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
57 /* Disk configuration info. */
58 #define IMSM_MAX_DEVICES 255
60 __u8 serial
[MAX_RAID_SERIAL_LEN
];/* 0xD8 - 0xE7 ascii serial number */
61 __u32 total_blocks
; /* 0xE8 - 0xEB total blocks */
62 __u32 scsi_id
; /* 0xEC - 0xEF scsi ID */
63 #define SPARE_DISK __cpu_to_le32(0x01) /* Spare */
64 #define CONFIGURED_DISK __cpu_to_le32(0x02) /* Member of some RaidDev */
65 #define FAILED_DISK __cpu_to_le32(0x04) /* Permanent failure */
66 #define USABLE_DISK __cpu_to_le32(0x08) /* Fully usable unless FAILED_DISK is set */
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
;
236 /* internal representation of IMSM metadata */
239 void *buf
; /* O_DIRECT buffer for reading/writing metadata */
240 struct imsm_super
*anchor
; /* immovable parameters */
242 size_t len
; /* size of the 'buf' allocation */
243 void *next_buf
; /* for realloc'ing buf from the manager */
245 int updates_pending
; /* count of pending updates for mdmon */
246 int creating_imsm
; /* flag to indicate container creation */
247 int current_vol
; /* index of raid device undergoing creation */
248 __u32 create_offset
; /* common start for 'current_vol' */
249 struct intel_dev
*devlist
;
253 __u8 serial
[MAX_RAID_SERIAL_LEN
];
256 struct imsm_disk disk
;
259 struct extent
*e
; /* for determining freespace @ create */
260 int raiddisk
; /* slot to fill in autolayout */
262 struct dl
*add
; /* list of disks to add while mdmon active */
263 struct dl
*missing
; /* disks removed while we weren't looking */
264 struct bbm_log
*bbm_log
;
265 const char *hba
; /* device path of the raid controller for this metadata */
266 const struct imsm_orom
*orom
; /* platform firmware support */
270 unsigned long long start
, size
;
273 /* definition of messages passed to imsm_process_update */
274 enum imsm_update_type
{
275 update_activate_spare
,
280 struct imsm_update_activate_spare
{
281 enum imsm_update_type type
;
285 struct imsm_update_activate_spare
*next
;
289 __u8 serial
[MAX_RAID_SERIAL_LEN
];
292 struct imsm_update_create_array
{
293 enum imsm_update_type type
;
298 struct imsm_update_add_disk
{
299 enum imsm_update_type type
;
302 static struct supertype
*match_metadata_desc_imsm(char *arg
)
304 struct supertype
*st
;
306 if (strcmp(arg
, "imsm") != 0 &&
307 strcmp(arg
, "default") != 0
311 st
= malloc(sizeof(*st
));
312 memset(st
, 0, sizeof(*st
));
313 st
->ss
= &super_imsm
;
314 st
->max_devs
= IMSM_MAX_DEVICES
;
315 st
->minor_version
= 0;
321 static __u8
*get_imsm_version(struct imsm_super
*mpb
)
323 return &mpb
->sig
[MPB_SIG_LEN
];
327 /* retrieve a disk directly from the anchor when the anchor is known to be
328 * up-to-date, currently only at load time
330 static struct imsm_disk
*__get_imsm_disk(struct imsm_super
*mpb
, __u8 index
)
332 if (index
>= mpb
->num_disks
)
334 return &mpb
->disk
[index
];
338 /* retrieve a disk from the parsed metadata */
339 static struct imsm_disk
*get_imsm_disk(struct intel_super
*super
, __u8 index
)
343 for (d
= super
->disks
; d
; d
= d
->next
)
344 if (d
->index
== index
)
351 /* generate a checksum directly from the anchor when the anchor is known to be
352 * up-to-date, currently only at load or write_super after coalescing
354 static __u32
__gen_imsm_checksum(struct imsm_super
*mpb
)
356 __u32 end
= mpb
->mpb_size
/ sizeof(end
);
357 __u32
*p
= (__u32
*) mpb
;
361 sum
+= __le32_to_cpu(*p
);
365 return sum
- __le32_to_cpu(mpb
->check_sum
);
368 static size_t sizeof_imsm_map(struct imsm_map
*map
)
370 return sizeof(struct imsm_map
) + sizeof(__u32
) * (map
->num_members
- 1);
373 struct imsm_map
*get_imsm_map(struct imsm_dev
*dev
, int second_map
)
375 struct imsm_map
*map
= &dev
->vol
.map
[0];
377 if (second_map
&& !dev
->vol
.migr_state
)
379 else if (second_map
) {
382 return ptr
+ sizeof_imsm_map(map
);
388 /* return the size of the device.
389 * migr_state increases the returned size if map[0] were to be duplicated
391 static size_t sizeof_imsm_dev(struct imsm_dev
*dev
, int migr_state
)
393 size_t size
= sizeof(*dev
) - sizeof(struct imsm_map
) +
394 sizeof_imsm_map(get_imsm_map(dev
, 0));
396 /* migrating means an additional map */
397 if (dev
->vol
.migr_state
)
398 size
+= sizeof_imsm_map(get_imsm_map(dev
, 1));
400 size
+= sizeof_imsm_map(get_imsm_map(dev
, 0));
406 /* retrieve disk serial number list from a metadata update */
407 static struct disk_info
*get_disk_info(struct imsm_update_create_array
*update
)
410 struct disk_info
*inf
;
412 inf
= u
+ sizeof(*update
) - sizeof(struct imsm_dev
) +
413 sizeof_imsm_dev(&update
->dev
, 0);
419 static struct imsm_dev
*__get_imsm_dev(struct imsm_super
*mpb
, __u8 index
)
425 if (index
>= mpb
->num_raid_devs
)
428 /* devices start after all disks */
429 offset
= ((void *) &mpb
->disk
[mpb
->num_disks
]) - _mpb
;
431 for (i
= 0; i
<= index
; i
++)
433 return _mpb
+ offset
;
435 offset
+= sizeof_imsm_dev(_mpb
+ offset
, 0);
440 static struct imsm_dev
*get_imsm_dev(struct intel_super
*super
, __u8 index
)
442 struct intel_dev
*dv
;
444 if (index
>= super
->anchor
->num_raid_devs
)
446 for (dv
= super
->devlist
; dv
; dv
= dv
->next
)
447 if (dv
->index
== index
)
452 static __u32
get_imsm_ord_tbl_ent(struct imsm_dev
*dev
, int slot
)
454 struct imsm_map
*map
;
456 if (dev
->vol
.migr_state
)
457 map
= get_imsm_map(dev
, 1);
459 map
= get_imsm_map(dev
, 0);
461 /* top byte identifies disk under rebuild */
462 return __le32_to_cpu(map
->disk_ord_tbl
[slot
]);
465 #define ord_to_idx(ord) (((ord) << 8) >> 8)
466 static __u32
get_imsm_disk_idx(struct imsm_dev
*dev
, int slot
)
468 __u32 ord
= get_imsm_ord_tbl_ent(dev
, slot
);
470 return ord_to_idx(ord
);
473 static void set_imsm_ord_tbl_ent(struct imsm_map
*map
, int slot
, __u32 ord
)
475 map
->disk_ord_tbl
[slot
] = __cpu_to_le32(ord
);
478 static int get_imsm_disk_slot(struct imsm_map
*map
, int idx
)
483 for (slot
= 0; slot
< map
->num_members
; slot
++) {
484 ord
= __le32_to_cpu(map
->disk_ord_tbl
[slot
]);
485 if (ord_to_idx(ord
) == idx
)
492 static int get_imsm_raid_level(struct imsm_map
*map
)
494 if (map
->raid_level
== 1) {
495 if (map
->num_members
== 2)
501 return map
->raid_level
;
504 static int cmp_extent(const void *av
, const void *bv
)
506 const struct extent
*a
= av
;
507 const struct extent
*b
= bv
;
508 if (a
->start
< b
->start
)
510 if (a
->start
> b
->start
)
515 static int count_memberships(struct dl
*dl
, struct intel_super
*super
)
520 for (i
= 0; i
< super
->anchor
->num_raid_devs
; i
++) {
521 struct imsm_dev
*dev
= get_imsm_dev(super
, i
);
522 struct imsm_map
*map
= get_imsm_map(dev
, 0);
524 if (get_imsm_disk_slot(map
, dl
->index
) >= 0)
531 static struct extent
*get_extents(struct intel_super
*super
, struct dl
*dl
)
533 /* find a list of used extents on the given physical device */
534 struct extent
*rv
, *e
;
536 int memberships
= count_memberships(dl
, super
);
537 __u32 reservation
= MPB_SECTOR_CNT
+ IMSM_RESERVED_SECTORS
;
539 rv
= malloc(sizeof(struct extent
) * (memberships
+ 1));
544 for (i
= 0; i
< super
->anchor
->num_raid_devs
; i
++) {
545 struct imsm_dev
*dev
= get_imsm_dev(super
, i
);
546 struct imsm_map
*map
= get_imsm_map(dev
, 0);
548 if (get_imsm_disk_slot(map
, dl
->index
) >= 0) {
549 e
->start
= __le32_to_cpu(map
->pba_of_lba0
);
550 e
->size
= __le32_to_cpu(map
->blocks_per_member
);
554 qsort(rv
, memberships
, sizeof(*rv
), cmp_extent
);
556 /* determine the start of the metadata
557 * when no raid devices are defined use the default
558 * ...otherwise allow the metadata to truncate the value
559 * as is the case with older versions of imsm
562 struct extent
*last
= &rv
[memberships
- 1];
565 remainder
= __le32_to_cpu(dl
->disk
.total_blocks
) -
566 (last
->start
+ last
->size
);
567 /* round down to 1k block to satisfy precision of the kernel
571 /* make sure remainder is still sane */
572 if (remainder
< ROUND_UP(super
->len
, 512) >> 9)
573 remainder
= ROUND_UP(super
->len
, 512) >> 9;
574 if (reservation
> remainder
)
575 reservation
= remainder
;
577 e
->start
= __le32_to_cpu(dl
->disk
.total_blocks
) - reservation
;
582 /* try to determine how much space is reserved for metadata from
583 * the last get_extents() entry, otherwise fallback to the
586 static __u32
imsm_reserved_sectors(struct intel_super
*super
, struct dl
*dl
)
592 /* for spares just return a minimal reservation which will grow
593 * once the spare is picked up by an array
596 return MPB_SECTOR_CNT
;
598 e
= get_extents(super
, dl
);
600 return MPB_SECTOR_CNT
+ IMSM_RESERVED_SECTORS
;
602 /* scroll to last entry */
603 for (i
= 0; e
[i
].size
; i
++)
606 rv
= __le32_to_cpu(dl
->disk
.total_blocks
) - e
[i
].start
;
614 static void print_imsm_dev(struct imsm_dev
*dev
, char *uuid
, int disk_idx
)
618 struct imsm_map
*map
= get_imsm_map(dev
, 0);
622 printf("[%.16s]:\n", dev
->volume
);
623 printf(" UUID : %s\n", uuid
);
624 printf(" RAID Level : %d\n", get_imsm_raid_level(map
));
625 printf(" Members : %d\n", map
->num_members
);
626 slot
= get_imsm_disk_slot(map
, disk_idx
);
628 ord
= get_imsm_ord_tbl_ent(dev
, slot
);
629 printf(" This Slot : %d%s\n", slot
,
630 ord
& IMSM_ORD_REBUILD
? " (out-of-sync)" : "");
632 printf(" This Slot : ?\n");
633 sz
= __le32_to_cpu(dev
->size_high
);
635 sz
+= __le32_to_cpu(dev
->size_low
);
636 printf(" Array Size : %llu%s\n", (unsigned long long)sz
,
637 human_size(sz
* 512));
638 sz
= __le32_to_cpu(map
->blocks_per_member
);
639 printf(" Per Dev Size : %llu%s\n", (unsigned long long)sz
,
640 human_size(sz
* 512));
641 printf(" Sector Offset : %u\n",
642 __le32_to_cpu(map
->pba_of_lba0
));
643 printf(" Num Stripes : %u\n",
644 __le32_to_cpu(map
->num_data_stripes
));
645 printf(" Chunk Size : %u KiB\n",
646 __le16_to_cpu(map
->blocks_per_strip
) / 2);
647 printf(" Reserved : %d\n", __le32_to_cpu(dev
->reserved_blocks
));
648 printf(" Migrate State : %s", dev
->vol
.migr_state
? "migrating" : "idle\n");
649 if (dev
->vol
.migr_state
) {
650 if (migr_type(dev
) == MIGR_INIT
)
651 printf(": initializing\n");
652 else if (migr_type(dev
) == MIGR_REBUILD
)
653 printf(": rebuilding\n");
654 else if (migr_type(dev
) == MIGR_VERIFY
)
656 else if (migr_type(dev
) == MIGR_GEN_MIGR
)
657 printf(": general migration\n");
658 else if (migr_type(dev
) == MIGR_STATE_CHANGE
)
659 printf(": state change\n");
660 else if (migr_type(dev
) == MIGR_REPAIR
)
661 printf(": repair\n");
663 printf(": <unknown:%d>\n", migr_type(dev
));
665 printf(" Map State : %s", map_state_str
[map
->map_state
]);
666 if (dev
->vol
.migr_state
) {
667 struct imsm_map
*map
= get_imsm_map(dev
, 1);
668 printf(" <-- %s", map_state_str
[map
->map_state
]);
671 printf(" Dirty State : %s\n", dev
->vol
.dirty
? "dirty" : "clean");
674 static void print_imsm_disk(struct imsm_super
*mpb
, int index
, __u32 reserved
)
676 struct imsm_disk
*disk
= __get_imsm_disk(mpb
, index
);
677 char str
[MAX_RAID_SERIAL_LEN
+ 1];
685 snprintf(str
, MAX_RAID_SERIAL_LEN
+ 1, "%s", disk
->serial
);
686 printf(" Disk%02d Serial : %s\n", index
, str
);
688 printf(" State :%s%s%s%s\n", s
&SPARE_DISK
? " spare" : "",
689 s
&CONFIGURED_DISK
? " active" : "",
690 s
&FAILED_DISK
? " failed" : "",
691 s
&USABLE_DISK
? " usable" : "");
692 printf(" Id : %08x\n", __le32_to_cpu(disk
->scsi_id
));
693 sz
= __le32_to_cpu(disk
->total_blocks
) - reserved
;
694 printf(" Usable Size : %llu%s\n", (unsigned long long)sz
,
695 human_size(sz
* 512));
698 static void getinfo_super_imsm(struct supertype
*st
, struct mdinfo
*info
);
700 static void examine_super_imsm(struct supertype
*st
, char *homehost
)
702 struct intel_super
*super
= st
->sb
;
703 struct imsm_super
*mpb
= super
->anchor
;
704 char str
