * ...otherwise allow the metadata to truncate the value
581 * as is the case with older versions of imsm
584 struct extent
*last
= &rv
[memberships
- 1];
587 remainder
= __le32_to_cpu(dl
->disk
.total_blocks
) -
588 (last
->start
+ last
->size
);
589 /* round down to 1k block to satisfy precision of the kernel
593 /* make sure remainder is still sane */
594 if (remainder
< (unsigned)ROUND_UP(super
->len
, 512) >> 9)
595 remainder
= ROUND_UP(super
->len
, 512) >> 9;
596 if (reservation
> remainder
)
597 reservation
= remainder
;
599 e
->start
= __le32_to_cpu(dl
->disk
.total_blocks
) - reservation
;
604 /* try to determine how much space is reserved for metadata from
605 * the last get_extents() entry, otherwise fallback to the
608 static __u32
imsm_reserved_sectors(struct intel_super
*super
, struct dl
*dl
)
614 /* for spares just return a minimal reservation which will grow
615 * once the spare is picked up by an array
618 return MPB_SECTOR_CNT
;
620 e
= get_extents(super
, dl
);
622 return MPB_SECTOR_CNT
+ IMSM_RESERVED_SECTORS
;
624 /* scroll to last entry */
625 for (i
= 0; e
[i
].size
; i
++)
628 rv
= __le32_to_cpu(dl
->disk
.total_blocks
) - e
[i
].start
;
635 static int is_spare(struct imsm_disk
*disk
)
637 return (disk
->status
& SPARE_DISK
) == SPARE_DISK
;
640 static int is_configured(struct imsm_disk
*disk
)
642 return (disk
->status
& CONFIGURED_DISK
) == CONFIGURED_DISK
;
645 static int is_failed(struct imsm_disk
*disk
)
647 return (disk
->status
& FAILED_DISK
) == FAILED_DISK
;
651 static __u64
blocks_per_migr_unit(struct imsm_dev
*dev
);
653 static void print_imsm_dev(struct imsm_dev
*dev
, char *uuid
, int disk_idx
)
657 struct imsm_map
*map
= get_imsm_map(dev
, 0);
661 printf("[%.16s]:\n", dev
->volume
);
662 printf(" UUID : %s\n", uuid
);
663 printf(" RAID Level : %d\n", get_imsm_raid_level(map
));
664 printf(" Members : %d\n", map
->num_members
);
665 printf(" Slots : [");
666 for (i
= 0; i
< map
->num_members
; i
++) {
667 ord
= get_imsm_ord_tbl_ent(dev
, i
);
668 printf("%s", ord
& IMSM_ORD_REBUILD
? "_" : "U");
671 slot
= get_imsm_disk_slot(map
, disk_idx
);
673 ord
= get_imsm_ord_tbl_ent(dev
, slot
);
674 printf(" This Slot : %d%s\n", slot
,
675 ord
& IMSM_ORD_REBUILD
? " (out-of-sync)" : "");
677 printf(" This Slot : ?\n");
678 sz
= __le32_to_cpu(dev
->size_high
);
680 sz
+= __le32_to_cpu(dev
->size_low
);
681 printf(" Array Size : %llu%s\n", (unsigned long long)sz
,
682 human_size(sz
* 512));
683 sz
= __le32_to_cpu(map
->blocks_per_member
);
684 printf(" Per Dev Size : %llu%s\n", (unsigned long long)sz
,
685 human_size(sz
* 512));
686 printf(" Sector Offset : %u\n",
687 __le32_to_cpu(map
->pba_of_lba0
));
688 printf(" Num Stripes : %u\n",
689 __le32_to_cpu(map
->num_data_stripes
));
690 printf(" Chunk Size : %u KiB\n",
691 __le16_to_cpu(map
->blocks_per_strip
) / 2);
692 printf(" Reserved : %d\n", __le32_to_cpu(dev
->reserved_blocks
));
693 printf(" Migrate State : ");
694 if (dev
->vol
.migr_state
) {
695 if (migr_type(dev
) == MIGR_INIT
)
696 printf("initialize\n");
697 else if (migr_type(dev
) == MIGR_REBUILD
)
699 else if (migr_type(dev
) == MIGR_VERIFY
)
701 else if (migr_type(dev
) == MIGR_GEN_MIGR
)
702 printf("general migration\n");
703 else if (migr_type(dev
) == MIGR_STATE_CHANGE
)
704 printf("state change\n");
705 else if (migr_type(dev
) == MIGR_REPAIR
)
708 printf("<unknown:%d>\n", migr_type(dev
));
711 printf(" Map State : %s", map_state_str
[map
->map_state
]);
712 if (dev
->vol
.migr_state
) {
713 struct imsm_map
*map
= get_imsm_map(dev
, 1);
715 printf(" <-- %s", map_state_str
[map
->map_state
]);
716 printf("\n Checkpoint : %u (%llu)",
717 __le32_to_cpu(dev
->vol
.curr_migr_unit
),
718 (unsigned long long)blocks_per_migr_unit(dev
));
721 printf(" Dirty State : %s\n", dev
->vol
.dirty
? "dirty" : "clean");
724 static void print_imsm_disk(struct imsm_super
*mpb
, int index
, __u32 reserved
)
726 struct imsm_disk
*disk
= __get_imsm_disk(mpb
, index
);
727 char str
[MAX_RAID_SERIAL_LEN
+ 1];
730 if (index
< 0 || !disk
)
734 snprintf(str
, MAX_RAID_SERIAL_LEN
+ 1, "%s", disk
->serial
);
735 printf(" Disk%02d Serial : %s\n", index
, str
);
736 printf(" State :%s%s%s\n", is_spare(disk
) ? " spare" : "",
737 is_configured(disk
) ? " active" : "",
738 is_failed(disk
) ? " failed" : "");
739 printf(" Id : %08x\n", __le32_to_cpu(disk
->scsi_id
));
740 sz
= __le32_to_cpu(disk
->total_blocks
) - reserved
;
741 printf(" Usable Size : %llu%s\n", (unsigned long long)sz
,
742 human_size(sz
* 512));
745 static void getinfo_super_imsm(struct supertype
*st
, struct mdinfo
*info
, char *map
);
747 static void examine_super_imsm(struct supertype
*st
, char *homehost
)
749 struct intel_super
*super
= st
->sb
;
750 struct imsm_super
*mpb
= super
->anchor
;
751 char str
[MAX_SIGNATURE_LENGTH
];
756 __u32 reserved
= imsm_reserved_sectors(super
, super
->disks
);
759 snprintf(str
, MPB_SIG_LEN
, "%s", mpb
->sig
);
760 printf(" Magic : %s\n", str
);
761 snprintf(str
, strlen(MPB_VERSION_RAID0
), "%s", get_imsm_version(mpb
));
762 printf(" Version : %s\n", get_imsm_version(mpb
));
763 printf(" Orig Family : %08x\n", __le32_to_cpu(mpb
->orig_family_num
));
764 printf(" Family : %08x\n", __le32_to_cpu(mpb
->family_num
));
765 printf(" Generation : %08x\n", __le32_to_cpu(mpb
->generation_num
));
766 getinfo_super_imsm(st
, &info
, NULL
);
767 fname_from_uuid(st
, &info
, nbuf
, ':');
768 printf(" UUID : %s\n", nbuf
+ 5);
769 sum
= __le32_to_cpu(mpb
->check_sum
);
770 printf(" Checksum : %08x %s\n", sum
,
771 __gen_imsm_checksum(mpb
) == sum
? "correct" : "incorrect");
772 printf(" MPB Sectors : %d\n", mpb_sectors(mpb
));
773 printf(" Disks : %d\n", mpb
->num_disks
);
774 printf(" RAID Devices : %d\n", mpb
->num_raid_devs
);
775 print_imsm_disk(mpb
, super
->disks
->index
, reserved
);
776 if (super
->bbm_log
) {
777 struct bbm_log
*log
= super
->bbm_log
;
780 printf("Bad Block Management Log:\n");
781 printf(" Log Size : %d\n", __le32_to_cpu(mpb
->bbm_log_size
));
782 printf(" Signature : %x\n", __le32_to_cpu(log
->signature
));
783 printf(" Entry Count : %d\n", __le32_to_cpu(log
->entry_count
));
784 printf(" Spare Blocks : %d\n", __le32_to_cpu(log
->reserved_spare_block_count
));
785 printf(" First Spare : %llx\n",
786 (unsigned long long) __le64_to_cpu(log
->first_spare_lba
));
788 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
790 struct imsm_dev
*dev
= __get_imsm_dev(mpb
, i
);
792 super
->current_vol
= i
;
793 getinfo_super_imsm(st
, &info
, NULL
);
794 fname_from_uuid(st
, &info
, nbuf
, ':');
795 print_imsm_dev(dev
, nbuf
+ 5, super
->disks
->index
);
797 for (i
= 0; i
< mpb
->num_disks
; i
++) {
798 if (i
== super
->disks
->index
)
800 print_imsm_disk(mpb
, i
, reserved
);
804 static void brief_examine_super_imsm(struct supertype
*st
, int verbose
)
806 /* We just write a generic IMSM ARRAY entry */
809 struct intel_super
*super
= st
->sb
;
811 if (!super
->anchor
->num_raid_devs
) {
812 printf("ARRAY metadata=imsm\n");
816 getinfo_super_imsm(st
, &info
, NULL
);
817 fname_from_uuid(st
, &info
, nbuf
, ':');
818 printf("ARRAY metadata=imsm UUID=%s\n", nbuf
+ 5);
821 static void brief_examine_subarrays_imsm(struct supertype
*st
, int verbose
)
823 /* We just write a generic IMSM ARRAY entry */
827 struct intel_super
*super
= st
->sb
;
830 if (!super
->anchor
->num_raid_devs
)
833 getinfo_super_imsm(st
, &info
, NULL
);
834 fname_from_uuid(st
, &info
, nbuf
, ':');
835 for (i
= 0; i
< super
->anchor
->num_raid_devs
; i
++) {
836 struct imsm_dev
*dev
= get_imsm_dev(super
, i
);
838 super
->current_vol
= i
;
839 getinfo_super_imsm(st
, &info
, NULL
);
840 fname_from_uuid(st
, &info
, nbuf1
, ':');
841 printf("ARRAY /dev/md/%.16s container=%s member=%d UUID=%s\n",
842 dev
->volume
, nbuf
+ 5, i
, nbuf1
+ 5);
846 static void export_examine_super_imsm(struct supertype
*st
)
848 struct intel_super
*super
= st
->sb
;
849 struct imsm_super
*mpb
= super
->anchor
;
853 getinfo_super_imsm(st
, &info
, NULL
);
854 fname_from_uuid(st
, &info
, nbuf
, ':');
855 printf("MD_METADATA=imsm\n");
856 printf("MD_LEVEL=container\n");
857 printf("MD_UUID=%s\n", nbuf
+5);
858 printf("MD_DEVICES=%u\n", mpb
->num_disks
);
861 static void detail_super_imsm(struct supertype
*st
, char *homehost
)
866 getinfo_super_imsm(st
, &info
, NULL
);
867 fname_from_uuid(st
, &info
, nbuf
, ':');
868 printf("\n UUID : %s\n", nbuf
+ 5);
871 static void brief_detail_super_imsm(struct supertype
*st
)
875 getinfo_super_imsm(st
, &info
, NULL
);
876 fname_from_uuid(st
, &info
, nbuf
, ':');
877 printf(" UUID=%s", nbuf
+ 5);
880 static int imsm_read_serial(int fd
, char *devname
, __u8
*serial
);
881 static void fd2devname(int fd
, char *name
);
883 static int imsm_enumerate_ports(const char *hba_path
, int port_count
, int host_base
, int verbose
)
885 /* dump an unsorted list of devices attached to ahci, as well as
886 * non-connected ports
888 int hba_len
= strlen(hba_path
) + 1;
893 unsigned long port_mask
= (1 << port_count
) - 1;
895 if (port_count
> (int)sizeof(port_mask
) * 8) {
897 fprintf(stderr
, Name
": port_count %d out of range\n", port_count
);
901 /* scroll through /sys/dev/block looking for devices attached to
904 dir
= opendir("/sys/dev/block");
905 for (ent
= dir
? readdir(dir
) : NULL
; ent
; ent
= readdir(dir
)) {
916 if (sscanf(ent
->d_name
, "%d:%d", &major
, &minor
) != 2)
918 path
= devt_to_devpath(makedev(major
, minor
));
921 if (!path_attached_to_hba(path
, hba_path
)) {
927 /* retrieve the scsi device type */
928 if (asprintf(&device
, "/sys/dev/block/%d:%d/device/xxxxxxx", major
, minor
) < 0) {
930 fprintf(stderr
, Name
": failed to allocate 'device'\n");
934 sprintf(device
, "/sys/dev/block/%d:%d/device/type", major
, minor
);
935 if (load_sys(device
, buf
) != 0) {
937 fprintf(stderr
, Name
": failed to read device type for %s\n",
943 type
= strtoul(buf
, NULL
, 10);
945 /* if it's not a disk print the vendor and model */
946 if (!(type
== 0 || type
== 7 || type
== 14)) {
949 sprintf(device
, "/sys/dev/block/%d:%d/device/vendor", major
, minor
);
950 if (load_sys(device
, buf
) == 0) {
951 strncpy(vendor
, buf
, sizeof(vendor
));
952 vendor
[sizeof(vendor
) - 1] = '\0';
953 c
= (char *) &vendor
[sizeof(vendor
) - 1];
954 while (isspace(*c
) || *c
== '\0')
958 sprintf(device
, "/sys/dev/block/%d:%d/device/model", major
, minor
);
959 if (load_sys(device
, buf
) == 0) {
960 strncpy(model
, buf
, sizeof(model
));
961 model
[sizeof(model
) - 1] = '\0';
962 c
= (char *) &model
[sizeof(model
) - 1];
963 while (isspace(*c
) || *c
== '\0')
967 if (vendor
[0] && model
[0])
968 sprintf(buf
, "%.64s %.64s", vendor
, model
);
970 switch (type
) { /* numbers from hald/linux/device.c */
971 case 1: sprintf(buf
, "tape"); break;
972 case 2: sprintf(buf
, "printer"); break;
973 case 3: sprintf(buf
, "processor"); break;
975 case 5: sprintf(buf
, "cdrom"); break;
976 case 6: sprintf(buf
, "scanner"); break;
977 case 8: sprintf(buf
, "media_changer"); break;
978 case 9: sprintf(buf
, "comm"); break;
979 case 12: sprintf(buf
, "raid"); break;
980 default: sprintf(buf
, "unknown");
986 /* chop device path to 'host%d' and calculate the port number */
987 c
= strchr(&path
[hba_len
], '/');
990 fprintf(stderr
, Name
": %s - invalid path name\n", path
+ hba_len
);
995 if (sscanf(&path
[hba_len
], "host%d", &port
) == 1)
999 *c
= '/'; /* repair the full string */
1000 fprintf(stderr
, Name
": failed to determine port number for %s\n",
1007 /* mark this port as used */
1008 port_mask
&= ~(1 << port
);
1010 /* print out the device information */
1012 printf(" Port%d : - non-disk device (%s) -\n", port
, buf
);
1016 fd
= dev_open(ent
->d_name
, O_RDONLY
);
1018 printf(" Port%d : - disk info unavailable -\n", port
);
1020 fd2devname(fd
, buf
);
1021 printf(" Port%d : %s", port
, buf
);
1022 if (imsm_read_serial(fd
, NULL
, (__u8
*) buf
) == 0)
1023 printf(" (%s)\n", buf
);
1038 for (i
= 0; i
< port_count
; i
++)
1039 if (port_mask
& (1 << i
))
1040 printf(" Port%d : - no device attached -\n", i
);
1046 static int detail_platform_imsm(int verbose
, int enumerate_only
)
1048 /* There are two components to imsm platform support, the ahci SATA
1049 * controller and the option-rom. To find the SATA controller we
1050 * simply look in /sys/bus/pci/drivers/ahci to see if an ahci
1051 * controller with the Intel vendor id is present. This approach
1052 * allows mdadm to leverage the kernel's ahci detection logic, with the
1053 * caveat that if ahci.ko is not loaded mdadm will not be able to
1054 * detect platform raid capabilities. The option-rom resides in a
1055 * platform "Adapter ROM". We scan for its signature to retrieve the
1056 * platform capabilities. If raid support is disabled in the BIOS the
1057 * option-rom capability structure will not be available.
