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
,
296 struct imsm_update_activate_spare
{
297 enum imsm_update_type type
;
301 struct imsm_update_activate_spare
*next
;
305 __u8 serial
[MAX_RAID_SERIAL_LEN
];
308 struct imsm_update_create_array
{
309 enum imsm_update_type type
;
314 struct imsm_update_kill_array
{
315 enum imsm_update_type type
;
319 struct imsm_update_rename_array
{
320 enum imsm_update_type type
;
321 __u8 name
[MAX_RAID_SERIAL_LEN
];
325 struct imsm_update_add_remove_disk
{
326 enum imsm_update_type type
;
329 static struct supertype
*match_metadata_desc_imsm(char *arg
)
331 struct supertype
*st
;
333 if (strcmp(arg
, "imsm") != 0 &&
334 strcmp(arg
, "default") != 0
338 st
= malloc(sizeof(*st
));
341 memset(st
, 0, sizeof(*st
));
342 st
->container_dev
= NoMdDev
;
343 st
->ss
= &super_imsm
;
344 st
->max_devs
= IMSM_MAX_DEVICES
;
345 st
->minor_version
= 0;
351 static __u8
*get_imsm_version(struct imsm_super
*mpb
)
353 return &mpb
->sig
[MPB_SIG_LEN
];
357 /* retrieve a disk directly from the anchor when the anchor is known to be
358 * up-to-date, currently only at load time
360 static struct imsm_disk
*__get_imsm_disk(struct imsm_super
*mpb
, __u8 index
)
362 if (index
>= mpb
->num_disks
)
364 return &mpb
->disk
[index
];
367 /* retrieve the disk description based on a index of the disk
370 static struct dl
*get_imsm_dl_disk(struct intel_super
*super
, __u8 index
)
374 for (d
= super
->disks
; d
; d
= d
->next
)
375 if (d
->index
== index
)
380 /* retrieve a disk from the parsed metadata */
381 static struct imsm_disk
*get_imsm_disk(struct intel_super
*super
, __u8 index
)
385 dl
= get_imsm_dl_disk(super
, index
);
392 /* generate a checksum directly from the anchor when the anchor is known to be
393 * up-to-date, currently only at load or write_super after coalescing
395 static __u32
__gen_imsm_checksum(struct imsm_super
*mpb
)
397 __u32 end
= mpb
->mpb_size
/ sizeof(end
);
398 __u32
*p
= (__u32
*) mpb
;
402 sum
+= __le32_to_cpu(*p
);
406 return sum
- __le32_to_cpu(mpb
->check_sum
);
409 static size_t sizeof_imsm_map(struct imsm_map
*map
)
411 return sizeof(struct imsm_map
) + sizeof(__u32
) * (map
->num_members
- 1);
414 struct imsm_map
*get_imsm_map(struct imsm_dev
*dev
, int second_map
)
416 struct imsm_map
*map
= &dev
->vol
.map
[0];
418 if (second_map
&& !dev
->vol
.migr_state
)
420 else if (second_map
) {
423 return ptr
+ sizeof_imsm_map(map
);
429 /* return the size of the device.
430 * migr_state increases the returned size if map[0] were to be duplicated
432 static size_t sizeof_imsm_dev(struct imsm_dev
*dev
, int migr_state
)
434 size_t size
= sizeof(*dev
) - sizeof(struct imsm_map
) +
435 sizeof_imsm_map(get_imsm_map(dev
, 0));
437 /* migrating means an additional map */
438 if (dev
->vol
.migr_state
)
439 size
+= sizeof_imsm_map(get_imsm_map(dev
, 1));
441 size
+= sizeof_imsm_map(get_imsm_map(dev
, 0));
447 /* retrieve disk serial number list from a metadata update */
448 static struct disk_info
*get_disk_info(struct imsm_update_create_array
*update
)
451 struct disk_info
*inf
;
453 inf
= u
+ sizeof(*update
) - sizeof(struct imsm_dev
) +
454 sizeof_imsm_dev(&update
->dev
, 0);
460 static struct imsm_dev
*__get_imsm_dev(struct imsm_super
*mpb
, __u8 index
)
466 if (index
>= mpb
->num_raid_devs
)
469 /* devices start after all disks */
470 offset
= ((void *) &mpb
->disk
[mpb
->num_disks
]) - _mpb
;
472 for (i
= 0; i
<= index
; i
++)
474 return _mpb
+ offset
;
476 offset
+= sizeof_imsm_dev(_mpb
+ offset
, 0);
481 static struct imsm_dev
*get_imsm_dev(struct intel_super
*super
, __u8 index
)
483 struct intel_dev
*dv
;
485 if (index
>= super
->anchor
->num_raid_devs
)
487 for (dv
= super
->devlist
; dv
; dv
= dv
->next
)
488 if (dv
->index
== index
)
493 static __u32
get_imsm_ord_tbl_ent(struct imsm_dev
*dev
, int slot
)
495 struct imsm_map
*map
;
497 if (dev
->vol
.migr_state
)
498 map
= get_imsm_map(dev
, 1);
500 map
= get_imsm_map(dev
, 0);
502 /* top byte identifies disk under rebuild */
503 return __le32_to_cpu(map
->disk_ord_tbl
[slot
]);
506 #define ord_to_idx(ord) (((ord) << 8) >> 8)
507 static __u32
get_imsm_disk_idx(struct imsm_dev
*dev
, int slot
)
509 __u32 ord
= get_imsm_ord_tbl_ent(dev
, slot
);
511 return ord_to_idx(ord
);
514 static void set_imsm_ord_tbl_ent(struct imsm_map
*map
, int slot
, __u32 ord
)
516 map
->disk_ord_tbl
[slot
] = __cpu_to_le32(ord
);
519 static int get_imsm_disk_slot(struct imsm_map
*map
, unsigned idx
)
524 for (slot
= 0; slot
< map
->num_members
; slot
++) {
525 ord
= __le32_to_cpu(map
->disk_ord_tbl
[slot
]);
526 if (ord_to_idx(ord
) == idx
)
533 static int get_imsm_raid_level(struct imsm_map
*map
)
535 if (map
->raid_level
== 1) {
536 if (map
->num_members
== 2)
542 return map
->raid_level
;
545 static int cmp_extent(const void *av
, const void *bv
)
547 const struct extent
*a
= av
;
548 const struct extent
*b
= bv
;
549 if (a
->start
< b
->start
)
551 if (a
->start
> b
->start
)
556 static int count_memberships(struct dl
*dl
, struct intel_super
*super
)
561 for (i
= 0; i
< super
->anchor
->num_raid_devs
; i
++) {
562 struct imsm_dev
*dev
= get_imsm_dev(super
, i
);
563 struct imsm_map
*map
= get_imsm_map(dev
, 0);
565 if (get_imsm_disk_slot(map
, dl
->index
) >= 0)
572 static struct extent
*get_extents(struct intel_super
*super
, struct dl
*dl
)
574 /* find a list of used extents on the given physical device */
575 struct extent
*rv
, *e
;
577 int memberships
= count_memberships(dl
, super
);
578 __u32 reservation
= MPB_SECTOR_CNT
+ IMSM_RESERVED_SECTORS
;
580 rv
= malloc(sizeof(struct extent
) * (memberships
+ 1));
585 for (i
= 0; i
< super
->anchor
->num_raid_devs
; i
++) {
586 struct imsm_dev
*dev
= get_imsm_dev(super
, i
);
587 struct imsm_map
*map
= get_imsm_map(dev
, 0);
589 if (get_imsm_disk_slot(map
, dl
->index
) >= 0) {
590 e
->start
= __le32_to_cpu(map
->pba_of_lba0
);
591 e
->size
= __le32_to_cpu(map
->blocks_per_member
);
595 qsort(rv
, memberships
, sizeof(*rv
), cmp_extent
);
597 /* determine the start of the metadata
598 * when no raid devices are defined use the default
599 * ...otherwise allow the metadata to truncate the value
600 * as is the case with older versions of imsm
603 struct extent
*last
= &rv
[memberships
- 1];
606 remainder
= __le32_to_cpu(dl
->disk
.total_blocks
) -
607 (last
->start
+ last
->size
);
608 /* round down to 1k block to satisfy precision of the kernel
612 /* make sure remainder is still sane */
613 if (remainder
< (unsigned)ROUND_UP(super
->len
, 512) >> 9)
614 remainder
= ROUND_UP(super
->len
, 512) >> 9;
615 if (reservation
> remainder
)
616 reservation
= remainder
;
618 e
->start
= __le32_to_cpu(dl
->disk
.total_blocks
) - reservation
;
623 /* try to determine how much space is reserved for metadata from
624 * the last get_extents() entry, otherwise fallback to the
627 static __u32
imsm_reserved_sectors(struct intel_super
*super
, struct dl
*dl
)
633 /* for spares just return a minimal reservation which will grow
634 * once the spare is picked up by an array
637 return MPB_SECTOR_CNT
;
639 e
= get_extents(super
, dl
);
641 return MPB_SECTOR_CNT
+ IMSM_RESERVED_SECTORS
;
643 /* scroll to last entry */
644 for (i
= 0; e
[i
].size
; i
++)
647 rv
= __le32_to_cpu(dl
->disk
.total_blocks
) - e
[i
].start
;
654 static int is_spare(struct imsm_disk
*disk
)
656 return (disk
->status
& SPARE_DISK
) == SPARE_DISK
;
659 static int is_configured(struct imsm_disk
*disk
)
661 return (disk
->status
& CONFIGURED_DISK
) == CONFIGURED_DISK
;
664 static int is_failed(struct imsm_disk
*disk
)
666 return (disk
->status
& FAILED_DISK
) == FAILED_DISK
;
669 /* Return minimum size of a spare that can be used in this array*/
670 static unsigned long long min_acceptable_spare_size_imsm(struct supertype
*st
)
672 struct intel_super
*super
= st
->sb
;
676 unsigned long long rv
= 0;
680 /* find first active disk in array */
682 while (dl
&& (is_failed(&dl
->disk
) || dl
->index
== -1))
686 /* find last lba used by subarrays */
687 e
= get_extents(super
, dl
);
690 for (i
= 0; e
[i
].size
; i
++)
693 rv
= e
[i
-1].start
+ e
[i
-1].size
;
695 /* add the amount of space needed for metadata */
696 rv
= rv
+ MPB_SECTOR_CNT
+ IMSM_RESERVED_SECTORS
;
701 static __u64
blocks_per_migr_unit(struct imsm_dev
*dev
);
703 static void print_imsm_dev(struct imsm_dev
*dev
, char *uuid
, int disk_idx
)
707 struct imsm_map
*map
= get_imsm_map(dev
, 0);
711 printf("[%.16s]:\n", dev
->volume
);
712 printf(" UUID : %s\n", uuid
);
713 printf(" RAID Level : %d\n", get_imsm_raid_level(map
));
714 printf(" Members : %d\n", map
->num_members
);
715 printf(" Slots : [");
716 for (i
= 0; i
< map
->num_members
; i
++) {
717 ord
= get_imsm_ord_tbl_ent(dev
, i
);
718 printf("%s", ord
& IMSM_ORD_REBUILD
? "_" : "U");
721 slot
= get_imsm_disk_slot(map
, disk_idx
);
723 ord
= get_imsm_ord_tbl_ent(dev
, slot
);
724 printf(" This Slot : %d%s\n", slot
,
725 ord
& IMSM_ORD_REBUILD
? " (out-of-sync)" : "");
727 printf(" This Slot : ?\n");
728 sz
= __le32_to_cpu(dev
->size_high
);
730 sz
+= __le32_to_cpu(dev
->size_low
);
731 printf(" Array Size : %llu%s\n", (unsigned long long)sz
,
732 human_size(sz
* 512));
733 sz
= __le32_to_cpu(map
->blocks_per_member
);
734 printf(" Per Dev Size : %llu%s\n", (unsigned long long)sz
,
735 human_size(sz
* 512));
736 printf(" Sector Offset : %u\n",
737 __le32_to_cpu(map
->pba_of_lba0
));
738 printf(" Num Stripes : %u\n",
739 __le32_to_cpu(map
->num_data_stripes
));
740 printf(" Chunk Size : %u KiB\n",
741 __le16_to_cpu(map
->blocks_per_strip
) / 2);
742 printf(" Reserved : %d\n", __le32_to_cpu(dev
->reserved_blocks
));
743 printf(" Migrate State : ");
744 if (dev
->vol
.migr_state
) {
745 if (migr_type(dev
) == MIGR_INIT
)
746 printf("initialize\n");
747 else if (migr_type(dev
) == MIGR_REBUILD
)
749 else if (migr_type(dev
) == MIGR_VERIFY
)
751 else if (migr_type(dev
) == MIGR_GEN_MIGR
)
752 printf("general migration\n");
753 else if (migr_type(dev
) == MIGR_STATE_CHANGE
)
754 printf("state change\n");
755 else if (migr_type(dev
) == MIGR_REPAIR
)
758 printf("<unknown:%d>\n", migr_type(dev
));
761 printf(" Map State : %s", map_state_str
[map
->map_state
]);
762 if (dev
->vol
.migr_state
) {
763 struct imsm_map
*map
= get_imsm_map(dev
, 1);
765 printf(" <-- %s", map_state_str
[map
->map_state
]);
766 printf("\n Checkpoint : %u (%llu)",
767 __le32_to_cpu(dev
->vol
.curr_migr_unit
),
768 (unsigned long long)blocks_per_migr_unit(dev
));
771 printf(" Dirty State : %s\n", dev
->vol
.dirty
? "dirty" : "clean");
774 static void print_imsm_disk(struct imsm_super
*mpb
, int index
, __u32 reserved
)
776 struct imsm_disk
*disk
= __get_imsm_disk(mpb
, index
);
777 char str
[MAX_RAID_SERIAL_LEN
+ 1];
780 if (index
< 0 || !disk
)
784 snprintf(str
, MAX_RAID_SERIAL_LEN
+ 1, "%s", disk
->serial
);
785 printf(" Disk%02d Serial : %s\n", index
, str
);
786 printf(" State :%s%s%s\n", is_spare(disk
) ? " spare" : "",
787 is_configured(disk
) ? " active" : "",
788 is_failed(disk
) ? " failed" : "");
789 printf(" Id : %08x\n", __le32_to_cpu(disk
->scsi_id
));
790 sz
= __le32_to_cpu(disk
->total_blocks
) - reserved
;
791 printf(" Usable Size : %llu%s\n", (unsigned long long)sz
,
792 human_size(sz
* 512));
795 static void getinfo_super_imsm(struct supertype
*st
, struct mdinfo
*info
, char *map
);
797 static void examine_super_imsm(struct supertype
*st
, char *homehost
)
799 struct intel_super
*super
= st
->sb
;
800 struct imsm_super
*mpb
= super
->anchor
;
801 char str
[MAX_SIGNATURE_LENGTH
];
806 __u32 reserved
= imsm_reserved_sectors(super
, super
->disks
);
809 snprintf(str
, MPB_SIG_LEN
, "%s", mpb
->sig
);
810 printf(" Magic : %s\n", str
);
811 snprintf(str
, strlen(MPB_VERSION_RAID0
), "%s", get_imsm_version(mpb
));
812 printf(" Version : %s\n", get_imsm_version(mpb
));
813 printf(" Orig Family : %08x\n", __le32_to_cpu(mpb
->orig_family_num
));
814 printf(" Family : %08x\n", __le32_to_cpu(mpb
->family_num
));
815 printf(" Generation : %08x\n", __le32_to_cpu(mpb
->generation_num
));
816 getinfo_super_imsm(st
, &info
, NULL
);
817 fname_from_uuid(st
, &info
, nbuf
, ':');
818 printf(" UUID : %s\n", nbuf
+ 5);
819 sum
= __le32_to_cpu(mpb
->check_sum
);
820 printf(" Checksum : %08x %s\n", sum
,
821 __gen_imsm_checksum(mpb
) == sum
? "correct" : "incorrect");
822 printf(" MPB Sectors : %d\n", mpb_sectors(mpb
));
823 printf(" Disks : %d\n", mpb
->num_disks
);
824 printf(" RAID Devices : %d\n", mpb
->num_raid_devs
);
825 print_imsm_disk(mpb
, super
->disks
->index
, reserved
);
826 if (super
->bbm_log
) {
827 struct bbm_log
*log
= super
->bbm_log
;
830 printf("Bad Block Management Log:\n");
831 printf(" Log Size : %d\n", __le32_to_cpu(mpb
->bbm_log_size
));
832 printf(" Signature : %x\n", __le32_to_cpu(log
->signature
));
833 printf(" Entry Count : %d\n", __le32_to_cpu(log
->entry_count
));
834 printf(" Spare Blocks : %d\n", __le32_to_cpu(log
->reserved_spare_block_count
));
835 printf(" First Spare : %llx\n",
836 (unsigned long long) __le64_to_cpu(log
->first_spare_lba
));
838 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
840 struct imsm_dev
*dev
= __get_imsm_dev(mpb
, i
);
842 super
->current_vol
= i
;
843 getinfo_super_imsm(st
, &info
, NULL
);
844 fname_from_uuid(st
, &info
, nbuf
, ':');
845 print_imsm_dev(dev
, nbuf
+ 5, super
->disks
->index
);
847 for (i
= 0; i
< mpb
->num_disks
; i
++) {
848 if (i
== super
->disks
->index
)
850 print_imsm_disk(mpb
, i
, reserved
);
854 static void brief_examine_super_imsm(struct supertype
*st
, int verbose
)
856 /* We just write a generic IMSM ARRAY entry */
859 struct intel_super
*super
= st
->sb
;
861 if (!super
->anchor
->num_raid_devs
) {
862 printf("ARRAY metadata=imsm\n");
866 getinfo_super_imsm(st
, &info
, NULL
);
867 fname_from_uuid(st
, &info
, nbuf
, ':');
868 printf("ARRAY metadata=imsm UUID=%s\n", nbuf
+ 5);
871 static void brief_examine_subarrays_imsm(struct supertype
*st
, int verbose
)
873 /* We just write a generic IMSM ARRAY entry */
877 struct intel_super
*super
= st
->sb
;
880 if (!super
->anchor
->num_raid_devs
)
883 getinfo_super_imsm(st
, &info
, NULL
);
884 fname_from_uuid(st
, &info
, nbuf
, ':');
885 for (i
= 0; i
< super
->anchor
->num_raid_devs
; i
++) {
886 struct imsm_dev
*dev
= get_imsm_dev(super
, i
);
888 super
->current_vol
= i
;
889 getinfo_super_imsm(st
, &info
, NULL
);
890 fname_from_uuid(st
, &info
, nbuf1
, ':');
891 printf("ARRAY /dev/md/%.16s container=%s member=%d UUID=%s\n",
892 dev
->volume
, nbuf
+ 5, i
, nbuf1
+ 5);
896 static void export_examine_super_imsm(struct supertype
*st
)
898 struct intel_super
*super
= st
->sb
;
899 struct imsm_super
*mpb
= super
->anchor
;
903 getinfo_super_imsm(st
, &info
, NULL
);
904 fname_from_uuid(st
, &info
, nbuf
, ':');
905 printf("MD_METADATA=imsm\n");
906 printf("MD_LEVEL=container\n");
907 printf("MD_UUID=%s\n", nbuf
+5);
908 printf("MD_DEVICES=%u\n", mpb
->num_disks
);
911 static void detail_super_imsm(struct supertype
*st
, char *homehost
)
916 getinfo_super_imsm(st
, &info
, NULL
);
917 fname_from_uuid(st
, &info
, nbuf
, ':');
918 printf("\n UUID : %s\n", nbuf
+ 5);
921 static void brief_detail_super_imsm(struct supertype
*st
)
925 getinfo_super_imsm(st
, &info
, NULL
);
926 fname_from_uuid(st
, &info
, nbuf
, ':');
927 printf(" UUID=%s", nbuf
+ 5);
930 static int imsm_read_serial(int fd
, char *devname
, __u8
*serial
);
931 static void fd2devname(int fd
, char *name
);
933 static int imsm_enumerate_ports(const char *hba_path
, int port_count
, int host_base
, int verbose
)
935 /* dump an unsorted list of devices attached to ahci, as well as
936 * non-connected ports
938 int hba_len
= strlen(hba_path
) + 1;
943 unsigned long port_mask
= (1 << port_count
) - 1;
945 if (port_count
> (int)sizeof(port_mask
) * 8) {
947 fprintf(stderr
, Name
": port_count %d out of range\n", port_count
);
951 /* scroll through /sys/dev/block looking for devices attached to
954 dir
= opendir("/sys/dev/block");
955 for (ent
= dir
? readdir(dir
) : NULL
; ent
; ent
= readdir(dir
)) {
966 if (sscanf(ent
->d_name
, "%d:%d", &major
, &minor
) != 2)
968 path
= devt_to_devpath(makedev(major
, minor
));
971 if (!path_attached_to_hba(path
, hba_path
)) {
977 /* retrieve the scsi device type */
978 if (asprintf(&device
, "/sys/dev/block/%d:%d/device/xxxxxxx", major
, minor
) < 0) {
980 fprintf(stderr
, Name
": failed to allocate 'device'\n");
984 sprintf(device
, "/sys/dev/block/%d:%d/device/type", major
, minor
);
985 if (load_sys(device
, buf
) != 0) {
987 fprintf(stderr
, Name
": failed to read device type for %s\n",
993 type
= strtoul(buf
, NULL
, 10);
995 /* if it's not a disk print the vendor and model */
996 if (!(type
== 0 || type
== 7 || type
== 14)) {
999 sprintf(device
, "/sys/dev/block/%d:%d/device/vendor", major
, minor
);
1000 if (load_sys(device
, buf
) == 0) {
1001 strncpy(vendor
, buf
, sizeof(vendor
));
1002 vendor
[sizeof(vendor
) - 1] = '\0';
1003 c
= (char *) &vendor
[sizeof(vendor
) - 1];
1004 while (isspace(*c
) || *c
== '\0')
1008 sprintf(device
, "/sys/dev/block/%d:%d/device/model", major
, minor
);
1009 if (load_sys(device
, buf
) == 0) {
1010 strncpy(model
, buf
, sizeof(model
));
1011 model
[sizeof(model
) - 1] = '\0';
1012 c
= (char *) &model
[sizeof(model
) - 1];
1013 while (isspace(*c
) || *c
== '\0')
1017 if (vendor
[0] && model
[0])
1018 sprintf(buf
, "%.64s %.64s", vendor
, model
);
1020 switch (type
) { /* numbers from hald/linux/device.c */
1021 case 1: sprintf(buf
, "tape"); break;
1022 case 2: sprintf(buf
, "printer"); break;
1023 case 3: sprintf(buf
, "processor"); break;
1025 case 5: sprintf(buf
, "cdrom"); break;
1026 case 6: sprintf(buf
, "scanner"); break;
1027 case 8: sprintf(buf
, "media_changer"); break;
1028 case 9: sprintf(buf
, "comm"); break;
1029 case 12: sprintf(buf
, "raid"); break;
1030 default: sprintf(buf
, "unknown");
1036 /* chop device path to 'host%d' and calculate the port number */
1037 c
= strchr(&path
[hba_len
], '/');
1040 fprintf(stderr
, Name
": %s - invalid path name\n", path
+ hba_len
);
1045 if (sscanf(&path
[hba_len
], "host%d", &port
) == 1)
1049 *c
= '/'; /* repair the full string */
1050 fprintf(stderr
, Name
": failed to determine port number for %s\n",
1057 /* mark this port as used */
1058 port_mask
&= ~(1 << port
);
1060 /* print out the device information */
1062 printf(" Port%d : - non-disk device (%s) -\n", port
, buf
);
1066 fd
= dev_open(ent
->d_name
, O_RDONLY
);
1068 printf(" Port%d : - disk info unavailable -\n", port
);
1070 fd2devname(fd
, buf
);
1071 printf(" Port%d : %s", port
, buf
);
1072 if (imsm_read_serial(fd
, NULL
, (__u8
*) buf
) == 0)
1073 printf(" (%s)\n", buf
);
1088 for (i
= 0; i
< port_count
; i
++)
1089 if (port_mask
& (1 << i
))
1090 printf(" Port%d : - no device attached -\n", i
);
1096 static int detail_platform_imsm(int verbose
, int enumerate_only
)
1098 /* There are two components to imsm platform support, the ahci SATA
1099 * controller and the option-rom. To find the SATA controller we
1100 * simply look in /sys/bus/pci/drivers/ahci to see if an ahci
1101 * controller with the Intel vendor id is present. This approach
1102 * allows mdadm to leverage the kernel's ahci detection logic, with the
1103 * caveat that if ahci.ko is not loaded mdadm will not be able to
1104 * detect platform raid capabilities. The option-rom resides in a
1105 * platform "Adapter ROM". We scan for its signature to retrieve the
1106 * platform capabilities. If raid support is disabled in the BIOS the
1107 * option-rom capability structure will not be available.
