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
{
301 /* definition of messages passed to imsm_process_update */
302 enum imsm_update_type
{
303 update_activate_spare
,
307 update_add_remove_disk
,
308 update_reshape_container_disks
,
312 struct imsm_update_activate_spare
{
313 enum imsm_update_type type
;
317 struct imsm_update_activate_spare
*next
;
330 enum takeover_direction
{
334 struct imsm_update_takeover
{
335 enum imsm_update_type type
;
337 enum takeover_direction direction
;
340 struct imsm_update_reshape
{
341 enum imsm_update_type type
;
344 int new_disks
[1]; /* new_raid_disks - old_raid_disks makedev number */
348 __u8 serial
[MAX_RAID_SERIAL_LEN
];
351 struct imsm_update_create_array
{
352 enum imsm_update_type type
;
357 struct imsm_update_kill_array
{
358 enum imsm_update_type type
;
362 struct imsm_update_rename_array
{
363 enum imsm_update_type type
;
364 __u8 name
[MAX_RAID_SERIAL_LEN
];
368 struct imsm_update_add_remove_disk
{
369 enum imsm_update_type type
;
373 static const char *_sys_dev_type
[] = {
374 [SYS_DEV_UNKNOWN
] = "Unknown",
375 [SYS_DEV_SAS
] = "SAS",
376 [SYS_DEV_SATA
] = "SATA"
379 const char *get_sys_dev_type(enum sys_dev_type type
)
381 if (type
>= SYS_DEV_MAX
)
382 type
= SYS_DEV_UNKNOWN
;
384 return _sys_dev_type
[type
];
388 static struct intel_hba
* alloc_intel_hba(struct sys_dev
*device
)
390 struct intel_hba
*result
= malloc(sizeof(*result
));
392 result
->type
= device
->type
;
393 result
->path
= strdup(device
->path
);
395 if (result
->path
&& (result
->pci_id
= strrchr(result
->path
, '/')) != NULL
)
401 static struct intel_hba
* find_intel_hba(struct intel_hba
*hba
, struct sys_dev
*device
)
403 struct intel_hba
*result
=NULL
;
404 for (result
= hba
; result
; result
= result
->next
) {
405 if (result
->type
== device
->type
&& strcmp(result
->path
, device
->path
) == 0)
412 static int attach_hba_to_super(struct intel_super
*super
, struct sys_dev
*device
,
415 struct intel_hba
*hba
;
417 /* check if disk attached to Intel HBA */
418 hba
= find_intel_hba(super
->hba
, device
);
421 /* Check if HBA is already attached to super */
422 if (super
->hba
== NULL
) {
423 super
->hba
= alloc_intel_hba(device
);
428 /* Intel metadata allows for all disks attached to the same type HBA.
429 * Do not sypport odf HBA types mixing
431 if (device
->type
!= hba
->type
)
437 hba
->next
= alloc_intel_hba(device
);
441 static struct sys_dev
* find_disk_attached_hba(int fd
, const char *devname
)
443 struct sys_dev
*list
, *elem
, *prev
;
446 if ((list
= find_intel_devices()) == NULL
)
450 disk_path
= (char *) devname
;
452 disk_path
= diskfd_to_devpath(fd
);
459 for (prev
= NULL
, elem
= list
; elem
; prev
= elem
, elem
= elem
->next
) {
460 if (path_attached_to_hba(disk_path
, elem
->path
)) {
464 prev
->next
= elem
->next
;
466 if (disk_path
!= devname
)
472 if (disk_path
!= devname
)
478 #endif /* MDASSEMBLE */
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 slot
= get_imsm_disk_slot(map
, disk_idx
);
903 ord
= get_imsm_ord_tbl_ent(dev
, slot
, -1);
904 printf(" This Slot : %d%s\n", slot
,
905 ord
& IMSM_ORD_REBUILD
? " (out-of-sync)" : "");
907 printf(" This Slot : ?\n");
908 sz
= __le32_to_cpu(dev
->size_high
);
910 sz
+= __le32_to_cpu(dev
->size_low
);
911 printf(" Array Size : %llu%s\n", (unsigned long long)sz
,
912 human_size(sz
* 512));
913 sz
= __le32_to_cpu(map
->blocks_per_member
);
914 printf(" Per Dev Size : %llu%s\n", (unsigned long long)sz
,
915 human_size(sz
* 512));
916 printf(" Sector Offset : %u\n",
917 __le32_to_cpu(map
->pba_of_lba0
));
918 printf(" Num Stripes : %u\n",
919 __le32_to_cpu(map
->num_data_stripes
));
920 printf(" Chunk Size : %u KiB",
921 __le16_to_cpu(map
->blocks_per_strip
) / 2);
923 printf(" <-- %u KiB",
924 __le16_to_cpu(map2
->blocks_per_strip
) / 2);
926 printf(" Reserved : %d\n", __le32_to_cpu(dev
->reserved_blocks
));
927 printf(" Migrate State : ");
928 if (dev
->vol
.migr_state
) {
929 if (migr_type(dev
) == MIGR_INIT
)
930 printf("initialize\n");
931 else if (migr_type(dev
) == MIGR_REBUILD
)
933 else if (migr_type(dev
) == MIGR_VERIFY
)
935 else if (migr_type(dev
) == MIGR_GEN_MIGR
)
936 printf("general migration\n");
937 else if (migr_type(dev
) == MIGR_STATE_CHANGE
)
938 printf("state change\n");
939 else if (migr_type(dev
) == MIGR_REPAIR
)
942 printf("<unknown:%d>\n", migr_type(dev
));
945 printf(" Map State : %s", map_state_str
[map
->map_state
]);
946 if (dev
->vol
.migr_state
) {
947 struct imsm_map
*map
= get_imsm_map(dev
, 1);
949 printf(" <-- %s", map_state_str
[map
->map_state
]);
950 printf("\n Checkpoint : %u (%llu)",
951 __le32_to_cpu(dev
->vol
.curr_migr_unit
),
952 (unsigned long long)blocks_per_migr_unit(dev
));
955 printf(" Dirty State : %s\n", dev
->vol
.dirty
? "dirty" : "clean");
958 static void print_imsm_disk(struct imsm_super
*mpb
, int index
, __u32 reserved
)
960 struct imsm_disk
*disk
= __get_imsm_disk(mpb
, index
);
961 char str
[MAX_RAID_SERIAL_LEN
+ 1];
964 if (index
< 0 || !disk
)
968 snprintf(str
, MAX_RAID_SERIAL_LEN
+ 1, "%s", disk
->serial
);
969 printf(" Disk%02d Serial : %s\n", index
, str
);
970 printf(" State :%s%s%s\n", is_spare(disk
) ? " spare" : "",
971 is_configured(disk
) ? " active" : "",
972 is_failed(disk
) ? " failed" : "");
973 printf(" Id : %08x\n", __le32_to_cpu(disk
->scsi_id
));
974 sz
= __le32_to_cpu(disk
->total_blocks
) - reserved
;
975 printf(" Usable Size : %llu%s\n", (unsigned long long)sz
,
976 human_size(sz
* 512));
979 static void getinfo_super_imsm(struct supertype
*st
, struct mdinfo
*info
, char *map
);
981 static void examine_super_imsm(struct supertype
*st
, char *homehost
)
983 struct intel_super
*super
= st
->sb
;
984 struct imsm_super
*mpb
= super
->anchor
;
985 char str
[MAX_SIGNATURE_LENGTH
];
990 __u32 reserved
= imsm_reserved_sectors(super
, super
->disks
);
993 snprintf(str
, MPB_SIG_LEN
, "%s", mpb
->sig
);
994 printf(" Magic : %s\n", str
);
995 snprintf(str
, strlen(MPB_VERSION_RAID0
), "%s", get_imsm_version(mpb
));
996 printf(" Version : %s\n", get_imsm_version(mpb
));
997 printf(" Orig Family : %08x\n", __le32_to_cpu(mpb
->orig_family_num
));
998 printf(" Family : %08x\n", __le32_to_cpu(mpb
->family_num
));
999 printf(" Generation : %08x\n", __le32_to_cpu(mpb
->generation_num
));
1000 getinfo_super_imsm(st
, &info
, NULL
);
1001 fname_from_uuid(st
, &info
, nbuf
, ':');
1002 printf(" UUID : %s\n", nbuf
+ 5);
1003 sum
= __le32_to_cpu(mpb
->check_sum
);
1004 printf(" Checksum : %08x %s\n", sum
,
1005 __gen_imsm_checksum(mpb
) == sum
? "correct" : "incorrect");
1006 printf(" MPB Sectors : %d\n", mpb_sectors(mpb
));
1007 printf(" Disks : %d\n", mpb
->num_disks
);
1008 printf(" RAID Devices : %d\n", mpb
->num_raid_devs
);
1009 print_imsm_disk(mpb
, super
->disks
->index
, reserved
);
1010 if (super
->bbm_log
) {
1011 struct bbm_log
*log
= super
->bbm_log
;
1014 printf("Bad Block Management Log:\n");
1015 printf(" Log Size : %d\n", __le32_to_cpu(mpb
->bbm_log_size
));
1016 printf(" Signature : %x\n", __le32_to_cpu(log
->signature
));
1017 printf(" Entry Count : %d\n", __le32_to_cpu(log
->entry_count
));
1018 printf(" Spare Blocks : %d\n", __le32_to_cpu(log
->reserved_spare_block_count
));
1019 printf(" First Spare : %llx\n",
1020 (unsigned long long) __le64_to_cpu(log
->first_spare_lba
));
1022 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
1024 struct imsm_dev
*dev
= __get_imsm_dev(mpb
, i
);
1026 super
->current_vol
= i
;
1027 getinfo_super_imsm(st
, &info
, NULL
);
1028 fname_from_uuid(st
, &info
, nbuf
, ':');
1029 print_imsm_dev(dev
, nbuf
+ 5, super
->disks
->index
);
1031 for (i
= 0; i
< mpb
->num_disks
; i
++) {
1032 if (i
== super
->disks
->index
)
1034 print_imsm_disk(mpb
, i
, reserved
);
1036 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
1037 struct imsm_disk
*disk
;
1038 char str
[MAX_RAID_SERIAL_LEN
+ 1];
1046 snprintf(str
, MAX_RAID_SERIAL_LEN
+ 1, "%s", disk
->serial
);
1047 printf(" Disk Serial : %s\n", str
);
1048 printf(" State :%s%s%s\n", is_spare(disk
) ? " spare" : "",
1049 is_configured(disk
) ? " active" : "",
1050 is_failed(disk
) ? " failed" : "");
1051 printf(" Id : %08x\n", __le32_to_cpu(disk
->scsi_id
));
1052 sz
= __le32_to_cpu(disk
->total_blocks
) - reserved
;
1053 printf(" Usable Size : %llu%s\n", (unsigned long long)sz
,
1054 human_size(sz
* 512));
1058 static void brief_examine_super_imsm(struct supertype
*st
, int verbose
)
1060 /* We just write a generic IMSM ARRAY entry */
1063 struct intel_super
*super
= st
->sb
;
1065 if (!super
->anchor
->num_raid_devs
) {
1066 printf("ARRAY metadata=imsm\n");
1070 getinfo_super_imsm(st
, &info
, NULL
);
1071 fname_from_uuid(st
, &info
, nbuf
, ':');
1072 printf("ARRAY metadata=imsm UUID=%s\n", nbuf
+ 5);
1075 static void brief_examine_subarrays_imsm(struct supertype
*st
, int verbose
)
1077 /* We just write a generic IMSM ARRAY entry */
1081 struct intel_super
*super
= st
->sb
;
1084 if (!super
->anchor
->num_raid_devs
)
1087 getinfo_super_imsm(st
, &info
, NULL
);
1088 fname_from_uuid(st
, &info
, nbuf
, ':');
1089 for (i
= 0; i
< super
->anchor
->num_raid_devs
; i
++) {
1090 struct imsm_dev
*dev
= get_imsm_dev(super
, i
);
1092 super
->current_vol
= i
;
1093 getinfo_super_imsm(st
, &info
, NULL
);
1094 fname_from_uuid(st
, &info
, nbuf1
, ':');
1095 printf("ARRAY /dev/md/%.16s container=%s member=%d UUID=%s\n",
1096 dev
->volume
, nbuf
+ 5, i
, nbuf1
+ 5);
1100 static void export_examine_super_imsm(struct supertype
*st
)
1102 struct intel_super
*super
= st
->sb
;
1103 struct imsm_super
*mpb
= super
->anchor
;
1107 getinfo_super_imsm(st
, &info
, NULL
);
1108 fname_from_uuid(st
, &info
, nbuf
, ':');
1109 printf("MD_METADATA=imsm\n");
1110 printf("MD_LEVEL=container\n");
1111 printf("MD_UUID=%s\n", nbuf
+5);
1112 printf("MD_DEVICES=%u\n", mpb
->num_disks
);
1115 static void detail_super_imsm(struct supertype
*st
, char *homehost
)
1120 getinfo_super_imsm(st
, &info
, NULL
);
1121 fname_from_uuid(st
, &info
, nbuf
, ':');
1122 printf("\n UUID : %s\n", nbuf
+ 5);
1125 static void brief_detail_super_imsm(struct supertype
*st
)
1129 getinfo_super_imsm(st
, &info
, NULL
);
1130 fname_from_uuid(st
, &info
, nbuf
, ':');
1131 printf(" UUID=%s", nbuf
+ 5);
1134 static int imsm_read_serial(int fd
, char *devname
, __u8
*serial
);
1135 static void fd2devname(int fd
, char *name
);
1137 static int ahci_enumerate_ports(const char *hba_path
, int port_count
, int host_base
, int verbose
)
1139 /* dump an unsorted list of devices attached to AHCI Intel storage
1140 * controller, as well as non-connected ports
1142 int hba_len
= strlen(hba_path
) + 1;
1147 unsigned long port_mask
= (1 << port_count
) - 1;
1149 if (port_count
> (int)sizeof(port_mask
) * 8) {
1151 fprintf(stderr
, Name
": port_count %d out of range\n", port_count
);
1155 /* scroll through /sys/dev/block looking for devices attached to
1158 dir
= opendir("/sys/dev/block");
1159 for (ent
= dir
? readdir(dir
) : NULL
; ent
; ent
= readdir(dir
)) {
1170 if (sscanf(ent
->d_name
, "%d:%d", &major
, &minor
) != 2)
1172 path
= devt_to_devpath(makedev(major
, minor
));
1175 if (!path_attached_to_hba(path
, hba_path
)) {
1181 /* retrieve the scsi device type */
1182 if (asprintf(&device
, "/sys/dev/block/%d:%d/device/xxxxxxx", major
, minor
) < 0) {
1184 fprintf(stderr
, Name
": failed to allocate 'device'\n");
1188 sprintf(device
, "/sys/dev/block/%d:%d/device/type", major
, minor
);
1189 if (load_sys(device
, buf
) != 0) {
1191 fprintf(stderr
, Name
": failed to read device type for %s\n",
1197 type
= strtoul(buf
, NULL
, 10);
1199 /* if it's not a disk print the vendor and model */
1200 if (!(type
== 0 || type
== 7 || type
== 14)) {
1203 sprintf(device
, "/sys/dev/block/%d:%d/device/vendor", major
, minor
);
1204 if (load_sys(device
, buf
) == 0) {
1205 strncpy(vendor
, buf
, sizeof(vendor
));
1206 vendor
[sizeof(vendor
) - 1] = '\0';
1207 c
= (char *) &vendor
[sizeof(vendor
) - 1];
1208 while (isspace(*c
) || *c
== '\0')
1212 sprintf(device
, "/sys/dev/block/%d:%d/device/model", major
, minor
);
1213 if (load_sys(device
, buf
) == 0) {
1214 strncpy(model
, buf
, sizeof(model
));
1215 model
[sizeof(model
) - 1] = '\0';
1216 c
= (char *) &model
[sizeof(model
) - 1];
1217 while (isspace(*c
) || *c
== '\0')
1221 if (vendor
[0] && model
[0])
1222 sprintf(buf
, "%.64s %.64s", vendor
, model
);
1224 switch (type
) { /* numbers from hald/linux/device.c */
1225 case 1: sprintf(buf
, "tape"); break;
1226 case 2: sprintf(buf
, "printer"); break;
1227 case 3: sprintf(buf
, "processor"); break;
1229 case 5: sprintf(buf
, "cdrom"); break;
1230 case 6: sprintf(buf
, "scanner"); break;
1231 case 8: sprintf(buf
, "media_changer"); break;
1232 case 9: sprintf(buf
, "comm"); break;
1233 case 12: sprintf(buf
, "raid"); break;
1234 default: sprintf(buf
, "unknown");
1240 /* chop device path to 'host%d' and calculate the port number */
1241 c
= strchr(&path
[hba_len
], '/');
1244 fprintf(stderr
, Name
": %s - invalid path name\n", path
+ hba_len
);
1249 if (sscanf(&path
[hba_len
], "host%d", &port
) == 1)
1253 *c
= '/'; /* repair the full string */
1254 fprintf(stderr
, Name
": failed to determine port number for %s\n",
1261 /* mark this port as used */
1262 port_mask
&= ~(1 << port
);
1264 /* print out the device information */
1266 printf(" Port%d : - non-disk device (%s) -\n", port
, buf
);
1270 fd
= dev_open(ent
->d_name
, O_RDONLY
);
1272 printf(" Port%d : - disk info unavailable -\n", port
);
1274 fd2devname(fd
, buf
);
1275 printf(" Port%d : %s", port
, buf
);
1276 if (imsm_read_serial(fd
, NULL
, (__u8
*) buf
) == 0)
1277 printf(" (%s)\n", buf
);
1292 for (i
= 0; i
< port_count
; i
++)
1293 if (port_mask
& (1 << i
))
1294 printf(" Port%d : - no device attached -\n", i
);
1302 static void print_found_intel_controllers(struct sys_dev
*elem
)
1304 for (; elem
; elem
= elem
->next
) {
1305 fprintf(stderr
, Name
": found Intel(R) ");
1306 if (elem
->type
== SYS_DEV_SATA
)
1307 fprintf(stderr
, "SATA ");
1308 else if (elem
->type
== SYS_DEV_SAS
)
1309 fprintf(stderr
, "SAS ");
1310 fprintf(stderr
, "RAID controller");
1312 fprintf(stderr
, " at %s", elem
->pci_id
);
1313 fprintf(stderr
, ".\n");
1318 static int ahci_get_port_count(const char *hba_path
, int *port_count
)
1325 if ((dir
= opendir(hba_path
)) == NULL
)
1328 for (ent
= readdir(dir
); ent
; ent
= readdir(dir
)) {
1331 if (sscanf(ent
->d_name
, "host%d", &host
) != 1)
1333 if (*port_count
== 0)
1335 else if (host
< host_base
)
1338 if (host
+ 1 > *port_count
+ host_base
)
1339 *port_count
= host
+ 1 - host_base
;
1345 static int detail_platform_imsm(int verbose
, int enumerate_only
)
1347 /* There are two components to imsm platform support, the ahci SATA
1348 * controller and the option-rom. To find the SATA controller we
1349 * simply look in /sys/bus/pci/drivers/ahci to see if an ahci
1350 * controller with the Intel vendor id is present. This approach
1351 * allows mdadm to leverage the kernel's ahci detection logic, with the
1352 * caveat that if ahci.ko is not loaded mdadm will not be able to
1353 * detect platform raid capabilities. The option-rom resides in a
1354 * platform "Adapter ROM". We scan for its signature to retrieve the
1355 * platform capabilities. If raid support is disabled in the BIOS the
1356 * option-rom capability structure will not be available.
1358 const struct imsm_orom
*orom
;
1359 struct sys_dev
*list
, *hba
;
1364 if (enumerate_only
) {
1365 if (check_env("IMSM_NO_PLATFORM") || find_imsm_orom())
1370 list
= find_intel_devices();
1373 fprintf(stderr
, Name
": no active Intel(R) RAID "
1374 "controller found.\n");
1375 free_sys_dev(&list
);
1378 print_found_intel_controllers(list
);
1380 orom
= find_imsm_orom();
1382 free_sys_dev(&list
);
1384 fprintf(stderr
, Name
": imsm option-rom not found\n");
1388 printf(" Platform : Intel(R) Matrix Storage Manager\n");
1389 printf(" Version : %d.%d.%d.%d\n", orom
->major_ver
, orom
->minor_ver
,
1390 orom
->hotfix_ver
, orom
->build
);
1391 printf(" RAID Levels :%s%s%s%s%s\n",
1392 imsm_orom_has_raid0(orom
) ? " raid0" : "",
1393 imsm_orom_has_raid1(orom
) ? " raid1" : "",
1394 imsm_orom_has_raid1e(orom
) ? " raid1e" : "",
1395 imsm_orom_has_raid10(orom
) ? " raid10" : "",
1396 imsm_orom_has_raid5(orom
) ? " raid5" : "");
1397 printf(" Chunk Sizes :%s%s%s%s%s%s%s%s%s%s%s%s%s%s%s%s\n",
1398 imsm_orom_has_chunk(orom
, 2) ? " 2k" : "",
1399 imsm_orom_has_chunk(orom
, 4) ? " 4k" : "",
1400 imsm_orom_has_chunk(orom
, 8) ? " 8k" : "",
1401 imsm_orom_has_chunk(orom
, 16) ? " 16k" : "",
1402 imsm_orom_has_chunk(orom
, 32) ? " 32k" : "",
1403 imsm_orom_has_chunk(orom
, 64) ? " 64k" : "",
1404 imsm_orom_has_chunk(orom
, 128) ? " 128k" : "",
1405 imsm_orom_has_chunk(orom
, 256) ? " 256k" : "",
1406 imsm_orom_has_chunk(orom
, 512) ? " 512k" : "",
1407 imsm_orom_has_chunk(orom
, 1024*1) ? " 1M" : "",
1408 imsm_orom_has_chunk(orom
, 1024*2) ? " 2M" : "",
1409 imsm_orom_has_chunk(orom
, 1024*4) ? " 4M" : "",
1410 imsm_orom_has_chunk(orom
, 1024*8) ? " 8M" : "",
1411 imsm_orom_has_chunk(orom
, 1024*16) ? " 16M" : "",
1412 imsm_orom_has_chunk(orom
, 1024*32) ? " 32M" : "",
1413 imsm_orom_has_chunk(orom
, 1024*64) ? " 64M" : "");
1414 printf(" Max Disks : %d\n", orom
->tds
);
1415 printf(" Max Volumes : %d\n", orom
->vpa
);
1417 for (hba
= list
; hba
; hba
= hba
->next
) {
1418 printf(" I/O Controller : %s (%s)\n",
1419 hba
->path
, get_sys_dev_type(hba
->type
));
1421 if (hba
->type
== SYS_DEV_SATA
) {
1422 host_base
= ahci_get_port_count(hba
->path
, &port_count
);
1423 if (ahci_enumerate_ports(hba
->path
, port_count
, host_base
, verbose
)) {
1425 fprintf(stderr
, Name
": failed to enumerate "
1426 "ports on SATA controller at %s.", hba
->pci_id
);
1429 } else if (hba
->type
== SYS_DEV_SAS
) {
1431 fprintf(stderr
, Name
": failed to enumerate "
1432 "devices on SAS controller at %s.", hba
->pci_id
);
1437 free_sys_dev(&list
);
1442 static int match_home_imsm(struct supertype
*st
, char *homehost
)
1444 /* the imsm metadata format does not specify any host
1445 * identification information. We return -1 since we can never
1446 * confirm nor deny whether a given array is "meant" for this
1447 * host. We rely on compare_super and the 'family_num' fields to
1448 * exclude member disks that do not belong, and we rely on
1449 * mdadm.conf to specify the arrays that should be assembled.
1450 * Auto-assembly may still pick up "foreign" arrays.
