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"
29 /* MPB == Metadata Parameter Block */
30 #define MPB_SIGNATURE "Intel Raid ISM Cfg Sig. "
31 #define MPB_SIG_LEN (strlen(MPB_SIGNATURE))
32 #define MPB_VERSION_RAID0 "1.0.00"
33 #define MPB_VERSION_RAID1 "1.1.00"
34 #define MPB_VERSION_MANY_VOLUMES_PER_ARRAY "1.2.00"
35 #define MPB_VERSION_3OR4_DISK_ARRAY "1.2.01"
36 #define MPB_VERSION_RAID5 "1.2.02"
37 #define MPB_VERSION_5OR6_DISK_ARRAY "1.2.04"
38 #define MPB_VERSION_CNG "1.2.06"
39 #define MPB_VERSION_ATTRIBS "1.3.00"
40 #define MAX_SIGNATURE_LENGTH 32
41 #define MAX_RAID_SERIAL_LEN 16
43 #define MPB_ATTRIB_CHECKSUM_VERIFY __cpu_to_le32(0x80000000)
44 #define MPB_ATTRIB_PM __cpu_to_le32(0x40000000)
45 #define MPB_ATTRIB_2TB __cpu_to_le32(0x20000000)
46 #define MPB_ATTRIB_RAID0 __cpu_to_le32(0x00000001)
47 #define MPB_ATTRIB_RAID1 __cpu_to_le32(0x00000002)
48 #define MPB_ATTRIB_RAID10 __cpu_to_le32(0x00000004)
49 #define MPB_ATTRIB_RAID1E __cpu_to_le32(0x00000008)
50 #define MPB_ATTRIB_RAID5 __cpu_to_le32(0x00000010)
51 #define MPB_ATTRIB_RAIDCNG __cpu_to_le32(0x00000020)
53 #define MPB_SECTOR_CNT 418
54 #define IMSM_RESERVED_SECTORS 4096
56 /* Disk configuration info. */
57 #define IMSM_MAX_DEVICES 255
59 __u8 serial
[MAX_RAID_SERIAL_LEN
];/* 0xD8 - 0xE7 ascii serial number */
60 __u32 total_blocks
; /* 0xE8 - 0xEB total blocks */
61 __u32 scsi_id
; /* 0xEC - 0xEF scsi ID */
62 #define SPARE_DISK __cpu_to_le32(0x01) /* Spare */
63 #define CONFIGURED_DISK __cpu_to_le32(0x02) /* Member of some RaidDev */
64 #define FAILED_DISK __cpu_to_le32(0x04) /* Permanent failure */
65 #define USABLE_DISK __cpu_to_le32(0x08) /* Fully usable unless FAILED_DISK is set */
66 __u32 status
; /* 0xF0 - 0xF3 */
67 __u32 owner_cfg_num
; /* which config 0,1,2... owns this disk */
68 #define IMSM_DISK_FILLERS 4
69 __u32 filler
[IMSM_DISK_FILLERS
]; /* 0xF4 - 0x107 MPB_DISK_FILLERS for future expansion */
72 /* RAID map configuration infos. */
74 __u32 pba_of_lba0
; /* start address of partition */
75 __u32 blocks_per_member
;/* blocks per member */
76 __u32 num_data_stripes
; /* number of data stripes */
77 __u16 blocks_per_strip
;
78 __u8 map_state
; /* Normal, Uninitialized, Degraded, Failed */
79 #define IMSM_T_STATE_NORMAL 0
80 #define IMSM_T_STATE_UNINITIALIZED 1
81 #define IMSM_T_STATE_DEGRADED 2
82 #define IMSM_T_STATE_FAILED 3
84 #define IMSM_T_RAID0 0
85 #define IMSM_T_RAID1 1
86 #define IMSM_T_RAID5 5 /* since metadata version 1.2.02 ? */
87 __u8 num_members
; /* number of member disks */
88 __u8 num_domains
; /* number of parity domains */
89 __u8 failed_disk_num
; /* valid only when state is degraded */
91 __u32 filler
[7]; /* expansion area */
92 #define IMSM_ORD_REBUILD (1 << 24)
93 __u32 disk_ord_tbl
[1]; /* disk_ord_tbl[num_members],
94 * top byte contains some flags
96 } __attribute__ ((packed
));
100 __u32 checkpoint_id
; /* id to access curr_migr_unit */
101 __u8 migr_state
; /* Normal or Migrating */
103 #define MIGR_REBUILD 1
104 #define MIGR_VERIFY 2 /* analagous to echo check > sync_action */
105 #define MIGR_GEN_MIGR 3
106 #define MIGR_STATE_CHANGE 4
107 __u8 migr_type
; /* Initializing, Rebuilding, ... */
109 __u8 fs_state
; /* fast-sync state for CnG (0xff == disabled) */
110 __u16 verify_errors
; /* number of mismatches */
111 __u16 bad_blocks
; /* number of bad blocks during verify */
113 struct imsm_map map
[1];
114 /* here comes another one if migr_state */
115 } __attribute__ ((packed
));
118 __u8 volume
[MAX_RAID_SERIAL_LEN
];
121 #define DEV_BOOTABLE __cpu_to_le32(0x01)
122 #define DEV_BOOT_DEVICE __cpu_to_le32(0x02)
123 #define DEV_READ_COALESCING __cpu_to_le32(0x04)
124 #define DEV_WRITE_COALESCING __cpu_to_le32(0x08)
125 #define DEV_LAST_SHUTDOWN_DIRTY __cpu_to_le32(0x10)
126 #define DEV_HIDDEN_AT_BOOT __cpu_to_le32(0x20)
127 #define DEV_CURRENTLY_HIDDEN __cpu_to_le32(0x40)
128 #define DEV_VERIFY_AND_FIX __cpu_to_le32(0x80)
129 #define DEV_MAP_STATE_UNINIT __cpu_to_le32(0x100)
130 #define DEV_NO_AUTO_RECOVERY __cpu_to_le32(0x200)
131 #define DEV_CLONE_N_GO __cpu_to_le32(0x400)
132 #define DEV_CLONE_MAN_SYNC __cpu_to_le32(0x800)
133 #define DEV_CNG_MASTER_DISK_NUM __cpu_to_le32(0x1000)
134 __u32 status
; /* Persistent RaidDev status */
135 __u32 reserved_blocks
; /* Reserved blocks at beginning of volume */
139 __u8 cng_master_disk
;
143 #define IMSM_DEV_FILLERS 10
144 __u32 filler
[IMSM_DEV_FILLERS
];
146 } __attribute__ ((packed
));
149 __u8 sig
[MAX_SIGNATURE_LENGTH
]; /* 0x00 - 0x1F */
150 __u32 check_sum
; /* 0x20 - 0x23 MPB Checksum */
151 __u32 mpb_size
; /* 0x24 - 0x27 Size of MPB */
152 __u32 family_num
; /* 0x28 - 0x2B Checksum from first time this config was written */
153 __u32 generation_num
; /* 0x2C - 0x2F Incremented each time this array's MPB is written */
154 __u32 error_log_size
; /* 0x30 - 0x33 in bytes */
155 __u32 attributes
; /* 0x34 - 0x37 */
156 __u8 num_disks
; /* 0x38 Number of configured disks */
157 __u8 num_raid_devs
; /* 0x39 Number of configured volumes */
158 __u8 error_log_pos
; /* 0x3A */
159 __u8 fill
[1]; /* 0x3B */
160 __u32 cache_size
; /* 0x3c - 0x40 in mb */
161 __u32 orig_family_num
; /* 0x40 - 0x43 original family num */
162 __u32 pwr_cycle_count
; /* 0x44 - 0x47 simulated power cycle count for array */
163 __u32 bbm_log_size
; /* 0x48 - 0x4B - size of bad Block Mgmt Log in bytes */
164 #define IMSM_FILLERS 35
165 __u32 filler
[IMSM_FILLERS
]; /* 0x4C - 0xD7 RAID_MPB_FILLERS */
166 struct imsm_disk disk
[1]; /* 0xD8 diskTbl[numDisks] */
167 /* here comes imsm_dev[num_raid_devs] */
168 /* here comes BBM logs */
169 } __attribute__ ((packed
));
171 #define BBM_LOG_MAX_ENTRIES 254
173 struct bbm_log_entry
{
174 __u64 defective_block_start
;
175 #define UNREADABLE 0xFFFFFFFF
176 __u32 spare_block_offset
;
177 __u16 remapped_marked_count
;
179 } __attribute__ ((__packed__
));
182 __u32 signature
; /* 0xABADB10C */
184 __u32 reserved_spare_block_count
; /* 0 */
185 __u32 reserved
; /* 0xFFFF */
186 __u64 first_spare_lba
;
187 struct bbm_log_entry mapped_block_entries
[BBM_LOG_MAX_ENTRIES
];
188 } __attribute__ ((__packed__
));
192 static char *map_state_str
[] = { "normal", "uninitialized", "degraded", "failed" };
195 static unsigned int sector_count(__u32 bytes
)
197 return ((bytes
+ (512-1)) & (~(512-1))) / 512;
200 static unsigned int mpb_sectors(struct imsm_super
*mpb
)
202 return sector_count(__le32_to_cpu(mpb
->mpb_size
));
205 /* internal representation of IMSM metadata */
208 void *buf
; /* O_DIRECT buffer for reading/writing metadata */
209 struct imsm_super
*anchor
; /* immovable parameters */
211 size_t len
; /* size of the 'buf' allocation */
212 void *next_buf
; /* for realloc'ing buf from the manager */
214 int updates_pending
; /* count of pending updates for mdmon */
215 int creating_imsm
; /* flag to indicate container creation */
216 int current_vol
; /* index of raid device undergoing creation */
217 __u32 create_offset
; /* common start for 'current_vol' */
218 #define IMSM_MAX_RAID_DEVS 2
219 struct imsm_dev
*dev_tbl
[IMSM_MAX_RAID_DEVS
];
223 __u8 serial
[MAX_RAID_SERIAL_LEN
];
226 struct imsm_disk disk
;
229 struct extent
*e
; /* for determining freespace @ create */
231 struct dl
*add
; /* list of disks to add while mdmon active */
232 struct dl
*missing
; /* disks removed while we weren't looking */
233 struct bbm_log
*bbm_log
;
234 const char *hba
; /* device path of the raid controller for this metadata */
235 const struct imsm_orom
*orom
; /* platform firmware support */
239 unsigned long long start
, size
;
242 /* definition of messages passed to imsm_process_update */
243 enum imsm_update_type
{
244 update_activate_spare
,
249 struct imsm_update_activate_spare
{
250 enum imsm_update_type type
;
254 struct imsm_update_activate_spare
*next
;
258 __u8 serial
[MAX_RAID_SERIAL_LEN
];
261 struct imsm_update_create_array
{
262 enum imsm_update_type type
;
267 struct imsm_update_add_disk
{
268 enum imsm_update_type type
;
271 static struct supertype
*match_metadata_desc_imsm(char *arg
)
273 struct supertype
*st
;
275 if (strcmp(arg
, "imsm") != 0 &&
276 strcmp(arg
, "default") != 0
280 st
= malloc(sizeof(*st
));
281 memset(st
, 0, sizeof(*st
));
282 st
->ss
= &super_imsm
;
283 st
->max_devs
= IMSM_MAX_DEVICES
;
284 st
->minor_version
= 0;
290 static __u8
*get_imsm_version(struct imsm_super
*mpb
)
292 return &mpb
->sig
[MPB_SIG_LEN
];
296 /* retrieve a disk directly from the anchor when the anchor is known to be
297 * up-to-date, currently only at load time
299 static struct imsm_disk
*__get_imsm_disk(struct imsm_super
*mpb
, __u8 index
)
301 if (index
>= mpb
->num_disks
)
303 return &mpb
->disk
[index
];
307 /* retrieve a disk from the parsed metadata */
308 static struct imsm_disk
*get_imsm_disk(struct intel_super
*super
, __u8 index
)
312 for (d
= super
->disks
; d
; d
= d
->next
)
313 if (d
->index
== index
)
320 /* generate a checksum directly from the anchor when the anchor is known to be
321 * up-to-date, currently only at load or write_super after coalescing
323 static __u32
__gen_imsm_checksum(struct imsm_super
*mpb
)
325 __u32 end
= mpb
->mpb_size
/ sizeof(end
);
326 __u32
*p
= (__u32
*) mpb
;
330 sum
+= __le32_to_cpu(*p
);
334 return sum
- __le32_to_cpu(mpb
->check_sum
);
337 static size_t sizeof_imsm_map(struct imsm_map
*map
)
339 return sizeof(struct imsm_map
) + sizeof(__u32
) * (map
->num_members
- 1);
342 struct imsm_map
*get_imsm_map(struct imsm_dev
*dev
, int second_map
)
344 struct imsm_map
*map
= &dev
->vol
.map
[0];
346 if (second_map
&& !dev
->vol
.migr_state
)
348 else if (second_map
) {
351 return ptr
+ sizeof_imsm_map(map
);
357 /* return the size of the device.
