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 /* definitions of reshape process types */
288 enum imsm_reshape_type
{
294 /* definition of messages passed to imsm_process_update */
295 enum imsm_update_type
{
296 update_activate_spare
,
300 update_add_remove_disk
,
301 update_reshape_container_disks
,
304 struct imsm_update_activate_spare
{
305 enum imsm_update_type type
;
309 struct imsm_update_activate_spare
*next
;
323 struct imsm_update_reshape
{
324 enum imsm_update_type type
;
327 int new_disks
[1]; /* new_raid_disks - old_raid_disks makedev number */
331 __u8 serial
[MAX_RAID_SERIAL_LEN
];
334 struct imsm_update_create_array
{
335 enum imsm_update_type type
;
340 struct imsm_update_kill_array
{
341 enum imsm_update_type type
;
345 struct imsm_update_rename_array
{
346 enum imsm_update_type type
;
347 __u8 name
[MAX_RAID_SERIAL_LEN
];
351 struct imsm_update_add_remove_disk
{
352 enum imsm_update_type type
;
355 static struct supertype
*match_metadata_desc_imsm(char *arg
)
357 struct supertype
*st
;
359 if (strcmp(arg
, "imsm") != 0 &&
360 strcmp(arg
, "default") != 0
364 st
= malloc(sizeof(*st
));
367 memset(st
, 0, sizeof(*st
));
368 st
->container_dev
= NoMdDev
;
369 st
->ss
= &super_imsm
;
370 st
->max_devs
= IMSM_MAX_DEVICES
;
371 st
->minor_version
= 0;
377 static __u8
*get_imsm_version(struct imsm_super
*mpb
)
379 return &mpb
->sig
[MPB_SIG_LEN
];
383 /* retrieve a disk directly from the anchor when the anchor is known to be
384 * up-to-date, currently only at load time
386 static struct imsm_disk
*__get_imsm_disk(struct imsm_super
*mpb
, __u8 index
)
388 if (index
>= mpb
->num_disks
)
390 return &mpb
->disk
[index
];
393 /* retrieve the disk description based on a index of the disk
396 static struct dl
*get_imsm_dl_disk(struct intel_super
*super
, __u8 index
)
400 for (d
= super
->disks
; d
; d
= d
->next
)
401 if (d
->index
== index
)
406 /* retrieve a disk from the parsed metadata */
407 static struct imsm_disk
*get_imsm_disk(struct intel_super
*super
, __u8 index
)
411 dl
= get_imsm_dl_disk(super
, index
);
418 /* generate a checksum directly from the anchor when the anchor is known to be
419 * up-to-date, currently only at load or write_super after coalescing
421 static __u32
__gen_imsm_checksum(struct imsm_super
*mpb
)
423 __u32 end
= mpb
->mpb_size
/ sizeof(end
);
424 __u32
*p
= (__u32
*) mpb
;
428 sum
+= __le32_to_cpu(*p
);
432 return sum
- __le32_to_cpu(mpb
->check_sum
);
435 static size_t sizeof_imsm_map(struct imsm_map
*map
)
437 return sizeof(struct imsm_map
) + sizeof(__u32
) * (map
->num_members
- 1);
440 struct imsm_map
*get_imsm_map(struct imsm_dev
*dev
, int second_map
)
442 struct imsm_map
*map
= &dev
->vol
.map
[0];
444 if (second_map
&& !dev
->vol
.migr_state
)
446 else if (second_map
) {
449 return ptr
+ sizeof_imsm_map(map
);
455 /* return the size of the device.
456 * migr_state increases the returned size if map[0] were to be duplicated
458 static size_t sizeof_imsm_dev(struct imsm_dev
*dev
, int migr_state
)
460 size_t size
= sizeof(*dev
) - sizeof(struct imsm_map
) +
461 sizeof_imsm_map(get_imsm_map(dev
, 0));
463 /* migrating means an additional map */
464 if (dev
->vol
.migr_state
)
465 size
+= sizeof_imsm_map(get_imsm_map(dev
, 1));
467 size
+= sizeof_imsm_map(get_imsm_map(dev
, 0));
473 /* retrieve disk serial number list from a metadata update */
474 static struct disk_info
*get_disk_info(struct imsm_update_create_array
*update
)
477 struct disk_info
*inf
;
479 inf
= u
+ sizeof(*update
) - sizeof(struct imsm_dev
) +
480 sizeof_imsm_dev(&update
->dev
, 0);
486 static struct imsm_dev
*__get_imsm_dev(struct imsm_super
*mpb
, __u8 index
)
492 if (index
>= mpb
->num_raid_devs
)
495 /* devices start after all disks */
496 offset
= ((void *) &mpb
->disk
[mpb
->num_disks
]) - _mpb
;
498 for (i
= 0; i
<= index
; i
++)
500 return _mpb
+ offset
;
502 offset
+= sizeof_imsm_dev(_mpb
+ offset
, 0);
507 static struct imsm_dev
*get_imsm_dev(struct intel_super
*super
, __u8 index
)
509 struct intel_dev
*dv
;
511 if (index
>= super
->anchor
->num_raid_devs
)
513 for (dv
= super
->devlist
; dv
; dv
= dv
->next
)
514 if (dv
->index
== index
)
522 * == 1 get second map
523 * == -1 than get map according to the current migr_state
525 static __u32
get_imsm_ord_tbl_ent(struct imsm_dev
*dev
,
529 struct imsm_map
*map
;
531 if (second_map
== -1) {
532 if (dev
->vol
.migr_state
)
533 map
= get_imsm_map(dev
, 1);
535 map
= get_imsm_map(dev
, 0);
537 map
= get_imsm_map(dev
, second_map
);
540 /* top byte identifies disk under rebuild */
541 return __le32_to_cpu(map
->disk_ord_tbl
[slot
]);
544 #define ord_to_idx(ord) (((ord) << 8) >> 8)
545 static __u32
get_imsm_disk_idx(struct imsm_dev
*dev
, int slot
, int second_map
)
547 __u32 ord
= get_imsm_ord_tbl_ent(dev
, slot
, second_map
);
549 return ord_to_idx(ord
);
552 static void set_imsm_ord_tbl_ent(struct imsm_map
*map
, int slot
, __u32 ord
)
554 map
->disk_ord_tbl
[slot
] = __cpu_to_le32(ord
);
557 static int get_imsm_disk_slot(struct imsm_map
*map
, unsigned idx
)
562 for (slot
= 0; slot
< map
->num_members
; slot
++) {
563 ord
= __le32_to_cpu(map
->disk_ord_tbl
[slot
]);
564 if (ord_to_idx(ord
) == idx
)
571 static int get_imsm_raid_level(struct imsm_map
*map
)
573 if (map
->raid_level
== 1) {
574 if (map
->num_members
== 2)
580 return map
->raid_level
;
583 static int cmp_extent(const void *av
, const void *bv
)
585 const struct extent
*a
= av
;
586 const struct extent
*b
= bv
;
587 if (a
->start
< b
->start
)
589 if (a
->start
> b
->start
)
594 static int count_memberships(struct dl
*dl
, struct intel_super
*super
)
599 for (i
= 0; i
< super
->anchor
->num_raid_devs
; i
++) {
600 struct imsm_dev
*dev
= get_imsm_dev(super
, i
);
601 struct imsm_map
*map
= get_imsm_map(dev
, 0);
603 if (get_imsm_disk_slot(map
, dl
->index
) >= 0)
610 static struct extent
*get_extents(struct intel_super
*super
, struct dl
*dl
)
612 /* find a list of used extents on the given physical device */
613 struct extent
*rv
, *e
;
615 int memberships
= count_memberships(dl
, super
);
616 __u32 reservation
= MPB_SECTOR_CNT
+ IMSM_RESERVED_SECTORS
;
618 rv
= malloc(sizeof(struct extent
) * (memberships
+ 1));
623 for (i
= 0; i
< super
->anchor
->num_raid_devs
; i
++) {
624 struct imsm_dev
*dev
= get_imsm_dev(super
, i
);
625 struct imsm_map
*map
= get_imsm_map(dev
, 0);
627 if (get_imsm_disk_slot(map
, dl
->index
) >= 0) {
628 e
->start
= __le32_to_cpu(map
->pba_of_lba0
);
629 e
->size
= __le32_to_cpu(map
->blocks_per_member
);
633 qsort(rv
, memberships
, sizeof(*rv
), cmp_extent
);
635 /* determine the start of the metadata
636 * when no raid devices are defined use the default
637 * ...otherwise allow the metadata to truncate the value
638 * as is the case with older versions of imsm
641 struct extent
*last
= &rv
[memberships
- 1];
644 remainder
= __le32_to_cpu(dl
->disk
.total_blocks
) -
645 (last
->start
+ last
->size
);
646 /* round down to 1k block to satisfy precision of the kernel
650 /* make sure remainder is still sane */
651 if (remainder
< (unsigned)ROUND_UP(super
->len
, 512) >> 9)
652 remainder
= ROUND_UP(super
->len
, 512) >> 9;
653 if (reservation
> remainder
)
654 reservation
= remainder
;
656 e
->start
= __le32_to_cpu(dl
->disk
.total_blocks
) - reservation
;
661 /* try to determine how much space is reserved for metadata from
662 * the last get_extents() entry, otherwise fallback to the
665 static __u32
imsm_reserved_sectors(struct intel_super
*super
, struct dl
*dl
)
671 /* for spares just return a minimal reservation which will grow
672 * once the spare is picked up by an array
675 return MPB_SECTOR_CNT
;
677 e
= get_extents(super
, dl
);
679 return MPB_SECTOR_CNT
+ IMSM_RESERVED_SECTORS
;
681 /* scroll to last entry */
682 for (i
= 0; e
[i
].size
; i
++)
685 rv
= __le32_to_cpu(dl
->disk
.total_blocks
) - e
[i
].start
;
692 static int is_spare(struct imsm_disk
*disk
)
694 return (disk
->status
& SPARE_DISK
) == SPARE_DISK
;
697 static int is_configured(struct imsm_disk
*disk
)
699 return (disk
->status
& CONFIGURED_DISK
) == CONFIGURED_DISK
;
702 static int is_failed(struct imsm_disk
*disk
)
704 return (disk
->status
& FAILED_DISK
) == FAILED_DISK
;
707 /* Return minimum size of a spare that can be used in this array*/
708 static unsigned long long min_acceptable_spare_size_imsm(struct supertype
*st
)
710 struct intel_super
*super
= st
->sb
;
714 unsigned long long rv
= 0;
718 /* find first active disk in array */
720 while (dl
&& (is_failed(&dl
->disk
) || dl
->index
== -1))
724 /* find last lba used by subarrays */
725 e
= get_extents(super
, dl
);
728 for (i
= 0; e
[i
].size
; i
++)
731 rv
= e
[i
-1].start
+ e
[i
-1].size
;
733 /* add the amount of space needed for metadata */
734 rv
= rv
+ MPB_SECTOR_CNT
+ IMSM_RESERVED_SECTORS
;
739 static __u64
blocks_per_migr_unit(struct imsm_dev
*dev
);
741 static void print_imsm_dev(struct imsm_dev
*dev
, char *uuid
, int disk_idx
)
745 struct imsm_map
*map
= get_imsm_map(dev
, 0);
746 struct imsm_map
*map2
= get_imsm_map(dev
, 1);
750 printf("[%.16s]:\n", dev
->volume
);
751 printf(" UUID : %s\n", uuid
);
752 printf(" RAID Level : %d", get_imsm_raid_level(map
));
754 printf(" <-- %d", get_imsm_raid_level(map2
));
756 printf(" Members : %d", map
->num_members
);
758 printf(" <-- %d", map2
->num_members
);
760 printf(" Slots : [");
761 for (i
= 0; i
< map
->num_members
; i
++) {
762 ord
= get_imsm_ord_tbl_ent(dev
, i
, 0);
763 printf("%s", ord
& IMSM_ORD_REBUILD
? "_" : "U");
768 for (i
= 0; i
< map2
->num_members
; i
++) {
769 ord
= get_imsm_ord_tbl_ent(dev
, i
, 1);
770 printf("%s", ord
& IMSM_ORD_REBUILD
? "_" : "U");
775 slot
= get_imsm_disk_slot(map
, disk_idx
);
777 ord
= get_imsm_ord_tbl_ent(dev
, slot
, -1);
778 printf(" This Slot : %d%s\n", slot
,
779 ord
& IMSM_ORD_REBUILD
? " (out-of-sync)" : "");
781 printf(" This Slot : ?\n");
782 sz
= __le32_to_cpu(dev
->size_high
);
784 sz
+= __le32_to_cpu(dev
->size_low
);
785 printf(" Array Size : %llu%s\n", (unsigned long long)sz
,
786 human_size(sz
* 512));
787 sz
= __le32_to_cpu(map
->blocks_per_member
);
788 printf(" Per Dev Size : %llu%s\n", (unsigned long long)sz
,
789 human_size(sz
* 512));
790 printf(" Sector Offset : %u\n",
791 __le32_to_cpu(map
->pba_of_lba0
));
792 printf(" Num Stripes : %u\n",
793 __le32_to_cpu(map
->num_data_stripes
));
794 printf(" Chunk Size : %u KiB",
795 __le16_to_cpu(map
->blocks_per_strip
) / 2);
797 printf(" <-- %u KiB",
798 __le16_to_cpu(map2
->blocks_per_strip
) / 2);
800 printf(" Reserved : %d\n", __le32_to_cpu(dev
->reserved_blocks
));
801 printf(" Migrate State : ");
802 if (dev
->vol
.migr_state
) {
803 if (migr_type(dev
) == MIGR_INIT
)
804 printf("initialize\n");
805 else if (migr_type(dev
) == MIGR_REBUILD
)
807 else if (migr_type(dev
) == MIGR_VERIFY
)
809 else if (migr_type(dev
) == MIGR_GEN_MIGR
)
810 printf("general migration\n");
811 else if (migr_type(dev
) == MIGR_STATE_CHANGE
)
812 printf("state change\n");
813 else if (migr_type(dev
) == MIGR_REPAIR
)
816 printf("<unknown:%d>\n", migr_type(dev
));
819 printf(" Map State : %s", map_state_str
[map
->map_state
]);
820 if (dev
->vol
.migr_state
) {
821 struct imsm_map
*map
= get_imsm_map(dev
, 1);
823 printf(" <-- %s", map_state_str
[map
->map_state
]);
824 printf("\n Checkpoint : %u (%llu)",
825 __le32_to_cpu(dev
->vol
.curr_migr_unit
),
826 (unsigned long long)blocks_per_migr_unit(dev
));
829 printf(" Dirty State : %s\n", dev
->vol
.dirty
? "dirty" : "clean");
832 static void print_imsm_disk(struct imsm_super
*mpb
, int index
, __u32 reserved
)
834 struct imsm_disk
*disk
= __get_imsm_disk(mpb
, index
);
835 char str
[MAX_RAID_SERIAL_LEN
+ 1];
838 if (index
< 0 || !disk
)
842 snprintf(str
, MAX_RAID_SERIAL_LEN
+ 1, "%s", disk
->serial
);
843 printf(" Disk%02d Serial : %s\n", index
, str
);
844 printf(" State :%s%s%s\n", is_spare(disk
) ? " spare" : "",
845 is_configured(disk
) ? " active" : "",
846 is_failed(disk
) ? " failed" : "");
847 printf(" Id : %08x\n", __le32_to_cpu(disk
->scsi_id
));
848 sz
= __le32_to_cpu(disk
->total_blocks
) - reserved
;
849 printf(" Usable Size : %llu%s\n", (unsigned long long)sz
,
850 human_size(sz
* 512));
853 static void getinfo_super_imsm(struct supertype
*st
, struct mdinfo
*info
, char *map
);
855 static void examine_super_imsm(struct supertype
*st
, char *homehost
)
857 struct intel_super
*super
= st
->sb
;
858 struct imsm_super
*mpb
= super
->anchor
;
859 char str
[MAX_SIGNATURE_LENGTH
];
864 __u32 reserved
= imsm_reserved_sectors(super
, super
->disks
);
867 snprintf(str
, MPB_SIG_LEN
, "%s", mpb
->sig
);
868 printf(" Magic : %s\n", str
);
869 snprintf(str
, strlen(MPB_VERSION_RAID0
), "%s", get_imsm_version(mpb
));
870 printf(" Version : %s\n", get_imsm_version(mpb
));
871 printf(" Orig Family : %08x\n", __le32_to_cpu(mpb
->orig_family_num
));
872 printf(" Family : %08x\n", __le32_to_cpu(mpb
->family_num
));
873 printf(" Generation : %08x\n", __le32_to_cpu(mpb
->generation_num
));
874 getinfo_super_imsm(st
, &info
, NULL
);
875 fname_from_uuid(st
, &info
, nbuf
, ':');
876 printf(" UUID : %s\n", nbuf
+ 5);
877 sum
= __le32_to_cpu(mpb
->check_sum
);
878 printf(" Checksum : %08x %s\n", sum
,
879 __gen_imsm_checksum(mpb
) == sum
? "correct" : "incorrect");
880 printf(" MPB Sectors : %d\n", mpb_sectors(mpb
));
881 printf(" Disks : %d\n", mpb
->num_disks
);
882 printf(" RAID Devices : %d\n", mpb
->num_raid_devs
);
883 print_imsm_disk(mpb
, super
->disks
->index
, reserved
);
884 if (super
->bbm_log
) {
885 struct bbm_log
*log
= super
->bbm_log
;
888 printf("Bad Block Management Log:\n");
889 printf(" Log Size : %d\n", __le32_to_cpu(mpb
->bbm_log_size
));
890 printf(" Signature : %x\n", __le32_to_cpu(log
->signature
));
891 printf(" Entry Count : %d\n", __le32_to_cpu(log
->entry_count
));
892 printf(" Spare Blocks : %d\n", __le32_to_cpu(log
->reserved_spare_block_count
));
893 printf(" First Spare : %llx\n",
894 (unsigned long long) __le64_to_cpu(log
->first_spare_lba
));
896 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
898 struct imsm_dev
*dev
= __get_imsm_dev(mpb
, i
);
900 super
->current_vol
= i
;
901 getinfo_super_imsm(st
, &info
, NULL
);
902 fname_from_uuid(st
, &info
, nbuf
, ':');
903 print_imsm_dev(dev
, nbuf
+ 5, super
->disks
->index
);
905 for (i
= 0; i
< mpb
->num_disks
; i
++) {
906 if (i
== super
->disks
->index
)
908 print_imsm_disk(mpb
, i
, reserved
);
910 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
911 struct imsm_disk
*disk
;
912 char str
[MAX_RAID_SERIAL_LEN
+ 1];
920 snprintf(str
, MAX_RAID_SERIAL_LEN
+ 1, "%s", disk
->serial
);
921 printf(" Disk Serial : %s\n", str
);
922 printf(" State :%s%s%s\n", is_spare(disk
) ? " spare" : "",
923 is_configured(disk
) ? " active" : "",
924 is_failed(disk
) ? " failed" : "");
925 printf(" Id : %08x\n", __le32_to_cpu(disk
->scsi_id
));
926 sz
= __le32_to_cpu(disk
->total_blocks
) - reserved
;
927 printf(" Usable Size : %llu%s\n", (unsigned long long)sz
,
928 human_size(sz
* 512));
932 static void brief_examine_super_imsm(struct supertype
*st
, int verbose
)
934 /* We just write a generic IMSM ARRAY entry */
937 struct intel_super
*super
= st
->sb
;
939 if (!super
->anchor
->num_raid_devs
) {
940 printf("ARRAY metadata=imsm\n");
944 getinfo_super_imsm(st
, &info
, NULL
);
945 fname_from_uuid(st
, &info
, nbuf
, ':');
946 printf("ARRAY metadata=imsm UUID=%s\n", nbuf
+ 5);
949 static void brief_examine_subarrays_imsm(struct supertype
*st
, int verbose
)
951 /* We just write a generic IMSM ARRAY entry */
955 struct intel_super
*super
= st
->sb
;
958 if (!super
->anchor
->num_raid_devs
)
961 getinfo_super_imsm(st
, &info
, NULL
);
962 fname_from_uuid(st
, &info
, nbuf
, ':');
963 for (i
= 0; i
< super
->anchor
->num_raid_devs
; i
++) {
964 struct imsm_dev
*dev
= get_imsm_dev(super
, i
);
966 super
->current_vol
= i
;
967 getinfo_super_imsm(st
, &info
, NULL
);
968 fname_from_uuid(st
, &info
, nbuf1
, ':');
969 printf("ARRAY /dev/md/%.16s container=%s member=%d UUID=%s\n",
970 dev
->volume
, nbuf
+ 5, i
, nbuf1
+ 5);
974 static void export_examine_super_imsm(struct supertype
*st
)
976 struct intel_super
*super
= st
->sb
;
977 struct imsm_super
*mpb
= super
->anchor
;
981 getinfo_super_imsm(st
, &info
, NULL
);
982 fname_from_uuid(st
, &info
, nbuf
, ':');
983 printf("MD_METADATA=imsm\n");
984 printf("MD_LEVEL=container\n");
985 printf("MD_UUID=%s\n", nbuf
+5);
986 printf("MD_DEVICES=%u\n", mpb
->num_disks
);
989 static void detail_super_imsm(struct supertype
*st
, char *homehost
)
994 getinfo_super_imsm(st
, &info
, NULL
);
995 fname_from_uuid(st
, &info
, nbuf
, ':');
996 printf("\n UUID : %s\n", nbuf
+ 5);
999 static void brief_detail_super_imsm(struct supertype
*st
)
1003 getinfo_super_imsm(st
, &info
, NULL
);
1004 fname_from_uuid(st
, &info
, nbuf
, ':');
1005 printf(" UUID=%s", nbuf
+ 5);
1008 static int imsm_read_serial(int fd
, char *devname
, __u8
*serial
);
1009 static void fd2devname(int fd
, char *name
);
1011 static int imsm_enumerate_ports(const char *hba_path
, int port_count
, int host_base
, int verbose
)
1013 /* dump an unsorted list of devices attached to ahci, as well as
1014 * non-connected ports
1016 int hba_len
= strlen(hba_path
) + 1;
1021 unsigned long port_mask
= (1 << port_count
) - 1;
1023 if (port_count
> (int)sizeof(port_mask
) * 8) {
1025 fprintf(stderr
, Name
": port_count %d out of range\n", port_count
);
1029 /* scroll through /sys/dev/block looking for devices attached to
1032 dir
= opendir("/sys/dev/block");
1033 for (ent
= dir
? readdir(dir
) : NULL
; ent
; ent
= readdir(dir
)) {
1044 if (sscanf(ent
->d_name
, "%d:%d", &major
, &minor
) != 2)
1046 path
= devt_to_devpath(makedev(major
, minor
));
1049 if (!path_attached_to_hba(path
, hba_path
)) {
1055 /* retrieve the scsi device type */
1056 if (asprintf(&device
, "/sys/dev/block/%d:%d/device/xxxxxxx", major
, minor
) < 0) {
1058 fprintf(stderr
, Name
": failed to allocate 'device'\n");
1062 sprintf(device
, "/sys/dev/block/%d:%d/device/type", major
, minor
);
1063 if (load_sys(device
, buf
) != 0) {
1065 fprintf(stderr
, Name
": failed to read device type for %s\n",
1071 type
= strtoul(buf
, NULL
, 10);
1073 /* if it's not a disk print the vendor and model */
1074 if (!(type
== 0 || type
== 7 || type
== 14)) {
1077 sprintf(device
, "/sys/dev/block/%d:%d/device/vendor", major
, minor
);
1078 if (load_sys(device
, buf
) == 0) {
1079 strncpy(vendor
, buf
, sizeof(vendor
));
1080 vendor
[sizeof(vendor
) - 1] = '\0';
1081 c
= (char *) &vendor
[sizeof(vendor
) - 1];
1082 while (isspace(*c
) || *c
== '\0')
1086 sprintf(device
, "/sys/dev/block/%d:%d/device/model", major
, minor
);
1087 if (load_sys(device
, buf
) == 0) {
1088 strncpy(model
, buf
, sizeof(model
));
1089 model
[sizeof(model
) - 1] = '\0';
1090 c
= (char *) &model
[sizeof(model
) - 1];
1091 while (isspace(*c
) || *c
== '\0')
1095 if (vendor
[0] && model
[0])
1096 sprintf(buf
, "%.64s %.64s", vendor
, model
);
1098 switch (type
) { /* numbers from hald/linux/device.c */
1099 case 1: sprintf(buf
, "tape"); break;
1100 case 2: sprintf(buf
, "printer"); break;
1101 case 3: sprintf(buf
, "processor"); break;
1103 case 5: sprintf(buf
, "cdrom"); break;
1104 case 6: sprintf(buf
, "scanner"); break;
1105 case 8: sprintf(buf
, "media_changer"); break;
1106 case 9: sprintf(buf
, "comm"); break;
1107 case 12: sprintf(buf
, "raid"); break;
1108 default: sprintf(buf
, "unknown");
1114 /* chop device path to 'host%d' and calculate the port number */
1115 c
= strchr(&path
[hba_len
], '/');
1118 fprintf(stderr
, Name
": %s - invalid path name\n", path
+ hba_len
);
1123 if (sscanf(&path
[hba_len
], "host%d", &port
) == 1)
1127 *c
= '/'; /* repair the full string */
1128 fprintf(stderr
, Name
": failed to determine port number for %s\n",
1135 /* mark this port as used */
1136 port_mask
&= ~(1 << port
);
1138 /* print out the device information */
1140 printf(" Port%d : - non-disk device (%s) -\n", port
, buf
);
1144 fd
= dev_open(ent
->d_name
, O_RDONLY
);
1146 printf(" Port%d : - disk info unavailable -\n", port
);
1148 fd2devname(fd
, buf
);
1149 printf(" Port%d : %s", port
, buf
);
1150 if (imsm_read_serial(fd
, NULL
, (__u8
*) buf
) == 0)
1151 printf(" (%s)\n", buf
);
1166 for (i
= 0; i
< port_count
; i
++)
1167 if (port_mask
& (1 << i
))
1168 printf(" Port%d : - no device attached -\n", i
);
1174 static int detail_platform_imsm(int verbose
, int enumerate_only
)
1176 /* There are two components to imsm platform support, the ahci SATA
1177 * controller and the option-rom. To find the SATA controller we
1178 * simply look in /sys/bus/pci/drivers/ahci to see if an ahci
1179 * controller with the Intel vendor id is present. This approach
1180 * allows mdadm to leverage the kernel's ahci detection logic, with the
1181 * caveat that if ahci.ko is not loaded mdadm will not be able to
1182 * detect platform raid capabilities. The option-rom resides in a
1183 * platform "Adapter ROM". We scan for its signature to retrieve the
1184 * platform capabilities. If raid support is disabled in the BIOS the
1185 * option-rom capability structure will not be available.