1109 const struct imsm_orom
*orom
;
1110 struct sys_dev
*list
, *hba
;
1113 const char *hba_path
;
1117 if (enumerate_only
) {
1118 if (check_env("IMSM_NO_PLATFORM") || find_imsm_orom())
1123 list
= find_driver_devices("pci", "ahci");
1124 for (hba
= list
; hba
; hba
= hba
->next
)
1125 if (devpath_to_vendor(hba
->path
) == 0x8086)
1130 fprintf(stderr
, Name
": unable to find active ahci controller\n");
1131 free_sys_dev(&list
);
1134 fprintf(stderr
, Name
": found Intel SATA AHCI Controller\n");
1135 hba_path
= hba
->path
;
1137 free_sys_dev(&list
);
1139 orom
= find_imsm_orom();
1142 fprintf(stderr
, Name
": imsm option-rom not found\n");
1146 printf(" Platform : Intel(R) Matrix Storage Manager\n");
1147 printf(" Version : %d.%d.%d.%d\n", orom
->major_ver
, orom
->minor_ver
,
1148 orom
->hotfix_ver
, orom
->build
);
1149 printf(" RAID Levels :%s%s%s%s%s\n",
1150 imsm_orom_has_raid0(orom
) ? " raid0" : "",
1151 imsm_orom_has_raid1(orom
) ? " raid1" : "",
1152 imsm_orom_has_raid1e(orom
) ? " raid1e" : "",
1153 imsm_orom_has_raid10(orom
) ? " raid10" : "",
1154 imsm_orom_has_raid5(orom
) ? " raid5" : "");
1155 printf(" Chunk Sizes :%s%s%s%s%s%s%s%s%s%s%s%s%s%s%s%s\n",
1156 imsm_orom_has_chunk(orom
, 2) ? " 2k" : "",
1157 imsm_orom_has_chunk(orom
, 4) ? " 4k" : "",
1158 imsm_orom_has_chunk(orom
, 8) ? " 8k" : "",
1159 imsm_orom_has_chunk(orom
, 16) ? " 16k" : "",
1160 imsm_orom_has_chunk(orom
, 32) ? " 32k" : "",
1161 imsm_orom_has_chunk(orom
, 64) ? " 64k" : "",
1162 imsm_orom_has_chunk(orom
, 128) ? " 128k" : "",
1163 imsm_orom_has_chunk(orom
, 256) ? " 256k" : "",
1164 imsm_orom_has_chunk(orom
, 512) ? " 512k" : "",
1165 imsm_orom_has_chunk(orom
, 1024*1) ? " 1M" : "",
1166 imsm_orom_has_chunk(orom
, 1024*2) ? " 2M" : "",
1167 imsm_orom_has_chunk(orom
, 1024*4) ? " 4M" : "",
1168 imsm_orom_has_chunk(orom
, 1024*8) ? " 8M" : "",
1169 imsm_orom_has_chunk(orom
, 1024*16) ? " 16M" : "",
1170 imsm_orom_has_chunk(orom
, 1024*32) ? " 32M" : "",
1171 imsm_orom_has_chunk(orom
, 1024*64) ? " 64M" : "");
1172 printf(" Max Disks : %d\n", orom
->tds
);
1173 printf(" Max Volumes : %d\n", orom
->vpa
);
1174 printf(" I/O Controller : %s\n", hba_path
);
1176 /* find the smallest scsi host number to determine a port number base */
1177 dir
= opendir(hba_path
);
1178 for (ent
= dir
? readdir(dir
) : NULL
; ent
; ent
= readdir(dir
)) {
1181 if (sscanf(ent
->d_name
, "host%d", &host
) != 1)
1183 if (port_count
== 0)
1185 else if (host
< host_base
)
1188 if (host
+ 1 > port_count
+ host_base
)
1189 port_count
= host
+ 1 - host_base
;
1195 if (!port_count
|| imsm_enumerate_ports(hba_path
, port_count
,
1196 host_base
, verbose
) != 0) {
1198 fprintf(stderr
, Name
": failed to enumerate ports\n");
1206 static int match_home_imsm(struct supertype
*st
, char *homehost
)
1208 /* the imsm metadata format does not specify any host
1209 * identification information. We return -1 since we can never
1210 * confirm nor deny whether a given array is "meant" for this
1211 * host. We rely on compare_super and the 'family_num' fields to
1212 * exclude member disks that do not belong, and we rely on
1213 * mdadm.conf to specify the arrays that should be assembled.
1214 * Auto-assembly may still pick up "foreign" arrays.
1220 static void uuid_from_super_imsm(struct supertype
*st
, int uuid
[4])
1222 /* The uuid returned here is used for:
1223 * uuid to put into bitmap file (Create, Grow)
1224 * uuid for backup header when saving critical section (Grow)
1225 * comparing uuids when re-adding a device into an array
1226 * In these cases the uuid required is that of the data-array,
1227 * not the device-set.
1228 * uuid to recognise same set when adding a missing device back
1229 * to an array. This is a uuid for the device-set.
1231 * For each of these we can make do with a truncated
1232 * or hashed uuid rather than the original, as long as
1234 * In each case the uuid required is that of the data-array,
1235 * not the device-set.
1237 /* imsm does not track uuid's so we synthesis one using sha1 on
1238 * - The signature (Which is constant for all imsm array, but no matter)
1239 * - the orig_family_num of the container
1240 * - the index number of the volume
1241 * - the 'serial' number of the volume.
1242 * Hopefully these are all constant.
1244 struct intel_super
*super
= st
->sb
;
1247 struct sha1_ctx ctx
;
1248 struct imsm_dev
*dev
= NULL
;
1251 /* some mdadm versions failed to set ->orig_family_num, in which
1252 * case fall back to ->family_num. orig_family_num will be
1253 * fixed up with the first metadata update.
1255 family_num
= super
->anchor
->orig_family_num
;
1256 if (family_num
== 0)
1257 family_num
= super
->anchor
->family_num
;
1258 sha1_init_ctx(&ctx
);
1259 sha1_process_bytes(super
->anchor
->sig
, MPB_SIG_LEN
, &ctx
);
1260 sha1_process_bytes(&family_num
, sizeof(__u32
), &ctx
);
1261 if (super
->current_vol
>= 0)
1262 dev
= get_imsm_dev(super
, super
->current_vol
);
1264 __u32 vol
= super
->current_vol
;
1265 sha1_process_bytes(&vol
, sizeof(vol
), &ctx
);
1266 sha1_process_bytes(dev
->volume
, MAX_RAID_SERIAL_LEN
, &ctx
);
1268 sha1_finish_ctx(&ctx
, buf
);
1269 memcpy(uuid
, buf
, 4*4);
1274 get_imsm_numerical_version(struct imsm_super
*mpb
, int *m
, int *p
)
1276 __u8
*v
= get_imsm_version(mpb
);
1277 __u8
*end
= mpb
->sig
+ MAX_SIGNATURE_LENGTH
;
1278 char major
[] = { 0, 0, 0 };
1279 char minor
[] = { 0 ,0, 0 };
1280 char patch
[] = { 0, 0, 0 };
1281 char *ver_parse
[] = { major
, minor
, patch
};
1285 while (*v
!= '\0' && v
< end
) {
1286 if (*v
!= '.' && j
< 2)
1287 ver_parse
[i
][j
++] = *v
;
1295 *m
= strtol(minor
, NULL
, 0);
1296 *p
= strtol(patch
, NULL
, 0);
1300 static __u32
migr_strip_blocks_resync(struct imsm_dev
*dev
)
1302 /* migr_strip_size when repairing or initializing parity */
1303 struct imsm_map
*map
= get_imsm_map(dev
, 0);
1304 __u32 chunk
= __le32_to_cpu(map
->blocks_per_strip
);
1306 switch (get_imsm_raid_level(map
)) {
1311 return 128*1024 >> 9;
1315 static __u32
migr_strip_blocks_rebuild(struct imsm_dev
*dev
)
1317 /* migr_strip_size when rebuilding a degraded disk, no idea why
1318 * this is different than migr_strip_size_resync(), but it's good
1321 struct imsm_map
*map
= get_imsm_map(dev
, 1);
1322 __u32 chunk
= __le32_to_cpu(map
->blocks_per_strip
);
1324 switch (get_imsm_raid_level(map
)) {
1327 if (map
->num_members
% map
->num_domains
== 0)
1328 return 128*1024 >> 9;
1332 return max((__u32
) 64*1024 >> 9, chunk
);
1334 return 128*1024 >> 9;
1338 static __u32
num_stripes_per_unit_resync(struct imsm_dev
*dev
)
1340 struct imsm_map
*lo
= get_imsm_map(dev
, 0);
1341 struct imsm_map
*hi
= get_imsm_map(dev
, 1);
1342 __u32 lo_chunk
= __le32_to_cpu(lo
->blocks_per_strip
);
1343 __u32 hi_chunk
= __le32_to_cpu(hi
->blocks_per_strip
);
1345 return max((__u32
) 1, hi_chunk
/ lo_chunk
);
1348 static __u32
num_stripes_per_unit_rebuild(struct imsm_dev
*dev
)
1350 struct imsm_map
*lo
= get_imsm_map(dev
, 0);
1351 int level
= get_imsm_raid_level(lo
);
1353 if (level
== 1 || level
== 10) {
1354 struct imsm_map
*hi
= get_imsm_map(dev
, 1);
1356 return hi
->num_domains
;
1358 return num_stripes_per_unit_resync(dev
);
1361 static __u8
imsm_num_data_members(struct imsm_dev
*dev
)
1363 /* named 'imsm_' because raid0, raid1 and raid10
1364 * counter-intuitively have the same number of data disks
1366 struct imsm_map
*map
= get_imsm_map(dev
, 0);
1368 switch (get_imsm_raid_level(map
)) {
1372 return map
->num_members
;
1374 return map
->num_members
- 1;
1376 dprintf("%s: unsupported raid level\n", __func__
);
1381 static __u32
parity_segment_depth(struct imsm_dev
*dev
)
1383 struct imsm_map
*map
= get_imsm_map(dev
, 0);
1384 __u32 chunk
= __le32_to_cpu(map
->blocks_per_strip
);
1386 switch(get_imsm_raid_level(map
)) {
1389 return chunk
* map
->num_domains
;
1391 return chunk
* map
->num_members
;
1397 static __u32
map_migr_block(struct imsm_dev
*dev
, __u32 block
)
1399 struct imsm_map
*map
= get_imsm_map(dev
, 1);
1400 __u32 chunk
= __le32_to_cpu(map
->blocks_per_strip
);
1401 __u32 strip
= block
/ chunk
;
1403 switch (get_imsm_raid_level(map
)) {
1406 __u32 vol_strip
= (strip
* map
->num_domains
) + 1;
1407 __u32 vol_stripe
= vol_strip
/ map
->num_members
;
1409 return vol_stripe
* chunk
+ block
% chunk
;
1411 __u32 stripe
= strip
/ (map
->num_members
- 1);
1413 return stripe
* chunk
+ block
% chunk
;
1420 static __u64
blocks_per_migr_unit(struct imsm_dev
*dev
)
1422 /* calculate the conversion factor between per member 'blocks'
1423 * (md/{resync,rebuild}_start) and imsm migration units, return
1424 * 0 for the 'not migrating' and 'unsupported migration' cases
1426 if (!dev
->vol
.migr_state
)
1429 switch (migr_type(dev
)) {
1433 struct imsm_map
*map
= get_imsm_map(dev
, 0);
1434 __u32 stripes_per_unit
;
1435 __u32 blocks_per_unit
;
1444 /* yes, this is really the translation of migr_units to
1445 * per-member blocks in the 'resync' case
1447 stripes_per_unit
= num_stripes_per_unit_resync(dev
);
1448 migr_chunk
= migr_strip_blocks_resync(dev
);
1449 disks
= imsm_num_data_members(dev
);
1450 blocks_per_unit
= stripes_per_unit
* migr_chunk
* disks
;
1451 stripe
= __le32_to_cpu(map
->blocks_per_strip
) * disks
;
1452 segment
= blocks_per_unit
/ stripe
;
1453 block_rel
= blocks_per_unit
- segment
* stripe
;
1454 parity_depth
= parity_segment_depth(dev
);
1455 block_map
= map_migr_block(dev
, block_rel
);
1456 return block_map
+ parity_depth
* segment
;
1458 case MIGR_REBUILD
: {
1459 __u32 stripes_per_unit
;
1462 stripes_per_unit
= num_stripes_per_unit_rebuild(dev
);
1463 migr_chunk
= migr_strip_blocks_rebuild(dev
);
1464 return migr_chunk
* stripes_per_unit
;
1467 case MIGR_STATE_CHANGE
:
1473 static int imsm_level_to_layout(int level
)
1481 return ALGORITHM_LEFT_ASYMMETRIC
;
1488 static void getinfo_super_imsm_volume(struct supertype
*st
, struct mdinfo
*info
, char *dmap
)
1490 struct intel_super
*super
= st
->sb
;
1491 struct imsm_dev
*dev
= get_imsm_dev(super
, super
->current_vol
);
1492 struct imsm_map
*map
= get_imsm_map(dev
, 0);
1495 int map_disks
= info
->array
.raid_disks
;
1497 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
1498 if (dl
->raiddisk
== info
->disk
.raid_disk
)
1500 info
->container_member
= super
->current_vol
;
1501 info
->array
.raid_disks
= map
->num_members
;
1502 info
->array
.level
= get_imsm_raid_level(map
);
1503 info
->array
.layout
= imsm_level_to_layout(info
->array
.level
);
1504 info
->array
.md_minor
= -1;
1505 info
->array
.ctime
= 0;
1506 info
->array
.utime
= 0;
1507 info
->array
.chunk_size
= __le16_to_cpu(map
->blocks_per_strip
) << 9;
1508 info
->array
.state
= !dev
->vol
.dirty
;
1509 info
->custom_array_size
= __le32_to_cpu(dev
->size_high
);
1510 info
->custom_array_size
<<= 32;
1511 info
->custom_array_size
|= __le32_to_cpu(dev
->size_low
);
1513 info
->disk
.major
= 0;
1514 info
->disk
.minor
= 0;
1516 info
->disk
.major
= dl
->major
;
1517 info
->disk
.minor
= dl
->minor
;
1520 info
->data_offset
= __le32_to_cpu(map
->pba_of_lba0
);
1521 info
->component_size
= __le32_to_cpu(map
->blocks_per_member
);
1522 memset(info
->uuid
, 0, sizeof(info
->uuid
));
1523 info
->recovery_start
= MaxSector
;
1524 info
->reshape_active
= 0;
1526 if (map
->map_state
== IMSM_T_STATE_UNINITIALIZED
|| dev
->vol
.dirty
) {
1527 info
->resync_start
= 0;
1528 } else if (dev
->vol
.migr_state
) {
1529 switch (migr_type(dev
)) {
1532 __u64 blocks_per_unit
= blocks_per_migr_unit(dev
);
1533 __u64 units
= __le32_to_cpu(dev
->vol
.curr_migr_unit
);
1535 info
->resync_start
= blocks_per_unit
* units
;
1539 /* we could emulate the checkpointing of
1540 * 'sync_action=check' migrations, but for now
1541 * we just immediately complete them
1544 /* this is handled by container_content_imsm() */
1546 case MIGR_STATE_CHANGE
:
1547 /* FIXME handle other migrations */
1549 /* we are not dirty, so... */
1550 info
->resync_start
= MaxSector
;
1553 info
->resync_start
= MaxSector
;
1555 strncpy(info
->name
, (char *) dev
->volume
, MAX_RAID_SERIAL_LEN
);
1556 info
->name
[MAX_RAID_SERIAL_LEN
] = 0;
1558 info
->array
.major_version
= -1;
1559 info
->array
.minor_version
= -2;
1560 devname
= devnum2devname(st
->container_dev
);
1561 *info
->text_version
= '\0';
1563 sprintf(info
->text_version
, "/%s/%d", devname
, info
->container_member
);
1565 info
->safe_mode_delay
= 4000; /* 4 secs like the Matrix driver */
1566 uuid_from_super_imsm(st
, info
->uuid
);
1570 for (i
=0; i
<map_disks
; i
++) {
1572 if (i
< info
->array
.raid_disks
) {
1573 struct imsm_disk
*dsk
;
1574 j
= get_imsm_disk_idx(dev
, i
);
1575 dsk
= get_imsm_disk(super
, j
);
1576 if (dsk
&& (dsk
->status
& CONFIGURED_DISK
))
1583 /* check the config file to see if we can return a real uuid for this spare */
1584 static void fixup_container_spare_uuid(struct mdinfo
*inf
)
1586 struct mddev_ident
*array_list
;
1588 if (inf
->array
.level
!= LEVEL_CONTAINER
||
1589 memcmp(inf
->uuid
, uuid_match_any
, sizeof(int[4])) != 0)
1592 array_list
= conf_get_ident(NULL
);
1594 for (; array_list
; array_list
= array_list
->next
) {
1595 if (array_list
->uuid_set
) {
1596 struct supertype
*_sst
; /* spare supertype */
1597 struct supertype
*_cst
; /* container supertype */
1599 _cst
= array_list
->st
;
1601 _sst
= _cst
->ss
->match_metadata_desc(inf
->text_version
);
1606 memcpy(inf
->uuid
, array_list
->uuid
, sizeof(int[4]));
1615 static __u8
imsm_check_degraded(struct intel_super
*super
, struct imsm_dev
*dev
, int failed
);
1616 static int imsm_count_failed(struct intel_super
*super
, struct imsm_dev
*dev
);
1618 static struct imsm_disk
*get_imsm_missing(struct intel_super
*super
, __u8 index
)
1622 for (d
= super
->missing
; d
; d
= d
->next
)
1623 if (d
->index
== index
)
1628 static void getinfo_super_imsm(struct supertype
*st
, struct mdinfo
*info
, char *map
)
1630 struct intel_super
*super
= st
->sb
;
1631 struct imsm_disk
*disk
;
1632 int map_disks
= info
->array
.raid_disks
;
1633 int max_enough
= -1;
1635 struct imsm_super
*mpb
;
1637 if (super
->current_vol
>= 0) {
1638 getinfo_super_imsm_volume(st
, info
, map
);
1642 /* Set raid_disks to zero so that Assemble will always pull in valid
1645 info
->array
.raid_disks
= 0;
1646 info
->array
.level
= LEVEL_CONTAINER
;
1647 info
->array
.layout
= 0;
1648 info
->array
.md_minor
= -1;
1649 info
->array
.ctime
= 0; /* N/A for imsm */
1650 info
->array
.utime
= 0;
1651 info
->array
.chunk_size
= 0;
1653 info
->disk
.major
= 0;
1654 info
->disk
.minor
= 0;
1655 info
->disk
.raid_disk
= -1;
1656 info
->reshape_active
= 0;
1657 info
->array
.major_version
= -1;
1658 info
->array
.minor_version
= -2;
1659 strcpy(info
->text_version
, "imsm");
1660 info
->safe_mode_delay
= 0;
1661 info
->disk
.number
= -1;
1662 info
->disk
.state
= 0;
1664 info
->recovery_start
= MaxSector
;
1666 /* do we have the all the insync disks that we expect? */
1667 mpb
= super
->anchor
;
1669 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
1670 struct imsm_dev
*dev
= get_imsm_dev(super
, i
);
1671 int failed
, enough
, j
, missing
= 0;
1672 struct imsm_map
*map
;
1675 failed
= imsm_count_failed(super
, dev
);
1676 state
= imsm_check_degraded(super
, dev
, failed
);
1677 map
= get_imsm_map(dev
, dev
->vol
.migr_state
);
1679 /* any newly missing disks?
