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
;
1207 info
->disk
.major
= 0;
1208 info
->disk
.minor
= 0;
1210 info
->disk
.major
= dl
->major
;
1211 info
->disk
.minor
= dl
->minor
;
1214 info
->data_offset
= __le32_to_cpu(map
->pba_of_lba0
);
1215 info
->component_size
= __le32_to_cpu(map
->blocks_per_member
);
1216 memset(info
->uuid
, 0, sizeof(info
->uuid
));
1218 if (map
->map_state
== IMSM_T_STATE_UNINITIALIZED
|| dev
->vol
.dirty
)
1219 info
->resync_start
= 0;
1220 else if (dev
->vol
.migr_state
)
1221 info
->resync_start
= __le32_to_cpu(dev
->vol
.curr_migr_unit
);
1223 info
->resync_start
= ~0ULL;
1225 strncpy(info
->name
, (char *) dev
->volume
, MAX_RAID_SERIAL_LEN
);
1226 info
->name
[MAX_RAID_SERIAL_LEN
] = 0;
1228 info
->array
.major_version
= -1;
1229 info
->array
.minor_version
= -2;
1230 sprintf(info
->text_version
, "/%s/%d",
1231 devnum2devname(st
->container_dev
),
1232 info
->container_member
);
1233 info
->safe_mode_delay
= 4000; /* 4 secs like the Matrix driver */
1234 uuid_from_super_imsm(st
, info
->uuid
);
1237 /* check the config file to see if we can return a real uuid for this spare */
1238 static void fixup_container_spare_uuid(struct mdinfo
*inf
)
1240 struct mddev_ident_s
*array_list
;
1242 if (inf
->array
.level
!= LEVEL_CONTAINER
||
1243 memcmp(inf
->uuid
, uuid_match_any
, sizeof(int[4])) != 0)
1246 array_list
= conf_get_ident(NULL
);
1248 for (; array_list
; array_list
= array_list
->next
) {
1249 if (array_list
->uuid_set
) {
1250 struct supertype
*_sst
; /* spare supertype */
1251 struct supertype
*_cst
; /* container supertype */
1253 _cst
= array_list
->st
;
1254 _sst
= _cst
->ss
->match_metadata_desc(inf
->text_version
);
1256 memcpy(inf
->uuid
, array_list
->uuid
, sizeof(int[4]));
1264 static void getinfo_super_imsm(struct supertype
*st
, struct mdinfo
*info
)
1266 struct intel_super
*super
= st
->sb
;
1267 struct imsm_disk
*disk
;
1270 if (super
->current_vol
>= 0) {
1271 getinfo_super_imsm_volume(st
, info
);
1275 /* Set raid_disks to zero so that Assemble will always pull in valid
1278 info
->array
.raid_disks
= 0;
1279 info
->array
.level
= LEVEL_CONTAINER
;
1280 info
->array
.layout
= 0;
1281 info
->array
.md_minor
= -1;
1282 info
->array
.ctime
= 0; /* N/A for imsm */
1283 info
->array
.utime
= 0;
1284 info
->array
.chunk_size
= 0;
1286 info
->disk
.major
= 0;
1287 info
->disk
.minor
= 0;
1288 info
->disk
.raid_disk
= -1;
1289 info
->reshape_active
= 0;
1290 info
->array
.major_version
= -1;
1291 info
->array
.minor_version
= -2;
1292 strcpy(info
->text_version
, "imsm");
1293 info
->safe_mode_delay
= 0;
1294 info
->disk
.number
= -1;
1295 info
->disk
.state
= 0;
1299 __u32 reserved
= imsm_reserved_sectors(super
, super
->disks
);
1301 disk
= &super
->disks
->disk
;
1302 info
->data_offset
= __le32_to_cpu(disk
->total_blocks
) - reserved
;
1303 info
->component_size
= reserved
;
1305 info
->disk
.state
= s
& CONFIGURED_DISK
? (1 << MD_DISK_ACTIVE
) : 0;
1306 /* we don't change info->disk.raid_disk here because
1307 * this state will be finalized in mdmon after we have
1308 * found the 'most fresh' version of the metadata
1310 info
->disk
.state
|= s
& FAILED_DISK
? (1 << MD_DISK_FAULTY
) : 0;
1311 info
->disk
.state
|= s
& SPARE_DISK
? 0 : (1 << MD_DISK_SYNC
);
1314 /* only call uuid_from_super_imsm when this disk is part of a populated container,
1315 * ->compare_super may have updated the 'num_raid_devs' field for spares
1317 if (info
->disk
.state
& (1 << MD_DISK_SYNC
) || super
->anchor
->num_raid_devs
)
1318 uuid_from_super_imsm(st
, info
->uuid
);
1320 memcpy(info
->uuid
, uuid_match_any
, sizeof(int[4]));
1321 fixup_container_spare_uuid(info
);
1325 static int update_super_imsm(struct supertype
*st
, struct mdinfo
*info
,
1326 char *update
, char *devname
, int verbose
,
1327 int uuid_set
, char *homehost
)
1331 /* For 'assemble' and 'force' we need to return non-zero if any
1332 * change was made. For others, the return value is ignored.
1333 * Update options are:
1334 * force-one : This device looks a bit old but needs to be included,
1335 * update age info appropriately.
1336 * assemble: clear any 'faulty' flag to allow this device to
1338 * force-array: Array is degraded but being forced, mark it clean
1339 * if that will be needed to assemble it.
1341 * newdev: not used ????
1342 * grow: Array has gained a new device - this is currently for
1344 * resync: mark as dirty so a resync will happen.
1345 * name: update the name - preserving the homehost
1347 * Following are not relevant for this imsm:
1348 * sparc2.2 : update from old dodgey metadata
1349 * super-minor: change the preferred_minor number
1350 * summaries: update redundant counters.
1351 * uuid: Change the uuid of the array to match watch is given
1352 * homehost: update the recorded homehost
1353 * _reshape_progress: record new reshape_progress position.
1356 //struct intel_super *super = st->sb;
1357 //struct imsm_super *mpb = super->mpb;
1359 if (strcmp(update
, "grow") == 0) {
1361 if (strcmp(update
, "resync") == 0) {
1362 /* dev->vol.dirty = 1; */
1365 /* IMSM has no concept of UUID or homehost */
1370 static size_t disks_to_mpb_size(int disks
)
1374 size
= sizeof(struct imsm_super
);
1375 size
+= (disks
- 1) * sizeof(struct imsm_disk
);
1376 size
+= 2 * sizeof(struct imsm_dev
);
1377 /* up to 2 maps per raid device (-2 for imsm_maps in imsm_dev */
1378 size
+= (4 - 2) * sizeof(struct imsm_map
);
1379 /* 4 possible disk_ord_tbl's */
1380 size
+= 4 * (disks
- 1) * sizeof(__u32
);
1385 static __u64
avail_size_imsm(struct supertype
*st
, __u64 devsize
)
1387 if (devsize
< (MPB_SECTOR_CNT
+ IMSM_RESERVED_SECTORS
))
1390 return devsize
- (MPB_SECTOR_CNT
+ IMSM_RESERVED_SECTORS
);
1393 static void free_devlist(struct intel_super
*super
)
1395 struct intel_dev
*dv
;
1397 while (super
->devlist
) {
1398 dv
= super
->devlist
->next
;
1399 free(super
->devlist
->dev
);
1400 free(super
->devlist
);
1401 super
->devlist
= dv
;
1405 static void imsm_copy_dev(struct imsm_dev
*dest
, struct imsm_dev
*src
)
1407 memcpy(dest
, src
, sizeof_imsm_dev(src
, 0));
1410 static int compare_super_imsm(struct supertype
*st
, struct supertype
*tst
)
1414 * 0 same, or first was empty, and second was copied
1415 * 1 second had wrong number
1417 * 3 wrong other info
1419 struct intel_super
*first
= st
->sb
;
1420 struct intel_super
*sec
= tst
->sb
;
1428 if (memcmp(first
->anchor
->sig
, sec
->anchor
->sig
, MAX_SIGNATURE_LENGTH
) != 0)
1431 /* if an anchor does not have num_raid_devs set then it is a free
1434 if (first
->anchor
->num_raid_devs
> 0 &&
1435 sec
->anchor
->num_raid_devs
> 0) {
1436 if (first
->anchor
->family_num
!= sec
->anchor
->family_num
)
1440 /* if 'first' is a spare promote it to a populated mpb with sec's
1443 if (first
->anchor
->num_raid_devs
== 0 &&
1444 sec
->anchor
->num_raid_devs
> 0) {
1446 struct intel_dev
*dv
;
1447 struct imsm_dev
*dev
;
1449 /* we need to copy raid device info from sec if an allocation
1450 * fails here we don't associate the spare
1452 for (i
= 0; i
< sec
->anchor
->num_raid_devs
; i
++) {
1453 dv
= malloc(sizeof(*dv
));
1456 dev
= malloc(sizeof_imsm_dev(get_imsm_dev(sec
, i
), 1));
1463 dv
->next
= first
->devlist
;
1464 first
->devlist
= dv
;
1466 if (i
<= sec
->anchor
->num_raid_devs
) {
1467 /* allocation failure */
1468 free_devlist(first
);
1469 fprintf(stderr
, "imsm: failed to associate spare\n");
1472 for (i
= 0; i
< sec
->anchor
->num_raid_devs
; i
++)
1473 imsm_copy_dev(get_imsm_dev(first
, i
), get_imsm_dev(sec
, i
));
1475 first
->anchor
->num_raid_devs
= sec
->anchor
->num_raid_devs
;
1476 first
->anchor
->family_num
= sec
->anchor
->family_num
;
1482 static void fd2devname(int fd
, char *name
)
1491 if (fstat(fd
, &st
) != 0)
1493 sprintf(path
, "/sys/dev/block/%d:%d",
1494 major(st
.st_rdev
), minor(st
.st_rdev
));
1496 rv
= readlink(path
, dname
, sizeof(dname
));
1501 nm
= strrchr(dname
, '/');
1503 snprintf(name
, MAX_RAID_SERIAL_LEN
, "/dev/%s", nm
);
1507 extern int scsi_get_serial(int fd
, void *buf
, size_t buf_len
);
1509 static int imsm_read_serial(int fd
, char *devname
,
1510 __u8 serial
[MAX_RAID_SERIAL_LEN
])
1512 unsigned char scsi_serial
[255];
1521 memset(scsi_serial
, 0, sizeof(scsi_serial
));
1523 rv
= scsi_get_serial(fd
, scsi_serial
, sizeof(scsi_serial
));
1525 if (rv
&& check_env("IMSM_DEVNAME_AS_SERIAL")) {
1526 memset(serial
, 0, MAX_RAID_SERIAL_LEN
);
1527 fd2devname(fd
, (char *) serial
);
1534 Name
": Failed to retrieve serial for %s\n",
1539 rsp_len
= scsi_serial
[3];
1543 Name
": Failed to retrieve serial for %s\n",
1547 rsp_buf
= (char *) &scsi_serial
[4];
1549 /* trim all whitespace and non-printable characters and convert
1552 for (i
= 0, dest
= rsp_buf
; i
< rsp_len
; i
++) {
1555 /* ':' is reserved for use in placeholder serial
1556 * numbers for missing disks
1564 len
= dest
- rsp_buf
;
1567 /* truncate leading characters */
1568 if (len
> MAX_RAID_SERIAL_LEN
) {
1569 dest
+= len
- MAX_RAID_SERIAL_LEN
;
1570 len
= MAX_RAID_SERIAL_LEN
;
1573 memset(serial
, 0, MAX_RAID_SERIAL_LEN
);
1574 memcpy(serial
, dest
, len
);
1579 static int serialcmp(__u8
*s1
, __u8
*s2
)
1581 return strncmp((char *) s1
, (char *) s2
, MAX_RAID_SERIAL_LEN
);
1584 static void serialcpy(__u8
*dest
, __u8
*src
)
1586 strncpy((char *) dest
, (char *) src
, MAX_RAID_SERIAL_LEN
);
1589 static struct dl
*serial_to_dl(__u8
*serial
, struct intel_super
*super
)
1593 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
1594 if (serialcmp(dl
->serial
, serial
) == 0)
1601 load_imsm_disk(int fd
, struct intel_super
*super
, char *devname
, int keep_fd
)
1608 __u8 serial
[MAX_RAID_SERIAL_LEN
];
1610 rv
= imsm_read_serial(fd
, devname
, serial
);
1615 /* check if this is a disk we have seen before. it may be a spare in
1616 * super->disks while the current anchor believes it is a raid member,
1617 * check if we need to update dl->index
1619 dl
= serial_to_dl(serial
, super
);
1621 dl
= malloc(sizeof(*dl
));
1628 Name
": failed to allocate disk buffer for %s\n",
1635 dl
->major
= major(stb
.st_rdev
);
1636 dl
->minor
= minor(stb
.st_rdev
);
1637 dl
->next
= super
->disks
;
1638 dl
->fd
= keep_fd
? fd
: -1;
1639 dl
->devname
= devname
? strdup(devname
) : NULL
;
1640 serialcpy(dl
->serial
, serial
);
1643 } else if (keep_fd
) {
1648 /* look up this disk's index in the current anchor */
1649 for (i
= 0; i
< super
->anchor
->num_disks
; i
++) {
1650 struct imsm_disk
*disk_iter
;
1652 disk_iter
= __get_imsm_disk(super
->anchor
, i
);
1654 if (serialcmp(disk_iter
->serial
, dl
->serial
) == 0) {
1655 dl
->disk
= *disk_iter
;
1656 /* only set index on disks that are a member of a
1657 * populated contianer, i.e. one with raid_devs
1659 if (dl
->disk
.status
& FAILED_DISK
)
1661 else if (dl
->disk
.status
& SPARE_DISK
)
1670 /* no match, maybe a stale failed drive */
1671 if (i
== super
->anchor
->num_disks
&& dl
->index
>= 0) {
1672 dl
->disk
= *__get_imsm_disk(super
->anchor
, dl
->index
);
1673 if (dl
->disk
.status
& FAILED_DISK
)
1684 /* When migrating map0 contains the 'destination' state while map1
1685 * contains the current state. When not migrating map0 contains the
1686 * current state. This routine assumes that map[0].map_state is set to
1687 * the current array state before being called.
