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
;
237 enum sys_dev_type type
;
240 struct intel_hba
*next
;
247 /* internal representation of IMSM metadata */
250 void *buf
; /* O_DIRECT buffer for reading/writing metadata */
251 struct imsm_super
*anchor
; /* immovable parameters */
253 size_t len
; /* size of the 'buf' allocation */
254 void *next_buf
; /* for realloc'ing buf from the manager */
256 int updates_pending
; /* count of pending updates for mdmon */
257 int current_vol
; /* index of raid device undergoing creation */
258 __u32 create_offset
; /* common start for 'current_vol' */
259 __u32 random
; /* random data for seeding new family numbers */
260 struct intel_dev
*devlist
;
264 __u8 serial
[MAX_RAID_SERIAL_LEN
];
267 struct imsm_disk disk
;
270 struct extent
*e
; /* for determining freespace @ create */
271 int raiddisk
; /* slot to fill in autolayout */
274 struct dl
*disk_mgmt_list
; /* list of disks to add/remove while mdmon
276 struct dl
*missing
; /* disks removed while we weren't looking */
277 struct bbm_log
*bbm_log
;
278 struct intel_hba
*hba
; /* device path of the raid controller for this metadata */
279 const struct imsm_orom
*orom
; /* platform firmware support */
280 struct intel_super
*next
; /* (temp) list for disambiguating family_num */
284 struct imsm_disk disk
;
285 #define IMSM_UNKNOWN_OWNER (-1)
287 struct intel_disk
*next
;
291 unsigned long long start
, size
;
294 /* definitions of reshape process types */
295 enum imsm_reshape_type
{
300 /* definition of messages passed to imsm_process_update */
301 enum imsm_update_type
{
302 update_activate_spare
,
306 update_add_remove_disk
,
307 update_reshape_container_disks
,
311 struct imsm_update_activate_spare
{
312 enum imsm_update_type type
;
316 struct imsm_update_activate_spare
*next
;
329 enum takeover_direction
{
333 struct imsm_update_takeover
{
334 enum imsm_update_type type
;
336 enum takeover_direction direction
;
339 struct imsm_update_reshape
{
340 enum imsm_update_type type
;
343 int new_disks
[1]; /* new_raid_disks - old_raid_disks makedev number */
347 __u8 serial
[MAX_RAID_SERIAL_LEN
];
350 struct imsm_update_create_array
{
351 enum imsm_update_type type
;
356 struct imsm_update_kill_array
{
357 enum imsm_update_type type
;
361 struct imsm_update_rename_array
{
362 enum imsm_update_type type
;
363 __u8 name
[MAX_RAID_SERIAL_LEN
];
367 struct imsm_update_add_remove_disk
{
368 enum imsm_update_type type
;
372 static const char *_sys_dev_type
[] = {
373 [SYS_DEV_UNKNOWN
] = "Unknown",
374 [SYS_DEV_SAS
] = "SAS",
375 [SYS_DEV_SATA
] = "SATA"
378 const char *get_sys_dev_type(enum sys_dev_type type
)
380 if (type
>= SYS_DEV_MAX
)
381 type
= SYS_DEV_UNKNOWN
;
383 return _sys_dev_type
[type
];
387 static struct intel_hba
* alloc_intel_hba(struct sys_dev
*device
)
389 struct intel_hba
*result
= malloc(sizeof(*result
));
391 result
->type
= device
->type
;
392 result
->path
= strdup(device
->path
);
394 if (result
->path
&& (result
->pci_id
= strrchr(result
->path
, '/')) != NULL
)
400 static struct intel_hba
* find_intel_hba(struct intel_hba
*hba
, struct sys_dev
*device
)
402 struct intel_hba
*result
=NULL
;
403 for (result
= hba
; result
; result
= result
->next
) {
404 if (result
->type
== device
->type
&& strcmp(result
->path
, device
->path
) == 0)
411 static int attach_hba_to_super(struct intel_super
*super
, struct sys_dev
*device
)
413 struct intel_hba
*hba
;
415 /* check if disk attached to Intel HBA */
416 hba
= find_intel_hba(super
->hba
, device
);
419 /* Check if HBA is already attached to super */
420 if (super
->hba
== NULL
) {
421 super
->hba
= alloc_intel_hba(device
);
426 /* Intel metadata allows for all disks attached to the same type HBA.
427 * Do not sypport odf HBA types mixing
429 if (device
->type
!= hba
->type
)
435 hba
->next
= alloc_intel_hba(device
);
439 static struct sys_dev
* find_disk_attached_hba(int fd
, const char *devname
)
441 struct sys_dev
*list
, *elem
, *prev
;
444 if ((list
= find_intel_devices()) == NULL
)
448 disk_path
= (char *) devname
;
450 disk_path
= diskfd_to_devpath(fd
);
457 for (prev
= NULL
, elem
= list
; elem
; prev
= elem
, elem
= elem
->next
) {
458 if (path_attached_to_hba(disk_path
, elem
->path
)) {
462 prev
->next
= elem
->next
;
464 if (disk_path
!= devname
)
470 if (disk_path
!= devname
)
476 #endif /* MDASSEMBLE */
479 static int find_intel_hba_capability(int fd
, struct intel_super
*super
, int verbose
);
481 static struct supertype
*match_metadata_desc_imsm(char *arg
)
483 struct supertype
*st
;
485 if (strcmp(arg
, "imsm") != 0 &&
486 strcmp(arg
, "default") != 0
490 st
= malloc(sizeof(*st
));
493 memset(st
, 0, sizeof(*st
));
494 st
->container_dev
= NoMdDev
;
495 st
->ss
= &super_imsm
;
496 st
->max_devs
= IMSM_MAX_DEVICES
;
497 st
->minor_version
= 0;
503 static __u8
*get_imsm_version(struct imsm_super
*mpb
)
505 return &mpb
->sig
[MPB_SIG_LEN
];
509 /* retrieve a disk directly from the anchor when the anchor is known to be
510 * up-to-date, currently only at load time
512 static struct imsm_disk
*__get_imsm_disk(struct imsm_super
*mpb
, __u8 index
)
514 if (index
>= mpb
->num_disks
)
516 return &mpb
->disk
[index
];
519 /* retrieve the disk description based on a index of the disk
522 static struct dl
*get_imsm_dl_disk(struct intel_super
*super
, __u8 index
)
526 for (d
= super
->disks
; d
; d
= d
->next
)
527 if (d
->index
== index
)
532 /* retrieve a disk from the parsed metadata */
533 static struct imsm_disk
*get_imsm_disk(struct intel_super
*super
, __u8 index
)
537 dl
= get_imsm_dl_disk(super
, index
);
544 /* generate a checksum directly from the anchor when the anchor is known to be
545 * up-to-date, currently only at load or write_super after coalescing
547 static __u32
__gen_imsm_checksum(struct imsm_super
*mpb
)
549 __u32 end
= mpb
->mpb_size
/ sizeof(end
);
550 __u32
*p
= (__u32
*) mpb
;
554 sum
+= __le32_to_cpu(*p
);
558 return sum
- __le32_to_cpu(mpb
->check_sum
);
561 static size_t sizeof_imsm_map(struct imsm_map
*map
)
563 return sizeof(struct imsm_map
) + sizeof(__u32
) * (map
->num_members
- 1);
566 struct imsm_map
*get_imsm_map(struct imsm_dev
*dev
, int second_map
)
568 /* A device can have 2 maps if it is in the middle of a migration.
570 * 0 - we return the first map
571 * 1 - we return the second map if it exists, else NULL
572 * -1 - we return the second map if it exists, else the first
574 struct imsm_map
*map
= &dev
->vol
.map
[0];
576 if (second_map
== 1 && !dev
->vol
.migr_state
)
578 else if (second_map
== 1 ||
579 (second_map
< 0 && dev
->vol
.migr_state
)) {
582 return ptr
+ sizeof_imsm_map(map
);
588 /* return the size of the device.
589 * migr_state increases the returned size if map[0] were to be duplicated
591 static size_t sizeof_imsm_dev(struct imsm_dev
*dev
, int migr_state
)
593 size_t size
= sizeof(*dev
) - sizeof(struct imsm_map
) +
594 sizeof_imsm_map(get_imsm_map(dev
, 0));
596 /* migrating means an additional map */
597 if (dev
->vol
.migr_state
)
598 size
+= sizeof_imsm_map(get_imsm_map(dev
, 1));
600 size
+= sizeof_imsm_map(get_imsm_map(dev
, 0));
606 /* retrieve disk serial number list from a metadata update */
607 static struct disk_info
*get_disk_info(struct imsm_update_create_array
*update
)
610 struct disk_info
*inf
;
612 inf
= u
+ sizeof(*update
) - sizeof(struct imsm_dev
) +
613 sizeof_imsm_dev(&update
->dev
, 0);
619 static struct imsm_dev
*__get_imsm_dev(struct imsm_super
*mpb
, __u8 index
)
625 if (index
>= mpb
->num_raid_devs
)
628 /* devices start after all disks */
629 offset
= ((void *) &mpb
->disk
[mpb
->num_disks
]) - _mpb
;
631 for (i
= 0; i
<= index
; i
++)
633 return _mpb
+ offset
;
635 offset
+= sizeof_imsm_dev(_mpb
+ offset
, 0);
640 static struct imsm_dev
*get_imsm_dev(struct intel_super
*super
, __u8 index
)
642 struct intel_dev
*dv
;
644 if (index
>= super
->anchor
->num_raid_devs
)
646 for (dv
= super
->devlist
; dv
; dv
= dv
->next
)
647 if (dv
->index
== index
)
655 * == 1 get second map
656 * == -1 than get map according to the current migr_state
658 static __u32
get_imsm_ord_tbl_ent(struct imsm_dev
*dev
,
662 struct imsm_map
*map
;
664 map
= get_imsm_map(dev
, second_map
);
666 /* top byte identifies disk under rebuild */
667 return __le32_to_cpu(map
->disk_ord_tbl
[slot
]);
670 #define ord_to_idx(ord) (((ord) << 8) >> 8)
671 static __u32
get_imsm_disk_idx(struct imsm_dev
*dev
, int slot
, int second_map
)
673 __u32 ord
= get_imsm_ord_tbl_ent(dev
, slot
, second_map
);
675 return ord_to_idx(ord
);
678 static void set_imsm_ord_tbl_ent(struct imsm_map
*map
, int slot
, __u32 ord
)
680 map
->disk_ord_tbl
[slot
] = __cpu_to_le32(ord
);
683 static int get_imsm_disk_slot(struct imsm_map
*map
, unsigned idx
)
688 for (slot
= 0; slot
< map
->num_members
; slot
++) {
689 ord
= __le32_to_cpu(map
->disk_ord_tbl
[slot
]);
690 if (ord_to_idx(ord
) == idx
)
697 static int get_imsm_raid_level(struct imsm_map
*map
)
699 if (map
->raid_level
== 1) {
700 if (map
->num_members
== 2)
706 return map
->raid_level
;
709 static int cmp_extent(const void *av
, const void *bv
)
711 const struct extent
*a
= av
;
712 const struct extent
*b
= bv
;
713 if (a
->start
< b
->start
)
715 if (a
->start
> b
->start
)
720 static int count_memberships(struct dl
*dl
, struct intel_super
*super
)
725 for (i
= 0; i
< super
->anchor
->num_raid_devs
; i
++) {
726 struct imsm_dev
*dev
= get_imsm_dev(super
, i
);
727 struct imsm_map
*map
= get_imsm_map(dev
, 0);
729 if (get_imsm_disk_slot(map
, dl
->index
) >= 0)
736 static struct extent
*get_extents(struct intel_super
*super
, struct dl
*dl
)
738 /* find a list of used extents on the given physical device */
739 struct extent
*rv
, *e
;
741 int memberships
= count_memberships(dl
, super
);
742 __u32 reservation
= MPB_SECTOR_CNT
+ IMSM_RESERVED_SECTORS
;
744 rv
= malloc(sizeof(struct extent
) * (memberships
+ 1));
749 for (i
= 0; i
< super
->anchor
->num_raid_devs
; i
++) {
750 struct imsm_dev
*dev
= get_imsm_dev(super
, i
);
751 struct imsm_map
*map
= get_imsm_map(dev
, 0);
753 if (get_imsm_disk_slot(map
, dl
->index
) >= 0) {
754 e
->start
= __le32_to_cpu(map
->pba_of_lba0
);
755 e
->size
= __le32_to_cpu(map
->blocks_per_member
);
759 qsort(rv
, memberships
, sizeof(*rv
), cmp_extent
);
761 /* determine the start of the metadata
762 * when no raid devices are defined use the default
763 * ...otherwise allow the metadata to truncate the value
764 * as is the case with older versions of imsm
767 struct extent
*last
= &rv
[memberships
- 1];
770 remainder
= __le32_to_cpu(dl
->disk
.total_blocks
) -
771 (last
->start
+ last
->size
);
772 /* round down to 1k block to satisfy precision of the kernel
776 /* make sure remainder is still sane */
777 if (remainder
< (unsigned)ROUND_UP(super
->len
, 512) >> 9)
778 remainder
= ROUND_UP(super
->len
, 512) >> 9;
779 if (reservation
> remainder
)
780 reservation
= remainder
;
782 e
->start
= __le32_to_cpu(dl
->disk
.total_blocks
) - reservation
;
787 /* try to determine how much space is reserved for metadata from
788 * the last get_extents() entry, otherwise fallback to the
791 static __u32
imsm_reserved_sectors(struct intel_super
*super
, struct dl
*dl
)
797 /* for spares just return a minimal reservation which will grow
798 * once the spare is picked up by an array
801 return MPB_SECTOR_CNT
;
803 e
= get_extents(super
, dl
);
805 return MPB_SECTOR_CNT
+ IMSM_RESERVED_SECTORS
;
807 /* scroll to last entry */
808 for (i
= 0; e
[i
].size
; i
++)
811 rv
= __le32_to_cpu(dl
->disk
.total_blocks
) - e
[i
].start
;
818 static int is_spare(struct imsm_disk
*disk
)
820 return (disk
->status
& SPARE_DISK
) == SPARE_DISK
;
823 static int is_configured(struct imsm_disk
*disk
)
825 return (disk
->status
& CONFIGURED_DISK
) == CONFIGURED_DISK
;
828 static int is_failed(struct imsm_disk
*disk
)
830 return (disk
->status
& FAILED_DISK
) == FAILED_DISK
;
833 /* Return minimum size of a spare that can be used in this array*/
834 static unsigned long long min_acceptable_spare_size_imsm(struct supertype
*st
)
836 struct intel_super
*super
= st
->sb
;
840 unsigned long long rv
= 0;
844 /* find first active disk in array */
846 while (dl
&& (is_failed(&dl
->disk
) || dl
->index
== -1))
850 /* find last lba used by subarrays */
851 e
= get_extents(super
, dl
);
854 for (i
= 0; e
[i
].size
; i
++)
857 rv
= e
[i
-1].start
+ e
[i
-1].size
;
859 /* add the amount of space needed for metadata */
860 rv
= rv
+ MPB_SECTOR_CNT
+ IMSM_RESERVED_SECTORS
;
865 static __u64
blocks_per_migr_unit(struct imsm_dev
*dev
);
867 static void print_imsm_dev(struct imsm_dev
*dev
, char *uuid
, int disk_idx
)
871 struct imsm_map
*map
= get_imsm_map(dev
, 0);
872 struct imsm_map
*map2
= get_imsm_map(dev
, 1);
876 printf("[%.16s]:\n", dev
->volume
);
877 printf(" UUID : %s\n", uuid
);
878 printf(" RAID Level : %d", get_imsm_raid_level(map
));
880 printf(" <-- %d", get_imsm_raid_level(map2
));
882 printf(" Members : %d", map
->num_members
);
884 printf(" <-- %d", map2
->num_members
);
886 printf(" Slots : [");
887 for (i
= 0; i
< map
->num_members
; i
++) {
888 ord
= get_imsm_ord_tbl_ent(dev
, i
, 0);
889 printf("%s", ord
& IMSM_ORD_REBUILD
? "_" : "U");
894 for (i
= 0; i
< map2
->num_members
; i
++) {
895 ord
= get_imsm_ord_tbl_ent(dev
, i
, 1);
896 printf("%s", ord
& IMSM_ORD_REBUILD
? "_" : "U");
901 printf(" Failed disk : ");
902 if (map
->failed_disk_num
== 0xff)
905 printf("%i", map
->failed_disk_num
);
907 slot
= get_imsm_disk_slot(map
, disk_idx
);
909 ord
= get_imsm_ord_tbl_ent(dev
, slot
, -1);
910 printf(" This Slot : %d%s\n", slot
,
911 ord
& IMSM_ORD_REBUILD
? " (out-of-sync)" : "");
913 printf(" This Slot : ?\n");
914 sz
= __le32_to_cpu(dev
->size_high
);
916 sz
+= __le32_to_cpu(dev
->size_low
);
917 printf(" Array Size : %llu%s\n", (unsigned long long)sz
,
918 human_size(sz
* 512));
919 sz
= __le32_to_cpu(map
->blocks_per_member
);
920 printf(" Per Dev Size : %llu%s\n", (unsigned long long)sz
,
921 human_size(sz
* 512));
922 printf(" Sector Offset : %u\n",
923 __le32_to_cpu(map
->pba_of_lba0
));
924 printf(" Num Stripes : %u\n",
925 __le32_to_cpu(map
->num_data_stripes
));
926 printf(" Chunk Size : %u KiB",
927 __le16_to_cpu(map
->blocks_per_strip
) / 2);
929 printf(" <-- %u KiB",
930 __le16_to_cpu(map2
->blocks_per_strip
) / 2);
932 printf(" Reserved : %d\n", __le32_to_cpu(dev
->reserved_blocks
));
933 printf(" Migrate State : ");
934 if (dev
->vol
.migr_state
) {
935 if (migr_type(dev
) == MIGR_INIT
)
936 printf("initialize\n");
937 else if (migr_type(dev
) == MIGR_REBUILD
)
939 else if (migr_type(dev
) == MIGR_VERIFY
)
941 else if (migr_type(dev
) == MIGR_GEN_MIGR
)
942 printf("general migration\n");
943 else if (migr_type(dev
) == MIGR_STATE_CHANGE
)
944 printf("state change\n");
945 else if (migr_type(dev
) == MIGR_REPAIR
)
948 printf("<unknown:%d>\n", migr_type(dev
));
951 printf(" Map State : %s", map_state_str
[map
->map_state
]);
952 if (dev
->vol
.migr_state
) {
953 struct imsm_map
*map
= get_imsm_map(dev
, 1);
955 printf(" <-- %s", map_state_str
[map
->map_state
]);
956 printf("\n Checkpoint : %u (%llu)",
957 __le32_to_cpu(dev
->vol
.curr_migr_unit
),
958 (unsigned long long)blocks_per_migr_unit(dev
));
961 printf(" Dirty State : %s\n", dev
->vol
.dirty
? "dirty" : "clean");
964 static void print_imsm_disk(struct imsm_super
*mpb
, int index
, __u32 reserved
)
966 struct imsm_disk
*disk
= __get_imsm_disk(mpb
, index
);
967 char str
[MAX_RAID_SERIAL_LEN
+ 1];
970 if (index
< 0 || !disk
)
974 snprintf(str
, MAX_RAID_SERIAL_LEN
+ 1, "%s", disk
->serial
);
975 printf(" Disk%02d Serial : %s\n", index
, str
);
976 printf(" State :%s%s%s\n", is_spare(disk
) ? " spare" : "",
977 is_configured(disk
) ? " active" : "",
978 is_failed(disk
) ? " failed" : "");
979 printf(" Id : %08x\n", __le32_to_cpu(disk
->scsi_id
));
980 sz
= __le32_to_cpu(disk
->total_blocks
) - reserved
;
981 printf(" Usable Size : %llu%s\n", (unsigned long long)sz
,
982 human_size(sz
* 512));
985 static void getinfo_super_imsm(struct supertype
*st
, struct mdinfo
*info
, char *map
);
987 static void examine_super_imsm(struct supertype
*st
, char *homehost
)
989 struct intel_super
*super
= st
->sb
;
990 struct imsm_super
*mpb
= super
->anchor
;
991 char str
[MAX_SIGNATURE_LENGTH
];
996 __u32 reserved
= imsm_reserved_sectors(super
, super
->disks
);
999 snprintf(str
, MPB_SIG_LEN
, "%s", mpb
->sig
);
1000 printf(" Magic : %s\n", str
);
1001 snprintf(str
, strlen(MPB_VERSION_RAID0
), "%s", get_imsm_version(mpb
));
1002 printf(" Version : %s\n", get_imsm_version(mpb
));
1003 printf(" Orig Family : %08x\n", __le32_to_cpu(mpb
->orig_family_num
));
1004 printf(" Family : %08x\n", __le32_to_cpu(mpb
->family_num
));
1005 printf(" Generation : %08x\n", __le32_to_cpu(mpb
->generation_num
));
1006 getinfo_super_imsm(st
, &info
, NULL
);
1007 fname_from_uuid(st
, &info
, nbuf
, ':');
1008 printf(" UUID : %s\n", nbuf
+ 5);
1009 sum
= __le32_to_cpu(mpb
->check_sum
);
1010 printf(" Checksum : %08x %s\n", sum
,
1011 __gen_imsm_checksum(mpb
) == sum
? "correct" : "incorrect");
1012 printf(" MPB Sectors : %d\n", mpb_sectors(mpb
));
1013 printf(" Disks : %d\n", mpb
->num_disks
);
1014 printf(" RAID Devices : %d\n", mpb
->num_raid_devs
);
1015 print_imsm_disk(mpb
, super
->disks
->index
, reserved
);
1016 if (super
->bbm_log
) {
1017 struct bbm_log
*log
= super
->bbm_log
;
1020 printf("Bad Block Management Log:\n");
1021 printf(" Log Size : %d\n", __le32_to_cpu(mpb
->bbm_log_size
));
1022 printf(" Signature : %x\n", __le32_to_cpu(log
->signature
));
1023 printf(" Entry Count : %d\n", __le32_to_cpu(log
->entry_count
));
1024 printf(" Spare Blocks : %d\n", __le32_to_cpu(log
->reserved_spare_block_count
));
1025 printf(" First Spare : %llx\n",
1026 (unsigned long long) __le64_to_cpu(log
->first_spare_lba
));
1028 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
1030 struct imsm_dev
*dev
= __get_imsm_dev(mpb
, i
);
1032 super
->current_vol
= i
;
1033 getinfo_super_imsm(st
, &info
, NULL
);
1034 fname_from_uuid(st
, &info
, nbuf
, ':');
1035 print_imsm_dev(dev
, nbuf
+ 5, super
->disks
->index
);
1037 for (i
= 0; i
< mpb
->num_disks
; i
++) {
1038 if (i
== super
->disks
->index
)
1040 print_imsm_disk(mpb
, i
, reserved
);
1042 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
1043 struct imsm_disk
*disk
;
1044 char str
[MAX_RAID_SERIAL_LEN
+ 1];
1052 snprintf(str
, MAX_RAID_SERIAL_LEN
+ 1, "%s", disk
->serial
);
1053 printf(" Disk Serial : %s\n", str
);
1054 printf(" State :%s%s%s\n", is_spare(disk
) ? " spare" : "",
1055 is_configured(disk
) ? " active" : "",
1056 is_failed(disk
) ? " failed" : "");
1057 printf(" Id : %08x\n", __le32_to_cpu(disk
->scsi_id
));
1058 sz
= __le32_to_cpu(disk
->total_blocks
) - reserved
;
1059 printf(" Usable Size : %llu%s\n", (unsigned long long)sz
,
1060 human_size(sz
* 512));
1064 static void brief_examine_super_imsm(struct supertype
*st
, int verbose
)
1066 /* We just write a generic IMSM ARRAY entry */
1069 struct intel_super
*super
= st
->sb
;
1071 if (!super
->anchor
->num_raid_devs
) {
1072 printf("ARRAY metadata=imsm\n");
1076 getinfo_super_imsm(st
, &info
, NULL
);
1077 fname_from_uuid(st
, &info
, nbuf
, ':');
1078 printf("ARRAY metadata=imsm UUID=%s\n", nbuf
+ 5);
1081 static void brief_examine_subarrays_imsm(struct supertype
*st
, int verbose
)
1083 /* We just write a generic IMSM ARRAY entry */
1087 struct intel_super
*super
= st
->sb
;
1090 if (!super
->anchor
->num_raid_devs
)
1093 getinfo_super_imsm(st
, &info
, NULL
);
1094 fname_from_uuid(st
, &info
, nbuf
, ':');
1095 for (i
= 0; i
< super
->anchor
->num_raid_devs
; i
++) {
1096 struct imsm_dev
*dev
= get_imsm_dev(super
, i
);
1098 super
->current_vol
= i
;
1099 getinfo_super_imsm(st
, &info
, NULL
);
1100 fname_from_uuid(st
, &info
, nbuf1
, ':');
1101 printf("ARRAY /dev/md/%.16s container=%s member=%d UUID=%s\n",
1102 dev
->volume
, nbuf
+ 5, i
, nbuf1
+ 5);
1106 static void export_examine_super_imsm(struct supertype
*st
)
1108 struct intel_super
*super
= st
->sb
;
1109 struct imsm_super
*mpb
= super
->anchor
;
1113 getinfo_super_imsm(st
, &info
, NULL
);
1114 fname_from_uuid(st
, &info
, nbuf
, ':');
1115 printf("MD_METADATA=imsm\n");
1116 printf("MD_LEVEL=container\n");
1117 printf("MD_UUID=%s\n", nbuf
+5);
1118 printf("MD_DEVICES=%u\n", mpb
->num_disks
);
1121 static void detail_super_imsm(struct supertype
*st
, char *homehost
)
1126 getinfo_super_imsm(st
, &info
, NULL
);
1127 fname_from_uuid(st
, &info
, nbuf
, ':');
1128 printf("\n UUID : %s\n", nbuf
+ 5);
1131 static void brief_detail_super_imsm(struct supertype
*st
)
1135 getinfo_super_imsm(st
, &info
, NULL
);
1136 fname_from_uuid(st
, &info
, nbuf
, ':');
1137 printf(" UUID=%s", nbuf
+ 5);
1140 static int imsm_read_serial(int fd
, char *devname
, __u8
*serial
);
1141 static void fd2devname(int fd
, char *name
);
1143 static int ahci_enumerate_ports(const char *hba_path
, int port_count
, int host_base
, int verbose
)
1145 /* dump an unsorted list of devices attached to AHCI Intel storage
1146 * controller, as well as non-connected ports
1148 int hba_len
= strlen(hba_path
) + 1;
1153 unsigned long port_mask
= (1 << port_count
) - 1;
1155 if (port_count
> (int)sizeof(port_mask
) * 8) {
1157 fprintf(stderr
, Name
": port_count %d out of range\n", port_count
);
1161 /* scroll through /sys/dev/block looking for devices attached to
1164 dir
= opendir("/sys/dev/block");
1165 for (ent
= dir
? readdir(dir
) : NULL
