2 * mdadm - Intel(R) Matrix Storage Manager Support
4 * Copyright (C) 2002-2008 Intel Corporation
6 * This program is free software; you can redistribute it and/or modify it
7 * under the terms and conditions of the GNU General Public License,
8 * version 2, as published by the Free Software Foundation.
10 * This program is distributed in the hope it will be useful, but WITHOUT
11 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
12 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for
15 * You should have received a copy of the GNU General Public License along with
16 * this program; if not, write to the Free Software Foundation, Inc.,
17 * 51 Franklin St - Fifth Floor, Boston, MA 02110-1301 USA.
20 #define HAVE_STDINT_H 1
24 #include "platform-intel.h"
30 /* MPB == Metadata Parameter Block */
31 #define MPB_SIGNATURE "Intel Raid ISM Cfg Sig. "
32 #define MPB_SIG_LEN (strlen(MPB_SIGNATURE))
33 #define MPB_VERSION_RAID0 "1.0.00"
34 #define MPB_VERSION_RAID1 "1.1.00"
35 #define MPB_VERSION_MANY_VOLUMES_PER_ARRAY "1.2.00"
36 #define MPB_VERSION_3OR4_DISK_ARRAY "1.2.01"
37 #define MPB_VERSION_RAID5 "1.2.02"
38 #define MPB_VERSION_5OR6_DISK_ARRAY "1.2.04"
39 #define MPB_VERSION_CNG "1.2.06"
40 #define MPB_VERSION_ATTRIBS "1.3.00"
41 #define MAX_SIGNATURE_LENGTH 32
42 #define MAX_RAID_SERIAL_LEN 16
44 #define MPB_ATTRIB_CHECKSUM_VERIFY __cpu_to_le32(0x80000000)
45 #define MPB_ATTRIB_PM __cpu_to_le32(0x40000000)
46 #define MPB_ATTRIB_2TB __cpu_to_le32(0x20000000)
47 #define MPB_ATTRIB_RAID0 __cpu_to_le32(0x00000001)
48 #define MPB_ATTRIB_RAID1 __cpu_to_le32(0x00000002)
49 #define MPB_ATTRIB_RAID10 __cpu_to_le32(0x00000004)
50 #define MPB_ATTRIB_RAID1E __cpu_to_le32(0x00000008)
51 #define MPB_ATTRIB_RAID5 __cpu_to_le32(0x00000010)
52 #define MPB_ATTRIB_RAIDCNG __cpu_to_le32(0x00000020)
54 #define MPB_SECTOR_CNT 418
55 #define IMSM_RESERVED_SECTORS 4096
57 /* Disk configuration info. */
58 #define IMSM_MAX_DEVICES 255
60 __u8 serial
[MAX_RAID_SERIAL_LEN
];/* 0xD8 - 0xE7 ascii serial number */
61 __u32 total_blocks
; /* 0xE8 - 0xEB total blocks */
62 __u32 scsi_id
; /* 0xEC - 0xEF scsi ID */
63 #define SPARE_DISK __cpu_to_le32(0x01) /* Spare */
64 #define CONFIGURED_DISK __cpu_to_le32(0x02) /* Member of some RaidDev */
65 #define FAILED_DISK __cpu_to_le32(0x04) /* Permanent failure */
66 #define USABLE_DISK __cpu_to_le32(0x08) /* Fully usable unless FAILED_DISK is set */
67 __u32 status
; /* 0xF0 - 0xF3 */
68 __u32 owner_cfg_num
; /* which config 0,1,2... owns this disk */
69 #define IMSM_DISK_FILLERS 4
70 __u32 filler
[IMSM_DISK_FILLERS
]; /* 0xF4 - 0x107 MPB_DISK_FILLERS for future expansion */
73 /* RAID map configuration infos. */
75 __u32 pba_of_lba0
; /* start address of partition */
76 __u32 blocks_per_member
;/* blocks per member */
77 __u32 num_data_stripes
; /* number of data stripes */
78 __u16 blocks_per_strip
;
79 __u8 map_state
; /* Normal, Uninitialized, Degraded, Failed */
80 #define IMSM_T_STATE_NORMAL 0
81 #define IMSM_T_STATE_UNINITIALIZED 1
82 #define IMSM_T_STATE_DEGRADED 2
83 #define IMSM_T_STATE_FAILED 3
85 #define IMSM_T_RAID0 0
86 #define IMSM_T_RAID1 1
87 #define IMSM_T_RAID5 5 /* since metadata version 1.2.02 ? */
88 __u8 num_members
; /* number of member disks */
89 __u8 num_domains
; /* number of parity domains */
90 __u8 failed_disk_num
; /* valid only when state is degraded */
92 __u32 filler
[7]; /* expansion area */
93 #define IMSM_ORD_REBUILD (1 << 24)
94 __u32 disk_ord_tbl
[1]; /* disk_ord_tbl[num_members],
95 * top byte contains some flags
97 } __attribute__ ((packed
));
100 __u32 curr_migr_unit
;
101 __u32 checkpoint_id
; /* id to access curr_migr_unit */
102 __u8 migr_state
; /* Normal or Migrating */
104 #define MIGR_REBUILD 1
105 #define MIGR_VERIFY 2 /* analagous to echo check > sync_action */
106 #define MIGR_GEN_MIGR 3
107 #define MIGR_STATE_CHANGE 4
108 __u8 migr_type
; /* Initializing, Rebuilding, ... */
110 __u8 fs_state
; /* fast-sync state for CnG (0xff == disabled) */
111 __u16 verify_errors
; /* number of mismatches */
112 __u16 bad_blocks
; /* number of bad blocks during verify */
114 struct imsm_map map
[1];
115 /* here comes another one if migr_state */
116 } __attribute__ ((packed
));
119 __u8 volume
[MAX_RAID_SERIAL_LEN
];
122 #define DEV_BOOTABLE __cpu_to_le32(0x01)
123 #define DEV_BOOT_DEVICE __cpu_to_le32(0x02)
124 #define DEV_READ_COALESCING __cpu_to_le32(0x04)
125 #define DEV_WRITE_COALESCING __cpu_to_le32(0x08)
126 #define DEV_LAST_SHUTDOWN_DIRTY __cpu_to_le32(0x10)
127 #define DEV_HIDDEN_AT_BOOT __cpu_to_le32(0x20)
128 #define DEV_CURRENTLY_HIDDEN __cpu_to_le32(0x40)
129 #define DEV_VERIFY_AND_FIX __cpu_to_le32(0x80)
130 #define DEV_MAP_STATE_UNINIT __cpu_to_le32(0x100)
131 #define DEV_NO_AUTO_RECOVERY __cpu_to_le32(0x200)
132 #define DEV_CLONE_N_GO __cpu_to_le32(0x400)
133 #define DEV_CLONE_MAN_SYNC __cpu_to_le32(0x800)
134 #define DEV_CNG_MASTER_DISK_NUM __cpu_to_le32(0x1000)
135 __u32 status
; /* Persistent RaidDev status */
136 __u32 reserved_blocks
; /* Reserved blocks at beginning of volume */
140 __u8 cng_master_disk
;
144 #define IMSM_DEV_FILLERS 10
145 __u32 filler
[IMSM_DEV_FILLERS
];
147 } __attribute__ ((packed
));
150 __u8 sig
[MAX_SIGNATURE_LENGTH
]; /* 0x00 - 0x1F */
151 __u32 check_sum
; /* 0x20 - 0x23 MPB Checksum */
152 __u32 mpb_size
; /* 0x24 - 0x27 Size of MPB */
153 __u32 family_num
; /* 0x28 - 0x2B Checksum from first time this config was written */
154 __u32 generation_num
; /* 0x2C - 0x2F Incremented each time this array's MPB is written */
155 __u32 error_log_size
; /* 0x30 - 0x33 in bytes */
156 __u32 attributes
; /* 0x34 - 0x37 */
157 __u8 num_disks
; /* 0x38 Number of configured disks */
158 __u8 num_raid_devs
; /* 0x39 Number of configured volumes */
159 __u8 error_log_pos
; /* 0x3A */
160 __u8 fill
[1]; /* 0x3B */
161 __u32 cache_size
; /* 0x3c - 0x40 in mb */
162 __u32 orig_family_num
; /* 0x40 - 0x43 original family num */
163 __u32 pwr_cycle_count
; /* 0x44 - 0x47 simulated power cycle count for array */
164 __u32 bbm_log_size
; /* 0x48 - 0x4B - size of bad Block Mgmt Log in bytes */
165 #define IMSM_FILLERS 35
166 __u32 filler
[IMSM_FILLERS
]; /* 0x4C - 0xD7 RAID_MPB_FILLERS */
167 struct imsm_disk disk
[1]; /* 0xD8 diskTbl[numDisks] */
168 /* here comes imsm_dev[num_raid_devs] */
169 /* here comes BBM logs */
170 } __attribute__ ((packed
));
172 #define BBM_LOG_MAX_ENTRIES 254
174 struct bbm_log_entry
{
175 __u64 defective_block_start
;
176 #define UNREADABLE 0xFFFFFFFF
177 __u32 spare_block_offset
;
178 __u16 remapped_marked_count
;
180 } __attribute__ ((__packed__
));
183 __u32 signature
; /* 0xABADB10C */
185 __u32 reserved_spare_block_count
; /* 0 */
186 __u32 reserved
; /* 0xFFFF */
187 __u64 first_spare_lba
;
188 struct bbm_log_entry mapped_block_entries
[BBM_LOG_MAX_ENTRIES
];
189 } __attribute__ ((__packed__
));
193 static char *map_state_str
[] = { "normal", "uninitialized", "degraded", "failed" };
196 static unsigned int sector_count(__u32 bytes
)
198 return ((bytes
+ (512-1)) & (~(512-1))) / 512;
201 static unsigned int mpb_sectors(struct imsm_super
*mpb
)
203 return sector_count(__le32_to_cpu(mpb
->mpb_size
));
207 struct imsm_dev
*dev
;
208 struct intel_dev
*next
;
212 /* internal representation of IMSM metadata */
215 void *buf
; /* O_DIRECT buffer for reading/writing metadata */
216 struct imsm_super
*anchor
; /* immovable parameters */
218 size_t len
; /* size of the 'buf' allocation */
219 void *next_buf
; /* for realloc'ing buf from the manager */
221 int updates_pending
; /* count of pending updates for mdmon */
222 int creating_imsm
; /* flag to indicate container creation */
223 int current_vol
; /* index of raid device undergoing creation */
224 __u32 create_offset
; /* common start for 'current_vol' */
225 struct intel_dev
*devlist
;
229 __u8 serial
[MAX_RAID_SERIAL_LEN
];
232 struct imsm_disk disk
;
235 struct extent
*e
; /* for determining freespace @ create */
237 struct dl
*add
; /* list of disks to add while mdmon active */
238 struct dl
*missing
; /* disks removed while we weren't looking */
239 struct bbm_log
*bbm_log
;
240 const char *hba
; /* device path of the raid controller for this metadata */
241 const struct imsm_orom
*orom
; /* platform firmware support */
245 unsigned long long start
, size
;
248 /* definition of messages passed to imsm_process_update */
249 enum imsm_update_type
{
250 update_activate_spare
,
255 struct imsm_update_activate_spare
{
256 enum imsm_update_type type
;
260 struct imsm_update_activate_spare
*next
;
264 __u8 serial
[MAX_RAID_SERIAL_LEN
];
267 struct imsm_update_create_array
{
268 enum imsm_update_type type
;
273 struct imsm_update_add_disk
{
274 enum imsm_update_type type
;
277 static struct supertype
*match_metadata_desc_imsm(char *arg
)
279 struct supertype
*st
;
281 if (strcmp(arg
, "imsm") != 0 &&
282 strcmp(arg
, "default") != 0
286 st
= malloc(sizeof(*st
));
287 memset(st
, 0, sizeof(*st
));
288 st
->ss
= &super_imsm
;
289 st
->max_devs
= IMSM_MAX_DEVICES
;
290 st
->minor_version
= 0;
296 static __u8
*get_imsm_version(struct imsm_super
*mpb
)
298 return &mpb
->sig
[MPB_SIG_LEN
];
302 /* retrieve a disk directly from the anchor when the anchor is known to be
303 * up-to-date, currently only at load time
305 static struct imsm_disk
*__get_imsm_disk(struct imsm_super
*mpb
, __u8 index
)
307 if (index
>= mpb
->num_disks
)
309 return &mpb
->disk
[index
];
313 /* retrieve a disk from the parsed metadata */
314 static struct imsm_disk
*get_imsm_disk(struct intel_super
*super
, __u8 index
)
318 for (d
= super
->disks
; d
; d
= d
->next
)
319 if (d
->index
== index
)
326 /* generate a checksum directly from the anchor when the anchor is known to be
327 * up-to-date, currently only at load or write_super after coalescing
329 static __u32
__gen_imsm_checksum(struct imsm_super
*mpb
)
331 __u32 end
= mpb
->mpb_size
/ sizeof(end
);
332 __u32
*p
= (__u32
*) mpb
;
336 sum
+= __le32_to_cpu(*p
);
340 return sum
- __le32_to_cpu(mpb
->check_sum
);
343 static size_t sizeof_imsm_map(struct imsm_map
*map
)
345 return sizeof(struct imsm_map
) + sizeof(__u32
) * (map
->num_members
- 1);
348 struct imsm_map
*get_imsm_map(struct imsm_dev
*dev
, int second_map
)
350 struct imsm_map
*map
= &dev
->vol
.map
[0];
352 if (second_map
&& !dev
->vol
.migr_state
)
354 else if (second_map
) {
357 return ptr
+ sizeof_imsm_map(map
);
363 /* return the size of the device.
