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 static __u8
imsm_check_degraded(struct intel_super
*super
, struct imsm_dev
*dev
, int failed
);
1629 static int imsm_count_failed(struct intel_super
*super
, struct imsm_dev
*dev
);
1631 static struct imsm_disk
*get_imsm_missing(struct intel_super
*super
, __u8 index
)
1635 for (d
= super
->missing
; d
; d
= d
->next
)
1636 if (d
->index
== index
)
1641 static void getinfo_super_imsm(struct supertype
*st
, struct mdinfo
*info
, char *map
)
1643 struct intel_super
*super
= st
->sb
;
1644 struct imsm_disk
*disk
;
1645 int map_disks
= info
->array
.raid_disks
;
1646 int max_enough
= -1;
1648 struct imsm_super
*mpb
;
1650 if (super
->current_vol
>= 0) {
1651 getinfo_super_imsm_volume(st
, info
, map
);
1655 /* Set raid_disks to zero so that Assemble will always pull in valid
1658 info
->array
.raid_disks
= 0;
1659 info
->array
.level
= LEVEL_CONTAINER
;
1660 info
->array
.layout
= 0;
1661 info
->array
.md_minor
= -1;
1662 info
->array
.ctime
= 0; /* N/A for imsm */
1663 info
->array
.utime
= 0;
1664 info
->array
.chunk_size
= 0;
1666 info
->disk
.major
= 0;
1667 info
->disk
.minor
= 0;
1668 info
->disk
.raid_disk
= -1;
1669 info
->reshape_active
= 0;
1670 info
->array
.major_version
= -1;
1671 info
->array
.minor_version
= -2;
1672 strcpy(info
->text_version
, "imsm");
1673 info
->safe_mode_delay
= 0;
1674 info
->disk
.number
= -1;
1675 info
->disk
.state
= 0;
1677 info
->recovery_start
= MaxSector
;
1679 /* do we have the all the insync disks that we expect? */
1680 mpb
= super
->anchor
;
1682 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
1683 struct imsm_dev
*dev
= get_imsm_dev(super
, i
);
1684 int failed
, enough
, j
, missing
= 0;
1685 struct imsm_map
*map
;
1688 failed
= imsm_count_failed(super
, dev
);
1689 state
= imsm_check_degraded(super
, dev
, failed
);
1690 map
= get_imsm_map(dev
, dev
->vol
.migr_state
);
1692 /* any newly missing disks?
1693 * (catches single-degraded vs double-degraded)
1695 for (j
= 0; j
< map
->num_members
; j
++) {
1696 __u32 ord
= get_imsm_ord_tbl_ent(dev
, i
);
1697 __u32 idx
= ord_to_idx(ord
);
1699 if (!(ord
& IMSM_ORD_REBUILD
) &&
1700 get_imsm_missing(super
, idx
)) {
1706 if (state
== IMSM_T_STATE_FAILED
)
1708 else if (state
== IMSM_T_STATE_DEGRADED
&&
1709 (state
!= map
->map_state
|| missing
))
1711 else /* we're normal, or already degraded */
1714 /* in the missing/failed disk case check to see
1715 * if at least one array is runnable
1717 max_enough
= max(max_enough
, enough
);
1719 dprintf("%s: enough: %d\n", __func__
, max_enough
);
1720 info
->container_enough
= max_enough
;
1723 __u32 reserved
= imsm_reserved_sectors(super
, super
->disks
);
1725 disk
= &super
->disks
->disk
;
1726 info
->data_offset
= __le32_to_cpu(disk
->total_blocks
) - reserved
;
1727 info
->component_size
= reserved
;
1728 info
->disk
.state
= is_configured(disk
) ? (1 << MD_DISK_ACTIVE
) : 0;
1729 /* we don't change info->disk.raid_disk here because
1730 * this state will be finalized in mdmon after we have
1731 * found the 'most fresh' version of the metadata
1733 info
->disk
.state
|= is_failed(disk
) ? (1 << MD_DISK_FAULTY
) : 0;
1734 info
->disk
.state
|= is_spare(disk
) ? 0 : (1 << MD_DISK_SYNC
);
1737 /* only call uuid_from_super_imsm when this disk is part of a populated container,
1738 * ->compare_super may have updated the 'num_raid_devs' field for spares
1740 if (info
->disk
.state
& (1 << MD_DISK_SYNC
) || super
->anchor
->num_raid_devs
)
1741 uuid_from_super_imsm(st
, info
->uuid
);
1743 memcpy(info
->uuid
, uuid_zero
, sizeof(uuid_zero
));
1745 /* I don't know how to compute 'map' on imsm, so use safe default */
1748 for (i
= 0; i
< map_disks
; i
++)
1754 /* allocates memory and fills disk in mdinfo structure
1755 * for each disk in array */
1756 struct mdinfo
*getinfo_super_disks_imsm(struct supertype
*st
)
1758 struct mdinfo
*mddev
= NULL
;
1759 struct intel_super
*super
= st
->sb
;
1760 struct imsm_disk
*disk
;
1763 if (!super
|| !super
->disks
)
1766 mddev
= malloc(sizeof(*mddev
));
1768 fprintf(stderr
, Name
": Failed to allocate memory.\n");
1771 memset(mddev
, 0, sizeof(*mddev
));
1775 tmp
= malloc(sizeof(*tmp
));
1777 fprintf(stderr
, Name
": Failed to allocate memory.\n");
1782 memset(tmp
, 0, sizeof(*tmp
));
1784 tmp
->next
= mddev
->devs
;
1786 tmp
->disk
.number
= count
++;
1787 tmp
->disk
.major
= dl
->major
;
1788 tmp
->disk
.minor
= dl
->minor
;
1789 tmp
->disk
.state
= is_configured(disk
) ?
1790 (1 << MD_DISK_ACTIVE
) : 0;
1791 tmp
->disk
.state
|= is_failed(disk
) ? (1 << MD_DISK_FAULTY
) : 0;
1792 tmp
->disk
.state
|= is_spare(disk
) ? 0 : (1 << MD_DISK_SYNC
);
1793 tmp
->disk
.raid_disk
= -1;
1799 static int update_super_imsm(struct supertype
*st
, struct mdinfo
*info
,
1800 char *update
, char *devname
, int verbose
,
1801 int uuid_set
, char *homehost
)
1803 /* For 'assemble' and 'force' we need to return non-zero if any
1804 * change was made. For others, the return value is ignored.
1805 * Update options are:
1806 * force-one : This device looks a bit old but needs to be included,
1807 * update age info appropriately.
1808 * assemble: clear any 'faulty' flag to allow this device to
1810 * force-array: Array is degraded but being forced, mark it clean
1811 * if that will be needed to assemble it.
1813 * newdev: not used ????
1814 * grow: Array has gained a new device - this is currently for
1816 * resync: mark as dirty so a resync will happen.
1817 * name: update the name - preserving the homehost
1818 * uuid: Change the uuid of the array to match watch is given
1820 * Following are not relevant for this imsm:
1821 * sparc2.2 : update from old dodgey metadata
1822 * super-minor: change the preferred_minor number
1823 * summaries: update redundant counters.
1824 * homehost: update the recorded homehost
1825 * _reshape_progress: record new reshape_progress position.
1828 struct intel_super
*super
= st
->sb
;
1829 struct imsm_super
*mpb
;
1831 /* we can only update container info */
1832 if (!super
|| super
->current_vol
>= 0 || !super
->anchor
)
1835 mpb
= super
->anchor
;
1837 if (strcmp(update
, "uuid") == 0 && uuid_set
&& !info
->update_private
)
1839 else if (strcmp(update
, "uuid") == 0 && uuid_set
&& info
->update_private
) {
1840 mpb
->orig_family_num
= *((__u32
*) info
->update_private
);
1842 } else if (strcmp(update
, "uuid") == 0) {
1843 __u32
*new_family
= malloc(sizeof(*new_family
));
1845 /* update orig_family_number with the incoming random
1846 * data, report the new effective uuid, and store the
1847 * new orig_family_num for future updates.
1850 memcpy(&mpb
->orig_family_num
, info
->uuid
, sizeof(__u32
));
1851 uuid_from_super_imsm(st
, info
->uuid
);
1852 *new_family
= mpb
->orig_family_num
;
1853 info
->update_private
= new_family
;
1856 } else if (strcmp(update
, "assemble") == 0)
1861 /* successful update? recompute checksum */
1863 mpb
->check_sum
= __le32_to_cpu(__gen_imsm_checksum(mpb
));
1868 static size_t disks_to_mpb_size(int disks
)
1872 size
= sizeof(struct imsm_super
);
1873 size
+= (disks
- 1) * sizeof(struct imsm_disk
);
1874 size
+= 2 * sizeof(struct imsm_dev
);
1875 /* up to 2 maps per raid device (-2 for imsm_maps in imsm_dev */
1876 size
+= (4 - 2) * sizeof(struct imsm_map
);
1877 /* 4 possible disk_ord_tbl's */
1878 size
+= 4 * (disks
- 1) * sizeof(__u32
);
1883 static __u64
avail_size_imsm(struct supertype
*st
, __u64 devsize
)
1885 if (devsize
< (MPB_SECTOR_CNT
+ IMSM_RESERVED_SECTORS
))
1888 return devsize
- (MPB_SECTOR_CNT
+ IMSM_RESERVED_SECTORS
);
1891 static void free_devlist(struct intel_super
*super
)
1893 struct intel_dev
*dv
;
1895 while (super
->devlist
) {
1896 dv
= super
->devlist
->next
;
1897 free(super
->devlist
->dev
);
1898 free(super
->devlist
);
1899 super
->devlist
= dv
;
1903 static void imsm_copy_dev(struct imsm_dev
*dest
, struct imsm_dev
*src
)
1905 memcpy(dest
, src
, sizeof_imsm_dev(src
, 0));
1908 static int compare_super_imsm(struct supertype
*st
, struct supertype
*tst
)
1912 * 0 same, or first was empty, and second was copied
1913 * 1 second had wrong number
1915 * 3 wrong other info
1917 struct intel_super
*first
= st
->sb
;
1918 struct intel_super
*sec
= tst
->sb
;
1926 /* if an anchor does not have num_raid_devs set then it is a free
1929 if (first
->anchor
->num_raid_devs
> 0 &&
1930 sec
->anchor
->num_raid_devs
> 0) {
1931 /* Determine if these disks might ever have been
1932 * related. Further disambiguation can only take place
1933 * in load_super_imsm_all
1935 __u32 first_family
= first
->anchor
->orig_family_num
;
1936 __u32 sec_family
= sec
->anchor
->orig_family_num
;
1938 if (memcmp(first
->anchor
->sig
, sec
->anchor
->sig
,
1939 MAX_SIGNATURE_LENGTH
) != 0)
1942 if (first_family
== 0)
1943 first_family
= first
->anchor
->family_num
;
1944 if (sec_family
== 0)
1945 sec_family
= sec
->anchor
->family_num
;
1947 if (first_family
!= sec_family
)
1953 /* if 'first' is a spare promote it to a populated mpb with sec's
1956 if (first
->anchor
->num_raid_devs
== 0 &&
1957 sec
->anchor
->num_raid_devs
> 0) {
1959 struct intel_dev
*dv
;
1960 struct imsm_dev
*dev
;
1962 /* we need to copy raid device info from sec if an allocation
1963 * fails here we don't associate the spare
1965 for (i
= 0; i
< sec
->anchor
->num_raid_devs
; i
++) {
1966 dv
= malloc(sizeof(*dv
));
1969 dev
= malloc(sizeof_imsm_dev(get_imsm_dev(sec
, i
), 1));
1976 dv
->next
= first
->devlist
;
1977 first
->devlist
= dv
;
1979 if (i
< sec
->anchor
->num_raid_devs
) {
1980 /* allocation failure */
1981 free_devlist(first
);
1982 fprintf(stderr
, "imsm: failed to associate spare\n");
1985 first
->anchor
->num_raid_devs
= sec
->anchor
->num_raid_devs
;
1986 first
->anchor
->orig_family_num
= sec
->anchor
->orig_family_num
;
1987 first
->anchor
->family_num
= sec
->anchor
->family_num
;
1988 memcpy(first
->anchor
->sig
, sec
->anchor
->sig
, MAX_SIGNATURE_LENGTH
);
1989 for (i
= 0; i
< sec
->anchor
->num_raid_devs
; i
++)
1990 imsm_copy_dev(get_imsm_dev(first
, i
), get_imsm_dev(sec
, i
));
1996 static void fd2devname(int fd
, char *name
)
2000 char dname
[PATH_MAX
];
2005 if (fstat(fd
, &st
) != 0)
2007 sprintf(path
, "/sys/dev/block/%d:%d",
2008 major(st
.st_rdev
), minor(st
.st_rdev
));
2010 rv
= readlink(path
, dname
, sizeof(dname
));
2015 nm
= strrchr(dname
, '/');
2017 snprintf(name
, MAX_RAID_SERIAL_LEN
, "/dev/%s", nm
);
2020 extern int scsi_get_serial(int fd
, void *buf
, size_t buf_len
);
2022 static int imsm_read_serial(int fd
, char *devname
,
2023 __u8 serial
[MAX_RAID_SERIAL_LEN
])
2025 unsigned char scsi_serial
[255];
2034 memset(scsi_serial
, 0, sizeof(scsi_serial
));
2036 rv
= scsi_get_serial(fd
, scsi_serial
, sizeof(scsi_serial
));
2038 if (rv
&& check_env("IMSM_DEVNAME_AS_SERIAL")) {
2039 memset(serial
, 0, MAX_RAID_SERIAL_LEN
);
2040 fd2devname(fd
, (char *) serial
);
2047 Name
": Failed to retrieve serial for %s\n",
2052 rsp_len
= scsi_serial
[3];
2056 Name
": Failed to retrieve serial for %s\n",
2060 rsp_buf
= (char *) &scsi_serial
[4];
2062 /* trim all whitespace and non-printable characters and convert
2065 for (i
= 0, dest
= rsp_buf
; i
< rsp_len
; i
++) {
2068 /* ':' is reserved for use in placeholder serial
2069 * numbers for missing disks
2077 len
= dest
- rsp_buf
;
2080 /* truncate leading characters */
2081 if (len
> MAX_RAID_SERIAL_LEN
) {
2082 dest
+= len
- MAX_RAID_SERIAL_LEN
;
2083 len
= MAX_RAID_SERIAL_LEN
;
2086 memset(serial
, 0, MAX_RAID_SERIAL_LEN
);
2087 memcpy(serial
, dest
, len
);
2092 static int serialcmp(__u8
*s1
, __u8
*s2
)
2094 return strncmp((char *) s1
, (char *) s2
, MAX_RAID_SERIAL_LEN
);
2097 static void serialcpy(__u8
*dest
, __u8
*src
)
2099 strncpy((char *) dest
, (char *) src
, MAX_RAID_SERIAL_LEN
);
2103 static struct dl
*serial_to_dl(__u8
*serial
, struct intel_super
*super
)
2107 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
2108 if (serialcmp(dl
->serial
, serial
) == 0)
2115 static struct imsm_disk
*
2116 __serial_to_disk(__u8
*serial
, struct imsm_super
*mpb
, int *idx
)
2120 for (i
= 0; i
< mpb
->num_disks
; i
++) {
2121 struct imsm_disk
*disk
= __get_imsm_disk(mpb
, i
);
2123 if (serialcmp(disk
->serial
, serial
) == 0) {
2134 load_imsm_disk(int fd
, struct intel_super
*super
, char *devname
, int keep_fd
)
2136 struct imsm_disk
*disk
;
2141 __u8 serial
[MAX_RAID_SERIAL_LEN
];
2143 rv
= imsm_read_serial(fd
, devname
, serial
);
2148 dl
= calloc(1, sizeof(*dl
));
2152 Name
": failed to allocate disk buffer for %s\n",
2158 dl
->major
= major(stb
.st_rdev
);
2159 dl
->minor
= minor(stb
.st_rdev
);
2160 dl
->next
= super
->disks
;
2161 dl
->fd
= keep_fd
? fd
: -1;
2162 assert(super
->disks
== NULL
);
2164 serialcpy(dl
->serial
, serial
);
2167 fd2devname(fd
, name
);
2169 dl
->devname
= strdup(devname
);
2171 dl
->devname
= strdup(name
);
2173 /* look up this disk's index in the current anchor */
2174 disk
= __serial_to_disk(dl
->serial
, super
->anchor
, &dl
->index
);
2177 /* only set index on disks that are a member of a
2178 * populated contianer, i.e. one with raid_devs
2180 if (is_failed(&dl
->disk
))
2182 else if (is_spare(&dl
->disk
))
2190 /* When migrating map0 contains the 'destination' state while map1
2191 * contains the current state. When not migrating map0 contains the
2192 * current state. This routine assumes that map[0].map_state is set to
2193 * the current array state before being called.