[MAX_SIGNATURE_LENGTH
];
709 __u32 reserved
= imsm_reserved_sectors(super
, super
->disks
);
712 snprintf(str
, MPB_SIG_LEN
, "%s", mpb
->sig
);
713 printf(" Magic : %s\n", str
);
714 snprintf(str
, strlen(MPB_VERSION_RAID0
), "%s", get_imsm_version(mpb
));
715 printf(" Version : %s\n", get_imsm_version(mpb
));
716 printf(" Family : %08x\n", __le32_to_cpu(mpb
->family_num
));
717 printf(" Generation : %08x\n", __le32_to_cpu(mpb
->generation_num
));
718 getinfo_super_imsm(st
, &info
);
719 fname_from_uuid(st
, &info
, nbuf
, ':');
720 printf(" UUID : %s\n", nbuf
+ 5);
721 sum
= __le32_to_cpu(mpb
->check_sum
);
722 printf(" Checksum : %08x %s\n", sum
,
723 __gen_imsm_checksum(mpb
) == sum
? "correct" : "incorrect");
724 printf(" MPB Sectors : %d\n", mpb_sectors(mpb
));
725 printf(" Disks : %d\n", mpb
->num_disks
);
726 printf(" RAID Devices : %d\n", mpb
->num_raid_devs
);
727 print_imsm_disk(mpb
, super
->disks
->index
, reserved
);
728 if (super
->bbm_log
) {
729 struct bbm_log
*log
= super
->bbm_log
;
732 printf("Bad Block Management Log:\n");
733 printf(" Log Size : %d\n", __le32_to_cpu(mpb
->bbm_log_size
));
734 printf(" Signature : %x\n", __le32_to_cpu(log
->signature
));
735 printf(" Entry Count : %d\n", __le32_to_cpu(log
->entry_count
));
736 printf(" Spare Blocks : %d\n", __le32_to_cpu(log
->reserved_spare_block_count
));
737 printf(" First Spare : %llx\n", __le64_to_cpu(log
->first_spare_lba
));
739 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
741 struct imsm_dev
*dev
= __get_imsm_dev(mpb
, i
);
743 super
->current_vol
= i
;
744 getinfo_super_imsm(st
, &info
);
745 fname_from_uuid(st
, &info
, nbuf
, ':');
746 print_imsm_dev(dev
, nbuf
+ 5, super
->disks
->index
);
748 for (i
= 0; i
< mpb
->num_disks
; i
++) {
749 if (i
== super
->disks
->index
)
751 print_imsm_disk(mpb
, i
, reserved
);
755 static void brief_examine_super_imsm(struct supertype
*st
)
757 /* We just write a generic IMSM ARRAY entry */
761 struct intel_super
*super
= st
->sb
;
764 if (!super
->anchor
->num_raid_devs
)
767 getinfo_super_imsm(st
, &info
);
768 fname_from_uuid(st
, &info
, nbuf
, ':');
769 printf("ARRAY metadata=imsm auto=md UUID=%s\n", nbuf
+ 5);
770 for (i
= 0; i
< super
->anchor
->num_raid_devs
; i
++) {
771 struct imsm_dev
*dev
= get_imsm_dev(super
, i
);
773 super
->current_vol
= i
;
774 getinfo_super_imsm(st
, &info
);
775 fname_from_uuid(st
, &info
, nbuf1
, ':');
776 printf("ARRAY /dev/md/%.16s container=%s\n"
777 " member=%d auto=mdp UUID=%s\n",
778 dev
->volume
, nbuf
+ 5, i
, nbuf1
+ 5);
782 static void export_examine_super_imsm(struct supertype
*st
)
784 struct intel_super
*super
= st
->sb
;
785 struct imsm_super
*mpb
= super
->anchor
;
789 getinfo_super_imsm(st
, &info
);
790 fname_from_uuid(st
, &info
, nbuf
, ':');
791 printf("MD_METADATA=imsm\n");
792 printf("MD_LEVEL=container\n");
793 printf("MD_UUID=%s\n", nbuf
+5);
794 printf("MD_DEVICES=%u\n", mpb
->num_disks
);
797 static void detail_super_imsm(struct supertype
*st
, char *homehost
)
802 getinfo_super_imsm(st
, &info
);
803 fname_from_uuid(st
, &info
, nbuf
, ':');
804 printf("\n UUID : %s\n", nbuf
+ 5);
807 static void brief_detail_super_imsm(struct supertype
*st
)
811 getinfo_super_imsm(st
, &info
);
812 fname_from_uuid(st
, &info
, nbuf
, ':');
813 printf(" UUID=%s", nbuf
+ 5);
816 static int imsm_read_serial(int fd
, char *devname
, __u8
*serial
);
817 static void fd2devname(int fd
, char *name
);
819 static int imsm_enumerate_ports(const char *hba_path
, int port_count
, int host_base
, int verbose
)
821 /* dump an unsorted list of devices attached to ahci, as well as
822 * non-connected ports
824 int hba_len
= strlen(hba_path
) + 1;
829 unsigned long port_mask
= (1 << port_count
) - 1;
831 if (port_count
> sizeof(port_mask
) * 8) {
833 fprintf(stderr
, Name
": port_count %d out of range\n", port_count
);
837 /* scroll through /sys/dev/block looking for devices attached to
840 dir
= opendir("/sys/dev/block");
841 for (ent
= dir
? readdir(dir
) : NULL
; ent
; ent
= readdir(dir
)) {
852 if (sscanf(ent
->d_name
, "%d:%d", &major
, &minor
) != 2)
854 path
= devt_to_devpath(makedev(major
, minor
));
857 if (!path_attached_to_hba(path
, hba_path
)) {
863 /* retrieve the scsi device type */
864 if (asprintf(&device
, "/sys/dev/block/%d:%d/device/xxxxxxx", major
, minor
) < 0) {
866 fprintf(stderr
, Name
": failed to allocate 'device'\n");
870 sprintf(device
, "/sys/dev/block/%d:%d/device/type", major
, minor
);
871 if (load_sys(device
, buf
) != 0) {
873 fprintf(stderr
, Name
": failed to read device type for %s\n",
879 type
= strtoul(buf
, NULL
, 10);
881 /* if it's not a disk print the vendor and model */
882 if (!(type
== 0 || type
== 7 || type
== 14)) {
885 sprintf(device
, "/sys/dev/block/%d:%d/device/vendor", major
, minor
);
886 if (load_sys(device
, buf
) == 0) {
887 strncpy(vendor
, buf
, sizeof(vendor
));
888 vendor
[sizeof(vendor
) - 1] = '\0';
889 c
= (char *) &vendor
[sizeof(vendor
) - 1];
890 while (isspace(*c
) || *c
== '\0')
894 sprintf(device
, "/sys/dev/block/%d:%d/device/model", major
, minor
);
895 if (load_sys(device
, buf
) == 0) {
896 strncpy(model
, buf
, sizeof(model
));
897 model
[sizeof(model
) - 1] = '\0';
898 c
= (char *) &model
[sizeof(model
) - 1];
899 while (isspace(*c
) || *c
== '\0')
903 if (vendor
[0] && model
[0])
904 sprintf(buf
, "%.64s %.64s", vendor
, model
);
906 switch (type
) { /* numbers from hald/linux/device.c */
907 case 1: sprintf(buf
, "tape"); break;
908 case 2: sprintf(buf
, "printer"); break;
909 case 3: sprintf(buf
, "processor"); break;
911 case 5: sprintf(buf
, "cdrom"); break;
912 case 6: sprintf(buf
, "scanner"); break;
913 case 8: sprintf(buf
, "media_changer"); break;
914 case 9: sprintf(buf
, "comm"); break;
915 case 12: sprintf(buf
, "raid"); break;
916 default: sprintf(buf
, "unknown");
922 /* chop device path to 'host%d' and calculate the port number */
923 c
= strchr(&path
[hba_len
], '/');
925 if (sscanf(&path
[hba_len
], "host%d", &port
) == 1)
929 *c
= '/'; /* repair the full string */
930 fprintf(stderr
, Name
": failed to determine port number for %s\n",
937 /* mark this port as used */
938 port_mask
&= ~(1 << port
);
940 /* print out the device information */
942 printf(" Port%d : - non-disk device (%s) -\n", port
, buf
);
946 fd
= dev_open(ent
->d_name
, O_RDONLY
);
948 printf(" Port%d : - disk info unavailable -\n", port
);
951 printf(" Port%d : %s", port
, buf
);
952 if (imsm_read_serial(fd
, NULL
, (__u8
*) buf
) == 0)
953 printf(" (%s)\n", buf
);
968 for (i
= 0; i
< port_count
; i
++)
969 if (port_mask
& (1 << i
))
970 printf(" Port%d : - no device attached -\n", i
);
976 static int detail_platform_imsm(int verbose
, int enumerate_only
)
978 /* There are two components to imsm platform support, the ahci SATA
979 * controller and the option-rom. To find the SATA controller we
980 * simply look in /sys/bus/pci/drivers/ahci to see if an ahci
981 * controller with the Intel vendor id is present. This approach
982 * allows mdadm to leverage the kernel's ahci detection logic, with the
983 * caveat that if ahci.ko is not loaded mdadm will not be able to
984 * detect platform raid capabilities. The option-rom resides in a
985 * platform "Adapter ROM". We scan for its signature to retrieve the
986 * platform capabilities. If raid support is disabled in the BIOS the
987 * option-rom capability structure will not be available.
989 const struct imsm_orom
*orom
;
990 struct sys_dev
*list
, *hba
;
993 const char *hba_path
;
997 if (enumerate_only
) {
998 if (check_env("IMSM_NO_PLATFORM") || find_imsm_orom())
1003 list
= find_driver_devices("pci", "ahci");
1004 for (hba
= list
; hba
; hba
= hba
->next
)
1005 if (devpath_to_vendor(hba
->path
) == 0x8086)
1010 fprintf(stderr
, Name
": unable to find active ahci controller\n");
1011 free_sys_dev(&list
);
1014 fprintf(stderr
, Name
": found Intel SATA AHCI Controller\n");
1015 hba_path
= hba
->path
;
1017 free_sys_dev(&list
);
1019 orom
= find_imsm_orom();
1022 fprintf(stderr
, Name
": imsm option-rom not found\n");
1026 printf(" Platform : Intel(R) Matrix Storage Manager\n");
1027 printf(" Version : %d.%d.%d.%d\n", orom
->major_ver
, orom
->minor_ver
,
1028 orom
->hotfix_ver
, orom
->build
);
1029 printf(" RAID Levels :%s%s%s%s%s\n",
1030 imsm_orom_has_raid0(orom
) ? " raid0" : "",
1031 imsm_orom_has_raid1(orom
) ? " raid1" : "",
1032 imsm_orom_has_raid1e(orom
) ? " raid1e" : "",
1033 imsm_orom_has_raid10(orom
) ? " raid10" : "",
1034 imsm_orom_has_raid5(orom
) ? " raid5" : "");
1035 printf(" Chunk Sizes :%s%s%s%s%s%s%s%s%s%s%s%s%s%s%s%s\n",
1036 imsm_orom_has_chunk(orom
, 2) ? " 2k" : "",
1037 imsm_orom_has_chunk(orom
, 4) ? " 4k" : "",
1038 imsm_orom_has_chunk(orom
, 8) ? " 8k" : "",
1039 imsm_orom_has_chunk(orom
, 16) ? " 16k" : "",
1040 imsm_orom_has_chunk(orom
, 32) ? " 32k" : "",
1041 imsm_orom_has_chunk(orom
, 64) ? " 64k" : "",
1042 imsm_orom_has_chunk(orom
, 128) ? " 128k" : "",
1043 imsm_orom_has_chunk(orom
, 256) ? " 256k" : "",
1044 imsm_orom_has_chunk(orom
, 512) ? " 512k" : "",
1045 imsm_orom_has_chunk(orom
, 1024*1) ? " 1M" : "",
1046 imsm_orom_has_chunk(orom
, 1024*2) ? " 2M" : "",
1047 imsm_orom_has_chunk(orom
, 1024*4) ? " 4M" : "",
1048 imsm_orom_has_chunk(orom
, 1024*8) ? " 8M" : "",
1049 imsm_orom_has_chunk(orom
, 1024*16) ? " 16M" : "",
1050 imsm_orom_has_chunk(orom
, 1024*32) ? " 32M" : "",
1051 imsm_orom_has_chunk(orom
, 1024*64) ? " 64M" : "");
1052 printf(" Max Disks : %d\n", orom
->tds
);
1053 printf(" Max Volumes : %d\n", orom
->vpa
);
1054 printf(" I/O Controller : %s\n", hba_path
);
1056 /* find the smallest scsi host number to determine a port number base */
1057 dir
= opendir(hba_path
);
1058 for (ent
= dir
? readdir(dir
) : NULL
; ent
; ent
= readdir(dir
)) {
1061 if (sscanf(ent
->d_name
, "host%d", &host
) != 1)
1063 if (port_count
== 0)
1065 else if (host
< host_base
)
1068 if (host
+ 1 > port_count
+ host_base
)
1069 port_count
= host
+ 1 - host_base
;
1075 if (!port_count
|| imsm_enumerate_ports(hba_path
, port_count
,
1076 host_base
, verbose
) != 0) {
1078 fprintf(stderr
, Name
": failed to enumerate ports\n");
1086 static int match_home_imsm(struct supertype
*st
, char *homehost
)
1088 /* the imsm metadata format does not specify any host
1089 * identification information. We return -1 since we can never
1090 * confirm nor deny whether a given array is "meant" for this
1091 * host. We rely on compare_super and the 'family_num' field to
1092 * exclude member disks that do not belong, and we rely on
1093 * mdadm.conf to specify the arrays that should be assembled.
1094 * Auto-assembly may still pick up "foreign" arrays.
1100 static void uuid_from_super_imsm(struct supertype
*st
, int uuid
[4])
1102 /* The uuid returned here is used for:
1103 * uuid to put into bitmap file (Create, Grow)
1104 * uuid for backup header when saving critical section (Grow)
1105 * comparing uuids when re-adding a device into an array
1106 * In these cases the uuid required is that of the data-array,
1107 * not the device-set.
1108 * uuid to recognise same set when adding a missing device back
1109 * to an array. This is a uuid for the device-set.
1111 * For each of these we can make do with a truncated
1112 * or hashed uuid rather than the original, as long as
1114 * In each case the uuid required is that of the data-array,
1115 * not the device-set.
1117 /* imsm does not track uuid's so we synthesis one using sha1 on
1118 * - The signature (Which is constant for all imsm array, but no matter)
1119 * - the family_num of the container
1120 * - the index number of the volume
1121 * - the 'serial' number of the volume.
1122 * Hopefully these are all constant.
