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
;
240 /* internal representation of IMSM metadata */
243 void *buf
; /* O_DIRECT buffer for reading/writing metadata */
244 struct imsm_super
*anchor
; /* immovable parameters */
246 size_t len
; /* size of the 'buf' allocation */
247 void *next_buf
; /* for realloc'ing buf from the manager */
249 int updates_pending
; /* count of pending updates for mdmon */
250 int current_vol
; /* index of raid device undergoing creation */
251 __u32 create_offset
; /* common start for 'current_vol' */
252 __u32 random
; /* random data for seeding new family numbers */
253 struct intel_dev
*devlist
;
257 __u8 serial
[MAX_RAID_SERIAL_LEN
];
260 struct imsm_disk disk
;
263 struct extent
*e
; /* for determining freespace @ create */
264 int raiddisk
; /* slot to fill in autolayout */
267 struct dl
*disk_mgmt_list
; /* list of disks to add/remove while mdmon
269 struct dl
*missing
; /* disks removed while we weren't looking */
270 struct bbm_log
*bbm_log
;
271 const char *hba
; /* device path of the raid controller for this metadata */
272 const struct imsm_orom
*orom
; /* platform firmware support */
273 struct intel_super
*next
; /* (temp) list for disambiguating family_num */
277 struct imsm_disk disk
;
278 #define IMSM_UNKNOWN_OWNER (-1)
280 struct intel_disk
*next
;
284 unsigned long long start
, size
;
287 /* definition of messages passed to imsm_process_update */
288 enum imsm_update_type
{
289 update_activate_spare
,
293 update_add_remove_disk
,
294 update_reshape_container_disks
,
297 struct imsm_update_activate_spare
{
298 enum imsm_update_type type
;
302 struct imsm_update_activate_spare
*next
;
316 struct imsm_update_reshape
{
317 enum imsm_update_type type
;
320 int new_disks
[1]; /* new_raid_disks - old_raid_disks makedev number */
324 __u8 serial
[MAX_RAID_SERIAL_LEN
];
327 struct imsm_update_create_array
{
328 enum imsm_update_type type
;
333 struct imsm_update_kill_array
{
334 enum imsm_update_type type
;
338 struct imsm_update_rename_array
{
339 enum imsm_update_type type
;
340 __u8 name
[MAX_RAID_SERIAL_LEN
];
344 struct imsm_update_add_remove_disk
{
345 enum imsm_update_type type
;
348 static struct supertype
*match_metadata_desc_imsm(char *arg
)
350 struct supertype
*st
;
352 if (strcmp(arg
, "imsm") != 0 &&
353 strcmp(arg
, "default") != 0
357 st
= malloc(sizeof(*st
));
360 memset(st
, 0, sizeof(*st
));
361 st
->container_dev
= NoMdDev
;
362 st
->ss
= &super_imsm
;
363 st
->max_devs
= IMSM_MAX_DEVICES
;
364 st
->minor_version
= 0;
370 static __u8
*get_imsm_version(struct imsm_super
*mpb
)
372 return &mpb
->sig
[MPB_SIG_LEN
];
376 /* retrieve a disk directly from the anchor when the anchor is known to be
377 * up-to-date, currently only at load time
379 static struct imsm_disk
*__get_imsm_disk(struct imsm_super
*mpb
, __u8 index
)
381 if (index
>= mpb
->num_disks
)
383 return &mpb
->disk
[index
];
386 /* retrieve the disk description based on a index of the disk
389 static struct dl
*get_imsm_dl_disk(struct intel_super
*super
, __u8 index
)
393 for (d
= super
->disks
; d
; d
= d
->next
)
394 if (d
->index
== index
)
399 /* retrieve a disk from the parsed metadata */
400 static struct imsm_disk
*get_imsm_disk(struct intel_super
*super
, __u8 index
)
404 dl
= get_imsm_dl_disk(super
, index
);
411 /* generate a checksum directly from the anchor when the anchor is known to be
412 * up-to-date, currently only at load or write_super after coalescing
414 static __u32
__gen_imsm_checksum(struct imsm_super
*mpb
)
416 __u32 end
= mpb
->mpb_size
/ sizeof(end
);
417 __u32
*p
= (__u32
*) mpb
;
421 sum
+= __le32_to_cpu(*p
);
425 return sum
- __le32_to_cpu(mpb
->check_sum
);
428 static size_t sizeof_imsm_map(struct imsm_map
*map
)
430 return sizeof(struct imsm_map
) + sizeof(__u32
) * (map
->num_members
- 1);
433 struct imsm_map
*get_imsm_map(struct imsm_dev
*dev
, int second_map
)
435 struct imsm_map
*map
= &dev
->vol
.map
[0];
437 if (second_map
&& !dev
->vol
.migr_state
)
439 else if (second_map
) {
442 return ptr
+ sizeof_imsm_map(map
);
448 /* return the size of the device.
449 * migr_state increases the returned size if map[0] were to be duplicated
451 static size_t sizeof_imsm_dev(struct imsm_dev
*dev
, int migr_state
)
453 size_t size
= sizeof(*dev
) - sizeof(struct imsm_map
) +
454 sizeof_imsm_map(get_imsm_map(dev
, 0));
456 /* migrating means an additional map */
457 if (dev
->vol
.migr_state
)
458 size
+= sizeof_imsm_map(get_imsm_map(dev
, 1));
460 size
+= sizeof_imsm_map(get_imsm_map(dev
, 0));
466 /* retrieve disk serial number list from a metadata update */
467 static struct disk_info
*get_disk_info(struct imsm_update_create_array
*update
)
470 struct disk_info
*inf
;
472 inf
= u
+ sizeof(*update
) - sizeof(struct imsm_dev
) +
473 sizeof_imsm_dev(&update
->dev
, 0);
479 static struct imsm_dev
*__get_imsm_dev(struct imsm_super
*mpb
, __u8 index
)
485 if (index
>= mpb
->num_raid_devs
)
488 /* devices start after all disks */
489 offset
= ((void *) &mpb
->disk
[mpb
->num_disks
]) - _mpb
;
491 for (i
= 0; i
<= index
; i
++)
493 return _mpb
+ offset
;
495 offset
+= sizeof_imsm_dev(_mpb
+ offset
, 0);
500 static struct imsm_dev
*get_imsm_dev(struct intel_super
*super
, __u8 index
)
502 struct intel_dev
*dv
;
504 if (index
>= super
->anchor
->num_raid_devs
)
506 for (dv
= super
->devlist
; dv
; dv
= dv
->next
)
507 if (dv
->index
== index
)
512 static __u32
get_imsm_ord_tbl_ent(struct imsm_dev
*dev
, int slot
)
514 struct imsm_map
*map
;
516 if (dev
->vol
.migr_state
)
517 map
= get_imsm_map(dev
, 1);
519 map
= get_imsm_map(dev
, 0);
521 /* top byte identifies disk under rebuild */
522 return __le32_to_cpu(map
->disk_ord_tbl
[slot
]);
525 #define ord_to_idx(ord) (((ord) << 8) >> 8)
526 static __u32
get_imsm_disk_idx(struct imsm_dev
*dev
, int slot
)
528 __u32 ord
= get_imsm_ord_tbl_ent(dev
, slot
);
530 return ord_to_idx(ord
);
533 static void set_imsm_ord_tbl_ent(struct imsm_map
*map
, int slot
, __u32 ord
)
535 map
->disk_ord_tbl
[slot
] = __cpu_to_le32(ord
);
538 static int get_imsm_disk_slot(struct imsm_map
*map
, unsigned idx
)
543 for (slot
= 0; slot
< map
->num_members
; slot
++) {
544 ord
= __le32_to_cpu(map
->disk_ord_tbl
[slot
]);
545 if (ord_to_idx(ord
) == idx
)
552 static int get_imsm_raid_level(struct imsm_map
*map
)
554 if (map
->raid_level
== 1) {
555 if (map
->num_members
== 2)
561 return map
->raid_level
;
564 static int cmp_extent(const void *av
, const void *bv
)
566 const struct extent
*a
= av
;
567 const struct extent
*b
= bv
;
568 if (a
->start
< b
->start
)
570 if (a
->start
> b
->start
)
575 static int count_memberships(struct dl
*dl
, struct intel_super
*super
)
580 for (i
= 0; i
< super
->anchor
->num_raid_devs
; i
++) {
581 struct imsm_dev
*dev
= get_imsm_dev(super
, i
);
582 struct imsm_map
*map
= get_imsm_map(dev
, 0);
584 if (get_imsm_disk_slot(map
, dl
->index
) >= 0)
591 static struct extent
*get_extents(struct intel_super
*super
, struct dl
*dl
)
593 /* find a list of used extents on the given physical device */
594 struct extent
*rv
, *e
;
596 int memberships
= count_memberships(dl
, super
);
597 __u32 reservation
= MPB_SECTOR_CNT
+ IMSM_RESERVED_SECTORS
;
599 rv
= malloc(sizeof(struct extent
) * (memberships
+ 1));
604 for (i
= 0; i
< super
->anchor
->num_raid_devs
; i
++) {
605 struct imsm_dev
*dev
= get_imsm_dev(super
, i
);
606 struct imsm_map
*map
= get_imsm_map(dev
, 0);
608 if (get_imsm_disk_slot(map
, dl
->index
) >= 0) {
609 e
->start
= __le32_to_cpu(map
->pba_of_lba0
);
610 e
->size
= __le32_to_cpu(map
->blocks_per_member
);
614 qsort(rv
, memberships
, sizeof(*rv
), cmp_extent
);
616 /* determine the start of the metadata
617 * when no raid devices are defined use the default
618 * ...otherwise allow the metadata to truncate the value
619 * as is the case with older versions of imsm
622 struct extent
*last
= &rv
[memberships
- 1];
625 remainder
= __le32_to_cpu(dl
->disk
.total_blocks
) -
626 (last
->start
+ last
->size
);
627 /* round down to 1k block to satisfy precision of the kernel
631 /* make sure remainder is still sane */
632 if (remainder
< (unsigned)ROUND_UP(super
->len
, 512) >> 9)
633 remainder
= ROUND_UP(super
->len
, 512) >> 9;
634 if (reservation
> remainder
)
635 reservation
= remainder
;
637 e
->start
= __le32_to_cpu(dl
->disk
.total_blocks
) - reservation
;
642 /* try to determine how much space is reserved for metadata from
643 * the last get_extents() entry, otherwise fallback to the
646 static __u32
imsm_reserved_sectors(struct intel_super
*super
, struct dl
*dl
)
652 /* for spares just return a minimal reservation which will grow
653 * once the spare is picked up by an array
656 return MPB_SECTOR_CNT
;
658 e
= get_extents(super
, dl
);
660 return MPB_SECTOR_CNT
+ IMSM_RESERVED_SECTORS
;
662 /* scroll to last entry */
663 for (i
= 0; e
[i
].size
; i
++)
666 rv
= __le32_to_cpu(dl
->disk
.total_blocks
) - e
[i
].start
;
673 static int is_spare(struct imsm_disk
*disk
)
675 return (disk
->status
& SPARE_DISK
) == SPARE_DISK
;
678 static int is_configured(struct imsm_disk
*disk
)
680 return (disk
->status
& CONFIGURED_DISK
) == CONFIGURED_DISK
;
683 static int is_failed(struct imsm_disk
*disk
)
685 return (disk
->status
& FAILED_DISK
) == FAILED_DISK
;
688 /* Return minimum size of a spare that can be used in this array*/
689 static unsigned long long min_acceptable_spare_size_imsm(struct supertype
*st
)
691 struct intel_super
*super
= st
->sb
;
695 unsigned long long rv
= 0;
699 /* find first active disk in array */
701 while (dl
&& (is_failed(&dl
->disk
) || dl
->index
== -1))
705 /* find last lba used by subarrays */
706 e
= get_extents(super
, dl
);
709 for (i
= 0; e
[i
].size
; i
++)
712 rv
= e
[i
-1].start
+ e
[i
-1].size
;
714 /* add the amount of space needed for metadata */
715 rv
= rv
+ MPB_SECTOR_CNT
+ IMSM_RESERVED_SECTORS
;
720 static __u64
blocks_per_migr_unit(struct imsm_dev
*dev
);
722 static void print_imsm_dev(struct imsm_dev
*dev
, char *uuid
, int disk_idx
)
726 struct imsm_map
*map
= get_imsm_map(dev
, 0);
730 printf("[%.16s]:\n", dev
->volume
);
731 printf(" UUID : %s\n", uuid
);
732 printf(" RAID Level : %d\n", get_imsm_raid_level(map
));
733 printf(" Members : %d\n", map
->num_members
);
734 printf(" Slots : [");
735 for (i
= 0; i
< map
->num_members
; i
++) {
736 ord
= get_imsm_ord_tbl_ent(dev
, i
);
737 printf("%s", ord
& IMSM_ORD_REBUILD
? "_" : "U");
740 slot
= get_imsm_disk_slot(map
, disk_idx
);
742 ord
= get_imsm_ord_tbl_ent(dev
, slot
);
743 printf(" This Slot : %d%s\n", slot
,
744 ord
& IMSM_ORD_REBUILD
? " (out-of-sync)" : "");
746 printf(" This Slot : ?\n");
747 sz
= __le32_to_cpu(dev
->size_high
);
749 sz
+= __le32_to_cpu(dev
->size_low
);
750 printf(" Array Size : %llu%s\n", (unsigned long long)sz
,
751 human_size(sz
* 512));
752 sz
= __le32_to_cpu(map
->blocks_per_member
);
753 printf(" Per Dev Size : %llu%s\n", (unsigned long long)sz
,
754 human_size(sz
* 512));
755 printf(" Sector Offset : %u\n",
756 __le32_to_cpu(map
->pba_of_lba0
));
757 printf(" Num Stripes : %u\n",
758 __le32_to_cpu(map
->num_data_stripes
));
759 printf(" Chunk Size : %u KiB\n",
760 __le16_to_cpu(map
->blocks_per_strip
) / 2);
761 printf(" Reserved : %d\n", __le32_to_cpu(dev
->reserved_blocks
));
762 printf(" Migrate State : ");
763 if (dev
->vol
.migr_state
) {
764 if (migr_type(dev
) == MIGR_INIT
)
765 printf("initialize\n");
766 else if (migr_type(dev
) == MIGR_REBUILD
)
768 else if (migr_type(dev
) == MIGR_VERIFY
)
770 else if (migr_type(dev
) == MIGR_GEN_MIGR
)
771 printf("general migration\n");
772 else if (migr_type(dev
) == MIGR_STATE_CHANGE
)
773 printf("state change\n");
774 else if (migr_type(dev
) == MIGR_REPAIR
)
777 printf("<unknown:%d>\n", migr_type(dev
));
780 printf(" Map State : %s", map_state_str
[map
->map_state
]);
781 if (dev
->vol
.migr_state
) {
782 struct imsm_map
*map
= get_imsm_map(dev
, 1);
784 printf(" <-- %s", map_state_str
[map
->map_state
]);
785 printf("\n Checkpoint : %u (%llu)",
786 __le32_to_cpu(dev
->vol
.curr_migr_unit
),
787 (unsigned long long)blocks_per_migr_unit(dev
));
790 printf(" Dirty State : %s\n", dev
->vol
.dirty
? "dirty" : "clean");
793 static void print_imsm_disk(struct imsm_super
*mpb
, int index
, __u32 reserved
)
795 struct imsm_disk
*disk
= __get_imsm_disk(mpb
, index
);
796 char str
[MAX_RAID_SERIAL_LEN
+ 1];
799 if (index
< 0 || !disk
)
803 snprintf(str
, MAX_RAID_SERIAL_LEN
+ 1, "%s", disk
->serial
);
804 printf(" Disk%02d Serial : %s\n", index
, str
);
805 printf(" State :%s%s%s\n", is_spare(disk
) ? " spare" : "",
806 is_configured(disk
) ? " active" : "",
807 is_failed(disk
) ? " failed" : "");
808 printf(" Id : %08x\n", __le32_to_cpu(disk
->scsi_id
));
809 sz
= __le32_to_cpu(disk
->total_blocks
) - reserved
;
810 printf(" Usable Size : %llu%s\n", (unsigned long long)sz
,
811 human_size(sz
* 512));
814 static void getinfo_super_imsm(struct supertype
*st
, struct mdinfo
*info
, char *map
);
816 static void examine_super_imsm(struct supertype
*st
, char *homehost
)
818 struct intel_super
*super
= st
->sb
;
819 struct imsm_super
*mpb
= super
->anchor
;
820 char str
[MAX_SIGNATURE_LENGTH
];
825 __u32 reserved
= imsm_reserved_sectors(super
, super
->disks
);
828 snprintf(str
, MPB_SIG_LEN
, "%s", mpb
->sig
);
829 printf(" Magic : %s\n", str
);
830 snprintf(str
, strlen(MPB_VERSION_RAID0
), "%s", get_imsm_version(mpb
));
831 printf(" Version : %s\n", get_imsm_version(mpb
));
832 printf(" Orig Family : %08x\n", __le32_to_cpu(mpb
->orig_family_num
));
833 printf(" Family : %08x\n", __le32_to_cpu(mpb
->family_num
));
834 printf(" Generation : %08x\n", __le32_to_cpu(mpb
->generation_num
));
835 getinfo_super_imsm(st
, &info
, NULL
);
836 fname_from_uuid(st
, &info
, nbuf
, ':');
837 printf(" UUID : %s\n", nbuf
+ 5);
838 sum
= __le32_to_cpu(mpb
->check_sum
);
839 printf(" Checksum : %08x %s\n", sum
,
840 __gen_imsm_checksum(mpb
) == sum
? "correct" : "incorrect");
841 printf(" MPB Sectors : %d\n", mpb_sectors(mpb
));
842 printf(" Disks : %d\n", mpb
->num_disks
);
843 printf(" RAID Devices : %d\n", mpb
->num_raid_devs
);
844 print_imsm_disk(mpb
, super
->disks
->index
, reserved
);
845 if (super
->bbm_log
) {
846 struct bbm_log
*log
= super
->bbm_log
;
849 printf("Bad Block Management Log:\n");
850 printf(" Log Size : %d\n", __le32_to_cpu(mpb
->bbm_log_size
));
851 printf(" Signature : %x\n", __le32_to_cpu(log
->signature
));
852 printf(" Entry Count : %d\n", __le32_to_cpu(log
->entry_count
));
853 printf(" Spare Blocks : %d\n", __le32_to_cpu(log
->reserved_spare_block_count
));
854 printf(" First Spare : %llx\n",
855 (unsigned long long) __le64_to_cpu(log
->first_spare_lba
));
857 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
859 struct imsm_dev
*dev
= __get_imsm_dev(mpb
, i
);
861 super
->current_vol
= i
;
862 getinfo_super_imsm(st
, &info
, NULL
);
863 fname_from_uuid(st
, &info
, nbuf
, ':');
864 print_imsm_dev(dev
, nbuf
+ 5, super
->disks
->index
);
866 for (i
= 0; i
< mpb
->num_disks
; i
++) {
867 if (i
== super
->disks
->index
)
869 print_imsm_disk(mpb
, i
, reserved
);
871 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
872 struct imsm_disk
*disk
;
873 char str
[MAX_RAID_SERIAL_LEN
+ 1];
881 snprintf(str
, MAX_RAID_SERIAL_LEN
+ 1, "%s", disk
->serial
);
882 printf(" Disk Serial : %s\n", str
);
883 printf(" State :%s%s%s\n", is_spare(disk
) ? " spare" : "",
884 is_configured(disk
) ? " active" : "",
885 is_failed(disk
) ? " failed" : "");
886 printf(" Id : %08x\n", __le32_to_cpu(disk
->scsi_id
));
887 sz
= __le32_to_cpu(disk
->total_blocks
) - reserved
;
888 printf(" Usable Size : %llu%s\n", (unsigned long long)sz
,
889 human_size(sz
* 512));
893 static void brief_examine_super_imsm(struct supertype
*st
, int verbose
)
895 /* We just write a generic IMSM ARRAY entry */
898 struct intel_super
*super
= st
->sb
;
900 if (!super
->anchor
->num_raid_devs
) {
901 printf("ARRAY metadata=imsm\n");
905 getinfo_super_imsm(st
, &info
, NULL
);
906 fname_from_uuid(st
, &info
, nbuf
, ':');
907 printf("ARRAY metadata=imsm UUID=%s\n", nbuf
+ 5);
910 static void brief_examine_subarrays_imsm(struct supertype
*st
, int verbose
)
912 /* We just write a generic IMSM ARRAY entry */
916 struct intel_super
*super
= st
->sb
;
919 if (!super
->anchor
->num_raid_devs
)
922 getinfo_super_imsm(st
, &info
, NULL
);
923 fname_from_uuid(st
, &info
, nbuf
, ':');
924 for (i
= 0; i
< super
->anchor
->num_raid_devs
; i
++) {
925 struct imsm_dev
*dev
= get_imsm_dev(super
, i
);
927 super
->current_vol
= i
;
928 getinfo_super_imsm(st
, &info
, NULL
);
929 fname_from_uuid(st
, &info
, nbuf1
, ':');
930 printf("ARRAY /dev/md/%.16s container=%s member=%d UUID=%s\n",
931 dev
->volume
, nbuf
+ 5, i
, nbuf1
+ 5);
935 static void export_examine_super_imsm(struct supertype
*st
)
937 struct intel_super
*super
= st
->sb
;
938 struct imsm_super
*mpb
= super
->anchor
;
942 getinfo_super_imsm(st
, &info
, NULL
);
943 fname_from_uuid(st
, &info
, nbuf
, ':');
944 printf("MD_METADATA=imsm\n");
945 printf("MD_LEVEL=container\n");
946 printf("MD_UUID=%s\n", nbuf
+5);
947 printf("MD_DEVICES=%u\n", mpb
->num_disks
);
950 static void detail_super_imsm(struct supertype
*st
, char *homehost
)
955 getinfo_super_imsm(st
, &info
, NULL
);
956 fname_from_uuid(st
, &info
, nbuf
, ':');
957 printf("\n UUID : %s\n", nbuf
+ 5);
960 static void brief_detail_super_imsm(struct supertype
*st
)
964 getinfo_super_imsm(st
, &info
, NULL
);
965 fname_from_uuid(st
, &info
, nbuf
, ':');
966 printf(" UUID=%s", nbuf
+ 5);
969 static int imsm_read_serial(int fd
, char *devname
, __u8
*serial
);
970 static void fd2devname(int fd
, char *name
);
972 static int imsm_enumerate_ports(const char *hba_path
, int port_count
, int host_base
, int verbose
)
974 /* dump an unsorted list of devices attached to ahci, as well as
975 * non-connected ports
977 int hba_len
= strlen(hba_path
) + 1;
982 unsigned long port_mask
= (1 << port_count
) - 1;
984 if (port_count
> (int)sizeof(port_mask
) * 8) {
986 fprintf(stderr
, Name
": port_count %d out of range\n", port_count
);
990 /* scroll through /sys/dev/block looking for devices attached to
993 dir
= opendir("/sys/dev/block");
994 for (ent
= dir
? readdir(dir
) : NULL
; ent
; ent
= readdir(dir
)) {
1005 if (sscanf(ent
->d_name
, "%d:%d", &major
, &minor
) != 2)
1007 path
= devt_to_devpath(makedev(major
, minor
));
1010 if (!path_attached_to_hba(path
, hba_path
)) {
1016 /* retrieve the scsi device type */
1017 if (asprintf(&device
, "/sys/dev/block/%d:%d/device/xxxxxxx", major
, minor
) < 0) {
1019 fprintf(stderr
, Name
": failed to allocate 'device'\n");
1023 sprintf(device
, "/sys/dev/block/%d:%d/device/type", major
, minor
);
1024 if (load_sys(device
, buf
) != 0) {
1026 fprintf(stderr
, Name
": failed to read device type for %s\n",
1032 type
= strtoul(buf
, NULL
, 10);
1034 /* if it's not a disk print the vendor and model */
1035 if (!(type
== 0 || type
== 7 || type
== 14)) {
1038 sprintf(device
, "/sys/dev/block/%d:%d/device/vendor", major
, minor
);
1039 if (load_sys(device
, buf
) == 0) {
1040 strncpy(vendor
, buf
, sizeof(vendor
));
1041 vendor
[sizeof(vendor
) - 1] = '\0';
1042 c
= (char *) &vendor
[sizeof(vendor
) - 1];
1043 while (isspace(*c
) || *c
== '\0')
1047 sprintf(device
, "/sys/dev/block/%d:%d/device/model", major
, minor
);
1048 if (load_sys(device
, buf
) == 0) {
1049 strncpy(model
, buf
, sizeof(model
));
1050 model
[sizeof(model
) - 1] = '\0';
1051 c
= (char *) &model
[sizeof(model
) - 1];
1052 while (isspace(*c
) || *c
== '\0')
1056 if (vendor
[0] && model
[0])
1057 sprintf(buf
, "%.64s %.64s", vendor
, model
);
1059 switch (type
) { /* numbers from hald/linux/device.c */
1060 case 1: sprintf(buf
, "tape"); break;
1061 case 2: sprintf(buf
, "printer"); break;
1062 case 3: sprintf(buf
, "processor"); break;
1064 case 5: sprintf(buf
, "cdrom"); break;
1065 case 6: sprintf(buf
, "scanner"); break;
1066 case 8: sprintf(buf
, "media_changer"); break;
1067 case 9: sprintf(buf
, "comm"); break;
1068 case 12: sprintf(buf
, "raid"); break;
1069 default: sprintf(buf
, "unknown");
1075 /* chop device path to 'host%d' and calculate the port number */
1076 c
= strchr(&path
[hba_len
], '/');
1079 fprintf(stderr
, Name
": %s - invalid path name\n", path
+ hba_len
);
1084 if (sscanf(&path
[hba_len
], "host%d", &port
) == 1)
1088 *c
= '/'; /* repair the full string */
1089 fprintf(stderr
, Name
": failed to determine port number for %s\n",
1096 /* mark this port as used */
1097 port_mask
&= ~(1 << port
);
1099 /* print out the device information */
1101 printf(" Port%d : - non-disk device (%s) -\n", port
, buf
);
1105 fd
= dev_open(ent
->d_name
, O_RDONLY
);
1107 printf(" Port%d : - disk info unavailable -\n", port
);
1109 fd2devname(fd
, buf
);
1110 printf(" Port%d : %s", port
, buf
);
1111 if (imsm_read_serial(fd
, NULL
, (__u8
*) buf
) == 0)
1112 printf(" (%s)\n", buf
);
1127 for (i
= 0; i
< port_count
; i
++)
1128 if (port_mask
& (1 << i
))
1129 printf(" Port%d : - no device attached -\n", i
);
1135 static int detail_platform_imsm(int verbose
, int enumerate_only
)
1137 /* There are two components to imsm platform support, the ahci SATA
1138 * controller and the option-rom. To find the SATA controller we
1139 * simply look in /sys/bus/pci/drivers/ahci to see if an ahci
1140 * controller with the Intel vendor id is present. This approach
1141 * allows mdadm to leverage the kernel's ahci detection logic, with the
1142 * caveat that if ahci.ko is not loaded mdadm will not be able to
1143 * detect platform raid capabilities. The option-rom resides in a
1144 * platform "Adapter ROM". We scan for its signature to retrieve the
1145 * platform capabilities. If raid support is disabled in the BIOS the
1146 * option-rom capability structure will not be available.
