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
];
387 static struct intel_hba
* alloc_intel_hba(struct sys_dev
*device
)
389 struct intel_hba
*result
= malloc(sizeof(*result
));
391 result
->type
= device
->type
;
392 result
->path
= strdup(device
->path
);
394 if (result
->path
&& (result
->pci_id
= strrchr(result
->path
, '/')) != NULL
)
400 static struct intel_hba
* find_intel_hba(struct intel_hba
*hba
, struct sys_dev
*device
)
402 struct intel_hba
*result
=NULL
;
403 for (result
= hba
; result
; result
= result
->next
) {
404 if (result
->type
== device
->type
&& strcmp(result
->path
, device
->path
) == 0)
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
)
480 static struct supertype
*match_metadata_desc_imsm(char *arg
)
482 struct supertype
*st
;
484 if (strcmp(arg
, "imsm") != 0 &&
485 strcmp(arg
, "default") != 0
489 st
= malloc(sizeof(*st
));
492 memset(st
, 0, sizeof(*st
));
493 st
->container_dev
= NoMdDev
;
494 st
->ss
= &super_imsm
;
495 st
->max_devs
= IMSM_MAX_DEVICES
;
496 st
->minor_version
= 0;
502 static __u8
*get_imsm_version(struct imsm_super
*mpb
)
504 return &mpb
->sig
[MPB_SIG_LEN
];
508 /* retrieve a disk directly from the anchor when the anchor is known to be
509 * up-to-date, currently only at load time
511 static struct imsm_disk
*__get_imsm_disk(struct imsm_super
*mpb
, __u8 index
)
513 if (index
>= mpb
->num_disks
)
515 return &mpb
->disk
[index
];
518 /* retrieve the disk description based on a index of the disk
521 static struct dl
*get_imsm_dl_disk(struct intel_super
*super
, __u8 index
)
525 for (d
= super
->disks
; d
; d
= d
->next
)
526 if (d
->index
== index
)
531 /* retrieve a disk from the parsed metadata */
532 static struct imsm_disk
*get_imsm_disk(struct intel_super
*super
, __u8 index
)
536 dl
= get_imsm_dl_disk(super
, index
);
543 /* generate a checksum directly from the anchor when the anchor is known to be
544 * up-to-date, currently only at load or write_super after coalescing
546 static __u32
__gen_imsm_checksum(struct imsm_super
*mpb
)
548 __u32 end
= mpb
->mpb_size
/ sizeof(end
);
549 __u32
*p
= (__u32
*) mpb
;
553 sum
+= __le32_to_cpu(*p
);
557 return sum
- __le32_to_cpu(mpb
->check_sum
);
560 static size_t sizeof_imsm_map(struct imsm_map
*map
)
562 return sizeof(struct imsm_map
) + sizeof(__u32
) * (map
->num_members
- 1);
565 struct imsm_map
*get_imsm_map(struct imsm_dev
*dev
, int second_map
)
567 struct imsm_map
*map
= &dev
->vol
.map
[0];
569 if (second_map
&& !dev
->vol
.migr_state
)
571 else if (second_map
) {
574 return ptr
+ sizeof_imsm_map(map
);
580 /* return the size of the device.
581 * migr_state increases the returned size if map[0] were to be duplicated
583 static size_t sizeof_imsm_dev(struct imsm_dev
*dev
, int migr_state
)
585 size_t size
= sizeof(*dev
) - sizeof(struct imsm_map
) +
586 sizeof_imsm_map(get_imsm_map(dev
, 0));
588 /* migrating means an additional map */
589 if (dev
->vol
.migr_state
)
590 size
+= sizeof_imsm_map(get_imsm_map(dev
, 1));
592 size
+= sizeof_imsm_map(get_imsm_map(dev
, 0));
598 /* retrieve disk serial number list from a metadata update */
599 static struct disk_info
*get_disk_info(struct imsm_update_create_array
*update
)
602 struct disk_info
*inf
;
604 inf
= u
+ sizeof(*update
) - sizeof(struct imsm_dev
) +
605 sizeof_imsm_dev(&update
->dev
, 0);
611 static struct imsm_dev
*__get_imsm_dev(struct imsm_super
*mpb
, __u8 index
)
617 if (index
>= mpb
->num_raid_devs
)
620 /* devices start after all disks */
621 offset
= ((void *) &mpb
->disk
[mpb
->num_disks
]) - _mpb
;
623 for (i
= 0; i
<= index
; i
++)
625 return _mpb
+ offset
;
627 offset
+= sizeof_imsm_dev(_mpb
+ offset
, 0);
632 static struct imsm_dev
*get_imsm_dev(struct intel_super
*super
, __u8 index
)
634 struct intel_dev
*dv
;
636 if (index
>= super
->anchor
->num_raid_devs
)
638 for (dv
= super
->devlist
; dv
; dv
= dv
->next
)
639 if (dv
->index
== index
)
647 * == 1 get second map
648 * == -1 than get map according to the current migr_state
650 static __u32
get_imsm_ord_tbl_ent(struct imsm_dev
*dev
,
654 struct imsm_map
*map
;
656 if (second_map
== -1) {
657 if (dev
->vol
.migr_state
)
658 map
= get_imsm_map(dev
, 1);
660 map
= get_imsm_map(dev
, 0);
662 map
= get_imsm_map(dev
, second_map
);
665 /* top byte identifies disk under rebuild */
666 return __le32_to_cpu(map
->disk_ord_tbl
[slot
]);
669 #define ord_to_idx(ord) (((ord) << 8) >> 8)
670 static __u32
get_imsm_disk_idx(struct imsm_dev
*dev
, int slot
, int second_map
)
672 __u32 ord
= get_imsm_ord_tbl_ent(dev
, slot
, second_map
);
674 return ord_to_idx(ord
);
677 static void set_imsm_ord_tbl_ent(struct imsm_map
*map
, int slot
, __u32 ord
)
679 map
->disk_ord_tbl
[slot
] = __cpu_to_le32(ord
);
682 static int get_imsm_disk_slot(struct imsm_map
*map
, unsigned idx
)
687 for (slot
= 0; slot
< map
->num_members
; slot
++) {
688 ord
= __le32_to_cpu(map
->disk_ord_tbl
[slot
]);
689 if (ord_to_idx(ord
) == idx
)
696 static int get_imsm_raid_level(struct imsm_map
*map
)
698 if (map
->raid_level
== 1) {
699 if (map
->num_members
== 2)
705 return map
->raid_level
;
708 static int cmp_extent(const void *av
, const void *bv
)
710 const struct extent
*a
= av
;
711 const struct extent
*b
= bv
;
712 if (a
->start
< b
->start
)
714 if (a
->start
> b
->start
)
719 static int count_memberships(struct dl
*dl
, struct intel_super
*super
)
724 for (i
= 0; i
< super
->anchor
->num_raid_devs
; i
++) {
725 struct imsm_dev
*dev
= get_imsm_dev(super
, i
);
726 struct imsm_map
*map
= get_imsm_map(dev
, 0);
728 if (get_imsm_disk_slot(map
, dl
->index
) >= 0)
735 static struct extent
*get_extents(struct intel_super
*super
, struct dl
*dl
)
737 /* find a list of used extents on the given physical device */
738 struct extent
*rv
, *e
;
740 int memberships
= count_memberships(dl
, super
);
741 __u32 reservation
= MPB_SECTOR_CNT
+ IMSM_RESERVED_SECTORS
;
743 rv
= malloc(sizeof(struct extent
) * (memberships
+ 1));
748 for (i
= 0; i
< super
->anchor
->num_raid_devs
; i
++) {
749 struct imsm_dev
*dev
= get_imsm_dev(super
, i
);
750 struct imsm_map
*map
= get_imsm_map(dev
, 0);
752 if (get_imsm_disk_slot(map
, dl
->index
) >= 0) {
753 e
->start
= __le32_to_cpu(map
->pba_of_lba0
);
754 e
->size
= __le32_to_cpu(map
->blocks_per_member
);
758 qsort(rv
, memberships
, sizeof(*rv
), cmp_extent
);
760 /* determine the start of the metadata
761 * when no raid devices are defined use the default
762 * ...otherwise allow the metadata to truncate the value
763 * as is the case with older versions of imsm
766 struct extent
*last
= &rv
[memberships
- 1];
769 remainder
= __le32_to_cpu(dl
->disk
.total_blocks
) -
770 (last
->start
+ last
->size
);
771 /* round down to 1k block to satisfy precision of the kernel
775 /* make sure remainder is still sane */
776 if (remainder
< (unsigned)ROUND_UP(super
->len
, 512) >> 9)
777 remainder
= ROUND_UP(super
->len
, 512) >> 9;
778 if (reservation
> remainder
)
779 reservation
= remainder
;
781 e
->start
= __le32_to_cpu(dl
->disk
.total_blocks
) - reservation
;
786 /* try to determine how much space is reserved for metadata from
787 * the last get_extents() entry, otherwise fallback to the
790 static __u32
imsm_reserved_sectors(struct intel_super
*super
, struct dl
*dl
)
796 /* for spares just return a minimal reservation which will grow
797 * once the spare is picked up by an array
800 return MPB_SECTOR_CNT
;
802 e
= get_extents(super
, dl
);
804 return MPB_SECTOR_CNT
+ IMSM_RESERVED_SECTORS
;
806 /* scroll to last entry */
807 for (i
= 0; e
[i
].size
; i
++)
810 rv
= __le32_to_cpu(dl
->disk
.total_blocks
) - e
[i
].start
;
817 static int is_spare(struct imsm_disk
*disk
)
819 return (disk
->status
& SPARE_DISK
) == SPARE_DISK
;
822 static int is_configured(struct imsm_disk
*disk
)
824 return (disk
->status
& CONFIGURED_DISK
) == CONFIGURED_DISK
;
827 static int is_failed(struct imsm_disk
*disk
)
829 return (disk
->status
& FAILED_DISK
) == FAILED_DISK
;
832 /* Return minimum size of a spare that can be used in this array*/
833 static unsigned long long min_acceptable_spare_size_imsm(struct supertype
*st
)
835 struct intel_super
*super
= st
->sb
;
839 unsigned long long rv
= 0;
843 /* find first active disk in array */
845 while (dl
&& (is_failed(&dl
->disk
) || dl
->index
== -1))
849 /* find last lba used by subarrays */
850 e
= get_extents(super
, dl
);
853 for (i
= 0; e
[i
].size
; i
++)
856 rv
= e
[i
-1].start
+ e
[i
-1].size
;
858 /* add the amount of space needed for metadata */
859 rv
= rv
+ MPB_SECTOR_CNT
+ IMSM_RESERVED_SECTORS
;
864 static __u64
blocks_per_migr_unit(struct imsm_dev
*dev
);
866 static void print_imsm_dev(struct imsm_dev
*dev
, char *uuid
, int disk_idx
)
870 struct imsm_map
*map
= get_imsm_map(dev
, 0);
871 struct imsm_map
*map2
= get_imsm_map(dev
, 1);
875 printf("[%.16s]:\n", dev
->volume
);
876 printf(" UUID : %s\n", uuid
);
877 printf(" RAID Level : %d", get_imsm_raid_level(map
));
879 printf(" <-- %d", get_imsm_raid_level(map2
));
881 printf(" Members : %d", map
->num_members
);
883 printf(" <-- %d", map2
->num_members
);
885 printf(" Slots : [");
886 for (i
= 0; i
< map
->num_members
; i
++) {
887 ord
= get_imsm_ord_tbl_ent(dev
, i
, 0);
888 printf("%s", ord
& IMSM_ORD_REBUILD
? "_" : "U");
893 for (i
= 0; i
< map2
->num_members
; i
++) {
894 ord
= get_imsm_ord_tbl_ent(dev
, i
, 1);
895 printf("%s", ord
& IMSM_ORD_REBUILD
? "_" : "U");
900 slot
= get_imsm_disk_slot(map
, disk_idx
);
902 ord
= get_imsm_ord_tbl_ent(dev
, slot
, -1);
903 printf(" This Slot : %d%s\n", slot
,
904 ord
& IMSM_ORD_REBUILD
? " (out-of-sync)" : "");
906 printf(" This Slot : ?\n");
907 sz
= __le32_to_cpu(dev
->size_high
);
909 sz
+= __le32_to_cpu(dev
->size_low
);
910 printf(" Array Size : %llu%s\n", (unsigned long long)sz
,
911 human_size(sz
* 512));
912 sz
= __le32_to_cpu(map
->blocks_per_member
);
913 printf(" Per Dev Size : %llu%s\n", (unsigned long long)sz
,
914 human_size(sz
* 512));
915 printf(" Sector Offset : %u\n",
916 __le32_to_cpu(map
->pba_of_lba0
));
917 printf(" Num Stripes : %u\n",
918 __le32_to_cpu(map
->num_data_stripes
));
919 printf(" Chunk Size : %u KiB",
920 __le16_to_cpu(map
->blocks_per_strip
) / 2);
922 printf(" <-- %u KiB",
923 __le16_to_cpu(map2
->blocks_per_strip
) / 2);
925 printf(" Reserved : %d\n", __le32_to_cpu(dev
->reserved_blocks
));
926 printf(" Migrate State : ");
927 if (dev
->vol
.migr_state
) {
928 if (migr_type(dev
) == MIGR_INIT
)
929 printf("initialize\n");
930 else if (migr_type(dev
) == MIGR_REBUILD
)
932 else if (migr_type(dev
) == MIGR_VERIFY
)
934 else if (migr_type(dev
) == MIGR_GEN_MIGR
)
935 printf("general migration\n");
936 else if (migr_type(dev
) == MIGR_STATE_CHANGE
)
937 printf("state change\n");
938 else if (migr_type(dev
) == MIGR_REPAIR
)
941 printf("<unknown:%d>\n", migr_type(dev
));
944 printf(" Map State : %s", map_state_str
[map
->map_state
]);
945 if (dev
->vol
.migr_state
) {
946 struct imsm_map
*map
= get_imsm_map(dev
, 1);
948 printf(" <-- %s", map_state_str
[map
->map_state
]);
949 printf("\n Checkpoint : %u (%llu)",
950 __le32_to_cpu(dev
->vol
.curr_migr_unit
),
951 (unsigned long long)blocks_per_migr_unit(dev
));
954 printf(" Dirty State : %s\n", dev
->vol
.dirty
? "dirty" : "clean");
957 static void print_imsm_disk(struct imsm_super
*mpb
, int index
, __u32 reserved
)
959 struct imsm_disk
*disk
= __get_imsm_disk(mpb
, index
);
960 char str
[MAX_RAID_SERIAL_LEN
+ 1];
963 if (index
< 0 || !disk
)
967 snprintf(str
, MAX_RAID_SERIAL_LEN
+ 1, "%s", disk
->serial
);
968 printf(" Disk%02d Serial : %s\n", index
, str
);
969 printf(" State :%s%s%s\n", is_spare(disk
) ? " spare" : "",
970 is_configured(disk
) ? " active" : "",
971 is_failed(disk
) ? " failed" : "");
972 printf(" Id : %08x\n", __le32_to_cpu(disk
->scsi_id
));
973 sz
= __le32_to_cpu(disk
->total_blocks
) - reserved
;
974 printf(" Usable Size : %llu%s\n", (unsigned long long)sz
,
975 human_size(sz
* 512));
978 static void getinfo_super_imsm(struct supertype
*st
, struct mdinfo
*info
, char *map
);
980 static void examine_super_imsm(struct supertype
*st
, char *homehost
)
982 struct intel_super
*super
= st
->sb
;
983 struct imsm_super
*mpb
= super
->anchor
;
984 char str
[MAX_SIGNATURE_LENGTH
];
989 __u32 reserved
= imsm_reserved_sectors(super
, super
->disks
);
992 snprintf(str
, MPB_SIG_LEN
, "%s", mpb
->sig
);
993 printf(" Magic : %s\n", str
);
994 snprintf(str
, strlen(MPB_VERSION_RAID0
), "%s", get_imsm_version(mpb
));
995 printf(" Version : %s\n", get_imsm_version(mpb
));
996 printf(" Orig Family : %08x\n", __le32_to_cpu(mpb
->orig_family_num
));
997 printf(" Family : %08x\n", __le32_to_cpu(mpb
->family_num
));
998 printf(" Generation : %08x\n", __le32_to_cpu(mpb
->generation_num
));
999 getinfo_super_imsm(st
, &info
, NULL
);
1000 fname_from_uuid(st
, &info
, nbuf
, ':');
1001 printf(" UUID : %s\n", nbuf
+ 5);
1002 sum
= __le32_to_cpu(mpb
->check_sum
);
1003 printf(" Checksum : %08x %s\n", sum
,
1004 __gen_imsm_checksum(mpb
) == sum
? "correct" : "incorrect");
1005 printf(" MPB Sectors : %d\n", mpb_sectors(mpb
));
1006 printf(" Disks : %d\n", mpb
->num_disks
);
1007 printf(" RAID Devices : %d\n", mpb
->num_raid_devs
);
1008 print_imsm_disk(mpb
, super
->disks
->index
, reserved
);
1009 if (super
->bbm_log
) {
1010 struct bbm_log
*log
= super
->bbm_log
;
1013 printf("Bad Block Management Log:\n");
1014 printf(" Log Size : %d\n", __le32_to_cpu(mpb
->bbm_log_size
));
1015 printf(" Signature : %x\n", __le32_to_cpu(log
->signature
));
1016 printf(" Entry Count : %d\n", __le32_to_cpu(log
->entry_count
));
1017 printf(" Spare Blocks : %d\n", __le32_to_cpu(log
->reserved_spare_block_count
));
1018 printf(" First Spare : %llx\n",
1019 (unsigned long long) __le64_to_cpu(log
->first_spare_lba
));
1021 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
1023 struct imsm_dev
*dev
= __get_imsm_dev(mpb
, i
);
1025 super
->current_vol
= i
;
1026 getinfo_super_imsm(st
, &info
, NULL
);
1027 fname_from_uuid(st
, &info
, nbuf
, ':');
1028 print_imsm_dev(dev
, nbuf
+ 5, super
->disks
->index
);
1030 for (i
= 0; i
< mpb
->num_disks
; i
++) {
1031 if (i
== super
->disks
->index
)
1033 print_imsm_disk(mpb
, i
, reserved
);
1035 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
1036 struct imsm_disk
*disk
;
1037 char str
[MAX_RAID_SERIAL_LEN
+ 1];
1045 snprintf(str
, MAX_RAID_SERIAL_LEN
+ 1, "%s", disk
->serial
);
1046 printf(" Disk Serial : %s\n", str
);
1047 printf(" State :%s%s%s\n", is_spare(disk
) ? " spare" : "",
1048 is_configured(disk
) ? " active" : "",
1049 is_failed(disk
) ? " failed" : "");
1050 printf(" Id : %08x\n", __le32_to_cpu(disk
->scsi_id
));
1051 sz
= __le32_to_cpu(disk
->total_blocks
) - reserved
;
1052 printf(" Usable Size : %llu%s\n", (unsigned long long)sz
,
1053 human_size(sz
* 512));
1057 static void brief_examine_super_imsm(struct supertype
*st
, int verbose
)
1059 /* We just write a generic IMSM ARRAY entry */
1062 struct intel_super
*super
= st
->sb
;
1064 if (!super
->anchor
->num_raid_devs
) {
1065 printf("ARRAY metadata=imsm\n");
1069 getinfo_super_imsm(st
, &info
, NULL
);
1070 fname_from_uuid(st
, &info
, nbuf
, ':');
1071 printf("ARRAY metadata=imsm UUID=%s\n", nbuf
+ 5);
1074 static void brief_examine_subarrays_imsm(struct supertype
*st
, int verbose
)
1076 /* We just write a generic IMSM ARRAY entry */
1080 struct intel_super
*super
= st
->sb
;
1083 if (!super
->anchor
->num_raid_devs
)
1086 getinfo_super_imsm(st
, &info
, NULL
);
1087 fname_from_uuid(st
, &info
, nbuf
, ':');
1088 for (i
= 0; i
< super
->anchor
->num_raid_devs
; i
++) {
1089 struct imsm_dev
*dev
= get_imsm_dev(super
, i
);
1091 super
->current_vol
= i
;
1092 getinfo_super_imsm(st
, &info
, NULL
);
1093 fname_from_uuid(st
, &info
, nbuf1
, ':');
1094 printf("ARRAY /dev/md/%.16s container=%s member=%d UUID=%s\n",
1095 dev
->volume
, nbuf
+ 5, i
, nbuf1
+ 5);
1099 static void export_examine_super_imsm(struct supertype
*st
)
1101 struct intel_super
*super
= st
->sb
;
1102 struct imsm_super
*mpb
= super
->anchor
;
1106 getinfo_super_imsm(st
, &info
, NULL
);
1107 fname_from_uuid(st
, &info
, nbuf
, ':');
1108 printf("MD_METADATA=imsm\n");
1109 printf("MD_LEVEL=container\n");
1110 printf("MD_UUID=%s\n", nbuf
+5);
1111 printf("MD_DEVICES=%u\n", mpb
->num_disks
);
1114 static void detail_super_imsm(struct supertype
*st
, char *homehost
)
1119 getinfo_super_imsm(st
, &info
, NULL
);
1120 fname_from_uuid(st
, &info
, nbuf
, ':');
1121 printf("\n UUID : %s\n", nbuf
+ 5);
1124 static void brief_detail_super_imsm(struct supertype
*st
)
1128 getinfo_super_imsm(st
, &info
, NULL
);
1129 fname_from_uuid(st
, &info
, nbuf
, ':');
1130 printf(" UUID=%s", nbuf
+ 5);
1133 static int imsm_read_serial(int fd
, char *devname
, __u8
*serial
);
1134 static void fd2devname(int fd
, char *name
);
1136 static int imsm_enumerate_ports(const char *hba_path
, int port_count
, int host_base
, int verbose
)
1138 /* dump an unsorted list of devices attached to ahci, as well as
1139 * non-connected ports
1141 int hba_len
= strlen(hba_path
) + 1;
1146 unsigned long port_mask
= (1 << port_count
) - 1;
1148 if (port_count
> (int)sizeof(port_mask
) * 8) {
1150 fprintf(stderr
, Name
": port_count %d out of range\n", port_count
);
1154 /* scroll through /sys/dev/block looking for devices attached to
1157 dir
= opendir("/sys/dev/block");
1158 for (ent
= dir
? readdir(dir
) : NULL
; ent
; ent
= readdir(dir
)) {
1169 if (sscanf(ent
->d_name
, "%d:%d", &major
, &minor
) != 2)
1171 path
= devt_to_devpath(makedev(major
, minor
));
1174 if (!path_attached_to_hba(path
, hba_path
)) {
1180 /* retrieve the scsi device type */
1181 if (asprintf(&device
, "/sys/dev/block/%d:%d/device/xxxxxxx", major
, minor
) < 0) {
1183 fprintf(stderr
, Name
": failed to allocate 'device'\n");
1187 sprintf(device
, "/sys/dev/block/%d:%d/device/type", major
, minor
);
1188 if (load_sys(device
, buf
) != 0) {
1190 fprintf(stderr
, Name
": failed to read device type for %s\n",
1196 type
= strtoul(buf
, NULL
, 10);
1198 /* if it's not a disk print the vendor and model */
1199 if (!(type
== 0 || type
== 7 || type
== 14)) {
1202 sprintf(device
, "/sys/dev/block/%d:%d/device/vendor", major
, minor
);
1203 if (load_sys(device
, buf
) == 0) {
1204 strncpy(vendor
, buf
, sizeof(vendor
));
1205 vendor
[sizeof(vendor
) - 1] = '\0';
1206 c
= (char *) &vendor
[sizeof(vendor
) - 1];
1207 while (isspace(*c
) || *c
== '\0')
1211 sprintf(device
, "/sys/dev/block/%d:%d/device/model", major
, minor
);
1212 if (load_sys(device
, buf
) == 0) {
1213 strncpy(model
, buf
, sizeof(model
));
1214 model
[sizeof(model
) - 1] = '\0';
1215 c
= (char *) &model
[sizeof(model
) - 1];
1216 while (isspace(*c
) || *c
== '\0')
1220 if (vendor
[0] && model
[0])
1221 sprintf(buf
, "%.64s %.64s", vendor
, model
);
1223 switch (type
) { /* numbers from hald/linux/device.c */
1224 case 1: sprintf(buf
, "tape"); break;
1225 case 2: sprintf(buf
, "printer"); break;
1226 case 3: sprintf(buf
, "processor"); break;
1228 case 5: sprintf(buf
, "cdrom"); break;
1229 case 6: sprintf(buf
, "scanner"); break;
1230 case 8: sprintf(buf
, "media_changer"); break;
1231 case 9: sprintf(buf
, "comm"); break;
1232 case 12: sprintf(buf
, "raid"); break;
1233 default: sprintf(buf
, "unknown");
1239 /* chop device path to 'host%d' and calculate the port number */
1240 c
= strchr(&path
[hba_len
], '/');
1243 fprintf(stderr
, Name
": %s - invalid path name\n", path
+ hba_len
);
1248 if (sscanf(&path
[hba_len
], "host%d", &port
) == 1)
1252 *c
= '/'; /* repair the full string */
1253 fprintf(stderr
, Name
": failed to determine port number for %s\n",
1260 /* mark this port as used */
1261 port_mask
&= ~(1 << port
);
1263 /* print out the device information */
1265 printf(" Port%d : - non-disk device (%s) -\n", port
, buf
);
1269 fd
= dev_open(ent
->d_name
, O_RDONLY
);
1271 printf(" Port%d : - disk info unavailable -\n", port
);
1273 fd2devname(fd
, buf
);
1274 printf(" Port%d : %s", port
, buf
);
1275 if (imsm_read_serial(fd
, NULL
, (__u8
*) buf
) == 0)
1276 printf(" (%s)\n", buf
);
1291 for (i
= 0; i
< port_count
; i
++)
1292 if (port_mask
& (1 << i
))
1293 printf(" Port%d : - no device attached -\n", i
);
1299 static int detail_platform_imsm(int verbose
, int enumerate_only
)
1301 /* There are two components to imsm platform support, the ahci SATA
1302 * controller and the option-rom. To find the SATA controller we
1303 * simply look in /sys/bus/pci/drivers/ahci to see if an ahci
1304 * controller with the Intel vendor id is present. This approach
1305 * allows mdadm to leverage the kernel's ahci detection logic, with the
1306 * caveat that if ahci.ko is not loaded mdadm will not be able to
1307 * detect platform raid capabilities. The option-rom resides in a
1308 * platform "Adapter ROM". We scan for its signature to retrieve the
1309 * platform capabilities. If raid support is disabled in the BIOS the
1310 * option-rom capability structure will not be available.
