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
,
414 struct intel_hba
*hba
;
416 /* check if disk attached to Intel HBA */
417 hba
= find_intel_hba(super
->hba
, device
);
420 /* Check if HBA is already attached to super */
421 if (super
->hba
== NULL
) {
422 super
->hba
= alloc_intel_hba(device
);
427 /* Intel metadata allows for all disks attached to the same type HBA.
428 * Do not sypport odf HBA types mixing
430 if (device
->type
!= hba
->type
)
436 hba
->next
= alloc_intel_hba(device
);
440 static struct sys_dev
* find_disk_attached_hba(int fd
, const char *devname
)
442 struct sys_dev
*list
, *elem
, *prev
;
445 if ((list
= find_intel_devices()) == NULL
)
449 disk_path
= (char *) devname
;
451 disk_path
= diskfd_to_devpath(fd
);
458 for (prev
= NULL
, elem
= list
; elem
; prev
= elem
, elem
= elem
->next
) {
459 if (path_attached_to_hba(disk_path
, elem
->path
)) {
463 prev
->next
= elem
->next
;
465 if (disk_path
!= devname
)
471 if (disk_path
!= devname
)
477 #endif /* MDASSEMBLE */
480 static struct supertype
*match_metadata_desc_imsm(char *arg
)
482 struct supertype
*st
;
484 if (strcmp(arg
, "imsm") != 0 &&
485 strcmp(arg
, "default") != 0
489 st
= malloc(sizeof(*st
));
492 memset(st
, 0, sizeof(*st
));
493 st
->container_dev
= NoMdDev
;
494 st
->ss
= &super_imsm
;
495 st
->max_devs
= IMSM_MAX_DEVICES
;
496 st
->minor_version
= 0;
502 static __u8
*get_imsm_version(struct imsm_super
*mpb
)
504 return &mpb
->sig
[MPB_SIG_LEN
];
508 /* retrieve a disk directly from the anchor when the anchor is known to be
509 * up-to-date, currently only at load time
511 static struct imsm_disk
*__get_imsm_disk(struct imsm_super
*mpb
, __u8 index
)
513 if (index
>= mpb
->num_disks
)
515 return &mpb
->disk
[index
];
518 /* retrieve the disk description based on a index of the disk
521 static struct dl
*get_imsm_dl_disk(struct intel_super
*super
, __u8 index
)
525 for (d
= super
->disks
; d
; d
= d
->next
)
526 if (d
->index
== index
)
531 /* retrieve a disk from the parsed metadata */
532 static struct imsm_disk
*get_imsm_disk(struct intel_super
*super
, __u8 index
)
536 dl
= get_imsm_dl_disk(super
, index
);
543 /* generate a checksum directly from the anchor when the anchor is known to be
544 * up-to-date, currently only at load or write_super after coalescing
546 static __u32
__gen_imsm_checksum(struct imsm_super
*mpb
)
548 __u32 end
= mpb
->mpb_size
/ sizeof(end
);
549 __u32
*p
= (__u32
*) mpb
;
553 sum
+= __le32_to_cpu(*p
);
557 return sum
- __le32_to_cpu(mpb
->check_sum
);
560 static size_t sizeof_imsm_map(struct imsm_map
*map
)
562 return sizeof(struct imsm_map
) + sizeof(__u32
) * (map
->num_members
- 1);
565 struct imsm_map
*get_imsm_map(struct imsm_dev
*dev
, int second_map
)
567 /* A device can have 2 maps if it is in the middle of a migration.
569 * 0 - we return the first map
570 * 1 - we return the second map if it exists, else NULL
571 * -1 - we return the second map if it exists, else the first
573 struct imsm_map
*map
= &dev
->vol
.map
[0];
575 if (second_map
== 1 && !dev
->vol
.migr_state
)
577 else if (second_map
== 1 ||
578 (second_map
< 0 && dev
->vol
.migr_state
)) {
581 return ptr
+ sizeof_imsm_map(map
);
587 /* return the size of the device.
588 * migr_state increases the returned size if map[0] were to be duplicated
590 static size_t sizeof_imsm_dev(struct imsm_dev
*dev
, int migr_state
)
592 size_t size
= sizeof(*dev
) - sizeof(struct imsm_map
) +
593 sizeof_imsm_map(get_imsm_map(dev
, 0));
595 /* migrating means an additional map */
596 if (dev
->vol
.migr_state
)
597 size
+= sizeof_imsm_map(get_imsm_map(dev
, 1));
599 size
+= sizeof_imsm_map(get_imsm_map(dev
, 0));
605 /* retrieve disk serial number list from a metadata update */
606 static struct disk_info
*get_disk_info(struct imsm_update_create_array
*update
)
609 struct disk_info
*inf
;
611 inf
= u
+ sizeof(*update
) - sizeof(struct imsm_dev
) +
612 sizeof_imsm_dev(&update
->dev
, 0);
618 static struct imsm_dev
*__get_imsm_dev(struct imsm_super
*mpb
, __u8 index
)
624 if (index
>= mpb
->num_raid_devs
)
627 /* devices start after all disks */
628 offset
= ((void *) &mpb
->disk
[mpb
->num_disks
]) - _mpb
;
630 for (i
= 0; i
<= index
; i
++)
632 return _mpb
+ offset
;
634 offset
+= sizeof_imsm_dev(_mpb
+ offset
, 0);
639 static struct imsm_dev
*get_imsm_dev(struct intel_super
*super
, __u8 index
)
641 struct intel_dev
*dv
;
643 if (index
>= super
->anchor
->num_raid_devs
)
645 for (dv
= super
->devlist
; dv
; dv
= dv
->next
)
646 if (dv
->index
== index
)
654 * == 1 get second map
655 * == -1 than get map according to the current migr_state
657 static __u32
get_imsm_ord_tbl_ent(struct imsm_dev
*dev
,
661 struct imsm_map
*map
;
663 map
= get_imsm_map(dev
, second_map
);
665 /* top byte identifies disk under rebuild */
666 return __le32_to_cpu(map
->disk_ord_tbl
[slot
]);
669 #define ord_to_idx(ord) (((ord) << 8) >> 8)
670 static __u32
get_imsm_disk_idx(struct imsm_dev
*dev
, int slot
, int second_map
)
672 __u32 ord
= get_imsm_ord_tbl_ent(dev
, slot
, second_map
);
674 return ord_to_idx(ord
);
677 static void set_imsm_ord_tbl_ent(struct imsm_map
*map
, int slot
, __u32 ord
)
679 map
->disk_ord_tbl
[slot
] = __cpu_to_le32(ord
);
682 static int get_imsm_disk_slot(struct imsm_map
*map
, unsigned idx
)
687 for (slot
= 0; slot
< map
->num_members
; slot
++) {
688 ord
= __le32_to_cpu(map
->disk_ord_tbl
[slot
]);
689 if (ord_to_idx(ord
) == idx
)
696 static int get_imsm_raid_level(struct imsm_map
*map
)
698 if (map
->raid_level
== 1) {
699 if (map
->num_members
== 2)
705 return map
->raid_level
;
708 static int cmp_extent(const void *av
, const void *bv
)
710 const struct extent
*a
= av
;
711 const struct extent
*b
= bv
;
712 if (a
->start
< b
->start
)
714 if (a
->start
> b
->start
)
719 static int count_memberships(struct dl
*dl
, struct intel_super
*super
)
724 for (i
= 0; i
< super
->anchor
->num_raid_devs
; i
++) {
725 struct imsm_dev
*dev
= get_imsm_dev(super
, i
);
726 struct imsm_map
*map
= get_imsm_map(dev
, 0);
728 if (get_imsm_disk_slot(map
, dl
->index
) >= 0)
735 static struct extent
*get_extents(struct intel_super
*super
, struct dl
*dl
)
737 /* find a list of used extents on the given physical device */
738 struct extent
*rv
, *e
;
740 int memberships
= count_memberships(dl
, super
);
741 __u32 reservation
= MPB_SECTOR_CNT
+ IMSM_RESERVED_SECTORS
;
743 rv
= malloc(sizeof(struct extent
) * (memberships
+ 1));
748 for (i
= 0; i
< super
->anchor
->num_raid_devs
; i
++) {
749 struct imsm_dev
*dev
= get_imsm_dev(super
, i
);
750 struct imsm_map
*map
= get_imsm_map(dev
, 0);
752 if (get_imsm_disk_slot(map
, dl
->index
) >= 0) {
753 e
->start
= __le32_to_cpu(map
->pba_of_lba0
);
754 e
->size
= __le32_to_cpu(map
->blocks_per_member
);
758 qsort(rv
, memberships
, sizeof(*rv
), cmp_extent
);
760 /* determine the start of the metadata
761 * when no raid devices are defined use the default
762 * ...otherwise allow the metadata to truncate the value
763 * as is the case with older versions of imsm
766 struct extent
*last
= &rv
[memberships
- 1];
769 remainder
= __le32_to_cpu(dl
->disk
.total_blocks
) -
770 (last
->start
+ last
->size
);
771 /* round down to 1k block to satisfy precision of the kernel
775 /* make sure remainder is still sane */
776 if (remainder
< (unsigned)ROUND_UP(super
->len
, 512) >> 9)
777 remainder
= ROUND_UP(super
->len
, 512) >> 9;
778 if (reservation
> remainder
)
779 reservation
= remainder
;
781 e
->start
= __le32_to_cpu(dl
->disk
.total_blocks
) - reservation
;
786 /* try to determine how much space is reserved for metadata from
787 * the last get_extents() entry, otherwise fallback to the
790 static __u32
imsm_reserved_sectors(struct intel_super
*super
, struct dl
*dl
)
796 /* for spares just return a minimal reservation which will grow
797 * once the spare is picked up by an array
800 return MPB_SECTOR_CNT
;
802 e
= get_extents(super
, dl
);
804 return MPB_SECTOR_CNT
+ IMSM_RESERVED_SECTORS
;
806 /* scroll to last entry */
807 for (i
= 0; e
[i
].size
; i
++)
810 rv
= __le32_to_cpu(dl
->disk
.total_blocks
) - e
[i
].start
;
817 static int is_spare(struct imsm_disk
*disk
)
819 return (disk
->status
& SPARE_DISK
) == SPARE_DISK
;
822 static int is_configured(struct imsm_disk
*disk
)
824 return (disk
->status
& CONFIGURED_DISK
) == CONFIGURED_DISK
;
827 static int is_failed(struct imsm_disk
*disk
)
829 return (disk
->status
& FAILED_DISK
) == FAILED_DISK
;
832 /* Return minimum size of a spare that can be used in this array*/
833 static unsigned long long min_acceptable_spare_size_imsm(struct supertype
*st
)
835 struct intel_super
*super
= st
->sb
;
839 unsigned long long rv
= 0;
843 /* find first active disk in array */
845 while (dl
&& (is_failed(&dl
->disk
) || dl
->index
== -1))
849 /* find last lba used by subarrays */
850 e
= get_extents(super
, dl
);
853 for (i
= 0; e
[i
].size
; i
++)
856 rv
= e
[i
-1].start
+ e
[i
-1].size
;
858 /* add the amount of space needed for metadata */
859 rv
= rv
+ MPB_SECTOR_CNT
+ IMSM_RESERVED_SECTORS
;
864 static __u64
blocks_per_migr_unit(struct imsm_dev
*dev
);
866 static void print_imsm_dev(struct imsm_dev
*dev
, char *uuid
, int disk_idx
)
870 struct imsm_map
*map
= get_imsm_map(dev
, 0);
871 struct imsm_map
*map2
= get_imsm_map(dev
, 1);
875 printf("[%.16s]:\n", dev
->volume
);
876 printf(" UUID : %s\n", uuid
);
877 printf(" RAID Level : %d", get_imsm_raid_level(map
));
879 printf(" <-- %d", get_imsm_raid_level(map2
));
881 printf(" Members : %d", map
->num_members
);
883 printf(" <-- %d", map2
->num_members
);
885 printf(" Slots : [");
886 for (i
= 0; i
< map
->num_members
; i
++) {
887 ord
= get_imsm_ord_tbl_ent(dev
, i
, 0);
888 printf("%s", ord
& IMSM_ORD_REBUILD
? "_" : "U");
893 for (i
= 0; i
< map2
->num_members
; i
++) {
894 ord
= get_imsm_ord_tbl_ent(dev
, i
, 1);
895 printf("%s", ord
& IMSM_ORD_REBUILD
? "_" : "U");
900 printf(" Failed disk : ");
901 if (map
->failed_disk_num
== 0xff)
904 printf("%i", map
->failed_disk_num
);
906 slot
= get_imsm_disk_slot(map
, disk_idx
);
908 ord
= get_imsm_ord_tbl_ent(dev
, slot
, -1);
909 printf(" This Slot : %d%s\n", slot
,
910 ord
& IMSM_ORD_REBUILD
? " (out-of-sync)" : "");
912 printf(" This Slot : ?\n");
913 sz
= __le32_to_cpu(dev
->size_high
);
915 sz
+= __le32_to_cpu(dev
->size_low
);
916 printf(" Array Size : %llu%s\n", (unsigned long long)sz
,
917 human_size(sz
* 512));
918 sz
= __le32_to_cpu(map
->blocks_per_member
);
919 printf(" Per Dev Size : %llu%s\n", (unsigned long long)sz
,
920 human_size(sz
* 512));
921 printf(" Sector Offset : %u\n",
922 __le32_to_cpu(map
->pba_of_lba0
));
923 printf(" Num Stripes : %u\n",
924 __le32_to_cpu(map
->num_data_stripes
));
925 printf(" Chunk Size : %u KiB",
926 __le16_to_cpu(map
->blocks_per_strip
) / 2);
928 printf(" <-- %u KiB",
929 __le16_to_cpu(map2
->blocks_per_strip
) / 2);
931 printf(" Reserved : %d\n", __le32_to_cpu(dev
->reserved_blocks
));
932 printf(" Migrate State : ");
933 if (dev
->vol
.migr_state
) {
934 if (migr_type(dev
) == MIGR_INIT
)
935 printf("initialize\n");
936 else if (migr_type(dev
) == MIGR_REBUILD
)
938 else if (migr_type(dev
) == MIGR_VERIFY
)
940 else if (migr_type(dev
) == MIGR_GEN_MIGR
)
941 printf("general migration\n");
942 else if (migr_type(dev
) == MIGR_STATE_CHANGE
)
943 printf("state change\n");
944 else if (migr_type(dev
) == MIGR_REPAIR
)
947 printf("<unknown:%d>\n", migr_type(dev
));
950 printf(" Map State : %s", map_state_str
[map
->map_state
]);
951 if (dev
->vol
.migr_state
) {
952 struct imsm_map
*map
= get_imsm_map(dev
, 1);
954 printf(" <-- %s", map_state_str
[map
->map_state
]);
955 printf("\n Checkpoint : %u (%llu)",
956 __le32_to_cpu(dev
->vol
.curr_migr_unit
),
957 (unsigned long long)blocks_per_migr_unit(dev
));
960 printf(" Dirty State : %s\n", dev
->vol
.dirty
? "dirty" : "clean");
963 static void print_imsm_disk(struct imsm_super
*mpb
, int index
, __u32 reserved
)
965 struct imsm_disk
*disk
= __get_imsm_disk(mpb
, index
);
966 char str
[MAX_RAID_SERIAL_LEN
+ 1];
969 if (index
< 0 || !disk
)
973 snprintf(str
, MAX_RAID_SERIAL_LEN
+ 1, "%s", disk
->serial
);
974 printf(" Disk%02d Serial : %s\n", index
, str
);
975 printf(" State :%s%s%s\n", is_spare(disk
) ? " spare" : "",
976 is_configured(disk
) ? " active" : "",
977 is_failed(disk
) ? " failed" : "");
978 printf(" Id : %08x\n", __le32_to_cpu(disk
->scsi_id
));
979 sz
= __le32_to_cpu(disk
->total_blocks
) - reserved
;
980 printf(" Usable Size : %llu%s\n", (unsigned long long)sz
,
981 human_size(sz
* 512));
984 static void getinfo_super_imsm(struct supertype
*st
, struct mdinfo
*info
, char *map
);
986 static void examine_super_imsm(struct supertype
*st
, char *homehost
)
988 struct intel_super
*super
= st
->sb
;
989 struct imsm_super
*mpb
= super
->anchor
;
990 char str
[MAX_SIGNATURE_LENGTH
];
995 __u32 reserved
= imsm_reserved_sectors(super
, super
->disks
);
998 snprintf(str
, MPB_SIG_LEN
, "%s", mpb
->sig
);
999 printf(" Magic : %s\n", str
);
1000 snprintf(str
, strlen(MPB_VERSION_RAID0
), "%s", get_imsm_version(mpb
));
1001 printf(" Version : %s\n", get_imsm_version(mpb
));
1002 printf(" Orig Family : %08x\n", __le32_to_cpu(mpb
->orig_family_num
));
1003 printf(" Family : %08x\n", __le32_to_cpu(mpb
->family_num
));
1004 printf(" Generation : %08x\n", __le32_to_cpu(mpb
->generation_num
));
1005 getinfo_super_imsm(st
, &info
, NULL
);
1006 fname_from_uuid(st
, &info
, nbuf
, ':');
1007 printf(" UUID : %s\n", nbuf
+ 5);
1008 sum
= __le32_to_cpu(mpb
->check_sum
);
1009 printf(" Checksum : %08x %s\n", sum
,
1010 __gen_imsm_checksum(mpb
) == sum
? "correct" : "incorrect");
1011 printf(" MPB Sectors : %d\n", mpb_sectors(mpb
));
1012 printf(" Disks : %d\n", mpb
->num_disks
);
1013 printf(" RAID Devices : %d\n", mpb
->num_raid_devs
);
1014 print_imsm_disk(mpb
, super
->disks
->index
, reserved
);
1015 if (super
->bbm_log
) {
1016 struct bbm_log
*log
= super
->bbm_log
;
1019 printf("Bad Block Management Log:\n");
1020 printf(" Log Size : %d\n", __le32_to_cpu(mpb
->bbm_log_size
));
1021 printf(" Signature : %x\n", __le32_to_cpu(log
->signature
));
1022 printf(" Entry Count : %d\n", __le32_to_cpu(log
->entry_count
));
1023 printf(" Spare Blocks : %d\n", __le32_to_cpu(log
->reserved_spare_block_count
));
1024 printf(" First Spare : %llx\n",
1025 (unsigned long long) __le64_to_cpu(log
->first_spare_lba
));
1027 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
1029 struct imsm_dev
*dev
= __get_imsm_dev(mpb
, i
);
1031 super
->current_vol
= i
;
1032 getinfo_super_imsm(st
, &info
, NULL
);
1033 fname_from_uuid(st
, &info
, nbuf
, ':');
1034 print_imsm_dev(dev
, nbuf
+ 5, super
->disks
->index
);
1036 for (i
= 0; i
< mpb
->num_disks
; i
++) {
1037 if (i
== super
->disks
->index
)
1039 print_imsm_disk(mpb
, i
, reserved
);
1041 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
1042 struct imsm_disk
*disk
;
1043 char str
[MAX_RAID_SERIAL_LEN
+ 1];
1051 snprintf(str
, MAX_RAID_SERIAL_LEN
+ 1, "%s", disk
->serial
);
1052 printf(" Disk Serial : %s\n", str
);
1053 printf(" State :%s%s%s\n", is_spare(disk
) ? " spare" : "",
1054 is_configured(disk
) ? " active" : "",
1055 is_failed(disk
) ? " failed" : "");
1056 printf(" Id : %08x\n", __le32_to_cpu(disk
->scsi_id
));
1057 sz
= __le32_to_cpu(disk
->total_blocks
) - reserved
;
1058 printf(" Usable Size : %llu%s\n", (unsigned long long)sz
,
1059 human_size(sz
* 512));
1063 static void brief_examine_super_imsm(struct supertype
*st
, int verbose
)
1065 /* We just write a generic IMSM ARRAY entry */
1068 struct intel_super
*super
= st
->sb
;
1070 if (!super
->anchor
->num_raid_devs
) {
1071 printf("ARRAY metadata=imsm\n");
1075 getinfo_super_imsm(st
, &info
, NULL
);
1076 fname_from_uuid(st
, &info
, nbuf
, ':');
1077 printf("ARRAY metadata=imsm UUID=%s\n", nbuf
+ 5);
1080 static void brief_examine_subarrays_imsm(struct supertype
*st
, int verbose
)
1082 /* We just write a generic IMSM ARRAY entry */
1086 struct intel_super
*super
= st
->sb
;
1089 if (!super
->anchor
->num_raid_devs
)
1092 getinfo_super_imsm(st
, &info
, NULL
);
1093 fname_from_uuid(st
, &info
, nbuf
, ':');
1094 for (i
= 0; i
< super
->anchor
->num_raid_devs
; i
++) {
1095 struct imsm_dev
*dev
= get_imsm_dev(super
, i
);
1097 super
->current_vol
= i
;
1098 getinfo_super_imsm(st
, &info
, NULL
);
1099 fname_from_uuid(st
, &info
, nbuf1
, ':');
1100 printf("ARRAY /dev/md/%.16s container=%s member=%d UUID=%s\n",
1101 dev
->volume
, nbuf
+ 5, i
, nbuf1
+ 5);
1105 static void export_examine_super_imsm(struct supertype
*st
)
1107 struct intel_super
*super
= st
->sb
;
1108 struct imsm_super
*mpb
= super
->anchor
;
1112 getinfo_super_imsm(st
, &info
, NULL
);
1113 fname_from_uuid(st
, &info
, nbuf
, ':');
1114 printf("MD_METADATA=imsm\n");
1115 printf("MD_LEVEL=container\n");
1116 printf("MD_UUID=%s\n", nbuf
+5);
1117 printf("MD_DEVICES=%u\n", mpb
->num_disks
);
1120 static void detail_super_imsm(struct supertype
*st
, char *homehost
)
1125 getinfo_super_imsm(st
, &info
, NULL
);
1126 fname_from_uuid(st
, &info
, nbuf
, ':');
1127 printf("\n UUID : %s\n", nbuf
+ 5);
1130 static void brief_detail_super_imsm(struct supertype
*st
)
1134 getinfo_super_imsm(st
, &info
, NULL
);
1135 fname_from_uuid(st
, &info
, nbuf
, ':');
1136 printf(" UUID=%s", nbuf
+ 5);
1139 static int imsm_read_serial(int fd
, char *devname
, __u8
*serial
);
1140 static void fd2devname(int fd
, char *name
);
1142 static int ahci_enumerate_ports(const char *hba_path
, int port_count
, int host_base
, int verbose
)
1144 /* dump an unsorted list of devices attached to AHCI Intel storage
1145 * controller, as well as non-connected ports
1147 int hba_len
= strlen(hba_path
) + 1;
1152 unsigned long port_mask
= (1 << port_count
) - 1;
1154 if (port_count
> (int)sizeof(port_mask
) * 8) {
1156 fprintf(stderr
, Name
": port_count %d out of range\n", port_count
);
1160 /* scroll through /sys/dev/block looking for devices attached to
1163 dir
= opendir("/sys/dev/block");
1164 for (ent
= dir
? readdir(dir
) : NULL
; ent
; ent
= readdir(dir
)) {
1175 if (sscanf(ent
->d_name
, "%d:%d", &major
, &minor
) != 2)
1177 path
= devt_to_devpath(makedev(major
, minor
));
1180 if (!path_attached_to_hba(path
, hba_path
)) {
1186 /* retrieve the scsi device type */
1187 if (asprintf(&device
, "/sys/dev/block/%d:%d/device/xxxxxxx", major
, minor
) < 0) {
1189 fprintf(stderr
, Name
": failed to allocate 'device'\n");
1193 sprintf(device
, "/sys/dev/block/%d:%d/device/type", major
, minor
);
1194 if (load_sys(device
, buf
) != 0) {
1196 fprintf(stderr
, Name
": failed to read device type for %s\n",
1202 type
= strtoul(buf
, NULL
, 10);
1204 /* if it's not a disk print the vendor and model */
1205 if (!(type
== 0 || type
== 7 || type
== 14)) {
1208 sprintf(device
, "/sys/dev/block/%d:%d/device/vendor", major
, minor
);
1209 if (load_sys(device
, buf
) == 0) {
1210 strncpy(vendor
, buf
, sizeof(vendor
));
1211 vendor
[sizeof(vendor
) - 1] = '\0';
1212 c
= (char *) &vendor
[sizeof(vendor
) - 1];
1213 while (isspace(*c
) || *c
== '\0')
1217 sprintf(device
, "/sys/dev/block/%d:%d/device/model", major
, minor
);
1218 if (load_sys(device
, buf
) == 0) {
1219 strncpy(model
, buf
, sizeof(model
));
1220 model
[sizeof(model
) - 1] = '\0';
1221 c
= (char *) &model
[sizeof(model
) - 1];
1222 while (isspace(*c
) || *c
== '\0')
1226 if (vendor
[0] && model
[0])
1227 sprintf(buf
, "%.64s %.64s", vendor
, model
);
1229 switch (type
) { /* numbers from hald/linux/device.c */
1230 case 1: sprintf(buf
, "tape"); break;
1231 case 2: sprintf(buf
, "printer"); break;
1232 case 3: sprintf(buf
, "processor"); break;
1234 case 5: sprintf(buf
, "cdrom"); break;
1235 case 6: sprintf(buf
, "scanner"); break;
1236 case 8: sprintf(buf
, "media_changer"); break;
1237 case 9: sprintf(buf
, "comm"); break;
1238 case 12: sprintf(buf
, "raid"); break;
1239 default: sprintf(buf
, "unknown");
1245 /* chop device path to 'host%d' and calculate the port number */
1246 c
= strchr(&path
[hba_len
], '/');
1249 fprintf(stderr
, Name
": %s - invalid path name\n", path
+ hba_len
);
1254 if (sscanf(&path
[hba_len
], "host%d", &port
) == 1)
1258 *c
= '/'; /* repair the full string */
1259 fprintf(stderr
, Name
": failed to determine port number for %s\n",
1266 /* mark this port as used */
1267 port_mask
&= ~(1 << port
);
1269 /* print out the device information */
1271 printf(" Port%d : - non-disk device (%s) -\n", port
, buf
);
1275 fd
= dev_open(ent
->d_name
, O_RDONLY
);
1277 printf(" Port%d : - disk info unavailable -\n", port
);
1279 fd2devname(fd
, buf
);
1280 printf(" Port%d : %s", port
, buf
);
1281 if (imsm_read_serial(fd
, NULL
, (__u8
*) buf
) == 0)
1282 printf(" (%s)\n", buf
);
1297 for (i
= 0; i
< port_count
; i
++)
1298 if (port_mask
& (1 << i
))
1299 printf(" Port%d : - no device attached -\n", i
);
1307 static void print_found_intel_controllers(struct sys_dev
*elem
)
1309 for (; elem
; elem
= elem
->next
) {
1310 fprintf(stderr
, Name
": found Intel(R) ");
1311 if (elem
->type
== SYS_DEV_SATA
)
1312 fprintf(stderr
, "SATA ");
1313 else if (elem
->type
== SYS_DEV_SAS
)
1314 fprintf(stderr
, "SAS ");
1315 fprintf(stderr
, "RAID controller");
1317 fprintf(stderr
, " at %s", elem
->pci_id
);
1318 fprintf(stderr
, ".\n");
1323 static int ahci_get_port_count(const char *hba_path
, int *port_count
)
1330 if ((dir
= opendir(hba_path
)) == NULL
)
1333 for (ent
= readdir(dir
); ent
; ent
= readdir(dir
)) {
1336 if (sscanf(ent
->d_name
, "host%d", &host
) != 1)
1338 if (*port_count
== 0)
1340 else if (host
< host_base
)
1343 if (host
+ 1 > *port_count
+ host_base
)
1344 *port_count
= host
+ 1 - host_base
;
1350 static int detail_platform_imsm(int verbose
, int enumerate_only
)
1352 /* There are two components to imsm platform support, the ahci SATA
1353 * controller and the option-rom. To find the SATA controller we
1354 * simply look in /sys/bus/pci/drivers/ahci to see if an ahci
1355 * controller with the Intel vendor id is present. This approach
1356 * allows mdadm to leverage the kernel's ahci detection logic, with the
1357 * caveat that if ahci.ko is not loaded mdadm will not be able to
1358 * detect platform raid capabilities. The option-rom resides in a
1359 * platform "Adapter ROM". We scan for its signature to retrieve the
1360 * platform capabilities. If raid support is disabled in the BIOS the
1361 * option-rom capability structure will not be available.