1148 const struct imsm_orom
*orom
;
1149 struct sys_dev
*list
, *hba
;
1152 const char *hba_path
;
1156 if (enumerate_only
) {
1157 if (check_env("IMSM_NO_PLATFORM") || find_imsm_orom())
1162 list
= find_driver_devices("pci", "ahci");
1163 for (hba
= list
; hba
; hba
= hba
->next
)
1164 if (devpath_to_vendor(hba
->path
) == 0x8086)
1169 fprintf(stderr
, Name
": unable to find active ahci controller\n");
1170 free_sys_dev(&list
);
1173 fprintf(stderr
, Name
": found Intel SATA AHCI Controller\n");
1174 hba_path
= hba
->path
;
1176 free_sys_dev(&list
);
1178 orom
= find_imsm_orom();
1181 fprintf(stderr
, Name
": imsm option-rom not found\n");
1185 printf(" Platform : Intel(R) Matrix Storage Manager\n");
1186 printf(" Version : %d.%d.%d.%d\n", orom
->major_ver
, orom
->minor_ver
,
1187 orom
->hotfix_ver
, orom
->build
);
1188 printf(" RAID Levels :%s%s%s%s%s\n",
1189 imsm_orom_has_raid0(orom
) ? " raid0" : "",
1190 imsm_orom_has_raid1(orom
) ? " raid1" : "",
1191 imsm_orom_has_raid1e(orom
) ? " raid1e" : "",
1192 imsm_orom_has_raid10(orom
) ? " raid10" : "",
1193 imsm_orom_has_raid5(orom
) ? " raid5" : "");
1194 printf(" Chunk Sizes :%s%s%s%s%s%s%s%s%s%s%s%s%s%s%s%s\n",
1195 imsm_orom_has_chunk(orom
, 2) ? " 2k" : "",
1196 imsm_orom_has_chunk(orom
, 4) ? " 4k" : "",
1197 imsm_orom_has_chunk(orom
, 8) ? " 8k" : "",
1198 imsm_orom_has_chunk(orom
, 16) ? " 16k" : "",
1199 imsm_orom_has_chunk(orom
, 32) ? " 32k" : "",
1200 imsm_orom_has_chunk(orom
, 64) ? " 64k" : "",
1201 imsm_orom_has_chunk(orom
, 128) ? " 128k" : "",
1202 imsm_orom_has_chunk(orom
, 256) ? " 256k" : "",
1203 imsm_orom_has_chunk(orom
, 512) ? " 512k" : "",
1204 imsm_orom_has_chunk(orom
, 1024*1) ? " 1M" : "",
1205 imsm_orom_has_chunk(orom
, 1024*2) ? " 2M" : "",
1206 imsm_orom_has_chunk(orom
, 1024*4) ? " 4M" : "",
1207 imsm_orom_has_chunk(orom
, 1024*8) ? " 8M" : "",
1208 imsm_orom_has_chunk(orom
, 1024*16) ? " 16M" : "",
1209 imsm_orom_has_chunk(orom
, 1024*32) ? " 32M" : "",
1210 imsm_orom_has_chunk(orom
, 1024*64) ? " 64M" : "");
1211 printf(" Max Disks : %d\n", orom
->tds
);
1212 printf(" Max Volumes : %d\n", orom
->vpa
);
1213 printf(" I/O Controller : %s\n", hba_path
);
1215 /* find the smallest scsi host number to determine a port number base */
1216 dir
= opendir(hba_path
);
1217 for (ent
= dir
? readdir(dir
) : NULL
; ent
; ent
= readdir(dir
)) {
1220 if (sscanf(ent
->d_name
, "host%d", &host
) != 1)
1222 if (port_count
== 0)
1224 else if (host
< host_base
)
1227 if (host
+ 1 > port_count
+ host_base
)
1228 port_count
= host
+ 1 - host_base
;
1234 if (!port_count
|| imsm_enumerate_ports(hba_path
, port_count
,
1235 host_base
, verbose
) != 0) {
1237 fprintf(stderr
, Name
": failed to enumerate ports\n");
1245 static int match_home_imsm(struct supertype
*st
, char *homehost
)
1247 /* the imsm metadata format does not specify any host
1248 * identification information. We return -1 since we can never
1249 * confirm nor deny whether a given array is "meant" for this
1250 * host. We rely on compare_super and the 'family_num' fields to
1251 * exclude member disks that do not belong, and we rely on
1252 * mdadm.conf to specify the arrays that should be assembled.
1253 * Auto-assembly may still pick up "foreign" arrays.
1259 static void uuid_from_super_imsm(struct supertype
*st
, int uuid
[4])
1261 /* The uuid returned here is used for:
1262 * uuid to put into bitmap file (Create, Grow)
1263 * uuid for backup header when saving critical section (Grow)
1264 * comparing uuids when re-adding a device into an array
1265 * In these cases the uuid required is that of the data-array,
1266 * not the device-set.
1267 * uuid to recognise same set when adding a missing device back
1268 * to an array. This is a uuid for the device-set.
1270 * For each of these we can make do with a truncated
1271 * or hashed uuid rather than the original, as long as
1273 * In each case the uuid required is that of the data-array,
1274 * not the device-set.
1276 /* imsm does not track uuid's so we synthesis one using sha1 on
1277 * - The signature (Which is constant for all imsm array, but no matter)
1278 * - the orig_family_num of the container
1279 * - the index number of the volume
1280 * - the 'serial' number of the volume.
1281 * Hopefully these are all constant.
1283 struct intel_super
*super
= st
->sb
;
1286 struct sha1_ctx ctx
;
1287 struct imsm_dev
*dev
= NULL
;
1290 /* some mdadm versions failed to set ->orig_family_num, in which
1291 * case fall back to ->family_num. orig_family_num will be
1292 * fixed up with the first metadata update.
1294 family_num
= super
->anchor
->orig_family_num
;
1295 if (family_num
== 0)
1296 family_num
= super
->anchor
->family_num
;
1297 sha1_init_ctx(&ctx
);
1298 sha1_process_bytes(super
->anchor
->sig
, MPB_SIG_LEN
, &ctx
);
1299 sha1_process_bytes(&family_num
, sizeof(__u32
), &ctx
);
1300 if (super
->current_vol
>= 0)
1301 dev
= get_imsm_dev(super
, super
->current_vol
);
1303 __u32 vol
= super
->current_vol
;
1304 sha1_process_bytes(&vol
, sizeof(vol
), &ctx
);
1305 sha1_process_bytes(dev
->volume
, MAX_RAID_SERIAL_LEN
, &ctx
);
1307 sha1_finish_ctx(&ctx
, buf
);
1308 memcpy(uuid
, buf
, 4*4);
1313 get_imsm_numerical_version(struct imsm_super
*mpb
, int *m
, int *p
)
1315 __u8
*v
= get_imsm_version(mpb
);
1316 __u8
*end
= mpb
->sig
+ MAX_SIGNATURE_LENGTH
;
1317 char major
[] = { 0, 0, 0 };
1318 char minor
[] = { 0 ,0, 0 };
1319 char patch
[] = { 0, 0, 0 };
1320 char *ver_parse
[] = { major
, minor
, patch
};
1324 while (*v
!= '\0' && v
< end
) {
1325 if (*v
!= '.' && j
< 2)
1326 ver_parse
[i
][j
++] = *v
;
1334 *m
= strtol(minor
, NULL
, 0);
1335 *p
= strtol(patch
, NULL
, 0);
1339 static __u32
migr_strip_blocks_resync(struct imsm_dev
*dev
)
1341 /* migr_strip_size when repairing or initializing parity */
1342 struct imsm_map
*map
= get_imsm_map(dev
, 0);
1343 __u32 chunk
= __le32_to_cpu(map
->blocks_per_strip
);
1345 switch (get_imsm_raid_level(map
)) {
1350 return 128*1024 >> 9;
1354 static __u32
migr_strip_blocks_rebuild(struct imsm_dev
*dev
)
1356 /* migr_strip_size when rebuilding a degraded disk, no idea why
1357 * this is different than migr_strip_size_resync(), but it's good
1360 struct imsm_map
*map
= get_imsm_map(dev
, 1);
1361 __u32 chunk
= __le32_to_cpu(map
->blocks_per_strip
);
1363 switch (get_imsm_raid_level(map
)) {
1366 if (map
->num_members
% map
->num_domains
== 0)
1367 return 128*1024 >> 9;
1371 return max((__u32
) 64*1024 >> 9, chunk
);
1373 return 128*1024 >> 9;
1377 static __u32
num_stripes_per_unit_resync(struct imsm_dev
*dev
)
1379 struct imsm_map
*lo
= get_imsm_map(dev
, 0);
1380 struct imsm_map
*hi
= get_imsm_map(dev
, 1);
1381 __u32 lo_chunk
= __le32_to_cpu(lo
->blocks_per_strip
);
1382 __u32 hi_chunk
= __le32_to_cpu(hi
->blocks_per_strip
);
1384 return max((__u32
) 1, hi_chunk
/ lo_chunk
);
1387 static __u32
num_stripes_per_unit_rebuild(struct imsm_dev
*dev
)
1389 struct imsm_map
*lo
= get_imsm_map(dev
, 0);
1390 int level
= get_imsm_raid_level(lo
);
1392 if (level
== 1 || level
== 10) {
1393 struct imsm_map
*hi
= get_imsm_map(dev
, 1);
1395 return hi
->num_domains
;
1397 return num_stripes_per_unit_resync(dev
);
1400 static __u8
imsm_num_data_members(struct imsm_dev
*dev
)
1402 /* named 'imsm_' because raid0, raid1 and raid10
1403 * counter-intuitively have the same number of data disks
1405 struct imsm_map
*map
= get_imsm_map(dev
, 0);
1407 switch (get_imsm_raid_level(map
)) {
1411 return map
->num_members
;
1413 return map
->num_members
- 1;
1415 dprintf("%s: unsupported raid level\n", __func__
);
1420 static __u32
parity_segment_depth(struct imsm_dev
*dev
)
1422 struct imsm_map
*map
= get_imsm_map(dev
, 0);
1423 __u32 chunk
= __le32_to_cpu(map
->blocks_per_strip
);
1425 switch(get_imsm_raid_level(map
)) {
1428 return chunk
* map
->num_domains
;
1430 return chunk
* map
->num_members
;
1436 static __u32
map_migr_block(struct imsm_dev
*dev
, __u32 block
)
1438 struct imsm_map
*map
= get_imsm_map(dev
, 1);
1439 __u32 chunk
= __le32_to_cpu(map
->blocks_per_strip
);
1440 __u32 strip
= block
/ chunk
;
1442 switch (get_imsm_raid_level(map
)) {
1445 __u32 vol_strip
= (strip
* map
->num_domains
) + 1;
1446 __u32 vol_stripe
= vol_strip
/ map
->num_members
;
1448 return vol_stripe
* chunk
+ block
% chunk
;
1450 __u32 stripe
= strip
/ (map
->num_members
- 1);
1452 return stripe
* chunk
+ block
% chunk
;
1459 static __u64
blocks_per_migr_unit(struct imsm_dev
*dev
)
1461 /* calculate the conversion factor between per member 'blocks'
1462 * (md/{resync,rebuild}_start) and imsm migration units, return
1463 * 0 for the 'not migrating' and 'unsupported migration' cases
1465 if (!dev
->vol
.migr_state
)
1468 switch (migr_type(dev
)) {
1472 struct imsm_map
*map
= get_imsm_map(dev
, 0);
1473 __u32 stripes_per_unit
;
1474 __u32 blocks_per_unit
;
1483 /* yes, this is really the translation of migr_units to
1484 * per-member blocks in the 'resync' case
1486 stripes_per_unit
= num_stripes_per_unit_resync(dev
);
1487 migr_chunk
= migr_strip_blocks_resync(dev
);
1488 disks
= imsm_num_data_members(dev
);
1489 blocks_per_unit
= stripes_per_unit
* migr_chunk
* disks
;
1490 stripe
= __le32_to_cpu(map
->blocks_per_strip
) * disks
;
1491 segment
= blocks_per_unit
/ stripe
;
1492 block_rel
= blocks_per_unit
- segment
* stripe
;
1493 parity_depth
= parity_segment_depth(dev
);
1494 block_map
= map_migr_block(dev
, block_rel
);
1495 return block_map
+ parity_depth
* segment
;
1497 case MIGR_REBUILD
: {
1498 __u32 stripes_per_unit
;
1501 stripes_per_unit
= num_stripes_per_unit_rebuild(dev
);
1502 migr_chunk
= migr_strip_blocks_rebuild(dev
);
1503 return migr_chunk
* stripes_per_unit
;
1506 /* FIXME I need a number here */
1507 case MIGR_STATE_CHANGE
:
1513 static int imsm_level_to_layout(int level
)
1521 return ALGORITHM_LEFT_ASYMMETRIC
;
1528 static void getinfo_super_imsm_volume(struct supertype
*st
, struct mdinfo
*info
, char *dmap
)
1530 struct intel_super
*super
= st
->sb
;
1531 struct imsm_dev
*dev
= get_imsm_dev(super
, super
->current_vol
);
1532 struct imsm_map
*map
= get_imsm_map(dev
, 0);
1533 struct imsm_map
*prev_map
= get_imsm_map(dev
, 1);
1536 int map_disks
= info
->array
.raid_disks
;
1538 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
1539 if (dl
->raiddisk
== info
->disk
.raid_disk
)
1541 info
->container_member
= super
->current_vol
;
1542 info
->array
.raid_disks
= map
->num_members
;
1543 info
->array
.level
= get_imsm_raid_level(map
);
1544 info
->array
.layout
= imsm_level_to_layout(info
->array
.level
);
1545 info
->array
.md_minor
= -1;
1546 info
->array
.ctime
= 0;
1547 info
->array
.utime
= 0;
1548 info
->array
.chunk_size
= __le16_to_cpu(map
->blocks_per_strip
) << 9;
1549 info
->array
.state
= !dev
->vol
.dirty
;
1550 info
->custom_array_size
= __le32_to_cpu(dev
->size_high
);
1551 info
->custom_array_size
<<= 32;
1552 info
->custom_array_size
|= __le32_to_cpu(dev
->size_low
);
1554 info
->disk
.major
= 0;
1555 info
->disk
.minor
= 0;
1557 info
->disk
.major
= dl
->major
;
1558 info
->disk
.minor
= dl
->minor
;
1561 info
->data_offset
= __le32_to_cpu(map
->pba_of_lba0
);
1562 info
->component_size
= __le32_to_cpu(map
->blocks_per_member
);
1563 memset(info
->uuid
, 0, sizeof(info
->uuid
));
1564 info
->recovery_start
= MaxSector
;
1565 info
->reshape_active
= (prev_map
!= NULL
);
1566 if (info
->reshape_active
)
1567 info
->delta_disks
= map
->num_members
- prev_map
->num_members
;
1569 info
->delta_disks
= 0;
1571 if (map
->map_state
== IMSM_T_STATE_UNINITIALIZED
|| dev
->vol
.dirty
) {
1572 info
->resync_start
= 0;
1573 } else if (dev
->vol
.migr_state
) {
1574 switch (migr_type(dev
)) {
1577 __u64 blocks_per_unit
= blocks_per_migr_unit(dev
);
1578 __u64 units
= __le32_to_cpu(dev
->vol
.curr_migr_unit
);
1580 info
->resync_start
= blocks_per_unit
* units
;
1584 /* we could emulate the checkpointing of
1585 * 'sync_action=check' migrations, but for now
1586 * we just immediately complete them
1589 /* this is handled by container_content_imsm() */
1591 case MIGR_STATE_CHANGE
:
1592 /* FIXME handle other migrations */
1594 /* we are not dirty, so... */
1595 info
->resync_start
= MaxSector
;
1598 info
->resync_start
= MaxSector
;
1600 strncpy(info
->name
, (char *) dev
->volume
, MAX_RAID_SERIAL_LEN
);
1601 info
->name
[MAX_RAID_SERIAL_LEN
] = 0;
1603 info
->array
.major_version
= -1;
1604 info
->array
.minor_version
= -2;
1605 devname
= devnum2devname(st
->container_dev
);
1606 *info
->text_version
= '\0';
1608 sprintf(info
->text_version
, "/%s/%d", devname
, info
->container_member
);
1610 info
->safe_mode_delay
= 4000; /* 4 secs like the Matrix driver */
1611 uuid_from_super_imsm(st
, info
->uuid
);
1615 for (i
=0; i
<map_disks
; i
++) {
1617 if (i
< info
->array
.raid_disks
) {
1618 struct imsm_disk
*dsk
;
1619 j
= get_imsm_disk_idx(dev
, i
);
1620 dsk
= get_imsm_disk(super
, j
);
1621 if (dsk
&& (dsk
->status
& CONFIGURED_DISK
))
1628 /* check the config file to see if we can return a real uuid for this spare */
1629 static void fixup_container_spare_uuid(struct mdinfo
*inf
)
1631 struct mddev_ident
*array_list
;
1633 if (inf
->array
.level
!= LEVEL_CONTAINER
||
1634 memcmp(inf
->uuid
, uuid_match_any
, sizeof(int[4])) != 0)
1637 array_list
= conf_get_ident(NULL
);
1639 for (; array_list
; array_list
= array_list
->next
) {
1640 if (array_list
->uuid_set
) {
1641 struct supertype
*_sst
; /* spare supertype */
1642 struct supertype
*_cst
; /* container supertype */
1644 _cst
= array_list
->st
;
1646 _sst
= _cst
->ss
->match_metadata_desc(inf
->text_version
);
1651 memcpy(inf
->uuid
, array_list
->uuid
, sizeof(int[4]));
1660 static __u8
imsm_check_degraded(struct intel_super
*super
, struct imsm_dev
*dev
, int failed
);
1661 static int imsm_count_failed(struct intel_super
*super
, struct imsm_dev
*dev
);
1663 static struct imsm_disk
*get_imsm_missing(struct intel_super
*super
, __u8 index
)
1667 for (d
= super
->missing
; d
; d
= d
->next
)
1668 if (d
->index
== index
)
1673 static void getinfo_super_imsm(struct supertype
*st
, struct mdinfo
*info
, char *map
)
1675 struct intel_super
*super
= st
->sb
;
1676 struct imsm_disk
*disk
;
1677 int map_disks
= info
->array
.raid_disks
;
1678 int max_enough
= -1;
1680 struct imsm_super
*mpb
;
1682 if (super
->current_vol
>= 0) {
1683 getinfo_super_imsm_volume(st
, info
, map
);
1687 /* Set raid_disks to zero so that Assemble will always pull in valid
1690 info
->array
.raid_disks
= 0;
1691 info
->array
.level
= LEVEL_CONTAINER
;
1692 info
->array
.layout
= 0;
1693 info
->array
.md_minor
= -1;
1694 info
->array
.ctime
= 0; /* N/A for imsm */
1695 info
->array
.utime
= 0;
1696 info
->array
.chunk_size
= 0;
1698 info
->disk
.major
= 0;
1699 info
->disk
.minor
= 0;
1700 info
->disk
.raid_disk
= -1;
1701 info
->reshape_active
= 0;
1702 info
->array
.major_version
= -1;
1703 info
->array
.minor_version
= -2;
1704 strcpy(info
->text_version
, "imsm");
1705 info
->safe_mode_delay
= 0;
1706 info
->disk
.number
= -1;
1707 info
->disk
.state
= 0;
1709 info
->recovery_start
= MaxSector
;
1711 /* do we have the all the insync disks that we expect? */
1712 mpb
= super
->anchor
;
1714 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
1715 struct imsm_dev
*dev
= get_imsm_dev(super
, i
);
1716 int failed
, enough
, j
, missing
= 0;
1717 struct imsm_map
*map
;
1720 failed
= imsm_count_failed(super
, dev
);
1721 state
= imsm_check_degraded(super
, dev
, failed
);
1722 map
= get_imsm_map(dev
, dev
->vol
.migr_state
);
1724 /* any newly missing disks?
