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 info
->reshape_progress
= 0;
1788 if (map_to_analyse
->map_state
== IMSM_T_STATE_UNINITIALIZED
||
1790 info
->resync_start
= 0;
1791 } else if (dev
->vol
.migr_state
) {
1792 switch (migr_type(dev
)) {
1795 __u64 blocks_per_unit
= blocks_per_migr_unit(dev
);
1796 __u64 units
= __le32_to_cpu(dev
->vol
.curr_migr_unit
);
1798 info
->resync_start
= blocks_per_unit
* units
;
1801 case MIGR_GEN_MIGR
: {
1802 __u64 blocks_per_unit
= blocks_per_migr_unit(dev
);
1803 __u64 units
= __le32_to_cpu(dev
->vol
.curr_migr_unit
);
1805 info
->reshape_progress
= blocks_per_unit
* units
;
1806 dprintf("IMSM: General Migration checkpoint : %llu "
1807 "(%llu) -> read reshape progress : %llu\n",
1808 units
, blocks_per_unit
, info
->reshape_progress
);
1811 /* we could emulate the checkpointing of
1812 * 'sync_action=check' migrations, but for now
1813 * we just immediately complete them
1816 /* this is handled by container_content_imsm() */
1817 case MIGR_STATE_CHANGE
:
1818 /* FIXME handle other migrations */
1820 /* we are not dirty, so... */
1821 info
->resync_start
= MaxSector
;
1824 info
->resync_start
= MaxSector
;
1826 strncpy(info
->name
, (char *) dev
->volume
, MAX_RAID_SERIAL_LEN
);
1827 info
->name
[MAX_RAID_SERIAL_LEN
] = 0;
1829 info
->array
.major_version
= -1;
1830 info
->array
.minor_version
= -2;
1831 devname
= devnum2devname(st
->container_dev
);
1832 *info
->text_version
= '\0';
1834 sprintf(info
->text_version
, "/%s/%d", devname
, info
->container_member
);
1836 info
->safe_mode_delay
= 4000; /* 4 secs like the Matrix driver */
1837 uuid_from_super_imsm(st
, info
->uuid
);
1841 for (i
=0; i
<map_disks
; i
++) {
1843 if (i
< info
->array
.raid_disks
) {
1844 struct imsm_disk
*dsk
;
1845 j
= get_imsm_disk_idx(dev
, i
, -1);
1846 dsk
= get_imsm_disk(super
, j
);
1847 if (dsk
&& (dsk
->status
& CONFIGURED_DISK
))
1854 static __u8
imsm_check_degraded(struct intel_super
*super
, struct imsm_dev
*dev
, int failed
);
1855 static int imsm_count_failed(struct intel_super
*super
, struct imsm_dev
*dev
);
1857 static struct imsm_disk
*get_imsm_missing(struct intel_super
*super
, __u8 index
)
1861 for (d
= super
->missing
; d
; d
= d
->next
)
1862 if (d
->index
== index
)
1867 static void getinfo_super_imsm(struct supertype
*st
, struct mdinfo
*info
, char *map
)
1869 struct intel_super
*super
= st
->sb
;
1870 struct imsm_disk
*disk
;
1871 int map_disks
= info
->array
.raid_disks
;
1872 int max_enough
= -1;
1874 struct imsm_super
*mpb
;
1876 if (super
->current_vol
>= 0) {
1877 getinfo_super_imsm_volume(st
, info
, map
);
1881 /* Set raid_disks to zero so that Assemble will always pull in valid
1884 info
->array
.raid_disks
= 0;
1885 info
->array
.level
= LEVEL_CONTAINER
;
1886 info
->array
.layout
= 0;
1887 info
->array
.md_minor
= -1;
1888 info
->array
.ctime
= 0; /* N/A for imsm */
1889 info
->array
.utime
= 0;
1890 info
->array
.chunk_size
= 0;
1892 info
->disk
.major
= 0;
1893 info
->disk
.minor
= 0;
1894 info
->disk
.raid_disk
= -1;
1895 info
->reshape_active
= 0;
1896 info
->array
.major_version
= -1;
1897 info
->array
.minor_version
= -2;
1898 strcpy(info
->text_version
, "imsm");
1899 info
->safe_mode_delay
= 0;
1900 info
->disk
.number
= -1;
1901 info
->disk
.state
= 0;
1903 info
->recovery_start
= MaxSector
;
1905 /* do we have the all the insync disks that we expect? */
1906 mpb
= super
->anchor
;
1908 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
1909 struct imsm_dev
*dev
= get_imsm_dev(super
, i
);
1910 int failed
, enough
, j
, missing
= 0;
1911 struct imsm_map
*map
;
1914 failed
= imsm_count_failed(super
, dev
);
1915 state
= imsm_check_degraded(super
, dev
, failed
);
1916 map
= get_imsm_map(dev
, dev
->vol
.migr_state
);
1918 /* any newly missing disks?
1919 * (catches single-degraded vs double-degraded)
1921 for (j
= 0; j
< map
->num_members
; j
++) {
1922 __u32 ord
= get_imsm_ord_tbl_ent(dev
, i
, -1);
1923 __u32 idx
= ord_to_idx(ord
);
1925 if (!(ord
& IMSM_ORD_REBUILD
) &&
1926 get_imsm_missing(super
, idx
)) {
1932 if (state
== IMSM_T_STATE_FAILED
)
1934 else if (state
== IMSM_T_STATE_DEGRADED
&&
1935 (state
!= map
->map_state
|| missing
))
1937 else /* we're normal, or already degraded */
1940 /* in the missing/failed disk case check to see
1941 * if at least one array is runnable
1943 max_enough
= max(max_enough
, enough
);
1945 dprintf("%s: enough: %d\n", __func__
, max_enough
);
1946 info
->container_enough
= max_enough
;
1949 __u32 reserved
= imsm_reserved_sectors(super
, super
->disks
);
1951 disk
= &super
->disks
->disk
;
1952 info
->data_offset
= __le32_to_cpu(disk
->total_blocks
) - reserved
;
1953 info
->component_size
= reserved
;
1954 info
->disk
.state
= is_configured(disk
) ? (1 << MD_DISK_ACTIVE
) : 0;
1955 /* we don't change info->disk.raid_disk here because
1956 * this state will be finalized in mdmon after we have
1957 * found the 'most fresh' version of the metadata
1959 info
->disk
.state
|= is_failed(disk
) ? (1 << MD_DISK_FAULTY
) : 0;
1960 info
->disk
.state
|= is_spare(disk
) ? 0 : (1 << MD_DISK_SYNC
);
1963 /* only call uuid_from_super_imsm when this disk is part of a populated container,
1964 * ->compare_super may have updated the 'num_raid_devs' field for spares
1966 if (info
->disk
.state
& (1 << MD_DISK_SYNC
) || super
->anchor
->num_raid_devs
)
1967 uuid_from_super_imsm(st
, info
->uuid
);
1969 memcpy(info
->uuid
, uuid_zero
, sizeof(uuid_zero
));
1971 /* I don't know how to compute 'map' on imsm, so use safe default */
1974 for (i
= 0; i
< map_disks
; i
++)
1980 /* allocates memory and fills disk in mdinfo structure
1981 * for each disk in array */
1982 struct mdinfo
*getinfo_super_disks_imsm(struct supertype
*st
)
1984 struct mdinfo
*mddev
= NULL
;
1985 struct intel_super
*super
= st
->sb
;
1986 struct imsm_disk
*disk
;
1989 if (!super
|| !super
->disks
)
1992 mddev
= malloc(sizeof(*mddev
));
1994 fprintf(stderr
, Name
": Failed to allocate memory.\n");
1997 memset(mddev
, 0, sizeof(*mddev
));
2001 tmp
= malloc(sizeof(*tmp
));
2003 fprintf(stderr
, Name
": Failed to allocate memory.\n");
2008 memset(tmp
, 0, sizeof(*tmp
));
2010 tmp
->next
= mddev
->devs
;
2012 tmp
->disk
.number
= count
++;
2013 tmp
->disk
.major
= dl
->major
;
2014 tmp
->disk
.minor
= dl
->minor
;
2015 tmp
->disk
.state
= is_configured(disk
) ?
2016 (1 << MD_DISK_ACTIVE
) : 0;
2017 tmp
->disk
.state
|= is_failed(disk
) ? (1 << MD_DISK_FAULTY
) : 0;
2018 tmp
->disk
.state
|= is_spare(disk
) ? 0 : (1 << MD_DISK_SYNC
);
2019 tmp
->disk
.raid_disk
= -1;
2025 static int update_super_imsm(struct supertype
*st
, struct mdinfo
*info
,
2026 char *update
, char *devname
, int verbose
,
2027 int uuid_set
, char *homehost
)
2029 /* For 'assemble' and 'force' we need to return non-zero if any
2030 * change was made. For others, the return value is ignored.
2031 * Update options are:
2032 * force-one : This device looks a bit old but needs to be included,
2033 * update age info appropriately.
2034 * assemble: clear any 'faulty' flag to allow this device to
2036 * force-array: Array is degraded but being forced, mark it clean
2037 * if that will be needed to assemble it.
2039 * newdev: not used ????
2040 * grow: Array has gained a new device - this is currently for
2042 * resync: mark as dirty so a resync will happen.
2043 * name: update the name - preserving the homehost
2044 * uuid: Change the uuid of the array to match watch is given
2046 * Following are not relevant for this imsm:
2047 * sparc2.2 : update from old dodgey metadata
2048 * super-minor: change the preferred_minor number
2049 * summaries: update redundant counters.
2050 * homehost: update the recorded homehost
2051 * _reshape_progress: record new reshape_progress position.
2054 struct intel_super
*super
= st
->sb
;
2055 struct imsm_super
*mpb
;
2057 /* we can only update container info */
2058 if (!super
|| super
->current_vol
>= 0 || !super
->anchor
)
2061 mpb
= super
->anchor
;
2063 if (strcmp(update
, "uuid") == 0 && uuid_set
&& !info
->update_private
)
2065 else if (strcmp(update
, "uuid") == 0 && uuid_set
&& info
->update_private
) {
2066 mpb
->orig_family_num
= *((__u32
*) info
->update_private
);
2068 } else if (strcmp(update
, "uuid") == 0) {
2069 __u32
*new_family
= malloc(sizeof(*new_family
));
2071 /* update orig_family_number with the incoming random
2072 * data, report the new effective uuid, and store the
2073 * new orig_family_num for future updates.
2076 memcpy(&mpb
->orig_family_num
, info
->uuid
, sizeof(__u32
));
2077 uuid_from_super_imsm(st
, info
->uuid
);
2078 *new_family
= mpb
->orig_family_num
;
2079 info
->update_private
= new_family
;
2082 } else if (strcmp(update
, "assemble") == 0)
2087 /* successful update? recompute checksum */
2089 mpb
->check_sum
= __le32_to_cpu(__gen_imsm_checksum(mpb
));
2094 static size_t disks_to_mpb_size(int disks
)
2098 size
= sizeof(struct imsm_super
);
2099 size
+= (disks
- 1) * sizeof(struct imsm_disk
);
2100 size
+= 2 * sizeof(struct imsm_dev
);
2101 /* up to 2 maps per raid device (-2 for imsm_maps in imsm_dev */
2102 size
+= (4 - 2) * sizeof(struct imsm_map
);
2103 /* 4 possible disk_ord_tbl's */
2104 size
+= 4 * (disks
- 1) * sizeof(__u32
);
2109 static __u64
avail_size_imsm(struct supertype
*st
, __u64 devsize
)
2111 if (devsize
< (MPB_SECTOR_CNT
+ IMSM_RESERVED_SECTORS
))
2114 return devsize
- (MPB_SECTOR_CNT
+ IMSM_RESERVED_SECTORS
);
2117 static void free_devlist(struct intel_super
*super
)
2119 struct intel_dev
*dv
;
2121 while (super
->devlist
) {
2122 dv
= super
->devlist
->next
;
2123 free(super
->devlist
->dev
);
2124 free(super
->devlist
);
2125 super
->devlist
= dv
;
2129 static void imsm_copy_dev(struct imsm_dev
*dest
, struct imsm_dev
*src
)
2131 memcpy(dest
, src
, sizeof_imsm_dev(src
, 0));
2134 static int compare_super_imsm(struct supertype
*st
, struct supertype
*tst
)
2138 * 0 same, or first was empty, and second was copied
2139 * 1 second had wrong number
2141 * 3 wrong other info
2143 struct intel_super
*first
= st
->sb
;
2144 struct intel_super
*sec
= tst
->sb
;
2152 /* if an anchor does not have num_raid_devs set then it is a free
2155 if (first
->anchor
->num_raid_devs
> 0 &&
2156 sec
->anchor
->num_raid_devs
> 0) {
2157 /* Determine if these disks might ever have been
2158 * related. Further disambiguation can only take place
2159 * in load_super_imsm_all
2161 __u32 first_family
= first
->anchor
->orig_family_num
;
2162 __u32 sec_family
= sec
->anchor
->orig_family_num
;
2164 if (memcmp(first
->anchor
->sig
, sec
->anchor
->sig
,
2165 MAX_SIGNATURE_LENGTH
) != 0)
2168 if (first_family
== 0)
2169 first_family
= first
->anchor
->family_num
;
2170 if (sec_family
== 0)
2171 sec_family
= sec
->anchor
->family_num
;
2173 if (first_family
!= sec_family
)
2179 /* if 'first' is a spare promote it to a populated mpb with sec's
2182 if (first
->anchor
->num_raid_devs
== 0 &&
2183 sec
->anchor
->num_raid_devs
> 0) {
2185 struct intel_dev
*dv
;
2186 struct imsm_dev
*dev
;
2188 /* we need to copy raid device info from sec if an allocation
2189 * fails here we don't associate the spare
2191 for (i
= 0; i
< sec
->anchor
->num_raid_devs
; i
++) {
2192 dv
= malloc(sizeof(*dv
));
2195 dev
= malloc(sizeof_imsm_dev(get_imsm_dev(sec
, i
), 1));
2202 dv
->next
= first
->devlist
;
2203 first
->devlist
= dv
;
2205 if (i
< sec
->anchor
->num_raid_devs
) {
2206 /* allocation failure */
2207 free_devlist(first
);
2208 fprintf(stderr
, "imsm: failed to associate spare\n");
2211 first
->anchor
->num_raid_devs
= sec
->anchor
->num_raid_devs
;
2212 first
->anchor
->orig_family_num
= sec
->anchor
->orig_family_num
;
2213 first
->anchor
->family_num
= sec
->anchor
->family_num
;
2214 memcpy(first
->anchor
->sig
, sec
->anchor
->sig
, MAX_SIGNATURE_LENGTH
);
2215 for (i
= 0; i
< sec
->anchor
->num_raid_devs
; i
++)
2216 imsm_copy_dev(get_imsm_dev(first
, i
), get_imsm_dev(sec
, i
));
2222 static void fd2devname(int fd
, char *name
)
2226 char dname
[PATH_MAX
];
2231 if (fstat(fd
, &st
) != 0)
2233 sprintf(path
, "/sys/dev/block/%d:%d",
2234 major(st
.st_rdev
), minor(st
.st_rdev
));
2236 rv
= readlink(path
, dname
, sizeof(dname
));
2241 nm
= strrchr(dname
, '/');
2243 snprintf(name
, MAX_RAID_SERIAL_LEN
, "/dev/%s", nm
);
2246 extern int scsi_get_serial(int fd
, void *buf
, size_t buf_len
);
2248 static int imsm_read_serial(int fd
, char *devname
,
2249 __u8 serial
[MAX_RAID_SERIAL_LEN
])
2251 unsigned char scsi_serial
[255];
2260 memset(scsi_serial
, 0, sizeof(scsi_serial
));
2262 rv
= scsi_get_serial(fd
, scsi_serial
, sizeof(scsi_serial
));
2264 if (rv
&& check_env("IMSM_DEVNAME_AS_SERIAL")) {
2265 memset(serial
, 0, MAX_RAID_SERIAL_LEN
);
2266 fd2devname(fd
, (char *) serial
);
2273 Name
": Failed to retrieve serial for %s\n",
2278 rsp_len
= scsi_serial
[3];
2282 Name
": Failed to retrieve serial for %s\n",
2286 rsp_buf
= (char *) &scsi_serial
[4];
2288 /* trim all whitespace and non-printable characters and convert
2291 for (i
= 0, dest
= rsp_buf
; i
< rsp_len
; i
++) {
2294 /* ':' is reserved for use in placeholder serial
2295 * numbers for missing disks
2303 len
= dest
- rsp_buf
;
2306 /* truncate leading characters */
2307 if (len
> MAX_RAID_SERIAL_LEN
) {
2308 dest
+= len
- MAX_RAID_SERIAL_LEN
;
2309 len
= MAX_RAID_SERIAL_LEN
;
2312 memset(serial
, 0, MAX_RAID_SERIAL_LEN
);
2313 memcpy(serial
, dest
, len
);
2318 static int serialcmp(__u8
*s1
, __u8
*s2
)
2320 return strncmp((char *) s1
, (char *) s2
, MAX_RAID_SERIAL_LEN
);
2323 static void serialcpy(__u8
*dest
, __u8
*src
)
2325 strncpy((char *) dest
, (char *) src
, MAX_RAID_SERIAL_LEN
);
2329 static struct dl
*serial_to_dl(__u8
*serial
, struct intel_super
*super
)
2333 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
2334 if (serialcmp(dl
->serial
, serial
) == 0)
2341 static struct imsm_disk
*
2342 __serial_to_disk(__u8
*serial
, struct imsm_super
*mpb
, int *idx
)
2346 for (i
= 0; i
< mpb
->num_disks
; i
++) {
2347 struct imsm_disk
*disk
= __get_imsm_disk(mpb
, i
);
2349 if (serialcmp(disk
->serial
, serial
) == 0) {
2360 load_imsm_disk(int fd
, struct intel_super
*super
, char *devname
, int keep_fd
)
2362 struct imsm_disk
*disk
;
2367 __u8 serial
[MAX_RAID_SERIAL_LEN
];
2369 rv
= imsm_read_serial(fd
, devname
, serial
);
2374 dl
= calloc(1, sizeof(*dl
));
2378 Name
": failed to allocate disk buffer for %s\n",
2384 dl
->major
= major(stb
.st_rdev
);
2385 dl
->minor
= minor(stb
.st_rdev
);
2386 dl
->next
= super
->disks
;
2387 dl
->fd
= keep_fd
? fd
: -1;
2388 assert(super
->disks
== NULL
);
2390 serialcpy(dl
->serial
, serial
);
2393 fd2devname(fd
, name
);
2395 dl
->devname
= strdup(devname
);
2397 dl
->devname
= strdup(name
);
2399 /* look up this disk's index in the current anchor */
2400 disk
= __serial_to_disk(dl
->serial
, super
->anchor
, &dl
->index
);
2403 /* only set index on disks that are a member of a
2404 * populated contianer, i.e. one with raid_devs
2406 if (is_failed(&dl
->disk
))
2408 else if (is_spare(&dl
->disk
))
2416 /* When migrating map0 contains the 'destination' state while map1
2417 * contains the current state. When not migrating map0 contains the
2418 * current state. This routine assumes that map[0].map_state is set to
2419 * the current array state before being called.