1312 const struct imsm_orom
*orom
;
1313 struct sys_dev
*list
, *hba
;
1316 const char *hba_path
;
1320 if (enumerate_only
) {
1321 if (check_env("IMSM_NO_PLATFORM") || find_imsm_orom())
1326 list
= find_driver_devices("pci", "ahci");
1327 for (hba
= list
; hba
; hba
= hba
->next
)
1328 if (devpath_to_vendor(hba
->path
) == 0x8086)
1333 fprintf(stderr
, Name
": unable to find active ahci controller\n");
1334 free_sys_dev(&list
);
1337 fprintf(stderr
, Name
": found Intel SATA AHCI Controller\n");
1338 hba_path
= hba
->path
;
1340 free_sys_dev(&list
);
1342 orom
= find_imsm_orom();
1345 fprintf(stderr
, Name
": imsm option-rom not found\n");
1349 printf(" Platform : Intel(R) Matrix Storage Manager\n");
1350 printf(" Version : %d.%d.%d.%d\n", orom
->major_ver
, orom
->minor_ver
,
1351 orom
->hotfix_ver
, orom
->build
);
1352 printf(" RAID Levels :%s%s%s%s%s\n",
1353 imsm_orom_has_raid0(orom
) ? " raid0" : "",
1354 imsm_orom_has_raid1(orom
) ? " raid1" : "",
1355 imsm_orom_has_raid1e(orom
) ? " raid1e" : "",
1356 imsm_orom_has_raid10(orom
) ? " raid10" : "",
1357 imsm_orom_has_raid5(orom
) ? " raid5" : "");
1358 printf(" Chunk Sizes :%s%s%s%s%s%s%s%s%s%s%s%s%s%s%s%s\n",
1359 imsm_orom_has_chunk(orom
, 2) ? " 2k" : "",
1360 imsm_orom_has_chunk(orom
, 4) ? " 4k" : "",
1361 imsm_orom_has_chunk(orom
, 8) ? " 8k" : "",
1362 imsm_orom_has_chunk(orom
, 16) ? " 16k" : "",
1363 imsm_orom_has_chunk(orom
, 32) ? " 32k" : "",
1364 imsm_orom_has_chunk(orom
, 64) ? " 64k" : "",
1365 imsm_orom_has_chunk(orom
, 128) ? " 128k" : "",
1366 imsm_orom_has_chunk(orom
, 256) ? " 256k" : "",
1367 imsm_orom_has_chunk(orom
, 512) ? " 512k" : "",
1368 imsm_orom_has_chunk(orom
, 1024*1) ? " 1M" : "",
1369 imsm_orom_has_chunk(orom
, 1024*2) ? " 2M" : "",
1370 imsm_orom_has_chunk(orom
, 1024*4) ? " 4M" : "",
1371 imsm_orom_has_chunk(orom
, 1024*8) ? " 8M" : "",
1372 imsm_orom_has_chunk(orom
, 1024*16) ? " 16M" : "",
1373 imsm_orom_has_chunk(orom
, 1024*32) ? " 32M" : "",
1374 imsm_orom_has_chunk(orom
, 1024*64) ? " 64M" : "");
1375 printf(" Max Disks : %d\n", orom
->tds
);
1376 printf(" Max Volumes : %d\n", orom
->vpa
);
1377 printf(" I/O Controller : %s\n", hba_path
);
1379 /* find the smallest scsi host number to determine a port number base */
1380 dir
= opendir(hba_path
);
1381 for (ent
= dir
? readdir(dir
) : NULL
; ent
; ent
= readdir(dir
)) {
1384 if (sscanf(ent
->d_name
, "host%d", &host
) != 1)
1386 if (port_count
== 0)
1388 else if (host
< host_base
)
1391 if (host
+ 1 > port_count
+ host_base
)
1392 port_count
= host
+ 1 - host_base
;
1398 if (!port_count
|| imsm_enumerate_ports(hba_path
, port_count
,
1399 host_base
, verbose
) != 0) {
1401 fprintf(stderr
, Name
": failed to enumerate ports\n");
1409 static int match_home_imsm(struct supertype
*st
, char *homehost
)
1411 /* the imsm metadata format does not specify any host
1412 * identification information. We return -1 since we can never
1413 * confirm nor deny whether a given array is "meant" for this
1414 * host. We rely on compare_super and the 'family_num' fields to
1415 * exclude member disks that do not belong, and we rely on
1416 * mdadm.conf to specify the arrays that should be assembled.
1417 * Auto-assembly may still pick up "foreign" arrays.
1423 static void uuid_from_super_imsm(struct supertype
*st
, int uuid
[4])
1425 /* The uuid returned here is used for:
1426 * uuid to put into bitmap file (Create, Grow)
1427 * uuid for backup header when saving critical section (Grow)
1428 * comparing uuids when re-adding a device into an array
1429 * In these cases the uuid required is that of the data-array,
1430 * not the device-set.
1431 * uuid to recognise same set when adding a missing device back
1432 * to an array. This is a uuid for the device-set.
1434 * For each of these we can make do with a truncated
1435 * or hashed uuid rather than the original, as long as
1437 * In each case the uuid required is that of the data-array,
1438 * not the device-set.
1440 /* imsm does not track uuid's so we synthesis one using sha1 on
1441 * - The signature (Which is constant for all imsm array, but no matter)
1442 * - the orig_family_num of the container
1443 * - the index number of the volume
1444 * - the 'serial' number of the volume.
1445 * Hopefully these are all constant.
1447 struct intel_super
*super
= st
->sb
;
1450 struct sha1_ctx ctx
;
1451 struct imsm_dev
*dev
= NULL
;
1454 /* some mdadm versions failed to set ->orig_family_num, in which
1455 * case fall back to ->family_num. orig_family_num will be
1456 * fixed up with the first metadata update.
1458 family_num
= super
->anchor
->orig_family_num
;
1459 if (family_num
== 0)
1460 family_num
= super
->anchor
->family_num
;
1461 sha1_init_ctx(&ctx
);
1462 sha1_process_bytes(super
->anchor
->sig
, MPB_SIG_LEN
, &ctx
);
1463 sha1_process_bytes(&family_num
, sizeof(__u32
), &ctx
);
1464 if (super
->current_vol
>= 0)
1465 dev
= get_imsm_dev(super
, super
->current_vol
);
1467 __u32 vol
= super
->current_vol
;
1468 sha1_process_bytes(&vol
, sizeof(vol
), &ctx
);
1469 sha1_process_bytes(dev
->volume
, MAX_RAID_SERIAL_LEN
, &ctx
);
1471 sha1_finish_ctx(&ctx
, buf
);
1472 memcpy(uuid
, buf
, 4*4);
1477 get_imsm_numerical_version(struct imsm_super
*mpb
, int *m
, int *p
)
1479 __u8
*v
= get_imsm_version(mpb
);
1480 __u8
*end
= mpb
->sig
+ MAX_SIGNATURE_LENGTH
;
1481 char major
[] = { 0, 0, 0 };
1482 char minor
[] = { 0 ,0, 0 };
1483 char patch
[] = { 0, 0, 0 };
1484 char *ver_parse
[] = { major
, minor
, patch
};
1488 while (*v
!= '\0' && v
< end
) {
1489 if (*v
!= '.' && j
< 2)
1490 ver_parse
[i
][j
++] = *v
;
1498 *m
= strtol(minor
, NULL
, 0);
1499 *p
= strtol(patch
, NULL
, 0);
1503 static __u32
migr_strip_blocks_resync(struct imsm_dev
*dev
)
1505 /* migr_strip_size when repairing or initializing parity */
1506 struct imsm_map
*map
= get_imsm_map(dev
, 0);
1507 __u32 chunk
= __le32_to_cpu(map
->blocks_per_strip
);
1509 switch (get_imsm_raid_level(map
)) {
1514 return 128*1024 >> 9;
1518 static __u32
migr_strip_blocks_rebuild(struct imsm_dev
*dev
)
1520 /* migr_strip_size when rebuilding a degraded disk, no idea why
1521 * this is different than migr_strip_size_resync(), but it's good
1524 struct imsm_map
*map
= get_imsm_map(dev
, 1);
1525 __u32 chunk
= __le32_to_cpu(map
->blocks_per_strip
);
1527 switch (get_imsm_raid_level(map
)) {
1530 if (map
->num_members
% map
->num_domains
== 0)
1531 return 128*1024 >> 9;
1535 return max((__u32
) 64*1024 >> 9, chunk
);
1537 return 128*1024 >> 9;
1541 static __u32
num_stripes_per_unit_resync(struct imsm_dev
*dev
)
1543 struct imsm_map
*lo
= get_imsm_map(dev
, 0);
1544 struct imsm_map
*hi
= get_imsm_map(dev
, 1);
1545 __u32 lo_chunk
= __le32_to_cpu(lo
->blocks_per_strip
);
1546 __u32 hi_chunk
= __le32_to_cpu(hi
->blocks_per_strip
);
1548 return max((__u32
) 1, hi_chunk
/ lo_chunk
);
1551 static __u32
num_stripes_per_unit_rebuild(struct imsm_dev
*dev
)
1553 struct imsm_map
*lo
= get_imsm_map(dev
, 0);
1554 int level
= get_imsm_raid_level(lo
);
1556 if (level
== 1 || level
== 10) {
1557 struct imsm_map
*hi
= get_imsm_map(dev
, 1);
1559 return hi
->num_domains
;
1561 return num_stripes_per_unit_resync(dev
);
1564 static __u8
imsm_num_data_members(struct imsm_dev
*dev
, int second_map
)
1566 /* named 'imsm_' because raid0, raid1 and raid10
1567 * counter-intuitively have the same number of data disks
1569 struct imsm_map
*map
= get_imsm_map(dev
, second_map
);
1571 switch (get_imsm_raid_level(map
)) {
1575 return map
->num_members
;
1577 return map
->num_members
- 1;
1579 dprintf("%s: unsupported raid level\n", __func__
);
1584 static __u32
parity_segment_depth(struct imsm_dev
*dev
)
1586 struct imsm_map
*map
= get_imsm_map(dev
, 0);
1587 __u32 chunk
= __le32_to_cpu(map
->blocks_per_strip
);
1589 switch(get_imsm_raid_level(map
)) {
1592 return chunk
* map
->num_domains
;
1594 return chunk
* map
->num_members
;
1600 static __u32
map_migr_block(struct imsm_dev
*dev
, __u32 block
)
1602 struct imsm_map
*map
= get_imsm_map(dev
, 1);
1603 __u32 chunk
= __le32_to_cpu(map
->blocks_per_strip
);
1604 __u32 strip
= block
/ chunk
;
1606 switch (get_imsm_raid_level(map
)) {
1609 __u32 vol_strip
= (strip
* map
->num_domains
) + 1;
1610 __u32 vol_stripe
= vol_strip
/ map
->num_members
;
1612 return vol_stripe
* chunk
+ block
% chunk
;
1614 __u32 stripe
= strip
/ (map
->num_members
- 1);
1616 return stripe
* chunk
+ block
% chunk
;
1623 static __u64
blocks_per_migr_unit(struct imsm_dev
*dev
)
1625 /* calculate the conversion factor between per member 'blocks'
1626 * (md/{resync,rebuild}_start) and imsm migration units, return
1627 * 0 for the 'not migrating' and 'unsupported migration' cases
1629 if (!dev
->vol
.migr_state
)
1632 switch (migr_type(dev
)) {
1637 struct imsm_map
*map
= get_imsm_map(dev
, 0);
1638 __u32 stripes_per_unit
;
1639 __u32 blocks_per_unit
;
1648 /* yes, this is really the translation of migr_units to
1649 * per-member blocks in the 'resync' case
1651 stripes_per_unit
= num_stripes_per_unit_resync(dev
);
1652 migr_chunk
= migr_strip_blocks_resync(dev
);
1653 disks
= imsm_num_data_members(dev
, 0);
1654 blocks_per_unit
= stripes_per_unit
* migr_chunk
* disks
;
1655 stripe
= __le32_to_cpu(map
->blocks_per_strip
) * disks
;
1656 segment
= blocks_per_unit
/ stripe
;
1657 block_rel
= blocks_per_unit
- segment
* stripe
;
1658 parity_depth
= parity_segment_depth(dev
);
1659 block_map
= map_migr_block(dev
, block_rel
);
1660 return block_map
+ parity_depth
* segment
;
1662 case MIGR_REBUILD
: {
1663 __u32 stripes_per_unit
;
1666 stripes_per_unit
= num_stripes_per_unit_rebuild(dev
);
1667 migr_chunk
= migr_strip_blocks_rebuild(dev
);
1668 return migr_chunk
* stripes_per_unit
;
1670 case MIGR_STATE_CHANGE
:
1676 static int imsm_level_to_layout(int level
)
1684 return ALGORITHM_LEFT_ASYMMETRIC
;
1691 static void getinfo_super_imsm_volume(struct supertype
*st
, struct mdinfo
*info
, char *dmap
)
1693 struct intel_super
*super
= st
->sb
;
1694 struct imsm_dev
*dev
= get_imsm_dev(super
, super
->current_vol
);
1695 struct imsm_map
*map
= get_imsm_map(dev
, 0);
1696 struct imsm_map
*prev_map
= get_imsm_map(dev
, 1);
1697 struct imsm_map
*map_to_analyse
= map
;
1700 int map_disks
= info
->array
.raid_disks
;
1703 map_to_analyse
= prev_map
;
1705 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
1706 if (dl
->raiddisk
== info
->disk
.raid_disk
)
1708 info
->container_member
= super
->current_vol
;
1709 info
->array
.raid_disks
= map_to_analyse
->num_members
;
1710 info
->array
.level
= get_imsm_raid_level(map_to_analyse
);
1711 info
->array
.layout
= imsm_level_to_layout(info
->array
.level
);
1712 info
->array
.md_minor
= -1;
1713 info
->array
.ctime
= 0;
1714 info
->array
.utime
= 0;
1715 info
->array
.chunk_size
=
1716 __le16_to_cpu(map_to_analyse
->blocks_per_strip
) << 9;
1717 info
->array
.state
= !dev
->vol
.dirty
;
1718 info
->custom_array_size
= __le32_to_cpu(dev
->size_high
);
1719 info
->custom_array_size
<<= 32;
1720 info
->custom_array_size
|= __le32_to_cpu(dev
->size_low
);
1722 info
->new_level
= get_imsm_raid_level(map
);
1723 info
->new_layout
= imsm_level_to_layout(info
->new_level
);
1724 info
->new_chunk
= __le16_to_cpu(map
->blocks_per_strip
) << 9;
1726 info
->new_level
= UnSet
;
1727 info
->new_layout
= UnSet
;
1728 info
->new_chunk
= info
->array
.chunk_size
;
1730 info
->disk
.major
= 0;
1731 info
->disk
.minor
= 0;
1733 info
->disk
.major
= dl
->major
;
1734 info
->disk
.minor
= dl
->minor
;
1737 info
->data_offset
= __le32_to_cpu(map_to_analyse
->pba_of_lba0
);
1738 info
->component_size
=
1739 __le32_to_cpu(map_to_analyse
->blocks_per_member
);
1740 memset(info
->uuid
, 0, sizeof(info
->uuid
));
1741 info
->recovery_start
= MaxSector
;
1742 info
->reshape_active
= (prev_map
!= NULL
);
1743 if (info
->reshape_active
)
1744 info
->delta_disks
= map
->num_members
- prev_map
->num_members
;
1746 info
->delta_disks
= 0;
1748 if (map_to_analyse
->map_state
== IMSM_T_STATE_UNINITIALIZED
||
1750 info
->resync_start
= 0;
1751 } else if (dev
->vol
.migr_state
) {
1752 switch (migr_type(dev
)) {
1755 __u64 blocks_per_unit
= blocks_per_migr_unit(dev
);
1756 __u64 units
= __le32_to_cpu(dev
->vol
.curr_migr_unit
);
1758 info
->resync_start
= blocks_per_unit
* units
;
1762 /* we could emulate the checkpointing of
1763 * 'sync_action=check' migrations, but for now
1764 * we just immediately complete them
1767 /* this is handled by container_content_imsm() */
1769 case MIGR_STATE_CHANGE
:
1770 /* FIXME handle other migrations */
1772 /* we are not dirty, so... */
1773 info
->resync_start
= MaxSector
;
1776 info
->resync_start
= MaxSector
;
1778 strncpy(info
->name
, (char *) dev
->volume
, MAX_RAID_SERIAL_LEN
);
1779 info
->name
[MAX_RAID_SERIAL_LEN
] = 0;
1781 info
->array
.major_version
= -1;
1782 info
->array
.minor_version
= -2;
1783 devname
= devnum2devname(st
->container_dev
);
1784 *info
->text_version
= '\0';
1786 sprintf(info
->text_version
, "/%s/%d", devname
, info
->container_member
);
1788 info
->safe_mode_delay
= 4000; /* 4 secs like the Matrix driver */
1789 uuid_from_super_imsm(st
, info
->uuid
);
1793 for (i
=0; i
<map_disks
; i
++) {
1795 if (i
< info
->array
.raid_disks
) {
1796 struct imsm_disk
*dsk
;
1797 j
= get_imsm_disk_idx(dev
, i
, -1);
1798 dsk
= get_imsm_disk(super
, j
);
1799 if (dsk
&& (dsk
->status
& CONFIGURED_DISK
))
1806 static __u8
imsm_check_degraded(struct intel_super
*super
, struct imsm_dev
*dev
, int failed
);
1807 static int imsm_count_failed(struct intel_super
*super
, struct imsm_dev
*dev
);
1809 static struct imsm_disk
*get_imsm_missing(struct intel_super
*super
, __u8 index
)
1813 for (d
= super
->missing
; d
; d
= d
->next
)
1814 if (d
->index
== index
)
1819 static void getinfo_super_imsm(struct supertype
*st
, struct mdinfo
*info
, char *map
)
1821 struct intel_super
*super
= st
->sb
;
1822 struct imsm_disk
*disk
;
1823 int map_disks
= info
->array
.raid_disks
;
1824 int max_enough
= -1;
1826 struct imsm_super
*mpb
;
1828 if (super
->current_vol
>= 0) {
1829 getinfo_super_imsm_volume(st
, info
, map
);
1833 /* Set raid_disks to zero so that Assemble will always pull in valid
1836 info
->array
.raid_disks
= 0;
1837 info
->array
.level
= LEVEL_CONTAINER
;
1838 info
->array
.layout
= 0;
1839 info
->array
.md_minor
= -1;
1840 info
->array
.ctime
= 0; /* N/A for imsm */
1841 info
->array
.utime
= 0;
1842 info
->array
.chunk_size
= 0;
1844 info
->disk
.major
= 0;
1845 info
->disk
.minor
= 0;
1846 info
->disk
.raid_disk
= -1;
1847 info
->reshape_active
= 0;
1848 info
->array
.major_version
= -1;
1849 info
->array
.minor_version
= -2;
1850 strcpy(info
->text_version
, "imsm");
1851 info
->safe_mode_delay
= 0;
1852 info
->disk
.number
= -1;
1853 info
->disk
.state
= 0;
1855 info
->recovery_start
= MaxSector
;
1857 /* do we have the all the insync disks that we expect? */
1858 mpb
= super
->anchor
;
1860 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
1861 struct imsm_dev
*dev
= get_imsm_dev(super
, i
);
1862 int failed
, enough
, j
, missing
= 0;
1863 struct imsm_map
*map
;
1866 failed
= imsm_count_failed(super
, dev
);
1867 state
= imsm_check_degraded(super
, dev
, failed
);
1868 map
= get_imsm_map(dev
, dev
->vol
.migr_state
);
1870 /* any newly missing disks?
1871 * (catches single-degraded vs double-degraded)
1873 for (j
= 0; j
< map
->num_members
; j
++) {
1874 __u32 ord
= get_imsm_ord_tbl_ent(dev
, i
, -1);
1875 __u32 idx
= ord_to_idx(ord
);
1877 if (!(ord
& IMSM_ORD_REBUILD
) &&
1878 get_imsm_missing(super
, idx
)) {
1884 if (state
== IMSM_T_STATE_FAILED
)
1886 else if (state
== IMSM_T_STATE_DEGRADED
&&
1887 (state
!= map
->map_state
|| missing
))
1889 else /* we're normal, or already degraded */
1892 /* in the missing/failed disk case check to see
1893 * if at least one array is runnable
1895 max_enough
= max(max_enough
, enough
);
1897 dprintf("%s: enough: %d\n", __func__
, max_enough
);
1898 info
->container_enough
= max_enough
;
1901 __u32 reserved
= imsm_reserved_sectors(super
, super
->disks
);
1903 disk
= &super
->disks
->disk
;
1904 info
->data_offset
= __le32_to_cpu(disk
->total_blocks
) - reserved
;
1905 info
->component_size
= reserved
;
1906 info
->disk
.state
= is_configured(disk
) ? (1 << MD_DISK_ACTIVE
) : 0;
1907 /* we don't change info->disk.raid_disk here because
1908 * this state will be finalized in mdmon after we have
1909 * found the 'most fresh' version of the metadata
1911 info
->disk
.state
|= is_failed(disk
) ? (1 << MD_DISK_FAULTY
) : 0;
1912 info
->disk
.state
|= is_spare(disk
) ? 0 : (1 << MD_DISK_SYNC
);
1915 /* only call uuid_from_super_imsm when this disk is part of a populated container,
1916 * ->compare_super may have updated the 'num_raid_devs' field for spares
1918 if (info
->disk
.state
& (1 << MD_DISK_SYNC
) || super
->anchor
->num_raid_devs
)
1919 uuid_from_super_imsm(st
, info
->uuid
);
1921 memcpy(info
->uuid
, uuid_zero
, sizeof(uuid_zero
));
1923 /* I don't know how to compute 'map' on imsm, so use safe default */
1926 for (i
= 0; i
< map_disks
; i
++)
1932 /* allocates memory and fills disk in mdinfo structure
1933 * for each disk in array */
1934 struct mdinfo
*getinfo_super_disks_imsm(struct supertype
*st
)
1936 struct mdinfo
*mddev
= NULL
;
1937 struct intel_super
*super
= st
->sb
;
1938 struct imsm_disk
*disk
;
1941 if (!super
|| !super
->disks
)
1944 mddev
= malloc(sizeof(*mddev
));
1946 fprintf(stderr
, Name
": Failed to allocate memory.\n");
1949 memset(mddev
, 0, sizeof(*mddev
));
1953 tmp
= malloc(sizeof(*tmp
));
1955 fprintf(stderr
, Name
": Failed to allocate memory.\n");
1960 memset(tmp
, 0, sizeof(*tmp
));
1962 tmp
->next
= mddev
->devs
;
1964 tmp
->disk
.number
= count
++;
1965 tmp
->disk
.major
= dl
->major
;
1966 tmp
->disk
.minor
= dl
->minor
;
1967 tmp
->disk
.state
= is_configured(disk
) ?
