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
;
650 /* Return minimum size of a spare that can be used in this array*/
651 static unsigned long long min_acceptable_spare_size_imsm(struct supertype
*st
)
653 struct intel_super
*super
= st
->sb
;
657 unsigned long long rv
= 0;
661 /* find first active disk in array */
663 while (dl
&& (is_failed(&dl
->disk
) || dl
->index
== -1))
667 /* find last lba used by subarrays */
668 e
= get_extents(super
, dl
);
671 for (i
= 0; e
[i
].size
; i
++)
674 rv
= e
[i
-1].start
+ e
[i
-1].size
;
676 /* add the amount of space needed for metadata */
677 rv
= rv
+ MPB_SECTOR_CNT
+ IMSM_RESERVED_SECTORS
;
682 static __u64
blocks_per_migr_unit(struct imsm_dev
*dev
);
684 static void print_imsm_dev(struct imsm_dev
*dev
, char *uuid
, int disk_idx
)
688 struct imsm_map
*map
= get_imsm_map(dev
, 0);
692 printf("[%.16s]:\n", dev
->volume
);
693 printf(" UUID : %s\n", uuid
);
694 printf(" RAID Level : %d\n", get_imsm_raid_level(map
));
695 printf(" Members : %d\n", map
->num_members
);
696 printf(" Slots : [");
697 for (i
= 0; i
< map
->num_members
; i
++) {
698 ord
= get_imsm_ord_tbl_ent(dev
, i
);
699 printf("%s", ord
& IMSM_ORD_REBUILD
? "_" : "U");
702 slot
= get_imsm_disk_slot(map
, disk_idx
);
704 ord
= get_imsm_ord_tbl_ent(dev
, slot
);
705 printf(" This Slot : %d%s\n", slot
,
706 ord
& IMSM_ORD_REBUILD
? " (out-of-sync)" : "");
708 printf(" This Slot : ?\n");
709 sz
= __le32_to_cpu(dev
->size_high
);
711 sz
+= __le32_to_cpu(dev
->size_low
);
712 printf(" Array Size : %llu%s\n", (unsigned long long)sz
,
713 human_size(sz
* 512));
714 sz
= __le32_to_cpu(map
->blocks_per_member
);
715 printf(" Per Dev Size : %llu%s\n", (unsigned long long)sz
,
716 human_size(sz
* 512));
717 printf(" Sector Offset : %u\n",
718 __le32_to_cpu(map
->pba_of_lba0
));
719 printf(" Num Stripes : %u\n",
720 __le32_to_cpu(map
->num_data_stripes
));
721 printf(" Chunk Size : %u KiB\n",
722 __le16_to_cpu(map
->blocks_per_strip
) / 2);
723 printf(" Reserved : %d\n", __le32_to_cpu(dev
->reserved_blocks
));
724 printf(" Migrate State : ");
725 if (dev
->vol
.migr_state
) {
726 if (migr_type(dev
) == MIGR_INIT
)
727 printf("initialize\n");
728 else if (migr_type(dev
) == MIGR_REBUILD
)
730 else if (migr_type(dev
) == MIGR_VERIFY
)
732 else if (migr_type(dev
) == MIGR_GEN_MIGR
)
733 printf("general migration\n");
734 else if (migr_type(dev
) == MIGR_STATE_CHANGE
)
735 printf("state change\n");
736 else if (migr_type(dev
) == MIGR_REPAIR
)
739 printf("<unknown:%d>\n", migr_type(dev
));
742 printf(" Map State : %s", map_state_str
[map
->map_state
]);
743 if (dev
->vol
.migr_state
) {
744 struct imsm_map
*map
= get_imsm_map(dev
, 1);
746 printf(" <-- %s", map_state_str
[map
->map_state
]);
747 printf("\n Checkpoint : %u (%llu)",
748 __le32_to_cpu(dev
->vol
.curr_migr_unit
),
749 (unsigned long long)blocks_per_migr_unit(dev
));
752 printf(" Dirty State : %s\n", dev
->vol
.dirty
? "dirty" : "clean");
755 static void print_imsm_disk(struct imsm_super
*mpb
, int index
, __u32 reserved
)
757 struct imsm_disk
*disk
= __get_imsm_disk(mpb
, index
);
758 char str
[MAX_RAID_SERIAL_LEN
+ 1];
761 if (index
< 0 || !disk
)
765 snprintf(str
, MAX_RAID_SERIAL_LEN
+ 1, "%s", disk
->serial
);
766 printf(" Disk%02d Serial : %s\n", index
, str
);
767 printf(" State :%s%s%s\n", is_spare(disk
) ? " spare" : "",
768 is_configured(disk
) ? " active" : "",
769 is_failed(disk
) ? " failed" : "");
770 printf(" Id : %08x\n", __le32_to_cpu(disk
->scsi_id
));
771 sz
= __le32_to_cpu(disk
->total_blocks
) - reserved
;
772 printf(" Usable Size : %llu%s\n", (unsigned long long)sz
,
773 human_size(sz
* 512));
776 static void getinfo_super_imsm(struct supertype
*st
, struct mdinfo
*info
, char *map
);
778 static void examine_super_imsm(struct supertype
*st
, char *homehost
)
780 struct intel_super
*super
= st
->sb
;
781 struct imsm_super
*mpb
= super
->anchor
;
782 char str
[MAX_SIGNATURE_LENGTH
];
787 __u32 reserved
= imsm_reserved_sectors(super
, super
->disks
);
790 snprintf(str
, MPB_SIG_LEN
, "%s", mpb
->sig
);
791 printf(" Magic : %s\n", str
);
792 snprintf(str
, strlen(MPB_VERSION_RAID0
), "%s", get_imsm_version(mpb
));
793 printf(" Version : %s\n", get_imsm_version(mpb
));
794 printf(" Orig Family : %08x\n", __le32_to_cpu(mpb
->orig_family_num
));
795 printf(" Family : %08x\n", __le32_to_cpu(mpb
->family_num
));
796 printf(" Generation : %08x\n", __le32_to_cpu(mpb
->generation_num
));
797 getinfo_super_imsm(st
, &info
, NULL
);
798 fname_from_uuid(st
, &info
, nbuf
, ':');
799 printf(" UUID : %s\n", nbuf
+ 5);
800 sum
= __le32_to_cpu(mpb
->check_sum
);
801 printf(" Checksum : %08x %s\n", sum
,
802 __gen_imsm_checksum(mpb
) == sum
? "correct" : "incorrect");
803 printf(" MPB Sectors : %d\n", mpb_sectors(mpb
));
804 printf(" Disks : %d\n", mpb
->num_disks
);
805 printf(" RAID Devices : %d\n", mpb
->num_raid_devs
);
806 print_imsm_disk(mpb
, super
->disks
->index
, reserved
);
807 if (super
->bbm_log
) {
808 struct bbm_log
*log
= super
->bbm_log
;
811 printf("Bad Block Management Log:\n");
812 printf(" Log Size : %d\n", __le32_to_cpu(mpb
->bbm_log_size
));
813 printf(" Signature : %x\n", __le32_to_cpu(log
->signature
));
814 printf(" Entry Count : %d\n", __le32_to_cpu(log
->entry_count
));
815 printf(" Spare Blocks : %d\n", __le32_to_cpu(log
->reserved_spare_block_count
));
816 printf(" First Spare : %llx\n",
817 (unsigned long long) __le64_to_cpu(log
->first_spare_lba
));
819 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
821 struct imsm_dev
*dev
= __get_imsm_dev(mpb
, i
);
823 super
->current_vol
= i
;
824 getinfo_super_imsm(st
, &info
, NULL
);
825 fname_from_uuid(st
, &info
, nbuf
, ':');
826 print_imsm_dev(dev
, nbuf
+ 5, super
->disks
->index
);
828 for (i
= 0; i
< mpb
->num_disks
; i
++) {
829 if (i
== super
->disks
->index
)
831 print_imsm_disk(mpb
, i
, reserved
);
835 static void brief_examine_super_imsm(struct supertype
*st
, int verbose
)
837 /* We just write a generic IMSM ARRAY entry */
840 struct intel_super
*super
= st
->sb
;
842 if (!super
->anchor
->num_raid_devs
) {
843 printf("ARRAY metadata=imsm\n");
847 getinfo_super_imsm(st
, &info
, NULL
);
848 fname_from_uuid(st
, &info
, nbuf
, ':');
849 printf("ARRAY metadata=imsm UUID=%s\n", nbuf
+ 5);
852 static void brief_examine_subarrays_imsm(struct supertype
*st
, int verbose
)
854 /* We just write a generic IMSM ARRAY entry */
858 struct intel_super
*super
= st
->sb
;
861 if (!super
->anchor
->num_raid_devs
)
864 getinfo_super_imsm(st
, &info
, NULL
);
865 fname_from_uuid(st
, &info
, nbuf
, ':');
866 for (i
= 0; i
< super
->anchor
->num_raid_devs
; i
++) {
867 struct imsm_dev
*dev
= get_imsm_dev(super
, i
);
869 super
->current_vol
= i
;
870 getinfo_super_imsm(st
, &info
, NULL
);
871 fname_from_uuid(st
, &info
, nbuf1
, ':');
872 printf("ARRAY /dev/md/%.16s container=%s member=%d UUID=%s\n",
873 dev
->volume
, nbuf
+ 5, i
, nbuf1
+ 5);
877 static void export_examine_super_imsm(struct supertype
*st
)
879 struct intel_super
*super
= st
->sb
;
880 struct imsm_super
*mpb
= super
->anchor
;
884 getinfo_super_imsm(st
, &info
, NULL
);
885 fname_from_uuid(st
, &info
, nbuf
, ':');
886 printf("MD_METADATA=imsm\n");
887 printf("MD_LEVEL=container\n");
888 printf("MD_UUID=%s\n", nbuf
+5);
889 printf("MD_DEVICES=%u\n", mpb
->num_disks
);
892 static void detail_super_imsm(struct supertype
*st
, char *homehost
)
897 getinfo_super_imsm(st
, &info
, NULL
);
898 fname_from_uuid(st
, &info
, nbuf
, ':');
899 printf("\n UUID : %s\n", nbuf
+ 5);
902 static void brief_detail_super_imsm(struct supertype
*st
)
906 getinfo_super_imsm(st
, &info
, NULL
);
907 fname_from_uuid(st
, &info
, nbuf
, ':');
908 printf(" UUID=%s", nbuf
+ 5);
911 static int imsm_read_serial(int fd
, char *devname
, __u8
*serial
);
912 static void fd2devname(int fd
, char *name
);
914 static int imsm_enumerate_ports(const char *hba_path
, int port_count
, int host_base
, int verbose
)
916 /* dump an unsorted list of devices attached to ahci, as well as
917 * non-connected ports
919 int hba_len
= strlen(hba_path
) + 1;
924 unsigned long port_mask
= (1 << port_count
) - 1;
926 if (port_count
> (int)sizeof(port_mask
) * 8) {
928 fprintf(stderr
, Name
": port_count %d out of range\n", port_count
);
932 /* scroll through /sys/dev/block looking for devices attached to
935 dir
= opendir("/sys/dev/block");
936 for (ent
= dir
? readdir(dir
) : NULL
; ent
; ent
= readdir(dir
)) {
947 if (sscanf(ent
->d_name
, "%d:%d", &major
, &minor
) != 2)
949 path
= devt_to_devpath(makedev(major
, minor
));
952 if (!path_attached_to_hba(path
, hba_path
)) {
958 /* retrieve the scsi device type */
959 if (asprintf(&device
, "/sys/dev/block/%d:%d/device/xxxxxxx", major
, minor
) < 0) {
961 fprintf(stderr
, Name
": failed to allocate 'device'\n");
965 sprintf(device
, "/sys/dev/block/%d:%d/device/type", major
, minor
);
966 if (load_sys(device
, buf
) != 0) {
968 fprintf(stderr
, Name
": failed to read device type for %s\n",
974 type
= strtoul(buf
, NULL
, 10);
976 /* if it's not a disk print the vendor and model */
977 if (!(type
== 0 || type
== 7 || type
== 14)) {
980 sprintf(device
, "/sys/dev/block/%d:%d/device/vendor", major
, minor
);
981 if (load_sys(device
, buf
) == 0) {
982 strncpy(vendor
, buf
, sizeof(vendor
));
983 vendor
[sizeof(vendor
) - 1] = '\0';
984 c
= (char *) &vendor
[sizeof(vendor
) - 1];
985 while (isspace(*c
) || *c
== '\0')
989 sprintf(device
, "/sys/dev/block/%d:%d/device/model", major
, minor
);
990 if (load_sys(device
, buf
) == 0) {
991 strncpy(model
, buf
, sizeof(model
));
992 model
[sizeof(model
) - 1] = '\0';
993 c
= (char *) &model
[sizeof(model
) - 1];
994 while (isspace(*c
) || *c
== '\0')
998 if (vendor
[0] && model
[0])
999 sprintf(buf
, "%.64s %.64s", vendor
, model
);
1001 switch (type
) { /* numbers from hald/linux/device.c */
1002 case 1: sprintf(buf
, "tape"); break;
1003 case 2: sprintf(buf
, "printer"); break;
1004 case 3: sprintf(buf
, "processor"); break;
1006 case 5: sprintf(buf
, "cdrom"); break;
1007 case 6: sprintf(buf
, "scanner"); break;
1008 case 8: sprintf(buf
, "media_changer"); break;
1009 case 9: sprintf(buf
, "comm"); break;
1010 case 12: sprintf(buf
, "raid"); break;
1011 default: sprintf(buf
, "unknown");
1017 /* chop device path to 'host%d' and calculate the port number */
1018 c
= strchr(&path
[hba_len
], '/');
1021 fprintf(stderr
, Name
": %s - invalid path name\n", path
+ hba_len
);
1026 if (sscanf(&path
[hba_len
], "host%d", &port
) == 1)
1030 *c
= '/'; /* repair the full string */
1031 fprintf(stderr
, Name
": failed to determine port number for %s\n",
1038 /* mark this port as used */
1039 port_mask
&= ~(1 << port
);
1041 /* print out the device information */
1043 printf(" Port%d : - non-disk device (%s) -\n", port
, buf
);
1047 fd
= dev_open(ent
->d_name
, O_RDONLY
);
1049 printf(" Port%d : - disk info unavailable -\n", port
);
1051 fd2devname(fd
, buf
);
1052 printf(" Port%d : %s", port
, buf
);
1053 if (imsm_read_serial(fd
, NULL
, (__u8
*) buf
) == 0)
1054 printf(" (%s)\n", buf
);
1069 for (i
= 0; i
< port_count
; i
++)
1070 if (port_mask
& (1 << i
))
1071 printf(" Port%d : - no device attached -\n", i
);
1077 static int detail_platform_imsm(int verbose
, int enumerate_only
)
1079 /* There are two components to imsm platform support, the ahci SATA
1080 * controller and the option-rom. To find the SATA controller we
1081 * simply look in /sys/bus/pci/drivers/ahci to see if an ahci
1082 * controller with the Intel vendor id is present. This approach
1083 * allows mdadm to leverage the kernel's ahci detection logic, with the
1084 * caveat that if ahci.ko is not loaded mdadm will not be able to
1085 * detect platform raid capabilities. The option-rom resides in a
1086 * platform "Adapter ROM". We scan for its signature to retrieve the
1087 * platform capabilities. If raid support is disabled in the BIOS the
1088 * option-rom capability structure will not be available.
1090 const struct imsm_orom
*orom
;
1091 struct sys_dev
*list
, *hba
;
1094 const char *hba_path
;
1098 if (enumerate_only
) {
1099 if (check_env("IMSM_NO_PLATFORM") || find_imsm_orom())
1104 list
= find_driver_devices("pci", "ahci");
1105 for (hba
= list
; hba
; hba
= hba
->next
)
1106 if (devpath_to_vendor(hba
->path
) == 0x8086)
1111 fprintf(stderr
, Name
": unable to find active ahci controller\n");
1112 free_sys_dev(&list
);
1115 fprintf(stderr
, Name
": found Intel SATA AHCI Controller\n");
1116 hba_path
= hba
->path
;
1118 free_sys_dev(&list
);
1120 orom
= find_imsm_orom();
1123 fprintf(stderr
, Name
": imsm option-rom not found\n");
1127 printf(" Platform : Intel(R) Matrix Storage Manager\n");
1128 printf(" Version : %d.%d.%d.%d\n", orom
->major_ver
, orom
->minor_ver
,
1129 orom
->hotfix_ver
, orom
->build
);
1130 printf(" RAID Levels :%s%s%s%s%s\n",
1131 imsm_orom_has_raid0(orom
) ? " raid0" : "",
1132 imsm_orom_has_raid1(orom
) ? " raid1" : "",
1133 imsm_orom_has_raid1e(orom
) ? " raid1e" : "",
1134 imsm_orom_has_raid10(orom
) ? " raid10" : "",
1135 imsm_orom_has_raid5(orom
) ? " raid5" : "");
1136 printf(" Chunk Sizes :%s%s%s%s%s%s%s%s%s%s%s%s%s%s%s%s\n",
1137 imsm_orom_has_chunk(orom
, 2) ? " 2k" : "",
1138 imsm_orom_has_chunk(orom
, 4) ? " 4k" : "",
1139 imsm_orom_has_chunk(orom
, 8) ? " 8k" : "",
1140 imsm_orom_has_chunk(orom
, 16) ? " 16k" : "",
1141 imsm_orom_has_chunk(orom
, 32) ? " 32k" : "",
1142 imsm_orom_has_chunk(orom
, 64) ? " 64k" : "",
1143 imsm_orom_has_chunk(orom
, 128) ? " 128k" : "",
1144 imsm_orom_has_chunk(orom
, 256) ? " 256k" : "",
1145 imsm_orom_has_chunk(orom
, 512) ? " 512k" : "",
1146 imsm_orom_has_chunk(orom
, 1024*1) ? " 1M" : "",
1147 imsm_orom_has_chunk(orom
, 1024*2) ? " 2M" : "",
1148 imsm_orom_has_chunk(orom
, 1024*4) ? " 4M" : "",
1149 imsm_orom_has_chunk(orom
, 1024*8) ? " 8M" : "",
1150 imsm_orom_has_chunk(orom
, 1024*16) ? " 16M" : "",
1151 imsm_orom_has_chunk(orom
, 1024*32) ? " 32M" : "",
1152 imsm_orom_has_chunk(orom
, 1024*64) ? " 64M" : "");
1153 printf(" Max Disks : %d\n", orom
->tds
);
1154 printf(" Max Volumes : %d\n", orom
->vpa
);
1155 printf(" I/O Controller : %s\n", hba_path
);
1157 /* find the smallest scsi host number to determine a port number base */
1158 dir
= opendir(hba_path
);
1159 for (ent
= dir
? readdir(dir
) : NULL
; ent
; ent
= readdir(dir
)) {
1162 if (sscanf(ent
->d_name
, "host%d", &host
) != 1)
1164 if (port_count
== 0)
1166 else if (host
< host_base
)
1169 if (host
+ 1 > port_count
+ host_base
)
1170 port_count
= host
+ 1 - host_base
;
1176 if (!port_count
|| imsm_enumerate_ports(hba_path
, port_count
,
1177 host_base
, verbose
) != 0) {
1179 fprintf(stderr
, Name
": failed to enumerate ports\n");
1187 static int match_home_imsm(struct supertype
*st
, char *homehost
)
1189 /* the imsm metadata format does not specify any host
1190 * identification information. We return -1 since we can never
1191 * confirm nor deny whether a given array is "meant" for this
1192 * host. We rely on compare_super and the 'family_num' fields to
1193 * exclude member disks that do not belong, and we rely on
1194 * mdadm.conf to specify the arrays that should be assembled.