; ent
; ent
= readdir(dir
)) {
1176 if (sscanf(ent
->d_name
, "%d:%d", &major
, &minor
) != 2)
1178 path
= devt_to_devpath(makedev(major
, minor
));
1181 if (!path_attached_to_hba(path
, hba_path
)) {
1187 /* retrieve the scsi device type */
1188 if (asprintf(&device
, "/sys/dev/block/%d:%d/device/xxxxxxx", major
, minor
) < 0) {
1190 fprintf(stderr
, Name
": failed to allocate 'device'\n");
1194 sprintf(device
, "/sys/dev/block/%d:%d/device/type", major
, minor
);
1195 if (load_sys(device
, buf
) != 0) {
1197 fprintf(stderr
, Name
": failed to read device type for %s\n",
1203 type
= strtoul(buf
, NULL
, 10);
1205 /* if it's not a disk print the vendor and model */
1206 if (!(type
== 0 || type
== 7 || type
== 14)) {
1209 sprintf(device
, "/sys/dev/block/%d:%d/device/vendor", major
, minor
);
1210 if (load_sys(device
, buf
) == 0) {
1211 strncpy(vendor
, buf
, sizeof(vendor
));
1212 vendor
[sizeof(vendor
) - 1] = '\0';
1213 c
= (char *) &vendor
[sizeof(vendor
) - 1];
1214 while (isspace(*c
) || *c
== '\0')
1218 sprintf(device
, "/sys/dev/block/%d:%d/device/model", major
, minor
);
1219 if (load_sys(device
, buf
) == 0) {
1220 strncpy(model
, buf
, sizeof(model
));
1221 model
[sizeof(model
) - 1] = '\0';
1222 c
= (char *) &model
[sizeof(model
) - 1];
1223 while (isspace(*c
) || *c
== '\0')
1227 if (vendor
[0] && model
[0])
1228 sprintf(buf
, "%.64s %.64s", vendor
, model
);
1230 switch (type
) { /* numbers from hald/linux/device.c */
1231 case 1: sprintf(buf
, "tape"); break;
1232 case 2: sprintf(buf
, "printer"); break;
1233 case 3: sprintf(buf
, "processor"); break;
1235 case 5: sprintf(buf
, "cdrom"); break;
1236 case 6: sprintf(buf
, "scanner"); break;
1237 case 8: sprintf(buf
, "media_changer"); break;
1238 case 9: sprintf(buf
, "comm"); break;
1239 case 12: sprintf(buf
, "raid"); break;
1240 default: sprintf(buf
, "unknown");
1246 /* chop device path to 'host%d' and calculate the port number */
1247 c
= strchr(&path
[hba_len
], '/');
1250 fprintf(stderr
, Name
": %s - invalid path name\n", path
+ hba_len
);
1255 if (sscanf(&path
[hba_len
], "host%d", &port
) == 1)
1259 *c
= '/'; /* repair the full string */
1260 fprintf(stderr
, Name
": failed to determine port number for %s\n",
1267 /* mark this port as used */
1268 port_mask
&= ~(1 << port
);
1270 /* print out the device information */
1272 printf(" Port%d : - non-disk device (%s) -\n", port
, buf
);
1276 fd
= dev_open(ent
->d_name
, O_RDONLY
);
1278 printf(" Port%d : - disk info unavailable -\n", port
);
1280 fd2devname(fd
, buf
);
1281 printf(" Port%d : %s", port
, buf
);
1282 if (imsm_read_serial(fd
, NULL
, (__u8
*) buf
) == 0)
1283 printf(" (%s)\n", buf
);
1298 for (i
= 0; i
< port_count
; i
++)
1299 if (port_mask
& (1 << i
))
1300 printf(" Port%d : - no device attached -\n", i
);
1308 static void print_found_intel_controllers(struct sys_dev
*elem
)
1310 for (; elem
; elem
= elem
->next
) {
1311 fprintf(stderr
, Name
": found Intel(R) ");
1312 if (elem
->type
== SYS_DEV_SATA
)
1313 fprintf(stderr
, "SATA ");
1314 else if (elem
->type
== SYS_DEV_SAS
)
1315 fprintf(stderr
, "SAS ");
1316 fprintf(stderr
, "RAID controller");
1318 fprintf(stderr
, " at %s", elem
->pci_id
);
1319 fprintf(stderr
, ".\n");
1324 static int ahci_get_port_count(const char *hba_path
, int *port_count
)
1331 if ((dir
= opendir(hba_path
)) == NULL
)
1334 for (ent
= readdir(dir
); ent
; ent
= readdir(dir
)) {
1337 if (sscanf(ent
->d_name
, "host%d", &host
) != 1)
1339 if (*port_count
== 0)
1341 else if (host
< host_base
)
1344 if (host
+ 1 > *port_count
+ host_base
)
1345 *port_count
= host
+ 1 - host_base
;
1351 static void print_imsm_capability(const struct imsm_orom
*orom
)
1353 printf(" Platform : Intel(R) Matrix Storage Manager\n");
1354 printf(" Version : %d.%d.%d.%d\n", orom
->major_ver
, orom
->minor_ver
,
1355 orom
->hotfix_ver
, orom
->build
);
1356 printf(" RAID Levels :%s%s%s%s%s\n",
1357 imsm_orom_has_raid0(orom
) ? " raid0" : "",
1358 imsm_orom_has_raid1(orom
) ? " raid1" : "",
1359 imsm_orom_has_raid1e(orom
) ? " raid1e" : "",
1360 imsm_orom_has_raid10(orom
) ? " raid10" : "",
1361 imsm_orom_has_raid5(orom
) ? " raid5" : "");
1362 printf(" Chunk Sizes :%s%s%s%s%s%s%s%s%s%s%s%s%s%s%s%s\n",
1363 imsm_orom_has_chunk(orom
, 2) ? " 2k" : "",
1364 imsm_orom_has_chunk(orom
, 4) ? " 4k" : "",
1365 imsm_orom_has_chunk(orom
, 8) ? " 8k" : "",
1366 imsm_orom_has_chunk(orom
, 16) ? " 16k" : "",
1367 imsm_orom_has_chunk(orom
, 32) ? " 32k" : "",
1368 imsm_orom_has_chunk(orom
, 64) ? " 64k" : "",
1369 imsm_orom_has_chunk(orom
, 128) ? " 128k" : "",
1370 imsm_orom_has_chunk(orom
, 256) ? " 256k" : "",
1371 imsm_orom_has_chunk(orom
, 512) ? " 512k" : "",
1372 imsm_orom_has_chunk(orom
, 1024*1) ? " 1M" : "",
1373 imsm_orom_has_chunk(orom
, 1024*2) ? " 2M" : "",
1374 imsm_orom_has_chunk(orom
, 1024*4) ? " 4M" : "",
1375 imsm_orom_has_chunk(orom
, 1024*8) ? " 8M" : "",
1376 imsm_orom_has_chunk(orom
, 1024*16) ? " 16M" : "",
1377 imsm_orom_has_chunk(orom
, 1024*32) ? " 32M" : "",
1378 imsm_orom_has_chunk(orom
, 1024*64) ? " 64M" : "");
1379 printf(" Max Disks : %d\n", orom
->tds
);
1380 printf(" Max Volumes : %d\n", orom
->vpa
);
1384 static int detail_platform_imsm(int verbose
, int enumerate_only
)
1386 /* There are two components to imsm platform support, the ahci SATA
1387 * controller and the option-rom. To find the SATA controller we
1388 * simply look in /sys/bus/pci/drivers/ahci to see if an ahci
1389 * controller with the Intel vendor id is present. This approach
1390 * allows mdadm to leverage the kernel's ahci detection logic, with the
1391 * caveat that if ahci.ko is not loaded mdadm will not be able to
1392 * detect platform raid capabilities. The option-rom resides in a
1393 * platform "Adapter ROM". We scan for its signature to retrieve the
1394 * platform capabilities. If raid support is disabled in the BIOS the
1395 * option-rom capability structure will not be available.
1397 const struct imsm_orom
*orom
;
1398 struct sys_dev
*list
, *hba
;
1403 if (enumerate_only
) {
1404 if (check_env("IMSM_NO_PLATFORM"))
1406 list
= find_intel_devices();
1409 for (hba
= list
; hba
; hba
= hba
->next
) {
1410 orom
= find_imsm_capability(hba
->type
);
1416 free_sys_dev(&list
);
1420 list
= find_intel_devices();
1423 fprintf(stderr
, Name
": no active Intel(R) RAID "
1424 "controller found.\n");
1425 free_sys_dev(&list
);
1428 print_found_intel_controllers(list
);
1430 for (hba
= list
; hba
; hba
= hba
->next
) {
1431 orom
= find_imsm_capability(hba
->type
);
1433 fprintf(stderr
, Name
": imsm capabilities not found for controller: %s (type %s)\n",
1434 hba
->path
, get_sys_dev_type(hba
->type
));
1436 print_imsm_capability(orom
);
1439 for (hba
= list
; hba
; hba
= hba
->next
) {
1440 printf(" I/O Controller : %s (%s)\n",
1441 hba
->path
, get_sys_dev_type(hba
->type
));
1443 if (hba
->type
== SYS_DEV_SATA
) {
1444 host_base
= ahci_get_port_count(hba
->path
, &port_count
);
1445 if (ahci_enumerate_ports(hba
->path
, port_count
, host_base
, verbose
)) {
1447 fprintf(stderr
, Name
": failed to enumerate "
1448 "ports on SATA controller at %s.", hba
->pci_id
);
1454 free_sys_dev(&list
);
1459 static int match_home_imsm(struct supertype
*st
, char *homehost
)
1461 /* the imsm metadata format does not specify any host
1462 * identification information. We return -1 since we can never
1463 * confirm nor deny whether a given array is "meant" for this
1464 * host. We rely on compare_super and the 'family_num' fields to
1465 * exclude member disks that do not belong, and we rely on
1466 * mdadm.conf to specify the arrays that should be assembled.
1467 * Auto-assembly may still pick up "foreign" arrays.
1473 static void uuid_from_super_imsm(struct supertype
*st
, int uuid
[4])
1475 /* The uuid returned here is used for:
1476 * uuid to put into bitmap file (Create, Grow)
1477 * uuid for backup header when saving critical section (Grow)
1478 * comparing uuids when re-adding a device into an array
1479 * In these cases the uuid required is that of the data-array,
1480 * not the device-set.
1481 * uuid to recognise same set when adding a missing device back
1482 * to an array. This is a uuid for the device-set.
1484 * For each of these we can make do with a truncated
1485 * or hashed uuid rather than the original, as long as
1487 * In each case the uuid required is that of the data-array,
1488 * not the device-set.
1490 /* imsm does not track uuid's so we synthesis one using sha1 on
1491 * - The signature (Which is constant for all imsm array, but no matter)
1492 * - the orig_family_num of the container
1493 * - the index number of the volume
1494 * - the 'serial' number of the volume.
1495 * Hopefully these are all constant.
1497 struct intel_super
*super
= st
->sb
;
1500 struct sha1_ctx ctx
;
1501 struct imsm_dev
*dev
= NULL
;
1504 /* some mdadm versions failed to set ->orig_family_num, in which
1505 * case fall back to ->family_num. orig_family_num will be
1506 * fixed up with the first metadata update.
1508 family_num
= super
->anchor
->orig_family_num
;
1509 if (family_num
== 0)
1510 family_num
= super
->anchor
->family_num
;
1511 sha1_init_ctx(&ctx
);
1512 sha1_process_bytes(super
->anchor
->sig
, MPB_SIG_LEN
, &ctx
);
1513 sha1_process_bytes(&family_num
, sizeof(__u32
), &ctx
);
1514 if (super
->current_vol
>= 0)
1515 dev
= get_imsm_dev(super
, super
->current_vol
);
1517 __u32 vol
= super
->current_vol
;
1518 sha1_process_bytes(&vol
, sizeof(vol
), &ctx
);
1519 sha1_process_bytes(dev
->volume
, MAX_RAID_SERIAL_LEN
, &ctx
);
1521 sha1_finish_ctx(&ctx
, buf
);
1522 memcpy(uuid
, buf
, 4*4);
1527 get_imsm_numerical_version(struct imsm_super
*mpb
, int *m
, int *p
)
1529 __u8
*v
= get_imsm_version(mpb
);
1530 __u8
*end
= mpb
->sig
+ MAX_SIGNATURE_LENGTH
;
1531 char major
[] = { 0, 0, 0 };
1532 char minor
[] = { 0 ,0, 0 };
1533 char patch
[] = { 0, 0, 0 };
1534 char *ver_parse
[] = { major
, minor
, patch
};
1538 while (*v
!= '\0' && v
< end
) {
1539 if (*v
!= '.' && j
< 2)
1540 ver_parse
[i
][j
++] = *v
;
1548 *m
= strtol(minor
, NULL
, 0);
1549 *p
= strtol(patch
, NULL
, 0);
1553 static __u32
migr_strip_blocks_resync(struct imsm_dev
*dev
)
1555 /* migr_strip_size when repairing or initializing parity */
1556 struct imsm_map
*map
= get_imsm_map(dev
, 0);
1557 __u32 chunk
= __le32_to_cpu(map
->blocks_per_strip
);
1559 switch (get_imsm_raid_level(map
)) {
1564 return 128*1024 >> 9;
1568 static __u32
migr_strip_blocks_rebuild(struct imsm_dev
*dev
)
1570 /* migr_strip_size when rebuilding a degraded disk, no idea why
1571 * this is different than migr_strip_size_resync(), but it's good
1574 struct imsm_map
*map
= get_imsm_map(dev
, 1);
1575 __u32 chunk
= __le32_to_cpu(map
->blocks_per_strip
);
1577 switch (get_imsm_raid_level(map
)) {
1580 if (map
->num_members
% map
->num_domains
== 0)
1581 return 128*1024 >> 9;
1585 return max((__u32
) 64*1024 >> 9, chunk
);
1587 return 128*1024 >> 9;
1591 static __u32
num_stripes_per_unit_resync(struct imsm_dev
*dev
)
1593 struct imsm_map
*lo
= get_imsm_map(dev
, 0);
1594 struct imsm_map
*hi
= get_imsm_map(dev
, 1);
1595 __u32 lo_chunk
= __le32_to_cpu(lo
->blocks_per_strip
);
1596 __u32 hi_chunk
= __le32_to_cpu(hi
->blocks_per_strip
);
1598 return max((__u32
) 1, hi_chunk
/ lo_chunk
);
1601 static __u32
num_stripes_per_unit_rebuild(struct imsm_dev
*dev
)
1603 struct imsm_map
*lo
= get_imsm_map(dev
, 0);
1604 int level
= get_imsm_raid_level(lo
);
1606 if (level
== 1 || level
== 10) {
1607 struct imsm_map
*hi
= get_imsm_map(dev
, 1);
1609 return hi
->num_domains
;
1611 return num_stripes_per_unit_resync(dev
);
1614 static __u8
imsm_num_data_members(struct imsm_dev
*dev
, int second_map
)
1616 /* named 'imsm_' because raid0, raid1 and raid10
1617 * counter-intuitively have the same number of data disks
1619 struct imsm_map
*map
= get_imsm_map(dev
, second_map
);
1621 switch (get_imsm_raid_level(map
)) {
1625 return map
->num_members
;
1627 return map
->num_members
- 1;
1629 dprintf("%s: unsupported raid level\n", __func__
);
1634 static __u32
parity_segment_depth(struct imsm_dev
*dev
)
1636 struct imsm_map
*map
= get_imsm_map(dev
, 0);
1637 __u32 chunk
= __le32_to_cpu(map
->blocks_per_strip
);
1639 switch(get_imsm_raid_level(map
)) {
1642 return chunk
* map
->num_domains
;
1644 return chunk
* map
->num_members
;
1650 static __u32
map_migr_block(struct imsm_dev
*dev
, __u32 block
)
1652 struct imsm_map
*map
= get_imsm_map(dev
, 1);
1653 __u32 chunk
= __le32_to_cpu(map
->blocks_per_strip
);
1654 __u32 strip
= block
/ chunk
;
1656 switch (get_imsm_raid_level(map
)) {
1659 __u32 vol_strip
= (strip
* map
->num_domains
) + 1;
1660 __u32 vol_stripe
= vol_strip
/ map
->num_members
;
1662 return vol_stripe
* chunk
+ block
% chunk
;
1664 __u32 stripe
= strip
/ (map
->num_members
- 1);
1666 return stripe
* chunk
+ block
% chunk
;
1673 static __u64
blocks_per_migr_unit(struct imsm_dev
*dev
)
1675 /* calculate the conversion factor between per member 'blocks'
1676 * (md/{resync,rebuild}_start) and imsm migration units, return
1677 * 0 for the 'not migrating' and 'unsupported migration' cases
1679 if (!dev
->vol
.migr_state
)
1682 switch (migr_type(dev
)) {
1687 struct imsm_map
*map
= get_imsm_map(dev
, 0);
1688 __u32 stripes_per_unit
;
1689 __u32 blocks_per_unit
;
1698 /* yes, this is really the translation of migr_units to
1699 * per-member blocks in the 'resync' case
1701 stripes_per_unit
= num_stripes_per_unit_resync(dev
);
1702 migr_chunk
= migr_strip_blocks_resync(dev
);
1703 disks
= imsm_num_data_members(dev
, 0);
1704 blocks_per_unit
= stripes_per_unit
* migr_chunk
* disks
;
1705 if (migr_type(dev
) == MIGR_GEN_MIGR
)
1706 return blocks_per_unit
;
1707 stripe
= __le32_to_cpu(map
->blocks_per_strip
) * disks
;
1708 segment
= blocks_per_unit
/ stripe
;
1709 block_rel
= blocks_per_unit
- segment
* stripe
;
1710 parity_depth
= parity_segment_depth(dev
);
1711 block_map
= map_migr_block(dev
, block_rel
);
1712 return block_map
+ parity_depth
* segment
;
1714 case MIGR_REBUILD
: {
1715 __u32 stripes_per_unit
;
1718 stripes_per_unit
= num_stripes_per_unit_rebuild(dev
);
1719 migr_chunk
= migr_strip_blocks_rebuild(dev
);
1720 return migr_chunk
* stripes_per_unit
;
1722 case MIGR_STATE_CHANGE
:
1728 static int imsm_level_to_layout(int level
)
1736 return ALGORITHM_LEFT_ASYMMETRIC
;
1743 static void getinfo_super_imsm_volume(struct supertype
*st
, struct mdinfo
*info
, char *dmap
)
1745 struct intel_super
*super
= st
->sb
;
1746 struct imsm_dev
*dev
= get_imsm_dev(super
, super
->current_vol
);
1747 struct imsm_map
*map
= get_imsm_map(dev
, 0);
1748 struct imsm_map
*prev_map
= get_imsm_map(dev
, 1);
1749 struct imsm_map
*map_to_analyse
= map
;
1752 int map_disks
= info
->array
.raid_disks
;
1755 map_to_analyse
= prev_map
;
1757 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
1758 if (dl
->raiddisk
== info
->disk
.raid_disk
)
1760 info
->container_member
= super
->current_vol
;
1761 info
->array
.raid_disks
= map_to_analyse
->num_members
;
1762 info
->array
.level
= get_imsm_raid_level(map_to_analyse
);
1763 info
->array
.layout
= imsm_level_to_layout(info
->array
.level
);
1764 info
->array
.md_minor
= -1;
1765 info
->array
.ctime
= 0;
1766 info
->array
.utime
= 0;
1767 info
->array
.chunk_size
=
1768 __le16_to_cpu(map_to_analyse
->blocks_per_strip
) << 9;
1769 info
->array
.state
= !dev
->vol
.dirty
;
1770 info
->custom_array_size
= __le32_to_cpu(dev
->size_high
);
1771 info
->custom_array_size
<<= 32;
1772 info
->custom_array_size
|= __le32_to_cpu(dev
->size_low
);
1773 if (prev_map
&& map
->map_state
== prev_map
->map_state
) {
1774 info
->reshape_active
= 1;
1775 info
->new_level
= get_imsm_raid_level(map
);
1776 info
->new_layout
= imsm_level_to_layout(info
->new_level
);
1777 info
->new_chunk
= __le16_to_cpu(map
->blocks_per_strip
) << 9;
1778 info
->delta_disks
= map
->num_members
- prev_map
->num_members
;
1779 if (info
->delta_disks
) {
1780 /* this needs to be applied to every array
1783 info
->reshape_active
= 2;
1785 /* We shape information that we give to md might have to be
1786 * modify to cope with md's requirement for reshaping arrays.
1787 * For example, when reshaping a RAID0, md requires it to be
1788 * presented as a degraded RAID4.
1789 * Also if a RAID0 is migrating to a RAID5 we need to specify
1790 * the array as already being RAID5, but the 'before' layout
1791 * is a RAID4-like layout.
1793 switch (info
->array
.level
) {
1795 switch(info
->new_level
) {
1797 /* conversion is happening as RAID4 */
1798 info
->array
.level
= 4;
1799 info
->array
.raid_disks
+= 1;
1802 /* conversion is happening as RAID5 */
1803 info
->array
.level
= 5;
1804 info
->array
.layout
= ALGORITHM_PARITY_N
;
1805 info
->array
.raid_disks
+= 1;
1806 info
->delta_disks
-= 1;
1809 /* FIXME error message */
1810 info
->array
.level
= UnSet
;
1816 info
->new_level
= UnSet
;
1817 info
->new_layout
= UnSet
;
1818 info
->new_chunk
= info
->array
.chunk_size
;
1819 info
->delta_disks
= 0;
1821 info
->disk
.major
= 0;
1822 info
->disk
.minor
= 0;
1824 info
->disk
.major
= dl
->major
;
1825 info
->disk
.minor
= dl
->minor
;
1828 info
->data_offset
= __le32_to_cpu(map_to_analyse
->pba_of_lba0
);
1829 info
->component_size
=
1830 __le32_to_cpu(map_to_analyse
->blocks_per_member
);
1831 memset(info
->uuid
, 0, sizeof(info
->uuid
));
1832 info
->recovery_start
= MaxSector
;
1834 info
->reshape_progress
= 0;
1835 info
->resync_start
= MaxSector
;
1836 if (map_to_analyse
->map_state
== IMSM_T_STATE_UNINITIALIZED
||
1838 info
->resync_start
= 0;
1840 if (dev
->vol
.migr_state
) {
1841 switch (migr_type(dev
)) {
1844 __u64 blocks_per_unit
= blocks_per_migr_unit(dev
);
1845 __u64 units
= __le32_to_cpu(dev
->vol
.curr_migr_unit
);
1847 info
->resync_start
= blocks_per_unit
* units
;
1850 case MIGR_GEN_MIGR
: {
1851 __u64 blocks_per_unit
= blocks_per_migr_unit(dev
);
1852 __u64 units
= __le32_to_cpu(dev
->vol
.curr_migr_unit
);
1853 unsigned long long array_blocks
;
1856 info
->reshape_progress
= blocks_per_unit
* units
;
1857 dprintf("IMSM: General Migration checkpoint : %llu "
1858 "(%llu) -> read reshape progress : %llu\n",
1859 units
, blocks_per_unit
, info
->reshape_progress
);
1861 used_disks
= imsm_num_data_members(dev
, 1);
1862 if (used_disks
> 0) {
1863 array_blocks
= map
->blocks_per_member
*
1865 /* round array size down to closest MB
1867 info
->custom_array_size
= (array_blocks
1868 >> SECT_PER_MB_SHIFT
)
1869 << SECT_PER_MB_SHIFT
;
1873 /* we could emulate the checkpointing of
1874 * 'sync_action=check' migrations, but for now
1875 * we just immediately complete them
1878 /* this is handled by container_content_imsm() */
1879 case MIGR_STATE_CHANGE
:
1880 /* FIXME handle other migrations */
1882 /* we are not dirty, so... */
1883 info
->resync_start
= MaxSector
;
1887 strncpy(info
->name
, (char *) dev
->volume
, MAX_RAID_SERIAL_LEN
);
1888 info
->name
[MAX_RAID_SERIAL_LEN
] = 0;
1890 info
->array
.major_version
= -1;
1891 info
->array
.minor_version
= -2;
1892 devname
= devnum2devname(st
->container_dev
);
1893 *info
->text_version
= '\0';
1895 sprintf(info
->text_version
, "/%s/%d", devname
, info
->container_member
);
1897 info
->safe_mode_delay
= 4000; /* 4 secs like the Matrix driver */
1898 uuid_from_super_imsm(st
, info
->uuid
);
1902 for (i
=0; i
<map_disks
; i
++) {
1904 if (i
< info
->array
.raid_disks
) {
1905 struct imsm_disk
*dsk
;
1906 j
= get_imsm_disk_idx(dev
, i
, -1);
1907 dsk
= get_imsm_disk(super
, j
);
1908 if (dsk
&& (dsk
->status
& CONFIGURED_DISK
))
1915 static __u8
imsm_check_degraded(struct intel_super
*super
, struct imsm_dev
*dev
, int failed
);
1916 static int imsm_count_failed(struct intel_super
*super
, struct imsm_dev
*dev
);
1918 static struct imsm_disk
*get_imsm_missing(struct intel_super
*super
, __u8 index
)
1922 for (d
= super
->missing
; d
; d
= d
->next
)
1923 if (d
->index
== index
)
1928 static void getinfo_super_imsm(struct supertype
*st
, struct mdinfo
*info
, char *map
)
1930 struct intel_super
*super
= st
->sb
;
1931 struct imsm_disk
*disk
;
1932 int map_disks
= info
->array
.raid_disks
;
1933 int max_enough
= -1;
1935 struct imsm_super
*mpb
;
1937 if (super
->current_vol
>= 0) {
1938 getinfo_super_imsm_volume(st
, info
, map
);
1942 /* Set raid_disks to zero so that Assemble will always pull in valid
1945 info
->array
.raid_disks
= 0;
1946 info
->array
.level
= LEVEL_CONTAINER
;
1947 info
->array
.layout
= 0;
1948 info
->array
.md_minor
= -1;
1949 info
->array
.ctime
= 0; /* N/A for imsm */
1950 info
->array
.utime
= 0;
1951 info
->array
.chunk_size
= 0;
1953 info
->disk
.major
= 0;
1954 info
->disk
.minor
= 0;
1955 info
->disk
.raid_disk
= -1;
1956 info
->reshape_active
= 0;
1957 info
->array
.major_version
= -1;
1958 info
->array
.minor_version
= -2;
1959 strcpy(info
->text_version
, "imsm");
1960 info
->safe_mode_delay
= 0;
1961 info
->disk
.number
= -1;
1962 info
->disk
.state
= 0;
1964 info
->recovery_start
= MaxSector
;
1966 /* do we have the all the insync disks that we expect? */
1967 mpb
= super
->anchor
;
1969 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
1970 struct imsm_dev
*dev
= get_imsm_dev(super
, i
);
1971 int failed
, enough
, j
, missing
= 0;
1972 struct imsm_map
*map
;
1975 failed
= imsm_count_failed(super
, dev
);
1976 state
= imsm_check_degraded(super
, dev
, failed
);
1977 map
= get_imsm_map(dev
, dev
->vol
.migr_state
);
1979 /* any newly missing disks?