1059 const struct imsm_orom
*orom
;
1060 struct sys_dev
*list
, *hba
;
1063 const char *hba_path
;
1067 if (enumerate_only
) {
1068 if (check_env("IMSM_NO_PLATFORM") || find_imsm_orom())
1073 list
= find_driver_devices("pci", "ahci");
1074 for (hba
= list
; hba
; hba
= hba
->next
)
1075 if (devpath_to_vendor(hba
->path
) == 0x8086)
1080 fprintf(stderr
, Name
": unable to find active ahci controller\n");
1081 free_sys_dev(&list
);
1084 fprintf(stderr
, Name
": found Intel SATA AHCI Controller\n");
1085 hba_path
= hba
->path
;
1087 free_sys_dev(&list
);
1089 orom
= find_imsm_orom();
1092 fprintf(stderr
, Name
": imsm option-rom not found\n");
1096 printf(" Platform : Intel(R) Matrix Storage Manager\n");
1097 printf(" Version : %d.%d.%d.%d\n", orom
->major_ver
, orom
->minor_ver
,
1098 orom
->hotfix_ver
, orom
->build
);
1099 printf(" RAID Levels :%s%s%s%s%s\n",
1100 imsm_orom_has_raid0(orom
) ? " raid0" : "",
1101 imsm_orom_has_raid1(orom
) ? " raid1" : "",
1102 imsm_orom_has_raid1e(orom
) ? " raid1e" : "",
1103 imsm_orom_has_raid10(orom
) ? " raid10" : "",
1104 imsm_orom_has_raid5(orom
) ? " raid5" : "");
1105 printf(" Chunk Sizes :%s%s%s%s%s%s%s%s%s%s%s%s%s%s%s%s\n",
1106 imsm_orom_has_chunk(orom
, 2) ? " 2k" : "",
1107 imsm_orom_has_chunk(orom
, 4) ? " 4k" : "",
1108 imsm_orom_has_chunk(orom
, 8) ? " 8k" : "",
1109 imsm_orom_has_chunk(orom
, 16) ? " 16k" : "",
1110 imsm_orom_has_chunk(orom
, 32) ? " 32k" : "",
1111 imsm_orom_has_chunk(orom
, 64) ? " 64k" : "",
1112 imsm_orom_has_chunk(orom
, 128) ? " 128k" : "",
1113 imsm_orom_has_chunk(orom
, 256) ? " 256k" : "",
1114 imsm_orom_has_chunk(orom
, 512) ? " 512k" : "",
1115 imsm_orom_has_chunk(orom
, 1024*1) ? " 1M" : "",
1116 imsm_orom_has_chunk(orom
, 1024*2) ? " 2M" : "",
1117 imsm_orom_has_chunk(orom
, 1024*4) ? " 4M" : "",
1118 imsm_orom_has_chunk(orom
, 1024*8) ? " 8M" : "",
1119 imsm_orom_has_chunk(orom
, 1024*16) ? " 16M" : "",
1120 imsm_orom_has_chunk(orom
, 1024*32) ? " 32M" : "",
1121 imsm_orom_has_chunk(orom
, 1024*64) ? " 64M" : "");
1122 printf(" Max Disks : %d\n", orom
->tds
);
1123 printf(" Max Volumes : %d\n", orom
->vpa
);
1124 printf(" I/O Controller : %s\n", hba_path
);
1126 /* find the smallest scsi host number to determine a port number base */
1127 dir
= opendir(hba_path
);
1128 for (ent
= dir
? readdir(dir
) : NULL
; ent
; ent
= readdir(dir
)) {
1131 if (sscanf(ent
->d_name
, "host%d", &host
) != 1)
1133 if (port_count
== 0)
1135 else if (host
< host_base
)
1138 if (host
+ 1 > port_count
+ host_base
)
1139 port_count
= host
+ 1 - host_base
;
1145 if (!port_count
|| imsm_enumerate_ports(hba_path
, port_count
,
1146 host_base
, verbose
) != 0) {
1148 fprintf(stderr
, Name
": failed to enumerate ports\n");
1156 static int match_home_imsm(struct supertype
*st
, char *homehost
)
1158 /* the imsm metadata format does not specify any host
1159 * identification information. We return -1 since we can never
1160 * confirm nor deny whether a given array is "meant" for this
1161 * host. We rely on compare_super and the 'family_num' fields to
1162 * exclude member disks that do not belong, and we rely on
1163 * mdadm.conf to specify the arrays that should be assembled.
1164 * Auto-assembly may still pick up "foreign" arrays.
1170 static void uuid_from_super_imsm(struct supertype
*st
, int uuid
[4])
1172 /* The uuid returned here is used for:
1173 * uuid to put into bitmap file (Create, Grow)
1174 * uuid for backup header when saving critical section (Grow)
1175 * comparing uuids when re-adding a device into an array
1176 * In these cases the uuid required is that of the data-array,
1177 * not the device-set.
1178 * uuid to recognise same set when adding a missing device back
1179 * to an array. This is a uuid for the device-set.
1181 * For each of these we can make do with a truncated
1182 * or hashed uuid rather than the original, as long as
1184 * In each case the uuid required is that of the data-array,
1185 * not the device-set.
1187 /* imsm does not track uuid's so we synthesis one using sha1 on
1188 * - The signature (Which is constant for all imsm array, but no matter)
1189 * - the orig_family_num of the container
1190 * - the index number of the volume
1191 * - the 'serial' number of the volume.
1192 * Hopefully these are all constant.
1194 struct intel_super
*super
= st
->sb
;
1197 struct sha1_ctx ctx
;
1198 struct imsm_dev
*dev
= NULL
;
1201 /* some mdadm versions failed to set ->orig_family_num, in which
1202 * case fall back to ->family_num. orig_family_num will be
1203 * fixed up with the first metadata update.
1205 family_num
= super
->anchor
->orig_family_num
;
1206 if (family_num
== 0)
1207 family_num
= super
->anchor
->family_num
;
1208 sha1_init_ctx(&ctx
);
1209 sha1_process_bytes(super
->anchor
->sig
, MPB_SIG_LEN
, &ctx
);
1210 sha1_process_bytes(&family_num
, sizeof(__u32
), &ctx
);
1211 if (super
->current_vol
>= 0)
1212 dev
= get_imsm_dev(super
, super
->current_vol
);
1214 __u32 vol
= super
->current_vol
;
1215 sha1_process_bytes(&vol
, sizeof(vol
), &ctx
);
1216 sha1_process_bytes(dev
->volume
, MAX_RAID_SERIAL_LEN
, &ctx
);
1218 sha1_finish_ctx(&ctx
, buf
);
1219 memcpy(uuid
, buf
, 4*4);
1224 get_imsm_numerical_version(struct imsm_super
*mpb
, int *m
, int *p
)
1226 __u8
*v
= get_imsm_version(mpb
);
1227 __u8
*end
= mpb
->sig
+ MAX_SIGNATURE_LENGTH
;
1228 char major
[] = { 0, 0, 0 };
1229 char minor
[] = { 0 ,0, 0 };
1230 char patch
[] = { 0, 0, 0 };
1231 char *ver_parse
[] = { major
, minor
, patch
};
1235 while (*v
!= '\0' && v
< end
) {
1236 if (*v
!= '.' && j
< 2)
1237 ver_parse
[i
][j
++] = *v
;
1245 *m
= strtol(minor
, NULL
, 0);
1246 *p
= strtol(patch
, NULL
, 0);
1250 static __u32
migr_strip_blocks_resync(struct imsm_dev
*dev
)
1252 /* migr_strip_size when repairing or initializing parity */
1253 struct imsm_map
*map
= get_imsm_map(dev
, 0);
1254 __u32 chunk
= __le32_to_cpu(map
->blocks_per_strip
);
1256 switch (get_imsm_raid_level(map
)) {
1261 return 128*1024 >> 9;
1265 static __u32
migr_strip_blocks_rebuild(struct imsm_dev
*dev
)
1267 /* migr_strip_size when rebuilding a degraded disk, no idea why
1268 * this is different than migr_strip_size_resync(), but it's good
1271 struct imsm_map
*map
= get_imsm_map(dev
, 1);
1272 __u32 chunk
= __le32_to_cpu(map
->blocks_per_strip
);
1274 switch (get_imsm_raid_level(map
)) {
1277 if (map
->num_members
% map
->num_domains
== 0)
1278 return 128*1024 >> 9;
1282 return max((__u32
) 64*1024 >> 9, chunk
);
1284 return 128*1024 >> 9;
1288 static __u32
num_stripes_per_unit_resync(struct imsm_dev
*dev
)
1290 struct imsm_map
*lo
= get_imsm_map(dev
, 0);
1291 struct imsm_map
*hi
= get_imsm_map(dev
, 1);
1292 __u32 lo_chunk
= __le32_to_cpu(lo
->blocks_per_strip
);
1293 __u32 hi_chunk
= __le32_to_cpu(hi
->blocks_per_strip
);
1295 return max((__u32
) 1, hi_chunk
/ lo_chunk
);
1298 static __u32
num_stripes_per_unit_rebuild(struct imsm_dev
*dev
)
1300 struct imsm_map
*lo
= get_imsm_map(dev
, 0);
1301 int level
= get_imsm_raid_level(lo
);
1303 if (level
== 1 || level
== 10) {
1304 struct imsm_map
*hi
= get_imsm_map(dev
, 1);
1306 return hi
->num_domains
;
1308 return num_stripes_per_unit_resync(dev
);
1311 static __u8
imsm_num_data_members(struct imsm_dev
*dev
)
1313 /* named 'imsm_' because raid0, raid1 and raid10
1314 * counter-intuitively have the same number of data disks
1316 struct imsm_map
*map
= get_imsm_map(dev
, 0);
1318 switch (get_imsm_raid_level(map
)) {
1322 return map
->num_members
;
1324 return map
->num_members
- 1;
1326 dprintf("%s: unsupported raid level\n", __func__
);
1331 static __u32
parity_segment_depth(struct imsm_dev
*dev
)
1333 struct imsm_map
*map
= get_imsm_map(dev
, 0);
1334 __u32 chunk
= __le32_to_cpu(map
->blocks_per_strip
);
1336 switch(get_imsm_raid_level(map
)) {
1339 return chunk
* map
->num_domains
;
1341 return chunk
* map
->num_members
;
1347 static __u32
map_migr_block(struct imsm_dev
*dev
, __u32 block
)
1349 struct imsm_map
*map
= get_imsm_map(dev
, 1);
1350 __u32 chunk
= __le32_to_cpu(map
->blocks_per_strip
);
1351 __u32 strip
= block
/ chunk
;
1353 switch (get_imsm_raid_level(map
)) {
1356 __u32 vol_strip
= (strip
* map
->num_domains
) + 1;
1357 __u32 vol_stripe
= vol_strip
/ map
->num_members
;
1359 return vol_stripe
* chunk
+ block
% chunk
;
1361 __u32 stripe
= strip
/ (map
->num_members
- 1);
1363 return stripe
* chunk
+ block
% chunk
;
1370 static __u64
blocks_per_migr_unit(struct imsm_dev
*dev
)
1372 /* calculate the conversion factor between per member 'blocks'
1373 * (md/{resync,rebuild}_start) and imsm migration units, return
1374 * 0 for the 'not migrating' and 'unsupported migration' cases
1376 if (!dev
->vol
.migr_state
)
1379 switch (migr_type(dev
)) {
1383 struct imsm_map
*map
= get_imsm_map(dev
, 0);
1384 __u32 stripes_per_unit
;
1385 __u32 blocks_per_unit
;
1394 /* yes, this is really the translation of migr_units to
1395 * per-member blocks in the 'resync' case
1397 stripes_per_unit
= num_stripes_per_unit_resync(dev
);
1398 migr_chunk
= migr_strip_blocks_resync(dev
);
1399 disks
= imsm_num_data_members(dev
);
1400 blocks_per_unit
= stripes_per_unit
* migr_chunk
* disks
;
1401 stripe
= __le32_to_cpu(map
->blocks_per_strip
) * disks
;
1402 segment
= blocks_per_unit
/ stripe
;
1403 block_rel
= blocks_per_unit
- segment
* stripe
;
1404 parity_depth
= parity_segment_depth(dev
);
1405 block_map
= map_migr_block(dev
, block_rel
);
1406 return block_map
+ parity_depth
* segment
;
1408 case MIGR_REBUILD
: {
1409 __u32 stripes_per_unit
;
1412 stripes_per_unit
= num_stripes_per_unit_rebuild(dev
);
1413 migr_chunk
= migr_strip_blocks_rebuild(dev
);
1414 return migr_chunk
* stripes_per_unit
;
1417 case MIGR_STATE_CHANGE
:
1423 static int imsm_level_to_layout(int level
)
1431 return ALGORITHM_LEFT_ASYMMETRIC
;
1438 static void getinfo_super_imsm_volume(struct supertype
*st
, struct mdinfo
*info
, char *dmap
)
1440 struct intel_super
*super
= st
->sb
;
1441 struct imsm_dev
*dev
= get_imsm_dev(super
, super
->current_vol
);
1442 struct imsm_map
*map
= get_imsm_map(dev
, 0);
1445 int map_disks
= info
->array
.raid_disks
;
1447 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
1448 if (dl
->raiddisk
== info
->disk
.raid_disk
)
1450 info
->container_member
= super
->current_vol
;
1451 info
->array
.raid_disks
= map
->num_members
;
1452 info
->array
.level
= get_imsm_raid_level(map
);
1453 info
->array
.layout
= imsm_level_to_layout(info
->array
.level
);
1454 info
->array
.md_minor
= -1;
1455 info
->array
.ctime
= 0;
1456 info
->array
.utime
= 0;
1457 info
->array
.chunk_size
= __le16_to_cpu(map
->blocks_per_strip
) << 9;
1458 info
->array
.state
= !dev
->vol
.dirty
;
1459 info
->custom_array_size
= __le32_to_cpu(dev
->size_high
);
1460 info
->custom_array_size
<<= 32;
1461 info
->custom_array_size
|= __le32_to_cpu(dev
->size_low
);
1463 info
->disk
.major
= 0;
1464 info
->disk
.minor
= 0;
1466 info
->disk
.major
= dl
->major
;
1467 info
->disk
.minor
= dl
->minor
;
1470 info
->data_offset
= __le32_to_cpu(map
->pba_of_lba0
);
1471 info
->component_size
= __le32_to_cpu(map
->blocks_per_member
);
1472 memset(info
->uuid
, 0, sizeof(info
->uuid
));
1473 info
->recovery_start
= MaxSector
;
1474 info
->reshape_active
= 0;
1476 if (map
->map_state
== IMSM_T_STATE_UNINITIALIZED
|| dev
->vol
.dirty
) {
1477 info
->resync_start
= 0;
1478 } else if (dev
->vol
.migr_state
) {
1479 switch (migr_type(dev
)) {
1482 __u64 blocks_per_unit
= blocks_per_migr_unit(dev
);
1483 __u64 units
= __le32_to_cpu(dev
->vol
.curr_migr_unit
);
1485 info
->resync_start
= blocks_per_unit
* units
;
1489 /* we could emulate the checkpointing of
1490 * 'sync_action=check' migrations, but for now
1491 * we just immediately complete them
1494 /* this is handled by container_content_imsm() */
1496 case MIGR_STATE_CHANGE
:
1497 /* FIXME handle other migrations */
1499 /* we are not dirty, so... */
1500 info
->resync_start
= MaxSector
;
1503 info
->resync_start
= MaxSector
;
1505 strncpy(info
->name
, (char *) dev
->volume
, MAX_RAID_SERIAL_LEN
);
1506 info
->name
[MAX_RAID_SERIAL_LEN
] = 0;
1508 info
->array
.major_version
= -1;
1509 info
->array
.minor_version
= -2;
1510 devname
= devnum2devname(st
->container_dev
);
1511 *info
->text_version
= '\0';
1513 sprintf(info
->text_version
, "/%s/%d", devname
, info
->container_member
);
1515 info
->safe_mode_delay
= 4000; /* 4 secs like the Matrix driver */
1516 uuid_from_super_imsm(st
, info
->uuid
);
1520 for (i
=0; i
<map_disks
; i
++) {
1522 if (i
< info
->array
.raid_disks
) {
1523 struct imsm_disk
*dsk
;
1524 j
= get_imsm_disk_idx(dev
, i
);
1525 dsk
= get_imsm_disk(super
, j
);
1526 if (dsk
&& (dsk
->status
& CONFIGURED_DISK
))
1533 /* check the config file to see if we can return a real uuid for this spare */
1534 static void fixup_container_spare_uuid(struct mdinfo
*inf
)
1536 struct mddev_ident
*array_list
;
1538 if (inf
->array
.level
!= LEVEL_CONTAINER
||
1539 memcmp(inf
->uuid
, uuid_match_any
, sizeof(int[4])) != 0)
1542 array_list
= conf_get_ident(NULL
);
1544 for (; array_list
; array_list
= array_list
->next
) {
1545 if (array_list
->uuid_set
) {
1546 struct supertype
*_sst
; /* spare supertype */
1547 struct supertype
*_cst
; /* container supertype */
1549 _cst
= array_list
->st
;
1551 _sst
= _cst
->ss
->match_metadata_desc(inf
->text_version
);
1556 memcpy(inf
->uuid
, array_list
->uuid
, sizeof(int[4]));
1565 static __u8
imsm_check_degraded(struct intel_super
*super
, struct imsm_dev
*dev
, int failed
);
1566 static int imsm_count_failed(struct intel_super
*super
, struct imsm_dev
*dev
);
1568 static struct imsm_disk
*get_imsm_missing(struct intel_super
*super
, __u8 index
)
1572 for (d
= super
->missing
; d
; d
= d
->next
)
1573 if (d
->index
== index
)
1578 static void getinfo_super_imsm(struct supertype
*st
, struct mdinfo
*info
, char *map
)
1580 struct intel_super
*super
= st
->sb
;
1581 struct imsm_disk
*disk
;
1582 int map_disks
= info
->array
.raid_disks
;
1583 int max_enough
= -1;
1585 struct imsm_super
*mpb
;
1587 if (super
->current_vol
>= 0) {
1588 getinfo_super_imsm_volume(st
, info
, map
);
1592 /* Set raid_disks to zero so that Assemble will always pull in valid
1595 info
->array
.raid_disks
= 0;
1596 info
->array
.level
= LEVEL_CONTAINER
;
1597 info
->array
.layout
= 0;
1598 info
->array
.md_minor
= -1;
1599 info
->array
.ctime
= 0; /* N/A for imsm */
1600 info
->array
.utime
= 0;
1601 info
->array
.chunk_size
= 0;
1603 info
->disk
.major
= 0;
1604 info
->disk
.minor
= 0;
1605 info
->disk
.raid_disk
= -1;
1606 info
->reshape_active
= 0;
1607 info
->array
.major_version
= -1;
1608 info
->array
.minor_version
= -2;
1609 strcpy(info
->text_version
, "imsm");
1610 info
->safe_mode_delay
= 0;
1611 info
->disk
.number
= -1;
1612 info
->disk
.state
= 0;
1614 info
->recovery_start
= MaxSector
;
1616 /* do we have the all the insync disks that we expect? */
1617 mpb
= super
->anchor
;
1619 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
1620 struct imsm_dev
*dev
= get_imsm_dev(super
, i
);
1621 int failed
, enough
, j
, missing
= 0;
1622 struct imsm_map
*map
;
1625 failed
= imsm_count_failed(super
, dev
);
1626 state
= imsm_check_degraded(super
, dev
, failed
);
1627 map
= get_imsm_map(dev
, dev
->vol
.migr_state
);
1629 /* any newly missing disks?