358 * migr_state increases the returned size if map[0] were to be duplicated
360 static size_t sizeof_imsm_dev(struct imsm_dev
*dev
, int migr_state
)
362 size_t size
= sizeof(*dev
) - sizeof(struct imsm_map
) +
363 sizeof_imsm_map(get_imsm_map(dev
, 0));
365 /* migrating means an additional map */
366 if (dev
->vol
.migr_state
)
367 size
+= sizeof_imsm_map(get_imsm_map(dev
, 1));
369 size
+= sizeof_imsm_map(get_imsm_map(dev
, 0));
375 /* retrieve disk serial number list from a metadata update */
376 static struct disk_info
*get_disk_info(struct imsm_update_create_array
*update
)
379 struct disk_info
*inf
;
381 inf
= u
+ sizeof(*update
) - sizeof(struct imsm_dev
) +
382 sizeof_imsm_dev(&update
->dev
, 0);
388 static struct imsm_dev
*__get_imsm_dev(struct imsm_super
*mpb
, __u8 index
)
394 if (index
>= mpb
->num_raid_devs
)
397 /* devices start after all disks */
398 offset
= ((void *) &mpb
->disk
[mpb
->num_disks
]) - _mpb
;
400 for (i
= 0; i
<= index
; i
++)
402 return _mpb
+ offset
;
404 offset
+= sizeof_imsm_dev(_mpb
+ offset
, 0);
409 static struct imsm_dev
*get_imsm_dev(struct intel_super
*super
, __u8 index
)
411 if (index
>= super
->anchor
->num_raid_devs
)
413 return super
->dev_tbl
[index
];
416 static __u32
get_imsm_ord_tbl_ent(struct imsm_dev
*dev
, int slot
)
418 struct imsm_map
*map
;
420 if (dev
->vol
.migr_state
)
421 map
= get_imsm_map(dev
, 1);
423 map
= get_imsm_map(dev
, 0);
425 /* top byte identifies disk under rebuild */
426 return __le32_to_cpu(map
->disk_ord_tbl
[slot
]);
429 #define ord_to_idx(ord) (((ord) << 8) >> 8)
430 static __u32
get_imsm_disk_idx(struct imsm_dev
*dev
, int slot
)
432 __u32 ord
= get_imsm_ord_tbl_ent(dev
, slot
);
434 return ord_to_idx(ord
);
437 static void set_imsm_ord_tbl_ent(struct imsm_map
*map
, int slot
, __u32 ord
)
439 map
->disk_ord_tbl
[slot
] = __cpu_to_le32(ord
);
442 static int get_imsm_raid_level(struct imsm_map
*map
)
444 if (map
->raid_level
== 1) {
445 if (map
->num_members
== 2)
451 return map
->raid_level
;
454 static int cmp_extent(const void *av
, const void *bv
)
456 const struct extent
*a
= av
;
457 const struct extent
*b
= bv
;
458 if (a
->start
< b
->start
)
460 if (a
->start
> b
->start
)
465 static int count_memberships(struct dl
*dl
, struct intel_super
*super
)
470 for (i
= 0; i
< super
->anchor
->num_raid_devs
; i
++) {
471 struct imsm_dev
*dev
= get_imsm_dev(super
, i
);
472 struct imsm_map
*map
= get_imsm_map(dev
, 0);
474 for (j
= 0; j
< map
->num_members
; j
++) {
475 __u32 index
= get_imsm_disk_idx(dev
, j
);
477 if (index
== dl
->index
)
485 static struct extent
*get_extents(struct intel_super
*super
, struct dl
*dl
)
487 /* find a list of used extents on the given physical device */
488 struct extent
*rv
, *e
;
490 int memberships
= count_memberships(dl
, super
);
491 __u32 reservation
= MPB_SECTOR_CNT
+ IMSM_RESERVED_SECTORS
;
493 rv
= malloc(sizeof(struct extent
) * (memberships
+ 1));
498 for (i
= 0; i
< super
->anchor
->num_raid_devs
; i
++) {
499 struct imsm_dev
*dev
= get_imsm_dev(super
, i
);
500 struct imsm_map
*map
= get_imsm_map(dev
, 0);
502 for (j
= 0; j
< map
->num_members
; j
++) {
503 __u32 index
= get_imsm_disk_idx(dev
, j
);
505 if (index
== dl
->index
) {
506 e
->start
= __le32_to_cpu(map
->pba_of_lba0
);
507 e
->size
= __le32_to_cpu(map
->blocks_per_member
);
512 qsort(rv
, memberships
, sizeof(*rv
), cmp_extent
);
514 /* determine the start of the metadata
515 * when no raid devices are defined use the default
516 * ...otherwise allow the metadata to truncate the value
517 * as is the case with older versions of imsm
520 struct extent
*last
= &rv
[memberships
- 1];
523 remainder
= __le32_to_cpu(dl
->disk
.total_blocks
) -
524 (last
->start
+ last
->size
);
525 /* round down to 1k block to satisfy precision of the kernel
529 /* make sure remainder is still sane */
530 if (remainder
< ROUND_UP(super
->len
, 512) >> 9)
531 remainder
= ROUND_UP(super
->len
, 512) >> 9;
532 if (reservation
> remainder
)
533 reservation
= remainder
;
535 e
->start
= __le32_to_cpu(dl
->disk
.total_blocks
) - reservation
;
540 /* try to determine how much space is reserved for metadata from
541 * the last get_extents() entry, otherwise fallback to the
544 static __u32
imsm_reserved_sectors(struct intel_super
*super
, struct dl
*dl
)
550 /* for spares just return a minimal reservation which will grow
551 * once the spare is picked up by an array
554 return MPB_SECTOR_CNT
;
556 e
= get_extents(super
, dl
);
558 return MPB_SECTOR_CNT
+ IMSM_RESERVED_SECTORS
;
560 /* scroll to last entry */
561 for (i
= 0; e
[i
].size
; i
++)
564 rv
= __le32_to_cpu(dl
->disk
.total_blocks
) - e
[i
].start
;
572 static void print_imsm_dev(struct imsm_dev
*dev
, char *uuid
, int disk_idx
)
576 struct imsm_map
*map
= get_imsm_map(dev
, 0);
580 printf("[%.16s]:\n", dev
->volume
);
581 printf(" UUID : %s\n", uuid
);
582 printf(" RAID Level : %d\n", get_imsm_raid_level(map
));
583 printf(" Members : %d\n", map
->num_members
);
584 for (slot
= 0; slot
< map
->num_members
; slot
++)
585 if (disk_idx
== get_imsm_disk_idx(dev
, slot
))
587 if (slot
< map
->num_members
) {
588 ord
= get_imsm_ord_tbl_ent(dev
, slot
);
589 printf(" This Slot : %d%s\n", slot
,
590 ord
& IMSM_ORD_REBUILD
? " (out-of-sync)" : "");
592 printf(" This Slot : ?\n");
593 sz
= __le32_to_cpu(dev
->size_high
);
595 sz
+= __le32_to_cpu(dev
->size_low
);
596 printf(" Array Size : %llu%s\n", (unsigned long long)sz
,
597 human_size(sz
* 512));
598 sz
= __le32_to_cpu(map
->blocks_per_member
);
599 printf(" Per Dev Size : %llu%s\n", (unsigned long long)sz
,
600 human_size(sz
* 512));
601 printf(" Sector Offset : %u\n",
602 __le32_to_cpu(map
->pba_of_lba0
));
603 printf(" Num Stripes : %u\n",
604 __le32_to_cpu(map
->num_data_stripes
));
605 printf(" Chunk Size : %u KiB\n",
606 __le16_to_cpu(map
->blocks_per_strip
) / 2);
607 printf(" Reserved : %d\n", __le32_to_cpu(dev
->reserved_blocks
));
608 printf(" Migrate State : %s", dev
->vol
.migr_state
? "migrating" : "idle");
609 if (dev
->vol
.migr_state
)
610 printf(": %s", dev
->vol
.migr_type
? "rebuilding" : "initializing");
612 printf(" Map State : %s", map_state_str
[map
->map_state
]);
613 if (dev
->vol
.migr_state
) {
614 struct imsm_map
*map
= get_imsm_map(dev
, 1);
615 printf(" <-- %s", map_state_str
[map
->map_state
]);
618 printf(" Dirty State : %s\n", dev
->vol
.dirty
? "dirty" : "clean");
621 static void print_imsm_disk(struct imsm_super
*mpb
, int index
, __u32 reserved
)
623 struct imsm_disk
*disk
= __get_imsm_disk(mpb
, index
);
624 char str
[MAX_RAID_SERIAL_LEN
+ 1];
632 snprintf(str
, MAX_RAID_SERIAL_LEN
+ 1, "%s", disk
->serial
);
633 printf(" Disk%02d Serial : %s\n", index
, str
);
635 printf(" State :%s%s%s%s\n", s
&SPARE_DISK
? " spare" : "",
636 s
&CONFIGURED_DISK
? " active" : "",
637 s
&FAILED_DISK
? " failed" : "",
638 s
&USABLE_DISK
? " usable" : "");
639 printf(" Id : %08x\n", __le32_to_cpu(disk
->scsi_id
));
640 sz
= __le32_to_cpu(disk
->total_blocks
) - reserved
;
641 printf(" Usable Size : %llu%s\n", (unsigned long long)sz
,
642 human_size(sz
* 512));
645 static void getinfo_super_imsm(struct supertype
*st
, struct mdinfo
*info
);
647 static void examine_super_imsm(struct supertype
*st
, char *homehost
)
649 struct intel_super
*super
= st
->sb
;
650 struct imsm_super
*mpb
= super
->anchor
;
651 char str
[MAX_SIGNATURE_LENGTH
];
656 __u32 reserved
= imsm_reserved_sectors(super
, super
->disks
);
659 snprintf(str
, MPB_SIG_LEN
, "%s", mpb
->sig
);
660 printf(" Magic : %s\n", str
);
661 snprintf(str
, strlen(MPB_VERSION_RAID0
), "%s", get_imsm_version(mpb
));
662 printf(" Version : %s\n", get_imsm_version(mpb
));
663 printf(" Family : %08x\n", __le32_to_cpu(mpb
->family_num
));
664 printf(" Generation : %08x\n", __le32_to_cpu(mpb
->generation_num
));
665 getinfo_super_imsm(st
, &info
);
666 fname_from_uuid(st
, &info
, nbuf
,'-');
667 printf(" UUID : %s\n", nbuf
+ 5);
668 sum
= __le32_to_cpu(mpb
->check_sum
);
669 printf(" Checksum : %08x %s\n", sum
,
670 __gen_imsm_checksum(mpb
) == sum
? "correct" : "incorrect");
671 printf(" MPB Sectors : %d\n", mpb_sectors(mpb
));
672 printf(" Disks : %d\n", mpb
->num_disks
);
673 printf(" RAID Devices : %d\n", mpb
->num_raid_devs
);
674 print_imsm_disk(mpb
, super
->disks
->index
, reserved
);
675 if (super
->bbm_log
) {
676 struct bbm_log
*log
= super
->bbm_log
;
679 printf("Bad Block Management Log:\n");
680 printf(" Log Size : %d\n", __le32_to_cpu(mpb
->bbm_log_size
));
681 printf(" Signature : %x\n", __le32_to_cpu(log
->signature
));
682 printf(" Entry Count : %d\n", __le32_to_cpu(log
->entry_count
));
683 printf(" Spare Blocks : %d\n", __le32_to_cpu(log
->reserved_spare_block_count
));
684 printf(" First Spare : %llx\n", __le64_to_cpu(log
->first_spare_lba
));
686 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
688 struct imsm_dev
*dev
= __get_imsm_dev(mpb
, i
);
690 super
->current_vol
= i
;
691 getinfo_super_imsm(st
, &info
);
692 fname_from_uuid(st
, &info
, nbuf
, '-');
693 print_imsm_dev(dev
, nbuf
+ 5, super
->disks
->index
);
695 for (i
= 0; i
< mpb
->num_disks
; i
++) {
696 if (i
== super
->disks
->index
)
698 print_imsm_disk(mpb
, i
, reserved
);
702 static void brief_examine_super_imsm(struct supertype
*st
)
704 /* We just write a generic IMSM ARRAY entry */
708 struct intel_super
*super
= st
->sb
;
711 if (!super
->anchor
->num_raid_devs
)
714 getinfo_super_imsm(st
, &info
);
715 fname_from_uuid(st
, &info
, nbuf
,'-');
716 printf("ARRAY metadata=imsm auto=md UUID=%s\n", nbuf
+ 5);
717 for (i
= 0; i
< super
->anchor
->num_raid_devs
; i
++) {
718 struct imsm_dev
*dev
= get_imsm_dev(super
, i
);
720 super
->current_vol
= i
;
721 getinfo_super_imsm(st
, &info
);
722 fname_from_uuid(st
, &info
, nbuf1
,'-');
723 printf("ARRAY /dev/md/%.16s container=%s\n"
724 " member=%d auto=mdp UUID=%s\n",
725 dev
->volume
, nbuf
+ 5, i
, nbuf1
+ 5);
729 static void detail_super_imsm(struct supertype
*st
, char *homehost
)
734 getinfo_super_imsm(st
, &info
);
735 fname_from_uuid(st
, &info
, nbuf
,'-');
736 printf("\n UUID : %s\n", nbuf
+ 5);
739 static void brief_detail_super_imsm(struct supertype
*st
)
743 getinfo_super_imsm(st
, &info
);
744 fname_from_uuid(st
, &info
, nbuf
,'-');
745 printf(" UUID=%s", nbuf
+ 5);
749 static int match_home_imsm(struct supertype
*st
, char *homehost
)
751 /* the imsm metadata format does not specify any host
752 * identification information. We return -1 since we can never
753 * confirm nor deny whether a given array is "meant" for this
754 * host. We rely on compare_super and the 'family_num' field to
755 * exclude member disks that do not belong, and we rely on
756 * mdadm.conf to specify the arrays that should be assembled.
757 * Auto-assembly may still pick up "foreign" arrays.
763 static void uuid_from_super_imsm(struct supertype
*st
, int uuid
[4])
765 /* The uuid returned here is used for:
766 * uuid to put into bitmap file (Create, Grow)
767 * uuid for backup header when saving critical section (Grow)
768 * comparing uuids when re-adding a device into an array
769 * In these cases the uuid required is that of the data-array,
770 * not the device-set.
771 * uuid to recognise same set when adding a missing device back
772 * to an array. This is a uuid for the device-set.
774 * For each of these we can make do with a truncated
775 * or hashed uuid rather than the original, as long as
777 * In each case the uuid required is that of the data-array,
778 * not the device-set.
780 /* imsm does not track uuid's so we synthesis one using sha1 on
781 * - The signature (Which is constant for all imsm array, but no matter)
782 * - the family_num of the container
783 * - the index number of the volume
784 * - the 'serial' number of the volume.
785 * Hopefully these are all constant.
