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 if (migr_type(dev
) == MIGR_GEN_MIGR
)
1694 return blocks_per_unit
;
1695 stripe
= __le32_to_cpu(map
->blocks_per_strip
) * disks
;
1696 segment
= blocks_per_unit
/ stripe
;
1697 block_rel
= blocks_per_unit
- segment
* stripe
;
1698 parity_depth
= parity_segment_depth(dev
);
1699 block_map
= map_migr_block(dev
, block_rel
);
1700 return block_map
+ parity_depth
* segment
;
1702 case MIGR_REBUILD
: {
1703 __u32 stripes_per_unit
;
1706 stripes_per_unit
= num_stripes_per_unit_rebuild(dev
);
1707 migr_chunk
= migr_strip_blocks_rebuild(dev
);
1708 return migr_chunk
* stripes_per_unit
;
1710 case MIGR_STATE_CHANGE
:
1716 static int imsm_level_to_layout(int level
)
1724 return ALGORITHM_LEFT_ASYMMETRIC
;
1731 static void getinfo_super_imsm_volume(struct supertype
*st
, struct mdinfo
*info
, char *dmap
)
1733 struct intel_super
*super
= st
->sb
;
1734 struct imsm_dev
*dev
= get_imsm_dev(super
, super
->current_vol
);
1735 struct imsm_map
*map
= get_imsm_map(dev
, 0);
1736 struct imsm_map
*prev_map
= get_imsm_map(dev
, 1);
1737 struct imsm_map
*map_to_analyse
= map
;
1740 int map_disks
= info
->array
.raid_disks
;
1743 map_to_analyse
= prev_map
;
1745 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
1746 if (dl
->raiddisk
== info
->disk
.raid_disk
)
1748 info
->container_member
= super
->current_vol
;
1749 info
->array
.raid_disks
= map_to_analyse
->num_members
;
1750 info
->array
.level
= get_imsm_raid_level(map_to_analyse
);
1751 info
->array
.layout
= imsm_level_to_layout(info
->array
.level
);
1752 info
->array
.md_minor
= -1;
1753 info
->array
.ctime
= 0;
1754 info
->array
.utime
= 0;
1755 info
->array
.chunk_size
=
1756 __le16_to_cpu(map_to_analyse
->blocks_per_strip
) << 9;
1757 info
->array
.state
= !dev
->vol
.dirty
;
1758 info
->custom_array_size
= __le32_to_cpu(dev
->size_high
);
1759 info
->custom_array_size
<<= 32;
1760 info
->custom_array_size
|= __le32_to_cpu(dev
->size_low
);
1761 if (prev_map
&& map
->map_state
== prev_map
->map_state
) {
1762 info
->reshape_active
= 1;
1763 info
->new_level
= get_imsm_raid_level(map
);
1764 info
->new_layout
= imsm_level_to_layout(info
->new_level
);
1765 info
->new_chunk
= __le16_to_cpu(map
->blocks_per_strip
) << 9;
1766 info
->delta_disks
= map
->num_members
- prev_map
->num_members
;
1767 /* We shape information that we give to md might have to be
1768 * modify to cope with md's requirement for reshaping arrays.
1769 * For example, when reshaping a RAID0, md requires it to be
1770 * presented as a degraded RAID4.
1771 * Also if a RAID0 is migrating to a RAID5 we need to specify
1772 * the array as already being RAID5, but the 'before' layout
1773 * is a RAID4-like layout.
1775 switch (info
->array
.level
) {
1777 switch(info
->new_level
) {
1779 /* conversion is happening as RAID4 */
1780 info
->array
.level
= 4;
1781 info
->array
.raid_disks
+= 1;
1784 /* conversion is happening as RAID5 */
1785 info
->array
.level
= 5;
1786 info
->array
.layout
= ALGORITHM_PARITY_N
;
1787 info
->array
.raid_disks
+= 1;
1788 info
->delta_disks
-= 1;
1791 /* FIXME error message */
1792 info
->array
.level
= UnSet
;
1798 info
->new_level
= UnSet
;
1799 info
->new_layout
= UnSet
;
1800 info
->new_chunk
= info
->array
.chunk_size
;
1801 info
->delta_disks
= 0;
1803 info
->disk
.major
= 0;
1804 info
->disk
.minor
= 0;
1806 info
->disk
.major
= dl
->major
;
1807 info
->disk
.minor
= dl
->minor
;
1810 info
->data_offset
= __le32_to_cpu(map_to_analyse
->pba_of_lba0
);
1811 info
->component_size
=
1812 __le32_to_cpu(map_to_analyse
->blocks_per_member
);
1813 memset(info
->uuid
, 0, sizeof(info
->uuid
));
1814 info
->recovery_start
= MaxSector
;
1816 info
->reshape_progress
= 0;
1817 if (map_to_analyse
->map_state
== IMSM_T_STATE_UNINITIALIZED
||
1819 info
->resync_start
= 0;
1820 } else if (dev
->vol
.migr_state
) {
1821 switch (migr_type(dev
)) {
1824 __u64 blocks_per_unit
= blocks_per_migr_unit(dev
);
1825 __u64 units
= __le32_to_cpu(dev
->vol
.curr_migr_unit
);
1827 info
->resync_start
= blocks_per_unit
* units
;
1830 case MIGR_GEN_MIGR
: {
1831 __u64 blocks_per_unit
= blocks_per_migr_unit(dev
);
1832 __u64 units
= __le32_to_cpu(dev
->vol
.curr_migr_unit
);
1833 unsigned long long array_blocks
;
1836 info
->reshape_progress
= blocks_per_unit
* units
;
1837 dprintf("IMSM: General Migration checkpoint : %llu "
1838 "(%llu) -> read reshape progress : %llu\n",
1839 units
, blocks_per_unit
, info
->reshape_progress
);
1841 used_disks
= imsm_num_data_members(dev
, 1);
1842 if (used_disks
> 0) {
1843 array_blocks
= map
->blocks_per_member
*
1845 /* round array size down to closest MB
1847 info
->custom_array_size
= (array_blocks
1848 >> SECT_PER_MB_SHIFT
)
1849 << SECT_PER_MB_SHIFT
;
1853 /* we could emulate the checkpointing of
1854 * 'sync_action=check' migrations, but for now
1855 * we just immediately complete them
1858 /* this is handled by container_content_imsm() */
1859 case MIGR_STATE_CHANGE
:
1860 /* FIXME handle other migrations */
1862 /* we are not dirty, so... */
1863 info
->resync_start
= MaxSector
;
1866 info
->resync_start
= MaxSector
;
1868 strncpy(info
->name
, (char *) dev
->volume
, MAX_RAID_SERIAL_LEN
);
1869 info
->name
[MAX_RAID_SERIAL_LEN
] = 0;
1871 info
->array
.major_version
= -1;
1872 info
->array
.minor_version
= -2;
1873 devname
= devnum2devname(st
->container_dev
);
1874 *info
->text_version
= '\0';
1876 sprintf(info
->text_version
, "/%s/%d", devname
, info
->container_member
);
1878 info
->safe_mode_delay
= 4000; /* 4 secs like the Matrix driver */
1879 uuid_from_super_imsm(st
, info
->uuid
);
1883 for (i
=0; i
<map_disks
; i
++) {
1885 if (i
< info
->array
.raid_disks
) {
1886 struct imsm_disk
*dsk
;
1887 j
= get_imsm_disk_idx(dev
, i
, -1);
1888 dsk
= get_imsm_disk(super
, j
);
1889 if (dsk
&& (dsk
->status
& CONFIGURED_DISK
))
1896 static __u8
imsm_check_degraded(struct intel_super
*super
, struct imsm_dev
*dev
, int failed
);
1897 static int imsm_count_failed(struct intel_super
*super
, struct imsm_dev
*dev
);
1899 static struct imsm_disk
*get_imsm_missing(struct intel_super
*super
, __u8 index
)
1903 for (d
= super
->missing
; d
; d
= d
->next
)
1904 if (d
->index
== index
)
1909 static void getinfo_super_imsm(struct supertype
*st
, struct mdinfo
*info
, char *map
)
1911 struct intel_super
*super
= st
->sb
;
1912 struct imsm_disk
*disk
;
1913 int map_disks
= info
->array
.raid_disks
;
1914 int max_enough
= -1;
1916 struct imsm_super
*mpb
;
1918 if (super
->current_vol
>= 0) {
1919 getinfo_super_imsm_volume(st
, info
, map
);
1923 /* Set raid_disks to zero so that Assemble will always pull in valid
1926 info
->array
.raid_disks
= 0;
1927 info
->array
.level
= LEVEL_CONTAINER
;
1928 info
->array
.layout
= 0;
1929 info
->array
.md_minor
= -1;
1930 info
->array
.ctime
= 0; /* N/A for imsm */
1931 info
->array
.utime
= 0;
1932 info
->array
.chunk_size
= 0;
1934 info
->disk
.major
= 0;
1935 info
->disk
.minor
= 0;
1936 info
->disk
.raid_disk
= -1;
1937 info
->reshape_active
= 0;
1938 info
->array
.major_version
= -1;
1939 info
->array
.minor_version
= -2;
1940 strcpy(info
->text_version
, "imsm");
1941 info
->safe_mode_delay
= 0;
1942 info
->disk
.number
= -1;
1943 info
->disk
.state
= 0;
1945 info
->recovery_start
= MaxSector
;
1947 /* do we have the all the insync disks that we expect? */
1948 mpb
= super
->anchor
;
1950 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
1951 struct imsm_dev
*dev
= get_imsm_dev(super
, i
);
1952 int failed
, enough
, j
, missing
= 0;
1953 struct imsm_map
*map
;
1956 failed
= imsm_count_failed(super
, dev
);
1957 state
= imsm_check_degraded(super
, dev
, failed
);
1958 map
= get_imsm_map(dev
, dev
->vol
.migr_state
);
1960 /* any newly missing disks?
1961 * (catches single-degraded vs double-degraded)
1963 for (j
= 0; j
< map
->num_members
; j
++) {
1964 __u32 ord
= get_imsm_ord_tbl_ent(dev
, i
, -1);
1965 __u32 idx
= ord_to_idx(ord
);
1967 if (!(ord
& IMSM_ORD_REBUILD
) &&
1968 get_imsm_missing(super
, idx
)) {
1974 if (state
== IMSM_T_STATE_FAILED
)
1976 else if (state
== IMSM_T_STATE_DEGRADED
&&
1977 (state
!= map
->map_state
|| missing
))
1979 else /* we're normal, or already degraded */
1982 /* in the missing/failed disk case check to see
1983 * if at least one array is runnable
1985 max_enough
= max(max_enough
, enough
);
1987 dprintf("%s: enough: %d\n", __func__
, max_enough
);
1988 info
->container_enough
= max_enough
;
1991 __u32 reserved
= imsm_reserved_sectors(super
, super
->disks
);
1993 disk
= &super
->disks
->disk
;
1994 info
->data_offset
= __le32_to_cpu(disk
->total_blocks
) - reserved
;
1995 info
->component_size
= reserved
;
1996 info
->disk
.state
= is_configured(disk
) ? (1 << MD_DISK_ACTIVE
) : 0;
1997 /* we don't change info->disk.raid_disk here because
1998 * this state will be finalized in mdmon after we have
1999 * found the 'most fresh' version of the metadata
2001 info
->disk
.state
|= is_failed(disk
) ? (1 << MD_DISK_FAULTY
) : 0;
2002 info
->disk
.state
|= is_spare(disk
) ? 0 : (1 << MD_DISK_SYNC
);
2005 /* only call uuid_from_super_imsm when this disk is part of a populated container,
2006 * ->compare_super may have updated the 'num_raid_devs' field for spares
2008 if (info
->disk
.state
& (1 << MD_DISK_SYNC
) || super
->anchor
->num_raid_devs
)
2009 uuid_from_super_imsm(st
, info
->uuid
);
2011 memcpy(info
->uuid
, uuid_zero
, sizeof(uuid_zero
));
2013 /* I don't know how to compute 'map' on imsm, so use safe default */
2016 for (i
= 0; i
< map_disks
; i
++)
2022 /* allocates memory and fills disk in mdinfo structure
2023 * for each disk in array */
2024 struct mdinfo
*getinfo_super_disks_imsm(struct supertype
*st
)
2026 struct mdinfo
*mddev
= NULL
;
2027 struct intel_super
*super
= st
->sb
;
2028 struct imsm_disk
*disk
;
2031 if (!super
|| !super
->disks
)
2034 mddev
= malloc(sizeof(*mddev
));
2036 fprintf(stderr
, Name
": Failed to allocate memory.\n");
2039 memset(mddev
, 0, sizeof(*mddev
));
2043 tmp
= malloc(sizeof(*tmp
));
2045 fprintf(stderr
, Name
": Failed to allocate memory.\n");
2050 memset(tmp
, 0, sizeof(*tmp
));
2052 tmp
->next
= mddev
->devs
;
2054 tmp
->disk
.number
= count
++;
2055 tmp
->disk
.major
= dl
->major
;
2056 tmp
->disk
.minor
= dl
->minor
;
2057 tmp
->disk
.state
= is_configured(disk
) ?
2058 (1 << MD_DISK_ACTIVE
) : 0;
2059 tmp
->disk
.state
|= is_failed(disk
) ? (1 << MD_DISK_FAULTY
) : 0;
2060 tmp
->disk
.state
|= is_spare(disk
) ? 0 : (1 << MD_DISK_SYNC
);
2061 tmp
->disk
.raid_disk
= -1;
2067 static int update_super_imsm(struct supertype
*st
, struct mdinfo
*info
,
2068 char *update
, char *devname
, int verbose
,
2069 int uuid_set
, char *homehost
)
2071 /* For 'assemble' and 'force' we need to return non-zero if any
2072 * change was made. For others, the return value is ignored.
2073 * Update options are:
2074 * force-one : This device looks a bit old but needs to be included,
2075 * update age info appropriately.
2076 * assemble: clear any 'faulty' flag to allow this device to
2078 * force-array: Array is degraded but being forced, mark it clean
2079 * if that will be needed to assemble it.
2081 * newdev: not used ????
2082 * grow: Array has gained a new device - this is currently for
2084 * resync: mark as dirty so a resync will happen.
2085 * name: update the name - preserving the homehost
2086 * uuid: Change the uuid of the array to match watch is given
2088 * Following are not relevant for this imsm:
2089 * sparc2.2 : update from old dodgey metadata
2090 * super-minor: change the preferred_minor number
2091 * summaries: update redundant counters.
2092 * homehost: update the recorded homehost
2093 * _reshape_progress: record new reshape_progress position.
2096 struct intel_super
*super
= st
->sb
;
2097 struct imsm_super
*mpb
;
2099 /* we can only update container info */
2100 if (!super
|| super
->current_vol
>= 0 || !super
->anchor
)
2103 mpb
= super
->anchor
;
2105 if (strcmp(update
, "uuid") == 0 && uuid_set
&& !info
->update_private
)
2107 else if (strcmp(update
, "uuid") == 0 && uuid_set
&& info
->update_private
) {
2108 mpb
->orig_family_num
= *((__u32
*) info
->update_private
);
2110 } else if (strcmp(update
, "uuid") == 0) {
2111 __u32
*new_family
= malloc(sizeof(*new_family
));
2113 /* update orig_family_number with the incoming random
2114 * data, report the new effective uuid, and store the
2115 * new orig_family_num for future updates.
2118 memcpy(&mpb
->orig_family_num
, info
->uuid
, sizeof(__u32
));
2119 uuid_from_super_imsm(st
, info
->uuid
);
2120 *new_family
= mpb
->orig_family_num
;
2121 info
->update_private
= new_family
;
2124 } else if (strcmp(update
, "assemble") == 0)
2129 /* successful update? recompute checksum */
2131 mpb
->check_sum
= __le32_to_cpu(__gen_imsm_checksum(mpb
));
2136 static size_t disks_to_mpb_size(int disks
)
2140 size
= sizeof(struct imsm_super
);
2141 size
+= (disks
- 1) * sizeof(struct imsm_disk
);
2142 size
+= 2 * sizeof(struct imsm_dev
);
2143 /* up to 2 maps per raid device (-2 for imsm_maps in imsm_dev */
2144 size
+= (4 - 2) * sizeof(struct imsm_map
);
2145 /* 4 possible disk_ord_tbl's */
2146 size
+= 4 * (disks
- 1) * sizeof(__u32
);
2151 static __u64
avail_size_imsm(struct supertype
*st
, __u64 devsize
)
2153 if (devsize
< (MPB_SECTOR_CNT
+ IMSM_RESERVED_SECTORS
))
2156 return devsize
- (MPB_SECTOR_CNT
+ IMSM_RESERVED_SECTORS
);
2159 static void free_devlist(struct intel_super
*super
)
2161 struct intel_dev
*dv
;
2163 while (super
->devlist
) {
2164 dv
= super
->devlist
->next
;
2165 free(super
->devlist
->dev
);
2166 free(super
->devlist
);
2167 super
->devlist
= dv
;
2171 static void imsm_copy_dev(struct imsm_dev
*dest
, struct imsm_dev
*src
)
2173 memcpy(dest
, src
, sizeof_imsm_dev(src
, 0));
2176 static int compare_super_imsm(struct supertype
*st
, struct supertype
*tst
)
2180 * 0 same, or first was empty, and second was copied
2181 * 1 second had wrong number
2183 * 3 wrong other info
2185 struct intel_super
*first
= st
->sb
;
2186 struct intel_super
*sec
= tst
->sb
;
2194 /* if an anchor does not have num_raid_devs set then it is a free
2197 if (first
->anchor
->num_raid_devs
> 0 &&
2198 sec
->anchor
->num_raid_devs
> 0) {
2199 /* Determine if these disks might ever have been
2200 * related. Further disambiguation can only take place
2201 * in load_super_imsm_all
2203 __u32 first_family
= first
->anchor
->orig_family_num
;
2204 __u32 sec_family
= sec
->anchor
->orig_family_num
;
2206 if (memcmp(first
->anchor
->sig
, sec
->anchor
->sig
,
2207 MAX_SIGNATURE_LENGTH
) != 0)
2210 if (first_family
== 0)
2211 first_family
= first
->anchor
->family_num
;
2212 if (sec_family
== 0)
2213 sec_family
= sec
->anchor
->family_num
;
2215 if (first_family
!= sec_family
)
2221 /* if 'first' is a spare promote it to a populated mpb with sec's
2224 if (first
->anchor
->num_raid_devs
== 0 &&
2225 sec
->anchor
->num_raid_devs
> 0) {
2227 struct intel_dev
*dv
;
2228 struct imsm_dev
*dev
;
2230 /* we need to copy raid device info from sec if an allocation
2231 * fails here we don't associate the spare
2233 for (i
= 0; i
< sec
->anchor
->num_raid_devs
; i
++) {
2234 dv
= malloc(sizeof(*dv
));
2237 dev
= malloc(sizeof_imsm_dev(get_imsm_dev(sec
, i
), 1));
2244 dv
->next
= first
->devlist
;
2245 first
->devlist
= dv
;
2247 if (i
< sec
->anchor
->num_raid_devs
) {
2248 /* allocation failure */
2249 free_devlist(first
);
2250 fprintf(stderr
, "imsm: failed to associate spare\n");
2253 first
->anchor
->num_raid_devs
= sec
->anchor
->num_raid_devs
;
2254 first
->anchor
->orig_family_num
= sec
->anchor
->orig_family_num
;
2255 first
->anchor
->family_num
= sec
->anchor
->family_num
;
2256 memcpy(first
->anchor
->sig
, sec
->anchor
->sig
, MAX_SIGNATURE_LENGTH
);
2257 for (i
= 0; i
< sec
->anchor
->num_raid_devs
; i
++)
2258 imsm_copy_dev(get_imsm_dev(first
, i
), get_imsm_dev(sec
, i
));
2264 static void fd2devname(int fd
, char *name
)
2268 char dname
[PATH_MAX
];
2273 if (fstat(fd
, &st
) != 0)
2275 sprintf(path
, "/sys/dev/block/%d:%d",
2276 major(st
.st_rdev
), minor(st
.st_rdev
));
2278 rv
= readlink(path
, dname
, sizeof(dname
));
2283 nm
= strrchr(dname
, '/');
2285 snprintf(name
, MAX_RAID_SERIAL_LEN
, "/dev/%s", nm
);
2288 extern int scsi_get_serial(int fd
, void *buf
, size_t buf_len
);
2290 static int imsm_read_serial(int fd
, char *devname
,
2291 __u8 serial
[MAX_RAID_SERIAL_LEN
])
2293 unsigned char scsi_serial
[255];
2302 memset(scsi_serial
, 0, sizeof(scsi_serial
));
2304 rv
= scsi_get_serial(fd
, scsi_serial
, sizeof(scsi_serial
));
2306 if (rv
&& check_env("IMSM_DEVNAME_AS_SERIAL")) {
2307 memset(serial
, 0, MAX_RAID_SERIAL_LEN
);
2308 fd2devname(fd
, (char *) serial
);
2315 Name
": Failed to retrieve serial for %s\n",
2320 rsp_len
= scsi_serial
[3];
2324 Name
": Failed to retrieve serial for %s\n",
2328 rsp_buf
= (char *) &scsi_serial
[4];
2330 /* trim all whitespace and non-printable characters and convert
2333 for (i
= 0, dest
= rsp_buf
; i
< rsp_len
; i
++) {
2336 /* ':' is reserved for use in placeholder serial
2337 * numbers for missing disks
2345 len
= dest
- rsp_buf
;
2348 /* truncate leading characters */
2349 if (len
> MAX_RAID_SERIAL_LEN
) {
2350 dest
+= len
- MAX_RAID_SERIAL_LEN
;
2351 len
= MAX_RAID_SERIAL_LEN
;
2354 memset(serial
, 0, MAX_RAID_SERIAL_LEN
);
2355 memcpy(serial
, dest
, len
);
2360 static int serialcmp(__u8
*s1
, __u8
*s2
)
2362 return strncmp((char *) s1
, (char *) s2
, MAX_RAID_SERIAL_LEN
);
2365 static void serialcpy(__u8
*dest
, __u8
*src
)
2367 strncpy((char *) dest
, (char *) src
, MAX_RAID_SERIAL_LEN
);
2371 static struct dl
*serial_to_dl(__u8
*serial
, struct intel_super
*super
)
2375 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
2376 if (serialcmp(dl
->serial
, serial
) == 0)
2383 static struct imsm_disk
*
2384 __serial_to_disk(__u8
*serial
, struct imsm_super
*mpb
, int *idx
)
2388 for (i
= 0; i
< mpb
->num_disks
; i
++) {
2389 struct imsm_disk
*disk
= __get_imsm_disk(mpb
, i
);
2391 if (serialcmp(disk
->serial
, serial
) == 0) {
2402 load_imsm_disk(int fd
, struct intel_super
*super
, char *devname
, int keep_fd
)
2404 struct imsm_disk
*disk
;
2409 __u8 serial
[MAX_RAID_SERIAL_LEN
];
2411 rv
= imsm_read_serial(fd
, devname
, serial
);
2416 dl
= calloc(1, sizeof(*dl
));
2420 Name
": failed to allocate disk buffer for %s\n",
2426 dl
->major
= major(stb
.st_rdev
);
2427 dl
->minor
= minor(stb
.st_rdev
);
2428 dl
->next
= super
->disks
;
2429 dl
->fd
= keep_fd
? fd
: -1;
2430 assert(super
->disks
== NULL
);
2432 serialcpy(dl
->serial
, serial
);
2435 fd2devname(fd
, name
);
2437 dl
->devname
= strdup(devname
);
2439 dl
->devname
= strdup(name
);
2441 /* look up this disk's index in the current anchor */
2442 disk
= __serial_to_disk(dl
->serial
, super
->anchor
, &dl
->index
);
2445 /* only set index on disks that are a member of a
2446 * populated contianer, i.e. one with raid_devs
2448 if (is_failed(&dl
->disk
))
2450 else if (is_spare(&dl
->disk
))
2458 /* When migrating map0 contains the 'destination' state while map1
2459 * contains the current state. When not migrating map0 contains the
2460 * current state. This routine assumes that map[0].map_state is set to
2461 * the current array state before being called.