1689 * Migration is indicated by one of the following states
1690 * 1/ Idle (migr_state=0 map0state=normal||unitialized||degraded||failed)
1691 * 2/ Initialize (migr_state=1 migr_type=MIGR_INIT map0state=normal
1692 * map1state=unitialized)
1693 * 3/ Repair (Resync) (migr_state=1 migr_type=MIGR_REPAIR map0state=normal
1695 * 4/ Rebuild (migr_state=1 migr_type=MIGR_REBUILD map0state=normal
1696 * map1state=degraded)
1698 static void migrate(struct imsm_dev
*dev
, __u8 to_state
, int migr_type
)
1700 struct imsm_map
*dest
;
1701 struct imsm_map
*src
= get_imsm_map(dev
, 0);
1703 dev
->vol
.migr_state
= 1;
1704 set_migr_type(dev
, migr_type
);
1705 dev
->vol
.curr_migr_unit
= 0;
1706 dest
= get_imsm_map(dev
, 1);
1708 /* duplicate and then set the target end state in map[0] */
1709 memcpy(dest
, src
, sizeof_imsm_map(src
));
1710 if (migr_type
== MIGR_REBUILD
) {
1714 for (i
= 0; i
< src
->num_members
; i
++) {
1715 ord
= __le32_to_cpu(src
->disk_ord_tbl
[i
]);
1716 set_imsm_ord_tbl_ent(src
, i
, ord_to_idx(ord
));
1720 src
->map_state
= to_state
;
1723 static void end_migration(struct imsm_dev
*dev
, __u8 map_state
)
1725 struct imsm_map
*map
= get_imsm_map(dev
, 0);
1726 struct imsm_map
*prev
= get_imsm_map(dev
, dev
->vol
.migr_state
);
1729 /* merge any IMSM_ORD_REBUILD bits that were not successfully
1730 * completed in the last migration.
1732 * FIXME add support for online capacity expansion and
1733 * raid-level-migration
1735 for (i
= 0; i
< prev
->num_members
; i
++)
1736 map
->disk_ord_tbl
[i
] |= prev
->disk_ord_tbl
[i
];
1738 dev
->vol
.migr_state
= 0;
1739 dev
->vol
.curr_migr_unit
= 0;
1740 map
->map_state
= map_state
;
1744 static int parse_raid_devices(struct intel_super
*super
)
1747 struct imsm_dev
*dev_new
;
1748 size_t len
, len_migr
;
1749 size_t space_needed
= 0;
1750 struct imsm_super
*mpb
= super
->anchor
;
1752 for (i
= 0; i
< super
->anchor
->num_raid_devs
; i
++) {
1753 struct imsm_dev
*dev_iter
= __get_imsm_dev(super
->anchor
, i
);
1754 struct intel_dev
*dv
;
1756 len
= sizeof_imsm_dev(dev_iter
, 0);
1757 len_migr
= sizeof_imsm_dev(dev_iter
, 1);
1759 space_needed
+= len_migr
- len
;
1761 dv
= malloc(sizeof(*dv
));
1764 dev_new
= malloc(len_migr
);
1769 imsm_copy_dev(dev_new
, dev_iter
);
1772 dv
->next
= super
->devlist
;
1773 super
->devlist
= dv
;
1776 /* ensure that super->buf is large enough when all raid devices
1779 if (__le32_to_cpu(mpb
->mpb_size
) + space_needed
> super
->len
) {
1782 len
= ROUND_UP(__le32_to_cpu(mpb
->mpb_size
) + space_needed
, 512);
1783 if (posix_memalign(&buf
, 512, len
) != 0)
1786 memcpy(buf
, super
->buf
, super
->len
);
1787 memset(buf
+ super
->len
, 0, len
- super
->len
);
1796 /* retrieve a pointer to the bbm log which starts after all raid devices */
1797 struct bbm_log
*__get_imsm_bbm_log(struct imsm_super
*mpb
)
1801 if (__le32_to_cpu(mpb
->bbm_log_size
)) {
1803 ptr
+= mpb
->mpb_size
- __le32_to_cpu(mpb
->bbm_log_size
);
1809 static void __free_imsm(struct intel_super
*super
, int free_disks
);
1811 /* load_imsm_mpb - read matrix metadata
1812 * allocates super->mpb to be freed by free_super
1814 static int load_imsm_mpb(int fd
, struct intel_super
*super
, char *devname
)
1816 unsigned long long dsize
;
1817 unsigned long long sectors
;
1819 struct imsm_super
*anchor
;
1823 get_dev_size(fd
, NULL
, &dsize
);
1825 if (lseek64(fd
, dsize
- (512 * 2), SEEK_SET
) < 0) {
1828 Name
": Cannot seek to anchor block on %s: %s\n",
1829 devname
, strerror(errno
));
1833 if (posix_memalign((void**)&anchor
, 512, 512) != 0) {
1836 Name
": Failed to allocate imsm anchor buffer"
1837 " on %s\n", devname
);
1840 if (read(fd
, anchor
, 512) != 512) {
1843 Name
": Cannot read anchor block on %s: %s\n",
1844 devname
, strerror(errno
));
1849 if (strncmp((char *) anchor
->sig
, MPB_SIGNATURE
, MPB_SIG_LEN
) != 0) {
1852 Name
": no IMSM anchor on %s\n", devname
);
1857 __free_imsm(super
, 0);
1858 super
->len
= ROUND_UP(anchor
->mpb_size
, 512);
1859 if (posix_memalign(&super
->buf
, 512, super
->len
) != 0) {
1862 Name
": unable to allocate %zu byte mpb buffer\n",
1867 memcpy(super
->buf
, anchor
, 512);
1869 sectors
= mpb_sectors(anchor
) - 1;
1872 check_sum
= __gen_imsm_checksum(super
->anchor
);
1873 if (check_sum
!= __le32_to_cpu(super
->anchor
->check_sum
)) {
1876 Name
": IMSM checksum %x != %x on %s\n",
1878 __le32_to_cpu(super
->anchor
->check_sum
),
1883 rc
= load_imsm_disk(fd
, super
, devname
, 0);
1885 rc
= parse_raid_devices(super
);
1889 /* read the extended mpb */
1890 if (lseek64(fd
, dsize
- (512 * (2 + sectors
)), SEEK_SET
) < 0) {
1893 Name
": Cannot seek to extended mpb on %s: %s\n",
1894 devname
, strerror(errno
));
1898 if (read(fd
, super
->buf
+ 512, super
->len
- 512) != super
->len
- 512) {
1901 Name
": Cannot read extended mpb on %s: %s\n",
1902 devname
, strerror(errno
));
1906 check_sum
= __gen_imsm_checksum(super
->anchor
);
1907 if (check_sum
!= __le32_to_cpu(super
->anchor
->check_sum
)) {
1910 Name
": IMSM checksum %x != %x on %s\n",
1911 check_sum
, __le32_to_cpu(super
->anchor
->check_sum
),
1916 /* FIXME the BBM log is disk specific so we cannot use this global
1917 * buffer for all disks. Ok for now since we only look at the global
1918 * bbm_log_size parameter to gate assembly
1920 super
->bbm_log
= __get_imsm_bbm_log(super
->anchor
);
1922 rc
= load_imsm_disk(fd
, super
, devname
, 0);
1924 rc
= parse_raid_devices(super
);
1929 static void __free_imsm_disk(struct dl
*d
)
1940 static void free_imsm_disks(struct intel_super
*super
)
1944 while (super
->disks
) {
1946 super
->disks
= d
->next
;
1947 __free_imsm_disk(d
);
1949 while (super
->missing
) {
1951 super
->missing
= d
->next
;
1952 __free_imsm_disk(d
);
1957 /* free all the pieces hanging off of a super pointer */
1958 static void __free_imsm(struct intel_super
*super
, int free_disks
)
1965 free_imsm_disks(super
);
1966 free_devlist(super
);
1968 free((void *) super
->hba
);
1973 static void free_imsm(struct intel_super
*super
)
1975 __free_imsm(super
, 1);
1979 static void free_super_imsm(struct supertype
*st
)
1981 struct intel_super
*super
= st
->sb
;
1990 static struct intel_super
*alloc_super(int creating_imsm
)
1992 struct intel_super
*super
= malloc(sizeof(*super
));
1995 memset(super
, 0, sizeof(*super
));
1996 super
->creating_imsm
= creating_imsm
;
1997 super
->current_vol
= -1;
1998 super
->create_offset
= ~((__u32
) 0);
1999 if (!check_env("IMSM_NO_PLATFORM"))
2000 super
->orom
= find_imsm_orom();
2001 if (super
->orom
&& !check_env("IMSM_TEST_OROM")) {
2002 struct sys_dev
*list
, *ent
;
2004 /* find the first intel ahci controller */
2005 list
= find_driver_devices("pci", "ahci");
2006 for (ent
= list
; ent
; ent
= ent
->next
)
2007 if (devpath_to_vendor(ent
->path
) == 0x8086)
2010 super
->hba
= ent
->path
;
2013 free_sys_dev(&list
);
2021 /* find_missing - helper routine for load_super_imsm_all that identifies
2022 * disks that have disappeared from the system. This routine relies on
2023 * the mpb being uptodate, which it is at load time.