1725 * (catches single-degraded vs double-degraded)
1727 for (j
= 0; j
< map
->num_members
; j
++) {
1728 __u32 ord
= get_imsm_ord_tbl_ent(dev
, i
);
1729 __u32 idx
= ord_to_idx(ord
);
1731 if (!(ord
& IMSM_ORD_REBUILD
) &&
1732 get_imsm_missing(super
, idx
)) {
1738 if (state
== IMSM_T_STATE_FAILED
)
1740 else if (state
== IMSM_T_STATE_DEGRADED
&&
1741 (state
!= map
->map_state
|| missing
))
1743 else /* we're normal, or already degraded */
1746 /* in the missing/failed disk case check to see
1747 * if at least one array is runnable
1749 max_enough
= max(max_enough
, enough
);
1751 dprintf("%s: enough: %d\n", __func__
, max_enough
);
1752 info
->container_enough
= max_enough
;
1755 __u32 reserved
= imsm_reserved_sectors(super
, super
->disks
);
1757 disk
= &super
->disks
->disk
;
1758 info
->data_offset
= __le32_to_cpu(disk
->total_blocks
) - reserved
;
1759 info
->component_size
= reserved
;
1760 info
->disk
.state
= is_configured(disk
) ? (1 << MD_DISK_ACTIVE
) : 0;
1761 /* we don't change info->disk.raid_disk here because
1762 * this state will be finalized in mdmon after we have
1763 * found the 'most fresh' version of the metadata
1765 info
->disk
.state
|= is_failed(disk
) ? (1 << MD_DISK_FAULTY
) : 0;
1766 info
->disk
.state
|= is_spare(disk
) ? 0 : (1 << MD_DISK_SYNC
);
1769 /* only call uuid_from_super_imsm when this disk is part of a populated container,
1770 * ->compare_super may have updated the 'num_raid_devs' field for spares
1772 if (info
->disk
.state
& (1 << MD_DISK_SYNC
) || super
->anchor
->num_raid_devs
)
1773 uuid_from_super_imsm(st
, info
->uuid
);
1775 memcpy(info
->uuid
, uuid_match_any
, sizeof(int[4]));
1776 fixup_container_spare_uuid(info
);
1779 /* I don't know how to compute 'map' on imsm, so use safe default */
1782 for (i
= 0; i
< map_disks
; i
++)
1788 /* allocates memory and fills disk in mdinfo structure
1789 * for each disk in array */
1790 struct mdinfo
*getinfo_super_disks_imsm(struct supertype
*st
)
1792 struct mdinfo
*mddev
= NULL
;
1793 struct intel_super
*super
= st
->sb
;
1794 struct imsm_disk
*disk
;
1797 if (!super
|| !super
->disks
)
1800 mddev
= malloc(sizeof(*mddev
));
1802 fprintf(stderr
, Name
": Failed to allocate memory.\n");
1805 memset(mddev
, 0, sizeof(*mddev
));
1809 tmp
= malloc(sizeof(*tmp
));
1811 fprintf(stderr
, Name
": Failed to allocate memory.\n");
1816 memset(tmp
, 0, sizeof(*tmp
));
1818 tmp
->next
= mddev
->devs
;
1820 tmp
->disk
.number
= count
++;
1821 tmp
->disk
.major
= dl
->major
;
1822 tmp
->disk
.minor
= dl
->minor
;
1823 tmp
->disk
.state
= is_configured(disk
) ?
1824 (1 << MD_DISK_ACTIVE
) : 0;
1825 tmp
->disk
.state
|= is_failed(disk
) ? (1 << MD_DISK_FAULTY
) : 0;
1826 tmp
->disk
.state
|= is_spare(disk
) ? 0 : (1 << MD_DISK_SYNC
);
1827 tmp
->disk
.raid_disk
= -1;
1833 static int update_super_imsm(struct supertype
*st
, struct mdinfo
*info
,
1834 char *update
, char *devname
, int verbose
,
1835 int uuid_set
, char *homehost
)
1837 /* For 'assemble' and 'force' we need to return non-zero if any
1838 * change was made. For others, the return value is ignored.
1839 * Update options are:
1840 * force-one : This device looks a bit old but needs to be included,
1841 * update age info appropriately.
1842 * assemble: clear any 'faulty' flag to allow this device to
1844 * force-array: Array is degraded but being forced, mark it clean
1845 * if that will be needed to assemble it.
1847 * newdev: not used ????
1848 * grow: Array has gained a new device - this is currently for
1850 * resync: mark as dirty so a resync will happen.
1851 * name: update the name - preserving the homehost
1852 * uuid: Change the uuid of the array to match watch is given
1854 * Following are not relevant for this imsm:
1855 * sparc2.2 : update from old dodgey metadata
1856 * super-minor: change the preferred_minor number
1857 * summaries: update redundant counters.
1858 * homehost: update the recorded homehost
1859 * _reshape_progress: record new reshape_progress position.
1862 struct intel_super
*super
= st
->sb
;
1863 struct imsm_super
*mpb
;
1865 /* we can only update container info */
1866 if (!super
|| super
->current_vol
>= 0 || !super
->anchor
)
1869 mpb
= super
->anchor
;
1871 if (strcmp(update
, "uuid") == 0 && uuid_set
&& !info
->update_private
)
1873 else if (strcmp(update
, "uuid") == 0 && uuid_set
&& info
->update_private
) {
1874 mpb
->orig_family_num
= *((__u32
*) info
->update_private
);
1876 } else if (strcmp(update
, "uuid") == 0) {
1877 __u32
*new_family
= malloc(sizeof(*new_family
));
1879 /* update orig_family_number with the incoming random
1880 * data, report the new effective uuid, and store the
1881 * new orig_family_num for future updates.
1884 memcpy(&mpb
->orig_family_num
, info
->uuid
, sizeof(__u32
));
1885 uuid_from_super_imsm(st
, info
->uuid
);
1886 *new_family
= mpb
->orig_family_num
;
1887 info
->update_private
= new_family
;
1890 } else if (strcmp(update
, "assemble") == 0)
1895 /* successful update? recompute checksum */
1897 mpb
->check_sum
= __le32_to_cpu(__gen_imsm_checksum(mpb
));
1902 static size_t disks_to_mpb_size(int disks
)
1906 size
= sizeof(struct imsm_super
);
1907 size
+= (disks
- 1) * sizeof(struct imsm_disk
);
1908 size
+= 2 * sizeof(struct imsm_dev
);
1909 /* up to 2 maps per raid device (-2 for imsm_maps in imsm_dev */
1910 size
+= (4 - 2) * sizeof(struct imsm_map
);
1911 /* 4 possible disk_ord_tbl's */
1912 size
+= 4 * (disks
- 1) * sizeof(__u32
);
1917 static __u64
avail_size_imsm(struct supertype
*st
, __u64 devsize
)
1919 if (devsize
< (MPB_SECTOR_CNT
+ IMSM_RESERVED_SECTORS
))
1922 return devsize
- (MPB_SECTOR_CNT
+ IMSM_RESERVED_SECTORS
);
1925 static void free_devlist(struct intel_super
*super
)
1927 struct intel_dev
*dv
;
1929 while (super
->devlist
) {
1930 dv
= super
->devlist
->next
;
1931 free(super
->devlist
->dev
);
1932 free(super
->devlist
);
1933 super
->devlist
= dv
;
1937 static void imsm_copy_dev(struct imsm_dev
*dest
, struct imsm_dev
*src
)
1939 memcpy(dest
, src
, sizeof_imsm_dev(src
, 0));
1942 static int compare_super_imsm(struct supertype
*st
, struct supertype
*tst
)
1946 * 0 same, or first was empty, and second was copied
1947 * 1 second had wrong number
1949 * 3 wrong other info
1951 struct intel_super
*first
= st
->sb
;
1952 struct intel_super
*sec
= tst
->sb
;
1960 /* if an anchor does not have num_raid_devs set then it is a free
1963 if (first
->anchor
->num_raid_devs
> 0 &&
1964 sec
->anchor
->num_raid_devs
> 0) {
1965 /* Determine if these disks might ever have been
1966 * related. Further disambiguation can only take place
1967 * in load_super_imsm_all
1969 __u32 first_family
= first
->anchor
->orig_family_num
;
1970 __u32 sec_family
= sec
->anchor
->orig_family_num
;
1972 if (memcmp(first
->anchor
->sig
, sec
->anchor
->sig
,
1973 MAX_SIGNATURE_LENGTH
) != 0)
1976 if (first_family
== 0)
1977 first_family
= first
->anchor
->family_num
;
1978 if (sec_family
== 0)
1979 sec_family
= sec
->anchor
->family_num
;
1981 if (first_family
!= sec_family
)
1987 /* if 'first' is a spare promote it to a populated mpb with sec's
1990 if (first
->anchor
->num_raid_devs
== 0 &&
1991 sec
->anchor
->num_raid_devs
> 0) {
1993 struct intel_dev
*dv
;
1994 struct imsm_dev
*dev
;
1996 /* we need to copy raid device info from sec if an allocation
1997 * fails here we don't associate the spare
1999 for (i
= 0; i
< sec
->anchor
->num_raid_devs
; i
++) {
2000 dv
= malloc(sizeof(*dv
));
2003 dev
= malloc(sizeof_imsm_dev(get_imsm_dev(sec
, i
), 1));
2010 dv
->next
= first
->devlist
;
2011 first
->devlist
= dv
;
2013 if (i
< sec
->anchor
->num_raid_devs
) {
2014 /* allocation failure */
2015 free_devlist(first
);
2016 fprintf(stderr
, "imsm: failed to associate spare\n");
2019 first
->anchor
->num_raid_devs
= sec
->anchor
->num_raid_devs
;
2020 first
->anchor
->orig_family_num
= sec
->anchor
->orig_family_num
;
2021 first
->anchor
->family_num
= sec
->anchor
->family_num
;
2022 memcpy(first
->anchor
->sig
, sec
->anchor
->sig
, MAX_SIGNATURE_LENGTH
);
2023 for (i
= 0; i
< sec
->anchor
->num_raid_devs
; i
++)
2024 imsm_copy_dev(get_imsm_dev(first
, i
), get_imsm_dev(sec
, i
));
2030 static void fd2devname(int fd
, char *name
)
2034 char dname
[PATH_MAX
];
2039 if (fstat(fd
, &st
) != 0)
2041 sprintf(path
, "/sys/dev/block/%d:%d",
2042 major(st
.st_rdev
), minor(st
.st_rdev
));
2044 rv
= readlink(path
, dname
, sizeof(dname
));
2049 nm
= strrchr(dname
, '/');
2051 snprintf(name
, MAX_RAID_SERIAL_LEN
, "/dev/%s", nm
);
2054 extern int scsi_get_serial(int fd
, void *buf
, size_t buf_len
);
2056 static int imsm_read_serial(int fd
, char *devname
,
2057 __u8 serial
[MAX_RAID_SERIAL_LEN
])
2059 unsigned char scsi_serial
[255];
2068 memset(scsi_serial
, 0, sizeof(scsi_serial
));
2070 rv
= scsi_get_serial(fd
, scsi_serial
, sizeof(scsi_serial
));
2072 if (rv
&& check_env("IMSM_DEVNAME_AS_SERIAL")) {
2073 memset(serial
, 0, MAX_RAID_SERIAL_LEN
);
2074 fd2devname(fd
, (char *) serial
);
2081 Name
": Failed to retrieve serial for %s\n",
2086 rsp_len
= scsi_serial
[3];
2090 Name
": Failed to retrieve serial for %s\n",
2094 rsp_buf
= (char *) &scsi_serial
[4];
2096 /* trim all whitespace and non-printable characters and convert
2099 for (i
= 0, dest
= rsp_buf
; i
< rsp_len
; i
++) {
2102 /* ':' is reserved for use in placeholder serial
2103 * numbers for missing disks
2111 len
= dest
- rsp_buf
;
2114 /* truncate leading characters */
2115 if (len
> MAX_RAID_SERIAL_LEN
) {
2116 dest
+= len
- MAX_RAID_SERIAL_LEN
;
2117 len
= MAX_RAID_SERIAL_LEN
;
2120 memset(serial
, 0, MAX_RAID_SERIAL_LEN
);
2121 memcpy(serial
, dest
, len
);
2126 static int serialcmp(__u8
*s1
, __u8
*s2
)
2128 return strncmp((char *) s1
, (char *) s2
, MAX_RAID_SERIAL_LEN
);
2131 static void serialcpy(__u8
*dest
, __u8
*src
)
2133 strncpy((char *) dest
, (char *) src
, MAX_RAID_SERIAL_LEN
);
2137 static struct dl
*serial_to_dl(__u8
*serial
, struct intel_super
*super
)
2141 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
2142 if (serialcmp(dl
->serial
, serial
) == 0)
2149 static struct imsm_disk
*
2150 __serial_to_disk(__u8
*serial
, struct imsm_super
*mpb
, int *idx
)
2154 for (i
= 0; i
< mpb
->num_disks
; i
++) {
2155 struct imsm_disk
*disk
= __get_imsm_disk(mpb
, i
);
2157 if (serialcmp(disk
->serial
, serial
) == 0) {
2168 load_imsm_disk(int fd
, struct intel_super
*super
, char *devname
, int keep_fd
)
2170 struct imsm_disk
*disk
;
2175 __u8 serial
[MAX_RAID_SERIAL_LEN
];
2177 rv
= imsm_read_serial(fd
, devname
, serial
);
2182 dl
= calloc(1, sizeof(*dl
));
2186 Name
": failed to allocate disk buffer for %s\n",
2192 dl
->major
= major(stb
.st_rdev
);
2193 dl
->minor
= minor(stb
.st_rdev
);
2194 dl
->next
= super
->disks
;
2195 dl
->fd
= keep_fd
? fd
: -1;
2196 assert(super
->disks
== NULL
);
2198 serialcpy(dl
->serial
, serial
);
2201 fd2devname(fd
, name
);
2203 dl
->devname
= strdup(devname
);
2205 dl
->devname
= strdup(name
);
2207 /* look up this disk's index in the current anchor */
2208 disk
= __serial_to_disk(dl
->serial
, super
->anchor
, &dl
->index
);
2211 /* only set index on disks that are a member of a
2212 * populated contianer, i.e. one with raid_devs
2214 if (is_failed(&dl
->disk
))
2216 else if (is_spare(&dl
->disk
))
2224 /* When migrating map0 contains the 'destination' state while map1
2225 * contains the current state. When not migrating map0 contains the
2226 * current state. This routine assumes that map[0].map_state is set to
2227 * the current array state before being called.
2229 * Migration is indicated by one of the following states
2230 * 1/ Idle (migr_state=0 map0state=normal||unitialized||degraded||failed)
2231 * 2/ Initialize (migr_state=1 migr_type=MIGR_INIT map0state=normal
2232 * map1state=unitialized)
2233 * 3/ Repair (Resync) (migr_state=1 migr_type=MIGR_REPAIR map0state=normal
2235 * 4/ Rebuild (migr_state=1 migr_type=MIGR_REBUILD map0state=normal
2236 * map1state=degraded)
2238 static void migrate(struct imsm_dev
*dev
, __u8 to_state
, int migr_type
)
2240 struct imsm_map
*dest
;
2241 struct imsm_map
*src
= get_imsm_map(dev
, 0);
2243 dev
->vol
.migr_state
= 1;
2244 set_migr_type(dev
, migr_type
);
2245 dev
->vol
.curr_migr_unit
= 0;
2246 dest
= get_imsm_map(dev
, 1);
2248 /* duplicate and then set the target end state in map[0] */
2249 memcpy(dest
, src
, sizeof_imsm_map(src
));
2250 if ((migr_type
== MIGR_REBUILD
) ||
2251 (migr_type
== MIGR_GEN_MIGR
)) {
2255 for (i
= 0; i
< src
->num_members
; i
++) {
2256 ord
= __le32_to_cpu(src
->disk_ord_tbl
[i
]);
2257 set_imsm_ord_tbl_ent(src
, i
, ord_to_idx(ord
));
2261 src
->map_state
= to_state
;
2264 static void end_migration(struct imsm_dev
*dev
, __u8 map_state
)
2266 struct imsm_map
*map
= get_imsm_map(dev
, 0);
2267 struct imsm_map
*prev
= get_imsm_map(dev
, dev
->vol
.migr_state
);
2270 /* merge any IMSM_ORD_REBUILD bits that were not successfully
2271 * completed in the last migration.
2273 * FIXME add support for raid-level-migration
2275 for (i
= 0; i
< prev
->num_members
; i
++)
2276 for (j
= 0; j
< map
->num_members
; j
++)
2277 /* during online capacity expansion
2278 * disks position can be changed if takeover is used
2280 if (ord_to_idx(map
->disk_ord_tbl
[j
]) ==
2281 ord_to_idx(prev
->disk_ord_tbl
[i
])) {
2282 map
->disk_ord_tbl
[j
] |= prev
->disk_ord_tbl
[i
];
2286 dev
->vol
.migr_state
= 0;
2287 dev
->vol
.migr_type
= 0;
2288 dev
->vol
.curr_migr_unit
= 0;
2289 map
->map_state
= map_state
;
2293 static int parse_raid_devices(struct intel_super
*super
)
2296 struct imsm_dev
*dev_new
;
2297 size_t len
, len_migr
;
2298 size_t space_needed
= 0;
2299 struct imsm_super
*mpb
= super
->anchor
;
2301 for (i
= 0; i
< super
->anchor
->num_raid_devs
; i
++) {
2302 struct imsm_dev
*dev_iter
= __get_imsm_dev(super
->anchor
, i
);
2303 struct intel_dev
*dv
;
2305 len
= sizeof_imsm_dev(dev_iter
, 0);
2306 len_migr
= sizeof_imsm_dev(dev_iter
, 1);
2308 space_needed
+= len_migr
- len
;
2310 dv
= malloc(sizeof(*dv
));
2313 dev_new
= malloc(len_migr
);
2318 imsm_copy_dev(dev_new
, dev_iter
);
2321 dv
->next
= super
->devlist
;
2322 super
->devlist
= dv
;
2325 /* ensure that super->buf is large enough when all raid devices
2328 if (__le32_to_cpu(mpb
->mpb_size
) + space_needed
> super
->len
) {
2331 len
= ROUND_UP(__le32_to_cpu(mpb
->mpb_size
) + space_needed
, 512);
2332 if (posix_memalign(&buf
, 512, len
) != 0)
2335 memcpy(buf
, super
->buf
, super
->len
);
2336 memset(buf
+ super
->len
, 0, len
- super
->len
);
2345 /* retrieve a pointer to the bbm log which starts after all raid devices */
2346 struct bbm_log
*__get_imsm_bbm_log(struct imsm_super
*mpb
)
2350 if (__le32_to_cpu(mpb
->bbm_log_size
)) {
2352 ptr
+= mpb
->mpb_size
- __le32_to_cpu(mpb
->bbm_log_size
);
2358 static void __free_imsm(struct intel_super
*super
, int free_disks
);
2360 /* load_imsm_mpb - read matrix metadata
2361 * allocates super->mpb to be freed by free_super
2363 static int load_imsm_mpb(int fd
, struct intel_super
*super
, char *devname
)
2365 unsigned long long dsize
;
2366 unsigned long long sectors
;
2368 struct imsm_super
*anchor
;
2371 get_dev_size(fd
, NULL
, &dsize
);
2375 Name
": %s: device to small for imsm\n",
2380 if (lseek64(fd
, dsize
- (512 * 2), SEEK_SET
) < 0) {
2383 Name
": Cannot seek to anchor block on %s: %s\n",
2384 devname
, strerror(errno
));
2388 if (posix_memalign((void**)&anchor
, 512, 512) != 0) {
2391 Name
": Failed to allocate imsm anchor buffer"
2392 " on %s\n", devname
);
2395 if (read(fd
, anchor
, 512) != 512) {
2398 Name
": Cannot read anchor block on %s: %s\n",
2399 devname
, strerror(errno
));
2404 if (strncmp((char *) anchor
->sig
, MPB_SIGNATURE
, MPB_SIG_LEN
) != 0) {
2407 Name
": no IMSM anchor on %s\n", devname
);
2412 __free_imsm(super
, 0);
2413 super
->len
= ROUND_UP(anchor
->mpb_size
, 512);
2414 if (posix_memalign(&super
->buf
, 512, super
->len
) != 0) {
2417 Name
": unable to allocate %zu byte mpb buffer\n",
2422 memcpy(super
->buf
, anchor
, 512);
2424 sectors
= mpb_sectors(anchor
) - 1;
2427 check_sum
= __gen_imsm_checksum(super
->anchor
);
2428 if (check_sum
!= __le32_to_cpu(super
->anchor
->check_sum
)) {
2431 Name
": IMSM checksum %x != %x on %s\n",
2433 __le32_to_cpu(super
->anchor
->check_sum
),
2441 /* read the extended mpb */
2442 if (lseek64(fd
, dsize
- (512 * (2 + sectors
)), SEEK_SET
) < 0) {
2445 Name
": Cannot seek to extended mpb on %s: %s\n",
2446 devname
, strerror(errno
));
2450 if ((unsigned)read(fd
, super
->buf
+ 512, super
->len
- 512) != super
->len
- 512) {
2453 Name
": Cannot read extended mpb on %s: %s\n",
2454 devname
, strerror(errno
));
2458 check_sum
= __gen_imsm_checksum(super
->anchor
);
2459 if (check_sum
!= __le32_to_cpu(super
->anchor
->check_sum
)) {
2462 Name
": IMSM checksum %x != %x on %s\n",
2463 check_sum
, __le32_to_cpu(super
->anchor
->check_sum
),
2468 /* FIXME the BBM log is disk specific so we cannot use this global
2469 * buffer for all disks. Ok for now since we only look at the global
2470 * bbm_log_size parameter to gate assembly
2472 super
->bbm_log
= __get_imsm_bbm_log(super
->anchor
);
2478 load_and_parse_mpb(int fd
, struct intel_super
*super
, char *devname
, int keep_fd
)
2482 err
= load_imsm_mpb(fd
, super
, devname
);
2485 err
= load_imsm_disk(fd
, super
, devname
, keep_fd
);
2488 err
= parse_raid_devices(super
);
2493 static void __free_imsm_disk(struct dl
*d
)
2505 static void free_imsm_disks(struct intel_super
*super
)
2509 while (super
->disks
) {
2511 super
->disks
= d
->next
;
2512 __free_imsm_disk(d
);
2514 while (super
->missing
) {
2516 super
->missing
= d
->next
;
2517 __free_imsm_disk(d
);
2522 /* free all the pieces hanging off of a super pointer */
2523 static void __free_imsm(struct intel_super
*super
, int free_disks
)
2530 free_imsm_disks(super
);
2531 free_devlist(super
);
2533 free((void *) super
->hba
);
2538 static void free_imsm(struct intel_super
*super
)
2540 __free_imsm(super
, 1);
2544 static void free_super_imsm(struct supertype
*st
)
2546 struct intel_super
*super
= st
->sb
;
2555 static struct intel_super
*alloc_super(void)
2557 struct intel_super
*super
= malloc(sizeof(*super
));
2560 memset(super
, 0, sizeof(*super
));
2561 super
->current_vol
= -1;
2562 super
->create_offset
= ~((__u32
) 0);
2563 if (!check_env("IMSM_NO_PLATFORM"))
2564 super
->orom
= find_imsm_orom();
2565 if (super
->orom
&& !check_env("IMSM_TEST_OROM")) {
2566 struct sys_dev
*list
, *ent
;
2568 /* find the first intel ahci controller */
2569 list
= find_driver_devices("pci", "ahci");
2570 for (ent
= list
; ent
; ent
= ent
->next
)
2571 if (devpath_to_vendor(ent
->path
) == 0x8086)
2574 super
->hba
= ent
->path
;
2577 free_sys_dev(&list
);
2585 /* find_missing - helper routine for load_super_imsm_all that identifies
2586 * disks that have disappeared from the system. This routine relies on
2587 * the mpb being uptodate, which it is at load time.