2421 * Migration is indicated by one of the following states
2422 * 1/ Idle (migr_state=0 map0state=normal||unitialized||degraded||failed)
2423 * 2/ Initialize (migr_state=1 migr_type=MIGR_INIT map0state=normal
2424 * map1state=unitialized)
2425 * 3/ Repair (Resync) (migr_state=1 migr_type=MIGR_REPAIR map0state=normal
2427 * 4/ Rebuild (migr_state=1 migr_type=MIGR_REBUILD map0state=normal
2428 * map1state=degraded)
2430 static void migrate(struct imsm_dev
*dev
, __u8 to_state
, int migr_type
)
2432 struct imsm_map
*dest
;
2433 struct imsm_map
*src
= get_imsm_map(dev
, 0);
2435 dev
->vol
.migr_state
= 1;
2436 set_migr_type(dev
, migr_type
);
2437 dev
->vol
.curr_migr_unit
= 0;
2438 dest
= get_imsm_map(dev
, 1);
2440 /* duplicate and then set the target end state in map[0] */
2441 memcpy(dest
, src
, sizeof_imsm_map(src
));
2442 if ((migr_type
== MIGR_REBUILD
) ||
2443 (migr_type
== MIGR_GEN_MIGR
)) {
2447 for (i
= 0; i
< src
->num_members
; i
++) {
2448 ord
= __le32_to_cpu(src
->disk_ord_tbl
[i
]);
2449 set_imsm_ord_tbl_ent(src
, i
, ord_to_idx(ord
));
2453 src
->map_state
= to_state
;
2456 static void end_migration(struct imsm_dev
*dev
, __u8 map_state
)
2458 struct imsm_map
*map
= get_imsm_map(dev
, 0);
2459 struct imsm_map
*prev
= get_imsm_map(dev
, dev
->vol
.migr_state
);
2462 /* merge any IMSM_ORD_REBUILD bits that were not successfully
2463 * completed in the last migration.
2465 * FIXME add support for raid-level-migration
2467 for (i
= 0; i
< prev
->num_members
; i
++)
2468 for (j
= 0; j
< map
->num_members
; j
++)
2469 /* during online capacity expansion
2470 * disks position can be changed if takeover is used
2472 if (ord_to_idx(map
->disk_ord_tbl
[j
]) ==
2473 ord_to_idx(prev
->disk_ord_tbl
[i
])) {
2474 map
->disk_ord_tbl
[j
] |= prev
->disk_ord_tbl
[i
];
2478 dev
->vol
.migr_state
= 0;
2479 dev
->vol
.migr_type
= 0;
2480 dev
->vol
.curr_migr_unit
= 0;
2481 map
->map_state
= map_state
;
2485 static int parse_raid_devices(struct intel_super
*super
)
2488 struct imsm_dev
*dev_new
;
2489 size_t len
, len_migr
;
2491 size_t space_needed
= 0;
2492 struct imsm_super
*mpb
= super
->anchor
;
2494 for (i
= 0; i
< super
->anchor
->num_raid_devs
; i
++) {
2495 struct imsm_dev
*dev_iter
= __get_imsm_dev(super
->anchor
, i
);
2496 struct intel_dev
*dv
;
2498 len
= sizeof_imsm_dev(dev_iter
, 0);
2499 len_migr
= sizeof_imsm_dev(dev_iter
, 1);
2501 space_needed
+= len_migr
- len
;
2503 dv
= malloc(sizeof(*dv
));
2506 if (max_len
< len_migr
)
2508 if (max_len
> len_migr
)
2509 space_needed
+= max_len
- len_migr
;
2510 dev_new
= malloc(max_len
);
2515 imsm_copy_dev(dev_new
, dev_iter
);
2518 dv
->next
= super
->devlist
;
2519 super
->devlist
= dv
;
2522 /* ensure that super->buf is large enough when all raid devices
2525 if (__le32_to_cpu(mpb
->mpb_size
) + space_needed
> super
->len
) {
2528 len
= ROUND_UP(__le32_to_cpu(mpb
->mpb_size
) + space_needed
, 512);
2529 if (posix_memalign(&buf
, 512, len
) != 0)
2532 memcpy(buf
, super
->buf
, super
->len
);
2533 memset(buf
+ super
->len
, 0, len
- super
->len
);
2542 /* retrieve a pointer to the bbm log which starts after all raid devices */
2543 struct bbm_log
*__get_imsm_bbm_log(struct imsm_super
*mpb
)
2547 if (__le32_to_cpu(mpb
->bbm_log_size
)) {
2549 ptr
+= mpb
->mpb_size
- __le32_to_cpu(mpb
->bbm_log_size
);
2555 static void __free_imsm(struct intel_super
*super
, int free_disks
);
2557 /* load_imsm_mpb - read matrix metadata
2558 * allocates super->mpb to be freed by free_super
2560 static int load_imsm_mpb(int fd
, struct intel_super
*super
, char *devname
)
2562 unsigned long long dsize
;
2563 unsigned long long sectors
;
2565 struct imsm_super
*anchor
;
2568 get_dev_size(fd
, NULL
, &dsize
);
2572 Name
": %s: device to small for imsm\n",
2577 if (lseek64(fd
, dsize
- (512 * 2), SEEK_SET
) < 0) {
2580 Name
": Cannot seek to anchor block on %s: %s\n",
2581 devname
, strerror(errno
));
2585 if (posix_memalign((void**)&anchor
, 512, 512) != 0) {
2588 Name
": Failed to allocate imsm anchor buffer"
2589 " on %s\n", devname
);
2592 if (read(fd
, anchor
, 512) != 512) {
2595 Name
": Cannot read anchor block on %s: %s\n",
2596 devname
, strerror(errno
));
2601 if (strncmp((char *) anchor
->sig
, MPB_SIGNATURE
, MPB_SIG_LEN
) != 0) {
2604 Name
": no IMSM anchor on %s\n", devname
);
2609 __free_imsm(super
, 0);
2610 super
->len
= ROUND_UP(anchor
->mpb_size
, 512);
2611 if (posix_memalign(&super
->buf
, 512, super
->len
) != 0) {
2614 Name
": unable to allocate %zu byte mpb buffer\n",
2619 memcpy(super
->buf
, anchor
, 512);
2621 sectors
= mpb_sectors(anchor
) - 1;
2624 check_sum
= __gen_imsm_checksum(super
->anchor
);
2625 if (check_sum
!= __le32_to_cpu(super
->anchor
->check_sum
)) {
2628 Name
": IMSM checksum %x != %x on %s\n",
2630 __le32_to_cpu(super
->anchor
->check_sum
),
2638 /* read the extended mpb */
2639 if (lseek64(fd
, dsize
- (512 * (2 + sectors
)), SEEK_SET
) < 0) {
2642 Name
": Cannot seek to extended mpb on %s: %s\n",
2643 devname
, strerror(errno
));
2647 if ((unsigned)read(fd
, super
->buf
+ 512, super
->len
- 512) != super
->len
- 512) {
2650 Name
": Cannot read extended mpb on %s: %s\n",
2651 devname
, strerror(errno
));
2655 check_sum
= __gen_imsm_checksum(super
->anchor
);
2656 if (check_sum
!= __le32_to_cpu(super
->anchor
->check_sum
)) {
2659 Name
": IMSM checksum %x != %x on %s\n",
2660 check_sum
, __le32_to_cpu(super
->anchor
->check_sum
),
2665 /* FIXME the BBM log is disk specific so we cannot use this global
2666 * buffer for all disks. Ok for now since we only look at the global
2667 * bbm_log_size parameter to gate assembly
2669 super
->bbm_log
= __get_imsm_bbm_log(super
->anchor
);
2675 load_and_parse_mpb(int fd
, struct intel_super
*super
, char *devname
, int keep_fd
)
2679 err
= load_imsm_mpb(fd
, super
, devname
);
2682 err
= load_imsm_disk(fd
, super
, devname
, keep_fd
);
2685 err
= parse_raid_devices(super
);
2690 static void __free_imsm_disk(struct dl
*d
)
2702 static void free_imsm_disks(struct intel_super
*super
)
2706 while (super
->disks
) {
2708 super
->disks
= d
->next
;
2709 __free_imsm_disk(d
);
2711 while (super
->disk_mgmt_list
) {
2712 d
= super
->disk_mgmt_list
;
2713 super
->disk_mgmt_list
= d
->next
;
2714 __free_imsm_disk(d
);
2716 while (super
->missing
) {
2718 super
->missing
= d
->next
;
2719 __free_imsm_disk(d
);
2724 /* free all the pieces hanging off of a super pointer */
2725 static void __free_imsm(struct intel_super
*super
, int free_disks
)
2727 struct intel_hba
*elem
, *next
;
2734 free_imsm_disks(super
);
2735 free_devlist(super
);
2739 free((void *)elem
->path
);
2747 static void free_imsm(struct intel_super
*super
)
2749 __free_imsm(super
, 1);
2753 static void free_super_imsm(struct supertype
*st
)
2755 struct intel_super
*super
= st
->sb
;
2764 static struct intel_super
*alloc_super(void)
2766 struct intel_super
*super
= malloc(sizeof(*super
));
2769 memset(super
, 0, sizeof(*super
));
2770 super
->current_vol
= -1;
2771 super
->create_offset
= ~((__u32
) 0);
2772 if (!check_env("IMSM_NO_PLATFORM"))
2773 super
->orom
= find_imsm_orom();
2780 /* find_missing - helper routine for load_super_imsm_all that identifies
2781 * disks that have disappeared from the system. This routine relies on
2782 * the mpb being uptodate, which it is at load time.
2784 static int find_missing(struct intel_super
*super
)
2787 struct imsm_super
*mpb
= super
->anchor
;
2789 struct imsm_disk
*disk
;
2791 for (i
= 0; i
< mpb
->num_disks
; i
++) {
2792 disk
= __get_imsm_disk(mpb
, i
);
2793 dl
= serial_to_dl(disk
->serial
, super
);
2797 dl
= malloc(sizeof(*dl
));
2803 dl
->devname
= strdup("missing");
2805 serialcpy(dl
->serial
, disk
->serial
);
2808 dl
->next
= super
->missing
;
2809 super
->missing
= dl
;
2815 static struct intel_disk
*disk_list_get(__u8
*serial
, struct intel_disk
*disk_list
)
2817 struct intel_disk
*idisk
= disk_list
;
2820 if (serialcmp(idisk
->disk
.serial
, serial
) == 0)
2822 idisk
= idisk
->next
;
2828 static int __prep_thunderdome(struct intel_super
**table
, int tbl_size
,
2829 struct intel_super
*super
,
2830 struct intel_disk
**disk_list
)
2832 struct imsm_disk
*d
= &super
->disks
->disk
;
2833 struct imsm_super
*mpb
= super
->anchor
;
2836 for (i
= 0; i
< tbl_size
; i
++) {
2837 struct imsm_super
*tbl_mpb
= table
[i
]->anchor
;
2838 struct imsm_disk
*tbl_d
= &table
[i
]->disks
->disk
;
2840 if (tbl_mpb
->family_num
== mpb
->family_num
) {
2841 if (tbl_mpb
->check_sum
== mpb
->check_sum
) {
2842 dprintf("%s: mpb from %d:%d matches %d:%d\n",
2843 __func__
, super
->disks
->major
,
2844 super
->disks
->minor
,
2845 table
[i
]->disks
->major
,
2846 table
[i
]->disks
->minor
);
2850 if (((is_configured(d
) && !is_configured(tbl_d
)) ||
2851 is_configured(d
) == is_configured(tbl_d
)) &&
2852 tbl_mpb
->generation_num
< mpb
->generation_num
) {
2853 /* current version of the mpb is a
2854 * better candidate than the one in
2855 * super_table, but copy over "cross
2856 * generational" status
2858 struct intel_disk
*idisk
;
2860 dprintf("%s: mpb from %d:%d replaces %d:%d\n",
2861 __func__
, super
->disks
->major
,
2862 super
->disks
->minor
,
2863 table
[i
]->disks
->major
,
2864 table
[i
]->disks
->minor
);
2866 idisk
= disk_list_get(tbl_d
->serial
, *disk_list
);
2867 if (idisk
&& is_failed(&idisk
->disk
))
2868 tbl_d
->status
|= FAILED_DISK
;
2871 struct intel_disk
*idisk
;
2872 struct imsm_disk
*disk
;
2874 /* tbl_mpb is more up to date, but copy
2875 * over cross generational status before
2878 disk
= __serial_to_disk(d
->serial
, mpb
, NULL
);
2879 if (disk
&& is_failed(disk
))
2880 d
->status
|= FAILED_DISK
;
2882 idisk
= disk_list_get(d
->serial
, *disk_list
);
2885 if (disk
&& is_configured(disk
))
2886 idisk
->disk
.status
|= CONFIGURED_DISK
;
2889 dprintf("%s: mpb from %d:%d prefer %d:%d\n",
2890 __func__
, super
->disks
->major
,
2891 super
->disks
->minor
,
2892 table
[i
]->disks
->major
,
2893 table
[i
]->disks
->minor
);
2901 table
[tbl_size
++] = super
;
2905 /* update/extend the merged list of imsm_disk records */
2906 for (j
= 0; j
< mpb
->num_disks
; j
++) {
2907 struct imsm_disk
*disk
= __get_imsm_disk(mpb
, j
);
2908 struct intel_disk
*idisk
;
2910 idisk
= disk_list_get(disk
->serial
, *disk_list
);
2912 idisk
->disk
.status
|= disk
->status
;
2913 if (is_configured(&idisk
->disk
) ||
2914 is_failed(&idisk
->disk
))
2915 idisk
->disk
.status
&= ~(SPARE_DISK
);
2917 idisk
= calloc(1, sizeof(*idisk
));
2920 idisk
->owner
= IMSM_UNKNOWN_OWNER
;
2921 idisk
->disk
= *disk
;
2922 idisk
->next
= *disk_list
;
2926 if (serialcmp(idisk
->disk
.serial
, d
->serial
) == 0)
2933 static struct intel_super
*
2934 validate_members(struct intel_super
*super
, struct intel_disk
*disk_list
,
2937 struct imsm_super
*mpb
= super
->anchor
;
2941 for (i
= 0; i
< mpb
->num_disks
; i
++) {
2942 struct imsm_disk
*disk
= __get_imsm_disk(mpb
, i
);
2943 struct intel_disk
*idisk
;
2945 idisk
= disk_list_get(disk
->serial
, disk_list
);
2947 if (idisk
->owner
== owner
||
2948 idisk
->owner
== IMSM_UNKNOWN_OWNER
)
2951 dprintf("%s: '%.16s' owner %d != %d\n",
2952 __func__
, disk
->serial
, idisk
->owner
,
2955 dprintf("%s: unknown disk %x [%d]: %.16s\n",
2956 __func__
, __le32_to_cpu(mpb
->family_num
), i
,
2962 if (ok_count
== mpb
->num_disks
)
2967 static void show_conflicts(__u32 family_num
, struct intel_super
*super_list
)
2969 struct intel_super
*s
;
2971 for (s
= super_list
; s
; s
= s
->next
) {
2972 if (family_num
!= s
->anchor
->family_num
)
2974 fprintf(stderr
, "Conflict, offlining family %#x on '%s'\n",
2975 __le32_to_cpu(family_num
), s
->disks
->devname
);
2979 static struct intel_super
*
2980 imsm_thunderdome(struct intel_super
**super_list
, int len
)
2982 struct intel_super
*super_table
[len
];
2983 struct intel_disk
*disk_list
= NULL
;
2984 struct intel_super
*champion
, *spare
;
2985 struct intel_super
*s
, **del
;
2990 memset(super_table
, 0, sizeof(super_table
));
2991 for (s
= *super_list
; s
; s
= s
->next
)
2992 tbl_size
= __prep_thunderdome(super_table
, tbl_size
, s
, &disk_list
);
2994 for (i
= 0; i
< tbl_size
; i
++) {
2995 struct imsm_disk
*d
;
2996 struct intel_disk
*idisk
;
2997 struct imsm_super
*mpb
= super_table
[i
]->anchor
;
3000 d
= &s
->disks
->disk
;
3002 /* 'd' must appear in merged disk list for its
3003 * configuration to be valid
3005 idisk
= disk_list_get(d
->serial
, disk_list
);
3006 if (idisk
&& idisk
->owner
== i
)
3007 s
= validate_members(s
, disk_list
, i
);
3012 dprintf("%s: marking family: %#x from %d:%d offline\n",
3013 __func__
, mpb
->family_num
,
3014 super_table
[i
]->disks
->major
,
3015 super_table
[i
]->disks
->minor
);
3019 /* This is where the mdadm implementation differs from the Windows
3020 * driver which has no strict concept of a container. We can only
3021 * assemble one family from a container, so when returning a prodigal
3022 * array member to this system the code will not be able to disambiguate
3023 * the container contents that should be assembled ("foreign" versus
3024 * "local"). It requires user intervention to set the orig_family_num
3025 * to a new value to establish a new container. The Windows driver in
3026 * this situation fixes up the volume name in place and manages the
3027 * foreign array as an independent entity.