1968 (1 << MD_DISK_ACTIVE
) : 0;
1969 tmp
->disk
.state
|= is_failed(disk
) ? (1 << MD_DISK_FAULTY
) : 0;
1970 tmp
->disk
.state
|= is_spare(disk
) ? 0 : (1 << MD_DISK_SYNC
);
1971 tmp
->disk
.raid_disk
= -1;
1977 static int update_super_imsm(struct supertype
*st
, struct mdinfo
*info
,
1978 char *update
, char *devname
, int verbose
,
1979 int uuid_set
, char *homehost
)
1981 /* For 'assemble' and 'force' we need to return non-zero if any
1982 * change was made. For others, the return value is ignored.
1983 * Update options are:
1984 * force-one : This device looks a bit old but needs to be included,
1985 * update age info appropriately.
1986 * assemble: clear any 'faulty' flag to allow this device to
1988 * force-array: Array is degraded but being forced, mark it clean
1989 * if that will be needed to assemble it.
1991 * newdev: not used ????
1992 * grow: Array has gained a new device - this is currently for
1994 * resync: mark as dirty so a resync will happen.
1995 * name: update the name - preserving the homehost
1996 * uuid: Change the uuid of the array to match watch is given
1998 * Following are not relevant for this imsm:
1999 * sparc2.2 : update from old dodgey metadata
2000 * super-minor: change the preferred_minor number
2001 * summaries: update redundant counters.
2002 * homehost: update the recorded homehost
2003 * _reshape_progress: record new reshape_progress position.
2006 struct intel_super
*super
= st
->sb
;
2007 struct imsm_super
*mpb
;
2009 /* we can only update container info */
2010 if (!super
|| super
->current_vol
>= 0 || !super
->anchor
)
2013 mpb
= super
->anchor
;
2015 if (strcmp(update
, "uuid") == 0 && uuid_set
&& !info
->update_private
)
2017 else if (strcmp(update
, "uuid") == 0 && uuid_set
&& info
->update_private
) {
2018 mpb
->orig_family_num
= *((__u32
*) info
->update_private
);
2020 } else if (strcmp(update
, "uuid") == 0) {
2021 __u32
*new_family
= malloc(sizeof(*new_family
));
2023 /* update orig_family_number with the incoming random
2024 * data, report the new effective uuid, and store the
2025 * new orig_family_num for future updates.
2028 memcpy(&mpb
->orig_family_num
, info
->uuid
, sizeof(__u32
));
2029 uuid_from_super_imsm(st
, info
->uuid
);
2030 *new_family
= mpb
->orig_family_num
;
2031 info
->update_private
= new_family
;
2034 } else if (strcmp(update
, "assemble") == 0)
2039 /* successful update? recompute checksum */
2041 mpb
->check_sum
= __le32_to_cpu(__gen_imsm_checksum(mpb
));
2046 static size_t disks_to_mpb_size(int disks
)
2050 size
= sizeof(struct imsm_super
);
2051 size
+= (disks
- 1) * sizeof(struct imsm_disk
);
2052 size
+= 2 * sizeof(struct imsm_dev
);
2053 /* up to 2 maps per raid device (-2 for imsm_maps in imsm_dev */
2054 size
+= (4 - 2) * sizeof(struct imsm_map
);
2055 /* 4 possible disk_ord_tbl's */
2056 size
+= 4 * (disks
- 1) * sizeof(__u32
);
2061 static __u64
avail_size_imsm(struct supertype
*st
, __u64 devsize
)
2063 if (devsize
< (MPB_SECTOR_CNT
+ IMSM_RESERVED_SECTORS
))
2066 return devsize
- (MPB_SECTOR_CNT
+ IMSM_RESERVED_SECTORS
);
2069 static void free_devlist(struct intel_super
*super
)
2071 struct intel_dev
*dv
;
2073 while (super
->devlist
) {
2074 dv
= super
->devlist
->next
;
2075 free(super
->devlist
->dev
);
2076 free(super
->devlist
);
2077 super
->devlist
= dv
;
2081 static void imsm_copy_dev(struct imsm_dev
*dest
, struct imsm_dev
*src
)
2083 memcpy(dest
, src
, sizeof_imsm_dev(src
, 0));
2086 static int compare_super_imsm(struct supertype
*st
, struct supertype
*tst
)
2090 * 0 same, or first was empty, and second was copied
2091 * 1 second had wrong number
2093 * 3 wrong other info
2095 struct intel_super
*first
= st
->sb
;
2096 struct intel_super
*sec
= tst
->sb
;
2104 /* if an anchor does not have num_raid_devs set then it is a free
2107 if (first
->anchor
->num_raid_devs
> 0 &&
2108 sec
->anchor
->num_raid_devs
> 0) {
2109 /* Determine if these disks might ever have been
2110 * related. Further disambiguation can only take place
2111 * in load_super_imsm_all
2113 __u32 first_family
= first
->anchor
->orig_family_num
;
2114 __u32 sec_family
= sec
->anchor
->orig_family_num
;
2116 if (memcmp(first
->anchor
->sig
, sec
->anchor
->sig
,
2117 MAX_SIGNATURE_LENGTH
) != 0)
2120 if (first_family
== 0)
2121 first_family
= first
->anchor
->family_num
;
2122 if (sec_family
== 0)
2123 sec_family
= sec
->anchor
->family_num
;
2125 if (first_family
!= sec_family
)
2131 /* if 'first' is a spare promote it to a populated mpb with sec's
2134 if (first
->anchor
->num_raid_devs
== 0 &&
2135 sec
->anchor
->num_raid_devs
> 0) {
2137 struct intel_dev
*dv
;
2138 struct imsm_dev
*dev
;
2140 /* we need to copy raid device info from sec if an allocation
2141 * fails here we don't associate the spare
2143 for (i
= 0; i
< sec
->anchor
->num_raid_devs
; i
++) {
2144 dv
= malloc(sizeof(*dv
));
2147 dev
= malloc(sizeof_imsm_dev(get_imsm_dev(sec
, i
), 1));
2154 dv
->next
= first
->devlist
;
2155 first
->devlist
= dv
;
2157 if (i
< sec
->anchor
->num_raid_devs
) {
2158 /* allocation failure */
2159 free_devlist(first
);
2160 fprintf(stderr
, "imsm: failed to associate spare\n");
2163 first
->anchor
->num_raid_devs
= sec
->anchor
->num_raid_devs
;
2164 first
->anchor
->orig_family_num
= sec
->anchor
->orig_family_num
;
2165 first
->anchor
->family_num
= sec
->anchor
->family_num
;
2166 memcpy(first
->anchor
->sig
, sec
->anchor
->sig
, MAX_SIGNATURE_LENGTH
);
2167 for (i
= 0; i
< sec
->anchor
->num_raid_devs
; i
++)
2168 imsm_copy_dev(get_imsm_dev(first
, i
), get_imsm_dev(sec
, i
));
2174 static void fd2devname(int fd
, char *name
)
2178 char dname
[PATH_MAX
];
2183 if (fstat(fd
, &st
) != 0)
2185 sprintf(path
, "/sys/dev/block/%d:%d",
2186 major(st
.st_rdev
), minor(st
.st_rdev
));
2188 rv
= readlink(path
, dname
, sizeof(dname
));
2193 nm
= strrchr(dname
, '/');
2195 snprintf(name
, MAX_RAID_SERIAL_LEN
, "/dev/%s", nm
);
2198 extern int scsi_get_serial(int fd
, void *buf
, size_t buf_len
);
2200 static int imsm_read_serial(int fd
, char *devname
,
2201 __u8 serial
[MAX_RAID_SERIAL_LEN
])
2203 unsigned char scsi_serial
[255];
2212 memset(scsi_serial
, 0, sizeof(scsi_serial
));
2214 rv
= scsi_get_serial(fd
, scsi_serial
, sizeof(scsi_serial
));
2216 if (rv
&& check_env("IMSM_DEVNAME_AS_SERIAL")) {
2217 memset(serial
, 0, MAX_RAID_SERIAL_LEN
);
2218 fd2devname(fd
, (char *) serial
);
2225 Name
": Failed to retrieve serial for %s\n",
2230 rsp_len
= scsi_serial
[3];
2234 Name
": Failed to retrieve serial for %s\n",
2238 rsp_buf
= (char *) &scsi_serial
[4];
2240 /* trim all whitespace and non-printable characters and convert
2243 for (i
= 0, dest
= rsp_buf
; i
< rsp_len
; i
++) {
2246 /* ':' is reserved for use in placeholder serial
2247 * numbers for missing disks
2255 len
= dest
- rsp_buf
;
2258 /* truncate leading characters */
2259 if (len
> MAX_RAID_SERIAL_LEN
) {
2260 dest
+= len
- MAX_RAID_SERIAL_LEN
;
2261 len
= MAX_RAID_SERIAL_LEN
;
2264 memset(serial
, 0, MAX_RAID_SERIAL_LEN
);
2265 memcpy(serial
, dest
, len
);
2270 static int serialcmp(__u8
*s1
, __u8
*s2
)
2272 return strncmp((char *) s1
, (char *) s2
, MAX_RAID_SERIAL_LEN
);
2275 static void serialcpy(__u8
*dest
, __u8
*src
)
2277 strncpy((char *) dest
, (char *) src
, MAX_RAID_SERIAL_LEN
);
2281 static struct dl
*serial_to_dl(__u8
*serial
, struct intel_super
*super
)
2285 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
2286 if (serialcmp(dl
->serial
, serial
) == 0)
2293 static struct imsm_disk
*
2294 __serial_to_disk(__u8
*serial
, struct imsm_super
*mpb
, int *idx
)
2298 for (i
= 0; i
< mpb
->num_disks
; i
++) {
2299 struct imsm_disk
*disk
= __get_imsm_disk(mpb
, i
);
2301 if (serialcmp(disk
->serial
, serial
) == 0) {
2312 load_imsm_disk(int fd
, struct intel_super
*super
, char *devname
, int keep_fd
)
2314 struct imsm_disk
*disk
;
2319 __u8 serial
[MAX_RAID_SERIAL_LEN
];
2321 rv
= imsm_read_serial(fd
, devname
, serial
);
2326 dl
= calloc(1, sizeof(*dl
));
2330 Name
": failed to allocate disk buffer for %s\n",
2336 dl
->major
= major(stb
.st_rdev
);
2337 dl
->minor
= minor(stb
.st_rdev
);
2338 dl
->next
= super
->disks
;
2339 dl
->fd
= keep_fd
? fd
: -1;
2340 assert(super
->disks
== NULL
);
2342 serialcpy(dl
->serial
, serial
);
2345 fd2devname(fd
, name
);
2347 dl
->devname
= strdup(devname
);
2349 dl
->devname
= strdup(name
);
2351 /* look up this disk's index in the current anchor */
2352 disk
= __serial_to_disk(dl
->serial
, super
->anchor
, &dl
->index
);
2355 /* only set index on disks that are a member of a
2356 * populated contianer, i.e. one with raid_devs
2358 if (is_failed(&dl
->disk
))
2360 else if (is_spare(&dl
->disk
))
2368 /* When migrating map0 contains the 'destination' state while map1
2369 * contains the current state. When not migrating map0 contains the
2370 * current state. This routine assumes that map[0].map_state is set to
2371 * the current array state before being called.
2373 * Migration is indicated by one of the following states
2374 * 1/ Idle (migr_state=0 map0state=normal||unitialized||degraded||failed)
2375 * 2/ Initialize (migr_state=1 migr_type=MIGR_INIT map0state=normal
2376 * map1state=unitialized)
2377 * 3/ Repair (Resync) (migr_state=1 migr_type=MIGR_REPAIR map0state=normal
2379 * 4/ Rebuild (migr_state=1 migr_type=MIGR_REBUILD map0state=normal
2380 * map1state=degraded)
2382 static void migrate(struct imsm_dev
*dev
, __u8 to_state
, int migr_type
)
2384 struct imsm_map
*dest
;
2385 struct imsm_map
*src
= get_imsm_map(dev
, 0);
2387 dev
->vol
.migr_state
= 1;
2388 set_migr_type(dev
, migr_type
);
2389 dev
->vol
.curr_migr_unit
= 0;
2390 dest
= get_imsm_map(dev
, 1);
2392 /* duplicate and then set the target end state in map[0] */
2393 memcpy(dest
, src
, sizeof_imsm_map(src
));
2394 if ((migr_type
== MIGR_REBUILD
) ||
2395 (migr_type
== MIGR_GEN_MIGR
)) {
2399 for (i
= 0; i
< src
->num_members
; i
++) {
2400 ord
= __le32_to_cpu(src
->disk_ord_tbl
[i
]);
2401 set_imsm_ord_tbl_ent(src
, i
, ord_to_idx(ord
));
2405 src
->map_state
= to_state
;
2408 static void end_migration(struct imsm_dev
*dev
, __u8 map_state
)
2410 struct imsm_map
*map
= get_imsm_map(dev
, 0);
2411 struct imsm_map
*prev
= get_imsm_map(dev
, dev
->vol
.migr_state
);
2414 /* merge any IMSM_ORD_REBUILD bits that were not successfully
2415 * completed in the last migration.
2417 * FIXME add support for raid-level-migration
2419 for (i
= 0; i
< prev
->num_members
; i
++)
2420 for (j
= 0; j
< map
->num_members
; j
++)
2421 /* during online capacity expansion
2422 * disks position can be changed if takeover is used
2424 if (ord_to_idx(map
->disk_ord_tbl
[j
]) ==
2425 ord_to_idx(prev
->disk_ord_tbl
[i
])) {
2426 map
->disk_ord_tbl
[j
] |= prev
->disk_ord_tbl
[i
];
2430 dev
->vol
.migr_state
= 0;
2431 dev
->vol
.migr_type
= 0;
2432 dev
->vol
.curr_migr_unit
= 0;
2433 map
->map_state
= map_state
;
2437 static int parse_raid_devices(struct intel_super
*super
)
2440 struct imsm_dev
*dev_new
;
2441 size_t len
, len_migr
;
2442 size_t space_needed
= 0;
2443 struct imsm_super
*mpb
= super
->anchor
;
2445 for (i
= 0; i
< super
->anchor
->num_raid_devs
; i
++) {
2446 struct imsm_dev
*dev_iter
= __get_imsm_dev(super
->anchor
, i
);
2447 struct intel_dev
*dv
;
2449 len
= sizeof_imsm_dev(dev_iter
, 0);
2450 len_migr
= sizeof_imsm_dev(dev_iter
, 1);
2452 space_needed
+= len_migr
- len
;
2454 dv
= malloc(sizeof(*dv
));
2457 dev_new
= malloc(len_migr
);
2462 imsm_copy_dev(dev_new
, dev_iter
);
2465 dv
->next
= super
->devlist
;
2466 super
->devlist
= dv
;
2469 /* ensure that super->buf is large enough when all raid devices
2472 if (__le32_to_cpu(mpb
->mpb_size
) + space_needed
> super
->len
) {
2475 len
= ROUND_UP(__le32_to_cpu(mpb
->mpb_size
) + space_needed
, 512);
2476 if (posix_memalign(&buf
, 512, len
) != 0)
2479 memcpy(buf
, super
->buf
, super
->len
);
2480 memset(buf
+ super
->len
, 0, len
- super
->len
);
2489 /* retrieve a pointer to the bbm log which starts after all raid devices */
2490 struct bbm_log
*__get_imsm_bbm_log(struct imsm_super
*mpb
)
2494 if (__le32_to_cpu(mpb
->bbm_log_size
)) {
2496 ptr
+= mpb
->mpb_size
- __le32_to_cpu(mpb
->bbm_log_size
);
2502 static void __free_imsm(struct intel_super
*super
, int free_disks
);
2504 /* load_imsm_mpb - read matrix metadata
2505 * allocates super->mpb to be freed by free_super
2507 static int load_imsm_mpb(int fd
, struct intel_super
*super
, char *devname
)
2509 unsigned long long dsize
;
2510 unsigned long long sectors
;
2512 struct imsm_super
*anchor
;
2515 get_dev_size(fd
, NULL
, &dsize
);
2519 Name
": %s: device to small for imsm\n",
2524 if (lseek64(fd
, dsize
- (512 * 2), SEEK_SET
) < 0) {
2527 Name
": Cannot seek to anchor block on %s: %s\n",
2528 devname
, strerror(errno
));
2532 if (posix_memalign((void**)&anchor
, 512, 512) != 0) {
2535 Name
": Failed to allocate imsm anchor buffer"
2536 " on %s\n", devname
);
2539 if (read(fd
, anchor
, 512) != 512) {
2542 Name
": Cannot read anchor block on %s: %s\n",
2543 devname
, strerror(errno
));
2548 if (strncmp((char *) anchor
->sig
, MPB_SIGNATURE
, MPB_SIG_LEN
) != 0) {
2551 Name
": no IMSM anchor on %s\n", devname
);
2556 __free_imsm(super
, 0);
2557 super
->len
= ROUND_UP(anchor
->mpb_size
, 512);
2558 if (posix_memalign(&super
->buf
, 512, super
->len
) != 0) {
2561 Name
": unable to allocate %zu byte mpb buffer\n",
2566 memcpy(super
->buf
, anchor
, 512);
2568 sectors
= mpb_sectors(anchor
) - 1;
2571 check_sum
= __gen_imsm_checksum(super
->anchor
);
2572 if (check_sum
!= __le32_to_cpu(super
->anchor
->check_sum
)) {
2575 Name
": IMSM checksum %x != %x on %s\n",
2577 __le32_to_cpu(super
->anchor
->check_sum
),
2585 /* read the extended mpb */
2586 if (lseek64(fd
, dsize
- (512 * (2 + sectors
)), SEEK_SET
) < 0) {
2589 Name
": Cannot seek to extended mpb on %s: %s\n",
2590 devname
, strerror(errno
));
2594 if ((unsigned)read(fd
, super
->buf
+ 512, super
->len
- 512) != super
->len
- 512) {
2597 Name
": Cannot read extended mpb on %s: %s\n",
2598 devname
, strerror(errno
));
2602 check_sum
= __gen_imsm_checksum(super
->anchor
);
2603 if (check_sum
!= __le32_to_cpu(super
->anchor
->check_sum
)) {
2606 Name
": IMSM checksum %x != %x on %s\n",
2607 check_sum
, __le32_to_cpu(super
->anchor
->check_sum
),
2612 /* FIXME the BBM log is disk specific so we cannot use this global
2613 * buffer for all disks. Ok for now since we only look at the global
2614 * bbm_log_size parameter to gate assembly
2616 super
->bbm_log
= __get_imsm_bbm_log(super
->anchor
);
2622 load_and_parse_mpb(int fd
, struct intel_super
*super
, char *devname
, int keep_fd
)
2626 err
= load_imsm_mpb(fd
, super
, devname
);
2629 err
= load_imsm_disk(fd
, super
, devname
, keep_fd
);
2632 err
= parse_raid_devices(super
);
2637 static void __free_imsm_disk(struct dl
*d
)
2649 static void free_imsm_disks(struct intel_super
*super
)
2653 while (super
->disks
) {
2655 super
->disks
= d
->next
;
2656 __free_imsm_disk(d
);
2658 while (super
->missing
) {
2660 super
->missing
= d
->next
;
2661 __free_imsm_disk(d
);
2666 /* free all the pieces hanging off of a super pointer */
2667 static void __free_imsm(struct intel_super
*super
, int free_disks
)
2669 struct intel_hba
*elem
, *next
;
2676 free_imsm_disks(super
);
2677 free_devlist(super
);
2681 free((void *)elem
->path
);
2689 static void free_imsm(struct intel_super
*super
)
2691 __free_imsm(super
, 1);
2695 static void free_super_imsm(struct supertype
*st
)
2697 struct intel_super
*super
= st
->sb
;
2706 static struct intel_super
*alloc_super(void)
2708 struct intel_super
*super
= malloc(sizeof(*super
));
2711 memset(super
, 0, sizeof(*super
));
2712 super
->current_vol
= -1;
2713 super
->create_offset
= ~((__u32
) 0);
2714 if (!check_env("IMSM_NO_PLATFORM"))
2715 super
->orom
= find_imsm_orom();
2722 /* find_missing - helper routine for load_super_imsm_all that identifies
2723 * disks that have disappeared from the system. This routine relies on
2724 * the mpb being uptodate, which it is at load time.
2726 static int find_missing(struct intel_super
*super
)
2729 struct imsm_super
*mpb
= super
->anchor
;
2731 struct imsm_disk
*disk
;
2733 for (i
= 0; i
< mpb
->num_disks
; i
++) {
2734 disk
= __get_imsm_disk(mpb
, i
);
2735 dl
= serial_to_dl(disk
->serial
, super
);
2739 dl
= malloc(sizeof(*dl
));
2745 dl
->devname
= strdup("missing");
2747 serialcpy(dl
->serial
, disk
->serial
);
2750 dl
->next
= super
->missing
;
2751 super
->missing
= dl
;
2757 static struct intel_disk
*disk_list_get(__u8
*serial
, struct intel_disk
*disk_list
)
2759 struct intel_disk
*idisk
= disk_list
;
2762 if (serialcmp(idisk
->disk
.serial
, serial
) == 0)
2764 idisk
= idisk
->next
;
2770 static int __prep_thunderdome(struct intel_super
**table
, int tbl_size
,
2771 struct intel_super
*super
,
2772 struct intel_disk
**disk_list
)
2774 struct imsm_disk
*d
= &super
->disks
->disk
;
2775 struct imsm_super
*mpb
= super
->anchor
;
2778 for (i
= 0; i
< tbl_size
; i
++) {
2779 struct imsm_super
*tbl_mpb
= table
[i
]->anchor
;
2780 struct imsm_disk
*tbl_d
= &table
[i
]->disks
->disk
;
2782 if (tbl_mpb
->family_num
== mpb
->family_num
) {
2783 if (tbl_mpb
->check_sum
== mpb
->check_sum
) {
2784 dprintf("%s: mpb from %d:%d matches %d:%d\n",
2785 __func__
, super
->disks
->major
,
2786 super
->disks
->minor
,
2787 table
[i
]->disks
->major
,
2788 table
[i
]->disks
->minor
);
2792 if (((is_configured(d
) && !is_configured(tbl_d
)) ||
2793 is_configured(d
) == is_configured(tbl_d
)) &&
2794 tbl_mpb
->generation_num
< mpb
->generation_num
) {
2795 /* current version of the mpb is a
2796 * better candidate than the one in
2797 * super_table, but copy over "cross
2798 * generational" status
2800 struct intel_disk
*idisk
;
2802 dprintf("%s: mpb from %d:%d replaces %d:%d\n",
2803 __func__
, super
->disks
->major
,
2804 super
->disks
->minor
,
2805 table
[i
]->disks
->major
,
2806 table
[i
]->disks
->minor
);
2808 idisk
= disk_list_get(tbl_d
->serial
, *disk_list
);
2809 if (idisk
&& is_failed(&idisk
->disk
))
2810 tbl_d
->status
|= FAILED_DISK
;
2813 struct intel_disk
*idisk
;
2814 struct imsm_disk
*disk
;
2816 /* tbl_mpb is more up to date, but copy
2817 * over cross generational status before
2820 disk
= __serial_to_disk(d
->serial
, mpb
, NULL
);
2821 if (disk
&& is_failed(disk
))
2822 d
->status
|= FAILED_DISK
;
2824 idisk
= disk_list_get(d
->serial
, *disk_list
);
2827 if (disk
&& is_configured(disk
))
2828 idisk
->disk
.status
|= CONFIGURED_DISK
;
2831 dprintf("%s: mpb from %d:%d prefer %d:%d\n",
2832 __func__
, super
->disks
->major
,
2833 super
->disks
->minor
,
2834 table
[i
]->disks
->major
,
2835 table
[i
]->disks
->minor
);
2843 table
[tbl_size
++] = super
;
2847 /* update/extend the merged list of imsm_disk records */
2848 for (j
= 0; j
< mpb
->num_disks
; j
++) {
2849 struct imsm_disk
*disk
= __get_imsm_disk(mpb
, j
);
2850 struct intel_disk
*idisk
;
2852 idisk
= disk_list_get(disk
->serial
, *disk_list
);
2854 idisk
->disk
.status
|= disk
->status
;
2855 if (is_configured(&idisk
->disk
) ||
2856 is_failed(&idisk
->disk
))
2857 idisk
->disk
.status
&= ~(SPARE_DISK
);
2859 idisk
= calloc(1, sizeof(*idisk
));
2862 idisk
->owner
= IMSM_UNKNOWN_OWNER
;
2863 idisk
->disk
= *disk
;
2864 idisk
->next
= *disk_list
;
2868 if (serialcmp(idisk
->disk
.serial
, d
->serial
) == 0)
2875 static struct intel_super
*
2876 validate_members(struct intel_super
*super
, struct intel_disk
*disk_list
,
2879 struct imsm_super
*mpb
= super
->anchor
;
2883 for (i
= 0; i
< mpb
->num_disks
; i
++) {
2884 struct imsm_disk
*disk
= __get_imsm_disk(mpb
, i
);
2885 struct intel_disk
*idisk
;
2887 idisk
= disk_list_get(disk
->serial
, disk_list
);
2889 if (idisk
->owner
== owner
||
2890 idisk
->owner
== IMSM_UNKNOWN_OWNER
)
2893 dprintf("%s: '%.16s' owner %d != %d\n",
2894 __func__
, disk
->serial
, idisk
->owner
,
2897 dprintf("%s: unknown disk %x [%d]: %.16s\n",
2898 __func__
, __le32_to_cpu(mpb
->family_num
), i
,
2904 if (ok_count
== mpb
->num_disks
)
2909 static void show_conflicts(__u32 family_num
, struct intel_super
*super_list
)
2911 struct intel_super
*s
;
2913 for (s
= super_list
; s
; s
= s
->next
) {
2914 if (family_num
!= s
->anchor
->family_num
)
2916 fprintf(stderr
, "Conflict, offlining family %#x on '%s'\n",
2917 __le32_to_cpu(family_num
), s
->disks
->devname
);
2921 static struct intel_super
*
2922 imsm_thunderdome(struct intel_super
**super_list
, int len
)
2924 struct intel_super
*super_table
[len
];
2925 struct intel_disk
*disk_list
= NULL
;
2926 struct intel_super
*champion
, *spare
;
2927 struct intel_super
*s
, **del
;
2932 memset(super_table
, 0, sizeof(super_table
));
2933 for (s
= *super_list
; s
; s
= s
->next
)
2934 tbl_size
= __prep_thunderdome(super_table
, tbl_size
, s
, &disk_list
);
2936 for (i
= 0; i
< tbl_size
; i
++) {
2937 struct imsm_disk
*d
;
2938 struct intel_disk
*idisk
;
2939 struct imsm_super
*mpb
= super_table
[i
]->anchor
;
2942 d
= &s
->disks
->disk
;
2944 /* 'd' must appear in merged disk list for its
2945 * configuration to be valid
2947 idisk
= disk_list_get(d
->serial
, disk_list
);
2948 if (idisk
&& idisk
->owner
== i
)
2949 s
= validate_members(s
, disk_list
, i
);
2954 dprintf("%s: marking family: %#x from %d:%d offline\n",
2955 __func__
, mpb
->family_num
,
2956 super_table
[i
]->disks
->major
,
2957 super_table
[i
]->disks
->minor
);
2961 /* This is where the mdadm implementation differs from the Windows
2962 * driver which has no strict concept of a container. We can only
2963 * assemble one family from a container, so when returning a prodigal
2964 * array member to this system the code will not be able to disambiguate
2965 * the container contents that should be assembled ("foreign" versus
2966 * "local"). It requires user intervention to set the orig_family_num
2967 * to a new value to establish a new container. The Windows driver in
2968 * this situation fixes up the volume name in place and manages the
2969 * foreign array as an independent entity.