787 struct intel_super
*super
= st
->sb
;
791 struct imsm_dev
*dev
= NULL
;
794 sha1_process_bytes(super
->anchor
->sig
, MPB_SIG_LEN
, &ctx
);
795 sha1_process_bytes(&super
->anchor
->family_num
, sizeof(__u32
), &ctx
);
796 if (super
->current_vol
>= 0)
797 dev
= get_imsm_dev(super
, super
->current_vol
);
799 __u32 vol
= super
->current_vol
;
800 sha1_process_bytes(&vol
, sizeof(vol
), &ctx
);
801 sha1_process_bytes(dev
->volume
, MAX_RAID_SERIAL_LEN
, &ctx
);
803 sha1_finish_ctx(&ctx
, buf
);
804 memcpy(uuid
, buf
, 4*4);
809 get_imsm_numerical_version(struct imsm_super
*mpb
, int *m
, int *p
)
811 __u8
*v
= get_imsm_version(mpb
);
812 __u8
*end
= mpb
->sig
+ MAX_SIGNATURE_LENGTH
;
813 char major
[] = { 0, 0, 0 };
814 char minor
[] = { 0 ,0, 0 };
815 char patch
[] = { 0, 0, 0 };
816 char *ver_parse
[] = { major
, minor
, patch
};
820 while (*v
!= '\0' && v
< end
) {
821 if (*v
!= '.' && j
< 2)
822 ver_parse
[i
][j
++] = *v
;
830 *m
= strtol(minor
, NULL
, 0);
831 *p
= strtol(patch
, NULL
, 0);
835 static int imsm_level_to_layout(int level
)
843 return ALGORITHM_LEFT_ASYMMETRIC
;
850 static void getinfo_super_imsm_volume(struct supertype
*st
, struct mdinfo
*info
)
852 struct intel_super
*super
= st
->sb
;
853 struct imsm_dev
*dev
= get_imsm_dev(super
, super
->current_vol
);
854 struct imsm_map
*map
= get_imsm_map(dev
, 0);
856 info
->container_member
= super
->current_vol
;
857 info
->array
.raid_disks
= map
->num_members
;
858 info
->array
.level
= get_imsm_raid_level(map
);
859 info
->array
.layout
= imsm_level_to_layout(info
->array
.level
);
860 info
->array
.md_minor
= -1;
861 info
->array
.ctime
= 0;
862 info
->array
.utime
= 0;
863 info
->array
.chunk_size
= __le16_to_cpu(map
->blocks_per_strip
) << 9;
864 info
->array
.state
= !dev
->vol
.dirty
;
866 info
->disk
.major
= 0;
867 info
->disk
.minor
= 0;
869 info
->data_offset
= __le32_to_cpu(map
->pba_of_lba0
);
870 info
->component_size
= __le32_to_cpu(map
->blocks_per_member
);
871 memset(info
->uuid
, 0, sizeof(info
->uuid
));
873 if (map
->map_state
== IMSM_T_STATE_UNINITIALIZED
|| dev
->vol
.dirty
)
874 info
->resync_start
= 0;
875 else if (dev
->vol
.migr_state
)
876 info
->resync_start
= __le32_to_cpu(dev
->vol
.curr_migr_unit
);
878 info
->resync_start
= ~0ULL;
880 strncpy(info
->name
, (char *) dev
->volume
, MAX_RAID_SERIAL_LEN
);
881 info
->name
[MAX_RAID_SERIAL_LEN
] = 0;
883 info
->array
.major_version
= -1;
884 info
->array
.minor_version
= -2;
885 sprintf(info
->text_version
, "/%s/%d",
886 devnum2devname(st
->container_dev
),
887 info
->container_member
);
888 info
->safe_mode_delay
= 4000; /* 4 secs like the Matrix driver */
889 uuid_from_super_imsm(st
, info
->uuid
);
893 static void getinfo_super_imsm(struct supertype
*st
, struct mdinfo
*info
)
895 struct intel_super
*super
= st
->sb
;
896 struct imsm_disk
*disk
;
899 if (super
->current_vol
>= 0) {
900 getinfo_super_imsm_volume(st
, info
);
904 /* Set raid_disks to zero so that Assemble will always pull in valid
907 info
->array
.raid_disks
= 0;
908 info
->array
.level
= LEVEL_CONTAINER
;
909 info
->array
.layout
= 0;
910 info
->array
.md_minor
= -1;
911 info
->array
.ctime
= 0; /* N/A for imsm */
912 info
->array
.utime
= 0;
913 info
->array
.chunk_size
= 0;
915 info
->disk
.major
= 0;
916 info
->disk
.minor
= 0;
917 info
->disk
.raid_disk
= -1;
918 info
->reshape_active
= 0;
919 info
->array
.major_version
= -1;
920 info
->array
.minor_version
= -2;
921 strcpy(info
->text_version
, "imsm");
922 info
->safe_mode_delay
= 0;
923 info
->disk
.number
= -1;
924 info
->disk
.state
= 0;
928 __u32 reserved
= imsm_reserved_sectors(super
, super
->disks
);
930 disk
= &super
->disks
->disk
;
931 info
->data_offset
= __le32_to_cpu(disk
->total_blocks
) - reserved
;
932 info
->component_size
= reserved
;
934 info
->disk
.state
= s
& CONFIGURED_DISK
? (1 << MD_DISK_ACTIVE
) : 0;
935 info
->disk
.state
|= s
& FAILED_DISK
? (1 << MD_DISK_FAULTY
) : 0;
936 info
->disk
.state
|= s
& SPARE_DISK
? 0 : (1 << MD_DISK_SYNC
);
939 /* only call uuid_from_super_imsm when this disk is part of a populated container,
940 * ->compare_super may have updated the 'num_raid_devs' field for spares
942 if (info
->disk
.state
& (1 << MD_DISK_SYNC
) || super
->anchor
->num_raid_devs
)
943 uuid_from_super_imsm(st
, info
->uuid
);
945 memcpy(info
->uuid
, uuid_match_any
, sizeof(int[4]));
948 static int update_super_imsm(struct supertype
*st
, struct mdinfo
*info
,
949 char *update
, char *devname
, int verbose
,
950 int uuid_set
, char *homehost
)
954 /* For 'assemble' and 'force' we need to return non-zero if any
955 * change was made. For others, the return value is ignored.
956 * Update options are:
957 * force-one : This device looks a bit old but needs to be included,
958 * update age info appropriately.
959 * assemble: clear any 'faulty' flag to allow this device to
961 * force-array: Array is degraded but being forced, mark it clean
962 * if that will be needed to assemble it.
964 * newdev: not used ????
965 * grow: Array has gained a new device - this is currently for
967 * resync: mark as dirty so a resync will happen.
968 * name: update the name - preserving the homehost
970 * Following are not relevant for this imsm:
971 * sparc2.2 : update from old dodgey metadata
972 * super-minor: change the preferred_minor number
973 * summaries: update redundant counters.
974 * uuid: Change the uuid of the array to match watch is given
975 * homehost: update the recorded homehost
976 * _reshape_progress: record new reshape_progress position.
979 //struct intel_super *super = st->sb;
980 //struct imsm_super *mpb = super->mpb;
982 if (strcmp(update
, "grow") == 0) {
984 if (strcmp(update
, "resync") == 0) {
985 /* dev->vol.dirty = 1; */
988 /* IMSM has no concept of UUID or homehost */
993 static size_t disks_to_mpb_size(int disks
)
997 size
= sizeof(struct imsm_super
);
998 size
+= (disks
- 1) * sizeof(struct imsm_disk
);
999 size
+= 2 * sizeof(struct imsm_dev
);
1000 /* up to 2 maps per raid device (-2 for imsm_maps in imsm_dev */
1001 size
+= (4 - 2) * sizeof(struct imsm_map
);
1002 /* 4 possible disk_ord_tbl's */
1003 size
+= 4 * (disks
- 1) * sizeof(__u32
);
1008 static __u64
avail_size_imsm(struct supertype
*st
, __u64 devsize
)
1010 if (devsize
< (MPB_SECTOR_CNT
+ IMSM_RESERVED_SECTORS
))
1013 return devsize
- (MPB_SECTOR_CNT
+ IMSM_RESERVED_SECTORS
);
1016 static int compare_super_imsm(struct supertype
*st
, struct supertype
*tst
)
1020 * 0 same, or first was empty, and second was copied
1021 * 1 second had wrong number
1023 * 3 wrong other info
1025 struct intel_super
*first
= st
->sb
;
1026 struct intel_super
*sec
= tst
->sb
;
1034 if (memcmp(first
->anchor
->sig
, sec
->anchor
->sig
, MAX_SIGNATURE_LENGTH
) != 0)
1037 /* if an anchor does not have num_raid_devs set then it is a free
1040 if (first
->anchor
->num_raid_devs
> 0 &&
1041 sec
->anchor
->num_raid_devs
> 0) {
1042 if (first
->anchor
->family_num
!= sec
->anchor
->family_num
)
1046 /* if 'first' is a spare promote it to a populated mpb with sec's
1049 if (first
->anchor
->num_raid_devs
== 0 &&
1050 sec
->anchor
->num_raid_devs
> 0) {
1053 /* we need to copy raid device info from sec if an allocation
1054 * fails here we don't associate the spare
1056 for (i
= 0; i
< sec
->anchor
->num_raid_devs
; i
++) {
1057 first
->dev_tbl
[i
] = malloc(sizeof(struct imsm_dev
));
1058 if (!first
->dev_tbl
) {
1060 free(first
->dev_tbl
[i
]);
1061 first
->dev_tbl
[i
] = NULL
;
1063 fprintf(stderr
, "imsm: failed to associate spare\n");
1066 *first
->dev_tbl
[i
] = *sec
->dev_tbl
[i
];
1069 first
->anchor
->num_raid_devs
= sec
->anchor
->num_raid_devs
;
1070 first
->anchor
->family_num
= sec
->anchor
->family_num
;
1076 static void fd2devname(int fd
, char *name
)
1085 if (fstat(fd
, &st
) != 0)
1087 sprintf(path
, "/sys/dev/block/%d:%d",
1088 major(st
.st_rdev
), minor(st
.st_rdev
));
1090 rv
= readlink(path
, dname
, sizeof(dname
));
1095 nm
= strrchr(dname
, '/');
1097 snprintf(name
, MAX_RAID_SERIAL_LEN
, "/dev/%s", nm
);
1101 extern int scsi_get_serial(int fd
, void *buf
, size_t buf_len
);
1103 static int imsm_read_serial(int fd
, char *devname
,
1104 __u8 serial
[MAX_RAID_SERIAL_LEN
])
1106 unsigned char scsi_serial
[255];
1112 memset(scsi_serial
, 0, sizeof(scsi_serial
));
1114 rv
= scsi_get_serial(fd
, scsi_serial
, sizeof(scsi_serial
));
1116 if (rv
&& check_env("IMSM_DEVNAME_AS_SERIAL")) {
1117 memset(serial
, 0, MAX_RAID_SERIAL_LEN
);
1118 fd2devname(fd
, (char *) serial
);
1125 Name
": Failed to retrieve serial for %s\n",
1130 /* trim leading whitespace */
1131 rsp_len
= scsi_serial
[3];
1132 rsp_buf
= (char *) &scsi_serial
[4];
1137 /* truncate len to the end of rsp_buf if necessary */
1138 if (c
+ MAX_RAID_SERIAL_LEN
> rsp_buf
+ rsp_len
)
1139 len
= rsp_len
- (c
- rsp_buf
);
1141 len
= MAX_RAID_SERIAL_LEN
;
1143 /* initialize the buffer and copy rsp_buf characters */
1144 memset(serial
, 0, MAX_RAID_SERIAL_LEN
);
1145 memcpy(serial
, c
, len
);
1147 /* trim trailing whitespace starting with the last character copied */
1148 c
= (char *) &serial
[len
- 1];
1149 while (isspace(*c
) || *c
== '\0')
1155 static int serialcmp(__u8
*s1
, __u8
*s2
)
1157 return strncmp((char *) s1
, (char *) s2
, MAX_RAID_SERIAL_LEN
);
1160 static void serialcpy(__u8
*dest
, __u8
*src
)
1162 strncpy((char *) dest
, (char *) src
, MAX_RAID_SERIAL_LEN
);
1165 static struct dl
*serial_to_dl(__u8
*serial
, struct intel_super
*super
)
1169 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
1170 if (serialcmp(dl
->serial
, serial
) == 0)
1177 load_imsm_disk(int fd
, struct intel_super
*super
, char *devname
, int keep_fd
)
1184 __u8 serial
[MAX_RAID_SERIAL_LEN
];
1186 rv
= imsm_read_serial(fd
, devname
, serial
);
1191 /* check if this is a disk we have seen before. it may be a spare in
1192 * super->disks while the current anchor believes it is a raid member,
1193 * check if we need to update dl->index
1195 dl
= serial_to_dl(serial
, super
);
1197 dl
= malloc(sizeof(*dl
));
1204 Name
": failed to allocate disk buffer for %s\n",
1211 dl
->major
= major(stb
.st_rdev
);
1212 dl
->minor
= minor(stb
.st_rdev
);
1213 dl
->next
= super
->disks
;
1214 dl
->fd
= keep_fd
? fd
: -1;
1215 dl
->devname
= devname
? strdup(devname
) : NULL
;
1216 serialcpy(dl
->serial
, serial
);
1219 } else if (keep_fd
) {
1224 /* look up this disk's index in the current anchor */
1225 for (i
= 0; i
< super
->anchor
->num_disks
; i
++) {
1226 struct imsm_disk
*disk_iter
;
1228 disk_iter
= __get_imsm_disk(super
->anchor
, i
);
1230 if (serialcmp(disk_iter
->serial
, dl
->serial
) == 0) {
1231 dl
->disk
= *disk_iter
;
1232 /* only set index on disks that are a member of a
1233 * populated contianer, i.e. one with raid_devs
1235 if (dl
->disk
.status
& FAILED_DISK
)
1237 else if (dl
->disk
.status
& SPARE_DISK
)
1246 /* no match, maybe a stale failed drive */
1247 if (i
== super
->anchor
->num_disks
&& dl
->index
>= 0) {
1248 dl
->disk
= *__get_imsm_disk(super
->anchor
, dl
->index
);
1249 if (dl
->disk
.status
& FAILED_DISK
)
1259 static void imsm_copy_dev(struct imsm_dev
*dest
, struct imsm_dev
*src
)
1261 memcpy(dest
, src
, sizeof_imsm_dev(src
, 0));
1265 /* When migrating map0 contains the 'destination' state while map1
1266 * contains the current state. When not migrating map0 contains the
1267 * current state. This routine assumes that map[0].map_state is set to
1268 * the current array state before being called.