1187 const struct imsm_orom
*orom
;
1188 struct sys_dev
*list
, *hba
;
1191 const char *hba_path
;
1195 if (enumerate_only
) {
1196 if (check_env("IMSM_NO_PLATFORM") || find_imsm_orom())
1201 list
= find_driver_devices("pci", "ahci");
1202 for (hba
= list
; hba
; hba
= hba
->next
)
1203 if (devpath_to_vendor(hba
->path
) == 0x8086)
1208 fprintf(stderr
, Name
": unable to find active ahci controller\n");
1209 free_sys_dev(&list
);
1212 fprintf(stderr
, Name
": found Intel SATA AHCI Controller\n");
1213 hba_path
= hba
->path
;
1215 free_sys_dev(&list
);
1217 orom
= find_imsm_orom();
1220 fprintf(stderr
, Name
": imsm option-rom not found\n");
1224 printf(" Platform : Intel(R) Matrix Storage Manager\n");
1225 printf(" Version : %d.%d.%d.%d\n", orom
->major_ver
, orom
->minor_ver
,
1226 orom
->hotfix_ver
, orom
->build
);
1227 printf(" RAID Levels :%s%s%s%s%s\n",
1228 imsm_orom_has_raid0(orom
) ? " raid0" : "",
1229 imsm_orom_has_raid1(orom
) ? " raid1" : "",
1230 imsm_orom_has_raid1e(orom
) ? " raid1e" : "",
1231 imsm_orom_has_raid10(orom
) ? " raid10" : "",
1232 imsm_orom_has_raid5(orom
) ? " raid5" : "");
1233 printf(" Chunk Sizes :%s%s%s%s%s%s%s%s%s%s%s%s%s%s%s%s\n",
1234 imsm_orom_has_chunk(orom
, 2) ? " 2k" : "",
1235 imsm_orom_has_chunk(orom
, 4) ? " 4k" : "",
1236 imsm_orom_has_chunk(orom
, 8) ? " 8k" : "",
1237 imsm_orom_has_chunk(orom
, 16) ? " 16k" : "",
1238 imsm_orom_has_chunk(orom
, 32) ? " 32k" : "",
1239 imsm_orom_has_chunk(orom
, 64) ? " 64k" : "",
1240 imsm_orom_has_chunk(orom
, 128) ? " 128k" : "",
1241 imsm_orom_has_chunk(orom
, 256) ? " 256k" : "",
1242 imsm_orom_has_chunk(orom
, 512) ? " 512k" : "",
1243 imsm_orom_has_chunk(orom
, 1024*1) ? " 1M" : "",
1244 imsm_orom_has_chunk(orom
, 1024*2) ? " 2M" : "",
1245 imsm_orom_has_chunk(orom
, 1024*4) ? " 4M" : "",
1246 imsm_orom_has_chunk(orom
, 1024*8) ? " 8M" : "",
1247 imsm_orom_has_chunk(orom
, 1024*16) ? " 16M" : "",
1248 imsm_orom_has_chunk(orom
, 1024*32) ? " 32M" : "",
1249 imsm_orom_has_chunk(orom
, 1024*64) ? " 64M" : "");
1250 printf(" Max Disks : %d\n", orom
->tds
);
1251 printf(" Max Volumes : %d\n", orom
->vpa
);
1252 printf(" I/O Controller : %s\n", hba_path
);
1254 /* find the smallest scsi host number to determine a port number base */
1255 dir
= opendir(hba_path
);
1256 for (ent
= dir
? readdir(dir
) : NULL
; ent
; ent
= readdir(dir
)) {
1259 if (sscanf(ent
->d_name
, "host%d", &host
) != 1)
1261 if (port_count
== 0)
1263 else if (host
< host_base
)
1266 if (host
+ 1 > port_count
+ host_base
)
1267 port_count
= host
+ 1 - host_base
;
1273 if (!port_count
|| imsm_enumerate_ports(hba_path
, port_count
,
1274 host_base
, verbose
) != 0) {
1276 fprintf(stderr
, Name
": failed to enumerate ports\n");
1284 static int match_home_imsm(struct supertype
*st
, char *homehost
)
1286 /* the imsm metadata format does not specify any host
1287 * identification information. We return -1 since we can never
1288 * confirm nor deny whether a given array is "meant" for this
1289 * host. We rely on compare_super and the 'family_num' fields to
1290 * exclude member disks that do not belong, and we rely on
1291 * mdadm.conf to specify the arrays that should be assembled.
1292 * Auto-assembly may still pick up "foreign" arrays.
1298 static void uuid_from_super_imsm(struct supertype
*st
, int uuid
[4])
1300 /* The uuid returned here is used for:
1301 * uuid to put into bitmap file (Create, Grow)
1302 * uuid for backup header when saving critical section (Grow)
1303 * comparing uuids when re-adding a device into an array
1304 * In these cases the uuid required is that of the data-array,
1305 * not the device-set.
1306 * uuid to recognise same set when adding a missing device back
1307 * to an array. This is a uuid for the device-set.
1309 * For each of these we can make do with a truncated
1310 * or hashed uuid rather than the original, as long as
1312 * In each case the uuid required is that of the data-array,
1313 * not the device-set.
1315 /* imsm does not track uuid's so we synthesis one using sha1 on
1316 * - The signature (Which is constant for all imsm array, but no matter)
1317 * - the orig_family_num of the container
1318 * - the index number of the volume
1319 * - the 'serial' number of the volume.
1320 * Hopefully these are all constant.
1322 struct intel_super
*super
= st
->sb
;
1325 struct sha1_ctx ctx
;
1326 struct imsm_dev
*dev
= NULL
;
1329 /* some mdadm versions failed to set ->orig_family_num, in which
1330 * case fall back to ->family_num. orig_family_num will be
1331 * fixed up with the first metadata update.
1333 family_num
= super
->anchor
->orig_family_num
;
1334 if (family_num
== 0)
1335 family_num
= super
->anchor
->family_num
;
1336 sha1_init_ctx(&ctx
);
1337 sha1_process_bytes(super
->anchor
->sig
, MPB_SIG_LEN
, &ctx
);
1338 sha1_process_bytes(&family_num
, sizeof(__u32
), &ctx
);
1339 if (super
->current_vol
>= 0)
1340 dev
= get_imsm_dev(super
, super
->current_vol
);
1342 __u32 vol
= super
->current_vol
;
1343 sha1_process_bytes(&vol
, sizeof(vol
), &ctx
);
1344 sha1_process_bytes(dev
->volume
, MAX_RAID_SERIAL_LEN
, &ctx
);
1346 sha1_finish_ctx(&ctx
, buf
);
1347 memcpy(uuid
, buf
, 4*4);
1352 get_imsm_numerical_version(struct imsm_super
*mpb
, int *m
, int *p
)
1354 __u8
*v
= get_imsm_version(mpb
);
1355 __u8
*end
= mpb
->sig
+ MAX_SIGNATURE_LENGTH
;
1356 char major
[] = { 0, 0, 0 };
1357 char minor
[] = { 0 ,0, 0 };
1358 char patch
[] = { 0, 0, 0 };
1359 char *ver_parse
[] = { major
, minor
, patch
};
1363 while (*v
!= '\0' && v
< end
) {
1364 if (*v
!= '.' && j
< 2)
1365 ver_parse
[i
][j
++] = *v
;
1373 *m
= strtol(minor
, NULL
, 0);
1374 *p
= strtol(patch
, NULL
, 0);
1378 static __u32
migr_strip_blocks_resync(struct imsm_dev
*dev
)
1380 /* migr_strip_size when repairing or initializing parity */
1381 struct imsm_map
*map
= get_imsm_map(dev
, 0);
1382 __u32 chunk
= __le32_to_cpu(map
->blocks_per_strip
);
1384 switch (get_imsm_raid_level(map
)) {
1389 return 128*1024 >> 9;
1393 static __u32
migr_strip_blocks_rebuild(struct imsm_dev
*dev
)
1395 /* migr_strip_size when rebuilding a degraded disk, no idea why
1396 * this is different than migr_strip_size_resync(), but it's good
1399 struct imsm_map
*map
= get_imsm_map(dev
, 1);
1400 __u32 chunk
= __le32_to_cpu(map
->blocks_per_strip
);
1402 switch (get_imsm_raid_level(map
)) {
1405 if (map
->num_members
% map
->num_domains
== 0)
1406 return 128*1024 >> 9;
1410 return max((__u32
) 64*1024 >> 9, chunk
);
1412 return 128*1024 >> 9;
1416 static __u32
num_stripes_per_unit_resync(struct imsm_dev
*dev
)
1418 struct imsm_map
*lo
= get_imsm_map(dev
, 0);
1419 struct imsm_map
*hi
= get_imsm_map(dev
, 1);
1420 __u32 lo_chunk
= __le32_to_cpu(lo
->blocks_per_strip
);
1421 __u32 hi_chunk
= __le32_to_cpu(hi
->blocks_per_strip
);
1423 return max((__u32
) 1, hi_chunk
/ lo_chunk
);
1426 static __u32
num_stripes_per_unit_rebuild(struct imsm_dev
*dev
)
1428 struct imsm_map
*lo
= get_imsm_map(dev
, 0);
1429 int level
= get_imsm_raid_level(lo
);
1431 if (level
== 1 || level
== 10) {
1432 struct imsm_map
*hi
= get_imsm_map(dev
, 1);
1434 return hi
->num_domains
;
1436 return num_stripes_per_unit_resync(dev
);
1439 static __u8
imsm_num_data_members(struct imsm_dev
*dev
, int second_map
)
1441 /* named 'imsm_' because raid0, raid1 and raid10
1442 * counter-intuitively have the same number of data disks
1444 struct imsm_map
*map
= get_imsm_map(dev
, second_map
);
1446 switch (get_imsm_raid_level(map
)) {
1450 return map
->num_members
;
1452 return map
->num_members
- 1;
1454 dprintf("%s: unsupported raid level\n", __func__
);
1459 static __u32
parity_segment_depth(struct imsm_dev
*dev
)
1461 struct imsm_map
*map
= get_imsm_map(dev
, 0);
1462 __u32 chunk
= __le32_to_cpu(map
->blocks_per_strip
);
1464 switch(get_imsm_raid_level(map
)) {
1467 return chunk
* map
->num_domains
;
1469 return chunk
* map
->num_members
;
1475 static __u32
map_migr_block(struct imsm_dev
*dev
, __u32 block
)
1477 struct imsm_map
*map
= get_imsm_map(dev
, 1);
1478 __u32 chunk
= __le32_to_cpu(map
->blocks_per_strip
);
1479 __u32 strip
= block
/ chunk
;
1481 switch (get_imsm_raid_level(map
)) {
1484 __u32 vol_strip
= (strip
* map
->num_domains
) + 1;
1485 __u32 vol_stripe
= vol_strip
/ map
->num_members
;
1487 return vol_stripe
* chunk
+ block
% chunk
;
1489 __u32 stripe
= strip
/ (map
->num_members
- 1);
1491 return stripe
* chunk
+ block
% chunk
;
1498 static __u64
blocks_per_migr_unit(struct imsm_dev
*dev
)
1500 /* calculate the conversion factor between per member 'blocks'
1501 * (md/{resync,rebuild}_start) and imsm migration units, return
1502 * 0 for the 'not migrating' and 'unsupported migration' cases
1504 if (!dev
->vol
.migr_state
)
1507 switch (migr_type(dev
)) {
1512 struct imsm_map
*map
= get_imsm_map(dev
, 0);
1513 __u32 stripes_per_unit
;
1514 __u32 blocks_per_unit
;
1523 /* yes, this is really the translation of migr_units to
1524 * per-member blocks in the 'resync' case
1526 stripes_per_unit
= num_stripes_per_unit_resync(dev
);
1527 migr_chunk
= migr_strip_blocks_resync(dev
);
1528 disks
= imsm_num_data_members(dev
, 0);
1529 blocks_per_unit
= stripes_per_unit
* migr_chunk
* disks
;
1530 stripe
= __le32_to_cpu(map
->blocks_per_strip
) * disks
;
1531 segment
= blocks_per_unit
/ stripe
;
1532 block_rel
= blocks_per_unit
- segment
* stripe
;
1533 parity_depth
= parity_segment_depth(dev
);
1534 block_map
= map_migr_block(dev
, block_rel
);
1535 return block_map
+ parity_depth
* segment
;
1537 case MIGR_REBUILD
: {
1538 __u32 stripes_per_unit
;
1541 stripes_per_unit
= num_stripes_per_unit_rebuild(dev
);
1542 migr_chunk
= migr_strip_blocks_rebuild(dev
);
1543 return migr_chunk
* stripes_per_unit
;
1545 case MIGR_STATE_CHANGE
:
1551 static int imsm_level_to_layout(int level
)
1559 return ALGORITHM_LEFT_ASYMMETRIC
;
1566 static void getinfo_super_imsm_volume(struct supertype
*st
, struct mdinfo
*info
, char *dmap
)
1568 struct intel_super
*super
= st
->sb
;
1569 struct imsm_dev
*dev
= get_imsm_dev(super
, super
->current_vol
);
1570 struct imsm_map
*map
= get_imsm_map(dev
, 0);
1571 struct imsm_map
*prev_map
= get_imsm_map(dev
, 1);
1572 struct imsm_map
*map_to_analyse
= map
;
1575 int map_disks
= info
->array
.raid_disks
;
1578 map_to_analyse
= prev_map
;
1580 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
1581 if (dl
->raiddisk
== info
->disk
.raid_disk
)
1583 info
->container_member
= super
->current_vol
;
1584 info
->array
.raid_disks
= map_to_analyse
->num_members
;
1585 info
->array
.level
= get_imsm_raid_level(map_to_analyse
);
1586 info
->array
.layout
= imsm_level_to_layout(info
->array
.level
);
1587 info
->array
.md_minor
= -1;
1588 info
->array
.ctime
= 0;
1589 info
->array
.utime
= 0;
1590 info
->array
.chunk_size
=
1591 __le16_to_cpu(map_to_analyse
->blocks_per_strip
) << 9;
1592 info
->array
.state
= !dev
->vol
.dirty
;
1593 info
->custom_array_size
= __le32_to_cpu(dev
->size_high
);
1594 info
->custom_array_size
<<= 32;
1595 info
->custom_array_size
|= __le32_to_cpu(dev
->size_low
);
1597 info
->new_level
= get_imsm_raid_level(map
);
1598 info
->new_layout
= imsm_level_to_layout(info
->new_level
);
1599 info
->new_chunk
= __le16_to_cpu(map
->blocks_per_strip
) << 9;
1601 info
->new_level
= UnSet
;
1602 info
->new_layout
= UnSet
;
1603 info
->new_chunk
= info
->array
.chunk_size
;
1605 info
->disk
.major
= 0;
1606 info
->disk
.minor
= 0;
1608 info
->disk
.major
= dl
->major
;
1609 info
->disk
.minor
= dl
->minor
;
1612 info
->data_offset
= __le32_to_cpu(map_to_analyse
->pba_of_lba0
);
1613 info
->component_size
=
1614 __le32_to_cpu(map_to_analyse
->blocks_per_member
);
1615 memset(info
->uuid
, 0, sizeof(info
->uuid
));
1616 info
->recovery_start
= MaxSector
;
1617 info
->reshape_active
= (prev_map
!= NULL
);
1618 if (info
->reshape_active
)
1619 info
->delta_disks
= map
->num_members
- prev_map
->num_members
;
1621 info
->delta_disks
= 0;
1623 if (map_to_analyse
->map_state
== IMSM_T_STATE_UNINITIALIZED
||
1625 info
->resync_start
= 0;
1626 } else if (dev
->vol
.migr_state
) {
1627 switch (migr_type(dev
)) {
1630 __u64 blocks_per_unit
= blocks_per_migr_unit(dev
);
1631 __u64 units
= __le32_to_cpu(dev
->vol
.curr_migr_unit
);
1633 info
->resync_start
= blocks_per_unit
* units
;
1637 /* we could emulate the checkpointing of
1638 * 'sync_action=check' migrations, but for now
1639 * we just immediately complete them
1642 /* this is handled by container_content_imsm() */
1644 case MIGR_STATE_CHANGE
:
1645 /* FIXME handle other migrations */
1647 /* we are not dirty, so... */
1648 info
->resync_start
= MaxSector
;
1651 info
->resync_start
= MaxSector
;
1653 strncpy(info
->name
, (char *) dev
->volume
, MAX_RAID_SERIAL_LEN
);
1654 info
->name
[MAX_RAID_SERIAL_LEN
] = 0;
1656 info
->array
.major_version
= -1;
1657 info
->array
.minor_version
= -2;
1658 devname
= devnum2devname(st
->container_dev
);
1659 *info
->text_version
= '\0';
1661 sprintf(info
->text_version
, "/%s/%d", devname
, info
->container_member
);
1663 info
->safe_mode_delay
= 4000; /* 4 secs like the Matrix driver */
1664 uuid_from_super_imsm(st
, info
->uuid
);
1668 for (i
=0; i
<map_disks
; i
++) {
1670 if (i
< info
->array
.raid_disks
) {
1671 struct imsm_disk
*dsk
;
1672 j
= get_imsm_disk_idx(dev
, i
, -1);
1673 dsk
= get_imsm_disk(super
, j
);
1674 if (dsk
&& (dsk
->status
& CONFIGURED_DISK
))
1681 static __u8
imsm_check_degraded(struct intel_super
*super
, struct imsm_dev
*dev
, int failed
);
1682 static int imsm_count_failed(struct intel_super
*super
, struct imsm_dev
*dev
);
1684 static struct imsm_disk
*get_imsm_missing(struct intel_super
*super
, __u8 index
)
1688 for (d
= super
->missing
; d
; d
= d
->next
)
1689 if (d
->index
== index
)
1694 static void getinfo_super_imsm(struct supertype
*st
, struct mdinfo
*info
, char *map
)
1696 struct intel_super
*super
= st
->sb
;
1697 struct imsm_disk
*disk
;
1698 int map_disks
= info
->array
.raid_disks
;
1699 int max_enough
= -1;
1701 struct imsm_super
*mpb
;
1703 if (super
->current_vol
>= 0) {
1704 getinfo_super_imsm_volume(st
, info
, map
);
1708 /* Set raid_disks to zero so that Assemble will always pull in valid
1711 info
->array
.raid_disks
= 0;
1712 info
->array
.level
= LEVEL_CONTAINER
;
1713 info
->array
.layout
= 0;
1714 info
->array
.md_minor
= -1;
1715 info
->array
.ctime
= 0; /* N/A for imsm */
1716 info
->array
.utime
= 0;
1717 info
->array
.chunk_size
= 0;
1719 info
->disk
.major
= 0;
1720 info
->disk
.minor
= 0;
1721 info
->disk
.raid_disk
= -1;
1722 info
->reshape_active
= 0;
1723 info
->array
.major_version
= -1;
1724 info
->array
.minor_version
= -2;
1725 strcpy(info
->text_version
, "imsm");
1726 info
->safe_mode_delay
= 0;
1727 info
->disk
.number
= -1;
1728 info
->disk
.state
= 0;
1730 info
->recovery_start
= MaxSector
;
1732 /* do we have the all the insync disks that we expect? */
1733 mpb
= super
->anchor
;
1735 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
1736 struct imsm_dev
*dev
= get_imsm_dev(super
, i
);
1737 int failed
, enough
, j
, missing
= 0;
1738 struct imsm_map
*map
;
1741 failed
= imsm_count_failed(super
, dev
);
1742 state
= imsm_check_degraded(super
, dev
, failed
);
1743 map
= get_imsm_map(dev
, dev
->vol
.migr_state
);
1745 /* any newly missing disks?