1680 * (catches single-degraded vs double-degraded)
1682 for (j
= 0; j
< map
->num_members
; j
++) {
1683 __u32 ord
= get_imsm_ord_tbl_ent(dev
, i
);
1684 __u32 idx
= ord_to_idx(ord
);
1686 if (!(ord
& IMSM_ORD_REBUILD
) &&
1687 get_imsm_missing(super
, idx
)) {
1693 if (state
== IMSM_T_STATE_FAILED
)
1695 else if (state
== IMSM_T_STATE_DEGRADED
&&
1696 (state
!= map
->map_state
|| missing
))
1698 else /* we're normal, or already degraded */
1701 /* in the missing/failed disk case check to see
1702 * if at least one array is runnable
1704 max_enough
= max(max_enough
, enough
);
1706 dprintf("%s: enough: %d\n", __func__
, max_enough
);
1707 info
->container_enough
= max_enough
;
1710 __u32 reserved
= imsm_reserved_sectors(super
, super
->disks
);
1712 disk
= &super
->disks
->disk
;
1713 info
->data_offset
= __le32_to_cpu(disk
->total_blocks
) - reserved
;
1714 info
->component_size
= reserved
;
1715 info
->disk
.state
= is_configured(disk
) ? (1 << MD_DISK_ACTIVE
) : 0;
1716 /* we don't change info->disk.raid_disk here because
1717 * this state will be finalized in mdmon after we have
1718 * found the 'most fresh' version of the metadata
1720 info
->disk
.state
|= is_failed(disk
) ? (1 << MD_DISK_FAULTY
) : 0;
1721 info
->disk
.state
|= is_spare(disk
) ? 0 : (1 << MD_DISK_SYNC
);
1724 /* only call uuid_from_super_imsm when this disk is part of a populated container,
1725 * ->compare_super may have updated the 'num_raid_devs' field for spares
1727 if (info
->disk
.state
& (1 << MD_DISK_SYNC
) || super
->anchor
->num_raid_devs
)
1728 uuid_from_super_imsm(st
, info
->uuid
);
1730 memcpy(info
->uuid
, uuid_match_any
, sizeof(int[4]));
1731 fixup_container_spare_uuid(info
);
1734 /* I don't know how to compute 'map' on imsm, so use safe default */
1737 for (i
= 0; i
< map_disks
; i
++)
1743 /* allocates memory and fills disk in mdinfo structure
1744 * for each disk in array */
1745 struct mdinfo
*getinfo_super_disks_imsm(struct supertype
*st
)
1747 struct mdinfo
*mddev
= NULL
;
1748 struct intel_super
*super
= st
->sb
;
1749 struct imsm_disk
*disk
;
1752 if (!super
|| !super
->disks
)
1755 mddev
= malloc(sizeof(*mddev
));
1757 fprintf(stderr
, Name
": Failed to allocate memory.\n");
1760 memset(mddev
, 0, sizeof(*mddev
));
1764 tmp
= malloc(sizeof(*tmp
));
1766 fprintf(stderr
, Name
": Failed to allocate memory.\n");
1771 memset(tmp
, 0, sizeof(*tmp
));
1773 tmp
->next
= mddev
->devs
;
1775 tmp
->disk
.number
= count
++;
1776 tmp
->disk
.major
= dl
->major
;
1777 tmp
->disk
.minor
= dl
->minor
;
1778 tmp
->disk
.state
= is_configured(disk
) ?
1779 (1 << MD_DISK_ACTIVE
) : 0;
1780 tmp
->disk
.state
|= is_failed(disk
) ? (1 << MD_DISK_FAULTY
) : 0;
1781 tmp
->disk
.state
|= is_spare(disk
) ? 0 : (1 << MD_DISK_SYNC
);
1782 tmp
->disk
.raid_disk
= -1;
1788 static int update_super_imsm(struct supertype
*st
, struct mdinfo
*info
,
1789 char *update
, char *devname
, int verbose
,
1790 int uuid_set
, char *homehost
)
1792 /* For 'assemble' and 'force' we need to return non-zero if any
1793 * change was made. For others, the return value is ignored.
1794 * Update options are:
1795 * force-one : This device looks a bit old but needs to be included,
1796 * update age info appropriately.
1797 * assemble: clear any 'faulty' flag to allow this device to
1799 * force-array: Array is degraded but being forced, mark it clean
1800 * if that will be needed to assemble it.
1802 * newdev: not used ????
1803 * grow: Array has gained a new device - this is currently for
1805 * resync: mark as dirty so a resync will happen.
1806 * name: update the name - preserving the homehost
1807 * uuid: Change the uuid of the array to match watch is given
1809 * Following are not relevant for this imsm:
1810 * sparc2.2 : update from old dodgey metadata
1811 * super-minor: change the preferred_minor number
1812 * summaries: update redundant counters.
1813 * homehost: update the recorded homehost
1814 * _reshape_progress: record new reshape_progress position.
1817 struct intel_super
*super
= st
->sb
;
1818 struct imsm_super
*mpb
;
1820 /* we can only update container info */
1821 if (!super
|| super
->current_vol
>= 0 || !super
->anchor
)
1824 mpb
= super
->anchor
;
1826 if (strcmp(update
, "uuid") == 0 && uuid_set
&& !info
->update_private
)
1828 else if (strcmp(update
, "uuid") == 0 && uuid_set
&& info
->update_private
) {
1829 mpb
->orig_family_num
= *((__u32
*) info
->update_private
);
1831 } else if (strcmp(update
, "uuid") == 0) {
1832 __u32
*new_family
= malloc(sizeof(*new_family
));
1834 /* update orig_family_number with the incoming random
1835 * data, report the new effective uuid, and store the
1836 * new orig_family_num for future updates.
1839 memcpy(&mpb
->orig_family_num
, info
->uuid
, sizeof(__u32
));
1840 uuid_from_super_imsm(st
, info
->uuid
);
1841 *new_family
= mpb
->orig_family_num
;
1842 info
->update_private
= new_family
;
1845 } else if (strcmp(update
, "assemble") == 0)
1850 /* successful update? recompute checksum */
1852 mpb
->check_sum
= __le32_to_cpu(__gen_imsm_checksum(mpb
));
1857 static size_t disks_to_mpb_size(int disks
)
1861 size
= sizeof(struct imsm_super
);
1862 size
+= (disks
- 1) * sizeof(struct imsm_disk
);
1863 size
+= 2 * sizeof(struct imsm_dev
);
1864 /* up to 2 maps per raid device (-2 for imsm_maps in imsm_dev */
1865 size
+= (4 - 2) * sizeof(struct imsm_map
);
1866 /* 4 possible disk_ord_tbl's */
1867 size
+= 4 * (disks
- 1) * sizeof(__u32
);
1872 static __u64
avail_size_imsm(struct supertype
*st
, __u64 devsize
)
1874 if (devsize
< (MPB_SECTOR_CNT
+ IMSM_RESERVED_SECTORS
))
1877 return devsize
- (MPB_SECTOR_CNT
+ IMSM_RESERVED_SECTORS
);
1880 static void free_devlist(struct intel_super
*super
)
1882 struct intel_dev
*dv
;
1884 while (super
->devlist
) {
1885 dv
= super
->devlist
->next
;
1886 free(super
->devlist
->dev
);
1887 free(super
->devlist
);
1888 super
->devlist
= dv
;
1892 static void imsm_copy_dev(struct imsm_dev
*dest
, struct imsm_dev
*src
)
1894 memcpy(dest
, src
, sizeof_imsm_dev(src
, 0));
1897 static int compare_super_imsm(struct supertype
*st
, struct supertype
*tst
)
1901 * 0 same, or first was empty, and second was copied
1902 * 1 second had wrong number
1904 * 3 wrong other info
1906 struct intel_super
*first
= st
->sb
;
1907 struct intel_super
*sec
= tst
->sb
;
1915 /* if an anchor does not have num_raid_devs set then it is a free
1918 if (first
->anchor
->num_raid_devs
> 0 &&
1919 sec
->anchor
->num_raid_devs
> 0) {
1920 /* Determine if these disks might ever have been
1921 * related. Further disambiguation can only take place
1922 * in load_super_imsm_all
1924 __u32 first_family
= first
->anchor
->orig_family_num
;
1925 __u32 sec_family
= sec
->anchor
->orig_family_num
;
1927 if (memcmp(first
->anchor
->sig
, sec
->anchor
->sig
,
1928 MAX_SIGNATURE_LENGTH
) != 0)
1931 if (first_family
== 0)
1932 first_family
= first
->anchor
->family_num
;
1933 if (sec_family
== 0)
1934 sec_family
= sec
->anchor
->family_num
;
1936 if (first_family
!= sec_family
)
1942 /* if 'first' is a spare promote it to a populated mpb with sec's
1945 if (first
->anchor
->num_raid_devs
== 0 &&
1946 sec
->anchor
->num_raid_devs
> 0) {
1948 struct intel_dev
*dv
;
1949 struct imsm_dev
*dev
;
1951 /* we need to copy raid device info from sec if an allocation
1952 * fails here we don't associate the spare
1954 for (i
= 0; i
< sec
->anchor
->num_raid_devs
; i
++) {
1955 dv
= malloc(sizeof(*dv
));
1958 dev
= malloc(sizeof_imsm_dev(get_imsm_dev(sec
, i
), 1));
1965 dv
->next
= first
->devlist
;
1966 first
->devlist
= dv
;
1968 if (i
< sec
->anchor
->num_raid_devs
) {
1969 /* allocation failure */
1970 free_devlist(first
);
1971 fprintf(stderr
, "imsm: failed to associate spare\n");
1974 first
->anchor
->num_raid_devs
= sec
->anchor
->num_raid_devs
;
1975 first
->anchor
->orig_family_num
= sec
->anchor
->orig_family_num
;
1976 first
->anchor
->family_num
= sec
->anchor
->family_num
;
1977 memcpy(first
->anchor
->sig
, sec
->anchor
->sig
, MAX_SIGNATURE_LENGTH
);
1978 for (i
= 0; i
< sec
->anchor
->num_raid_devs
; i
++)
1979 imsm_copy_dev(get_imsm_dev(first
, i
), get_imsm_dev(sec
, i
));
1985 static void fd2devname(int fd
, char *name
)
1989 char dname
[PATH_MAX
];
1994 if (fstat(fd
, &st
) != 0)
1996 sprintf(path
, "/sys/dev/block/%d:%d",
1997 major(st
.st_rdev
), minor(st
.st_rdev
));
1999 rv
= readlink(path
, dname
, sizeof(dname
));
2004 nm
= strrchr(dname
, '/');
2006 snprintf(name
, MAX_RAID_SERIAL_LEN
, "/dev/%s", nm
);
2009 extern int scsi_get_serial(int fd
, void *buf
, size_t buf_len
);
2011 static int imsm_read_serial(int fd
, char *devname
,
2012 __u8 serial
[MAX_RAID_SERIAL_LEN
])
2014 unsigned char scsi_serial
[255];
2023 memset(scsi_serial
, 0, sizeof(scsi_serial
));
2025 rv
= scsi_get_serial(fd
, scsi_serial
, sizeof(scsi_serial
));
2027 if (rv
&& check_env("IMSM_DEVNAME_AS_SERIAL")) {
2028 memset(serial
, 0, MAX_RAID_SERIAL_LEN
);
2029 fd2devname(fd
, (char *) serial
);
2036 Name
": Failed to retrieve serial for %s\n",
2041 rsp_len
= scsi_serial
[3];
2045 Name
": Failed to retrieve serial for %s\n",
2049 rsp_buf
= (char *) &scsi_serial
[4];
2051 /* trim all whitespace and non-printable characters and convert
2054 for (i
= 0, dest
= rsp_buf
; i
< rsp_len
; i
++) {
2057 /* ':' is reserved for use in placeholder serial
2058 * numbers for missing disks
2066 len
= dest
- rsp_buf
;
2069 /* truncate leading characters */
2070 if (len
> MAX_RAID_SERIAL_LEN
) {
2071 dest
+= len
- MAX_RAID_SERIAL_LEN
;
2072 len
= MAX_RAID_SERIAL_LEN
;
2075 memset(serial
, 0, MAX_RAID_SERIAL_LEN
);
2076 memcpy(serial
, dest
, len
);
2081 static int serialcmp(__u8
*s1
, __u8
*s2
)
2083 return strncmp((char *) s1
, (char *) s2
, MAX_RAID_SERIAL_LEN
);
2086 static void serialcpy(__u8
*dest
, __u8
*src
)
2088 strncpy((char *) dest
, (char *) src
, MAX_RAID_SERIAL_LEN
);
2092 static struct dl
*serial_to_dl(__u8
*serial
, struct intel_super
*super
)
2096 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
2097 if (serialcmp(dl
->serial
, serial
) == 0)
2104 static struct imsm_disk
*
2105 __serial_to_disk(__u8
*serial
, struct imsm_super
*mpb
, int *idx
)
2109 for (i
= 0; i
< mpb
->num_disks
; i
++) {
2110 struct imsm_disk
*disk
= __get_imsm_disk(mpb
, i
);
2112 if (serialcmp(disk
->serial
, serial
) == 0) {
2123 load_imsm_disk(int fd
, struct intel_super
*super
, char *devname
, int keep_fd
)
2125 struct imsm_disk
*disk
;
2130 __u8 serial
[MAX_RAID_SERIAL_LEN
];
2132 rv
= imsm_read_serial(fd
, devname
, serial
);
2137 dl
= calloc(1, sizeof(*dl
));
2141 Name
": failed to allocate disk buffer for %s\n",
2147 dl
->major
= major(stb
.st_rdev
);
2148 dl
->minor
= minor(stb
.st_rdev
);
2149 dl
->next
= super
->disks
;
2150 dl
->fd
= keep_fd
? fd
: -1;
2151 assert(super
->disks
== NULL
);
2153 serialcpy(dl
->serial
, serial
);
2156 fd2devname(fd
, name
);
2158 dl
->devname
= strdup(devname
);
2160 dl
->devname
= strdup(name
);
2162 /* look up this disk's index in the current anchor */
2163 disk
= __serial_to_disk(dl
->serial
, super
->anchor
, &dl
->index
);
2166 /* only set index on disks that are a member of a
2167 * populated contianer, i.e. one with raid_devs
2169 if (is_failed(&dl
->disk
))
2171 else if (is_spare(&dl
->disk
))
2179 /* When migrating map0 contains the 'destination' state while map1
2180 * contains the current state. When not migrating map0 contains the
2181 * current state. This routine assumes that map[0].map_state is set to
2182 * the current array state before being called.
2184 * Migration is indicated by one of the following states
2185 * 1/ Idle (migr_state=0 map0state=normal||unitialized||degraded||failed)
2186 * 2/ Initialize (migr_state=1 migr_type=MIGR_INIT map0state=normal
2187 * map1state=unitialized)
2188 * 3/ Repair (Resync) (migr_state=1 migr_type=MIGR_REPAIR map0state=normal
2190 * 4/ Rebuild (migr_state=1 migr_type=MIGR_REBUILD map0state=normal
2191 * map1state=degraded)
2193 static void migrate(struct imsm_dev
*dev
, __u8 to_state
, int migr_type
)
2195 struct imsm_map
*dest
;
2196 struct imsm_map
*src
= get_imsm_map(dev
, 0);
2198 dev
->vol
.migr_state
= 1;
2199 set_migr_type(dev
, migr_type
);
2200 dev
->vol
.curr_migr_unit
= 0;
2201 dest
= get_imsm_map(dev
, 1);
2203 /* duplicate and then set the target end state in map[0] */
2204 memcpy(dest
, src
, sizeof_imsm_map(src
));
2205 if ((migr_type
== MIGR_REBUILD
) ||
2206 (migr_type
== MIGR_GEN_MIGR
)) {
2210 for (i
= 0; i
< src
->num_members
; i
++) {
2211 ord
= __le32_to_cpu(src
->disk_ord_tbl
[i
]);
2212 set_imsm_ord_tbl_ent(src
, i
, ord_to_idx(ord
));
2216 src
->map_state
= to_state
;
2219 static void end_migration(struct imsm_dev
*dev
, __u8 map_state
)
2221 struct imsm_map
*map
= get_imsm_map(dev
, 0);
2222 struct imsm_map
*prev
= get_imsm_map(dev
, dev
->vol
.migr_state
);
2225 /* merge any IMSM_ORD_REBUILD bits that were not successfully
2226 * completed in the last migration.