2195 * Migration is indicated by one of the following states
2196 * 1/ Idle (migr_state=0 map0state=normal||unitialized||degraded||failed)
2197 * 2/ Initialize (migr_state=1 migr_type=MIGR_INIT map0state=normal
2198 * map1state=unitialized)
2199 * 3/ Repair (Resync) (migr_state=1 migr_type=MIGR_REPAIR map0state=normal
2201 * 4/ Rebuild (migr_state=1 migr_type=MIGR_REBUILD map0state=normal
2202 * map1state=degraded)
2204 static void migrate(struct imsm_dev
*dev
, __u8 to_state
, int migr_type
)
2206 struct imsm_map
*dest
;
2207 struct imsm_map
*src
= get_imsm_map(dev
, 0);
2209 dev
->vol
.migr_state
= 1;
2210 set_migr_type(dev
, migr_type
);
2211 dev
->vol
.curr_migr_unit
= 0;
2212 dest
= get_imsm_map(dev
, 1);
2214 /* duplicate and then set the target end state in map[0] */
2215 memcpy(dest
, src
, sizeof_imsm_map(src
));
2216 if ((migr_type
== MIGR_REBUILD
) ||
2217 (migr_type
== MIGR_GEN_MIGR
)) {
2221 for (i
= 0; i
< src
->num_members
; i
++) {
2222 ord
= __le32_to_cpu(src
->disk_ord_tbl
[i
]);
2223 set_imsm_ord_tbl_ent(src
, i
, ord_to_idx(ord
));
2227 src
->map_state
= to_state
;
2230 static void end_migration(struct imsm_dev
*dev
, __u8 map_state
)
2232 struct imsm_map
*map
= get_imsm_map(dev
, 0);
2233 struct imsm_map
*prev
= get_imsm_map(dev
, dev
->vol
.migr_state
);
2236 /* merge any IMSM_ORD_REBUILD bits that were not successfully
2237 * completed in the last migration.
2239 * FIXME add support for raid-level-migration
2241 for (i
= 0; i
< prev
->num_members
; i
++)
2242 for (j
= 0; j
< map
->num_members
; j
++)
2243 /* during online capacity expansion
2244 * disks position can be changed if takeover is used
2246 if (ord_to_idx(map
->disk_ord_tbl
[j
]) ==
2247 ord_to_idx(prev
->disk_ord_tbl
[i
])) {
2248 map
->disk_ord_tbl
[j
] |= prev
->disk_ord_tbl
[i
];
2252 dev
->vol
.migr_state
= 0;
2253 dev
->vol
.migr_type
= 0;
2254 dev
->vol
.curr_migr_unit
= 0;
2255 map
->map_state
= map_state
;
2259 static int parse_raid_devices(struct intel_super
*super
)
2262 struct imsm_dev
*dev_new
;
2263 size_t len
, len_migr
;
2264 size_t space_needed
= 0;
2265 struct imsm_super
*mpb
= super
->anchor
;
2267 for (i
= 0; i
< super
->anchor
->num_raid_devs
; i
++) {
2268 struct imsm_dev
*dev_iter
= __get_imsm_dev(super
->anchor
, i
);
2269 struct intel_dev
*dv
;
2271 len
= sizeof_imsm_dev(dev_iter
, 0);
2272 len_migr
= sizeof_imsm_dev(dev_iter
, 1);
2274 space_needed
+= len_migr
- len
;
2276 dv
= malloc(sizeof(*dv
));
2279 dev_new
= malloc(len_migr
);
2284 imsm_copy_dev(dev_new
, dev_iter
);
2287 dv
->next
= super
->devlist
;
2288 super
->devlist
= dv
;
2291 /* ensure that super->buf is large enough when all raid devices
2294 if (__le32_to_cpu(mpb
->mpb_size
) + space_needed
> super
->len
) {
2297 len
= ROUND_UP(__le32_to_cpu(mpb
->mpb_size
) + space_needed
, 512);
2298 if (posix_memalign(&buf
, 512, len
) != 0)
2301 memcpy(buf
, super
->buf
, super
->len
);
2302 memset(buf
+ super
->len
, 0, len
- super
->len
);
2311 /* retrieve a pointer to the bbm log which starts after all raid devices */
2312 struct bbm_log
*__get_imsm_bbm_log(struct imsm_super
*mpb
)
2316 if (__le32_to_cpu(mpb
->bbm_log_size
)) {
2318 ptr
+= mpb
->mpb_size
- __le32_to_cpu(mpb
->bbm_log_size
);
2324 static void __free_imsm(struct intel_super
*super
, int free_disks
);
2326 /* load_imsm_mpb - read matrix metadata
2327 * allocates super->mpb to be freed by free_super
2329 static int load_imsm_mpb(int fd
, struct intel_super
*super
, char *devname
)
2331 unsigned long long dsize
;
2332 unsigned long long sectors
;
2334 struct imsm_super
*anchor
;
2337 get_dev_size(fd
, NULL
, &dsize
);
2341 Name
": %s: device to small for imsm\n",
2346 if (lseek64(fd
, dsize
- (512 * 2), SEEK_SET
) < 0) {
2349 Name
": Cannot seek to anchor block on %s: %s\n",
2350 devname
, strerror(errno
));
2354 if (posix_memalign((void**)&anchor
, 512, 512) != 0) {
2357 Name
": Failed to allocate imsm anchor buffer"
2358 " on %s\n", devname
);
2361 if (read(fd
, anchor
, 512) != 512) {
2364 Name
": Cannot read anchor block on %s: %s\n",
2365 devname
, strerror(errno
));
2370 if (strncmp((char *) anchor
->sig
, MPB_SIGNATURE
, MPB_SIG_LEN
) != 0) {
2373 Name
": no IMSM anchor on %s\n", devname
);
2378 __free_imsm(super
, 0);
2379 super
->len
= ROUND_UP(anchor
->mpb_size
, 512);
2380 if (posix_memalign(&super
->buf
, 512, super
->len
) != 0) {
2383 Name
": unable to allocate %zu byte mpb buffer\n",
2388 memcpy(super
->buf
, anchor
, 512);
2390 sectors
= mpb_sectors(anchor
) - 1;
2393 check_sum
= __gen_imsm_checksum(super
->anchor
);
2394 if (check_sum
!= __le32_to_cpu(super
->anchor
->check_sum
)) {
2397 Name
": IMSM checksum %x != %x on %s\n",
2399 __le32_to_cpu(super
->anchor
->check_sum
),
2407 /* read the extended mpb */
2408 if (lseek64(fd
, dsize
- (512 * (2 + sectors
)), SEEK_SET
) < 0) {
2411 Name
": Cannot seek to extended mpb on %s: %s\n",
2412 devname
, strerror(errno
));
2416 if ((unsigned)read(fd
, super
->buf
+ 512, super
->len
- 512) != super
->len
- 512) {
2419 Name
": Cannot read extended mpb on %s: %s\n",
2420 devname
, strerror(errno
));
2424 check_sum
= __gen_imsm_checksum(super
->anchor
);
2425 if (check_sum
!= __le32_to_cpu(super
->anchor
->check_sum
)) {
2428 Name
": IMSM checksum %x != %x on %s\n",
2429 check_sum
, __le32_to_cpu(super
->anchor
->check_sum
),
2434 /* FIXME the BBM log is disk specific so we cannot use this global
2435 * buffer for all disks. Ok for now since we only look at the global
2436 * bbm_log_size parameter to gate assembly
2438 super
->bbm_log
= __get_imsm_bbm_log(super
->anchor
);
2444 load_and_parse_mpb(int fd
, struct intel_super
*super
, char *devname
, int keep_fd
)
2448 err
= load_imsm_mpb(fd
, super
, devname
);
2451 err
= load_imsm_disk(fd
, super
, devname
, keep_fd
);
2454 err
= parse_raid_devices(super
);
2459 static void __free_imsm_disk(struct dl
*d
)
2471 static void free_imsm_disks(struct intel_super
*super
)
2475 while (super
->disks
) {
2477 super
->disks
= d
->next
;
2478 __free_imsm_disk(d
);
2480 while (super
->missing
) {
2482 super
->missing
= d
->next
;
2483 __free_imsm_disk(d
);
2488 /* free all the pieces hanging off of a super pointer */
2489 static void __free_imsm(struct intel_super
*super
, int free_disks
)
2496 free_imsm_disks(super
);
2497 free_devlist(super
);
2499 free((void *) super
->hba
);
2504 static void free_imsm(struct intel_super
*super
)
2506 __free_imsm(super
, 1);
2510 static void free_super_imsm(struct supertype
*st
)
2512 struct intel_super
*super
= st
->sb
;
2521 static struct intel_super
*alloc_super(void)
2523 struct intel_super
*super
= malloc(sizeof(*super
));
2526 memset(super
, 0, sizeof(*super
));
2527 super
->current_vol
= -1;
2528 super
->create_offset
= ~((__u32
) 0);
2529 if (!check_env("IMSM_NO_PLATFORM"))
2530 super
->orom
= find_imsm_orom();
2531 if (super
->orom
&& !check_env("IMSM_TEST_OROM")) {
2532 struct sys_dev
*list
, *ent
;
2534 /* find the first intel ahci controller */
2535 list
= find_driver_devices("pci", "ahci");
2536 for (ent
= list
; ent
; ent
= ent
->next
)
2537 if (devpath_to_vendor(ent
->path
) == 0x8086)
2540 super
->hba
= ent
->path
;
2543 free_sys_dev(&list
);
2551 /* find_missing - helper routine for load_super_imsm_all that identifies
2552 * disks that have disappeared from the system. This routine relies on
2553 * the mpb being uptodate, which it is at load time.
2555 static int find_missing(struct intel_super
*super
)
2558 struct imsm_super
*mpb
= super
->anchor
;
2560 struct imsm_disk
*disk
;
2562 for (i
= 0; i
< mpb
->num_disks
; i
++) {
2563 disk
= __get_imsm_disk(mpb
, i
);
2564 dl
= serial_to_dl(disk
->serial
, super
);
2568 dl
= malloc(sizeof(*dl
));
2574 dl
->devname
= strdup("missing");
2576 serialcpy(dl
->serial
, disk
->serial
);
2579 dl
->next
= super
->missing
;
2580 super
->missing
= dl
;
2586 static struct intel_disk
*disk_list_get(__u8
*serial
, struct intel_disk
*disk_list
)
2588 struct intel_disk
*idisk
= disk_list
;
2591 if (serialcmp(idisk
->disk
.serial
, serial
) == 0)
2593 idisk
= idisk
->next
;
2599 static int __prep_thunderdome(struct intel_super
**table
, int tbl_size
,
2600 struct intel_super
*super
,
2601 struct intel_disk
**disk_list
)
2603 struct imsm_disk
*d
= &super
->disks
->disk
;
2604 struct imsm_super
*mpb
= super
->anchor
;
2607 for (i
= 0; i
< tbl_size
; i
++) {
2608 struct imsm_super
*tbl_mpb
= table
[i
]->anchor
;
2609 struct imsm_disk
*tbl_d
= &table
[i
]->disks
->disk
;
2611 if (tbl_mpb
->family_num
== mpb
->family_num
) {
2612 if (tbl_mpb
->check_sum
== mpb
->check_sum
) {
2613 dprintf("%s: mpb from %d:%d matches %d:%d\n",
2614 __func__
, super
->disks
->major
,
2615 super
->disks
->minor
,
2616 table
[i
]->disks
->major
,
2617 table
[i
]->disks
->minor
);
2621 if (((is_configured(d
) && !is_configured(tbl_d
)) ||
2622 is_configured(d
) == is_configured(tbl_d
)) &&
2623 tbl_mpb
->generation_num
< mpb
->generation_num
) {
2624 /* current version of the mpb is a
2625 * better candidate than the one in
2626 * super_table, but copy over "cross
2627 * generational" status
2629 struct intel_disk
*idisk
;
2631 dprintf("%s: mpb from %d:%d replaces %d:%d\n",
2632 __func__
, super
->disks
->major
,
2633 super
->disks
->minor
,
2634 table
[i
]->disks
->major
,
2635 table
[i
]->disks
->minor
);
2637 idisk
= disk_list_get(tbl_d
->serial
, *disk_list
);
2638 if (idisk
&& is_failed(&idisk
->disk
))
2639 tbl_d
->status
|= FAILED_DISK
;
2642 struct intel_disk
*idisk
;
2643 struct imsm_disk
*disk
;
2645 /* tbl_mpb is more up to date, but copy
2646 * over cross generational status before
2649 disk
= __serial_to_disk(d
->serial
, mpb
, NULL
);
2650 if (disk
&& is_failed(disk
))
2651 d
->status
|= FAILED_DISK
;
2653 idisk
= disk_list_get(d
->serial
, *disk_list
);
2656 if (disk
&& is_configured(disk
))
2657 idisk
->disk
.status
|= CONFIGURED_DISK
;
2660 dprintf("%s: mpb from %d:%d prefer %d:%d\n",
2661 __func__
, super
->disks
->major
,
2662 super
->disks
->minor
,
2663 table
[i
]->disks
->major
,
2664 table
[i
]->disks
->minor
);
2672 table
[tbl_size
++] = super
;
2676 /* update/extend the merged list of imsm_disk records */
2677 for (j
= 0; j
< mpb
->num_disks
; j
++) {
2678 struct imsm_disk
*disk
= __get_imsm_disk(mpb
, j
);
2679 struct intel_disk
*idisk
;
2681 idisk
= disk_list_get(disk
->serial
, *disk_list
);
2683 idisk
->disk
.status
|= disk
->status
;
2684 if (is_configured(&idisk
->disk
) ||
2685 is_failed(&idisk
->disk
))
2686 idisk
->disk
.status
&= ~(SPARE_DISK
);
2688 idisk
= calloc(1, sizeof(*idisk
));
2691 idisk
->owner
= IMSM_UNKNOWN_OWNER
;
2692 idisk
->disk
= *disk
;
2693 idisk
->next
= *disk_list
;
2697 if (serialcmp(idisk
->disk
.serial
, d
->serial
) == 0)
2704 static struct intel_super
*
2705 validate_members(struct intel_super
*super
, struct intel_disk
*disk_list
,
2708 struct imsm_super
*mpb
= super
->anchor
;
2712 for (i
= 0; i
< mpb
->num_disks
; i
++) {
2713 struct imsm_disk
*disk
= __get_imsm_disk(mpb
, i
);
2714 struct intel_disk
*idisk
;
2716 idisk
= disk_list_get(disk
->serial
, disk_list
);
2718 if (idisk
->owner
== owner
||
2719 idisk
->owner
== IMSM_UNKNOWN_OWNER
)
2722 dprintf("%s: '%.16s' owner %d != %d\n",
2723 __func__
, disk
->serial
, idisk
->owner
,
2726 dprintf("%s: unknown disk %x [%d]: %.16s\n",
2727 __func__
, __le32_to_cpu(mpb
->family_num
), i
,
2733 if (ok_count
== mpb
->num_disks
)
2738 static void show_conflicts(__u32 family_num
, struct intel_super
*super_list
)
2740 struct intel_super
*s
;
2742 for (s
= super_list
; s
; s
= s
->next
) {
2743 if (family_num
!= s
->anchor
->family_num
)
2745 fprintf(stderr
, "Conflict, offlining family %#x on '%s'\n",
2746 __le32_to_cpu(family_num
), s
->disks
->devname
);
2750 static struct intel_super
*
2751 imsm_thunderdome(struct intel_super
**super_list
, int len
)
2753 struct intel_super
*super_table
[len
];
2754 struct intel_disk
*disk_list
= NULL
;
2755 struct intel_super
*champion
, *spare
;
2756 struct intel_super
*s
, **del
;
2761 memset(super_table
, 0, sizeof(super_table
));
2762 for (s
= *super_list
; s
; s
= s
->next
)
2763 tbl_size
= __prep_thunderdome(super_table
, tbl_size
, s
, &disk_list
);
2765 for (i
= 0; i
< tbl_size
; i
++) {
2766 struct imsm_disk
*d
;
2767 struct intel_disk
*idisk
;
2768 struct imsm_super
*mpb
= super_table
[i
]->anchor
;
2771 d
= &s
->disks
->disk
;
2773 /* 'd' must appear in merged disk list for its
2774 * configuration to be valid
2776 idisk
= disk_list_get(d
->serial
, disk_list
);
2777 if (idisk
&& idisk
->owner
== i
)
2778 s
= validate_members(s
, disk_list
, i
);
2783 dprintf("%s: marking family: %#x from %d:%d offline\n",
2784 __func__
, mpb
->family_num
,
2785 super_table
[i
]->disks
->major
,
2786 super_table
[i
]->disks
->minor
);
2790 /* This is where the mdadm implementation differs from the Windows
2791 * driver which has no strict concept of a container. We can only
2792 * assemble one family from a container, so when returning a prodigal
2793 * array member to this system the code will not be able to disambiguate
2794 * the container contents that should be assembled ("foreign" versus
2795 * "local"). It requires user intervention to set the orig_family_num
2796 * to a new value to establish a new container. The Windows driver in
2797 * this situation fixes up the volume name in place and manages the
2798 * foreign array as an independent entity.