1980 * (catches single-degraded vs double-degraded)
1982 for (j
= 0; j
< map
->num_members
; j
++) {
1983 __u32 ord
= get_imsm_ord_tbl_ent(dev
, i
, -1);
1984 __u32 idx
= ord_to_idx(ord
);
1986 if (!(ord
& IMSM_ORD_REBUILD
) &&
1987 get_imsm_missing(super
, idx
)) {
1993 if (state
== IMSM_T_STATE_FAILED
)
1995 else if (state
== IMSM_T_STATE_DEGRADED
&&
1996 (state
!= map
->map_state
|| missing
))
1998 else /* we're normal, or already degraded */
2001 /* in the missing/failed disk case check to see
2002 * if at least one array is runnable
2004 max_enough
= max(max_enough
, enough
);
2006 dprintf("%s: enough: %d\n", __func__
, max_enough
);
2007 info
->container_enough
= max_enough
;
2010 __u32 reserved
= imsm_reserved_sectors(super
, super
->disks
);
2012 disk
= &super
->disks
->disk
;
2013 info
->data_offset
= __le32_to_cpu(disk
->total_blocks
) - reserved
;
2014 info
->component_size
= reserved
;
2015 info
->disk
.state
= is_configured(disk
) ? (1 << MD_DISK_ACTIVE
) : 0;
2016 /* we don't change info->disk.raid_disk here because
2017 * this state will be finalized in mdmon after we have
2018 * found the 'most fresh' version of the metadata
2020 info
->disk
.state
|= is_failed(disk
) ? (1 << MD_DISK_FAULTY
) : 0;
2021 info
->disk
.state
|= is_spare(disk
) ? 0 : (1 << MD_DISK_SYNC
);
2024 /* only call uuid_from_super_imsm when this disk is part of a populated container,
2025 * ->compare_super may have updated the 'num_raid_devs' field for spares
2027 if (info
->disk
.state
& (1 << MD_DISK_SYNC
) || super
->anchor
->num_raid_devs
)
2028 uuid_from_super_imsm(st
, info
->uuid
);
2030 memcpy(info
->uuid
, uuid_zero
, sizeof(uuid_zero
));
2032 /* I don't know how to compute 'map' on imsm, so use safe default */
2035 for (i
= 0; i
< map_disks
; i
++)
2041 /* allocates memory and fills disk in mdinfo structure
2042 * for each disk in array */
2043 struct mdinfo
*getinfo_super_disks_imsm(struct supertype
*st
)
2045 struct mdinfo
*mddev
= NULL
;
2046 struct intel_super
*super
= st
->sb
;
2047 struct imsm_disk
*disk
;
2050 if (!super
|| !super
->disks
)
2053 mddev
= malloc(sizeof(*mddev
));
2055 fprintf(stderr
, Name
": Failed to allocate memory.\n");
2058 memset(mddev
, 0, sizeof(*mddev
));
2062 tmp
= malloc(sizeof(*tmp
));
2064 fprintf(stderr
, Name
": Failed to allocate memory.\n");
2069 memset(tmp
, 0, sizeof(*tmp
));
2071 tmp
->next
= mddev
->devs
;
2073 tmp
->disk
.number
= count
++;
2074 tmp
->disk
.major
= dl
->major
;
2075 tmp
->disk
.minor
= dl
->minor
;
2076 tmp
->disk
.state
= is_configured(disk
) ?
2077 (1 << MD_DISK_ACTIVE
) : 0;
2078 tmp
->disk
.state
|= is_failed(disk
) ? (1 << MD_DISK_FAULTY
) : 0;
2079 tmp
->disk
.state
|= is_spare(disk
) ? 0 : (1 << MD_DISK_SYNC
);
2080 tmp
->disk
.raid_disk
= -1;
2086 static int update_super_imsm(struct supertype
*st
, struct mdinfo
*info
,
2087 char *update
, char *devname
, int verbose
,
2088 int uuid_set
, char *homehost
)
2090 /* For 'assemble' and 'force' we need to return non-zero if any
2091 * change was made. For others, the return value is ignored.
2092 * Update options are:
2093 * force-one : This device looks a bit old but needs to be included,
2094 * update age info appropriately.
2095 * assemble: clear any 'faulty' flag to allow this device to
2097 * force-array: Array is degraded but being forced, mark it clean
2098 * if that will be needed to assemble it.
2100 * newdev: not used ????
2101 * grow: Array has gained a new device - this is currently for
2103 * resync: mark as dirty so a resync will happen.
2104 * name: update the name - preserving the homehost
2105 * uuid: Change the uuid of the array to match watch is given
2107 * Following are not relevant for this imsm:
2108 * sparc2.2 : update from old dodgey metadata
2109 * super-minor: change the preferred_minor number
2110 * summaries: update redundant counters.
2111 * homehost: update the recorded homehost
2112 * _reshape_progress: record new reshape_progress position.
2115 struct intel_super
*super
= st
->sb
;
2116 struct imsm_super
*mpb
;
2118 /* we can only update container info */
2119 if (!super
|| super
->current_vol
>= 0 || !super
->anchor
)
2122 mpb
= super
->anchor
;
2124 if (strcmp(update
, "uuid") == 0 && uuid_set
&& !info
->update_private
)
2126 else if (strcmp(update
, "uuid") == 0 && uuid_set
&& info
->update_private
) {
2127 mpb
->orig_family_num
= *((__u32
*) info
->update_private
);
2129 } else if (strcmp(update
, "uuid") == 0) {
2130 __u32
*new_family
= malloc(sizeof(*new_family
));
2132 /* update orig_family_number with the incoming random
2133 * data, report the new effective uuid, and store the
2134 * new orig_family_num for future updates.
2137 memcpy(&mpb
->orig_family_num
, info
->uuid
, sizeof(__u32
));
2138 uuid_from_super_imsm(st
, info
->uuid
);
2139 *new_family
= mpb
->orig_family_num
;
2140 info
->update_private
= new_family
;
2143 } else if (strcmp(update
, "assemble") == 0)
2148 /* successful update? recompute checksum */
2150 mpb
->check_sum
= __le32_to_cpu(__gen_imsm_checksum(mpb
));
2155 static size_t disks_to_mpb_size(int disks
)
2159 size
= sizeof(struct imsm_super
);
2160 size
+= (disks
- 1) * sizeof(struct imsm_disk
);
2161 size
+= 2 * sizeof(struct imsm_dev
);
2162 /* up to 2 maps per raid device (-2 for imsm_maps in imsm_dev */
2163 size
+= (4 - 2) * sizeof(struct imsm_map
);
2164 /* 4 possible disk_ord_tbl's */
2165 size
+= 4 * (disks
- 1) * sizeof(__u32
);
2170 static __u64
avail_size_imsm(struct supertype
*st
, __u64 devsize
)
2172 if (devsize
< (MPB_SECTOR_CNT
+ IMSM_RESERVED_SECTORS
))
2175 return devsize
- (MPB_SECTOR_CNT
+ IMSM_RESERVED_SECTORS
);
2178 static void free_devlist(struct intel_super
*super
)
2180 struct intel_dev
*dv
;
2182 while (super
->devlist
) {
2183 dv
= super
->devlist
->next
;
2184 free(super
->devlist
->dev
);
2185 free(super
->devlist
);
2186 super
->devlist
= dv
;
2190 static void imsm_copy_dev(struct imsm_dev
*dest
, struct imsm_dev
*src
)
2192 memcpy(dest
, src
, sizeof_imsm_dev(src
, 0));
2195 static int compare_super_imsm(struct supertype
*st
, struct supertype
*tst
)
2199 * 0 same, or first was empty, and second was copied
2200 * 1 second had wrong number
2202 * 3 wrong other info
2204 struct intel_super
*first
= st
->sb
;
2205 struct intel_super
*sec
= tst
->sb
;
2212 /* in platform dependent environment test if the disks
2213 * use the same Intel hba
2215 if (!check_env("IMSM_NO_PLATFORM")) {
2216 if (first
->hba
->type
!= sec
->hba
->type
) {
2218 "HBAs of devices does not match %s != %s\n",
2219 get_sys_dev_type(first
->hba
->type
),
2220 get_sys_dev_type(sec
->hba
->type
));
2225 /* if an anchor does not have num_raid_devs set then it is a free
2228 if (first
->anchor
->num_raid_devs
> 0 &&
2229 sec
->anchor
->num_raid_devs
> 0) {
2230 /* Determine if these disks might ever have been
2231 * related. Further disambiguation can only take place
2232 * in load_super_imsm_all
2234 __u32 first_family
= first
->anchor
->orig_family_num
;
2235 __u32 sec_family
= sec
->anchor
->orig_family_num
;
2237 if (memcmp(first
->anchor
->sig
, sec
->anchor
->sig
,
2238 MAX_SIGNATURE_LENGTH
) != 0)
2241 if (first_family
== 0)
2242 first_family
= first
->anchor
->family_num
;
2243 if (sec_family
== 0)
2244 sec_family
= sec
->anchor
->family_num
;
2246 if (first_family
!= sec_family
)
2252 /* if 'first' is a spare promote it to a populated mpb with sec's
2255 if (first
->anchor
->num_raid_devs
== 0 &&
2256 sec
->anchor
->num_raid_devs
> 0) {
2258 struct intel_dev
*dv
;
2259 struct imsm_dev
*dev
;
2261 /* we need to copy raid device info from sec if an allocation
2262 * fails here we don't associate the spare
2264 for (i
= 0; i
< sec
->anchor
->num_raid_devs
; i
++) {
2265 dv
= malloc(sizeof(*dv
));
2268 dev
= malloc(sizeof_imsm_dev(get_imsm_dev(sec
, i
), 1));
2275 dv
->next
= first
->devlist
;
2276 first
->devlist
= dv
;
2278 if (i
< sec
->anchor
->num_raid_devs
) {
2279 /* allocation failure */
2280 free_devlist(first
);
2281 fprintf(stderr
, "imsm: failed to associate spare\n");
2284 first
->anchor
->num_raid_devs
= sec
->anchor
->num_raid_devs
;
2285 first
->anchor
->orig_family_num
= sec
->anchor
->orig_family_num
;
2286 first
->anchor
->family_num
= sec
->anchor
->family_num
;
2287 memcpy(first
->anchor
->sig
, sec
->anchor
->sig
, MAX_SIGNATURE_LENGTH
);
2288 for (i
= 0; i
< sec
->anchor
->num_raid_devs
; i
++)
2289 imsm_copy_dev(get_imsm_dev(first
, i
), get_imsm_dev(sec
, i
));
2295 static void fd2devname(int fd
, char *name
)
2299 char dname
[PATH_MAX
];
2304 if (fstat(fd
, &st
) != 0)
2306 sprintf(path
, "/sys/dev/block/%d:%d",
2307 major(st
.st_rdev
), minor(st
.st_rdev
));
2309 rv
= readlink(path
, dname
, sizeof(dname
));
2314 nm
= strrchr(dname
, '/');
2316 snprintf(name
, MAX_RAID_SERIAL_LEN
, "/dev/%s", nm
);
2319 extern int scsi_get_serial(int fd
, void *buf
, size_t buf_len
);
2321 static int imsm_read_serial(int fd
, char *devname
,
2322 __u8 serial
[MAX_RAID_SERIAL_LEN
])
2324 unsigned char scsi_serial
[255];
2333 memset(scsi_serial
, 0, sizeof(scsi_serial
));
2335 rv
= scsi_get_serial(fd
, scsi_serial
, sizeof(scsi_serial
));
2337 if (rv
&& check_env("IMSM_DEVNAME_AS_SERIAL")) {
2338 memset(serial
, 0, MAX_RAID_SERIAL_LEN
);
2339 fd2devname(fd
, (char *) serial
);
2346 Name
": Failed to retrieve serial for %s\n",
2351 rsp_len
= scsi_serial
[3];
2355 Name
": Failed to retrieve serial for %s\n",
2359 rsp_buf
= (char *) &scsi_serial
[4];
2361 /* trim all whitespace and non-printable characters and convert
2364 for (i
= 0, dest
= rsp_buf
; i
< rsp_len
; i
++) {
2367 /* ':' is reserved for use in placeholder serial
2368 * numbers for missing disks
2376 len
= dest
- rsp_buf
;
2379 /* truncate leading characters */
2380 if (len
> MAX_RAID_SERIAL_LEN
) {
2381 dest
+= len
- MAX_RAID_SERIAL_LEN
;
2382 len
= MAX_RAID_SERIAL_LEN
;
2385 memset(serial
, 0, MAX_RAID_SERIAL_LEN
);
2386 memcpy(serial
, dest
, len
);
2391 static int serialcmp(__u8
*s1
, __u8
*s2
)
2393 return strncmp((char *) s1
, (char *) s2
, MAX_RAID_SERIAL_LEN
);
2396 static void serialcpy(__u8
*dest
, __u8
*src
)
2398 strncpy((char *) dest
, (char *) src
, MAX_RAID_SERIAL_LEN
);
2402 static struct dl
*serial_to_dl(__u8
*serial
, struct intel_super
*super
)
2406 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
2407 if (serialcmp(dl
->serial
, serial
) == 0)
2414 static struct imsm_disk
*
2415 __serial_to_disk(__u8
*serial
, struct imsm_super
*mpb
, int *idx
)
2419 for (i
= 0; i
< mpb
->num_disks
; i
++) {
2420 struct imsm_disk
*disk
= __get_imsm_disk(mpb
, i
);
2422 if (serialcmp(disk
->serial
, serial
) == 0) {
2433 load_imsm_disk(int fd
, struct intel_super
*super
, char *devname
, int keep_fd
)
2435 struct imsm_disk
*disk
;
2440 __u8 serial
[MAX_RAID_SERIAL_LEN
];
2442 rv
= imsm_read_serial(fd
, devname
, serial
);
2447 dl
= calloc(1, sizeof(*dl
));
2451 Name
": failed to allocate disk buffer for %s\n",
2457 dl
->major
= major(stb
.st_rdev
);
2458 dl
->minor
= minor(stb
.st_rdev
);
2459 dl
->next
= super
->disks
;
2460 dl
->fd
= keep_fd
? fd
: -1;
2461 assert(super
->disks
== NULL
);
2463 serialcpy(dl
->serial
, serial
);
2466 fd2devname(fd
, name
);
2468 dl
->devname
= strdup(devname
);
2470 dl
->devname
= strdup(name
);
2472 /* look up this disk's index in the current anchor */
2473 disk
= __serial_to_disk(dl
->serial
, super
->anchor
, &dl
->index
);
2476 /* only set index on disks that are a member of a
2477 * populated contianer, i.e. one with raid_devs
2479 if (is_failed(&dl
->disk
))
2481 else if (is_spare(&dl
->disk
))
2489 /* When migrating map0 contains the 'destination' state while map1
2490 * contains the current state. When not migrating map0 contains the
2491 * current state. This routine assumes that map[0].map_state is set to
2492 * the current array state before being called.
2494 * Migration is indicated by one of the following states
2495 * 1/ Idle (migr_state=0 map0state=normal||unitialized||degraded||failed)
2496 * 2/ Initialize (migr_state=1 migr_type=MIGR_INIT map0state=normal
2497 * map1state=unitialized)
2498 * 3/ Repair (Resync) (migr_state=1 migr_type=MIGR_REPAIR map0state=normal
2500 * 4/ Rebuild (migr_state=1 migr_type=MIGR_REBUILD map0state=normal
2501 * map1state=degraded)
2503 static void migrate(struct imsm_dev
*dev
, __u8 to_state
, int migr_type
)
2505 struct imsm_map
*dest
;
2506 struct imsm_map
*src
= get_imsm_map(dev
, 0);
2508 dev
->vol
.migr_state
= 1;
2509 set_migr_type(dev
, migr_type
);
2510 dev
->vol
.curr_migr_unit
= 0;
2511 dest
= get_imsm_map(dev
, 1);
2513 /* duplicate and then set the target end state in map[0] */
2514 memcpy(dest
, src
, sizeof_imsm_map(src
));
2515 if ((migr_type
== MIGR_REBUILD
) ||
2516 (migr_type
== MIGR_GEN_MIGR
)) {
2520 for (i
= 0; i
< src
->num_members
; i
++) {
2521 ord
= __le32_to_cpu(src
->disk_ord_tbl
[i
]);
2522 set_imsm_ord_tbl_ent(src
, i
, ord_to_idx(ord
));
2526 src
->map_state
= to_state
;
2529 static void end_migration(struct imsm_dev
*dev
, __u8 map_state
)
2531 struct imsm_map
*map
= get_imsm_map(dev
, 0);
2532 struct imsm_map
*prev
= get_imsm_map(dev
, dev
->vol
.migr_state
);
2535 /* merge any IMSM_ORD_REBUILD bits that were not successfully
2536 * completed in the last migration.
2538 * FIXME add support for raid-level-migration
2540 for (i
= 0; i
< prev
->num_members
; i
++)
2541 for (j
= 0; j
< map
->num_members
; j
++)
2542 /* during online capacity expansion
2543 * disks position can be changed if takeover is used
2545 if (ord_to_idx(map
->disk_ord_tbl
[j
]) ==
2546 ord_to_idx(prev
->disk_ord_tbl
[i
])) {
2547 map
->disk_ord_tbl
[j
] |= prev
->disk_ord_tbl
[i
];
2551 dev
->vol
.migr_state
= 0;
2552 dev
->vol
.migr_type
= 0;
2553 dev
->vol
.curr_migr_unit
= 0;
2554 map
->map_state
= map_state
;
2558 static int parse_raid_devices(struct intel_super
*super
)
2561 struct imsm_dev
*dev_new
;
2562 size_t len
, len_migr
;
2564 size_t space_needed
= 0;
2565 struct imsm_super
*mpb
= super
->anchor
;
2567 for (i
= 0; i
< super
->anchor
->num_raid_devs
; i
++) {
2568 struct imsm_dev
*dev_iter
= __get_imsm_dev(super
->anchor
, i
);
2569 struct intel_dev
*dv
;
2571 len
= sizeof_imsm_dev(dev_iter
, 0);
2572 len_migr
= sizeof_imsm_dev(dev_iter
, 1);
2574 space_needed
+= len_migr
- len
;
2576 dv
= malloc(sizeof(*dv
));
2579 if (max_len
< len_migr
)
2581 if (max_len
> len_migr
)
2582 space_needed
+= max_len
- len_migr
;
2583 dev_new
= malloc(max_len
);
2588 imsm_copy_dev(dev_new
, dev_iter
);
2591 dv
->next
= super
->devlist
;
2592 super
->devlist
= dv
;
2595 /* ensure that super->buf is large enough when all raid devices
2598 if (__le32_to_cpu(mpb
->mpb_size
) + space_needed
> super
->len
) {
2601 len
= ROUND_UP(__le32_to_cpu(mpb
->mpb_size
) + space_needed
, 512);
2602 if (posix_memalign(&buf
, 512, len
) != 0)
2605 memcpy(buf
, super
->buf
, super
->len
);
2606 memset(buf
+ super
->len
, 0, len
- super
->len
);
2615 /* retrieve a pointer to the bbm log which starts after all raid devices */
2616 struct bbm_log
*__get_imsm_bbm_log(struct imsm_super
*mpb
)
2620 if (__le32_to_cpu(mpb
->bbm_log_size
)) {
2622 ptr
+= mpb
->mpb_size
- __le32_to_cpu(mpb
->bbm_log_size
);
2628 static void __free_imsm(struct intel_super
*super
, int free_disks
);
2630 /* load_imsm_mpb - read matrix metadata
2631 * allocates super->mpb to be freed by free_imsm
2633 static int load_imsm_mpb(int fd
, struct intel_super
*super
, char *devname
)
2635 unsigned long long dsize
;
2636 unsigned long long sectors
;
2638 struct imsm_super
*anchor
;
2641 get_dev_size(fd
, NULL
, &dsize
);
2645 Name
": %s: device to small for imsm\n",
2650 if (lseek64(fd
, dsize
- (512 * 2), SEEK_SET
) < 0) {
2653 Name
": Cannot seek to anchor block on %s: %s\n",
2654 devname
, strerror(errno
));
2658 if (posix_memalign((void**)&anchor
, 512, 512) != 0) {
2661 Name
": Failed to allocate imsm anchor buffer"
2662 " on %s\n", devname
);
2665 if (read(fd
, anchor
, 512) != 512) {
2668 Name
": Cannot read anchor block on %s: %s\n",
2669 devname
, strerror(errno
));
2674 if (strncmp((char *) anchor
->sig
, MPB_SIGNATURE
, MPB_SIG_LEN
) != 0) {
2677 Name
": no IMSM anchor on %s\n", devname
);
2682 __free_imsm(super
, 0);
2683 /* reload capability and hba */
2685 /* capability and hba must be updated with new super allocation */
2686 find_intel_hba_capability(fd
, super
, 0);
2687 super
->len
= ROUND_UP(anchor
->mpb_size
, 512);
2688 if (posix_memalign(&super
->buf
, 512, super
->len
) != 0) {
2691 Name
": unable to allocate %zu byte mpb buffer\n",
2696 memcpy(super
->buf
, anchor
, 512);
2698 sectors
= mpb_sectors(anchor
) - 1;
2701 check_sum
= __gen_imsm_checksum(super
->anchor
);
2702 if (check_sum
!= __le32_to_cpu(super
->anchor
->check_sum
)) {
2705 Name
": IMSM checksum %x != %x on %s\n",
2707 __le32_to_cpu(super
->anchor
->check_sum
),
2715 /* read the extended mpb */
2716 if (lseek64(fd
, dsize
- (512 * (2 + sectors
)), SEEK_SET
) < 0) {
2719 Name
": Cannot seek to extended mpb on %s: %s\n",
2720 devname
, strerror(errno
));
2724 if ((unsigned)read(fd
, super
->buf
+ 512, super
->len
- 512) != super
->len
- 512) {
2727 Name
": Cannot read extended mpb on %s: %s\n",
2728 devname
, strerror(errno
));
2732 check_sum
= __gen_imsm_checksum(super
->anchor
);
2733 if (check_sum
!= __le32_to_cpu(super
->anchor
->check_sum
)) {
2736 Name
": IMSM checksum %x != %x on %s\n",
2737 check_sum
, __le32_to_cpu(super
->anchor
->check_sum
),
2742 /* FIXME the BBM log is disk specific so we cannot use this global
2743 * buffer for all disks. Ok for now since we only look at the global
2744 * bbm_log_size parameter to gate assembly
2746 super
->bbm_log
= __get_imsm_bbm_log(super
->anchor
);
2752 load_and_parse_mpb(int fd
, struct intel_super
*super
, char *devname
, int keep_fd
)
2756 err
= load_imsm_mpb(fd
, super
, devname
);
2759 err
= load_imsm_disk(fd
, super
, devname
, keep_fd
);
2762 err
= parse_raid_devices(super
);
2767 static void __free_imsm_disk(struct dl
*d
)
2779 static void free_imsm_disks(struct intel_super
*super
)
2783 while (super
->disks
) {
2785 super
->disks
= d
->next
;
2786 __free_imsm_disk(d
);
2788 while (super
->disk_mgmt_list
) {
2789 d
= super
->disk_mgmt_list
;
2790 super
->disk_mgmt_list
= d
->next
;
2791 __free_imsm_disk(d
);
2793 while (super
->missing
) {
2795 super
->missing
= d
->next
;
2796 __free_imsm_disk(d
);
2801 /* free all the pieces hanging off of a super pointer */
2802 static void __free_imsm(struct intel_super
*super
, int free_disks
)
2804 struct intel_hba
*elem
, *next
;
2810 /* unlink capability description */
2813 free_imsm_disks(super
);
2814 free_devlist(super
);
2818 free((void *)elem
->path
);
2826 static void free_imsm(struct intel_super
*super
)
2828 __free_imsm(super
, 1);
2832 static void free_super_imsm(struct supertype
*st
)
2834 struct intel_super
*super
= st
->sb
;
2843 static struct intel_super
*alloc_super(void)
2845 struct intel_super
*super
= malloc(sizeof(*super
));
2848 memset(super
, 0, sizeof(*super
));
2849 super
->current_vol
= -1;
2850 super
->create_offset
= ~((__u32
) 0);
2856 * find and allocate hba and OROM/EFI based on valid fd of RAID component device
2858 static int find_intel_hba_capability(int fd
, struct intel_super
*super
, int verbose
)
2860 struct sys_dev
*hba_name
;
2863 if ((fd
< 0) || check_env("IMSM_NO_PLATFORM")) {
2868 hba_name
= find_disk_attached_hba(fd
, NULL
);
2873 fd2devname(fd
, str
);
2875 Name
": %s is not attached to Intel(R) RAID controller.\n",
2880 rv
= attach_hba_to_super(super
, hba_name
);
2885 fd2devname(fd
, str
);
2886 fprintf(stderr
, Name
": %s is attached to Intel(R) %s RAID "
2887 "controller (%s),\n"
2888 " but the container is assigned to Intel(R) "
2889 "%s RAID controller (",
2892 hba_name
->pci_id
? : "Err!",
2893 get_sys_dev_type(hba_name
->type
));
2895 struct intel_hba
*hba
= super
->hba
;
2897 fprintf(stderr
, "%s", hba
->pci_id
? : "Err!");
2899 fprintf(stderr
, ", ");
2903 fprintf(stderr
, ").\n"
2904 " Mixing devices attached to different controllers "
2905 "is not allowed.\n");
2907 free_sys_dev(&hba_name
);
2910 super
->orom
= find_imsm_capability(hba_name
->type
);
2911 free_sys_dev(&hba_name
);
2918 /* find_missing - helper routine for load_super_imsm_all that identifies
2919 * disks that have disappeared from the system. This routine relies on
2920 * the mpb being uptodate, which it is at load time.
