2 * mdadm - Intel(R) Matrix Storage Manager Support
4 * Copyright (C) 2002-2008 Intel Corporation
6 * This program is free software; you can redistribute it and/or modify it
7 * under the terms and conditions of the GNU General Public License,
8 * version 2, as published by the Free Software Foundation.
10 * This program is distributed in the hope it will be useful, but WITHOUT
11 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
12 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for
15 * You should have received a copy of the GNU General Public License along with
16 * this program; if not, write to the Free Software Foundation, Inc.,
17 * 51 Franklin St - Fifth Floor, Boston, MA 02110-1301 USA.
20 #define HAVE_STDINT_H 1
24 #include "platform-intel.h"
30 /* MPB == Metadata Parameter Block */
31 #define MPB_SIGNATURE "Intel Raid ISM Cfg Sig. "
32 #define MPB_SIG_LEN (strlen(MPB_SIGNATURE))
33 #define MPB_VERSION_RAID0 "1.0.00"
34 #define MPB_VERSION_RAID1 "1.1.00"
35 #define MPB_VERSION_MANY_VOLUMES_PER_ARRAY "1.2.00"
36 #define MPB_VERSION_3OR4_DISK_ARRAY "1.2.01"
37 #define MPB_VERSION_RAID5 "1.2.02"
38 #define MPB_VERSION_5OR6_DISK_ARRAY "1.2.04"
39 #define MPB_VERSION_CNG "1.2.06"
40 #define MPB_VERSION_ATTRIBS "1.3.00"
41 #define MAX_SIGNATURE_LENGTH 32
42 #define MAX_RAID_SERIAL_LEN 16
44 #define MPB_ATTRIB_CHECKSUM_VERIFY __cpu_to_le32(0x80000000)
45 #define MPB_ATTRIB_PM __cpu_to_le32(0x40000000)
46 #define MPB_ATTRIB_2TB __cpu_to_le32(0x20000000)
47 #define MPB_ATTRIB_RAID0 __cpu_to_le32(0x00000001)
48 #define MPB_ATTRIB_RAID1 __cpu_to_le32(0x00000002)
49 #define MPB_ATTRIB_RAID10 __cpu_to_le32(0x00000004)
50 #define MPB_ATTRIB_RAID1E __cpu_to_le32(0x00000008)
51 #define MPB_ATTRIB_RAID5 __cpu_to_le32(0x00000010)
52 #define MPB_ATTRIB_RAIDCNG __cpu_to_le32(0x00000020)
54 #define MPB_SECTOR_CNT 418
55 #define IMSM_RESERVED_SECTORS 4096
56 #define SECT_PER_MB_SHIFT 11
58 /* Disk configuration info. */
59 #define IMSM_MAX_DEVICES 255
61 __u8 serial
[MAX_RAID_SERIAL_LEN
];/* 0xD8 - 0xE7 ascii serial number */
62 __u32 total_blocks
; /* 0xE8 - 0xEB total blocks */
63 __u32 scsi_id
; /* 0xEC - 0xEF scsi ID */
64 #define SPARE_DISK __cpu_to_le32(0x01) /* Spare */
65 #define CONFIGURED_DISK __cpu_to_le32(0x02) /* Member of some RaidDev */
66 #define FAILED_DISK __cpu_to_le32(0x04) /* Permanent failure */
67 __u32 status
; /* 0xF0 - 0xF3 */
68 __u32 owner_cfg_num
; /* which config 0,1,2... owns this disk */
69 #define IMSM_DISK_FILLERS 4
70 __u32 filler
[IMSM_DISK_FILLERS
]; /* 0xF4 - 0x107 MPB_DISK_FILLERS for future expansion */
73 /* RAID map configuration infos. */
75 __u32 pba_of_lba0
; /* start address of partition */
76 __u32 blocks_per_member
;/* blocks per member */
77 __u32 num_data_stripes
; /* number of data stripes */
78 __u16 blocks_per_strip
;
79 __u8 map_state
; /* Normal, Uninitialized, Degraded, Failed */
80 #define IMSM_T_STATE_NORMAL 0
81 #define IMSM_T_STATE_UNINITIALIZED 1
82 #define IMSM_T_STATE_DEGRADED 2
83 #define IMSM_T_STATE_FAILED 3
85 #define IMSM_T_RAID0 0
86 #define IMSM_T_RAID1 1
87 #define IMSM_T_RAID5 5 /* since metadata version 1.2.02 ? */
88 __u8 num_members
; /* number of member disks */
89 __u8 num_domains
; /* number of parity domains */
90 __u8 failed_disk_num
; /* valid only when state is degraded */
92 __u32 filler
[7]; /* expansion area */
93 #define IMSM_ORD_REBUILD (1 << 24)
94 __u32 disk_ord_tbl
[1]; /* disk_ord_tbl[num_members],
95 * top byte contains some flags
97 } __attribute__ ((packed
));
100 __u32 curr_migr_unit
;
101 __u32 checkpoint_id
; /* id to access curr_migr_unit */
102 __u8 migr_state
; /* Normal or Migrating */
104 #define MIGR_REBUILD 1
105 #define MIGR_VERIFY 2 /* analagous to echo check > sync_action */
106 #define MIGR_GEN_MIGR 3
107 #define MIGR_STATE_CHANGE 4
108 #define MIGR_REPAIR 5
109 __u8 migr_type
; /* Initializing, Rebuilding, ... */
111 __u8 fs_state
; /* fast-sync state for CnG (0xff == disabled) */
112 __u16 verify_errors
; /* number of mismatches */
113 __u16 bad_blocks
; /* number of bad blocks during verify */
115 struct imsm_map map
[1];
116 /* here comes another one if migr_state */
117 } __attribute__ ((packed
));
120 __u8 volume
[MAX_RAID_SERIAL_LEN
];
123 #define DEV_BOOTABLE __cpu_to_le32(0x01)
124 #define DEV_BOOT_DEVICE __cpu_to_le32(0x02)
125 #define DEV_READ_COALESCING __cpu_to_le32(0x04)
126 #define DEV_WRITE_COALESCING __cpu_to_le32(0x08)
127 #define DEV_LAST_SHUTDOWN_DIRTY __cpu_to_le32(0x10)
128 #define DEV_HIDDEN_AT_BOOT __cpu_to_le32(0x20)
129 #define DEV_CURRENTLY_HIDDEN __cpu_to_le32(0x40)
130 #define DEV_VERIFY_AND_FIX __cpu_to_le32(0x80)
131 #define DEV_MAP_STATE_UNINIT __cpu_to_le32(0x100)
132 #define DEV_NO_AUTO_RECOVERY __cpu_to_le32(0x200)
133 #define DEV_CLONE_N_GO __cpu_to_le32(0x400)
134 #define DEV_CLONE_MAN_SYNC __cpu_to_le32(0x800)
135 #define DEV_CNG_MASTER_DISK_NUM __cpu_to_le32(0x1000)
136 __u32 status
; /* Persistent RaidDev status */
137 __u32 reserved_blocks
; /* Reserved blocks at beginning of volume */
141 __u8 cng_master_disk
;
145 #define IMSM_DEV_FILLERS 10
146 __u32 filler
[IMSM_DEV_FILLERS
];
148 } __attribute__ ((packed
));
151 __u8 sig
[MAX_SIGNATURE_LENGTH
]; /* 0x00 - 0x1F */
152 __u32 check_sum
; /* 0x20 - 0x23 MPB Checksum */
153 __u32 mpb_size
; /* 0x24 - 0x27 Size of MPB */
154 __u32 family_num
; /* 0x28 - 0x2B Checksum from first time this config was written */
155 __u32 generation_num
; /* 0x2C - 0x2F Incremented each time this array's MPB is written */
156 __u32 error_log_size
; /* 0x30 - 0x33 in bytes */
157 __u32 attributes
; /* 0x34 - 0x37 */
158 __u8 num_disks
; /* 0x38 Number of configured disks */
159 __u8 num_raid_devs
; /* 0x39 Number of configured volumes */
160 __u8 error_log_pos
; /* 0x3A */
161 __u8 fill
[1]; /* 0x3B */
162 __u32 cache_size
; /* 0x3c - 0x40 in mb */
163 __u32 orig_family_num
; /* 0x40 - 0x43 original family num */
164 __u32 pwr_cycle_count
; /* 0x44 - 0x47 simulated power cycle count for array */
165 __u32 bbm_log_size
; /* 0x48 - 0x4B - size of bad Block Mgmt Log in bytes */
166 #define IMSM_FILLERS 35
167 __u32 filler
[IMSM_FILLERS
]; /* 0x4C - 0xD7 RAID_MPB_FILLERS */
168 struct imsm_disk disk
[1]; /* 0xD8 diskTbl[numDisks] */
169 /* here comes imsm_dev[num_raid_devs] */
170 /* here comes BBM logs */
171 } __attribute__ ((packed
));
173 #define BBM_LOG_MAX_ENTRIES 254
175 struct bbm_log_entry
{
176 __u64 defective_block_start
;
177 #define UNREADABLE 0xFFFFFFFF
178 __u32 spare_block_offset
;
179 __u16 remapped_marked_count
;
181 } __attribute__ ((__packed__
));
184 __u32 signature
; /* 0xABADB10C */
186 __u32 reserved_spare_block_count
; /* 0 */
187 __u32 reserved
; /* 0xFFFF */
188 __u64 first_spare_lba
;
189 struct bbm_log_entry mapped_block_entries
[BBM_LOG_MAX_ENTRIES
];
190 } __attribute__ ((__packed__
));
194 static char *map_state_str
[] = { "normal", "uninitialized", "degraded", "failed" };
197 static __u8
migr_type(struct imsm_dev
*dev
)
199 if (dev
->vol
.migr_type
== MIGR_VERIFY
&&
200 dev
->status
& DEV_VERIFY_AND_FIX
)
203 return dev
->vol
.migr_type
;
206 static void set_migr_type(struct imsm_dev
*dev
, __u8 migr_type
)
208 /* for compatibility with older oroms convert MIGR_REPAIR, into
209 * MIGR_VERIFY w/ DEV_VERIFY_AND_FIX status
211 if (migr_type
== MIGR_REPAIR
) {
212 dev
->vol
.migr_type
= MIGR_VERIFY
;
213 dev
->status
|= DEV_VERIFY_AND_FIX
;
215 dev
->vol
.migr_type
= migr_type
;
216 dev
->status
&= ~DEV_VERIFY_AND_FIX
;
220 static unsigned int sector_count(__u32 bytes
)
222 return ((bytes
+ (512-1)) & (~(512-1))) / 512;
225 static unsigned int mpb_sectors(struct imsm_super
*mpb
)
227 return sector_count(__le32_to_cpu(mpb
->mpb_size
));
231 struct imsm_dev
*dev
;
232 struct intel_dev
*next
;
237 enum sys_dev_type type
;
240 struct intel_hba
*next
;
247 /* internal representation of IMSM metadata */
250 void *buf
; /* O_DIRECT buffer for reading/writing metadata */
251 struct imsm_super
*anchor
; /* immovable parameters */
253 size_t len
; /* size of the 'buf' allocation */
254 void *next_buf
; /* for realloc'ing buf from the manager */
256 int updates_pending
; /* count of pending updates for mdmon */
257 int current_vol
; /* index of raid device undergoing creation */
258 __u32 create_offset
; /* common start for 'current_vol' */
259 __u32 random
; /* random data for seeding new family numbers */
260 struct intel_dev
*devlist
;
264 __u8 serial
[MAX_RAID_SERIAL_LEN
];
267 struct imsm_disk disk
;
270 struct extent
*e
; /* for determining freespace @ create */
271 int raiddisk
; /* slot to fill in autolayout */
274 struct dl
*disk_mgmt_list
; /* list of disks to add/remove while mdmon
276 struct dl
*missing
; /* disks removed while we weren't looking */
277 struct bbm_log
*bbm_log
;
278 struct intel_hba
*hba
; /* device path of the raid controller for this metadata */
279 const struct imsm_orom
*orom
; /* platform firmware support */
280 struct intel_super
*next
; /* (temp) list for disambiguating family_num */
284 struct imsm_disk disk
;
285 #define IMSM_UNKNOWN_OWNER (-1)
287 struct intel_disk
*next
;
291 unsigned long long start
, size
;
294 /* definitions of reshape process types */
295 enum imsm_reshape_type
{
300 /* definition of messages passed to imsm_process_update */
301 enum imsm_update_type
{
302 update_activate_spare
,
306 update_add_remove_disk
,
307 update_reshape_container_disks
,
311 struct imsm_update_activate_spare
{
312 enum imsm_update_type type
;
316 struct imsm_update_activate_spare
*next
;
329 enum takeover_direction
{
333 struct imsm_update_takeover
{
334 enum imsm_update_type type
;
336 enum takeover_direction direction
;
339 struct imsm_update_reshape
{
340 enum imsm_update_type type
;
343 int new_disks
[1]; /* new_raid_disks - old_raid_disks makedev number */
347 __u8 serial
[MAX_RAID_SERIAL_LEN
];
350 struct imsm_update_create_array
{
351 enum imsm_update_type type
;
356 struct imsm_update_kill_array
{
357 enum imsm_update_type type
;
361 struct imsm_update_rename_array
{
362 enum imsm_update_type type
;
363 __u8 name
[MAX_RAID_SERIAL_LEN
];
367 struct imsm_update_add_remove_disk
{
368 enum imsm_update_type type
;
372 static const char *_sys_dev_type
[] = {
373 [SYS_DEV_UNKNOWN
] = "Unknown",
374 [SYS_DEV_SAS
] = "SAS",
375 [SYS_DEV_SATA
] = "SATA"
378 const char *get_sys_dev_type(enum sys_dev_type type
)
380 if (type
>= SYS_DEV_MAX
)
381 type
= SYS_DEV_UNKNOWN
;
383 return _sys_dev_type
[type
];
387 static struct intel_hba
* alloc_intel_hba(struct sys_dev
*device
)
389 struct intel_hba
*result
= malloc(sizeof(*result
));
391 result
->type
= device
->type
;
392 result
->path
= strdup(device
->path
);
394 if (result
->path
&& (result
->pci_id
= strrchr(result
->path
, '/')) != NULL
)
400 static struct intel_hba
* find_intel_hba(struct intel_hba
*hba
, struct sys_dev
*device
)
402 struct intel_hba
*result
=NULL
;
403 for (result
= hba
; result
; result
= result
->next
) {
404 if (result
->type
== device
->type
&& strcmp(result
->path
, device
->path
) == 0)
411 static int attach_hba_to_super(struct intel_super
*super
, struct sys_dev
*device
)
413 struct intel_hba
*hba
;
415 /* check if disk attached to Intel HBA */
416 hba
= find_intel_hba(super
->hba
, device
);
419 /* Check if HBA is already attached to super */
420 if (super
->hba
== NULL
) {
421 super
->hba
= alloc_intel_hba(device
);
426 /* Intel metadata allows for all disks attached to the same type HBA.
427 * Do not sypport odf HBA types mixing
429 if (device
->type
!= hba
->type
)
435 hba
->next
= alloc_intel_hba(device
);
439 static struct sys_dev
* find_disk_attached_hba(int fd
, const char *devname
)
441 struct sys_dev
*list
, *elem
, *prev
;
444 if ((list
= find_intel_devices()) == NULL
)
448 disk_path
= (char *) devname
;
450 disk_path
= diskfd_to_devpath(fd
);
457 for (prev
= NULL
, elem
= list
; elem
; prev
= elem
, elem
= elem
->next
) {
458 if (path_attached_to_hba(disk_path
, elem
->path
)) {
462 prev
->next
= elem
->next
;
464 if (disk_path
!= devname
)
470 if (disk_path
!= devname
)
476 #endif /* MDASSEMBLE */
479 static struct supertype
*match_metadata_desc_imsm(char *arg
)
481 struct supertype
*st
;
483 if (strcmp(arg
, "imsm") != 0 &&
484 strcmp(arg
, "default") != 0
488 st
= malloc(sizeof(*st
));
491 memset(st
, 0, sizeof(*st
));
492 st
->container_dev
= NoMdDev
;
493 st
->ss
= &super_imsm
;
494 st
->max_devs
= IMSM_MAX_DEVICES
;
495 st
->minor_version
= 0;
501 static __u8
*get_imsm_version(struct imsm_super
*mpb
)
503 return &mpb
->sig
[MPB_SIG_LEN
];
507 /* retrieve a disk directly from the anchor when the anchor is known to be
508 * up-to-date, currently only at load time
510 static struct imsm_disk
*__get_imsm_disk(struct imsm_super
*mpb
, __u8 index
)
512 if (index
>= mpb
->num_disks
)
514 return &mpb
->disk
[index
];
517 /* retrieve the disk description based on a index of the disk
520 static struct dl
*get_imsm_dl_disk(struct intel_super
*super
, __u8 index
)
524 for (d
= super
->disks
; d
; d
= d
->next
)
525 if (d
->index
== index
)
530 /* retrieve a disk from the parsed metadata */
531 static struct imsm_disk
*get_imsm_disk(struct intel_super
*super
, __u8 index
)
535 dl
= get_imsm_dl_disk(super
, index
);
542 /* generate a checksum directly from the anchor when the anchor is known to be
543 * up-to-date, currently only at load or write_super after coalescing
545 static __u32
__gen_imsm_checksum(struct imsm_super
*mpb
)
547 __u32 end
= mpb
->mpb_size
/ sizeof(end
);
548 __u32
*p
= (__u32
*) mpb
;
552 sum
+= __le32_to_cpu(*p
);
556 return sum
- __le32_to_cpu(mpb
->check_sum
);
559 static size_t sizeof_imsm_map(struct imsm_map
*map
)
561 return sizeof(struct imsm_map
) + sizeof(__u32
) * (map
->num_members
- 1);
564 struct imsm_map
*get_imsm_map(struct imsm_dev
*dev
, int second_map
)
566 /* A device can have 2 maps if it is in the middle of a migration.
568 * 0 - we return the first map
569 * 1 - we return the second map if it exists, else NULL
570 * -1 - we return the second map if it exists, else the first
572 struct imsm_map
*map
= &dev
->vol
.map
[0];
574 if (second_map
== 1 && !dev
->vol
.migr_state
)
576 else if (second_map
== 1 ||
577 (second_map
< 0 && dev
->vol
.migr_state
)) {
580 return ptr
+ sizeof_imsm_map(map
);
586 /* return the size of the device.
587 * migr_state increases the returned size if map[0] were to be duplicated
589 static size_t sizeof_imsm_dev(struct imsm_dev
*dev
, int migr_state
)
591 size_t size
= sizeof(*dev
) - sizeof(struct imsm_map
) +
592 sizeof_imsm_map(get_imsm_map(dev
, 0));
594 /* migrating means an additional map */
595 if (dev
->vol
.migr_state
)
596 size
+= sizeof_imsm_map(get_imsm_map(dev
, 1));
598 size
+= sizeof_imsm_map(get_imsm_map(dev
, 0));
604 /* retrieve disk serial number list from a metadata update */
605 static struct disk_info
*get_disk_info(struct imsm_update_create_array
*update
)
608 struct disk_info
*inf
;
610 inf
= u
+ sizeof(*update
) - sizeof(struct imsm_dev
) +
611 sizeof_imsm_dev(&update
->dev
, 0);
617 static struct imsm_dev
*__get_imsm_dev(struct imsm_super
*mpb
, __u8 index
)
623 if (index
>= mpb
->num_raid_devs
)
626 /* devices start after all disks */
627 offset
= ((void *) &mpb
->disk
[mpb
->num_disks
]) - _mpb
;
629 for (i
= 0; i
<= index
; i
++)
631 return _mpb
+ offset
;
633 offset
+= sizeof_imsm_dev(_mpb
+ offset
, 0);
638 static struct imsm_dev
*get_imsm_dev(struct intel_super
*super
, __u8 index
)
640 struct intel_dev
*dv
;
642 if (index
>= super
->anchor
->num_raid_devs
)
644 for (dv
= super
->devlist
; dv
; dv
= dv
->next
)
645 if (dv
->index
== index
)
653 * == 1 get second map
654 * == -1 than get map according to the current migr_state
656 static __u32
get_imsm_ord_tbl_ent(struct imsm_dev
*dev
,
660 struct imsm_map
*map
;
662 map
= get_imsm_map(dev
, second_map
);
664 /* top byte identifies disk under rebuild */
665 return __le32_to_cpu(map
->disk_ord_tbl
[slot
]);
668 #define ord_to_idx(ord) (((ord) << 8) >> 8)
669 static __u32
get_imsm_disk_idx(struct imsm_dev
*dev
, int slot
, int second_map
)
671 __u32 ord
= get_imsm_ord_tbl_ent(dev
, slot
, second_map
);
673 return ord_to_idx(ord
);
676 static void set_imsm_ord_tbl_ent(struct imsm_map
*map
, int slot
, __u32 ord
)
678 map
->disk_ord_tbl
[slot
] = __cpu_to_le32(ord
);
681 static int get_imsm_disk_slot(struct imsm_map
*map
, unsigned idx
)
686 for (slot
= 0; slot
< map
->num_members
; slot
++) {
687 ord
= __le32_to_cpu(map
->disk_ord_tbl
[slot
]);
688 if (ord_to_idx(ord
) == idx
)
695 static int get_imsm_raid_level(struct imsm_map
*map
)
697 if (map
->raid_level
== 1) {
698 if (map
->num_members
== 2)
704 return map
->raid_level
;
707 static int cmp_extent(const void *av
, const void *bv
)
709 const struct extent
*a
= av
;
710 const struct extent
*b
= bv
;
711 if (a
->start
< b
->start
)
713 if (a
->start
> b
->start
)
718 static int count_memberships(struct dl
*dl
, struct intel_super
*super
)
723 for (i
= 0; i
< super
->anchor
->num_raid_devs
; i
++) {
724 struct imsm_dev
*dev
= get_imsm_dev(super
, i
);
725 struct imsm_map
*map
= get_imsm_map(dev
, 0);
727 if (get_imsm_disk_slot(map
, dl
->index
) >= 0)
734 static struct extent
*get_extents(struct intel_super
*super
, struct dl
*dl
)
736 /* find a list of used extents on the given physical device */
737 struct extent
*rv
, *e
;
739 int memberships
= count_memberships(dl
, super
);
740 __u32 reservation
= MPB_SECTOR_CNT
+ IMSM_RESERVED_SECTORS
;
742 rv
= malloc(sizeof(struct extent
) * (memberships
+ 1));
747 for (i
= 0; i
< super
->anchor
->num_raid_devs
; i
++) {
748 struct imsm_dev
*dev
= get_imsm_dev(super
, i
);
749 struct imsm_map
*map
= get_imsm_map(dev
, 0);
751 if (get_imsm_disk_slot(map
, dl
->index
) >= 0) {
752 e
->start
= __le32_to_cpu(map
->pba_of_lba0
);
753 e
->size
= __le32_to_cpu(map
->blocks_per_member
);
757 qsort(rv
, memberships
, sizeof(*rv
), cmp_extent
);
759 /* determine the start of the metadata
760 * when no raid devices are defined use the default
761 * ...otherwise allow the metadata to truncate the value
762 * as is the case with older versions of imsm
765 struct extent
*last
= &rv
[memberships
- 1];
768 remainder
= __le32_to_cpu(dl
->disk
.total_blocks
) -
769 (last
->start
+ last
->size
);
770 /* round down to 1k block to satisfy precision of the kernel
774 /* make sure remainder is still sane */
775 if (remainder
< (unsigned)ROUND_UP(super
->len
, 512) >> 9)
776 remainder
= ROUND_UP(super
->len
, 512) >> 9;
777 if (reservation
> remainder
)
778 reservation
= remainder
;
780 e
->start
= __le32_to_cpu(dl
->disk
.total_blocks
) - reservation
;
785 /* try to determine how much space is reserved for metadata from
786 * the last get_extents() entry, otherwise fallback to the
789 static __u32
imsm_reserved_sectors(struct intel_super
*super
, struct dl
*dl
)
795 /* for spares just return a minimal reservation which will grow
796 * once the spare is picked up by an array
799 return MPB_SECTOR_CNT
;
801 e
= get_extents(super
, dl
);
803 return MPB_SECTOR_CNT
+ IMSM_RESERVED_SECTORS
;
805 /* scroll to last entry */
806 for (i
= 0; e
[i
].size
; i
++)
809 rv
= __le32_to_cpu(dl
->disk
.total_blocks
) - e
[i
].start
;
816 static int is_spare(struct imsm_disk
*disk
)
818 return (disk
->status
& SPARE_DISK
) == SPARE_DISK
;
821 static int is_configured(struct imsm_disk
*disk
)
823 return (disk
->status
& CONFIGURED_DISK
) == CONFIGURED_DISK
;
826 static int is_failed(struct imsm_disk
*disk
)
828 return (disk
->status
& FAILED_DISK
) == FAILED_DISK
;
831 /* Return minimum size of a spare that can be used in this array*/
832 static unsigned long long min_acceptable_spare_size_imsm(struct supertype
*st
)
834 struct intel_super
*super
= st
->sb
;
838 unsigned long long rv
= 0;
842 /* find first active disk in array */
844 while (dl
&& (is_failed(&dl
->disk
) || dl
->index
== -1))
848 /* find last lba used by subarrays */
849 e
= get_extents(super
, dl
);
852 for (i
= 0; e
[i
].size
; i
++)
855 rv
= e
[i
-1].start
+ e
[i
-1].size
;
857 /* add the amount of space needed for metadata */
858 rv
= rv
+ MPB_SECTOR_CNT
+ IMSM_RESERVED_SECTORS
;
863 static __u64
blocks_per_migr_unit(struct imsm_dev
*dev
);
865 static void print_imsm_dev(struct imsm_dev
*dev
, char *uuid
, int disk_idx
)
869 struct imsm_map
*map
= get_imsm_map(dev
, 0);
870 struct imsm_map
*map2
= get_imsm_map(dev
, 1);
874 printf("[%.16s]:\n", dev
->volume
);
875 printf(" UUID : %s\n", uuid
);
876 printf(" RAID Level : %d", get_imsm_raid_level(map
));
878 printf(" <-- %d", get_imsm_raid_level(map2
));
880 printf(" Members : %d", map
->num_members
);
882 printf(" <-- %d", map2
->num_members
);
884 printf(" Slots : [");
885 for (i
= 0; i
< map
->num_members
; i
++) {
886 ord
= get_imsm_ord_tbl_ent(dev
, i
, 0);
887 printf("%s", ord
& IMSM_ORD_REBUILD
? "_" : "U");
892 for (i
= 0; i
< map2
->num_members
; i
++) {
893 ord
= get_imsm_ord_tbl_ent(dev
, i
, 1);
894 printf("%s", ord
& IMSM_ORD_REBUILD
? "_" : "U");
899 printf(" Failed disk : ");
900 if (map
->failed_disk_num
== 0xff)
903 printf("%i", map
->failed_disk_num
);
905 slot
= get_imsm_disk_slot(map
, disk_idx
);
907 ord
= get_imsm_ord_tbl_ent(dev
, slot
, -1);
908 printf(" This Slot : %d%s\n", slot
,
909 ord
& IMSM_ORD_REBUILD
? " (out-of-sync)" : "");
911 printf(" This Slot : ?\n");
912 sz
= __le32_to_cpu(dev
->size_high
);
914 sz
+= __le32_to_cpu(dev
->size_low
);
915 printf(" Array Size : %llu%s\n", (unsigned long long)sz
,
916 human_size(sz
* 512));
917 sz
= __le32_to_cpu(map
->blocks_per_member
);
918 printf(" Per Dev Size : %llu%s\n", (unsigned long long)sz
,
919 human_size(sz
* 512));
920 printf(" Sector Offset : %u\n",
921 __le32_to_cpu(map
->pba_of_lba0
));
922 printf(" Num Stripes : %u\n",
923 __le32_to_cpu(map
->num_data_stripes
));
924 printf(" Chunk Size : %u KiB",
925 __le16_to_cpu(map
->blocks_per_strip
) / 2);
927 printf(" <-- %u KiB",
928 __le16_to_cpu(map2
->blocks_per_strip
) / 2);
930 printf(" Reserved : %d\n", __le32_to_cpu(dev
->reserved_blocks
));
931 printf(" Migrate State : ");
932 if (dev
->vol
.migr_state
) {
933 if (migr_type(dev
) == MIGR_INIT
)
934 printf("initialize\n");
935 else if (migr_type(dev
) == MIGR_REBUILD
)
937 else if (migr_type(dev
) == MIGR_VERIFY
)
939 else if (migr_type(dev
) == MIGR_GEN_MIGR
)
940 printf("general migration\n");
941 else if (migr_type(dev
) == MIGR_STATE_CHANGE
)
942 printf("state change\n");
943 else if (migr_type(dev
) == MIGR_REPAIR
)
946 printf("<unknown:%d>\n", migr_type(dev
));
949 printf(" Map State : %s", map_state_str
[map
->map_state
]);
950 if (dev
->vol
.migr_state
) {
951 struct imsm_map
*map
= get_imsm_map(dev