2974 for (i
= 0; i
< tbl_size
; i
++) {
2975 struct intel_super
*tbl_ent
= super_table
[i
];
2981 if (tbl_ent
->anchor
->num_raid_devs
== 0) {
2986 if (s
&& !is_spare
) {
2987 show_conflicts(tbl_ent
->anchor
->family_num
, *super_list
);
2989 } else if (!s
&& !is_spare
)
3002 fprintf(stderr
, "Chose family %#x on '%s', "
3003 "assemble conflicts to new container with '--update=uuid'\n",
3004 __le32_to_cpu(s
->anchor
->family_num
), s
->disks
->devname
);
3006 /* collect all dl's onto 'champion', and update them to
3007 * champion's version of the status
3009 for (s
= *super_list
; s
; s
= s
->next
) {
3010 struct imsm_super
*mpb
= champion
->anchor
;
3011 struct dl
*dl
= s
->disks
;
3016 for (i
= 0; i
< mpb
->num_disks
; i
++) {
3017 struct imsm_disk
*disk
;
3019 disk
= __serial_to_disk(dl
->serial
, mpb
, &dl
->index
);
3022 /* only set index on disks that are a member of
3023 * a populated contianer, i.e. one with
3026 if (is_failed(&dl
->disk
))
3028 else if (is_spare(&dl
->disk
))
3034 if (i
>= mpb
->num_disks
) {
3035 struct intel_disk
*idisk
;
3037 idisk
= disk_list_get(dl
->serial
, disk_list
);
3038 if (idisk
&& is_spare(&idisk
->disk
) &&
3039 !is_failed(&idisk
->disk
) && !is_configured(&idisk
->disk
))
3047 dl
->next
= champion
->disks
;
3048 champion
->disks
= dl
;
3052 /* delete 'champion' from super_list */
3053 for (del
= super_list
; *del
; ) {
3054 if (*del
== champion
) {
3055 *del
= (*del
)->next
;
3058 del
= &(*del
)->next
;
3060 champion
->next
= NULL
;
3064 struct intel_disk
*idisk
= disk_list
;
3066 disk_list
= disk_list
->next
;
3073 static int load_super_imsm_all(struct supertype
*st
, int fd
, void **sbp
,
3077 struct intel_super
*super_list
= NULL
;
3078 struct intel_super
*super
= NULL
;
3079 int devnum
= fd2devnum(fd
);
3085 /* check if 'fd' an opened container */
3086 sra
= sysfs_read(fd
, 0, GET_LEVEL
|GET_VERSION
|GET_DEVS
|GET_STATE
);
3090 if (sra
->array
.major_version
!= -1 ||
3091 sra
->array
.minor_version
!= -2 ||
3092 strcmp(sra
->text_version
, "imsm") != 0) {
3097 for (sd
= sra
->devs
, i
= 0; sd
; sd
= sd
->next
, i
++) {
3098 struct intel_super
*s
= alloc_super();
3105 s
->next
= super_list
;
3109 sprintf(nm
, "%d:%d", sd
->disk
.major
, sd
->disk
.minor
);
3110 dfd
= dev_open(nm
, O_RDWR
);
3114 err
= load_and_parse_mpb(dfd
, s
, NULL
, 1);
3116 /* retry the load if we might have raced against mdmon */
3117 if (err
== 3 && mdmon_running(devnum
))
3118 for (retry
= 0; retry
< 3; retry
++) {
3120 err
= load_and_parse_mpb(dfd
, s
, NULL
, 1);
3128 /* all mpbs enter, maybe one leaves */
3129 super
= imsm_thunderdome(&super_list
, i
);
3135 if (find_missing(super
) != 0) {
3143 while (super_list
) {
3144 struct intel_super
*s
= super_list
;
3146 super_list
= super_list
->next
;
3155 st
->container_dev
= devnum
;
3156 if (err
== 0 && st
->ss
== NULL
) {
3157 st
->ss
= &super_imsm
;
3158 st
->minor_version
= 0;
3159 st
->max_devs
= IMSM_MAX_DEVICES
;
3164 static int load_container_imsm(struct supertype
*st
, int fd
, char *devname
)
3166 return load_super_imsm_all(st
, fd
, &st
->sb
, devname
);
3170 static int load_super_imsm(struct supertype
*st
, int fd
, char *devname
)
3172 struct intel_super
*super
;
3175 if (test_partition(fd
))
3176 /* IMSM not allowed on partitions */
3179 free_super_imsm(st
);
3181 super
= alloc_super();
3184 Name
": malloc of %zu failed.\n",
3189 rv
= load_and_parse_mpb(fd
, super
, devname
, 0);
3194 Name
": Failed to load all information "
3195 "sections on %s\n", devname
);
3201 if (st
->ss
== NULL
) {
3202 st
->ss
= &super_imsm
;
3203 st
->minor_version
= 0;
3204 st
->max_devs
= IMSM_MAX_DEVICES
;
3209 static __u16
info_to_blocks_per_strip(mdu_array_info_t
*info
)
3211 if (info
->level
== 1)
3213 return info
->chunk_size
>> 9;
3216 static __u32
info_to_num_data_stripes(mdu_array_info_t
*info
, int num_domains
)
3220 num_stripes
= (info
->size
* 2) / info_to_blocks_per_strip(info
);
3221 num_stripes
/= num_domains
;
3226 static __u32
info_to_blocks_per_member(mdu_array_info_t
*info
)
3228 if (info
->level
== 1)
3229 return info
->size
* 2;
3231 return (info
->size
* 2) & ~(info_to_blocks_per_strip(info
) - 1);
3234 static void imsm_update_version_info(struct intel_super
*super
)
3236 /* update the version and attributes */
3237 struct imsm_super
*mpb
= super
->anchor
;
3239 struct imsm_dev
*dev
;
3240 struct imsm_map
*map
;
3243 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
3244 dev
= get_imsm_dev(super
, i
);
3245 map
= get_imsm_map(dev
, 0);
3246 if (__le32_to_cpu(dev
->size_high
) > 0)
3247 mpb
->attributes
|= MPB_ATTRIB_2TB
;
3249 /* FIXME detect when an array spans a port multiplier */
3251 mpb
->attributes
|= MPB_ATTRIB_PM
;
3254 if (mpb
->num_raid_devs
> 1 ||
3255 mpb
->attributes
!= MPB_ATTRIB_CHECKSUM_VERIFY
) {
3256 version
= MPB_VERSION_ATTRIBS
;
3257 switch (get_imsm_raid_level(map
)) {
3258 case 0: mpb
->attributes
|= MPB_ATTRIB_RAID0
; break;
3259 case 1: mpb
->attributes
|= MPB_ATTRIB_RAID1
; break;
3260 case 10: mpb
->attributes
|= MPB_ATTRIB_RAID10
; break;
3261 case 5: mpb
->attributes
|= MPB_ATTRIB_RAID5
; break;
3264 if (map
->num_members
>= 5)
3265 version
= MPB_VERSION_5OR6_DISK_ARRAY
;
3266 else if (dev
->status
== DEV_CLONE_N_GO
)
3267 version
= MPB_VERSION_CNG
;
3268 else if (get_imsm_raid_level(map
) == 5)
3269 version
= MPB_VERSION_RAID5
;
3270 else if (map
->num_members
>= 3)
3271 version
= MPB_VERSION_3OR4_DISK_ARRAY
;
3272 else if (get_imsm_raid_level(map
) == 1)
3273 version
= MPB_VERSION_RAID1
;
3275 version
= MPB_VERSION_RAID0
;
3277 strcpy(((char *) mpb
->sig
) + strlen(MPB_SIGNATURE
), version
);
3281 static int check_name(struct intel_super
*super
, char *name
, int quiet
)
3283 struct imsm_super
*mpb
= super
->anchor
;
3284 char *reason
= NULL
;
3287 if (strlen(name
) > MAX_RAID_SERIAL_LEN
)
3288 reason
= "must be 16 characters or less";
3290 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
3291 struct imsm_dev
*dev
= get_imsm_dev(super
, i
);
3293 if (strncmp((char *) dev
->volume
, name
, MAX_RAID_SERIAL_LEN
) == 0) {
3294 reason
= "already exists";
3299 if (reason
&& !quiet
)
3300 fprintf(stderr
, Name
": imsm volume name %s\n", reason
);
3305 static int init_super_imsm_volume(struct supertype
*st
, mdu_array_info_t
*info
,
3306 unsigned long long size
, char *name
,
3307 char *homehost
, int *uuid
)
3309 /* We are creating a volume inside a pre-existing container.
3310 * so st->sb is already set.
3312 struct intel_super
*super
= st
->sb
;
3313 struct imsm_super
*mpb
= super
->anchor
;
3314 struct intel_dev
*dv
;
3315 struct imsm_dev
*dev
;
3316 struct imsm_vol
*vol
;
3317 struct imsm_map
*map
;
3318 int idx
= mpb
->num_raid_devs
;
3320 unsigned long long array_blocks
;
3321 size_t size_old
, size_new
;
3322 __u32 num_data_stripes
;
3324 if (super
->orom
&& mpb
->num_raid_devs
>= super
->orom
->vpa
) {
3325 fprintf(stderr
, Name
": This imsm-container already has the "
3326 "maximum of %d volumes\n", super
->orom
->vpa
);
3330 /* ensure the mpb is large enough for the new data */
3331 size_old
= __le32_to_cpu(mpb
->mpb_size
);
3332 size_new
= disks_to_mpb_size(info
->nr_disks
);
3333 if (size_new
> size_old
) {
3335 size_t size_round
= ROUND_UP(size_new
, 512);
3337 if (posix_memalign(&mpb_new
, 512, size_round
) != 0) {
3338 fprintf(stderr
, Name
": could not allocate new mpb\n");
3341 memcpy(mpb_new
, mpb
, size_old
);
3344 super
->anchor
= mpb_new
;
3345 mpb
->mpb_size
= __cpu_to_le32(size_new
);
3346 memset(mpb_new
+ size_old
, 0, size_round
- size_old
);
3348 super
->current_vol
= idx
;
3349 /* when creating the first raid device in this container set num_disks
3350 * to zero, i.e. delete this spare and add raid member devices in
3351 * add_to_super_imsm_volume()
3353 if (super
->current_vol
== 0)
3356 if (!check_name(super
, name
, 0))
3358 dv
= malloc(sizeof(*dv
));
3360 fprintf(stderr
, Name
": failed to allocate device list entry\n");
3363 dev
= malloc(sizeof(*dev
) + sizeof(__u32
) * (info
->raid_disks
- 1));
3366 fprintf(stderr
, Name
": could not allocate raid device\n");
3369 strncpy((char *) dev
->volume
, name
, MAX_RAID_SERIAL_LEN
);
3370 if (info
->level
== 1)
3371 array_blocks
= info_to_blocks_per_member(info
);
3373 array_blocks
= calc_array_size(info
->level
, info
->raid_disks
,
3374 info
->layout
, info
->chunk_size
,
3376 /* round array size down to closest MB */
3377 array_blocks
= (array_blocks
>> SECT_PER_MB_SHIFT
) << SECT_PER_MB_SHIFT
;
3379 dev
->size_low
= __cpu_to_le32((__u32
) array_blocks
);
3380 dev
->size_high
= __cpu_to_le32((__u32
) (array_blocks
>> 32));
3381 dev
->status
= __cpu_to_le32(0);
3382 dev
->reserved_blocks
= __cpu_to_le32(0);
3384 vol
->migr_state
= 0;
3385 set_migr_type(dev
, MIGR_INIT
);
3387 vol
->curr_migr_unit
= 0;
3388 map
= get_imsm_map(dev
, 0);
3389 map
->pba_of_lba0
= __cpu_to_le32(super
->create_offset
);
3390 map
->blocks_per_member
= __cpu_to_le32(info_to_blocks_per_member(info
));
3391 map
->blocks_per_strip
= __cpu_to_le16(info_to_blocks_per_strip(info
));
3392 map
->failed_disk_num
= ~0;
3393 map
->map_state
= info
->level
? IMSM_T_STATE_UNINITIALIZED
:
3394 IMSM_T_STATE_NORMAL
;
3397 if (info
->level
== 1 && info
->raid_disks
> 2) {
3400 fprintf(stderr
, Name
": imsm does not support more than 2 disks"
3401 "in a raid1 volume\n");
3405 map
->raid_level
= info
->level
;
3406 if (info
->level
== 10) {
3407 map
->raid_level
= 1;
3408 map
->num_domains
= info
->raid_disks
/ 2;
3409 } else if (info
->level
== 1)
3410 map
->num_domains
= info
->raid_disks
;
3412 map
->num_domains
= 1;
3414 num_data_stripes
= info_to_num_data_stripes(info
, map
->num_domains
);
3415 map
->num_data_stripes
= __cpu_to_le32(num_data_stripes
);
3417 map
->num_members
= info
->raid_disks
;
3418 for (i
= 0; i
< map
->num_members
; i
++) {
3419 /* initialized in add_to_super */
3420 set_imsm_ord_tbl_ent(map
, i
, IMSM_ORD_REBUILD
);
3422 mpb
->num_raid_devs
++;
3425 dv
->index
= super
->current_vol
;
3426 dv
->next
= super
->devlist
;
3427 super
->devlist
= dv
;
3429 imsm_update_version_info(super
);
3434 static int init_super_imsm(struct supertype
*st
, mdu_array_info_t
*info
,
3435 unsigned long long size
, char *name
,
3436 char *homehost
, int *uuid
)
3438 /* This is primarily called by Create when creating a new array.
3439 * We will then get add_to_super called for each component, and then
3440 * write_init_super called to write it out to each device.
3441 * For IMSM, Create can create on fresh devices or on a pre-existing
3443 * To create on a pre-existing array a different method will be called.
3444 * This one is just for fresh drives.
3446 struct intel_super
*super
;
3447 struct imsm_super
*mpb
;
3452 return init_super_imsm_volume(st
, info
, size
, name
, homehost
, uuid
);
3455 mpb_size
= disks_to_mpb_size(info
->nr_disks
);
3459 super
= alloc_super();
3460 if (super
&& posix_memalign(&super
->buf
, 512, mpb_size
) != 0) {
3465 fprintf(stderr
, Name
3466 ": %s could not allocate superblock\n", __func__
);
3469 memset(super
->buf
, 0, mpb_size
);
3471 mpb
->mpb_size
= __cpu_to_le32(mpb_size
);
3475 /* zeroing superblock */
3479 mpb
->attributes
= MPB_ATTRIB_CHECKSUM_VERIFY
;
3481 version
= (char *) mpb
->sig
;
3482 strcpy(version
, MPB_SIGNATURE
);
3483 version
+= strlen(MPB_SIGNATURE
);
3484 strcpy(version
, MPB_VERSION_RAID0
);
3490 static int add_to_super_imsm_volume(struct supertype
*st
, mdu_disk_info_t
*dk
,
3491 int fd
, char *devname
)
3493 struct intel_super
*super
= st
->sb
;
3494 struct imsm_super
*mpb
= super
->anchor
;
3496 struct imsm_dev
*dev
;
3497 struct imsm_map
*map
;
3500 dev
= get_imsm_dev(super
, super
->current_vol
);
3501 map
= get_imsm_map(dev
, 0);
3503 if (! (dk
->state
& (1<<MD_DISK_SYNC
))) {
3504 fprintf(stderr
, Name
": %s: Cannot add spare devices to IMSM volume\n",
3510 /* we're doing autolayout so grab the pre-marked (in
3511 * validate_geometry) raid_disk
3513 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
3514 if (dl
->raiddisk
== dk
->raid_disk
)
3517 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
3518 if (dl
->major
== dk
->major
&&
3519 dl
->minor
== dk
->minor
)
3524 fprintf(stderr
, Name
": %s is not a member of the same container\n", devname
);
3528 /* add a pristine spare to the metadata */
3529 if (dl
->index
< 0) {
3530 dl
->index
= super
->anchor
->num_disks
;
3531 super
->anchor
->num_disks
++;
3533 /* Check the device has not already been added */
3534 slot
= get_imsm_disk_slot(map
, dl
->index
);
3536 (get_imsm_ord_tbl_ent(dev
, slot
, -1) & IMSM_ORD_REBUILD
) == 0) {
3537 fprintf(stderr
, Name
": %s has been included in this array twice\n",
3541 set_imsm_ord_tbl_ent(map
, dk
->number
, dl
->index
);
3542 dl
->disk
.status
= CONFIGURED_DISK
;
3544 /* if we are creating the first raid device update the family number */
3545 if (super
->current_vol
== 0) {
3547 struct imsm_dev
*_dev
= __get_imsm_dev(mpb
, 0);
3548 struct imsm_disk
*_disk
= __get_imsm_disk(mpb
, dl
->index
);
3550 if (!_dev
|| !_disk
) {
3551 fprintf(stderr
, Name
": BUG mpb setup error\n");
3557 sum
+= __gen_imsm_checksum(mpb
);
3558 mpb
->family_num
= __cpu_to_le32(sum
);
3559 mpb
->orig_family_num
= mpb
->family_num
;
3566 static int add_to_super_imsm(struct supertype
*st
, mdu_disk_info_t
*dk
,
3567 int fd
, char *devname
)
3569 struct intel_super
*super
= st
->sb
;
3571 unsigned long long size
;
3576 /* If we are on an RAID enabled platform check that the disk is
3577 * attached to the raid controller.
3578 * We do not need to test disks attachment for container based additions,
3579 * they shall be already tested when container was created/assembled.