2922 static int find_missing(struct intel_super
*super
)
2925 struct imsm_super
*mpb
= super
->anchor
;
2927 struct imsm_disk
*disk
;
2929 for (i
= 0; i
< mpb
->num_disks
; i
++) {
2930 disk
= __get_imsm_disk(mpb
, i
);
2931 dl
= serial_to_dl(disk
->serial
, super
);
2935 dl
= malloc(sizeof(*dl
));
2941 dl
->devname
= strdup("missing");
2943 serialcpy(dl
->serial
, disk
->serial
);
2946 dl
->next
= super
->missing
;
2947 super
->missing
= dl
;
2953 static struct intel_disk
*disk_list_get(__u8
*serial
, struct intel_disk
*disk_list
)
2955 struct intel_disk
*idisk
= disk_list
;
2958 if (serialcmp(idisk
->disk
.serial
, serial
) == 0)
2960 idisk
= idisk
->next
;
2966 static int __prep_thunderdome(struct intel_super
**table
, int tbl_size
,
2967 struct intel_super
*super
,
2968 struct intel_disk
**disk_list
)
2970 struct imsm_disk
*d
= &super
->disks
->disk
;
2971 struct imsm_super
*mpb
= super
->anchor
;
2974 for (i
= 0; i
< tbl_size
; i
++) {
2975 struct imsm_super
*tbl_mpb
= table
[i
]->anchor
;
2976 struct imsm_disk
*tbl_d
= &table
[i
]->disks
->disk
;
2978 if (tbl_mpb
->family_num
== mpb
->family_num
) {
2979 if (tbl_mpb
->check_sum
== mpb
->check_sum
) {
2980 dprintf("%s: mpb from %d:%d matches %d:%d\n",
2981 __func__
, super
->disks
->major
,
2982 super
->disks
->minor
,
2983 table
[i
]->disks
->major
,
2984 table
[i
]->disks
->minor
);
2988 if (((is_configured(d
) && !is_configured(tbl_d
)) ||
2989 is_configured(d
) == is_configured(tbl_d
)) &&
2990 tbl_mpb
->generation_num
< mpb
->generation_num
) {
2991 /* current version of the mpb is a
2992 * better candidate than the one in
2993 * super_table, but copy over "cross
2994 * generational" status
2996 struct intel_disk
*idisk
;
2998 dprintf("%s: mpb from %d:%d replaces %d:%d\n",
2999 __func__
, super
->disks
->major
,
3000 super
->disks
->minor
,
3001 table
[i
]->disks
->major
,
3002 table
[i
]->disks
->minor
);
3004 idisk
= disk_list_get(tbl_d
->serial
, *disk_list
);
3005 if (idisk
&& is_failed(&idisk
->disk
))
3006 tbl_d
->status
|= FAILED_DISK
;
3009 struct intel_disk
*idisk
;
3010 struct imsm_disk
*disk
;
3012 /* tbl_mpb is more up to date, but copy
3013 * over cross generational status before
3016 disk
= __serial_to_disk(d
->serial
, mpb
, NULL
);
3017 if (disk
&& is_failed(disk
))
3018 d
->status
|= FAILED_DISK
;
3020 idisk
= disk_list_get(d
->serial
, *disk_list
);
3023 if (disk
&& is_configured(disk
))
3024 idisk
->disk
.status
|= CONFIGURED_DISK
;
3027 dprintf("%s: mpb from %d:%d prefer %d:%d\n",
3028 __func__
, super
->disks
->major
,
3029 super
->disks
->minor
,
3030 table
[i
]->disks
->major
,
3031 table
[i
]->disks
->minor
);
3039 table
[tbl_size
++] = super
;
3043 /* update/extend the merged list of imsm_disk records */
3044 for (j
= 0; j
< mpb
->num_disks
; j
++) {
3045 struct imsm_disk
*disk
= __get_imsm_disk(mpb
, j
);
3046 struct intel_disk
*idisk
;
3048 idisk
= disk_list_get(disk
->serial
, *disk_list
);
3050 idisk
->disk
.status
|= disk
->status
;
3051 if (is_configured(&idisk
->disk
) ||
3052 is_failed(&idisk
->disk
))
3053 idisk
->disk
.status
&= ~(SPARE_DISK
);
3055 idisk
= calloc(1, sizeof(*idisk
));
3058 idisk
->owner
= IMSM_UNKNOWN_OWNER
;
3059 idisk
->disk
= *disk
;
3060 idisk
->next
= *disk_list
;
3064 if (serialcmp(idisk
->disk
.serial
, d
->serial
) == 0)
3071 static struct intel_super
*
3072 validate_members(struct intel_super
*super
, struct intel_disk
*disk_list
,
3075 struct imsm_super
*mpb
= super
->anchor
;
3079 for (i
= 0; i
< mpb
->num_disks
; i
++) {
3080 struct imsm_disk
*disk
= __get_imsm_disk(mpb
, i
);
3081 struct intel_disk
*idisk
;
3083 idisk
= disk_list_get(disk
->serial
, disk_list
);
3085 if (idisk
->owner
== owner
||
3086 idisk
->owner
== IMSM_UNKNOWN_OWNER
)
3089 dprintf("%s: '%.16s' owner %d != %d\n",
3090 __func__
, disk
->serial
, idisk
->owner
,
3093 dprintf("%s: unknown disk %x [%d]: %.16s\n",
3094 __func__
, __le32_to_cpu(mpb
->family_num
), i
,
3100 if (ok_count
== mpb
->num_disks
)
3105 static void show_conflicts(__u32 family_num
, struct intel_super
*super_list
)
3107 struct intel_super
*s
;
3109 for (s
= super_list
; s
; s
= s
->next
) {
3110 if (family_num
!= s
->anchor
->family_num
)
3112 fprintf(stderr
, "Conflict, offlining family %#x on '%s'\n",
3113 __le32_to_cpu(family_num
), s
->disks
->devname
);
3117 static struct intel_super
*
3118 imsm_thunderdome(struct intel_super
**super_list
, int len
)
3120 struct intel_super
*super_table
[len
];
3121 struct intel_disk
*disk_list
= NULL
;
3122 struct intel_super
*champion
, *spare
;
3123 struct intel_super
*s
, **del
;
3128 memset(super_table
, 0, sizeof(super_table
));
3129 for (s
= *super_list
; s
; s
= s
->next
)
3130 tbl_size
= __prep_thunderdome(super_table
, tbl_size
, s
, &disk_list
);
3132 for (i
= 0; i
< tbl_size
; i
++) {
3133 struct imsm_disk
*d
;
3134 struct intel_disk
*idisk
;
3135 struct imsm_super
*mpb
= super_table
[i
]->anchor
;
3138 d
= &s
->disks
->disk
;
3140 /* 'd' must appear in merged disk list for its
3141 * configuration to be valid
3143 idisk
= disk_list_get(d
->serial
, disk_list
);
3144 if (idisk
&& idisk
->owner
== i
)
3145 s
= validate_members(s
, disk_list
, i
);
3150 dprintf("%s: marking family: %#x from %d:%d offline\n",
3151 __func__
, mpb
->family_num
,
3152 super_table
[i
]->disks
->major
,
3153 super_table
[i
]->disks
->minor
);
3157 /* This is where the mdadm implementation differs from the Windows
3158 * driver which has no strict concept of a container. We can only
3159 * assemble one family from a container, so when returning a prodigal
3160 * array member to this system the code will not be able to disambiguate
3161 * the container contents that should be assembled ("foreign" versus
3162 * "local"). It requires user intervention to set the orig_family_num
3163 * to a new value to establish a new container. The Windows driver in
3164 * this situation fixes up the volume name in place and manages the
3165 * foreign array as an independent entity.
3170 for (i
= 0; i
< tbl_size
; i
++) {
3171 struct intel_super
*tbl_ent
= super_table
[i
];
3177 if (tbl_ent
->anchor
->num_raid_devs
== 0) {
3182 if (s
&& !is_spare
) {
3183 show_conflicts(tbl_ent
->anchor
->family_num
, *super_list
);
3185 } else if (!s
&& !is_spare
)
3198 fprintf(stderr
, "Chose family %#x on '%s', "
3199 "assemble conflicts to new container with '--update=uuid'\n",
3200 __le32_to_cpu(s
->anchor
->family_num
), s
->disks
->devname
);
3202 /* collect all dl's onto 'champion', and update them to
3203 * champion's version of the status
3205 for (s
= *super_list
; s
; s
= s
->next
) {
3206 struct imsm_super
*mpb
= champion
->anchor
;
3207 struct dl
*dl
= s
->disks
;
3212 for (i
= 0; i
< mpb
->num_disks
; i
++) {
3213 struct imsm_disk
*disk
;
3215 disk
= __serial_to_disk(dl
->serial
, mpb
, &dl
->index
);
3218 /* only set index on disks that are a member of
3219 * a populated contianer, i.e. one with
3222 if (is_failed(&dl
->disk
))
3224 else if (is_spare(&dl
->disk
))
3230 if (i
>= mpb
->num_disks
) {
3231 struct intel_disk
*idisk
;
3233 idisk
= disk_list_get(dl
->serial
, disk_list
);
3234 if (idisk
&& is_spare(&idisk
->disk
) &&
3235 !is_failed(&idisk
->disk
) && !is_configured(&idisk
->disk
))
3243 dl
->next
= champion
->disks
;
3244 champion
->disks
= dl
;
3248 /* delete 'champion' from super_list */
3249 for (del
= super_list
; *del
; ) {
3250 if (*del
== champion
) {
3251 *del
= (*del
)->next
;
3254 del
= &(*del
)->next
;
3256 champion
->next
= NULL
;
3260 struct intel_disk
*idisk
= disk_list
;
3262 disk_list
= disk_list
->next
;
3269 static int load_super_imsm_all(struct supertype
*st
, int fd
, void **sbp
,
3273 struct intel_super
*super_list
= NULL
;
3274 struct intel_super
*super
= NULL
;
3275 int devnum
= fd2devnum(fd
);
3281 /* check if 'fd' an opened container */
3282 sra
= sysfs_read(fd
, 0, GET_LEVEL
|GET_VERSION
|GET_DEVS
|GET_STATE
);
3286 if (sra
->array
.major_version
!= -1 ||
3287 sra
->array
.minor_version
!= -2 ||
3288 strcmp(sra
->text_version
, "imsm") != 0) {
3293 for (sd
= sra
->devs
, i
= 0; sd
; sd
= sd
->next
, i
++) {
3294 struct intel_super
*s
= alloc_super();
3302 s
->next
= super_list
;
3306 sprintf(nm
, "%d:%d", sd
->disk
.major
, sd
->disk
.minor
);
3307 dfd
= dev_open(nm
, O_RDWR
);
3311 rv
= find_intel_hba_capability(dfd
, s
, 1);
3312 /* no orom/efi or non-intel hba of the disk */
3316 err
= load_and_parse_mpb(dfd
, s
, NULL
, 1);
3318 /* retry the load if we might have raced against mdmon */
3319 if (err
== 3 && mdmon_running(devnum
))
3320 for (retry
= 0; retry
< 3; retry
++) {
3322 err
= load_and_parse_mpb(dfd
, s
, NULL
, 1);
3330 /* all mpbs enter, maybe one leaves */
3331 super
= imsm_thunderdome(&super_list
, i
);
3337 if (find_missing(super
) != 0) {
3345 while (super_list
) {
3346 struct intel_super
*s
= super_list
;
3348 super_list
= super_list
->next
;
3357 st
->container_dev
= devnum
;
3358 if (err
== 0 && st
->ss
== NULL
) {
3359 st
->ss
= &super_imsm
;
3360 st
->minor_version
= 0;
3361 st
->max_devs
= IMSM_MAX_DEVICES
;
3366 static int load_container_imsm(struct supertype
*st
, int fd
, char *devname
)
3368 return load_super_imsm_all(st
, fd
, &st
->sb
, devname
);
3372 static int load_super_imsm(struct supertype
*st
, int fd
, char *devname
)
3374 struct intel_super
*super
;
3377 if (test_partition(fd
))
3378 /* IMSM not allowed on partitions */
3381 free_super_imsm(st
);
3383 super
= alloc_super();
3386 Name
": malloc of %zu failed.\n",
3390 rv
= find_intel_hba_capability(fd
, super
, 1);
3391 /* no orom/efi or non-intel hba of the disk */
3395 Name
": No OROM/EFI properties for %s\n", devname
);
3400 rv
= load_and_parse_mpb(fd
, super
, devname
, 0);
3405 Name
": Failed to load all information "
3406 "sections on %s\n", devname
);
3412 if (st
->ss
== NULL
) {
3413 st
->ss
= &super_imsm
;
3414 st
->minor_version
= 0;
3415 st
->max_devs
= IMSM_MAX_DEVICES
;
3420 static __u16
info_to_blocks_per_strip(mdu_array_info_t
*info
)
3422 if (info
->level
== 1)
3424 return info
->chunk_size
>> 9;
3427 static __u32
info_to_num_data_stripes(mdu_array_info_t
*info
, int num_domains
)
3431 num_stripes
= (info
->size
* 2) / info_to_blocks_per_strip(info
);
3432 num_stripes
/= num_domains
;
3437 static __u32
info_to_blocks_per_member(mdu_array_info_t
*info
)
3439 if (info
->level
== 1)
3440 return info
->size
* 2;
3442 return (info
->size
* 2) & ~(info_to_blocks_per_strip(info
) - 1);
3445 static void imsm_update_version_info(struct intel_super
*super
)
3447 /* update the version and attributes */
3448 struct imsm_super
*mpb
= super
->anchor
;
3450 struct imsm_dev
*dev
;
3451 struct imsm_map
*map
;
3454 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
3455 dev
= get_imsm_dev(super
, i
);
3456 map
= get_imsm_map(dev
, 0);
3457 if (__le32_to_cpu(dev
->size_high
) > 0)
3458 mpb
->attributes
|= MPB_ATTRIB_2TB
;
3460 /* FIXME detect when an array spans a port multiplier */
3462 mpb
->attributes
|= MPB_ATTRIB_PM
;
3465 if (mpb
->num_raid_devs
> 1 ||
3466 mpb
->attributes
!= MPB_ATTRIB_CHECKSUM_VERIFY
) {
3467 version
= MPB_VERSION_ATTRIBS
;
3468 switch (get_imsm_raid_level(map
)) {
3469 case 0: mpb
->attributes
|= MPB_ATTRIB_RAID0
; break;
3470 case 1: mpb
->attributes
|= MPB_ATTRIB_RAID1
; break;
3471 case 10: mpb
->attributes
|= MPB_ATTRIB_RAID10
; break;
3472 case 5: mpb
->attributes
|= MPB_ATTRIB_RAID5
; break;
3475 if (map
->num_members
>= 5)
3476 version
= MPB_VERSION_5OR6_DISK_ARRAY
;
3477 else if (dev
->status
== DEV_CLONE_N_GO
)
3478 version
= MPB_VERSION_CNG
;
3479 else if (get_imsm_raid_level(map
) == 5)
3480 version
= MPB_VERSION_RAID5
;
3481 else if (map
->num_members
>= 3)
3482 version
= MPB_VERSION_3OR4_DISK_ARRAY
;
3483 else if (get_imsm_raid_level(map
) == 1)
3484 version
= MPB_VERSION_RAID1
;
3486 version
= MPB_VERSION_RAID0
;
3488 strcpy(((char *) mpb
->sig
) + strlen(MPB_SIGNATURE
), version
);
3492 static int check_name(struct intel_super
*super
, char *name
, int quiet
)
3494 struct imsm_super
*mpb
= super
->anchor
;
3495 char *reason
= NULL
;
3498 if (strlen(name
) > MAX_RAID_SERIAL_LEN
)
3499 reason
= "must be 16 characters or less";
3501 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
3502 struct imsm_dev
*dev
= get_imsm_dev(super
, i
);
3504 if (strncmp((char *) dev
->volume
, name
, MAX_RAID_SERIAL_LEN
) == 0) {
3505 reason
= "already exists";
3510 if (reason
&& !quiet
)
3511 fprintf(stderr
, Name
": imsm volume name %s\n", reason
);
3516 static int init_super_imsm_volume(struct supertype
*st
, mdu_array_info_t
*info
,
3517 unsigned long long size
, char *name
,
3518 char *homehost
, int *uuid
)
3520 /* We are creating a volume inside a pre-existing container.
3521 * so st->sb is already set.
3523 struct intel_super
*super
= st
->sb
;
3524 struct imsm_super
*mpb
= super
->anchor
;
3525 struct intel_dev
*dv
;
3526 struct imsm_dev
*dev
;
3527 struct imsm_vol
*vol
;
3528 struct imsm_map
*map
;
3529 int idx
= mpb
->num_raid_devs
;
3531 unsigned long long array_blocks
;
3532 size_t size_old
, size_new
;
3533 __u32 num_data_stripes
;
3535 if (super
->orom
&& mpb
->num_raid_devs
>= super
->orom
->vpa
) {
3536 fprintf(stderr
, Name
": This imsm-container already has the "
3537 "maximum of %d volumes\n", super
->orom
->vpa
);
3541 /* ensure the mpb is large enough for the new data */
3542 size_old
= __le32_to_cpu(mpb
->mpb_size
);
3543 size_new
= disks_to_mpb_size(info
->nr_disks
);
3544 if (size_new
> size_old
) {
3546 size_t size_round
= ROUND_UP(size_new
, 512);
3548 if (posix_memalign(&mpb_new
, 512, size_round
) != 0) {
3549 fprintf(stderr
, Name
": could not allocate new mpb\n");
3552 memcpy(mpb_new
, mpb
, size_old
);
3555 super
->anchor
= mpb_new
;
3556 mpb
->mpb_size
= __cpu_to_le32(size_new
);
3557 memset(mpb_new
+ size_old
, 0, size_round
- size_old
);
3559 super
->current_vol
= idx
;
3560 /* when creating the first raid device in this container set num_disks
3561 * to zero, i.e. delete this spare and add raid member devices in
3562 * add_to_super_imsm_volume()
3564 if (super
->current_vol
== 0)
3567 if (!check_name(super
, name
, 0))
3569 dv
= malloc(sizeof(*dv
));
3571 fprintf(stderr
, Name
": failed to allocate device list entry\n");
3574 dev
= calloc(1, sizeof(*dev
) + sizeof(__u32
) * (info
->raid_disks
- 1));
3577 fprintf(stderr
, Name
": could not allocate raid device\n");
3581 strncpy((char *) dev
->volume
, name
, MAX_RAID_SERIAL_LEN
);
3582 if (info
->level
== 1)
3583 array_blocks
= info_to_blocks_per_member(info
);
3585 array_blocks
= calc_array_size(info
->level
, info
->raid_disks
,
3586 info
->layout
, info
->chunk_size
,
3588 /* round array size down to closest MB */
3589 array_blocks
= (array_blocks
>> SECT_PER_MB_SHIFT
) << SECT_PER_MB_SHIFT
;
3591 dev
->size_low
= __cpu_to_le32((__u32
) array_blocks
);
3592 dev
->size_high
= __cpu_to_le32((__u32
) (array_blocks
>> 32));
3593 dev
->status
= (DEV_READ_COALESCING
| DEV_WRITE_COALESCING
);
3595 vol
->migr_state
= 0;
3596 set_migr_type(dev
, MIGR_INIT
);
3598 vol
->curr_migr_unit
= 0;
3599 map
= get_imsm_map(dev
, 0);
3600 map
->pba_of_lba0
= __cpu_to_le32(super
->create_offset
);
3601 map
->blocks_per_member
= __cpu_to_le32(info_to_blocks_per_member(info
));
3602 map
->blocks_per_strip
= __cpu_to_le16(info_to_blocks_per_strip(info
));
3603 map
->failed_disk_num
= ~0;
3604 map
->map_state
= info
->level
? IMSM_T_STATE_UNINITIALIZED
:
3605 IMSM_T_STATE_NORMAL
;
3608 if (info
->level
== 1 && info
->raid_disks
> 2) {
3611 fprintf(stderr
, Name
": imsm does not support more than 2 disks"
3612 "in a raid1 volume\n");
3616 map
->raid_level
= info
->level
;
3617 if (info
->level
== 10) {
3618 map
->raid_level
= 1;
3619 map
->num_domains
= info
->raid_disks
/ 2;
3620 } else if (info
->level
== 1)
3621 map
->num_domains
= info
->raid_disks
;
3623 map
->num_domains
= 1;
3625 num_data_stripes
= info_to_num_data_stripes(info
, map
->num_domains
);
3626 map
->num_data_stripes
= __cpu_to_le32(num_data_stripes
);
3628 map
->num_members
= info
->raid_disks
;
3629 for (i
= 0; i
< map
->num_members
; i
++) {
3630 /* initialized in add_to_super */
3631 set_imsm_ord_tbl_ent(map
, i
, IMSM_ORD_REBUILD
);
3633 mpb
->num_raid_devs
++;
3636 dv
->index
= super
->current_vol
;
3637 dv
->next
= super
->devlist
;
3638 super
->devlist
= dv
;
3640 imsm_update_version_info(super
);
3645 static int init_super_imsm(struct supertype
*st
, mdu_array_info_t
*info
,
3646 unsigned long long size
, char *name
,
3647 char *homehost
, int *uuid
)
3649 /* This is primarily called by Create when creating a new array.
3650 * We will then get add_to_super called for each component, and then
3651 * write_init_super called to write it out to each device.
3652 * For IMSM, Create can create on fresh devices or on a pre-existing
3654 * To create on a pre-existing array a different method will be called.
3655 * This one is just for fresh drives.
3657 struct intel_super
*super
;
3658 struct imsm_super
*mpb
;
3663 return init_super_imsm_volume(st
, info
, size
, name
, homehost
, uuid
);
3666 mpb_size
= disks_to_mpb_size(info
->nr_disks
);
3670 super
= alloc_super();
3671 if (super
&& posix_memalign(&super
->buf
, 512, mpb_size
) != 0) {
3676 fprintf(stderr
, Name
3677 ": %s could not allocate superblock\n", __func__
);
3680 memset(super
->buf
, 0, mpb_size
);
3682 mpb
->mpb_size
= __cpu_to_le32(mpb_size
);
3686 /* zeroing superblock */
3690 mpb
->attributes
= MPB_ATTRIB_CHECKSUM_VERIFY
;
3692 version
= (char *) mpb
->sig
;
3693 strcpy(version
, MPB_SIGNATURE
);
3694 version
+= strlen(MPB_SIGNATURE
);
3695 strcpy(version
, MPB_VERSION_RAID0
);
3701 static int add_to_super_imsm_volume(struct supertype
*st
, mdu_disk_info_t
*dk
,
3702 int fd
, char *devname
)
3704 struct intel_super
*super
= st
->sb
;
3705 struct imsm_super
*mpb
= super
->anchor
;
3707 struct imsm_dev
*dev
;
3708 struct imsm_map
*map
;
3711 dev
= get_imsm_dev(super
, super
->current_vol
);
3712 map
= get_imsm_map(dev
, 0);
3714 if (! (dk
->state
& (1<<MD_DISK_SYNC
))) {
3715 fprintf(stderr
, Name
": %s: Cannot add spare devices to IMSM volume\n",
3721 /* we're doing autolayout so grab the pre-marked (in
3722 * validate_geometry) raid_disk
3724 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
3725 if (dl
->raiddisk
== dk
->raid_disk
)
3728 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
3729 if (dl
->major
== dk
->major
&&
3730 dl
->minor
== dk
->minor
)
3735 fprintf(stderr
, Name
": %s is not a member of the same container\n", devname
);
3739 /* add a pristine spare to the metadata */
3740 if (dl
->index
< 0) {
3741 dl
->index
= super
->anchor
->num_disks
;
3742 super
->anchor
->num_disks
++;
3744 /* Check the device has not already been added */
3745 slot
= get_imsm_disk_slot(map
, dl
->index
);
3747 (get_imsm_ord_tbl_ent(dev
, slot
, -1) & IMSM_ORD_REBUILD
) == 0) {
3748 fprintf(stderr
, Name
": %s has been included in this array twice\n",
3752 set_imsm_ord_tbl_ent(map
, dk
->number
, dl
->index
);
3753 dl
->disk
.status
= CONFIGURED_DISK
;
3755 /* if we are creating the first raid device update the family number */
3756 if (super
->current_vol
== 0) {
3758 struct imsm_dev
*_dev
= __get_imsm_dev(mpb
, 0);
3759 struct imsm_disk
*_disk
= __get_imsm_disk(mpb
, dl
->index
);
3761 if (!_dev
|| !_disk
) {
3762 fprintf(stderr
, Name
": BUG mpb setup error\n");
3768 sum
+= __gen_imsm_checksum(mpb
);
3769 mpb
->family_num
= __cpu_to_le32(sum
);
3770 mpb
->orig_family_num
= mpb
->family_num
;
3777 static int add_to_super_imsm(struct supertype
*st
, mdu_disk_info_t
*dk
,
3778 int fd
, char *devname
)
3780 struct intel_super
*super
= st
->sb
;
3782 unsigned long long size
;
3787 /* If we are on an RAID enabled platform check that the disk is
3788 * attached to the raid controller.