2025 static int find_missing(struct intel_super
*super
)
2028 struct imsm_super
*mpb
= super
->anchor
;
2030 struct imsm_disk
*disk
;
2032 for (i
= 0; i
< mpb
->num_disks
; i
++) {
2033 disk
= __get_imsm_disk(mpb
, i
);
2034 dl
= serial_to_dl(disk
->serial
, super
);
2038 dl
= malloc(sizeof(*dl
));
2044 dl
->devname
= strdup("missing");
2046 serialcpy(dl
->serial
, disk
->serial
);
2049 dl
->next
= super
->missing
;
2050 super
->missing
= dl
;
2056 static int load_super_imsm_all(struct supertype
*st
, int fd
, void **sbp
,
2057 char *devname
, int keep_fd
)
2060 struct intel_super
*super
;
2061 struct mdinfo
*sd
, *best
= NULL
;
2067 int devnum
= fd2devnum(fd
);
2069 enum sysfs_read_flags flags
;
2071 flags
= GET_LEVEL
|GET_VERSION
|GET_DEVS
|GET_STATE
;
2072 if (mdmon_running(devnum
))
2073 flags
|= SKIP_GONE_DEVS
;
2075 /* check if 'fd' an opened container */
2076 sra
= sysfs_read(fd
, 0, flags
);
2080 if (sra
->array
.major_version
!= -1 ||
2081 sra
->array
.minor_version
!= -2 ||
2082 strcmp(sra
->text_version
, "imsm") != 0)
2085 super
= alloc_super(0);
2089 /* find the most up to date disk in this array, skipping spares */
2090 for (sd
= sra
->devs
; sd
; sd
= sd
->next
) {
2091 sprintf(nm
, "%d:%d", sd
->disk
.major
, sd
->disk
.minor
);
2092 dfd
= dev_open(nm
, keep_fd
? O_RDWR
: O_RDONLY
);
2097 rv
= load_imsm_mpb(dfd
, super
, NULL
);
2099 /* retry the load if we might have raced against mdmon */
2100 if (rv
== 3 && mdmon_running(devnum
))
2101 for (retry
= 0; retry
< 3; retry
++) {
2103 rv
= load_imsm_mpb(dfd
, super
, NULL
);
2110 if (super
->anchor
->num_raid_devs
== 0)
2113 gen
= __le32_to_cpu(super
->anchor
->generation_num
);
2114 if (!best
|| gen
> bestgen
) {
2129 /* load the most up to date anchor */
2130 sprintf(nm
, "%d:%d", best
->disk
.major
, best
->disk
.minor
);
2131 dfd
= dev_open(nm
, O_RDONLY
);
2136 rv
= load_imsm_mpb(dfd
, super
, NULL
);
2143 /* re-parse the disk list with the current anchor */
2144 for (sd
= sra
->devs
; sd
; sd
= sd
->next
) {
2145 sprintf(nm
, "%d:%d", sd
->disk
.major
, sd
->disk
.minor
);
2146 dfd
= dev_open(nm
, keep_fd
? O_RDWR
: O_RDONLY
);
2151 load_imsm_disk(dfd
, super
, NULL
, keep_fd
);
2157 if (find_missing(super
) != 0) {
2162 if (st
->subarray
[0]) {
2163 if (atoi(st
->subarray
) <= super
->anchor
->num_raid_devs
)
2164 super
->current_vol
= atoi(st
->subarray
);
2170 st
->container_dev
= devnum
;
2171 if (st
->ss
== NULL
) {
2172 st
->ss
= &super_imsm
;
2173 st
->minor_version
= 0;
2174 st
->max_devs
= IMSM_MAX_DEVICES
;
2176 st
->loaded_container
= 1;
2182 static int load_super_imsm(struct supertype
*st
, int fd
, char *devname
)
2184 struct intel_super
*super
;
2188 if (load_super_imsm_all(st
, fd
, &st
->sb
, devname
, 1) == 0)
2191 if (st
->subarray
[0])
2192 return 1; /* FIXME */
2194 super
= alloc_super(0);
2197 Name
": malloc of %zu failed.\n",
2202 rv
= load_imsm_mpb(fd
, super
, devname
);
2207 Name
": Failed to load all information "
2208 "sections on %s\n", devname
);
2214 if (st
->ss
== NULL
) {
2215 st
->ss
= &super_imsm
;
2216 st
->minor_version
= 0;
2217 st
->max_devs
= IMSM_MAX_DEVICES
;
2219 st
->loaded_container
= 0;
2224 static __u16
info_to_blocks_per_strip(mdu_array_info_t
*info
)
2226 if (info
->level
== 1)
2228 return info
->chunk_size
>> 9;
2231 static __u32
info_to_num_data_stripes(mdu_array_info_t
*info
, int num_domains
)
2235 num_stripes
= (info
->size
* 2) / info_to_blocks_per_strip(info
);
2236 num_stripes
/= num_domains
;
2241 static __u32
info_to_blocks_per_member(mdu_array_info_t
*info
)
2243 if (info
->level
== 1)
2244 return info
->size
* 2;
2246 return (info
->size
* 2) & ~(info_to_blocks_per_strip(info
) - 1);
2249 static void imsm_update_version_info(struct intel_super
*super
)
2251 /* update the version and attributes */
2252 struct imsm_super
*mpb
= super
->anchor
;
2254 struct imsm_dev
*dev
;
2255 struct imsm_map
*map
;
2258 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
2259 dev
= get_imsm_dev(super
, i
);
2260 map
= get_imsm_map(dev
, 0);
2261 if (__le32_to_cpu(dev
->size_high
) > 0)
2262 mpb
->attributes
|= MPB_ATTRIB_2TB
;
2264 /* FIXME detect when an array spans a port multiplier */
2266 mpb
->attributes
|= MPB_ATTRIB_PM
;
2269 if (mpb
->num_raid_devs
> 1 ||
2270 mpb
->attributes
!= MPB_ATTRIB_CHECKSUM_VERIFY
) {
2271 version
= MPB_VERSION_ATTRIBS
;
2272 switch (get_imsm_raid_level(map
)) {
2273 case 0: mpb
->attributes
|= MPB_ATTRIB_RAID0
; break;
2274 case 1: mpb
->attributes
|= MPB_ATTRIB_RAID1
; break;
2275 case 10: mpb
->attributes
|= MPB_ATTRIB_RAID10
; break;
2276 case 5: mpb
->attributes
|= MPB_ATTRIB_RAID5
; break;
2279 if (map
->num_members
>= 5)
2280 version
= MPB_VERSION_5OR6_DISK_ARRAY
;
2281 else if (dev
->status
== DEV_CLONE_N_GO
)
2282 version
= MPB_VERSION_CNG
;
2283 else if (get_imsm_raid_level(map
) == 5)
2284 version
= MPB_VERSION_RAID5
;
2285 else if (map
->num_members
>= 3)
2286 version
= MPB_VERSION_3OR4_DISK_ARRAY
;
2287 else if (get_imsm_raid_level(map
) == 1)
2288 version
= MPB_VERSION_RAID1
;
2290 version
= MPB_VERSION_RAID0
;
2292 strcpy(((char *) mpb
->sig
) + strlen(MPB_SIGNATURE
), version
);
2296 static int init_super_imsm_volume(struct supertype
*st
, mdu_array_info_t
*info
,
2297 unsigned long long size
, char *name
,
2298 char *homehost
, int *uuid
)
2300 /* We are creating a volume inside a pre-existing container.
2301 * so st->sb is already set.
2303 struct intel_super
*super
= st
->sb
;
2304 struct imsm_super
*mpb
= super
->anchor
;
2305 struct intel_dev
*dv
;
2306 struct imsm_dev
*dev
;
2307 struct imsm_vol
*vol
;
2308 struct imsm_map
*map
;
2309 int idx
= mpb
->num_raid_devs
;
2311 unsigned long long array_blocks
;
2312 size_t size_old
, size_new
;
2313 __u32 num_data_stripes
;
2315 if (super
->orom
&& mpb
->num_raid_devs
>= super
->orom
->vpa
) {
2316 fprintf(stderr
, Name
": This imsm-container already has the "
2317 "maximum of %d volumes\n", super
->orom
->vpa
);
2321 /* ensure the mpb is large enough for the new data */
2322 size_old
= __le32_to_cpu(mpb
->mpb_size
);
2323 size_new
= disks_to_mpb_size(info
->nr_disks
);
2324 if (size_new
> size_old
) {
2326 size_t size_round
= ROUND_UP(size_new
, 512);
2328 if (posix_memalign(&mpb_new
, 512, size_round
) != 0) {
2329 fprintf(stderr
, Name
": could not allocate new mpb\n");
2332 memcpy(mpb_new
, mpb
, size_old
);
2335 super
->anchor
= mpb_new
;
2336 mpb
->mpb_size
= __cpu_to_le32(size_new
);
2337 memset(mpb_new
+ size_old
, 0, size_round
- size_old
);
2339 super
->current_vol
= idx
;
2340 /* when creating the first raid device in this container set num_disks
2341 * to zero, i.e. delete this spare and add raid member devices in
2342 * add_to_super_imsm_volume()
2344 if (super
->current_vol
== 0)
2347 for (i
= 0; i
< super
->current_vol
; i
++) {
2348 dev
= get_imsm_dev(super
, i
);
2349 if (strncmp((char *) dev
->volume
, name
,
2350 MAX_RAID_SERIAL_LEN
) == 0) {
2351 fprintf(stderr
, Name
": '%s' is already defined for this container\n",
2357 sprintf(st
->subarray
, "%d", idx
);
2358 dv
= malloc(sizeof(*dv
));
2360 fprintf(stderr
, Name
": failed to allocate device list entry\n");
2363 dev
= malloc(sizeof(*dev
) + sizeof(__u32
) * (info
->raid_disks
- 1));
2366 fprintf(stderr
, Name
": could not allocate raid device\n");
2369 strncpy((char *) dev
->volume
, name
, MAX_RAID_SERIAL_LEN
);
2370 if (info
->level
== 1)
2371 array_blocks
= info_to_blocks_per_member(info
);
2373 array_blocks
= calc_array_size(info
->level
, info
->raid_disks
,
2374 info
->layout
, info
->chunk_size
,
2376 dev
->size_low
= __cpu_to_le32((__u32
) array_blocks
);
2377 dev
->size_high
= __cpu_to_le32((__u32
) (array_blocks
>> 32));
2378 dev
->status
= __cpu_to_le32(0);
2379 dev
->reserved_blocks
= __cpu_to_le32(0);
2381 vol
->migr_state
= 0;
2382 set_migr_type(dev
, MIGR_INIT
);
2384 vol
->curr_migr_unit
= 0;
2385 map
= get_imsm_map(dev
, 0);
2386 map
->pba_of_lba0
= __cpu_to_le32(super
->create_offset
);
2387 map
->blocks_per_member
= __cpu_to_le32(info_to_blocks_per_member(info
));
2388 map
->blocks_per_strip
= __cpu_to_le16(info_to_blocks_per_strip(info
));
2389 map
->failed_disk_num
= ~0;
2390 map
->map_state
= info
->level
? IMSM_T_STATE_UNINITIALIZED
:
2391 IMSM_T_STATE_NORMAL
;
2393 if (info
->level
== 1 && info
->raid_disks
> 2) {
2394 fprintf(stderr
, Name
": imsm does not support more than 2 disks"
2395 "in a raid1 volume\n");
2398 if (info
->level
== 10) {
2399 map
->raid_level
= 1;
2400 map
->num_domains
= info
->raid_disks
/ 2;
2402 map
->raid_level
= info
->level
;
2403 map
->num_domains
= 1;
2405 num_data_stripes
= info_to_num_data_stripes(info
, map
->num_domains
);
2406 map
->num_data_stripes
= __cpu_to_le32(num_data_stripes
);
2408 map
->num_members
= info
->raid_disks
;
2409 for (i
= 0; i
< map
->num_members
; i
++) {
2410 /* initialized in add_to_super */
2411 set_imsm_ord_tbl_ent(map
, i
, 0);
2413 mpb
->num_raid_devs
++;
2416 dv
->index
= super
->current_vol
;
2417 dv
->next
= super
->devlist
;
2418 super
->devlist
= dv
;
2420 imsm_update_version_info(super
);
2425 static int init_super_imsm(struct supertype
*st
, mdu_array_info_t
*info
,
2426 unsigned long long size
, char *name
,
2427 char *homehost
, int *uuid
)
2429 /* This is primarily called by Create when creating a new array.