3032 for (i
= 0; i
< tbl_size
; i
++) {
3033 struct intel_super
*tbl_ent
= super_table
[i
];
3039 if (tbl_ent
->anchor
->num_raid_devs
== 0) {
3044 if (s
&& !is_spare
) {
3045 show_conflicts(tbl_ent
->anchor
->family_num
, *super_list
);
3047 } else if (!s
&& !is_spare
)
3060 fprintf(stderr
, "Chose family %#x on '%s', "
3061 "assemble conflicts to new container with '--update=uuid'\n",
3062 __le32_to_cpu(s
->anchor
->family_num
), s
->disks
->devname
);
3064 /* collect all dl's onto 'champion', and update them to
3065 * champion's version of the status
3067 for (s
= *super_list
; s
; s
= s
->next
) {
3068 struct imsm_super
*mpb
= champion
->anchor
;
3069 struct dl
*dl
= s
->disks
;
3074 for (i
= 0; i
< mpb
->num_disks
; i
++) {
3075 struct imsm_disk
*disk
;
3077 disk
= __serial_to_disk(dl
->serial
, mpb
, &dl
->index
);
3080 /* only set index on disks that are a member of
3081 * a populated contianer, i.e. one with
3084 if (is_failed(&dl
->disk
))
3086 else if (is_spare(&dl
->disk
))
3092 if (i
>= mpb
->num_disks
) {
3093 struct intel_disk
*idisk
;
3095 idisk
= disk_list_get(dl
->serial
, disk_list
);
3096 if (idisk
&& is_spare(&idisk
->disk
) &&
3097 !is_failed(&idisk
->disk
) && !is_configured(&idisk
->disk
))
3105 dl
->next
= champion
->disks
;
3106 champion
->disks
= dl
;
3110 /* delete 'champion' from super_list */
3111 for (del
= super_list
; *del
; ) {
3112 if (*del
== champion
) {
3113 *del
= (*del
)->next
;
3116 del
= &(*del
)->next
;
3118 champion
->next
= NULL
;
3122 struct intel_disk
*idisk
= disk_list
;
3124 disk_list
= disk_list
->next
;
3131 static int load_super_imsm_all(struct supertype
*st
, int fd
, void **sbp
,
3135 struct intel_super
*super_list
= NULL
;
3136 struct intel_super
*super
= NULL
;
3137 int devnum
= fd2devnum(fd
);
3143 /* check if 'fd' an opened container */
3144 sra
= sysfs_read(fd
, 0, GET_LEVEL
|GET_VERSION
|GET_DEVS
|GET_STATE
);
3148 if (sra
->array
.major_version
!= -1 ||
3149 sra
->array
.minor_version
!= -2 ||
3150 strcmp(sra
->text_version
, "imsm") != 0) {
3155 for (sd
= sra
->devs
, i
= 0; sd
; sd
= sd
->next
, i
++) {
3156 struct intel_super
*s
= alloc_super();
3163 s
->next
= super_list
;
3167 sprintf(nm
, "%d:%d", sd
->disk
.major
, sd
->disk
.minor
);
3168 dfd
= dev_open(nm
, O_RDWR
);
3172 err
= load_and_parse_mpb(dfd
, s
, NULL
, 1);
3174 /* retry the load if we might have raced against mdmon */
3175 if (err
== 3 && mdmon_running(devnum
))
3176 for (retry
= 0; retry
< 3; retry
++) {
3178 err
= load_and_parse_mpb(dfd
, s
, NULL
, 1);
3186 /* all mpbs enter, maybe one leaves */
3187 super
= imsm_thunderdome(&super_list
, i
);
3193 if (find_missing(super
) != 0) {
3201 while (super_list
) {
3202 struct intel_super
*s
= super_list
;
3204 super_list
= super_list
->next
;
3213 st
->container_dev
= devnum
;
3214 if (err
== 0 && st
->ss
== NULL
) {
3215 st
->ss
= &super_imsm
;
3216 st
->minor_version
= 0;
3217 st
->max_devs
= IMSM_MAX_DEVICES
;
3222 static int load_container_imsm(struct supertype
*st
, int fd
, char *devname
)
3224 return load_super_imsm_all(st
, fd
, &st
->sb
, devname
);
3228 static int load_super_imsm(struct supertype
*st
, int fd
, char *devname
)
3230 struct intel_super
*super
;
3233 if (test_partition(fd
))
3234 /* IMSM not allowed on partitions */
3237 free_super_imsm(st
);
3239 super
= alloc_super();
3242 Name
": malloc of %zu failed.\n",
3247 rv
= load_and_parse_mpb(fd
, super
, devname
, 0);
3252 Name
": Failed to load all information "
3253 "sections on %s\n", devname
);
3259 if (st
->ss
== NULL
) {
3260 st
->ss
= &super_imsm
;
3261 st
->minor_version
= 0;
3262 st
->max_devs
= IMSM_MAX_DEVICES
;
3267 static __u16
info_to_blocks_per_strip(mdu_array_info_t
*info
)
3269 if (info
->level
== 1)
3271 return info
->chunk_size
>> 9;
3274 static __u32
info_to_num_data_stripes(mdu_array_info_t
*info
, int num_domains
)
3278 num_stripes
= (info
->size
* 2) / info_to_blocks_per_strip(info
);
3279 num_stripes
/= num_domains
;
3284 static __u32
info_to_blocks_per_member(mdu_array_info_t
*info
)
3286 if (info
->level
== 1)
3287 return info
->size
* 2;
3289 return (info
->size
* 2) & ~(info_to_blocks_per_strip(info
) - 1);
3292 static void imsm_update_version_info(struct intel_super
*super
)
3294 /* update the version and attributes */
3295 struct imsm_super
*mpb
= super
->anchor
;
3297 struct imsm_dev
*dev
;
3298 struct imsm_map
*map
;
3301 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
3302 dev
= get_imsm_dev(super
, i
);
3303 map
= get_imsm_map(dev
, 0);
3304 if (__le32_to_cpu(dev
->size_high
) > 0)
3305 mpb
->attributes
|= MPB_ATTRIB_2TB
;
3307 /* FIXME detect when an array spans a port multiplier */
3309 mpb
->attributes
|= MPB_ATTRIB_PM
;
3312 if (mpb
->num_raid_devs
> 1 ||
3313 mpb
->attributes
!= MPB_ATTRIB_CHECKSUM_VERIFY
) {
3314 version
= MPB_VERSION_ATTRIBS
;
3315 switch (get_imsm_raid_level(map
)) {
3316 case 0: mpb
->attributes
|= MPB_ATTRIB_RAID0
; break;
3317 case 1: mpb
->attributes
|= MPB_ATTRIB_RAID1
; break;
3318 case 10: mpb
->attributes
|= MPB_ATTRIB_RAID10
; break;
3319 case 5: mpb
->attributes
|= MPB_ATTRIB_RAID5
; break;
3322 if (map
->num_members
>= 5)
3323 version
= MPB_VERSION_5OR6_DISK_ARRAY
;
3324 else if (dev
->status
== DEV_CLONE_N_GO
)
3325 version
= MPB_VERSION_CNG
;
3326 else if (get_imsm_raid_level(map
) == 5)
3327 version
= MPB_VERSION_RAID5
;
3328 else if (map
->num_members
>= 3)
3329 version
= MPB_VERSION_3OR4_DISK_ARRAY
;
3330 else if (get_imsm_raid_level(map
) == 1)
3331 version
= MPB_VERSION_RAID1
;
3333 version
= MPB_VERSION_RAID0
;
3335 strcpy(((char *) mpb
->sig
) + strlen(MPB_SIGNATURE
), version
);
3339 static int check_name(struct intel_super
*super
, char *name
, int quiet
)
3341 struct imsm_super
*mpb
= super
->anchor
;
3342 char *reason
= NULL
;
3345 if (strlen(name
) > MAX_RAID_SERIAL_LEN
)
3346 reason
= "must be 16 characters or less";
3348 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
3349 struct imsm_dev
*dev
= get_imsm_dev(super
, i
);
3351 if (strncmp((char *) dev
->volume
, name
, MAX_RAID_SERIAL_LEN
) == 0) {
3352 reason
= "already exists";
3357 if (reason
&& !quiet
)
3358 fprintf(stderr
, Name
": imsm volume name %s\n", reason
);
3363 static int init_super_imsm_volume(struct supertype
*st
, mdu_array_info_t
*info
,
3364 unsigned long long size
, char *name
,
3365 char *homehost
, int *uuid
)
3367 /* We are creating a volume inside a pre-existing container.
3368 * so st->sb is already set.
3370 struct intel_super
*super
= st
->sb
;
3371 struct imsm_super
*mpb
= super
->anchor
;
3372 struct intel_dev
*dv
;
3373 struct imsm_dev
*dev
;
3374 struct imsm_vol
*vol
;
3375 struct imsm_map
*map
;
3376 int idx
= mpb
->num_raid_devs
;
3378 unsigned long long array_blocks
;
3379 size_t size_old
, size_new
;
3380 __u32 num_data_stripes
;
3382 if (super
->orom
&& mpb
->num_raid_devs
>= super
->orom
->vpa
) {
3383 fprintf(stderr
, Name
": This imsm-container already has the "
3384 "maximum of %d volumes\n", super
->orom
->vpa
);
3388 /* ensure the mpb is large enough for the new data */
3389 size_old
= __le32_to_cpu(mpb
->mpb_size
);
3390 size_new
= disks_to_mpb_size(info
->nr_disks
);
3391 if (size_new
> size_old
) {
3393 size_t size_round
= ROUND_UP(size_new
, 512);
3395 if (posix_memalign(&mpb_new
, 512, size_round
) != 0) {
3396 fprintf(stderr
, Name
": could not allocate new mpb\n");
3399 memcpy(mpb_new
, mpb
, size_old
);
3402 super
->anchor
= mpb_new
;
3403 mpb
->mpb_size
= __cpu_to_le32(size_new
);
3404 memset(mpb_new
+ size_old
, 0, size_round
- size_old
);
3406 super
->current_vol
= idx
;
3407 /* when creating the first raid device in this container set num_disks
3408 * to zero, i.e. delete this spare and add raid member devices in
3409 * add_to_super_imsm_volume()
3411 if (super
->current_vol
== 0)
3414 if (!check_name(super
, name
, 0))
3416 dv
= malloc(sizeof(*dv
));
3418 fprintf(stderr
, Name
": failed to allocate device list entry\n");
3421 dev
= malloc(sizeof(*dev
) + sizeof(__u32
) * (info
->raid_disks
- 1));
3424 fprintf(stderr
, Name
": could not allocate raid device\n");
3427 strncpy((char *) dev
->volume
, name
, MAX_RAID_SERIAL_LEN
);
3428 if (info
->level
== 1)
3429 array_blocks
= info_to_blocks_per_member(info
);
3431 array_blocks
= calc_array_size(info
->level
, info
->raid_disks
,
3432 info
->layout
, info
->chunk_size
,
3434 /* round array size down to closest MB */
3435 array_blocks
= (array_blocks
>> SECT_PER_MB_SHIFT
) << SECT_PER_MB_SHIFT
;
3437 dev
->size_low
= __cpu_to_le32((__u32
) array_blocks
);
3438 dev
->size_high
= __cpu_to_le32((__u32
) (array_blocks
>> 32));
3439 dev
->status
= __cpu_to_le32(0);
3440 dev
->reserved_blocks
= __cpu_to_le32(0);
3442 vol
->migr_state
= 0;
3443 set_migr_type(dev
, MIGR_INIT
);
3445 vol
->curr_migr_unit
= 0;
3446 map
= get_imsm_map(dev
, 0);
3447 map
->pba_of_lba0
= __cpu_to_le32(super
->create_offset
);
3448 map
->blocks_per_member
= __cpu_to_le32(info_to_blocks_per_member(info
));
3449 map
->blocks_per_strip
= __cpu_to_le16(info_to_blocks_per_strip(info
));
3450 map
->failed_disk_num
= ~0;
3451 map
->map_state
= info
->level
? IMSM_T_STATE_UNINITIALIZED
:
3452 IMSM_T_STATE_NORMAL
;
3455 if (info
->level
== 1 && info
->raid_disks
> 2) {
3458 fprintf(stderr
, Name
": imsm does not support more than 2 disks"
3459 "in a raid1 volume\n");
3463 map
->raid_level
= info
->level
;
3464 if (info
->level
== 10) {
3465 map
->raid_level
= 1;
3466 map
->num_domains
= info
->raid_disks
/ 2;
3467 } else if (info
->level
== 1)
3468 map
->num_domains
= info
->raid_disks
;
3470 map
->num_domains
= 1;
3472 num_data_stripes
= info_to_num_data_stripes(info
, map
->num_domains
);
3473 map
->num_data_stripes
= __cpu_to_le32(num_data_stripes
);
3475 map
->num_members
= info
->raid_disks
;
3476 for (i
= 0; i
< map
->num_members
; i
++) {
3477 /* initialized in add_to_super */
3478 set_imsm_ord_tbl_ent(map
, i
, IMSM_ORD_REBUILD
);
3480 mpb
->num_raid_devs
++;
3483 dv
->index
= super
->current_vol
;
3484 dv
->next
= super
->devlist
;
3485 super
->devlist
= dv
;
3487 imsm_update_version_info(super
);
3492 static int init_super_imsm(struct supertype
*st
, mdu_array_info_t
*info
,
3493 unsigned long long size
, char *name
,
3494 char *homehost
, int *uuid
)
3496 /* This is primarily called by Create when creating a new array.
3497 * We will then get add_to_super called for each component, and then
3498 * write_init_super called to write it out to each device.
3499 * For IMSM, Create can create on fresh devices or on a pre-existing
3501 * To create on a pre-existing array a different method will be called.
3502 * This one is just for fresh drives.
3504 struct intel_super
*super
;
3505 struct imsm_super
*mpb
;
3510 return init_super_imsm_volume(st
, info
, size
, name
, homehost
, uuid
);
3513 mpb_size
= disks_to_mpb_size(info
->nr_disks
);
3517 super
= alloc_super();
3518 if (super
&& posix_memalign(&super
->buf
, 512, mpb_size
) != 0) {
3523 fprintf(stderr
, Name
3524 ": %s could not allocate superblock\n", __func__
);
3527 memset(super
->buf
, 0, mpb_size
);
3529 mpb
->mpb_size
= __cpu_to_le32(mpb_size
);
3533 /* zeroing superblock */
3537 mpb
->attributes
= MPB_ATTRIB_CHECKSUM_VERIFY
;
3539 version
= (char *) mpb
->sig
;
3540 strcpy(version
, MPB_SIGNATURE
);
3541 version
+= strlen(MPB_SIGNATURE
);
3542 strcpy(version
, MPB_VERSION_RAID0
);
3548 static int add_to_super_imsm_volume(struct supertype
*st
, mdu_disk_info_t
*dk
,
3549 int fd
, char *devname
)
3551 struct intel_super
*super
= st
->sb
;
3552 struct imsm_super
*mpb
= super
->anchor
;
3554 struct imsm_dev
*dev
;
3555 struct imsm_map
*map
;
3558 dev
= get_imsm_dev(super
, super
->current_vol
);
3559 map
= get_imsm_map(dev
, 0);
3561 if (! (dk
->state
& (1<<MD_DISK_SYNC
))) {
3562 fprintf(stderr
, Name
": %s: Cannot add spare devices to IMSM volume\n",
3568 /* we're doing autolayout so grab the pre-marked (in
3569 * validate_geometry) raid_disk
3571 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
3572 if (dl
->raiddisk
== dk
->raid_disk
)
3575 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
3576 if (dl
->major
== dk
->major
&&
3577 dl
->minor
== dk
->minor
)
3582 fprintf(stderr
, Name
": %s is not a member of the same container\n", devname
);
3586 /* add a pristine spare to the metadata */
3587 if (dl
->index
< 0) {
3588 dl
->index
= super
->anchor
->num_disks
;
3589 super
->anchor
->num_disks
++;
3591 /* Check the device has not already been added */
3592 slot
= get_imsm_disk_slot(map
, dl
->index
);
3594 (get_imsm_ord_tbl_ent(dev
, slot
, -1) & IMSM_ORD_REBUILD
) == 0) {
3595 fprintf(stderr
, Name
": %s has been included in this array twice\n",
3599 set_imsm_ord_tbl_ent(map
, dk
->number
, dl
->index
);
3600 dl
->disk
.status
= CONFIGURED_DISK
;
3602 /* if we are creating the first raid device update the family number */
3603 if (super
->current_vol
== 0) {
3605 struct imsm_dev
*_dev
= __get_imsm_dev(mpb
, 0);
3606 struct imsm_disk
*_disk
= __get_imsm_disk(mpb
, dl
->index
);
3608 if (!_dev
|| !_disk
) {
3609 fprintf(stderr
, Name
": BUG mpb setup error\n");
3615 sum
+= __gen_imsm_checksum(mpb
);
3616 mpb
->family_num
= __cpu_to_le32(sum
);
3617 mpb
->orig_family_num
= mpb
->family_num
;
3624 static int add_to_super_imsm(struct supertype
*st
, mdu_disk_info_t
*dk
,
3625 int fd
, char *devname
)
3627 struct intel_super
*super
= st
->sb
;
3629 unsigned long long size
;
3634 /* If we are on an RAID enabled platform check that the disk is
3635 * attached to the raid controller.
3636 * We do not need to test disks attachment for container based additions,
3637 * they shall be already tested when container was created/assembled.