1363 const struct imsm_orom
*orom
;
1364 struct sys_dev
*list
, *hba
;
1369 if (enumerate_only
) {
1370 if (check_env("IMSM_NO_PLATFORM") || find_imsm_orom())
1375 list
= find_intel_devices();
1378 fprintf(stderr
, Name
": no active Intel(R) RAID "
1379 "controller found.\n");
1380 free_sys_dev(&list
);
1383 print_found_intel_controllers(list
);
1385 orom
= find_imsm_orom();
1387 free_sys_dev(&list
);
1389 fprintf(stderr
, Name
": imsm option-rom not found\n");
1393 printf(" Platform : Intel(R) Matrix Storage Manager\n");
1394 printf(" Version : %d.%d.%d.%d\n", orom
->major_ver
, orom
->minor_ver
,
1395 orom
->hotfix_ver
, orom
->build
);
1396 printf(" RAID Levels :%s%s%s%s%s\n",
1397 imsm_orom_has_raid0(orom
) ? " raid0" : "",
1398 imsm_orom_has_raid1(orom
) ? " raid1" : "",
1399 imsm_orom_has_raid1e(orom
) ? " raid1e" : "",
1400 imsm_orom_has_raid10(orom
) ? " raid10" : "",
1401 imsm_orom_has_raid5(orom
) ? " raid5" : "");
1402 printf(" Chunk Sizes :%s%s%s%s%s%s%s%s%s%s%s%s%s%s%s%s\n",
1403 imsm_orom_has_chunk(orom
, 2) ? " 2k" : "",
1404 imsm_orom_has_chunk(orom
, 4) ? " 4k" : "",
1405 imsm_orom_has_chunk(orom
, 8) ? " 8k" : "",
1406 imsm_orom_has_chunk(orom
, 16) ? " 16k" : "",
1407 imsm_orom_has_chunk(orom
, 32) ? " 32k" : "",
1408 imsm_orom_has_chunk(orom
, 64) ? " 64k" : "",
1409 imsm_orom_has_chunk(orom
, 128) ? " 128k" : "",
1410 imsm_orom_has_chunk(orom
, 256) ? " 256k" : "",
1411 imsm_orom_has_chunk(orom
, 512) ? " 512k" : "",
1412 imsm_orom_has_chunk(orom
, 1024*1) ? " 1M" : "",
1413 imsm_orom_has_chunk(orom
, 1024*2) ? " 2M" : "",
1414 imsm_orom_has_chunk(orom
, 1024*4) ? " 4M" : "",
1415 imsm_orom_has_chunk(orom
, 1024*8) ? " 8M" : "",
1416 imsm_orom_has_chunk(orom
, 1024*16) ? " 16M" : "",
1417 imsm_orom_has_chunk(orom
, 1024*32) ? " 32M" : "",
1418 imsm_orom_has_chunk(orom
, 1024*64) ? " 64M" : "");
1419 printf(" Max Disks : %d\n", orom
->tds
);
1420 printf(" Max Volumes : %d\n", orom
->vpa
);
1422 for (hba
= list
; hba
; hba
= hba
->next
) {
1423 printf(" I/O Controller : %s (%s)\n",
1424 hba
->path
, get_sys_dev_type(hba
->type
));
1426 if (hba
->type
== SYS_DEV_SATA
) {
1427 host_base
= ahci_get_port_count(hba
->path
, &port_count
);
1428 if (ahci_enumerate_ports(hba
->path
, port_count
, host_base
, verbose
)) {
1430 fprintf(stderr
, Name
": failed to enumerate "
1431 "ports on SATA controller at %s.", hba
->pci_id
);
1434 } else if (hba
->type
== SYS_DEV_SAS
) {
1436 fprintf(stderr
, Name
": failed to enumerate "
1437 "devices on SAS controller at %s.", hba
->pci_id
);
1442 free_sys_dev(&list
);
1447 static int match_home_imsm(struct supertype
*st
, char *homehost
)
1449 /* the imsm metadata format does not specify any host
1450 * identification information. We return -1 since we can never
1451 * confirm nor deny whether a given array is "meant" for this
1452 * host. We rely on compare_super and the 'family_num' fields to
1453 * exclude member disks that do not belong, and we rely on
1454 * mdadm.conf to specify the arrays that should be assembled.
1455 * Auto-assembly may still pick up "foreign" arrays.
1461 static void uuid_from_super_imsm(struct supertype
*st
, int uuid
[4])
1463 /* The uuid returned here is used for:
1464 * uuid to put into bitmap file (Create, Grow)
1465 * uuid for backup header when saving critical section (Grow)
1466 * comparing uuids when re-adding a device into an array
1467 * In these cases the uuid required is that of the data-array,
1468 * not the device-set.
1469 * uuid to recognise same set when adding a missing device back
1470 * to an array. This is a uuid for the device-set.
1472 * For each of these we can make do with a truncated
1473 * or hashed uuid rather than the original, as long as
1475 * In each case the uuid required is that of the data-array,
1476 * not the device-set.
1478 /* imsm does not track uuid's so we synthesis one using sha1 on
1479 * - The signature (Which is constant for all imsm array, but no matter)
1480 * - the orig_family_num of the container
1481 * - the index number of the volume
1482 * - the 'serial' number of the volume.
1483 * Hopefully these are all constant.
1485 struct intel_super
*super
= st
->sb
;
1488 struct sha1_ctx ctx
;
1489 struct imsm_dev
*dev
= NULL
;
1492 /* some mdadm versions failed to set ->orig_family_num, in which
1493 * case fall back to ->family_num. orig_family_num will be
1494 * fixed up with the first metadata update.
1496 family_num
= super
->anchor
->orig_family_num
;
1497 if (family_num
== 0)
1498 family_num
= super
->anchor
->family_num
;
1499 sha1_init_ctx(&ctx
);
1500 sha1_process_bytes(super
->anchor
->sig
, MPB_SIG_LEN
, &ctx
);
1501 sha1_process_bytes(&family_num
, sizeof(__u32
), &ctx
);
1502 if (super
->current_vol
>= 0)
1503 dev
= get_imsm_dev(super
, super
->current_vol
);
1505 __u32 vol
= super
->current_vol
;
1506 sha1_process_bytes(&vol
, sizeof(vol
), &ctx
);
1507 sha1_process_bytes(dev
->volume
, MAX_RAID_SERIAL_LEN
, &ctx
);
1509 sha1_finish_ctx(&ctx
, buf
);
1510 memcpy(uuid
, buf
, 4*4);
1515 get_imsm_numerical_version(struct imsm_super
*mpb
, int *m
, int *p
)
1517 __u8
*v
= get_imsm_version(mpb
);
1518 __u8
*end
= mpb
->sig
+ MAX_SIGNATURE_LENGTH
;
1519 char major
[] = { 0, 0, 0 };
1520 char minor
[] = { 0 ,0, 0 };
1521 char patch
[] = { 0, 0, 0 };
1522 char *ver_parse
[] = { major
, minor
, patch
};
1526 while (*v
!= '\0' && v
< end
) {
1527 if (*v
!= '.' && j
< 2)
1528 ver_parse
[i
][j
++] = *v
;
1536 *m
= strtol(minor
, NULL
, 0);
1537 *p
= strtol(patch
, NULL
, 0);
1541 static __u32
migr_strip_blocks_resync(struct imsm_dev
*dev
)
1543 /* migr_strip_size when repairing or initializing parity */
1544 struct imsm_map
*map
= get_imsm_map(dev
, 0);
1545 __u32 chunk
= __le32_to_cpu(map
->blocks_per_strip
);
1547 switch (get_imsm_raid_level(map
)) {
1552 return 128*1024 >> 9;
1556 static __u32
migr_strip_blocks_rebuild(struct imsm_dev
*dev
)
1558 /* migr_strip_size when rebuilding a degraded disk, no idea why
1559 * this is different than migr_strip_size_resync(), but it's good
1562 struct imsm_map
*map
= get_imsm_map(dev
, 1);
1563 __u32 chunk
= __le32_to_cpu(map
->blocks_per_strip
);
1565 switch (get_imsm_raid_level(map
)) {
1568 if (map
->num_members
% map
->num_domains
== 0)
1569 return 128*1024 >> 9;
1573 return max((__u32
) 64*1024 >> 9, chunk
);
1575 return 128*1024 >> 9;
1579 static __u32
num_stripes_per_unit_resync(struct imsm_dev
*dev
)
1581 struct imsm_map
*lo
= get_imsm_map(dev
, 0);
1582 struct imsm_map
*hi
= get_imsm_map(dev
, 1);
1583 __u32 lo_chunk
= __le32_to_cpu(lo
->blocks_per_strip
);
1584 __u32 hi_chunk
= __le32_to_cpu(hi
->blocks_per_strip
);
1586 return max((__u32
) 1, hi_chunk
/ lo_chunk
);
1589 static __u32
num_stripes_per_unit_rebuild(struct imsm_dev
*dev
)
1591 struct imsm_map
*lo
= get_imsm_map(dev
, 0);
1592 int level
= get_imsm_raid_level(lo
);
1594 if (level
== 1 || level
== 10) {
1595 struct imsm_map
*hi
= get_imsm_map(dev
, 1);
1597 return hi
->num_domains
;
1599 return num_stripes_per_unit_resync(dev
);
1602 static __u8
imsm_num_data_members(struct imsm_dev
*dev
, int second_map
)
1604 /* named 'imsm_' because raid0, raid1 and raid10
1605 * counter-intuitively have the same number of data disks
1607 struct imsm_map
*map
= get_imsm_map(dev
, second_map
);
1609 switch (get_imsm_raid_level(map
)) {
1613 return map
->num_members
;
1615 return map
->num_members
- 1;
1617 dprintf("%s: unsupported raid level\n", __func__
);
1622 static __u32
parity_segment_depth(struct imsm_dev
*dev
)
1624 struct imsm_map
*map
= get_imsm_map(dev
, 0);
1625 __u32 chunk
= __le32_to_cpu(map
->blocks_per_strip
);
1627 switch(get_imsm_raid_level(map
)) {
1630 return chunk
* map
->num_domains
;
1632 return chunk
* map
->num_members
;
1638 static __u32
map_migr_block(struct imsm_dev
*dev
, __u32 block
)
1640 struct imsm_map
*map
= get_imsm_map(dev
, 1);
1641 __u32 chunk
= __le32_to_cpu(map
->blocks_per_strip
);
1642 __u32 strip
= block
/ chunk
;
1644 switch (get_imsm_raid_level(map
)) {
1647 __u32 vol_strip
= (strip
* map
->num_domains
) + 1;
1648 __u32 vol_stripe
= vol_strip
/ map
->num_members
;
1650 return vol_stripe
* chunk
+ block
% chunk
;
1652 __u32 stripe
= strip
/ (map
->num_members
- 1);
1654 return stripe
* chunk
+ block
% chunk
;
1661 static __u64
blocks_per_migr_unit(struct imsm_dev
*dev
)
1663 /* calculate the conversion factor between per member 'blocks'
1664 * (md/{resync,rebuild}_start) and imsm migration units, return
1665 * 0 for the 'not migrating' and 'unsupported migration' cases
1667 if (!dev
->vol
.migr_state
)
1670 switch (migr_type(dev
)) {
1675 struct imsm_map
*map
= get_imsm_map(dev
, 0);
1676 __u32 stripes_per_unit
;
1677 __u32 blocks_per_unit
;
1686 /* yes, this is really the translation of migr_units to
1687 * per-member blocks in the 'resync' case
1689 stripes_per_unit
= num_stripes_per_unit_resync(dev
);
1690 migr_chunk
= migr_strip_blocks_resync(dev
);
1691 disks
= imsm_num_data_members(dev
, 0);
1692 blocks_per_unit
= stripes_per_unit
* migr_chunk
* disks
;
1693 stripe
= __le32_to_cpu(map
->blocks_per_strip
) * disks
;
1694 segment
= blocks_per_unit
/ stripe
;
1695 block_rel
= blocks_per_unit
- segment
* stripe
;
1696 parity_depth
= parity_segment_depth(dev
);
1697 block_map
= map_migr_block(dev
, block_rel
);
1698 return block_map
+ parity_depth
* segment
;
1700 case MIGR_REBUILD
: {
1701 __u32 stripes_per_unit
;
1704 stripes_per_unit
= num_stripes_per_unit_rebuild(dev
);
1705 migr_chunk
= migr_strip_blocks_rebuild(dev
);
1706 return migr_chunk
* stripes_per_unit
;
1708 case MIGR_STATE_CHANGE
:
1714 static int imsm_level_to_layout(int level
)
1722 return ALGORITHM_LEFT_ASYMMETRIC
;
1729 static void getinfo_super_imsm_volume(struct supertype
*st
, struct mdinfo
*info
, char *dmap
)
1731 struct intel_super
*super
= st
->sb
;
1732 struct imsm_dev
*dev
= get_imsm_dev(super
, super
->current_vol
);
1733 struct imsm_map
*map
= get_imsm_map(dev
, 0);
1734 struct imsm_map
*prev_map
= get_imsm_map(dev
, 1);
1735 struct imsm_map
*map_to_analyse
= map
;
1738 int map_disks
= info
->array
.raid_disks
;
1741 map_to_analyse
= prev_map
;
1743 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
1744 if (dl
->raiddisk
== info
->disk
.raid_disk
)
1746 info
->container_member
= super
->current_vol
;
1747 info
->array
.raid_disks
= map_to_analyse
->num_members
;
1748 info
->array
.level
= get_imsm_raid_level(map_to_analyse
);
1749 info
->array
.layout
= imsm_level_to_layout(info
->array
.level
);
1750 info
->array
.md_minor
= -1;
1751 info
->array
.ctime
= 0;
1752 info
->array
.utime
= 0;
1753 info
->array
.chunk_size
=
1754 __le16_to_cpu(map_to_analyse
->blocks_per_strip
) << 9;
1755 info
->array
.state
= !dev
->vol
.dirty
;
1756 info
->custom_array_size
= __le32_to_cpu(dev
->size_high
);
1757 info
->custom_array_size
<<= 32;
1758 info
->custom_array_size
|= __le32_to_cpu(dev
->size_low
);
1760 info
->new_level
= get_imsm_raid_level(map
);
1761 info
->new_layout
= imsm_level_to_layout(info
->new_level
);
1762 info
->new_chunk
= __le16_to_cpu(map
->blocks_per_strip
) << 9;
1764 info
->new_level
= UnSet
;
1765 info
->new_layout
= UnSet
;
1766 info
->new_chunk
= info
->array
.chunk_size
;
1768 info
->disk
.major
= 0;
1769 info
->disk
.minor
= 0;
1771 info
->disk
.major
= dl
->major
;
1772 info
->disk
.minor
= dl
->minor
;
1775 info
->data_offset
= __le32_to_cpu(map_to_analyse
->pba_of_lba0
);
1776 info
->component_size
=
1777 __le32_to_cpu(map_to_analyse
->blocks_per_member
);
1778 memset(info
->uuid
, 0, sizeof(info
->uuid
));
1779 info
->recovery_start
= MaxSector
;
1780 info
->reshape_active
= (prev_map
!= NULL
) &&
1781 (map
->map_state
== prev_map
->map_state
);
1782 if (info
->reshape_active
)
1783 info
->delta_disks
= map
->num_members
- prev_map
->num_members
;
1785 info
->delta_disks
= 0;
1787 if (map_to_analyse
->map_state
== IMSM_T_STATE_UNINITIALIZED
||
1789 info
->resync_start
= 0;
1790 } else if (dev
->vol
.migr_state
) {
1791 switch (migr_type(dev
)) {
1794 __u64 blocks_per_unit
= blocks_per_migr_unit(dev
);
1795 __u64 units
= __le32_to_cpu(dev
->vol
.curr_migr_unit
);
1797 info
->resync_start
= blocks_per_unit
* units
;
1801 /* we could emulate the checkpointing of
1802 * 'sync_action=check' migrations, but for now
1803 * we just immediately complete them
1806 /* this is handled by container_content_imsm() */
1808 case MIGR_STATE_CHANGE
:
1809 /* FIXME handle other migrations */
1811 /* we are not dirty, so... */
1812 info
->resync_start
= MaxSector
;
1815 info
->resync_start
= MaxSector
;
1817 strncpy(info
->name
, (char *) dev
->volume
, MAX_RAID_SERIAL_LEN
);
1818 info
->name
[MAX_RAID_SERIAL_LEN
] = 0;
1820 info
->array
.major_version
= -1;
1821 info
->array
.minor_version
= -2;
1822 devname
= devnum2devname(st
->container_dev
);
1823 *info
->text_version
= '\0';
1825 sprintf(info
->text_version
, "/%s/%d", devname
, info
->container_member
);
1827 info
->safe_mode_delay
= 4000; /* 4 secs like the Matrix driver */
1828 uuid_from_super_imsm(st
, info
->uuid
);
1832 for (i
=0; i
<map_disks
; i
++) {
1834 if (i
< info
->array
.raid_disks
) {
1835 struct imsm_disk
*dsk
;
1836 j
= get_imsm_disk_idx(dev
, i
, -1);
1837 dsk
= get_imsm_disk(super
, j
);
1838 if (dsk
&& (dsk
->status
& CONFIGURED_DISK
))
1845 static __u8
imsm_check_degraded(struct intel_super
*super
, struct imsm_dev
*dev
, int failed
);
1846 static int imsm_count_failed(struct intel_super
*super
, struct imsm_dev
*dev
);
1848 static struct imsm_disk
*get_imsm_missing(struct intel_super
*super
, __u8 index
)
1852 for (d
= super
->missing
; d
; d
= d
->next
)
1853 if (d
->index
== index
)
1858 static void getinfo_super_imsm(struct supertype
*st
, struct mdinfo
*info
, char *map
)
1860 struct intel_super
*super
= st
->sb
;
1861 struct imsm_disk
*disk
;
1862 int map_disks
= info
->array
.raid_disks
;
1863 int max_enough
= -1;
1865 struct imsm_super
*mpb
;
1867 if (super
->current_vol
>= 0) {
1868 getinfo_super_imsm_volume(st
, info
, map
);
1872 /* Set raid_disks to zero so that Assemble will always pull in valid
1875 info
->array
.raid_disks
= 0;
1876 info
->array
.level
= LEVEL_CONTAINER
;
1877 info
->array
.layout
= 0;
1878 info
->array
.md_minor
= -1;
1879 info
->array
.ctime
= 0; /* N/A for imsm */
1880 info
->array
.utime
= 0;
1881 info
->array
.chunk_size
= 0;
1883 info
->disk
.major
= 0;
1884 info
->disk
.minor
= 0;
1885 info
->disk
.raid_disk
= -1;
1886 info
->reshape_active
= 0;
1887 info
->array
.major_version
= -1;
1888 info
->array
.minor_version
= -2;
1889 strcpy(info
->text_version
, "imsm");
1890 info
->safe_mode_delay
= 0;
1891 info
->disk
.number
= -1;
1892 info
->disk
.state
= 0;
1894 info
->recovery_start
= MaxSector
;
1896 /* do we have the all the insync disks that we expect? */
1897 mpb
= super
->anchor
;
1899 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
1900 struct imsm_dev
*dev
= get_imsm_dev(super
, i
);
1901 int failed
, enough
, j
, missing
= 0;
1902 struct imsm_map
*map
;
1905 failed
= imsm_count_failed(super
, dev
);
1906 state
= imsm_check_degraded(super
, dev
, failed
);
1907 map
= get_imsm_map(dev
, dev
->vol
.migr_state
);
1909 /* any newly missing disks?
1910 * (catches single-degraded vs double-degraded)
1912 for (j
= 0; j
< map
->num_members
; j
++) {
1913 __u32 ord
= get_imsm_ord_tbl_ent(dev
, i
, -1);
1914 __u32 idx
= ord_to_idx(ord
);
1916 if (!(ord
& IMSM_ORD_REBUILD
) &&
1917 get_imsm_missing(super
, idx
)) {
1923 if (state
== IMSM_T_STATE_FAILED
)
1925 else if (state
== IMSM_T_STATE_DEGRADED
&&
1926 (state
!= map
->map_state
|| missing
))
1928 else /* we're normal, or already degraded */
1931 /* in the missing/failed disk case check to see
1932 * if at least one array is runnable
1934 max_enough
= max(max_enough
, enough
);
1936 dprintf("%s: enough: %d\n", __func__
, max_enough
);
1937 info
->container_enough
= max_enough
;
1940 __u32 reserved
= imsm_reserved_sectors(super
, super
->disks
);
1942 disk
= &super
->disks
->disk
;
1943 info
->data_offset
= __le32_to_cpu(disk
->total_blocks
) - reserved
;
1944 info
->component_size
= reserved
;
1945 info
->disk
.state
= is_configured(disk
) ? (1 << MD_DISK_ACTIVE
) : 0;
1946 /* we don't change info->disk.raid_disk here because
1947 * this state will be finalized in mdmon after we have
1948 * found the 'most fresh' version of the metadata
1950 info
->disk
.state
|= is_failed(disk
) ? (1 << MD_DISK_FAULTY
) : 0;
1951 info
->disk
.state
|= is_spare(disk
) ? 0 : (1 << MD_DISK_SYNC
);
1954 /* only call uuid_from_super_imsm when this disk is part of a populated container,
1955 * ->compare_super may have updated the 'num_raid_devs' field for spares
1957 if (info
->disk
.state
& (1 << MD_DISK_SYNC
) || super
->anchor
->num_raid_devs
)
1958 uuid_from_super_imsm(st
, info
->uuid
);
1960 memcpy(info
->uuid
, uuid_zero
, sizeof(uuid_zero
));
1962 /* I don't know how to compute 'map' on imsm, so use safe default */
1965 for (i
= 0; i
< map_disks
; i
++)
1971 /* allocates memory and fills disk in mdinfo structure
1972 * for each disk in array */
1973 struct mdinfo
*getinfo_super_disks_imsm(struct supertype
*st
)
1975 struct mdinfo
*mddev
= NULL
;
1976 struct intel_super
*super
= st
->sb
;
1977 struct imsm_disk
*disk
;
1980 if (!super
|| !super
->disks
)
1983 mddev
= malloc(sizeof(*mddev
));
1985 fprintf(stderr
, Name
": Failed to allocate memory.\n");
1988 memset(mddev
, 0, sizeof(*mddev
));
1992 tmp
= malloc(sizeof(*tmp
));
1994 fprintf(stderr
, Name
": Failed to allocate memory.\n");
1999 memset(tmp
, 0, sizeof(*tmp
));
2001 tmp
->next
= mddev
->devs
;
2003 tmp
->disk
.number
= count
++;
2004 tmp
->disk
.major
= dl
->major
;
2005 tmp
->disk
.minor
= dl
->minor
;
2006 tmp
->disk
.state
= is_configured(disk
) ?