2803 for (i
= 0; i
< tbl_size
; i
++) {
2804 struct intel_super
*tbl_ent
= super_table
[i
];
2810 if (tbl_ent
->anchor
->num_raid_devs
== 0) {
2815 if (s
&& !is_spare
) {
2816 show_conflicts(tbl_ent
->anchor
->family_num
, *super_list
);
2818 } else if (!s
&& !is_spare
)
2831 fprintf(stderr
, "Chose family %#x on '%s', "
2832 "assemble conflicts to new container with '--update=uuid'\n",
2833 __le32_to_cpu(s
->anchor
->family_num
), s
->disks
->devname
);
2835 /* collect all dl's onto 'champion', and update them to
2836 * champion's version of the status
2838 for (s
= *super_list
; s
; s
= s
->next
) {
2839 struct imsm_super
*mpb
= champion
->anchor
;
2840 struct dl
*dl
= s
->disks
;
2845 for (i
= 0; i
< mpb
->num_disks
; i
++) {
2846 struct imsm_disk
*disk
;
2848 disk
= __serial_to_disk(dl
->serial
, mpb
, &dl
->index
);
2851 /* only set index on disks that are a member of
2852 * a populated contianer, i.e. one with
2855 if (is_failed(&dl
->disk
))
2857 else if (is_spare(&dl
->disk
))
2863 if (i
>= mpb
->num_disks
) {
2864 struct intel_disk
*idisk
;
2866 idisk
= disk_list_get(dl
->serial
, disk_list
);
2867 if (idisk
&& is_spare(&idisk
->disk
) &&
2868 !is_failed(&idisk
->disk
) && !is_configured(&idisk
->disk
))
2876 dl
->next
= champion
->disks
;
2877 champion
->disks
= dl
;
2881 /* delete 'champion' from super_list */
2882 for (del
= super_list
; *del
; ) {
2883 if (*del
== champion
) {
2884 *del
= (*del
)->next
;
2887 del
= &(*del
)->next
;
2889 champion
->next
= NULL
;
2893 struct intel_disk
*idisk
= disk_list
;
2895 disk_list
= disk_list
->next
;
2902 static int load_super_imsm_all(struct supertype
*st
, int fd
, void **sbp
,
2906 struct intel_super
*super_list
= NULL
;
2907 struct intel_super
*super
= NULL
;
2908 int devnum
= fd2devnum(fd
);
2914 /* check if 'fd' an opened container */
2915 sra
= sysfs_read(fd
, 0, GET_LEVEL
|GET_VERSION
|GET_DEVS
|GET_STATE
);
2919 if (sra
->array
.major_version
!= -1 ||
2920 sra
->array
.minor_version
!= -2 ||
2921 strcmp(sra
->text_version
, "imsm") != 0) {
2926 for (sd
= sra
->devs
, i
= 0; sd
; sd
= sd
->next
, i
++) {
2927 struct intel_super
*s
= alloc_super();
2934 s
->next
= super_list
;
2938 sprintf(nm
, "%d:%d", sd
->disk
.major
, sd
->disk
.minor
);
2939 dfd
= dev_open(nm
, O_RDWR
);
2943 err
= load_and_parse_mpb(dfd
, s
, NULL
, 1);
2945 /* retry the load if we might have raced against mdmon */
2946 if (err
== 3 && mdmon_running(devnum
))
2947 for (retry
= 0; retry
< 3; retry
++) {
2949 err
= load_and_parse_mpb(dfd
, s
, NULL
, 1);
2957 /* all mpbs enter, maybe one leaves */
2958 super
= imsm_thunderdome(&super_list
, i
);
2964 if (find_missing(super
) != 0) {
2972 while (super_list
) {
2973 struct intel_super
*s
= super_list
;
2975 super_list
= super_list
->next
;
2984 st
->container_dev
= devnum
;
2985 if (err
== 0 && st
->ss
== NULL
) {
2986 st
->ss
= &super_imsm
;
2987 st
->minor_version
= 0;
2988 st
->max_devs
= IMSM_MAX_DEVICES
;
2993 static int load_container_imsm(struct supertype
*st
, int fd
, char *devname
)
2995 return load_super_imsm_all(st
, fd
, &st
->sb
, devname
);
2999 static int load_super_imsm(struct supertype
*st
, int fd
, char *devname
)
3001 struct intel_super
*super
;
3005 if (load_super_imsm_all(st
, fd
, &st
->sb
, devname
) == 0)
3009 if (test_partition(fd
))
3010 /* IMSM not allowed on partitions */
3013 free_super_imsm(st
);
3015 super
= alloc_super();
3018 Name
": malloc of %zu failed.\n",
3023 rv
= load_and_parse_mpb(fd
, super
, devname
, 0);
3028 Name
": Failed to load all information "
3029 "sections on %s\n", devname
);
3035 if (st
->ss
== NULL
) {
3036 st
->ss
= &super_imsm
;
3037 st
->minor_version
= 0;
3038 st
->max_devs
= IMSM_MAX_DEVICES
;
3043 static __u16
info_to_blocks_per_strip(mdu_array_info_t
*info
)
3045 if (info
->level
== 1)
3047 return info
->chunk_size
>> 9;
3050 static __u32
info_to_num_data_stripes(mdu_array_info_t
*info
, int num_domains
)
3054 num_stripes
= (info
->size
* 2) / info_to_blocks_per_strip(info
);
3055 num_stripes
/= num_domains
;
3060 static __u32
info_to_blocks_per_member(mdu_array_info_t
*info
)
3062 if (info
->level
== 1)
3063 return info
->size
* 2;
3065 return (info
->size
* 2) & ~(info_to_blocks_per_strip(info
) - 1);
3068 static void imsm_update_version_info(struct intel_super
*super
)
3070 /* update the version and attributes */
3071 struct imsm_super
*mpb
= super
->anchor
;
3073 struct imsm_dev
*dev
;
3074 struct imsm_map
*map
;
3077 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
3078 dev
= get_imsm_dev(super
, i
);
3079 map
= get_imsm_map(dev
, 0);
3080 if (__le32_to_cpu(dev
->size_high
) > 0)
3081 mpb
->attributes
|= MPB_ATTRIB_2TB
;
3083 /* FIXME detect when an array spans a port multiplier */
3085 mpb
->attributes
|= MPB_ATTRIB_PM
;
3088 if (mpb
->num_raid_devs
> 1 ||
3089 mpb
->attributes
!= MPB_ATTRIB_CHECKSUM_VERIFY
) {
3090 version
= MPB_VERSION_ATTRIBS
;
3091 switch (get_imsm_raid_level(map
)) {
3092 case 0: mpb
->attributes
|= MPB_ATTRIB_RAID0
; break;
3093 case 1: mpb
->attributes
|= MPB_ATTRIB_RAID1
; break;
3094 case 10: mpb
->attributes
|= MPB_ATTRIB_RAID10
; break;
3095 case 5: mpb
->attributes
|= MPB_ATTRIB_RAID5
; break;
3098 if (map
->num_members
>= 5)
3099 version
= MPB_VERSION_5OR6_DISK_ARRAY
;
3100 else if (dev
->status
== DEV_CLONE_N_GO
)
3101 version
= MPB_VERSION_CNG
;
3102 else if (get_imsm_raid_level(map
) == 5)
3103 version
= MPB_VERSION_RAID5
;
3104 else if (map
->num_members
>= 3)
3105 version
= MPB_VERSION_3OR4_DISK_ARRAY
;
3106 else if (get_imsm_raid_level(map
) == 1)
3107 version
= MPB_VERSION_RAID1
;
3109 version
= MPB_VERSION_RAID0
;
3111 strcpy(((char *) mpb
->sig
) + strlen(MPB_SIGNATURE
), version
);
3115 static int check_name(struct intel_super
*super
, char *name
, int quiet
)
3117 struct imsm_super
*mpb
= super
->anchor
;
3118 char *reason
= NULL
;
3121 if (strlen(name
) > MAX_RAID_SERIAL_LEN
)
3122 reason
= "must be 16 characters or less";
3124 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
3125 struct imsm_dev
*dev
= get_imsm_dev(super
, i
);
3127 if (strncmp((char *) dev
->volume
, name
, MAX_RAID_SERIAL_LEN
) == 0) {
3128 reason
= "already exists";
3133 if (reason
&& !quiet
)
3134 fprintf(stderr
, Name
": imsm volume name %s\n", reason
);
3139 static int init_super_imsm_volume(struct supertype
*st
, mdu_array_info_t
*info
,
3140 unsigned long long size
, char *name
,
3141 char *homehost
, int *uuid
)
3143 /* We are creating a volume inside a pre-existing container.
3144 * so st->sb is already set.
3146 struct intel_super
*super
= st
->sb
;
3147 struct imsm_super
*mpb
= super
->anchor
;
3148 struct intel_dev
*dv
;
3149 struct imsm_dev
*dev
;
3150 struct imsm_vol
*vol
;
3151 struct imsm_map
*map
;
3152 int idx
= mpb
->num_raid_devs
;
3154 unsigned long long array_blocks
;
3155 size_t size_old
, size_new
;
3156 __u32 num_data_stripes
;
3158 if (super
->orom
&& mpb
->num_raid_devs
>= super
->orom
->vpa
) {
3159 fprintf(stderr
, Name
": This imsm-container already has the "
3160 "maximum of %d volumes\n", super
->orom
->vpa
);
3164 /* ensure the mpb is large enough for the new data */
3165 size_old
= __le32_to_cpu(mpb
->mpb_size
);
3166 size_new
= disks_to_mpb_size(info
->nr_disks
);
3167 if (size_new
> size_old
) {
3169 size_t size_round
= ROUND_UP(size_new
, 512);
3171 if (posix_memalign(&mpb_new
, 512, size_round
) != 0) {
3172 fprintf(stderr
, Name
": could not allocate new mpb\n");
3175 memcpy(mpb_new
, mpb
, size_old
);
3178 super
->anchor
= mpb_new
;
3179 mpb
->mpb_size
= __cpu_to_le32(size_new
);
3180 memset(mpb_new
+ size_old
, 0, size_round
- size_old
);
3182 super
->current_vol
= idx
;
3183 /* when creating the first raid device in this container set num_disks
3184 * to zero, i.e. delete this spare and add raid member devices in
3185 * add_to_super_imsm_volume()
3187 if (super
->current_vol
== 0)
3190 if (!check_name(super
, name
, 0))
3192 dv
= malloc(sizeof(*dv
));
3194 fprintf(stderr
, Name
": failed to allocate device list entry\n");
3197 dev
= malloc(sizeof(*dev
) + sizeof(__u32
) * (info
->raid_disks
- 1));
3200 fprintf(stderr
, Name
": could not allocate raid device\n");
3203 strncpy((char *) dev
->volume
, name
, MAX_RAID_SERIAL_LEN
);
3204 if (info
->level
== 1)
3205 array_blocks
= info_to_blocks_per_member(info
);
3207 array_blocks
= calc_array_size(info
->level
, info
->raid_disks
,
3208 info
->layout
, info
->chunk_size
,
3210 /* round array size down to closest MB */
3211 array_blocks
= (array_blocks
>> SECT_PER_MB_SHIFT
) << SECT_PER_MB_SHIFT
;
3213 dev
->size_low
= __cpu_to_le32((__u32
) array_blocks
);
3214 dev
->size_high
= __cpu_to_le32((__u32
) (array_blocks
>> 32));
3215 dev
->status
= __cpu_to_le32(0);
3216 dev
->reserved_blocks
= __cpu_to_le32(0);
3218 vol
->migr_state
= 0;
3219 set_migr_type(dev
, MIGR_INIT
);
3221 vol
->curr_migr_unit
= 0;
3222 map
= get_imsm_map(dev
, 0);
3223 map
->pba_of_lba0
= __cpu_to_le32(super
->create_offset
);
3224 map
->blocks_per_member
= __cpu_to_le32(info_to_blocks_per_member(info
));
3225 map
->blocks_per_strip
= __cpu_to_le16(info_to_blocks_per_strip(info
));
3226 map
->failed_disk_num
= ~0;
3227 map
->map_state
= info
->level
? IMSM_T_STATE_UNINITIALIZED
:
3228 IMSM_T_STATE_NORMAL
;
3231 if (info
->level
== 1 && info
->raid_disks
> 2) {
3234 fprintf(stderr
, Name
": imsm does not support more than 2 disks"
3235 "in a raid1 volume\n");
3239 map
->raid_level
= info
->level
;
3240 if (info
->level
== 10) {
3241 map
->raid_level
= 1;
3242 map
->num_domains
= info
->raid_disks
/ 2;
3243 } else if (info
->level
== 1)
3244 map
->num_domains
= info
->raid_disks
;
3246 map
->num_domains
= 1;
3248 num_data_stripes
= info_to_num_data_stripes(info
, map
->num_domains
);
3249 map
->num_data_stripes
= __cpu_to_le32(num_data_stripes
);
3251 map
->num_members
= info
->raid_disks
;
3252 for (i
= 0; i
< map
->num_members
; i
++) {
3253 /* initialized in add_to_super */
3254 set_imsm_ord_tbl_ent(map
, i
, IMSM_ORD_REBUILD
);
3256 mpb
->num_raid_devs
++;
3259 dv
->index
= super
->current_vol
;
3260 dv
->next
= super
->devlist
;
3261 super
->devlist
= dv
;
3263 imsm_update_version_info(super
);
3268 static int init_super_imsm(struct supertype
*st
, mdu_array_info_t
*info
,
3269 unsigned long long size
, char *name
,
3270 char *homehost
, int *uuid
)
3272 /* This is primarily called by Create when creating a new array.
3273 * We will then get add_to_super called for each component, and then
3274 * write_init_super called to write it out to each device.
3275 * For IMSM, Create can create on fresh devices or on a pre-existing
3277 * To create on a pre-existing array a different method will be called.
3278 * This one is just for fresh drives.