3639 if ((fd
!= -1) && !check_env("IMSM_NO_PLATFORM")) {
3640 struct sys_dev
*hba_name
;
3641 struct intel_hba
*hba
;
3643 hba_name
= find_disk_attached_hba(fd
, NULL
);
3646 Name
": %s is not attached to Intel(R) RAID controller.\n",
3647 devname
? : "disk");
3650 rv
= attach_hba_to_super(super
, hba_name
, devname
);
3653 fprintf(stderr
, Name
": %s is attached to Intel(R) %s RAID "
3654 "controller (%s),\n but the container is assigned to Intel(R) "
3655 "%s RAID controller (",
3657 get_sys_dev_type(hba_name
->type
),
3658 hba_name
->pci_id
? : "Err!",
3659 get_sys_dev_type(hba_name
->type
));
3663 fprintf(stderr
, "%s", hba
->pci_id
? : "Err!");
3665 fprintf(stderr
, ", ");
3669 fprintf(stderr
, ").\n"
3670 " Mixing devices attached to different controllers "
3671 "is not allowed.\n");
3672 free_sys_dev(&hba_name
);
3675 free_sys_dev(&hba_name
);
3678 if (super
->current_vol
>= 0)
3679 return add_to_super_imsm_volume(st
, dk
, fd
, devname
);
3682 dd
= malloc(sizeof(*dd
));
3685 Name
": malloc failed %s:%d.\n", __func__
, __LINE__
);
3688 memset(dd
, 0, sizeof(*dd
));
3689 dd
->major
= major(stb
.st_rdev
);
3690 dd
->minor
= minor(stb
.st_rdev
);
3692 dd
->devname
= devname
? strdup(devname
) : NULL
;
3695 dd
->action
= DISK_ADD
;
3696 rv
= imsm_read_serial(fd
, devname
, dd
->serial
);
3699 Name
": failed to retrieve scsi serial, aborting\n");
3704 get_dev_size(fd
, NULL
, &size
);
3706 serialcpy(dd
->disk
.serial
, dd
->serial
);
3707 dd
->disk
.total_blocks
= __cpu_to_le32(size
);
3708 dd
->disk
.status
= SPARE_DISK
;
3709 if (sysfs_disk_to_scsi_id(fd
, &id
) == 0)
3710 dd
->disk
.scsi_id
= __cpu_to_le32(id
);
3712 dd
->disk
.scsi_id
= __cpu_to_le32(0);
3714 if (st
->update_tail
) {
3715 dd
->next
= super
->disk_mgmt_list
;
3716 super
->disk_mgmt_list
= dd
;
3718 dd
->next
= super
->disks
;
3726 static int remove_from_super_imsm(struct supertype
*st
, mdu_disk_info_t
*dk
)
3728 struct intel_super
*super
= st
->sb
;
3731 /* remove from super works only in mdmon - for communication
3732 * manager - monitor. Check if communication memory buffer
3735 if (!st
->update_tail
) {
3737 Name
": %s shall be used in mdmon context only"
3738 "(line %d).\n", __func__
, __LINE__
);
3741 dd
= malloc(sizeof(*dd
));
3744 Name
": malloc failed %s:%d.\n", __func__
, __LINE__
);
3747 memset(dd
, 0, sizeof(*dd
));
3748 dd
->major
= dk
->major
;
3749 dd
->minor
= dk
->minor
;
3752 dd
->disk
.status
= SPARE_DISK
;
3753 dd
->action
= DISK_REMOVE
;
3755 dd
->next
= super
->disk_mgmt_list
;
3756 super
->disk_mgmt_list
= dd
;
3762 static int store_imsm_mpb(int fd
, struct imsm_super
*mpb
);
3766 struct imsm_super anchor
;
3767 } spare_record
__attribute__ ((aligned(512)));
3769 /* spare records have their own family number and do not have any defined raid
3772 static int write_super_imsm_spares(struct intel_super
*super
, int doclose
)
3774 struct imsm_super
*mpb
= super
->anchor
;
3775 struct imsm_super
*spare
= &spare_record
.anchor
;
3779 spare
->mpb_size
= __cpu_to_le32(sizeof(struct imsm_super
)),
3780 spare
->generation_num
= __cpu_to_le32(1UL),
3781 spare
->attributes
= MPB_ATTRIB_CHECKSUM_VERIFY
;
3782 spare
->num_disks
= 1,
3783 spare
->num_raid_devs
= 0,
3784 spare
->cache_size
= mpb
->cache_size
,
3785 spare
->pwr_cycle_count
= __cpu_to_le32(1),
3787 snprintf((char *) spare
->sig
, MAX_SIGNATURE_LENGTH
,
3788 MPB_SIGNATURE MPB_VERSION_RAID0
);
3790 for (d
= super
->disks
; d
; d
= d
->next
) {
3794 spare
->disk
[0] = d
->disk
;
3795 sum
= __gen_imsm_checksum(spare
);
3796 spare
->family_num
= __cpu_to_le32(sum
);
3797 spare
->orig_family_num
= 0;
3798 sum
= __gen_imsm_checksum(spare
);
3799 spare
->check_sum
= __cpu_to_le32(sum
);
3801 if (store_imsm_mpb(d
->fd
, spare
)) {
3802 fprintf(stderr
, "%s: failed for device %d:%d %s\n",
3803 __func__
, d
->major
, d
->minor
, strerror(errno
));
3815 static int write_super_imsm(struct supertype
*st
, int doclose
)
3817 struct intel_super
*super
= st
->sb
;
3818 struct imsm_super
*mpb
= super
->anchor
;
3824 __u32 mpb_size
= sizeof(struct imsm_super
) - sizeof(struct imsm_disk
);
3827 /* 'generation' is incremented everytime the metadata is written */
3828 generation
= __le32_to_cpu(mpb
->generation_num
);
3830 mpb
->generation_num
= __cpu_to_le32(generation
);
3832 /* fix up cases where previous mdadm releases failed to set
3835 if (mpb
->orig_family_num
== 0)
3836 mpb
->orig_family_num
= mpb
->family_num
;
3838 for (d
= super
->disks
; d
; d
= d
->next
) {
3842 mpb
->disk
[d
->index
] = d
->disk
;
3846 for (d
= super
->missing
; d
; d
= d
->next
) {
3847 mpb
->disk
[d
->index
] = d
->disk
;
3850 mpb
->num_disks
= num_disks
;
3851 mpb_size
+= sizeof(struct imsm_disk
) * mpb
->num_disks
;
3853 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
3854 struct imsm_dev
*dev
= __get_imsm_dev(mpb
, i
);
3855 struct imsm_dev
*dev2
= get_imsm_dev(super
, i
);
3857 imsm_copy_dev(dev
, dev2
);
3858 mpb_size
+= sizeof_imsm_dev(dev
, 0);
3861 mpb_size
+= __le32_to_cpu(mpb
->bbm_log_size
);
3862 mpb
->mpb_size
= __cpu_to_le32(mpb_size
);
3864 /* recalculate checksum */
3865 sum
= __gen_imsm_checksum(mpb
);
3866 mpb
->check_sum
= __cpu_to_le32(sum
);
3868 /* write the mpb for disks that compose raid devices */
3869 for (d
= super
->disks
; d
; d
= d
->next
) {
3872 if (store_imsm_mpb(d
->fd
, mpb
))
3873 fprintf(stderr
, "%s: failed for device %d:%d %s\n",
3874 __func__
, d
->major
, d
->minor
, strerror(errno
));
3882 return write_super_imsm_spares(super
, doclose
);
3888 static int create_array(struct supertype
*st
, int dev_idx
)
3891 struct imsm_update_create_array
*u
;
3892 struct intel_super
*super
= st
->sb
;
3893 struct imsm_dev
*dev
= get_imsm_dev(super
, dev_idx
);
3894 struct imsm_map
*map
= get_imsm_map(dev
, 0);
3895 struct disk_info
*inf
;
3896 struct imsm_disk
*disk
;
3899 len
= sizeof(*u
) - sizeof(*dev
) + sizeof_imsm_dev(dev
, 0) +
3900 sizeof(*inf
) * map
->num_members
;
3903 fprintf(stderr
, "%s: failed to allocate update buffer\n",
3908 u
->type
= update_create_array
;
3909 u
->dev_idx
= dev_idx
;
3910 imsm_copy_dev(&u
->dev
, dev
);
3911 inf
= get_disk_info(u
);
3912 for (i
= 0; i
< map
->num_members
; i
++) {
3913 int idx
= get_imsm_disk_idx(dev
, i
, -1);
3915 disk
= get_imsm_disk(super
, idx
);
3916 serialcpy(inf
[i
].serial
, disk
->serial
);
3918 append_metadata_update(st
, u
, len
);
3923 static int mgmt_disk(struct supertype
*st
)
3925 struct intel_super
*super
= st
->sb
;
3927 struct imsm_update_add_remove_disk
*u
;
3929 if (!super
->disk_mgmt_list
)
3935 fprintf(stderr
, "%s: failed to allocate update buffer\n",
3940 u
->type
= update_add_remove_disk
;
3941 append_metadata_update(st
, u
, len
);
3946 static int write_init_super_imsm(struct supertype
*st
)
3948 struct intel_super
*super
= st
->sb
;
3949 int current_vol
= super
->current_vol
;
3951 /* we are done with current_vol reset it to point st at the container */
3952 super
->current_vol
= -1;
3954 if (st
->update_tail
) {
3955 /* queue the recently created array / added disk
3956 * as a metadata update */
3959 /* determine if we are creating a volume or adding a disk */
3960 if (current_vol
< 0) {
3961 /* in the mgmt (add/remove) disk case we are running
3962 * in mdmon context, so don't close fd's
3964 return mgmt_disk(st
);
3966 rv
= create_array(st
, current_vol
);
3971 for (d
= super
->disks
; d
; d
= d
->next
)
3972 Kill(d
->devname
, NULL
, 0, 1, 1);
3973 return write_super_imsm(st
, 1);
3978 static int store_super_imsm(struct supertype
*st
, int fd
)
3980 struct intel_super
*super
= st
->sb
;
3981 struct imsm_super
*mpb
= super
? super
->anchor
: NULL
;
3987 return store_imsm_mpb(fd
, mpb
);
3993 static int imsm_bbm_log_size(struct imsm_super
*mpb
)
3995 return __le32_to_cpu(mpb
->bbm_log_size
);
3999 static int validate_geometry_imsm_container(struct supertype
*st
, int level
,
4000 int layout
, int raiddisks
, int chunk
,
4001 unsigned long long size
, char *dev
,
4002 unsigned long long *freesize
,
4006 unsigned long long ldsize
;
4007 const struct imsm_orom
*orom
;
4009 if (level
!= LEVEL_CONTAINER
)
4014 if (check_env("IMSM_NO_PLATFORM"))
4017 orom
= find_imsm_orom();
4018 if (orom
&& raiddisks
> orom
->tds
) {
4020 fprintf(stderr
, Name
": %d exceeds maximum number of"
4021 " platform supported disks: %d\n",
4022 raiddisks
, orom
->tds
);
4026 fd
= open(dev
, O_RDONLY
|O_EXCL
, 0);
4029 fprintf(stderr
, Name
": imsm: Cannot open %s: %s\n",
4030 dev
, strerror(errno
));
4033 if (!get_dev_size(fd
, dev
, &ldsize
)) {
4039 *freesize
= avail_size_imsm(st
, ldsize
>> 9);
4044 static unsigned long long find_size(struct extent
*e
, int *idx
, int num_extents
)
4046 const unsigned long long base_start
= e
[*idx
].start
;
4047 unsigned long long end
= base_start
+ e
[*idx
].size
;
4050 if (base_start
== end
)
4054 for (i
= *idx
; i
< num_extents
; i
++) {
4055 /* extend overlapping extents */
4056 if (e
[i
].start
>= base_start
&&
4057 e
[i
].start
<= end
) {
4060 if (e
[i
].start
+ e
[i
].size
> end
)
4061 end
= e
[i
].start
+ e
[i
].size
;
4062 } else if (e
[i
].start
> end
) {
4068 return end
- base_start
;
4071 static unsigned long long merge_extents(struct intel_super
*super
, int sum_extents
)
4073 /* build a composite disk with all known extents and generate a new
4074 * 'maxsize' given the "all disks in an array must share a common start
4075 * offset" constraint
4077 struct extent
*e
= calloc(sum_extents
, sizeof(*e
));
4081 unsigned long long pos
;
4082 unsigned long long start
= 0;
4083 unsigned long long maxsize
;
4084 unsigned long reserve
;
4089 /* coalesce and sort all extents. also, check to see if we need to
4090 * reserve space between member arrays
4093 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
4096 for (i
= 0; i
< dl
->extent_cnt
; i
++)
4099 qsort(e
, sum_extents
, sizeof(*e
), cmp_extent
);
4104 while (i
< sum_extents
) {
4105 e
[j
].start
= e
[i
].start
;
4106 e
[j
].size
= find_size(e
, &i
, sum_extents
);
4108 if (e
[j
-1].size
== 0)
4117 unsigned long long esize
;
4119 esize
= e
[i
].start
- pos
;
4120 if (esize
>= maxsize
) {
4125 pos
= e
[i
].start
+ e
[i
].size
;
4127 } while (e
[i
-1].size
);
4133 /* FIXME assumes volume at offset 0 is the first volume in a
4136 if (start_extent
> 0)
4137 reserve
= IMSM_RESERVED_SECTORS
; /* gap between raid regions */
4141 if (maxsize
< reserve
)
4144 super
->create_offset
= ~((__u32
) 0);
4145 if (start
+ reserve
> super
->create_offset
)
4146 return 0; /* start overflows create_offset */
4147 super
->create_offset
= start
+ reserve
;
4149 return maxsize
- reserve
;
4152 static int is_raid_level_supported(const struct imsm_orom
*orom
, int level
, int raiddisks
)
4154 if (level
< 0 || level
== 6 || level
== 4)
4157 /* if we have an orom prevent invalid raid levels */
4160 case 0: return imsm_orom_has_raid0(orom
);
4163 return imsm_orom_has_raid1e(orom
);
4164 return imsm_orom_has_raid1(orom
) && raiddisks
== 2;
4165 case 10: return imsm_orom_has_raid10(orom
) && raiddisks
== 4;
4166 case 5: return imsm_orom_has_raid5(orom
) && raiddisks
> 2;
4169 return 1; /* not on an Intel RAID platform so anything goes */
4174 #define pr_vrb(fmt, arg...) (void) (verbose && fprintf(stderr, Name fmt, ##arg))
4176 validate_geometry_imsm_orom(struct intel_super
*super
, int level
, int layout
,
4177 int raiddisks
, int *chunk
, int verbose
)
4179 if (!is_raid_level_supported(super
->orom
, level
, raiddisks
)) {
4180 pr_vrb(": platform does not support raid%d with %d disk%s\n",
4181 level
, raiddisks
, raiddisks
> 1 ? "s" : "");
4184 if (super
->orom
&& level
!= 1) {
4185 if (chunk
&& (*chunk
== 0 || *chunk
== UnSet
))
4186 *chunk
= imsm_orom_default_chunk(super
->orom
);
4187 else if (chunk
&& !imsm_orom_has_chunk(super
->orom
, *chunk
)) {
4188 pr_vrb(": platform does not support a chunk size of: "
4193 if (layout
!= imsm_level_to_layout(level
)) {
4195 pr_vrb(": imsm raid 5 only supports the left-asymmetric layout\n");
4196 else if (level
== 10)
4197 pr_vrb(": imsm raid 10 only supports the n2 layout\n");
4199 pr_vrb(": imsm unknown layout %#x for this raid level %d\n",
4207 /* validate_geometry_imsm_volume - lifted from validate_geometry_ddf_bvd
4208 * FIX ME add ahci details
4210 static int validate_geometry_imsm_volume(struct supertype
*st
, int level
,
4211 int layout
, int raiddisks
, int *chunk
,
4212 unsigned long long size
, char *dev
,
4213 unsigned long long *freesize
,
4217 struct intel_super
*super
= st
->sb
;
4218 struct imsm_super
*mpb
= super
->anchor
;
4220 unsigned long long pos
= 0;
4221 unsigned long long maxsize
;
4225 /* We must have the container info already read in. */
4229 if (!validate_geometry_imsm_orom(super
, level
, layout
, raiddisks
, chunk
, verbose
))
4233 /* General test: make sure there is space for
4234 * 'raiddisks' device extents of size 'size' at a given
4237 unsigned long long minsize
= size
;
4238 unsigned long long start_offset
= MaxSector
;
4241 minsize
= MPB_SECTOR_CNT
+ IMSM_RESERVED_SECTORS
;
4242 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
4247 e
= get_extents(super
, dl
);
4250 unsigned long long esize
;
4251 esize
= e
[i
].start
- pos
;
4252 if (esize
>= minsize
)
4254 if (found
&& start_offset
== MaxSector
) {
4257 } else if (found
&& pos
!= start_offset
) {
4261 pos
= e
[i
].start
+ e
[i
].size
;
4263 } while (e
[i
-1].size
);
4268 if (dcnt
< raiddisks
) {
4270 fprintf(stderr
, Name
": imsm: Not enough "
4271 "devices with space for this array "
4279 /* This device must be a member of the set */
4280 if (stat(dev
, &stb
) < 0)
4282 if ((S_IFMT
& stb
.st_mode
) != S_IFBLK
)
4284 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
4285 if (dl
->major
== (int)major(stb
.st_rdev
) &&
4286 dl
->minor
== (int)minor(stb
.st_rdev
))
4291 fprintf(stderr
, Name
": %s is not in the "
4292 "same imsm set\n", dev
);
4294 } else if (super
->orom
&& dl
->index
< 0 && mpb
->num_raid_devs
) {
4295 /* If a volume is present then the current creation attempt
4296 * cannot incorporate new spares because the orom may not
4297 * understand this configuration (all member disks must be
4298 * members of each array in the container).
4300 fprintf(stderr
, Name
": %s is a spare and a volume"
4301 " is already defined for this container\n", dev
);
4302 fprintf(stderr
, Name
": The option-rom requires all member"
4303 " disks to be a member of all volumes\n");
4307 /* retrieve the largest free space block */
4308 e
= get_extents(super
, dl
);
4313 unsigned long long esize
;
4315 esize
= e
[i
].start
- pos
;
4316 if (esize
>= maxsize
)
4318 pos
= e
[i
].start
+ e
[i
].size
;
4320 } while (e
[i
-1].size
);
4325 fprintf(stderr
, Name
": unable to determine free space for: %s\n",
4329 if (maxsize
< size
) {
4331 fprintf(stderr
, Name
": %s not enough space (%llu < %llu)\n",
4332 dev
, maxsize
, size
);
4336 /* count total number of extents for merge */
4338 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
4340 i
+= dl
->extent_cnt
;
4342 maxsize
= merge_extents(super
, i
);
4343 if (maxsize
< size
|| maxsize
== 0) {
4345 fprintf(stderr
, Name
": not enough space after merge (%llu < %llu)\n",
4350 *freesize
= maxsize
;
4355 static int reserve_space(struct supertype
*st
, int raiddisks
,
4356 unsigned long long size
, int chunk
,
4357 unsigned long long *freesize
)
4359 struct intel_super
*super
= st
->sb
;
4360 struct imsm_super
*mpb
= super
->anchor
;
4365 unsigned long long maxsize
;
4366 unsigned long long minsize
;
4370 /* find the largest common start free region of the possible disks */
4374 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
4380 /* don't activate new spares if we are orom constrained
4381 * and there is already a volume active in the container
4383 if (super
->orom
&& dl
->index
< 0 && mpb
->num_raid_devs
)
4386 e
= get_extents(super
, dl
);
4389 for (i
= 1; e
[i
-1].size
; i
++)
4397 maxsize
= merge_extents(super
, extent_cnt
);
4401 minsize
= chunk
* 2;
4403 if (cnt
< raiddisks
||
4404 (super
->orom
&& used
&& used
!= raiddisks
) ||
4405 maxsize
< minsize
||
4407 fprintf(stderr
, Name
": not enough devices with space to create array.\n");
4408 return 0; /* No enough free spaces large enough */
4420 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
4422 dl
->raiddisk
= cnt
++;
4429 static int validate_geometry_imsm(struct supertype
*st
, int level
, int layout
,
4430 int raiddisks
, int *chunk
, unsigned long long size
,
4431 char *dev
, unsigned long long *freesize
,
4438 /* if given unused devices create a container
4439 * if given given devices in a container create a member volume
4441 if (level
== LEVEL_CONTAINER
) {
4442 /* Must be a fresh device to add to a container */
4443 return validate_geometry_imsm_container(st
, level
, layout
,
4445 chunk
?*chunk
:0, size
,
4451 if (st
->sb
&& freesize
) {
4452 /* we are being asked to automatically layout a
4453 * new volume based on the current contents of
4454 * the container. If the the parameters can be
4455 * satisfied reserve_space will record the disks,
4456 * start offset, and size of the volume to be
4457 * created. add_to_super and getinfo_super
4458 * detect when autolayout is in progress.
4460 if (!validate_geometry_imsm_orom(st
->sb
, level
, layout
,
4464 return reserve_space(st
, raiddisks
, size
,
4465 chunk
?*chunk
:0, freesize
);
4470 /* creating in a given container */
4471 return validate_geometry_imsm_volume(st
, level
, layout
,
4472 raiddisks
, chunk
, size
,
4473 dev
, freesize
, verbose
);
4476 /* This device needs to be a device in an 'imsm' container */
4477 fd
= open(dev
, O_RDONLY
|O_EXCL
, 0);
4481 Name
": Cannot create this array on device %s\n",
4486 if (errno
!= EBUSY
|| (fd
= open(dev
, O_RDONLY
, 0)) < 0) {
4488 fprintf(stderr
, Name
": Cannot open %s: %s\n",
4489 dev
, strerror(errno
));
4492 /* Well, it is in use by someone, maybe an 'imsm' container. */
4493 cfd
= open_container(fd
);
4497 fprintf(stderr
, Name
": Cannot use %s: It is busy\n",
4501 sra
= sysfs_read(cfd
, 0, GET_VERSION
);
4502 if (sra
&& sra
->array
.major_version
== -1 &&
4503 strcmp(sra
->text_version
, "imsm") == 0)
4507 /* This is a member of a imsm container. Load the container
4508 * and try to create a volume
4510 struct intel_super
*super
;
4512 if (load_super_imsm_all(st
, cfd
, (void **) &super
, NULL
) == 0) {
4514 st
->container_dev
= fd2devnum(cfd
);
4516 return validate_geometry_imsm_volume(st
, level
, layout
,
4524 fprintf(stderr
, Name
": failed container membership check\n");
4530 static void default_geometry_imsm(struct supertype
*st
, int *level
, int *layout
, int *chunk
)
4532 struct intel_super
*super
= st
->sb
;
4534 if (level
&& *level
== UnSet
)
4535 *level
= LEVEL_CONTAINER
;
4537 if (level
&& layout
&& *layout
== UnSet
)
4538 *layout
= imsm_level_to_layout(*level
);
4540 if (chunk
&& (*chunk
== UnSet
|| *chunk
== 0) &&
4541 super
&& super
->orom
)
4542 *chunk
= imsm_orom_default_chunk(super
->orom
);
4545 static void handle_missing(struct intel_super
*super
, struct imsm_dev
*dev
);
4547 static int kill_subarray_imsm(struct supertype
*st
)
4549 /* remove the subarray currently referenced by ->current_vol */
4551 struct intel_dev
**dp
;
4552 struct intel_super
*super
= st
->sb
;
4553 __u8 current_vol
= super
->current_vol
;
4554 struct imsm_super
*mpb
= super
->anchor
;
4556 if (super
->current_vol
< 0)
4558 super
->current_vol
= -1; /* invalidate subarray cursor */
4560 /* block deletions that would change the uuid of active subarrays
4562 * FIXME when immutable ids are available, but note that we'll
4563 * also need to fixup the invalidated/active subarray indexes in
4566 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
4569 if (i
< current_vol
)
4571 sprintf(subarray
, "%u", i
);
4572 if (is_subarray_active(subarray
, st
->devname
)) {
4574 Name
": deleting subarray-%d would change the UUID of active subarray-%d, aborting\n",
4581 if (st
->update_tail
) {
4582 struct imsm_update_kill_array
*u
= malloc(sizeof(*u
));
4586 u
->type
= update_kill_array
;
4587 u
->dev_idx
= current_vol
;
4588 append_metadata_update(st
, u
, sizeof(*u
));
4593 for (dp
= &super
->devlist
; *dp
;)
4594 if ((*dp
)->index
== current_vol
) {
4597 handle_missing(super
, (*dp
)->dev
);
4598 if ((*dp
)->index
> current_vol
)
4603 /* no more raid devices, all active components are now spares,
4604 * but of course failed are still failed
4606 if (--mpb
->num_raid_devs
== 0) {
4609 for (d
= super
->disks
; d
; d
= d
->next
)
4610 if (d
->index
> -2) {
4612 d
->disk
.status
= SPARE_DISK
;
4616 super
->updates_pending
++;
4621 static int update_subarray_imsm(struct supertype
*st
, char *subarray
,
4622 char *update
, struct mddev_ident
*ident
)
4624 /* update the subarray currently referenced by ->current_vol */
4625 struct intel_super
*super
= st
->sb
;
4626 struct imsm_super
*mpb
= super
->anchor
;
4628 if (strcmp(update
, "name") == 0) {
4629 char *name
= ident
->name
;
4633 if (is_subarray_active(subarray
, st
->devname
)) {
4635 Name
": Unable to update name of active subarray\n");
4639 if (!check_name(super
, name
, 0))
4642 vol
= strtoul(subarray
, &ep
, 10);
4643 if (*ep
!= '\0' || vol
>= super
->anchor
->num_raid_devs
)
4646 if (st
->update_tail
) {
4647 struct imsm_update_rename_array
*u
= malloc(sizeof(*u
));
4651 u
->type
= update_rename_array
;
4653 snprintf((char *) u
->name
, MAX_RAID_SERIAL_LEN
, "%s", name
);
4654 append_metadata_update(st
, u
, sizeof(*u
));
4656 struct imsm_dev
*dev
;
4659 dev
= get_imsm_dev(super
, vol
);
4660 snprintf((char *) dev
->volume
, MAX_RAID_SERIAL_LEN
, "%s", name
);
4661 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
4662 dev
= get_imsm_dev(super
, i
);
4663 handle_missing(super
, dev
);
4665 super
->updates_pending
++;
4673 static int is_gen_migration(struct imsm_dev
*dev
)
4675 if (!dev
->vol
.migr_state
)
4678 if (migr_type(dev
) == MIGR_GEN_MIGR
)
4683 #endif /* MDASSEMBLE */
4685 static int is_rebuilding(struct imsm_dev
*dev
)
4687 struct imsm_map
*migr_map
;
4689 if (!dev
->vol
.migr_state
)
4692 if (migr_type(dev
) != MIGR_REBUILD
)
4695 migr_map
= get_imsm_map(dev
, 1);
4697 if (migr_map
->map_state
== IMSM_T_STATE_DEGRADED
)
4703 static void update_recovery_start(struct imsm_dev
*dev
, struct mdinfo
*array
)
4705 struct mdinfo
*rebuild
= NULL
;
4709 if (!is_rebuilding(dev
))
4712 /* Find the rebuild target, but punt on the dual rebuild case */
4713 for (d
= array
->devs
; d
; d
= d
->next
)
4714 if (d
->recovery_start
== 0) {
4721 /* (?) none of the disks are marked with
4722 * IMSM_ORD_REBUILD, so assume they are missing and the
4723 * disk_ord_tbl was not correctly updated
4725 dprintf("%s: failed to locate out-of-sync disk\n", __func__
);
4729 units
= __le32_to_cpu(dev
->vol
.curr_migr_unit
);
4730 rebuild
->recovery_start
= units
* blocks_per_migr_unit(dev
);
4734 static struct mdinfo
*container_content_imsm(struct supertype
*st
, char *subarray
)
4736 /* Given a container loaded by load_super_imsm_all,
4737 * extract information about all the arrays into
4739 * If 'subarray' is given, just extract info about that array.