, 1);
953 printf(" <-- %s", map_state_str
[map
->map_state
]);
954 printf("\n Checkpoint : %u (%llu)",
955 __le32_to_cpu(dev
->vol
.curr_migr_unit
),
956 (unsigned long long)blocks_per_migr_unit(dev
));
959 printf(" Dirty State : %s\n", dev
->vol
.dirty
? "dirty" : "clean");
962 static void print_imsm_disk(struct imsm_super
*mpb
, int index
, __u32 reserved
)
964 struct imsm_disk
*disk
= __get_imsm_disk(mpb
, index
);
965 char str
[MAX_RAID_SERIAL_LEN
+ 1];
968 if (index
< 0 || !disk
)
972 snprintf(str
, MAX_RAID_SERIAL_LEN
+ 1, "%s", disk
->serial
);
973 printf(" Disk%02d Serial : %s\n", index
, str
);
974 printf(" State :%s%s%s\n", is_spare(disk
) ? " spare" : "",
975 is_configured(disk
) ? " active" : "",
976 is_failed(disk
) ? " failed" : "");
977 printf(" Id : %08x\n", __le32_to_cpu(disk
->scsi_id
));
978 sz
= __le32_to_cpu(disk
->total_blocks
) - reserved
;
979 printf(" Usable Size : %llu%s\n", (unsigned long long)sz
,
980 human_size(sz
* 512));
983 static void getinfo_super_imsm(struct supertype
*st
, struct mdinfo
*info
, char *map
);
985 static void examine_super_imsm(struct supertype
*st
, char *homehost
)
987 struct intel_super
*super
= st
->sb
;
988 struct imsm_super
*mpb
= super
->anchor
;
989 char str
[MAX_SIGNATURE_LENGTH
];
994 __u32 reserved
= imsm_reserved_sectors(super
, super
->disks
);
997 snprintf(str
, MPB_SIG_LEN
, "%s", mpb
->sig
);
998 printf(" Magic : %s\n", str
);
999 snprintf(str
, strlen(MPB_VERSION_RAID0
), "%s", get_imsm_version(mpb
));
1000 printf(" Version : %s\n", get_imsm_version(mpb
));
1001 printf(" Orig Family : %08x\n", __le32_to_cpu(mpb
->orig_family_num
));
1002 printf(" Family : %08x\n", __le32_to_cpu(mpb
->family_num
));
1003 printf(" Generation : %08x\n", __le32_to_cpu(mpb
->generation_num
));
1004 getinfo_super_imsm(st
, &info
, NULL
);
1005 fname_from_uuid(st
, &info
, nbuf
, ':');
1006 printf(" UUID : %s\n", nbuf
+ 5);
1007 sum
= __le32_to_cpu(mpb
->check_sum
);
1008 printf(" Checksum : %08x %s\n", sum
,
1009 __gen_imsm_checksum(mpb
) == sum
? "correct" : "incorrect");
1010 printf(" MPB Sectors : %d\n", mpb_sectors(mpb
));
1011 printf(" Disks : %d\n", mpb
->num_disks
);
1012 printf(" RAID Devices : %d\n", mpb
->num_raid_devs
);
1013 print_imsm_disk(mpb
, super
->disks
->index
, reserved
);
1014 if (super
->bbm_log
) {
1015 struct bbm_log
*log
= super
->bbm_log
;
1018 printf("Bad Block Management Log:\n");
1019 printf(" Log Size : %d\n", __le32_to_cpu(mpb
->bbm_log_size
));
1020 printf(" Signature : %x\n", __le32_to_cpu(log
->signature
));
1021 printf(" Entry Count : %d\n", __le32_to_cpu(log
->entry_count
));
1022 printf(" Spare Blocks : %d\n", __le32_to_cpu(log
->reserved_spare_block_count
));
1023 printf(" First Spare : %llx\n",
1024 (unsigned long long) __le64_to_cpu(log
->first_spare_lba
));
1026 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
1028 struct imsm_dev
*dev
= __get_imsm_dev(mpb
, i
);
1030 super
->current_vol
= i
;
1031 getinfo_super_imsm(st
, &info
, NULL
);
1032 fname_from_uuid(st
, &info
, nbuf
, ':');
1033 print_imsm_dev(dev
, nbuf
+ 5, super
->disks
->index
);
1035 for (i
= 0; i
< mpb
->num_disks
; i
++) {
1036 if (i
== super
->disks
->index
)
1038 print_imsm_disk(mpb
, i
, reserved
);
1040 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
1041 struct imsm_disk
*disk
;
1042 char str
[MAX_RAID_SERIAL_LEN
+ 1];
1050 snprintf(str
, MAX_RAID_SERIAL_LEN
+ 1, "%s", disk
->serial
);
1051 printf(" Disk Serial : %s\n", str
);
1052 printf(" State :%s%s%s\n", is_spare(disk
) ? " spare" : "",
1053 is_configured(disk
) ? " active" : "",
1054 is_failed(disk
) ? " failed" : "");
1055 printf(" Id : %08x\n", __le32_to_cpu(disk
->scsi_id
));
1056 sz
= __le32_to_cpu(disk
->total_blocks
) - reserved
;
1057 printf(" Usable Size : %llu%s\n", (unsigned long long)sz
,
1058 human_size(sz
* 512));
1062 static void brief_examine_super_imsm(struct supertype
*st
, int verbose
)
1064 /* We just write a generic IMSM ARRAY entry */
1067 struct intel_super
*super
= st
->sb
;
1069 if (!super
->anchor
->num_raid_devs
) {
1070 printf("ARRAY metadata=imsm\n");
1074 getinfo_super_imsm(st
, &info
, NULL
);
1075 fname_from_uuid(st
, &info
, nbuf
, ':');
1076 printf("ARRAY metadata=imsm UUID=%s\n", nbuf
+ 5);
1079 static void brief_examine_subarrays_imsm(struct supertype
*st
, int verbose
)
1081 /* We just write a generic IMSM ARRAY entry */
1085 struct intel_super
*super
= st
->sb
;
1088 if (!super
->anchor
->num_raid_devs
)
1091 getinfo_super_imsm(st
, &info
, NULL
);
1092 fname_from_uuid(st
, &info
, nbuf
, ':');
1093 for (i
= 0; i
< super
->anchor
->num_raid_devs
; i
++) {
1094 struct imsm_dev
*dev
= get_imsm_dev(super
, i
);
1096 super
->current_vol
= i
;
1097 getinfo_super_imsm(st
, &info
, NULL
);
1098 fname_from_uuid(st
, &info
, nbuf1
, ':');
1099 printf("ARRAY /dev/md/%.16s container=%s member=%d UUID=%s\n",
1100 dev
->volume
, nbuf
+ 5, i
, nbuf1
+ 5);
1104 static void export_examine_super_imsm(struct supertype
*st
)
1106 struct intel_super
*super
= st
->sb
;
1107 struct imsm_super
*mpb
= super
->anchor
;
1111 getinfo_super_imsm(st
, &info
, NULL
);
1112 fname_from_uuid(st
, &info
, nbuf
, ':');
1113 printf("MD_METADATA=imsm\n");
1114 printf("MD_LEVEL=container\n");
1115 printf("MD_UUID=%s\n", nbuf
+5);
1116 printf("MD_DEVICES=%u\n", mpb
->num_disks
);
1119 static void detail_super_imsm(struct supertype
*st
, char *homehost
)
1124 getinfo_super_imsm(st
, &info
, NULL
);
1125 fname_from_uuid(st
, &info
, nbuf
, ':');
1126 printf("\n UUID : %s\n", nbuf
+ 5);
1129 static void brief_detail_super_imsm(struct supertype
*st
)
1133 getinfo_super_imsm(st
, &info
, NULL
);
1134 fname_from_uuid(st
, &info
, nbuf
, ':');
1135 printf(" UUID=%s", nbuf
+ 5);
1138 static int imsm_read_serial(int fd
, char *devname
, __u8
*serial
);
1139 static void fd2devname(int fd
, char *name
);
1141 static int ahci_enumerate_ports(const char *hba_path
, int port_count
, int host_base
, int verbose
)
1143 /* dump an unsorted list of devices attached to AHCI Intel storage
1144 * controller, as well as non-connected ports
1146 int hba_len
= strlen(hba_path
) + 1;
1151 unsigned long port_mask
= (1 << port_count
) - 1;
1153 if (port_count
> (int)sizeof(port_mask
) * 8) {
1155 fprintf(stderr
, Name
": port_count %d out of range\n", port_count
);
1159 /* scroll through /sys/dev/block looking for devices attached to
1162 dir
= opendir("/sys/dev/block");
1163 for (ent
= dir
? readdir(dir
) : NULL
; ent
; ent
= readdir(dir
)) {
1174 if (sscanf(ent
->d_name
, "%d:%d", &major
, &minor
) != 2)
1176 path
= devt_to_devpath(makedev(major
, minor
));
1179 if (!path_attached_to_hba(path
, hba_path
)) {
1185 /* retrieve the scsi device type */
1186 if (asprintf(&device
, "/sys/dev/block/%d:%d/device/xxxxxxx", major
, minor
) < 0) {
1188 fprintf(stderr
, Name
": failed to allocate 'device'\n");
1192 sprintf(device
, "/sys/dev/block/%d:%d/device/type", major
, minor
);
1193 if (load_sys(device
, buf
) != 0) {
1195 fprintf(stderr
, Name
": failed to read device type for %s\n",
1201 type
= strtoul(buf
, NULL
, 10);
1203 /* if it's not a disk print the vendor and model */
1204 if (!(type
== 0 || type
== 7 || type
== 14)) {
1207 sprintf(device
, "/sys/dev/block/%d:%d/device/vendor", major
, minor
);
1208 if (load_sys(device
, buf
) == 0) {
1209 strncpy(vendor
, buf
, sizeof(vendor
));
1210 vendor
[sizeof(vendor
) - 1] = '\0';
1211 c
= (char *) &vendor
[sizeof(vendor
) - 1];
1212 while (isspace(*c
) || *c
== '\0')
1216 sprintf(device
, "/sys/dev/block/%d:%d/device/model", major
, minor
);
1217 if (load_sys(device
, buf
) == 0) {
1218 strncpy(model
, buf
, sizeof(model
));
1219 model
[sizeof(model
) - 1] = '\0';
1220 c
= (char *) &model
[sizeof(model
) - 1];
1221 while (isspace(*c
) || *c
== '\0')
1225 if (vendor
[0] && model
[0])
1226 sprintf(buf
, "%.64s %.64s", vendor
, model
);
1228 switch (type
) { /* numbers from hald/linux/device.c */
1229 case 1: sprintf(buf
, "tape"); break;
1230 case 2: sprintf(buf
, "printer"); break;
1231 case 3: sprintf(buf
, "processor"); break;
1233 case 5: sprintf(buf
, "cdrom"); break;
1234 case 6: sprintf(buf
, "scanner"); break;
1235 case 8: sprintf(buf
, "media_changer"); break;
1236 case 9: sprintf(buf
, "comm"); break;
1237 case 12: sprintf(buf
, "raid"); break;
1238 default: sprintf(buf
, "unknown");
1244 /* chop device path to 'host%d' and calculate the port number */
1245 c
= strchr(&path
[hba_len
], '/');
1248 fprintf(stderr
, Name
": %s - invalid path name\n", path
+ hba_len
);
1253 if (sscanf(&path
[hba_len
], "host%d", &port
) == 1)
1257 *c
= '/'; /* repair the full string */
1258 fprintf(stderr
, Name
": failed to determine port number for %s\n",
1265 /* mark this port as used */
1266 port_mask
&= ~(1 << port
);
1268 /* print out the device information */
1270 printf(" Port%d : - non-disk device (%s) -\n", port
, buf
);
1274 fd
= dev_open(ent
->d_name
, O_RDONLY
);
1276 printf(" Port%d : - disk info unavailable -\n", port
);
1278 fd2devname(fd
, buf
);
1279 printf(" Port%d : %s", port
, buf
);
1280 if (imsm_read_serial(fd
, NULL
, (__u8
*) buf
) == 0)
1281 printf(" (%s)\n", buf
);
1296 for (i
= 0; i
< port_count
; i
++)
1297 if (port_mask
& (1 << i
))
1298 printf(" Port%d : - no device attached -\n", i
);
1306 static void print_found_intel_controllers(struct sys_dev
*elem
)
1308 for (; elem
; elem
= elem
->next
) {
1309 fprintf(stderr
, Name
": found Intel(R) ");
1310 if (elem
->type
== SYS_DEV_SATA
)
1311 fprintf(stderr
, "SATA ");
1312 else if (elem
->type
== SYS_DEV_SAS
)
1313 fprintf(stderr
, "SAS ");
1314 fprintf(stderr
, "RAID controller");
1316 fprintf(stderr
, " at %s", elem
->pci_id
);
1317 fprintf(stderr
, ".\n");
1322 static int ahci_get_port_count(const char *hba_path
, int *port_count
)
1329 if ((dir
= opendir(hba_path
)) == NULL
)
1332 for (ent
= readdir(dir
); ent
; ent
= readdir(dir
)) {
1335 if (sscanf(ent
->d_name
, "host%d", &host
) != 1)
1337 if (*port_count
== 0)
1339 else if (host
< host_base
)
1342 if (host
+ 1 > *port_count
+ host_base
)
1343 *port_count
= host
+ 1 - host_base
;
1349 static void print_imsm_capability(const struct imsm_orom
*orom
)
1351 printf(" Platform : Intel(R) Matrix Storage Manager\n");
1352 printf(" Version : %d.%d.%d.%d\n", orom
->major_ver
, orom
->minor_ver
,
1353 orom
->hotfix_ver
, orom
->build
);
1354 printf(" RAID Levels :%s%s%s%s%s\n",
1355 imsm_orom_has_raid0(orom
) ? " raid0" : "",
1356 imsm_orom_has_raid1(orom
) ? " raid1" : "",
1357 imsm_orom_has_raid1e(orom
) ? " raid1e" : "",
1358 imsm_orom_has_raid10(orom
) ? " raid10" : "",
1359 imsm_orom_has_raid5(orom
) ? " raid5" : "");
1360 printf(" Chunk Sizes :%s%s%s%s%s%s%s%s%s%s%s%s%s%s%s%s\n",
1361 imsm_orom_has_chunk(orom
, 2) ? " 2k" : "",
1362 imsm_orom_has_chunk(orom
, 4) ? " 4k" : "",
1363 imsm_orom_has_chunk(orom
, 8) ? " 8k" : "",
1364 imsm_orom_has_chunk(orom
, 16) ? " 16k" : "",
1365 imsm_orom_has_chunk(orom
, 32) ? " 32k" : "",
1366 imsm_orom_has_chunk(orom
, 64) ? " 64k" : "",
1367 imsm_orom_has_chunk(orom
, 128) ? " 128k" : "",
1368 imsm_orom_has_chunk(orom
, 256) ? " 256k" : "",
1369 imsm_orom_has_chunk(orom
, 512) ? " 512k" : "",
1370 imsm_orom_has_chunk(orom
, 1024*1) ? " 1M" : "",
1371 imsm_orom_has_chunk(orom
, 1024*2) ? " 2M" : "",
1372 imsm_orom_has_chunk(orom
, 1024*4) ? " 4M" : "",
1373 imsm_orom_has_chunk(orom
, 1024*8) ? " 8M" : "",
1374 imsm_orom_has_chunk(orom
, 1024*16) ? " 16M" : "",
1375 imsm_orom_has_chunk(orom
, 1024*32) ? " 32M" : "",
1376 imsm_orom_has_chunk(orom
, 1024*64) ? " 64M" : "");
1377 printf(" Max Disks : %d\n", orom
->tds
);
1378 printf(" Max Volumes : %d\n", orom
->vpa
);
1382 static int detail_platform_imsm(int verbose
, int enumerate_only
)
1384 /* There are two components to imsm platform support, the ahci SATA
1385 * controller and the option-rom. To find the SATA controller we
1386 * simply look in /sys/bus/pci/drivers/ahci to see if an ahci
1387 * controller with the Intel vendor id is present. This approach
1388 * allows mdadm to leverage the kernel's ahci detection logic, with the
1389 * caveat that if ahci.ko is not loaded mdadm will not be able to
1390 * detect platform raid capabilities. The option-rom resides in a
1391 * platform "Adapter ROM". We scan for its signature to retrieve the
1392 * platform capabilities. If raid support is disabled in the BIOS the
1393 * option-rom capability structure will not be available.
1395 const struct imsm_orom
*orom
;
1396 struct sys_dev
*list
, *hba
;
1401 if (enumerate_only
) {
1402 if (check_env("IMSM_NO_PLATFORM"))
1404 list
= find_intel_devices();
1407 for (hba
= list
; hba
; hba
= hba
->next
) {
1408 orom
= find_imsm_capability(hba
->type
);
1414 free_sys_dev(&list
);
1418 list
= find_intel_devices();
1421 fprintf(stderr
, Name
": no active Intel(R) RAID "
1422 "controller found.\n");
1423 free_sys_dev(&list
);
1426 print_found_intel_controllers(list
);
1428 for (hba
= list
; hba
; hba
= hba
->next
) {
1429 orom
= find_imsm_capability(hba
->type
);
1431 fprintf(stderr
, Name
": imsm capabilities not found for controller: %s (type %s)\n",
1432 hba
->path
, get_sys_dev_type(hba
->type
));
1434 print_imsm_capability(orom
);
1437 for (hba
= list
; hba
; hba
= hba
->next
) {
1438 printf(" I/O Controller : %s (%s)\n",
1439 hba
->path
, get_sys_dev_type(hba
->type
));
1441 if (hba
->type
== SYS_DEV_SATA
) {
1442 host_base
= ahci_get_port_count(hba
->path
, &port_count
);
1443 if (ahci_enumerate_ports(hba
->path
, port_count
, host_base
, verbose
)) {
1445 fprintf(stderr
, Name
": failed to enumerate "
1446 "ports on SATA controller at %s.", hba
->pci_id
);
1452 free_sys_dev(&list
);
1457 static int match_home_imsm(struct supertype
*st
, char *homehost
)
1459 /* the imsm metadata format does not specify any host
1460 * identification information. We return -1 since we can never
1461 * confirm nor deny whether a given array is "meant" for this
1462 * host. We rely on compare_super and the 'family_num' fields to
1463 * exclude member disks that do not belong, and we rely on
1464 * mdadm.conf to specify the arrays that should be assembled.
1465 * Auto-assembly may still pick up "foreign" arrays.
1471 static void uuid_from_super_imsm(struct supertype
*st
, int uuid
[4])
1473 /* The uuid returned here is used for:
1474 * uuid to put into bitmap file (Create, Grow)
1475 * uuid for backup header when saving critical section (Grow)
1476 * comparing uuids when re-adding a device into an array
1477 * In these cases the uuid required is that of the data-array,
1478 * not the device-set.
1479 * uuid to recognise same set when adding a missing device back
1480 * to an array. This is a uuid for the device-set.
1482 * For each of these we can make do with a truncated
1483 * or hashed uuid rather than the original, as long as
1485 * In each case the uuid required is that of the data-array,
1486 * not the device-set.
1488 /* imsm does not track uuid's so we synthesis one using sha1 on
1489 * - The signature (Which is constant for all imsm array, but no matter)
1490 * - the orig_family_num of the container
1491 * - the index number of the volume
1492 * - the 'serial' number of the volume.
1493 * Hopefully these are all constant.
1495 struct intel_super
*super
= st
->sb
;
1498 struct sha1_ctx ctx
;
1499 struct imsm_dev
*dev
= NULL
;
1502 /* some mdadm versions failed to set ->orig_family_num, in which
1503 * case fall back to ->family_num. orig_family_num will be
1504 * fixed up with the first metadata update.
1506 family_num
= super
->anchor
->orig_family_num
;
1507 if (family_num
== 0)
1508 family_num
= super
->anchor
->family_num
;
1509 sha1_init_ctx(&ctx
);
1510 sha1_process_bytes(super
->anchor
->sig
, MPB_SIG_LEN
, &ctx
);
1511 sha1_process_bytes(&family_num
, sizeof(__u32
), &ctx
);
1512 if (super
->current_vol
>= 0)
1513 dev
= get_imsm_dev(super
, super
->current_vol
);
1515 __u32 vol
= super
->current_vol
;
1516 sha1_process_bytes(&vol
, sizeof(vol
), &ctx
);
1517 sha1_process_bytes(dev
->volume
, MAX_RAID_SERIAL_LEN
, &ctx
);
1519 sha1_finish_ctx(&ctx
, buf
);
1520 memcpy(uuid
, buf
, 4*4);
1525 get_imsm_numerical_version(struct imsm_super
*mpb
, int *m
, int *p
)
1527 __u8
*v
= get_imsm_version(mpb
);
1528 __u8
*end
= mpb
->sig
+ MAX_SIGNATURE_LENGTH
;
1529 char major
[] = { 0, 0, 0 };
1530 char minor
[] = { 0 ,0, 0 };
1531 char patch
[] = { 0, 0, 0 };
1532 char *ver_parse
[] = { major
, minor
, patch
};
1536 while (*v
!= '\0' && v
< end
) {
1537 if (*v
!= '.' && j
< 2)
1538 ver_parse
[i
][j
++] = *v
;
1546 *m
= strtol(minor
, NULL
, 0);
1547 *p
= strtol(patch
, NULL
, 0);
1551 static __u32
migr_strip_blocks_resync(struct imsm_dev
*dev
)
1553 /* migr_strip_size when repairing or initializing parity */
1554 struct imsm_map
*map
= get_imsm_map(dev
, 0);
1555 __u32 chunk
= __le32_to_cpu(map
->blocks_per_strip
);
1557 switch (get_imsm_raid_level(map
)) {
1562 return 128*1024 >> 9;
1566 static __u32
migr_strip_blocks_rebuild(struct imsm_dev
*dev
)
1568 /* migr_strip_size when rebuilding a degraded disk, no idea why
1569 * this is different than migr_strip_size_resync(), but it's good
1572 struct imsm_map
*map
= get_imsm_map(dev
, 1);
1573 __u32 chunk
= __le32_to_cpu(map
->blocks_per_strip
);
1575 switch (get_imsm_raid_level(map
)) {
1578 if (map
->num_members
% map
->num_domains
== 0)
1579 return 128*1024 >> 9;
1583 return max((__u32
) 64*1024 >> 9, chunk
);
1585 return 128*1024 >> 9;
1589 static __u32
num_stripes_per_unit_resync(struct imsm_dev
*dev
)
1591 struct imsm_map
*lo
= get_imsm_map(dev
, 0);
1592 struct imsm_map
*hi
= get_imsm_map(dev
, 1);
1593 __u32 lo_chunk
= __le32_to_cpu(lo
->blocks_per_strip
);
1594 __u32 hi_chunk
= __le32_to_cpu(hi
->blocks_per_strip
);
1596 return max((__u32
) 1, hi_chunk
/ lo_chunk
);
1599 static __u32
num_stripes_per_unit_rebuild(struct imsm_dev
*dev
)
1601 struct imsm_map
*lo
= get_imsm_map(dev
, 0);
1602 int level
= get_imsm_raid_level(lo
);
1604 if (level
== 1 || level
== 10) {
1605 struct imsm_map
*hi
= get_imsm_map(dev
, 1);
1607 return hi
->num_domains
;
1609 return num_stripes_per_unit_resync(dev
);
1612 static __u8
imsm_num_data_members(struct imsm_dev
*dev
, int second_map
)
1614 /* named 'imsm_' because raid0, raid1 and raid10
1615 * counter-intuitively have the same number of data disks
1617 struct imsm_map
*map
= get_imsm_map(dev
, second_map
);
1619 switch (get_imsm_raid_level(map
)) {
1623 return map
->num_members
;
1625 return map
->num_members
- 1;
1627 dprintf("%s: unsupported raid level\n", __func__
);
1632 static __u32
parity_segment_depth(struct imsm_dev
*dev
)
1634 struct imsm_map
*map
= get_imsm_map(dev
, 0);
1635 __u32 chunk
= __le32_to_cpu(map
->blocks_per_strip
);
1637 switch(get_imsm_raid_level(map
)) {
1640 return chunk
* map
->num_domains
;
1642 return chunk
* map
->num_members
;
1648 static __u32
map_migr_block(struct imsm_dev
*dev
, __u32 block
)
1650 struct imsm_map
*map
= get_imsm_map(dev
, 1);
1651 __u32 chunk
= __le32_to_cpu(map
->blocks_per_strip
);
1652 __u32 strip
= block
/ chunk
;
1654 switch (get_imsm_raid_level(map
)) {
1657 __u32 vol_strip
= (strip
* map
->num_domains
) + 1;
1658 __u32 vol_stripe
= vol_strip
/ map
->num_members
;
1660 return vol_stripe
* chunk
+ block
% chunk
;
1662 __u32 stripe
= strip
/ (map
->num_members
- 1);
1664 return stripe
* chunk
+ block
% chunk
;
1671 static __u64
blocks_per_migr_unit(struct imsm_dev
*dev
)
1673 /* calculate the conversion factor between per member 'blocks'
1674 * (md/{resync,rebuild}_start) and imsm migration units, return
1675 * 0 for the 'not migrating' and 'unsupported migration' cases
1677 if (!dev
->vol
.migr_state
)
1680 switch (migr_type(dev
)) {
1685 struct imsm_map
*map
= get_imsm_map(dev
, 0);
1686 __u32 stripes_per_unit
;
1687 __u32 blocks_per_unit
;
1696 /* yes, this is really the translation of migr_units to
1697 * per-member blocks in the 'resync' case
1699 stripes_per_unit
= num_stripes_per_unit_resync(dev
);
1700 migr_chunk
= migr_strip_blocks_resync(dev
);
1701 disks
= imsm_num_data_members(dev
, 0);
1702 blocks_per_unit
= stripes_per_unit
* migr_chunk
* disks
;
1703 if (migr_type(dev
) == MIGR_GEN_MIGR
)
1704 return blocks_per_unit
;
1705 stripe
= __le32_to_cpu(map
->blocks_per_strip
) * disks
;
1706 segment
= blocks_per_unit
/ stripe
;
1707 block_rel
= blocks_per_unit
- segment
* stripe
;
1708 parity_depth
= parity_segment_depth(dev
);
1709 block_map
= map_migr_block(dev
, block_rel
);
1710 return block_map
+ parity_depth
* segment
;
1712 case MIGR_REBUILD
: {
1713 __u32 stripes_per_unit
;
1716 stripes_per_unit
= num_stripes_per_unit_rebuild(dev
);
1717 migr_chunk
= migr_strip_blocks_rebuild(dev
);
1718 return migr_chunk
* stripes_per_unit
;
1720 case MIGR_STATE_CHANGE
:
1726 static int imsm_level_to_layout(int level
)
1734 return ALGORITHM_LEFT_ASYMMETRIC
;
1741 static void getinfo_super_imsm_volume(struct supertype
*st
, struct mdinfo
*info
, char *dmap
)
1743 struct intel_super
*super
= st
->sb
;
1744 struct imsm_dev
*dev
= get_imsm_dev(super
, super
->current_vol
);
1745 struct imsm_map
*map
= get_imsm_map(dev
, 0);
1746 struct imsm_map
*prev_map
= get_imsm_map(dev
, 1);
1747 struct imsm_map
*map_to_analyse
= map
;
1750 int map_disks
= info
->array
.raid_disks
;
1753 map_to_analyse
= prev_map
;
1755 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
1756 if (dl
->raiddisk
== info
->disk
.raid_disk
)
1758 info
->container_member
= super
->current_vol
;
1759 info
->array
.raid_disks
= map_to_analyse
->num_members
;
1760 info
->array
.level
= get_imsm_raid_level(map_to_analyse
);
1761 info
->array
.layout
= imsm_level_to_layout(info
->array
.level
);
1762 info
->array
.md_minor
= -1;
1763 info
->array
.ctime
= 0;
1764 info
->array
.utime
= 0;
1765 info
->array
.chunk_size
=
1766 __le16_to_cpu(map_to_analyse
->blocks_per_strip
) << 9;
1767 info
->array
.state
= !dev
->vol
.dirty
;
1768 info
->custom_array_size
= __le32_to_cpu(dev
->size_high
);
1769 info
->custom_array_size
<<= 32;
1770 info
->custom_array_size
|= __le32_to_cpu(dev
->size_low
);
1771 if (prev_map
&& map
->map_state
== prev_map
->map_state
) {
1772 info
->reshape_active
= 1;
1773 info
->new_level
= get_imsm_raid_level(map
);
1774 info
->new_layout
= imsm_level_to_layout(info
->new_level
);
1775 info
->new_chunk
= __le16_to_cpu(map
->blocks_per_strip
) << 9;
1776 info
->delta_disks
= map
->num_members
- prev_map
->num_members
;
1777 if (info
->delta_disks
) {
1778 /* this needs to be applied to every array
1781 info
->reshape_active
= 2;
1783 /* We shape information that we give to md might have to be
1784 * modify to cope with md's requirement for reshaping arrays.