364 * migr_state increases the returned size if map[0] were to be duplicated
366 static size_t sizeof_imsm_dev(struct imsm_dev
*dev
, int migr_state
)
368 size_t size
= sizeof(*dev
) - sizeof(struct imsm_map
) +
369 sizeof_imsm_map(get_imsm_map(dev
, 0));
371 /* migrating means an additional map */
372 if (dev
->vol
.migr_state
)
373 size
+= sizeof_imsm_map(get_imsm_map(dev
, 1));
375 size
+= sizeof_imsm_map(get_imsm_map(dev
, 0));
381 /* retrieve disk serial number list from a metadata update */
382 static struct disk_info
*get_disk_info(struct imsm_update_create_array
*update
)
385 struct disk_info
*inf
;
387 inf
= u
+ sizeof(*update
) - sizeof(struct imsm_dev
) +
388 sizeof_imsm_dev(&update
->dev
, 0);
394 static struct imsm_dev
*__get_imsm_dev(struct imsm_super
*mpb
, __u8 index
)
400 if (index
>= mpb
->num_raid_devs
)
403 /* devices start after all disks */
404 offset
= ((void *) &mpb
->disk
[mpb
->num_disks
]) - _mpb
;
406 for (i
= 0; i
<= index
; i
++)
408 return _mpb
+ offset
;
410 offset
+= sizeof_imsm_dev(_mpb
+ offset
, 0);
415 static struct imsm_dev
*get_imsm_dev(struct intel_super
*super
, __u8 index
)
417 struct intel_dev
*dv
;
419 if (index
>= super
->anchor
->num_raid_devs
)
421 for (dv
= super
->devlist
; dv
; dv
= dv
->next
)
422 if (dv
->index
== index
)
427 static __u32
get_imsm_ord_tbl_ent(struct imsm_dev
*dev
, int slot
)
429 struct imsm_map
*map
;
431 if (dev
->vol
.migr_state
)
432 map
= get_imsm_map(dev
, 1);
434 map
= get_imsm_map(dev
, 0);
436 /* top byte identifies disk under rebuild */
437 return __le32_to_cpu(map
->disk_ord_tbl
[slot
]);
440 #define ord_to_idx(ord) (((ord) << 8) >> 8)
441 static __u32
get_imsm_disk_idx(struct imsm_dev
*dev
, int slot
)
443 __u32 ord
= get_imsm_ord_tbl_ent(dev
, slot
);
445 return ord_to_idx(ord
);
448 static void set_imsm_ord_tbl_ent(struct imsm_map
*map
, int slot
, __u32 ord
)
450 map
->disk_ord_tbl
[slot
] = __cpu_to_le32(ord
);
453 static int get_imsm_disk_slot(struct imsm_map
*map
, int idx
)
458 for (slot
= 0; slot
< map
->num_members
; slot
++) {
459 ord
= __le32_to_cpu(map
->disk_ord_tbl
[slot
]);
460 if (ord_to_idx(ord
) == idx
)
467 static int get_imsm_raid_level(struct imsm_map
*map
)
469 if (map
->raid_level
== 1) {
470 if (map
->num_members
== 2)
476 return map
->raid_level
;
479 static int cmp_extent(const void *av
, const void *bv
)
481 const struct extent
*a
= av
;
482 const struct extent
*b
= bv
;
483 if (a
->start
< b
->start
)
485 if (a
->start
> b
->start
)
490 static int count_memberships(struct dl
*dl
, struct intel_super
*super
)
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 if (get_imsm_disk_slot(map
, dl
->index
) >= 0)
506 static struct extent
*get_extents(struct intel_super
*super
, struct dl
*dl
)
508 /* find a list of used extents on the given physical device */
509 struct extent
*rv
, *e
;
511 int memberships
= count_memberships(dl
, super
);
512 __u32 reservation
= MPB_SECTOR_CNT
+ IMSM_RESERVED_SECTORS
;
514 rv
= malloc(sizeof(struct extent
) * (memberships
+ 1));
519 for (i
= 0; i
< super
->anchor
->num_raid_devs
; i
++) {
520 struct imsm_dev
*dev
= get_imsm_dev(super
, i
);
521 struct imsm_map
*map
= get_imsm_map(dev
, 0);
523 if (get_imsm_disk_slot(map
, dl
->index
) >= 0) {
524 e
->start
= __le32_to_cpu(map
->pba_of_lba0
);
525 e
->size
= __le32_to_cpu(map
->blocks_per_member
);
529 qsort(rv
, memberships
, sizeof(*rv
), cmp_extent
);
531 /* determine the start of the metadata
532 * when no raid devices are defined use the default
533 * ...otherwise allow the metadata to truncate the value
534 * as is the case with older versions of imsm
537 struct extent
*last
= &rv
[memberships
- 1];
540 remainder
= __le32_to_cpu(dl
->disk
.total_blocks
) -
541 (last
->start
+ last
->size
);
542 /* round down to 1k block to satisfy precision of the kernel
546 /* make sure remainder is still sane */
547 if (remainder
< ROUND_UP(super
->len
, 512) >> 9)
548 remainder
= ROUND_UP(super
->len
, 512) >> 9;
549 if (reservation
> remainder
)
550 reservation
= remainder
;
552 e
->start
= __le32_to_cpu(dl
->disk
.total_blocks
) - reservation
;
557 /* try to determine how much space is reserved for metadata from
558 * the last get_extents() entry, otherwise fallback to the
561 static __u32
imsm_reserved_sectors(struct intel_super
*super
, struct dl
*dl
)
567 /* for spares just return a minimal reservation which will grow
568 * once the spare is picked up by an array
571 return MPB_SECTOR_CNT
;
573 e
= get_extents(super
, dl
);
575 return MPB_SECTOR_CNT
+ IMSM_RESERVED_SECTORS
;
577 /* scroll to last entry */
578 for (i
= 0; e
[i
].size
; i
++)
581 rv
= __le32_to_cpu(dl
->disk
.total_blocks
) - e
[i
].start
;
589 static void print_imsm_dev(struct imsm_dev
*dev
, char *uuid
, int disk_idx
)
593 struct imsm_map
*map
= get_imsm_map(dev
, 0);
597 printf("[%.16s]:\n", dev
->volume
);
598 printf(" UUID : %s\n", uuid
);
599 printf(" RAID Level : %d\n", get_imsm_raid_level(map
));
600 printf(" Members : %d\n", map
->num_members
);
601 slot
= get_imsm_disk_slot(map
, disk_idx
);
603 ord
= get_imsm_ord_tbl_ent(dev
, slot
);
604 printf(" This Slot : %d%s\n", slot
,
605 ord
& IMSM_ORD_REBUILD
? " (out-of-sync)" : "");
607 printf(" This Slot : ?\n");
608 sz
= __le32_to_cpu(dev
->size_high
);
610 sz
+= __le32_to_cpu(dev
->size_low
);
611 printf(" Array Size : %llu%s\n", (unsigned long long)sz
,
612 human_size(sz
* 512));
613 sz
= __le32_to_cpu(map
->blocks_per_member
);
614 printf(" Per Dev Size : %llu%s\n", (unsigned long long)sz
,
615 human_size(sz
* 512));
616 printf(" Sector Offset : %u\n",
617 __le32_to_cpu(map
->pba_of_lba0
));
618 printf(" Num Stripes : %u\n",
619 __le32_to_cpu(map
->num_data_stripes
));
620 printf(" Chunk Size : %u KiB\n",
621 __le16_to_cpu(map
->blocks_per_strip
) / 2);
622 printf(" Reserved : %d\n", __le32_to_cpu(dev
->reserved_blocks
));
623 printf(" Migrate State : %s", dev
->vol
.migr_state
? "migrating" : "idle");
624 if (dev
->vol
.migr_state
)
625 printf(": %s", dev
->vol
.migr_type
? "rebuilding" : "initializing");
627 printf(" Map State : %s", map_state_str
[map
->map_state
]);
628 if (dev
->vol
.migr_state
) {
629 struct imsm_map
*map
= get_imsm_map(dev
, 1);
630 printf(" <-- %s", map_state_str
[map
->map_state
]);
633 printf(" Dirty State : %s\n", dev
->vol
.dirty
? "dirty" : "clean");
636 static void print_imsm_disk(struct imsm_super
*mpb
, int index
, __u32 reserved
)
638 struct imsm_disk
*disk
= __get_imsm_disk(mpb
, index
);
639 char str
[MAX_RAID_SERIAL_LEN
+ 1];
647 snprintf(str
, MAX_RAID_SERIAL_LEN
+ 1, "%s", disk
->serial
);
648 printf(" Disk%02d Serial : %s\n", index
, str
);
650 printf(" State :%s%s%s%s\n", s
&SPARE_DISK
? " spare" : "",
651 s
&CONFIGURED_DISK
? " active" : "",
652 s
&FAILED_DISK
? " failed" : "",
653 s
&USABLE_DISK
? " usable" : "");
654 printf(" Id : %08x\n", __le32_to_cpu(disk
->scsi_id
));
655 sz
= __le32_to_cpu(disk
->total_blocks
) - reserved
;
656 printf(" Usable Size : %llu%s\n", (unsigned long long)sz
,
657 human_size(sz
* 512));
660 static void getinfo_super_imsm(struct supertype
*st
, struct mdinfo
*info
);
662 static void examine_super_imsm(struct supertype