4741 * For each imsm_dev create an mdinfo, fill it in,
4742 * then look for matching devices in super->disks
4743 * and create appropriate device mdinfo.
4745 struct intel_super
*super
= st
->sb
;
4746 struct imsm_super
*mpb
= super
->anchor
;
4747 struct mdinfo
*rest
= NULL
;
4751 int spare_disks
= 0;
4753 /* check for bad blocks */
4754 if (imsm_bbm_log_size(super
->anchor
))
4757 /* count spare devices, not used in maps
4759 for (d
= super
->disks
; d
; d
= d
->next
)
4763 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
4764 struct imsm_dev
*dev
;
4765 struct imsm_map
*map
;
4766 struct imsm_map
*map2
;
4767 struct mdinfo
*this;
4772 (i
!= strtoul(subarray
, &ep
, 10) || *ep
!= '\0'))
4775 dev
= get_imsm_dev(super
, i
);
4776 map
= get_imsm_map(dev
, 0);
4777 map2
= get_imsm_map(dev
, 1);
4779 /* do not publish arrays that are in the middle of an
4780 * unsupported migration
4782 if (dev
->vol
.migr_state
&&
4783 (migr_type(dev
) == MIGR_STATE_CHANGE
)) {
4784 fprintf(stderr
, Name
": cannot assemble volume '%.16s':"
4785 " unsupported migration in progress\n",
4790 this = malloc(sizeof(*this));
4792 fprintf(stderr
, Name
": failed to allocate %zu bytes\n",
4796 memset(this, 0, sizeof(*this));
4799 super
->current_vol
= i
;
4800 getinfo_super_imsm_volume(st
, this, NULL
);
4801 for (slot
= 0 ; slot
< map
->num_members
; slot
++) {
4802 unsigned long long recovery_start
;
4803 struct mdinfo
*info_d
;
4810 idx
= get_imsm_disk_idx(dev
, slot
, 0);
4811 ord
= get_imsm_ord_tbl_ent(dev
, slot
, -1);
4812 for (d
= super
->disks
; d
; d
= d
->next
)
4813 if (d
->index
== idx
)
4816 recovery_start
= MaxSector
;
4819 if (d
&& is_failed(&d
->disk
))
4821 if (ord
& IMSM_ORD_REBUILD
)
4825 * if we skip some disks the array will be assmebled degraded;
4826 * reset resync start to avoid a dirty-degraded
4827 * situation when performing the intial sync
4829 * FIXME handle dirty degraded
4831 if ((skip
|| recovery_start
== 0) && !dev
->vol
.dirty
)
4832 this->resync_start
= MaxSector
;
4836 info_d
= calloc(1, sizeof(*info_d
));
4838 fprintf(stderr
, Name
": failed to allocate disk"
4839 " for volume %.16s\n", dev
->volume
);
4840 info_d
= this->devs
;
4842 struct mdinfo
*d
= info_d
->next
;
4851 info_d
->next
= this->devs
;
4852 this->devs
= info_d
;
4854 info_d
->disk
.number
= d
->index
;
4855 info_d
->disk
.major
= d
->major
;
4856 info_d
->disk
.minor
= d
->minor
;
4857 info_d
->disk
.raid_disk
= slot
;
4858 info_d
->recovery_start
= recovery_start
;
4860 if (slot
< map2
->num_members
)
4861 info_d
->disk
.state
= (1 << MD_DISK_ACTIVE
);
4863 this->array
.spare_disks
++;
4865 if (slot
< map
->num_members
)
4866 info_d
->disk
.state
= (1 << MD_DISK_ACTIVE
);
4868 this->array
.spare_disks
++;
4870 if (info_d
->recovery_start
== MaxSector
)
4871 this->array
.working_disks
++;
4873 info_d
->events
= __le32_to_cpu(mpb
->generation_num
);
4874 info_d
->data_offset
= __le32_to_cpu(map
->pba_of_lba0
);
4875 info_d
->component_size
= __le32_to_cpu(map
->blocks_per_member
);
4877 /* now that the disk list is up-to-date fixup recovery_start */
4878 update_recovery_start(dev
, this);
4879 this->array
.spare_disks
+= spare_disks
;
4883 /* if array has bad blocks, set suitable bit in array status */
4885 rest
->array
.state
|= (1<<MD_SB_BBM_ERRORS
);
4891 static __u8
imsm_check_degraded(struct intel_super
*super
, struct imsm_dev
*dev
, int failed
)
4893 struct imsm_map
*map
= get_imsm_map(dev
, 0);
4896 return map
->map_state
== IMSM_T_STATE_UNINITIALIZED
?
4897 IMSM_T_STATE_UNINITIALIZED
: IMSM_T_STATE_NORMAL
;
4899 switch (get_imsm_raid_level(map
)) {
4901 return IMSM_T_STATE_FAILED
;
4904 if (failed
< map
->num_members
)
4905 return IMSM_T_STATE_DEGRADED
;
4907 return IMSM_T_STATE_FAILED
;
4912 * check to see if any mirrors have failed, otherwise we
4913 * are degraded. Even numbered slots are mirrored on
4917 /* gcc -Os complains that this is unused */
4918 int insync
= insync
;
4920 for (i
= 0; i
< map
->num_members
; i
++) {
4921 __u32 ord
= get_imsm_ord_tbl_ent(dev
, i
, -1);
4922 int idx
= ord_to_idx(ord
);
4923 struct imsm_disk
*disk
;
4925 /* reset the potential in-sync count on even-numbered
4926 * slots. num_copies is always 2 for imsm raid10
4931 disk
= get_imsm_disk(super
, idx
);
4932 if (!disk
|| is_failed(disk
) || ord
& IMSM_ORD_REBUILD
)
4935 /* no in-sync disks left in this mirror the
4939 return IMSM_T_STATE_FAILED
;
4942 return IMSM_T_STATE_DEGRADED
;
4946 return IMSM_T_STATE_DEGRADED
;
4948 return IMSM_T_STATE_FAILED
;
4954 return map
->map_state
;
4957 static int imsm_count_failed(struct intel_super
*super
, struct imsm_dev
*dev
)
4961 struct imsm_disk
*disk
;
4962 struct imsm_map
*map
= get_imsm_map(dev
, 0);
4963 struct imsm_map
*prev
= get_imsm_map(dev
, dev
->vol
.migr_state
);
4967 /* at the beginning of migration we set IMSM_ORD_REBUILD on
4968 * disks that are being rebuilt. New failures are recorded to
4969 * map[0]. So we look through all the disks we started with and
4970 * see if any failures are still present, or if any new ones
4973 * FIXME add support for online capacity expansion and
4974 * raid-level-migration
4976 for (i
= 0; i
< prev
->num_members
; i
++) {
4977 ord
= __le32_to_cpu(prev
->disk_ord_tbl
[i
]);
4978 ord
|= __le32_to_cpu(map
->disk_ord_tbl
[i
]);
4979 idx
= ord_to_idx(ord
);
4981 disk
= get_imsm_disk(super
, idx
);
4982 if (!disk
|| is_failed(disk
) || ord
& IMSM_ORD_REBUILD
)
4990 static int imsm_open_new(struct supertype
*c
, struct active_array
*a
,
4993 struct intel_super
*super
= c
->sb
;
4994 struct imsm_super
*mpb
= super
->anchor
;
4996 if (atoi(inst
) >= mpb
->num_raid_devs
) {
4997 fprintf(stderr
, "%s: subarry index %d, out of range\n",
4998 __func__
, atoi(inst
));
5002 dprintf("imsm: open_new %s\n", inst
);
5003 a
->info
.container_member
= atoi(inst
);
5007 static int is_resyncing(struct imsm_dev
*dev
)
5009 struct imsm_map
*migr_map
;
5011 if (!dev
->vol
.migr_state
)
5014 if (migr_type(dev
) == MIGR_INIT
||
5015 migr_type(dev
) == MIGR_REPAIR
)
5018 if (migr_type(dev
) == MIGR_GEN_MIGR
)
5021 migr_map
= get_imsm_map(dev
, 1);
5023 if ((migr_map
->map_state
== IMSM_T_STATE_NORMAL
) &&
5024 (dev
->vol
.migr_type
!= MIGR_GEN_MIGR
))
5030 /* return true if we recorded new information */
5031 static int mark_failure(struct imsm_dev
*dev
, struct imsm_disk
*disk
, int idx
)
5035 struct imsm_map
*map
;
5037 /* new failures are always set in map[0] */
5038 map
= get_imsm_map(dev
, 0);
5040 slot
= get_imsm_disk_slot(map
, idx
);
5044 ord
= __le32_to_cpu(map
->disk_ord_tbl
[slot
]);
5045 if (is_failed(disk
) && (ord
& IMSM_ORD_REBUILD
))
5048 disk
->status
|= FAILED_DISK
;
5049 disk
->status
&= ~CONFIGURED_DISK
;
5050 set_imsm_ord_tbl_ent(map
, slot
, idx
| IMSM_ORD_REBUILD
);
5051 if (map
->failed_disk_num
== 0xff)
5052 map
->failed_disk_num
= slot
;
5056 static void mark_missing(struct imsm_dev
*dev
, struct imsm_disk
*disk
, int idx
)
5058 mark_failure(dev
, disk
, idx
);
5060 if (disk
->scsi_id
== __cpu_to_le32(~(__u32
)0))
5063 disk
->scsi_id
= __cpu_to_le32(~(__u32
)0);
5064 memmove(&disk
->serial
[0], &disk
->serial
[1], MAX_RAID_SERIAL_LEN
- 1);
5067 static void handle_missing(struct intel_super
*super
, struct imsm_dev
*dev
)
5073 if (!super
->missing
)
5075 failed
= imsm_count_failed(super
, dev
);
5076 map_state
= imsm_check_degraded(super
, dev
, failed
);
5078 dprintf("imsm: mark missing\n");
5079 end_migration(dev
, map_state
);
5080 for (dl
= super
->missing
; dl
; dl
= dl
->next
)
5081 mark_missing(dev
, &dl
->disk
, dl
->index
);
5082 super
->updates_pending
++;
5085 static unsigned long long imsm_set_array_size(struct imsm_dev
*dev
)
5087 int used_disks
= imsm_num_data_members(dev
, 0);
5088 unsigned long long array_blocks
;
5089 struct imsm_map
*map
;
5091 if (used_disks
== 0) {
5092 /* when problems occures
5093 * return current array_blocks value
5095 array_blocks
= __le32_to_cpu(dev
->size_high
);
5096 array_blocks
= array_blocks
<< 32;
5097 array_blocks
+= __le32_to_cpu(dev
->size_low
);
5099 return array_blocks
;
5102 /* set array size in metadata
5104 map
= get_imsm_map(dev
, 0);
5105 array_blocks
= map
->blocks_per_member
* used_disks
;
5107 /* round array size down to closest MB
5109 array_blocks
= (array_blocks
>> SECT_PER_MB_SHIFT
) << SECT_PER_MB_SHIFT
;
5110 dev
->size_low
= __cpu_to_le32((__u32
)array_blocks
);
5111 dev
->size_high
= __cpu_to_le32((__u32
)(array_blocks
>> 32));
5113 return array_blocks
;
5116 static void imsm_set_disk(struct active_array
*a
, int n
, int state
);
5118 static void imsm_progress_container_reshape(struct intel_super
*super
)
5120 /* if no device has a migr_state, but some device has a
5121 * different number of members than the previous device, start
5122 * changing the number of devices in this device to match
5125 struct imsm_super
*mpb
= super
->anchor
;
5126 int prev_disks
= -1;
5130 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
5131 struct imsm_dev
*dev
= get_imsm_dev(super
, i
);
5132 struct imsm_map
*map
= get_imsm_map(dev
, 0);
5133 struct imsm_map
*map2
;
5134 int prev_num_members
;
5136 if (dev
->vol
.migr_state
)
5139 if (prev_disks
== -1)
5140 prev_disks
= map
->num_members
;
5141 if (prev_disks
== map
->num_members
)
5144 /* OK, this array needs to enter reshape mode.
5145 * i.e it needs a migr_state
5148 copy_map_size
= sizeof_imsm_map(map
);
5149 prev_num_members
= map
->num_members
;
5150 map
->num_members
= prev_disks
;
5151 dev
->vol
.migr_state
= 1;
5152 dev
->vol
.curr_migr_unit
= 0;
5153 dev
->vol
.migr_type
= MIGR_GEN_MIGR
;
5154 for (i
= prev_num_members
;
5155 i
< map
->num_members
; i
++)
5156 set_imsm_ord_tbl_ent(map
, i
, i
);
5157 map2
= get_imsm_map(dev
, 1);
5158 /* Copy the current map */
5159 memcpy(map2
, map
, copy_map_size
);
5160 map2
->num_members
= prev_num_members
;
5162 imsm_set_array_size(dev
);
5163 super
->updates_pending
++;
5167 /* Handle dirty -> clean transititions, resync and reshape. Degraded and rebuild
5168 * states are handled in imsm_set_disk() with one exception, when a
5169 * resync is stopped due to a new failure this routine will set the
5170 * 'degraded' state for the array.
5172 static int imsm_set_array_state(struct active_array
*a
, int consistent
)
5174 int inst
= a
->info
.container_member
;
5175 struct intel_super
*super
= a
->container
->sb
;
5176 struct imsm_dev
*dev
= get_imsm_dev(super
, inst
);
5177 struct imsm_map
*map
= get_imsm_map(dev
, 0);
5178 int failed
= imsm_count_failed(super
, dev
);
5179 __u8 map_state
= imsm_check_degraded(super
, dev
, failed
);
5180 __u32 blocks_per_unit
;
5182 if (dev
->vol
.migr_state
&&
5183 dev
->vol
.migr_type
== MIGR_GEN_MIGR
) {
5184 /* array state change is blocked due to reshape action
5186 * - abort the reshape (if last_checkpoint is 0 and action!= reshape)
5187 * - finish the reshape (if last_checkpoint is big and action != reshape)
5188 * - update curr_migr_unit
5190 if (a
->curr_action
== reshape
) {
5191 /* still reshaping, maybe update curr_migr_unit */
5192 long long blocks_per_unit
= blocks_per_migr_unit(dev
);
5193 long long unit
= a
->last_checkpoint
;
5194 if (blocks_per_unit
) {
5195 unit
/= blocks_per_unit
;
5197 __le32_to_cpu(dev
->vol
.curr_migr_unit
)) {
5198 dev
->vol
.curr_migr_unit
=
5199 __cpu_to_le32(unit
);
5200 super
->updates_pending
++;
5205 if (a
->last_checkpoint
== 0 && a
->prev_action
== reshape
) {
5206 /* for some reason we aborted the reshape.