1630 * (catches single-degraded vs double-degraded)
1632 for (j
= 0; j
< map
->num_members
; j
++) {
1633 __u32 ord
= get_imsm_ord_tbl_ent(dev
, i
);
1634 __u32 idx
= ord_to_idx(ord
);
1636 if (!(ord
& IMSM_ORD_REBUILD
) &&
1637 get_imsm_missing(super
, idx
)) {
1643 if (state
== IMSM_T_STATE_FAILED
)
1645 else if (state
== IMSM_T_STATE_DEGRADED
&&
1646 (state
!= map
->map_state
|| missing
))
1648 else /* we're normal, or already degraded */
1651 /* in the missing/failed disk case check to see
1652 * if at least one array is runnable
1654 max_enough
= max(max_enough
, enough
);
1656 dprintf("%s: enough: %d\n", __func__
, max_enough
);
1657 info
->container_enough
= max_enough
;
1660 __u32 reserved
= imsm_reserved_sectors(super
, super
->disks
);
1662 disk
= &super
->disks
->disk
;
1663 info
->data_offset
= __le32_to_cpu(disk
->total_blocks
) - reserved
;
1664 info
->component_size
= reserved
;
1665 info
->disk
.state
= is_configured(disk
) ? (1 << MD_DISK_ACTIVE
) : 0;
1666 /* we don't change info->disk.raid_disk here because
1667 * this state will be finalized in mdmon after we have
1668 * found the 'most fresh' version of the metadata
1670 info
->disk
.state
|= is_failed(disk
) ? (1 << MD_DISK_FAULTY
) : 0;
1671 info
->disk
.state
|= is_spare(disk
) ? 0 : (1 << MD_DISK_SYNC
);
1674 /* only call uuid_from_super_imsm when this disk is part of a populated container,
1675 * ->compare_super may have updated the 'num_raid_devs' field for spares
1677 if (info
->disk
.state
& (1 << MD_DISK_SYNC
) || super
->anchor
->num_raid_devs
)
1678 uuid_from_super_imsm(st
, info
->uuid
);
1680 memcpy(info
->uuid
, uuid_match_any
, sizeof(int[4]));
1681 fixup_container_spare_uuid(info
);
1684 /* I don't know how to compute 'map' on imsm, so use safe default */
1687 for (i
= 0; i
< map_disks
; i
++)
1693 static int update_super_imsm(struct supertype
*st
, struct mdinfo
*info
,
1694 char *update
, char *devname
, int verbose
,
1695 int uuid_set
, char *homehost
)
1697 /* For 'assemble' and 'force' we need to return non-zero if any
1698 * change was made. For others, the return value is ignored.
1699 * Update options are:
1700 * force-one : This device looks a bit old but needs to be included,
1701 * update age info appropriately.
1702 * assemble: clear any 'faulty' flag to allow this device to
1704 * force-array: Array is degraded but being forced, mark it clean
1705 * if that will be needed to assemble it.
1707 * newdev: not used ????
1708 * grow: Array has gained a new device - this is currently for
1710 * resync: mark as dirty so a resync will happen.
1711 * name: update the name - preserving the homehost
1712 * uuid: Change the uuid of the array to match watch is given
1714 * Following are not relevant for this imsm:
1715 * sparc2.2 : update from old dodgey metadata
1716 * super-minor: change the preferred_minor number
1717 * summaries: update redundant counters.
1718 * homehost: update the recorded homehost
1719 * _reshape_progress: record new reshape_progress position.
1722 struct intel_super
*super
= st
->sb
;
1723 struct imsm_super
*mpb
;
1725 /* we can only update container info */
1726 if (!super
|| super
->current_vol
>= 0 || !super
->anchor
)
1729 mpb
= super
->anchor
;
1731 if (strcmp(update
, "uuid") == 0 && uuid_set
&& !info
->update_private
)
1733 else if (strcmp(update
, "uuid") == 0 && uuid_set
&& info
->update_private
) {
1734 mpb
->orig_family_num
= *((__u32
*) info
->update_private
);
1736 } else if (strcmp(update
, "uuid") == 0) {
1737 __u32
*new_family
= malloc(sizeof(*new_family
));
1739 /* update orig_family_number with the incoming random
1740 * data, report the new effective uuid, and store the
1741 * new orig_family_num for future updates.
1744 memcpy(&mpb
->orig_family_num
, info
->uuid
, sizeof(__u32
));
1745 uuid_from_super_imsm(st
, info
->uuid
);
1746 *new_family
= mpb
->orig_family_num
;
1747 info
->update_private
= new_family
;
1750 } else if (strcmp(update
, "assemble") == 0)
1755 /* successful update? recompute checksum */
1757 mpb
->check_sum
= __le32_to_cpu(__gen_imsm_checksum(mpb
));
1762 static size_t disks_to_mpb_size(int disks
)
1766 size
= sizeof(struct imsm_super
);
1767 size
+= (disks
- 1) * sizeof(struct imsm_disk
);
1768 size
+= 2 * sizeof(struct imsm_dev
);
1769 /* up to 2 maps per raid device (-2 for imsm_maps in imsm_dev */
1770 size
+= (4 - 2) * sizeof(struct imsm_map
);
1771 /* 4 possible disk_ord_tbl's */
1772 size
+= 4 * (disks
- 1) * sizeof(__u32
);
1777 static __u64
avail_size_imsm(struct supertype
*st
, __u64 devsize
)
1779 if (devsize
< (MPB_SECTOR_CNT
+ IMSM_RESERVED_SECTORS
))
1782 return devsize
- (MPB_SECTOR_CNT
+ IMSM_RESERVED_SECTORS
);
1785 static void free_devlist(struct intel_super
*super
)
1787 struct intel_dev
*dv
;
1789 while (super
->devlist
) {
1790 dv
= super
->devlist
->next
;
1791 free(super
->devlist
->dev
);
1792 free(super
->devlist
);
1793 super
->devlist
= dv
;
1797 static void imsm_copy_dev(struct imsm_dev
*dest
, struct imsm_dev
*src
)
1799 memcpy(dest
, src
, sizeof_imsm_dev(src
, 0));
1802 static int compare_super_imsm(struct supertype
*st
, struct supertype
*tst
)
1806 * 0 same, or first was empty, and second was copied
1807 * 1 second had wrong number
1809 * 3 wrong other info
1811 struct intel_super
*first
= st
->sb
;
1812 struct intel_super
*sec
= tst
->sb
;
1820 /* if an anchor does not have num_raid_devs set then it is a free
1823 if (first
->anchor
->num_raid_devs
> 0 &&
1824 sec
->anchor
->num_raid_devs
> 0) {
1825 /* Determine if these disks might ever have been
1826 * related. Further disambiguation can only take place
1827 * in load_super_imsm_all
1829 __u32 first_family
= first
->anchor
->orig_family_num
;
1830 __u32 sec_family
= sec
->anchor
->orig_family_num
;
1832 if (memcmp(first
->anchor
->sig
, sec
->anchor
->sig
,
1833 MAX_SIGNATURE_LENGTH
) != 0)
1836 if (first_family
== 0)
1837 first_family
= first
->anchor
->family_num
;
1838 if (sec_family
== 0)
1839 sec_family
= sec
->anchor
->family_num
;
1841 if (first_family
!= sec_family
)
1847 /* if 'first' is a spare promote it to a populated mpb with sec's
1850 if (first
->anchor
->num_raid_devs
== 0 &&
1851 sec
->anchor
->num_raid_devs
> 0) {
1853 struct intel_dev
*dv
;
1854 struct imsm_dev
*dev
;
1856 /* we need to copy raid device info from sec if an allocation
1857 * fails here we don't associate the spare
1859 for (i
= 0; i
< sec
->anchor
->num_raid_devs
; i
++) {
1860 dv
= malloc(sizeof(*dv
));
1863 dev
= malloc(sizeof_imsm_dev(get_imsm_dev(sec
, i
), 1));
1870 dv
->next
= first
->devlist
;
1871 first
->devlist
= dv
;
1873 if (i
< sec
->anchor
->num_raid_devs
) {
1874 /* allocation failure */
1875 free_devlist(first
);
1876 fprintf(stderr
, "imsm: failed to associate spare\n");
1879 first
->anchor
->num_raid_devs
= sec
->anchor
->num_raid_devs
;
1880 first
->anchor
->orig_family_num
= sec
->anchor
->orig_family_num
;
1881 first
->anchor
->family_num
= sec
->anchor
->family_num
;
1882 memcpy(first
->anchor
->sig
, sec
->anchor
->sig
, MAX_SIGNATURE_LENGTH
);
1883 for (i
= 0; i
< sec
->anchor
->num_raid_devs
; i
++)
1884 imsm_copy_dev(get_imsm_dev(first
, i
), get_imsm_dev(sec
, i
));
1890 static void fd2devname(int fd
, char *name
)
1894 char dname
[PATH_MAX
];
1899 if (fstat(fd
, &st
) != 0)
1901 sprintf(path
, "/sys/dev/block/%d:%d",
1902 major(st
.st_rdev
), minor(st
.st_rdev
));
1904 rv
= readlink(path
, dname
, sizeof(dname
));
1909 nm
= strrchr(dname
, '/');
1911 snprintf(name
, MAX_RAID_SERIAL_LEN
, "/dev/%s", nm
);
1914 extern int scsi_get_serial(int fd
, void *buf
, size_t buf_len
);
1916 static int imsm_read_serial(int fd
, char *devname
,
1917 __u8 serial
[MAX_RAID_SERIAL_LEN
])
1919 unsigned char scsi_serial
[255];
1928 memset(scsi_serial
, 0, sizeof(scsi_serial
));
1930 rv
= scsi_get_serial(fd
, scsi_serial
, sizeof(scsi_serial
));
1932 if (rv
&& check_env("IMSM_DEVNAME_AS_SERIAL")) {
1933 memset(serial
, 0, MAX_RAID_SERIAL_LEN
);
1934 fd2devname(fd
, (char *) serial
);
1941 Name
": Failed to retrieve serial for %s\n",
1946 rsp_len
= scsi_serial
[3];
1950 Name
": Failed to retrieve serial for %s\n",
1954 rsp_buf
= (char *) &scsi_serial
[4];
1956 /* trim all whitespace and non-printable characters and convert
1959 for (i
= 0, dest
= rsp_buf
; i
< rsp_len
; i
++) {
1962 /* ':' is reserved for use in placeholder serial
1963 * numbers for missing disks
1971 len
= dest
- rsp_buf
;
1974 /* truncate leading characters */
1975 if (len
> MAX_RAID_SERIAL_LEN
) {
1976 dest
+= len
- MAX_RAID_SERIAL_LEN
;
1977 len
= MAX_RAID_SERIAL_LEN
;
1980 memset(serial
, 0, MAX_RAID_SERIAL_LEN
);
1981 memcpy(serial
, dest
, len
);
1986 static int serialcmp(__u8
*s1
, __u8
*s2
)
1988 return strncmp((char *) s1
, (char *) s2
, MAX_RAID_SERIAL_LEN
);
1991 static void serialcpy(__u8
*dest
, __u8
*src
)
1993 strncpy((char *) dest
, (char *) src
, MAX_RAID_SERIAL_LEN
);
1997 static struct dl
*serial_to_dl(__u8
*serial
, struct intel_super
*super
)
2001 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
2002 if (serialcmp(dl
->serial
, serial
) == 0)
2009 static struct imsm_disk
*
2010 __serial_to_disk(__u8
*serial
, struct imsm_super
*mpb
, int *idx
)
2014 for (i
= 0; i
< mpb
->num_disks
; i
++) {
2015 struct imsm_disk
*disk
= __get_imsm_disk(mpb
, i
);
2017 if (serialcmp(disk
->serial
, serial
) == 0) {
2028 load_imsm_disk(int fd
, struct intel_super
*super
, char *devname
, int keep_fd
)
2030 struct imsm_disk
*disk
;
2035 __u8 serial
[MAX_RAID_SERIAL_LEN
];
2037 rv
= imsm_read_serial(fd
, devname
, serial
);
2042 dl
= calloc(1, sizeof(*dl
));
2046 Name
": failed to allocate disk buffer for %s\n",
2052 dl
->major
= major(stb
.st_rdev
);
2053 dl
->minor
= minor(stb
.st_rdev
);
2054 dl
->next
= super
->disks
;
2055 dl
->fd
= keep_fd
? fd
: -1;
2056 assert(super
->disks
== NULL
);
2058 serialcpy(dl
->serial
, serial
);
2061 fd2devname(fd
, name
);
2063 dl
->devname
= strdup(devname
);
2065 dl
->devname
= strdup(name
);
2067 /* look up this disk's index in the current anchor */
2068 disk
= __serial_to_disk(dl
->serial
, super
->anchor
, &dl
->index
);
2071 /* only set index on disks that are a member of a
2072 * populated contianer, i.e. one with raid_devs
2074 if (is_failed(&dl
->disk
))
2076 else if (is_spare(&dl
->disk
))
2084 /* When migrating map0 contains the 'destination' state while map1
2085 * contains the current state. When not migrating map0 contains the
2086 * current state. This routine assumes that map[0].map_state is set to
2087 * the current array state before being called.
2089 * Migration is indicated by one of the following states
2090 * 1/ Idle (migr_state=0 map0state=normal||unitialized||degraded||failed)
2091 * 2/ Initialize (migr_state=1 migr_type=MIGR_INIT map0state=normal
2092 * map1state=unitialized)
2093 * 3/ Repair (Resync) (migr_state=1 migr_type=MIGR_REPAIR map0state=normal
2095 * 4/ Rebuild (migr_state=1 migr_type=MIGR_REBUILD map0state=normal
2096 * map1state=degraded)
2098 static void migrate(struct imsm_dev
*dev
, __u8 to_state
, int migr_type
)
2100 struct imsm_map
*dest
;
2101 struct imsm_map
*src
= get_imsm_map(dev
, 0);
2103 dev
->vol
.migr_state
= 1;
2104 set_migr_type(dev
, migr_type
);
2105 dev
->vol
.curr_migr_unit
= 0;
2106 dest
= get_imsm_map(dev
, 1);
2108 /* duplicate and then set the target end state in map[0] */
2109 memcpy(dest
, src
, sizeof_imsm_map(src
));
2110 if (migr_type
== MIGR_REBUILD
) {
2114 for (i
= 0; i
< src
->num_members
; i
++) {
2115 ord
= __le32_to_cpu(src
->disk_ord_tbl
[i
]);
2116 set_imsm_ord_tbl_ent(src
, i
, ord_to_idx(ord
));
2120 src
->map_state
= to_state
;
2123 static void end_migration(struct imsm_dev
*dev
, __u8 map_state
)
2125 struct imsm_map
*map
= get_imsm_map(dev
, 0);
2126 struct imsm_map
*prev
= get_imsm_map(dev
, dev
->vol
.migr_state
);
2129 /* merge any IMSM_ORD_REBUILD bits that were not successfully
2130 * completed in the last migration.
2132 * FIXME add support for online capacity expansion and
2133 * raid-level-migration
2135 for (i
= 0; i
< prev
->num_members
; i
++)
2136 map
->disk_ord_tbl
[i
] |= prev
->disk_ord_tbl
[i
];
2138 dev
->vol
.migr_state
= 0;
2139 dev
->vol
.curr_migr_unit
= 0;
2140 map
->map_state
= map_state
;
2144 static int parse_raid_devices(struct intel_super
*super
)
2147 struct imsm_dev
*dev_new
;
2148 size_t len
, len_migr
;
2149 size_t space_needed
= 0;
2150 struct imsm_super
*mpb
= super
->anchor
;
2152 for (i
= 0; i
< super
->anchor
->num_raid_devs
; i
++) {
2153 struct imsm_dev
*dev_iter
= __get_imsm_dev(super
->anchor
, i
);
2154 struct intel_dev
*dv
;
2156 len
= sizeof_imsm_dev(dev_iter
, 0);
2157 len_migr
= sizeof_imsm_dev(dev_iter
, 1);
2159 space_needed
+= len_migr
- len
;
2161 dv
= malloc(sizeof(*dv
));
2164 dev_new
= malloc(len_migr
);
2169 imsm_copy_dev(dev_new
, dev_iter
);
2172 dv
->next
= super
->devlist
;
2173 super
->devlist
= dv
;
2176 /* ensure that super->buf is large enough when all raid devices
2179 if (__le32_to_cpu(mpb
->mpb_size
) + space_needed
> super
->len
) {
2182 len
= ROUND_UP(__le32_to_cpu(mpb
->mpb_size
) + space_needed
, 512);
2183 if (posix_memalign(&buf
, 512, len
) != 0)
2186 memcpy(buf
, super
->buf
, super
->len
);
2187 memset(buf
+ super
->len
, 0, len
- super
->len
);
2196 /* retrieve a pointer to the bbm log which starts after all raid devices */
2197 struct bbm_log
*__get_imsm_bbm_log(struct imsm_super
*mpb
)
2201 if (__le32_to_cpu(mpb
->bbm_log_size
)) {
2203 ptr
+= mpb
->mpb_size
- __le32_to_cpu(mpb
->bbm_log_size
);
2209 static void __free_imsm(struct intel_super
*super
, int free_disks
);
2211 /* load_imsm_mpb - read matrix metadata
2212 * allocates super->mpb to be freed by free_super
2214 static int load_imsm_mpb(int fd
, struct intel_super
*super
, char *devname
)
2216 unsigned long long dsize
;
2217 unsigned long long sectors
;
2219 struct imsm_super
*anchor
;
2222 get_dev_size(fd
, NULL
, &dsize
);
2226 Name
": %s: device to small for imsm\n",
2231 if (lseek64(fd
, dsize
- (512 * 2), SEEK_SET
) < 0) {
2234 Name
": Cannot seek to anchor block on %s: %s\n",
2235 devname
, strerror(errno
));
2239 if (posix_memalign((void**)&anchor
, 512, 512) != 0) {
2242 Name
": Failed to allocate imsm anchor buffer"
2243 " on %s\n", devname
);
2246 if (read(fd
, anchor
, 512) != 512) {
2249 Name
": Cannot read anchor block on %s: %s\n",
2250 devname
, strerror(errno
));
2255 if (strncmp((char *) anchor
->sig
, MPB_SIGNATURE
, MPB_SIG_LEN
) != 0) {
2258 Name
": no IMSM anchor on %s\n", devname
);
2263 __free_imsm(super
, 0);
2264 super
->len
= ROUND_UP(anchor
->mpb_size
, 512);
2265 if (posix_memalign(&super
->buf
, 512, super
->len
) != 0) {
2268 Name
": unable to allocate %zu byte mpb buffer\n",
2273 memcpy(super
->buf
, anchor
, 512);
2275 sectors
= mpb_sectors(anchor
) - 1;
2278 check_sum
= __gen_imsm_checksum(super
->anchor
);
2279 if (check_sum
!= __le32_to_cpu(super
->anchor
->check_sum
)) {
2282 Name
": IMSM checksum %x != %x on %s\n",
2284 __le32_to_cpu(super
->anchor
->check_sum
),
2292 /* read the extended mpb */
2293 if (lseek64(fd
, dsize
- (512 * (2 + sectors
)), SEEK_SET
) < 0) {
2296 Name
": Cannot seek to extended mpb on %s: %s\n",
2297 devname
, strerror(errno
));
2301 if ((unsigned)read(fd
, super
->buf
+ 512, super
->len
- 512) != super
->len
- 512) {
2304 Name
": Cannot read extended mpb on %s: %s\n",
2305 devname
, strerror(errno
));
2309 check_sum
= __gen_imsm_checksum(super
->anchor
);
2310 if (check_sum
!= __le32_to_cpu(super
->anchor
->check_sum
)) {
2313 Name
": IMSM checksum %x != %x on %s\n",
2314 check_sum
, __le32_to_cpu(super
->anchor
->check_sum
),
2319 /* FIXME the BBM log is disk specific so we cannot use this global
2320 * buffer for all disks. Ok for now since we only look at the global
2321 * bbm_log_size parameter to gate assembly
2323 super
->bbm_log
= __get_imsm_bbm_log(super
->anchor
);
2329 load_and_parse_mpb(int fd
, struct intel_super
*super
, char *devname
, int keep_fd
)
2333 err
= load_imsm_mpb(fd
, super
, devname
);
2336 err
= load_imsm_disk(fd
, super
, devname
, keep_fd
);
2339 err
= parse_raid_devices(super
);
2344 static void __free_imsm_disk(struct dl
*d
)
2355 static void free_imsm_disks(struct intel_super
*super
)
2359 while (super
->disks
) {
2361 super
->disks
= d
->next
;
2362 __free_imsm_disk(d
);
2364 while (super
->missing
) {
2366 super
->missing
= d
->next
;
2367 __free_imsm_disk(d
);
2372 /* free all the pieces hanging off of a super pointer */
2373 static void __free_imsm(struct intel_super
*super
, int free_disks
)
2380 free_imsm_disks(super
);
2381 free_devlist(super
);
2383 free((void *) super
->hba
);
2388 static void free_imsm(struct intel_super
*super
)
2390 __free_imsm(super
, 1);
2394 static void free_super_imsm(struct supertype
*st
)
2396 struct intel_super
*super
= st
->sb
;
2405 static struct intel_super
*alloc_super(void)
2407 struct intel_super
*super
= malloc(sizeof(*super
));
2410 memset(super
, 0, sizeof(*super
));
2411 super
->current_vol
= -1;
2412 super
->create_offset
= ~((__u32
) 0);
2413 if (!check_env("IMSM_NO_PLATFORM"))
2414 super
->orom
= find_imsm_orom();
2415 if (super
->orom
&& !check_env("IMSM_TEST_OROM")) {
2416 struct sys_dev
*list
, *ent
;
2418 /* find the first intel ahci controller */
2419 list
= find_driver_devices("pci", "ahci");
2420 for (ent
= list
; ent
; ent
= ent
->next
)
2421 if (devpath_to_vendor(ent
->path
) == 0x8086)
2424 super
->hba
= ent
->path
;
2427 free_sys_dev(&list
);
2435 /* find_missing - helper routine for load_super_imsm_all that identifies
2436 * disks that have disappeared from the system. This routine relies on
2437 * the mpb being uptodate, which it is at load time.
