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
28 /* MPB == Metadata Parameter Block */
29 #define MPB_SIGNATURE "Intel Raid ISM Cfg Sig. "
30 #define MPB_SIG_LEN (strlen(MPB_SIGNATURE))
31 #define MPB_VERSION_RAID0 "1.0.00"
32 #define MPB_VERSION_RAID1 "1.1.00"
33 #define MPB_VERSION_MANY_VOLUMES_PER_ARRAY "1.2.00"
34 #define MPB_VERSION_3OR4_DISK_ARRAY "1.2.01"
35 #define MPB_VERSION_RAID5 "1.2.02"
36 #define MPB_VERSION_5OR6_DISK_ARRAY "1.2.04"
37 #define MPB_VERSION_CNG "1.2.06"
38 #define MPB_VERSION_ATTRIBS "1.3.00"
39 #define MAX_SIGNATURE_LENGTH 32
40 #define MAX_RAID_SERIAL_LEN 16
42 #define MPB_ATTRIB_CHECKSUM_VERIFY __cpu_to_le32(0x80000000)
43 #define MPB_ATTRIB_PM __cpu_to_le32(0x40000000)
44 #define MPB_ATTRIB_2TB __cpu_to_le32(0x20000000)
45 #define MPB_ATTRIB_RAID0 __cpu_to_le32(0x00000001)
46 #define MPB_ATTRIB_RAID1 __cpu_to_le32(0x00000002)
47 #define MPB_ATTRIB_RAID10 __cpu_to_le32(0x00000004)
48 #define MPB_ATTRIB_RAID1E __cpu_to_le32(0x00000008)
49 #define MPB_ATTRIB_RAID5 __cpu_to_le32(0x00000010)
50 #define MPB_ATTRIB_RAIDCNG __cpu_to_le32(0x00000020)
52 #define MPB_SECTOR_CNT 418
53 #define IMSM_RESERVED_SECTORS 4096
55 /* Disk configuration info. */
56 #define IMSM_MAX_DEVICES 255
58 __u8 serial
[MAX_RAID_SERIAL_LEN
];/* 0xD8 - 0xE7 ascii serial number */
59 __u32 total_blocks
; /* 0xE8 - 0xEB total blocks */
60 __u32 scsi_id
; /* 0xEC - 0xEF scsi ID */
61 #define SPARE_DISK __cpu_to_le32(0x01) /* Spare */
62 #define CONFIGURED_DISK __cpu_to_le32(0x02) /* Member of some RaidDev */
63 #define FAILED_DISK __cpu_to_le32(0x04) /* Permanent failure */
64 #define USABLE_DISK __cpu_to_le32(0x08) /* Fully usable unless FAILED_DISK is set */
65 __u32 status
; /* 0xF0 - 0xF3 */
66 __u32 owner_cfg_num
; /* which config 0,1,2... owns this disk */
67 #define IMSM_DISK_FILLERS 4
68 __u32 filler
[IMSM_DISK_FILLERS
]; /* 0xF4 - 0x107 MPB_DISK_FILLERS for future expansion */
71 /* RAID map configuration infos. */
73 __u32 pba_of_lba0
; /* start address of partition */
74 __u32 blocks_per_member
;/* blocks per member */
75 __u32 num_data_stripes
; /* number of data stripes */
76 __u16 blocks_per_strip
;
77 __u8 map_state
; /* Normal, Uninitialized, Degraded, Failed */
78 #define IMSM_T_STATE_NORMAL 0
79 #define IMSM_T_STATE_UNINITIALIZED 1
80 #define IMSM_T_STATE_DEGRADED 2
81 #define IMSM_T_STATE_FAILED 3
83 #define IMSM_T_RAID0 0
84 #define IMSM_T_RAID1 1
85 #define IMSM_T_RAID5 5 /* since metadata version 1.2.02 ? */
86 __u8 num_members
; /* number of member disks */
87 __u8 num_domains
; /* number of parity domains */
88 __u8 failed_disk_num
; /* valid only when state is degraded */
90 __u32 filler
[7]; /* expansion area */
91 #define IMSM_ORD_REBUILD (1 << 24)
92 __u32 disk_ord_tbl
[1]; /* disk_ord_tbl[num_members],
93 * top byte contains some flags
95 } __attribute__ ((packed
));
99 __u32 checkpoint_id
; /* id to access curr_migr_unit */
100 __u8 migr_state
; /* Normal or Migrating */
102 #define MIGR_REBUILD 1
103 #define MIGR_VERIFY 2 /* analagous to echo check > sync_action */
104 #define MIGR_GEN_MIGR 3
105 #define MIGR_STATE_CHANGE 4
106 __u8 migr_type
; /* Initializing, Rebuilding, ... */
108 __u8 fs_state
; /* fast-sync state for CnG (0xff == disabled) */
109 __u16 verify_errors
; /* number of mismatches */
110 __u16 bad_blocks
; /* number of bad blocks during verify */
112 struct imsm_map map
[1];
113 /* here comes another one if migr_state */
114 } __attribute__ ((packed
));
117 __u8 volume
[MAX_RAID_SERIAL_LEN
];
120 #define DEV_BOOTABLE __cpu_to_le32(0x01)
121 #define DEV_BOOT_DEVICE __cpu_to_le32(0x02)
122 #define DEV_READ_COALESCING __cpu_to_le32(0x04)
123 #define DEV_WRITE_COALESCING __cpu_to_le32(0x08)
124 #define DEV_LAST_SHUTDOWN_DIRTY __cpu_to_le32(0x10)
125 #define DEV_HIDDEN_AT_BOOT __cpu_to_le32(0x20)
126 #define DEV_CURRENTLY_HIDDEN __cpu_to_le32(0x40)
127 #define DEV_VERIFY_AND_FIX __cpu_to_le32(0x80)
128 #define DEV_MAP_STATE_UNINIT __cpu_to_le32(0x100)
129 #define DEV_NO_AUTO_RECOVERY __cpu_to_le32(0x200)
130 #define DEV_CLONE_N_GO __cpu_to_le32(0x400)
131 #define DEV_CLONE_MAN_SYNC __cpu_to_le32(0x800)
132 #define DEV_CNG_MASTER_DISK_NUM __cpu_to_le32(0x1000)
133 __u32 status
; /* Persistent RaidDev status */
134 __u32 reserved_blocks
; /* Reserved blocks at beginning of volume */
138 __u8 cng_master_disk
;
142 #define IMSM_DEV_FILLERS 10
143 __u32 filler
[IMSM_DEV_FILLERS
];
145 } __attribute__ ((packed
));
148 __u8 sig
[MAX_SIGNATURE_LENGTH
]; /* 0x00 - 0x1F */
149 __u32 check_sum
; /* 0x20 - 0x23 MPB Checksum */
150 __u32 mpb_size
; /* 0x24 - 0x27 Size of MPB */
151 __u32 family_num
; /* 0x28 - 0x2B Checksum from first time this config was written */
152 __u32 generation_num
; /* 0x2C - 0x2F Incremented each time this array's MPB is written */
153 __u32 error_log_size
; /* 0x30 - 0x33 in bytes */
154 __u32 attributes
; /* 0x34 - 0x37 */
155 __u8 num_disks
; /* 0x38 Number of configured disks */
156 __u8 num_raid_devs
; /* 0x39 Number of configured volumes */
157 __u8 error_log_pos
; /* 0x3A */
158 __u8 fill
[1]; /* 0x3B */
159 __u32 cache_size
; /* 0x3c - 0x40 in mb */
160 __u32 orig_family_num
; /* 0x40 - 0x43 original family num */
161 __u32 pwr_cycle_count
; /* 0x44 - 0x47 simulated power cycle count for array */
162 __u32 bbm_log_size
; /* 0x48 - 0x4B - size of bad Block Mgmt Log in bytes */
163 #define IMSM_FILLERS 35
164 __u32 filler
[IMSM_FILLERS
]; /* 0x4C - 0xD7 RAID_MPB_FILLERS */
165 struct imsm_disk disk
[1]; /* 0xD8 diskTbl[numDisks] */
166 /* here comes imsm_dev[num_raid_devs] */
167 /* here comes BBM logs */
168 } __attribute__ ((packed
));
170 #define BBM_LOG_MAX_ENTRIES 254
172 struct bbm_log_entry
{
173 __u64 defective_block_start
;
174 #define UNREADABLE 0xFFFFFFFF
175 __u32 spare_block_offset
;
176 __u16 remapped_marked_count
;
178 } __attribute__ ((__packed__
));
181 __u32 signature
; /* 0xABADB10C */
183 __u32 reserved_spare_block_count
; /* 0 */
184 __u32 reserved
; /* 0xFFFF */
185 __u64 first_spare_lba
;
186 struct bbm_log_entry mapped_block_entries
[BBM_LOG_MAX_ENTRIES
];
187 } __attribute__ ((__packed__
));
191 static char *map_state_str
[] = { "normal", "uninitialized", "degraded", "failed" };
194 static unsigned int sector_count(__u32 bytes
)
196 return ((bytes
+ (512-1)) & (~(512-1))) / 512;
199 static unsigned int mpb_sectors(struct imsm_super
*mpb
)
201 return sector_count(__le32_to_cpu(mpb
->mpb_size
));
204 /* internal representation of IMSM metadata */
207 void *buf
; /* O_DIRECT buffer for reading/writing metadata */
208 struct imsm_super
*anchor
; /* immovable parameters */
210 size_t len
; /* size of the 'buf' allocation */
211 void *next_buf
; /* for realloc'ing buf from the manager */
213 int updates_pending
; /* count of pending updates for mdmon */
214 int creating_imsm
; /* flag to indicate container creation */
215 int current_vol
; /* index of raid device undergoing creation */
216 __u32 create_offset
; /* common start for 'current_vol' */
217 #define IMSM_MAX_RAID_DEVS 2
218 struct imsm_dev
*dev_tbl
[IMSM_MAX_RAID_DEVS
];
222 __u8 serial
[MAX_RAID_SERIAL_LEN
];
225 struct imsm_disk disk
;
228 struct extent
*e
; /* for determining freespace @ create */
230 struct dl
*add
; /* list of disks to add while mdmon active */
231 struct dl
*missing
; /* disks removed while we weren't looking */
232 struct bbm_log
*bbm_log
;
236 unsigned long long start
, size
;
239 /* definition of messages passed to imsm_process_update */
240 enum imsm_update_type
{
241 update_activate_spare
,
246 struct imsm_update_activate_spare
{
247 enum imsm_update_type type
;
251 struct imsm_update_activate_spare
*next
;
255 __u8 serial
[MAX_RAID_SERIAL_LEN
];
258 struct imsm_update_create_array
{
259 enum imsm_update_type type
;
264 struct imsm_update_add_disk
{
265 enum imsm_update_type type
;
268 static struct supertype
*match_metadata_desc_imsm(char *arg
)
270 struct supertype
*st
;
272 if (strcmp(arg
, "imsm") != 0 &&
273 strcmp(arg
, "default") != 0
277 st
= malloc(sizeof(*st
));
278 memset(st
, 0, sizeof(*st
));
279 st
->ss
= &super_imsm
;
280 st
->max_devs
= IMSM_MAX_DEVICES
;
281 st
->minor_version
= 0;
287 static __u8
*get_imsm_version(struct imsm_super
*mpb
)
289 return &mpb
->sig
[MPB_SIG_LEN
];
293 /* retrieve a disk directly from the anchor when the anchor is known to be
294 * up-to-date, currently only at load time
296 static struct imsm_disk
*__get_imsm_disk(struct imsm_super
*mpb
, __u8 index
)
298 if (index
>= mpb
->num_disks
)
300 return &mpb
->disk
[index
];
304 /* retrieve a disk from the parsed metadata */
305 static struct imsm_disk
*get_imsm_disk(struct intel_super
*super
, __u8 index
)
309 for (d
= super
->disks
; d
; d
= d
->next
)
310 if (d
->index
== index
)
317 /* generate a checksum directly from the anchor when the anchor is known to be
318 * up-to-date, currently only at load or write_super after coalescing
320 static __u32
__gen_imsm_checksum(struct imsm_super
*mpb
)
322 __u32 end
= mpb
->mpb_size
/ sizeof(end
);
323 __u32
*p
= (__u32
*) mpb
;
327 sum
+= __le32_to_cpu(*p
);
331 return sum
- __le32_to_cpu(mpb
->check_sum
);
334 static size_t sizeof_imsm_map(struct imsm_map
*map
)
336 return sizeof(struct imsm_map
) + sizeof(__u32
) * (map
->num_members
- 1);
339 struct imsm_map
*get_imsm_map(struct imsm_dev
*dev
, int second_map
)
341 struct imsm_map
*map
= &dev
->vol
.map
[0];
343 if (second_map
&& !dev
->vol
.migr_state
)
345 else if (second_map
) {
348 return ptr
+ sizeof_imsm_map(map
);
354 /* return the size of the device.