1785 * For example, when reshaping a RAID0, md requires it to be
1786 * presented as a degraded RAID4.
1787 * Also if a RAID0 is migrating to a RAID5 we need to specify
1788 * the array as already being RAID5, but the 'before' layout
1789 * is a RAID4-like layout.
1791 switch (info
->array
.level
) {
1793 switch(info
->new_level
) {
1795 /* conversion is happening as RAID4 */
1796 info
->array
.level
= 4;
1797 info
->array
.raid_disks
+= 1;
1800 /* conversion is happening as RAID5 */
1801 info
->array
.level
= 5;
1802 info
->array
.layout
= ALGORITHM_PARITY_N
;
1803 info
->array
.raid_disks
+= 1;
1804 info
->delta_disks
-= 1;
1807 /* FIXME error message */
1808 info
->array
.level
= UnSet
;
1814 info
->new_level
= UnSet
;
1815 info
->new_layout
= UnSet
;
1816 info
->new_chunk
= info
->array
.chunk_size
;
1817 info
->delta_disks
= 0;
1819 info
->disk
.major
= 0;
1820 info
->disk
.minor
= 0;
1822 info
->disk
.major
= dl
->major
;
1823 info
->disk
.minor
= dl
->minor
;
1826 info
->data_offset
= __le32_to_cpu(map_to_analyse
->pba_of_lba0
);
1827 info
->component_size
=
1828 __le32_to_cpu(map_to_analyse
->blocks_per_member
);
1829 memset(info
->uuid
, 0, sizeof(info
->uuid
));
1830 info
->recovery_start
= MaxSector
;
1832 info
->reshape_progress
= 0;
1833 info
->resync_start
= MaxSector
;
1834 if (map_to_analyse
->map_state
== IMSM_T_STATE_UNINITIALIZED
||
1836 info
->resync_start
= 0;
1838 if (dev
->vol
.migr_state
) {
1839 switch (migr_type(dev
)) {
1842 __u64 blocks_per_unit
= blocks_per_migr_unit(dev
);
1843 __u64 units
= __le32_to_cpu(dev
->vol
.curr_migr_unit
);
1845 info
->resync_start
= blocks_per_unit
* units
;
1848 case MIGR_GEN_MIGR
: {
1849 __u64 blocks_per_unit
= blocks_per_migr_unit(dev
);
1850 __u64 units
= __le32_to_cpu(dev
->vol
.curr_migr_unit
);
1851 unsigned long long array_blocks
;
1854 info
->reshape_progress
= blocks_per_unit
* units
;
1855 dprintf("IMSM: General Migration checkpoint : %llu "
1856 "(%llu) -> read reshape progress : %llu\n",
1857 units
, blocks_per_unit
, info
->reshape_progress
);
1859 used_disks
= imsm_num_data_members(dev
, 1);
1860 if (used_disks
> 0) {
1861 array_blocks
= map
->blocks_per_member
*
1863 /* round array size down to closest MB
1865 info
->custom_array_size
= (array_blocks
1866 >> SECT_PER_MB_SHIFT
)
1867 << SECT_PER_MB_SHIFT
;
1871 /* we could emulate the checkpointing of
1872 * 'sync_action=check' migrations, but for now
1873 * we just immediately complete them
1876 /* this is handled by container_content_imsm() */
1877 case MIGR_STATE_CHANGE
:
1878 /* FIXME handle other migrations */
1880 /* we are not dirty, so... */
1881 info
->resync_start
= MaxSector
;
1885 strncpy(info
->name
, (char *) dev
->volume
, MAX_RAID_SERIAL_LEN
);
1886 info
->name
[MAX_RAID_SERIAL_LEN
] = 0;
1888 info
->array
.major_version
= -1;
1889 info
->array
.minor_version
= -2;
1890 devname
= devnum2devname(st
->container_dev
);
1891 *info
->text_version
= '\0';
1893 sprintf(info
->text_version
, "/%s/%d", devname
, info
->container_member
);
1895 info
->safe_mode_delay
= 4000; /* 4 secs like the Matrix driver */
1896 uuid_from_super_imsm(st
, info
->uuid
);
1900 for (i
=0; i
<map_disks
; i
++) {
1902 if (i
< info
->array
.raid_disks
) {
1903 struct imsm_disk
*dsk
;
1904 j
= get_imsm_disk_idx(dev
, i
, -1);
1905 dsk
= get_imsm_disk(super
, j
);
1906 if (dsk
&& (dsk
->status
& CONFIGURED_DISK
))
1913 static __u8
imsm_check_degraded(struct intel_super
*super
, struct imsm_dev
*dev
, int failed
);
1914 static int imsm_count_failed(struct intel_super
*super
, struct imsm_dev
*dev
);
1916 static struct imsm_disk
*get_imsm_missing(struct intel_super
*super
, __u8 index
)
1920 for (d
= super
->missing
; d
; d
= d
->next
)
1921 if (d
->index
== index
)
1926 static void getinfo_super_imsm(struct supertype
*st
, struct mdinfo
*info
, char *map
)
1928 struct intel_super
*super
= st
->sb
;
1929 struct imsm_disk
*disk
;
1930 int map_disks
= info
->array
.raid_disks
;
1931 int max_enough
= -1;
1933 struct imsm_super
*mpb
;
1935 if (super
->current_vol
>= 0) {
1936 getinfo_super_imsm_volume(st
, info
, map
);
1940 /* Set raid_disks to zero so that Assemble will always pull in valid
1943 info
->array
.raid_disks
= 0;
1944 info
->array
.level
= LEVEL_CONTAINER
;
1945 info
->array
.layout
= 0;
1946 info
->array
.md_minor
= -1;
1947 info
->array
.ctime
= 0; /* N/A for imsm */
1948 info
->array
.utime
= 0;
1949 info
->array
.chunk_size
= 0;
1951 info
->disk
.major
= 0;
1952 info
->disk
.minor
= 0;
1953 info
->disk
.raid_disk
= -1;
1954 info
->reshape_active
= 0;
1955 info
->array
.major_version
= -1;
1956 info
->array
.minor_version
= -2;
1957 strcpy(info
->text_version
, "imsm");
1958 info
->safe_mode_delay
= 0;
1959 info
->disk
.number
= -1;
1960 info
->disk
.state
= 0;
1962 info
->recovery_start
= MaxSector
;
1964 /* do we have the all the insync disks that we expect? */
1965 mpb
= super
->anchor
;
1967 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
1968 struct imsm_dev
*dev
= get_imsm_dev(super
, i
);
1969 int failed
, enough
, j
, missing
= 0;
1970 struct imsm_map
*map
;
1973 failed
= imsm_count_failed(super
, dev
);
1974 state
= imsm_check_degraded(super
, dev
, failed
);
1975 map
= get_imsm_map(dev
, dev
->vol
.migr_state
);
1977 /* any newly missing disks?
1978 * (catches single-degraded vs double-degraded)
1980 for (j
= 0; j
< map
->num_members
; j
++) {
1981 __u32 ord
= get_imsm_ord_tbl_ent(dev
, i
, -1);
1982 __u32 idx
= ord_to_idx(ord
);
1984 if (!(ord
& IMSM_ORD_REBUILD
) &&
1985 get_imsm_missing(super
, idx
)) {
1991 if (state
== IMSM_T_STATE_FAILED
)
1993 else if (state
== IMSM_T_STATE_DEGRADED
&&
1994 (state
!= map
->map_state
|| missing
))
1996 else /* we're normal, or already degraded */
1999 /* in the missing/failed disk case check to see
2000 * if at least one array is runnable
2002 max_enough
= max(max_enough
, enough
);
2004 dprintf("%s: enough: %d\n", __func__
, max_enough
);
2005 info
->container_enough
= max_enough
;
2008 __u32 reserved
= imsm_reserved_sectors(super
, super
->disks
);
2010 disk
= &super
->disks
->disk
;
2011 info
->data_offset
= __le32_to_cpu(disk
->total_blocks
) - reserved
;
2012 info
->component_size
= reserved
;
2013 info
->disk
.state
= is_configured(disk
) ? (1 << MD_DISK_ACTIVE
) : 0;
2014 /* we don't change info->disk.raid_disk here because
2015 * this state will be finalized in mdmon after we have
2016 * found the 'most fresh' version of the metadata
2018 info
->disk
.state
|= is_failed(disk
) ? (1 << MD_DISK_FAULTY
) : 0;
2019 info
->disk
.state
|= is_spare(disk
) ? 0 : (1 << MD_DISK_SYNC
);
2022 /* only call uuid_from_super_imsm when this disk is part of a populated container,
2023 * ->compare_super may have updated the 'num_raid_devs' field for spares
2025 if (info
->disk
.state
& (1 << MD_DISK_SYNC
) || super
->anchor
->num_raid_devs
)
2026 uuid_from_super_imsm(st
, info
->uuid
);
2028 memcpy(info
->uuid
, uuid_zero
, sizeof(uuid_zero
));
2030 /* I don't know how to compute 'map' on imsm, so use safe default */
2033 for (i
= 0; i
< map_disks
; i
++)
2039 /* allocates memory and fills disk in mdinfo structure
2040 * for each disk in array */
2041 struct mdinfo
*getinfo_super_disks_imsm(struct supertype
*st
)
2043 struct mdinfo
*mddev
= NULL
;
2044 struct intel_super
*super
= st
->sb
;
2045 struct imsm_disk
*disk
;
2048 if (!super
|| !super
->disks
)
2051 mddev
= malloc(sizeof(*mddev
));
2053 fprintf(stderr
, Name
": Failed to allocate memory.\n");
2056 memset(mddev
, 0, sizeof(*mddev
));
2060 tmp
= malloc(sizeof(*tmp
));
2062 fprintf(stderr
, Name
": Failed to allocate memory.\n");
2067 memset(tmp
, 0, sizeof(*tmp
));
2069 tmp
->next
= mddev
->devs
;
2071 tmp
->disk
.number
= count
++;
2072 tmp
->disk
.major
= dl
->major
;
2073 tmp
->disk
.minor
= dl
->minor
;
2074 tmp
->disk
.state
= is_configured(disk
) ?
2075 (1 << MD_DISK_ACTIVE
) : 0;
2076 tmp
->disk
.state
|= is_failed(disk
) ? (1 << MD_DISK_FAULTY
) : 0;
2077 tmp
->disk
.state
|= is_spare(disk
) ? 0 : (1 << MD_DISK_SYNC
);
2078 tmp
->disk
.raid_disk
= -1;
2084 static int update_super_imsm(struct supertype
*st
, struct mdinfo
*info
,
2085 char *update
, char *devname
, int verbose
,
2086 int uuid_set
, char *homehost
)
2088 /* For 'assemble' and 'force' we need to return non-zero if any
2089 * change was made. For others, the return value is ignored.
2090 * Update options are:
2091 * force-one : This device looks a bit old but needs to be included,
2092 * update age info appropriately.
2093 * assemble: clear any 'faulty' flag to allow this device to
2095 * force-array: Array is degraded but being forced, mark it clean
2096 * if that will be needed to assemble it.
2098 * newdev: not used ????
2099 * grow: Array has gained a new device - this is currently for
2101 * resync: mark as dirty so a resync will happen.
2102 * name: update the name - preserving the homehost
2103 * uuid: Change the uuid of the array to match watch is given
2105 * Following are not relevant for this imsm:
2106 * sparc2.2 : update from old dodgey metadata
2107 * super-minor: change the preferred_minor number
2108 * summaries: update redundant counters.
2109 * homehost: update the recorded homehost
2110 * _reshape_progress: record new reshape_progress position.
2113 struct intel_super
*super
= st
->sb
;
2114 struct imsm_super
*mpb
;
2116 /* we can only update container info */
2117 if (!super
|| super
->current_vol
>= 0 || !super
->anchor
)
2120 mpb
= super
->anchor
;
2122 if (strcmp(update
, "uuid") == 0 && uuid_set
&& !info
->update_private
)
2124 else if (strcmp(update
, "uuid") == 0 && uuid_set
&& info
->update_private
) {
2125 mpb
->orig_family_num
= *((__u32
*) info
->update_private
);
2127 } else if (strcmp(update
, "uuid") == 0) {
2128 __u32
*new_family
= malloc(sizeof(*new_family
));
2130 /* update orig_family_number with the incoming random
2131 * data, report the new effective uuid, and store the
2132 * new orig_family_num for future updates.
2135 memcpy(&mpb
->orig_family_num
, info
->uuid
, sizeof(__u32
));
2136 uuid_from_super_imsm(st
, info
->uuid
);
2137 *new_family
= mpb
->orig_family_num
;
2138 info
->update_private
= new_family
;
2141 } else if (strcmp(update
, "assemble") == 0)
2146 /* successful update? recompute checksum */
2148 mpb
->check_sum
= __le32_to_cpu(__gen_imsm_checksum(mpb
));
2153 static size_t disks_to_mpb_size(int disks
)
2157 size
= sizeof(struct imsm_super
);
2158 size
+= (disks
- 1) * sizeof(struct imsm_disk
);
2159 size
+= 2 * sizeof(struct imsm_dev
);
2160 /* up to 2 maps per raid device (-2 for imsm_maps in imsm_dev */
2161 size
+= (4 - 2) * sizeof(struct imsm_map
);
2162 /* 4 possible disk_ord_tbl's */
2163 size
+= 4 * (disks
- 1) * sizeof(__u32
);
2168 static __u64
avail_size_imsm(struct supertype
*st
, __u64 devsize
)
2170 if (devsize
< (MPB_SECTOR_CNT
+ IMSM_RESERVED_SECTORS
))
2173 return devsize
- (MPB_SECTOR_CNT
+ IMSM_RESERVED_SECTORS
);
2176 static void free_devlist(struct intel_super
*super
)
2178 struct intel_dev
*dv
;
2180 while (super
->devlist
) {
2181 dv
= super
->devlist
->next
;
2182 free(super
->devlist
->dev
);
2183 free(super
->devlist
);
2184 super
->devlist
= dv
;
2188 static void imsm_copy_dev(struct imsm_dev
*dest
, struct imsm_dev
*src
)
2190 memcpy(dest
, src
, sizeof_imsm_dev(src
, 0));
2193 static int compare_super_imsm(struct supertype
*st
, struct supertype
*tst
)
2197 * 0 same, or first was empty, and second was copied
2198 * 1 second had wrong number
2200 * 3 wrong other info
2202 struct intel_super
*first
= st
->sb
;
2203 struct intel_super
*sec
= tst
->sb
;
2210 /* in platform dependent environment test if the disks
2211 * use the same Intel hba
2213 if (!check_env("IMSM_NO_PLATFORM")) {
2214 if (first
->hba
->type
!= sec
->hba
->type
) {
2216 "HBAs of devices does not match %s != %s\n",
2217 get_sys_dev_type(first
->hba
->type
),
2218 get_sys_dev_type(sec
->hba
->type
));
2223 /* if an anchor does not have num_raid_devs set then it is a free
2226 if (first
->anchor
->num_raid_devs
> 0 &&
2227 sec
->anchor
->num_raid_devs
> 0) {
2228 /* Determine if these disks might ever have been
2229 * related. Further disambiguation can only take place
2230 * in load_super_imsm_all
2232 __u32 first_family
= first
->anchor
->orig_family_num
;
2233 __u32 sec_family
= sec
->anchor
->orig_family_num
;
2235 if (memcmp(first
->anchor
->sig
, sec
->anchor
->sig
,
2236 MAX_SIGNATURE_LENGTH
) != 0)
2239 if (first_family
== 0)
2240 first_family
= first
->anchor
->family_num
;
2241 if (sec_family
== 0)
2242 sec_family
= sec
->anchor
->family_num
;
2244 if (first_family
!= sec_family
)
2250 /* if 'first' is a spare promote it to a populated mpb with sec's
2253 if (first
->anchor
->num_raid_devs
== 0 &&
2254 sec
->anchor
->num_raid_devs
> 0) {
2256 struct intel_dev
*dv
;
2257 struct imsm_dev
*dev
;
2259 /* we need to copy raid device info from sec if an allocation
2260 * fails here we don't associate the spare
2262 for (i
= 0; i
< sec
->anchor
->num_raid_devs
; i
++) {
2263 dv
= malloc(sizeof(*dv
));
2266 dev
= malloc(sizeof_imsm_dev(get_imsm_dev(sec
, i
), 1));
2273 dv
->next
= first
->devlist
;
2274 first
->devlist
= dv
;
2276 if (i
< sec
->anchor
->num_raid_devs
) {
2277 /* allocation failure */
2278 free_devlist(first
);
2279 fprintf(stderr
, "imsm: failed to associate spare\n");
2282 first
->anchor
->num_raid_devs
= sec
->anchor
->num_raid_devs
;
2283 first
->anchor
->orig_family_num
= sec
->anchor
->orig_family_num
;
2284 first
->anchor
->family_num
= sec
->anchor
->family_num
;
2285 memcpy(first
->anchor
->sig
, sec
->anchor
->sig
, MAX_SIGNATURE_LENGTH
);
2286 for (i
= 0; i
< sec
->anchor
->num_raid_devs
; i
++)
2287 imsm_copy_dev(get_imsm_dev(first
, i
), get_imsm_dev(sec
, i
));
2293 static void fd2devname(int fd
, char *name
)
2297 char dname
[PATH_MAX
];
2302 if (fstat(fd
, &st
) != 0)
2304 sprintf(path
, "/sys/dev/block/%d:%d",
2305 major(st
.st_rdev
), minor(st
.st_rdev
));
2307 rv
= readlink(path
, dname
, sizeof(dname
));
2312 nm
= strrchr(dname
, '/');
2314 snprintf(name
, MAX_RAID_SERIAL_LEN
, "/dev/%s", nm
);
2317 extern int scsi_get_serial(int fd
, void *buf
, size_t buf_len
);
2319 static int imsm_read_serial(int fd
, char *devname
,
2320 __u8 serial
[MAX_RAID_SERIAL_LEN
])
2322 unsigned char scsi_serial
[255];
2331 memset(scsi_serial
, 0, sizeof(scsi_serial
));
2333 rv
= scsi_get_serial(fd
, scsi_serial
, sizeof(scsi_serial
));
2335 if (rv
&& check_env("IMSM_DEVNAME_AS_SERIAL")) {
2336 memset(serial
, 0, MAX_RAID_SERIAL_LEN
);
2337 fd2devname(fd
, (char *) serial
);
2344 Name
": Failed to retrieve serial for %s\n",
2349 rsp_len
= scsi_serial
[3];
2353 Name
": Failed to retrieve serial for %s\n",
2357 rsp_buf
= (char *) &scsi_serial
[4];
2359 /* trim all whitespace and non-printable characters and convert
2362 for (i
= 0, dest
= rsp_buf
; i
< rsp_len
; i
++) {
2365 /* ':' is reserved for use in placeholder serial
2366 * numbers for missing disks
2374 len
= dest
- rsp_buf
;
2377 /* truncate leading characters */
2378 if (len
> MAX_RAID_SERIAL_LEN
) {
2379 dest
+= len
- MAX_RAID_SERIAL_LEN
;
2380 len
= MAX_RAID_SERIAL_LEN
;
2383 memset(serial
, 0, MAX_RAID_SERIAL_LEN
);
2384 memcpy(serial
, dest
, len
);
2389 static int serialcmp(__u8
*s1
, __u8
*s2
)
2391 return strncmp((char *) s1
, (char *) s2
, MAX_RAID_SERIAL_LEN
);
2394 static void serialcpy(__u8
*dest
, __u8
*src
)
2396 strncpy((char *) dest
, (char *) src
, MAX_RAID_SERIAL_LEN
);
2400 static struct dl
*serial_to_dl(__u8
*serial
, struct intel_super
*super
)
2404 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
2405 if (serialcmp(dl
->serial
, serial
) == 0)
2412 static struct imsm_disk
*
2413 __serial_to_disk(__u8
*serial
, struct imsm_super
*mpb
, int *idx
)
2417 for (i
= 0; i
< mpb
->num_disks
; i
++) {
2418 struct imsm_disk
*disk
= __get_imsm_disk(mpb
, i
);
2420 if (serialcmp(disk
->serial
, serial
) == 0) {
2431 load_imsm_disk(int fd
, struct intel_super
*super
, char *devname
, int keep_fd
)
2433 struct imsm_disk
*disk
;
2438 __u8 serial
[MAX_RAID_SERIAL_LEN
];
2440 rv
= imsm_read_serial(fd
, devname
, serial
);
2445 dl
= calloc(1, sizeof(*dl
));
2449 Name
": failed to allocate disk buffer for %s\n",
2455 dl
->major
= major(stb
.st_rdev
);
2456 dl
->minor
= minor(stb
.st_rdev
);
2457 dl
->next
= super
->disks
;
2458 dl
->fd
= keep_fd
? fd
: -1;
2459 assert(super
->disks
== NULL
);
2461 serialcpy(dl
->serial
, serial
);
2464 fd2devname(fd
, name
);
2466 dl
->devname
= strdup(devname
);
2468 dl
->devname
= strdup(name
);
2470 /* look up this disk's index in the current anchor */
2471 disk
= __serial_to_disk(dl
->serial
, super
->anchor
, &dl
->index
);
2474 /* only set index on disks that are a member of a
2475 * populated contianer, i.e. one with raid_devs
2477 if (is_failed(&dl
->disk
))
2479 else if (is_spare(&dl
->disk
))
2487 /* When migrating map0 contains the 'destination' state while map1
2488 * contains the current state. When not migrating map0 contains the
2489 * current state. This routine assumes that map[0].map_state is set to
2490 * the current array state before being called.
2492 * Migration is indicated by one of the following states
2493 * 1/ Idle (migr_state=0 map0state=normal||unitialized||degraded||failed)
2494 * 2/ Initialize (migr_state=1 migr_type=MIGR_INIT map0state=normal
2495 * map1state=unitialized)
2496 * 3/ Repair (Resync) (migr_state=1 migr_type=MIGR_REPAIR map0state=normal
2498 * 4/ Rebuild (migr_state=1 migr_type=MIGR_REBUILD map0state=normal
2499 * map1state=degraded)
2501 static void migrate(struct imsm_dev
*dev
, __u8 to_state
, int migr_type
)
2503 struct imsm_map
*dest
;
2504 struct imsm_map
*src
= get_imsm_map(dev
, 0);
2506 dev
->vol
.migr_state
= 1;
2507 set_migr_type(dev
, migr_type
);
2508 dev
->vol
.curr_migr_unit
= 0;
2509 dest
= get_imsm_map(dev
, 1);
2511 /* duplicate and then set the target end state in map[0] */
2512 memcpy(dest
, src
, sizeof_imsm_map(src
));
2513 if ((migr_type
== MIGR_REBUILD
) ||
2514 (migr_type
== MIGR_GEN_MIGR
)) {
2518 for (i
= 0; i
< src
->num_members
; i
++) {
2519 ord
= __le32_to_cpu(src
->disk_ord_tbl
[i
]);
2520 set_imsm_ord_tbl_ent(src
, i
, ord_to_idx(ord
));
2524 src
->map_state
= to_state
;
2527 static void end_migration(struct imsm_dev
*dev
, __u8 map_state
)
2529 struct imsm_map
*map
= get_imsm_map(dev
, 0);
2530 struct imsm_map
*prev
= get_imsm_map(dev
, dev
->vol
.migr_state
);
2533 /* merge any IMSM_ORD_REBUILD bits that were not successfully
2534 * completed in the last migration.