2007 (1 << MD_DISK_ACTIVE
) : 0;
2008 tmp
->disk
.state
|= is_failed(disk
) ? (1 << MD_DISK_FAULTY
) : 0;
2009 tmp
->disk
.state
|= is_spare(disk
) ? 0 : (1 << MD_DISK_SYNC
);
2010 tmp
->disk
.raid_disk
= -1;
2016 static int update_super_imsm(struct supertype
*st
, struct mdinfo
*info
,
2017 char *update
, char *devname
, int verbose
,
2018 int uuid_set
, char *homehost
)
2020 /* For 'assemble' and 'force' we need to return non-zero if any
2021 * change was made. For others, the return value is ignored.
2022 * Update options are:
2023 * force-one : This device looks a bit old but needs to be included,
2024 * update age info appropriately.
2025 * assemble: clear any 'faulty' flag to allow this device to
2027 * force-array: Array is degraded but being forced, mark it clean
2028 * if that will be needed to assemble it.
2030 * newdev: not used ????
2031 * grow: Array has gained a new device - this is currently for
2033 * resync: mark as dirty so a resync will happen.
2034 * name: update the name - preserving the homehost
2035 * uuid: Change the uuid of the array to match watch is given
2037 * Following are not relevant for this imsm:
2038 * sparc2.2 : update from old dodgey metadata
2039 * super-minor: change the preferred_minor number
2040 * summaries: update redundant counters.
2041 * homehost: update the recorded homehost
2042 * _reshape_progress: record new reshape_progress position.
2045 struct intel_super
*super
= st
->sb
;
2046 struct imsm_super
*mpb
;
2048 /* we can only update container info */
2049 if (!super
|| super
->current_vol
>= 0 || !super
->anchor
)
2052 mpb
= super
->anchor
;
2054 if (strcmp(update
, "uuid") == 0 && uuid_set
&& !info
->update_private
)
2056 else if (strcmp(update
, "uuid") == 0 && uuid_set
&& info
->update_private
) {
2057 mpb
->orig_family_num
= *((__u32
*) info
->update_private
);
2059 } else if (strcmp(update
, "uuid") == 0) {
2060 __u32
*new_family
= malloc(sizeof(*new_family
));
2062 /* update orig_family_number with the incoming random
2063 * data, report the new effective uuid, and store the
2064 * new orig_family_num for future updates.
2067 memcpy(&mpb
->orig_family_num
, info
->uuid
, sizeof(__u32
));
2068 uuid_from_super_imsm(st
, info
->uuid
);
2069 *new_family
= mpb
->orig_family_num
;
2070 info
->update_private
= new_family
;
2073 } else if (strcmp(update
, "assemble") == 0)
2078 /* successful update? recompute checksum */
2080 mpb
->check_sum
= __le32_to_cpu(__gen_imsm_checksum(mpb
));
2085 static size_t disks_to_mpb_size(int disks
)
2089 size
= sizeof(struct imsm_super
);
2090 size
+= (disks
- 1) * sizeof(struct imsm_disk
);
2091 size
+= 2 * sizeof(struct imsm_dev
);
2092 /* up to 2 maps per raid device (-2 for imsm_maps in imsm_dev */
2093 size
+= (4 - 2) * sizeof(struct imsm_map
);
2094 /* 4 possible disk_ord_tbl's */
2095 size
+= 4 * (disks
- 1) * sizeof(__u32
);
2100 static __u64
avail_size_imsm(struct supertype
*st
, __u64 devsize
)
2102 if (devsize
< (MPB_SECTOR_CNT
+ IMSM_RESERVED_SECTORS
))
2105 return devsize
- (MPB_SECTOR_CNT
+ IMSM_RESERVED_SECTORS
);
2108 static void free_devlist(struct intel_super
*super
)
2110 struct intel_dev
*dv
;
2112 while (super
->devlist
) {
2113 dv
= super
->devlist
->next
;
2114 free(super
->devlist
->dev
);
2115 free(super
->devlist
);
2116 super
->devlist
= dv
;
2120 static void imsm_copy_dev(struct imsm_dev
*dest
, struct imsm_dev
*src
)
2122 memcpy(dest
, src
, sizeof_imsm_dev(src
, 0));
2125 static int compare_super_imsm(struct supertype
*st
, struct supertype
*tst
)
2129 * 0 same, or first was empty, and second was copied
2130 * 1 second had wrong number
2132 * 3 wrong other info
2134 struct intel_super
*first
= st
->sb
;
2135 struct intel_super
*sec
= tst
->sb
;
2143 /* if an anchor does not have num_raid_devs set then it is a free
2146 if (first
->anchor
->num_raid_devs
> 0 &&
2147 sec
->anchor
->num_raid_devs
> 0) {
2148 /* Determine if these disks might ever have been
2149 * related. Further disambiguation can only take place
2150 * in load_super_imsm_all
2152 __u32 first_family
= first
->anchor
->orig_family_num
;
2153 __u32 sec_family
= sec
->anchor
->orig_family_num
;
2155 if (memcmp(first
->anchor
->sig
, sec
->anchor
->sig
,
2156 MAX_SIGNATURE_LENGTH
) != 0)
2159 if (first_family
== 0)
2160 first_family
= first
->anchor
->family_num
;
2161 if (sec_family
== 0)
2162 sec_family
= sec
->anchor
->family_num
;
2164 if (first_family
!= sec_family
)
2170 /* if 'first' is a spare promote it to a populated mpb with sec's
2173 if (first
->anchor
->num_raid_devs
== 0 &&
2174 sec
->anchor
->num_raid_devs
> 0) {
2176 struct intel_dev
*dv
;
2177 struct imsm_dev
*dev
;
2179 /* we need to copy raid device info from sec if an allocation
2180 * fails here we don't associate the spare
2182 for (i
= 0; i
< sec
->anchor
->num_raid_devs
; i
++) {
2183 dv
= malloc(sizeof(*dv
));
2186 dev
= malloc(sizeof_imsm_dev(get_imsm_dev(sec
, i
), 1));
2193 dv
->next
= first
->devlist
;
2194 first
->devlist
= dv
;
2196 if (i
< sec
->anchor
->num_raid_devs
) {
2197 /* allocation failure */
2198 free_devlist(first
);
2199 fprintf(stderr
, "imsm: failed to associate spare\n");
2202 first
->anchor
->num_raid_devs
= sec
->anchor
->num_raid_devs
;
2203 first
->anchor
->orig_family_num
= sec
->anchor
->orig_family_num
;
2204 first
->anchor
->family_num
= sec
->anchor
->family_num
;
2205 memcpy(first
->anchor
->sig
, sec
->anchor
->sig
, MAX_SIGNATURE_LENGTH
);
2206 for (i
= 0; i
< sec
->anchor
->num_raid_devs
; i
++)
2207 imsm_copy_dev(get_imsm_dev(first
, i
), get_imsm_dev(sec
, i
));
2213 static void fd2devname(int fd
, char *name
)
2217 char dname
[PATH_MAX
];
2222 if (fstat(fd
, &st
) != 0)
2224 sprintf(path
, "/sys/dev/block/%d:%d",
2225 major(st
.st_rdev
), minor(st
.st_rdev
));
2227 rv
= readlink(path
, dname
, sizeof(dname
));
2232 nm
= strrchr(dname
, '/');
2234 snprintf(name
, MAX_RAID_SERIAL_LEN
, "/dev/%s", nm
);
2237 extern int scsi_get_serial(int fd
, void *buf
, size_t buf_len
);
2239 static int imsm_read_serial(int fd
, char *devname
,
2240 __u8 serial
[MAX_RAID_SERIAL_LEN
])
2242 unsigned char scsi_serial
[255];
2251 memset(scsi_serial
, 0, sizeof(scsi_serial
));
2253 rv
= scsi_get_serial(fd
, scsi_serial
, sizeof(scsi_serial
));
2255 if (rv
&& check_env("IMSM_DEVNAME_AS_SERIAL")) {
2256 memset(serial
, 0, MAX_RAID_SERIAL_LEN
);
2257 fd2devname(fd
, (char *) serial
);
2264 Name
": Failed to retrieve serial for %s\n",
2269 rsp_len
= scsi_serial
[3];
2273 Name
": Failed to retrieve serial for %s\n",
2277 rsp_buf
= (char *) &scsi_serial
[4];
2279 /* trim all whitespace and non-printable characters and convert
2282 for (i
= 0, dest
= rsp_buf
; i
< rsp_len
; i
++) {
2285 /* ':' is reserved for use in placeholder serial
2286 * numbers for missing disks
2294 len
= dest
- rsp_buf
;
2297 /* truncate leading characters */
2298 if (len
> MAX_RAID_SERIAL_LEN
) {
2299 dest
+= len
- MAX_RAID_SERIAL_LEN
;
2300 len
= MAX_RAID_SERIAL_LEN
;
2303 memset(serial
, 0, MAX_RAID_SERIAL_LEN
);
2304 memcpy(serial
, dest
, len
);
2309 static int serialcmp(__u8
*s1
, __u8
*s2
)
2311 return strncmp((char *) s1
, (char *) s2
, MAX_RAID_SERIAL_LEN
);
2314 static void serialcpy(__u8
*dest
, __u8
*src
)
2316 strncpy((char *) dest
, (char *) src
, MAX_RAID_SERIAL_LEN
);
2320 static struct dl
*serial_to_dl(__u8
*serial
, struct intel_super
*super
)
2324 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
2325 if (serialcmp(dl
->serial
, serial
) == 0)
2332 static struct imsm_disk
*
2333 __serial_to_disk(__u8
*serial
, struct imsm_super
*mpb
, int *idx
)
2337 for (i
= 0; i
< mpb
->num_disks
; i
++) {
2338 struct imsm_disk
*disk
= __get_imsm_disk(mpb
, i
);
2340 if (serialcmp(disk
->serial
, serial
) == 0) {
2351 load_imsm_disk(int fd
, struct intel_super
*super
, char *devname
, int keep_fd
)
2353 struct imsm_disk
*disk
;
2358 __u8 serial
[MAX_RAID_SERIAL_LEN
];
2360 rv
= imsm_read_serial(fd
, devname
, serial
);
2365 dl
= calloc(1, sizeof(*dl
));
2369 Name
": failed to allocate disk buffer for %s\n",
2375 dl
->major
= major(stb
.st_rdev
);
2376 dl
->minor
= minor(stb
.st_rdev
);
2377 dl
->next
= super
->disks
;
2378 dl
->fd
= keep_fd
? fd
: -1;
2379 assert(super
->disks
== NULL
);
2381 serialcpy(dl
->serial
, serial
);
2384 fd2devname(fd
, name
);
2386 dl
->devname
= strdup(devname
);
2388 dl
->devname
= strdup(name
);
2390 /* look up this disk's index in the current anchor */
2391 disk
= __serial_to_disk(dl
->serial
, super
->anchor
, &dl
->index
);
2394 /* only set index on disks that are a member of a
2395 * populated contianer, i.e. one with raid_devs
2397 if (is_failed(&dl
->disk
))
2399 else if (is_spare(&dl
->disk
))
2407 /* When migrating map0 contains the 'destination' state while map1
2408 * contains the current state. When not migrating map0 contains the
2409 * current state. This routine assumes that map[0].map_state is set to
2410 * the current array state before being called.
2412 * Migration is indicated by one of the following states
2413 * 1/ Idle (migr_state=0 map0state=normal||unitialized||degraded||failed)
2414 * 2/ Initialize (migr_state=1 migr_type=MIGR_INIT map0state=normal
2415 * map1state=unitialized)
2416 * 3/ Repair (Resync) (migr_state=1 migr_type=MIGR_REPAIR map0state=normal
2418 * 4/ Rebuild (migr_state=1 migr_type=MIGR_REBUILD map0state=normal
2419 * map1state=degraded)
2421 static void migrate(struct imsm_dev
*dev
, __u8 to_state
, int migr_type
)
2423 struct imsm_map
*dest
;
2424 struct imsm_map
*src
= get_imsm_map(dev
, 0);
2426 dev
->vol
.migr_state
= 1;
2427 set_migr_type(dev
, migr_type
);
2428 dev
->vol
.curr_migr_unit
= 0;
2429 dest
= get_imsm_map(dev
, 1);
2431 /* duplicate and then set the target end state in map[0] */
2432 memcpy(dest
, src
, sizeof_imsm_map(src
));
2433 if ((migr_type
== MIGR_REBUILD
) ||
2434 (migr_type
== MIGR_GEN_MIGR
)) {
2438 for (i
= 0; i
< src
->num_members
; i
++) {
2439 ord
= __le32_to_cpu(src
->disk_ord_tbl
[i
]);
2440 set_imsm_ord_tbl_ent(src
, i
, ord_to_idx(ord
));
2444 src
->map_state
= to_state
;
2447 static void end_migration(struct imsm_dev
*dev
, __u8 map_state
)
2449 struct imsm_map
*map
= get_imsm_map(dev
, 0);
2450 struct imsm_map
*prev
= get_imsm_map(dev
, dev
->vol
.migr_state
);
2453 /* merge any IMSM_ORD_REBUILD bits that were not successfully
2454 * completed in the last migration.
2456 * FIXME add support for raid-level-migration
2458 for (i
= 0; i
< prev
->num_members
; i
++)
2459 for (j
= 0; j
< map
->num_members
; j
++)
2460 /* during online capacity expansion
2461 * disks position can be changed if takeover is used
2463 if (ord_to_idx(map
->disk_ord_tbl
[j
]) ==
2464 ord_to_idx(prev
->disk_ord_tbl
[i
])) {
2465 map
->disk_ord_tbl
[j
] |= prev
->disk_ord_tbl
[i
];
2469 dev
->vol
.migr_state
= 0;
2470 dev
->vol
.migr_type
= 0;
2471 dev
->vol
.curr_migr_unit
= 0;
2472 map
->map_state
= map_state
;
2476 static int parse_raid_devices(struct intel_super
*super
)
2479 struct imsm_dev
*dev_new
;
2480 size_t len
, len_migr
;
2482 size_t space_needed
= 0;
2483 struct imsm_super
*mpb
= super
->anchor
;
2485 for (i
= 0; i
< super
->anchor
->num_raid_devs
; i
++) {
2486 struct imsm_dev
*dev_iter
= __get_imsm_dev(super
->anchor
, i
);
2487 struct intel_dev
*dv
;
2489 len
= sizeof_imsm_dev(dev_iter
, 0);
2490 len_migr
= sizeof_imsm_dev(dev_iter
, 1);
2492 space_needed
+= len_migr
- len
;
2494 dv
= malloc(sizeof(*dv
));
2497 if (max_len
< len_migr
)
2499 if (max_len
> len_migr
)
2500 space_needed
+= max_len
- len_migr
;
2501 dev_new
= malloc(max_len
);
2506 imsm_copy_dev(dev_new
, dev_iter
);
2509 dv
->next
= super
->devlist
;
2510 super
->devlist
= dv
;
2513 /* ensure that super->buf is large enough when all raid devices
2516 if (__le32_to_cpu(mpb
->mpb_size
) + space_needed
> super
->len
) {
2519 len
= ROUND_UP(__le32_to_cpu(mpb
->mpb_size
) + space_needed
, 512);
2520 if (posix_memalign(&buf
, 512, len
) != 0)
2523 memcpy(buf
, super
->buf
, super
->len
);
2524 memset(buf
+ super
->len
, 0, len
- super
->len
);
2533 /* retrieve a pointer to the bbm log which starts after all raid devices */
2534 struct bbm_log
*__get_imsm_bbm_log(struct imsm_super
*mpb
)
2538 if (__le32_to_cpu(mpb
->bbm_log_size
)) {
2540 ptr
+= mpb
->mpb_size
- __le32_to_cpu(mpb
->bbm_log_size
);
2546 static void __free_imsm(struct intel_super
*super
, int free_disks
);
2548 /* load_imsm_mpb - read matrix metadata
2549 * allocates super->mpb to be freed by free_super
2551 static int load_imsm_mpb(int fd
, struct intel_super
*super
, char *devname
)
2553 unsigned long long dsize
;
2554 unsigned long long sectors
;
2556 struct imsm_super
*anchor
;
2559 get_dev_size(fd
, NULL
, &dsize
);
2563 Name
": %s: device to small for imsm\n",
2568 if (lseek64(fd
, dsize
- (512 * 2), SEEK_SET
) < 0) {
2571 Name
": Cannot seek to anchor block on %s: %s\n",
2572 devname
, strerror(errno
));
2576 if (posix_memalign((void**)&anchor
, 512, 512) != 0) {
2579 Name
": Failed to allocate imsm anchor buffer"
2580 " on %s\n", devname
);
2583 if (read(fd
, anchor
, 512) != 512) {
2586 Name
": Cannot read anchor block on %s: %s\n",
2587 devname
, strerror(errno
));
2592 if (strncmp((char *) anchor
->sig
, MPB_SIGNATURE
, MPB_SIG_LEN
) != 0) {
2595 Name
": no IMSM anchor on %s\n", devname
);
2600 __free_imsm(super
, 0);
2601 super
->len
= ROUND_UP(anchor
->mpb_size
, 512);
2602 if (posix_memalign(&super
->buf
, 512, super
->len
) != 0) {
2605 Name
": unable to allocate %zu byte mpb buffer\n",
2610 memcpy(super
->buf
, anchor
, 512);
2612 sectors
= mpb_sectors(anchor
) - 1;
2615 check_sum
= __gen_imsm_checksum(super
->anchor
);
2616 if (check_sum
!= __le32_to_cpu(super
->anchor
->check_sum
)) {
2619 Name
": IMSM checksum %x != %x on %s\n",
2621 __le32_to_cpu(super
->anchor
->check_sum
),
2629 /* read the extended mpb */
2630 if (lseek64(fd
, dsize
- (512 * (2 + sectors
)), SEEK_SET
) < 0) {
2633 Name
": Cannot seek to extended mpb on %s: %s\n",
2634 devname
, strerror(errno
));
2638 if ((unsigned)read(fd
, super
->buf
+ 512, super
->len
- 512) != super
->len
- 512) {
2641 Name
": Cannot read extended mpb on %s: %s\n",
2642 devname
, strerror(errno
));
2646 check_sum
= __gen_imsm_checksum(super
->anchor
);
2647 if (check_sum
!= __le32_to_cpu(super
->anchor
->check_sum
)) {
2650 Name
": IMSM checksum %x != %x on %s\n",
2651 check_sum
, __le32_to_cpu(super
->anchor
->check_sum
),
2656 /* FIXME the BBM log is disk specific so we cannot use this global
2657 * buffer for all disks. Ok for now since we only look at the global
2658 * bbm_log_size parameter to gate assembly
2660 super
->bbm_log
= __get_imsm_bbm_log(super
->anchor
);
2666 load_and_parse_mpb(int fd
, struct intel_super
*super
, char *devname
, int keep_fd
)
2670 err
= load_imsm_mpb(fd
, super
, devname
);
2673 err
= load_imsm_disk(fd
, super
, devname
, keep_fd
);
2676 err
= parse_raid_devices(super
);
2681 static void __free_imsm_disk(struct dl
*d
)
2693 static void free_imsm_disks(struct intel_super
*super
)
2697 while (super
->disks
) {
2699 super
->disks
= d
->next
;
2700 __free_imsm_disk(d
);
2702 while (super
->disk_mgmt_list
) {
2703 d
= super
->disk_mgmt_list
;
2704 super
->disk_mgmt_list
= d
->next
;
2705 __free_imsm_disk(d
);
2707 while (super
->missing
) {
2709 super
->missing
= d
->next
;
2710 __free_imsm_disk(d
);
2715 /* free all the pieces hanging off of a super pointer */
2716 static void __free_imsm(struct intel_super
*super
, int free_disks
)
2718 struct intel_hba
*elem
, *next
;
2725 free_imsm_disks(super
);
2726 free_devlist(super
);
2730 free((void *)elem
->path
);
2738 static void free_imsm(struct intel_super
*super
)
2740 __free_imsm(super
, 1);
2744 static void free_super_imsm(struct supertype
*st
)
2746 struct intel_super
*super
= st
->sb
;
2755 static struct intel_super
*alloc_super(void)
2757 struct intel_super
*super
= malloc(sizeof(*super
));
2760 memset(super
, 0, sizeof(*super
));
2761 super
->current_vol
= -1;
2762 super
->create_offset
= ~((__u32
) 0);
2763 if (!check_env("IMSM_NO_PLATFORM"))
2764 super
->orom
= find_imsm_orom();
2771 /* find_missing - helper routine for load_super_imsm_all that identifies
2772 * disks that have disappeared from the system. This routine relies on
2773 * the mpb being uptodate, which it is at load time.