2439 static int find_missing(struct intel_super
*super
)
2442 struct imsm_super
*mpb
= super
->anchor
;
2444 struct imsm_disk
*disk
;
2446 for (i
= 0; i
< mpb
->num_disks
; i
++) {
2447 disk
= __get_imsm_disk(mpb
, i
);
2448 dl
= serial_to_dl(disk
->serial
, super
);
2452 dl
= malloc(sizeof(*dl
));
2458 dl
->devname
= strdup("missing");
2460 serialcpy(dl
->serial
, disk
->serial
);
2463 dl
->next
= super
->missing
;
2464 super
->missing
= dl
;
2470 static struct intel_disk
*disk_list_get(__u8
*serial
, struct intel_disk
*disk_list
)
2472 struct intel_disk
*idisk
= disk_list
;
2475 if (serialcmp(idisk
->disk
.serial
, serial
) == 0)
2477 idisk
= idisk
->next
;
2483 static int __prep_thunderdome(struct intel_super
**table
, int tbl_size
,
2484 struct intel_super
*super
,
2485 struct intel_disk
**disk_list
)
2487 struct imsm_disk
*d
= &super
->disks
->disk
;
2488 struct imsm_super
*mpb
= super
->anchor
;
2491 for (i
= 0; i
< tbl_size
; i
++) {
2492 struct imsm_super
*tbl_mpb
= table
[i
]->anchor
;
2493 struct imsm_disk
*tbl_d
= &table
[i
]->disks
->disk
;
2495 if (tbl_mpb
->family_num
== mpb
->family_num
) {
2496 if (tbl_mpb
->check_sum
== mpb
->check_sum
) {
2497 dprintf("%s: mpb from %d:%d matches %d:%d\n",
2498 __func__
, super
->disks
->major
,
2499 super
->disks
->minor
,
2500 table
[i
]->disks
->major
,
2501 table
[i
]->disks
->minor
);
2505 if (((is_configured(d
) && !is_configured(tbl_d
)) ||
2506 is_configured(d
) == is_configured(tbl_d
)) &&
2507 tbl_mpb
->generation_num
< mpb
->generation_num
) {
2508 /* current version of the mpb is a
2509 * better candidate than the one in
2510 * super_table, but copy over "cross
2511 * generational" status
2513 struct intel_disk
*idisk
;
2515 dprintf("%s: mpb from %d:%d replaces %d:%d\n",
2516 __func__
, super
->disks
->major
,
2517 super
->disks
->minor
,
2518 table
[i
]->disks
->major
,
2519 table
[i
]->disks
->minor
);
2521 idisk
= disk_list_get(tbl_d
->serial
, *disk_list
);
2522 if (idisk
&& is_failed(&idisk
->disk
))
2523 tbl_d
->status
|= FAILED_DISK
;
2526 struct intel_disk
*idisk
;
2527 struct imsm_disk
*disk
;
2529 /* tbl_mpb is more up to date, but copy
2530 * over cross generational status before
2533 disk
= __serial_to_disk(d
->serial
, mpb
, NULL
);
2534 if (disk
&& is_failed(disk
))
2535 d
->status
|= FAILED_DISK
;
2537 idisk
= disk_list_get(d
->serial
, *disk_list
);
2540 if (disk
&& is_configured(disk
))
2541 idisk
->disk
.status
|= CONFIGURED_DISK
;
2544 dprintf("%s: mpb from %d:%d prefer %d:%d\n",
2545 __func__
, super
->disks
->major
,
2546 super
->disks
->minor
,
2547 table
[i
]->disks
->major
,
2548 table
[i
]->disks
->minor
);
2556 table
[tbl_size
++] = super
;
2560 /* update/extend the merged list of imsm_disk records */
2561 for (j
= 0; j
< mpb
->num_disks
; j
++) {
2562 struct imsm_disk
*disk
= __get_imsm_disk(mpb
, j
);
2563 struct intel_disk
*idisk
;
2565 idisk
= disk_list_get(disk
->serial
, *disk_list
);
2567 idisk
->disk
.status
|= disk
->status
;
2568 if (is_configured(&idisk
->disk
) ||
2569 is_failed(&idisk
->disk
))
2570 idisk
->disk
.status
&= ~(SPARE_DISK
);
2572 idisk
= calloc(1, sizeof(*idisk
));
2575 idisk
->owner
= IMSM_UNKNOWN_OWNER
;
2576 idisk
->disk
= *disk
;
2577 idisk
->next
= *disk_list
;
2581 if (serialcmp(idisk
->disk
.serial
, d
->serial
) == 0)
2588 static struct intel_super
*
2589 validate_members(struct intel_super
*super
, struct intel_disk
*disk_list
,
2592 struct imsm_super
*mpb
= super
->anchor
;
2596 for (i
= 0; i
< mpb
->num_disks
; i
++) {
2597 struct imsm_disk
*disk
= __get_imsm_disk(mpb
, i
);
2598 struct intel_disk
*idisk
;
2600 idisk
= disk_list_get(disk
->serial
, disk_list
);
2602 if (idisk
->owner
== owner
||
2603 idisk
->owner
== IMSM_UNKNOWN_OWNER
)
2606 dprintf("%s: '%.16s' owner %d != %d\n",
2607 __func__
, disk
->serial
, idisk
->owner
,
2610 dprintf("%s: unknown disk %x [%d]: %.16s\n",
2611 __func__
, __le32_to_cpu(mpb
->family_num
), i
,
2617 if (ok_count
== mpb
->num_disks
)
2622 static void show_conflicts(__u32 family_num
, struct intel_super
*super_list
)
2624 struct intel_super
*s
;
2626 for (s
= super_list
; s
; s
= s
->next
) {
2627 if (family_num
!= s
->anchor
->family_num
)
2629 fprintf(stderr
, "Conflict, offlining family %#x on '%s'\n",
2630 __le32_to_cpu(family_num
), s
->disks
->devname
);
2634 static struct intel_super
*
2635 imsm_thunderdome(struct intel_super
**super_list
, int len
)
2637 struct intel_super
*super_table
[len
];
2638 struct intel_disk
*disk_list
= NULL
;
2639 struct intel_super
*champion
, *spare
;
2640 struct intel_super
*s
, **del
;
2645 memset(super_table
, 0, sizeof(super_table
));
2646 for (s
= *super_list
; s
; s
= s
->next
)
2647 tbl_size
= __prep_thunderdome(super_table
, tbl_size
, s
, &disk_list
);
2649 for (i
= 0; i
< tbl_size
; i
++) {
2650 struct imsm_disk
*d
;
2651 struct intel_disk
*idisk
;
2652 struct imsm_super
*mpb
= super_table
[i
]->anchor
;
2655 d
= &s
->disks
->disk
;
2657 /* 'd' must appear in merged disk list for its
2658 * configuration to be valid
2660 idisk
= disk_list_get(d
->serial
, disk_list
);
2661 if (idisk
&& idisk
->owner
== i
)
2662 s
= validate_members(s
, disk_list
, i
);
2667 dprintf("%s: marking family: %#x from %d:%d offline\n",
2668 __func__
, mpb
->family_num
,
2669 super_table
[i
]->disks
->major
,
2670 super_table
[i
]->disks
->minor
);
2674 /* This is where the mdadm implementation differs from the Windows
2675 * driver which has no strict concept of a container. We can only
2676 * assemble one family from a container, so when returning a prodigal
2677 * array member to this system the code will not be able to disambiguate
2678 * the container contents that should be assembled ("foreign" versus
2679 * "local"). It requires user intervention to set the orig_family_num
2680 * to a new value to establish a new container. The Windows driver in
2681 * this situation fixes up the volume name in place and manages the
2682 * foreign array as an independent entity.
2687 for (i
= 0; i
< tbl_size
; i
++) {
2688 struct intel_super
*tbl_ent
= super_table
[i
];
2694 if (tbl_ent
->anchor
->num_raid_devs
== 0) {
2699 if (s
&& !is_spare
) {
2700 show_conflicts(tbl_ent
->anchor
->family_num
, *super_list
);
2702 } else if (!s
&& !is_spare
)
2715 fprintf(stderr
, "Chose family %#x on '%s', "
2716 "assemble conflicts to new container with '--update=uuid'\n",
2717 __le32_to_cpu(s
->anchor
->family_num
), s
->disks
->devname
);
2719 /* collect all dl's onto 'champion', and update them to
2720 * champion's version of the status
2722 for (s
= *super_list
; s
; s
= s
->next
) {
2723 struct imsm_super
*mpb
= champion
->anchor
;
2724 struct dl
*dl
= s
->disks
;
2729 for (i
= 0; i
< mpb
->num_disks
; i
++) {
2730 struct imsm_disk
*disk
;
2732 disk
= __serial_to_disk(dl
->serial
, mpb
, &dl
->index
);
2735 /* only set index on disks that are a member of
2736 * a populated contianer, i.e. one with
2739 if (is_failed(&dl
->disk
))
2741 else if (is_spare(&dl
->disk
))
2747 if (i
>= mpb
->num_disks
) {
2748 struct intel_disk
*idisk
;
2750 idisk
= disk_list_get(dl
->serial
, disk_list
);
2751 if (idisk
&& is_spare(&idisk
->disk
) &&
2752 !is_failed(&idisk
->disk
) && !is_configured(&idisk
->disk
))
2760 dl
->next
= champion
->disks
;
2761 champion
->disks
= dl
;
2765 /* delete 'champion' from super_list */
2766 for (del
= super_list
; *del
; ) {
2767 if (*del
== champion
) {
2768 *del
= (*del
)->next
;
2771 del
= &(*del
)->next
;
2773 champion
->next
= NULL
;
2777 struct intel_disk
*idisk
= disk_list
;
2779 disk_list
= disk_list
->next
;
2786 static int load_super_imsm_all(struct supertype
*st
, int fd
, void **sbp
,
2790 struct intel_super
*super_list
= NULL
;
2791 struct intel_super
*super
= NULL
;
2792 int devnum
= fd2devnum(fd
);
2798 /* check if 'fd' an opened container */
2799 sra
= sysfs_read(fd
, 0, GET_LEVEL
|GET_VERSION
|GET_DEVS
|GET_STATE
);
2803 if (sra
->array
.major_version
!= -1 ||
2804 sra
->array
.minor_version
!= -2 ||
2805 strcmp(sra
->text_version
, "imsm") != 0) {
2810 for (sd
= sra
->devs
, i
= 0; sd
; sd
= sd
->next
, i
++) {
2811 struct intel_super
*s
= alloc_super();
2818 s
->next
= super_list
;
2822 sprintf(nm
, "%d:%d", sd
->disk
.major
, sd
->disk
.minor
);
2823 dfd
= dev_open(nm
, O_RDWR
);
2827 err
= load_and_parse_mpb(dfd
, s
, NULL
, 1);
2829 /* retry the load if we might have raced against mdmon */
2830 if (err
== 3 && mdmon_running(devnum
))
2831 for (retry
= 0; retry
< 3; retry
++) {
2833 err
= load_and_parse_mpb(dfd
, s
, NULL
, 1);
2841 /* all mpbs enter, maybe one leaves */
2842 super
= imsm_thunderdome(&super_list
, i
);
2848 if (find_missing(super
) != 0) {
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
;
2877 static int load_container_imsm(struct supertype
*st
, int fd
, char *devname
)
2879 return load_super_imsm_all(st
, fd
, &st
->sb
, devname
);
2883 static int load_super_imsm(struct supertype
*st
, int fd
, char *devname
)
2885 struct intel_super
*super
;
2889 if (load_super_imsm_all(st
, fd
, &st
->sb
, devname
) == 0)
2893 if (test_partition(fd
))
2894 /* IMSM not allowed on partitions */
2897 free_super_imsm(st
);
2899 super
= alloc_super();
2902 Name
": malloc of %zu failed.\n",
2907 rv
= load_and_parse_mpb(fd
, super
, devname
, 0);
2912 Name
": Failed to load all information "
2913 "sections on %s\n", devname
);
2919 if (st
->ss
== NULL
) {
2920 st
->ss
= &super_imsm
;
2921 st
->minor_version
= 0;
2922 st
->max_devs
= IMSM_MAX_DEVICES
;
2927 static __u16
info_to_blocks_per_strip(mdu_array_info_t
*info
)
2929 if (info
->level
== 1)
2931 return info
->chunk_size
>> 9;
2934 static __u32
info_to_num_data_stripes(mdu_array_info_t
*info
, int num_domains
)
2938 num_stripes
= (info
->size
* 2) / info_to_blocks_per_strip(info
);
2939 num_stripes
/= num_domains
;
2944 static __u32
info_to_blocks_per_member(mdu_array_info_t
*info
)
2946 if (info
->level
== 1)
2947 return info
->size
* 2;
2949 return (info
->size
* 2) & ~(info_to_blocks_per_strip(info
) - 1);
2952 static void imsm_update_version_info(struct intel_super
*super
)
2954 /* update the version and attributes */
2955 struct imsm_super
*mpb
= super
->anchor
;
2957 struct imsm_dev
*dev
;
2958 struct imsm_map
*map
;
2961 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
2962 dev
= get_imsm_dev(super
, i
);
2963 map
= get_imsm_map(dev
, 0);
2964 if (__le32_to_cpu(dev
->size_high
) > 0)
2965 mpb
->attributes
|= MPB_ATTRIB_2TB
;
2967 /* FIXME detect when an array spans a port multiplier */
2969 mpb
->attributes
|= MPB_ATTRIB_PM
;
2972 if (mpb
->num_raid_devs
> 1 ||
2973 mpb
->attributes
!= MPB_ATTRIB_CHECKSUM_VERIFY
) {
2974 version
= MPB_VERSION_ATTRIBS
;
2975 switch (get_imsm_raid_level(map
)) {
2976 case 0: mpb
->attributes
|= MPB_ATTRIB_RAID0
; break;
2977 case 1: mpb
->attributes
|= MPB_ATTRIB_RAID1
; break;
2978 case 10: mpb
->attributes
|= MPB_ATTRIB_RAID10
; break;
2979 case 5: mpb
->attributes
|= MPB_ATTRIB_RAID5
; break;
2982 if (map
->num_members
>= 5)
2983 version
= MPB_VERSION_5OR6_DISK_ARRAY
;
2984 else if (dev
->status
== DEV_CLONE_N_GO
)
2985 version
= MPB_VERSION_CNG
;
2986 else if (get_imsm_raid_level(map
) == 5)
2987 version
= MPB_VERSION_RAID5
;
2988 else if (map
->num_members
>= 3)
2989 version
= MPB_VERSION_3OR4_DISK_ARRAY
;
2990 else if (get_imsm_raid_level(map
) == 1)
2991 version
= MPB_VERSION_RAID1
;
2993 version
= MPB_VERSION_RAID0
;
2995 strcpy(((char *) mpb
->sig
) + strlen(MPB_SIGNATURE
), version
);
2999 static int check_name(struct intel_super
*super
, char *name
, int quiet
)
3001 struct imsm_super
*mpb
= super
->anchor
;
3002 char *reason
= NULL
;
3005 if (strlen(name
) > MAX_RAID_SERIAL_LEN
)
3006 reason
= "must be 16 characters or less";
3008 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
3009 struct imsm_dev
*dev
= get_imsm_dev(super
, i
);
3011 if (strncmp((char *) dev
->volume
, name
, MAX_RAID_SERIAL_LEN
) == 0) {
3012 reason
= "already exists";
3017 if (reason
&& !quiet
)
3018 fprintf(stderr
, Name
": imsm volume name %s\n", reason
);
3023 static int init_super_imsm_volume(struct supertype
*st
, mdu_array_info_t
*info
,
3024 unsigned long long size
, char *name
,
3025 char *homehost
, int *uuid
)
3027 /* We are creating a volume inside a pre-existing container.
3028 * so st->sb is already set.