*st
, char *homehost
)
664 struct intel_super
*super
= st
->sb
;
665 struct imsm_super
*mpb
= super
->anchor
;
666 char str
[MAX_SIGNATURE_LENGTH
];
671 __u32 reserved
= imsm_reserved_sectors(super
, super
->disks
);
674 snprintf(str
, MPB_SIG_LEN
, "%s", mpb
->sig
);
675 printf(" Magic : %s\n", str
);
676 snprintf(str
, strlen(MPB_VERSION_RAID0
), "%s", get_imsm_version(mpb
));
677 printf(" Version : %s\n", get_imsm_version(mpb
));
678 printf(" Family : %08x\n", __le32_to_cpu(mpb
->family_num
));
679 printf(" Generation : %08x\n", __le32_to_cpu(mpb
->generation_num
));
680 getinfo_super_imsm(st
, &info
);
681 fname_from_uuid(st
, &info
, nbuf
,'-');
682 printf(" UUID : %s\n", nbuf
+ 5);
683 sum
= __le32_to_cpu(mpb
->check_sum
);
684 printf(" Checksum : %08x %s\n", sum
,
685 __gen_imsm_checksum(mpb
) == sum
? "correct" : "incorrect");
686 printf(" MPB Sectors : %d\n", mpb_sectors(mpb
));
687 printf(" Disks : %d\n", mpb
->num_disks
);
688 printf(" RAID Devices : %d\n", mpb
->num_raid_devs
);
689 print_imsm_disk(mpb
, super
->disks
->index
, reserved
);
690 if (super
->bbm_log
) {
691 struct bbm_log
*log
= super
->bbm_log
;
694 printf("Bad Block Management Log:\n");
695 printf(" Log Size : %d\n", __le32_to_cpu(mpb
->bbm_log_size
));
696 printf(" Signature : %x\n", __le32_to_cpu(log
->signature
));
697 printf(" Entry Count : %d\n", __le32_to_cpu(log
->entry_count
));
698 printf(" Spare Blocks : %d\n", __le32_to_cpu(log
->reserved_spare_block_count
));
699 printf(" First Spare : %llx\n", __le64_to_cpu(log
->first_spare_lba
));
701 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
703 struct imsm_dev
*dev
= __get_imsm_dev(mpb
, i
);
705 super
->current_vol
= i
;
706 getinfo_super_imsm(st
, &info
);
707 fname_from_uuid(st
, &info
, nbuf
, '-');
708 print_imsm_dev(dev
, nbuf
+ 5, super
->disks
->index
);
710 for (i
= 0; i
< mpb
->num_disks
; i
++) {
711 if (i
== super
->disks
->index
)
713 print_imsm_disk(mpb
, i
, reserved
);
717 static void brief_examine_super_imsm(struct supertype
*st
)
719 /* We just write a generic IMSM ARRAY entry */
723 struct intel_super
*super
= st
->sb
;
726 if (!super
->anchor
->num_raid_devs
)
729 getinfo_super_imsm(st
, &info
);
730 fname_from_uuid(st
, &info
, nbuf
,'-');
731 printf("ARRAY metadata=imsm auto=md UUID=%s\n", nbuf
+ 5);
732 for (i
= 0; i
< super
->anchor
->num_raid_devs
; i
++) {
733 struct imsm_dev
*dev
= get_imsm_dev(super
, i
);
735 super
->current_vol
= i
;
736 getinfo_super_imsm(st
, &info
);
737 fname_from_uuid(st
, &info
, nbuf1
,'-');
738 printf("ARRAY /dev/md/%.16s container=%s\n"
739 " member=%d auto=mdp UUID=%s\n",
740 dev
->volume
, nbuf
+ 5, i
, nbuf1
+ 5);
744 static void detail_super_imsm(struct supertype
*st
, char *homehost
)
749 getinfo_super_imsm(st
, &info
);
750 fname_from_uuid(st
, &info
, nbuf
,'-');
751 printf("\n UUID : %s\n", nbuf
+ 5);
754 static void brief_detail_super_imsm(struct supertype
*st
)
758 getinfo_super_imsm(st
, &info
);
759 fname_from_uuid(st
, &info
, nbuf
,'-');
760 printf(" UUID=%s", nbuf
+ 5);
763 static int imsm_read_serial(int fd
, char *devname
, __u8
*serial
);
764 static void fd2devname(int fd
, char *name
);
766 static int imsm_enumerate_ports(const char *hba_path
, int port_count
, int host_base
, int verbose
)
768 /* dump an unsorted list of devices attached to ahci, as well as
769 * non-connected ports
771 int hba_len
= strlen(hba_path
) + 1;
776 unsigned long port_mask
= (1 << port_count
) - 1;
778 if (port_count
> sizeof(port_mask
) * 8) {
780 fprintf(stderr
, Name
": port_count %d out of range\n", port_count
);
784 /* scroll through /sys/dev/block looking for devices attached to
787 dir
= opendir("/sys/dev/block");
788 for (ent
= dir
? readdir(dir
) : NULL
; ent
; ent
= readdir(dir
)) {
799 if (sscanf(ent
->d_name
, "%d:%d", &major
, &minor
) != 2)
801 path
= devt_to_devpath(makedev(major
, minor
));
804 if (!path_attached_to_hba(path
, hba_path
)) {
810 /* retrieve the scsi device type */
811 if (asprintf(&device
, "/sys/dev/block/%d:%d/device/xxxxxxx", major
, minor
) < 0) {
813 fprintf(stderr
, Name
": failed to allocate 'device'\n");
817 sprintf(device
, "/sys/dev/block/%d:%d/device/type", major
, minor
);
818 if (load_sys(device
, buf
) != 0) {
820 fprintf(stderr
, Name
": failed to read device type for %s\n",
826 type
= strtoul(buf
, NULL
, 10);
828 /* if it's not a disk print the vendor and model */
829 if (!(type
== 0 || type
== 7 || type
== 14)) {
832 sprintf(device
, "/sys/dev/block/%d:%d/device/vendor", major
, minor
);
833 if (load_sys(device
, buf
) == 0) {
834 strncpy(vendor
, buf
, sizeof(vendor
));
835 vendor
[sizeof(vendor
) - 1] = '\0';
836 c
= (char *) &vendor
[sizeof(vendor
) - 1];
837 while (isspace(*c
) || *c
== '\0')
841 sprintf(device
, "/sys/dev/block/%d:%d/device/model", major
, minor
);
842 if (load_sys(device
, buf
) == 0) {
843 strncpy(model
, buf
, sizeof(model
));
844 model
[sizeof(model
) - 1] = '\0';
845 c
= (char *) &model
[sizeof(model
) - 1];
846 while (isspace(*c
) || *c
== '\0')
850 if (vendor
[0] && model
[0])
851 sprintf(buf
, "%.64s %.64s", vendor
, model
);
853 switch (type
) { /* numbers from hald/linux/device.c */
854 case 1: sprintf(buf
, "tape"); break;
855 case 2: sprintf(buf
, "printer"); break;
856 case 3: sprintf(buf
, "processor"); break;
858 case 5: sprintf(buf
, "cdrom"); break;
859 case 6: sprintf(buf
, "scanner"); break;
860 case 8: sprintf(buf
, "media_changer"); break;
861 case 9: sprintf(buf
, "comm"); break;
862 case 12: sprintf(buf
, "raid"); break;
863 default: sprintf(buf
, "unknown");
869 /* chop device path to 'host%d' and calculate the port number */
870 c
= strchr(&path
[hba_len
], '/');
872 if (sscanf(&path
[hba_len
], "host%d", &port
) == 1)
876 *c
= '/'; /* repair the full string */
877 fprintf(stderr
, Name
": failed to determine port number for %s\n",
884 /* mark this port as used */
885 port_mask
&= ~(1 << port
);
887 /* print out the device information */
889 printf(" Port%d : - non-disk device (%s) -\n", port
, buf
);
893 fd
= dev_open(ent
->d_name
, O_RDONLY
);
895 printf(" Port%d : - disk info unavailable -\n", port
);
898 printf(" Port%d : %s", port
, buf
);
899 if (imsm_read_serial(fd
, NULL
, (__u8
*) buf
) == 0)
900 printf(" (%s)\n", buf
);
915 for (i
= 0; i
< port_count
; i
++)
916 if (port_mask
& (1 << i
))
917 printf(" Port%d : - no device attached -\n", i
);
923 static int detail_platform_imsm(int verbose
, int enumerate_only
)
925 /* There are two components to imsm platform support, the ahci SATA
926 * controller and the option-rom. To find the SATA controller we
927 * simply look in /sys/bus/pci/drivers/ahci to see if an ahci
928 * controller with the Intel vendor id is present. This approach
929 * allows mdadm to leverage the kernel's ahci detection logic, with the
930 * caveat that if ahci.ko is not loaded mdadm will not be able to
931 * detect platform raid capabilities. The option-rom resides in a
932 * platform "Adapter ROM". We scan for its signature to retrieve the
933 * platform capabilities. If raid support is disabled in the BIOS the
934 * option-rom capability structure will not be available.