1195 * Auto-assembly may still pick up "foreign" arrays.
1201 static void uuid_from_super_imsm(struct supertype
*st
, int uuid
[4])
1203 /* The uuid returned here is used for:
1204 * uuid to put into bitmap file (Create, Grow)
1205 * uuid for backup header when saving critical section (Grow)
1206 * comparing uuids when re-adding a device into an array
1207 * In these cases the uuid required is that of the data-array,
1208 * not the device-set.
1209 * uuid to recognise same set when adding a missing device back
1210 * to an array. This is a uuid for the device-set.
1212 * For each of these we can make do with a truncated
1213 * or hashed uuid rather than the original, as long as
1215 * In each case the uuid required is that of the data-array,
1216 * not the device-set.
1218 /* imsm does not track uuid's so we synthesis one using sha1 on
1219 * - The signature (Which is constant for all imsm array, but no matter)
1220 * - the orig_family_num of the container
1221 * - the index number of the volume
1222 * - the 'serial' number of the volume.
1223 * Hopefully these are all constant.
1225 struct intel_super
*super
= st
->sb
;
1228 struct sha1_ctx ctx
;
1229 struct imsm_dev
*dev
= NULL
;
1232 /* some mdadm versions failed to set ->orig_family_num, in which
1233 * case fall back to ->family_num. orig_family_num will be
1234 * fixed up with the first metadata update.
1236 family_num
= super
->anchor
->orig_family_num
;
1237 if (family_num
== 0)
1238 family_num
= super
->anchor
->family_num
;
1239 sha1_init_ctx(&ctx
);
1240 sha1_process_bytes(super
->anchor
->sig
, MPB_SIG_LEN
, &ctx
);
1241 sha1_process_bytes(&family_num
, sizeof(__u32
), &ctx
);
1242 if (super
->current_vol
>= 0)
1243 dev
= get_imsm_dev(super
, super
->current_vol
);
1245 __u32 vol
= super
->current_vol
;
1246 sha1_process_bytes(&vol
, sizeof(vol
), &ctx
);
1247 sha1_process_bytes(dev
->volume
, MAX_RAID_SERIAL_LEN
, &ctx
);
1249 sha1_finish_ctx(&ctx
, buf
);
1250 memcpy(uuid
, buf
, 4*4);
1255 get_imsm_numerical_version(struct imsm_super
*mpb
, int *m
, int *p
)
1257 __u8
*v
= get_imsm_version(mpb
);
1258 __u8
*end
= mpb
->sig
+ MAX_SIGNATURE_LENGTH
;
1259 char major
[] = { 0, 0, 0 };
1260 char minor
[] = { 0 ,0, 0 };
1261 char patch
[] = { 0, 0, 0 };
1262 char *ver_parse
[] = { major
, minor
, patch
};
1266 while (*v
!= '\0' && v
< end
) {
1267 if (*v
!= '.' && j
< 2)
1268 ver_parse
[i
][j
++] = *v
;
1276 *m
= strtol(minor
, NULL
, 0);
1277 *p
= strtol(patch
, NULL
, 0);
1281 static __u32
migr_strip_blocks_resync(struct imsm_dev
*dev
)
1283 /* migr_strip_size when repairing or initializing parity */
1284 struct imsm_map
*map
= get_imsm_map(dev
, 0);
1285 __u32 chunk
= __le32_to_cpu(map
->blocks_per_strip
);
1287 switch (get_imsm_raid_level(map
)) {
1292 return 128*1024 >> 9;
1296 static __u32
migr_strip_blocks_rebuild(struct imsm_dev
*dev
)
1298 /* migr_strip_size when rebuilding a degraded disk, no idea why
1299 * this is different than migr_strip_size_resync(), but it's good
1302 struct imsm_map
*map
= get_imsm_map(dev
, 1);
1303 __u32 chunk
= __le32_to_cpu(map
->blocks_per_strip
);
1305 switch (get_imsm_raid_level(map
)) {
1308 if (map
->num_members
% map
->num_domains
== 0)
1309 return 128*1024 >> 9;
1313 return max((__u32
) 64*1024 >> 9, chunk
);
1315 return 128*1024 >> 9;
1319 static __u32
num_stripes_per_unit_resync(struct imsm_dev
*dev
)
1321 struct imsm_map
*lo
= get_imsm_map(dev
, 0);
1322 struct imsm_map
*hi
= get_imsm_map(dev
, 1);
1323 __u32 lo_chunk
= __le32_to_cpu(lo
->blocks_per_strip
);
1324 __u32 hi_chunk
= __le32_to_cpu(hi
->blocks_per_strip
);
1326 return max((__u32
) 1, hi_chunk
/ lo_chunk
);
1329 static __u32
num_stripes_per_unit_rebuild(struct imsm_dev
*dev
)
1331 struct imsm_map
*lo
= get_imsm_map(dev
, 0);
1332 int level
= get_imsm_raid_level(lo
);
1334 if (level
== 1 || level
== 10) {
1335 struct imsm_map
*hi
= get_imsm_map(dev
, 1);
1337 return hi
->num_domains
;
1339 return num_stripes_per_unit_resync(dev
);
1342 static __u8
imsm_num_data_members(struct imsm_dev
*dev
)
1344 /* named 'imsm_' because raid0, raid1 and raid10
1345 * counter-intuitively have the same number of data disks
1347 struct imsm_map
*map
= get_imsm_map(dev
, 0);
1349 switch (get_imsm_raid_level(map
)) {
1353 return map
->num_members
;
1355 return map
->num_members
- 1;
1357 dprintf("%s: unsupported raid level\n", __func__
);
1362 static __u32
parity_segment_depth(struct imsm_dev
*dev
)
1364 struct imsm_map
*map
= get_imsm_map(dev
, 0);
1365 __u32 chunk
= __le32_to_cpu(map
->blocks_per_strip
);
1367 switch(get_imsm_raid_level(map
)) {
1370 return chunk
* map
->num_domains
;
1372 return chunk
* map
->num_members
;
1378 static __u32
map_migr_block(struct imsm_dev
*dev
, __u32 block
)
1380 struct imsm_map
*map
= get_imsm_map(dev
, 1);
1381 __u32 chunk
= __le32_to_cpu(map
->blocks_per_strip
);
1382 __u32 strip
= block
/ chunk
;
1384 switch (get_imsm_raid_level(map
)) {
1387 __u32 vol_strip
= (strip
* map
->num_domains
) + 1;
1388 __u32 vol_stripe
= vol_strip
/ map
->num_members
;
1390 return vol_stripe
* chunk
+ block
% chunk
;
1392 __u32 stripe
= strip
/ (map
->num_members
- 1);
1394 return stripe
* chunk
+ block
% chunk
;
1401 static __u64
blocks_per_migr_unit(struct imsm_dev
*dev
)
1403 /* calculate the conversion factor between per member 'blocks'
1404 * (md/{resync,rebuild}_start) and imsm migration units, return
1405 * 0 for the 'not migrating' and 'unsupported migration' cases
1407 if (!dev
->vol
.migr_state
)
1410 switch (migr_type(dev
)) {
1414 struct imsm_map
*map
= get_imsm_map(dev
, 0);
1415 __u32 stripes_per_unit
;
1416 __u32 blocks_per_unit
;
1425 /* yes, this is really the translation of migr_units to
1426 * per-member blocks in the 'resync' case
1428 stripes_per_unit
= num_stripes_per_unit_resync(dev
);
1429 migr_chunk
= migr_strip_blocks_resync(dev
);
1430 disks
= imsm_num_data_members(dev
);
1431 blocks_per_unit
= stripes_per_unit
* migr_chunk
* disks
;
1432 stripe
= __le32_to_cpu(map
->blocks_per_strip
) * disks
;
1433 segment
= blocks_per_unit
/ stripe
;
1434 block_rel
= blocks_per_unit
- segment
* stripe
;
1435 parity_depth
= parity_segment_depth(dev
);
1436 block_map
= map_migr_block(dev
, block_rel
);
1437 return block_map
+ parity_depth
* segment
;
1439 case MIGR_REBUILD
: {
1440 __u32 stripes_per_unit
;
1443 stripes_per_unit
= num_stripes_per_unit_rebuild(dev
);
1444 migr_chunk
= migr_strip_blocks_rebuild(dev
);
1445 return migr_chunk
* stripes_per_unit
;
1448 case MIGR_STATE_CHANGE
:
1454 static int imsm_level_to_layout(int level
)
1462 return ALGORITHM_LEFT_ASYMMETRIC
;
1469 static void getinfo_super_imsm_volume(struct supertype
*st
, struct mdinfo
*info
, char *dmap
)
1471 struct intel_super
*super
= st
->sb
;
1472 struct imsm_dev
*dev
= get_imsm_dev(super
, super
->current_vol
);
1473 struct imsm_map
*map
= get_imsm_map(dev
, 0);
1476 int map_disks
= info
->array
.raid_disks
;
1478 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
1479 if (dl
->raiddisk
== info
->disk
.raid_disk
)
1481 info
->container_member
= super
->current_vol
;
1482 info
->array
.raid_disks
= map
->num_members
;
1483 info
->array
.level
= get_imsm_raid_level(map
);
1484 info
->array
.layout
= imsm_level_to_layout(info
->array
.level
);
1485 info
->array
.md_minor
= -1;
1486 info
->array
.ctime
= 0;
1487 info
->array
.utime
= 0;
1488 info
->array
.chunk_size
= __le16_to_cpu(map
->blocks_per_strip
) << 9;
1489 info
->array
.state
= !dev
->vol
.dirty
;
1490 info
->custom_array_size
= __le32_to_cpu(dev
->size_high
);
1491 info
->custom_array_size
<<= 32;
1492 info
->custom_array_size
|= __le32_to_cpu(dev
->size_low
);
1494 info
->disk
.major
= 0;
1495 info
->disk
.minor
= 0;
1497 info
->disk
.major
= dl
->major
;
1498 info
->disk
.minor
= dl
->minor
;
1501 info
->data_offset
= __le32_to_cpu(map
->pba_of_lba0
);
1502 info
->component_size
= __le32_to_cpu(map
->blocks_per_member
);
1503 memset(info
->uuid
, 0, sizeof(info
->uuid
));
1504 info
->recovery_start
= MaxSector
;
1505 info
->reshape_active
= 0;
1507 if (map
->map_state
== IMSM_T_STATE_UNINITIALIZED
|| dev
->vol
.dirty
) {
1508 info
->resync_start
= 0;
1509 } else if (dev
->vol
.migr_state
) {
1510 switch (migr_type(dev
)) {
1513 __u64 blocks_per_unit
= blocks_per_migr_unit(dev
);
1514 __u64 units
= __le32_to_cpu(dev
->vol
.curr_migr_unit
);
1516 info
->resync_start
= blocks_per_unit
* units
;
1520 /* we could emulate the checkpointing of
1521 * 'sync_action=check' migrations, but for now
1522 * we just immediately complete them
1525 /* this is handled by container_content_imsm() */
1527 case MIGR_STATE_CHANGE
:
1528 /* FIXME handle other migrations */
1530 /* we are not dirty, so... */
1531 info
->resync_start
= MaxSector
;
1534 info
->resync_start
= MaxSector
;
1536 strncpy(info
->name
, (char *) dev
->volume
, MAX_RAID_SERIAL_LEN
);
1537 info
->name
[MAX_RAID_SERIAL_LEN
] = 0;
1539 info
->array
.major_version
= -1;
1540 info
->array
.minor_version
= -2;
1541 devname
= devnum2devname(st
->container_dev
);
1542 *info
->text_version
= '\0';
1544 sprintf(info
->text_version
, "/%s/%d", devname
, info
->container_member
);
1546 info
->safe_mode_delay
= 4000; /* 4 secs like the Matrix driver */
1547 uuid_from_super_imsm(st
, info
->uuid
);
1551 for (i
=0; i
<map_disks
; i
++) {
1553 if (i
< info
->array
.raid_disks
) {
1554 struct imsm_disk
*dsk
;
1555 j
= get_imsm_disk_idx(dev
, i
);
1556 dsk
= get_imsm_disk(super
, j
);
1557 if (dsk
&& (dsk
->status
& CONFIGURED_DISK
))
1564 /* check the config file to see if we can return a real uuid for this spare */
1565 static void fixup_container_spare_uuid(struct mdinfo
*inf
)
1567 struct mddev_ident
*array_list
;
1569 if (inf
->array
.level
!= LEVEL_CONTAINER
||
1570 memcmp(inf
->uuid
, uuid_match_any
, sizeof(int[4])) != 0)
1573 array_list
= conf_get_ident(NULL
);
1575 for (; array_list
; array_list
= array_list
->next
) {
1576 if (array_list
->uuid_set
) {
1577 struct supertype
*_sst
; /* spare supertype */
1578 struct supertype
*_cst
; /* container supertype */
1580 _cst
= array_list
->st
;
1582 _sst
= _cst
->ss
->match_metadata_desc(inf
->text_version
);
1587 memcpy(inf
->uuid
, array_list
->uuid
, sizeof(int[4]));
1596 static __u8
imsm_check_degraded(struct intel_super
*super
, struct imsm_dev
*dev
, int failed
);
1597 static int imsm_count_failed(struct intel_super
*super
, struct imsm_dev
*dev
);
1599 static struct imsm_disk
*get_imsm_missing(struct intel_super
*super
, __u8 index
)
1603 for (d
= super
->missing
; d
; d
= d
->next
)
1604 if (d
->index
== index
)
1609 static void getinfo_super_imsm(struct supertype
*st
, struct mdinfo
*info
, char *map
)
1611 struct intel_super
*super
= st
->sb
;
1612 struct imsm_disk
*disk
;
1613 int map_disks
= info
->array
.raid_disks
;
1614 int max_enough
= -1;
1616 struct imsm_super
*mpb
;
1618 if (super
->current_vol
>= 0) {
1619 getinfo_super_imsm_volume(st
, info
, map
);
1623 /* Set raid_disks to zero so that Assemble will always pull in valid
1626 info
->array
.raid_disks
= 0;
1627 info
->array
.level
= LEVEL_CONTAINER
;
1628 info
->array
.layout
= 0;
1629 info
->array
.md_minor
= -1;
1630 info
->array
.ctime
= 0; /* N/A for imsm */
1631 info
->array
.utime
= 0;
1632 info
->array
.chunk_size
= 0;
1634 info
->disk
.major
= 0;
1635 info
->disk
.minor
= 0;
1636 info
->disk
.raid_disk
= -1;
1637 info
->reshape_active
= 0;
1638 info
->array
.major_version
= -1;
1639 info
->array
.minor_version
= -2;
1640 strcpy(info
->text_version
, "imsm");
1641 info
->safe_mode_delay
= 0;
1642 info
->disk
.number
= -1;
1643 info
->disk
.state
= 0;
1645 info
->recovery_start
= MaxSector
;
1647 /* do we have the all the insync disks that we expect? */
1648 mpb
= super
->anchor
;
1650 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
1651 struct imsm_dev
*dev
= get_imsm_dev(super
, i
);
1652 int failed
, enough
, j
, missing
= 0;
1653 struct imsm_map
*map
;
1656 failed
= imsm_count_failed(super
, dev
);
1657 state
= imsm_check_degraded(super
, dev
, failed
);
1658 map
= get_imsm_map(dev
, dev
->vol
.migr_state
);
1660 /* any newly missing disks?
1661 * (catches single-degraded vs double-degraded)
1663 for (j
= 0; j
< map
->num_members
; j
++) {
1664 __u32 ord
= get_imsm_ord_tbl_ent(dev
, i
);
1665 __u32 idx
= ord_to_idx(ord
);
1667 if (!(ord
& IMSM_ORD_REBUILD
) &&
1668 get_imsm_missing(super
, idx
)) {
1674 if (state
== IMSM_T_STATE_FAILED
)
1676 else if (state
== IMSM_T_STATE_DEGRADED
&&
1677 (state
!= map
->map_state
|| missing
))
1679 else /* we're normal, or already degraded */
1682 /* in the missing/failed disk case check to see
1683 * if at least one array is runnable
1685 max_enough
= max(max_enough
, enough
);
1687 dprintf("%s: enough: %d\n", __func__
, max_enough
);
1688 info
->container_enough
= max_enough
;
1691 __u32 reserved
= imsm_reserved_sectors(super
, super
->disks
);
1693 disk
= &super
->disks
->disk
;
1694 info
->data_offset
= __le32_to_cpu(disk
->total_blocks
) - reserved
;
1695 info
->component_size
= reserved
;
1696 info
->disk
.state
= is_configured(disk
) ? (1 << MD_DISK_ACTIVE
) : 0;
1697 /* we don't change info->disk.raid_disk here because
1698 * this state will be finalized in mdmon after we have
1699 * found the 'most fresh' version of the metadata
1701 info
->disk
.state
|= is_failed(disk
) ? (1 << MD_DISK_FAULTY
) : 0;
1702 info
->disk
.state
|= is_spare(disk
) ? 0 : (1 << MD_DISK_SYNC
);
1705 /* only call uuid_from_super_imsm when this disk is part of a populated container,
1706 * ->compare_super may have updated the 'num_raid_devs' field for spares
1708 if (info
->disk
.state
& (1 << MD_DISK_SYNC
) || super
->anchor
->num_raid_devs
)
1709 uuid_from_super_imsm(st
, info
->uuid
);
1711 memcpy(info
->uuid
, uuid_match_any
, sizeof(int[4]));
1712 fixup_container_spare_uuid(info
);
1715 /* I don't know how to compute 'map' on imsm, so use safe default */
1718 for (i
= 0; i
< map_disks
; i
++)
1724 /* allocates memory and fills disk in mdinfo structure
1725 * for each disk in array */
1726 struct mdinfo
*getinfo_super_disks_imsm(struct supertype
*st
)
1728 struct mdinfo
*mddev
= NULL
;
1729 struct intel_super
*super
= st
->sb
;
1730 struct imsm_disk
*disk
;
1733 if (!super
|| !super
->disks
)
1736 mddev
= malloc(sizeof(*mddev
));
1738 fprintf(stderr
, Name
": Failed to allocate memory.\n");
1741 memset(mddev
, 0, sizeof(*mddev
));
1745 tmp
= malloc(sizeof(*tmp
));
1747 fprintf(stderr
, Name
": Failed to allocate memory.\n");
1752 memset(tmp
, 0, sizeof(*tmp
));
1754 tmp
->next
= mddev
->devs
;
1756 tmp
->disk
.number
= count
++;
1757 tmp
->disk
.major
= dl
->major
;
1758 tmp
->disk
.minor
= dl
->minor
;
1759 tmp
->disk
.state
= is_configured(disk
) ?