2228 * FIXME add support for raid-level-migration
2230 for (i
= 0; i
< prev
->num_members
; i
++)
2231 for (j
= 0; j
< map
->num_members
; j
++)
2232 /* during online capacity expansion
2233 * disks position can be changed if takeover is used
2235 if (ord_to_idx(map
->disk_ord_tbl
[j
]) ==
2236 ord_to_idx(prev
->disk_ord_tbl
[i
])) {
2237 map
->disk_ord_tbl
[j
] |= prev
->disk_ord_tbl
[i
];
2241 dev
->vol
.migr_state
= 0;
2242 dev
->vol
.migr_type
= 0;
2243 dev
->vol
.curr_migr_unit
= 0;
2244 map
->map_state
= map_state
;
2248 static int parse_raid_devices(struct intel_super
*super
)
2251 struct imsm_dev
*dev_new
;
2252 size_t len
, len_migr
;
2253 size_t space_needed
= 0;
2254 struct imsm_super
*mpb
= super
->anchor
;
2256 for (i
= 0; i
< super
->anchor
->num_raid_devs
; i
++) {
2257 struct imsm_dev
*dev_iter
= __get_imsm_dev(super
->anchor
, i
);
2258 struct intel_dev
*dv
;
2260 len
= sizeof_imsm_dev(dev_iter
, 0);
2261 len_migr
= sizeof_imsm_dev(dev_iter
, 1);
2263 space_needed
+= len_migr
- len
;
2265 dv
= malloc(sizeof(*dv
));
2268 dev_new
= malloc(len_migr
);
2273 imsm_copy_dev(dev_new
, dev_iter
);
2276 dv
->next
= super
->devlist
;
2277 super
->devlist
= dv
;
2280 /* ensure that super->buf is large enough when all raid devices
2283 if (__le32_to_cpu(mpb
->mpb_size
) + space_needed
> super
->len
) {
2286 len
= ROUND_UP(__le32_to_cpu(mpb
->mpb_size
) + space_needed
, 512);
2287 if (posix_memalign(&buf
, 512, len
) != 0)
2290 memcpy(buf
, super
->buf
, super
->len
);
2291 memset(buf
+ super
->len
, 0, len
- super
->len
);
2300 /* retrieve a pointer to the bbm log which starts after all raid devices */
2301 struct bbm_log
*__get_imsm_bbm_log(struct imsm_super
*mpb
)
2305 if (__le32_to_cpu(mpb
->bbm_log_size
)) {
2307 ptr
+= mpb
->mpb_size
- __le32_to_cpu(mpb
->bbm_log_size
);
2313 static void __free_imsm(struct intel_super
*super
, int free_disks
);
2315 /* load_imsm_mpb - read matrix metadata
2316 * allocates super->mpb to be freed by free_super
2318 static int load_imsm_mpb(int fd
, struct intel_super
*super
, char *devname
)
2320 unsigned long long dsize
;
2321 unsigned long long sectors
;
2323 struct imsm_super
*anchor
;
2326 get_dev_size(fd
, NULL
, &dsize
);
2330 Name
": %s: device to small for imsm\n",
2335 if (lseek64(fd
, dsize
- (512 * 2), SEEK_SET
) < 0) {
2338 Name
": Cannot seek to anchor block on %s: %s\n",
2339 devname
, strerror(errno
));
2343 if (posix_memalign((void**)&anchor
, 512, 512) != 0) {
2346 Name
": Failed to allocate imsm anchor buffer"
2347 " on %s\n", devname
);
2350 if (read(fd
, anchor
, 512) != 512) {
2353 Name
": Cannot read anchor block on %s: %s\n",
2354 devname
, strerror(errno
));
2359 if (strncmp((char *) anchor
->sig
, MPB_SIGNATURE
, MPB_SIG_LEN
) != 0) {
2362 Name
": no IMSM anchor on %s\n", devname
);
2367 __free_imsm(super
, 0);
2368 super
->len
= ROUND_UP(anchor
->mpb_size
, 512);
2369 if (posix_memalign(&super
->buf
, 512, super
->len
) != 0) {
2372 Name
": unable to allocate %zu byte mpb buffer\n",
2377 memcpy(super
->buf
, anchor
, 512);
2379 sectors
= mpb_sectors(anchor
) - 1;
2382 check_sum
= __gen_imsm_checksum(super
->anchor
);
2383 if (check_sum
!= __le32_to_cpu(super
->anchor
->check_sum
)) {
2386 Name
": IMSM checksum %x != %x on %s\n",
2388 __le32_to_cpu(super
->anchor
->check_sum
),
2396 /* read the extended mpb */
2397 if (lseek64(fd
, dsize
- (512 * (2 + sectors
)), SEEK_SET
) < 0) {
2400 Name
": Cannot seek to extended mpb on %s: %s\n",
2401 devname
, strerror(errno
));
2405 if ((unsigned)read(fd
, super
->buf
+ 512, super
->len
- 512) != super
->len
- 512) {
2408 Name
": Cannot read extended mpb on %s: %s\n",
2409 devname
, strerror(errno
));
2413 check_sum
= __gen_imsm_checksum(super
->anchor
);
2414 if (check_sum
!= __le32_to_cpu(super
->anchor
->check_sum
)) {
2417 Name
": IMSM checksum %x != %x on %s\n",
2418 check_sum
, __le32_to_cpu(super
->anchor
->check_sum
),
2423 /* FIXME the BBM log is disk specific so we cannot use this global
2424 * buffer for all disks. Ok for now since we only look at the global
2425 * bbm_log_size parameter to gate assembly
2427 super
->bbm_log
= __get_imsm_bbm_log(super
->anchor
);
2433 load_and_parse_mpb(int fd
, struct intel_super
*super
, char *devname
, int keep_fd
)
2437 err
= load_imsm_mpb(fd
, super
, devname
);
2440 err
= load_imsm_disk(fd
, super
, devname
, keep_fd
);
2443 err
= parse_raid_devices(super
);
2448 static void __free_imsm_disk(struct dl
*d
)
2460 static void free_imsm_disks(struct intel_super
*super
)
2464 while (super
->disks
) {
2466 super
->disks
= d
->next
;
2467 __free_imsm_disk(d
);
2469 while (super
->missing
) {
2471 super
->missing
= d
->next
;
2472 __free_imsm_disk(d
);
2477 /* free all the pieces hanging off of a super pointer */
2478 static void __free_imsm(struct intel_super
*super
, int free_disks
)
2485 free_imsm_disks(super
);
2486 free_devlist(super
);
2488 free((void *) super
->hba
);
2493 static void free_imsm(struct intel_super
*super
)
2495 __free_imsm(super
, 1);
2499 static void free_super_imsm(struct supertype
*st
)
2501 struct intel_super
*super
= st
->sb
;
2510 static struct intel_super
*alloc_super(void)
2512 struct intel_super
*super
= malloc(sizeof(*super
));
2515 memset(super
, 0, sizeof(*super
));
2516 super
->current_vol
= -1;
2517 super
->create_offset
= ~((__u32
) 0);
2518 if (!check_env("IMSM_NO_PLATFORM"))
2519 super
->orom
= find_imsm_orom();
2520 if (super
->orom
&& !check_env("IMSM_TEST_OROM")) {
2521 struct sys_dev
*list
, *ent
;
2523 /* find the first intel ahci controller */
2524 list
= find_driver_devices("pci", "ahci");
2525 for (ent
= list
; ent
; ent
= ent
->next
)
2526 if (devpath_to_vendor(ent
->path
) == 0x8086)
2529 super
->hba
= ent
->path
;
2532 free_sys_dev(&list
);
2540 /* find_missing - helper routine for load_super_imsm_all that identifies
2541 * disks that have disappeared from the system. This routine relies on
2542 * the mpb being uptodate, which it is at load time.
2544 static int find_missing(struct intel_super
*super
)
2547 struct imsm_super
*mpb
= super
->anchor
;
2549 struct imsm_disk
*disk
;
2551 for (i
= 0; i
< mpb
->num_disks
; i
++) {
2552 disk
= __get_imsm_disk(mpb
, i
);
2553 dl
= serial_to_dl(disk
->serial
, super
);
2557 dl
= malloc(sizeof(*dl
));
2563 dl
->devname
= strdup("missing");
2565 serialcpy(dl
->serial
, disk
->serial
);
2568 dl
->next
= super
->missing
;
2569 super
->missing
= dl
;
2575 static struct intel_disk
*disk_list_get(__u8
*serial
, struct intel_disk
*disk_list
)
2577 struct intel_disk
*idisk
= disk_list
;
2580 if (serialcmp(idisk
->disk
.serial
, serial
) == 0)
2582 idisk
= idisk
->next
;
2588 static int __prep_thunderdome(struct intel_super
**table
, int tbl_size
,
2589 struct intel_super
*super
,
2590 struct intel_disk
**disk_list
)
2592 struct imsm_disk
*d
= &super
->disks
->disk
;
2593 struct imsm_super
*mpb
= super
->anchor
;
2596 for (i
= 0; i
< tbl_size
; i
++) {
2597 struct imsm_super
*tbl_mpb
= table
[i
]->anchor
;
2598 struct imsm_disk
*tbl_d
= &table
[i
]->disks
->disk
;
2600 if (tbl_mpb
->family_num
== mpb
->family_num
) {
2601 if (tbl_mpb
->check_sum
== mpb
->check_sum
) {
2602 dprintf("%s: mpb from %d:%d matches %d:%d\n",
2603 __func__
, super
->disks
->major
,
2604 super
->disks
->minor
,
2605 table
[i
]->disks
->major
,
2606 table
[i
]->disks
->minor
);
2610 if (((is_configured(d
) && !is_configured(tbl_d
)) ||
2611 is_configured(d
) == is_configured(tbl_d
)) &&
2612 tbl_mpb
->generation_num
< mpb
->generation_num
) {
2613 /* current version of the mpb is a
2614 * better candidate than the one in
2615 * super_table, but copy over "cross
2616 * generational" status
2618 struct intel_disk
*idisk
;
2620 dprintf("%s: mpb from %d:%d replaces %d:%d\n",
2621 __func__
, super
->disks
->major
,
2622 super
->disks
->minor
,
2623 table
[i
]->disks
->major
,
2624 table
[i
]->disks
->minor
);
2626 idisk
= disk_list_get(tbl_d
->serial
, *disk_list
);
2627 if (idisk
&& is_failed(&idisk
->disk
))
2628 tbl_d
->status
|= FAILED_DISK
;
2631 struct intel_disk
*idisk
;
2632 struct imsm_disk
*disk
;
2634 /* tbl_mpb is more up to date, but copy
2635 * over cross generational status before
2638 disk
= __serial_to_disk(d
->serial
, mpb
, NULL
);
2639 if (disk
&& is_failed(disk
))
2640 d
->status
|= FAILED_DISK
;
2642 idisk
= disk_list_get(d
->serial
, *disk_list
);
2645 if (disk
&& is_configured(disk
))
2646 idisk
->disk
.status
|= CONFIGURED_DISK
;
2649 dprintf("%s: mpb from %d:%d prefer %d:%d\n",
2650 __func__
, super
->disks
->major
,
2651 super
->disks
->minor
,
2652 table
[i
]->disks
->major
,
2653 table
[i
]->disks
->minor
);
2661 table
[tbl_size
++] = super
;
2665 /* update/extend the merged list of imsm_disk records */
2666 for (j
= 0; j
< mpb
->num_disks
; j
++) {
2667 struct imsm_disk
*disk
= __get_imsm_disk(mpb
, j
);
2668 struct intel_disk
*idisk
;
2670 idisk
= disk_list_get(disk
->serial
, *disk_list
);
2672 idisk
->disk
.status
|= disk
->status
;
2673 if (is_configured(&idisk
->disk
) ||
2674 is_failed(&idisk
->disk
))
2675 idisk
->disk
.status
&= ~(SPARE_DISK
);
2677 idisk
= calloc(1, sizeof(*idisk
));
2680 idisk
->owner
= IMSM_UNKNOWN_OWNER
;
2681 idisk
->disk
= *disk
;
2682 idisk
->next
= *disk_list
;
2686 if (serialcmp(idisk
->disk
.serial
, d
->serial
) == 0)
2693 static struct intel_super
*
2694 validate_members(struct intel_super
*super
, struct intel_disk
*disk_list
,
2697 struct imsm_super
*mpb
= super
->anchor
;
2701 for (i
= 0; i
< mpb
->num_disks
; i
++) {
2702 struct imsm_disk
*disk
= __get_imsm_disk(mpb
, i
);
2703 struct intel_disk
*idisk
;
2705 idisk
= disk_list_get(disk
->serial
, disk_list
);
2707 if (idisk
->owner
== owner
||
2708 idisk
->owner
== IMSM_UNKNOWN_OWNER
)
2711 dprintf("%s: '%.16s' owner %d != %d\n",
2712 __func__
, disk
->serial
, idisk
->owner
,
2715 dprintf("%s: unknown disk %x [%d]: %.16s\n",
2716 __func__
, __le32_to_cpu(mpb
->family_num
), i
,
2722 if (ok_count
== mpb
->num_disks
)
2727 static void show_conflicts(__u32 family_num
, struct intel_super
*super_list
)
2729 struct intel_super
*s
;
2731 for (s
= super_list
; s
; s
= s
->next
) {
2732 if (family_num
!= s
->anchor
->family_num
)
2734 fprintf(stderr
, "Conflict, offlining family %#x on '%s'\n",
2735 __le32_to_cpu(family_num
), s
->disks
->devname
);
2739 static struct intel_super
*
2740 imsm_thunderdome(struct intel_super
**super_list
, int len
)
2742 struct intel_super
*super_table
[len
];
2743 struct intel_disk
*disk_list
= NULL
;
2744 struct intel_super
*champion
, *spare
;
2745 struct intel_super
*s
, **del
;
2750 memset(super_table
, 0, sizeof(super_table
));
2751 for (s
= *super_list
; s
; s
= s
->next
)
2752 tbl_size
= __prep_thunderdome(super_table
, tbl_size
, s
, &disk_list
);
2754 for (i
= 0; i
< tbl_size
; i
++) {
2755 struct imsm_disk
*d
;
2756 struct intel_disk
*idisk
;
2757 struct imsm_super
*mpb
= super_table
[i
]->anchor
;
2760 d
= &s
->disks
->disk
;
2762 /* 'd' must appear in merged disk list for its
2763 * configuration to be valid
2765 idisk
= disk_list_get(d
->serial
, disk_list
);
2766 if (idisk
&& idisk
->owner
== i
)
2767 s
= validate_members(s
, disk_list
, i
);
2772 dprintf("%s: marking family: %#x from %d:%d offline\n",
2773 __func__
, mpb
->family_num
,
2774 super_table
[i
]->disks
->major
,
2775 super_table
[i
]->disks
->minor
);
2779 /* This is where the mdadm implementation differs from the Windows
2780 * driver which has no strict concept of a container. We can only
2781 * assemble one family from a container, so when returning a prodigal
2782 * array member to this system the code will not be able to disambiguate
2783 * the container contents that should be assembled ("foreign" versus
2784 * "local"). It requires user intervention to set the orig_family_num
2785 * to a new value to establish a new container. The Windows driver in
2786 * this situation fixes up the volume name in place and manages the
2787 * foreign array as an independent entity.
2792 for (i
= 0; i
< tbl_size
; i
++) {
2793 struct intel_super
*tbl_ent
= super_table
[i
];
2799 if (tbl_ent
->anchor
->num_raid_devs
== 0) {
2804 if (s
&& !is_spare
) {
2805 show_conflicts(tbl_ent
->anchor
->family_num
, *super_list
);
2807 } else if (!s
&& !is_spare
)
2820 fprintf(stderr
, "Chose family %#x on '%s', "
2821 "assemble conflicts to new container with '--update=uuid'\n",
2822 __le32_to_cpu(s
->anchor
->family_num
), s
->disks
->devname
);
2824 /* collect all dl's onto 'champion', and update them to
2825 * champion's version of the status
2827 for (s
= *super_list
; s
; s
= s
->next
) {
2828 struct imsm_super
*mpb
= champion
->anchor
;
2829 struct dl
*dl
= s
->disks
;
2834 for (i
= 0; i
< mpb
->num_disks
; i
++) {
2835 struct imsm_disk
*disk
;
2837 disk
= __serial_to_disk(dl
->serial
, mpb
, &dl
->index
);
2840 /* only set index on disks that are a member of
2841 * a populated contianer, i.e. one with
2844 if (is_failed(&dl
->disk
))
2846 else if (is_spare(&dl
->disk
))
2852 if (i
>= mpb
->num_disks
) {
2853 struct intel_disk
*idisk
;
2855 idisk
= disk_list_get(dl
->serial
, disk_list
);
2856 if (idisk
&& is_spare(&idisk
->disk
) &&
2857 !is_failed(&idisk
->disk
) && !is_configured(&idisk
->disk
))
2865 dl
->next
= champion
->disks
;
2866 champion
->disks
= dl
;
2870 /* delete 'champion' from super_list */
2871 for (del
= super_list
; *del
; ) {
2872 if (*del
== champion
) {
2873 *del
= (*del
)->next
;
2876 del
= &(*del
)->next
;
2878 champion
->next
= NULL
;
2882 struct intel_disk
*idisk
= disk_list
;
2884 disk_list
= disk_list
->next
;
2891 static int load_super_imsm_all(struct supertype
*st
, int fd
, void **sbp
,
2895 struct intel_super
*super_list
= NULL
;
2896 struct intel_super
*super
= NULL
;
2897 int devnum
= fd2devnum(fd
);
2903 /* check if 'fd' an opened container */
2904 sra
= sysfs_read(fd
, 0, GET_LEVEL
|GET_VERSION
|GET_DEVS
|GET_STATE
);
2908 if (sra
->array
.major_version
!= -1 ||
2909 sra
->array
.minor_version
!= -2 ||
2910 strcmp(sra
->text_version
, "imsm") != 0) {
2915 for (sd
= sra
->devs
, i
= 0; sd
; sd
= sd
->next
, i
++) {
2916 struct intel_super
*s
= alloc_super();
2923 s
->next
= super_list
;
2927 sprintf(nm
, "%d:%d", sd
->disk
.major
, sd
->disk
.minor
);
2928 dfd
= dev_open(nm
, O_RDWR
);
2932 err
= load_and_parse_mpb(dfd
, s
, NULL
, 1);
2934 /* retry the load if we might have raced against mdmon */
2935 if (err
== 3 && mdmon_running(devnum
))
2936 for (retry
= 0; retry
< 3; retry
++) {
2938 err
= load_and_parse_mpb(dfd
, s
, NULL
, 1);
2946 /* all mpbs enter, maybe one leaves */
2947 super
= imsm_thunderdome(&super_list
, i
);
2953 if (find_missing(super
) != 0) {
2961 while (super_list
) {
2962 struct intel_super
*s
= super_list
;
2964 super_list
= super_list
->next
;
2973 st
->container_dev
= devnum
;
2974 if (err
== 0 && st
->ss
== NULL
) {
2975 st
->ss
= &super_imsm
;
2976 st
->minor_version
= 0;
2977 st
->max_devs
= IMSM_MAX_DEVICES
;
2982 static int load_container_imsm(struct supertype
*st
, int fd
, char *devname
)
2984 return load_super_imsm_all(st
, fd
, &st
->sb
, devname
);
2988 static int load_super_imsm(struct supertype
*st
, int fd
, char *devname
)
2990 struct intel_super
*super
;
2994 if (load_super_imsm_all(st
, fd
, &st
->sb
, devname
) == 0)
2998 if (test_partition(fd
))
2999 /* IMSM not allowed on partitions */
3002 free_super_imsm(st
);
3004 super
= alloc_super();
3007 Name
": malloc of %zu failed.\n",
3012 rv
= load_and_parse_mpb(fd
, super
, devname
, 0);
3017 Name
": Failed to load all information "
3018 "sections on %s\n", devname
);
3024 if (st
->ss
== NULL
) {
3025 st
->ss
= &super_imsm
;
3026 st
->minor_version
= 0;
3027 st
->max_devs
= IMSM_MAX_DEVICES
;
3032 static __u16
info_to_blocks_per_strip(mdu_array_info_t
*info
)
3034 if (info
->level
== 1)
3036 return info
->chunk_size
>> 9;
3039 static __u32
info_to_num_data_stripes(mdu_array_info_t
*info
, int num_domains
)
3043 num_stripes
= (info
->size
* 2) / info_to_blocks_per_strip(info
);
3044 num_stripes
/= num_domains
;
3049 static __u32
info_to_blocks_per_member(mdu_array_info_t
*info
)
3051 if (info
->level
== 1)
3052 return info
->size
* 2;
3054 return (info
->size
* 2) & ~(info_to_blocks_per_strip(info
) - 1);
3057 static void imsm_update_version_info(struct intel_super
*super
)
3059 /* update the version and attributes */
3060 struct imsm_super
*mpb
= super
->anchor
;
3062 struct imsm_dev
*dev
;
3063 struct imsm_map
*map
;
3066 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
3067 dev
= get_imsm_dev(super
, i
);
3068 map
= get_imsm_map(dev
, 0);
3069 if (__le32_to_cpu(dev
->size_high
) > 0)
3070 mpb
->attributes
|= MPB_ATTRIB_2TB
;
3072 /* FIXME detect when an array spans a port multiplier */
3074 mpb
->attributes
|= MPB_ATTRIB_PM
;
3077 if (mpb
->num_raid_devs
> 1 ||
3078 mpb
->attributes
!= MPB_ATTRIB_CHECKSUM_VERIFY
) {
3079 version
= MPB_VERSION_ATTRIBS
;
3080 switch (get_imsm_raid_level(map
)) {
3081 case 0: mpb
->attributes
|= MPB_ATTRIB_RAID0
; break;
3082 case 1: mpb
->attributes
|= MPB_ATTRIB_RAID1
; break;
3083 case 10: mpb
->attributes
|= MPB_ATTRIB_RAID10
; break;
3084 case 5: mpb
->attributes
|= MPB_ATTRIB_RAID5
; break;
3087 if (map
->num_members
>= 5)
3088 version
= MPB_VERSION_5OR6_DISK_ARRAY
;
3089 else if (dev
->status
== DEV_CLONE_N_GO
)
3090 version
= MPB_VERSION_CNG
;
3091 else if (get_imsm_raid_level(map
) == 5)
3092 version
= MPB_VERSION_RAID5
;
3093 else if (map
->num_members
>= 3)
3094 version
= MPB_VERSION_3OR4_DISK_ARRAY
;
3095 else if (get_imsm_raid_level(map
) == 1)
3096 version
= MPB_VERSION_RAID1
;
3098 version
= MPB_VERSION_RAID0
;
3100 strcpy(((char *) mpb
->sig
) + strlen(MPB_SIGNATURE
), version
);
3104 static int check_name(struct intel_super
*super
, char *name
, int quiet
)
3106 struct imsm_super
*mpb
= super
->anchor
;
3107 char *reason
= NULL
;
3110 if (strlen(name
) > MAX_RAID_SERIAL_LEN
)
3111 reason
= "must be 16 characters or less";
3113 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
3114 struct imsm_dev
*dev
= get_imsm_dev(super
, i
);
3116 if (strncmp((char *) dev
->volume
, name
, MAX_RAID_SERIAL_LEN
) == 0) {
3117 reason
= "already exists";
3122 if (reason
&& !quiet
)
3123 fprintf(stderr
, Name
": imsm volume name %s\n", reason
);
3128 static int init_super_imsm_volume(struct supertype
*st
, mdu_array_info_t
*info
,
3129 unsigned long long size
, char *name
,
3130 char *homehost
, int *uuid
)
3132 /* We are creating a volume inside a pre-existing container.