3789 * We do not need to test disks attachment for container based additions,
3790 * they shall be already tested when container was created/assembled.
3792 rv
= find_intel_hba_capability(fd
, super
, 1);
3793 /* no orom/efi or non-intel hba of the disk */
3795 dprintf("capability: %p fd: %d ret: %d\n",
3796 super
->orom
, fd
, rv
);
3800 if (super
->current_vol
>= 0)
3801 return add_to_super_imsm_volume(st
, dk
, fd
, devname
);
3804 dd
= malloc(sizeof(*dd
));
3807 Name
": malloc failed %s:%d.\n", __func__
, __LINE__
);
3810 memset(dd
, 0, sizeof(*dd
));
3811 dd
->major
= major(stb
.st_rdev
);
3812 dd
->minor
= minor(stb
.st_rdev
);
3814 dd
->devname
= devname
? strdup(devname
) : NULL
;
3817 dd
->action
= DISK_ADD
;
3818 rv
= imsm_read_serial(fd
, devname
, dd
->serial
);
3821 Name
": failed to retrieve scsi serial, aborting\n");
3826 get_dev_size(fd
, NULL
, &size
);
3828 serialcpy(dd
->disk
.serial
, dd
->serial
);
3829 dd
->disk
.total_blocks
= __cpu_to_le32(size
);
3830 dd
->disk
.status
= SPARE_DISK
;
3831 if (sysfs_disk_to_scsi_id(fd
, &id
) == 0)
3832 dd
->disk
.scsi_id
= __cpu_to_le32(id
);
3834 dd
->disk
.scsi_id
= __cpu_to_le32(0);
3836 if (st
->update_tail
) {
3837 dd
->next
= super
->disk_mgmt_list
;
3838 super
->disk_mgmt_list
= dd
;
3840 dd
->next
= super
->disks
;
3848 static int remove_from_super_imsm(struct supertype
*st
, mdu_disk_info_t
*dk
)
3850 struct intel_super
*super
= st
->sb
;
3853 /* remove from super works only in mdmon - for communication
3854 * manager - monitor. Check if communication memory buffer
3857 if (!st
->update_tail
) {
3859 Name
": %s shall be used in mdmon context only"
3860 "(line %d).\n", __func__
, __LINE__
);
3863 dd
= malloc(sizeof(*dd
));
3866 Name
": malloc failed %s:%d.\n", __func__
, __LINE__
);
3869 memset(dd
, 0, sizeof(*dd
));
3870 dd
->major
= dk
->major
;
3871 dd
->minor
= dk
->minor
;
3874 dd
->disk
.status
= SPARE_DISK
;
3875 dd
->action
= DISK_REMOVE
;
3877 dd
->next
= super
->disk_mgmt_list
;
3878 super
->disk_mgmt_list
= dd
;
3884 static int store_imsm_mpb(int fd
, struct imsm_super
*mpb
);
3888 struct imsm_super anchor
;
3889 } spare_record
__attribute__ ((aligned(512)));
3891 /* spare records have their own family number and do not have any defined raid
3894 static int write_super_imsm_spares(struct intel_super
*super
, int doclose
)
3896 struct imsm_super
*mpb
= super
->anchor
;
3897 struct imsm_super
*spare
= &spare_record
.anchor
;
3901 spare
->mpb_size
= __cpu_to_le32(sizeof(struct imsm_super
)),
3902 spare
->generation_num
= __cpu_to_le32(1UL),
3903 spare
->attributes
= MPB_ATTRIB_CHECKSUM_VERIFY
;
3904 spare
->num_disks
= 1,
3905 spare
->num_raid_devs
= 0,
3906 spare
->cache_size
= mpb
->cache_size
,
3907 spare
->pwr_cycle_count
= __cpu_to_le32(1),
3909 snprintf((char *) spare
->sig
, MAX_SIGNATURE_LENGTH
,
3910 MPB_SIGNATURE MPB_VERSION_RAID0
);
3912 for (d
= super
->disks
; d
; d
= d
->next
) {
3916 spare
->disk
[0] = d
->disk
;
3917 sum
= __gen_imsm_checksum(spare
);
3918 spare
->family_num
= __cpu_to_le32(sum
);
3919 spare
->orig_family_num
= 0;
3920 sum
= __gen_imsm_checksum(spare
);
3921 spare
->check_sum
= __cpu_to_le32(sum
);
3923 if (store_imsm_mpb(d
->fd
, spare
)) {
3924 fprintf(stderr
, "%s: failed for device %d:%d %s\n",
3925 __func__
, d
->major
, d
->minor
, strerror(errno
));
3937 static int write_super_imsm(struct supertype
*st
, int doclose
)
3939 struct intel_super
*super
= st
->sb
;
3940 struct imsm_super
*mpb
= super
->anchor
;
3946 __u32 mpb_size
= sizeof(struct imsm_super
) - sizeof(struct imsm_disk
);
3949 /* 'generation' is incremented everytime the metadata is written */
3950 generation
= __le32_to_cpu(mpb
->generation_num
);
3952 mpb
->generation_num
= __cpu_to_le32(generation
);
3954 /* fix up cases where previous mdadm releases failed to set
3957 if (mpb
->orig_family_num
== 0)
3958 mpb
->orig_family_num
= mpb
->family_num
;
3960 for (d
= super
->disks
; d
; d
= d
->next
) {
3964 mpb
->disk
[d
->index
] = d
->disk
;
3968 for (d
= super
->missing
; d
; d
= d
->next
) {
3969 mpb
->disk
[d
->index
] = d
->disk
;
3972 mpb
->num_disks
= num_disks
;
3973 mpb_size
+= sizeof(struct imsm_disk
) * mpb
->num_disks
;
3975 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
3976 struct imsm_dev
*dev
= __get_imsm_dev(mpb
, i
);
3977 struct imsm_dev
*dev2
= get_imsm_dev(super
, i
);
3979 imsm_copy_dev(dev
, dev2
);
3980 mpb_size
+= sizeof_imsm_dev(dev
, 0);
3983 mpb_size
+= __le32_to_cpu(mpb
->bbm_log_size
);
3984 mpb
->mpb_size
= __cpu_to_le32(mpb_size
);
3986 /* recalculate checksum */
3987 sum
= __gen_imsm_checksum(mpb
);
3988 mpb
->check_sum
= __cpu_to_le32(sum
);
3990 /* write the mpb for disks that compose raid devices */
3991 for (d
= super
->disks
; d
; d
= d
->next
) {
3994 if (store_imsm_mpb(d
->fd
, mpb
))
3995 fprintf(stderr
, "%s: failed for device %d:%d %s\n",
3996 __func__
, d
->major
, d
->minor
, strerror(errno
));
4004 return write_super_imsm_spares(super
, doclose
);
4010 static int create_array(struct supertype
*st
, int dev_idx
)
4013 struct imsm_update_create_array
*u
;
4014 struct intel_super
*super
= st
->sb
;
4015 struct imsm_dev
*dev
= get_imsm_dev(super
, dev_idx
);
4016 struct imsm_map
*map
= get_imsm_map(dev
, 0);
4017 struct disk_info
*inf
;
4018 struct imsm_disk
*disk
;
4021 len
= sizeof(*u
) - sizeof(*dev
) + sizeof_imsm_dev(dev
, 0) +
4022 sizeof(*inf
) * map
->num_members
;
4025 fprintf(stderr
, "%s: failed to allocate update buffer\n",
4030 u
->type
= update_create_array
;
4031 u
->dev_idx
= dev_idx
;
4032 imsm_copy_dev(&u
->dev
, dev
);
4033 inf
= get_disk_info(u
);
4034 for (i
= 0; i
< map
->num_members
; i
++) {
4035 int idx
= get_imsm_disk_idx(dev
, i
, -1);
4037 disk
= get_imsm_disk(super
, idx
);
4038 serialcpy(inf
[i
].serial
, disk
->serial
);
4040 append_metadata_update(st
, u
, len
);
4045 static int mgmt_disk(struct supertype
*st
)
4047 struct intel_super
*super
= st
->sb
;
4049 struct imsm_update_add_remove_disk
*u
;
4051 if (!super
->disk_mgmt_list
)
4057 fprintf(stderr
, "%s: failed to allocate update buffer\n",
4062 u
->type
= update_add_remove_disk
;
4063 append_metadata_update(st
, u
, len
);
4068 static int write_init_super_imsm(struct supertype
*st
)
4070 struct intel_super
*super
= st
->sb
;
4071 int current_vol
= super
->current_vol
;
4073 /* we are done with current_vol reset it to point st at the container */
4074 super
->current_vol
= -1;
4076 if (st
->update_tail
) {
4077 /* queue the recently created array / added disk
4078 * as a metadata update */
4081 /* determine if we are creating a volume or adding a disk */
4082 if (current_vol
< 0) {
4083 /* in the mgmt (add/remove) disk case we are running
4084 * in mdmon context, so don't close fd's
4086 return mgmt_disk(st
);
4088 rv
= create_array(st
, current_vol
);
4093 for (d
= super
->disks
; d
; d
= d
->next
)
4094 Kill(d
->devname
, NULL
, 0, 1, 1);
4095 return write_super_imsm(st
, 1);
4100 static int store_super_imsm(struct supertype
*st
, int fd
)
4102 struct intel_super
*super
= st
->sb
;
4103 struct imsm_super
*mpb
= super
? super
->anchor
: NULL
;
4109 return store_imsm_mpb(fd
, mpb
);
4115 static int imsm_bbm_log_size(struct imsm_super
*mpb
)
4117 return __le32_to_cpu(mpb
->bbm_log_size
);
4121 static int validate_geometry_imsm_container(struct supertype
*st
, int level
,
4122 int layout
, int raiddisks
, int chunk
,
4123 unsigned long long size
, char *dev
,
4124 unsigned long long *freesize
,
4128 unsigned long long ldsize
;
4129 struct intel_super
*super
=NULL
;
4132 if (level
!= LEVEL_CONTAINER
)
4137 fd
= open(dev
, O_RDONLY
|O_EXCL
, 0);
4140 fprintf(stderr
, Name
": imsm: Cannot open %s: %s\n",
4141 dev
, strerror(errno
));
4144 if (!get_dev_size(fd
, dev
, &ldsize
)) {
4149 /* capabilities retrieve could be possible
4150 * note that there is no fd for the disks in array.
4152 super
= alloc_super();
4155 Name
": malloc of %zu failed.\n",
4161 rv
= find_intel_hba_capability(fd
, super
, verbose
);
4165 fd2devname(fd
, str
);
4166 dprintf("validate_geometry_imsm_container: fd: %d %s orom: %p rv: %d raiddisk: %d\n",
4167 fd
, str
, super
->orom
, rv
, raiddisks
);
4169 /* no orom/efi or non-intel hba of the disk */
4175 if (super
->orom
&& raiddisks
> super
->orom
->tds
) {
4177 fprintf(stderr
, Name
": %d exceeds maximum number of"
4178 " platform supported disks: %d\n",
4179 raiddisks
, super
->orom
->tds
);
4185 *freesize
= avail_size_imsm(st
, ldsize
>> 9);
4191 static unsigned long long find_size(struct extent
*e
, int *idx
, int num_extents
)
4193 const unsigned long long base_start
= e
[*idx
].start
;
4194 unsigned long long end
= base_start
+ e
[*idx
].size
;
4197 if (base_start
== end
)
4201 for (i
= *idx
; i
< num_extents
; i
++) {
4202 /* extend overlapping extents */
4203 if (e
[i
].start
>= base_start
&&
4204 e
[i
].start
<= end
) {
4207 if (e
[i
].start
+ e
[i
].size
> end
)
4208 end
= e
[i
].start
+ e
[i
].size
;
4209 } else if (e
[i
].start
> end
) {
4215 return end
- base_start
;
4218 static unsigned long long merge_extents(struct intel_super
*super
, int sum_extents
)
4220 /* build a composite disk with all known extents and generate a new
4221 * 'maxsize' given the "all disks in an array must share a common start
4222 * offset" constraint
4224 struct extent
*e
= calloc(sum_extents
, sizeof(*e
));
4228 unsigned long long pos
;
4229 unsigned long long start
= 0;
4230 unsigned long long maxsize
;
4231 unsigned long reserve
;
4236 /* coalesce and sort all extents. also, check to see if we need to
4237 * reserve space between member arrays
4240 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
4243 for (i
= 0; i
< dl
->extent_cnt
; i
++)
4246 qsort(e
, sum_extents
, sizeof(*e
), cmp_extent
);
4251 while (i
< sum_extents
) {
4252 e
[j
].start
= e
[i
].start
;
4253 e
[j
].size
= find_size(e
, &i
, sum_extents
);
4255 if (e
[j
-1].size
== 0)
4264 unsigned long long esize
;
4266 esize
= e
[i
].start
- pos
;
4267 if (esize
>= maxsize
) {
4272 pos
= e
[i
].start
+ e
[i
].size
;
4274 } while (e
[i
-1].size
);
4280 /* FIXME assumes volume at offset 0 is the first volume in a
4283 if (start_extent
> 0)
4284 reserve
= IMSM_RESERVED_SECTORS
; /* gap between raid regions */
4288 if (maxsize
< reserve
)
4291 super
->create_offset
= ~((__u32
) 0);
4292 if (start
+ reserve
> super
->create_offset
)
4293 return 0; /* start overflows create_offset */
4294 super
->create_offset
= start
+ reserve
;
4296 return maxsize
- reserve
;
4299 static int is_raid_level_supported(const struct imsm_orom
*orom
, int level
, int raiddisks
)
4301 if (level
< 0 || level
== 6 || level
== 4)
4304 /* if we have an orom prevent invalid raid levels */
4307 case 0: return imsm_orom_has_raid0(orom
);
4310 return imsm_orom_has_raid1e(orom
);
4311 return imsm_orom_has_raid1(orom
) && raiddisks
== 2;
4312 case 10: return imsm_orom_has_raid10(orom
) && raiddisks
== 4;
4313 case 5: return imsm_orom_has_raid5(orom
) && raiddisks
> 2;
4316 return 1; /* not on an Intel RAID platform so anything goes */
4322 #define pr_vrb(fmt, arg...) (void) (verbose && fprintf(stderr, Name fmt, ##arg))
4324 * validate volume parameters with OROM/EFI capabilities
4327 validate_geometry_imsm_orom(struct intel_super
*super
, int level
, int layout
,
4328 int raiddisks
, int *chunk
, int verbose
)
4333 /* validate container capabilities */
4334 if (super
->orom
&& raiddisks
> super
->orom
->tds
) {
4336 fprintf(stderr
, Name
": %d exceeds maximum number of"
4337 " platform supported disks: %d\n",
4338 raiddisks
, super
->orom
->tds
);
4342 /* capabilities of OROM tested - copied from validate_geometry_imsm_volume */
4343 if (super
->orom
&& (!is_raid_level_supported(super
->orom
, level
,
4345 pr_vrb(": platform does not support raid%d with %d disk%s\n",
4346 level
, raiddisks
, raiddisks
> 1 ? "s" : "");
4349 if (super
->orom
&& level
!= 1) {
4350 if (chunk
&& (*chunk
== 0 || *chunk
== UnSet
))
4351 *chunk
= imsm_orom_default_chunk(super
->orom
);
4352 else if (chunk
&& !imsm_orom_has_chunk(super
->orom
, *chunk
)) {
4353 pr_vrb(": platform does not support a chunk size of: "
4358 if (layout
!= imsm_level_to_layout(level
)) {
4360 pr_vrb(": imsm raid 5 only supports the left-asymmetric layout\n");
4361 else if (level
== 10)
4362 pr_vrb(": imsm raid 10 only supports the n2 layout\n");
4364 pr_vrb(": imsm unknown layout %#x for this raid level %d\n",
4371 /* validate_geometry_imsm_volume - lifted from validate_geometry_ddf_bvd
4372 * FIX ME add ahci details
4374 static int validate_geometry_imsm_volume(struct supertype
*st
, int level
,
4375 int layout
, int raiddisks
, int *chunk
,
4376 unsigned long long size
, char *dev
,
4377 unsigned long long *freesize
,
4381 struct intel_super
*super
= st
->sb
;
4382 struct imsm_super
*mpb
= super
->anchor
;
4384 unsigned long long pos
= 0;
4385 unsigned long long maxsize
;
4389 /* We must have the container info already read in. */
4393 if (!validate_geometry_imsm_orom(super
, level
, layout
, raiddisks
, chunk
, verbose
)) {
4394 fprintf(stderr
, Name
": RAID gemetry validation failed. "
4395 "Cannot proceed with the action(s).\n");
4399 /* General test: make sure there is space for
4400 * 'raiddisks' device extents of size 'size' at a given
4403 unsigned long long minsize
= size
;
4404 unsigned long long start_offset
= MaxSector
;
4407 minsize
= MPB_SECTOR_CNT
+ IMSM_RESERVED_SECTORS
;
4408 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
4413 e
= get_extents(super
, dl
);
4416 unsigned long long esize
;
4417 esize
= e
[i
].start
- pos
;
4418 if (esize
>= minsize
)
4420 if (found
&& start_offset
== MaxSector
) {
4423 } else if (found
&& pos
!= start_offset
) {
4427 pos
= e
[i
].start
+ e
[i
].size
;
4429 } while (e
[i
-1].size
);
4434 if (dcnt
< raiddisks
) {
4436 fprintf(stderr
, Name
": imsm: Not enough "
4437 "devices with space for this array "
4445 /* This device must be a member of the set */
4446 if (stat(dev
, &stb
) < 0)
4448 if ((S_IFMT
& stb
.st_mode
) != S_IFBLK
)
4450 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
4451 if (dl
->major
== (int)major(stb
.st_rdev
) &&
4452 dl
->minor
== (int)minor(stb
.st_rdev
))
4457 fprintf(stderr
, Name
": %s is not in the "
4458 "same imsm set\n", dev
);
4460 } else if (super
->orom
&& dl
->index
< 0 && mpb
->num_raid_devs
) {
4461 /* If a volume is present then the current creation attempt
4462 * cannot incorporate new spares because the orom may not
4463 * understand this configuration (all member disks must be
4464 * members of each array in the container).
4466 fprintf(stderr
, Name
": %s is a spare and a volume"
4467 " is already defined for this container\n", dev
);
4468 fprintf(stderr
, Name
": The option-rom requires all member"
4469 " disks to be a member of all volumes\n");
4473 /* retrieve the largest free space block */
4474 e
= get_extents(super
, dl
);
4479 unsigned long long esize
;
4481 esize
= e
[i
].start
- pos
;
4482 if (esize
>= maxsize
)
4484 pos
= e
[i
].start
+ e
[i
].size
;
4486 } while (e
[i
-1].size
);
4491 fprintf(stderr
, Name
": unable to determine free space for: %s\n",
4495 if (maxsize
< size
) {
4497 fprintf(stderr
, Name
": %s not enough space (%llu < %llu)\n",
4498 dev
, maxsize
, size
);
4502 /* count total number of extents for merge */
4504 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
4506 i
+= dl
->extent_cnt
;
4508 maxsize
= merge_extents(super
, i
);
4509 if (maxsize
< size
|| maxsize
== 0) {
4511 fprintf(stderr
, Name
": not enough space after merge (%llu < %llu)\n",
4516 *freesize
= maxsize
;
4521 static int reserve_space(struct supertype
*st
, int raiddisks
,
4522 unsigned long long size
, int chunk
,
4523 unsigned long long *freesize
)
4525 struct intel_super
*super
= st
->sb
;
4526 struct imsm_super
*mpb
= super
->anchor
;
4531 unsigned long long maxsize
;
4532 unsigned long long minsize
;
4536 /* find the largest common start free region of the possible disks */
4540 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
4546 /* don't activate new spares if we are orom constrained
4547 * and there is already a volume active in the container
4549 if (super
->orom
&& dl
->index
< 0 && mpb
->num_raid_devs
)
4552 e
= get_extents(super
, dl
);
4555 for (i
= 1; e
[i
-1].size
; i
++)
4563 maxsize
= merge_extents(super
, extent_cnt
);
4567 minsize
= chunk
* 2;
4569 if (cnt
< raiddisks
||
4570 (super
->orom
&& used
&& used
!= raiddisks
) ||
4571 maxsize
< minsize
||
4573 fprintf(stderr
, Name
": not enough devices with space to create array.\n");
4574 return 0; /* No enough free spaces large enough */
4586 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
4588 dl
->raiddisk
= cnt
++;
4595 static int validate_geometry_imsm(struct supertype
*st
, int level
, int layout
,
4596 int raiddisks
, int *chunk
, unsigned long long size
,
4597 char *dev
, unsigned long long *freesize
,
4605 * if given unused devices create a container
4606 * if given given devices in a container create a member volume
4608 if (level
== LEVEL_CONTAINER
) {
4609 /* Must be a fresh device to add to a container */
4610 return validate_geometry_imsm_container(st
, level
, layout
,
4612 chunk
?*chunk
:0, size
,
4618 if (st
->sb
&& freesize
) {
4619 /* we are being asked to automatically layout a
4620 * new volume based on the current contents of
4621 * the container. If the the parameters can be
4622 * satisfied reserve_space will record the disks,
4623 * start offset, and size of the volume to be
4624 * created. add_to_super and getinfo_super
4625 * detect when autolayout is in progress.
4627 if (!validate_geometry_imsm_orom(st
->sb
, level
, layout
,
4631 return reserve_space(st
, raiddisks
, size
,
4632 chunk
?*chunk
:0, freesize
);
4637 /* creating in a given container */
4638 return validate_geometry_imsm_volume(st
, level
, layout
,
4639 raiddisks
, chunk
, size
,
4640 dev
, freesize
, verbose
);
4643 /* This device needs to be a device in an 'imsm' container */
4644 fd
= open(dev
, O_RDONLY
|O_EXCL
, 0);
4648 Name
": Cannot create this array on device %s\n",
4653 if (errno
!= EBUSY
|| (fd
= open(dev
, O_RDONLY
, 0)) < 0) {
4655 fprintf(stderr
, Name
": Cannot open %s: %s\n",
4656 dev
, strerror(errno
));
4659 /* Well, it is in use by someone, maybe an 'imsm' container. */
4660 cfd
= open_container(fd
);
4664 fprintf(stderr
, Name
": Cannot use %s: It is busy\n",
4668 sra
= sysfs_read(cfd
, 0, GET_VERSION
);
4669 if (sra
&& sra
->array
.major_version
== -1 &&
4670 strcmp(sra
->text_version
, "imsm") == 0)
4674 /* This is a member of a imsm container. Load the container
4675 * and try to create a volume
4677 struct intel_super
*super
;
4679 if (load_super_imsm_all(st
, cfd
, (void **) &super
, NULL
) == 0) {
4681 st
->container_dev
= fd2devnum(cfd
);
4683 return validate_geometry_imsm_volume(st
, level
, layout
,
4691 fprintf(stderr
, Name
": failed container membership check\n");
4697 static void default_geometry_imsm(struct supertype
*st
, int *level
, int *layout
, int *chunk
)
4699 struct intel_super
*super
= st
->sb
;
4701 if (level
&& *level
== UnSet
)
4702 *level
= LEVEL_CONTAINER
;
4704 if (level
&& layout
&& *layout
== UnSet
)
4705 *layout
= imsm_level_to_layout(*level
);
4707 if (chunk
&& (*chunk
== UnSet
|| *chunk
== 0) &&
4708 super
&& super
->orom
)
4709 *chunk
= imsm_orom_default_chunk(super
->orom
);
4712 static void handle_missing(struct intel_super
*super
, struct imsm_dev
*dev
);
4714 static int kill_subarray_imsm(struct supertype
*st
)
4716 /* remove the subarray currently referenced by ->current_vol */
4718 struct intel_dev
**dp
;
4719 struct intel_super
*super
= st
->sb
;
4720 __u8 current_vol
= super
->current_vol
;
4721 struct imsm_super
*mpb
= super
->anchor
;
4723 if (super
->current_vol
< 0)
4725 super
->current_vol
= -1; /* invalidate subarray cursor */
4727 /* block deletions that would change the uuid of active subarrays
4729 * FIXME when immutable ids are available, but note that we'll
4730 * also need to fixup the invalidated/active subarray indexes in
4733 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
4736 if (i
< current_vol
)
4738 sprintf(subarray
, "%u", i
);
4739 if (is_subarray_active(subarray
, st
->devname
)) {
4741 Name
": deleting subarray-%d would change the UUID of active subarray-%d, aborting\n",
4748 if (st
->update_tail
) {
4749 struct imsm_update_kill_array
*u
= malloc(sizeof(*u
));
4753 u
->type
= update_kill_array
;
4754 u
->dev_idx
= current_vol
;
4755 append_metadata_update(st
, u
, sizeof(*u
));
4760 for (dp
= &super
->devlist
; *dp
;)
4761 if ((*dp
)->index
== current_vol
) {
4764 handle_missing(super
, (*dp
)->dev
);
4765 if ((*dp
)->index
> current_vol
)
4770 /* no more raid devices, all active components are now spares,
4771 * but of course failed are still failed
4773 if (--mpb
->num_raid_devs
== 0) {
4776 for (d
= super
->disks
; d
; d
= d
->next
)
4777 if (d
->index
> -2) {
4779 d
->disk
.status
= SPARE_DISK
;
4783 super
->updates_pending
++;
4788 static int update_subarray_imsm(struct supertype
*st
, char *subarray
,
4789 char *update
, struct mddev_ident
*ident
)
4791 /* update the subarray currently referenced by ->current_vol */
4792 struct intel_super
*super
= st
->sb
;
4793 struct imsm_super
*mpb
= super
->anchor
;
4795 if (strcmp(update
, "name") == 0) {
4796 char *name
= ident
->name
;
4800 if (is_subarray_active(subarray
, st
->devname
)) {
4802 Name
": Unable to update name of active subarray\n");
4806 if (!check_name(super
, name
, 0))
4809 vol
= strtoul(subarray
, &ep
, 10);
4810 if (*ep
!= '\0' || vol
>= super
->anchor
->num_raid_devs
)
4813 if (st
->update_tail
) {
4814 struct imsm_update_rename_array
*u
= malloc(sizeof(*u
));
4818 u
->type
= update_rename_array
;
4820 snprintf((char *) u
->name
, MAX_RAID_SERIAL_LEN
, "%s", name
);
4821 append_metadata_update(st
, u
, sizeof(*u
));
4823 struct imsm_dev
*dev
;
4826 dev
= get_imsm_dev(super
, vol
);
4827 snprintf((char *) dev
->volume
, MAX_RAID_SERIAL_LEN
, "%s", name
);
4828 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
4829 dev
= get_imsm_dev(super
, i
);
4830 handle_missing(super
, dev
);
4832 super
->updates_pending
++;
4840 static int is_gen_migration(struct imsm_dev
*dev
)
4842 if (!dev
->vol
.migr_state
)
4845 if (migr_type(dev
) == MIGR_GEN_MIGR
)
4850 #endif /* MDASSEMBLE */
4852 static int is_rebuilding(struct imsm_dev
*dev
)
4854 struct imsm_map
*migr_map
;
4856 if (!dev
->vol
.migr_state
)
4859 if (migr_type(dev
) != MIGR_REBUILD
)
4862 migr_map
= get_imsm_map(dev
, 1);
4864 if (migr_map
->map_state
== IMSM_T_STATE_DEGRADED
)
4870 static void update_recovery_start(struct imsm_dev
*dev
, struct mdinfo
*array
)
4872 struct mdinfo
*rebuild
= NULL
;
4876 if (!is_rebuilding(dev
))
4879 /* Find the rebuild target, but punt on the dual rebuild case */
4880 for (d
= array
->devs
; d
; d
= d
->next
)
4881 if (d
->recovery_start
== 0) {
4888 /* (?) none of the disks are marked with
4889 * IMSM_ORD_REBUILD, so assume they are missing and the
4890 * disk_ord_tbl was not correctly updated
4892 dprintf("%s: failed to locate out-of-sync disk\n", __func__
);
4896 units
= __le32_to_cpu(dev
->vol
.curr_migr_unit
);
4897 rebuild
->recovery_start
= units
* blocks_per_migr_unit(dev
);
4901 static struct mdinfo
*container_content_imsm(struct supertype
*st
, char *subarray
)
4903 /* Given a container loaded by load_super_imsm_all,
4904 * extract information about all the arrays into
4906 * If 'subarray' is given, just extract info about that array.