936 const struct imsm_orom
*orom
;
937 struct sys_dev
*list
, *hba
;
940 const char *hba_path
;
944 if (enumerate_only
) {
945 if (check_env("IMSM_NO_PLATFORM") || find_imsm_orom())
950 list
= find_driver_devices("pci", "ahci");
951 for (hba
= list
; hba
; hba
= hba
->next
)
952 if (devpath_to_vendor(hba
->path
) == 0x8086)
957 fprintf(stderr
, Name
": unable to find active ahci controller\n");
961 fprintf(stderr
, Name
": found Intel SATA AHCI Controller\n");
962 hba_path
= hba
->path
;
966 orom
= find_imsm_orom();
969 fprintf(stderr
, Name
": imsm option-rom not found\n");
973 printf(" Platform : Intel(R) Matrix Storage Manager\n");
974 printf(" Version : %d.%d.%d.%d\n", orom
->major_ver
, orom
->minor_ver
,
975 orom
->hotfix_ver
, orom
->build
);
976 printf(" RAID Levels :%s%s%s%s%s\n",
977 imsm_orom_has_raid0(orom
) ? " raid0" : "",
978 imsm_orom_has_raid1(orom
) ? " raid1" : "",
979 imsm_orom_has_raid1e(orom
) ? " raid1e" : "",
980 imsm_orom_has_raid10(orom
) ? " raid10" : "",
981 imsm_orom_has_raid5(orom
) ? " raid5" : "");
982 printf(" Max Disks : %d\n", orom
->tds
);
983 printf(" Max Volumes : %d\n", orom
->vpa
);
984 printf(" I/O Controller : %s\n", hba_path
);
986 /* find the smallest scsi host number to determine a port number base */
987 dir
= opendir(hba_path
);
988 for (ent
= dir
? readdir(dir
) : NULL
; ent
; ent
= readdir(dir
)) {
991 if (sscanf(ent
->d_name
, "host%d", &host
) != 1)
995 else if (host
< host_base
)
998 if (host
+ 1 > port_count
+ host_base
)
999 port_count
= host
+ 1 - host_base
;
1005 if (!port_count
|| imsm_enumerate_ports(hba_path
, port_count
,
1006 host_base
, verbose
) != 0) {
1008 fprintf(stderr
, Name
": failed to enumerate ports\n");
1016 static int match_home_imsm(struct supertype
*st
, char *homehost
)
1018 /* the imsm metadata format does not specify any host
1019 * identification information. We return -1 since we can never
1020 * confirm nor deny whether a given array is "meant" for this
1021 * host. We rely on compare_super and the 'family_num' field to
1022 * exclude member disks that do not belong, and we rely on
1023 * mdadm.conf to specify the arrays that should be assembled.
1024 * Auto-assembly may still pick up "foreign" arrays.
1030 static void uuid_from_super_imsm(struct supertype
*st
, int uuid
[4])
1032 /* The uuid returned here is used for:
1033 * uuid to put into bitmap file (Create, Grow)
1034 * uuid for backup header when saving critical section (Grow)
1035 * comparing uuids when re-adding a device into an array
1036 * In these cases the uuid required is that of the data-array,
1037 * not the device-set.
1038 * uuid to recognise same set when adding a missing device back
1039 * to an array. This is a uuid for the device-set.
1041 * For each of these we can make do with a truncated
1042 * or hashed uuid rather than the original, as long as
1044 * In each case the uuid required is that of the data-array,
1045 * not the device-set.
1047 /* imsm does not track uuid's so we synthesis one using sha1 on
1048 * - The signature (Which is constant for all imsm array, but no matter)
1049 * - the family_num of the container
1050 * - the index number of the volume
1051 * - the 'serial' number of the volume.
1052 * Hopefully these are all constant.
1054 struct intel_super
*super
= st
->sb
;
1057 struct sha1_ctx ctx
;
1058 struct imsm_dev
*dev
= NULL
;
1060 sha1_init_ctx(&ctx
);
1061 sha1_process_bytes(super
->anchor
->sig
, MPB_SIG_LEN
, &ctx
);
1062 sha1_process_bytes(&super
->anchor
->family_num
, sizeof(__u32
), &ctx
);
1063 if (super
->current_vol
>= 0)
1064 dev
= get_imsm_dev(super
, super
->current_vol
);
1066 __u32 vol
= super
->current_vol
;
1067 sha1_process_bytes(&vol
, sizeof(vol
), &ctx
);
1068 sha1_process_bytes(dev
->volume
, MAX_RAID_SERIAL_LEN
, &ctx
);
1070 sha1_finish_ctx(&ctx
, buf
);
1071 memcpy(uuid
, buf
, 4*4);
1076 get_imsm_numerical_version(struct imsm_super
*mpb
, int *m
, int *p
)
1078 __u8
*v
= get_imsm_version(mpb
);
1079 __u8
*end
= mpb
->sig
+ MAX_SIGNATURE_LENGTH
;
1080 char major
[] = { 0, 0, 0 };
1081 char minor
[] = { 0 ,0, 0 };
1082 char patch
[] = { 0, 0, 0 };
1083 char *ver_parse
[] = { major
, minor
, patch
};
1087 while (*v
!= '\0' && v
< end
) {
1088 if (*v
!= '.' && j
< 2)
1089 ver_parse
[i
][j
++] = *v
;
1097 *m
= strtol(minor
, NULL
, 0);
1098 *p
= strtol(patch
, NULL
, 0);
1102 static int imsm_level_to_layout(int level
)
1110 return ALGORITHM_LEFT_ASYMMETRIC
;
1117 static void getinfo_super_imsm_volume(struct supertype
*st
, struct mdinfo
*info
)
1119 struct intel_super
*super
= st
->sb
;
1120 struct imsm_dev
*dev
= get_imsm_dev(super
, super
->current_vol
);
1121 struct imsm_map
*map
= get_imsm_map(dev
, 0);
1123 info
->container_member
= super
->current_vol
;
1124 info
->array
.raid_disks
= map
->num_members
;
1125 info
->array
.level
= get_imsm_raid_level(map
);
1126 info
->array
.layout
= imsm_level_to_layout(info
->array
.level
);
1127 info
->array
.md_minor
= -1;
1128 info
->array
.ctime
= 0;
1129 info
->array
.utime
= 0;
1130 info
->array
.chunk_size
= __le16_to_cpu(map
->blocks_per_strip
) << 9;
1131 info
->array
.state
= !dev
->vol
.dirty
;
1133 info
->disk
.major
= 0;
1134 info
->disk
.minor
= 0;
1136 info
->data_offset
= __le32_to_cpu(map
->pba_of_lba0
);
1137 info
->component_size
= __le32_to_cpu(map
->blocks_per_member
);
1138 memset(info
->uuid
, 0, sizeof(info
->uuid
));
1140 if (map
->map_state
== IMSM_T_STATE_UNINITIALIZED
|| dev
->vol
.dirty
)
1141 info
->resync_start
= 0;
1142 else if (dev
->vol
.migr_state
)
1143 info
->resync_start
= __le32_to_cpu(dev
->vol
.curr_migr_unit
);
1145 info
->resync_start
= ~0ULL;
1147 strncpy(info
->name
, (char *) dev
->volume
, MAX_RAID_SERIAL_LEN
);
1148 info
->name
[MAX_RAID_SERIAL_LEN
] = 0;
1150 info
->array
.major_version
= -1;
1151 info
->array
.minor_version
= -2;
1152 sprintf(info
->text_version
, "/%s/%d",
1153 devnum2devname(st
->container_dev
),
1154 info
->container_member
);
1155 info
->safe_mode_delay
= 4000; /* 4 secs like the Matrix driver */
1156 uuid_from_super_imsm(st
, info
->uuid
);
1159 /* check the config file to see if we can return a real uuid for this spare */
1160 static void fixup_container_spare_uuid(struct mdinfo
*inf
)
1162 struct mddev_ident_s
*array_list
;
1164 if (inf
->array
.level
!= LEVEL_CONTAINER
||
1165 memcmp(inf
->uuid
, uuid_match_any
, sizeof(int[4])) != 0)
1168 array_list
= conf_get_ident(NULL
);
1170 for (; array_list
; array_list
= array_list
->next
) {
1171 if (array_list
->uuid_set
) {
1172 struct supertype
*_sst
; /* spare supertype */
1173 struct supertype
*_cst
; /* container supertype */
1175 _cst
= array_list
->st
;
1176 _sst
= _cst
->ss
->match_metadata_desc(inf
->text_version
);
1178 memcpy(inf
->uuid
, array_list
->uuid
, sizeof(int[4]));
1186 static void getinfo_super_imsm(struct supertype
*st
, struct mdinfo
*info
)
1188 struct intel_super
*super
= st
->sb
;
1189 struct imsm_disk
*disk
;
1192 if (super
->current_vol
>= 0) {
1193 getinfo_super_imsm_volume(st
, info
);
1197 /* Set raid_disks to zero so that Assemble will always pull in valid
1200 info
->array
.raid_disks
= 0;
1201 info
->array
.level
= LEVEL_CONTAINER
;
1202 info
->array
.layout
= 0;
1203 info
->array
.md_minor
= -1;
1204 info
->array
.ctime
= 0; /* N/A for imsm */
1205 info
->array
.utime
= 0;
1206 info
->array
.chunk_size
= 0;
1208 info
->disk
.major
= 0;
1209 info
->disk
.minor
= 0;
1210 info
->disk
.raid_disk
= -1;
1211 info
->reshape_active
= 0;
1212 info
->array
.major_version
= -1;
1213 info
->array
.minor_version
= -2;
1214 strcpy(info
->text_version
, "imsm");
1215 info
->safe_mode_delay
= 0;
1216 info
->disk
.number
= -1;
1217 info
->disk
.state
= 0;
1221 __u32 reserved
= imsm_reserved_sectors(super
, super
->disks
);
1223 disk
= &super
->disks
->disk
;
1224 info
->data_offset
= __le32_to_cpu(disk
->total_blocks
) - reserved
;
1225 info
->component_size
= reserved
;
1227 info
->disk
.state
= s
& CONFIGURED_DISK
? (1 << MD_DISK_ACTIVE
) : 0;
1228 /* we don't change info->disk.raid_disk here because
1229 * this state will be finalized in mdmon after we have
1230 * found the 'most fresh' version of the metadata
1232 info
->disk
.state
|= s
& FAILED_DISK
? (1 << MD_DISK_FAULTY
) : 0;
1233 info
->disk
.state
|= s
& SPARE_DISK
? 0 : (1 << MD_DISK_SYNC
);
1236 /* only call uuid_from_super_imsm when this disk is part of a populated container,
1237 * ->compare_super may have updated the 'num_raid_devs' field for spares
1239 if (info
->disk
.state
& (1 << MD_DISK_SYNC
) || super
->anchor
->num_raid_devs
)
1240 uuid_from_super_imsm(st
, info
->uuid
);
1242 memcpy(info
->uuid
, uuid_match_any
, sizeof(int[4]));
1243 fixup_container_spare_uuid(info
);
1247 static int update_super_imsm(struct supertype
*st
, struct mdinfo
*info
,
1248 char *update
, char *devname
, int verbose
,
1249 int uuid_set
, char *homehost
)
1253 /* For 'assemble' and 'force' we need to return non-zero if any
1254 * change was made. For others, the return value is ignored.