3280 struct intel_super
*super
;
3281 struct imsm_super
*mpb
;
3286 return init_super_imsm_volume(st
, info
, size
, name
, homehost
, uuid
);
3289 mpb_size
= disks_to_mpb_size(info
->nr_disks
);
3293 super
= alloc_super();
3294 if (super
&& posix_memalign(&super
->buf
, 512, mpb_size
) != 0) {
3299 fprintf(stderr
, Name
3300 ": %s could not allocate superblock\n", __func__
);
3303 memset(super
->buf
, 0, mpb_size
);
3305 mpb
->mpb_size
= __cpu_to_le32(mpb_size
);
3309 /* zeroing superblock */
3313 mpb
->attributes
= MPB_ATTRIB_CHECKSUM_VERIFY
;
3315 version
= (char *) mpb
->sig
;
3316 strcpy(version
, MPB_SIGNATURE
);
3317 version
+= strlen(MPB_SIGNATURE
);
3318 strcpy(version
, MPB_VERSION_RAID0
);
3324 static int add_to_super_imsm_volume(struct supertype
*st
, mdu_disk_info_t
*dk
,
3325 int fd
, char *devname
)
3327 struct intel_super
*super
= st
->sb
;
3328 struct imsm_super
*mpb
= super
->anchor
;
3330 struct imsm_dev
*dev
;
3331 struct imsm_map
*map
;
3334 dev
= get_imsm_dev(super
, super
->current_vol
);
3335 map
= get_imsm_map(dev
, 0);
3337 if (! (dk
->state
& (1<<MD_DISK_SYNC
))) {
3338 fprintf(stderr
, Name
": %s: Cannot add spare devices to IMSM volume\n",
3344 /* we're doing autolayout so grab the pre-marked (in
3345 * validate_geometry) raid_disk
3347 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
3348 if (dl
->raiddisk
== dk
->raid_disk
)
3351 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
3352 if (dl
->major
== dk
->major
&&
3353 dl
->minor
== dk
->minor
)
3358 fprintf(stderr
, Name
": %s is not a member of the same container\n", devname
);
3362 /* add a pristine spare to the metadata */
3363 if (dl
->index
< 0) {
3364 dl
->index
= super
->anchor
->num_disks
;
3365 super
->anchor
->num_disks
++;
3367 /* Check the device has not already been added */
3368 slot
= get_imsm_disk_slot(map
, dl
->index
);
3370 (get_imsm_ord_tbl_ent(dev
, slot
) & IMSM_ORD_REBUILD
) == 0) {
3371 fprintf(stderr
, Name
": %s has been included in this array twice\n",
3375 set_imsm_ord_tbl_ent(map
, dk
->number
, dl
->index
);
3376 dl
->disk
.status
= CONFIGURED_DISK
;
3378 /* if we are creating the first raid device update the family number */
3379 if (super
->current_vol
== 0) {
3381 struct imsm_dev
*_dev
= __get_imsm_dev(mpb
, 0);
3382 struct imsm_disk
*_disk
= __get_imsm_disk(mpb
, dl
->index
);
3384 if (!_dev
|| !_disk
) {
3385 fprintf(stderr
, Name
": BUG mpb setup error\n");
3391 sum
+= __gen_imsm_checksum(mpb
);
3392 mpb
->family_num
= __cpu_to_le32(sum
);
3393 mpb
->orig_family_num
= mpb
->family_num
;
3399 static int add_to_super_imsm(struct supertype
*st
, mdu_disk_info_t
*dk
,
3400 int fd
, char *devname
)
3402 struct intel_super
*super
= st
->sb
;
3404 unsigned long long size
;
3409 /* if we are on an RAID enabled platform check that the disk is
3410 * attached to the raid controller
3412 if (super
->hba
&& !disk_attached_to_hba(fd
, super
->hba
)) {
3414 Name
": %s is not attached to the raid controller: %s\n",
3415 devname
? : "disk", super
->hba
);
3419 if (super
->current_vol
>= 0)
3420 return add_to_super_imsm_volume(st
, dk
, fd
, devname
);
3423 dd
= malloc(sizeof(*dd
));
3426 Name
": malloc failed %s:%d.\n", __func__
, __LINE__
);
3429 memset(dd
, 0, sizeof(*dd
));
3430 dd
->major
= major(stb
.st_rdev
);
3431 dd
->minor
= minor(stb
.st_rdev
);
3433 dd
->devname
= devname
? strdup(devname
) : NULL
;
3436 dd
->action
= DISK_ADD
;
3437 rv
= imsm_read_serial(fd
, devname
, dd
->serial
);
3440 Name
": failed to retrieve scsi serial, aborting\n");
3445 get_dev_size(fd
, NULL
, &size
);
3447 serialcpy(dd
->disk
.serial
, dd
->serial
);
3448 dd
->disk
.total_blocks
= __cpu_to_le32(size
);
3449 dd
->disk
.status
= SPARE_DISK
;
3450 if (sysfs_disk_to_scsi_id(fd
, &id
) == 0)
3451 dd
->disk
.scsi_id
= __cpu_to_le32(id
);
3453 dd
->disk
.scsi_id
= __cpu_to_le32(0);
3455 if (st
->update_tail
) {
3456 dd
->next
= super
->disk_mgmt_list
;
3457 super
->disk_mgmt_list
= dd
;
3459 dd
->next
= super
->disks
;
3467 static int remove_from_super_imsm(struct supertype
*st
, mdu_disk_info_t
*dk
)
3469 struct intel_super
*super
= st
->sb
;
3472 /* remove from super works only in mdmon - for communication
3473 * manager - monitor. Check if communication memory buffer
3476 if (!st
->update_tail
) {
3478 Name
": %s shall be used in mdmon context only"
3479 "(line %d).\n", __func__
, __LINE__
);
3482 dd
= malloc(sizeof(*dd
));
3485 Name
": malloc failed %s:%d.\n", __func__
, __LINE__
);
3488 memset(dd
, 0, sizeof(*dd
));
3489 dd
->major
= dk
->major
;
3490 dd
->minor
= dk
->minor
;
3493 dd
->disk
.status
= SPARE_DISK
;
3494 dd
->action
= DISK_REMOVE
;
3496 dd
->next
= super
->disk_mgmt_list
;
3497 super
->disk_mgmt_list
= dd
;
3503 static int store_imsm_mpb(int fd
, struct imsm_super
*mpb
);
3507 struct imsm_super anchor
;
3508 } spare_record
__attribute__ ((aligned(512)));
3510 /* spare records have their own family number and do not have any defined raid
3513 static int write_super_imsm_spares(struct intel_super
*super
, int doclose
)
3515 struct imsm_super
*mpb
= super
->anchor
;
3516 struct imsm_super
*spare
= &spare_record
.anchor
;
3520 spare
->mpb_size
= __cpu_to_le32(sizeof(struct imsm_super
)),
3521 spare
->generation_num
= __cpu_to_le32(1UL),
3522 spare
->attributes
= MPB_ATTRIB_CHECKSUM_VERIFY
;
3523 spare
->num_disks
= 1,
3524 spare
->num_raid_devs
= 0,
3525 spare
->cache_size
= mpb
->cache_size
,
3526 spare
->pwr_cycle_count
= __cpu_to_le32(1),
3528 snprintf((char *) spare
->sig
, MAX_SIGNATURE_LENGTH
,
3529 MPB_SIGNATURE MPB_VERSION_RAID0
);
3531 for (d
= super
->disks
; d
; d
= d
->next
) {
3535 spare
->disk
[0] = d
->disk
;
3536 sum
= __gen_imsm_checksum(spare
);
3537 spare
->family_num
= __cpu_to_le32(sum
);
3538 spare
->orig_family_num
= 0;
3539 sum
= __gen_imsm_checksum(spare
);
3540 spare
->check_sum
= __cpu_to_le32(sum
);
3542 if (store_imsm_mpb(d
->fd
, spare
)) {
3543 fprintf(stderr
, "%s: failed for device %d:%d %s\n",
3544 __func__
, d
->major
, d
->minor
, strerror(errno
));
3556 static int write_super_imsm(struct supertype
*st
, int doclose
)
3558 struct intel_super
*super
= st
->sb
;
3559 struct imsm_super
*mpb
= super
->anchor
;
3565 __u32 mpb_size
= sizeof(struct imsm_super
) - sizeof(struct imsm_disk
);
3568 /* 'generation' is incremented everytime the metadata is written */
3569 generation
= __le32_to_cpu(mpb
->generation_num
);
3571 mpb
->generation_num
= __cpu_to_le32(generation
);
3573 /* fix up cases where previous mdadm releases failed to set
3576 if (mpb
->orig_family_num
== 0)
3577 mpb
->orig_family_num
= mpb
->family_num
;
3579 for (d
= super
->disks
; d
; d
= d
->next
) {
3583 mpb
->disk
[d
->index
] = d
->disk
;
3587 for (d
= super
->missing
; d
; d
= d
->next
) {
3588 mpb
->disk
[d
->index
] = d
->disk
;
3591 mpb
->num_disks
= num_disks
;
3592 mpb_size
+= sizeof(struct imsm_disk
) * mpb
->num_disks
;
3594 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
3595 struct imsm_dev
*dev
= __get_imsm_dev(mpb
, i
);
3596 struct imsm_dev
*dev2
= get_imsm_dev(super
, i
);
3598 imsm_copy_dev(dev
, dev2
);
3599 mpb_size
+= sizeof_imsm_dev(dev
, 0);
3602 mpb_size
+= __le32_to_cpu(mpb
->bbm_log_size
);
3603 mpb
->mpb_size
= __cpu_to_le32(mpb_size
);
3605 /* recalculate checksum */
3606 sum
= __gen_imsm_checksum(mpb
);
3607 mpb
->check_sum
= __cpu_to_le32(sum
);
3609 /* write the mpb for disks that compose raid devices */
3610 for (d
= super
->disks
; d
; d
= d
->next
) {
3613 if (store_imsm_mpb(d
->fd
, mpb
))
3614 fprintf(stderr
, "%s: failed for device %d:%d %s\n",
3615 __func__
, d
->major
, d
->minor
, strerror(errno
));
3623 return write_super_imsm_spares(super
, doclose
);
3629 static int create_array(struct supertype
*st
, int dev_idx
)
3632 struct imsm_update_create_array
*u
;
3633 struct intel_super
*super
= st
->sb
;
3634 struct imsm_dev
*dev
= get_imsm_dev(super
, dev_idx
);
3635 struct imsm_map
*map
= get_imsm_map(dev
, 0);
3636 struct disk_info
*inf
;
3637 struct imsm_disk
*disk
;
3640 len
= sizeof(*u
) - sizeof(*dev
) + sizeof_imsm_dev(dev
, 0) +
3641 sizeof(*inf
) * map
->num_members
;
3644 fprintf(stderr
, "%s: failed to allocate update buffer\n",
3649 u
->type
= update_create_array
;
3650 u
->dev_idx
= dev_idx
;
3651 imsm_copy_dev(&u
->dev
, dev
);
3652 inf
= get_disk_info(u
);
3653 for (i
= 0; i
< map
->num_members
; i
++) {
3654 int idx
= get_imsm_disk_idx(dev
, i
);
3656 disk
= get_imsm_disk(super
, idx
);
3657 serialcpy(inf
[i
].serial
, disk
->serial
);
3659 append_metadata_update(st
, u
, len
);
3664 static int mgmt_disk(struct supertype
*st
)
3666 struct intel_super
*super
= st
->sb
;
3668 struct imsm_update_add_remove_disk
*u
;
3670 if (!super
->disk_mgmt_list
)
3676 fprintf(stderr
, "%s: failed to allocate update buffer\n",
3681 u
->type
= update_add_remove_disk
;
3682 append_metadata_update(st
, u
, len
);
3687 static int write_init_super_imsm(struct supertype
*st
)
3689 struct intel_super
*super
= st
->sb
;
3690 int current_vol
= super
->current_vol
;
3692 /* we are done with current_vol reset it to point st at the container */
3693 super
->current_vol
= -1;
3695 if (st
->update_tail
) {
3696 /* queue the recently created array / added disk
3697 * as a metadata update */
3701 /* determine if we are creating a volume or adding a disk */
3702 if (current_vol
< 0) {
3703 /* in the mgmt (add/remove) disk case we are running
3704 * in mdmon context, so don't close fd's
3706 return mgmt_disk(st
);
3708 rv
= create_array(st
, current_vol
);
3710 for (d
= super
->disks
; d
; d
= d
->next
) {
3718 for (d
= super
->disks
; d
; d
= d
->next
)
3719 Kill(d
->devname
, NULL
, 0, 1, 1);
3720 return write_super_imsm(st
, 1);
3725 static int store_super_imsm(struct supertype
*st
, int fd
)
3727 struct intel_super
*super
= st
->sb
;
3728 struct imsm_super
*mpb
= super
? super
->anchor
: NULL
;
3734 return store_imsm_mpb(fd
, mpb
);
3740 static int imsm_bbm_log_size(struct imsm_super
*mpb
)
3742 return __le32_to_cpu(mpb
->bbm_log_size
);
3746 static int validate_geometry_imsm_container(struct supertype
*st
, int level
,
3747 int layout
, int raiddisks
, int chunk
,
3748 unsigned long long size
, char *dev
,
3749 unsigned long long *freesize
,
3753 unsigned long long ldsize
;
3754 const struct imsm_orom
*orom
;
3756 if (level
!= LEVEL_CONTAINER
)
3761 if (check_env("IMSM_NO_PLATFORM"))
3764 orom
= find_imsm_orom();
3765 if (orom
&& raiddisks
> orom
->tds
) {
3767 fprintf(stderr
, Name
": %d exceeds maximum number of"
3768 " platform supported disks: %d\n",
3769 raiddisks
, orom
->tds
);
3773 fd
= open(dev
, O_RDONLY
|O_EXCL
, 0);
3776 fprintf(stderr
, Name
": imsm: Cannot open %s: %s\n",
3777 dev
, strerror(errno
));
3780 if (!get_dev_size(fd
, dev
, &ldsize
)) {
3786 *freesize
= avail_size_imsm(st
, ldsize
>> 9);
3791 static unsigned long long find_size(struct extent
*e
, int *idx
, int num_extents
)
3793 const unsigned long long base_start
= e
[*idx
].start
;
3794 unsigned long long end
= base_start
+ e
[*idx
].size
;
3797 if (base_start
== end
)
3801 for (i
= *idx
; i
< num_extents
; i
++) {
3802 /* extend overlapping extents */
3803 if (e
[i
].start
>= base_start
&&
3804 e
[i
].start
<= end
) {
3807 if (e
[i
].start
+ e
[i
].size
> end
)
3808 end
= e
[i
].start
+ e
[i
].size
;
3809 } else if (e
[i
].start
> end
) {
3815 return end
- base_start
;
3818 static unsigned long long merge_extents(struct intel_super
*super
, int sum_extents
)
3820 /* build a composite disk with all known extents and generate a new
3821 * 'maxsize' given the "all disks in an array must share a common start
3822 * offset" constraint
3824 struct extent
*e
= calloc(sum_extents
, sizeof(*e
));
3828 unsigned long long pos
;
3829 unsigned long long start
= 0;
3830 unsigned long long maxsize
;
3831 unsigned long reserve
;
3836 /* coalesce and sort all extents. also, check to see if we need to
3837 * reserve space between member arrays
3840 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
3843 for (i
= 0; i
< dl
->extent_cnt
; i
++)
3846 qsort(e
, sum_extents
, sizeof(*e
), cmp_extent
);
3851 while (i
< sum_extents
) {
3852 e
[j
].start
= e
[i
].start
;
3853 e
[j
].size
= find_size(e
, &i
, sum_extents
);
3855 if (e
[j
-1].size
== 0)
3864 unsigned long long esize
;
3866 esize
= e
[i
].start
- pos
;
3867 if (esize
>= maxsize
) {
3872 pos
= e
[i
].start
+ e
[i
].size
;
3874 } while (e
[i
-1].size
);
3880 /* FIXME assumes volume at offset 0 is the first volume in a
3883 if (start_extent
> 0)
3884 reserve
= IMSM_RESERVED_SECTORS
; /* gap between raid regions */
3888 if (maxsize
< reserve
)
3891 super
->create_offset
= ~((__u32
) 0);
3892 if (start
+ reserve
> super
->create_offset
)
3893 return 0; /* start overflows create_offset */
3894 super
->create_offset
= start
+ reserve
;
3896 return maxsize
- reserve
;
3899 static int is_raid_level_supported(const struct imsm_orom
*orom
, int level
, int raiddisks
)
3901 if (level
< 0 || level
== 6 || level
== 4)
3904 /* if we have an orom prevent invalid raid levels */
3907 case 0: return imsm_orom_has_raid0(orom
);
3910 return imsm_orom_has_raid1e(orom
);
3911 return imsm_orom_has_raid1(orom
) && raiddisks
== 2;
3912 case 10: return imsm_orom_has_raid10(orom
) && raiddisks
== 4;
3913 case 5: return imsm_orom_has_raid5(orom
) && raiddisks
> 2;
3916 return 1; /* not on an Intel RAID platform so anything goes */
3921 #define pr_vrb(fmt, arg...) (void) (verbose && fprintf(stderr, Name fmt, ##arg))
3923 validate_geometry_imsm_orom(struct intel_super
*super
, int level
, int layout
,
3924 int raiddisks
, int chunk
, int verbose
)
3926 if (!is_raid_level_supported(super
->orom
, level
, raiddisks
)) {
3927 pr_vrb(": platform does not support raid%d with %d disk%s\n",
3928 level
, raiddisks
, raiddisks
> 1 ? "s" : "");
3931 if (super
->orom
&& level
!= 1 &&
3932 !imsm_orom_has_chunk(super
->orom
, chunk
)) {
3933 pr_vrb(": platform does not support a chunk size of: %d\n", chunk
);
3936 if (layout
!= imsm_level_to_layout(level
)) {
3938 pr_vrb(": imsm raid 5 only supports the left-asymmetric layout\n");
3939 else if (level
== 10)
3940 pr_vrb(": imsm raid 10 only supports the n2 layout\n");
3942 pr_vrb(": imsm unknown layout %#x for this raid level %d\n",
3950 /* validate_geometry_imsm_volume - lifted from validate_geometry_ddf_bvd
3951 * FIX ME add ahci details
3953 static int validate_geometry_imsm_volume(struct supertype
*st
, int level
,
3954 int layout
, int raiddisks
, int chunk
,
3955 unsigned long long size
, char *dev
,
3956 unsigned long long *freesize
,
3960 struct intel_super
*super
= st
->sb
;
3961 struct imsm_super
*mpb
= super
->anchor
;
3963 unsigned long long pos
= 0;
3964 unsigned long long maxsize
;
3968 /* We must have the container info already read in. */
3972 if (!validate_geometry_imsm_orom(super
, level
, layout
, raiddisks
, chunk
, verbose
))
3976 /* General test: make sure there is space for
3977 * 'raiddisks' device extents of size 'size' at a given
3980 unsigned long long minsize
= size
;
3981 unsigned long long start_offset
= MaxSector
;
3984 minsize
= MPB_SECTOR_CNT
+ IMSM_RESERVED_SECTORS
;
3985 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
3990 e
= get_extents(super
, dl
);
3993 unsigned long long esize
;
3994 esize
= e
[i
].start
- pos
;
3995 if (esize
>= minsize
)
3997 if (found
&& start_offset
== MaxSector
) {
4000 } else if (found
&& pos
!= start_offset
) {
4004 pos
= e
[i
].start
+ e
[i
].size
;
4006 } while (e
[i
-1].size
);
4011 if (dcnt
< raiddisks
) {
4013 fprintf(stderr
, Name
": imsm: Not enough "
4014 "devices with space for this array "
4022 /* This device must be a member of the set */
4023 if (stat(dev
, &stb
) < 0)
4025 if ((S_IFMT
& stb
.st_mode
) != S_IFBLK
)
4027 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
4028 if (dl
->major
== (int)major(stb
.st_rdev
) &&
4029 dl
->minor
== (int)minor(stb
.st_rdev
))
4034 fprintf(stderr
, Name
": %s is not in the "
4035 "same imsm set\n", dev
);
4037 } else if (super
->orom
&& dl
->index
< 0 && mpb
->num_raid_devs
) {
4038 /* If a volume is present then the current creation attempt
4039 * cannot incorporate new spares because the orom may not
4040 * understand this configuration (all member disks must be
4041 * members of each array in the container).
4043 fprintf(stderr
, Name
": %s is a spare and a volume"
4044 " is already defined for this container\n", dev
);
4045 fprintf(stderr
, Name
": The option-rom requires all member"
4046 " disks to be a member of all volumes\n");
4050 /* retrieve the largest free space block */
4051 e
= get_extents(super
, dl
);
4056 unsigned long long esize
;
4058 esize
= e
[i
].start
- pos
;
4059 if (esize
>= maxsize
)
4061 pos
= e
[i
].start
+ e
[i
].size
;
4063 } while (e
[i
-1].size
);
4068 fprintf(stderr
, Name
": unable to determine free space for: %s\n",
4072 if (maxsize
< size
) {
4074 fprintf(stderr
, Name
": %s not enough space (%llu < %llu)\n",
4075 dev
, maxsize
, size
);
4079 /* count total number of extents for merge */
4081 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
4083 i
+= dl
->extent_cnt
;
4085 maxsize
= merge_extents(super
, i
);
4086 if (maxsize
< size
|| maxsize
== 0) {
4088 fprintf(stderr
, Name
": not enough space after merge (%llu < %llu)\n",
4093 *freesize
= maxsize
;
4098 static int reserve_space(struct supertype
*st
, int raiddisks
,
4099 unsigned long long size
, int chunk
,
4100 unsigned long long *freesize
)
4102 struct intel_super
*super
= st
->sb
;
4103 struct imsm_super
*mpb
= super
->anchor
;
4108 unsigned long long maxsize
;
4109 unsigned long long minsize
;
4113 /* find the largest common start free region of the possible disks */
4117 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
4123 /* don't activate new spares if we are orom constrained
4124 * and there is already a volume active in the container
4126 if (super
->orom
&& dl
->index
< 0 && mpb
->num_raid_devs
)
4129 e
= get_extents(super
, dl
);
4132 for (i
= 1; e
[i
-1].size
; i
++)
4140 maxsize
= merge_extents(super
, extent_cnt
);
4145 if (cnt
< raiddisks
||
4146 (super
->orom
&& used
&& used
!= raiddisks
) ||
4147 maxsize
< minsize
||
4149 fprintf(stderr
, Name
": not enough devices with space to create array.\n");
4150 return 0; /* No enough free spaces large enough */
4162 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
4164 dl
->raiddisk
= cnt
++;
4171 static int validate_geometry_imsm(struct supertype
*st
, int level
, int layout
,
4172 int raiddisks
, int chunk
, unsigned long long size
,
4173 char *dev
, unsigned long long *freesize
,
4180 /* if given unused devices create a container
4181 * if given given devices in a container create a member volume
4183 if (level
== LEVEL_CONTAINER
) {
4184 /* Must be a fresh device to add to a container */
4185 return validate_geometry_imsm_container(st
, level
, layout
,
4186 raiddisks
, chunk
, size
,
4192 if (st
->sb
&& freesize
) {
4193 /* we are being asked to automatically layout a
4194 * new volume based on the current contents of
4195 * the container. If the the parameters can be
4196 * satisfied reserve_space will record the disks,
4197 * start offset, and size of the volume to be
4198 * created. add_to_super and getinfo_super
4199 * detect when autolayout is in progress.