1124 struct intel_super
*super
= st
->sb
;
1127 struct sha1_ctx ctx
;
1128 struct imsm_dev
*dev
= NULL
;
1130 sha1_init_ctx(&ctx
);
1131 sha1_process_bytes(super
->anchor
->sig
, MPB_SIG_LEN
, &ctx
);
1132 sha1_process_bytes(&super
->anchor
->family_num
, sizeof(__u32
), &ctx
);
1133 if (super
->current_vol
>= 0)
1134 dev
= get_imsm_dev(super
, super
->current_vol
);
1136 __u32 vol
= super
->current_vol
;
1137 sha1_process_bytes(&vol
, sizeof(vol
), &ctx
);
1138 sha1_process_bytes(dev
->volume
, MAX_RAID_SERIAL_LEN
, &ctx
);
1140 sha1_finish_ctx(&ctx
, buf
);
1141 memcpy(uuid
, buf
, 4*4);
1146 get_imsm_numerical_version(struct imsm_super
*mpb
, int *m
, int *p
)
1148 __u8
*v
= get_imsm_version(mpb
);
1149 __u8
*end
= mpb
->sig
+ MAX_SIGNATURE_LENGTH
;
1150 char major
[] = { 0, 0, 0 };
1151 char minor
[] = { 0 ,0, 0 };
1152 char patch
[] = { 0, 0, 0 };
1153 char *ver_parse
[] = { major
, minor
, patch
};
1157 while (*v
!= '\0' && v
< end
) {
1158 if (*v
!= '.' && j
< 2)
1159 ver_parse
[i
][j
++] = *v
;
1167 *m
= strtol(minor
, NULL
, 0);
1168 *p
= strtol(patch
, NULL
, 0);
1172 static int imsm_level_to_layout(int level
)
1180 return ALGORITHM_LEFT_ASYMMETRIC
;
1187 static void getinfo_super_imsm_volume(struct supertype
*st
, struct mdinfo
*info
)
1189 struct intel_super
*super
= st
->sb
;
1190 struct imsm_dev
*dev
= get_imsm_dev(super
, super
->current_vol
);
1191 struct imsm_map
*map
= get_imsm_map(dev
, 0);
1194 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
1195 if (dl
->raiddisk
== info
->disk
.raid_disk
)
1197 info
->container_member
= super
->current_vol
;
1198 info
->array
.raid_disks
= map
->num_members
;
1199 info
->array
.level
= get_imsm_raid_level(map
);
1200 info
->array
.layout
= imsm_level_to_layout(info
->array
.level
);
1201 info
->array
.md_minor
= -1;
1202 info
->array
.ctime
= 0;
1203 info
->array
.utime
= 0;
1204 info
->array
.chunk_size
= __le16_to_cpu(map
->blocks_per_strip
) << 9;
1205 info
->array
.state
= !dev
->vol
.dirty
;
1206 info
->custom_array_size
= __le32_to_cpu(dev
->size_high
);
1207 info
->custom_array_size
<<= 32;
1208 info
->custom_array_size
|= __le32_to_cpu(dev
->size_low
);
1210 info
->disk
.major
= 0;
1211 info
->disk
.minor
= 0;
1213 info
->disk
.major
= dl
->major
;
1214 info
->disk
.minor
= dl
->minor
;
1217 info
->data_offset
= __le32_to_cpu(map
->pba_of_lba0
);
1218 info
->component_size
= __le32_to_cpu(map
->blocks_per_member
);
1219 memset(info
->uuid
, 0, sizeof(info
->uuid
));
1221 if (map
->map_state
== IMSM_T_STATE_UNINITIALIZED
|| dev
->vol
.dirty
)
1222 info
->resync_start
= 0;
1223 else if (dev
->vol
.migr_state
)
1224 /* FIXME add curr_migr_unit to resync_start conversion */
1225 info
->resync_start
= 0;
1227 info
->resync_start
= ~0ULL;
1229 strncpy(info
->name
, (char *) dev
->volume
, MAX_RAID_SERIAL_LEN
);
1230 info
->name
[MAX_RAID_SERIAL_LEN
] = 0;
1232 info
->array
.major_version
= -1;
1233 info
->array
.minor_version
= -2;
1234 sprintf(info
->text_version
, "/%s/%d",
1235 devnum2devname(st
->container_dev
),
1236 info
->container_member
);
1237 info
->safe_mode_delay
= 4000; /* 4 secs like the Matrix driver */
1238 uuid_from_super_imsm(st
, info
->uuid
);
1241 /* check the config file to see if we can return a real uuid for this spare */
1242 static void fixup_container_spare_uuid(struct mdinfo
*inf
)
1244 struct mddev_ident_s
*array_list
;
1246 if (inf
->array
.level
!= LEVEL_CONTAINER
||
1247 memcmp(inf
->uuid
, uuid_match_any
, sizeof(int[4])) != 0)
1250 array_list
= conf_get_ident(NULL
);
1252 for (; array_list
; array_list
= array_list
->next
) {
1253 if (array_list
->uuid_set
) {
1254 struct supertype
*_sst
; /* spare supertype */
1255 struct supertype
*_cst
; /* container supertype */
1257 _cst
= array_list
->st
;
1258 _sst
= _cst
->ss
->match_metadata_desc(inf
->text_version
);
1260 memcpy(inf
->uuid
, array_list
->uuid
, sizeof(int[4]));
1268 static void getinfo_super_imsm(struct supertype
*st
, struct mdinfo
*info
)
1270 struct intel_super
*super
= st
->sb
;
1271 struct imsm_disk
*disk
;
1274 if (super
->current_vol
>= 0) {
1275 getinfo_super_imsm_volume(st
, info
);
1279 /* Set raid_disks to zero so that Assemble will always pull in valid
1282 info
->array
.raid_disks
= 0;
1283 info
->array
.level
= LEVEL_CONTAINER
;
1284 info
->array
.layout
= 0;
1285 info
->array
.md_minor
= -1;
1286 info
->array
.ctime
= 0; /* N/A for imsm */
1287 info
->array
.utime
= 0;
1288 info
->array
.chunk_size
= 0;
1290 info
->disk
.major
= 0;
1291 info
->disk
.minor
= 0;
1292 info
->disk
.raid_disk
= -1;
1293 info
->reshape_active
= 0;
1294 info
->array
.major_version
= -1;
1295 info
->array
.minor_version
= -2;
1296 strcpy(info
->text_version
, "imsm");
1297 info
->safe_mode_delay
= 0;
1298 info
->disk
.number
= -1;
1299 info
->disk
.state
= 0;
1303 __u32 reserved
= imsm_reserved_sectors(super
, super
->disks
);
1305 disk
= &super
->disks
->disk
;
1306 info
->data_offset
= __le32_to_cpu(disk
->total_blocks
) - reserved
;
1307 info
->component_size
= reserved
;
1309 info
->disk
.state
= s
& CONFIGURED_DISK
? (1 << MD_DISK_ACTIVE
) : 0;
1310 /* we don't change info->disk.raid_disk here because
1311 * this state will be finalized in mdmon after we have
1312 * found the 'most fresh' version of the metadata
1314 info
->disk
.state
|= s
& FAILED_DISK
? (1 << MD_DISK_FAULTY
) : 0;
1315 info
->disk
.state
|= s
& SPARE_DISK
? 0 : (1 << MD_DISK_SYNC
);
1318 /* only call uuid_from_super_imsm when this disk is part of a populated container,
1319 * ->compare_super may have updated the 'num_raid_devs' field for spares
1321 if (info
->disk
.state
& (1 << MD_DISK_SYNC
) || super
->anchor
->num_raid_devs
)
1322 uuid_from_super_imsm(st
, info
->uuid
);
1324 memcpy(info
->uuid
, uuid_match_any
, sizeof(int[4]));
1325 fixup_container_spare_uuid(info
);
1329 static int update_super_imsm(struct supertype
*st
, struct mdinfo
*info
,
1330 char *update
, char *devname
, int verbose
,
1331 int uuid_set
, char *homehost
)
1335 /* For 'assemble' and 'force' we need to return non-zero if any
1336 * change was made. For others, the return value is ignored.
1337 * Update options are:
1338 * force-one : This device looks a bit old but needs to be included,
1339 * update age info appropriately.
1340 * assemble: clear any 'faulty' flag to allow this device to
1342 * force-array: Array is degraded but being forced, mark it clean
1343 * if that will be needed to assemble it.
1345 * newdev: not used ????
1346 * grow: Array has gained a new device - this is currently for
1348 * resync: mark as dirty so a resync will happen.
1349 * name: update the name - preserving the homehost
1351 * Following are not relevant for this imsm:
1352 * sparc2.2 : update from old dodgey metadata
1353 * super-minor: change the preferred_minor number
1354 * summaries: update redundant counters.
1355 * uuid: Change the uuid of the array to match watch is given
1356 * homehost: update the recorded homehost
1357 * _reshape_progress: record new reshape_progress position.
1360 //struct intel_super *super = st->sb;
1361 //struct imsm_super *mpb = super->mpb;
1363 if (strcmp(update
, "grow") == 0) {
1365 if (strcmp(update
, "resync") == 0) {
1366 /* dev->vol.dirty = 1; */
1369 /* IMSM has no concept of UUID or homehost */
1374 static size_t disks_to_mpb_size(int disks
)
1378 size
= sizeof(struct imsm_super
);
1379 size
+= (disks
- 1) * sizeof(struct imsm_disk
);
1380 size
+= 2 * sizeof(struct imsm_dev
);
1381 /* up to 2 maps per raid device (-2 for imsm_maps in imsm_dev */
1382 size
+= (4 - 2) * sizeof(struct imsm_map
);
1383 /* 4 possible disk_ord_tbl's */
1384 size
+= 4 * (disks
- 1) * sizeof(__u32
);
1389 static __u64
avail_size_imsm(struct supertype
*st
, __u64 devsize
)
1391 if (devsize
< (MPB_SECTOR_CNT
+ IMSM_RESERVED_SECTORS
))
1394 return devsize
- (MPB_SECTOR_CNT
+ IMSM_RESERVED_SECTORS
);
1397 static void free_devlist(struct intel_super
*super
)
1399 struct intel_dev
*dv
;
1401 while (super
->devlist
) {
1402 dv
= super
->devlist
->next
;
1403 free(super
->devlist
->dev
);
1404 free(super
->devlist
);
1405 super
->devlist
= dv
;
1409 static void imsm_copy_dev(struct imsm_dev
*dest
, struct imsm_dev
*src
)
1411 memcpy(dest
, src
, sizeof_imsm_dev(src
, 0));
1414 static int compare_super_imsm(struct supertype
*st
, struct supertype
*tst
)
1418 * 0 same, or first was empty, and second was copied
1419 * 1 second had wrong number
1421 * 3 wrong other info
1423 struct intel_super
*first
= st
->sb
;
1424 struct intel_super
*sec
= tst
->sb
;
1432 if (memcmp(first
->anchor
->sig
, sec
->anchor
->sig
, MAX_SIGNATURE_LENGTH
) != 0)
1435 /* if an anchor does not have num_raid_devs set then it is a free
1438 if (first
->anchor
->num_raid_devs
> 0 &&
1439 sec
->anchor
->num_raid_devs
> 0) {
1440 if (first
->anchor
->family_num
!= sec
->anchor
->family_num
)
1444 /* if 'first' is a spare promote it to a populated mpb with sec's
1447 if (first
->anchor
->num_raid_devs
== 0 &&
1448 sec
->anchor
->num_raid_devs
> 0) {
1450 struct intel_dev
*dv
;
1451 struct imsm_dev
*dev
;
1453 /* we need to copy raid device info from sec if an allocation
1454 * fails here we don't associate the spare
1456 for (i
= 0; i
< sec
->anchor
->num_raid_devs
; i
++) {
1457 dv
= malloc(sizeof(*dv
));
1460 dev
= malloc(sizeof_imsm_dev(get_imsm_dev(sec
, i
), 1));
1467 dv
->next
= first
->devlist
;
1468 first
->devlist
= dv
;
1470 if (i
<= sec
->anchor
->num_raid_devs
) {
1471 /* allocation failure */
1472 free_devlist(first
);
1473 fprintf(stderr
, "imsm: failed to associate spare\n");
1476 for (i
= 0; i
< sec
->anchor
->num_raid_devs
; i
++)
1477 imsm_copy_dev(get_imsm_dev(first
, i
), get_imsm_dev(sec
, i
));
1479 first
->anchor
->num_raid_devs
= sec
->anchor
->num_raid_devs
;
1480 first
->anchor
->family_num
= sec
->anchor
->family_num
;
1486 static void fd2devname(int fd
, char *name
)
1495 if (fstat(fd
, &st
) != 0)
1497 sprintf(path
, "/sys/dev/block/%d:%d",
1498 major(st
.st_rdev
), minor(st
.st_rdev
));
1500 rv
= readlink(path
, dname
, sizeof(dname
));
1505 nm
= strrchr(dname
, '/');
1507 snprintf(name
, MAX_RAID_SERIAL_LEN
, "/dev/%s", nm
);
1511 extern int scsi_get_serial(int fd
, void *buf
, size_t buf_len
);
1513 static int imsm_read_serial(int fd
, char *devname
,
1514 __u8 serial
[MAX_RAID_SERIAL_LEN
])
1516 unsigned char scsi_serial
[255];
1525 memset(scsi_serial
, 0, sizeof(scsi_serial
));
1527 rv
= scsi_get_serial(fd
, scsi_serial
, sizeof(scsi_serial
));
1529 if (rv
&& check_env("IMSM_DEVNAME_AS_SERIAL")) {
1530 memset(serial
, 0, MAX_RAID_SERIAL_LEN
);
1531 fd2devname(fd
, (char *) serial
);
1538 Name
": Failed to retrieve serial for %s\n",
1543 rsp_len
= scsi_serial
[3];
1547 Name
": Failed to retrieve serial for %s\n",
1551 rsp_buf
= (char *) &scsi_serial
[4];
1553 /* trim all whitespace and non-printable characters and convert
1556 for (i
= 0, dest
= rsp_buf
; i
< rsp_len
; i
++) {
1559 /* ':' is reserved for use in placeholder serial
1560 * numbers for missing disks
1568 len
= dest
- rsp_buf
;
1571 /* truncate leading characters */
1572 if (len
> MAX_RAID_SERIAL_LEN
) {
1573 dest
+= len
- MAX_RAID_SERIAL_LEN
;
1574 len
= MAX_RAID_SERIAL_LEN
;
1577 memset(serial
, 0, MAX_RAID_SERIAL_LEN
);
1578 memcpy(serial
, dest
, len
);
1583 static int serialcmp(__u8
*s1
, __u8
*s2
)
1585 return strncmp((char *) s1
, (char *) s2
, MAX_RAID_SERIAL_LEN
);
1588 static void serialcpy(__u8
*dest
, __u8
*src
)
1590 strncpy((char *) dest
, (char *) src
, MAX_RAID_SERIAL_LEN
);
1593 static struct dl
*serial_to_dl(__u8
*serial
, struct intel_super
*super
)
1597 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
1598 if (serialcmp(dl
->serial
, serial
) == 0)
1605 load_imsm_disk(int fd
, struct intel_super
*super
, char *devname
, int keep_fd
)
1612 __u8 serial
[MAX_RAID_SERIAL_LEN
];
1614 rv
= imsm_read_serial(fd
, devname
, serial
);
1619 /* check if this is a disk we have seen before. it may be a spare in
1620 * super->disks while the current anchor believes it is a raid member,
1621 * check if we need to update dl->index
1623 dl
= serial_to_dl(serial
, super
);
1625 dl
= malloc(sizeof(*dl
));
1632 Name
": failed to allocate disk buffer for %s\n",
1639 dl
->major
= major(stb
.st_rdev
);
1640 dl
->minor
= minor(stb
.st_rdev
);
1641 dl
->next
= super
->disks
;
1642 dl
->fd
= keep_fd
? fd
: -1;
1643 dl
->devname
= devname
? strdup(devname
) : NULL
;
1644 serialcpy(dl
->serial
, serial
);
1647 } else if (keep_fd
) {
1652 /* look up this disk's index in the current anchor */
1653 for (i
= 0; i
< super
->anchor
->num_disks
; i
++) {
1654 struct imsm_disk
*disk_iter
;
1656 disk_iter
= __get_imsm_disk(super
->anchor
, i
);
1658 if (serialcmp(disk_iter
->serial
, dl
->serial
) == 0) {
1659 dl
->disk
= *disk_iter
;
1660 /* only set index on disks that are a member of a
1661 * populated contianer, i.e. one with raid_devs
1663 if (dl
->disk
.status
& FAILED_DISK
)
1665 else if (dl
->disk
.status
& SPARE_DISK
)
1674 /* no match, maybe a stale failed drive */
1675 if (i
== super
->anchor
->num_disks
&& dl
->index
>= 0) {
1676 dl
->disk
= *__get_imsm_disk(super
->anchor
, dl
->index
);
1677 if (dl
->disk
.status
& FAILED_DISK
)
1688 /* When migrating map0 contains the 'destination' state while map1
1689 * contains the current state. When not migrating map0 contains the
1690 * current state. This routine assumes that map[0].map_state is set to
1691 * the current array state before being called.
1693 * Migration is indicated by one of the following states
1694 * 1/ Idle (migr_state=0 map0state=normal||unitialized||degraded||failed)
1695 * 2/ Initialize (migr_state=1 migr_type=MIGR_INIT map0state=normal
1696 * map1state=unitialized)
1697 * 3/ Repair (Resync) (migr_state=1 migr_type=MIGR_REPAIR map0state=normal
1699 * 4/ Rebuild (migr_state=1 migr_type=MIGR_REBUILD map0state=normal
1700 * map1state=degraded)
1702 static void migrate(struct imsm_dev
*dev
, __u8 to_state
, int migr_type
)
1704 struct imsm_map
*dest
;
1705 struct imsm_map
*src
= get_imsm_map(dev
, 0);
1707 dev
->vol
.migr_state
= 1;
1708 set_migr_type(dev
, migr_type
);
1709 dev
->vol
.curr_migr_unit
= 0;
1710 dest
= get_imsm_map(dev
, 1);
1712 /* duplicate and then set the target end state in map[0] */
1713 memcpy(dest
, src
, sizeof_imsm_map(src
));
1714 if (migr_type
== MIGR_REBUILD
) {
1718 for (i
= 0; i
< src
->num_members
; i
++) {
1719 ord
= __le32_to_cpu(src
->disk_ord_tbl
[i
]);
1720 set_imsm_ord_tbl_ent(src
, i
, ord_to_idx(ord
));
1724 src
->map_state
= to_state
;
1727 static void end_migration(struct imsm_dev
*dev
, __u8 map_state
)
1729 struct imsm_map
*map
= get_imsm_map(dev
, 0);
1730 struct imsm_map
*prev
= get_imsm_map(dev
, dev
->vol
.migr_state
);
1733 /* merge any IMSM_ORD_REBUILD bits that were not successfully
1734 * completed in the last migration.