1760 (1 << MD_DISK_ACTIVE
) : 0;
1761 tmp
->disk
.state
|= is_failed(disk
) ? (1 << MD_DISK_FAULTY
) : 0;
1762 tmp
->disk
.state
|= is_spare(disk
) ? 0 : (1 << MD_DISK_SYNC
);
1763 tmp
->disk
.raid_disk
= -1;
1769 static int update_super_imsm(struct supertype
*st
, struct mdinfo
*info
,
1770 char *update
, char *devname
, int verbose
,
1771 int uuid_set
, char *homehost
)
1773 /* For 'assemble' and 'force' we need to return non-zero if any
1774 * change was made. For others, the return value is ignored.
1775 * Update options are:
1776 * force-one : This device looks a bit old but needs to be included,
1777 * update age info appropriately.
1778 * assemble: clear any 'faulty' flag to allow this device to
1780 * force-array: Array is degraded but being forced, mark it clean
1781 * if that will be needed to assemble it.
1783 * newdev: not used ????
1784 * grow: Array has gained a new device - this is currently for
1786 * resync: mark as dirty so a resync will happen.
1787 * name: update the name - preserving the homehost
1788 * uuid: Change the uuid of the array to match watch is given
1790 * Following are not relevant for this imsm:
1791 * sparc2.2 : update from old dodgey metadata
1792 * super-minor: change the preferred_minor number
1793 * summaries: update redundant counters.
1794 * homehost: update the recorded homehost
1795 * _reshape_progress: record new reshape_progress position.
1798 struct intel_super
*super
= st
->sb
;
1799 struct imsm_super
*mpb
;
1801 /* we can only update container info */
1802 if (!super
|| super
->current_vol
>= 0 || !super
->anchor
)
1805 mpb
= super
->anchor
;
1807 if (strcmp(update
, "uuid") == 0 && uuid_set
&& !info
->update_private
)
1809 else if (strcmp(update
, "uuid") == 0 && uuid_set
&& info
->update_private
) {
1810 mpb
->orig_family_num
= *((__u32
*) info
->update_private
);
1812 } else if (strcmp(update
, "uuid") == 0) {
1813 __u32
*new_family
= malloc(sizeof(*new_family
));
1815 /* update orig_family_number with the incoming random
1816 * data, report the new effective uuid, and store the
1817 * new orig_family_num for future updates.
1820 memcpy(&mpb
->orig_family_num
, info
->uuid
, sizeof(__u32
));
1821 uuid_from_super_imsm(st
, info
->uuid
);
1822 *new_family
= mpb
->orig_family_num
;
1823 info
->update_private
= new_family
;
1826 } else if (strcmp(update
, "assemble") == 0)
1831 /* successful update? recompute checksum */
1833 mpb
->check_sum
= __le32_to_cpu(__gen_imsm_checksum(mpb
));
1838 static size_t disks_to_mpb_size(int disks
)
1842 size
= sizeof(struct imsm_super
);
1843 size
+= (disks
- 1) * sizeof(struct imsm_disk
);
1844 size
+= 2 * sizeof(struct imsm_dev
);
1845 /* up to 2 maps per raid device (-2 for imsm_maps in imsm_dev */
1846 size
+= (4 - 2) * sizeof(struct imsm_map
);
1847 /* 4 possible disk_ord_tbl's */
1848 size
+= 4 * (disks
- 1) * sizeof(__u32
);
1853 static __u64
avail_size_imsm(struct supertype
*st
, __u64 devsize
)
1855 if (devsize
< (MPB_SECTOR_CNT
+ IMSM_RESERVED_SECTORS
))
1858 return devsize
- (MPB_SECTOR_CNT
+ IMSM_RESERVED_SECTORS
);
1861 static void free_devlist(struct intel_super
*super
)
1863 struct intel_dev
*dv
;
1865 while (super
->devlist
) {
1866 dv
= super
->devlist
->next
;
1867 free(super
->devlist
->dev
);
1868 free(super
->devlist
);
1869 super
->devlist
= dv
;
1873 static void imsm_copy_dev(struct imsm_dev
*dest
, struct imsm_dev
*src
)
1875 memcpy(dest
, src
, sizeof_imsm_dev(src
, 0));
1878 static int compare_super_imsm(struct supertype
*st
, struct supertype
*tst
)
1882 * 0 same, or first was empty, and second was copied
1883 * 1 second had wrong number
1885 * 3 wrong other info
1887 struct intel_super
*first
= st
->sb
;
1888 struct intel_super
*sec
= tst
->sb
;
1896 /* if an anchor does not have num_raid_devs set then it is a free
1899 if (first
->anchor
->num_raid_devs
> 0 &&
1900 sec
->anchor
->num_raid_devs
> 0) {
1901 /* Determine if these disks might ever have been
1902 * related. Further disambiguation can only take place
1903 * in load_super_imsm_all
1905 __u32 first_family
= first
->anchor
->orig_family_num
;
1906 __u32 sec_family
= sec
->anchor
->orig_family_num
;
1908 if (memcmp(first
->anchor
->sig
, sec
->anchor
->sig
,
1909 MAX_SIGNATURE_LENGTH
) != 0)
1912 if (first_family
== 0)
1913 first_family
= first
->anchor
->family_num
;
1914 if (sec_family
== 0)
1915 sec_family
= sec
->anchor
->family_num
;
1917 if (first_family
!= sec_family
)
1923 /* if 'first' is a spare promote it to a populated mpb with sec's
1926 if (first
->anchor
->num_raid_devs
== 0 &&
1927 sec
->anchor
->num_raid_devs
> 0) {
1929 struct intel_dev
*dv
;
1930 struct imsm_dev
*dev
;
1932 /* we need to copy raid device info from sec if an allocation
1933 * fails here we don't associate the spare
1935 for (i
= 0; i
< sec
->anchor
->num_raid_devs
; i
++) {
1936 dv
= malloc(sizeof(*dv
));
1939 dev
= malloc(sizeof_imsm_dev(get_imsm_dev(sec
, i
), 1));
1946 dv
->next
= first
->devlist
;
1947 first
->devlist
= dv
;
1949 if (i
< sec
->anchor
->num_raid_devs
) {
1950 /* allocation failure */
1951 free_devlist(first
);
1952 fprintf(stderr
, "imsm: failed to associate spare\n");
1955 first
->anchor
->num_raid_devs
= sec
->anchor
->num_raid_devs
;
1956 first
->anchor
->orig_family_num
= sec
->anchor
->orig_family_num
;
1957 first
->anchor
->family_num
= sec
->anchor
->family_num
;
1958 memcpy(first
->anchor
->sig
, sec
->anchor
->sig
, MAX_SIGNATURE_LENGTH
);
1959 for (i
= 0; i
< sec
->anchor
->num_raid_devs
; i
++)
1960 imsm_copy_dev(get_imsm_dev(first
, i
), get_imsm_dev(sec
, i
));
1966 static void fd2devname(int fd
, char *name
)
1970 char dname
[PATH_MAX
];
1975 if (fstat(fd
, &st
) != 0)
1977 sprintf(path
, "/sys/dev/block/%d:%d",
1978 major(st
.st_rdev
), minor(st
.st_rdev
));
1980 rv
= readlink(path
, dname
, sizeof(dname
));
1985 nm
= strrchr(dname
, '/');
1987 snprintf(name
, MAX_RAID_SERIAL_LEN
, "/dev/%s", nm
);
1990 extern int scsi_get_serial(int fd
, void *buf
, size_t buf_len
);
1992 static int imsm_read_serial(int fd
, char *devname
,
1993 __u8 serial
[MAX_RAID_SERIAL_LEN
])
1995 unsigned char scsi_serial
[255];
2004 memset(scsi_serial
, 0, sizeof(scsi_serial
));
2006 rv
= scsi_get_serial(fd
, scsi_serial
, sizeof(scsi_serial
));
2008 if (rv
&& check_env("IMSM_DEVNAME_AS_SERIAL")) {
2009 memset(serial
, 0, MAX_RAID_SERIAL_LEN
);
2010 fd2devname(fd
, (char *) serial
);
2017 Name
": Failed to retrieve serial for %s\n",
2022 rsp_len
= scsi_serial
[3];
2026 Name
": Failed to retrieve serial for %s\n",
2030 rsp_buf
= (char *) &scsi_serial
[4];
2032 /* trim all whitespace and non-printable characters and convert
2035 for (i
= 0, dest
= rsp_buf
; i
< rsp_len
; i
++) {
2038 /* ':' is reserved for use in placeholder serial
2039 * numbers for missing disks
2047 len
= dest
- rsp_buf
;
2050 /* truncate leading characters */
2051 if (len
> MAX_RAID_SERIAL_LEN
) {
2052 dest
+= len
- MAX_RAID_SERIAL_LEN
;
2053 len
= MAX_RAID_SERIAL_LEN
;
2056 memset(serial
, 0, MAX_RAID_SERIAL_LEN
);
2057 memcpy(serial
, dest
, len
);
2062 static int serialcmp(__u8
*s1
, __u8
*s2
)
2064 return strncmp((char *) s1
, (char *) s2
, MAX_RAID_SERIAL_LEN
);
2067 static void serialcpy(__u8
*dest
, __u8
*src
)
2069 strncpy((char *) dest
, (char *) src
, MAX_RAID_SERIAL_LEN
);
2073 static struct dl
*serial_to_dl(__u8
*serial
, struct intel_super
*super
)
2077 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
2078 if (serialcmp(dl
->serial
, serial
) == 0)
2085 static struct imsm_disk
*
2086 __serial_to_disk(__u8
*serial
, struct imsm_super
*mpb
, int *idx
)
2090 for (i
= 0; i
< mpb
->num_disks
; i
++) {
2091 struct imsm_disk
*disk
= __get_imsm_disk(mpb
, i
);
2093 if (serialcmp(disk
->serial
, serial
) == 0) {
2104 load_imsm_disk(int fd
, struct intel_super
*super
, char *devname
, int keep_fd
)
2106 struct imsm_disk
*disk
;
2111 __u8 serial
[MAX_RAID_SERIAL_LEN
];
2113 rv
= imsm_read_serial(fd
, devname
, serial
);
2118 dl
= calloc(1, sizeof(*dl
));
2122 Name
": failed to allocate disk buffer for %s\n",
2128 dl
->major
= major(stb
.st_rdev
);
2129 dl
->minor
= minor(stb
.st_rdev
);
2130 dl
->next
= super
->disks
;
2131 dl
->fd
= keep_fd
? fd
: -1;
2132 assert(super
->disks
== NULL
);
2134 serialcpy(dl
->serial
, serial
);
2137 fd2devname(fd
, name
);
2139 dl
->devname
= strdup(devname
);
2141 dl
->devname
= strdup(name
);
2143 /* look up this disk's index in the current anchor */
2144 disk
= __serial_to_disk(dl
->serial
, super
->anchor
, &dl
->index
);
2147 /* only set index on disks that are a member of a
2148 * populated contianer, i.e. one with raid_devs
2150 if (is_failed(&dl
->disk
))
2152 else if (is_spare(&dl
->disk
))
2160 /* When migrating map0 contains the 'destination' state while map1
2161 * contains the current state. When not migrating map0 contains the
2162 * current state. This routine assumes that map[0].map_state is set to
2163 * the current array state before being called.
2165 * Migration is indicated by one of the following states
2166 * 1/ Idle (migr_state=0 map0state=normal||unitialized||degraded||failed)
2167 * 2/ Initialize (migr_state=1 migr_type=MIGR_INIT map0state=normal
2168 * map1state=unitialized)
2169 * 3/ Repair (Resync) (migr_state=1 migr_type=MIGR_REPAIR map0state=normal
2171 * 4/ Rebuild (migr_state=1 migr_type=MIGR_REBUILD map0state=normal
2172 * map1state=degraded)
2174 static void migrate(struct imsm_dev
*dev
, __u8 to_state
, int migr_type
)
2176 struct imsm_map
*dest
;
2177 struct imsm_map
*src
= get_imsm_map(dev
, 0);
2179 dev
->vol
.migr_state
= 1;
2180 set_migr_type(dev
, migr_type
);
2181 dev
->vol
.curr_migr_unit
= 0;
2182 dest
= get_imsm_map(dev
, 1);
2184 /* duplicate and then set the target end state in map[0] */
2185 memcpy(dest
, src
, sizeof_imsm_map(src
));
2186 if (migr_type
== MIGR_REBUILD
) {
2190 for (i
= 0; i
< src
->num_members
; i
++) {
2191 ord
= __le32_to_cpu(src
->disk_ord_tbl
[i
]);
2192 set_imsm_ord_tbl_ent(src
, i
, ord_to_idx(ord
));
2196 src
->map_state
= to_state
;
2199 static void end_migration(struct imsm_dev
*dev
, __u8 map_state
)
2201 struct imsm_map
*map
= get_imsm_map(dev
, 0);
2202 struct imsm_map
*prev
= get_imsm_map(dev
, dev
->vol
.migr_state
);
2205 /* merge any IMSM_ORD_REBUILD bits that were not successfully
2206 * completed in the last migration.
2208 * FIXME add support for online capacity expansion and
2209 * raid-level-migration
2211 for (i
= 0; i
< prev
->num_members
; i
++)
2212 map
->disk_ord_tbl
[i
] |= prev
->disk_ord_tbl
[i
];
2214 dev
->vol
.migr_state
= 0;
2215 dev
->vol
.curr_migr_unit
= 0;
2216 map
->map_state
= map_state
;
2220 static int parse_raid_devices(struct intel_super
*super
)
2223 struct imsm_dev
*dev_new
;
2224 size_t len
, len_migr
;
2225 size_t space_needed
= 0;
2226 struct imsm_super
*mpb
= super
->anchor
;
2228 for (i
= 0; i
< super
->anchor
->num_raid_devs
; i
++) {
2229 struct imsm_dev
*dev_iter
= __get_imsm_dev(super
->anchor
, i
);
2230 struct intel_dev
*dv
;
2232 len
= sizeof_imsm_dev(dev_iter
, 0);
2233 len_migr
= sizeof_imsm_dev(dev_iter
, 1);
2235 space_needed
+= len_migr
- len
;
2237 dv
= malloc(sizeof(*dv
));
2240 dev_new
= malloc(len_migr
);
2245 imsm_copy_dev(dev_new
, dev_iter
);
2248 dv
->next
= super
->devlist
;
2249 super
->devlist
= dv
;
2252 /* ensure that super->buf is large enough when all raid devices
2255 if (__le32_to_cpu(mpb
->mpb_size
) + space_needed
> super
->len
) {
2258 len
= ROUND_UP(__le32_to_cpu(mpb
->mpb_size
) + space_needed
, 512);
2259 if (posix_memalign(&buf
, 512, len
) != 0)
2262 memcpy(buf
, super
->buf
, super
->len
);
2263 memset(buf
+ super
->len
, 0, len
- super
->len
);
2272 /* retrieve a pointer to the bbm log which starts after all raid devices */
2273 struct bbm_log
*__get_imsm_bbm_log(struct imsm_super
*mpb
)
2277 if (__le32_to_cpu(mpb
->bbm_log_size
)) {
2279 ptr
+= mpb
->mpb_size
- __le32_to_cpu(mpb
->bbm_log_size
);
2285 static void __free_imsm(struct intel_super
*super
, int free_disks
);
2287 /* load_imsm_mpb - read matrix metadata
2288 * allocates super->mpb to be freed by free_super
2290 static int load_imsm_mpb(int fd
, struct intel_super
*super
, char *devname
)
2292 unsigned long long dsize
;
2293 unsigned long long sectors
;
2295 struct imsm_super
*anchor
;
2298 get_dev_size(fd
, NULL
, &dsize
);
2302 Name
": %s: device to small for imsm\n",
2307 if (lseek64(fd
, dsize
- (512 * 2), SEEK_SET
) < 0) {
2310 Name
": Cannot seek to anchor block on %s: %s\n",
2311 devname
, strerror(errno
));
2315 if (posix_memalign((void**)&anchor
, 512, 512) != 0) {
2318 Name
": Failed to allocate imsm anchor buffer"
2319 " on %s\n", devname
);
2322 if (read(fd
, anchor
, 512) != 512) {
2325 Name
": Cannot read anchor block on %s: %s\n",
2326 devname
, strerror(errno
));
2331 if (strncmp((char *) anchor
->sig
, MPB_SIGNATURE
, MPB_SIG_LEN
) != 0) {
2334 Name
": no IMSM anchor on %s\n", devname
);
2339 __free_imsm(super
, 0);
2340 super
->len
= ROUND_UP(anchor
->mpb_size
, 512);
2341 if (posix_memalign(&super
->buf
, 512, super
->len
) != 0) {
2344 Name
": unable to allocate %zu byte mpb buffer\n",
2349 memcpy(super
->buf
, anchor
, 512);
2351 sectors
= mpb_sectors(anchor
) - 1;
2354 check_sum
= __gen_imsm_checksum(super
->anchor
);
2355 if (check_sum
!= __le32_to_cpu(super
->anchor
->check_sum
)) {
2358 Name
": IMSM checksum %x != %x on %s\n",
2360 __le32_to_cpu(super
->anchor
->check_sum
),
2368 /* read the extended mpb */
2369 if (lseek64(fd
, dsize
- (512 * (2 + sectors
)), SEEK_SET
) < 0) {
2372 Name
": Cannot seek to extended mpb on %s: %s\n",
2373 devname
, strerror(errno
));
2377 if ((unsigned)read(fd
, super
->buf
+ 512, super
->len
- 512) != super
->len
- 512) {
2380 Name
": Cannot read extended mpb on %s: %s\n",
2381 devname
, strerror(errno
));
2385 check_sum
= __gen_imsm_checksum(super
->anchor
);
2386 if (check_sum
!= __le32_to_cpu(super
->anchor
->check_sum
)) {
2389 Name
": IMSM checksum %x != %x on %s\n",
2390 check_sum
, __le32_to_cpu(super
->anchor
->check_sum
),
2395 /* FIXME the BBM log is disk specific so we cannot use this global
2396 * buffer for all disks. Ok for now since we only look at the global
2397 * bbm_log_size parameter to gate assembly
2399 super
->bbm_log
= __get_imsm_bbm_log(super
->anchor
);
2405 load_and_parse_mpb(int fd
, struct intel_super
*super
, char *devname
, int keep_fd
)
2409 err
= load_imsm_mpb(fd
, super
, devname
);
2412 err
= load_imsm_disk(fd
, super
, devname
, keep_fd
);
2415 err
= parse_raid_devices(super
);
2420 static void __free_imsm_disk(struct dl
*d
)
2431 static void free_imsm_disks(struct intel_super
*super
)
2435 while (super
->disks
) {
2437 super
->disks
= d
->next
;
2438 __free_imsm_disk(d
);
2440 while (super
->missing
) {
2442 super
->missing
= d
->next
;
2443 __free_imsm_disk(d
);
2448 /* free all the pieces hanging off of a super pointer */
2449 static void __free_imsm(struct intel_super
*super
, int free_disks
)
2456 free_imsm_disks(super
);
2457 free_devlist(super
);
2459 free((void *) super
->hba
);
2464 static void free_imsm(struct intel_super
*super
)
2466 __free_imsm(super
, 1);
2470 static void free_super_imsm(struct supertype
*st
)
2472 struct intel_super
*super
= st
->sb
;
2481 static struct intel_super
*alloc_super(void)
2483 struct intel_super
*super
= malloc(sizeof(*super
));
2486 memset(super
, 0, sizeof(*super
));
2487 super
->current_vol
= -1;
2488 super
->create_offset
= ~((__u32
) 0);
2489 if (!check_env("IMSM_NO_PLATFORM"))
2490 super
->orom
= find_imsm_orom();
2491 if (super
->orom
&& !check_env("IMSM_TEST_OROM")) {
2492 struct sys_dev
*list
, *ent
;
2494 /* find the first intel ahci controller */
2495 list
= find_driver_devices("pci", "ahci");
2496 for (ent
= list
; ent
; ent
= ent
->next
)
2497 if (devpath_to_vendor(ent
->path
) == 0x8086)
2500 super
->hba
= ent
->path
;
2503 free_sys_dev(&list
);
2511 /* find_missing - helper routine for load_super_imsm_all that identifies
2512 * disks that have disappeared from the system. This routine relies on
2513 * the mpb being uptodate, which it is at load time.