355 * migr_state increases the returned size if map[0] were to be duplicated
357 static size_t sizeof_imsm_dev(struct imsm_dev
*dev
, int migr_state
)
359 size_t size
= sizeof(*dev
) - sizeof(struct imsm_map
) +
360 sizeof_imsm_map(get_imsm_map(dev
, 0));
362 /* migrating means an additional map */
363 if (dev
->vol
.migr_state
)
364 size
+= sizeof_imsm_map(get_imsm_map(dev
, 1));
366 size
+= sizeof_imsm_map(get_imsm_map(dev
, 0));
372 /* retrieve disk serial number list from a metadata update */
373 static struct disk_info
*get_disk_info(struct imsm_update_create_array
*update
)
376 struct disk_info
*inf
;
378 inf
= u
+ sizeof(*update
) - sizeof(struct imsm_dev
) +
379 sizeof_imsm_dev(&update
->dev
, 0);
385 static struct imsm_dev
*__get_imsm_dev(struct imsm_super
*mpb
, __u8 index
)
391 if (index
>= mpb
->num_raid_devs
)
394 /* devices start after all disks */
395 offset
= ((void *) &mpb
->disk
[mpb
->num_disks
]) - _mpb
;
397 for (i
= 0; i
<= index
; i
++)
399 return _mpb
+ offset
;
401 offset
+= sizeof_imsm_dev(_mpb
+ offset
, 0);
406 static struct imsm_dev
*get_imsm_dev(struct intel_super
*super
, __u8 index
)
408 if (index
>= super
->anchor
->num_raid_devs
)
410 return super
->dev_tbl
[index
];
413 static __u32
get_imsm_ord_tbl_ent(struct imsm_dev
*dev
, int slot
)
415 struct imsm_map
*map
;
417 if (dev
->vol
.migr_state
)
418 map
= get_imsm_map(dev
, 1);
420 map
= get_imsm_map(dev
, 0);
422 /* top byte identifies disk under rebuild */
423 return __le32_to_cpu(map
->disk_ord_tbl
[slot
]);
426 #define ord_to_idx(ord) (((ord) << 8) >> 8)
427 static __u32
get_imsm_disk_idx(struct imsm_dev
*dev
, int slot
)
429 __u32 ord
= get_imsm_ord_tbl_ent(dev
, slot
);
431 return ord_to_idx(ord
);
434 static void set_imsm_ord_tbl_ent(struct imsm_map
*map
, int slot
, __u32 ord
)
436 map
->disk_ord_tbl
[slot
] = __cpu_to_le32(ord
);
439 static int get_imsm_raid_level(struct imsm_map
*map
)
441 if (map
->raid_level
== 1) {
442 if (map
->num_members
== 2)
448 return map
->raid_level
;
451 static int cmp_extent(const void *av
, const void *bv
)
453 const struct extent
*a
= av
;
454 const struct extent
*b
= bv
;
455 if (a
->start
< b
->start
)
457 if (a
->start
> b
->start
)
462 static int count_memberships(struct dl
*dl
, struct intel_super
*super
)
467 for (i
= 0; i
< super
->anchor
->num_raid_devs
; i
++) {
468 struct imsm_dev
*dev
= get_imsm_dev(super
, i
);
469 struct imsm_map
*map
= get_imsm_map(dev
, 0);
471 for (j
= 0; j
< map
->num_members
; j
++) {
472 __u32 index
= get_imsm_disk_idx(dev
, j
);
474 if (index
== dl
->index
)
482 static struct extent
*get_extents(struct intel_super
*super
, struct dl
*dl
)
484 /* find a list of used extents on the given physical device */
485 struct extent
*rv
, *e
;
487 int memberships
= count_memberships(dl
, super
);
488 __u32 reservation
= MPB_SECTOR_CNT
+ IMSM_RESERVED_SECTORS
;
490 rv
= malloc(sizeof(struct extent
) * (memberships
+ 1));
495 for (i
= 0; i
< super
->anchor
->num_raid_devs
; i
++) {
496 struct imsm_dev
*dev
= get_imsm_dev(super
, i
);
497 struct imsm_map
*map
= get_imsm_map(dev
, 0);
499 for (j
= 0; j
< map
->num_members
; j
++) {
500 __u32 index
= get_imsm_disk_idx(dev
, j
);
502 if (index
== dl
->index
) {
503 e
->start
= __le32_to_cpu(map
->pba_of_lba0
);
504 e
->size
= __le32_to_cpu(map
->blocks_per_member
);
509 qsort(rv
, memberships
, sizeof(*rv
), cmp_extent
);
511 /* determine the start of the metadata
512 * when no raid devices are defined use the default
513 * ...otherwise allow the metadata to truncate the value
514 * as is the case with older versions of imsm
517 struct extent
*last
= &rv
[memberships
- 1];
520 remainder
= __le32_to_cpu(dl
->disk
.total_blocks
) -
521 (last
->start
+ last
->size
);
522 /* round down to 1k block to satisfy precision of the kernel
526 /* make sure remainder is still sane */
527 if (remainder
< ROUND_UP(super
->len
, 512) >> 9)
528 remainder
= ROUND_UP(super
->len
, 512) >> 9;
529 if (reservation
> remainder
)
530 reservation
= remainder
;
532 e
->start
= __le32_to_cpu(dl
->disk
.total_blocks
) - reservation
;
537 /* try to determine how much space is reserved for metadata from
538 * the last get_extents() entry, otherwise fallback to the
541 static __u32
imsm_reserved_sectors(struct intel_super
*super
, struct dl
*dl
)
547 /* for spares just return a minimal reservation which will grow
548 * once the spare is picked up by an array
551 return MPB_SECTOR_CNT
;
553 e
= get_extents(super
, dl
);
555 return MPB_SECTOR_CNT
+ IMSM_RESERVED_SECTORS
;
557 /* scroll to last entry */
558 for (i
= 0; e
[i
].size
; i
++)
561 rv
= __le32_to_cpu(dl
->disk
.total_blocks
) - e
[i
].start
;
569 static void print_imsm_dev(struct imsm_dev
*dev
, char *uuid
, int disk_idx
)
573 struct imsm_map
*map
= get_imsm_map(dev
, 0);
577 printf("[%.16s]:\n", dev
->volume
);
578 printf(" UUID : %s\n", uuid
);
579 printf(" RAID Level : %d\n", get_imsm_raid_level(map
));
580 printf(" Members : %d\n", map
->num_members
);
581 for (slot
= 0; slot
< map
->num_members
; slot
++)
582 if (disk_idx
== get_imsm_disk_idx(dev
, slot
))
584 if (slot
< map
->num_members
) {
585 ord
= get_imsm_ord_tbl_ent(dev
, slot
);
586 printf(" This Slot : %d%s\n", slot
,
587 ord
& IMSM_ORD_REBUILD
? " (out-of-sync)" : "");
589 printf(" This Slot : ?\n");
590 sz
= __le32_to_cpu(dev
->size_high
);
592 sz
+= __le32_to_cpu(dev
->size_low
);
593 printf(" Array Size : %llu%s\n", (unsigned long long)sz
,
594 human_size(sz
* 512));
595 sz
= __le32_to_cpu(map
->blocks_per_member
);
596 printf(" Per Dev Size : %llu%s\n", (unsigned long long)sz
,
597 human_size(sz
* 512));
598 printf(" Sector Offset : %u\n",
599 __le32_to_cpu(map
->pba_of_lba0
));
600 printf(" Num Stripes : %u\n",
601 __le32_to_cpu(map
->num_data_stripes
));
602 printf(" Chunk Size : %u KiB\n",
603 __le16_to_cpu(map
->blocks_per_strip
) / 2);
604 printf(" Reserved : %d\n", __le32_to_cpu(dev
->reserved_blocks
));
605 printf(" Migrate State : %s", dev
->vol
.migr_state
? "migrating" : "idle");
606 if (dev
->vol
.migr_state
)
607 printf(": %s", dev
->vol
.migr_type
? "rebuilding" : "initializing");
609 printf(" Map State : %s", map_state_str
[map
->map_state
]);
610 if (dev
->vol
.migr_state
) {
611 struct imsm_map
*map
= get_imsm_map(dev
, 1);
612 printf(" <-- %s", map_state_str
[map
->map_state
]);
615 printf(" Dirty State : %s\n", dev
->vol
.dirty
? "dirty" : "clean");
618 static void print_imsm_disk(struct imsm_super
*mpb
, int index
, __u32 reserved
)
620 struct imsm_disk
*disk
= __get_imsm_disk(mpb
, index
);
621 char str
[MAX_RAID_SERIAL_LEN
+ 1];
629 snprintf(str
, MAX_RAID_SERIAL_LEN
+ 1, "%s", disk
->serial
);
630 printf(" Disk%02d Serial : %s\n", index
, str
);
632 printf(" State :%s%s%s%s\n", s
&SPARE_DISK
? " spare" : "",
633 s
&CONFIGURED_DISK
? " active" : "",
634 s
&FAILED_DISK
? " failed" : "",
635 s
&USABLE_DISK
? " usable" : "");
636 printf(" Id : %08x\n", __le32_to_cpu(disk
->scsi_id
));
637 sz
= __le32_to_cpu(disk
->total_blocks
) - reserved
;
638 printf(" Usable Size : %llu%s\n", (unsigned long long)sz
,
639 human_size(sz
* 512));
642 static void getinfo_super_imsm(struct supertype
*st
, struct mdinfo
*info
);
644 static void examine_super_imsm(struct supertype
*st
, char *homehost
)
646 struct intel_super
*super
= st
->sb
;
647 struct imsm_super
*mpb
= super
->anchor
;
648 char str
[MAX_SIGNATURE_LENGTH
];
653 __u32 reserved
= imsm_reserved_sectors(super
, super
->disks
);
656 snprintf(str
, MPB_SIG_LEN
, "%s", mpb
->sig
);
657 printf(" Magic : %s\n", str
);
658 snprintf(str
, strlen(MPB_VERSION_RAID0
), "%s", get_imsm_version(mpb
));
659 printf(" Version : %s\n", get_imsm_version(mpb
));
660 printf(" Family : %08x\n", __le32_to_cpu(mpb
->family_num
));
661 printf(" Generation : %08x\n", __le32_to_cpu(mpb
->generation_num
));
662 getinfo_super_imsm(st
, &info
);
663 fname_from_uuid(st
, &info
, nbuf
,'-');
664 printf(" UUID : %s\n", nbuf
+ 5);
665 sum
= __le32_to_cpu(mpb
->check_sum
);
666 printf(" Checksum : %08x %s\n", sum
,
667 __gen_imsm_checksum(mpb
) == sum
? "correct" : "incorrect");
668 printf(" MPB Sectors : %d\n", mpb_sectors(mpb
));
669 printf(" Disks : %d\n", mpb
->num_disks
);
670 printf(" RAID Devices : %d\n", mpb
->num_raid_devs
);
671 print_imsm_disk(mpb
, super
->disks
->index
, reserved
);
672 if (super
->bbm_log
) {
673 struct bbm_log
*log
= super
->bbm_log
;
676 printf("Bad Block Management Log:\n");
677 printf(" Log Size : %d\n", __le32_to_cpu(mpb
->bbm_log_size
));
678 printf(" Signature : %x\n", __le32_to_cpu(log
->signature
));
679 printf(" Entry Count : %d\n", __le32_to_cpu(log
->entry_count
));
680 printf(" Spare Blocks : %d\n", __le32_to_cpu(log
->reserved_spare_block_count
));
681 printf(" First Spare : %llx\n", __le64_to_cpu(log
->first_spare_lba
));
683 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
685 struct imsm_dev
*dev
= __get_imsm_dev(mpb
, i
);
687 super
->current_vol
= i
;
688 getinfo_super_imsm(st
, &info
);
689 fname_from_uuid(st
, &info
, nbuf
, '-');
690 print_imsm_dev(dev
, nbuf
+ 5, super
->disks
->index
);
692 for (i
= 0; i
< mpb
->num_disks
; i
++) {
693 if (i
== super
->disks
->index
)
695 print_imsm_disk(mpb
, i
, reserved
);
699 static void brief_examine_super_imsm(struct supertype
*st
)
701 /* We just write a generic IMSM ARRAY entry */
705 struct intel_super
*super
= st
->sb
;
708 if (!super
->anchor
->num_raid_devs
)
711 getinfo_super_imsm(st
, &info
);
712 fname_from_uuid(st
, &info
, nbuf
,'-');
713 printf("ARRAY metadata=imsm auto=md UUID=%s\n", nbuf
+ 5);
714 for (i
= 0; i
< super
->anchor
->num_raid_devs
; i
++) {
715 struct imsm_dev
*dev
= get_imsm_dev(super
, i
);
717 super
->current_vol
= i
;
718 getinfo_super_imsm(st
, &info
);
719 fname_from_uuid(st
, &info
, nbuf1
,'-');
720 printf("ARRAY /dev/md/%.16s container=%s\n"
721 " member=%d auto=mdp UUID=%s\n",
722 dev
->volume
, nbuf
+ 5, i
, nbuf1
+ 5);
726 static void detail_super_imsm(struct supertype
*st
, char *homehost
)
731 getinfo_super_imsm(st
, &info
);
732 fname_from_uuid(st
, &info
, nbuf
,'-');
733 printf("\n UUID : %s\n", nbuf
+ 5);
736 static void brief_detail_super_imsm(struct supertype
*st
)
740 getinfo_super_imsm(st
, &info
);
741 fname_from_uuid(st
, &info
, nbuf
,'-');
742 printf(" UUID=%s", nbuf
+ 5);
746 static int match_home_imsm(struct supertype
*st
, char *homehost
)
748 /* the imsm metadata format does not specify any host
749 * identification information. We return -1 since we can never
750 * confirm nor deny whether a given array is "meant" for this
751 * host. We rely on compare_super and the 'family_num' field to
752 * exclude member disks that do not belong, and we rely on
753 * mdadm.conf to specify the arrays that should be assembled.
754 * Auto-assembly may still pick up "foreign" arrays.
760 static void uuid_from_super_imsm(struct supertype
*st
, int uuid
[4])
762 /* The uuid returned here is used for:
763 * uuid to put into bitmap file (Create, Grow)
764 * uuid for backup header when saving critical section (Grow)
765 * comparing uuids when re-adding a device into an array
766 * In these cases the uuid required is that of the data-array,
767 * not the device-set.
768 * uuid to recognise same set when adding a missing device back
769 * to an array. This is a uuid for the device-set.
771 * For each of these we can make do with a truncated
772 * or hashed uuid rather than the original, as long as
774 * In each case the uuid required is that of the data-array,
775 * not the device-set.
777 /* imsm does not track uuid's so we synthesis one using sha1 on
778 * - The signature (Which is constant for all imsm array, but no matter)
779 * - the family_num of the container
780 * - the index number of the volume
781 * - the 'serial' number of the volume.
782 * Hopefully these are all constant.