2775 static int find_missing(struct intel_super
*super
)
2778 struct imsm_super
*mpb
= super
->anchor
;
2780 struct imsm_disk
*disk
;
2782 for (i
= 0; i
< mpb
->num_disks
; i
++) {
2783 disk
= __get_imsm_disk(mpb
, i
);
2784 dl
= serial_to_dl(disk
->serial
, super
);
2788 dl
= malloc(sizeof(*dl
));
2794 dl
->devname
= strdup("missing");
2796 serialcpy(dl
->serial
, disk
->serial
);
2799 dl
->next
= super
->missing
;
2800 super
->missing
= dl
;
2806 static struct intel_disk
*disk_list_get(__u8
*serial
, struct intel_disk
*disk_list
)
2808 struct intel_disk
*idisk
= disk_list
;
2811 if (serialcmp(idisk
->disk
.serial
, serial
) == 0)
2813 idisk
= idisk
->next
;
2819 static int __prep_thunderdome(struct intel_super
**table
, int tbl_size
,
2820 struct intel_super
*super
,
2821 struct intel_disk
**disk_list
)
2823 struct imsm_disk
*d
= &super
->disks
->disk
;
2824 struct imsm_super
*mpb
= super
->anchor
;
2827 for (i
= 0; i
< tbl_size
; i
++) {
2828 struct imsm_super
*tbl_mpb
= table
[i
]->anchor
;
2829 struct imsm_disk
*tbl_d
= &table
[i
]->disks
->disk
;
2831 if (tbl_mpb
->family_num
== mpb
->family_num
) {
2832 if (tbl_mpb
->check_sum
== mpb
->check_sum
) {
2833 dprintf("%s: mpb from %d:%d matches %d:%d\n",
2834 __func__
, super
->disks
->major
,
2835 super
->disks
->minor
,
2836 table
[i
]->disks
->major
,
2837 table
[i
]->disks
->minor
);
2841 if (((is_configured(d
) && !is_configured(tbl_d
)) ||
2842 is_configured(d
) == is_configured(tbl_d
)) &&
2843 tbl_mpb
->generation_num
< mpb
->generation_num
) {
2844 /* current version of the mpb is a
2845 * better candidate than the one in
2846 * super_table, but copy over "cross
2847 * generational" status
2849 struct intel_disk
*idisk
;
2851 dprintf("%s: mpb from %d:%d replaces %d:%d\n",
2852 __func__
, super
->disks
->major
,
2853 super
->disks
->minor
,
2854 table
[i
]->disks
->major
,
2855 table
[i
]->disks
->minor
);
2857 idisk
= disk_list_get(tbl_d
->serial
, *disk_list
);
2858 if (idisk
&& is_failed(&idisk
->disk
))
2859 tbl_d
->status
|= FAILED_DISK
;
2862 struct intel_disk
*idisk
;
2863 struct imsm_disk
*disk
;
2865 /* tbl_mpb is more up to date, but copy
2866 * over cross generational status before
2869 disk
= __serial_to_disk(d
->serial
, mpb
, NULL
);
2870 if (disk
&& is_failed(disk
))
2871 d
->status
|= FAILED_DISK
;
2873 idisk
= disk_list_get(d
->serial
, *disk_list
);
2876 if (disk
&& is_configured(disk
))
2877 idisk
->disk
.status
|= CONFIGURED_DISK
;
2880 dprintf("%s: mpb from %d:%d prefer %d:%d\n",
2881 __func__
, super
->disks
->major
,
2882 super
->disks
->minor
,
2883 table
[i
]->disks
->major
,
2884 table
[i
]->disks
->minor
);
2892 table
[tbl_size
++] = super
;
2896 /* update/extend the merged list of imsm_disk records */
2897 for (j
= 0; j
< mpb
->num_disks
; j
++) {
2898 struct imsm_disk
*disk
= __get_imsm_disk(mpb
, j
);
2899 struct intel_disk
*idisk
;
2901 idisk
= disk_list_get(disk
->serial
, *disk_list
);
2903 idisk
->disk
.status
|= disk
->status
;
2904 if (is_configured(&idisk
->disk
) ||
2905 is_failed(&idisk
->disk
))
2906 idisk
->disk
.status
&= ~(SPARE_DISK
);
2908 idisk
= calloc(1, sizeof(*idisk
));
2911 idisk
->owner
= IMSM_UNKNOWN_OWNER
;
2912 idisk
->disk
= *disk
;
2913 idisk
->next
= *disk_list
;
2917 if (serialcmp(idisk
->disk
.serial
, d
->serial
) == 0)
2924 static struct intel_super
*
2925 validate_members(struct intel_super
*super
, struct intel_disk
*disk_list
,
2928 struct imsm_super
*mpb
= super
->anchor
;
2932 for (i
= 0; i
< mpb
->num_disks
; i
++) {
2933 struct imsm_disk
*disk
= __get_imsm_disk(mpb
, i
);
2934 struct intel_disk
*idisk
;
2936 idisk
= disk_list_get(disk
->serial
, disk_list
);
2938 if (idisk
->owner
== owner
||
2939 idisk
->owner
== IMSM_UNKNOWN_OWNER
)
2942 dprintf("%s: '%.16s' owner %d != %d\n",
2943 __func__
, disk
->serial
, idisk
->owner
,
2946 dprintf("%s: unknown disk %x [%d]: %.16s\n",
2947 __func__
, __le32_to_cpu(mpb
->family_num
), i
,
2953 if (ok_count
== mpb
->num_disks
)
2958 static void show_conflicts(__u32 family_num
, struct intel_super
*super_list
)
2960 struct intel_super
*s
;
2962 for (s
= super_list
; s
; s
= s
->next
) {
2963 if (family_num
!= s
->anchor
->family_num
)
2965 fprintf(stderr
, "Conflict, offlining family %#x on '%s'\n",
2966 __le32_to_cpu(family_num
), s
->disks
->devname
);
2970 static struct intel_super
*
2971 imsm_thunderdome(struct intel_super
**super_list
, int len
)
2973 struct intel_super
*super_table
[len
];
2974 struct intel_disk
*disk_list
= NULL
;
2975 struct intel_super
*champion
, *spare
;
2976 struct intel_super
*s
, **del
;
2981 memset(super_table
, 0, sizeof(super_table
));
2982 for (s
= *super_list
; s
; s
= s
->next
)
2983 tbl_size
= __prep_thunderdome(super_table
, tbl_size
, s
, &disk_list
);
2985 for (i
= 0; i
< tbl_size
; i
++) {
2986 struct imsm_disk
*d
;
2987 struct intel_disk
*idisk
;
2988 struct imsm_super
*mpb
= super_table
[i
]->anchor
;
2991 d
= &s
->disks
->disk
;
2993 /* 'd' must appear in merged disk list for its
2994 * configuration to be valid
2996 idisk
= disk_list_get(d
->serial
, disk_list
);
2997 if (idisk
&& idisk
->owner
== i
)
2998 s
= validate_members(s
, disk_list
, i
);
3003 dprintf("%s: marking family: %#x from %d:%d offline\n",
3004 __func__
, mpb
->family_num
,
3005 super_table
[i
]->disks
->major
,
3006 super_table
[i
]->disks
->minor
);
3010 /* This is where the mdadm implementation differs from the Windows
3011 * driver which has no strict concept of a container. We can only
3012 * assemble one family from a container, so when returning a prodigal
3013 * array member to this system the code will not be able to disambiguate
3014 * the container contents that should be assembled ("foreign" versus
3015 * "local"). It requires user intervention to set the orig_family_num
3016 * to a new value to establish a new container. The Windows driver in
3017 * this situation fixes up the volume name in place and manages the
3018 * foreign array as an independent entity.
3023 for (i
= 0; i
< tbl_size
; i
++) {
3024 struct intel_super
*tbl_ent
= super_table
[i
];
3030 if (tbl_ent
->anchor
->num_raid_devs
== 0) {
3035 if (s
&& !is_spare
) {
3036 show_conflicts(tbl_ent
->anchor
->family_num
, *super_list
);
3038 } else if (!s
&& !is_spare
)
3051 fprintf(stderr
, "Chose family %#x on '%s', "
3052 "assemble conflicts to new container with '--update=uuid'\n",
3053 __le32_to_cpu(s
->anchor
->family_num
), s
->disks
->devname
);
3055 /* collect all dl's onto 'champion', and update them to
3056 * champion's version of the status
3058 for (s
= *super_list
; s
; s
= s
->next
) {
3059 struct imsm_super
*mpb
= champion
->anchor
;
3060 struct dl
*dl
= s
->disks
;
3065 for (i
= 0; i
< mpb
->num_disks
; i
++) {
3066 struct imsm_disk
*disk
;
3068 disk
= __serial_to_disk(dl
->serial
, mpb
, &dl
->index
);
3071 /* only set index on disks that are a member of
3072 * a populated contianer, i.e. one with
3075 if (is_failed(&dl
->disk
))
3077 else if (is_spare(&dl
->disk
))
3083 if (i
>= mpb
->num_disks
) {
3084 struct intel_disk
*idisk
;
3086 idisk
= disk_list_get(dl
->serial
, disk_list
);
3087 if (idisk
&& is_spare(&idisk
->disk
) &&
3088 !is_failed(&idisk
->disk
) && !is_configured(&idisk
->disk
))
3096 dl
->next
= champion
->disks
;
3097 champion
->disks
= dl
;
3101 /* delete 'champion' from super_list */
3102 for (del
= super_list
; *del
; ) {
3103 if (*del
== champion
) {
3104 *del
= (*del
)->next
;
3107 del
= &(*del
)->next
;
3109 champion
->next
= NULL
;
3113 struct intel_disk
*idisk
= disk_list
;
3115 disk_list
= disk_list
->next
;
3122 static int load_super_imsm_all(struct supertype
*st
, int fd
, void **sbp
,
3126 struct intel_super
*super_list
= NULL
;
3127 struct intel_super
*super
= NULL
;
3128 int devnum
= fd2devnum(fd
);
3134 /* check if 'fd' an opened container */
3135 sra
= sysfs_read(fd
, 0, GET_LEVEL
|GET_VERSION
|GET_DEVS
|GET_STATE
);
3139 if (sra
->array
.major_version
!= -1 ||
3140 sra
->array
.minor_version
!= -2 ||
3141 strcmp(sra
->text_version
, "imsm") != 0) {
3146 for (sd
= sra
->devs
, i
= 0; sd
; sd
= sd
->next
, i
++) {
3147 struct intel_super
*s
= alloc_super();
3154 s
->next
= super_list
;
3158 sprintf(nm
, "%d:%d", sd
->disk
.major
, sd
->disk
.minor
);
3159 dfd
= dev_open(nm
, O_RDWR
);
3163 err
= load_and_parse_mpb(dfd
, s
, NULL
, 1);
3165 /* retry the load if we might have raced against mdmon */
3166 if (err
== 3 && mdmon_running(devnum
))
3167 for (retry
= 0; retry
< 3; retry
++) {
3169 err
= load_and_parse_mpb(dfd
, s
, NULL
, 1);
3177 /* all mpbs enter, maybe one leaves */
3178 super
= imsm_thunderdome(&super_list
, i
);
3184 if (find_missing(super
) != 0) {
3192 while (super_list
) {
3193 struct intel_super
*s
= super_list
;
3195 super_list
= super_list
->next
;
3204 st
->container_dev
= devnum
;
3205 if (err
== 0 && st
->ss
== NULL
) {
3206 st
->ss
= &super_imsm
;
3207 st
->minor_version
= 0;
3208 st
->max_devs
= IMSM_MAX_DEVICES
;
3213 static int load_container_imsm(struct supertype
*st
, int fd
, char *devname
)
3215 return load_super_imsm_all(st
, fd
, &st
->sb
, devname
);
3219 static int load_super_imsm(struct supertype
*st
, int fd
, char *devname
)
3221 struct intel_super
*super
;
3224 if (test_partition(fd
))
3225 /* IMSM not allowed on partitions */
3228 free_super_imsm(st
);
3230 super
= alloc_super();
3233 Name
": malloc of %zu failed.\n",
3238 rv
= load_and_parse_mpb(fd
, super
, devname
, 0);
3243 Name
": Failed to load all information "
3244 "sections on %s\n", devname
);
3250 if (st
->ss
== NULL
) {
3251 st
->ss
= &super_imsm
;
3252 st
->minor_version
= 0;
3253 st
->max_devs
= IMSM_MAX_DEVICES
;
3258 static __u16
info_to_blocks_per_strip(mdu_array_info_t
*info
)
3260 if (info
->level
== 1)
3262 return info
->chunk_size
>> 9;
3265 static __u32
info_to_num_data_stripes(mdu_array_info_t
*info
, int num_domains
)
3269 num_stripes
= (info
->size
* 2) / info_to_blocks_per_strip(info
);
3270 num_stripes
/= num_domains
;
3275 static __u32
info_to_blocks_per_member(mdu_array_info_t
*info
)
3277 if (info
->level
== 1)
3278 return info
->size
* 2;
3280 return (info
->size
* 2) & ~(info_to_blocks_per_strip(info
) - 1);
3283 static void imsm_update_version_info(struct intel_super
*super
)
3285 /* update the version and attributes */
3286 struct imsm_super
*mpb
= super
->anchor
;
3288 struct imsm_dev
*dev
;
3289 struct imsm_map
*map
;
3292 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
3293 dev
= get_imsm_dev(super
, i
);
3294 map
= get_imsm_map(dev
, 0);
3295 if (__le32_to_cpu(dev
->size_high
) > 0)
3296 mpb
->attributes
|= MPB_ATTRIB_2TB
;
3298 /* FIXME detect when an array spans a port multiplier */
3300 mpb
->attributes
|= MPB_ATTRIB_PM
;
3303 if (mpb
->num_raid_devs
> 1 ||
3304 mpb
->attributes
!= MPB_ATTRIB_CHECKSUM_VERIFY
) {
3305 version
= MPB_VERSION_ATTRIBS
;
3306 switch (get_imsm_raid_level(map
)) {
3307 case 0: mpb
->attributes
|= MPB_ATTRIB_RAID0
; break;
3308 case 1: mpb
->attributes
|= MPB_ATTRIB_RAID1
; break;
3309 case 10: mpb
->attributes
|= MPB_ATTRIB_RAID10
; break;
3310 case 5: mpb
->attributes
|= MPB_ATTRIB_RAID5
; break;
3313 if (map
->num_members
>= 5)
3314 version
= MPB_VERSION_5OR6_DISK_ARRAY
;
3315 else if (dev
->status
== DEV_CLONE_N_GO
)
3316 version
= MPB_VERSION_CNG
;
3317 else if (get_imsm_raid_level(map
) == 5)
3318 version
= MPB_VERSION_RAID5
;
3319 else if (map
->num_members
>= 3)
3320 version
= MPB_VERSION_3OR4_DISK_ARRAY
;
3321 else if (get_imsm_raid_level(map
) == 1)
3322 version
= MPB_VERSION_RAID1
;
3324 version
= MPB_VERSION_RAID0
;
3326 strcpy(((char *) mpb
->sig
) + strlen(MPB_SIGNATURE
), version
);
3330 static int check_name(struct intel_super
*super
, char *name
, int quiet
)
3332 struct imsm_super
*mpb
= super
->anchor
;
3333 char *reason
= NULL
;
3336 if (strlen(name
) > MAX_RAID_SERIAL_LEN
)
3337 reason
= "must be 16 characters or less";
3339 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
3340 struct imsm_dev
*dev
= get_imsm_dev(super
, i
);
3342 if (strncmp((char *) dev
->volume
, name
, MAX_RAID_SERIAL_LEN
) == 0) {
3343 reason
= "already exists";
3348 if (reason
&& !quiet
)
3349 fprintf(stderr
, Name
": imsm volume name %s\n", reason
);
3354 static int init_super_imsm_volume(struct supertype
*st
, mdu_array_info_t
*info
,
3355 unsigned long long size
, char *name
,
3356 char *homehost
, int *uuid
)
3358 /* We are creating a volume inside a pre-existing container.
3359 * so st->sb is already set.
3361 struct intel_super
*super
= st
->sb
;
3362 struct imsm_super
*mpb
= super
->anchor
;
3363 struct intel_dev
*dv
;
3364 struct imsm_dev
*dev
;
3365 struct imsm_vol
*vol
;
3366 struct imsm_map
*map
;
3367 int idx
= mpb
->num_raid_devs
;
3369 unsigned long long array_blocks
;
3370 size_t size_old
, size_new
;
3371 __u32 num_data_stripes
;
3373 if (super
->orom
&& mpb
->num_raid_devs
>= super
->orom
->vpa
) {
3374 fprintf(stderr
, Name
": This imsm-container already has the "
3375 "maximum of %d volumes\n", super
->orom
->vpa
);
3379 /* ensure the mpb is large enough for the new data */
3380 size_old
= __le32_to_cpu(mpb
->mpb_size
);
3381 size_new
= disks_to_mpb_size(info
->nr_disks
);
3382 if (size_new
> size_old
) {
3384 size_t size_round
= ROUND_UP(size_new
, 512);
3386 if (posix_memalign(&mpb_new
, 512, size_round
) != 0) {
3387 fprintf(stderr
, Name
": could not allocate new mpb\n");
3390 memcpy(mpb_new
, mpb
, size_old
);
3393 super
->anchor
= mpb_new
;
3394 mpb
->mpb_size
= __cpu_to_le32(size_new
);
3395 memset(mpb_new
+ size_old
, 0, size_round
- size_old
);
3397 super
->current_vol
= idx
;
3398 /* when creating the first raid device in this container set num_disks
3399 * to zero, i.e. delete this spare and add raid member devices in
3400 * add_to_super_imsm_volume()
3402 if (super
->current_vol
== 0)
3405 if (!check_name(super
, name
, 0))
3407 dv
= malloc(sizeof(*dv
));
3409 fprintf(stderr
, Name
": failed to allocate device list entry\n");
3412 dev
= malloc(sizeof(*dev
) + sizeof(__u32
) * (info
->raid_disks
- 1));
3415 fprintf(stderr
, Name
": could not allocate raid device\n");
3418 strncpy((char *) dev
->volume
, name
, MAX_RAID_SERIAL_LEN
);
3419 if (info
->level
== 1)
3420 array_blocks
= info_to_blocks_per_member(info
);
3422 array_blocks
= calc_array_size(info
->level
, info
->raid_disks
,
3423 info
->layout
, info
->chunk_size
,
3425 /* round array size down to closest MB */
3426 array_blocks
= (array_blocks
>> SECT_PER_MB_SHIFT
) << SECT_PER_MB_SHIFT
;
3428 dev
->size_low
= __cpu_to_le32((__u32
) array_blocks
);
3429 dev
->size_high
= __cpu_to_le32((__u32
) (array_blocks
>> 32));
3430 dev
->status
= __cpu_to_le32(0);
3431 dev
->reserved_blocks
= __cpu_to_le32(0);
3433 vol
->migr_state
= 0;
3434 set_migr_type(dev
, MIGR_INIT
);
3436 vol
->curr_migr_unit
= 0;
3437 map
= get_imsm_map(dev
, 0);
3438 map
->pba_of_lba0
= __cpu_to_le32(super
->create_offset
);
3439 map
->blocks_per_member
= __cpu_to_le32(info_to_blocks_per_member(info
));
3440 map
->blocks_per_strip
= __cpu_to_le16(info_to_blocks_per_strip(info
));
3441 map
->failed_disk_num
= ~0;
3442 map
->map_state
= info
->level
? IMSM_T_STATE_UNINITIALIZED
:
3443 IMSM_T_STATE_NORMAL
;
3446 if (info
->level
== 1 && info
->raid_disks
> 2) {
3449 fprintf(stderr
, Name
": imsm does not support more than 2 disks"
3450 "in a raid1 volume\n");
3454 map
->raid_level
= info
->level
;
3455 if (info
->level
== 10) {
3456 map
->raid_level
= 1;
3457 map
->num_domains
= info
->raid_disks
/ 2;
3458 } else if (info
->level
== 1)
3459 map
->num_domains
= info
->raid_disks
;
3461 map
->num_domains
= 1;
3463 num_data_stripes
= info_to_num_data_stripes(info
, map
->num_domains
);
3464 map
->num_data_stripes
= __cpu_to_le32(num_data_stripes
);
3466 map
->num_members
= info
->raid_disks
;
3467 for (i
= 0; i
< map
->num_members
; i
++) {
3468 /* initialized in add_to_super */
3469 set_imsm_ord_tbl_ent(map
, i
, IMSM_ORD_REBUILD
);
3471 mpb
->num_raid_devs
++;
3474 dv
->index
= super
->current_vol
;
3475 dv
->next
= super
->devlist
;
3476 super
->devlist
= dv
;
3478 imsm_update_version_info(super
);
3483 static int init_super_imsm(struct supertype
*st
, mdu_array_info_t
*info
,
3484 unsigned long long size
, char *name
,
3485 char *homehost
, int *uuid
)
3487 /* This is primarily called by Create when creating a new array.
3488 * We will then get add_to_super called for each component, and then
3489 * write_init_super called to write it out to each device.
3490 * For IMSM, Create can create on fresh devices or on a pre-existing
3492 * To create on a pre-existing array a different method will be called.
3493 * This one is just for fresh drives.
3495 struct intel_super
*super
;
3496 struct imsm_super
*mpb
;
3501 return init_super_imsm_volume(st
, info
, size
, name
, homehost
, uuid
);
3504 mpb_size
= disks_to_mpb_size(info
->nr_disks
);
3508 super
= alloc_super();
3509 if (super
&& posix_memalign(&super
->buf
, 512, mpb_size
) != 0) {
3514 fprintf(stderr
, Name
3515 ": %s could not allocate superblock\n", __func__
);
3518 memset(super
->buf
, 0, mpb_size
);
3520 mpb
->mpb_size
= __cpu_to_le32(mpb_size
);
3524 /* zeroing superblock */
3528 mpb
->attributes
= MPB_ATTRIB_CHECKSUM_VERIFY
;
3530 version
= (char *) mpb
->sig
;
3531 strcpy(version
, MPB_SIGNATURE
);
3532 version
+= strlen(MPB_SIGNATURE
);
3533 strcpy(version
, MPB_VERSION_RAID0
);
3539 static int add_to_super_imsm_volume(struct supertype
*st
, mdu_disk_info_t
*dk
,
3540 int fd
, char *devname
)
3542 struct intel_super
*super
= st
->sb
;
3543 struct imsm_super
*mpb
= super
->anchor
;
3545 struct imsm_dev
*dev
;
3546 struct imsm_map
*map
;
3549 dev
= get_imsm_dev(super
, super
->current_vol
);
3550 map
= get_imsm_map(dev
, 0);
3552 if (! (dk
->state
& (1<<MD_DISK_SYNC
))) {
3553 fprintf(stderr
, Name
": %s: Cannot add spare devices to IMSM volume\n",
3559 /* we're doing autolayout so grab the pre-marked (in
3560 * validate_geometry) raid_disk
3562 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
3563 if (dl
->raiddisk
== dk
->raid_disk
)
3566 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
3567 if (dl
->major
== dk
->major
&&
3568 dl
->minor
== dk
->minor
)
3573 fprintf(stderr
, Name
": %s is not a member of the same container\n", devname
);
3577 /* add a pristine spare to the metadata */
3578 if (dl
->index
< 0) {
3579 dl
->index
= super
->anchor
->num_disks
;
3580 super
->anchor
->num_disks
++;
3582 /* Check the device has not already been added */
3583 slot
= get_imsm_disk_slot(map
, dl
->index
);
3585 (get_imsm_ord_tbl_ent(dev
, slot
, -1) & IMSM_ORD_REBUILD
) == 0) {
3586 fprintf(stderr
, Name
": %s has been included in this array twice\n",
3590 set_imsm_ord_tbl_ent(map
, dk
->number
, dl
->index
);
3591 dl
->disk
.status
= CONFIGURED_DISK
;
3593 /* if we are creating the first raid device update the family number */
3594 if (super
->current_vol
== 0) {
3596 struct imsm_dev
*_dev
= __get_imsm_dev(mpb
, 0);
3597 struct imsm_disk
*_disk
= __get_imsm_disk(mpb
, dl
->index
);
3599 if (!_dev
|| !_disk
) {
3600 fprintf(stderr
, Name
": BUG mpb setup error\n");
3606 sum
+= __gen_imsm_checksum(mpb
);
3607 mpb
->family_num
= __cpu_to_le32(sum
);
3608 mpb
->orig_family_num
= mpb
->family_num
;
3615 static int add_to_super_imsm(struct supertype
*st
, mdu_disk_info_t
*dk
,
3616 int fd
, char *devname
)
3618 struct intel_super
*super
= st
->sb
;
3620 unsigned long long size
;
3625 /* If we are on an RAID enabled platform check that the disk is
3626 * attached to the raid controller.
3627 * We do not need to test disks attachment for container based additions,
3628 * they shall be already tested when container was created/assembled.