2515 static int find_missing(struct intel_super
*super
)
2518 struct imsm_super
*mpb
= super
->anchor
;
2520 struct imsm_disk
*disk
;
2522 for (i
= 0; i
< mpb
->num_disks
; i
++) {
2523 disk
= __get_imsm_disk(mpb
, i
);
2524 dl
= serial_to_dl(disk
->serial
, super
);
2528 dl
= malloc(sizeof(*dl
));
2534 dl
->devname
= strdup("missing");
2536 serialcpy(dl
->serial
, disk
->serial
);
2539 dl
->next
= super
->missing
;
2540 super
->missing
= dl
;
2546 static struct intel_disk
*disk_list_get(__u8
*serial
, struct intel_disk
*disk_list
)
2548 struct intel_disk
*idisk
= disk_list
;
2551 if (serialcmp(idisk
->disk
.serial
, serial
) == 0)
2553 idisk
= idisk
->next
;
2559 static int __prep_thunderdome(struct intel_super
**table
, int tbl_size
,
2560 struct intel_super
*super
,
2561 struct intel_disk
**disk_list
)
2563 struct imsm_disk
*d
= &super
->disks
->disk
;
2564 struct imsm_super
*mpb
= super
->anchor
;
2567 for (i
= 0; i
< tbl_size
; i
++) {
2568 struct imsm_super
*tbl_mpb
= table
[i
]->anchor
;
2569 struct imsm_disk
*tbl_d
= &table
[i
]->disks
->disk
;
2571 if (tbl_mpb
->family_num
== mpb
->family_num
) {
2572 if (tbl_mpb
->check_sum
== mpb
->check_sum
) {
2573 dprintf("%s: mpb from %d:%d matches %d:%d\n",
2574 __func__
, super
->disks
->major
,
2575 super
->disks
->minor
,
2576 table
[i
]->disks
->major
,
2577 table
[i
]->disks
->minor
);
2581 if (((is_configured(d
) && !is_configured(tbl_d
)) ||
2582 is_configured(d
) == is_configured(tbl_d
)) &&
2583 tbl_mpb
->generation_num
< mpb
->generation_num
) {
2584 /* current version of the mpb is a
2585 * better candidate than the one in
2586 * super_table, but copy over "cross
2587 * generational" status
2589 struct intel_disk
*idisk
;
2591 dprintf("%s: mpb from %d:%d replaces %d:%d\n",
2592 __func__
, super
->disks
->major
,
2593 super
->disks
->minor
,
2594 table
[i
]->disks
->major
,
2595 table
[i
]->disks
->minor
);
2597 idisk
= disk_list_get(tbl_d
->serial
, *disk_list
);
2598 if (idisk
&& is_failed(&idisk
->disk
))
2599 tbl_d
->status
|= FAILED_DISK
;
2602 struct intel_disk
*idisk
;
2603 struct imsm_disk
*disk
;
2605 /* tbl_mpb is more up to date, but copy
2606 * over cross generational status before
2609 disk
= __serial_to_disk(d
->serial
, mpb
, NULL
);
2610 if (disk
&& is_failed(disk
))
2611 d
->status
|= FAILED_DISK
;
2613 idisk
= disk_list_get(d
->serial
, *disk_list
);
2616 if (disk
&& is_configured(disk
))
2617 idisk
->disk
.status
|= CONFIGURED_DISK
;
2620 dprintf("%s: mpb from %d:%d prefer %d:%d\n",
2621 __func__
, super
->disks
->major
,
2622 super
->disks
->minor
,
2623 table
[i
]->disks
->major
,
2624 table
[i
]->disks
->minor
);
2632 table
[tbl_size
++] = super
;
2636 /* update/extend the merged list of imsm_disk records */
2637 for (j
= 0; j
< mpb
->num_disks
; j
++) {
2638 struct imsm_disk
*disk
= __get_imsm_disk(mpb
, j
);
2639 struct intel_disk
*idisk
;
2641 idisk
= disk_list_get(disk
->serial
, *disk_list
);
2643 idisk
->disk
.status
|= disk
->status
;
2644 if (is_configured(&idisk
->disk
) ||
2645 is_failed(&idisk
->disk
))
2646 idisk
->disk
.status
&= ~(SPARE_DISK
);
2648 idisk
= calloc(1, sizeof(*idisk
));
2651 idisk
->owner
= IMSM_UNKNOWN_OWNER
;
2652 idisk
->disk
= *disk
;
2653 idisk
->next
= *disk_list
;
2657 if (serialcmp(idisk
->disk
.serial
, d
->serial
) == 0)
2664 static struct intel_super
*
2665 validate_members(struct intel_super
*super
, struct intel_disk
*disk_list
,
2668 struct imsm_super
*mpb
= super
->anchor
;
2672 for (i
= 0; i
< mpb
->num_disks
; i
++) {
2673 struct imsm_disk
*disk
= __get_imsm_disk(mpb
, i
);
2674 struct intel_disk
*idisk
;
2676 idisk
= disk_list_get(disk
->serial
, disk_list
);
2678 if (idisk
->owner
== owner
||
2679 idisk
->owner
== IMSM_UNKNOWN_OWNER
)
2682 dprintf("%s: '%.16s' owner %d != %d\n",
2683 __func__
, disk
->serial
, idisk
->owner
,
2686 dprintf("%s: unknown disk %x [%d]: %.16s\n",
2687 __func__
, __le32_to_cpu(mpb
->family_num
), i
,
2693 if (ok_count
== mpb
->num_disks
)
2698 static void show_conflicts(__u32 family_num
, struct intel_super
*super_list
)
2700 struct intel_super
*s
;
2702 for (s
= super_list
; s
; s
= s
->next
) {
2703 if (family_num
!= s
->anchor
->family_num
)
2705 fprintf(stderr
, "Conflict, offlining family %#x on '%s'\n",
2706 __le32_to_cpu(family_num
), s
->disks
->devname
);
2710 static struct intel_super
*
2711 imsm_thunderdome(struct intel_super
**super_list
, int len
)
2713 struct intel_super
*super_table
[len
];
2714 struct intel_disk
*disk_list
= NULL
;
2715 struct intel_super
*champion
, *spare
;
2716 struct intel_super
*s
, **del
;
2721 memset(super_table
, 0, sizeof(super_table
));
2722 for (s
= *super_list
; s
; s
= s
->next
)
2723 tbl_size
= __prep_thunderdome(super_table
, tbl_size
, s
, &disk_list
);
2725 for (i
= 0; i
< tbl_size
; i
++) {
2726 struct imsm_disk
*d
;
2727 struct intel_disk
*idisk
;
2728 struct imsm_super
*mpb
= super_table
[i
]->anchor
;
2731 d
= &s
->disks
->disk
;
2733 /* 'd' must appear in merged disk list for its
2734 * configuration to be valid
2736 idisk
= disk_list_get(d
->serial
, disk_list
);
2737 if (idisk
&& idisk
->owner
== i
)
2738 s
= validate_members(s
, disk_list
, i
);
2743 dprintf("%s: marking family: %#x from %d:%d offline\n",
2744 __func__
, mpb
->family_num
,
2745 super_table
[i
]->disks
->major
,
2746 super_table
[i
]->disks
->minor
);
2750 /* This is where the mdadm implementation differs from the Windows
2751 * driver which has no strict concept of a container. We can only
2752 * assemble one family from a container, so when returning a prodigal
2753 * array member to this system the code will not be able to disambiguate
2754 * the container contents that should be assembled ("foreign" versus
2755 * "local"). It requires user intervention to set the orig_family_num
2756 * to a new value to establish a new container. The Windows driver in
2757 * this situation fixes up the volume name in place and manages the
2758 * foreign array as an independent entity.
2763 for (i
= 0; i
< tbl_size
; i
++) {
2764 struct intel_super
*tbl_ent
= super_table
[i
];
2770 if (tbl_ent
->anchor
->num_raid_devs
== 0) {
2775 if (s
&& !is_spare
) {
2776 show_conflicts(tbl_ent
->anchor
->family_num
, *super_list
);
2778 } else if (!s
&& !is_spare
)
2791 fprintf(stderr
, "Chose family %#x on '%s', "
2792 "assemble conflicts to new container with '--update=uuid'\n",
2793 __le32_to_cpu(s
->anchor
->family_num
), s
->disks
->devname
);
2795 /* collect all dl's onto 'champion', and update them to
2796 * champion's version of the status
2798 for (s
= *super_list
; s
; s
= s
->next
) {
2799 struct imsm_super
*mpb
= champion
->anchor
;
2800 struct dl
*dl
= s
->disks
;
2805 for (i
= 0; i
< mpb
->num_disks
; i
++) {
2806 struct imsm_disk
*disk
;
2808 disk
= __serial_to_disk(dl
->serial
, mpb
, &dl
->index
);
2811 /* only set index on disks that are a member of
2812 * a populated contianer, i.e. one with
2815 if (is_failed(&dl
->disk
))
2817 else if (is_spare(&dl
->disk
))
2823 if (i
>= mpb
->num_disks
) {
2824 struct intel_disk
*idisk
;
2826 idisk
= disk_list_get(dl
->serial
, disk_list
);
2827 if (idisk
&& is_spare(&idisk
->disk
) &&
2828 !is_failed(&idisk
->disk
) && !is_configured(&idisk
->disk
))
2836 dl
->next
= champion
->disks
;
2837 champion
->disks
= dl
;
2841 /* delete 'champion' from super_list */
2842 for (del
= super_list
; *del
; ) {
2843 if (*del
== champion
) {
2844 *del
= (*del
)->next
;
2847 del
= &(*del
)->next
;
2849 champion
->next
= NULL
;
2853 struct intel_disk
*idisk
= disk_list
;
2855 disk_list
= disk_list
->next
;
2862 static int load_super_imsm_all(struct supertype
*st
, int fd
, void **sbp
,
2866 struct intel_super
*super_list
= NULL
;
2867 struct intel_super
*super
= NULL
;
2868 int devnum
= fd2devnum(fd
);
2874 /* check if 'fd' an opened container */
2875 sra
= sysfs_read(fd
, 0, GET_LEVEL
|GET_VERSION
|GET_DEVS
|GET_STATE
);
2879 if (sra
->array
.major_version
!= -1 ||
2880 sra
->array
.minor_version
!= -2 ||
2881 strcmp(sra
->text_version
, "imsm") != 0) {
2886 for (sd
= sra
->devs
, i
= 0; sd
; sd
= sd
->next
, i
++) {
2887 struct intel_super
*s
= alloc_super();
2894 s
->next
= super_list
;
2898 sprintf(nm
, "%d:%d", sd
->disk
.major
, sd
->disk
.minor
);
2899 dfd
= dev_open(nm
, O_RDWR
);
2903 err
= load_and_parse_mpb(dfd
, s
, NULL
, 1);
2905 /* retry the load if we might have raced against mdmon */
2906 if (err
== 3 && mdmon_running(devnum
))
2907 for (retry
= 0; retry
< 3; retry
++) {
2909 err
= load_and_parse_mpb(dfd
, s
, NULL
, 1);
2917 /* all mpbs enter, maybe one leaves */
2918 super
= imsm_thunderdome(&super_list
, i
);
2924 if (find_missing(super
) != 0) {
2932 while (super_list
) {
2933 struct intel_super
*s
= super_list
;
2935 super_list
= super_list
->next
;
2944 st
->container_dev
= devnum
;
2945 if (err
== 0 && st
->ss
== NULL
) {
2946 st
->ss
= &super_imsm
;
2947 st
->minor_version
= 0;
2948 st
->max_devs
= IMSM_MAX_DEVICES
;
2953 static int load_container_imsm(struct supertype
*st
, int fd
, char *devname
)
2955 return load_super_imsm_all(st
, fd
, &st
->sb
, devname
);
2959 static int load_super_imsm(struct supertype
*st
, int fd
, char *devname
)
2961 struct intel_super
*super
;
2965 if (load_super_imsm_all(st
, fd
, &st
->sb
, devname
) == 0)
2969 if (test_partition(fd
))
2970 /* IMSM not allowed on partitions */
2973 free_super_imsm(st
);
2975 super
= alloc_super();
2978 Name
": malloc of %zu failed.\n",
2983 rv
= load_and_parse_mpb(fd
, super
, devname
, 0);
2988 Name
": Failed to load all information "
2989 "sections on %s\n", devname
);
2995 if (st
->ss
== NULL
) {
2996 st
->ss
= &super_imsm
;
2997 st
->minor_version
= 0;
2998 st
->max_devs
= IMSM_MAX_DEVICES
;
3003 static __u16
info_to_blocks_per_strip(mdu_array_info_t
*info
)
3005 if (info
->level
== 1)
3007 return info
->chunk_size
>> 9;
3010 static __u32
info_to_num_data_stripes(mdu_array_info_t
*info
, int num_domains
)
3014 num_stripes
= (info
->size
* 2) / info_to_blocks_per_strip(info
);
3015 num_stripes
/= num_domains
;
3020 static __u32
info_to_blocks_per_member(mdu_array_info_t
*info
)
3022 if (info
->level
== 1)
3023 return info
->size
* 2;
3025 return (info
->size
* 2) & ~(info_to_blocks_per_strip(info
) - 1);
3028 static void imsm_update_version_info(struct intel_super
*super
)
3030 /* update the version and attributes */
3031 struct imsm_super
*mpb
= super
->anchor
;
3033 struct imsm_dev
*dev
;
3034 struct imsm_map
*map
;
3037 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
3038 dev
= get_imsm_dev(super
, i
);
3039 map
= get_imsm_map(dev
, 0);
3040 if (__le32_to_cpu(dev
->size_high
) > 0)
3041 mpb
->attributes
|= MPB_ATTRIB_2TB
;
3043 /* FIXME detect when an array spans a port multiplier */
3045 mpb
->attributes
|= MPB_ATTRIB_PM
;
3048 if (mpb
->num_raid_devs
> 1 ||
3049 mpb
->attributes
!= MPB_ATTRIB_CHECKSUM_VERIFY
) {
3050 version
= MPB_VERSION_ATTRIBS
;
3051 switch (get_imsm_raid_level(map
)) {
3052 case 0: mpb
->attributes
|= MPB_ATTRIB_RAID0
; break;
3053 case 1: mpb
->attributes
|= MPB_ATTRIB_RAID1
; break;
3054 case 10: mpb
->attributes
|= MPB_ATTRIB_RAID10
; break;
3055 case 5: mpb
->attributes
|= MPB_ATTRIB_RAID5
; break;
3058 if (map
->num_members
>= 5)
3059 version
= MPB_VERSION_5OR6_DISK_ARRAY
;
3060 else if (dev
->status
== DEV_CLONE_N_GO
)
3061 version
= MPB_VERSION_CNG
;
3062 else if (get_imsm_raid_level(map
) == 5)
3063 version
= MPB_VERSION_RAID5
;
3064 else if (map
->num_members
>= 3)
3065 version
= MPB_VERSION_3OR4_DISK_ARRAY
;
3066 else if (get_imsm_raid_level(map
) == 1)
3067 version
= MPB_VERSION_RAID1
;
3069 version
= MPB_VERSION_RAID0
;
3071 strcpy(((char *) mpb
->sig
) + strlen(MPB_SIGNATURE
), version
);
3075 static int check_name(struct intel_super
*super
, char *name
, int quiet
)
3077 struct imsm_super
*mpb
= super
->anchor
;
3078 char *reason
= NULL
;
3081 if (strlen(name
) > MAX_RAID_SERIAL_LEN
)
3082 reason
= "must be 16 characters or less";
3084 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
3085 struct imsm_dev
*dev
= get_imsm_dev(super
, i
);
3087 if (strncmp((char *) dev
->volume
, name
, MAX_RAID_SERIAL_LEN
) == 0) {
3088 reason
= "already exists";
3093 if (reason
&& !quiet
)
3094 fprintf(stderr
, Name
": imsm volume name %s\n", reason
);
3099 static int init_super_imsm_volume(struct supertype
*st
, mdu_array_info_t
*info
,
3100 unsigned long long size
, char *name
,
3101 char *homehost
, int *uuid
)
3103 /* We are creating a volume inside a pre-existing container.
3104 * so st->sb is already set.