1270 * Migration is indicated by one of the following states
1271 * 1/ Idle (migr_state=0 map0state=normal||unitialized||degraded||failed)
1272 * 2/ Initialize (migr_state=1 migr_type=MIGR_INIT map0state=normal
1273 * map1state=unitialized)
1274 * 3/ Verify (Resync) (migr_state=1 migr_type=MIGR_REBUILD map0state=normal
1276 * 4/ Rebuild (migr_state=1 migr_type=MIGR_REBUILD map0state=normal
1277 * map1state=degraded)
1279 static void migrate(struct imsm_dev
*dev
, __u8 to_state
, int rebuild_resync
)
1281 struct imsm_map
*dest
;
1282 struct imsm_map
*src
= get_imsm_map(dev
, 0);
1284 dev
->vol
.migr_state
= 1;
1285 dev
->vol
.migr_type
= rebuild_resync
;
1286 dev
->vol
.curr_migr_unit
= 0;
1287 dest
= get_imsm_map(dev
, 1);
1289 memcpy(dest
, src
, sizeof_imsm_map(src
));
1290 src
->map_state
= to_state
;
1293 static void end_migration(struct imsm_dev
*dev
, __u8 map_state
)
1295 struct imsm_map
*map
= get_imsm_map(dev
, 0);
1297 dev
->vol
.migr_state
= 0;
1298 dev
->vol
.curr_migr_unit
= 0;
1299 map
->map_state
= map_state
;
1303 static int parse_raid_devices(struct intel_super
*super
)
1306 struct imsm_dev
*dev_new
;
1307 size_t len
, len_migr
;
1308 size_t space_needed
= 0;
1309 struct imsm_super
*mpb
= super
->anchor
;
1311 for (i
= 0; i
< super
->anchor
->num_raid_devs
; i
++) {
1312 struct imsm_dev
*dev_iter
= __get_imsm_dev(super
->anchor
, i
);
1314 len
= sizeof_imsm_dev(dev_iter
, 0);
1315 len_migr
= sizeof_imsm_dev(dev_iter
, 1);
1317 space_needed
+= len_migr
- len
;
1319 dev_new
= malloc(len_migr
);
1322 imsm_copy_dev(dev_new
, dev_iter
);
1323 super
->dev_tbl
[i
] = dev_new
;
1326 /* ensure that super->buf is large enough when all raid devices
1329 if (__le32_to_cpu(mpb
->mpb_size
) + space_needed
> super
->len
) {
1332 len
= ROUND_UP(__le32_to_cpu(mpb
->mpb_size
) + space_needed
, 512);
1333 if (posix_memalign(&buf
, 512, len
) != 0)
1336 memcpy(buf
, super
->buf
, len
);
1345 /* retrieve a pointer to the bbm log which starts after all raid devices */
1346 struct bbm_log
*__get_imsm_bbm_log(struct imsm_super
*mpb
)
1350 if (__le32_to_cpu(mpb
->bbm_log_size
)) {
1352 ptr
+= mpb
->mpb_size
- __le32_to_cpu(mpb
->bbm_log_size
);
1358 static void __free_imsm(struct intel_super
*super
, int free_disks
);
1360 /* load_imsm_mpb - read matrix metadata
1361 * allocates super->mpb to be freed by free_super
1363 static int load_imsm_mpb(int fd
, struct intel_super
*super
, char *devname
)
1365 unsigned long long dsize
;
1366 unsigned long long sectors
;
1368 struct imsm_super
*anchor
;
1372 get_dev_size(fd
, NULL
, &dsize
);
1374 if (lseek64(fd
, dsize
- (512 * 2), SEEK_SET
) < 0) {
1377 Name
": Cannot seek to anchor block on %s: %s\n",
1378 devname
, strerror(errno
));
1382 if (posix_memalign((void**)&anchor
, 512, 512) != 0) {
1385 Name
": Failed to allocate imsm anchor buffer"
1386 " on %s\n", devname
);
1389 if (read(fd
, anchor
, 512) != 512) {
1392 Name
": Cannot read anchor block on %s: %s\n",
1393 devname
, strerror(errno
));
1398 if (strncmp((char *) anchor
->sig
, MPB_SIGNATURE
, MPB_SIG_LEN
) != 0) {
1401 Name
": no IMSM anchor on %s\n", devname
);
1406 __free_imsm(super
, 0);
1407 super
->len
= ROUND_UP(anchor
->mpb_size
, 512);
1408 if (posix_memalign(&super
->buf
, 512, super
->len
) != 0) {
1411 Name
": unable to allocate %zu byte mpb buffer\n",
1416 memcpy(super
->buf
, anchor
, 512);
1418 sectors
= mpb_sectors(anchor
) - 1;
1421 rc
= load_imsm_disk(fd
, super
, devname
, 0);
1423 rc
= parse_raid_devices(super
);
1427 /* read the extended mpb */
1428 if (lseek64(fd
, dsize
- (512 * (2 + sectors
)), SEEK_SET
) < 0) {
1431 Name
": Cannot seek to extended mpb on %s: %s\n",
1432 devname
, strerror(errno
));
1436 if (read(fd
, super
->buf
+ 512, super
->len
- 512) != super
->len
- 512) {
1439 Name
": Cannot read extended mpb on %s: %s\n",
1440 devname
, strerror(errno
));
1444 check_sum
= __gen_imsm_checksum(super
->anchor
);
1445 if (check_sum
!= __le32_to_cpu(super
->anchor
->check_sum
)) {
1448 Name
": IMSM checksum %x != %x on %s\n",
1449 check_sum
, __le32_to_cpu(super
->anchor
->check_sum
),
1454 /* FIXME the BBM log is disk specific so we cannot use this global
1455 * buffer for all disks. Ok for now since we only look at the global
1456 * bbm_log_size parameter to gate assembly
1458 super
->bbm_log
= __get_imsm_bbm_log(super
->anchor
);
1460 rc
= load_imsm_disk(fd
, super
, devname
, 0);
1462 rc
= parse_raid_devices(super
);
1467 static void __free_imsm_disk(struct dl
*d
)
1478 static void free_imsm_disks(struct intel_super
*super
)
1482 while (super
->disks
) {
1484 super
->disks
= d
->next
;
1485 __free_imsm_disk(d
);
1487 while (super
->missing
) {
1489 super
->missing
= d
->next
;
1490 __free_imsm_disk(d
);
1495 /* free all the pieces hanging off of a super pointer */
1496 static void __free_imsm(struct intel_super
*super
, int free_disks
)
1505 free_imsm_disks(super
);
1506 for (i
= 0; i
< IMSM_MAX_RAID_DEVS
; i
++)
1507 if (super
->dev_tbl
[i
]) {
1508 free(super
->dev_tbl
[i
]);
1509 super
->dev_tbl
[i
] = NULL
;
1512 free((void *) super
->hba
);
1517 static void free_imsm(struct intel_super
*super
)
1519 __free_imsm(super
, 1);
1523 static void free_super_imsm(struct supertype
*st
)
1525 struct intel_super
*super
= st
->sb
;
1534 static struct intel_super
*alloc_super(int creating_imsm
)
1536 struct intel_super
*super
= malloc(sizeof(*super
));
1539 memset(super
, 0, sizeof(*super
));
1540 super
->creating_imsm
= creating_imsm
;
1541 super
->current_vol
= -1;
1542 super
->create_offset
= ~((__u32
) 0);
1543 if (!check_env("IMSM_NO_PLATFORM"))
1544 super
->orom
= find_imsm_orom();
1546 struct sys_dev
*list
, *ent
;
1548 /* find the first intel ahci controller */
1549 list
= find_driver_devices("pci", "ahci");
1550 for (ent
= list
; ent
; ent
= ent
->next
)
1551 if (devpath_to_vendor(ent
->path
) == 0x8086)
1554 super
->hba
= ent
->path
;
1557 free_sys_dev(&list
);
1565 /* find_missing - helper routine for load_super_imsm_all that identifies
1566 * disks that have disappeared from the system. This routine relies on
1567 * the mpb being uptodate, which it is at load time.
1569 static int find_missing(struct intel_super
*super
)
1572 struct imsm_super
*mpb
= super
->anchor
;
1574 struct imsm_disk
*disk
;
1576 for (i
= 0; i
< mpb
->num_disks
; i
++) {
1577 disk
= __get_imsm_disk(mpb
, i
);
1578 dl
= serial_to_dl(disk
->serial
, super
);
1581 /* ok we have a 'disk' without a live entry in
1584 if (disk
->status
& FAILED_DISK
|| !(disk
->status
& USABLE_DISK
))
1585 continue; /* never mind, already marked */
1587 dl
= malloc(sizeof(*dl
));
1593 dl
->devname
= strdup("missing");
1595 serialcpy(dl
->serial
, disk
->serial
);
1597 dl
->next
= super
->missing
;
1598 super
->missing
= dl
;
1604 static int load_super_imsm_all(struct supertype
*st
, int fd
, void **sbp
,
1605 char *devname
, int keep_fd
)
1608 struct intel_super
*super
;
1609 struct mdinfo
*sd
, *best
= NULL
;
1616 /* check if this disk is a member of an active array */
1617 sra
= sysfs_read(fd
, 0, GET_LEVEL
|GET_VERSION
|GET_DEVS
|GET_STATE
);
1621 if (sra
->array
.major_version
!= -1 ||
1622 sra
->array
.minor_version
!= -2 ||
1623 strcmp(sra
->text_version
, "imsm") != 0)
1626 super
= alloc_super(0);
1630 /* find the most up to date disk in this array, skipping spares */
1631 for (sd
= sra
->devs
; sd
; sd
= sd
->next
) {
1632 sprintf(nm
, "%d:%d", sd
->disk
.major
, sd
->disk
.minor
);
1633 dfd
= dev_open(nm
, keep_fd
? O_RDWR
: O_RDONLY
);
1638 rv
= load_imsm_mpb(dfd
, super
, NULL
);
1642 if (super
->anchor
->num_raid_devs
== 0)
1645 gen
= __le32_to_cpu(super
->anchor
->generation_num
);
1646 if (!best
|| gen
> bestgen
) {
1661 /* load the most up to date anchor */
1662 sprintf(nm
, "%d:%d", best
->disk
.major
, best
->disk
.minor
);
1663 dfd
= dev_open(nm
, O_RDONLY
);
1668 rv
= load_imsm_mpb(dfd
, super
, NULL
);
1675 /* re-parse the disk list with the current anchor */
1676 for (sd
= sra
->devs
; sd
; sd
= sd
->next
) {
1677 sprintf(nm
, "%d:%d", sd
->disk
.major
, sd
->disk
.minor
);
1678 dfd
= dev_open(nm
, keep_fd
? O_RDWR
: O_RDONLY
);
1683 load_imsm_disk(dfd
, super
, NULL
, keep_fd
);
1689 if (find_missing(super
) != 0) {
1694 if (st
->subarray
[0]) {
1695 if (atoi(st
->subarray
) <= super
->anchor
->num_raid_devs
)
1696 super
->current_vol
= atoi(st
->subarray
);
1702 st
->container_dev
= fd2devnum(fd
);
1703 if (st
->ss
== NULL
) {
1704 st
->ss
= &super_imsm
;
1705 st
->minor_version
= 0;
1706 st
->max_devs
= IMSM_MAX_DEVICES
;
1708 st
->loaded_container
= 1;
1714 static int load_super_imsm(struct supertype
*st
, int fd
, char *devname
)
1716 struct intel_super
*super
;
1720 if (load_super_imsm_all(st
, fd
, &st
->sb
, devname
, 1) == 0)
1723 if (st
->subarray
[0])
1724 return 1; /* FIXME */
1726 super
= alloc_super(0);
1729 Name
": malloc of %zu failed.\n",
1734 rv
= load_imsm_mpb(fd
, super
, devname
);
1739 Name
": Failed to load all information "
1740 "sections on %s\n", devname
);
1746 if (st
->ss
== NULL
) {
1747 st
->ss
= &super_imsm
;
1748 st
->minor_version
= 0;
1749 st
->max_devs
= IMSM_MAX_DEVICES
;
1751 st
->loaded_container
= 0;
1756 static __u16
info_to_blocks_per_strip(mdu_array_info_t
*info
)
1758 if (info
->level
== 1)
1760 return info
->chunk_size
>> 9;
1763 static __u32
info_to_num_data_stripes(mdu_array_info_t
*info
)
1767 num_stripes
= (info
->size
* 2) / info_to_blocks_per_strip(info
);
1768 if (info
->level
== 1)
1774 static __u32
info_to_blocks_per_member(mdu_array_info_t
*info
)
1776 return (info
->size
* 2) & ~(info_to_blocks_per_strip(info
) - 1);
1779 static void imsm_update_version_info(struct intel_super
*super
)
1781 /* update the version and attributes */
1782 struct imsm_super
*mpb
= super
->anchor
;
1784 struct imsm_dev
*dev
;
1785 struct imsm_map
*map
;
1788 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
1789 dev
= get_imsm_dev(super
, i
);
1790 map
= get_imsm_map(dev
, 0);
1791 if (__le32_to_cpu(dev
->size_high
) > 0)
1792 mpb
->attributes
|= MPB_ATTRIB_2TB
;
1794 /* FIXME detect when an array spans a port multiplier */
1796 mpb
->attributes
|= MPB_ATTRIB_PM
;
1799 if (mpb
->num_raid_devs
> 1 ||
1800 mpb
->attributes
!= MPB_ATTRIB_CHECKSUM_VERIFY
) {
1801 version
= MPB_VERSION_ATTRIBS
;
1802 switch (get_imsm_raid_level(map
)) {
1803 case 0: mpb
->attributes
|= MPB_ATTRIB_RAID0
; break;
1804 case 1: mpb
->attributes
|= MPB_ATTRIB_RAID1
; break;
1805 case 10: mpb
->attributes
|= MPB_ATTRIB_RAID10
; break;
1806 case 5: mpb
->attributes
|= MPB_ATTRIB_RAID5
; break;
1809 if (map
->num_members
>= 5)
1810 version
= MPB_VERSION_5OR6_DISK_ARRAY
;
1811 else if (dev
->status
== DEV_CLONE_N_GO
)
1812 version
= MPB_VERSION_CNG
;
1813 else if (get_imsm_raid_level(map
) == 5)
1814 version
= MPB_VERSION_RAID5
;
1815 else if (map
->num_members
>= 3)
1816 version
= MPB_VERSION_3OR4_DISK_ARRAY
;
1817 else if (get_imsm_raid_level(map
) == 1)
1818 version
= MPB_VERSION_RAID1
;
1820 version
= MPB_VERSION_RAID0
;
1822 strcpy(((char *) mpb
->sig
) + strlen(MPB_SIGNATURE
), version
);
1826 static int init_super_imsm_volume(struct supertype
*st
, mdu_array_info_t
*info
,
1827 unsigned long long size
, char *name
,
1828 char *homehost
, int *uuid
)
1830 /* We are creating a volume inside a pre-existing container.
1831 * so st->sb is already set.