784 struct intel_super
*super
= st
->sb
;
788 struct imsm_dev
*dev
= NULL
;
791 sha1_process_bytes(super
->anchor
->sig
, MPB_SIG_LEN
, &ctx
);
792 sha1_process_bytes(&super
->anchor
->family_num
, sizeof(__u32
), &ctx
);
793 if (super
->current_vol
>= 0)
794 dev
= get_imsm_dev(super
, super
->current_vol
);
796 __u32 vol
= super
->current_vol
;
797 sha1_process_bytes(&vol
, sizeof(vol
), &ctx
);
798 sha1_process_bytes(dev
->volume
, MAX_RAID_SERIAL_LEN
, &ctx
);
800 sha1_finish_ctx(&ctx
, buf
);
801 memcpy(uuid
, buf
, 4*4);
806 get_imsm_numerical_version(struct imsm_super
*mpb
, int *m
, int *p
)
808 __u8
*v
= get_imsm_version(mpb
);
809 __u8
*end
= mpb
->sig
+ MAX_SIGNATURE_LENGTH
;
810 char major
[] = { 0, 0, 0 };
811 char minor
[] = { 0 ,0, 0 };
812 char patch
[] = { 0, 0, 0 };
813 char *ver_parse
[] = { major
, minor
, patch
};
817 while (*v
!= '\0' && v
< end
) {
818 if (*v
!= '.' && j
< 2)
819 ver_parse
[i
][j
++] = *v
;
827 *m
= strtol(minor
, NULL
, 0);
828 *p
= strtol(patch
, NULL
, 0);
832 static int imsm_level_to_layout(int level
)
840 return ALGORITHM_LEFT_ASYMMETRIC
;
847 static void getinfo_super_imsm_volume(struct supertype
*st
, struct mdinfo
*info
)
849 struct intel_super
*super
= st
->sb
;
850 struct imsm_dev
*dev
= get_imsm_dev(super
, super
->current_vol
);
851 struct imsm_map
*map
= get_imsm_map(dev
, 0);
853 info
->container_member
= super
->current_vol
;
854 info
->array
.raid_disks
= map
->num_members
;
855 info
->array
.level
= get_imsm_raid_level(map
);
856 info
->array
.layout
= imsm_level_to_layout(info
->array
.level
);
857 info
->array
.md_minor
= -1;
858 info
->array
.ctime
= 0;
859 info
->array
.utime
= 0;
860 info
->array
.chunk_size
= __le16_to_cpu(map
->blocks_per_strip
) << 9;
861 info
->array
.state
= !dev
->vol
.dirty
;
863 info
->disk
.major
= 0;
864 info
->disk
.minor
= 0;
866 info
->data_offset
= __le32_to_cpu(map
->pba_of_lba0
);
867 info
->component_size
= __le32_to_cpu(map
->blocks_per_member
);
868 memset(info
->uuid
, 0, sizeof(info
->uuid
));
870 if (map
->map_state
== IMSM_T_STATE_UNINITIALIZED
|| dev
->vol
.dirty
)
871 info
->resync_start
= 0;
872 else if (dev
->vol
.migr_state
)
873 info
->resync_start
= __le32_to_cpu(dev
->vol
.curr_migr_unit
);
875 info
->resync_start
= ~0ULL;
877 strncpy(info
->name
, (char *) dev
->volume
, MAX_RAID_SERIAL_LEN
);
878 info
->name
[MAX_RAID_SERIAL_LEN
] = 0;
880 info
->array
.major_version
= -1;
881 info
->array
.minor_version
= -2;
882 sprintf(info
->text_version
, "/%s/%d",
883 devnum2devname(st
->container_dev
),
884 info
->container_member
);
885 info
->safe_mode_delay
= 4000; /* 4 secs like the Matrix driver */
886 uuid_from_super_imsm(st
, info
->uuid
);
890 static void getinfo_super_imsm(struct supertype
*st
, struct mdinfo
*info
)
892 struct intel_super
*super
= st
->sb
;
893 struct imsm_disk
*disk
;
896 if (super
->current_vol
>= 0) {
897 getinfo_super_imsm_volume(st
, info
);
901 /* Set raid_disks to zero so that Assemble will always pull in valid
904 info
->array
.raid_disks
= 0;
905 info
->array
.level
= LEVEL_CONTAINER
;
906 info
->array
.layout
= 0;
907 info
->array
.md_minor
= -1;
908 info
->array
.ctime
= 0; /* N/A for imsm */
909 info
->array
.utime
= 0;
910 info
->array
.chunk_size
= 0;
912 info
->disk
.major
= 0;
913 info
->disk
.minor
= 0;
914 info
->disk
.raid_disk
= -1;
915 info
->reshape_active
= 0;
916 info
->array
.major_version
= -1;
917 info
->array
.minor_version
= -2;
918 strcpy(info
->text_version
, "imsm");
919 info
->safe_mode_delay
= 0;
920 info
->disk
.number
= -1;
921 info
->disk
.state
= 0;
925 __u32 reserved
= imsm_reserved_sectors(super
, super
->disks
);
927 disk
= &super
->disks
->disk
;
928 info
->data_offset
= __le32_to_cpu(disk
->total_blocks
) - reserved
;
929 info
->component_size
= reserved
;
931 info
->disk
.state
= s
& CONFIGURED_DISK
? (1 << MD_DISK_ACTIVE
) : 0;
932 info
->disk
.state
|= s
& FAILED_DISK
? (1 << MD_DISK_FAULTY
) : 0;
933 info
->disk
.state
|= s
& SPARE_DISK
? 0 : (1 << MD_DISK_SYNC
);
936 /* only call uuid_from_super_imsm when this disk is part of a populated container,
937 * ->compare_super may have updated the 'num_raid_devs' field for spares
939 if (info
->disk
.state
& (1 << MD_DISK_SYNC
) || super
->anchor
->num_raid_devs
)
940 uuid_from_super_imsm(st
, info
->uuid
);
942 memcpy(info
->uuid
, uuid_match_any
, sizeof(int[4]));
945 static int update_super_imsm(struct supertype
*st
, struct mdinfo
*info
,
946 char *update
, char *devname
, int verbose
,
947 int uuid_set
, char *homehost
)
951 /* For 'assemble' and 'force' we need to return non-zero if any
952 * change was made. For others, the return value is ignored.
953 * Update options are:
954 * force-one : This device looks a bit old but needs to be included,
955 * update age info appropriately.
956 * assemble: clear any 'faulty' flag to allow this device to
958 * force-array: Array is degraded but being forced, mark it clean
959 * if that will be needed to assemble it.
961 * newdev: not used ????
962 * grow: Array has gained a new device - this is currently for
964 * resync: mark as dirty so a resync will happen.
965 * name: update the name - preserving the homehost
967 * Following are not relevant for this imsm:
968 * sparc2.2 : update from old dodgey metadata
969 * super-minor: change the preferred_minor number
970 * summaries: update redundant counters.
971 * uuid: Change the uuid of the array to match watch is given
972 * homehost: update the recorded homehost
973 * _reshape_progress: record new reshape_progress position.
976 //struct intel_super *super = st->sb;
977 //struct imsm_super *mpb = super->mpb;
979 if (strcmp(update
, "grow") == 0) {
981 if (strcmp(update
, "resync") == 0) {
982 /* dev->vol.dirty = 1; */
985 /* IMSM has no concept of UUID or homehost */
990 static size_t disks_to_mpb_size(int disks
)
994 size
= sizeof(struct imsm_super
);
995 size
+= (disks
- 1) * sizeof(struct imsm_disk
);
996 size
+= 2 * sizeof(struct imsm_dev
);
997 /* up to 2 maps per raid device (-2 for imsm_maps in imsm_dev */
998 size
+= (4 - 2) * sizeof(struct imsm_map
);
999 /* 4 possible disk_ord_tbl's */
1000 size
+= 4 * (disks
- 1) * sizeof(__u32
);
1005 static __u64
avail_size_imsm(struct supertype
*st
, __u64 devsize
)
1007 if (devsize
< (MPB_SECTOR_CNT
+ IMSM_RESERVED_SECTORS
))
1010 return devsize
- (MPB_SECTOR_CNT
+ IMSM_RESERVED_SECTORS
);
1013 static int compare_super_imsm(struct supertype
*st
, struct supertype
*tst
)
1017 * 0 same, or first was empty, and second was copied
1018 * 1 second had wrong number
1020 * 3 wrong other info
1022 struct intel_super
*first
= st
->sb
;
1023 struct intel_super
*sec
= tst
->sb
;
1031 if (memcmp(first
->anchor
->sig
, sec
->anchor
->sig
, MAX_SIGNATURE_LENGTH
) != 0)
1034 /* if an anchor does not have num_raid_devs set then it is a free
1037 if (first
->anchor
->num_raid_devs
> 0 &&
1038 sec
->anchor
->num_raid_devs
> 0) {
1039 if (first
->anchor
->family_num
!= sec
->anchor
->family_num
)
1043 /* if 'first' is a spare promote it to a populated mpb with sec's
1046 if (first
->anchor
->num_raid_devs
== 0 &&
1047 sec
->anchor
->num_raid_devs
> 0) {
1050 /* we need to copy raid device info from sec if an allocation
1051 * fails here we don't associate the spare
1053 for (i
= 0; i
< sec
->anchor
->num_raid_devs
; i
++) {
1054 first
->dev_tbl
[i
] = malloc(sizeof(struct imsm_dev
));
1055 if (!first
->dev_tbl
) {
1057 free(first
->dev_tbl
[i
]);
1058 first
->dev_tbl
[i
] = NULL
;
1060 fprintf(stderr
, "imsm: failed to associate spare\n");
1063 *first
->dev_tbl
[i
] = *sec
->dev_tbl
[i
];
1066 first
->anchor
->num_raid_devs
= sec
->anchor
->num_raid_devs
;
1067 first
->anchor
->family_num
= sec
->anchor
->family_num
;
1073 static void fd2devname(int fd
, char *name
)
1082 if (fstat(fd
, &st
) != 0)
1084 sprintf(path
, "/sys/dev/block/%d:%d",
1085 major(st
.st_rdev
), minor(st
.st_rdev
));
1087 rv
= readlink(path
, dname
, sizeof(dname
));
1092 nm
= strrchr(dname
, '/');
1094 snprintf(name
, MAX_RAID_SERIAL_LEN
, "/dev/%s", nm
);
1098 extern int scsi_get_serial(int fd
, void *buf
, size_t buf_len
);
1100 static int imsm_read_serial(int fd
, char *devname
,
1101 __u8 serial
[MAX_RAID_SERIAL_LEN
])
1103 unsigned char scsi_serial
[255];
1109 memset(scsi_serial
, 0, sizeof(scsi_serial
));
1111 rv
= scsi_get_serial(fd
, scsi_serial
, sizeof(scsi_serial
));
1113 if (rv
&& check_env("IMSM_DEVNAME_AS_SERIAL")) {
1114 memset(serial
, 0, MAX_RAID_SERIAL_LEN
);
1115 fd2devname(fd
, (char *) serial
);
1122 Name
": Failed to retrieve serial for %s\n",
1127 /* trim leading whitespace */
1128 rsp_len
= scsi_serial
[3];
1129 rsp_buf
= (char *) &scsi_serial
[4];
1134 /* truncate len to the end of rsp_buf if necessary */
1135 if (c
+ MAX_RAID_SERIAL_LEN
> rsp_buf
+ rsp_len
)
1136 len
= rsp_len
- (c
- rsp_buf
);
1138 len
= MAX_RAID_SERIAL_LEN
;
1140 /* initialize the buffer and copy rsp_buf characters */
1141 memset(serial
, 0, MAX_RAID_SERIAL_LEN
);
1142 memcpy(serial
, c
, len
);
1144 /* trim trailing whitespace starting with the last character copied */
1145 c
= (char *) &serial
[len
- 1];
1146 while (isspace(*c
) || *c
== '\0')
1152 static int serialcmp(__u8
*s1
, __u8
*s2
)
1154 return strncmp((char *) s1
, (char *) s2
, MAX_RAID_SERIAL_LEN
);
1157 static void serialcpy(__u8
*dest
, __u8
*src
)
1159 strncpy((char *) dest
, (char *) src
, MAX_RAID_SERIAL_LEN
);
1162 static struct dl
*serial_to_dl(__u8
*serial
, struct intel_super
*super
)
1166 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
1167 if (serialcmp(dl
->serial
, serial
) == 0)
1174 load_imsm_disk(int fd
, struct intel_super
*super
, char *devname
, int keep_fd
)
1181 __u8 serial
[MAX_RAID_SERIAL_LEN
];
1183 rv
= imsm_read_serial(fd
, devname
, serial
);
1188 /* check if this is a disk we have seen before. it may be a spare in
1189 * super->disks while the current anchor believes it is a raid member,
1190 * check if we need to update dl->index
1192 dl
= serial_to_dl(serial
, super
);
1194 dl
= malloc(sizeof(*dl
));
1201 Name
": failed to allocate disk buffer for %s\n",
1208 dl
->major
= major(stb
.st_rdev
);
1209 dl
->minor
= minor(stb
.st_rdev
);
1210 dl
->next
= super
->disks
;
1211 dl
->fd
= keep_fd
? fd
: -1;
1212 dl
->devname
= devname
? strdup(devname
) : NULL
;
1213 serialcpy(dl
->serial
, serial
);
1216 } else if (keep_fd
) {
1221 /* look up this disk's index in the current anchor */
1222 for (i
= 0; i
< super
->anchor
->num_disks
; i
++) {
1223 struct imsm_disk
*disk_iter
;
1225 disk_iter
= __get_imsm_disk(super
->anchor
, i
);
1227 if (serialcmp(disk_iter
->serial
, dl
->serial
) == 0) {
1228 dl
->disk
= *disk_iter
;
1229 /* only set index on disks that are a member of a
1230 * populated contianer, i.e. one with raid_devs
1232 if (dl
->disk
.status
& FAILED_DISK
)
1234 else if (dl
->disk
.status
& SPARE_DISK
)
1243 /* no match, maybe a stale failed drive */
1244 if (i
== super
->anchor
->num_disks
&& dl
->index
>= 0) {
1245 dl
->disk
= *__get_imsm_disk(super
->anchor
, dl
->index
);
1246 if (dl
->disk
.status
& FAILED_DISK
)
1256 static void imsm_copy_dev(struct imsm_dev
*dest
, struct imsm_dev
*src
)
1258 memcpy(dest
, src
, sizeof_imsm_dev(src
, 0));
1262 /* When migrating map0 contains the 'destination' state while map1
1263 * contains the current state. When not migrating map0 contains the
1264 * current state. This routine assumes that map[0].map_state is set to
1265 * the current array state before being called.