1746 * (catches single-degraded vs double-degraded)
1748 for (j
= 0; j
< map
->num_members
; j
++) {
1749 __u32 ord
= get_imsm_ord_tbl_ent(dev
, i
, -1);
1750 __u32 idx
= ord_to_idx(ord
);
1752 if (!(ord
& IMSM_ORD_REBUILD
) &&
1753 get_imsm_missing(super
, idx
)) {
1759 if (state
== IMSM_T_STATE_FAILED
)
1761 else if (state
== IMSM_T_STATE_DEGRADED
&&
1762 (state
!= map
->map_state
|| missing
))
1764 else /* we're normal, or already degraded */
1767 /* in the missing/failed disk case check to see
1768 * if at least one array is runnable
1770 max_enough
= max(max_enough
, enough
);
1772 dprintf("%s: enough: %d\n", __func__
, max_enough
);
1773 info
->container_enough
= max_enough
;
1776 __u32 reserved
= imsm_reserved_sectors(super
, super
->disks
);
1778 disk
= &super
->disks
->disk
;
1779 info
->data_offset
= __le32_to_cpu(disk
->total_blocks
) - reserved
;
1780 info
->component_size
= reserved
;
1781 info
->disk
.state
= is_configured(disk
) ? (1 << MD_DISK_ACTIVE
) : 0;
1782 /* we don't change info->disk.raid_disk here because
1783 * this state will be finalized in mdmon after we have
1784 * found the 'most fresh' version of the metadata
1786 info
->disk
.state
|= is_failed(disk
) ? (1 << MD_DISK_FAULTY
) : 0;
1787 info
->disk
.state
|= is_spare(disk
) ? 0 : (1 << MD_DISK_SYNC
);
1790 /* only call uuid_from_super_imsm when this disk is part of a populated container,
1791 * ->compare_super may have updated the 'num_raid_devs' field for spares
1793 if (info
->disk
.state
& (1 << MD_DISK_SYNC
) || super
->anchor
->num_raid_devs
)
1794 uuid_from_super_imsm(st
, info
->uuid
);
1796 memcpy(info
->uuid
, uuid_zero
, sizeof(uuid_zero
));
1798 /* I don't know how to compute 'map' on imsm, so use safe default */
1801 for (i
= 0; i
< map_disks
; i
++)
1807 /* allocates memory and fills disk in mdinfo structure
1808 * for each disk in array */
1809 struct mdinfo
*getinfo_super_disks_imsm(struct supertype
*st
)
1811 struct mdinfo
*mddev
= NULL
;
1812 struct intel_super
*super
= st
->sb
;
1813 struct imsm_disk
*disk
;
1816 if (!super
|| !super
->disks
)
1819 mddev
= malloc(sizeof(*mddev
));
1821 fprintf(stderr
, Name
": Failed to allocate memory.\n");
1824 memset(mddev
, 0, sizeof(*mddev
));
1828 tmp
= malloc(sizeof(*tmp
));
1830 fprintf(stderr
, Name
": Failed to allocate memory.\n");
1835 memset(tmp
, 0, sizeof(*tmp
));
1837 tmp
->next
= mddev
->devs
;
1839 tmp
->disk
.number
= count
++;
1840 tmp
->disk
.major
= dl
->major
;
1841 tmp
->disk
.minor
= dl
->minor
;
1842 tmp
->disk
.state
= is_configured(disk
) ?
1843 (1 << MD_DISK_ACTIVE
) : 0;
1844 tmp
->disk
.state
|= is_failed(disk
) ? (1 << MD_DISK_FAULTY
) : 0;
1845 tmp
->disk
.state
|= is_spare(disk
) ? 0 : (1 << MD_DISK_SYNC
);
1846 tmp
->disk
.raid_disk
= -1;
1852 static int update_super_imsm(struct supertype
*st
, struct mdinfo
*info
,
1853 char *update
, char *devname
, int verbose
,
1854 int uuid_set
, char *homehost
)
1856 /* For 'assemble' and 'force' we need to return non-zero if any
1857 * change was made. For others, the return value is ignored.
1858 * Update options are:
1859 * force-one : This device looks a bit old but needs to be included,
1860 * update age info appropriately.
1861 * assemble: clear any 'faulty' flag to allow this device to
1863 * force-array: Array is degraded but being forced, mark it clean
1864 * if that will be needed to assemble it.
1866 * newdev: not used ????
1867 * grow: Array has gained a new device - this is currently for
1869 * resync: mark as dirty so a resync will happen.
1870 * name: update the name - preserving the homehost
1871 * uuid: Change the uuid of the array to match watch is given
1873 * Following are not relevant for this imsm:
1874 * sparc2.2 : update from old dodgey metadata
1875 * super-minor: change the preferred_minor number
1876 * summaries: update redundant counters.
1877 * homehost: update the recorded homehost
1878 * _reshape_progress: record new reshape_progress position.
1881 struct intel_super
*super
= st
->sb
;
1882 struct imsm_super
*mpb
;
1884 /* we can only update container info */
1885 if (!super
|| super
->current_vol
>= 0 || !super
->anchor
)
1888 mpb
= super
->anchor
;
1890 if (strcmp(update
, "uuid") == 0 && uuid_set
&& !info
->update_private
)
1892 else if (strcmp(update
, "uuid") == 0 && uuid_set
&& info
->update_private
) {
1893 mpb
->orig_family_num
= *((__u32
*) info
->update_private
);
1895 } else if (strcmp(update
, "uuid") == 0) {
1896 __u32
*new_family
= malloc(sizeof(*new_family
));
1898 /* update orig_family_number with the incoming random
1899 * data, report the new effective uuid, and store the
1900 * new orig_family_num for future updates.
1903 memcpy(&mpb
->orig_family_num
, info
->uuid
, sizeof(__u32
));
1904 uuid_from_super_imsm(st
, info
->uuid
);
1905 *new_family
= mpb
->orig_family_num
;
1906 info
->update_private
= new_family
;
1909 } else if (strcmp(update
, "assemble") == 0)
1914 /* successful update? recompute checksum */
1916 mpb
->check_sum
= __le32_to_cpu(__gen_imsm_checksum(mpb
));
1921 static size_t disks_to_mpb_size(int disks
)
1925 size
= sizeof(struct imsm_super
);
1926 size
+= (disks
- 1) * sizeof(struct imsm_disk
);
1927 size
+= 2 * sizeof(struct imsm_dev
);
1928 /* up to 2 maps per raid device (-2 for imsm_maps in imsm_dev */
1929 size
+= (4 - 2) * sizeof(struct imsm_map
);
1930 /* 4 possible disk_ord_tbl's */
1931 size
+= 4 * (disks
- 1) * sizeof(__u32
);
1936 static __u64
avail_size_imsm(struct supertype
*st
, __u64 devsize
)
1938 if (devsize
< (MPB_SECTOR_CNT
+ IMSM_RESERVED_SECTORS
))
1941 return devsize
- (MPB_SECTOR_CNT
+ IMSM_RESERVED_SECTORS
);
1944 static void free_devlist(struct intel_super
*super
)
1946 struct intel_dev
*dv
;
1948 while (super
->devlist
) {
1949 dv
= super
->devlist
->next
;
1950 free(super
->devlist
->dev
);
1951 free(super
->devlist
);
1952 super
->devlist
= dv
;
1956 static void imsm_copy_dev(struct imsm_dev
*dest
, struct imsm_dev
*src
)
1958 memcpy(dest
, src
, sizeof_imsm_dev(src
, 0));
1961 static int compare_super_imsm(struct supertype
*st
, struct supertype
*tst
)
1965 * 0 same, or first was empty, and second was copied
1966 * 1 second had wrong number
1968 * 3 wrong other info
1970 struct intel_super
*first
= st
->sb
;
1971 struct intel_super
*sec
= tst
->sb
;
1979 /* if an anchor does not have num_raid_devs set then it is a free
1982 if (first
->anchor
->num_raid_devs
> 0 &&
1983 sec
->anchor
->num_raid_devs
> 0) {
1984 /* Determine if these disks might ever have been
1985 * related. Further disambiguation can only take place
1986 * in load_super_imsm_all
1988 __u32 first_family
= first
->anchor
->orig_family_num
;
1989 __u32 sec_family
= sec
->anchor
->orig_family_num
;
1991 if (memcmp(first
->anchor
->sig
, sec
->anchor
->sig
,
1992 MAX_SIGNATURE_LENGTH
) != 0)
1995 if (first_family
== 0)
1996 first_family
= first
->anchor
->family_num
;
1997 if (sec_family
== 0)
1998 sec_family
= sec
->anchor
->family_num
;
2000 if (first_family
!= sec_family
)
2006 /* if 'first' is a spare promote it to a populated mpb with sec's
2009 if (first
->anchor
->num_raid_devs
== 0 &&
2010 sec
->anchor
->num_raid_devs
> 0) {
2012 struct intel_dev
*dv
;
2013 struct imsm_dev
*dev
;
2015 /* we need to copy raid device info from sec if an allocation
2016 * fails here we don't associate the spare
2018 for (i
= 0; i
< sec
->anchor
->num_raid_devs
; i
++) {
2019 dv
= malloc(sizeof(*dv
));
2022 dev
= malloc(sizeof_imsm_dev(get_imsm_dev(sec
, i
), 1));
2029 dv
->next
= first
->devlist
;
2030 first
->devlist
= dv
;
2032 if (i
< sec
->anchor
->num_raid_devs
) {
2033 /* allocation failure */
2034 free_devlist(first
);
2035 fprintf(stderr
, "imsm: failed to associate spare\n");
2038 first
->anchor
->num_raid_devs
= sec
->anchor
->num_raid_devs
;
2039 first
->anchor
->orig_family_num
= sec
->anchor
->orig_family_num
;
2040 first
->anchor
->family_num
= sec
->anchor
->family_num
;
2041 memcpy(first
->anchor
->sig
, sec
->anchor
->sig
, MAX_SIGNATURE_LENGTH
);
2042 for (i
= 0; i
< sec
->anchor
->num_raid_devs
; i
++)
2043 imsm_copy_dev(get_imsm_dev(first
, i
), get_imsm_dev(sec
, i
));
2049 static void fd2devname(int fd
, char *name
)
2053 char dname
[PATH_MAX
];
2058 if (fstat(fd
, &st
) != 0)
2060 sprintf(path
, "/sys/dev/block/%d:%d",
2061 major(st
.st_rdev
), minor(st
.st_rdev
));
2063 rv
= readlink(path
, dname
, sizeof(dname
));
2068 nm
= strrchr(dname
, '/');
2070 snprintf(name
, MAX_RAID_SERIAL_LEN
, "/dev/%s", nm
);
2073 extern int scsi_get_serial(int fd
, void *buf
, size_t buf_len
);
2075 static int imsm_read_serial(int fd
, char *devname
,
2076 __u8 serial
[MAX_RAID_SERIAL_LEN
])
2078 unsigned char scsi_serial
[255];
2087 memset(scsi_serial
, 0, sizeof(scsi_serial
));
2089 rv
= scsi_get_serial(fd
, scsi_serial
, sizeof(scsi_serial
));
2091 if (rv
&& check_env("IMSM_DEVNAME_AS_SERIAL")) {
2092 memset(serial
, 0, MAX_RAID_SERIAL_LEN
);
2093 fd2devname(fd
, (char *) serial
);
2100 Name
": Failed to retrieve serial for %s\n",
2105 rsp_len
= scsi_serial
[3];
2109 Name
": Failed to retrieve serial for %s\n",
2113 rsp_buf
= (char *) &scsi_serial
[4];
2115 /* trim all whitespace and non-printable characters and convert
2118 for (i
= 0, dest
= rsp_buf
; i
< rsp_len
; i
++) {
2121 /* ':' is reserved for use in placeholder serial
2122 * numbers for missing disks
2130 len
= dest
- rsp_buf
;
2133 /* truncate leading characters */
2134 if (len
> MAX_RAID_SERIAL_LEN
) {
2135 dest
+= len
- MAX_RAID_SERIAL_LEN
;
2136 len
= MAX_RAID_SERIAL_LEN
;
2139 memset(serial
, 0, MAX_RAID_SERIAL_LEN
);
2140 memcpy(serial
, dest
, len
);
2145 static int serialcmp(__u8
*s1
, __u8
*s2
)
2147 return strncmp((char *) s1
, (char *) s2
, MAX_RAID_SERIAL_LEN
);
2150 static void serialcpy(__u8
*dest
, __u8
*src
)
2152 strncpy((char *) dest
, (char *) src
, MAX_RAID_SERIAL_LEN
);
2156 static struct dl
*serial_to_dl(__u8
*serial
, struct intel_super
*super
)
2160 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
2161 if (serialcmp(dl
->serial
, serial
) == 0)
2168 static struct imsm_disk
*
2169 __serial_to_disk(__u8
*serial
, struct imsm_super
*mpb
, int *idx
)
2173 for (i
= 0; i
< mpb
->num_disks
; i
++) {
2174 struct imsm_disk
*disk
= __get_imsm_disk(mpb
, i
);
2176 if (serialcmp(disk
->serial
, serial
) == 0) {
2187 load_imsm_disk(int fd
, struct intel_super
*super
, char *devname
, int keep_fd
)
2189 struct imsm_disk
*disk
;
2194 __u8 serial
[MAX_RAID_SERIAL_LEN
];
2196 rv
= imsm_read_serial(fd
, devname
, serial
);
2201 dl
= calloc(1, sizeof(*dl
));
2205 Name
": failed to allocate disk buffer for %s\n",
2211 dl
->major
= major(stb
.st_rdev
);
2212 dl
->minor
= minor(stb
.st_rdev
);
2213 dl
->next
= super
->disks
;
2214 dl
->fd
= keep_fd
? fd
: -1;
2215 assert(super
->disks
== NULL
);
2217 serialcpy(dl
->serial
, serial
);
2220 fd2devname(fd
, name
);
2222 dl
->devname
= strdup(devname
);
2224 dl
->devname
= strdup(name
);
2226 /* look up this disk's index in the current anchor */
2227 disk
= __serial_to_disk(dl
->serial
, super
->anchor
, &dl
->index
);
2230 /* only set index on disks that are a member of a
2231 * populated contianer, i.e. one with raid_devs
2233 if (is_failed(&dl
->disk
))
2235 else if (is_spare(&dl
->disk
))
2243 /* When migrating map0 contains the 'destination' state while map1
2244 * contains the current state. When not migrating map0 contains the
2245 * current state. This routine assumes that map[0].map_state is set to
2246 * the current array state before being called.
2248 * Migration is indicated by one of the following states
2249 * 1/ Idle (migr_state=0 map0state=normal||unitialized||degraded||failed)
2250 * 2/ Initialize (migr_state=1 migr_type=MIGR_INIT map0state=normal
2251 * map1state=unitialized)
2252 * 3/ Repair (Resync) (migr_state=1 migr_type=MIGR_REPAIR map0state=normal
2254 * 4/ Rebuild (migr_state=1 migr_type=MIGR_REBUILD map0state=normal
2255 * map1state=degraded)
2257 static void migrate(struct imsm_dev
*dev
, __u8 to_state
, int migr_type
)
2259 struct imsm_map
*dest
;
2260 struct imsm_map
*src
= get_imsm_map(dev
, 0);
2262 dev
->vol
.migr_state
= 1;
2263 set_migr_type(dev
, migr_type
);
2264 dev
->vol
.curr_migr_unit
= 0;
2265 dest
= get_imsm_map(dev
, 1);
2267 /* duplicate and then set the target end state in map[0] */
2268 memcpy(dest
, src
, sizeof_imsm_map(src
));
2269 if ((migr_type
== MIGR_REBUILD
) ||
2270 (migr_type
== MIGR_GEN_MIGR
)) {
2274 for (i
= 0; i
< src
->num_members
; i
++) {
2275 ord
= __le32_to_cpu(src
->disk_ord_tbl
[i
]);
2276 set_imsm_ord_tbl_ent(src
, i
, ord_to_idx(ord
));
2280 src
->map_state
= to_state
;
2283 static void end_migration(struct imsm_dev
*dev
, __u8 map_state
)
2285 struct imsm_map
*map
= get_imsm_map(dev
, 0);
2286 struct imsm_map
*prev
= get_imsm_map(dev
, dev
->vol
.migr_state
);
2289 /* merge any IMSM_ORD_REBUILD bits that were not successfully
2290 * completed in the last migration.
2292 * FIXME add support for raid-level-migration
2294 for (i
= 0; i
< prev
->num_members
; i
++)
2295 for (j
= 0; j
< map
->num_members
; j
++)
2296 /* during online capacity expansion
2297 * disks position can be changed if takeover is used
2299 if (ord_to_idx(map
->disk_ord_tbl
[j
]) ==
2300 ord_to_idx(prev
->disk_ord_tbl
[i
])) {
2301 map
->disk_ord_tbl
[j
] |= prev
->disk_ord_tbl
[i
];
2305 dev
->vol
.migr_state
= 0;
2306 dev
->vol
.migr_type
= 0;
2307 dev
->vol
.curr_migr_unit
= 0;
2308 map
->map_state
= map_state
;
2312 static int parse_raid_devices(struct intel_super
*super
)
2315 struct imsm_dev
*dev_new
;
2316 size_t len
, len_migr
;
2317 size_t space_needed
= 0;
2318 struct imsm_super
*mpb
= super
->anchor
;
2320 for (i
= 0; i
< super
->anchor
->num_raid_devs
; i
++) {
2321 struct imsm_dev
*dev_iter
= __get_imsm_dev(super
->anchor
, i
);
2322 struct intel_dev
*dv
;
2324 len
= sizeof_imsm_dev(dev_iter
, 0);
2325 len_migr
= sizeof_imsm_dev(dev_iter
, 1);
2327 space_needed
+= len_migr
- len
;
2329 dv
= malloc(sizeof(*dv
));
2332 dev_new
= malloc(len_migr
);
2337 imsm_copy_dev(dev_new
, dev_iter
);
2340 dv
->next
= super
->devlist
;
2341 super
->devlist
= dv
;
2344 /* ensure that super->buf is large enough when all raid devices
2347 if (__le32_to_cpu(mpb
->mpb_size
) + space_needed
> super
->len
) {
2350 len
= ROUND_UP(__le32_to_cpu(mpb
->mpb_size
) + space_needed
, 512);
2351 if (posix_memalign(&buf
, 512, len
) != 0)
2354 memcpy(buf
, super
->buf
, super
->len
);
2355 memset(buf
+ super
->len
, 0, len
- super
->len
);
2364 /* retrieve a pointer to the bbm log which starts after all raid devices */
2365 struct bbm_log
*__get_imsm_bbm_log(struct imsm_super
*mpb
)
2369 if (__le32_to_cpu(mpb
->bbm_log_size
)) {
2371 ptr
+= mpb
->mpb_size
- __le32_to_cpu(mpb
->bbm_log_size
);
2377 static void __free_imsm(struct intel_super
*super
, int free_disks
);
2379 /* load_imsm_mpb - read matrix metadata
2380 * allocates super->mpb to be freed by free_super
2382 static int load_imsm_mpb(int fd
, struct intel_super
*super
, char *devname
)
2384 unsigned long long dsize
;
2385 unsigned long long sectors
;
2387 struct imsm_super
*anchor
;
2390 get_dev_size(fd
, NULL
, &dsize
);
2394 Name
": %s: device to small for imsm\n",
2399 if (lseek64(fd
, dsize
- (512 * 2), SEEK_SET
) < 0) {
2402 Name
": Cannot seek to anchor block on %s: %s\n",
2403 devname
, strerror(errno
));
2407 if (posix_memalign((void**)&anchor
, 512, 512) != 0) {
2410 Name
": Failed to allocate imsm anchor buffer"
2411 " on %s\n", devname
);
2414 if (read(fd
, anchor
, 512) != 512) {
2417 Name
": Cannot read anchor block on %s: %s\n",
2418 devname
, strerror(errno
));
2423 if (strncmp((char *) anchor
->sig
, MPB_SIGNATURE
, MPB_SIG_LEN
) != 0) {
2426 Name
": no IMSM anchor on %s\n", devname
);
2431 __free_imsm(super
, 0);
2432 super
->len
= ROUND_UP(anchor
->mpb_size
, 512);
2433 if (posix_memalign(&super
->buf
, 512, super
->len
) != 0) {
2436 Name
": unable to allocate %zu byte mpb buffer\n",
2441 memcpy(super
->buf
, anchor
, 512);
2443 sectors
= mpb_sectors(anchor
) - 1;
2446 check_sum
= __gen_imsm_checksum(super
->anchor
);
2447 if (check_sum
!= __le32_to_cpu(super
->anchor
->check_sum
)) {
2450 Name
": IMSM checksum %x != %x on %s\n",
2452 __le32_to_cpu(super
->anchor
->check_sum
),
2460 /* read the extended mpb */
2461 if (lseek64(fd
, dsize
- (512 * (2 + sectors
)), SEEK_SET
) < 0) {
2464 Name
": Cannot seek to extended mpb on %s: %s\n",
2465 devname
, strerror(errno
));
2469 if ((unsigned)read(fd
, super
->buf
+ 512, super
->len
- 512) != super
->len
- 512) {
2472 Name
": Cannot read extended mpb on %s: %s\n",
2473 devname
, strerror(errno
));
2477 check_sum
= __gen_imsm_checksum(super
->anchor
);
2478 if (check_sum
!= __le32_to_cpu(super
->anchor
->check_sum
)) {
2481 Name
": IMSM checksum %x != %x on %s\n",
2482 check_sum
, __le32_to_cpu(super
->anchor
->check_sum
),
2487 /* FIXME the BBM log is disk specific so we cannot use this global
2488 * buffer for all disks. Ok for now since we only look at the global
2489 * bbm_log_size parameter to gate assembly
2491 super
->bbm_log
= __get_imsm_bbm_log(super
->anchor
);
2497 load_and_parse_mpb(int fd
, struct intel_super
*super
, char *devname
, int keep_fd
)
2501 err
= load_imsm_mpb(fd
, super
, devname
);
2504 err
= load_imsm_disk(fd
, super
, devname
, keep_fd
);
2507 err
= parse_raid_devices(super
);
2512 static void __free_imsm_disk(struct dl
*d
)
2524 static void free_imsm_disks(struct intel_super
*super
)
2528 while (super
->disks
) {
2530 super
->disks
= d
->next
;
2531 __free_imsm_disk(d
);
2533 while (super
->missing
) {
2535 super
->missing
= d
->next
;
2536 __free_imsm_disk(d
);
2541 /* free all the pieces hanging off of a super pointer */
2542 static void __free_imsm(struct intel_super
*super
, int free_disks
)
2549 free_imsm_disks(super
);
2550 free_devlist(super
);
2552 free((void *) super
->hba
);
2557 static void free_imsm(struct intel_super
*super
)
2559 __free_imsm(super
, 1);
2563 static void free_super_imsm(struct supertype
*st
)
2565 struct intel_super
*super
= st
->sb
;
2574 static struct intel_super
*alloc_super(void)
2576 struct intel_super
*super
= malloc(sizeof(*super
));
2579 memset(super
, 0, sizeof(*super
));
2580 super
->current_vol
= -1;
2581 super
->create_offset
= ~((__u32
) 0);
2582 if (!check_env("IMSM_NO_PLATFORM"))
2583 super
->orom
= find_imsm_orom();
2584 if (super
->orom
&& !check_env("IMSM_TEST_OROM")) {
2585 struct sys_dev
*list
, *ent
;
2587 /* find the first intel ahci controller */
2588 list
= find_driver_devices("pci", "ahci");
2589 for (ent
= list
; ent
; ent
= ent
->next
)
2590 if (devpath_to_vendor(ent
->path
) == 0x8086)
2593 super
->hba
= ent
->path
;
2596 free_sys_dev(&list
);
2604 /* find_missing - helper routine for load_super_imsm_all that identifies
2605 * disks that have disappeared from the system. This routine relies on
2606 * the mpb being uptodate, which it is at load time.