1736 * FIXME add support for online capacity expansion and
1737 * raid-level-migration
1739 for (i
= 0; i
< prev
->num_members
; i
++)
1740 map
->disk_ord_tbl
[i
] |= prev
->disk_ord_tbl
[i
];
1742 dev
->vol
.migr_state
= 0;
1743 dev
->vol
.curr_migr_unit
= 0;
1744 map
->map_state
= map_state
;
1748 static int parse_raid_devices(struct intel_super
*super
)
1751 struct imsm_dev
*dev_new
;
1752 size_t len
, len_migr
;
1753 size_t space_needed
= 0;
1754 struct imsm_super
*mpb
= super
->anchor
;
1756 for (i
= 0; i
< super
->anchor
->num_raid_devs
; i
++) {
1757 struct imsm_dev
*dev_iter
= __get_imsm_dev(super
->anchor
, i
);
1758 struct intel_dev
*dv
;
1760 len
= sizeof_imsm_dev(dev_iter
, 0);
1761 len_migr
= sizeof_imsm_dev(dev_iter
, 1);
1763 space_needed
+= len_migr
- len
;
1765 dv
= malloc(sizeof(*dv
));
1768 dev_new
= malloc(len_migr
);
1773 imsm_copy_dev(dev_new
, dev_iter
);
1776 dv
->next
= super
->devlist
;
1777 super
->devlist
= dv
;
1780 /* ensure that super->buf is large enough when all raid devices
1783 if (__le32_to_cpu(mpb
->mpb_size
) + space_needed
> super
->len
) {
1786 len
= ROUND_UP(__le32_to_cpu(mpb
->mpb_size
) + space_needed
, 512);
1787 if (posix_memalign(&buf
, 512, len
) != 0)
1790 memcpy(buf
, super
->buf
, super
->len
);
1791 memset(buf
+ super
->len
, 0, len
- super
->len
);
1800 /* retrieve a pointer to the bbm log which starts after all raid devices */
1801 struct bbm_log
*__get_imsm_bbm_log(struct imsm_super
*mpb
)
1805 if (__le32_to_cpu(mpb
->bbm_log_size
)) {
1807 ptr
+= mpb
->mpb_size
- __le32_to_cpu(mpb
->bbm_log_size
);
1813 static void __free_imsm(struct intel_super
*super
, int free_disks
);
1815 /* load_imsm_mpb - read matrix metadata
1816 * allocates super->mpb to be freed by free_super
1818 static int load_imsm_mpb(int fd
, struct intel_super
*super
, char *devname
)
1820 unsigned long long dsize
;
1821 unsigned long long sectors
;
1823 struct imsm_super
*anchor
;
1827 get_dev_size(fd
, NULL
, &dsize
);
1829 if (lseek64(fd
, dsize
- (512 * 2), SEEK_SET
) < 0) {
1832 Name
": Cannot seek to anchor block on %s: %s\n",
1833 devname
, strerror(errno
));
1837 if (posix_memalign((void**)&anchor
, 512, 512) != 0) {
1840 Name
": Failed to allocate imsm anchor buffer"
1841 " on %s\n", devname
);
1844 if (read(fd
, anchor
, 512) != 512) {
1847 Name
": Cannot read anchor block on %s: %s\n",
1848 devname
, strerror(errno
));
1853 if (strncmp((char *) anchor
->sig
, MPB_SIGNATURE
, MPB_SIG_LEN
) != 0) {
1856 Name
": no IMSM anchor on %s\n", devname
);
1861 __free_imsm(super
, 0);
1862 super
->len
= ROUND_UP(anchor
->mpb_size
, 512);
1863 if (posix_memalign(&super
->buf
, 512, super
->len
) != 0) {
1866 Name
": unable to allocate %zu byte mpb buffer\n",
1871 memcpy(super
->buf
, anchor
, 512);
1873 sectors
= mpb_sectors(anchor
) - 1;
1876 check_sum
= __gen_imsm_checksum(super
->anchor
);
1877 if (check_sum
!= __le32_to_cpu(super
->anchor
->check_sum
)) {
1880 Name
": IMSM checksum %x != %x on %s\n",
1882 __le32_to_cpu(super
->anchor
->check_sum
),
1887 rc
= load_imsm_disk(fd
, super
, devname
, 0);
1889 rc
= parse_raid_devices(super
);
1893 /* read the extended mpb */
1894 if (lseek64(fd
, dsize
- (512 * (2 + sectors
)), SEEK_SET
) < 0) {
1897 Name
": Cannot seek to extended mpb on %s: %s\n",
1898 devname
, strerror(errno
));
1902 if (read(fd
, super
->buf
+ 512, super
->len
- 512) != super
->len
- 512) {
1905 Name
": Cannot read extended mpb on %s: %s\n",
1906 devname
, strerror(errno
));
1910 check_sum
= __gen_imsm_checksum(super
->anchor
);
1911 if (check_sum
!= __le32_to_cpu(super
->anchor
->check_sum
)) {
1914 Name
": IMSM checksum %x != %x on %s\n",
1915 check_sum
, __le32_to_cpu(super
->anchor
->check_sum
),
1920 /* FIXME the BBM log is disk specific so we cannot use this global
1921 * buffer for all disks. Ok for now since we only look at the global
1922 * bbm_log_size parameter to gate assembly
1924 super
->bbm_log
= __get_imsm_bbm_log(super
->anchor
);
1926 rc
= load_imsm_disk(fd
, super
, devname
, 0);
1928 rc
= parse_raid_devices(super
);
1933 static void __free_imsm_disk(struct dl
*d
)
1944 static void free_imsm_disks(struct intel_super
*super
)
1948 while (super
->disks
) {
1950 super
->disks
= d
->next
;
1951 __free_imsm_disk(d
);
1953 while (super
->missing
) {
1955 super
->missing
= d
->next
;
1956 __free_imsm_disk(d
);
1961 /* free all the pieces hanging off of a super pointer */
1962 static void __free_imsm(struct intel_super
*super
, int free_disks
)
1969 free_imsm_disks(super
);
1970 free_devlist(super
);
1972 free((void *) super
->hba
);
1977 static void free_imsm(struct intel_super
*super
)
1979 __free_imsm(super
, 1);
1983 static void free_super_imsm(struct supertype
*st
)
1985 struct intel_super
*super
= st
->sb
;
1994 static struct intel_super
*alloc_super(int creating_imsm
)
1996 struct intel_super
*super
= malloc(sizeof(*super
));
1999 memset(super
, 0, sizeof(*super
));
2000 super
->creating_imsm
= creating_imsm
;
2001 super
->current_vol
= -1;
2002 super
->create_offset
= ~((__u32
) 0);
2003 if (!check_env("IMSM_NO_PLATFORM"))
2004 super
->orom
= find_imsm_orom();
2005 if (super
->orom
&& !check_env("IMSM_TEST_OROM")) {
2006 struct sys_dev
*list
, *ent
;
2008 /* find the first intel ahci controller */
2009 list
= find_driver_devices("pci", "ahci");
2010 for (ent
= list
; ent
; ent
= ent
->next
)
2011 if (devpath_to_vendor(ent
->path
) == 0x8086)
2014 super
->hba
= ent
->path
;
2017 free_sys_dev(&list
);
2025 /* find_missing - helper routine for load_super_imsm_all that identifies
2026 * disks that have disappeared from the system. This routine relies on
2027 * the mpb being uptodate, which it is at load time.
2029 static int find_missing(struct intel_super
*super
)
2032 struct imsm_super
*mpb
= super
->anchor
;
2034 struct imsm_disk
*disk
;
2036 for (i
= 0; i
< mpb
->num_disks
; i
++) {
2037 disk
= __get_imsm_disk(mpb
, i
);
2038 dl
= serial_to_dl(disk
->serial
, super
);
2042 dl
= malloc(sizeof(*dl
));
2048 dl
->devname
= strdup("missing");
2050 serialcpy(dl
->serial
, disk
->serial
);
2053 dl
->next
= super
->missing
;
2054 super
->missing
= dl
;
2060 static int load_super_imsm_all(struct supertype
*st
, int fd
, void **sbp
,
2061 char *devname
, int keep_fd
)
2064 struct intel_super
*super
;
2065 struct mdinfo
*sd
, *best
= NULL
;
2071 int devnum
= fd2devnum(fd
);
2073 enum sysfs_read_flags flags
;
2075 flags
= GET_LEVEL
|GET_VERSION
|GET_DEVS
|GET_STATE
;
2076 if (mdmon_running(devnum
))
2077 flags
|= SKIP_GONE_DEVS
;
2079 /* check if 'fd' an opened container */
2080 sra
= sysfs_read(fd
, 0, flags
);
2084 if (sra
->array
.major_version
!= -1 ||
2085 sra
->array
.minor_version
!= -2 ||
2086 strcmp(sra
->text_version
, "imsm") != 0)
2089 super
= alloc_super(0);
2093 /* find the most up to date disk in this array, skipping spares */
2094 for (sd
= sra
->devs
; sd
; sd
= sd
->next
) {
2095 sprintf(nm
, "%d:%d", sd
->disk
.major
, sd
->disk
.minor
);
2096 dfd
= dev_open(nm
, keep_fd
? O_RDWR
: O_RDONLY
);
2101 rv
= load_imsm_mpb(dfd
, super
, NULL
);
2103 /* retry the load if we might have raced against mdmon */
2104 if (rv
== 3 && mdmon_running(devnum
))
2105 for (retry
= 0; retry
< 3; retry
++) {
2107 rv
= load_imsm_mpb(dfd
, super
, NULL
);
2114 if (super
->anchor
->num_raid_devs
== 0)
2117 gen
= __le32_to_cpu(super
->anchor
->generation_num
);
2118 if (!best
|| gen
> bestgen
) {
2133 /* load the most up to date anchor */
2134 sprintf(nm
, "%d:%d", best
->disk
.major
, best
->disk
.minor
);
2135 dfd
= dev_open(nm
, O_RDONLY
);
2140 rv
= load_imsm_mpb(dfd
, super
, NULL
);
2147 /* re-parse the disk list with the current anchor */
2148 for (sd
= sra
->devs
; sd
; sd
= sd
->next
) {
2149 sprintf(nm
, "%d:%d", sd
->disk
.major
, sd
->disk
.minor
);
2150 dfd
= dev_open(nm
, keep_fd
? O_RDWR
: O_RDONLY
);
2155 load_imsm_disk(dfd
, super
, NULL
, keep_fd
);
2161 if (find_missing(super
) != 0) {
2166 if (st
->subarray
[0]) {
2167 if (atoi(st
->subarray
) <= super
->anchor
->num_raid_devs
)
2168 super
->current_vol
= atoi(st
->subarray
);
2174 st
->container_dev
= devnum
;
2175 if (st
->ss
== NULL
) {
2176 st
->ss
= &super_imsm
;
2177 st
->minor_version
= 0;
2178 st
->max_devs
= IMSM_MAX_DEVICES
;
2180 st
->loaded_container
= 1;
2186 static int load_super_imsm(struct supertype
*st
, int fd
, char *devname
)
2188 struct intel_super
*super
;
2192 if (load_super_imsm_all(st
, fd
, &st
->sb
, devname
, 1) == 0)
2195 if (st
->subarray
[0])
2196 return 1; /* FIXME */
2198 super
= alloc_super(0);
2201 Name
": malloc of %zu failed.\n",
2206 rv
= load_imsm_mpb(fd
, super
, devname
);
2211 Name
": Failed to load all information "
2212 "sections on %s\n", devname
);
2218 if (st
->ss
== NULL
) {
2219 st
->ss
= &super_imsm
;
2220 st
->minor_version
= 0;
2221 st
->max_devs
= IMSM_MAX_DEVICES
;
2223 st
->loaded_container
= 0;
2228 static __u16
info_to_blocks_per_strip(mdu_array_info_t
*info
)
2230 if (info
->level
== 1)
2232 return info
->chunk_size
>> 9;
2235 static __u32
info_to_num_data_stripes(mdu_array_info_t
*info
, int num_domains
)
2239 num_stripes
= (info
->size
* 2) / info_to_blocks_per_strip(info
);
2240 num_stripes
/= num_domains
;
2245 static __u32
info_to_blocks_per_member(mdu_array_info_t
*info
)
2247 if (info
->level
== 1)
2248 return info
->size
* 2;
2250 return (info
->size
* 2) & ~(info_to_blocks_per_strip(info
) - 1);
2253 static void imsm_update_version_info(struct intel_super
*super
)
2255 /* update the version and attributes */
2256 struct imsm_super
*mpb
= super
->anchor
;
2258 struct imsm_dev
*dev
;
2259 struct imsm_map
*map
;
2262 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
2263 dev
= get_imsm_dev(super
, i
);
2264 map
= get_imsm_map(dev
, 0);
2265 if (__le32_to_cpu(dev
->size_high
) > 0)
2266 mpb
->attributes
|= MPB_ATTRIB_2TB
;
2268 /* FIXME detect when an array spans a port multiplier */
2270 mpb
->attributes
|= MPB_ATTRIB_PM
;
2273 if (mpb
->num_raid_devs
> 1 ||
2274 mpb
->attributes
!= MPB_ATTRIB_CHECKSUM_VERIFY
) {
2275 version
= MPB_VERSION_ATTRIBS
;
2276 switch (get_imsm_raid_level(map
)) {
2277 case 0: mpb
->attributes
|= MPB_ATTRIB_RAID0
; break;
2278 case 1: mpb
->attributes
|= MPB_ATTRIB_RAID1
; break;
2279 case 10: mpb
->attributes
|= MPB_ATTRIB_RAID10
; break;
2280 case 5: mpb
->attributes
|= MPB_ATTRIB_RAID5
; break;
2283 if (map
->num_members
>= 5)
2284 version
= MPB_VERSION_5OR6_DISK_ARRAY
;
2285 else if (dev
->status
== DEV_CLONE_N_GO
)
2286 version
= MPB_VERSION_CNG
;
2287 else if (get_imsm_raid_level(map
) == 5)
2288 version
= MPB_VERSION_RAID5
;
2289 else if (map
->num_members
>= 3)
2290 version
= MPB_VERSION_3OR4_DISK_ARRAY
;
2291 else if (get_imsm_raid_level(map
) == 1)
2292 version
= MPB_VERSION_RAID1
;
2294 version
= MPB_VERSION_RAID0
;
2296 strcpy(((char *) mpb
->sig
) + strlen(MPB_SIGNATURE
), version
);
2300 static int init_super_imsm_volume(struct supertype
*st
, mdu_array_info_t
*info
,
2301 unsigned long long size
, char *name
,
2302 char *homehost
, int *uuid
)
2304 /* We are creating a volume inside a pre-existing container.
2305 * so st->sb is already set.