2463 * Migration is indicated by one of the following states
2464 * 1/ Idle (migr_state=0 map0state=normal||unitialized||degraded||failed)
2465 * 2/ Initialize (migr_state=1 migr_type=MIGR_INIT map0state=normal
2466 * map1state=unitialized)
2467 * 3/ Repair (Resync) (migr_state=1 migr_type=MIGR_REPAIR map0state=normal
2469 * 4/ Rebuild (migr_state=1 migr_type=MIGR_REBUILD map0state=normal
2470 * map1state=degraded)
2472 static void migrate(struct imsm_dev
*dev
, __u8 to_state
, int migr_type
)
2474 struct imsm_map
*dest
;
2475 struct imsm_map
*src
= get_imsm_map(dev
, 0);
2477 dev
->vol
.migr_state
= 1;
2478 set_migr_type(dev
, migr_type
);
2479 dev
->vol
.curr_migr_unit
= 0;
2480 dest
= get_imsm_map(dev
, 1);
2482 /* duplicate and then set the target end state in map[0] */
2483 memcpy(dest
, src
, sizeof_imsm_map(src
));
2484 if ((migr_type
== MIGR_REBUILD
) ||
2485 (migr_type
== MIGR_GEN_MIGR
)) {
2489 for (i
= 0; i
< src
->num_members
; i
++) {
2490 ord
= __le32_to_cpu(src
->disk_ord_tbl
[i
]);
2491 set_imsm_ord_tbl_ent(src
, i
, ord_to_idx(ord
));
2495 src
->map_state
= to_state
;
2498 static void end_migration(struct imsm_dev
*dev
, __u8 map_state
)
2500 struct imsm_map
*map
= get_imsm_map(dev
, 0);
2501 struct imsm_map
*prev
= get_imsm_map(dev
, dev
->vol
.migr_state
);
2504 /* merge any IMSM_ORD_REBUILD bits that were not successfully
2505 * completed in the last migration.
2507 * FIXME add support for raid-level-migration
2509 for (i
= 0; i
< prev
->num_members
; i
++)
2510 for (j
= 0; j
< map
->num_members
; j
++)
2511 /* during online capacity expansion
2512 * disks position can be changed if takeover is used
2514 if (ord_to_idx(map
->disk_ord_tbl
[j
]) ==
2515 ord_to_idx(prev
->disk_ord_tbl
[i
])) {
2516 map
->disk_ord_tbl
[j
] |= prev
->disk_ord_tbl
[i
];
2520 dev
->vol
.migr_state
= 0;
2521 dev
->vol
.migr_type
= 0;
2522 dev
->vol
.curr_migr_unit
= 0;
2523 map
->map_state
= map_state
;
2527 static int parse_raid_devices(struct intel_super
*super
)
2530 struct imsm_dev
*dev_new
;
2531 size_t len
, len_migr
;
2533 size_t space_needed
= 0;
2534 struct imsm_super
*mpb
= super
->anchor
;
2536 for (i
= 0; i
< super
->anchor
->num_raid_devs
; i
++) {
2537 struct imsm_dev
*dev_iter
= __get_imsm_dev(super
->anchor
, i
);
2538 struct intel_dev
*dv
;
2540 len
= sizeof_imsm_dev(dev_iter
, 0);
2541 len_migr
= sizeof_imsm_dev(dev_iter
, 1);
2543 space_needed
+= len_migr
- len
;
2545 dv
= malloc(sizeof(*dv
));
2548 if (max_len
< len_migr
)
2550 if (max_len
> len_migr
)
2551 space_needed
+= max_len
- len_migr
;
2552 dev_new
= malloc(max_len
);
2557 imsm_copy_dev(dev_new
, dev_iter
);
2560 dv
->next
= super
->devlist
;
2561 super
->devlist
= dv
;
2564 /* ensure that super->buf is large enough when all raid devices
2567 if (__le32_to_cpu(mpb
->mpb_size
) + space_needed
> super
->len
) {
2570 len
= ROUND_UP(__le32_to_cpu(mpb
->mpb_size
) + space_needed
, 512);
2571 if (posix_memalign(&buf
, 512, len
) != 0)
2574 memcpy(buf
, super
->buf
, super
->len
);
2575 memset(buf
+ super
->len
, 0, len
- super
->len
);
2584 /* retrieve a pointer to the bbm log which starts after all raid devices */
2585 struct bbm_log
*__get_imsm_bbm_log(struct imsm_super
*mpb
)
2589 if (__le32_to_cpu(mpb
->bbm_log_size
)) {
2591 ptr
+= mpb
->mpb_size
- __le32_to_cpu(mpb
->bbm_log_size
);
2597 static void __free_imsm(struct intel_super
*super
, int free_disks
);
2599 /* load_imsm_mpb - read matrix metadata
2600 * allocates super->mpb to be freed by free_super
2602 static int load_imsm_mpb(int fd
, struct intel_super
*super
, char *devname
)
2604 unsigned long long dsize
;
2605 unsigned long long sectors
;
2607 struct imsm_super
*anchor
;
2610 get_dev_size(fd
, NULL
, &dsize
);
2614 Name
": %s: device to small for imsm\n",
2619 if (lseek64(fd
, dsize
- (512 * 2), SEEK_SET
) < 0) {
2622 Name
": Cannot seek to anchor block on %s: %s\n",
2623 devname
, strerror(errno
));
2627 if (posix_memalign((void**)&anchor
, 512, 512) != 0) {
2630 Name
": Failed to allocate imsm anchor buffer"
2631 " on %s\n", devname
);
2634 if (read(fd
, anchor
, 512) != 512) {
2637 Name
": Cannot read anchor block on %s: %s\n",
2638 devname
, strerror(errno
));
2643 if (strncmp((char *) anchor
->sig
, MPB_SIGNATURE
, MPB_SIG_LEN
) != 0) {
2646 Name
": no IMSM anchor on %s\n", devname
);
2651 __free_imsm(super
, 0);
2652 super
->len
= ROUND_UP(anchor
->mpb_size
, 512);
2653 if (posix_memalign(&super
->buf
, 512, super
->len
) != 0) {
2656 Name
": unable to allocate %zu byte mpb buffer\n",
2661 memcpy(super
->buf
, anchor
, 512);
2663 sectors
= mpb_sectors(anchor
) - 1;
2666 check_sum
= __gen_imsm_checksum(super
->anchor
);
2667 if (check_sum
!= __le32_to_cpu(super
->anchor
->check_sum
)) {
2670 Name
": IMSM checksum %x != %x on %s\n",
2672 __le32_to_cpu(super
->anchor
->check_sum
),
2680 /* read the extended mpb */
2681 if (lseek64(fd
, dsize
- (512 * (2 + sectors
)), SEEK_SET
) < 0) {
2684 Name
": Cannot seek to extended mpb on %s: %s\n",
2685 devname
, strerror(errno
));
2689 if ((unsigned)read(fd
, super
->buf
+ 512, super
->len
- 512) != super
->len
- 512) {
2692 Name
": Cannot read extended mpb on %s: %s\n",
2693 devname
, strerror(errno
));
2697 check_sum
= __gen_imsm_checksum(super
->anchor
);
2698 if (check_sum
!= __le32_to_cpu(super
->anchor
->check_sum
)) {
2701 Name
": IMSM checksum %x != %x on %s\n",
2702 check_sum
, __le32_to_cpu(super
->anchor
->check_sum
),
2707 /* FIXME the BBM log is disk specific so we cannot use this global
2708 * buffer for all disks. Ok for now since we only look at the global
2709 * bbm_log_size parameter to gate assembly
2711 super
->bbm_log
= __get_imsm_bbm_log(super
->anchor
);
2717 load_and_parse_mpb(int fd
, struct intel_super
*super
, char *devname
, int keep_fd
)
2721 err
= load_imsm_mpb(fd
, super
, devname
);
2724 err
= load_imsm_disk(fd
, super
, devname
, keep_fd
);
2727 err
= parse_raid_devices(super
);
2732 static void __free_imsm_disk(struct dl
*d
)
2744 static void free_imsm_disks(struct intel_super
*super
)
2748 while (super
->disks
) {
2750 super
->disks
= d
->next
;
2751 __free_imsm_disk(d
);
2753 while (super
->disk_mgmt_list
) {
2754 d
= super
->disk_mgmt_list
;
2755 super
->disk_mgmt_list
= d
->next
;
2756 __free_imsm_disk(d
);
2758 while (super
->missing
) {
2760 super
->missing
= d
->next
;
2761 __free_imsm_disk(d
);
2766 /* free all the pieces hanging off of a super pointer */
2767 static void __free_imsm(struct intel_super
*super
, int free_disks
)
2769 struct intel_hba
*elem
, *next
;
2776 free_imsm_disks(super
);
2777 free_devlist(super
);
2781 free((void *)elem
->path
);
2789 static void free_imsm(struct intel_super
*super
)
2791 __free_imsm(super
, 1);
2795 static void free_super_imsm(struct supertype
*st
)
2797 struct intel_super
*super
= st
->sb
;
2806 static struct intel_super
*alloc_super(void)
2808 struct intel_super
*super
= malloc(sizeof(*super
));
2811 memset(super
, 0, sizeof(*super
));
2812 super
->current_vol
= -1;
2813 super
->create_offset
= ~((__u32
) 0);
2814 if (!check_env("IMSM_NO_PLATFORM"))
2815 super
->orom
= find_imsm_orom();
2822 /* find_missing - helper routine for load_super_imsm_all that identifies
2823 * disks that have disappeared from the system. This routine relies on
2824 * the mpb being uptodate, which it is at load time.
2826 static int find_missing(struct intel_super
*super
)
2829 struct imsm_super
*mpb
= super
->anchor
;
2831 struct imsm_disk
*disk
;
2833 for (i
= 0; i
< mpb
->num_disks
; i
++) {
2834 disk
= __get_imsm_disk(mpb
, i
);
2835 dl
= serial_to_dl(disk
->serial
, super
);
2839 dl
= malloc(sizeof(*dl
));
2845 dl
->devname
= strdup("missing");
2847 serialcpy(dl
->serial
, disk
->serial
);
2850 dl
->next
= super
->missing
;
2851 super
->missing
= dl
;
2857 static struct intel_disk
*disk_list_get(__u8
*serial
, struct intel_disk
*disk_list
)
2859 struct intel_disk
*idisk
= disk_list
;
2862 if (serialcmp(idisk
->disk
.serial
, serial
) == 0)
2864 idisk
= idisk
->next
;
2870 static int __prep_thunderdome(struct intel_super
**table
, int tbl_size
,
2871 struct intel_super
*super
,
2872 struct intel_disk
**disk_list
)
2874 struct imsm_disk
*d
= &super
->disks
->disk
;
2875 struct imsm_super
*mpb
= super
->anchor
;
2878 for (i
= 0; i
< tbl_size
; i
++) {
2879 struct imsm_super
*tbl_mpb
= table
[i
]->anchor
;
2880 struct imsm_disk
*tbl_d
= &table
[i
]->disks
->disk
;
2882 if (tbl_mpb
->family_num
== mpb
->family_num
) {
2883 if (tbl_mpb
->check_sum
== mpb
->check_sum
) {
2884 dprintf("%s: mpb from %d:%d matches %d:%d\n",
2885 __func__
, super
->disks
->major
,
2886 super
->disks
->minor
,
2887 table
[i
]->disks
->major
,
2888 table
[i
]->disks
->minor
);
2892 if (((is_configured(d
) && !is_configured(tbl_d
)) ||
2893 is_configured(d
) == is_configured(tbl_d
)) &&
2894 tbl_mpb
->generation_num
< mpb
->generation_num
) {
2895 /* current version of the mpb is a
2896 * better candidate than the one in
2897 * super_table, but copy over "cross
2898 * generational" status
2900 struct intel_disk
*idisk
;
2902 dprintf("%s: mpb from %d:%d replaces %d:%d\n",
2903 __func__
, super
->disks
->major
,
2904 super
->disks
->minor
,
2905 table
[i
]->disks
->major
,
2906 table
[i
]->disks
->minor
);
2908 idisk
= disk_list_get(tbl_d
->serial
, *disk_list
);
2909 if (idisk
&& is_failed(&idisk
->disk
))
2910 tbl_d
->status
|= FAILED_DISK
;
2913 struct intel_disk
*idisk
;
2914 struct imsm_disk
*disk
;
2916 /* tbl_mpb is more up to date, but copy
2917 * over cross generational status before
2920 disk
= __serial_to_disk(d
->serial
, mpb
, NULL
);
2921 if (disk
&& is_failed(disk
))
2922 d
->status
|= FAILED_DISK
;
2924 idisk
= disk_list_get(d
->serial
, *disk_list
);
2927 if (disk
&& is_configured(disk
))
2928 idisk
->disk
.status
|= CONFIGURED_DISK
;
2931 dprintf("%s: mpb from %d:%d prefer %d:%d\n",
2932 __func__
, super
->disks
->major
,
2933 super
->disks
->minor
,
2934 table
[i
]->disks
->major
,
2935 table
[i
]->disks
->minor
);
2943 table
[tbl_size
++] = super
;
2947 /* update/extend the merged list of imsm_disk records */
2948 for (j
= 0; j
< mpb
->num_disks
; j
++) {
2949 struct imsm_disk
*disk
= __get_imsm_disk(mpb
, j
);
2950 struct intel_disk
*idisk
;
2952 idisk
= disk_list_get(disk
->serial
, *disk_list
);
2954 idisk
->disk
.status
|= disk
->status
;
2955 if (is_configured(&idisk
->disk
) ||
2956 is_failed(&idisk
->disk
))
2957 idisk
->disk
.status
&= ~(SPARE_DISK
);
2959 idisk
= calloc(1, sizeof(*idisk
));
2962 idisk
->owner
= IMSM_UNKNOWN_OWNER
;
2963 idisk
->disk
= *disk
;
2964 idisk
->next
= *disk_list
;
2968 if (serialcmp(idisk
->disk
.serial
, d
->serial
) == 0)
2975 static struct intel_super
*
2976 validate_members(struct intel_super
*super
, struct intel_disk
*disk_list
,
2979 struct imsm_super
*mpb
= super
->anchor
;
2983 for (i
= 0; i
< mpb
->num_disks
; i
++) {
2984 struct imsm_disk
*disk
= __get_imsm_disk(mpb
, i
);
2985 struct intel_disk
*idisk
;
2987 idisk
= disk_list_get(disk
->serial
, disk_list
);
2989 if (idisk
->owner
== owner
||
2990 idisk
->owner
== IMSM_UNKNOWN_OWNER
)
2993 dprintf("%s: '%.16s' owner %d != %d\n",
2994 __func__
, disk
->serial
, idisk
->owner
,
2997 dprintf("%s: unknown disk %x [%d]: %.16s\n",
2998 __func__
, __le32_to_cpu(mpb
->family_num
), i
,
3004 if (ok_count
== mpb
->num_disks
)
3009 static void show_conflicts(__u32 family_num
, struct intel_super
*super_list
)
3011 struct intel_super
*s
;
3013 for (s
= super_list
; s
; s
= s
->next
) {
3014 if (family_num
!= s
->anchor
->family_num
)
3016 fprintf(stderr
, "Conflict, offlining family %#x on '%s'\n",
3017 __le32_to_cpu(family_num
), s
->disks
->devname
);
3021 static struct intel_super
*
3022 imsm_thunderdome(struct intel_super
**super_list
, int len
)
3024 struct intel_super
*super_table
[len
];
3025 struct intel_disk
*disk_list
= NULL
;
3026 struct intel_super
*champion
, *spare
;
3027 struct intel_super
*s
, **del
;
3032 memset(super_table
, 0, sizeof(super_table
));
3033 for (s
= *super_list
; s
; s
= s
->next
)
3034 tbl_size
= __prep_thunderdome(super_table
, tbl_size
, s
, &disk_list
);
3036 for (i
= 0; i
< tbl_size
; i
++) {
3037 struct imsm_disk
*d
;
3038 struct intel_disk
*idisk
;
3039 struct imsm_super
*mpb
= super_table
[i
]->anchor
;
3042 d
= &s
->disks
->disk
;
3044 /* 'd' must appear in merged disk list for its
3045 * configuration to be valid
3047 idisk
= disk_list_get(d
->serial
, disk_list
);
3048 if (idisk
&& idisk
->owner
== i
)
3049 s
= validate_members(s
, disk_list
, i
);
3054 dprintf("%s: marking family: %#x from %d:%d offline\n",
3055 __func__
, mpb
->family_num
,
3056 super_table
[i
]->disks
->major
,
3057 super_table
[i
]->disks
->minor
);
3061 /* This is where the mdadm implementation differs from the Windows
3062 * driver which has no strict concept of a container. We can only
3063 * assemble one family from a container, so when returning a prodigal
3064 * array member to this system the code will not be able to disambiguate
3065 * the container contents that should be assembled ("foreign" versus
3066 * "local"). It requires user intervention to set the orig_family_num
3067 * to a new value to establish a new container. The Windows driver in
3068 * this situation fixes up the volume name in place and manages the
3069 * foreign array as an independent entity.