2589 static int find_missing(struct intel_super
*super
)
2592 struct imsm_super
*mpb
= super
->anchor
;
2594 struct imsm_disk
*disk
;
2596 for (i
= 0; i
< mpb
->num_disks
; i
++) {
2597 disk
= __get_imsm_disk(mpb
, i
);
2598 dl
= serial_to_dl(disk
->serial
, super
);
2602 dl
= malloc(sizeof(*dl
));
2608 dl
->devname
= strdup("missing");
2610 serialcpy(dl
->serial
, disk
->serial
);
2613 dl
->next
= super
->missing
;
2614 super
->missing
= dl
;
2620 static struct intel_disk
*disk_list_get(__u8
*serial
, struct intel_disk
*disk_list
)
2622 struct intel_disk
*idisk
= disk_list
;
2625 if (serialcmp(idisk
->disk
.serial
, serial
) == 0)
2627 idisk
= idisk
->next
;
2633 static int __prep_thunderdome(struct intel_super
**table
, int tbl_size
,
2634 struct intel_super
*super
,
2635 struct intel_disk
**disk_list
)
2637 struct imsm_disk
*d
= &super
->disks
->disk
;
2638 struct imsm_super
*mpb
= super
->anchor
;
2641 for (i
= 0; i
< tbl_size
; i
++) {
2642 struct imsm_super
*tbl_mpb
= table
[i
]->anchor
;
2643 struct imsm_disk
*tbl_d
= &table
[i
]->disks
->disk
;
2645 if (tbl_mpb
->family_num
== mpb
->family_num
) {
2646 if (tbl_mpb
->check_sum
== mpb
->check_sum
) {
2647 dprintf("%s: mpb from %d:%d matches %d:%d\n",
2648 __func__
, super
->disks
->major
,
2649 super
->disks
->minor
,
2650 table
[i
]->disks
->major
,
2651 table
[i
]->disks
->minor
);
2655 if (((is_configured(d
) && !is_configured(tbl_d
)) ||
2656 is_configured(d
) == is_configured(tbl_d
)) &&
2657 tbl_mpb
->generation_num
< mpb
->generation_num
) {
2658 /* current version of the mpb is a
2659 * better candidate than the one in
2660 * super_table, but copy over "cross
2661 * generational" status
2663 struct intel_disk
*idisk
;
2665 dprintf("%s: mpb from %d:%d replaces %d:%d\n",
2666 __func__
, super
->disks
->major
,
2667 super
->disks
->minor
,
2668 table
[i
]->disks
->major
,
2669 table
[i
]->disks
->minor
);
2671 idisk
= disk_list_get(tbl_d
->serial
, *disk_list
);
2672 if (idisk
&& is_failed(&idisk
->disk
))
2673 tbl_d
->status
|= FAILED_DISK
;
2676 struct intel_disk
*idisk
;
2677 struct imsm_disk
*disk
;
2679 /* tbl_mpb is more up to date, but copy
2680 * over cross generational status before
2683 disk
= __serial_to_disk(d
->serial
, mpb
, NULL
);
2684 if (disk
&& is_failed(disk
))
2685 d
->status
|= FAILED_DISK
;
2687 idisk
= disk_list_get(d
->serial
, *disk_list
);
2690 if (disk
&& is_configured(disk
))
2691 idisk
->disk
.status
|= CONFIGURED_DISK
;
2694 dprintf("%s: mpb from %d:%d prefer %d:%d\n",
2695 __func__
, super
->disks
->major
,
2696 super
->disks
->minor
,
2697 table
[i
]->disks
->major
,
2698 table
[i
]->disks
->minor
);
2706 table
[tbl_size
++] = super
;
2710 /* update/extend the merged list of imsm_disk records */
2711 for (j
= 0; j
< mpb
->num_disks
; j
++) {
2712 struct imsm_disk
*disk
= __get_imsm_disk(mpb
, j
);
2713 struct intel_disk
*idisk
;
2715 idisk
= disk_list_get(disk
->serial
, *disk_list
);
2717 idisk
->disk
.status
|= disk
->status
;
2718 if (is_configured(&idisk
->disk
) ||
2719 is_failed(&idisk
->disk
))
2720 idisk
->disk
.status
&= ~(SPARE_DISK
);
2722 idisk
= calloc(1, sizeof(*idisk
));
2725 idisk
->owner
= IMSM_UNKNOWN_OWNER
;
2726 idisk
->disk
= *disk
;
2727 idisk
->next
= *disk_list
;
2731 if (serialcmp(idisk
->disk
.serial
, d
->serial
) == 0)
2738 static struct intel_super
*
2739 validate_members(struct intel_super
*super
, struct intel_disk
*disk_list
,
2742 struct imsm_super
*mpb
= super
->anchor
;
2746 for (i
= 0; i
< mpb
->num_disks
; i
++) {
2747 struct imsm_disk
*disk
= __get_imsm_disk(mpb
, i
);
2748 struct intel_disk
*idisk
;
2750 idisk
= disk_list_get(disk
->serial
, disk_list
);
2752 if (idisk
->owner
== owner
||
2753 idisk
->owner
== IMSM_UNKNOWN_OWNER
)
2756 dprintf("%s: '%.16s' owner %d != %d\n",
2757 __func__
, disk
->serial
, idisk
->owner
,
2760 dprintf("%s: unknown disk %x [%d]: %.16s\n",
2761 __func__
, __le32_to_cpu(mpb
->family_num
), i
,
2767 if (ok_count
== mpb
->num_disks
)
2772 static void show_conflicts(__u32 family_num
, struct intel_super
*super_list
)
2774 struct intel_super
*s
;
2776 for (s
= super_list
; s
; s
= s
->next
) {
2777 if (family_num
!= s
->anchor
->family_num
)
2779 fprintf(stderr
, "Conflict, offlining family %#x on '%s'\n",
2780 __le32_to_cpu(family_num
), s
->disks
->devname
);
2784 static struct intel_super
*
2785 imsm_thunderdome(struct intel_super
**super_list
, int len
)
2787 struct intel_super
*super_table
[len
];
2788 struct intel_disk
*disk_list
= NULL
;
2789 struct intel_super
*champion
, *spare
;
2790 struct intel_super
*s
, **del
;
2795 memset(super_table
, 0, sizeof(super_table
));
2796 for (s
= *super_list
; s
; s
= s
->next
)
2797 tbl_size
= __prep_thunderdome(super_table
, tbl_size
, s
, &disk_list
);
2799 for (i
= 0; i
< tbl_size
; i
++) {
2800 struct imsm_disk
*d
;
2801 struct intel_disk
*idisk
;
2802 struct imsm_super
*mpb
= super_table
[i
]->anchor
;
2805 d
= &s
->disks
->disk
;
2807 /* 'd' must appear in merged disk list for its
2808 * configuration to be valid
2810 idisk
= disk_list_get(d
->serial
, disk_list
);
2811 if (idisk
&& idisk
->owner
== i
)
2812 s
= validate_members(s
, disk_list
, i
);
2817 dprintf("%s: marking family: %#x from %d:%d offline\n",
2818 __func__
, mpb
->family_num
,
2819 super_table
[i
]->disks
->major
,
2820 super_table
[i
]->disks
->minor
);
2824 /* This is where the mdadm implementation differs from the Windows
2825 * driver which has no strict concept of a container. We can only
2826 * assemble one family from a container, so when returning a prodigal
2827 * array member to this system the code will not be able to disambiguate
2828 * the container contents that should be assembled ("foreign" versus
2829 * "local"). It requires user intervention to set the orig_family_num
2830 * to a new value to establish a new container. The Windows driver in
2831 * this situation fixes up the volume name in place and manages the
2832 * foreign array as an independent entity.
2837 for (i
= 0; i
< tbl_size
; i
++) {
2838 struct intel_super
*tbl_ent
= super_table
[i
];
2844 if (tbl_ent
->anchor
->num_raid_devs
== 0) {
2849 if (s
&& !is_spare
) {
2850 show_conflicts(tbl_ent
->anchor
->family_num
, *super_list
);
2852 } else if (!s
&& !is_spare
)
2865 fprintf(stderr
, "Chose family %#x on '%s', "
2866 "assemble conflicts to new container with '--update=uuid'\n",
2867 __le32_to_cpu(s
->anchor
->family_num
), s
->disks
->devname
);
2869 /* collect all dl's onto 'champion', and update them to
2870 * champion's version of the status
2872 for (s
= *super_list
; s
; s
= s
->next
) {
2873 struct imsm_super
*mpb
= champion
->anchor
;
2874 struct dl
*dl
= s
->disks
;
2879 for (i
= 0; i
< mpb
->num_disks
; i
++) {
2880 struct imsm_disk
*disk
;
2882 disk
= __serial_to_disk(dl
->serial
, mpb
, &dl
->index
);
2885 /* only set index on disks that are a member of
2886 * a populated contianer, i.e. one with
2889 if (is_failed(&dl
->disk
))
2891 else if (is_spare(&dl
->disk
))
2897 if (i
>= mpb
->num_disks
) {
2898 struct intel_disk
*idisk
;
2900 idisk
= disk_list_get(dl
->serial
, disk_list
);
2901 if (idisk
&& is_spare(&idisk
->disk
) &&
2902 !is_failed(&idisk
->disk
) && !is_configured(&idisk
->disk
))
2910 dl
->next
= champion
->disks
;
2911 champion
->disks
= dl
;
2915 /* delete 'champion' from super_list */
2916 for (del
= super_list
; *del
; ) {
2917 if (*del
== champion
) {
2918 *del
= (*del
)->next
;
2921 del
= &(*del
)->next
;
2923 champion
->next
= NULL
;
2927 struct intel_disk
*idisk
= disk_list
;
2929 disk_list
= disk_list
->next
;
2936 static int load_super_imsm_all(struct supertype
*st
, int fd
, void **sbp
,
2940 struct intel_super
*super_list
= NULL
;
2941 struct intel_super
*super
= NULL
;
2942 int devnum
= fd2devnum(fd
);
2948 /* check if 'fd' an opened container */
2949 sra
= sysfs_read(fd
, 0, GET_LEVEL
|GET_VERSION
|GET_DEVS
|GET_STATE
);
2953 if (sra
->array
.major_version
!= -1 ||
2954 sra
->array
.minor_version
!= -2 ||
2955 strcmp(sra
->text_version
, "imsm") != 0) {
2960 for (sd
= sra
->devs
, i
= 0; sd
; sd
= sd
->next
, i
++) {
2961 struct intel_super
*s
= alloc_super();
2968 s
->next
= super_list
;
2972 sprintf(nm
, "%d:%d", sd
->disk
.major
, sd
->disk
.minor
);
2973 dfd
= dev_open(nm
, O_RDWR
);
2977 err
= load_and_parse_mpb(dfd
, s
, NULL
, 1);
2979 /* retry the load if we might have raced against mdmon */
2980 if (err
== 3 && mdmon_running(devnum
))
2981 for (retry
= 0; retry
< 3; retry
++) {
2983 err
= load_and_parse_mpb(dfd
, s
, NULL
, 1);
2991 /* all mpbs enter, maybe one leaves */
2992 super
= imsm_thunderdome(&super_list
, i
);
2998 if (find_missing(super
) != 0) {
3006 while (super_list
) {
3007 struct intel_super
*s
= super_list
;
3009 super_list
= super_list
->next
;
3018 st
->container_dev
= devnum
;
3019 if (err
== 0 && st
->ss
== NULL
) {
3020 st
->ss
= &super_imsm
;
3021 st
->minor_version
= 0;
3022 st
->max_devs
= IMSM_MAX_DEVICES
;
3027 static int load_container_imsm(struct supertype
*st
, int fd
, char *devname
)
3029 return load_super_imsm_all(st
, fd
, &st
->sb
, devname
);
3033 static int load_super_imsm(struct supertype
*st
, int fd
, char *devname
)
3035 struct intel_super
*super
;
3039 if (load_super_imsm_all(st
, fd
, &st
->sb
, devname
) == 0)
3043 if (test_partition(fd
))
3044 /* IMSM not allowed on partitions */
3047 free_super_imsm(st
);
3049 super
= alloc_super();
3052 Name
": malloc of %zu failed.\n",
3057 rv
= load_and_parse_mpb(fd
, super
, devname
, 0);
3062 Name
": Failed to load all information "
3063 "sections on %s\n", devname
);
3069 if (st
->ss
== NULL
) {
3070 st
->ss
= &super_imsm
;
3071 st
->minor_version
= 0;
3072 st
->max_devs
= IMSM_MAX_DEVICES
;
3077 static __u16
info_to_blocks_per_strip(mdu_array_info_t
*info
)
3079 if (info
->level
== 1)
3081 return info
->chunk_size
>> 9;
3084 static __u32
info_to_num_data_stripes(mdu_array_info_t
*info
, int num_domains
)
3088 num_stripes
= (info
->size
* 2) / info_to_blocks_per_strip(info
);
3089 num_stripes
/= num_domains
;
3094 static __u32
info_to_blocks_per_member(mdu_array_info_t
*info
)
3096 if (info
->level
== 1)
3097 return info
->size
* 2;
3099 return (info
->size
* 2) & ~(info_to_blocks_per_strip(info
) - 1);
3102 static void imsm_update_version_info(struct intel_super
*super
)
3104 /* update the version and attributes */
3105 struct imsm_super
*mpb
= super
->anchor
;
3107 struct imsm_dev
*dev
;
3108 struct imsm_map
*map
;
3111 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
3112 dev
= get_imsm_dev(super
, i
);
3113 map
= get_imsm_map(dev
, 0);
3114 if (__le32_to_cpu(dev
->size_high
) > 0)
3115 mpb
->attributes
|= MPB_ATTRIB_2TB
;
3117 /* FIXME detect when an array spans a port multiplier */
3119 mpb
->attributes
|= MPB_ATTRIB_PM
;
3122 if (mpb
->num_raid_devs
> 1 ||
3123 mpb
->attributes
!= MPB_ATTRIB_CHECKSUM_VERIFY
) {
3124 version
= MPB_VERSION_ATTRIBS
;
3125 switch (get_imsm_raid_level(map
)) {
3126 case 0: mpb
->attributes
|= MPB_ATTRIB_RAID0
; break;
3127 case 1: mpb
->attributes
|= MPB_ATTRIB_RAID1
; break;
3128 case 10: mpb
->attributes
|= MPB_ATTRIB_RAID10
; break;
3129 case 5: mpb
->attributes
|= MPB_ATTRIB_RAID5
; break;
3132 if (map
->num_members
>= 5)
3133 version
= MPB_VERSION_5OR6_DISK_ARRAY
;
3134 else if (dev
->status
== DEV_CLONE_N_GO
)
3135 version
= MPB_VERSION_CNG
;
3136 else if (get_imsm_raid_level(map
) == 5)
3137 version
= MPB_VERSION_RAID5
;
3138 else if (map
->num_members
>= 3)
3139 version
= MPB_VERSION_3OR4_DISK_ARRAY
;
3140 else if (get_imsm_raid_level(map
) == 1)
3141 version
= MPB_VERSION_RAID1
;
3143 version
= MPB_VERSION_RAID0
;
3145 strcpy(((char *) mpb
->sig
) + strlen(MPB_SIGNATURE
), version
);
3149 static int check_name(struct intel_super
*super
, char *name
, int quiet
)
3151 struct imsm_super
*mpb
= super
->anchor
;
3152 char *reason
= NULL
;
3155 if (strlen(name
) > MAX_RAID_SERIAL_LEN
)
3156 reason
= "must be 16 characters or less";
3158 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
3159 struct imsm_dev
*dev
= get_imsm_dev(super
, i
);
3161 if (strncmp((char *) dev
->volume
, name
, MAX_RAID_SERIAL_LEN
) == 0) {
3162 reason
= "already exists";
3167 if (reason
&& !quiet
)
3168 fprintf(stderr
, Name
": imsm volume name %s\n", reason
);
3173 static int init_super_imsm_volume(struct supertype
*st
, mdu_array_info_t
*info
,
3174 unsigned long long size
, char *name
,
3175 char *homehost
, int *uuid
)
3177 /* We are creating a volume inside a pre-existing container.
3178 * so st->sb is already set.
3180 struct intel_super
*super
= st
->sb
;
3181 struct imsm_super
*mpb
= super
->anchor
;
3182 struct intel_dev
*dv
;
3183 struct imsm_dev
*dev
;
3184 struct imsm_vol
*vol
;
3185 struct imsm_map
*map
;
3186 int idx
= mpb
->num_raid_devs
;
3188 unsigned long long array_blocks
;
3189 size_t size_old
, size_new
;
3190 __u32 num_data_stripes
;
3192 if (super
->orom
&& mpb
->num_raid_devs
>= super
->orom
->vpa
) {
3193 fprintf(stderr
, Name
": This imsm-container already has the "
3194 "maximum of %d volumes\n", super
->orom
->vpa
);
3198 /* ensure the mpb is large enough for the new data */
3199 size_old
= __le32_to_cpu(mpb
->mpb_size
);
3200 size_new
= disks_to_mpb_size(info
->nr_disks
);
3201 if (size_new
> size_old
) {
3203 size_t size_round
= ROUND_UP(size_new
, 512);
3205 if (posix_memalign(&mpb_new
, 512, size_round
) != 0) {
3206 fprintf(stderr
, Name
": could not allocate new mpb\n");
3209 memcpy(mpb_new
, mpb
, size_old
);
3212 super
->anchor
= mpb_new
;
3213 mpb
->mpb_size
= __cpu_to_le32(size_new
);
3214 memset(mpb_new
+ size_old
, 0, size_round
- size_old
);
3216 super
->current_vol
= idx
;
3217 /* when creating the first raid device in this container set num_disks
3218 * to zero, i.e. delete this spare and add raid member devices in
3219 * add_to_super_imsm_volume()
3221 if (super
->current_vol
== 0)
3224 if (!check_name(super
, name
, 0))
3226 dv
= malloc(sizeof(*dv
));
3228 fprintf(stderr
, Name
": failed to allocate device list entry\n");
3231 dev
= malloc(sizeof(*dev
) + sizeof(__u32
) * (info
->raid_disks
- 1));
3234 fprintf(stderr
, Name
": could not allocate raid device\n");
3237 strncpy((char *) dev
->volume
, name
, MAX_RAID_SERIAL_LEN
);
3238 if (info
->level
== 1)
3239 array_blocks
= info_to_blocks_per_member(info
);
3241 array_blocks
= calc_array_size(info
->level
, info
->raid_disks
,
3242 info
->layout
, info
->chunk_size
,
3244 /* round array size down to closest MB */
3245 array_blocks
= (array_blocks
>> SECT_PER_MB_SHIFT
) << SECT_PER_MB_SHIFT
;
3247 dev
->size_low
= __cpu_to_le32((__u32
) array_blocks
);
3248 dev
->size_high
= __cpu_to_le32((__u32
) (array_blocks
>> 32));
3249 dev
->status
= __cpu_to_le32(0);
3250 dev
->reserved_blocks
= __cpu_to_le32(0);
3252 vol
->migr_state
= 0;
3253 set_migr_type(dev
, MIGR_INIT
);
3255 vol
->curr_migr_unit
= 0;
3256 map
= get_imsm_map(dev
, 0);
3257 map
->pba_of_lba0
= __cpu_to_le32(super
->create_offset
);
3258 map
->blocks_per_member
= __cpu_to_le32(info_to_blocks_per_member(info
));
3259 map
->blocks_per_strip
= __cpu_to_le16(info_to_blocks_per_strip(info
));
3260 map
->failed_disk_num
= ~0;
3261 map
->map_state
= info
->level
? IMSM_T_STATE_UNINITIALIZED
:
3262 IMSM_T_STATE_NORMAL
;
3265 if (info
->level
== 1 && info
->raid_disks
> 2) {
3268 fprintf(stderr
, Name
": imsm does not support more than 2 disks"
3269 "in a raid1 volume\n");
3273 map
->raid_level
= info
->level
;
3274 if (info
->level
== 10) {
3275 map
->raid_level
= 1;
3276 map
->num_domains
= info
->raid_disks
/ 2;
3277 } else if (info
->level
== 1)
3278 map
->num_domains
= info
->raid_disks
;
3280 map
->num_domains
= 1;
3282 num_data_stripes
= info_to_num_data_stripes(info
, map
->num_domains
);
3283 map
->num_data_stripes
= __cpu_to_le32(num_data_stripes
);
3285 map
->num_members
= info
->raid_disks
;
3286 for (i
= 0; i
< map
->num_members
; i
++) {
3287 /* initialized in add_to_super */
3288 set_imsm_ord_tbl_ent(map
, i
, IMSM_ORD_REBUILD
);
3290 mpb
->num_raid_devs
++;
3293 dv
->index
= super
->current_vol
;
3294 dv
->next
= super
->devlist
;
3295 super
->devlist
= dv
;
3297 imsm_update_version_info(super
);
3302 static int init_super_imsm(struct supertype
*st
, mdu_array_info_t
*info
,
3303 unsigned long long size
, char *name
,
3304 char *homehost
, int *uuid
)
3306 /* This is primarily called by Create when creating a new array.
3307 * We will then get add_to_super called for each component, and then
3308 * write_init_super called to write it out to each device.