5209 struct imsm_map
*map2
= get_imsm_map(dev
, 1);
5210 dev
->vol
.migr_state
= 0;
5211 dev
->vol
.migr_type
= 0;
5212 dev
->vol
.curr_migr_unit
= 0;
5213 memcpy(map
, map2
, sizeof_imsm_map(map2
));
5214 super
->updates_pending
++;
5216 if (a
->last_checkpoint
>= a
->info
.component_size
) {
5217 unsigned long long array_blocks
;
5221 used_disks
= imsm_num_data_members(dev
, 0);
5222 if (used_disks
> 0) {
5224 map
->blocks_per_member
*
5226 /* round array size down to closest MB
5228 array_blocks
= (array_blocks
5229 >> SECT_PER_MB_SHIFT
)
5230 << SECT_PER_MB_SHIFT
;
5231 a
->info
.custom_array_size
= array_blocks
;
5232 /* encourage manager to update array
5236 a
->check_reshape
= 1;
5238 /* finalize online capacity expansion/reshape */
5239 for (mdi
= a
->info
.devs
; mdi
; mdi
= mdi
->next
)
5241 mdi
->disk
.raid_disk
,
5244 imsm_progress_container_reshape(super
);
5249 /* before we activate this array handle any missing disks */
5250 if (consistent
== 2)
5251 handle_missing(super
, dev
);
5253 if (consistent
== 2 &&
5254 (!is_resync_complete(&a
->info
) ||
5255 map_state
!= IMSM_T_STATE_NORMAL
||
5256 dev
->vol
.migr_state
))
5259 if (is_resync_complete(&a
->info
)) {
5260 /* complete intialization / resync,
5261 * recovery and interrupted recovery is completed in
5264 if (is_resyncing(dev
)) {
5265 dprintf("imsm: mark resync done\n");
5266 end_migration(dev
, map_state
);
5267 super
->updates_pending
++;
5268 a
->last_checkpoint
= 0;
5270 } else if (!is_resyncing(dev
) && !failed
) {
5271 /* mark the start of the init process if nothing is failed */
5272 dprintf("imsm: mark resync start\n");
5273 if (map
->map_state
== IMSM_T_STATE_UNINITIALIZED
)
5274 migrate(dev
, IMSM_T_STATE_NORMAL
, MIGR_INIT
);
5276 migrate(dev
, IMSM_T_STATE_NORMAL
, MIGR_REPAIR
);
5277 super
->updates_pending
++;
5280 /* check if we can update curr_migr_unit from resync_start, recovery_start */
5281 blocks_per_unit
= blocks_per_migr_unit(dev
);
5282 if (blocks_per_unit
) {
5286 units
= a
->last_checkpoint
/ blocks_per_unit
;
5289 /* check that we did not overflow 32-bits, and that
5290 * curr_migr_unit needs updating
5292 if (units32
== units
&&
5293 __le32_to_cpu(dev
->vol
.curr_migr_unit
) != units32
) {
5294 dprintf("imsm: mark checkpoint (%u)\n", units32
);
5295 dev
->vol
.curr_migr_unit
= __cpu_to_le32(units32
);
5296 super
->updates_pending
++;
5300 /* mark dirty / clean */
5301 if (dev
->vol
.dirty
!= !consistent
) {
5302 dprintf("imsm: mark '%s'\n", consistent
? "clean" : "dirty");
5307 super
->updates_pending
++;
5313 static void imsm_set_disk(struct active_array
*a
, int n
, int state
)
5315 int inst
= a
->info
.container_member
;
5316 struct intel_super
*super
= a
->container
->sb
;
5317 struct imsm_dev
*dev
= get_imsm_dev(super
, inst
);
5318 struct imsm_map
*map
= get_imsm_map(dev
, 0);
5319 struct imsm_disk
*disk
;
5324 if (n
> map
->num_members
)
5325 fprintf(stderr
, "imsm: set_disk %d out of range 0..%d\n",
5326 n
, map
->num_members
- 1);
5331 dprintf("imsm: set_disk %d:%x\n", n
, state
);
5333 ord
= get_imsm_ord_tbl_ent(dev
, n
, -1);
5334 disk
= get_imsm_disk(super
, ord_to_idx(ord
));
5336 /* check for new failures */
5337 if (state
& DS_FAULTY
) {
5338 if (mark_failure(dev
, disk
, ord_to_idx(ord
)))
5339 super
->updates_pending
++;
5342 /* check if in_sync */
5343 if (state
& DS_INSYNC
&& ord
& IMSM_ORD_REBUILD
&& is_rebuilding(dev
)) {
5344 struct imsm_map
*migr_map
= get_imsm_map(dev
, 1);
5346 set_imsm_ord_tbl_ent(migr_map
, n
, ord_to_idx(ord
));
5347 super
->updates_pending
++;
5350 failed
= imsm_count_failed(super
, dev
);
5351 map_state
= imsm_check_degraded(super
, dev
, failed
);
5353 /* check if recovery complete, newly degraded, or failed */
5354 if (map_state
== IMSM_T_STATE_NORMAL
&& is_rebuilding(dev
)) {
5355 end_migration(dev
, map_state
);
5356 map
= get_imsm_map(dev
, 0);
5357 map
->failed_disk_num
= ~0;
5358 super
->updates_pending
++;
5359 a
->last_checkpoint
= 0;
5360 } else if (map_state
== IMSM_T_STATE_DEGRADED
&&
5361 map
->map_state
!= map_state
&&
5362 !dev
->vol
.migr_state
) {
5363 dprintf("imsm: mark degraded\n");
5364 map
->map_state
= map_state
;
5365 super
->updates_pending
++;
5366 a
->last_checkpoint
= 0;
5367 } else if (map_state
== IMSM_T_STATE_FAILED
&&
5368 map
->map_state
!= map_state
) {
5369 dprintf("imsm: mark failed\n");
5370 end_migration(dev
, map_state
);
5371 super
->updates_pending
++;
5372 a
->last_checkpoint
= 0;
5373 } else if (is_gen_migration(dev
)) {
5374 dprintf("imsm: Detected General Migration in state: ");
5375 if (map_state
== IMSM_T_STATE_NORMAL
) {
5376 end_migration(dev
, map_state
);
5377 map
= get_imsm_map(dev
, 0);
5378 map
->failed_disk_num
= ~0;
5379 dprintf("normal\n");
5381 if (map_state
== IMSM_T_STATE_DEGRADED
) {
5382 printf("degraded\n");
5383 end_migration(dev
, map_state
);
5385 dprintf("failed\n");
5387 map
->map_state
= map_state
;
5389 super
->updates_pending
++;
5393 static int store_imsm_mpb(int fd
, struct imsm_super
*mpb
)
5396 __u32 mpb_size
= __le32_to_cpu(mpb
->mpb_size
);
5397 unsigned long long dsize
;
5398 unsigned long long sectors
;
5400 get_dev_size(fd
, NULL
, &dsize
);
5402 if (mpb_size
> 512) {
5403 /* -1 to account for anchor */
5404 sectors
= mpb_sectors(mpb
) - 1;
5406 /* write the extended mpb to the sectors preceeding the anchor */
5407 if (lseek64(fd
, dsize
- (512 * (2 + sectors
)), SEEK_SET
) < 0)
5410 if ((unsigned long long)write(fd
, buf
+ 512, 512 * sectors
)
5415 /* first block is stored on second to last sector of the disk */
5416 if (lseek64(fd
, dsize
- (512 * 2), SEEK_SET
) < 0)
5419 if (write(fd
, buf
, 512) != 512)
5425 static void imsm_sync_metadata(struct supertype
*container
)
5427 struct intel_super
*super
= container
->sb
;
5429 dprintf("sync metadata: %d\n", super
->updates_pending
);
5430 if (!super
->updates_pending
)
5433 write_super_imsm(container
, 0);
5435 super
->updates_pending
= 0;
5438 static struct dl
*imsm_readd(struct intel_super
*super
, int idx
, struct active_array
*a
)
5440 struct imsm_dev
*dev
= get_imsm_dev(super
, a
->info
.container_member
);
5441 int i
= get_imsm_disk_idx(dev
, idx
, -1);
5444 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
5448 if (dl
&& is_failed(&dl
->disk
))
5452 dprintf("%s: found %x:%x\n", __func__
, dl
->major
, dl
->minor
);
5457 static struct dl
*imsm_add_spare(struct intel_super
*super
, int slot
,
5458 struct active_array
*a
, int activate_new
,
5459 struct mdinfo
*additional_test_list
)
5461 struct imsm_dev
*dev
= get_imsm_dev(super
, a
->info
.container_member
);
5462 int idx
= get_imsm_disk_idx(dev
, slot
, -1);
5463 struct imsm_super
*mpb
= super
->anchor
;
5464 struct imsm_map
*map
;
5465 unsigned long long pos
;
5470 __u32 array_start
= 0;
5471 __u32 array_end
= 0;
5473 struct mdinfo
*test_list
;
5475 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
5476 /* If in this array, skip */
5477 for (d
= a
->info
.devs
; d
; d
= d
->next
)
5478 if (d
->state_fd
>= 0 &&
5479 d
->disk
.major
== dl
->major
&&
5480 d
->disk
.minor
== dl
->minor
) {
5481 dprintf("%x:%x already in array\n",
5482 dl
->major
, dl
->minor
);
5487 test_list
= additional_test_list
;
5489 if (test_list
->disk
.major
== dl
->major
&&
5490 test_list
->disk
.minor
== dl
->minor
) {
5491 dprintf("%x:%x already in additional test list\n",
5492 dl
->major
, dl
->minor
);
5495 test_list
= test_list
->next
;
5500 /* skip in use or failed drives */
5501 if (is_failed(&dl
->disk
) || idx
== dl
->index
||
5503 dprintf("%x:%x status (failed: %d index: %d)\n",
5504 dl
->major
, dl
->minor
, is_failed(&dl
->disk
), idx
);
5508 /* skip pure spares when we are looking for partially
5509 * assimilated drives
5511 if (dl
->index
== -1 && !activate_new
)
5514 /* Does this unused device have the requisite free space?
5515 * It needs to be able to cover all member volumes
5517 ex
= get_extents(super
, dl
);
5519 dprintf("cannot get extents\n");
5522 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
5523 dev
= get_imsm_dev(super
, i
);
5524 map
= get_imsm_map(dev
, 0);
5526 /* check if this disk is already a member of
5529 if (get_imsm_disk_slot(map
, dl
->index
) >= 0)
5535 array_start
= __le32_to_cpu(map
->pba_of_lba0
);
5536 array_end
= array_start
+
5537 __le32_to_cpu(map
->blocks_per_member
) - 1;
5540 /* check that we can start at pba_of_lba0 with
5541 * blocks_per_member of space
5543 if (array_start
>= pos
&& array_end
< ex
[j
].start
) {
5547 pos
= ex
[j
].start
+ ex
[j
].size
;
5549 } while (ex
[j
-1].size
);
5556 if (i
< mpb
->num_raid_devs
) {
5557 dprintf("%x:%x does not have %u to %u available\n",
5558 dl
->major
, dl
->minor
, array_start
, array_end
);
5569 static int imsm_rebuild_allowed(struct supertype
*cont
, int dev_idx
, int failed
)
5571 struct imsm_dev
*dev2
;
5572 struct imsm_map
*map
;
5578 dev2
= get_imsm_dev(cont
->sb
, dev_idx
);
5580 state
= imsm_check_degraded(cont
->sb
, dev2
, failed
);
5581 if (state
== IMSM_T_STATE_FAILED
) {
5582 map
= get_imsm_map(dev2
, 0);
5585 for (slot
= 0; slot
< map
->num_members
; slot
++) {
5587 * Check if failed disks are deleted from intel
5588 * disk list or are marked to be deleted
5590 idx
= get_imsm_disk_idx(dev2
, slot
, -1);
5591 idisk
= get_imsm_dl_disk(cont
->sb
, idx
);
5593 * Do not rebuild the array if failed disks
5594 * from failed sub-array are not removed from
5598 is_failed(&idisk
->disk
) &&
5599 (idisk
->action
!= DISK_REMOVE
))
5607 static struct mdinfo
*imsm_activate_spare(struct active_array
*a
,
5608 struct metadata_update
**updates
)
5611 * Find a device with unused free space and use it to replace a
5612 * failed/vacant region in an array. We replace failed regions one a
5613 * array at a time. The result is that a new spare disk will be added
5614 * to the first failed array and after the monitor has finished
5615 * propagating failures the remainder will be consumed.
5617 * FIXME add a capability for mdmon to request spares from another
5621 struct intel_super
*super
= a
->container
->sb
;
5622 int inst
= a
->info
.container_member
;
5623 struct imsm_dev
*dev
= get_imsm_dev(super
, inst
);
5624 struct imsm_map
*map
= get_imsm_map(dev
, 0);
5625 int failed
= a
->info
.array
.raid_disks
;
5626 struct mdinfo
*rv
= NULL
;
5629 struct metadata_update
*mu
;
5631 struct imsm_update_activate_spare
*u
;
5636 for (d
= a
->info
.devs
; d
; d
= d
->next
) {
5637 if ((d
->curr_state
& DS_FAULTY
) &&
5639 /* wait for Removal to happen */
5641 if (d
->state_fd
>= 0)
5645 dprintf("imsm: activate spare: inst=%d failed=%d (%d) level=%d\n",
5646 inst
, failed
, a
->info
.array
.raid_disks
, a
->info
.array
.level
);
5648 if (dev
->vol
.migr_state
&&
5649 dev
->vol
.migr_type
== MIGR_GEN_MIGR
)
5650 /* No repair during migration */
5653 if (a
->info
.array
.level
== 4)
5654 /* No repair for takeovered array
5655 * imsm doesn't support raid4
5659 if (imsm_check_degraded(super
, dev
, failed
) != IMSM_T_STATE_DEGRADED
)
5663 * If there are any failed disks check state of the other volume.
5664 * Block rebuild if the another one is failed until failed disks
5665 * are removed from container.
5668 dprintf("found failed disks in %s, check if there another"
5669 "failed sub-array.\n",
5671 /* check if states of the other volumes allow for rebuild */
5672 for (i
= 0; i
< super
->anchor
->num_raid_devs
; i
++) {
5674 allowed
= imsm_rebuild_allowed(a
->container
,
5682 /* For each slot, if it is not working, find a spare */
5683 for (i
= 0; i
< a
->info
.array
.raid_disks
; i
++) {
5684 for (d
= a
->info
.devs
; d
; d
= d
->next
)
5685 if (d
->disk
.raid_disk
== i
)
5687 dprintf("found %d: %p %x\n", i
, d
, d
?d
->curr_state
:0);
5688 if (d
&& (d
->state_fd
>= 0))
5692 * OK, this device needs recovery. Try to re-add the
5693 * previous occupant of this slot, if this fails see if
5694 * we can continue the assimilation of a spare that was
5695 * partially assimilated, finally try to activate a new
5698 dl
= imsm_readd(super
, i
, a
);
5700 dl
= imsm_add_spare(super
, i
, a
, 0, NULL
);
5702 dl
= imsm_add_spare(super
, i
, a
, 1, NULL
);
5706 /* found a usable disk with enough space */
5707 di
= malloc(sizeof(*di
));
5710 memset(di
, 0, sizeof(*di
));
5712 /* dl->index will be -1 in the case we are activating a
5713 * pristine spare. imsm_process_update() will create a
5714 * new index in this case. Once a disk is found to be
5715 * failed in all member arrays it is kicked from the
5718 di
->disk
.number
= dl
->index
;
5720 /* (ab)use di->devs to store a pointer to the device
5723 di
->devs
= (struct mdinfo
*) dl
;
5725 di
->disk
.raid_disk
= i
;
5726 di
->disk
.major
= dl
->major
;
5727 di
->disk
.minor
= dl
->minor
;
5729 di
->recovery_start
= 0;
5730 di
->data_offset
= __le32_to_cpu(map
->pba_of_lba0
);
5731 di
->component_size
= a
->info
.component_size
;
5732 di
->container_member
= inst
;
5733 super
->random
= random32();
5737 dprintf("%x:%x to be %d at %llu\n", dl
->major
, dl
->minor
,
5738 i
, di
->data_offset
);
5744 /* No spares found */
5746 /* Now 'rv' has a list of devices to return.