2430 * We will then get add_to_super called for each component, and then
2431 * write_init_super called to write it out to each device.
2432 * For IMSM, Create can create on fresh devices or on a pre-existing
2434 * To create on a pre-existing array a different method will be called.
2435 * This one is just for fresh drives.
2437 struct intel_super
*super
;
2438 struct imsm_super
*mpb
;
2447 return init_super_imsm_volume(st
, info
, size
, name
, homehost
,
2450 super
= alloc_super(1);
2453 mpb_size
= disks_to_mpb_size(info
->nr_disks
);
2454 if (posix_memalign(&super
->buf
, 512, mpb_size
) != 0) {
2459 memset(mpb
, 0, mpb_size
);
2461 mpb
->attributes
= MPB_ATTRIB_CHECKSUM_VERIFY
;
2463 version
= (char *) mpb
->sig
;
2464 strcpy(version
, MPB_SIGNATURE
);
2465 version
+= strlen(MPB_SIGNATURE
);
2466 strcpy(version
, MPB_VERSION_RAID0
);
2467 mpb
->mpb_size
= mpb_size
;
2474 static int add_to_super_imsm_volume(struct supertype
*st
, mdu_disk_info_t
*dk
,
2475 int fd
, char *devname
)
2477 struct intel_super
*super
= st
->sb
;
2478 struct imsm_super
*mpb
= super
->anchor
;
2480 struct imsm_dev
*dev
;
2481 struct imsm_map
*map
;
2483 dev
= get_imsm_dev(super
, super
->current_vol
);
2484 map
= get_imsm_map(dev
, 0);
2486 if (! (dk
->state
& (1<<MD_DISK_SYNC
))) {
2487 fprintf(stderr
, Name
": %s: Cannot add spare devices to IMSM volume\n",
2493 /* we're doing autolayout so grab the pre-marked (in
2494 * validate_geometry) raid_disk
2496 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
2497 if (dl
->raiddisk
== dk
->raid_disk
)
2500 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
2501 if (dl
->major
== dk
->major
&&
2502 dl
->minor
== dk
->minor
)
2507 fprintf(stderr
, Name
": %s is not a member of the same container\n", devname
);
2511 /* add a pristine spare to the metadata */
2512 if (dl
->index
< 0) {
2513 dl
->index
= super
->anchor
->num_disks
;
2514 super
->anchor
->num_disks
++;
2516 set_imsm_ord_tbl_ent(map
, dk
->number
, dl
->index
);
2517 dl
->disk
.status
= CONFIGURED_DISK
| USABLE_DISK
;
2519 /* if we are creating the first raid device update the family number */
2520 if (super
->current_vol
== 0) {
2522 struct imsm_dev
*_dev
= __get_imsm_dev(mpb
, 0);
2523 struct imsm_disk
*_disk
= __get_imsm_disk(mpb
, dl
->index
);
2527 sum
= __gen_imsm_checksum(mpb
);
2528 mpb
->family_num
= __cpu_to_le32(sum
);
2534 static int add_to_super_imsm(struct supertype
*st
, mdu_disk_info_t
*dk
,
2535 int fd
, char *devname
)
2537 struct intel_super
*super
= st
->sb
;
2539 unsigned long long size
;
2544 /* if we are on an RAID enabled platform check that the disk is
2545 * attached to the raid controller
2547 if (super
->hba
&& !disk_attached_to_hba(fd
, super
->hba
)) {
2549 Name
": %s is not attached to the raid controller: %s\n",
2550 devname
? : "disk", super
->hba
);
2554 if (super
->current_vol
>= 0)
2555 return add_to_super_imsm_volume(st
, dk
, fd
, devname
);
2558 dd
= malloc(sizeof(*dd
));
2561 Name
": malloc failed %s:%d.\n", __func__
, __LINE__
);
2564 memset(dd
, 0, sizeof(*dd
));
2565 dd
->major
= major(stb
.st_rdev
);
2566 dd
->minor
= minor(stb
.st_rdev
);
2568 dd
->devname
= devname
? strdup(devname
) : NULL
;
2571 rv
= imsm_read_serial(fd
, devname
, dd
->serial
);
2574 Name
": failed to retrieve scsi serial, aborting\n");
2579 get_dev_size(fd
, NULL
, &size
);
2581 serialcpy(dd
->disk
.serial
, dd
->serial
);
2582 dd
->disk
.total_blocks
= __cpu_to_le32(size
);
2583 dd
->disk
.status
= USABLE_DISK
| SPARE_DISK
;
2584 if (sysfs_disk_to_scsi_id(fd
, &id
) == 0)
2585 dd
->disk
.scsi_id
= __cpu_to_le32(id
);
2587 dd
->disk
.scsi_id
= __cpu_to_le32(0);
2589 if (st
->update_tail
) {
2590 dd
->next
= super
->add
;
2593 dd
->next
= super
->disks
;
2600 static int store_imsm_mpb(int fd
, struct intel_super
*super
);
2602 /* spare records have their own family number and do not have any defined raid
2605 static int write_super_imsm_spares(struct intel_super
*super
, int doclose
)
2607 struct imsm_super mpb_save
;
2608 struct imsm_super
*mpb
= super
->anchor
;
2613 mpb
->num_raid_devs
= 0;
2615 mpb
->mpb_size
= sizeof(struct imsm_super
);
2616 mpb
->generation_num
= __cpu_to_le32(1UL);
2618 for (d
= super
->disks
; d
; d
= d
->next
) {
2622 mpb
->disk
[0] = d
->disk
;
2623 sum
= __gen_imsm_checksum(mpb
);
2624 mpb
->family_num
= __cpu_to_le32(sum
);
2625 sum
= __gen_imsm_checksum(mpb
);
2626 mpb
->check_sum
= __cpu_to_le32(sum
);
2628 if (store_imsm_mpb(d
->fd
, super
)) {
2629 fprintf(stderr
, "%s: failed for device %d:%d %s\n",
2630 __func__
, d
->major
, d
->minor
, strerror(errno
));
2644 static int write_super_imsm(struct intel_super
*super
, int doclose
)
2646 struct imsm_super
*mpb
= super
->anchor
;
2652 __u32 mpb_size
= sizeof(struct imsm_super
) - sizeof(struct imsm_disk
);
2654 /* 'generation' is incremented everytime the metadata is written */
2655 generation
= __le32_to_cpu(mpb
->generation_num
);
2657 mpb
->generation_num
= __cpu_to_le32(generation
);
2659 mpb_size
+= sizeof(struct imsm_disk
) * mpb
->num_disks
;
2660 for (d
= super
->disks
; d
; d
= d
->next
) {
2664 mpb
->disk
[d
->index
] = d
->disk
;
2666 for (d
= super
->missing
; d
; d
= d
->next
)
2667 mpb
->disk
[d
->index
] = d
->disk
;
2669 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
2670 struct imsm_dev
*dev
= __get_imsm_dev(mpb
, i
);
2672 imsm_copy_dev(dev
, get_imsm_dev(super
, i
));
2673 mpb_size
+= sizeof_imsm_dev(dev
, 0);
2675 mpb_size
+= __le32_to_cpu(mpb
->bbm_log_size
);
2676 mpb
->mpb_size
= __cpu_to_le32(mpb_size
);
2678 /* recalculate checksum */
2679 sum
= __gen_imsm_checksum(mpb
);
2680 mpb
->check_sum
= __cpu_to_le32(sum
);
2682 /* write the mpb for disks that compose raid devices */
2683 for (d
= super
->disks
; d
; d
= d
->next
) {
2686 if (store_imsm_mpb(d
->fd
, super
))
2687 fprintf(stderr
, "%s: failed for device %d:%d %s\n",
2688 __func__
, d
->major
, d
->minor
, strerror(errno
));
2696 return write_super_imsm_spares(super
, doclose
);
2702 static int create_array(struct supertype
*st
)
2705 struct imsm_update_create_array
*u
;
2706 struct intel_super
*super
= st
->sb
;
2707 struct imsm_dev
*dev
= get_imsm_dev(super
, super
->current_vol
);
2708 struct imsm_map
*map
= get_imsm_map(dev
, 0);
2709 struct disk_info
*inf
;
2710 struct imsm_disk
*disk
;
2714 len
= sizeof(*u
) - sizeof(*dev
) + sizeof_imsm_dev(dev
, 0) +
2715 sizeof(*inf
) * map
->num_members
;
2718 fprintf(stderr
, "%s: failed to allocate update buffer\n",
2723 u
->type
= update_create_array
;
2724 u
->dev_idx
= super
->current_vol
;
2725 imsm_copy_dev(&u
->dev
, dev
);
2726 inf
= get_disk_info(u
);
2727 for (i
= 0; i
< map
->num_members
; i
++) {
2728 idx
= get_imsm_disk_idx(dev
, i
);
2729 disk
= get_imsm_disk(super
, idx
);
2730 serialcpy(inf
[i
].serial
, disk
->serial
);
2732 append_metadata_update(st
, u
, len
);
2737 static int _add_disk(struct supertype
*st
)
2739 struct intel_super
*super
= st
->sb
;
2741 struct imsm_update_add_disk
*u
;
2749 fprintf(stderr
, "%s: failed to allocate update buffer\n",
2754 u
->type
= update_add_disk
;
2755 append_metadata_update(st
, u
, len
);
2760 static int write_init_super_imsm(struct supertype
*st
)
2762 if (st
->update_tail
) {
2763 /* queue the recently created array / added disk
2764 * as a metadata update */
2765 struct intel_super
*super
= st
->sb
;
2769 /* determine if we are creating a volume or adding a disk */
2770 if (super
->current_vol
< 0) {
2771 /* in the add disk case we are running in mdmon
2772 * context, so don't close fd's
2774 return _add_disk(st
);
2776 rv
= create_array(st
);
2778 for (d
= super
->disks
; d
; d
= d
->next
) {
2785 return write_super_imsm(st
->sb
, 1);
2789 static int store_zero_imsm(struct supertype
*st
, int fd
)
2791 unsigned long long dsize
;
2794 get_dev_size(fd
, NULL
, &dsize
);
2796 /* first block is stored on second to last sector of the disk */
2797 if (lseek64(fd
, dsize
- (512 * 2), SEEK_SET
) < 0)
2800 if (posix_memalign(&buf
, 512, 512) != 0)
2803 memset(buf
, 0, 512);
2804 if (write(fd
, buf
, 512) != 512)
2809 static int imsm_bbm_log_size(struct imsm_super
*mpb
)
2811 return __le32_to_cpu(mpb
->bbm_log_size
);
2815 static int validate_geometry_imsm_container(struct supertype
*st
, int level
,
2816 int layout
, int raiddisks
, int chunk
,
2817 unsigned long long size
, char *dev
,
2818 unsigned long long *freesize
,
2822 unsigned long long ldsize
;
2823 const struct imsm_orom
*orom
;
2825 if (level
!= LEVEL_CONTAINER
)
2830 if (check_env("IMSM_NO_PLATFORM"))
2833 orom
= find_imsm_orom();
2834 if (orom
&& raiddisks
> orom
->tds
) {
2836 fprintf(stderr
, Name
": %d exceeds maximum number of"
2837 " platform supported disks: %d\n",
2838 raiddisks
, orom
->tds
);
2842 fd
= open(dev
, O_RDONLY
|O_EXCL
, 0);
2845 fprintf(stderr
, Name
": imsm: Cannot open %s: %s\n",
2846 dev
, strerror(errno
));
2849 if (!get_dev_size(fd
, dev
, &ldsize
)) {
2855 *freesize
= avail_size_imsm(st
, ldsize
>> 9);
2860 static unsigned long long find_size(struct extent
*e
, int *idx
, int num_extents
)
2862 const unsigned long long base_start
= e
[*idx
].start
;
2863 unsigned long long end
= base_start
+ e
[*idx
].size
;
2866 if (base_start
== end
)
2870 for (i
= *idx
; i
< num_extents
; i
++) {
2871 /* extend overlapping extents */
2872 if (e
[i
].start
>= base_start
&&
2873 e
[i
].start
<= end
) {
2876 if (e
[i
].start
+ e
[i
].size
> end
)
2877 end
= e
[i
].start
+ e
[i
].size
;