3309 * For IMSM, Create can create on fresh devices or on a pre-existing
3311 * To create on a pre-existing array a different method will be called.
3312 * This one is just for fresh drives.
3314 struct intel_super
*super
;
3315 struct imsm_super
*mpb
;
3320 return init_super_imsm_volume(st
, info
, size
, name
, homehost
, uuid
);
3323 mpb_size
= disks_to_mpb_size(info
->nr_disks
);
3327 super
= alloc_super();
3328 if (super
&& posix_memalign(&super
->buf
, 512, mpb_size
) != 0) {
3333 fprintf(stderr
, Name
3334 ": %s could not allocate superblock\n", __func__
);
3337 memset(super
->buf
, 0, mpb_size
);
3339 mpb
->mpb_size
= __cpu_to_le32(mpb_size
);
3343 /* zeroing superblock */
3347 mpb
->attributes
= MPB_ATTRIB_CHECKSUM_VERIFY
;
3349 version
= (char *) mpb
->sig
;
3350 strcpy(version
, MPB_SIGNATURE
);
3351 version
+= strlen(MPB_SIGNATURE
);
3352 strcpy(version
, MPB_VERSION_RAID0
);
3358 static int add_to_super_imsm_volume(struct supertype
*st
, mdu_disk_info_t
*dk
,
3359 int fd
, char *devname
)
3361 struct intel_super
*super
= st
->sb
;
3362 struct imsm_super
*mpb
= super
->anchor
;
3364 struct imsm_dev
*dev
;
3365 struct imsm_map
*map
;
3368 dev
= get_imsm_dev(super
, super
->current_vol
);
3369 map
= get_imsm_map(dev
, 0);
3371 if (! (dk
->state
& (1<<MD_DISK_SYNC
))) {
3372 fprintf(stderr
, Name
": %s: Cannot add spare devices to IMSM volume\n",
3378 /* we're doing autolayout so grab the pre-marked (in
3379 * validate_geometry) raid_disk
3381 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
3382 if (dl
->raiddisk
== dk
->raid_disk
)
3385 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
3386 if (dl
->major
== dk
->major
&&
3387 dl
->minor
== dk
->minor
)
3392 fprintf(stderr
, Name
": %s is not a member of the same container\n", devname
);
3396 /* add a pristine spare to the metadata */
3397 if (dl
->index
< 0) {
3398 dl
->index
= super
->anchor
->num_disks
;
3399 super
->anchor
->num_disks
++;
3401 /* Check the device has not already been added */
3402 slot
= get_imsm_disk_slot(map
, dl
->index
);
3404 (get_imsm_ord_tbl_ent(dev
, slot
) & IMSM_ORD_REBUILD
) == 0) {
3405 fprintf(stderr
, Name
": %s has been included in this array twice\n",
3409 set_imsm_ord_tbl_ent(map
, dk
->number
, dl
->index
);
3410 dl
->disk
.status
= CONFIGURED_DISK
;
3412 /* if we are creating the first raid device update the family number */
3413 if (super
->current_vol
== 0) {
3415 struct imsm_dev
*_dev
= __get_imsm_dev(mpb
, 0);
3416 struct imsm_disk
*_disk
= __get_imsm_disk(mpb
, dl
->index
);
3418 if (!_dev
|| !_disk
) {
3419 fprintf(stderr
, Name
": BUG mpb setup error\n");
3425 sum
+= __gen_imsm_checksum(mpb
);
3426 mpb
->family_num
= __cpu_to_le32(sum
);
3427 mpb
->orig_family_num
= mpb
->family_num
;
3433 static int add_to_super_imsm(struct supertype
*st
, mdu_disk_info_t
*dk
,
3434 int fd
, char *devname
)
3436 struct intel_super
*super
= st
->sb
;
3438 unsigned long long size
;
3443 /* if we are on an RAID enabled platform check that the disk is
3444 * attached to the raid controller
3446 if (super
->hba
&& !disk_attached_to_hba(fd
, super
->hba
)) {
3448 Name
": %s is not attached to the raid controller: %s\n",
3449 devname
? : "disk", super
->hba
);
3453 if (super
->current_vol
>= 0)
3454 return add_to_super_imsm_volume(st
, dk
, fd
, devname
);
3457 dd
= malloc(sizeof(*dd
));
3460 Name
": malloc failed %s:%d.\n", __func__
, __LINE__
);
3463 memset(dd
, 0, sizeof(*dd
));
3464 dd
->major
= major(stb
.st_rdev
);
3465 dd
->minor
= minor(stb
.st_rdev
);
3467 dd
->devname
= devname
? strdup(devname
) : NULL
;
3470 dd
->action
= DISK_ADD
;
3471 rv
= imsm_read_serial(fd
, devname
, dd
->serial
);
3474 Name
": failed to retrieve scsi serial, aborting\n");
3479 get_dev_size(fd
, NULL
, &size
);
3481 serialcpy(dd
->disk
.serial
, dd
->serial
);
3482 dd
->disk
.total_blocks
= __cpu_to_le32(size
);
3483 dd
->disk
.status
= SPARE_DISK
;
3484 if (sysfs_disk_to_scsi_id(fd
, &id
) == 0)
3485 dd
->disk
.scsi_id
= __cpu_to_le32(id
);
3487 dd
->disk
.scsi_id
= __cpu_to_le32(0);
3489 if (st
->update_tail
) {
3490 dd
->next
= super
->disk_mgmt_list
;
3491 super
->disk_mgmt_list
= dd
;
3493 dd
->next
= super
->disks
;
3501 static int remove_from_super_imsm(struct supertype
*st
, mdu_disk_info_t
*dk
)
3503 struct intel_super
*super
= st
->sb
;
3506 /* remove from super works only in mdmon - for communication
3507 * manager - monitor. Check if communication memory buffer
3510 if (!st
->update_tail
) {
3512 Name
": %s shall be used in mdmon context only"
3513 "(line %d).\n", __func__
, __LINE__
);
3516 dd
= malloc(sizeof(*dd
));
3519 Name
": malloc failed %s:%d.\n", __func__
, __LINE__
);
3522 memset(dd
, 0, sizeof(*dd
));
3523 dd
->major
= dk
->major
;
3524 dd
->minor
= dk
->minor
;
3527 dd
->disk
.status
= SPARE_DISK
;
3528 dd
->action
= DISK_REMOVE
;
3530 dd
->next
= super
->disk_mgmt_list
;
3531 super
->disk_mgmt_list
= dd
;
3537 static int store_imsm_mpb(int fd
, struct imsm_super
*mpb
);
3541 struct imsm_super anchor
;
3542 } spare_record
__attribute__ ((aligned(512)));
3544 /* spare records have their own family number and do not have any defined raid
3547 static int write_super_imsm_spares(struct intel_super
*super
, int doclose
)
3549 struct imsm_super
*mpb
= super
->anchor
;
3550 struct imsm_super
*spare
= &spare_record
.anchor
;
3554 spare
->mpb_size
= __cpu_to_le32(sizeof(struct imsm_super
)),
3555 spare
->generation_num
= __cpu_to_le32(1UL),
3556 spare
->attributes
= MPB_ATTRIB_CHECKSUM_VERIFY
;
3557 spare
->num_disks
= 1,
3558 spare
->num_raid_devs
= 0,
3559 spare
->cache_size
= mpb
->cache_size
,
3560 spare
->pwr_cycle_count
= __cpu_to_le32(1),
3562 snprintf((char *) spare
->sig
, MAX_SIGNATURE_LENGTH
,
3563 MPB_SIGNATURE MPB_VERSION_RAID0
);
3565 for (d
= super
->disks
; d
; d
= d
->next
) {
3569 spare
->disk
[0] = d
->disk
;
3570 sum
= __gen_imsm_checksum(spare
);
3571 spare
->family_num
= __cpu_to_le32(sum
);
3572 spare
->orig_family_num
= 0;
3573 sum
= __gen_imsm_checksum(spare
);
3574 spare
->check_sum
= __cpu_to_le32(sum
);
3576 if (store_imsm_mpb(d
->fd
, spare
)) {
3577 fprintf(stderr
, "%s: failed for device %d:%d %s\n",
3578 __func__
, d
->major
, d
->minor
, strerror(errno
));
3590 static int write_super_imsm(struct supertype
*st
, int doclose
)
3592 struct intel_super
*super
= st
->sb
;
3593 struct imsm_super
*mpb
= super
->anchor
;
3599 __u32 mpb_size
= sizeof(struct imsm_super
) - sizeof(struct imsm_disk
);
3602 /* 'generation' is incremented everytime the metadata is written */
3603 generation
= __le32_to_cpu(mpb
->generation_num
);
3605 mpb
->generation_num
= __cpu_to_le32(generation
);
3607 /* fix up cases where previous mdadm releases failed to set
3610 if (mpb
->orig_family_num
== 0)
3611 mpb
->orig_family_num
= mpb
->family_num
;
3613 for (d
= super
->disks
; d
; d
= d
->next
) {
3617 mpb
->disk
[d
->index
] = d
->disk
;
3621 for (d
= super
->missing
; d
; d
= d
->next
) {
3622 mpb
->disk
[d
->index
] = d
->disk
;
3625 mpb
->num_disks
= num_disks
;
3626 mpb_size
+= sizeof(struct imsm_disk
) * mpb
->num_disks
;
3628 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
3629 struct imsm_dev
*dev
= __get_imsm_dev(mpb
, i
);
3630 struct imsm_dev
*dev2
= get_imsm_dev(super
, i
);
3632 imsm_copy_dev(dev
, dev2
);
3633 mpb_size
+= sizeof_imsm_dev(dev
, 0);
3636 mpb_size
+= __le32_to_cpu(mpb
->bbm_log_size
);
3637 mpb
->mpb_size
= __cpu_to_le32(mpb_size
);
3639 /* recalculate checksum */
3640 sum
= __gen_imsm_checksum(mpb
);
3641 mpb
->check_sum
= __cpu_to_le32(sum
);
3643 /* write the mpb for disks that compose raid devices */
3644 for (d
= super
->disks
; d
; d
= d
->next
) {
3647 if (store_imsm_mpb(d
->fd
, mpb
))
3648 fprintf(stderr
, "%s: failed for device %d:%d %s\n",
3649 __func__
, d
->major
, d
->minor
, strerror(errno
));
3657 return write_super_imsm_spares(super
, doclose
);
3663 static int create_array(struct supertype
*st
, int dev_idx
)
3666 struct imsm_update_create_array
*u
;
3667 struct intel_super
*super
= st
->sb
;
3668 struct imsm_dev
*dev
= get_imsm_dev(super
, dev_idx
);
3669 struct imsm_map
*map
= get_imsm_map(dev
, 0);
3670 struct disk_info
*inf
;
3671 struct imsm_disk
*disk
;
3674 len
= sizeof(*u
) - sizeof(*dev
) + sizeof_imsm_dev(dev
, 0) +
3675 sizeof(*inf
) * map
->num_members
;
3678 fprintf(stderr
, "%s: failed to allocate update buffer\n",
3683 u
->type
= update_create_array
;
3684 u
->dev_idx
= dev_idx
;
3685 imsm_copy_dev(&u
->dev
, dev
);
3686 inf
= get_disk_info(u
);
3687 for (i
= 0; i
< map
->num_members
; i
++) {
3688 int idx
= get_imsm_disk_idx(dev
, i
);
3690 disk
= get_imsm_disk(super
, idx
);
3691 serialcpy(inf
[i
].serial
, disk
->serial
);
3693 append_metadata_update(st
, u
, len
);
3698 static int mgmt_disk(struct supertype
*st
)
3700 struct intel_super
*super
= st
->sb
;
3702 struct imsm_update_add_remove_disk
*u
;
3704 if (!super
->disk_mgmt_list
)
3710 fprintf(stderr
, "%s: failed to allocate update buffer\n",
3715 u
->type
= update_add_remove_disk
;
3716 append_metadata_update(st
, u
, len
);
3721 static int write_init_super_imsm(struct supertype
*st
)
3723 struct intel_super
*super
= st
->sb
;
3724 int current_vol
= super
->current_vol
;
3726 /* we are done with current_vol reset it to point st at the container */
3727 super
->current_vol
= -1;
3729 if (st
->update_tail
) {
3730 /* queue the recently created array / added disk
3731 * as a metadata update */
3735 /* determine if we are creating a volume or adding a disk */
3736 if (current_vol
< 0) {
3737 /* in the mgmt (add/remove) disk case we are running
3738 * in mdmon context, so don't close fd's
3740 return mgmt_disk(st
);
3742 rv
= create_array(st
, current_vol
);
3744 for (d
= super
->disks
; d
; d
= d
->next
) {
3752 for (d
= super
->disks
; d
; d
= d
->next
)
3753 Kill(d
->devname
, NULL
, 0, 1, 1);
3754 return write_super_imsm(st
, 1);
3759 static int store_super_imsm(struct supertype
*st
, int fd
)
3761 struct intel_super
*super
= st
->sb
;
3762 struct imsm_super
*mpb
= super
? super
->anchor
: NULL
;
3768 return store_imsm_mpb(fd
, mpb
);
3774 static int imsm_bbm_log_size(struct imsm_super
*mpb
)
3776 return __le32_to_cpu(mpb
->bbm_log_size
);
3780 static int validate_geometry_imsm_container(struct supertype
*st
, int level
,
3781 int layout
, int raiddisks
, int chunk
,
3782 unsigned long long size
, char *dev
,
3783 unsigned long long *freesize
,
3787 unsigned long long ldsize
;
3788 const struct imsm_orom
*orom
;
3790 if (level
!= LEVEL_CONTAINER
)
3795 if (check_env("IMSM_NO_PLATFORM"))
3798 orom
= find_imsm_orom();
3799 if (orom
&& raiddisks
> orom
->tds
) {
3801 fprintf(stderr
, Name
": %d exceeds maximum number of"
3802 " platform supported disks: %d\n",
3803 raiddisks
, orom
->tds
);
3807 fd
= open(dev
, O_RDONLY
|O_EXCL
, 0);
3810 fprintf(stderr
, Name
": imsm: Cannot open %s: %s\n",
3811 dev
, strerror(errno
));
3814 if (!get_dev_size(fd
, dev
, &ldsize
)) {
3820 *freesize
= avail_size_imsm(st
, ldsize
>> 9);
3825 static unsigned long long find_size(struct extent
*e
, int *idx
, int num_extents
)
3827 const unsigned long long base_start
= e
[*idx
].start
;
3828 unsigned long long end
= base_start
+ e
[*idx
].size
;
3831 if (base_start
== end
)
3835 for (i
= *idx
; i
< num_extents
; i
++) {
3836 /* extend overlapping extents */
3837 if (e
[i
].start
>= base_start
&&
3838 e
[i
].start
<= end
) {
3841 if (e
[i
].start
+ e
[i
].size
> end
)
3842 end
= e
[i
].start
+ e
[i
].size
;
3843 } else if (e
[i
].start
> end
) {
3849 return end
- base_start
;
3852 static unsigned long long merge_extents(struct intel_super
*super
, int sum_extents
)
3854 /* build a composite disk with all known extents and generate a new
3855 * 'maxsize' given the "all disks in an array must share a common start
3856 * offset" constraint
3858 struct extent
*e
= calloc(sum_extents
, sizeof(*e
));
3862 unsigned long long pos
;
3863 unsigned long long start
= 0;
3864 unsigned long long maxsize
;
3865 unsigned long reserve
;
3870 /* coalesce and sort all extents. also, check to see if we need to
3871 * reserve space between member arrays
3874 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
3877 for (i
= 0; i
< dl
->extent_cnt
; i
++)
3880 qsort(e
, sum_extents
, sizeof(*e
), cmp_extent
);
3885 while (i
< sum_extents
) {
3886 e
[j
].start
= e
[i
].start
;
3887 e
[j
].size
= find_size(e
, &i
, sum_extents
);
3889 if (e
[j
-1].size
== 0)
3898 unsigned long long esize
;
3900 esize
= e
[i
].start
- pos
;
3901 if (esize
>= maxsize
) {
3906 pos
= e
[i
].start
+ e
[i
].size
;
3908 } while (e
[i
-1].size
);
3914 /* FIXME assumes volume at offset 0 is the first volume in a
3917 if (start_extent
> 0)
3918 reserve
= IMSM_RESERVED_SECTORS
; /* gap between raid regions */
3922 if (maxsize
< reserve
)
3925 super
->create_offset
= ~((__u32
) 0);
3926 if (start
+ reserve
> super
->create_offset
)
3927 return 0; /* start overflows create_offset */
3928 super
->create_offset
= start
+ reserve
;
3930 return maxsize
- reserve
;
3933 static int is_raid_level_supported(const struct imsm_orom
*orom
, int level
, int raiddisks
)
3935 if (level
< 0 || level
== 6 || level
== 4)
3938 /* if we have an orom prevent invalid raid levels */
3941 case 0: return imsm_orom_has_raid0(orom
);
3944 return imsm_orom_has_raid1e(orom
);
3945 return imsm_orom_has_raid1(orom
) && raiddisks
== 2;
3946 case 10: return imsm_orom_has_raid10(orom
) && raiddisks
== 4;
3947 case 5: return imsm_orom_has_raid5(orom
) && raiddisks
> 2;
3950 return 1; /* not on an Intel RAID platform so anything goes */
3955 #define pr_vrb(fmt, arg...) (void) (verbose && fprintf(stderr, Name fmt, ##arg))
3957 validate_geometry_imsm_orom(struct intel_super
*super
, int level
, int layout
,
3958 int raiddisks
, int chunk
, int verbose
)
3960 if (!is_raid_level_supported(super
->orom
, level
, raiddisks
)) {
3961 pr_vrb(": platform does not support raid%d with %d disk%s\n",
3962 level
, raiddisks
, raiddisks
> 1 ? "s" : "");
3965 if (super
->orom
&& level
!= 1 &&
3966 !imsm_orom_has_chunk(super
->orom
, chunk
)) {
3967 pr_vrb(": platform does not support a chunk size of: %d\n", chunk
);
3970 if (layout
!= imsm_level_to_layout(level
)) {
3972 pr_vrb(": imsm raid 5 only supports the left-asymmetric layout\n");
3973 else if (level
== 10)
3974 pr_vrb(": imsm raid 10 only supports the n2 layout\n");
3976 pr_vrb(": imsm unknown layout %#x for this raid level %d\n",
3984 /* validate_geometry_imsm_volume - lifted from validate_geometry_ddf_bvd
3985 * FIX ME add ahci details
3987 static int validate_geometry_imsm_volume(struct supertype
*st
, int level
,
3988 int layout
, int raiddisks
, int chunk
,
3989 unsigned long long size
, char *dev
,
3990 unsigned long long *freesize
,
3994 struct intel_super
*super
= st
->sb
;
3995 struct imsm_super
*mpb
= super
->anchor
;
3997 unsigned long long pos
= 0;
3998 unsigned long long maxsize
;
4002 /* We must have the container info already read in. */
4006 if (!validate_geometry_imsm_orom(super
, level
, layout
, raiddisks
, chunk
, verbose
))
4010 /* General test: make sure there is space for
4011 * 'raiddisks' device extents of size 'size' at a given
4014 unsigned long long minsize
= size
;
4015 unsigned long long start_offset
= MaxSector
;
4018 minsize
= MPB_SECTOR_CNT
+ IMSM_RESERVED_SECTORS
;
4019 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
4024 e
= get_extents(super
, dl
);
4027 unsigned long long esize
;
4028 esize
= e
[i
].start
- pos
;
4029 if (esize
>= minsize
)
4031 if (found
&& start_offset
== MaxSector
) {
4034 } else if (found
&& pos
!= start_offset
) {
4038 pos
= e
[i
].start
+ e
[i
].size
;
4040 } while (e
[i
-1].size
);
4045 if (dcnt
< raiddisks
) {
4047 fprintf(stderr
, Name
": imsm: Not enough "
4048 "devices with space for this array "
4056 /* This device must be a member of the set */
4057 if (stat(dev
, &stb
) < 0)
4059 if ((S_IFMT
& stb
.st_mode
) != S_IFBLK
)
4061 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
4062 if (dl
->major
== (int)major(stb
.st_rdev
) &&
4063 dl
->minor
== (int)minor(stb
.st_rdev
))
4068 fprintf(stderr
, Name
": %s is not in the "
4069 "same imsm set\n", dev
);
4071 } else if (super
->orom
&& dl
->index
< 0 && mpb
->num_raid_devs
) {
4072 /* If a volume is present then the current creation attempt
4073 * cannot incorporate new spares because the orom may not
4074 * understand this configuration (all member disks must be
4075 * members of each array in the container).
4077 fprintf(stderr
, Name
": %s is a spare and a volume"
4078 " is already defined for this container\n", dev
);
4079 fprintf(stderr
, Name
": The option-rom requires all member"
4080 " disks to be a member of all volumes\n");
4084 /* retrieve the largest free space block */
4085 e
= get_extents(super
, dl
);
4090 unsigned long long esize
;
4092 esize
= e
[i
].start
- pos
;
4093 if (esize
>= maxsize
)
4095 pos
= e
[i
].start
+ e
[i
].size
;
4097 } while (e
[i
-1].size
);
4102 fprintf(stderr
, Name
": unable to determine free space for: %s\n",
4106 if (maxsize
< size
) {
4108 fprintf(stderr
, Name
": %s not enough space (%llu < %llu)\n",
4109 dev
, maxsize
, size
);
4113 /* count total number of extents for merge */
4115 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
4117 i
+= dl
->extent_cnt
;
4119 maxsize
= merge_extents(super
, i
);
4120 if (maxsize
< size
|| maxsize
== 0) {
4122 fprintf(stderr
, Name
": not enough space after merge (%llu < %llu)\n",
4127 *freesize
= maxsize
;
4132 static int reserve_space(struct supertype
*st
, int raiddisks
,
4133 unsigned long long size
, int chunk
,
4134 unsigned long long *freesize
)
4136 struct intel_super
*super
= st
->sb
;
4137 struct imsm_super
*mpb
= super
->anchor
;
4142 unsigned long long maxsize
;
4143 unsigned long long minsize
;
4147 /* find the largest common start free region of the possible disks */
4151 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
4157 /* don't activate new spares if we are orom constrained
4158 * and there is already a volume active in the container
4160 if (super
->orom
&& dl
->index
< 0 && mpb
->num_raid_devs
)
4163 e
= get_extents(super
, dl
);
4166 for (i
= 1; e
[i
-1].size
; i
++)
4174 maxsize
= merge_extents(super
, extent_cnt
);
4179 if (cnt
< raiddisks
||
4180 (super
->orom
&& used
&& used
!= raiddisks
) ||
4181 maxsize
< minsize
||
4183 fprintf(stderr
, Name
": not enough devices with space to create array.\n");
4184 return 0; /* No enough free spaces large enough */
4196 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
4198 dl
->raiddisk
= cnt
++;
4205 static int validate_geometry_imsm(struct supertype
*st
, int level
, int layout
,
4206 int raiddisks
, int chunk
, unsigned long long size
,
4207 char *dev
, unsigned long long *freesize
,
4214 /* if given unused devices create a container
4215 * if given given devices in a container create a member volume
4217 if (level
== LEVEL_CONTAINER
) {
4218 /* Must be a fresh device to add to a container */
4219 return validate_geometry_imsm_container(st
, level
, layout
,
4220 raiddisks
, chunk
, size
,
4226 if (st
->sb
&& freesize
) {
4227 /* we are being asked to automatically layout a
4228 * new volume based on the current contents of
4229 * the container. If the the parameters can be
4230 * satisfied reserve_space will record the disks,
4231 * start offset, and size of the volume to be
4232 * created. add_to_super and getinfo_super
4233 * detect when autolayout is in progress.