3630 if ((fd
!= -1) && !check_env("IMSM_NO_PLATFORM")) {
3631 struct sys_dev
*hba_name
;
3632 struct intel_hba
*hba
;
3634 hba_name
= find_disk_attached_hba(fd
, NULL
);
3637 Name
": %s is not attached to Intel(R) RAID controller.\n",
3638 devname
? : "disk");
3641 rv
= attach_hba_to_super(super
, hba_name
, devname
);
3644 fprintf(stderr
, Name
": %s is attached to Intel(R) %s RAID "
3645 "controller (%s),\n but the container is assigned to Intel(R) "
3646 "%s RAID controller (",
3648 get_sys_dev_type(hba_name
->type
),
3649 hba_name
->pci_id
? : "Err!",
3650 get_sys_dev_type(hba_name
->type
));
3654 fprintf(stderr
, "%s", hba
->pci_id
? : "Err!");
3656 fprintf(stderr
, ", ");
3660 fprintf(stderr
, ").\n"
3661 " Mixing devices attached to different controllers "
3662 "is not allowed.\n");
3663 free_sys_dev(&hba_name
);
3666 free_sys_dev(&hba_name
);
3669 if (super
->current_vol
>= 0)
3670 return add_to_super_imsm_volume(st
, dk
, fd
, devname
);
3673 dd
= malloc(sizeof(*dd
));
3676 Name
": malloc failed %s:%d.\n", __func__
, __LINE__
);
3679 memset(dd
, 0, sizeof(*dd
));
3680 dd
->major
= major(stb
.st_rdev
);
3681 dd
->minor
= minor(stb
.st_rdev
);
3683 dd
->devname
= devname
? strdup(devname
) : NULL
;
3686 dd
->action
= DISK_ADD
;
3687 rv
= imsm_read_serial(fd
, devname
, dd
->serial
);
3690 Name
": failed to retrieve scsi serial, aborting\n");
3695 get_dev_size(fd
, NULL
, &size
);
3697 serialcpy(dd
->disk
.serial
, dd
->serial
);
3698 dd
->disk
.total_blocks
= __cpu_to_le32(size
);
3699 dd
->disk
.status
= SPARE_DISK
;
3700 if (sysfs_disk_to_scsi_id(fd
, &id
) == 0)
3701 dd
->disk
.scsi_id
= __cpu_to_le32(id
);
3703 dd
->disk
.scsi_id
= __cpu_to_le32(0);
3705 if (st
->update_tail
) {
3706 dd
->next
= super
->disk_mgmt_list
;
3707 super
->disk_mgmt_list
= dd
;
3709 dd
->next
= super
->disks
;
3717 static int remove_from_super_imsm(struct supertype
*st
, mdu_disk_info_t
*dk
)
3719 struct intel_super
*super
= st
->sb
;
3722 /* remove from super works only in mdmon - for communication
3723 * manager - monitor. Check if communication memory buffer
3726 if (!st
->update_tail
) {
3728 Name
": %s shall be used in mdmon context only"
3729 "(line %d).\n", __func__
, __LINE__
);
3732 dd
= malloc(sizeof(*dd
));
3735 Name
": malloc failed %s:%d.\n", __func__
, __LINE__
);
3738 memset(dd
, 0, sizeof(*dd
));
3739 dd
->major
= dk
->major
;
3740 dd
->minor
= dk
->minor
;
3743 dd
->disk
.status
= SPARE_DISK
;
3744 dd
->action
= DISK_REMOVE
;
3746 dd
->next
= super
->disk_mgmt_list
;
3747 super
->disk_mgmt_list
= dd
;
3753 static int store_imsm_mpb(int fd
, struct imsm_super
*mpb
);
3757 struct imsm_super anchor
;
3758 } spare_record
__attribute__ ((aligned(512)));
3760 /* spare records have their own family number and do not have any defined raid
3763 static int write_super_imsm_spares(struct intel_super
*super
, int doclose
)
3765 struct imsm_super
*mpb
= super
->anchor
;
3766 struct imsm_super
*spare
= &spare_record
.anchor
;
3770 spare
->mpb_size
= __cpu_to_le32(sizeof(struct imsm_super
)),
3771 spare
->generation_num
= __cpu_to_le32(1UL),
3772 spare
->attributes
= MPB_ATTRIB_CHECKSUM_VERIFY
;
3773 spare
->num_disks
= 1,
3774 spare
->num_raid_devs
= 0,
3775 spare
->cache_size
= mpb
->cache_size
,
3776 spare
->pwr_cycle_count
= __cpu_to_le32(1),
3778 snprintf((char *) spare
->sig
, MAX_SIGNATURE_LENGTH
,
3779 MPB_SIGNATURE MPB_VERSION_RAID0
);
3781 for (d
= super
->disks
; d
; d
= d
->next
) {
3785 spare
->disk
[0] = d
->disk
;
3786 sum
= __gen_imsm_checksum(spare
);
3787 spare
->family_num
= __cpu_to_le32(sum
);
3788 spare
->orig_family_num
= 0;
3789 sum
= __gen_imsm_checksum(spare
);
3790 spare
->check_sum
= __cpu_to_le32(sum
);
3792 if (store_imsm_mpb(d
->fd
, spare
)) {
3793 fprintf(stderr
, "%s: failed for device %d:%d %s\n",
3794 __func__
, d
->major
, d
->minor
, strerror(errno
));
3806 static int write_super_imsm(struct supertype
*st
, int doclose
)
3808 struct intel_super
*super
= st
->sb
;
3809 struct imsm_super
*mpb
= super
->anchor
;
3815 __u32 mpb_size
= sizeof(struct imsm_super
) - sizeof(struct imsm_disk
);
3818 /* 'generation' is incremented everytime the metadata is written */
3819 generation
= __le32_to_cpu(mpb
->generation_num
);
3821 mpb
->generation_num
= __cpu_to_le32(generation
);
3823 /* fix up cases where previous mdadm releases failed to set
3826 if (mpb
->orig_family_num
== 0)
3827 mpb
->orig_family_num
= mpb
->family_num
;
3829 for (d
= super
->disks
; d
; d
= d
->next
) {
3833 mpb
->disk
[d
->index
] = d
->disk
;
3837 for (d
= super
->missing
; d
; d
= d
->next
) {
3838 mpb
->disk
[d
->index
] = d
->disk
;
3841 mpb
->num_disks
= num_disks
;
3842 mpb_size
+= sizeof(struct imsm_disk
) * mpb
->num_disks
;
3844 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
3845 struct imsm_dev
*dev
= __get_imsm_dev(mpb
, i
);
3846 struct imsm_dev
*dev2
= get_imsm_dev(super
, i
);
3848 imsm_copy_dev(dev
, dev2
);
3849 mpb_size
+= sizeof_imsm_dev(dev
, 0);
3852 mpb_size
+= __le32_to_cpu(mpb
->bbm_log_size
);
3853 mpb
->mpb_size
= __cpu_to_le32(mpb_size
);
3855 /* recalculate checksum */
3856 sum
= __gen_imsm_checksum(mpb
);
3857 mpb
->check_sum
= __cpu_to_le32(sum
);
3859 /* write the mpb for disks that compose raid devices */
3860 for (d
= super
->disks
; d
; d
= d
->next
) {
3863 if (store_imsm_mpb(d
->fd
, mpb
))
3864 fprintf(stderr
, "%s: failed for device %d:%d %s\n",
3865 __func__
, d
->major
, d
->minor
, strerror(errno
));
3873 return write_super_imsm_spares(super
, doclose
);
3879 static int create_array(struct supertype
*st
, int dev_idx
)
3882 struct imsm_update_create_array
*u
;
3883 struct intel_super
*super
= st
->sb
;
3884 struct imsm_dev
*dev
= get_imsm_dev(super
, dev_idx
);
3885 struct imsm_map
*map
= get_imsm_map(dev
, 0);
3886 struct disk_info
*inf
;
3887 struct imsm_disk
*disk
;
3890 len
= sizeof(*u
) - sizeof(*dev
) + sizeof_imsm_dev(dev
, 0) +
3891 sizeof(*inf
) * map
->num_members
;
3894 fprintf(stderr
, "%s: failed to allocate update buffer\n",
3899 u
->type
= update_create_array
;
3900 u
->dev_idx
= dev_idx
;
3901 imsm_copy_dev(&u
->dev
, dev
);
3902 inf
= get_disk_info(u
);
3903 for (i
= 0; i
< map
->num_members
; i
++) {
3904 int idx
= get_imsm_disk_idx(dev
, i
, -1);
3906 disk
= get_imsm_disk(super
, idx
);
3907 serialcpy(inf
[i
].serial
, disk
->serial
);
3909 append_metadata_update(st
, u
, len
);
3914 static int mgmt_disk(struct supertype
*st
)
3916 struct intel_super
*super
= st
->sb
;
3918 struct imsm_update_add_remove_disk
*u
;
3920 if (!super
->disk_mgmt_list
)
3926 fprintf(stderr
, "%s: failed to allocate update buffer\n",
3931 u
->type
= update_add_remove_disk
;
3932 append_metadata_update(st
, u
, len
);
3937 static int write_init_super_imsm(struct supertype
*st
)
3939 struct intel_super
*super
= st
->sb
;
3940 int current_vol
= super
->current_vol
;
3942 /* we are done with current_vol reset it to point st at the container */
3943 super
->current_vol
= -1;
3945 if (st
->update_tail
) {
3946 /* queue the recently created array / added disk
3947 * as a metadata update */
3950 /* determine if we are creating a volume or adding a disk */
3951 if (current_vol
< 0) {
3952 /* in the mgmt (add/remove) disk case we are running
3953 * in mdmon context, so don't close fd's
3955 return mgmt_disk(st
);
3957 rv
= create_array(st
, current_vol
);
3962 for (d
= super
->disks
; d
; d
= d
->next
)
3963 Kill(d
->devname
, NULL
, 0, 1, 1);
3964 return write_super_imsm(st
, 1);
3969 static int store_super_imsm(struct supertype
*st
, int fd
)
3971 struct intel_super
*super
= st
->sb
;
3972 struct imsm_super
*mpb
= super
? super
->anchor
: NULL
;
3978 return store_imsm_mpb(fd
, mpb
);
3984 static int imsm_bbm_log_size(struct imsm_super
*mpb
)
3986 return __le32_to_cpu(mpb
->bbm_log_size
);
3990 static int validate_geometry_imsm_container(struct supertype
*st
, int level
,
3991 int layout
, int raiddisks
, int chunk
,
3992 unsigned long long size
, char *dev
,
3993 unsigned long long *freesize
,
3997 unsigned long long ldsize
;
3998 const struct imsm_orom
*orom
;
4000 if (level
!= LEVEL_CONTAINER
)
4005 if (check_env("IMSM_NO_PLATFORM"))
4008 orom
= find_imsm_orom();
4009 if (orom
&& raiddisks
> orom
->tds
) {
4011 fprintf(stderr
, Name
": %d exceeds maximum number of"
4012 " platform supported disks: %d\n",
4013 raiddisks
, orom
->tds
);
4017 fd
= open(dev
, O_RDONLY
|O_EXCL
, 0);
4020 fprintf(stderr
, Name
": imsm: Cannot open %s: %s\n",
4021 dev
, strerror(errno
));
4024 if (!get_dev_size(fd
, dev
, &ldsize
)) {
4030 *freesize
= avail_size_imsm(st
, ldsize
>> 9);
4035 static unsigned long long find_size(struct extent
*e
, int *idx
, int num_extents
)
4037 const unsigned long long base_start
= e
[*idx
].start
;
4038 unsigned long long end
= base_start
+ e
[*idx
].size
;
4041 if (base_start
== end
)
4045 for (i
= *idx
; i
< num_extents
; i
++) {
4046 /* extend overlapping extents */
4047 if (e
[i
].start
>= base_start
&&
4048 e
[i
].start
<= end
) {
4051 if (e
[i
].start
+ e
[i
].size
> end
)
4052 end
= e
[i
].start
+ e
[i
].size
;
4053 } else if (e
[i
].start
> end
) {
4059 return end
- base_start
;
4062 static unsigned long long merge_extents(struct intel_super
*super
, int sum_extents
)
4064 /* build a composite disk with all known extents and generate a new
4065 * 'maxsize' given the "all disks in an array must share a common start
4066 * offset" constraint
4068 struct extent
*e
= calloc(sum_extents
, sizeof(*e
));
4072 unsigned long long pos
;
4073 unsigned long long start
= 0;
4074 unsigned long long maxsize
;
4075 unsigned long reserve
;
4080 /* coalesce and sort all extents. also, check to see if we need to
4081 * reserve space between member arrays
4084 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
4087 for (i
= 0; i
< dl
->extent_cnt
; i
++)
4090 qsort(e
, sum_extents
, sizeof(*e
), cmp_extent
);
4095 while (i
< sum_extents
) {
4096 e
[j
].start
= e
[i
].start
;
4097 e
[j
].size
= find_size(e
, &i
, sum_extents
);
4099 if (e
[j
-1].size
== 0)
4108 unsigned long long esize
;
4110 esize
= e
[i
].start
- pos
;
4111 if (esize
>= maxsize
) {
4116 pos
= e
[i
].start
+ e
[i
].size
;
4118 } while (e
[i
-1].size
);
4124 /* FIXME assumes volume at offset 0 is the first volume in a
4127 if (start_extent
> 0)
4128 reserve
= IMSM_RESERVED_SECTORS
; /* gap between raid regions */
4132 if (maxsize
< reserve
)
4135 super
->create_offset
= ~((__u32
) 0);
4136 if (start
+ reserve
> super
->create_offset
)
4137 return 0; /* start overflows create_offset */
4138 super
->create_offset
= start
+ reserve
;
4140 return maxsize
- reserve
;
4143 static int is_raid_level_supported(const struct imsm_orom
*orom
, int level
, int raiddisks
)
4145 if (level
< 0 || level
== 6 || level
== 4)
4148 /* if we have an orom prevent invalid raid levels */
4151 case 0: return imsm_orom_has_raid0(orom
);
4154 return imsm_orom_has_raid1e(orom
);
4155 return imsm_orom_has_raid1(orom
) && raiddisks
== 2;
4156 case 10: return imsm_orom_has_raid10(orom
) && raiddisks
== 4;
4157 case 5: return imsm_orom_has_raid5(orom
) && raiddisks
> 2;
4160 return 1; /* not on an Intel RAID platform so anything goes */
4165 #define pr_vrb(fmt, arg...) (void) (verbose && fprintf(stderr, Name fmt, ##arg))
4167 validate_geometry_imsm_orom(struct intel_super
*super
, int level
, int layout
,
4168 int raiddisks
, int chunk
, int verbose
)
4170 if (!is_raid_level_supported(super
->orom
, level
, raiddisks
)) {
4171 pr_vrb(": platform does not support raid%d with %d disk%s\n",
4172 level
, raiddisks
, raiddisks
> 1 ? "s" : "");
4175 if (super
->orom
&& level
!= 1 &&
4176 !imsm_orom_has_chunk(super
->orom
, chunk
)) {
4177 pr_vrb(": platform does not support a chunk size of: %d\n", chunk
);
4180 if (layout
!= imsm_level_to_layout(level
)) {
4182 pr_vrb(": imsm raid 5 only supports the left-asymmetric layout\n");
4183 else if (level
== 10)
4184 pr_vrb(": imsm raid 10 only supports the n2 layout\n");
4186 pr_vrb(": imsm unknown layout %#x for this raid level %d\n",
4194 /* validate_geometry_imsm_volume - lifted from validate_geometry_ddf_bvd
4195 * FIX ME add ahci details
4197 static int validate_geometry_imsm_volume(struct supertype
*st
, int level
,
4198 int layout
, int raiddisks
, int chunk
,
4199 unsigned long long size
, char *dev
,
4200 unsigned long long *freesize
,
4204 struct intel_super
*super
= st
->sb
;
4205 struct imsm_super
*mpb
= super
->anchor
;
4207 unsigned long long pos
= 0;
4208 unsigned long long maxsize
;
4212 /* We must have the container info already read in. */
4216 if (!validate_geometry_imsm_orom(super
, level
, layout
, raiddisks
, chunk
, verbose
))
4220 /* General test: make sure there is space for
4221 * 'raiddisks' device extents of size 'size' at a given
4224 unsigned long long minsize
= size
;
4225 unsigned long long start_offset
= MaxSector
;
4228 minsize
= MPB_SECTOR_CNT
+ IMSM_RESERVED_SECTORS
;
4229 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
4234 e
= get_extents(super
, dl
);
4237 unsigned long long esize
;
4238 esize
= e
[i
].start
- pos
;
4239 if (esize
>= minsize
)
4241 if (found
&& start_offset
== MaxSector
) {
4244 } else if (found
&& pos
!= start_offset
) {
4248 pos
= e
[i
].start
+ e
[i
].size
;
4250 } while (e
[i
-1].size
);
4255 if (dcnt
< raiddisks
) {
4257 fprintf(stderr
, Name
": imsm: Not enough "
4258 "devices with space for this array "
4266 /* This device must be a member of the set */
4267 if (stat(dev
, &stb
) < 0)
4269 if ((S_IFMT
& stb
.st_mode
) != S_IFBLK
)
4271 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
4272 if (dl
->major
== (int)major(stb
.st_rdev
) &&
4273 dl
->minor
== (int)minor(stb
.st_rdev
))
4278 fprintf(stderr
, Name
": %s is not in the "
4279 "same imsm set\n", dev
);
4281 } else if (super
->orom
&& dl
->index
< 0 && mpb
->num_raid_devs
) {
4282 /* If a volume is present then the current creation attempt
4283 * cannot incorporate new spares because the orom may not
4284 * understand this configuration (all member disks must be
4285 * members of each array in the container).
4287 fprintf(stderr
, Name
": %s is a spare and a volume"
4288 " is already defined for this container\n", dev
);
4289 fprintf(stderr
, Name
": The option-rom requires all member"
4290 " disks to be a member of all volumes\n");
4294 /* retrieve the largest free space block */
4295 e
= get_extents(super
, dl
);
4300 unsigned long long esize
;
4302 esize
= e
[i
].start
- pos
;
4303 if (esize
>= maxsize
)
4305 pos
= e
[i
].start
+ e
[i
].size
;
4307 } while (e
[i
-1].size
);
4312 fprintf(stderr
, Name
": unable to determine free space for: %s\n",
4316 if (maxsize
< size
) {
4318 fprintf(stderr
, Name
": %s not enough space (%llu < %llu)\n",
4319 dev
, maxsize
, size
);
4323 /* count total number of extents for merge */
4325 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
4327 i
+= dl
->extent_cnt
;
4329 maxsize
= merge_extents(super
, i
);
4330 if (maxsize
< size
|| maxsize
== 0) {
4332 fprintf(stderr
, Name
": not enough space after merge (%llu < %llu)\n",
4337 *freesize
= maxsize
;
4342 static int reserve_space(struct supertype
*st
, int raiddisks
,
4343 unsigned long long size
, int chunk
,
4344 unsigned long long *freesize
)
4346 struct intel_super
*super
= st
->sb
;
4347 struct imsm_super
*mpb
= super
->anchor
;
4352 unsigned long long maxsize
;
4353 unsigned long long minsize
;
4357 /* find the largest common start free region of the possible disks */
4361 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
4367 /* don't activate new spares if we are orom constrained
4368 * and there is already a volume active in the container
4370 if (super
->orom
&& dl
->index
< 0 && mpb
->num_raid_devs
)
4373 e
= get_extents(super
, dl
);
4376 for (i
= 1; e
[i
-1].size
; i
++)
4384 maxsize
= merge_extents(super
, extent_cnt
);
4388 minsize
= chunk
* 2;
4390 if (cnt
< raiddisks
||
4391 (super
->orom
&& used
&& used
!= raiddisks
) ||
4392 maxsize
< minsize
||
4394 fprintf(stderr
, Name
": not enough devices with space to create array.\n");
4395 return 0; /* No enough free spaces large enough */
4407 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
4409 dl
->raiddisk
= cnt
++;
4416 static int validate_geometry_imsm(struct supertype
*st
, int level
, int layout
,
4417 int raiddisks
, int chunk
, unsigned long long size
,
4418 char *dev
, unsigned long long *freesize
,
4425 /* if given unused devices create a container
4426 * if given given devices in a container create a member volume
4428 if (level
== LEVEL_CONTAINER
) {
4429 /* Must be a fresh device to add to a container */
4430 return validate_geometry_imsm_container(st
, level
, layout
,
4431 raiddisks
, chunk
, size
,
4437 if (st
->sb
&& freesize
) {
4438 /* we are being asked to automatically layout a
4439 * new volume based on the current contents of
4440 * the container. If the the parameters can be
4441 * satisfied reserve_space will record the disks,
4442 * start offset, and size of the volume to be
4443 * created. add_to_super and getinfo_super
4444 * detect when autolayout is in progress.
4446 if (!validate_geometry_imsm_orom(st
->sb
, level
, layout
,
4450 return reserve_space(st
, raiddisks
, size
, chunk
, freesize
);
4455 /* creating in a given container */
4456 return validate_geometry_imsm_volume(st
, level
, layout
,
4457 raiddisks
, chunk
, size
,
4458 dev
, freesize
, verbose
);
4461 /* This device needs to be a device in an 'imsm' container */
4462 fd
= open(dev
, O_RDONLY
|O_EXCL
, 0);
4466 Name
": Cannot create this array on device %s\n",
4471 if (errno
!= EBUSY
|| (fd
= open(dev
, O_RDONLY
, 0)) < 0) {
4473 fprintf(stderr
, Name
": Cannot open %s: %s\n",
4474 dev
, strerror(errno
));
4477 /* Well, it is in use by someone, maybe an 'imsm' container. */
4478 cfd
= open_container(fd
);
4482 fprintf(stderr
, Name
": Cannot use %s: It is busy\n",
4486 sra
= sysfs_read(cfd
, 0, GET_VERSION
);
4487 if (sra
&& sra
->array
.major_version
== -1 &&
4488 strcmp(sra
->text_version
, "imsm") == 0)
4492 /* This is a member of a imsm container. Load the container
4493 * and try to create a volume
4495 struct intel_super
*super
;
4497 if (load_super_imsm_all(st
, cfd
, (void **) &super
, NULL
) == 0) {
4499 st
->container_dev
= fd2devnum(cfd
);
4501 return validate_geometry_imsm_volume(st
, level
, layout
,
4509 fprintf(stderr
, Name
": failed container membership check\n");
4515 static void default_geometry_imsm(struct supertype
*st
, int *level
, int *layout
, int *chunk
)
4517 struct intel_super
*super
= st
->sb
;
4519 if (level
&& *level
== UnSet
)
4520 *level
= LEVEL_CONTAINER
;
4522 if (level
&& layout
&& *layout
== UnSet
)
4523 *layout
= imsm_level_to_layout(*level
);
4525 if (chunk
&& (*chunk
== UnSet
|| *chunk
== 0) &&
4526 super
&& super
->orom
)
4527 *chunk
= imsm_orom_default_chunk(super
->orom
);
4530 static void handle_missing(struct intel_super
*super
, struct imsm_dev
*dev
);
4532 static int kill_subarray_imsm(struct supertype
*st
)
4534 /* remove the subarray currently referenced by ->current_vol */
4536 struct intel_dev
**dp
;
4537 struct intel_super
*super
= st
->sb
;
4538 __u8 current_vol
= super
->current_vol
;
4539 struct imsm_super
*mpb
= super
->anchor
;
4541 if (super
->current_vol
< 0)
4543 super
->current_vol
= -1; /* invalidate subarray cursor */
4545 /* block deletions that would change the uuid of active subarrays
4547 * FIXME when immutable ids are available, but note that we'll
4548 * also need to fixup the invalidated/active subarray indexes in
4551 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
4554 if (i
< current_vol
)
4556 sprintf(subarray
, "%u", i
);
4557 if (is_subarray_active(subarray
, st
->devname
)) {
4559 Name
": deleting subarray-%d would change the UUID of active subarray-%d, aborting\n",
4566 if (st
->update_tail
) {
4567 struct imsm_update_kill_array
*u
= malloc(sizeof(*u
));
4571 u
->type
= update_kill_array
;
4572 u
->dev_idx
= current_vol
;
4573 append_metadata_update(st
, u
, sizeof(*u
));
4578 for (dp
= &super
->devlist
; *dp
;)
4579 if ((*dp
)->index
== current_vol
) {
4582 handle_missing(super
, (*dp
)->dev
);
4583 if ((*dp
)->index
> current_vol
)
4588 /* no more raid devices, all active components are now spares,
4589 * but of course failed are still failed
4591 if (--mpb
->num_raid_devs
== 0) {
4594 for (d
= super
->disks
; d
; d
= d
->next
)
4595 if (d
->index
> -2) {
4597 d
->disk
.status
= SPARE_DISK
;
4601 super
->updates_pending
++;
4606 static int update_subarray_imsm(struct supertype
*st
, char *subarray
,
4607 char *update
, struct mddev_ident
*ident
)
4609 /* update the subarray currently referenced by ->current_vol */
4610 struct intel_super
*super
= st
->sb
;
4611 struct imsm_super
*mpb
= super
->anchor
;
4613 if (strcmp(update
, "name") == 0) {
4614 char *name
= ident
->name
;
4618 if (is_subarray_active(subarray
, st
->devname
)) {
4620 Name
": Unable to update name of active subarray\n");
4624 if (!check_name(super
, name
, 0))
4627 vol
= strtoul(subarray
, &ep
, 10);
4628 if (*ep
!= '\0' || vol
>= super
->anchor
->num_raid_devs
)
4631 if (st
->update_tail
) {
4632 struct imsm_update_rename_array
*u
= malloc(sizeof(*u
));
4636 u
->type
= update_rename_array
;
4638 snprintf((char *) u
->name
, MAX_RAID_SERIAL_LEN
, "%s", name
);
4639 append_metadata_update(st
, u
, sizeof(*u
));
4641 struct imsm_dev
*dev
;
4644 dev
= get_imsm_dev(super
, vol
);
4645 snprintf((char *) dev
->volume
, MAX_RAID_SERIAL_LEN
, "%s", name
);
4646 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
4647 dev
= get_imsm_dev(super
, i
);
4648 handle_missing(super
, dev
);
4650 super
->updates_pending
++;
4658 static int is_gen_migration(struct imsm_dev
*dev
)
4660 if (!dev
->vol
.migr_state
)
4663 if (migr_type(dev
) == MIGR_GEN_MIGR
)
4668 #endif /* MDASSEMBLE */
4670 static int is_rebuilding(struct imsm_dev
*dev
)
4672 struct imsm_map
*migr_map
;
4674 if (!dev
->vol
.migr_state
)
4677 if (migr_type(dev
) != MIGR_REBUILD
)
4680 migr_map
= get_imsm_map(dev
, 1);
4682 if (migr_map
->map_state
== IMSM_T_STATE_DEGRADED
)
4688 static void update_recovery_start(struct imsm_dev
*dev
, struct mdinfo
*array
)
4690 struct mdinfo
*rebuild
= NULL
;
4694 if (!is_rebuilding(dev
))
4697 /* Find the rebuild target, but punt on the dual rebuild case */
4698 for (d
= array
->devs
; d
; d
= d
->next
)
4699 if (d
->recovery_start
== 0) {
4706 /* (?) none of the disks are marked with
4707 * IMSM_ORD_REBUILD, so assume they are missing and the
4708 * disk_ord_tbl was not correctly updated
4710 dprintf("%s: failed to locate out-of-sync disk\n", __func__
);
4714 units
= __le32_to_cpu(dev
->vol
.curr_migr_unit
);
4715 rebuild
->recovery_start
= units
* blocks_per_migr_unit(dev
);
4719 static struct mdinfo
*container_content_imsm(struct supertype
*st
, char *subarray
)
4721 /* Given a container loaded by load_super_imsm_all,
4722 * extract information about all the arrays into
4724 * If 'subarray' is given, just extract info about that array.
4726 * For each imsm_dev create an mdinfo, fill it in,
4727 * then look for matching devices in super->disks
4728 * and create appropriate device mdinfo.