2878 } else if (e
[i
].start
> end
) {
2884 return end
- base_start
;
2887 static unsigned long long merge_extents(struct intel_super
*super
, int sum_extents
)
2889 /* build a composite disk with all known extents and generate a new
2890 * 'maxsize' given the "all disks in an array must share a common start
2891 * offset" constraint
2893 struct extent
*e
= calloc(sum_extents
, sizeof(*e
));
2897 unsigned long long pos
;
2898 unsigned long long start
= 0;
2899 unsigned long long maxsize
;
2900 unsigned long reserve
;
2903 return ~0ULL; /* error */
2905 /* coalesce and sort all extents. also, check to see if we need to
2906 * reserve space between member arrays
2909 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
2912 for (i
= 0; i
< dl
->extent_cnt
; i
++)
2915 qsort(e
, sum_extents
, sizeof(*e
), cmp_extent
);
2920 while (i
< sum_extents
) {
2921 e
[j
].start
= e
[i
].start
;
2922 e
[j
].size
= find_size(e
, &i
, sum_extents
);
2924 if (e
[j
-1].size
== 0)
2933 unsigned long long esize
;
2935 esize
= e
[i
].start
- pos
;
2936 if (esize
>= maxsize
) {
2941 pos
= e
[i
].start
+ e
[i
].size
;
2943 } while (e
[i
-1].size
);
2946 if (start_extent
> 0)
2947 reserve
= IMSM_RESERVED_SECTORS
; /* gap between raid regions */
2951 if (maxsize
< reserve
)
2954 super
->create_offset
= ~((__u32
) 0);
2955 if (start
+ reserve
> super
->create_offset
)
2956 return ~0ULL; /* start overflows create_offset */
2957 super
->create_offset
= start
+ reserve
;
2959 return maxsize
- reserve
;
2962 static int is_raid_level_supported(const struct imsm_orom
*orom
, int level
, int raiddisks
)
2964 if (level
< 0 || level
== 6 || level
== 4)
2967 /* if we have an orom prevent invalid raid levels */
2970 case 0: return imsm_orom_has_raid0(orom
);
2973 return imsm_orom_has_raid1e(orom
);
2974 return imsm_orom_has_raid1(orom
) && raiddisks
== 2;
2975 case 10: return imsm_orom_has_raid10(orom
) && raiddisks
== 4;
2976 case 5: return imsm_orom_has_raid5(orom
) && raiddisks
> 2;
2979 return 1; /* not on an Intel RAID platform so anything goes */
2984 #define pr_vrb(fmt, arg...) (void) (verbose && fprintf(stderr, Name fmt, ##arg))
2985 /* validate_geometry_imsm_volume - lifted from validate_geometry_ddf_bvd
2986 * FIX ME add ahci details
2988 static int validate_geometry_imsm_volume(struct supertype
*st
, int level
,
2989 int layout
, int raiddisks
, int chunk
,
2990 unsigned long long size
, char *dev
,
2991 unsigned long long *freesize
,
2995 struct intel_super
*super
= st
->sb
;
2996 struct imsm_super
*mpb
= super
->anchor
;
2998 unsigned long long pos
= 0;
2999 unsigned long long maxsize
;
3003 /* We must have the container info already read in. */
3007 if (!is_raid_level_supported(super
->orom
, level
, raiddisks
)) {
3008 pr_vrb(": platform does not support raid%d with %d disk%s\n",
3009 level
, raiddisks
, raiddisks
> 1 ? "s" : "");
3012 if (super
->orom
&& level
!= 1 &&
3013 !imsm_orom_has_chunk(super
->orom
, chunk
)) {
3014 pr_vrb(": platform does not support a chunk size of: %d\n", chunk
);
3017 if (layout
!= imsm_level_to_layout(level
)) {
3019 pr_vrb(": imsm raid 5 only supports the left-asymmetric layout\n");
3020 else if (level
== 10)
3021 pr_vrb(": imsm raid 10 only supports the n2 layout\n");
3023 pr_vrb(": imsm unknown layout %#x for this raid level %d\n",
3029 /* General test: make sure there is space for
3030 * 'raiddisks' device extents of size 'size' at a given
3033 unsigned long long minsize
= size
;
3034 unsigned long long start_offset
= ~0ULL;
3037 minsize
= MPB_SECTOR_CNT
+ IMSM_RESERVED_SECTORS
;
3038 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
3043 e
= get_extents(super
, dl
);
3046 unsigned long long esize
;
3047 esize
= e
[i
].start
- pos
;
3048 if (esize
>= minsize
)
3050 if (found
&& start_offset
== ~0ULL) {
3053 } else if (found
&& pos
!= start_offset
) {
3057 pos
= e
[i
].start
+ e
[i
].size
;
3059 } while (e
[i
-1].size
);
3064 if (dcnt
< raiddisks
) {
3066 fprintf(stderr
, Name
": imsm: Not enough "
3067 "devices with space for this array "
3075 /* This device must be a member of the set */
3076 if (stat(dev
, &stb
) < 0)
3078 if ((S_IFMT
& stb
.st_mode
) != S_IFBLK
)
3080 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
3081 if (dl
->major
== major(stb
.st_rdev
) &&
3082 dl
->minor
== minor(stb
.st_rdev
))
3087 fprintf(stderr
, Name
": %s is not in the "
3088 "same imsm set\n", dev
);
3090 } else if (super
->orom
&& dl
->index
< 0 && mpb
->num_raid_devs
) {
3091 /* If a volume is present then the current creation attempt
3092 * cannot incorporate new spares because the orom may not
3093 * understand this configuration (all member disks must be
3094 * members of each array in the container).
3096 fprintf(stderr
, Name
": %s is a spare and a volume"
3097 " is already defined for this container\n", dev
);
3098 fprintf(stderr
, Name
": The option-rom requires all member"
3099 " disks to be a member of all volumes\n");
3103 /* retrieve the largest free space block */
3104 e
= get_extents(super
, dl
);
3109 unsigned long long esize
;
3111 esize
= e
[i
].start
- pos
;
3112 if (esize
>= maxsize
)
3114 pos
= e
[i
].start
+ e
[i
].size
;
3116 } while (e
[i
-1].size
);
3121 fprintf(stderr
, Name
": unable to determine free space for: %s\n",
3125 if (maxsize
< size
) {
3127 fprintf(stderr
, Name
": %s not enough space (%llu < %llu)\n",
3128 dev
, maxsize
, size
);
3132 /* count total number of extents for merge */
3134 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
3136 i
+= dl
->extent_cnt
;
3138 maxsize
= merge_extents(super
, i
);
3139 if (maxsize
< size
) {
3141 fprintf(stderr
, Name
": not enough space after merge (%llu < %llu)\n",
3144 } else if (maxsize
== ~0ULL) {
3146 fprintf(stderr
, Name
": failed to merge %d extents\n", i
);
3150 *freesize
= maxsize
;
3155 static int reserve_space(struct supertype
*st
, int raiddisks
,
3156 unsigned long long size
, int chunk
,
3157 unsigned long long *freesize
)
3159 struct intel_super
*super
= st
->sb
;
3160 struct imsm_super
*mpb
= super
->anchor
;
3165 unsigned long long maxsize
;
3166 unsigned long long minsize
;
3170 /* find the largest common start free region of the possible disks */
3174 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
3180 /* don't activate new spares if we are orom constrained
3181 * and there is already a volume active in the container
3183 if (super
->orom
&& dl
->index
< 0 && mpb
->num_raid_devs
)
3186 e
= get_extents(super
, dl
);
3189 for (i
= 1; e
[i
-1].size
; i
++)
3197 maxsize
= merge_extents(super
, extent_cnt
);
3202 if (cnt
< raiddisks
||
3203 (super
->orom
&& used
&& used
!= raiddisks
) ||
3204 maxsize
< minsize
) {
3205 fprintf(stderr
, Name
": not enough devices with space to create array.\n");
3206 return 0; /* No enough free spaces large enough */
3218 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
3220 dl
->raiddisk
= cnt
++;
3227 static int validate_geometry_imsm(struct supertype
*st
, int level
, int layout
,
3228 int raiddisks
, int chunk
, unsigned long long size
,
3229 char *dev
, unsigned long long *freesize
,
3235 /* if given unused devices create a container
3236 * if given given devices in a container create a member volume
3238 if (level
== LEVEL_CONTAINER
) {
3239 /* Must be a fresh device to add to a container */
3240 return validate_geometry_imsm_container(st
, level
, layout
,
3241 raiddisks
, chunk
, size
,
3247 if (st
->sb
&& freesize
) {
3248 /* we are being asked to automatically layout a
3249 * new volume based on the current contents of
3250 * the container. If the the parameters can be
3251 * satisfied reserve_space will record the disks,
3252 * start offset, and size of the volume to be
3253 * created. add_to_super and getinfo_super
3254 * detect when autolayout is in progress.
3256 return reserve_space(st
, raiddisks
, size
, chunk
, freesize
);
3261 /* creating in a given container */
3262 return validate_geometry_imsm_volume(st
, level
, layout
,
3263 raiddisks
, chunk
, size
,
3264 dev
, freesize
, verbose
);
3267 /* limit creation to the following levels */
3279 /* This device needs to be a device in an 'imsm' container */
3280 fd
= open(dev
, O_RDONLY
|O_EXCL
, 0);
3284 Name
": Cannot create this array on device %s\n",
3289 if (errno
!= EBUSY
|| (fd
= open(dev
, O_RDONLY
, 0)) < 0) {
3291 fprintf(stderr
, Name
": Cannot open %s: %s\n",
3292 dev
, strerror(errno
));
3295 /* Well, it is in use by someone, maybe an 'imsm' container. */
3296 cfd
= open_container(fd
);
3300 fprintf(stderr
, Name
": Cannot use %s: It is busy\n",
3304 sra
= sysfs_read(cfd
, 0, GET_VERSION
);
3306 if (sra
&& sra
->array
.major_version
== -1 &&
3307 strcmp(sra
->text_version
, "imsm") == 0) {
3308 /* This is a member of a imsm container. Load the container
3309 * and try to create a volume
3311 struct intel_super
*super
;
3313 if (load_super_imsm_all(st
, cfd
, (void **) &super
, NULL
, 1) == 0) {
3315 st
->container_dev
= fd2devnum(cfd
);
3317 return validate_geometry_imsm_volume(st
, level
, layout
,
3323 } else /* may belong to another container */
3328 #endif /* MDASSEMBLE */
3330 static struct mdinfo
*container_content_imsm(struct supertype
*st
)
3332 /* Given a container loaded by load_super_imsm_all,
3333 * extract information about all the arrays into
3336 * For each imsm_dev create an mdinfo, fill it in,
3337 * then look for matching devices in super->disks
3338 * and create appropriate device mdinfo.