1833 struct intel_super
*super
= st
->sb
;
1834 struct imsm_super
*mpb
= super
->anchor
;
1835 struct imsm_dev
*dev
;
1836 struct imsm_vol
*vol
;
1837 struct imsm_map
*map
;
1838 int idx
= mpb
->num_raid_devs
;
1840 unsigned long long array_blocks
;
1841 size_t size_old
, size_new
;
1843 if (super
->orom
&& mpb
->num_raid_devs
>= super
->orom
->vpa
) {
1844 fprintf(stderr
, Name
": This imsm-container already has the "
1845 "maximum of %d volumes\n", super
->orom
->vpa
);
1849 /* ensure the mpb is large enough for the new data */
1850 size_old
= __le32_to_cpu(mpb
->mpb_size
);
1851 size_new
= disks_to_mpb_size(info
->nr_disks
);
1852 if (size_new
> size_old
) {
1854 size_t size_round
= ROUND_UP(size_new
, 512);
1856 if (posix_memalign(&mpb_new
, 512, size_round
) != 0) {
1857 fprintf(stderr
, Name
": could not allocate new mpb\n");
1860 memcpy(mpb_new
, mpb
, size_old
);
1863 super
->anchor
= mpb_new
;
1864 mpb
->mpb_size
= __cpu_to_le32(size_new
);
1865 memset(mpb_new
+ size_old
, 0, size_round
- size_old
);
1867 super
->current_vol
= idx
;
1868 /* when creating the first raid device in this container set num_disks
1869 * to zero, i.e. delete this spare and add raid member devices in
1870 * add_to_super_imsm_volume()
1872 if (super
->current_vol
== 0)
1874 sprintf(st
->subarray
, "%d", idx
);
1875 dev
= malloc(sizeof(*dev
) + sizeof(__u32
) * (info
->raid_disks
- 1));
1877 fprintf(stderr
, Name
": could not allocate raid device\n");
1880 strncpy((char *) dev
->volume
, name
, MAX_RAID_SERIAL_LEN
);
1881 if (info
->level
== 1)
1882 array_blocks
= info_to_blocks_per_member(info
);
1884 array_blocks
= calc_array_size(info
->level
, info
->raid_disks
,
1885 info
->layout
, info
->chunk_size
,
1887 dev
->size_low
= __cpu_to_le32((__u32
) array_blocks
);
1888 dev
->size_high
= __cpu_to_le32((__u32
) (array_blocks
>> 32));
1889 dev
->status
= __cpu_to_le32(0);
1890 dev
->reserved_blocks
= __cpu_to_le32(0);
1892 vol
->migr_state
= 0;
1893 vol
->migr_type
= MIGR_INIT
;
1895 vol
->curr_migr_unit
= 0;
1896 map
= get_imsm_map(dev
, 0);
1897 map
->pba_of_lba0
= __cpu_to_le32(super
->create_offset
);
1898 map
->blocks_per_member
= __cpu_to_le32(info_to_blocks_per_member(info
));
1899 map
->blocks_per_strip
= __cpu_to_le16(info_to_blocks_per_strip(info
));
1900 map
->num_data_stripes
= __cpu_to_le32(info_to_num_data_stripes(info
));
1901 map
->map_state
= info
->level
? IMSM_T_STATE_UNINITIALIZED
:
1902 IMSM_T_STATE_NORMAL
;
1904 if (info
->level
== 1 && info
->raid_disks
> 2) {
1905 fprintf(stderr
, Name
": imsm does not support more than 2 disks"
1906 "in a raid1 volume\n");
1909 if (info
->level
== 10) {
1910 map
->raid_level
= 1;
1911 map
->num_domains
= info
->raid_disks
/ 2;
1913 map
->raid_level
= info
->level
;
1914 map
->num_domains
= !!map
->raid_level
;
1917 map
->num_members
= info
->raid_disks
;
1918 for (i
= 0; i
< map
->num_members
; i
++) {
1919 /* initialized in add_to_super */
1920 set_imsm_ord_tbl_ent(map
, i
, 0);
1922 mpb
->num_raid_devs
++;
1923 super
->dev_tbl
[super
->current_vol
] = dev
;
1925 imsm_update_version_info(super
);
1930 static int init_super_imsm(struct supertype
*st
, mdu_array_info_t
*info
,
1931 unsigned long long size
, char *name
,
1932 char *homehost
, int *uuid
)
1934 /* This is primarily called by Create when creating a new array.
1935 * We will then get add_to_super called for each component, and then
1936 * write_init_super called to write it out to each device.
1937 * For IMSM, Create can create on fresh devices or on a pre-existing
1939 * To create on a pre-existing array a different method will be called.
1940 * This one is just for fresh drives.
1942 struct intel_super
*super
;
1943 struct imsm_super
*mpb
;
1952 return init_super_imsm_volume(st
, info
, size
, name
, homehost
,
1955 super
= alloc_super(1);
1958 mpb_size
= disks_to_mpb_size(info
->nr_disks
);
1959 if (posix_memalign(&super
->buf
, 512, mpb_size
) != 0) {
1964 memset(mpb
, 0, mpb_size
);
1966 mpb
->attributes
= MPB_ATTRIB_CHECKSUM_VERIFY
;
1968 version
= (char *) mpb
->sig
;
1969 strcpy(version
, MPB_SIGNATURE
);
1970 version
+= strlen(MPB_SIGNATURE
);
1971 strcpy(version
, MPB_VERSION_RAID0
);
1972 mpb
->mpb_size
= mpb_size
;
1979 static int add_to_super_imsm_volume(struct supertype
*st
, mdu_disk_info_t
*dk
,
1980 int fd
, char *devname
)
1982 struct intel_super
*super
= st
->sb
;
1983 struct imsm_super
*mpb
= super
->anchor
;
1985 struct imsm_dev
*dev
;
1986 struct imsm_map
*map
;
1988 dev
= get_imsm_dev(super
, super
->current_vol
);
1989 map
= get_imsm_map(dev
, 0);
1991 if (! (dk
->state
& (1<<MD_DISK_SYNC
))) {
1992 fprintf(stderr
, Name
": %s: Cannot add spare devices to IMSM volume\n",
1997 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
1998 if (dl
->major
== dk
->major
&&
1999 dl
->minor
== dk
->minor
)
2003 fprintf(stderr
, Name
": %s is not a member of the same container\n", devname
);
2007 /* add a pristine spare to the metadata */
2008 if (dl
->index
< 0) {
2009 dl
->index
= super
->anchor
->num_disks
;
2010 super
->anchor
->num_disks
++;
2012 set_imsm_ord_tbl_ent(map
, dk
->number
, dl
->index
);
2013 dl
->disk
.status
= CONFIGURED_DISK
| USABLE_DISK
;
2015 /* if we are creating the first raid device update the family number */
2016 if (super
->current_vol
== 0) {
2018 struct imsm_dev
*_dev
= __get_imsm_dev(mpb
, 0);
2019 struct imsm_disk
*_disk
= __get_imsm_disk(mpb
, dl
->index
);
2023 sum
= __gen_imsm_checksum(mpb
);
2024 mpb
->family_num
= __cpu_to_le32(sum
);
2030 static int add_to_super_imsm(struct supertype
*st
, mdu_disk_info_t
*dk
,
2031 int fd
, char *devname
)
2033 struct intel_super
*super
= st
->sb
;
2035 unsigned long long size
;
2040 /* if we are on an RAID enabled platform check that the disk is
2041 * attached to the raid controller
2043 if (super
->hba
&& !disk_attached_to_hba(fd
, super
->hba
)) {
2045 Name
": %s is not attached to the raid controller: %s\n",
2046 devname
? : "disk", super
->hba
);
2050 if (super
->current_vol
>= 0)
2051 return add_to_super_imsm_volume(st
, dk
, fd
, devname
);
2054 dd
= malloc(sizeof(*dd
));
2057 Name
": malloc failed %s:%d.\n", __func__
, __LINE__
);
2060 memset(dd
, 0, sizeof(*dd
));
2061 dd
->major
= major(stb
.st_rdev
);
2062 dd
->minor
= minor(stb
.st_rdev
);
2064 dd
->devname
= devname
? strdup(devname
) : NULL
;
2066 rv
= imsm_read_serial(fd
, devname
, dd
->serial
);
2069 Name
": failed to retrieve scsi serial, aborting\n");
2074 get_dev_size(fd
, NULL
, &size
);
2076 serialcpy(dd
->disk
.serial
, dd
->serial
);
2077 dd
->disk
.total_blocks
= __cpu_to_le32(size
);
2078 dd
->disk
.status
= USABLE_DISK
| SPARE_DISK
;
2079 if (sysfs_disk_to_scsi_id(fd
, &id
) == 0)
2080 dd
->disk
.scsi_id
= __cpu_to_le32(id
);
2082 dd
->disk
.scsi_id
= __cpu_to_le32(0);
2084 if (st
->update_tail
) {
2085 dd
->next
= super
->add
;
2088 dd
->next
= super
->disks
;
2095 static int store_imsm_mpb(int fd
, struct intel_super
*super
);
2097 /* spare records have their own family number and do not have any defined raid
2100 static int write_super_imsm_spares(struct intel_super
*super
, int doclose
)
2102 struct imsm_super mpb_save
;
2103 struct imsm_super
*mpb
= super
->anchor
;
2108 mpb
->num_raid_devs
= 0;
2110 mpb
->mpb_size
= sizeof(struct imsm_super
);
2111 mpb
->generation_num
= __cpu_to_le32(1UL);
2113 for (d
= super
->disks
; d
; d
= d
->next
) {
2117 mpb
->disk
[0] = d
->disk
;
2118 sum
= __gen_imsm_checksum(mpb
);
2119 mpb
->family_num
= __cpu_to_le32(sum
);
2120 sum
= __gen_imsm_checksum(mpb
);
2121 mpb
->check_sum
= __cpu_to_le32(sum
);
2123 if (store_imsm_mpb(d
->fd
, super
)) {
2124 fprintf(stderr
, "%s: failed for device %d:%d %s\n",
2125 __func__
, d
->major
, d
->minor
, strerror(errno
));
2139 static int write_super_imsm(struct intel_super
*super
, int doclose
)
2141 struct imsm_super
*mpb
= super
->anchor
;
2147 __u32 mpb_size
= sizeof(struct imsm_super
) - sizeof(struct imsm_disk
);
2149 /* 'generation' is incremented everytime the metadata is written */
2150 generation
= __le32_to_cpu(mpb
->generation_num
);
2152 mpb
->generation_num
= __cpu_to_le32(generation
);
2154 mpb_size
+= sizeof(struct imsm_disk
) * mpb
->num_disks
;
2155 for (d
= super
->disks
; d
; d
= d
->next
) {
2159 mpb
->disk
[d
->index
] = d
->disk
;
2161 for (d
= super
->missing
; d
; d
= d
->next
)
2162 mpb
->disk
[d
->index
] = d
->disk
;
2164 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
2165 struct imsm_dev
*dev
= __get_imsm_dev(mpb
, i
);
2167 imsm_copy_dev(dev
, super
->dev_tbl
[i
]);
2168 mpb_size
+= sizeof_imsm_dev(dev
, 0);
2170 mpb_size
+= __le32_to_cpu(mpb
->bbm_log_size
);
2171 mpb
->mpb_size
= __cpu_to_le32(mpb_size
);
2173 /* recalculate checksum */
2174 sum
= __gen_imsm_checksum(mpb
);
2175 mpb
->check_sum
= __cpu_to_le32(sum
);
2177 /* write the mpb for disks that compose raid devices */
2178 for (d
= super
->disks
; d
; d
= d
->next
) {
2181 if (store_imsm_mpb(d
->fd
, super
))
2182 fprintf(stderr
, "%s: failed for device %d:%d %s\n",
2183 __func__
, d
->major
, d
->minor
, strerror(errno
));
2191 return write_super_imsm_spares(super
, doclose
);
2197 static int create_array(struct supertype
*st
)
2200 struct imsm_update_create_array
*u
;
2201 struct intel_super
*super
= st
->sb
;
2202 struct imsm_dev
*dev
= get_imsm_dev(super
, super
->current_vol
);
2203 struct imsm_map
*map
= get_imsm_map(dev
, 0);
2204 struct disk_info
*inf
;
2205 struct imsm_disk
*disk
;
2209 len
= sizeof(*u
) - sizeof(*dev
) + sizeof_imsm_dev(dev
, 0) +
2210 sizeof(*inf
) * map
->num_members
;
2213 fprintf(stderr
, "%s: failed to allocate update buffer\n",
2218 u
->type
= update_create_array
;
2219 u
->dev_idx
= super
->current_vol
;
2220 imsm_copy_dev(&u
->dev
, dev
);
2221 inf
= get_disk_info(u
);
2222 for (i
= 0; i
< map
->num_members
; i
++) {
2223 idx
= get_imsm_disk_idx(dev
, i
);
2224 disk
= get_imsm_disk(super
, idx
);
2225 serialcpy(inf
[i
].serial
, disk
->serial
);
2227 append_metadata_update(st
, u
, len
);
2232 static int _add_disk(struct supertype
*st
)
2234 struct intel_super
*super
= st
->sb
;
2236 struct imsm_update_add_disk
*u
;
2244 fprintf(stderr
, "%s: failed to allocate update buffer\n",
2249 u
->type
= update_add_disk
;
2250 append_metadata_update(st
, u
, len
);
2255 static int write_init_super_imsm(struct supertype
*st
)
2257 if (st
->update_tail
) {
2258 /* queue the recently created array / added disk
2259 * as a metadata update */
2260 struct intel_super
*super
= st
->sb
;
2264 /* determine if we are creating a volume or adding a disk */
2265 if (super
->current_vol
< 0) {
2266 /* in the add disk case we are running in mdmon
2267 * context, so don't close fd's
2269 return _add_disk(st
);
2271 rv
= create_array(st
);
2273 for (d
= super
->disks
; d
; d
= d
->next
) {
2280 return write_super_imsm(st
->sb
, 1);
2284 static int store_zero_imsm(struct supertype
*st
, int fd
)
2286 unsigned long long dsize
;
2289 get_dev_size(fd
, NULL
, &dsize
);
2291 /* first block is stored on second to last sector of the disk */
2292 if (lseek64(fd
, dsize
- (512 * 2), SEEK_SET
) < 0)
2295 if (posix_memalign(&buf
, 512, 512) != 0)
2298 memset(buf
, 0, 512);
2299 if (write(fd
, buf