1456 static void uuid_from_super_imsm(struct supertype
*st
, int uuid
[4])
1458 /* The uuid returned here is used for:
1459 * uuid to put into bitmap file (Create, Grow)
1460 * uuid for backup header when saving critical section (Grow)
1461 * comparing uuids when re-adding a device into an array
1462 * In these cases the uuid required is that of the data-array,
1463 * not the device-set.
1464 * uuid to recognise same set when adding a missing device back
1465 * to an array. This is a uuid for the device-set.
1467 * For each of these we can make do with a truncated
1468 * or hashed uuid rather than the original, as long as
1470 * In each case the uuid required is that of the data-array,
1471 * not the device-set.
1473 /* imsm does not track uuid's so we synthesis one using sha1 on
1474 * - The signature (Which is constant for all imsm array, but no matter)
1475 * - the orig_family_num of the container
1476 * - the index number of the volume
1477 * - the 'serial' number of the volume.
1478 * Hopefully these are all constant.
1480 struct intel_super
*super
= st
->sb
;
1483 struct sha1_ctx ctx
;
1484 struct imsm_dev
*dev
= NULL
;
1487 /* some mdadm versions failed to set ->orig_family_num, in which
1488 * case fall back to ->family_num. orig_family_num will be
1489 * fixed up with the first metadata update.
1491 family_num
= super
->anchor
->orig_family_num
;
1492 if (family_num
== 0)
1493 family_num
= super
->anchor
->family_num
;
1494 sha1_init_ctx(&ctx
);
1495 sha1_process_bytes(super
->anchor
->sig
, MPB_SIG_LEN
, &ctx
);
1496 sha1_process_bytes(&family_num
, sizeof(__u32
), &ctx
);
1497 if (super
->current_vol
>= 0)
1498 dev
= get_imsm_dev(super
, super
->current_vol
);
1500 __u32 vol
= super
->current_vol
;
1501 sha1_process_bytes(&vol
, sizeof(vol
), &ctx
);
1502 sha1_process_bytes(dev
->volume
, MAX_RAID_SERIAL_LEN
, &ctx
);
1504 sha1_finish_ctx(&ctx
, buf
);
1505 memcpy(uuid
, buf
, 4*4);
1510 get_imsm_numerical_version(struct imsm_super
*mpb
, int *m
, int *p
)
1512 __u8
*v
= get_imsm_version(mpb
);
1513 __u8
*end
= mpb
->sig
+ MAX_SIGNATURE_LENGTH
;
1514 char major
[] = { 0, 0, 0 };
1515 char minor
[] = { 0 ,0, 0 };
1516 char patch
[] = { 0, 0, 0 };
1517 char *ver_parse
[] = { major
, minor
, patch
};
1521 while (*v
!= '\0' && v
< end
) {
1522 if (*v
!= '.' && j
< 2)
1523 ver_parse
[i
][j
++] = *v
;
1531 *m
= strtol(minor
, NULL
, 0);
1532 *p
= strtol(patch
, NULL
, 0);
1536 static __u32
migr_strip_blocks_resync(struct imsm_dev
*dev
)
1538 /* migr_strip_size when repairing or initializing parity */
1539 struct imsm_map
*map
= get_imsm_map(dev
, 0);
1540 __u32 chunk
= __le32_to_cpu(map
->blocks_per_strip
);
1542 switch (get_imsm_raid_level(map
)) {
1547 return 128*1024 >> 9;
1551 static __u32
migr_strip_blocks_rebuild(struct imsm_dev
*dev
)
1553 /* migr_strip_size when rebuilding a degraded disk, no idea why
1554 * this is different than migr_strip_size_resync(), but it's good
1557 struct imsm_map
*map
= get_imsm_map(dev
, 1);
1558 __u32 chunk
= __le32_to_cpu(map
->blocks_per_strip
);
1560 switch (get_imsm_raid_level(map
)) {
1563 if (map
->num_members
% map
->num_domains
== 0)
1564 return 128*1024 >> 9;
1568 return max((__u32
) 64*1024 >> 9, chunk
);
1570 return 128*1024 >> 9;
1574 static __u32
num_stripes_per_unit_resync(struct imsm_dev
*dev
)
1576 struct imsm_map
*lo
= get_imsm_map(dev
, 0);
1577 struct imsm_map
*hi
= get_imsm_map(dev
, 1);
1578 __u32 lo_chunk
= __le32_to_cpu(lo
->blocks_per_strip
);
1579 __u32 hi_chunk
= __le32_to_cpu(hi
->blocks_per_strip
);
1581 return max((__u32
) 1, hi_chunk
/ lo_chunk
);
1584 static __u32
num_stripes_per_unit_rebuild(struct imsm_dev
*dev
)
1586 struct imsm_map
*lo
= get_imsm_map(dev
, 0);
1587 int level
= get_imsm_raid_level(lo
);
1589 if (level
== 1 || level
== 10) {
1590 struct imsm_map
*hi
= get_imsm_map(dev
, 1);
1592 return hi
->num_domains
;
1594 return num_stripes_per_unit_resync(dev
);
1597 static __u8
imsm_num_data_members(struct imsm_dev
*dev
, int second_map
)
1599 /* named 'imsm_' because raid0, raid1 and raid10
1600 * counter-intuitively have the same number of data disks
1602 struct imsm_map
*map
= get_imsm_map(dev
, second_map
);
1604 switch (get_imsm_raid_level(map
)) {
1608 return map
->num_members
;
1610 return map
->num_members
- 1;
1612 dprintf("%s: unsupported raid level\n", __func__
);
1617 static __u32
parity_segment_depth(struct imsm_dev
*dev
)
1619 struct imsm_map
*map
= get_imsm_map(dev
, 0);
1620 __u32 chunk
= __le32_to_cpu(map
->blocks_per_strip
);
1622 switch(get_imsm_raid_level(map
)) {
1625 return chunk
* map
->num_domains
;
1627 return chunk
* map
->num_members
;
1633 static __u32
map_migr_block(struct imsm_dev
*dev
, __u32 block
)
1635 struct imsm_map
*map
= get_imsm_map(dev
, 1);
1636 __u32 chunk
= __le32_to_cpu(map
->blocks_per_strip
);
1637 __u32 strip
= block
/ chunk
;
1639 switch (get_imsm_raid_level(map
)) {
1642 __u32 vol_strip
= (strip
* map
->num_domains
) + 1;
1643 __u32 vol_stripe
= vol_strip
/ map
->num_members
;
1645 return vol_stripe
* chunk
+ block
% chunk
;
1647 __u32 stripe
= strip
/ (map
->num_members
- 1);
1649 return stripe
* chunk
+ block
% chunk
;
1656 static __u64
blocks_per_migr_unit(struct imsm_dev
*dev
)
1658 /* calculate the conversion factor between per member 'blocks'
1659 * (md/{resync,rebuild}_start) and imsm migration units, return
1660 * 0 for the 'not migrating' and 'unsupported migration' cases
1662 if (!dev
->vol
.migr_state
)
1665 switch (migr_type(dev
)) {
1670 struct imsm_map
*map
= get_imsm_map(dev
, 0);
1671 __u32 stripes_per_unit
;
1672 __u32 blocks_per_unit
;
1681 /* yes, this is really the translation of migr_units to
1682 * per-member blocks in the 'resync' case
1684 stripes_per_unit
= num_stripes_per_unit_resync(dev
);
1685 migr_chunk
= migr_strip_blocks_resync(dev
);
1686 disks
= imsm_num_data_members(dev
, 0);
1687 blocks_per_unit
= stripes_per_unit
* migr_chunk
* disks
;
1688 stripe
= __le32_to_cpu(map
->blocks_per_strip
) * disks
;
1689 segment
= blocks_per_unit
/ stripe
;
1690 block_rel
= blocks_per_unit
- segment
* stripe
;
1691 parity_depth
= parity_segment_depth(dev
);
1692 block_map
= map_migr_block(dev
, block_rel
);
1693 return block_map
+ parity_depth
* segment
;
1695 case MIGR_REBUILD
: {
1696 __u32 stripes_per_unit
;
1699 stripes_per_unit
= num_stripes_per_unit_rebuild(dev
);
1700 migr_chunk
= migr_strip_blocks_rebuild(dev
);
1701 return migr_chunk
* stripes_per_unit
;
1703 case MIGR_STATE_CHANGE
:
1709 static int imsm_level_to_layout(int level
)
1717 return ALGORITHM_LEFT_ASYMMETRIC
;
1724 static void getinfo_super_imsm_volume(struct supertype
*st
, struct mdinfo
*info
, char *dmap
)
1726 struct intel_super
*super
= st
->sb
;
1727 struct imsm_dev
*dev
= get_imsm_dev(super
, super
->current_vol
);
1728 struct imsm_map
*map
= get_imsm_map(dev
, 0);
1729 struct imsm_map
*prev_map
= get_imsm_map(dev
, 1);
1730 struct imsm_map
*map_to_analyse
= map
;
1733 int map_disks
= info
->array
.raid_disks
;
1736 map_to_analyse
= prev_map
;
1738 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
1739 if (dl
->raiddisk
== info
->disk
.raid_disk
)
1741 info
->container_member
= super
->current_vol
;
1742 info
->array
.raid_disks
= map_to_analyse
->num_members
;
1743 info
->array
.level
= get_imsm_raid_level(map_to_analyse
);
1744 info
->array
.layout
= imsm_level_to_layout(info
->array
.level
);
1745 info
->array
.md_minor
= -1;
1746 info
->array
.ctime
= 0;
1747 info
->array
.utime
= 0;
1748 info
->array
.chunk_size
=
1749 __le16_to_cpu(map_to_analyse
->blocks_per_strip
) << 9;
1750 info
->array
.state
= !dev
->vol
.dirty
;
1751 info
->custom_array_size
= __le32_to_cpu(dev
->size_high
);
1752 info
->custom_array_size
<<= 32;
1753 info
->custom_array_size
|= __le32_to_cpu(dev
->size_low
);
1755 info
->new_level
= get_imsm_raid_level(map
);
1756 info
->new_layout
= imsm_level_to_layout(info
->new_level
);
1757 info
->new_chunk
= __le16_to_cpu(map
->blocks_per_strip
) << 9;
1759 info
->new_level
= UnSet
;
1760 info
->new_layout
= UnSet
;
1761 info
->new_chunk
= info
->array
.chunk_size
;
1763 info
->disk
.major
= 0;
1764 info
->disk
.minor
= 0;
1766 info
->disk
.major
= dl
->major
;
1767 info
->disk
.minor
= dl
->minor
;
1770 info
->data_offset
= __le32_to_cpu(map_to_analyse
->pba_of_lba0
);
1771 info
->component_size
=
1772 __le32_to_cpu(map_to_analyse
->blocks_per_member
);
1773 memset(info
->uuid
, 0, sizeof(info
->uuid
));
1774 info
->recovery_start
= MaxSector
;
1775 info
->reshape_active
= (prev_map
!= NULL
) &&
1776 (map
->map_state
== prev_map
->map_state
);
1777 if (info
->reshape_active
)
1778 info
->delta_disks
= map
->num_members
- prev_map
->num_members
;
1780 info
->delta_disks
= 0;
1782 if (map_to_analyse
->map_state
== IMSM_T_STATE_UNINITIALIZED
||
1784 info
->resync_start
= 0;
1785 } else if (dev
->vol
.migr_state
) {
1786 switch (migr_type(dev
)) {
1789 __u64 blocks_per_unit
= blocks_per_migr_unit(dev
);
1790 __u64 units
= __le32_to_cpu(dev
->vol
.curr_migr_unit
);
1792 info
->resync_start
= blocks_per_unit
* units
;
1796 /* we could emulate the checkpointing of
1797 * 'sync_action=check' migrations, but for now
1798 * we just immediately complete them
1801 /* this is handled by container_content_imsm() */
1803 case MIGR_STATE_CHANGE
:
1804 /* FIXME handle other migrations */
1806 /* we are not dirty, so... */
1807 info
->resync_start
= MaxSector
;
1810 info
->resync_start
= MaxSector
;
1812 strncpy(info
->name
, (char *) dev
->volume
, MAX_RAID_SERIAL_LEN
);
1813 info
->name
[MAX_RAID_SERIAL_LEN
] = 0;
1815 info
->array
.major_version
= -1;
1816 info
->array
.minor_version
= -2;
1817 devname
= devnum2devname(st
->container_dev
);
1818 *info
->text_version
= '\0';
1820 sprintf(info
->text_version
, "/%s/%d", devname
, info
->container_member
);
1822 info
->safe_mode_delay
= 4000; /* 4 secs like the Matrix driver */
1823 uuid_from_super_imsm(st
, info
->uuid
);
1827 for (i
=0; i
<map_disks
; i
++) {
1829 if (i
< info
->array
.raid_disks
) {
1830 struct imsm_disk
*dsk
;
1831 j
= get_imsm_disk_idx(dev
, i
, -1);
1832 dsk
= get_imsm_disk(super
, j
);
1833 if (dsk
&& (dsk
->status
& CONFIGURED_DISK
))
1840 static __u8
imsm_check_degraded(struct intel_super
*super
, struct imsm_dev
*dev
, int failed
);
1841 static int imsm_count_failed(struct intel_super
*super
, struct imsm_dev
*dev
);
1843 static struct imsm_disk
*get_imsm_missing(struct intel_super
*super
, __u8 index
)
1847 for (d
= super
->missing
; d
; d
= d
->next
)
1848 if (d
->index
== index
)
1853 static void getinfo_super_imsm(struct supertype
*st
, struct mdinfo
*info
, char *map
)
1855 struct intel_super
*super
= st
->sb
;
1856 struct imsm_disk
*disk
;
1857 int map_disks
= info
->array
.raid_disks
;
1858 int max_enough
= -1;
1860 struct imsm_super
*mpb
;
1862 if (super
->current_vol
>= 0) {
1863 getinfo_super_imsm_volume(st
, info
, map
);
1867 /* Set raid_disks to zero so that Assemble will always pull in valid
1870 info
->array
.raid_disks
= 0;
1871 info
->array
.level
= LEVEL_CONTAINER
;
1872 info
->array
.layout
= 0;
1873 info
->array
.md_minor
= -1;
1874 info
->array
.ctime
= 0; /* N/A for imsm */
1875 info
->array
.utime
= 0;
1876 info
->array
.chunk_size
= 0;
1878 info
->disk
.major
= 0;
1879 info
->disk
.minor
= 0;
1880 info
->disk
.raid_disk
= -1;
1881 info
->reshape_active
= 0;
1882 info
->array
.major_version
= -1;
1883 info
->array
.minor_version
= -2;
1884 strcpy(info
->text_version
, "imsm");
1885 info
->safe_mode_delay
= 0;
1886 info
->disk
.number
= -1;
1887 info
->disk
.state
= 0;
1889 info
->recovery_start
= MaxSector
;
1891 /* do we have the all the insync disks that we expect? */
1892 mpb
= super
->anchor
;
1894 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
1895 struct imsm_dev
*dev
= get_imsm_dev(super
, i
);
1896 int failed
, enough
, j
, missing
= 0;
1897 struct imsm_map
*map
;
1900 failed
= imsm_count_failed(super
, dev
);
1901 state
= imsm_check_degraded(super
, dev
, failed
);
1902 map
= get_imsm_map(dev
, dev
->vol
.migr_state
);
1904 /* any newly missing disks?
1905 * (catches single-degraded vs double-degraded)
1907 for (j
= 0; j
< map
->num_members
; j
++) {
1908 __u32 ord
= get_imsm_ord_tbl_ent(dev
, i
, -1);
1909 __u32 idx
= ord_to_idx(ord
);
1911 if (!(ord
& IMSM_ORD_REBUILD
) &&
1912 get_imsm_missing(super
, idx
)) {
1918 if (state
== IMSM_T_STATE_FAILED
)
1920 else if (state
== IMSM_T_STATE_DEGRADED
&&
1921 (state
!= map
->map_state
|| missing
))
1923 else /* we're normal, or already degraded */
1926 /* in the missing/failed disk case check to see
1927 * if at least one array is runnable
1929 max_enough
= max(max_enough
, enough
);
1931 dprintf("%s: enough: %d\n", __func__
, max_enough
);
1932 info
->container_enough
= max_enough
;
1935 __u32 reserved
= imsm_reserved_sectors(super
, super
->disks
);
1937 disk
= &super
->disks
->disk
;
1938 info
->data_offset
= __le32_to_cpu(disk
->total_blocks
) - reserved
;
1939 info
->component_size
= reserved
;
1940 info
->disk
.state
= is_configured(disk
) ? (1 << MD_DISK_ACTIVE
) : 0;
1941 /* we don't change info->disk.raid_disk here because
1942 * this state will be finalized in mdmon after we have
1943 * found the 'most fresh' version of the metadata
1945 info
->disk
.state
|= is_failed(disk
) ? (1 << MD_DISK_FAULTY
) : 0;
1946 info
->disk
.state
|= is_spare(disk
) ? 0 : (1 << MD_DISK_SYNC
);
1949 /* only call uuid_from_super_imsm when this disk is part of a populated container,
1950 * ->compare_super may have updated the 'num_raid_devs' field for spares
1952 if (info
->disk
.state
& (1 << MD_DISK_SYNC
) || super
->anchor
->num_raid_devs
)
1953 uuid_from_super_imsm(st
, info
->uuid
);
1955 memcpy(info
->uuid
, uuid_zero
, sizeof(uuid_zero
));
1957 /* I don't know how to compute 'map' on imsm, so use safe default */
1960 for (i
= 0; i
< map_disks
; i
++)
1966 /* allocates memory and fills disk in mdinfo structure
1967 * for each disk in array */
1968 struct mdinfo
*getinfo_super_disks_imsm(struct supertype
*st
)
1970 struct mdinfo
*mddev
= NULL
;
1971 struct intel_super
*super
= st
->sb
;
1972 struct imsm_disk
*disk
;
1975 if (!super
|| !super
->disks
)
1978 mddev
= malloc(sizeof(*mddev
));
1980 fprintf(stderr
, Name
": Failed to allocate memory.\n");
1983 memset(mddev
, 0, sizeof(*mddev
));
1987 tmp
= malloc(sizeof(*tmp
));
1989 fprintf(stderr
, Name
": Failed to allocate memory.\n");
1994 memset(tmp
, 0, sizeof(*tmp
));
1996 tmp
->next
= mddev
->devs
;
1998 tmp
->disk
.number
= count
++;
1999 tmp
->disk
.major
= dl
->major
;
2000 tmp
->disk
.minor
= dl
->minor
;
2001 tmp
->disk
.state
= is_configured(disk
) ?
2002 (1 << MD_DISK_ACTIVE
) : 0;
2003 tmp
->disk
.state
|= is_failed(disk
) ? (1 << MD_DISK_FAULTY
) : 0;
2004 tmp
->disk
.state
|= is_spare(disk
) ? 0 : (1 << MD_DISK_SYNC
);
2005 tmp
->disk
.raid_disk
= -1;
2011 static int update_super_imsm(struct supertype
*st
, struct mdinfo
*info
,
2012 char *update
, char *devname
, int verbose
,
2013 int uuid_set
, char *homehost
)
2015 /* For 'assemble' and 'force' we need to return non-zero if any
2016 * change was made. For others, the return value is ignored.
2017 * Update options are:
2018 * force-one : This device looks a bit old but needs to be included,
2019 * update age info appropriately.
2020 * assemble: clear any 'faulty' flag to allow this device to
2022 * force-array: Array is degraded but being forced, mark it clean
2023 * if that will be needed to assemble it.
2025 * newdev: not used ????
2026 * grow: Array has gained a new device - this is currently for
2028 * resync: mark as dirty so a resync will happen.
2029 * name: update the name - preserving the homehost
2030 * uuid: Change the uuid of the array to match watch is given
2032 * Following are not relevant for this imsm:
2033 * sparc2.2 : update from old dodgey metadata
2034 * super-minor: change the preferred_minor number
2035 * summaries: update redundant counters.
2036 * homehost: update the recorded homehost
2037 * _reshape_progress: record new reshape_progress position.
2040 struct intel_super
*super
= st
->sb
;
2041 struct imsm_super
*mpb
;
2043 /* we can only update container info */
2044 if (!super
|| super
->current_vol
>= 0 || !super
->anchor
)
2047 mpb
= super
->anchor
;
2049 if (strcmp(update
, "uuid") == 0 && uuid_set
&& !info
->update_private
)
2051 else if (strcmp(update
, "uuid") == 0 && uuid_set
&& info
->update_private
) {
2052 mpb
->orig_family_num
= *((__u32
*) info
->update_private
);
2054 } else if (strcmp(update
, "uuid") == 0) {
2055 __u32
*new_family
= malloc(sizeof(*new_family
));
2057 /* update orig_family_number with the incoming random
2058 * data, report the new effective uuid, and store the
2059 * new orig_family_num for future updates.