3581 if ((fd
!= -1) && !check_env("IMSM_NO_PLATFORM")) {
3582 struct sys_dev
*hba_name
;
3583 struct intel_hba
*hba
;
3585 hba_name
= find_disk_attached_hba(fd
, NULL
);
3588 Name
": %s is not attached to Intel(R) RAID controller.\n",
3589 devname
? : "disk");
3592 rv
= attach_hba_to_super(super
, hba_name
, devname
);
3595 fprintf(stderr
, Name
": %s is attached to Intel(R) %s RAID "
3596 "controller (%s),\n but the container is assigned to Intel(R) "
3597 "%s RAID controller (",
3599 get_sys_dev_type(hba_name
->type
),
3600 hba_name
->pci_id
? : "Err!",
3601 get_sys_dev_type(hba_name
->type
));
3605 fprintf(stderr
, "%s", hba
->pci_id
? : "Err!");
3607 fprintf(stderr
, ", ");
3611 fprintf(stderr
, ").\n"
3612 " Mixing devices attached to different controllers "
3613 "is not allowed.\n");
3614 free_sys_dev(&hba_name
);
3617 free_sys_dev(&hba_name
);
3620 if (super
->current_vol
>= 0)
3621 return add_to_super_imsm_volume(st
, dk
, fd
, devname
);
3624 dd
= malloc(sizeof(*dd
));
3627 Name
": malloc failed %s:%d.\n", __func__
, __LINE__
);
3630 memset(dd
, 0, sizeof(*dd
));
3631 dd
->major
= major(stb
.st_rdev
);
3632 dd
->minor
= minor(stb
.st_rdev
);
3634 dd
->devname
= devname
? strdup(devname
) : NULL
;
3637 dd
->action
= DISK_ADD
;
3638 rv
= imsm_read_serial(fd
, devname
, dd
->serial
);
3641 Name
": failed to retrieve scsi serial, aborting\n");
3646 get_dev_size(fd
, NULL
, &size
);
3648 serialcpy(dd
->disk
.serial
, dd
->serial
);
3649 dd
->disk
.total_blocks
= __cpu_to_le32(size
);
3650 dd
->disk
.status
= SPARE_DISK
;
3651 if (sysfs_disk_to_scsi_id(fd
, &id
) == 0)
3652 dd
->disk
.scsi_id
= __cpu_to_le32(id
);
3654 dd
->disk
.scsi_id
= __cpu_to_le32(0);
3656 if (st
->update_tail
) {
3657 dd
->next
= super
->disk_mgmt_list
;
3658 super
->disk_mgmt_list
= dd
;
3660 dd
->next
= super
->disks
;
3668 static int remove_from_super_imsm(struct supertype
*st
, mdu_disk_info_t
*dk
)
3670 struct intel_super
*super
= st
->sb
;
3673 /* remove from super works only in mdmon - for communication
3674 * manager - monitor. Check if communication memory buffer
3677 if (!st
->update_tail
) {
3679 Name
": %s shall be used in mdmon context only"
3680 "(line %d).\n", __func__
, __LINE__
);
3683 dd
= malloc(sizeof(*dd
));
3686 Name
": malloc failed %s:%d.\n", __func__
, __LINE__
);
3689 memset(dd
, 0, sizeof(*dd
));
3690 dd
->major
= dk
->major
;
3691 dd
->minor
= dk
->minor
;
3694 dd
->disk
.status
= SPARE_DISK
;
3695 dd
->action
= DISK_REMOVE
;
3697 dd
->next
= super
->disk_mgmt_list
;
3698 super
->disk_mgmt_list
= dd
;
3704 static int store_imsm_mpb(int fd
, struct imsm_super
*mpb
);
3708 struct imsm_super anchor
;
3709 } spare_record
__attribute__ ((aligned(512)));
3711 /* spare records have their own family number and do not have any defined raid
3714 static int write_super_imsm_spares(struct intel_super
*super
, int doclose
)
3716 struct imsm_super
*mpb
= super
->anchor
;
3717 struct imsm_super
*spare
= &spare_record
.anchor
;
3721 spare
->mpb_size
= __cpu_to_le32(sizeof(struct imsm_super
)),
3722 spare
->generation_num
= __cpu_to_le32(1UL),
3723 spare
->attributes
= MPB_ATTRIB_CHECKSUM_VERIFY
;
3724 spare
->num_disks
= 1,
3725 spare
->num_raid_devs
= 0,
3726 spare
->cache_size
= mpb
->cache_size
,
3727 spare
->pwr_cycle_count
= __cpu_to_le32(1),
3729 snprintf((char *) spare
->sig
, MAX_SIGNATURE_LENGTH
,
3730 MPB_SIGNATURE MPB_VERSION_RAID0
);
3732 for (d
= super
->disks
; d
; d
= d
->next
) {
3736 spare
->disk
[0] = d
->disk
;
3737 sum
= __gen_imsm_checksum(spare
);
3738 spare
->family_num
= __cpu_to_le32(sum
);
3739 spare
->orig_family_num
= 0;
3740 sum
= __gen_imsm_checksum(spare
);
3741 spare
->check_sum
= __cpu_to_le32(sum
);
3743 if (store_imsm_mpb(d
->fd
, spare
)) {
3744 fprintf(stderr
, "%s: failed for device %d:%d %s\n",
3745 __func__
, d
->major
, d
->minor
, strerror(errno
));
3757 static int write_super_imsm(struct supertype
*st
, int doclose
)
3759 struct intel_super
*super
= st
->sb
;
3760 struct imsm_super
*mpb
= super
->anchor
;
3766 __u32 mpb_size
= sizeof(struct imsm_super
) - sizeof(struct imsm_disk
);
3769 /* 'generation' is incremented everytime the metadata is written */
3770 generation
= __le32_to_cpu(mpb
->generation_num
);
3772 mpb
->generation_num
= __cpu_to_le32(generation
);
3774 /* fix up cases where previous mdadm releases failed to set
3777 if (mpb
->orig_family_num
== 0)
3778 mpb
->orig_family_num
= mpb
->family_num
;
3780 for (d
= super
->disks
; d
; d
= d
->next
) {
3784 mpb
->disk
[d
->index
] = d
->disk
;
3788 for (d
= super
->missing
; d
; d
= d
->next
) {
3789 mpb
->disk
[d
->index
] = d
->disk
;
3792 mpb
->num_disks
= num_disks
;
3793 mpb_size
+= sizeof(struct imsm_disk
) * mpb
->num_disks
;
3795 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
3796 struct imsm_dev
*dev
= __get_imsm_dev(mpb
, i
);
3797 struct imsm_dev
*dev2
= get_imsm_dev(super
, i
);
3799 imsm_copy_dev(dev
, dev2
);
3800 mpb_size
+= sizeof_imsm_dev(dev
, 0);
3803 mpb_size
+= __le32_to_cpu(mpb
->bbm_log_size
);
3804 mpb
->mpb_size
= __cpu_to_le32(mpb_size
);
3806 /* recalculate checksum */
3807 sum
= __gen_imsm_checksum(mpb
);
3808 mpb
->check_sum
= __cpu_to_le32(sum
);
3810 /* write the mpb for disks that compose raid devices */
3811 for (d
= super
->disks
; d
; d
= d
->next
) {
3814 if (store_imsm_mpb(d
->fd
, mpb
))
3815 fprintf(stderr
, "%s: failed for device %d:%d %s\n",
3816 __func__
, d
->major
, d
->minor
, strerror(errno
));
3824 return write_super_imsm_spares(super
, doclose
);
3830 static int create_array(struct supertype
*st
, int dev_idx
)
3833 struct imsm_update_create_array
*u
;
3834 struct intel_super
*super
= st
->sb
;
3835 struct imsm_dev
*dev
= get_imsm_dev(super
, dev_idx
);
3836 struct imsm_map
*map
= get_imsm_map(dev
, 0);
3837 struct disk_info
*inf
;
3838 struct imsm_disk
*disk
;
3841 len
= sizeof(*u
) - sizeof(*dev
) + sizeof_imsm_dev(dev
, 0) +
3842 sizeof(*inf
) * map
->num_members
;
3845 fprintf(stderr
, "%s: failed to allocate update buffer\n",
3850 u
->type
= update_create_array
;
3851 u
->dev_idx
= dev_idx
;
3852 imsm_copy_dev(&u
->dev
, dev
);
3853 inf
= get_disk_info(u
);
3854 for (i
= 0; i
< map
->num_members
; i
++) {
3855 int idx
= get_imsm_disk_idx(dev
, i
, -1);
3857 disk
= get_imsm_disk(super
, idx
);
3858 serialcpy(inf
[i
].serial
, disk
->serial
);
3860 append_metadata_update(st
, u
, len
);
3865 static int mgmt_disk(struct supertype
*st
)
3867 struct intel_super
*super
= st
->sb
;
3869 struct imsm_update_add_remove_disk
*u
;
3871 if (!super
->disk_mgmt_list
)
3877 fprintf(stderr
, "%s: failed to allocate update buffer\n",
3882 u
->type
= update_add_remove_disk
;
3883 append_metadata_update(st
, u
, len
);
3888 static int write_init_super_imsm(struct supertype
*st
)
3890 struct intel_super
*super
= st
->sb
;
3891 int current_vol
= super
->current_vol
;
3893 /* we are done with current_vol reset it to point st at the container */
3894 super
->current_vol
= -1;
3896 if (st
->update_tail
) {
3897 /* queue the recently created array / added disk
3898 * as a metadata update */
3901 /* determine if we are creating a volume or adding a disk */
3902 if (current_vol
< 0) {
3903 /* in the mgmt (add/remove) disk case we are running
3904 * in mdmon context, so don't close fd's
3906 return mgmt_disk(st
);
3908 rv
= create_array(st
, current_vol
);
3913 for (d
= super
->disks
; d
; d
= d
->next
)
3914 Kill(d
->devname
, NULL
, 0, 1, 1);
3915 return write_super_imsm(st
, 1);
3920 static int store_super_imsm(struct supertype
*st
, int fd
)
3922 struct intel_super
*super
= st
->sb
;
3923 struct imsm_super
*mpb
= super
? super
->anchor
: NULL
;
3929 return store_imsm_mpb(fd
, mpb
);
3935 static int imsm_bbm_log_size(struct imsm_super
*mpb
)
3937 return __le32_to_cpu(mpb
->bbm_log_size
);
3941 static int validate_geometry_imsm_container(struct supertype
*st
, int level
,
3942 int layout
, int raiddisks
, int chunk
,
3943 unsigned long long size
, char *dev
,
3944 unsigned long long *freesize
,
3948 unsigned long long ldsize
;
3949 const struct imsm_orom
*orom
;
3951 if (level
!= LEVEL_CONTAINER
)
3956 if (check_env("IMSM_NO_PLATFORM"))
3959 orom
= find_imsm_orom();
3960 if (orom
&& raiddisks
> orom
->tds
) {
3962 fprintf(stderr
, Name
": %d exceeds maximum number of"
3963 " platform supported disks: %d\n",
3964 raiddisks
, orom
->tds
);
3968 fd
= open(dev
, O_RDONLY
|O_EXCL
, 0);
3971 fprintf(stderr
, Name
": imsm: Cannot open %s: %s\n",
3972 dev
, strerror(errno
));
3975 if (!get_dev_size(fd
, dev
, &ldsize
)) {
3981 *freesize
= avail_size_imsm(st
, ldsize
>> 9);
3986 static unsigned long long find_size(struct extent
*e
, int *idx
, int num_extents
)
3988 const unsigned long long base_start
= e
[*idx
].start
;
3989 unsigned long long end
= base_start
+ e
[*idx
].size
;
3992 if (base_start
== end
)
3996 for (i
= *idx
; i
< num_extents
; i
++) {
3997 /* extend overlapping extents */
3998 if (e
[i
].start
>= base_start
&&
3999 e
[i
].start
<= end
) {
4002 if (e
[i
].start
+ e
[i
].size
> end
)
4003 end
= e
[i
].start
+ e
[i
].size
;
4004 } else if (e
[i
].start
> end
) {
4010 return end
- base_start
;
4013 static unsigned long long merge_extents(struct intel_super
*super
, int sum_extents
)
4015 /* build a composite disk with all known extents and generate a new
4016 * 'maxsize' given the "all disks in an array must share a common start
4017 * offset" constraint
4019 struct extent
*e
= calloc(sum_extents
, sizeof(*e
));
4023 unsigned long long pos
;
4024 unsigned long long start
= 0;
4025 unsigned long long maxsize
;
4026 unsigned long reserve
;
4031 /* coalesce and sort all extents. also, check to see if we need to
4032 * reserve space between member arrays
4035 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
4038 for (i
= 0; i
< dl
->extent_cnt
; i
++)
4041 qsort(e
, sum_extents
, sizeof(*e
), cmp_extent
);
4046 while (i
< sum_extents
) {
4047 e
[j
].start
= e
[i
].start
;
4048 e
[j
].size
= find_size(e
, &i
, sum_extents
);
4050 if (e
[j
-1].size
== 0)
4059 unsigned long long esize
;
4061 esize
= e
[i
].start
- pos
;
4062 if (esize
>= maxsize
) {
4067 pos
= e
[i
].start
+ e
[i
].size
;
4069 } while (e
[i
-1].size
);
4075 /* FIXME assumes volume at offset 0 is the first volume in a
4078 if (start_extent
> 0)
4079 reserve
= IMSM_RESERVED_SECTORS
; /* gap between raid regions */
4083 if (maxsize
< reserve
)
4086 super
->create_offset
= ~((__u32
) 0);
4087 if (start
+ reserve
> super
->create_offset
)
4088 return 0; /* start overflows create_offset */
4089 super
->create_offset
= start
+ reserve
;
4091 return maxsize
- reserve
;
4094 static int is_raid_level_supported(const struct imsm_orom
*orom
, int level
, int raiddisks
)
4096 if (level
< 0 || level
== 6 || level
== 4)
4099 /* if we have an orom prevent invalid raid levels */
4102 case 0: return imsm_orom_has_raid0(orom
);
4105 return imsm_orom_has_raid1e(orom
);
4106 return imsm_orom_has_raid1(orom
) && raiddisks
== 2;
4107 case 10: return imsm_orom_has_raid10(orom
) && raiddisks
== 4;
4108 case 5: return imsm_orom_has_raid5(orom
) && raiddisks
> 2;
4111 return 1; /* not on an Intel RAID platform so anything goes */
4116 #define pr_vrb(fmt, arg...) (void) (verbose && fprintf(stderr, Name fmt, ##arg))
4118 validate_geometry_imsm_orom(struct intel_super
*super
, int level
, int layout
,
4119 int raiddisks
, int chunk
, int verbose
)
4121 if (!is_raid_level_supported(super
->orom
, level
, raiddisks
)) {
4122 pr_vrb(": platform does not support raid%d with %d disk%s\n",
4123 level
, raiddisks
, raiddisks
> 1 ? "s" : "");
4126 if (super
->orom
&& level
!= 1 &&
4127 !imsm_orom_has_chunk(super
->orom
, chunk
)) {
4128 pr_vrb(": platform does not support a chunk size of: %d\n", chunk
);
4131 if (layout
!= imsm_level_to_layout(level
)) {
4133 pr_vrb(": imsm raid 5 only supports the left-asymmetric layout\n");
4134 else if (level
== 10)
4135 pr_vrb(": imsm raid 10 only supports the n2 layout\n");
4137 pr_vrb(": imsm unknown layout %#x for this raid level %d\n",
4145 /* validate_geometry_imsm_volume - lifted from validate_geometry_ddf_bvd
4146 * FIX ME add ahci details
4148 static int validate_geometry_imsm_volume(struct supertype
*st
, int level
,
4149 int layout
, int raiddisks
, int chunk
,
4150 unsigned long long size
, char *dev
,
4151 unsigned long long *freesize
,
4155 struct intel_super
*super
= st
->sb
;
4156 struct imsm_super
*mpb
= super
->anchor
;
4158 unsigned long long pos
= 0;
4159 unsigned long long maxsize
;
4163 /* We must have the container info already read in. */
4167 if (!validate_geometry_imsm_orom(super
, level
, layout
, raiddisks
, chunk
, verbose
))
4171 /* General test: make sure there is space for
4172 * 'raiddisks' device extents of size 'size' at a given
4175 unsigned long long minsize
= size
;
4176 unsigned long long start_offset
= MaxSector
;
4179 minsize
= MPB_SECTOR_CNT
+ IMSM_RESERVED_SECTORS
;
4180 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
4185 e
= get_extents(super
, dl
);
4188 unsigned long long esize
;
4189 esize
= e
[i
].start
- pos
;
4190 if (esize
>= minsize
)
4192 if (found
&& start_offset
== MaxSector
) {
4195 } else if (found
&& pos
!= start_offset
) {
4199 pos
= e
[i
].start
+ e
[i
].size
;
4201 } while (e
[i
-1].size
);
4206 if (dcnt
< raiddisks
) {
4208 fprintf(stderr
, Name
": imsm: Not enough "
4209 "devices with space for this array "
4217 /* This device must be a member of the set */
4218 if (stat(dev
, &stb
) < 0)
4220 if ((S_IFMT
& stb
.st_mode
) != S_IFBLK
)
4222 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
4223 if (dl
->major
== (int)major(stb
.st_rdev
) &&
4224 dl
->minor
== (int)minor(stb
.st_rdev
))
4229 fprintf(stderr
, Name
": %s is not in the "
4230 "same imsm set\n", dev
);
4232 } else if (super
->orom
&& dl
->index
< 0 && mpb
->num_raid_devs
) {
4233 /* If a volume is present then the current creation attempt
4234 * cannot incorporate new spares because the orom may not
4235 * understand this configuration (all member disks must be
4236 * members of each array in the container).
4238 fprintf(stderr
, Name
": %s is a spare and a volume"
4239 " is already defined for this container\n", dev
);
4240 fprintf(stderr
, Name
": The option-rom requires all member"
4241 " disks to be a member of all volumes\n");
4245 /* retrieve the largest free space block */
4246 e
= get_extents(super
, dl
);
4251 unsigned long long esize
;
4253 esize
= e
[i
].start
- pos
;
4254 if (esize
>= maxsize
)
4256 pos
= e
[i
].start
+ e
[i
].size
;
4258 } while (e
[i
-1].size
);
4263 fprintf(stderr
, Name
": unable to determine free space for: %s\n",
4267 if (maxsize
< size
) {
4269 fprintf(stderr
, Name
": %s not enough space (%llu < %llu)\n",
4270 dev
, maxsize
, size
);
4274 /* count total number of extents for merge */
4276 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
4278 i
+= dl
->extent_cnt
;
4280 maxsize
= merge_extents(super
, i
);
4281 if (maxsize
< size
|| maxsize
== 0) {
4283 fprintf(stderr
, Name
": not enough space after merge (%llu < %llu)\n",
4288 *freesize
= maxsize
;
4293 static int reserve_space(struct supertype
*st
, int raiddisks
,
4294 unsigned long long size
, int chunk
,
4295 unsigned long long *freesize
)
4297 struct intel_super
*super
= st
->sb
;
4298 struct imsm_super
*mpb
= super
->anchor
;
4303 unsigned long long maxsize
;
4304 unsigned long long minsize
;
4308 /* find the largest common start free region of the possible disks */
4312 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
4318 /* don't activate new spares if we are orom constrained
4319 * and there is already a volume active in the container
4321 if (super
->orom
&& dl
->index
< 0 && mpb
->num_raid_devs
)
4324 e
= get_extents(super
, dl
);
4327 for (i
= 1; e
[i
-1].size
; i
++)
4335 maxsize
= merge_extents(super
, extent_cnt
);
4340 if (cnt
< raiddisks
||
4341 (super
->orom
&& used
&& used
!= raiddisks
) ||
4342 maxsize
< minsize
||
4344 fprintf(stderr
, Name
": not enough devices with space to create array.\n");
4345 return 0; /* No enough free spaces large enough */
4357 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
4359 dl
->raiddisk
= cnt
++;
4366 static int validate_geometry_imsm(struct supertype
*st
, int level
, int layout
,
4367 int raiddisks
, int chunk
, unsigned long long size
,
4368 char *dev
, unsigned long long *freesize
,
4375 /* if given unused devices create a container
4376 * if given given devices in a container create a member volume
4378 if (level
== LEVEL_CONTAINER
) {
4379 /* Must be a fresh device to add to a container */
4380 return validate_geometry_imsm_container(st
, level
, layout
,
4381 raiddisks
, chunk
, size
,
4387 if (st
->sb
&& freesize
) {
4388 /* we are being asked to automatically layout a
4389 * new volume based on the current contents of
4390 * the container. If the the parameters can be
4391 * satisfied reserve_space will record the disks,
4392 * start offset, and size of the volume to be
4393 * created. add_to_super and getinfo_super
4394 * detect when autolayout is in progress.
4396 if (!validate_geometry_imsm_orom(st
->sb
, level
, layout
,
4400 return reserve_space(st
, raiddisks
, size
, chunk
, freesize
);
4405 /* creating in a given container */
4406 return validate_geometry_imsm_volume(st
, level
, layout
,
4407 raiddisks
, chunk
, size
,
4408 dev
, freesize
, verbose
);
4411 /* This device needs to be a device in an 'imsm' container */
4412 fd
= open(dev
, O_RDONLY
|O_EXCL
, 0);
4416 Name
": Cannot create this array on device %s\n",
4421 if (errno
!= EBUSY
|| (fd
= open(dev
, O_RDONLY
, 0)) < 0) {
4423 fprintf(stderr
, Name
": Cannot open %s: %s\n",
4424 dev
, strerror(errno
));
4427 /* Well, it is in use by someone, maybe an 'imsm' container. */
4428 cfd
= open_container(fd
);
4432 fprintf(stderr
, Name
": Cannot use %s: It is busy\n",
4436 sra
= sysfs_read(cfd
, 0, GET_VERSION
);
4437 if (sra
&& sra
->array
.major_version
== -1 &&
4438 strcmp(sra
->text_version
, "imsm") == 0)
4442 /* This is a member of a imsm container. Load the container
4443 * and try to create a volume
4445 struct intel_super
*super
;
4447 if (load_super_imsm_all(st
, cfd
, (void **) &super
, NULL
) == 0) {
4449 st
->container_dev
= fd2devnum(cfd
);
4451 return validate_geometry_imsm_volume(st
, level
, layout
,
4459 fprintf(stderr
, Name
": failed container membership check\n");
4465 static void default_geometry_imsm(struct supertype
*st
, int *level
, int *layout
, int *chunk
)
4467 struct intel_super
*super
= st
->sb
;
4469 if (level
&& *level
== UnSet
)
4470 *level
= LEVEL_CONTAINER
;
4472 if (level
&& layout
&& *layout
== UnSet
)
4473 *layout
= imsm_level_to_layout(*level
);
4475 if (chunk
&& (*chunk
== UnSet
|| *chunk
== 0) &&
4476 super
&& super
->orom
)
4477 *chunk
= imsm_orom_default_chunk(super
->orom
);
4480 static void handle_missing(struct intel_super
*super
, struct imsm_dev
*dev
);
4482 static int kill_subarray_imsm(struct supertype
*st
)
4484 /* remove the subarray currently referenced by ->current_vol */
4486 struct intel_dev
**dp
;
4487 struct intel_super
*super
= st
->sb
;
4488 __u8 current_vol
= super
->current_vol
;
4489 struct imsm_super
*mpb
= super
->anchor
;
4491 if (super
->current_vol
< 0)
4493 super
->current_vol
= -1; /* invalidate subarray cursor */
4495 /* block deletions that would change the uuid of active subarrays
4497 * FIXME when immutable ids are available, but note that we'll
4498 * also need to fixup the invalidated/active subarray indexes in
4501 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
4504 if (i
< current_vol
)
4506 sprintf(subarray
, "%u", i
);
4507 if (is_subarray_active(subarray
, st
->devname
)) {
4509 Name
": deleting subarray-%d would change the UUID of active subarray-%d, aborting\n",
4516 if (st
->update_tail
) {
4517 struct imsm_update_kill_array
*u
= malloc(sizeof(*u
));
4521 u
->type
= update_kill_array
;
4522 u
->dev_idx
= current_vol
;
4523 append_metadata_update(st
, u
, sizeof(*u
));
4528 for (dp
= &super
->devlist
; *dp
;)
4529 if ((*dp
)->index
== current_vol
) {
4532 handle_missing(super
, (*dp
)->dev
);
4533 if ((*dp
)->index
> current_vol
)
4538 /* no more raid devices, all active components are now spares,
4539 * but of course failed are still failed
4541 if (--mpb
->num_raid_devs
== 0) {
4544 for (d
= super
->disks
; d
; d
= d
->next
)
4545 if (d
->index
> -2) {
4547 d
->disk
.status
= SPARE_DISK
;
4551 super
->updates_pending
++;
4556 static int update_subarray_imsm(struct supertype
*st
, char *subarray
,
4557 char *update
, struct mddev_ident
*ident
)
4559 /* update the subarray currently referenced by ->current_vol */
4560 struct intel_super
*super
= st
->sb
;
4561 struct imsm_super
*mpb
= super
->anchor
;
4563 if (strcmp(update
, "name") == 0) {
4564 char *name
= ident
->name
;
4568 if (is_subarray_active(subarray
, st
->devname
)) {
4570 Name
": Unable to update name of active subarray\n");
4574 if (!check_name(super
, name
, 0))
4577 vol
= strtoul(subarray
, &ep
, 10);
4578 if (*ep
!= '\0' || vol
>= super
->anchor
->num_raid_devs
)
4581 if (st
->update_tail
) {
4582 struct imsm_update_rename_array
*u
= malloc(sizeof(*u
));
4586 u
->type
= update_rename_array
;
4588 snprintf((char *) u
->name
, MAX_RAID_SERIAL_LEN
, "%s", name
);
4589 append_metadata_update(st
, u
, sizeof(*u
));
4591 struct imsm_dev
*dev
;
4594 dev
= get_imsm_dev(super
, vol
);
4595 snprintf((char *) dev
->volume
, MAX_RAID_SERIAL_LEN
, "%s", name
);
4596 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
4597 dev
= get_imsm_dev(super
, i
);
4598 handle_missing(super
, dev
);
4600 super
->updates_pending
++;
4607 #endif /* MDASSEMBLE */
4609 static int is_gen_migration(struct imsm_dev
*dev
)
4611 if (!dev
->vol
.migr_state
)
4614 if (migr_type(dev
) == MIGR_GEN_MIGR
)
4620 static int is_rebuilding(struct imsm_dev
*dev
)
4622 struct imsm_map
*migr_map
;
4624 if (!dev
->vol
.migr_state
)
4627 if (migr_type(dev
) != MIGR_REBUILD
)
4630 migr_map
= get_imsm_map(dev
, 1);
4632 if (migr_map
->map_state
== IMSM_T_STATE_DEGRADED
)
4638 static void update_recovery_start(struct imsm_dev
*dev
, struct mdinfo
*array
)
4640 struct mdinfo
*rebuild
= NULL
;
4644 if (!is_rebuilding(dev
))
4647 /* Find the rebuild target, but punt on the dual rebuild case */
4648 for (d
= array
->devs
; d
; d
= d
->next
)
4649 if (d
->recovery_start
== 0) {
4656 /* (?) none of the disks are marked with
4657 * IMSM_ORD_REBUILD, so assume they are missing and the
4658 * disk_ord_tbl was not correctly updated
4660 dprintf("%s: failed to locate out-of-sync disk\n", __func__
);
4664 units
= __le32_to_cpu(dev
->vol
.curr_migr_unit
);
4665 rebuild
->recovery_start
= units
* blocks_per_migr_unit(dev
);
4669 static struct mdinfo
*container_content_imsm(struct supertype
*st
, char *subarray
)
4671 /* Given a container loaded by load_super_imsm_all,
4672 * extract information about all the arrays into
4674 * If 'subarray' is given, just extract info about that array.
4676 * For each imsm_dev create an mdinfo, fill it in,
4677 * then look for matching devices in super->disks
4678 * and create appropriate device mdinfo.