2307 struct intel_super
*super
= st
->sb
;
2308 struct imsm_super
*mpb
= super
->anchor
;
2309 struct intel_dev
*dv
;
2310 struct imsm_dev
*dev
;
2311 struct imsm_vol
*vol
;
2312 struct imsm_map
*map
;
2313 int idx
= mpb
->num_raid_devs
;
2315 unsigned long long array_blocks
;
2316 size_t size_old
, size_new
;
2317 __u32 num_data_stripes
;
2319 if (super
->orom
&& mpb
->num_raid_devs
>= super
->orom
->vpa
) {
2320 fprintf(stderr
, Name
": This imsm-container already has the "
2321 "maximum of %d volumes\n", super
->orom
->vpa
);
2325 /* ensure the mpb is large enough for the new data */
2326 size_old
= __le32_to_cpu(mpb
->mpb_size
);
2327 size_new
= disks_to_mpb_size(info
->nr_disks
);
2328 if (size_new
> size_old
) {
2330 size_t size_round
= ROUND_UP(size_new
, 512);
2332 if (posix_memalign(&mpb_new
, 512, size_round
) != 0) {
2333 fprintf(stderr
, Name
": could not allocate new mpb\n");
2336 memcpy(mpb_new
, mpb
, size_old
);
2339 super
->anchor
= mpb_new
;
2340 mpb
->mpb_size
= __cpu_to_le32(size_new
);
2341 memset(mpb_new
+ size_old
, 0, size_round
- size_old
);
2343 super
->current_vol
= idx
;
2344 /* when creating the first raid device in this container set num_disks
2345 * to zero, i.e. delete this spare and add raid member devices in
2346 * add_to_super_imsm_volume()
2348 if (super
->current_vol
== 0)
2351 for (i
= 0; i
< super
->current_vol
; i
++) {
2352 dev
= get_imsm_dev(super
, i
);
2353 if (strncmp((char *) dev
->volume
, name
,
2354 MAX_RAID_SERIAL_LEN
) == 0) {
2355 fprintf(stderr
, Name
": '%s' is already defined for this container\n",
2361 sprintf(st
->subarray
, "%d", idx
);
2362 dv
= malloc(sizeof(*dv
));
2364 fprintf(stderr
, Name
": failed to allocate device list entry\n");
2367 dev
= malloc(sizeof(*dev
) + sizeof(__u32
) * (info
->raid_disks
- 1));
2370 fprintf(stderr
, Name
": could not allocate raid device\n");
2373 strncpy((char *) dev
->volume
, name
, MAX_RAID_SERIAL_LEN
);
2374 if (info
->level
== 1)
2375 array_blocks
= info_to_blocks_per_member(info
);
2377 array_blocks
= calc_array_size(info
->level
, info
->raid_disks
,
2378 info
->layout
, info
->chunk_size
,
2380 dev
->size_low
= __cpu_to_le32((__u32
) array_blocks
);
2381 dev
->size_high
= __cpu_to_le32((__u32
) (array_blocks
>> 32));
2382 dev
->status
= __cpu_to_le32(0);
2383 dev
->reserved_blocks
= __cpu_to_le32(0);
2385 vol
->migr_state
= 0;
2386 set_migr_type(dev
, MIGR_INIT
);
2388 vol
->curr_migr_unit
= 0;
2389 map
= get_imsm_map(dev
, 0);
2390 map
->pba_of_lba0
= __cpu_to_le32(super
->create_offset
);
2391 map
->blocks_per_member
= __cpu_to_le32(info_to_blocks_per_member(info
));
2392 map
->blocks_per_strip
= __cpu_to_le16(info_to_blocks_per_strip(info
));
2393 map
->failed_disk_num
= ~0;
2394 map
->map_state
= info
->level
? IMSM_T_STATE_UNINITIALIZED
:
2395 IMSM_T_STATE_NORMAL
;
2397 if (info
->level
== 1 && info
->raid_disks
> 2) {
2398 fprintf(stderr
, Name
": imsm does not support more than 2 disks"
2399 "in a raid1 volume\n");
2402 if (info
->level
== 10) {
2403 map
->raid_level
= 1;
2404 map
->num_domains
= info
->raid_disks
/ 2;
2406 map
->raid_level
= info
->level
;
2407 map
->num_domains
= 1;
2409 num_data_stripes
= info_to_num_data_stripes(info
, map
->num_domains
);
2410 map
->num_data_stripes
= __cpu_to_le32(num_data_stripes
);
2412 map
->num_members
= info
->raid_disks
;
2413 for (i
= 0; i
< map
->num_members
; i
++) {
2414 /* initialized in add_to_super */
2415 set_imsm_ord_tbl_ent(map
, i
, 0);
2417 mpb
->num_raid_devs
++;
2420 dv
->index
= super
->current_vol
;
2421 dv
->next
= super
->devlist
;
2422 super
->devlist
= dv
;
2424 imsm_update_version_info(super
);
2429 static int init_super_imsm(struct supertype
*st
, mdu_array_info_t
*info
,
2430 unsigned long long size
, char *name
,
2431 char *homehost
, int *uuid
)
2433 /* This is primarily called by Create when creating a new array.
2434 * We will then get add_to_super called for each component, and then
2435 * write_init_super called to write it out to each device.
2436 * For IMSM, Create can create on fresh devices or on a pre-existing
2438 * To create on a pre-existing array a different method will be called.
2439 * This one is just for fresh drives.
2441 struct intel_super
*super
;
2442 struct imsm_super
*mpb
;
2451 return init_super_imsm_volume(st
, info
, size
, name
, homehost
,
2454 super
= alloc_super(1);
2457 mpb_size
= disks_to_mpb_size(info
->nr_disks
);
2458 if (posix_memalign(&super
->buf
, 512, mpb_size
) != 0) {
2463 memset(mpb
, 0, mpb_size
);
2465 mpb
->attributes
= MPB_ATTRIB_CHECKSUM_VERIFY
;
2467 version
= (char *) mpb
->sig
;
2468 strcpy(version
, MPB_SIGNATURE
);
2469 version
+= strlen(MPB_SIGNATURE
);
2470 strcpy(version
, MPB_VERSION_RAID0
);
2471 mpb
->mpb_size
= mpb_size
;
2478 static int add_to_super_imsm_volume(struct supertype
*st
, mdu_disk_info_t
*dk
,
2479 int fd
, char *devname
)
2481 struct intel_super
*super
= st
->sb
;
2482 struct imsm_super
*mpb
= super
->anchor
;
2484 struct imsm_dev
*dev
;
2485 struct imsm_map
*map
;
2487 dev
= get_imsm_dev(super
, super
->current_vol
);
2488 map
= get_imsm_map(dev
, 0);
2490 if (! (dk
->state
& (1<<MD_DISK_SYNC
))) {
2491 fprintf(stderr
, Name
": %s: Cannot add spare devices to IMSM volume\n",
2497 /* we're doing autolayout so grab the pre-marked (in
2498 * validate_geometry) raid_disk
2500 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
2501 if (dl
->raiddisk
== dk
->raid_disk
)
2504 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
2505 if (dl
->major
== dk
->major
&&
2506 dl
->minor
== dk
->minor
)
2511 fprintf(stderr
, Name
": %s is not a member of the same container\n", devname
);
2515 /* add a pristine spare to the metadata */
2516 if (dl
->index
< 0) {
2517 dl
->index
= super
->anchor
->num_disks
;
2518 super
->anchor
->num_disks
++;
2520 set_imsm_ord_tbl_ent(map
, dk
->number
, dl
->index
);
2521 dl
->disk
.status
= CONFIGURED_DISK
| USABLE_DISK
;
2523 /* if we are creating the first raid device update the family number */
2524 if (super
->current_vol
== 0) {
2526 struct imsm_dev
*_dev
= __get_imsm_dev(mpb
, 0);
2527 struct imsm_disk
*_disk
= __get_imsm_disk(mpb
, dl
->index
);
2531 sum
= __gen_imsm_checksum(mpb
);
2532 mpb
->family_num
= __cpu_to_le32(sum
);
2538 static int add_to_super_imsm(struct supertype
*st
, mdu_disk_info_t
*dk
,
2539 int fd
, char *devname
)
2541 struct intel_super
*super
= st
->sb
;
2543 unsigned long long size
;
2548 /* if we are on an RAID enabled platform check that the disk is
2549 * attached to the raid controller
2551 if (super
->hba
&& !disk_attached_to_hba(fd
, super
->hba
)) {
2553 Name
": %s is not attached to the raid controller: %s\n",
2554 devname
? : "disk", super
->hba
);
2558 if (super
->current_vol
>= 0)
2559 return add_to_super_imsm_volume(st
, dk
, fd
, devname
);
2562 dd
= malloc(sizeof(*dd
));
2565 Name
": malloc failed %s:%d.\n", __func__
, __LINE__
);
2568 memset(dd
, 0, sizeof(*dd
));
2569 dd
->major
= major(stb
.st_rdev
);
2570 dd
->minor
= minor(stb
.st_rdev
);
2572 dd
->devname
= devname
? strdup(devname
) : NULL
;
2575 rv
= imsm_read_serial(fd
, devname
, dd
->serial
);
2578 Name
": failed to retrieve scsi serial, aborting\n");
2583 get_dev_size(fd
, NULL
, &size
);
2585 serialcpy(dd
->disk
.serial
, dd
->serial
);
2586 dd
->disk
.total_blocks
= __cpu_to_le32(size
);
2587 dd
->disk
.status
= USABLE_DISK
| SPARE_DISK
;
2588 if (sysfs_disk_to_scsi_id(fd
, &id
) == 0)
2589 dd
->disk
.scsi_id
= __cpu_to_le32(id
);
2591 dd
->disk
.scsi_id
= __cpu_to_le32(0);
2593 if (st
->update_tail
) {
2594 dd
->next
= super
->add
;
2597 dd
->next
= super
->disks
;
2604 static int store_imsm_mpb(int fd
, struct intel_super
*super
);
2606 /* spare records have their own family number and do not have any defined raid
2609 static int write_super_imsm_spares(struct intel_super
*super
, int doclose
)
2611 struct imsm_super mpb_save
;
2612 struct imsm_super
*mpb
= super
->anchor
;
2617 mpb
->num_raid_devs
= 0;
2619 mpb
->mpb_size
= sizeof(struct imsm_super
);
2620 mpb
->generation_num
= __cpu_to_le32(1UL);
2622 for (d
= super
->disks
; d
; d
= d
->next
) {
2626 mpb
->disk
[0] = d
->disk
;
2627 sum
= __gen_imsm_checksum(mpb
);
2628 mpb
->family_num
= __cpu_to_le32(sum
);
2629 sum
= __gen_imsm_checksum(mpb
);
2630 mpb
->check_sum
= __cpu_to_le32(sum
);
2632 if (store_imsm_mpb(d
->fd
, super
)) {
2633 fprintf(stderr
, "%s: failed for device %d:%d %s\n",
2634 __func__
, d
->major
, d
->minor
, strerror(errno
));
2648 static int write_super_imsm(struct intel_super
*super
, int doclose
)
2650 struct imsm_super
*mpb
= super
->anchor
;
2656 __u32 mpb_size
= sizeof(struct imsm_super
) - sizeof(struct imsm_disk
);
2658 /* 'generation' is incremented everytime the metadata is written */
2659 generation
= __le32_to_cpu(mpb
->generation_num
);
2661 mpb
->generation_num
= __cpu_to_le32(generation
);
2663 mpb_size
+= sizeof(struct imsm_disk
) * mpb
->num_disks
;
2664 for (d
= super
->disks
; d
; d
= d
->next
) {
2668 mpb
->disk
[d
->index
] = d
->disk
;
2670 for (d
= super
->missing
; d
; d
= d
->next
)
2671 mpb
->disk
[d
->index
] = d
->disk
;
2673 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
2674 struct imsm_dev
*dev
= __get_imsm_dev(mpb
, i
);
2676 imsm_copy_dev(dev
, get_imsm_dev(super
, i
));
2677 mpb_size
+= sizeof_imsm_dev(dev
, 0);
2679 mpb_size
+= __le32_to_cpu(mpb
->bbm_log_size
);
2680 mpb
->mpb_size
= __cpu_to_le32(mpb_size
);
2682 /* recalculate checksum */
2683 sum
= __gen_imsm_checksum(mpb
);
2684 mpb
->check_sum
= __cpu_to_le32(sum
);
2686 /* write the mpb for disks that compose raid devices */
2687 for (d
= super
->disks
; d
; d
= d
->next
) {
2690 if (store_imsm_mpb(d
->fd
, super
))
2691 fprintf(stderr
, "%s: failed for device %d:%d %s\n",
2692 __func__
, d
->major
, d
->minor
, strerror(errno
));
2700 return write_super_imsm_spares(super
, doclose
);
2706 static int create_array(struct supertype
*st
)
2709 struct imsm_update_create_array
*u
;
2710 struct intel_super
*super
= st
->sb
;
2711 struct imsm_dev
*dev
= get_imsm_dev(super
, super
->current_vol
);
2712 struct imsm_map
*map
= get_imsm_map(dev
, 0);
2713 struct disk_info
*inf
;
2714 struct imsm_disk
*disk
;
2718 len
= sizeof(*u
) - sizeof(*dev
) + sizeof_imsm_dev(dev
, 0) +
2719 sizeof(*inf
) * map
->num_members
;
2722 fprintf(stderr
, "%s: failed to allocate update buffer\n",
2727 u
->type
= update_create_array
;
2728 u
->dev_idx
= super
->current_vol
;
2729 imsm_copy_dev(&u
->dev
, dev
);
2730 inf
= get_disk_info(u
);
2731 for (i
= 0; i
< map
->num_members
; i
++) {
2732 idx
= get_imsm_disk_idx(dev
, i
);
2733 disk
= get_imsm_disk(super
, idx
);
2734 serialcpy(inf
[i
].serial
, disk
->serial
);
2736 append_metadata_update(st
, u
, len
);
2741 static int _add_disk(struct supertype
*st
)
2743 struct intel_super
*super
= st
->sb
;
2745 struct imsm_update_add_disk
*u
;
2753 fprintf(stderr
, "%s: failed to allocate update buffer\n",
2758 u
->type
= update_add_disk
;
2759 append_metadata_update(st
, u
, len
);
2764 static int write_init_super_imsm(struct supertype
*st
)
2766 if (st
->update_tail
) {
2767 /* queue the recently created array / added disk
2768 * as a metadata update */
2769 struct intel_super
*super
= st
->sb
;
2773 /* determine if we are creating a volume or adding a disk */
2774 if (super
->current_vol
< 0) {
2775 /* in the add disk case we are running in mdmon
2776 * context, so don't close fd's
2778 return _add_disk(st
);
2780 rv
= create_array(st
);
2782 for (d
= super
->disks
; d
; d
= d
->next
) {
2789 return write_super_imsm(st
->sb
, 1);
2793 static int store_zero_imsm(struct supertype
*st
, int fd
)
2795 unsigned long long dsize
;
2798 get_dev_size(fd
, NULL
, &dsize
);
2800 /* first block is stored on second to last sector of the disk */
2801 if (lseek64(fd
, dsize
- (512 * 2), SEEK_SET
) < 0)
2804 if (posix_memalign(&buf
, 512, 512) != 0)
2807 memset(buf
, 0, 512);
2808 if (write(fd
, buf
, 512) != 512)
2813 static int imsm_bbm_log_size(struct imsm_super
*mpb
)
2815 return __le32_to_cpu(mpb
->bbm_log_size
);
2819 static int validate_geometry_imsm_container(struct supertype
*st
, int level
,
2820 int layout
, int raiddisks
, int chunk
,
2821 unsigned long long size
, char *dev
,
2822 unsigned long long *freesize
,
2826 unsigned long long ldsize
;
2827 const struct imsm_orom
*orom
;
2829 if (level
!= LEVEL_CONTAINER
)
2834 if (check_env("IMSM_NO_PLATFORM"))
2837 orom
= find_imsm_orom();
2838 if (orom
&& raiddisks
> orom
->tds
) {
2840 fprintf(stderr
, Name
": %d exceeds maximum number of"
2841 " platform supported disks: %d\n",
2842 raiddisks
, orom
->tds
);
2846 fd
= open(dev
, O_RDONLY
|O_EXCL
, 0);
2849 fprintf(stderr
, Name
": imsm: Cannot open %s: %s\n",
2850 dev
, strerror(errno
));
2853 if (!get_dev_size(fd
, dev
, &ldsize
)) {
2859 *freesize
= avail_size_imsm(st