3074 for (i
= 0; i
< tbl_size
; i
++) {
3075 struct intel_super
*tbl_ent
= super_table
[i
];
3081 if (tbl_ent
->anchor
->num_raid_devs
== 0) {
3086 if (s
&& !is_spare
) {
3087 show_conflicts(tbl_ent
->anchor
->family_num
, *super_list
);
3089 } else if (!s
&& !is_spare
)
3102 fprintf(stderr
, "Chose family %#x on '%s', "
3103 "assemble conflicts to new container with '--update=uuid'\n",
3104 __le32_to_cpu(s
->anchor
->family_num
), s
->disks
->devname
);
3106 /* collect all dl's onto 'champion', and update them to
3107 * champion's version of the status
3109 for (s
= *super_list
; s
; s
= s
->next
) {
3110 struct imsm_super
*mpb
= champion
->anchor
;
3111 struct dl
*dl
= s
->disks
;
3116 for (i
= 0; i
< mpb
->num_disks
; i
++) {
3117 struct imsm_disk
*disk
;
3119 disk
= __serial_to_disk(dl
->serial
, mpb
, &dl
->index
);
3122 /* only set index on disks that are a member of
3123 * a populated contianer, i.e. one with
3126 if (is_failed(&dl
->disk
))
3128 else if (is_spare(&dl
->disk
))
3134 if (i
>= mpb
->num_disks
) {
3135 struct intel_disk
*idisk
;
3137 idisk
= disk_list_get(dl
->serial
, disk_list
);
3138 if (idisk
&& is_spare(&idisk
->disk
) &&
3139 !is_failed(&idisk
->disk
) && !is_configured(&idisk
->disk
))
3147 dl
->next
= champion
->disks
;
3148 champion
->disks
= dl
;
3152 /* delete 'champion' from super_list */
3153 for (del
= super_list
; *del
; ) {
3154 if (*del
== champion
) {
3155 *del
= (*del
)->next
;
3158 del
= &(*del
)->next
;
3160 champion
->next
= NULL
;
3164 struct intel_disk
*idisk
= disk_list
;
3166 disk_list
= disk_list
->next
;
3173 static int load_super_imsm_all(struct supertype
*st
, int fd
, void **sbp
,
3177 struct intel_super
*super_list
= NULL
;
3178 struct intel_super
*super
= NULL
;
3179 int devnum
= fd2devnum(fd
);
3185 /* check if 'fd' an opened container */
3186 sra
= sysfs_read(fd
, 0, GET_LEVEL
|GET_VERSION
|GET_DEVS
|GET_STATE
);
3190 if (sra
->array
.major_version
!= -1 ||
3191 sra
->array
.minor_version
!= -2 ||
3192 strcmp(sra
->text_version
, "imsm") != 0) {
3197 for (sd
= sra
->devs
, i
= 0; sd
; sd
= sd
->next
, i
++) {
3198 struct intel_super
*s
= alloc_super();
3205 s
->next
= super_list
;
3209 sprintf(nm
, "%d:%d", sd
->disk
.major
, sd
->disk
.minor
);
3210 dfd
= dev_open(nm
, O_RDWR
);
3214 err
= load_and_parse_mpb(dfd
, s
, NULL
, 1);
3216 /* retry the load if we might have raced against mdmon */
3217 if (err
== 3 && mdmon_running(devnum
))
3218 for (retry
= 0; retry
< 3; retry
++) {
3220 err
= load_and_parse_mpb(dfd
, s
, NULL
, 1);
3228 /* all mpbs enter, maybe one leaves */
3229 super
= imsm_thunderdome(&super_list
, i
);
3235 if (find_missing(super
) != 0) {
3243 while (super_list
) {
3244 struct intel_super
*s
= super_list
;
3246 super_list
= super_list
->next
;
3255 st
->container_dev
= devnum
;
3256 if (err
== 0 && st
->ss
== NULL
) {
3257 st
->ss
= &super_imsm
;
3258 st
->minor_version
= 0;
3259 st
->max_devs
= IMSM_MAX_DEVICES
;
3264 static int load_container_imsm(struct supertype
*st
, int fd
, char *devname
)
3266 return load_super_imsm_all(st
, fd
, &st
->sb
, devname
);
3270 static int load_super_imsm(struct supertype
*st
, int fd
, char *devname
)
3272 struct intel_super
*super
;
3275 if (test_partition(fd
))
3276 /* IMSM not allowed on partitions */
3279 free_super_imsm(st
);
3281 super
= alloc_super();
3284 Name
": malloc of %zu failed.\n",
3289 rv
= load_and_parse_mpb(fd
, super
, devname
, 0);
3294 Name
": Failed to load all information "
3295 "sections on %s\n", devname
);
3301 if (st
->ss
== NULL
) {
3302 st
->ss
= &super_imsm
;
3303 st
->minor_version
= 0;
3304 st
->max_devs
= IMSM_MAX_DEVICES
;
3309 static __u16
info_to_blocks_per_strip(mdu_array_info_t
*info
)
3311 if (info
->level
== 1)
3313 return info
->chunk_size
>> 9;
3316 static __u32
info_to_num_data_stripes(mdu_array_info_t
*info
, int num_domains
)
3320 num_stripes
= (info
->size
* 2) / info_to_blocks_per_strip(info
);
3321 num_stripes
/= num_domains
;
3326 static __u32
info_to_blocks_per_member(mdu_array_info_t
*info
)
3328 if (info
->level
== 1)
3329 return info
->size
* 2;
3331 return (info
->size
* 2) & ~(info_to_blocks_per_strip(info
) - 1);
3334 static void imsm_update_version_info(struct intel_super
*super
)
3336 /* update the version and attributes */
3337 struct imsm_super
*mpb
= super
->anchor
;
3339 struct imsm_dev
*dev
;
3340 struct imsm_map
*map
;
3343 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
3344 dev
= get_imsm_dev(super
, i
);
3345 map
= get_imsm_map(dev
, 0);
3346 if (__le32_to_cpu(dev
->size_high
) > 0)
3347 mpb
->attributes
|= MPB_ATTRIB_2TB
;
3349 /* FIXME detect when an array spans a port multiplier */
3351 mpb
->attributes
|= MPB_ATTRIB_PM
;
3354 if (mpb
->num_raid_devs
> 1 ||
3355 mpb
->attributes
!= MPB_ATTRIB_CHECKSUM_VERIFY
) {
3356 version
= MPB_VERSION_ATTRIBS
;
3357 switch (get_imsm_raid_level(map
)) {
3358 case 0: mpb
->attributes
|= MPB_ATTRIB_RAID0
; break;
3359 case 1: mpb
->attributes
|= MPB_ATTRIB_RAID1
; break;
3360 case 10: mpb
->attributes
|= MPB_ATTRIB_RAID10
; break;
3361 case 5: mpb
->attributes
|= MPB_ATTRIB_RAID5
; break;
3364 if (map
->num_members
>= 5)
3365 version
= MPB_VERSION_5OR6_DISK_ARRAY
;
3366 else if (dev
->status
== DEV_CLONE_N_GO
)
3367 version
= MPB_VERSION_CNG
;
3368 else if (get_imsm_raid_level(map
) == 5)
3369 version
= MPB_VERSION_RAID5
;
3370 else if (map
->num_members
>= 3)
3371 version
= MPB_VERSION_3OR4_DISK_ARRAY
;
3372 else if (get_imsm_raid_level(map
) == 1)
3373 version
= MPB_VERSION_RAID1
;
3375 version
= MPB_VERSION_RAID0
;
3377 strcpy(((char *) mpb
->sig
) + strlen(MPB_SIGNATURE
), version
);
3381 static int check_name(struct intel_super
*super
, char *name
, int quiet
)
3383 struct imsm_super
*mpb
= super
->anchor
;
3384 char *reason
= NULL
;
3387 if (strlen(name
) > MAX_RAID_SERIAL_LEN
)
3388 reason
= "must be 16 characters or less";
3390 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
3391 struct imsm_dev
*dev
= get_imsm_dev(super
, i
);
3393 if (strncmp((char *) dev
->volume
, name
, MAX_RAID_SERIAL_LEN
) == 0) {
3394 reason
= "already exists";
3399 if (reason
&& !quiet
)
3400 fprintf(stderr
, Name
": imsm volume name %s\n", reason
);
3405 static int init_super_imsm_volume(struct supertype
*st
, mdu_array_info_t
*info
,
3406 unsigned long long size
, char *name
,
3407 char *homehost
, int *uuid
)
3409 /* We are creating a volume inside a pre-existing container.
3410 * so st->sb is already set.
3412 struct intel_super
*super
= st
->sb
;
3413 struct imsm_super
*mpb
= super
->anchor
;
3414 struct intel_dev
*dv
;
3415 struct imsm_dev
*dev
;
3416 struct imsm_vol
*vol
;
3417 struct imsm_map
*map
;
3418 int idx
= mpb
->num_raid_devs
;
3420 unsigned long long array_blocks
;
3421 size_t size_old
, size_new
;
3422 __u32 num_data_stripes
;
3424 if (super
->orom
&& mpb
->num_raid_devs
>= super
->orom
->vpa
) {
3425 fprintf(stderr
, Name
": This imsm-container already has the "
3426 "maximum of %d volumes\n", super
->orom
->vpa
);
3430 /* ensure the mpb is large enough for the new data */
3431 size_old
= __le32_to_cpu(mpb
->mpb_size
);
3432 size_new
= disks_to_mpb_size(info
->nr_disks
);
3433 if (size_new
> size_old
) {
3435 size_t size_round
= ROUND_UP(size_new
, 512);
3437 if (posix_memalign(&mpb_new
, 512, size_round
) != 0) {
3438 fprintf(stderr
, Name
": could not allocate new mpb\n");
3441 memcpy(mpb_new
, mpb
, size_old
);
3444 super
->anchor
= mpb_new
;
3445 mpb
->mpb_size
= __cpu_to_le32(size_new
);
3446 memset(mpb_new
+ size_old
, 0, size_round
- size_old
);
3448 super
->current_vol
= idx
;
3449 /* when creating the first raid device in this container set num_disks
3450 * to zero, i.e. delete this spare and add raid member devices in
3451 * add_to_super_imsm_volume()
3453 if (super
->current_vol
== 0)
3456 if (!check_name(super
, name
, 0))
3458 dv
= malloc(sizeof(*dv
));
3460 fprintf(stderr
, Name
": failed to allocate device list entry\n");
3463 dev
= calloc(1, sizeof(*dev
) + sizeof(__u32
) * (info
->raid_disks
- 1));
3466 fprintf(stderr
, Name
": could not allocate raid device\n");
3470 strncpy((char *) dev
->volume
, name
, MAX_RAID_SERIAL_LEN
);
3471 if (info
->level
== 1)
3472 array_blocks
= info_to_blocks_per_member(info
);
3474 array_blocks
= calc_array_size(info
->level
, info
->raid_disks
,
3475 info
->layout
, info
->chunk_size
,
3477 /* round array size down to closest MB */
3478 array_blocks
= (array_blocks
>> SECT_PER_MB_SHIFT
) << SECT_PER_MB_SHIFT
;
3480 dev
->size_low
= __cpu_to_le32((__u32
) array_blocks
);
3481 dev
->size_high
= __cpu_to_le32((__u32
) (array_blocks
>> 32));
3482 dev
->status
= (DEV_READ_COALESCING
| DEV_WRITE_COALESCING
);
3484 vol
->migr_state
= 0;
3485 set_migr_type(dev
, MIGR_INIT
);
3487 vol
->curr_migr_unit
= 0;
3488 map
= get_imsm_map(dev
, 0);
3489 map
->pba_of_lba0
= __cpu_to_le32(super
->create_offset
);
3490 map
->blocks_per_member
= __cpu_to_le32(info_to_blocks_per_member(info
));
3491 map
->blocks_per_strip
= __cpu_to_le16(info_to_blocks_per_strip(info
));
3492 map
->failed_disk_num
= ~0;
3493 map
->map_state
= info
->level
? IMSM_T_STATE_UNINITIALIZED
:
3494 IMSM_T_STATE_NORMAL
;
3497 if (info
->level
== 1 && info
->raid_disks
> 2) {
3500 fprintf(stderr
, Name
": imsm does not support more than 2 disks"
3501 "in a raid1 volume\n");
3505 map
->raid_level
= info
->level
;
3506 if (info
->level
== 10) {
3507 map
->raid_level
= 1;
3508 map
->num_domains
= info
->raid_disks
/ 2;
3509 } else if (info
->level
== 1)
3510 map
->num_domains
= info
->raid_disks
;
3512 map
->num_domains
= 1;
3514 num_data_stripes
= info_to_num_data_stripes(info
, map
->num_domains
);
3515 map
->num_data_stripes
= __cpu_to_le32(num_data_stripes
);
3517 map
->num_members
= info
->raid_disks
;
3518 for (i
= 0; i
< map
->num_members
; i
++) {
3519 /* initialized in add_to_super */
3520 set_imsm_ord_tbl_ent(map
, i
, IMSM_ORD_REBUILD
);
3522 mpb
->num_raid_devs
++;
3525 dv
->index
= super
->current_vol
;
3526 dv
->next
= super
->devlist
;
3527 super
->devlist
= dv
;
3529 imsm_update_version_info(super
);
3534 static int init_super_imsm(struct supertype
*st
, mdu_array_info_t
*info
,
3535 unsigned long long size
, char *name
,
3536 char *homehost
, int *uuid
)
3538 /* This is primarily called by Create when creating a new array.
3539 * We will then get add_to_super called for each component, and then
3540 * write_init_super called to write it out to each device.
3541 * For IMSM, Create can create on fresh devices or on a pre-existing
3543 * To create on a pre-existing array a different method will be called.
3544 * This one is just for fresh drives.
3546 struct intel_super
*super
;
3547 struct imsm_super
*mpb
;
3552 return init_super_imsm_volume(st
, info
, size
, name
, homehost
, uuid
);
3555 mpb_size
= disks_to_mpb_size(info
->nr_disks
);
3559 super
= alloc_super();
3560 if (super
&& posix_memalign(&super
->buf
, 512, mpb_size
) != 0) {
3565 fprintf(stderr
, Name
3566 ": %s could not allocate superblock\n", __func__
);
3569 memset(super
->buf
, 0, mpb_size
);
3571 mpb
->mpb_size
= __cpu_to_le32(mpb_size
);
3575 /* zeroing superblock */
3579 mpb
->attributes
= MPB_ATTRIB_CHECKSUM_VERIFY
;
3581 version
= (char *) mpb
->sig
;
3582 strcpy(version
, MPB_SIGNATURE
);
3583 version
+= strlen(MPB_SIGNATURE
);
3584 strcpy(version
, MPB_VERSION_RAID0
);
3590 static int add_to_super_imsm_volume(struct supertype
*st
, mdu_disk_info_t
*dk
,
3591 int fd
, char *devname
)
3593 struct intel_super
*super
= st
->sb
;
3594 struct imsm_super
*mpb
= super
->anchor
;
3596 struct imsm_dev
*dev
;
3597 struct imsm_map
*map
;
3600 dev
= get_imsm_dev(super
, super
->current_vol
);
3601 map
= get_imsm_map(dev
, 0);
3603 if (! (dk
->state
& (1<<MD_DISK_SYNC
))) {
3604 fprintf(stderr
, Name
": %s: Cannot add spare devices to IMSM volume\n",
3610 /* we're doing autolayout so grab the pre-marked (in
3611 * validate_geometry) raid_disk
3613 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
3614 if (dl
->raiddisk
== dk
->raid_disk
)
3617 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
3618 if (dl
->major
== dk
->major
&&
3619 dl
->minor
== dk
->minor
)
3624 fprintf(stderr
, Name
": %s is not a member of the same container\n", devname
);
3628 /* add a pristine spare to the metadata */
3629 if (dl
->index
< 0) {
3630 dl
->index
= super
->anchor
->num_disks
;
3631 super
->anchor
->num_disks
++;
3633 /* Check the device has not already been added */
3634 slot
= get_imsm_disk_slot(map
, dl
->index
);
3636 (get_imsm_ord_tbl_ent(dev
, slot
, -1) & IMSM_ORD_REBUILD
) == 0) {
3637 fprintf(stderr
, Name
": %s has been included in this array twice\n",
3641 set_imsm_ord_tbl_ent(map
, dk
->number
, dl
->index
);
3642 dl
->disk
.status
= CONFIGURED_DISK
;
3644 /* if we are creating the first raid device update the family number */
3645 if (super
->current_vol
== 0) {
3647 struct imsm_dev
*_dev
= __get_imsm_dev(mpb
, 0);
3648 struct imsm_disk
*_disk
= __get_imsm_disk(mpb
, dl
->index
);
3650 if (!_dev
|| !_disk
) {
3651 fprintf(stderr
, Name
": BUG mpb setup error\n");
3657 sum
+= __gen_imsm_checksum(mpb
);
3658 mpb
->family_num
= __cpu_to_le32(sum
);
3659 mpb
->orig_family_num
= mpb
->family_num
;
3666 static int add_to_super_imsm(struct supertype
*st
, mdu_disk_info_t
*dk
,
3667 int fd
, char *devname
)
3669 struct intel_super
*super
= st
->sb
;
3671 unsigned long long size
;
3676 /* If we are on an RAID enabled platform check that the disk is
3677 * attached to the raid controller.
3678 * We do not need to test disks attachment for container based additions,
3679 * they shall be already tested when container was created/assembled.