5747 * Create a metadata_update record to update the
5748 * disk_ord_tbl for the array
5750 mu
= malloc(sizeof(*mu
));
5752 mu
->buf
= malloc(sizeof(struct imsm_update_activate_spare
) * num_spares
);
5753 if (mu
->buf
== NULL
) {
5760 struct mdinfo
*n
= rv
->next
;
5769 mu
->space_list
= NULL
;
5770 mu
->len
= sizeof(struct imsm_update_activate_spare
) * num_spares
;
5771 mu
->next
= *updates
;
5772 u
= (struct imsm_update_activate_spare
*) mu
->buf
;
5774 for (di
= rv
; di
; di
= di
->next
) {
5775 u
->type
= update_activate_spare
;
5776 u
->dl
= (struct dl
*) di
->devs
;
5778 u
->slot
= di
->disk
.raid_disk
;
5789 static int disks_overlap(struct intel_super
*super
, int idx
, struct imsm_update_create_array
*u
)
5791 struct imsm_dev
*dev
= get_imsm_dev(super
, idx
);
5792 struct imsm_map
*map
= get_imsm_map(dev
, 0);
5793 struct imsm_map
*new_map
= get_imsm_map(&u
->dev
, 0);
5794 struct disk_info
*inf
= get_disk_info(u
);
5795 struct imsm_disk
*disk
;
5799 for (i
= 0; i
< map
->num_members
; i
++) {
5800 disk
= get_imsm_disk(super
, get_imsm_disk_idx(dev
, i
, -1));
5801 for (j
= 0; j
< new_map
->num_members
; j
++)
5802 if (serialcmp(disk
->serial
, inf
[j
].serial
) == 0)
5810 static struct dl
*get_disk_super(struct intel_super
*super
, int major
, int minor
)
5812 struct dl
*dl
= NULL
;
5813 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
5814 if ((dl
->major
== major
) && (dl
->minor
== minor
))
5819 static int remove_disk_super(struct intel_super
*super
, int major
, int minor
)
5821 struct dl
*prev
= NULL
;
5825 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
5826 if ((dl
->major
== major
) && (dl
->minor
== minor
)) {
5829 prev
->next
= dl
->next
;
5831 super
->disks
= dl
->next
;
5833 __free_imsm_disk(dl
);
5834 dprintf("%s: removed %x:%x\n",
5835 __func__
, major
, minor
);
5843 static void imsm_delete(struct intel_super
*super
, struct dl
**dlp
, unsigned index
);
5845 static int add_remove_disk_update(struct intel_super
*super
)
5847 int check_degraded
= 0;
5848 struct dl
*disk
= NULL
;
5849 /* add/remove some spares to/from the metadata/contrainer */
5850 while (super
->disk_mgmt_list
) {
5851 struct dl
*disk_cfg
;
5853 disk_cfg
= super
->disk_mgmt_list
;
5854 super
->disk_mgmt_list
= disk_cfg
->next
;
5855 disk_cfg
->next
= NULL
;
5857 if (disk_cfg
->action
== DISK_ADD
) {
5858 disk_cfg
->next
= super
->disks
;
5859 super
->disks
= disk_cfg
;
5861 dprintf("%s: added %x:%x\n",
5862 __func__
, disk_cfg
->major
,
5864 } else if (disk_cfg
->action
== DISK_REMOVE
) {
5865 dprintf("Disk remove action processed: %x.%x\n",
5866 disk_cfg
->major
, disk_cfg
->minor
);
5867 disk
= get_disk_super(super
,
5871 /* store action status */
5872 disk
->action
= DISK_REMOVE
;
5873 /* remove spare disks only */
5874 if (disk
->index
== -1) {
5875 remove_disk_super(super
,
5880 /* release allocate disk structure */
5881 __free_imsm_disk(disk_cfg
);
5884 return check_degraded
;
5887 static int apply_reshape_container_disks_update(struct imsm_update_reshape
*u
,
5888 struct intel_super
*super
,
5891 struct dl
*new_disk
;
5892 struct intel_dev
*id
;
5894 int delta_disks
= u
->new_raid_disks
- u
->old_raid_disks
;
5895 int disk_count
= u
->old_raid_disks
;
5896 void **tofree
= NULL
;
5897 int devices_to_reshape
= 1;
5898 struct imsm_super
*mpb
= super
->anchor
;
5900 unsigned int dev_id
;
5902 dprintf("imsm: apply_reshape_container_disks_update()\n");
5904 /* enable spares to use in array */
5905 for (i
= 0; i
< delta_disks
; i
++) {
5906 new_disk
= get_disk_super(super
,
5907 major(u
->new_disks
[i
]),
5908 minor(u
->new_disks
[i
]));
5909 dprintf("imsm: new disk for reshape is: %i:%i "
5910 "(%p, index = %i)\n",
5911 major(u
->new_disks
[i
]), minor(u
->new_disks
[i
]),
5912 new_disk
, new_disk
->index
);
5913 if ((new_disk
== NULL
) ||
5914 ((new_disk
->index
>= 0) &&
5915 (new_disk
->index
< u
->old_raid_disks
)))
5916 goto update_reshape_exit
;
5917 new_disk
->index
= disk_count
++;
5918 /* slot to fill in autolayout
5920 new_disk
->raiddisk
= new_disk
->index
;
5921 new_disk
->disk
.status
|=
5923 new_disk
->disk
.status
&= ~SPARE_DISK
;
5926 dprintf("imsm: volume set mpb->num_raid_devs = %i\n",
5927 mpb
->num_raid_devs
);
5928 /* manage changes in volume
5930 for (dev_id
= 0; dev_id
< mpb
->num_raid_devs
; dev_id
++) {
5931 void **sp
= *space_list
;
5932 struct imsm_dev
*newdev
;
5933 struct imsm_map
*newmap
, *oldmap
;
5935 for (id
= super
->devlist
; id
; id
= id
->next
) {
5936 if (id
->index
== dev_id
)
5945 /* Copy the dev, but not (all of) the map */
5946 memcpy(newdev
, id
->dev
, sizeof(*newdev
));
5947 oldmap
= get_imsm_map(id
->dev
, 0);
5948 newmap
= get_imsm_map(newdev
, 0);
5949 /* Copy the current map */
5950 memcpy(newmap
, oldmap
, sizeof_imsm_map(oldmap
));
5951 /* update one device only
5953 if (devices_to_reshape
) {
5954 dprintf("imsm: modifying subdev: %i\n",
5956 devices_to_reshape
--;
5957 newdev
->vol
.migr_state
= 1;
5958 newdev
->vol
.curr_migr_unit
= 0;
5959 newdev
->vol
.migr_type
= MIGR_GEN_MIGR
;
5960 newmap
->num_members
= u
->new_raid_disks
;
5961 for (i
= 0; i
< delta_disks
; i
++) {
5962 set_imsm_ord_tbl_ent(newmap
,
5963 u
->old_raid_disks
+ i
,
5964 u
->old_raid_disks
+ i
);
5966 /* New map is correct, now need to save old map
5968 newmap
= get_imsm_map(newdev
, 1);
5969 memcpy(newmap
, oldmap
, sizeof_imsm_map(oldmap
));
5971 imsm_set_array_size(newdev
);
5974 sp
= (void **)id
->dev
;
5980 *space_list
= tofree
;
5983 update_reshape_exit
:
5988 static int apply_takeover_update(struct imsm_update_takeover
*u
,
5989 struct intel_super
*super
,
5992 struct imsm_dev
*dev
= NULL
;
5993 struct intel_dev
*dv
;
5994 struct imsm_dev
*dev_new
;
5995 struct imsm_map
*map
;
5999 for (dv
= super
->devlist
; dv
; dv
= dv
->next
)
6000 if (dv
->index
== (unsigned int)u
->subarray
) {
6008 map
= get_imsm_map(dev
, 0);
6010 if (u
->direction
== R10_TO_R0
) {
6011 /* Number of failed disks must be half of initial disk number */
6012 if (imsm_count_failed(super
, dev
) != (map
->num_members
/ 2))
6015 /* iterate through devices to mark removed disks as spare */
6016 for (dm
= super
->disks
; dm
; dm
= dm
->next
) {
6017 if (dm
->disk
.status
& FAILED_DISK
) {
6018 int idx
= dm
->index
;
6019 /* update indexes on the disk list */
6020 /* FIXME this loop-with-the-loop looks wrong, I'm not convinced
6021 the index values will end up being correct.... NB */
6022 for (du
= super
->disks
; du
; du
= du
->next
)
6023 if (du
->index
> idx
)
6025 /* mark as spare disk */
6026 dm
->disk
.status
= SPARE_DISK
;
6031 map
->num_members
= map
->num_members
/ 2;
6032 map
->map_state
= IMSM_T_STATE_NORMAL
;
6033 map
->num_domains
= 1;
6034 map
->raid_level
= 0;
6035 map
->failed_disk_num
= -1;
6038 if (u
->direction
== R0_TO_R10
) {
6040 /* update slots in current disk list */
6041 for (dm
= super
->disks
; dm
; dm
= dm
->next
) {
6045 /* create new *missing* disks */
6046 for (i
= 0; i
< map
->num_members
; i
++) {
6047 space
= *space_list
;
6050 *space_list
= *space
;
6052 memcpy(du
, super
->disks
, sizeof(*du
));
6053 du
->disk
.status
= FAILED_DISK
;
6054 du
->disk
.scsi_id
= 0;
6058 du
->index
= (i
* 2) + 1;
6059 sprintf((char *)du
->disk
.serial
,
6060 " MISSING_%d", du
->index
);
6061 sprintf((char *)du
->serial
,
6062 "MISSING_%d", du
->index
);
6063 du
->next
= super
->missing
;
6064 super
->missing
= du
;
6066 /* create new dev and map */
6067 space
= *space_list
;
6070 *space_list
= *space
;
6071 dev_new
= (void *)space
;
6072 memcpy(dev_new
, dev
, sizeof(*dev
));
6073 /* update new map */
6074 map
= get_imsm_map(dev_new
, 0);
6075 map
->failed_disk_num
= map
->num_members
;
6076 map
->num_members
= map
->num_members
* 2;
6077 map
->map_state
= IMSM_T_STATE_NORMAL
;
6078 map
->num_domains
= 2;
6079 map
->raid_level
= 1;
6080 /* replace dev<->dev_new */
6083 /* update disk order table */
6084 for (du
= super
->disks
; du
; du
= du
->next
)
6086 set_imsm_ord_tbl_ent(map
, du
->index
, du
->index
);
6087 for (du
= super
->missing
; du
; du
= du
->next
)
6089 set_imsm_ord_tbl_ent(map
, du
->index
,
6090 du
->index
| IMSM_ORD_REBUILD
);
6095 static void imsm_process_update(struct supertype
*st
,
6096 struct metadata_update
*update
)
6099 * crack open the metadata_update envelope to find the update record
6100 * update can be one of:
6101 * update_reshape_container_disks - all the arrays in the container
6102 * are being reshaped to have more devices. We need to mark
6103 * the arrays for general migration and convert selected spares
6104 * into active devices.
6105 * update_activate_spare - a spare device has replaced a failed
6106 * device in an array, update the disk_ord_tbl. If this disk is
6107 * present in all member arrays then also clear the SPARE_DISK
6109 * update_create_array
6111 * update_rename_array
6112 * update_add_remove_disk
6114 struct intel_super
*super
= st
->sb
;
6115 struct imsm_super
*mpb
;
6116 enum imsm_update_type type
= *(enum imsm_update_type
*) update
->buf
;
6118 /* update requires a larger buf but the allocation failed */
6119 if (super
->next_len
&& !super
->next_buf
) {
6120 super
->next_len
= 0;
6124 if (super
->next_buf
) {
6125 memcpy(super
->next_buf
, super
->buf
, super
->len
);
6127 super
->len
= super
->next_len
;
6128 super
->buf
= super
->next_buf
;
6130 super
->next_len
= 0;
6131 super
->next_buf
= NULL
;
6134 mpb
= super
->anchor
;
6137 case update_takeover
: {
6138 struct imsm_update_takeover
*u
= (void *)update
->buf
;
6139 if (apply_takeover_update(u
, super
, &update
->space_list
))
6140 super
->updates_pending
++;
6144 case update_reshape_container_disks
: {
6145 struct imsm_update_reshape
*u
= (void *)update
->buf
;
6146 if (apply_reshape_container_disks_update(
6147 u
, super
, &update
->space_list
))
6148 super
->updates_pending
++;
6151 case update_activate_spare
: {
6152 struct imsm_update_activate_spare
*u
= (void *) update
->buf
;
6153 struct imsm_dev
*dev
= get_imsm_dev(super
, u
->array
);
6154 struct imsm_map
*map
= get_imsm_map(dev
, 0);
6155 struct imsm_map
*migr_map
;
6156 struct active_array
*a
;
6157 struct imsm_disk
*disk
;
6162 int victim
= get_imsm_disk_idx(dev
, u
->slot
, -1);
6165 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
6170 fprintf(stderr
, "error: imsm_activate_spare passed "
6171 "an unknown disk (index: %d)\n",
6176 super
->updates_pending
++;
6178 /* count failures (excluding rebuilds and the victim)
6179 * to determine map[0] state
6182 for (i
= 0; i
< map
->num_members
; i
++) {
6185 disk
= get_imsm_disk(super
,
6186 get_imsm_disk_idx(dev
, i
, -1));
6187 if (!disk
|| is_failed(disk
))
6191 /* adding a pristine spare, assign a new index */
6192 if (dl
->index
< 0) {
6193 dl
->index
= super
->anchor
->num_disks
;
6194 super
->anchor
->num_disks
++;
6197 disk
->status
|= CONFIGURED_DISK
;
6198 disk
->status
&= ~SPARE_DISK
;
6201 to_state
= imsm_check_degraded(super
, dev
, failed
);
6202 map
->map_state
= IMSM_T_STATE_DEGRADED
;
6203 migrate(dev
, to_state
, MIGR_REBUILD
);
6204 migr_map
= get_imsm_map(dev
, 1);
6205 set_imsm_ord_tbl_ent(map
, u
->slot
, dl
->index
);
6206 set_imsm_ord_tbl_ent(migr_map
, u
->slot
, dl
->index
| IMSM_ORD_REBUILD
);
6208 /* update the family_num to mark a new container
6209 * generation, being careful to record the existing
6210 * family_num in orig_family_num to clean up after
6211 * earlier mdadm versions that neglected to set it.
6213 if (mpb
->orig_family_num
== 0)
6214 mpb
->orig_family_num
= mpb
->family_num
;
6215 mpb
->family_num
+= super
->random
;
6217 /* count arrays using the victim in the metadata */
6219 for (a
= st
->arrays
; a
; a
= a
->next
) {
6220 dev
= get_imsm_dev(super
, a
->info
.container_member
);
6221 map
= get_imsm_map(dev
, 0);
6223 if (get_imsm_disk_slot(map
, victim
) >= 0)
6227 /* delete the victim if it is no longer being
6233 /* We know that 'manager' isn't touching anything,
6234 * so it is safe to delete
6236 for (dlp
= &super
->disks
; *dlp
; dlp
= &(*dlp
)->next
)
6237 if ((*dlp
)->index
== victim
)
6240 /* victim may be on the missing list */
6242 for (dlp
= &super
->missing
; *dlp
; dlp
= &(*dlp
)->next
)
6243 if ((*dlp
)->index
== victim
)
6245 imsm_delete(super
, dlp
, victim
);
6249 case update_create_array
: {
6250 /* someone wants to create a new array, we need to be aware of
6251 * a few races/collisions:
6252 * 1/ 'Create' called by two separate instances of mdadm
6253 * 2/ 'Create' versus 'activate_spare': mdadm has chosen
6254 * devices that have since been assimilated via
6256 * In the event this update can not be carried out mdadm will
6257 * (FIX ME) notice that its update did not take hold.
6259 struct imsm_update_create_array
*u
= (void *) update
->buf
;
6260 struct intel_dev
*dv
;
6261 struct imsm_dev
*dev
;
6262 struct imsm_map
*map
, *new_map
;
6263 unsigned long long start
, end
;
6264 unsigned long long new_start
, new_end
;
6266 struct disk_info
*inf
;
6269 /* handle racing creates: first come first serve */
6270 if (u
->dev_idx
< mpb
->num_raid_devs
) {
6271 dprintf("%s: subarray %d already defined\n",
6272 __func__
, u
->dev_idx
);
6276 /* check update is next in sequence */
6277 if (u
->dev_idx
!= mpb
->num_raid_devs
) {
6278 dprintf("%s: can not create array %d expected index %d\n",
6279 __func__
, u
->dev_idx
, mpb
->num_raid_devs
);
6283 new_map
= get_imsm_map(&u
->dev
, 0);
6284 new_start
= __le32_to_cpu(new_map
->pba_of_lba0
);
6285 new_end
= new_start
+ __le32_to_cpu(new_map
->blocks_per_member
);
6286 inf
= get_disk_info(u
);
6288 /* handle activate_spare versus create race:
6289 * check to make sure that overlapping arrays do not include
6292 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
6293 dev
= get_imsm_dev(super
, i
);
6294 map
= get_imsm_map(dev
, 0);
6295 start
= __le32_to_cpu(map
->pba_of_lba0
);
6296 end
= start
+ __le32_to_cpu(map
->blocks_per_member
);
6297 if ((new_start
>= start
&& new_start
<= end
) ||
6298 (start
>= new_start
&& start
<= new_end
))
6303 if (disks_overlap(super
, i
, u
)) {
6304 dprintf("%s: arrays overlap\n", __func__
);
6309 /* check that prepare update was successful */
6310 if (!update
->space
) {
6311 dprintf("%s: prepare update failed\n", __func__
);
6315 /* check that all disks are still active before committing
6316 * changes. FIXME: could we instead handle this by creating a
6317 * degraded array? That's probably not what the user expects,
6318 * so better to drop this update on the floor.
6320 for (i
= 0; i
< new_map
->num_members
; i
++) {
6321 dl
= serial_to_dl(inf
[i
].serial
, super
);
6323 dprintf("%s: disk disappeared\n", __func__
);
6328 super
->updates_pending
++;
6330 /* convert spares to members and fixup ord_tbl */
6331 for (i
= 0; i
< new_map
->num_members
; i
++) {
6332 dl
= serial_to_dl(inf
[i
].serial
, super
);
6333 if (dl
->index
== -1) {
6334 dl
->index
= mpb
->num_disks
;
6336 dl
->disk
.status
|= CONFIGURED_DISK
;
6337 dl
->disk
.status
&= ~SPARE_DISK
;
6339 set_imsm_ord_tbl_ent(new_map
, i
, dl
->index
);
6344 update
->space
= NULL
;
6345 imsm_copy_dev(dev
, &u
->dev
);
6346 dv
->index
= u
->dev_idx
;
6347 dv
->next
= super
->devlist
;
6348 super
->devlist
= dv
;
6349 mpb
->num_raid_devs
++;
6351 imsm_update_version_info(super
);
6354 /* mdmon knows how to release update->space, but not
6355 * ((struct intel_dev *) update->space)->dev
6357 if (update
->space
) {
6363 case update_kill_array
: {
6364 struct imsm_update_kill_array
*u
= (void *) update
->buf
;
6365 int victim
= u
->dev_idx
;
6366 struct active_array
*a
;
6367 struct intel_dev
**dp
;
6368 struct imsm_dev
*dev
;
6370 /* sanity check that we are not affecting the uuid of
6371 * active arrays, or deleting an active array
6373 * FIXME when immutable ids are available, but note that
6374 * we'll also need to fixup the invalidated/active
6375 * subarray indexes in mdstat
6377 for (a
= st
->arrays
; a
; a
= a
->next
)
6378 if (a
->info
.container_member
>= victim
)
6380 /* by definition if mdmon is running at least one array
6381 * is active in the container, so checking
6382 * mpb->num_raid_devs is just extra paranoia
6384 dev
= get_imsm_dev(super
, victim
);
6385 if (a
|| !dev
|| mpb
->num_raid_devs
== 1) {
6386 dprintf("failed to delete subarray-%d\n", victim
);
6390 for (dp
= &super
->devlist
; *dp
;)
6391 if ((*dp
)->index
== (unsigned)super
->current_vol
) {
6394 if ((*dp
)->index
> (unsigned)victim
)
6398 mpb
->num_raid_devs
--;
6399 super
->updates_pending
++;
6402 case update_rename_array
: {
6403 struct imsm_update_rename_array
*u
= (void *) update
->buf
;
6404 char name
[MAX_RAID_SERIAL_LEN
+1];
6405 int target
= u
->dev_idx
;
6406 struct active_array
*a
;
6407 struct imsm_dev
*dev
;
6409 /* sanity check that we are not affecting the uuid of
6412 snprintf(name
, MAX_RAID_SERIAL_LEN
, "%s", (char *) u
->name
);
6413 name
[MAX_RAID_SERIAL_LEN
] = '\0';
6414 for (a
= st
->arrays
; a
; a
= a
->next
)
6415 if (a
->info
.container_member
== target
)
6417 dev
= get_imsm_dev(super
, u
->dev_idx
);
6418 if (a
|| !dev
|| !check_name(super
, name
, 1)) {
6419 dprintf("failed to rename subarray-%d\n", target
);
6423 snprintf((char *) dev
->volume
, MAX_RAID_SERIAL_LEN
, "%s", name
);
6424 super
->updates_pending
++;
6427 case update_add_remove_disk
: {
6428 /* we may be able to repair some arrays if disks are
6429 * being added, check teh status of add_remove_disk
6430 * if discs has been added.
6432 if (add_remove_disk_update(super
)) {
6433 struct active_array
*a
;
6435 super
->updates_pending
++;
6436 for (a
= st
->arrays
; a
; a
= a
->next
)
6437 a
->check_degraded
= 1;
6442 fprintf(stderr
, "error: unsuported process update type:"
6443 "(type: %d)\n", type
);
6447 static void imsm_prepare_update(struct supertype
*st
,
6448 struct metadata_update
*update
)
6451 * Allocate space to hold new disk entries, raid-device entries or a new
6452 * mpb if necessary. The manager synchronously waits for updates to
6453 * complete in the monitor, so new mpb buffers allocated here can be
6454 * integrated by the monitor thread without worrying about live pointers
6455 * in the manager thread.
6457 enum imsm_update_type type
= *(enum imsm_update_type
*) update
->buf
;
6458 struct intel_super
*super
= st
->sb
;
6459 struct imsm_super
*mpb
= super
->anchor
;
6464 case update_takeover
: {
6465 struct imsm_update_takeover
*u
= (void *)update
->buf
;
6466 if (u
->direction
== R0_TO_R10
) {
6467 void **tail
= (void **)&update
->space_list
;
6468 struct imsm_dev
*dev
= get_imsm_dev(super
, u
->subarray
);
6469 struct imsm_map
*map
= get_imsm_map(dev
, 0);
6470 int num_members
= map
->num_members
;
6474 /* allocate memory for added disks */
6475 for (i
= 0; i
< num_members
; i
++) {
6476 size
= sizeof(struct dl
);
6477 space
= malloc(size
);
6486 /* allocate memory for new device */
6487 size
= sizeof_imsm_dev(super
->devlist
->dev
, 0) +
6488 (num_members
* sizeof(__u32
));
6489 space
= malloc(size
);
6498 len
= disks_to_mpb_size(num_members
* 2);
6500 /* if allocation didn't success, free buffer */
6501 while (update
->space_list
) {
6502 void **sp
= update
->space_list
;
6503 update
->space_list
= *sp
;
6511 case update_reshape_container_disks
: {
6512 /* Every raid device in the container is about to
6513 * gain some more devices, and we will enter a
6515 * So each 'imsm_map' will be bigger, and the imsm_vol
6516 * will now hold 2 of them.
6517 * Thus we need new 'struct imsm_dev' allocations sized
6518 * as sizeof_imsm_dev but with more devices in both maps.