2536 * FIXME add support for raid-level-migration
2538 for (i
= 0; i
< prev
->num_members
; i
++)
2539 for (j
= 0; j
< map
->num_members
; j
++)
2540 /* during online capacity expansion
2541 * disks position can be changed if takeover is used
2543 if (ord_to_idx(map
->disk_ord_tbl
[j
]) ==
2544 ord_to_idx(prev
->disk_ord_tbl
[i
])) {
2545 map
->disk_ord_tbl
[j
] |= prev
->disk_ord_tbl
[i
];
2549 dev
->vol
.migr_state
= 0;
2550 dev
->vol
.migr_type
= 0;
2551 dev
->vol
.curr_migr_unit
= 0;
2552 map
->map_state
= map_state
;
2556 static int parse_raid_devices(struct intel_super
*super
)
2559 struct imsm_dev
*dev_new
;
2560 size_t len
, len_migr
;
2562 size_t space_needed
= 0;
2563 struct imsm_super
*mpb
= super
->anchor
;
2565 for (i
= 0; i
< super
->anchor
->num_raid_devs
; i
++) {
2566 struct imsm_dev
*dev_iter
= __get_imsm_dev(super
->anchor
, i
);
2567 struct intel_dev
*dv
;
2569 len
= sizeof_imsm_dev(dev_iter
, 0);
2570 len_migr
= sizeof_imsm_dev(dev_iter
, 1);
2572 space_needed
+= len_migr
- len
;
2574 dv
= malloc(sizeof(*dv
));
2577 if (max_len
< len_migr
)
2579 if (max_len
> len_migr
)
2580 space_needed
+= max_len
- len_migr
;
2581 dev_new
= malloc(max_len
);
2586 imsm_copy_dev(dev_new
, dev_iter
);
2589 dv
->next
= super
->devlist
;
2590 super
->devlist
= dv
;
2593 /* ensure that super->buf is large enough when all raid devices
2596 if (__le32_to_cpu(mpb
->mpb_size
) + space_needed
> super
->len
) {
2599 len
= ROUND_UP(__le32_to_cpu(mpb
->mpb_size
) + space_needed
, 512);
2600 if (posix_memalign(&buf
, 512, len
) != 0)
2603 memcpy(buf
, super
->buf
, super
->len
);
2604 memset(buf
+ super
->len
, 0, len
- super
->len
);
2613 /* retrieve a pointer to the bbm log which starts after all raid devices */
2614 struct bbm_log
*__get_imsm_bbm_log(struct imsm_super
*mpb
)
2618 if (__le32_to_cpu(mpb
->bbm_log_size
)) {
2620 ptr
+= mpb
->mpb_size
- __le32_to_cpu(mpb
->bbm_log_size
);
2626 static void __free_imsm(struct intel_super
*super
, int free_disks
);
2628 /* load_imsm_mpb - read matrix metadata
2629 * allocates super->mpb to be freed by free_super
2631 static int load_imsm_mpb(int fd
, struct intel_super
*super
, char *devname
)
2633 unsigned long long dsize
;
2634 unsigned long long sectors
;
2636 struct imsm_super
*anchor
;
2639 get_dev_size(fd
, NULL
, &dsize
);
2643 Name
": %s: device to small for imsm\n",
2648 if (lseek64(fd
, dsize
- (512 * 2), SEEK_SET
) < 0) {
2651 Name
": Cannot seek to anchor block on %s: %s\n",
2652 devname
, strerror(errno
));
2656 if (posix_memalign((void**)&anchor
, 512, 512) != 0) {
2659 Name
": Failed to allocate imsm anchor buffer"
2660 " on %s\n", devname
);
2663 if (read(fd
, anchor
, 512) != 512) {
2666 Name
": Cannot read anchor block on %s: %s\n",
2667 devname
, strerror(errno
));
2672 if (strncmp((char *) anchor
->sig
, MPB_SIGNATURE
, MPB_SIG_LEN
) != 0) {
2675 Name
": no IMSM anchor on %s\n", devname
);
2680 __free_imsm(super
, 0);
2681 super
->len
= ROUND_UP(anchor
->mpb_size
, 512);
2682 if (posix_memalign(&super
->buf
, 512, super
->len
) != 0) {
2685 Name
": unable to allocate %zu byte mpb buffer\n",
2690 memcpy(super
->buf
, anchor
, 512);
2692 sectors
= mpb_sectors(anchor
) - 1;
2695 check_sum
= __gen_imsm_checksum(super
->anchor
);
2696 if (check_sum
!= __le32_to_cpu(super
->anchor
->check_sum
)) {
2699 Name
": IMSM checksum %x != %x on %s\n",
2701 __le32_to_cpu(super
->anchor
->check_sum
),
2709 /* read the extended mpb */
2710 if (lseek64(fd
, dsize
- (512 * (2 + sectors
)), SEEK_SET
) < 0) {
2713 Name
": Cannot seek to extended mpb on %s: %s\n",
2714 devname
, strerror(errno
));
2718 if ((unsigned)read(fd
, super
->buf
+ 512, super
->len
- 512) != super
->len
- 512) {
2721 Name
": Cannot read extended mpb on %s: %s\n",
2722 devname
, strerror(errno
));
2726 check_sum
= __gen_imsm_checksum(super
->anchor
);
2727 if (check_sum
!= __le32_to_cpu(super
->anchor
->check_sum
)) {
2730 Name
": IMSM checksum %x != %x on %s\n",
2731 check_sum
, __le32_to_cpu(super
->anchor
->check_sum
),
2736 /* FIXME the BBM log is disk specific so we cannot use this global
2737 * buffer for all disks. Ok for now since we only look at the global
2738 * bbm_log_size parameter to gate assembly
2740 super
->bbm_log
= __get_imsm_bbm_log(super
->anchor
);
2746 load_and_parse_mpb(int fd
, struct intel_super
*super
, char *devname
, int keep_fd
)
2750 err
= load_imsm_mpb(fd
, super
, devname
);
2753 err
= load_imsm_disk(fd
, super
, devname
, keep_fd
);
2756 err
= parse_raid_devices(super
);
2761 static void __free_imsm_disk(struct dl
*d
)
2773 static void free_imsm_disks(struct intel_super
*super
)
2777 while (super
->disks
) {
2779 super
->disks
= d
->next
;
2780 __free_imsm_disk(d
);
2782 while (super
->disk_mgmt_list
) {
2783 d
= super
->disk_mgmt_list
;
2784 super
->disk_mgmt_list
= d
->next
;
2785 __free_imsm_disk(d
);
2787 while (super
->missing
) {
2789 super
->missing
= d
->next
;
2790 __free_imsm_disk(d
);
2795 /* free all the pieces hanging off of a super pointer */
2796 static void __free_imsm(struct intel_super
*super
, int free_disks
)
2798 struct intel_hba
*elem
, *next
;
2805 free_imsm_disks(super
);
2806 free_devlist(super
);
2810 free((void *)elem
->path
);
2818 static void free_imsm(struct intel_super
*super
)
2820 __free_imsm(super
, 1);
2824 static void free_super_imsm(struct supertype
*st
)
2826 struct intel_super
*super
= st
->sb
;
2835 static struct intel_super
*alloc_super(void)
2837 struct intel_super
*super
= malloc(sizeof(*super
));
2840 memset(super
, 0, sizeof(*super
));
2841 super
->current_vol
= -1;
2842 super
->create_offset
= ~((__u32
) 0);
2843 if (!check_env("IMSM_NO_PLATFORM"))
2844 super
->orom
= find_imsm_orom();
2851 /* find_missing - helper routine for load_super_imsm_all that identifies
2852 * disks that have disappeared from the system. This routine relies on
2853 * the mpb being uptodate, which it is at load time.
2855 static int find_missing(struct intel_super
*super
)
2858 struct imsm_super
*mpb
= super
->anchor
;
2860 struct imsm_disk
*disk
;
2862 for (i
= 0; i
< mpb
->num_disks
; i
++) {
2863 disk
= __get_imsm_disk(mpb
, i
);
2864 dl
= serial_to_dl(disk
->serial
, super
);
2868 dl
= malloc(sizeof(*dl
));
2874 dl
->devname
= strdup("missing");
2876 serialcpy(dl
->serial
, disk
->serial
);
2879 dl
->next
= super
->missing
;
2880 super
->missing
= dl
;
2886 static struct intel_disk
*disk_list_get(__u8
*serial
, struct intel_disk
*disk_list
)
2888 struct intel_disk
*idisk
= disk_list
;
2891 if (serialcmp(idisk
->disk
.serial
, serial
) == 0)
2893 idisk
= idisk
->next
;
2899 static int __prep_thunderdome(struct intel_super
**table
, int tbl_size
,
2900 struct intel_super
*super
,
2901 struct intel_disk
**disk_list
)
2903 struct imsm_disk
*d
= &super
->disks
->disk
;
2904 struct imsm_super
*mpb
= super
->anchor
;
2907 for (i
= 0; i
< tbl_size
; i
++) {
2908 struct imsm_super
*tbl_mpb
= table
[i
]->anchor
;
2909 struct imsm_disk
*tbl_d
= &table
[i
]->disks
->disk
;
2911 if (tbl_mpb
->family_num
== mpb
->family_num
) {
2912 if (tbl_mpb
->check_sum
== mpb
->check_sum
) {
2913 dprintf("%s: mpb from %d:%d matches %d:%d\n",
2914 __func__
, super
->disks
->major
,
2915 super
->disks
->minor
,
2916 table
[i
]->disks
->major
,
2917 table
[i
]->disks
->minor
);
2921 if (((is_configured(d
) && !is_configured(tbl_d
)) ||
2922 is_configured(d
) == is_configured(tbl_d
)) &&
2923 tbl_mpb
->generation_num
< mpb
->generation_num
) {
2924 /* current version of the mpb is a
2925 * better candidate than the one in
2926 * super_table, but copy over "cross
2927 * generational" status
2929 struct intel_disk
*idisk
;
2931 dprintf("%s: mpb from %d:%d replaces %d:%d\n",
2932 __func__
, super
->disks
->major
,
2933 super
->disks
->minor
,
2934 table
[i
]->disks
->major
,
2935 table
[i
]->disks
->minor
);
2937 idisk
= disk_list_get(tbl_d
->serial
, *disk_list
);
2938 if (idisk
&& is_failed(&idisk
->disk
))
2939 tbl_d
->status
|= FAILED_DISK
;
2942 struct intel_disk
*idisk
;
2943 struct imsm_disk
*disk
;
2945 /* tbl_mpb is more up to date, but copy
2946 * over cross generational status before
2949 disk
= __serial_to_disk(d
->serial
, mpb
, NULL
);
2950 if (disk
&& is_failed(disk
))
2951 d
->status
|= FAILED_DISK
;
2953 idisk
= disk_list_get(d
->serial
, *disk_list
);
2956 if (disk
&& is_configured(disk
))
2957 idisk
->disk
.status
|= CONFIGURED_DISK
;
2960 dprintf("%s: mpb from %d:%d prefer %d:%d\n",
2961 __func__
, super
->disks
->major
,
2962 super
->disks
->minor
,
2963 table
[i
]->disks
->major
,
2964 table
[i
]->disks
->minor
);
2972 table
[tbl_size
++] = super
;
2976 /* update/extend the merged list of imsm_disk records */
2977 for (j
= 0; j
< mpb
->num_disks
; j
++) {
2978 struct imsm_disk
*disk
= __get_imsm_disk(mpb
, j
);
2979 struct intel_disk
*idisk
;
2981 idisk
= disk_list_get(disk
->serial
, *disk_list
);
2983 idisk
->disk
.status
|= disk
->status
;
2984 if (is_configured(&idisk
->disk
) ||
2985 is_failed(&idisk
->disk
))
2986 idisk
->disk
.status
&= ~(SPARE_DISK
);
2988 idisk
= calloc(1, sizeof(*idisk
));
2991 idisk
->owner
= IMSM_UNKNOWN_OWNER
;
2992 idisk
->disk
= *disk
;
2993 idisk
->next
= *disk_list
;
2997 if (serialcmp(idisk
->disk
.serial
, d
->serial
) == 0)
3004 static struct intel_super
*
3005 validate_members(struct intel_super
*super
, struct intel_disk
*disk_list
,
3008 struct imsm_super
*mpb
= super
->anchor
;
3012 for (i
= 0; i
< mpb
->num_disks
; i
++) {
3013 struct imsm_disk
*disk
= __get_imsm_disk(mpb
, i
);
3014 struct intel_disk
*idisk
;
3016 idisk
= disk_list_get(disk
->serial
, disk_list
);
3018 if (idisk
->owner
== owner
||
3019 idisk
->owner
== IMSM_UNKNOWN_OWNER
)
3022 dprintf("%s: '%.16s' owner %d != %d\n",
3023 __func__
, disk
->serial
, idisk
->owner
,
3026 dprintf("%s: unknown disk %x [%d]: %.16s\n",
3027 __func__
, __le32_to_cpu(mpb
->family_num
), i
,
3033 if (ok_count
== mpb
->num_disks
)
3038 static void show_conflicts(__u32 family_num
, struct intel_super
*super_list
)
3040 struct intel_super
*s
;
3042 for (s
= super_list
; s
; s
= s
->next
) {
3043 if (family_num
!= s
->anchor
->family_num
)
3045 fprintf(stderr
, "Conflict, offlining family %#x on '%s'\n",
3046 __le32_to_cpu(family_num
), s
->disks
->devname
);
3050 static struct intel_super
*
3051 imsm_thunderdome(struct intel_super
**super_list
, int len
)
3053 struct intel_super
*super_table
[len
];
3054 struct intel_disk
*disk_list
= NULL
;
3055 struct intel_super
*champion
, *spare
;
3056 struct intel_super
*s
, **del
;
3061 memset(super_table
, 0, sizeof(super_table
));
3062 for (s
= *super_list
; s
; s
= s
->next
)
3063 tbl_size
= __prep_thunderdome(super_table
, tbl_size
, s
, &disk_list
);
3065 for (i
= 0; i
< tbl_size
; i
++) {
3066 struct imsm_disk
*d
;
3067 struct intel_disk
*idisk
;
3068 struct imsm_super
*mpb
= super_table
[i
]->anchor
;
3071 d
= &s
->disks
->disk
;
3073 /* 'd' must appear in merged disk list for its
3074 * configuration to be valid
3076 idisk
= disk_list_get(d
->serial
, disk_list
);
3077 if (idisk
&& idisk
->owner
== i
)
3078 s
= validate_members(s
, disk_list
, i
);
3083 dprintf("%s: marking family: %#x from %d:%d offline\n",
3084 __func__
, mpb
->family_num
,
3085 super_table
[i
]->disks
->major
,
3086 super_table
[i
]->disks
->minor
);
3090 /* This is where the mdadm implementation differs from the Windows
3091 * driver which has no strict concept of a container. We can only
3092 * assemble one family from a container, so when returning a prodigal
3093 * array member to this system the code will not be able to disambiguate
3094 * the container contents that should be assembled ("foreign" versus
3095 * "local"). It requires user intervention to set the orig_family_num
3096 * to a new value to establish a new container. The Windows driver in
3097 * this situation fixes up the volume name in place and manages the
3098 * foreign array as an independent entity.
3103 for (i
= 0; i
< tbl_size
; i
++) {
3104 struct intel_super
*tbl_ent
= super_table
[i
];
3110 if (tbl_ent
->anchor
->num_raid_devs
== 0) {
3115 if (s
&& !is_spare
) {
3116 show_conflicts(tbl_ent
->anchor
->family_num
, *super_list
);
3118 } else if (!s
&& !is_spare
)
3131 fprintf(stderr
, "Chose family %#x on '%s', "
3132 "assemble conflicts to new container with '--update=uuid'\n",
3133 __le32_to_cpu(s
->anchor
->family_num
), s
->disks
->devname
);
3135 /* collect all dl's onto 'champion', and update them to
3136 * champion's version of the status
3138 for (s
= *super_list
; s
; s
= s
->next
) {
3139 struct imsm_super
*mpb
= champion
->anchor
;
3140 struct dl
*dl
= s
->disks
;
3145 for (i
= 0; i
< mpb
->num_disks
; i
++) {
3146 struct imsm_disk
*disk
;
3148 disk
= __serial_to_disk(dl
->serial
, mpb
, &dl
->index
);
3151 /* only set index on disks that are a member of
3152 * a populated contianer, i.e. one with
3155 if (is_failed(&dl
->disk
))
3157 else if (is_spare(&dl
->disk
))
3163 if (i
>= mpb
->num_disks
) {
3164 struct intel_disk
*idisk
;
3166 idisk
= disk_list_get(dl
->serial
, disk_list
);
3167 if (idisk
&& is_spare(&idisk
->disk
) &&
3168 !is_failed(&idisk
->disk
) && !is_configured(&idisk
->disk
))
3176 dl
->next
= champion
->disks
;
3177 champion
->disks
= dl
;
3181 /* delete 'champion' from super_list */
3182 for (del
= super_list
; *del
; ) {
3183 if (*del
== champion
) {
3184 *del
= (*del
)->next
;
3187 del
= &(*del
)->next
;
3189 champion
->next
= NULL
;
3193 struct intel_disk
*idisk
= disk_list
;
3195 disk_list
= disk_list
->next
;
3202 static int load_super_imsm_all(struct supertype
*st
, int fd
, void **sbp
,
3206 struct intel_super
*super_list
= NULL
;
3207 struct intel_super
*super
= NULL
;
3208 int devnum
= fd2devnum(fd
);
3214 /* check if 'fd' an opened container */
3215 sra
= sysfs_read(fd
, 0, GET_LEVEL
|GET_VERSION
|GET_DEVS
|GET_STATE
);
3219 if (sra
->array
.major_version
!= -1 ||
3220 sra
->array
.minor_version
!= -2 ||
3221 strcmp(sra
->text_version
, "imsm") != 0) {
3226 for (sd
= sra
->devs
, i
= 0; sd
; sd
= sd
->next
, i
++) {
3227 struct intel_super
*s
= alloc_super();
3234 s
->next
= super_list
;
3238 sprintf(nm
, "%d:%d", sd
->disk
.major
, sd
->disk
.minor
);
3239 dfd
= dev_open(nm
, O_RDWR
);
3243 err
= load_and_parse_mpb(dfd
, s
, NULL
, 1);
3245 /* retry the load if we might have raced against mdmon */
3246 if (err
== 3 && mdmon_running(devnum
))
3247 for (retry
= 0; retry
< 3; retry
++) {
3249 err
= load_and_parse_mpb(dfd
, s
, NULL
, 1);
3257 /* all mpbs enter, maybe one leaves */
3258 super
= imsm_thunderdome(&super_list
, i
);
3264 if (find_missing(super
) != 0) {
3272 while (super_list
) {
3273 struct intel_super
*s
= super_list
;
3275 super_list
= super_list
->next
;
3284 st
->container_dev
= devnum
;
3285 if (err
== 0 && st
->ss
== NULL
) {
3286 st
->ss
= &super_imsm
;
3287 st
->minor_version
= 0;
3288 st
->max_devs
= IMSM_MAX_DEVICES
;
3293 static int load_container_imsm(struct supertype
*st
, int fd
, char *devname
)
3295 return load_super_imsm_all(st
, fd
, &st
->sb
, devname
);
3299 static int load_super_imsm(struct supertype
*st
, int fd
, char *devname
)
3301 struct intel_super
*super
;
3304 if (test_partition(fd
))
3305 /* IMSM not allowed on partitions */
3308 free_super_imsm(st
);
3310 super
= alloc_super();
3313 Name
": malloc of %zu failed.\n",
3318 rv
= load_and_parse_mpb(fd
, super
, devname
, 0);
3323 Name
": Failed to load all information "
3324 "sections on %s\n", devname
);
3330 if (st
->ss
== NULL
) {
3331 st
->ss
= &super_imsm
;
3332 st
->minor_version
= 0;
3333 st
->max_devs
= IMSM_MAX_DEVICES
;
3338 static __u16
info_to_blocks_per_strip(mdu_array_info_t
*info
)
3340 if (info
->level
== 1)
3342 return info
->chunk_size
>> 9;
3345 static __u32
info_to_num_data_stripes(mdu_array_info_t
*info
, int num_domains
)
3349 num_stripes
= (info
->size
* 2) / info_to_blocks_per_strip(info
);
3350 num_stripes
/= num_domains
;
3355 static __u32
info_to_blocks_per_member(mdu_array_info_t
*info
)
3357 if (info
->level
== 1)
3358 return info
->size
* 2;
3360 return (info
->size
* 2) & ~(info_to_blocks_per_strip(info
) - 1);
3363 static void imsm_update_version_info(struct intel_super
*super
)
3365 /* update the version and attributes */
3366 struct imsm_super
*mpb
= super
->anchor
;
3368 struct imsm_dev
*dev
;
3369 struct imsm_map
*map
;
3372 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
3373 dev
= get_imsm_dev(super
, i
);
3374 map
= get_imsm_map(dev
, 0);
3375 if (__le32_to_cpu(dev
->size_high
) > 0)
3376 mpb
->attributes
|= MPB_ATTRIB_2TB
;
3378 /* FIXME detect when an array spans a port multiplier */
3380 mpb
->attributes
|= MPB_ATTRIB_PM
;
3383 if (mpb
->num_raid_devs
> 1 ||
3384 mpb
->attributes
!= MPB_ATTRIB_CHECKSUM_VERIFY
) {
3385 version
= MPB_VERSION_ATTRIBS
;
3386 switch (get_imsm_raid_level(map
)) {
3387 case 0: mpb
->attributes
|= MPB_ATTRIB_RAID0
; break;
3388 case 1: mpb
->attributes
|= MPB_ATTRIB_RAID1
; break;
3389 case 10: mpb
->attributes
|= MPB_ATTRIB_RAID10
; break;
3390 case 5: mpb
->attributes
|= MPB_ATTRIB_RAID5
; break;
3393 if (map
->num_members
>= 5)
3394 version
= MPB_VERSION_5OR6_DISK_ARRAY
;
3395 else if (dev
->status
== DEV_CLONE_N_GO
)
3396 version
= MPB_VERSION_CNG
;
3397 else if (get_imsm_raid_level(map
) == 5)
3398 version
= MPB_VERSION_RAID5
;
3399 else if (map
->num_members
>= 3)
3400 version
= MPB_VERSION_3OR4_DISK_ARRAY
;
3401 else if (get_imsm_raid_level(map
) == 1)
3402 version
= MPB_VERSION_RAID1
;
3404 version
= MPB_VERSION_RAID0
;
3406 strcpy(((char *) mpb
->sig
) + strlen(MPB_SIGNATURE
), version
);
3410 static int check_name(struct intel_super
*super
, char *name
, int quiet
)
3412 struct imsm_super
*mpb
= super
->anchor
;
3413 char *reason
= NULL
;
3416 if (strlen(name
) > MAX_RAID_SERIAL_LEN
)
3417 reason
= "must be 16 characters or less";
3419 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
3420 struct imsm_dev
*dev
= get_imsm_dev(super
, i
);
3422 if (strncmp((char *) dev
->volume
, name
, MAX_RAID_SERIAL_LEN
) == 0) {
3423 reason
= "already exists";
3428 if (reason
&& !quiet
)
3429 fprintf(stderr
, Name
": imsm volume name %s\n", reason
);
3434 static int init_super_imsm_volume(struct supertype
*st
, mdu_array_info_t
*info
,
3435 unsigned long long size
, char *name
,
3436 char *homehost
, int *uuid
)
3438 /* We are creating a volume inside a pre-existing container.
3439 * so st->sb is already set.
3441 struct intel_super
*super
= st
->sb
;
3442 struct imsm_super
*mpb
= super
->anchor
;
3443 struct intel_dev
*dv
;
3444 struct imsm_dev
*dev
;
3445 struct imsm_vol
*vol
;
3446 struct imsm_map
*map
;
3447 int idx
= mpb
->num_raid_devs
;
3449 unsigned long long array_blocks
;
3450 size_t size_old
, size_new
;
3451 __u32 num_data_stripes
;
3453 if (super
->orom
&& mpb
->num_raid_devs
>= super
->orom
->vpa
) {
3454 fprintf(stderr
, Name
": This imsm-container already has the "
3455 "maximum of %d volumes\n", super
->orom
->vpa
);
3459 /* ensure the mpb is large enough for the new data */
3460 size_old
= __le32_to_cpu(mpb
->mpb_size
);
3461 size_new
= disks_to_mpb_size(info
->nr_disks
);
3462 if (size_new
> size_old
) {
3464 size_t size_round
= ROUND_UP(size_new
, 512);
3466 if (posix_memalign(&mpb_new
, 512, size_round
) != 0) {
3467 fprintf(stderr
, Name
": could not allocate new mpb\n");
3470 memcpy(mpb_new
, mpb
, size_old
);
3473 super
->anchor
= mpb_new
;
3474 mpb
->mpb_size
= __cpu_to_le32(size_new
);
3475 memset(mpb_new
+ size_old
, 0, size_round
- size_old
);
3477 super
->current_vol
= idx
;
3478 /* when creating the first raid device in this container set num_disks
3479 * to zero, i.e. delete this spare and add raid member devices in
3480 * add_to_super_imsm_volume()
3482 if (super
->current_vol
== 0)
3485 if (!check_name(super
, name
, 0))
3487 dv
= malloc(sizeof(*dv
));
3489 fprintf(stderr
, Name
": failed to allocate device list entry\n");
3492 dev
= calloc(1, sizeof(*dev
) + sizeof(__u32
) * (info
->raid_disks
- 1));
3495 fprintf(stderr
, Name
": could not allocate raid device\n");
3499 strncpy((char *) dev
->volume
, name
, MAX_RAID_SERIAL_LEN
);
3500 if (info
->level
== 1)
3501 array_blocks
= info_to_blocks_per_member(info
);
3503 array_blocks
= calc_array_size(info
->level
, info
->raid_disks
,
3504 info
->layout
, info
->chunk_size
,
3506 /* round array size down to closest MB */
3507 array_blocks
= (array_blocks
>> SECT_PER_MB_SHIFT
) << SECT_PER_MB_SHIFT
;
3509 dev
->size_low
= __cpu_to_le32((__u32
) array_blocks
);
3510 dev
->size_high
= __cpu_to_le32((__u32
) (array_blocks
>> 32));
3511 dev
->status
= (DEV_READ_COALESCING
| DEV_WRITE_COALESCING
);
3513 vol
->migr_state
= 0;
3514 set_migr_type(dev
, MIGR_INIT
);
3516 vol
->curr_migr_unit
= 0;
3517 map
= get_imsm_map(dev
, 0);
3518 map
->pba_of_lba0
= __cpu_to_le32(super
->create_offset
);
3519 map
->blocks_per_member
= __cpu_to_le32(info_to_blocks_per_member(info
));
3520 map
->blocks_per_strip
= __cpu_to_le16(info_to_blocks_per_strip(info
));
3521 map
->failed_disk_num
= ~0;
3522 map
->map_state
= info
->level
? IMSM_T_STATE_UNINITIALIZED
:
3523 IMSM_T_STATE_NORMAL
;
3526 if (info
->level
== 1 && info
->raid_disks
> 2) {
3529 fprintf(stderr
, Name
": imsm does not support more than 2 disks"
3530 "in a raid1 volume\n");
3534 map
->raid_level
= info
->level
;
3535 if (info
->level
== 10) {
3536 map
->raid_level
= 1;
3537 map
->num_domains
= info
->raid_disks
/ 2;
3538 } else if (info
->level
== 1)
3539 map
->num_domains
= info
->raid_disks
;
3541 map
->num_domains
= 1;
3543 num_data_stripes
= info_to_num_data_stripes(info
, map
->num_domains
);
3544 map
->num_data_stripes
= __cpu_to_le32(num_data_stripes
);
3546 map
->num_members
= info
->raid_disks
;
3547 for (i
= 0; i
< map
->num_members
; i
++) {
3548 /* initialized in add_to_super */
3549 set_imsm_ord_tbl_ent(map
, i
, IMSM_ORD_REBUILD
);
3551 mpb
->num_raid_devs
++;
3554 dv
->index
= super
->current_vol
;
3555 dv
->next
= super
->devlist
;
3556 super
->devlist
= dv
;
3558 imsm_update_version_info(super
);
3563 static int init_super_imsm(struct supertype
*st
, mdu_array_info_t
*info
,
3564 unsigned long long size
, char *name
,
3565 char *homehost
, int *uuid
)
3567 /* This is primarily called by Create when creating a new array.