1255 * Update options are:
1256 * force-one : This device looks a bit old but needs to be included,
1257 * update age info appropriately.
1258 * assemble: clear any 'faulty' flag to allow this device to
1260 * force-array: Array is degraded but being forced, mark it clean
1261 * if that will be needed to assemble it.
1263 * newdev: not used ????
1264 * grow: Array has gained a new device - this is currently for
1266 * resync: mark as dirty so a resync will happen.
1267 * name: update the name - preserving the homehost
1269 * Following are not relevant for this imsm:
1270 * sparc2.2 : update from old dodgey metadata
1271 * super-minor: change the preferred_minor number
1272 * summaries: update redundant counters.
1273 * uuid: Change the uuid of the array to match watch is given
1274 * homehost: update the recorded homehost
1275 * _reshape_progress: record new reshape_progress position.
1278 //struct intel_super *super = st->sb;
1279 //struct imsm_super *mpb = super->mpb;
1281 if (strcmp(update
, "grow") == 0) {
1283 if (strcmp(update
, "resync") == 0) {
1284 /* dev->vol.dirty = 1; */
1287 /* IMSM has no concept of UUID or homehost */
1292 static size_t disks_to_mpb_size(int disks
)
1296 size
= sizeof(struct imsm_super
);
1297 size
+= (disks
- 1) * sizeof(struct imsm_disk
);
1298 size
+= 2 * sizeof(struct imsm_dev
);
1299 /* up to 2 maps per raid device (-2 for imsm_maps in imsm_dev */
1300 size
+= (4 - 2) * sizeof(struct imsm_map
);
1301 /* 4 possible disk_ord_tbl's */
1302 size
+= 4 * (disks
- 1) * sizeof(__u32
);
1307 static __u64
avail_size_imsm(struct supertype
*st
, __u64 devsize
)
1309 if (devsize
< (MPB_SECTOR_CNT
+ IMSM_RESERVED_SECTORS
))
1312 return devsize
- (MPB_SECTOR_CNT
+ IMSM_RESERVED_SECTORS
);
1315 static void free_devlist(struct intel_super
*super
)
1317 struct intel_dev
*dv
;
1319 while (super
->devlist
) {
1320 dv
= super
->devlist
->next
;
1321 free(super
->devlist
->dev
);
1322 free(super
->devlist
);
1323 super
->devlist
= dv
;
1327 static void imsm_copy_dev(struct imsm_dev
*dest
, struct imsm_dev
*src
)
1329 memcpy(dest
, src
, sizeof_imsm_dev(src
, 0));
1332 static int compare_super_imsm(struct supertype
*st
, struct supertype
*tst
)
1336 * 0 same, or first was empty, and second was copied
1337 * 1 second had wrong number
1339 * 3 wrong other info
1341 struct intel_super
*first
= st
->sb
;
1342 struct intel_super
*sec
= tst
->sb
;
1350 if (memcmp(first
->anchor
->sig
, sec
->anchor
->sig
, MAX_SIGNATURE_LENGTH
) != 0)
1353 /* if an anchor does not have num_raid_devs set then it is a free
1356 if (first
->anchor
->num_raid_devs
> 0 &&
1357 sec
->anchor
->num_raid_devs
> 0) {
1358 if (first
->anchor
->family_num
!= sec
->anchor
->family_num
)
1362 /* if 'first' is a spare promote it to a populated mpb with sec's
1365 if (first
->anchor
->num_raid_devs
== 0 &&
1366 sec
->anchor
->num_raid_devs
> 0) {
1368 struct intel_dev
*dv
;
1369 struct imsm_dev
*dev
;
1371 /* we need to copy raid device info from sec if an allocation
1372 * fails here we don't associate the spare
1374 for (i
= 0; i
< sec
->anchor
->num_raid_devs
; i
++) {
1375 dv
= malloc(sizeof(*dv
));
1378 dev
= malloc(sizeof_imsm_dev(get_imsm_dev(sec
, i
), 1));
1385 dv
->next
= first
->devlist
;
1386 first
->devlist
= dv
;
1388 if (i
<= sec
->anchor
->num_raid_devs
) {
1389 /* allocation failure */
1390 free_devlist(first
);
1391 fprintf(stderr
, "imsm: failed to associate spare\n");
1394 for (i
= 0; i
< sec
->anchor
->num_raid_devs
; i
++)
1395 imsm_copy_dev(get_imsm_dev(first
, i
), get_imsm_dev(sec
, i
));
1397 first
->anchor
->num_raid_devs
= sec
->anchor
->num_raid_devs
;
1398 first
->anchor
->family_num
= sec
->anchor
->family_num
;
1404 static void fd2devname(int fd
, char *name
)
1413 if (fstat(fd
, &st
) != 0)
1415 sprintf(path
, "/sys/dev/block/%d:%d",
1416 major(st
.st_rdev
), minor(st
.st_rdev
));
1418 rv
= readlink(path
, dname
, sizeof(dname
));
1423 nm
= strrchr(dname
, '/');
1425 snprintf(name
, MAX_RAID_SERIAL_LEN
, "/dev/%s", nm
);
1429 extern int scsi_get_serial(int fd
, void *buf
, size_t buf_len
);
1431 static int imsm_read_serial(int fd
, char *devname
,
1432 __u8 serial
[MAX_RAID_SERIAL_LEN
])
1434 unsigned char scsi_serial
[255];
1440 memset(scsi_serial
, 0, sizeof(scsi_serial
));
1442 rv
= scsi_get_serial(fd
, scsi_serial
, sizeof(scsi_serial
));
1444 if (rv
&& check_env("IMSM_DEVNAME_AS_SERIAL")) {
1445 memset(serial
, 0, MAX_RAID_SERIAL_LEN
);
1446 fd2devname(fd
, (char *) serial
);
1453 Name
": Failed to retrieve serial for %s\n",
1458 /* trim leading whitespace */
1459 rsp_len
= scsi_serial
[3];
1463 Name
": Failed to retrieve serial for %s\n",
1467 rsp_buf
= (char *) &scsi_serial
[4];
1472 /* truncate len to the end of rsp_buf if necessary */
1473 if (c
+ MAX_RAID_SERIAL_LEN
> rsp_buf
+ rsp_len
)
1474 len
= rsp_len
- (c
- rsp_buf
);
1476 len
= MAX_RAID_SERIAL_LEN
;
1478 /* initialize the buffer and copy rsp_buf characters */
1479 memset(serial
, 0, MAX_RAID_SERIAL_LEN
);
1480 memcpy(serial
, c
, len
);
1482 /* trim trailing whitespace starting with the last character copied */
1483 c
= (char *) &serial
[len
- 1];
1484 while (isspace(*c
) || *c
== '\0')
1490 static int serialcmp(__u8
*s1
, __u8
*s2
)
1492 return strncmp((char *) s1
, (char *) s2
, MAX_RAID_SERIAL_LEN
);
1495 static void serialcpy(__u8
*dest
, __u8
*src
)
1497 strncpy((char *) dest
, (char *) src
, MAX_RAID_SERIAL_LEN
);
1500 static struct dl
*serial_to_dl(__u8
*serial
, struct intel_super
*super
)
1504 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
1505 if (serialcmp(dl
->serial
, serial
) == 0)
1512 load_imsm_disk(int fd
, struct intel_super
*super
, char *devname
, int keep_fd
)
1519 __u8 serial
[MAX_RAID_SERIAL_LEN
];
1521 rv
= imsm_read_serial(fd
, devname
, serial
);
1526 /* check if this is a disk we have seen before. it may be a spare in
1527 * super->disks while the current anchor believes it is a raid member,
1528 * check if we need to update dl->index
1530 dl
= serial_to_dl(serial
, super
);
1532 dl
= malloc(sizeof(*dl
));
1539 Name
": failed to allocate disk buffer for %s\n",
1546 dl
->major
= major(stb
.st_rdev
);
1547 dl
->minor
= minor(stb
.st_rdev
);
1548 dl
->next
= super
->disks
;
1549 dl
->fd
= keep_fd
? fd
: -1;
1550 dl
->devname
= devname
? strdup(devname
) : NULL
;
1551 serialcpy(dl
->serial
, serial
);
1554 } else if (keep_fd
) {
1559 /* look up this disk's index in the current anchor */
1560 for (i
= 0; i
< super
->anchor
->num_disks
; i
++) {
1561 struct imsm_disk
*disk_iter
;
1563 disk_iter
= __get_imsm_disk(super
->anchor
, i
);
1565 if (serialcmp(disk_iter
->serial
, dl
->serial
) == 0) {
1566 dl
->disk
= *disk_iter
;
1567 /* only set index on disks that are a member of a
1568 * populated contianer, i.e. one with raid_devs
1570 if (dl
->disk
.status
& FAILED_DISK
)
1572 else if (dl
->disk
.status
& SPARE_DISK
)
1581 /* no match, maybe a stale failed drive */
1582 if (i
== super
->anchor
->num_disks
&& dl
->index
>= 0) {
1583 dl
->disk
= *__get_imsm_disk(super
->anchor
, dl
->index
);
1584 if (dl
->disk
.status
& FAILED_DISK
)
1595 /* When migrating map0 contains the 'destination' state while map1
1596 * contains the current state. When not migrating map0 contains the
1597 * current state. This routine assumes that map[0].map_state is set to
1598 * the current array state before being called.