3681 if ((fd
!= -1) && !check_env("IMSM_NO_PLATFORM")) {
3682 struct sys_dev
*hba_name
;
3683 struct intel_hba
*hba
;
3685 hba_name
= find_disk_attached_hba(fd
, NULL
);
3688 Name
": %s is not attached to Intel(R) RAID controller.\n",
3689 devname
? : "disk");
3692 rv
= attach_hba_to_super(super
, hba_name
, devname
);
3695 fprintf(stderr
, Name
": %s is attached to Intel(R) %s RAID "
3696 "controller (%s),\n but the container is assigned to Intel(R) "
3697 "%s RAID controller (",
3699 get_sys_dev_type(hba_name
->type
),
3700 hba_name
->pci_id
? : "Err!",
3701 get_sys_dev_type(hba_name
->type
));
3705 fprintf(stderr
, "%s", hba
->pci_id
? : "Err!");
3707 fprintf(stderr
, ", ");
3711 fprintf(stderr
, ").\n"
3712 " Mixing devices attached to different controllers "
3713 "is not allowed.\n");
3714 free_sys_dev(&hba_name
);
3717 free_sys_dev(&hba_name
);
3720 if (super
->current_vol
>= 0)
3721 return add_to_super_imsm_volume(st
, dk
, fd
, devname
);
3724 dd
= malloc(sizeof(*dd
));
3727 Name
": malloc failed %s:%d.\n", __func__
, __LINE__
);
3730 memset(dd
, 0, sizeof(*dd
));
3731 dd
->major
= major(stb
.st_rdev
);
3732 dd
->minor
= minor(stb
.st_rdev
);
3734 dd
->devname
= devname
? strdup(devname
) : NULL
;
3737 dd
->action
= DISK_ADD
;
3738 rv
= imsm_read_serial(fd
, devname
, dd
->serial
);
3741 Name
": failed to retrieve scsi serial, aborting\n");
3746 get_dev_size(fd
, NULL
, &size
);
3748 serialcpy(dd
->disk
.serial
, dd
->serial
);
3749 dd
->disk
.total_blocks
= __cpu_to_le32(size
);
3750 dd
->disk
.status
= SPARE_DISK
;
3751 if (sysfs_disk_to_scsi_id(fd
, &id
) == 0)
3752 dd
->disk
.scsi_id
= __cpu_to_le32(id
);
3754 dd
->disk
.scsi_id
= __cpu_to_le32(0);
3756 if (st
->update_tail
) {
3757 dd
->next
= super
->disk_mgmt_list
;
3758 super
->disk_mgmt_list
= dd
;
3760 dd
->next
= super
->disks
;
3768 static int remove_from_super_imsm(struct supertype
*st
, mdu_disk_info_t
*dk
)
3770 struct intel_super
*super
= st
->sb
;
3773 /* remove from super works only in mdmon - for communication
3774 * manager - monitor. Check if communication memory buffer
3777 if (!st
->update_tail
) {
3779 Name
": %s shall be used in mdmon context only"
3780 "(line %d).\n", __func__
, __LINE__
);
3783 dd
= malloc(sizeof(*dd
));
3786 Name
": malloc failed %s:%d.\n", __func__
, __LINE__
);
3789 memset(dd
, 0, sizeof(*dd
));
3790 dd
->major
= dk
->major
;
3791 dd
->minor
= dk
->minor
;
3794 dd
->disk
.status
= SPARE_DISK
;
3795 dd
->action
= DISK_REMOVE
;
3797 dd
->next
= super
->disk_mgmt_list
;
3798 super
->disk_mgmt_list
= dd
;
3804 static int store_imsm_mpb(int fd
, struct imsm_super
*mpb
);
3808 struct imsm_super anchor
;
3809 } spare_record
__attribute__ ((aligned(512)));
3811 /* spare records have their own family number and do not have any defined raid
3814 static int write_super_imsm_spares(struct intel_super
*super
, int doclose
)
3816 struct imsm_super
*mpb
= super
->anchor
;
3817 struct imsm_super
*spare
= &spare_record
.anchor
;
3821 spare
->mpb_size
= __cpu_to_le32(sizeof(struct imsm_super
)),
3822 spare
->generation_num
= __cpu_to_le32(1UL),
3823 spare
->attributes
= MPB_ATTRIB_CHECKSUM_VERIFY
;
3824 spare
->num_disks
= 1,
3825 spare
->num_raid_devs
= 0,
3826 spare
->cache_size
= mpb
->cache_size
,
3827 spare
->pwr_cycle_count
= __cpu_to_le32(1),
3829 snprintf((char *) spare
->sig
, MAX_SIGNATURE_LENGTH
,
3830 MPB_SIGNATURE MPB_VERSION_RAID0
);
3832 for (d
= super
->disks
; d
; d
= d
->next
) {
3836 spare
->disk
[0] = d
->disk
;
3837 sum
= __gen_imsm_checksum(spare
);
3838 spare
->family_num
= __cpu_to_le32(sum
);
3839 spare
->orig_family_num
= 0;
3840 sum
= __gen_imsm_checksum(spare
);
3841 spare
->check_sum
= __cpu_to_le32(sum
);
3843 if (store_imsm_mpb(d
->fd
, spare
)) {
3844 fprintf(stderr
, "%s: failed for device %d:%d %s\n",
3845 __func__
, d
->major
, d
->minor
, strerror(errno
));
3857 static int write_super_imsm(struct supertype
*st
, int doclose
)
3859 struct intel_super
*super
= st
->sb
;
3860 struct imsm_super
*mpb
= super
->anchor
;
3866 __u32 mpb_size
= sizeof(struct imsm_super
) - sizeof(struct imsm_disk
);
3869 /* 'generation' is incremented everytime the metadata is written */
3870 generation
= __le32_to_cpu(mpb
->generation_num
);
3872 mpb
->generation_num
= __cpu_to_le32(generation
);
3874 /* fix up cases where previous mdadm releases failed to set
3877 if (mpb
->orig_family_num
== 0)
3878 mpb
->orig_family_num
= mpb
->family_num
;
3880 for (d
= super
->disks
; d
; d
= d
->next
) {
3884 mpb
->disk
[d
->index
] = d
->disk
;
3888 for (d
= super
->missing
; d
; d
= d
->next
) {
3889 mpb
->disk
[d
->index
] = d
->disk
;
3892 mpb
->num_disks
= num_disks
;
3893 mpb_size
+= sizeof(struct imsm_disk
) * mpb
->num_disks
;
3895 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
3896 struct imsm_dev
*dev
= __get_imsm_dev(mpb
, i
);
3897 struct imsm_dev
*dev2
= get_imsm_dev(super
, i
);
3899 imsm_copy_dev(dev
, dev2
);
3900 mpb_size
+= sizeof_imsm_dev(dev
, 0);
3903 mpb_size
+= __le32_to_cpu(mpb
->bbm_log_size
);
3904 mpb
->mpb_size
= __cpu_to_le32(mpb_size
);
3906 /* recalculate checksum */
3907 sum
= __gen_imsm_checksum(mpb
);
3908 mpb
->check_sum
= __cpu_to_le32(sum
);
3910 /* write the mpb for disks that compose raid devices */
3911 for (d
= super
->disks
; d
; d
= d
->next
) {
3914 if (store_imsm_mpb(d
->fd
, mpb
))
3915 fprintf(stderr
, "%s: failed for device %d:%d %s\n",
3916 __func__
, d
->major
, d
->minor
, strerror(errno
));
3924 return write_super_imsm_spares(super
, doclose
);
3930 static int create_array(struct supertype
*st
, int dev_idx
)
3933 struct imsm_update_create_array
*u
;
3934 struct intel_super
*super
= st
->sb
;
3935 struct imsm_dev
*dev
= get_imsm_dev(super
, dev_idx
);
3936 struct imsm_map
*map
= get_imsm_map(dev
, 0);
3937 struct disk_info
*inf
;
3938 struct imsm_disk
*disk
;
3941 len
= sizeof(*u
) - sizeof(*dev
) + sizeof_imsm_dev(dev
, 0) +
3942 sizeof(*inf
) * map
->num_members
;
3945 fprintf(stderr
, "%s: failed to allocate update buffer\n",
3950 u
->type
= update_create_array
;
3951 u
->dev_idx
= dev_idx
;
3952 imsm_copy_dev(&u
->dev
, dev
);
3953 inf
= get_disk_info(u
);
3954 for (i
= 0; i
< map
->num_members
; i
++) {
3955 int idx
= get_imsm_disk_idx(dev
, i
, -1);
3957 disk
= get_imsm_disk(super
, idx
);
3958 serialcpy(inf
[i
].serial
, disk
->serial
);
3960 append_metadata_update(st
, u
, len
);
3965 static int mgmt_disk(struct supertype
*st
)
3967 struct intel_super
*super
= st
->sb
;
3969 struct imsm_update_add_remove_disk
*u
;
3971 if (!super
->disk_mgmt_list
)
3977 fprintf(stderr
, "%s: failed to allocate update buffer\n",
3982 u
->type
= update_add_remove_disk
;
3983 append_metadata_update(st
, u
, len
);
3988 static int write_init_super_imsm(struct supertype
*st
)
3990 struct intel_super
*super
= st
->sb
;
3991 int current_vol
= super
->current_vol
;
3993 /* we are done with current_vol reset it to point st at the container */
3994 super
->current_vol
= -1;
3996 if (st
->update_tail
) {
3997 /* queue the recently created array / added disk
3998 * as a metadata update */
4001 /* determine if we are creating a volume or adding a disk */
4002 if (current_vol
< 0) {
4003 /* in the mgmt (add/remove) disk case we are running
4004 * in mdmon context, so don't close fd's
4006 return mgmt_disk(st
);
4008 rv
= create_array(st
, current_vol
);
4013 for (d
= super
->disks
; d
; d
= d
->next
)
4014 Kill(d
->devname
, NULL
, 0, 1, 1);
4015 return write_super_imsm(st
, 1);
4020 static int store_super_imsm(struct supertype
*st
, int fd
)
4022 struct intel_super
*super
= st
->sb
;
4023 struct imsm_super
*mpb
= super
? super
->anchor
: NULL
;
4029 return store_imsm_mpb(fd
, mpb
);
4035 static int imsm_bbm_log_size(struct imsm_super
*mpb
)
4037 return __le32_to_cpu(mpb
->bbm_log_size
);
4041 static int validate_geometry_imsm_container(struct supertype
*st
, int level
,
4042 int layout
, int raiddisks
, int chunk
,
4043 unsigned long long size
, char *dev
,
4044 unsigned long long *freesize
,
4048 unsigned long long ldsize
;
4049 const struct imsm_orom
*orom
;
4051 if (level
!= LEVEL_CONTAINER
)
4056 if (check_env("IMSM_NO_PLATFORM"))
4059 orom
= find_imsm_orom();
4060 if (orom
&& raiddisks
> orom
->tds
) {
4062 fprintf(stderr
, Name
": %d exceeds maximum number of"
4063 " platform supported disks: %d\n",
4064 raiddisks
, orom
->tds
);
4068 fd
= open(dev
, O_RDONLY
|O_EXCL
, 0);
4071 fprintf(stderr
, Name
": imsm: Cannot open %s: %s\n",
4072 dev
, strerror(errno
));
4075 if (!get_dev_size(fd
, dev
, &ldsize
)) {
4081 *freesize
= avail_size_imsm(st
, ldsize
>> 9);
4086 static unsigned long long find_size(struct extent
*e
, int *idx
, int num_extents
)
4088 const unsigned long long base_start
= e
[*idx
].start
;
4089 unsigned long long end
= base_start
+ e
[*idx
].size
;
4092 if (base_start
== end
)
4096 for (i
= *idx
; i
< num_extents
; i
++) {
4097 /* extend overlapping extents */
4098 if (e
[i
].start
>= base_start
&&
4099 e
[i
].start
<= end
) {
4102 if (e
[i
].start
+ e
[i
].size
> end
)
4103 end
= e
[i
].start
+ e
[i
].size
;
4104 } else if (e
[i
].start
> end
) {
4110 return end
- base_start
;
4113 static unsigned long long merge_extents(struct intel_super
*super
, int sum_extents
)
4115 /* build a composite disk with all known extents and generate a new
4116 * 'maxsize' given the "all disks in an array must share a common start
4117 * offset" constraint
4119 struct extent
*e
= calloc(sum_extents
, sizeof(*e
));
4123 unsigned long long pos
;
4124 unsigned long long start
= 0;
4125 unsigned long long maxsize
;
4126 unsigned long reserve
;
4131 /* coalesce and sort all extents. also, check to see if we need to
4132 * reserve space between member arrays
4135 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
4138 for (i
= 0; i
< dl
->extent_cnt
; i
++)
4141 qsort(e
, sum_extents
, sizeof(*e
), cmp_extent
);
4146 while (i
< sum_extents
) {
4147 e
[j
].start
= e
[i
].start
;
4148 e
[j
].size
= find_size(e
, &i
, sum_extents
);
4150 if (e
[j
-1].size
== 0)
4159 unsigned long long esize
;
4161 esize
= e
[i
].start
- pos
;
4162 if (esize
>= maxsize
) {
4167 pos
= e
[i
].start
+ e
[i
].size
;
4169 } while (e
[i
-1].size
);
4175 /* FIXME assumes volume at offset 0 is the first volume in a
4178 if (start_extent
> 0)
4179 reserve
= IMSM_RESERVED_SECTORS
; /* gap between raid regions */
4183 if (maxsize
< reserve
)
4186 super
->create_offset
= ~((__u32
) 0);
4187 if (start
+ reserve
> super
->create_offset
)
4188 return 0; /* start overflows create_offset */
4189 super
->create_offset
= start
+ reserve
;
4191 return maxsize
- reserve
;
4194 static int is_raid_level_supported(const struct imsm_orom
*orom
, int level
, int raiddisks
)
4196 if (level
< 0 || level
== 6 || level
== 4)
4199 /* if we have an orom prevent invalid raid levels */
4202 case 0: return imsm_orom_has_raid0(orom
);
4205 return imsm_orom_has_raid1e(orom
);
4206 return imsm_orom_has_raid1(orom
) && raiddisks
== 2;
4207 case 10: return imsm_orom_has_raid10(orom
) && raiddisks
== 4;
4208 case 5: return imsm_orom_has_raid5(orom
) && raiddisks
> 2;
4211 return 1; /* not on an Intel RAID platform so anything goes */
4216 #define pr_vrb(fmt, arg...) (void) (verbose && fprintf(stderr, Name fmt, ##arg))
4218 validate_geometry_imsm_orom(struct intel_super
*super
, int level
, int layout
,
4219 int raiddisks
, int *chunk
, int verbose
)
4221 if (!is_raid_level_supported(super
->orom
, level
, raiddisks
)) {
4222 pr_vrb(": platform does not support raid%d with %d disk%s\n",
4223 level
, raiddisks
, raiddisks
> 1 ? "s" : "");
4226 if (super
->orom
&& level
!= 1) {
4227 if (chunk
&& (*chunk
== 0 || *chunk
== UnSet
))
4228 *chunk
= imsm_orom_default_chunk(super
->orom
);
4229 else if (chunk
&& !imsm_orom_has_chunk(super
->orom
, *chunk
)) {
4230 pr_vrb(": platform does not support a chunk size of: "
4235 if (layout
!= imsm_level_to_layout(level
)) {
4237 pr_vrb(": imsm raid 5 only supports the left-asymmetric layout\n");
4238 else if (level
== 10)
4239 pr_vrb(": imsm raid 10 only supports the n2 layout\n");
4241 pr_vrb(": imsm unknown layout %#x for this raid level %d\n",
4249 /* validate_geometry_imsm_volume - lifted from validate_geometry_ddf_bvd
4250 * FIX ME add ahci details
4252 static int validate_geometry_imsm_volume(struct supertype
*st
, int level
,
4253 int layout
, int raiddisks
, int *chunk
,
4254 unsigned long long size
, char *dev
,
4255 unsigned long long *freesize
,
4259 struct intel_super
*super
= st
->sb
;
4260 struct imsm_super
*mpb
= super
->anchor
;
4262 unsigned long long pos
= 0;
4263 unsigned long long maxsize
;
4267 /* We must have the container info already read in. */
4271 if (!validate_geometry_imsm_orom(super
, level
, layout
, raiddisks
, chunk
, verbose
))
4275 /* General test: make sure there is space for
4276 * 'raiddisks' device extents of size 'size' at a given
4279 unsigned long long minsize
= size
;
4280 unsigned long long start_offset
= MaxSector
;
4283 minsize
= MPB_SECTOR_CNT
+ IMSM_RESERVED_SECTORS
;
4284 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
4289 e
= get_extents(super
, dl
);
4292 unsigned long long esize
;
4293 esize
= e
[i
].start
- pos
;
4294 if (esize
>= minsize
)
4296 if (found
&& start_offset
== MaxSector
) {
4299 } else if (found
&& pos
!= start_offset
) {
4303 pos
= e
[i
].start
+ e
[i
].size
;
4305 } while (e
[i
-1].size
);
4310 if (dcnt
< raiddisks
) {
4312 fprintf(stderr
, Name
": imsm: Not enough "
4313 "devices with space for this array "
4321 /* This device must be a member of the set */
4322 if (stat(dev
, &stb
) < 0)
4324 if ((S_IFMT
& stb
.st_mode
) != S_IFBLK
)
4326 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
4327 if (dl
->major
== (int)major(stb
.st_rdev
) &&
4328 dl
->minor
== (int)minor(stb
.st_rdev
))
4333 fprintf(stderr
, Name
": %s is not in the "
4334 "same imsm set\n", dev
);
4336 } else if (super
->orom
&& dl
->index
< 0 && mpb
->num_raid_devs
) {
4337 /* If a volume is present then the current creation attempt
4338 * cannot incorporate new spares because the orom may not
4339 * understand this configuration (all member disks must be
4340 * members of each array in the container).
4342 fprintf(stderr
, Name
": %s is a spare and a volume"
4343 " is already defined for this container\n", dev
);
4344 fprintf(stderr
, Name
": The option-rom requires all member"
4345 " disks to be a member of all volumes\n");
4349 /* retrieve the largest free space block */
4350 e
= get_extents(super
, dl
);
4355 unsigned long long esize
;
4357 esize
= e
[i
].start
- pos
;
4358 if (esize
>= maxsize
)
4360 pos
= e
[i
].start
+ e
[i
].size
;
4362 } while (e
[i
-1].size
);
4367 fprintf(stderr
, Name
": unable to determine free space for: %s\n",
4371 if (maxsize
< size
) {
4373 fprintf(stderr
, Name
": %s not enough space (%llu < %llu)\n",
4374 dev
, maxsize
, size
);
4378 /* count total number of extents for merge */
4380 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
4382 i
+= dl
->extent_cnt
;
4384 maxsize
= merge_extents(super
, i
);
4385 if (maxsize
< size
|| maxsize
== 0) {
4387 fprintf(stderr
, Name
": not enough space after merge (%llu < %llu)\n",
4392 *freesize
= maxsize
;
4397 static int reserve_space(struct supertype
*st
, int raiddisks
,
4398 unsigned long long size
, int chunk
,
4399 unsigned long long *freesize
)
4401 struct intel_super
*super
= st
->sb
;
4402 struct imsm_super
*mpb
= super
->anchor
;
4407 unsigned long long maxsize
;
4408 unsigned long long minsize
;
4412 /* find the largest common start free region of the possible disks */
4416 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
4422 /* don't activate new spares if we are orom constrained
4423 * and there is already a volume active in the container
4425 if (super
->orom
&& dl
->index
< 0 && mpb
->num_raid_devs
)
4428 e
= get_extents(super
, dl
);
4431 for (i
= 1; e
[i
-1].size
; i
++)
4439 maxsize
= merge_extents(super
, extent_cnt
);
4443 minsize
= chunk
* 2;
4445 if (cnt
< raiddisks
||
4446 (super
->orom
&& used
&& used
!= raiddisks
) ||
4447 maxsize
< minsize
||
4449 fprintf(stderr
, Name
": not enough devices with space to create array.\n");
4450 return 0; /* No enough free spaces large enough */
4462 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
4464 dl
->raiddisk
= cnt
++;
4471 static int validate_geometry_imsm(struct supertype
*st
, int level
, int layout
,
4472 int raiddisks
, int *chunk
, unsigned long long size
,
4473 char *dev
, unsigned long long *freesize
,
4480 /* if given unused devices create a container
4481 * if given given devices in a container create a member volume
4483 if (level
== LEVEL_CONTAINER
) {
4484 /* Must be a fresh device to add to a container */
4485 return validate_geometry_imsm_container(st
, level
, layout
,
4487 chunk
?*chunk
:0, size
,
4493 if (st
->sb
&& freesize
) {
4494 /* we are being asked to automatically layout a
4495 * new volume based on the current contents of
4496 * the container. If the the parameters can be
4497 * satisfied reserve_space will record the disks,
4498 * start offset, and size of the volume to be
4499 * created. add_to_super and getinfo_super
4500 * detect when autolayout is in progress.