3030 struct intel_super
*super
= st
->sb
;
3031 struct imsm_super
*mpb
= super
->anchor
;
3032 struct intel_dev
*dv
;
3033 struct imsm_dev
*dev
;
3034 struct imsm_vol
*vol
;
3035 struct imsm_map
*map
;
3036 int idx
= mpb
->num_raid_devs
;
3038 unsigned long long array_blocks
;
3039 size_t size_old
, size_new
;
3040 __u32 num_data_stripes
;
3042 if (super
->orom
&& mpb
->num_raid_devs
>= super
->orom
->vpa
) {
3043 fprintf(stderr
, Name
": This imsm-container already has the "
3044 "maximum of %d volumes\n", super
->orom
->vpa
);
3048 /* ensure the mpb is large enough for the new data */
3049 size_old
= __le32_to_cpu(mpb
->mpb_size
);
3050 size_new
= disks_to_mpb_size(info
->nr_disks
);
3051 if (size_new
> size_old
) {
3053 size_t size_round
= ROUND_UP(size_new
, 512);
3055 if (posix_memalign(&mpb_new
, 512, size_round
) != 0) {
3056 fprintf(stderr
, Name
": could not allocate new mpb\n");
3059 memcpy(mpb_new
, mpb
, size_old
);
3062 super
->anchor
= mpb_new
;
3063 mpb
->mpb_size
= __cpu_to_le32(size_new
);
3064 memset(mpb_new
+ size_old
, 0, size_round
- size_old
);
3066 super
->current_vol
= idx
;
3067 /* when creating the first raid device in this container set num_disks
3068 * to zero, i.e. delete this spare and add raid member devices in
3069 * add_to_super_imsm_volume()
3071 if (super
->current_vol
== 0)
3074 if (!check_name(super
, name
, 0))
3076 dv
= malloc(sizeof(*dv
));
3078 fprintf(stderr
, Name
": failed to allocate device list entry\n");
3081 dev
= malloc(sizeof(*dev
) + sizeof(__u32
) * (info
->raid_disks
- 1));
3084 fprintf(stderr
, Name
": could not allocate raid device\n");
3087 strncpy((char *) dev
->volume
, name
, MAX_RAID_SERIAL_LEN
);
3088 if (info
->level
== 1)
3089 array_blocks
= info_to_blocks_per_member(info
);
3091 array_blocks
= calc_array_size(info
->level
, info
->raid_disks
,
3092 info
->layout
, info
->chunk_size
,
3094 /* round array size down to closest MB */
3095 array_blocks
= (array_blocks
>> SECT_PER_MB_SHIFT
) << SECT_PER_MB_SHIFT
;
3097 dev
->size_low
= __cpu_to_le32((__u32
) array_blocks
);
3098 dev
->size_high
= __cpu_to_le32((__u32
) (array_blocks
>> 32));
3099 dev
->status
= __cpu_to_le32(0);
3100 dev
->reserved_blocks
= __cpu_to_le32(0);
3102 vol
->migr_state
= 0;
3103 set_migr_type(dev
, MIGR_INIT
);
3105 vol
->curr_migr_unit
= 0;
3106 map
= get_imsm_map(dev
, 0);
3107 map
->pba_of_lba0
= __cpu_to_le32(super
->create_offset
);
3108 map
->blocks_per_member
= __cpu_to_le32(info_to_blocks_per_member(info
));
3109 map
->blocks_per_strip
= __cpu_to_le16(info_to_blocks_per_strip(info
));
3110 map
->failed_disk_num
= ~0;
3111 map
->map_state
= info
->level
? IMSM_T_STATE_UNINITIALIZED
:
3112 IMSM_T_STATE_NORMAL
;
3115 if (info
->level
== 1 && info
->raid_disks
> 2) {
3118 fprintf(stderr
, Name
": imsm does not support more than 2 disks"
3119 "in a raid1 volume\n");
3123 map
->raid_level
= info
->level
;
3124 if (info
->level
== 10) {
3125 map
->raid_level
= 1;
3126 map
->num_domains
= info
->raid_disks
/ 2;
3127 } else if (info
->level
== 1)
3128 map
->num_domains
= info
->raid_disks
;
3130 map
->num_domains
= 1;
3132 num_data_stripes
= info_to_num_data_stripes(info
, map
->num_domains
);
3133 map
->num_data_stripes
= __cpu_to_le32(num_data_stripes
);
3135 map
->num_members
= info
->raid_disks
;
3136 for (i
= 0; i
< map
->num_members
; i
++) {
3137 /* initialized in add_to_super */
3138 set_imsm_ord_tbl_ent(map
, i
, IMSM_ORD_REBUILD
);
3140 mpb
->num_raid_devs
++;
3143 dv
->index
= super
->current_vol
;
3144 dv
->next
= super
->devlist
;
3145 super
->devlist
= dv
;
3147 imsm_update_version_info(super
);
3152 static int init_super_imsm(struct supertype
*st
, mdu_array_info_t
*info
,
3153 unsigned long long size
, char *name
,
3154 char *homehost
, int *uuid
)
3156 /* This is primarily called by Create when creating a new array.
3157 * We will then get add_to_super called for each component, and then
3158 * write_init_super called to write it out to each device.
3159 * For IMSM, Create can create on fresh devices or on a pre-existing
3161 * To create on a pre-existing array a different method will be called.
3162 * This one is just for fresh drives.
3164 struct intel_super
*super
;
3165 struct imsm_super
*mpb
;
3170 return init_super_imsm_volume(st
, info
, size
, name
, homehost
, uuid
);
3173 mpb_size
= disks_to_mpb_size(info
->nr_disks
);
3177 super
= alloc_super();
3178 if (super
&& posix_memalign(&super
->buf
, 512, mpb_size
) != 0) {
3183 fprintf(stderr
, Name
3184 ": %s could not allocate superblock\n", __func__
);
3187 memset(super
->buf
, 0, mpb_size
);
3189 mpb
->mpb_size
= __cpu_to_le32(mpb_size
);
3193 /* zeroing superblock */
3197 mpb
->attributes
= MPB_ATTRIB_CHECKSUM_VERIFY
;
3199 version
= (char *) mpb
->sig
;
3200 strcpy(version
, MPB_SIGNATURE
);
3201 version
+= strlen(MPB_SIGNATURE
);
3202 strcpy(version
, MPB_VERSION_RAID0
);
3208 static int add_to_super_imsm_volume(struct supertype
*st
, mdu_disk_info_t
*dk
,
3209 int fd
, char *devname
)
3211 struct intel_super
*super
= st
->sb
;
3212 struct imsm_super
*mpb
= super
->anchor
;
3214 struct imsm_dev
*dev
;
3215 struct imsm_map
*map
;
3218 dev
= get_imsm_dev(super
, super
->current_vol
);
3219 map
= get_imsm_map(dev
, 0);
3221 if (! (dk
->state
& (1<<MD_DISK_SYNC
))) {
3222 fprintf(stderr
, Name
": %s: Cannot add spare devices to IMSM volume\n",
3228 /* we're doing autolayout so grab the pre-marked (in
3229 * validate_geometry) raid_disk
3231 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
3232 if (dl
->raiddisk
== dk
->raid_disk
)
3235 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
3236 if (dl
->major
== dk
->major
&&
3237 dl
->minor
== dk
->minor
)
3242 fprintf(stderr
, Name
": %s is not a member of the same container\n", devname
);
3246 /* add a pristine spare to the metadata */
3247 if (dl
->index
< 0) {
3248 dl
->index
= super
->anchor
->num_disks
;
3249 super
->anchor
->num_disks
++;
3251 /* Check the device has not already been added */
3252 slot
= get_imsm_disk_slot(map
, dl
->index
);
3254 (get_imsm_ord_tbl_ent(dev
, slot
) & IMSM_ORD_REBUILD
) == 0) {
3255 fprintf(stderr
, Name
": %s has been included in this array twice\n",
3259 set_imsm_ord_tbl_ent(map
, dk
->number
, dl
->index
);
3260 dl
->disk
.status
= CONFIGURED_DISK
;
3262 /* if we are creating the first raid device update the family number */
3263 if (super
->current_vol
== 0) {
3265 struct imsm_dev
*_dev
= __get_imsm_dev(mpb
, 0);
3266 struct imsm_disk
*_disk
= __get_imsm_disk(mpb
, dl
->index
);
3268 if (!_dev
|| !_disk
) {
3269 fprintf(stderr
, Name
": BUG mpb setup error\n");
3275 sum
+= __gen_imsm_checksum(mpb
);
3276 mpb
->family_num
= __cpu_to_le32(sum
);
3277 mpb
->orig_family_num
= mpb
->family_num
;
3283 static int add_to_super_imsm(struct supertype
*st
, mdu_disk_info_t
*dk
,
3284 int fd
, char *devname
)
3286 struct intel_super
*super
= st
->sb
;
3288 unsigned long long size
;
3293 /* if we are on an RAID enabled platform check that the disk is
3294 * attached to the raid controller
3296 if (super
->hba
&& !disk_attached_to_hba(fd
, super
->hba
)) {
3298 Name
": %s is not attached to the raid controller: %s\n",
3299 devname
? : "disk", super
->hba
);
3303 if (super
->current_vol
>= 0)
3304 return add_to_super_imsm_volume(st
, dk
, fd
, devname
);
3307 dd
= malloc(sizeof(*dd
));
3310 Name
": malloc failed %s:%d.\n", __func__
, __LINE__
);
3313 memset(dd
, 0, sizeof(*dd
));
3314 dd
->major
= major(stb
.st_rdev
);
3315 dd
->minor
= minor(stb
.st_rdev
);
3317 dd
->devname
= devname
? strdup(devname
) : NULL
;
3320 rv
= imsm_read_serial(fd
, devname
, dd
->serial
);
3323 Name
": failed to retrieve scsi serial, aborting\n");
3328 get_dev_size(fd
, NULL
, &size
);
3330 serialcpy(dd
->disk
.serial
, dd
->serial
);
3331 dd
->disk
.total_blocks
= __cpu_to_le32(size
);
3332 dd
->disk
.status
= SPARE_DISK
;
3333 if (sysfs_disk_to_scsi_id(fd
, &id
) == 0)
3334 dd
->disk
.scsi_id
= __cpu_to_le32(id
);
3336 dd
->disk
.scsi_id
= __cpu_to_le32(0);
3338 if (st
->update_tail
) {
3339 dd
->next
= super
->add
;
3342 dd
->next
= super
->disks
;
3349 static int store_imsm_mpb(int fd
, struct imsm_super
*mpb
);
3353 struct imsm_super anchor
;
3354 } spare_record
__attribute__ ((aligned(512)));
3356 /* spare records have their own family number and do not have any defined raid
3359 static int write_super_imsm_spares(struct intel_super
*super
, int doclose
)
3361 struct imsm_super
*mpb
= super
->anchor
;
3362 struct imsm_super
*spare
= &spare_record
.anchor
;
3366 spare
->mpb_size
= __cpu_to_le32(sizeof(struct imsm_super
)),
3367 spare
->generation_num
= __cpu_to_le32(1UL),
3368 spare
->attributes
= MPB_ATTRIB_CHECKSUM_VERIFY
;
3369 spare
->num_disks
= 1,
3370 spare
->num_raid_devs
= 0,
3371 spare
->cache_size
= mpb
->cache_size
,
3372 spare
->pwr_cycle_count
= __cpu_to_le32(1),
3374 snprintf((char *) spare
->sig
, MAX_SIGNATURE_LENGTH
,
3375 MPB_SIGNATURE MPB_VERSION_RAID0
);
3377 for (d
= super
->disks
; d
; d
= d
->next
) {
3381 spare
->disk
[0] = d
->disk
;
3382 sum
= __gen_imsm_checksum(spare
);
3383 spare
->family_num
= __cpu_to_le32(sum
);
3384 spare
->orig_family_num
= 0;
3385 sum
= __gen_imsm_checksum(spare
);
3386 spare
->check_sum
= __cpu_to_le32(sum
);
3388 if (store_imsm_mpb(d
->fd
, spare
)) {
3389 fprintf(stderr
, "%s: failed for device %d:%d %s\n",
3390 __func__
, d
->major
, d
->minor
, strerror(errno
));
3402 static int write_super_imsm(struct intel_super
*super
, int doclose
)
3404 struct imsm_super
*mpb
= super
->anchor
;
3410 __u32 mpb_size
= sizeof(struct imsm_super
) - sizeof(struct imsm_disk
);
3412 /* 'generation' is incremented everytime the metadata is written */
3413 generation
= __le32_to_cpu(mpb
->generation_num
);
3415 mpb
->generation_num
= __cpu_to_le32(generation
);
3417 /* fix up cases where previous mdadm releases failed to set
3420 if (mpb
->orig_family_num
== 0)
3421 mpb
->orig_family_num
= mpb
->family_num
;
3423 mpb_size
+= sizeof(struct imsm_disk
) * mpb
->num_disks
;
3424 for (d
= super
->disks
; d
; d
= d
->next
) {
3428 mpb
->disk
[d
->index
] = d
->disk
;
3430 for (d
= super
->missing
; d
; d
= d
->next
)
3431 mpb
->disk
[d
->index
] = d
->disk
;
3433 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
3434 struct imsm_dev
*dev
= __get_imsm_dev(mpb
, i
);
3436 imsm_copy_dev(dev
, get_imsm_dev(super
, i
));
3437 mpb_size
+= sizeof_imsm_dev(dev
, 0);
3439 mpb_size
+= __le32_to_cpu(mpb
->bbm_log_size
);
3440 mpb
->mpb_size
= __cpu_to_le32(mpb_size
);
3442 /* recalculate checksum */
3443 sum
= __gen_imsm_checksum(mpb
);
3444 mpb
->check_sum
= __cpu_to_le32(sum
);
3446 /* write the mpb for disks that compose raid devices */
3447 for (d
= super
->disks
; d
; d
= d
->next
) {
3450 if (store_imsm_mpb(d
->fd
, mpb
))
3451 fprintf(stderr
, "%s: failed for device %d:%d %s\n",
3452 __func__
, d
->major
, d
->minor
, strerror(errno
));
3460 return write_super_imsm_spares(super
, doclose
);
3466 static int create_array(struct supertype
*st
, int dev_idx
)
3469 struct imsm_update_create_array
*u
;
3470 struct intel_super
*super
= st
->sb
;
3471 struct imsm_dev
*dev
= get_imsm_dev(super
, dev_idx
);
3472 struct imsm_map
*map
= get_imsm_map(dev
, 0);
3473 struct disk_info
*inf
;
3474 struct imsm_disk
*disk
;
3477 len
= sizeof(*u
) - sizeof(*dev
) + sizeof_imsm_dev(dev
, 0) +
3478 sizeof(*inf
) * map
->num_members
;
3481 fprintf(stderr
, "%s: failed to allocate update buffer\n",
3486 u
->type
= update_create_array
;
3487 u
->dev_idx
= dev_idx
;
3488 imsm_copy_dev(&u
->dev
, dev
);
3489 inf
= get_disk_info(u
);
3490 for (i
= 0; i
< map
->num_members
; i
++) {
3491 int idx
= get_imsm_disk_idx(dev
, i
);
3493 disk
= get_imsm_disk(super
, idx
);
3494 serialcpy(inf
[i
].serial
, disk
->serial
);
3496 append_metadata_update(st
, u
, len
);
3501 static int _add_disk(struct supertype
*st
)
3503 struct intel_super
*super
= st
->sb
;
3505 struct imsm_update_add_disk
*u
;
3513 fprintf(stderr
, "%s: failed to allocate update buffer\n",
3518 u
->type
= update_add_disk
;
3519 append_metadata_update(st
, u
, len
);
3524 static int write_init_super_imsm(struct supertype
*st
)
3526 struct intel_super
*super
= st
->sb
;
3527 int current_vol
= super
->current_vol
;
3529 /* we are done with current_vol reset it to point st at the container */
3530 super
->current_vol
= -1;
3532 if (st
->update_tail
) {
3533 /* queue the recently created array / added disk
3534 * as a metadata update */
3538 /* determine if we are creating a volume or adding a disk */
3539 if (current_vol
< 0) {
3540 /* in the add disk case we are running in mdmon
3541 * context, so don't close fd's
3543 return _add_disk(st
);
3545 rv
= create_array(st
, current_vol
);
3547 for (d
= super
->disks
; d
; d
= d
->next
) {
3555 for (d
= super
->disks
; d
; d
= d
->next
)
3556 Kill(d
->devname
, NULL
, 0, 1, 1);
3557 return write_super_imsm(st
->sb
, 1);
3562 static int store_super_imsm(struct supertype
*st
, int fd
)
3564 struct intel_super
*super
= st
->sb
;
3565 struct imsm_super
*mpb
= super
? super
->anchor
: NULL
;
3571 return store_imsm_mpb(fd
, mpb
);
3577 static int imsm_bbm_log_size(struct imsm_super
*mpb
)
3579 return __le32_to_cpu(mpb
->bbm_log_size
);
3583 static int validate_geometry_imsm_container(struct supertype
*st
, int level
,
3584 int layout
, int raiddisks
, int chunk
,
3585 unsigned long long size
, char *dev
,
3586 unsigned long long *freesize
,
3590 unsigned long long ldsize
;
3591 const struct imsm_orom
*orom
;
3593 if (level
!= LEVEL_CONTAINER
)
3598 if (check_env("IMSM_NO_PLATFORM"))
3601 orom
= find_imsm_orom();
3602 if (orom
&& raiddisks
> orom
->tds
) {
3604 fprintf(stderr
, Name
": %d exceeds maximum number of"
3605 " platform supported disks: %d\n",
3606 raiddisks
, orom
->tds
);
3610 fd
= open(dev
, O_RDONLY
|O_EXCL
, 0);
3613 fprintf(stderr
, Name
": imsm: Cannot open %s: %s\n",
3614 dev
, strerror(errno
));
3617 if (!get_dev_size(fd
, dev
, &ldsize
)) {
3623 *freesize
= avail_size_imsm(st
, ldsize
>> 9);
3628 static unsigned long long find_size(struct extent
*e
, int *idx
, int num_extents
)
3630 const unsigned long long base_start
= e
[*idx
].start
;
3631 unsigned long long end
= base_start
+ e
[*idx
].size
;
3634 if (base_start
== end
)
3638 for (i
= *idx
; i
< num_extents
; i
++) {
3639 /* extend overlapping extents */
3640 if (e
[i
].start
>= base_start
&&
3641 e
[i
].start
<= end
) {
3644 if (e
[i
].start
+ e
[i
].size
> end
)
3645 end
= e
[i
].start
+ e
[i
].size
;
3646 } else if (e
[i
].start
> end
) {
3652 return end
- base_start
;
3655 static unsigned long long merge_extents(struct intel_super
*super
, int sum_extents
)
3657 /* build a composite disk with all known extents and generate a new
3658 * 'maxsize' given the "all disks in an array must share a common start
3659 * offset" constraint
3661 struct extent
*e
= calloc(sum_extents
, sizeof(*e
));
3665 unsigned long long pos
;
3666 unsigned long long start
= 0;
3667 unsigned long long maxsize
;
3668 unsigned long reserve
;
3673 /* coalesce and sort all extents. also, check to see if we need to
3674 * reserve space between member arrays
3677 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
3680 for (i
= 0; i
< dl
->extent_cnt
; i
++)
3683 qsort(e
, sum_extents
, sizeof(*e
), cmp_extent
);
3688 while (i
< sum_extents
) {
3689 e
[j
].start
= e
[i
].start
;
3690 e
[j
].size
= find_size(e
, &i
, sum_extents
);
3692 if (e
[j
-1].size
== 0)
3701 unsigned long long esize
;
3703 esize
= e
[i
].start
- pos
;
3704 if (esize