4680 struct intel_super
*super
= st
->sb
;
4681 struct imsm_super
*mpb
= super
->anchor
;
4682 struct mdinfo
*rest
= NULL
;
4686 /* check for bad blocks */
4687 if (imsm_bbm_log_size(super
->anchor
))
4690 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
4691 struct imsm_dev
*dev
;
4692 struct imsm_map
*map
;
4693 struct imsm_map
*map2
;
4694 struct mdinfo
*this;
4699 (i
!= strtoul(subarray
, &ep
, 10) || *ep
!= '\0'))
4702 dev
= get_imsm_dev(super
, i
);
4703 map
= get_imsm_map(dev
, 0);
4704 map2
= get_imsm_map(dev
, 1);
4706 /* do not publish arrays that are in the middle of an
4707 * unsupported migration
4709 if (dev
->vol
.migr_state
&&
4710 (migr_type(dev
) == MIGR_STATE_CHANGE
)) {
4711 fprintf(stderr
, Name
": cannot assemble volume '%.16s':"
4712 " unsupported migration in progress\n",
4717 this = malloc(sizeof(*this));
4719 fprintf(stderr
, Name
": failed to allocate %zu bytes\n",
4723 memset(this, 0, sizeof(*this));
4726 super
->current_vol
= i
;
4727 getinfo_super_imsm_volume(st
, this, NULL
);
4728 for (slot
= 0 ; slot
< map
->num_members
; slot
++) {
4729 unsigned long long recovery_start
;
4730 struct mdinfo
*info_d
;
4737 idx
= get_imsm_disk_idx(dev
, slot
, 0);
4738 ord
= get_imsm_ord_tbl_ent(dev
, slot
, 0);
4739 for (d
= super
->disks
; d
; d
= d
->next
)
4740 if (d
->index
== idx
)
4743 recovery_start
= MaxSector
;
4746 if (d
&& is_failed(&d
->disk
))
4748 if (ord
& IMSM_ORD_REBUILD
)
4752 * if we skip some disks the array will be assmebled degraded;
4753 * reset resync start to avoid a dirty-degraded
4754 * situation when performing the intial sync
4756 * FIXME handle dirty degraded
4758 if ((skip
|| recovery_start
== 0) && !dev
->vol
.dirty
)
4759 this->resync_start
= MaxSector
;
4763 info_d
= calloc(1, sizeof(*info_d
));
4765 fprintf(stderr
, Name
": failed to allocate disk"
4766 " for volume %.16s\n", dev
->volume
);
4767 info_d
= this->devs
;
4769 struct mdinfo
*d
= info_d
->next
;
4778 info_d
->next
= this->devs
;
4779 this->devs
= info_d
;
4781 info_d
->disk
.number
= d
->index
;
4782 info_d
->disk
.major
= d
->major
;
4783 info_d
->disk
.minor
= d
->minor
;
4784 info_d
->disk
.raid_disk
= slot
;
4785 info_d
->recovery_start
= recovery_start
;
4787 if (slot
< map2
->num_members
)
4788 info_d
->disk
.state
= (1 << MD_DISK_ACTIVE
);
4790 this->array
.spare_disks
++;
4792 if (slot
< map
->num_members
)
4793 info_d
->disk
.state
= (1 << MD_DISK_ACTIVE
);
4795 this->array
.spare_disks
++;
4797 if (info_d
->recovery_start
== MaxSector
)
4798 this->array
.working_disks
++;
4800 info_d
->events
= __le32_to_cpu(mpb
->generation_num
);
4801 info_d
->data_offset
= __le32_to_cpu(map
->pba_of_lba0
);
4802 info_d
->component_size
= __le32_to_cpu(map
->blocks_per_member
);
4804 /* now that the disk list is up-to-date fixup recovery_start */
4805 update_recovery_start(dev
, this);
4809 /* if array has bad blocks, set suitable bit in array status */
4811 rest
->array
.state
|= (1<<MD_SB_BBM_ERRORS
);
4817 static __u8
imsm_check_degraded(struct intel_super
*super
, struct imsm_dev
*dev
, int failed
)
4819 struct imsm_map
*map
= get_imsm_map(dev
, 0);
4822 return map
->map_state
== IMSM_T_STATE_UNINITIALIZED
?
4823 IMSM_T_STATE_UNINITIALIZED
: IMSM_T_STATE_NORMAL
;
4825 switch (get_imsm_raid_level(map
)) {
4827 return IMSM_T_STATE_FAILED
;
4830 if (failed
< map
->num_members
)
4831 return IMSM_T_STATE_DEGRADED
;
4833 return IMSM_T_STATE_FAILED
;
4838 * check to see if any mirrors have failed, otherwise we
4839 * are degraded. Even numbered slots are mirrored on
4843 /* gcc -Os complains that this is unused */
4844 int insync
= insync
;
4846 for (i
= 0; i
< map
->num_members
; i
++) {
4847 __u32 ord
= get_imsm_ord_tbl_ent(dev
, i
, -1);
4848 int idx
= ord_to_idx(ord
);
4849 struct imsm_disk
*disk
;
4851 /* reset the potential in-sync count on even-numbered
4852 * slots. num_copies is always 2 for imsm raid10
4857 disk
= get_imsm_disk(super
, idx
);
4858 if (!disk
|| is_failed(disk
) || ord
& IMSM_ORD_REBUILD
)
4861 /* no in-sync disks left in this mirror the
4865 return IMSM_T_STATE_FAILED
;
4868 return IMSM_T_STATE_DEGRADED
;
4872 return IMSM_T_STATE_DEGRADED
;
4874 return IMSM_T_STATE_FAILED
;
4880 return map
->map_state
;
4883 static int imsm_count_failed(struct intel_super
*super
, struct imsm_dev
*dev
)
4887 struct imsm_disk
*disk
;
4888 struct imsm_map
*map
= get_imsm_map(dev
, 0);
4889 struct imsm_map
*prev
= get_imsm_map(dev
, dev
->vol
.migr_state
);
4893 /* at the beginning of migration we set IMSM_ORD_REBUILD on
4894 * disks that are being rebuilt. New failures are recorded to
4895 * map[0]. So we look through all the disks we started with and
4896 * see if any failures are still present, or if any new ones
4899 * FIXME add support for online capacity expansion and
4900 * raid-level-migration
4902 for (i
= 0; i
< prev
->num_members
; i
++) {
4903 ord
= __le32_to_cpu(prev
->disk_ord_tbl
[i
]);
4904 ord
|= __le32_to_cpu(map
->disk_ord_tbl
[i
]);
4905 idx
= ord_to_idx(ord
);
4907 disk
= get_imsm_disk(super
, idx
);
4908 if (!disk
|| is_failed(disk
) || ord
& IMSM_ORD_REBUILD
)
4916 static int imsm_open_new(struct supertype
*c
, struct active_array
*a
,
4919 struct intel_super
*super
= c
->sb
;
4920 struct imsm_super
*mpb
= super
->anchor
;
4922 if (atoi(inst
) >= mpb
->num_raid_devs
) {
4923 fprintf(stderr
, "%s: subarry index %d, out of range\n",
4924 __func__
, atoi(inst
));
4928 dprintf("imsm: open_new %s\n", inst
);
4929 a
->info
.container_member
= atoi(inst
);
4933 static int is_resyncing(struct imsm_dev
*dev
)
4935 struct imsm_map
*migr_map
;
4937 if (!dev
->vol
.migr_state
)
4940 if (migr_type(dev
) == MIGR_INIT
||
4941 migr_type(dev
) == MIGR_REPAIR
)
4944 if (migr_type(dev
) == MIGR_GEN_MIGR
)
4947 migr_map
= get_imsm_map(dev
, 1);
4949 if ((migr_map
->map_state
== IMSM_T_STATE_NORMAL
) &&
4950 (dev
->vol
.migr_type
!= MIGR_GEN_MIGR
))
4956 /* return true if we recorded new information */
4957 static int mark_failure(struct imsm_dev
*dev
, struct imsm_disk
*disk
, int idx
)
4961 struct imsm_map
*map
;
4963 /* new failures are always set in map[0] */
4964 map
= get_imsm_map(dev
, 0);
4966 slot
= get_imsm_disk_slot(map
, idx
);
4970 ord
= __le32_to_cpu(map
->disk_ord_tbl
[slot
]);
4971 if (is_failed(disk
) && (ord
& IMSM_ORD_REBUILD
))
4974 disk
->status
|= FAILED_DISK
;
4975 disk
->status
&= ~CONFIGURED_DISK
;
4976 set_imsm_ord_tbl_ent(map
, slot
, idx
| IMSM_ORD_REBUILD
);
4977 if (map
->failed_disk_num
== 0xff)
4978 map
->failed_disk_num
= slot
;
4982 static void mark_missing(struct imsm_dev
*dev
, struct imsm_disk
*disk
, int idx
)
4984 mark_failure(dev
, disk
, idx
);
4986 if (disk
->scsi_id
== __cpu_to_le32(~(__u32
)0))
4989 disk
->scsi_id
= __cpu_to_le32(~(__u32
)0);
4990 memmove(&disk
->serial
[0], &disk
->serial
[1], MAX_RAID_SERIAL_LEN
- 1);
4993 static void handle_missing(struct intel_super
*super
, struct imsm_dev
*dev
)
4999 if (!super
->missing
)
5001 failed
= imsm_count_failed(super
, dev
);
5002 map_state
= imsm_check_degraded(super
, dev
, failed
);
5004 dprintf("imsm: mark missing\n");
5005 end_migration(dev
, map_state
);
5006 for (dl
= super
->missing
; dl
; dl
= dl
->next
)
5007 mark_missing(dev
, &dl
->disk
, dl
->index
);
5008 super
->updates_pending
++;
5011 static void imsm_set_disk(struct active_array
*a
, int n
, int state
);
5013 static void imsm_progress_container_reshape(struct intel_super
*super
)
5015 /* if no device has a migr_state, but some device has a
5016 * different number of members than the previous device, start
5017 * changing the number of devices in this device to match
5020 struct imsm_super
*mpb
= super
->anchor
;
5021 int prev_disks
= -1;
5024 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
5025 struct imsm_dev
*dev
= get_imsm_dev(super
, i
);
5026 struct imsm_map
*map
= get_imsm_map(dev
, 0);
5027 struct imsm_map
*map2
;
5028 int prev_num_members
;
5031 if (dev
->vol
.migr_state
)
5034 if (prev_disks
== -1)
5035 prev_disks
= map
->num_members
;
5036 if (prev_disks
== map
->num_members
)
5039 /* OK, this array needs to enter reshape mode.
5040 * i.e it needs a migr_state
5043 prev_num_members
= map
->num_members
;
5044 map
->num_members
= prev_disks
;
5045 dev
->vol
.migr_state
= 1;
5046 dev
->vol
.curr_migr_unit
= 0;
5047 dev
->vol
.migr_type
= MIGR_GEN_MIGR
;
5048 for (i
= prev_num_members
;
5049 i
< map
->num_members
; i
++)
5050 set_imsm_ord_tbl_ent(map
, i
, i
);
5051 map2
= get_imsm_map(dev
, 1);
5052 /* Copy the current map */
5053 memcpy(map2
, map
, sizeof_imsm_map(map
));
5054 map2
->num_members
= prev_num_members
;
5056 /* calculate new size
5058 used_disks
= imsm_num_data_members(dev
, 0);
5060 unsigned long long array_blocks
;
5063 map
->blocks_per_member
5065 /* round array size down to closest MB
5067 array_blocks
= (array_blocks
5068 >> SECT_PER_MB_SHIFT
)
5069 << SECT_PER_MB_SHIFT
;
5071 __cpu_to_le32((__u32
)array_blocks
);
5074 (__u32
)(array_blocks
>> 32));
5076 super
->updates_pending
++;
5080 /* Handle dirty -> clean transititions, resync and reshape. Degraded and rebuild
5081 * states are handled in imsm_set_disk() with one exception, when a
5082 * resync is stopped due to a new failure this routine will set the
5083 * 'degraded' state for the array.
5085 static int imsm_set_array_state(struct active_array
*a
, int consistent
)
5087 int inst
= a
->info
.container_member
;
5088 struct intel_super
*super
= a
->container
->sb
;
5089 struct imsm_dev
*dev
= get_imsm_dev(super
, inst
);
5090 struct imsm_map
*map
= get_imsm_map(dev
, 0);
5091 int failed
= imsm_count_failed(super
, dev
);
5092 __u8 map_state
= imsm_check_degraded(super
, dev
, failed
);
5093 __u32 blocks_per_unit
;
5095 if (dev
->vol
.migr_state
&&
5096 dev
->vol
.migr_type
== MIGR_GEN_MIGR
) {
5097 /* array state change is blocked due to reshape action
5099 * - abort the reshape (if last_checkpoint is 0 and action!= reshape)
5100 * - finish the reshape (if last_checkpoint is big and action != reshape)
5101 * - update curr_migr_unit
5103 if (a
->curr_action
== reshape
) {
5104 /* still reshaping, maybe update curr_migr_unit */
5105 long long blocks_per_unit
= blocks_per_migr_unit(dev
);
5106 long long unit
= a
->last_checkpoint
;
5107 if (blocks_per_unit
) {
5108 unit
/= blocks_per_unit
;
5110 __le32_to_cpu(dev
->vol
.curr_migr_unit
)) {
5111 dev
->vol
.curr_migr_unit
=
5112 __cpu_to_le32(unit
);
5113 super
->updates_pending
++;
5118 if (a
->last_checkpoint
== 0 && a
->prev_action
== reshape
) {
5119 /* for some reason we aborted the reshape.
5122 struct imsm_map
*map2
= get_imsm_map(dev
, 1);
5123 dev
->vol
.migr_state
= 0;
5124 dev
->vol
.migr_type
= 0;
5125 dev
->vol
.curr_migr_unit
= 0;
5126 memcpy(map
, map2
, sizeof_imsm_map(map2
));
5127 super
->updates_pending
++;
5129 if (a
->last_checkpoint
>= a
->info
.component_size
) {
5130 unsigned long long array_blocks
;
5132 /* it seems the reshape is all done */
5133 dev
->vol
.migr_state
= 0;
5134 dev
->vol
.migr_type
= 0;
5135 dev
->vol
.curr_migr_unit
= 0;
5137 used_disks
= imsm_num_data_members(dev
, -1);
5138 array_blocks
= map
->blocks_per_member
* used_disks
;
5139 /* round array size down to closest MB */
5140 array_blocks
= (array_blocks
>> SECT_PER_MB_SHIFT
)
5141 << SECT_PER_MB_SHIFT
;
5142 dev
->size_low
= __cpu_to_le32((__u32
) array_blocks
);
5143 dev
->size_high
= __cpu_to_le32((__u32
) (array_blocks
>> 32));
5144 a
->info
.custom_array_size
= array_blocks
;
5145 a
->check_reshape
= 1; /* encourage manager to update
5148 super
->updates_pending
++;
5149 imsm_progress_container_reshape(super
);
5154 /* before we activate this array handle any missing disks */
5155 if (consistent
== 2)
5156 handle_missing(super
, dev
);
5158 if (consistent
== 2 &&
5159 (!is_resync_complete(&a
->info
) ||
5160 map_state
!= IMSM_T_STATE_NORMAL
||
5161 dev
->vol
.migr_state
))
5164 if (is_resync_complete(&a
->info
)) {
5165 /* complete intialization / resync,
5166 * recovery and interrupted recovery is completed in
5169 if (is_resyncing(dev
)) {
5170 dprintf("imsm: mark resync done\n");
5171 end_migration(dev
, map_state
);
5172 super
->updates_pending
++;
5173 a
->last_checkpoint
= 0;
5175 } else if (!is_resyncing(dev
) && !failed
) {
5176 /* mark the start of the init process if nothing is failed */
5177 dprintf("imsm: mark resync start\n");
5178 if (map
->map_state
== IMSM_T_STATE_UNINITIALIZED
)
5179 migrate(dev
, IMSM_T_STATE_NORMAL
, MIGR_INIT
);
5181 migrate(dev
, IMSM_T_STATE_NORMAL
, MIGR_REPAIR
);
5182 super
->updates_pending
++;
5185 /* check if we can update curr_migr_unit from resync_start, recovery_start */
5186 blocks_per_unit
= blocks_per_migr_unit(dev
);
5187 if (blocks_per_unit
) {
5191 units
= a
->last_checkpoint
/ blocks_per_unit
;
5194 /* check that we did not overflow 32-bits, and that
5195 * curr_migr_unit needs updating
5197 if (units32
== units
&&
5198 __le32_to_cpu(dev
->vol
.curr_migr_unit
) != units32
) {
5199 dprintf("imsm: mark checkpoint (%u)\n", units32
);
5200 dev
->vol
.curr_migr_unit
= __cpu_to_le32(units32
);
5201 super
->updates_pending
++;
5205 /* mark dirty / clean */
5206 if (dev
->vol
.dirty
!= !consistent
) {
5207 dprintf("imsm: mark '%s'\n", consistent
? "clean" : "dirty");
5212 super
->updates_pending
++;
5215 /* finalize online capacity expansion/reshape */
5216 if ((a
->curr_action
!= reshape
) &&
5217 (a
->prev_action
== reshape
)) {
5220 for (mdi
= a
->info
.devs
; mdi
; mdi
= mdi
->next
)
5221 imsm_set_disk(a
, mdi
->disk
.raid_disk
, mdi
->curr_state
);
5227 static void imsm_set_disk(struct active_array
*a
, int n
, int state
)
5229 int inst
= a
->info
.container_member
;
5230 struct intel_super
*super
= a
->container
->sb
;
5231 struct imsm_dev
*dev
= get_imsm_dev(super
, inst
);
5232 struct imsm_map
*map
= get_imsm_map(dev
, 0);
5233 struct imsm_disk
*disk
;
5238 if (n
> map
->num_members
)
5239 fprintf(stderr
, "imsm: set_disk %d out of range 0..%d\n",
5240 n
, map
->num_members
- 1);
5245 dprintf("imsm: set_disk %d:%x\n", n
, state
);
5247 ord
= get_imsm_ord_tbl_ent(dev
, n
, -1);
5248 disk
= get_imsm_disk(super
, ord_to_idx(ord
));
5250 /* check for new failures */
5251 if (state
& DS_FAULTY
) {
5252 if (mark_failure(dev
, disk
, ord_to_idx(ord
)))
5253 super
->updates_pending
++;
5256 /* check if in_sync */
5257 if (state
& DS_INSYNC
&& ord
& IMSM_ORD_REBUILD
&& is_rebuilding(dev
)) {
5258 struct imsm_map
*migr_map
= get_imsm_map(dev
, 1);
5260 set_imsm_ord_tbl_ent(migr_map
, n
, ord_to_idx(ord
));
5261 super
->updates_pending
++;
5264 failed
= imsm_count_failed(super
, dev
);
5265 map_state
= imsm_check_degraded(super
, dev
, failed
);
5267 /* check if recovery complete, newly degraded, or failed */
5268 if (map_state
== IMSM_T_STATE_NORMAL
&& is_rebuilding(dev
)) {
5269 end_migration(dev
, map_state
);
5270 map
= get_imsm_map(dev
, 0);
5271 map
->failed_disk_num
= ~0;
5272 super
->updates_pending
++;
5273 a
->last_checkpoint
= 0;
5274 } else if (map_state
== IMSM_T_STATE_DEGRADED
&&
5275 map
->map_state
!= map_state
&&
5276 !dev
->vol
.migr_state
) {
5277 dprintf("imsm: mark degraded\n");
5278 map
->map_state
= map_state
;
5279 super
->updates_pending
++;
5280 a
->last_checkpoint
= 0;
5281 } else if (map_state
== IMSM_T_STATE_FAILED
&&
5282 map
->map_state
!= map_state
) {
5283 dprintf("imsm: mark failed\n");
5284 end_migration(dev
, map_state
);
5285 super
->updates_pending
++;
5286 a
->last_checkpoint
= 0;
5287 } else if (is_gen_migration(dev
)) {
5288 dprintf("imsm: Detected General Migration in state: ");
5289 if (map_state
== IMSM_T_STATE_NORMAL
) {
5290 end_migration(dev
, map_state
);
5291 map
= get_imsm_map(dev
, 0);
5292 map
->failed_disk_num
= ~0;
5293 dprintf("normal\n");
5295 if (map_state
== IMSM_T_STATE_DEGRADED
) {
5296 printf("degraded\n");
5297 end_migration(dev
, map_state
);
5299 dprintf("failed\n");
5301 map
->map_state
= map_state
;
5303 super
->updates_pending
++;
5307 static int store_imsm_mpb(int fd
, struct imsm_super
*mpb
)
5310 __u32 mpb_size
= __le32_to_cpu(mpb
->mpb_size
);
5311 unsigned long long dsize
;
5312 unsigned long long sectors
;
5314 get_dev_size(fd
, NULL
, &dsize
);
5316 if (mpb_size
> 512) {
5317 /* -1 to account for anchor */
5318 sectors
= mpb_sectors(mpb
) - 1;
5320 /* write the extended mpb to the sectors preceeding the anchor */
5321 if (lseek64(fd
, dsize
- (512 * (2 + sectors
)), SEEK_SET
) < 0)
5324 if ((unsigned long long)write(fd
, buf
+ 512, 512 * sectors
)
5329 /* first block is stored on second to last sector of the disk */
5330 if (lseek64(fd
, dsize
- (512 * 2), SEEK_SET
) < 0)
5333 if (write(fd
, buf
, 512) != 512)
5339 static void imsm_sync_metadata(struct supertype
*container
)
5341 struct intel_super
*super
= container
->sb
;
5343 dprintf("sync metadata: %d\n", super
->updates_pending
);
5344 if (!super
->updates_pending
)
5347 write_super_imsm(container
, 0);
5349 super
->updates_pending
= 0;
5352 static struct dl
*imsm_readd(struct intel_super
*super
, int idx
, struct active_array
*a
)
5354 struct imsm_dev
*dev
= get_imsm_dev(super
, a
->info
.container_member
);
5355 int i
= get_imsm_disk_idx(dev
, idx
, -1);
5358 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
5362 if (dl
&& is_failed(&dl
->disk
))
5366 dprintf("%s: found %x:%x\n", __func__
, dl
->major
, dl
->minor
);
5371 static struct dl
*imsm_add_spare(struct intel_super
*super
, int slot
,
5372 struct active_array
*a
, int activate_new
,
5373 struct mdinfo
*additional_test_list
)
5375 struct imsm_dev
*dev
= get_imsm_dev(super
, a
->info
.container_member
);
5376 int idx
= get_imsm_disk_idx(dev
, slot
, -1);
5377 struct imsm_super
*mpb
= super
->anchor
;
5378 struct imsm_map
*map
;
5379 unsigned long long pos
;
5384 __u32 array_start
= 0;
5385 __u32 array_end
= 0;
5387 struct mdinfo
*test_list
;
5389 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
5390 /* If in this array, skip */
5391 for (d
= a
->info
.devs
; d
; d
= d
->next
)
5392 if (d
->state_fd
>= 0 &&
5393 d
->disk
.major
== dl
->major
&&
5394 d
->disk
.minor
== dl
->minor
) {
5395 dprintf("%x:%x already in array\n",
5396 dl
->major
, dl
->minor
);
5401 test_list
= additional_test_list
;
5403 if (test_list
->disk
.major
== dl
->major
&&
5404 test_list
->disk
.minor
== dl
->minor
) {
5405 dprintf("%x:%x already in additional test list\n",
5406 dl
->major
, dl
->minor
);
5409 test_list
= test_list
->next
;
5414 /* skip in use or failed drives */
5415 if (is_failed(&dl
->disk
) || idx
== dl
->index
||
5417 dprintf("%x:%x status (failed: %d index: %d)\n",
5418 dl
->major
, dl
->minor
, is_failed(&dl
->disk
), idx
);
5422 /* skip pure spares when we are looking for partially
5423 * assimilated drives
5425 if (dl
->index
== -1 && !activate_new
)
5428 /* Does this unused device have the requisite free space?
5429 * It needs to be able to cover all member volumes
5431 ex
= get_extents(super
, dl
);
5433 dprintf("cannot get extents\n");
5436 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
5437 dev
= get_imsm_dev(super
, i
);
5438 map
= get_imsm_map(dev
, 0);
5440 /* check if this disk is already a member of
5443 if (get_imsm_disk_slot(map
, dl
->index
) >= 0)
5449 array_start
= __le32_to_cpu(map
->pba_of_lba0
);
5450 array_end
= array_start
+
5451 __le32_to_cpu(map
->blocks_per_member
) - 1;
5454 /* check that we can start at pba_of_lba0 with
5455 * blocks_per_member of space
5457 if (array_start
>= pos
&& array_end
< ex
[j
].start
) {
5461 pos
= ex
[j
].start
+ ex
[j
].size
;
5463 } while (ex
[j
-1].size
);
5470 if (i
< mpb
->num_raid_devs
) {
5471 dprintf("%x:%x does not have %u to %u available\n",
5472 dl
->major
, dl
->minor
, array_start
, array_end
);
5483 static int imsm_rebuild_allowed(struct supertype
*cont
, int dev_idx
, int failed
)
5485 struct imsm_dev
*dev2
;
5486 struct imsm_map
*map
;
5492 dev2
= get_imsm_dev(cont
->sb
, dev_idx
);
5494 state
= imsm_check_degraded(cont
->sb
, dev2
, failed
);
5495 if (state
== IMSM_T_STATE_FAILED
) {
5496 map
= get_imsm_map(dev2
, 0);
5499 for (slot
= 0; slot
< map
->num_members
; slot
++) {
5501 * Check if failed disks are deleted from intel
5502 * disk list or are marked to be deleted
5504 idx
= get_imsm_disk_idx(dev2
, slot
, -1);
5505 idisk
= get_imsm_dl_disk(cont
->sb
, idx
);
5507 * Do not rebuild the array if failed disks
5508 * from failed sub-array are not removed from
5512 is_failed(&idisk
->disk
) &&
5513 (idisk
->action
!= DISK_REMOVE
))
5521 static struct mdinfo
*imsm_activate_spare(struct active_array
*a
,
5522 struct metadata_update
**updates
)
5525 * Find a device with unused free space and use it to replace a
5526 * failed/vacant region in an array. We replace failed regions one a
5527 * array at a time. The result is that a new spare disk will be added
5528 * to the first failed array and after the monitor has finished
5529 * propagating failures the remainder will be consumed.