4730 struct intel_super
*super
= st
->sb
;
4731 struct imsm_super
*mpb
= super
->anchor
;
4732 struct mdinfo
*rest
= NULL
;
4736 int spare_disks
= 0;
4738 /* check for bad blocks */
4739 if (imsm_bbm_log_size(super
->anchor
))
4742 /* count spare devices, not used in maps
4744 for (d
= super
->disks
; d
; d
= d
->next
)
4748 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
4749 struct imsm_dev
*dev
;
4750 struct imsm_map
*map
;
4751 struct imsm_map
*map2
;
4752 struct mdinfo
*this;
4757 (i
!= strtoul(subarray
, &ep
, 10) || *ep
!= '\0'))
4760 dev
= get_imsm_dev(super
, i
);
4761 map
= get_imsm_map(dev
, 0);
4762 map2
= get_imsm_map(dev
, 1);
4764 /* do not publish arrays that are in the middle of an
4765 * unsupported migration
4767 if (dev
->vol
.migr_state
&&
4768 (migr_type(dev
) == MIGR_STATE_CHANGE
)) {
4769 fprintf(stderr
, Name
": cannot assemble volume '%.16s':"
4770 " unsupported migration in progress\n",
4775 this = malloc(sizeof(*this));
4777 fprintf(stderr
, Name
": failed to allocate %zu bytes\n",
4781 memset(this, 0, sizeof(*this));
4784 super
->current_vol
= i
;
4785 getinfo_super_imsm_volume(st
, this, NULL
);
4786 for (slot
= 0 ; slot
< map
->num_members
; slot
++) {
4787 unsigned long long recovery_start
;
4788 struct mdinfo
*info_d
;
4795 idx
= get_imsm_disk_idx(dev
, slot
, 0);
4796 ord
= get_imsm_ord_tbl_ent(dev
, slot
, -1);
4797 for (d
= super
->disks
; d
; d
= d
->next
)
4798 if (d
->index
== idx
)
4801 recovery_start
= MaxSector
;
4804 if (d
&& is_failed(&d
->disk
))
4806 if (ord
& IMSM_ORD_REBUILD
)
4810 * if we skip some disks the array will be assmebled degraded;
4811 * reset resync start to avoid a dirty-degraded
4812 * situation when performing the intial sync
4814 * FIXME handle dirty degraded
4816 if ((skip
|| recovery_start
== 0) && !dev
->vol
.dirty
)
4817 this->resync_start
= MaxSector
;
4821 info_d
= calloc(1, sizeof(*info_d
));
4823 fprintf(stderr
, Name
": failed to allocate disk"
4824 " for volume %.16s\n", dev
->volume
);
4825 info_d
= this->devs
;
4827 struct mdinfo
*d
= info_d
->next
;
4836 info_d
->next
= this->devs
;
4837 this->devs
= info_d
;
4839 info_d
->disk
.number
= d
->index
;
4840 info_d
->disk
.major
= d
->major
;
4841 info_d
->disk
.minor
= d
->minor
;
4842 info_d
->disk
.raid_disk
= slot
;
4843 info_d
->recovery_start
= recovery_start
;
4845 if (slot
< map2
->num_members
)
4846 info_d
->disk
.state
= (1 << MD_DISK_ACTIVE
);
4848 this->array
.spare_disks
++;
4850 if (slot
< map
->num_members
)
4851 info_d
->disk
.state
= (1 << MD_DISK_ACTIVE
);
4853 this->array
.spare_disks
++;
4855 if (info_d
->recovery_start
== MaxSector
)
4856 this->array
.working_disks
++;
4858 info_d
->events
= __le32_to_cpu(mpb
->generation_num
);
4859 info_d
->data_offset
= __le32_to_cpu(map
->pba_of_lba0
);
4860 info_d
->component_size
= __le32_to_cpu(map
->blocks_per_member
);
4862 /* now that the disk list is up-to-date fixup recovery_start */
4863 update_recovery_start(dev
, this);
4864 this->array
.spare_disks
+= spare_disks
;
4868 /* if array has bad blocks, set suitable bit in array status */
4870 rest
->array
.state
|= (1<<MD_SB_BBM_ERRORS
);
4876 static __u8
imsm_check_degraded(struct intel_super
*super
, struct imsm_dev
*dev
, int failed
)
4878 struct imsm_map
*map
= get_imsm_map(dev
, 0);
4881 return map
->map_state
== IMSM_T_STATE_UNINITIALIZED
?
4882 IMSM_T_STATE_UNINITIALIZED
: IMSM_T_STATE_NORMAL
;
4884 switch (get_imsm_raid_level(map
)) {
4886 return IMSM_T_STATE_FAILED
;
4889 if (failed
< map
->num_members
)
4890 return IMSM_T_STATE_DEGRADED
;
4892 return IMSM_T_STATE_FAILED
;
4897 * check to see if any mirrors have failed, otherwise we
4898 * are degraded. Even numbered slots are mirrored on
4902 /* gcc -Os complains that this is unused */
4903 int insync
= insync
;
4905 for (i
= 0; i
< map
->num_members
; i
++) {
4906 __u32 ord
= get_imsm_ord_tbl_ent(dev
, i
, -1);
4907 int idx
= ord_to_idx(ord
);
4908 struct imsm_disk
*disk
;
4910 /* reset the potential in-sync count on even-numbered
4911 * slots. num_copies is always 2 for imsm raid10
4916 disk
= get_imsm_disk(super
, idx
);
4917 if (!disk
|| is_failed(disk
) || ord
& IMSM_ORD_REBUILD
)
4920 /* no in-sync disks left in this mirror the
4924 return IMSM_T_STATE_FAILED
;
4927 return IMSM_T_STATE_DEGRADED
;
4931 return IMSM_T_STATE_DEGRADED
;
4933 return IMSM_T_STATE_FAILED
;
4939 return map
->map_state
;
4942 static int imsm_count_failed(struct intel_super
*super
, struct imsm_dev
*dev
)
4946 struct imsm_disk
*disk
;
4947 struct imsm_map
*map
= get_imsm_map(dev
, 0);
4948 struct imsm_map
*prev
= get_imsm_map(dev
, dev
->vol
.migr_state
);
4952 /* at the beginning of migration we set IMSM_ORD_REBUILD on
4953 * disks that are being rebuilt. New failures are recorded to
4954 * map[0]. So we look through all the disks we started with and
4955 * see if any failures are still present, or if any new ones
4958 * FIXME add support for online capacity expansion and
4959 * raid-level-migration
4961 for (i
= 0; i
< prev
->num_members
; i
++) {
4962 ord
= __le32_to_cpu(prev
->disk_ord_tbl
[i
]);
4963 ord
|= __le32_to_cpu(map
->disk_ord_tbl
[i
]);
4964 idx
= ord_to_idx(ord
);
4966 disk
= get_imsm_disk(super
, idx
);
4967 if (!disk
|| is_failed(disk
) || ord
& IMSM_ORD_REBUILD
)
4975 static int imsm_open_new(struct supertype
*c
, struct active_array
*a
,
4978 struct intel_super
*super
= c
->sb
;
4979 struct imsm_super
*mpb
= super
->anchor
;
4981 if (atoi(inst
) >= mpb
->num_raid_devs
) {
4982 fprintf(stderr
, "%s: subarry index %d, out of range\n",
4983 __func__
, atoi(inst
));
4987 dprintf("imsm: open_new %s\n", inst
);
4988 a
->info
.container_member
= atoi(inst
);
4992 static int is_resyncing(struct imsm_dev
*dev
)
4994 struct imsm_map
*migr_map
;
4996 if (!dev
->vol
.migr_state
)
4999 if (migr_type(dev
) == MIGR_INIT
||
5000 migr_type(dev
) == MIGR_REPAIR
)
5003 if (migr_type(dev
) == MIGR_GEN_MIGR
)
5006 migr_map
= get_imsm_map(dev
, 1);
5008 if ((migr_map
->map_state
== IMSM_T_STATE_NORMAL
) &&
5009 (dev
->vol
.migr_type
!= MIGR_GEN_MIGR
))
5015 /* return true if we recorded new information */
5016 static int mark_failure(struct imsm_dev
*dev
, struct imsm_disk
*disk
, int idx
)
5020 struct imsm_map
*map
;
5022 /* new failures are always set in map[0] */
5023 map
= get_imsm_map(dev
, 0);
5025 slot
= get_imsm_disk_slot(map
, idx
);
5029 ord
= __le32_to_cpu(map
->disk_ord_tbl
[slot
]);
5030 if (is_failed(disk
) && (ord
& IMSM_ORD_REBUILD
))
5033 disk
->status
|= FAILED_DISK
;
5034 disk
->status
&= ~CONFIGURED_DISK
;
5035 set_imsm_ord_tbl_ent(map
, slot
, idx
| IMSM_ORD_REBUILD
);
5036 if (map
->failed_disk_num
== 0xff)
5037 map
->failed_disk_num
= slot
;
5041 static void mark_missing(struct imsm_dev
*dev
, struct imsm_disk
*disk
, int idx
)
5043 mark_failure(dev
, disk
, idx
);
5045 if (disk
->scsi_id
== __cpu_to_le32(~(__u32
)0))
5048 disk
->scsi_id
= __cpu_to_le32(~(__u32
)0);
5049 memmove(&disk
->serial
[0], &disk
->serial
[1], MAX_RAID_SERIAL_LEN
- 1);
5052 static void handle_missing(struct intel_super
*super
, struct imsm_dev
*dev
)
5058 if (!super
->missing
)
5060 failed
= imsm_count_failed(super
, dev
);
5061 map_state
= imsm_check_degraded(super
, dev
, failed
);
5063 dprintf("imsm: mark missing\n");
5064 end_migration(dev
, map_state
);
5065 for (dl
= super
->missing
; dl
; dl
= dl
->next
)
5066 mark_missing(dev
, &dl
->disk
, dl
->index
);
5067 super
->updates_pending
++;
5070 static unsigned long long imsm_set_array_size(struct imsm_dev
*dev
)
5072 int used_disks
= imsm_num_data_members(dev
, 0);
5073 unsigned long long array_blocks
;
5074 struct imsm_map
*map
;
5076 if (used_disks
== 0) {
5077 /* when problems occures
5078 * return current array_blocks value
5080 array_blocks
= __le32_to_cpu(dev
->size_high
);
5081 array_blocks
= array_blocks
<< 32;
5082 array_blocks
+= __le32_to_cpu(dev
->size_low
);
5084 return array_blocks
;
5087 /* set array size in metadata
5089 map
= get_imsm_map(dev
, 0);
5090 array_blocks
= map
->blocks_per_member
* used_disks
;
5092 /* round array size down to closest MB
5094 array_blocks
= (array_blocks
>> SECT_PER_MB_SHIFT
) << SECT_PER_MB_SHIFT
;
5095 dev
->size_low
= __cpu_to_le32((__u32
)array_blocks
);
5096 dev
->size_high
= __cpu_to_le32((__u32
)(array_blocks
>> 32));
5098 return array_blocks
;
5101 static void imsm_set_disk(struct active_array
*a
, int n
, int state
);
5103 static void imsm_progress_container_reshape(struct intel_super
*super
)
5105 /* if no device has a migr_state, but some device has a
5106 * different number of members than the previous device, start
5107 * changing the number of devices in this device to match
5110 struct imsm_super
*mpb
= super
->anchor
;
5111 int prev_disks
= -1;
5115 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
5116 struct imsm_dev
*dev
= get_imsm_dev(super
, i
);
5117 struct imsm_map
*map
= get_imsm_map(dev
, 0);
5118 struct imsm_map
*map2
;
5119 int prev_num_members
;
5121 if (dev
->vol
.migr_state
)
5124 if (prev_disks
== -1)
5125 prev_disks
= map
->num_members
;
5126 if (prev_disks
== map
->num_members
)
5129 /* OK, this array needs to enter reshape mode.
5130 * i.e it needs a migr_state
5133 copy_map_size
= sizeof_imsm_map(map
);
5134 prev_num_members
= map
->num_members
;
5135 map
->num_members
= prev_disks
;
5136 dev
->vol
.migr_state
= 1;
5137 dev
->vol
.curr_migr_unit
= 0;
5138 dev
->vol
.migr_type
= MIGR_GEN_MIGR
;
5139 for (i
= prev_num_members
;
5140 i
< map
->num_members
; i
++)
5141 set_imsm_ord_tbl_ent(map
, i
, i
);
5142 map2
= get_imsm_map(dev
, 1);
5143 /* Copy the current map */
5144 memcpy(map2
, map
, copy_map_size
);
5145 map2
->num_members
= prev_num_members
;
5147 imsm_set_array_size(dev
);
5148 super
->updates_pending
++;
5152 /* Handle dirty -> clean transititions, resync and reshape. Degraded and rebuild
5153 * states are handled in imsm_set_disk() with one exception, when a
5154 * resync is stopped due to a new failure this routine will set the
5155 * 'degraded' state for the array.
5157 static int imsm_set_array_state(struct active_array
*a
, int consistent
)
5159 int inst
= a
->info
.container_member
;
5160 struct intel_super
*super
= a
->container
->sb
;
5161 struct imsm_dev
*dev
= get_imsm_dev(super
, inst
);
5162 struct imsm_map
*map
= get_imsm_map(dev
, 0);
5163 int failed
= imsm_count_failed(super
, dev
);
5164 __u8 map_state
= imsm_check_degraded(super
, dev
, failed
);
5165 __u32 blocks_per_unit
;
5167 if (dev
->vol
.migr_state
&&
5168 dev
->vol
.migr_type
== MIGR_GEN_MIGR
) {
5169 /* array state change is blocked due to reshape action
5171 * - abort the reshape (if last_checkpoint is 0 and action!= reshape)
5172 * - finish the reshape (if last_checkpoint is big and action != reshape)
5173 * - update curr_migr_unit
5175 if (a
->curr_action
== reshape
) {
5176 /* still reshaping, maybe update curr_migr_unit */
5177 long long blocks_per_unit
= blocks_per_migr_unit(dev
);
5178 long long unit
= a
->last_checkpoint
;
5179 if (blocks_per_unit
) {
5180 unit
/= blocks_per_unit
;
5182 __le32_to_cpu(dev
->vol
.curr_migr_unit
)) {
5183 dev
->vol
.curr_migr_unit
=
5184 __cpu_to_le32(unit
);
5185 super
->updates_pending
++;
5190 if (a
->last_checkpoint
== 0 && a
->prev_action
== reshape
) {
5191 /* for some reason we aborted the reshape.
5194 struct imsm_map
*map2
= get_imsm_map(dev
, 1);
5195 dev
->vol
.migr_state
= 0;
5196 dev
->vol
.migr_type
= 0;
5197 dev
->vol
.curr_migr_unit
= 0;
5198 memcpy(map
, map2
, sizeof_imsm_map(map2
));
5199 super
->updates_pending
++;
5201 if (a
->last_checkpoint
>= a
->info
.component_size
) {
5202 unsigned long long array_blocks
;
5206 used_disks
= imsm_num_data_members(dev
, 0);
5207 if (used_disks
> 0) {
5209 map
->blocks_per_member
*
5211 /* round array size down to closest MB
5213 array_blocks
= (array_blocks
5214 >> SECT_PER_MB_SHIFT
)
5215 << SECT_PER_MB_SHIFT
;
5216 a
->info
.custom_array_size
= array_blocks
;
5217 /* encourage manager to update array
5221 a
->check_reshape
= 1;
5223 /* finalize online capacity expansion/reshape */
5224 for (mdi
= a
->info
.devs
; mdi
; mdi
= mdi
->next
)
5226 mdi
->disk
.raid_disk
,
5229 imsm_progress_container_reshape(super
);
5234 /* before we activate this array handle any missing disks */
5235 if (consistent
== 2)
5236 handle_missing(super
, dev
);
5238 if (consistent
== 2 &&
5239 (!is_resync_complete(&a
->info
) ||
5240 map_state
!= IMSM_T_STATE_NORMAL
||
5241 dev
->vol
.migr_state
))
5244 if (is_resync_complete(&a
->info
)) {
5245 /* complete intialization / resync,
5246 * recovery and interrupted recovery is completed in
5249 if (is_resyncing(dev
)) {
5250 dprintf("imsm: mark resync done\n");
5251 end_migration(dev
, map_state
);
5252 super
->updates_pending
++;
5253 a
->last_checkpoint
= 0;
5255 } else if (!is_resyncing(dev
) && !failed
) {
5256 /* mark the start of the init process if nothing is failed */
5257 dprintf("imsm: mark resync start\n");
5258 if (map
->map_state
== IMSM_T_STATE_UNINITIALIZED
)
5259 migrate(dev
, IMSM_T_STATE_NORMAL
, MIGR_INIT
);
5261 migrate(dev
, IMSM_T_STATE_NORMAL
, MIGR_REPAIR
);
5262 super
->updates_pending
++;
5265 /* check if we can update curr_migr_unit from resync_start, recovery_start */
5266 blocks_per_unit
= blocks_per_migr_unit(dev
);
5267 if (blocks_per_unit
) {
5271 units
= a
->last_checkpoint
/ blocks_per_unit
;
5274 /* check that we did not overflow 32-bits, and that
5275 * curr_migr_unit needs updating
5277 if (units32
== units
&&
5278 __le32_to_cpu(dev
->vol
.curr_migr_unit
) != units32
) {
5279 dprintf("imsm: mark checkpoint (%u)\n", units32
);
5280 dev
->vol
.curr_migr_unit
= __cpu_to_le32(units32
);
5281 super
->updates_pending
++;
5285 /* mark dirty / clean */
5286 if (dev
->vol
.dirty
!= !consistent
) {
5287 dprintf("imsm: mark '%s'\n", consistent
? "clean" : "dirty");
5292 super
->updates_pending
++;
5298 static void imsm_set_disk(struct active_array
*a
, int n
, int state
)
5300 int inst
= a
->info
.container_member
;
5301 struct intel_super
*super
= a
->container
->sb
;
5302 struct imsm_dev
*dev
= get_imsm_dev(super
, inst
);
5303 struct imsm_map
*map
= get_imsm_map(dev
, 0);
5304 struct imsm_disk
*disk
;
5309 if (n
> map
->num_members
)
5310 fprintf(stderr
, "imsm: set_disk %d out of range 0..%d\n",
5311 n
, map
->num_members
- 1);
5316 dprintf("imsm: set_disk %d:%x\n", n
, state
);
5318 ord
= get_imsm_ord_tbl_ent(dev
, n
, -1);
5319 disk
= get_imsm_disk(super
, ord_to_idx(ord
));
5321 /* check for new failures */
5322 if (state
& DS_FAULTY
) {
5323 if (mark_failure(dev
, disk
, ord_to_idx(ord
)))
5324 super
->updates_pending
++;
5327 /* check if in_sync */
5328 if (state
& DS_INSYNC
&& ord
& IMSM_ORD_REBUILD
&& is_rebuilding(dev
)) {
5329 struct imsm_map
*migr_map
= get_imsm_map(dev
, 1);
5331 set_imsm_ord_tbl_ent(migr_map
, n
, ord_to_idx(ord
));
5332 super
->updates_pending
++;
5335 failed
= imsm_count_failed(super
, dev
);
5336 map_state
= imsm_check_degraded(super
, dev
, failed
);
5338 /* check if recovery complete, newly degraded, or failed */
5339 if (map_state
== IMSM_T_STATE_NORMAL
&& is_rebuilding(dev
)) {
5340 end_migration(dev
, map_state
);
5341 map
= get_imsm_map(dev
, 0);
5342 map
->failed_disk_num
= ~0;
5343 super
->updates_pending
++;
5344 a
->last_checkpoint
= 0;
5345 } else if (map_state
== IMSM_T_STATE_DEGRADED
&&
5346 map
->map_state
!= map_state
&&
5347 !dev
->vol
.migr_state
) {
5348 dprintf("imsm: mark degraded\n");
5349 map
->map_state
= map_state
;
5350 super
->updates_pending
++;
5351 a
->last_checkpoint
= 0;
5352 } else if (map_state
== IMSM_T_STATE_FAILED
&&
5353 map
->map_state
!= map_state
) {
5354 dprintf("imsm: mark failed\n");
5355 end_migration(dev
, map_state
);
5356 super
->updates_pending
++;
5357 a
->last_checkpoint
= 0;
5358 } else if (is_gen_migration(dev
)) {
5359 dprintf("imsm: Detected General Migration in state: ");
5360 if (map_state
== IMSM_T_STATE_NORMAL
) {
5361 end_migration(dev
, map_state
);
5362 map
= get_imsm_map(dev
, 0);
5363 map
->failed_disk_num
= ~0;
5364 dprintf("normal\n");
5366 if (map_state
== IMSM_T_STATE_DEGRADED
) {
5367 printf("degraded\n");
5368 end_migration(dev
, map_state
);
5370 dprintf("failed\n");
5372 map
->map_state
= map_state
;
5374 super
->updates_pending
++;
5378 static int store_imsm_mpb(int fd
, struct imsm_super
*mpb
)
5381 __u32 mpb_size
= __le32_to_cpu(mpb
->mpb_size
);
5382 unsigned long long dsize
;
5383 unsigned long long sectors
;
5385 get_dev_size(fd
, NULL
, &dsize
);
5387 if (mpb_size
> 512) {
5388 /* -1 to account for anchor */
5389 sectors
= mpb_sectors(mpb
) - 1;
5391 /* write the extended mpb to the sectors preceeding the anchor */
5392 if (lseek64(fd
, dsize
- (512 * (2 + sectors
)), SEEK_SET
) < 0)
5395 if ((unsigned long long)write(fd
, buf
+ 512, 512 * sectors
)
5400 /* first block is stored on second to last sector of the disk */
5401 if (lseek64(fd
, dsize
- (512 * 2), SEEK_SET
) < 0)
5404 if (write(fd
, buf
, 512) != 512)
5410 static void imsm_sync_metadata(struct supertype
*container
)
5412 struct intel_super
*super
= container
->sb
;
5414 dprintf("sync metadata: %d\n", super
->updates_pending
);
5415 if (!super
->updates_pending
)
5418 write_super_imsm(container
, 0);
5420 super
->updates_pending
= 0;
5423 static struct dl
*imsm_readd(struct intel_super
*super
, int idx
, struct active_array
*a
)
5425 struct imsm_dev
*dev
= get_imsm_dev(super
, a
->info
.container_member
);
5426 int i
= get_imsm_disk_idx(dev
, idx
, -1);
5429 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
5433 if (dl
&& is_failed(&dl
->disk
))
5437 dprintf("%s: found %x:%x\n", __func__
, dl
->major
, dl
->minor
);
5442 static struct dl
*imsm_add_spare(struct intel_super
*super
, int slot
,
5443 struct active_array
*a
, int activate_new
,
5444 struct mdinfo
*additional_test_list
)
5446 struct imsm_dev
*dev
= get_imsm_dev(super
, a
->info
.container_member
);
5447 int idx
= get_imsm_disk_idx(dev
, slot
, -1);
5448 struct imsm_super
*mpb
= super
->anchor
;
5449 struct imsm_map
*map
;
5450 unsigned long long pos
;
5455 __u32 array_start
= 0;
5456 __u32 array_end
= 0;
5458 struct mdinfo
*test_list
;
5460 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
5461 /* If in this array, skip */
5462 for (d
= a
->info
.devs
; d
; d
= d
->next
)
5463 if (d
->state_fd
>= 0 &&
5464 d
->disk
.major
== dl
->major
&&
5465 d
->disk
.minor
== dl
->minor
) {
5466 dprintf("%x:%x already in array\n",
5467 dl
->major
, dl
->minor
);
5472 test_list
= additional_test_list
;
5474 if (test_list
->disk
.major
== dl
->major
&&
5475 test_list
->disk
.minor
== dl
->minor
) {
5476 dprintf("%x:%x already in additional test list\n",
5477 dl
->major
, dl
->minor
);
5480 test_list
= test_list
->next
;
5485 /* skip in use or failed drives */
5486 if (is_failed(&dl
->disk
) || idx
== dl
->index
||
5488 dprintf("%x:%x status (failed: %d index: %d)\n",
5489 dl
->major
, dl
->minor
, is_failed(&dl
->disk
), idx
);
5493 /* skip pure spares when we are looking for partially
5494 * assimilated drives
5496 if (dl
->index
== -1 && !activate_new
)
5499 /* Does this unused device have the requisite free space?