3340 struct intel_super
*super
= st
->sb
;
3341 struct imsm_super
*mpb
= super
->anchor
;
3342 struct mdinfo
*rest
= NULL
;
3345 /* do not assemble arrays that might have bad blocks */
3346 if (imsm_bbm_log_size(super
->anchor
)) {
3347 fprintf(stderr
, Name
": BBM log found in metadata. "
3348 "Cannot activate array(s).\n");
3352 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
3353 struct imsm_dev
*dev
= get_imsm_dev(super
, i
);
3354 struct imsm_map
*map
= get_imsm_map(dev
, 0);
3355 struct mdinfo
*this;
3358 this = malloc(sizeof(*this));
3359 memset(this, 0, sizeof(*this));
3362 super
->current_vol
= i
;
3363 getinfo_super_imsm_volume(st
, this);
3364 for (slot
= 0 ; slot
< map
->num_members
; slot
++) {
3365 struct mdinfo
*info_d
;
3373 idx
= get_imsm_disk_idx(dev
, slot
);
3374 ord
= get_imsm_ord_tbl_ent(dev
, slot
);
3375 for (d
= super
->disks
; d
; d
= d
->next
)
3376 if (d
->index
== idx
)
3382 s
= d
? d
->disk
.status
: 0;
3383 if (s
& FAILED_DISK
)
3385 if (!(s
& USABLE_DISK
))
3387 if (ord
& IMSM_ORD_REBUILD
)
3391 * if we skip some disks the array will be assmebled degraded;
3392 * reset resync start to avoid a dirty-degraded situation
3394 * FIXME handle dirty degraded
3396 if (skip
&& !dev
->vol
.dirty
)
3397 this->resync_start
= ~0ULL;
3401 info_d
= malloc(sizeof(*info_d
));
3403 fprintf(stderr
, Name
": failed to allocate disk"
3404 " for volume %s\n", (char *) dev
->volume
);
3409 memset(info_d
, 0, sizeof(*info_d
));
3410 info_d
->next
= this->devs
;
3411 this->devs
= info_d
;
3413 info_d
->disk
.number
= d
->index
;
3414 info_d
->disk
.major
= d
->major
;
3415 info_d
->disk
.minor
= d
->minor
;
3416 info_d
->disk
.raid_disk
= slot
;
3418 this->array
.working_disks
++;
3420 info_d
->events
= __le32_to_cpu(mpb
->generation_num
);
3421 info_d
->data_offset
= __le32_to_cpu(map
->pba_of_lba0
);
3422 info_d
->component_size
= __le32_to_cpu(map
->blocks_per_member
);
3424 strcpy(info_d
->name
, d
->devname
);
3434 static int imsm_open_new(struct supertype
*c
, struct active_array
*a
,
3437 struct intel_super
*super
= c
->sb
;
3438 struct imsm_super
*mpb
= super
->anchor
;
3440 if (atoi(inst
) >= mpb
->num_raid_devs
) {
3441 fprintf(stderr
, "%s: subarry index %d, out of range\n",
3442 __func__
, atoi(inst
));
3446 dprintf("imsm: open_new %s\n", inst
);
3447 a
->info
.container_member
= atoi(inst
);
3451 static __u8
imsm_check_degraded(struct intel_super
*super
, struct imsm_dev
*dev
, int failed
)
3453 struct imsm_map
*map
= get_imsm_map(dev
, 0);
3456 return map
->map_state
== IMSM_T_STATE_UNINITIALIZED
?
3457 IMSM_T_STATE_UNINITIALIZED
: IMSM_T_STATE_NORMAL
;
3459 switch (get_imsm_raid_level(map
)) {
3461 return IMSM_T_STATE_FAILED
;
3464 if (failed
< map
->num_members
)
3465 return IMSM_T_STATE_DEGRADED
;
3467 return IMSM_T_STATE_FAILED
;
3472 * check to see if any mirrors have failed, otherwise we
3473 * are degraded. Even numbered slots are mirrored on
3477 /* gcc -Os complains that this is unused */
3478 int insync
= insync
;
3480 for (i
= 0; i
< map
->num_members
; i
++) {
3481 __u32 ord
= get_imsm_ord_tbl_ent(dev
, i
);
3482 int idx
= ord_to_idx(ord
);
3483 struct imsm_disk
*disk
;
3485 /* reset the potential in-sync count on even-numbered
3486 * slots. num_copies is always 2 for imsm raid10
3491 disk
= get_imsm_disk(super
, idx
);
3492 if (!disk
|| disk
->status
& FAILED_DISK
||
3493 ord
& IMSM_ORD_REBUILD
)
3496 /* no in-sync disks left in this mirror the
3500 return IMSM_T_STATE_FAILED
;
3503 return IMSM_T_STATE_DEGRADED
;
3507 return IMSM_T_STATE_DEGRADED
;
3509 return IMSM_T_STATE_FAILED
;
3515 return map
->map_state
;
3518 static int imsm_count_failed(struct intel_super
*super
, struct imsm_dev
*dev
)
3522 struct imsm_disk
*disk
;
3523 struct imsm_map
*map
= get_imsm_map(dev
, 0);
3524 struct imsm_map
*prev
= get_imsm_map(dev
, dev
->vol
.migr_state
);
3528 /* at the beginning of migration we set IMSM_ORD_REBUILD on
3529 * disks that are being rebuilt. New failures are recorded to
3530 * map[0]. So we look through all the disks we started with and
3531 * see if any failures are still present, or if any new ones
3534 * FIXME add support for online capacity expansion and
3535 * raid-level-migration
3537 for (i
= 0; i
< prev
->num_members
; i
++) {
3538 ord
= __le32_to_cpu(prev
->disk_ord_tbl
[i
]);
3539 ord
|= __le32_to_cpu(map
->disk_ord_tbl
[i
]);
3540 idx
= ord_to_idx(ord
);
3542 disk
= get_imsm_disk(super
, idx
);
3543 if (!disk
|| disk
->status
& FAILED_DISK
||
3544 ord
& IMSM_ORD_REBUILD
)
3551 static int is_resyncing(struct imsm_dev
*dev
)
3553 struct imsm_map
*migr_map
;
3555 if (!dev
->vol
.migr_state
)
3558 if (migr_type(dev
) == MIGR_INIT
||
3559 migr_type(dev
) == MIGR_REPAIR
)
3562 migr_map
= get_imsm_map(dev
, 1);
3564 if (migr_map
->map_state
== IMSM_T_STATE_NORMAL
)
3570 static int is_rebuilding(struct imsm_dev
*dev
)
3572 struct imsm_map
*migr_map
;
3574 if (!dev
->vol
.migr_state
)
3577 if (migr_type(dev
) != MIGR_REBUILD
)
3580 migr_map
= get_imsm_map(dev
, 1);
3582 if (migr_map
->map_state
== IMSM_T_STATE_DEGRADED
)
3588 /* return true if we recorded new information */
3589 static int mark_failure(struct imsm_dev
*dev
, struct imsm_disk
*disk
, int idx
)
3593 struct imsm_map
*map
;
3595 /* new failures are always set in map[0] */
3596 map
= get_imsm_map(dev
, 0);
3598 slot
= get_imsm_disk_slot(map
, idx
);
3602 ord
= __le32_to_cpu(map
->disk_ord_tbl
[slot
]);
3603 if ((disk
->status
& FAILED_DISK
) && (ord
& IMSM_ORD_REBUILD
))
3606 disk
->status
|= FAILED_DISK
;
3607 set_imsm_ord_tbl_ent(map
, slot
, idx
| IMSM_ORD_REBUILD
);
3608 if (map
->failed_disk_num
== ~0)
3609 map
->failed_disk_num
= slot
;
3613 static void mark_missing(struct imsm_dev
*dev
, struct imsm_disk
*disk
, int idx
)
3615 mark_failure(dev
, disk
, idx
);
3617 if (disk
->scsi_id
== __cpu_to_le32(~(__u32
)0))
3620 disk
->scsi_id
= __cpu_to_le32(~(__u32
)0);
3621 memmove(&disk
->serial
[0], &disk
->serial
[1], MAX_RAID_SERIAL_LEN
- 1);
3624 /* Handle dirty -> clean transititions and resync. Degraded and rebuild
3625 * states are handled in imsm_set_disk() with one exception, when a
3626 * resync is stopped due to a new failure this routine will set the
3627 * 'degraded' state for the array.
3629 static int imsm_set_array_state(struct active_array
*a
, int consistent
)
3631 int inst
= a
->info
.container_member
;
3632 struct intel_super
*super
= a
->container
->sb
;
3633 struct imsm_dev
*dev
= get_imsm_dev(super
, inst
);
3634 struct imsm_map
*map
= get_imsm_map(dev
, 0);
3635 int failed
= imsm_count_failed(super
, dev
);
3636 __u8 map_state
= imsm_check_degraded(super
, dev
, failed
);
3638 /* before we activate this array handle any missing disks */
3639 if (consistent
== 2 && super
->missing
) {
3642 dprintf("imsm: mark missing\n");
3643 end_migration(dev
, map_state
);
3644 for (dl
= super
->missing
; dl
; dl
= dl
->next
)
3645 mark_missing(dev
, &dl
->disk
, dl
->index
);
3646 super
->updates_pending
++;
3649 if (consistent
== 2 &&
3650 (!is_resync_complete(a
) ||
3651 map_state
!= IMSM_T_STATE_NORMAL
||
3652 dev
->vol
.migr_state
))
3655 if (is_resync_complete(a
)) {
3656 /* complete intialization / resync,
3657 * recovery and interrupted recovery is completed in
3660 if (is_resyncing(dev
)) {
3661 dprintf("imsm: mark resync done\n");
3662 end_migration(dev
, map_state
);
3663 super
->updates_pending
++;
3665 } else if (!is_resyncing(dev
) && !failed
) {
3666 /* mark the start of the init process if nothing is failed */
3667 dprintf("imsm: mark resync start (%llu)\n", a
->resync_start
);
3668 if (map
->map_state
== IMSM_T_STATE_UNINITIALIZED
)
3669 migrate(dev
, IMSM_T_STATE_NORMAL
, MIGR_INIT
);
3671 migrate(dev
, IMSM_T_STATE_NORMAL
, MIGR_REPAIR
);
3672 super
->updates_pending
++;
3675 /* check if we can update the migration checkpoint */
3676 if (dev
->vol
.migr_state
&&
3677 __le32_to_cpu(dev
->vol
.curr_migr_unit
) != a
->resync_start
) {
3678 dprintf("imsm: checkpoint migration (%llu)\n", a
->resync_start
);
3679 dev
->vol
.curr_migr_unit
= __cpu_to_le32(a
->resync_start
);
3680 super
->updates_pending
++;
3683 /* mark dirty / clean */
3684 if (dev
->vol
.dirty
!= !consistent
) {
3685 dprintf("imsm: mark '%s' (%llu)\n",
3686 consistent
? "clean" : "dirty", a
->resync_start
);
3691 super
->updates_pending
++;
3696 static void imsm_set_disk(struct active_array
*a
, int n
, int state
)
3698 int inst
= a
->info
.container_member
;
3699 struct intel_super
*super
= a
->container
->sb
;
3700 struct imsm_dev
*dev
= get_imsm_dev(super
, inst
);
3701 struct imsm_map
*map
= get_imsm_map(dev
, 0);
3702 struct imsm_disk
*disk
;
3707 if (n
> map
->num_members
)
3708 fprintf(stderr
, "imsm: set_disk %d out of range 0..%d\n",
3709 n
, map
->num_members
- 1);
3714 dprintf("imsm: set_disk %d:%x\n", n
, state
);
3716 ord
= get_imsm_ord_tbl_ent(dev
, n
);
3717 disk
= get_imsm_disk(super
, ord_to_idx(ord
));
3719 /* check for new failures */
3720 if (state
& DS_FAULTY
) {
3721 if (mark_failure(dev
, disk
, ord_to_idx(ord
)))
3722 super
->updates_pending
++;
3725 /* check if in_sync */
3726 if (state
& DS_INSYNC
&& ord
& IMSM_ORD_REBUILD
&& is_rebuilding(dev
)) {
3727 struct imsm_map
*migr_map
= get_imsm_map(dev
, 1);
3729 set_imsm_ord_tbl_ent(migr_map
, n
, ord_to_idx(ord
));
3730 super
->updates_pending
++;
3733 failed
= imsm_count_failed(super
, dev
);
3734 map_state
= imsm_check_degraded(super
, dev
, failed
);
3736 /* check if recovery complete, newly degraded, or failed */
3737 if (map_state
== IMSM_T_STATE_NORMAL
&& is_rebuilding(dev
)) {
3738 end_migration(dev
, map_state
);
3739 map
= get_imsm_map(dev
, 0);
3740 map
->failed_disk_num
= ~0;
3741 super
->updates_pending
++;
3742 } else if (map_state
== IMSM_T_STATE_DEGRADED
&&
3743 map
->map_state
!= map_state
&&
3744 !dev
->vol
.migr_state
) {
3745 dprintf("imsm: mark degraded\n");
3746 map
->map_state
= map_state
;
3747 super
->updates_pending
++;
3748 } else if (map_state
== IMSM_T_STATE_FAILED
&&
3749 map
->map_state
!= map_state
) {
3750 dprintf("imsm: mark failed\n");
3751 end_migration(dev
, map_state
);
3752 super
->updates_pending
++;
3756 static int store_imsm_mpb(int fd
, struct intel_super
*super
)
3758 struct imsm_super
*mpb
= super
->anchor
;
3759 __u32 mpb_size
= __le32_to_cpu(mpb
->mpb_size
);
3760 unsigned long long dsize
;
3761 unsigned long long sectors
;
3763 get_dev_size(fd
, NULL
, &dsize
);
3765 if (mpb_size
> 512) {
3766 /* -1 to account for anchor */
3767 sectors
= mpb_sectors(mpb
) - 1;
3769 /* write the extended mpb to the sectors preceeding the anchor */