, ldsize
>> 9);
2864 static unsigned long long find_size(struct extent
*e
, int *idx
, int num_extents
)
2866 const unsigned long long base_start
= e
[*idx
].start
;
2867 unsigned long long end
= base_start
+ e
[*idx
].size
;
2870 if (base_start
== end
)
2874 for (i
= *idx
; i
< num_extents
; i
++) {
2875 /* extend overlapping extents */
2876 if (e
[i
].start
>= base_start
&&
2877 e
[i
].start
<= end
) {
2880 if (e
[i
].start
+ e
[i
].size
> end
)
2881 end
= e
[i
].start
+ e
[i
].size
;
2882 } else if (e
[i
].start
> end
) {
2888 return end
- base_start
;
2891 static unsigned long long merge_extents(struct intel_super
*super
, int sum_extents
)
2893 /* build a composite disk with all known extents and generate a new
2894 * 'maxsize' given the "all disks in an array must share a common start
2895 * offset" constraint
2897 struct extent
*e
= calloc(sum_extents
, sizeof(*e
));
2901 unsigned long long pos
;
2902 unsigned long long start
= 0;
2903 unsigned long long maxsize
;
2904 unsigned long reserve
;
2907 return ~0ULL; /* error */
2909 /* coalesce and sort all extents. also, check to see if we need to
2910 * reserve space between member arrays
2913 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
2916 for (i
= 0; i
< dl
->extent_cnt
; i
++)
2919 qsort(e
, sum_extents
, sizeof(*e
), cmp_extent
);
2924 while (i
< sum_extents
) {
2925 e
[j
].start
= e
[i
].start
;
2926 e
[j
].size
= find_size(e
, &i
, sum_extents
);
2928 if (e
[j
-1].size
== 0)
2937 unsigned long long esize
;
2939 esize
= e
[i
].start
- pos
;
2940 if (esize
>= maxsize
) {
2945 pos
= e
[i
].start
+ e
[i
].size
;
2947 } while (e
[i
-1].size
);
2950 if (start_extent
> 0)
2951 reserve
= IMSM_RESERVED_SECTORS
; /* gap between raid regions */
2955 if (maxsize
< reserve
)
2958 super
->create_offset
= ~((__u32
) 0);
2959 if (start
+ reserve
> super
->create_offset
)
2960 return ~0ULL; /* start overflows create_offset */
2961 super
->create_offset
= start
+ reserve
;
2963 return maxsize
- reserve
;
2966 static int is_raid_level_supported(const struct imsm_orom
*orom
, int level
, int raiddisks
)
2968 if (level
< 0 || level
== 6 || level
== 4)
2971 /* if we have an orom prevent invalid raid levels */
2974 case 0: return imsm_orom_has_raid0(orom
);
2977 return imsm_orom_has_raid1e(orom
);
2978 return imsm_orom_has_raid1(orom
) && raiddisks
== 2;
2979 case 10: return imsm_orom_has_raid10(orom
) && raiddisks
== 4;
2980 case 5: return imsm_orom_has_raid5(orom
) && raiddisks
> 2;
2983 return 1; /* not on an Intel RAID platform so anything goes */
2988 #define pr_vrb(fmt, arg...) (void) (verbose && fprintf(stderr, Name fmt, ##arg))
2989 /* validate_geometry_imsm_volume - lifted from validate_geometry_ddf_bvd
2990 * FIX ME add ahci details
2992 static int validate_geometry_imsm_volume(struct supertype
*st
, int level
,
2993 int layout
, int raiddisks
, int chunk
,
2994 unsigned long long size
, char *dev
,
2995 unsigned long long *freesize
,
2999 struct intel_super
*super
= st
->sb
;
3000 struct imsm_super
*mpb
= super
->anchor
;
3002 unsigned long long pos
= 0;
3003 unsigned long long maxsize
;
3007 /* We must have the container info already read in. */
3011 if (!is_raid_level_supported(super
->orom
, level
, raiddisks
)) {
3012 pr_vrb(": platform does not support raid%d with %d disk%s\n",
3013 level
, raiddisks
, raiddisks
> 1 ? "s" : "");
3016 if (super
->orom
&& level
!= 1 &&
3017 !imsm_orom_has_chunk(super
->orom
, chunk
)) {
3018 pr_vrb(": platform does not support a chunk size of: %d\n", chunk
);
3021 if (layout
!= imsm_level_to_layout(level
)) {
3023 pr_vrb(": imsm raid 5 only supports the left-asymmetric layout\n");
3024 else if (level
== 10)
3025 pr_vrb(": imsm raid 10 only supports the n2 layout\n");
3027 pr_vrb(": imsm unknown layout %#x for this raid level %d\n",
3033 /* General test: make sure there is space for
3034 * 'raiddisks' device extents of size 'size' at a given
3037 unsigned long long minsize
= size
;
3038 unsigned long long start_offset
= ~0ULL;
3041 minsize
= MPB_SECTOR_CNT
+ IMSM_RESERVED_SECTORS
;
3042 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
3047 e
= get_extents(super
, dl
);
3050 unsigned long long esize
;
3051 esize
= e
[i
].start
- pos
;
3052 if (esize
>= minsize
)
3054 if (found
&& start_offset
== ~0ULL) {
3057 } else if (found
&& pos
!= start_offset
) {
3061 pos
= e
[i
].start
+ e
[i
].size
;
3063 } while (e
[i
-1].size
);
3068 if (dcnt
< raiddisks
) {
3070 fprintf(stderr
, Name
": imsm: Not enough "
3071 "devices with space for this array "
3079 /* This device must be a member of the set */
3080 if (stat(dev
, &stb
) < 0)
3082 if ((S_IFMT
& stb
.st_mode
) != S_IFBLK
)
3084 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
3085 if (dl
->major
== major(stb
.st_rdev
) &&
3086 dl
->minor
== minor(stb
.st_rdev
))
3091 fprintf(stderr
, Name
": %s is not in the "
3092 "same imsm set\n", dev
);
3094 } else if (super
->orom
&& dl
->index
< 0 && mpb
->num_raid_devs
) {
3095 /* If a volume is present then the current creation attempt
3096 * cannot incorporate new spares because the orom may not
3097 * understand this configuration (all member disks must be
3098 * members of each array in the container).
3100 fprintf(stderr
, Name
": %s is a spare and a volume"
3101 " is already defined for this container\n", dev
);
3102 fprintf(stderr
, Name
": The option-rom requires all member"
3103 " disks to be a member of all volumes\n");
3107 /* retrieve the largest free space block */
3108 e
= get_extents(super
, dl
);
3113 unsigned long long esize
;
3115 esize
= e
[i
].start
- pos
;
3116 if (esize
>= maxsize
)
3118 pos
= e
[i
].start
+ e
[i
].size
;
3120 } while (e
[i
-1].size
);
3125 fprintf(stderr
, Name
": unable to determine free space for: %s\n",
3129 if (maxsize
< size
) {
3131 fprintf(stderr
, Name
": %s not enough space (%llu < %llu)\n",
3132 dev
, maxsize
, size
);
3136 /* count total number of extents for merge */
3138 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
3140 i
+= dl
->extent_cnt
;
3142 maxsize
= merge_extents(super
, i
);
3143 if (maxsize
< size
) {
3145 fprintf(stderr
, Name
": not enough space after merge (%llu < %llu)\n",
3148 } else if (maxsize
== ~0ULL) {
3150 fprintf(stderr
, Name
": failed to merge %d extents\n", i
);
3154 *freesize
= maxsize
;
3159 static int reserve_space(struct supertype
*st
, int raiddisks
,
3160 unsigned long long size
, int chunk
,
3161 unsigned long long *freesize
)
3163 struct intel_super
*super
= st
->sb
;
3164 struct imsm_super
*mpb
= super
->anchor
;
3169 unsigned long long maxsize
;
3170 unsigned long long minsize
;
3174 /* find the largest common start free region of the possible disks */
3178 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
3184 /* don't activate new spares if we are orom constrained
3185 * and there is already a volume active in the container
3187 if (super
->orom
&& dl
->index
< 0 && mpb
->num_raid_devs
)
3190 e
= get_extents(super
, dl
);
3193 for (i
= 1; e
[i
-1].size
; i
++)
3201 maxsize
= merge_extents(super
, extent_cnt
);
3206 if (cnt
< raiddisks
||
3207 (super
->orom
&& used
&& used
!= raiddisks
) ||
3208 maxsize
< minsize
) {
3209 fprintf(stderr
, Name
": not enough devices with space to create array.\n");
3210 return 0; /* No enough free spaces large enough */
3222 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
3224 dl
->raiddisk
= cnt
++;
3231 static int validate_geometry_imsm(struct supertype
*st
, int level
, int layout
,
3232 int raiddisks
, int chunk
, unsigned long long size
,
3233 char *dev
, unsigned long long *freesize
,
3239 /* if given unused devices create a container
3240 * if given given devices in a container create a member volume
3242 if (level
== LEVEL_CONTAINER
) {
3243 /* Must be a fresh device to add to a container */
3244 return validate_geometry_imsm_container(st
, level
, layout
,
3245 raiddisks
, chunk
, size
,
3251 if (st
->sb
&& freesize
) {
3252 /* we are being asked to automatically layout a
3253 * new volume based on the current contents of
3254 * the container. If the the parameters can be
3255 * satisfied reserve_space will record the disks,
3256 * start offset, and size of the volume to be
3257 * created. add_to_super and getinfo_super
3258 * detect when autolayout is in progress.
3260 return reserve_space(st
, raiddisks
, size
, chunk
, freesize
);
3265 /* creating in a given container */
3266 return validate_geometry_imsm_volume(st
, level
, layout
,
3267 raiddisks
, chunk
, size
,
3268 dev
, freesize
, verbose
);
3271 /* limit creation to the following levels */
3283 /* This device needs to be a device in an 'imsm' container */
3284 fd
= open(dev
, O_RDONLY
|O_EXCL
, 0);
3288 Name
": Cannot create this array on device %s\n",
3293 if (errno
!= EBUSY
|| (fd
= open(dev
, O_RDONLY
, 0)) < 0) {
3295 fprintf(stderr
, Name
": Cannot open %s: %s\n",
3296 dev
, strerror(errno
));
3299 /* Well, it is in use by someone, maybe an 'imsm' container. */
3300 cfd
= open_container(fd
);
3304 fprintf(stderr
, Name
": Cannot use %s: It is busy\n",
3308 sra
= sysfs_read(cfd
, 0, GET_VERSION
);
3310 if (sra
&& sra
->array
.major_version
== -1 &&
3311 strcmp(sra
->text_version
, "imsm") == 0) {
3312 /* This is a member of a imsm container. Load the container
3313 * and try to create a volume
3315 struct intel_super
*super
;
3317 if (load_super_imsm_all(st
, cfd
, (void **) &super
, NULL
, 1) == 0) {
3319 st
->container_dev
= fd2devnum(cfd
);
3321 return validate_geometry_imsm_volume(st
, level
, layout
,
3327 } else /* may belong to another container */
3332 #endif /* MDASSEMBLE */
3334 static struct mdinfo
*container_content_imsm(struct supertype
*st
)
3336 /* Given a container loaded by load_super_imsm_all,
3337 * extract information about all the arrays into
3340 * For each imsm_dev create an mdinfo, fill it in,
3341 * then look for matching devices in super->disks
3342 * and create appropriate device mdinfo.
3344 struct intel_super
*super
= st
->sb
;
3345 struct imsm_super
*mpb
= super
->anchor
;
3346 struct mdinfo
*rest
= NULL
;
3349 /* do not assemble arrays that might have bad blocks */
3350 if (imsm_bbm_log_size(super
->anchor
)) {
3351 fprintf(stderr
, Name
": BBM log found in metadata. "
3352 "Cannot activate array(s).\n");
3356 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
3357 struct imsm_dev
*dev
= get_imsm_dev(super
, i
);
3358 struct imsm_map
*map
= get_imsm_map(dev
, 0);
3359 struct mdinfo
*this;
3362 /* do not publish arrays that are in the middle of an
3363 * unsupported migration
3365 if (dev
->vol
.migr_state
&&
3366 (migr_type(dev
) == MIGR_GEN_MIGR
||
3367 migr_type(dev
) == MIGR_STATE_CHANGE
)) {
3368 fprintf(stderr
, Name
": cannot assemble volume '%.16s':"
3369 " unsupported migration in progress\n",
3374 this = malloc(sizeof(*this));
3375 memset(this, 0, sizeof(*this));
3378 super
->current_vol
= i
;
3379 getinfo_super_imsm_volume(st
, this);
3380 for (slot
= 0 ; slot
< map
->num_members
; slot
++) {
3381 struct mdinfo
*info_d
;
3389 idx
= get_imsm_disk_idx(dev
, slot
);
3390 ord
= get_imsm_ord_tbl_ent(dev
, slot
);
3391 for (d
= super
->disks
; d
; d
= d
->next
)
3392 if (d
->index
== idx
)
3398 s
= d
? d
->disk
.status
: 0;
3399 if (s
& FAILED_DISK
)
3401 if (!(s
& USABLE_DISK
))
3403 if (ord
& IMSM_ORD_REBUILD
)
3407 * if we skip some disks the array will be assmebled degraded;
3408 * reset resync start to avoid a dirty-degraded situation
3410 * FIXME handle dirty degraded
3412 if (skip
&& !dev
->vol
.dirty
)
3413 this->resync_start
= ~0ULL;
3417 info_d
= malloc(sizeof(*info_d
));
3419 fprintf(stderr
, Name
": failed to allocate disk"
3420 " for volume %.16s\n", dev
->volume
);
3425 memset(info_d
, 0, sizeof(*info_d
));
3426 info_d
->next
= this->devs
;
3427 this->devs
= info_d
;
3429 info_d
->disk
.number
= d
->index
;
3430 info_d
->disk
.major
= d
->major
;
3431 info_d
->disk
.minor
= d
->minor
;
3432 info_d
->disk
.raid_disk
= slot
;
3434 this->array
.working_disks
++;
3436 info_d
->events
= __le32_to_cpu(mpb
->generation_num
);
3437 info_d
->data_offset
= __le32_to_cpu(map
->pba_of_lba0
);
3438 info_d
->component_size
= __le32_to_cpu(map
->blocks_per_member
);
3440 strcpy(info_d
->name
, d
->devname
);
3450 static int imsm_open_new(struct supertype
*c
, struct active_array
*a
,
3453 struct intel_super
*super
= c
->sb
;
3454 struct imsm_super
*mpb
= super
->anchor
;
3456 if (atoi(inst
) >= mpb
->num_raid_devs
) {
3457 fprintf(stderr
, "%s: subarry index %d, out of range\n",
3458 __func__
, atoi(inst
));
3462 dprintf("imsm: open_new %s\n", inst
);
3463 a
->info
.container_member
= atoi(inst
);
3467 static __u8
imsm_check_degraded(struct intel_super
*super
, struct imsm_dev
*dev
, int failed
)
3469 struct imsm_map
*map
= get_imsm_map(dev
, 0);
3472 return map
->map_state
== IMSM_T_STATE_UNINITIALIZED
?