1267 * Migration is indicated by one of the following states
1268 * 1/ Idle (migr_state=0 map0state=normal||unitialized||degraded||failed)
1269 * 2/ Initialize (migr_state=1 migr_type=MIGR_INIT map0state=normal
1270 * map1state=unitialized)
1271 * 3/ Verify (Resync) (migr_state=1 migr_type=MIGR_REBUILD map0state=normal
1273 * 4/ Rebuild (migr_state=1 migr_type=MIGR_REBUILD map0state=normal
1274 * map1state=degraded)
1276 static void migrate(struct imsm_dev
*dev
, __u8 to_state
, int rebuild_resync
)
1278 struct imsm_map
*dest
;
1279 struct imsm_map
*src
= get_imsm_map(dev
, 0);
1281 dev
->vol
.migr_state
= 1;
1282 dev
->vol
.migr_type
= rebuild_resync
;
1283 dev
->vol
.curr_migr_unit
= 0;
1284 dest
= get_imsm_map(dev
, 1);
1286 memcpy(dest
, src
, sizeof_imsm_map(src
));
1287 src
->map_state
= to_state
;
1290 static void end_migration(struct imsm_dev
*dev
, __u8 map_state
)
1292 struct imsm_map
*map
= get_imsm_map(dev
, 0);
1294 dev
->vol
.migr_state
= 0;
1295 dev
->vol
.curr_migr_unit
= 0;
1296 map
->map_state
= map_state
;
1300 static int parse_raid_devices(struct intel_super
*super
)
1303 struct imsm_dev
*dev_new
;
1304 size_t len
, len_migr
;
1305 size_t space_needed
= 0;
1306 struct imsm_super
*mpb
= super
->anchor
;
1308 for (i
= 0; i
< super
->anchor
->num_raid_devs
; i
++) {
1309 struct imsm_dev
*dev_iter
= __get_imsm_dev(super
->anchor
, i
);
1311 len
= sizeof_imsm_dev(dev_iter
, 0);
1312 len_migr
= sizeof_imsm_dev(dev_iter
, 1);
1314 space_needed
+= len_migr
- len
;
1316 dev_new
= malloc(len_migr
);
1319 imsm_copy_dev(dev_new
, dev_iter
);
1320 super
->dev_tbl
[i
] = dev_new
;
1323 /* ensure that super->buf is large enough when all raid devices
1326 if (__le32_to_cpu(mpb
->mpb_size
) + space_needed
> super
->len
) {
1329 len
= ROUND_UP(__le32_to_cpu(mpb
->mpb_size
) + space_needed
, 512);
1330 if (posix_memalign(&buf
, 512, len
) != 0)
1333 memcpy(buf
, super
->buf
, len
);
1342 /* retrieve a pointer to the bbm log which starts after all raid devices */
1343 struct bbm_log
*__get_imsm_bbm_log(struct imsm_super
*mpb
)
1347 if (__le32_to_cpu(mpb
->bbm_log_size
)) {
1349 ptr
+= mpb
->mpb_size
- __le32_to_cpu(mpb
->bbm_log_size
);
1355 static void __free_imsm(struct intel_super
*super
, int free_disks
);
1357 /* load_imsm_mpb - read matrix metadata
1358 * allocates super->mpb to be freed by free_super
1360 static int load_imsm_mpb(int fd
, struct intel_super
*super
, char *devname
)
1362 unsigned long long dsize
;
1363 unsigned long long sectors
;
1365 struct imsm_super
*anchor
;
1369 get_dev_size(fd
, NULL
, &dsize
);
1371 if (lseek64(fd
, dsize
- (512 * 2), SEEK_SET
) < 0) {
1374 Name
": Cannot seek to anchor block on %s: %s\n",
1375 devname
, strerror(errno
));
1379 if (posix_memalign((void**)&anchor
, 512, 512) != 0) {
1382 Name
": Failed to allocate imsm anchor buffer"
1383 " on %s\n", devname
);
1386 if (read(fd
, anchor
, 512) != 512) {
1389 Name
": Cannot read anchor block on %s: %s\n",
1390 devname
, strerror(errno
));
1395 if (strncmp((char *) anchor
->sig
, MPB_SIGNATURE
, MPB_SIG_LEN
) != 0) {
1398 Name
": no IMSM anchor on %s\n", devname
);
1403 __free_imsm(super
, 0);
1404 super
->len
= ROUND_UP(anchor
->mpb_size
, 512);
1405 if (posix_memalign(&super
->buf
, 512, super
->len
) != 0) {
1408 Name
": unable to allocate %zu byte mpb buffer\n",
1413 memcpy(super
->buf
, anchor
, 512);
1415 sectors
= mpb_sectors(anchor
) - 1;
1418 rc
= load_imsm_disk(fd
, super
, devname
, 0);
1420 rc
= parse_raid_devices(super
);
1424 /* read the extended mpb */
1425 if (lseek64(fd
, dsize
- (512 * (2 + sectors
)), SEEK_SET
) < 0) {
1428 Name
": Cannot seek to extended mpb on %s: %s\n",
1429 devname
, strerror(errno
));
1433 if (read(fd
, super
->buf
+ 512, super
->len
- 512) != super
->len
- 512) {
1436 Name
": Cannot read extended mpb on %s: %s\n",
1437 devname
, strerror(errno
));
1441 check_sum
= __gen_imsm_checksum(super
->anchor
);
1442 if (check_sum
!= __le32_to_cpu(super
->anchor
->check_sum
)) {
1445 Name
": IMSM checksum %x != %x on %s\n",
1446 check_sum
, __le32_to_cpu(super
->anchor
->check_sum
),
1451 /* FIXME the BBM log is disk specific so we cannot use this global
1452 * buffer for all disks. Ok for now since we only look at the global
1453 * bbm_log_size parameter to gate assembly
1455 super
->bbm_log
= __get_imsm_bbm_log(super
->anchor
);
1457 rc
= load_imsm_disk(fd
, super
, devname
, 0);
1459 rc
= parse_raid_devices(super
);
1464 static void __free_imsm_disk(struct dl
*d
)
1475 static void free_imsm_disks(struct intel_super
*super
)
1479 while (super
->disks
) {
1481 super
->disks
= d
->next
;
1482 __free_imsm_disk(d
);
1484 while (super
->missing
) {
1486 super
->missing
= d
->next
;
1487 __free_imsm_disk(d
);
1492 /* free all the pieces hanging off of a super pointer */
1493 static void __free_imsm(struct intel_super
*super
, int free_disks
)
1502 free_imsm_disks(super
);
1503 for (i
= 0; i
< IMSM_MAX_RAID_DEVS
; i
++)
1504 if (super
->dev_tbl
[i
]) {
1505 free(super
->dev_tbl
[i
]);
1506 super
->dev_tbl
[i
] = NULL
;
1510 static void free_imsm(struct intel_super
*super
)
1512 __free_imsm(super
, 1);
1516 static void free_super_imsm(struct supertype
*st
)
1518 struct intel_super
*super
= st
->sb
;
1527 static struct intel_super
*alloc_super(int creating_imsm
)
1529 struct intel_super
*super
= malloc(sizeof(*super
));
1532 memset(super
, 0, sizeof(*super
));
1533 super
->creating_imsm
= creating_imsm
;
1534 super
->current_vol
= -1;
1535 super
->create_offset
= ~((__u32
) 0);
1542 /* find_missing - helper routine for load_super_imsm_all that identifies
1543 * disks that have disappeared from the system. This routine relies on
1544 * the mpb being uptodate, which it is at load time.
1546 static int find_missing(struct intel_super
*super
)
1549 struct imsm_super
*mpb
= super
->anchor
;
1551 struct imsm_disk
*disk
;
1553 for (i
= 0; i
< mpb
->num_disks
; i
++) {
1554 disk
= __get_imsm_disk(mpb
, i
);
1555 dl
= serial_to_dl(disk
->serial
, super
);
1558 /* ok we have a 'disk' without a live entry in
1561 if (disk
->status
& FAILED_DISK
|| !(disk
->status
& USABLE_DISK
))
1562 continue; /* never mind, already marked */
1564 dl
= malloc(sizeof(*dl
));
1570 dl
->devname
= strdup("missing");
1572 serialcpy(dl
->serial
, disk
->serial
);
1574 dl
->next
= super
->missing
;
1575 super
->missing
= dl
;
1581 static int load_super_imsm_all(struct supertype
*st
, int fd
, void **sbp
,
1582 char *devname
, int keep_fd
)
1585 struct intel_super
*super
;
1586 struct mdinfo
*sd
, *best
= NULL
;
1593 /* check if this disk is a member of an active array */
1594 sra
= sysfs_read(fd
, 0, GET_LEVEL
|GET_VERSION
|GET_DEVS
|GET_STATE
);
1598 if (sra
->array
.major_version
!= -1 ||
1599 sra
->array
.minor_version
!= -2 ||
1600 strcmp(sra
->text_version
, "imsm") != 0)
1603 super
= alloc_super(0);
1607 /* find the most up to date disk in this array, skipping spares */
1608 for (sd
= sra
->devs
; sd
; sd
= sd
->next
) {
1609 sprintf(nm
, "%d:%d", sd
->disk
.major
, sd
->disk
.minor
);
1610 dfd
= dev_open(nm
, keep_fd
? O_RDWR
: O_RDONLY
);
1615 rv
= load_imsm_mpb(dfd
, super
, NULL
);
1619 if (super
->anchor
->num_raid_devs
== 0)
1622 gen
= __le32_to_cpu(super
->anchor
->generation_num
);
1623 if (!best
|| gen
> bestgen
) {
1638 /* load the most up to date anchor */
1639 sprintf(nm
, "%d:%d", best
->disk
.major
, best
->disk
.minor
);
1640 dfd
= dev_open(nm
, O_RDONLY
);
1645 rv
= load_imsm_mpb(dfd
, super
, NULL
);
1652 /* re-parse the disk list with the current anchor */
1653 for (sd
= sra
->devs
; sd
; sd
= sd
->next
) {
1654 sprintf(nm
, "%d:%d", sd
->disk
.major
, sd
->disk
.minor
);
1655 dfd
= dev_open(nm
, keep_fd
? O_RDWR
: O_RDONLY
);
1660 load_imsm_disk(dfd
, super
, NULL
, keep_fd
);
1666 if (find_missing(super
) != 0) {
1671 if (st
->subarray
[0]) {
1672 if (atoi(st
->subarray
) <= super
->anchor
->num_raid_devs
)
1673 super
->current_vol
= atoi(st
->subarray
);
1679 st
->container_dev
= fd2devnum(fd
);
1680 if (st
->ss
== NULL
) {
1681 st
->ss
= &super_imsm
;
1682 st
->minor_version
= 0;
1683 st
->max_devs
= IMSM_MAX_DEVICES
;
1685 st
->loaded_container
= 1;
1691 static int load_super_imsm(struct supertype
*st
, int fd
, char *devname
)
1693 struct intel_super
*super
;
1697 if (load_super_imsm_all(st
, fd
, &st
->sb
, devname
, 1) == 0)
1700 if (st
->subarray
[0])
1701 return 1; /* FIXME */
1703 super
= alloc_super(0);
1706 Name
": malloc of %zu failed.\n",
1711 rv
= load_imsm_mpb(fd
, super
, devname
);
1716 Name
": Failed to load all information "
1717 "sections on %s\n", devname
);
1723 if (st
->ss
== NULL
) {
1724 st
->ss
= &super_imsm
;
1725 st
->minor_version
= 0;
1726 st
->max_devs
= IMSM_MAX_DEVICES
;
1728 st
->loaded_container
= 0;
1733 static __u16
info_to_blocks_per_strip(mdu_array_info_t
*info
)
1735 if (info
->level
== 1)
1737 return info
->chunk_size
>> 9;
1740 static __u32
info_to_num_data_stripes(mdu_array_info_t
*info
)
1744 num_stripes
= (info
->size
* 2) / info_to_blocks_per_strip(info
);
1745 if (info
->level
== 1)
1751 static __u32
info_to_blocks_per_member(mdu_array_info_t
*info
)
1753 return (info
->size
* 2) & ~(info_to_blocks_per_strip(info
) - 1);
1756 static void imsm_update_version_info(struct intel_super
*super
)
1758 /* update the version and attributes */
1759 struct imsm_super
*mpb
= super
->anchor
;
1761 struct imsm_dev
*dev
;
1762 struct imsm_map
*map
;
1765 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
1766 dev
= get_imsm_dev(super
, i
);
1767 map
= get_imsm_map(dev
, 0);
1768 if (__le32_to_cpu(dev
->size_high
) > 0)
1769 mpb
->attributes
|= MPB_ATTRIB_2TB
;
1771 /* FIXME detect when an array spans a port multiplier */
1773 mpb
->attributes
|= MPB_ATTRIB_PM
;
1776 if (mpb
->num_raid_devs
> 1 ||
1777 mpb
->attributes
!= MPB_ATTRIB_CHECKSUM_VERIFY
) {
1778 version
= MPB_VERSION_ATTRIBS
;
1779 switch (get_imsm_raid_level(map
)) {
1780 case 0: mpb
->attributes
|= MPB_ATTRIB_RAID0
; break;
1781 case 1: mpb
->attributes
|= MPB_ATTRIB_RAID1
; break;
1782 case 10: mpb
->attributes
|= MPB_ATTRIB_RAID10
; break;
1783 case 5: mpb
->attributes
|= MPB_ATTRIB_RAID5
; break;
1786 if (map
->num_members
>= 5)
1787 version
= MPB_VERSION_5OR6_DISK_ARRAY
;
1788 else if (dev
->status
== DEV_CLONE_N_GO
)
1789 version
= MPB_VERSION_CNG
;
1790 else if (get_imsm_raid_level(map
) == 5)
1791 version
= MPB_VERSION_RAID5
;
1792 else if (map
->num_members
>= 3)
1793 version
= MPB_VERSION_3OR4_DISK_ARRAY
;
1794 else if (get_imsm_raid_level(map
) == 1)
1795 version
= MPB_VERSION_RAID1
;
1797 version
= MPB_VERSION_RAID0
;
1799 strcpy(((char *) mpb
->sig
) + strlen(MPB_SIGNATURE
), version
);
1803 static int init_super_imsm_volume(struct supertype
*st
, mdu_array_info_t
*info
,
1804 unsigned long long size
, char *name
,
1805 char *homehost
, int *uuid
)
1807 /* We are creating a volume inside a pre-existing container.