4502 if (!validate_geometry_imsm_orom(st
->sb
, level
, layout
,
4506 return reserve_space(st
, raiddisks
, size
,
4507 chunk
?*chunk
:0, freesize
);
4512 /* creating in a given container */
4513 return validate_geometry_imsm_volume(st
, level
, layout
,
4514 raiddisks
, chunk
, size
,
4515 dev
, freesize
, verbose
);
4518 /* This device needs to be a device in an 'imsm' container */
4519 fd
= open(dev
, O_RDONLY
|O_EXCL
, 0);
4523 Name
": Cannot create this array on device %s\n",
4528 if (errno
!= EBUSY
|| (fd
= open(dev
, O_RDONLY
, 0)) < 0) {
4530 fprintf(stderr
, Name
": Cannot open %s: %s\n",
4531 dev
, strerror(errno
));
4534 /* Well, it is in use by someone, maybe an 'imsm' container. */
4535 cfd
= open_container(fd
);
4539 fprintf(stderr
, Name
": Cannot use %s: It is busy\n",
4543 sra
= sysfs_read(cfd
, 0, GET_VERSION
);
4544 if (sra
&& sra
->array
.major_version
== -1 &&
4545 strcmp(sra
->text_version
, "imsm") == 0)
4549 /* This is a member of a imsm container. Load the container
4550 * and try to create a volume
4552 struct intel_super
*super
;
4554 if (load_super_imsm_all(st
, cfd
, (void **) &super
, NULL
) == 0) {
4556 st
->container_dev
= fd2devnum(cfd
);
4558 return validate_geometry_imsm_volume(st
, level
, layout
,
4566 fprintf(stderr
, Name
": failed container membership check\n");
4572 static void default_geometry_imsm(struct supertype
*st
, int *level
, int *layout
, int *chunk
)
4574 struct intel_super
*super
= st
->sb
;
4576 if (level
&& *level
== UnSet
)
4577 *level
= LEVEL_CONTAINER
;
4579 if (level
&& layout
&& *layout
== UnSet
)
4580 *layout
= imsm_level_to_layout(*level
);
4582 if (chunk
&& (*chunk
== UnSet
|| *chunk
== 0) &&
4583 super
&& super
->orom
)
4584 *chunk
= imsm_orom_default_chunk(super
->orom
);
4587 static void handle_missing(struct intel_super
*super
, struct imsm_dev
*dev
);
4589 static int kill_subarray_imsm(struct supertype
*st
)
4591 /* remove the subarray currently referenced by ->current_vol */
4593 struct intel_dev
**dp
;
4594 struct intel_super
*super
= st
->sb
;
4595 __u8 current_vol
= super
->current_vol
;
4596 struct imsm_super
*mpb
= super
->anchor
;
4598 if (super
->current_vol
< 0)
4600 super
->current_vol
= -1; /* invalidate subarray cursor */
4602 /* block deletions that would change the uuid of active subarrays
4604 * FIXME when immutable ids are available, but note that we'll
4605 * also need to fixup the invalidated/active subarray indexes in
4608 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
4611 if (i
< current_vol
)
4613 sprintf(subarray
, "%u", i
);
4614 if (is_subarray_active(subarray
, st
->devname
)) {
4616 Name
": deleting subarray-%d would change the UUID of active subarray-%d, aborting\n",
4623 if (st
->update_tail
) {
4624 struct imsm_update_kill_array
*u
= malloc(sizeof(*u
));
4628 u
->type
= update_kill_array
;
4629 u
->dev_idx
= current_vol
;
4630 append_metadata_update(st
, u
, sizeof(*u
));
4635 for (dp
= &super
->devlist
; *dp
;)
4636 if ((*dp
)->index
== current_vol
) {
4639 handle_missing(super
, (*dp
)->dev
);
4640 if ((*dp
)->index
> current_vol
)
4645 /* no more raid devices, all active components are now spares,
4646 * but of course failed are still failed
4648 if (--mpb
->num_raid_devs
== 0) {
4651 for (d
= super
->disks
; d
; d
= d
->next
)
4652 if (d
->index
> -2) {
4654 d
->disk
.status
= SPARE_DISK
;
4658 super
->updates_pending
++;
4663 static int update_subarray_imsm(struct supertype
*st
, char *subarray
,
4664 char *update
, struct mddev_ident
*ident
)
4666 /* update the subarray currently referenced by ->current_vol */
4667 struct intel_super
*super
= st
->sb
;
4668 struct imsm_super
*mpb
= super
->anchor
;
4670 if (strcmp(update
, "name") == 0) {
4671 char *name
= ident
->name
;
4675 if (is_subarray_active(subarray
, st
->devname
)) {
4677 Name
": Unable to update name of active subarray\n");
4681 if (!check_name(super
, name
, 0))
4684 vol
= strtoul(subarray
, &ep
, 10);
4685 if (*ep
!= '\0' || vol
>= super
->anchor
->num_raid_devs
)
4688 if (st
->update_tail
) {
4689 struct imsm_update_rename_array
*u
= malloc(sizeof(*u
));
4693 u
->type
= update_rename_array
;
4695 snprintf((char *) u
->name
, MAX_RAID_SERIAL_LEN
, "%s", name
);
4696 append_metadata_update(st
, u
, sizeof(*u
));
4698 struct imsm_dev
*dev
;
4701 dev
= get_imsm_dev(super
, vol
);
4702 snprintf((char *) dev
->volume
, MAX_RAID_SERIAL_LEN
, "%s", name
);
4703 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
4704 dev
= get_imsm_dev(super
, i
);
4705 handle_missing(super
, dev
);
4707 super
->updates_pending
++;
4715 static int is_gen_migration(struct imsm_dev
*dev
)
4717 if (!dev
->vol
.migr_state
)
4720 if (migr_type(dev
) == MIGR_GEN_MIGR
)
4725 #endif /* MDASSEMBLE */
4727 static int is_rebuilding(struct imsm_dev
*dev
)
4729 struct imsm_map
*migr_map
;
4731 if (!dev
->vol
.migr_state
)
4734 if (migr_type(dev
) != MIGR_REBUILD
)
4737 migr_map
= get_imsm_map(dev
, 1);
4739 if (migr_map
->map_state
== IMSM_T_STATE_DEGRADED
)
4745 static void update_recovery_start(struct imsm_dev
*dev
, struct mdinfo
*array
)
4747 struct mdinfo
*rebuild
= NULL
;
4751 if (!is_rebuilding(dev
))
4754 /* Find the rebuild target, but punt on the dual rebuild case */
4755 for (d
= array
->devs
; d
; d
= d
->next
)
4756 if (d
->recovery_start
== 0) {
4763 /* (?) none of the disks are marked with
4764 * IMSM_ORD_REBUILD, so assume they are missing and the
4765 * disk_ord_tbl was not correctly updated
4767 dprintf("%s: failed to locate out-of-sync disk\n", __func__
);
4771 units
= __le32_to_cpu(dev
->vol
.curr_migr_unit
);
4772 rebuild
->recovery_start
= units
* blocks_per_migr_unit(dev
);
4776 static struct mdinfo
*container_content_imsm(struct supertype
*st
, char *subarray
)
4778 /* Given a container loaded by load_super_imsm_all,
4779 * extract information about all the arrays into
4781 * If 'subarray' is given, just extract info about that array.
4783 * For each imsm_dev create an mdinfo, fill it in,
4784 * then look for matching devices in super->disks
4785 * and create appropriate device mdinfo.
4787 struct intel_super
*super
= st
->sb
;
4788 struct imsm_super
*mpb
= super
->anchor
;
4789 struct mdinfo
*rest
= NULL
;
4793 int spare_disks
= 0;
4795 /* check for bad blocks */
4796 if (imsm_bbm_log_size(super
->anchor
))
4799 /* count spare devices, not used in maps
4801 for (d
= super
->disks
; d
; d
= d
->next
)
4805 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
4806 struct imsm_dev
*dev
;
4807 struct imsm_map
*map
;
4808 struct imsm_map
*map2
;
4809 struct mdinfo
*this;
4814 (i
!= strtoul(subarray
, &ep
, 10) || *ep
!= '\0'))
4817 dev
= get_imsm_dev(super
, i
);
4818 map
= get_imsm_map(dev
, 0);
4819 map2
= get_imsm_map(dev
, 1);
4821 /* do not publish arrays that are in the middle of an
4822 * unsupported migration
4824 if (dev
->vol
.migr_state
&&
4825 (migr_type(dev
) == MIGR_STATE_CHANGE
)) {
4826 fprintf(stderr
, Name
": cannot assemble volume '%.16s':"
4827 " unsupported migration in progress\n",
4832 this = malloc(sizeof(*this));
4834 fprintf(stderr
, Name
": failed to allocate %zu bytes\n",
4838 memset(this, 0, sizeof(*this));
4841 super
->current_vol
= i
;
4842 getinfo_super_imsm_volume(st
, this, NULL
);
4843 for (slot
= 0 ; slot
< map
->num_members
; slot
++) {
4844 unsigned long long recovery_start
;
4845 struct mdinfo
*info_d
;
4852 idx
= get_imsm_disk_idx(dev
, slot
, 0);
4853 ord
= get_imsm_ord_tbl_ent(dev
, slot
, -1);
4854 for (d
= super
->disks
; d
; d
= d
->next
)
4855 if (d
->index
== idx
)
4858 recovery_start
= MaxSector
;
4861 if (d
&& is_failed(&d
->disk
))
4863 if (ord
& IMSM_ORD_REBUILD
)
4867 * if we skip some disks the array will be assmebled degraded;
4868 * reset resync start to avoid a dirty-degraded
4869 * situation when performing the intial sync
4871 * FIXME handle dirty degraded
4873 if ((skip
|| recovery_start
== 0) && !dev
->vol
.dirty
)
4874 this->resync_start
= MaxSector
;
4878 info_d
= calloc(1, sizeof(*info_d
));
4880 fprintf(stderr
, Name
": failed to allocate disk"
4881 " for volume %.16s\n", dev
->volume
);
4882 info_d
= this->devs
;
4884 struct mdinfo
*d
= info_d
->next
;
4893 info_d
->next
= this->devs
;
4894 this->devs
= info_d
;
4896 info_d
->disk
.number
= d
->index
;
4897 info_d
->disk
.major
= d
->major
;
4898 info_d
->disk
.minor
= d
->minor
;
4899 info_d
->disk
.raid_disk
= slot
;
4900 info_d
->recovery_start
= recovery_start
;
4902 if (slot
< map2
->num_members
)
4903 info_d
->disk
.state
= (1 << MD_DISK_ACTIVE
);
4905 this->array
.spare_disks
++;
4907 if (slot
< map
->num_members
)
4908 info_d
->disk
.state
= (1 << MD_DISK_ACTIVE
);
4910 this->array
.spare_disks
++;
4912 if (info_d
->recovery_start
== MaxSector
)
4913 this->array
.working_disks
++;
4915 info_d
->events
= __le32_to_cpu(mpb
->generation_num
);
4916 info_d
->data_offset
= __le32_to_cpu(map
->pba_of_lba0
);
4917 info_d
->component_size
= __le32_to_cpu(map
->blocks_per_member
);
4919 /* now that the disk list is up-to-date fixup recovery_start */
4920 update_recovery_start(dev
, this);
4921 this->array
.spare_disks
+= spare_disks
;
4925 /* if array has bad blocks, set suitable bit in array status */
4927 rest
->array
.state
|= (1<<MD_SB_BBM_ERRORS
);
4933 static __u8
imsm_check_degraded(struct intel_super
*super
, struct imsm_dev
*dev
, int failed
)
4935 struct imsm_map
*map
= get_imsm_map(dev
, 0);
4938 return map
->map_state
== IMSM_T_STATE_UNINITIALIZED
?
4939 IMSM_T_STATE_UNINITIALIZED
: IMSM_T_STATE_NORMAL
;
4941 switch (get_imsm_raid_level(map
)) {
4943 return IMSM_T_STATE_FAILED
;
4946 if (failed
< map
->num_members
)
4947 return IMSM_T_STATE_DEGRADED
;
4949 return IMSM_T_STATE_FAILED
;
4954 * check to see if any mirrors have failed, otherwise we
4955 * are degraded. Even numbered slots are mirrored on
4959 /* gcc -Os complains that this is unused */
4960 int insync
= insync
;
4962 for (i
= 0; i
< map
->num_members
; i
++) {
4963 __u32 ord
= get_imsm_ord_tbl_ent(dev
, i
, -1);
4964 int idx
= ord_to_idx(ord
);
4965 struct imsm_disk
*disk
;
4967 /* reset the potential in-sync count on even-numbered
4968 * slots. num_copies is always 2 for imsm raid10
4973 disk
= get_imsm_disk(super
, idx
);
4974 if (!disk
|| is_failed(disk
) || ord
& IMSM_ORD_REBUILD
)
4977 /* no in-sync disks left in this mirror the
4981 return IMSM_T_STATE_FAILED
;
4984 return IMSM_T_STATE_DEGRADED
;
4988 return IMSM_T_STATE_DEGRADED
;
4990 return IMSM_T_STATE_FAILED
;
4996 return map
->map_state
;
4999 static int imsm_count_failed(struct intel_super
*super
, struct imsm_dev
*dev
)
5003 struct imsm_disk
*disk
;
5004 struct imsm_map
*map
= get_imsm_map(dev
, 0);
5005 struct imsm_map
*prev
= get_imsm_map(dev
, dev
->vol
.migr_state
);
5009 /* at the beginning of migration we set IMSM_ORD_REBUILD on
5010 * disks that are being rebuilt. New failures are recorded to
5011 * map[0]. So we look through all the disks we started with and
5012 * see if any failures are still present, or if any new ones
5015 * FIXME add support for online capacity expansion and
5016 * raid-level-migration
5018 for (i
= 0; i
< prev
->num_members
; i
++) {
5019 ord
= __le32_to_cpu(prev
->disk_ord_tbl
[i
]);
5020 ord
|= __le32_to_cpu(map
->disk_ord_tbl
[i
]);
5021 idx
= ord_to_idx(ord
);
5023 disk
= get_imsm_disk(super
, idx
);
5024 if (!disk
|| is_failed(disk
) || ord
& IMSM_ORD_REBUILD
)
5032 static int imsm_open_new(struct supertype
*c
, struct active_array
*a
,
5035 struct intel_super
*super
= c
->sb
;
5036 struct imsm_super
*mpb
= super
->anchor
;
5038 if (atoi(inst
) >= mpb
->num_raid_devs
) {
5039 fprintf(stderr
, "%s: subarry index %d, out of range\n",
5040 __func__
, atoi(inst
));
5044 dprintf("imsm: open_new %s\n", inst
);
5045 a
->info
.container_member
= atoi(inst
);
5049 static int is_resyncing(struct imsm_dev
*dev
)
5051 struct imsm_map
*migr_map
;
5053 if (!dev
->vol
.migr_state
)
5056 if (migr_type(dev
) == MIGR_INIT
||
5057 migr_type(dev
) == MIGR_REPAIR
)
5060 if (migr_type(dev
) == MIGR_GEN_MIGR
)
5063 migr_map
= get_imsm_map(dev
, 1);
5065 if ((migr_map
->map_state
== IMSM_T_STATE_NORMAL
) &&
5066 (dev
->vol
.migr_type
!= MIGR_GEN_MIGR
))
5072 /* return true if we recorded new information */
5073 static int mark_failure(struct imsm_dev
*dev
, struct imsm_disk
*disk
, int idx
)
5077 struct imsm_map
*map
;
5078 char buf
[MAX_RAID_SERIAL_LEN
+3];
5079 unsigned int len
, shift
= 0;
5081 /* new failures are always set in map[0] */
5082 map
= get_imsm_map(dev
, 0);
5084 slot
= get_imsm_disk_slot(map
, idx
);
5088 ord
= __le32_to_cpu(map
->disk_ord_tbl
[slot
]);
5089 if (is_failed(disk
) && (ord
& IMSM_ORD_REBUILD
))
5092 sprintf(buf
, "%s:0", disk
->serial
);
5093 if ((len
= strlen(buf
)) >= MAX_RAID_SERIAL_LEN
)
5094 shift
= len
- MAX_RAID_SERIAL_LEN
+ 1;
5095 strncpy((char *)disk
->serial
, &buf
[shift
], MAX_RAID_SERIAL_LEN
);
5097 disk
->status
|= FAILED_DISK
;
5098 set_imsm_ord_tbl_ent(map
, slot
, idx
| IMSM_ORD_REBUILD
);
5099 if (map
->failed_disk_num
== 0xff)
5100 map
->failed_disk_num
= slot
;
5104 static void mark_missing(struct imsm_dev
*dev
, struct imsm_disk
*disk
, int idx
)
5106 mark_failure(dev
, disk
, idx
);
5108 if (disk
->scsi_id
== __cpu_to_le32(~(__u32
)0))
5111 disk
->scsi_id
= __cpu_to_le32(~(__u32
)0);
5112 memmove(&disk
->serial
[0], &disk
->serial
[1], MAX_RAID_SERIAL_LEN
- 1);
5115 static void handle_missing(struct intel_super
*super
, struct imsm_dev
*dev
)
5121 if (!super
->missing
)
5123 failed
= imsm_count_failed(super
, dev
);
5124 map_state
= imsm_check_degraded(super
, dev
, failed
);
5126 dprintf("imsm: mark missing\n");
5127 end_migration(dev
, map_state
);
5128 for (dl
= super
->missing
; dl
; dl
= dl
->next
)
5129 mark_missing(dev
, &dl
->disk
, dl
->index
);
5130 super
->updates_pending
++;
5133 static unsigned long long imsm_set_array_size(struct imsm_dev
*dev
)
5135 int used_disks
= imsm_num_data_members(dev
, 0);
5136 unsigned long long array_blocks
;
5137 struct imsm_map
*map
;
5139 if (used_disks
== 0) {
5140 /* when problems occures
5141 * return current array_blocks value
5143 array_blocks
= __le32_to_cpu(dev
->size_high
);
5144 array_blocks
= array_blocks
<< 32;
5145 array_blocks
+= __le32_to_cpu(dev
->size_low
);
5147 return array_blocks
;
5150 /* set array size in metadata
5152 map
= get_imsm_map(dev
, 0);
5153 array_blocks
= map
->blocks_per_member
* used_disks
;
5155 /* round array size down to closest MB
5157 array_blocks
= (array_blocks
>> SECT_PER_MB_SHIFT
) << SECT_PER_MB_SHIFT
;
5158 dev
->size_low
= __cpu_to_le32((__u32
)array_blocks
);
5159 dev
->size_high
= __cpu_to_le32((__u32
)(array_blocks
>> 32));
5161 return array_blocks
;
5164 static void imsm_set_disk(struct active_array
*a
, int n
, int state
);
5166 static void imsm_progress_container_reshape(struct intel_super
*super
)
5168 /* if no device has a migr_state, but some device has a
5169 * different number of members than the previous device, start
5170 * changing the number of devices in this device to match
5173 struct imsm_super
*mpb
= super
->anchor
;
5174 int prev_disks
= -1;
5178 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
5179 struct imsm_dev
*dev
= get_imsm_dev(super
, i
);
5180 struct imsm_map
*map
= get_imsm_map(dev
, 0);
5181 struct imsm_map
*map2
;
5182 int prev_num_members
;
5184 if (dev
->vol
.migr_state
)
5187 if (prev_disks
== -1)
5188 prev_disks
= map
->num_members
;
5189 if (prev_disks
== map
->num_members
)
5192 /* OK, this array needs to enter reshape mode.
5193 * i.e it needs a migr_state
5196 copy_map_size
= sizeof_imsm_map(map
);
5197 prev_num_members
= map
->num_members
;
5198 map
->num_members
= prev_disks
;
5199 dev
->vol
.migr_state
= 1;
5200 dev
->vol
.curr_migr_unit
= 0;
5201 dev
->vol
.migr_type
= MIGR_GEN_MIGR
;
5202 for (i
= prev_num_members
;
5203 i
< map
->num_members
; i
++)
5204 set_imsm_ord_tbl_ent(map
, i
, i
);
5205 map2
= get_imsm_map(dev
, 1);
5206 /* Copy the current map */
5207 memcpy(map2
, map
, copy_map_size
);
5208 map2
->num_members
= prev_num_members
;
5210 imsm_set_array_size(dev
);
5211 super
->updates_pending
++;
5215 /* Handle dirty -> clean transititions, resync and reshape. Degraded and rebuild
5216 * states are handled in imsm_set_disk() with one exception, when a
5217 * resync is stopped due to a new failure this routine will set the
5218 * 'degraded' state for the array.
5220 static int imsm_set_array_state(struct active_array
*a
, int consistent
)
5222 int inst
= a
->info
.container_member
;
5223 struct intel_super
*super
= a
->container
->sb
;
5224 struct imsm_dev
*dev
= get_imsm_dev(super
, inst
);
5225 struct imsm_map
*map
= get_imsm_map(dev
, 0);
5226 int failed
= imsm_count_failed(super
, dev
);
5227 __u8 map_state
= imsm_check_degraded(super
, dev
, failed
);
5228 __u32 blocks_per_unit
;
5230 if (dev
->vol
.migr_state
&&
5231 dev
->vol
.migr_type
== MIGR_GEN_MIGR
) {
5232 /* array state change is blocked due to reshape action
5234 * - abort the reshape (if last_checkpoint is 0 and action!= reshape)
5235 * - finish the reshape (if last_checkpoint is big and action != reshape)
5236 * - update curr_migr_unit
5238 if (a
->curr_action
== reshape
) {
5239 /* still reshaping, maybe update curr_migr_unit */
5240 long long blocks_per_unit
= blocks_per_migr_unit(dev
);
5241 long long unit
= a
->last_checkpoint
;
5242 if (blocks_per_unit
) {
5243 unit
/= blocks_per_unit
;
5245 __le32_to_cpu(dev
->vol
.curr_migr_unit
)) {
5246 dev
->vol
.curr_migr_unit
=
5247 __cpu_to_le32(unit
);
5248 super
->updates_pending
++;
5253 if (a
->last_checkpoint
== 0 && a
->prev_action
== reshape
) {
5254 /* for some reason we aborted the reshape.