2062 memcpy(&mpb
->orig_family_num
, info
->uuid
, sizeof(__u32
));
2063 uuid_from_super_imsm(st
, info
->uuid
);
2064 *new_family
= mpb
->orig_family_num
;
2065 info
->update_private
= new_family
;
2068 } else if (strcmp(update
, "assemble") == 0)
2073 /* successful update? recompute checksum */
2075 mpb
->check_sum
= __le32_to_cpu(__gen_imsm_checksum(mpb
));
2080 static size_t disks_to_mpb_size(int disks
)
2084 size
= sizeof(struct imsm_super
);
2085 size
+= (disks
- 1) * sizeof(struct imsm_disk
);
2086 size
+= 2 * sizeof(struct imsm_dev
);
2087 /* up to 2 maps per raid device (-2 for imsm_maps in imsm_dev */
2088 size
+= (4 - 2) * sizeof(struct imsm_map
);
2089 /* 4 possible disk_ord_tbl's */
2090 size
+= 4 * (disks
- 1) * sizeof(__u32
);
2095 static __u64
avail_size_imsm(struct supertype
*st
, __u64 devsize
)
2097 if (devsize
< (MPB_SECTOR_CNT
+ IMSM_RESERVED_SECTORS
))
2100 return devsize
- (MPB_SECTOR_CNT
+ IMSM_RESERVED_SECTORS
);
2103 static void free_devlist(struct intel_super
*super
)
2105 struct intel_dev
*dv
;
2107 while (super
->devlist
) {
2108 dv
= super
->devlist
->next
;
2109 free(super
->devlist
->dev
);
2110 free(super
->devlist
);
2111 super
->devlist
= dv
;
2115 static void imsm_copy_dev(struct imsm_dev
*dest
, struct imsm_dev
*src
)
2117 memcpy(dest
, src
, sizeof_imsm_dev(src
, 0));
2120 static int compare_super_imsm(struct supertype
*st
, struct supertype
*tst
)
2124 * 0 same, or first was empty, and second was copied
2125 * 1 second had wrong number
2127 * 3 wrong other info
2129 struct intel_super
*first
= st
->sb
;
2130 struct intel_super
*sec
= tst
->sb
;
2138 /* if an anchor does not have num_raid_devs set then it is a free
2141 if (first
->anchor
->num_raid_devs
> 0 &&
2142 sec
->anchor
->num_raid_devs
> 0) {
2143 /* Determine if these disks might ever have been
2144 * related. Further disambiguation can only take place
2145 * in load_super_imsm_all
2147 __u32 first_family
= first
->anchor
->orig_family_num
;
2148 __u32 sec_family
= sec
->anchor
->orig_family_num
;
2150 if (memcmp(first
->anchor
->sig
, sec
->anchor
->sig
,
2151 MAX_SIGNATURE_LENGTH
) != 0)
2154 if (first_family
== 0)
2155 first_family
= first
->anchor
->family_num
;
2156 if (sec_family
== 0)
2157 sec_family
= sec
->anchor
->family_num
;
2159 if (first_family
!= sec_family
)
2165 /* if 'first' is a spare promote it to a populated mpb with sec's
2168 if (first
->anchor
->num_raid_devs
== 0 &&
2169 sec
->anchor
->num_raid_devs
> 0) {
2171 struct intel_dev
*dv
;
2172 struct imsm_dev
*dev
;
2174 /* we need to copy raid device info from sec if an allocation
2175 * fails here we don't associate the spare
2177 for (i
= 0; i
< sec
->anchor
->num_raid_devs
; i
++) {
2178 dv
= malloc(sizeof(*dv
));
2181 dev
= malloc(sizeof_imsm_dev(get_imsm_dev(sec
, i
), 1));
2188 dv
->next
= first
->devlist
;
2189 first
->devlist
= dv
;
2191 if (i
< sec
->anchor
->num_raid_devs
) {
2192 /* allocation failure */
2193 free_devlist(first
);
2194 fprintf(stderr
, "imsm: failed to associate spare\n");
2197 first
->anchor
->num_raid_devs
= sec
->anchor
->num_raid_devs
;
2198 first
->anchor
->orig_family_num
= sec
->anchor
->orig_family_num
;
2199 first
->anchor
->family_num
= sec
->anchor
->family_num
;
2200 memcpy(first
->anchor
->sig
, sec
->anchor
->sig
, MAX_SIGNATURE_LENGTH
);
2201 for (i
= 0; i
< sec
->anchor
->num_raid_devs
; i
++)
2202 imsm_copy_dev(get_imsm_dev(first
, i
), get_imsm_dev(sec
, i
));
2208 static void fd2devname(int fd
, char *name
)
2212 char dname
[PATH_MAX
];
2217 if (fstat(fd
, &st
) != 0)
2219 sprintf(path
, "/sys/dev/block/%d:%d",
2220 major(st
.st_rdev
), minor(st
.st_rdev
));
2222 rv
= readlink(path
, dname
, sizeof(dname
));
2227 nm
= strrchr(dname
, '/');
2229 snprintf(name
, MAX_RAID_SERIAL_LEN
, "/dev/%s", nm
);
2232 extern int scsi_get_serial(int fd
, void *buf
, size_t buf_len
);
2234 static int imsm_read_serial(int fd
, char *devname
,
2235 __u8 serial
[MAX_RAID_SERIAL_LEN
])
2237 unsigned char scsi_serial
[255];
2246 memset(scsi_serial
, 0, sizeof(scsi_serial
));
2248 rv
= scsi_get_serial(fd
, scsi_serial
, sizeof(scsi_serial
));
2250 if (rv
&& check_env("IMSM_DEVNAME_AS_SERIAL")) {
2251 memset(serial
, 0, MAX_RAID_SERIAL_LEN
);
2252 fd2devname(fd
, (char *) serial
);
2259 Name
": Failed to retrieve serial for %s\n",
2264 rsp_len
= scsi_serial
[3];
2268 Name
": Failed to retrieve serial for %s\n",
2272 rsp_buf
= (char *) &scsi_serial
[4];
2274 /* trim all whitespace and non-printable characters and convert
2277 for (i
= 0, dest
= rsp_buf
; i
< rsp_len
; i
++) {
2280 /* ':' is reserved for use in placeholder serial
2281 * numbers for missing disks
2289 len
= dest
- rsp_buf
;
2292 /* truncate leading characters */
2293 if (len
> MAX_RAID_SERIAL_LEN
) {
2294 dest
+= len
- MAX_RAID_SERIAL_LEN
;
2295 len
= MAX_RAID_SERIAL_LEN
;
2298 memset(serial
, 0, MAX_RAID_SERIAL_LEN
);
2299 memcpy(serial
, dest
, len
);
2304 static int serialcmp(__u8
*s1
, __u8
*s2
)
2306 return strncmp((char *) s1
, (char *) s2
, MAX_RAID_SERIAL_LEN
);
2309 static void serialcpy(__u8
*dest
, __u8
*src
)
2311 strncpy((char *) dest
, (char *) src
, MAX_RAID_SERIAL_LEN
);
2315 static struct dl
*serial_to_dl(__u8
*serial
, struct intel_super
*super
)
2319 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
2320 if (serialcmp(dl
->serial
, serial
) == 0)
2327 static struct imsm_disk
*
2328 __serial_to_disk(__u8
*serial
, struct imsm_super
*mpb
, int *idx
)
2332 for (i
= 0; i
< mpb
->num_disks
; i
++) {
2333 struct imsm_disk
*disk
= __get_imsm_disk(mpb
, i
);
2335 if (serialcmp(disk
->serial
, serial
) == 0) {
2346 load_imsm_disk(int fd
, struct intel_super
*super
, char *devname
, int keep_fd
)
2348 struct imsm_disk
*disk
;
2353 __u8 serial
[MAX_RAID_SERIAL_LEN
];
2355 rv
= imsm_read_serial(fd
, devname
, serial
);
2360 dl
= calloc(1, sizeof(*dl
));
2364 Name
": failed to allocate disk buffer for %s\n",
2370 dl
->major
= major(stb
.st_rdev
);
2371 dl
->minor
= minor(stb
.st_rdev
);
2372 dl
->next
= super
->disks
;
2373 dl
->fd
= keep_fd
? fd
: -1;
2374 assert(super
->disks
== NULL
);
2376 serialcpy(dl
->serial
, serial
);
2379 fd2devname(fd
, name
);
2381 dl
->devname
= strdup(devname
);
2383 dl
->devname
= strdup(name
);
2385 /* look up this disk's index in the current anchor */
2386 disk
= __serial_to_disk(dl
->serial
, super
->anchor
, &dl
->index
);
2389 /* only set index on disks that are a member of a
2390 * populated contianer, i.e. one with raid_devs
2392 if (is_failed(&dl
->disk
))
2394 else if (is_spare(&dl
->disk
))
2402 /* When migrating map0 contains the 'destination' state while map1
2403 * contains the current state. When not migrating map0 contains the
2404 * current state. This routine assumes that map[0].map_state is set to
2405 * the current array state before being called.
2407 * Migration is indicated by one of the following states
2408 * 1/ Idle (migr_state=0 map0state=normal||unitialized||degraded||failed)
2409 * 2/ Initialize (migr_state=1 migr_type=MIGR_INIT map0state=normal
2410 * map1state=unitialized)
2411 * 3/ Repair (Resync) (migr_state=1 migr_type=MIGR_REPAIR map0state=normal
2413 * 4/ Rebuild (migr_state=1 migr_type=MIGR_REBUILD map0state=normal
2414 * map1state=degraded)
2416 static void migrate(struct imsm_dev
*dev
, __u8 to_state
, int migr_type
)
2418 struct imsm_map
*dest
;
2419 struct imsm_map
*src
= get_imsm_map(dev
, 0);
2421 dev
->vol
.migr_state
= 1;
2422 set_migr_type(dev
, migr_type
);
2423 dev
->vol
.curr_migr_unit
= 0;
2424 dest
= get_imsm_map(dev
, 1);
2426 /* duplicate and then set the target end state in map[0] */
2427 memcpy(dest
, src
, sizeof_imsm_map(src
));
2428 if ((migr_type
== MIGR_REBUILD
) ||
2429 (migr_type
== MIGR_GEN_MIGR
)) {
2433 for (i
= 0; i
< src
->num_members
; i
++) {
2434 ord
= __le32_to_cpu(src
->disk_ord_tbl
[i
]);
2435 set_imsm_ord_tbl_ent(src
, i
, ord_to_idx(ord
));
2439 src
->map_state
= to_state
;
2442 static void end_migration(struct imsm_dev
*dev
, __u8 map_state
)
2444 struct imsm_map
*map
= get_imsm_map(dev
, 0);
2445 struct imsm_map
*prev
= get_imsm_map(dev
, dev
->vol
.migr_state
);
2448 /* merge any IMSM_ORD_REBUILD bits that were not successfully
2449 * completed in the last migration.
2451 * FIXME add support for raid-level-migration
2453 for (i
= 0; i
< prev
->num_members
; i
++)
2454 for (j
= 0; j
< map
->num_members
; j
++)
2455 /* during online capacity expansion
2456 * disks position can be changed if takeover is used
2458 if (ord_to_idx(map
->disk_ord_tbl
[j
]) ==
2459 ord_to_idx(prev
->disk_ord_tbl
[i
])) {
2460 map
->disk_ord_tbl
[j
] |= prev
->disk_ord_tbl
[i
];
2464 dev
->vol
.migr_state
= 0;
2465 dev
->vol
.migr_type
= 0;
2466 dev
->vol
.curr_migr_unit
= 0;
2467 map
->map_state
= map_state
;
2471 static int parse_raid_devices(struct intel_super
*super
)
2474 struct imsm_dev
*dev_new
;
2475 size_t len
, len_migr
;
2477 size_t space_needed
= 0;
2478 struct imsm_super
*mpb
= super
->anchor
;
2480 for (i
= 0; i
< super
->anchor
->num_raid_devs
; i
++) {
2481 struct imsm_dev
*dev_iter
= __get_imsm_dev(super
->anchor
, i
);
2482 struct intel_dev
*dv
;
2484 len
= sizeof_imsm_dev(dev_iter
, 0);
2485 len_migr
= sizeof_imsm_dev(dev_iter
, 1);
2487 space_needed
+= len_migr
- len
;
2489 dv
= malloc(sizeof(*dv
));
2492 if (max_len
< len_migr
)
2494 if (max_len
> len_migr
)
2495 space_needed
+= max_len
- len_migr
;
2496 dev_new
= malloc(max_len
);
2501 imsm_copy_dev(dev_new
, dev_iter
);
2504 dv
->next
= super
->devlist
;
2505 super
->devlist
= dv
;
2508 /* ensure that super->buf is large enough when all raid devices
2511 if (__le32_to_cpu(mpb
->mpb_size
) + space_needed
> super
->len
) {
2514 len
= ROUND_UP(__le32_to_cpu(mpb
->mpb_size
) + space_needed
, 512);
2515 if (posix_memalign(&buf
, 512, len
) != 0)
2518 memcpy(buf
, super
->buf
, super
->len
);
2519 memset(buf
+ super
->len
, 0, len
- super
->len
);
2528 /* retrieve a pointer to the bbm log which starts after all raid devices */
2529 struct bbm_log
*__get_imsm_bbm_log(struct imsm_super
*mpb
)
2533 if (__le32_to_cpu(mpb
->bbm_log_size
)) {
2535 ptr
+= mpb
->mpb_size
- __le32_to_cpu(mpb
->bbm_log_size
);
2541 static void __free_imsm(struct intel_super
*super
, int free_disks
);
2543 /* load_imsm_mpb - read matrix metadata
2544 * allocates super->mpb to be freed by free_super
2546 static int load_imsm_mpb(int fd
, struct intel_super
*super
, char *devname
)
2548 unsigned long long dsize
;
2549 unsigned long long sectors
;
2551 struct imsm_super
*anchor
;
2554 get_dev_size(fd
, NULL
, &dsize
);
2558 Name
": %s: device to small for imsm\n",
2563 if (lseek64(fd
, dsize
- (512 * 2), SEEK_SET
) < 0) {
2566 Name
": Cannot seek to anchor block on %s: %s\n",
2567 devname
, strerror(errno
));
2571 if (posix_memalign((void**)&anchor
, 512, 512) != 0) {
2574 Name
": Failed to allocate imsm anchor buffer"
2575 " on %s\n", devname
);
2578 if (read(fd
, anchor
, 512) != 512) {
2581 Name
": Cannot read anchor block on %s: %s\n",
2582 devname
, strerror(errno
));
2587 if (strncmp((char *) anchor
->sig
, MPB_SIGNATURE
, MPB_SIG_LEN
) != 0) {
2590 Name
": no IMSM anchor on %s\n", devname
);
2595 __free_imsm(super
, 0);
2596 super
->len
= ROUND_UP(anchor
->mpb_size
, 512);
2597 if (posix_memalign(&super
->buf
, 512, super
->len
) != 0) {
2600 Name
": unable to allocate %zu byte mpb buffer\n",
2605 memcpy(super
->buf
, anchor
, 512);
2607 sectors
= mpb_sectors(anchor
) - 1;
2610 check_sum
= __gen_imsm_checksum(super
->anchor
);
2611 if (check_sum
!= __le32_to_cpu(super
->anchor
->check_sum
)) {
2614 Name
": IMSM checksum %x != %x on %s\n",
2616 __le32_to_cpu(super
->anchor
->check_sum
),
2624 /* read the extended mpb */
2625 if (lseek64(fd
, dsize
- (512 * (2 + sectors
)), SEEK_SET
) < 0) {
2628 Name
": Cannot seek to extended mpb on %s: %s\n",
2629 devname
, strerror(errno
));
2633 if ((unsigned)read(fd
, super
->buf
+ 512, super
->len
- 512) != super
->len
- 512) {
2636 Name
": Cannot read extended mpb on %s: %s\n",
2637 devname
, strerror(errno
));
2641 check_sum
= __gen_imsm_checksum(super
->anchor
);
2642 if (check_sum
!= __le32_to_cpu(super
->anchor
->check_sum
)) {
2645 Name
": IMSM checksum %x != %x on %s\n",
2646 check_sum
, __le32_to_cpu(super
->anchor
->check_sum
),
2651 /* FIXME the BBM log is disk specific so we cannot use this global
2652 * buffer for all disks. Ok for now since we only look at the global
2653 * bbm_log_size parameter to gate assembly
2655 super
->bbm_log
= __get_imsm_bbm_log(super
->anchor
);
2661 load_and_parse_mpb(int fd
, struct intel_super
*super
, char *devname
, int keep_fd
)
2665 err
= load_imsm_mpb(fd
, super
, devname
);
2668 err
= load_imsm_disk(fd
, super
, devname
, keep_fd
);
2671 err
= parse_raid_devices(super
);
2676 static void __free_imsm_disk(struct dl
*d
)
2688 static void free_imsm_disks(struct intel_super
*super
)
2692 while (super
->disks
) {
2694 super
->disks
= d
->next
;
2695 __free_imsm_disk(d
);
2697 while (super
->disk_mgmt_list
) {
2698 d
= super
->disk_mgmt_list
;
2699 super
->disk_mgmt_list
= d
->next
;
2700 __free_imsm_disk(d
);
2702 while (super
->missing
) {
2704 super
->missing
= d
->next
;
2705 __free_imsm_disk(d
);
2710 /* free all the pieces hanging off of a super pointer */
2711 static void __free_imsm(struct intel_super
*super
, int free_disks
)
2713 struct intel_hba
*elem
, *next
;
2720 free_imsm_disks(super
);
2721 free_devlist(super
);
2725 free((void *)elem
->path
);
2733 static void free_imsm(struct intel_super
*super
)
2735 __free_imsm(super
, 1);
2739 static void free_super_imsm(struct supertype
*st
)
2741 struct intel_super
*super
= st
->sb
;
2750 static struct intel_super
*alloc_super(void)
2752 struct intel_super
*super
= malloc(sizeof(*super
));
2755 memset(super
, 0, sizeof(*super
));
2756 super
->current_vol
= -1;
2757 super
->create_offset
= ~((__u32
) 0);
2758 if (!check_env("IMSM_NO_PLATFORM"))
2759 super
->orom
= find_imsm_orom();
2766 /* find_missing - helper routine for load_super_imsm_all that identifies
2767 * disks that have disappeared from the system. This routine relies on
2768 * the mpb being uptodate, which it is at load time.
2770 static int find_missing(struct intel_super
*super
)
2773 struct imsm_super
*mpb
= super
->anchor
;
2775 struct imsm_disk
*disk
;
2777 for (i
= 0; i
< mpb
->num_disks
; i
++) {
2778 disk
= __get_imsm_disk(mpb
, i
);
2779 dl
= serial_to_dl(disk
->serial
, super
);
2783 dl
= malloc(sizeof(*dl
));
2789 dl
->devname
= strdup("missing");
2791 serialcpy(dl
->serial
, disk
->serial
);
2794 dl
->next
= super
->missing
;
2795 super
->missing
= dl
;
2801 static struct intel_disk
*disk_list_get(__u8
*serial
, struct intel_disk
*disk_list
)
2803 struct intel_disk
*idisk
= disk_list
;
2806 if (serialcmp(idisk
->disk
.serial
, serial
) == 0)
2808 idisk
= idisk
->next
;
2814 static int __prep_thunderdome(struct intel_super
**table
, int tbl_size
,
2815 struct intel_super
*super
,
2816 struct intel_disk
**disk_list
)
2818 struct imsm_disk
*d
= &super
->disks
->disk
;
2819 struct imsm_super
*mpb
= super
->anchor
;
2822 for (i
= 0; i
< tbl_size
; i
++) {
2823 struct imsm_super
*tbl_mpb
= table
[i
]->anchor
;
2824 struct imsm_disk
*tbl_d
= &table
[i
]->disks
->disk
;
2826 if (tbl_mpb
->family_num
== mpb
->family_num
) {
2827 if (tbl_mpb
->check_sum
== mpb
->check_sum
) {
2828 dprintf("%s: mpb from %d:%d matches %d:%d\n",
2829 __func__
, super
->disks
->major
,
2830 super
->disks
->minor
,
2831 table
[i
]->disks
->major
,
2832 table
[i
]->disks
->minor
);
2836 if (((is_configured(d
) && !is_configured(tbl_d
)) ||
2837 is_configured(d
) == is_configured(tbl_d
)) &&
2838 tbl_mpb
->generation_num
< mpb
->generation_num
) {
2839 /* current version of the mpb is a
2840 * better candidate than the one in
2841 * super_table, but copy over "cross
2842 * generational" status
2844 struct intel_disk
*idisk
;
2846 dprintf("%s: mpb from %d:%d replaces %d:%d\n",
2847 __func__
, super
->disks
->major
,
2848 super
->disks
->minor
,
2849 table
[i
]->disks
->major
,
2850 table
[i
]->disks
->minor
);
2852 idisk
= disk_list_get(tbl_d
->serial
, *disk_list
);
2853 if (idisk
&& is_failed(&idisk
->disk
))
2854 tbl_d
->status
|= FAILED_DISK
;
2857 struct intel_disk
*idisk
;
2858 struct imsm_disk
*disk
;
2860 /* tbl_mpb is more up to date, but copy
2861 * over cross generational status before
2864 disk
= __serial_to_disk(d
->serial
, mpb
, NULL
);
2865 if (disk
&& is_failed(disk
))
2866 d
->status
|= FAILED_DISK
;
2868 idisk
= disk_list_get(d
->serial
, *disk_list
);
2871 if (disk
&& is_configured(disk
))
2872 idisk
->disk
.status
|= CONFIGURED_DISK
;
2875 dprintf("%s: mpb from %d:%d prefer %d:%d\n",
2876 __func__
, super
->disks
->major
,
2877 super
->disks
->minor
,
2878 table
[i
]->disks
->major
,
2879 table
[i
]->disks
->minor
);
2887 table
[tbl_size
++] = super
;
2891 /* update/extend the merged list of imsm_disk records */
2892 for (j
= 0; j
< mpb
->num_disks
; j
++) {
2893 struct imsm_disk
*disk
= __get_imsm_disk(mpb
, j
);
2894 struct intel_disk
*idisk
;
2896 idisk
= disk_list_get(disk
->serial
, *disk_list
);
2898 idisk
->disk
.status
|= disk
->status
;
2899 if (is_configured(&idisk
->disk
) ||
2900 is_failed(&idisk
->disk
))
2901 idisk
->disk
.status
&= ~(SPARE_DISK
);
2903 idisk
= calloc(1, sizeof(*idisk
));
2906 idisk
->owner
= IMSM_UNKNOWN_OWNER
;
2907 idisk
->disk
= *disk
;
2908 idisk
->next
= *disk_list
;
2912 if (serialcmp(idisk
->disk
.serial
, d
->serial
) == 0)
2919 static struct intel_super
*
2920 validate_members(struct intel_super
*super
, struct intel_disk
*disk_list
,
2923 struct imsm_super
*mpb
= super
->anchor
;
2927 for (i
= 0; i
< mpb
->num_disks
; i
++) {
2928 struct imsm_disk
*disk
= __get_imsm_disk(mpb
, i
);
2929 struct intel_disk
*idisk
;
2931 idisk
= disk_list_get(disk
->serial
, disk_list
);
2933 if (idisk
->owner
== owner
||
2934 idisk
->owner
== IMSM_UNKNOWN_OWNER
)
2937 dprintf("%s: '%.16s' owner %d != %d\n",
2938 __func__
, disk
->serial
, idisk
->owner
,
2941 dprintf("%s: unknown disk %x [%d]: %.16s\n",
2942 __func__
, __le32_to_cpu(mpb
->family_num
), i
,
2948 if (ok_count
== mpb
->num_disks
)
2953 static void show_conflicts(__u32 family_num
, struct intel_super
*super_list
)
2955 struct intel_super
*s
;
2957 for (s
= super_list
; s
; s
= s
->next
) {
2958 if (family_num
!= s
->anchor
->family_num
)
2960 fprintf(stderr
, "Conflict, offlining family %#x on '%s'\n",
2961 __le32_to_cpu(family_num
), s
->disks
->devname
);
2965 static struct intel_super
*
2966 imsm_thunderdome(struct intel_super
**super_list
, int len
)
2968 struct intel_super
*super_table
[len
];
2969 struct intel_disk
*disk_list
= NULL
;
2970 struct intel_super
*champion
, *spare
;
2971 struct intel_super
*s
, **del
;
2976 memset(super_table
, 0, sizeof(super_table
));
2977 for (s
= *super_list
; s
; s
= s
->next
)
2978 tbl_size
= __prep_thunderdome(super_table
, tbl_size
, s
, &disk_list
);
2980 for (i
= 0; i
< tbl_size
; i
++) {
2981 struct imsm_disk
*d
;
2982 struct intel_disk
*idisk
;
2983 struct imsm_super
*mpb
= super_table
[i
]->anchor
;
2986 d
= &s
->disks
->disk
;
2988 /* 'd' must appear in merged disk list for its
2989 * configuration to be valid
2991 idisk
= disk_list_get(d
->serial
, disk_list
);
2992 if (idisk
&& idisk
->owner
== i
)
2993 s
= validate_members(s
, disk_list
, i
);
2998 dprintf("%s: marking family: %#x from %d:%d offline\n",
2999 __func__
, mpb
->family_num
,
3000 super_table
[i
]->disks
->major
,
3001 super_table
[i
]->disks
->minor
);
3005 /* This is where the mdadm implementation differs from the Windows
3006 * driver which has no strict concept of a container. We can only
3007 * assemble one family from a container, so when returning a prodigal
3008 * array member to this system the code will not be able to disambiguate
3009 * the container contents that should be assembled ("foreign" versus
3010 * "local"). It requires user intervention to set the orig_family_num
3011 * to a new value to establish a new container. The Windows driver in
3012 * this situation fixes up the volume name in place and manages the
3013 * foreign array as an independent entity.