3770 if (lseek64(fd
, dsize
- (512 * (2 + sectors
)), SEEK_SET
) < 0)
3773 if (write(fd
, super
->buf
+ 512, 512 * sectors
) != 512 * sectors
)
3777 /* first block is stored on second to last sector of the disk */
3778 if (lseek64(fd
, dsize
- (512 * 2), SEEK_SET
) < 0)
3781 if (write(fd
, super
->buf
, 512) != 512)
3787 static void imsm_sync_metadata(struct supertype
*container
)
3789 struct intel_super
*super
= container
->sb
;
3791 if (!super
->updates_pending
)
3794 write_super_imsm(super
, 0);
3796 super
->updates_pending
= 0;
3799 static struct dl
*imsm_readd(struct intel_super
*super
, int idx
, struct active_array
*a
)
3801 struct imsm_dev
*dev
= get_imsm_dev(super
, a
->info
.container_member
);
3802 int i
= get_imsm_disk_idx(dev
, idx
);
3805 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
3809 if (dl
&& dl
->disk
.status
& FAILED_DISK
)
3813 dprintf("%s: found %x:%x\n", __func__
, dl
->major
, dl
->minor
);
3818 static struct dl
*imsm_add_spare(struct intel_super
*super
, int slot
,
3819 struct active_array
*a
, int activate_new
)
3821 struct imsm_dev
*dev
= get_imsm_dev(super
, a
->info
.container_member
);
3822 int idx
= get_imsm_disk_idx(dev
, slot
);
3823 struct imsm_super
*mpb
= super
->anchor
;
3824 struct imsm_map
*map
;
3825 unsigned long long esize
;
3826 unsigned long long pos
;
3835 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
3836 /* If in this array, skip */
3837 for (d
= a
->info
.devs
; d
; d
= d
->next
)
3838 if (d
->state_fd
>= 0 &&
3839 d
->disk
.major
== dl
->major
&&
3840 d
->disk
.minor
== dl
->minor
) {
3841 dprintf("%x:%x already in array\n", dl
->major
, dl
->minor
);
3847 /* skip in use or failed drives */
3848 if (dl
->disk
.status
& FAILED_DISK
|| idx
== dl
->index
||
3850 dprintf("%x:%x status (failed: %d index: %d)\n",
3851 dl
->major
, dl
->minor
,
3852 (dl
->disk
.status
& FAILED_DISK
) == FAILED_DISK
, idx
);
3856 /* skip pure spares when we are looking for partially
3857 * assimilated drives
3859 if (dl
->index
== -1 && !activate_new
)
3862 /* Does this unused device have the requisite free space?
3863 * It needs to be able to cover all member volumes
3865 ex
= get_extents(super
, dl
);
3867 dprintf("cannot get extents\n");
3870 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
3871 dev
= get_imsm_dev(super
, i
);
3872 map
= get_imsm_map(dev
, 0);
3874 /* check if this disk is already a member of
3877 if (get_imsm_disk_slot(map
, dl
->index
) >= 0)
3883 array_start
= __le32_to_cpu(map
->pba_of_lba0
);
3884 blocks
= __le32_to_cpu(map
->blocks_per_member
);
3887 /* check that we can start at pba_of_lba0 with
3888 * blocks_per_member of space
3890 esize
= ex
[j
].start
- pos
;
3891 if (array_start
>= pos
&&
3892 array_start
+ blocks
< ex
[j
].start
) {
3896 pos
= ex
[j
].start
+ ex
[j
].size
;
3898 } while (ex
[j
-1].size
);
3905 if (i
< mpb
->num_raid_devs
) {
3906 dprintf("%x:%x does not have %u at %u\n",
3907 dl
->major
, dl
->minor
,
3908 blocks
, array_start
);
3918 static struct mdinfo
*imsm_activate_spare(struct active_array
*a
,
3919 struct metadata_update
**updates
)
3922 * Find a device with unused free space and use it to replace a
3923 * failed/vacant region in an array. We replace failed regions one a
3924 * array at a time. The result is that a new spare disk will be added
3925 * to the first failed array and after the monitor has finished
3926 * propagating failures the remainder will be consumed.
3928 * FIXME add a capability for mdmon to request spares from another
3932 struct intel_super
*super
= a
->container
->sb
;
3933 int inst
= a
->info
.container_member
;
3934 struct imsm_dev
*dev
= get_imsm_dev(super
, inst
);
3935 struct imsm_map
*map
= get_imsm_map(dev
, 0);
3936 int failed
= a
->info
.array
.raid_disks
;
3937 struct mdinfo
*rv
= NULL
;
3940 struct metadata_update
*mu
;
3942 struct imsm_update_activate_spare
*u
;
3946 for (d
= a
->info
.devs
; d
; d
= d
->next
) {
3947 if ((d
->curr_state
& DS_FAULTY
) &&
3949 /* wait for Removal to happen */
3951 if (d
->state_fd
>= 0)
3955 dprintf("imsm: activate spare: inst=%d failed=%d (%d) level=%d\n",
3956 inst
, failed
, a
->info
.array
.raid_disks
, a
->info
.array
.level
);
3957 if (imsm_check_degraded(super
, dev
, failed
) != IMSM_T_STATE_DEGRADED
)
3960 /* For each slot, if it is not working, find a spare */
3961 for (i
= 0; i
< a
->info
.array
.raid_disks
; i
++) {
3962 for (d
= a
->info
.devs
; d
; d
= d
->next
)
3963 if (d
->disk
.raid_disk
== i
)
3965 dprintf("found %d: %p %x\n", i
, d
, d
?d
->curr_state
:0);
3966 if (d
&& (d
->state_fd
>= 0))
3970 * OK, this device needs recovery. Try to re-add the
3971 * previous occupant of this slot, if this fails see if
3972 * we can continue the assimilation of a spare that was
3973 * partially assimilated, finally try to activate a new
3976 dl
= imsm_readd(super
, i
, a
);
3978 dl
= imsm_add_spare(super
, i
, a
, 0);
3980 dl
= imsm_add_spare(super
, i
, a
, 1);
3984 /* found a usable disk with enough space */
3985 di
= malloc(sizeof(*di
));
3988 memset(di
, 0, sizeof(*di
));
3990 /* dl->index will be -1 in the case we are activating a
3991 * pristine spare. imsm_process_update() will create a
3992 * new index in this case. Once a disk is found to be
3993 * failed in all member arrays it is kicked from the
3996 di
->disk
.number
= dl
->index
;
3998 /* (ab)use di->devs to store a pointer to the device
4001 di
->devs
= (struct mdinfo
*) dl
;
4003 di
->disk
.raid_disk
= i
;
4004 di
->disk
.major
= dl
->major
;
4005 di
->disk
.minor
= dl
->minor
;
4007 di
->data_offset
= __le32_to_cpu(map
->pba_of_lba0
);
4008 di
->component_size
= a
->info
.component_size
;
4009 di
->container_member
= inst
;
4013 dprintf("%x:%x to be %d at %llu\n", dl
->major
, dl
->minor
,
4014 i
, di
->data_offset
);
4020 /* No spares found */
4022 /* Now 'rv' has a list of devices to return.
4023 * Create a metadata_update record to update the
4024 * disk_ord_tbl for the array
4026 mu
= malloc(sizeof(*mu
));
4028 mu
->buf
= malloc(sizeof(struct imsm_update_activate_spare
) * num_spares
);
4029 if (mu
->buf
== NULL
) {
4036 struct mdinfo
*n
= rv
->next
;
4045 mu
->len
= sizeof(struct imsm_update_activate_spare
) * num_spares
;
4046 mu
->next
= *updates
;
4047 u
= (struct imsm_update_activate_spare
*) mu
->buf
;
4049 for (di
= rv
; di
; di
= di
->next
) {
4050 u
->type
= update_activate_spare
;
4051 u
->dl
= (struct dl
*) di
->devs
;
4053 u
->slot
= di
->disk
.raid_disk
;
4064 static int disks_overlap(struct intel_super
*super
, int idx
, struct imsm_update_create_array
*u
)
4066 struct imsm_dev
*dev
= get_imsm_dev(super
, idx
);
4067 struct imsm_map
*map
= get_imsm_map(dev
, 0);
4068 struct imsm_map
*new_map
= get_imsm_map(&u
->dev
, 0);
4069 struct disk_info
*inf
= get_disk_info(u
);
4070 struct imsm_disk
*disk
;
4074 for (i
= 0; i
< map
->num_members
; i
++) {
4075 disk
= get_imsm_disk(super
, get_imsm_disk_idx(dev
, i
));
4076 for (j
= 0; j
< new_map
->num_members
; j
++)
4077 if (serialcmp(disk
->serial
, inf
[j
].serial
) == 0)
4084 static void imsm_delete(struct intel_super
*super
, struct dl
**dlp
, int index
);
4086 static void imsm_process_update(struct supertype
*st
,
4087 struct metadata_update
*update
)
4090 * crack open the metadata_update envelope to find the update record
4091 * update can be one of:
4092 * update_activate_spare - a spare device has replaced a failed
4093 * device in an array, update the disk_ord_tbl. If this disk is
4094 * present in all member arrays then also clear the SPARE_DISK
4097 struct intel_super
*super
= st
->sb
;
4098 struct imsm_super
*mpb
;
4099 enum imsm_update_type type
= *(enum imsm_update_type
*) update
->buf
;
4101 /* update requires a larger buf but the allocation failed */
4102 if (super
->next_len
&& !super
->next_buf
) {
4103 super
->next_len
= 0;
4107 if (super
->next_buf
) {
4108 memcpy(super
->next_buf
, super
->buf
, super
->len
);
4110 super
->len
= super
->next_len
;
4111 super
->buf
= super
->next_buf
;
4113 super
->next_len
= 0;
4114 super
->next_buf
= NULL
;
4117 mpb
= super
->anchor
;
4120 case update_activate_spare
: {
4121 struct imsm_update_activate_spare
*u
= (void *) update
->buf
;
4122 struct imsm_dev
*dev
= get_imsm_dev(super
, u
->array
);
4123 struct imsm_map
*map
= get_imsm_map(dev
, 0);
4124 struct imsm_map
*migr_map
;
4125 struct active_array
*a
;
4126 struct imsm_disk
*disk
;
4131 int victim
= get_imsm_disk_idx(dev
, u
->slot
);
4134 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
4139 fprintf(stderr
, "error: imsm_activate_spare passed "
4140 "an unknown disk (index: %d)\n",
4145 super
->updates_pending
++;
4147 /* count failures (excluding rebuilds and the victim)
4148 * to determine map[0] state
4151 for (i
= 0; i
< map
->num_members
; i
++) {
4154 disk
= get_imsm_disk(super
, get_imsm_disk_idx(dev
, i
));
4155 if (!disk
|| disk
->status
& FAILED_DISK
)
4159 /* adding a pristine spare, assign a new index */
4160 if (dl
->index
< 0) {
4161 dl
->index
= super
->anchor
->num_disks
;
4162 super
->anchor
->num_disks
++;
4165 disk
->status
|= CONFIGURED_DISK
;
4166 disk
->status
&= ~SPARE_DISK
;
4169 to_state
= imsm_check_degraded(super
, dev
, failed
);
4170 map
->map_state
= IMSM_T_STATE_DEGRADED
;
4171 migrate(dev
, to_state
, MIGR_REBUILD
);
4172 migr_map
= get_imsm_map(dev
, 1);
4173 set_imsm_ord_tbl_ent(map
, u
->slot
, dl
->index
);
4174 set_imsm_ord_tbl_ent(migr_map
, u
->slot
, dl
->index
| IMSM_ORD_REBUILD
);
4176 /* count arrays using the victim in the metadata */
4178 for (a
= st
->arrays
; a
; a
= a
->next
) {
4179 dev
= get_imsm_dev(super
, a
->info
.container_member
);
4180 map
= get_imsm_map(dev
, 0);
4182 if (get_imsm_disk_slot(map
, victim
) >= 0)
4186 /* delete the victim if it is no longer being
4192 /* We know that 'manager' isn't touching anything,
4193 * so it is safe to delete
4195 for (dlp
= &super
->disks
; *dlp
; dlp
= &(*dlp
)->next
)
4196 if ((*dlp
)->index
== victim
)
4199 /* victim may be on the missing list */
4201 for (dlp
= &super
->missing
; *dlp
; dlp
= &(*dlp
)->next
)
4202 if ((*dlp
)->index
== victim
)
4204 imsm_delete(super
, dlp
, victim
);
4208 case update_create_array
: {
4209 /* someone wants to create a new array, we need to be aware of
4210 * a few races/collisions:
4211 * 1/ 'Create' called by two separate instances of mdadm
4212 * 2/ 'Create' versus 'activate_spare': mdadm has chosen
4213 * devices that have since been assimilated via
4215 * In the event this update can not be carried out mdadm will
4216 * (FIX ME) notice that its update did not take hold.