5257 struct imsm_map
*map2
= get_imsm_map(dev
, 1);
5258 dev
->vol
.migr_state
= 0;
5259 dev
->vol
.migr_type
= 0;
5260 dev
->vol
.curr_migr_unit
= 0;
5261 memcpy(map
, map2
, sizeof_imsm_map(map2
));
5262 super
->updates_pending
++;
5264 if (a
->last_checkpoint
>= a
->info
.component_size
) {
5265 unsigned long long array_blocks
;
5269 used_disks
= imsm_num_data_members(dev
, 0);
5270 if (used_disks
> 0) {
5272 map
->blocks_per_member
*
5274 /* round array size down to closest MB
5276 array_blocks
= (array_blocks
5277 >> SECT_PER_MB_SHIFT
)
5278 << SECT_PER_MB_SHIFT
;
5279 a
->info
.custom_array_size
= array_blocks
;
5280 /* encourage manager to update array
5284 a
->check_reshape
= 1;
5286 /* finalize online capacity expansion/reshape */
5287 for (mdi
= a
->info
.devs
; mdi
; mdi
= mdi
->next
)
5289 mdi
->disk
.raid_disk
,
5292 imsm_progress_container_reshape(super
);
5297 /* before we activate this array handle any missing disks */
5298 if (consistent
== 2)
5299 handle_missing(super
, dev
);
5301 if (consistent
== 2 &&
5302 (!is_resync_complete(&a
->info
) ||
5303 map_state
!= IMSM_T_STATE_NORMAL
||
5304 dev
->vol
.migr_state
))
5307 if (is_resync_complete(&a
->info
)) {
5308 /* complete intialization / resync,
5309 * recovery and interrupted recovery is completed in
5312 if (is_resyncing(dev
)) {
5313 dprintf("imsm: mark resync done\n");
5314 end_migration(dev
, map_state
);
5315 super
->updates_pending
++;
5316 a
->last_checkpoint
= 0;
5318 } else if (!is_resyncing(dev
) && !failed
) {
5319 /* mark the start of the init process if nothing is failed */
5320 dprintf("imsm: mark resync start\n");
5321 if (map
->map_state
== IMSM_T_STATE_UNINITIALIZED
)
5322 migrate(dev
, IMSM_T_STATE_NORMAL
, MIGR_INIT
);
5324 migrate(dev
, IMSM_T_STATE_NORMAL
, MIGR_REPAIR
);
5325 super
->updates_pending
++;
5328 /* check if we can update curr_migr_unit from resync_start, recovery_start */
5329 blocks_per_unit
= blocks_per_migr_unit(dev
);
5330 if (blocks_per_unit
) {
5334 units
= a
->last_checkpoint
/ blocks_per_unit
;
5337 /* check that we did not overflow 32-bits, and that
5338 * curr_migr_unit needs updating
5340 if (units32
== units
&&
5341 __le32_to_cpu(dev
->vol
.curr_migr_unit
) != units32
) {
5342 dprintf("imsm: mark checkpoint (%u)\n", units32
);
5343 dev
->vol
.curr_migr_unit
= __cpu_to_le32(units32
);
5344 super
->updates_pending
++;
5348 /* mark dirty / clean */
5349 if (dev
->vol
.dirty
!= !consistent
) {
5350 dprintf("imsm: mark '%s'\n", consistent
? "clean" : "dirty");
5355 super
->updates_pending
++;
5361 static void imsm_set_disk(struct active_array
*a
, int n
, int state
)
5363 int inst
= a
->info
.container_member
;
5364 struct intel_super
*super
= a
->container
->sb
;
5365 struct imsm_dev
*dev
= get_imsm_dev(super
, inst
);
5366 struct imsm_map
*map
= get_imsm_map(dev
, 0);
5367 struct imsm_disk
*disk
;
5372 if (n
> map
->num_members
)
5373 fprintf(stderr
, "imsm: set_disk %d out of range 0..%d\n",
5374 n
, map
->num_members
- 1);
5379 dprintf("imsm: set_disk %d:%x\n", n
, state
);
5381 ord
= get_imsm_ord_tbl_ent(dev
, n
, -1);
5382 disk
= get_imsm_disk(super
, ord_to_idx(ord
));
5384 /* check for new failures */
5385 if (state
& DS_FAULTY
) {
5386 if (mark_failure(dev
, disk
, ord_to_idx(ord
)))
5387 super
->updates_pending
++;
5390 /* check if in_sync */
5391 if (state
& DS_INSYNC
&& ord
& IMSM_ORD_REBUILD
&& is_rebuilding(dev
)) {
5392 struct imsm_map
*migr_map
= get_imsm_map(dev
, 1);
5394 set_imsm_ord_tbl_ent(migr_map
, n
, ord_to_idx(ord
));
5395 super
->updates_pending
++;
5398 failed
= imsm_count_failed(super
, dev
);
5399 map_state
= imsm_check_degraded(super
, dev
, failed
);
5401 /* check if recovery complete, newly degraded, or failed */
5402 if (map_state
== IMSM_T_STATE_NORMAL
&& is_rebuilding(dev
)) {
5403 end_migration(dev
, map_state
);
5404 map
= get_imsm_map(dev
, 0);
5405 map
->failed_disk_num
= ~0;
5406 super
->updates_pending
++;
5407 a
->last_checkpoint
= 0;
5408 } else if (map_state
== IMSM_T_STATE_DEGRADED
&&
5409 map
->map_state
!= map_state
&&
5410 !dev
->vol
.migr_state
) {
5411 dprintf("imsm: mark degraded\n");
5412 map
->map_state
= map_state
;
5413 super
->updates_pending
++;
5414 a
->last_checkpoint
= 0;
5415 } else if (map_state
== IMSM_T_STATE_FAILED
&&
5416 map
->map_state
!= map_state
) {
5417 dprintf("imsm: mark failed\n");
5418 end_migration(dev
, map_state
);
5419 super
->updates_pending
++;
5420 a
->last_checkpoint
= 0;
5421 } else if (is_gen_migration(dev
)) {
5422 dprintf("imsm: Detected General Migration in state: ");
5423 if (map_state
== IMSM_T_STATE_NORMAL
) {
5424 end_migration(dev
, map_state
);
5425 map
= get_imsm_map(dev
, 0);
5426 map
->failed_disk_num
= ~0;
5427 dprintf("normal\n");
5429 if (map_state
== IMSM_T_STATE_DEGRADED
) {
5430 printf("degraded\n");
5431 end_migration(dev
, map_state
);
5433 dprintf("failed\n");
5435 map
->map_state
= map_state
;
5437 super
->updates_pending
++;
5441 static int store_imsm_mpb(int fd
, struct imsm_super
*mpb
)
5444 __u32 mpb_size
= __le32_to_cpu(mpb
->mpb_size
);
5445 unsigned long long dsize
;
5446 unsigned long long sectors
;
5448 get_dev_size(fd
, NULL
, &dsize
);
5450 if (mpb_size
> 512) {
5451 /* -1 to account for anchor */
5452 sectors
= mpb_sectors(mpb
) - 1;
5454 /* write the extended mpb to the sectors preceeding the anchor */
5455 if (lseek64(fd
, dsize
- (512 * (2 + sectors
)), SEEK_SET
) < 0)
5458 if ((unsigned long long)write(fd
, buf
+ 512, 512 * sectors
)
5463 /* first block is stored on second to last sector of the disk */
5464 if (lseek64(fd
, dsize
- (512 * 2), SEEK_SET
) < 0)
5467 if (write(fd
, buf
, 512) != 512)
5473 static void imsm_sync_metadata(struct supertype
*container
)
5475 struct intel_super
*super
= container
->sb
;
5477 dprintf("sync metadata: %d\n", super
->updates_pending
);
5478 if (!super
->updates_pending
)
5481 write_super_imsm(container
, 0);
5483 super
->updates_pending
= 0;
5486 static struct dl
*imsm_readd(struct intel_super
*super
, int idx
, struct active_array
*a
)
5488 struct imsm_dev
*dev
= get_imsm_dev(super
, a
->info
.container_member
);
5489 int i
= get_imsm_disk_idx(dev
, idx
, -1);
5492 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
5496 if (dl
&& is_failed(&dl
->disk
))
5500 dprintf("%s: found %x:%x\n", __func__
, dl
->major
, dl
->minor
);
5505 static struct dl
*imsm_add_spare(struct intel_super
*super
, int slot
,
5506 struct active_array
*a
, int activate_new
,
5507 struct mdinfo
*additional_test_list
)
5509 struct imsm_dev
*dev
= get_imsm_dev(super
, a
->info
.container_member
);
5510 int idx
= get_imsm_disk_idx(dev
, slot
, -1);
5511 struct imsm_super
*mpb
= super
->anchor
;
5512 struct imsm_map
*map
;
5513 unsigned long long pos
;
5518 __u32 array_start
= 0;
5519 __u32 array_end
= 0;
5521 struct mdinfo
*test_list
;
5523 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
5524 /* If in this array, skip */
5525 for (d
= a
->info
.devs
; d
; d
= d
->next
)
5526 if (d
->state_fd
>= 0 &&
5527 d
->disk
.major
== dl
->major
&&
5528 d
->disk
.minor
== dl
->minor
) {
5529 dprintf("%x:%x already in array\n",
5530 dl
->major
, dl
->minor
);
5535 test_list
= additional_test_list
;
5537 if (test_list
->disk
.major
== dl
->major
&&
5538 test_list
->disk
.minor
== dl
->minor
) {
5539 dprintf("%x:%x already in additional test list\n",
5540 dl
->major
, dl
->minor
);
5543 test_list
= test_list
->next
;
5548 /* skip in use or failed drives */
5549 if (is_failed(&dl
->disk
) || idx
== dl
->index
||
5551 dprintf("%x:%x status (failed: %d index: %d)\n",
5552 dl
->major
, dl
->minor
, is_failed(&dl
->disk
), idx
);
5556 /* skip pure spares when we are looking for partially
5557 * assimilated drives
5559 if (dl
->index
== -1 && !activate_new
)
5562 /* Does this unused device have the requisite free space?
5563 * It needs to be able to cover all member volumes
5565 ex
= get_extents(super
, dl
);
5567 dprintf("cannot get extents\n");
5570 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
5571 dev
= get_imsm_dev(super
, i
);
5572 map
= get_imsm_map(dev
, 0);
5574 /* check if this disk is already a member of
5577 if (get_imsm_disk_slot(map
, dl
->index
) >= 0)
5583 array_start
= __le32_to_cpu(map
->pba_of_lba0
);
5584 array_end
= array_start
+
5585 __le32_to_cpu(map
->blocks_per_member
) - 1;
5588 /* check that we can start at pba_of_lba0 with
5589 * blocks_per_member of space
5591 if (array_start
>= pos
&& array_end
< ex
[j
].start
) {
5595 pos
= ex
[j
].start
+ ex
[j
].size
;
5597 } while (ex
[j
-1].size
);
5604 if (i
< mpb
->num_raid_devs
) {
5605 dprintf("%x:%x does not have %u to %u available\n",
5606 dl
->major
, dl
->minor
, array_start
, array_end
);
5617 static int imsm_rebuild_allowed(struct supertype
*cont
, int dev_idx
, int failed
)
5619 struct imsm_dev
*dev2
;
5620 struct imsm_map
*map
;
5626 dev2
= get_imsm_dev(cont
->sb
, dev_idx
);
5628 state
= imsm_check_degraded(cont
->sb
, dev2
, failed
);
5629 if (state
== IMSM_T_STATE_FAILED
) {
5630 map
= get_imsm_map(dev2
, 0);
5633 for (slot
= 0; slot
< map
->num_members
; slot
++) {
5635 * Check if failed disks are deleted from intel
5636 * disk list or are marked to be deleted
5638 idx
= get_imsm_disk_idx(dev2
, slot
, -1);
5639 idisk
= get_imsm_dl_disk(cont
->sb
, idx
);
5641 * Do not rebuild the array if failed disks
5642 * from failed sub-array are not removed from
5646 is_failed(&idisk
->disk
) &&
5647 (idisk
->action
!= DISK_REMOVE
))
5655 static struct mdinfo
*imsm_activate_spare(struct active_array
*a
,
5656 struct metadata_update
**updates
)
5659 * Find a device with unused free space and use it to replace a
5660 * failed/vacant region in an array. We replace failed regions one a
5661 * array at a time. The result is that a new spare disk will be added
5662 * to the first failed array and after the monitor has finished
5663 * propagating failures the remainder will be consumed.
5665 * FIXME add a capability for mdmon to request spares from another
5669 struct intel_super
*super
= a
->container
->sb
;
5670 int inst
= a
->info
.container_member
;
5671 struct imsm_dev
*dev
= get_imsm_dev(super
, inst
);
5672 struct imsm_map
*map
= get_imsm_map(dev
, 0);
5673 int failed
= a
->info
.array
.raid_disks
;
5674 struct mdinfo
*rv
= NULL
;
5677 struct metadata_update
*mu
;
5679 struct imsm_update_activate_spare
*u
;
5684 for (d
= a
->info
.devs
; d
; d
= d
->next
) {
5685 if ((d
->curr_state
& DS_FAULTY
) &&
5687 /* wait for Removal to happen */
5689 if (d
->state_fd
>= 0)
5693 dprintf("imsm: activate spare: inst=%d failed=%d (%d) level=%d\n",
5694 inst
, failed
, a
->info
.array
.raid_disks
, a
->info
.array
.level
);
5696 if (dev
->vol
.migr_state
&&
5697 dev
->vol
.migr_type
== MIGR_GEN_MIGR
)
5698 /* No repair during migration */
5701 if (a
->info
.array
.level
== 4)
5702 /* No repair for takeovered array
5703 * imsm doesn't support raid4
5707 if (imsm_check_degraded(super
, dev
, failed
) != IMSM_T_STATE_DEGRADED
)
5711 * If there are any failed disks check state of the other volume.
5712 * Block rebuild if the another one is failed until failed disks
5713 * are removed from container.
5716 dprintf("found failed disks in %s, check if there another"
5717 "failed sub-array.\n",
5719 /* check if states of the other volumes allow for rebuild */
5720 for (i
= 0; i
< super
->anchor
->num_raid_devs
; i
++) {
5722 allowed
= imsm_rebuild_allowed(a
->container
,
5730 /* For each slot, if it is not working, find a spare */
5731 for (i
= 0; i
< a
->info
.array
.raid_disks
; i
++) {
5732 for (d
= a
->info
.devs
; d
; d
= d
->next
)
5733 if (d
->disk
.raid_disk
== i
)
5735 dprintf("found %d: %p %x\n", i
, d
, d
?d
->curr_state
:0);
5736 if (d
&& (d
->state_fd
>= 0))
5740 * OK, this device needs recovery. Try to re-add the
5741 * previous occupant of this slot, if this fails see if
5742 * we can continue the assimilation of a spare that was
5743 * partially assimilated, finally try to activate a new
5746 dl
= imsm_readd(super
, i
, a
);
5748 dl
= imsm_add_spare(super
, i
, a
, 0, NULL
);
5750 dl
= imsm_add_spare(super
, i
, a
, 1, NULL
);
5754 /* found a usable disk with enough space */
5755 di
= malloc(sizeof(*di
));
5758 memset(di
, 0, sizeof(*di
));
5760 /* dl->index will be -1 in the case we are activating a
5761 * pristine spare. imsm_process_update() will create a
5762 * new index in this case. Once a disk is found to be
5763 * failed in all member arrays it is kicked from the
5766 di
->disk
.number
= dl
->index
;
5768 /* (ab)use di->devs to store a pointer to the device
5771 di
->devs
= (struct mdinfo
*) dl
;
5773 di
->disk
.raid_disk
= i
;
5774 di
->disk
.major
= dl
->major
;
5775 di
->disk
.minor
= dl
->minor
;
5777 di
->recovery_start
= 0;
5778 di
->data_offset
= __le32_to_cpu(map
->pba_of_lba0
);
5779 di
->component_size
= a
->info
.component_size
;
5780 di
->container_member
= inst
;
5781 super
->random
= random32();
5785 dprintf("%x:%x to be %d at %llu\n", dl
->major
, dl
->minor
,
5786 i
, di
->data_offset
);
5792 /* No spares found */
5794 /* Now 'rv' has a list of devices to return.