3473 IMSM_T_STATE_UNINITIALIZED
: IMSM_T_STATE_NORMAL
;
3475 switch (get_imsm_raid_level(map
)) {
3477 return IMSM_T_STATE_FAILED
;
3480 if (failed
< map
->num_members
)
3481 return IMSM_T_STATE_DEGRADED
;
3483 return IMSM_T_STATE_FAILED
;
3488 * check to see if any mirrors have failed, otherwise we
3489 * are degraded. Even numbered slots are mirrored on
3493 /* gcc -Os complains that this is unused */
3494 int insync
= insync
;
3496 for (i
= 0; i
< map
->num_members
; i
++) {
3497 __u32 ord
= get_imsm_ord_tbl_ent(dev
, i
);
3498 int idx
= ord_to_idx(ord
);
3499 struct imsm_disk
*disk
;
3501 /* reset the potential in-sync count on even-numbered
3502 * slots. num_copies is always 2 for imsm raid10
3507 disk
= get_imsm_disk(super
, idx
);
3508 if (!disk
|| disk
->status
& FAILED_DISK
||
3509 ord
& IMSM_ORD_REBUILD
)
3512 /* no in-sync disks left in this mirror the
3516 return IMSM_T_STATE_FAILED
;
3519 return IMSM_T_STATE_DEGRADED
;
3523 return IMSM_T_STATE_DEGRADED
;
3525 return IMSM_T_STATE_FAILED
;
3531 return map
->map_state
;
3534 static int imsm_count_failed(struct intel_super
*super
, struct imsm_dev
*dev
)
3538 struct imsm_disk
*disk
;
3539 struct imsm_map
*map
= get_imsm_map(dev
, 0);
3540 struct imsm_map
*prev
= get_imsm_map(dev
, dev
->vol
.migr_state
);
3544 /* at the beginning of migration we set IMSM_ORD_REBUILD on
3545 * disks that are being rebuilt. New failures are recorded to
3546 * map[0]. So we look through all the disks we started with and
3547 * see if any failures are still present, or if any new ones
3550 * FIXME add support for online capacity expansion and
3551 * raid-level-migration
3553 for (i
= 0; i
< prev
->num_members
; i
++) {
3554 ord
= __le32_to_cpu(prev
->disk_ord_tbl
[i
]);
3555 ord
|= __le32_to_cpu(map
->disk_ord_tbl
[i
]);
3556 idx
= ord_to_idx(ord
);
3558 disk
= get_imsm_disk(super
, idx
);
3559 if (!disk
|| disk
->status
& FAILED_DISK
||
3560 ord
& IMSM_ORD_REBUILD
)
3567 static int is_resyncing(struct imsm_dev
*dev
)
3569 struct imsm_map
*migr_map
;
3571 if (!dev
->vol
.migr_state
)
3574 if (migr_type(dev
) == MIGR_INIT
||
3575 migr_type(dev
) == MIGR_REPAIR
)
3578 migr_map
= get_imsm_map(dev
, 1);
3580 if (migr_map
->map_state
== IMSM_T_STATE_NORMAL
)
3586 static int is_rebuilding(struct imsm_dev
*dev
)
3588 struct imsm_map
*migr_map
;
3590 if (!dev
->vol
.migr_state
)
3593 if (migr_type(dev
) != MIGR_REBUILD
)
3596 migr_map
= get_imsm_map(dev
, 1);
3598 if (migr_map
->map_state
== IMSM_T_STATE_DEGRADED
)
3604 /* return true if we recorded new information */
3605 static int mark_failure(struct imsm_dev
*dev
, struct imsm_disk
*disk
, int idx
)
3609 struct imsm_map
*map
;
3611 /* new failures are always set in map[0] */
3612 map
= get_imsm_map(dev
, 0);
3614 slot
= get_imsm_disk_slot(map
, idx
);
3618 ord
= __le32_to_cpu(map
->disk_ord_tbl
[slot
]);
3619 if ((disk
->status
& FAILED_DISK
) && (ord
& IMSM_ORD_REBUILD
))
3622 disk
->status
|= FAILED_DISK
;
3623 set_imsm_ord_tbl_ent(map
, slot
, idx
| IMSM_ORD_REBUILD
);
3624 if (map
->failed_disk_num
== ~0)
3625 map
->failed_disk_num
= slot
;
3629 static void mark_missing(struct imsm_dev
*dev
, struct imsm_disk
*disk
, int idx
)
3631 mark_failure(dev
, disk
, idx
);
3633 if (disk
->scsi_id
== __cpu_to_le32(~(__u32
)0))
3636 disk
->scsi_id
= __cpu_to_le32(~(__u32
)0);
3637 memmove(&disk
->serial
[0], &disk
->serial
[1], MAX_RAID_SERIAL_LEN
- 1);
3640 /* Handle dirty -> clean transititions and resync. Degraded and rebuild
3641 * states are handled in imsm_set_disk() with one exception, when a
3642 * resync is stopped due to a new failure this routine will set the
3643 * 'degraded' state for the array.
3645 static int imsm_set_array_state(struct active_array
*a
, int consistent
)
3647 int inst
= a
->info
.container_member
;
3648 struct intel_super
*super
= a
->container
->sb
;
3649 struct imsm_dev
*dev
= get_imsm_dev(super
, inst
);
3650 struct imsm_map
*map
= get_imsm_map(dev
, 0);
3651 int failed
= imsm_count_failed(super
, dev
);
3652 __u8 map_state
= imsm_check_degraded(super
, dev
, failed
);
3654 /* before we activate this array handle any missing disks */
3655 if (consistent
== 2 && super
->missing
) {
3658 dprintf("imsm: mark missing\n");
3659 end_migration(dev
, map_state
);
3660 for (dl
= super
->missing
; dl
; dl
= dl
->next
)
3661 mark_missing(dev
, &dl
->disk
, dl
->index
);
3662 super
->updates_pending
++;
3665 if (consistent
== 2 &&
3666 (!is_resync_complete(a
) ||
3667 map_state
!= IMSM_T_STATE_NORMAL
||
3668 dev
->vol
.migr_state
))
3671 if (is_resync_complete(a
)) {
3672 /* complete intialization / resync,
3673 * recovery and interrupted recovery is completed in
3676 if (is_resyncing(dev
)) {
3677 dprintf("imsm: mark resync done\n");
3678 end_migration(dev
, map_state
);
3679 super
->updates_pending
++;
3681 } else if (!is_resyncing(dev
) && !failed
) {
3682 /* mark the start of the init process if nothing is failed */
3683 dprintf("imsm: mark resync start (%llu)\n", a
->resync_start
);
3684 if (map
->map_state
== IMSM_T_STATE_UNINITIALIZED
)
3685 migrate(dev
, IMSM_T_STATE_NORMAL
, MIGR_INIT
);
3687 migrate(dev
, IMSM_T_STATE_NORMAL
, MIGR_REPAIR
);
3688 super
->updates_pending
++;
3691 /* FIXME check if we can update curr_migr_unit from resync_start */
3693 /* mark dirty / clean */
3694 if (dev
->vol
.dirty
!= !consistent
) {
3695 dprintf("imsm: mark '%s' (%llu)\n",
3696 consistent
? "clean" : "dirty", a
->resync_start
);
3701 super
->updates_pending
++;
3706 static void imsm_set_disk(struct active_array
*a
, int n
, int state
)
3708 int inst
= a
->info
.container_member
;
3709 struct intel_super
*super
= a
->container
->sb
;
3710 struct imsm_dev
*dev
= get_imsm_dev(super
, inst
);
3711 struct imsm_map
*map
= get_imsm_map(dev
, 0);
3712 struct imsm_disk
*disk
;
3717 if (n
> map
->num_members
)
3718 fprintf(stderr
, "imsm: set_disk %d out of range 0..%d\n",
3719 n
, map
->num_members
- 1);
3724 dprintf("imsm: set_disk %d:%x\n", n
, state
);
3726 ord
= get_imsm_ord_tbl_ent(dev
, n
);
3727 disk
= get_imsm_disk(super
, ord_to_idx(ord
));
3729 /* check for new failures */
3730 if (state
& DS_FAULTY
) {
3731 if (mark_failure(dev
, disk
, ord_to_idx(ord
)))
3732 super
->updates_pending
++;
3735 /* check if in_sync */
3736 if (state
& DS_INSYNC
&& ord
& IMSM_ORD_REBUILD
&& is_rebuilding(dev
)) {
3737 struct imsm_map
*migr_map
= get_imsm_map(dev
, 1);
3739 set_imsm_ord_tbl_ent(migr_map
, n
, ord_to_idx(ord
));
3740 super
->updates_pending
++;
3743 failed
= imsm_count_failed(super
, dev
);
3744 map_state
= imsm_check_degraded(super
, dev
, failed
);
3746 /* check if recovery complete, newly degraded, or failed */
3747 if (map_state
== IMSM_T_STATE_NORMAL
&& is_rebuilding(dev
)) {
3748 end_migration(dev
, map_state
);
3749 map
= get_imsm_map(dev
, 0);
3750 map
->failed_disk_num
= ~0;
3751 super
->updates_pending
++;
3752 } else if (map_state
== IMSM_T_STATE_DEGRADED
&&
3753 map
->map_state
!= map_state
&&
3754 !dev
->vol
.migr_state
) {
3755 dprintf("imsm: mark degraded\n");
3756 map
->map_state
= map_state
;
3757 super
->updates_pending
++;
3758 } else if (map_state
== IMSM_T_STATE_FAILED
&&
3759 map
->map_state
!= map_state
) {
3760 dprintf("imsm: mark failed\n");
3761 end_migration(dev
, map_state
);
3762 super
->updates_pending
++;
3766 static int store_imsm_mpb(int fd
, struct intel_super
*super
)
3768 struct imsm_super
*mpb
= super
->anchor
;
3769 __u32 mpb_size
= __le32_to_cpu(mpb
->mpb_size
);
3770 unsigned long long dsize
;
3771 unsigned long long sectors
;
3773 get_dev_size(fd
, NULL
, &dsize
);
3775 if (mpb_size
> 512) {
3776 /* -1 to account for anchor */
3777 sectors
= mpb_sectors(mpb
) - 1;
3779 /* write the extended mpb to the sectors preceeding the anchor */
3780 if (lseek64(fd
, dsize
- (512 * (2 + sectors
)), SEEK_SET
) < 0)
3783 if (write(fd
, super
->buf
+ 512, 512 * sectors
) != 512 * sectors
)
3787 /* first block is stored on second to last sector of the disk */
3788 if (lseek64(fd
, dsize
- (512 * 2), SEEK_SET
) < 0)
3791 if (write(fd
, super
->buf
, 512) != 512)
3797 static void imsm_sync_metadata(struct supertype
*container
)
3799 struct intel_super
*super
= container
->sb
;
3801 if (!super
->updates_pending
)
3804 write_super_imsm(super
, 0);
3806 super
->updates_pending
= 0;
3809 static struct dl
*imsm_readd(struct intel_super
*super
, int idx
, struct active_array
*a
)
3811 struct imsm_dev
*dev
= get_imsm_dev(super
, a
->info
.container_member
);
3812 int i
= get_imsm_disk_idx(dev
, idx
);
3815 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
3819 if (dl
&& dl
->disk
.status
& FAILED_DISK
)
3823 dprintf("%s: found %x:%x\n", __func__
, dl
->major
, dl
->minor
);
3828 static struct dl
*imsm_add_spare(struct intel_super
*super
, int slot
,
3829 struct active_array
*a
, int activate_new
)
3831 struct imsm_dev
*dev
= get_imsm_dev(super
, a
->info
.container_member
);
3832 int idx
= get_imsm_disk_idx(dev
, slot
);
3833 struct imsm_super
*mpb
= super
->anchor
;
3834 struct imsm_map
*map
;
3835 unsigned long long esize
;
3836 unsigned long long pos
;
3845 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
3846 /* If in this array, skip */
3847 for (d
= a
->info
.devs
; d
; d
= d
->next
)
3848 if (d
->state_fd
>= 0 &&
3849 d
->disk
.major
== dl
->major
&&
3850 d
->disk
.minor
== dl
->minor
) {
3851 dprintf("%x:%x already in array\n", dl
->major
, dl
->minor
);
3857 /* skip in use or failed drives */
3858 if (dl
->disk
.status
& FAILED_DISK
|| idx
== dl
->index
||
3860 dprintf("%x:%x status (failed: %d index: %d)\n",
3861 dl
->major
, dl
->minor
,
3862 (dl
->disk
.status
& FAILED_DISK
) == FAILED_DISK
, idx
);
3866 /* skip pure spares when we are looking for partially
3867 * assimilated drives
3869 if (dl
->index
== -1 && !activate_new
)
3872 /* Does this unused device have the requisite free space?
3873 * It needs to be able to cover all member volumes
3875 ex
= get_extents(super
, dl
);
3877 dprintf("cannot get extents\n");
3880 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
3881 dev
= get_imsm_dev(super
, i
);
3882 map
= get_imsm_map(dev
, 0);
3884 /* check if this disk is already a member of
3887 if (get_imsm_disk_slot(map
, dl
->index
) >= 0)
3893 array_start
= __le32_to_cpu(map
->pba_of_lba0
);
3894 blocks
= __le32_to_cpu(map
->blocks_per_member
);
3897 /* check that we can start at pba_of_lba0 with
3898 * blocks_per_member of space
3900 esize
= ex
[j
].start
- pos
;
3901 if (array_start
>= pos
&&
3902 array_start
+ blocks
< ex
[j
].start
) {
3906 pos
= ex
[j
].start
+ ex
[j
].size
;
3908 } while (ex
[j
-1].size
);
3915 if (i
< mpb
->num_raid_devs
) {
3916 dprintf("%x:%x does not have %u at %u\n",
3917 dl
->major
, dl
->minor
,
3918 blocks
, array_start
);
3928 static struct mdinfo
*imsm_activate_spare(struct active_array
*a
,
3929 struct metadata_update
**updates
)
3932 * Find a device with unused free space and use it to replace a
3933 * failed/vacant region in an array. We replace failed regions one a
3934 * array at a time. The result is that a new spare disk will be added
3935 * to the first failed array and after the monitor has finished
3936 * propagating failures the remainder will be consumed.
3938 * FIXME add a capability for mdmon to request spares from another
3942 struct intel_super
*super
= a
->container
->sb
;
3943 int inst
= a
->info
.container_member
;
3944 struct imsm_dev
*dev
= get_imsm_dev(super
, inst
);
3945 struct imsm_map
*map
= get_imsm_map(dev
, 0);
3946 int failed
= a
->info
.array
.raid_disks
;
3947 struct mdinfo
*rv
= NULL
;
3950 struct metadata_update
*mu
;
3952 struct imsm_update_activate_spare
*u
;
3956 for (d
= a
->info
.devs
; d
; d
= d
->next
) {
3957 if ((d
->curr_state
& DS_FAULTY
) &&
3959 /* wait for Removal to happen */
3961 if (d
->state_fd
>= 0)
3965 dprintf("imsm: activate spare: inst=%d failed=%d (%d) level=%d\n",
3966 inst
, failed
, a
->info
.array
.raid_disks
, a
->info
.array
.level
);
3967 if (imsm_check_degraded(super
, dev
, failed
) != IMSM_T_STATE_DEGRADED
)
3970 /* For each slot, if it is not working, find a spare */
3971 for (i
= 0; i
< a
->info
.array
.raid_disks
; i
++) {
3972 for (d
= a
->info
.devs
; d
; d
= d
->next
)
3973 if (d
->disk
.raid_disk
== i
)
3975 dprintf("found %d: %p %x\n", i
, d
, d
?d
->curr_state
:0);
3976 if (d
&& (d
->state_fd
>= 0))
3980 * OK, this device needs recovery. Try to re-add the
3981 * previous occupant of this slot, if this fails see if
3982 * we can continue the assimilation of a spare that was
3983 * partially assimilated, finally try to activate a new
3986 dl
= imsm_readd(super
, i
, a
);
3988 dl
= imsm_add_spare(super
, i
, a
, 0);
3990 dl
= imsm_add_spare(super
, i
, a
, 1);
3994 /* found a usable disk with enough space */
3995 di
= malloc(sizeof(*di
));
3998 memset(di
, 0, sizeof(*di
));
4000 /* dl->index will be -1 in the case we are activating a
4001 * pristine spare. imsm_process_update() will create a
4002 * new index in this case. Once a disk is found to be
4003 * failed in all member arrays it is kicked from the
4006 di
->disk
.number
= dl
->index
;
4008 /* (ab)use di->devs to store a pointer to the device
4011 di
->devs
= (struct mdinfo
*) dl
;
4013 di
->disk
.raid_disk
= i
;
4014 di
->disk
.major
= dl
->major
;
4015 di
->disk
.minor
= dl
->minor
;
4017 di
->data_offset
= __le32_to_cpu(map
->pba_of_lba0
);
4018 di
->component_size
= a
->info
.component_size
;
4019 di
->container_member
= inst
;
4023 dprintf("%x:%x to be %d at %llu\n", dl
->major
, dl
->minor
,
4024 i
, di
->data_offset
);
4030 /* No spares found */
4032 /* Now 'rv' has a list of devices to return.