5500 * It needs to be able to cover all member volumes
5502 ex
= get_extents(super
, dl
);
5504 dprintf("cannot get extents\n");
5507 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
5508 dev
= get_imsm_dev(super
, i
);
5509 map
= get_imsm_map(dev
, 0);
5511 /* check if this disk is already a member of
5514 if (get_imsm_disk_slot(map
, dl
->index
) >= 0)
5520 array_start
= __le32_to_cpu(map
->pba_of_lba0
);
5521 array_end
= array_start
+
5522 __le32_to_cpu(map
->blocks_per_member
) - 1;
5525 /* check that we can start at pba_of_lba0 with
5526 * blocks_per_member of space
5528 if (array_start
>= pos
&& array_end
< ex
[j
].start
) {
5532 pos
= ex
[j
].start
+ ex
[j
].size
;
5534 } while (ex
[j
-1].size
);
5541 if (i
< mpb
->num_raid_devs
) {
5542 dprintf("%x:%x does not have %u to %u available\n",
5543 dl
->major
, dl
->minor
, array_start
, array_end
);
5554 static int imsm_rebuild_allowed(struct supertype
*cont
, int dev_idx
, int failed
)
5556 struct imsm_dev
*dev2
;
5557 struct imsm_map
*map
;
5563 dev2
= get_imsm_dev(cont
->sb
, dev_idx
);
5565 state
= imsm_check_degraded(cont
->sb
, dev2
, failed
);
5566 if (state
== IMSM_T_STATE_FAILED
) {
5567 map
= get_imsm_map(dev2
, 0);
5570 for (slot
= 0; slot
< map
->num_members
; slot
++) {
5572 * Check if failed disks are deleted from intel
5573 * disk list or are marked to be deleted
5575 idx
= get_imsm_disk_idx(dev2
, slot
, -1);
5576 idisk
= get_imsm_dl_disk(cont
->sb
, idx
);
5578 * Do not rebuild the array if failed disks
5579 * from failed sub-array are not removed from
5583 is_failed(&idisk
->disk
) &&
5584 (idisk
->action
!= DISK_REMOVE
))
5592 static struct mdinfo
*imsm_activate_spare(struct active_array
*a
,
5593 struct metadata_update
**updates
)
5596 * Find a device with unused free space and use it to replace a
5597 * failed/vacant region in an array. We replace failed regions one a
5598 * array at a time. The result is that a new spare disk will be added
5599 * to the first failed array and after the monitor has finished
5600 * propagating failures the remainder will be consumed.
5602 * FIXME add a capability for mdmon to request spares from another
5606 struct intel_super
*super
= a
->container
->sb
;
5607 int inst
= a
->info
.container_member
;
5608 struct imsm_dev
*dev
= get_imsm_dev(super
, inst
);
5609 struct imsm_map
*map
= get_imsm_map(dev
, 0);
5610 int failed
= a
->info
.array
.raid_disks
;
5611 struct mdinfo
*rv
= NULL
;
5614 struct metadata_update
*mu
;
5616 struct imsm_update_activate_spare
*u
;
5621 for (d
= a
->info
.devs
; d
; d
= d
->next
) {
5622 if ((d
->curr_state
& DS_FAULTY
) &&
5624 /* wait for Removal to happen */
5626 if (d
->state_fd
>= 0)
5630 dprintf("imsm: activate spare: inst=%d failed=%d (%d) level=%d\n",
5631 inst
, failed
, a
->info
.array
.raid_disks
, a
->info
.array
.level
);
5633 if (dev
->vol
.migr_state
&&
5634 dev
->vol
.migr_type
== MIGR_GEN_MIGR
)
5635 /* No repair during migration */
5638 if (a
->info
.array
.level
== 4)
5639 /* No repair for takeovered array
5640 * imsm doesn't support raid4
5644 if (imsm_check_degraded(super
, dev
, failed
) != IMSM_T_STATE_DEGRADED
)
5648 * If there are any failed disks check state of the other volume.
5649 * Block rebuild if the another one is failed until failed disks
5650 * are removed from container.
5653 dprintf("found failed disks in %s, check if there another"
5654 "failed sub-array.\n",
5656 /* check if states of the other volumes allow for rebuild */
5657 for (i
= 0; i
< super
->anchor
->num_raid_devs
; i
++) {
5659 allowed
= imsm_rebuild_allowed(a
->container
,
5667 /* For each slot, if it is not working, find a spare */
5668 for (i
= 0; i
< a
->info
.array
.raid_disks
; i
++) {
5669 for (d
= a
->info
.devs
; d
; d
= d
->next
)
5670 if (d
->disk
.raid_disk
== i
)
5672 dprintf("found %d: %p %x\n", i
, d
, d
?d
->curr_state
:0);
5673 if (d
&& (d
->state_fd
>= 0))
5677 * OK, this device needs recovery. Try to re-add the
5678 * previous occupant of this slot, if this fails see if
5679 * we can continue the assimilation of a spare that was
5680 * partially assimilated, finally try to activate a new
5683 dl
= imsm_readd(super
, i
, a
);
5685 dl
= imsm_add_spare(super
, i
, a
, 0, NULL
);
5687 dl
= imsm_add_spare(super
, i
, a
, 1, NULL
);
5691 /* found a usable disk with enough space */
5692 di
= malloc(sizeof(*di
));
5695 memset(di
, 0, sizeof(*di
));
5697 /* dl->index will be -1 in the case we are activating a
5698 * pristine spare. imsm_process_update() will create a
5699 * new index in this case. Once a disk is found to be
5700 * failed in all member arrays it is kicked from the
5703 di
->disk
.number
= dl
->index
;
5705 /* (ab)use di->devs to store a pointer to the device
5708 di
->devs
= (struct mdinfo
*) dl
;
5710 di
->disk
.raid_disk
= i
;
5711 di
->disk
.major
= dl
->major
;
5712 di
->disk
.minor
= dl
->minor
;
5714 di
->recovery_start
= 0;
5715 di
->data_offset
= __le32_to_cpu(map
->pba_of_lba0
);
5716 di
->component_size
= a
->info
.component_size
;
5717 di
->container_member
= inst
;
5718 super
->random
= random32();
5722 dprintf("%x:%x to be %d at %llu\n", dl
->major
, dl
->minor
,
5723 i
, di
->data_offset
);
5729 /* No spares found */
5731 /* Now 'rv' has a list of devices to return.
5732 * Create a metadata_update record to update the
5733 * disk_ord_tbl for the array
5735 mu
= malloc(sizeof(*mu
));
5737 mu
->buf
= malloc(sizeof(struct imsm_update_activate_spare
) * num_spares
);
5738 if (mu
->buf
== NULL
) {
5745 struct mdinfo
*n
= rv
->next
;
5754 mu
->space_list
= NULL
;
5755 mu
->len
= sizeof(struct imsm_update_activate_spare
) * num_spares
;
5756 mu
->next
= *updates
;
5757 u
= (struct imsm_update_activate_spare
*) mu
->buf
;
5759 for (di
= rv
; di
; di
= di
->next
) {
5760 u
->type
= update_activate_spare
;
5761 u
->dl
= (struct dl
*) di
->devs
;
5763 u
->slot
= di
->disk
.raid_disk
;
5774 static int disks_overlap(struct intel_super
*super
, int idx
, struct imsm_update_create_array
*u
)
5776 struct imsm_dev
*dev
= get_imsm_dev(super
, idx
);
5777 struct imsm_map
*map
= get_imsm_map(dev
, 0);
5778 struct imsm_map
*new_map
= get_imsm_map(&u
->dev
, 0);
5779 struct disk_info
*inf
= get_disk_info(u
);
5780 struct imsm_disk
*disk
;
5784 for (i
= 0; i
< map
->num_members
; i
++) {
5785 disk
= get_imsm_disk(super
, get_imsm_disk_idx(dev
, i
, -1));
5786 for (j
= 0; j
< new_map
->num_members
; j
++)
5787 if (serialcmp(disk
->serial
, inf
[j
].serial
) == 0)
5795 static struct dl
*get_disk_super(struct intel_super
*super
, int major
, int minor
)
5797 struct dl
*dl
= NULL
;
5798 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
5799 if ((dl
->major
== major
) && (dl
->minor
== minor
))
5804 static int remove_disk_super(struct intel_super
*super
, int major
, int minor
)
5806 struct dl
*prev
= NULL
;
5810 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
5811 if ((dl
->major
== major
) && (dl
->minor
== minor
)) {
5814 prev
->next
= dl
->next
;
5816 super
->disks
= dl
->next
;
5818 __free_imsm_disk(dl
);
5819 dprintf("%s: removed %x:%x\n",
5820 __func__
, major
, minor
);
5828 static void imsm_delete(struct intel_super
*super
, struct dl
**dlp
, unsigned index
);
5830 static int add_remove_disk_update(struct intel_super
*super
)
5832 int check_degraded
= 0;
5833 struct dl
*disk
= NULL
;
5834 /* add/remove some spares to/from the metadata/contrainer */
5835 while (super
->disk_mgmt_list
) {
5836 struct dl
*disk_cfg
;
5838 disk_cfg
= super
->disk_mgmt_list
;
5839 super
->disk_mgmt_list
= disk_cfg
->next
;
5840 disk_cfg
->next
= NULL
;
5842 if (disk_cfg
->action
== DISK_ADD
) {
5843 disk_cfg
->next
= super
->disks
;
5844 super
->disks
= disk_cfg
;
5846 dprintf("%s: added %x:%x\n",
5847 __func__
, disk_cfg
->major
,
5849 } else if (disk_cfg
->action
== DISK_REMOVE
) {
5850 dprintf("Disk remove action processed: %x.%x\n",
5851 disk_cfg
->major
, disk_cfg
->minor
);
5852 disk
= get_disk_super(super
,
5856 /* store action status */
5857 disk
->action
= DISK_REMOVE
;
5858 /* remove spare disks only */
5859 if (disk
->index
== -1) {
5860 remove_disk_super(super
,
5865 /* release allocate disk structure */
5866 __free_imsm_disk(disk_cfg
);
5869 return check_degraded
;
5872 static int apply_reshape_container_disks_update(struct imsm_update_reshape
*u
,
5873 struct intel_super
*super
,
5876 struct dl
*new_disk
;
5877 struct intel_dev
*id
;
5879 int delta_disks
= u
->new_raid_disks
- u
->old_raid_disks
;
5880 int disk_count
= u
->old_raid_disks
;
5881 void **tofree
= NULL
;
5882 int devices_to_reshape
= 1;
5883 struct imsm_super
*mpb
= super
->anchor
;
5885 unsigned int dev_id
;
5887 dprintf("imsm: apply_reshape_container_disks_update()\n");
5889 /* enable spares to use in array */
5890 for (i
= 0; i
< delta_disks
; i
++) {
5891 new_disk
= get_disk_super(super
,
5892 major(u
->new_disks
[i
]),
5893 minor(u
->new_disks
[i
]));
5894 dprintf("imsm: new disk for reshape is: %i:%i "
5895 "(%p, index = %i)\n",
5896 major(u
->new_disks
[i
]), minor(u
->new_disks
[i
]),
5897 new_disk
, new_disk
->index
);
5898 if ((new_disk
== NULL
) ||
5899 ((new_disk
->index
>= 0) &&
5900 (new_disk
->index
< u
->old_raid_disks
)))
5901 goto update_reshape_exit
;
5902 new_disk
->index
= disk_count
++;
5903 /* slot to fill in autolayout
5905 new_disk
->raiddisk
= new_disk
->index
;
5906 new_disk
->disk
.status
|=
5908 new_disk
->disk
.status
&= ~SPARE_DISK
;
5911 dprintf("imsm: volume set mpb->num_raid_devs = %i\n",
5912 mpb
->num_raid_devs
);
5913 /* manage changes in volume
5915 for (dev_id
= 0; dev_id
< mpb
->num_raid_devs
; dev_id
++) {
5916 void **sp
= *space_list
;
5917 struct imsm_dev
*newdev
;
5918 struct imsm_map
*newmap
, *oldmap
;
5920 for (id
= super
->devlist
; id
; id
= id
->next
) {
5921 if (id
->index
== dev_id
)
5930 /* Copy the dev, but not (all of) the map */
5931 memcpy(newdev
, id
->dev
, sizeof(*newdev
));
5932 oldmap
= get_imsm_map(id
->dev
, 0);
5933 newmap
= get_imsm_map(newdev
, 0);
5934 /* Copy the current map */
5935 memcpy(newmap
, oldmap
, sizeof_imsm_map(oldmap
));
5936 /* update one device only
5938 if (devices_to_reshape
) {
5939 dprintf("imsm: modifying subdev: %i\n",
5941 devices_to_reshape
--;
5942 newdev
->vol
.migr_state
= 1;
5943 newdev
->vol
.curr_migr_unit
= 0;
5944 newdev
->vol
.migr_type
= MIGR_GEN_MIGR
;
5945 newmap
->num_members
= u
->new_raid_disks
;
5946 for (i
= 0; i
< delta_disks
; i
++) {
5947 set_imsm_ord_tbl_ent(newmap
,
5948 u
->old_raid_disks
+ i
,
5949 u
->old_raid_disks
+ i
);
5951 /* New map is correct, now need to save old map
5953 newmap
= get_imsm_map(newdev
, 1);
5954 memcpy(newmap
, oldmap
, sizeof_imsm_map(oldmap
));
5956 imsm_set_array_size(newdev
);
5959 sp
= (void **)id
->dev
;
5965 *space_list
= tofree
;
5968 update_reshape_exit
:
5973 static int apply_takeover_update(struct imsm_update_takeover
*u
,
5974 struct intel_super
*super
,
5977 struct imsm_dev
*dev
= NULL
;
5978 struct intel_dev
*dv
;
5979 struct imsm_dev
*dev_new
;
5980 struct imsm_map
*map
;
5984 for (dv
= super
->devlist
; dv
; dv
= dv
->next
)
5985 if (dv
->index
== (unsigned int)u
->subarray
) {
5993 map
= get_imsm_map(dev
, 0);
5995 if (u
->direction
== R10_TO_R0
) {
5996 /* Number of failed disks must be half of initial disk number */
5997 if (imsm_count_failed(super
, dev
) != (map
->num_members
/ 2))
6000 /* iterate through devices to mark removed disks as spare */
6001 for (dm
= super
->disks
; dm
; dm
= dm
->next
) {
6002 if (dm
->disk
.status
& FAILED_DISK
) {
6003 int idx
= dm
->index
;
6004 /* update indexes on the disk list */
6005 /* FIXME this loop-with-the-loop looks wrong, I'm not convinced
6006 the index values will end up being correct.... NB */
6007 for (du
= super
->disks
; du
; du
= du
->next
)
6008 if (du
->index
> idx
)
6010 /* mark as spare disk */
6011 dm
->disk
.status
= SPARE_DISK
;
6016 map
->num_members
= map
->num_members
/ 2;
6017 map
->map_state
= IMSM_T_STATE_NORMAL
;
6018 map
->num_domains
= 1;
6019 map
->raid_level
= 0;
6020 map
->failed_disk_num
= -1;
6023 if (u
->direction
== R0_TO_R10
) {
6025 /* update slots in current disk list */
6026 for (dm
= super
->disks
; dm
; dm
= dm
->next
) {
6030 /* create new *missing* disks */
6031 for (i
= 0; i
< map
->num_members
; i
++) {
6032 space
= *space_list
;
6035 *space_list
= *space
;
6037 memcpy(du
, super
->disks
, sizeof(*du
));
6038 du
->disk
.status
= FAILED_DISK
;
6039 du
->disk
.scsi_id
= 0;
6043 du
->index
= (i
* 2) + 1;
6044 sprintf((char *)du
->disk
.serial
,
6045 " MISSING_%d", du
->index
);
6046 sprintf((char *)du
->serial
,
6047 "MISSING_%d", du
->index
);
6048 du
->next
= super
->missing
;
6049 super
->missing
= du
;
6051 /* create new dev and map */
6052 space
= *space_list
;
6055 *space_list
= *space
;
6056 dev_new
= (void *)space
;
6057 memcpy(dev_new
, dev
, sizeof(*dev
));
6058 /* update new map */
6059 map
= get_imsm_map(dev_new
, 0);
6060 map
->failed_disk_num
= map
->num_members
;
6061 map
->num_members
= map
->num_members
* 2;
6062 map
->map_state
= IMSM_T_STATE_NORMAL
;
6063 map
->num_domains
= 2;
6064 map
->raid_level
= 1;
6065 /* replace dev<->dev_new */
6068 /* update disk order table */
6069 for (du
= super
->disks
; du
; du
= du
->next
)
6071 set_imsm_ord_tbl_ent(map
, du
->index
, du
->index
);
6072 for (du
= super
->missing
; du
; du
= du
->next
)
6074 set_imsm_ord_tbl_ent(map
, du
->index
,
6075 du
->index
| IMSM_ORD_REBUILD
);
6080 static void imsm_process_update(struct supertype
*st
,
6081 struct metadata_update
*update
)
6084 * crack open the metadata_update envelope to find the update record
6085 * update can be one of:
6086 * update_reshape_container_disks - all the arrays in the container
6087 * are being reshaped to have more devices. We need to mark
6088 * the arrays for general migration and convert selected spares
6089 * into active devices.
6090 * update_activate_spare - a spare device has replaced a failed
6091 * device in an array, update the disk_ord_tbl. If this disk is
6092 * present in all member arrays then also clear the SPARE_DISK
6094 * update_create_array
6096 * update_rename_array
6097 * update_add_remove_disk
6099 struct intel_super
*super
= st
->sb
;
6100 struct imsm_super
*mpb
;
6101 enum imsm_update_type type
= *(enum imsm_update_type
*) update
->buf
;
6103 /* update requires a larger buf but the allocation failed */
6104 if (super
->next_len
&& !super
->next_buf
) {
6105 super
->next_len
= 0;
6109 if (super
->next_buf
) {
6110 memcpy(super
->next_buf
, super
->buf
, super
->len
);
6112 super
->len
= super
->next_len
;
6113 super
->buf
= super
->next_buf
;
6115 super
->next_len
= 0;
6116 super
->next_buf
= NULL
;
6119 mpb
= super
->anchor
;
6122 case update_takeover
: {
6123 struct imsm_update_takeover
*u
= (void *)update
->buf
;
6124 if (apply_takeover_update(u
, super
, &update
->space_list
))
6125 super
->updates_pending
++;
6129 case update_reshape_container_disks
: {
6130 struct imsm_update_reshape
*u
= (void *)update
->buf
;
6131 if (apply_reshape_container_disks_update(
6132 u
, super
, &update
->space_list
))
6133 super
->updates_pending
++;
6136 case update_activate_spare
: {
6137 struct imsm_update_activate_spare
*u
= (void *) update
->buf
;
6138 struct imsm_dev
*dev
= get_imsm_dev(super
, u
->array
);
6139 struct imsm_map
*map
= get_imsm_map(dev
, 0);
6140 struct imsm_map
*migr_map
;
6141 struct active_array
*a
;
6142 struct imsm_disk
*disk
;
6147 int victim
= get_imsm_disk_idx(dev
, u
->slot
, -1);
6150 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
6155 fprintf(stderr
, "error: imsm_activate_spare passed "
6156 "an unknown disk (index: %d)\n",
6161 super
->updates_pending
++;
6163 /* count failures (excluding rebuilds and the victim)
6164 * to determine map[0] state
6167 for (i
= 0; i
< map
->num_members
; i
++) {
6170 disk
= get_imsm_disk(super
,
6171 get_imsm_disk_idx(dev
, i
, -1));
6172 if (!disk
|| is_failed(disk
))
6176 /* adding a pristine spare, assign a new index */
6177 if (dl
->index
< 0) {
6178 dl
->index
= super
->anchor
->num_disks
;
6179 super
->anchor
->num_disks
++;
6182 disk
->status
|= CONFIGURED_DISK
;
6183 disk
->status
&= ~SPARE_DISK
;
6186 to_state
= imsm_check_degraded(super
, dev
, failed
);
6187 map
->map_state
= IMSM_T_STATE_DEGRADED
;
6188 migrate(dev
, to_state
, MIGR_REBUILD
);
6189 migr_map
= get_imsm_map(dev
, 1);
6190 set_imsm_ord_tbl_ent(map
, u
->slot
, dl
->index
);
6191 set_imsm_ord_tbl_ent(migr_map
, u
->slot
, dl
->index
| IMSM_ORD_REBUILD
);
6193 /* update the family_num to mark a new container
6194 * generation, being careful to record the existing
6195 * family_num in orig_family_num to clean up after
6196 * earlier mdadm versions that neglected to set it.
6198 if (mpb
->orig_family_num
== 0)
6199 mpb
->orig_family_num
= mpb
->family_num
;
6200 mpb
->family_num
+= super
->random
;
6202 /* count arrays using the victim in the metadata */
6204 for (a
= st
->arrays
; a
; a
= a
->next
) {
6205 dev
= get_imsm_dev(super
, a
->info
.container_member
);
6206 map
= get_imsm_map(dev
, 0);
6208 if (get_imsm_disk_slot(map
, victim
) >= 0)
6212 /* delete the victim if it is no longer being
6218 /* We know that 'manager' isn't touching anything,
6219 * so it is safe to delete
6221 for (dlp
= &super
->disks
; *dlp
; dlp
= &(*dlp
)->next
)
6222 if ((*dlp
)->index
== victim
)
6225 /* victim may be on the missing list */
6227 for (dlp
= &super
->missing
; *dlp
; dlp
= &(*dlp
)->next
)
6228 if ((*dlp
)->index
== victim
)
6230 imsm_delete(super
, dlp
, victim
);
6234 case update_create_array
: {
6235 /* someone wants to create a new array, we need to be aware of
6236 * a few races/collisions:
6237 * 1/ 'Create' called by two separate instances of mdadm
6238 * 2/ 'Create' versus 'activate_spare': mdadm has chosen
6239 * devices that have since been assimilated via
6241 * In the event this update can not be carried out mdadm will
6242 * (FIX ME) notice that its update did not take hold.
6244 struct imsm_update_create_array
*u
= (void *) update
->buf
;
6245 struct intel_dev
*dv
;
6246 struct imsm_dev
*dev
;
6247 struct imsm_map
*map
, *new_map
;
6248 unsigned long long start
, end
;
6249 unsigned long long new_start
, new_end
;
6251 struct disk_info
*inf
;
6254 /* handle racing creates: first come first serve */
6255 if (u
->dev_idx
< mpb
->num_raid_devs
) {
6256 dprintf("%s: subarray %d already defined\n",
6257 __func__
, u
->dev_idx
);
6261 /* check update is next in sequence */
6262 if (u
->dev_idx
!= mpb
->num_raid_devs
) {
6263 dprintf("%s: can not create array %d expected index %d\n",
6264 __func__
, u
->dev_idx
, mpb
->num_raid_devs
);
6268 new_map
= get_imsm_map(&u
->dev
, 0);
6269 new_start
= __le32_to_cpu(new_map
->pba_of_lba0
);
6270 new_end
= new_start
+ __le32_to_cpu(new_map
->blocks_per_member
);
6271 inf
= get_disk_info(u
);
6273 /* handle activate_spare versus create race:
6274 * check to make sure that overlapping arrays do not include
6277 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
6278 dev
= get_imsm_dev(super
, i
);
6279 map
= get_imsm_map(dev
, 0);
6280 start
= __le32_to_cpu(map
->pba_of_lba0
);
6281 end
= start
+ __le32_to_cpu(map
->blocks_per_member
);
6282 if ((new_start
>= start
&& new_start
<= end
) ||
6283 (start
>= new_start
&& start
<= new_end
))
6288 if (disks_overlap(super
, i
, u
)) {
6289 dprintf("%s: arrays overlap\n", __func__
);
6294 /* check that prepare update was successful */
6295 if (!update
->space
) {
6296 dprintf("%s: prepare update failed\n", __func__
);
6300 /* check that all disks are still active before committing
6301 * changes. FIXME: could we instead handle this by creating a
6302 * degraded array? That's probably not what the user expects,
6303 * so better to drop this update on the floor.