5795 * Create a metadata_update record to update the
5796 * disk_ord_tbl for the array
5798 mu
= malloc(sizeof(*mu
));
5800 mu
->buf
= malloc(sizeof(struct imsm_update_activate_spare
) * num_spares
);
5801 if (mu
->buf
== NULL
) {
5808 struct mdinfo
*n
= rv
->next
;
5817 mu
->space_list
= NULL
;
5818 mu
->len
= sizeof(struct imsm_update_activate_spare
) * num_spares
;
5819 mu
->next
= *updates
;
5820 u
= (struct imsm_update_activate_spare
*) mu
->buf
;
5822 for (di
= rv
; di
; di
= di
->next
) {
5823 u
->type
= update_activate_spare
;
5824 u
->dl
= (struct dl
*) di
->devs
;
5826 u
->slot
= di
->disk
.raid_disk
;
5837 static int disks_overlap(struct intel_super
*super
, int idx
, struct imsm_update_create_array
*u
)
5839 struct imsm_dev
*dev
= get_imsm_dev(super
, idx
);
5840 struct imsm_map
*map
= get_imsm_map(dev
, 0);
5841 struct imsm_map
*new_map
= get_imsm_map(&u
->dev
, 0);
5842 struct disk_info
*inf
= get_disk_info(u
);
5843 struct imsm_disk
*disk
;
5847 for (i
= 0; i
< map
->num_members
; i
++) {
5848 disk
= get_imsm_disk(super
, get_imsm_disk_idx(dev
, i
, -1));
5849 for (j
= 0; j
< new_map
->num_members
; j
++)
5850 if (serialcmp(disk
->serial
, inf
[j
].serial
) == 0)
5858 static struct dl
*get_disk_super(struct intel_super
*super
, int major
, int minor
)
5860 struct dl
*dl
= NULL
;
5861 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
5862 if ((dl
->major
== major
) && (dl
->minor
== minor
))
5867 static int remove_disk_super(struct intel_super
*super
, int major
, int minor
)
5869 struct dl
*prev
= NULL
;
5873 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
5874 if ((dl
->major
== major
) && (dl
->minor
== minor
)) {
5877 prev
->next
= dl
->next
;
5879 super
->disks
= dl
->next
;
5881 __free_imsm_disk(dl
);
5882 dprintf("%s: removed %x:%x\n",
5883 __func__
, major
, minor
);
5891 static void imsm_delete(struct intel_super
*super
, struct dl
**dlp
, unsigned index
);
5893 static int add_remove_disk_update(struct intel_super
*super
)
5895 int check_degraded
= 0;
5896 struct dl
*disk
= NULL
;
5897 /* add/remove some spares to/from the metadata/contrainer */
5898 while (super
->disk_mgmt_list
) {
5899 struct dl
*disk_cfg
;
5901 disk_cfg
= super
->disk_mgmt_list
;
5902 super
->disk_mgmt_list
= disk_cfg
->next
;
5903 disk_cfg
->next
= NULL
;
5905 if (disk_cfg
->action
== DISK_ADD
) {
5906 disk_cfg
->next
= super
->disks
;
5907 super
->disks
= disk_cfg
;
5909 dprintf("%s: added %x:%x\n",
5910 __func__
, disk_cfg
->major
,
5912 } else if (disk_cfg
->action
== DISK_REMOVE
) {
5913 dprintf("Disk remove action processed: %x.%x\n",
5914 disk_cfg
->major
, disk_cfg
->minor
);
5915 disk
= get_disk_super(super
,
5919 /* store action status */
5920 disk
->action
= DISK_REMOVE
;
5921 /* remove spare disks only */
5922 if (disk
->index
== -1) {
5923 remove_disk_super(super
,
5928 /* release allocate disk structure */
5929 __free_imsm_disk(disk_cfg
);
5932 return check_degraded
;
5935 static int apply_reshape_container_disks_update(struct imsm_update_reshape
*u
,
5936 struct intel_super
*super
,
5939 struct dl
*new_disk
;
5940 struct intel_dev
*id
;
5942 int delta_disks
= u
->new_raid_disks
- u
->old_raid_disks
;
5943 int disk_count
= u
->old_raid_disks
;
5944 void **tofree
= NULL
;
5945 int devices_to_reshape
= 1;
5946 struct imsm_super
*mpb
= super
->anchor
;
5948 unsigned int dev_id
;
5950 dprintf("imsm: apply_reshape_container_disks_update()\n");
5952 /* enable spares to use in array */
5953 for (i
= 0; i
< delta_disks
; i
++) {
5954 new_disk
= get_disk_super(super
,
5955 major(u
->new_disks
[i
]),
5956 minor(u
->new_disks
[i
]));
5957 dprintf("imsm: new disk for reshape is: %i:%i "
5958 "(%p, index = %i)\n",
5959 major(u
->new_disks
[i
]), minor(u
->new_disks
[i
]),
5960 new_disk
, new_disk
->index
);
5961 if ((new_disk
== NULL
) ||
5962 ((new_disk
->index
>= 0) &&
5963 (new_disk
->index
< u
->old_raid_disks
)))
5964 goto update_reshape_exit
;
5965 new_disk
->index
= disk_count
++;
5966 /* slot to fill in autolayout
5968 new_disk
->raiddisk
= new_disk
->index
;
5969 new_disk
->disk
.status
|=
5971 new_disk
->disk
.status
&= ~SPARE_DISK
;
5974 dprintf("imsm: volume set mpb->num_raid_devs = %i\n",
5975 mpb
->num_raid_devs
);
5976 /* manage changes in volume
5978 for (dev_id
= 0; dev_id
< mpb
->num_raid_devs
; dev_id
++) {
5979 void **sp
= *space_list
;
5980 struct imsm_dev
*newdev
;
5981 struct imsm_map
*newmap
, *oldmap
;
5983 for (id
= super
->devlist
; id
; id
= id
->next
) {
5984 if (id
->index
== dev_id
)
5993 /* Copy the dev, but not (all of) the map */
5994 memcpy(newdev
, id
->dev
, sizeof(*newdev
));
5995 oldmap
= get_imsm_map(id
->dev
, 0);
5996 newmap
= get_imsm_map(newdev
, 0);
5997 /* Copy the current map */
5998 memcpy(newmap
, oldmap
, sizeof_imsm_map(oldmap
));
5999 /* update one device only
6001 if (devices_to_reshape
) {
6002 dprintf("imsm: modifying subdev: %i\n",
6004 devices_to_reshape
--;
6005 newdev
->vol
.migr_state
= 1;
6006 newdev
->vol
.curr_migr_unit
= 0;
6007 newdev
->vol
.migr_type
= MIGR_GEN_MIGR
;
6008 newmap
->num_members
= u
->new_raid_disks
;
6009 for (i
= 0; i
< delta_disks
; i
++) {
6010 set_imsm_ord_tbl_ent(newmap
,
6011 u
->old_raid_disks
+ i
,
6012 u
->old_raid_disks
+ i
);
6014 /* New map is correct, now need to save old map
6016 newmap
= get_imsm_map(newdev
, 1);
6017 memcpy(newmap
, oldmap
, sizeof_imsm_map(oldmap
));
6019 imsm_set_array_size(newdev
);
6022 sp
= (void **)id
->dev
;
6028 *space_list
= tofree
;
6031 update_reshape_exit
:
6036 static int apply_takeover_update(struct imsm_update_takeover
*u
,
6037 struct intel_super
*super
,
6040 struct imsm_dev
*dev
= NULL
;
6041 struct intel_dev
*dv
;
6042 struct imsm_dev
*dev_new
;
6043 struct imsm_map
*map
;
6047 for (dv
= super
->devlist
; dv
; dv
= dv
->next
)
6048 if (dv
->index
== (unsigned int)u
->subarray
) {
6056 map
= get_imsm_map(dev
, 0);
6058 if (u
->direction
== R10_TO_R0
) {
6059 /* Number of failed disks must be half of initial disk number */
6060 if (imsm_count_failed(super
, dev
) != (map
->num_members
/ 2))
6063 /* iterate through devices to mark removed disks as spare */
6064 for (dm
= super
->disks
; dm
; dm
= dm
->next
) {
6065 if (dm
->disk
.status
& FAILED_DISK
) {
6066 int idx
= dm
->index
;
6067 /* update indexes on the disk list */
6068 /* FIXME this loop-with-the-loop looks wrong, I'm not convinced
6069 the index values will end up being correct.... NB */
6070 for (du
= super
->disks
; du
; du
= du
->next
)
6071 if (du
->index
> idx
)
6073 /* mark as spare disk */
6074 dm
->disk
.status
= SPARE_DISK
;
6079 map
->num_members
= map
->num_members
/ 2;
6080 map
->map_state
= IMSM_T_STATE_NORMAL
;
6081 map
->num_domains
= 1;
6082 map
->raid_level
= 0;
6083 map
->failed_disk_num
= -1;
6086 if (u
->direction
== R0_TO_R10
) {
6088 /* update slots in current disk list */
6089 for (dm
= super
->disks
; dm
; dm
= dm
->next
) {
6093 /* create new *missing* disks */
6094 for (i
= 0; i
< map
->num_members
; i
++) {
6095 space
= *space_list
;
6098 *space_list
= *space
;
6100 memcpy(du
, super
->disks
, sizeof(*du
));
6104 du
->index
= (i
* 2) + 1;
6105 sprintf((char *)du
->disk
.serial
,
6106 " MISSING_%d", du
->index
);
6107 sprintf((char *)du
->serial
,
6108 "MISSING_%d", du
->index
);
6109 du
->next
= super
->missing
;
6110 super
->missing
= du
;
6112 /* create new dev and map */
6113 space
= *space_list
;
6116 *space_list
= *space
;
6117 dev_new
= (void *)space
;
6118 memcpy(dev_new
, dev
, sizeof(*dev
));
6119 /* update new map */
6120 map
= get_imsm_map(dev_new
, 0);
6121 map
->num_members
= map
->num_members
* 2;
6122 map
->map_state
= IMSM_T_STATE_DEGRADED
;
6123 map
->num_domains
= 2;
6124 map
->raid_level
= 1;
6125 /* replace dev<->dev_new */
6128 /* update disk order table */
6129 for (du
= super
->disks
; du
; du
= du
->next
)
6131 set_imsm_ord_tbl_ent(map
, du
->index
, du
->index
);
6132 for (du
= super
->missing
; du
; du
= du
->next
)
6133 if (du
->index
>= 0) {
6134 set_imsm_ord_tbl_ent(map
, du
->index
, du
->index
);
6135 mark_missing(dev_new
, &du
->disk
, du
->index
);
6141 static void imsm_process_update(struct supertype
*st
,
6142 struct metadata_update
*update
)
6145 * crack open the metadata_update envelope to find the update record
6146 * update can be one of:
6147 * update_reshape_container_disks - all the arrays in the container
6148 * are being reshaped to have more devices. We need to mark
6149 * the arrays for general migration and convert selected spares
6150 * into active devices.
6151 * update_activate_spare - a spare device has replaced a failed
6152 * device in an array, update the disk_ord_tbl. If this disk is
6153 * present in all member arrays then also clear the SPARE_DISK
6155 * update_create_array
6157 * update_rename_array
6158 * update_add_remove_disk
6160 struct intel_super
*super
= st
->sb
;
6161 struct imsm_super
*mpb
;
6162 enum imsm_update_type type
= *(enum imsm_update_type
*) update
->buf
;
6164 /* update requires a larger buf but the allocation failed */
6165 if (super
->next_len
&& !super
->next_buf
) {
6166 super
->next_len
= 0;
6170 if (super
->next_buf
) {
6171 memcpy(super
->next_buf
, super
->buf
, super
->len
);
6173 super
->len
= super
->next_len
;
6174 super
->buf
= super
->next_buf
;
6176 super
->next_len
= 0;
6177 super
->next_buf
= NULL
;
6180 mpb
= super
->anchor
;
6183 case update_takeover
: {
6184 struct imsm_update_takeover
*u
= (void *)update
->buf
;
6185 if (apply_takeover_update(u
, super
, &update
->space_list
)) {
6186 imsm_update_version_info(super
);
6187 super
->updates_pending
++;
6192 case update_reshape_container_disks
: {
6193 struct imsm_update_reshape
*u
= (void *)update
->buf
;
6194 if (apply_reshape_container_disks_update(
6195 u
, super
, &update
->space_list
))
6196 super
->updates_pending
++;
6199 case update_activate_spare
: {
6200 struct imsm_update_activate_spare
*u
= (void *) update
->buf
;
6201 struct imsm_dev
*dev
= get_imsm_dev(super
, u
->array
);
6202 struct imsm_map
*map
= get_imsm_map(dev
, 0);
6203 struct imsm_map
*migr_map
;
6204 struct active_array
*a
;
6205 struct imsm_disk
*disk
;
6210 int victim
= get_imsm_disk_idx(dev
, u
->slot
, -1);
6213 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
6218 fprintf(stderr
, "error: imsm_activate_spare passed "
6219 "an unknown disk (index: %d)\n",
6224 super
->updates_pending
++;
6226 /* count failures (excluding rebuilds and the victim)
6227 * to determine map[0] state
6230 for (i
= 0; i
< map
->num_members
; i
++) {
6233 disk
= get_imsm_disk(super
,
6234 get_imsm_disk_idx(dev
, i
, -1));
6235 if (!disk
|| is_failed(disk
))
6239 /* adding a pristine spare, assign a new index */
6240 if (dl
->index
< 0) {
6241 dl
->index
= super
->anchor
->num_disks
;
6242 super
->anchor
->num_disks
++;
6245 disk
->status
|= CONFIGURED_DISK
;
6246 disk
->status
&= ~SPARE_DISK
;
6249 to_state
= imsm_check_degraded(super
, dev
, failed
);
6250 map
->map_state
= IMSM_T_STATE_DEGRADED
;
6251 migrate(dev
, to_state
, MIGR_REBUILD
);
6252 migr_map
= get_imsm_map(dev
, 1);
6253 set_imsm_ord_tbl_ent(map
, u
->slot
, dl
->index
);
6254 set_imsm_ord_tbl_ent(migr_map
, u
->slot
, dl
->index
| IMSM_ORD_REBUILD
);
6256 /* update the family_num to mark a new container
6257 * generation, being careful to record the existing
6258 * family_num in orig_family_num to clean up after
6259 * earlier mdadm versions that neglected to set it.
6261 if (mpb
->orig_family_num
== 0)
6262 mpb
->orig_family_num
= mpb
->family_num
;
6263 mpb
->family_num
+= super
->random
;
6265 /* count arrays using the victim in the metadata */
6267 for (a
= st
->arrays
; a
; a
= a
->next
) {
6268 dev
= get_imsm_dev(super
, a
->info
.container_member
);
6269 map
= get_imsm_map(dev
, 0);
6271 if (get_imsm_disk_slot(map
, victim
) >= 0)
6275 /* delete the victim if it is no longer being
6281 /* We know that 'manager' isn't touching anything,
6282 * so it is safe to delete
6284 for (dlp
= &super
->disks
; *dlp
; dlp
= &(*dlp
)->next
)
6285 if ((*dlp
)->index
== victim
)
6288 /* victim may be on the missing list */
6290 for (dlp
= &super
->missing
; *dlp
; dlp
= &(*dlp
)->next
)
6291 if ((*dlp
)->index
== victim
)
6293 imsm_delete(super
, dlp
, victim
);
6297 case update_create_array
: {
6298 /* someone wants to create a new array, we need to be aware of
6299 * a few races/collisions:
6300 * 1/ 'Create' called by two separate instances of mdadm
6301 * 2/ 'Create' versus 'activate_spare': mdadm has chosen
6302 * devices that have since been assimilated via
6304 * In the event this update can not be carried out mdadm will
6305 * (FIX ME) notice that its update did not take hold.
6307 struct imsm_update_create_array
*u
= (void *) update
->buf
;
6308 struct intel_dev
*dv
;
6309 struct imsm_dev
*dev
;
6310 struct imsm_map
*map
, *new_map
;
6311 unsigned long long start
, end
;
6312 unsigned long long new_start
, new_end
;
6314 struct disk_info
*inf
;
6317 /* handle racing creates: first come first serve */
6318 if (u
->dev_idx
< mpb
->num_raid_devs
) {
6319 dprintf("%s: subarray %d already defined\n",
6320 __func__
, u
->dev_idx
);
6324 /* check update is next in sequence */
6325 if (u
->dev_idx
!= mpb
->num_raid_devs
) {
6326 dprintf("%s: can not create array %d expected index %d\n",
6327 __func__
, u
->dev_idx
, mpb
->num_raid_devs
);
6331 new_map
= get_imsm_map(&u
->dev
, 0);
6332 new_start
= __le32_to_cpu(new_map
->pba_of_lba0
);
6333 new_end
= new_start
+ __le32_to_cpu(new_map
->blocks_per_member
);
6334 inf
= get_disk_info(u
);
6336 /* handle activate_spare versus create race:
6337 * check to make sure that overlapping arrays do not include
6340 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
6341 dev
= get_imsm_dev(super
, i
);
6342 map
= get_imsm_map(dev
, 0);
6343 start
= __le32_to_cpu(map
->pba_of_lba0
);
6344 end
= start
+ __le32_to_cpu(map
->blocks_per_member
);
6345 if ((new_start
>= start
&& new_start
<= end
) ||
6346 (start
>= new_start
&& start
<= new_end
))
6351 if (disks_overlap(super
, i
, u
)) {
6352 dprintf("%s: arrays overlap\n", __func__
);
6357 /* check that prepare update was successful */
6358 if (!update
->space
) {
6359 dprintf("%s: prepare update failed\n", __func__
);
6363 /* check that all disks are still active before committing
6364 * changes. FIXME: could we instead handle this by creating a
6365 * degraded array? That's probably not what the user expects,
6366 * so better to drop this update on the floor.
6368 for (i
= 0; i
< new_map
->num_members
; i
++) {
6369 dl
= serial_to_dl(inf
[i
].serial
, super
);
6371 dprintf("%s: disk disappeared\n", __func__
);
6376 super
->updates_pending
++;
6378 /* convert spares to members and fixup ord_tbl */
6379 for (i
= 0; i
< new_map
->num_members
; i
++) {
6380 dl
= serial_to_dl(inf
[i
].serial
, super
);
6381 if (dl
->index
== -1) {
6382 dl
->index
= mpb
->num_disks
;
6384 dl
->disk
.status
|= CONFIGURED_DISK
;
6385 dl
->disk
.status
&= ~SPARE_DISK
;
6387 set_imsm_ord_tbl_ent(new_map
, i
, dl
->index
);
6392 update
->space
= NULL
;
6393 imsm_copy_dev(dev
, &u
->dev
);
6394 dv
->index
= u
->dev_idx
;
6395 dv
->next
= super
->devlist
;
6396 super
->devlist
= dv
;
6397 mpb
->num_raid_devs
++;
6399 imsm_update_version_info(super
);
6402 /* mdmon knows how to release update->space, but not
6403 * ((struct intel_dev *) update->space)->dev
6405 if (update
->space
) {
6411 case update_kill_array
: {
6412 struct imsm_update_kill_array
*u
= (void *) update
->buf
;
6413 int victim
= u
->dev_idx
;
6414 struct active_array
*a
;
6415 struct intel_dev
**dp
;
6416 struct imsm_dev
*dev
;
6418 /* sanity check that we are not affecting the uuid of
6419 * active arrays, or deleting an active array
6421 * FIXME when immutable ids are available, but note that
6422 * we'll also need to fixup the invalidated/active
6423 * subarray indexes in mdstat
6425 for (a
= st
->arrays
; a
; a
= a
->next
)
6426 if (a
->info
.container_member
>= victim
)
6428 /* by definition if mdmon is running at least one array
6429 * is active in the container, so checking
6430 * mpb->num_raid_devs is just extra paranoia
6432 dev
= get_imsm_dev(super
, victim
);
6433 if (a
|| !dev
|| mpb
->num_raid_devs
== 1) {
6434 dprintf("failed to delete subarray-%d\n", victim
);
6438 for (dp
= &super
->devlist
; *dp
;)
6439 if ((*dp
)->index
== (unsigned)super
->current_vol
) {
6442 if ((*dp
)->index
> (unsigned)victim
)
6446 mpb
->num_raid_devs
--;
6447 super
->updates_pending
++;
6450 case update_rename_array
: {
6451 struct imsm_update_rename_array
*u
= (void *) update
->buf
;
6452 char name
[MAX_RAID_SERIAL_LEN
+1];
6453 int target
= u
->dev_idx
;
6454 struct active_array
*a
;
6455 struct imsm_dev
*dev
;
6457 /* sanity check that we are not affecting the uuid of
6460 snprintf(name
, MAX_RAID_SERIAL_LEN
, "%s", (char *) u
->name
);
6461 name
[MAX_RAID_SERIAL_LEN
] = '\0';
6462 for (a
= st
->arrays
; a
; a
= a
->next
)
6463 if (a
->info
.container_member
== target
)
6465 dev
= get_imsm_dev(super
, u
->dev_idx
);
6466 if (a
|| !dev
|| !check_name(super
, name
, 1)) {
6467 dprintf("failed to rename subarray-%d\n", target
);
6471 snprintf((char *) dev
->volume
, MAX_RAID_SERIAL_LEN
, "%s", name
);
6472 super
->updates_pending
++;
6475 case update_add_remove_disk
: {
6476 /* we may be able to repair some arrays if disks are
6477 * being added, check teh status of add_remove_disk
6478 * if discs has been added.
6480 if (add_remove_disk_update(super
)) {
6481 struct active_array
*a
;
6483 super
->updates_pending
++;
6484 for (a
= st
->arrays
; a
; a
= a
->next
)
6485 a
->check_degraded
= 1;
6490 fprintf(stderr
, "error: unsuported process update type:"
6491 "(type: %d)\n", type
);
6495 static void imsm_prepare_update(struct supertype
*st
,
6496 struct metadata_update
*update
)
6499 * Allocate space to hold new disk entries, raid-device entries or a new
6500 * mpb if necessary. The manager synchronously waits for updates to
6501 * complete in the monitor, so new mpb buffers allocated here can be
6502 * integrated by the monitor thread without worrying about live pointers
6503 * in the manager thread.
6505 enum imsm_update_type type
= *(enum imsm_update_type
*) update
->buf
;
6506 struct intel_super
*super
= st
->sb
;
6507 struct imsm_super
*mpb
= super
->anchor
;
6512 case update_takeover
: {
6513 struct imsm_update_takeover
*u
= (void *)update
->buf
;
6514 if (u
->direction
== R0_TO_R10
) {
6515 void **tail
= (void **)&update
->space_list
;
6516 struct imsm_dev
*dev
= get_imsm_dev(super
, u
->subarray
);
6517 struct imsm_map
*map
= get_imsm_map(dev
, 0);
6518 int num_members
= map
->num_members
;
6522 /* allocate memory for added disks */
6523 for (i
= 0; i
< num_members
; i
++) {
6524 size
= sizeof(struct dl
);
6525 space
= malloc(size
);
6534 /* allocate memory for new device */
6535 size
= sizeof_imsm_dev(super
->devlist
->dev
, 0) +
6536 (num_members
* sizeof(__u32
));
6537 space
= malloc(size
);
6546 len
= disks_to_mpb_size(num_members
* 2);
6548 /* if allocation didn't success, free buffer */
6549 while (update
->space_list
) {
6550 void **sp
= update
->space_list
;
6551 update
->space_list
= *sp
;
6559 case update_reshape_container_disks
: {
6560 /* Every raid device in the container is about to
6561 * gain some more devices, and we will enter a
6563 * So each 'imsm_map' will be bigger, and the imsm_vol
6564 * will now hold 2 of them.
6565 * Thus we need new 'struct imsm_dev' allocations sized
6566 * as sizeof_imsm_dev but with more devices in both maps.