3106 struct intel_super
*super
= st
->sb
;
3107 struct imsm_super
*mpb
= super
->anchor
;
3108 struct intel_dev
*dv
;
3109 struct imsm_dev
*dev
;
3110 struct imsm_vol
*vol
;
3111 struct imsm_map
*map
;
3112 int idx
= mpb
->num_raid_devs
;
3114 unsigned long long array_blocks
;
3115 size_t size_old
, size_new
;
3116 __u32 num_data_stripes
;
3118 if (super
->orom
&& mpb
->num_raid_devs
>= super
->orom
->vpa
) {
3119 fprintf(stderr
, Name
": This imsm-container already has the "
3120 "maximum of %d volumes\n", super
->orom
->vpa
);
3124 /* ensure the mpb is large enough for the new data */
3125 size_old
= __le32_to_cpu(mpb
->mpb_size
);
3126 size_new
= disks_to_mpb_size(info
->nr_disks
);
3127 if (size_new
> size_old
) {
3129 size_t size_round
= ROUND_UP(size_new
, 512);
3131 if (posix_memalign(&mpb_new
, 512, size_round
) != 0) {
3132 fprintf(stderr
, Name
": could not allocate new mpb\n");
3135 memcpy(mpb_new
, mpb
, size_old
);
3138 super
->anchor
= mpb_new
;
3139 mpb
->mpb_size
= __cpu_to_le32(size_new
);
3140 memset(mpb_new
+ size_old
, 0, size_round
- size_old
);
3142 super
->current_vol
= idx
;
3143 /* when creating the first raid device in this container set num_disks
3144 * to zero, i.e. delete this spare and add raid member devices in
3145 * add_to_super_imsm_volume()
3147 if (super
->current_vol
== 0)
3150 if (!check_name(super
, name
, 0))
3152 dv
= malloc(sizeof(*dv
));
3154 fprintf(stderr
, Name
": failed to allocate device list entry\n");
3157 dev
= malloc(sizeof(*dev
) + sizeof(__u32
) * (info
->raid_disks
- 1));
3160 fprintf(stderr
, Name
": could not allocate raid device\n");
3163 strncpy((char *) dev
->volume
, name
, MAX_RAID_SERIAL_LEN
);
3164 if (info
->level
== 1)
3165 array_blocks
= info_to_blocks_per_member(info
);
3167 array_blocks
= calc_array_size(info
->level
, info
->raid_disks
,
3168 info
->layout
, info
->chunk_size
,
3170 /* round array size down to closest MB */
3171 array_blocks
= (array_blocks
>> SECT_PER_MB_SHIFT
) << SECT_PER_MB_SHIFT
;
3173 dev
->size_low
= __cpu_to_le32((__u32
) array_blocks
);
3174 dev
->size_high
= __cpu_to_le32((__u32
) (array_blocks
>> 32));
3175 dev
->status
= __cpu_to_le32(0);
3176 dev
->reserved_blocks
= __cpu_to_le32(0);
3178 vol
->migr_state
= 0;
3179 set_migr_type(dev
, MIGR_INIT
);
3181 vol
->curr_migr_unit
= 0;
3182 map
= get_imsm_map(dev
, 0);
3183 map
->pba_of_lba0
= __cpu_to_le32(super
->create_offset
);
3184 map
->blocks_per_member
= __cpu_to_le32(info_to_blocks_per_member(info
));
3185 map
->blocks_per_strip
= __cpu_to_le16(info_to_blocks_per_strip(info
));
3186 map
->failed_disk_num
= ~0;
3187 map
->map_state
= info
->level
? IMSM_T_STATE_UNINITIALIZED
:
3188 IMSM_T_STATE_NORMAL
;
3191 if (info
->level
== 1 && info
->raid_disks
> 2) {
3194 fprintf(stderr
, Name
": imsm does not support more than 2 disks"
3195 "in a raid1 volume\n");
3199 map
->raid_level
= info
->level
;
3200 if (info
->level
== 10) {
3201 map
->raid_level
= 1;
3202 map
->num_domains
= info
->raid_disks
/ 2;
3203 } else if (info
->level
== 1)
3204 map
->num_domains
= info
->raid_disks
;
3206 map
->num_domains
= 1;
3208 num_data_stripes
= info_to_num_data_stripes(info
, map
->num_domains
);
3209 map
->num_data_stripes
= __cpu_to_le32(num_data_stripes
);
3211 map
->num_members
= info
->raid_disks
;
3212 for (i
= 0; i
< map
->num_members
; i
++) {
3213 /* initialized in add_to_super */
3214 set_imsm_ord_tbl_ent(map
, i
, IMSM_ORD_REBUILD
);
3216 mpb
->num_raid_devs
++;
3219 dv
->index
= super
->current_vol
;
3220 dv
->next
= super
->devlist
;
3221 super
->devlist
= dv
;
3223 imsm_update_version_info(super
);
3228 static int init_super_imsm(struct supertype
*st
, mdu_array_info_t
*info
,
3229 unsigned long long size
, char *name
,
3230 char *homehost
, int *uuid
)
3232 /* This is primarily called by Create when creating a new array.
3233 * We will then get add_to_super called for each component, and then
3234 * write_init_super called to write it out to each device.
3235 * For IMSM, Create can create on fresh devices or on a pre-existing
3237 * To create on a pre-existing array a different method will be called.
3238 * This one is just for fresh drives.
3240 struct intel_super
*super
;
3241 struct imsm_super
*mpb
;
3246 return init_super_imsm_volume(st
, info
, size
, name
, homehost
, uuid
);
3249 mpb_size
= disks_to_mpb_size(info
->nr_disks
);
3253 super
= alloc_super();
3254 if (super
&& posix_memalign(&super
->buf
, 512, mpb_size
) != 0) {
3259 fprintf(stderr
, Name
3260 ": %s could not allocate superblock\n", __func__
);
3263 memset(super
->buf
, 0, mpb_size
);
3265 mpb
->mpb_size
= __cpu_to_le32(mpb_size
);
3269 /* zeroing superblock */
3273 mpb
->attributes
= MPB_ATTRIB_CHECKSUM_VERIFY
;
3275 version
= (char *) mpb
->sig
;
3276 strcpy(version
, MPB_SIGNATURE
);
3277 version
+= strlen(MPB_SIGNATURE
);
3278 strcpy(version
, MPB_VERSION_RAID0
);
3284 static int add_to_super_imsm_volume(struct supertype
*st
, mdu_disk_info_t
*dk
,
3285 int fd
, char *devname
)
3287 struct intel_super
*super
= st
->sb
;
3288 struct imsm_super
*mpb
= super
->anchor
;
3290 struct imsm_dev
*dev
;
3291 struct imsm_map
*map
;
3294 dev
= get_imsm_dev(super
, super
->current_vol
);
3295 map
= get_imsm_map(dev
, 0);
3297 if (! (dk
->state
& (1<<MD_DISK_SYNC
))) {
3298 fprintf(stderr
, Name
": %s: Cannot add spare devices to IMSM volume\n",
3304 /* we're doing autolayout so grab the pre-marked (in
3305 * validate_geometry) raid_disk
3307 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
3308 if (dl
->raiddisk
== dk
->raid_disk
)
3311 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
3312 if (dl
->major
== dk
->major
&&
3313 dl
->minor
== dk
->minor
)
3318 fprintf(stderr
, Name
": %s is not a member of the same container\n", devname
);
3322 /* add a pristine spare to the metadata */
3323 if (dl
->index
< 0) {
3324 dl
->index
= super
->anchor
->num_disks
;
3325 super
->anchor
->num_disks
++;
3327 /* Check the device has not already been added */
3328 slot
= get_imsm_disk_slot(map
, dl
->index
);
3330 (get_imsm_ord_tbl_ent(dev
, slot
) & IMSM_ORD_REBUILD
) == 0) {
3331 fprintf(stderr
, Name
": %s has been included in this array twice\n",
3335 set_imsm_ord_tbl_ent(map
, dk
->number
, dl
->index
);
3336 dl
->disk
.status
= CONFIGURED_DISK
;
3338 /* if we are creating the first raid device update the family number */
3339 if (super
->current_vol
== 0) {
3341 struct imsm_dev
*_dev
= __get_imsm_dev(mpb
, 0);
3342 struct imsm_disk
*_disk
= __get_imsm_disk(mpb
, dl
->index
);
3344 if (!_dev
|| !_disk
) {
3345 fprintf(stderr
, Name
": BUG mpb setup error\n");
3351 sum
+= __gen_imsm_checksum(mpb
);
3352 mpb
->family_num
= __cpu_to_le32(sum
);
3353 mpb
->orig_family_num
= mpb
->family_num
;
3359 static int add_to_super_imsm(struct supertype
*st
, mdu_disk_info_t
*dk
,
3360 int fd
, char *devname
)
3362 struct intel_super
*super
= st
->sb
;
3364 unsigned long long size
;
3369 /* if we are on an RAID enabled platform check that the disk is
3370 * attached to the raid controller
3372 if (super
->hba
&& !disk_attached_to_hba(fd
, super
->hba
)) {
3374 Name
": %s is not attached to the raid controller: %s\n",
3375 devname
? : "disk", super
->hba
);
3379 if (super
->current_vol
>= 0)
3380 return add_to_super_imsm_volume(st
, dk
, fd
, devname
);
3383 dd
= malloc(sizeof(*dd
));
3386 Name
": malloc failed %s:%d.\n", __func__
, __LINE__
);
3389 memset(dd
, 0, sizeof(*dd
));
3390 dd
->major
= major(stb
.st_rdev
);
3391 dd
->minor
= minor(stb
.st_rdev
);
3393 dd
->devname
= devname
? strdup(devname
) : NULL
;
3396 rv
= imsm_read_serial(fd
, devname
, dd
->serial
);
3399 Name
": failed to retrieve scsi serial, aborting\n");
3404 get_dev_size(fd
, NULL
, &size
);
3406 serialcpy(dd
->disk
.serial
, dd
->serial
);
3407 dd
->disk
.total_blocks
= __cpu_to_le32(size
);
3408 dd
->disk
.status
= SPARE_DISK
;
3409 if (sysfs_disk_to_scsi_id(fd
, &id
) == 0)
3410 dd
->disk
.scsi_id
= __cpu_to_le32(id
);
3412 dd
->disk
.scsi_id
= __cpu_to_le32(0);
3414 if (st
->update_tail
) {
3415 dd
->next
= super
->add
;
3418 dd
->next
= super
->disks
;
3425 static int store_imsm_mpb(int fd
, struct imsm_super
*mpb
);
3429 struct imsm_super anchor
;
3430 } spare_record
__attribute__ ((aligned(512)));
3432 /* spare records have their own family number and do not have any defined raid
3435 static int write_super_imsm_spares(struct intel_super
*super
, int doclose
)
3437 struct imsm_super
*mpb
= super
->anchor
;
3438 struct imsm_super
*spare
= &spare_record
.anchor
;
3442 spare
->mpb_size
= __cpu_to_le32(sizeof(struct imsm_super
)),
3443 spare
->generation_num
= __cpu_to_le32(1UL),
3444 spare
->attributes
= MPB_ATTRIB_CHECKSUM_VERIFY
;
3445 spare
->num_disks
= 1,
3446 spare
->num_raid_devs
= 0,
3447 spare
->cache_size
= mpb
->cache_size
,
3448 spare
->pwr_cycle_count
= __cpu_to_le32(1),
3450 snprintf((char *) spare
->sig
, MAX_SIGNATURE_LENGTH
,
3451 MPB_SIGNATURE MPB_VERSION_RAID0
);
3453 for (d
= super
->disks
; d
; d
= d
->next
) {
3457 spare
->disk
[0] = d
->disk
;
3458 sum
= __gen_imsm_checksum(spare
);
3459 spare
->family_num
= __cpu_to_le32(sum
);
3460 spare
->orig_family_num
= 0;
3461 sum
= __gen_imsm_checksum(spare
);
3462 spare
->check_sum
= __cpu_to_le32(sum
);
3464 if (store_imsm_mpb(d
->fd
, spare
)) {
3465 fprintf(stderr
, "%s: failed for device %d:%d %s\n",
3466 __func__
, d
->major
, d
->minor
, strerror(errno
));
3478 static int write_super_imsm(struct intel_super
*super
, int doclose
)
3480 struct imsm_super
*mpb
= super
->anchor
;
3486 __u32 mpb_size
= sizeof(struct imsm_super
) - sizeof(struct imsm_disk
);
3488 /* 'generation' is incremented everytime the metadata is written */
3489 generation
= __le32_to_cpu(mpb
->generation_num
);
3491 mpb
->generation_num
= __cpu_to_le32(generation
);
3493 /* fix up cases where previous mdadm releases failed to set
3496 if (mpb
->orig_family_num
== 0)
3497 mpb
->orig_family_num
= mpb
->family_num
;
3499 mpb_size
+= sizeof(struct imsm_disk
) * mpb
->num_disks
;
3500 for (d
= super
->disks
; d
; d
= d
->next
) {
3504 mpb
->disk
[d
->index
] = d
->disk
;
3506 for (d
= super
->missing
; d
; d
= d
->next
)
3507 mpb
->disk
[d
->index
] = d
->disk
;
3509 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
3510 struct imsm_dev
*dev
= __get_imsm_dev(mpb
, i
);
3512 imsm_copy_dev(dev
, get_imsm_dev(super
, i
));
3513 mpb_size
+= sizeof_imsm_dev(dev
, 0);
3515 mpb_size
+= __le32_to_cpu(mpb
->bbm_log_size
);
3516 mpb
->mpb_size
= __cpu_to_le32(mpb_size
);
3518 /* recalculate checksum */
3519 sum
= __gen_imsm_checksum(mpb
);
3520 mpb
->check_sum
= __cpu_to_le32(sum
);
3522 /* write the mpb for disks that compose raid devices */
3523 for (d
= super
->disks
; d
; d
= d
->next
) {
3526 if (store_imsm_mpb(d
->fd
, mpb
))
3527 fprintf(stderr
, "%s: failed for device %d:%d %s\n",
3528 __func__
, d
->major
, d
->minor
, strerror(errno
));
3536 return write_super_imsm_spares(super
, doclose
);
3542 static int create_array(struct supertype
*st
, int dev_idx
)
3545 struct imsm_update_create_array
*u
;
3546 struct intel_super
*super
= st
->sb
;
3547 struct imsm_dev
*dev
= get_imsm_dev(super
, dev_idx
);
3548 struct imsm_map
*map
= get_imsm_map(dev
, 0);
3549 struct disk_info
*inf
;
3550 struct imsm_disk
*disk
;
3553 len
= sizeof(*u
) - sizeof(*dev
) + sizeof_imsm_dev(dev
, 0) +
3554 sizeof(*inf
) * map
->num_members
;
3557 fprintf(stderr
, "%s: failed to allocate update buffer\n",
3562 u
->type
= update_create_array
;
3563 u
->dev_idx
= dev_idx
;
3564 imsm_copy_dev(&u
->dev
, dev
);
3565 inf
= get_disk_info(u
);
3566 for (i
= 0; i
< map
->num_members
; i
++) {
3567 int idx
= get_imsm_disk_idx(dev
, i
);
3569 disk
= get_imsm_disk(super
, idx
);
3570 serialcpy(inf
[i
].serial
, disk
->serial
);
3572 append_metadata_update(st
, u
, len
);
3577 static int _add_disk(struct supertype
*st
)
3579 struct intel_super
*super
= st
->sb
;
3581 struct imsm_update_add_disk
*u
;
3589 fprintf(stderr
, "%s: failed to allocate update buffer\n",
3594 u
->type
= update_add_disk
;
3595 append_metadata_update(st
, u
, len
);
3600 static int write_init_super_imsm(struct supertype
*st
)
3602 struct intel_super
*super
= st
->sb
;
3603 int current_vol
= super
->current_vol
;
3605 /* we are done with current_vol reset it to point st at the container */
3606 super
->current_vol
= -1;
3608 if (st
->update_tail
) {
3609 /* queue the recently created array / added disk
3610 * as a metadata update */
3614 /* determine if we are creating a volume or adding a disk */
3615 if (current_vol
< 0) {
3616 /* in the add disk case we are running in mdmon
3617 * context, so don't close fd's
3619 return _add_disk(st
);
3621 rv
= create_array(st
, current_vol
);
3623 for (d
= super
->disks
; d
; d
= d
->next
) {
3631 for (d
= super
->disks
; d
; d
= d
->next
)
3632 Kill(d
->devname
, NULL
, 0, 1, 1);
3633 return write_super_imsm(st
->sb
, 1);
3638 static int store_super_imsm(struct supertype
*st
, int fd
)
3640 struct intel_super
*super
= st
->sb
;
3641 struct imsm_super
*mpb
= super
? super
->anchor
: NULL
;
3647 return store_imsm_mpb(fd
, mpb
);
3653 static int imsm_bbm_log_size(struct imsm_super
*mpb
)
3655 return __le32_to_cpu(mpb
->bbm_log_size
);
3659 static int validate_geometry_imsm_container(struct supertype
*st
, int level
,
3660 int layout
, int raiddisks
, int chunk
,
3661 unsigned long long size
, char *dev
,
3662 unsigned long long *freesize
,
3666 unsigned long long ldsize
;
3667 const struct imsm_orom
*orom
;
3669 if (level
!= LEVEL_CONTAINER
)
3674 if (check_env("IMSM_NO_PLATFORM"))
3677 orom
= find_imsm_orom();
3678 if (orom
&& raiddisks
> orom
->tds
) {
3680 fprintf(stderr
, Name
": %d exceeds maximum number of"
3681 " platform supported disks: %d\n",
3682 raiddisks
, orom
->tds
);
3686 fd
= open(dev
, O_RDONLY
|O_EXCL
, 0);
3689 fprintf(stderr
, Name
": imsm: Cannot open %s: %s\n",
3690 dev
, strerror(errno
));
3693 if (!get_dev_size(fd
, dev
, &ldsize
)) {
3699 *freesize
= avail_size_imsm(st
, ldsize
>> 9);
3704 static unsigned long long find_size(struct extent
*e
, int *idx
, int num_extents
)
3706 const unsigned long long base_start
= e
[*idx
].start
;
3707 unsigned long long end
= base_start
+ e
[*idx
].size
;
3710 if (base_start
== end
)
3714 for (i
= *idx
; i
< num_extents
; i
++) {
3715 /* extend overlapping extents */
3716 if (e
[i
].start
>= base_start
&&
3717 e
[i
].start
<= end
) {
3720 if (e
[i
].start
+ e
[i
].size
> end
)
3721 end
= e
[i
].start
+ e
[i
].size
;
3722 } else if (e
[i
].start
> end
) {
3728 return end
- base_start
;
3731 static unsigned long long merge_extents(struct intel_super
*super
, int sum_extents
)
3733 /* build a composite disk with all known extents and generate a new
3734 * 'maxsize' given the "all disks in an array must share a common start
3735 * offset" constraint
3737 struct extent
*e
= calloc(sum_extents
, sizeof(*e
));
3741 unsigned long long pos
;
3742 unsigned long long start
= 0;
3743 unsigned long long maxsize
;
3744 unsigned long reserve
;
3749 /* coalesce and sort all extents. also, check to see if we need to
3750 * reserve space between member arrays
3753 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
3756 for (i
= 0; i
< dl
->extent_cnt
; i
++)
3759 qsort(e
, sum_extents
, sizeof(*e
), cmp_extent
);
3764 while (i
< sum_extents
) {
3765 e
[j
].start
= e
[i
].start
;
3766 e
[j
].size
= find_size(e
, &i
, sum_extents
);
3768 if (e
[j
-1].size
== 0)
3777 unsigned long long esize
;
3779 esize
= e
[i
].start
- pos
;
3780 if (esize
>= maxsize
) {
3785 pos
= e
[i
].start
+ e
[i
].size
;
3787 } while (e
[i
-1].size
);
3793 /* FIXME assumes volume at offset 0 is the first volume in a
3796 if (start_extent
> 0)
3797 reserve
= IMSM_RESERVED_SECTORS
; /* gap between raid regions */
3801 if (maxsize
< reserve
)
3804 super
->create_offset
= ~((__u32
) 0);
3805 if (start
+ reserve
> super
->create_offset
)
3806 return 0; /* start overflows create_offset */
3807 super
->create_offset
= start
+ reserve
;
3809 return maxsize
- reserve
;
3812 static int is_raid_level_supported(const struct imsm_orom
*orom
, int level
, int raiddisks
)
3814 if (level
< 0 || level
== 6 || level
== 4)
3817 /* if we have an orom prevent invalid raid levels */
3820 case 0: return imsm_orom_has_raid0(orom
);
3823 return imsm_orom_has_raid1e(orom
);
3824 return imsm_orom_has_raid1(orom
) && raiddisks
== 2;
3825 case 10: return imsm_orom_has_raid10(orom
) && raiddisks
== 4;
3826 case 5: return imsm_orom_has_raid5(orom
) && raiddisks
> 2;
3829 return 1; /* not on an Intel RAID platform so anything goes */
3834 #define pr_vrb(fmt, arg...) (void) (verbose && fprintf(stderr, Name fmt, ##arg))
3836 validate_geometry_imsm_orom(struct intel_super
*super
, int level
, int layout
,
3837 int raiddisks
, int chunk
, int verbose
)
3839 if (!is_raid_level_supported(super
->orom
, level
, raiddisks
)) {
3840 pr_vrb(": platform does not support raid%d with %d disk%s\n",
3841 level
, raiddisks
, raiddisks
> 1 ? "s" : "");
3844 if (super
->orom
&& level
!= 1 &&
3845 !imsm_orom_has_chunk(super
->orom
, chunk
)) {
3846 pr_vrb(": platform does not support a chunk size of: %d\n", chunk
);
3849 if (layout
!= imsm_level_to_layout(level
)) {
3851 pr_vrb(": imsm raid 5 only supports the left-asymmetric layout\n");
3852 else if (level
== 10)
3853 pr_vrb(": imsm raid 10 only supports the n2 layout\n");
3855 pr_vrb(": imsm unknown layout %#x for this raid level %d\n",
3863 /* validate_geometry_imsm_volume - lifted from validate_geometry_ddf_bvd
3864 * FIX ME add ahci details
3866 static int validate_geometry_imsm_volume(struct supertype
*st
, int level
,
3867 int layout
, int raiddisks
, int chunk
,
3868 unsigned long long size
, char *dev
,
3869 unsigned long long *freesize
,
3873 struct intel_super
*super
= st
->sb
;
3874 struct imsm_super
*mpb
= super
->anchor
;
3876 unsigned long long pos
= 0;
3877 unsigned long long maxsize
;
3881 /* We must have the container info already read in. */
3885 if (!validate_geometry_imsm_orom(super
, level
, layout
, raiddisks
, chunk
, verbose
))
3889 /* General test: make sure there is space for
3890 * 'raiddisks' device extents of size 'size' at a given
3893 unsigned long long minsize
= size
;
3894 unsigned long long start_offset
= MaxSector
;
3897 minsize
= MPB_SECTOR_CNT
+ IMSM_RESERVED_SECTORS
;
3898 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
3903 e
= get_extents(super
, dl
);
3906 unsigned long long esize
;
3907 esize
= e
[i
].start
- pos
;
3908 if (esize
>= minsize
)
3910 if (found
&& start_offset
== MaxSector
) {
3913 } else if (found
&& pos
!= start_offset
) {
3917 pos
= e
[i
].start
+ e
[i
].size
;
3919 } while (e
[i
-1].size
);
3924 if (dcnt
< raiddisks
) {
3926 fprintf(stderr
, Name
": imsm: Not enough "
3927 "devices with space for this array "
3935 /* This device must be a member of the set */
3936 if (stat(dev
, &stb
) < 0)
3938 if ((S_IFMT
& stb
.st_mode
) != S_IFBLK
)
3940 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
3941 if (dl
->major
== (int)major(stb
.st_rdev
) &&
3942 dl
->minor
== (int)minor(stb
.st_rdev
))
3947 fprintf(stderr
, Name
": %s is not in the "
3948 "same imsm set\n", dev
);
3950 } else if (super
->orom
&& dl
->index
< 0 && mpb
->num_raid_devs
) {
3951 /* If a volume is present then the current creation attempt
3952 * cannot incorporate new spares because the orom may not
3953 * understand this configuration (all member disks must be
3954 * members of each array in the container).