>= maxsize
) {
3709 pos
= e
[i
].start
+ e
[i
].size
;
3711 } while (e
[i
-1].size
);
3717 /* FIXME assumes volume at offset 0 is the first volume in a
3720 if (start_extent
> 0)
3721 reserve
= IMSM_RESERVED_SECTORS
; /* gap between raid regions */
3725 if (maxsize
< reserve
)
3728 super
->create_offset
= ~((__u32
) 0);
3729 if (start
+ reserve
> super
->create_offset
)
3730 return 0; /* start overflows create_offset */
3731 super
->create_offset
= start
+ reserve
;
3733 return maxsize
- reserve
;
3736 static int is_raid_level_supported(const struct imsm_orom
*orom
, int level
, int raiddisks
)
3738 if (level
< 0 || level
== 6 || level
== 4)
3741 /* if we have an orom prevent invalid raid levels */
3744 case 0: return imsm_orom_has_raid0(orom
);
3747 return imsm_orom_has_raid1e(orom
);
3748 return imsm_orom_has_raid1(orom
) && raiddisks
== 2;
3749 case 10: return imsm_orom_has_raid10(orom
) && raiddisks
== 4;
3750 case 5: return imsm_orom_has_raid5(orom
) && raiddisks
> 2;
3753 return 1; /* not on an Intel RAID platform so anything goes */
3758 #define pr_vrb(fmt, arg...) (void) (verbose && fprintf(stderr, Name fmt, ##arg))
3760 validate_geometry_imsm_orom(struct intel_super
*super
, int level
, int layout
,
3761 int raiddisks
, int chunk
, int verbose
)
3763 if (!is_raid_level_supported(super
->orom
, level
, raiddisks
)) {
3764 pr_vrb(": platform does not support raid%d with %d disk%s\n",
3765 level
, raiddisks
, raiddisks
> 1 ? "s" : "");
3768 if (super
->orom
&& level
!= 1 &&
3769 !imsm_orom_has_chunk(super
->orom
, chunk
)) {
3770 pr_vrb(": platform does not support a chunk size of: %d\n", chunk
);
3773 if (layout
!= imsm_level_to_layout(level
)) {
3775 pr_vrb(": imsm raid 5 only supports the left-asymmetric layout\n");
3776 else if (level
== 10)
3777 pr_vrb(": imsm raid 10 only supports the n2 layout\n");
3779 pr_vrb(": imsm unknown layout %#x for this raid level %d\n",
3787 /* validate_geometry_imsm_volume - lifted from validate_geometry_ddf_bvd
3788 * FIX ME add ahci details
3790 static int validate_geometry_imsm_volume(struct supertype
*st
, int level
,
3791 int layout
, int raiddisks
, int chunk
,
3792 unsigned long long size
, char *dev
,
3793 unsigned long long *freesize
,
3797 struct intel_super
*super
= st
->sb
;
3798 struct imsm_super
*mpb
= super
->anchor
;
3800 unsigned long long pos
= 0;
3801 unsigned long long maxsize
;
3805 /* We must have the container info already read in. */
3809 if (!validate_geometry_imsm_orom(super
, level
, layout
, raiddisks
, chunk
, verbose
))
3813 /* General test: make sure there is space for
3814 * 'raiddisks' device extents of size 'size' at a given
3817 unsigned long long minsize
= size
;
3818 unsigned long long start_offset
= MaxSector
;
3821 minsize
= MPB_SECTOR_CNT
+ IMSM_RESERVED_SECTORS
;
3822 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
3827 e
= get_extents(super
, dl
);
3830 unsigned long long esize
;
3831 esize
= e
[i
].start
- pos
;
3832 if (esize
>= minsize
)
3834 if (found
&& start_offset
== MaxSector
) {
3837 } else if (found
&& pos
!= start_offset
) {
3841 pos
= e
[i
].start
+ e
[i
].size
;
3843 } while (e
[i
-1].size
);
3848 if (dcnt
< raiddisks
) {
3850 fprintf(stderr
, Name
": imsm: Not enough "
3851 "devices with space for this array "
3859 /* This device must be a member of the set */
3860 if (stat(dev
, &stb
) < 0)
3862 if ((S_IFMT
& stb
.st_mode
) != S_IFBLK
)
3864 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
3865 if (dl
->major
== (int)major(stb
.st_rdev
) &&
3866 dl
->minor
== (int)minor(stb
.st_rdev
))
3871 fprintf(stderr
, Name
": %s is not in the "
3872 "same imsm set\n", dev
);
3874 } else if (super
->orom
&& dl
->index
< 0 && mpb
->num_raid_devs
) {
3875 /* If a volume is present then the current creation attempt
3876 * cannot incorporate new spares because the orom may not
3877 * understand this configuration (all member disks must be
3878 * members of each array in the container).
3880 fprintf(stderr
, Name
": %s is a spare and a volume"
3881 " is already defined for this container\n", dev
);
3882 fprintf(stderr
, Name
": The option-rom requires all member"
3883 " disks to be a member of all volumes\n");
3887 /* retrieve the largest free space block */
3888 e
= get_extents(super
, dl
);
3893 unsigned long long esize
;
3895 esize
= e
[i
].start
- pos
;
3896 if (esize
>= maxsize
)
3898 pos
= e
[i
].start
+ e
[i
].size
;
3900 } while (e
[i
-1].size
);
3905 fprintf(stderr
, Name
": unable to determine free space for: %s\n",
3909 if (maxsize
< size
) {
3911 fprintf(stderr
, Name
": %s not enough space (%llu < %llu)\n",
3912 dev
, maxsize
, size
);
3916 /* count total number of extents for merge */
3918 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
3920 i
+= dl
->extent_cnt
;
3922 maxsize
= merge_extents(super
, i
);
3923 if (maxsize
< size
|| maxsize
== 0) {
3925 fprintf(stderr
, Name
": not enough space after merge (%llu < %llu)\n",
3930 *freesize
= maxsize
;
3935 static int reserve_space(struct supertype
*st
, int raiddisks
,
3936 unsigned long long size
, int chunk
,
3937 unsigned long long *freesize
)
3939 struct intel_super
*super
= st
->sb
;
3940 struct imsm_super
*mpb
= super
->anchor
;
3945 unsigned long long maxsize
;
3946 unsigned long long minsize
;
3950 /* find the largest common start free region of the possible disks */
3954 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
3960 /* don't activate new spares if we are orom constrained
3961 * and there is already a volume active in the container
3963 if (super
->orom
&& dl
->index
< 0 && mpb
->num_raid_devs
)
3966 e
= get_extents(super
, dl
);
3969 for (i
= 1; e
[i
-1].size
; i
++)
3977 maxsize
= merge_extents(super
, extent_cnt
);
3982 if (cnt
< raiddisks
||
3983 (super
->orom
&& used
&& used
!= raiddisks
) ||
3984 maxsize
< minsize
||
3986 fprintf(stderr
, Name
": not enough devices with space to create array.\n");
3987 return 0; /* No enough free spaces large enough */
3999 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
4001 dl
->raiddisk
= cnt
++;
4008 static int validate_geometry_imsm(struct supertype
*st
, int level
, int layout
,
4009 int raiddisks
, int chunk
, unsigned long long size
,
4010 char *dev
, unsigned long long *freesize
,
4017 /* if given unused devices create a container
4018 * if given given devices in a container create a member volume
4020 if (level
== LEVEL_CONTAINER
) {
4021 /* Must be a fresh device to add to a container */
4022 return validate_geometry_imsm_container(st
, level
, layout
,
4023 raiddisks
, chunk
, size
,
4029 if (st
->sb
&& freesize
) {
4030 /* we are being asked to automatically layout a
4031 * new volume based on the current contents of
4032 * the container. If the the parameters can be
4033 * satisfied reserve_space will record the disks,
4034 * start offset, and size of the volume to be
4035 * created. add_to_super and getinfo_super
4036 * detect when autolayout is in progress.
4038 if (!validate_geometry_imsm_orom(st
->sb
, level
, layout
,
4042 return reserve_space(st
, raiddisks
, size
, chunk
, freesize
);
4047 /* creating in a given container */
4048 return validate_geometry_imsm_volume(st
, level
, layout
,
4049 raiddisks
, chunk
, size
,
4050 dev
, freesize
, verbose
);
4053 /* This device needs to be a device in an 'imsm' container */
4054 fd
= open(dev
, O_RDONLY
|O_EXCL
, 0);
4058 Name
": Cannot create this array on device %s\n",
4063 if (errno
!= EBUSY
|| (fd
= open(dev
, O_RDONLY
, 0)) < 0) {
4065 fprintf(stderr
, Name
": Cannot open %s: %s\n",
4066 dev
, strerror(errno
));
4069 /* Well, it is in use by someone, maybe an 'imsm' container. */
4070 cfd
= open_container(fd
);
4074 fprintf(stderr
, Name
": Cannot use %s: It is busy\n",
4078 sra
= sysfs_read(cfd
, 0, GET_VERSION
);
4079 if (sra
&& sra
->array
.major_version
== -1 &&
4080 strcmp(sra
->text_version
, "imsm") == 0)
4084 /* This is a member of a imsm container. Load the container
4085 * and try to create a volume
4087 struct intel_super
*super
;
4089 if (load_super_imsm_all(st
, cfd
, (void **) &super
, NULL
) == 0) {
4091 st
->container_dev
= fd2devnum(cfd
);
4093 return validate_geometry_imsm_volume(st
, level
, layout
,
4101 fprintf(stderr
, Name
": failed container membership check\n");
4107 static int default_chunk_imsm(struct supertype
*st
)
4109 struct intel_super
*super
= st
->sb
;
4114 return imsm_orom_default_chunk(super
->orom
);
4117 static void handle_missing(struct intel_super
*super
, struct imsm_dev
*dev
);
4119 static int kill_subarray_imsm(struct supertype
*st
)
4121 /* remove the subarray currently referenced by ->current_vol */
4123 struct intel_dev
**dp
;
4124 struct intel_super
*super
= st
->sb
;
4125 __u8 current_vol
= super
->current_vol
;
4126 struct imsm_super
*mpb
= super
->anchor
;
4128 if (super
->current_vol
< 0)
4130 super
->current_vol
= -1; /* invalidate subarray cursor */
4132 /* block deletions that would change the uuid of active subarrays
4134 * FIXME when immutable ids are available, but note that we'll
4135 * also need to fixup the invalidated/active subarray indexes in
4138 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
4141 if (i
< current_vol
)
4143 sprintf(subarray
, "%u", i
);
4144 if (is_subarray_active(subarray
, st
->devname
)) {
4146 Name
": deleting subarray-%d would change the UUID of active subarray-%d, aborting\n",
4153 if (st
->update_tail
) {
4154 struct imsm_update_kill_array
*u
= malloc(sizeof(*u
));
4158 u
->type
= update_kill_array
;
4159 u
->dev_idx
= current_vol
;
4160 append_metadata_update(st
, u
, sizeof(*u
));
4165 for (dp
= &super
->devlist
; *dp
;)
4166 if ((*dp
)->index
== current_vol
) {
4169 handle_missing(super
, (*dp
)->dev
);
4170 if ((*dp
)->index
> current_vol
)
4175 /* no more raid devices, all active components are now spares,
4176 * but of course failed are still failed
4178 if (--mpb
->num_raid_devs
== 0) {
4181 for (d
= super
->disks
; d
; d
= d
->next
)
4182 if (d
->index
> -2) {
4184 d
->disk
.status
= SPARE_DISK
;
4188 super
->updates_pending
++;
4193 static int update_subarray_imsm(struct supertype
*st
, char *subarray
,
4194 char *update
, struct mddev_ident
*ident
)
4196 /* update the subarray currently referenced by ->current_vol */
4197 struct intel_super
*super
= st
->sb
;
4198 struct imsm_super
*mpb
= super
->anchor
;
4200 if (strcmp(update
, "name") == 0) {
4201 char *name
= ident
->name
;
4205 if (is_subarray_active(subarray
, st
->devname
)) {
4207 Name
": Unable to update name of active subarray\n");
4211 if (!check_name(super
, name
, 0))
4214 vol
= strtoul(subarray
, &ep
, 10);
4215 if (*ep
!= '\0' || vol
>= super
->anchor
->num_raid_devs
)
4218 if (st
->update_tail
) {
4219 struct imsm_update_rename_array
*u
= malloc(sizeof(*u
));
4223 u
->type
= update_rename_array
;
4225 snprintf((char *) u
->name
, MAX_RAID_SERIAL_LEN
, "%s", name
);
4226 append_metadata_update(st
, u
, sizeof(*u
));
4228 struct imsm_dev
*dev
;
4231 dev
= get_imsm_dev(super
, vol
);
4232 snprintf((char *) dev
->volume
, MAX_RAID_SERIAL_LEN
, "%s", name
);
4233 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
4234 dev
= get_imsm_dev(super
, i
);
4235 handle_missing(super
, dev
);
4237 super
->updates_pending
++;
4244 #endif /* MDASSEMBLE */
4246 static int is_rebuilding(struct imsm_dev
*dev
)
4248 struct imsm_map
*migr_map
;
4250 if (!dev
->vol
.migr_state
)
4253 if (migr_type(dev
) != MIGR_REBUILD
)
4256 migr_map
= get_imsm_map(dev
, 1);
4258 if (migr_map
->map_state
== IMSM_T_STATE_DEGRADED
)
4264 static void update_recovery_start(struct imsm_dev
*dev
, struct mdinfo
*array
)
4266 struct mdinfo
*rebuild
= NULL
;
4270 if (!is_rebuilding(dev
))
4273 /* Find the rebuild target, but punt on the dual rebuild case */
4274 for (d
= array
->devs
; d
; d
= d
->next
)
4275 if (d
->recovery_start
== 0) {
4282 /* (?) none of the disks are marked with
4283 * IMSM_ORD_REBUILD, so assume they are missing and the
4284 * disk_ord_tbl was not correctly updated
4286 dprintf("%s: failed to locate out-of-sync disk\n", __func__
);
4290 units
= __le32_to_cpu(dev
->vol
.curr_migr_unit
);
4291 rebuild
->recovery_start
= units
* blocks_per_migr_unit(dev
);
4295 static struct mdinfo
*container_content_imsm(struct supertype
*st
, char *subarray
)
4297 /* Given a container loaded by load_super_imsm_all,
4298 * extract information about all the arrays into
4300 * If 'subarray' is given, just extract info about that array.
4302 * For each imsm_dev create an mdinfo, fill it in,
4303 * then look for matching devices in super->disks
4304 * and create appropriate device mdinfo.
4306 struct intel_super
*super
= st
->sb
;
4307 struct imsm_super
*mpb
= super
->anchor
;
4308 struct mdinfo
*rest
= NULL
;
4311 /* do not assemble arrays that might have bad blocks */
4312 if (imsm_bbm_log_size(super
->anchor
)) {
4313 fprintf(stderr
, Name
": BBM log found in metadata. "
4314 "Cannot activate array(s).\n");
4318 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
4319 struct imsm_dev
*dev
;
4320 struct imsm_map
*map
;
4321 struct mdinfo
*this;
4326 (i
!= strtoul(subarray
, &ep
, 10) || *ep
!= '\0'))
4329 dev
= get_imsm_dev(super
, i
);
4330 map
= get_imsm_map(dev
, 0);
4332 /* do not publish arrays that are in the middle of an
4333 * unsupported migration
4335 if (dev
->vol
.migr_state
&&
4336 (migr_type(dev
) == MIGR_GEN_MIGR
||
4337 migr_type(dev
) == MIGR_STATE_CHANGE
)) {
4338 fprintf(stderr
, Name
": cannot assemble volume '%.16s':"
4339 " unsupported migration in progress\n",
4344 this = malloc(sizeof(*this));
4346 fprintf(stderr
, Name
": failed to allocate %zu bytes\n",
4350 memset(this, 0, sizeof(*this));
4353 super
->current_vol
= i
;
4354 getinfo_super_imsm_volume(st
, this, NULL
);
4355 for (slot
= 0 ; slot
< map
->num_members
; slot
++) {
4356 unsigned long long recovery_start
;
4357 struct mdinfo
*info_d
;
4364 idx
= get_imsm_disk_idx(dev
, slot
);
4365 ord
= get_imsm_ord_tbl_ent(dev
, slot
);
4366 for (d
= super
->disks
; d
; d
= d
->next
)
4367 if (d
->index
== idx
)
4370 recovery_start
= MaxSector
;
4373 if (d
&& is_failed(&d
->disk
))
4375 if (ord
& IMSM_ORD_REBUILD
)
4379 * if we skip some disks the array will be assmebled degraded;
4380 * reset resync start to avoid a dirty-degraded
4381 * situation when performing the intial sync
4383 * FIXME handle dirty degraded
4385 if ((skip
|| recovery_start
== 0) && !dev
->vol
.dirty
)
4386 this->resync_start
= MaxSector
;
4390 info_d
= calloc(1, sizeof(*info_d
));
4392 fprintf(stderr
, Name
": failed to allocate disk"
4393 " for volume %.16s\n", dev
->volume
);
4394 info_d
= this->devs
;
4396 struct mdinfo
*d
= info_d
->next
;
4405 info_d
->next
= this->devs
;
4406 this->devs
= info_d
;
4408 info_d
->disk
.number
= d
->index
;
4409 info_d
->disk
.major
= d
->major
;
4410 info_d
->disk
.minor
= d
->minor
;
4411 info_d
->disk
.raid_disk
= slot
;
4412 info_d
->recovery_start
= recovery_start
;
4414 if (info_d
->recovery_start
== MaxSector
)
4415 this->array
.working_disks
++;
4417 info_d
->events
= __le32_to_cpu(mpb
->generation_num
);
4418 info_d
->data_offset
= __le32_to_cpu(map
->pba_of_lba0
);
4419 info_d
->component_size
= __le32_to_cpu(map
->blocks_per_member
);
4421 /* now that the disk list is up-to-date fixup recovery_start */
4422 update_recovery_start(dev
, this);
4430 static __u8
imsm_check_degraded(struct intel_super
*super
, struct imsm_dev
*dev
, int failed
)
4432 struct imsm_map
*map
= get_imsm_map(dev
, 0);
4435 return map
->map_state
== IMSM_T_STATE_UNINITIALIZED
?