3568 * We will then get add_to_super called for each component, and then
3569 * write_init_super called to write it out to each device.
3570 * For IMSM, Create can create on fresh devices or on a pre-existing
3572 * To create on a pre-existing array a different method will be called.
3573 * This one is just for fresh drives.
3575 struct intel_super
*super
;
3576 struct imsm_super
*mpb
;
3581 return init_super_imsm_volume(st
, info
, size
, name
, homehost
, uuid
);
3584 mpb_size
= disks_to_mpb_size(info
->nr_disks
);
3588 super
= alloc_super();
3589 if (super
&& posix_memalign(&super
->buf
, 512, mpb_size
) != 0) {
3594 fprintf(stderr
, Name
3595 ": %s could not allocate superblock\n", __func__
);
3598 memset(super
->buf
, 0, mpb_size
);
3600 mpb
->mpb_size
= __cpu_to_le32(mpb_size
);
3604 /* zeroing superblock */
3608 mpb
->attributes
= MPB_ATTRIB_CHECKSUM_VERIFY
;
3610 version
= (char *) mpb
->sig
;
3611 strcpy(version
, MPB_SIGNATURE
);
3612 version
+= strlen(MPB_SIGNATURE
);
3613 strcpy(version
, MPB_VERSION_RAID0
);
3619 static int add_to_super_imsm_volume(struct supertype
*st
, mdu_disk_info_t
*dk
,
3620 int fd
, char *devname
)
3622 struct intel_super
*super
= st
->sb
;
3623 struct imsm_super
*mpb
= super
->anchor
;
3625 struct imsm_dev
*dev
;
3626 struct imsm_map
*map
;
3629 dev
= get_imsm_dev(super
, super
->current_vol
);
3630 map
= get_imsm_map(dev
, 0);
3632 if (! (dk
->state
& (1<<MD_DISK_SYNC
))) {
3633 fprintf(stderr
, Name
": %s: Cannot add spare devices to IMSM volume\n",
3639 /* we're doing autolayout so grab the pre-marked (in
3640 * validate_geometry) raid_disk
3642 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
3643 if (dl
->raiddisk
== dk
->raid_disk
)
3646 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
3647 if (dl
->major
== dk
->major
&&
3648 dl
->minor
== dk
->minor
)
3653 fprintf(stderr
, Name
": %s is not a member of the same container\n", devname
);
3657 /* add a pristine spare to the metadata */
3658 if (dl
->index
< 0) {
3659 dl
->index
= super
->anchor
->num_disks
;
3660 super
->anchor
->num_disks
++;
3662 /* Check the device has not already been added */
3663 slot
= get_imsm_disk_slot(map
, dl
->index
);
3665 (get_imsm_ord_tbl_ent(dev
, slot
, -1) & IMSM_ORD_REBUILD
) == 0) {
3666 fprintf(stderr
, Name
": %s has been included in this array twice\n",
3670 set_imsm_ord_tbl_ent(map
, dk
->number
, dl
->index
);
3671 dl
->disk
.status
= CONFIGURED_DISK
;
3673 /* if we are creating the first raid device update the family number */
3674 if (super
->current_vol
== 0) {
3676 struct imsm_dev
*_dev
= __get_imsm_dev(mpb
, 0);
3677 struct imsm_disk
*_disk
= __get_imsm_disk(mpb
, dl
->index
);
3679 if (!_dev
|| !_disk
) {
3680 fprintf(stderr
, Name
": BUG mpb setup error\n");
3686 sum
+= __gen_imsm_checksum(mpb
);
3687 mpb
->family_num
= __cpu_to_le32(sum
);
3688 mpb
->orig_family_num
= mpb
->family_num
;
3695 static int add_to_super_imsm(struct supertype
*st
, mdu_disk_info_t
*dk
,
3696 int fd
, char *devname
)
3698 struct intel_super
*super
= st
->sb
;
3700 unsigned long long size
;
3705 /* If we are on an RAID enabled platform check that the disk is
3706 * attached to the raid controller.
3707 * We do not need to test disks attachment for container based additions,
3708 * they shall be already tested when container was created/assembled.
3710 if ((fd
!= -1) && !check_env("IMSM_NO_PLATFORM")) {
3711 struct sys_dev
*hba_name
;
3712 struct intel_hba
*hba
;
3714 hba_name
= find_disk_attached_hba(fd
, NULL
);
3717 Name
": %s is not attached to Intel(R) RAID controller.\n",
3718 devname
? : "disk");
3721 rv
= attach_hba_to_super(super
, hba_name
);
3724 fprintf(stderr
, Name
": %s is attached to Intel(R) %s RAID "
3725 "controller (%s),\n but the container is assigned to Intel(R) "
3726 "%s RAID controller (",
3728 get_sys_dev_type(hba_name
->type
),
3729 hba_name
->pci_id
? : "Err!",
3730 get_sys_dev_type(hba_name
->type
));
3734 fprintf(stderr
, "%s", hba
->pci_id
? : "Err!");
3736 fprintf(stderr
, ", ");
3740 fprintf(stderr
, ").\n"
3741 " Mixing devices attached to different controllers "
3742 "is not allowed.\n");
3743 free_sys_dev(&hba_name
);
3746 free_sys_dev(&hba_name
);
3749 if (super
->current_vol
>= 0)
3750 return add_to_super_imsm_volume(st
, dk
, fd
, devname
);
3753 dd
= malloc(sizeof(*dd
));
3756 Name
": malloc failed %s:%d.\n", __func__
, __LINE__
);
3759 memset(dd
, 0, sizeof(*dd
));
3760 dd
->major
= major(stb
.st_rdev
);
3761 dd
->minor
= minor(stb
.st_rdev
);
3763 dd
->devname
= devname
? strdup(devname
) : NULL
;
3766 dd
->action
= DISK_ADD
;
3767 rv
= imsm_read_serial(fd
, devname
, dd
->serial
);
3770 Name
": failed to retrieve scsi serial, aborting\n");
3775 get_dev_size(fd
, NULL
, &size
);
3777 serialcpy(dd
->disk
.serial
, dd
->serial
);
3778 dd
->disk
.total_blocks
= __cpu_to_le32(size
);
3779 dd
->disk
.status
= SPARE_DISK
;
3780 if (sysfs_disk_to_scsi_id(fd
, &id
) == 0)
3781 dd
->disk
.scsi_id
= __cpu_to_le32(id
);
3783 dd
->disk
.scsi_id
= __cpu_to_le32(0);
3785 if (st
->update_tail
) {
3786 dd
->next
= super
->disk_mgmt_list
;
3787 super
->disk_mgmt_list
= dd
;
3789 dd
->next
= super
->disks
;
3797 static int remove_from_super_imsm(struct supertype
*st
, mdu_disk_info_t
*dk
)
3799 struct intel_super
*super
= st
->sb
;
3802 /* remove from super works only in mdmon - for communication
3803 * manager - monitor. Check if communication memory buffer
3806 if (!st
->update_tail
) {
3808 Name
": %s shall be used in mdmon context only"
3809 "(line %d).\n", __func__
, __LINE__
);
3812 dd
= malloc(sizeof(*dd
));
3815 Name
": malloc failed %s:%d.\n", __func__
, __LINE__
);
3818 memset(dd
, 0, sizeof(*dd
));
3819 dd
->major
= dk
->major
;
3820 dd
->minor
= dk
->minor
;
3823 dd
->disk
.status
= SPARE_DISK
;
3824 dd
->action
= DISK_REMOVE
;
3826 dd
->next
= super
->disk_mgmt_list
;
3827 super
->disk_mgmt_list
= dd
;
3833 static int store_imsm_mpb(int fd
, struct imsm_super
*mpb
);
3837 struct imsm_super anchor
;
3838 } spare_record
__attribute__ ((aligned(512)));
3840 /* spare records have their own family number and do not have any defined raid
3843 static int write_super_imsm_spares(struct intel_super
*super
, int doclose
)
3845 struct imsm_super
*mpb
= super
->anchor
;
3846 struct imsm_super
*spare
= &spare_record
.anchor
;
3850 spare
->mpb_size
= __cpu_to_le32(sizeof(struct imsm_super
)),
3851 spare
->generation_num
= __cpu_to_le32(1UL),
3852 spare
->attributes
= MPB_ATTRIB_CHECKSUM_VERIFY
;
3853 spare
->num_disks
= 1,
3854 spare
->num_raid_devs
= 0,
3855 spare
->cache_size
= mpb
->cache_size
,
3856 spare
->pwr_cycle_count
= __cpu_to_le32(1),
3858 snprintf((char *) spare
->sig
, MAX_SIGNATURE_LENGTH
,
3859 MPB_SIGNATURE MPB_VERSION_RAID0
);
3861 for (d
= super
->disks
; d
; d
= d
->next
) {
3865 spare
->disk
[0] = d
->disk
;
3866 sum
= __gen_imsm_checksum(spare
);
3867 spare
->family_num
= __cpu_to_le32(sum
);
3868 spare
->orig_family_num
= 0;
3869 sum
= __gen_imsm_checksum(spare
);
3870 spare
->check_sum
= __cpu_to_le32(sum
);
3872 if (store_imsm_mpb(d
->fd
, spare
)) {
3873 fprintf(stderr
, "%s: failed for device %d:%d %s\n",
3874 __func__
, d
->major
, d
->minor
, strerror(errno
));
3886 static int write_super_imsm(struct supertype
*st
, int doclose
)
3888 struct intel_super
*super
= st
->sb
;
3889 struct imsm_super
*mpb
= super
->anchor
;
3895 __u32 mpb_size
= sizeof(struct imsm_super
) - sizeof(struct imsm_disk
);
3898 /* 'generation' is incremented everytime the metadata is written */
3899 generation
= __le32_to_cpu(mpb
->generation_num
);
3901 mpb
->generation_num
= __cpu_to_le32(generation
);
3903 /* fix up cases where previous mdadm releases failed to set
3906 if (mpb
->orig_family_num
== 0)
3907 mpb
->orig_family_num
= mpb
->family_num
;
3909 for (d
= super
->disks
; d
; d
= d
->next
) {
3913 mpb
->disk
[d
->index
] = d
->disk
;
3917 for (d
= super
->missing
; d
; d
= d
->next
) {
3918 mpb
->disk
[d
->index
] = d
->disk
;
3921 mpb
->num_disks
= num_disks
;
3922 mpb_size
+= sizeof(struct imsm_disk
) * mpb
->num_disks
;
3924 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
3925 struct imsm_dev
*dev
= __get_imsm_dev(mpb
, i
);
3926 struct imsm_dev
*dev2
= get_imsm_dev(super
, i
);
3928 imsm_copy_dev(dev
, dev2
);
3929 mpb_size
+= sizeof_imsm_dev(dev
, 0);
3932 mpb_size
+= __le32_to_cpu(mpb
->bbm_log_size
);
3933 mpb
->mpb_size
= __cpu_to_le32(mpb_size
);
3935 /* recalculate checksum */
3936 sum
= __gen_imsm_checksum(mpb
);
3937 mpb
->check_sum
= __cpu_to_le32(sum
);
3939 /* write the mpb for disks that compose raid devices */
3940 for (d
= super
->disks
; d
; d
= d
->next
) {
3943 if (store_imsm_mpb(d
->fd
, mpb
))
3944 fprintf(stderr
, "%s: failed for device %d:%d %s\n",
3945 __func__
, d
->major
, d
->minor
, strerror(errno
));
3953 return write_super_imsm_spares(super
, doclose
);
3959 static int create_array(struct supertype
*st
, int dev_idx
)
3962 struct imsm_update_create_array
*u
;
3963 struct intel_super
*super
= st
->sb
;
3964 struct imsm_dev
*dev
= get_imsm_dev(super
, dev_idx
);
3965 struct imsm_map
*map
= get_imsm_map(dev
, 0);
3966 struct disk_info
*inf
;
3967 struct imsm_disk
*disk
;
3970 len
= sizeof(*u
) - sizeof(*dev
) + sizeof_imsm_dev(dev
, 0) +
3971 sizeof(*inf
) * map
->num_members
;
3974 fprintf(stderr
, "%s: failed to allocate update buffer\n",
3979 u
->type
= update_create_array
;
3980 u
->dev_idx
= dev_idx
;
3981 imsm_copy_dev(&u
->dev
, dev
);
3982 inf
= get_disk_info(u
);
3983 for (i
= 0; i
< map
->num_members
; i
++) {
3984 int idx
= get_imsm_disk_idx(dev
, i
, -1);
3986 disk
= get_imsm_disk(super
, idx
);
3987 serialcpy(inf
[i
].serial
, disk
->serial
);
3989 append_metadata_update(st
, u
, len
);
3994 static int mgmt_disk(struct supertype
*st
)
3996 struct intel_super
*super
= st
->sb
;
3998 struct imsm_update_add_remove_disk
*u
;
4000 if (!super
->disk_mgmt_list
)
4006 fprintf(stderr
, "%s: failed to allocate update buffer\n",
4011 u
->type
= update_add_remove_disk
;
4012 append_metadata_update(st
, u
, len
);
4017 static int write_init_super_imsm(struct supertype
*st
)
4019 struct intel_super
*super
= st
->sb
;
4020 int current_vol
= super
->current_vol
;
4022 /* we are done with current_vol reset it to point st at the container */
4023 super
->current_vol
= -1;
4025 if (st
->update_tail
) {
4026 /* queue the recently created array / added disk
4027 * as a metadata update */
4030 /* determine if we are creating a volume or adding a disk */
4031 if (current_vol
< 0) {
4032 /* in the mgmt (add/remove) disk case we are running
4033 * in mdmon context, so don't close fd's
4035 return mgmt_disk(st
);
4037 rv
= create_array(st
, current_vol
);
4042 for (d
= super
->disks
; d
; d
= d
->next
)
4043 Kill(d
->devname
, NULL
, 0, 1, 1);
4044 return write_super_imsm(st
, 1);
4049 static int store_super_imsm(struct supertype
*st
, int fd
)
4051 struct intel_super
*super
= st
->sb
;
4052 struct imsm_super
*mpb
= super
? super
->anchor
: NULL
;
4058 return store_imsm_mpb(fd
, mpb
);
4064 static int imsm_bbm_log_size(struct imsm_super
*mpb
)
4066 return __le32_to_cpu(mpb
->bbm_log_size
);
4070 static int validate_geometry_imsm_container(struct supertype
*st
, int level
,
4071 int layout
, int raiddisks
, int chunk
,
4072 unsigned long long size
, char *dev
,
4073 unsigned long long *freesize
,
4077 unsigned long long ldsize
;
4078 const struct imsm_orom
*orom
;
4080 if (level
!= LEVEL_CONTAINER
)
4085 if (check_env("IMSM_NO_PLATFORM"))
4088 orom
= find_imsm_orom();
4089 if (orom
&& raiddisks
> orom
->tds
) {
4091 fprintf(stderr
, Name
": %d exceeds maximum number of"
4092 " platform supported disks: %d\n",
4093 raiddisks
, orom
->tds
);
4097 fd
= open(dev
, O_RDONLY
|O_EXCL
, 0);
4100 fprintf(stderr
, Name
": imsm: Cannot open %s: %s\n",
4101 dev
, strerror(errno
));
4104 if (!get_dev_size(fd
, dev
, &ldsize
)) {
4110 *freesize
= avail_size_imsm(st
, ldsize
>> 9);
4115 static unsigned long long find_size(struct extent
*e
, int *idx
, int num_extents
)
4117 const unsigned long long base_start
= e
[*idx
].start
;
4118 unsigned long long end
= base_start
+ e
[*idx
].size
;
4121 if (base_start
== end
)
4125 for (i
= *idx
; i
< num_extents
; i
++) {
4126 /* extend overlapping extents */
4127 if (e
[i
].start
>= base_start
&&
4128 e
[i
].start
<= end
) {
4131 if (e
[i
].start
+ e
[i
].size
> end
)
4132 end
= e
[i
].start
+ e
[i
].size
;
4133 } else if (e
[i
].start
> end
) {
4139 return end
- base_start
;
4142 static unsigned long long merge_extents(struct intel_super
*super
, int sum_extents
)
4144 /* build a composite disk with all known extents and generate a new
4145 * 'maxsize' given the "all disks in an array must share a common start
4146 * offset" constraint
4148 struct extent
*e
= calloc(sum_extents
, sizeof(*e
));
4152 unsigned long long pos
;
4153 unsigned long long start
= 0;
4154 unsigned long long maxsize
;
4155 unsigned long reserve
;
4160 /* coalesce and sort all extents. also, check to see if we need to
4161 * reserve space between member arrays
4164 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
4167 for (i
= 0; i
< dl
->extent_cnt
; i
++)
4170 qsort(e
, sum_extents
, sizeof(*e
), cmp_extent
);
4175 while (i
< sum_extents
) {
4176 e
[j
].start
= e
[i
].start
;
4177 e
[j
].size
= find_size(e
, &i
, sum_extents
);
4179 if (e
[j
-1].size
== 0)
4188 unsigned long long esize
;
4190 esize
= e
[i
].start
- pos
;
4191 if (esize
>= maxsize
) {
4196 pos
= e
[i
].start
+ e
[i
].size
;
4198 } while (e
[i
-1].size
);
4204 /* FIXME assumes volume at offset 0 is the first volume in a
4207 if (start_extent
> 0)
4208 reserve
= IMSM_RESERVED_SECTORS
; /* gap between raid regions */
4212 if (maxsize
< reserve
)
4215 super
->create_offset
= ~((__u32
) 0);
4216 if (start
+ reserve
> super
->create_offset
)
4217 return 0; /* start overflows create_offset */
4218 super
->create_offset
= start
+ reserve
;
4220 return maxsize
- reserve
;
4223 static int is_raid_level_supported(const struct imsm_orom
*orom
, int level
, int raiddisks
)
4225 if (level
< 0 || level
== 6 || level
== 4)
4228 /* if we have an orom prevent invalid raid levels */
4231 case 0: return imsm_orom_has_raid0(orom
);
4234 return imsm_orom_has_raid1e(orom
);
4235 return imsm_orom_has_raid1(orom
) && raiddisks
== 2;
4236 case 10: return imsm_orom_has_raid10(orom
) && raiddisks
== 4;
4237 case 5: return imsm_orom_has_raid5(orom
) && raiddisks
> 2;
4240 return 1; /* not on an Intel RAID platform so anything goes */
4246 #define pr_vrb(fmt, arg...) (void) (verbose && fprintf(stderr, Name fmt, ##arg))
4248 * validate volume parameters with OROM/EFI capabilities
4251 validate_geometry_imsm_orom(struct intel_super
*super
, int level
, int layout
,
4252 int raiddisks
, int *chunk
, int verbose
)
4257 /* validate container capabilities */
4258 if (super
->orom
&& raiddisks
> super
->orom
->tds
) {
4260 fprintf(stderr
, Name
": %d exceeds maximum number of"
4261 " platform supported disks: %d\n",
4262 raiddisks
, super
->orom
->tds
);
4266 /* capabilities of OROM tested - copied from validate_geometry_imsm_volume */
4267 if (super
->orom
&& (!is_raid_level_supported(super
->orom
, level
,
4269 pr_vrb(": platform does not support raid%d with %d disk%s\n",
4270 level
, raiddisks
, raiddisks
> 1 ? "s" : "");
4273 if (super
->orom
&& level
!= 1) {
4274 if (chunk
&& (*chunk
== 0 || *chunk
== UnSet
))
4275 *chunk
= imsm_orom_default_chunk(super
->orom
);
4276 else if (chunk
&& !imsm_orom_has_chunk(super
->orom
, *chunk
)) {
4277 pr_vrb(": platform does not support a chunk size of: "
4282 if (layout
!= imsm_level_to_layout(level
)) {
4284 pr_vrb(": imsm raid 5 only supports the left-asymmetric layout\n");
4285 else if (level
== 10)
4286 pr_vrb(": imsm raid 10 only supports the n2 layout\n");
4288 pr_vrb(": imsm unknown layout %#x for this raid level %d\n",
4295 /* validate_geometry_imsm_volume - lifted from validate_geometry_ddf_bvd
4296 * FIX ME add ahci details
4298 static int validate_geometry_imsm_volume(struct supertype
*st
, int level
,
4299 int layout
, int raiddisks
, int *chunk
,
4300 unsigned long long size
, char *dev
,
4301 unsigned long long *freesize
,
4305 struct intel_super
*super
= st
->sb
;
4306 struct imsm_super
*mpb
= super
->anchor
;
4308 unsigned long long pos
= 0;
4309 unsigned long long maxsize
;
4313 /* We must have the container info already read in. */
4317 if (!validate_geometry_imsm_orom(super
, level
, layout
, raiddisks
, chunk
, verbose
)) {
4318 fprintf(stderr
, Name
": RAID gemetry validation failed. "
4319 "Cannot proceed with the action(s).\n");
4323 /* General test: make sure there is space for
4324 * 'raiddisks' device extents of size 'size' at a given
4327 unsigned long long minsize
= size
;
4328 unsigned long long start_offset
= MaxSector
;
4331 minsize
= MPB_SECTOR_CNT
+ IMSM_RESERVED_SECTORS
;
4332 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
4337 e
= get_extents(super
, dl
);
4340 unsigned long long esize
;
4341 esize
= e
[i
].start
- pos
;
4342 if (esize
>= minsize
)
4344 if (found
&& start_offset
== MaxSector
) {
4347 } else if (found
&& pos
!= start_offset
) {
4351 pos
= e
[i
].start
+ e
[i
].size
;
4353 } while (e
[i
-1].size
);
4358 if (dcnt
< raiddisks
) {
4360 fprintf(stderr
, Name
": imsm: Not enough "
4361 "devices with space for this array "
4369 /* This device must be a member of the set */
4370 if (stat(dev
, &stb
) < 0)
4372 if ((S_IFMT
& stb
.st_mode
) != S_IFBLK
)
4374 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
4375 if (dl
->major
== (int)major(stb
.st_rdev
) &&
4376 dl
->minor
== (int)minor(stb
.st_rdev
))
4381 fprintf(stderr
, Name
": %s is not in the "
4382 "same imsm set\n", dev
);
4384 } else if (super
->orom
&& dl
->index
< 0 && mpb
->num_raid_devs
) {
4385 /* If a volume is present then the current creation attempt
4386 * cannot incorporate new spares because the orom may not
4387 * understand this configuration (all member disks must be
4388 * members of each array in the container).
4390 fprintf(stderr
, Name
": %s is a spare and a volume"
4391 " is already defined for this container\n", dev
);
4392 fprintf(stderr
, Name
": The option-rom requires all member"
4393 " disks to be a member of all volumes\n");
4397 /* retrieve the largest free space block */
4398 e
= get_extents(super
, dl
);
4403 unsigned long long esize
;
4405 esize
= e
[i
].start
- pos
;
4406 if (esize
>= maxsize
)
4408 pos
= e
[i
].start
+ e
[i
].size
;
4410 } while (e
[i
-1].size
);
4415 fprintf(stderr
, Name
": unable to determine free space for: %s\n",
4419 if (maxsize
< size
) {
4421 fprintf(stderr
, Name
": %s not enough space (%llu < %llu)\n",
4422 dev
, maxsize
, size
);
4426 /* count total number of extents for merge */
4428 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
4430 i
+= dl
->extent_cnt
;
4432 maxsize
= merge_extents(super
, i
);
4433 if (maxsize
< size
|| maxsize
== 0) {
4435 fprintf(stderr
, Name
": not enough space after merge (%llu < %llu)\n",
4440 *freesize
= maxsize
;
4445 static int reserve_space(struct supertype
*st
, int raiddisks
,
4446 unsigned long long size
, int chunk
,
4447 unsigned long long *freesize
)
4449 struct intel_super
*super
= st
->sb
;
4450 struct imsm_super
*mpb
= super
->anchor
;
4455 unsigned long long maxsize
;
4456 unsigned long long minsize
;
4460 /* find the largest common start free region of the possible disks */
4464 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
4470 /* don't activate new spares if we are orom constrained
4471 * and there is already a volume active in the container
4473 if (super
->orom
&& dl
->index
< 0 && mpb
->num_raid_devs
)
4476 e
= get_extents(super
, dl
);
4479 for (i
= 1; e
[i
-1].size
; i
++)
4487 maxsize
= merge_extents(super
, extent_cnt
);
4491 minsize
= chunk
* 2;
4493 if (cnt
< raiddisks
||
4494 (super
->orom
&& used
&& used
!= raiddisks
) ||
4495 maxsize
< minsize
||
4497 fprintf(stderr
, Name
": not enough devices with space to create array.\n");
4498 return 0; /* No enough free spaces large enough */
4510 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
4512 dl
->raiddisk
= cnt
++;
4519 static int validate_geometry_imsm(struct supertype
*st
, int level
, int layout
,
4520 int raiddisks
, int *chunk
, unsigned long long size
,
4521 char *dev
, unsigned long long *freesize
,
4529 * if given unused devices create a container
4530 * if given given devices in a container create a member volume
4532 if (level
== LEVEL_CONTAINER
) {
4533 /* Must be a fresh device to add to a container */
4534 return validate_geometry_imsm_container(st
, level
, layout
,
4536 chunk
?*chunk
:0, size
,
4542 if (st
->sb
&& freesize
) {
4543 /* we are being asked to automatically layout a
4544 * new volume based on the current contents of
4545 * the container. If the the parameters can be
4546 * satisfied reserve_space will record the disks,
4547 * start offset, and size of the volume to be
4548 * created. add_to_super and getinfo_super
4549 * detect when autolayout is in progress.