3018 for (i
= 0; i
< tbl_size
; i
++) {
3019 struct intel_super
*tbl_ent
= super_table
[i
];
3025 if (tbl_ent
->anchor
->num_raid_devs
== 0) {
3030 if (s
&& !is_spare
) {
3031 show_conflicts(tbl_ent
->anchor
->family_num
, *super_list
);
3033 } else if (!s
&& !is_spare
)
3046 fprintf(stderr
, "Chose family %#x on '%s', "
3047 "assemble conflicts to new container with '--update=uuid'\n",
3048 __le32_to_cpu(s
->anchor
->family_num
), s
->disks
->devname
);
3050 /* collect all dl's onto 'champion', and update them to
3051 * champion's version of the status
3053 for (s
= *super_list
; s
; s
= s
->next
) {
3054 struct imsm_super
*mpb
= champion
->anchor
;
3055 struct dl
*dl
= s
->disks
;
3060 for (i
= 0; i
< mpb
->num_disks
; i
++) {
3061 struct imsm_disk
*disk
;
3063 disk
= __serial_to_disk(dl
->serial
, mpb
, &dl
->index
);
3066 /* only set index on disks that are a member of
3067 * a populated contianer, i.e. one with
3070 if (is_failed(&dl
->disk
))
3072 else if (is_spare(&dl
->disk
))
3078 if (i
>= mpb
->num_disks
) {
3079 struct intel_disk
*idisk
;
3081 idisk
= disk_list_get(dl
->serial
, disk_list
);
3082 if (idisk
&& is_spare(&idisk
->disk
) &&
3083 !is_failed(&idisk
->disk
) && !is_configured(&idisk
->disk
))
3091 dl
->next
= champion
->disks
;
3092 champion
->disks
= dl
;
3096 /* delete 'champion' from super_list */
3097 for (del
= super_list
; *del
; ) {
3098 if (*del
== champion
) {
3099 *del
= (*del
)->next
;
3102 del
= &(*del
)->next
;
3104 champion
->next
= NULL
;
3108 struct intel_disk
*idisk
= disk_list
;
3110 disk_list
= disk_list
->next
;
3117 static int load_super_imsm_all(struct supertype
*st
, int fd
, void **sbp
,
3121 struct intel_super
*super_list
= NULL
;
3122 struct intel_super
*super
= NULL
;
3123 int devnum
= fd2devnum(fd
);
3129 /* check if 'fd' an opened container */
3130 sra
= sysfs_read(fd
, 0, GET_LEVEL
|GET_VERSION
|GET_DEVS
|GET_STATE
);
3134 if (sra
->array
.major_version
!= -1 ||
3135 sra
->array
.minor_version
!= -2 ||
3136 strcmp(sra
->text_version
, "imsm") != 0) {
3141 for (sd
= sra
->devs
, i
= 0; sd
; sd
= sd
->next
, i
++) {
3142 struct intel_super
*s
= alloc_super();
3149 s
->next
= super_list
;
3153 sprintf(nm
, "%d:%d", sd
->disk
.major
, sd
->disk
.minor
);
3154 dfd
= dev_open(nm
, O_RDWR
);
3158 err
= load_and_parse_mpb(dfd
, s
, NULL
, 1);
3160 /* retry the load if we might have raced against mdmon */
3161 if (err
== 3 && mdmon_running(devnum
))
3162 for (retry
= 0; retry
< 3; retry
++) {
3164 err
= load_and_parse_mpb(dfd
, s
, NULL
, 1);
3172 /* all mpbs enter, maybe one leaves */
3173 super
= imsm_thunderdome(&super_list
, i
);
3179 if (find_missing(super
) != 0) {
3187 while (super_list
) {
3188 struct intel_super
*s
= super_list
;
3190 super_list
= super_list
->next
;
3199 st
->container_dev
= devnum
;
3200 if (err
== 0 && st
->ss
== NULL
) {
3201 st
->ss
= &super_imsm
;
3202 st
->minor_version
= 0;
3203 st
->max_devs
= IMSM_MAX_DEVICES
;
3208 static int load_container_imsm(struct supertype
*st
, int fd
, char *devname
)
3210 return load_super_imsm_all(st
, fd
, &st
->sb
, devname
);
3214 static int load_super_imsm(struct supertype
*st
, int fd
, char *devname
)
3216 struct intel_super
*super
;
3219 if (test_partition(fd
))
3220 /* IMSM not allowed on partitions */
3223 free_super_imsm(st
);
3225 super
= alloc_super();
3228 Name
": malloc of %zu failed.\n",
3233 rv
= load_and_parse_mpb(fd
, super
, devname
, 0);
3238 Name
": Failed to load all information "
3239 "sections on %s\n", devname
);
3245 if (st
->ss
== NULL
) {
3246 st
->ss
= &super_imsm
;
3247 st
->minor_version
= 0;
3248 st
->max_devs
= IMSM_MAX_DEVICES
;
3253 static __u16
info_to_blocks_per_strip(mdu_array_info_t
*info
)
3255 if (info
->level
== 1)
3257 return info
->chunk_size
>> 9;
3260 static __u32
info_to_num_data_stripes(mdu_array_info_t
*info
, int num_domains
)
3264 num_stripes
= (info
->size
* 2) / info_to_blocks_per_strip(info
);
3265 num_stripes
/= num_domains
;
3270 static __u32
info_to_blocks_per_member(mdu_array_info_t
*info
)
3272 if (info
->level
== 1)
3273 return info
->size
* 2;
3275 return (info
->size
* 2) & ~(info_to_blocks_per_strip(info
) - 1);
3278 static void imsm_update_version_info(struct intel_super
*super
)
3280 /* update the version and attributes */
3281 struct imsm_super
*mpb
= super
->anchor
;
3283 struct imsm_dev
*dev
;
3284 struct imsm_map
*map
;
3287 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
3288 dev
= get_imsm_dev(super
, i
);
3289 map
= get_imsm_map(dev
, 0);
3290 if (__le32_to_cpu(dev
->size_high
) > 0)
3291 mpb
->attributes
|= MPB_ATTRIB_2TB
;
3293 /* FIXME detect when an array spans a port multiplier */
3295 mpb
->attributes
|= MPB_ATTRIB_PM
;
3298 if (mpb
->num_raid_devs
> 1 ||
3299 mpb
->attributes
!= MPB_ATTRIB_CHECKSUM_VERIFY
) {
3300 version
= MPB_VERSION_ATTRIBS
;
3301 switch (get_imsm_raid_level(map
)) {
3302 case 0: mpb
->attributes
|= MPB_ATTRIB_RAID0
; break;
3303 case 1: mpb
->attributes
|= MPB_ATTRIB_RAID1
; break;
3304 case 10: mpb
->attributes
|= MPB_ATTRIB_RAID10
; break;
3305 case 5: mpb
->attributes
|= MPB_ATTRIB_RAID5
; break;
3308 if (map
->num_members
>= 5)
3309 version
= MPB_VERSION_5OR6_DISK_ARRAY
;
3310 else if (dev
->status
== DEV_CLONE_N_GO
)
3311 version
= MPB_VERSION_CNG
;
3312 else if (get_imsm_raid_level(map
) == 5)
3313 version
= MPB_VERSION_RAID5
;
3314 else if (map
->num_members
>= 3)
3315 version
= MPB_VERSION_3OR4_DISK_ARRAY
;
3316 else if (get_imsm_raid_level(map
) == 1)
3317 version
= MPB_VERSION_RAID1
;
3319 version
= MPB_VERSION_RAID0
;
3321 strcpy(((char *) mpb
->sig
) + strlen(MPB_SIGNATURE
), version
);
3325 static int check_name(struct intel_super
*super
, char *name
, int quiet
)
3327 struct imsm_super
*mpb
= super
->anchor
;
3328 char *reason
= NULL
;
3331 if (strlen(name
) > MAX_RAID_SERIAL_LEN
)
3332 reason
= "must be 16 characters or less";
3334 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
3335 struct imsm_dev
*dev
= get_imsm_dev(super
, i
);
3337 if (strncmp((char *) dev
->volume
, name
, MAX_RAID_SERIAL_LEN
) == 0) {
3338 reason
= "already exists";
3343 if (reason
&& !quiet
)
3344 fprintf(stderr
, Name
": imsm volume name %s\n", reason
);
3349 static int init_super_imsm_volume(struct supertype
*st
, mdu_array_info_t
*info
,
3350 unsigned long long size
, char *name
,
3351 char *homehost
, int *uuid
)
3353 /* We are creating a volume inside a pre-existing container.
3354 * so st->sb is already set.
3356 struct intel_super
*super
= st
->sb
;
3357 struct imsm_super
*mpb
= super
->anchor
;
3358 struct intel_dev
*dv
;
3359 struct imsm_dev
*dev
;
3360 struct imsm_vol
*vol
;
3361 struct imsm_map
*map
;
3362 int idx
= mpb
->num_raid_devs
;
3364 unsigned long long array_blocks
;
3365 size_t size_old
, size_new
;
3366 __u32 num_data_stripes
;
3368 if (super
->orom
&& mpb
->num_raid_devs
>= super
->orom
->vpa
) {
3369 fprintf(stderr
, Name
": This imsm-container already has the "
3370 "maximum of %d volumes\n", super
->orom
->vpa
);
3374 /* ensure the mpb is large enough for the new data */
3375 size_old
= __le32_to_cpu(mpb
->mpb_size
);
3376 size_new
= disks_to_mpb_size(info
->nr_disks
);
3377 if (size_new
> size_old
) {
3379 size_t size_round
= ROUND_UP(size_new
, 512);
3381 if (posix_memalign(&mpb_new
, 512, size_round
) != 0) {
3382 fprintf(stderr
, Name
": could not allocate new mpb\n");
3385 memcpy(mpb_new
, mpb
, size_old
);
3388 super
->anchor
= mpb_new
;
3389 mpb
->mpb_size
= __cpu_to_le32(size_new
);
3390 memset(mpb_new
+ size_old
, 0, size_round
- size_old
);
3392 super
->current_vol
= idx
;
3393 /* when creating the first raid device in this container set num_disks
3394 * to zero, i.e. delete this spare and add raid member devices in
3395 * add_to_super_imsm_volume()
3397 if (super
->current_vol
== 0)
3400 if (!check_name(super
, name
, 0))
3402 dv
= malloc(sizeof(*dv
));
3404 fprintf(stderr
, Name
": failed to allocate device list entry\n");
3407 dev
= malloc(sizeof(*dev
) + sizeof(__u32
) * (info
->raid_disks
- 1));
3410 fprintf(stderr
, Name
": could not allocate raid device\n");
3413 strncpy((char *) dev
->volume
, name
, MAX_RAID_SERIAL_LEN
);
3414 if (info
->level
== 1)
3415 array_blocks
= info_to_blocks_per_member(info
);
3417 array_blocks
= calc_array_size(info
->level
, info
->raid_disks
,
3418 info
->layout
, info
->chunk_size
,
3420 /* round array size down to closest MB */
3421 array_blocks
= (array_blocks
>> SECT_PER_MB_SHIFT
) << SECT_PER_MB_SHIFT
;
3423 dev
->size_low
= __cpu_to_le32((__u32
) array_blocks
);
3424 dev
->size_high
= __cpu_to_le32((__u32
) (array_blocks
>> 32));
3425 dev
->status
= __cpu_to_le32(0);
3426 dev
->reserved_blocks
= __cpu_to_le32(0);
3428 vol
->migr_state
= 0;
3429 set_migr_type(dev
, MIGR_INIT
);
3431 vol
->curr_migr_unit
= 0;
3432 map
= get_imsm_map(dev
, 0);
3433 map
->pba_of_lba0
= __cpu_to_le32(super
->create_offset
);
3434 map
->blocks_per_member
= __cpu_to_le32(info_to_blocks_per_member(info
));
3435 map
->blocks_per_strip
= __cpu_to_le16(info_to_blocks_per_strip(info
));
3436 map
->failed_disk_num
= ~0;
3437 map
->map_state
= info
->level
? IMSM_T_STATE_UNINITIALIZED
:
3438 IMSM_T_STATE_NORMAL
;
3441 if (info
->level
== 1 && info
->raid_disks
> 2) {
3444 fprintf(stderr
, Name
": imsm does not support more than 2 disks"
3445 "in a raid1 volume\n");
3449 map
->raid_level
= info
->level
;
3450 if (info
->level
== 10) {
3451 map
->raid_level
= 1;
3452 map
->num_domains
= info
->raid_disks
/ 2;
3453 } else if (info
->level
== 1)
3454 map
->num_domains
= info
->raid_disks
;
3456 map
->num_domains
= 1;
3458 num_data_stripes
= info_to_num_data_stripes(info
, map
->num_domains
);
3459 map
->num_data_stripes
= __cpu_to_le32(num_data_stripes
);
3461 map
->num_members
= info
->raid_disks
;
3462 for (i
= 0; i
< map
->num_members
; i
++) {
3463 /* initialized in add_to_super */
3464 set_imsm_ord_tbl_ent(map
, i
, IMSM_ORD_REBUILD
);
3466 mpb
->num_raid_devs
++;
3469 dv
->index
= super
->current_vol
;
3470 dv
->next
= super
->devlist
;
3471 super
->devlist
= dv
;
3473 imsm_update_version_info(super
);
3478 static int init_super_imsm(struct supertype
*st
, mdu_array_info_t
*info
,
3479 unsigned long long size
, char *name
,
3480 char *homehost
, int *uuid
)
3482 /* This is primarily called by Create when creating a new array.
3483 * We will then get add_to_super called for each component, and then
3484 * write_init_super called to write it out to each device.
3485 * For IMSM, Create can create on fresh devices or on a pre-existing
3487 * To create on a pre-existing array a different method will be called.
3488 * This one is just for fresh drives.
3490 struct intel_super
*super
;
3491 struct imsm_super
*mpb
;
3496 return init_super_imsm_volume(st
, info
, size
, name
, homehost
, uuid
);
3499 mpb_size
= disks_to_mpb_size(info
->nr_disks
);
3503 super
= alloc_super();
3504 if (super
&& posix_memalign(&super
->buf
, 512, mpb_size
) != 0) {
3509 fprintf(stderr
, Name
3510 ": %s could not allocate superblock\n", __func__
);
3513 memset(super
->buf
, 0, mpb_size
);
3515 mpb
->mpb_size
= __cpu_to_le32(mpb_size
);
3519 /* zeroing superblock */
3523 mpb
->attributes
= MPB_ATTRIB_CHECKSUM_VERIFY
;
3525 version
= (char *) mpb
->sig
;
3526 strcpy(version
, MPB_SIGNATURE
);
3527 version
+= strlen(MPB_SIGNATURE
);
3528 strcpy(version
, MPB_VERSION_RAID0
);
3534 static int add_to_super_imsm_volume(struct supertype
*st
, mdu_disk_info_t
*dk
,
3535 int fd
, char *devname
)
3537 struct intel_super
*super
= st
->sb
;
3538 struct imsm_super
*mpb
= super
->anchor
;
3540 struct imsm_dev
*dev
;
3541 struct imsm_map
*map
;
3544 dev
= get_imsm_dev(super
, super
->current_vol
);
3545 map
= get_imsm_map(dev
, 0);
3547 if (! (dk
->state
& (1<<MD_DISK_SYNC
))) {
3548 fprintf(stderr
, Name
": %s: Cannot add spare devices to IMSM volume\n",
3554 /* we're doing autolayout so grab the pre-marked (in
3555 * validate_geometry) raid_disk
3557 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
3558 if (dl
->raiddisk
== dk
->raid_disk
)
3561 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
3562 if (dl
->major
== dk
->major
&&
3563 dl
->minor
== dk
->minor
)
3568 fprintf(stderr
, Name
": %s is not a member of the same container\n", devname
);
3572 /* add a pristine spare to the metadata */
3573 if (dl
->index
< 0) {
3574 dl
->index
= super
->anchor
->num_disks
;
3575 super
->anchor
->num_disks
++;
3577 /* Check the device has not already been added */
3578 slot
= get_imsm_disk_slot(map
, dl
->index
);
3580 (get_imsm_ord_tbl_ent(dev
, slot
, -1) & IMSM_ORD_REBUILD
) == 0) {
3581 fprintf(stderr
, Name
": %s has been included in this array twice\n",
3585 set_imsm_ord_tbl_ent(map
, dk
->number
, dl
->index
);
3586 dl
->disk
.status
= CONFIGURED_DISK
;
3588 /* if we are creating the first raid device update the family number */
3589 if (super
->current_vol
== 0) {
3591 struct imsm_dev
*_dev
= __get_imsm_dev(mpb
, 0);
3592 struct imsm_disk
*_disk
= __get_imsm_disk(mpb
, dl
->index
);
3594 if (!_dev
|| !_disk
) {
3595 fprintf(stderr
, Name
": BUG mpb setup error\n");
3601 sum
+= __gen_imsm_checksum(mpb
);
3602 mpb
->family_num
= __cpu_to_le32(sum
);
3603 mpb
->orig_family_num
= mpb
->family_num
;
3610 static int add_to_super_imsm(struct supertype
*st
, mdu_disk_info_t
*dk
,
3611 int fd
, char *devname
)
3613 struct intel_super
*super
= st
->sb
;
3615 unsigned long long size
;
3620 /* If we are on an RAID enabled platform check that the disk is
3621 * attached to the raid controller.
3622 * We do not need to test disks attachment for container based additions,
3623 * they shall be already tested when container was created/assembled.
3625 if ((fd
!= -1) && !check_env("IMSM_NO_PLATFORM")) {
3626 struct sys_dev
*hba_name
;
3627 struct intel_hba
*hba
;
3629 hba_name
= find_disk_attached_hba(fd
, NULL
);
3632 Name
": %s is not attached to Intel(R) RAID controller.\n",
3633 devname
? : "disk");
3636 rv
= attach_hba_to_super(super
, hba_name
, devname
);
3639 fprintf(stderr
, Name
": %s is attached to Intel(R) %s RAID "
3640 "controller (%s),\n but the container is assigned to Intel(R) "
3641 "%s RAID controller (",
3643 get_sys_dev_type(hba_name
->type
),
3644 hba_name
->pci_id
? : "Err!",
3645 get_sys_dev_type(hba_name
->type
));
3649 fprintf(stderr
, "%s", hba
->pci_id
? : "Err!");
3651 fprintf(stderr
, ", ");
3655 fprintf(stderr
, ").\n"
3656 " Mixing devices attached to different controllers "
3657 "is not allowed.\n");
3658 free_sys_dev(&hba_name
);
3661 free_sys_dev(&hba_name
);
3664 if (super
->current_vol
>= 0)
3665 return add_to_super_imsm_volume(st
, dk
, fd
, devname
);
3668 dd
= malloc(sizeof(*dd
));
3671 Name
": malloc failed %s:%d.\n", __func__
, __LINE__
);
3674 memset(dd
, 0, sizeof(*dd
));
3675 dd
->major
= major(stb
.st_rdev
);
3676 dd
->minor
= minor(stb
.st_rdev
);
3678 dd
->devname
= devname
? strdup(devname
) : NULL
;
3681 dd
->action
= DISK_ADD
;
3682 rv
= imsm_read_serial(fd
, devname
, dd
->serial
);
3685 Name
": failed to retrieve scsi serial, aborting\n");
3690 get_dev_size(fd
, NULL
, &size
);
3692 serialcpy(dd
->disk
.serial
, dd
->serial
);
3693 dd
->disk
.total_blocks
= __cpu_to_le32(size
);
3694 dd
->disk
.status
= SPARE_DISK
;
3695 if (sysfs_disk_to_scsi_id(fd
, &id
) == 0)
3696 dd
->disk
.scsi_id
= __cpu_to_le32(id
);
3698 dd
->disk
.scsi_id
= __cpu_to_le32(0);
3700 if (st
->update_tail
) {
3701 dd
->next
= super
->disk_mgmt_list
;
3702 super
->disk_mgmt_list
= dd
;
3704 dd
->next
= super
->disks
;
3712 static int remove_from_super_imsm(struct supertype
*st
, mdu_disk_info_t
*dk
)
3714 struct intel_super
*super
= st
->sb
;
3717 /* remove from super works only in mdmon - for communication
3718 * manager - monitor. Check if communication memory buffer
3721 if (!st
->update_tail
) {
3723 Name
": %s shall be used in mdmon context only"
3724 "(line %d).\n", __func__
, __LINE__
);
3727 dd
= malloc(sizeof(*dd
));
3730 Name
": malloc failed %s:%d.\n", __func__
, __LINE__
);
3733 memset(dd
, 0, sizeof(*dd
));
3734 dd
->major
= dk
->major
;
3735 dd
->minor
= dk
->minor
;
3738 dd
->disk
.status
= SPARE_DISK
;
3739 dd
->action
= DISK_REMOVE
;
3741 dd
->next
= super
->disk_mgmt_list
;
3742 super
->disk_mgmt_list
= dd
;
3748 static int store_imsm_mpb(int fd
, struct imsm_super
*mpb
);
3752 struct imsm_super anchor
;
3753 } spare_record
__attribute__ ((aligned(512)));
3755 /* spare records have their own family number and do not have any defined raid
3758 static int write_super_imsm_spares(struct intel_super
*super
, int doclose
)
3760 struct imsm_super
*mpb
= super
->anchor
;
3761 struct imsm_super
*spare
= &spare_record
.anchor
;
3765 spare
->mpb_size
= __cpu_to_le32(sizeof(struct imsm_super
)),
3766 spare
->generation_num
= __cpu_to_le32(1UL),
3767 spare
->attributes
= MPB_ATTRIB_CHECKSUM_VERIFY
;
3768 spare
->num_disks
= 1,
3769 spare
->num_raid_devs
= 0,
3770 spare
->cache_size
= mpb
->cache_size
,
3771 spare
->pwr_cycle_count
= __cpu_to_le32(1),
3773 snprintf((char *) spare
->sig
, MAX_SIGNATURE_LENGTH
,
3774 MPB_SIGNATURE MPB_VERSION_RAID0
);
3776 for (d
= super
->disks
; d
; d
= d
->next
) {
3780 spare
->disk
[0] = d
->disk
;
3781 sum
= __gen_imsm_checksum(spare
);
3782 spare
->family_num
= __cpu_to_le32(sum
);
3783 spare
->orig_family_num
= 0;
3784 sum
= __gen_imsm_checksum(spare
);
3785 spare
->check_sum
= __cpu_to_le32(sum
);
3787 if (store_imsm_mpb(d
->fd
, spare
)) {
3788 fprintf(stderr
, "%s: failed for device %d:%d %s\n",
3789 __func__
, d
->major
, d
->minor
, strerror(errno
));
3801 static int write_super_imsm(struct supertype
*st
, int doclose
)
3803 struct intel_super
*super
= st
->sb
;
3804 struct imsm_super
*mpb
= super
->anchor
;
3810 __u32 mpb_size
= sizeof(struct imsm_super
) - sizeof(struct imsm_disk
);
3813 /* 'generation' is incremented everytime the metadata is written */
3814 generation
= __le32_to_cpu(mpb
->generation_num
);
3816 mpb
->generation_num
= __cpu_to_le32(generation
);
3818 /* fix up cases where previous mdadm releases failed to set
3821 if (mpb
->orig_family_num
== 0)
3822 mpb
->orig_family_num
= mpb
->family_num
;
3824 for (d
= super
->disks
; d
; d
= d
->next
) {
3828 mpb
->disk
[d
->index
] = d
->disk
;
3832 for (d
= super
->missing
; d
; d
= d
->next
) {
3833 mpb
->disk
[d
->index
] = d
->disk
;
3836 mpb
->num_disks
= num_disks
;
3837 mpb_size
+= sizeof(struct imsm_disk
) * mpb
->num_disks
;
3839 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
3840 struct imsm_dev
*dev
= __get_imsm_dev(mpb
, i
);
3841 struct imsm_dev
*dev2
= get_imsm_dev(super
, i
);
3843 imsm_copy_dev(dev
, dev2
);
3844 mpb_size
+= sizeof_imsm_dev(dev
, 0);
3847 mpb_size
+= __le32_to_cpu(mpb
->bbm_log_size
);
3848 mpb
->mpb_size
= __cpu_to_le32(mpb_size
);
3850 /* recalculate checksum */
3851 sum
= __gen_imsm_checksum(mpb
);
3852 mpb
->check_sum
= __cpu_to_le32(sum
);
3854 /* write the mpb for disks that compose raid devices */
3855 for (d
= super
->disks
; d
; d
= d
->next
) {
3858 if (store_imsm_mpb(d
->fd
, mpb
))
3859 fprintf(stderr
, "%s: failed for device %d:%d %s\n",
3860 __func__
, d
->major
, d
->minor
, strerror(errno
));
3868 return write_super_imsm_spares(super
, doclose
);
3874 static int create_array(struct supertype
*st
, int dev_idx
)
3877 struct imsm_update_create_array
*u
;
3878 struct intel_super
*super
= st
->sb
;
3879 struct imsm_dev
*dev
= get_imsm_dev(super
, dev_idx
);
3880 struct imsm_map
*map
= get_imsm_map(dev
, 0);
3881 struct disk_info
*inf
;
3882 struct imsm_disk
*disk
;
3885 len
= sizeof(*u
) - sizeof(*dev
) + sizeof_imsm_dev(dev
, 0) +
3886 sizeof(*inf
) * map
->num_members
;
3889 fprintf(stderr
, "%s: failed to allocate update buffer\n",
3894 u
->type
= update_create_array
;
3895 u
->dev_idx
= dev_idx
;
3896 imsm_copy_dev(&u
->dev
, dev
);
3897 inf
= get_disk_info(u
);
3898 for (i
= 0; i
< map
->num_members
; i
++) {
3899 int idx
= get_imsm_disk_idx(dev
, i
, -1);
3901 disk
= get_imsm_disk(super
, idx
);
3902 serialcpy(inf
[i
].serial
, disk