3956 fprintf(stderr
, Name
": %s is a spare and a volume"
3957 " is already defined for this container\n", dev
);
3958 fprintf(stderr
, Name
": The option-rom requires all member"
3959 " disks to be a member of all volumes\n");
3963 /* retrieve the largest free space block */
3964 e
= get_extents(super
, dl
);
3969 unsigned long long esize
;
3971 esize
= e
[i
].start
- pos
;
3972 if (esize
>= maxsize
)
3974 pos
= e
[i
].start
+ e
[i
].size
;
3976 } while (e
[i
-1].size
);
3981 fprintf(stderr
, Name
": unable to determine free space for: %s\n",
3985 if (maxsize
< size
) {
3987 fprintf(stderr
, Name
": %s not enough space (%llu < %llu)\n",
3988 dev
, maxsize
, size
);
3992 /* count total number of extents for merge */
3994 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
3996 i
+= dl
->extent_cnt
;
3998 maxsize
= merge_extents(super
, i
);
3999 if (maxsize
< size
|| maxsize
== 0) {
4001 fprintf(stderr
, Name
": not enough space after merge (%llu < %llu)\n",
4006 *freesize
= maxsize
;
4011 static int reserve_space(struct supertype
*st
, int raiddisks
,
4012 unsigned long long size
, int chunk
,
4013 unsigned long long *freesize
)
4015 struct intel_super
*super
= st
->sb
;
4016 struct imsm_super
*mpb
= super
->anchor
;
4021 unsigned long long maxsize
;
4022 unsigned long long minsize
;
4026 /* find the largest common start free region of the possible disks */
4030 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
4036 /* don't activate new spares if we are orom constrained
4037 * and there is already a volume active in the container
4039 if (super
->orom
&& dl
->index
< 0 && mpb
->num_raid_devs
)
4042 e
= get_extents(super
, dl
);
4045 for (i
= 1; e
[i
-1].size
; i
++)
4053 maxsize
= merge_extents(super
, extent_cnt
);
4058 if (cnt
< raiddisks
||
4059 (super
->orom
&& used
&& used
!= raiddisks
) ||
4060 maxsize
< minsize
||
4062 fprintf(stderr
, Name
": not enough devices with space to create array.\n");
4063 return 0; /* No enough free spaces large enough */
4075 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
4077 dl
->raiddisk
= cnt
++;
4084 static int validate_geometry_imsm(struct supertype
*st
, int level
, int layout
,
4085 int raiddisks
, int chunk
, unsigned long long size
,
4086 char *dev
, unsigned long long *freesize
,
4093 /* if given unused devices create a container
4094 * if given given devices in a container create a member volume
4096 if (level
== LEVEL_CONTAINER
) {
4097 /* Must be a fresh device to add to a container */
4098 return validate_geometry_imsm_container(st
, level
, layout
,
4099 raiddisks
, chunk
, size
,
4105 if (st
->sb
&& freesize
) {
4106 /* we are being asked to automatically layout a
4107 * new volume based on the current contents of
4108 * the container. If the the parameters can be
4109 * satisfied reserve_space will record the disks,
4110 * start offset, and size of the volume to be
4111 * created. add_to_super and getinfo_super
4112 * detect when autolayout is in progress.
4114 if (!validate_geometry_imsm_orom(st
->sb
, level
, layout
,
4118 return reserve_space(st
, raiddisks
, size
, chunk
, freesize
);
4123 /* creating in a given container */
4124 return validate_geometry_imsm_volume(st
, level
, layout
,
4125 raiddisks
, chunk
, size
,
4126 dev
, freesize
, verbose
);
4129 /* This device needs to be a device in an 'imsm' container */
4130 fd
= open(dev
, O_RDONLY
|O_EXCL
, 0);
4134 Name
": Cannot create this array on device %s\n",
4139 if (errno
!= EBUSY
|| (fd
= open(dev
, O_RDONLY
, 0)) < 0) {
4141 fprintf(stderr
, Name
": Cannot open %s: %s\n",
4142 dev
, strerror(errno
));
4145 /* Well, it is in use by someone, maybe an 'imsm' container. */
4146 cfd
= open_container(fd
);
4150 fprintf(stderr
, Name
": Cannot use %s: It is busy\n",
4154 sra
= sysfs_read(cfd
, 0, GET_VERSION
);
4155 if (sra
&& sra
->array
.major_version
== -1 &&
4156 strcmp(sra
->text_version
, "imsm") == 0)
4160 /* This is a member of a imsm container. Load the container
4161 * and try to create a volume
4163 struct intel_super
*super
;
4165 if (load_super_imsm_all(st
, cfd
, (void **) &super
, NULL
) == 0) {
4167 st
->container_dev
= fd2devnum(cfd
);
4169 return validate_geometry_imsm_volume(st
, level
, layout
,
4177 fprintf(stderr
, Name
": failed container membership check\n");
4183 static int default_chunk_imsm(struct supertype
*st
)
4185 struct intel_super
*super
= st
->sb
;
4190 return imsm_orom_default_chunk(super
->orom
);
4193 static void handle_missing(struct intel_super
*super
, struct imsm_dev
*dev
);
4195 static int kill_subarray_imsm(struct supertype
*st
)
4197 /* remove the subarray currently referenced by ->current_vol */
4199 struct intel_dev
**dp
;
4200 struct intel_super
*super
= st
->sb
;
4201 __u8 current_vol
= super
->current_vol
;
4202 struct imsm_super
*mpb
= super
->anchor
;
4204 if (super
->current_vol
< 0)
4206 super
->current_vol
= -1; /* invalidate subarray cursor */
4208 /* block deletions that would change the uuid of active subarrays
4210 * FIXME when immutable ids are available, but note that we'll
4211 * also need to fixup the invalidated/active subarray indexes in
4214 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
4217 if (i
< current_vol
)
4219 sprintf(subarray
, "%u", i
);
4220 if (is_subarray_active(subarray
, st
->devname
)) {
4222 Name
": deleting subarray-%d would change the UUID of active subarray-%d, aborting\n",
4229 if (st
->update_tail
) {
4230 struct imsm_update_kill_array
*u
= malloc(sizeof(*u
));
4234 u
->type
= update_kill_array
;
4235 u
->dev_idx
= current_vol
;
4236 append_metadata_update(st
, u
, sizeof(*u
));
4241 for (dp
= &super
->devlist
; *dp
;)
4242 if ((*dp
)->index
== current_vol
) {
4245 handle_missing(super
, (*dp
)->dev
);
4246 if ((*dp
)->index
> current_vol
)
4251 /* no more raid devices, all active components are now spares,
4252 * but of course failed are still failed
4254 if (--mpb
->num_raid_devs
== 0) {
4257 for (d
= super
->disks
; d
; d
= d
->next
)
4258 if (d
->index
> -2) {
4260 d
->disk
.status
= SPARE_DISK
;
4264 super
->updates_pending
++;
4269 static int update_subarray_imsm(struct supertype
*st
, char *subarray
,
4270 char *update
, struct mddev_ident
*ident
)
4272 /* update the subarray currently referenced by ->current_vol */
4273 struct intel_super
*super
= st
->sb
;
4274 struct imsm_super
*mpb
= super
->anchor
;
4276 if (strcmp(update
, "name") == 0) {
4277 char *name
= ident
->name
;
4281 if (is_subarray_active(subarray
, st
->devname
)) {
4283 Name
": Unable to update name of active subarray\n");
4287 if (!check_name(super
, name
, 0))
4290 vol
= strtoul(subarray
, &ep
, 10);
4291 if (*ep
!= '\0' || vol
>= super
->anchor
->num_raid_devs
)
4294 if (st
->update_tail
) {
4295 struct imsm_update_rename_array
*u
= malloc(sizeof(*u
));
4299 u
->type
= update_rename_array
;
4301 snprintf((char *) u
->name
, MAX_RAID_SERIAL_LEN
, "%s", name
);
4302 append_metadata_update(st
, u
, sizeof(*u
));
4304 struct imsm_dev
*dev
;
4307 dev
= get_imsm_dev(super
, vol
);
4308 snprintf((char *) dev
->volume
, MAX_RAID_SERIAL_LEN
, "%s", name
);
4309 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
4310 dev
= get_imsm_dev(super
, i
);
4311 handle_missing(super
, dev
);
4313 super
->updates_pending
++;
4320 #endif /* MDASSEMBLE */
4322 static int is_rebuilding(struct imsm_dev
*dev
)
4324 struct imsm_map
*migr_map
;
4326 if (!dev
->vol
.migr_state
)
4329 if (migr_type(dev
) != MIGR_REBUILD
)
4332 migr_map
= get_imsm_map(dev
, 1);
4334 if (migr_map
->map_state
== IMSM_T_STATE_DEGRADED
)
4340 static void update_recovery_start(struct imsm_dev
*dev
, struct mdinfo
*array
)
4342 struct mdinfo
*rebuild
= NULL
;
4346 if (!is_rebuilding(dev
))
4349 /* Find the rebuild target, but punt on the dual rebuild case */
4350 for (d
= array
->devs
; d
; d
= d
->next
)
4351 if (d
->recovery_start
== 0) {
4358 /* (?) none of the disks are marked with
4359 * IMSM_ORD_REBUILD, so assume they are missing and the
4360 * disk_ord_tbl was not correctly updated
4362 dprintf("%s: failed to locate out-of-sync disk\n", __func__
);
4366 units
= __le32_to_cpu(dev
->vol
.curr_migr_unit
);
4367 rebuild
->recovery_start
= units
* blocks_per_migr_unit(dev
);
4371 static struct mdinfo
*container_content_imsm(struct supertype
*st
, char *subarray
)
4373 /* Given a container loaded by load_super_imsm_all,
4374 * extract information about all the arrays into
4376 * If 'subarray' is given, just extract info about that array.
4378 * For each imsm_dev create an mdinfo, fill it in,
4379 * then look for matching devices in super->disks
4380 * and create appropriate device mdinfo.
4382 struct intel_super
*super
= st
->sb
;
4383 struct imsm_super
*mpb
= super
->anchor
;
4384 struct mdinfo
*rest
= NULL
;
4387 /* do not assemble arrays that might have bad blocks */
4388 if (imsm_bbm_log_size(super
->anchor
)) {
4389 fprintf(stderr
, Name
": BBM log found in metadata. "
4390 "Cannot activate array(s).\n");
4394 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
4395 struct imsm_dev
*dev
;
4396 struct imsm_map
*map
;
4397 struct mdinfo
*this;
4402 (i
!= strtoul(subarray
, &ep
, 10) || *ep
!= '\0'))
4405 dev
= get_imsm_dev(super
, i
);
4406 map
= get_imsm_map(dev
, 0);
4408 /* do not publish arrays that are in the middle of an
4409 * unsupported migration
4411 if (dev
->vol
.migr_state
&&
4412 (migr_type(dev
) == MIGR_GEN_MIGR
||
4413 migr_type(dev
) == MIGR_STATE_CHANGE
)) {
4414 fprintf(stderr
, Name
": cannot assemble volume '%.16s':"
4415 " unsupported migration in progress\n",
4420 this = malloc(sizeof(*this));
4422 fprintf(stderr
, Name
": failed to allocate %zu bytes\n",
4426 memset(this, 0, sizeof(*this));
4429 super
->current_vol
= i
;
4430 getinfo_super_imsm_volume(st
, this, NULL
);
4431 for (slot
= 0 ; slot
< map
->num_members
; slot
++) {
4432 unsigned long long recovery_start
;
4433 struct mdinfo
*info_d
;
4440 idx
= get_imsm_disk_idx(dev
, slot
);
4441 ord
= get_imsm_ord_tbl_ent(dev
, slot
);
4442 for (d
= super
->disks
; d
; d
= d
->next
)
4443 if (d
->index
== idx
)
4446 recovery_start
= MaxSector
;
4449 if (d
&& is_failed(&d
->disk
))
4451 if (ord
& IMSM_ORD_REBUILD
)
4455 * if we skip some disks the array will be assmebled degraded;
4456 * reset resync start to avoid a dirty-degraded
4457 * situation when performing the intial sync
4459 * FIXME handle dirty degraded
4461 if ((skip
|| recovery_start
== 0) && !dev
->vol
.dirty
)
4462 this->resync_start
= MaxSector
;
4466 info_d
= calloc(1, sizeof(*info_d
));
4468 fprintf(stderr
, Name
": failed to allocate disk"
4469 " for volume %.16s\n", dev
->volume
);
4470 info_d
= this->devs
;
4472 struct mdinfo
*d
= info_d
->next
;
4481 info_d
->next
= this->devs
;
4482 this->devs
= info_d
;
4484 info_d
->disk
.number
= d
->index
;
4485 info_d
->disk
.major
= d
->major
;
4486 info_d
->disk
.minor
= d
->minor
;
4487 info_d
->disk
.raid_disk
= slot
;
4488 info_d
->recovery_start
= recovery_start
;
4490 if (info_d
->recovery_start
== MaxSector
)
4491 this->array
.working_disks
++;
4493 info_d
->events
= __le32_to_cpu(mpb
->generation_num
);
4494 info_d
->data_offset
= __le32_to_cpu(map
->pba_of_lba0
);
4495 info_d
->component_size
= __le32_to_cpu(map
->blocks_per_member
);
4497 /* now that the disk list is up-to-date fixup recovery_start */
4498 update_recovery_start(dev
, this);
4506 static __u8
imsm_check_degraded(struct intel_super
*super
, struct imsm_dev
*dev
, int failed
)
4508 struct imsm_map
*map
= get_imsm_map(dev
, 0);
4511 return map
->map_state
== IMSM_T_STATE_UNINITIALIZED
?