2608 static int find_missing(struct intel_super
*super
)
2611 struct imsm_super
*mpb
= super
->anchor
;
2613 struct imsm_disk
*disk
;
2615 for (i
= 0; i
< mpb
->num_disks
; i
++) {
2616 disk
= __get_imsm_disk(mpb
, i
);
2617 dl
= serial_to_dl(disk
->serial
, super
);
2621 dl
= malloc(sizeof(*dl
));
2627 dl
->devname
= strdup("missing");
2629 serialcpy(dl
->serial
, disk
->serial
);
2632 dl
->next
= super
->missing
;
2633 super
->missing
= dl
;
2639 static struct intel_disk
*disk_list_get(__u8
*serial
, struct intel_disk
*disk_list
)
2641 struct intel_disk
*idisk
= disk_list
;
2644 if (serialcmp(idisk
->disk
.serial
, serial
) == 0)
2646 idisk
= idisk
->next
;
2652 static int __prep_thunderdome(struct intel_super
**table
, int tbl_size
,
2653 struct intel_super
*super
,
2654 struct intel_disk
**disk_list
)
2656 struct imsm_disk
*d
= &super
->disks
->disk
;
2657 struct imsm_super
*mpb
= super
->anchor
;
2660 for (i
= 0; i
< tbl_size
; i
++) {
2661 struct imsm_super
*tbl_mpb
= table
[i
]->anchor
;
2662 struct imsm_disk
*tbl_d
= &table
[i
]->disks
->disk
;
2664 if (tbl_mpb
->family_num
== mpb
->family_num
) {
2665 if (tbl_mpb
->check_sum
== mpb
->check_sum
) {
2666 dprintf("%s: mpb from %d:%d matches %d:%d\n",
2667 __func__
, super
->disks
->major
,
2668 super
->disks
->minor
,
2669 table
[i
]->disks
->major
,
2670 table
[i
]->disks
->minor
);
2674 if (((is_configured(d
) && !is_configured(tbl_d
)) ||
2675 is_configured(d
) == is_configured(tbl_d
)) &&
2676 tbl_mpb
->generation_num
< mpb
->generation_num
) {
2677 /* current version of the mpb is a
2678 * better candidate than the one in
2679 * super_table, but copy over "cross
2680 * generational" status
2682 struct intel_disk
*idisk
;
2684 dprintf("%s: mpb from %d:%d replaces %d:%d\n",
2685 __func__
, super
->disks
->major
,
2686 super
->disks
->minor
,
2687 table
[i
]->disks
->major
,
2688 table
[i
]->disks
->minor
);
2690 idisk
= disk_list_get(tbl_d
->serial
, *disk_list
);
2691 if (idisk
&& is_failed(&idisk
->disk
))
2692 tbl_d
->status
|= FAILED_DISK
;
2695 struct intel_disk
*idisk
;
2696 struct imsm_disk
*disk
;
2698 /* tbl_mpb is more up to date, but copy
2699 * over cross generational status before
2702 disk
= __serial_to_disk(d
->serial
, mpb
, NULL
);
2703 if (disk
&& is_failed(disk
))
2704 d
->status
|= FAILED_DISK
;
2706 idisk
= disk_list_get(d
->serial
, *disk_list
);
2709 if (disk
&& is_configured(disk
))
2710 idisk
->disk
.status
|= CONFIGURED_DISK
;
2713 dprintf("%s: mpb from %d:%d prefer %d:%d\n",
2714 __func__
, super
->disks
->major
,
2715 super
->disks
->minor
,
2716 table
[i
]->disks
->major
,
2717 table
[i
]->disks
->minor
);
2725 table
[tbl_size
++] = super
;
2729 /* update/extend the merged list of imsm_disk records */
2730 for (j
= 0; j
< mpb
->num_disks
; j
++) {
2731 struct imsm_disk
*disk
= __get_imsm_disk(mpb
, j
);
2732 struct intel_disk
*idisk
;
2734 idisk
= disk_list_get(disk
->serial
, *disk_list
);
2736 idisk
->disk
.status
|= disk
->status
;
2737 if (is_configured(&idisk
->disk
) ||
2738 is_failed(&idisk
->disk
))
2739 idisk
->disk
.status
&= ~(SPARE_DISK
);
2741 idisk
= calloc(1, sizeof(*idisk
));
2744 idisk
->owner
= IMSM_UNKNOWN_OWNER
;
2745 idisk
->disk
= *disk
;
2746 idisk
->next
= *disk_list
;
2750 if (serialcmp(idisk
->disk
.serial
, d
->serial
) == 0)
2757 static struct intel_super
*
2758 validate_members(struct intel_super
*super
, struct intel_disk
*disk_list
,
2761 struct imsm_super
*mpb
= super
->anchor
;
2765 for (i
= 0; i
< mpb
->num_disks
; i
++) {
2766 struct imsm_disk
*disk
= __get_imsm_disk(mpb
, i
);
2767 struct intel_disk
*idisk
;
2769 idisk
= disk_list_get(disk
->serial
, disk_list
);
2771 if (idisk
->owner
== owner
||
2772 idisk
->owner
== IMSM_UNKNOWN_OWNER
)
2775 dprintf("%s: '%.16s' owner %d != %d\n",
2776 __func__
, disk
->serial
, idisk
->owner
,
2779 dprintf("%s: unknown disk %x [%d]: %.16s\n",
2780 __func__
, __le32_to_cpu(mpb
->family_num
), i
,
2786 if (ok_count
== mpb
->num_disks
)
2791 static void show_conflicts(__u32 family_num
, struct intel_super
*super_list
)
2793 struct intel_super
*s
;
2795 for (s
= super_list
; s
; s
= s
->next
) {
2796 if (family_num
!= s
->anchor
->family_num
)
2798 fprintf(stderr
, "Conflict, offlining family %#x on '%s'\n",
2799 __le32_to_cpu(family_num
), s
->disks
->devname
);
2803 static struct intel_super
*
2804 imsm_thunderdome(struct intel_super
**super_list
, int len
)
2806 struct intel_super
*super_table
[len
];
2807 struct intel_disk
*disk_list
= NULL
;
2808 struct intel_super
*champion
, *spare
;
2809 struct intel_super
*s
, **del
;
2814 memset(super_table
, 0, sizeof(super_table
));
2815 for (s
= *super_list
; s
; s
= s
->next
)
2816 tbl_size
= __prep_thunderdome(super_table
, tbl_size
, s
, &disk_list
);
2818 for (i
= 0; i
< tbl_size
; i
++) {
2819 struct imsm_disk
*d
;
2820 struct intel_disk
*idisk
;
2821 struct imsm_super
*mpb
= super_table
[i
]->anchor
;
2824 d
= &s
->disks
->disk
;
2826 /* 'd' must appear in merged disk list for its
2827 * configuration to be valid
2829 idisk
= disk_list_get(d
->serial
, disk_list
);
2830 if (idisk
&& idisk
->owner
== i
)
2831 s
= validate_members(s
, disk_list
, i
);
2836 dprintf("%s: marking family: %#x from %d:%d offline\n",
2837 __func__
, mpb
->family_num
,
2838 super_table
[i
]->disks
->major
,
2839 super_table
[i
]->disks
->minor
);
2843 /* This is where the mdadm implementation differs from the Windows
2844 * driver which has no strict concept of a container. We can only
2845 * assemble one family from a container, so when returning a prodigal
2846 * array member to this system the code will not be able to disambiguate
2847 * the container contents that should be assembled ("foreign" versus
2848 * "local"). It requires user intervention to set the orig_family_num
2849 * to a new value to establish a new container. The Windows driver in
2850 * this situation fixes up the volume name in place and manages the
2851 * foreign array as an independent entity.
2856 for (i
= 0; i
< tbl_size
; i
++) {
2857 struct intel_super
*tbl_ent
= super_table
[i
];
2863 if (tbl_ent
->anchor
->num_raid_devs
== 0) {
2868 if (s
&& !is_spare
) {
2869 show_conflicts(tbl_ent
->anchor
->family_num
, *super_list
);
2871 } else if (!s
&& !is_spare
)
2884 fprintf(stderr
, "Chose family %#x on '%s', "
2885 "assemble conflicts to new container with '--update=uuid'\n",
2886 __le32_to_cpu(s
->anchor
->family_num
), s
->disks
->devname
);
2888 /* collect all dl's onto 'champion', and update them to
2889 * champion's version of the status
2891 for (s
= *super_list
; s
; s
= s
->next
) {
2892 struct imsm_super
*mpb
= champion
->anchor
;
2893 struct dl
*dl
= s
->disks
;
2898 for (i
= 0; i
< mpb
->num_disks
; i
++) {
2899 struct imsm_disk
*disk
;
2901 disk
= __serial_to_disk(dl
->serial
, mpb
, &dl
->index
);
2904 /* only set index on disks that are a member of
2905 * a populated contianer, i.e. one with
2908 if (is_failed(&dl
->disk
))
2910 else if (is_spare(&dl
->disk
))
2916 if (i
>= mpb
->num_disks
) {
2917 struct intel_disk
*idisk
;
2919 idisk
= disk_list_get(dl
->serial
, disk_list
);
2920 if (idisk
&& is_spare(&idisk
->disk
) &&
2921 !is_failed(&idisk
->disk
) && !is_configured(&idisk
->disk
))
2929 dl
->next
= champion
->disks
;
2930 champion
->disks
= dl
;
2934 /* delete 'champion' from super_list */
2935 for (del
= super_list
; *del
; ) {
2936 if (*del
== champion
) {
2937 *del
= (*del
)->next
;
2940 del
= &(*del
)->next
;
2942 champion
->next
= NULL
;
2946 struct intel_disk
*idisk
= disk_list
;
2948 disk_list
= disk_list
->next
;
2955 static int load_super_imsm_all(struct supertype
*st
, int fd
, void **sbp
,
2959 struct intel_super
*super_list
= NULL
;
2960 struct intel_super
*super
= NULL
;
2961 int devnum
= fd2devnum(fd
);
2967 /* check if 'fd' an opened container */
2968 sra
= sysfs_read(fd
, 0, GET_LEVEL
|GET_VERSION
|GET_DEVS
|GET_STATE
);
2972 if (sra
->array
.major_version
!= -1 ||
2973 sra
->array
.minor_version
!= -2 ||
2974 strcmp(sra
->text_version
, "imsm") != 0) {
2979 for (sd
= sra
->devs
, i
= 0; sd
; sd
= sd
->next
, i
++) {
2980 struct intel_super
*s
= alloc_super();
2987 s
->next
= super_list
;
2991 sprintf(nm
, "%d:%d", sd
->disk
.major
, sd
->disk
.minor
);
2992 dfd
= dev_open(nm
, O_RDWR
);
2996 err
= load_and_parse_mpb(dfd
, s
, NULL
, 1);
2998 /* retry the load if we might have raced against mdmon */
2999 if (err
== 3 && mdmon_running(devnum
))
3000 for (retry
= 0; retry
< 3; retry
++) {
3002 err
= load_and_parse_mpb(dfd
, s
, NULL
, 1);
3010 /* all mpbs enter, maybe one leaves */
3011 super
= imsm_thunderdome(&super_list
, i
);
3017 if (find_missing(super
) != 0) {
3025 while (super_list
) {
3026 struct intel_super
*s
= super_list
;
3028 super_list
= super_list
->next
;
3037 st
->container_dev
= devnum
;
3038 if (err
== 0 && st
->ss
== NULL
) {
3039 st
->ss
= &super_imsm
;
3040 st
->minor_version
= 0;
3041 st
->max_devs
= IMSM_MAX_DEVICES
;
3046 static int load_container_imsm(struct supertype
*st
, int fd
, char *devname
)
3048 return load_super_imsm_all(st
, fd
, &st
->sb
, devname
);
3052 static int load_super_imsm(struct supertype
*st
, int fd
, char *devname
)
3054 struct intel_super
*super
;
3057 if (test_partition(fd
))
3058 /* IMSM not allowed on partitions */
3061 free_super_imsm(st
);
3063 super
= alloc_super();
3066 Name
": malloc of %zu failed.\n",
3071 rv
= load_and_parse_mpb(fd
, super
, devname
, 0);
3076 Name
": Failed to load all information "
3077 "sections on %s\n", devname
);
3083 if (st
->ss
== NULL
) {
3084 st
->ss
= &super_imsm
;
3085 st
->minor_version
= 0;
3086 st
->max_devs
= IMSM_MAX_DEVICES
;
3091 static __u16
info_to_blocks_per_strip(mdu_array_info_t
*info
)
3093 if (info
->level
== 1)
3095 return info
->chunk_size
>> 9;
3098 static __u32
info_to_num_data_stripes(mdu_array_info_t
*info
, int num_domains
)
3102 num_stripes
= (info
->size
* 2) / info_to_blocks_per_strip(info
);
3103 num_stripes
/= num_domains
;
3108 static __u32
info_to_blocks_per_member(mdu_array_info_t
*info
)
3110 if (info
->level
== 1)
3111 return info
->size
* 2;
3113 return (info
->size
* 2) & ~(info_to_blocks_per_strip(info
) - 1);
3116 static void imsm_update_version_info(struct intel_super
*super
)
3118 /* update the version and attributes */
3119 struct imsm_super
*mpb
= super
->anchor
;
3121 struct imsm_dev
*dev
;
3122 struct imsm_map
*map
;
3125 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
3126 dev
= get_imsm_dev(super
, i
);
3127 map
= get_imsm_map(dev
, 0);
3128 if (__le32_to_cpu(dev
->size_high
) > 0)
3129 mpb
->attributes
|= MPB_ATTRIB_2TB
;
3131 /* FIXME detect when an array spans a port multiplier */
3133 mpb
->attributes
|= MPB_ATTRIB_PM
;
3136 if (mpb
->num_raid_devs
> 1 ||
3137 mpb
->attributes
!= MPB_ATTRIB_CHECKSUM_VERIFY
) {
3138 version
= MPB_VERSION_ATTRIBS
;
3139 switch (get_imsm_raid_level(map
)) {
3140 case 0: mpb
->attributes
|= MPB_ATTRIB_RAID0
; break;
3141 case 1: mpb
->attributes
|= MPB_ATTRIB_RAID1
; break;
3142 case 10: mpb
->attributes
|= MPB_ATTRIB_RAID10
; break;
3143 case 5: mpb
->attributes
|= MPB_ATTRIB_RAID5
; break;
3146 if (map
->num_members
>= 5)
3147 version
= MPB_VERSION_5OR6_DISK_ARRAY
;
3148 else if (dev
->status
== DEV_CLONE_N_GO
)
3149 version
= MPB_VERSION_CNG
;
3150 else if (get_imsm_raid_level(map
) == 5)
3151 version
= MPB_VERSION_RAID5
;
3152 else if (map
->num_members
>= 3)
3153 version
= MPB_VERSION_3OR4_DISK_ARRAY
;
3154 else if (get_imsm_raid_level(map
) == 1)
3155 version
= MPB_VERSION_RAID1
;
3157 version
= MPB_VERSION_RAID0
;
3159 strcpy(((char *) mpb
->sig
) + strlen(MPB_SIGNATURE
), version
);
3163 static int check_name(struct intel_super
*super
, char *name
, int quiet
)
3165 struct imsm_super
*mpb
= super
->anchor
;
3166 char *reason
= NULL
;
3169 if (strlen(name
) > MAX_RAID_SERIAL_LEN
)
3170 reason
= "must be 16 characters or less";
3172 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
3173 struct imsm_dev
*dev
= get_imsm_dev(super
, i
);
3175 if (strncmp((char *) dev
->volume
, name
, MAX_RAID_SERIAL_LEN
) == 0) {
3176 reason
= "already exists";
3181 if (reason
&& !quiet
)
3182 fprintf(stderr
, Name
": imsm volume name %s\n", reason
);
3187 static int init_super_imsm_volume(struct supertype
*st
, mdu_array_info_t
*info
,
3188 unsigned long long size
, char *name
,
3189 char *homehost
, int *uuid
)
3191 /* We are creating a volume inside a pre-existing container.
3192 * so st->sb is already set.
3194 struct intel_super
*super
= st
->sb
;
3195 struct imsm_super
*mpb
= super
->anchor
;
3196 struct intel_dev
*dv
;
3197 struct imsm_dev
*dev
;
3198 struct imsm_vol
*vol
;
3199 struct imsm_map
*map
;
3200 int idx
= mpb
->num_raid_devs
;
3202 unsigned long long array_blocks
;
3203 size_t size_old
, size_new
;
3204 __u32 num_data_stripes
;
3206 if (super
->orom
&& mpb
->num_raid_devs
>= super
->orom
->vpa
) {
3207 fprintf(stderr
, Name
": This imsm-container already has the "
3208 "maximum of %d volumes\n", super
->orom
->vpa
);
3212 /* ensure the mpb is large enough for the new data */
3213 size_old
= __le32_to_cpu(mpb
->mpb_size
);
3214 size_new
= disks_to_mpb_size(info
->nr_disks
);
3215 if (size_new
> size_old
) {
3217 size_t size_round
= ROUND_UP(size_new
, 512);
3219 if (posix_memalign(&mpb_new
, 512, size_round
) != 0) {
3220 fprintf(stderr
, Name
": could not allocate new mpb\n");
3223 memcpy(mpb_new
, mpb
, size_old
);
3226 super
->anchor
= mpb_new
;
3227 mpb
->mpb_size
= __cpu_to_le32(size_new
);
3228 memset(mpb_new
+ size_old
, 0, size_round
- size_old
);
3230 super
->current_vol
= idx
;
3231 /* when creating the first raid device in this container set num_disks
3232 * to zero, i.e. delete this spare and add raid member devices in
3233 * add_to_super_imsm_volume()
3235 if (super
->current_vol
== 0)
3238 if (!check_name(super
, name
, 0))
3240 dv
= malloc(sizeof(*dv
));
3242 fprintf(stderr
, Name
": failed to allocate device list entry\n");
3245 dev
= malloc(sizeof(*dev
) + sizeof(__u32
) * (info
->raid_disks
- 1));
3248 fprintf(stderr
, Name
": could not allocate raid device\n");
3251 strncpy((char *) dev
->volume
, name
, MAX_RAID_SERIAL_LEN
);
3252 if (info
->level
== 1)
3253 array_blocks
= info_to_blocks_per_member(info
);
3255 array_blocks
= calc_array_size(info
->level
, info
->raid_disks
,
3256 info
->layout
, info
->chunk_size
,
3258 /* round array size down to closest MB */
3259 array_blocks
= (array_blocks
>> SECT_PER_MB_SHIFT
) << SECT_PER_MB_SHIFT
;
3261 dev
->size_low
= __cpu_to_le32((__u32
) array_blocks
);
3262 dev
->size_high
= __cpu_to_le32((__u32
) (array_blocks
>> 32));
3263 dev
->status
= __cpu_to_le32(0);
3264 dev
->reserved_blocks
= __cpu_to_le32(0);
3266 vol
->migr_state
= 0;
3267 set_migr_type(dev
, MIGR_INIT
);
3269 vol
->curr_migr_unit
= 0;
3270 map
= get_imsm_map(dev
, 0);
3271 map
->pba_of_lba0
= __cpu_to_le32(super
->create_offset
);
3272 map
->blocks_per_member
= __cpu_to_le32(info_to_blocks_per_member(info
));
3273 map
->blocks_per_strip
= __cpu_to_le16(info_to_blocks_per_strip(info
));
3274 map
->failed_disk_num
= ~0;
3275 map
->map_state
= info
->level
? IMSM_T_STATE_UNINITIALIZED
:
3276 IMSM_T_STATE_NORMAL
;
3279 if (info
->level
== 1 && info
->raid_disks
> 2) {
3282 fprintf(stderr
, Name
": imsm does not support more than 2 disks"
3283 "in a raid1 volume\n");
3287 map
->raid_level
= info
->level
;
3288 if (info
->level
== 10) {
3289 map
->raid_level
= 1;
3290 map
->num_domains
= info
->raid_disks
/ 2;
3291 } else if (info
->level
== 1)
3292 map
->num_domains
= info
->raid_disks
;
3294 map
->num_domains
= 1;
3296 num_data_stripes
= info_to_num_data_stripes(info
, map
->num_domains
);
3297 map
->num_data_stripes
= __cpu_to_le32(num_data_stripes
);
3299 map
->num_members
= info
->raid_disks
;
3300 for (i
= 0; i
< map
->num_members
; i
++) {
3301 /* initialized in add_to_super */
3302 set_imsm_ord_tbl_ent(map
, i
, IMSM_ORD_REBUILD
);
3304 mpb
->num_raid_devs
++;
3307 dv
->index
= super
->current_vol
;
3308 dv
->next
= super
->devlist
;
3309 super
->devlist
= dv
;
3311 imsm_update_version_info(super
);
3316 static int init_super_imsm(struct supertype
*st
, mdu_array_info_t
*info
,
3317 unsigned long long size
, char *name
,
3318 char *homehost
, int *uuid
)
3320 /* This is primarily called by Create when creating a new array.
3321 * We will then get add_to_super called for each component, and then
3322 * write_init_super called to write it out to each device.
3323 * For IMSM, Create can create on fresh devices or on a pre-existing
3325 * To create on a pre-existing array a different method will be called.
3326 * This one is just for fresh drives.