->serial
);
3904 append_metadata_update(st
, u
, len
);
3909 static int mgmt_disk(struct supertype
*st
)
3911 struct intel_super
*super
= st
->sb
;
3913 struct imsm_update_add_remove_disk
*u
;
3915 if (!super
->disk_mgmt_list
)
3921 fprintf(stderr
, "%s: failed to allocate update buffer\n",
3926 u
->type
= update_add_remove_disk
;
3927 append_metadata_update(st
, u
, len
);
3932 static int write_init_super_imsm(struct supertype
*st
)
3934 struct intel_super
*super
= st
->sb
;
3935 int current_vol
= super
->current_vol
;
3937 /* we are done with current_vol reset it to point st at the container */
3938 super
->current_vol
= -1;
3940 if (st
->update_tail
) {
3941 /* queue the recently created array / added disk
3942 * as a metadata update */
3945 /* determine if we are creating a volume or adding a disk */
3946 if (current_vol
< 0) {
3947 /* in the mgmt (add/remove) disk case we are running
3948 * in mdmon context, so don't close fd's
3950 return mgmt_disk(st
);
3952 rv
= create_array(st
, current_vol
);
3957 for (d
= super
->disks
; d
; d
= d
->next
)
3958 Kill(d
->devname
, NULL
, 0, 1, 1);
3959 return write_super_imsm(st
, 1);
3964 static int store_super_imsm(struct supertype
*st
, int fd
)
3966 struct intel_super
*super
= st
->sb
;
3967 struct imsm_super
*mpb
= super
? super
->anchor
: NULL
;
3973 return store_imsm_mpb(fd
, mpb
);
3979 static int imsm_bbm_log_size(struct imsm_super
*mpb
)
3981 return __le32_to_cpu(mpb
->bbm_log_size
);
3985 static int validate_geometry_imsm_container(struct supertype
*st
, int level
,
3986 int layout
, int raiddisks
, int chunk
,
3987 unsigned long long size
, char *dev
,
3988 unsigned long long *freesize
,
3992 unsigned long long ldsize
;
3993 const struct imsm_orom
*orom
;
3995 if (level
!= LEVEL_CONTAINER
)
4000 if (check_env("IMSM_NO_PLATFORM"))
4003 orom
= find_imsm_orom();
4004 if (orom
&& raiddisks
> orom
->tds
) {
4006 fprintf(stderr
, Name
": %d exceeds maximum number of"
4007 " platform supported disks: %d\n",
4008 raiddisks
, orom
->tds
);
4012 fd
= open(dev
, O_RDONLY
|O_EXCL
, 0);
4015 fprintf(stderr
, Name
": imsm: Cannot open %s: %s\n",
4016 dev
, strerror(errno
));
4019 if (!get_dev_size(fd
, dev
, &ldsize
)) {
4025 *freesize
= avail_size_imsm(st
, ldsize
>> 9);
4030 static unsigned long long find_size(struct extent
*e
, int *idx
, int num_extents
)
4032 const unsigned long long base_start
= e
[*idx
].start
;
4033 unsigned long long end
= base_start
+ e
[*idx
].size
;
4036 if (base_start
== end
)
4040 for (i
= *idx
; i
< num_extents
; i
++) {
4041 /* extend overlapping extents */
4042 if (e
[i
].start
>= base_start
&&
4043 e
[i
].start
<= end
) {
4046 if (e
[i
].start
+ e
[i
].size
> end
)
4047 end
= e
[i
].start
+ e
[i
].size
;
4048 } else if (e
[i
].start
> end
) {
4054 return end
- base_start
;
4057 static unsigned long long merge_extents(struct intel_super
*super
, int sum_extents
)
4059 /* build a composite disk with all known extents and generate a new
4060 * 'maxsize' given the "all disks in an array must share a common start
4061 * offset" constraint
4063 struct extent
*e
= calloc(sum_extents
, sizeof(*e
));
4067 unsigned long long pos
;
4068 unsigned long long start
= 0;
4069 unsigned long long maxsize
;
4070 unsigned long reserve
;
4075 /* coalesce and sort all extents. also, check to see if we need to
4076 * reserve space between member arrays
4079 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
4082 for (i
= 0; i
< dl
->extent_cnt
; i
++)
4085 qsort(e
, sum_extents
, sizeof(*e
), cmp_extent
);
4090 while (i
< sum_extents
) {
4091 e
[j
].start
= e
[i
].start
;
4092 e
[j
].size
= find_size(e
, &i
, sum_extents
);
4094 if (e
[j
-1].size
== 0)
4103 unsigned long long esize
;
4105 esize
= e
[i
].start
- pos
;
4106 if (esize
>= maxsize
) {
4111 pos
= e
[i
].start
+ e
[i
].size
;
4113 } while (e
[i
-1].size
);
4119 /* FIXME assumes volume at offset 0 is the first volume in a
4122 if (start_extent
> 0)
4123 reserve
= IMSM_RESERVED_SECTORS
; /* gap between raid regions */
4127 if (maxsize
< reserve
)
4130 super
->create_offset
= ~((__u32
) 0);
4131 if (start
+ reserve
> super
->create_offset
)
4132 return 0; /* start overflows create_offset */
4133 super
->create_offset
= start
+ reserve
;
4135 return maxsize
- reserve
;
4138 static int is_raid_level_supported(const struct imsm_orom
*orom
, int level
, int raiddisks
)
4140 if (level
< 0 || level
== 6 || level
== 4)
4143 /* if we have an orom prevent invalid raid levels */
4146 case 0: return imsm_orom_has_raid0(orom
);
4149 return imsm_orom_has_raid1e(orom
);
4150 return imsm_orom_has_raid1(orom
) && raiddisks
== 2;
4151 case 10: return imsm_orom_has_raid10(orom
) && raiddisks
== 4;
4152 case 5: return imsm_orom_has_raid5(orom
) && raiddisks
> 2;
4155 return 1; /* not on an Intel RAID platform so anything goes */
4160 #define pr_vrb(fmt, arg...) (void) (verbose && fprintf(stderr, Name fmt, ##arg))
4162 validate_geometry_imsm_orom(struct intel_super
*super
, int level
, int layout
,
4163 int raiddisks
, int chunk
, int verbose
)
4165 if (!is_raid_level_supported(super
->orom
, level
, raiddisks
)) {
4166 pr_vrb(": platform does not support raid%d with %d disk%s\n",
4167 level
, raiddisks
, raiddisks
> 1 ? "s" : "");
4170 if (super
->orom
&& level
!= 1 &&
4171 !imsm_orom_has_chunk(super
->orom
, chunk
)) {
4172 pr_vrb(": platform does not support a chunk size of: %d\n", chunk
);
4175 if (layout
!= imsm_level_to_layout(level
)) {
4177 pr_vrb(": imsm raid 5 only supports the left-asymmetric layout\n");
4178 else if (level
== 10)
4179 pr_vrb(": imsm raid 10 only supports the n2 layout\n");
4181 pr_vrb(": imsm unknown layout %#x for this raid level %d\n",
4189 /* validate_geometry_imsm_volume - lifted from validate_geometry_ddf_bvd
4190 * FIX ME add ahci details
4192 static int validate_geometry_imsm_volume(struct supertype
*st
, int level
,
4193 int layout
, int raiddisks
, int chunk
,
4194 unsigned long long size
, char *dev
,
4195 unsigned long long *freesize
,
4199 struct intel_super
*super
= st
->sb
;
4200 struct imsm_super
*mpb
= super
->anchor
;
4202 unsigned long long pos
= 0;
4203 unsigned long long maxsize
;
4207 /* We must have the container info already read in. */
4211 if (!validate_geometry_imsm_orom(super
, level
, layout
, raiddisks
, chunk
, verbose
))
4215 /* General test: make sure there is space for
4216 * 'raiddisks' device extents of size 'size' at a given
4219 unsigned long long minsize
= size
;
4220 unsigned long long start_offset
= MaxSector
;
4223 minsize
= MPB_SECTOR_CNT
+ IMSM_RESERVED_SECTORS
;
4224 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
4229 e
= get_extents(super
, dl
);
4232 unsigned long long esize
;
4233 esize
= e
[i
].start
- pos
;
4234 if (esize
>= minsize
)
4236 if (found
&& start_offset
== MaxSector
) {
4239 } else if (found
&& pos
!= start_offset
) {
4243 pos
= e
[i
].start
+ e
[i
].size
;
4245 } while (e
[i
-1].size
);
4250 if (dcnt
< raiddisks
) {
4252 fprintf(stderr
, Name
": imsm: Not enough "
4253 "devices with space for this array "
4261 /* This device must be a member of the set */
4262 if (stat(dev
, &stb
) < 0)
4264 if ((S_IFMT
& stb
.st_mode
) != S_IFBLK
)
4266 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
4267 if (dl
->major
== (int)major(stb
.st_rdev
) &&
4268 dl
->minor
== (int)minor(stb
.st_rdev
))
4273 fprintf(stderr
, Name
": %s is not in the "
4274 "same imsm set\n", dev
);
4276 } else if (super
->orom
&& dl
->index
< 0 && mpb
->num_raid_devs
) {
4277 /* If a volume is present then the current creation attempt
4278 * cannot incorporate new spares because the orom may not
4279 * understand this configuration (all member disks must be
4280 * members of each array in the container).
4282 fprintf(stderr
, Name
": %s is a spare and a volume"
4283 " is already defined for this container\n", dev
);
4284 fprintf(stderr
, Name
": The option-rom requires all member"
4285 " disks to be a member of all volumes\n");
4289 /* retrieve the largest free space block */
4290 e
= get_extents(super
, dl
);
4295 unsigned long long esize
;
4297 esize
= e
[i
].start
- pos
;
4298 if (esize
>= maxsize
)
4300 pos
= e
[i
].start
+ e
[i
].size
;
4302 } while (e
[i
-1].size
);
4307 fprintf(stderr
, Name
": unable to determine free space for: %s\n",
4311 if (maxsize
< size
) {
4313 fprintf(stderr
, Name
": %s not enough space (%llu < %llu)\n",
4314 dev
, maxsize
, size
);
4318 /* count total number of extents for merge */
4320 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
4322 i
+= dl
->extent_cnt
;
4324 maxsize
= merge_extents(super
, i
);
4325 if (maxsize
< size
|| maxsize
== 0) {
4327 fprintf(stderr
, Name
": not enough space after merge (%llu < %llu)\n",
4332 *freesize
= maxsize
;
4337 static int reserve_space(struct supertype
*st
, int raiddisks
,
4338 unsigned long long size
, int chunk
,
4339 unsigned long long *freesize
)
4341 struct intel_super
*super
= st
->sb
;
4342 struct imsm_super
*mpb
= super
->anchor
;
4347 unsigned long long maxsize
;
4348 unsigned long long minsize
;
4352 /* find the largest common start free region of the possible disks */
4356 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
4362 /* don't activate new spares if we are orom constrained
4363 * and there is already a volume active in the container
4365 if (super
->orom
&& dl
->index
< 0 && mpb
->num_raid_devs
)
4368 e
= get_extents(super
, dl
);
4371 for (i
= 1; e
[i
-1].size
; i
++)
4379 maxsize
= merge_extents(super
, extent_cnt
);
4384 if (cnt
< raiddisks
||
4385 (super
->orom
&& used
&& used
!= raiddisks
) ||
4386 maxsize
< minsize
||
4388 fprintf(stderr
, Name
": not enough devices with space to create array.\n");
4389 return 0; /* No enough free spaces large enough */
4401 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
4403 dl
->raiddisk
= cnt
++;
4410 static int validate_geometry_imsm(struct supertype
*st
, int level
, int layout
,
4411 int raiddisks
, int chunk
, unsigned long long size
,
4412 char *dev
, unsigned long long *freesize
,
4419 /* if given unused devices create a container
4420 * if given given devices in a container create a member volume
4422 if (level
== LEVEL_CONTAINER
) {
4423 /* Must be a fresh device to add to a container */
4424 return validate_geometry_imsm_container(st
, level
, layout
,
4425 raiddisks
, chunk
, size
,
4431 if (st
->sb
&& freesize
) {
4432 /* we are being asked to automatically layout a
4433 * new volume based on the current contents of
4434 * the container. If the the parameters can be
4435 * satisfied reserve_space will record the disks,
4436 * start offset, and size of the volume to be
4437 * created. add_to_super and getinfo_super
4438 * detect when autolayout is in progress.
4440 if (!validate_geometry_imsm_orom(st
->sb
, level
, layout
,
4444 return reserve_space(st
, raiddisks
, size
, chunk
, freesize
);
4449 /* creating in a given container */
4450 return validate_geometry_imsm_volume(st
, level
, layout
,
4451 raiddisks
, chunk
, size
,
4452 dev
, freesize
, verbose
);
4455 /* This device needs to be a device in an 'imsm' container */
4456 fd
= open(dev
, O_RDONLY
|O_EXCL
, 0);
4460 Name
": Cannot create this array on device %s\n",
4465 if (errno
!= EBUSY
|| (fd
= open(dev
, O_RDONLY
, 0)) < 0) {
4467 fprintf(stderr
, Name
": Cannot open %s: %s\n",
4468 dev
, strerror(errno
));
4471 /* Well, it is in use by someone, maybe an 'imsm' container. */
4472 cfd
= open_container(fd
);
4476 fprintf(stderr
, Name
": Cannot use %s: It is busy\n",
4480 sra
= sysfs_read(cfd
, 0, GET_VERSION
);
4481 if (sra
&& sra
->array
.major_version
== -1 &&
4482 strcmp(sra
->text_version
, "imsm") == 0)
4486 /* This is a member of a imsm container. Load the container
4487 * and try to create a volume
4489 struct intel_super
*super
;
4491 if (load_super_imsm_all(st
, cfd
, (void **) &super
, NULL
) == 0) {
4493 st
->container_dev
= fd2devnum(cfd
);
4495 return validate_geometry_imsm_volume(st
, level
, layout
,
4503 fprintf(stderr
, Name
": failed container membership check\n");
4509 static void default_geometry_imsm(struct supertype
*st
, int *level
, int *layout
, int *chunk
)
4511 struct intel_super
*super
= st
->sb
;
4513 if (level
&& *level
== UnSet
)
4514 *level
= LEVEL_CONTAINER
;
4516 if (level
&& layout
&& *layout
== UnSet
)
4517 *layout
= imsm_level_to_layout(*level
);
4519 if (chunk
&& (*chunk
== UnSet
|| *chunk
== 0) &&
4520 super
&& super
->orom
)
4521 *chunk
= imsm_orom_default_chunk(super
->orom
);
4524 static void handle_missing(struct intel_super
*super
, struct imsm_dev
*dev
);
4526 static int kill_subarray_imsm(struct supertype
*st
)
4528 /* remove the subarray currently referenced by ->current_vol */
4530 struct intel_dev
**dp
;
4531 struct intel_super
*super
= st
->sb
;
4532 __u8 current_vol
= super
->current_vol
;
4533 struct imsm_super
*mpb
= super
->anchor
;
4535 if (super
->current_vol
< 0)
4537 super
->current_vol
= -1; /* invalidate subarray cursor */
4539 /* block deletions that would change the uuid of active subarrays
4541 * FIXME when immutable ids are available, but note that we'll
4542 * also need to fixup the invalidated/active subarray indexes in
4545 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
4548 if (i
< current_vol
)
4550 sprintf(subarray
, "%u", i
);
4551 if (is_subarray_active(subarray
, st
->devname
)) {
4553 Name
": deleting subarray-%d would change the UUID of active subarray-%d, aborting\n",
4560 if (st
->update_tail
) {
4561 struct imsm_update_kill_array
*u
= malloc(sizeof(*u
));
4565 u
->type
= update_kill_array
;
4566 u
->dev_idx
= current_vol
;
4567 append_metadata_update(st
, u
, sizeof(*u
));
4572 for (dp
= &super
->devlist
; *dp
;)
4573 if ((*dp
)->index
== current_vol
) {
4576 handle_missing(super
, (*dp
)->dev
);
4577 if ((*dp
)->index
> current_vol
)
4582 /* no more raid devices, all active components are now spares,
4583 * but of course failed are still failed
4585 if (--mpb
->num_raid_devs
== 0) {
4588 for (d
= super
->disks
; d
; d
= d
->next
)
4589 if (d
->index
> -2) {
4591 d
->disk
.status
= SPARE_DISK
;
4595 super
->updates_pending
++;
4600 static int update_subarray_imsm(struct supertype
*st
, char *subarray
,
4601 char *update
, struct mddev_ident
*ident
)
4603 /* update the subarray currently referenced by ->current_vol */
4604 struct intel_super
*super
= st
->sb
;
4605 struct imsm_super
*mpb
= super
->anchor
;
4607 if (strcmp(update
, "name") == 0) {
4608 char *name
= ident
->name
;
4612 if (is_subarray_active(subarray
, st
->devname
)) {
4614 Name
": Unable to update name of active subarray\n");
4618 if (!check_name(super
, name
, 0))
4621 vol
= strtoul(subarray
, &ep
, 10);
4622 if (*ep
!= '\0' || vol
>= super
->anchor
->num_raid_devs
)
4625 if (st
->update_tail
) {
4626 struct imsm_update_rename_array
*u
= malloc(sizeof(*u
));
4630 u
->type
= update_rename_array
;
4632 snprintf((char *) u
->name
, MAX_RAID_SERIAL_LEN
, "%s", name
);
4633 append_metadata_update(st
, u
, sizeof(*u
));
4635 struct imsm_dev
*dev
;
4638 dev
= get_imsm_dev(super
, vol
);
4639 snprintf((char *) dev
->volume
, MAX_RAID_SERIAL_LEN
, "%s", name
);
4640 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
4641 dev
= get_imsm_dev(super
, i
);
4642 handle_missing(super
, dev
);
4644 super
->updates_pending
++;
4652 static int is_gen_migration(struct imsm_dev
*dev
)
4654 if (!dev
->vol
.migr_state
)
4657 if (migr_type(dev
) == MIGR_GEN_MIGR
)
4662 #endif /* MDASSEMBLE */
4664 static int is_rebuilding(struct imsm_dev
*dev
)
4666 struct imsm_map
*migr_map
;
4668 if (!dev
->vol
.migr_state
)
4671 if (migr_type(dev
) != MIGR_REBUILD
)
4674 migr_map
= get_imsm_map(dev
, 1);
4676 if (migr_map
->map_state
== IMSM_T_STATE_DEGRADED
)
4682 static void update_recovery_start(struct imsm_dev
*dev
, struct mdinfo
*array
)
4684 struct mdinfo
*rebuild
= NULL
;
4688 if (!is_rebuilding(dev
))
4691 /* Find the rebuild target, but punt on the dual rebuild case */
4692 for (d
= array
->devs
; d
; d
= d
->next
)
4693 if (d
->recovery_start
== 0) {
4700 /* (?) none of the disks are marked with
4701 * IMSM_ORD_REBUILD, so assume they are missing and the
4702 * disk_ord_tbl was not correctly updated
4704 dprintf("%s: failed to locate out-of-sync disk\n", __func__
);
4708 units
= __le32_to_cpu(dev
->vol
.curr_migr_unit
);
4709 rebuild
->recovery_start
= units
* blocks_per_migr_unit(dev
);
4713 static struct mdinfo
*container_content_imsm(struct supertype
*st
, char *subarray
)
4715 /* Given a container loaded by load_super_imsm_all,
4716 * extract information about all the arrays into
4718 * If 'subarray' is given, just extract info about that array.
4720 * For each imsm_dev create an mdinfo, fill it in,
4721 * then look for matching devices in super->disks
4722 * and create appropriate device mdinfo.
4724 struct intel_super
*super
= st
->sb
;
4725 struct imsm_super
*mpb
= super
->anchor
;
4726 struct mdinfo
*rest
= NULL
;
4730 /* check for bad blocks */
4731 if (imsm_bbm_log_size(super
->anchor
))
4734 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
4735 struct imsm_dev
*dev
;
4736 struct imsm_map
*map
;
4737 struct imsm_map
*map2
;
4738 struct mdinfo
*this;
4743 (i
!= strtoul(subarray
, &ep
, 10) || *ep
!= '\0'))
4746 dev
= get_imsm_dev(super
, i
);
4747 map
= get_imsm_map(dev
, 0);
4748 map2
= get_imsm_map(dev
, 1);
4750 /* do not publish arrays that are in the middle of an
4751 * unsupported migration
4753 if (dev
->vol
.migr_state
&&
4754 (migr_type(dev
) == MIGR_STATE_CHANGE
)) {
4755 fprintf(stderr
, Name
": cannot assemble volume '%.16s':"
4756 " unsupported migration in progress\n",
4761 this = malloc(sizeof(*this));
4763 fprintf(stderr
, Name
": failed to allocate %zu bytes\n",
4767 memset(this, 0, sizeof(*this));
4770 super
->current_vol
= i
;
4771 getinfo_super_imsm_volume(st
, this, NULL
);
4772 for (slot
= 0 ; slot
< map
->num_members
; slot
++) {
4773 unsigned long long recovery_start
;
4774 struct mdinfo
*info_d
;
4781 idx
= get_imsm_disk_idx(dev
, slot
, 0);
4782 ord
= get_imsm_ord_tbl_ent(dev
, slot
, 0);
4783 for (d
= super
->disks
; d
; d
= d
->next
)
4784 if (d
->index
== idx
)
4787 recovery_start
= MaxSector
;
4790 if (d
&& is_failed(&d
->disk
))
4792 if (ord
& IMSM_ORD_REBUILD
)
4796 * if we skip some disks the array will be assmebled degraded;
4797 * reset resync start to avoid a dirty-degraded
4798 * situation when performing the intial sync
4800 * FIXME handle dirty degraded
4802 if ((skip
|| recovery_start
== 0) && !dev
->vol
.dirty
)
4803 this->resync_start
= MaxSector
;
4807 info_d
= calloc(1, sizeof(*info_d
));
4809 fprintf(stderr
, Name
": failed to allocate disk"
4810 " for volume %.16s\n", dev
->volume
);
4811 info_d
= this->devs
;
4813 struct mdinfo
*d
= info_d
->next
;
4822 info_d
->next
= this->devs
;
4823 this->devs
= info_d
;
4825 info_d
->disk
.number
= d
->index
;
4826 info_d
->disk
.major
= d
->major
;
4827 info_d
->disk
.minor
= d
->minor
;
4828 info_d
->disk
.raid_disk
= slot
;
4829 info_d
->recovery_start
= recovery_start
;
4831 if (slot
< map2
->num_members
)
4832 info_d
->disk
.state
= (1 << MD_DISK_ACTIVE
);
4834 this->array
.spare_disks
++;
4836 if (slot
< map
->num_members
)
4837 info_d
->disk
.state
= (1 << MD_DISK_ACTIVE
);
4839 this->array
.spare_disks
++;
4841 if (info_d
->recovery_start
== MaxSector
)
4842 this->array
.working_disks
++;
4844 info_d
->events
= __le32_to_cpu(mpb
->generation_num
);
4845 info_d
->data_offset
= __le32_to_cpu(map
->pba_of_lba0
);
4846 info_d
->component_size
= __le32_to_cpu(map
->blocks_per_member
);
4848 /* now that the disk list is up-to-date fixup recovery_start */
4849 update_recovery_start(dev
, this);
4853 /* if array has bad blocks, set suitable bit in array status */
4855 rest
->array
.state
|= (1<<MD_SB_BBM_ERRORS
);
4861 static __u8
imsm_check_degraded(struct intel_super
*super
, struct imsm_dev
*dev
, int failed
)
4863 struct imsm_map
*map
= get_imsm_map(dev
, 0);
4866 return map
->map_state
== IMSM_T_STATE_UNINITIALIZED
?