4908 * For each imsm_dev create an mdinfo, fill it in,
4909 * then look for matching devices in super->disks
4910 * and create appropriate device mdinfo.
4912 struct intel_super
*super
= st
->sb
;
4913 struct imsm_super
*mpb
= super
->anchor
;
4914 struct mdinfo
*rest
= NULL
;
4918 int spare_disks
= 0;
4920 /* check for bad blocks */
4921 if (imsm_bbm_log_size(super
->anchor
))
4924 /* count spare devices, not used in maps
4926 for (d
= super
->disks
; d
; d
= d
->next
)
4930 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
4931 struct imsm_dev
*dev
;
4932 struct imsm_map
*map
;
4933 struct imsm_map
*map2
;
4934 struct mdinfo
*this;
4939 (i
!= strtoul(subarray
, &ep
, 10) || *ep
!= '\0'))
4942 dev
= get_imsm_dev(super
, i
);
4943 map
= get_imsm_map(dev
, 0);
4944 map2
= get_imsm_map(dev
, 1);
4946 /* do not publish arrays that are in the middle of an
4947 * unsupported migration
4949 if (dev
->vol
.migr_state
&&
4950 (migr_type(dev
) == MIGR_STATE_CHANGE
)) {
4951 fprintf(stderr
, Name
": cannot assemble volume '%.16s':"
4952 " unsupported migration in progress\n",
4956 /* do not publish arrays that are not support by controller's
4960 chunk
= __le16_to_cpu(map
->blocks_per_strip
) >> 1;
4961 if (!validate_geometry_imsm_orom(super
,
4962 get_imsm_raid_level(map
), /* RAID level */
4963 imsm_level_to_layout(get_imsm_raid_level(map
)),
4964 map
->num_members
, /* raid disks */
4967 fprintf(stderr
, Name
": RAID gemetry validation failed. "
4968 "Cannot proceed with the action(s).\n");
4971 this = malloc(sizeof(*this));
4973 fprintf(stderr
, Name
": failed to allocate %zu bytes\n",
4977 memset(this, 0, sizeof(*this));
4980 super
->current_vol
= i
;
4981 getinfo_super_imsm_volume(st
, this, NULL
);
4982 for (slot
= 0 ; slot
< map
->num_members
; slot
++) {
4983 unsigned long long recovery_start
;
4984 struct mdinfo
*info_d
;
4991 idx
= get_imsm_disk_idx(dev
, slot
, 0);
4992 ord
= get_imsm_ord_tbl_ent(dev
, slot
, -1);
4993 for (d
= super
->disks
; d
; d
= d
->next
)
4994 if (d
->index
== idx
)
4997 recovery_start
= MaxSector
;
5000 if (d
&& is_failed(&d
->disk
))
5002 if (ord
& IMSM_ORD_REBUILD
)
5006 * if we skip some disks the array will be assmebled degraded;
5007 * reset resync start to avoid a dirty-degraded
5008 * situation when performing the intial sync
5010 * FIXME handle dirty degraded
5012 if ((skip
|| recovery_start
== 0) && !dev
->vol
.dirty
)
5013 this->resync_start
= MaxSector
;
5017 info_d
= calloc(1, sizeof(*info_d
));
5019 fprintf(stderr
, Name
": failed to allocate disk"
5020 " for volume %.16s\n", dev
->volume
);
5021 info_d
= this->devs
;
5023 struct mdinfo
*d
= info_d
->next
;
5032 info_d
->next
= this->devs
;
5033 this->devs
= info_d
;
5035 info_d
->disk
.number
= d
->index
;
5036 info_d
->disk
.major
= d
->major
;
5037 info_d
->disk
.minor
= d
->minor
;
5038 info_d
->disk
.raid_disk
= slot
;
5039 info_d
->recovery_start
= recovery_start
;
5041 if (slot
< map2
->num_members
)
5042 info_d
->disk
.state
= (1 << MD_DISK_ACTIVE
);
5044 this->array
.spare_disks
++;
5046 if (slot
< map
->num_members
)
5047 info_d
->disk
.state
= (1 << MD_DISK_ACTIVE
);
5049 this->array
.spare_disks
++;
5051 if (info_d
->recovery_start
== MaxSector
)
5052 this->array
.working_disks
++;
5054 info_d
->events
= __le32_to_cpu(mpb
->generation_num
);
5055 info_d
->data_offset
= __le32_to_cpu(map
->pba_of_lba0
);
5056 info_d
->component_size
= __le32_to_cpu(map
->blocks_per_member
);
5058 /* now that the disk list is up-to-date fixup recovery_start */
5059 update_recovery_start(dev
, this);
5060 this->array
.spare_disks
+= spare_disks
;
5064 /* if array has bad blocks, set suitable bit in array status */
5066 rest
->array
.state
|= (1<<MD_SB_BBM_ERRORS
);
5072 static __u8
imsm_check_degraded(struct intel_super
*super
, struct imsm_dev
*dev
, int failed
)
5074 struct imsm_map
*map
= get_imsm_map(dev
, 0);
5077 return map
->map_state
== IMSM_T_STATE_UNINITIALIZED
?
5078 IMSM_T_STATE_UNINITIALIZED
: IMSM_T_STATE_NORMAL
;
5080 switch (get_imsm_raid_level(map
)) {
5082 return IMSM_T_STATE_FAILED
;
5085 if (failed
< map
->num_members
)
5086 return IMSM_T_STATE_DEGRADED
;
5088 return IMSM_T_STATE_FAILED
;
5093 * check to see if any mirrors have failed, otherwise we
5094 * are degraded. Even numbered slots are mirrored on
5098 /* gcc -Os complains that this is unused */
5099 int insync
= insync
;
5101 for (i
= 0; i
< map
->num_members
; i
++) {
5102 __u32 ord
= get_imsm_ord_tbl_ent(dev
, i
, -1);
5103 int idx
= ord_to_idx(ord
);
5104 struct imsm_disk
*disk
;
5106 /* reset the potential in-sync count on even-numbered
5107 * slots. num_copies is always 2 for imsm raid10
5112 disk
= get_imsm_disk(super
, idx
);
5113 if (!disk
|| is_failed(disk
) || ord
& IMSM_ORD_REBUILD
)
5116 /* no in-sync disks left in this mirror the
5120 return IMSM_T_STATE_FAILED
;
5123 return IMSM_T_STATE_DEGRADED
;
5127 return IMSM_T_STATE_DEGRADED
;
5129 return IMSM_T_STATE_FAILED
;
5135 return map
->map_state
;
5138 static int imsm_count_failed(struct intel_super
*super
, struct imsm_dev
*dev
)
5142 struct imsm_disk
*disk
;
5143 struct imsm_map
*map
= get_imsm_map(dev
, 0);
5144 struct imsm_map
*prev
= get_imsm_map(dev
, dev
->vol
.migr_state
);
5148 /* at the beginning of migration we set IMSM_ORD_REBUILD on
5149 * disks that are being rebuilt. New failures are recorded to
5150 * map[0]. So we look through all the disks we started with and
5151 * see if any failures are still present, or if any new ones
5154 * FIXME add support for online capacity expansion and
5155 * raid-level-migration
5157 for (i
= 0; i
< prev
->num_members
; i
++) {
5158 ord
= __le32_to_cpu(prev
->disk_ord_tbl
[i
]);
5159 ord
|= __le32_to_cpu(map
->disk_ord_tbl
[i
]);
5160 idx
= ord_to_idx(ord
);
5162 disk
= get_imsm_disk(super
, idx
);
5163 if (!disk
|| is_failed(disk
) || ord
& IMSM_ORD_REBUILD
)
5171 static int imsm_open_new(struct supertype
*c
, struct active_array
*a
,
5174 struct intel_super
*super
= c
->sb
;
5175 struct imsm_super
*mpb
= super
->anchor
;
5177 if (atoi(inst
) >= mpb
->num_raid_devs
) {
5178 fprintf(stderr
, "%s: subarry index %d, out of range\n",
5179 __func__
, atoi(inst
));
5183 dprintf("imsm: open_new %s\n", inst
);
5184 a
->info
.container_member
= atoi(inst
);
5188 static int is_resyncing(struct imsm_dev
*dev
)
5190 struct imsm_map
*migr_map
;
5192 if (!dev
->vol
.migr_state
)
5195 if (migr_type(dev
) == MIGR_INIT
||
5196 migr_type(dev
) == MIGR_REPAIR
)
5199 if (migr_type(dev
) == MIGR_GEN_MIGR
)
5202 migr_map
= get_imsm_map(dev
, 1);
5204 if ((migr_map
->map_state
== IMSM_T_STATE_NORMAL
) &&
5205 (dev
->vol
.migr_type
!= MIGR_GEN_MIGR
))
5211 /* return true if we recorded new information */
5212 static int mark_failure(struct imsm_dev
*dev
, struct imsm_disk
*disk
, int idx
)
5216 struct imsm_map
*map
;
5217 char buf
[MAX_RAID_SERIAL_LEN
+3];
5218 unsigned int len
, shift
= 0;
5220 /* new failures are always set in map[0] */
5221 map
= get_imsm_map(dev
, 0);
5223 slot
= get_imsm_disk_slot(map
, idx
);
5227 ord
= __le32_to_cpu(map
->disk_ord_tbl
[slot
]);
5228 if (is_failed(disk
) && (ord
& IMSM_ORD_REBUILD
))
5231 sprintf(buf
, "%s:0", disk
->serial
);
5232 if ((len
= strlen(buf
)) >= MAX_RAID_SERIAL_LEN
)
5233 shift
= len
- MAX_RAID_SERIAL_LEN
+ 1;
5234 strncpy((char *)disk
->serial
, &buf
[shift
], MAX_RAID_SERIAL_LEN
);
5236 disk
->status
|= FAILED_DISK
;
5237 set_imsm_ord_tbl_ent(map
, slot
, idx
| IMSM_ORD_REBUILD
);
5238 if (map
->failed_disk_num
== 0xff)
5239 map
->failed_disk_num
= slot
;
5243 static void mark_missing(struct imsm_dev
*dev
, struct imsm_disk
*disk
, int idx
)
5245 mark_failure(dev
, disk
, idx
);
5247 if (disk
->scsi_id
== __cpu_to_le32(~(__u32
)0))
5250 disk
->scsi_id
= __cpu_to_le32(~(__u32
)0);
5251 memmove(&disk
->serial
[0], &disk
->serial
[1], MAX_RAID_SERIAL_LEN
- 1);
5254 static void handle_missing(struct intel_super
*super
, struct imsm_dev
*dev
)
5260 if (!super
->missing
)
5262 failed
= imsm_count_failed(super
, dev
);
5263 map_state
= imsm_check_degraded(super
, dev
, failed
);
5265 dprintf("imsm: mark missing\n");
5266 end_migration(dev
, map_state
);
5267 for (dl
= super
->missing
; dl
; dl
= dl
->next
)
5268 mark_missing(dev
, &dl
->disk
, dl
->index
);
5269 super
->updates_pending
++;
5272 static unsigned long long imsm_set_array_size(struct imsm_dev
*dev
)
5274 int used_disks
= imsm_num_data_members(dev
, 0);
5275 unsigned long long array_blocks
;
5276 struct imsm_map
*map
;
5278 if (used_disks
== 0) {
5279 /* when problems occures
5280 * return current array_blocks value
5282 array_blocks
= __le32_to_cpu(dev
->size_high
);
5283 array_blocks
= array_blocks
<< 32;
5284 array_blocks
+= __le32_to_cpu(dev
->size_low
);
5286 return array_blocks
;
5289 /* set array size in metadata
5291 map
= get_imsm_map(dev
, 0);
5292 array_blocks
= map
->blocks_per_member
* used_disks
;
5294 /* round array size down to closest MB
5296 array_blocks
= (array_blocks
>> SECT_PER_MB_SHIFT
) << SECT_PER_MB_SHIFT
;
5297 dev
->size_low
= __cpu_to_le32((__u32
)array_blocks
);
5298 dev
->size_high
= __cpu_to_le32((__u32
)(array_blocks
>> 32));
5300 return array_blocks
;
5303 static void imsm_set_disk(struct active_array
*a
, int n
, int state
);
5305 static void imsm_progress_container_reshape(struct intel_super
*super
)
5307 /* if no device has a migr_state, but some device has a
5308 * different number of members than the previous device, start
5309 * changing the number of devices in this device to match
5312 struct imsm_super
*mpb
= super
->anchor
;
5313 int prev_disks
= -1;
5317 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
5318 struct imsm_dev
*dev
= get_imsm_dev(super
, i
);
5319 struct imsm_map
*map
= get_imsm_map(dev
, 0);
5320 struct imsm_map
*map2
;
5321 int prev_num_members
;
5323 if (dev
->vol
.migr_state
)
5326 if (prev_disks
== -1)
5327 prev_disks
= map
->num_members
;
5328 if (prev_disks
== map
->num_members
)
5331 /* OK, this array needs to enter reshape mode.
5332 * i.e it needs a migr_state
5335 copy_map_size
= sizeof_imsm_map(map
);
5336 prev_num_members
= map
->num_members
;
5337 map
->num_members
= prev_disks
;
5338 dev
->vol
.migr_state
= 1;
5339 dev
->vol
.curr_migr_unit
= 0;
5340 dev
->vol
.migr_type
= MIGR_GEN_MIGR
;
5341 for (i
= prev_num_members
;
5342 i
< map
->num_members
; i
++)
5343 set_imsm_ord_tbl_ent(map
, i
, i
);
5344 map2
= get_imsm_map(dev
, 1);
5345 /* Copy the current map */
5346 memcpy(map2
, map
, copy_map_size
);
5347 map2
->num_members
= prev_num_members
;
5349 imsm_set_array_size(dev
);
5350 super
->updates_pending
++;
5354 /* Handle dirty -> clean transititions, resync and reshape. Degraded and rebuild
5355 * states are handled in imsm_set_disk() with one exception, when a
5356 * resync is stopped due to a new failure this routine will set the
5357 * 'degraded' state for the array.
5359 static int imsm_set_array_state(struct active_array
*a
, int consistent
)
5361 int inst
= a
->info
.container_member
;
5362 struct intel_super
*super
= a
->container
->sb
;
5363 struct imsm_dev
*dev
= get_imsm_dev(super
, inst
);
5364 struct imsm_map
*map
= get_imsm_map(dev
, 0);
5365 int failed
= imsm_count_failed(super
, dev
);
5366 __u8 map_state
= imsm_check_degraded(super
, dev
, failed
);
5367 __u32 blocks_per_unit
;
5369 if (dev
->vol
.migr_state
&&
5370 dev
->vol
.migr_type
== MIGR_GEN_MIGR
) {
5371 /* array state change is blocked due to reshape action
5373 * - abort the reshape (if last_checkpoint is 0 and action!= reshape)
5374 * - finish the reshape (if last_checkpoint is big and action != reshape)
5375 * - update curr_migr_unit
5377 if (a
->curr_action
== reshape
) {
5378 /* still reshaping, maybe update curr_migr_unit */
5379 goto mark_checkpoint
;
5381 if (a
->last_checkpoint
== 0 && a
->prev_action
== reshape
) {
5382 /* for some reason we aborted the reshape.
5385 struct imsm_map
*map2
= get_imsm_map(dev
, 1);
5386 dev
->vol
.migr_state
= 0;
5387 dev
->vol
.migr_type
= 0;
5388 dev
->vol
.curr_migr_unit
= 0;
5389 memcpy(map
, map2
, sizeof_imsm_map(map2
));
5390 super
->updates_pending
++;
5392 if (a
->last_checkpoint
>= a
->info
.component_size
) {
5393 unsigned long long array_blocks
;
5397 used_disks
= imsm_num_data_members(dev
, 0);
5398 if (used_disks
> 0) {
5400 map
->blocks_per_member
*
5402 /* round array size down to closest MB
5404 array_blocks
= (array_blocks
5405 >> SECT_PER_MB_SHIFT
)
5406 << SECT_PER_MB_SHIFT
;
5407 a
->info
.custom_array_size
= array_blocks
;
5408 /* encourage manager to update array
5412 a
->check_reshape
= 1;
5414 /* finalize online capacity expansion/reshape */
5415 for (mdi
= a
->info
.devs
; mdi
; mdi
= mdi
->next
)
5417 mdi
->disk
.raid_disk
,
5420 imsm_progress_container_reshape(super
);
5425 /* before we activate this array handle any missing disks */
5426 if (consistent
== 2)
5427 handle_missing(super
, dev
);
5429 if (consistent
== 2 &&
5430 (!is_resync_complete(&a
->info
) ||
5431 map_state
!= IMSM_T_STATE_NORMAL
||
5432 dev
->vol
.migr_state
))
5435 if (is_resync_complete(&a
->info
)) {
5436 /* complete intialization / resync,
5437 * recovery and interrupted recovery is completed in
5440 if (is_resyncing(dev
)) {
5441 dprintf("imsm: mark resync done\n");
5442 end_migration(dev
, map_state
);
5443 super
->updates_pending
++;
5444 a
->last_checkpoint
= 0;
5446 } else if (!is_resyncing(dev
) && !failed
) {
5447 /* mark the start of the init process if nothing is failed */
5448 dprintf("imsm: mark resync start\n");
5449 if (map
->map_state
== IMSM_T_STATE_UNINITIALIZED
)
5450 migrate(dev
, IMSM_T_STATE_NORMAL
, MIGR_INIT
);
5452 migrate(dev
, IMSM_T_STATE_NORMAL
, MIGR_REPAIR
);
5453 super
->updates_pending
++;
5457 /* check if we can update curr_migr_unit from resync_start, recovery_start */
5458 blocks_per_unit
= blocks_per_migr_unit(dev
);
5459 if (blocks_per_unit
) {
5463 units
= a
->last_checkpoint
/ blocks_per_unit
;
5466 /* check that we did not overflow 32-bits, and that
5467 * curr_migr_unit needs updating
5469 if (units32
== units
&&
5470 __le32_to_cpu(dev
->vol
.curr_migr_unit
) != units32
) {
5471 dprintf("imsm: mark checkpoint (%u)\n", units32
);
5472 dev
->vol
.curr_migr_unit
= __cpu_to_le32(units32
);
5473 super
->updates_pending
++;
5477 /* mark dirty / clean */
5478 if (dev
->vol
.dirty
!= !consistent
) {
5479 dprintf("imsm: mark '%s'\n", consistent
? "clean" : "dirty");
5484 super
->updates_pending
++;
5490 static void imsm_set_disk(struct active_array
*a
, int n
, int state
)
5492 int inst
= a
->info
.container_member
;
5493 struct intel_super
*super
= a
->container
->sb
;
5494 struct imsm_dev
*dev
= get_imsm_dev(super
, inst
);
5495 struct imsm_map
*map
= get_imsm_map(dev
, 0);
5496 struct imsm_disk
*disk
;
5501 if (n
> map
->num_members
)
5502 fprintf(stderr
, "imsm: set_disk %d out of range 0..%d\n",
5503 n
, map
->num_members
- 1);
5508 dprintf("imsm: set_disk %d:%x\n", n
, state
);
5510 ord
= get_imsm_ord_tbl_ent(dev
, n
, -1);
5511 disk
= get_imsm_disk(super
, ord_to_idx(ord
));
5513 /* check for new failures */
5514 if (state
& DS_FAULTY
) {
5515 if (mark_failure(dev
, disk
, ord_to_idx(ord
)))
5516 super
->updates_pending
++;
5519 /* check if in_sync */
5520 if (state
& DS_INSYNC
&& ord
& IMSM_ORD_REBUILD
&& is_rebuilding(dev
)) {
5521 struct imsm_map
*migr_map
= get_imsm_map(dev
, 1);
5523 set_imsm_ord_tbl_ent(migr_map
, n
, ord_to_idx(ord
));
5524 super
->updates_pending
++;
5527 failed
= imsm_count_failed(super
, dev
);
5528 map_state
= imsm_check_degraded(super
, dev
, failed
);
5530 /* check if recovery complete, newly degraded, or failed */
5531 if (map_state
== IMSM_T_STATE_NORMAL
&& is_rebuilding(dev
)) {
5532 end_migration(dev
, map_state
);
5533 map
= get_imsm_map(dev
, 0);
5534 map
->failed_disk_num
= ~0;
5535 super
->updates_pending
++;
5536 a
->last_checkpoint
= 0;
5537 } else if (map_state
== IMSM_T_STATE_DEGRADED
&&
5538 map
->map_state
!= map_state
&&
5539 !dev
->vol
.migr_state
) {
5540 dprintf("imsm: mark degraded\n");
5541 map
->map_state
= map_state
;
5542 super
->updates_pending
++;
5543 a
->last_checkpoint
= 0;
5544 } else if (map_state
== IMSM_T_STATE_FAILED
&&
5545 map
->map_state
!= map_state
) {
5546 dprintf("imsm: mark failed\n");
5547 end_migration(dev
, map_state
);
5548 super
->updates_pending
++;
5549 a
->last_checkpoint
= 0;
5550 } else if (is_gen_migration(dev
)) {
5551 dprintf("imsm: Detected General Migration in state: ");
5552 if (map_state
== IMSM_T_STATE_NORMAL
) {
5553 end_migration(dev
, map_state
);
5554 map
= get_imsm_map(dev
, 0);
5555 map
->failed_disk_num
= ~0;
5556 dprintf("normal\n");
5558 if (map_state
== IMSM_T_STATE_DEGRADED
) {
5559 printf("degraded\n");
5560 end_migration(dev
, map_state
);
5562 dprintf("failed\n");
5564 map
->map_state
= map_state
;
5566 super
->updates_pending
++;
5570 static int store_imsm_mpb(int fd
, struct imsm_super
*mpb
)
5573 __u32 mpb_size
= __le32_to_cpu(mpb
->mpb_size
);
5574 unsigned long long dsize
;
5575 unsigned long long sectors
;
5577 get_dev_size(fd
, NULL
, &dsize
);
5579 if (mpb_size
> 512) {
5580 /* -1 to account for anchor */
5581 sectors
= mpb_sectors(mpb
) - 1;
5583 /* write the extended mpb to the sectors preceeding the anchor */
5584 if (lseek64(fd
, dsize
- (512 * (2 + sectors
)), SEEK_SET
) < 0)
5587 if ((unsigned long long)write(fd
, buf
+ 512, 512 * sectors
)
5592 /* first block is stored on second to last sector of the disk */
5593 if (lseek64(fd
, dsize
- (512 * 2), SEEK_SET
) < 0)
5596 if (write(fd
, buf
, 512) != 512)
5602 static void imsm_sync_metadata(struct supertype
*container
)
5604 struct intel_super
*super
= container
->sb
;
5606 dprintf("sync metadata: %d\n", super
->updates_pending
);
5607 if (!super
->updates_pending
)
5610 write_super_imsm(container
, 0);
5612 super
->updates_pending
= 0;
5615 static struct dl
*imsm_readd(struct intel_super
*super
, int idx
, struct active_array
*a
)
5617 struct imsm_dev
*dev
= get_imsm_dev(super
, a
->info
.container_member
);
5618 int i
= get_imsm_disk_idx(dev
, idx
, -1);
5621 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
5625 if (dl
&& is_failed(&dl
->disk
))
5629 dprintf("%s: found %x:%x\n", __func__
, dl
->major
, dl
->minor
);
5634 static struct dl
*imsm_add_spare(struct intel_super
*super
, int slot
,
5635 struct active_array
*a
, int activate_new
,
5636 struct mdinfo
*additional_test_list
)
5638 struct imsm_dev
*dev
= get_imsm_dev(super
, a
->info
.container_member
);
5639 int idx
= get_imsm_disk_idx(dev
, slot
, -1);
5640 struct imsm_super
*mpb
= super
->anchor
;
5641 struct imsm_map
*map
;
5642 unsigned long long pos
;
5647 __u32 array_start
= 0;
5648 __u32 array_end
= 0;
5650 struct mdinfo
*test_list
;
5652 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
5653 /* If in this array, skip */
5654 for (d
= a
->info
.devs
; d
; d
= d
->next
)
5655 if (d
->state_fd
>= 0 &&
5656 d
->disk
.major
== dl
->major
&&
5657 d
->disk
.minor
== dl
->minor
) {
5658 dprintf("%x:%x already in array\n",
5659 dl
->major
, dl
->minor
);
5664 test_list
= additional_test_list
;
5666 if (test_list
->disk
.major
== dl
->major
&&
5667 test_list
->disk
.minor
== dl
->minor
) {
5668 dprintf("%x:%x already in additional test list\n",
5669 dl
->major
, dl
->minor
);
5672 test_list
= test_list
->next
;
5677 /* skip in use or failed drives */
5678 if (is_failed(&dl
->disk
) || idx
== dl
->index
||
5680 dprintf("%x:%x status (failed: %d index: %d)\n",
5681 dl
->major
, dl
->minor
, is_failed(&dl
->disk
), idx
);
5685 /* skip pure spares when we are looking for partially
5686 * assimilated drives
5688 if (dl
->index
== -1 && !activate_new
)
5691 /* Does this unused device have the requisite free space?