, 512) != 512)
2304 static int imsm_bbm_log_size(struct imsm_super
*mpb
)
2306 return __le32_to_cpu(mpb
->bbm_log_size
);
2310 static int validate_geometry_imsm_container(struct supertype
*st
, int level
,
2311 int layout
, int raiddisks
, int chunk
,
2312 unsigned long long size
, char *dev
,
2313 unsigned long long *freesize
,
2317 unsigned long long ldsize
;
2318 const struct imsm_orom
*orom
;
2320 if (level
!= LEVEL_CONTAINER
)
2325 if (check_env("IMSM_NO_PLATFORM"))
2328 orom
= find_imsm_orom();
2329 if (orom
&& raiddisks
> orom
->tds
) {
2331 fprintf(stderr
, Name
": %d exceeds maximum number of"
2332 " platform supported disks: %d\n",
2333 raiddisks
, orom
->tds
);
2337 fd
= open(dev
, O_RDONLY
|O_EXCL
, 0);
2340 fprintf(stderr
, Name
": imsm: Cannot open %s: %s\n",
2341 dev
, strerror(errno
));
2344 if (!get_dev_size(fd
, dev
, &ldsize
)) {
2350 *freesize
= avail_size_imsm(st
, ldsize
>> 9);
2355 static unsigned long long find_size(struct extent
*e
, int *idx
, int num_extents
)
2357 const unsigned long long base_start
= e
[*idx
].start
;
2358 unsigned long long end
= base_start
+ e
[*idx
].size
;
2361 if (base_start
== end
)
2365 for (i
= *idx
; i
< num_extents
; i
++) {
2366 /* extend overlapping extents */
2367 if (e
[i
].start
>= base_start
&&
2368 e
[i
].start
<= end
) {
2371 if (e
[i
].start
+ e
[i
].size
> end
)
2372 end
= e
[i
].start
+ e
[i
].size
;
2373 } else if (e
[i
].start
> end
) {
2379 return end
- base_start
;
2382 static unsigned long long merge_extents(struct intel_super
*super
, int sum_extents
)
2384 /* build a composite disk with all known extents and generate a new
2385 * 'maxsize' given the "all disks in an array must share a common start
2386 * offset" constraint
2388 struct extent
*e
= calloc(sum_extents
, sizeof(*e
));
2392 unsigned long long pos
;
2393 unsigned long long start
;
2394 unsigned long long maxsize
;
2395 unsigned long reserve
;
2398 return ~0ULL; /* error */
2400 /* coalesce and sort all extents. also, check to see if we need to
2401 * reserve space between member arrays
2404 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
2407 for (i
= 0; i
< dl
->extent_cnt
; i
++)
2410 qsort(e
, sum_extents
, sizeof(*e
), cmp_extent
);
2415 while (i
< sum_extents
) {
2416 e
[j
].start
= e
[i
].start
;
2417 e
[j
].size
= find_size(e
, &i
, sum_extents
);
2419 if (e
[j
-1].size
== 0)
2428 unsigned long long esize
;
2430 esize
= e
[i
].start
- pos
;
2431 if (esize
>= maxsize
) {
2436 pos
= e
[i
].start
+ e
[i
].size
;
2438 } while (e
[i
-1].size
);
2441 if (start_extent
> 0)
2442 reserve
= IMSM_RESERVED_SECTORS
; /* gap between raid regions */
2446 if (maxsize
< reserve
)
2449 super
->create_offset
= ~((__u32
) 0);
2450 if (start
+ reserve
> super
->create_offset
)
2451 return ~0ULL; /* start overflows create_offset */
2452 super
->create_offset
= start
+ reserve
;
2454 return maxsize
- reserve
;
2457 static int is_raid_level_supported(const struct imsm_orom
*orom
, int level
, int raiddisks
)
2459 if (level
< 0 || level
== 6 || level
== 4)
2462 /* if we have an orom prevent invalid raid levels */
2465 case 0: return imsm_orom_has_raid0(orom
);
2468 return imsm_orom_has_raid1e(orom
);
2470 return imsm_orom_has_raid1(orom
);
2471 case 10: return imsm_orom_has_raid10(orom
);
2472 case 5: return imsm_orom_has_raid5(orom
);
2475 return 1; /* not on an Intel RAID platform so anything goes */
2480 #define vprintf(fmt, arg...) (void) (verbose && fprintf(stderr, Name fmt, ##arg))
2481 /* validate_geometry_imsm_volume - lifted from validate_geometry_ddf_bvd
2482 * FIX ME add ahci details
2484 static int validate_geometry_imsm_volume(struct supertype
*st
, int level
,
2485 int layout
, int raiddisks
, int chunk
,
2486 unsigned long long size
, char *dev
,
2487 unsigned long long *freesize
,
2491 struct intel_super
*super
= st
->sb
;
2493 unsigned long long pos
= 0;
2494 unsigned long long maxsize
;
2498 /* We must have the container info already read in. */
2502 if (!is_raid_level_supported(super
->orom
, level
, raiddisks
)) {
2503 vprintf(": platform does not support raid level: %d\n", level
);
2506 if (super
->orom
&& !imsm_orom_has_chunk(super
->orom
, chunk
)) {
2507 vprintf(": platform does not support a chunk size of: %d\n", chunk
);
2510 if (layout
!= imsm_level_to_layout(level
)) {
2512 vprintf(": imsm raid 5 only supports the left-asymmetric layout\n");
2513 else if (level
== 10)
2514 vprintf(": imsm raid 10 only supports the n2 layout\n");
2516 vprintf(": imsm unknown layout %#x for this raid level %d\n",
2522 /* General test: make sure there is space for
2523 * 'raiddisks' device extents of size 'size' at a given
2526 unsigned long long minsize
= size
*2 /* convert to blocks */;
2527 unsigned long long start_offset
= ~0ULL;
2530 minsize
= MPB_SECTOR_CNT
+ IMSM_RESERVED_SECTORS
;
2531 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
2536 e
= get_extents(super
, dl
);
2539 unsigned long long esize
;
2540 esize
= e
[i
].start
- pos
;
2541 if (esize
>= minsize
)
2543 if (found
&& start_offset
== ~0ULL) {
2546 } else if (found
&& pos
!= start_offset
) {
2550 pos
= e
[i
].start
+ e
[i
].size
;
2552 } while (e
[i
-1].size
);
2557 if (dcnt
< raiddisks
) {
2559 fprintf(stderr
, Name
": imsm: Not enough "
2560 "devices with space for this array "
2568 /* This device must be a member of the set */
2569 if (stat(dev
, &stb
) < 0)
2571 if ((S_IFMT
& stb
.st_mode
) != S_IFBLK
)
2573 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
2574 if (dl
->major
== major(stb
.st_rdev
) &&
2575 dl
->minor
== minor(stb
.st_rdev
))
2580 fprintf(stderr
, Name
": %s is not in the "
2581 "same imsm set\n", dev
);
2585 /* retrieve the largest free space block */
2586 e
= get_extents(super
, dl
);
2591 unsigned long long esize
;
2593 esize
= e
[i
].start
- pos
;
2594 if (esize
>= maxsize
)
2596 pos
= e
[i
].start
+ e
[i
].size
;
2598 } while (e
[i
-1].size
);
2603 fprintf(stderr
, Name
": unable to determine free space for: %s\n",
2607 if (maxsize
< size
) {
2609 fprintf(stderr
, Name
": %s not enough space (%llu < %llu)\n",
2610 dev
, maxsize
, size
);
2614 /* count total number of extents for merge */
2616 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
2618 i
+= dl
->extent_cnt
;
2620 maxsize
= merge_extents(super
, i
);
2621 if (maxsize
< size
) {
2623 fprintf(stderr
, Name
": not enough space after merge (%llu < %llu)\n",
2626 } else if (maxsize
== ~0ULL) {
2628 fprintf(stderr
, Name
": failed to merge %d extents\n", i
);
2632 *freesize
= maxsize
;
2637 static int validate_geometry_imsm(struct supertype
*st
, int level
, int layout
,
2638 int raiddisks
, int chunk
, unsigned long long size
,
2639 char *dev
, unsigned long long *freesize
,
2645 /* if given unused devices create a container
2646 * if given given devices in a container create a member volume
2648 if (level
== LEVEL_CONTAINER
) {
2649 /* Must be a fresh device to add to a container */
2650 return validate_geometry_imsm_container(st
, level
, layout
,
2651 raiddisks
, chunk
, size
,
2657 /* creating in a given container */
2658 return validate_geometry_imsm_volume(st
, level
, layout
,
2659 raiddisks
, chunk
, size
,
2660 dev
, freesize
, verbose
);
2663 /* limit creation to the following levels */
2675 /* This device needs to be a device in an 'imsm' container */
2676 fd
= open(dev
, O_RDONLY
|O_EXCL
, 0);
2680 Name
": Cannot create this array on device %s\n",
2685 if (errno
!= EBUSY
|| (fd
= open(dev
, O_RDONLY
, 0)) < 0) {
2687 fprintf(stderr
, Name
": Cannot open %s: %s\n",
2688 dev
, strerror(errno
));
2691 /* Well, it is in use by someone, maybe an 'imsm' container. */
2692 cfd
= open_container(fd
);
2696 fprintf(stderr
, Name
": Cannot use %s: It is busy\n",
2700 sra
= sysfs_read(cfd
, 0, GET_VERSION
);
2702 if (sra
&& sra
->array
.major_version
== -1 &&
2703 strcmp(sra
->text_version
, "imsm") == 0) {
2704 /* This is a member of a imsm container. Load the container
2705 * and try to create a volume
2707 struct intel_super
*super
;
2709 if (load_super_imsm_all(st
, cfd
, (void **) &super
, NULL
, 1) == 0) {
2711 st
->container_dev
= fd2devnum(cfd
);
2713 return validate_geometry_imsm_volume(st
, level
, layout
,
2719 } else /* may belong to another container */
2724 #endif /* MDASSEMBLE */
2726 static struct mdinfo
*container_content_imsm(struct supertype
*st
)
2728 /* Given a container loaded by load_super_imsm_all,
2729 * extract information about all the arrays into
2732 * For each imsm_dev create an mdinfo, fill it in,
2733 * then look for matching devices in super->disks
2734 * and create appropriate device mdinfo.
2736 struct intel_super
*super
= st
->sb
;
2737 struct imsm_super
*mpb
= super
->anchor
;
2738 struct mdinfo
*rest
= NULL
;
2741 /* do not assemble arrays that might have bad blocks */
2742 if (imsm_bbm_log_size(super
->anchor
)) {
2743 fprintf(stderr
, Name
": BBM log found in metadata. "
2744 "Cannot activate array(s).\n");
2748 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
2749 struct imsm_dev
*dev
= get_imsm_dev(super
, i
);
2750 struct imsm_map
*map
= get_imsm_map(dev
, 0);
2751 struct mdinfo
*this;
2754 this = malloc(sizeof(*this));
2755 memset(this, 0, sizeof(*this));
2758 super
->current_vol
= i
;
2759 getinfo_super_imsm_volume(st
, this);
2760 for (slot
= 0 ; slot
< map
->num_members
; slot
++) {
2761 struct mdinfo
*info_d
;
2769 idx
= get_imsm_disk_idx(dev
, slot
);
2770 ord
= get_imsm_ord_tbl_ent(dev
, slot
);
2771 for (d
= super
->disks
; d
; d
= d
->next
)
2772 if (d
->index
== idx
)
2778 s
= d
? d
->disk
.status
: 0;
2779 if (s
& FAILED_DISK
)
2781 if (!(s
& USABLE_DISK
))
2783 if (ord
& IMSM_ORD_REBUILD
)
2787 * if we skip some disks the array will be assmebled degraded;
2788 * reset resync start to avoid a dirty-degraded situation
2790 * FIXME handle dirty degraded
2792 if (skip
&& !dev
->vol
.dirty
)
2793 this->resync_start
= ~0ULL;
2797 info_d
= malloc(sizeof(*info_d
));
2799 fprintf(stderr
, Name
": failed to allocate disk"
2800 " for volume %s\n", (char *) dev
->volume
);
2805 memset(info_d
, 0, sizeof(*info_d
));
2806 info_d
->next
= this->devs
;
2807 this->devs
= info_d
;
2809 info_d
->disk
.number
= d
->index
;
2810 info_d
->disk
.major
= d
->major
;
2811 info_d
->disk
.minor
= d
->minor
;
2812 info_d
->disk
.raid_disk
= slot
;
2814 this->array
.working_disks
++;
2816 info_d
->events
= __le32_to_cpu(mpb
->generation_num
);
2817 info_d
->data_offset
= __le32_to_cpu(map
->pba_of_lba0
);
2818 info_d
->component_size
= __le32_to_cpu(map
->blocks_per_member
);
2820 strcpy(info_d
->name
, d
->devname
);
2830 static int imsm_open_new(struct supertype
*c
, struct active_array
*a
,
2833 struct intel_super
*super
= c
->sb
;
2834 struct imsm_super
*mpb
= super
->anchor
;
2836 if (atoi(inst
) >= mpb
->num_raid_devs
) {
2837 fprintf(stderr
, "%s: subarry index %d, out of range\n",
2838 __func__
, atoi(inst
));
2842 dprintf("imsm: open_new %s\n", inst
);
2843 a
->info
.container_member
= atoi(inst
);
2847 static __u8
imsm_check_degraded(struct intel_super
*super
, struct imsm_dev
*dev
, int failed
)
2849 struct imsm_map
*map
= get_imsm_map(dev
, 0);
2852 return map
->map_state
== IMSM_T_STATE_UNINITIALIZED
?