4551 if (!validate_geometry_imsm_orom(st
->sb
, level
, layout
,
4555 return reserve_space(st
, raiddisks
, size
,
4556 chunk
?*chunk
:0, freesize
);
4561 /* creating in a given container */
4562 return validate_geometry_imsm_volume(st
, level
, layout
,
4563 raiddisks
, chunk
, size
,
4564 dev
, freesize
, verbose
);
4567 /* This device needs to be a device in an 'imsm' container */
4568 fd
= open(dev
, O_RDONLY
|O_EXCL
, 0);
4572 Name
": Cannot create this array on device %s\n",
4577 if (errno
!= EBUSY
|| (fd
= open(dev
, O_RDONLY
, 0)) < 0) {
4579 fprintf(stderr
, Name
": Cannot open %s: %s\n",
4580 dev
, strerror(errno
));
4583 /* Well, it is in use by someone, maybe an 'imsm' container. */
4584 cfd
= open_container(fd
);
4588 fprintf(stderr
, Name
": Cannot use %s: It is busy\n",
4592 sra
= sysfs_read(cfd
, 0, GET_VERSION
);
4593 if (sra
&& sra
->array
.major_version
== -1 &&
4594 strcmp(sra
->text_version
, "imsm") == 0)
4598 /* This is a member of a imsm container. Load the container
4599 * and try to create a volume
4601 struct intel_super
*super
;
4603 if (load_super_imsm_all(st
, cfd
, (void **) &super
, NULL
) == 0) {
4605 st
->container_dev
= fd2devnum(cfd
);
4607 return validate_geometry_imsm_volume(st
, level
, layout
,
4615 fprintf(stderr
, Name
": failed container membership check\n");
4621 static void default_geometry_imsm(struct supertype
*st
, int *level
, int *layout
, int *chunk
)
4623 struct intel_super
*super
= st
->sb
;
4625 if (level
&& *level
== UnSet
)
4626 *level
= LEVEL_CONTAINER
;
4628 if (level
&& layout
&& *layout
== UnSet
)
4629 *layout
= imsm_level_to_layout(*level
);
4631 if (chunk
&& (*chunk
== UnSet
|| *chunk
== 0) &&
4632 super
&& super
->orom
)
4633 *chunk
= imsm_orom_default_chunk(super
->orom
);
4636 static void handle_missing(struct intel_super
*super
, struct imsm_dev
*dev
);
4638 static int kill_subarray_imsm(struct supertype
*st
)
4640 /* remove the subarray currently referenced by ->current_vol */
4642 struct intel_dev
**dp
;
4643 struct intel_super
*super
= st
->sb
;
4644 __u8 current_vol
= super
->current_vol
;
4645 struct imsm_super
*mpb
= super
->anchor
;
4647 if (super
->current_vol
< 0)
4649 super
->current_vol
= -1; /* invalidate subarray cursor */
4651 /* block deletions that would change the uuid of active subarrays
4653 * FIXME when immutable ids are available, but note that we'll
4654 * also need to fixup the invalidated/active subarray indexes in
4657 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
4660 if (i
< current_vol
)
4662 sprintf(subarray
, "%u", i
);
4663 if (is_subarray_active(subarray
, st
->devname
)) {
4665 Name
": deleting subarray-%d would change the UUID of active subarray-%d, aborting\n",
4672 if (st
->update_tail
) {
4673 struct imsm_update_kill_array
*u
= malloc(sizeof(*u
));
4677 u
->type
= update_kill_array
;
4678 u
->dev_idx
= current_vol
;
4679 append_metadata_update(st
, u
, sizeof(*u
));
4684 for (dp
= &super
->devlist
; *dp
;)
4685 if ((*dp
)->index
== current_vol
) {
4688 handle_missing(super
, (*dp
)->dev
);
4689 if ((*dp
)->index
> current_vol
)
4694 /* no more raid devices, all active components are now spares,
4695 * but of course failed are still failed
4697 if (--mpb
->num_raid_devs
== 0) {
4700 for (d
= super
->disks
; d
; d
= d
->next
)
4701 if (d
->index
> -2) {
4703 d
->disk
.status
= SPARE_DISK
;
4707 super
->updates_pending
++;
4712 static int update_subarray_imsm(struct supertype
*st
, char *subarray
,
4713 char *update
, struct mddev_ident
*ident
)
4715 /* update the subarray currently referenced by ->current_vol */
4716 struct intel_super
*super
= st
->sb
;
4717 struct imsm_super
*mpb
= super
->anchor
;
4719 if (strcmp(update
, "name") == 0) {
4720 char *name
= ident
->name
;
4724 if (is_subarray_active(subarray
, st
->devname
)) {
4726 Name
": Unable to update name of active subarray\n");
4730 if (!check_name(super
, name
, 0))
4733 vol
= strtoul(subarray
, &ep
, 10);
4734 if (*ep
!= '\0' || vol
>= super
->anchor
->num_raid_devs
)
4737 if (st
->update_tail
) {
4738 struct imsm_update_rename_array
*u
= malloc(sizeof(*u
));
4742 u
->type
= update_rename_array
;
4744 snprintf((char *) u
->name
, MAX_RAID_SERIAL_LEN
, "%s", name
);
4745 append_metadata_update(st
, u
, sizeof(*u
));
4747 struct imsm_dev
*dev
;
4750 dev
= get_imsm_dev(super
, vol
);
4751 snprintf((char *) dev
->volume
, MAX_RAID_SERIAL_LEN
, "%s", name
);
4752 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
4753 dev
= get_imsm_dev(super
, i
);
4754 handle_missing(super
, dev
);
4756 super
->updates_pending
++;
4764 static int is_gen_migration(struct imsm_dev
*dev
)
4766 if (!dev
->vol
.migr_state
)
4769 if (migr_type(dev
) == MIGR_GEN_MIGR
)
4774 #endif /* MDASSEMBLE */
4776 static int is_rebuilding(struct imsm_dev
*dev
)
4778 struct imsm_map
*migr_map
;
4780 if (!dev
->vol
.migr_state
)
4783 if (migr_type(dev
) != MIGR_REBUILD
)
4786 migr_map
= get_imsm_map(dev
, 1);
4788 if (migr_map
->map_state
== IMSM_T_STATE_DEGRADED
)
4794 static void update_recovery_start(struct imsm_dev
*dev
, struct mdinfo
*array
)
4796 struct mdinfo
*rebuild
= NULL
;
4800 if (!is_rebuilding(dev
))
4803 /* Find the rebuild target, but punt on the dual rebuild case */
4804 for (d
= array
->devs
; d
; d
= d
->next
)
4805 if (d
->recovery_start
== 0) {
4812 /* (?) none of the disks are marked with
4813 * IMSM_ORD_REBUILD, so assume they are missing and the
4814 * disk_ord_tbl was not correctly updated
4816 dprintf("%s: failed to locate out-of-sync disk\n", __func__
);
4820 units
= __le32_to_cpu(dev
->vol
.curr_migr_unit
);
4821 rebuild
->recovery_start
= units
* blocks_per_migr_unit(dev
);
4825 static struct mdinfo
*container_content_imsm(struct supertype
*st
, char *subarray
)
4827 /* Given a container loaded by load_super_imsm_all,
4828 * extract information about all the arrays into
4830 * If 'subarray' is given, just extract info about that array.
4832 * For each imsm_dev create an mdinfo, fill it in,
4833 * then look for matching devices in super->disks
4834 * and create appropriate device mdinfo.
4836 struct intel_super
*super
= st
->sb
;
4837 struct imsm_super
*mpb
= super
->anchor
;
4838 struct mdinfo
*rest
= NULL
;
4842 int spare_disks
= 0;
4844 /* check for bad blocks */
4845 if (imsm_bbm_log_size(super
->anchor
))
4848 /* count spare devices, not used in maps
4850 for (d
= super
->disks
; d
; d
= d
->next
)
4854 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
4855 struct imsm_dev
*dev
;
4856 struct imsm_map
*map
;
4857 struct imsm_map
*map2
;
4858 struct mdinfo
*this;
4863 (i
!= strtoul(subarray
, &ep
, 10) || *ep
!= '\0'))
4866 dev
= get_imsm_dev(super
, i
);
4867 map
= get_imsm_map(dev
, 0);
4868 map2
= get_imsm_map(dev
, 1);
4870 /* do not publish arrays that are in the middle of an
4871 * unsupported migration
4873 if (dev
->vol
.migr_state
&&
4874 (migr_type(dev
) == MIGR_STATE_CHANGE
)) {
4875 fprintf(stderr
, Name
": cannot assemble volume '%.16s':"
4876 " unsupported migration in progress\n",
4880 /* do not publish arrays that are not support by controller's
4884 chunk
= __le16_to_cpu(map
->blocks_per_strip
) >> 1;
4885 if (!validate_geometry_imsm_orom(super
,
4886 get_imsm_raid_level(map
), /* RAID level */
4887 imsm_level_to_layout(get_imsm_raid_level(map
)),
4888 map
->num_members
, /* raid disks */
4891 fprintf(stderr
, Name
": RAID gemetry validation failed. "
4892 "Cannot proceed with the action(s).\n");
4895 this = malloc(sizeof(*this));
4897 fprintf(stderr
, Name
": failed to allocate %zu bytes\n",
4901 memset(this, 0, sizeof(*this));
4904 super
->current_vol
= i
;
4905 getinfo_super_imsm_volume(st
, this, NULL
);
4906 for (slot
= 0 ; slot
< map
->num_members
; slot
++) {
4907 unsigned long long recovery_start
;
4908 struct mdinfo
*info_d
;
4915 idx
= get_imsm_disk_idx(dev
, slot
, 0);
4916 ord
= get_imsm_ord_tbl_ent(dev
, slot
, -1);
4917 for (d
= super
->disks
; d
; d
= d
->next
)
4918 if (d
->index
== idx
)
4921 recovery_start
= MaxSector
;
4924 if (d
&& is_failed(&d
->disk
))
4926 if (ord
& IMSM_ORD_REBUILD
)
4930 * if we skip some disks the array will be assmebled degraded;
4931 * reset resync start to avoid a dirty-degraded
4932 * situation when performing the intial sync
4934 * FIXME handle dirty degraded
4936 if ((skip
|| recovery_start
== 0) && !dev
->vol
.dirty
)
4937 this->resync_start
= MaxSector
;
4941 info_d
= calloc(1, sizeof(*info_d
));
4943 fprintf(stderr
, Name
": failed to allocate disk"
4944 " for volume %.16s\n", dev
->volume
);
4945 info_d
= this->devs
;
4947 struct mdinfo
*d
= info_d
->next
;
4956 info_d
->next
= this->devs
;
4957 this->devs
= info_d
;
4959 info_d
->disk
.number
= d
->index
;
4960 info_d
->disk
.major
= d
->major
;
4961 info_d
->disk
.minor
= d
->minor
;
4962 info_d
->disk
.raid_disk
= slot
;
4963 info_d
->recovery_start
= recovery_start
;
4965 if (slot
< map2
->num_members
)
4966 info_d
->disk
.state
= (1 << MD_DISK_ACTIVE
);
4968 this->array
.spare_disks
++;
4970 if (slot
< map
->num_members
)
4971 info_d
->disk
.state
= (1 << MD_DISK_ACTIVE
);
4973 this->array
.spare_disks
++;
4975 if (info_d
->recovery_start
== MaxSector
)
4976 this->array
.working_disks
++;
4978 info_d
->events
= __le32_to_cpu(mpb
->generation_num
);
4979 info_d
->data_offset
= __le32_to_cpu(map
->pba_of_lba0
);
4980 info_d
->component_size
= __le32_to_cpu(map
->blocks_per_member
);
4982 /* now that the disk list is up-to-date fixup recovery_start */
4983 update_recovery_start(dev
, this);
4984 this->array
.spare_disks
+= spare_disks
;
4988 /* if array has bad blocks, set suitable bit in array status */
4990 rest
->array
.state
|= (1<<MD_SB_BBM_ERRORS
);
4996 static __u8
imsm_check_degraded(struct intel_super
*super
, struct imsm_dev
*dev
, int failed
)
4998 struct imsm_map
*map
= get_imsm_map(dev
, 0);
5001 return map
->map_state
== IMSM_T_STATE_UNINITIALIZED
?
5002 IMSM_T_STATE_UNINITIALIZED
: IMSM_T_STATE_NORMAL
;
5004 switch (get_imsm_raid_level(map
)) {
5006 return IMSM_T_STATE_FAILED
;
5009 if (failed
< map
->num_members
)
5010 return IMSM_T_STATE_DEGRADED
;
5012 return IMSM_T_STATE_FAILED
;
5017 * check to see if any mirrors have failed, otherwise we
5018 * are degraded. Even numbered slots are mirrored on
5022 /* gcc -Os complains that this is unused */
5023 int insync
= insync
;
5025 for (i
= 0; i
< map
->num_members
; i
++) {
5026 __u32 ord
= get_imsm_ord_tbl_ent(dev
, i
, -1);
5027 int idx
= ord_to_idx(ord
);
5028 struct imsm_disk
*disk
;
5030 /* reset the potential in-sync count on even-numbered
5031 * slots. num_copies is always 2 for imsm raid10
5036 disk
= get_imsm_disk(super
, idx
);
5037 if (!disk
|| is_failed(disk
) || ord
& IMSM_ORD_REBUILD
)
5040 /* no in-sync disks left in this mirror the
5044 return IMSM_T_STATE_FAILED
;
5047 return IMSM_T_STATE_DEGRADED
;
5051 return IMSM_T_STATE_DEGRADED
;
5053 return IMSM_T_STATE_FAILED
;
5059 return map
->map_state
;
5062 static int imsm_count_failed(struct intel_super
*super
, struct imsm_dev
*dev
)
5066 struct imsm_disk
*disk
;
5067 struct imsm_map
*map
= get_imsm_map(dev
, 0);
5068 struct imsm_map
*prev
= get_imsm_map(dev
, dev
->vol
.migr_state
);
5072 /* at the beginning of migration we set IMSM_ORD_REBUILD on
5073 * disks that are being rebuilt. New failures are recorded to
5074 * map[0]. So we look through all the disks we started with and
5075 * see if any failures are still present, or if any new ones
5078 * FIXME add support for online capacity expansion and
5079 * raid-level-migration
5081 for (i
= 0; i
< prev
->num_members
; i
++) {
5082 ord
= __le32_to_cpu(prev
->disk_ord_tbl
[i
]);
5083 ord
|= __le32_to_cpu(map
->disk_ord_tbl
[i
]);
5084 idx
= ord_to_idx(ord
);
5086 disk
= get_imsm_disk(super
, idx
);
5087 if (!disk
|| is_failed(disk
) || ord
& IMSM_ORD_REBUILD
)
5095 static int imsm_open_new(struct supertype
*c
, struct active_array
*a
,
5098 struct intel_super
*super
= c
->sb
;
5099 struct imsm_super
*mpb
= super
->anchor
;
5101 if (atoi(inst
) >= mpb
->num_raid_devs
) {
5102 fprintf(stderr
, "%s: subarry index %d, out of range\n",
5103 __func__
, atoi(inst
));
5107 dprintf("imsm: open_new %s\n", inst
);
5108 a
->info
.container_member
= atoi(inst
);
5112 static int is_resyncing(struct imsm_dev
*dev
)
5114 struct imsm_map
*migr_map
;
5116 if (!dev
->vol
.migr_state
)
5119 if (migr_type(dev
) == MIGR_INIT
||
5120 migr_type(dev
) == MIGR_REPAIR
)
5123 if (migr_type(dev
) == MIGR_GEN_MIGR
)
5126 migr_map
= get_imsm_map(dev
, 1);
5128 if ((migr_map
->map_state
== IMSM_T_STATE_NORMAL
) &&
5129 (dev
->vol
.migr_type
!= MIGR_GEN_MIGR
))
5135 /* return true if we recorded new information */
5136 static int mark_failure(struct imsm_dev
*dev
, struct imsm_disk
*disk
, int idx
)
5140 struct imsm_map
*map
;
5141 char buf
[MAX_RAID_SERIAL_LEN
+3];
5142 unsigned int len
, shift
= 0;
5144 /* new failures are always set in map[0] */
5145 map
= get_imsm_map(dev
, 0);
5147 slot
= get_imsm_disk_slot(map
, idx
);
5151 ord
= __le32_to_cpu(map
->disk_ord_tbl
[slot
]);
5152 if (is_failed(disk
) && (ord
& IMSM_ORD_REBUILD
))
5155 sprintf(buf
, "%s:0", disk
->serial
);
5156 if ((len
= strlen(buf
)) >= MAX_RAID_SERIAL_LEN
)
5157 shift
= len
- MAX_RAID_SERIAL_LEN
+ 1;
5158 strncpy((char *)disk
->serial
, &buf
[shift
], MAX_RAID_SERIAL_LEN
);
5160 disk
->status
|= FAILED_DISK
;
5161 set_imsm_ord_tbl_ent(map
, slot
, idx
| IMSM_ORD_REBUILD
);
5162 if (map
->failed_disk_num
== 0xff)
5163 map
->failed_disk_num
= slot
;
5167 static void mark_missing(struct imsm_dev
*dev
, struct imsm_disk
*disk
, int idx
)
5169 mark_failure(dev
, disk
, idx
);
5171 if (disk
->scsi_id
== __cpu_to_le32(~(__u32
)0))
5174 disk
->scsi_id
= __cpu_to_le32(~(__u32
)0);
5175 memmove(&disk
->serial
[0], &disk
->serial
[1], MAX_RAID_SERIAL_LEN
- 1);
5178 static void handle_missing(struct intel_super
*super
, struct imsm_dev
*dev
)
5184 if (!super
->missing
)
5186 failed
= imsm_count_failed(super
, dev
);
5187 map_state
= imsm_check_degraded(super
, dev
, failed
);
5189 dprintf("imsm: mark missing\n");
5190 end_migration(dev
, map_state
);
5191 for (dl
= super
->missing
; dl
; dl
= dl
->next
)
5192 mark_missing(dev
, &dl
->disk
, dl
->index
);
5193 super
->updates_pending
++;
5196 static unsigned long long imsm_set_array_size(struct imsm_dev
*dev
)
5198 int used_disks
= imsm_num_data_members(dev
, 0);
5199 unsigned long long array_blocks
;
5200 struct imsm_map
*map
;
5202 if (used_disks
== 0) {
5203 /* when problems occures
5204 * return current array_blocks value
5206 array_blocks
= __le32_to_cpu(dev
->size_high
);
5207 array_blocks
= array_blocks
<< 32;
5208 array_blocks
+= __le32_to_cpu(dev
->size_low
);
5210 return array_blocks
;
5213 /* set array size in metadata
5215 map
= get_imsm_map(dev
, 0);
5216 array_blocks
= map
->blocks_per_member
* used_disks
;
5218 /* round array size down to closest MB
5220 array_blocks
= (array_blocks
>> SECT_PER_MB_SHIFT
) << SECT_PER_MB_SHIFT
;
5221 dev
->size_low
= __cpu_to_le32((__u32
)array_blocks
);
5222 dev
->size_high
= __cpu_to_le32((__u32
)(array_blocks
>> 32));
5224 return array_blocks
;
5227 static void imsm_set_disk(struct active_array
*a
, int n
, int state
);
5229 static void imsm_progress_container_reshape(struct intel_super
*super
)
5231 /* if no device has a migr_state, but some device has a
5232 * different number of members than the previous device, start
5233 * changing the number of devices in this device to match
5236 struct imsm_super
*mpb
= super
->anchor
;
5237 int prev_disks
= -1;
5241 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
5242 struct imsm_dev
*dev
= get_imsm_dev(super
, i
);
5243 struct imsm_map
*map
= get_imsm_map(dev
, 0);
5244 struct imsm_map
*map2
;
5245 int prev_num_members
;
5247 if (dev
->vol
.migr_state
)
5250 if (prev_disks
== -1)
5251 prev_disks
= map
->num_members
;
5252 if (prev_disks
== map
->num_members
)
5255 /* OK, this array needs to enter reshape mode.
5256 * i.e it needs a migr_state
5259 copy_map_size
= sizeof_imsm_map(map
);
5260 prev_num_members
= map
->num_members
;
5261 map
->num_members
= prev_disks
;
5262 dev
->vol
.migr_state
= 1;
5263 dev
->vol
.curr_migr_unit
= 0;
5264 dev
->vol
.migr_type
= MIGR_GEN_MIGR
;
5265 for (i
= prev_num_members
;
5266 i
< map
->num_members
; i
++)
5267 set_imsm_ord_tbl_ent(map
, i
, i
);
5268 map2
= get_imsm_map(dev
, 1);
5269 /* Copy the current map */
5270 memcpy(map2
, map
, copy_map_size
);
5271 map2
->num_members
= prev_num_members
;
5273 imsm_set_array_size(dev
);
5274 super
->updates_pending
++;
5278 /* Handle dirty -> clean transititions, resync and reshape. Degraded and rebuild
5279 * states are handled in imsm_set_disk() with one exception, when a
5280 * resync is stopped due to a new failure this routine will set the
5281 * 'degraded' state for the array.
5283 static int imsm_set_array_state(struct active_array
*a
, int consistent
)
5285 int inst
= a
->info
.container_member
;
5286 struct intel_super
*super
= a
->container
->sb
;
5287 struct imsm_dev
*dev
= get_imsm_dev(super
, inst
);
5288 struct imsm_map
*map
= get_imsm_map(dev
, 0);
5289 int failed
= imsm_count_failed(super
, dev
);
5290 __u8 map_state
= imsm_check_degraded(super
, dev
, failed
);
5291 __u32 blocks_per_unit
;
5293 if (dev
->vol
.migr_state
&&
5294 dev
->vol
.migr_type
== MIGR_GEN_MIGR
) {
5295 /* array state change is blocked due to reshape action
5297 * - abort the reshape (if last_checkpoint is 0 and action!= reshape)
5298 * - finish the reshape (if last_checkpoint is big and action != reshape)
5299 * - update curr_migr_unit
5301 if (a
->curr_action
== reshape
) {
5302 /* still reshaping, maybe update curr_migr_unit */
5303 goto mark_checkpoint
;
5305 if (a
->last_checkpoint
== 0 && a
->prev_action
== reshape
) {
5306 /* for some reason we aborted the reshape.
5309 struct imsm_map
*map2
= get_imsm_map(dev
, 1);
5310 dev
->vol
.migr_state
= 0;
5311 dev
->vol
.migr_type
= 0;
5312 dev
->vol
.curr_migr_unit
= 0;
5313 memcpy(map
, map2
, sizeof_imsm_map(map2
));
5314 super
->updates_pending
++;
5316 if (a
->last_checkpoint
>= a
->info
.component_size
) {
5317 unsigned long long array_blocks
;
5321 used_disks
= imsm_num_data_members(dev
, 0);
5322 if (used_disks
> 0) {
5324 map
->blocks_per_member
*
5326 /* round array size down to closest MB
5328 array_blocks
= (array_blocks
5329 >> SECT_PER_MB_SHIFT
)
5330 << SECT_PER_MB_SHIFT
;
5331 a
->info
.custom_array_size
= array_blocks
;
5332 /* encourage manager to update array
5336 a
->check_reshape
= 1;
5338 /* finalize online capacity expansion/reshape */
5339 for (mdi
= a
->info
.devs
; mdi
; mdi
= mdi
->next
)
5341 mdi
->disk
.raid_disk
,
5344 imsm_progress_container_reshape(super
);
5349 /* before we activate this array handle any missing disks */
5350 if (consistent
== 2)
5351 handle_missing(super
, dev
);
5353 if (consistent
== 2 &&
5354 (!is_resync_complete(&a
->info
) ||
5355 map_state
!= IMSM_T_STATE_NORMAL
||
5356 dev
->vol
.migr_state
))
5359 if (is_resync_complete(&a
->info
)) {
5360 /* complete intialization / resync,
5361 * recovery and interrupted recovery is completed in
5364 if (is_resyncing(dev
)) {
5365 dprintf("imsm: mark resync done\n");
5366 end_migration(dev
, map_state
);
5367 super
->updates_pending
++;
5368 a
->last_checkpoint
= 0;
5370 } else if (!is_resyncing(dev
) && !failed
) {
5371 /* mark the start of the init process if nothing is failed */
5372 dprintf("imsm: mark resync start\n");
5373 if (map
->map_state
== IMSM_T_STATE_UNINITIALIZED
)
5374 migrate(dev
, IMSM_T_STATE_NORMAL
, MIGR_INIT
);
5376 migrate(dev
, IMSM_T_STATE_NORMAL
, MIGR_REPAIR
);
5377 super
->updates_pending
++;
5381 /* check if we can update curr_migr_unit from resync_start, recovery_start */
5382 blocks_per_unit
= blocks_per_migr_unit(dev
);
5383 if (blocks_per_unit
) {
5387 units
= a
->last_checkpoint
/ blocks_per_unit
;
5390 /* check that we did not overflow 32-bits, and that
5391 * curr_migr_unit needs updating
5393 if (units32
== units
&&
5394 __le32_to_cpu(dev
->vol
.curr_migr_unit
) != units32
) {
5395 dprintf("imsm: mark checkpoint (%u)\n", units32
);
5396 dev
->vol
.curr_migr_unit
= __cpu_to_le32(units32
);
5397 super
->updates_pending
++;
5401 /* mark dirty / clean */
5402 if (dev
->vol
.dirty
!= !consistent
) {
5403 dprintf("imsm: mark '%s'\n", consistent
? "clean" : "dirty");
5408 super
->updates_pending
++;
5414 static void imsm_set_disk(struct active_array
*a
, int n
, int state
)
5416 int inst
= a
->info
.container_member
;
5417 struct intel_super
*super
= a
->container
->sb
;
5418 struct imsm_dev
*dev
= get_imsm_dev(super
, inst
);
5419 struct imsm_map
*map
= get_imsm_map(dev
, 0);
5420 struct imsm_disk
*disk
;
5425 if (n
> map
->num_members
)
5426 fprintf(stderr
, "imsm: set_disk %d out of range 0..%d\n",
5427 n
, map
->num_members
- 1);
5432 dprintf("imsm: set_disk %d:%x\n", n
, state
);
5434 ord
= get_imsm_ord_tbl_ent(dev
, n
, -1);
5435 disk
= get_imsm_disk(super
, ord_to_idx(ord
));
5437 /* check for new failures */
5438 if (state
& DS_FAULTY
) {
5439 if (mark_failure(dev
, disk
, ord_to_idx(ord
)))
5440 super
->updates_pending
++;
5443 /* check if in_sync */
5444 if (state
& DS_INSYNC
&& ord
& IMSM_ORD_REBUILD
&& is_rebuilding(dev
)) {
5445 struct imsm_map
*migr_map
= get_imsm_map(dev
, 1);
5447 set_imsm_ord_tbl_ent(migr_map
, n
, ord_to_idx(ord
));
5448 super
->updates_pending
++;
5451 failed
= imsm_count_failed(super
, dev
);
5452 map_state
= imsm_check_degraded(super
, dev
, failed
);
5454 /* check if recovery complete, newly degraded, or failed */
5455 if (map_state
== IMSM_T_STATE_NORMAL
&& is_rebuilding(dev
)) {
5456 end_migration(dev
, map_state
);
5457 map
= get_imsm_map(dev
, 0);
5458 map
->failed_disk_num
= ~0;
5459 super
->updates_pending
++;
5460 a
->last_checkpoint
= 0;
5461 } else if (map_state
== IMSM_T_STATE_DEGRADED
&&
5462 map
->map_state
!= map_state
&&
5463 !dev
->vol
.migr_state
) {
5464 dprintf("imsm: mark degraded\n");
5465 map
->map_state
= map_state
;
5466 super
->updates_pending
++;
5467 a
->last_checkpoint
= 0;
5468 } else if (map_state
== IMSM_T_STATE_FAILED
&&
5469 map
->map_state
!= map_state
) {
5470 dprintf("imsm: mark failed\n");
5471 end_migration(dev
, map_state
);
5472 super
->updates_pending
++;
5473 a
->last_checkpoint
= 0;
5474 } else if (is_gen_migration(dev
)) {
5475 dprintf("imsm: Detected General Migration in state: ");
5476 if (map_state
== IMSM_T_STATE_NORMAL
) {
5477 end_migration(dev
, map_state
);
5478 map
= get_imsm_map(dev
, 0);
5479 map
->failed_disk_num
= ~0;
5480 dprintf("normal\n");
5482 if (map_state
== IMSM_T_STATE_DEGRADED
) {
5483 printf("degraded\n");
5484 end_migration(dev
, map_state
);
5486 dprintf("failed\n");
5488 map
->map_state
= map_state
;
5490 super
->updates_pending
++;
5494 static int store_imsm_mpb(int fd
, struct imsm_super
*mpb
)
5497 __u32 mpb_size
= __le32_to_cpu(mpb
->mpb_size
);
5498 unsigned long long dsize
;
5499 unsigned long long sectors
;
5501 get_dev_size(fd
, NULL
, &dsize
);
5503 if (mpb_size
> 512) {
5504 /* -1 to account for anchor */
5505 sectors
= mpb_sectors(mpb
) - 1;
5507 /* write the extended mpb to the sectors preceeding the anchor */
5508 if (lseek64(fd
, dsize
- (512 * (2 + sectors
)), SEEK_SET
) < 0)
5511 if ((unsigned long long)write(fd
, buf
+ 512, 512 * sectors
)
5516 /* first block is stored on second to last sector of the disk */
5517 if (lseek64(fd
, dsize
- (512 * 2), SEEK_SET
) < 0)
5520 if (write(fd
, buf
, 512) != 512)
5526 static void imsm_sync_metadata(struct supertype
*container
)
5528 struct intel_super
*super
= container
->sb
;
5530 dprintf("sync metadata: %d\n", super
->updates_pending
);
5531 if (!super
->updates_pending
)
5534 write_super_imsm(container
, 0);
5536 super
->updates_pending
= 0;
5539 static struct dl
*imsm_readd(struct intel_super
*super
, int idx
, struct active_array
*a
)
5541 struct imsm_dev
*dev
= get_imsm_dev(super
, a
->info
.container_member
);
5542 int i
= get_imsm_disk_idx(dev
, idx
, -1);
5545 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
5549 if (dl
&& is_failed(&dl
->disk
))
5553 dprintf("%s: found %x:%x\n", __func__
, dl
->major
, dl
->minor
);
5558 static struct dl
*imsm_add_spare(struct intel_super
*super
, int slot
,
5559 struct active_array
*a
, int activate_new
,
5560 struct mdinfo
*additional_test_list
)
5562 struct imsm_dev
*dev
= get_imsm_dev(super
, a
->info
.container_member
);
5563 int idx
= get_imsm_disk_idx(dev
, slot
, -1);
5564 struct imsm_super
*mpb
= super
->anchor
;
5565 struct imsm_map
*map
;
5566 unsigned long long pos
;
5571 __u32 array_start
= 0;
5572 __u32 array_end
= 0;
5574 struct mdinfo
*test_list
;
5576 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
5577 /* If in this array, skip */
5578 for (d
= a
->info
.devs
; d
; d
= d
->next
)
5579 if (d
->state_fd
>= 0 &&
5580 d
->disk
.major
== dl
->major
&&
5581 d
->disk
.minor
== dl
->minor
) {
5582 dprintf("%x:%x already in array\n",
5583 dl
->major
, dl
->minor
);
5588 test_list
= additional_test_list
;
5590 if (test_list
->disk
.major
== dl
->major
&&
5591 test_list
->disk
.minor
== dl
->minor
) {
5592 dprintf("%x:%x already in additional test list\n",
5593 dl
->major
, dl
->minor
);
5596 test_list
= test_list
->next
;
5601 /* skip in use or failed drives */
5602 if (is_failed(&dl
->disk
) || idx
== dl
->index
||
5604 dprintf("%x:%x status (failed: %d index: %d)\n",
5605 dl
->major
, dl
->minor
, is_failed(&dl
->disk
), idx
);
5609 /* skip pure spares when we are looking for partially
5610 * assimilated drives
5612 if (dl
->index
== -1 && !activate_new
)
5615 /* Does this unused device have the requisite free space?