5531 * FIXME add a capability for mdmon to request spares from another
5535 struct intel_super
*super
= a
->container
->sb
;
5536 int inst
= a
->info
.container_member
;
5537 struct imsm_dev
*dev
= get_imsm_dev(super
, inst
);
5538 struct imsm_map
*map
= get_imsm_map(dev
, 0);
5539 int failed
= a
->info
.array
.raid_disks
;
5540 struct mdinfo
*rv
= NULL
;
5543 struct metadata_update
*mu
;
5545 struct imsm_update_activate_spare
*u
;
5550 for (d
= a
->info
.devs
; d
; d
= d
->next
) {
5551 if ((d
->curr_state
& DS_FAULTY
) &&
5553 /* wait for Removal to happen */
5555 if (d
->state_fd
>= 0)
5559 dprintf("imsm: activate spare: inst=%d failed=%d (%d) level=%d\n",
5560 inst
, failed
, a
->info
.array
.raid_disks
, a
->info
.array
.level
);
5562 if (dev
->vol
.migr_state
&&
5563 dev
->vol
.migr_type
== MIGR_GEN_MIGR
)
5564 /* No repair during migration */
5567 if (a
->info
.array
.level
== 4)
5568 /* No repair for takeovered array
5569 * imsm doesn't support raid4
5573 if (imsm_check_degraded(super
, dev
, failed
) != IMSM_T_STATE_DEGRADED
)
5577 * If there are any failed disks check state of the other volume.
5578 * Block rebuild if the another one is failed until failed disks
5579 * are removed from container.
5582 dprintf("found failed disks in %s, check if there another"
5583 "failed sub-array.\n",
5585 /* check if states of the other volumes allow for rebuild */
5586 for (i
= 0; i
< super
->anchor
->num_raid_devs
; i
++) {
5588 allowed
= imsm_rebuild_allowed(a
->container
,
5596 /* For each slot, if it is not working, find a spare */
5597 for (i
= 0; i
< a
->info
.array
.raid_disks
; i
++) {
5598 for (d
= a
->info
.devs
; d
; d
= d
->next
)
5599 if (d
->disk
.raid_disk
== i
)
5601 dprintf("found %d: %p %x\n", i
, d
, d
?d
->curr_state
:0);
5602 if (d
&& (d
->state_fd
>= 0))
5606 * OK, this device needs recovery. Try to re-add the
5607 * previous occupant of this slot, if this fails see if
5608 * we can continue the assimilation of a spare that was
5609 * partially assimilated, finally try to activate a new
5612 dl
= imsm_readd(super
, i
, a
);
5614 dl
= imsm_add_spare(super
, i
, a
, 0, NULL
);
5616 dl
= imsm_add_spare(super
, i
, a
, 1, NULL
);
5620 /* found a usable disk with enough space */
5621 di
= malloc(sizeof(*di
));
5624 memset(di
, 0, sizeof(*di
));
5626 /* dl->index will be -1 in the case we are activating a
5627 * pristine spare. imsm_process_update() will create a
5628 * new index in this case. Once a disk is found to be
5629 * failed in all member arrays it is kicked from the
5632 di
->disk
.number
= dl
->index
;
5634 /* (ab)use di->devs to store a pointer to the device
5637 di
->devs
= (struct mdinfo
*) dl
;
5639 di
->disk
.raid_disk
= i
;
5640 di
->disk
.major
= dl
->major
;
5641 di
->disk
.minor
= dl
->minor
;
5643 di
->recovery_start
= 0;
5644 di
->data_offset
= __le32_to_cpu(map
->pba_of_lba0
);
5645 di
->component_size
= a
->info
.component_size
;
5646 di
->container_member
= inst
;
5647 super
->random
= random32();
5651 dprintf("%x:%x to be %d at %llu\n", dl
->major
, dl
->minor
,
5652 i
, di
->data_offset
);
5658 /* No spares found */
5660 /* Now 'rv' has a list of devices to return.
5661 * Create a metadata_update record to update the
5662 * disk_ord_tbl for the array
5664 mu
= malloc(sizeof(*mu
));
5666 mu
->buf
= malloc(sizeof(struct imsm_update_activate_spare
) * num_spares
);
5667 if (mu
->buf
== NULL
) {
5674 struct mdinfo
*n
= rv
->next
;
5683 mu
->space_list
= NULL
;
5684 mu
->len
= sizeof(struct imsm_update_activate_spare
) * num_spares
;
5685 mu
->next
= *updates
;
5686 u
= (struct imsm_update_activate_spare
*) mu
->buf
;
5688 for (di
= rv
; di
; di
= di
->next
) {
5689 u
->type
= update_activate_spare
;
5690 u
->dl
= (struct dl
*) di
->devs
;
5692 u
->slot
= di
->disk
.raid_disk
;
5703 static int disks_overlap(struct intel_super
*super
, int idx
, struct imsm_update_create_array
*u
)
5705 struct imsm_dev
*dev
= get_imsm_dev(super
, idx
);
5706 struct imsm_map
*map
= get_imsm_map(dev
, 0);
5707 struct imsm_map
*new_map
= get_imsm_map(&u
->dev
, 0);
5708 struct disk_info
*inf
= get_disk_info(u
);
5709 struct imsm_disk
*disk
;
5713 for (i
= 0; i
< map
->num_members
; i
++) {
5714 disk
= get_imsm_disk(super
, get_imsm_disk_idx(dev
, i
, -1));
5715 for (j
= 0; j
< new_map
->num_members
; j
++)
5716 if (serialcmp(disk
->serial
, inf
[j
].serial
) == 0)
5724 static struct dl
*get_disk_super(struct intel_super
*super
, int major
, int minor
)
5726 struct dl
*dl
= NULL
;
5727 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
5728 if ((dl
->major
== major
) && (dl
->minor
== minor
))
5733 static int remove_disk_super(struct intel_super
*super
, int major
, int minor
)
5735 struct dl
*prev
= NULL
;
5739 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
5740 if ((dl
->major
== major
) && (dl
->minor
== minor
)) {
5743 prev
->next
= dl
->next
;
5745 super
->disks
= dl
->next
;
5747 __free_imsm_disk(dl
);
5748 dprintf("%s: removed %x:%x\n",
5749 __func__
, major
, minor
);
5757 static void imsm_delete(struct intel_super
*super
, struct dl
**dlp
, unsigned index
);
5759 static int add_remove_disk_update(struct intel_super
*super
)
5761 int check_degraded
= 0;
5762 struct dl
*disk
= NULL
;
5763 /* add/remove some spares to/from the metadata/contrainer */
5764 while (super
->disk_mgmt_list
) {
5765 struct dl
*disk_cfg
;
5767 disk_cfg
= super
->disk_mgmt_list
;
5768 super
->disk_mgmt_list
= disk_cfg
->next
;
5769 disk_cfg
->next
= NULL
;
5771 if (disk_cfg
->action
== DISK_ADD
) {
5772 disk_cfg
->next
= super
->disks
;
5773 super
->disks
= disk_cfg
;
5775 dprintf("%s: added %x:%x\n",
5776 __func__
, disk_cfg
->major
,
5778 } else if (disk_cfg
->action
== DISK_REMOVE
) {
5779 dprintf("Disk remove action processed: %x.%x\n",
5780 disk_cfg
->major
, disk_cfg
->minor
);
5781 disk
= get_disk_super(super
,
5785 /* store action status */
5786 disk
->action
= DISK_REMOVE
;
5787 /* remove spare disks only */
5788 if (disk
->index
== -1) {
5789 remove_disk_super(super
,
5794 /* release allocate disk structure */
5795 __free_imsm_disk(disk_cfg
);
5798 return check_degraded
;
5801 static int apply_reshape_container_disks_update(struct imsm_update_reshape
*u
,
5802 struct intel_super
*super
,
5805 struct dl
*new_disk
;
5806 struct intel_dev
*id
;
5808 int delta_disks
= u
->new_raid_disks
- u
->old_raid_disks
;
5809 int disk_count
= u
->old_raid_disks
;
5810 void **tofree
= NULL
;
5811 int devices_to_reshape
= 1;
5812 struct imsm_super
*mpb
= super
->anchor
;
5815 dprintf("imsm: imsm_process_update() for update_reshape\n");
5817 /* enable spares to use in array */
5818 for (i
= 0; i
< delta_disks
; i
++) {
5819 new_disk
= get_disk_super(super
,
5820 major(u
->new_disks
[i
]),
5821 minor(u
->new_disks
[i
]));
5822 dprintf("imsm: imsm_process_update(): new disk "
5823 "for reshape is: %i:%i (%p, index = %i)\n",
5824 major(u
->new_disks
[i
]), minor(u
->new_disks
[i
]),
5825 new_disk
, new_disk
->index
);
5826 if ((new_disk
== NULL
) ||
5827 ((new_disk
->index
>= 0) &&
5828 (new_disk
->index
< u
->old_raid_disks
)))
5829 goto update_reshape_exit
;
5830 new_disk
->index
= disk_count
++;
5831 /* slot to fill in autolayout
5833 new_disk
->raiddisk
= new_disk
->index
;
5834 new_disk
->disk
.status
|=
5836 new_disk
->disk
.status
&= ~SPARE_DISK
;
5839 dprintf("imsm: process_update(): update_reshape: volume set"
5840 " mpb->num_raid_devs = %i\n", mpb
->num_raid_devs
);
5841 /* manage changes in volume
5843 for (id
= super
->devlist
; id
; id
= id
->next
) {
5844 void **sp
= *space_list
;
5845 struct imsm_dev
*newdev
;
5846 struct imsm_map
*newmap
, *oldmap
;
5852 /* Copy the dev, but not (all of) the map */
5853 memcpy(newdev
, id
->dev
, sizeof(*newdev
));
5854 oldmap
= get_imsm_map(id
->dev
, 0);
5855 newmap
= get_imsm_map(newdev
, 0);
5856 /* Copy the current map */
5857 memcpy(newmap
, oldmap
, sizeof_imsm_map(oldmap
));
5858 /* update one device only
5860 if (devices_to_reshape
) {
5863 dprintf("process_update(): modifying "
5864 "subdev: %i\n", id
->index
);
5865 devices_to_reshape
--;
5866 newdev
->vol
.migr_state
= 1;
5867 newdev
->vol
.curr_migr_unit
= 0;
5868 newdev
->vol
.migr_type
= MIGR_GEN_MIGR
;
5869 newmap
->num_members
= u
->new_raid_disks
;
5870 for (i
= 0; i
< delta_disks
; i
++) {
5871 set_imsm_ord_tbl_ent(newmap
,
5872 u
->old_raid_disks
+ i
,
5873 u
->old_raid_disks
+ i
);
5875 /* New map is correct, now need to save old map
5877 newmap
= get_imsm_map(newdev
, 1);
5878 memcpy(newmap
, oldmap
, sizeof_imsm_map(oldmap
));
5880 /* calculate new size
5882 used_disks
= imsm_num_data_members(newdev
, 0);
5884 unsigned long long array_blocks
;
5887 newmap
->blocks_per_member
* used_disks
;
5888 /* round array size down to closest MB
5890 array_blocks
= (array_blocks
5891 >> SECT_PER_MB_SHIFT
)
5892 << SECT_PER_MB_SHIFT
;
5894 __cpu_to_le32((__u32
)array_blocks
);
5896 __cpu_to_le32((__u32
)(array_blocks
>> 32));
5900 sp
= (void **)id
->dev
;
5906 *space_list
= tofree
;
5909 update_reshape_exit
:
5914 static int apply_takeover_update(struct imsm_update_takeover
*u
,
5915 struct intel_super
*super
,
5918 struct imsm_dev
*dev
= NULL
;
5919 struct intel_dev
*dv
;
5920 struct imsm_dev
*dev_new
;
5921 struct imsm_map
*map
;
5925 for (dv
= super
->devlist
; dv
; dv
= dv
->next
)
5926 if (dv
->index
== (unsigned int)u
->subarray
) {
5934 map
= get_imsm_map(dev
, 0);
5936 if (u
->direction
== R10_TO_R0
) {
5937 /* Number of failed disks must be half of initial disk number */
5938 if (imsm_count_failed(super
, dev
) != (map
->num_members
/ 2))
5941 /* iterate through devices to mark removed disks as spare */
5942 for (dm
= super
->disks
; dm
; dm
= dm
->next
) {
5943 if (dm
->disk
.status
& FAILED_DISK
) {
5944 int idx
= dm
->index
;
5945 /* update indexes on the disk list */
5946 /* FIXME this loop-with-the-loop looks wrong, I'm not convinced
5947 the index values will end up being correct.... NB */
5948 for (du
= super
->disks
; du
; du
= du
->next
)
5949 if (du
->index
> idx
)
5951 /* mark as spare disk */
5952 dm
->disk
.status
= SPARE_DISK
;
5957 map
->num_members
= map
->num_members
/ 2;
5958 map
->map_state
= IMSM_T_STATE_NORMAL
;
5959 map
->num_domains
= 1;
5960 map
->raid_level
= 0;
5961 map
->failed_disk_num
= -1;
5964 if (u
->direction
== R0_TO_R10
) {
5966 /* update slots in current disk list */
5967 for (dm
= super
->disks
; dm
; dm
= dm
->next
) {
5971 /* create new *missing* disks */
5972 for (i
= 0; i
< map
->num_members
; i
++) {
5973 space
= *space_list
;
5976 *space_list
= *space
;
5978 memcpy(du
, super
->disks
, sizeof(*du
));
5979 du
->disk
.status
= FAILED_DISK
;
5980 du
->disk
.scsi_id
= 0;
5984 du
->index
= (i
* 2) + 1;
5985 sprintf((char *)du
->disk
.serial
,
5986 " MISSING_%d", du
->index
);
5987 sprintf((char *)du
->serial
,
5988 "MISSING_%d", du
->index
);
5989 du
->next
= super
->missing
;
5990 super
->missing
= du
;
5992 /* create new dev and map */
5993 space
= *space_list
;
5996 *space_list
= *space
;
5997 dev_new
= (void *)space
;
5998 memcpy(dev_new
, dev
, sizeof(*dev
));
5999 /* update new map */
6000 map
= get_imsm_map(dev_new
, 0);
6001 map
->failed_disk_num
= map
->num_members
;
6002 map
->num_members
= map
->num_members
* 2;
6003 map
->map_state
= IMSM_T_STATE_NORMAL
;
6004 map
->num_domains
= 2;
6005 map
->raid_level
= 1;
6006 /* replace dev<->dev_new */
6009 /* update disk order table */
6010 for (du
= super
->disks
; du
; du
= du
->next
)
6012 set_imsm_ord_tbl_ent(map
, du
->index
, du
->index
);
6013 for (du
= super
->missing
; du
; du
= du
->next
)
6015 set_imsm_ord_tbl_ent(map
, du
->index
,
6016 du
->index
| IMSM_ORD_REBUILD
);
6021 static void imsm_process_update(struct supertype
*st
,
6022 struct metadata_update
*update
)
6025 * crack open the metadata_update envelope to find the update record
6026 * update can be one of:
6027 * update_reshape_container_disks - all the arrays in the container
6028 * are being reshaped to have more devices. We need to mark
6029 * the arrays for general migration and convert selected spares
6030 * into active devices.
6031 * update_activate_spare - a spare device has replaced a failed
6032 * device in an array, update the disk_ord_tbl. If this disk is
6033 * present in all member arrays then also clear the SPARE_DISK
6035 * update_create_array
6037 * update_rename_array
6038 * update_add_remove_disk
6040 struct intel_super
*super
= st
->sb
;
6041 struct imsm_super
*mpb
;
6042 enum imsm_update_type type
= *(enum imsm_update_type
*) update
->buf
;
6044 /* update requires a larger buf but the allocation failed */
6045 if (super
->next_len
&& !super
->next_buf
) {
6046 super
->next_len
= 0;
6050 if (super
->next_buf
) {
6051 memcpy(super
->next_buf
, super
->buf
, super
->len
);
6053 super
->len
= super
->next_len
;
6054 super
->buf
= super
->next_buf
;
6056 super
->next_len
= 0;
6057 super
->next_buf
= NULL
;
6060 mpb
= super
->anchor
;
6063 case update_takeover
: {
6064 struct imsm_update_takeover
*u
= (void *)update
->buf
;
6065 if (apply_takeover_update(u
, super
, &update
->space_list
))
6066 super
->updates_pending
++;
6070 case update_reshape_container_disks
: {
6071 struct imsm_update_reshape
*u
= (void *)update
->buf
;
6072 if (apply_reshape_container_disks_update(
6073 u
, super
, &update
->space_list
))
6074 super
->updates_pending
++;
6077 case update_activate_spare
: {
6078 struct imsm_update_activate_spare
*u
= (void *) update
->buf
;
6079 struct imsm_dev
*dev
= get_imsm_dev(super
, u
->array
);
6080 struct imsm_map
*map
= get_imsm_map(dev
, 0);
6081 struct imsm_map
*migr_map
;
6082 struct active_array
*a
;
6083 struct imsm_disk
*disk
;
6088 int victim
= get_imsm_disk_idx(dev
, u
->slot
, -1);
6091 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
6096 fprintf(stderr
, "error: imsm_activate_spare passed "
6097 "an unknown disk (index: %d)\n",
6102 super
->updates_pending
++;
6104 /* count failures (excluding rebuilds and the victim)
6105 * to determine map[0] state
6108 for (i
= 0; i
< map
->num_members
; i
++) {
6111 disk
= get_imsm_disk(super
,
6112 get_imsm_disk_idx(dev
, i
, -1));
6113 if (!disk
|| is_failed(disk
))
6117 /* adding a pristine spare, assign a new index */
6118 if (dl
->index
< 0) {
6119 dl
->index
= super
->anchor
->num_disks
;
6120 super
->anchor
->num_disks
++;
6123 disk
->status
|= CONFIGURED_DISK
;
6124 disk
->status
&= ~SPARE_DISK
;
6127 to_state
= imsm_check_degraded(super
, dev
, failed
);
6128 map
->map_state
= IMSM_T_STATE_DEGRADED
;
6129 migrate(dev
, to_state
, MIGR_REBUILD
);
6130 migr_map
= get_imsm_map(dev
, 1);
6131 set_imsm_ord_tbl_ent(map
, u
->slot
, dl
->index
);
6132 set_imsm_ord_tbl_ent(migr_map
, u
->slot
, dl
->index
| IMSM_ORD_REBUILD
);
6134 /* update the family_num to mark a new container
6135 * generation, being careful to record the existing
6136 * family_num in orig_family_num to clean up after
6137 * earlier mdadm versions that neglected to set it.
6139 if (mpb
->orig_family_num
== 0)
6140 mpb
->orig_family_num
= mpb
->family_num
;
6141 mpb
->family_num
+= super
->random
;
6143 /* count arrays using the victim in the metadata */
6145 for (a
= st
->arrays
; a
; a
= a
->next
) {
6146 dev
= get_imsm_dev(super
, a
->info
.container_member
);
6147 map
= get_imsm_map(dev
, 0);
6149 if (get_imsm_disk_slot(map
, victim
) >= 0)
6153 /* delete the victim if it is no longer being
6159 /* We know that 'manager' isn't touching anything,
6160 * so it is safe to delete
6162 for (dlp
= &super
->disks
; *dlp
; dlp
= &(*dlp
)->next
)
6163 if ((*dlp
)->index
== victim
)
6166 /* victim may be on the missing list */
6168 for (dlp
= &super
->missing
; *dlp
; dlp
= &(*dlp
)->next
)
6169 if ((*dlp
)->index
== victim
)
6171 imsm_delete(super
, dlp
, victim
);
6175 case update_create_array
: {
6176 /* someone wants to create a new array, we need to be aware of
6177 * a few races/collisions:
6178 * 1/ 'Create' called by two separate instances of mdadm
6179 * 2/ 'Create' versus 'activate_spare': mdadm has chosen
6180 * devices that have since been assimilated via
6182 * In the event this update can not be carried out mdadm will
6183 * (FIX ME) notice that its update did not take hold.
6185 struct imsm_update_create_array
*u
= (void *) update
->buf
;
6186 struct intel_dev
*dv
;
6187 struct imsm_dev
*dev
;
6188 struct imsm_map
*map
, *new_map
;
6189 unsigned long long start
, end
;
6190 unsigned long long new_start
, new_end
;
6192 struct disk_info
*inf
;
6195 /* handle racing creates: first come first serve */
6196 if (u
->dev_idx
< mpb
->num_raid_devs
) {
6197 dprintf("%s: subarray %d already defined\n",
6198 __func__
, u
->dev_idx
);
6202 /* check update is next in sequence */
6203 if (u
->dev_idx
!= mpb
->num_raid_devs
) {
6204 dprintf("%s: can not create array %d expected index %d\n",
6205 __func__
, u
->dev_idx
, mpb
->num_raid_devs
);
6209 new_map
= get_imsm_map(&u
->dev
, 0);
6210 new_start
= __le32_to_cpu(new_map
->pba_of_lba0
);
6211 new_end
= new_start
+ __le32_to_cpu(new_map
->blocks_per_member
);
6212 inf
= get_disk_info(u
);
6214 /* handle activate_spare versus create race:
6215 * check to make sure that overlapping arrays do not include
6218 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
6219 dev
= get_imsm_dev(super
, i
);
6220 map
= get_imsm_map(dev
, 0);
6221 start
= __le32_to_cpu(map
->pba_of_lba0
);
6222 end
= start
+ __le32_to_cpu(map
->blocks_per_member
);
6223 if ((new_start
>= start
&& new_start
<= end
) ||
6224 (start
>= new_start
&& start
<= new_end
))
6229 if (disks_overlap(super
, i
, u
)) {
6230 dprintf("%s: arrays overlap\n", __func__
);
6235 /* check that prepare update was successful */
6236 if (!update
->space
) {
6237 dprintf("%s: prepare update failed\n", __func__
);
6241 /* check that all disks are still active before committing
6242 * changes. FIXME: could we instead handle this by creating a
6243 * degraded array? That's probably not what the user expects,
6244 * so better to drop this update on the floor.
6246 for (i
= 0; i
< new_map
->num_members
; i
++) {
6247 dl
= serial_to_dl(inf
[i
].serial
, super
);
6249 dprintf("%s: disk disappeared\n", __func__
);
6254 super
->updates_pending
++;
6256 /* convert spares to members and fixup ord_tbl */
6257 for (i
= 0; i
< new_map
->num_members
; i
++) {
6258 dl
= serial_to_dl(inf
[i
].serial
, super
);
6259 if (dl
->index
== -1) {
6260 dl
->index
= mpb
->num_disks
;
6262 dl
->disk
.status
|= CONFIGURED_DISK
;
6263 dl
->disk
.status
&= ~SPARE_DISK
;
6265 set_imsm_ord_tbl_ent(new_map
, i
, dl
->index
);
6270 update
->space
= NULL
;
6271 imsm_copy_dev(dev
, &u
->dev
);
6272 dv
->index
= u
->dev_idx
;
6273 dv
->next
= super
->devlist
;
6274 super
->devlist
= dv
;
6275 mpb
->num_raid_devs
++;
6277 imsm_update_version_info(super
);
6280 /* mdmon knows how to release update->space, but not
6281 * ((struct intel_dev *) update->space)->dev
6283 if (update
->space
) {
6289 case update_kill_array
: {
6290 struct imsm_update_kill_array
*u
= (void *) update
->buf
;
6291 int victim
= u
->dev_idx
;
6292 struct active_array
*a
;
6293 struct intel_dev
**dp
;
6294 struct imsm_dev
*dev
;
6296 /* sanity check that we are not affecting the uuid of
6297 * active arrays, or deleting an active array
6299 * FIXME when immutable ids are available, but note that
6300 * we'll also need to fixup the invalidated/active
6301 * subarray indexes in mdstat
6303 for (a
= st
->arrays
; a
; a
= a
->next
)
6304 if (a
->info
.container_member
>= victim
)
6306 /* by definition if mdmon is running at least one array
6307 * is active in the container, so checking
6308 * mpb->num_raid_devs is just extra paranoia
6310 dev
= get_imsm_dev(super
, victim
);
6311 if (a
|| !dev
|| mpb
->num_raid_devs
== 1) {
6312 dprintf("failed to delete subarray-%d\n", victim
);
6316 for (dp
= &super
->devlist
; *dp
;)
6317 if ((*dp
)->index
== (unsigned)super
->current_vol
) {
6320 if ((*dp
)->index
> (unsigned)victim
)
6324 mpb
->num_raid_devs
--;
6325 super
->updates_pending
++;
6328 case update_rename_array
: {
6329 struct imsm_update_rename_array
*u
= (void *) update
->buf
;
6330 char name
[MAX_RAID_SERIAL_LEN
+1];
6331 int target
= u
->dev_idx
;
6332 struct active_array
*a
;
6333 struct imsm_dev
*dev
;
6335 /* sanity check that we are not affecting the uuid of
6338 snprintf(name
, MAX_RAID_SERIAL_LEN
, "%s", (char *) u
->name
);
6339 name
[MAX_RAID_SERIAL_LEN
] = '\0';
6340 for (a
= st
->arrays
; a
; a
= a
->next
)
6341 if (a
->info
.container_member
== target
)
6343 dev
= get_imsm_dev(super
, u
->dev_idx
);
6344 if (a
|| !dev
|| !check_name(super
, name
, 1)) {
6345 dprintf("failed to rename subarray-%d\n", target
);
6349 snprintf((char *) dev
->volume
, MAX_RAID_SERIAL_LEN
, "%s", name
);
6350 super
->updates_pending
++;
6353 case update_add_remove_disk
: {
6354 /* we may be able to repair some arrays if disks are
6355 * being added, check teh status of add_remove_disk
6356 * if discs has been added.