4235 if (!validate_geometry_imsm_orom(st
->sb
, level
, layout
,
4239 return reserve_space(st
, raiddisks
, size
, chunk
, freesize
);
4244 /* creating in a given container */
4245 return validate_geometry_imsm_volume(st
, level
, layout
,
4246 raiddisks
, chunk
, size
,
4247 dev
, freesize
, verbose
);
4250 /* This device needs to be a device in an 'imsm' container */
4251 fd
= open(dev
, O_RDONLY
|O_EXCL
, 0);
4255 Name
": Cannot create this array on device %s\n",
4260 if (errno
!= EBUSY
|| (fd
= open(dev
, O_RDONLY
, 0)) < 0) {
4262 fprintf(stderr
, Name
": Cannot open %s: %s\n",
4263 dev
, strerror(errno
));
4266 /* Well, it is in use by someone, maybe an 'imsm' container. */
4267 cfd
= open_container(fd
);
4271 fprintf(stderr
, Name
": Cannot use %s: It is busy\n",
4275 sra
= sysfs_read(cfd
, 0, GET_VERSION
);
4276 if (sra
&& sra
->array
.major_version
== -1 &&
4277 strcmp(sra
->text_version
, "imsm") == 0)
4281 /* This is a member of a imsm container. Load the container
4282 * and try to create a volume
4284 struct intel_super
*super
;
4286 if (load_super_imsm_all(st
, cfd
, (void **) &super
, NULL
) == 0) {
4288 st
->container_dev
= fd2devnum(cfd
);
4290 return validate_geometry_imsm_volume(st
, level
, layout
,
4298 fprintf(stderr
, Name
": failed container membership check\n");
4304 static void default_geometry_imsm(struct supertype
*st
, int *level
, int *layout
, int *chunk
)
4306 struct intel_super
*super
= st
->sb
;
4308 if (level
&& *level
== UnSet
)
4309 *level
= LEVEL_CONTAINER
;
4311 if (level
&& layout
&& *layout
== UnSet
)
4312 *layout
= imsm_level_to_layout(*level
);
4314 if (chunk
&& (*chunk
== UnSet
|| *chunk
== 0) &&
4315 super
&& super
->orom
)
4316 *chunk
= imsm_orom_default_chunk(super
->orom
);
4319 static void handle_missing(struct intel_super
*super
, struct imsm_dev
*dev
);
4321 static int kill_subarray_imsm(struct supertype
*st
)
4323 /* remove the subarray currently referenced by ->current_vol */
4325 struct intel_dev
**dp
;
4326 struct intel_super
*super
= st
->sb
;
4327 __u8 current_vol
= super
->current_vol
;
4328 struct imsm_super
*mpb
= super
->anchor
;
4330 if (super
->current_vol
< 0)
4332 super
->current_vol
= -1; /* invalidate subarray cursor */
4334 /* block deletions that would change the uuid of active subarrays
4336 * FIXME when immutable ids are available, but note that we'll
4337 * also need to fixup the invalidated/active subarray indexes in
4340 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
4343 if (i
< current_vol
)
4345 sprintf(subarray
, "%u", i
);
4346 if (is_subarray_active(subarray
, st
->devname
)) {
4348 Name
": deleting subarray-%d would change the UUID of active subarray-%d, aborting\n",
4355 if (st
->update_tail
) {
4356 struct imsm_update_kill_array
*u
= malloc(sizeof(*u
));
4360 u
->type
= update_kill_array
;
4361 u
->dev_idx
= current_vol
;
4362 append_metadata_update(st
, u
, sizeof(*u
));
4367 for (dp
= &super
->devlist
; *dp
;)
4368 if ((*dp
)->index
== current_vol
) {
4371 handle_missing(super
, (*dp
)->dev
);
4372 if ((*dp
)->index
> current_vol
)
4377 /* no more raid devices, all active components are now spares,
4378 * but of course failed are still failed
4380 if (--mpb
->num_raid_devs
== 0) {
4383 for (d
= super
->disks
; d
; d
= d
->next
)
4384 if (d
->index
> -2) {
4386 d
->disk
.status
= SPARE_DISK
;
4390 super
->updates_pending
++;
4395 static int update_subarray_imsm(struct supertype
*st
, char *subarray
,
4396 char *update
, struct mddev_ident
*ident
)
4398 /* update the subarray currently referenced by ->current_vol */
4399 struct intel_super
*super
= st
->sb
;
4400 struct imsm_super
*mpb
= super
->anchor
;
4402 if (strcmp(update
, "name") == 0) {
4403 char *name
= ident
->name
;
4407 if (is_subarray_active(subarray
, st
->devname
)) {
4409 Name
": Unable to update name of active subarray\n");
4413 if (!check_name(super
, name
, 0))
4416 vol
= strtoul(subarray
, &ep
, 10);
4417 if (*ep
!= '\0' || vol
>= super
->anchor
->num_raid_devs
)
4420 if (st
->update_tail
) {
4421 struct imsm_update_rename_array
*u
= malloc(sizeof(*u
));
4425 u
->type
= update_rename_array
;
4427 snprintf((char *) u
->name
, MAX_RAID_SERIAL_LEN
, "%s", name
);
4428 append_metadata_update(st
, u
, sizeof(*u
));
4430 struct imsm_dev
*dev
;
4433 dev
= get_imsm_dev(super
, vol
);
4434 snprintf((char *) dev
->volume
, MAX_RAID_SERIAL_LEN
, "%s", name
);
4435 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
4436 dev
= get_imsm_dev(super
, i
);
4437 handle_missing(super
, dev
);
4439 super
->updates_pending
++;
4446 #endif /* MDASSEMBLE */
4448 static int is_gen_migration(struct imsm_dev
*dev
)
4450 if (!dev
->vol
.migr_state
)
4453 if (migr_type(dev
) == MIGR_GEN_MIGR
)
4459 static int is_rebuilding(struct imsm_dev
*dev
)
4461 struct imsm_map
*migr_map
;
4463 if (!dev
->vol
.migr_state
)
4466 if (migr_type(dev
) != MIGR_REBUILD
)
4469 migr_map
= get_imsm_map(dev
, 1);
4471 if (migr_map
->map_state
== IMSM_T_STATE_DEGRADED
)
4477 static void update_recovery_start(struct imsm_dev
*dev
, struct mdinfo
*array
)
4479 struct mdinfo
*rebuild
= NULL
;
4483 if (!is_rebuilding(dev
))
4486 /* Find the rebuild target, but punt on the dual rebuild case */
4487 for (d
= array
->devs
; d
; d
= d
->next
)
4488 if (d
->recovery_start
== 0) {
4495 /* (?) none of the disks are marked with
4496 * IMSM_ORD_REBUILD, so assume they are missing and the
4497 * disk_ord_tbl was not correctly updated
4499 dprintf("%s: failed to locate out-of-sync disk\n", __func__
);
4503 units
= __le32_to_cpu(dev
->vol
.curr_migr_unit
);
4504 rebuild
->recovery_start
= units
* blocks_per_migr_unit(dev
);
4508 static struct mdinfo
*container_content_imsm(struct supertype
*st
, char *subarray
)
4510 /* Given a container loaded by load_super_imsm_all,
4511 * extract information about all the arrays into
4513 * If 'subarray' is given, just extract info about that array.
4515 * For each imsm_dev create an mdinfo, fill it in,
4516 * then look for matching devices in super->disks
4517 * and create appropriate device mdinfo.
4519 struct intel_super
*super
= st
->sb
;
4520 struct imsm_super
*mpb
= super
->anchor
;
4521 struct mdinfo
*rest
= NULL
;
4525 /* check for bad blocks */
4526 if (imsm_bbm_log_size(super
->anchor
))
4529 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
4530 struct imsm_dev
*dev
;
4531 struct imsm_map
*map
;
4532 struct mdinfo
*this;
4537 (i
!= strtoul(subarray
, &ep
, 10) || *ep
!= '\0'))
4540 dev
= get_imsm_dev(super
, i
);
4541 map
= get_imsm_map(dev
, 0);
4543 /* do not publish arrays that are in the middle of an
4544 * unsupported migration
4546 if (dev
->vol
.migr_state
&&
4547 (migr_type(dev
) == MIGR_STATE_CHANGE
)) {
4548 fprintf(stderr
, Name
": cannot assemble volume '%.16s':"
4549 " unsupported migration in progress\n",
4554 this = malloc(sizeof(*this));
4556 fprintf(stderr
, Name
": failed to allocate %zu bytes\n",
4560 memset(this, 0, sizeof(*this));
4563 super
->current_vol
= i
;
4564 getinfo_super_imsm_volume(st
, this, NULL
);
4565 for (slot
= 0 ; slot
< map
->num_members
; slot
++) {
4566 unsigned long long recovery_start
;
4567 struct mdinfo
*info_d
;
4574 idx
= get_imsm_disk_idx(dev
, slot
);
4575 ord
= get_imsm_ord_tbl_ent(dev
, slot
);
4576 for (d
= super
->disks
; d
; d
= d
->next
)
4577 if (d
->index
== idx
)
4580 recovery_start
= MaxSector
;
4583 if (d
&& is_failed(&d
->disk
))
4585 if (ord
& IMSM_ORD_REBUILD
)
4589 * if we skip some disks the array will be assmebled degraded;
4590 * reset resync start to avoid a dirty-degraded
4591 * situation when performing the intial sync
4593 * FIXME handle dirty degraded
4595 if ((skip
|| recovery_start
== 0) && !dev
->vol
.dirty
)
4596 this->resync_start
= MaxSector
;
4600 info_d
= calloc(1, sizeof(*info_d
));
4602 fprintf(stderr
, Name
": failed to allocate disk"
4603 " for volume %.16s\n", dev
->volume
);
4604 info_d
= this->devs
;
4606 struct mdinfo
*d
= info_d
->next
;
4615 info_d
->next
= this->devs
;
4616 this->devs
= info_d
;
4618 info_d
->disk
.number
= d
->index
;
4619 info_d
->disk
.major
= d
->major
;
4620 info_d
->disk
.minor
= d
->minor
;
4621 info_d
->disk
.raid_disk
= slot
;
4622 info_d
->recovery_start
= recovery_start
;
4624 if (info_d
->recovery_start
== MaxSector
)
4625 this->array
.working_disks
++;
4627 info_d
->events
= __le32_to_cpu(mpb
->generation_num
);
4628 info_d
->data_offset
= __le32_to_cpu(map
->pba_of_lba0
);
4629 info_d
->component_size
= __le32_to_cpu(map
->blocks_per_member
);
4631 /* now that the disk list is up-to-date fixup recovery_start */
4632 update_recovery_start(dev
, this);
4636 /* if array has bad blocks, set suitable bit in array status */
4638 rest
->array
.state
|= (1<<MD_SB_BBM_ERRORS
);
4644 static __u8
imsm_check_degraded(struct intel_super
*super
, struct imsm_dev
*dev
, int failed
)
4646 struct imsm_map
*map
= get_imsm_map(dev
, 0);
4649 return map
->map_state
== IMSM_T_STATE_UNINITIALIZED
?
4650 IMSM_T_STATE_UNINITIALIZED
: IMSM_T_STATE_NORMAL
;
4652 switch (get_imsm_raid_level(map
)) {
4654 return IMSM_T_STATE_FAILED
;
4657 if (failed
< map
->num_members
)
4658 return IMSM_T_STATE_DEGRADED
;
4660 return IMSM_T_STATE_FAILED
;
4665 * check to see if any mirrors have failed, otherwise we
4666 * are degraded. Even numbered slots are mirrored on
4670 /* gcc -Os complains that this is unused */
4671 int insync
= insync
;
4673 for (i
= 0; i
< map
->num_members
; i
++) {
4674 __u32 ord
= get_imsm_ord_tbl_ent(dev
, i
);
4675 int idx
= ord_to_idx(ord
);
4676 struct imsm_disk
*disk
;
4678 /* reset the potential in-sync count on even-numbered
4679 * slots. num_copies is always 2 for imsm raid10
4684 disk
= get_imsm_disk(super
, idx
);
4685 if (!disk
|| is_failed(disk
) || ord
& IMSM_ORD_REBUILD
)
4688 /* no in-sync disks left in this mirror the
4692 return IMSM_T_STATE_FAILED
;
4695 return IMSM_T_STATE_DEGRADED
;
4699 return IMSM_T_STATE_DEGRADED
;
4701 return IMSM_T_STATE_FAILED
;
4707 return map
->map_state
;
4710 static int imsm_count_failed(struct intel_super
*super
, struct imsm_dev
*dev
)
4714 struct imsm_disk
*disk
;
4715 struct imsm_map
*map
= get_imsm_map(dev
, 0);
4716 struct imsm_map
*prev
= get_imsm_map(dev
, dev
->vol
.migr_state
);
4720 /* at the beginning of migration we set IMSM_ORD_REBUILD on
4721 * disks that are being rebuilt. New failures are recorded to
4722 * map[0]. So we look through all the disks we started with and
4723 * see if any failures are still present, or if any new ones
4726 * FIXME add support for online capacity expansion and
4727 * raid-level-migration
4729 for (i
= 0; i
< prev
->num_members
; i
++) {
4730 ord
= __le32_to_cpu(prev
->disk_ord_tbl
[i
]);
4731 ord
|= __le32_to_cpu(map
->disk_ord_tbl
[i
]);
4732 idx
= ord_to_idx(ord
);
4734 disk
= get_imsm_disk(super
, idx
);
4735 if (!disk
|| is_failed(disk
) || ord
& IMSM_ORD_REBUILD
)
4743 static int imsm_open_new(struct supertype
*c
, struct active_array
*a
,
4746 struct intel_super
*super
= c
->sb
;
4747 struct imsm_super
*mpb
= super
->anchor
;
4749 if (atoi(inst
) >= mpb
->num_raid_devs
) {
4750 fprintf(stderr
, "%s: subarry index %d, out of range\n",
4751 __func__
, atoi(inst
));
4755 dprintf("imsm: open_new %s\n", inst
);
4756 a
->info
.container_member
= atoi(inst
);
4760 static int is_resyncing(struct imsm_dev
*dev
)
4762 struct imsm_map
*migr_map
;
4764 if (!dev
->vol
.migr_state
)
4767 if (migr_type(dev
) == MIGR_INIT
||
4768 migr_type(dev
) == MIGR_REPAIR
)
4771 if (migr_type(dev
) == MIGR_GEN_MIGR
)
4774 migr_map
= get_imsm_map(dev
, 1);
4776 if ((migr_map
->map_state
== IMSM_T_STATE_NORMAL
) &&
4777 (dev
->vol
.migr_type
!= MIGR_GEN_MIGR
))
4783 /* return true if we recorded new information */
4784 static int mark_failure(struct imsm_dev
*dev
, struct imsm_disk
*disk
, int idx
)
4788 struct imsm_map
*map
;
4790 /* new failures are always set in map[0] */
4791 map
= get_imsm_map(dev
, 0);
4793 slot
= get_imsm_disk_slot(map
, idx
);
4797 ord
= __le32_to_cpu(map
->disk_ord_tbl
[slot
]);
4798 if (is_failed(disk
) && (ord
& IMSM_ORD_REBUILD
))
4801 disk
->status
|= FAILED_DISK
;
4802 disk
->status
&= ~CONFIGURED_DISK
;
4803 set_imsm_ord_tbl_ent(map
, slot
, idx
| IMSM_ORD_REBUILD
);
4804 if (map
->failed_disk_num
== 0xff)
4805 map
->failed_disk_num
= slot
;
4809 static void mark_missing(struct imsm_dev
*dev
, struct imsm_disk
*disk
, int idx
)
4811 mark_failure(dev
, disk
, idx
);
4813 if (disk
->scsi_id
== __cpu_to_le32(~(__u32
)0))
4816 disk
->scsi_id
= __cpu_to_le32(~(__u32
)0);
4817 memmove(&disk
->serial
[0], &disk
->serial
[1], MAX_RAID_SERIAL_LEN
- 1);
4820 static void handle_missing(struct intel_super
*super
, struct imsm_dev
*dev
)
4826 if (!super
->missing
)
4828 failed
= imsm_count_failed(super
, dev
);
4829 map_state
= imsm_check_degraded(super
, dev
, failed
);
4831 dprintf("imsm: mark missing\n");
4832 end_migration(dev
, map_state
);
4833 for (dl
= super
->missing
; dl
; dl
= dl
->next
)
4834 mark_missing(dev
, &dl
->disk
, dl
->index
);
4835 super
->updates_pending
++;
4838 static void imsm_set_disk(struct active_array
*a
, int n
, int state
);
4840 /* Handle dirty -> clean transititions, resync and reshape. Degraded and rebuild
4841 * states are handled in imsm_set_disk() with one exception, when a
4842 * resync is stopped due to a new failure this routine will set the
4843 * 'degraded' state for the array.
4845 static int imsm_set_array_state(struct active_array
*a
, int consistent
)
4847 int inst
= a
->info
.container_member
;
4848 struct intel_super
*super
= a
->container
->sb
;
4849 struct imsm_dev
*dev
= get_imsm_dev(super
, inst
);
4850 struct imsm_map
*map
= get_imsm_map(dev
, 0);
4851 int failed
= imsm_count_failed(super
, dev
);
4852 __u8 map_state
= imsm_check_degraded(super
, dev
, failed
);
4853 __u32 blocks_per_unit
;
4855 if (dev
->vol
.migr_state
&&
4856 dev
->vol
.migr_type
== MIGR_GEN_MIGR
) {
4857 /* array state change is blocked due to reshape action
4859 * - abort the reshape (if last_checkpoint is 0 and action!= reshape)
4860 * - finish the reshape (if last_checkpoint is big and action != reshape)
4861 * - update curr_migr_unit
4863 if (a
->curr_action
== reshape
) {
4864 /* still reshaping, maybe update curr_migr_unit */
4865 long long blocks_per_unit
= blocks_per_migr_unit(dev
);
4866 long long unit
= a
->last_checkpoint
;
4867 unit
/= blocks_per_unit
;
4868 if (unit
> __le32_to_cpu(dev
->vol
.curr_migr_unit
)) {
4869 dev
->vol
.curr_migr_unit
= __cpu_to_le32(unit
);
4870 super
->updates_pending
++;
4873 if (a
->last_checkpoint
== 0 && a
->prev_action
== reshape
) {
4874 /* for some reason we aborted the reshape.