5616 * It needs to be able to cover all member volumes
5618 ex
= get_extents(super
, dl
);
5620 dprintf("cannot get extents\n");
5623 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
5624 dev
= get_imsm_dev(super
, i
);
5625 map
= get_imsm_map(dev
, 0);
5627 /* check if this disk is already a member of
5630 if (get_imsm_disk_slot(map
, dl
->index
) >= 0)
5636 array_start
= __le32_to_cpu(map
->pba_of_lba0
);
5637 array_end
= array_start
+
5638 __le32_to_cpu(map
->blocks_per_member
) - 1;
5641 /* check that we can start at pba_of_lba0 with
5642 * blocks_per_member of space
5644 if (array_start
>= pos
&& array_end
< ex
[j
].start
) {
5648 pos
= ex
[j
].start
+ ex
[j
].size
;
5650 } while (ex
[j
-1].size
);
5657 if (i
< mpb
->num_raid_devs
) {
5658 dprintf("%x:%x does not have %u to %u available\n",
5659 dl
->major
, dl
->minor
, array_start
, array_end
);
5670 static int imsm_rebuild_allowed(struct supertype
*cont
, int dev_idx
, int failed
)
5672 struct imsm_dev
*dev2
;
5673 struct imsm_map
*map
;
5679 dev2
= get_imsm_dev(cont
->sb
, dev_idx
);
5681 state
= imsm_check_degraded(cont
->sb
, dev2
, failed
);
5682 if (state
== IMSM_T_STATE_FAILED
) {
5683 map
= get_imsm_map(dev2
, 0);
5686 for (slot
= 0; slot
< map
->num_members
; slot
++) {
5688 * Check if failed disks are deleted from intel
5689 * disk list or are marked to be deleted
5691 idx
= get_imsm_disk_idx(dev2
, slot
, -1);
5692 idisk
= get_imsm_dl_disk(cont
->sb
, idx
);
5694 * Do not rebuild the array if failed disks
5695 * from failed sub-array are not removed from
5699 is_failed(&idisk
->disk
) &&
5700 (idisk
->action
!= DISK_REMOVE
))
5708 static struct mdinfo
*imsm_activate_spare(struct active_array
*a
,
5709 struct metadata_update
**updates
)
5712 * Find a device with unused free space and use it to replace a
5713 * failed/vacant region in an array. We replace failed regions one a
5714 * array at a time. The result is that a new spare disk will be added
5715 * to the first failed array and after the monitor has finished
5716 * propagating failures the remainder will be consumed.
5718 * FIXME add a capability for mdmon to request spares from another
5722 struct intel_super
*super
= a
->container
->sb
;
5723 int inst
= a
->info
.container_member
;
5724 struct imsm_dev
*dev
= get_imsm_dev(super
, inst
);
5725 struct imsm_map
*map
= get_imsm_map(dev
, 0);
5726 int failed
= a
->info
.array
.raid_disks
;
5727 struct mdinfo
*rv
= NULL
;
5730 struct metadata_update
*mu
;
5732 struct imsm_update_activate_spare
*u
;
5737 for (d
= a
->info
.devs
; d
; d
= d
->next
) {
5738 if ((d
->curr_state
& DS_FAULTY
) &&
5740 /* wait for Removal to happen */
5742 if (d
->state_fd
>= 0)
5746 dprintf("imsm: activate spare: inst=%d failed=%d (%d) level=%d\n",
5747 inst
, failed
, a
->info
.array
.raid_disks
, a
->info
.array
.level
);
5749 if (dev
->vol
.migr_state
&&
5750 dev
->vol
.migr_type
== MIGR_GEN_MIGR
)
5751 /* No repair during migration */
5754 if (a
->info
.array
.level
== 4)
5755 /* No repair for takeovered array
5756 * imsm doesn't support raid4
5760 if (imsm_check_degraded(super
, dev
, failed
) != IMSM_T_STATE_DEGRADED
)
5764 * If there are any failed disks check state of the other volume.
5765 * Block rebuild if the another one is failed until failed disks
5766 * are removed from container.
5769 dprintf("found failed disks in %s, check if there another"
5770 "failed sub-array.\n",
5772 /* check if states of the other volumes allow for rebuild */
5773 for (i
= 0; i
< super
->anchor
->num_raid_devs
; i
++) {
5775 allowed
= imsm_rebuild_allowed(a
->container
,
5783 /* For each slot, if it is not working, find a spare */
5784 for (i
= 0; i
< a
->info
.array
.raid_disks
; i
++) {
5785 for (d
= a
->info
.devs
; d
; d
= d
->next
)
5786 if (d
->disk
.raid_disk
== i
)
5788 dprintf("found %d: %p %x\n", i
, d
, d
?d
->curr_state
:0);
5789 if (d
&& (d
->state_fd
>= 0))
5793 * OK, this device needs recovery. Try to re-add the
5794 * previous occupant of this slot, if this fails see if
5795 * we can continue the assimilation of a spare that was
5796 * partially assimilated, finally try to activate a new
5799 dl
= imsm_readd(super
, i
, a
);
5801 dl
= imsm_add_spare(super
, i
, a
, 0, NULL
);
5803 dl
= imsm_add_spare(super
, i
, a
, 1, NULL
);
5807 /* found a usable disk with enough space */
5808 di
= malloc(sizeof(*di
));
5811 memset(di
, 0, sizeof(*di
));
5813 /* dl->index will be -1 in the case we are activating a
5814 * pristine spare. imsm_process_update() will create a
5815 * new index in this case. Once a disk is found to be
5816 * failed in all member arrays it is kicked from the
5819 di
->disk
.number
= dl
->index
;
5821 /* (ab)use di->devs to store a pointer to the device
5824 di
->devs
= (struct mdinfo
*) dl
;
5826 di
->disk
.raid_disk
= i
;
5827 di
->disk
.major
= dl
->major
;
5828 di
->disk
.minor
= dl
->minor
;
5830 di
->recovery_start
= 0;
5831 di
->data_offset
= __le32_to_cpu(map
->pba_of_lba0
);
5832 di
->component_size
= a
->info
.component_size
;
5833 di
->container_member
= inst
;
5834 super
->random
= random32();
5838 dprintf("%x:%x to be %d at %llu\n", dl
->major
, dl
->minor
,
5839 i
, di
->data_offset
);
5845 /* No spares found */
5847 /* Now 'rv' has a list of devices to return.
5848 * Create a metadata_update record to update the
5849 * disk_ord_tbl for the array
5851 mu
= malloc(sizeof(*mu
));
5853 mu
->buf
= malloc(sizeof(struct imsm_update_activate_spare
) * num_spares
);
5854 if (mu
->buf
== NULL
) {
5861 struct mdinfo
*n
= rv
->next
;
5870 mu
->space_list
= NULL
;
5871 mu
->len
= sizeof(struct imsm_update_activate_spare
) * num_spares
;
5872 mu
->next
= *updates
;
5873 u
= (struct imsm_update_activate_spare
*) mu
->buf
;
5875 for (di
= rv
; di
; di
= di
->next
) {
5876 u
->type
= update_activate_spare
;
5877 u
->dl
= (struct dl
*) di
->devs
;
5879 u
->slot
= di
->disk
.raid_disk
;
5890 static int disks_overlap(struct intel_super
*super
, int idx
, struct imsm_update_create_array
*u
)
5892 struct imsm_dev
*dev
= get_imsm_dev(super
, idx
);
5893 struct imsm_map
*map
= get_imsm_map(dev
, 0);
5894 struct imsm_map
*new_map
= get_imsm_map(&u
->dev
, 0);
5895 struct disk_info
*inf
= get_disk_info(u
);
5896 struct imsm_disk
*disk
;
5900 for (i
= 0; i
< map
->num_members
; i
++) {
5901 disk
= get_imsm_disk(super
, get_imsm_disk_idx(dev
, i
, -1));
5902 for (j
= 0; j
< new_map
->num_members
; j
++)
5903 if (serialcmp(disk
->serial
, inf
[j
].serial
) == 0)
5911 static struct dl
*get_disk_super(struct intel_super
*super
, int major
, int minor
)
5913 struct dl
*dl
= NULL
;
5914 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
5915 if ((dl
->major
== major
) && (dl
->minor
== minor
))
5920 static int remove_disk_super(struct intel_super
*super
, int major
, int minor
)
5922 struct dl
*prev
= NULL
;
5926 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
5927 if ((dl
->major
== major
) && (dl
->minor
== minor
)) {
5930 prev
->next
= dl
->next
;
5932 super
->disks
= dl
->next
;
5934 __free_imsm_disk(dl
);
5935 dprintf("%s: removed %x:%x\n",
5936 __func__
, major
, minor
);
5944 static void imsm_delete(struct intel_super
*super
, struct dl
**dlp
, unsigned index
);
5946 static int add_remove_disk_update(struct intel_super
*super
)
5948 int check_degraded
= 0;
5949 struct dl
*disk
= NULL
;
5950 /* add/remove some spares to/from the metadata/contrainer */
5951 while (super
->disk_mgmt_list
) {
5952 struct dl
*disk_cfg
;
5954 disk_cfg
= super
->disk_mgmt_list
;
5955 super
->disk_mgmt_list
= disk_cfg
->next
;
5956 disk_cfg
->next
= NULL
;
5958 if (disk_cfg
->action
== DISK_ADD
) {
5959 disk_cfg
->next
= super
->disks
;
5960 super
->disks
= disk_cfg
;
5962 dprintf("%s: added %x:%x\n",
5963 __func__
, disk_cfg
->major
,
5965 } else if (disk_cfg
->action
== DISK_REMOVE
) {
5966 dprintf("Disk remove action processed: %x.%x\n",
5967 disk_cfg
->major
, disk_cfg
->minor
);
5968 disk
= get_disk_super(super
,
5972 /* store action status */
5973 disk
->action
= DISK_REMOVE
;
5974 /* remove spare disks only */
5975 if (disk
->index
== -1) {
5976 remove_disk_super(super
,
5981 /* release allocate disk structure */
5982 __free_imsm_disk(disk_cfg
);
5985 return check_degraded
;
5988 static int apply_reshape_container_disks_update(struct imsm_update_reshape
*u
,
5989 struct intel_super
*super
,
5992 struct dl
*new_disk
;
5993 struct intel_dev
*id
;
5995 int delta_disks
= u
->new_raid_disks
- u
->old_raid_disks
;
5996 int disk_count
= u
->old_raid_disks
;
5997 void **tofree
= NULL
;
5998 int devices_to_reshape
= 1;
5999 struct imsm_super
*mpb
= super
->anchor
;
6001 unsigned int dev_id
;
6003 dprintf("imsm: apply_reshape_container_disks_update()\n");
6005 /* enable spares to use in array */
6006 for (i
= 0; i
< delta_disks
; i
++) {
6007 new_disk
= get_disk_super(super
,
6008 major(u
->new_disks
[i
]),
6009 minor(u
->new_disks
[i
]));
6010 dprintf("imsm: new disk for reshape is: %i:%i "
6011 "(%p, index = %i)\n",
6012 major(u
->new_disks
[i
]), minor(u
->new_disks
[i
]),
6013 new_disk
, new_disk
->index
);
6014 if ((new_disk
== NULL
) ||
6015 ((new_disk
->index
>= 0) &&
6016 (new_disk
->index
< u
->old_raid_disks
)))
6017 goto update_reshape_exit
;
6018 new_disk
->index
= disk_count
++;
6019 /* slot to fill in autolayout
6021 new_disk
->raiddisk
= new_disk
->index
;
6022 new_disk
->disk
.status
|=
6024 new_disk
->disk
.status
&= ~SPARE_DISK
;
6027 dprintf("imsm: volume set mpb->num_raid_devs = %i\n",
6028 mpb
->num_raid_devs
);
6029 /* manage changes in volume
6031 for (dev_id
= 0; dev_id
< mpb
->num_raid_devs
; dev_id
++) {
6032 void **sp
= *space_list
;
6033 struct imsm_dev
*newdev
;
6034 struct imsm_map
*newmap
, *oldmap
;
6036 for (id
= super
->devlist
; id
; id
= id
->next
) {
6037 if (id
->index
== dev_id
)
6046 /* Copy the dev, but not (all of) the map */
6047 memcpy(newdev
, id
->dev
, sizeof(*newdev
));
6048 oldmap
= get_imsm_map(id
->dev
, 0);
6049 newmap
= get_imsm_map(newdev
, 0);
6050 /* Copy the current map */
6051 memcpy(newmap
, oldmap
, sizeof_imsm_map(oldmap
));
6052 /* update one device only
6054 if (devices_to_reshape
) {
6055 dprintf("imsm: modifying subdev: %i\n",
6057 devices_to_reshape
--;
6058 newdev
->vol
.migr_state
= 1;
6059 newdev
->vol
.curr_migr_unit
= 0;
6060 newdev
->vol
.migr_type
= MIGR_GEN_MIGR
;
6061 newmap
->num_members
= u
->new_raid_disks
;
6062 for (i
= 0; i
< delta_disks
; i
++) {
6063 set_imsm_ord_tbl_ent(newmap
,
6064 u
->old_raid_disks
+ i
,
6065 u
->old_raid_disks
+ i
);
6067 /* New map is correct, now need to save old map
6069 newmap
= get_imsm_map(newdev
, 1);
6070 memcpy(newmap
, oldmap
, sizeof_imsm_map(oldmap
));
6072 imsm_set_array_size(newdev
);
6075 sp
= (void **)id
->dev
;
6081 *space_list
= tofree
;
6084 update_reshape_exit
:
6089 static int apply_takeover_update(struct imsm_update_takeover
*u
,
6090 struct intel_super
*super
,
6093 struct imsm_dev
*dev
= NULL
;
6094 struct intel_dev
*dv
;
6095 struct imsm_dev
*dev_new
;
6096 struct imsm_map
*map
;
6100 for (dv
= super
->devlist
; dv
; dv
= dv
->next
)
6101 if (dv
->index
== (unsigned int)u
->subarray
) {
6109 map
= get_imsm_map(dev
, 0);
6111 if (u
->direction
== R10_TO_R0
) {
6112 /* Number of failed disks must be half of initial disk number */
6113 if (imsm_count_failed(super
, dev
) != (map
->num_members
/ 2))
6116 /* iterate through devices to mark removed disks as spare */
6117 for (dm
= super
->disks
; dm
; dm
= dm
->next
) {
6118 if (dm
->disk
.status
& FAILED_DISK
) {
6119 int idx
= dm
->index
;
6120 /* update indexes on the disk list */
6121 /* FIXME this loop-with-the-loop looks wrong, I'm not convinced
6122 the index values will end up being correct.... NB */
6123 for (du
= super
->disks
; du
; du
= du
->next
)
6124 if (du
->index
> idx
)
6126 /* mark as spare disk */
6127 dm
->disk
.status
= SPARE_DISK
;
6132 map
->num_members
= map
->num_members
/ 2;
6133 map
->map_state
= IMSM_T_STATE_NORMAL
;
6134 map
->num_domains
= 1;
6135 map
->raid_level
= 0;
6136 map
->failed_disk_num
= -1;
6139 if (u
->direction
== R0_TO_R10
) {
6141 /* update slots in current disk list */
6142 for (dm
= super
->disks
; dm
; dm
= dm
->next
) {
6146 /* create new *missing* disks */
6147 for (i
= 0; i
< map
->num_members
; i
++) {
6148 space
= *space_list
;
6151 *space_list
= *space
;
6153 memcpy(du
, super
->disks
, sizeof(*du
));
6157 du
->index
= (i
* 2) + 1;
6158 sprintf((char *)du
->disk
.serial
,
6159 " MISSING_%d", du
->index
);
6160 sprintf((char *)du
->serial
,
6161 "MISSING_%d", du
->index
);
6162 du
->next
= super
->missing
;
6163 super
->missing
= du
;
6165 /* create new dev and map */
6166 space
= *space_list
;
6169 *space_list
= *space
;
6170 dev_new
= (void *)space
;
6171 memcpy(dev_new
, dev
, sizeof(*dev
));
6172 /* update new map */
6173 map
= get_imsm_map(dev_new
, 0);
6174 map
->num_members
= map
->num_members
* 2;
6175 map
->map_state
= IMSM_T_STATE_DEGRADED
;
6176 map
->num_domains
= 2;
6177 map
->raid_level
= 1;
6178 /* replace dev<->dev_new */
6181 /* update disk order table */
6182 for (du
= super
->disks
; du
; du
= du
->next
)
6184 set_imsm_ord_tbl_ent(map
, du
->index
, du
->index
);
6185 for (du
= super
->missing
; du
; du
= du
->next
)
6186 if (du
->index
>= 0) {
6187 set_imsm_ord_tbl_ent(map
, du
->index
, du
->index
);
6188 mark_missing(dev_new
, &du
->disk
, du
->index
);
6194 static void imsm_process_update(struct supertype
*st
,
6195 struct metadata_update
*update
)
6198 * crack open the metadata_update envelope to find the update record
6199 * update can be one of:
6200 * update_reshape_container_disks - all the arrays in the container
6201 * are being reshaped to have more devices. We need to mark
6202 * the arrays for general migration and convert selected spares
6203 * into active devices.
6204 * update_activate_spare - a spare device has replaced a failed
6205 * device in an array, update the disk_ord_tbl. If this disk is
6206 * present in all member arrays then also clear the SPARE_DISK
6208 * update_create_array
6210 * update_rename_array
6211 * update_add_remove_disk
6213 struct intel_super
*super
= st
->sb
;
6214 struct imsm_super
*mpb
;
6215 enum imsm_update_type type
= *(enum imsm_update_type
*) update
->buf
;
6217 /* update requires a larger buf but the allocation failed */
6218 if (super
->next_len
&& !super
->next_buf
) {
6219 super
->next_len
= 0;
6223 if (super
->next_buf
) {
6224 memcpy(super
->next_buf
, super
->buf
, super
->len
);
6226 super
->len
= super
->next_len
;
6227 super
->buf
= super
->next_buf
;
6229 super
->next_len
= 0;
6230 super
->next_buf
= NULL
;
6233 mpb
= super
->anchor
;
6236 case update_takeover
: {
6237 struct imsm_update_takeover
*u
= (void *)update
->buf
;
6238 if (apply_takeover_update(u
, super
, &update
->space_list
)) {
6239 imsm_update_version_info(super
);
6240 super
->updates_pending
++;
6245 case update_reshape_container_disks
: {
6246 struct imsm_update_reshape
*u
= (void *)update
->buf
;
6247 if (apply_reshape_container_disks_update(
6248 u
, super
, &update
->space_list
))
6249 super
->updates_pending
++;
6252 case update_activate_spare
: {
6253 struct imsm_update_activate_spare
*u
= (void *) update
->buf
;
6254 struct imsm_dev
*dev
= get_imsm_dev(super
, u
->array
);
6255 struct imsm_map
*map
= get_imsm_map(dev
, 0);
6256 struct imsm_map
*migr_map
;
6257 struct active_array
*a
;
6258 struct imsm_disk
*disk
;
6263 int victim
= get_imsm_disk_idx(dev
, u
->slot
, -1);
6266 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
6271 fprintf(stderr
, "error: imsm_activate_spare passed "
6272 "an unknown disk (index: %d)\n",
6277 super
->updates_pending
++;
6279 /* count failures (excluding rebuilds and the victim)
6280 * to determine map[0] state
6283 for (i
= 0; i
< map
->num_members
; i
++) {
6286 disk
= get_imsm_disk(super
,
6287 get_imsm_disk_idx(dev
, i
, -1));
6288 if (!disk
|| is_failed(disk
))
6292 /* adding a pristine spare, assign a new index */
6293 if (dl
->index
< 0) {
6294 dl
->index
= super
->anchor
->num_disks
;
6295 super
->anchor
->num_disks
++;
6298 disk
->status
|= CONFIGURED_DISK
;
6299 disk
->status
&= ~SPARE_DISK
;
6302 to_state
= imsm_check_degraded(super
, dev
, failed
);
6303 map
->map_state
= IMSM_T_STATE_DEGRADED
;
6304 migrate(dev
, to_state
, MIGR_REBUILD
);
6305 migr_map
= get_imsm_map(dev
, 1);
6306 set_imsm_ord_tbl_ent(map
, u
->slot
, dl
->index
);
6307 set_imsm_ord_tbl_ent(migr_map
, u
->slot
, dl
->index
| IMSM_ORD_REBUILD
);
6309 /* update the family_num to mark a new container
6310 * generation, being careful to record the existing
6311 * family_num in orig_family_num to clean up after
6312 * earlier mdadm versions that neglected to set it.
6314 if (mpb
->orig_family_num
== 0)
6315 mpb
->orig_family_num
= mpb
->family_num
;
6316 mpb
->family_num
+= super
->random
;
6318 /* count arrays using the victim in the metadata */
6320 for (a
= st
->arrays
; a
; a
= a
->next
) {
6321 dev
= get_imsm_dev(super
, a
->info
.container_member
);
6322 map
= get_imsm_map(dev
, 0);
6324 if (get_imsm_disk_slot(map
, victim
) >= 0)
6328 /* delete the victim if it is no longer being
6334 /* We know that 'manager' isn't touching anything,
6335 * so it is safe to delete
6337 for (dlp
= &super
->disks
; *dlp
; dlp
= &(*dlp
)->next
)
6338 if ((*dlp
)->index
== victim
)
6341 /* victim may be on the missing list */
6343 for (dlp
= &super
->missing
; *dlp
; dlp
= &(*dlp
)->next
)
6344 if ((*dlp
)->index
== victim
)
6346 imsm_delete(super
, dlp
, victim
);
6350 case update_create_array
: {
6351 /* someone wants to create a new array, we need to be aware of
6352 * a few races/collisions:
6353 * 1/ 'Create' called by two separate instances of mdadm
6354 * 2/ 'Create' versus 'activate_spare': mdadm has chosen
6355 * devices that have since been assimilated via
6357 * In the event this update can not be carried out mdadm will
6358 * (FIX ME) notice that its update did not take hold.
6360 struct imsm_update_create_array
*u
= (void *) update
->buf
;
6361 struct intel_dev
*dv
;
6362 struct imsm_dev
*dev
;
6363 struct imsm_map
*map
, *new_map
;
6364 unsigned long long start
, end
;
6365 unsigned long long new_start
, new_end
;
6367 struct disk_info
*inf
;
6370 /* handle racing creates: first come first serve */
6371 if (u
->dev_idx
< mpb
->num_raid_devs
) {
6372 dprintf("%s: subarray %d already defined\n",
6373 __func__
, u
->dev_idx
);
6377 /* check update is next in sequence */
6378 if (u
->dev_idx
!= mpb
->num_raid_devs
) {
6379 dprintf("%s: can not create array %d expected index %d\n",
6380 __func__
, u
->dev_idx
, mpb
->num_raid_devs
);
6384 new_map
= get_imsm_map(&u
->dev
, 0);
6385 new_start
= __le32_to_cpu(new_map
->pba_of_lba0
);
6386 new_end
= new_start
+ __le32_to_cpu(new_map
->blocks_per_member
);
6387 inf
= get_disk_info(u
);
6389 /* handle activate_spare versus create race:
6390 * check to make sure that overlapping arrays do not include
6393 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
6394 dev
= get_imsm_dev(super
, i
);
6395 map
= get_imsm_map(dev
, 0);
6396 start
= __le32_to_cpu(map
->pba_of_lba0
);
6397 end
= start
+ __le32_to_cpu(map
->blocks_per_member
);
6398 if ((new_start
>= start
&& new_start
<= end
) ||
6399 (start
>= new_start
&& start
<= new_end
))
6404 if (disks_overlap(super
, i
, u
)) {
6405 dprintf("%s: arrays overlap\n", __func__
);
6410 /* check that prepare update was successful */
6411 if (!update
->space
) {
6412 dprintf("%s: prepare update failed\n", __func__
);
6416 /* check that all disks are still active before committing
6417 * changes. FIXME: could we instead handle this by creating a
6418 * degraded array? That's probably not what the user expects,
6419 * so better to drop this update on the floor.