4867 IMSM_T_STATE_UNINITIALIZED
: IMSM_T_STATE_NORMAL
;
4869 switch (get_imsm_raid_level(map
)) {
4871 return IMSM_T_STATE_FAILED
;
4874 if (failed
< map
->num_members
)
4875 return IMSM_T_STATE_DEGRADED
;
4877 return IMSM_T_STATE_FAILED
;
4882 * check to see if any mirrors have failed, otherwise we
4883 * are degraded. Even numbered slots are mirrored on
4887 /* gcc -Os complains that this is unused */
4888 int insync
= insync
;
4890 for (i
= 0; i
< map
->num_members
; i
++) {
4891 __u32 ord
= get_imsm_ord_tbl_ent(dev
, i
, -1);
4892 int idx
= ord_to_idx(ord
);
4893 struct imsm_disk
*disk
;
4895 /* reset the potential in-sync count on even-numbered
4896 * slots. num_copies is always 2 for imsm raid10
4901 disk
= get_imsm_disk(super
, idx
);
4902 if (!disk
|| is_failed(disk
) || ord
& IMSM_ORD_REBUILD
)
4905 /* no in-sync disks left in this mirror the
4909 return IMSM_T_STATE_FAILED
;
4912 return IMSM_T_STATE_DEGRADED
;
4916 return IMSM_T_STATE_DEGRADED
;
4918 return IMSM_T_STATE_FAILED
;
4924 return map
->map_state
;
4927 static int imsm_count_failed(struct intel_super
*super
, struct imsm_dev
*dev
)
4931 struct imsm_disk
*disk
;
4932 struct imsm_map
*map
= get_imsm_map(dev
, 0);
4933 struct imsm_map
*prev
= get_imsm_map(dev
, dev
->vol
.migr_state
);
4937 /* at the beginning of migration we set IMSM_ORD_REBUILD on
4938 * disks that are being rebuilt. New failures are recorded to
4939 * map[0]. So we look through all the disks we started with and
4940 * see if any failures are still present, or if any new ones
4943 * FIXME add support for online capacity expansion and
4944 * raid-level-migration
4946 for (i
= 0; i
< prev
->num_members
; i
++) {
4947 ord
= __le32_to_cpu(prev
->disk_ord_tbl
[i
]);
4948 ord
|= __le32_to_cpu(map
->disk_ord_tbl
[i
]);
4949 idx
= ord_to_idx(ord
);
4951 disk
= get_imsm_disk(super
, idx
);
4952 if (!disk
|| is_failed(disk
) || ord
& IMSM_ORD_REBUILD
)
4960 static int imsm_open_new(struct supertype
*c
, struct active_array
*a
,
4963 struct intel_super
*super
= c
->sb
;
4964 struct imsm_super
*mpb
= super
->anchor
;
4966 if (atoi(inst
) >= mpb
->num_raid_devs
) {
4967 fprintf(stderr
, "%s: subarry index %d, out of range\n",
4968 __func__
, atoi(inst
));
4972 dprintf("imsm: open_new %s\n", inst
);
4973 a
->info
.container_member
= atoi(inst
);
4977 static int is_resyncing(struct imsm_dev
*dev
)
4979 struct imsm_map
*migr_map
;
4981 if (!dev
->vol
.migr_state
)
4984 if (migr_type(dev
) == MIGR_INIT
||
4985 migr_type(dev
) == MIGR_REPAIR
)
4988 if (migr_type(dev
) == MIGR_GEN_MIGR
)
4991 migr_map
= get_imsm_map(dev
, 1);
4993 if ((migr_map
->map_state
== IMSM_T_STATE_NORMAL
) &&
4994 (dev
->vol
.migr_type
!= MIGR_GEN_MIGR
))
5000 /* return true if we recorded new information */
5001 static int mark_failure(struct imsm_dev
*dev
, struct imsm_disk
*disk
, int idx
)
5005 struct imsm_map
*map
;
5007 /* new failures are always set in map[0] */
5008 map
= get_imsm_map(dev
, 0);
5010 slot
= get_imsm_disk_slot(map
, idx
);
5014 ord
= __le32_to_cpu(map
->disk_ord_tbl
[slot
]);
5015 if (is_failed(disk
) && (ord
& IMSM_ORD_REBUILD
))
5018 disk
->status
|= FAILED_DISK
;
5019 disk
->status
&= ~CONFIGURED_DISK
;
5020 set_imsm_ord_tbl_ent(map
, slot
, idx
| IMSM_ORD_REBUILD
);
5021 if (map
->failed_disk_num
== 0xff)
5022 map
->failed_disk_num
= slot
;
5026 static void mark_missing(struct imsm_dev
*dev
, struct imsm_disk
*disk
, int idx
)
5028 mark_failure(dev
, disk
, idx
);
5030 if (disk
->scsi_id
== __cpu_to_le32(~(__u32
)0))
5033 disk
->scsi_id
= __cpu_to_le32(~(__u32
)0);
5034 memmove(&disk
->serial
[0], &disk
->serial
[1], MAX_RAID_SERIAL_LEN
- 1);
5037 static void handle_missing(struct intel_super
*super
, struct imsm_dev
*dev
)
5043 if (!super
->missing
)
5045 failed
= imsm_count_failed(super
, dev
);
5046 map_state
= imsm_check_degraded(super
, dev
, failed
);
5048 dprintf("imsm: mark missing\n");
5049 end_migration(dev
, map_state
);
5050 for (dl
= super
->missing
; dl
; dl
= dl
->next
)
5051 mark_missing(dev
, &dl
->disk
, dl
->index
);
5052 super
->updates_pending
++;
5055 static unsigned long long imsm_set_array_size(struct imsm_dev
*dev
)
5057 int used_disks
= imsm_num_data_members(dev
, 0);
5058 unsigned long long array_blocks
;
5059 struct imsm_map
*map
;
5061 if (used_disks
== 0) {
5062 /* when problems occures
5063 * return current array_blocks value
5065 array_blocks
= __le32_to_cpu(dev
->size_high
);
5066 array_blocks
= array_blocks
<< 32;
5067 array_blocks
+= __le32_to_cpu(dev
->size_low
);
5069 return array_blocks
;
5072 /* set array size in metadata
5074 map
= get_imsm_map(dev
, 0);
5075 array_blocks
= map
->blocks_per_member
* used_disks
;
5077 /* round array size down to closest MB
5079 array_blocks
= (array_blocks
>> SECT_PER_MB_SHIFT
) << SECT_PER_MB_SHIFT
;
5080 dev
->size_low
= __cpu_to_le32((__u32
)array_blocks
);
5081 dev
->size_high
= __cpu_to_le32((__u32
)(array_blocks
>> 32));
5083 return array_blocks
;
5086 static void imsm_set_disk(struct active_array
*a
, int n
, int state
);
5088 static void imsm_progress_container_reshape(struct intel_super
*super
)
5090 /* if no device has a migr_state, but some device has a
5091 * different number of members than the previous device, start
5092 * changing the number of devices in this device to match
5095 struct imsm_super
*mpb
= super
->anchor
;
5096 int prev_disks
= -1;
5100 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
5101 struct imsm_dev
*dev
= get_imsm_dev(super
, i
);
5102 struct imsm_map
*map
= get_imsm_map(dev
, 0);
5103 struct imsm_map
*map2
;
5104 int prev_num_members
;
5106 if (dev
->vol
.migr_state
)
5109 if (prev_disks
== -1)
5110 prev_disks
= map
->num_members
;
5111 if (prev_disks
== map
->num_members
)
5114 /* OK, this array needs to enter reshape mode.
5115 * i.e it needs a migr_state
5118 copy_map_size
= sizeof_imsm_map(map
);
5119 prev_num_members
= map
->num_members
;
5120 map
->num_members
= prev_disks
;
5121 dev
->vol
.migr_state
= 1;
5122 dev
->vol
.curr_migr_unit
= 0;
5123 dev
->vol
.migr_type
= MIGR_GEN_MIGR
;
5124 for (i
= prev_num_members
;
5125 i
< map
->num_members
; i
++)
5126 set_imsm_ord_tbl_ent(map
, i
, i
);
5127 map2
= get_imsm_map(dev
, 1);
5128 /* Copy the current map */
5129 memcpy(map2
, map
, copy_map_size
);
5130 map2
->num_members
= prev_num_members
;
5132 imsm_set_array_size(dev
);
5133 super
->updates_pending
++;
5137 /* Handle dirty -> clean transititions, resync and reshape. Degraded and rebuild
5138 * states are handled in imsm_set_disk() with one exception, when a
5139 * resync is stopped due to a new failure this routine will set the
5140 * 'degraded' state for the array.
5142 static int imsm_set_array_state(struct active_array
*a
, int consistent
)
5144 int inst
= a
->info
.container_member
;
5145 struct intel_super
*super
= a
->container
->sb
;
5146 struct imsm_dev
*dev
= get_imsm_dev(super
, inst
);
5147 struct imsm_map
*map
= get_imsm_map(dev
, 0);
5148 int failed
= imsm_count_failed(super
, dev
);
5149 __u8 map_state
= imsm_check_degraded(super
, dev
, failed
);
5150 __u32 blocks_per_unit
;
5152 if (dev
->vol
.migr_state
&&
5153 dev
->vol
.migr_type
== MIGR_GEN_MIGR
) {
5154 /* array state change is blocked due to reshape action
5156 * - abort the reshape (if last_checkpoint is 0 and action!= reshape)
5157 * - finish the reshape (if last_checkpoint is big and action != reshape)
5158 * - update curr_migr_unit
5160 if (a
->curr_action
== reshape
) {
5161 /* still reshaping, maybe update curr_migr_unit */
5162 long long blocks_per_unit
= blocks_per_migr_unit(dev
);
5163 long long unit
= a
->last_checkpoint
;
5164 if (blocks_per_unit
) {
5165 unit
/= blocks_per_unit
;
5167 __le32_to_cpu(dev
->vol
.curr_migr_unit
)) {
5168 dev
->vol
.curr_migr_unit
=
5169 __cpu_to_le32(unit
);
5170 super
->updates_pending
++;
5175 if (a
->last_checkpoint
== 0 && a
->prev_action
== reshape
) {
5176 /* for some reason we aborted the reshape.
5179 struct imsm_map
*map2
= get_imsm_map(dev
, 1);
5180 dev
->vol
.migr_state
= 0;
5181 dev
->vol
.migr_type
= 0;
5182 dev
->vol
.curr_migr_unit
= 0;
5183 memcpy(map
, map2
, sizeof_imsm_map(map2
));
5184 super
->updates_pending
++;
5186 if (a
->last_checkpoint
>= a
->info
.component_size
) {
5187 unsigned long long array_blocks
;
5191 used_disks
= imsm_num_data_members(dev
, 0);
5192 if (used_disks
> 0) {
5194 map
->blocks_per_member
*
5196 /* round array size down to closest MB
5198 array_blocks
= (array_blocks
5199 >> SECT_PER_MB_SHIFT
)
5200 << SECT_PER_MB_SHIFT
;
5201 a
->info
.custom_array_size
= array_blocks
;
5202 /* encourage manager to update array
5206 a
->check_reshape
= 1;
5208 /* finalize online capacity expansion/reshape */
5209 for (mdi
= a
->info
.devs
; mdi
; mdi
= mdi
->next
)
5211 mdi
->disk
.raid_disk
,
5214 imsm_progress_container_reshape(super
);
5219 /* before we activate this array handle any missing disks */
5220 if (consistent
== 2)
5221 handle_missing(super
, dev
);
5223 if (consistent
== 2 &&
5224 (!is_resync_complete(&a
->info
) ||
5225 map_state
!= IMSM_T_STATE_NORMAL
||
5226 dev
->vol
.migr_state
))
5229 if (is_resync_complete(&a
->info
)) {
5230 /* complete intialization / resync,
5231 * recovery and interrupted recovery is completed in
5234 if (is_resyncing(dev
)) {
5235 dprintf("imsm: mark resync done\n");
5236 end_migration(dev
, map_state
);
5237 super
->updates_pending
++;
5238 a
->last_checkpoint
= 0;
5240 } else if (!is_resyncing(dev
) && !failed
) {
5241 /* mark the start of the init process if nothing is failed */
5242 dprintf("imsm: mark resync start\n");
5243 if (map
->map_state
== IMSM_T_STATE_UNINITIALIZED
)
5244 migrate(dev
, IMSM_T_STATE_NORMAL
, MIGR_INIT
);
5246 migrate(dev
, IMSM_T_STATE_NORMAL
, MIGR_REPAIR
);
5247 super
->updates_pending
++;
5250 /* check if we can update curr_migr_unit from resync_start, recovery_start */
5251 blocks_per_unit
= blocks_per_migr_unit(dev
);
5252 if (blocks_per_unit
) {
5256 units
= a
->last_checkpoint
/ blocks_per_unit
;
5259 /* check that we did not overflow 32-bits, and that
5260 * curr_migr_unit needs updating
5262 if (units32
== units
&&
5263 __le32_to_cpu(dev
->vol
.curr_migr_unit
) != units32
) {
5264 dprintf("imsm: mark checkpoint (%u)\n", units32
);
5265 dev
->vol
.curr_migr_unit
= __cpu_to_le32(units32
);
5266 super
->updates_pending
++;
5270 /* mark dirty / clean */
5271 if (dev
->vol
.dirty
!= !consistent
) {
5272 dprintf("imsm: mark '%s'\n", consistent
? "clean" : "dirty");
5277 super
->updates_pending
++;
5283 static void imsm_set_disk(struct active_array
*a
, int n
, int state
)
5285 int inst
= a
->info
.container_member
;
5286 struct intel_super
*super
= a
->container
->sb
;
5287 struct imsm_dev
*dev
= get_imsm_dev(super
, inst
);
5288 struct imsm_map
*map
= get_imsm_map(dev
, 0);
5289 struct imsm_disk
*disk
;
5294 if (n
> map
->num_members
)
5295 fprintf(stderr
, "imsm: set_disk %d out of range 0..%d\n",
5296 n
, map
->num_members
- 1);
5301 dprintf("imsm: set_disk %d:%x\n", n
, state
);
5303 ord
= get_imsm_ord_tbl_ent(dev
, n
, -1);
5304 disk
= get_imsm_disk(super
, ord_to_idx(ord
));
5306 /* check for new failures */
5307 if (state
& DS_FAULTY
) {
5308 if (mark_failure(dev
, disk
, ord_to_idx(ord
)))
5309 super
->updates_pending
++;
5312 /* check if in_sync */
5313 if (state
& DS_INSYNC
&& ord
& IMSM_ORD_REBUILD
&& is_rebuilding(dev
)) {
5314 struct imsm_map
*migr_map
= get_imsm_map(dev
, 1);
5316 set_imsm_ord_tbl_ent(migr_map
, n
, ord_to_idx(ord
));
5317 super
->updates_pending
++;
5320 failed
= imsm_count_failed(super
, dev
);
5321 map_state
= imsm_check_degraded(super
, dev
, failed
);
5323 /* check if recovery complete, newly degraded, or failed */
5324 if (map_state
== IMSM_T_STATE_NORMAL
&& is_rebuilding(dev
)) {
5325 end_migration(dev
, map_state
);
5326 map
= get_imsm_map(dev
, 0);
5327 map
->failed_disk_num
= ~0;
5328 super
->updates_pending
++;
5329 a
->last_checkpoint
= 0;
5330 } else if (map_state
== IMSM_T_STATE_DEGRADED
&&
5331 map
->map_state
!= map_state
&&
5332 !dev
->vol
.migr_state
) {
5333 dprintf("imsm: mark degraded\n");
5334 map
->map_state
= map_state
;
5335 super
->updates_pending
++;
5336 a
->last_checkpoint
= 0;
5337 } else if (map_state
== IMSM_T_STATE_FAILED
&&
5338 map
->map_state
!= map_state
) {
5339 dprintf("imsm: mark failed\n");
5340 end_migration(dev
, map_state
);
5341 super
->updates_pending
++;
5342 a
->last_checkpoint
= 0;
5343 } else if (is_gen_migration(dev
)) {
5344 dprintf("imsm: Detected General Migration in state: ");
5345 if (map_state
== IMSM_T_STATE_NORMAL
) {
5346 end_migration(dev
, map_state
);
5347 map
= get_imsm_map(dev
, 0);
5348 map
->failed_disk_num
= ~0;
5349 dprintf("normal\n");
5351 if (map_state
== IMSM_T_STATE_DEGRADED
) {
5352 printf("degraded\n");
5353 end_migration(dev
, map_state
);
5355 dprintf("failed\n");
5357 map
->map_state
= map_state
;
5359 super
->updates_pending
++;
5363 static int store_imsm_mpb(int fd
, struct imsm_super
*mpb
)
5366 __u32 mpb_size
= __le32_to_cpu(mpb
->mpb_size
);
5367 unsigned long long dsize
;
5368 unsigned long long sectors
;
5370 get_dev_size(fd
, NULL
, &dsize
);
5372 if (mpb_size
> 512) {
5373 /* -1 to account for anchor */
5374 sectors
= mpb_sectors(mpb
) - 1;
5376 /* write the extended mpb to the sectors preceeding the anchor */
5377 if (lseek64(fd
, dsize
- (512 * (2 + sectors
)), SEEK_SET
) < 0)
5380 if ((unsigned long long)write(fd
, buf
+ 512, 512 * sectors
)
5385 /* first block is stored on second to last sector of the disk */
5386 if (lseek64(fd
, dsize
- (512 * 2), SEEK_SET
) < 0)
5389 if (write(fd
, buf
, 512) != 512)
5395 static void imsm_sync_metadata(struct supertype
*container
)
5397 struct intel_super
*super
= container
->sb
;
5399 dprintf("sync metadata: %d\n", super
->updates_pending
);
5400 if (!super
->updates_pending
)
5403 write_super_imsm(container
, 0);
5405 super
->updates_pending
= 0;
5408 static struct dl
*imsm_readd(struct intel_super
*super
, int idx
, struct active_array
*a
)
5410 struct imsm_dev
*dev
= get_imsm_dev(super
, a
->info
.container_member
);
5411 int i
= get_imsm_disk_idx(dev
, idx
, -1);
5414 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
5418 if (dl
&& is_failed(&dl
->disk
))
5422 dprintf("%s: found %x:%x\n", __func__
, dl
->major
, dl
->minor
);
5427 static struct dl
*imsm_add_spare(struct intel_super
*super
, int slot
,
5428 struct active_array
*a
, int activate_new
,
5429 struct mdinfo
*additional_test_list
)
5431 struct imsm_dev
*dev
= get_imsm_dev(super
, a
->info
.container_member
);
5432 int idx
= get_imsm_disk_idx(dev
, slot
, -1);
5433 struct imsm_super
*mpb
= super
->anchor
;
5434 struct imsm_map
*map
;
5435 unsigned long long pos
;
5440 __u32 array_start
= 0;
5441 __u32 array_end
= 0;
5443 struct mdinfo
*test_list
;
5445 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
5446 /* If in this array, skip */
5447 for (d
= a
->info
.devs
; d
; d
= d
->next
)
5448 if (d
->state_fd
>= 0 &&
5449 d
->disk
.major
== dl
->major
&&
5450 d
->disk
.minor
== dl
->minor
) {
5451 dprintf("%x:%x already in array\n",
5452 dl
->major
, dl
->minor
);
5457 test_list
= additional_test_list
;
5459 if (test_list
->disk
.major
== dl
->major
&&
5460 test_list
->disk
.minor
== dl
->minor
) {
5461 dprintf("%x:%x already in additional test list\n",
5462 dl
->major
, dl
->minor
);
5465 test_list
= test_list
->next
;
5470 /* skip in use or failed drives */
5471 if (is_failed(&dl
->disk
) || idx
== dl
->index
||
5473 dprintf("%x:%x status (failed: %d index: %d)\n",
5474 dl
->major
, dl
->minor
, is_failed(&dl
->disk
), idx
);
5478 /* skip pure spares when we are looking for partially
5479 * assimilated drives
5481 if (dl
->index
== -1 && !activate_new
)
5484 /* Does this unused device have the requisite free space?
5485 * It needs to be able to cover all member volumes
5487 ex
= get_extents(super
, dl
);
5489 dprintf("cannot get extents\n");
5492 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
5493 dev
= get_imsm_dev(super
, i
);
5494 map
= get_imsm_map(dev
, 0);
5496 /* check if this disk is already a member of
5499 if (get_imsm_disk_slot(map
, dl
->index
) >= 0)
5505 array_start
= __le32_to_cpu(map
->pba_of_lba0
);
5506 array_end
= array_start
+
5507 __le32_to_cpu(map
->blocks_per_member
) - 1;
5510 /* check that we can start at pba_of_lba0 with
5511 * blocks_per_member of space
5513 if (array_start
>= pos
&& array_end
< ex
[j
].start
) {
5517 pos
= ex
[j
].start
+ ex
[j
].size
;
5519 } while (ex
[j
-1].size
);
5526 if (i
< mpb
->num_raid_devs
) {
5527 dprintf("%x:%x does not have %u to %u available\n",
5528 dl
->major
, dl
->minor
, array_start
, array_end
);
5539 static int imsm_rebuild_allowed(struct supertype
*cont
, int dev_idx
, int failed
)
5541 struct imsm_dev
*dev2
;
5542 struct imsm_map
*map
;
5548 dev2
= get_imsm_dev(cont
->sb
, dev_idx
);
5550 state
= imsm_check_degraded(cont
->sb
, dev2
, failed
);
5551 if (state
== IMSM_T_STATE_FAILED
) {
5552 map
= get_imsm_map(dev2
, 0);
5555 for (slot
= 0; slot
< map
->num_members
; slot
++) {
5557 * Check if failed disks are deleted from intel
5558 * disk list or are marked to be deleted
5560 idx
= get_imsm_disk_idx(dev2
, slot
, -1);
5561 idisk
= get_imsm_dl_disk(cont
->sb
, idx
);
5563 * Do not rebuild the array if failed disks
5564 * from failed sub-array are not removed from
5568 is_failed(&idisk
->disk
) &&
5569 (idisk
->action
!= DISK_REMOVE
))
5577 static struct mdinfo
*imsm_activate_spare(struct active_array
*a
,
5578 struct metadata_update
**updates
)
5581 * Find a device with unused free space and use it to replace a
5582 * failed/vacant region in an array. We replace failed regions one a
5583 * array at a time. The result is that a new spare disk will be added
5584 * to the first failed array and after the monitor has finished
5585 * propagating failures the remainder will be consumed.