3328 struct intel_super
*super
;
3329 struct imsm_super
*mpb
;
3334 return init_super_imsm_volume(st
, info
, size
, name
, homehost
, uuid
);
3337 mpb_size
= disks_to_mpb_size(info
->nr_disks
);
3341 super
= alloc_super();
3342 if (super
&& posix_memalign(&super
->buf
, 512, mpb_size
) != 0) {
3347 fprintf(stderr
, Name
3348 ": %s could not allocate superblock\n", __func__
);
3351 memset(super
->buf
, 0, mpb_size
);
3353 mpb
->mpb_size
= __cpu_to_le32(mpb_size
);
3357 /* zeroing superblock */
3361 mpb
->attributes
= MPB_ATTRIB_CHECKSUM_VERIFY
;
3363 version
= (char *) mpb
->sig
;
3364 strcpy(version
, MPB_SIGNATURE
);
3365 version
+= strlen(MPB_SIGNATURE
);
3366 strcpy(version
, MPB_VERSION_RAID0
);
3372 static int add_to_super_imsm_volume(struct supertype
*st
, mdu_disk_info_t
*dk
,
3373 int fd
, char *devname
)
3375 struct intel_super
*super
= st
->sb
;
3376 struct imsm_super
*mpb
= super
->anchor
;
3378 struct imsm_dev
*dev
;
3379 struct imsm_map
*map
;
3382 dev
= get_imsm_dev(super
, super
->current_vol
);
3383 map
= get_imsm_map(dev
, 0);
3385 if (! (dk
->state
& (1<<MD_DISK_SYNC
))) {
3386 fprintf(stderr
, Name
": %s: Cannot add spare devices to IMSM volume\n",
3392 /* we're doing autolayout so grab the pre-marked (in
3393 * validate_geometry) raid_disk
3395 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
3396 if (dl
->raiddisk
== dk
->raid_disk
)
3399 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
3400 if (dl
->major
== dk
->major
&&
3401 dl
->minor
== dk
->minor
)
3406 fprintf(stderr
, Name
": %s is not a member of the same container\n", devname
);
3410 /* add a pristine spare to the metadata */
3411 if (dl
->index
< 0) {
3412 dl
->index
= super
->anchor
->num_disks
;
3413 super
->anchor
->num_disks
++;
3415 /* Check the device has not already been added */
3416 slot
= get_imsm_disk_slot(map
, dl
->index
);
3418 (get_imsm_ord_tbl_ent(dev
, slot
, -1) & IMSM_ORD_REBUILD
) == 0) {
3419 fprintf(stderr
, Name
": %s has been included in this array twice\n",
3423 set_imsm_ord_tbl_ent(map
, dk
->number
, dl
->index
);
3424 dl
->disk
.status
= CONFIGURED_DISK
;
3426 /* if we are creating the first raid device update the family number */
3427 if (super
->current_vol
== 0) {
3429 struct imsm_dev
*_dev
= __get_imsm_dev(mpb
, 0);
3430 struct imsm_disk
*_disk
= __get_imsm_disk(mpb
, dl
->index
);
3432 if (!_dev
|| !_disk
) {
3433 fprintf(stderr
, Name
": BUG mpb setup error\n");
3439 sum
+= __gen_imsm_checksum(mpb
);
3440 mpb
->family_num
= __cpu_to_le32(sum
);
3441 mpb
->orig_family_num
= mpb
->family_num
;
3447 static int add_to_super_imsm(struct supertype
*st
, mdu_disk_info_t
*dk
,
3448 int fd
, char *devname
)
3450 struct intel_super
*super
= st
->sb
;
3452 unsigned long long size
;
3457 /* if we are on an RAID enabled platform check that the disk is
3458 * attached to the raid controller
3460 if (super
->hba
&& !disk_attached_to_hba(fd
, super
->hba
)) {
3462 Name
": %s is not attached to the raid controller: %s\n",
3463 devname
? : "disk", super
->hba
);
3467 if (super
->current_vol
>= 0)
3468 return add_to_super_imsm_volume(st
, dk
, fd
, devname
);
3471 dd
= malloc(sizeof(*dd
));
3474 Name
": malloc failed %s:%d.\n", __func__
, __LINE__
);
3477 memset(dd
, 0, sizeof(*dd
));
3478 dd
->major
= major(stb
.st_rdev
);
3479 dd
->minor
= minor(stb
.st_rdev
);
3481 dd
->devname
= devname
? strdup(devname
) : NULL
;
3484 dd
->action
= DISK_ADD
;
3485 rv
= imsm_read_serial(fd
, devname
, dd
->serial
);
3488 Name
": failed to retrieve scsi serial, aborting\n");
3493 get_dev_size(fd
, NULL
, &size
);
3495 serialcpy(dd
->disk
.serial
, dd
->serial
);
3496 dd
->disk
.total_blocks
= __cpu_to_le32(size
);
3497 dd
->disk
.status
= SPARE_DISK
;
3498 if (sysfs_disk_to_scsi_id(fd
, &id
) == 0)
3499 dd
->disk
.scsi_id
= __cpu_to_le32(id
);
3501 dd
->disk
.scsi_id
= __cpu_to_le32(0);
3503 if (st
->update_tail
) {
3504 dd
->next
= super
->disk_mgmt_list
;
3505 super
->disk_mgmt_list
= dd
;
3507 dd
->next
= super
->disks
;
3515 static int remove_from_super_imsm(struct supertype
*st
, mdu_disk_info_t
*dk
)
3517 struct intel_super
*super
= st
->sb
;
3520 /* remove from super works only in mdmon - for communication
3521 * manager - monitor. Check if communication memory buffer
3524 if (!st
->update_tail
) {
3526 Name
": %s shall be used in mdmon context only"
3527 "(line %d).\n", __func__
, __LINE__
);
3530 dd
= malloc(sizeof(*dd
));
3533 Name
": malloc failed %s:%d.\n", __func__
, __LINE__
);
3536 memset(dd
, 0, sizeof(*dd
));
3537 dd
->major
= dk
->major
;
3538 dd
->minor
= dk
->minor
;
3541 dd
->disk
.status
= SPARE_DISK
;
3542 dd
->action
= DISK_REMOVE
;
3544 dd
->next
= super
->disk_mgmt_list
;
3545 super
->disk_mgmt_list
= dd
;
3551 static int store_imsm_mpb(int fd
, struct imsm_super
*mpb
);
3555 struct imsm_super anchor
;
3556 } spare_record
__attribute__ ((aligned(512)));
3558 /* spare records have their own family number and do not have any defined raid
3561 static int write_super_imsm_spares(struct intel_super
*super
, int doclose
)
3563 struct imsm_super
*mpb
= super
->anchor
;
3564 struct imsm_super
*spare
= &spare_record
.anchor
;
3568 spare
->mpb_size
= __cpu_to_le32(sizeof(struct imsm_super
)),
3569 spare
->generation_num
= __cpu_to_le32(1UL),
3570 spare
->attributes
= MPB_ATTRIB_CHECKSUM_VERIFY
;
3571 spare
->num_disks
= 1,
3572 spare
->num_raid_devs
= 0,
3573 spare
->cache_size
= mpb
->cache_size
,
3574 spare
->pwr_cycle_count
= __cpu_to_le32(1),
3576 snprintf((char *) spare
->sig
, MAX_SIGNATURE_LENGTH
,
3577 MPB_SIGNATURE MPB_VERSION_RAID0
);
3579 for (d
= super
->disks
; d
; d
= d
->next
) {
3583 spare
->disk
[0] = d
->disk
;
3584 sum
= __gen_imsm_checksum(spare
);
3585 spare
->family_num
= __cpu_to_le32(sum
);
3586 spare
->orig_family_num
= 0;
3587 sum
= __gen_imsm_checksum(spare
);
3588 spare
->check_sum
= __cpu_to_le32(sum
);
3590 if (store_imsm_mpb(d
->fd
, spare
)) {
3591 fprintf(stderr
, "%s: failed for device %d:%d %s\n",
3592 __func__
, d
->major
, d
->minor
, strerror(errno
));
3604 static int write_super_imsm(struct supertype
*st
, int doclose
)
3606 struct intel_super
*super
= st
->sb
;
3607 struct imsm_super
*mpb
= super
->anchor
;
3613 __u32 mpb_size
= sizeof(struct imsm_super
) - sizeof(struct imsm_disk
);
3616 /* 'generation' is incremented everytime the metadata is written */
3617 generation
= __le32_to_cpu(mpb
->generation_num
);
3619 mpb
->generation_num
= __cpu_to_le32(generation
);
3621 /* fix up cases where previous mdadm releases failed to set
3624 if (mpb
->orig_family_num
== 0)
3625 mpb
->orig_family_num
= mpb
->family_num
;
3627 for (d
= super
->disks
; d
; d
= d
->next
) {
3631 mpb
->disk
[d
->index
] = d
->disk
;
3635 for (d
= super
->missing
; d
; d
= d
->next
) {
3636 mpb
->disk
[d
->index
] = d
->disk
;
3639 mpb
->num_disks
= num_disks
;
3640 mpb_size
+= sizeof(struct imsm_disk
) * mpb
->num_disks
;
3642 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
3643 struct imsm_dev
*dev
= __get_imsm_dev(mpb
, i
);
3644 struct imsm_dev
*dev2
= get_imsm_dev(super
, i
);
3646 imsm_copy_dev(dev
, dev2
);
3647 mpb_size
+= sizeof_imsm_dev(dev
, 0);
3650 mpb_size
+= __le32_to_cpu(mpb
->bbm_log_size
);
3651 mpb
->mpb_size
= __cpu_to_le32(mpb_size
);
3653 /* recalculate checksum */
3654 sum
= __gen_imsm_checksum(mpb
);
3655 mpb
->check_sum
= __cpu_to_le32(sum
);
3657 /* write the mpb for disks that compose raid devices */
3658 for (d
= super
->disks
; d
; d
= d
->next
) {
3661 if (store_imsm_mpb(d
->fd
, mpb
))
3662 fprintf(stderr
, "%s: failed for device %d:%d %s\n",
3663 __func__
, d
->major
, d
->minor
, strerror(errno
));
3671 return write_super_imsm_spares(super
, doclose
);
3677 static int create_array(struct supertype
*st
, int dev_idx
)
3680 struct imsm_update_create_array
*u
;
3681 struct intel_super
*super
= st
->sb
;
3682 struct imsm_dev
*dev
= get_imsm_dev(super
, dev_idx
);
3683 struct imsm_map
*map
= get_imsm_map(dev
, 0);
3684 struct disk_info
*inf
;
3685 struct imsm_disk
*disk
;
3688 len
= sizeof(*u
) - sizeof(*dev
) + sizeof_imsm_dev(dev
, 0) +
3689 sizeof(*inf
) * map
->num_members
;
3692 fprintf(stderr
, "%s: failed to allocate update buffer\n",
3697 u
->type
= update_create_array
;
3698 u
->dev_idx
= dev_idx
;
3699 imsm_copy_dev(&u
->dev
, dev
);
3700 inf
= get_disk_info(u
);
3701 for (i
= 0; i
< map
->num_members
; i
++) {
3702 int idx
= get_imsm_disk_idx(dev
, i
, -1);
3704 disk
= get_imsm_disk(super
, idx
);
3705 serialcpy(inf
[i
].serial
, disk
->serial
);
3707 append_metadata_update(st
, u
, len
);
3712 static int mgmt_disk(struct supertype
*st
)
3714 struct intel_super
*super
= st
->sb
;
3716 struct imsm_update_add_remove_disk
*u
;
3718 if (!super
->disk_mgmt_list
)
3724 fprintf(stderr
, "%s: failed to allocate update buffer\n",
3729 u
->type
= update_add_remove_disk
;
3730 append_metadata_update(st
, u
, len
);
3735 static int write_init_super_imsm(struct supertype
*st
)
3737 struct intel_super
*super
= st
->sb
;
3738 int current_vol
= super
->current_vol
;
3740 /* we are done with current_vol reset it to point st at the container */
3741 super
->current_vol
= -1;
3743 if (st
->update_tail
) {
3744 /* queue the recently created array / added disk
3745 * as a metadata update */
3749 /* determine if we are creating a volume or adding a disk */
3750 if (current_vol
< 0) {
3751 /* in the mgmt (add/remove) disk case we are running
3752 * in mdmon context, so don't close fd's
3754 return mgmt_disk(st
);
3756 rv
= create_array(st
, current_vol
);
3758 for (d
= super
->disks
; d
; d
= d
->next
) {
3766 for (d
= super
->disks
; d
; d
= d
->next
)
3767 Kill(d
->devname
, NULL
, 0, 1, 1);
3768 return write_super_imsm(st
, 1);
3773 static int store_super_imsm(struct supertype
*st
, int fd
)
3775 struct intel_super
*super
= st
->sb
;
3776 struct imsm_super
*mpb
= super
? super
->anchor
: NULL
;
3782 return store_imsm_mpb(fd
, mpb
);
3788 static int imsm_bbm_log_size(struct imsm_super
*mpb
)
3790 return __le32_to_cpu(mpb
->bbm_log_size
);
3794 static int validate_geometry_imsm_container(struct supertype
*st
, int level
,
3795 int layout
, int raiddisks
, int chunk
,
3796 unsigned long long size
, char *dev
,
3797 unsigned long long *freesize
,
3801 unsigned long long ldsize
;
3802 const struct imsm_orom
*orom
;
3804 if (level
!= LEVEL_CONTAINER
)
3809 if (check_env("IMSM_NO_PLATFORM"))
3812 orom
= find_imsm_orom();
3813 if (orom
&& raiddisks
> orom
->tds
) {
3815 fprintf(stderr
, Name
": %d exceeds maximum number of"
3816 " platform supported disks: %d\n",
3817 raiddisks
, orom
->tds
);
3821 fd
= open(dev
, O_RDONLY
|O_EXCL
, 0);
3824 fprintf(stderr
, Name
": imsm: Cannot open %s: %s\n",
3825 dev
, strerror(errno
));
3828 if (!get_dev_size(fd
, dev
, &ldsize
)) {
3834 *freesize
= avail_size_imsm(st
, ldsize
>> 9);
3839 static unsigned long long find_size(struct extent
*e
, int *idx
, int num_extents
)
3841 const unsigned long long base_start
= e
[*idx
].start
;
3842 unsigned long long end
= base_start
+ e
[*idx
].size
;
3845 if (base_start
== end
)
3849 for (i
= *idx
; i
< num_extents
; i
++) {
3850 /* extend overlapping extents */
3851 if (e
[i
].start
>= base_start
&&
3852 e
[i
].start
<= end
) {
3855 if (e
[i
].start
+ e
[i
].size
> end
)
3856 end
= e
[i
].start
+ e
[i
].size
;
3857 } else if (e
[i
].start
> end
) {
3863 return end
- base_start
;
3866 static unsigned long long merge_extents(struct intel_super
*super
, int sum_extents
)
3868 /* build a composite disk with all known extents and generate a new
3869 * 'maxsize' given the "all disks in an array must share a common start
3870 * offset" constraint
3872 struct extent
*e
= calloc(sum_extents
, sizeof(*e
));
3876 unsigned long long pos
;
3877 unsigned long long start
= 0;
3878 unsigned long long maxsize
;
3879 unsigned long reserve
;
3884 /* coalesce and sort all extents. also, check to see if we need to
3885 * reserve space between member arrays
3888 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
3891 for (i
= 0; i
< dl
->extent_cnt
; i
++)
3894 qsort(e
, sum_extents
, sizeof(*e
), cmp_extent
);
3899 while (i
< sum_extents
) {
3900 e
[j
].start
= e
[i
].start
;
3901 e
[j
].size
= find_size(e
, &i
, sum_extents
);
3903 if (e
[j
-1].size
== 0)
3912 unsigned long long esize
;
3914 esize
= e
[i
].start
- pos
;
3915 if (esize
>= maxsize
) {
3920 pos
= e
[i
].start
+ e
[i
].size
;
3922 } while (e
[i
-1].size
);
3928 /* FIXME assumes volume at offset 0 is the first volume in a
3931 if (start_extent
> 0)
3932 reserve
= IMSM_RESERVED_SECTORS
; /* gap between raid regions */
3936 if (maxsize
< reserve
)
3939 super
->create_offset
= ~((__u32
) 0);
3940 if (start
+ reserve
> super
->create_offset
)
3941 return 0; /* start overflows create_offset */
3942 super
->create_offset
= start
+ reserve
;
3944 return maxsize
- reserve
;
3947 static int is_raid_level_supported(const struct imsm_orom
*orom
, int level
, int raiddisks
)
3949 if (level
< 0 || level
== 6 || level
== 4)
3952 /* if we have an orom prevent invalid raid levels */
3955 case 0: return imsm_orom_has_raid0(orom
);
3958 return imsm_orom_has_raid1e(orom
);
3959 return imsm_orom_has_raid1(orom
) && raiddisks
== 2;
3960 case 10: return imsm_orom_has_raid10(orom
) && raiddisks
== 4;
3961 case 5: return imsm_orom_has_raid5(orom
) && raiddisks
> 2;
3964 return 1; /* not on an Intel RAID platform so anything goes */
3969 #define pr_vrb(fmt, arg...) (void) (verbose && fprintf(stderr, Name fmt, ##arg))
3971 validate_geometry_imsm_orom(struct intel_super
*super
, int level
, int layout
,
3972 int raiddisks
, int chunk
, int verbose
)
3974 if (!is_raid_level_supported(super
->orom
, level
, raiddisks
)) {
3975 pr_vrb(": platform does not support raid%d with %d disk%s\n",
3976 level
, raiddisks
, raiddisks
> 1 ? "s" : "");
3979 if (super
->orom
&& level
!= 1 &&
3980 !imsm_orom_has_chunk(super
->orom
, chunk
)) {
3981 pr_vrb(": platform does not support a chunk size of: %d\n", chunk
);
3984 if (layout
!= imsm_level_to_layout(level
)) {
3986 pr_vrb(": imsm raid 5 only supports the left-asymmetric layout\n");
3987 else if (level
== 10)
3988 pr_vrb(": imsm raid 10 only supports the n2 layout\n");
3990 pr_vrb(": imsm unknown layout %#x for this raid level %d\n",
3998 /* validate_geometry_imsm_volume - lifted from validate_geometry_ddf_bvd
3999 * FIX ME add ahci details
4001 static int validate_geometry_imsm_volume(struct supertype
*st
, int level
,
4002 int layout
, int raiddisks
, int chunk
,
4003 unsigned long long size
, char *dev
,
4004 unsigned long long *freesize
,
4008 struct intel_super
*super
= st
->sb
;
4009 struct imsm_super
*mpb
= super
->anchor
;
4011 unsigned long long pos
= 0;
4012 unsigned long long maxsize
;
4016 /* We must have the container info already read in. */
4020 if (!validate_geometry_imsm_orom(super
, level
, layout
, raiddisks
, chunk
, verbose
))
4024 /* General test: make sure there is space for
4025 * 'raiddisks' device extents of size 'size' at a given
4028 unsigned long long minsize
= size
;
4029 unsigned long long start_offset
= MaxSector
;
4032 minsize
= MPB_SECTOR_CNT
+ IMSM_RESERVED_SECTORS
;
4033 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
4038 e
= get_extents(super
, dl
);
4041 unsigned long long esize
;
4042 esize
= e
[i
].start
- pos
;
4043 if (esize
>= minsize
)
4045 if (found
&& start_offset
== MaxSector
) {
4048 } else if (found
&& pos
!= start_offset
) {
4052 pos
= e
[i
].start
+ e
[i
].size
;
4054 } while (e
[i
-1].size
);
4059 if (dcnt
< raiddisks
) {
4061 fprintf(stderr
, Name
": imsm: Not enough "
4062 "devices with space for this array "
4070 /* This device must be a member of the set */
4071 if (stat(dev
, &stb
) < 0)
4073 if ((S_IFMT
& stb
.st_mode
) != S_IFBLK
)
4075 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
4076 if (dl
->major
== (int)major(stb
.st_rdev
) &&
4077 dl
->minor
== (int)minor(stb
.st_rdev
))
4082 fprintf(stderr
, Name
": %s is not in the "
4083 "same imsm set\n", dev
);
4085 } else if (super
->orom
&& dl
->index
< 0 && mpb
->num_raid_devs
) {
4086 /* If a volume is present then the current creation attempt
4087 * cannot incorporate new spares because the orom may not
4088 * understand this configuration (all member disks must be
4089 * members of each array in the container).
4091 fprintf(stderr
, Name
": %s is a spare and a volume"
4092 " is already defined for this container\n", dev
);
4093 fprintf(stderr
, Name
": The option-rom requires all member"
4094 " disks to be a member of all volumes\n");
4098 /* retrieve the largest free space block */
4099 e
= get_extents(super
, dl
);
4104 unsigned long long esize
;
4106 esize
= e
[i
].start
- pos
;
4107 if (esize
>= maxsize
)
4109 pos
= e
[i
].start
+ e
[i
].size
;
4111 } while (e
[i
-1].size
);
4116 fprintf(stderr
, Name
": unable to determine free space for: %s\n",
4120 if (maxsize
< size
) {
4122 fprintf(stderr
, Name
": %s not enough space (%llu < %llu)\n",
4123 dev
, maxsize
, size
);
4127 /* count total number of extents for merge */
4129 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
4131 i
+= dl
->extent_cnt
;
4133 maxsize
= merge_extents(super
, i
);
4134 if (maxsize
< size
|| maxsize
== 0) {
4136 fprintf(stderr
, Name
": not enough space after merge (%llu < %llu)\n",
4141 *freesize
= maxsize
;
4146 static int reserve_space(struct supertype
*st
, int raiddisks
,
4147 unsigned long long size
, int chunk
,
4148 unsigned long long *freesize
)
4150 struct intel_super
*super
= st
->sb
;
4151 struct imsm_super
*mpb
= super
->anchor
;
4156 unsigned long long maxsize
;
4157 unsigned long long minsize
;
4161 /* find the largest common start free region of the possible disks */
4165 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
4171 /* don't activate new spares if we are orom constrained
4172 * and there is already a volume active in the container
4174 if (super
->orom
&& dl
->index
< 0 && mpb
->num_raid_devs
)
4177 e
= get_extents(super
, dl
);
4180 for (i
= 1; e
[i
-1].size
; i
++)
4188 maxsize
= merge_extents(super
, extent_cnt
);
4193 if (cnt
< raiddisks
||
4194 (super
->orom
&& used
&& used
!= raiddisks
) ||
4195 maxsize
< minsize
||
4197 fprintf(stderr
, Name
": not enough devices with space to create array.\n");
4198 return 0; /* No enough free spaces large enough */
4210 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
4212 dl
->raiddisk
= cnt
++;
4219 static int validate_geometry_imsm(struct supertype
*st
, int level
, int layout
,
4220 int raiddisks
, int chunk
, unsigned long long size
,
4221 char *dev
, unsigned long long *freesize
,
4228 /* if given unused devices create a container
4229 * if given given devices in a container create a member volume
4231 if (level
== LEVEL_CONTAINER
) {
4232 /* Must be a fresh device to add to a container */
4233 return validate_geometry_imsm_container(st
, level
, layout
,
4234 raiddisks
, chunk
, size
,
4240 if (st
->sb
&& freesize
) {
4241 /* we are being asked to automatically layout a
4242 * new volume based on the current contents of
4243 * the container. If the the parameters can be
4244 * satisfied reserve_space will record the disks,
4245 * start offset, and size of the volume to be
4246 * created. add_to_super and getinfo_super
4247 * detect when autolayout is in progress.
4249 if (!validate_geometry_imsm_orom(st
->sb
, level
, layout
,
4253 return reserve_space(st
, raiddisks
, size
, chunk
, freesize
);
4258 /* creating in a given container */
4259 return validate_geometry_imsm_volume(st
, level
, layout
,
4260 raiddisks
, chunk
, size
,
4261 dev
, freesize
, verbose
);
4264 /* This device needs to be a device in an 'imsm' container */
4265 fd
= open(dev
, O_RDONLY
|O_EXCL
, 0);
4269 Name
": Cannot create this array on device %s\n",
4274 if (errno
!= EBUSY
|| (fd
= open(dev
, O_RDONLY
, 0)) < 0) {
4276 fprintf(stderr
, Name
": Cannot open %s: %s\n",
4277 dev
, strerror(errno
));
4280 /* Well, it is in use by someone, maybe an 'imsm' container. */
4281 cfd
= open_container(fd
);
4285 fprintf(stderr
, Name
": Cannot use %s: It is busy\n",
4289 sra
= sysfs_read(cfd
, 0, GET_VERSION
);
4290 if (sra
&& sra
->array
.major_version
== -1 &&
4291 strcmp(sra
->text_version
, "imsm") == 0)
4295 /* This is a member of a imsm container. Load the container
4296 * and try to create a volume
4298 struct intel_super
*super
;
4300 if (load_super_imsm_all(st
, cfd
, (void **) &super
, NULL
) == 0) {
4302 st
->container_dev
= fd2devnum(cfd
);
4304 return validate_geometry_imsm_volume(st
, level
, layout
,
4312 fprintf(stderr
, Name
": failed container membership check\n");
4318 static void default_geometry_imsm(struct supertype
*st
, int *level
, int *layout
, int *chunk
)
4320 struct intel_super
*super
= st
->sb
;
4322 if (level
&& *level
== UnSet
)
4323 *level
= LEVEL_CONTAINER
;
4325 if (level
&& layout
&& *layout
== UnSet
)
4326 *layout
= imsm_level_to_layout(*level
);
4328 if (chunk
&& (*chunk
== UnSet
|| *chunk
== 0) &&
4329 super
&& super
->orom
)
4330 *chunk
= imsm_orom_default_chunk(super
->orom
);
4333 static void handle_missing(struct intel_super
*super
, struct imsm_dev
*dev
);
4335 static int kill_subarray_imsm(struct supertype
*st
)
4337 /* remove the subarray currently referenced by ->current_vol */
4339 struct intel_dev
**dp
;
4340 struct intel_super
*super
= st
->sb
;
4341 __u8 current_vol
= super
->current_vol
;
4342 struct imsm_super
*mpb
= super
->anchor
;
4344 if (super
->current_vol
< 0)
4346 super
->current_vol
= -1; /* invalidate subarray cursor */
4348 /* block deletions that would change the uuid of active subarrays
4350 * FIXME when immutable ids are available, but note that we'll
4351 * also need to fixup the invalidated/active subarray indexes in
4354 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
4357 if (i
< current_vol
)
4359 sprintf(subarray
, "%u", i
);
4360 if (is_subarray_active(subarray
, st
->devname
)) {
4362 Name
": deleting subarray-%d would change the UUID of active subarray-%d, aborting\n",
4369 if (st
->update_tail
) {
4370 struct imsm_update_kill_array
*u
= malloc(sizeof(*u
));
4374 u
->type
= update_kill_array
;
4375 u
->dev_idx
= current_vol
;
4376 append_metadata_update(st
, u
, sizeof(*u
));
4381 for (dp
= &super
->devlist
; *dp
;)
4382 if ((*dp
)->index
== current_vol
) {
4385 handle_missing(super
, (*dp
)->dev
);
4386 if ((*dp
)->index
> current_vol
)
4391 /* no more raid devices, all active components are now spares,
4392 * but of course failed are still failed
4394 if (--mpb
->num_raid_devs
== 0) {
4397 for (d
= super
->disks
; d
; d
= d
->next
)
4398 if (d
->index
> -2) {
4400 d
->disk
.status
= SPARE_DISK
;
4404 super
->updates_pending
++;
4409 static int update_subarray_imsm(struct supertype
*st
, char *subarray
,
4410 char *update
, struct mddev_ident
*ident
)
4412 /* update the subarray currently referenced by ->current_vol */
4413 struct intel_super
*super
= st
->sb
;
4414 struct imsm_super
*mpb
= super
->anchor
;
4416 if (strcmp(update
, "name") == 0) {
4417 char *name
= ident
->name
;
4421 if (is_subarray_active(subarray
, st
->devname
)) {
4423 Name
": Unable to update name of active subarray\n");
4427 if (!check_name(super
, name
, 0))
4430 vol
= strtoul(subarray
, &ep
, 10);
4431 if (*ep
!= '\0' || vol
>= super
->anchor
->num_raid_devs
)
4434 if (st
->update_tail
) {
4435 struct imsm_update_rename_array
*u
= malloc(sizeof(*u
));
4439 u
->type
= update_rename_array
;
4441 snprintf((char *) u
->name
, MAX_RAID_SERIAL_LEN
, "%s", name
);
4442 append_metadata_update(st
, u
, sizeof(*u
));
4444 struct imsm_dev
*dev
;
4447 dev
= get_imsm_dev(super
, vol
);
4448 snprintf((char *) dev
->volume
, MAX_RAID_SERIAL_LEN
, "%s", name
);
4449 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
4450 dev
= get_imsm_dev(super
, i
);
4451 handle_missing(super
, dev
);
4453 super
->updates_pending
++;
4460 #endif /* MDASSEMBLE */
4462 static int is_gen_migration(struct imsm_dev
*dev
)
4464 if (!dev
->vol
.migr_state
)
4467 if (migr_type(dev
) == MIGR_GEN_MIGR
)
4473 static int is_rebuilding(struct imsm_dev
*dev
)
4475 struct imsm_map
*migr_map
;
4477 if (!dev
->vol
.migr_state
)
4480 if (migr_type(dev
) != MIGR_REBUILD
)
4483 migr_map
= get_imsm_map(dev
, 1);
4485 if (migr_map
->map_state
== IMSM_T_STATE_DEGRADED
)
4491 static void update_recovery_start(struct imsm_dev
*dev
, struct mdinfo
*array
)
4493 struct mdinfo
*rebuild
= NULL
;
4497 if (!is_rebuilding(dev
))
4500 /* Find the rebuild target, but punt on the dual rebuild case */
4501 for (d
= array
->devs
; d
; d
= d
->next
)
4502 if (d
->recovery_start
== 0) {
4509 /* (?) none of the disks are marked with
4510 * IMSM_ORD_REBUILD, so assume they are missing and the
4511 * disk_ord_tbl was not correctly updated
4513 dprintf("%s: failed to locate out-of-sync disk\n", __func__
);
4517 units
= __le32_to_cpu(dev
->vol
.curr_migr_unit
);
4518 rebuild
->recovery_start
= units
* blocks_per_migr_unit(dev
);
4522 static struct mdinfo
*container_content_imsm(struct supertype
*st
, char *subarray
)
4524 /* Given a container loaded by load_super_imsm_all,
4525 * extract information about all the arrays into
4527 * If 'subarray' is given, just extract info about that array.