2853 IMSM_T_STATE_UNINITIALIZED
: IMSM_T_STATE_NORMAL
;
2855 switch (get_imsm_raid_level(map
)) {
2857 return IMSM_T_STATE_FAILED
;
2860 if (failed
< map
->num_members
)
2861 return IMSM_T_STATE_DEGRADED
;
2863 return IMSM_T_STATE_FAILED
;
2868 * check to see if any mirrors have failed, otherwise we
2869 * are degraded. Even numbered slots are mirrored on
2873 /* gcc -Os complains that this is unused */
2874 int insync
= insync
;
2876 for (i
= 0; i
< map
->num_members
; i
++) {
2877 __u32 ord
= get_imsm_ord_tbl_ent(dev
, i
);
2878 int idx
= ord_to_idx(ord
);
2879 struct imsm_disk
*disk
;
2881 /* reset the potential in-sync count on even-numbered
2882 * slots. num_copies is always 2 for imsm raid10
2887 disk
= get_imsm_disk(super
, idx
);
2888 if (!disk
|| disk
->status
& FAILED_DISK
||
2889 ord
& IMSM_ORD_REBUILD
)
2892 /* no in-sync disks left in this mirror the
2896 return IMSM_T_STATE_FAILED
;
2899 return IMSM_T_STATE_DEGRADED
;
2903 return IMSM_T_STATE_DEGRADED
;
2905 return IMSM_T_STATE_FAILED
;
2911 return map
->map_state
;
2914 static int imsm_count_failed(struct intel_super
*super
, struct imsm_dev
*dev
)
2918 struct imsm_disk
*disk
;
2919 struct imsm_map
*map
= get_imsm_map(dev
, 0);
2921 for (i
= 0; i
< map
->num_members
; i
++) {
2922 __u32 ord
= get_imsm_ord_tbl_ent(dev
, i
);
2923 int idx
= ord_to_idx(ord
);
2925 disk
= get_imsm_disk(super
, idx
);
2926 if (!disk
|| disk
->status
& FAILED_DISK
||
2927 ord
& IMSM_ORD_REBUILD
)
2934 static int is_resyncing(struct imsm_dev
*dev
)
2936 struct imsm_map
*migr_map
;
2938 if (!dev
->vol
.migr_state
)
2941 if (dev
->vol
.migr_type
== MIGR_INIT
)
2944 migr_map
= get_imsm_map(dev
, 1);
2946 if (migr_map
->map_state
== IMSM_T_STATE_NORMAL
)
2952 static int is_rebuilding(struct imsm_dev
*dev
)
2954 struct imsm_map
*migr_map
;
2956 if (!dev
->vol
.migr_state
)
2959 if (dev
->vol
.migr_type
!= MIGR_REBUILD
)
2962 migr_map
= get_imsm_map(dev
, 1);
2964 if (migr_map
->map_state
== IMSM_T_STATE_DEGRADED
)
2970 static void mark_failure(struct imsm_disk
*disk
)
2972 if (disk
->status
& FAILED_DISK
)
2974 disk
->status
|= FAILED_DISK
;
2975 disk
->scsi_id
= __cpu_to_le32(~(__u32
)0);
2976 memmove(&disk
->serial
[0], &disk
->serial
[1], MAX_RAID_SERIAL_LEN
- 1);
2979 /* Handle dirty -> clean transititions and resync. Degraded and rebuild
2980 * states are handled in imsm_set_disk() with one exception, when a
2981 * resync is stopped due to a new failure this routine will set the
2982 * 'degraded' state for the array.
2984 static int imsm_set_array_state(struct active_array
*a
, int consistent
)
2986 int inst
= a
->info
.container_member
;
2987 struct intel_super
*super
= a
->container
->sb
;
2988 struct imsm_dev
*dev
= get_imsm_dev(super
, inst
);
2989 struct imsm_map
*map
= get_imsm_map(dev
, 0);
2990 int failed
= imsm_count_failed(super
, dev
);
2991 __u8 map_state
= imsm_check_degraded(super
, dev
, failed
);
2993 /* before we activate this array handle any missing disks */
2994 if (consistent
== 2 && super
->missing
) {
2997 dprintf("imsm: mark missing\n");
2998 end_migration(dev
, map_state
);
2999 for (dl
= super
->missing
; dl
; dl
= dl
->next
)
3000 mark_failure(&dl
->disk
);
3001 super
->updates_pending
++;
3004 if (consistent
== 2 &&
3005 (!is_resync_complete(a
) ||
3006 map_state
!= IMSM_T_STATE_NORMAL
||
3007 dev
->vol
.migr_state
))
3010 if (is_resync_complete(a
)) {
3011 /* complete intialization / resync,
3012 * recovery is completed in ->set_disk
3014 if (is_resyncing(dev
)) {
3015 dprintf("imsm: mark resync done\n");
3016 end_migration(dev
, map_state
);
3017 super
->updates_pending
++;
3019 } else if (!is_resyncing(dev
) && !failed
) {
3020 /* mark the start of the init process if nothing is failed */
3021 dprintf("imsm: mark resync start (%llu)\n", a
->resync_start
);
3022 if (map
->map_state
== IMSM_T_STATE_NORMAL
)
3023 migrate(dev
, IMSM_T_STATE_NORMAL
, MIGR_REBUILD
);
3025 migrate(dev
, IMSM_T_STATE_NORMAL
, MIGR_INIT
);
3026 super
->updates_pending
++;
3029 /* check if we can update the migration checkpoint */
3030 if (dev
->vol
.migr_state
&&
3031 __le32_to_cpu(dev
->vol
.curr_migr_unit
) != a
->resync_start
) {
3032 dprintf("imsm: checkpoint migration (%llu)\n", a
->resync_start
);
3033 dev
->vol
.curr_migr_unit
= __cpu_to_le32(a
->resync_start
);
3034 super
->updates_pending
++;
3037 /* mark dirty / clean */
3038 if (dev
->vol
.dirty
!= !consistent
) {
3039 dprintf("imsm: mark '%s' (%llu)\n",
3040 consistent
? "clean" : "dirty", a
->resync_start
);
3045 super
->updates_pending
++;
3050 static void imsm_set_disk(struct active_array
*a
, int n
, int state
)
3052 int inst
= a
->info
.container_member
;
3053 struct intel_super
*super
= a
->container
->sb
;
3054 struct imsm_dev
*dev
= get_imsm_dev(super
, inst
);
3055 struct imsm_map
*map
= get_imsm_map(dev
, 0);
3056 struct imsm_disk
*disk
;
3061 if (n
> map
->num_members
)
3062 fprintf(stderr
, "imsm: set_disk %d out of range 0..%d\n",
3063 n
, map
->num_members
- 1);
3068 dprintf("imsm: set_disk %d:%x\n", n
, state
);
3070 ord
= get_imsm_ord_tbl_ent(dev
, n
);
3071 disk
= get_imsm_disk(super
, ord_to_idx(ord
));
3073 /* check for new failures */
3074 if ((state
& DS_FAULTY
) && !(disk
->status
& FAILED_DISK
)) {
3076 super
->updates_pending
++;
3079 /* check if in_sync */
3080 if (state
& DS_INSYNC
&& ord
& IMSM_ORD_REBUILD
) {
3081 struct imsm_map
*migr_map
= get_imsm_map(dev
, 1);
3083 set_imsm_ord_tbl_ent(migr_map
, n
, ord_to_idx(ord
));
3084 super
->updates_pending
++;
3087 failed
= imsm_count_failed(super
, dev
);
3088 map_state
= imsm_check_degraded(super
, dev
, failed
);
3090 /* check if recovery complete, newly degraded, or failed */
3091 if (map_state
== IMSM_T_STATE_NORMAL
&& is_rebuilding(dev
)) {
3092 end_migration(dev
, map_state
);
3093 super
->updates_pending
++;
3094 } else if (map_state
== IMSM_T_STATE_DEGRADED
&&
3095 map
->map_state
!= map_state
&&
3096 !dev
->vol
.migr_state
) {
3097 dprintf("imsm: mark degraded\n");
3098 map
->map_state
= map_state
;
3099 super
->updates_pending
++;
3100 } else if (map_state
== IMSM_T_STATE_FAILED
&&
3101 map
->map_state
!= map_state
) {
3102 dprintf("imsm: mark failed\n");
3103 end_migration(dev
, map_state
);
3104 super
->updates_pending
++;
3108 static int store_imsm_mpb(int fd
, struct intel_super
*super
)
3110 struct imsm_super
*mpb
= super
->anchor
;
3111 __u32 mpb_size
= __le32_to_cpu(mpb
->mpb_size
);
3112 unsigned long long dsize
;
3113 unsigned long long sectors
;
3115 get_dev_size(fd
, NULL
, &dsize
);
3117 if (mpb_size
> 512) {
3118 /* -1 to account for anchor */
3119 sectors
= mpb_sectors(mpb
) - 1;
3121 /* write the extended mpb to the sectors preceeding the anchor */
3122 if (lseek64(fd
, dsize
- (512 * (2 + sectors
)), SEEK_SET
) < 0)
3125 if (write(fd
, super
->buf
+ 512, 512 * sectors
) != 512 * sectors
)
3129 /* first block is stored on second to last sector of the disk */
3130 if (lseek64(fd
, dsize
- (512 * 2), SEEK_SET
) < 0)
3133 if (write(fd
, super
->buf
, 512) != 512)
3139 static void imsm_sync_metadata(struct supertype
*container
)
3141 struct intel_super
*super
= container
->sb
;
3143 if (!super
->updates_pending
)
3146 write_super_imsm(super
, 0);
3148 super
->updates_pending
= 0;
3151 static struct dl
*imsm_readd(struct intel_super
*super
, int idx
, struct active_array
*a
)
3153 struct imsm_dev
*dev
= get_imsm_dev(super
, a
->info
.container_member
);
3154 int i
= get_imsm_disk_idx(dev
, idx
);
3157 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
3161 if (dl
&& dl
->disk
.status
& FAILED_DISK
)
3165 dprintf("%s: found %x:%x\n", __func__
, dl
->major
, dl
->minor
);
3170 static struct dl
*imsm_add_spare(struct intel_super
*super
, int slot
, struct active_array
*a
)
3172 struct imsm_dev
*dev
= get_imsm_dev(super
, a
->info
.container_member
);
3173 int idx
= get_imsm_disk_idx(dev
, slot
);
3174 struct imsm_map
*map
= get_imsm_map(dev
, 0);
3175 unsigned long long esize
;
3176 unsigned long long pos
;
3184 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
3185 /* If in this array, skip */
3186 for (d
= a
->info
.devs
; d
; d
= d
->next
)
3187 if (d
->state_fd
>= 0 &&
3188 d
->disk
.major
== dl
->major
&&
3189 d
->disk
.minor
== dl
->minor
) {
3190 dprintf("%x:%x already in array\n", dl
->major
, dl
->minor
);
3196 /* skip in use or failed drives */
3197 if (dl
->disk
.status
& FAILED_DISK
|| idx
== dl
->index
) {
3198 dprintf("%x:%x status ( %s%s)\n",
3199 dl
->major
, dl
->minor
,
3200 dl
->disk
.status
& FAILED_DISK
? "failed " : "",
3201 idx
== dl
->index
? "in use " : "");
3205 /* Does this unused device have the requisite free space?
3206 * We need a->info.component_size sectors
3208 ex
= get_extents(super
, dl
);
3210 dprintf("cannot get extents\n");
3216 array_start
= __le32_to_cpu(map
->pba_of_lba0
);
3219 /* check that we can start at pba_of_lba0 with
3220 * a->info.component_size of space
3222 esize
= ex
[j
].start
- pos
;
3223 if (array_start
>= pos
&&
3224 array_start
+ a
->info
.component_size
< ex
[j
].start
) {
3228 pos
= ex
[j
].start
+ ex
[j
].size
;
3231 } while (ex
[j
-1].size
);
3235 dprintf("%x:%x does not have %llu at %d\n",
3236 dl
->major
, dl
->minor
,
3237 a
->info
.component_size
,
3238 __le32_to_cpu(map
->pba_of_lba0
));
3248 static struct mdinfo
*imsm_activate_spare(struct active_array
*a
,
3249 struct metadata_update
**updates
)
3252 * Find a device with unused free space and use it to replace a
3253 * failed/vacant region in an array. We replace failed regions one a
3254 * array at a time. The result is that a new spare disk will be added
3255 * to the first failed array and after the monitor has finished
3256 * propagating failures the remainder will be consumed.
3258 * FIXME add a capability for mdmon to request spares from another
3262 struct intel_super
*super
= a
->container
->sb
;
3263 int inst
= a
->info
.container_member
;
3264 struct imsm_dev
*dev
= get_imsm_dev(super
, inst
);
3265 struct imsm_map
*map
= get_imsm_map(dev
, 0);
3266 int failed
= a
->info
.array
.raid_disks
;
3267 struct mdinfo
*rv
= NULL
;
3270 struct metadata_update
*mu
;
3272 struct imsm_update_activate_spare
*u
;
3276 for (d
= a
->info
.devs
; d
; d
= d
->next
) {
3277 if ((d
->curr_state
& DS_FAULTY
) &&
3279 /* wait for Removal to happen */
3281 if (d
->state_fd
>= 0)
3285 dprintf("imsm: activate spare: inst=%d failed=%d (%d) level=%d\n",
3286 inst
, failed
, a
->info
.array
.raid_disks
, a
->info
.array
.level
);
3287 if (imsm_check_degraded(super
, dev
, failed
) != IMSM_T_STATE_DEGRADED
)
3290 /* For each slot, if it is not working, find a spare */
3291 for (i
= 0; i
< a
->info
.array
.raid_disks
; i
++) {
3292 for (d
= a
->info
.devs
; d
; d
= d
->next
)
3293 if (d
->disk
.raid_disk
== i
)
3295 dprintf("found %d: %p %x\n", i
, d
, d
?d
->curr_state
:0);
3296 if (d
&& (d
->state_fd
>= 0))
3300 * OK, this device needs recovery. Try to re-add the previous
3301 * occupant of this slot, if this fails add a new spare
3303 dl
= imsm_readd(super
, i
, a
);
3305 dl
= imsm_add_spare(super
, i
, a
);
3309 /* found a usable disk with enough space */
3310 di
= malloc(sizeof(*di
));
3313 memset(di
, 0, sizeof(*di
));
3315 /* dl->index will be -1 in the case we are activating a
3316 * pristine spare. imsm_process_update() will create a
3317 * new index in this case. Once a disk is found to be
3318 * failed in all member arrays it is kicked from the
3321 di
->disk
.number
= dl
->index
;
3323 /* (ab)use di->devs to store a pointer to the device
3326 di
->devs
= (struct mdinfo
*) dl
;
3328 di
->disk
.raid_disk
= i
;
3329 di
->disk
.major
= dl
->major
;
3330 di
->disk
.minor
= dl
->minor
;
3332 di
->data_offset
= __le32_to_cpu(map
->pba_of_lba0
);
3333 di
->component_size
= a
->info
.component_size
;
3334 di
->container_member
= inst
;
3338 dprintf("%x:%x to be %d at %llu\n", dl
->major
, dl
->minor
,
3339 i
, di
->data_offset
);
3345 /* No spares found */
3347 /* Now 'rv' has a list of devices to return.