6421 for (i
= 0; i
< new_map
->num_members
; i
++) {
6422 dl
= serial_to_dl(inf
[i
].serial
, super
);
6424 dprintf("%s: disk disappeared\n", __func__
);
6429 super
->updates_pending
++;
6431 /* convert spares to members and fixup ord_tbl */
6432 for (i
= 0; i
< new_map
->num_members
; i
++) {
6433 dl
= serial_to_dl(inf
[i
].serial
, super
);
6434 if (dl
->index
== -1) {
6435 dl
->index
= mpb
->num_disks
;
6437 dl
->disk
.status
|= CONFIGURED_DISK
;
6438 dl
->disk
.status
&= ~SPARE_DISK
;
6440 set_imsm_ord_tbl_ent(new_map
, i
, dl
->index
);
6445 update
->space
= NULL
;
6446 imsm_copy_dev(dev
, &u
->dev
);
6447 dv
->index
= u
->dev_idx
;
6448 dv
->next
= super
->devlist
;
6449 super
->devlist
= dv
;
6450 mpb
->num_raid_devs
++;
6452 imsm_update_version_info(super
);
6455 /* mdmon knows how to release update->space, but not
6456 * ((struct intel_dev *) update->space)->dev
6458 if (update
->space
) {
6464 case update_kill_array
: {
6465 struct imsm_update_kill_array
*u
= (void *) update
->buf
;
6466 int victim
= u
->dev_idx
;
6467 struct active_array
*a
;
6468 struct intel_dev
**dp
;
6469 struct imsm_dev
*dev
;
6471 /* sanity check that we are not affecting the uuid of
6472 * active arrays, or deleting an active array
6474 * FIXME when immutable ids are available, but note that
6475 * we'll also need to fixup the invalidated/active
6476 * subarray indexes in mdstat
6478 for (a
= st
->arrays
; a
; a
= a
->next
)
6479 if (a
->info
.container_member
>= victim
)
6481 /* by definition if mdmon is running at least one array
6482 * is active in the container, so checking
6483 * mpb->num_raid_devs is just extra paranoia
6485 dev
= get_imsm_dev(super
, victim
);
6486 if (a
|| !dev
|| mpb
->num_raid_devs
== 1) {
6487 dprintf("failed to delete subarray-%d\n", victim
);
6491 for (dp
= &super
->devlist
; *dp
;)
6492 if ((*dp
)->index
== (unsigned)super
->current_vol
) {
6495 if ((*dp
)->index
> (unsigned)victim
)
6499 mpb
->num_raid_devs
--;
6500 super
->updates_pending
++;
6503 case update_rename_array
: {
6504 struct imsm_update_rename_array
*u
= (void *) update
->buf
;
6505 char name
[MAX_RAID_SERIAL_LEN
+1];
6506 int target
= u
->dev_idx
;
6507 struct active_array
*a
;
6508 struct imsm_dev
*dev
;
6510 /* sanity check that we are not affecting the uuid of
6513 snprintf(name
, MAX_RAID_SERIAL_LEN
, "%s", (char *) u
->name
);
6514 name
[MAX_RAID_SERIAL_LEN
] = '\0';
6515 for (a
= st
->arrays
; a
; a
= a
->next
)
6516 if (a
->info
.container_member
== target
)
6518 dev
= get_imsm_dev(super
, u
->dev_idx
);
6519 if (a
|| !dev
|| !check_name(super
, name
, 1)) {
6520 dprintf("failed to rename subarray-%d\n", target
);
6524 snprintf((char *) dev
->volume
, MAX_RAID_SERIAL_LEN
, "%s", name
);
6525 super
->updates_pending
++;
6528 case update_add_remove_disk
: {
6529 /* we may be able to repair some arrays if disks are
6530 * being added, check teh status of add_remove_disk
6531 * if discs has been added.
6533 if (add_remove_disk_update(super
)) {
6534 struct active_array
*a
;
6536 super
->updates_pending
++;
6537 for (a
= st
->arrays
; a
; a
= a
->next
)
6538 a
->check_degraded
= 1;
6543 fprintf(stderr
, "error: unsuported process update type:"
6544 "(type: %d)\n", type
);
6548 static void imsm_prepare_update(struct supertype
*st
,
6549 struct metadata_update
*update
)
6552 * Allocate space to hold new disk entries, raid-device entries or a new
6553 * mpb if necessary. The manager synchronously waits for updates to
6554 * complete in the monitor, so new mpb buffers allocated here can be
6555 * integrated by the monitor thread without worrying about live pointers
6556 * in the manager thread.
6558 enum imsm_update_type type
= *(enum imsm_update_type
*) update
->buf
;
6559 struct intel_super
*super
= st
->sb
;
6560 struct imsm_super
*mpb
= super
->anchor
;
6565 case update_takeover
: {
6566 struct imsm_update_takeover
*u
= (void *)update
->buf
;
6567 if (u
->direction
== R0_TO_R10
) {
6568 void **tail
= (void **)&update
->space_list
;
6569 struct imsm_dev
*dev
= get_imsm_dev(super
, u
->subarray
);
6570 struct imsm_map
*map
= get_imsm_map(dev
, 0);
6571 int num_members
= map
->num_members
;
6575 /* allocate memory for added disks */
6576 for (i
= 0; i
< num_members
; i
++) {
6577 size
= sizeof(struct dl
);
6578 space
= malloc(size
);
6587 /* allocate memory for new device */
6588 size
= sizeof_imsm_dev(super
->devlist
->dev
, 0) +
6589 (num_members
* sizeof(__u32
));
6590 space
= malloc(size
);
6599 len
= disks_to_mpb_size(num_members
* 2);
6601 /* if allocation didn't success, free buffer */
6602 while (update
->space_list
) {
6603 void **sp
= update
->space_list
;
6604 update
->space_list
= *sp
;
6612 case update_reshape_container_disks
: {
6613 /* Every raid device in the container is about to
6614 * gain some more devices, and we will enter a
6616 * So each 'imsm_map' will be bigger, and the imsm_vol
6617 * will now hold 2 of them.
6618 * Thus we need new 'struct imsm_dev' allocations sized
6619 * as sizeof_imsm_dev but with more devices in both maps.
6621 struct imsm_update_reshape
*u
= (void *)update
->buf
;
6622 struct intel_dev
*dl
;
6623 void **space_tail
= (void**)&update
->space_list
;
6625 dprintf("imsm: imsm_prepare_update() for update_reshape\n");
6627 for (dl
= super
->devlist
; dl
; dl
= dl
->next
) {
6628 int size
= sizeof_imsm_dev(dl
->dev
, 1);
6630 if (u
->new_raid_disks
> u
->old_raid_disks
)
6631 size
+= sizeof(__u32
)*2*
6632 (u
->new_raid_disks
- u
->old_raid_disks
);
6641 len
= disks_to_mpb_size(u
->new_raid_disks
);
6642 dprintf("New anchor length is %llu\n", (unsigned long long)len
);
6645 case update_create_array
: {
6646 struct imsm_update_create_array
*u
= (void *) update
->buf
;
6647 struct intel_dev
*dv
;
6648 struct imsm_dev
*dev
= &u
->dev
;
6649 struct imsm_map
*map
= get_imsm_map(dev
, 0);
6651 struct disk_info
*inf
;
6655 inf
= get_disk_info(u
);
6656 len
= sizeof_imsm_dev(dev
, 1);
6657 /* allocate a new super->devlist entry */
6658 dv
= malloc(sizeof(*dv
));
6660 dv
->dev
= malloc(len
);
6665 update
->space
= NULL
;
6669 /* count how many spares will be converted to members */
6670 for (i
= 0; i
< map
->num_members
; i
++) {
6671 dl
= serial_to_dl(inf
[i
].serial
, super
);
6673 /* hmm maybe it failed?, nothing we can do about
6678 if (count_memberships(dl
, super
) == 0)
6681 len
+= activate
* sizeof(struct imsm_disk
);
6688 /* check if we need a larger metadata buffer */
6689 if (super
->next_buf
)
6690 buf_len
= super
->next_len
;
6692 buf_len
= super
->len
;
6694 if (__le32_to_cpu(mpb
->mpb_size
) + len
> buf_len
) {
6695 /* ok we need a larger buf than what is currently allocated
6696 * if this allocation fails process_update will notice that
6697 * ->next_len is set and ->next_buf is NULL
6699 buf_len
= ROUND_UP(__le32_to_cpu(mpb
->mpb_size
) + len
, 512);
6700 if (super
->next_buf
)
6701 free(super
->next_buf
);
6703 super
->next_len
= buf_len
;
6704 if (posix_memalign(&super
->next_buf
, 512, buf_len
) == 0)
6705 memset(super
->next_buf
, 0, buf_len
);
6707 super
->next_buf
= NULL
;
6711 /* must be called while manager is quiesced */
6712 static void imsm_delete(struct intel_super
*super
, struct dl
**dlp
, unsigned index
)
6714 struct imsm_super
*mpb
= super
->anchor
;
6716 struct imsm_dev
*dev
;
6717 struct imsm_map
*map
;
6718 int i
, j
, num_members
;
6721 dprintf("%s: deleting device[%d] from imsm_super\n",
6724 /* shift all indexes down one */
6725 for (iter
= super
->disks
; iter
; iter
= iter
->next
)
6726 if (iter
->index
> (int)index
)
6728 for (iter
= super
->missing
; iter
; iter
= iter
->next
)
6729 if (iter
->index
> (int)index
)
6732 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
6733 dev
= get_imsm_dev(super
, i
);
6734 map
= get_imsm_map(dev
, 0);
6735 num_members
= map
->num_members
;
6736 for (j
= 0; j
< num_members
; j
++) {
6737 /* update ord entries being careful not to propagate
6738 * ord-flags to the first map
6740 ord
= get_imsm_ord_tbl_ent(dev
, j
, -1);
6742 if (ord_to_idx(ord
) <= index
)
6745 map
= get_imsm_map(dev
, 0);
6746 set_imsm_ord_tbl_ent(map
, j
, ord_to_idx(ord
- 1));
6747 map
= get_imsm_map(dev
, 1);
6749 set_imsm_ord_tbl_ent(map
, j
, ord
- 1);
6754 super
->updates_pending
++;
6756 struct dl
*dl
= *dlp
;
6758 *dlp
= (*dlp
)->next
;
6759 __free_imsm_disk(dl
);
6763 static char disk_by_path
[] = "/dev/disk/by-path/";
6765 static const char *imsm_get_disk_controller_domain(const char *path
)
6767 char disk_path
[PATH_MAX
];
6771 strncpy(disk_path
, disk_by_path
, PATH_MAX
- 1);
6772 strncat(disk_path
, path
, PATH_MAX
- strlen(disk_path
) - 1);
6773 if (stat(disk_path
, &st
) == 0) {
6774 struct sys_dev
* hba
;
6777 path
= devt_to_devpath(st
.st_rdev
);
6780 hba
= find_disk_attached_hba(-1, path
);
6781 if (hba
&& hba
->type
== SYS_DEV_SAS
)
6783 else if (hba
&& hba
->type
== SYS_DEV_SATA
)
6787 dprintf("path: %s hba: %s attached: %s\n",
6788 path
, (hba
) ? hba
->path
: "NULL", drv
);
6796 static int imsm_find_array_minor_by_subdev(int subdev
, int container
, int *minor
)
6798 char subdev_name
[20];
6799 struct mdstat_ent
*mdstat
;
6801 sprintf(subdev_name
, "%d", subdev
);
6802 mdstat
= mdstat_by_subdev(subdev_name
, container
);
6806 *minor
= mdstat
->devnum
;
6807 free_mdstat(mdstat
);
6811 static int imsm_reshape_is_allowed_on_container(struct supertype
*st
,
6812 struct geo_params
*geo
,
6813 int *old_raid_disks
)
6815 /* currently we only support increasing the number of devices
6816 * for a container. This increases the number of device for each
6817 * member array. They must all be RAID0 or RAID5.
6820 struct mdinfo
*info
, *member
;
6821 int devices_that_can_grow
= 0;
6823 dprintf("imsm: imsm_reshape_is_allowed_on_container(ENTER): "
6824 "st->devnum = (%i)\n",
6827 if (geo
->size
!= -1 ||
6828 geo
->level
!= UnSet
||
6829 geo
->layout
!= UnSet
||
6830 geo
->chunksize
!= 0 ||
6831 geo
->raid_disks
== UnSet
) {
6832 dprintf("imsm: Container operation is allowed for "
6833 "raid disks number change only.\n");
6837 info
= container_content_imsm(st
, NULL
);
6838 for (member
= info
; member
; member
= member
->next
) {
6842 dprintf("imsm: checking device_num: %i\n",
6843 member
->container_member
);
6845 if (geo
->raid_disks
<= member
->array
.raid_disks
) {
6846 /* we work on container for Online Capacity Expansion
6847 * only so raid_disks has to grow
6849 dprintf("imsm: for container operation raid disks "
6850 "increase is required\n");
6854 if ((info
->array
.level
!= 0) &&
6855 (info
->array
.level
!= 5)) {
6856 /* we cannot use this container with other raid level
6858 dprintf("imsm: for container operation wrong"
6859 " raid level (%i) detected\n",
6863 /* check for platform support
6864 * for this raid level configuration
6866 struct intel_super
*super
= st
->sb
;
6867 if (!is_raid_level_supported(super
->orom
,
6868 member
->array
.level
,
6870 dprintf("platform does not support raid%d with"
6874 geo
->raid_disks
> 1 ? "s" : "");
6879 if (*old_raid_disks
&&
6880 info
->array
.raid_disks
!= *old_raid_disks
)
6882 *old_raid_disks
= info
->array
.raid_disks
;
6884 /* All raid5 and raid0 volumes in container
6885 * have to be ready for Online Capacity Expansion
6886 * so they need to be assembled. We have already
6887 * checked that no recovery etc is happening.
6889 result
= imsm_find_array_minor_by_subdev(member
->container_member
,
6893 dprintf("imsm: cannot find array\n");
6896 devices_that_can_grow
++;
6899 if (!member
&& devices_that_can_grow
)
6903 dprintf("\tContainer operation allowed\n");
6905 dprintf("\tError: %i\n", ret_val
);
6910 /* Function: get_spares_for_grow
6911 * Description: Allocates memory and creates list of spare devices
6912 * avaliable in container. Checks if spare drive size is acceptable.
6913 * Parameters: Pointer to the supertype structure
6914 * Returns: Pointer to the list of spare devices (mdinfo structure) on success,
6917 static struct mdinfo
*get_spares_for_grow(struct supertype
*st
)
6919 unsigned long long min_size
= min_acceptable_spare_size_imsm(st
);
6920 return container_choose_spares(st
, min_size
, NULL
, NULL
, NULL
, 0);
6923 /******************************************************************************
6924 * function: imsm_create_metadata_update_for_reshape
6925 * Function creates update for whole IMSM container.
6927 ******************************************************************************/
6928 static int imsm_create_metadata_update_for_reshape(
6929 struct supertype
*st
,
6930 struct geo_params
*geo
,
6932 struct imsm_update_reshape
**updatep
)
6934 struct intel_super
*super
= st
->sb
;
6935 struct imsm_super
*mpb
= super
->anchor
;
6936 int update_memory_size
= 0;
6937 struct imsm_update_reshape
*u
= NULL
;
6938 struct mdinfo
*spares
= NULL
;
6940 int delta_disks
= 0;
6943 dprintf("imsm_update_metadata_for_reshape(enter) raid_disks = %i\n",
6946 delta_disks
= geo
->raid_disks
- old_raid_disks
;
6948 /* size of all update data without anchor */
6949 update_memory_size
= sizeof(struct imsm_update_reshape
);
6951 /* now add space for spare disks that we need to add. */
6952 update_memory_size
+= sizeof(u
->new_disks
[0]) * (delta_disks
- 1);
6954 u
= calloc(1, update_memory_size
);
6957 "cannot get memory for imsm_update_reshape update\n");
6960 u
->type
= update_reshape_container_disks
;
6961 u
->old_raid_disks
= old_raid_disks
;
6962 u
->new_raid_disks
= geo
->raid_disks
;
6964 /* now get spare disks list
6966 spares
= get_spares_for_grow(st
);
6969 || delta_disks
> spares
->array
.spare_disks
) {
6970 fprintf(stderr
, Name
": imsm: ERROR: Cannot get spare devices "
6971 "for %s.\n", geo
->dev_name
);
6975 /* we have got spares
6976 * update disk list in imsm_disk list table in anchor
6978 dprintf("imsm: %i spares are available.\n\n",
6979 spares
->array
.spare_disks
);
6982 for (i
= 0; i
< delta_disks
; i
++) {
6987 u
->new_disks
[i
] = makedev(dev
->disk
.major
,
6989 dl
= get_disk_super(super
, dev
->disk
.major
, dev
->disk
.minor
);
6990 dl
->index
= mpb
->num_disks
;
7000 dprintf("imsm: reshape update preparation :");
7001 if (i
== delta_disks
) {
7004 return update_memory_size
;
7007 dprintf(" Error\n");
7012 static void imsm_update_metadata_locally(struct supertype
*st
,
7015 struct metadata_update mu
;
7020 mu
.space_list
= NULL
;
7022 imsm_prepare_update(st
, &mu
);
7023 imsm_process_update(st
, &mu
);
7025 while (mu
.space_list
) {
7026 void **space
= mu
.space_list
;
7027 mu
.space_list
= *space
;
7032 /***************************************************************************
7033 * Function: imsm_analyze_change
7034 * Description: Function analyze change for single volume
7035 * and validate if transition is supported
7036 * Parameters: Geometry parameters, supertype structure
7037 * Returns: Operation type code on success, -1 if fail
7038 ****************************************************************************/
7039 enum imsm_reshape_type
imsm_analyze_change(struct supertype
*st
,
7040 struct geo_params
*geo
)
7047 getinfo_super_imsm_volume(st
, &info
, NULL
);
7049 if ((geo
->level
!= info
.array
.level
) &&
7050 (geo
->level
>= 0) &&
7051 (geo
->level
!= UnSet
)) {
7052 switch (info
.array
.level
) {
7054 if (geo
->level
== 5) {
7055 change
= CH_MIGRATION
;
7058 if (geo
->level
== 10) {
7059 change
= CH_TAKEOVER
;
7064 if (geo
->level
== 0) {
7065 change
= CH_TAKEOVER
;
7070 if (geo
->level
== 0)
7071 change
= CH_MIGRATION
;
7074 if (geo
->level
== 0) {
7075 change
= CH_TAKEOVER
;
7082 Name
" Error. Level Migration from %d to %d "
7084 info
.array
.level
, geo
->level
);
7085 goto analyse_change_exit
;
7088 geo
->level
= info
.array
.level
;
7090 if ((geo
->layout
!= info
.array
.layout
)
7091 && ((geo
->layout
!= UnSet
) && (geo
->layout
!= -1))) {
7092 change
= CH_MIGRATION
;
7093 if ((info
.array
.layout
== 0)
7094 && (info
.array
.level
== 5)
7095 && (geo
->layout
== 5)) {
7096 /* reshape 5 -> 4 */
7097 } else if ((info
.array
.layout
== 5)
7098 && (info
.array
.level
== 5)
7099 && (geo
->layout
== 0)) {
7100 /* reshape 4 -> 5 */
7105 Name
" Error. Layout Migration from %d to %d "
7107 info
.array
.layout
, geo
->layout
);
7109 goto analyse_change_exit
;
7112 geo
->layout
= info
.array
.layout
;
7114 if ((geo
->chunksize
> 0) && (geo
->chunksize
!= UnSet
)
7115 && (geo
->chunksize
!= info
.array
.chunk_size
))
7116 change
= CH_MIGRATION
;
7118 geo
->chunksize
= info
.array
.chunk_size
;
7120 chunk
= geo
->chunksize
/ 1024;
7121 if (!validate_geometry_imsm(st
,
7131 struct intel_super
*super
= st
->sb
;
7132 struct imsm_super
*mpb
= super
->anchor
;
7134 if (mpb
->num_raid_devs
> 1) {
7136 Name
" Error. Cannot perform operation on %s"
7137 "- for this operation it MUST be single "
7138 "array in container\n",
7144 analyse_change_exit
:
7149 int imsm_takeover(struct supertype
*st
, struct geo_params
*geo
)
7151 struct intel_super
*super
= st
->sb
;
7152 struct imsm_update_takeover
*u
;
7154 u
= malloc(sizeof(struct imsm_update_takeover
));
7158 u
->type
= update_takeover
;
7159 u
->subarray
= super
->current_vol
;
7161 /* 10->0 transition */
7162 if (geo
->level
== 0)
7163 u
->direction
= R10_TO_R0
;
7165 /* 0->10 transition */
7166 if (geo
->level
== 10)
7167 u
->direction
= R0_TO_R10
;
7169 /* update metadata locally */
7170 imsm_update_metadata_locally(st
, u
,
7171 sizeof(struct imsm_update_takeover
));
7172 /* and possibly remotely */
7173 if (st
->update_tail
)
7174 append_metadata_update(st
, u
,
7175 sizeof(struct imsm_update_takeover
));
7182 static int imsm_reshape_super(struct supertype
*st
, long long size
, int level
,
7183 int layout
, int chunksize
, int raid_disks
,
7184 int delta_disks
, char *backup
, char *dev
,
7188 struct geo_params geo
;
7190 dprintf("imsm: reshape_super called.\n");
7192 memset(&geo
, 0, sizeof(struct geo_params
));
7195 geo
.dev_id
= st
->devnum
;
7198 geo
.layout
= layout
;
7199 geo
.chunksize
= chunksize
;
7200 geo
.raid_disks
= raid_disks
;
7201 if (delta_disks
!= UnSet
)
7202 geo
.raid_disks
+= delta_disks
;
7204 dprintf("\tfor level : %i\n", geo
.level
);
7205 dprintf("\tfor raid_disks : %i\n", geo
.raid_disks
);
7207 if (experimental() == 0)
7210 if (st
->container_dev
== st
->devnum
) {
7211 /* On container level we can only increase number of devices. */
7212 dprintf("imsm: info: Container operation\n");
7213 int old_raid_disks
= 0;
7214 if (imsm_reshape_is_allowed_on_container(
7215 st
, &geo
, &old_raid_disks
)) {
7216 struct imsm_update_reshape
*u
= NULL
;
7219 len
= imsm_create_metadata_update_for_reshape(
7220 st
, &geo
, old_raid_disks
, &u
);
7223 dprintf("imsm: Cannot prepare update\n");
7224 goto exit_imsm_reshape_super
;
7228 /* update metadata locally */
7229 imsm_update_metadata_locally(st
, u
, len
);
7230 /* and possibly remotely */
7231 if (st
->update_tail
)
7232 append_metadata_update(st
, u
, len
);
7237 fprintf(stderr
, Name
": (imsm) Operation "
7238 "is not allowed on this container\n");
7241 /* On volume level we support following operations
7242 * - takeover: raid10 -> raid0; raid0 -> raid10
7243 * - chunk size migration
7244 * - migration: raid5 -> raid0; raid0 -> raid5
7246 struct intel_super
*super
= st
->sb
;
7247 struct intel_dev
*dev
= super
->devlist
;
7249 dprintf("imsm: info: Volume operation\n");
7250 /* find requested device */
7252 imsm_find_array_minor_by_subdev(dev
->index
, st
->container_dev
, &devnum
);
7253 if (devnum
== geo
.dev_id
)
7258 fprintf(stderr
, Name
" Cannot find %s (%i) subarray\n",
7259 geo
.dev_name
, geo
.dev_id
);
7260 goto exit_imsm_reshape_super
;
7262 super
->current_vol
= dev
->index
;
7263 change
= imsm_analyze_change(st
, &geo
);
7266 ret_val
= imsm_takeover(st
, &geo
);
7276 exit_imsm_reshape_super
:
7277 dprintf("imsm: reshape_super Exit code = %i\n", ret_val
);
7281 static int imsm_manage_reshape(
7282 int afd
, struct mdinfo
*sra
, struct reshape
*reshape
,
7283 struct supertype
*st
, unsigned long stripes
,
7284 int *fds
, unsigned long long *offsets
,
7285 int dests
, int *destfd
, unsigned long long *destoffsets
)
7287 /* Just use child_monitor for now */
7288 return child_monitor(
7289 afd
, sra
, reshape
, st
, stripes
,
7290 fds
, offsets
, dests
, destfd
, destoffsets
);
7292 #endif /* MDASSEMBLE */
7294 struct superswitch super_imsm
= {
7296 .examine_super
= examine_super_imsm
,
7297 .brief_examine_super
= brief_examine_super_imsm
,
7298 .brief_examine_subarrays
= brief_examine_subarrays_imsm
,
7299 .export_examine_super
= export_examine_super_imsm
,
7300 .detail_super
= detail_super_imsm
,
7301 .brief_detail_super
= brief_detail_super_imsm
,
7302 .write_init_super
= write_init_super_imsm
,
7303 .validate_geometry
= validate_geometry_imsm
,
7304 .add_to_super
= add_to_super_imsm
,
7305 .remove_from_super
= remove_from_super_imsm
,
7306 .detail_platform
= detail_platform_imsm
,
7307 .kill_subarray
= kill_subarray_imsm
,
7308 .update_subarray
= update_subarray_imsm
,
7309 .load_container
= load_container_imsm
,
7310 .default_geometry
= default_geometry_imsm
,
7311 .get_disk_controller_domain
= imsm_get_disk_controller_domain
,
7312 .reshape_super
= imsm_reshape_super
,
7313 .manage_reshape
= imsm_manage_reshape
,
7315 .match_home
= match_home_imsm
,
7316 .uuid_from_super
= uuid_from_super_imsm
,
7317 .getinfo_super
= getinfo_super_imsm
,
7318 .getinfo_super_disks
= getinfo_super_disks_imsm
,
7319 .update_super
= update_super_imsm
,
7321 .avail_size
= avail_size_imsm
,
7322 .min_acceptable_spare_size
= min_acceptable_spare_size_imsm
,
7324 .compare_super
= compare_super_imsm
,
7326 .load_super
= load_super_imsm
,
7327 .init_super
= init_super_imsm
,
7328 .store_super
= store_super_imsm
,
7329 .free_super
= free_super_imsm
,
7330 .match_metadata_desc
= match_metadata_desc_imsm
,
7331 .container_content
= container_content_imsm
,
7338 .open_new
= imsm_open_new
,
7339 .set_array_state
= imsm_set_array_state
,
7340 .set_disk
= imsm_set_disk
,
7341 .sync_metadata
= imsm_sync_metadata
,
7342 .activate_spare
= imsm_activate_spare
,
7343 .process_update
= imsm_process_update
,
7344 .prepare_update
= imsm_prepare_update
,
7345 #endif /* MDASSEMBLE */