4529 * For each imsm_dev create an mdinfo, fill it in,
4530 * then look for matching devices in super->disks
4531 * and create appropriate device mdinfo.
4533 struct intel_super
*super
= st
->sb
;
4534 struct imsm_super
*mpb
= super
->anchor
;
4535 struct mdinfo
*rest
= NULL
;
4539 /* check for bad blocks */
4540 if (imsm_bbm_log_size(super
->anchor
))
4543 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
4544 struct imsm_dev
*dev
;
4545 struct imsm_map
*map
;
4546 struct imsm_map
*map2
;
4547 struct mdinfo
*this;
4552 (i
!= strtoul(subarray
, &ep
, 10) || *ep
!= '\0'))
4555 dev
= get_imsm_dev(super
, i
);
4556 map
= get_imsm_map(dev
, 0);
4557 map2
= get_imsm_map(dev
, 1);
4559 /* do not publish arrays that are in the middle of an
4560 * unsupported migration
4562 if (dev
->vol
.migr_state
&&
4563 (migr_type(dev
) == MIGR_STATE_CHANGE
)) {
4564 fprintf(stderr
, Name
": cannot assemble volume '%.16s':"
4565 " unsupported migration in progress\n",
4570 this = malloc(sizeof(*this));
4572 fprintf(stderr
, Name
": failed to allocate %zu bytes\n",
4576 memset(this, 0, sizeof(*this));
4579 super
->current_vol
= i
;
4580 getinfo_super_imsm_volume(st
, this, NULL
);
4581 for (slot
= 0 ; slot
< map
->num_members
; slot
++) {
4582 unsigned long long recovery_start
;
4583 struct mdinfo
*info_d
;
4590 idx
= get_imsm_disk_idx(dev
, slot
, 0);
4591 ord
= get_imsm_ord_tbl_ent(dev
, slot
, 0);
4592 for (d
= super
->disks
; d
; d
= d
->next
)
4593 if (d
->index
== idx
)
4596 recovery_start
= MaxSector
;
4599 if (d
&& is_failed(&d
->disk
))
4601 if (ord
& IMSM_ORD_REBUILD
)
4605 * if we skip some disks the array will be assmebled degraded;
4606 * reset resync start to avoid a dirty-degraded
4607 * situation when performing the intial sync
4609 * FIXME handle dirty degraded
4611 if ((skip
|| recovery_start
== 0) && !dev
->vol
.dirty
)
4612 this->resync_start
= MaxSector
;
4616 info_d
= calloc(1, sizeof(*info_d
));
4618 fprintf(stderr
, Name
": failed to allocate disk"
4619 " for volume %.16s\n", dev
->volume
);
4620 info_d
= this->devs
;
4622 struct mdinfo
*d
= info_d
->next
;
4631 info_d
->next
= this->devs
;
4632 this->devs
= info_d
;
4634 info_d
->disk
.number
= d
->index
;
4635 info_d
->disk
.major
= d
->major
;
4636 info_d
->disk
.minor
= d
->minor
;
4637 info_d
->disk
.raid_disk
= slot
;
4638 info_d
->recovery_start
= recovery_start
;
4640 if (slot
< map2
->num_members
)
4641 info_d
->disk
.state
= (1 << MD_DISK_ACTIVE
);
4643 this->array
.spare_disks
++;
4645 if (slot
< map
->num_members
)
4646 info_d
->disk
.state
= (1 << MD_DISK_ACTIVE
);
4648 this->array
.spare_disks
++;
4650 if (info_d
->recovery_start
== MaxSector
)
4651 this->array
.working_disks
++;
4653 info_d
->events
= __le32_to_cpu(mpb
->generation_num
);
4654 info_d
->data_offset
= __le32_to_cpu(map
->pba_of_lba0
);
4655 info_d
->component_size
= __le32_to_cpu(map
->blocks_per_member
);
4657 /* now that the disk list is up-to-date fixup recovery_start */
4658 update_recovery_start(dev
, this);
4662 /* if array has bad blocks, set suitable bit in array status */
4664 rest
->array
.state
|= (1<<MD_SB_BBM_ERRORS
);
4670 static __u8
imsm_check_degraded(struct intel_super
*super
, struct imsm_dev
*dev
, int failed
)
4672 struct imsm_map
*map
= get_imsm_map(dev
, 0);
4675 return map
->map_state
== IMSM_T_STATE_UNINITIALIZED
?
4676 IMSM_T_STATE_UNINITIALIZED
: IMSM_T_STATE_NORMAL
;
4678 switch (get_imsm_raid_level(map
)) {
4680 return IMSM_T_STATE_FAILED
;
4683 if (failed
< map
->num_members
)
4684 return IMSM_T_STATE_DEGRADED
;
4686 return IMSM_T_STATE_FAILED
;
4691 * check to see if any mirrors have failed, otherwise we
4692 * are degraded. Even numbered slots are mirrored on
4696 /* gcc -Os complains that this is unused */
4697 int insync
= insync
;
4699 for (i
= 0; i
< map
->num_members
; i
++) {
4700 __u32 ord
= get_imsm_ord_tbl_ent(dev
, i
, -1);
4701 int idx
= ord_to_idx(ord
);
4702 struct imsm_disk
*disk
;
4704 /* reset the potential in-sync count on even-numbered
4705 * slots. num_copies is always 2 for imsm raid10
4710 disk
= get_imsm_disk(super
, idx
);
4711 if (!disk
|| is_failed(disk
) || ord
& IMSM_ORD_REBUILD
)
4714 /* no in-sync disks left in this mirror the
4718 return IMSM_T_STATE_FAILED
;
4721 return IMSM_T_STATE_DEGRADED
;
4725 return IMSM_T_STATE_DEGRADED
;
4727 return IMSM_T_STATE_FAILED
;
4733 return map
->map_state
;
4736 static int imsm_count_failed(struct intel_super
*super
, struct imsm_dev
*dev
)
4740 struct imsm_disk
*disk
;
4741 struct imsm_map
*map
= get_imsm_map(dev
, 0);
4742 struct imsm_map
*prev
= get_imsm_map(dev
, dev
->vol
.migr_state
);
4746 /* at the beginning of migration we set IMSM_ORD_REBUILD on
4747 * disks that are being rebuilt. New failures are recorded to
4748 * map[0]. So we look through all the disks we started with and
4749 * see if any failures are still present, or if any new ones
4752 * FIXME add support for online capacity expansion and
4753 * raid-level-migration
4755 for (i
= 0; i
< prev
->num_members
; i
++) {
4756 ord
= __le32_to_cpu(prev
->disk_ord_tbl
[i
]);
4757 ord
|= __le32_to_cpu(map
->disk_ord_tbl
[i
]);
4758 idx
= ord_to_idx(ord
);
4760 disk
= get_imsm_disk(super
, idx
);
4761 if (!disk
|| is_failed(disk
) || ord
& IMSM_ORD_REBUILD
)
4769 static int imsm_open_new(struct supertype
*c
, struct active_array
*a
,
4772 struct intel_super
*super
= c
->sb
;
4773 struct imsm_super
*mpb
= super
->anchor
;
4775 if (atoi(inst
) >= mpb
->num_raid_devs
) {
4776 fprintf(stderr
, "%s: subarry index %d, out of range\n",
4777 __func__
, atoi(inst
));
4781 dprintf("imsm: open_new %s\n", inst
);
4782 a
->info
.container_member
= atoi(inst
);
4786 static int is_resyncing(struct imsm_dev
*dev
)
4788 struct imsm_map
*migr_map
;
4790 if (!dev
->vol
.migr_state
)
4793 if (migr_type(dev
) == MIGR_INIT
||
4794 migr_type(dev
) == MIGR_REPAIR
)
4797 if (migr_type(dev
) == MIGR_GEN_MIGR
)
4800 migr_map
= get_imsm_map(dev
, 1);
4802 if ((migr_map
->map_state
== IMSM_T_STATE_NORMAL
) &&
4803 (dev
->vol
.migr_type
!= MIGR_GEN_MIGR
))
4809 /* return true if we recorded new information */
4810 static int mark_failure(struct imsm_dev
*dev
, struct imsm_disk
*disk
, int idx
)
4814 struct imsm_map
*map
;
4816 /* new failures are always set in map[0] */
4817 map
= get_imsm_map(dev
, 0);
4819 slot
= get_imsm_disk_slot(map
, idx
);
4823 ord
= __le32_to_cpu(map
->disk_ord_tbl
[slot
]);
4824 if (is_failed(disk
) && (ord
& IMSM_ORD_REBUILD
))
4827 disk
->status
|= FAILED_DISK
;
4828 disk
->status
&= ~CONFIGURED_DISK
;
4829 set_imsm_ord_tbl_ent(map
, slot
, idx
| IMSM_ORD_REBUILD
);
4830 if (map
->failed_disk_num
== 0xff)
4831 map
->failed_disk_num
= slot
;
4835 static void mark_missing(struct imsm_dev
*dev
, struct imsm_disk
*disk
, int idx
)
4837 mark_failure(dev
, disk
, idx
);
4839 if (disk
->scsi_id
== __cpu_to_le32(~(__u32
)0))
4842 disk
->scsi_id
= __cpu_to_le32(~(__u32
)0);
4843 memmove(&disk
->serial
[0], &disk
->serial
[1], MAX_RAID_SERIAL_LEN
- 1);
4846 static void handle_missing(struct intel_super
*super
, struct imsm_dev
*dev
)
4852 if (!super
->missing
)
4854 failed
= imsm_count_failed(super
, dev
);
4855 map_state
= imsm_check_degraded(super
, dev
, failed
);
4857 dprintf("imsm: mark missing\n");
4858 end_migration(dev
, map_state
);
4859 for (dl
= super
->missing
; dl
; dl
= dl
->next
)
4860 mark_missing(dev
, &dl
->disk
, dl
->index
);
4861 super
->updates_pending
++;
4864 static void imsm_set_disk(struct active_array
*a
, int n
, int state
);
4866 static void imsm_progress_container_reshape(struct intel_super
*super
)
4868 /* if no device has a migr_state, but some device has a
4869 * different number of members than the previous device, start
4870 * changing the number of devices in this device to match
4873 struct imsm_super
*mpb
= super
->anchor
;
4874 int prev_disks
= -1;
4877 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
4878 struct imsm_dev
*dev
= get_imsm_dev(super
, i
);
4879 struct imsm_map
*map
= get_imsm_map(dev
, 0);
4880 struct imsm_map
*map2
;
4881 int prev_num_members
;
4884 if (dev
->vol
.migr_state
)
4887 if (prev_disks
== -1)
4888 prev_disks
= map
->num_members
;
4889 if (prev_disks
== map
->num_members
)
4892 /* OK, this array needs to enter reshape mode.
4893 * i.e it needs a migr_state
4896 prev_num_members
= map
->num_members
;
4897 map
->num_members
= prev_disks
;
4898 dev
->vol
.migr_state
= 1;
4899 dev
->vol
.curr_migr_unit
= 0;
4900 dev
->vol
.migr_type
= MIGR_GEN_MIGR
;
4901 for (i
= prev_num_members
;
4902 i
< map
->num_members
; i
++)
4903 set_imsm_ord_tbl_ent(map
, i
, i
);
4904 map2
= get_imsm_map(dev
, 1);
4905 /* Copy the current map */
4906 memcpy(map2
, map
, sizeof_imsm_map(map
));
4907 map2
->num_members
= prev_num_members
;
4909 /* calculate new size
4911 used_disks
= imsm_num_data_members(dev
, 0);
4913 unsigned long long array_blocks
;
4916 map
->blocks_per_member
4918 /* round array size down to closest MB
4920 array_blocks
= (array_blocks
4921 >> SECT_PER_MB_SHIFT
)
4922 << SECT_PER_MB_SHIFT
;
4924 __cpu_to_le32((__u32
)array_blocks
);
4927 (__u32
)(array_blocks
>> 32));
4929 super
->updates_pending
++;
4933 /* Handle dirty -> clean transititions, resync and reshape. Degraded and rebuild
4934 * states are handled in imsm_set_disk() with one exception, when a
4935 * resync is stopped due to a new failure this routine will set the
4936 * 'degraded' state for the array.
4938 static int imsm_set_array_state(struct active_array
*a
, int consistent
)
4940 int inst
= a
->info
.container_member
;
4941 struct intel_super
*super
= a
->container
->sb
;
4942 struct imsm_dev
*dev
= get_imsm_dev(super
, inst
);
4943 struct imsm_map
*map
= get_imsm_map(dev
, 0);
4944 int failed
= imsm_count_failed(super
, dev
);
4945 __u8 map_state
= imsm_check_degraded(super
, dev
, failed
);
4946 __u32 blocks_per_unit
;
4948 if (dev
->vol
.migr_state
&&
4949 dev
->vol
.migr_type
== MIGR_GEN_MIGR
) {
4950 /* array state change is blocked due to reshape action
4952 * - abort the reshape (if last_checkpoint is 0 and action!= reshape)
4953 * - finish the reshape (if last_checkpoint is big and action != reshape)
4954 * - update curr_migr_unit
4956 if (a
->curr_action
== reshape
) {
4957 /* still reshaping, maybe update curr_migr_unit */
4958 long long blocks_per_unit
= blocks_per_migr_unit(dev
);
4959 long long unit
= a
->last_checkpoint
;
4960 if (blocks_per_unit
) {
4961 unit
/= blocks_per_unit
;
4963 __le32_to_cpu(dev
->vol
.curr_migr_unit
)) {
4964 dev
->vol
.curr_migr_unit
=
4965 __cpu_to_le32(unit
);
4966 super
->updates_pending
++;
4971 if (a
->last_checkpoint
== 0 && a
->prev_action
== reshape
) {
4972 /* for some reason we aborted the reshape.
4975 struct imsm_map
*map2
= get_imsm_map(dev
, 1);
4976 dev
->vol
.migr_state
= 0;
4977 dev
->vol
.migr_type
= 0;
4978 dev
->vol
.curr_migr_unit
= 0;
4979 memcpy(map
, map2
, sizeof_imsm_map(map2
));
4980 super
->updates_pending
++;
4982 if (a
->last_checkpoint
>= a
->info
.component_size
) {
4983 unsigned long long array_blocks
;
4985 /* it seems the reshape is all done */
4986 dev
->vol
.migr_state
= 0;
4987 dev
->vol
.migr_type
= 0;
4988 dev
->vol
.curr_migr_unit
= 0;
4990 used_disks
= imsm_num_data_members(dev
, -1);
4991 array_blocks
= map
->blocks_per_member
* used_disks
;
4992 /* round array size down to closest MB */
4993 array_blocks
= (array_blocks
>> SECT_PER_MB_SHIFT
)
4994 << SECT_PER_MB_SHIFT
;
4995 dev
->size_low
= __cpu_to_le32((__u32
) array_blocks
);
4996 dev
->size_high
= __cpu_to_le32((__u32
) (array_blocks
>> 32));
4997 a
->info
.custom_array_size
= array_blocks
;
4998 a
->check_reshape
= 1; /* encourage manager to update
5001 super
->updates_pending
++;
5002 imsm_progress_container_reshape(super
);
5007 /* before we activate this array handle any missing disks */
5008 if (consistent
== 2)
5009 handle_missing(super
, dev
);
5011 if (consistent
== 2 &&
5012 (!is_resync_complete(&a
->info
) ||
5013 map_state
!= IMSM_T_STATE_NORMAL
||
5014 dev
->vol
.migr_state
))
5017 if (is_resync_complete(&a
->info
)) {
5018 /* complete intialization / resync,
5019 * recovery and interrupted recovery is completed in
5022 if (is_resyncing(dev
)) {
5023 dprintf("imsm: mark resync done\n");
5024 end_migration(dev
, map_state
);
5025 super
->updates_pending
++;
5026 a
->last_checkpoint
= 0;
5028 } else if (!is_resyncing(dev
) && !failed
) {
5029 /* mark the start of the init process if nothing is failed */
5030 dprintf("imsm: mark resync start\n");
5031 if (map
->map_state
== IMSM_T_STATE_UNINITIALIZED
)
5032 migrate(dev
, IMSM_T_STATE_NORMAL
, MIGR_INIT
);
5034 migrate(dev
, IMSM_T_STATE_NORMAL
, MIGR_REPAIR
);
5035 super
->updates_pending
++;
5038 /* check if we can update curr_migr_unit from resync_start, recovery_start */
5039 blocks_per_unit
= blocks_per_migr_unit(dev
);
5040 if (blocks_per_unit
) {
5044 units
= a
->last_checkpoint
/ blocks_per_unit
;
5047 /* check that we did not overflow 32-bits, and that
5048 * curr_migr_unit needs updating
5050 if (units32
== units
&&
5051 __le32_to_cpu(dev
->vol
.curr_migr_unit
) != units32
) {
5052 dprintf("imsm: mark checkpoint (%u)\n", units32
);
5053 dev
->vol
.curr_migr_unit
= __cpu_to_le32(units32
);
5054 super
->updates_pending
++;
5058 /* mark dirty / clean */
5059 if (dev
->vol
.dirty
!= !consistent
) {
5060 dprintf("imsm: mark '%s'\n", consistent
? "clean" : "dirty");
5065 super
->updates_pending
++;
5068 /* finalize online capacity expansion/reshape */
5069 if ((a
->curr_action
!= reshape
) &&
5070 (a
->prev_action
== reshape
)) {
5073 for (mdi
= a
->info
.devs
; mdi
; mdi
= mdi
->next
)
5074 imsm_set_disk(a
, mdi
->disk
.raid_disk
, mdi
->curr_state
);
5080 static void imsm_set_disk(struct active_array
*a
, int n
, int state
)
5082 int inst
= a
->info
.container_member
;
5083 struct intel_super
*super
= a
->container
->sb
;
5084 struct imsm_dev
*dev
= get_imsm_dev(super
, inst
);
5085 struct imsm_map
*map
= get_imsm_map(dev
, 0);
5086 struct imsm_disk
*disk
;
5091 if (n
> map
->num_members
)
5092 fprintf(stderr
, "imsm: set_disk %d out of range 0..%d\n",
5093 n
, map
->num_members
- 1);
5098 dprintf("imsm: set_disk %d:%x\n", n
, state
);
5100 ord
= get_imsm_ord_tbl_ent(dev
, n
, -1);
5101 disk
= get_imsm_disk(super
, ord_to_idx(ord
));
5103 /* check for new failures */
5104 if (state
& DS_FAULTY
) {
5105 if (mark_failure(dev
, disk
, ord_to_idx(ord
)))
5106 super
->updates_pending
++;
5109 /* check if in_sync */
5110 if (state
& DS_INSYNC
&& ord
& IMSM_ORD_REBUILD
&& is_rebuilding(dev
)) {
5111 struct imsm_map
*migr_map
= get_imsm_map(dev
, 1);
5113 set_imsm_ord_tbl_ent(migr_map
, n
, ord_to_idx(ord
));
5114 super
->updates_pending
++;
5117 failed
= imsm_count_failed(super
, dev
);
5118 map_state
= imsm_check_degraded(super
, dev
, failed
);
5120 /* check if recovery complete, newly degraded, or failed */
5121 if (map_state
== IMSM_T_STATE_NORMAL
&& is_rebuilding(dev
)) {
5122 end_migration(dev
, map_state
);
5123 map
= get_imsm_map(dev
, 0);
5124 map
->failed_disk_num
= ~0;
5125 super
->updates_pending
++;
5126 a
->last_checkpoint
= 0;
5127 } else if (map_state
== IMSM_T_STATE_DEGRADED
&&
5128 map
->map_state
!= map_state
&&
5129 !dev
->vol
.migr_state
) {
5130 dprintf("imsm: mark degraded\n");
5131 map
->map_state
= map_state
;
5132 super
->updates_pending
++;
5133 a
->last_checkpoint
= 0;
5134 } else if (map_state
== IMSM_T_STATE_FAILED
&&
5135 map
->map_state
!= map_state
) {
5136 dprintf("imsm: mark failed\n");
5137 end_migration(dev
, map_state
);
5138 super
->updates_pending
++;
5139 a
->last_checkpoint
= 0;
5140 } else if (is_gen_migration(dev
)) {
5141 dprintf("imsm: Detected General Migration in state: ");
5142 if (map_state
== IMSM_T_STATE_NORMAL
) {
5143 end_migration(dev
, map_state
);
5144 map
= get_imsm_map(dev
, 0);
5145 map
->failed_disk_num
= ~0;
5146 dprintf("normal\n");
5148 if (map_state
== IMSM_T_STATE_DEGRADED
) {
5149 printf("degraded\n");
5150 end_migration(dev
, map_state
);
5152 dprintf("failed\n");
5154 map
->map_state
= map_state
;
5156 super
->updates_pending
++;
5160 static int store_imsm_mpb(int fd
, struct imsm_super
*mpb
)
5163 __u32 mpb_size
= __le32_to_cpu(mpb
->mpb_size
);
5164 unsigned long long dsize
;
5165 unsigned long long sectors
;
5167 get_dev_size(fd
, NULL
, &dsize
);
5169 if (mpb_size
> 512) {
5170 /* -1 to account for anchor */
5171 sectors
= mpb_sectors(mpb
) - 1;
5173 /* write the extended mpb to the sectors preceeding the anchor */
5174 if (lseek64(fd
, dsize
- (512 * (2 + sectors
)), SEEK_SET
) < 0)
5177 if ((unsigned long long)write(fd
, buf
+ 512, 512 * sectors
)
5182 /* first block is stored on second to last sector of the disk */
5183 if (lseek64(fd
, dsize
- (512 * 2), SEEK_SET
) < 0)
5186 if (write(fd
, buf
, 512) != 512)
5192 static void imsm_sync_metadata(struct supertype
*container
)
5194 struct intel_super
*super
= container
->sb
;
5196 dprintf("sync metadata: %d\n", super
->updates_pending
);
5197 if (!super
->updates_pending
)
5200 write_super_imsm(container
, 0);
5202 super
->updates_pending
= 0;
5205 static struct dl
*imsm_readd(struct intel_super
*super
, int idx
, struct active_array
*a
)
5207 struct imsm_dev
*dev
= get_imsm_dev(super
, a
->info
.container_member
);
5208 int i
= get_imsm_disk_idx(dev
, idx
, -1);
5211 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
5215 if (dl
&& is_failed(&dl
->disk
))
5219 dprintf("%s: found %x:%x\n", __func__
, dl
->major
, dl
->minor
);
5224 static struct dl
*imsm_add_spare(struct intel_super
*super
, int slot
,
5225 struct active_array
*a
, int activate_new
,
5226 struct mdinfo
*additional_test_list
)
5228 struct imsm_dev
*dev
= get_imsm_dev(super
, a
->info
.container_member
);
5229 int idx
= get_imsm_disk_idx(dev
, slot
, -1);
5230 struct imsm_super
*mpb
= super
->anchor
;
5231 struct imsm_map
*map
;
5232 unsigned long long pos
;
5237 __u32 array_start
= 0;
5238 __u32 array_end
= 0;
5240 struct mdinfo
*test_list
;
5242 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
5243 /* If in this array, skip */
5244 for (d
= a
->info
.devs
; d
; d
= d
->next
)
5245 if (d
->state_fd
>= 0 &&
5246 d
->disk
.major
== dl
->major
&&
5247 d
->disk
.minor
== dl
->minor
) {
5248 dprintf("%x:%x already in array\n",
5249 dl
->major
, dl
->minor
);
5254 test_list
= additional_test_list
;
5256 if (test_list
->disk
.major
== dl
->major
&&
5257 test_list
->disk
.minor
== dl
->minor
) {
5258 dprintf("%x:%x already in additional test list\n",
5259 dl
->major
, dl
->minor
);
5262 test_list
= test_list
->next
;
5267 /* skip in use or failed drives */
5268 if (is_failed(&dl
->disk
) || idx
== dl
->index
||
5270 dprintf("%x:%x status (failed: %d index: %d)\n",
5271 dl
->major
, dl
->minor
, is_failed(&dl
->disk
), idx
);
5275 /* skip pure spares when we are looking for partially
5276 * assimilated drives
5278 if (dl
->index
== -1 && !activate_new
)
5281 /* Does this unused device have the requisite free space?