1600 * Migration is indicated by one of the following states
1601 * 1/ Idle (migr_state=0 map0state=normal||unitialized||degraded||failed)
1602 * 2/ Initialize (migr_state=1 migr_type=MIGR_INIT map0state=normal
1603 * map1state=unitialized)
1604 * 3/ Verify (Resync) (migr_state=1 migr_type=MIGR_REBUILD map0state=normal
1606 * 4/ Rebuild (migr_state=1 migr_type=MIGR_REBUILD map0state=normal
1607 * map1state=degraded)
1609 static void migrate(struct imsm_dev
*dev
, __u8 to_state
, int migr_type
)
1611 struct imsm_map
*dest
;
1612 struct imsm_map
*src
= get_imsm_map(dev
, 0);
1614 dev
->vol
.migr_state
= 1;
1615 dev
->vol
.migr_type
= migr_type
;
1616 dev
->vol
.curr_migr_unit
= 0;
1617 dest
= get_imsm_map(dev
, 1);
1619 /* duplicate and then set the target end state in map[0] */
1620 memcpy(dest
, src
, sizeof_imsm_map(src
));
1621 if (migr_type
== MIGR_REBUILD
) {
1625 for (i
= 0; i
< src
->num_members
; i
++) {
1626 ord
= __le32_to_cpu(src
->disk_ord_tbl
[i
]);
1627 set_imsm_ord_tbl_ent(src
, i
, ord_to_idx(ord
));
1631 src
->map_state
= to_state
;
1634 static void end_migration(struct imsm_dev
*dev
, __u8 map_state
)
1636 struct imsm_map
*map
= get_imsm_map(dev
, 0);
1637 struct imsm_map
*prev
= get_imsm_map(dev
, dev
->vol
.migr_state
);
1640 /* merge any IMSM_ORD_REBUILD bits that were not successfully
1641 * completed in the last migration.
1643 * FIXME add support for online capacity expansion and
1644 * raid-level-migration
1646 for (i
= 0; i
< prev
->num_members
; i
++)
1647 map
->disk_ord_tbl
[i
] |= prev
->disk_ord_tbl
[i
];
1649 dev
->vol
.migr_state
= 0;
1650 dev
->vol
.curr_migr_unit
= 0;
1651 map
->map_state
= map_state
;
1655 static int parse_raid_devices(struct intel_super
*super
)
1658 struct imsm_dev
*dev_new
;
1659 size_t len
, len_migr
;
1660 size_t space_needed
= 0;
1661 struct imsm_super
*mpb
= super
->anchor
;
1663 for (i
= 0; i
< super
->anchor
->num_raid_devs
; i
++) {
1664 struct imsm_dev
*dev_iter
= __get_imsm_dev(super
->anchor
, i
);
1665 struct intel_dev
*dv
;
1667 len
= sizeof_imsm_dev(dev_iter
, 0);
1668 len_migr
= sizeof_imsm_dev(dev_iter
, 1);
1670 space_needed
+= len_migr
- len
;
1672 dv
= malloc(sizeof(*dv
));
1675 dev_new
= malloc(len_migr
);
1680 imsm_copy_dev(dev_new
, dev_iter
);
1683 dv
->next
= super
->devlist
;
1684 super
->devlist
= dv
;
1687 /* ensure that super->buf is large enough when all raid devices
1690 if (__le32_to_cpu(mpb
->mpb_size
) + space_needed
> super
->len
) {
1693 len
= ROUND_UP(__le32_to_cpu(mpb
->mpb_size
) + space_needed
, 512);
1694 if (posix_memalign(&buf
, 512, len
) != 0)
1697 memcpy(buf
, super
->buf
, len
);
1706 /* retrieve a pointer to the bbm log which starts after all raid devices */
1707 struct bbm_log
*__get_imsm_bbm_log(struct imsm_super
*mpb
)
1711 if (__le32_to_cpu(mpb
->bbm_log_size
)) {
1713 ptr
+= mpb
->mpb_size
- __le32_to_cpu(mpb
->bbm_log_size
);
1719 static void __free_imsm(struct intel_super
*super
, int free_disks
);
1721 /* load_imsm_mpb - read matrix metadata
1722 * allocates super->mpb to be freed by free_super
1724 static int load_imsm_mpb(int fd
, struct intel_super
*super
, char *devname
)
1726 unsigned long long dsize
;
1727 unsigned long long sectors
;
1729 struct imsm_super
*anchor
;
1733 get_dev_size(fd
, NULL
, &dsize
);
1735 if (lseek64(fd
, dsize
- (512 * 2), SEEK_SET
) < 0) {
1738 Name
": Cannot seek to anchor block on %s: %s\n",
1739 devname
, strerror(errno
));
1743 if (posix_memalign((void**)&anchor
, 512, 512) != 0) {
1746 Name
": Failed to allocate imsm anchor buffer"
1747 " on %s\n", devname
);
1750 if (read(fd
, anchor
, 512) != 512) {
1753 Name
": Cannot read anchor block on %s: %s\n",
1754 devname
, strerror(errno
));
1759 if (strncmp((char *) anchor
->sig
, MPB_SIGNATURE
, MPB_SIG_LEN
) != 0) {
1762 Name
": no IMSM anchor on %s\n", devname
);
1767 __free_imsm(super
, 0);
1768 super
->len
= ROUND_UP(anchor
->mpb_size
, 512);
1769 if (posix_memalign(&super
->buf
, 512, super
->len
) != 0) {
1772 Name
": unable to allocate %zu byte mpb buffer\n",
1777 memcpy(super
->buf
, anchor
, 512);
1779 sectors
= mpb_sectors(anchor
) - 1;
1782 check_sum
= __gen_imsm_checksum(super
->anchor
);
1783 if (check_sum
!= __le32_to_cpu(super
->anchor
->check_sum
)) {
1786 Name
": IMSM checksum %x != %x on %s\n",
1788 __le32_to_cpu(super
->anchor
->check_sum
),
1793 rc
= load_imsm_disk(fd
, super
, devname
, 0);
1795 rc
= parse_raid_devices(super
);
1799 /* read the extended mpb */
1800 if (lseek64(fd
, dsize
- (512 * (2 + sectors
)), SEEK_SET
) < 0) {
1803 Name
": Cannot seek to extended mpb on %s: %s\n",
1804 devname
, strerror(errno
));
1808 if (read(fd
, super
->buf
+ 512, super
->len
- 512) != super
->len
- 512) {
1811 Name
": Cannot read extended mpb on %s: %s\n",
1812 devname
, strerror(errno
));
1816 check_sum
= __gen_imsm_checksum(super
->anchor
);
1817 if (check_sum
!= __le32_to_cpu(super
->anchor
->check_sum
)) {
1820 Name
": IMSM checksum %x != %x on %s\n",
1821 check_sum
, __le32_to_cpu(super
->anchor
->check_sum
),
1826 /* FIXME the BBM log is disk specific so we cannot use this global
1827 * buffer for all disks. Ok for now since we only look at the global
1828 * bbm_log_size parameter to gate assembly
1830 super
->bbm_log
= __get_imsm_bbm_log(super
->anchor
);
1832 rc
= load_imsm_disk(fd
, super
, devname
, 0);
1834 rc
= parse_raid_devices(super
);
1839 static void __free_imsm_disk(struct dl
*d
)
1850 static void free_imsm_disks(struct intel_super
*super
)
1854 while (super
->disks
) {
1856 super
->disks
= d
->next
;
1857 __free_imsm_disk(d
);
1859 while (super
->missing
) {
1861 super
->missing
= d
->next
;
1862 __free_imsm_disk(d
);
1867 /* free all the pieces hanging off of a super pointer */
1868 static void __free_imsm(struct intel_super
*super
, int free_disks
)
1875 free_imsm_disks(super
);
1876 free_devlist(super
);
1878 free((void *) super
->hba
);
1883 static void free_imsm(struct intel_super
*super
)
1885 __free_imsm(super
, 1);
1889 static void free_super_imsm(struct supertype
*st
)
1891 struct intel_super
*super
= st
->sb
;
1900 static struct intel_super
*alloc_super(int creating_imsm
)
1902 struct intel_super
*super
= malloc(sizeof(*super
));
1905 memset(super
, 0, sizeof(*super
));
1906 super
->creating_imsm
= creating_imsm
;
1907 super
->current_vol
= -1;
1908 super
->create_offset
= ~((__u32
) 0);
1909 if (!check_env("IMSM_NO_PLATFORM"))
1910 super
->orom
= find_imsm_orom();
1911 if (super
->orom
&& !check_env("IMSM_TEST_OROM")) {
1912 struct sys_dev
*list
, *ent
;
1914 /* find the first intel ahci controller */
1915 list
= find_driver_devices("pci", "ahci");
1916 for (ent
= list
; ent
; ent
= ent
->next
)
1917 if (devpath_to_vendor(ent
->path
) == 0x8086)
1920 super
->hba
= ent
->path
;
1923 free_sys_dev(&list
);
1931 /* find_missing - helper routine for load_super_imsm_all that identifies
1932 * disks that have disappeared from the system. This routine relies on
1933 * the mpb being uptodate, which it is at load time.