5692 * It needs to be able to cover all member volumes
5694 ex
= get_extents(super
, dl
);
5696 dprintf("cannot get extents\n");
5699 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
5700 dev
= get_imsm_dev(super
, i
);
5701 map
= get_imsm_map(dev
, 0);
5703 /* check if this disk is already a member of
5706 if (get_imsm_disk_slot(map
, dl
->index
) >= 0)
5712 array_start
= __le32_to_cpu(map
->pba_of_lba0
);
5713 array_end
= array_start
+
5714 __le32_to_cpu(map
->blocks_per_member
) - 1;
5717 /* check that we can start at pba_of_lba0 with
5718 * blocks_per_member of space
5720 if (array_start
>= pos
&& array_end
< ex
[j
].start
) {
5724 pos
= ex
[j
].start
+ ex
[j
].size
;
5726 } while (ex
[j
-1].size
);
5733 if (i
< mpb
->num_raid_devs
) {
5734 dprintf("%x:%x does not have %u to %u available\n",
5735 dl
->major
, dl
->minor
, array_start
, array_end
);
5746 static int imsm_rebuild_allowed(struct supertype
*cont
, int dev_idx
, int failed
)
5748 struct imsm_dev
*dev2
;
5749 struct imsm_map
*map
;
5755 dev2
= get_imsm_dev(cont
->sb
, dev_idx
);
5757 state
= imsm_check_degraded(cont
->sb
, dev2
, failed
);
5758 if (state
== IMSM_T_STATE_FAILED
) {
5759 map
= get_imsm_map(dev2
, 0);
5762 for (slot
= 0; slot
< map
->num_members
; slot
++) {
5764 * Check if failed disks are deleted from intel
5765 * disk list or are marked to be deleted
5767 idx
= get_imsm_disk_idx(dev2
, slot
, -1);
5768 idisk
= get_imsm_dl_disk(cont
->sb
, idx
);
5770 * Do not rebuild the array if failed disks
5771 * from failed sub-array are not removed from
5775 is_failed(&idisk
->disk
) &&
5776 (idisk
->action
!= DISK_REMOVE
))
5784 static struct mdinfo
*imsm_activate_spare(struct active_array
*a
,
5785 struct metadata_update
**updates
)
5788 * Find a device with unused free space and use it to replace a
5789 * failed/vacant region in an array. We replace failed regions one a
5790 * array at a time. The result is that a new spare disk will be added
5791 * to the first failed array and after the monitor has finished
5792 * propagating failures the remainder will be consumed.
5794 * FIXME add a capability for mdmon to request spares from another
5798 struct intel_super
*super
= a
->container
->sb
;
5799 int inst
= a
->info
.container_member
;
5800 struct imsm_dev
*dev
= get_imsm_dev(super
, inst
);
5801 struct imsm_map
*map
= get_imsm_map(dev
, 0);
5802 int failed
= a
->info
.array
.raid_disks
;
5803 struct mdinfo
*rv
= NULL
;
5806 struct metadata_update
*mu
;
5808 struct imsm_update_activate_spare
*u
;
5813 for (d
= a
->info
.devs
; d
; d
= d
->next
) {
5814 if ((d
->curr_state
& DS_FAULTY
) &&
5816 /* wait for Removal to happen */
5818 if (d
->state_fd
>= 0)
5822 dprintf("imsm: activate spare: inst=%d failed=%d (%d) level=%d\n",
5823 inst
, failed
, a
->info
.array
.raid_disks
, a
->info
.array
.level
);
5825 if (dev
->vol
.migr_state
&&
5826 dev
->vol
.migr_type
== MIGR_GEN_MIGR
)
5827 /* No repair during migration */
5830 if (a
->info
.array
.level
== 4)
5831 /* No repair for takeovered array
5832 * imsm doesn't support raid4
5836 if (imsm_check_degraded(super
, dev
, failed
) != IMSM_T_STATE_DEGRADED
)
5840 * If there are any failed disks check state of the other volume.
5841 * Block rebuild if the another one is failed until failed disks
5842 * are removed from container.
5845 dprintf("found failed disks in %s, check if there another"
5846 "failed sub-array.\n",
5848 /* check if states of the other volumes allow for rebuild */
5849 for (i
= 0; i
< super
->anchor
->num_raid_devs
; i
++) {
5851 allowed
= imsm_rebuild_allowed(a
->container
,
5859 /* For each slot, if it is not working, find a spare */
5860 for (i
= 0; i
< a
->info
.array
.raid_disks
; i
++) {
5861 for (d
= a
->info
.devs
; d
; d
= d
->next
)
5862 if (d
->disk
.raid_disk
== i
)
5864 dprintf("found %d: %p %x\n", i
, d
, d
?d
->curr_state
:0);
5865 if (d
&& (d
->state_fd
>= 0))
5869 * OK, this device needs recovery. Try to re-add the
5870 * previous occupant of this slot, if this fails see if
5871 * we can continue the assimilation of a spare that was
5872 * partially assimilated, finally try to activate a new
5875 dl
= imsm_readd(super
, i
, a
);
5877 dl
= imsm_add_spare(super
, i
, a
, 0, NULL
);
5879 dl
= imsm_add_spare(super
, i
, a
, 1, NULL
);
5883 /* found a usable disk with enough space */
5884 di
= malloc(sizeof(*di
));
5887 memset(di
, 0, sizeof(*di
));
5889 /* dl->index will be -1 in the case we are activating a
5890 * pristine spare. imsm_process_update() will create a
5891 * new index in this case. Once a disk is found to be
5892 * failed in all member arrays it is kicked from the
5895 di
->disk
.number
= dl
->index
;
5897 /* (ab)use di->devs to store a pointer to the device
5900 di
->devs
= (struct mdinfo
*) dl
;
5902 di
->disk
.raid_disk
= i
;
5903 di
->disk
.major
= dl
->major
;
5904 di
->disk
.minor
= dl
->minor
;
5906 di
->recovery_start
= 0;
5907 di
->data_offset
= __le32_to_cpu(map
->pba_of_lba0
);
5908 di
->component_size
= a
->info
.component_size
;
5909 di
->container_member
= inst
;
5910 super
->random
= random32();
5914 dprintf("%x:%x to be %d at %llu\n", dl
->major
, dl
->minor
,
5915 i
, di
->data_offset
);
5921 /* No spares found */
5923 /* Now 'rv' has a list of devices to return.
5924 * Create a metadata_update record to update the
5925 * disk_ord_tbl for the array
5927 mu
= malloc(sizeof(*mu
));
5929 mu
->buf
= malloc(sizeof(struct imsm_update_activate_spare
) * num_spares
);
5930 if (mu
->buf
== NULL
) {
5937 struct mdinfo
*n
= rv
->next
;
5946 mu
->space_list
= NULL
;
5947 mu
->len
= sizeof(struct imsm_update_activate_spare
) * num_spares
;
5948 mu
->next
= *updates
;
5949 u
= (struct imsm_update_activate_spare
*) mu
->buf
;
5951 for (di
= rv
; di
; di
= di
->next
) {
5952 u
->type
= update_activate_spare
;
5953 u
->dl
= (struct dl
*) di
->devs
;
5955 u
->slot
= di
->disk
.raid_disk
;
5966 static int disks_overlap(struct intel_super
*super
, int idx
, struct imsm_update_create_array
*u
)
5968 struct imsm_dev
*dev
= get_imsm_dev(super
, idx
);
5969 struct imsm_map
*map
= get_imsm_map(dev
, 0);
5970 struct imsm_map
*new_map
= get_imsm_map(&u
->dev
, 0);
5971 struct disk_info
*inf
= get_disk_info(u
);
5972 struct imsm_disk
*disk
;
5976 for (i
= 0; i
< map
->num_members
; i
++) {
5977 disk
= get_imsm_disk(super
, get_imsm_disk_idx(dev
, i
, -1));
5978 for (j
= 0; j
< new_map
->num_members
; j
++)
5979 if (serialcmp(disk
->serial
, inf
[j
].serial
) == 0)
5987 static struct dl
*get_disk_super(struct intel_super
*super
, int major
, int minor
)
5989 struct dl
*dl
= NULL
;
5990 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
5991 if ((dl
->major
== major
) && (dl
->minor
== minor
))
5996 static int remove_disk_super(struct intel_super
*super
, int major
, int minor
)
5998 struct dl
*prev
= NULL
;
6002 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
6003 if ((dl
->major
== major
) && (dl
->minor
== minor
)) {
6006 prev
->next
= dl
->next
;
6008 super
->disks
= dl
->next
;
6010 __free_imsm_disk(dl
);
6011 dprintf("%s: removed %x:%x\n",
6012 __func__
, major
, minor
);
6020 static void imsm_delete(struct intel_super
*super
, struct dl
**dlp
, unsigned index
);
6022 static int add_remove_disk_update(struct intel_super
*super
)
6024 int check_degraded
= 0;
6025 struct dl
*disk
= NULL
;
6026 /* add/remove some spares to/from the metadata/contrainer */
6027 while (super
->disk_mgmt_list
) {
6028 struct dl
*disk_cfg
;
6030 disk_cfg
= super
->disk_mgmt_list
;
6031 super
->disk_mgmt_list
= disk_cfg
->next
;
6032 disk_cfg
->next
= NULL
;
6034 if (disk_cfg
->action
== DISK_ADD
) {
6035 disk_cfg
->next
= super
->disks
;
6036 super
->disks
= disk_cfg
;
6038 dprintf("%s: added %x:%x\n",
6039 __func__
, disk_cfg
->major
,
6041 } else if (disk_cfg
->action
== DISK_REMOVE
) {
6042 dprintf("Disk remove action processed: %x.%x\n",
6043 disk_cfg
->major
, disk_cfg
->minor
);
6044 disk
= get_disk_super(super
,
6048 /* store action status */
6049 disk
->action
= DISK_REMOVE
;
6050 /* remove spare disks only */
6051 if (disk
->index
== -1) {
6052 remove_disk_super(super
,
6057 /* release allocate disk structure */
6058 __free_imsm_disk(disk_cfg
);
6061 return check_degraded
;
6064 static int apply_reshape_container_disks_update(struct imsm_update_reshape
*u
,
6065 struct intel_super
*super
,
6068 struct dl
*new_disk
;
6069 struct intel_dev
*id
;
6071 int delta_disks
= u
->new_raid_disks
- u
->old_raid_disks
;
6072 int disk_count
= u
->old_raid_disks
;
6073 void **tofree
= NULL
;
6074 int devices_to_reshape
= 1;
6075 struct imsm_super
*mpb
= super
->anchor
;
6077 unsigned int dev_id
;
6079 dprintf("imsm: apply_reshape_container_disks_update()\n");
6081 /* enable spares to use in array */
6082 for (i
= 0; i
< delta_disks
; i
++) {
6083 new_disk
= get_disk_super(super
,
6084 major(u
->new_disks
[i
]),
6085 minor(u
->new_disks
[i
]));
6086 dprintf("imsm: new disk for reshape is: %i:%i "
6087 "(%p, index = %i)\n",
6088 major(u
->new_disks
[i
]), minor(u
->new_disks
[i
]),
6089 new_disk
, new_disk
->index
);
6090 if ((new_disk
== NULL
) ||
6091 ((new_disk
->index
>= 0) &&
6092 (new_disk
->index
< u
->old_raid_disks
)))
6093 goto update_reshape_exit
;
6094 new_disk
->index
= disk_count
++;
6095 /* slot to fill in autolayout
6097 new_disk
->raiddisk
= new_disk
->index
;
6098 new_disk
->disk
.status
|=
6100 new_disk
->disk
.status
&= ~SPARE_DISK
;
6103 dprintf("imsm: volume set mpb->num_raid_devs = %i\n",
6104 mpb
->num_raid_devs
);
6105 /* manage changes in volume
6107 for (dev_id
= 0; dev_id
< mpb
->num_raid_devs
; dev_id
++) {
6108 void **sp
= *space_list
;
6109 struct imsm_dev
*newdev
;
6110 struct imsm_map
*newmap
, *oldmap
;
6112 for (id
= super
->devlist
; id
; id
= id
->next
) {
6113 if (id
->index
== dev_id
)
6122 /* Copy the dev, but not (all of) the map */
6123 memcpy(newdev
, id
->dev
, sizeof(*newdev
));
6124 oldmap
= get_imsm_map(id
->dev
, 0);
6125 newmap
= get_imsm_map(newdev
, 0);
6126 /* Copy the current map */
6127 memcpy(newmap
, oldmap
, sizeof_imsm_map(oldmap
));
6128 /* update one device only
6130 if (devices_to_reshape
) {
6131 dprintf("imsm: modifying subdev: %i\n",
6133 devices_to_reshape
--;
6134 newdev
->vol
.migr_state
= 1;
6135 newdev
->vol
.curr_migr_unit
= 0;
6136 newdev
->vol
.migr_type
= MIGR_GEN_MIGR
;
6137 newmap
->num_members
= u
->new_raid_disks
;
6138 for (i
= 0; i
< delta_disks
; i
++) {
6139 set_imsm_ord_tbl_ent(newmap
,
6140 u
->old_raid_disks
+ i
,
6141 u
->old_raid_disks
+ i
);
6143 /* New map is correct, now need to save old map
6145 newmap
= get_imsm_map(newdev
, 1);
6146 memcpy(newmap
, oldmap
, sizeof_imsm_map(oldmap
));
6148 imsm_set_array_size(newdev
);
6151 sp
= (void **)id
->dev
;
6157 *space_list
= tofree
;
6160 update_reshape_exit
:
6165 static int apply_takeover_update(struct imsm_update_takeover
*u
,
6166 struct intel_super
*super
,
6169 struct imsm_dev
*dev
= NULL
;
6170 struct intel_dev
*dv
;
6171 struct imsm_dev
*dev_new
;
6172 struct imsm_map
*map
;
6176 for (dv
= super
->devlist
; dv
; dv
= dv
->next
)
6177 if (dv
->index
== (unsigned int)u
->subarray
) {
6185 map
= get_imsm_map(dev
, 0);
6187 if (u
->direction
== R10_TO_R0
) {
6188 /* Number of failed disks must be half of initial disk number */
6189 if (imsm_count_failed(super
, dev
) != (map
->num_members
/ 2))
6192 /* iterate through devices to mark removed disks as spare */
6193 for (dm
= super
->disks
; dm
; dm
= dm
->next
) {
6194 if (dm
->disk
.status
& FAILED_DISK
) {
6195 int idx
= dm
->index
;
6196 /* update indexes on the disk list */
6197 /* FIXME this loop-with-the-loop looks wrong, I'm not convinced
6198 the index values will end up being correct.... NB */
6199 for (du
= super
->disks
; du
; du
= du
->next
)
6200 if (du
->index
> idx
)
6202 /* mark as spare disk */
6203 dm
->disk
.status
= SPARE_DISK
;
6208 map
->num_members
= map
->num_members
/ 2;
6209 map
->map_state
= IMSM_T_STATE_NORMAL
;
6210 map
->num_domains
= 1;
6211 map
->raid_level
= 0;
6212 map
->failed_disk_num
= -1;
6215 if (u
->direction
== R0_TO_R10
) {
6217 /* update slots in current disk list */
6218 for (dm
= super
->disks
; dm
; dm
= dm
->next
) {
6222 /* create new *missing* disks */
6223 for (i
= 0; i
< map
->num_members
; i
++) {
6224 space
= *space_list
;
6227 *space_list
= *space
;
6229 memcpy(du
, super
->disks
, sizeof(*du
));
6233 du
->index
= (i
* 2) + 1;
6234 sprintf((char *)du
->disk
.serial
,
6235 " MISSING_%d", du
->index
);
6236 sprintf((char *)du
->serial
,
6237 "MISSING_%d", du
->index
);
6238 du
->next
= super
->missing
;
6239 super
->missing
= du
;
6241 /* create new dev and map */
6242 space
= *space_list
;
6245 *space_list
= *space
;
6246 dev_new
= (void *)space
;
6247 memcpy(dev_new
, dev
, sizeof(*dev
));
6248 /* update new map */
6249 map
= get_imsm_map(dev_new
, 0);
6250 map
->num_members
= map
->num_members
* 2;
6251 map
->map_state
= IMSM_T_STATE_DEGRADED
;
6252 map
->num_domains
= 2;
6253 map
->raid_level
= 1;
6254 /* replace dev<->dev_new */
6257 /* update disk order table */
6258 for (du
= super
->disks
; du
; du
= du
->next
)
6260 set_imsm_ord_tbl_ent(map
, du
->index
, du
->index
);
6261 for (du
= super
->missing
; du
; du
= du
->next
)
6262 if (du
->index
>= 0) {
6263 set_imsm_ord_tbl_ent(map
, du
->index
, du
->index
);
6264 mark_missing(dev_new
, &du
->disk
, du
->index
);
6270 static void imsm_process_update(struct supertype
*st
,
6271 struct metadata_update
*update
)
6274 * crack open the metadata_update envelope to find the update record
6275 * update can be one of:
6276 * update_reshape_container_disks - all the arrays in the container
6277 * are being reshaped to have more devices. We need to mark
6278 * the arrays for general migration and convert selected spares
6279 * into active devices.
6280 * update_activate_spare - a spare device has replaced a failed
6281 * device in an array, update the disk_ord_tbl. If this disk is
6282 * present in all member arrays then also clear the SPARE_DISK
6284 * update_create_array
6286 * update_rename_array
6287 * update_add_remove_disk
6289 struct intel_super
*super
= st
->sb
;
6290 struct imsm_super
*mpb
;
6291 enum imsm_update_type type
= *(enum imsm_update_type
*) update
->buf
;
6293 /* update requires a larger buf but the allocation failed */
6294 if (super
->next_len
&& !super
->next_buf
) {
6295 super
->next_len
= 0;
6299 if (super
->next_buf
) {
6300 memcpy(super
->next_buf
, super
->buf
, super
->len
);
6302 super
->len
= super
->next_len
;
6303 super
->buf
= super
->next_buf
;
6305 super
->next_len
= 0;
6306 super
->next_buf
= NULL
;
6309 mpb
= super
->anchor
;
6312 case update_takeover
: {
6313 struct imsm_update_takeover
*u
= (void *)update
->buf
;
6314 if (apply_takeover_update(u
, super
, &update
->space_list
)) {
6315 imsm_update_version_info(super
);
6316 super
->updates_pending
++;
6321 case update_reshape_container_disks
: {
6322 struct imsm_update_reshape
*u
= (void *)update
->buf
;
6323 if (apply_reshape_container_disks_update(
6324 u
, super
, &update
->space_list
))
6325 super
->updates_pending
++;
6328 case update_activate_spare
: {
6329 struct imsm_update_activate_spare
*u
= (void *) update
->buf
;
6330 struct imsm_dev
*dev
= get_imsm_dev(super
, u
->array
);
6331 struct imsm_map
*map
= get_imsm_map(dev
, 0);
6332 struct imsm_map
*migr_map
;
6333 struct active_array
*a
;
6334 struct imsm_disk
*disk
;
6339 int victim
= get_imsm_disk_idx(dev
, u
->slot
, -1);
6342 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
6347 fprintf(stderr
, "error: imsm_activate_spare passed "
6348 "an unknown disk (index: %d)\n",
6353 super
->updates_pending
++;
6355 /* count failures (excluding rebuilds and the victim)
6356 * to determine map[0] state
6359 for (i
= 0; i
< map
->num_members
; i
++) {
6362 disk
= get_imsm_disk(super
,
6363 get_imsm_disk_idx(dev
, i
, -1));
6364 if (!disk
|| is_failed(disk
))
6368 /* adding a pristine spare, assign a new index */
6369 if (dl
->index
< 0) {
6370 dl
->index
= super
->anchor
->num_disks
;
6371 super
->anchor
->num_disks
++;
6374 disk
->status
|= CONFIGURED_DISK
;
6375 disk
->status
&= ~SPARE_DISK
;
6378 to_state
= imsm_check_degraded(super
, dev
, failed
);
6379 map
->map_state
= IMSM_T_STATE_DEGRADED
;
6380 migrate(dev
, to_state
, MIGR_REBUILD
);
6381 migr_map
= get_imsm_map(dev
, 1);
6382 set_imsm_ord_tbl_ent(map
, u
->slot
, dl
->index
);
6383 set_imsm_ord_tbl_ent(migr_map
, u
->slot
, dl
->index
| IMSM_ORD_REBUILD
);
6385 /* update the family_num to mark a new container
6386 * generation, being careful to record the existing
6387 * family_num in orig_family_num to clean up after
6388 * earlier mdadm versions that neglected to set it.
6390 if (mpb
->orig_family_num
== 0)
6391 mpb
->orig_family_num
= mpb
->family_num
;
6392 mpb
->family_num
+= super
->random
;
6394 /* count arrays using the victim in the metadata */
6396 for (a
= st
->arrays
; a
; a
= a
->next
) {
6397 dev
= get_imsm_dev(super
, a
->info
.container_member
);
6398 map
= get_imsm_map(dev
, 0);
6400 if (get_imsm_disk_slot(map
, victim
) >= 0)
6404 /* delete the victim if it is no longer being
6410 /* We know that 'manager' isn't touching anything,
6411 * so it is safe to delete
6413 for (dlp
= &super
->disks
; *dlp
; dlp
= &(*dlp
)->next
)
6414 if ((*dlp
)->index
== victim
)
6417 /* victim may be on the missing list */
6419 for (dlp
= &super
->missing
; *dlp
; dlp
= &(*dlp
)->next
)
6420 if ((*dlp
)->index
== victim
)
6422 imsm_delete(super
, dlp
, victim
);
6426 case update_create_array
: {
6427 /* someone wants to create a new array, we need to be aware of
6428 * a few races/collisions:
6429 * 1/ 'Create' called by two separate instances of mdadm
6430 * 2/ 'Create' versus 'activate_spare': mdadm has chosen
6431 * devices that have since been assimilated via
6433 * In the event this update can not be carried out mdadm will
6434 * (FIX ME) notice that its update did not take hold.
6436 struct imsm_update_create_array
*u
= (void *) update
->buf
;
6437 struct intel_dev
*dv
;
6438 struct imsm_dev
*dev
;
6439 struct imsm_map
*map
, *new_map
;
6440 unsigned long long start
, end
;
6441 unsigned long long new_start
, new_end
;
6443 struct disk_info
*inf
;
6446 /* handle racing creates: first come first serve */
6447 if (u
->dev_idx
< mpb
->num_raid_devs
) {
6448 dprintf("%s: subarray %d already defined\n",
6449 __func__
, u
->dev_idx
);
6453 /* check update is next in sequence */
6454 if (u
->dev_idx
!= mpb
->num_raid_devs
) {
6455 dprintf("%s: can not create array %d expected index %d\n",
6456 __func__
, u
->dev_idx
, mpb
->num_raid_devs
);
6460 new_map
= get_imsm_map(&u
->dev
, 0);
6461 new_start
= __le32_to_cpu(new_map
->pba_of_lba0
);
6462 new_end
= new_start
+ __le32_to_cpu(new_map
->blocks_per_member
);
6463 inf
= get_disk_info(u
);
6465 /* handle activate_spare versus create race:
6466 * check to make sure that overlapping arrays do not include
6469 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
6470 dev
= get_imsm_dev(super
, i
);
6471 map
= get_imsm_map(dev
, 0);
6472 start
= __le32_to_cpu(map
->pba_of_lba0
);
6473 end
= start
+ __le32_to_cpu(map
->blocks_per_member
);
6474 if ((new_start
>= start
&& new_start
<= end
) ||
6475 (start
>= new_start
&& start
<= new_end
))
6480 if (disks_overlap(super
, i
, u
)) {
6481 dprintf("%s: arrays overlap\n", __func__
);
6486 /* check that prepare update was successful */
6487 if (!update
->space
) {
6488 dprintf("%s: prepare update failed\n", __func__
);
6492 /* check that all disks are still active before committing
6493 * changes. FIXME: could we instead handle this by creating a
6494 * degraded array? That's probably not what the user expects,
6495 * so better to drop this update on the floor.