4201 if (!validate_geometry_imsm_orom(st
->sb
, level
, layout
,
4205 return reserve_space(st
, raiddisks
, size
, chunk
, freesize
);
4210 /* creating in a given container */
4211 return validate_geometry_imsm_volume(st
, level
, layout
,
4212 raiddisks
, chunk
, size
,
4213 dev
, freesize
, verbose
);
4216 /* This device needs to be a device in an 'imsm' container */
4217 fd
= open(dev
, O_RDONLY
|O_EXCL
, 0);
4221 Name
": Cannot create this array on device %s\n",
4226 if (errno
!= EBUSY
|| (fd
= open(dev
, O_RDONLY
, 0)) < 0) {
4228 fprintf(stderr
, Name
": Cannot open %s: %s\n",
4229 dev
, strerror(errno
));
4232 /* Well, it is in use by someone, maybe an 'imsm' container. */
4233 cfd
= open_container(fd
);
4237 fprintf(stderr
, Name
": Cannot use %s: It is busy\n",
4241 sra
= sysfs_read(cfd
, 0, GET_VERSION
);
4242 if (sra
&& sra
->array
.major_version
== -1 &&
4243 strcmp(sra
->text_version
, "imsm") == 0)
4247 /* This is a member of a imsm container. Load the container
4248 * and try to create a volume
4250 struct intel_super
*super
;
4252 if (load_super_imsm_all(st
, cfd
, (void **) &super
, NULL
) == 0) {
4254 st
->container_dev
= fd2devnum(cfd
);
4256 return validate_geometry_imsm_volume(st
, level
, layout
,
4264 fprintf(stderr
, Name
": failed container membership check\n");
4270 static void default_geometry_imsm(struct supertype
*st
, int *level
, int *layout
, int *chunk
)
4272 struct intel_super
*super
= st
->sb
;
4274 if (level
&& *level
== UnSet
)
4275 *level
= LEVEL_CONTAINER
;
4277 if (level
&& layout
&& *layout
== UnSet
)
4278 *layout
= imsm_level_to_layout(*level
);
4280 if (chunk
&& (*chunk
== UnSet
|| *chunk
== 0) &&
4281 super
&& super
->orom
)
4282 *chunk
= imsm_orom_default_chunk(super
->orom
);
4285 static void handle_missing(struct intel_super
*super
, struct imsm_dev
*dev
);
4287 static int kill_subarray_imsm(struct supertype
*st
)
4289 /* remove the subarray currently referenced by ->current_vol */
4291 struct intel_dev
**dp
;
4292 struct intel_super
*super
= st
->sb
;
4293 __u8 current_vol
= super
->current_vol
;
4294 struct imsm_super
*mpb
= super
->anchor
;
4296 if (super
->current_vol
< 0)
4298 super
->current_vol
= -1; /* invalidate subarray cursor */
4300 /* block deletions that would change the uuid of active subarrays
4302 * FIXME when immutable ids are available, but note that we'll
4303 * also need to fixup the invalidated/active subarray indexes in
4306 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
4309 if (i
< current_vol
)
4311 sprintf(subarray
, "%u", i
);
4312 if (is_subarray_active(subarray
, st
->devname
)) {
4314 Name
": deleting subarray-%d would change the UUID of active subarray-%d, aborting\n",
4321 if (st
->update_tail
) {
4322 struct imsm_update_kill_array
*u
= malloc(sizeof(*u
));
4326 u
->type
= update_kill_array
;
4327 u
->dev_idx
= current_vol
;
4328 append_metadata_update(st
, u
, sizeof(*u
));
4333 for (dp
= &super
->devlist
; *dp
;)
4334 if ((*dp
)->index
== current_vol
) {
4337 handle_missing(super
, (*dp
)->dev
);
4338 if ((*dp
)->index
> current_vol
)
4343 /* no more raid devices, all active components are now spares,
4344 * but of course failed are still failed
4346 if (--mpb
->num_raid_devs
== 0) {
4349 for (d
= super
->disks
; d
; d
= d
->next
)
4350 if (d
->index
> -2) {
4352 d
->disk
.status
= SPARE_DISK
;
4356 super
->updates_pending
++;
4361 static int update_subarray_imsm(struct supertype
*st
, char *subarray
,
4362 char *update
, struct mddev_ident
*ident
)
4364 /* update the subarray currently referenced by ->current_vol */
4365 struct intel_super
*super
= st
->sb
;
4366 struct imsm_super
*mpb
= super
->anchor
;
4368 if (strcmp(update
, "name") == 0) {
4369 char *name
= ident
->name
;
4373 if (is_subarray_active(subarray
, st
->devname
)) {
4375 Name
": Unable to update name of active subarray\n");
4379 if (!check_name(super
, name
, 0))
4382 vol
= strtoul(subarray
, &ep
, 10);
4383 if (*ep
!= '\0' || vol
>= super
->anchor
->num_raid_devs
)
4386 if (st
->update_tail
) {
4387 struct imsm_update_rename_array
*u
= malloc(sizeof(*u
));
4391 u
->type
= update_rename_array
;
4393 snprintf((char *) u
->name
, MAX_RAID_SERIAL_LEN
, "%s", name
);
4394 append_metadata_update(st
, u
, sizeof(*u
));
4396 struct imsm_dev
*dev
;
4399 dev
= get_imsm_dev(super
, vol
);
4400 snprintf((char *) dev
->volume
, MAX_RAID_SERIAL_LEN
, "%s", name
);
4401 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
4402 dev
= get_imsm_dev(super
, i
);
4403 handle_missing(super
, dev
);
4405 super
->updates_pending
++;
4412 #endif /* MDASSEMBLE */
4414 static int is_gen_migration(struct imsm_dev
*dev
)
4416 if (!dev
->vol
.migr_state
)
4419 if (migr_type(dev
) == MIGR_GEN_MIGR
)
4425 static int is_rebuilding(struct imsm_dev
*dev
)
4427 struct imsm_map
*migr_map
;
4429 if (!dev
->vol
.migr_state
)
4432 if (migr_type(dev
) != MIGR_REBUILD
)
4435 migr_map
= get_imsm_map(dev
, 1);
4437 if (migr_map
->map_state
== IMSM_T_STATE_DEGRADED
)
4443 static void update_recovery_start(struct imsm_dev
*dev
, struct mdinfo
*array
)
4445 struct mdinfo
*rebuild
= NULL
;
4449 if (!is_rebuilding(dev
))
4452 /* Find the rebuild target, but punt on the dual rebuild case */
4453 for (d
= array
->devs
; d
; d
= d
->next
)
4454 if (d
->recovery_start
== 0) {
4461 /* (?) none of the disks are marked with
4462 * IMSM_ORD_REBUILD, so assume they are missing and the
4463 * disk_ord_tbl was not correctly updated
4465 dprintf("%s: failed to locate out-of-sync disk\n", __func__
);
4469 units
= __le32_to_cpu(dev
->vol
.curr_migr_unit
);
4470 rebuild
->recovery_start
= units
* blocks_per_migr_unit(dev
);
4474 static struct mdinfo
*container_content_imsm(struct supertype
*st
, char *subarray
)
4476 /* Given a container loaded by load_super_imsm_all,
4477 * extract information about all the arrays into
4479 * If 'subarray' is given, just extract info about that array.
4481 * For each imsm_dev create an mdinfo, fill it in,
4482 * then look for matching devices in super->disks
4483 * and create appropriate device mdinfo.
4485 struct intel_super
*super
= st
->sb
;
4486 struct imsm_super
*mpb
= super
->anchor
;
4487 struct mdinfo
*rest
= NULL
;
4491 /* check for bad blocks */
4492 if (imsm_bbm_log_size(super
->anchor
))
4495 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
4496 struct imsm_dev
*dev
;
4497 struct imsm_map
*map
;
4498 struct mdinfo
*this;
4503 (i
!= strtoul(subarray
, &ep
, 10) || *ep
!= '\0'))
4506 dev
= get_imsm_dev(super
, i
);
4507 map
= get_imsm_map(dev
, 0);
4509 /* do not publish arrays that are in the middle of an
4510 * unsupported migration
4512 if (dev
->vol
.migr_state
&&
4513 (migr_type(dev
) == MIGR_STATE_CHANGE
)) {
4514 fprintf(stderr
, Name
": cannot assemble volume '%.16s':"
4515 " unsupported migration in progress\n",
4520 this = malloc(sizeof(*this));
4522 fprintf(stderr
, Name
": failed to allocate %zu bytes\n",
4526 memset(this, 0, sizeof(*this));
4529 super
->current_vol
= i
;
4530 getinfo_super_imsm_volume(st
, this, NULL
);
4531 for (slot
= 0 ; slot
< map
->num_members
; slot
++) {
4532 unsigned long long recovery_start
;
4533 struct mdinfo
*info_d
;
4540 idx
= get_imsm_disk_idx(dev
, slot
);
4541 ord
= get_imsm_ord_tbl_ent(dev
, slot
);
4542 for (d
= super
->disks
; d
; d
= d
->next
)
4543 if (d
->index
== idx
)
4546 recovery_start
= MaxSector
;
4549 if (d
&& is_failed(&d
->disk
))
4551 if (ord
& IMSM_ORD_REBUILD
)
4555 * if we skip some disks the array will be assmebled degraded;
4556 * reset resync start to avoid a dirty-degraded
4557 * situation when performing the intial sync
4559 * FIXME handle dirty degraded
4561 if ((skip
|| recovery_start
== 0) && !dev
->vol
.dirty
)
4562 this->resync_start
= MaxSector
;
4566 info_d
= calloc(1, sizeof(*info_d
));
4568 fprintf(stderr
, Name
": failed to allocate disk"
4569 " for volume %.16s\n", dev
->volume
);
4570 info_d
= this->devs
;
4572 struct mdinfo
*d
= info_d
->next
;
4581 info_d
->next
= this->devs
;
4582 this->devs
= info_d
;
4584 info_d
->disk
.number
= d
->index
;
4585 info_d
->disk
.major
= d
->major
;
4586 info_d
->disk
.minor
= d
->minor
;
4587 info_d
->disk
.raid_disk
= slot
;
4588 info_d
->recovery_start
= recovery_start
;
4590 if (info_d
->recovery_start
== MaxSector
)
4591 this->array
.working_disks
++;
4593 info_d
->events
= __le32_to_cpu(mpb
->generation_num
);
4594 info_d
->data_offset
= __le32_to_cpu(map
->pba_of_lba0
);
4595 info_d
->component_size
= __le32_to_cpu(map
->blocks_per_member
);
4597 /* now that the disk list is up-to-date fixup recovery_start */
4598 update_recovery_start(dev
, this);
4602 /* if array has bad blocks, set suitable bit in array status */
4604 rest
->array
.state
|= (1<<MD_SB_BBM_ERRORS
);
4610 static __u8
imsm_check_degraded(struct intel_super
*super
, struct imsm_dev
*dev
, int failed
)
4612 struct imsm_map
*map
= get_imsm_map(dev
, 0);
4615 return map
->map_state
== IMSM_T_STATE_UNINITIALIZED
?
4616 IMSM_T_STATE_UNINITIALIZED
: IMSM_T_STATE_NORMAL
;
4618 switch (get_imsm_raid_level(map
)) {
4620 return IMSM_T_STATE_FAILED
;
4623 if (failed
< map
->num_members
)
4624 return IMSM_T_STATE_DEGRADED
;
4626 return IMSM_T_STATE_FAILED
;
4631 * check to see if any mirrors have failed, otherwise we
4632 * are degraded. Even numbered slots are mirrored on
4636 /* gcc -Os complains that this is unused */
4637 int insync
= insync
;
4639 for (i
= 0; i
< map
->num_members
; i
++) {
4640 __u32 ord
= get_imsm_ord_tbl_ent(dev
, i
);
4641 int idx
= ord_to_idx(ord
);
4642 struct imsm_disk
*disk
;
4644 /* reset the potential in-sync count on even-numbered
4645 * slots. num_copies is always 2 for imsm raid10
4650 disk
= get_imsm_disk(super
, idx
);
4651 if (!disk
|| is_failed(disk
) || ord
& IMSM_ORD_REBUILD
)
4654 /* no in-sync disks left in this mirror the
4658 return IMSM_T_STATE_FAILED
;
4661 return IMSM_T_STATE_DEGRADED
;
4665 return IMSM_T_STATE_DEGRADED
;
4667 return IMSM_T_STATE_FAILED
;
4673 return map
->map_state
;
4676 static int imsm_count_failed(struct intel_super
*super
, struct imsm_dev
*dev
)
4680 struct imsm_disk
*disk
;
4681 struct imsm_map
*map
= get_imsm_map(dev
, 0);
4682 struct imsm_map
*prev
= get_imsm_map(dev
, dev
->vol
.migr_state
);
4686 /* at the beginning of migration we set IMSM_ORD_REBUILD on
4687 * disks that are being rebuilt. New failures are recorded to
4688 * map[0]. So we look through all the disks we started with and
4689 * see if any failures are still present, or if any new ones
4692 * FIXME add support for online capacity expansion and
4693 * raid-level-migration
4695 for (i
= 0; i
< prev
->num_members
; i
++) {
4696 ord
= __le32_to_cpu(prev
->disk_ord_tbl
[i
]);
4697 ord
|= __le32_to_cpu(map
->disk_ord_tbl
[i
]);
4698 idx
= ord_to_idx(ord
);
4700 disk
= get_imsm_disk(super
, idx
);
4701 if (!disk
|| is_failed(disk
) || ord
& IMSM_ORD_REBUILD
)
4709 static int imsm_open_new(struct supertype
*c
, struct active_array
*a
,
4712 struct intel_super
*super
= c
->sb
;
4713 struct imsm_super
*mpb
= super
->anchor
;
4715 if (atoi(inst
) >= mpb
->num_raid_devs
) {
4716 fprintf(stderr
, "%s: subarry index %d, out of range\n",
4717 __func__
, atoi(inst
));
4721 dprintf("imsm: open_new %s\n", inst
);
4722 a
->info
.container_member
= atoi(inst
);
4726 static int is_resyncing(struct imsm_dev
*dev
)
4728 struct imsm_map
*migr_map
;
4730 if (!dev
->vol
.migr_state
)
4733 if (migr_type(dev
) == MIGR_INIT
||
4734 migr_type(dev
) == MIGR_REPAIR
)
4737 if (migr_type(dev
) == MIGR_GEN_MIGR
)
4740 migr_map
= get_imsm_map(dev
, 1);
4742 if ((migr_map
->map_state
== IMSM_T_STATE_NORMAL
) &&
4743 (dev
->vol
.migr_type
!= MIGR_GEN_MIGR
))
4749 /* return true if we recorded new information */
4750 static int mark_failure(struct imsm_dev
*dev
, struct imsm_disk
*disk
, int idx
)
4754 struct imsm_map
*map
;
4756 /* new failures are always set in map[0] */
4757 map
= get_imsm_map(dev
, 0);
4759 slot
= get_imsm_disk_slot(map
, idx
);
4763 ord
= __le32_to_cpu(map
->disk_ord_tbl
[slot
]);
4764 if (is_failed(disk
) && (ord
& IMSM_ORD_REBUILD
))
4767 disk
->status
|= FAILED_DISK
;
4768 disk
->status
&= ~CONFIGURED_DISK
;
4769 set_imsm_ord_tbl_ent(map
, slot
, idx
| IMSM_ORD_REBUILD
);
4770 if (map
->failed_disk_num
== 0xff)
4771 map
->failed_disk_num
= slot
;
4775 static void mark_missing(struct imsm_dev
*dev
, struct imsm_disk
*disk
, int idx
)
4777 mark_failure(dev
, disk
, idx
);
4779 if (disk
->scsi_id
== __cpu_to_le32(~(__u32
)0))
4782 disk
->scsi_id
= __cpu_to_le32(~(__u32
)0);
4783 memmove(&disk
->serial
[0], &disk
->serial
[1], MAX_RAID_SERIAL_LEN
- 1);
4786 static void handle_missing(struct intel_super
*super
, struct imsm_dev
*dev
)
4792 if (!super
->missing
)
4794 failed
= imsm_count_failed(super
, dev
);
4795 map_state
= imsm_check_degraded(super
, dev
, failed
);
4797 dprintf("imsm: mark missing\n");
4798 end_migration(dev
, map_state
);
4799 for (dl
= super
->missing
; dl
; dl
= dl
->next
)
4800 mark_missing(dev
, &dl
->disk
, dl
->index
);
4801 super
->updates_pending
++;
4804 static void imsm_set_disk(struct active_array
*a
, int n
, int state
);
4806 /* Handle dirty -> clean transititions, resync and reshape. Degraded and rebuild
4807 * states are handled in imsm_set_disk() with one exception, when a
4808 * resync is stopped due to a new failure this routine will set the
4809 * 'degraded' state for the array.