1808 * so st->sb is already set.
1810 struct intel_super
*super
= st
->sb
;
1811 struct imsm_super
*mpb
= super
->anchor
;
1812 struct imsm_dev
*dev
;
1813 struct imsm_vol
*vol
;
1814 struct imsm_map
*map
;
1815 int idx
= mpb
->num_raid_devs
;
1817 unsigned long long array_blocks
;
1818 size_t size_old
, size_new
;
1820 if (mpb
->num_raid_devs
>= 2) {
1821 fprintf(stderr
, Name
": This imsm-container already has the "
1822 "maximum of 2 volumes\n");
1826 /* ensure the mpb is large enough for the new data */
1827 size_old
= __le32_to_cpu(mpb
->mpb_size
);
1828 size_new
= disks_to_mpb_size(info
->nr_disks
);
1829 if (size_new
> size_old
) {
1831 size_t size_round
= ROUND_UP(size_new
, 512);
1833 if (posix_memalign(&mpb_new
, 512, size_round
) != 0) {
1834 fprintf(stderr
, Name
": could not allocate new mpb\n");
1837 memcpy(mpb_new
, mpb
, size_old
);
1840 super
->anchor
= mpb_new
;
1841 mpb
->mpb_size
= __cpu_to_le32(size_new
);
1842 memset(mpb_new
+ size_old
, 0, size_round
- size_old
);
1844 super
->current_vol
= idx
;
1845 /* when creating the first raid device in this container set num_disks
1846 * to zero, i.e. delete this spare and add raid member devices in
1847 * add_to_super_imsm_volume()
1849 if (super
->current_vol
== 0)
1851 sprintf(st
->subarray
, "%d", idx
);
1852 dev
= malloc(sizeof(*dev
) + sizeof(__u32
) * (info
->raid_disks
- 1));
1854 fprintf(stderr
, Name
": could not allocate raid device\n");
1857 strncpy((char *) dev
->volume
, name
, MAX_RAID_SERIAL_LEN
);
1858 if (info
->level
== 1)
1859 array_blocks
= info_to_blocks_per_member(info
);
1861 array_blocks
= calc_array_size(info
->level
, info
->raid_disks
,
1862 info
->layout
, info
->chunk_size
,
1864 dev
->size_low
= __cpu_to_le32((__u32
) array_blocks
);
1865 dev
->size_high
= __cpu_to_le32((__u32
) (array_blocks
>> 32));
1866 dev
->status
= __cpu_to_le32(0);
1867 dev
->reserved_blocks
= __cpu_to_le32(0);
1869 vol
->migr_state
= 0;
1870 vol
->migr_type
= MIGR_INIT
;
1872 vol
->curr_migr_unit
= 0;
1873 map
= get_imsm_map(dev
, 0);
1874 map
->pba_of_lba0
= __cpu_to_le32(super
->create_offset
);
1875 map
->blocks_per_member
= __cpu_to_le32(info_to_blocks_per_member(info
));
1876 map
->blocks_per_strip
= __cpu_to_le16(info_to_blocks_per_strip(info
));
1877 map
->num_data_stripes
= __cpu_to_le32(info_to_num_data_stripes(info
));
1878 map
->map_state
= info
->level
? IMSM_T_STATE_UNINITIALIZED
:
1879 IMSM_T_STATE_NORMAL
;
1881 if (info
->level
== 1 && info
->raid_disks
> 2) {
1882 fprintf(stderr
, Name
": imsm does not support more than 2 disks"
1883 "in a raid1 volume\n");
1886 if (info
->level
== 10) {
1887 map
->raid_level
= 1;
1888 map
->num_domains
= info
->raid_disks
/ 2;
1890 map
->raid_level
= info
->level
;
1891 map
->num_domains
= !!map
->raid_level
;
1894 map
->num_members
= info
->raid_disks
;
1895 for (i
= 0; i
< map
->num_members
; i
++) {
1896 /* initialized in add_to_super */
1897 set_imsm_ord_tbl_ent(map
, i
, 0);
1899 mpb
->num_raid_devs
++;
1900 super
->dev_tbl
[super
->current_vol
] = dev
;
1902 imsm_update_version_info(super
);
1907 static int init_super_imsm(struct supertype
*st
, mdu_array_info_t
*info
,
1908 unsigned long long size
, char *name
,
1909 char *homehost
, int *uuid
)
1911 /* This is primarily called by Create when creating a new array.
1912 * We will then get add_to_super called for each component, and then
1913 * write_init_super called to write it out to each device.
1914 * For IMSM, Create can create on fresh devices or on a pre-existing
1916 * To create on a pre-existing array a different method will be called.
1917 * This one is just for fresh drives.
1919 struct intel_super
*super
;
1920 struct imsm_super
*mpb
;
1929 return init_super_imsm_volume(st
, info
, size
, name
, homehost
,
1932 super
= alloc_super(1);
1935 mpb_size
= disks_to_mpb_size(info
->nr_disks
);
1936 if (posix_memalign(&super
->buf
, 512, mpb_size
) != 0) {
1941 memset(mpb
, 0, mpb_size
);
1943 mpb
->attributes
= MPB_ATTRIB_CHECKSUM_VERIFY
;
1945 version
= (char *) mpb
->sig
;
1946 strcpy(version
, MPB_SIGNATURE
);
1947 version
+= strlen(MPB_SIGNATURE
);
1948 strcpy(version
, MPB_VERSION_RAID0
);
1949 mpb
->mpb_size
= mpb_size
;
1956 static int add_to_super_imsm_volume(struct supertype
*st
, mdu_disk_info_t
*dk
,
1957 int fd
, char *devname
)
1959 struct intel_super
*super
= st
->sb
;
1960 struct imsm_super
*mpb
= super
->anchor
;
1962 struct imsm_dev
*dev
;
1963 struct imsm_map
*map
;
1965 dev
= get_imsm_dev(super
, super
->current_vol
);
1966 map
= get_imsm_map(dev
, 0);
1968 if (! (dk
->state
& (1<<MD_DISK_SYNC
))) {
1969 fprintf(stderr
, Name
": %s: Cannot add spare devices to IMSM volume\n",
1974 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
1975 if (dl
->major
== dk
->major
&&
1976 dl
->minor
== dk
->minor
)
1980 fprintf(stderr
, Name
": %s is not a member of the same container\n", devname
);
1984 /* add a pristine spare to the metadata */
1985 if (dl
->index
< 0) {
1986 dl
->index
= super
->anchor
->num_disks
;
1987 super
->anchor
->num_disks
++;
1989 set_imsm_ord_tbl_ent(map
, dk
->number
, dl
->index
);
1990 dl
->disk
.status
= CONFIGURED_DISK
| USABLE_DISK
;
1992 /* if we are creating the first raid device update the family number */
1993 if (super
->current_vol
== 0) {
1995 struct imsm_dev
*_dev
= __get_imsm_dev(mpb
, 0);
1996 struct imsm_disk
*_disk
= __get_imsm_disk(mpb
, dl
->index
);
2000 sum
= __gen_imsm_checksum(mpb
);
2001 mpb
->family_num
= __cpu_to_le32(sum
);
2007 static int add_to_super_imsm(struct supertype
*st
, mdu_disk_info_t
*dk
,
2008 int fd
, char *devname
)
2010 struct intel_super
*super
= st
->sb
;
2012 unsigned long long size
;
2017 if (super
->current_vol
>= 0)
2018 return add_to_super_imsm_volume(st
, dk
, fd
, devname
);
2021 dd
= malloc(sizeof(*dd
));
2024 Name
": malloc failed %s:%d.\n", __func__
, __LINE__
);
2027 memset(dd
, 0, sizeof(*dd
));
2028 dd
->major
= major(stb
.st_rdev
);
2029 dd
->minor
= minor(stb
.st_rdev
);
2031 dd
->devname
= devname
? strdup(devname
) : NULL
;
2033 rv
= imsm_read_serial(fd
, devname
, dd
->serial
);
2036 Name
": failed to retrieve scsi serial, aborting\n");
2041 get_dev_size(fd
, NULL
, &size
);
2043 serialcpy(dd
->disk
.serial
, dd
->serial
);
2044 dd
->disk
.total_blocks
= __cpu_to_le32(size
);
2045 dd
->disk
.status
= USABLE_DISK
| SPARE_DISK
;
2046 if (sysfs_disk_to_scsi_id(fd
, &id
) == 0)
2047 dd
->disk
.scsi_id
= __cpu_to_le32(id
);
2049 dd
->disk
.scsi_id
= __cpu_to_le32(0);
2051 if (st
->update_tail
) {
2052 dd
->next
= super
->add
;
2055 dd
->next
= super
->disks
;
2062 static int store_imsm_mpb(int fd
, struct intel_super
*super
);
2064 /* spare records have their own family number and do not have any defined raid
2067 static int write_super_imsm_spares(struct intel_super
*super
, int doclose
)
2069 struct imsm_super mpb_save
;
2070 struct imsm_super
*mpb
= super
->anchor
;
2075 mpb
->num_raid_devs
= 0;
2077 mpb
->mpb_size
= sizeof(struct imsm_super
);
2078 mpb
->generation_num
= __cpu_to_le32(1UL);
2080 for (d
= super
->disks
; d
; d
= d
->next
) {
2084 mpb
->disk
[0] = d
->disk
;
2085 sum
= __gen_imsm_checksum(mpb
);
2086 mpb
->family_num
= __cpu_to_le32(sum
);
2087 sum
= __gen_imsm_checksum(mpb
);
2088 mpb
->check_sum
= __cpu_to_le32(sum
);
2090 if (store_imsm_mpb(d
->fd
, super
)) {
2091 fprintf(stderr
, "%s: failed for device %d:%d %s\n",
2092 __func__
, d
->major
, d
->minor
, strerror(errno
));
2106 static int write_super_imsm(struct intel_super
*super
, int doclose
)
2108 struct imsm_super
*mpb
= super
->anchor
;
2114 __u32 mpb_size
= sizeof(struct imsm_super
) - sizeof(struct imsm_disk
);
2116 /* 'generation' is incremented everytime the metadata is written */
2117 generation
= __le32_to_cpu(mpb
->generation_num
);
2119 mpb
->generation_num
= __cpu_to_le32(generation
);
2121 mpb_size
+= sizeof(struct imsm_disk
) * mpb
->num_disks
;
2122 for (d
= super
->disks
; d
; d
= d
->next
) {
2126 mpb
->disk
[d
->index
] = d
->disk
;
2128 for (d
= super
->missing
; d
; d
= d
->next
)
2129 mpb
->disk
[d
->index
] = d
->disk
;
2131 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
2132 struct imsm_dev
*dev
= __get_imsm_dev(mpb
, i
);
2134 imsm_copy_dev(dev
, super
->dev_tbl
[i
]);
2135 mpb_size
+= sizeof_imsm_dev(dev
, 0);
2137 mpb_size
+= __le32_to_cpu(mpb
->bbm_log_size
);
2138 mpb
->mpb_size
= __cpu_to_le32(mpb_size
);
2140 /* recalculate checksum */
2141 sum
= __gen_imsm_checksum(mpb
);
2142 mpb
->check_sum
= __cpu_to_le32(sum
);
2144 /* write the mpb for disks that compose raid devices */
2145 for (d
= super
->disks
; d
; d
= d
->next
) {
2148 if (store_imsm_mpb(d
->fd
, super
))
2149 fprintf(stderr
, "%s: failed for device %d:%d %s\n",
2150 __func__
, d
->major
, d
->minor
, strerror(errno
));
2158 return write_super_imsm_spares(super
, doclose
);
2164 static int create_array(struct supertype
*st
)
2167 struct imsm_update_create_array
*u
;
2168 struct intel_super
*super
= st
->sb
;
2169 struct imsm_dev
*dev
= get_imsm_dev(super
, super
->current_vol
);
2170 struct imsm_map
*map
= get_imsm_map(dev
, 0);
2171 struct disk_info
*inf
;
2172 struct imsm_disk
*disk
;
2176 len
= sizeof(*u
) - sizeof(*dev
) + sizeof_imsm_dev(dev
, 0) +
2177 sizeof(*inf
) * map
->num_members
;
2180 fprintf(stderr
, "%s: failed to allocate update buffer\n",
2185 u
->type
= update_create_array
;
2186 u
->dev_idx
= super
->current_vol
;
2187 imsm_copy_dev(&u
->dev
, dev
);
2188 inf
= get_disk_info(u
);
2189 for (i
= 0; i
< map
->num_members
; i
++) {
2190 idx
= get_imsm_disk_idx(dev
, i
);
2191 disk
= get_imsm_disk(super
, idx
);
2192 serialcpy(inf
[i
].serial
, disk
->serial
);
2194 append_metadata_update(st
, u
, len
);
2199 static int _add_disk(struct supertype
*st
)
2201 struct intel_super
*super
= st
->sb
;
2203 struct imsm_update_add_disk
*u
;
2211 fprintf(stderr
, "%s: failed to allocate update buffer\n",
2216 u
->type
= update_add_disk
;
2217 append_metadata_update(st
, u
, len
);
2222 static int write_init_super_imsm(struct supertype
*st
)
2224 if (st
->update_tail
) {
2225 /* queue the recently created array / added disk
2226 * as a metadata update */
2227 struct intel_super
*super
= st
->sb
;
2231 /* determine if we are creating a volume or adding a disk */
2232 if (super
->current_vol
< 0) {
2233 /* in the add disk case we are running in mdmon
2234 * context, so don't close fd's
2236 return _add_disk(st
);
2238 rv
= create_array(st
);
2240 for (d
= super
->disks
; d
; d
= d
->next
) {
2247 return write_super_imsm(st
->sb
, 1);
2251 static int store_zero_imsm(struct supertype
*st
, int fd
)
2253 unsigned long long dsize
;
2256 get_dev_size(fd
, NULL
, &dsize
);
2258 /* first block is stored on second to last sector of the disk */
2259 if (lseek64(fd
, dsize
- (512 * 2), SEEK_SET
) < 0)
2262 if (posix_memalign(&buf
, 512, 512) != 0)
2265 memset(buf
, 0, 512);
2266 if (write(fd
, buf
, 512) != 512)
2271 static int imsm_bbm_log_size(struct imsm_super
*mpb
)
2273 return __le32_to_cpu(mpb
->bbm_log_size
);
2277 static int validate_geometry_imsm_container(struct supertype
*st
, int level
,
2278 int layout
, int raiddisks
, int chunk
,
2279 unsigned long long size
, char *dev
,
2280 unsigned long long *freesize
,
2284 unsigned long long ldsize
;
2286 if (level
!= LEVEL_CONTAINER
)
2291 fd
= open(dev
, O_RDONLY
|O_EXCL
, 0);
2294 fprintf(stderr
, Name
": imsm: Cannot open %s: %s\n",
2295 dev
, strerror(errno
));
2298 if (!get_dev_size(fd
, dev
, &ldsize
)) {
2304 *freesize
= avail_size_imsm(st
, ldsize
>> 9);
2309 static unsigned long long find_size(struct extent
*e
, int *idx
, int num_extents
)
2311 const unsigned long long base_start
= e
[*idx