4218 struct imsm_update_create_array
*u
= (void *) update
->buf
;
4219 struct intel_dev
*dv
;
4220 struct imsm_dev
*dev
;
4221 struct imsm_map
*map
, *new_map
;
4222 unsigned long long start
, end
;
4223 unsigned long long new_start
, new_end
;
4225 struct disk_info
*inf
;
4228 /* handle racing creates: first come first serve */
4229 if (u
->dev_idx
< mpb
->num_raid_devs
) {
4230 dprintf("%s: subarray %d already defined\n",
4231 __func__
, u
->dev_idx
);
4235 /* check update is next in sequence */
4236 if (u
->dev_idx
!= mpb
->num_raid_devs
) {
4237 dprintf("%s: can not create array %d expected index %d\n",
4238 __func__
, u
->dev_idx
, mpb
->num_raid_devs
);
4242 new_map
= get_imsm_map(&u
->dev
, 0);
4243 new_start
= __le32_to_cpu(new_map
->pba_of_lba0
);
4244 new_end
= new_start
+ __le32_to_cpu(new_map
->blocks_per_member
);
4245 inf
= get_disk_info(u
);
4247 /* handle activate_spare versus create race:
4248 * check to make sure that overlapping arrays do not include
4251 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
4252 dev
= get_imsm_dev(super
, i
);
4253 map
= get_imsm_map(dev
, 0);
4254 start
= __le32_to_cpu(map
->pba_of_lba0
);
4255 end
= start
+ __le32_to_cpu(map
->blocks_per_member
);
4256 if ((new_start
>= start
&& new_start
<= end
) ||
4257 (start
>= new_start
&& start
<= new_end
))
4262 if (disks_overlap(super
, i
, u
)) {
4263 dprintf("%s: arrays overlap\n", __func__
);
4268 /* check that prepare update was successful */
4269 if (!update
->space
) {
4270 dprintf("%s: prepare update failed\n", __func__
);
4274 /* check that all disks are still active before committing
4275 * changes. FIXME: could we instead handle this by creating a
4276 * degraded array? That's probably not what the user expects,
4277 * so better to drop this update on the floor.
4279 for (i
= 0; i
< new_map
->num_members
; i
++) {
4280 dl
= serial_to_dl(inf
[i
].serial
, super
);
4282 dprintf("%s: disk disappeared\n", __func__
);
4287 super
->updates_pending
++;
4289 /* convert spares to members and fixup ord_tbl */
4290 for (i
= 0; i
< new_map
->num_members
; i
++) {
4291 dl
= serial_to_dl(inf
[i
].serial
, super
);
4292 if (dl
->index
== -1) {
4293 dl
->index
= mpb
->num_disks
;
4295 dl
->disk
.status
|= CONFIGURED_DISK
;
4296 dl
->disk
.status
&= ~SPARE_DISK
;
4298 set_imsm_ord_tbl_ent(new_map
, i
, dl
->index
);
4303 update
->space
= NULL
;
4304 imsm_copy_dev(dev
, &u
->dev
);
4305 dv
->index
= u
->dev_idx
;
4306 dv
->next
= super
->devlist
;
4307 super
->devlist
= dv
;
4308 mpb
->num_raid_devs
++;
4310 imsm_update_version_info(super
);
4313 /* mdmon knows how to release update->space, but not
4314 * ((struct intel_dev *) update->space)->dev
4316 if (update
->space
) {
4322 case update_add_disk
:
4324 /* we may be able to repair some arrays if disks are
4327 struct active_array
*a
;
4329 super
->updates_pending
++;
4330 for (a
= st
->arrays
; a
; a
= a
->next
)
4331 a
->check_degraded
= 1;
4333 /* add some spares to the metadata */
4334 while (super
->add
) {
4338 super
->add
= al
->next
;
4339 al
->next
= super
->disks
;
4341 dprintf("%s: added %x:%x\n",
4342 __func__
, al
->major
, al
->minor
);
4349 static void imsm_prepare_update(struct supertype
*st
,
4350 struct metadata_update
*update
)
4353 * Allocate space to hold new disk entries, raid-device entries or a new
4354 * mpb if necessary. The manager synchronously waits for updates to
4355 * complete in the monitor, so new mpb buffers allocated here can be
4356 * integrated by the monitor thread without worrying about live pointers
4357 * in the manager thread.
4359 enum imsm_update_type type
= *(enum imsm_update_type
*) update
->buf
;
4360 struct intel_super
*super
= st
->sb
;
4361 struct imsm_super
*mpb
= super
->anchor
;
4366 case update_create_array
: {
4367 struct imsm_update_create_array
*u
= (void *) update
->buf
;
4368 struct intel_dev
*dv
;
4369 struct imsm_dev
*dev
= &u
->dev
;
4370 struct imsm_map
*map
= get_imsm_map(dev
, 0);
4372 struct disk_info
*inf
;
4376 inf
= get_disk_info(u
);
4377 len
= sizeof_imsm_dev(dev
, 1);
4378 /* allocate a new super->devlist entry */
4379 dv
= malloc(sizeof(*dv
));
4381 dv
->dev
= malloc(len
);
4386 update
->space
= NULL
;
4390 /* count how many spares will be converted to members */
4391 for (i
= 0; i
< map
->num_members
; i
++) {
4392 dl
= serial_to_dl(inf
[i
].serial
, super
);
4394 /* hmm maybe it failed?, nothing we can do about
4399 if (count_memberships(dl
, super
) == 0)
4402 len
+= activate
* sizeof(struct imsm_disk
);
4409 /* check if we need a larger metadata buffer */
4410 if (super
->next_buf
)
4411 buf_len
= super
->next_len
;
4413 buf_len
= super
->len
;
4415 if (__le32_to_cpu(mpb
->mpb_size
) + len
> buf_len
) {
4416 /* ok we need a larger buf than what is currently allocated
4417 * if this allocation fails process_update will notice that
4418 * ->next_len is set and ->next_buf is NULL
4420 buf_len
= ROUND_UP(__le32_to_cpu(mpb
->mpb_size
) + len
, 512);
4421 if (super
->next_buf
)
4422 free(super
->next_buf
);
4424 super
->next_len
= buf_len
;
4425 if (posix_memalign(&super
->next_buf
, 512, buf_len
) == 0)
4426 memset(super
->next_buf
, 0, buf_len
);
4428 super
->next_buf
= NULL
;
4432 /* must be called while manager is quiesced */
4433 static void imsm_delete(struct intel_super
*super
, struct dl
**dlp
, int index
)
4435 struct imsm_super
*mpb
= super
->anchor
;
4437 struct imsm_dev
*dev
;
4438 struct imsm_map
*map
;
4439 int i
, j
, num_members
;
4442 dprintf("%s: deleting device[%d] from imsm_super\n",
4445 /* shift all indexes down one */
4446 for (iter
= super
->disks
; iter
; iter
= iter
->next
)
4447 if (iter
->index
> index
)
4449 for (iter
= super
->missing
; iter
; iter
= iter
->next
)
4450 if (iter
->index
> index
)
4453 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
4454 dev
= get_imsm_dev(super
, i
);
4455 map
= get_imsm_map(dev
, 0);
4456 num_members
= map
->num_members
;
4457 for (j
= 0; j
< num_members
; j
++) {
4458 /* update ord entries being careful not to propagate
4459 * ord-flags to the first map
4461 ord
= get_imsm_ord_tbl_ent(dev
, j
);
4463 if (ord_to_idx(ord
) <= index
)
4466 map
= get_imsm_map(dev
, 0);
4467 set_imsm_ord_tbl_ent(map
, j
, ord_to_idx(ord
- 1));
4468 map
= get_imsm_map(dev
, 1);
4470 set_imsm_ord_tbl_ent(map
, j
, ord
- 1);
4475 super
->updates_pending
++;
4477 struct dl
*dl
= *dlp
;
4479 *dlp
= (*dlp
)->next
;
4480 __free_imsm_disk(dl
);
4483 #endif /* MDASSEMBLE */
4485 struct superswitch super_imsm
= {
4487 .examine_super
= examine_super_imsm
,
4488 .brief_examine_super
= brief_examine_super_imsm
,
4489 .export_examine_super
= export_examine_super_imsm
,
4490 .detail_super
= detail_super_imsm
,
4491 .brief_detail_super
= brief_detail_super_imsm
,
4492 .write_init_super
= write_init_super_imsm
,
4493 .validate_geometry
= validate_geometry_imsm
,
4494 .add_to_super
= add_to_super_imsm
,
4495 .detail_platform
= detail_platform_imsm
,
4497 .match_home
= match_home_imsm
,
4498 .uuid_from_super
= uuid_from_super_imsm
,
4499 .getinfo_super
= getinfo_super_imsm
,
4500 .update_super
= update_super_imsm
,
4502 .avail_size
= avail_size_imsm
,
4504 .compare_super
= compare_super_imsm
,
4506 .load_super
= load_super_imsm
,
4507 .init_super
= init_super_imsm
,
4508 .store_super
= store_zero_imsm
,
4509 .free_super
= free_super_imsm
,
4510 .match_metadata_desc
= match_metadata_desc_imsm
,
4511 .container_content
= container_content_imsm
,
4512 .default_layout
= imsm_level_to_layout
,
4519 .open_new
= imsm_open_new
,
4520 .load_super
= load_super_imsm
,
4521 .set_array_state
= imsm_set_array_state
,
4522 .set_disk
= imsm_set_disk
,
4523 .sync_metadata
= imsm_sync_metadata
,
4524 .activate_spare
= imsm_activate_spare
,
4525 .process_update
= imsm_process_update
,
4526 .prepare_update
= imsm_prepare_update
,
4527 #endif /* MDASSEMBLE */