6358 if (add_remove_disk_update(super
)) {
6359 struct active_array
*a
;
6361 super
->updates_pending
++;
6362 for (a
= st
->arrays
; a
; a
= a
->next
)
6363 a
->check_degraded
= 1;
6368 fprintf(stderr
, "error: unsuported process update type:"
6369 "(type: %d)\n", type
);
6373 static void imsm_prepare_update(struct supertype
*st
,
6374 struct metadata_update
*update
)
6377 * Allocate space to hold new disk entries, raid-device entries or a new
6378 * mpb if necessary. The manager synchronously waits for updates to
6379 * complete in the monitor, so new mpb buffers allocated here can be
6380 * integrated by the monitor thread without worrying about live pointers
6381 * in the manager thread.
6383 enum imsm_update_type type
= *(enum imsm_update_type
*) update
->buf
;
6384 struct intel_super
*super
= st
->sb
;
6385 struct imsm_super
*mpb
= super
->anchor
;
6390 case update_takeover
: {
6391 struct imsm_update_takeover
*u
= (void *)update
->buf
;
6392 if (u
->direction
== R0_TO_R10
) {
6393 void **tail
= (void **)&update
->space_list
;
6394 struct imsm_dev
*dev
= get_imsm_dev(super
, u
->subarray
);
6395 struct imsm_map
*map
= get_imsm_map(dev
, 0);
6396 int num_members
= map
->num_members
;
6400 /* allocate memory for added disks */
6401 for (i
= 0; i
< num_members
; i
++) {
6402 size
= sizeof(struct dl
);
6403 space
= malloc(size
);
6412 /* allocate memory for new device */
6413 size
= sizeof_imsm_dev(super
->devlist
->dev
, 0) +
6414 (num_members
* sizeof(__u32
));
6415 space
= malloc(size
);
6424 len
= disks_to_mpb_size(num_members
* 2);
6426 /* if allocation didn't success, free buffer */
6427 while (update
->space_list
) {
6428 void **sp
= update
->space_list
;
6429 update
->space_list
= *sp
;
6437 case update_reshape_container_disks
: {
6438 /* Every raid device in the container is about to
6439 * gain some more devices, and we will enter a
6441 * So each 'imsm_map' will be bigger, and the imsm_vol
6442 * will now hold 2 of them.
6443 * Thus we need new 'struct imsm_dev' allocations sized
6444 * as sizeof_imsm_dev but with more devices in both maps.
6446 struct imsm_update_reshape
*u
= (void *)update
->buf
;
6447 struct intel_dev
*dl
;
6448 void **space_tail
= (void**)&update
->space_list
;
6450 dprintf("imsm: imsm_prepare_update() for update_reshape\n");
6452 for (dl
= super
->devlist
; dl
; dl
= dl
->next
) {
6453 int size
= sizeof_imsm_dev(dl
->dev
, 1);
6455 if (u
->new_raid_disks
> u
->old_raid_disks
)
6456 size
+= sizeof(__u32
)*2*
6457 (u
->new_raid_disks
- u
->old_raid_disks
);
6466 len
= disks_to_mpb_size(u
->new_raid_disks
);
6467 dprintf("New anchor length is %llu\n", (unsigned long long)len
);
6470 case update_create_array
: {
6471 struct imsm_update_create_array
*u
= (void *) update
->buf
;
6472 struct intel_dev
*dv
;
6473 struct imsm_dev
*dev
= &u
->dev
;
6474 struct imsm_map
*map
= get_imsm_map(dev
, 0);
6476 struct disk_info
*inf
;
6480 inf
= get_disk_info(u
);
6481 len
= sizeof_imsm_dev(dev
, 1);
6482 /* allocate a new super->devlist entry */
6483 dv
= malloc(sizeof(*dv
));
6485 dv
->dev
= malloc(len
);
6490 update
->space
= NULL
;
6494 /* count how many spares will be converted to members */
6495 for (i
= 0; i
< map
->num_members
; i
++) {
6496 dl
= serial_to_dl(inf
[i
].serial
, super
);
6498 /* hmm maybe it failed?, nothing we can do about
6503 if (count_memberships(dl
, super
) == 0)
6506 len
+= activate
* sizeof(struct imsm_disk
);
6513 /* check if we need a larger metadata buffer */
6514 if (super
->next_buf
)
6515 buf_len
= super
->next_len
;
6517 buf_len
= super
->len
;
6519 if (__le32_to_cpu(mpb
->mpb_size
) + len
> buf_len
) {
6520 /* ok we need a larger buf than what is currently allocated
6521 * if this allocation fails process_update will notice that
6522 * ->next_len is set and ->next_buf is NULL
6524 buf_len
= ROUND_UP(__le32_to_cpu(mpb
->mpb_size
) + len
, 512);
6525 if (super
->next_buf
)
6526 free(super
->next_buf
);
6528 super
->next_len
= buf_len
;
6529 if (posix_memalign(&super
->next_buf
, 512, buf_len
) == 0)
6530 memset(super
->next_buf
, 0, buf_len
);
6532 super
->next_buf
= NULL
;
6536 /* must be called while manager is quiesced */
6537 static void imsm_delete(struct intel_super
*super
, struct dl
**dlp
, unsigned index
)
6539 struct imsm_super
*mpb
= super
->anchor
;
6541 struct imsm_dev
*dev
;
6542 struct imsm_map
*map
;
6543 int i
, j
, num_members
;
6546 dprintf("%s: deleting device[%d] from imsm_super\n",
6549 /* shift all indexes down one */
6550 for (iter
= super
->disks
; iter
; iter
= iter
->next
)
6551 if (iter
->index
> (int)index
)
6553 for (iter
= super
->missing
; iter
; iter
= iter
->next
)
6554 if (iter
->index
> (int)index
)
6557 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
6558 dev
= get_imsm_dev(super
, i
);
6559 map
= get_imsm_map(dev
, 0);
6560 num_members
= map
->num_members
;
6561 for (j
= 0; j
< num_members
; j
++) {
6562 /* update ord entries being careful not to propagate
6563 * ord-flags to the first map
6565 ord
= get_imsm_ord_tbl_ent(dev
, j
, -1);
6567 if (ord_to_idx(ord
) <= index
)
6570 map
= get_imsm_map(dev
, 0);
6571 set_imsm_ord_tbl_ent(map
, j
, ord_to_idx(ord
- 1));
6572 map
= get_imsm_map(dev
, 1);
6574 set_imsm_ord_tbl_ent(map
, j
, ord
- 1);
6579 super
->updates_pending
++;
6581 struct dl
*dl
= *dlp
;
6583 *dlp
= (*dlp
)->next
;
6584 __free_imsm_disk(dl
);
6587 #endif /* MDASSEMBLE */
6589 static char disk_by_path
[] = "/dev/disk/by-path/";
6591 static const char *imsm_get_disk_controller_domain(const char *path
)
6593 struct sys_dev
*list
, *hba
= NULL
;
6594 char disk_path
[PATH_MAX
];
6598 list
= find_driver_devices("pci", "ahci");
6599 for (hba
= list
; hba
; hba
= hba
->next
)
6600 if (devpath_to_vendor(hba
->path
) == 0x8086)
6606 strncpy(disk_path
, disk_by_path
, PATH_MAX
- 1);
6607 strncat(disk_path
, path
, PATH_MAX
- strlen(disk_path
) - 1);
6608 if (stat(disk_path
, &st
) == 0) {
6609 dpath
= devt_to_devpath(st
.st_rdev
);
6611 ahci
= path_attached_to_hba(dpath
, hba
->path
);
6614 dprintf("path: %s(%s) hba: %s attached: %d\n",
6615 path
, dpath
, (hba
) ? hba
->path
: "NULL", ahci
);
6616 free_sys_dev(&list
);
6623 static int imsm_find_array_minor_by_subdev(int subdev
, int container
, int *minor
)
6625 char subdev_name
[20];
6626 struct mdstat_ent
*mdstat
;
6628 sprintf(subdev_name
, "%d", subdev
);
6629 mdstat
= mdstat_by_subdev(subdev_name
, container
);
6633 *minor
= mdstat
->devnum
;
6634 free_mdstat(mdstat
);
6638 static int imsm_reshape_is_allowed_on_container(struct supertype
*st
,
6639 struct geo_params
*geo
,
6640 int *old_raid_disks
)
6642 /* currently we only support increasing the number of devices
6643 * for a container. This increases the number of device for each
6644 * member array. They must all be RAID0 or RAID5.
6647 struct mdinfo
*info
, *member
;
6648 int devices_that_can_grow
= 0;
6650 dprintf("imsm: imsm_reshape_is_allowed_on_container(ENTER): "
6651 "st->devnum = (%i)\n",
6654 if (geo
->size
!= -1 ||
6655 geo
->level
!= UnSet
||
6656 geo
->layout
!= UnSet
||
6657 geo
->chunksize
!= 0 ||
6658 geo
->raid_disks
== UnSet
) {
6659 dprintf("imsm: Container operation is allowed for "
6660 "raid disks number change only.\n");
6664 info
= container_content_imsm(st
, NULL
);
6665 for (member
= info
; member
; member
= member
->next
) {
6669 dprintf("imsm: checking device_num: %i\n",
6670 member
->container_member
);
6672 if (geo
->raid_disks
< member
->array
.raid_disks
) {
6673 /* we work on container for Online Capacity Expansion
6674 * only so raid_disks has to grow
6676 dprintf("imsm: for container operation raid disks "
6677 "increase is required\n");
6681 if ((info
->array
.level
!= 0) &&
6682 (info
->array
.level
!= 5)) {
6683 /* we cannot use this container with other raid level
6685 dprintf("imsm: for container operation wrong"
6686 " raid level (%i) detected\n",
6690 /* check for platform support
6691 * for this raid level configuration
6693 struct intel_super
*super
= st
->sb
;
6694 if (!is_raid_level_supported(super
->orom
,
6695 member
->array
.level
,
6697 dprintf("platform does not support raid%d with"
6701 geo
->raid_disks
> 1 ? "s" : "");
6706 if (*old_raid_disks
&&
6707 info
->array
.raid_disks
!= *old_raid_disks
)
6709 *old_raid_disks
= info
->array
.raid_disks
;
6711 /* All raid5 and raid0 volumes in container
6712 * have to be ready for Online Capacity Expansion
6713 * so they need to be assembled. We have already
6714 * checked that no recovery etc is happening.
6716 result
= imsm_find_array_minor_by_subdev(member
->container_member
,
6720 dprintf("imsm: cannot find array\n");
6723 devices_that_can_grow
++;
6726 if (!member
&& devices_that_can_grow
)
6730 dprintf("\tContainer operation allowed\n");
6732 dprintf("\tError: %i\n", ret_val
);
6737 /* Function: get_spares_for_grow
6738 * Description: Allocates memory and creates list of spare devices
6739 * avaliable in container. Checks if spare drive size is acceptable.
6740 * Parameters: Pointer to the supertype structure
6741 * Returns: Pointer to the list of spare devices (mdinfo structure) on success,
6744 static struct mdinfo
*get_spares_for_grow(struct supertype
*st
)
6746 unsigned long long min_size
= min_acceptable_spare_size_imsm(st
);
6747 return container_choose_spares(st
, min_size
, NULL
, NULL
, NULL
, 0);
6750 /******************************************************************************
6751 * function: imsm_create_metadata_update_for_reshape
6752 * Function creates update for whole IMSM container.
6754 ******************************************************************************/
6755 static int imsm_create_metadata_update_for_reshape(
6756 struct supertype
*st
,
6757 struct geo_params
*geo
,
6759 struct imsm_update_reshape
**updatep
)
6761 struct intel_super
*super
= st
->sb
;
6762 struct imsm_super
*mpb
= super
->anchor
;
6763 int update_memory_size
= 0;
6764 struct imsm_update_reshape
*u
= NULL
;
6765 struct mdinfo
*spares
= NULL
;
6767 int delta_disks
= 0;
6770 dprintf("imsm_update_metadata_for_reshape(enter) raid_disks = %i\n",
6773 delta_disks
= geo
->raid_disks
- old_raid_disks
;
6775 /* size of all update data without anchor */
6776 update_memory_size
= sizeof(struct imsm_update_reshape
);
6778 /* now add space for spare disks that we need to add. */
6779 update_memory_size
+= sizeof(u
->new_disks
[0]) * (delta_disks
- 1);
6781 u
= calloc(1, update_memory_size
);
6784 "cannot get memory for imsm_update_reshape update\n");
6787 u
->type
= update_reshape_container_disks
;
6788 u
->old_raid_disks
= old_raid_disks
;
6789 u
->new_raid_disks
= geo
->raid_disks
;
6791 /* now get spare disks list
6793 spares
= get_spares_for_grow(st
);
6796 || delta_disks
> spares
->array
.spare_disks
) {
6797 dprintf("imsm: ERROR: Cannot get spare devices.\n");
6801 /* we have got spares
6802 * update disk list in imsm_disk list table in anchor
6804 dprintf("imsm: %i spares are available.\n\n",
6805 spares
->array
.spare_disks
);
6808 for (i
= 0; i
< delta_disks
; i
++) {
6813 u
->new_disks
[i
] = makedev(dev
->disk
.major
,
6815 dl
= get_disk_super(super
, dev
->disk
.major
, dev
->disk
.minor
);
6816 dl
->index
= mpb
->num_disks
;
6826 dprintf("imsm: reshape update preparation :");
6827 if (i
== delta_disks
) {
6830 return update_memory_size
;
6833 dprintf(" Error\n");
6838 static void imsm_update_metadata_locally(struct supertype
*st
,
6841 struct metadata_update mu
;
6846 mu
.space_list
= NULL
;
6848 imsm_prepare_update(st
, &mu
);
6849 imsm_process_update(st
, &mu
);
6851 while (mu
.space_list
) {
6852 void **space
= mu
.space_list
;
6853 mu
.space_list
= *space
;
6858 /***************************************************************************
6859 * Function: imsm_analyze_change
6860 * Description: Function analyze change for single volume
6861 * and validate if transition is supported
6862 * Parameters: Geometry parameters, supertype structure
6863 * Returns: Operation type code on success, -1 if fail
6864 ****************************************************************************/
6865 enum imsm_reshape_type
imsm_analyze_change(struct supertype
*st
,
6866 struct geo_params
*geo
)
6872 getinfo_super_imsm_volume(st
, &info
, NULL
);
6874 if ((geo
->level
!= info
.array
.level
) &&
6875 (geo
->level
>= 0) &&
6876 (geo
->level
!= UnSet
)) {
6877 switch (info
.array
.level
) {
6879 if (geo
->level
== 5) {
6880 change
= CH_LEVEL_MIGRATION
;
6883 if (geo
->level
== 10) {
6884 change
= CH_TAKEOVER
;
6889 if (geo
->level
!= 0)
6890 change
= CH_LEVEL_MIGRATION
;
6893 if (geo
->level
== 0) {
6894 change
= CH_TAKEOVER
;
6901 Name
" Error. Level Migration from %d to %d "
6903 info
.array
.level
, geo
->level
);
6904 goto analyse_change_exit
;
6907 geo
->level
= info
.array
.level
;
6909 if ((geo
->layout
!= info
.array
.layout
)
6910 && ((geo
->layout
!= UnSet
) && (geo
->layout
!= -1))) {
6911 change
= CH_LEVEL_MIGRATION
;
6912 if ((info
.array
.layout
== 0)
6913 && (info
.array
.level
== 5)
6914 && (geo
->layout
== 5)) {
6915 /* reshape 5 -> 4 */
6916 } else if ((info
.array
.layout
== 5)
6917 && (info
.array
.level
== 5)
6918 && (geo
->layout
== 0)) {
6919 /* reshape 4 -> 5 */
6924 Name
" Error. Layout Migration from %d to %d "
6926 info
.array
.layout
, geo
->layout
);
6928 goto analyse_change_exit
;
6931 geo
->layout
= info
.array
.layout
;
6933 if ((geo
->chunksize
> 0) && (geo
->chunksize
!= UnSet
)
6934 && (geo
->chunksize
!= info
.array
.chunk_size
))
6935 change
= CH_CHUNK_MIGR
;
6937 geo
->chunksize
= info
.array
.chunk_size
;
6939 if (!validate_geometry_imsm(st
,
6943 (geo
->chunksize
/ 1024),
6949 struct intel_super
*super
= st
->sb
;
6950 struct imsm_super
*mpb
= super
->anchor
;
6952 if (mpb
->num_raid_devs
> 1) {
6954 Name
" Error. Cannot perform operation on %s"
6955 "- for this operation it MUST be single "
6956 "array in container\n",
6962 analyse_change_exit
:
6967 int imsm_takeover(struct supertype
*st
, struct geo_params
*geo
)
6969 struct intel_super
*super
= st
->sb
;
6970 struct imsm_update_takeover
*u
;
6972 u
= malloc(sizeof(struct imsm_update_takeover
));
6976 u
->type
= update_takeover
;
6977 u
->subarray
= super
->current_vol
;
6979 /* 10->0 transition */
6980 if (geo
->level
== 0)
6981 u
->direction
= R10_TO_R0
;
6983 /* 0->10 transition */
6984 if (geo
->level
== 10)
6985 u
->direction
= R0_TO_R10
;
6987 /* update metadata locally */
6988 imsm_update_metadata_locally(st
, u
,
6989 sizeof(struct imsm_update_takeover
));
6990 /* and possibly remotely */
6991 if (st
->update_tail
)
6992 append_metadata_update(st
, u
,
6993 sizeof(struct imsm_update_takeover
));
7000 static int imsm_reshape_super(struct supertype
*st
, long long size
, int level
,
7001 int layout
, int chunksize
, int raid_disks
,
7002 char *backup
, char *dev
, int verbose
)
7005 struct geo_params geo
;
7007 dprintf("imsm: reshape_super called.\n");
7009 memset(&geo
, sizeof(struct geo_params
), 0);
7012 geo
.dev_id
= st
->devnum
;
7015 geo
.layout
= layout
;
7016 geo
.chunksize
= chunksize
;
7017 geo
.raid_disks
= raid_disks
;
7019 dprintf("\tfor level : %i\n", geo
.level
);
7020 dprintf("\tfor raid_disks : %i\n", geo
.raid_disks
);
7022 if (experimental() == 0)
7025 if (st
->container_dev
== st
->devnum
) {
7026 /* On container level we can only increase number of devices. */
7027 dprintf("imsm: info: Container operation\n");
7028 int old_raid_disks
= 0;
7029 if (imsm_reshape_is_allowed_on_container(
7030 st
, &geo
, &old_raid_disks
)) {
7031 struct imsm_update_reshape
*u
= NULL
;
7034 len
= imsm_create_metadata_update_for_reshape(
7035 st
, &geo
, old_raid_disks
, &u
);
7038 dprintf("imsm: Cannot prepare update\n");
7039 goto exit_imsm_reshape_super
;
7043 /* update metadata locally */
7044 imsm_update_metadata_locally(st
, u
, len
);
7045 /* and possibly remotely */
7046 if (st
->update_tail
)
7047 append_metadata_update(st
, u
, len
);
7052 fprintf(stderr
, Name
"imsm: Operation is not allowed "
7053 "on this container\n");
7056 /* On volume level we support following operations
7057 * - takeover: raid10 -> raid0; raid0 -> raid10
7058 * - chunk size migration
7059 * - migration: raid5 -> raid0; raid0 -> raid5
7061 struct intel_super
*super
= st
->sb
;
7062 struct intel_dev
*dev
= super
->devlist
;
7064 dprintf("imsm: info: Volume operation\n");
7065 /* find requested device */
7067 imsm_find_array_minor_by_subdev(dev
->index
, st
->container_dev
, &devnum
);
7068 if (devnum
== geo
.dev_id
)
7073 fprintf(stderr
, Name
" Cannot find %s (%i) subarray\n",
7074 geo
.dev_name
, geo
.dev_id
);
7075 goto exit_imsm_reshape_super
;
7077 super
->current_vol
= dev
->index
;
7078 change
= imsm_analyze_change(st
, &geo
);
7081 ret_val
= imsm_takeover(st
, &geo
);
7086 case CH_LEVEL_MIGRATION
:
7094 exit_imsm_reshape_super
:
7095 dprintf("imsm: reshape_super Exit code = %i\n", ret_val
);
7099 static int imsm_manage_reshape(
7100 int afd
, struct mdinfo
*sra
, struct reshape
*reshape
,
7101 struct supertype
*st
, unsigned long stripes
,
7102 int *fds
, unsigned long long *offsets
,
7103 int dests
, int *destfd
, unsigned long long *destoffsets
)
7105 /* Just use child_monitor for now */
7106 return child_monitor(
7107 afd
, sra
, reshape
, st
, stripes
,
7108 fds
, offsets
, dests
, destfd
, destoffsets
);
7111 struct superswitch super_imsm
= {
7113 .examine_super
= examine_super_imsm
,
7114 .brief_examine_super
= brief_examine_super_imsm
,
7115 .brief_examine_subarrays
= brief_examine_subarrays_imsm
,
7116 .export_examine_super
= export_examine_super_imsm
,
7117 .detail_super
= detail_super_imsm
,
7118 .brief_detail_super
= brief_detail_super_imsm
,
7119 .write_init_super
= write_init_super_imsm
,
7120 .validate_geometry
= validate_geometry_imsm
,
7121 .add_to_super
= add_to_super_imsm
,
7122 .remove_from_super
= remove_from_super_imsm
,
7123 .detail_platform
= detail_platform_imsm
,
7124 .kill_subarray
= kill_subarray_imsm
,
7125 .update_subarray
= update_subarray_imsm
,
7126 .load_container
= load_container_imsm
,
7128 .match_home
= match_home_imsm
,
7129 .uuid_from_super
= uuid_from_super_imsm
,
7130 .getinfo_super
= getinfo_super_imsm
,
7131 .getinfo_super_disks
= getinfo_super_disks_imsm
,
7132 .update_super
= update_super_imsm
,
7134 .avail_size
= avail_size_imsm
,
7135 .min_acceptable_spare_size
= min_acceptable_spare_size_imsm
,
7137 .compare_super
= compare_super_imsm
,
7139 .load_super
= load_super_imsm
,
7140 .init_super
= init_super_imsm
,
7141 .store_super
= store_super_imsm
,
7142 .free_super
= free_super_imsm
,
7143 .match_metadata_desc
= match_metadata_desc_imsm
,
7144 .container_content
= container_content_imsm
,
7145 .default_geometry
= default_geometry_imsm
,
7146 .get_disk_controller_domain
= imsm_get_disk_controller_domain
,
7147 .reshape_super
= imsm_reshape_super
,
7148 .manage_reshape
= imsm_manage_reshape
,
7155 .open_new
= imsm_open_new
,
7156 .set_array_state
= imsm_set_array_state
,
7157 .set_disk
= imsm_set_disk
,
7158 .sync_metadata
= imsm_sync_metadata
,
7159 .activate_spare
= imsm_activate_spare
,
7160 .process_update
= imsm_process_update
,
7161 .prepare_update
= imsm_prepare_update
,
7162 #endif /* MDASSEMBLE */