4877 struct imsm_map
*map2
= get_imsm_map(dev
, 1);
4878 dev
->vol
.migr_state
= 0;
4879 dev
->vol
.migr_type
= 0;
4880 dev
->vol
.curr_migr_unit
= 0;
4881 memcpy(map
, map2
, sizeof_imsm_map(map2
));
4882 super
->updates_pending
++;
4884 if (a
->last_checkpoint
>= a
->info
.component_size
) {
4885 unsigned long long array_blocks
;
4887 /* it seems the reshape is all done */
4888 dev
->vol
.migr_state
= 0;
4889 dev
->vol
.migr_type
= 0;
4890 dev
->vol
.curr_migr_unit
= 0;
4892 used_disks
= imsm_num_data_members(dev
);
4893 array_blocks
= map
->blocks_per_member
* used_disks
;
4894 /* round array size down to closest MB */
4895 array_blocks
= (array_blocks
>> SECT_PER_MB_SHIFT
)
4896 << SECT_PER_MB_SHIFT
;
4897 dev
->size_low
= __cpu_to_le32((__u32
) array_blocks
);
4898 dev
->size_high
= __cpu_to_le32((__u32
) (array_blocks
>> 32));
4899 a
->info
.custom_array_size
= array_blocks
;
4900 a
->check_reshape
= 1; /* encourage manager to update
4903 super
->updates_pending
++;
4909 /* before we activate this array handle any missing disks */
4910 if (consistent
== 2)
4911 handle_missing(super
, dev
);
4913 if (consistent
== 2 &&
4914 (!is_resync_complete(&a
->info
) ||
4915 map_state
!= IMSM_T_STATE_NORMAL
||
4916 dev
->vol
.migr_state
))
4919 if (is_resync_complete(&a
->info
)) {
4920 /* complete intialization / resync,
4921 * recovery and interrupted recovery is completed in
4924 if (is_resyncing(dev
)) {
4925 dprintf("imsm: mark resync done\n");
4926 end_migration(dev
, map_state
);
4927 super
->updates_pending
++;
4928 a
->last_checkpoint
= 0;
4930 } else if (!is_resyncing(dev
) && !failed
) {
4931 /* mark the start of the init process if nothing is failed */
4932 dprintf("imsm: mark resync start\n");
4933 if (map
->map_state
== IMSM_T_STATE_UNINITIALIZED
)
4934 migrate(dev
, IMSM_T_STATE_NORMAL
, MIGR_INIT
);
4936 migrate(dev
, IMSM_T_STATE_NORMAL
, MIGR_REPAIR
);
4937 super
->updates_pending
++;
4940 /* check if we can update curr_migr_unit from resync_start, recovery_start */
4941 blocks_per_unit
= blocks_per_migr_unit(dev
);
4942 if (blocks_per_unit
) {
4946 units
= a
->last_checkpoint
/ blocks_per_unit
;
4949 /* check that we did not overflow 32-bits, and that
4950 * curr_migr_unit needs updating
4952 if (units32
== units
&&
4953 __le32_to_cpu(dev
->vol
.curr_migr_unit
) != units32
) {
4954 dprintf("imsm: mark checkpoint (%u)\n", units32
);
4955 dev
->vol
.curr_migr_unit
= __cpu_to_le32(units32
);
4956 super
->updates_pending
++;
4960 /* mark dirty / clean */
4961 if (dev
->vol
.dirty
!= !consistent
) {
4962 dprintf("imsm: mark '%s'\n", consistent
? "clean" : "dirty");
4967 super
->updates_pending
++;
4970 /* finalize online capacity expansion/reshape */
4971 if ((a
->curr_action
!= reshape
) &&
4972 (a
->prev_action
== reshape
)) {
4975 for (mdi
= a
->info
.devs
; mdi
; mdi
= mdi
->next
)
4976 imsm_set_disk(a
, mdi
->disk
.raid_disk
, mdi
->curr_state
);
4982 static void imsm_set_disk(struct active_array
*a
, int n
, int state
)
4984 int inst
= a
->info
.container_member
;
4985 struct intel_super
*super
= a
->container
->sb
;
4986 struct imsm_dev
*dev
= get_imsm_dev(super
, inst
);
4987 struct imsm_map
*map
= get_imsm_map(dev
, 0);
4988 struct imsm_disk
*disk
;
4993 if (n
> map
->num_members
)
4994 fprintf(stderr
, "imsm: set_disk %d out of range 0..%d\n",
4995 n
, map
->num_members
- 1);
5000 dprintf("imsm: set_disk %d:%x\n", n
, state
);
5002 ord
= get_imsm_ord_tbl_ent(dev
, n
);
5003 disk
= get_imsm_disk(super
, ord_to_idx(ord
));
5005 /* check for new failures */
5006 if (state
& DS_FAULTY
) {
5007 if (mark_failure(dev
, disk
, ord_to_idx(ord
)))
5008 super
->updates_pending
++;
5011 /* check if in_sync */
5012 if (state
& DS_INSYNC
&& ord
& IMSM_ORD_REBUILD
&& is_rebuilding(dev
)) {
5013 struct imsm_map
*migr_map
= get_imsm_map(dev
, 1);
5015 set_imsm_ord_tbl_ent(migr_map
, n
, ord_to_idx(ord
));
5016 super
->updates_pending
++;
5019 failed
= imsm_count_failed(super
, dev
);
5020 map_state
= imsm_check_degraded(super
, dev
, failed
);
5022 /* check if recovery complete, newly degraded, or failed */
5023 if (map_state
== IMSM_T_STATE_NORMAL
&& is_rebuilding(dev
)) {
5024 end_migration(dev
, map_state
);
5025 map
= get_imsm_map(dev
, 0);
5026 map
->failed_disk_num
= ~0;
5027 super
->updates_pending
++;
5028 a
->last_checkpoint
= 0;
5029 } else if (map_state
== IMSM_T_STATE_DEGRADED
&&
5030 map
->map_state
!= map_state
&&
5031 !dev
->vol
.migr_state
) {
5032 dprintf("imsm: mark degraded\n");
5033 map
->map_state
= map_state
;
5034 super
->updates_pending
++;
5035 a
->last_checkpoint
= 0;
5036 } else if (map_state
== IMSM_T_STATE_FAILED
&&
5037 map
->map_state
!= map_state
) {
5038 dprintf("imsm: mark failed\n");
5039 end_migration(dev
, map_state
);
5040 super
->updates_pending
++;
5041 a
->last_checkpoint
= 0;
5042 } else if (is_gen_migration(dev
)) {
5043 dprintf("imsm: Detected General Migration in state: ");
5044 if (map_state
== IMSM_T_STATE_NORMAL
) {
5045 end_migration(dev
, map_state
);
5046 map
= get_imsm_map(dev
, 0);
5047 map
->failed_disk_num
= ~0;
5048 dprintf("normal\n");
5050 if (map_state
== IMSM_T_STATE_DEGRADED
) {
5051 printf("degraded\n");
5052 end_migration(dev
, map_state
);
5054 dprintf("failed\n");
5056 map
->map_state
= map_state
;
5058 super
->updates_pending
++;
5062 static int store_imsm_mpb(int fd
, struct imsm_super
*mpb
)
5065 __u32 mpb_size
= __le32_to_cpu(mpb
->mpb_size
);
5066 unsigned long long dsize
;
5067 unsigned long long sectors
;
5069 get_dev_size(fd
, NULL
, &dsize
);
5071 if (mpb_size
> 512) {
5072 /* -1 to account for anchor */
5073 sectors
= mpb_sectors(mpb
) - 1;
5075 /* write the extended mpb to the sectors preceeding the anchor */
5076 if (lseek64(fd
, dsize
- (512 * (2 + sectors
)), SEEK_SET
) < 0)
5079 if ((unsigned long long)write(fd
, buf
+ 512, 512 * sectors
)
5084 /* first block is stored on second to last sector of the disk */
5085 if (lseek64(fd
, dsize
- (512 * 2), SEEK_SET
) < 0)
5088 if (write(fd
, buf
, 512) != 512)
5094 static void imsm_sync_metadata(struct supertype
*container
)
5096 struct intel_super
*super
= container
->sb
;
5098 dprintf("sync metadata: %d\n", super
->updates_pending
);
5099 if (!super
->updates_pending
)
5102 write_super_imsm(container
, 0);
5104 super
->updates_pending
= 0;
5107 static struct dl
*imsm_readd(struct intel_super
*super
, int idx
, struct active_array
*a
)
5109 struct imsm_dev
*dev
= get_imsm_dev(super
, a
->info
.container_member
);
5110 int i
= get_imsm_disk_idx(dev
, idx
);
5113 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
5117 if (dl
&& is_failed(&dl
->disk
))
5121 dprintf("%s: found %x:%x\n", __func__
, dl
->major
, dl
->minor
);
5126 static struct dl
*imsm_add_spare(struct intel_super
*super
, int slot
,
5127 struct active_array
*a
, int activate_new
,
5128 struct mdinfo
*additional_test_list
)
5130 struct imsm_dev
*dev
= get_imsm_dev(super
, a
->info
.container_member
);
5131 int idx
= get_imsm_disk_idx(dev
, slot
);
5132 struct imsm_super
*mpb
= super
->anchor
;
5133 struct imsm_map
*map
;
5134 unsigned long long pos
;
5139 __u32 array_start
= 0;
5140 __u32 array_end
= 0;
5142 struct mdinfo
*test_list
;
5144 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
5145 /* If in this array, skip */
5146 for (d
= a
->info
.devs
; d
; d
= d
->next
)
5147 if (d
->state_fd
>= 0 &&
5148 d
->disk
.major
== dl
->major
&&
5149 d
->disk
.minor
== dl
->minor
) {
5150 dprintf("%x:%x already in array\n",
5151 dl
->major
, dl
->minor
);
5156 test_list
= additional_test_list
;
5158 if (test_list
->disk
.major
== dl
->major
&&
5159 test_list
->disk
.minor
== dl
->minor
) {
5160 dprintf("%x:%x already in additional test list\n",
5161 dl
->major
, dl
->minor
);
5164 test_list
= test_list
->next
;
5169 /* skip in use or failed drives */
5170 if (is_failed(&dl
->disk
) || idx
== dl
->index
||
5172 dprintf("%x:%x status (failed: %d index: %d)\n",
5173 dl
->major
, dl
->minor
, is_failed(&dl
->disk
), idx
);
5177 /* skip pure spares when we are looking for partially
5178 * assimilated drives
5180 if (dl
->index
== -1 && !activate_new
)
5183 /* Does this unused device have the requisite free space?
5184 * It needs to be able to cover all member volumes
5186 ex
= get_extents(super
, dl
);
5188 dprintf("cannot get extents\n");
5191 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
5192 dev
= get_imsm_dev(super
, i
);
5193 map
= get_imsm_map(dev
, 0);
5195 /* check if this disk is already a member of
5198 if (get_imsm_disk_slot(map
, dl
->index
) >= 0)
5204 array_start
= __le32_to_cpu(map
->pba_of_lba0
);
5205 array_end
= array_start
+
5206 __le32_to_cpu(map
->blocks_per_member
) - 1;
5209 /* check that we can start at pba_of_lba0 with
5210 * blocks_per_member of space
5212 if (array_start
>= pos
&& array_end
< ex
[j
].start
) {
5216 pos
= ex
[j
].start
+ ex
[j
].size
;
5218 } while (ex
[j
-1].size
);
5225 if (i
< mpb
->num_raid_devs
) {
5226 dprintf("%x:%x does not have %u to %u available\n",
5227 dl
->major
, dl
->minor
, array_start
, array_end
);
5238 static int imsm_rebuild_allowed(struct supertype
*cont
, int dev_idx
, int failed
)
5240 struct imsm_dev
*dev2
;
5241 struct imsm_map
*map
;
5247 dev2
= get_imsm_dev(cont
->sb
, dev_idx
);
5249 state
= imsm_check_degraded(cont
->sb
, dev2
, failed
);
5250 if (state
== IMSM_T_STATE_FAILED
) {
5251 map
= get_imsm_map(dev2
, 0);
5254 for (slot
= 0; slot
< map
->num_members
; slot
++) {
5256 * Check if failed disks are deleted from intel
5257 * disk list or are marked to be deleted
5259 idx
= get_imsm_disk_idx(dev2
, slot
);
5260 idisk
= get_imsm_dl_disk(cont
->sb
, idx
);
5262 * Do not rebuild the array if failed disks
5263 * from failed sub-array are not removed from
5267 is_failed(&idisk
->disk
) &&
5268 (idisk
->action
!= DISK_REMOVE
))
5276 static struct mdinfo
*imsm_activate_spare(struct active_array
*a
,
5277 struct metadata_update
**updates
)
5280 * Find a device with unused free space and use it to replace a
5281 * failed/vacant region in an array. We replace failed regions one a
5282 * array at a time. The result is that a new spare disk will be added
5283 * to the first failed array and after the monitor has finished
5284 * propagating failures the remainder will be consumed.
5286 * FIXME add a capability for mdmon to request spares from another
5290 struct intel_super
*super
= a
->container
->sb
;
5291 int inst
= a
->info
.container_member
;
5292 struct imsm_dev
*dev
= get_imsm_dev(super
, inst
);
5293 struct imsm_map
*map
= get_imsm_map(dev
, 0);
5294 int failed
= a
->info
.array
.raid_disks
;
5295 struct mdinfo
*rv
= NULL
;
5298 struct metadata_update
*mu
;
5300 struct imsm_update_activate_spare
*u
;
5305 for (d
= a
->info
.devs
; d
; d
= d
->next
) {
5306 if ((d
->curr_state
& DS_FAULTY
) &&
5308 /* wait for Removal to happen */
5310 if (d
->state_fd
>= 0)
5314 dprintf("imsm: activate spare: inst=%d failed=%d (%d) level=%d\n",
5315 inst
, failed
, a
->info
.array
.raid_disks
, a
->info
.array
.level
);
5317 if (dev
->vol
.migr_state
&&
5318 dev
->vol
.migr_type
== MIGR_GEN_MIGR
)
5319 /* No repair during migration */
5322 if (imsm_check_degraded(super
, dev
, failed
) != IMSM_T_STATE_DEGRADED
)
5326 * If there are any failed disks check state of the other volume.
5327 * Block rebuild if the another one is failed until failed disks
5328 * are removed from container.
5331 dprintf("found failed disks in %s, check if there another"
5332 "failed sub-array.\n",
5334 /* check if states of the other volumes allow for rebuild */
5335 for (i
= 0; i
< super
->anchor
->num_raid_devs
; i
++) {
5337 allowed
= imsm_rebuild_allowed(a
->container
,
5345 /* For each slot, if it is not working, find a spare */
5346 for (i
= 0; i
< a
->info
.array
.raid_disks
; i
++) {
5347 for (d
= a
->info
.devs
; d
; d
= d
->next
)
5348 if (d
->disk
.raid_disk
== i
)
5350 dprintf("found %d: %p %x\n", i
, d
, d
?d
->curr_state
:0);
5351 if (d
&& (d
->state_fd
>= 0))
5355 * OK, this device needs recovery. Try to re-add the
5356 * previous occupant of this slot, if this fails see if
5357 * we can continue the assimilation of a spare that was
5358 * partially assimilated, finally try to activate a new
5361 dl
= imsm_readd(super
, i
, a
);
5363 dl
= imsm_add_spare(super
, i
, a
, 0, NULL
);
5365 dl
= imsm_add_spare(super
, i
, a
, 1, NULL
);
5369 /* found a usable disk with enough space */
5370 di
= malloc(sizeof(*di
));
5373 memset(di
, 0, sizeof(*di
));
5375 /* dl->index will be -1 in the case we are activating a
5376 * pristine spare. imsm_process_update() will create a
5377 * new index in this case. Once a disk is found to be
5378 * failed in all member arrays it is kicked from the
5381 di
->disk
.number
= dl
->index
;
5383 /* (ab)use di->devs to store a pointer to the device
5386 di
->devs
= (struct mdinfo
*) dl
;
5388 di
->disk
.raid_disk
= i
;
5389 di
->disk
.major
= dl
->major
;
5390 di
->disk
.minor
= dl
->minor
;
5392 di
->recovery_start
= 0;
5393 di
->data_offset
= __le32_to_cpu(map
->pba_of_lba0
);
5394 di
->component_size
= a
->info
.component_size
;
5395 di
->container_member
= inst
;
5396 super
->random
= random32();
5400 dprintf("%x:%x to be %d at %llu\n", dl
->major
, dl
->minor
,
5401 i
, di
->data_offset
);
5407 /* No spares found */
5409 /* Now 'rv' has a list of devices to return.
5410 * Create a metadata_update record to update the
5411 * disk_ord_tbl for the array
5413 mu
= malloc(sizeof(*mu
));
5415 mu
->buf
= malloc(sizeof(struct imsm_update_activate_spare
) * num_spares
);
5416 if (mu
->buf
== NULL
) {
5423 struct mdinfo
*n
= rv
->next
;
5432 mu
->space_list
= NULL
;
5433 mu
->len
= sizeof(struct imsm_update_activate_spare
) * num_spares
;
5434 mu
->next
= *updates
;
5435 u
= (struct imsm_update_activate_spare
*) mu
->buf
;
5437 for (di
= rv
; di
; di
= di
->next
) {
5438 u
->type
= update_activate_spare
;
5439 u
->dl
= (struct dl
*) di
->devs
;
5441 u
->slot
= di
->disk
.raid_disk
;
5452 static int disks_overlap(struct intel_super
*super
, int idx
, struct imsm_update_create_array
*u
)
5454 struct imsm_dev
*dev
= get_imsm_dev(super
, idx
);
5455 struct imsm_map
*map
= get_imsm_map(dev
, 0);
5456 struct imsm_map
*new_map
= get_imsm_map(&u
->dev
, 0);
5457 struct disk_info
*inf
= get_disk_info(u
);
5458 struct imsm_disk
*disk
;
5462 for (i
= 0; i
< map
->num_members
; i
++) {
5463 disk
= get_imsm_disk(super
, get_imsm_disk_idx(dev
, i
));
5464 for (j
= 0; j
< new_map
->num_members
; j
++)
5465 if (serialcmp(disk
->serial
, inf
[j
].serial
) == 0)
5473 static struct dl
*get_disk_super(struct intel_super
*super
, int major
, int minor
)
5475 struct dl
*dl
= NULL
;
5476 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
5477 if ((dl
->major
== major
) && (dl
->minor
== minor
))
5482 static int remove_disk_super(struct intel_super
*super
, int major
, int minor
)
5484 struct dl
*prev
= NULL
;
5488 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
5489 if ((dl
->major
== major
) && (dl
->minor
== minor
)) {
5492 prev
->next
= dl
->next
;
5494 super
->disks
= dl
->next
;
5496 __free_imsm_disk(dl
);
5497 dprintf("%s: removed %x:%x\n",
5498 __func__
, major
, minor
);
5506 static void imsm_delete(struct intel_super
*super
, struct dl
**dlp
, unsigned index
);
5508 static int add_remove_disk_update(struct intel_super
*super
)
5510 int check_degraded
= 0;
5511 struct dl
*disk
= NULL
;
5512 /* add/remove some spares to/from the metadata/contrainer */
5513 while (super
->disk_mgmt_list
) {
5514 struct dl
*disk_cfg
;
5516 disk_cfg
= super
->disk_mgmt_list
;
5517 super
->disk_mgmt_list
= disk_cfg
->next
;
5518 disk_cfg
->next
= NULL
;
5520 if (disk_cfg
->action
== DISK_ADD
) {
5521 disk_cfg
->next
= super
->disks
;
5522 super
->disks
= disk_cfg
;
5524 dprintf("%s: added %x:%x\n",
5525 __func__
, disk_cfg
->major
,
5527 } else if (disk_cfg
->action
== DISK_REMOVE
) {
5528 dprintf("Disk remove action processed: %x.%x\n",
5529 disk_cfg
->major
, disk_cfg
->minor
);
5530 disk
= get_disk_super(super
,
5534 /* store action status */
5535 disk
->action
= DISK_REMOVE
;
5536 /* remove spare disks only */
5537 if (disk
->index
== -1) {
5538 remove_disk_super(super
,
5543 /* release allocate disk structure */
5544 __free_imsm_disk(disk_cfg
);
5547 return check_degraded
;
5550 static void imsm_process_update(struct supertype
*st
,
5551 struct metadata_update
*update
)
5554 * crack open the metadata_update envelope to find the update record
5555 * update can be one of:
5556 * update_reshape_container_disks - all the arrays in the container
5557 * are being reshaped to have more devices. We need to mark
5558 * the arrays for general migration and convert selected spares
5559 * into active devices.
5560 * update_activate_spare - a spare device has replaced a failed
5561 * device in an array, update the disk_ord_tbl. If this disk is
5562 * present in all member arrays then also clear the SPARE_DISK
5564 * update_create_array
5566 * update_rename_array
5567 * update_add_remove_disk
5569 struct intel_super
*super
= st
->sb
;
5570 struct imsm_super
*mpb
;
5571 enum imsm_update_type type
= *(enum imsm_update_type
*) update
->buf
;
5573 /* update requires a larger buf but the allocation failed */
5574 if (super
->next_len
&& !super
->next_buf
) {
5575 super
->next_len
= 0;
5579 if (super
->next_buf
) {
5580 memcpy(super
->next_buf
, super
->buf
, super
->len
);
5582 super
->len
= super
->next_len
;
5583 super
->buf
= super
->next_buf
;
5585 super
->next_len
= 0;
5586 super
->next_buf
= NULL
;
5589 mpb
= super
->anchor
;
5592 case update_reshape_container_disks
: {
5593 struct imsm_update_reshape
*u
= (void *)update
->buf
;
5594 struct dl
*new_disk
;
5595 struct intel_dev
*id
;
5597 int delta_disks
= u
->new_raid_disks
- u
->old_raid_disks
;
5598 void **tofree
= NULL
;
5600 dprintf("imsm: imsm_process_update() for update_reshape\n");
5602 /* enable spares to use in array */
5603 for (i
= 0; i
< delta_disks
; i
++) {
5605 new_disk
= get_disk_super(super
,
5606 major(u
->new_disks
[i
]),
5607 minor(u
->new_disks
[i
]));
5608 if (new_disk
== NULL
|| new_disk
->index
< 0)
5609 goto update_reshape_exit
;
5611 new_disk
->index
= mpb
->num_disks
++;
5612 /* slot to fill in autolayout */
5613 new_disk
->raiddisk
= new_disk
->index
;
5614 new_disk
->disk
.status
|=
5616 new_disk
->disk
.status
&= ~SPARE_DISK
;
5619 dprintf("imsm: process_update(): update_reshape: volume set"\
5620 " mpb->num_raid_devs = %i\n", mpb
->num_raid_devs
);
5621 /* manage changes in volumes
5623 for (id
= super
->devlist
; id
; id
= id
->next
) {
5624 void **sp
= update
->space_list
;
5625 struct imsm_dev
*newdev
;
5626 struct imsm_map
*newmap
, *oldmap
;
5630 update
->space_list
= *sp
;
5632 /* Copy the dev, but not (all of) the map */
5633 memcpy(newdev
, id
->dev
, sizeof(*newdev
));
5634 oldmap
= get_imsm_map(id
->dev
, 0);
5635 newmap
= get_imsm_map(newdev
, 0);
5636 /* Copy the current map */
5637 memcpy(newmap
, oldmap
, sizeof_imsm_map(oldmap
));
5638 newdev
->vol
.migr_state
= 1;
5639 newdev
->vol
.curr_migr_unit
= 0;
5640 newdev
->vol
.migr_type
= MIGR_GEN_MIGR
;
5641 newmap
->num_members
= u
->new_raid_disks
;
5642 for (i
= 0; i
< delta_disks
; i
++) {
5643 set_imsm_ord_tbl_ent(newmap
,
5644 u
->old_raid_disks
+ i
,
5645 u
->old_raid_disks
+ i
);
5647 /* New map is correct, now need to save old map */
5648 oldmap
= get_imsm_map(newdev
, 1);
5649 memcpy(newmap
, oldmap
, sizeof_imsm_map(oldmap
));
5651 sp
= (void **)id
->dev
;
5657 update
->space_list
= tofree
;
5658 super
->updates_pending
++;
5659 update_reshape_exit
:
5662 case update_activate_spare
: {
5663 struct imsm_update_activate_spare
*u
= (void *) update
->buf
;
5664 struct imsm_dev
*dev
= get_imsm_dev(super
, u
->array
);
5665 struct imsm_map
*map
= get_imsm_map(dev
, 0);
5666 struct imsm_map
*migr_map
;
5667 struct active_array
*a
;
5668 struct imsm_disk
*disk
;
5673 int victim
= get_imsm_disk_idx(dev
, u
->slot
);
5676 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
5681 fprintf(stderr
, "error: imsm_activate_spare passed "
5682 "an unknown disk (index: %d)\n",
5687 super
->updates_pending
++;
5689 /* count failures (excluding rebuilds and the victim)
5690 * to determine map[0] state
5693 for (i
= 0; i
< map
->num_members
; i
++) {
5696 disk
= get_imsm_disk(super
, get_imsm_disk_idx(dev
, i
));
5697 if (!disk
|| is_failed(disk
))
5701 /* adding a pristine spare, assign a new index */
5702 if (dl
->index
< 0) {
5703 dl
->index
= super
->anchor
->num_disks
;
5704 super
->anchor
->num_disks
++;
5707 disk
->status
|= CONFIGURED_DISK
;
5708 disk
->status
&= ~SPARE_DISK
;
5711 to_state
= imsm_check_degraded(super
, dev
, failed
);
5712 map
->map_state
= IMSM_T_STATE_DEGRADED
;
5713 migrate(dev
, to_state
, MIGR_REBUILD
);
5714 migr_map
= get_imsm_map(dev
, 1);
5715 set_imsm_ord_tbl_ent(map
, u
->slot
, dl
->index
);
5716 set_imsm_ord_tbl_ent(migr_map
, u
->slot
, dl
->index
| IMSM_ORD_REBUILD
);
5718 /* update the family_num to mark a new container
5719 * generation, being careful to record the existing
5720 * family_num in orig_family_num to clean up after
5721 * earlier mdadm versions that neglected to set it.
5723 if (mpb
->orig_family_num
== 0)
5724 mpb
->orig_family_num
= mpb
->family_num
;
5725 mpb
->family_num
+= super
->random
;
5727 /* count arrays using the victim in the metadata */
5729 for (a
= st
->arrays
; a
; a
= a
->next
) {
5730 dev
= get_imsm_dev(super
, a
->info
.container_member
);
5731 map
= get_imsm_map(dev
, 0);
5733 if (get_imsm_disk_slot(map
, victim
) >= 0)
5737 /* delete the victim if it is no longer being
5743 /* We know that 'manager' isn't touching anything,
5744 * so it is safe to delete
5746 for (dlp
= &super
->disks
; *dlp
; dlp
= &(*dlp
)->next
)
5747 if ((*dlp
)->index
== victim
)
5750 /* victim may be on the missing list */
5752 for (dlp
= &super
->missing
; *dlp
; dlp
= &(*dlp
)->next
)
5753 if ((*dlp
)->index
== victim
)
5755 imsm_delete(super
, dlp
, victim
);
5759 case update_create_array
: {
5760 /* someone wants to create a new array, we need to be aware of
5761 * a few races/collisions:
5762 * 1/ 'Create' called by two separate instances of mdadm
5763 * 2/ 'Create' versus 'activate_spare': mdadm has chosen
5764 * devices that have since been assimilated via
5766 * In the event this update can not be carried out mdadm will
5767 * (FIX ME) notice that its update did not take hold.