].start
;
2312 unsigned long long end
= base_start
+ e
[*idx
].size
;
2315 if (base_start
== end
)
2319 for (i
= *idx
; i
< num_extents
; i
++) {
2320 /* extend overlapping extents */
2321 if (e
[i
].start
>= base_start
&&
2322 e
[i
].start
<= end
) {
2325 if (e
[i
].start
+ e
[i
].size
> end
)
2326 end
= e
[i
].start
+ e
[i
].size
;
2327 } else if (e
[i
].start
> end
) {
2333 return end
- base_start
;
2336 static unsigned long long merge_extents(struct intel_super
*super
, int sum_extents
)
2338 /* build a composite disk with all known extents and generate a new
2339 * 'maxsize' given the "all disks in an array must share a common start
2340 * offset" constraint
2342 struct extent
*e
= calloc(sum_extents
, sizeof(*e
));
2346 unsigned long long pos
;
2347 unsigned long long start
;
2348 unsigned long long maxsize
;
2349 unsigned long reserve
;
2352 return ~0ULL; /* error */
2354 /* coalesce and sort all extents. also, check to see if we need to
2355 * reserve space between member arrays
2358 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
2361 for (i
= 0; i
< dl
->extent_cnt
; i
++)
2364 qsort(e
, sum_extents
, sizeof(*e
), cmp_extent
);
2369 while (i
< sum_extents
) {
2370 e
[j
].start
= e
[i
].start
;
2371 e
[j
].size
= find_size(e
, &i
, sum_extents
);
2373 if (e
[j
-1].size
== 0)
2382 unsigned long long esize
;
2384 esize
= e
[i
].start
- pos
;
2385 if (esize
>= maxsize
) {
2390 pos
= e
[i
].start
+ e
[i
].size
;
2392 } while (e
[i
-1].size
);
2395 if (start_extent
> 0)
2396 reserve
= IMSM_RESERVED_SECTORS
; /* gap between raid regions */
2400 if (maxsize
< reserve
)
2403 super
->create_offset
= ~((__u32
) 0);
2404 if (start
+ reserve
> super
->create_offset
)
2405 return ~0ULL; /* start overflows create_offset */
2406 super
->create_offset
= start
+ reserve
;
2408 return maxsize
- reserve
;
2411 /* validate_geometry_imsm_volume - lifted from validate_geometry_ddf_bvd
2412 * FIX ME add ahci details
2414 static int validate_geometry_imsm_volume(struct supertype
*st
, int level
,
2415 int layout
, int raiddisks
, int chunk
,
2416 unsigned long long size
, char *dev
,
2417 unsigned long long *freesize
,
2421 struct intel_super
*super
= st
->sb
;
2423 unsigned long long pos
= 0;
2424 unsigned long long maxsize
;
2428 if (level
== LEVEL_CONTAINER
)
2431 if (level
== 1 && raiddisks
> 2) {
2433 fprintf(stderr
, Name
": imsm does not support more "
2434 "than 2 in a raid1 configuration\n");
2438 /* We must have the container info already read in. */
2443 /* General test: make sure there is space for
2444 * 'raiddisks' device extents of size 'size' at a given
2447 unsigned long long minsize
= size
*2 /* convert to blocks */;
2448 unsigned long long start_offset
= ~0ULL;
2451 minsize
= MPB_SECTOR_CNT
+ IMSM_RESERVED_SECTORS
;
2452 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
2457 e
= get_extents(super
, dl
);
2460 unsigned long long esize
;
2461 esize
= e
[i
].start
- pos
;
2462 if (esize
>= minsize
)
2464 if (found
&& start_offset
== ~0ULL) {
2467 } else if (found
&& pos
!= start_offset
) {
2471 pos
= e
[i
].start
+ e
[i
].size
;
2473 } while (e
[i
-1].size
);
2478 if (dcnt
< raiddisks
) {
2480 fprintf(stderr
, Name
": imsm: Not enough "
2481 "devices with space for this array "
2489 /* This device must be a member of the set */
2490 if (stat(dev
, &stb
) < 0)
2492 if ((S_IFMT
& stb
.st_mode
) != S_IFBLK
)
2494 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
2495 if (dl
->major
== major(stb
.st_rdev
) &&
2496 dl
->minor
== minor(stb
.st_rdev
))
2501 fprintf(stderr
, Name
": %s is not in the "
2502 "same imsm set\n", dev
);
2506 /* retrieve the largest free space block */
2507 e
= get_extents(super
, dl
);
2512 unsigned long long esize
;
2514 esize
= e
[i
].start
- pos
;
2515 if (esize
>= maxsize
)
2517 pos
= e
[i
].start
+ e
[i
].size
;
2519 } while (e
[i
-1].size
);
2524 fprintf(stderr
, Name
": unable to determine free space for: %s\n",
2528 if (maxsize
< size
) {
2530 fprintf(stderr
, Name
": %s not enough space (%llu < %llu)\n",
2531 dev
, maxsize
, size
);
2535 /* count total number of extents for merge */
2537 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
2539 i
+= dl
->extent_cnt
;
2541 maxsize
= merge_extents(super
, i
);
2542 if (maxsize
< size
) {
2544 fprintf(stderr
, Name
": not enough space after merge (%llu < %llu)\n",
2547 } else if (maxsize
== ~0ULL) {
2549 fprintf(stderr
, Name
": failed to merge %d extents\n", i
);
2553 *freesize
= maxsize
;
2558 static int validate_geometry_imsm(struct supertype
*st
, int level
, int layout
,
2559 int raiddisks
, int chunk
, unsigned long long size
,
2560 char *dev
, unsigned long long *freesize
,
2566 /* if given unused devices create a container
2567 * if given given devices in a container create a member volume
2569 if (level
== LEVEL_CONTAINER
) {
2570 /* Must be a fresh device to add to a container */
2571 return validate_geometry_imsm_container(st
, level
, layout
,
2572 raiddisks
, chunk
, size
,
2578 /* creating in a given container */
2579 return validate_geometry_imsm_volume(st
, level
, layout
,
2580 raiddisks
, chunk
, size
,
2581 dev
, freesize
, verbose
);
2584 /* limit creation to the following levels */
2596 /* This device needs to be a device in an 'imsm' container */
2597 fd
= open(dev
, O_RDONLY
|O_EXCL
, 0);
2601 Name
": Cannot create this array on device %s\n",
2606 if (errno
!= EBUSY
|| (fd
= open(dev
, O_RDONLY
, 0)) < 0) {
2608 fprintf(stderr
, Name
": Cannot open %s: %s\n",
2609 dev
, strerror(errno
));
2612 /* Well, it is in use by someone, maybe an 'imsm' container. */
2613 cfd
= open_container(fd
);
2617 fprintf(stderr
, Name
": Cannot use %s: It is busy\n",
2621 sra
= sysfs_read(cfd
, 0, GET_VERSION
);
2623 if (sra
&& sra
->array
.major_version
== -1 &&
2624 strcmp(sra
->text_version
, "imsm") == 0) {
2625 /* This is a member of a imsm container. Load the container
2626 * and try to create a volume
2628 struct intel_super
*super
;
2630 if (load_super_imsm_all(st
, cfd
, (void **) &super
, NULL
, 1) == 0) {
2632 st
->container_dev
= fd2devnum(cfd
);
2634 return validate_geometry_imsm_volume(st
, level
, layout
,
2640 } else /* may belong to another container */
2645 #endif /* MDASSEMBLE */
2647 static struct mdinfo
*container_content_imsm(struct supertype
*st
)
2649 /* Given a container loaded by load_super_imsm_all,
2650 * extract information about all the arrays into
2653 * For each imsm_dev create an mdinfo, fill it in,
2654 * then look for matching devices in super->disks
2655 * and create appropriate device mdinfo.
2657 struct intel_super
*super
= st
->sb
;
2658 struct imsm_super
*mpb
= super
->anchor
;
2659 struct mdinfo
*rest
= NULL
;
2662 /* do not assemble arrays that might have bad blocks */
2663 if (imsm_bbm_log_size(super
->anchor
)) {
2664 fprintf(stderr
, Name
": BBM log found in metadata. "
2665 "Cannot activate array(s).\n");
2669 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
2670 struct imsm_dev
*dev
= get_imsm_dev(super
, i
);
2671 struct imsm_map
*map
= get_imsm_map(dev
, 0);
2672 struct mdinfo
*this;
2675 this = malloc(sizeof(*this));
2676 memset(this, 0, sizeof(*this));
2679 super
->current_vol
= i
;
2680 getinfo_super_imsm_volume(st
, this);
2681 for (slot
= 0 ; slot
< map
->num_members
; slot
++) {
2682 struct mdinfo
*info_d
;
2690 idx
= get_imsm_disk_idx(dev
, slot
);
2691 ord
= get_imsm_ord_tbl_ent(dev
, slot
);
2692 for (d
= super
->disks
; d
; d
= d
->next
)
2693 if (d
->index
== idx
)
2699 s
= d
? d
->disk
.status
: 0;
2700 if (s
& FAILED_DISK
)
2702 if (!(s
& USABLE_DISK
))
2704 if (ord
& IMSM_ORD_REBUILD
)
2708 * if we skip some disks the array will be assmebled degraded;
2709 * reset resync start to avoid a dirty-degraded situation
2711 * FIXME handle dirty degraded
2713 if (skip
&& !dev
->vol
.dirty
)
2714 this->resync_start
= ~0ULL;
2718 info_d
= malloc(sizeof(*info_d
));
2720 fprintf(stderr
, Name
": failed to allocate disk"
2721 " for volume %s\n", (char *) dev
->volume
);
2726 memset(info_d
, 0, sizeof(*info_d
));
2727 info_d
->next
= this->devs
;
2728 this->devs
= info_d
;
2730 info_d
->disk
.number
= d
->index
;
2731 info_d
->disk
.major
= d
->major
;
2732 info_d
->disk
.minor
= d
->minor
;
2733 info_d
->disk
.raid_disk
= slot
;
2735 this->array
.working_disks
++;
2737 info_d
->events
= __le32_to_cpu(mpb
->generation_num
);
2738 info_d
->data_offset
= __le32_to_cpu(map
->pba_of_lba0
);
2739 info_d
->component_size
= __le32_to_cpu(map
->blocks_per_member
);
2741 strcpy(info_d
->name
, d
->devname
);
2751 static int imsm_open_new(struct supertype
*c
, struct active_array
*a
,
2754 struct intel_super
*super
= c
->sb
;
2755 struct imsm_super
*mpb
= super
->anchor
;
2757 if (atoi(inst
) >= mpb
->num_raid_devs
) {
2758 fprintf(stderr
, "%s: subarry index %d, out of range\n",
2759 __func__
, atoi(inst
));
2763 dprintf("imsm: open_new %s\n", inst
);
2764 a
->info
.container_member
= atoi(inst
);
2768 static __u8
imsm_check_degraded(struct intel_super
*super
, struct imsm_dev
*dev
, int failed
)
2770 struct imsm_map
*map
= get_imsm_map(dev
, 0);
2773 return map
->map_state
== IMSM_T_STATE_UNINITIALIZED
?
2774 IMSM_T_STATE_UNINITIALIZED
: IMSM_T_STATE_NORMAL
;
2776 switch (get_imsm_raid_level(map
)) {
2778 return IMSM_T_STATE_FAILED
;
2781 if (failed
< map
->num_members
)
2782 return IMSM_T_STATE_DEGRADED
;
2784 return IMSM_T_STATE_FAILED
;
2789 * check to see if any mirrors have failed, otherwise we
2790 * are degraded. Even numbered slots are mirrored on
2794 /* gcc -Os complains that this is unused */
2795 int insync
= insync
;
2797 for (i
= 0; i
< map
->num_members
; i
++) {
2798 __u32 ord
= get_imsm_ord_tbl_ent(dev
, i
);
2799 int idx
= ord_to_idx(ord
);
2800 struct imsm_disk
*disk
;
2802 /* reset the potential in-sync count on even-numbered
2803 * slots. num_copies is always 2 for imsm raid10
2808 disk
= get_imsm_disk(super
, idx
);
2809 if (!disk
|| disk
->status
& FAILED_DISK
||
2810 ord
& IMSM_ORD_REBUILD
)
2813 /* no in-sync disks left in this mirror the
2817 return IMSM_T_STATE_FAILED
;
2820 return IMSM_T_STATE_DEGRADED
;
2824 return IMSM_T_STATE_DEGRADED
;
2826 return IMSM_T_STATE_FAILED
;
2832 return map
->map_state
;
2835 static int imsm_count_failed(struct intel_super
*super
, struct imsm_dev
*dev
)
2839 struct imsm_disk
*disk
;
2840 struct imsm_map
*map
= get_imsm_map(dev
, 0);
2842 for (i
= 0; i
< map
->num_members
; i
++) {
2843 __u32 ord
= get_imsm_ord_tbl_ent(dev
, i
);
2844 int idx
= ord_to_idx(ord
);
2846 disk
= get_imsm_disk(super
, idx
);
2847 if (!disk
|| disk
->status
& FAILED_DISK
||
2848 ord
& IMSM_ORD_REBUILD
)
2855 static int is_resyncing(struct imsm_dev
*dev
)
2857 struct imsm_map
*migr_map
;
2859 if (!dev
->vol
.migr_state
)
2862 if (dev
->vol
.migr_type
== MIGR_INIT
)
2865 migr_map
= get_imsm_map(dev
, 1);
2867 if (migr_map
->map_state
== IMSM_T_STATE_NORMAL
)
2873 static int is_rebuilding(struct imsm_dev
*dev
)
2875 struct imsm_map
*migr_map
;
2877 if (!dev
->vol
.migr_state
)
2880 if (dev
->vol
.migr_type
!= MIGR_REBUILD
)
2883 migr_map
= get_imsm_map(dev
, 1);
2885 if (migr_map
->map_state
== IMSM_T_STATE_DEGRADED
)
2891 static void mark_failure(struct imsm_disk
*disk
)
2893 if (disk
->status
& FAILED_DISK
)
2895 disk
->status
|= FAILED_DISK
;
2896 disk
->scsi_id
= __cpu_to_le32(~(__u32
)0);
2897 memmove(&disk
->serial
[0], &disk
->serial
[1], MAX_RAID_SERIAL_LEN
- 1);
2900 /* Handle dirty -> clean transititions and resync. Degraded and rebuild
2901 * states are handled in imsm_set_disk() with one exception, when a
2902 * resync is stopped due to a new failure this routine will set the
2903 * 'degraded' state for the array.