4436 IMSM_T_STATE_UNINITIALIZED
: IMSM_T_STATE_NORMAL
;
4438 switch (get_imsm_raid_level(map
)) {
4440 return IMSM_T_STATE_FAILED
;
4443 if (failed
< map
->num_members
)
4444 return IMSM_T_STATE_DEGRADED
;
4446 return IMSM_T_STATE_FAILED
;
4451 * check to see if any mirrors have failed, otherwise we
4452 * are degraded. Even numbered slots are mirrored on
4456 /* gcc -Os complains that this is unused */
4457 int insync
= insync
;
4459 for (i
= 0; i
< map
->num_members
; i
++) {
4460 __u32 ord
= get_imsm_ord_tbl_ent(dev
, i
);
4461 int idx
= ord_to_idx(ord
);
4462 struct imsm_disk
*disk
;
4464 /* reset the potential in-sync count on even-numbered
4465 * slots. num_copies is always 2 for imsm raid10
4470 disk
= get_imsm_disk(super
, idx
);
4471 if (!disk
|| is_failed(disk
) || ord
& IMSM_ORD_REBUILD
)
4474 /* no in-sync disks left in this mirror the
4478 return IMSM_T_STATE_FAILED
;
4481 return IMSM_T_STATE_DEGRADED
;
4485 return IMSM_T_STATE_DEGRADED
;
4487 return IMSM_T_STATE_FAILED
;
4493 return map
->map_state
;
4496 static int imsm_count_failed(struct intel_super
*super
, struct imsm_dev
*dev
)
4500 struct imsm_disk
*disk
;
4501 struct imsm_map
*map
= get_imsm_map(dev
, 0);
4502 struct imsm_map
*prev
= get_imsm_map(dev
, dev
->vol
.migr_state
);
4506 /* at the beginning of migration we set IMSM_ORD_REBUILD on
4507 * disks that are being rebuilt. New failures are recorded to
4508 * map[0]. So we look through all the disks we started with and
4509 * see if any failures are still present, or if any new ones
4512 * FIXME add support for online capacity expansion and
4513 * raid-level-migration
4515 for (i
= 0; i
< prev
->num_members
; i
++) {
4516 ord
= __le32_to_cpu(prev
->disk_ord_tbl
[i
]);
4517 ord
|= __le32_to_cpu(map
->disk_ord_tbl
[i
]);
4518 idx
= ord_to_idx(ord
);
4520 disk
= get_imsm_disk(super
, idx
);
4521 if (!disk
|| is_failed(disk
) || ord
& IMSM_ORD_REBUILD
)
4529 static int imsm_open_new(struct supertype
*c
, struct active_array
*a
,
4532 struct intel_super
*super
= c
->sb
;
4533 struct imsm_super
*mpb
= super
->anchor
;
4535 if (atoi(inst
) >= mpb
->num_raid_devs
) {
4536 fprintf(stderr
, "%s: subarry index %d, out of range\n",
4537 __func__
, atoi(inst
));
4541 dprintf("imsm: open_new %s\n", inst
);
4542 a
->info
.container_member
= atoi(inst
);
4546 static int is_resyncing(struct imsm_dev
*dev
)
4548 struct imsm_map
*migr_map
;
4550 if (!dev
->vol
.migr_state
)
4553 if (migr_type(dev
) == MIGR_INIT
||
4554 migr_type(dev
) == MIGR_REPAIR
)
4557 migr_map
= get_imsm_map(dev
, 1);
4559 if (migr_map
->map_state
== IMSM_T_STATE_NORMAL
)
4565 /* return true if we recorded new information */
4566 static int mark_failure(struct imsm_dev
*dev
, struct imsm_disk
*disk
, int idx
)
4570 struct imsm_map
*map
;
4572 /* new failures are always set in map[0] */
4573 map
= get_imsm_map(dev
, 0);
4575 slot
= get_imsm_disk_slot(map
, idx
);
4579 ord
= __le32_to_cpu(map
->disk_ord_tbl
[slot
]);
4580 if (is_failed(disk
) && (ord
& IMSM_ORD_REBUILD
))
4583 disk
->status
|= FAILED_DISK
;
4584 disk
->status
&= ~CONFIGURED_DISK
;
4585 set_imsm_ord_tbl_ent(map
, slot
, idx
| IMSM_ORD_REBUILD
);
4586 if (map
->failed_disk_num
== 0xff)
4587 map
->failed_disk_num
= slot
;
4591 static void mark_missing(struct imsm_dev
*dev
, struct imsm_disk
*disk
, int idx
)
4593 mark_failure(dev
, disk
, idx
);
4595 if (disk
->scsi_id
== __cpu_to_le32(~(__u32
)0))
4598 disk
->scsi_id
= __cpu_to_le32(~(__u32
)0);
4599 memmove(&disk
->serial
[0], &disk
->serial
[1], MAX_RAID_SERIAL_LEN
- 1);
4602 static void handle_missing(struct intel_super
*super
, struct imsm_dev
*dev
)
4608 if (!super
->missing
)
4610 failed
= imsm_count_failed(super
, dev
);
4611 map_state
= imsm_check_degraded(super
, dev
, failed
);
4613 dprintf("imsm: mark missing\n");
4614 end_migration(dev
, map_state
);
4615 for (dl
= super
->missing
; dl
; dl
= dl
->next
)
4616 mark_missing(dev
, &dl
->disk
, dl
->index
);
4617 super
->updates_pending
++;
4620 /* Handle dirty -> clean transititions and resync. Degraded and rebuild
4621 * states are handled in imsm_set_disk() with one exception, when a
4622 * resync is stopped due to a new failure this routine will set the
4623 * 'degraded' state for the array.
4625 static int imsm_set_array_state(struct active_array
*a
, int consistent
)
4627 int inst
= a
->info
.container_member
;
4628 struct intel_super
*super
= a
->container
->sb
;
4629 struct imsm_dev
*dev
= get_imsm_dev(super
, inst
);
4630 struct imsm_map
*map
= get_imsm_map(dev
, 0);
4631 int failed
= imsm_count_failed(super
, dev
);
4632 __u8 map_state
= imsm_check_degraded(super
, dev
, failed
);
4633 __u32 blocks_per_unit
;
4635 /* before we activate this array handle any missing disks */
4636 if (consistent
== 2)
4637 handle_missing(super
, dev
);
4639 if (consistent
== 2 &&
4640 (!is_resync_complete(&a
->info
) ||
4641 map_state
!= IMSM_T_STATE_NORMAL
||
4642 dev
->vol
.migr_state
))
4645 if (is_resync_complete(&a
->info
)) {
4646 /* complete intialization / resync,
4647 * recovery and interrupted recovery is completed in
4650 if (is_resyncing(dev
)) {
4651 dprintf("imsm: mark resync done\n");
4652 end_migration(dev
, map_state
);
4653 super
->updates_pending
++;
4654 a
->last_checkpoint
= 0;
4656 } else if (!is_resyncing(dev
) && !failed
) {
4657 /* mark the start of the init process if nothing is failed */
4658 dprintf("imsm: mark resync start\n");
4659 if (map
->map_state
== IMSM_T_STATE_UNINITIALIZED
)
4660 migrate(dev
, IMSM_T_STATE_NORMAL
, MIGR_INIT
);
4662 migrate(dev
, IMSM_T_STATE_NORMAL
, MIGR_REPAIR
);
4663 super
->updates_pending
++;
4666 /* check if we can update curr_migr_unit from resync_start, recovery_start */
4667 blocks_per_unit
= blocks_per_migr_unit(dev
);
4668 if (blocks_per_unit
) {
4672 units
= a
->last_checkpoint
/ blocks_per_unit
;
4675 /* check that we did not overflow 32-bits, and that
4676 * curr_migr_unit needs updating
4678 if (units32
== units
&&
4679 __le32_to_cpu(dev
->vol
.curr_migr_unit
) != units32
) {
4680 dprintf("imsm: mark checkpoint (%u)\n", units32
);
4681 dev
->vol
.curr_migr_unit
= __cpu_to_le32(units32
);
4682 super
->updates_pending
++;
4686 /* mark dirty / clean */
4687 if (dev
->vol
.dirty
!= !consistent
) {
4688 dprintf("imsm: mark '%s'\n", consistent
? "clean" : "dirty");
4693 super
->updates_pending
++;
4698 static void imsm_set_disk(struct active_array
*a
, int n
, int state
)
4700 int inst
= a
->info
.container_member
;
4701 struct intel_super
*super
= a
->container
->sb
;
4702 struct imsm_dev
*dev
= get_imsm_dev(super
, inst
);
4703 struct imsm_map
*map
= get_imsm_map(dev
, 0);
4704 struct imsm_disk
*disk
;
4709 if (n
> map
->num_members
)
4710 fprintf(stderr
, "imsm: set_disk %d out of range 0..%d\n",
4711 n
, map
->num_members
- 1);
4716 dprintf("imsm: set_disk %d:%x\n", n
, state
);
4718 ord
= get_imsm_ord_tbl_ent(dev
, n
);
4719 disk
= get_imsm_disk(super
, ord_to_idx(ord
));
4721 /* check for new failures */
4722 if (state
& DS_FAULTY
) {
4723 if (mark_failure(dev
, disk
, ord_to_idx(ord
)))
4724 super
->updates_pending
++;
4727 /* check if in_sync */
4728 if (state
& DS_INSYNC
&& ord
& IMSM_ORD_REBUILD
&& is_rebuilding(dev
)) {
4729 struct imsm_map
*migr_map
= get_imsm_map(dev
, 1);
4731 set_imsm_ord_tbl_ent(migr_map
, n
, ord_to_idx(ord
));
4732 super
->updates_pending
++;
4735 failed
= imsm_count_failed(super
, dev
);
4736 map_state
= imsm_check_degraded(super
, dev
, failed
);
4738 /* check if recovery complete, newly degraded, or failed */
4739 if (map_state
== IMSM_T_STATE_NORMAL
&& is_rebuilding(dev
)) {
4740 end_migration(dev
, map_state
);
4741 map
= get_imsm_map(dev
, 0);
4742 map
->failed_disk_num
= ~0;
4743 super
->updates_pending
++;
4744 a
->last_checkpoint
= 0;
4745 } else if (map_state
== IMSM_T_STATE_DEGRADED
&&
4746 map
->map_state
!= map_state
&&
4747 !dev
->vol
.migr_state
) {
4748 dprintf("imsm: mark degraded\n");
4749 map
->map_state
= map_state
;
4750 super
->updates_pending
++;
4751 a
->last_checkpoint
= 0;
4752 } else if (map_state
== IMSM_T_STATE_FAILED
&&
4753 map
->map_state
!= map_state
) {
4754 dprintf("imsm: mark failed\n");
4755 end_migration(dev
, map_state
);
4756 super
->updates_pending
++;
4757 a
->last_checkpoint
= 0;
4761 static int store_imsm_mpb(int fd
, struct imsm_super
*mpb
)
4764 __u32 mpb_size
= __le32_to_cpu(mpb
->mpb_size
);
4765 unsigned long long dsize
;
4766 unsigned long long sectors
;
4768 get_dev_size(fd
, NULL
, &dsize
);
4770 if (mpb_size
> 512) {
4771 /* -1 to account for anchor */
4772 sectors
= mpb_sectors(mpb
) - 1;
4774 /* write the extended mpb to the sectors preceeding the anchor */
4775 if (lseek64(fd
, dsize
- (512 * (2 + sectors
)), SEEK_SET
) < 0)
4778 if ((unsigned long long)write(fd
, buf
+ 512, 512 * sectors
)
4783 /* first block is stored on second to last sector of the disk */
4784 if (lseek64(fd
, dsize
- (512 * 2), SEEK_SET
) < 0)
4787 if (write(fd
, buf
, 512) != 512)
4793 static void imsm_sync_metadata(struct supertype
*container
)
4795 struct intel_super
*super
= container
->sb
;
4797 if (!super
->updates_pending
)
4800 write_super_imsm(super
, 0);
4802 super
->updates_pending
= 0;
4805 static struct dl
*imsm_readd(struct intel_super
*super
, int idx
, struct active_array
*a
)
4807 struct imsm_dev
*dev
= get_imsm_dev(super
, a
->info
.container_member
);
4808 int i
= get_imsm_disk_idx(dev
, idx
);
4811 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
4815 if (dl
&& is_failed(&dl
->disk
))
4819 dprintf("%s: found %x:%x\n", __func__
, dl
->major
, dl
->minor
);
4824 static struct dl
*imsm_add_spare(struct intel_super
*super
, int slot
,
4825 struct active_array
*a
, int activate_new
)
4827 struct imsm_dev
*dev
= get_imsm_dev(super
, a
->info
.container_member
);
4828 int idx
= get_imsm_disk_idx(dev
, slot
);
4829 struct imsm_super
*mpb
= super
->anchor
;
4830 struct imsm_map
*map
;
4831 unsigned long long pos
;
4836 __u32 array_start
= 0;
4837 __u32 array_end
= 0;
4840 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
4841 /* If in this array, skip */
4842 for (d
= a
->info
.devs
; d
; d
= d
->next
)
4843 if (d
->state_fd
>= 0 &&
4844 d
->disk
.major
== dl
->major
&&
4845 d
->disk
.minor
== dl
->minor
) {
4846 dprintf("%x:%x already in array\n", dl
->major
, dl
->minor
);
4852 /* skip in use or failed drives */
4853 if (is_failed(&dl
->disk
) || idx
== dl
->index
||
4855 dprintf("%x:%x status (failed: %d index: %d)\n",
4856 dl
->major
, dl
->minor
, is_failed(&dl
->disk
), idx
);
4860 /* skip pure spares when we are looking for partially
4861 * assimilated drives
4863 if (dl
->index
== -1 && !activate_new
)
4866 /* Does this unused device have the requisite free space?
4867 * It needs to be able to cover all member volumes
4869 ex
= get_extents(super
, dl
);
4871 dprintf("cannot get extents\n");
4874 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
4875 dev
= get_imsm_dev(super
, i
);
4876 map
= get_imsm_map(dev
, 0);
4878 /* check if this disk is already a member of
4881 if (get_imsm_disk_slot(map
, dl
->index
) >= 0)
4887 array_start
= __le32_to_cpu(map
->pba_of_lba0
);
4888 array_end
= array_start
+
4889 __le32_to_cpu(map
->blocks_per_member
) - 1;
4892 /* check that we can start at pba_of_lba0 with
4893 * blocks_per_member of space
4895 if (array_start
>= pos
&& array_end
< ex
[j
].start
) {
4899 pos
= ex
[j
].start
+ ex
[j
].size
;
4901 } while (ex
[j
-1].size
);
4908 if (i
< mpb
->num_raid_devs
) {
4909 dprintf("%x:%x does not have %u to %u available\n",
4910 dl
->major
, dl
->minor
, array_start
, array_end
);
4920 static struct mdinfo
*imsm_activate_spare(struct active_array
*a
,
4921 struct metadata_update
**updates
)
4924 * Find a device with unused free space and use it to replace a
4925 * failed/vacant region in an array. We replace failed regions one a
4926 * array at a time. The result is that a new spare disk will be added
4927 * to the first failed array and after the monitor has finished
4928 * propagating failures the remainder will be consumed.
4930 * FIXME add a capability for mdmon to request spares from another
4934 struct intel_super
*super
= a
->container
->sb
;
4935 int inst
= a
->info
.container_member
;
4936 struct imsm_dev
*dev
= get_imsm_dev(super
, inst
);
4937 struct imsm_map
*map
= get_imsm_map(dev
, 0);
4938 int failed
= a
->info
.array
.raid_disks
;
4939 struct mdinfo
*rv
= NULL
;
4942 struct metadata_update
*mu
;
4944 struct imsm_update_activate_spare
*u
;
4948 for (d
= a
->info
.devs
; d
; d
= d
->next
) {
4949 if ((d
->curr_state
& DS_FAULTY
) &&
4951 /* wait for Removal to happen */
4953 if (d
->state_fd
>= 0)
4957 dprintf("imsm: activate spare: inst=%d failed=%d (%d) level=%d\n",
4958 inst
, failed
, a
->info
.array
.raid_disks
, a
->info
.array
.level
);
4959 if (imsm_check_degraded(super
, dev
, failed
) != IMSM_T_STATE_DEGRADED
)
4962 /* For each slot, if it is not working, find a spare */
4963 for (i
= 0; i
< a
->info
.array
.raid_disks
; i
++) {
4964 for (d
= a
->info
.devs
; d
; d
= d
->next
)
4965 if (d
->disk
.raid_disk
== i
)
4967 dprintf("found %d: %p %x\n", i
, d
, d
?d
->curr_state
:0);
4968 if (d
&& (d
->state_fd
>= 0))
4972 * OK, this device needs recovery. Try to re-add the
4973 * previous occupant of this slot, if this fails see if
4974 * we can continue the assimilation of a spare that was
4975 * partially assimilated, finally try to activate a new
4978 dl
= imsm_readd(super
, i
, a
);
4980 dl
= imsm_add_spare(super
, i
, a
, 0);
4982 dl
= imsm_add_spare(super
, i
, a
, 1);
4986 /* found a usable disk with enough space */
4987 di
= malloc(sizeof(*di
));
4990 memset(di
, 0, sizeof(*di
));
4992 /* dl->index will be -1 in the case we are activating a
4993 * pristine spare. imsm_process_update() will create a
4994 * new index in this case. Once a disk is found to be
4995 * failed in all member arrays it is kicked from the
4998 di
->disk
.number
= dl
->index
;
5000 /* (ab)use di->devs to store a pointer to the device
5003 di
->devs
= (struct mdinfo
*) dl
;
5005 di
->disk
.raid_disk
= i
;
5006 di
->disk
.major
= dl
->major
;
5007 di
->disk
.minor
= dl
->minor
;
5009 di
->recovery_start
= 0;
5010 di
->data_offset
= __le32_to_cpu(map
->pba_of_lba0
);
5011 di
->component_size
= a
->info
.component_size
;
5012 di
->container_member
= inst
;
5013 super
->random
= random32();
5017 dprintf("%x:%x to be %d at %llu\n", dl
->major
, dl
->minor
,
5018 i
, di
->data_offset
);
5024 /* No spares found */
5026 /* Now 'rv' has a list of devices to return.