4512 IMSM_T_STATE_UNINITIALIZED
: IMSM_T_STATE_NORMAL
;
4514 switch (get_imsm_raid_level(map
)) {
4516 return IMSM_T_STATE_FAILED
;
4519 if (failed
< map
->num_members
)
4520 return IMSM_T_STATE_DEGRADED
;
4522 return IMSM_T_STATE_FAILED
;
4527 * check to see if any mirrors have failed, otherwise we
4528 * are degraded. Even numbered slots are mirrored on
4532 /* gcc -Os complains that this is unused */
4533 int insync
= insync
;
4535 for (i
= 0; i
< map
->num_members
; i
++) {
4536 __u32 ord
= get_imsm_ord_tbl_ent(dev
, i
);
4537 int idx
= ord_to_idx(ord
);
4538 struct imsm_disk
*disk
;
4540 /* reset the potential in-sync count on even-numbered
4541 * slots. num_copies is always 2 for imsm raid10
4546 disk
= get_imsm_disk(super
, idx
);
4547 if (!disk
|| is_failed(disk
) || ord
& IMSM_ORD_REBUILD
)
4550 /* no in-sync disks left in this mirror the
4554 return IMSM_T_STATE_FAILED
;
4557 return IMSM_T_STATE_DEGRADED
;
4561 return IMSM_T_STATE_DEGRADED
;
4563 return IMSM_T_STATE_FAILED
;
4569 return map
->map_state
;
4572 static int imsm_count_failed(struct intel_super
*super
, struct imsm_dev
*dev
)
4576 struct imsm_disk
*disk
;
4577 struct imsm_map
*map
= get_imsm_map(dev
, 0);
4578 struct imsm_map
*prev
= get_imsm_map(dev
, dev
->vol
.migr_state
);
4582 /* at the beginning of migration we set IMSM_ORD_REBUILD on
4583 * disks that are being rebuilt. New failures are recorded to
4584 * map[0]. So we look through all the disks we started with and
4585 * see if any failures are still present, or if any new ones
4588 * FIXME add support for online capacity expansion and
4589 * raid-level-migration
4591 for (i
= 0; i
< prev
->num_members
; i
++) {
4592 ord
= __le32_to_cpu(prev
->disk_ord_tbl
[i
]);
4593 ord
|= __le32_to_cpu(map
->disk_ord_tbl
[i
]);
4594 idx
= ord_to_idx(ord
);
4596 disk
= get_imsm_disk(super
, idx
);
4597 if (!disk
|| is_failed(disk
) || ord
& IMSM_ORD_REBUILD
)
4605 static int imsm_open_new(struct supertype
*c
, struct active_array
*a
,
4608 struct intel_super
*super
= c
->sb
;
4609 struct imsm_super
*mpb
= super
->anchor
;
4611 if (atoi(inst
) >= mpb
->num_raid_devs
) {
4612 fprintf(stderr
, "%s: subarry index %d, out of range\n",
4613 __func__
, atoi(inst
));
4617 dprintf("imsm: open_new %s\n", inst
);
4618 a
->info
.container_member
= atoi(inst
);
4622 static int is_resyncing(struct imsm_dev
*dev
)
4624 struct imsm_map
*migr_map
;
4626 if (!dev
->vol
.migr_state
)
4629 if (migr_type(dev
) == MIGR_INIT
||
4630 migr_type(dev
) == MIGR_REPAIR
)
4633 migr_map
= get_imsm_map(dev
, 1);
4635 if (migr_map
->map_state
== IMSM_T_STATE_NORMAL
)
4641 /* return true if we recorded new information */
4642 static int mark_failure(struct imsm_dev
*dev
, struct imsm_disk
*disk
, int idx
)
4646 struct imsm_map
*map
;
4648 /* new failures are always set in map[0] */
4649 map
= get_imsm_map(dev
, 0);
4651 slot
= get_imsm_disk_slot(map
, idx
);
4655 ord
= __le32_to_cpu(map
->disk_ord_tbl
[slot
]);
4656 if (is_failed(disk
) && (ord
& IMSM_ORD_REBUILD
))
4659 disk
->status
|= FAILED_DISK
;
4660 disk
->status
&= ~CONFIGURED_DISK
;
4661 set_imsm_ord_tbl_ent(map
, slot
, idx
| IMSM_ORD_REBUILD
);
4662 if (map
->failed_disk_num
== 0xff)
4663 map
->failed_disk_num
= slot
;
4667 static void mark_missing(struct imsm_dev
*dev
, struct imsm_disk
*disk
, int idx
)
4669 mark_failure(dev
, disk
, idx
);
4671 if (disk
->scsi_id
== __cpu_to_le32(~(__u32
)0))
4674 disk
->scsi_id
= __cpu_to_le32(~(__u32
)0);
4675 memmove(&disk
->serial
[0], &disk
->serial
[1], MAX_RAID_SERIAL_LEN
- 1);
4678 static void handle_missing(struct intel_super
*super
, struct imsm_dev
*dev
)
4684 if (!super
->missing
)
4686 failed
= imsm_count_failed(super
, dev
);
4687 map_state
= imsm_check_degraded(super
, dev
, failed
);
4689 dprintf("imsm: mark missing\n");
4690 end_migration(dev
, map_state
);
4691 for (dl
= super
->missing
; dl
; dl
= dl
->next
)
4692 mark_missing(dev
, &dl
->disk
, dl
->index
);
4693 super
->updates_pending
++;
4696 /* Handle dirty -> clean transititions and resync. Degraded and rebuild
4697 * states are handled in imsm_set_disk() with one exception, when a
4698 * resync is stopped due to a new failure this routine will set the
4699 * 'degraded' state for the array.
4701 static int imsm_set_array_state(struct active_array
*a
, int consistent
)
4703 int inst
= a
->info
.container_member
;
4704 struct intel_super
*super
= a
->container
->sb
;
4705 struct imsm_dev
*dev
= get_imsm_dev(super
, inst
);
4706 struct imsm_map
*map
= get_imsm_map(dev
, 0);
4707 int failed
= imsm_count_failed(super
, dev
);
4708 __u8 map_state
= imsm_check_degraded(super
, dev
, failed
);
4709 __u32 blocks_per_unit
;
4711 /* before we activate this array handle any missing disks */
4712 if (consistent
== 2)
4713 handle_missing(super
, dev
);
4715 if (consistent
== 2 &&
4716 (!is_resync_complete(&a
->info
) ||
4717 map_state
!= IMSM_T_STATE_NORMAL
||
4718 dev
->vol
.migr_state
))
4721 if (is_resync_complete(&a
->info
)) {
4722 /* complete intialization / resync,
4723 * recovery and interrupted recovery is completed in
4726 if (is_resyncing(dev
)) {
4727 dprintf("imsm: mark resync done\n");
4728 end_migration(dev
, map_state
);
4729 super
->updates_pending
++;
4730 a
->last_checkpoint
= 0;
4732 } else if (!is_resyncing(dev
) && !failed
) {
4733 /* mark the start of the init process if nothing is failed */
4734 dprintf("imsm: mark resync start\n");
4735 if (map
->map_state
== IMSM_T_STATE_UNINITIALIZED
)
4736 migrate(dev
, IMSM_T_STATE_NORMAL
, MIGR_INIT
);
4738 migrate(dev
, IMSM_T_STATE_NORMAL
, MIGR_REPAIR
);
4739 super
->updates_pending
++;
4742 /* check if we can update curr_migr_unit from resync_start, recovery_start */
4743 blocks_per_unit
= blocks_per_migr_unit(dev
);
4744 if (blocks_per_unit
) {
4748 units
= a
->last_checkpoint
/ blocks_per_unit
;
4751 /* check that we did not overflow 32-bits, and that
4752 * curr_migr_unit needs updating
4754 if (units32
== units
&&
4755 __le32_to_cpu(dev
->vol
.curr_migr_unit
) != units32
) {
4756 dprintf("imsm: mark checkpoint (%u)\n", units32
);
4757 dev
->vol
.curr_migr_unit
= __cpu_to_le32(units32
);
4758 super
->updates_pending
++;
4762 /* mark dirty / clean */
4763 if (dev
->vol
.dirty
!= !consistent
) {
4764 dprintf("imsm: mark '%s'\n", consistent
? "clean" : "dirty");
4769 super
->updates_pending
++;
4774 static void imsm_set_disk(struct active_array
*a
, int n
, int state
)
4776 int inst
= a
->info
.container_member
;
4777 struct intel_super
*super
= a
->container
->sb
;
4778 struct imsm_dev
*dev
= get_imsm_dev(super
, inst
);
4779 struct imsm_map
*map
= get_imsm_map(dev
, 0);
4780 struct imsm_disk
*disk
;
4785 if (n
> map
->num_members
)
4786 fprintf(stderr
, "imsm: set_disk %d out of range 0..%d\n",
4787 n
, map
->num_members
- 1);
4792 dprintf("imsm: set_disk %d:%x\n", n
, state
);
4794 ord
= get_imsm_ord_tbl_ent(dev
, n
);
4795 disk
= get_imsm_disk(super
, ord_to_idx(ord
));
4797 /* check for new failures */
4798 if (state
& DS_FAULTY
) {
4799 if (mark_failure(dev
, disk
, ord_to_idx(ord
)))
4800 super
->updates_pending
++;
4803 /* check if in_sync */
4804 if (state
& DS_INSYNC
&& ord
& IMSM_ORD_REBUILD
&& is_rebuilding(dev
)) {
4805 struct imsm_map
*migr_map
= get_imsm_map(dev
, 1);
4807 set_imsm_ord_tbl_ent(migr_map
, n
, ord_to_idx(ord
));
4808 super
->updates_pending
++;
4811 failed
= imsm_count_failed(super
, dev
);
4812 map_state
= imsm_check_degraded(super
, dev
, failed
);
4814 /* check if recovery complete, newly degraded, or failed */
4815 if (map_state
== IMSM_T_STATE_NORMAL
&& is_rebuilding(dev
)) {
4816 end_migration(dev
, map_state
);
4817 map
= get_imsm_map(dev
, 0);
4818 map
->failed_disk_num
= ~0;
4819 super
->updates_pending
++;
4820 a
->last_checkpoint
= 0;
4821 } else if (map_state
== IMSM_T_STATE_DEGRADED
&&
4822 map
->map_state
!= map_state
&&
4823 !dev
->vol
.migr_state
) {
4824 dprintf("imsm: mark degraded\n");
4825 map
->map_state
= map_state
;
4826 super
->updates_pending
++;
4827 a
->last_checkpoint
= 0;
4828 } else if (map_state
== IMSM_T_STATE_FAILED
&&
4829 map
->map_state
!= map_state
) {
4830 dprintf("imsm: mark failed\n");
4831 end_migration(dev
, map_state
);
4832 super
->updates_pending
++;
4833 a
->last_checkpoint
= 0;
4837 static int store_imsm_mpb(int fd
, struct imsm_super
*mpb
)
4840 __u32 mpb_size
= __le32_to_cpu(mpb
->mpb_size
);
4841 unsigned long long dsize
;
4842 unsigned long long sectors
;
4844 get_dev_size(fd
, NULL
, &dsize
);
4846 if (mpb_size
> 512) {
4847 /* -1 to account for anchor */
4848 sectors
= mpb_sectors(mpb
) - 1;
4850 /* write the extended mpb to the sectors preceeding the anchor */
4851 if (lseek64(fd
, dsize
- (512 * (2 + sectors
)), SEEK_SET
) < 0)
4854 if ((unsigned long long)write(fd
, buf
+ 512, 512 * sectors
)
4859 /* first block is stored on second to last sector of the disk */
4860 if (lseek64(fd
, dsize
- (512 * 2), SEEK_SET
) < 0)
4863 if (write(fd
, buf
, 512) != 512)
4869 static void imsm_sync_metadata(struct supertype
*container
)
4871 struct intel_super
*super
= container
->sb
;
4873 if (!super
->updates_pending
)
4876 write_super_imsm(super
, 0);
4878 super
->updates_pending
= 0;
4881 static struct dl
*imsm_readd(struct intel_super
*super
, int idx
, struct active_array
*a
)
4883 struct imsm_dev
*dev
= get_imsm_dev(super
, a
->info
.container_member
);
4884 int i
= get_imsm_disk_idx(dev
, idx
);
4887 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
4891 if (dl
&& is_failed(&dl
->disk
))
4895 dprintf("%s: found %x:%x\n", __func__
, dl
->major
, dl
->minor
);
4900 static struct dl
*imsm_add_spare(struct intel_super
*super
, int slot
,
4901 struct active_array
*a
, int activate_new
)
4903 struct imsm_dev
*dev
= get_imsm_dev(super
, a
->info
.container_member
);
4904 int idx
= get_imsm_disk_idx(dev
, slot
);
4905 struct imsm_super
*mpb
= super
->anchor
;
4906 struct imsm_map
*map
;
4907 unsigned long long pos
;
4912 __u32 array_start
= 0;
4913 __u32 array_end
= 0;
4916 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
4917 /* If in this array, skip */
4918 for (d
= a
->info
.devs
; d
; d
= d
->next
)
4919 if (d
->state_fd
>= 0 &&
4920 d
->disk
.major
== dl
->major
&&
4921 d
->disk
.minor
== dl
->minor
) {
4922 dprintf("%x:%x already in array\n", dl
->major
, dl
->minor
);
4928 /* skip in use or failed drives */
4929 if (is_failed(&dl
->disk
) || idx
== dl
->index
||
4931 dprintf("%x:%x status (failed: %d index: %d)\n",
4932 dl
->major
, dl
->minor
, is_failed(&dl
->disk
), idx
);
4936 /* skip pure spares when we are looking for partially
4937 * assimilated drives
4939 if (dl
->index
== -1 && !activate_new
)
4942 /* Does this unused device have the requisite free space?
4943 * It needs to be able to cover all member volumes
4945 ex
= get_extents(super
, dl
);
4947 dprintf("cannot get extents\n");
4950 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
4951 dev
= get_imsm_dev(super
, i
);
4952 map
= get_imsm_map(dev
, 0);
4954 /* check if this disk is already a member of
4957 if (get_imsm_disk_slot(map
, dl
->index
) >= 0)
4963 array_start
= __le32_to_cpu(map
->pba_of_lba0
);
4964 array_end
= array_start
+
4965 __le32_to_cpu(map
->blocks_per_member
) - 1;
4968 /* check that we can start at pba_of_lba0 with
4969 * blocks_per_member of space
4971 if (array_start
>= pos
&& array_end
< ex
[j
].start
) {
4975 pos
= ex
[j
].start
+ ex
[j
].size
;
4977 } while (ex
[j
-1].size
);
4984 if (i
< mpb
->num_raid_devs
) {
4985 dprintf("%x:%x does not have %u to %u available\n",
4986 dl
->major
, dl
->minor
, array_start
, array_end
);
4996 static struct mdinfo
*imsm_activate_spare(struct active_array
*a
,
4997 struct metadata_update
**updates
)
5000 * Find a device with unused free space and use it to replace a
5001 * failed/vacant region in an array. We replace failed regions one a
5002 * array at a time. The result is that a new spare disk will be added
5003 * to the first failed array and after the monitor has finished
5004 * propagating failures the remainder will be consumed.
5006 * FIXME add a capability for mdmon to request spares from another
5010 struct intel_super
*super
= a
->container
->sb
;
5011 int inst
= a
->info
.container_member
;
5012 struct imsm_dev
*dev
= get_imsm_dev(super
, inst
);
5013 struct imsm_map
*map
= get_imsm_map(dev
, 0);
5014 int failed
= a
->info
.array
.raid_disks
;
5015 struct mdinfo
*rv
= NULL
;
5018 struct metadata_update
*mu
;
5020 struct imsm_update_activate_spare
*u
;
5024 for (d
= a
->info
.devs
; d
; d
= d
->next
) {
5025 if ((d
->curr_state
& DS_FAULTY
) &&
5027 /* wait for Removal to happen */
5029 if (d
->state_fd
>= 0)
5033 dprintf("imsm: activate spare: inst=%d failed=%d (%d) level=%d\n",
5034 inst
, failed
, a
->info
.array
.raid_disks
, a
->info
.array
.level
);
5035 if (imsm_check_degraded(super
, dev
, failed
) != IMSM_T_STATE_DEGRADED
)
5038 /* For each slot, if it is not working, find a spare */
5039 for (i
= 0; i
< a
->info
.array
.raid_disks
; i
++) {
5040 for (d
= a
->info
.devs
; d
; d
= d
->next
)
5041 if (d
->disk
.raid_disk
== i
)
5043 dprintf("found %d: %p %x\n", i
, d
, d
?d
->curr_state
:0);
5044 if (d
&& (d
->state_fd
>= 0))
5048 * OK, this device needs recovery. Try to re-add the
5049 * previous occupant of this slot, if this fails see if
5050 * we can continue the assimilation of a spare that was
5051 * partially assimilated, finally try to activate a new
5054 dl
= imsm_readd(super
, i
, a
);
5056 dl
= imsm_add_spare(super
, i
, a
, 0);
5058 dl
= imsm_add_spare(super
, i
, a
, 1);
5062 /* found a usable disk with enough space */
5063 di
= malloc(sizeof(*di
));
5066 memset(di
, 0, sizeof(*di
));
5068 /* dl->index will be -1 in the case we are activating a
5069 * pristine spare. imsm_process_update() will create a
5070 * new index in this case. Once a disk is found to be
5071 * failed in all member arrays it is kicked from the
5074 di
->disk
.number
= dl
->index
;
5076 /* (ab)use di->devs to store a pointer to the device
5079 di
->devs
= (struct mdinfo
*) dl
;
5081 di
->disk
.raid_disk
= i
;
5082 di
->disk
.major
= dl
->major
;
5083 di
->disk
.minor
= dl
->minor
;
5085 di
->recovery_start
= 0;
5086 di
->data_offset
= __le32_to_cpu(map
->pba_of_lba0
);
5087 di
->component_size
= a
->info
.component_size
;
5088 di
->container_member
= inst
;
5089 super
->random
= random32();
5093 dprintf("%x:%x to be %d at %llu\n", dl
->major
, dl
->minor
,
5094 i
, di
->data_offset
);
5100 /* No spares found */
5102 /* Now 'rv' has a list of devices to return.