2905 static int imsm_set_array_state(struct active_array
*a
, int consistent
)
2907 int inst
= a
->info
.container_member
;
2908 struct intel_super
*super
= a
->container
->sb
;
2909 struct imsm_dev
*dev
= get_imsm_dev(super
, inst
);
2910 struct imsm_map
*map
= get_imsm_map(dev
, 0);
2911 int failed
= imsm_count_failed(super
, dev
);
2912 __u8 map_state
= imsm_check_degraded(super
, dev
, failed
);
2914 /* before we activate this array handle any missing disks */
2915 if (consistent
== 2 && super
->missing
) {
2918 dprintf("imsm: mark missing\n");
2919 end_migration(dev
, map_state
);
2920 for (dl
= super
->missing
; dl
; dl
= dl
->next
)
2921 mark_failure(&dl
->disk
);
2922 super
->updates_pending
++;
2925 if (consistent
== 2 &&
2926 (!is_resync_complete(a
) ||
2927 map_state
!= IMSM_T_STATE_NORMAL
||
2928 dev
->vol
.migr_state
))
2931 if (is_resync_complete(a
)) {
2932 /* complete intialization / resync,
2933 * recovery is completed in ->set_disk
2935 if (is_resyncing(dev
)) {
2936 dprintf("imsm: mark resync done\n");
2937 end_migration(dev
, map_state
);
2938 super
->updates_pending
++;
2940 } else if (!is_resyncing(dev
) && !failed
) {
2941 /* mark the start of the init process if nothing is failed */
2942 dprintf("imsm: mark resync start (%llu)\n", a
->resync_start
);
2943 if (map
->map_state
== IMSM_T_STATE_NORMAL
)
2944 migrate(dev
, IMSM_T_STATE_NORMAL
, MIGR_REBUILD
);
2946 migrate(dev
, IMSM_T_STATE_NORMAL
, MIGR_INIT
);
2947 super
->updates_pending
++;
2950 /* check if we can update the migration checkpoint */
2951 if (dev
->vol
.migr_state
&&
2952 __le32_to_cpu(dev
->vol
.curr_migr_unit
) != a
->resync_start
) {
2953 dprintf("imsm: checkpoint migration (%llu)\n", a
->resync_start
);
2954 dev
->vol
.curr_migr_unit
= __cpu_to_le32(a
->resync_start
);
2955 super
->updates_pending
++;
2958 /* mark dirty / clean */
2959 if (dev
->vol
.dirty
!= !consistent
) {
2960 dprintf("imsm: mark '%s' (%llu)\n",
2961 consistent
? "clean" : "dirty", a
->resync_start
);
2966 super
->updates_pending
++;
2971 static void imsm_set_disk(struct active_array
*a
, int n
, int state
)
2973 int inst
= a
->info
.container_member
;
2974 struct intel_super
*super
= a
->container
->sb
;
2975 struct imsm_dev
*dev
= get_imsm_dev(super
, inst
);
2976 struct imsm_map
*map
= get_imsm_map(dev
, 0);
2977 struct imsm_disk
*disk
;
2982 if (n
> map
->num_members
)
2983 fprintf(stderr
, "imsm: set_disk %d out of range 0..%d\n",
2984 n
, map
->num_members
- 1);
2989 dprintf("imsm: set_disk %d:%x\n", n
, state
);
2991 ord
= get_imsm_ord_tbl_ent(dev
, n
);
2992 disk
= get_imsm_disk(super
, ord_to_idx(ord
));
2994 /* check for new failures */
2995 if ((state
& DS_FAULTY
) && !(disk
->status
& FAILED_DISK
)) {
2997 super
->updates_pending
++;
3000 /* check if in_sync */
3001 if (state
& DS_INSYNC
&& ord
& IMSM_ORD_REBUILD
) {
3002 struct imsm_map
*migr_map
= get_imsm_map(dev
, 1);
3004 set_imsm_ord_tbl_ent(migr_map
, n
, ord_to_idx(ord
));
3005 super
->updates_pending
++;
3008 failed
= imsm_count_failed(super
, dev
);
3009 map_state
= imsm_check_degraded(super
, dev
, failed
);
3011 /* check if recovery complete, newly degraded, or failed */
3012 if (map_state
== IMSM_T_STATE_NORMAL
&& is_rebuilding(dev
)) {
3013 end_migration(dev
, map_state
);
3014 super
->updates_pending
++;
3015 } else if (map_state
== IMSM_T_STATE_DEGRADED
&&
3016 map
->map_state
!= map_state
&&
3017 !dev
->vol
.migr_state
) {
3018 dprintf("imsm: mark degraded\n");
3019 map
->map_state
= map_state
;
3020 super
->updates_pending
++;
3021 } else if (map_state
== IMSM_T_STATE_FAILED
&&
3022 map
->map_state
!= map_state
) {
3023 dprintf("imsm: mark failed\n");
3024 end_migration(dev
, map_state
);
3025 super
->updates_pending
++;
3029 static int store_imsm_mpb(int fd
, struct intel_super
*super
)
3031 struct imsm_super
*mpb
= super
->anchor
;
3032 __u32 mpb_size
= __le32_to_cpu(mpb
->mpb_size
);
3033 unsigned long long dsize
;
3034 unsigned long long sectors
;
3036 get_dev_size(fd
, NULL
, &dsize
);
3038 if (mpb_size
> 512) {
3039 /* -1 to account for anchor */
3040 sectors
= mpb_sectors(mpb
) - 1;
3042 /* write the extended mpb to the sectors preceeding the anchor */
3043 if (lseek64(fd
, dsize
- (512 * (2 + sectors
)), SEEK_SET
) < 0)
3046 if (write(fd
, super
->buf
+ 512, 512 * sectors
) != 512 * sectors
)
3050 /* first block is stored on second to last sector of the disk */
3051 if (lseek64(fd
, dsize
- (512 * 2), SEEK_SET
) < 0)
3054 if (write(fd
, super
->buf
, 512) != 512)
3060 static void imsm_sync_metadata(struct supertype
*container
)
3062 struct intel_super
*super
= container
->sb
;
3064 if (!super
->updates_pending
)
3067 write_super_imsm(super
, 0);
3069 super
->updates_pending
= 0;
3072 static struct dl
*imsm_readd(struct intel_super
*super
, int idx
, struct active_array
*a
)
3074 struct imsm_dev
*dev
= get_imsm_dev(super
, a
->info
.container_member
);
3075 int i
= get_imsm_disk_idx(dev
, idx
);
3078 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
3082 if (dl
&& dl
->disk
.status
& FAILED_DISK
)
3086 dprintf("%s: found %x:%x\n", __func__
, dl
->major
, dl
->minor
);
3091 static struct dl
*imsm_add_spare(struct intel_super
*super
, int slot
, struct active_array
*a
)
3093 struct imsm_dev
*dev
= get_imsm_dev(super
, a
->info
.container_member
);
3094 int idx
= get_imsm_disk_idx(dev
, slot
);
3095 struct imsm_map
*map
= get_imsm_map(dev
, 0);
3096 unsigned long long esize
;
3097 unsigned long long pos
;
3105 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
3106 /* If in this array, skip */
3107 for (d
= a
->info
.devs
; d
; d
= d
->next
)
3108 if (d
->state_fd
>= 0 &&
3109 d
->disk
.major
== dl
->major
&&
3110 d
->disk
.minor
== dl
->minor
) {
3111 dprintf("%x:%x already in array\n", dl
->major
, dl
->minor
);
3117 /* skip in use or failed drives */
3118 if (dl
->disk
.status
& FAILED_DISK
|| idx
== dl
->index
) {
3119 dprintf("%x:%x status ( %s%s)\n",
3120 dl
->major
, dl
->minor
,
3121 dl
->disk
.status
& FAILED_DISK
? "failed " : "",
3122 idx
== dl
->index
? "in use " : "");
3126 /* Does this unused device have the requisite free space?
3127 * We need a->info.component_size sectors
3129 ex
= get_extents(super
, dl
);
3131 dprintf("cannot get extents\n");
3137 array_start
= __le32_to_cpu(map
->pba_of_lba0
);
3140 /* check that we can start at pba_of_lba0 with
3141 * a->info.component_size of space
3143 esize
= ex
[j
].start
- pos
;
3144 if (array_start
>= pos
&&
3145 array_start
+ a
->info
.component_size
< ex
[j
].start
) {
3149 pos
= ex
[j
].start
+ ex
[j
].size
;
3152 } while (ex
[j
-1].size
);
3156 dprintf("%x:%x does not have %llu at %d\n",
3157 dl
->major
, dl
->minor
,
3158 a
->info
.component_size
,
3159 __le32_to_cpu(map
->pba_of_lba0
));
3169 static struct mdinfo
*imsm_activate_spare(struct active_array
*a
,
3170 struct metadata_update
**updates
)
3173 * Find a device with unused free space and use it to replace a
3174 * failed/vacant region in an array. We replace failed regions one a
3175 * array at a time. The result is that a new spare disk will be added
3176 * to the first failed array and after the monitor has finished
3177 * propagating failures the remainder will be consumed.
3179 * FIXME add a capability for mdmon to request spares from another
3183 struct intel_super
*super
= a
->container
->sb
;
3184 int inst
= a
->info
.container_member
;
3185 struct imsm_dev
*dev
= get_imsm_dev(super
, inst
);
3186 struct imsm_map
*map
= get_imsm_map(dev
, 0);
3187 int failed
= a
->info
.array
.raid_disks
;
3188 struct mdinfo
*rv
= NULL
;
3191 struct metadata_update
*mu
;
3193 struct imsm_update_activate_spare
*u
;
3197 for (d
= a
->info
.devs
; d
; d
= d
->next
) {
3198 if ((d
->curr_state
& DS_FAULTY
) &&
3200 /* wait for Removal to happen */
3202 if (d
->state_fd
>= 0)
3206 dprintf("imsm: activate spare: inst=%d failed=%d (%d) level=%d\n",
3207 inst
, failed
, a
->info
.array
.raid_disks
, a
->info
.array
.level
);
3208 if (imsm_check_degraded(super
, dev
, failed
) != IMSM_T_STATE_DEGRADED
)
3211 /* For each slot, if it is not working, find a spare */
3212 for (i
= 0; i
< a
->info
.array
.raid_disks
; i
++) {
3213 for (d
= a
->info
.devs
; d
; d
= d
->next
)
3214 if (d
->disk
.raid_disk
== i
)
3216 dprintf("found %d: %p %x\n", i
, d
, d
?d
->curr_state
:0);
3217 if (d
&& (d
->state_fd
>= 0))
3221 * OK, this device needs recovery. Try to re-add the previous
3222 * occupant of this slot, if this fails add a new spare
3224 dl
= imsm_readd(super
, i
, a
);
3226 dl
= imsm_add_spare(super
, i
, a
);
3230 /* found a usable disk with enough space */
3231 di
= malloc(sizeof(*di
));
3234 memset(di
, 0, sizeof(*di
));
3236 /* dl->index will be -1 in the case we are activating a
3237 * pristine spare. imsm_process_update() will create a
3238 * new index in this case. Once a disk is found to be
3239 * failed in all member arrays it is kicked from the
3242 di
->disk
.number
= dl
->index
;
3244 /* (ab)use di->devs to store a pointer to the device
3247 di
->devs
= (struct mdinfo
*) dl
;
3249 di
->disk
.raid_disk
= i
;
3250 di
->disk
.major
= dl
->major
;
3251 di
->disk
.minor
= dl
->minor
;
3253 di
->data_offset
= __le32_to_cpu(map
->pba_of_lba0
);
3254 di
->component_size
= a
->info
.component_size
;
3255 di
->container_member
= inst
;
3259 dprintf("%x:%x to be %d at %llu\n", dl
->major
, dl
->minor
,
3260 i
, di
->data_offset
);
3266 /* No spares found */
3268 /* Now 'rv' has a list of devices to return.