6520 struct imsm_update_reshape
*u
= (void *)update
->buf
;
6521 struct intel_dev
*dl
;
6522 void **space_tail
= (void**)&update
->space_list
;
6524 dprintf("imsm: imsm_prepare_update() for update_reshape\n");
6526 for (dl
= super
->devlist
; dl
; dl
= dl
->next
) {
6527 int size
= sizeof_imsm_dev(dl
->dev
, 1);
6529 if (u
->new_raid_disks
> u
->old_raid_disks
)
6530 size
+= sizeof(__u32
)*2*
6531 (u
->new_raid_disks
- u
->old_raid_disks
);
6540 len
= disks_to_mpb_size(u
->new_raid_disks
);
6541 dprintf("New anchor length is %llu\n", (unsigned long long)len
);
6544 case update_create_array
: {
6545 struct imsm_update_create_array
*u
= (void *) update
->buf
;
6546 struct intel_dev
*dv
;
6547 struct imsm_dev
*dev
= &u
->dev
;
6548 struct imsm_map
*map
= get_imsm_map(dev
, 0);
6550 struct disk_info
*inf
;
6554 inf
= get_disk_info(u
);
6555 len
= sizeof_imsm_dev(dev
, 1);
6556 /* allocate a new super->devlist entry */
6557 dv
= malloc(sizeof(*dv
));
6559 dv
->dev
= malloc(len
);
6564 update
->space
= NULL
;
6568 /* count how many spares will be converted to members */
6569 for (i
= 0; i
< map
->num_members
; i
++) {
6570 dl
= serial_to_dl(inf
[i
].serial
, super
);
6572 /* hmm maybe it failed?, nothing we can do about
6577 if (count_memberships(dl
, super
) == 0)
6580 len
+= activate
* sizeof(struct imsm_disk
);
6587 /* check if we need a larger metadata buffer */
6588 if (super
->next_buf
)
6589 buf_len
= super
->next_len
;
6591 buf_len
= super
->len
;
6593 if (__le32_to_cpu(mpb
->mpb_size
) + len
> buf_len
) {
6594 /* ok we need a larger buf than what is currently allocated
6595 * if this allocation fails process_update will notice that
6596 * ->next_len is set and ->next_buf is NULL
6598 buf_len
= ROUND_UP(__le32_to_cpu(mpb
->mpb_size
) + len
, 512);
6599 if (super
->next_buf
)
6600 free(super
->next_buf
);
6602 super
->next_len
= buf_len
;
6603 if (posix_memalign(&super
->next_buf
, 512, buf_len
) == 0)
6604 memset(super
->next_buf
, 0, buf_len
);
6606 super
->next_buf
= NULL
;
6610 /* must be called while manager is quiesced */
6611 static void imsm_delete(struct intel_super
*super
, struct dl
**dlp
, unsigned index
)
6613 struct imsm_super
*mpb
= super
->anchor
;
6615 struct imsm_dev
*dev
;
6616 struct imsm_map
*map
;
6617 int i
, j
, num_members
;
6620 dprintf("%s: deleting device[%d] from imsm_super\n",
6623 /* shift all indexes down one */
6624 for (iter
= super
->disks
; iter
; iter
= iter
->next
)
6625 if (iter
->index
> (int)index
)
6627 for (iter
= super
->missing
; iter
; iter
= iter
->next
)
6628 if (iter
->index
> (int)index
)
6631 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
6632 dev
= get_imsm_dev(super
, i
);
6633 map
= get_imsm_map(dev
, 0);
6634 num_members
= map
->num_members
;
6635 for (j
= 0; j
< num_members
; j
++) {
6636 /* update ord entries being careful not to propagate
6637 * ord-flags to the first map
6639 ord
= get_imsm_ord_tbl_ent(dev
, j
, -1);
6641 if (ord_to_idx(ord
) <= index
)
6644 map
= get_imsm_map(dev
, 0);
6645 set_imsm_ord_tbl_ent(map
, j
, ord_to_idx(ord
- 1));
6646 map
= get_imsm_map(dev
, 1);
6648 set_imsm_ord_tbl_ent(map
, j
, ord
- 1);
6653 super
->updates_pending
++;
6655 struct dl
*dl
= *dlp
;
6657 *dlp
= (*dlp
)->next
;
6658 __free_imsm_disk(dl
);
6662 static char disk_by_path
[] = "/dev/disk/by-path/";
6664 static const char *imsm_get_disk_controller_domain(const char *path
)
6666 char disk_path
[PATH_MAX
];
6670 strncpy(disk_path
, disk_by_path
, PATH_MAX
- 1);
6671 strncat(disk_path
, path
, PATH_MAX
- strlen(disk_path
) - 1);
6672 if (stat(disk_path
, &st
) == 0) {
6673 struct sys_dev
* hba
;
6676 path
= devt_to_devpath(st
.st_rdev
);
6679 hba
= find_disk_attached_hba(-1, path
);
6680 if (hba
&& hba
->type
== SYS_DEV_SAS
)
6682 else if (hba
&& hba
->type
== SYS_DEV_SATA
)
6686 dprintf("path: %s hba: %s attached: %s\n",
6687 path
, (hba
) ? hba
->path
: "NULL", drv
);
6695 static int imsm_find_array_minor_by_subdev(int subdev
, int container
, int *minor
)
6697 char subdev_name
[20];
6698 struct mdstat_ent
*mdstat
;
6700 sprintf(subdev_name
, "%d", subdev
);
6701 mdstat
= mdstat_by_subdev(subdev_name
, container
);
6705 *minor
= mdstat
->devnum
;
6706 free_mdstat(mdstat
);
6710 static int imsm_reshape_is_allowed_on_container(struct supertype
*st
,
6711 struct geo_params
*geo
,
6712 int *old_raid_disks
)
6714 /* currently we only support increasing the number of devices
6715 * for a container. This increases the number of device for each
6716 * member array. They must all be RAID0 or RAID5.
6719 struct mdinfo
*info
, *member
;
6720 int devices_that_can_grow
= 0;
6722 dprintf("imsm: imsm_reshape_is_allowed_on_container(ENTER): "
6723 "st->devnum = (%i)\n",
6726 if (geo
->size
!= -1 ||
6727 geo
->level
!= UnSet
||
6728 geo
->layout
!= UnSet
||
6729 geo
->chunksize
!= 0 ||
6730 geo
->raid_disks
== UnSet
) {
6731 dprintf("imsm: Container operation is allowed for "
6732 "raid disks number change only.\n");
6736 info
= container_content_imsm(st
, NULL
);
6737 for (member
= info
; member
; member
= member
->next
) {
6741 dprintf("imsm: checking device_num: %i\n",
6742 member
->container_member
);
6744 if (geo
->raid_disks
<= member
->array
.raid_disks
) {
6745 /* we work on container for Online Capacity Expansion
6746 * only so raid_disks has to grow
6748 dprintf("imsm: for container operation raid disks "
6749 "increase is required\n");
6753 if ((info
->array
.level
!= 0) &&
6754 (info
->array
.level
!= 5)) {
6755 /* we cannot use this container with other raid level
6757 dprintf("imsm: for container operation wrong"
6758 " raid level (%i) detected\n",
6762 /* check for platform support
6763 * for this raid level configuration
6765 struct intel_super
*super
= st
->sb
;
6766 if (!is_raid_level_supported(super
->orom
,
6767 member
->array
.level
,
6769 dprintf("platform does not support raid%d with"
6773 geo
->raid_disks
> 1 ? "s" : "");
6778 if (*old_raid_disks
&&
6779 info
->array
.raid_disks
!= *old_raid_disks
)
6781 *old_raid_disks
= info
->array
.raid_disks
;
6783 /* All raid5 and raid0 volumes in container
6784 * have to be ready for Online Capacity Expansion
6785 * so they need to be assembled. We have already
6786 * checked that no recovery etc is happening.
6788 result
= imsm_find_array_minor_by_subdev(member
->container_member
,
6792 dprintf("imsm: cannot find array\n");
6795 devices_that_can_grow
++;
6798 if (!member
&& devices_that_can_grow
)
6802 dprintf("\tContainer operation allowed\n");
6804 dprintf("\tError: %i\n", ret_val
);
6809 /* Function: get_spares_for_grow
6810 * Description: Allocates memory and creates list of spare devices
6811 * avaliable in container. Checks if spare drive size is acceptable.
6812 * Parameters: Pointer to the supertype structure
6813 * Returns: Pointer to the list of spare devices (mdinfo structure) on success,
6816 static struct mdinfo
*get_spares_for_grow(struct supertype
*st
)
6818 unsigned long long min_size
= min_acceptable_spare_size_imsm(st
);
6819 return container_choose_spares(st
, min_size
, NULL
, NULL
, NULL
, 0);
6822 /******************************************************************************
6823 * function: imsm_create_metadata_update_for_reshape
6824 * Function creates update for whole IMSM container.
6826 ******************************************************************************/
6827 static int imsm_create_metadata_update_for_reshape(
6828 struct supertype
*st
,
6829 struct geo_params
*geo
,
6831 struct imsm_update_reshape
**updatep
)
6833 struct intel_super
*super
= st
->sb
;
6834 struct imsm_super
*mpb
= super
->anchor
;
6835 int update_memory_size
= 0;
6836 struct imsm_update_reshape
*u
= NULL
;
6837 struct mdinfo
*spares
= NULL
;
6839 int delta_disks
= 0;
6842 dprintf("imsm_update_metadata_for_reshape(enter) raid_disks = %i\n",
6845 delta_disks
= geo
->raid_disks
- old_raid_disks
;
6847 /* size of all update data without anchor */
6848 update_memory_size
= sizeof(struct imsm_update_reshape
);
6850 /* now add space for spare disks that we need to add. */
6851 update_memory_size
+= sizeof(u
->new_disks
[0]) * (delta_disks
- 1);
6853 u
= calloc(1, update_memory_size
);
6856 "cannot get memory for imsm_update_reshape update\n");
6859 u
->type
= update_reshape_container_disks
;
6860 u
->old_raid_disks
= old_raid_disks
;
6861 u
->new_raid_disks
= geo
->raid_disks
;
6863 /* now get spare disks list
6865 spares
= get_spares_for_grow(st
);
6868 || delta_disks
> spares
->array
.spare_disks
) {
6869 dprintf("imsm: ERROR: Cannot get spare devices.\n");
6873 /* we have got spares
6874 * update disk list in imsm_disk list table in anchor
6876 dprintf("imsm: %i spares are available.\n\n",
6877 spares
->array
.spare_disks
);
6880 for (i
= 0; i
< delta_disks
; i
++) {
6885 u
->new_disks
[i
] = makedev(dev
->disk
.major
,
6887 dl
= get_disk_super(super
, dev
->disk
.major
, dev
->disk
.minor
);
6888 dl
->index
= mpb
->num_disks
;
6898 dprintf("imsm: reshape update preparation :");
6899 if (i
== delta_disks
) {
6902 return update_memory_size
;
6905 dprintf(" Error\n");
6910 static void imsm_update_metadata_locally(struct supertype
*st
,
6913 struct metadata_update mu
;
6918 mu
.space_list
= NULL
;
6920 imsm_prepare_update(st
, &mu
);
6921 imsm_process_update(st
, &mu
);
6923 while (mu
.space_list
) {
6924 void **space
= mu
.space_list
;
6925 mu
.space_list
= *space
;
6930 /***************************************************************************
6931 * Function: imsm_analyze_change
6932 * Description: Function analyze change for single volume
6933 * and validate if transition is supported
6934 * Parameters: Geometry parameters, supertype structure
6935 * Returns: Operation type code on success, -1 if fail
6936 ****************************************************************************/
6937 enum imsm_reshape_type
imsm_analyze_change(struct supertype
*st
,
6938 struct geo_params
*geo
)
6945 getinfo_super_imsm_volume(st
, &info
, NULL
);
6947 if ((geo
->level
!= info
.array
.level
) &&
6948 (geo
->level
>= 0) &&
6949 (geo
->level
!= UnSet
)) {
6950 switch (info
.array
.level
) {
6952 if (geo
->level
== 5) {
6953 change
= CH_MIGRATION
;
6956 if (geo
->level
== 10) {
6957 change
= CH_TAKEOVER
;
6962 if (geo
->level
== 0) {
6963 change
= CH_TAKEOVER
;
6968 if (geo
->level
== 0)
6969 change
= CH_MIGRATION
;
6972 if (geo
->level
== 0) {
6973 change
= CH_TAKEOVER
;
6980 Name
" Error. Level Migration from %d to %d "
6982 info
.array
.level
, geo
->level
);
6983 goto analyse_change_exit
;
6986 geo
->level
= info
.array
.level
;
6988 if ((geo
->layout
!= info
.array
.layout
)
6989 && ((geo
->layout
!= UnSet
) && (geo
->layout
!= -1))) {
6990 change
= CH_MIGRATION
;
6991 if ((info
.array
.layout
== 0)
6992 && (info
.array
.level
== 5)
6993 && (geo
->layout
== 5)) {
6994 /* reshape 5 -> 4 */
6995 } else if ((info
.array
.layout
== 5)
6996 && (info
.array
.level
== 5)
6997 && (geo
->layout
== 0)) {
6998 /* reshape 4 -> 5 */
7003 Name
" Error. Layout Migration from %d to %d "
7005 info
.array
.layout
, geo
->layout
);
7007 goto analyse_change_exit
;
7010 geo
->layout
= info
.array
.layout
;
7012 if ((geo
->chunksize
> 0) && (geo
->chunksize
!= UnSet
)
7013 && (geo
->chunksize
!= info
.array
.chunk_size
))
7014 change
= CH_MIGRATION
;
7016 geo
->chunksize
= info
.array
.chunk_size
;
7018 chunk
= geo
->chunksize
/ 1024;
7019 if (!validate_geometry_imsm(st
,
7029 struct intel_super
*super
= st
->sb
;
7030 struct imsm_super
*mpb
= super
->anchor
;
7032 if (mpb
->num_raid_devs
> 1) {
7034 Name
" Error. Cannot perform operation on %s"
7035 "- for this operation it MUST be single "
7036 "array in container\n",
7042 analyse_change_exit
:
7047 int imsm_takeover(struct supertype
*st
, struct geo_params
*geo
)
7049 struct intel_super
*super
= st
->sb
;
7050 struct imsm_update_takeover
*u
;
7052 u
= malloc(sizeof(struct imsm_update_takeover
));
7056 u
->type
= update_takeover
;
7057 u
->subarray
= super
->current_vol
;
7059 /* 10->0 transition */
7060 if (geo
->level
== 0)
7061 u
->direction
= R10_TO_R0
;
7063 /* 0->10 transition */
7064 if (geo
->level
== 10)
7065 u
->direction
= R0_TO_R10
;
7067 /* update metadata locally */
7068 imsm_update_metadata_locally(st
, u
,
7069 sizeof(struct imsm_update_takeover
));
7070 /* and possibly remotely */
7071 if (st
->update_tail
)
7072 append_metadata_update(st
, u
,
7073 sizeof(struct imsm_update_takeover
));
7080 static int imsm_reshape_super(struct supertype
*st
, long long size
, int level
,
7081 int layout
, int chunksize
, int raid_disks
,
7082 int delta_disks
, char *backup
, char *dev
,
7086 struct geo_params geo
;
7088 dprintf("imsm: reshape_super called.\n");
7090 memset(&geo
, 0, sizeof(struct geo_params
));
7093 geo
.dev_id
= st
->devnum
;
7096 geo
.layout
= layout
;
7097 geo
.chunksize
= chunksize
;
7098 geo
.raid_disks
= raid_disks
;
7099 if (delta_disks
!= UnSet
)
7100 geo
.raid_disks
+= delta_disks
;
7102 dprintf("\tfor level : %i\n", geo
.level
);
7103 dprintf("\tfor raid_disks : %i\n", geo
.raid_disks
);
7105 if (experimental() == 0)
7108 if (st
->container_dev
== st
->devnum
) {
7109 /* On container level we can only increase number of devices. */
7110 dprintf("imsm: info: Container operation\n");
7111 int old_raid_disks
= 0;
7112 if (imsm_reshape_is_allowed_on_container(
7113 st
, &geo
, &old_raid_disks
)) {
7114 struct imsm_update_reshape
*u
= NULL
;
7117 len
= imsm_create_metadata_update_for_reshape(
7118 st
, &geo
, old_raid_disks
, &u
);
7121 dprintf("imsm: Cannot prepare update\n");
7122 goto exit_imsm_reshape_super
;
7126 /* update metadata locally */
7127 imsm_update_metadata_locally(st
, u
, len
);
7128 /* and possibly remotely */
7129 if (st
->update_tail
)
7130 append_metadata_update(st
, u
, len
);
7135 fprintf(stderr
, Name
": (imsm) Operation "
7136 "is not allowed on this container\n");
7139 /* On volume level we support following operations
7140 * - takeover: raid10 -> raid0; raid0 -> raid10
7141 * - chunk size migration
7142 * - migration: raid5 -> raid0; raid0 -> raid5
7144 struct intel_super
*super
= st
->sb
;
7145 struct intel_dev
*dev
= super
->devlist
;
7147 dprintf("imsm: info: Volume operation\n");
7148 /* find requested device */
7150 imsm_find_array_minor_by_subdev(dev
->index
, st
->container_dev
, &devnum
);
7151 if (devnum
== geo
.dev_id
)
7156 fprintf(stderr
, Name
" Cannot find %s (%i) subarray\n",
7157 geo
.dev_name
, geo
.dev_id
);
7158 goto exit_imsm_reshape_super
;
7160 super
->current_vol
= dev
->index
;
7161 change
= imsm_analyze_change(st
, &geo
);
7164 ret_val
= imsm_takeover(st
, &geo
);
7174 exit_imsm_reshape_super
:
7175 dprintf("imsm: reshape_super Exit code = %i\n", ret_val
);
7179 static int imsm_manage_reshape(
7180 int afd
, struct mdinfo
*sra
, struct reshape
*reshape
,
7181 struct supertype
*st
, unsigned long stripes
,
7182 int *fds
, unsigned long long *offsets
,
7183 int dests
, int *destfd
, unsigned long long *destoffsets
)
7185 /* Just use child_monitor for now */
7186 return child_monitor(
7187 afd
, sra
, reshape
, st
, stripes
,
7188 fds
, offsets
, dests
, destfd
, destoffsets
);
7190 #endif /* MDASSEMBLE */
7192 struct superswitch super_imsm
= {
7194 .examine_super
= examine_super_imsm
,
7195 .brief_examine_super
= brief_examine_super_imsm
,
7196 .brief_examine_subarrays
= brief_examine_subarrays_imsm
,
7197 .export_examine_super
= export_examine_super_imsm
,
7198 .detail_super
= detail_super_imsm
,
7199 .brief_detail_super
= brief_detail_super_imsm
,
7200 .write_init_super
= write_init_super_imsm
,
7201 .validate_geometry
= validate_geometry_imsm
,
7202 .add_to_super
= add_to_super_imsm
,
7203 .remove_from_super
= remove_from_super_imsm
,
7204 .detail_platform
= detail_platform_imsm
,
7205 .kill_subarray
= kill_subarray_imsm
,
7206 .update_subarray
= update_subarray_imsm
,
7207 .load_container
= load_container_imsm
,
7208 .default_geometry
= default_geometry_imsm
,
7209 .get_disk_controller_domain
= imsm_get_disk_controller_domain
,
7210 .reshape_super
= imsm_reshape_super
,
7211 .manage_reshape
= imsm_manage_reshape
,
7213 .match_home
= match_home_imsm
,
7214 .uuid_from_super
= uuid_from_super_imsm
,
7215 .getinfo_super
= getinfo_super_imsm
,
7216 .getinfo_super_disks
= getinfo_super_disks_imsm
,
7217 .update_super
= update_super_imsm
,
7219 .avail_size
= avail_size_imsm
,
7220 .min_acceptable_spare_size
= min_acceptable_spare_size_imsm
,
7222 .compare_super
= compare_super_imsm
,
7224 .load_super
= load_super_imsm
,
7225 .init_super
= init_super_imsm
,
7226 .store_super
= store_super_imsm
,
7227 .free_super
= free_super_imsm
,
7228 .match_metadata_desc
= match_metadata_desc_imsm
,
7229 .container_content
= container_content_imsm
,
7236 .open_new
= imsm_open_new
,
7237 .set_array_state
= imsm_set_array_state
,
7238 .set_disk
= imsm_set_disk
,
7239 .sync_metadata
= imsm_sync_metadata
,
7240 .activate_spare
= imsm_activate_spare
,
7241 .process_update
= imsm_process_update
,
7242 .prepare_update
= imsm_prepare_update
,
7243 #endif /* MDASSEMBLE */