3133 * so st->sb is already set.
3135 struct intel_super
*super
= st
->sb
;
3136 struct imsm_super
*mpb
= super
->anchor
;
3137 struct intel_dev
*dv
;
3138 struct imsm_dev
*dev
;
3139 struct imsm_vol
*vol
;
3140 struct imsm_map
*map
;
3141 int idx
= mpb
->num_raid_devs
;
3143 unsigned long long array_blocks
;
3144 size_t size_old
, size_new
;
3145 __u32 num_data_stripes
;
3147 if (super
->orom
&& mpb
->num_raid_devs
>= super
->orom
->vpa
) {
3148 fprintf(stderr
, Name
": This imsm-container already has the "
3149 "maximum of %d volumes\n", super
->orom
->vpa
);
3153 /* ensure the mpb is large enough for the new data */
3154 size_old
= __le32_to_cpu(mpb
->mpb_size
);
3155 size_new
= disks_to_mpb_size(info
->nr_disks
);
3156 if (size_new
> size_old
) {
3158 size_t size_round
= ROUND_UP(size_new
, 512);
3160 if (posix_memalign(&mpb_new
, 512, size_round
) != 0) {
3161 fprintf(stderr
, Name
": could not allocate new mpb\n");
3164 memcpy(mpb_new
, mpb
, size_old
);
3167 super
->anchor
= mpb_new
;
3168 mpb
->mpb_size
= __cpu_to_le32(size_new
);
3169 memset(mpb_new
+ size_old
, 0, size_round
- size_old
);
3171 super
->current_vol
= idx
;
3172 /* when creating the first raid device in this container set num_disks
3173 * to zero, i.e. delete this spare and add raid member devices in
3174 * add_to_super_imsm_volume()
3176 if (super
->current_vol
== 0)
3179 if (!check_name(super
, name
, 0))
3181 dv
= malloc(sizeof(*dv
));
3183 fprintf(stderr
, Name
": failed to allocate device list entry\n");
3186 dev
= malloc(sizeof(*dev
) + sizeof(__u32
) * (info
->raid_disks
- 1));
3189 fprintf(stderr
, Name
": could not allocate raid device\n");
3192 strncpy((char *) dev
->volume
, name
, MAX_RAID_SERIAL_LEN
);
3193 if (info
->level
== 1)
3194 array_blocks
= info_to_blocks_per_member(info
);
3196 array_blocks
= calc_array_size(info
->level
, info
->raid_disks
,
3197 info
->layout
, info
->chunk_size
,
3199 /* round array size down to closest MB */
3200 array_blocks
= (array_blocks
>> SECT_PER_MB_SHIFT
) << SECT_PER_MB_SHIFT
;
3202 dev
->size_low
= __cpu_to_le32((__u32
) array_blocks
);
3203 dev
->size_high
= __cpu_to_le32((__u32
) (array_blocks
>> 32));
3204 dev
->status
= __cpu_to_le32(0);
3205 dev
->reserved_blocks
= __cpu_to_le32(0);
3207 vol
->migr_state
= 0;
3208 set_migr_type(dev
, MIGR_INIT
);
3210 vol
->curr_migr_unit
= 0;
3211 map
= get_imsm_map(dev
, 0);
3212 map
->pba_of_lba0
= __cpu_to_le32(super
->create_offset
);
3213 map
->blocks_per_member
= __cpu_to_le32(info_to_blocks_per_member(info
));
3214 map
->blocks_per_strip
= __cpu_to_le16(info_to_blocks_per_strip(info
));
3215 map
->failed_disk_num
= ~0;
3216 map
->map_state
= info
->level
? IMSM_T_STATE_UNINITIALIZED
:
3217 IMSM_T_STATE_NORMAL
;
3220 if (info
->level
== 1 && info
->raid_disks
> 2) {
3223 fprintf(stderr
, Name
": imsm does not support more than 2 disks"
3224 "in a raid1 volume\n");
3228 map
->raid_level
= info
->level
;
3229 if (info
->level
== 10) {
3230 map
->raid_level
= 1;
3231 map
->num_domains
= info
->raid_disks
/ 2;
3232 } else if (info
->level
== 1)
3233 map
->num_domains
= info
->raid_disks
;
3235 map
->num_domains
= 1;
3237 num_data_stripes
= info_to_num_data_stripes(info
, map
->num_domains
);
3238 map
->num_data_stripes
= __cpu_to_le32(num_data_stripes
);
3240 map
->num_members
= info
->raid_disks
;
3241 for (i
= 0; i
< map
->num_members
; i
++) {
3242 /* initialized in add_to_super */
3243 set_imsm_ord_tbl_ent(map
, i
, IMSM_ORD_REBUILD
);
3245 mpb
->num_raid_devs
++;
3248 dv
->index
= super
->current_vol
;
3249 dv
->next
= super
->devlist
;
3250 super
->devlist
= dv
;
3252 imsm_update_version_info(super
);
3257 static int init_super_imsm(struct supertype
*st
, mdu_array_info_t
*info
,
3258 unsigned long long size
, char *name
,
3259 char *homehost
, int *uuid
)
3261 /* This is primarily called by Create when creating a new array.
3262 * We will then get add_to_super called for each component, and then
3263 * write_init_super called to write it out to each device.
3264 * For IMSM, Create can create on fresh devices or on a pre-existing
3266 * To create on a pre-existing array a different method will be called.
3267 * This one is just for fresh drives.
3269 struct intel_super
*super
;
3270 struct imsm_super
*mpb
;
3275 return init_super_imsm_volume(st
, info
, size
, name
, homehost
, uuid
);
3278 mpb_size
= disks_to_mpb_size(info
->nr_disks
);
3282 super
= alloc_super();
3283 if (super
&& posix_memalign(&super
->buf
, 512, mpb_size
) != 0) {
3288 fprintf(stderr
, Name
3289 ": %s could not allocate superblock\n", __func__
);
3292 memset(super
->buf
, 0, mpb_size
);
3294 mpb
->mpb_size
= __cpu_to_le32(mpb_size
);
3298 /* zeroing superblock */
3302 mpb
->attributes
= MPB_ATTRIB_CHECKSUM_VERIFY
;
3304 version
= (char *) mpb
->sig
;
3305 strcpy(version
, MPB_SIGNATURE
);
3306 version
+= strlen(MPB_SIGNATURE
);
3307 strcpy(version
, MPB_VERSION_RAID0
);
3313 static int add_to_super_imsm_volume(struct supertype
*st
, mdu_disk_info_t
*dk
,
3314 int fd
, char *devname
)
3316 struct intel_super
*super
= st
->sb
;
3317 struct imsm_super
*mpb
= super
->anchor
;
3319 struct imsm_dev
*dev
;
3320 struct imsm_map
*map
;
3323 dev
= get_imsm_dev(super
, super
->current_vol
);
3324 map
= get_imsm_map(dev
, 0);
3326 if (! (dk
->state
& (1<<MD_DISK_SYNC
))) {
3327 fprintf(stderr
, Name
": %s: Cannot add spare devices to IMSM volume\n",
3333 /* we're doing autolayout so grab the pre-marked (in
3334 * validate_geometry) raid_disk
3336 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
3337 if (dl
->raiddisk
== dk
->raid_disk
)
3340 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
3341 if (dl
->major
== dk
->major
&&
3342 dl
->minor
== dk
->minor
)
3347 fprintf(stderr
, Name
": %s is not a member of the same container\n", devname
);
3351 /* add a pristine spare to the metadata */
3352 if (dl
->index
< 0) {
3353 dl
->index
= super
->anchor
->num_disks
;
3354 super
->anchor
->num_disks
++;
3356 /* Check the device has not already been added */
3357 slot
= get_imsm_disk_slot(map
, dl
->index
);
3359 (get_imsm_ord_tbl_ent(dev
, slot
) & IMSM_ORD_REBUILD
) == 0) {
3360 fprintf(stderr
, Name
": %s has been included in this array twice\n",
3364 set_imsm_ord_tbl_ent(map
, dk
->number
, dl
->index
);
3365 dl
->disk
.status
= CONFIGURED_DISK
;
3367 /* if we are creating the first raid device update the family number */
3368 if (super
->current_vol
== 0) {
3370 struct imsm_dev
*_dev
= __get_imsm_dev(mpb
, 0);
3371 struct imsm_disk
*_disk
= __get_imsm_disk(mpb
, dl
->index
);
3373 if (!_dev
|| !_disk
) {
3374 fprintf(stderr
, Name
": BUG mpb setup error\n");
3380 sum
+= __gen_imsm_checksum(mpb
);
3381 mpb
->family_num
= __cpu_to_le32(sum
);
3382 mpb
->orig_family_num
= mpb
->family_num
;
3388 static int add_to_super_imsm(struct supertype
*st
, mdu_disk_info_t
*dk
,
3389 int fd
, char *devname
)
3391 struct intel_super
*super
= st
->sb
;
3393 unsigned long long size
;
3398 /* if we are on an RAID enabled platform check that the disk is
3399 * attached to the raid controller
3401 if (super
->hba
&& !disk_attached_to_hba(fd
, super
->hba
)) {
3403 Name
": %s is not attached to the raid controller: %s\n",
3404 devname
? : "disk", super
->hba
);
3408 if (super
->current_vol
>= 0)
3409 return add_to_super_imsm_volume(st
, dk
, fd
, devname
);
3412 dd
= malloc(sizeof(*dd
));
3415 Name
": malloc failed %s:%d.\n", __func__
, __LINE__
);
3418 memset(dd
, 0, sizeof(*dd
));
3419 dd
->major
= major(stb
.st_rdev
);
3420 dd
->minor
= minor(stb
.st_rdev
);
3422 dd
->devname
= devname
? strdup(devname
) : NULL
;
3425 dd
->action
= DISK_ADD
;
3426 rv
= imsm_read_serial(fd
, devname
, dd
->serial
);
3429 Name
": failed to retrieve scsi serial, aborting\n");
3434 get_dev_size(fd
, NULL
, &size
);
3436 serialcpy(dd
->disk
.serial
, dd
->serial
);
3437 dd
->disk
.total_blocks
= __cpu_to_le32(size
);
3438 dd
->disk
.status
= SPARE_DISK
;
3439 if (sysfs_disk_to_scsi_id(fd
, &id
) == 0)
3440 dd
->disk
.scsi_id
= __cpu_to_le32(id
);
3442 dd
->disk
.scsi_id
= __cpu_to_le32(0);
3444 if (st
->update_tail
) {
3445 dd
->next
= super
->disk_mgmt_list
;
3446 super
->disk_mgmt_list
= dd
;
3448 dd
->next
= super
->disks
;
3456 static int remove_from_super_imsm(struct supertype
*st
, mdu_disk_info_t
*dk
)
3458 struct intel_super
*super
= st
->sb
;
3461 /* remove from super works only in mdmon - for communication
3462 * manager - monitor. Check if communication memory buffer
3465 if (!st
->update_tail
) {
3467 Name
": %s shall be used in mdmon context only"
3468 "(line %d).\n", __func__
, __LINE__
);
3471 dd
= malloc(sizeof(*dd
));
3474 Name
": malloc failed %s:%d.\n", __func__
, __LINE__
);
3477 memset(dd
, 0, sizeof(*dd
));
3478 dd
->major
= dk
->major
;
3479 dd
->minor
= dk
->minor
;
3482 dd
->disk
.status
= SPARE_DISK
;
3483 dd
->action
= DISK_REMOVE
;
3485 dd
->next
= super
->disk_mgmt_list
;
3486 super
->disk_mgmt_list
= dd
;
3492 static int store_imsm_mpb(int fd
, struct imsm_super
*mpb
);
3496 struct imsm_super anchor
;
3497 } spare_record
__attribute__ ((aligned(512)));
3499 /* spare records have their own family number and do not have any defined raid
3502 static int write_super_imsm_spares(struct intel_super
*super
, int doclose
)
3504 struct imsm_super
*mpb
= super
->anchor
;
3505 struct imsm_super
*spare
= &spare_record
.anchor
;
3509 spare
->mpb_size
= __cpu_to_le32(sizeof(struct imsm_super
)),
3510 spare
->generation_num
= __cpu_to_le32(1UL),
3511 spare
->attributes
= MPB_ATTRIB_CHECKSUM_VERIFY
;
3512 spare
->num_disks
= 1,
3513 spare
->num_raid_devs
= 0,
3514 spare
->cache_size
= mpb
->cache_size
,
3515 spare
->pwr_cycle_count
= __cpu_to_le32(1),
3517 snprintf((char *) spare
->sig
, MAX_SIGNATURE_LENGTH
,
3518 MPB_SIGNATURE MPB_VERSION_RAID0
);
3520 for (d
= super
->disks
; d
; d
= d
->next
) {
3524 spare
->disk
[0] = d
->disk
;
3525 sum
= __gen_imsm_checksum(spare
);
3526 spare
->family_num
= __cpu_to_le32(sum
);
3527 spare
->orig_family_num
= 0;
3528 sum
= __gen_imsm_checksum(spare
);
3529 spare
->check_sum
= __cpu_to_le32(sum
);
3531 if (store_imsm_mpb(d
->fd
, spare
)) {
3532 fprintf(stderr
, "%s: failed for device %d:%d %s\n",
3533 __func__
, d
->major
, d
->minor
, strerror(errno
));
3545 static int write_super_imsm(struct supertype
*st
, int doclose
)
3547 struct intel_super
*super
= st
->sb
;
3548 struct imsm_super
*mpb
= super
->anchor
;
3554 __u32 mpb_size
= sizeof(struct imsm_super
) - sizeof(struct imsm_disk
);
3557 /* 'generation' is incremented everytime the metadata is written */
3558 generation
= __le32_to_cpu(mpb
->generation_num
);
3560 mpb
->generation_num
= __cpu_to_le32(generation
);
3562 /* fix up cases where previous mdadm releases failed to set
3565 if (mpb
->orig_family_num
== 0)
3566 mpb
->orig_family_num
= mpb
->family_num
;
3568 for (d
= super
->disks
; d
; d
= d
->next
) {
3572 mpb
->disk
[d
->index
] = d
->disk
;
3576 for (d
= super
->missing
; d
; d
= d
->next
) {
3577 mpb
->disk
[d
->index
] = d
->disk
;
3580 mpb
->num_disks
= num_disks
;
3581 mpb_size
+= sizeof(struct imsm_disk
) * mpb
->num_disks
;
3583 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
3584 struct imsm_dev
*dev
= __get_imsm_dev(mpb
, i
);
3585 struct imsm_dev
*dev2
= get_imsm_dev(super
, i
);
3587 imsm_copy_dev(dev
, dev2
);
3588 mpb_size
+= sizeof_imsm_dev(dev
, 0);
3591 mpb_size
+= __le32_to_cpu(mpb
->bbm_log_size
);
3592 mpb
->mpb_size
= __cpu_to_le32(mpb_size
);
3594 /* recalculate checksum */
3595 sum
= __gen_imsm_checksum(mpb
);
3596 mpb
->check_sum
= __cpu_to_le32(sum
);
3598 /* write the mpb for disks that compose raid devices */
3599 for (d
= super
->disks
; d
; d
= d
->next
) {
3602 if (store_imsm_mpb(d
->fd
, mpb
))
3603 fprintf(stderr
, "%s: failed for device %d:%d %s\n",
3604 __func__
, d
->major
, d
->minor
, strerror(errno
));
3612 return write_super_imsm_spares(super
, doclose
);
3618 static int create_array(struct supertype
*st
, int dev_idx
)
3621 struct imsm_update_create_array
*u
;
3622 struct intel_super
*super
= st
->sb
;
3623 struct imsm_dev
*dev
= get_imsm_dev(super
, dev_idx
);
3624 struct imsm_map
*map
= get_imsm_map(dev
, 0);
3625 struct disk_info
*inf
;
3626 struct imsm_disk
*disk
;
3629 len
= sizeof(*u
) - sizeof(*dev
) + sizeof_imsm_dev(dev
, 0) +
3630 sizeof(*inf
) * map
->num_members
;
3633 fprintf(stderr
, "%s: failed to allocate update buffer\n",
3638 u
->type
= update_create_array
;
3639 u
->dev_idx
= dev_idx
;
3640 imsm_copy_dev(&u
->dev
, dev
);
3641 inf
= get_disk_info(u
);
3642 for (i
= 0; i
< map
->num_members
; i
++) {
3643 int idx
= get_imsm_disk_idx(dev
, i
);
3645 disk
= get_imsm_disk(super
, idx
);
3646 serialcpy(inf
[i
].serial
, disk
->serial
);
3648 append_metadata_update(st
, u
, len
);
3653 static int mgmt_disk(struct supertype
*st
)
3655 struct intel_super
*super
= st
->sb
;
3657 struct imsm_update_add_remove_disk
*u
;
3659 if (!super
->disk_mgmt_list
)
3665 fprintf(stderr
, "%s: failed to allocate update buffer\n",
3670 u
->type
= update_add_remove_disk
;
3671 append_metadata_update(st
, u
, len
);
3676 static int write_init_super_imsm(struct supertype
*st
)
3678 struct intel_super
*super
= st
->sb
;
3679 int current_vol
= super
->current_vol
;
3681 /* we are done with current_vol reset it to point st at the container */
3682 super
->current_vol
= -1;
3684 if (st
->update_tail
) {
3685 /* queue the recently created array / added disk
3686 * as a metadata update */
3690 /* determine if we are creating a volume or adding a disk */
3691 if (current_vol
< 0) {
3692 /* in the mgmt (add/remove) disk case we are running
3693 * in mdmon context, so don't close fd's
3695 return mgmt_disk(st
);
3697 rv
= create_array(st
, current_vol
);
3699 for (d
= super
->disks
; d
; d
= d
->next
) {
3707 for (d
= super
->disks
; d
; d
= d
->next
)
3708 Kill(d
->devname
, NULL
, 0, 1, 1);
3709 return write_super_imsm(st
, 1);
3714 static int store_super_imsm(struct supertype
*st
, int fd
)
3716 struct intel_super
*super
= st
->sb
;
3717 struct imsm_super
*mpb
= super
? super
->anchor
: NULL
;
3723 return store_imsm_mpb(fd
, mpb
);
3729 static int imsm_bbm_log_size(struct imsm_super
*mpb
)
3731 return __le32_to_cpu(mpb
->bbm_log_size
);
3735 static int validate_geometry_imsm_container(struct supertype
*st
, int level
,
3736 int layout
, int raiddisks
, int chunk
,
3737 unsigned long long size
, char *dev
,
3738 unsigned long long *freesize
,
3742 unsigned long long ldsize
;
3743 const struct imsm_orom
*orom
;
3745 if (level
!= LEVEL_CONTAINER
)
3750 if (check_env("IMSM_NO_PLATFORM"))
3753 orom
= find_imsm_orom();
3754 if (orom
&& raiddisks
> orom
->tds
) {
3756 fprintf(stderr
, Name
": %d exceeds maximum number of"
3757 " platform supported disks: %d\n",
3758 raiddisks
, orom
->tds
);
3762 fd
= open(dev
, O_RDONLY
|O_EXCL
, 0);
3765 fprintf(stderr
, Name
": imsm: Cannot open %s: %s\n",
3766 dev
, strerror(errno
));
3769 if (!get_dev_size(fd
, dev
, &ldsize
)) {
3775 *freesize
= avail_size_imsm(st
, ldsize
>> 9);
3780 static unsigned long long find_size(struct extent
*e
, int *idx
, int num_extents
)
3782 const unsigned long long base_start
= e
[*idx
].start
;
3783 unsigned long long end
= base_start
+ e
[*idx
].size
;
3786 if (base_start
== end
)
3790 for (i
= *idx
; i
< num_extents
; i
++) {
3791 /* extend overlapping extents */
3792 if (e
[i
].start
>= base_start
&&
3793 e
[i
].start
<= end
) {
3796 if (e
[i
].start
+ e
[i
].size
> end
)
3797 end
= e
[i
].start
+ e
[i
].size
;
3798 } else if (e
[i
].start
> end
) {
3804 return end
- base_start
;
3807 static unsigned long long merge_extents(struct intel_super
*super
, int sum_extents
)
3809 /* build a composite disk with all known extents and generate a new
3810 * 'maxsize' given the "all disks in an array must share a common start
3811 * offset" constraint
3813 struct extent
*e
= calloc(sum_extents
, sizeof(*e
));
3817 unsigned long long pos
;
3818 unsigned long long start
= 0;
3819 unsigned long long maxsize
;
3820 unsigned long reserve
;
3825 /* coalesce and sort all extents. also, check to see if we need to
3826 * reserve space between member arrays
3829 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
3832 for (i
= 0; i
< dl
->extent_cnt
; i
++)
3835 qsort(e
, sum_extents
, sizeof(*e
), cmp_extent
);
3840 while (i
< sum_extents
) {
3841 e
[j
].start
= e
[i
].start
;
3842 e
[j
].size
= find_size(e
, &i
, sum_extents
);
3844 if (e
[j
-1].size
== 0)
3853 unsigned long long esize
;
3855 esize
= e
[i
].start
- pos
;
3856 if (esize
>= maxsize
) {
3861 pos
= e
[i
].start
+ e
[i
].size
;
3863 } while (e
[i
-1].size
);
3869 /* FIXME assumes volume at offset 0 is the first volume in a
3872 if (start_extent
> 0)
3873 reserve
= IMSM_RESERVED_SECTORS
; /* gap between raid regions */
3877 if (maxsize
< reserve
)
3880 super
->create_offset
= ~((__u32
) 0);
3881 if (start
+ reserve
> super
->create_offset
)
3882 return 0; /* start overflows create_offset */
3883 super
->create_offset
= start
+ reserve
;
3885 return maxsize
- reserve
;
3888 static int is_raid_level_supported(const struct imsm_orom
*orom
, int level
, int raiddisks
)
3890 if (level
< 0 || level
== 6 || level
== 4)
3893 /* if we have an orom prevent invalid raid levels */
3896 case 0: return imsm_orom_has_raid0(orom
);
3899 return imsm_orom_has_raid1e(orom
);
3900 return imsm_orom_has_raid1(orom
) && raiddisks
== 2;
3901 case 10: return imsm_orom_has_raid10(orom
) && raiddisks
== 4;
3902 case 5: return imsm_orom_has_raid5(orom
) && raiddisks
> 2;
3905 return 1; /* not on an Intel RAID platform so anything goes */
3910 #define pr_vrb(fmt, arg...) (void) (verbose && fprintf(stderr, Name fmt, ##arg))
3912 validate_geometry_imsm_orom(struct intel_super
*super
, int level
, int layout
,
3913 int raiddisks
, int chunk
, int verbose
)
3915 if (!is_raid_level_supported(super
->orom
, level
, raiddisks
)) {
3916 pr_vrb(": platform does not support raid%d with %d disk%s\n",
3917 level
, raiddisks
, raiddisks
> 1 ? "s" : "");
3920 if (super
->orom
&& level
!= 1 &&
3921 !imsm_orom_has_chunk(super
->orom
, chunk
)) {
3922 pr_vrb(": platform does not support a chunk size of: %d\n", chunk
);
3925 if (layout
!= imsm_level_to_layout(level
)) {
3927 pr_vrb(": imsm raid 5 only supports the left-asymmetric layout\n");
3928 else if (level
== 10)
3929 pr_vrb(": imsm raid 10 only supports the n2 layout\n");
3931 pr_vrb(": imsm unknown layout %#x for this raid level %d\n",
3939 /* validate_geometry_imsm_volume - lifted from validate_geometry_ddf_bvd
3940 * FIX ME add ahci details
3942 static int validate_geometry_imsm_volume(struct supertype
*st
, int level
,
3943 int layout
, int raiddisks
, int chunk
,
3944 unsigned long long size
, char *dev
,
3945 unsigned long long *freesize
,
3949 struct intel_super
*super
= st
->sb
;
3950 struct imsm_super
*mpb
= super
->anchor
;
3952 unsigned long long pos
= 0;
3953 unsigned long long maxsize
;
3957 /* We must have the container info already read in. */
3961 if (!validate_geometry_imsm_orom(super
, level
, layout
, raiddisks
, chunk
, verbose
))
3965 /* General test: make sure there is space for
3966 * 'raiddisks' device extents of size 'size' at a given
3969 unsigned long long minsize
= size
;
3970 unsigned long long start_offset
= MaxSector
;
3973 minsize
= MPB_SECTOR_CNT
+ IMSM_RESERVED_SECTORS
;
3974 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
3979 e
= get_extents(super
, dl
);
3982 unsigned long long esize
;
3983 esize
= e
[i
].start
- pos
;
3984 if (esize
>= minsize
)
3986 if (found
&& start_offset
== MaxSector
) {
3989 } else if (found
&& pos
!= start_offset
) {
3993 pos
= e
[i
].start
+ e
[i
].size
;
3995 } while (e
[i
-1].size
);
4000 if (dcnt
< raiddisks
) {
4002 fprintf(stderr
, Name
": imsm: Not enough "
4003 "devices with space for this array "
4011 /* This device must be a member of the set */
4012 if (stat(dev
, &stb
) < 0)
4014 if ((S_IFMT
& stb
.st_mode
) != S_IFBLK
)
4016 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
4017 if (dl
->major
== (int)major(stb
.st_rdev
) &&
4018 dl
->minor
== (int)minor(stb
.st_rdev
))
4023 fprintf(stderr
, Name
": %s is not in the "
4024 "same imsm set\n", dev
);
4026 } else if (super
->orom
&& dl
->index
< 0 && mpb
->num_raid_devs
) {
4027 /* If a volume is present then the current creation attempt
4028 * cannot incorporate new spares because the orom may not
4029 * understand this configuration (all member disks must be
4030 * members of each array in the container).