5027 * Create a metadata_update record to update the
5028 * disk_ord_tbl for the array
5030 mu
= malloc(sizeof(*mu
));
5032 mu
->buf
= malloc(sizeof(struct imsm_update_activate_spare
) * num_spares
);
5033 if (mu
->buf
== NULL
) {
5040 struct mdinfo
*n
= rv
->next
;
5049 mu
->len
= sizeof(struct imsm_update_activate_spare
) * num_spares
;
5050 mu
->next
= *updates
;
5051 u
= (struct imsm_update_activate_spare
*) mu
->buf
;
5053 for (di
= rv
; di
; di
= di
->next
) {
5054 u
->type
= update_activate_spare
;
5055 u
->dl
= (struct dl
*) di
->devs
;
5057 u
->slot
= di
->disk
.raid_disk
;
5068 static int disks_overlap(struct intel_super
*super
, int idx
, struct imsm_update_create_array
*u
)
5070 struct imsm_dev
*dev
= get_imsm_dev(super
, idx
);
5071 struct imsm_map
*map
= get_imsm_map(dev
, 0);
5072 struct imsm_map
*new_map
= get_imsm_map(&u
->dev
, 0);
5073 struct disk_info
*inf
= get_disk_info(u
);
5074 struct imsm_disk
*disk
;
5078 for (i
= 0; i
< map
->num_members
; i
++) {
5079 disk
= get_imsm_disk(super
, get_imsm_disk_idx(dev
, i
));
5080 for (j
= 0; j
< new_map
->num_members
; j
++)
5081 if (serialcmp(disk
->serial
, inf
[j
].serial
) == 0)
5088 static void imsm_delete(struct intel_super
*super
, struct dl
**dlp
, unsigned index
);
5090 static void imsm_process_update(struct supertype
*st
,
5091 struct metadata_update
*update
)
5094 * crack open the metadata_update envelope to find the update record
5095 * update can be one of:
5096 * update_activate_spare - a spare device has replaced a failed
5097 * device in an array, update the disk_ord_tbl. If this disk is
5098 * present in all member arrays then also clear the SPARE_DISK
5101 struct intel_super
*super
= st
->sb
;
5102 struct imsm_super
*mpb
;
5103 enum imsm_update_type type
= *(enum imsm_update_type
*) update
->buf
;
5105 /* update requires a larger buf but the allocation failed */
5106 if (super
->next_len
&& !super
->next_buf
) {
5107 super
->next_len
= 0;
5111 if (super
->next_buf
) {
5112 memcpy(super
->next_buf
, super
->buf
, super
->len
);
5114 super
->len
= super
->next_len
;
5115 super
->buf
= super
->next_buf
;
5117 super
->next_len
= 0;
5118 super
->next_buf
= NULL
;
5121 mpb
= super
->anchor
;
5124 case update_activate_spare
: {
5125 struct imsm_update_activate_spare
*u
= (void *) update
->buf
;
5126 struct imsm_dev
*dev
= get_imsm_dev(super
, u
->array
);
5127 struct imsm_map
*map
= get_imsm_map(dev
, 0);
5128 struct imsm_map
*migr_map
;
5129 struct active_array
*a
;
5130 struct imsm_disk
*disk
;
5135 int victim
= get_imsm_disk_idx(dev
, u
->slot
);
5138 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
5143 fprintf(stderr
, "error: imsm_activate_spare passed "
5144 "an unknown disk (index: %d)\n",
5149 super
->updates_pending
++;
5151 /* count failures (excluding rebuilds and the victim)
5152 * to determine map[0] state
5155 for (i
= 0; i
< map
->num_members
; i
++) {
5158 disk
= get_imsm_disk(super
, get_imsm_disk_idx(dev
, i
));
5159 if (!disk
|| is_failed(disk
))
5163 /* adding a pristine spare, assign a new index */
5164 if (dl
->index
< 0) {
5165 dl
->index
= super
->anchor
->num_disks
;
5166 super
->anchor
->num_disks
++;
5169 disk
->status
|= CONFIGURED_DISK
;
5170 disk
->status
&= ~SPARE_DISK
;
5173 to_state
= imsm_check_degraded(super
, dev
, failed
);
5174 map
->map_state
= IMSM_T_STATE_DEGRADED
;
5175 migrate(dev
, to_state
, MIGR_REBUILD
);
5176 migr_map
= get_imsm_map(dev
, 1);
5177 set_imsm_ord_tbl_ent(map
, u
->slot
, dl
->index
);
5178 set_imsm_ord_tbl_ent(migr_map
, u
->slot
, dl
->index
| IMSM_ORD_REBUILD
);
5180 /* update the family_num to mark a new container
5181 * generation, being careful to record the existing
5182 * family_num in orig_family_num to clean up after
5183 * earlier mdadm versions that neglected to set it.
5185 if (mpb
->orig_family_num
== 0)
5186 mpb
->orig_family_num
= mpb
->family_num
;
5187 mpb
->family_num
+= super
->random
;
5189 /* count arrays using the victim in the metadata */
5191 for (a
= st
->arrays
; a
; a
= a
->next
) {
5192 dev
= get_imsm_dev(super
, a
->info
.container_member
);
5193 map
= get_imsm_map(dev
, 0);
5195 if (get_imsm_disk_slot(map
, victim
) >= 0)
5199 /* delete the victim if it is no longer being
5205 /* We know that 'manager' isn't touching anything,
5206 * so it is safe to delete
5208 for (dlp
= &super
->disks
; *dlp
; dlp
= &(*dlp
)->next
)
5209 if ((*dlp
)->index
== victim
)
5212 /* victim may be on the missing list */
5214 for (dlp
= &super
->missing
; *dlp
; dlp
= &(*dlp
)->next
)
5215 if ((*dlp
)->index
== victim
)
5217 imsm_delete(super
, dlp
, victim
);
5221 case update_create_array
: {
5222 /* someone wants to create a new array, we need to be aware of
5223 * a few races/collisions:
5224 * 1/ 'Create' called by two separate instances of mdadm
5225 * 2/ 'Create' versus 'activate_spare': mdadm has chosen
5226 * devices that have since been assimilated via
5228 * In the event this update can not be carried out mdadm will
5229 * (FIX ME) notice that its update did not take hold.
5231 struct imsm_update_create_array
*u
= (void *) update
->buf
;
5232 struct intel_dev
*dv
;
5233 struct imsm_dev
*dev
;
5234 struct imsm_map
*map
, *new_map
;
5235 unsigned long long start
, end
;
5236 unsigned long long new_start
, new_end
;
5238 struct disk_info
*inf
;
5241 /* handle racing creates: first come first serve */
5242 if (u
->dev_idx
< mpb
->num_raid_devs
) {
5243 dprintf("%s: subarray %d already defined\n",
5244 __func__
, u
->dev_idx
);
5248 /* check update is next in sequence */
5249 if (u
->dev_idx
!= mpb
->num_raid_devs
) {
5250 dprintf("%s: can not create array %d expected index %d\n",
5251 __func__
, u
->dev_idx
, mpb
->num_raid_devs
);
5255 new_map
= get_imsm_map(&u
->dev
, 0);
5256 new_start
= __le32_to_cpu(new_map
->pba_of_lba0
);
5257 new_end
= new_start
+ __le32_to_cpu(new_map
->blocks_per_member
);
5258 inf
= get_disk_info(u
);
5260 /* handle activate_spare versus create race:
5261 * check to make sure that overlapping arrays do not include
5264 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
5265 dev
= get_imsm_dev(super
, i
);
5266 map
= get_imsm_map(dev
, 0);
5267 start
= __le32_to_cpu(map
->pba_of_lba0
);
5268 end
= start
+ __le32_to_cpu(map
->blocks_per_member
);
5269 if ((new_start
>= start
&& new_start
<= end
) ||
5270 (start
>= new_start
&& start
<= new_end
))
5275 if (disks_overlap(super
, i
, u
)) {
5276 dprintf("%s: arrays overlap\n", __func__
);
5281 /* check that prepare update was successful */
5282 if (!update
->space
) {
5283 dprintf("%s: prepare update failed\n", __func__
);
5287 /* check that all disks are still active before committing
5288 * changes. FIXME: could we instead handle this by creating a
5289 * degraded array? That's probably not what the user expects,
5290 * so better to drop this update on the floor.
5292 for (i
= 0; i
< new_map
->num_members
; i
++) {
5293 dl
= serial_to_dl(inf
[i
].serial
, super
);
5295 dprintf("%s: disk disappeared\n", __func__
);
5300 super
->updates_pending
++;
5302 /* convert spares to members and fixup ord_tbl */
5303 for (i
= 0; i
< new_map
->num_members
; i
++) {
5304 dl
= serial_to_dl(inf
[i
].serial
, super
);
5305 if (dl
->index
== -1) {
5306 dl
->index
= mpb
->num_disks
;
5308 dl
->disk
.status
|= CONFIGURED_DISK
;
5309 dl
->disk
.status
&= ~SPARE_DISK
;
5311 set_imsm_ord_tbl_ent(new_map
, i
, dl
->index
);
5316 update
->space
= NULL
;
5317 imsm_copy_dev(dev
, &u
->dev
);
5318 dv
->index
= u
->dev_idx
;
5319 dv
->next
= super
->devlist
;
5320 super
->devlist
= dv
;
5321 mpb
->num_raid_devs
++;
5323 imsm_update_version_info(super
);
5326 /* mdmon knows how to release update->space, but not
5327 * ((struct intel_dev *) update->space)->dev
5329 if (update
->space
) {
5335 case update_kill_array
: {
5336 struct imsm_update_kill_array
*u
= (void *) update
->buf
;
5337 int victim
= u
->dev_idx
;
5338 struct active_array
*a
;
5339 struct intel_dev
**dp
;
5340 struct imsm_dev
*dev
;
5342 /* sanity check that we are not affecting the uuid of
5343 * active arrays, or deleting an active array
5345 * FIXME when immutable ids are available, but note that
5346 * we'll also need to fixup the invalidated/active
5347 * subarray indexes in mdstat
5349 for (a
= st
->arrays
; a
; a
= a
->next
)
5350 if (a
->info
.container_member
>= victim
)
5352 /* by definition if mdmon is running at least one array
5353 * is active in the container, so checking
5354 * mpb->num_raid_devs is just extra paranoia
5356 dev
= get_imsm_dev(super
, victim
);
5357 if (a
|| !dev
|| mpb
->num_raid_devs
== 1) {
5358 dprintf("failed to delete subarray-%d\n", victim
);
5362 for (dp
= &super
->devlist
; *dp
;)
5363 if ((*dp
)->index
== (unsigned)super
->current_vol
) {
5366 if ((*dp
)->index
> (unsigned)victim
)
5370 mpb
->num_raid_devs
--;
5371 super
->updates_pending
++;
5374 case update_rename_array
: {
5375 struct imsm_update_rename_array
*u
= (void *) update
->buf
;
5376 char name
[MAX_RAID_SERIAL_LEN
+1];
5377 int target
= u
->dev_idx
;
5378 struct active_array
*a
;
5379 struct imsm_dev
*dev
;
5381 /* sanity check that we are not affecting the uuid of
5384 snprintf(name
, MAX_RAID_SERIAL_LEN
, "%s", (char *) u
->name
);
5385 name
[MAX_RAID_SERIAL_LEN
] = '\0';
5386 for (a
= st
->arrays
; a
; a
= a
->next
)
5387 if (a
->info
.container_member
== target
)
5389 dev
= get_imsm_dev(super
, u
->dev_idx
);
5390 if (a
|| !dev
|| !check_name(super
, name
, 1)) {
5391 dprintf("failed to rename subarray-%d\n", target
);
5395 snprintf((char *) dev
->volume
, MAX_RAID_SERIAL_LEN
, "%s", name
);
5396 super
->updates_pending
++;
5399 case update_add_disk
:
5401 /* we may be able to repair some arrays if disks are
5404 struct active_array
*a
;
5406 super
->updates_pending
++;
5407 for (a
= st
->arrays
; a
; a
= a
->next
)
5408 a
->check_degraded
= 1;
5410 /* add some spares to the metadata */
5411 while (super
->add
) {
5415 super
->add
= al
->next
;
5416 al
->next
= super
->disks
;
5418 dprintf("%s: added %x:%x\n",
5419 __func__
, al
->major
, al
->minor
);
5426 static void imsm_prepare_update(struct supertype
*st
,
5427 struct metadata_update
*update
)
5430 * Allocate space to hold new disk entries, raid-device entries or a new
5431 * mpb if necessary. The manager synchronously waits for updates to
5432 * complete in the monitor, so new mpb buffers allocated here can be
5433 * integrated by the monitor thread without worrying about live pointers
5434 * in the manager thread.
5436 enum imsm_update_type type
= *(enum imsm_update_type
*) update
->buf
;
5437 struct intel_super
*super
= st
->sb
;
5438 struct imsm_super
*mpb
= super
->anchor
;
5443 case update_create_array
: {
5444 struct imsm_update_create_array
*u
= (void *) update
->buf
;
5445 struct intel_dev
*dv
;
5446 struct imsm_dev
*dev
= &u
->dev
;
5447 struct imsm_map
*map
= get_imsm_map(dev
, 0);
5449 struct disk_info
*inf
;
5453 inf
= get_disk_info(u
);
5454 len
= sizeof_imsm_dev(dev
, 1);
5455 /* allocate a new super->devlist entry */
5456 dv
= malloc(sizeof(*dv
));
5458 dv
->dev
= malloc(len
);
5463 update
->space
= NULL
;
5467 /* count how many spares will be converted to members */
5468 for (i
= 0; i
< map
->num_members
; i
++) {
5469 dl
= serial_to_dl(inf
[i
].serial
, super
);
5471 /* hmm maybe it failed?, nothing we can do about
5476 if (count_memberships(dl
, super
) == 0)
5479 len
+= activate
* sizeof(struct imsm_disk
);
5486 /* check if we need a larger metadata buffer */
5487 if (super
->next_buf
)
5488 buf_len
= super
->next_len
;
5490 buf_len
= super
->len
;
5492 if (__le32_to_cpu(mpb
->mpb_size
) + len
> buf_len
) {
5493 /* ok we need a larger buf than what is currently allocated
5494 * if this allocation fails process_update will notice that
5495 * ->next_len is set and ->next_buf is NULL
5497 buf_len
= ROUND_UP(__le32_to_cpu(mpb
->mpb_size
) + len
, 512);
5498 if (super
->next_buf
)
5499 free(super
->next_buf
);
5501 super
->next_len
= buf_len
;
5502 if (posix_memalign(&super
->next_buf
, 512, buf_len
) == 0)
5503 memset(super
->next_buf
, 0, buf_len
);
5505 super
->next_buf
= NULL
;
5509 /* must be called while manager is quiesced */
5510 static void imsm_delete(struct intel_super
*super
, struct dl
**dlp
, unsigned index
)
5512 struct imsm_super
*mpb
= super
->anchor
;
5514 struct imsm_dev
*dev
;
5515 struct imsm_map
*map
;
5516 int i
, j
, num_members
;
5519 dprintf("%s: deleting device[%d] from imsm_super\n",
5522 /* shift all indexes down one */
5523 for (iter
= super
->disks
; iter
; iter
= iter
->next
)
5524 if (iter
->index
> (int)index
)
5526 for (iter
= super
->missing
; iter
; iter
= iter
->next
)
5527 if (iter
->index
> (int)index
)
5530 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
5531 dev
= get_imsm_dev(super
, i
);
5532 map
= get_imsm_map(dev
, 0);
5533 num_members
= map
->num_members
;
5534 for (j
= 0; j
< num_members
; j
++) {
5535 /* update ord entries being careful not to propagate
5536 * ord-flags to the first map
5538 ord
= get_imsm_ord_tbl_ent(dev
, j
);
5540 if (ord_to_idx(ord
) <= index
)
5543 map
= get_imsm_map(dev
, 0);
5544 set_imsm_ord_tbl_ent(map
, j
, ord_to_idx(ord
- 1));
5545 map
= get_imsm_map(dev
, 1);
5547 set_imsm_ord_tbl_ent(map
, j
, ord
- 1);
5552 super
->updates_pending
++;
5554 struct dl
*dl
= *dlp
;
5556 *dlp
= (*dlp
)->next
;
5557 __free_imsm_disk(dl
);
5560 #endif /* MDASSEMBLE */
5562 struct superswitch super_imsm
= {
5564 .examine_super
= examine_super_imsm
,
5565 .brief_examine_super
= brief_examine_super_imsm
,
5566 .brief_examine_subarrays
= brief_examine_subarrays_imsm
,
5567 .export_examine_super
= export_examine_super_imsm
,
5568 .detail_super
= detail_super_imsm
,
5569 .brief_detail_super
= brief_detail_super_imsm
,
5570 .write_init_super
= write_init_super_imsm
,
5571 .validate_geometry
= validate_geometry_imsm
,
5572 .default_chunk
= default_chunk_imsm
,
5573 .add_to_super
= add_to_super_imsm
,
5574 .detail_platform
= detail_platform_imsm
,
5575 .kill_subarray
= kill_subarray_imsm
,
5576 .update_subarray
= update_subarray_imsm
,
5577 .load_container
= load_container_imsm
,
5579 .match_home
= match_home_imsm
,
5580 .uuid_from_super
= uuid_from_super_imsm
,
5581 .getinfo_super
= getinfo_super_imsm
,
5582 .update_super
= update_super_imsm
,
5584 .avail_size
= avail_size_imsm
,
5586 .compare_super
= compare_super_imsm
,
5588 .load_super
= load_super_imsm
,
5589 .init_super
= init_super_imsm
,
5590 .store_super
= store_super_imsm
,
5591 .free_super
= free_super_imsm
,
5592 .match_metadata_desc
= match_metadata_desc_imsm
,
5593 .container_content
= container_content_imsm
,
5594 .default_layout
= imsm_level_to_layout
,
5601 .open_new
= imsm_open_new
,
5602 .set_array_state
= imsm_set_array_state
,
5603 .set_disk
= imsm_set_disk
,
5604 .sync_metadata
= imsm_sync_metadata
,
5605 .activate_spare
= imsm_activate_spare
,
5606 .process_update
= imsm_process_update
,
5607 .prepare_update
= imsm_prepare_update
,
5608 #endif /* MDASSEMBLE */