5282 * It needs to be able to cover all member volumes
5284 ex
= get_extents(super
, dl
);
5286 dprintf("cannot get extents\n");
5289 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
5290 dev
= get_imsm_dev(super
, i
);
5291 map
= get_imsm_map(dev
, 0);
5293 /* check if this disk is already a member of
5296 if (get_imsm_disk_slot(map
, dl
->index
) >= 0)
5302 array_start
= __le32_to_cpu(map
->pba_of_lba0
);
5303 array_end
= array_start
+
5304 __le32_to_cpu(map
->blocks_per_member
) - 1;
5307 /* check that we can start at pba_of_lba0 with
5308 * blocks_per_member of space
5310 if (array_start
>= pos
&& array_end
< ex
[j
].start
) {
5314 pos
= ex
[j
].start
+ ex
[j
].size
;
5316 } while (ex
[j
-1].size
);
5323 if (i
< mpb
->num_raid_devs
) {
5324 dprintf("%x:%x does not have %u to %u available\n",
5325 dl
->major
, dl
->minor
, array_start
, array_end
);
5336 static int imsm_rebuild_allowed(struct supertype
*cont
, int dev_idx
, int failed
)
5338 struct imsm_dev
*dev2
;
5339 struct imsm_map
*map
;
5345 dev2
= get_imsm_dev(cont
->sb
, dev_idx
);
5347 state
= imsm_check_degraded(cont
->sb
, dev2
, failed
);
5348 if (state
== IMSM_T_STATE_FAILED
) {
5349 map
= get_imsm_map(dev2
, 0);
5352 for (slot
= 0; slot
< map
->num_members
; slot
++) {
5354 * Check if failed disks are deleted from intel
5355 * disk list or are marked to be deleted
5357 idx
= get_imsm_disk_idx(dev2
, slot
, -1);
5358 idisk
= get_imsm_dl_disk(cont
->sb
, idx
);
5360 * Do not rebuild the array if failed disks
5361 * from failed sub-array are not removed from
5365 is_failed(&idisk
->disk
) &&
5366 (idisk
->action
!= DISK_REMOVE
))
5374 static struct mdinfo
*imsm_activate_spare(struct active_array
*a
,
5375 struct metadata_update
**updates
)
5378 * Find a device with unused free space and use it to replace a
5379 * failed/vacant region in an array. We replace failed regions one a
5380 * array at a time. The result is that a new spare disk will be added
5381 * to the first failed array and after the monitor has finished
5382 * propagating failures the remainder will be consumed.
5384 * FIXME add a capability for mdmon to request spares from another
5388 struct intel_super
*super
= a
->container
->sb
;
5389 int inst
= a
->info
.container_member
;
5390 struct imsm_dev
*dev
= get_imsm_dev(super
, inst
);
5391 struct imsm_map
*map
= get_imsm_map(dev
, 0);
5392 int failed
= a
->info
.array
.raid_disks
;
5393 struct mdinfo
*rv
= NULL
;
5396 struct metadata_update
*mu
;
5398 struct imsm_update_activate_spare
*u
;
5403 for (d
= a
->info
.devs
; d
; d
= d
->next
) {
5404 if ((d
->curr_state
& DS_FAULTY
) &&
5406 /* wait for Removal to happen */
5408 if (d
->state_fd
>= 0)
5412 dprintf("imsm: activate spare: inst=%d failed=%d (%d) level=%d\n",
5413 inst
, failed
, a
->info
.array
.raid_disks
, a
->info
.array
.level
);
5415 if (dev
->vol
.migr_state
&&
5416 dev
->vol
.migr_type
== MIGR_GEN_MIGR
)
5417 /* No repair during migration */
5420 if (a
->info
.array
.level
== 4)
5421 /* No repair for takeovered array
5422 * imsm doesn't support raid4
5426 if (imsm_check_degraded(super
, dev
, failed
) != IMSM_T_STATE_DEGRADED
)
5430 * If there are any failed disks check state of the other volume.
5431 * Block rebuild if the another one is failed until failed disks
5432 * are removed from container.
5435 dprintf("found failed disks in %s, check if there another"
5436 "failed sub-array.\n",
5438 /* check if states of the other volumes allow for rebuild */
5439 for (i
= 0; i
< super
->anchor
->num_raid_devs
; i
++) {
5441 allowed
= imsm_rebuild_allowed(a
->container
,
5449 /* For each slot, if it is not working, find a spare */
5450 for (i
= 0; i
< a
->info
.array
.raid_disks
; i
++) {
5451 for (d
= a
->info
.devs
; d
; d
= d
->next
)
5452 if (d
->disk
.raid_disk
== i
)
5454 dprintf("found %d: %p %x\n", i
, d
, d
?d
->curr_state
:0);
5455 if (d
&& (d
->state_fd
>= 0))
5459 * OK, this device needs recovery. Try to re-add the
5460 * previous occupant of this slot, if this fails see if
5461 * we can continue the assimilation of a spare that was
5462 * partially assimilated, finally try to activate a new
5465 dl
= imsm_readd(super
, i
, a
);
5467 dl
= imsm_add_spare(super
, i
, a
, 0, NULL
);
5469 dl
= imsm_add_spare(super
, i
, a
, 1, NULL
);
5473 /* found a usable disk with enough space */
5474 di
= malloc(sizeof(*di
));
5477 memset(di
, 0, sizeof(*di
));
5479 /* dl->index will be -1 in the case we are activating a
5480 * pristine spare. imsm_process_update() will create a
5481 * new index in this case. Once a disk is found to be
5482 * failed in all member arrays it is kicked from the
5485 di
->disk
.number
= dl
->index
;
5487 /* (ab)use di->devs to store a pointer to the device
5490 di
->devs
= (struct mdinfo
*) dl
;
5492 di
->disk
.raid_disk
= i
;
5493 di
->disk
.major
= dl
->major
;
5494 di
->disk
.minor
= dl
->minor
;
5496 di
->recovery_start
= 0;
5497 di
->data_offset
= __le32_to_cpu(map
->pba_of_lba0
);
5498 di
->component_size
= a
->info
.component_size
;
5499 di
->container_member
= inst
;
5500 super
->random
= random32();
5504 dprintf("%x:%x to be %d at %llu\n", dl
->major
, dl
->minor
,
5505 i
, di
->data_offset
);
5511 /* No spares found */
5513 /* Now 'rv' has a list of devices to return.
5514 * Create a metadata_update record to update the
5515 * disk_ord_tbl for the array
5517 mu
= malloc(sizeof(*mu
));
5519 mu
->buf
= malloc(sizeof(struct imsm_update_activate_spare
) * num_spares
);
5520 if (mu
->buf
== NULL
) {
5527 struct mdinfo
*n
= rv
->next
;
5536 mu
->space_list
= NULL
;
5537 mu
->len
= sizeof(struct imsm_update_activate_spare
) * num_spares
;
5538 mu
->next
= *updates
;
5539 u
= (struct imsm_update_activate_spare
*) mu
->buf
;
5541 for (di
= rv
; di
; di
= di
->next
) {
5542 u
->type
= update_activate_spare
;
5543 u
->dl
= (struct dl
*) di
->devs
;
5545 u
->slot
= di
->disk
.raid_disk
;
5556 static int disks_overlap(struct intel_super
*super
, int idx
, struct imsm_update_create_array
*u
)
5558 struct imsm_dev
*dev
= get_imsm_dev(super
, idx
);
5559 struct imsm_map
*map
= get_imsm_map(dev
, 0);
5560 struct imsm_map
*new_map
= get_imsm_map(&u
->dev
, 0);
5561 struct disk_info
*inf
= get_disk_info(u
);
5562 struct imsm_disk
*disk
;
5566 for (i
= 0; i
< map
->num_members
; i
++) {
5567 disk
= get_imsm_disk(super
, get_imsm_disk_idx(dev
, i
, -1));
5568 for (j
= 0; j
< new_map
->num_members
; j
++)
5569 if (serialcmp(disk
->serial
, inf
[j
].serial
) == 0)
5577 static struct dl
*get_disk_super(struct intel_super
*super
, int major
, int minor
)
5579 struct dl
*dl
= NULL
;
5580 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
5581 if ((dl
->major
== major
) && (dl
->minor
== minor
))
5586 static int remove_disk_super(struct intel_super
*super
, int major
, int minor
)
5588 struct dl
*prev
= NULL
;
5592 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
5593 if ((dl
->major
== major
) && (dl
->minor
== minor
)) {
5596 prev
->next
= dl
->next
;
5598 super
->disks
= dl
->next
;
5600 __free_imsm_disk(dl
);
5601 dprintf("%s: removed %x:%x\n",
5602 __func__
, major
, minor
);
5610 static void imsm_delete(struct intel_super
*super
, struct dl
**dlp
, unsigned index
);
5612 static int add_remove_disk_update(struct intel_super
*super
)
5614 int check_degraded
= 0;
5615 struct dl
*disk
= NULL
;
5616 /* add/remove some spares to/from the metadata/contrainer */
5617 while (super
->disk_mgmt_list
) {
5618 struct dl
*disk_cfg
;
5620 disk_cfg
= super
->disk_mgmt_list
;
5621 super
->disk_mgmt_list
= disk_cfg
->next
;
5622 disk_cfg
->next
= NULL
;
5624 if (disk_cfg
->action
== DISK_ADD
) {
5625 disk_cfg
->next
= super
->disks
;
5626 super
->disks
= disk_cfg
;
5628 dprintf("%s: added %x:%x\n",
5629 __func__
, disk_cfg
->major
,
5631 } else if (disk_cfg
->action
== DISK_REMOVE
) {
5632 dprintf("Disk remove action processed: %x.%x\n",
5633 disk_cfg
->major
, disk_cfg
->minor
);
5634 disk
= get_disk_super(super
,
5638 /* store action status */
5639 disk
->action
= DISK_REMOVE
;
5640 /* remove spare disks only */
5641 if (disk
->index
== -1) {
5642 remove_disk_super(super
,
5647 /* release allocate disk structure */
5648 __free_imsm_disk(disk_cfg
);
5651 return check_degraded
;
5654 static int apply_reshape_container_disks_update(struct imsm_update_reshape
*u
,
5655 struct intel_super
*super
,
5658 struct dl
*new_disk
;
5659 struct intel_dev
*id
;
5661 int delta_disks
= u
->new_raid_disks
- u
->old_raid_disks
;
5662 int disk_count
= u
->old_raid_disks
;
5663 void **tofree
= NULL
;
5664 int devices_to_reshape
= 1;
5665 struct imsm_super
*mpb
= super
->anchor
;
5668 dprintf("imsm: imsm_process_update() for update_reshape\n");
5670 /* enable spares to use in array */
5671 for (i
= 0; i
< delta_disks
; i
++) {
5672 new_disk
= get_disk_super(super
,
5673 major(u
->new_disks
[i
]),
5674 minor(u
->new_disks
[i
]));
5675 dprintf("imsm: imsm_process_update(): new disk "
5676 "for reshape is: %i:%i (%p, index = %i)\n",
5677 major(u
->new_disks
[i
]), minor(u
->new_disks
[i
]),
5678 new_disk
, new_disk
->index
);
5679 if ((new_disk
== NULL
) ||
5680 ((new_disk
->index
>= 0) &&
5681 (new_disk
->index
< u
->old_raid_disks
)))
5682 goto update_reshape_exit
;
5683 new_disk
->index
= disk_count
++;
5684 /* slot to fill in autolayout
5686 new_disk
->raiddisk
= new_disk
->index
;
5687 new_disk
->disk
.status
|=
5689 new_disk
->disk
.status
&= ~SPARE_DISK
;
5692 dprintf("imsm: process_update(): update_reshape: volume set"
5693 " mpb->num_raid_devs = %i\n", mpb
->num_raid_devs
);
5694 /* manage changes in volume
5696 for (id
= super
->devlist
; id
; id
= id
->next
) {
5697 void **sp
= *space_list
;
5698 struct imsm_dev
*newdev
;
5699 struct imsm_map
*newmap
, *oldmap
;
5705 /* Copy the dev, but not (all of) the map */
5706 memcpy(newdev
, id
->dev
, sizeof(*newdev
));
5707 oldmap
= get_imsm_map(id
->dev
, 0);
5708 newmap
= get_imsm_map(newdev
, 0);
5709 /* Copy the current map */
5710 memcpy(newmap
, oldmap
, sizeof_imsm_map(oldmap
));
5711 /* update one device only
5713 if (devices_to_reshape
) {
5716 dprintf("process_update(): modifying "
5717 "subdev: %i\n", id
->index
);
5718 devices_to_reshape
--;
5719 newdev
->vol
.migr_state
= 1;
5720 newdev
->vol
.curr_migr_unit
= 0;
5721 newdev
->vol
.migr_type
= MIGR_GEN_MIGR
;
5722 newmap
->num_members
= u
->new_raid_disks
;
5723 for (i
= 0; i
< delta_disks
; i
++) {
5724 set_imsm_ord_tbl_ent(newmap
,
5725 u
->old_raid_disks
+ i
,
5726 u
->old_raid_disks
+ i
);
5728 /* New map is correct, now need to save old map
5730 newmap
= get_imsm_map(newdev
, 1);
5731 memcpy(newmap
, oldmap
, sizeof_imsm_map(oldmap
));
5733 /* calculate new size
5735 used_disks
= imsm_num_data_members(newdev
, 0);
5737 unsigned long long array_blocks
;
5740 newmap
->blocks_per_member
* used_disks
;
5741 /* round array size down to closest MB
5743 array_blocks
= (array_blocks
5744 >> SECT_PER_MB_SHIFT
)
5745 << SECT_PER_MB_SHIFT
;
5747 __cpu_to_le32((__u32
)array_blocks
);
5749 __cpu_to_le32((__u32
)(array_blocks
>> 32));
5753 sp
= (void **)id
->dev
;
5759 *space_list
= tofree
;
5762 update_reshape_exit
:
5767 static void imsm_process_update(struct supertype
*st
,
5768 struct metadata_update
*update
)
5771 * crack open the metadata_update envelope to find the update record
5772 * update can be one of:
5773 * update_reshape_container_disks - all the arrays in the container
5774 * are being reshaped to have more devices. We need to mark
5775 * the arrays for general migration and convert selected spares
5776 * into active devices.
5777 * update_activate_spare - a spare device has replaced a failed
5778 * device in an array, update the disk_ord_tbl. If this disk is
5779 * present in all member arrays then also clear the SPARE_DISK
5781 * update_create_array
5783 * update_rename_array
5784 * update_add_remove_disk
5786 struct intel_super
*super
= st
->sb
;
5787 struct imsm_super
*mpb
;
5788 enum imsm_update_type type
= *(enum imsm_update_type
*) update
->buf
;
5790 /* update requires a larger buf but the allocation failed */
5791 if (super
->next_len
&& !super
->next_buf
) {
5792 super
->next_len
= 0;
5796 if (super
->next_buf
) {
5797 memcpy(super
->next_buf
, super
->buf
, super
->len
);
5799 super
->len
= super
->next_len
;
5800 super
->buf
= super
->next_buf
;
5802 super
->next_len
= 0;
5803 super
->next_buf
= NULL
;
5806 mpb
= super
->anchor
;
5809 case update_reshape_container_disks
: {
5810 struct imsm_update_reshape
*u
= (void *)update
->buf
;
5811 if (apply_reshape_container_disks_update(
5812 u
, super
, &update
->space_list
))
5813 super
->updates_pending
++;
5816 case update_activate_spare
: {
5817 struct imsm_update_activate_spare
*u
= (void *) update
->buf
;
5818 struct imsm_dev
*dev
= get_imsm_dev(super
, u
->array
);
5819 struct imsm_map
*map
= get_imsm_map(dev
, 0);
5820 struct imsm_map
*migr_map
;
5821 struct active_array
*a
;
5822 struct imsm_disk
*disk
;
5827 int victim
= get_imsm_disk_idx(dev
, u
->slot
, -1);
5830 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
5835 fprintf(stderr
, "error: imsm_activate_spare passed "
5836 "an unknown disk (index: %d)\n",
5841 super
->updates_pending
++;
5843 /* count failures (excluding rebuilds and the victim)
5844 * to determine map[0] state
5847 for (i
= 0; i
< map
->num_members
; i
++) {
5850 disk
= get_imsm_disk(super
,
5851 get_imsm_disk_idx(dev
, i
, -1));
5852 if (!disk
|| is_failed(disk
))
5856 /* adding a pristine spare, assign a new index */
5857 if (dl
->index
< 0) {
5858 dl
->index
= super
->anchor
->num_disks
;
5859 super
->anchor
->num_disks
++;
5862 disk
->status
|= CONFIGURED_DISK
;
5863 disk
->status
&= ~SPARE_DISK
;
5866 to_state
= imsm_check_degraded(super
, dev
, failed
);
5867 map
->map_state
= IMSM_T_STATE_DEGRADED
;
5868 migrate(dev
, to_state
, MIGR_REBUILD
);
5869 migr_map
= get_imsm_map(dev
, 1);
5870 set_imsm_ord_tbl_ent(map
, u
->slot
, dl
->index
);
5871 set_imsm_ord_tbl_ent(migr_map
, u
->slot
, dl
->index
| IMSM_ORD_REBUILD
);
5873 /* update the family_num to mark a new container
5874 * generation, being careful to record the existing
5875 * family_num in orig_family_num to clean up after
5876 * earlier mdadm versions that neglected to set it.
5878 if (mpb
->orig_family_num
== 0)
5879 mpb
->orig_family_num
= mpb
->family_num
;
5880 mpb
->family_num
+= super
->random
;
5882 /* count arrays using the victim in the metadata */
5884 for (a
= st
->arrays
; a
; a
= a
->next
) {
5885 dev
= get_imsm_dev(super
, a
->info
.container_member
);
5886 map
= get_imsm_map(dev
, 0);
5888 if (get_imsm_disk_slot(map
, victim
) >= 0)
5892 /* delete the victim if it is no longer being
5898 /* We know that 'manager' isn't touching anything,
5899 * so it is safe to delete
5901 for (dlp
= &super
->disks
; *dlp
; dlp
= &(*dlp
)->next
)
5902 if ((*dlp
)->index
== victim
)
5905 /* victim may be on the missing list */
5907 for (dlp
= &super
->missing
; *dlp
; dlp
= &(*dlp
)->next
)
5908 if ((*dlp
)->index
== victim
)
5910 imsm_delete(super
, dlp
, victim
);
5914 case update_create_array
: {
5915 /* someone wants to create a new array, we need to be aware of
5916 * a few races/collisions:
5917 * 1/ 'Create' called by two separate instances of mdadm
5918 * 2/ 'Create' versus 'activate_spare': mdadm has chosen
5919 * devices that have since been assimilated via
5921 * In the event this update can not be carried out mdadm will
5922 * (FIX ME) notice that its update did not take hold.
5924 struct imsm_update_create_array
*u
= (void *) update
->buf
;
5925 struct intel_dev
*dv
;
5926 struct imsm_dev
*dev
;
5927 struct imsm_map
*map
, *new_map
;
5928 unsigned long long start
, end
;
5929 unsigned long long new_start
, new_end
;
5931 struct disk_info
*inf
;
5934 /* handle racing creates: first come first serve */
5935 if (u
->dev_idx
< mpb
->num_raid_devs
) {
5936 dprintf("%s: subarray %d already defined\n",
5937 __func__
, u
->dev_idx
);
5941 /* check update is next in sequence */
5942 if (u
->dev_idx
!= mpb
->num_raid_devs
) {
5943 dprintf("%s: can not create array %d expected index %d\n",
5944 __func__
, u
->dev_idx
, mpb
->num_raid_devs
);
5948 new_map
= get_imsm_map(&u
->dev
, 0);
5949 new_start
= __le32_to_cpu(new_map
->pba_of_lba0
);
5950 new_end
= new_start
+ __le32_to_cpu(new_map
->blocks_per_member
);
5951 inf
= get_disk_info(u
);
5953 /* handle activate_spare versus create race:
5954 * check to make sure that overlapping arrays do not include
5957 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
5958 dev
= get_imsm_dev(super
, i
);
5959 map
= get_imsm_map(dev
, 0);
5960 start
= __le32_to_cpu(map
->pba_of_lba0
);
5961 end
= start
+ __le32_to_cpu(map
->blocks_per_member
);
5962 if ((new_start
>= start
&& new_start
<= end
) ||
5963 (start
>= new_start
&& start
<= new_end
))
5968 if (disks_overlap(super
, i
, u
)) {
5969 dprintf("%s: arrays overlap\n", __func__
);
5974 /* check that prepare update was successful */
5975 if (!update
->space
) {
5976 dprintf("%s: prepare update failed\n", __func__
);
5980 /* check that all disks are still active before committing
5981 * changes. FIXME: could we instead handle this by creating a
5982 * degraded array? That's probably not what the user expects,
5983 * so better to drop this update on the floor.