4032 fprintf(stderr
, Name
": %s is a spare and a volume"
4033 " is already defined for this container\n", dev
);
4034 fprintf(stderr
, Name
": The option-rom requires all member"
4035 " disks to be a member of all volumes\n");
4039 /* retrieve the largest free space block */
4040 e
= get_extents(super
, dl
);
4045 unsigned long long esize
;
4047 esize
= e
[i
].start
- pos
;
4048 if (esize
>= maxsize
)
4050 pos
= e
[i
].start
+ e
[i
].size
;
4052 } while (e
[i
-1].size
);
4057 fprintf(stderr
, Name
": unable to determine free space for: %s\n",
4061 if (maxsize
< size
) {
4063 fprintf(stderr
, Name
": %s not enough space (%llu < %llu)\n",
4064 dev
, maxsize
, size
);
4068 /* count total number of extents for merge */
4070 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
4072 i
+= dl
->extent_cnt
;
4074 maxsize
= merge_extents(super
, i
);
4075 if (maxsize
< size
|| maxsize
== 0) {
4077 fprintf(stderr
, Name
": not enough space after merge (%llu < %llu)\n",
4082 *freesize
= maxsize
;
4087 static int reserve_space(struct supertype
*st
, int raiddisks
,
4088 unsigned long long size
, int chunk
,
4089 unsigned long long *freesize
)
4091 struct intel_super
*super
= st
->sb
;
4092 struct imsm_super
*mpb
= super
->anchor
;
4097 unsigned long long maxsize
;
4098 unsigned long long minsize
;
4102 /* find the largest common start free region of the possible disks */
4106 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
4112 /* don't activate new spares if we are orom constrained
4113 * and there is already a volume active in the container
4115 if (super
->orom
&& dl
->index
< 0 && mpb
->num_raid_devs
)
4118 e
= get_extents(super
, dl
);
4121 for (i
= 1; e
[i
-1].size
; i
++)
4129 maxsize
= merge_extents(super
, extent_cnt
);
4134 if (cnt
< raiddisks
||
4135 (super
->orom
&& used
&& used
!= raiddisks
) ||
4136 maxsize
< minsize
||
4138 fprintf(stderr
, Name
": not enough devices with space to create array.\n");
4139 return 0; /* No enough free spaces large enough */
4151 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
4153 dl
->raiddisk
= cnt
++;
4160 static int validate_geometry_imsm(struct supertype
*st
, int level
, int layout
,
4161 int raiddisks
, int chunk
, unsigned long long size
,
4162 char *dev
, unsigned long long *freesize
,
4169 /* if given unused devices create a container
4170 * if given given devices in a container create a member volume
4172 if (level
== LEVEL_CONTAINER
) {
4173 /* Must be a fresh device to add to a container */
4174 return validate_geometry_imsm_container(st
, level
, layout
,
4175 raiddisks
, chunk
, size
,
4181 if (st
->sb
&& freesize
) {
4182 /* we are being asked to automatically layout a
4183 * new volume based on the current contents of
4184 * the container. If the the parameters can be
4185 * satisfied reserve_space will record the disks,
4186 * start offset, and size of the volume to be
4187 * created. add_to_super and getinfo_super
4188 * detect when autolayout is in progress.
4190 if (!validate_geometry_imsm_orom(st
->sb
, level
, layout
,
4194 return reserve_space(st
, raiddisks
, size
, chunk
, freesize
);
4199 /* creating in a given container */
4200 return validate_geometry_imsm_volume(st
, level
, layout
,
4201 raiddisks
, chunk
, size
,
4202 dev
, freesize
, verbose
);
4205 /* This device needs to be a device in an 'imsm' container */
4206 fd
= open(dev
, O_RDONLY
|O_EXCL
, 0);
4210 Name
": Cannot create this array on device %s\n",
4215 if (errno
!= EBUSY
|| (fd
= open(dev
, O_RDONLY
, 0)) < 0) {
4217 fprintf(stderr
, Name
": Cannot open %s: %s\n",
4218 dev
, strerror(errno
));
4221 /* Well, it is in use by someone, maybe an 'imsm' container. */
4222 cfd
= open_container(fd
);
4226 fprintf(stderr
, Name
": Cannot use %s: It is busy\n",
4230 sra
= sysfs_read(cfd
, 0, GET_VERSION
);
4231 if (sra
&& sra
->array
.major_version
== -1 &&
4232 strcmp(sra
->text_version
, "imsm") == 0)
4236 /* This is a member of a imsm container. Load the container
4237 * and try to create a volume
4239 struct intel_super
*super
;
4241 if (load_super_imsm_all(st
, cfd
, (void **) &super
, NULL
) == 0) {
4243 st
->container_dev
= fd2devnum(cfd
);
4245 return validate_geometry_imsm_volume(st
, level
, layout
,
4253 fprintf(stderr
, Name
": failed container membership check\n");
4259 static void default_geometry_imsm(struct supertype
*st
, int *level
, int *layout
, int *chunk
)
4261 struct intel_super
*super
= st
->sb
;
4263 if (level
&& *level
== UnSet
)
4264 *level
= LEVEL_CONTAINER
;
4266 if (level
&& layout
&& *layout
== UnSet
)
4267 *layout
= imsm_level_to_layout(*level
);
4269 if (chunk
&& (*chunk
== UnSet
|| *chunk
== 0) &&
4270 super
&& super
->orom
)
4271 *chunk
= imsm_orom_default_chunk(super
->orom
);
4274 static void handle_missing(struct intel_super
*super
, struct imsm_dev
*dev
);
4276 static int kill_subarray_imsm(struct supertype
*st
)
4278 /* remove the subarray currently referenced by ->current_vol */
4280 struct intel_dev
**dp
;
4281 struct intel_super
*super
= st
->sb
;
4282 __u8 current_vol
= super
->current_vol
;
4283 struct imsm_super
*mpb
= super
->anchor
;
4285 if (super
->current_vol
< 0)
4287 super
->current_vol
= -1; /* invalidate subarray cursor */
4289 /* block deletions that would change the uuid of active subarrays
4291 * FIXME when immutable ids are available, but note that we'll
4292 * also need to fixup the invalidated/active subarray indexes in
4295 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
4298 if (i
< current_vol
)
4300 sprintf(subarray
, "%u", i
);
4301 if (is_subarray_active(subarray
, st
->devname
)) {
4303 Name
": deleting subarray-%d would change the UUID of active subarray-%d, aborting\n",
4310 if (st
->update_tail
) {
4311 struct imsm_update_kill_array
*u
= malloc(sizeof(*u
));
4315 u
->type
= update_kill_array
;
4316 u
->dev_idx
= current_vol
;
4317 append_metadata_update(st
, u
, sizeof(*u
));
4322 for (dp
= &super
->devlist
; *dp
;)
4323 if ((*dp
)->index
== current_vol
) {
4326 handle_missing(super
, (*dp
)->dev
);
4327 if ((*dp
)->index
> current_vol
)
4332 /* no more raid devices, all active components are now spares,
4333 * but of course failed are still failed
4335 if (--mpb
->num_raid_devs
== 0) {
4338 for (d
= super
->disks
; d
; d
= d
->next
)
4339 if (d
->index
> -2) {
4341 d
->disk
.status
= SPARE_DISK
;
4345 super
->updates_pending
++;
4350 static int update_subarray_imsm(struct supertype
*st
, char *subarray
,
4351 char *update
, struct mddev_ident
*ident
)
4353 /* update the subarray currently referenced by ->current_vol */
4354 struct intel_super
*super
= st
->sb
;
4355 struct imsm_super
*mpb
= super
->anchor
;
4357 if (strcmp(update
, "name") == 0) {
4358 char *name
= ident
->name
;
4362 if (is_subarray_active(subarray
, st
->devname
)) {
4364 Name
": Unable to update name of active subarray\n");
4368 if (!check_name(super
, name
, 0))
4371 vol
= strtoul(subarray
, &ep
, 10);
4372 if (*ep
!= '\0' || vol
>= super
->anchor
->num_raid_devs
)
4375 if (st
->update_tail
) {
4376 struct imsm_update_rename_array
*u
= malloc(sizeof(*u
));
4380 u
->type
= update_rename_array
;
4382 snprintf((char *) u
->name
, MAX_RAID_SERIAL_LEN
, "%s", name
);
4383 append_metadata_update(st
, u
, sizeof(*u
));
4385 struct imsm_dev
*dev
;
4388 dev
= get_imsm_dev(super
, vol
);
4389 snprintf((char *) dev
->volume
, MAX_RAID_SERIAL_LEN
, "%s", name
);
4390 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
4391 dev
= get_imsm_dev(super
, i
);
4392 handle_missing(super
, dev
);
4394 super
->updates_pending
++;
4401 #endif /* MDASSEMBLE */
4403 static int is_gen_migration(struct imsm_dev
*dev
)
4405 if (!dev
->vol
.migr_state
)
4408 if (migr_type(dev
) == MIGR_GEN_MIGR
)
4414 static int is_rebuilding(struct imsm_dev
*dev
)
4416 struct imsm_map
*migr_map
;
4418 if (!dev
->vol
.migr_state
)
4421 if (migr_type(dev
) != MIGR_REBUILD
)
4424 migr_map
= get_imsm_map(dev
, 1);
4426 if (migr_map
->map_state
== IMSM_T_STATE_DEGRADED
)
4432 static void update_recovery_start(struct imsm_dev
*dev
, struct mdinfo
*array
)
4434 struct mdinfo
*rebuild
= NULL
;
4438 if (!is_rebuilding(dev
))
4441 /* Find the rebuild target, but punt on the dual rebuild case */
4442 for (d
= array
->devs
; d
; d
= d
->next
)
4443 if (d
->recovery_start
== 0) {
4450 /* (?) none of the disks are marked with
4451 * IMSM_ORD_REBUILD, so assume they are missing and the
4452 * disk_ord_tbl was not correctly updated
4454 dprintf("%s: failed to locate out-of-sync disk\n", __func__
);
4458 units
= __le32_to_cpu(dev
->vol
.curr_migr_unit
);
4459 rebuild
->recovery_start
= units
* blocks_per_migr_unit(dev
);
4463 static struct mdinfo
*container_content_imsm(struct supertype
*st
, char *subarray
)
4465 /* Given a container loaded by load_super_imsm_all,
4466 * extract information about all the arrays into
4468 * If 'subarray' is given, just extract info about that array.
4470 * For each imsm_dev create an mdinfo, fill it in,
4471 * then look for matching devices in super->disks
4472 * and create appropriate device mdinfo.
4474 struct intel_super
*super
= st
->sb
;
4475 struct imsm_super
*mpb
= super
->anchor
;
4476 struct mdinfo
*rest
= NULL
;
4479 /* do not assemble arrays that might have bad blocks */
4480 if (imsm_bbm_log_size(super
->anchor
)) {
4481 fprintf(stderr
, Name
": BBM log found in metadata. "
4482 "Cannot activate array(s).\n");
4486 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
4487 struct imsm_dev
*dev
;
4488 struct imsm_map
*map
;
4489 struct mdinfo
*this;
4494 (i
!= strtoul(subarray
, &ep
, 10) || *ep
!= '\0'))
4497 dev
= get_imsm_dev(super
, i
);
4498 map
= get_imsm_map(dev
, 0);
4500 /* do not publish arrays that are in the middle of an
4501 * unsupported migration
4503 if (dev
->vol
.migr_state
&&
4504 (migr_type(dev
) == MIGR_STATE_CHANGE
)) {
4505 fprintf(stderr
, Name
": cannot assemble volume '%.16s':"
4506 " unsupported migration in progress\n",
4511 this = malloc(sizeof(*this));
4513 fprintf(stderr
, Name
": failed to allocate %zu bytes\n",
4517 memset(this, 0, sizeof(*this));
4520 super
->current_vol
= i
;
4521 getinfo_super_imsm_volume(st
, this, NULL
);
4522 for (slot
= 0 ; slot
< map
->num_members
; slot
++) {
4523 unsigned long long recovery_start
;
4524 struct mdinfo
*info_d
;
4531 idx
= get_imsm_disk_idx(dev
, slot
);
4532 ord
= get_imsm_ord_tbl_ent(dev
, slot
);
4533 for (d
= super
->disks
; d
; d
= d
->next
)
4534 if (d
->index
== idx
)
4537 recovery_start
= MaxSector
;
4540 if (d
&& is_failed(&d
->disk
))
4542 if (ord
& IMSM_ORD_REBUILD
)
4546 * if we skip some disks the array will be assmebled degraded;
4547 * reset resync start to avoid a dirty-degraded
4548 * situation when performing the intial sync
4550 * FIXME handle dirty degraded
4552 if ((skip
|| recovery_start
== 0) && !dev
->vol
.dirty
)
4553 this->resync_start
= MaxSector
;
4557 info_d
= calloc(1, sizeof(*info_d
));
4559 fprintf(stderr
, Name
": failed to allocate disk"
4560 " for volume %.16s\n", dev
->volume
);
4561 info_d
= this->devs
;
4563 struct mdinfo
*d
= info_d
->next
;
4572 info_d
->next
= this->devs
;
4573 this->devs
= info_d
;
4575 info_d
->disk
.number
= d
->index
;
4576 info_d
->disk
.major
= d
->major
;
4577 info_d
->disk
.minor
= d
->minor
;
4578 info_d
->disk
.raid_disk
= slot
;
4579 info_d
->recovery_start
= recovery_start
;
4581 if (info_d
->recovery_start
== MaxSector
)
4582 this->array
.working_disks
++;
4584 info_d
->events
= __le32_to_cpu(mpb
->generation_num
);
4585 info_d
->data_offset
= __le32_to_cpu(map
->pba_of_lba0
);
4586 info_d
->component_size
= __le32_to_cpu(map
->blocks_per_member
);
4588 /* now that the disk list is up-to-date fixup recovery_start */
4589 update_recovery_start(dev
, this);
4597 static __u8
imsm_check_degraded(struct intel_super
*super
, struct imsm_dev
*dev
, int failed
)
4599 struct imsm_map
*map
= get_imsm_map(dev
, 0);
4602 return map
->map_state
== IMSM_T_STATE_UNINITIALIZED
?