6497 for (i
= 0; i
< new_map
->num_members
; i
++) {
6498 dl
= serial_to_dl(inf
[i
].serial
, super
);
6500 dprintf("%s: disk disappeared\n", __func__
);
6505 super
->updates_pending
++;
6507 /* convert spares to members and fixup ord_tbl */
6508 for (i
= 0; i
< new_map
->num_members
; i
++) {
6509 dl
= serial_to_dl(inf
[i
].serial
, super
);
6510 if (dl
->index
== -1) {
6511 dl
->index
= mpb
->num_disks
;
6513 dl
->disk
.status
|= CONFIGURED_DISK
;
6514 dl
->disk
.status
&= ~SPARE_DISK
;
6516 set_imsm_ord_tbl_ent(new_map
, i
, dl
->index
);
6521 update
->space
= NULL
;
6522 imsm_copy_dev(dev
, &u
->dev
);
6523 dv
->index
= u
->dev_idx
;
6524 dv
->next
= super
->devlist
;
6525 super
->devlist
= dv
;
6526 mpb
->num_raid_devs
++;
6528 imsm_update_version_info(super
);
6531 /* mdmon knows how to release update->space, but not
6532 * ((struct intel_dev *) update->space)->dev
6534 if (update
->space
) {
6540 case update_kill_array
: {
6541 struct imsm_update_kill_array
*u
= (void *) update
->buf
;
6542 int victim
= u
->dev_idx
;
6543 struct active_array
*a
;
6544 struct intel_dev
**dp
;
6545 struct imsm_dev
*dev
;
6547 /* sanity check that we are not affecting the uuid of
6548 * active arrays, or deleting an active array
6550 * FIXME when immutable ids are available, but note that
6551 * we'll also need to fixup the invalidated/active
6552 * subarray indexes in mdstat
6554 for (a
= st
->arrays
; a
; a
= a
->next
)
6555 if (a
->info
.container_member
>= victim
)
6557 /* by definition if mdmon is running at least one array
6558 * is active in the container, so checking
6559 * mpb->num_raid_devs is just extra paranoia
6561 dev
= get_imsm_dev(super
, victim
);
6562 if (a
|| !dev
|| mpb
->num_raid_devs
== 1) {
6563 dprintf("failed to delete subarray-%d\n", victim
);
6567 for (dp
= &super
->devlist
; *dp
;)
6568 if ((*dp
)->index
== (unsigned)super
->current_vol
) {
6571 if ((*dp
)->index
> (unsigned)victim
)
6575 mpb
->num_raid_devs
--;
6576 super
->updates_pending
++;
6579 case update_rename_array
: {
6580 struct imsm_update_rename_array
*u
= (void *) update
->buf
;
6581 char name
[MAX_RAID_SERIAL_LEN
+1];
6582 int target
= u
->dev_idx
;
6583 struct active_array
*a
;
6584 struct imsm_dev
*dev
;
6586 /* sanity check that we are not affecting the uuid of
6589 snprintf(name
, MAX_RAID_SERIAL_LEN
, "%s", (char *) u
->name
);
6590 name
[MAX_RAID_SERIAL_LEN
] = '\0';
6591 for (a
= st
->arrays
; a
; a
= a
->next
)
6592 if (a
->info
.container_member
== target
)
6594 dev
= get_imsm_dev(super
, u
->dev_idx
);
6595 if (a
|| !dev
|| !check_name(super
, name
, 1)) {
6596 dprintf("failed to rename subarray-%d\n", target
);
6600 snprintf((char *) dev
->volume
, MAX_RAID_SERIAL_LEN
, "%s", name
);
6601 super
->updates_pending
++;
6604 case update_add_remove_disk
: {
6605 /* we may be able to repair some arrays if disks are
6606 * being added, check teh status of add_remove_disk
6607 * if discs has been added.
6609 if (add_remove_disk_update(super
)) {
6610 struct active_array
*a
;
6612 super
->updates_pending
++;
6613 for (a
= st
->arrays
; a
; a
= a
->next
)
6614 a
->check_degraded
= 1;
6619 fprintf(stderr
, "error: unsuported process update type:"
6620 "(type: %d)\n", type
);
6624 static void imsm_prepare_update(struct supertype
*st
,
6625 struct metadata_update
*update
)
6628 * Allocate space to hold new disk entries, raid-device entries or a new
6629 * mpb if necessary. The manager synchronously waits for updates to
6630 * complete in the monitor, so new mpb buffers allocated here can be
6631 * integrated by the monitor thread without worrying about live pointers
6632 * in the manager thread.
6634 enum imsm_update_type type
= *(enum imsm_update_type
*) update
->buf
;
6635 struct intel_super
*super
= st
->sb
;
6636 struct imsm_super
*mpb
= super
->anchor
;
6641 case update_takeover
: {
6642 struct imsm_update_takeover
*u
= (void *)update
->buf
;
6643 if (u
->direction
== R0_TO_R10
) {
6644 void **tail
= (void **)&update
->space_list
;
6645 struct imsm_dev
*dev
= get_imsm_dev(super
, u
->subarray
);
6646 struct imsm_map
*map
= get_imsm_map(dev
, 0);
6647 int num_members
= map
->num_members
;
6651 /* allocate memory for added disks */
6652 for (i
= 0; i
< num_members
; i
++) {
6653 size
= sizeof(struct dl
);
6654 space
= malloc(size
);
6663 /* allocate memory for new device */
6664 size
= sizeof_imsm_dev(super
->devlist
->dev
, 0) +
6665 (num_members
* sizeof(__u32
));
6666 space
= malloc(size
);
6675 len
= disks_to_mpb_size(num_members
* 2);
6677 /* if allocation didn't success, free buffer */
6678 while (update
->space_list
) {
6679 void **sp
= update
->space_list
;
6680 update
->space_list
= *sp
;
6688 case update_reshape_container_disks
: {
6689 /* Every raid device in the container is about to
6690 * gain some more devices, and we will enter a
6692 * So each 'imsm_map' will be bigger, and the imsm_vol
6693 * will now hold 2 of them.
6694 * Thus we need new 'struct imsm_dev' allocations sized
6695 * as sizeof_imsm_dev but with more devices in both maps.
6697 struct imsm_update_reshape
*u
= (void *)update
->buf
;
6698 struct intel_dev
*dl
;
6699 void **space_tail
= (void**)&update
->space_list
;
6701 dprintf("imsm: imsm_prepare_update() for update_reshape\n");
6703 for (dl
= super
->devlist
; dl
; dl
= dl
->next
) {
6704 int size
= sizeof_imsm_dev(dl
->dev
, 1);
6706 if (u
->new_raid_disks
> u
->old_raid_disks
)
6707 size
+= sizeof(__u32
)*2*
6708 (u
->new_raid_disks
- u
->old_raid_disks
);
6717 len
= disks_to_mpb_size(u
->new_raid_disks
);
6718 dprintf("New anchor length is %llu\n", (unsigned long long)len
);
6721 case update_create_array
: {
6722 struct imsm_update_create_array
*u
= (void *) update
->buf
;
6723 struct intel_dev
*dv
;
6724 struct imsm_dev
*dev
= &u
->dev
;
6725 struct imsm_map
*map
= get_imsm_map(dev
, 0);
6727 struct disk_info
*inf
;
6731 inf
= get_disk_info(u
);
6732 len
= sizeof_imsm_dev(dev
, 1);
6733 /* allocate a new super->devlist entry */
6734 dv
= malloc(sizeof(*dv
));
6736 dv
->dev
= malloc(len
);
6741 update
->space
= NULL
;
6745 /* count how many spares will be converted to members */
6746 for (i
= 0; i
< map
->num_members
; i
++) {
6747 dl
= serial_to_dl(inf
[i
].serial
, super
);
6749 /* hmm maybe it failed?, nothing we can do about
6754 if (count_memberships(dl
, super
) == 0)
6757 len
+= activate
* sizeof(struct imsm_disk
);
6764 /* check if we need a larger metadata buffer */
6765 if (super
->next_buf
)
6766 buf_len
= super
->next_len
;
6768 buf_len
= super
->len
;
6770 if (__le32_to_cpu(mpb
->mpb_size
) + len
> buf_len
) {
6771 /* ok we need a larger buf than what is currently allocated
6772 * if this allocation fails process_update will notice that
6773 * ->next_len is set and ->next_buf is NULL
6775 buf_len
= ROUND_UP(__le32_to_cpu(mpb
->mpb_size
) + len
, 512);
6776 if (super
->next_buf
)
6777 free(super
->next_buf
);
6779 super
->next_len
= buf_len
;
6780 if (posix_memalign(&super
->next_buf
, 512, buf_len
) == 0)
6781 memset(super
->next_buf
, 0, buf_len
);
6783 super
->next_buf
= NULL
;
6787 /* must be called while manager is quiesced */
6788 static void imsm_delete(struct intel_super
*super
, struct dl
**dlp
, unsigned index
)
6790 struct imsm_super
*mpb
= super
->anchor
;
6792 struct imsm_dev
*dev
;
6793 struct imsm_map
*map
;
6794 int i
, j
, num_members
;
6797 dprintf("%s: deleting device[%d] from imsm_super\n",
6800 /* shift all indexes down one */
6801 for (iter
= super
->disks
; iter
; iter
= iter
->next
)
6802 if (iter
->index
> (int)index
)
6804 for (iter
= super
->missing
; iter
; iter
= iter
->next
)
6805 if (iter
->index
> (int)index
)
6808 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
6809 dev
= get_imsm_dev(super
, i
);
6810 map
= get_imsm_map(dev
, 0);
6811 num_members
= map
->num_members
;
6812 for (j
= 0; j
< num_members
; j
++) {
6813 /* update ord entries being careful not to propagate
6814 * ord-flags to the first map
6816 ord
= get_imsm_ord_tbl_ent(dev
, j
, -1);
6818 if (ord_to_idx(ord
) <= index
)
6821 map
= get_imsm_map(dev
, 0);
6822 set_imsm_ord_tbl_ent(map
, j
, ord_to_idx(ord
- 1));
6823 map
= get_imsm_map(dev
, 1);
6825 set_imsm_ord_tbl_ent(map
, j
, ord
- 1);
6830 super
->updates_pending
++;
6832 struct dl
*dl
= *dlp
;
6834 *dlp
= (*dlp
)->next
;
6835 __free_imsm_disk(dl
);
6839 static char disk_by_path
[] = "/dev/disk/by-path/";
6841 static const char *imsm_get_disk_controller_domain(const char *path
)
6843 char disk_path
[PATH_MAX
];
6847 strncpy(disk_path
, disk_by_path
, PATH_MAX
- 1);
6848 strncat(disk_path
, path
, PATH_MAX
- strlen(disk_path
) - 1);
6849 if (stat(disk_path
, &st
) == 0) {
6850 struct sys_dev
* hba
;
6853 path
= devt_to_devpath(st
.st_rdev
);
6856 hba
= find_disk_attached_hba(-1, path
);
6857 if (hba
&& hba
->type
== SYS_DEV_SAS
)
6859 else if (hba
&& hba
->type
== SYS_DEV_SATA
)
6863 dprintf("path: %s hba: %s attached: %s\n",
6864 path
, (hba
) ? hba
->path
: "NULL", drv
);
6872 static int imsm_find_array_minor_by_subdev(int subdev
, int container
, int *minor
)
6874 char subdev_name
[20];
6875 struct mdstat_ent
*mdstat
;
6877 sprintf(subdev_name
, "%d", subdev
);
6878 mdstat
= mdstat_by_subdev(subdev_name
, container
);
6882 *minor
= mdstat
->devnum
;
6883 free_mdstat(mdstat
);
6887 static int imsm_reshape_is_allowed_on_container(struct supertype
*st
,
6888 struct geo_params
*geo
,
6889 int *old_raid_disks
)
6891 /* currently we only support increasing the number of devices
6892 * for a container. This increases the number of device for each
6893 * member array. They must all be RAID0 or RAID5.
6896 struct mdinfo
*info
, *member
;
6897 int devices_that_can_grow
= 0;
6899 dprintf("imsm: imsm_reshape_is_allowed_on_container(ENTER): "
6900 "st->devnum = (%i)\n",
6903 if (geo
->size
!= -1 ||
6904 geo
->level
!= UnSet
||
6905 geo
->layout
!= UnSet
||
6906 geo
->chunksize
!= 0 ||
6907 geo
->raid_disks
== UnSet
) {
6908 dprintf("imsm: Container operation is allowed for "
6909 "raid disks number change only.\n");
6913 info
= container_content_imsm(st
, NULL
);
6914 for (member
= info
; member
; member
= member
->next
) {
6918 dprintf("imsm: checking device_num: %i\n",
6919 member
->container_member
);
6921 if (geo
->raid_disks
<= member
->array
.raid_disks
) {
6922 /* we work on container for Online Capacity Expansion
6923 * only so raid_disks has to grow
6925 dprintf("imsm: for container operation raid disks "
6926 "increase is required\n");
6930 if ((info
->array
.level
!= 0) &&
6931 (info
->array
.level
!= 5)) {
6932 /* we cannot use this container with other raid level
6934 dprintf("imsm: for container operation wrong"
6935 " raid level (%i) detected\n",
6939 /* check for platform support
6940 * for this raid level configuration
6942 struct intel_super
*super
= st
->sb
;
6943 if (!is_raid_level_supported(super
->orom
,
6944 member
->array
.level
,
6946 dprintf("platform does not support raid%d with"
6950 geo
->raid_disks
> 1 ? "s" : "");
6955 if (*old_raid_disks
&&
6956 info
->array
.raid_disks
!= *old_raid_disks
)
6958 *old_raid_disks
= info
->array
.raid_disks
;
6960 /* All raid5 and raid0 volumes in container
6961 * have to be ready for Online Capacity Expansion
6962 * so they need to be assembled. We have already
6963 * checked that no recovery etc is happening.
6965 result
= imsm_find_array_minor_by_subdev(member
->container_member
,
6969 dprintf("imsm: cannot find array\n");
6972 devices_that_can_grow
++;
6975 if (!member
&& devices_that_can_grow
)
6979 dprintf("\tContainer operation allowed\n");
6981 dprintf("\tError: %i\n", ret_val
);
6986 /* Function: get_spares_for_grow
6987 * Description: Allocates memory and creates list of spare devices
6988 * avaliable in container. Checks if spare drive size is acceptable.
6989 * Parameters: Pointer to the supertype structure
6990 * Returns: Pointer to the list of spare devices (mdinfo structure) on success,
6993 static struct mdinfo
*get_spares_for_grow(struct supertype
*st
)
6995 unsigned long long min_size
= min_acceptable_spare_size_imsm(st
);
6996 return container_choose_spares(st
, min_size
, NULL
, NULL
, NULL
, 0);
6999 /******************************************************************************
7000 * function: imsm_create_metadata_update_for_reshape
7001 * Function creates update for whole IMSM container.
7003 ******************************************************************************/
7004 static int imsm_create_metadata_update_for_reshape(
7005 struct supertype
*st
,
7006 struct geo_params
*geo
,
7008 struct imsm_update_reshape
**updatep
)
7010 struct intel_super
*super
= st
->sb
;
7011 struct imsm_super
*mpb
= super
->anchor
;
7012 int update_memory_size
= 0;
7013 struct imsm_update_reshape
*u
= NULL
;
7014 struct mdinfo
*spares
= NULL
;
7016 int delta_disks
= 0;
7019 dprintf("imsm_update_metadata_for_reshape(enter) raid_disks = %i\n",
7022 delta_disks
= geo
->raid_disks
- old_raid_disks
;
7024 /* size of all update data without anchor */
7025 update_memory_size
= sizeof(struct imsm_update_reshape
);
7027 /* now add space for spare disks that we need to add. */
7028 update_memory_size
+= sizeof(u
->new_disks
[0]) * (delta_disks
- 1);
7030 u
= calloc(1, update_memory_size
);
7033 "cannot get memory for imsm_update_reshape update\n");
7036 u
->type
= update_reshape_container_disks
;
7037 u
->old_raid_disks
= old_raid_disks
;
7038 u
->new_raid_disks
= geo
->raid_disks
;
7040 /* now get spare disks list
7042 spares
= get_spares_for_grow(st
);
7045 || delta_disks
> spares
->array
.spare_disks
) {
7046 fprintf(stderr
, Name
": imsm: ERROR: Cannot get spare devices "
7047 "for %s.\n", geo
->dev_name
);
7051 /* we have got spares
7052 * update disk list in imsm_disk list table in anchor
7054 dprintf("imsm: %i spares are available.\n\n",
7055 spares
->array
.spare_disks
);
7058 for (i
= 0; i
< delta_disks
; i
++) {
7063 u
->new_disks
[i
] = makedev(dev
->disk
.major
,
7065 dl
= get_disk_super(super
, dev
->disk
.major
, dev
->disk
.minor
);
7066 dl
->index
= mpb
->num_disks
;
7076 dprintf("imsm: reshape update preparation :");
7077 if (i
== delta_disks
) {
7080 return update_memory_size
;
7083 dprintf(" Error\n");
7088 static void imsm_update_metadata_locally(struct supertype
*st
,
7091 struct metadata_update mu
;
7096 mu
.space_list
= NULL
;
7098 imsm_prepare_update(st
, &mu
);
7099 imsm_process_update(st
, &mu
);
7101 while (mu
.space_list
) {
7102 void **space
= mu
.space_list
;
7103 mu
.space_list
= *space
;
7108 /***************************************************************************
7109 * Function: imsm_analyze_change
7110 * Description: Function analyze change for single volume
7111 * and validate if transition is supported
7112 * Parameters: Geometry parameters, supertype structure
7113 * Returns: Operation type code on success, -1 if fail
7114 ****************************************************************************/
7115 enum imsm_reshape_type
imsm_analyze_change(struct supertype
*st
,
7116 struct geo_params
*geo
)
7123 getinfo_super_imsm_volume(st
, &info
, NULL
);
7125 if ((geo
->level
!= info
.array
.level
) &&
7126 (geo
->level
>= 0) &&
7127 (geo
->level
!= UnSet
)) {
7128 switch (info
.array
.level
) {
7130 if (geo
->level
== 5) {
7131 change
= CH_MIGRATION
;
7134 if (geo
->level
== 10) {
7135 change
= CH_TAKEOVER
;
7140 if (geo
->level
== 0) {
7141 change
= CH_TAKEOVER
;
7146 if (geo
->level
== 0)
7147 change
= CH_MIGRATION
;
7150 if (geo
->level
== 0) {
7151 change
= CH_TAKEOVER
;
7158 Name
" Error. Level Migration from %d to %d "
7160 info
.array
.level
, geo
->level
);
7161 goto analyse_change_exit
;
7164 geo
->level
= info
.array
.level
;
7166 if ((geo
->layout
!= info
.array
.layout
)
7167 && ((geo
->layout
!= UnSet
) && (geo
->layout
!= -1))) {
7168 change
= CH_MIGRATION
;
7169 if ((info
.array
.layout
== 0)
7170 && (info
.array
.level
== 5)
7171 && (geo
->layout
== 5)) {
7172 /* reshape 5 -> 4 */
7173 } else if ((info
.array
.layout
== 5)
7174 && (info
.array
.level
== 5)
7175 && (geo
->layout
== 0)) {
7176 /* reshape 4 -> 5 */
7181 Name
" Error. Layout Migration from %d to %d "
7183 info
.array
.layout
, geo
->layout
);
7185 goto analyse_change_exit
;
7188 geo
->layout
= info
.array
.layout
;
7190 if ((geo
->chunksize
> 0) && (geo
->chunksize
!= UnSet
)
7191 && (geo
->chunksize
!= info
.array
.chunk_size
))
7192 change
= CH_MIGRATION
;
7194 geo
->chunksize
= info
.array
.chunk_size
;
7196 chunk
= geo
->chunksize
/ 1024;
7197 if (!validate_geometry_imsm(st
,
7207 struct intel_super
*super
= st
->sb
;
7208 struct imsm_super
*mpb
= super
->anchor
;
7210 if (mpb
->num_raid_devs
> 1) {
7212 Name
" Error. Cannot perform operation on %s"
7213 "- for this operation it MUST be single "
7214 "array in container\n",
7220 analyse_change_exit
:
7225 int imsm_takeover(struct supertype
*st
, struct geo_params
*geo
)
7227 struct intel_super
*super
= st
->sb
;
7228 struct imsm_update_takeover
*u
;
7230 u
= malloc(sizeof(struct imsm_update_takeover
));
7234 u
->type
= update_takeover
;
7235 u
->subarray
= super
->current_vol
;
7237 /* 10->0 transition */
7238 if (geo
->level
== 0)
7239 u
->direction
= R10_TO_R0
;
7241 /* 0->10 transition */
7242 if (geo
->level
== 10)
7243 u
->direction
= R0_TO_R10
;
7245 /* update metadata locally */
7246 imsm_update_metadata_locally(st
, u
,
7247 sizeof(struct imsm_update_takeover
));
7248 /* and possibly remotely */
7249 if (st
->update_tail
)
7250 append_metadata_update(st
, u
,
7251 sizeof(struct imsm_update_takeover
));
7258 static int imsm_reshape_super(struct supertype
*st
, long long size
, int level
,
7259 int layout
, int chunksize
, int raid_disks
,
7260 int delta_disks
, char *backup
, char *dev
,
7264 struct geo_params geo
;
7266 dprintf("imsm: reshape_super called.\n");
7268 memset(&geo
, 0, sizeof(struct geo_params
));
7271 geo
.dev_id
= st
->devnum
;
7274 geo
.layout
= layout
;
7275 geo
.chunksize
= chunksize
;
7276 geo
.raid_disks
= raid_disks
;
7277 if (delta_disks
!= UnSet
)
7278 geo
.raid_disks
+= delta_disks
;
7280 dprintf("\tfor level : %i\n", geo
.level
);
7281 dprintf("\tfor raid_disks : %i\n", geo
.raid_disks
);
7283 if (experimental() == 0)
7286 if (st
->container_dev
== st
->devnum
) {
7287 /* On container level we can only increase number of devices. */
7288 dprintf("imsm: info: Container operation\n");
7289 int old_raid_disks
= 0;
7290 if (imsm_reshape_is_allowed_on_container(
7291 st
, &geo
, &old_raid_disks
)) {
7292 struct imsm_update_reshape
*u
= NULL
;
7295 len
= imsm_create_metadata_update_for_reshape(
7296 st
, &geo
, old_raid_disks
, &u
);
7299 dprintf("imsm: Cannot prepare update\n");
7300 goto exit_imsm_reshape_super
;
7304 /* update metadata locally */
7305 imsm_update_metadata_locally(st
, u
, len
);
7306 /* and possibly remotely */
7307 if (st
->update_tail
)
7308 append_metadata_update(st
, u
, len
);
7313 fprintf(stderr
, Name
": (imsm) Operation "
7314 "is not allowed on this container\n");
7317 /* On volume level we support following operations
7318 * - takeover: raid10 -> raid0; raid0 -> raid10
7319 * - chunk size migration
7320 * - migration: raid5 -> raid0; raid0 -> raid5
7322 struct intel_super
*super
= st
->sb
;
7323 struct intel_dev
*dev
= super
->devlist
;
7325 dprintf("imsm: info: Volume operation\n");
7326 /* find requested device */
7328 imsm_find_array_minor_by_subdev(dev
->index
, st
->container_dev
, &devnum
);
7329 if (devnum
== geo
.dev_id
)
7334 fprintf(stderr
, Name
" Cannot find %s (%i) subarray\n",
7335 geo
.dev_name
, geo
.dev_id
);
7336 goto exit_imsm_reshape_super
;
7338 super
->current_vol
= dev
->index
;
7339 change
= imsm_analyze_change(st
, &geo
);
7342 ret_val
= imsm_takeover(st
, &geo
);
7352 exit_imsm_reshape_super
:
7353 dprintf("imsm: reshape_super Exit code = %i\n", ret_val
);
7357 static int imsm_manage_reshape(
7358 int afd
, struct mdinfo
*sra
, struct reshape
*reshape
,
7359 struct supertype
*st
, unsigned long stripes
,
7360 int *fds
, unsigned long long *offsets
,
7361 int dests
, int *destfd
, unsigned long long *destoffsets
)
7363 /* Just use child_monitor for now */
7364 return child_monitor(
7365 afd
, sra
, reshape
, st
, stripes
,
7366 fds
, offsets
, dests
, destfd
, destoffsets
);
7368 #endif /* MDASSEMBLE */
7370 struct superswitch super_imsm
= {
7372 .examine_super
= examine_super_imsm
,
7373 .brief_examine_super
= brief_examine_super_imsm
,
7374 .brief_examine_subarrays
= brief_examine_subarrays_imsm
,
7375 .export_examine_super
= export_examine_super_imsm
,
7376 .detail_super
= detail_super_imsm
,
7377 .brief_detail_super
= brief_detail_super_imsm
,
7378 .write_init_super
= write_init_super_imsm
,
7379 .validate_geometry
= validate_geometry_imsm
,
7380 .add_to_super
= add_to_super_imsm
,
7381 .remove_from_super
= remove_from_super_imsm
,
7382 .detail_platform
= detail_platform_imsm
,
7383 .kill_subarray
= kill_subarray_imsm
,
7384 .update_subarray
= update_subarray_imsm
,
7385 .load_container
= load_container_imsm
,
7386 .default_geometry
= default_geometry_imsm
,
7387 .get_disk_controller_domain
= imsm_get_disk_controller_domain
,
7388 .reshape_super
= imsm_reshape_super
,
7389 .manage_reshape
= imsm_manage_reshape
,
7391 .match_home
= match_home_imsm
,
7392 .uuid_from_super
= uuid_from_super_imsm
,
7393 .getinfo_super
= getinfo_super_imsm
,
7394 .getinfo_super_disks
= getinfo_super_disks_imsm
,
7395 .update_super
= update_super_imsm
,
7397 .avail_size
= avail_size_imsm
,
7398 .min_acceptable_spare_size
= min_acceptable_spare_size_imsm
,
7400 .compare_super
= compare_super_imsm
,
7402 .load_super
= load_super_imsm
,
7403 .init_super
= init_super_imsm
,
7404 .store_super
= store_super_imsm
,
7405 .free_super
= free_super_imsm
,
7406 .match_metadata_desc
= match_metadata_desc_imsm
,
7407 .container_content
= container_content_imsm
,
7414 .open_new
= imsm_open_new
,
7415 .set_array_state
= imsm_set_array_state
,
7416 .set_disk
= imsm_set_disk
,
7417 .sync_metadata
= imsm_sync_metadata
,
7418 .activate_spare
= imsm_activate_spare
,
7419 .process_update
= imsm_process_update
,
7420 .prepare_update
= imsm_prepare_update
,
7421 #endif /* MDASSEMBLE */