4811 static int imsm_set_array_state(struct active_array
*a
, int consistent
)
4813 int inst
= a
->info
.container_member
;
4814 struct intel_super
*super
= a
->container
->sb
;
4815 struct imsm_dev
*dev
= get_imsm_dev(super
, inst
);
4816 struct imsm_map
*map
= get_imsm_map(dev
, 0);
4817 int failed
= imsm_count_failed(super
, dev
);
4818 __u8 map_state
= imsm_check_degraded(super
, dev
, failed
);
4819 __u32 blocks_per_unit
;
4821 if (dev
->vol
.migr_state
&&
4822 dev
->vol
.migr_type
== MIGR_GEN_MIGR
) {
4823 /* array state change is blocked due to reshape action
4825 * - abort the reshape (if last_checkpoint is 0 and action!= reshape)
4826 * - finish the reshape (if last_checkpoint is big and action != reshape)
4827 * - update curr_migr_unit
4829 if (a
->curr_action
== reshape
) {
4830 /* still reshaping, maybe update curr_migr_unit */
4831 long long blocks_per_unit
= blocks_per_migr_unit(dev
);
4832 long long unit
= a
->last_checkpoint
;
4833 unit
/= blocks_per_unit
;
4834 if (unit
> __le32_to_cpu(dev
->vol
.curr_migr_unit
)) {
4835 dev
->vol
.curr_migr_unit
= __cpu_to_le32(unit
);
4836 super
->updates_pending
++;
4839 if (a
->last_checkpoint
== 0 && a
->prev_action
== reshape
) {
4840 /* for some reason we aborted the reshape.
4843 struct imsm_map
*map2
= get_imsm_map(dev
, 1);
4844 dev
->vol
.migr_state
= 0;
4845 dev
->vol
.migr_type
= 0;
4846 dev
->vol
.curr_migr_unit
= 0;
4847 memcpy(map
, map2
, sizeof_imsm_map(map2
));
4848 super
->updates_pending
++;
4850 if (a
->last_checkpoint
>= a
->info
.component_size
) {
4851 unsigned long long array_blocks
;
4853 /* it seems the reshape is all done */
4854 dev
->vol
.migr_state
= 0;
4855 dev
->vol
.migr_type
= 0;
4856 dev
->vol
.curr_migr_unit
= 0;
4858 used_disks
= imsm_num_data_members(dev
);
4859 array_blocks
= map
->blocks_per_member
* used_disks
;
4860 /* round array size down to closest MB */
4861 array_blocks
= (array_blocks
>> SECT_PER_MB_SHIFT
)
4862 << SECT_PER_MB_SHIFT
;
4863 dev
->size_low
= __cpu_to_le32((__u32
) array_blocks
);
4864 dev
->size_high
= __cpu_to_le32((__u32
) (array_blocks
>> 32));
4865 a
->info
.custom_array_size
= array_blocks
;
4866 a
->check_reshape
= 1; /* encourage manager to update
4869 super
->updates_pending
++;
4875 /* before we activate this array handle any missing disks */
4876 if (consistent
== 2)
4877 handle_missing(super
, dev
);
4879 if (consistent
== 2 &&
4880 (!is_resync_complete(&a
->info
) ||
4881 map_state
!= IMSM_T_STATE_NORMAL
||
4882 dev
->vol
.migr_state
))
4885 if (is_resync_complete(&a
->info
)) {
4886 /* complete intialization / resync,
4887 * recovery and interrupted recovery is completed in
4890 if (is_resyncing(dev
)) {
4891 dprintf("imsm: mark resync done\n");
4892 end_migration(dev
, map_state
);
4893 super
->updates_pending
++;
4894 a
->last_checkpoint
= 0;
4896 } else if (!is_resyncing(dev
) && !failed
) {
4897 /* mark the start of the init process if nothing is failed */
4898 dprintf("imsm: mark resync start\n");
4899 if (map
->map_state
== IMSM_T_STATE_UNINITIALIZED
)
4900 migrate(dev
, IMSM_T_STATE_NORMAL
, MIGR_INIT
);
4902 migrate(dev
, IMSM_T_STATE_NORMAL
, MIGR_REPAIR
);
4903 super
->updates_pending
++;
4906 /* check if we can update curr_migr_unit from resync_start, recovery_start */
4907 blocks_per_unit
= blocks_per_migr_unit(dev
);
4908 if (blocks_per_unit
) {
4912 units
= a
->last_checkpoint
/ blocks_per_unit
;
4915 /* check that we did not overflow 32-bits, and that
4916 * curr_migr_unit needs updating
4918 if (units32
== units
&&
4919 __le32_to_cpu(dev
->vol
.curr_migr_unit
) != units32
) {
4920 dprintf("imsm: mark checkpoint (%u)\n", units32
);
4921 dev
->vol
.curr_migr_unit
= __cpu_to_le32(units32
);
4922 super
->updates_pending
++;
4926 /* mark dirty / clean */
4927 if (dev
->vol
.dirty
!= !consistent
) {
4928 dprintf("imsm: mark '%s'\n", consistent
? "clean" : "dirty");
4933 super
->updates_pending
++;
4936 /* finalize online capacity expansion/reshape */
4937 if ((a
->curr_action
!= reshape
) &&
4938 (a
->prev_action
== reshape
)) {
4941 for (mdi
= a
->info
.devs
; mdi
; mdi
= mdi
->next
)
4942 imsm_set_disk(a
, mdi
->disk
.raid_disk
, mdi
->curr_state
);
4948 static void imsm_set_disk(struct active_array
*a
, int n
, int state
)
4950 int inst
= a
->info
.container_member
;
4951 struct intel_super
*super
= a
->container
->sb
;
4952 struct imsm_dev
*dev
= get_imsm_dev(super
, inst
);
4953 struct imsm_map
*map
= get_imsm_map(dev
, 0);
4954 struct imsm_disk
*disk
;
4959 if (n
> map
->num_members
)
4960 fprintf(stderr
, "imsm: set_disk %d out of range 0..%d\n",
4961 n
, map
->num_members
- 1);
4966 dprintf("imsm: set_disk %d:%x\n", n
, state
);
4968 ord
= get_imsm_ord_tbl_ent(dev
, n
);
4969 disk
= get_imsm_disk(super
, ord_to_idx(ord
));
4971 /* check for new failures */
4972 if (state
& DS_FAULTY
) {
4973 if (mark_failure(dev
, disk
, ord_to_idx(ord
)))
4974 super
->updates_pending
++;
4977 /* check if in_sync */
4978 if (state
& DS_INSYNC
&& ord
& IMSM_ORD_REBUILD
&& is_rebuilding(dev
)) {
4979 struct imsm_map
*migr_map
= get_imsm_map(dev
, 1);
4981 set_imsm_ord_tbl_ent(migr_map
, n
, ord_to_idx(ord
));
4982 super
->updates_pending
++;
4985 failed
= imsm_count_failed(super
, dev
);
4986 map_state
= imsm_check_degraded(super
, dev
, failed
);
4988 /* check if recovery complete, newly degraded, or failed */
4989 if (map_state
== IMSM_T_STATE_NORMAL
&& is_rebuilding(dev
)) {
4990 end_migration(dev
, map_state
);
4991 map
= get_imsm_map(dev
, 0);
4992 map
->failed_disk_num
= ~0;
4993 super
->updates_pending
++;
4994 a
->last_checkpoint
= 0;
4995 } else if (map_state
== IMSM_T_STATE_DEGRADED
&&
4996 map
->map_state
!= map_state
&&
4997 !dev
->vol
.migr_state
) {
4998 dprintf("imsm: mark degraded\n");
4999 map
->map_state
= map_state
;
5000 super
->updates_pending
++;
5001 a
->last_checkpoint
= 0;
5002 } else if (map_state
== IMSM_T_STATE_FAILED
&&
5003 map
->map_state
!= map_state
) {
5004 dprintf("imsm: mark failed\n");
5005 end_migration(dev
, map_state
);
5006 super
->updates_pending
++;
5007 a
->last_checkpoint
= 0;
5008 } else if (is_gen_migration(dev
)) {
5009 dprintf("imsm: Detected General Migration in state: ");
5010 if (map_state
== IMSM_T_STATE_NORMAL
) {
5011 end_migration(dev
, map_state
);
5012 map
= get_imsm_map(dev
, 0);
5013 map
->failed_disk_num
= ~0;
5014 dprintf("normal\n");
5016 if (map_state
== IMSM_T_STATE_DEGRADED
) {
5017 printf("degraded\n");
5018 end_migration(dev
, map_state
);
5020 dprintf("failed\n");
5022 map
->map_state
= map_state
;
5024 super
->updates_pending
++;
5028 static int store_imsm_mpb(int fd
, struct imsm_super
*mpb
)
5031 __u32 mpb_size
= __le32_to_cpu(mpb
->mpb_size
);
5032 unsigned long long dsize
;
5033 unsigned long long sectors
;
5035 get_dev_size(fd
, NULL
, &dsize
);
5037 if (mpb_size
> 512) {
5038 /* -1 to account for anchor */
5039 sectors
= mpb_sectors(mpb
) - 1;
5041 /* write the extended mpb to the sectors preceeding the anchor */
5042 if (lseek64(fd
, dsize
- (512 * (2 + sectors
)), SEEK_SET
) < 0)
5045 if ((unsigned long long)write(fd
, buf
+ 512, 512 * sectors
)
5050 /* first block is stored on second to last sector of the disk */
5051 if (lseek64(fd
, dsize
- (512 * 2), SEEK_SET
) < 0)
5054 if (write(fd
, buf
, 512) != 512)
5060 static void imsm_sync_metadata(struct supertype
*container
)
5062 struct intel_super
*super
= container
->sb
;
5064 dprintf("sync metadata: %d\n", super
->updates_pending
);
5065 if (!super
->updates_pending
)
5068 write_super_imsm(container
, 0);
5070 super
->updates_pending
= 0;
5073 static struct dl
*imsm_readd(struct intel_super
*super
, int idx
, struct active_array
*a
)
5075 struct imsm_dev
*dev
= get_imsm_dev(super
, a
->info
.container_member
);
5076 int i
= get_imsm_disk_idx(dev
, idx
);
5079 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
5083 if (dl
&& is_failed(&dl
->disk
))
5087 dprintf("%s: found %x:%x\n", __func__
, dl
->major
, dl
->minor
);
5092 static struct dl
*imsm_add_spare(struct intel_super
*super
, int slot
,
5093 struct active_array
*a
, int activate_new
,
5094 struct mdinfo
*additional_test_list
)
5096 struct imsm_dev
*dev
= get_imsm_dev(super
, a
->info
.container_member
);
5097 int idx
= get_imsm_disk_idx(dev
, slot
);
5098 struct imsm_super
*mpb
= super
->anchor
;
5099 struct imsm_map
*map
;
5100 unsigned long long pos
;
5105 __u32 array_start
= 0;
5106 __u32 array_end
= 0;
5108 struct mdinfo
*test_list
;
5110 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
5111 /* If in this array, skip */
5112 for (d
= a
->info
.devs
; d
; d
= d
->next
)
5113 if (d
->state_fd
>= 0 &&
5114 d
->disk
.major
== dl
->major
&&
5115 d
->disk
.minor
== dl
->minor
) {
5116 dprintf("%x:%x already in array\n",
5117 dl
->major
, dl
->minor
);
5122 test_list
= additional_test_list
;
5124 if (test_list
->disk
.major
== dl
->major
&&
5125 test_list
->disk
.minor
== dl
->minor
) {
5126 dprintf("%x:%x already in additional test list\n",
5127 dl
->major
, dl
->minor
);
5130 test_list
= test_list
->next
;
5135 /* skip in use or failed drives */
5136 if (is_failed(&dl
->disk
) || idx
== dl
->index
||
5138 dprintf("%x:%x status (failed: %d index: %d)\n",
5139 dl
->major
, dl
->minor
, is_failed(&dl
->disk
), idx
);
5143 /* skip pure spares when we are looking for partially
5144 * assimilated drives
5146 if (dl
->index
== -1 && !activate_new
)
5149 /* Does this unused device have the requisite free space?
5150 * It needs to be able to cover all member volumes
5152 ex
= get_extents(super
, dl
);
5154 dprintf("cannot get extents\n");
5157 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
5158 dev
= get_imsm_dev(super
, i
);
5159 map
= get_imsm_map(dev
, 0);
5161 /* check if this disk is already a member of
5164 if (get_imsm_disk_slot(map
, dl
->index
) >= 0)
5170 array_start
= __le32_to_cpu(map
->pba_of_lba0
);
5171 array_end
= array_start
+
5172 __le32_to_cpu(map
->blocks_per_member
) - 1;
5175 /* check that we can start at pba_of_lba0 with
5176 * blocks_per_member of space
5178 if (array_start
>= pos
&& array_end
< ex
[j
].start
) {
5182 pos
= ex
[j
].start
+ ex
[j
].size
;
5184 } while (ex
[j
-1].size
);
5191 if (i
< mpb
->num_raid_devs
) {
5192 dprintf("%x:%x does not have %u to %u available\n",
5193 dl
->major
, dl
->minor
, array_start
, array_end
);
5204 static int imsm_rebuild_allowed(struct supertype
*cont
, int dev_idx
, int failed
)
5206 struct imsm_dev
*dev2
;
5207 struct imsm_map
*map
;
5213 dev2
= get_imsm_dev(cont
->sb
, dev_idx
);
5215 state
= imsm_check_degraded(cont
->sb
, dev2
, failed
);
5216 if (state
== IMSM_T_STATE_FAILED
) {
5217 map
= get_imsm_map(dev2
, 0);
5220 for (slot
= 0; slot
< map
->num_members
; slot
++) {
5222 * Check if failed disks are deleted from intel
5223 * disk list or are marked to be deleted
5225 idx
= get_imsm_disk_idx(dev2
, slot
);
5226 idisk
= get_imsm_dl_disk(cont
->sb
, idx
);
5228 * Do not rebuild the array if failed disks
5229 * from failed sub-array are not removed from
5233 is_failed(&idisk
->disk
) &&
5234 (idisk
->action
!= DISK_REMOVE
))
5242 static struct mdinfo
*imsm_activate_spare(struct active_array
*a
,
5243 struct metadata_update
**updates
)
5246 * Find a device with unused free space and use it to replace a
5247 * failed/vacant region in an array. We replace failed regions one a
5248 * array at a time. The result is that a new spare disk will be added
5249 * to the first failed array and after the monitor has finished
5250 * propagating failures the remainder will be consumed.
5252 * FIXME add a capability for mdmon to request spares from another
5256 struct intel_super
*super
= a
->container
->sb
;
5257 int inst
= a
->info
.container_member
;
5258 struct imsm_dev
*dev
= get_imsm_dev(super
, inst
);
5259 struct imsm_map
*map
= get_imsm_map(dev
, 0);
5260 int failed
= a
->info
.array
.raid_disks
;
5261 struct mdinfo
*rv
= NULL
;
5264 struct metadata_update
*mu
;
5266 struct imsm_update_activate_spare
*u
;
5271 for (d
= a
->info
.devs
; d
; d
= d
->next
) {
5272 if ((d
->curr_state
& DS_FAULTY
) &&
5274 /* wait for Removal to happen */
5276 if (d
->state_fd
>= 0)
5280 dprintf("imsm: activate spare: inst=%d failed=%d (%d) level=%d\n",
5281 inst
, failed
, a
->info
.array
.raid_disks
, a
->info
.array
.level
);
5283 if (dev
->vol
.migr_state
&&
5284 dev
->vol
.migr_type
== MIGR_GEN_MIGR
)
5285 /* No repair during migration */
5288 if (imsm_check_degraded(super
, dev
, failed
) != IMSM_T_STATE_DEGRADED
)
5292 * If there are any failed disks check state of the other volume.
5293 * Block rebuild if the another one is failed until failed disks
5294 * are removed from container.