4603 IMSM_T_STATE_UNINITIALIZED
: IMSM_T_STATE_NORMAL
;
4605 switch (get_imsm_raid_level(map
)) {
4607 return IMSM_T_STATE_FAILED
;
4610 if (failed
< map
->num_members
)
4611 return IMSM_T_STATE_DEGRADED
;
4613 return IMSM_T_STATE_FAILED
;
4618 * check to see if any mirrors have failed, otherwise we
4619 * are degraded. Even numbered slots are mirrored on
4623 /* gcc -Os complains that this is unused */
4624 int insync
= insync
;
4626 for (i
= 0; i
< map
->num_members
; i
++) {
4627 __u32 ord
= get_imsm_ord_tbl_ent(dev
, i
);
4628 int idx
= ord_to_idx(ord
);
4629 struct imsm_disk
*disk
;
4631 /* reset the potential in-sync count on even-numbered
4632 * slots. num_copies is always 2 for imsm raid10
4637 disk
= get_imsm_disk(super
, idx
);
4638 if (!disk
|| is_failed(disk
) || ord
& IMSM_ORD_REBUILD
)
4641 /* no in-sync disks left in this mirror the
4645 return IMSM_T_STATE_FAILED
;
4648 return IMSM_T_STATE_DEGRADED
;
4652 return IMSM_T_STATE_DEGRADED
;
4654 return IMSM_T_STATE_FAILED
;
4660 return map
->map_state
;
4663 static int imsm_count_failed(struct intel_super
*super
, struct imsm_dev
*dev
)
4667 struct imsm_disk
*disk
;
4668 struct imsm_map
*map
= get_imsm_map(dev
, 0);
4669 struct imsm_map
*prev
= get_imsm_map(dev
, dev
->vol
.migr_state
);
4673 /* at the beginning of migration we set IMSM_ORD_REBUILD on
4674 * disks that are being rebuilt. New failures are recorded to
4675 * map[0]. So we look through all the disks we started with and
4676 * see if any failures are still present, or if any new ones
4679 * FIXME add support for online capacity expansion and
4680 * raid-level-migration
4682 for (i
= 0; i
< prev
->num_members
; i
++) {
4683 ord
= __le32_to_cpu(prev
->disk_ord_tbl
[i
]);
4684 ord
|= __le32_to_cpu(map
->disk_ord_tbl
[i
]);
4685 idx
= ord_to_idx(ord
);
4687 disk
= get_imsm_disk(super
, idx
);
4688 if (!disk
|| is_failed(disk
) || ord
& IMSM_ORD_REBUILD
)
4696 static int imsm_open_new(struct supertype
*c
, struct active_array
*a
,
4699 struct intel_super
*super
= c
->sb
;
4700 struct imsm_super
*mpb
= super
->anchor
;
4702 if (atoi(inst
) >= mpb
->num_raid_devs
) {
4703 fprintf(stderr
, "%s: subarry index %d, out of range\n",
4704 __func__
, atoi(inst
));
4708 dprintf("imsm: open_new %s\n", inst
);
4709 a
->info
.container_member
= atoi(inst
);
4713 static int is_resyncing(struct imsm_dev
*dev
)
4715 struct imsm_map
*migr_map
;
4717 if (!dev
->vol
.migr_state
)
4720 if (migr_type(dev
) == MIGR_INIT
||
4721 migr_type(dev
) == MIGR_REPAIR
)
4724 migr_map
= get_imsm_map(dev
, 1);
4726 if (migr_map
->map_state
== IMSM_T_STATE_NORMAL
)
4732 /* return true if we recorded new information */
4733 static int mark_failure(struct imsm_dev
*dev
, struct imsm_disk
*disk
, int idx
)
4737 struct imsm_map
*map
;
4739 /* new failures are always set in map[0] */
4740 map
= get_imsm_map(dev
, 0);
4742 slot
= get_imsm_disk_slot(map
, idx
);
4746 ord
= __le32_to_cpu(map
->disk_ord_tbl
[slot
]);
4747 if (is_failed(disk
) && (ord
& IMSM_ORD_REBUILD
))
4750 disk
->status
|= FAILED_DISK
;
4751 disk
->status
&= ~CONFIGURED_DISK
;
4752 set_imsm_ord_tbl_ent(map
, slot
, idx
| IMSM_ORD_REBUILD
);
4753 if (map
->failed_disk_num
== 0xff)
4754 map
->failed_disk_num
= slot
;
4758 static void mark_missing(struct imsm_dev
*dev
, struct imsm_disk
*disk
, int idx
)
4760 mark_failure(dev
, disk
, idx
);
4762 if (disk
->scsi_id
== __cpu_to_le32(~(__u32
)0))
4765 disk
->scsi_id
= __cpu_to_le32(~(__u32
)0);
4766 memmove(&disk
->serial
[0], &disk
->serial
[1], MAX_RAID_SERIAL_LEN
- 1);
4769 static void handle_missing(struct intel_super
*super
, struct imsm_dev
*dev
)
4775 if (!super
->missing
)
4777 failed
= imsm_count_failed(super
, dev
);
4778 map_state
= imsm_check_degraded(super
, dev
, failed
);
4780 dprintf("imsm: mark missing\n");
4781 end_migration(dev
, map_state
);
4782 for (dl
= super
->missing
; dl
; dl
= dl
->next
)
4783 mark_missing(dev
, &dl
->disk
, dl
->index
);
4784 super
->updates_pending
++;
4787 static void imsm_set_disk(struct active_array
*a
, int n
, int state
);
4789 /* Handle dirty -> clean transititions and resync. Degraded and rebuild
4790 * states are handled in imsm_set_disk() with one exception, when a
4791 * resync is stopped due to a new failure this routine will set the
4792 * 'degraded' state for the array.
4794 static int imsm_set_array_state(struct active_array
*a
, int consistent
)
4796 int inst
= a
->info
.container_member
;
4797 struct intel_super
*super
= a
->container
->sb
;
4798 struct imsm_dev
*dev
= get_imsm_dev(super
, inst
);
4799 struct imsm_map
*map
= get_imsm_map(dev
, 0);
4800 int failed
= imsm_count_failed(super
, dev
);
4801 __u8 map_state
= imsm_check_degraded(super
, dev
, failed
);
4802 __u32 blocks_per_unit
;
4804 /* before we activate this array handle any missing disks */
4805 if (consistent
== 2)
4806 handle_missing(super
, dev
);
4808 if (consistent
== 2 &&
4809 (!is_resync_complete(&a
->info
) ||
4810 map_state
!= IMSM_T_STATE_NORMAL
||
4811 dev
->vol
.migr_state
))
4814 if (is_resync_complete(&a
->info
)) {
4815 /* complete intialization / resync,
4816 * recovery and interrupted recovery is completed in
4819 if (is_resyncing(dev
)) {
4820 dprintf("imsm: mark resync done\n");
4821 end_migration(dev
, map_state
);
4822 super
->updates_pending
++;
4823 a
->last_checkpoint
= 0;
4825 } else if (!is_resyncing(dev
) && !failed
) {
4826 /* mark the start of the init process if nothing is failed */
4827 dprintf("imsm: mark resync start\n");
4828 if (map
->map_state
== IMSM_T_STATE_UNINITIALIZED
)
4829 migrate(dev
, IMSM_T_STATE_NORMAL
, MIGR_INIT
);
4831 migrate(dev
, IMSM_T_STATE_NORMAL
, MIGR_REPAIR
);
4832 super
->updates_pending
++;
4835 /* check if we can update curr_migr_unit from resync_start, recovery_start */
4836 blocks_per_unit
= blocks_per_migr_unit(dev
);
4837 if (blocks_per_unit
) {
4841 units
= a
->last_checkpoint
/ blocks_per_unit
;
4844 /* check that we did not overflow 32-bits, and that
4845 * curr_migr_unit needs updating
4847 if (units32
== units
&&
4848 __le32_to_cpu(dev
->vol
.curr_migr_unit
) != units32
) {
4849 dprintf("imsm: mark checkpoint (%u)\n", units32
);
4850 dev
->vol
.curr_migr_unit
= __cpu_to_le32(units32
);
4851 super
->updates_pending
++;
4855 /* mark dirty / clean */
4856 if (dev
->vol
.dirty
!= !consistent
) {
4857 dprintf("imsm: mark '%s'\n", consistent
? "clean" : "dirty");
4862 super
->updates_pending
++;
4865 /* finalize online capacity expansion/reshape */
4866 if ((a
->curr_action
!= reshape
) &&
4867 (a
->prev_action
== reshape
)) {
4870 for (mdi
= a
->info
.devs
; mdi
; mdi
= mdi
->next
)
4871 imsm_set_disk(a
, mdi
->disk
.raid_disk
, mdi
->curr_state
);
4877 static void imsm_set_disk(struct active_array
*a
, int n
, int state
)
4879 int inst
= a
->info
.container_member
;
4880 struct intel_super
*super
= a
->container
->sb
;
4881 struct imsm_dev
*dev
= get_imsm_dev(super
, inst
);
4882 struct imsm_map
*map
= get_imsm_map(dev
, 0);
4883 struct imsm_disk
*disk
;
4888 if (n
> map
->num_members
)
4889 fprintf(stderr
, "imsm: set_disk %d out of range 0..%d\n",
4890 n
, map
->num_members
- 1);
4895 dprintf("imsm: set_disk %d:%x\n", n
, state
);
4897 ord
= get_imsm_ord_tbl_ent(dev
, n
);
4898 disk
= get_imsm_disk(super
, ord_to_idx(ord
));
4900 /* check for new failures */
4901 if (state
& DS_FAULTY
) {
4902 if (mark_failure(dev
, disk
, ord_to_idx(ord
)))
4903 super
->updates_pending
++;
4906 /* check if in_sync */
4907 if (state
& DS_INSYNC
&& ord
& IMSM_ORD_REBUILD
&& is_rebuilding(dev
)) {
4908 struct imsm_map
*migr_map
= get_imsm_map(dev
, 1);
4910 set_imsm_ord_tbl_ent(migr_map
, n
, ord_to_idx(ord
));
4911 super
->updates_pending
++;
4914 failed
= imsm_count_failed(super
, dev
);
4915 map_state
= imsm_check_degraded(super
, dev
, failed
);
4917 /* check if recovery complete, newly degraded, or failed */
4918 if (map_state
== IMSM_T_STATE_NORMAL
&& is_rebuilding(dev
)) {
4919 end_migration(dev
, map_state
);
4920 map
= get_imsm_map(dev
, 0);
4921 map
->failed_disk_num
= ~0;
4922 super
->updates_pending
++;
4923 a
->last_checkpoint
= 0;
4924 } else if (map_state
== IMSM_T_STATE_DEGRADED
&&
4925 map
->map_state
!= map_state
&&
4926 !dev
->vol
.migr_state
) {
4927 dprintf("imsm: mark degraded\n");
4928 map
->map_state
= map_state
;
4929 super
->updates_pending
++;
4930 a
->last_checkpoint
= 0;
4931 } else if (map_state
== IMSM_T_STATE_FAILED
&&
4932 map
->map_state
!= map_state
) {
4933 dprintf("imsm: mark failed\n");
4934 end_migration(dev
, map_state
);
4935 super
->updates_pending
++;
4936 a
->last_checkpoint
= 0;
4937 } else if (is_gen_migration(dev
)) {
4938 dprintf("imsm: Detected General Migration in state: ");
4939 if (map_state
== IMSM_T_STATE_NORMAL
) {
4940 end_migration(dev
, map_state
);
4941 map
= get_imsm_map(dev
, 0);
4942 map
->failed_disk_num
= ~0;
4943 dprintf("normal\n");
4945 if (map_state
== IMSM_T_STATE_DEGRADED
) {
4946 printf("degraded\n");
4947 end_migration(dev
, map_state
);
4949 dprintf("failed\n");
4951 map
->map_state
= map_state
;
4953 super
->updates_pending
++;
4957 static int store_imsm_mpb(int fd
, struct imsm_super
*mpb
)
4960 __u32 mpb_size
= __le32_to_cpu(mpb
->mpb_size
);
4961 unsigned long long dsize
;
4962 unsigned long long sectors
;
4964 get_dev_size(fd
, NULL
, &dsize
);
4966 if (mpb_size
> 512) {
4967 /* -1 to account for anchor */
4968 sectors
= mpb_sectors(mpb
) - 1;
4970 /* write the extended mpb to the sectors preceeding the anchor */
4971 if (lseek64(fd
, dsize
- (512 * (2 + sectors
)), SEEK_SET
) < 0)
4974 if ((unsigned long long)write(fd
, buf
+ 512, 512 * sectors
)
4979 /* first block is stored on second to last sector of the disk */
4980 if (lseek64(fd
, dsize
- (512 * 2), SEEK_SET
) < 0)
4983 if (write(fd
, buf
, 512) != 512)
4989 static void imsm_sync_metadata(struct supertype
*container
)
4991 struct intel_super
*super
= container
->sb
;
4993 dprintf("sync metadata: %d\n", super
->updates_pending
);
4994 if (!super
->updates_pending
)
4997 write_super_imsm(container
, 0);
4999 super
->updates_pending
= 0;
5002 static struct dl
*imsm_readd(struct intel_super
*super
, int idx
, struct active_array
*a
)
5004 struct imsm_dev
*dev
= get_imsm_dev(super
, a
->info
.container_member
);
5005 int i
= get_imsm_disk_idx(dev
, idx
);
5008 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
5012 if (dl
&& is_failed(&dl
->disk
))
5016 dprintf("%s: found %x:%x\n", __func__
, dl
->major
, dl
->minor
);
5021 static struct dl
*imsm_add_spare(struct intel_super
*super
, int slot
,
5022 struct active_array
*a
, int activate_new
,
5023 struct mdinfo
*additional_test_list
)
5025 struct imsm_dev
*dev
= get_imsm_dev(super
, a
->info
.container_member
);
5026 int idx
= get_imsm_disk_idx(dev
, slot
);
5027 struct imsm_super
*mpb
= super
->anchor
;
5028 struct imsm_map
*map
;
5029 unsigned long long pos
;
5034 __u32 array_start
= 0;
5035 __u32 array_end
= 0;
5038 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
5039 /* If in this array, skip */
5040 for (d
= a
->info
.devs
; d
; d
= d
->next
)
5041 if (d
->state_fd
>= 0 &&
5042 d
->disk
.major
== dl
->major
&&
5043 d
->disk
.minor
== dl
->minor
) {
5044 dprintf("%x:%x already in array\n",
5045 dl
->major
, dl
->minor
);
5050 while (additional_test_list
) {
5051 if (additional_test_list
->disk
.major
== dl
->major
&&
5052 additional_test_list
->disk
.minor
== dl
->minor
) {
5053 dprintf("%x:%x already in additional test list\n",
5054 dl
->major
, dl
->minor
);
5057 additional_test_list
= additional_test_list
->next
;
5059 if (additional_test_list
)
5062 /* skip in use or failed drives */
5063 if (is_failed(&dl
->disk
) || idx
== dl
->index
||
5065 dprintf("%x:%x status (failed: %d index: %d)\n",
5066 dl
->major
, dl
->minor
, is_failed(&dl
->disk
), idx
);
5070 /* skip pure spares when we are looking for partially
5071 * assimilated drives
5073 if (dl
->index
== -1 && !activate_new
)
5076 /* Does this unused device have the requisite free space?
5077 * It needs to be able to cover all member volumes
5079 ex
= get_extents(super
, dl
);
5081 dprintf("cannot get extents\n");
5084 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
5085 dev
= get_imsm_dev(super
, i
);
5086 map
= get_imsm_map(dev
, 0);
5088 /* check if this disk is already a member of
5091 if (get_imsm_disk_slot(map
, dl
->index
) >= 0)
5097 array_start
= __le32_to_cpu(map
->pba_of_lba0
);
5098 array_end
= array_start
+
5099 __le32_to_cpu(map
->blocks_per_member
) - 1;
5102 /* check that we can start at pba_of_lba0 with
5103 * blocks_per_member of space
5105 if (array_start
>= pos
&& array_end
< ex
[j
].start
) {
5109 pos
= ex
[j
].start
+ ex
[j
].size
;
5111 } while (ex
[j
-1].size
);
5118 if (i
< mpb
->num_raid_devs
) {
5119 dprintf("%x:%x does not have %u to %u available\n",
5120 dl
->major
, dl
->minor
, array_start
, array_end
);
5131 static int imsm_rebuild_allowed(struct supertype
*cont
, int dev_idx
, int failed
)
5133 struct imsm_dev
*dev2
;
5134 struct imsm_map
*map
;
5140 dev2
= get_imsm_dev(cont
->sb
, dev_idx
);
5142 state
= imsm_check_degraded(cont
->sb
, dev2
, failed
);
5143 if (state
== IMSM_T_STATE_FAILED
) {
5144 map
= get_imsm_map(dev2
, 0);
5147 for (slot
= 0; slot
< map
->num_members
; slot
++) {
5149 * Check if failed disks are deleted from intel
5150 * disk list or are marked to be deleted
5152 idx
= get_imsm_disk_idx(dev2
, slot
);
5153 idisk
= get_imsm_dl_disk(cont
->sb
, idx
);
5155 * Do not rebuild the array if failed disks
5156 * from failed sub-array are not removed from
5160 is_failed(&idisk
->disk
) &&
5161 (idisk
->action
!= DISK_REMOVE
))
5169 static struct mdinfo
*imsm_activate_spare(struct active_array
*a
,
5170 struct metadata_update
**updates
)
5173 * Find a device with unused free space and use it to replace a
5174 * failed/vacant region in an array. We replace failed regions one a
5175 * array at a time. The result is that a new spare disk will be added
5176 * to the first failed array and after the monitor has finished
5177 * propagating failures the remainder will be consumed.
5179 * FIXME add a capability for mdmon to request spares from another
5183 struct intel_super
*super
= a
->container
->sb
;
5184 int inst
= a
->info
.container_member
;
5185 struct imsm_dev
*dev
= get_imsm_dev(super
, inst
);
5186 struct imsm_map
*map
= get_imsm_map(dev
, 0);
5187 int failed
= a
->info
.array
.raid_disks
;
5188 struct mdinfo
*rv
= NULL
;
5191 struct metadata_update
*mu
;
5193 struct imsm_update_activate_spare
*u
;
5198 for (d
= a
->info
.devs
; d
; d
= d
->next
) {
5199 if ((d
->curr_state
& DS_FAULTY
) &&
5201 /* wait for Removal to happen */
5203 if (d
->state_fd
>= 0)
5207 dprintf("imsm: activate spare: inst=%d failed=%d (%d) level=%d\n",
5208 inst
, failed
, a
->info
.array
.raid_disks
, a
->info
.array
.level
);
5209 if (imsm_check_degraded(super
, dev
, failed
) != IMSM_T_STATE_DEGRADED
)
5213 * If there are any failed disks check state of the other volume.
5214 * Block rebuild if the another one is failed until failed disks
5215 * are removed from container.
5218 dprintf("found failed disks in %s, check if there another"
5219 "failed sub-array.\n",
5221 /* check if states of the other volumes allow for rebuild */
5222 for (i
= 0; i
< super
->anchor
->num_raid_devs
; i
++) {
5224 allowed
= imsm_rebuild_allowed(a
->container
,
5232 /* For each slot, if it is not working, find a spare */
5233 for (i
= 0; i
< a
->info
.array
.raid_disks
; i
++) {
5234 for (d
= a
->info
.devs
; d
; d
= d
->next
)
5235 if (d
->disk
.raid_disk
== i
)
5237 dprintf("found %d: %p %x\n", i
, d
, d
?d
->curr_state
:0);
5238 if (d
&& (d
->state_fd
>= 0))
5242 * OK, this device needs recovery. Try to re-add the
5243 * previous occupant of this slot, if this fails see if
5244 * we can continue the assimilation of a spare that was
5245 * partially assimilated, finally try to activate a new
5248 dl
= imsm_readd(super
, i
, a
);
5250 dl
= imsm_add_spare(super
, i
, a
, 0, NULL
);
5252 dl
= imsm_add_spare(super
, i
, a
, 1, NULL
);
5256 /* found a usable disk with enough space */
5257 di
= malloc(sizeof(*di
));
5260 memset(di
, 0, sizeof(*di
));
5262 /* dl->index will be -1 in the case we are activating a
5263 * pristine spare. imsm_process_update() will create a
5264 * new index in this case. Once a disk is found to be
5265 * failed in all member arrays it is kicked from the
5268 di
->disk
.number
= dl
->index
;
5270 /* (ab)use di->devs to store a pointer to the device
5273 di
->devs
= (struct mdinfo
*) dl
;
5275 di
->disk
.raid_disk
= i
;
5276 di
->disk
.major
= dl
->major
;
5277 di
->disk
.minor
= dl
->minor
;
5279 di
->recovery_start
= 0;
5280 di
->data_offset
= __le32_to_cpu(map
->pba_of_lba0
);
5281 di
->component_size
= a
->info
.component_size
;
5282 di
->container_member
= inst
;
5283 super
->random
= random32();
5287 dprintf("%x:%x to be %d at %llu\n", dl
->major
, dl
->minor
,
5288 i
, di
->data_offset
);
5294 /* No spares found */
5296 /* Now 'rv' has a list of devices to return.