5587 * FIXME add a capability for mdmon to request spares from another
5591 struct intel_super
*super
= a
->container
->sb
;
5592 int inst
= a
->info
.container_member
;
5593 struct imsm_dev
*dev
= get_imsm_dev(super
, inst
);
5594 struct imsm_map
*map
= get_imsm_map(dev
, 0);
5595 int failed
= a
->info
.array
.raid_disks
;
5596 struct mdinfo
*rv
= NULL
;
5599 struct metadata_update
*mu
;
5601 struct imsm_update_activate_spare
*u
;
5606 for (d
= a
->info
.devs
; d
; d
= d
->next
) {
5607 if ((d
->curr_state
& DS_FAULTY
) &&
5609 /* wait for Removal to happen */
5611 if (d
->state_fd
>= 0)
5615 dprintf("imsm: activate spare: inst=%d failed=%d (%d) level=%d\n",
5616 inst
, failed
, a
->info
.array
.raid_disks
, a
->info
.array
.level
);
5618 if (dev
->vol
.migr_state
&&
5619 dev
->vol
.migr_type
== MIGR_GEN_MIGR
)
5620 /* No repair during migration */
5623 if (a
->info
.array
.level
== 4)
5624 /* No repair for takeovered array
5625 * imsm doesn't support raid4
5629 if (imsm_check_degraded(super
, dev
, failed
) != IMSM_T_STATE_DEGRADED
)
5633 * If there are any failed disks check state of the other volume.
5634 * Block rebuild if the another one is failed until failed disks
5635 * are removed from container.
5638 dprintf("found failed disks in %s, check if there another"
5639 "failed sub-array.\n",
5641 /* check if states of the other volumes allow for rebuild */
5642 for (i
= 0; i
< super
->anchor
->num_raid_devs
; i
++) {
5644 allowed
= imsm_rebuild_allowed(a
->container
,
5652 /* For each slot, if it is not working, find a spare */
5653 for (i
= 0; i
< a
->info
.array
.raid_disks
; i
++) {
5654 for (d
= a
->info
.devs
; d
; d
= d
->next
)
5655 if (d
->disk
.raid_disk
== i
)
5657 dprintf("found %d: %p %x\n", i
, d
, d
?d
->curr_state
:0);
5658 if (d
&& (d
->state_fd
>= 0))
5662 * OK, this device needs recovery. Try to re-add the
5663 * previous occupant of this slot, if this fails see if
5664 * we can continue the assimilation of a spare that was
5665 * partially assimilated, finally try to activate a new
5668 dl
= imsm_readd(super
, i
, a
);
5670 dl
= imsm_add_spare(super
, i
, a
, 0, NULL
);
5672 dl
= imsm_add_spare(super
, i
, a
, 1, NULL
);
5676 /* found a usable disk with enough space */
5677 di
= malloc(sizeof(*di
));
5680 memset(di
, 0, sizeof(*di
));
5682 /* dl->index will be -1 in the case we are activating a
5683 * pristine spare. imsm_process_update() will create a
5684 * new index in this case. Once a disk is found to be
5685 * failed in all member arrays it is kicked from the
5688 di
->disk
.number
= dl
->index
;
5690 /* (ab)use di->devs to store a pointer to the device
5693 di
->devs
= (struct mdinfo
*) dl
;
5695 di
->disk
.raid_disk
= i
;
5696 di
->disk
.major
= dl
->major
;
5697 di
->disk
.minor
= dl
->minor
;
5699 di
->recovery_start
= 0;
5700 di
->data_offset
= __le32_to_cpu(map
->pba_of_lba0
);
5701 di
->component_size
= a
->info
.component_size
;
5702 di
->container_member
= inst
;
5703 super
->random
= random32();
5707 dprintf("%x:%x to be %d at %llu\n", dl
->major
, dl
->minor
,
5708 i
, di
->data_offset
);
5714 /* No spares found */
5716 /* Now 'rv' has a list of devices to return.
5717 * Create a metadata_update record to update the
5718 * disk_ord_tbl for the array
5720 mu
= malloc(sizeof(*mu
));
5722 mu
->buf
= malloc(sizeof(struct imsm_update_activate_spare
) * num_spares
);
5723 if (mu
->buf
== NULL
) {
5730 struct mdinfo
*n
= rv
->next
;
5739 mu
->space_list
= NULL
;
5740 mu
->len
= sizeof(struct imsm_update_activate_spare
) * num_spares
;
5741 mu
->next
= *updates
;
5742 u
= (struct imsm_update_activate_spare
*) mu
->buf
;
5744 for (di
= rv
; di
; di
= di
->next
) {
5745 u
->type
= update_activate_spare
;
5746 u
->dl
= (struct dl
*) di
->devs
;
5748 u
->slot
= di
->disk
.raid_disk
;
5759 static int disks_overlap(struct intel_super
*super
, int idx
, struct imsm_update_create_array
*u
)
5761 struct imsm_dev
*dev
= get_imsm_dev(super
, idx
);
5762 struct imsm_map
*map
= get_imsm_map(dev
, 0);
5763 struct imsm_map
*new_map
= get_imsm_map(&u
->dev
, 0);
5764 struct disk_info
*inf
= get_disk_info(u
);
5765 struct imsm_disk
*disk
;
5769 for (i
= 0; i
< map
->num_members
; i
++) {
5770 disk
= get_imsm_disk(super
, get_imsm_disk_idx(dev
, i
, -1));
5771 for (j
= 0; j
< new_map
->num_members
; j
++)
5772 if (serialcmp(disk
->serial
, inf
[j
].serial
) == 0)
5780 static struct dl
*get_disk_super(struct intel_super
*super
, int major
, int minor
)
5782 struct dl
*dl
= NULL
;
5783 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
5784 if ((dl
->major
== major
) && (dl
->minor
== minor
))
5789 static int remove_disk_super(struct intel_super
*super
, int major
, int minor
)
5791 struct dl
*prev
= NULL
;
5795 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
5796 if ((dl
->major
== major
) && (dl
->minor
== minor
)) {
5799 prev
->next
= dl
->next
;
5801 super
->disks
= dl
->next
;
5803 __free_imsm_disk(dl
);
5804 dprintf("%s: removed %x:%x\n",
5805 __func__
, major
, minor
);
5813 static void imsm_delete(struct intel_super
*super
, struct dl
**dlp
, unsigned index
);
5815 static int add_remove_disk_update(struct intel_super
*super
)
5817 int check_degraded
= 0;
5818 struct dl
*disk
= NULL
;
5819 /* add/remove some spares to/from the metadata/contrainer */
5820 while (super
->disk_mgmt_list
) {
5821 struct dl
*disk_cfg
;
5823 disk_cfg
= super
->disk_mgmt_list
;
5824 super
->disk_mgmt_list
= disk_cfg
->next
;
5825 disk_cfg
->next
= NULL
;
5827 if (disk_cfg
->action
== DISK_ADD
) {
5828 disk_cfg
->next
= super
->disks
;
5829 super
->disks
= disk_cfg
;
5831 dprintf("%s: added %x:%x\n",
5832 __func__
, disk_cfg
->major
,
5834 } else if (disk_cfg
->action
== DISK_REMOVE
) {
5835 dprintf("Disk remove action processed: %x.%x\n",
5836 disk_cfg
->major
, disk_cfg
->minor
);
5837 disk
= get_disk_super(super
,
5841 /* store action status */
5842 disk
->action
= DISK_REMOVE
;
5843 /* remove spare disks only */
5844 if (disk
->index
== -1) {
5845 remove_disk_super(super
,
5850 /* release allocate disk structure */
5851 __free_imsm_disk(disk_cfg
);
5854 return check_degraded
;
5857 static int apply_reshape_container_disks_update(struct imsm_update_reshape
*u
,
5858 struct intel_super
*super
,
5861 struct dl
*new_disk
;
5862 struct intel_dev
*id
;
5864 int delta_disks
= u
->new_raid_disks
- u
->old_raid_disks
;
5865 int disk_count
= u
->old_raid_disks
;
5866 void **tofree
= NULL
;
5867 int devices_to_reshape
= 1;
5868 struct imsm_super
*mpb
= super
->anchor
;
5870 unsigned int dev_id
;
5872 dprintf("imsm: apply_reshape_container_disks_update()\n");
5874 /* enable spares to use in array */
5875 for (i
= 0; i
< delta_disks
; i
++) {
5876 new_disk
= get_disk_super(super
,
5877 major(u
->new_disks
[i
]),
5878 minor(u
->new_disks
[i
]));
5879 dprintf("imsm: new disk for reshape is: %i:%i "
5880 "(%p, index = %i)\n",
5881 major(u
->new_disks
[i
]), minor(u
->new_disks
[i
]),
5882 new_disk
, new_disk
->index
);
5883 if ((new_disk
== NULL
) ||
5884 ((new_disk
->index
>= 0) &&
5885 (new_disk
->index
< u
->old_raid_disks
)))
5886 goto update_reshape_exit
;
5887 new_disk
->index
= disk_count
++;
5888 /* slot to fill in autolayout
5890 new_disk
->raiddisk
= new_disk
->index
;
5891 new_disk
->disk
.status
|=
5893 new_disk
->disk
.status
&= ~SPARE_DISK
;
5896 dprintf("imsm: volume set mpb->num_raid_devs = %i\n",
5897 mpb
->num_raid_devs
);
5898 /* manage changes in volume
5900 for (dev_id
= 0; dev_id
< mpb
->num_raid_devs
; dev_id
++) {
5901 void **sp
= *space_list
;
5902 struct imsm_dev
*newdev
;
5903 struct imsm_map
*newmap
, *oldmap
;
5905 for (id
= super
->devlist
; id
; id
= id
->next
) {
5906 if (id
->index
== dev_id
)
5915 /* Copy the dev, but not (all of) the map */
5916 memcpy(newdev
, id
->dev
, sizeof(*newdev
));
5917 oldmap
= get_imsm_map(id
->dev
, 0);
5918 newmap
= get_imsm_map(newdev
, 0);
5919 /* Copy the current map */
5920 memcpy(newmap
, oldmap
, sizeof_imsm_map(oldmap
));
5921 /* update one device only
5923 if (devices_to_reshape
) {
5924 dprintf("imsm: modifying subdev: %i\n",
5926 devices_to_reshape
--;
5927 newdev
->vol
.migr_state
= 1;
5928 newdev
->vol
.curr_migr_unit
= 0;
5929 newdev
->vol
.migr_type
= MIGR_GEN_MIGR
;
5930 newmap
->num_members
= u
->new_raid_disks
;
5931 for (i
= 0; i
< delta_disks
; i
++) {
5932 set_imsm_ord_tbl_ent(newmap
,
5933 u
->old_raid_disks
+ i
,
5934 u
->old_raid_disks
+ i
);
5936 /* New map is correct, now need to save old map
5938 newmap
= get_imsm_map(newdev
, 1);
5939 memcpy(newmap
, oldmap
, sizeof_imsm_map(oldmap
));
5941 imsm_set_array_size(newdev
);
5944 sp
= (void **)id
->dev
;
5950 *space_list
= tofree
;
5953 update_reshape_exit
:
5958 static int apply_takeover_update(struct imsm_update_takeover
*u
,
5959 struct intel_super
*super
,
5962 struct imsm_dev
*dev
= NULL
;
5963 struct intel_dev
*dv
;
5964 struct imsm_dev
*dev_new
;
5965 struct imsm_map
*map
;
5969 for (dv
= super
->devlist
; dv
; dv
= dv
->next
)
5970 if (dv
->index
== (unsigned int)u
->subarray
) {
5978 map
= get_imsm_map(dev
, 0);
5980 if (u
->direction
== R10_TO_R0
) {
5981 /* Number of failed disks must be half of initial disk number */
5982 if (imsm_count_failed(super
, dev
) != (map
->num_members
/ 2))
5985 /* iterate through devices to mark removed disks as spare */
5986 for (dm
= super
->disks
; dm
; dm
= dm
->next
) {
5987 if (dm
->disk
.status
& FAILED_DISK
) {
5988 int idx
= dm
->index
;
5989 /* update indexes on the disk list */
5990 /* FIXME this loop-with-the-loop looks wrong, I'm not convinced
5991 the index values will end up being correct.... NB */
5992 for (du
= super
->disks
; du
; du
= du
->next
)
5993 if (du
->index
> idx
)
5995 /* mark as spare disk */
5996 dm
->disk
.status
= SPARE_DISK
;
6001 map
->num_members
= map
->num_members
/ 2;
6002 map
->map_state
= IMSM_T_STATE_NORMAL
;
6003 map
->num_domains
= 1;
6004 map
->raid_level
= 0;
6005 map
->failed_disk_num
= -1;
6008 if (u
->direction
== R0_TO_R10
) {
6010 /* update slots in current disk list */
6011 for (dm
= super
->disks
; dm
; dm
= dm
->next
) {
6015 /* create new *missing* disks */
6016 for (i
= 0; i
< map
->num_members
; i
++) {
6017 space
= *space_list
;
6020 *space_list
= *space
;
6022 memcpy(du
, super
->disks
, sizeof(*du
));
6023 du
->disk
.status
= FAILED_DISK
;
6024 du
->disk
.scsi_id
= 0;
6028 du
->index
= (i
* 2) + 1;
6029 sprintf((char *)du
->disk
.serial
,
6030 " MISSING_%d", du
->index
);
6031 sprintf((char *)du
->serial
,
6032 "MISSING_%d", du
->index
);
6033 du
->next
= super
->missing
;
6034 super
->missing
= du
;
6036 /* create new dev and map */
6037 space
= *space_list
;
6040 *space_list
= *space
;
6041 dev_new
= (void *)space
;
6042 memcpy(dev_new
, dev
, sizeof(*dev
));
6043 /* update new map */
6044 map
= get_imsm_map(dev_new
, 0);
6045 map
->failed_disk_num
= map
->num_members
;
6046 map
->num_members
= map
->num_members
* 2;
6047 map
->map_state
= IMSM_T_STATE_NORMAL
;
6048 map
->num_domains
= 2;
6049 map
->raid_level
= 1;
6050 /* replace dev<->dev_new */
6053 /* update disk order table */
6054 for (du
= super
->disks
; du
; du
= du
->next
)
6056 set_imsm_ord_tbl_ent(map
, du
->index
, du
->index
);
6057 for (du
= super
->missing
; du
; du
= du
->next
)
6059 set_imsm_ord_tbl_ent(map
, du
->index
,
6060 du
->index
| IMSM_ORD_REBUILD
);
6065 static void imsm_process_update(struct supertype
*st
,
6066 struct metadata_update
*update
)
6069 * crack open the metadata_update envelope to find the update record
6070 * update can be one of:
6071 * update_reshape_container_disks - all the arrays in the container
6072 * are being reshaped to have more devices. We need to mark
6073 * the arrays for general migration and convert selected spares
6074 * into active devices.
6075 * update_activate_spare - a spare device has replaced a failed
6076 * device in an array, update the disk_ord_tbl. If this disk is
6077 * present in all member arrays then also clear the SPARE_DISK
6079 * update_create_array
6081 * update_rename_array
6082 * update_add_remove_disk
6084 struct intel_super
*super
= st
->sb
;
6085 struct imsm_super
*mpb
;
6086 enum imsm_update_type type
= *(enum imsm_update_type
*) update
->buf
;
6088 /* update requires a larger buf but the allocation failed */
6089 if (super
->next_len
&& !super
->next_buf
) {
6090 super
->next_len
= 0;
6094 if (super
->next_buf
) {
6095 memcpy(super
->next_buf
, super
->buf
, super
->len
);
6097 super
->len
= super
->next_len
;
6098 super
->buf
= super
->next_buf
;
6100 super
->next_len
= 0;
6101 super
->next_buf
= NULL
;
6104 mpb
= super
->anchor
;
6107 case update_takeover
: {
6108 struct imsm_update_takeover
*u
= (void *)update
->buf
;
6109 if (apply_takeover_update(u
, super
, &update
->space_list
))
6110 super
->updates_pending
++;
6114 case update_reshape_container_disks
: {
6115 struct imsm_update_reshape
*u
= (void *)update
->buf
;
6116 if (apply_reshape_container_disks_update(
6117 u
, super
, &update
->space_list
))
6118 super
->updates_pending
++;
6121 case update_activate_spare
: {
6122 struct imsm_update_activate_spare
*u
= (void *) update
->buf
;
6123 struct imsm_dev
*dev
= get_imsm_dev(super
, u
->array
);
6124 struct imsm_map
*map
= get_imsm_map(dev
, 0);
6125 struct imsm_map
*migr_map
;
6126 struct active_array
*a
;
6127 struct imsm_disk
*disk
;
6132 int victim
= get_imsm_disk_idx(dev
, u
->slot
, -1);
6135 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
6140 fprintf(stderr
, "error: imsm_activate_spare passed "
6141 "an unknown disk (index: %d)\n",
6146 super
->updates_pending
++;
6148 /* count failures (excluding rebuilds and the victim)
6149 * to determine map[0] state
6152 for (i
= 0; i
< map
->num_members
; i
++) {
6155 disk
= get_imsm_disk(super
,
6156 get_imsm_disk_idx(dev
, i
, -1));
6157 if (!disk
|| is_failed(disk
))
6161 /* adding a pristine spare, assign a new index */
6162 if (dl
->index
< 0) {
6163 dl
->index
= super
->anchor
->num_disks
;
6164 super
->anchor
->num_disks
++;
6167 disk
->status
|= CONFIGURED_DISK
;
6168 disk
->status
&= ~SPARE_DISK
;
6171 to_state
= imsm_check_degraded(super
, dev
, failed
);
6172 map
->map_state
= IMSM_T_STATE_DEGRADED
;
6173 migrate(dev
, to_state
, MIGR_REBUILD
);
6174 migr_map
= get_imsm_map(dev
, 1);
6175 set_imsm_ord_tbl_ent(map
, u
->slot
, dl
->index
);
6176 set_imsm_ord_tbl_ent(migr_map
, u
->slot
, dl
->index
| IMSM_ORD_REBUILD
);
6178 /* update the family_num to mark a new container
6179 * generation, being careful to record the existing
6180 * family_num in orig_family_num to clean up after
6181 * earlier mdadm versions that neglected to set it.
6183 if (mpb
->orig_family_num
== 0)
6184 mpb
->orig_family_num
= mpb
->family_num
;
6185 mpb
->family_num
+= super
->random
;
6187 /* count arrays using the victim in the metadata */
6189 for (a
= st
->arrays
; a
; a
= a
->next
) {
6190 dev
= get_imsm_dev(super
, a
->info
.container_member
);
6191 map
= get_imsm_map(dev
, 0);
6193 if (get_imsm_disk_slot(map
, victim
) >= 0)
6197 /* delete the victim if it is no longer being
6203 /* We know that 'manager' isn't touching anything,
6204 * so it is safe to delete
6206 for (dlp
= &super
->disks
; *dlp
; dlp
= &(*dlp
)->next
)
6207 if ((*dlp
)->index
== victim
)
6210 /* victim may be on the missing list */
6212 for (dlp
= &super
->missing
; *dlp
; dlp
= &(*dlp
)->next
)
6213 if ((*dlp
)->index
== victim
)
6215 imsm_delete(super
, dlp
, victim
);
6219 case update_create_array
: {
6220 /* someone wants to create a new array, we need to be aware of
6221 * a few races/collisions:
6222 * 1/ 'Create' called by two separate instances of mdadm
6223 * 2/ 'Create' versus 'activate_spare': mdadm has chosen
6224 * devices that have since been assimilated via
6226 * In the event this update can not be carried out mdadm will
6227 * (FIX ME) notice that its update did not take hold.
6229 struct imsm_update_create_array
*u
= (void *) update
->buf
;
6230 struct intel_dev
*dv
;
6231 struct imsm_dev
*dev
;
6232 struct imsm_map
*map
, *new_map
;
6233 unsigned long long start
, end
;
6234 unsigned long long new_start
, new_end
;
6236 struct disk_info
*inf
;
6239 /* handle racing creates: first come first serve */
6240 if (u
->dev_idx
< mpb
->num_raid_devs
) {
6241 dprintf("%s: subarray %d already defined\n",
6242 __func__
, u
->dev_idx
);
6246 /* check update is next in sequence */
6247 if (u
->dev_idx
!= mpb
->num_raid_devs
) {
6248 dprintf("%s: can not create array %d expected index %d\n",
6249 __func__
, u
->dev_idx
, mpb
->num_raid_devs
);
6253 new_map
= get_imsm_map(&u
->dev
, 0);
6254 new_start
= __le32_to_cpu(new_map
->pba_of_lba0
);
6255 new_end
= new_start
+ __le32_to_cpu(new_map
->blocks_per_member
);
6256 inf
= get_disk_info(u
);
6258 /* handle activate_spare versus create race:
6259 * check to make sure that overlapping arrays do not include
6262 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
6263 dev
= get_imsm_dev(super
, i
);
6264 map
= get_imsm_map(dev
, 0);
6265 start
= __le32_to_cpu(map
->pba_of_lba0
);
6266 end
= start
+ __le32_to_cpu(map
->blocks_per_member
);
6267 if ((new_start
>= start
&& new_start
<= end
) ||
6268 (start
>= new_start
&& start
<= new_end
))
6273 if (disks_overlap(super
, i
, u
)) {
6274 dprintf("%s: arrays overlap\n", __func__
);
6279 /* check that prepare update was successful */
6280 if (!update
->space
) {
6281 dprintf("%s: prepare update failed\n", __func__
);
6285 /* check that all disks are still active before committing
6286 * changes. FIXME: could we instead handle this by creating a
6287 * degraded array? That's probably not what the user expects,
6288 * so better to drop this update on the floor.
6290 for (i
= 0; i
< new_map
->num_members
; i
++) {
6291 dl
= serial_to_dl(inf
[i
].serial
, super
);
6293 dprintf("%s: disk disappeared\n", __func__
);
6298 super
->updates_pending
++;
6300 /* convert spares to members and fixup ord_tbl */
6301 for (i
= 0; i
< new_map
->num_members
; i
++) {
6302 dl
= serial_to_dl(inf
[i
].serial
, super
);
6303 if (dl
->index
== -1) {
6304 dl
->index
= mpb
->num_disks
;
6306 dl
->disk
.status
|= CONFIGURED_DISK
;
6307 dl
->disk
.status
&= ~SPARE_DISK
;
6309 set_imsm_ord_tbl_ent(new_map
, i
, dl
->index
);
6314 update
->space
= NULL
;
6315 imsm_copy_dev(dev
, &u
->dev
);
6316 dv
->index
= u
->dev_idx
;
6317 dv
->next
= super
->devlist
;
6318 super
->devlist
= dv
;
6319 mpb
->num_raid_devs
++;
6321 imsm_update_version_info(super
);
6324 /* mdmon knows how to release update->space, but not
6325 * ((struct intel_dev *) update->space)->dev
6327 if (update
->space
) {
6333 case update_kill_array
: {
6334 struct imsm_update_kill_array
*u
= (void *) update
->buf
;
6335 int victim
= u
->dev_idx
;
6336 struct active_array
*a
;
6337 struct intel_dev
**dp
;
6338 struct imsm_dev
*dev
;
6340 /* sanity check that we are not affecting the uuid of
6341 * active arrays, or deleting an active array
6343 * FIXME when immutable ids are available, but note that
6344 * we'll also need to fixup the invalidated/active
6345 * subarray indexes in mdstat
6347 for (a
= st
->arrays
; a
; a
= a
->next
)
6348 if (a
->info
.container_member
>= victim
)
6350 /* by definition if mdmon is running at least one array
6351 * is active in the container, so checking
6352 * mpb->num_raid_devs is just extra paranoia
6354 dev
= get_imsm_dev(super
, victim
);
6355 if (a
|| !dev
|| mpb
->num_raid_devs
== 1) {
6356 dprintf("failed to delete subarray-%d\n", victim
);
6360 for (dp
= &super
->devlist
; *dp
;)
6361 if ((*dp
)->index
== (unsigned)super
->current_vol
) {
6364 if ((*dp
)->index
> (unsigned)victim
)
6368 mpb
->num_raid_devs
--;
6369 super
->updates_pending
++;
6372 case update_rename_array
: {
6373 struct imsm_update_rename_array
*u
= (void *) update
->buf
;
6374 char name
[MAX_RAID_SERIAL_LEN
+1];
6375 int target
= u
->dev_idx
;
6376 struct active_array
*a
;
6377 struct imsm_dev
*dev
;
6379 /* sanity check that we are not affecting the uuid of
6382 snprintf(name
, MAX_RAID_SERIAL_LEN
, "%s", (char *) u
->name
);
6383 name
[MAX_RAID_SERIAL_LEN
] = '\0';
6384 for (a
= st
->arrays
; a
; a
= a
->next
)
6385 if (a
->info
.container_member
== target
)
6387 dev
= get_imsm_dev(super
, u
->dev_idx
);
6388 if (a
|| !dev
|| !check_name(super
, name
, 1)) {
6389 dprintf("failed to rename subarray-%d\n", target
);
6393 snprintf((char *) dev
->volume
, MAX_RAID_SERIAL_LEN
, "%s", name
);
6394 super
->updates_pending
++;
6397 case update_add_remove_disk
: {
6398 /* we may be able to repair some arrays if disks are
6399 * being added, check teh status of add_remove_disk
6400 * if discs has been added.