5297 dprintf("found failed disks in %s, check if there another"
5298 "failed sub-array.\n",
5300 /* check if states of the other volumes allow for rebuild */
5301 for (i
= 0; i
< super
->anchor
->num_raid_devs
; i
++) {
5303 allowed
= imsm_rebuild_allowed(a
->container
,
5311 /* For each slot, if it is not working, find a spare */
5312 for (i
= 0; i
< a
->info
.array
.raid_disks
; i
++) {
5313 for (d
= a
->info
.devs
; d
; d
= d
->next
)
5314 if (d
->disk
.raid_disk
== i
)
5316 dprintf("found %d: %p %x\n", i
, d
, d
?d
->curr_state
:0);
5317 if (d
&& (d
->state_fd
>= 0))
5321 * OK, this device needs recovery. Try to re-add the
5322 * previous occupant of this slot, if this fails see if
5323 * we can continue the assimilation of a spare that was
5324 * partially assimilated, finally try to activate a new
5327 dl
= imsm_readd(super
, i
, a
);
5329 dl
= imsm_add_spare(super
, i
, a
, 0, NULL
);
5331 dl
= imsm_add_spare(super
, i
, a
, 1, NULL
);
5335 /* found a usable disk with enough space */
5336 di
= malloc(sizeof(*di
));
5339 memset(di
, 0, sizeof(*di
));
5341 /* dl->index will be -1 in the case we are activating a
5342 * pristine spare. imsm_process_update() will create a
5343 * new index in this case. Once a disk is found to be
5344 * failed in all member arrays it is kicked from the
5347 di
->disk
.number
= dl
->index
;
5349 /* (ab)use di->devs to store a pointer to the device
5352 di
->devs
= (struct mdinfo
*) dl
;
5354 di
->disk
.raid_disk
= i
;
5355 di
->disk
.major
= dl
->major
;
5356 di
->disk
.minor
= dl
->minor
;
5358 di
->recovery_start
= 0;
5359 di
->data_offset
= __le32_to_cpu(map
->pba_of_lba0
);
5360 di
->component_size
= a
->info
.component_size
;
5361 di
->container_member
= inst
;
5362 super
->random
= random32();
5366 dprintf("%x:%x to be %d at %llu\n", dl
->major
, dl
->minor
,
5367 i
, di
->data_offset
);
5373 /* No spares found */
5375 /* Now 'rv' has a list of devices to return.
5376 * Create a metadata_update record to update the
5377 * disk_ord_tbl for the array
5379 mu
= malloc(sizeof(*mu
));
5381 mu
->buf
= malloc(sizeof(struct imsm_update_activate_spare
) * num_spares
);
5382 if (mu
->buf
== NULL
) {
5389 struct mdinfo
*n
= rv
->next
;
5398 mu
->space_list
= NULL
;
5399 mu
->len
= sizeof(struct imsm_update_activate_spare
) * num_spares
;
5400 mu
->next
= *updates
;
5401 u
= (struct imsm_update_activate_spare
*) mu
->buf
;
5403 for (di
= rv
; di
; di
= di
->next
) {
5404 u
->type
= update_activate_spare
;
5405 u
->dl
= (struct dl
*) di
->devs
;
5407 u
->slot
= di
->disk
.raid_disk
;
5418 static int disks_overlap(struct intel_super
*super
, int idx
, struct imsm_update_create_array
*u
)
5420 struct imsm_dev
*dev
= get_imsm_dev(super
, idx
);
5421 struct imsm_map
*map
= get_imsm_map(dev
, 0);
5422 struct imsm_map
*new_map
= get_imsm_map(&u
->dev
, 0);
5423 struct disk_info
*inf
= get_disk_info(u
);
5424 struct imsm_disk
*disk
;
5428 for (i
= 0; i
< map
->num_members
; i
++) {
5429 disk
= get_imsm_disk(super
, get_imsm_disk_idx(dev
, i
));
5430 for (j
= 0; j
< new_map
->num_members
; j
++)
5431 if (serialcmp(disk
->serial
, inf
[j
].serial
) == 0)
5439 static struct dl
*get_disk_super(struct intel_super
*super
, int major
, int minor
)
5441 struct dl
*dl
= NULL
;
5442 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
5443 if ((dl
->major
== major
) && (dl
->minor
== minor
))
5448 static int remove_disk_super(struct intel_super
*super
, int major
, int minor
)
5450 struct dl
*prev
= NULL
;
5454 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
5455 if ((dl
->major
== major
) && (dl
->minor
== minor
)) {
5458 prev
->next
= dl
->next
;
5460 super
->disks
= dl
->next
;
5462 __free_imsm_disk(dl
);
5463 dprintf("%s: removed %x:%x\n",
5464 __func__
, major
, minor
);
5472 static void imsm_delete(struct intel_super
*super
, struct dl
**dlp
, unsigned index
);
5474 static int add_remove_disk_update(struct intel_super
*super
)
5476 int check_degraded
= 0;
5477 struct dl
*disk
= NULL
;
5478 /* add/remove some spares to/from the metadata/contrainer */
5479 while (super
->disk_mgmt_list
) {
5480 struct dl
*disk_cfg
;
5482 disk_cfg
= super
->disk_mgmt_list
;
5483 super
->disk_mgmt_list
= disk_cfg
->next
;
5484 disk_cfg
->next
= NULL
;
5486 if (disk_cfg
->action
== DISK_ADD
) {
5487 disk_cfg
->next
= super
->disks
;
5488 super
->disks
= disk_cfg
;
5490 dprintf("%s: added %x:%x\n",
5491 __func__
, disk_cfg
->major
,
5493 } else if (disk_cfg
->action
== DISK_REMOVE
) {
5494 dprintf("Disk remove action processed: %x.%x\n",
5495 disk_cfg
->major
, disk_cfg
->minor
);
5496 disk
= get_disk_super(super
,
5500 /* store action status */
5501 disk
->action
= DISK_REMOVE
;
5502 /* remove spare disks only */
5503 if (disk
->index
== -1) {
5504 remove_disk_super(super
,
5509 /* release allocate disk structure */
5510 __free_imsm_disk(disk_cfg
);
5513 return check_degraded
;
5516 static void imsm_process_update(struct supertype
*st
,
5517 struct metadata_update
*update
)
5520 * crack open the metadata_update envelope to find the update record
5521 * update can be one of:
5522 * update_reshape_container_disks - all the arrays in the container
5523 * are being reshaped to have more devices. We need to mark
5524 * the arrays for general migration and convert selected spares
5525 * into active devices.
5526 * update_activate_spare - a spare device has replaced a failed
5527 * device in an array, update the disk_ord_tbl. If this disk is
5528 * present in all member arrays then also clear the SPARE_DISK
5530 * update_create_array
5532 * update_rename_array
5533 * update_add_remove_disk
5535 struct intel_super
*super
= st
->sb
;
5536 struct imsm_super
*mpb
;
5537 enum imsm_update_type type
= *(enum imsm_update_type
*) update
->buf
;
5539 /* update requires a larger buf but the allocation failed */
5540 if (super
->next_len
&& !super
->next_buf
) {
5541 super
->next_len
= 0;
5545 if (super
->next_buf
) {
5546 memcpy(super
->next_buf
, super
->buf
, super
->len
);
5548 super
->len
= super
->next_len
;
5549 super
->buf
= super
->next_buf
;
5551 super
->next_len
= 0;
5552 super
->next_buf
= NULL
;
5555 mpb
= super
->anchor
;
5558 case update_reshape_container_disks
: {
5559 struct imsm_update_reshape
*u
= (void *)update
->buf
;
5560 struct dl
*new_disk
;
5561 struct intel_dev
*id
;
5563 int delta_disks
= u
->new_raid_disks
- u
->old_raid_disks
;
5564 void **tofree
= NULL
;
5566 dprintf("imsm: imsm_process_update() for update_reshape\n");
5568 /* enable spares to use in array */
5569 for (i
= 0; i
< delta_disks
; i
++) {
5571 new_disk
= get_disk_super(super
,
5572 major(u
->new_disks
[i
]),
5573 minor(u
->new_disks
[i
]));
5574 if (new_disk
== NULL
|| new_disk
->index
< 0)
5575 goto update_reshape_exit
;
5577 new_disk
->index
= mpb
->num_disks
++;
5578 /* slot to fill in autolayout */
5579 new_disk
->raiddisk
= new_disk
->index
;
5580 new_disk
->disk
.status
|=
5582 new_disk
->disk
.status
&= ~SPARE_DISK
;
5585 dprintf("imsm: process_update(): update_reshape: volume set"\
5586 " mpb->num_raid_devs = %i\n", mpb
->num_raid_devs
);
5587 /* manage changes in volumes
5589 for (id
= super
->devlist
; id
; id
= id
->next
) {
5590 void **sp
= update
->space_list
;
5591 struct imsm_dev
*newdev
;
5592 struct imsm_map
*newmap
, *oldmap
;
5596 update
->space_list
= *sp
;
5598 /* Copy the dev, but not (all of) the map */
5599 memcpy(newdev
, id
->dev
, sizeof(*newdev
));
5600 oldmap
= get_imsm_map(id
->dev
, 0);
5601 newmap
= get_imsm_map(newdev
, 0);
5602 /* Copy the current map */
5603 memcpy(newmap
, oldmap
, sizeof_imsm_map(oldmap
));
5604 newdev
->vol
.migr_state
= 1;
5605 newdev
->vol
.curr_migr_unit
= 0;
5606 newdev
->vol
.migr_type
= MIGR_GEN_MIGR
;
5607 newmap
->num_members
= u
->new_raid_disks
;
5608 for (i
= 0; i
< delta_disks
; i
++) {
5609 set_imsm_ord_tbl_ent(newmap
,
5610 u
->old_raid_disks
+ i
,
5611 u
->old_raid_disks
+ i
);
5613 /* New map is correct, now need to save old map */
5614 oldmap
= get_imsm_map(newdev
, 1);
5615 memcpy(newmap
, oldmap
, sizeof_imsm_map(oldmap
));
5617 sp
= (void **)id
->dev
;
5623 update
->space_list
= tofree
;
5624 super
->updates_pending
++;
5625 update_reshape_exit
:
5628 case update_activate_spare
: {
5629 struct imsm_update_activate_spare
*u
= (void *) update
->buf
;
5630 struct imsm_dev
*dev
= get_imsm_dev(super
, u
->array
);
5631 struct imsm_map
*map
= get_imsm_map(dev
, 0);
5632 struct imsm_map
*migr_map
;
5633 struct active_array
*a
;
5634 struct imsm_disk
*disk
;
5639 int victim
= get_imsm_disk_idx(dev
, u
->slot
);
5642 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
5647 fprintf(stderr
, "error: imsm_activate_spare passed "
5648 "an unknown disk (index: %d)\n",
5653 super
->updates_pending
++;
5655 /* count failures (excluding rebuilds and the victim)
5656 * to determine map[0] state
5659 for (i
= 0; i
< map
->num_members
; i
++) {
5662 disk
= get_imsm_disk(super
, get_imsm_disk_idx(dev
, i
));
5663 if (!disk
|| is_failed(disk
))
5667 /* adding a pristine spare, assign a new index */
5668 if (dl
->index
< 0) {
5669 dl
->index
= super
->anchor
->num_disks
;
5670 super
->anchor
->num_disks
++;
5673 disk
->status
|= CONFIGURED_DISK
;
5674 disk
->status
&= ~SPARE_DISK
;
5677 to_state
= imsm_check_degraded(super
, dev
, failed
);
5678 map
->map_state
= IMSM_T_STATE_DEGRADED
;
5679 migrate(dev
, to_state
, MIGR_REBUILD
);
5680 migr_map
= get_imsm_map(dev
, 1);
5681 set_imsm_ord_tbl_ent(map
, u
->slot
, dl
->index
);
5682 set_imsm_ord_tbl_ent(migr_map
, u
->slot
, dl
->index
| IMSM_ORD_REBUILD
);
5684 /* update the family_num to mark a new container
5685 * generation, being careful to record the existing
5686 * family_num in orig_family_num to clean up after
5687 * earlier mdadm versions that neglected to set it.
5689 if (mpb
->orig_family_num
== 0)
5690 mpb
->orig_family_num
= mpb
->family_num
;
5691 mpb
->family_num
+= super
->random
;
5693 /* count arrays using the victim in the metadata */
5695 for (a
= st
->arrays
; a
; a
= a
->next
) {
5696 dev
= get_imsm_dev(super
, a
->info
.container_member
);
5697 map
= get_imsm_map(dev
, 0);
5699 if (get_imsm_disk_slot(map
, victim
) >= 0)
5703 /* delete the victim if it is no longer being
5709 /* We know that 'manager' isn't touching anything,
5710 * so it is safe to delete
5712 for (dlp
= &super
->disks
; *dlp
; dlp
= &(*dlp
)->next
)
5713 if ((*dlp
)->index
== victim
)
5716 /* victim may be on the missing list */
5718 for (dlp
= &super
->missing
; *dlp
; dlp
= &(*dlp
)->next
)
5719 if ((*dlp
)->index
== victim
)
5721 imsm_delete(super
, dlp
, victim
);
5725 case update_create_array
: {
5726 /* someone wants to create a new array, we need to be aware of
5727 * a few races/collisions:
5728 * 1/ 'Create' called by two separate instances of mdadm
5729 * 2/ 'Create' versus 'activate_spare': mdadm has chosen
5730 * devices that have since been assimilated via
5732 * In the event this update can not be carried out mdadm will
5733 * (FIX ME) notice that its update did not take hold.
5735 struct imsm_update_create_array
*u
= (void *) update
->buf
;
5736 struct intel_dev
*dv
;
5737 struct imsm_dev
*dev
;
5738 struct imsm_map
*map
, *new_map
;
5739 unsigned long long start
, end
;
5740 unsigned long long new_start
, new_end
;
5742 struct disk_info
*inf
;
5745 /* handle racing creates: first come first serve */
5746 if (u
->dev_idx
< mpb
->num_raid_devs
) {
5747 dprintf("%s: subarray %d already defined\n",
5748 __func__
, u
->dev_idx
);
5752 /* check update is next in sequence */
5753 if (u
->dev_idx
!= mpb
->num_raid_devs
) {
5754 dprintf("%s: can not create array %d expected index %d\n",
5755 __func__
, u
->dev_idx
, mpb
->num_raid_devs
);
5759 new_map
= get_imsm_map(&u
->dev
, 0);
5760 new_start
= __le32_to_cpu(new_map
->pba_of_lba0
);
5761 new_end
= new_start
+ __le32_to_cpu(new_map
->blocks_per_member
);
5762 inf
= get_disk_info(u
);
5764 /* handle activate_spare versus create race:
5765 * check to make sure that overlapping arrays do not include
5768 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
5769 dev
= get_imsm_dev(super
, i
);
5770 map
= get_imsm_map(dev
, 0);
5771 start
= __le32_to_cpu(map
->pba_of_lba0
);
5772 end
= start
+ __le32_to_cpu(map
->blocks_per_member
);
5773 if ((new_start
>= start
&& new_start
<= end
) ||
5774 (start
>= new_start
&& start
<= new_end
))
5779 if (disks_overlap(super
, i
, u
)) {
5780 dprintf("%s: arrays overlap\n", __func__
);
5785 /* check that prepare update was successful */
5786 if (!update
->space
) {
5787 dprintf("%s: prepare update failed\n", __func__
);
5791 /* check that all disks are still active before committing
5792 * changes. FIXME: could we instead handle this by creating a
5793 * degraded array? That's probably not what the user expects,
5794 * so better to drop this update on the floor.
5796 for (i
= 0; i
< new_map
->num_members
; i
++) {
5797 dl
= serial_to_dl(inf
[i
].serial
, super
);
5799 dprintf("%s: disk disappeared\n", __func__
);
5804 super
->updates_pending
++;
5806 /* convert spares to members and fixup ord_tbl */
5807 for (i
= 0; i
< new_map
->num_members
; i
++) {
5808 dl
= serial_to_dl(inf
[i
].serial
, super
);
5809 if (dl
->index
== -1) {
5810 dl
->index
= mpb
->num_disks
;
5812 dl
->disk
.status
|= CONFIGURED_DISK
;
5813 dl
->disk
.status
&= ~SPARE_DISK
;
5815 set_imsm_ord_tbl_ent(new_map
, i
, dl
->index
);
5820 update
->space
= NULL
;
5821 imsm_copy_dev(dev
, &u
->dev
);
5822 dv
->index
= u
->dev_idx
;
5823 dv
->next
= super
->devlist
;
5824 super
->devlist
= dv
;
5825 mpb
->num_raid_devs
++;
5827 imsm_update_version_info(super
);
5830 /* mdmon knows how to release update->space, but not
5831 * ((struct intel_dev *) update->space)->dev
5833 if (update
->space
) {
5839 case update_kill_array
: {
5840 struct imsm_update_kill_array
*u
= (void *) update
->buf
;
5841 int victim
= u
->dev_idx
;
5842 struct active_array
*a
;
5843 struct intel_dev
**dp
;
5844 struct imsm_dev
*dev
;
5846 /* sanity check that we are not affecting the uuid of
5847 * active arrays, or deleting an active array
5849 * FIXME when immutable ids are available, but note that
5850 * we'll also need to fixup the invalidated/active
5851 * subarray indexes in mdstat
5853 for (a
= st
->arrays
; a
; a
= a
->next
)
5854 if (a
->info
.container_member
>= victim
)
5856 /* by definition if mdmon is running at least one array
5857 * is active in the container, so checking
5858 * mpb->num_raid_devs is just extra paranoia
5860 dev
= get_imsm_dev(super
, victim
);
5861 if (a
|| !dev
|| mpb
->num_raid_devs
== 1) {
5862 dprintf("failed to delete subarray-%d\n", victim
);
5866 for (dp
= &super
->devlist
; *dp
;)
5867 if ((*dp
)->index
== (unsigned)super
->current_vol
) {
5870 if ((*dp
)->index
> (unsigned)victim
)
5874 mpb
->num_raid_devs
--;
5875 super
->updates_pending
++;
5878 case update_rename_array
: {
5879 struct imsm_update_rename_array
*u
= (void *) update
->buf
;
5880 char name
[MAX_RAID_SERIAL_LEN
+1];
5881 int target
= u
->dev_idx
;
5882 struct active_array
*a
;
5883 struct imsm_dev
*dev
;
5885 /* sanity check that we are not affecting the uuid of
5888 snprintf(name
, MAX_RAID_SERIAL_LEN
, "%s", (char *) u
->name
);
5889 name
[MAX_RAID_SERIAL_LEN
] = '\0';
5890 for (a
= st
->arrays
; a
; a
= a
->next
)
5891 if (a
->info
.container_member
== target
)
5893 dev
= get_imsm_dev(super
, u
->dev_idx
);
5894 if (a
|| !dev
|| !check_name(super
, name
, 1)) {
5895 dprintf("failed to rename subarray-%d\n", target
);
5899 snprintf((char *) dev
->volume
, MAX_RAID_SERIAL_LEN
, "%s", name
);
5900 super
->updates_pending
++;
5903 case update_add_remove_disk
: {
5904 /* we may be able to repair some arrays if disks are
5905 * being added, check teh status of add_remove_disk
5906 * if discs has been added.