3269 * Create a metadata_update record to update the
3270 * disk_ord_tbl for the array
3272 mu
= malloc(sizeof(*mu
));
3274 mu
->buf
= malloc(sizeof(struct imsm_update_activate_spare
) * num_spares
);
3275 if (mu
->buf
== NULL
) {
3282 struct mdinfo
*n
= rv
->next
;
3291 mu
->len
= sizeof(struct imsm_update_activate_spare
) * num_spares
;
3292 mu
->next
= *updates
;
3293 u
= (struct imsm_update_activate_spare
*) mu
->buf
;
3295 for (di
= rv
; di
; di
= di
->next
) {
3296 u
->type
= update_activate_spare
;
3297 u
->dl
= (struct dl
*) di
->devs
;
3299 u
->slot
= di
->disk
.raid_disk
;
3310 static int disks_overlap(struct intel_super
*super
, int idx
, struct imsm_update_create_array
*u
)
3312 struct imsm_dev
*dev
= get_imsm_dev(super
, idx
);
3313 struct imsm_map
*map
= get_imsm_map(dev
, 0);
3314 struct imsm_map
*new_map
= get_imsm_map(&u
->dev
, 0);
3315 struct disk_info
*inf
= get_disk_info(u
);
3316 struct imsm_disk
*disk
;
3320 for (i
= 0; i
< map
->num_members
; i
++) {
3321 disk
= get_imsm_disk(super
, get_imsm_disk_idx(dev
, i
));
3322 for (j
= 0; j
< new_map
->num_members
; j
++)
3323 if (serialcmp(disk
->serial
, inf
[j
].serial
) == 0)
3330 static void imsm_delete(struct intel_super
*super
, struct dl
**dlp
, int index
);
3332 static void imsm_process_update(struct supertype
*st
,
3333 struct metadata_update
*update
)
3336 * crack open the metadata_update envelope to find the update record
3337 * update can be one of:
3338 * update_activate_spare - a spare device has replaced a failed
3339 * device in an array, update the disk_ord_tbl. If this disk is
3340 * present in all member arrays then also clear the SPARE_DISK
3343 struct intel_super
*super
= st
->sb
;
3344 struct imsm_super
*mpb
;
3345 enum imsm_update_type type
= *(enum imsm_update_type
*) update
->buf
;
3347 /* update requires a larger buf but the allocation failed */
3348 if (super
->next_len
&& !super
->next_buf
) {
3349 super
->next_len
= 0;
3353 if (super
->next_buf
) {
3354 memcpy(super
->next_buf
, super
->buf
, super
->len
);
3356 super
->len
= super
->next_len
;
3357 super
->buf
= super
->next_buf
;
3359 super
->next_len
= 0;
3360 super
->next_buf
= NULL
;
3363 mpb
= super
->anchor
;
3366 case update_activate_spare
: {
3367 struct imsm_update_activate_spare
*u
= (void *) update
->buf
;
3368 struct imsm_dev
*dev
= get_imsm_dev(super
, u
->array
);
3369 struct imsm_map
*map
= get_imsm_map(dev
, 0);
3370 struct imsm_map
*migr_map
;
3371 struct active_array
*a
;
3372 struct imsm_disk
*disk
;
3377 int victim
= get_imsm_disk_idx(dev
, u
->slot
);
3380 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
3385 fprintf(stderr
, "error: imsm_activate_spare passed "
3386 "an unknown disk (index: %d)\n",
3391 super
->updates_pending
++;
3393 /* count failures (excluding rebuilds and the victim)
3394 * to determine map[0] state
3397 for (i
= 0; i
< map
->num_members
; i
++) {
3400 disk
= get_imsm_disk(super
, get_imsm_disk_idx(dev
, i
));
3401 if (!disk
|| disk
->status
& FAILED_DISK
)
3405 /* adding a pristine spare, assign a new index */
3406 if (dl
->index
< 0) {
3407 dl
->index
= super
->anchor
->num_disks
;
3408 super
->anchor
->num_disks
++;
3411 disk
->status
|= CONFIGURED_DISK
;
3412 disk
->status
&= ~SPARE_DISK
;
3415 to_state
= imsm_check_degraded(super
, dev
, failed
);
3416 map
->map_state
= IMSM_T_STATE_DEGRADED
;
3417 migrate(dev
, to_state
, MIGR_REBUILD
);
3418 migr_map
= get_imsm_map(dev
, 1);
3419 set_imsm_ord_tbl_ent(map
, u
->slot
, dl
->index
);
3420 set_imsm_ord_tbl_ent(migr_map
, u
->slot
, dl
->index
| IMSM_ORD_REBUILD
);
3422 /* count arrays using the victim in the metadata */
3424 for (a
= st
->arrays
; a
; a
= a
->next
) {
3425 dev
= get_imsm_dev(super
, a
->info
.container_member
);
3426 for (i
= 0; i
< map
->num_members
; i
++)
3427 if (victim
== get_imsm_disk_idx(dev
, i
))
3431 /* delete the victim if it is no longer being
3437 /* We know that 'manager' isn't touching anything,
3438 * so it is safe to delete
3440 for (dlp
= &super
->disks
; *dlp
; dlp
= &(*dlp
)->next
)
3441 if ((*dlp
)->index
== victim
)
3444 /* victim may be on the missing list */
3446 for (dlp
= &super
->missing
; *dlp
; dlp
= &(*dlp
)->next
)
3447 if ((*dlp
)->index
== victim
)
3449 imsm_delete(super
, dlp
, victim
);
3453 case update_create_array
: {
3454 /* someone wants to create a new array, we need to be aware of
3455 * a few races/collisions:
3456 * 1/ 'Create' called by two separate instances of mdadm
3457 * 2/ 'Create' versus 'activate_spare': mdadm has chosen
3458 * devices that have since been assimilated via
3460 * In the event this update can not be carried out mdadm will
3461 * (FIX ME) notice that its update did not take hold.
3463 struct imsm_update_create_array
*u
= (void *) update
->buf
;
3464 struct imsm_dev
*dev
;
3465 struct imsm_map
*map
, *new_map
;
3466 unsigned long long start
, end
;
3467 unsigned long long new_start
, new_end
;
3469 struct disk_info
*inf
;
3472 /* handle racing creates: first come first serve */
3473 if (u
->dev_idx
< mpb
->num_raid_devs
) {
3474 dprintf("%s: subarray %d already defined\n",
3475 __func__
, u
->dev_idx
);
3479 /* check update is next in sequence */
3480 if (u
->dev_idx
!= mpb
->num_raid_devs
) {
3481 dprintf("%s: can not create array %d expected index %d\n",
3482 __func__
, u
->dev_idx
, mpb
->num_raid_devs
);
3486 new_map
= get_imsm_map(&u
->dev
, 0);
3487 new_start
= __le32_to_cpu(new_map
->pba_of_lba0
);
3488 new_end
= new_start
+ __le32_to_cpu(new_map
->blocks_per_member
);
3489 inf
= get_disk_info(u
);
3491 /* handle activate_spare versus create race:
3492 * check to make sure that overlapping arrays do not include
3495 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
3496 dev
= get_imsm_dev(super
, i
);
3497 map
= get_imsm_map(dev
, 0);
3498 start
= __le32_to_cpu(map
->pba_of_lba0
);
3499 end
= start
+ __le32_to_cpu(map
->blocks_per_member
);
3500 if ((new_start
>= start
&& new_start
<= end
) ||
3501 (start
>= new_start
&& start
<= new_end
))
3506 if (disks_overlap(super
, i
, u
)) {
3507 dprintf("%s: arrays overlap\n", __func__
);
3512 /* check that prepare update was successful */
3513 if (!update
->space
) {
3514 dprintf("%s: prepare update failed\n", __func__
);
3518 /* check that all disks are still active before committing
3519 * changes. FIXME: could we instead handle this by creating a
3520 * degraded array? That's probably not what the user expects,
3521 * so better to drop this update on the floor.
3523 for (i
= 0; i
< new_map
->num_members
; i
++) {
3524 dl
= serial_to_dl(inf
[i
].serial
, super
);
3526 dprintf("%s: disk disappeared\n", __func__
);
3531 super
->updates_pending
++;
3533 /* convert spares to members and fixup ord_tbl */
3534 for (i
= 0; i
< new_map
->num_members
; i
++) {
3535 dl
= serial_to_dl(inf
[i
].serial
, super
);
3536 if (dl
->index
== -1) {
3537 dl
->index
= mpb
->num_disks
;
3539 dl
->disk
.status
|= CONFIGURED_DISK
;
3540 dl
->disk
.status
&= ~SPARE_DISK
;
3542 set_imsm_ord_tbl_ent(new_map
, i
, dl
->index
);
3545 dev
= update
->space
;
3546 update
->space
= NULL
;
3547 imsm_copy_dev(dev
, &u
->dev
);
3548 super
->dev_tbl
[u
->dev_idx
] = dev
;
3549 mpb
->num_raid_devs
++;
3551 imsm_update_version_info(super
);
3554 case update_add_disk
:
3556 /* we may be able to repair some arrays if disks are
3559 struct active_array
*a
;
3561 super
->updates_pending
++;
3562 for (a
= st
->arrays
; a
; a
= a
->next
)
3563 a
->check_degraded
= 1;
3565 /* add some spares to the metadata */
3566 while (super
->add
) {
3570 super
->add
= al
->next
;
3571 al
->next
= super
->disks
;
3573 dprintf("%s: added %x:%x\n",
3574 __func__
, al
->major
, al
->minor
);
3581 static void imsm_prepare_update(struct supertype
*st
,
3582 struct metadata_update
*update
)
3585 * Allocate space to hold new disk entries, raid-device entries or a new
3586 * mpb if necessary. The manager synchronously waits for updates to
3587 * complete in the monitor, so new mpb buffers allocated here can be
3588 * integrated by the monitor thread without worrying about live pointers
3589 * in the manager thread.
3591 enum imsm_update_type type
= *(enum imsm_update_type
*) update
->buf
;
3592 struct intel_super
*super
= st
->sb
;
3593 struct imsm_super
*mpb
= super
->anchor
;
3598 case update_create_array
: {
3599 struct imsm_update_create_array
*u
= (void *) update
->buf
;
3600 struct imsm_dev
*dev
= &u
->dev
;
3601 struct imsm_map
*map
= get_imsm_map(dev
, 0);
3603 struct disk_info
*inf
;
3607 inf
= get_disk_info(u
);
3608 len
= sizeof_imsm_dev(dev
, 1);
3609 /* allocate a new super->dev_tbl entry */
3610 update
->space
= malloc(len
);
3612 /* count how many spares will be converted to members */
3613 for (i
= 0; i
< map
->num_members
; i
++) {
3614 dl
= serial_to_dl(inf
[i
].serial
, super
);
3616 /* hmm maybe it failed?, nothing we can do about
3621 if (count_memberships(dl
, super
) == 0)
3624 len
+= activate
* sizeof(struct imsm_disk
);
3631 /* check if we need a larger metadata buffer */
3632 if (super
->next_buf
)
3633 buf_len
= super
->next_len
;
3635 buf_len
= super
->len
;
3637 if (__le32_to_cpu(mpb
->mpb_size
) + len
> buf_len
) {
3638 /* ok we need a larger buf than what is currently allocated
3639 * if this allocation fails process_update will notice that
3640 * ->next_len is set and ->next_buf is NULL
3642 buf_len
= ROUND_UP(__le32_to_cpu(mpb
->mpb_size
) + len
, 512);
3643 if (super
->next_buf
)
3644 free(super
->next_buf
);
3646 super
->next_len
= buf_len
;
3647 if (posix_memalign(&super
->next_buf
, buf_len
, 512) != 0)
3648 super
->next_buf
= NULL
;
3652 /* must be called while manager is quiesced */
3653 static void imsm_delete(struct intel_super
*super
, struct dl
**dlp
, int index
)
3655 struct imsm_super
*mpb
= super
->anchor
;
3657 struct imsm_dev
*dev
;
3658 struct imsm_map
*map
;
3659 int i
, j
, num_members
;
3662 dprintf("%s: deleting device[%d] from imsm_super\n",
3665 /* shift all indexes down one */
3666 for (iter
= super
->disks
; iter
; iter
= iter
->next
)
3667 if (iter
->index
> index
)
3669 for (iter
= super
->missing
; iter
; iter
= iter
->next
)
3670 if (iter
->index
> index
)
3673 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
3674 dev
= get_imsm_dev(super
, i
);
3675 map
= get_imsm_map(dev
, 0);
3676 num_members
= map
->num_members
;
3677 for (j
= 0; j
< num_members
; j
++) {
3678 /* update ord entries being careful not to propagate
3679 * ord-flags to the first map
3681 ord
= get_imsm_ord_tbl_ent(dev
, j
);
3683 if (ord_to_idx(ord
) <= index
)
3686 map
= get_imsm_map(dev
, 0);
3687 set_imsm_ord_tbl_ent(map
, j
, ord_to_idx(ord
- 1));
3688 map
= get_imsm_map(dev
, 1);
3690 set_imsm_ord_tbl_ent(map
, j
, ord
- 1);
3695 super
->updates_pending
++;
3697 struct dl
*dl
= *dlp
;
3699 *dlp
= (*dlp
)->next
;
3700 __free_imsm_disk(dl
);
3703 #endif /* MDASSEMBLE */
3705 struct superswitch super_imsm
= {
3707 .examine_super
= examine_super_imsm
,
3708 .brief_examine_super
= brief_examine_super_imsm
,
3709 .detail_super
= detail_super_imsm
,
3710 .brief_detail_super
= brief_detail_super_imsm
,
3711 .write_init_super
= write_init_super_imsm
,
3712 .validate_geometry
= validate_geometry_imsm
,
3713 .add_to_super
= add_to_super_imsm
,
3715 .match_home
= match_home_imsm
,
3716 .uuid_from_super
= uuid_from_super_imsm
,
3717 .getinfo_super
= getinfo_super_imsm
,
3718 .update_super
= update_super_imsm
,
3720 .avail_size
= avail_size_imsm
,
3722 .compare_super
= compare_super_imsm
,
3724 .load_super
= load_super_imsm
,
3725 .init_super
= init_super_imsm
,
3726 .store_super
= store_zero_imsm
,
3727 .free_super
= free_super_imsm
,
3728 .match_metadata_desc
= match_metadata_desc_imsm
,
3729 .container_content
= container_content_imsm
,
3735 .open_new
= imsm_open_new
,
3736 .load_super
= load_super_imsm
,
3737 .set_array_state
= imsm_set_array_state
,
3738 .set_disk
= imsm_set_disk
,
3739 .sync_metadata
= imsm_sync_metadata
,
3740 .activate_spare
= imsm_activate_spare
,
3741 .process_update
= imsm_process_update
,
3742 .prepare_update
= imsm_prepare_update
,
3743 #endif /* MDASSEMBLE */