1935 static int find_missing(struct intel_super
*super
)
1938 struct imsm_super
*mpb
= super
->anchor
;
1940 struct imsm_disk
*disk
;
1942 for (i
= 0; i
< mpb
->num_disks
; i
++) {
1943 disk
= __get_imsm_disk(mpb
, i
);
1944 dl
= serial_to_dl(disk
->serial
, super
);
1948 dl
= malloc(sizeof(*dl
));
1954 dl
->devname
= strdup("missing");
1956 serialcpy(dl
->serial
, disk
->serial
);
1959 dl
->next
= super
->missing
;
1960 super
->missing
= dl
;
1966 static int load_super_imsm_all(struct supertype
*st
, int fd
, void **sbp
,
1967 char *devname
, int keep_fd
)
1970 struct intel_super
*super
;
1971 struct mdinfo
*sd
, *best
= NULL
;
1977 int devnum
= fd2devnum(fd
);
1979 enum sysfs_read_flags flags
;
1981 flags
= GET_LEVEL
|GET_VERSION
|GET_DEVS
|GET_STATE
;
1982 if (mdmon_running(devnum
))
1983 flags
|= SKIP_GONE_DEVS
;
1985 /* check if 'fd' an opened container */
1986 sra
= sysfs_read(fd
, 0, flags
);
1990 if (sra
->array
.major_version
!= -1 ||
1991 sra
->array
.minor_version
!= -2 ||
1992 strcmp(sra
->text_version
, "imsm") != 0)
1995 super
= alloc_super(0);
1999 /* find the most up to date disk in this array, skipping spares */
2000 for (sd
= sra
->devs
; sd
; sd
= sd
->next
) {
2001 sprintf(nm
, "%d:%d", sd
->disk
.major
, sd
->disk
.minor
);
2002 dfd
= dev_open(nm
, keep_fd
? O_RDWR
: O_RDONLY
);
2007 rv
= load_imsm_mpb(dfd
, super
, NULL
);
2009 /* retry the load if we might have raced against mdmon */
2010 if (rv
== 3 && mdmon_running(devnum
))
2011 for (retry
= 0; retry
< 3; retry
++) {
2013 rv
= load_imsm_mpb(dfd
, super
, NULL
);
2020 if (super
->anchor
->num_raid_devs
== 0)
2023 gen
= __le32_to_cpu(super
->anchor
->generation_num
);
2024 if (!best
|| gen
> bestgen
) {
2039 /* load the most up to date anchor */
2040 sprintf(nm
, "%d:%d", best
->disk
.major
, best
->disk
.minor
);
2041 dfd
= dev_open(nm
, O_RDONLY
);
2046 rv
= load_imsm_mpb(dfd
, super
, NULL
);
2053 /* re-parse the disk list with the current anchor */
2054 for (sd
= sra
->devs
; sd
; sd
= sd
->next
) {
2055 sprintf(nm
, "%d:%d", sd
->disk
.major
, sd
->disk
.minor
);
2056 dfd
= dev_open(nm
, keep_fd
? O_RDWR
: O_RDONLY
);
2061 load_imsm_disk(dfd
, super
, NULL
, keep_fd
);
2067 if (find_missing(super
) != 0) {
2072 if (st
->subarray
[0]) {
2073 if (atoi(st
->subarray
) <= super
->anchor
->num_raid_devs
)
2074 super
->current_vol
= atoi(st
->subarray
);
2080 st
->container_dev
= devnum
;
2081 if (st
->ss
== NULL
) {
2082 st
->ss
= &super_imsm
;
2083 st
->minor_version
= 0;
2084 st
->max_devs
= IMSM_MAX_DEVICES
;
2086 st
->loaded_container
= 1;
2092 static int load_super_imsm(struct supertype
*st
, int fd
, char *devname
)
2094 struct intel_super
*super
;
2098 if (load_super_imsm_all(st
, fd
, &st
->sb
, devname
, 1) == 0)
2101 if (st
->subarray
[0])
2102 return 1; /* FIXME */
2104 super
= alloc_super(0);
2107 Name
": malloc of %zu failed.\n",
2112 rv
= load_imsm_mpb(fd
, super
, devname
);
2117 Name
": Failed to load all information "
2118 "sections on %s\n", devname
);
2124 if (st
->ss
== NULL
) {
2125 st
->ss
= &super_imsm
;
2126 st
->minor_version
= 0;
2127 st
->max_devs
= IMSM_MAX_DEVICES
;
2129 st
->loaded_container
= 0;
2134 static __u16
info_to_blocks_per_strip(mdu_array_info_t
*info
)
2136 if (info
->level
== 1)
2138 return info
->chunk_size
>> 9;
2141 static __u32
info_to_num_data_stripes(mdu_array_info_t
*info
)
2145 num_stripes
= (info
->size
* 2) / info_to_blocks_per_strip(info
);
2146 if (info
->level
== 1)
2152 static __u32
info_to_blocks_per_member(mdu_array_info_t
*info
)
2154 if (info
->level
== 1)
2155 return info
->size
* 2;
2157 return (info
->size
* 2) & ~(info_to_blocks_per_strip(info
) - 1);
2160 static void imsm_update_version_info(struct intel_super
*super
)
2162 /* update the version and attributes */
2163 struct imsm_super
*mpb
= super
->anchor
;
2165 struct imsm_dev
*dev
;
2166 struct imsm_map
*map
;
2169 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
2170 dev
= get_imsm_dev(super
, i
);
2171 map
= get_imsm_map(dev
, 0);
2172 if (__le32_to_cpu(dev
->size_high
) > 0)
2173 mpb
->attributes
|= MPB_ATTRIB_2TB
;
2175 /* FIXME detect when an array spans a port multiplier */
2177 mpb
->attributes
|= MPB_ATTRIB_PM
;
2180 if (mpb
->num_raid_devs
> 1 ||
2181 mpb
->attributes
!= MPB_ATTRIB_CHECKSUM_VERIFY
) {
2182 version
= MPB_VERSION_ATTRIBS
;
2183 switch (get_imsm_raid_level(map
)) {
2184 case 0: mpb
->attributes
|= MPB_ATTRIB_RAID0
; break;
2185 case 1: mpb
->attributes
|= MPB_ATTRIB_RAID1
; break;
2186 case 10: mpb
->attributes
|= MPB_ATTRIB_RAID10
; break;
2187 case 5: mpb
->attributes
|= MPB_ATTRIB_RAID5
; break;
2190 if (map
->num_members
>= 5)
2191 version
= MPB_VERSION_5OR6_DISK_ARRAY
;
2192 else if (dev
->status
== DEV_CLONE_N_GO
)
2193 version
= MPB_VERSION_CNG
;
2194 else if (get_imsm_raid_level(map
) == 5)
2195 version
= MPB_VERSION_RAID5
;
2196 else if (map
->num_members
>= 3)
2197 version
= MPB_VERSION_3OR4_DISK_ARRAY
;
2198 else if (get_imsm_raid_level(map
) == 1)
2199 version
= MPB_VERSION_RAID1
;
2201 version
= MPB_VERSION_RAID0
;
2203 strcpy(((char *) mpb
->sig
) + strlen(MPB_SIGNATURE
), version
);
2207 static int init_super_imsm_volume(struct supertype
*st
, mdu_array_info_t
*info
,
2208 unsigned long long size
, char *name
,
2209 char *homehost
, int *uuid
)
2211 /* We are creating a volume inside a pre-existing container.
2212 * so st->sb is already set.
2214 struct intel_super
*super
= st
->sb
;
2215 struct imsm_super
*mpb
= super
->anchor
;
2216 struct intel_dev
*dv
;
2217 struct imsm_dev
*dev
;
2218 struct imsm_vol
*vol
;
2219 struct imsm_map
*map
;
2220 int idx
= mpb
->num_raid_devs
;
2222 unsigned long long array_blocks
;
2223 size_t size_old
, size_new
;
2225 if (super
->orom
&& mpb
->num_raid_devs
>= super
->orom
->vpa
) {
2226 fprintf(stderr
, Name
": This imsm-container already has the "
2227 "maximum of %d volumes\n", super
->orom
->vpa
);
2231 /* ensure the mpb is large enough for the new data */
2232 size_old
= __le32_to_cpu(mpb
->mpb_size
);
2233 size_new
= disks_to_mpb_size(info
->nr_disks
);
2234 if (size_new
> size_old
) {
2236 size_t size_round
= ROUND_UP(size_new
, 512);
2238 if (posix_memalign(&mpb_new
, 512, size_round
) != 0) {
2239 fprintf(stderr
, Name
": could not allocate new mpb\n");
2242 memcpy(mpb_new
, mpb
, size_old
);
2245 super
->anchor
= mpb_new
;
2246 mpb
->mpb_size
= __cpu_to_le32(size_new
);
2247 memset(mpb_new
+ size_old
, 0, size_round
- size_old
);
2249 super
->current_vol
= idx
;
2250 /* when creating the first raid device in this container set num_disks
2251 * to zero, i.e. delete this spare and add raid member devices in
2252 * add_to_super_imsm_volume()
2254 if (super
->current_vol
== 0)
2257 for (i
= 0; i
< super
->current_vol
; i
++) {
2258 dev
= get_imsm_dev(super
, i
);
2259 if (strncmp((char *) dev
->volume
, name
,
2260 MAX_RAID_SERIAL_LEN
) == 0) {
2261 fprintf(stderr
, Name
": '%s' is already defined for this container\n",
2267 sprintf(st
->subarray
, "%d", idx
);
2268 dv
= malloc(sizeof(*dv
));
2270 fprintf(stderr
, Name
": failed to allocate device list entry\n");
2273 dev
= malloc(sizeof(*dev
) + sizeof(__u32
) * (info
->raid_disks
- 1));
2276 fprintf(stderr
, Name
": could not allocate raid device\n");
2279 strncpy((char *) dev
->volume
, name
, MAX_RAID_SERIAL_LEN
);
2280 if (info
->level
== 1)
2281 array_blocks
= info_to_blocks_per_member(info
);
2283 array_blocks
= calc_array_size(info
->level
, info
->raid_disks
,
2284 info
->layout
, info
->chunk_size
,
2286 dev
->size_low
= __cpu_to_le32((__u32
) array_blocks
);
2287 dev
->size_high
= __cpu_to_le32((__u32
) (array_blocks
>> 32));
2288 dev
->status
= __cpu_to_le32(0);
2289 dev
->reserved_blocks
= __cpu_to_le32(0);
2291 vol
->migr_state
= 0;
2292 vol
->migr_type
= MIGR_INIT
;
2294 vol
->curr_migr_unit
= 0;
2295 map
= get_imsm_map(dev
, 0);
2296 map
->pba_of_lba0
= __cpu_to_le32(super
->create_offset
);
2297 map
->blocks_per_member
= __cpu_to_le32(info_to_blocks_per_member(info
));
2298 map
->blocks_per_strip
= __cpu_to_le16(info_to_blocks_per_strip(info
));
2299 map
->num_data_stripes
= __cpu_to_le32(info_to_num_data_stripes(info
));
2300 map
->failed_disk_num
= ~0;
2301 map
->map_state
= info
->level
? IMSM_T_STATE_UNINITIALIZED
:
2302 IMSM_T_STATE_NORMAL
;
2304 if (info
->level
== 1 && info
->raid_disks
> 2) {
2305 fprintf(stderr
, Name
": imsm does not support more than 2 disks"
2306 "in a raid1 volume\n");
2309 if (info
->level
== 10) {
2310 map
->raid_level
= 1;
2311 map
->num_domains
= info
->raid_disks
/ 2;
2313 map
->raid_level
= info
->level
;
2314 map
->num_domains
= !!map
->raid_level
;
2317 map
->num_members
= info
->raid_disks
;
2318 for (i
= 0; i
< map
->num_members
; i
++) {
2319 /* initialized in add_to_super */
2320 set_imsm_ord_tbl_ent(map
, i
, 0);
2322 mpb
->num_raid_devs
++;
2325 dv
->index
= super
->current_vol
;
2326 dv
->next
= super
->devlist
;
2327 super
->devlist
= dv
;
2329 imsm_update_version_info(super
);
2334 static int init_super_imsm(struct supertype
*st
, mdu_array_info_t
*info
,
2335 unsigned long long size
, char *name
,
2336 char *homehost
, int *uuid
)
2338 /* This is primarily called by Create when creating a new array.
2339 * We will then get add_to_super called for each component, and then
2340 * write_init_super called to write it out to each device.
2341 * For IMSM, Create can create on fresh devices or on a pre-existing
2343 * To create on a pre-existing array a different method will be called.
2344 * This one is just for fresh drives.