6305 for (i
= 0; i
< new_map
->num_members
; i
++) {
6306 dl
= serial_to_dl(inf
[i
].serial
, super
);
6308 dprintf("%s: disk disappeared\n", __func__
);
6313 super
->updates_pending
++;
6315 /* convert spares to members and fixup ord_tbl */
6316 for (i
= 0; i
< new_map
->num_members
; i
++) {
6317 dl
= serial_to_dl(inf
[i
].serial
, super
);
6318 if (dl
->index
== -1) {
6319 dl
->index
= mpb
->num_disks
;
6321 dl
->disk
.status
|= CONFIGURED_DISK
;
6322 dl
->disk
.status
&= ~SPARE_DISK
;
6324 set_imsm_ord_tbl_ent(new_map
, i
, dl
->index
);
6329 update
->space
= NULL
;
6330 imsm_copy_dev(dev
, &u
->dev
);
6331 dv
->index
= u
->dev_idx
;
6332 dv
->next
= super
->devlist
;
6333 super
->devlist
= dv
;
6334 mpb
->num_raid_devs
++;
6336 imsm_update_version_info(super
);
6339 /* mdmon knows how to release update->space, but not
6340 * ((struct intel_dev *) update->space)->dev
6342 if (update
->space
) {
6348 case update_kill_array
: {
6349 struct imsm_update_kill_array
*u
= (void *) update
->buf
;
6350 int victim
= u
->dev_idx
;
6351 struct active_array
*a
;
6352 struct intel_dev
**dp
;
6353 struct imsm_dev
*dev
;
6355 /* sanity check that we are not affecting the uuid of
6356 * active arrays, or deleting an active array
6358 * FIXME when immutable ids are available, but note that
6359 * we'll also need to fixup the invalidated/active
6360 * subarray indexes in mdstat
6362 for (a
= st
->arrays
; a
; a
= a
->next
)
6363 if (a
->info
.container_member
>= victim
)
6365 /* by definition if mdmon is running at least one array
6366 * is active in the container, so checking
6367 * mpb->num_raid_devs is just extra paranoia
6369 dev
= get_imsm_dev(super
, victim
);
6370 if (a
|| !dev
|| mpb
->num_raid_devs
== 1) {
6371 dprintf("failed to delete subarray-%d\n", victim
);
6375 for (dp
= &super
->devlist
; *dp
;)
6376 if ((*dp
)->index
== (unsigned)super
->current_vol
) {
6379 if ((*dp
)->index
> (unsigned)victim
)
6383 mpb
->num_raid_devs
--;
6384 super
->updates_pending
++;
6387 case update_rename_array
: {
6388 struct imsm_update_rename_array
*u
= (void *) update
->buf
;
6389 char name
[MAX_RAID_SERIAL_LEN
+1];
6390 int target
= u
->dev_idx
;
6391 struct active_array
*a
;
6392 struct imsm_dev
*dev
;
6394 /* sanity check that we are not affecting the uuid of
6397 snprintf(name
, MAX_RAID_SERIAL_LEN
, "%s", (char *) u
->name
);
6398 name
[MAX_RAID_SERIAL_LEN
] = '\0';
6399 for (a
= st
->arrays
; a
; a
= a
->next
)
6400 if (a
->info
.container_member
== target
)
6402 dev
= get_imsm_dev(super
, u
->dev_idx
);
6403 if (a
|| !dev
|| !check_name(super
, name
, 1)) {
6404 dprintf("failed to rename subarray-%d\n", target
);
6408 snprintf((char *) dev
->volume
, MAX_RAID_SERIAL_LEN
, "%s", name
);
6409 super
->updates_pending
++;
6412 case update_add_remove_disk
: {
6413 /* we may be able to repair some arrays if disks are
6414 * being added, check teh status of add_remove_disk
6415 * if discs has been added.
6417 if (add_remove_disk_update(super
)) {
6418 struct active_array
*a
;
6420 super
->updates_pending
++;
6421 for (a
= st
->arrays
; a
; a
= a
->next
)
6422 a
->check_degraded
= 1;
6427 fprintf(stderr
, "error: unsuported process update type:"
6428 "(type: %d)\n", type
);
6432 static void imsm_prepare_update(struct supertype
*st
,
6433 struct metadata_update
*update
)
6436 * Allocate space to hold new disk entries, raid-device entries or a new
6437 * mpb if necessary. The manager synchronously waits for updates to
6438 * complete in the monitor, so new mpb buffers allocated here can be
6439 * integrated by the monitor thread without worrying about live pointers
6440 * in the manager thread.
6442 enum imsm_update_type type
= *(enum imsm_update_type
*) update
->buf
;
6443 struct intel_super
*super
= st
->sb
;
6444 struct imsm_super
*mpb
= super
->anchor
;
6449 case update_takeover
: {
6450 struct imsm_update_takeover
*u
= (void *)update
->buf
;
6451 if (u
->direction
== R0_TO_R10
) {
6452 void **tail
= (void **)&update
->space_list
;
6453 struct imsm_dev
*dev
= get_imsm_dev(super
, u
->subarray
);
6454 struct imsm_map
*map
= get_imsm_map(dev
, 0);
6455 int num_members
= map
->num_members
;
6459 /* allocate memory for added disks */
6460 for (i
= 0; i
< num_members
; i
++) {
6461 size
= sizeof(struct dl
);
6462 space
= malloc(size
);
6471 /* allocate memory for new device */
6472 size
= sizeof_imsm_dev(super
->devlist
->dev
, 0) +
6473 (num_members
* sizeof(__u32
));
6474 space
= malloc(size
);
6483 len
= disks_to_mpb_size(num_members
* 2);
6485 /* if allocation didn't success, free buffer */
6486 while (update
->space_list
) {
6487 void **sp
= update
->space_list
;
6488 update
->space_list
= *sp
;
6496 case update_reshape_container_disks
: {
6497 /* Every raid device in the container is about to
6498 * gain some more devices, and we will enter a
6500 * So each 'imsm_map' will be bigger, and the imsm_vol
6501 * will now hold 2 of them.
6502 * Thus we need new 'struct imsm_dev' allocations sized
6503 * as sizeof_imsm_dev but with more devices in both maps.
6505 struct imsm_update_reshape
*u
= (void *)update
->buf
;
6506 struct intel_dev
*dl
;
6507 void **space_tail
= (void**)&update
->space_list
;
6509 dprintf("imsm: imsm_prepare_update() for update_reshape\n");
6511 for (dl
= super
->devlist
; dl
; dl
= dl
->next
) {
6512 int size
= sizeof_imsm_dev(dl
->dev
, 1);
6514 if (u
->new_raid_disks
> u
->old_raid_disks
)
6515 size
+= sizeof(__u32
)*2*
6516 (u
->new_raid_disks
- u
->old_raid_disks
);
6525 len
= disks_to_mpb_size(u
->new_raid_disks
);
6526 dprintf("New anchor length is %llu\n", (unsigned long long)len
);
6529 case update_create_array
: {
6530 struct imsm_update_create_array
*u
= (void *) update
->buf
;
6531 struct intel_dev
*dv
;
6532 struct imsm_dev
*dev
= &u
->dev
;
6533 struct imsm_map
*map
= get_imsm_map(dev
, 0);
6535 struct disk_info
*inf
;
6539 inf
= get_disk_info(u
);
6540 len
= sizeof_imsm_dev(dev
, 1);
6541 /* allocate a new super->devlist entry */
6542 dv
= malloc(sizeof(*dv
));
6544 dv
->dev
= malloc(len
);
6549 update
->space
= NULL
;
6553 /* count how many spares will be converted to members */
6554 for (i
= 0; i
< map
->num_members
; i
++) {
6555 dl
= serial_to_dl(inf
[i
].serial
, super
);
6557 /* hmm maybe it failed?, nothing we can do about
6562 if (count_memberships(dl
, super
) == 0)
6565 len
+= activate
* sizeof(struct imsm_disk
);
6572 /* check if we need a larger metadata buffer */
6573 if (super
->next_buf
)
6574 buf_len
= super
->next_len
;
6576 buf_len
= super
->len
;
6578 if (__le32_to_cpu(mpb
->mpb_size
) + len
> buf_len
) {
6579 /* ok we need a larger buf than what is currently allocated
6580 * if this allocation fails process_update will notice that
6581 * ->next_len is set and ->next_buf is NULL
6583 buf_len
= ROUND_UP(__le32_to_cpu(mpb
->mpb_size
) + len
, 512);
6584 if (super
->next_buf
)
6585 free(super
->next_buf
);
6587 super
->next_len
= buf_len
;
6588 if (posix_memalign(&super
->next_buf
, 512, buf_len
) == 0)
6589 memset(super
->next_buf
, 0, buf_len
);
6591 super
->next_buf
= NULL
;
6595 /* must be called while manager is quiesced */
6596 static void imsm_delete(struct intel_super
*super
, struct dl
**dlp
, unsigned index
)
6598 struct imsm_super
*mpb
= super
->anchor
;
6600 struct imsm_dev
*dev
;
6601 struct imsm_map
*map
;
6602 int i
, j
, num_members
;
6605 dprintf("%s: deleting device[%d] from imsm_super\n",
6608 /* shift all indexes down one */
6609 for (iter
= super
->disks
; iter
; iter
= iter
->next
)
6610 if (iter
->index
> (int)index
)
6612 for (iter
= super
->missing
; iter
; iter
= iter
->next
)
6613 if (iter
->index
> (int)index
)
6616 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
6617 dev
= get_imsm_dev(super
, i
);
6618 map
= get_imsm_map(dev
, 0);
6619 num_members
= map
->num_members
;
6620 for (j
= 0; j
< num_members
; j
++) {
6621 /* update ord entries being careful not to propagate
6622 * ord-flags to the first map
6624 ord
= get_imsm_ord_tbl_ent(dev
, j
, -1);
6626 if (ord_to_idx(ord
) <= index
)
6629 map
= get_imsm_map(dev
, 0);
6630 set_imsm_ord_tbl_ent(map
, j
, ord_to_idx(ord
- 1));
6631 map
= get_imsm_map(dev
, 1);
6633 set_imsm_ord_tbl_ent(map
, j
, ord
- 1);
6638 super
->updates_pending
++;
6640 struct dl
*dl
= *dlp
;
6642 *dlp
= (*dlp
)->next
;
6643 __free_imsm_disk(dl
);
6647 static char disk_by_path
[] = "/dev/disk/by-path/";
6649 static const char *imsm_get_disk_controller_domain(const char *path
)
6651 char disk_path
[PATH_MAX
];
6655 strncpy(disk_path
, disk_by_path
, PATH_MAX
- 1);
6656 strncat(disk_path
, path
, PATH_MAX
- strlen(disk_path
) - 1);
6657 if (stat(disk_path
, &st
) == 0) {
6658 struct sys_dev
* hba
;
6661 path
= devt_to_devpath(st
.st_rdev
);
6664 hba
= find_disk_attached_hba(-1, path
);
6665 if (hba
&& hba
->type
== SYS_DEV_SAS
)
6667 else if (hba
&& hba
->type
== SYS_DEV_SATA
)
6671 dprintf("path: %s hba: %s attached: %s\n",
6672 path
, (hba
) ? hba
->path
: "NULL", drv
);
6680 static int imsm_find_array_minor_by_subdev(int subdev
, int container
, int *minor
)
6682 char subdev_name
[20];
6683 struct mdstat_ent
*mdstat
;
6685 sprintf(subdev_name
, "%d", subdev
);
6686 mdstat
= mdstat_by_subdev(subdev_name
, container
);
6690 *minor
= mdstat
->devnum
;
6691 free_mdstat(mdstat
);
6695 static int imsm_reshape_is_allowed_on_container(struct supertype
*st
,
6696 struct geo_params
*geo
,
6697 int *old_raid_disks
)
6699 /* currently we only support increasing the number of devices
6700 * for a container. This increases the number of device for each
6701 * member array. They must all be RAID0 or RAID5.
6704 struct mdinfo
*info
, *member
;
6705 int devices_that_can_grow
= 0;
6707 dprintf("imsm: imsm_reshape_is_allowed_on_container(ENTER): "
6708 "st->devnum = (%i)\n",
6711 if (geo
->size
!= -1 ||
6712 geo
->level
!= UnSet
||
6713 geo
->layout
!= UnSet
||
6714 geo
->chunksize
!= 0 ||
6715 geo
->raid_disks
== UnSet
) {
6716 dprintf("imsm: Container operation is allowed for "
6717 "raid disks number change only.\n");
6721 info
= container_content_imsm(st
, NULL
);
6722 for (member
= info
; member
; member
= member
->next
) {
6726 dprintf("imsm: checking device_num: %i\n",
6727 member
->container_member
);
6729 if (geo
->raid_disks
<= member
->array
.raid_disks
) {
6730 /* we work on container for Online Capacity Expansion
6731 * only so raid_disks has to grow
6733 dprintf("imsm: for container operation raid disks "
6734 "increase is required\n");
6738 if ((info
->array
.level
!= 0) &&
6739 (info
->array
.level
!= 5)) {
6740 /* we cannot use this container with other raid level
6742 dprintf("imsm: for container operation wrong"
6743 " raid level (%i) detected\n",
6747 /* check for platform support
6748 * for this raid level configuration
6750 struct intel_super
*super
= st
->sb
;
6751 if (!is_raid_level_supported(super
->orom
,
6752 member
->array
.level
,
6754 dprintf("platform does not support raid%d with"
6758 geo
->raid_disks
> 1 ? "s" : "");
6763 if (*old_raid_disks
&&
6764 info
->array
.raid_disks
!= *old_raid_disks
)
6766 *old_raid_disks
= info
->array
.raid_disks
;
6768 /* All raid5 and raid0 volumes in container
6769 * have to be ready for Online Capacity Expansion
6770 * so they need to be assembled. We have already
6771 * checked that no recovery etc is happening.
6773 result
= imsm_find_array_minor_by_subdev(member
->container_member
,
6777 dprintf("imsm: cannot find array\n");
6780 devices_that_can_grow
++;
6783 if (!member
&& devices_that_can_grow
)
6787 dprintf("\tContainer operation allowed\n");
6789 dprintf("\tError: %i\n", ret_val
);
6794 /* Function: get_spares_for_grow
6795 * Description: Allocates memory and creates list of spare devices
6796 * avaliable in container. Checks if spare drive size is acceptable.
6797 * Parameters: Pointer to the supertype structure
6798 * Returns: Pointer to the list of spare devices (mdinfo structure) on success,
6801 static struct mdinfo
*get_spares_for_grow(struct supertype
*st
)
6803 unsigned long long min_size
= min_acceptable_spare_size_imsm(st
);
6804 return container_choose_spares(st
, min_size
, NULL
, NULL
, NULL
, 0);
6807 /******************************************************************************
6808 * function: imsm_create_metadata_update_for_reshape
6809 * Function creates update for whole IMSM container.
6811 ******************************************************************************/
6812 static int imsm_create_metadata_update_for_reshape(
6813 struct supertype
*st
,
6814 struct geo_params
*geo
,
6816 struct imsm_update_reshape
**updatep
)
6818 struct intel_super
*super
= st
->sb
;
6819 struct imsm_super
*mpb
= super
->anchor
;
6820 int update_memory_size
= 0;
6821 struct imsm_update_reshape
*u
= NULL
;
6822 struct mdinfo
*spares
= NULL
;
6824 int delta_disks
= 0;
6827 dprintf("imsm_update_metadata_for_reshape(enter) raid_disks = %i\n",
6830 delta_disks
= geo
->raid_disks
- old_raid_disks
;
6832 /* size of all update data without anchor */
6833 update_memory_size
= sizeof(struct imsm_update_reshape
);
6835 /* now add space for spare disks that we need to add. */
6836 update_memory_size
+= sizeof(u
->new_disks
[0]) * (delta_disks
- 1);
6838 u
= calloc(1, update_memory_size
);
6841 "cannot get memory for imsm_update_reshape update\n");
6844 u
->type
= update_reshape_container_disks
;
6845 u
->old_raid_disks
= old_raid_disks
;
6846 u
->new_raid_disks
= geo
->raid_disks
;
6848 /* now get spare disks list
6850 spares
= get_spares_for_grow(st
);
6853 || delta_disks
> spares
->array
.spare_disks
) {
6854 dprintf("imsm: ERROR: Cannot get spare devices.\n");
6858 /* we have got spares
6859 * update disk list in imsm_disk list table in anchor
6861 dprintf("imsm: %i spares are available.\n\n",
6862 spares
->array
.spare_disks
);
6865 for (i
= 0; i
< delta_disks
; i
++) {
6870 u
->new_disks
[i
] = makedev(dev
->disk
.major
,
6872 dl
= get_disk_super(super
, dev
->disk
.major
, dev
->disk
.minor
);
6873 dl
->index
= mpb
->num_disks
;
6883 dprintf("imsm: reshape update preparation :");
6884 if (i
== delta_disks
) {
6887 return update_memory_size
;
6890 dprintf(" Error\n");
6895 static void imsm_update_metadata_locally(struct supertype
*st
,
6898 struct metadata_update mu
;
6903 mu
.space_list
= NULL
;
6905 imsm_prepare_update(st
, &mu
);
6906 imsm_process_update(st
, &mu
);
6908 while (mu
.space_list
) {
6909 void **space
= mu
.space_list
;
6910 mu
.space_list
= *space
;
6915 /***************************************************************************
6916 * Function: imsm_analyze_change
6917 * Description: Function analyze change for single volume
6918 * and validate if transition is supported
6919 * Parameters: Geometry parameters, supertype structure
6920 * Returns: Operation type code on success, -1 if fail
6921 ****************************************************************************/
6922 enum imsm_reshape_type
imsm_analyze_change(struct supertype
*st
,
6923 struct geo_params
*geo
)
6929 getinfo_super_imsm_volume(st
, &info
, NULL
);
6931 if ((geo
->level
!= info
.array
.level
) &&
6932 (geo
->level
>= 0) &&
6933 (geo
->level
!= UnSet
)) {
6934 switch (info
.array
.level
) {
6936 if (geo
->level
== 5) {
6937 change
= CH_MIGRATION
;
6940 if (geo
->level
== 10) {
6941 change
= CH_TAKEOVER
;
6946 if (geo
->level
== 0) {
6947 change
= CH_TAKEOVER
;
6952 if (geo
->level
== 0)
6953 change
= CH_MIGRATION
;
6956 if (geo
->level
== 0) {
6957 change
= CH_TAKEOVER
;
6964 Name
" Error. Level Migration from %d to %d "
6966 info
.array
.level
, geo
->level
);
6967 goto analyse_change_exit
;
6970 geo
->level
= info
.array
.level
;
6972 if ((geo
->layout
!= info
.array
.layout
)
6973 && ((geo
->layout
!= UnSet
) && (geo
->layout
!= -1))) {
6974 change
= CH_MIGRATION
;
6975 if ((info
.array
.layout
== 0)
6976 && (info
.array
.level
== 5)
6977 && (geo
->layout
== 5)) {
6978 /* reshape 5 -> 4 */
6979 } else if ((info
.array
.layout
== 5)
6980 && (info
.array
.level
== 5)
6981 && (geo
->layout
== 0)) {
6982 /* reshape 4 -> 5 */
6987 Name
" Error. Layout Migration from %d to %d "
6989 info
.array
.layout
, geo
->layout
);
6991 goto analyse_change_exit
;
6994 geo
->layout
= info
.array
.layout
;
6996 if ((geo
->chunksize
> 0) && (geo
->chunksize
!= UnSet
)
6997 && (geo
->chunksize
!= info
.array
.chunk_size
))
6998 change
= CH_MIGRATION
;
7000 geo
->chunksize
= info
.array
.chunk_size
;
7002 if (!validate_geometry_imsm(st
,
7006 (geo
->chunksize
/ 1024),
7012 struct intel_super
*super
= st
->sb
;
7013 struct imsm_super
*mpb
= super
->anchor
;
7015 if (mpb
->num_raid_devs
> 1) {
7017 Name
" Error. Cannot perform operation on %s"
7018 "- for this operation it MUST be single "
7019 "array in container\n",
7025 analyse_change_exit
:
7030 int imsm_takeover(struct supertype
*st
, struct geo_params
*geo
)
7032 struct intel_super
*super
= st
->sb
;
7033 struct imsm_update_takeover
*u
;
7035 u
= malloc(sizeof(struct imsm_update_takeover
));
7039 u
->type
= update_takeover
;
7040 u
->subarray
= super
->current_vol
;
7042 /* 10->0 transition */
7043 if (geo
->level
== 0)
7044 u
->direction
= R10_TO_R0
;
7046 /* 0->10 transition */
7047 if (geo
->level
== 10)
7048 u
->direction
= R0_TO_R10
;
7050 /* update metadata locally */
7051 imsm_update_metadata_locally(st
, u
,
7052 sizeof(struct imsm_update_takeover
));
7053 /* and possibly remotely */
7054 if (st
->update_tail
)
7055 append_metadata_update(st
, u
,
7056 sizeof(struct imsm_update_takeover
));
7063 static int imsm_reshape_super(struct supertype
*st
, long long size
, int level
,
7064 int layout
, int chunksize
, int raid_disks
,
7065 int delta_disks
, char *backup
, char *dev
,
7069 struct geo_params geo
;
7071 dprintf("imsm: reshape_super called.\n");
7073 memset(&geo
, 0, sizeof(struct geo_params
));
7076 geo
.dev_id
= st
->devnum
;
7079 geo
.layout
= layout
;
7080 geo
.chunksize
= chunksize
;
7081 geo
.raid_disks
= raid_disks
;
7082 if (delta_disks
!= UnSet
)
7083 geo
.raid_disks
+= delta_disks
;
7085 dprintf("\tfor level : %i\n", geo
.level
);
7086 dprintf("\tfor raid_disks : %i\n", geo
.raid_disks
);
7088 if (experimental() == 0)
7091 if (st
->container_dev
== st
->devnum
) {
7092 /* On container level we can only increase number of devices. */
7093 dprintf("imsm: info: Container operation\n");
7094 int old_raid_disks
= 0;
7095 if (imsm_reshape_is_allowed_on_container(
7096 st
, &geo
, &old_raid_disks
)) {
7097 struct imsm_update_reshape
*u
= NULL
;
7100 len
= imsm_create_metadata_update_for_reshape(
7101 st
, &geo
, old_raid_disks
, &u
);
7104 dprintf("imsm: Cannot prepare update\n");
7105 goto exit_imsm_reshape_super
;
7109 /* update metadata locally */
7110 imsm_update_metadata_locally(st
, u
, len
);
7111 /* and possibly remotely */
7112 if (st
->update_tail
)
7113 append_metadata_update(st
, u
, len
);
7118 fprintf(stderr
, Name
": (imsm) Operation "
7119 "is not allowed on this container\n");
7122 /* On volume level we support following operations
7123 * - takeover: raid10 -> raid0; raid0 -> raid10
7124 * - chunk size migration
7125 * - migration: raid5 -> raid0; raid0 -> raid5
7127 struct intel_super
*super
= st
->sb
;
7128 struct intel_dev
*dev
= super
->devlist
;
7130 dprintf("imsm: info: Volume operation\n");
7131 /* find requested device */
7133 imsm_find_array_minor_by_subdev(dev
->index
, st
->container_dev
, &devnum
);
7134 if (devnum
== geo
.dev_id
)
7139 fprintf(stderr
, Name
" Cannot find %s (%i) subarray\n",
7140 geo
.dev_name
, geo
.dev_id
);
7141 goto exit_imsm_reshape_super
;
7143 super
->current_vol
= dev
->index
;
7144 change
= imsm_analyze_change(st
, &geo
);
7147 ret_val
= imsm_takeover(st
, &geo
);
7157 exit_imsm_reshape_super
:
7158 dprintf("imsm: reshape_super Exit code = %i\n", ret_val
);
7162 static int imsm_manage_reshape(
7163 int afd
, struct mdinfo
*sra
, struct reshape
*reshape
,
7164 struct supertype
*st
, unsigned long stripes
,
7165 int *fds
, unsigned long long *offsets
,
7166 int dests
, int *destfd
, unsigned long long *destoffsets
)
7168 /* Just use child_monitor for now */
7169 return child_monitor(
7170 afd
, sra
, reshape
, st
, stripes
,
7171 fds
, offsets
, dests
, destfd
, destoffsets
);
7173 #endif /* MDASSEMBLE */
7175 struct superswitch super_imsm
= {
7177 .examine_super
= examine_super_imsm
,
7178 .brief_examine_super
= brief_examine_super_imsm
,
7179 .brief_examine_subarrays
= brief_examine_subarrays_imsm
,
7180 .export_examine_super
= export_examine_super_imsm
,
7181 .detail_super
= detail_super_imsm
,
7182 .brief_detail_super
= brief_detail_super_imsm
,
7183 .write_init_super
= write_init_super_imsm
,
7184 .validate_geometry
= validate_geometry_imsm
,
7185 .add_to_super
= add_to_super_imsm
,
7186 .remove_from_super
= remove_from_super_imsm
,
7187 .detail_platform
= detail_platform_imsm
,
7188 .kill_subarray
= kill_subarray_imsm
,
7189 .update_subarray
= update_subarray_imsm
,
7190 .load_container
= load_container_imsm
,
7191 .default_geometry
= default_geometry_imsm
,
7192 .get_disk_controller_domain
= imsm_get_disk_controller_domain
,
7193 .reshape_super
= imsm_reshape_super
,
7194 .manage_reshape
= imsm_manage_reshape
,
7196 .match_home
= match_home_imsm
,
7197 .uuid_from_super
= uuid_from_super_imsm
,
7198 .getinfo_super
= getinfo_super_imsm
,
7199 .getinfo_super_disks
= getinfo_super_disks_imsm
,
7200 .update_super
= update_super_imsm
,
7202 .avail_size
= avail_size_imsm
,
7203 .min_acceptable_spare_size
= min_acceptable_spare_size_imsm
,
7205 .compare_super
= compare_super_imsm
,
7207 .load_super
= load_super_imsm
,
7208 .init_super
= init_super_imsm
,
7209 .store_super
= store_super_imsm
,
7210 .free_super
= free_super_imsm
,
7211 .match_metadata_desc
= match_metadata_desc_imsm
,
7212 .container_content
= container_content_imsm
,
7219 .open_new
= imsm_open_new
,
7220 .set_array_state
= imsm_set_array_state
,
7221 .set_disk
= imsm_set_disk
,
7222 .sync_metadata
= imsm_sync_metadata
,
7223 .activate_spare
= imsm_activate_spare
,
7224 .process_update
= imsm_process_update
,
7225 .prepare_update
= imsm_prepare_update
,
7226 #endif /* MDASSEMBLE */