5985 for (i
= 0; i
< new_map
->num_members
; i
++) {
5986 dl
= serial_to_dl(inf
[i
].serial
, super
);
5988 dprintf("%s: disk disappeared\n", __func__
);
5993 super
->updates_pending
++;
5995 /* convert spares to members and fixup ord_tbl */
5996 for (i
= 0; i
< new_map
->num_members
; i
++) {
5997 dl
= serial_to_dl(inf
[i
].serial
, super
);
5998 if (dl
->index
== -1) {
5999 dl
->index
= mpb
->num_disks
;
6001 dl
->disk
.status
|= CONFIGURED_DISK
;
6002 dl
->disk
.status
&= ~SPARE_DISK
;
6004 set_imsm_ord_tbl_ent(new_map
, i
, dl
->index
);
6009 update
->space
= NULL
;
6010 imsm_copy_dev(dev
, &u
->dev
);
6011 dv
->index
= u
->dev_idx
;
6012 dv
->next
= super
->devlist
;
6013 super
->devlist
= dv
;
6014 mpb
->num_raid_devs
++;
6016 imsm_update_version_info(super
);
6019 /* mdmon knows how to release update->space, but not
6020 * ((struct intel_dev *) update->space)->dev
6022 if (update
->space
) {
6028 case update_kill_array
: {
6029 struct imsm_update_kill_array
*u
= (void *) update
->buf
;
6030 int victim
= u
->dev_idx
;
6031 struct active_array
*a
;
6032 struct intel_dev
**dp
;
6033 struct imsm_dev
*dev
;
6035 /* sanity check that we are not affecting the uuid of
6036 * active arrays, or deleting an active array
6038 * FIXME when immutable ids are available, but note that
6039 * we'll also need to fixup the invalidated/active
6040 * subarray indexes in mdstat
6042 for (a
= st
->arrays
; a
; a
= a
->next
)
6043 if (a
->info
.container_member
>= victim
)
6045 /* by definition if mdmon is running at least one array
6046 * is active in the container, so checking
6047 * mpb->num_raid_devs is just extra paranoia
6049 dev
= get_imsm_dev(super
, victim
);
6050 if (a
|| !dev
|| mpb
->num_raid_devs
== 1) {
6051 dprintf("failed to delete subarray-%d\n", victim
);
6055 for (dp
= &super
->devlist
; *dp
;)
6056 if ((*dp
)->index
== (unsigned)super
->current_vol
) {
6059 if ((*dp
)->index
> (unsigned)victim
)
6063 mpb
->num_raid_devs
--;
6064 super
->updates_pending
++;
6067 case update_rename_array
: {
6068 struct imsm_update_rename_array
*u
= (void *) update
->buf
;
6069 char name
[MAX_RAID_SERIAL_LEN
+1];
6070 int target
= u
->dev_idx
;
6071 struct active_array
*a
;
6072 struct imsm_dev
*dev
;
6074 /* sanity check that we are not affecting the uuid of
6077 snprintf(name
, MAX_RAID_SERIAL_LEN
, "%s", (char *) u
->name
);
6078 name
[MAX_RAID_SERIAL_LEN
] = '\0';
6079 for (a
= st
->arrays
; a
; a
= a
->next
)
6080 if (a
->info
.container_member
== target
)
6082 dev
= get_imsm_dev(super
, u
->dev_idx
);
6083 if (a
|| !dev
|| !check_name(super
, name
, 1)) {
6084 dprintf("failed to rename subarray-%d\n", target
);
6088 snprintf((char *) dev
->volume
, MAX_RAID_SERIAL_LEN
, "%s", name
);
6089 super
->updates_pending
++;
6092 case update_add_remove_disk
: {
6093 /* we may be able to repair some arrays if disks are
6094 * being added, check teh status of add_remove_disk
6095 * if discs has been added.
6097 if (add_remove_disk_update(super
)) {
6098 struct active_array
*a
;
6100 super
->updates_pending
++;
6101 for (a
= st
->arrays
; a
; a
= a
->next
)
6102 a
->check_degraded
= 1;
6107 fprintf(stderr
, "error: unsuported process update type:"
6108 "(type: %d)\n", type
);
6112 static void imsm_prepare_update(struct supertype
*st
,
6113 struct metadata_update
*update
)
6116 * Allocate space to hold new disk entries, raid-device entries or a new
6117 * mpb if necessary. The manager synchronously waits for updates to
6118 * complete in the monitor, so new mpb buffers allocated here can be
6119 * integrated by the monitor thread without worrying about live pointers
6120 * in the manager thread.
6122 enum imsm_update_type type
= *(enum imsm_update_type
*) update
->buf
;
6123 struct intel_super
*super
= st
->sb
;
6124 struct imsm_super
*mpb
= super
->anchor
;
6129 case update_reshape_container_disks
: {
6130 /* Every raid device in the container is about to
6131 * gain some more devices, and we will enter a
6133 * So each 'imsm_map' will be bigger, and the imsm_vol
6134 * will now hold 2 of them.
6135 * Thus we need new 'struct imsm_dev' allocations sized
6136 * as sizeof_imsm_dev but with more devices in both maps.
6138 struct imsm_update_reshape
*u
= (void *)update
->buf
;
6139 struct intel_dev
*dl
;
6140 void **space_tail
= (void**)&update
->space_list
;
6142 dprintf("imsm: imsm_prepare_update() for update_reshape\n");
6144 for (dl
= super
->devlist
; dl
; dl
= dl
->next
) {
6145 int size
= sizeof_imsm_dev(dl
->dev
, 1);
6147 if (u
->new_raid_disks
> u
->old_raid_disks
)
6148 size
+= sizeof(__u32
)*2*
6149 (u
->new_raid_disks
- u
->old_raid_disks
);
6158 len
= disks_to_mpb_size(u
->new_raid_disks
);
6159 dprintf("New anchor length is %llu\n", (unsigned long long)len
);
6162 case update_create_array
: {
6163 struct imsm_update_create_array
*u
= (void *) update
->buf
;
6164 struct intel_dev
*dv
;
6165 struct imsm_dev
*dev
= &u
->dev
;
6166 struct imsm_map
*map
= get_imsm_map(dev
, 0);
6168 struct disk_info
*inf
;
6172 inf
= get_disk_info(u
);
6173 len
= sizeof_imsm_dev(dev
, 1);
6174 /* allocate a new super->devlist entry */
6175 dv
= malloc(sizeof(*dv
));
6177 dv
->dev
= malloc(len
);
6182 update
->space
= NULL
;
6186 /* count how many spares will be converted to members */
6187 for (i
= 0; i
< map
->num_members
; i
++) {
6188 dl
= serial_to_dl(inf
[i
].serial
, super
);
6190 /* hmm maybe it failed?, nothing we can do about
6195 if (count_memberships(dl
, super
) == 0)
6198 len
+= activate
* sizeof(struct imsm_disk
);
6205 /* check if we need a larger metadata buffer */
6206 if (super
->next_buf
)
6207 buf_len
= super
->next_len
;
6209 buf_len
= super
->len
;
6211 if (__le32_to_cpu(mpb
->mpb_size
) + len
> buf_len
) {
6212 /* ok we need a larger buf than what is currently allocated
6213 * if this allocation fails process_update will notice that
6214 * ->next_len is set and ->next_buf is NULL
6216 buf_len
= ROUND_UP(__le32_to_cpu(mpb
->mpb_size
) + len
, 512);
6217 if (super
->next_buf
)
6218 free(super
->next_buf
);
6220 super
->next_len
= buf_len
;
6221 if (posix_memalign(&super
->next_buf
, 512, buf_len
) == 0)
6222 memset(super
->next_buf
, 0, buf_len
);
6224 super
->next_buf
= NULL
;
6228 /* must be called while manager is quiesced */
6229 static void imsm_delete(struct intel_super
*super
, struct dl
**dlp
, unsigned index
)
6231 struct imsm_super
*mpb
= super
->anchor
;
6233 struct imsm_dev
*dev
;
6234 struct imsm_map
*map
;
6235 int i
, j
, num_members
;
6238 dprintf("%s: deleting device[%d] from imsm_super\n",
6241 /* shift all indexes down one */
6242 for (iter
= super
->disks
; iter
; iter
= iter
->next
)
6243 if (iter
->index
> (int)index
)
6245 for (iter
= super
->missing
; iter
; iter
= iter
->next
)
6246 if (iter
->index
> (int)index
)
6249 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
6250 dev
= get_imsm_dev(super
, i
);
6251 map
= get_imsm_map(dev
, 0);
6252 num_members
= map
->num_members
;
6253 for (j
= 0; j
< num_members
; j
++) {
6254 /* update ord entries being careful not to propagate
6255 * ord-flags to the first map
6257 ord
= get_imsm_ord_tbl_ent(dev
, j
, -1);
6259 if (ord_to_idx(ord
) <= index
)
6262 map
= get_imsm_map(dev
, 0);
6263 set_imsm_ord_tbl_ent(map
, j
, ord_to_idx(ord
- 1));
6264 map
= get_imsm_map(dev
, 1);
6266 set_imsm_ord_tbl_ent(map
, j
, ord
- 1);
6271 super
->updates_pending
++;
6273 struct dl
*dl
= *dlp
;
6275 *dlp
= (*dlp
)->next
;
6276 __free_imsm_disk(dl
);
6279 #endif /* MDASSEMBLE */
6281 static char disk_by_path
[] = "/dev/disk/by-path/";
6283 static const char *imsm_get_disk_controller_domain(const char *path
)
6285 struct sys_dev
*list
, *hba
= NULL
;
6286 char disk_path
[PATH_MAX
];
6290 list
= find_driver_devices("pci", "ahci");
6291 for (hba
= list
; hba
; hba
= hba
->next
)
6292 if (devpath_to_vendor(hba
->path
) == 0x8086)
6298 strncpy(disk_path
, disk_by_path
, PATH_MAX
- 1);
6299 strncat(disk_path
, path
, PATH_MAX
- strlen(disk_path
) - 1);
6300 if (stat(disk_path
, &st
) == 0) {
6301 dpath
= devt_to_devpath(st
.st_rdev
);
6303 ahci
= path_attached_to_hba(dpath
, hba
->path
);
6306 dprintf("path: %s(%s) hba: %s attached: %d\n",
6307 path
, dpath
, (hba
) ? hba
->path
: "NULL", ahci
);
6308 free_sys_dev(&list
);
6315 static int imsm_find_array_minor_by_subdev(int subdev
, int container
, int *minor
)
6317 char subdev_name
[20];
6318 struct mdstat_ent
*mdstat
;
6320 sprintf(subdev_name
, "%d", subdev
);
6321 mdstat
= mdstat_by_subdev(subdev_name
, container
);
6325 *minor
= mdstat
->devnum
;
6326 free_mdstat(mdstat
);
6330 static int imsm_reshape_is_allowed_on_container(struct supertype
*st
,
6331 struct geo_params
*geo
,
6332 int *old_raid_disks
)
6334 /* currently we only support increasing the number of devices
6335 * for a container. This increases the number of device for each
6336 * member array. They must all be RAID0 or RAID5.
6339 struct mdinfo
*info
, *member
;
6340 int devices_that_can_grow
= 0;
6342 dprintf("imsm: imsm_reshape_is_allowed_on_container(ENTER): "
6343 "st->devnum = (%i)\n",
6346 if (geo
->size
!= -1 ||
6347 geo
->level
!= UnSet
||
6348 geo
->layout
!= UnSet
||
6349 geo
->chunksize
!= 0 ||
6350 geo
->raid_disks
== UnSet
) {
6351 dprintf("imsm: Container operation is allowed for "
6352 "raid disks number change only.\n");
6356 info
= container_content_imsm(st
, NULL
);
6357 for (member
= info
; member
; member
= member
->next
) {
6361 dprintf("imsm: checking device_num: %i\n",
6362 member
->container_member
);
6364 if (geo
->raid_disks
< member
->array
.raid_disks
) {
6365 /* we work on container for Online Capacity Expansion
6366 * only so raid_disks has to grow
6368 dprintf("imsm: for container operation raid disks "
6369 "increase is required\n");
6373 if ((info
->array
.level
!= 0) &&
6374 (info
->array
.level
!= 5)) {
6375 /* we cannot use this container with other raid level
6377 dprintf("imsm: for container operation wrong"
6378 " raid level (%i) detected\n",
6382 /* check for platform support
6383 * for this raid level configuration
6385 struct intel_super
*super
= st
->sb
;
6386 if (!is_raid_level_supported(super
->orom
,
6387 member
->array
.level
,
6389 dprintf("platform does not support raid%d with"
6393 geo
->raid_disks
> 1 ? "s" : "");
6398 if (*old_raid_disks
&&
6399 info
->array
.raid_disks
!= *old_raid_disks
)
6401 *old_raid_disks
= info
->array
.raid_disks
;
6403 /* All raid5 and raid0 volumes in container
6404 * have to be ready for Online Capacity Expansion
6405 * so they need to be assembled. We have already
6406 * checked that no recovery etc is happening.
6408 result
= imsm_find_array_minor_by_subdev(member
->container_member
,
6412 dprintf("imsm: cannot find array\n");
6415 devices_that_can_grow
++;
6418 if (!member
&& devices_that_can_grow
)
6422 dprintf("\tContainer operation allowed\n");
6424 dprintf("\tError: %i\n", ret_val
);
6429 /* Function: get_spares_for_grow
6430 * Description: Allocates memory and creates list of spare devices
6431 * avaliable in container. Checks if spare drive size is acceptable.
6432 * Parameters: Pointer to the supertype structure
6433 * Returns: Pointer to the list of spare devices (mdinfo structure) on success,
6436 static struct mdinfo
*get_spares_for_grow(struct supertype
*st
)
6438 unsigned long long min_size
= min_acceptable_spare_size_imsm(st
);
6439 return container_choose_spares(st
, min_size
, NULL
, NULL
, NULL
, 0);
6442 /******************************************************************************
6443 * function: imsm_create_metadata_update_for_reshape
6444 * Function creates update for whole IMSM container.
6446 ******************************************************************************/
6447 static int imsm_create_metadata_update_for_reshape(
6448 struct supertype
*st
,
6449 struct geo_params
*geo
,
6451 struct imsm_update_reshape
**updatep
)
6453 struct intel_super
*super
= st
->sb
;
6454 struct imsm_super
*mpb
= super
->anchor
;
6455 int update_memory_size
= 0;
6456 struct imsm_update_reshape
*u
= NULL
;
6457 struct mdinfo
*spares
= NULL
;
6459 int delta_disks
= 0;
6462 dprintf("imsm_update_metadata_for_reshape(enter) raid_disks = %i\n",
6465 delta_disks
= geo
->raid_disks
- old_raid_disks
;
6467 /* size of all update data without anchor */
6468 update_memory_size
= sizeof(struct imsm_update_reshape
);
6470 /* now add space for spare disks that we need to add. */
6471 update_memory_size
+= sizeof(u
->new_disks
[0]) * (delta_disks
- 1);
6473 u
= calloc(1, update_memory_size
);
6476 "cannot get memory for imsm_update_reshape update\n");
6479 u
->type
= update_reshape_container_disks
;
6480 u
->old_raid_disks
= old_raid_disks
;
6481 u
->new_raid_disks
= geo
->raid_disks
;
6483 /* now get spare disks list
6485 spares
= get_spares_for_grow(st
);
6488 || delta_disks
> spares
->array
.spare_disks
) {
6489 dprintf("imsm: ERROR: Cannot get spare devices.\n");
6493 /* we have got spares
6494 * update disk list in imsm_disk list table in anchor
6496 dprintf("imsm: %i spares are available.\n\n",
6497 spares
->array
.spare_disks
);
6500 for (i
= 0; i
< delta_disks
; i
++) {
6505 u
->new_disks
[i
] = makedev(dev
->disk
.major
,
6507 dl
= get_disk_super(super
, dev
->disk
.major
, dev
->disk
.minor
);
6508 dl
->index
= mpb
->num_disks
;
6518 dprintf("imsm: reshape update preparation :");
6519 if (i
== delta_disks
) {
6522 return update_memory_size
;
6525 dprintf(" Error\n");
6530 static void imsm_update_metadata_locally(struct supertype
*st
,
6533 struct metadata_update mu
;
6538 mu
.space_list
= NULL
;
6540 imsm_prepare_update(st
, &mu
);
6541 imsm_process_update(st
, &mu
);
6543 while (mu
.space_list
) {
6544 void **space
= mu
.space_list
;
6545 mu
.space_list
= *space
;
6550 /***************************************************************************
6551 * Function: imsm_analyze_change
6552 * Description: Function analyze change for single volume
6553 * and validate if transition is supported
6554 * Parameters: Geometry parameters, supertype structure
6555 * Returns: Operation type code on success, -1 if fail
6556 ****************************************************************************/
6557 enum imsm_reshape_type
imsm_analyze_change(struct supertype
*st
,
6558 struct geo_params
*geo
)
6564 getinfo_super_imsm_volume(st
, &info
, NULL
);
6566 if ((geo
->level
!= info
.array
.level
) &&
6567 (geo
->level
>= 0) &&
6568 (geo
->level
!= UnSet
)) {
6569 switch (info
.array
.level
) {
6571 if (geo
->level
== 5) {
6572 change
= CH_LEVEL_MIGRATION
;
6575 if (geo
->level
== 10) {
6576 change
= CH_TAKEOVER
;
6581 if (geo
->level
!= 0)
6582 change
= CH_LEVEL_MIGRATION
;
6585 if (geo
->level
== 0) {
6586 change
= CH_TAKEOVER
;
6593 Name
" Error. Level Migration from %d to %d "
6595 info
.array
.level
, geo
->level
);
6596 goto analyse_change_exit
;
6599 geo
->level
= info
.array
.level
;
6601 if ((geo
->layout
!= info
.array
.layout
)
6602 && ((geo
->layout
!= UnSet
) && (geo
->layout
!= -1))) {
6603 change
= CH_LEVEL_MIGRATION
;
6604 if ((info
.array
.layout
== 0)
6605 && (info
.array
.level
== 5)
6606 && (geo
->layout
== 5)) {
6607 /* reshape 5 -> 4 */
6608 } else if ((info
.array
.layout
== 5)
6609 && (info
.array
.level
== 5)
6610 && (geo
->layout
== 0)) {
6611 /* reshape 4 -> 5 */
6616 Name
" Error. Layout Migration from %d to %d "
6618 info
.array
.layout
, geo
->layout
);
6620 goto analyse_change_exit
;
6623 geo
->layout
= info
.array
.layout
;
6625 if ((geo
->chunksize
> 0) && (geo
->chunksize
!= UnSet
)
6626 && (geo
->chunksize
!= info
.array
.chunk_size
))
6627 change
= CH_CHUNK_MIGR
;
6629 geo
->chunksize
= info
.array
.chunk_size
;
6631 if (!validate_geometry_imsm(st
,
6635 (geo
->chunksize
/ 1024),
6641 struct intel_super
*super
= st
->sb
;
6642 struct imsm_super
*mpb
= super
->anchor
;
6644 if (mpb
->num_raid_devs
> 1) {
6646 Name
" Error. Cannot perform operation on %s"
6647 "- for this operation it MUST be single "
6648 "array in container\n",
6654 analyse_change_exit
:
6659 static int imsm_reshape_super(struct supertype
*st
, long long size
, int level
,
6660 int layout
, int chunksize
, int raid_disks
,
6661 char *backup
, char *dev
, int verbose
)
6664 struct geo_params geo
;
6666 dprintf("imsm: reshape_super called.\n");
6668 memset(&geo
, sizeof(struct geo_params
), 0);
6671 geo
.dev_id
= st
->devnum
;
6674 geo
.layout
= layout
;
6675 geo
.chunksize
= chunksize
;
6676 geo
.raid_disks
= raid_disks
;
6678 dprintf("\tfor level : %i\n", geo
.level
);
6679 dprintf("\tfor raid_disks : %i\n", geo
.raid_disks
);
6681 if (experimental() == 0)
6684 if (st
->container_dev
== st
->devnum
) {
6685 /* On container level we can only increase number of devices. */
6686 dprintf("imsm: info: Container operation\n");
6687 int old_raid_disks
= 0;
6688 if (imsm_reshape_is_allowed_on_container(
6689 st
, &geo
, &old_raid_disks
)) {
6690 struct imsm_update_reshape
*u
= NULL
;
6693 len
= imsm_create_metadata_update_for_reshape(
6694 st
, &geo
, old_raid_disks
, &u
);
6697 dprintf("imsm: Cannot prepare update\n");
6698 goto exit_imsm_reshape_super
;
6702 /* update metadata locally */
6703 imsm_update_metadata_locally(st
, u
, len
);
6704 /* and possibly remotely */
6705 if (st
->update_tail
)
6706 append_metadata_update(st
, u
, len
);
6711 fprintf(stderr
, Name
"imsm: Operation is not allowed "
6712 "on this container\n");
6715 /* On volume level we support following operations
6716 * - takeover: raid10 -> raid0; raid0 -> raid10
6717 * - chunk size migration
6718 * - migration: raid5 -> raid0; raid0 -> raid5
6720 struct intel_super
*super
= st
->sb
;
6721 struct intel_dev
*dev
= super
->devlist
;
6723 dprintf("imsm: info: Volume operation\n");
6724 /* find requested device */
6726 imsm_find_array_minor_by_subdev(dev
->index
, st
->container_dev
, &devnum
);
6727 if (devnum
== geo
.dev_id
)
6732 fprintf(stderr
, Name
" Cannot find %s (%i) subarray\n",
6733 geo
.dev_name
, geo
.dev_id
);
6734 goto exit_imsm_reshape_super
;
6736 super
->current_vol
= dev
->index
;
6737 change
= imsm_analyze_change(st
, &geo
);
6745 case CH_LEVEL_MIGRATION
:
6753 exit_imsm_reshape_super
:
6754 dprintf("imsm: reshape_super Exit code = %i\n", ret_val
);
6758 static int imsm_manage_reshape(
6759 int afd
, struct mdinfo
*sra
, struct reshape
*reshape
,
6760 struct supertype
*st
, unsigned long stripes
,
6761 int *fds
, unsigned long long *offsets
,
6762 int dests
, int *destfd
, unsigned long long *destoffsets
)
6764 /* Just use child_monitor for now */
6765 return child_monitor(
6766 afd
, sra
, reshape
, st
, stripes
,
6767 fds
, offsets
, dests
, destfd
, destoffsets
);
6770 struct superswitch super_imsm
= {
6772 .examine_super
= examine_super_imsm
,
6773 .brief_examine_super
= brief_examine_super_imsm
,
6774 .brief_examine_subarrays
= brief_examine_subarrays_imsm
,
6775 .export_examine_super
= export_examine_super_imsm
,
6776 .detail_super
= detail_super_imsm
,
6777 .brief_detail_super
= brief_detail_super_imsm
,
6778 .write_init_super
= write_init_super_imsm
,
6779 .validate_geometry
= validate_geometry_imsm
,
6780 .add_to_super
= add_to_super_imsm
,
6781 .remove_from_super
= remove_from_super_imsm
,
6782 .detail_platform
= detail_platform_imsm
,
6783 .kill_subarray
= kill_subarray_imsm
,
6784 .update_subarray
= update_subarray_imsm
,
6785 .load_container
= load_container_imsm
,
6787 .match_home
= match_home_imsm
,
6788 .uuid_from_super
= uuid_from_super_imsm
,
6789 .getinfo_super
= getinfo_super_imsm
,
6790 .getinfo_super_disks
= getinfo_super_disks_imsm
,
6791 .update_super
= update_super_imsm
,
6793 .avail_size
= avail_size_imsm
,
6794 .min_acceptable_spare_size
= min_acceptable_spare_size_imsm
,
6796 .compare_super
= compare_super_imsm
,
6798 .load_super
= load_super_imsm
,
6799 .init_super
= init_super_imsm
,
6800 .store_super
= store_super_imsm
,
6801 .free_super
= free_super_imsm
,
6802 .match_metadata_desc
= match_metadata_desc_imsm
,
6803 .container_content
= container_content_imsm
,
6804 .default_geometry
= default_geometry_imsm
,
6805 .get_disk_controller_domain
= imsm_get_disk_controller_domain
,
6806 .reshape_super
= imsm_reshape_super
,
6807 .manage_reshape
= imsm_manage_reshape
,
6814 .open_new
= imsm_open_new
,
6815 .set_array_state
= imsm_set_array_state
,
6816 .set_disk
= imsm_set_disk
,
6817 .sync_metadata
= imsm_sync_metadata
,
6818 .activate_spare
= imsm_activate_spare
,
6819 .process_update
= imsm_process_update
,
6820 .prepare_update
= imsm_prepare_update
,
6821 #endif /* MDASSEMBLE */