6568 struct imsm_update_reshape
*u
= (void *)update
->buf
;
6569 struct intel_dev
*dl
;
6570 void **space_tail
= (void**)&update
->space_list
;
6572 dprintf("imsm: imsm_prepare_update() for update_reshape\n");
6574 for (dl
= super
->devlist
; dl
; dl
= dl
->next
) {
6575 int size
= sizeof_imsm_dev(dl
->dev
, 1);
6577 if (u
->new_raid_disks
> u
->old_raid_disks
)
6578 size
+= sizeof(__u32
)*2*
6579 (u
->new_raid_disks
- u
->old_raid_disks
);
6588 len
= disks_to_mpb_size(u
->new_raid_disks
);
6589 dprintf("New anchor length is %llu\n", (unsigned long long)len
);
6592 case update_create_array
: {
6593 struct imsm_update_create_array
*u
= (void *) update
->buf
;
6594 struct intel_dev
*dv
;
6595 struct imsm_dev
*dev
= &u
->dev
;
6596 struct imsm_map
*map
= get_imsm_map(dev
, 0);
6598 struct disk_info
*inf
;
6602 inf
= get_disk_info(u
);
6603 len
= sizeof_imsm_dev(dev
, 1);
6604 /* allocate a new super->devlist entry */
6605 dv
= malloc(sizeof(*dv
));
6607 dv
->dev
= malloc(len
);
6612 update
->space
= NULL
;
6616 /* count how many spares will be converted to members */
6617 for (i
= 0; i
< map
->num_members
; i
++) {
6618 dl
= serial_to_dl(inf
[i
].serial
, super
);
6620 /* hmm maybe it failed?, nothing we can do about
6625 if (count_memberships(dl
, super
) == 0)
6628 len
+= activate
* sizeof(struct imsm_disk
);
6635 /* check if we need a larger metadata buffer */
6636 if (super
->next_buf
)
6637 buf_len
= super
->next_len
;
6639 buf_len
= super
->len
;
6641 if (__le32_to_cpu(mpb
->mpb_size
) + len
> buf_len
) {
6642 /* ok we need a larger buf than what is currently allocated
6643 * if this allocation fails process_update will notice that
6644 * ->next_len is set and ->next_buf is NULL
6646 buf_len
= ROUND_UP(__le32_to_cpu(mpb
->mpb_size
) + len
, 512);
6647 if (super
->next_buf
)
6648 free(super
->next_buf
);
6650 super
->next_len
= buf_len
;
6651 if (posix_memalign(&super
->next_buf
, 512, buf_len
) == 0)
6652 memset(super
->next_buf
, 0, buf_len
);
6654 super
->next_buf
= NULL
;
6658 /* must be called while manager is quiesced */
6659 static void imsm_delete(struct intel_super
*super
, struct dl
**dlp
, unsigned index
)
6661 struct imsm_super
*mpb
= super
->anchor
;
6663 struct imsm_dev
*dev
;
6664 struct imsm_map
*map
;
6665 int i
, j
, num_members
;
6668 dprintf("%s: deleting device[%d] from imsm_super\n",
6671 /* shift all indexes down one */
6672 for (iter
= super
->disks
; iter
; iter
= iter
->next
)
6673 if (iter
->index
> (int)index
)
6675 for (iter
= super
->missing
; iter
; iter
= iter
->next
)
6676 if (iter
->index
> (int)index
)
6679 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
6680 dev
= get_imsm_dev(super
, i
);
6681 map
= get_imsm_map(dev
, 0);
6682 num_members
= map
->num_members
;
6683 for (j
= 0; j
< num_members
; j
++) {
6684 /* update ord entries being careful not to propagate
6685 * ord-flags to the first map
6687 ord
= get_imsm_ord_tbl_ent(dev
, j
, -1);
6689 if (ord_to_idx(ord
) <= index
)
6692 map
= get_imsm_map(dev
, 0);
6693 set_imsm_ord_tbl_ent(map
, j
, ord_to_idx(ord
- 1));
6694 map
= get_imsm_map(dev
, 1);
6696 set_imsm_ord_tbl_ent(map
, j
, ord
- 1);
6701 super
->updates_pending
++;
6703 struct dl
*dl
= *dlp
;
6705 *dlp
= (*dlp
)->next
;
6706 __free_imsm_disk(dl
);
6710 static char disk_by_path
[] = "/dev/disk/by-path/";
6712 static const char *imsm_get_disk_controller_domain(const char *path
)
6714 char disk_path
[PATH_MAX
];
6718 strncpy(disk_path
, disk_by_path
, PATH_MAX
- 1);
6719 strncat(disk_path
, path
, PATH_MAX
- strlen(disk_path
) - 1);
6720 if (stat(disk_path
, &st
) == 0) {
6721 struct sys_dev
* hba
;
6724 path
= devt_to_devpath(st
.st_rdev
);
6727 hba
= find_disk_attached_hba(-1, path
);
6728 if (hba
&& hba
->type
== SYS_DEV_SAS
)
6730 else if (hba
&& hba
->type
== SYS_DEV_SATA
)
6734 dprintf("path: %s hba: %s attached: %s\n",
6735 path
, (hba
) ? hba
->path
: "NULL", drv
);
6743 static int imsm_find_array_minor_by_subdev(int subdev
, int container
, int *minor
)
6745 char subdev_name
[20];
6746 struct mdstat_ent
*mdstat
;
6748 sprintf(subdev_name
, "%d", subdev
);
6749 mdstat
= mdstat_by_subdev(subdev_name
, container
);
6753 *minor
= mdstat
->devnum
;
6754 free_mdstat(mdstat
);
6758 static int imsm_reshape_is_allowed_on_container(struct supertype
*st
,
6759 struct geo_params
*geo
,
6760 int *old_raid_disks
)
6762 /* currently we only support increasing the number of devices
6763 * for a container. This increases the number of device for each
6764 * member array. They must all be RAID0 or RAID5.
6767 struct mdinfo
*info
, *member
;
6768 int devices_that_can_grow
= 0;
6770 dprintf("imsm: imsm_reshape_is_allowed_on_container(ENTER): "
6771 "st->devnum = (%i)\n",
6774 if (geo
->size
!= -1 ||
6775 geo
->level
!= UnSet
||
6776 geo
->layout
!= UnSet
||
6777 geo
->chunksize
!= 0 ||
6778 geo
->raid_disks
== UnSet
) {
6779 dprintf("imsm: Container operation is allowed for "
6780 "raid disks number change only.\n");
6784 info
= container_content_imsm(st
, NULL
);
6785 for (member
= info
; member
; member
= member
->next
) {
6789 dprintf("imsm: checking device_num: %i\n",
6790 member
->container_member
);
6792 if (geo
->raid_disks
<= member
->array
.raid_disks
) {
6793 /* we work on container for Online Capacity Expansion
6794 * only so raid_disks has to grow
6796 dprintf("imsm: for container operation raid disks "
6797 "increase is required\n");
6801 if ((info
->array
.level
!= 0) &&
6802 (info
->array
.level
!= 5)) {
6803 /* we cannot use this container with other raid level
6805 dprintf("imsm: for container operation wrong"
6806 " raid level (%i) detected\n",
6810 /* check for platform support
6811 * for this raid level configuration
6813 struct intel_super
*super
= st
->sb
;
6814 if (!is_raid_level_supported(super
->orom
,
6815 member
->array
.level
,
6817 dprintf("platform does not support raid%d with"
6821 geo
->raid_disks
> 1 ? "s" : "");
6826 if (*old_raid_disks
&&
6827 info
->array
.raid_disks
!= *old_raid_disks
)
6829 *old_raid_disks
= info
->array
.raid_disks
;
6831 /* All raid5 and raid0 volumes in container
6832 * have to be ready for Online Capacity Expansion
6833 * so they need to be assembled. We have already
6834 * checked that no recovery etc is happening.
6836 result
= imsm_find_array_minor_by_subdev(member
->container_member
,
6840 dprintf("imsm: cannot find array\n");
6843 devices_that_can_grow
++;
6846 if (!member
&& devices_that_can_grow
)
6850 dprintf("\tContainer operation allowed\n");
6852 dprintf("\tError: %i\n", ret_val
);
6857 /* Function: get_spares_for_grow
6858 * Description: Allocates memory and creates list of spare devices
6859 * avaliable in container. Checks if spare drive size is acceptable.
6860 * Parameters: Pointer to the supertype structure
6861 * Returns: Pointer to the list of spare devices (mdinfo structure) on success,
6864 static struct mdinfo
*get_spares_for_grow(struct supertype
*st
)
6866 unsigned long long min_size
= min_acceptable_spare_size_imsm(st
);
6867 return container_choose_spares(st
, min_size
, NULL
, NULL
, NULL
, 0);
6870 /******************************************************************************
6871 * function: imsm_create_metadata_update_for_reshape
6872 * Function creates update for whole IMSM container.
6874 ******************************************************************************/
6875 static int imsm_create_metadata_update_for_reshape(
6876 struct supertype
*st
,
6877 struct geo_params
*geo
,
6879 struct imsm_update_reshape
**updatep
)
6881 struct intel_super
*super
= st
->sb
;
6882 struct imsm_super
*mpb
= super
->anchor
;
6883 int update_memory_size
= 0;
6884 struct imsm_update_reshape
*u
= NULL
;
6885 struct mdinfo
*spares
= NULL
;
6887 int delta_disks
= 0;
6890 dprintf("imsm_update_metadata_for_reshape(enter) raid_disks = %i\n",
6893 delta_disks
= geo
->raid_disks
- old_raid_disks
;
6895 /* size of all update data without anchor */
6896 update_memory_size
= sizeof(struct imsm_update_reshape
);
6898 /* now add space for spare disks that we need to add. */
6899 update_memory_size
+= sizeof(u
->new_disks
[0]) * (delta_disks
- 1);
6901 u
= calloc(1, update_memory_size
);
6904 "cannot get memory for imsm_update_reshape update\n");
6907 u
->type
= update_reshape_container_disks
;
6908 u
->old_raid_disks
= old_raid_disks
;
6909 u
->new_raid_disks
= geo
->raid_disks
;
6911 /* now get spare disks list
6913 spares
= get_spares_for_grow(st
);
6916 || delta_disks
> spares
->array
.spare_disks
) {
6917 fprintf(stderr
, Name
": imsm: ERROR: Cannot get spare devices "
6918 "for %s.\n", geo
->dev_name
);
6922 /* we have got spares
6923 * update disk list in imsm_disk list table in anchor
6925 dprintf("imsm: %i spares are available.\n\n",
6926 spares
->array
.spare_disks
);
6929 for (i
= 0; i
< delta_disks
; i
++) {
6934 u
->new_disks
[i
] = makedev(dev
->disk
.major
,
6936 dl
= get_disk_super(super
, dev
->disk
.major
, dev
->disk
.minor
);
6937 dl
->index
= mpb
->num_disks
;
6947 dprintf("imsm: reshape update preparation :");
6948 if (i
== delta_disks
) {
6951 return update_memory_size
;
6954 dprintf(" Error\n");
6959 static void imsm_update_metadata_locally(struct supertype
*st
,
6962 struct metadata_update mu
;
6967 mu
.space_list
= NULL
;
6969 imsm_prepare_update(st
, &mu
);
6970 imsm_process_update(st
, &mu
);
6972 while (mu
.space_list
) {
6973 void **space
= mu
.space_list
;
6974 mu
.space_list
= *space
;
6979 /***************************************************************************
6980 * Function: imsm_analyze_change
6981 * Description: Function analyze change for single volume
6982 * and validate if transition is supported
6983 * Parameters: Geometry parameters, supertype structure
6984 * Returns: Operation type code on success, -1 if fail
6985 ****************************************************************************/
6986 enum imsm_reshape_type
imsm_analyze_change(struct supertype
*st
,
6987 struct geo_params
*geo
)
6994 getinfo_super_imsm_volume(st
, &info
, NULL
);
6996 if ((geo
->level
!= info
.array
.level
) &&
6997 (geo
->level
>= 0) &&
6998 (geo
->level
!= UnSet
)) {
6999 switch (info
.array
.level
) {
7001 if (geo
->level
== 5) {
7002 change
= CH_MIGRATION
;
7005 if (geo
->level
== 10) {
7006 change
= CH_TAKEOVER
;
7011 if (geo
->level
== 0) {
7012 change
= CH_TAKEOVER
;
7017 if (geo
->level
== 0)
7018 change
= CH_MIGRATION
;
7021 if (geo
->level
== 0) {
7022 change
= CH_TAKEOVER
;
7029 Name
" Error. Level Migration from %d to %d "
7031 info
.array
.level
, geo
->level
);
7032 goto analyse_change_exit
;
7035 geo
->level
= info
.array
.level
;
7037 if ((geo
->layout
!= info
.array
.layout
)
7038 && ((geo
->layout
!= UnSet
) && (geo
->layout
!= -1))) {
7039 change
= CH_MIGRATION
;
7040 if ((info
.array
.layout
== 0)
7041 && (info
.array
.level
== 5)
7042 && (geo
->layout
== 5)) {
7043 /* reshape 5 -> 4 */
7044 } else if ((info
.array
.layout
== 5)
7045 && (info
.array
.level
== 5)
7046 && (geo
->layout
== 0)) {
7047 /* reshape 4 -> 5 */
7052 Name
" Error. Layout Migration from %d to %d "
7054 info
.array
.layout
, geo
->layout
);
7056 goto analyse_change_exit
;
7059 geo
->layout
= info
.array
.layout
;
7061 if ((geo
->chunksize
> 0) && (geo
->chunksize
!= UnSet
)
7062 && (geo
->chunksize
!= info
.array
.chunk_size
))
7063 change
= CH_MIGRATION
;
7065 geo
->chunksize
= info
.array
.chunk_size
;
7067 chunk
= geo
->chunksize
/ 1024;
7068 if (!validate_geometry_imsm(st
,
7078 struct intel_super
*super
= st
->sb
;
7079 struct imsm_super
*mpb
= super
->anchor
;
7081 if (mpb
->num_raid_devs
> 1) {
7083 Name
" Error. Cannot perform operation on %s"
7084 "- for this operation it MUST be single "
7085 "array in container\n",
7091 analyse_change_exit
:
7096 int imsm_takeover(struct supertype
*st
, struct geo_params
*geo
)
7098 struct intel_super
*super
= st
->sb
;
7099 struct imsm_update_takeover
*u
;
7101 u
= malloc(sizeof(struct imsm_update_takeover
));
7105 u
->type
= update_takeover
;
7106 u
->subarray
= super
->current_vol
;
7108 /* 10->0 transition */
7109 if (geo
->level
== 0)
7110 u
->direction
= R10_TO_R0
;
7112 /* 0->10 transition */
7113 if (geo
->level
== 10)
7114 u
->direction
= R0_TO_R10
;
7116 /* update metadata locally */
7117 imsm_update_metadata_locally(st
, u
,
7118 sizeof(struct imsm_update_takeover
));
7119 /* and possibly remotely */
7120 if (st
->update_tail
)
7121 append_metadata_update(st
, u
,
7122 sizeof(struct imsm_update_takeover
));
7129 static int imsm_reshape_super(struct supertype
*st
, long long size
, int level
,
7130 int layout
, int chunksize
, int raid_disks
,
7131 int delta_disks
, char *backup
, char *dev
,
7135 struct geo_params geo
;
7137 dprintf("imsm: reshape_super called.\n");
7139 memset(&geo
, 0, sizeof(struct geo_params
));
7142 geo
.dev_id
= st
->devnum
;
7145 geo
.layout
= layout
;
7146 geo
.chunksize
= chunksize
;
7147 geo
.raid_disks
= raid_disks
;
7148 if (delta_disks
!= UnSet
)
7149 geo
.raid_disks
+= delta_disks
;
7151 dprintf("\tfor level : %i\n", geo
.level
);
7152 dprintf("\tfor raid_disks : %i\n", geo
.raid_disks
);
7154 if (experimental() == 0)
7157 if (st
->container_dev
== st
->devnum
) {
7158 /* On container level we can only increase number of devices. */
7159 dprintf("imsm: info: Container operation\n");
7160 int old_raid_disks
= 0;
7161 if (imsm_reshape_is_allowed_on_container(
7162 st
, &geo
, &old_raid_disks
)) {
7163 struct imsm_update_reshape
*u
= NULL
;
7166 len
= imsm_create_metadata_update_for_reshape(
7167 st
, &geo
, old_raid_disks
, &u
);
7170 dprintf("imsm: Cannot prepare update\n");
7171 goto exit_imsm_reshape_super
;
7175 /* update metadata locally */
7176 imsm_update_metadata_locally(st
, u
, len
);
7177 /* and possibly remotely */
7178 if (st
->update_tail
)
7179 append_metadata_update(st
, u
, len
);
7184 fprintf(stderr
, Name
": (imsm) Operation "
7185 "is not allowed on this container\n");
7188 /* On volume level we support following operations
7189 * - takeover: raid10 -> raid0; raid0 -> raid10
7190 * - chunk size migration
7191 * - migration: raid5 -> raid0; raid0 -> raid5
7193 struct intel_super
*super
= st
->sb
;
7194 struct intel_dev
*dev
= super
->devlist
;
7196 dprintf("imsm: info: Volume operation\n");
7197 /* find requested device */
7199 imsm_find_array_minor_by_subdev(dev
->index
, st
->container_dev
, &devnum
);
7200 if (devnum
== geo
.dev_id
)
7205 fprintf(stderr
, Name
" Cannot find %s (%i) subarray\n",
7206 geo
.dev_name
, geo
.dev_id
);
7207 goto exit_imsm_reshape_super
;
7209 super
->current_vol
= dev
->index
;
7210 change
= imsm_analyze_change(st
, &geo
);
7213 ret_val
= imsm_takeover(st
, &geo
);
7223 exit_imsm_reshape_super
:
7224 dprintf("imsm: reshape_super Exit code = %i\n", ret_val
);
7228 static int imsm_manage_reshape(
7229 int afd
, struct mdinfo
*sra
, struct reshape
*reshape
,
7230 struct supertype
*st
, unsigned long stripes
,
7231 int *fds
, unsigned long long *offsets
,
7232 int dests
, int *destfd
, unsigned long long *destoffsets
)
7234 /* Just use child_monitor for now */
7235 return child_monitor(
7236 afd
, sra
, reshape
, st
, stripes
,
7237 fds
, offsets
, dests
, destfd
, destoffsets
);
7239 #endif /* MDASSEMBLE */
7241 struct superswitch super_imsm
= {
7243 .examine_super
= examine_super_imsm
,
7244 .brief_examine_super
= brief_examine_super_imsm
,
7245 .brief_examine_subarrays
= brief_examine_subarrays_imsm
,
7246 .export_examine_super
= export_examine_super_imsm
,
7247 .detail_super
= detail_super_imsm
,
7248 .brief_detail_super
= brief_detail_super_imsm
,
7249 .write_init_super
= write_init_super_imsm
,
7250 .validate_geometry
= validate_geometry_imsm
,
7251 .add_to_super
= add_to_super_imsm
,
7252 .remove_from_super
= remove_from_super_imsm
,
7253 .detail_platform
= detail_platform_imsm
,
7254 .kill_subarray
= kill_subarray_imsm
,
7255 .update_subarray
= update_subarray_imsm
,
7256 .load_container
= load_container_imsm
,
7257 .default_geometry
= default_geometry_imsm
,
7258 .get_disk_controller_domain
= imsm_get_disk_controller_domain
,
7259 .reshape_super
= imsm_reshape_super
,
7260 .manage_reshape
= imsm_manage_reshape
,
7262 .match_home
= match_home_imsm
,
7263 .uuid_from_super
= uuid_from_super_imsm
,
7264 .getinfo_super
= getinfo_super_imsm
,
7265 .getinfo_super_disks
= getinfo_super_disks_imsm
,
7266 .update_super
= update_super_imsm
,
7268 .avail_size
= avail_size_imsm
,
7269 .min_acceptable_spare_size
= min_acceptable_spare_size_imsm
,
7271 .compare_super
= compare_super_imsm
,
7273 .load_super
= load_super_imsm
,
7274 .init_super
= init_super_imsm
,
7275 .store_super
= store_super_imsm
,
7276 .free_super
= free_super_imsm
,
7277 .match_metadata_desc
= match_metadata_desc_imsm
,
7278 .container_content
= container_content_imsm
,
7285 .open_new
= imsm_open_new
,
7286 .set_array_state
= imsm_set_array_state
,
7287 .set_disk
= imsm_set_disk
,
7288 .sync_metadata
= imsm_sync_metadata
,
7289 .activate_spare
= imsm_activate_spare
,
7290 .process_update
= imsm_process_update
,
7291 .prepare_update
= imsm_prepare_update
,
7292 #endif /* MDASSEMBLE */