3348 * Create a metadata_update record to update the
3349 * disk_ord_tbl for the array
3351 mu
= malloc(sizeof(*mu
));
3353 mu
->buf
= malloc(sizeof(struct imsm_update_activate_spare
) * num_spares
);
3354 if (mu
->buf
== NULL
) {
3361 struct mdinfo
*n
= rv
->next
;
3370 mu
->len
= sizeof(struct imsm_update_activate_spare
) * num_spares
;
3371 mu
->next
= *updates
;
3372 u
= (struct imsm_update_activate_spare
*) mu
->buf
;
3374 for (di
= rv
; di
; di
= di
->next
) {
3375 u
->type
= update_activate_spare
;
3376 u
->dl
= (struct dl
*) di
->devs
;
3378 u
->slot
= di
->disk
.raid_disk
;
3389 static int disks_overlap(struct intel_super
*super
, int idx
, struct imsm_update_create_array
*u
)
3391 struct imsm_dev
*dev
= get_imsm_dev(super
, idx
);
3392 struct imsm_map
*map
= get_imsm_map(dev
, 0);
3393 struct imsm_map
*new_map
= get_imsm_map(&u
->dev
, 0);
3394 struct disk_info
*inf
= get_disk_info(u
);
3395 struct imsm_disk
*disk
;
3399 for (i
= 0; i
< map
->num_members
; i
++) {
3400 disk
= get_imsm_disk(super
, get_imsm_disk_idx(dev
, i
));
3401 for (j
= 0; j
< new_map
->num_members
; j
++)
3402 if (serialcmp(disk
->serial
, inf
[j
].serial
) == 0)
3409 static void imsm_delete(struct intel_super
*super
, struct dl
**dlp
, int index
);
3411 static void imsm_process_update(struct supertype
*st
,
3412 struct metadata_update
*update
)
3415 * crack open the metadata_update envelope to find the update record
3416 * update can be one of:
3417 * update_activate_spare - a spare device has replaced a failed
3418 * device in an array, update the disk_ord_tbl. If this disk is
3419 * present in all member arrays then also clear the SPARE_DISK
3422 struct intel_super
*super
= st
->sb
;
3423 struct imsm_super
*mpb
;
3424 enum imsm_update_type type
= *(enum imsm_update_type
*) update
->buf
;
3426 /* update requires a larger buf but the allocation failed */
3427 if (super
->next_len
&& !super
->next_buf
) {
3428 super
->next_len
= 0;
3432 if (super
->next_buf
) {
3433 memcpy(super
->next_buf
, super
->buf
, super
->len
);
3435 super
->len
= super
->next_len
;
3436 super
->buf
= super
->next_buf
;
3438 super
->next_len
= 0;
3439 super
->next_buf
= NULL
;
3442 mpb
= super
->anchor
;
3445 case update_activate_spare
: {
3446 struct imsm_update_activate_spare
*u
= (void *) update
->buf
;
3447 struct imsm_dev
*dev
= get_imsm_dev(super
, u
->array
);
3448 struct imsm_map
*map
= get_imsm_map(dev
, 0);
3449 struct imsm_map
*migr_map
;
3450 struct active_array
*a
;
3451 struct imsm_disk
*disk
;
3456 int victim
= get_imsm_disk_idx(dev
, u
->slot
);
3459 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
3464 fprintf(stderr
, "error: imsm_activate_spare passed "
3465 "an unknown disk (index: %d)\n",
3470 super
->updates_pending
++;
3472 /* count failures (excluding rebuilds and the victim)
3473 * to determine map[0] state
3476 for (i
= 0; i
< map
->num_members
; i
++) {
3479 disk
= get_imsm_disk(super
, get_imsm_disk_idx(dev
, i
));
3480 if (!disk
|| disk
->status
& FAILED_DISK
)
3484 /* adding a pristine spare, assign a new index */
3485 if (dl
->index
< 0) {
3486 dl
->index
= super
->anchor
->num_disks
;
3487 super
->anchor
->num_disks
++;
3490 disk
->status
|= CONFIGURED_DISK
;
3491 disk
->status
&= ~SPARE_DISK
;
3494 to_state
= imsm_check_degraded(super
, dev
, failed
);
3495 map
->map_state
= IMSM_T_STATE_DEGRADED
;
3496 migrate(dev
, to_state
, MIGR_REBUILD
);
3497 migr_map
= get_imsm_map(dev
, 1);
3498 set_imsm_ord_tbl_ent(map
, u
->slot
, dl
->index
);
3499 set_imsm_ord_tbl_ent(migr_map
, u
->slot
, dl
->index
| IMSM_ORD_REBUILD
);
3501 /* count arrays using the victim in the metadata */
3503 for (a
= st
->arrays
; a
; a
= a
->next
) {
3504 dev
= get_imsm_dev(super
, a
->info
.container_member
);
3505 for (i
= 0; i
< map
->num_members
; i
++)
3506 if (victim
== get_imsm_disk_idx(dev
, i
))
3510 /* delete the victim if it is no longer being
3516 /* We know that 'manager' isn't touching anything,
3517 * so it is safe to delete
3519 for (dlp
= &super
->disks
; *dlp
; dlp
= &(*dlp
)->next
)
3520 if ((*dlp
)->index
== victim
)
3523 /* victim may be on the missing list */
3525 for (dlp
= &super
->missing
; *dlp
; dlp
= &(*dlp
)->next
)
3526 if ((*dlp
)->index
== victim
)
3528 imsm_delete(super
, dlp
, victim
);
3532 case update_create_array
: {
3533 /* someone wants to create a new array, we need to be aware of
3534 * a few races/collisions:
3535 * 1/ 'Create' called by two separate instances of mdadm
3536 * 2/ 'Create' versus 'activate_spare': mdadm has chosen
3537 * devices that have since been assimilated via
3539 * In the event this update can not be carried out mdadm will
3540 * (FIX ME) notice that its update did not take hold.
3542 struct imsm_update_create_array
*u
= (void *) update
->buf
;
3543 struct imsm_dev
*dev
;
3544 struct imsm_map
*map
, *new_map
;
3545 unsigned long long start
, end
;
3546 unsigned long long new_start
, new_end
;
3548 struct disk_info
*inf
;
3551 /* handle racing creates: first come first serve */
3552 if (u
->dev_idx
< mpb
->num_raid_devs
) {
3553 dprintf("%s: subarray %d already defined\n",
3554 __func__
, u
->dev_idx
);
3558 /* check update is next in sequence */
3559 if (u
->dev_idx
!= mpb
->num_raid_devs
) {
3560 dprintf("%s: can not create array %d expected index %d\n",
3561 __func__
, u
->dev_idx
, mpb
->num_raid_devs
);
3565 new_map
= get_imsm_map(&u
->dev
, 0);
3566 new_start
= __le32_to_cpu(new_map
->pba_of_lba0
);
3567 new_end
= new_start
+ __le32_to_cpu(new_map
->blocks_per_member
);
3568 inf
= get_disk_info(u
);
3570 /* handle activate_spare versus create race:
3571 * check to make sure that overlapping arrays do not include
3574 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
3575 dev
= get_imsm_dev(super
, i
);
3576 map
= get_imsm_map(dev
, 0);
3577 start
= __le32_to_cpu(map
->pba_of_lba0
);
3578 end
= start
+ __le32_to_cpu(map
->blocks_per_member
);
3579 if ((new_start
>= start
&& new_start
<= end
) ||
3580 (start
>= new_start
&& start
<= new_end
))
3585 if (disks_overlap(super
, i
, u
)) {
3586 dprintf("%s: arrays overlap\n", __func__
);
3591 /* check that prepare update was successful */
3592 if (!update
->space
) {
3593 dprintf("%s: prepare update failed\n", __func__
);
3597 /* check that all disks are still active before committing
3598 * changes. FIXME: could we instead handle this by creating a
3599 * degraded array? That's probably not what the user expects,
3600 * so better to drop this update on the floor.
3602 for (i
= 0; i
< new_map
->num_members
; i
++) {
3603 dl
= serial_to_dl(inf
[i
].serial
, super
);
3605 dprintf("%s: disk disappeared\n", __func__
);
3610 super
->updates_pending
++;
3612 /* convert spares to members and fixup ord_tbl */
3613 for (i
= 0; i
< new_map
->num_members
; i
++) {
3614 dl
= serial_to_dl(inf
[i
].serial
, super
);
3615 if (dl
->index
== -1) {
3616 dl
->index
= mpb
->num_disks
;
3618 dl
->disk
.status
|= CONFIGURED_DISK
;
3619 dl
->disk
.status
&= ~SPARE_DISK
;
3621 set_imsm_ord_tbl_ent(new_map
, i
, dl
->index
);
3624 dev
= update
->space
;
3625 update
->space
= NULL
;
3626 imsm_copy_dev(dev
, &u
->dev
);
3627 super
->dev_tbl
[u
->dev_idx
] = dev
;
3628 mpb
->num_raid_devs
++;
3630 imsm_update_version_info(super
);
3633 case update_add_disk
:
3635 /* we may be able to repair some arrays if disks are
3638 struct active_array
*a
;
3640 super
->updates_pending
++;
3641 for (a
= st
->arrays
; a
; a
= a
->next
)
3642 a
->check_degraded
= 1;
3644 /* add some spares to the metadata */
3645 while (super
->add
) {
3649 super
->add
= al
->next
;
3650 al
->next
= super
->disks
;
3652 dprintf("%s: added %x:%x\n",
3653 __func__
, al
->major
, al
->minor
);
3660 static void imsm_prepare_update(struct supertype
*st
,
3661 struct metadata_update
*update
)
3664 * Allocate space to hold new disk entries, raid-device entries or a new
3665 * mpb if necessary. The manager synchronously waits for updates to
3666 * complete in the monitor, so new mpb buffers allocated here can be
3667 * integrated by the monitor thread without worrying about live pointers
3668 * in the manager thread.
3670 enum imsm_update_type type
= *(enum imsm_update_type
*) update
->buf
;
3671 struct intel_super
*super
= st
->sb
;
3672 struct imsm_super
*mpb
= super
->anchor
;
3677 case update_create_array
: {
3678 struct imsm_update_create_array
*u
= (void *) update
->buf
;
3679 struct imsm_dev
*dev
= &u
->dev
;
3680 struct imsm_map
*map
= get_imsm_map(dev
, 0);
3682 struct disk_info
*inf
;
3686 inf
= get_disk_info(u
);
3687 len
= sizeof_imsm_dev(dev
, 1);
3688 /* allocate a new super->dev_tbl entry */
3689 update
->space
= malloc(len
);
3691 /* count how many spares will be converted to members */
3692 for (i
= 0; i
< map
->num_members
; i
++) {
3693 dl
= serial_to_dl(inf
[i
].serial
, super
);
3695 /* hmm maybe it failed?, nothing we can do about
3700 if (count_memberships(dl
, super
) == 0)
3703 len
+= activate
* sizeof(struct imsm_disk
);
3710 /* check if we need a larger metadata buffer */
3711 if (super
->next_buf
)
3712 buf_len
= super
->next_len
;
3714 buf_len
= super
->len
;
3716 if (__le32_to_cpu(mpb
->mpb_size
) + len
> buf_len
) {
3717 /* ok we need a larger buf than what is currently allocated
3718 * if this allocation fails process_update will notice that
3719 * ->next_len is set and ->next_buf is NULL
3721 buf_len
= ROUND_UP(__le32_to_cpu(mpb
->mpb_size
) + len
, 512);
3722 if (super
->next_buf
)
3723 free(super
->next_buf
);
3725 super
->next_len
= buf_len
;
3726 if (posix_memalign(&super
->next_buf
, buf_len
, 512) != 0)
3727 super
->next_buf
= NULL
;
3731 /* must be called while manager is quiesced */
3732 static void imsm_delete(struct intel_super
*super
, struct dl
**dlp
, int index
)
3734 struct imsm_super
*mpb
= super
->anchor
;
3736 struct imsm_dev
*dev
;
3737 struct imsm_map
*map
;
3738 int i
, j
, num_members
;
3741 dprintf("%s: deleting device[%d] from imsm_super\n",
3744 /* shift all indexes down one */
3745 for (iter
= super
->disks
; iter
; iter
= iter
->next
)
3746 if (iter
->index
> index
)
3748 for (iter
= super
->missing
; iter
; iter
= iter
->next
)
3749 if (iter
->index
> index
)
3752 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
3753 dev
= get_imsm_dev(super
, i
);
3754 map
= get_imsm_map(dev
, 0);
3755 num_members
= map
->num_members
;
3756 for (j
= 0; j
< num_members
; j
++) {
3757 /* update ord entries being careful not to propagate
3758 * ord-flags to the first map
3760 ord
= get_imsm_ord_tbl_ent(dev
, j
);
3762 if (ord_to_idx(ord
) <= index
)
3765 map
= get_imsm_map(dev
, 0);
3766 set_imsm_ord_tbl_ent(map
, j
, ord_to_idx(ord
- 1));
3767 map
= get_imsm_map(dev
, 1);
3769 set_imsm_ord_tbl_ent(map
, j
, ord
- 1);
3774 super
->updates_pending
++;
3776 struct dl
*dl
= *dlp
;
3778 *dlp
= (*dlp
)->next
;
3779 __free_imsm_disk(dl
);
3782 #endif /* MDASSEMBLE */
3784 struct superswitch super_imsm
= {
3786 .examine_super
= examine_super_imsm
,
3787 .brief_examine_super
= brief_examine_super_imsm
,
3788 .detail_super
= detail_super_imsm
,
3789 .brief_detail_super
= brief_detail_super_imsm
,
3790 .write_init_super
= write_init_super_imsm
,
3791 .validate_geometry
= validate_geometry_imsm
,
3792 .add_to_super
= add_to_super_imsm
,
3794 .match_home
= match_home_imsm
,
3795 .uuid_from_super
= uuid_from_super_imsm
,
3796 .getinfo_super
= getinfo_super_imsm
,
3797 .update_super
= update_super_imsm
,
3799 .avail_size
= avail_size_imsm
,
3801 .compare_super
= compare_super_imsm
,
3803 .load_super
= load_super_imsm
,
3804 .init_super
= init_super_imsm
,
3805 .store_super
= store_zero_imsm
,
3806 .free_super
= free_super_imsm
,
3807 .match_metadata_desc
= match_metadata_desc_imsm
,
3808 .container_content
= container_content_imsm
,
3814 .open_new
= imsm_open_new
,
3815 .load_super
= load_super_imsm
,
3816 .set_array_state
= imsm_set_array_state
,
3817 .set_disk
= imsm_set_disk
,
3818 .sync_metadata
= imsm_sync_metadata
,
3819 .activate_spare
= imsm_activate_spare
,
3820 .process_update
= imsm_process_update
,
3821 .prepare_update
= imsm_prepare_update
,
3822 #endif /* MDASSEMBLE */