5769 struct imsm_update_create_array
*u
= (void *) update
->buf
;
5770 struct intel_dev
*dv
;
5771 struct imsm_dev
*dev
;
5772 struct imsm_map
*map
, *new_map
;
5773 unsigned long long start
, end
;
5774 unsigned long long new_start
, new_end
;
5776 struct disk_info
*inf
;
5779 /* handle racing creates: first come first serve */
5780 if (u
->dev_idx
< mpb
->num_raid_devs
) {
5781 dprintf("%s: subarray %d already defined\n",
5782 __func__
, u
->dev_idx
);
5786 /* check update is next in sequence */
5787 if (u
->dev_idx
!= mpb
->num_raid_devs
) {
5788 dprintf("%s: can not create array %d expected index %d\n",
5789 __func__
, u
->dev_idx
, mpb
->num_raid_devs
);
5793 new_map
= get_imsm_map(&u
->dev
, 0);
5794 new_start
= __le32_to_cpu(new_map
->pba_of_lba0
);
5795 new_end
= new_start
+ __le32_to_cpu(new_map
->blocks_per_member
);
5796 inf
= get_disk_info(u
);
5798 /* handle activate_spare versus create race:
5799 * check to make sure that overlapping arrays do not include
5802 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
5803 dev
= get_imsm_dev(super
, i
);
5804 map
= get_imsm_map(dev
, 0);
5805 start
= __le32_to_cpu(map
->pba_of_lba0
);
5806 end
= start
+ __le32_to_cpu(map
->blocks_per_member
);
5807 if ((new_start
>= start
&& new_start
<= end
) ||
5808 (start
>= new_start
&& start
<= new_end
))
5813 if (disks_overlap(super
, i
, u
)) {
5814 dprintf("%s: arrays overlap\n", __func__
);
5819 /* check that prepare update was successful */
5820 if (!update
->space
) {
5821 dprintf("%s: prepare update failed\n", __func__
);
5825 /* check that all disks are still active before committing
5826 * changes. FIXME: could we instead handle this by creating a
5827 * degraded array? That's probably not what the user expects,
5828 * so better to drop this update on the floor.
5830 for (i
= 0; i
< new_map
->num_members
; i
++) {
5831 dl
= serial_to_dl(inf
[i
].serial
, super
);
5833 dprintf("%s: disk disappeared\n", __func__
);
5838 super
->updates_pending
++;
5840 /* convert spares to members and fixup ord_tbl */
5841 for (i
= 0; i
< new_map
->num_members
; i
++) {
5842 dl
= serial_to_dl(inf
[i
].serial
, super
);
5843 if (dl
->index
== -1) {
5844 dl
->index
= mpb
->num_disks
;
5846 dl
->disk
.status
|= CONFIGURED_DISK
;
5847 dl
->disk
.status
&= ~SPARE_DISK
;
5849 set_imsm_ord_tbl_ent(new_map
, i
, dl
->index
);
5854 update
->space
= NULL
;
5855 imsm_copy_dev(dev
, &u
->dev
);
5856 dv
->index
= u
->dev_idx
;
5857 dv
->next
= super
->devlist
;
5858 super
->devlist
= dv
;
5859 mpb
->num_raid_devs
++;
5861 imsm_update_version_info(super
);
5864 /* mdmon knows how to release update->space, but not
5865 * ((struct intel_dev *) update->space)->dev
5867 if (update
->space
) {
5873 case update_kill_array
: {
5874 struct imsm_update_kill_array
*u
= (void *) update
->buf
;
5875 int victim
= u
->dev_idx
;
5876 struct active_array
*a
;
5877 struct intel_dev
**dp
;
5878 struct imsm_dev
*dev
;
5880 /* sanity check that we are not affecting the uuid of
5881 * active arrays, or deleting an active array
5883 * FIXME when immutable ids are available, but note that
5884 * we'll also need to fixup the invalidated/active
5885 * subarray indexes in mdstat
5887 for (a
= st
->arrays
; a
; a
= a
->next
)
5888 if (a
->info
.container_member
>= victim
)
5890 /* by definition if mdmon is running at least one array
5891 * is active in the container, so checking
5892 * mpb->num_raid_devs is just extra paranoia
5894 dev
= get_imsm_dev(super
, victim
);
5895 if (a
|| !dev
|| mpb
->num_raid_devs
== 1) {
5896 dprintf("failed to delete subarray-%d\n", victim
);
5900 for (dp
= &super
->devlist
; *dp
;)
5901 if ((*dp
)->index
== (unsigned)super
->current_vol
) {
5904 if ((*dp
)->index
> (unsigned)victim
)
5908 mpb
->num_raid_devs
--;
5909 super
->updates_pending
++;
5912 case update_rename_array
: {
5913 struct imsm_update_rename_array
*u
= (void *) update
->buf
;
5914 char name
[MAX_RAID_SERIAL_LEN
+1];
5915 int target
= u
->dev_idx
;
5916 struct active_array
*a
;
5917 struct imsm_dev
*dev
;
5919 /* sanity check that we are not affecting the uuid of
5922 snprintf(name
, MAX_RAID_SERIAL_LEN
, "%s", (char *) u
->name
);
5923 name
[MAX_RAID_SERIAL_LEN
] = '\0';
5924 for (a
= st
->arrays
; a
; a
= a
->next
)
5925 if (a
->info
.container_member
== target
)
5927 dev
= get_imsm_dev(super
, u
->dev_idx
);
5928 if (a
|| !dev
|| !check_name(super
, name
, 1)) {
5929 dprintf("failed to rename subarray-%d\n", target
);
5933 snprintf((char *) dev
->volume
, MAX_RAID_SERIAL_LEN
, "%s", name
);
5934 super
->updates_pending
++;
5937 case update_add_remove_disk
: {
5938 /* we may be able to repair some arrays if disks are
5939 * being added, check teh status of add_remove_disk
5940 * if discs has been added.
5942 if (add_remove_disk_update(super
)) {
5943 struct active_array
*a
;
5945 super
->updates_pending
++;
5946 for (a
= st
->arrays
; a
; a
= a
->next
)
5947 a
->check_degraded
= 1;
5952 fprintf(stderr
, "error: unsuported process update type:"
5953 "(type: %d)\n", type
);
5957 static void imsm_prepare_update(struct supertype
*st
,
5958 struct metadata_update
*update
)
5961 * Allocate space to hold new disk entries, raid-device entries or a new
5962 * mpb if necessary. The manager synchronously waits for updates to
5963 * complete in the monitor, so new mpb buffers allocated here can be
5964 * integrated by the monitor thread without worrying about live pointers
5965 * in the manager thread.
5967 enum imsm_update_type type
= *(enum imsm_update_type
*) update
->buf
;
5968 struct intel_super
*super
= st
->sb
;
5969 struct imsm_super
*mpb
= super
->anchor
;
5974 case update_reshape_container_disks
: {
5975 /* Every raid device in the container is about to
5976 * gain some more devices, and we will enter a
5978 * So each 'imsm_map' will be bigger, and the imsm_vol
5979 * will now hold 2 of them.
5980 * Thus we need new 'struct imsm_dev' allocations sized
5981 * as sizeof_imsm_dev but with more devices in both maps.
5983 struct imsm_update_reshape
*u
= (void *)update
->buf
;
5984 struct intel_dev
*dl
;
5985 void **space_tail
= (void**)&update
->space_list
;
5987 dprintf("imsm: imsm_prepare_update() for update_reshape\n");
5989 for (dl
= super
->devlist
; dl
; dl
= dl
->next
) {
5990 int size
= sizeof_imsm_dev(dl
->dev
, 1);
5992 size
+= sizeof(__u32
) * 2 *
5993 (u
->new_raid_disks
- u
->old_raid_disks
);
6002 len
= disks_to_mpb_size(u
->new_raid_disks
);
6003 dprintf("New anchor length is %llu\n", (unsigned long long)len
);
6006 case update_create_array
: {
6007 struct imsm_update_create_array
*u
= (void *) update
->buf
;
6008 struct intel_dev
*dv
;
6009 struct imsm_dev
*dev
= &u
->dev
;
6010 struct imsm_map
*map
= get_imsm_map(dev
, 0);
6012 struct disk_info
*inf
;
6016 inf
= get_disk_info(u
);
6017 len
= sizeof_imsm_dev(dev
, 1);
6018 /* allocate a new super->devlist entry */
6019 dv
= malloc(sizeof(*dv
));
6021 dv
->dev
= malloc(len
);
6026 update
->space
= NULL
;
6030 /* count how many spares will be converted to members */
6031 for (i
= 0; i
< map
->num_members
; i
++) {
6032 dl
= serial_to_dl(inf
[i
].serial
, super
);
6034 /* hmm maybe it failed?, nothing we can do about
6039 if (count_memberships(dl
, super
) == 0)
6042 len
+= activate
* sizeof(struct imsm_disk
);
6049 /* check if we need a larger metadata buffer */
6050 if (super
->next_buf
)
6051 buf_len
= super
->next_len
;
6053 buf_len
= super
->len
;
6055 if (__le32_to_cpu(mpb
->mpb_size
) + len
> buf_len
) {
6056 /* ok we need a larger buf than what is currently allocated
6057 * if this allocation fails process_update will notice that
6058 * ->next_len is set and ->next_buf is NULL
6060 buf_len
= ROUND_UP(__le32_to_cpu(mpb
->mpb_size
) + len
, 512);
6061 if (super
->next_buf
)
6062 free(super
->next_buf
);
6064 super
->next_len
= buf_len
;
6065 if (posix_memalign(&super
->next_buf
, 512, buf_len
) == 0)
6066 memset(super
->next_buf
, 0, buf_len
);
6068 super
->next_buf
= NULL
;
6072 /* must be called while manager is quiesced */
6073 static void imsm_delete(struct intel_super
*super
, struct dl
**dlp
, unsigned index
)
6075 struct imsm_super
*mpb
= super
->anchor
;
6077 struct imsm_dev
*dev
;
6078 struct imsm_map
*map
;
6079 int i
, j
, num_members
;
6082 dprintf("%s: deleting device[%d] from imsm_super\n",
6085 /* shift all indexes down one */
6086 for (iter
= super
->disks
; iter
; iter
= iter
->next
)
6087 if (iter
->index
> (int)index
)
6089 for (iter
= super
->missing
; iter
; iter
= iter
->next
)
6090 if (iter
->index
> (int)index
)
6093 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
6094 dev
= get_imsm_dev(super
, i
);
6095 map
= get_imsm_map(dev
, 0);
6096 num_members
= map
->num_members
;
6097 for (j
= 0; j
< num_members
; j
++) {
6098 /* update ord entries being careful not to propagate
6099 * ord-flags to the first map
6101 ord
= get_imsm_ord_tbl_ent(dev
, j
);
6103 if (ord_to_idx(ord
) <= index
)
6106 map
= get_imsm_map(dev
, 0);
6107 set_imsm_ord_tbl_ent(map
, j
, ord_to_idx(ord
- 1));
6108 map
= get_imsm_map(dev
, 1);
6110 set_imsm_ord_tbl_ent(map
, j
, ord
- 1);
6115 super
->updates_pending
++;
6117 struct dl
*dl
= *dlp
;
6119 *dlp
= (*dlp
)->next
;
6120 __free_imsm_disk(dl
);
6123 #endif /* MDASSEMBLE */
6125 static char disk_by_path
[] = "/dev/disk/by-path/";
6127 static const char *imsm_get_disk_controller_domain(const char *path
)
6129 struct sys_dev
*list
, *hba
= NULL
;
6130 char disk_path
[PATH_MAX
];
6134 list
= find_driver_devices("pci", "ahci");
6135 for (hba
= list
; hba
; hba
= hba
->next
)
6136 if (devpath_to_vendor(hba
->path
) == 0x8086)
6142 strncpy(disk_path
, disk_by_path
, PATH_MAX
- 1);
6143 strncat(disk_path
, path
, PATH_MAX
- strlen(disk_path
) - 1);
6144 if (stat(disk_path
, &st
) == 0) {
6145 dpath
= devt_to_devpath(st
.st_rdev
);
6147 ahci
= path_attached_to_hba(dpath
, hba
->path
);
6150 dprintf("path: %s(%s) hba: %s attached: %d\n",
6151 path
, dpath
, (hba
) ? hba
->path
: "NULL", ahci
);
6152 free_sys_dev(&list
);
6159 static int imsm_find_array_minor_by_subdev(int subdev
, int container
, int *minor
)
6161 char subdev_name
[20];
6162 struct mdstat_ent
*mdstat
;
6164 sprintf(subdev_name
, "%d", subdev
);
6165 mdstat
= mdstat_by_subdev(subdev_name
, container
);
6169 *minor
= mdstat
->devnum
;
6170 free_mdstat(mdstat
);
6174 static int imsm_reshape_is_allowed_on_container(struct supertype
*st
,
6175 struct geo_params
*geo
,
6176 int *old_raid_disks
)
6179 struct mdinfo
*info
, *member
;
6180 int devices_that_can_grow
= 0;
6182 dprintf("imsm: imsm_reshape_is_allowed_on_container(ENTER): "
6183 "st->devnum = (%i)\n",
6186 if (geo
->size
!= -1 ||
6187 geo
->level
!= UnSet
||
6188 geo
->layout
!= UnSet
||
6189 geo
->chunksize
!= 0 ||
6190 geo
->raid_disks
== UnSet
) {
6191 dprintf("imsm: Container operation is allowed for "
6192 "raid disks number change only.\n");
6196 info
= container_content_imsm(st
, NULL
);
6197 for (member
= info
; member
; member
= member
->next
) {
6201 dprintf("imsm: checking device_num: %i\n",
6202 member
->container_member
);
6204 if (geo
->raid_disks
< member
->array
.raid_disks
) {
6205 /* we work on container for Online Capacity Expansion
6206 * only so raid_disks has to grow
6208 dprintf("imsm: for container operation raid disks "
6209 "increase is required\n");
6213 if ((info
->array
.level
!= 0) &&
6214 (info
->array
.level
!= 5)) {
6215 /* we cannot use this container with other raid level
6217 dprintf("imsm: for container operation wrong"\
6218 " raid level (%i) detected\n",
6222 /* check for platform support
6223 * for this raid level configuration
6225 struct intel_super
*super
= st
->sb
;
6226 if (!is_raid_level_supported(super
->orom
,
6227 member
->array
.level
,
6229 dprintf("platform does not support raid%d with"\
6233 geo
->raid_disks
> 1 ? "s" : "");
6238 if (*old_raid_disks
&&
6239 info
->array
.raid_disks
!= *old_raid_disks
)
6241 *old_raid_disks
= info
->array
.raid_disks
;
6243 /* All raid5 and raid0 volumes in container
6244 * have to be ready for Online Capacity Expansion
6245 * so they need to be assembled. We have already
6246 * checked that no recovery etc is happening.
6248 result
= imsm_find_array_minor_by_subdev(member
->container_member
,
6252 dprintf("imsm: cannot find array\n");
6255 devices_that_can_grow
++;
6258 if (!member
&& devices_that_can_grow
)
6262 dprintf("\tContainer operation allowed\n");
6264 dprintf("\tError: %i\n", ret_val
);
6269 /* Function: get_spares_for_grow
6270 * Description: Allocates memory and creates list of spare devices
6271 * avaliable in container. Checks if spare drive size is acceptable.
6272 * Parameters: Pointer to the supertype structure
6273 * Returns: Pointer to the list of spare devices (mdinfo structure) on success,
6276 static struct mdinfo
*get_spares_for_grow(struct supertype
*st
)
6279 struct mdinfo
*disks
, *d
, **dp
;
6280 unsigned long long min_size
= min_acceptable_spare_size_imsm(st
);
6282 /* get list of alldisks in container */
6283 disks
= getinfo_super_disks_imsm(st
);
6287 /* find spare devices on the list */
6289 disks
->array
.spare_disks
= 0;
6293 if (d
->disk
.state
== 0) {
6294 /* check if size is acceptable */
6295 unsigned long long dev_size
;
6296 dev
= makedev(d
->disk
.major
,d
->disk
.minor
);
6298 dev_size_from_id(dev
, &dev_size
) &&
6299 dev_size
>= min_size
) {
6306 disks
->array
.spare_disks
++;
6316 /******************************************************************************
6317 * function: imsm_create_metadata_update_for_reshape
6318 * Function creates update for whole IMSM container.
6320 ******************************************************************************/
6321 static int imsm_create_metadata_update_for_reshape(
6322 struct supertype
*st
,
6323 struct geo_params
*geo
,
6325 struct imsm_update_reshape
**updatep
)
6327 struct intel_super
*super
= st
->sb
;
6328 struct imsm_super
*mpb
= super
->anchor
;
6329 int update_memory_size
= 0;
6330 struct imsm_update_reshape
*u
= NULL
;
6331 struct mdinfo
*spares
= NULL
;
6333 int delta_disks
= 0;
6335 dprintf("imsm_update_metadata_for_reshape(enter) raid_disks = %i\n",
6338 delta_disks
= geo
->raid_disks
- old_raid_disks
;
6340 /* size of all update data without anchor */
6341 update_memory_size
= sizeof(struct imsm_update_reshape
);
6343 /* now add space for spare disks that we need to add. */
6344 update_memory_size
+= sizeof(u
->new_disks
[0]) * (delta_disks
- 1);
6346 u
= calloc(1, update_memory_size
);
6349 "cannot get memory for imsm_update_reshape update\n");
6352 u
->type
= update_reshape_container_disks
;
6353 u
->old_raid_disks
= old_raid_disks
;
6354 u
->new_raid_disks
= geo
->raid_disks
;
6356 /* now get spare disks list
6358 spares
= get_spares_for_grow(st
);
6361 || delta_disks
> spares
->array
.spare_disks
) {
6362 dprintf("imsm: ERROR: Cannot get spare devices.\n");
6366 /* we have got spares
6367 * update disk list in imsm_disk list table in anchor
6369 dprintf("imsm: %i spares are available.\n\n",
6370 spares
->array
.spare_disks
);
6372 for (i
= 0; i
< delta_disks
; i
++) {
6373 struct mdinfo
*dev
= spares
->devs
;
6376 u
->new_disks
[i
] = makedev(dev
->disk
.major
,
6378 dl
= get_disk_super(super
, dev
->disk
.major
, dev
->disk
.minor
);
6379 dl
->index
= mpb
->num_disks
++;
6381 /* Now update the metadata so that container_content will find
6384 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
6386 struct imsm_dev
*dev
= get_imsm_dev(super
, i
);
6387 struct imsm_map
*map
= get_imsm_map(dev
, 0);
6388 map
->num_members
= geo
->raid_disks
;
6389 for (d
= 0; d
< delta_disks
; d
++) {
6390 set_imsm_ord_tbl_ent(map
, old_raid_disks
+ d
,
6391 mpb
->num_disks
- delta_disks
+ d
);
6400 if (i
== delta_disks
) {
6402 return update_memory_size
;
6410 static int imsm_reshape_super(struct supertype
*st
, long long size
, int level
,
6411 int layout
, int chunksize
, int raid_disks
,
6412 char *backup
, char *dev
, int verbouse
)
6414 /* currently we only support increasing the number of devices
6415 * for a container. This increases the number of device for each
6416 * member array. They must all be RAID0 or RAID5.
6420 struct geo_params geo
;
6422 dprintf("imsm: reshape_super called.\n");
6424 memset(&geo
, sizeof(struct geo_params
), 0);
6429 geo
.layout
= layout
;
6430 geo
.chunksize
= chunksize
;
6431 geo
.raid_disks
= raid_disks
;
6433 dprintf("\tfor level : %i\n", geo
.level
);
6434 dprintf("\tfor raid_disks : %i\n", geo
.raid_disks
);
6436 if (experimental() == 0)
6439 /* verify reshape conditions
6440 * on container level we can only increase number of devices. */
6441 if (st
->container_dev
== st
->devnum
) {
6442 /* check for delta_disks > 0
6443 *and supported raid levels 0 and 5 only in container */
6444 int old_raid_disks
= 0;
6445 if (imsm_reshape_is_allowed_on_container(
6446 st
, &geo
, &old_raid_disks
)) {
6447 struct imsm_update_reshape
*u
= NULL
;
6450 len
= imsm_create_metadata_update_for_reshape(
6451 st
, &geo
, old_raid_disks
, &u
);
6455 append_metadata_update(st
, u
, len
);
6457 dprintf("imsm: Cannot prepare "\
6460 dprintf("imsm: Operation is not allowed "\
6461 "on this container\n");
6463 dprintf("imsm: not a container operation\n");
6465 dprintf("imsm: reshape_super Exit code = %i\n", ret_val
);
6469 struct superswitch super_imsm
= {
6471 .examine_super
= examine_super_imsm
,
6472 .brief_examine_super
= brief_examine_super_imsm
,
6473 .brief_examine_subarrays
= brief_examine_subarrays_imsm
,
6474 .export_examine_super
= export_examine_super_imsm
,
6475 .detail_super
= detail_super_imsm
,
6476 .brief_detail_super
= brief_detail_super_imsm
,
6477 .write_init_super
= write_init_super_imsm
,
6478 .validate_geometry
= validate_geometry_imsm
,
6479 .add_to_super
= add_to_super_imsm
,
6480 .remove_from_super
= remove_from_super_imsm
,
6481 .detail_platform
= detail_platform_imsm
,
6482 .kill_subarray
= kill_subarray_imsm
,
6483 .update_subarray
= update_subarray_imsm
,
6484 .load_container
= load_container_imsm
,
6486 .match_home
= match_home_imsm
,
6487 .uuid_from_super
= uuid_from_super_imsm
,
6488 .getinfo_super
= getinfo_super_imsm
,
6489 .getinfo_super_disks
= getinfo_super_disks_imsm
,
6490 .update_super
= update_super_imsm
,
6492 .avail_size
= avail_size_imsm
,
6493 .min_acceptable_spare_size
= min_acceptable_spare_size_imsm
,
6495 .compare_super
= compare_super_imsm
,
6497 .load_super
= load_super_imsm
,
6498 .init_super
= init_super_imsm
,
6499 .store_super
= store_super_imsm
,
6500 .free_super
= free_super_imsm
,
6501 .match_metadata_desc
= match_metadata_desc_imsm
,
6502 .container_content
= container_content_imsm
,
6503 .default_geometry
= default_geometry_imsm
,
6504 .get_disk_controller_domain
= imsm_get_disk_controller_domain
,
6505 .reshape_super
= imsm_reshape_super
,
6512 .open_new
= imsm_open_new
,
6513 .set_array_state
= imsm_set_array_state
,
6514 .set_disk
= imsm_set_disk
,
6515 .sync_metadata
= imsm_sync_metadata
,
6516 .activate_spare
= imsm_activate_spare
,
6517 .process_update
= imsm_process_update
,
6518 .prepare_update
= imsm_prepare_update
,
6519 #endif /* MDASSEMBLE */