6402 if (add_remove_disk_update(super
)) {
6403 struct active_array
*a
;
6405 super
->updates_pending
++;
6406 for (a
= st
->arrays
; a
; a
= a
->next
)
6407 a
->check_degraded
= 1;
6412 fprintf(stderr
, "error: unsuported process update type:"
6413 "(type: %d)\n", type
);
6417 static void imsm_prepare_update(struct supertype
*st
,
6418 struct metadata_update
*update
)
6421 * Allocate space to hold new disk entries, raid-device entries or a new
6422 * mpb if necessary. The manager synchronously waits for updates to
6423 * complete in the monitor, so new mpb buffers allocated here can be
6424 * integrated by the monitor thread without worrying about live pointers
6425 * in the manager thread.
6427 enum imsm_update_type type
= *(enum imsm_update_type
*) update
->buf
;
6428 struct intel_super
*super
= st
->sb
;
6429 struct imsm_super
*mpb
= super
->anchor
;
6434 case update_takeover
: {
6435 struct imsm_update_takeover
*u
= (void *)update
->buf
;
6436 if (u
->direction
== R0_TO_R10
) {
6437 void **tail
= (void **)&update
->space_list
;
6438 struct imsm_dev
*dev
= get_imsm_dev(super
, u
->subarray
);
6439 struct imsm_map
*map
= get_imsm_map(dev
, 0);
6440 int num_members
= map
->num_members
;
6444 /* allocate memory for added disks */
6445 for (i
= 0; i
< num_members
; i
++) {
6446 size
= sizeof(struct dl
);
6447 space
= malloc(size
);
6456 /* allocate memory for new device */
6457 size
= sizeof_imsm_dev(super
->devlist
->dev
, 0) +
6458 (num_members
* sizeof(__u32
));
6459 space
= malloc(size
);
6468 len
= disks_to_mpb_size(num_members
* 2);
6470 /* if allocation didn't success, free buffer */
6471 while (update
->space_list
) {
6472 void **sp
= update
->space_list
;
6473 update
->space_list
= *sp
;
6481 case update_reshape_container_disks
: {
6482 /* Every raid device in the container is about to
6483 * gain some more devices, and we will enter a
6485 * So each 'imsm_map' will be bigger, and the imsm_vol
6486 * will now hold 2 of them.
6487 * Thus we need new 'struct imsm_dev' allocations sized
6488 * as sizeof_imsm_dev but with more devices in both maps.
6490 struct imsm_update_reshape
*u
= (void *)update
->buf
;
6491 struct intel_dev
*dl
;
6492 void **space_tail
= (void**)&update
->space_list
;
6494 dprintf("imsm: imsm_prepare_update() for update_reshape\n");
6496 for (dl
= super
->devlist
; dl
; dl
= dl
->next
) {
6497 int size
= sizeof_imsm_dev(dl
->dev
, 1);
6499 if (u
->new_raid_disks
> u
->old_raid_disks
)
6500 size
+= sizeof(__u32
)*2*
6501 (u
->new_raid_disks
- u
->old_raid_disks
);
6510 len
= disks_to_mpb_size(u
->new_raid_disks
);
6511 dprintf("New anchor length is %llu\n", (unsigned long long)len
);
6514 case update_create_array
: {
6515 struct imsm_update_create_array
*u
= (void *) update
->buf
;
6516 struct intel_dev
*dv
;
6517 struct imsm_dev
*dev
= &u
->dev
;
6518 struct imsm_map
*map
= get_imsm_map(dev
, 0);
6520 struct disk_info
*inf
;
6524 inf
= get_disk_info(u
);
6525 len
= sizeof_imsm_dev(dev
, 1);
6526 /* allocate a new super->devlist entry */
6527 dv
= malloc(sizeof(*dv
));
6529 dv
->dev
= malloc(len
);
6534 update
->space
= NULL
;
6538 /* count how many spares will be converted to members */
6539 for (i
= 0; i
< map
->num_members
; i
++) {
6540 dl
= serial_to_dl(inf
[i
].serial
, super
);
6542 /* hmm maybe it failed?, nothing we can do about
6547 if (count_memberships(dl
, super
) == 0)
6550 len
+= activate
* sizeof(struct imsm_disk
);
6557 /* check if we need a larger metadata buffer */
6558 if (super
->next_buf
)
6559 buf_len
= super
->next_len
;
6561 buf_len
= super
->len
;
6563 if (__le32_to_cpu(mpb
->mpb_size
) + len
> buf_len
) {
6564 /* ok we need a larger buf than what is currently allocated
6565 * if this allocation fails process_update will notice that
6566 * ->next_len is set and ->next_buf is NULL
6568 buf_len
= ROUND_UP(__le32_to_cpu(mpb
->mpb_size
) + len
, 512);
6569 if (super
->next_buf
)
6570 free(super
->next_buf
);
6572 super
->next_len
= buf_len
;
6573 if (posix_memalign(&super
->next_buf
, 512, buf_len
) == 0)
6574 memset(super
->next_buf
, 0, buf_len
);
6576 super
->next_buf
= NULL
;
6580 /* must be called while manager is quiesced */
6581 static void imsm_delete(struct intel_super
*super
, struct dl
**dlp
, unsigned index
)
6583 struct imsm_super
*mpb
= super
->anchor
;
6585 struct imsm_dev
*dev
;
6586 struct imsm_map
*map
;
6587 int i
, j
, num_members
;
6590 dprintf("%s: deleting device[%d] from imsm_super\n",
6593 /* shift all indexes down one */
6594 for (iter
= super
->disks
; iter
; iter
= iter
->next
)
6595 if (iter
->index
> (int)index
)
6597 for (iter
= super
->missing
; iter
; iter
= iter
->next
)
6598 if (iter
->index
> (int)index
)
6601 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
6602 dev
= get_imsm_dev(super
, i
);
6603 map
= get_imsm_map(dev
, 0);
6604 num_members
= map
->num_members
;
6605 for (j
= 0; j
< num_members
; j
++) {
6606 /* update ord entries being careful not to propagate
6607 * ord-flags to the first map
6609 ord
= get_imsm_ord_tbl_ent(dev
, j
, -1);
6611 if (ord_to_idx(ord
) <= index
)
6614 map
= get_imsm_map(dev
, 0);
6615 set_imsm_ord_tbl_ent(map
, j
, ord_to_idx(ord
- 1));
6616 map
= get_imsm_map(dev
, 1);
6618 set_imsm_ord_tbl_ent(map
, j
, ord
- 1);
6623 super
->updates_pending
++;
6625 struct dl
*dl
= *dlp
;
6627 *dlp
= (*dlp
)->next
;
6628 __free_imsm_disk(dl
);
6632 static char disk_by_path
[] = "/dev/disk/by-path/";
6634 static const char *imsm_get_disk_controller_domain(const char *path
)
6636 char disk_path
[PATH_MAX
];
6640 strncpy(disk_path
, disk_by_path
, PATH_MAX
- 1);
6641 strncat(disk_path
, path
, PATH_MAX
- strlen(disk_path
) - 1);
6642 if (stat(disk_path
, &st
) == 0) {
6643 struct sys_dev
* hba
;
6646 path
= devt_to_devpath(st
.st_rdev
);
6649 hba
= find_disk_attached_hba(-1, path
);
6650 if (hba
&& hba
->type
== SYS_DEV_SAS
)
6652 else if (hba
&& hba
->type
== SYS_DEV_SATA
)
6656 dprintf("path: %s hba: %s attached: %s\n",
6657 path
, (hba
) ? hba
->path
: "NULL", drv
);
6665 static int imsm_find_array_minor_by_subdev(int subdev
, int container
, int *minor
)
6667 char subdev_name
[20];
6668 struct mdstat_ent
*mdstat
;
6670 sprintf(subdev_name
, "%d", subdev
);
6671 mdstat
= mdstat_by_subdev(subdev_name
, container
);
6675 *minor
= mdstat
->devnum
;
6676 free_mdstat(mdstat
);
6680 static int imsm_reshape_is_allowed_on_container(struct supertype
*st
,
6681 struct geo_params
*geo
,
6682 int *old_raid_disks
)
6684 /* currently we only support increasing the number of devices
6685 * for a container. This increases the number of device for each
6686 * member array. They must all be RAID0 or RAID5.
6689 struct mdinfo
*info
, *member
;
6690 int devices_that_can_grow
= 0;
6692 dprintf("imsm: imsm_reshape_is_allowed_on_container(ENTER): "
6693 "st->devnum = (%i)\n",
6696 if (geo
->size
!= -1 ||
6697 geo
->level
!= UnSet
||
6698 geo
->layout
!= UnSet
||
6699 geo
->chunksize
!= 0 ||
6700 geo
->raid_disks
== UnSet
) {
6701 dprintf("imsm: Container operation is allowed for "
6702 "raid disks number change only.\n");
6706 info
= container_content_imsm(st
, NULL
);
6707 for (member
= info
; member
; member
= member
->next
) {
6711 dprintf("imsm: checking device_num: %i\n",
6712 member
->container_member
);
6714 if (geo
->raid_disks
<= member
->array
.raid_disks
) {
6715 /* we work on container for Online Capacity Expansion
6716 * only so raid_disks has to grow
6718 dprintf("imsm: for container operation raid disks "
6719 "increase is required\n");
6723 if ((info
->array
.level
!= 0) &&
6724 (info
->array
.level
!= 5)) {
6725 /* we cannot use this container with other raid level
6727 dprintf("imsm: for container operation wrong"
6728 " raid level (%i) detected\n",
6732 /* check for platform support
6733 * for this raid level configuration
6735 struct intel_super
*super
= st
->sb
;
6736 if (!is_raid_level_supported(super
->orom
,
6737 member
->array
.level
,
6739 dprintf("platform does not support raid%d with"
6743 geo
->raid_disks
> 1 ? "s" : "");
6748 if (*old_raid_disks
&&
6749 info
->array
.raid_disks
!= *old_raid_disks
)
6751 *old_raid_disks
= info
->array
.raid_disks
;
6753 /* All raid5 and raid0 volumes in container
6754 * have to be ready for Online Capacity Expansion
6755 * so they need to be assembled. We have already
6756 * checked that no recovery etc is happening.
6758 result
= imsm_find_array_minor_by_subdev(member
->container_member
,
6762 dprintf("imsm: cannot find array\n");
6765 devices_that_can_grow
++;
6768 if (!member
&& devices_that_can_grow
)
6772 dprintf("\tContainer operation allowed\n");
6774 dprintf("\tError: %i\n", ret_val
);
6779 /* Function: get_spares_for_grow
6780 * Description: Allocates memory and creates list of spare devices
6781 * avaliable in container. Checks if spare drive size is acceptable.
6782 * Parameters: Pointer to the supertype structure
6783 * Returns: Pointer to the list of spare devices (mdinfo structure) on success,
6786 static struct mdinfo
*get_spares_for_grow(struct supertype
*st
)
6788 unsigned long long min_size
= min_acceptable_spare_size_imsm(st
);
6789 return container_choose_spares(st
, min_size
, NULL
, NULL
, NULL
, 0);
6792 /******************************************************************************
6793 * function: imsm_create_metadata_update_for_reshape
6794 * Function creates update for whole IMSM container.
6796 ******************************************************************************/
6797 static int imsm_create_metadata_update_for_reshape(
6798 struct supertype
*st
,
6799 struct geo_params
*geo
,
6801 struct imsm_update_reshape
**updatep
)
6803 struct intel_super
*super
= st
->sb
;
6804 struct imsm_super
*mpb
= super
->anchor
;
6805 int update_memory_size
= 0;
6806 struct imsm_update_reshape
*u
= NULL
;
6807 struct mdinfo
*spares
= NULL
;
6809 int delta_disks
= 0;
6812 dprintf("imsm_update_metadata_for_reshape(enter) raid_disks = %i\n",
6815 delta_disks
= geo
->raid_disks
- old_raid_disks
;
6817 /* size of all update data without anchor */
6818 update_memory_size
= sizeof(struct imsm_update_reshape
);
6820 /* now add space for spare disks that we need to add. */
6821 update_memory_size
+= sizeof(u
->new_disks
[0]) * (delta_disks
- 1);
6823 u
= calloc(1, update_memory_size
);
6826 "cannot get memory for imsm_update_reshape update\n");
6829 u
->type
= update_reshape_container_disks
;
6830 u
->old_raid_disks
= old_raid_disks
;
6831 u
->new_raid_disks
= geo
->raid_disks
;
6833 /* now get spare disks list
6835 spares
= get_spares_for_grow(st
);
6838 || delta_disks
> spares
->array
.spare_disks
) {
6839 dprintf("imsm: ERROR: Cannot get spare devices.\n");
6843 /* we have got spares
6844 * update disk list in imsm_disk list table in anchor
6846 dprintf("imsm: %i spares are available.\n\n",
6847 spares
->array
.spare_disks
);
6850 for (i
= 0; i
< delta_disks
; i
++) {
6855 u
->new_disks
[i
] = makedev(dev
->disk
.major
,
6857 dl
= get_disk_super(super
, dev
->disk
.major
, dev
->disk
.minor
);
6858 dl
->index
= mpb
->num_disks
;
6868 dprintf("imsm: reshape update preparation :");
6869 if (i
== delta_disks
) {
6872 return update_memory_size
;
6875 dprintf(" Error\n");
6880 static void imsm_update_metadata_locally(struct supertype
*st
,
6883 struct metadata_update mu
;
6888 mu
.space_list
= NULL
;
6890 imsm_prepare_update(st
, &mu
);
6891 imsm_process_update(st
, &mu
);
6893 while (mu
.space_list
) {
6894 void **space
= mu
.space_list
;
6895 mu
.space_list
= *space
;
6900 /***************************************************************************
6901 * Function: imsm_analyze_change
6902 * Description: Function analyze change for single volume
6903 * and validate if transition is supported
6904 * Parameters: Geometry parameters, supertype structure
6905 * Returns: Operation type code on success, -1 if fail
6906 ****************************************************************************/
6907 enum imsm_reshape_type
imsm_analyze_change(struct supertype
*st
,
6908 struct geo_params
*geo
)
6914 getinfo_super_imsm_volume(st
, &info
, NULL
);
6916 if ((geo
->level
!= info
.array
.level
) &&
6917 (geo
->level
>= 0) &&
6918 (geo
->level
!= UnSet
)) {
6919 switch (info
.array
.level
) {
6921 if (geo
->level
== 5) {
6922 change
= CH_LEVEL_MIGRATION
;
6925 if (geo
->level
== 10) {
6926 change
= CH_TAKEOVER
;
6931 if (geo
->level
== 0) {
6932 change
= CH_TAKEOVER
;
6937 if (geo
->level
!= 0)
6938 change
= CH_LEVEL_MIGRATION
;
6941 if (geo
->level
== 0) {
6942 change
= CH_TAKEOVER
;
6949 Name
" Error. Level Migration from %d to %d "
6951 info
.array
.level
, geo
->level
);
6952 goto analyse_change_exit
;
6955 geo
->level
= info
.array
.level
;
6957 if ((geo
->layout
!= info
.array
.layout
)
6958 && ((geo
->layout
!= UnSet
) && (geo
->layout
!= -1))) {
6959 change
= CH_LEVEL_MIGRATION
;
6960 if ((info
.array
.layout
== 0)
6961 && (info
.array
.level
== 5)
6962 && (geo
->layout
== 5)) {
6963 /* reshape 5 -> 4 */
6964 } else if ((info
.array
.layout
== 5)
6965 && (info
.array
.level
== 5)
6966 && (geo
->layout
== 0)) {
6967 /* reshape 4 -> 5 */
6972 Name
" Error. Layout Migration from %d to %d "
6974 info
.array
.layout
, geo
->layout
);
6976 goto analyse_change_exit
;
6979 geo
->layout
= info
.array
.layout
;
6981 if ((geo
->chunksize
> 0) && (geo
->chunksize
!= UnSet
)
6982 && (geo
->chunksize
!= info
.array
.chunk_size
))
6983 change
= CH_CHUNK_MIGR
;
6985 geo
->chunksize
= info
.array
.chunk_size
;
6987 if (!validate_geometry_imsm(st
,
6991 (geo
->chunksize
/ 1024),
6997 struct intel_super
*super
= st
->sb
;
6998 struct imsm_super
*mpb
= super
->anchor
;
7000 if (mpb
->num_raid_devs
> 1) {
7002 Name
" Error. Cannot perform operation on %s"
7003 "- for this operation it MUST be single "
7004 "array in container\n",
7010 analyse_change_exit
:
7015 int imsm_takeover(struct supertype
*st
, struct geo_params
*geo
)
7017 struct intel_super
*super
= st
->sb
;
7018 struct imsm_update_takeover
*u
;
7020 u
= malloc(sizeof(struct imsm_update_takeover
));
7024 u
->type
= update_takeover
;
7025 u
->subarray
= super
->current_vol
;
7027 /* 10->0 transition */
7028 if (geo
->level
== 0)
7029 u
->direction
= R10_TO_R0
;
7031 /* 0->10 transition */
7032 if (geo
->level
== 10)
7033 u
->direction
= R0_TO_R10
;
7035 /* update metadata locally */
7036 imsm_update_metadata_locally(st
, u
,
7037 sizeof(struct imsm_update_takeover
));
7038 /* and possibly remotely */
7039 if (st
->update_tail
)
7040 append_metadata_update(st
, u
,
7041 sizeof(struct imsm_update_takeover
));
7048 static int imsm_reshape_super(struct supertype
*st
, long long size
, int level
,
7049 int layout
, int chunksize
, int raid_disks
,
7050 char *backup
, char *dev
, int verbose
)
7053 struct geo_params geo
;
7055 dprintf("imsm: reshape_super called.\n");
7057 memset(&geo
, 0, sizeof(struct geo_params
));
7060 geo
.dev_id
= st
->devnum
;
7063 geo
.layout
= layout
;
7064 geo
.chunksize
= chunksize
;
7065 geo
.raid_disks
= raid_disks
;
7067 dprintf("\tfor level : %i\n", geo
.level
);
7068 dprintf("\tfor raid_disks : %i\n", geo
.raid_disks
);
7070 if (experimental() == 0)
7073 if (st
->container_dev
== st
->devnum
) {
7074 /* On container level we can only increase number of devices. */
7075 dprintf("imsm: info: Container operation\n");
7076 int old_raid_disks
= 0;
7077 if (imsm_reshape_is_allowed_on_container(
7078 st
, &geo
, &old_raid_disks
)) {
7079 struct imsm_update_reshape
*u
= NULL
;
7082 len
= imsm_create_metadata_update_for_reshape(
7083 st
, &geo
, old_raid_disks
, &u
);
7086 dprintf("imsm: Cannot prepare update\n");
7087 goto exit_imsm_reshape_super
;
7091 /* update metadata locally */
7092 imsm_update_metadata_locally(st
, u
, len
);
7093 /* and possibly remotely */
7094 if (st
->update_tail
)
7095 append_metadata_update(st
, u
, len
);
7100 fprintf(stderr
, Name
": (imsm) Operation "
7101 "is not allowed on this container\n");
7104 /* On volume level we support following operations
7105 * - takeover: raid10 -> raid0; raid0 -> raid10
7106 * - chunk size migration
7107 * - migration: raid5 -> raid0; raid0 -> raid5
7109 struct intel_super
*super
= st
->sb
;
7110 struct intel_dev
*dev
= super
->devlist
;
7112 dprintf("imsm: info: Volume operation\n");
7113 /* find requested device */
7115 imsm_find_array_minor_by_subdev(dev
->index
, st
->container_dev
, &devnum
);
7116 if (devnum
== geo
.dev_id
)
7121 fprintf(stderr
, Name
" Cannot find %s (%i) subarray\n",
7122 geo
.dev_name
, geo
.dev_id
);
7123 goto exit_imsm_reshape_super
;
7125 super
->current_vol
= dev
->index
;
7126 change
= imsm_analyze_change(st
, &geo
);
7129 ret_val
= imsm_takeover(st
, &geo
);
7134 case CH_LEVEL_MIGRATION
:
7142 exit_imsm_reshape_super
:
7143 dprintf("imsm: reshape_super Exit code = %i\n", ret_val
);
7147 static int imsm_manage_reshape(
7148 int afd
, struct mdinfo
*sra
, struct reshape
*reshape
,
7149 struct supertype
*st
, unsigned long stripes
,
7150 int *fds
, unsigned long long *offsets
,
7151 int dests
, int *destfd
, unsigned long long *destoffsets
)
7153 /* Just use child_monitor for now */
7154 return child_monitor(
7155 afd
, sra
, reshape
, st
, stripes
,
7156 fds
, offsets
, dests
, destfd
, destoffsets
);
7158 #endif /* MDASSEMBLE */
7160 struct superswitch super_imsm
= {
7162 .examine_super
= examine_super_imsm
,
7163 .brief_examine_super
= brief_examine_super_imsm
,
7164 .brief_examine_subarrays
= brief_examine_subarrays_imsm
,
7165 .export_examine_super
= export_examine_super_imsm
,
7166 .detail_super
= detail_super_imsm
,
7167 .brief_detail_super
= brief_detail_super_imsm
,
7168 .write_init_super
= write_init_super_imsm
,
7169 .validate_geometry
= validate_geometry_imsm
,
7170 .add_to_super
= add_to_super_imsm
,
7171 .remove_from_super
= remove_from_super_imsm
,
7172 .detail_platform
= detail_platform_imsm
,
7173 .kill_subarray
= kill_subarray_imsm
,
7174 .update_subarray
= update_subarray_imsm
,
7175 .load_container
= load_container_imsm
,
7176 .default_geometry
= default_geometry_imsm
,
7177 .get_disk_controller_domain
= imsm_get_disk_controller_domain
,
7178 .reshape_super
= imsm_reshape_super
,
7179 .manage_reshape
= imsm_manage_reshape
,
7181 .match_home
= match_home_imsm
,
7182 .uuid_from_super
= uuid_from_super_imsm
,
7183 .getinfo_super
= getinfo_super_imsm
,
7184 .getinfo_super_disks
= getinfo_super_disks_imsm
,
7185 .update_super
= update_super_imsm
,
7187 .avail_size
= avail_size_imsm
,
7188 .min_acceptable_spare_size
= min_acceptable_spare_size_imsm
,
7190 .compare_super
= compare_super_imsm
,
7192 .load_super
= load_super_imsm
,
7193 .init_super
= init_super_imsm
,
7194 .store_super
= store_super_imsm
,
7195 .free_super
= free_super_imsm
,
7196 .match_metadata_desc
= match_metadata_desc_imsm
,
7197 .container_content
= container_content_imsm
,
7204 .open_new
= imsm_open_new
,
7205 .set_array_state
= imsm_set_array_state
,
7206 .set_disk
= imsm_set_disk
,
7207 .sync_metadata
= imsm_sync_metadata
,
7208 .activate_spare
= imsm_activate_spare
,
7209 .process_update
= imsm_process_update
,
7210 .prepare_update
= imsm_prepare_update
,
7211 #endif /* MDASSEMBLE */