5103 * Create a metadata_update record to update the
5104 * disk_ord_tbl for the array
5106 mu
= malloc(sizeof(*mu
));
5108 mu
->buf
= malloc(sizeof(struct imsm_update_activate_spare
) * num_spares
);
5109 if (mu
->buf
== NULL
) {
5116 struct mdinfo
*n
= rv
->next
;
5125 mu
->len
= sizeof(struct imsm_update_activate_spare
) * num_spares
;
5126 mu
->next
= *updates
;
5127 u
= (struct imsm_update_activate_spare
*) mu
->buf
;
5129 for (di
= rv
; di
; di
= di
->next
) {
5130 u
->type
= update_activate_spare
;
5131 u
->dl
= (struct dl
*) di
->devs
;
5133 u
->slot
= di
->disk
.raid_disk
;
5144 static int disks_overlap(struct intel_super
*super
, int idx
, struct imsm_update_create_array
*u
)
5146 struct imsm_dev
*dev
= get_imsm_dev(super
, idx
);
5147 struct imsm_map
*map
= get_imsm_map(dev
, 0);
5148 struct imsm_map
*new_map
= get_imsm_map(&u
->dev
, 0);
5149 struct disk_info
*inf
= get_disk_info(u
);
5150 struct imsm_disk
*disk
;
5154 for (i
= 0; i
< map
->num_members
; i
++) {
5155 disk
= get_imsm_disk(super
, get_imsm_disk_idx(dev
, i
));
5156 for (j
= 0; j
< new_map
->num_members
; j
++)
5157 if (serialcmp(disk
->serial
, inf
[j
].serial
) == 0)
5164 static void imsm_delete(struct intel_super
*super
, struct dl
**dlp
, unsigned index
);
5166 static void imsm_process_update(struct supertype
*st
,
5167 struct metadata_update
*update
)
5170 * crack open the metadata_update envelope to find the update record
5171 * update can be one of:
5172 * update_activate_spare - a spare device has replaced a failed
5173 * device in an array, update the disk_ord_tbl. If this disk is
5174 * present in all member arrays then also clear the SPARE_DISK
5177 struct intel_super
*super
= st
->sb
;
5178 struct imsm_super
*mpb
;
5179 enum imsm_update_type type
= *(enum imsm_update_type
*) update
->buf
;
5181 /* update requires a larger buf but the allocation failed */
5182 if (super
->next_len
&& !super
->next_buf
) {
5183 super
->next_len
= 0;
5187 if (super
->next_buf
) {
5188 memcpy(super
->next_buf
, super
->buf
, super
->len
);
5190 super
->len
= super
->next_len
;
5191 super
->buf
= super
->next_buf
;
5193 super
->next_len
= 0;
5194 super
->next_buf
= NULL
;
5197 mpb
= super
->anchor
;
5200 case update_activate_spare
: {
5201 struct imsm_update_activate_spare
*u
= (void *) update
->buf
;
5202 struct imsm_dev
*dev
= get_imsm_dev(super
, u
->array
);
5203 struct imsm_map
*map
= get_imsm_map(dev
, 0);
5204 struct imsm_map
*migr_map
;
5205 struct active_array
*a
;
5206 struct imsm_disk
*disk
;
5211 int victim
= get_imsm_disk_idx(dev
, u
->slot
);
5214 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
5219 fprintf(stderr
, "error: imsm_activate_spare passed "
5220 "an unknown disk (index: %d)\n",
5225 super
->updates_pending
++;
5227 /* count failures (excluding rebuilds and the victim)
5228 * to determine map[0] state
5231 for (i
= 0; i
< map
->num_members
; i
++) {
5234 disk
= get_imsm_disk(super
, get_imsm_disk_idx(dev
, i
));
5235 if (!disk
|| is_failed(disk
))
5239 /* adding a pristine spare, assign a new index */
5240 if (dl
->index
< 0) {
5241 dl
->index
= super
->anchor
->num_disks
;
5242 super
->anchor
->num_disks
++;
5245 disk
->status
|= CONFIGURED_DISK
;
5246 disk
->status
&= ~SPARE_DISK
;
5249 to_state
= imsm_check_degraded(super
, dev
, failed
);
5250 map
->map_state
= IMSM_T_STATE_DEGRADED
;
5251 migrate(dev
, to_state
, MIGR_REBUILD
);
5252 migr_map
= get_imsm_map(dev
, 1);
5253 set_imsm_ord_tbl_ent(map
, u
->slot
, dl
->index
);
5254 set_imsm_ord_tbl_ent(migr_map
, u
->slot
, dl
->index
| IMSM_ORD_REBUILD
);
5256 /* update the family_num to mark a new container
5257 * generation, being careful to record the existing
5258 * family_num in orig_family_num to clean up after
5259 * earlier mdadm versions that neglected to set it.
5261 if (mpb
->orig_family_num
== 0)
5262 mpb
->orig_family_num
= mpb
->family_num
;
5263 mpb
->family_num
+= super
->random
;
5265 /* count arrays using the victim in the metadata */
5267 for (a
= st
->arrays
; a
; a
= a
->next
) {
5268 dev
= get_imsm_dev(super
, a
->info
.container_member
);
5269 map
= get_imsm_map(dev
, 0);
5271 if (get_imsm_disk_slot(map
, victim
) >= 0)
5275 /* delete the victim if it is no longer being
5281 /* We know that 'manager' isn't touching anything,
5282 * so it is safe to delete
5284 for (dlp
= &super
->disks
; *dlp
; dlp
= &(*dlp
)->next
)
5285 if ((*dlp
)->index
== victim
)
5288 /* victim may be on the missing list */
5290 for (dlp
= &super
->missing
; *dlp
; dlp
= &(*dlp
)->next
)
5291 if ((*dlp
)->index
== victim
)
5293 imsm_delete(super
, dlp
, victim
);
5297 case update_create_array
: {
5298 /* someone wants to create a new array, we need to be aware of
5299 * a few races/collisions:
5300 * 1/ 'Create' called by two separate instances of mdadm
5301 * 2/ 'Create' versus 'activate_spare': mdadm has chosen
5302 * devices that have since been assimilated via
5304 * In the event this update can not be carried out mdadm will
5305 * (FIX ME) notice that its update did not take hold.
5307 struct imsm_update_create_array
*u
= (void *) update
->buf
;
5308 struct intel_dev
*dv
;
5309 struct imsm_dev
*dev
;
5310 struct imsm_map
*map
, *new_map
;
5311 unsigned long long start
, end
;
5312 unsigned long long new_start
, new_end
;
5314 struct disk_info
*inf
;
5317 /* handle racing creates: first come first serve */
5318 if (u
->dev_idx
< mpb
->num_raid_devs
) {
5319 dprintf("%s: subarray %d already defined\n",
5320 __func__
, u
->dev_idx
);
5324 /* check update is next in sequence */
5325 if (u
->dev_idx
!= mpb
->num_raid_devs
) {
5326 dprintf("%s: can not create array %d expected index %d\n",
5327 __func__
, u
->dev_idx
, mpb
->num_raid_devs
);
5331 new_map
= get_imsm_map(&u
->dev
, 0);
5332 new_start
= __le32_to_cpu(new_map
->pba_of_lba0
);
5333 new_end
= new_start
+ __le32_to_cpu(new_map
->blocks_per_member
);
5334 inf
= get_disk_info(u
);
5336 /* handle activate_spare versus create race:
5337 * check to make sure that overlapping arrays do not include
5340 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
5341 dev
= get_imsm_dev(super
, i
);
5342 map
= get_imsm_map(dev
, 0);
5343 start
= __le32_to_cpu(map
->pba_of_lba0
);
5344 end
= start
+ __le32_to_cpu(map
->blocks_per_member
);
5345 if ((new_start
>= start
&& new_start
<= end
) ||
5346 (start
>= new_start
&& start
<= new_end
))
5351 if (disks_overlap(super
, i
, u
)) {
5352 dprintf("%s: arrays overlap\n", __func__
);
5357 /* check that prepare update was successful */
5358 if (!update
->space
) {
5359 dprintf("%s: prepare update failed\n", __func__
);
5363 /* check that all disks are still active before committing
5364 * changes. FIXME: could we instead handle this by creating a
5365 * degraded array? That's probably not what the user expects,
5366 * so better to drop this update on the floor.
5368 for (i
= 0; i
< new_map
->num_members
; i
++) {
5369 dl
= serial_to_dl(inf
[i
].serial
, super
);
5371 dprintf("%s: disk disappeared\n", __func__
);
5376 super
->updates_pending
++;
5378 /* convert spares to members and fixup ord_tbl */
5379 for (i
= 0; i
< new_map
->num_members
; i
++) {
5380 dl
= serial_to_dl(inf
[i
].serial
, super
);
5381 if (dl
->index
== -1) {
5382 dl
->index
= mpb
->num_disks
;
5384 dl
->disk
.status
|= CONFIGURED_DISK
;
5385 dl
->disk
.status
&= ~SPARE_DISK
;
5387 set_imsm_ord_tbl_ent(new_map
, i
, dl
->index
);
5392 update
->space
= NULL
;
5393 imsm_copy_dev(dev
, &u
->dev
);
5394 dv
->index
= u
->dev_idx
;
5395 dv
->next
= super
->devlist
;
5396 super
->devlist
= dv
;
5397 mpb
->num_raid_devs
++;
5399 imsm_update_version_info(super
);
5402 /* mdmon knows how to release update->space, but not
5403 * ((struct intel_dev *) update->space)->dev
5405 if (update
->space
) {
5411 case update_kill_array
: {
5412 struct imsm_update_kill_array
*u
= (void *) update
->buf
;
5413 int victim
= u
->dev_idx
;
5414 struct active_array
*a
;
5415 struct intel_dev
**dp
;
5416 struct imsm_dev
*dev
;
5418 /* sanity check that we are not affecting the uuid of
5419 * active arrays, or deleting an active array
5421 * FIXME when immutable ids are available, but note that
5422 * we'll also need to fixup the invalidated/active
5423 * subarray indexes in mdstat
5425 for (a
= st
->arrays
; a
; a
= a
->next
)
5426 if (a
->info
.container_member
>= victim
)
5428 /* by definition if mdmon is running at least one array
5429 * is active in the container, so checking
5430 * mpb->num_raid_devs is just extra paranoia
5432 dev
= get_imsm_dev(super
, victim
);
5433 if (a
|| !dev
|| mpb
->num_raid_devs
== 1) {
5434 dprintf("failed to delete subarray-%d\n", victim
);
5438 for (dp
= &super
->devlist
; *dp
;)
5439 if ((*dp
)->index
== (unsigned)super
->current_vol
) {
5442 if ((*dp
)->index
> (unsigned)victim
)
5446 mpb
->num_raid_devs
--;
5447 super
->updates_pending
++;
5450 case update_rename_array
: {
5451 struct imsm_update_rename_array
*u
= (void *) update
->buf
;
5452 char name
[MAX_RAID_SERIAL_LEN
+1];
5453 int target
= u
->dev_idx
;
5454 struct active_array
*a
;
5455 struct imsm_dev
*dev
;
5457 /* sanity check that we are not affecting the uuid of
5460 snprintf(name
, MAX_RAID_SERIAL_LEN
, "%s", (char *) u
->name
);
5461 name
[MAX_RAID_SERIAL_LEN
] = '\0';
5462 for (a
= st
->arrays
; a
; a
= a
->next
)
5463 if (a
->info
.container_member
== target
)
5465 dev
= get_imsm_dev(super
, u
->dev_idx
);
5466 if (a
|| !dev
|| !check_name(super
, name
, 1)) {
5467 dprintf("failed to rename subarray-%d\n", target
);
5471 snprintf((char *) dev
->volume
, MAX_RAID_SERIAL_LEN
, "%s", name
);
5472 super
->updates_pending
++;
5475 case update_add_disk
:
5477 /* we may be able to repair some arrays if disks are
5480 struct active_array
*a
;
5482 super
->updates_pending
++;
5483 for (a
= st
->arrays
; a
; a
= a
->next
)
5484 a
->check_degraded
= 1;
5486 /* add some spares to the metadata */
5487 while (super
->add
) {
5491 super
->add
= al
->next
;
5492 al
->next
= super
->disks
;
5494 dprintf("%s: added %x:%x\n",
5495 __func__
, al
->major
, al
->minor
);
5502 static void imsm_prepare_update(struct supertype
*st
,
5503 struct metadata_update
*update
)
5506 * Allocate space to hold new disk entries, raid-device entries or a new
5507 * mpb if necessary. The manager synchronously waits for updates to
5508 * complete in the monitor, so new mpb buffers allocated here can be
5509 * integrated by the monitor thread without worrying about live pointers
5510 * in the manager thread.
5512 enum imsm_update_type type
= *(enum imsm_update_type
*) update
->buf
;
5513 struct intel_super
*super
= st
->sb
;
5514 struct imsm_super
*mpb
= super
->anchor
;
5519 case update_create_array
: {
5520 struct imsm_update_create_array
*u
= (void *) update
->buf
;
5521 struct intel_dev
*dv
;
5522 struct imsm_dev
*dev
= &u
->dev
;
5523 struct imsm_map
*map
= get_imsm_map(dev
, 0);
5525 struct disk_info
*inf
;
5529 inf
= get_disk_info(u
);
5530 len
= sizeof_imsm_dev(dev
, 1);
5531 /* allocate a new super->devlist entry */
5532 dv
= malloc(sizeof(*dv
));
5534 dv
->dev
= malloc(len
);
5539 update
->space
= NULL
;
5543 /* count how many spares will be converted to members */
5544 for (i
= 0; i
< map
->num_members
; i
++) {
5545 dl
= serial_to_dl(inf
[i
].serial
, super
);
5547 /* hmm maybe it failed?, nothing we can do about
5552 if (count_memberships(dl
, super
) == 0)
5555 len
+= activate
* sizeof(struct imsm_disk
);
5562 /* check if we need a larger metadata buffer */
5563 if (super
->next_buf
)
5564 buf_len
= super
->next_len
;
5566 buf_len
= super
->len
;
5568 if (__le32_to_cpu(mpb
->mpb_size
) + len
> buf_len
) {
5569 /* ok we need a larger buf than what is currently allocated
5570 * if this allocation fails process_update will notice that
5571 * ->next_len is set and ->next_buf is NULL
5573 buf_len
= ROUND_UP(__le32_to_cpu(mpb
->mpb_size
) + len
, 512);
5574 if (super
->next_buf
)
5575 free(super
->next_buf
);
5577 super
->next_len
= buf_len
;
5578 if (posix_memalign(&super
->next_buf
, 512, buf_len
) == 0)
5579 memset(super
->next_buf
, 0, buf_len
);
5581 super
->next_buf
= NULL
;
5585 /* must be called while manager is quiesced */
5586 static void imsm_delete(struct intel_super
*super
, struct dl
**dlp
, unsigned index
)
5588 struct imsm_super
*mpb
= super
->anchor
;
5590 struct imsm_dev
*dev
;
5591 struct imsm_map
*map
;
5592 int i
, j
, num_members
;
5595 dprintf("%s: deleting device[%d] from imsm_super\n",
5598 /* shift all indexes down one */
5599 for (iter
= super
->disks
; iter
; iter
= iter
->next
)
5600 if (iter
->index
> (int)index
)
5602 for (iter
= super
->missing
; iter
; iter
= iter
->next
)
5603 if (iter
->index
> (int)index
)
5606 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
5607 dev
= get_imsm_dev(super
, i
);
5608 map
= get_imsm_map(dev
, 0);
5609 num_members
= map
->num_members
;
5610 for (j
= 0; j
< num_members
; j
++) {
5611 /* update ord entries being careful not to propagate
5612 * ord-flags to the first map
5614 ord
= get_imsm_ord_tbl_ent(dev
, j
);
5616 if (ord_to_idx(ord
) <= index
)
5619 map
= get_imsm_map(dev
, 0);
5620 set_imsm_ord_tbl_ent(map
, j
, ord_to_idx(ord
- 1));
5621 map
= get_imsm_map(dev
, 1);
5623 set_imsm_ord_tbl_ent(map
, j
, ord
- 1);
5628 super
->updates_pending
++;
5630 struct dl
*dl
= *dlp
;
5632 *dlp
= (*dlp
)->next
;
5633 __free_imsm_disk(dl
);
5636 #endif /* MDASSEMBLE */
5638 static char disk_by_path
[] = "/dev/disk/by-path/";
5640 static const char *imsm_get_disk_controller_domain(const char *path
)
5642 struct sys_dev
*list
, *hba
= NULL
;
5643 char disk_path
[PATH_MAX
];
5647 list
= find_driver_devices("pci", "ahci");
5648 for (hba
= list
; hba
; hba
= hba
->next
)
5649 if (devpath_to_vendor(hba
->path
) == 0x8086)
5655 strncpy(disk_path
, disk_by_path
, PATH_MAX
- 1);
5656 strncat(disk_path
, path
, PATH_MAX
- strlen(disk_path
) - 1);
5657 if (stat(disk_path
, &st
) == 0) {
5658 dpath
= devt_to_devpath(st
.st_rdev
);
5660 ahci
= path_attached_to_hba(dpath
, hba
->path
);
5663 dprintf("path: %s(%s) hba: %s attached: %d\n",
5664 path
, dpath
, (hba
) ? hba
->path
: "NULL", ahci
);
5665 free_sys_dev(&list
);
5673 struct superswitch super_imsm
= {
5675 .examine_super
= examine_super_imsm
,
5676 .brief_examine_super
= brief_examine_super_imsm
,
5677 .brief_examine_subarrays
= brief_examine_subarrays_imsm
,
5678 .export_examine_super
= export_examine_super_imsm
,
5679 .detail_super
= detail_super_imsm
,
5680 .brief_detail_super
= brief_detail_super_imsm
,
5681 .write_init_super
= write_init_super_imsm
,
5682 .validate_geometry
= validate_geometry_imsm
,
5683 .default_chunk
= default_chunk_imsm
,
5684 .add_to_super
= add_to_super_imsm
,
5685 .detail_platform
= detail_platform_imsm
,
5686 .kill_subarray
= kill_subarray_imsm
,
5687 .update_subarray
= update_subarray_imsm
,
5688 .load_container
= load_container_imsm
,
5690 .match_home
= match_home_imsm
,
5691 .uuid_from_super
= uuid_from_super_imsm
,
5692 .getinfo_super
= getinfo_super_imsm
,
5693 .getinfo_super_disks
= getinfo_super_disks_imsm
,
5694 .update_super
= update_super_imsm
,
5696 .avail_size
= avail_size_imsm
,
5697 .min_acceptable_spare_size
= min_acceptable_spare_size_imsm
,
5699 .compare_super
= compare_super_imsm
,
5701 .load_super
= load_super_imsm
,
5702 .init_super
= init_super_imsm
,
5703 .store_super
= store_super_imsm
,
5704 .free_super
= free_super_imsm
,
5705 .match_metadata_desc
= match_metadata_desc_imsm
,
5706 .container_content
= container_content_imsm
,
5707 .default_layout
= imsm_level_to_layout
,
5708 .get_disk_controller_domain
= imsm_get_disk_controller_domain
,
5715 .open_new
= imsm_open_new
,
5716 .set_array_state
= imsm_set_array_state
,
5717 .set_disk
= imsm_set_disk
,
5718 .sync_metadata
= imsm_sync_metadata
,
5719 .activate_spare
= imsm_activate_spare
,
5720 .process_update
= imsm_process_update
,
5721 .prepare_update
= imsm_prepare_update
,
5722 #endif /* MDASSEMBLE */