5908 if (add_remove_disk_update(super
)) {
5909 struct active_array
*a
;
5911 super
->updates_pending
++;
5912 for (a
= st
->arrays
; a
; a
= a
->next
)
5913 a
->check_degraded
= 1;
5918 fprintf(stderr
, "error: unsuported process update type:"
5919 "(type: %d)\n", type
);
5923 static void imsm_prepare_update(struct supertype
*st
,
5924 struct metadata_update
*update
)
5927 * Allocate space to hold new disk entries, raid-device entries or a new
5928 * mpb if necessary. The manager synchronously waits for updates to
5929 * complete in the monitor, so new mpb buffers allocated here can be
5930 * integrated by the monitor thread without worrying about live pointers
5931 * in the manager thread.
5933 enum imsm_update_type type
= *(enum imsm_update_type
*) update
->buf
;
5934 struct intel_super
*super
= st
->sb
;
5935 struct imsm_super
*mpb
= super
->anchor
;
5940 case update_reshape_container_disks
: {
5941 /* Every raid device in the container is about to
5942 * gain some more devices, and we will enter a
5944 * So each 'imsm_map' will be bigger, and the imsm_vol
5945 * will now hold 2 of them.
5946 * Thus we need new 'struct imsm_dev' allocations sized
5947 * as sizeof_imsm_dev but with more devices in both maps.
5949 struct imsm_update_reshape
*u
= (void *)update
->buf
;
5950 struct intel_dev
*dl
;
5951 void **space_tail
= (void**)&update
->space_list
;
5953 dprintf("imsm: imsm_prepare_update() for update_reshape\n");
5955 for (dl
= super
->devlist
; dl
; dl
= dl
->next
) {
5956 int size
= sizeof_imsm_dev(dl
->dev
, 1);
5958 size
+= sizeof(__u32
) * 2 *
5959 (u
->new_raid_disks
- u
->old_raid_disks
);
5968 len
= disks_to_mpb_size(u
->new_raid_disks
);
5969 dprintf("New anchor length is %llu\n", (unsigned long long)len
);
5972 case update_create_array
: {
5973 struct imsm_update_create_array
*u
= (void *) update
->buf
;
5974 struct intel_dev
*dv
;
5975 struct imsm_dev
*dev
= &u
->dev
;
5976 struct imsm_map
*map
= get_imsm_map(dev
, 0);
5978 struct disk_info
*inf
;
5982 inf
= get_disk_info(u
);
5983 len
= sizeof_imsm_dev(dev
, 1);
5984 /* allocate a new super->devlist entry */
5985 dv
= malloc(sizeof(*dv
));
5987 dv
->dev
= malloc(len
);
5992 update
->space
= NULL
;
5996 /* count how many spares will be converted to members */
5997 for (i
= 0; i
< map
->num_members
; i
++) {
5998 dl
= serial_to_dl(inf
[i
].serial
, super
);
6000 /* hmm maybe it failed?, nothing we can do about
6005 if (count_memberships(dl
, super
) == 0)
6008 len
+= activate
* sizeof(struct imsm_disk
);
6015 /* check if we need a larger metadata buffer */
6016 if (super
->next_buf
)
6017 buf_len
= super
->next_len
;
6019 buf_len
= super
->len
;
6021 if (__le32_to_cpu(mpb
->mpb_size
) + len
> buf_len
) {
6022 /* ok we need a larger buf than what is currently allocated
6023 * if this allocation fails process_update will notice that
6024 * ->next_len is set and ->next_buf is NULL
6026 buf_len
= ROUND_UP(__le32_to_cpu(mpb
->mpb_size
) + len
, 512);
6027 if (super
->next_buf
)
6028 free(super
->next_buf
);
6030 super
->next_len
= buf_len
;
6031 if (posix_memalign(&super
->next_buf
, 512, buf_len
) == 0)
6032 memset(super
->next_buf
, 0, buf_len
);
6034 super
->next_buf
= NULL
;
6038 /* must be called while manager is quiesced */
6039 static void imsm_delete(struct intel_super
*super
, struct dl
**dlp
, unsigned index
)
6041 struct imsm_super
*mpb
= super
->anchor
;
6043 struct imsm_dev
*dev
;
6044 struct imsm_map
*map
;
6045 int i
, j
, num_members
;
6048 dprintf("%s: deleting device[%d] from imsm_super\n",
6051 /* shift all indexes down one */
6052 for (iter
= super
->disks
; iter
; iter
= iter
->next
)
6053 if (iter
->index
> (int)index
)
6055 for (iter
= super
->missing
; iter
; iter
= iter
->next
)
6056 if (iter
->index
> (int)index
)
6059 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
6060 dev
= get_imsm_dev(super
, i
);
6061 map
= get_imsm_map(dev
, 0);
6062 num_members
= map
->num_members
;
6063 for (j
= 0; j
< num_members
; j
++) {
6064 /* update ord entries being careful not to propagate
6065 * ord-flags to the first map
6067 ord
= get_imsm_ord_tbl_ent(dev
, j
);
6069 if (ord_to_idx(ord
) <= index
)
6072 map
= get_imsm_map(dev
, 0);
6073 set_imsm_ord_tbl_ent(map
, j
, ord_to_idx(ord
- 1));
6074 map
= get_imsm_map(dev
, 1);
6076 set_imsm_ord_tbl_ent(map
, j
, ord
- 1);
6081 super
->updates_pending
++;
6083 struct dl
*dl
= *dlp
;
6085 *dlp
= (*dlp
)->next
;
6086 __free_imsm_disk(dl
);
6089 #endif /* MDASSEMBLE */
6091 static char disk_by_path
[] = "/dev/disk/by-path/";
6093 static const char *imsm_get_disk_controller_domain(const char *path
)
6095 struct sys_dev
*list
, *hba
= NULL
;
6096 char disk_path
[PATH_MAX
];
6100 list
= find_driver_devices("pci", "ahci");
6101 for (hba
= list
; hba
; hba
= hba
->next
)
6102 if (devpath_to_vendor(hba
->path
) == 0x8086)
6108 strncpy(disk_path
, disk_by_path
, PATH_MAX
- 1);
6109 strncat(disk_path
, path
, PATH_MAX
- strlen(disk_path
) - 1);
6110 if (stat(disk_path
, &st
) == 0) {
6111 dpath
= devt_to_devpath(st
.st_rdev
);
6113 ahci
= path_attached_to_hba(dpath
, hba
->path
);
6116 dprintf("path: %s(%s) hba: %s attached: %d\n",
6117 path
, dpath
, (hba
) ? hba
->path
: "NULL", ahci
);
6118 free_sys_dev(&list
);
6125 static int imsm_find_array_minor_by_subdev(int subdev
, int container
, int *minor
)
6127 char subdev_name
[20];
6128 struct mdstat_ent
*mdstat
;
6130 sprintf(subdev_name
, "%d", subdev
);
6131 mdstat
= mdstat_by_subdev(subdev_name
, container
);
6135 *minor
= mdstat
->devnum
;
6136 free_mdstat(mdstat
);
6140 static int imsm_reshape_is_allowed_on_container(struct supertype
*st
,
6141 struct geo_params
*geo
,
6142 int *old_raid_disks
)
6145 struct mdinfo
*info
, *member
;
6146 int devices_that_can_grow
= 0;
6148 dprintf("imsm: imsm_reshape_is_allowed_on_container(ENTER): "
6149 "st->devnum = (%i)\n",
6152 if (geo
->size
!= -1 ||
6153 geo
->level
!= UnSet
||
6154 geo
->layout
!= UnSet
||
6155 geo
->chunksize
!= 0 ||
6156 geo
->raid_disks
== UnSet
) {
6157 dprintf("imsm: Container operation is allowed for "
6158 "raid disks number change only.\n");
6162 info
= container_content_imsm(st
, NULL
);
6163 for (member
= info
; member
; member
= member
->next
) {
6167 dprintf("imsm: checking device_num: %i\n",
6168 member
->container_member
);
6170 if (geo
->raid_disks
< member
->array
.raid_disks
) {
6171 /* we work on container for Online Capacity Expansion
6172 * only so raid_disks has to grow
6174 dprintf("imsm: for container operation raid disks "
6175 "increase is required\n");
6179 if ((info
->array
.level
!= 0) &&
6180 (info
->array
.level
!= 5)) {
6181 /* we cannot use this container with other raid level
6183 dprintf("imsm: for container operation wrong"\
6184 " raid level (%i) detected\n",
6188 /* check for platform support
6189 * for this raid level configuration
6191 struct intel_super
*super
= st
->sb
;
6192 if (!is_raid_level_supported(super
->orom
,
6193 member
->array
.level
,
6195 dprintf("platform does not support raid%d with"\
6199 geo
->raid_disks
> 1 ? "s" : "");
6204 if (*old_raid_disks
&&
6205 info
->array
.raid_disks
!= *old_raid_disks
)
6207 *old_raid_disks
= info
->array
.raid_disks
;
6209 /* All raid5 and raid0 volumes in container
6210 * have to be ready for Online Capacity Expansion
6211 * so they need to be assembled. We have already
6212 * checked that no recovery etc is happening.
6214 result
= imsm_find_array_minor_by_subdev(member
->container_member
,
6218 dprintf("imsm: cannot find array\n");
6221 devices_that_can_grow
++;
6224 if (!member
&& devices_that_can_grow
)
6228 dprintf("\tContainer operation allowed\n");
6230 dprintf("\tError: %i\n", ret_val
);
6235 /* Function: get_spares_for_grow
6236 * Description: Allocates memory and creates list of spare devices
6237 * avaliable in container. Checks if spare drive size is acceptable.
6238 * Parameters: Pointer to the supertype structure
6239 * Returns: Pointer to the list of spare devices (mdinfo structure) on success,
6242 static struct mdinfo
*get_spares_for_grow(struct supertype
*st
)
6244 unsigned long long min_size
= min_acceptable_spare_size_imsm(st
);
6245 return container_choose_spares(st
, min_size
, NULL
, NULL
, NULL
, 0);
6248 /******************************************************************************
6249 * function: imsm_create_metadata_update_for_reshape
6250 * Function creates update for whole IMSM container.
6252 ******************************************************************************/
6253 static int imsm_create_metadata_update_for_reshape(
6254 struct supertype
*st
,
6255 struct geo_params
*geo
,
6257 struct imsm_update_reshape
**updatep
)
6259 struct intel_super
*super
= st
->sb
;
6260 struct imsm_super
*mpb
= super
->anchor
;
6261 int update_memory_size
= 0;
6262 struct imsm_update_reshape
*u
= NULL
;
6263 struct mdinfo
*spares
= NULL
;
6265 int delta_disks
= 0;
6267 dprintf("imsm_update_metadata_for_reshape(enter) raid_disks = %i\n",
6270 delta_disks
= geo
->raid_disks
- old_raid_disks
;
6272 /* size of all update data without anchor */
6273 update_memory_size
= sizeof(struct imsm_update_reshape
);
6275 /* now add space for spare disks that we need to add. */
6276 update_memory_size
+= sizeof(u
->new_disks
[0]) * (delta_disks
- 1);
6278 u
= calloc(1, update_memory_size
);
6281 "cannot get memory for imsm_update_reshape update\n");
6284 u
->type
= update_reshape_container_disks
;
6285 u
->old_raid_disks
= old_raid_disks
;
6286 u
->new_raid_disks
= geo
->raid_disks
;
6288 /* now get spare disks list
6290 spares
= get_spares_for_grow(st
);
6293 || delta_disks
> spares
->array
.spare_disks
) {
6294 dprintf("imsm: ERROR: Cannot get spare devices.\n");
6298 /* we have got spares
6299 * update disk list in imsm_disk list table in anchor
6301 dprintf("imsm: %i spares are available.\n\n",
6302 spares
->array
.spare_disks
);
6304 for (i
= 0; i
< delta_disks
; i
++) {
6305 struct mdinfo
*dev
= spares
->devs
;
6308 u
->new_disks
[i
] = makedev(dev
->disk
.major
,
6310 dl
= get_disk_super(super
, dev
->disk
.major
, dev
->disk
.minor
);
6311 dl
->index
= mpb
->num_disks
++;
6313 /* Now update the metadata so that container_content will find
6316 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
6318 struct imsm_dev
*dev
= get_imsm_dev(super
, i
);
6319 struct imsm_map
*map
= get_imsm_map(dev
, 0);
6320 map
->num_members
= geo
->raid_disks
;
6321 for (d
= 0; d
< delta_disks
; d
++) {
6322 set_imsm_ord_tbl_ent(map
, old_raid_disks
+ d
,
6323 mpb
->num_disks
- delta_disks
+ d
);
6332 if (i
== delta_disks
) {
6334 return update_memory_size
;
6342 static int imsm_reshape_super(struct supertype
*st
, long long size
, int level
,
6343 int layout
, int chunksize
, int raid_disks
,
6344 char *backup
, char *dev
, int verbouse
)
6346 /* currently we only support increasing the number of devices
6347 * for a container. This increases the number of device for each
6348 * member array. They must all be RAID0 or RAID5.
6352 struct geo_params geo
;
6354 dprintf("imsm: reshape_super called.\n");
6356 memset(&geo
, sizeof(struct geo_params
), 0);
6361 geo
.layout
= layout
;
6362 geo
.chunksize
= chunksize
;
6363 geo
.raid_disks
= raid_disks
;
6365 dprintf("\tfor level : %i\n", geo
.level
);
6366 dprintf("\tfor raid_disks : %i\n", geo
.raid_disks
);
6368 if (experimental() == 0)
6371 /* verify reshape conditions
6372 * on container level we can only increase number of devices. */
6373 if (st
->container_dev
== st
->devnum
) {
6374 /* check for delta_disks > 0
6375 *and supported raid levels 0 and 5 only in container */
6376 int old_raid_disks
= 0;
6377 if (imsm_reshape_is_allowed_on_container(
6378 st
, &geo
, &old_raid_disks
)) {
6379 struct imsm_update_reshape
*u
= NULL
;
6382 len
= imsm_create_metadata_update_for_reshape(
6383 st
, &geo
, old_raid_disks
, &u
);
6387 append_metadata_update(st
, u
, len
);
6389 dprintf("imsm: Cannot prepare "\
6392 dprintf("imsm: Operation is not allowed "\
6393 "on this container\n");
6395 dprintf("imsm: not a container operation\n");
6397 dprintf("imsm: reshape_super Exit code = %i\n", ret_val
);
6401 struct superswitch super_imsm
= {
6403 .examine_super
= examine_super_imsm
,
6404 .brief_examine_super
= brief_examine_super_imsm
,
6405 .brief_examine_subarrays
= brief_examine_subarrays_imsm
,
6406 .export_examine_super
= export_examine_super_imsm
,
6407 .detail_super
= detail_super_imsm
,
6408 .brief_detail_super
= brief_detail_super_imsm
,
6409 .write_init_super
= write_init_super_imsm
,
6410 .validate_geometry
= validate_geometry_imsm
,
6411 .add_to_super
= add_to_super_imsm
,
6412 .remove_from_super
= remove_from_super_imsm
,
6413 .detail_platform
= detail_platform_imsm
,
6414 .kill_subarray
= kill_subarray_imsm
,
6415 .update_subarray
= update_subarray_imsm
,
6416 .load_container
= load_container_imsm
,
6418 .match_home
= match_home_imsm
,
6419 .uuid_from_super
= uuid_from_super_imsm
,
6420 .getinfo_super
= getinfo_super_imsm
,
6421 .getinfo_super_disks
= getinfo_super_disks_imsm
,
6422 .update_super
= update_super_imsm
,
6424 .avail_size
= avail_size_imsm
,
6425 .min_acceptable_spare_size
= min_acceptable_spare_size_imsm
,
6427 .compare_super
= compare_super_imsm
,
6429 .load_super
= load_super_imsm
,
6430 .init_super
= init_super_imsm
,
6431 .store_super
= store_super_imsm
,
6432 .free_super
= free_super_imsm
,
6433 .match_metadata_desc
= match_metadata_desc_imsm
,
6434 .container_content
= container_content_imsm
,
6435 .default_geometry
= default_geometry_imsm
,
6436 .get_disk_controller_domain
= imsm_get_disk_controller_domain
,
6437 .reshape_super
= imsm_reshape_super
,
6444 .open_new
= imsm_open_new
,
6445 .set_array_state
= imsm_set_array_state
,
6446 .set_disk
= imsm_set_disk
,
6447 .sync_metadata
= imsm_sync_metadata
,
6448 .activate_spare
= imsm_activate_spare
,
6449 .process_update
= imsm_process_update
,
6450 .prepare_update
= imsm_prepare_update
,
6451 #endif /* MDASSEMBLE */