5297 * Create a metadata_update record to update the
5298 * disk_ord_tbl for the array
5300 mu
= malloc(sizeof(*mu
));
5302 mu
->buf
= malloc(sizeof(struct imsm_update_activate_spare
) * num_spares
);
5303 if (mu
->buf
== NULL
) {
5310 struct mdinfo
*n
= rv
->next
;
5319 mu
->len
= sizeof(struct imsm_update_activate_spare
) * num_spares
;
5320 mu
->next
= *updates
;
5321 u
= (struct imsm_update_activate_spare
*) mu
->buf
;
5323 for (di
= rv
; di
; di
= di
->next
) {
5324 u
->type
= update_activate_spare
;
5325 u
->dl
= (struct dl
*) di
->devs
;
5327 u
->slot
= di
->disk
.raid_disk
;
5338 static int disks_overlap(struct intel_super
*super
, int idx
, struct imsm_update_create_array
*u
)
5340 struct imsm_dev
*dev
= get_imsm_dev(super
, idx
);
5341 struct imsm_map
*map
= get_imsm_map(dev
, 0);
5342 struct imsm_map
*new_map
= get_imsm_map(&u
->dev
, 0);
5343 struct disk_info
*inf
= get_disk_info(u
);
5344 struct imsm_disk
*disk
;
5348 for (i
= 0; i
< map
->num_members
; i
++) {
5349 disk
= get_imsm_disk(super
, get_imsm_disk_idx(dev
, i
));
5350 for (j
= 0; j
< new_map
->num_members
; j
++)
5351 if (serialcmp(disk
->serial
, inf
[j
].serial
) == 0)
5359 static struct dl
*get_disk_super(struct intel_super
*super
, int major
, int minor
)
5361 struct dl
*dl
= NULL
;
5362 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
5363 if ((dl
->major
== major
) && (dl
->minor
== minor
))
5368 static int remove_disk_super(struct intel_super
*super
, int major
, int minor
)
5370 struct dl
*prev
= NULL
;
5374 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
5375 if ((dl
->major
== major
) && (dl
->minor
== minor
)) {
5378 prev
->next
= dl
->next
;
5380 super
->disks
= dl
->next
;
5382 __free_imsm_disk(dl
);
5383 dprintf("%s: removed %x:%x\n",
5384 __func__
, major
, minor
);
5392 static void imsm_delete(struct intel_super
*super
, struct dl
**dlp
, unsigned index
);
5394 static int add_remove_disk_update(struct intel_super
*super
)
5396 int check_degraded
= 0;
5397 struct dl
*disk
= NULL
;
5398 /* add/remove some spares to/from the metadata/contrainer */
5399 while (super
->disk_mgmt_list
) {
5400 struct dl
*disk_cfg
;
5402 disk_cfg
= super
->disk_mgmt_list
;
5403 super
->disk_mgmt_list
= disk_cfg
->next
;
5404 disk_cfg
->next
= NULL
;
5406 if (disk_cfg
->action
== DISK_ADD
) {
5407 disk_cfg
->next
= super
->disks
;
5408 super
->disks
= disk_cfg
;
5410 dprintf("%s: added %x:%x\n",
5411 __func__
, disk_cfg
->major
,
5413 } else if (disk_cfg
->action
== DISK_REMOVE
) {
5414 dprintf("Disk remove action processed: %x.%x\n",
5415 disk_cfg
->major
, disk_cfg
->minor
);
5416 disk
= get_disk_super(super
,
5420 /* store action status */
5421 disk
->action
= DISK_REMOVE
;
5422 /* remove spare disks only */
5423 if (disk
->index
== -1) {
5424 remove_disk_super(super
,
5429 /* release allocate disk structure */
5430 __free_imsm_disk(disk_cfg
);
5433 return check_degraded
;
5436 static void imsm_process_update(struct supertype
*st
,
5437 struct metadata_update
*update
)
5440 * crack open the metadata_update envelope to find the update record
5441 * update can be one of:
5442 * update_activate_spare - a spare device has replaced a failed
5443 * device in an array, update the disk_ord_tbl. If this disk is
5444 * present in all member arrays then also clear the SPARE_DISK
5447 struct intel_super
*super
= st
->sb
;
5448 struct imsm_super
*mpb
;
5449 enum imsm_update_type type
= *(enum imsm_update_type
*) update
->buf
;
5451 /* update requires a larger buf but the allocation failed */
5452 if (super
->next_len
&& !super
->next_buf
) {
5453 super
->next_len
= 0;
5457 if (super
->next_buf
) {
5458 memcpy(super
->next_buf
, super
->buf
, super
->len
);
5460 super
->len
= super
->next_len
;
5461 super
->buf
= super
->next_buf
;
5463 super
->next_len
= 0;
5464 super
->next_buf
= NULL
;
5467 mpb
= super
->anchor
;
5470 case update_activate_spare
: {
5471 struct imsm_update_activate_spare
*u
= (void *) update
->buf
;
5472 struct imsm_dev
*dev
= get_imsm_dev(super
, u
->array
);
5473 struct imsm_map
*map
= get_imsm_map(dev
, 0);
5474 struct imsm_map
*migr_map
;
5475 struct active_array
*a
;
5476 struct imsm_disk
*disk
;
5481 int victim
= get_imsm_disk_idx(dev
, u
->slot
);
5484 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
5489 fprintf(stderr
, "error: imsm_activate_spare passed "
5490 "an unknown disk (index: %d)\n",
5495 super
->updates_pending
++;
5497 /* count failures (excluding rebuilds and the victim)
5498 * to determine map[0] state
5501 for (i
= 0; i
< map
->num_members
; i
++) {
5504 disk
= get_imsm_disk(super
, get_imsm_disk_idx(dev
, i
));
5505 if (!disk
|| is_failed(disk
))
5509 /* adding a pristine spare, assign a new index */
5510 if (dl
->index
< 0) {
5511 dl
->index
= super
->anchor
->num_disks
;
5512 super
->anchor
->num_disks
++;
5515 disk
->status
|= CONFIGURED_DISK
;
5516 disk
->status
&= ~SPARE_DISK
;
5519 to_state
= imsm_check_degraded(super
, dev
, failed
);
5520 map
->map_state
= IMSM_T_STATE_DEGRADED
;
5521 migrate(dev
, to_state
, MIGR_REBUILD
);
5522 migr_map
= get_imsm_map(dev
, 1);
5523 set_imsm_ord_tbl_ent(map
, u
->slot
, dl
->index
);
5524 set_imsm_ord_tbl_ent(migr_map
, u
->slot
, dl
->index
| IMSM_ORD_REBUILD
);
5526 /* update the family_num to mark a new container
5527 * generation, being careful to record the existing
5528 * family_num in orig_family_num to clean up after
5529 * earlier mdadm versions that neglected to set it.
5531 if (mpb
->orig_family_num
== 0)
5532 mpb
->orig_family_num
= mpb
->family_num
;
5533 mpb
->family_num
+= super
->random
;
5535 /* count arrays using the victim in the metadata */
5537 for (a
= st
->arrays
; a
; a
= a
->next
) {
5538 dev
= get_imsm_dev(super
, a
->info
.container_member
);
5539 map
= get_imsm_map(dev
, 0);
5541 if (get_imsm_disk_slot(map
, victim
) >= 0)
5545 /* delete the victim if it is no longer being
5551 /* We know that 'manager' isn't touching anything,
5552 * so it is safe to delete
5554 for (dlp
= &super
->disks
; *dlp
; dlp
= &(*dlp
)->next
)
5555 if ((*dlp
)->index
== victim
)
5558 /* victim may be on the missing list */
5560 for (dlp
= &super
->missing
; *dlp
; dlp
= &(*dlp
)->next
)
5561 if ((*dlp
)->index
== victim
)
5563 imsm_delete(super
, dlp
, victim
);
5567 case update_create_array
: {
5568 /* someone wants to create a new array, we need to be aware of
5569 * a few races/collisions:
5570 * 1/ 'Create' called by two separate instances of mdadm
5571 * 2/ 'Create' versus 'activate_spare': mdadm has chosen
5572 * devices that have since been assimilated via
5574 * In the event this update can not be carried out mdadm will
5575 * (FIX ME) notice that its update did not take hold.
5577 struct imsm_update_create_array
*u
= (void *) update
->buf
;
5578 struct intel_dev
*dv
;
5579 struct imsm_dev
*dev
;
5580 struct imsm_map
*map
, *new_map
;
5581 unsigned long long start
, end
;
5582 unsigned long long new_start
, new_end
;
5584 struct disk_info
*inf
;
5587 /* handle racing creates: first come first serve */
5588 if (u
->dev_idx
< mpb
->num_raid_devs
) {
5589 dprintf("%s: subarray %d already defined\n",
5590 __func__
, u
->dev_idx
);
5594 /* check update is next in sequence */
5595 if (u
->dev_idx
!= mpb
->num_raid_devs
) {
5596 dprintf("%s: can not create array %d expected index %d\n",
5597 __func__
, u
->dev_idx
, mpb
->num_raid_devs
);
5601 new_map
= get_imsm_map(&u
->dev
, 0);
5602 new_start
= __le32_to_cpu(new_map
->pba_of_lba0
);
5603 new_end
= new_start
+ __le32_to_cpu(new_map
->blocks_per_member
);
5604 inf
= get_disk_info(u
);
5606 /* handle activate_spare versus create race:
5607 * check to make sure that overlapping arrays do not include
5610 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
5611 dev
= get_imsm_dev(super
, i
);
5612 map
= get_imsm_map(dev
, 0);
5613 start
= __le32_to_cpu(map
->pba_of_lba0
);
5614 end
= start
+ __le32_to_cpu(map
->blocks_per_member
);
5615 if ((new_start
>= start
&& new_start
<= end
) ||
5616 (start
>= new_start
&& start
<= new_end
))
5621 if (disks_overlap(super
, i
, u
)) {
5622 dprintf("%s: arrays overlap\n", __func__
);
5627 /* check that prepare update was successful */
5628 if (!update
->space
) {
5629 dprintf("%s: prepare update failed\n", __func__
);
5633 /* check that all disks are still active before committing
5634 * changes. FIXME: could we instead handle this by creating a
5635 * degraded array? That's probably not what the user expects,
5636 * so better to drop this update on the floor.
5638 for (i
= 0; i
< new_map
->num_members
; i
++) {
5639 dl
= serial_to_dl(inf
[i
].serial
, super
);
5641 dprintf("%s: disk disappeared\n", __func__
);
5646 super
->updates_pending
++;
5648 /* convert spares to members and fixup ord_tbl */
5649 for (i
= 0; i
< new_map
->num_members
; i
++) {
5650 dl
= serial_to_dl(inf
[i
].serial
, super
);
5651 if (dl
->index
== -1) {
5652 dl
->index
= mpb
->num_disks
;
5654 dl
->disk
.status
|= CONFIGURED_DISK
;
5655 dl
->disk
.status
&= ~SPARE_DISK
;
5657 set_imsm_ord_tbl_ent(new_map
, i
, dl
->index
);
5662 update
->space
= NULL
;
5663 imsm_copy_dev(dev
, &u
->dev
);
5664 dv
->index
= u
->dev_idx
;
5665 dv
->next
= super
->devlist
;
5666 super
->devlist
= dv
;
5667 mpb
->num_raid_devs
++;
5669 imsm_update_version_info(super
);
5672 /* mdmon knows how to release update->space, but not
5673 * ((struct intel_dev *) update->space)->dev
5675 if (update
->space
) {
5681 case update_kill_array
: {
5682 struct imsm_update_kill_array
*u
= (void *) update
->buf
;
5683 int victim
= u
->dev_idx
;
5684 struct active_array
*a
;
5685 struct intel_dev
**dp
;
5686 struct imsm_dev
*dev
;
5688 /* sanity check that we are not affecting the uuid of
5689 * active arrays, or deleting an active array
5691 * FIXME when immutable ids are available, but note that
5692 * we'll also need to fixup the invalidated/active
5693 * subarray indexes in mdstat
5695 for (a
= st
->arrays
; a
; a
= a
->next
)
5696 if (a
->info
.container_member
>= victim
)
5698 /* by definition if mdmon is running at least one array
5699 * is active in the container, so checking
5700 * mpb->num_raid_devs is just extra paranoia
5702 dev
= get_imsm_dev(super
, victim
);
5703 if (a
|| !dev
|| mpb
->num_raid_devs
== 1) {
5704 dprintf("failed to delete subarray-%d\n", victim
);
5708 for (dp
= &super
->devlist
; *dp
;)
5709 if ((*dp
)->index
== (unsigned)super
->current_vol
) {
5712 if ((*dp
)->index
> (unsigned)victim
)
5716 mpb
->num_raid_devs
--;
5717 super
->updates_pending
++;
5720 case update_rename_array
: {
5721 struct imsm_update_rename_array
*u
= (void *) update
->buf
;
5722 char name
[MAX_RAID_SERIAL_LEN
+1];
5723 int target
= u
->dev_idx
;
5724 struct active_array
*a
;
5725 struct imsm_dev
*dev
;
5727 /* sanity check that we are not affecting the uuid of
5730 snprintf(name
, MAX_RAID_SERIAL_LEN
, "%s", (char *) u
->name
);
5731 name
[MAX_RAID_SERIAL_LEN
] = '\0';
5732 for (a
= st
->arrays
; a
; a
= a
->next
)
5733 if (a
->info
.container_member
== target
)
5735 dev
= get_imsm_dev(super
, u
->dev_idx
);
5736 if (a
|| !dev
|| !check_name(super
, name
, 1)) {
5737 dprintf("failed to rename subarray-%d\n", target
);
5741 snprintf((char *) dev
->volume
, MAX_RAID_SERIAL_LEN
, "%s", name
);
5742 super
->updates_pending
++;
5745 case update_add_remove_disk
: {
5746 /* we may be able to repair some arrays if disks are
5747 * being added, check teh status of add_remove_disk
5748 * if discs has been added.
5750 if (add_remove_disk_update(super
)) {
5751 struct active_array
*a
;
5753 super
->updates_pending
++;
5754 for (a
= st
->arrays
; a
; a
= a
->next
)
5755 a
->check_degraded
= 1;
5760 fprintf(stderr
, "error: unsuported process update type:"
5761 "(type: %d)\n", type
);
5765 static void imsm_prepare_update(struct supertype
*st
,
5766 struct metadata_update
*update
)
5769 * Allocate space to hold new disk entries, raid-device entries or a new
5770 * mpb if necessary. The manager synchronously waits for updates to
5771 * complete in the monitor, so new mpb buffers allocated here can be
5772 * integrated by the monitor thread without worrying about live pointers
5773 * in the manager thread.
5775 enum imsm_update_type type
= *(enum imsm_update_type
*) update
->buf
;
5776 struct intel_super
*super
= st
->sb
;
5777 struct imsm_super
*mpb
= super
->anchor
;
5782 case update_create_array
: {
5783 struct imsm_update_create_array
*u
= (void *) update
->buf
;
5784 struct intel_dev
*dv
;
5785 struct imsm_dev
*dev
= &u
->dev
;
5786 struct imsm_map
*map
= get_imsm_map(dev
, 0);
5788 struct disk_info
*inf
;
5792 inf
= get_disk_info(u
);
5793 len
= sizeof_imsm_dev(dev
, 1);
5794 /* allocate a new super->devlist entry */
5795 dv
= malloc(sizeof(*dv
));
5797 dv
->dev
= malloc(len
);
5802 update
->space
= NULL
;
5806 /* count how many spares will be converted to members */
5807 for (i
= 0; i
< map
->num_members
; i
++) {
5808 dl
= serial_to_dl(inf
[i
].serial
, super
);
5810 /* hmm maybe it failed?, nothing we can do about
5815 if (count_memberships(dl
, super
) == 0)
5818 len
+= activate
* sizeof(struct imsm_disk
);
5825 /* check if we need a larger metadata buffer */
5826 if (super
->next_buf
)
5827 buf_len
= super
->next_len
;
5829 buf_len
= super
->len
;
5831 if (__le32_to_cpu(mpb
->mpb_size
) + len
> buf_len
) {
5832 /* ok we need a larger buf than what is currently allocated
5833 * if this allocation fails process_update will notice that
5834 * ->next_len is set and ->next_buf is NULL
5836 buf_len
= ROUND_UP(__le32_to_cpu(mpb
->mpb_size
) + len
, 512);
5837 if (super
->next_buf
)
5838 free(super
->next_buf
);
5840 super
->next_len
= buf_len
;
5841 if (posix_memalign(&super
->next_buf
, 512, buf_len
) == 0)
5842 memset(super
->next_buf
, 0, buf_len
);
5844 super
->next_buf
= NULL
;
5848 /* must be called while manager is quiesced */
5849 static void imsm_delete(struct intel_super
*super
, struct dl
**dlp
, unsigned index
)
5851 struct imsm_super
*mpb
= super
->anchor
;
5853 struct imsm_dev
*dev
;
5854 struct imsm_map
*map
;
5855 int i
, j
, num_members
;
5858 dprintf("%s: deleting device[%d] from imsm_super\n",
5861 /* shift all indexes down one */
5862 for (iter
= super
->disks
; iter
; iter
= iter
->next
)
5863 if (iter
->index
> (int)index
)
5865 for (iter
= super
->missing
; iter
; iter
= iter
->next
)
5866 if (iter
->index
> (int)index
)
5869 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
5870 dev
= get_imsm_dev(super
, i
);
5871 map
= get_imsm_map(dev
, 0);
5872 num_members
= map
->num_members
;
5873 for (j
= 0; j
< num_members
; j
++) {
5874 /* update ord entries being careful not to propagate
5875 * ord-flags to the first map
5877 ord
= get_imsm_ord_tbl_ent(dev
, j
);
5879 if (ord_to_idx(ord
) <= index
)
5882 map
= get_imsm_map(dev
, 0);
5883 set_imsm_ord_tbl_ent(map
, j
, ord_to_idx(ord
- 1));
5884 map
= get_imsm_map(dev
, 1);
5886 set_imsm_ord_tbl_ent(map
, j
, ord
- 1);
5891 super
->updates_pending
++;
5893 struct dl
*dl
= *dlp
;
5895 *dlp
= (*dlp
)->next
;
5896 __free_imsm_disk(dl
);
5899 #endif /* MDASSEMBLE */
5901 static char disk_by_path
[] = "/dev/disk/by-path/";
5903 static const char *imsm_get_disk_controller_domain(const char *path
)
5905 struct sys_dev
*list
, *hba
= NULL
;
5906 char disk_path
[PATH_MAX
];
5910 list
= find_driver_devices("pci", "ahci");
5911 for (hba
= list
; hba
; hba
= hba
->next
)
5912 if (devpath_to_vendor(hba
->path
) == 0x8086)
5918 strncpy(disk_path
, disk_by_path
, PATH_MAX
- 1);
5919 strncat(disk_path
, path
, PATH_MAX
- strlen(disk_path
) - 1);
5920 if (stat(disk_path
, &st
) == 0) {
5921 dpath
= devt_to_devpath(st
.st_rdev
);
5923 ahci
= path_attached_to_hba(dpath
, hba
->path
);
5926 dprintf("path: %s(%s) hba: %s attached: %d\n",
5927 path
, dpath
, (hba
) ? hba
->path
: "NULL", ahci
);
5928 free_sys_dev(&list
);
5936 struct superswitch super_imsm
= {
5938 .examine_super
= examine_super_imsm
,
5939 .brief_examine_super
= brief_examine_super_imsm
,
5940 .brief_examine_subarrays
= brief_examine_subarrays_imsm
,
5941 .export_examine_super
= export_examine_super_imsm
,
5942 .detail_super
= detail_super_imsm
,
5943 .brief_detail_super
= brief_detail_super_imsm
,
5944 .write_init_super
= write_init_super_imsm
,
5945 .validate_geometry
= validate_geometry_imsm
,
5946 .add_to_super
= add_to_super_imsm
,
5947 .remove_from_super
= remove_from_super_imsm
,
5948 .detail_platform
= detail_platform_imsm
,
5949 .kill_subarray
= kill_subarray_imsm
,
5950 .update_subarray
= update_subarray_imsm
,
5951 .load_container
= load_container_imsm
,
5953 .match_home
= match_home_imsm
,
5954 .uuid_from_super
= uuid_from_super_imsm
,
5955 .getinfo_super
= getinfo_super_imsm
,
5956 .getinfo_super_disks
= getinfo_super_disks_imsm
,
5957 .update_super
= update_super_imsm
,
5959 .avail_size
= avail_size_imsm
,
5960 .min_acceptable_spare_size
= min_acceptable_spare_size_imsm
,
5962 .compare_super
= compare_super_imsm
,
5964 .load_super
= load_super_imsm
,
5965 .init_super
= init_super_imsm
,
5966 .store_super
= store_super_imsm
,
5967 .free_super
= free_super_imsm
,
5968 .match_metadata_desc
= match_metadata_desc_imsm
,
5969 .container_content
= container_content_imsm
,
5970 .default_geometry
= default_geometry_imsm
,
5971 .get_disk_controller_domain
= imsm_get_disk_controller_domain
,
5978 .open_new
= imsm_open_new
,
5979 .set_array_state
= imsm_set_array_state
,
5980 .set_disk
= imsm_set_disk
,
5981 .sync_metadata
= imsm_sync_metadata
,
5982 .activate_spare
= imsm_activate_spare
,
5983 .process_update
= imsm_process_update
,
5984 .prepare_update
= imsm_prepare_update
,
5985 #endif /* MDASSEMBLE */