4033 * Create a metadata_update record to update the
4034 * disk_ord_tbl for the array
4036 mu
= malloc(sizeof(*mu
));
4038 mu
->buf
= malloc(sizeof(struct imsm_update_activate_spare
) * num_spares
);
4039 if (mu
->buf
== NULL
) {
4046 struct mdinfo
*n
= rv
->next
;
4055 mu
->len
= sizeof(struct imsm_update_activate_spare
) * num_spares
;
4056 mu
->next
= *updates
;
4057 u
= (struct imsm_update_activate_spare
*) mu
->buf
;
4059 for (di
= rv
; di
; di
= di
->next
) {
4060 u
->type
= update_activate_spare
;
4061 u
->dl
= (struct dl
*) di
->devs
;
4063 u
->slot
= di
->disk
.raid_disk
;
4074 static int disks_overlap(struct intel_super
*super
, int idx
, struct imsm_update_create_array
*u
)
4076 struct imsm_dev
*dev
= get_imsm_dev(super
, idx
);
4077 struct imsm_map
*map
= get_imsm_map(dev
, 0);
4078 struct imsm_map
*new_map
= get_imsm_map(&u
->dev
, 0);
4079 struct disk_info
*inf
= get_disk_info(u
);
4080 struct imsm_disk
*disk
;
4084 for (i
= 0; i
< map
->num_members
; i
++) {
4085 disk
= get_imsm_disk(super
, get_imsm_disk_idx(dev
, i
));
4086 for (j
= 0; j
< new_map
->num_members
; j
++)
4087 if (serialcmp(disk
->serial
, inf
[j
].serial
) == 0)
4094 static void imsm_delete(struct intel_super
*super
, struct dl
**dlp
, int index
);
4096 static void imsm_process_update(struct supertype
*st
,
4097 struct metadata_update
*update
)
4100 * crack open the metadata_update envelope to find the update record
4101 * update can be one of:
4102 * update_activate_spare - a spare device has replaced a failed
4103 * device in an array, update the disk_ord_tbl. If this disk is
4104 * present in all member arrays then also clear the SPARE_DISK
4107 struct intel_super
*super
= st
->sb
;
4108 struct imsm_super
*mpb
;
4109 enum imsm_update_type type
= *(enum imsm_update_type
*) update
->buf
;
4111 /* update requires a larger buf but the allocation failed */
4112 if (super
->next_len
&& !super
->next_buf
) {
4113 super
->next_len
= 0;
4117 if (super
->next_buf
) {
4118 memcpy(super
->next_buf
, super
->buf
, super
->len
);
4120 super
->len
= super
->next_len
;
4121 super
->buf
= super
->next_buf
;
4123 super
->next_len
= 0;
4124 super
->next_buf
= NULL
;
4127 mpb
= super
->anchor
;
4130 case update_activate_spare
: {
4131 struct imsm_update_activate_spare
*u
= (void *) update
->buf
;
4132 struct imsm_dev
*dev
= get_imsm_dev(super
, u
->array
);
4133 struct imsm_map
*map
= get_imsm_map(dev
, 0);
4134 struct imsm_map
*migr_map
;
4135 struct active_array
*a
;
4136 struct imsm_disk
*disk
;
4141 int victim
= get_imsm_disk_idx(dev
, u
->slot
);
4144 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
4149 fprintf(stderr
, "error: imsm_activate_spare passed "
4150 "an unknown disk (index: %d)\n",
4155 super
->updates_pending
++;
4157 /* count failures (excluding rebuilds and the victim)
4158 * to determine map[0] state
4161 for (i
= 0; i
< map
->num_members
; i
++) {
4164 disk
= get_imsm_disk(super
, get_imsm_disk_idx(dev
, i
));
4165 if (!disk
|| disk
->status
& FAILED_DISK
)
4169 /* adding a pristine spare, assign a new index */
4170 if (dl
->index
< 0) {
4171 dl
->index
= super
->anchor
->num_disks
;
4172 super
->anchor
->num_disks
++;
4175 disk
->status
|= CONFIGURED_DISK
;
4176 disk
->status
&= ~SPARE_DISK
;
4179 to_state
= imsm_check_degraded(super
, dev
, failed
);
4180 map
->map_state
= IMSM_T_STATE_DEGRADED
;
4181 migrate(dev
, to_state
, MIGR_REBUILD
);
4182 migr_map
= get_imsm_map(dev
, 1);
4183 set_imsm_ord_tbl_ent(map
, u
->slot
, dl
->index
);
4184 set_imsm_ord_tbl_ent(migr_map
, u
->slot
, dl
->index
| IMSM_ORD_REBUILD
);
4186 /* count arrays using the victim in the metadata */
4188 for (a
= st
->arrays
; a
; a
= a
->next
) {
4189 dev
= get_imsm_dev(super
, a
->info
.container_member
);
4190 map
= get_imsm_map(dev
, 0);
4192 if (get_imsm_disk_slot(map
, victim
) >= 0)
4196 /* delete the victim if it is no longer being
4202 /* We know that 'manager' isn't touching anything,
4203 * so it is safe to delete
4205 for (dlp
= &super
->disks
; *dlp
; dlp
= &(*dlp
)->next
)
4206 if ((*dlp
)->index
== victim
)
4209 /* victim may be on the missing list */
4211 for (dlp
= &super
->missing
; *dlp
; dlp
= &(*dlp
)->next
)
4212 if ((*dlp
)->index
== victim
)
4214 imsm_delete(super
, dlp
, victim
);
4218 case update_create_array
: {
4219 /* someone wants to create a new array, we need to be aware of
4220 * a few races/collisions:
4221 * 1/ 'Create' called by two separate instances of mdadm
4222 * 2/ 'Create' versus 'activate_spare': mdadm has chosen
4223 * devices that have since been assimilated via
4225 * In the event this update can not be carried out mdadm will
4226 * (FIX ME) notice that its update did not take hold.
4228 struct imsm_update_create_array
*u
= (void *) update
->buf
;
4229 struct intel_dev
*dv
;
4230 struct imsm_dev
*dev
;
4231 struct imsm_map
*map
, *new_map
;
4232 unsigned long long start
, end
;
4233 unsigned long long new_start
, new_end
;
4235 struct disk_info
*inf
;
4238 /* handle racing creates: first come first serve */
4239 if (u
->dev_idx
< mpb
->num_raid_devs
) {
4240 dprintf("%s: subarray %d already defined\n",
4241 __func__
, u
->dev_idx
);
4245 /* check update is next in sequence */
4246 if (u
->dev_idx
!= mpb
->num_raid_devs
) {
4247 dprintf("%s: can not create array %d expected index %d\n",
4248 __func__
, u
->dev_idx
, mpb
->num_raid_devs
);
4252 new_map
= get_imsm_map(&u
->dev
, 0);
4253 new_start
= __le32_to_cpu(new_map
->pba_of_lba0
);
4254 new_end
= new_start
+ __le32_to_cpu(new_map
->blocks_per_member
);
4255 inf
= get_disk_info(u
);
4257 /* handle activate_spare versus create race:
4258 * check to make sure that overlapping arrays do not include
4261 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
4262 dev
= get_imsm_dev(super
, i
);
4263 map
= get_imsm_map(dev
, 0);
4264 start
= __le32_to_cpu(map
->pba_of_lba0
);
4265 end
= start
+ __le32_to_cpu(map
->blocks_per_member
);
4266 if ((new_start
>= start
&& new_start
<= end
) ||
4267 (start
>= new_start
&& start
<= new_end
))
4272 if (disks_overlap(super
, i
, u
)) {
4273 dprintf("%s: arrays overlap\n", __func__
);
4278 /* check that prepare update was successful */
4279 if (!update
->space
) {
4280 dprintf("%s: prepare update failed\n", __func__
);
4284 /* check that all disks are still active before committing
4285 * changes. FIXME: could we instead handle this by creating a
4286 * degraded array? That's probably not what the user expects,
4287 * so better to drop this update on the floor.
4289 for (i
= 0; i
< new_map
->num_members
; i
++) {
4290 dl
= serial_to_dl(inf
[i
].serial
, super
);
4292 dprintf("%s: disk disappeared\n", __func__
);
4297 super
->updates_pending
++;
4299 /* convert spares to members and fixup ord_tbl */
4300 for (i
= 0; i
< new_map
->num_members
; i
++) {
4301 dl
= serial_to_dl(inf
[i
].serial
, super
);
4302 if (dl
->index
== -1) {
4303 dl
->index
= mpb
->num_disks
;
4305 dl
->disk
.status
|= CONFIGURED_DISK
;
4306 dl
->disk
.status
&= ~SPARE_DISK
;
4308 set_imsm_ord_tbl_ent(new_map
, i
, dl
->index
);
4313 update
->space
= NULL
;
4314 imsm_copy_dev(dev
, &u
->dev
);
4315 dv
->index
= u
->dev_idx
;
4316 dv
->next
= super
->devlist
;
4317 super
->devlist
= dv
;
4318 mpb
->num_raid_devs
++;
4320 imsm_update_version_info(super
);
4323 /* mdmon knows how to release update->space, but not
4324 * ((struct intel_dev *) update->space)->dev
4326 if (update
->space
) {
4332 case update_add_disk
:
4334 /* we may be able to repair some arrays if disks are
4337 struct active_array
*a
;
4339 super
->updates_pending
++;
4340 for (a
= st
->arrays
; a
; a
= a
->next
)
4341 a
->check_degraded
= 1;
4343 /* add some spares to the metadata */
4344 while (super
->add
) {
4348 super
->add
= al
->next
;
4349 al
->next
= super
->disks
;
4351 dprintf("%s: added %x:%x\n",
4352 __func__
, al
->major
, al
->minor
);
4359 static void imsm_prepare_update(struct supertype
*st
,
4360 struct metadata_update
*update
)
4363 * Allocate space to hold new disk entries, raid-device entries or a new
4364 * mpb if necessary. The manager synchronously waits for updates to
4365 * complete in the monitor, so new mpb buffers allocated here can be
4366 * integrated by the monitor thread without worrying about live pointers
4367 * in the manager thread.
4369 enum imsm_update_type type
= *(enum imsm_update_type
*) update
->buf
;
4370 struct intel_super
*super
= st
->sb
;
4371 struct imsm_super
*mpb
= super
->anchor
;
4376 case update_create_array
: {
4377 struct imsm_update_create_array
*u
= (void *) update
->buf
;
4378 struct intel_dev
*dv
;
4379 struct imsm_dev
*dev
= &u
->dev
;
4380 struct imsm_map
*map
= get_imsm_map(dev
, 0);
4382 struct disk_info
*inf
;
4386 inf
= get_disk_info(u
);
4387 len
= sizeof_imsm_dev(dev
, 1);
4388 /* allocate a new super->devlist entry */
4389 dv
= malloc(sizeof(*dv
));
4391 dv
->dev
= malloc(len
);
4396 update
->space
= NULL
;
4400 /* count how many spares will be converted to members */
4401 for (i
= 0; i
< map
->num_members
; i
++) {
4402 dl
= serial_to_dl(inf
[i
].serial
, super
);
4404 /* hmm maybe it failed?, nothing we can do about
4409 if (count_memberships(dl
, super
) == 0)
4412 len
+= activate
* sizeof(struct imsm_disk
);
4419 /* check if we need a larger metadata buffer */
4420 if (super
->next_buf
)
4421 buf_len
= super
->next_len
;
4423 buf_len
= super
->len
;
4425 if (__le32_to_cpu(mpb
->mpb_size
) + len
> buf_len
) {
4426 /* ok we need a larger buf than what is currently allocated
4427 * if this allocation fails process_update will notice that
4428 * ->next_len is set and ->next_buf is NULL
4430 buf_len
= ROUND_UP(__le32_to_cpu(mpb
->mpb_size
) + len
, 512);
4431 if (super
->next_buf
)
4432 free(super
->next_buf
);
4434 super
->next_len
= buf_len
;
4435 if (posix_memalign(&super
->next_buf
, 512, buf_len
) == 0)
4436 memset(super
->next_buf
, 0, buf_len
);
4438 super
->next_buf
= NULL
;
4442 /* must be called while manager is quiesced */
4443 static void imsm_delete(struct intel_super
*super
, struct dl
**dlp
, int index
)
4445 struct imsm_super
*mpb
= super
->anchor
;
4447 struct imsm_dev
*dev
;
4448 struct imsm_map
*map
;
4449 int i
, j
, num_members
;
4452 dprintf("%s: deleting device[%d] from imsm_super\n",
4455 /* shift all indexes down one */
4456 for (iter
= super
->disks
; iter
; iter
= iter
->next
)
4457 if (iter
->index
> index
)
4459 for (iter
= super
->missing
; iter
; iter
= iter
->next
)
4460 if (iter
->index
> index
)
4463 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
4464 dev
= get_imsm_dev(super
, i
);
4465 map
= get_imsm_map(dev
, 0);
4466 num_members
= map
->num_members
;
4467 for (j
= 0; j
< num_members
; j
++) {
4468 /* update ord entries being careful not to propagate
4469 * ord-flags to the first map
4471 ord
= get_imsm_ord_tbl_ent(dev
, j
);
4473 if (ord_to_idx(ord
) <= index
)
4476 map
= get_imsm_map(dev
, 0);
4477 set_imsm_ord_tbl_ent(map
, j
, ord_to_idx(ord
- 1));
4478 map
= get_imsm_map(dev
, 1);
4480 set_imsm_ord_tbl_ent(map
, j
, ord
- 1);
4485 super
->updates_pending
++;
4487 struct dl
*dl
= *dlp
;
4489 *dlp
= (*dlp
)->next
;
4490 __free_imsm_disk(dl
);
4493 #endif /* MDASSEMBLE */
4495 struct superswitch super_imsm
= {
4497 .examine_super
= examine_super_imsm
,
4498 .brief_examine_super
= brief_examine_super_imsm
,
4499 .export_examine_super
= export_examine_super_imsm
,
4500 .detail_super
= detail_super_imsm
,
4501 .brief_detail_super
= brief_detail_super_imsm
,
4502 .write_init_super
= write_init_super_imsm
,
4503 .validate_geometry
= validate_geometry_imsm
,
4504 .add_to_super
= add_to_super_imsm
,
4505 .detail_platform
= detail_platform_imsm
,
4507 .match_home
= match_home_imsm
,
4508 .uuid_from_super
= uuid_from_super_imsm
,
4509 .getinfo_super
= getinfo_super_imsm
,
4510 .update_super
= update_super_imsm
,
4512 .avail_size
= avail_size_imsm
,
4514 .compare_super
= compare_super_imsm
,
4516 .load_super
= load_super_imsm
,
4517 .init_super
= init_super_imsm
,
4518 .store_super
= store_zero_imsm
,
4519 .free_super
= free_super_imsm
,
4520 .match_metadata_desc
= match_metadata_desc_imsm
,
4521 .container_content
= container_content_imsm
,
4522 .default_layout
= imsm_level_to_layout
,
4529 .open_new
= imsm_open_new
,
4530 .load_super
= load_super_imsm
,
4531 .set_array_state
= imsm_set_array_state
,
4532 .set_disk
= imsm_set_disk
,
4533 .sync_metadata
= imsm_sync_metadata
,
4534 .activate_spare
= imsm_activate_spare
,
4535 .process_update
= imsm_process_update
,
4536 .prepare_update
= imsm_prepare_update
,
4537 #endif /* MDASSEMBLE */