2346 struct intel_super
*super
;
2347 struct imsm_super
*mpb
;
2356 return init_super_imsm_volume(st
, info
, size
, name
, homehost
,
2359 super
= alloc_super(1);
2362 mpb_size
= disks_to_mpb_size(info
->nr_disks
);
2363 if (posix_memalign(&super
->buf
, 512, mpb_size
) != 0) {
2368 memset(mpb
, 0, mpb_size
);
2370 mpb
->attributes
= MPB_ATTRIB_CHECKSUM_VERIFY
;
2372 version
= (char *) mpb
->sig
;
2373 strcpy(version
, MPB_SIGNATURE
);
2374 version
+= strlen(MPB_SIGNATURE
);
2375 strcpy(version
, MPB_VERSION_RAID0
);
2376 mpb
->mpb_size
= mpb_size
;
2383 static int add_to_super_imsm_volume(struct supertype
*st
, mdu_disk_info_t
*dk
,
2384 int fd
, char *devname
)
2386 struct intel_super
*super
= st
->sb
;
2387 struct imsm_super
*mpb
= super
->anchor
;
2389 struct imsm_dev
*dev
;
2390 struct imsm_map
*map
;
2392 dev
= get_imsm_dev(super
, super
->current_vol
);
2393 map
= get_imsm_map(dev
, 0);
2395 if (! (dk
->state
& (1<<MD_DISK_SYNC
))) {
2396 fprintf(stderr
, Name
": %s: Cannot add spare devices to IMSM volume\n",
2401 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
2402 if (dl
->major
== dk
->major
&&
2403 dl
->minor
== dk
->minor
)
2407 fprintf(stderr
, Name
": %s is not a member of the same container\n", devname
);
2411 /* add a pristine spare to the metadata */
2412 if (dl
->index
< 0) {
2413 dl
->index
= super
->anchor
->num_disks
;
2414 super
->anchor
->num_disks
++;
2416 set_imsm_ord_tbl_ent(map
, dk
->number
, dl
->index
);
2417 dl
->disk
.status
= CONFIGURED_DISK
| USABLE_DISK
;
2419 /* if we are creating the first raid device update the family number */
2420 if (super
->current_vol
== 0) {
2422 struct imsm_dev
*_dev
= __get_imsm_dev(mpb
, 0);
2423 struct imsm_disk
*_disk
= __get_imsm_disk(mpb
, dl
->index
);
2427 sum
= __gen_imsm_checksum(mpb
);
2428 mpb
->family_num
= __cpu_to_le32(sum
);
2434 static int add_to_super_imsm(struct supertype
*st
, mdu_disk_info_t
*dk
,
2435 int fd
, char *devname
)
2437 struct intel_super
*super
= st
->sb
;
2439 unsigned long long size
;
2444 /* if we are on an RAID enabled platform check that the disk is
2445 * attached to the raid controller
2447 if (super
->hba
&& !disk_attached_to_hba(fd
, super
->hba
)) {
2449 Name
": %s is not attached to the raid controller: %s\n",
2450 devname
? : "disk", super
->hba
);
2454 if (super
->current_vol
>= 0)
2455 return add_to_super_imsm_volume(st
, dk
, fd
, devname
);
2458 dd
= malloc(sizeof(*dd
));
2461 Name
": malloc failed %s:%d.\n", __func__
, __LINE__
);
2464 memset(dd
, 0, sizeof(*dd
));
2465 dd
->major
= major(stb
.st_rdev
);
2466 dd
->minor
= minor(stb
.st_rdev
);
2468 dd
->devname
= devname
? strdup(devname
) : NULL
;
2471 rv
= imsm_read_serial(fd
, devname
, dd
->serial
);
2474 Name
": failed to retrieve scsi serial, aborting\n");
2479 get_dev_size(fd
, NULL
, &size
);
2481 serialcpy(dd
->disk
.serial
, dd
->serial
);
2482 dd
->disk
.total_blocks
= __cpu_to_le32(size
);
2483 dd
->disk
.status
= USABLE_DISK
| SPARE_DISK
;
2484 if (sysfs_disk_to_scsi_id(fd
, &id
) == 0)
2485 dd
->disk
.scsi_id
= __cpu_to_le32(id
);
2487 dd
->disk
.scsi_id
= __cpu_to_le32(0);
2489 if (st
->update_tail
) {
2490 dd
->next
= super
->add
;
2493 dd
->next
= super
->disks
;
2500 static int store_imsm_mpb(int fd
, struct intel_super
*super
);
2502 /* spare records have their own family number and do not have any defined raid
2505 static int write_super_imsm_spares(struct intel_super
*super
, int doclose
)
2507 struct imsm_super mpb_save
;
2508 struct imsm_super
*mpb
= super
->anchor
;
2513 mpb
->num_raid_devs
= 0;
2515 mpb
->mpb_size
= sizeof(struct imsm_super
);
2516 mpb
->generation_num
= __cpu_to_le32(1UL);
2518 for (d
= super
->disks
; d
; d
= d
->next
) {
2522 mpb
->disk
[0] = d
->disk
;
2523 sum
= __gen_imsm_checksum(mpb
);
2524 mpb
->family_num
= __cpu_to_le32(sum
);
2525 sum
= __gen_imsm_checksum(mpb
);
2526 mpb
->check_sum
= __cpu_to_le32(sum
);
2528 if (store_imsm_mpb(d
->fd
, super
)) {
2529 fprintf(stderr
, "%s: failed for device %d:%d %s\n",
2530 __func__
, d
->major
, d
->minor
, strerror(errno
));
2544 static int write_super_imsm(struct intel_super
*super
, int doclose
)
2546 struct imsm_super
*mpb
= super
->anchor
;
2552 __u32 mpb_size
= sizeof(struct imsm_super
) - sizeof(struct imsm_disk
);
2554 /* 'generation' is incremented everytime the metadata is written */
2555 generation
= __le32_to_cpu(mpb
->generation_num
);
2557 mpb
->generation_num
= __cpu_to_le32(generation
);
2559 mpb_size
+= sizeof(struct imsm_disk
) * mpb
->num_disks
;
2560 for (d
= super
->disks
; d
; d
= d
->next
) {
2564 mpb
->disk
[d
->index
] = d
->disk
;
2566 for (d
= super
->missing
; d
; d
= d
->next
)
2567 mpb
->disk
[d
->index
] = d
->disk
;
2569 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
2570 struct imsm_dev
*dev
= __get_imsm_dev(mpb
, i
);
2572 imsm_copy_dev(dev
, get_imsm_dev(super
, i
));
2573 mpb_size
+= sizeof_imsm_dev(dev
, 0);
2575 mpb_size
+= __le32_to_cpu(mpb
->bbm_log_size
);
2576 mpb
->mpb_size
= __cpu_to_le32(mpb_size
);
2578 /* recalculate checksum */
2579 sum
= __gen_imsm_checksum(mpb
);
2580 mpb
->check_sum
= __cpu_to_le32(sum
);
2582 /* write the mpb for disks that compose raid devices */
2583 for (d
= super
->disks
; d
; d
= d
->next
) {
2586 if (store_imsm_mpb(d
->fd
, super
))
2587 fprintf(stderr
, "%s: failed for device %d:%d %s\n",
2588 __func__
, d
->major
, d
->minor
, strerror(errno
));
2596 return write_super_imsm_spares(super
, doclose
);
2602 static int create_array(struct supertype
*st
)
2605 struct imsm_update_create_array
*u
;
2606 struct intel_super
*super
= st
->sb
;
2607 struct imsm_dev
*dev
= get_imsm_dev(super
, super
->current_vol
);
2608 struct imsm_map
*map
= get_imsm_map(dev
, 0);
2609 struct disk_info
*inf
;
2610 struct imsm_disk
*disk
;
2614 len
= sizeof(*u
) - sizeof(*dev
) + sizeof_imsm_dev(dev
, 0) +
2615 sizeof(*inf
) * map
->num_members
;
2618 fprintf(stderr
, "%s: failed to allocate update buffer\n",
2623 u
->type
= update_create_array
;
2624 u
->dev_idx
= super
->current_vol
;
2625 imsm_copy_dev(&u
->dev
, dev
);
2626 inf
= get_disk_info(u
);
2627 for (i
= 0; i
< map
->num_members
; i
++) {
2628 idx
= get_imsm_disk_idx(dev
, i
);
2629 disk
= get_imsm_disk(super
, idx
);
2630 serialcpy(inf
[i
].serial
, disk
->serial
);
2632 append_metadata_update(st
, u
, len
);
2637 static int _add_disk(struct supertype
*st
)
2639 struct intel_super
*super
= st
->sb
;
2641 struct imsm_update_add_disk
*u
;
2649 fprintf(stderr
, "%s: failed to allocate update buffer\n",
2654 u
->type
= update_add_disk
;
2655 append_metadata_update(st
, u
, len
);
2660 static int write_init_super_imsm(struct supertype
*st
)
2662 if (st
->update_tail
) {
2663 /* queue the recently created array / added disk
2664 * as a metadata update */
2665 struct intel_super
*super
= st
->sb
;
2669 /* determine if we are creating a volume or adding a disk */
2670 if (super
->current_vol
< 0) {
2671 /* in the add disk case we are running in mdmon
2672 * context, so don't close fd's
2674 return _add_disk(st
);
2676 rv
= create_array(st
);
2678 for (d
= super
->disks
; d
; d
= d
->next
) {
2685 return write_super_imsm(st
->sb
, 1);
2689 static int store_zero_imsm(struct supertype
*st
, int fd
)
2691 unsigned long long dsize
;
2694 get_dev_size(fd
, NULL
, &dsize
);
2696 /* first block is stored on second to last sector of the disk */
2697 if (lseek64(fd
, dsize
- (512 * 2), SEEK_SET
) < 0)
2700 if (posix_memalign(&buf
, 512, 512) != 0)
2703 memset(buf
, 0, 512);
2704 if (write(fd
, buf
, 512) != 512)
2709 static int imsm_bbm_log_size(struct imsm_super
*mpb
)
2711 return __le32_to_cpu(mpb
->bbm_log_size
);
2715 static int validate_geometry_imsm_container(struct supertype
*st
, int level
,
2716 int layout
, int raiddisks
, int chunk
,
2717 unsigned long long size
, char *dev
,
2718 unsigned long long *freesize
,
2722 unsigned long long ldsize
;
2723 const struct imsm_orom
*orom
;
2725 if (level
!= LEVEL_CONTAINER
)
2730 if (check_env("IMSM_NO_PLATFORM"))
2733 orom
= find_imsm_orom();
2734 if (orom
&& raiddisks
> orom
->tds
) {
2736 fprintf(stderr
, Name
": %d exceeds maximum number of"
2737 " platform supported disks: %d\n",
2738 raiddisks
, orom
->tds
);
2742 fd
= open(dev
, O_RDONLY
|O_EXCL
, 0);
2745 fprintf(stderr
, Name
": imsm: Cannot open %s: %s\n",
2746 dev
, strerror(errno
));
2749 if (!get_dev_size(fd
, dev
, &ldsize
)) {
2755 *freesize
= avail_size_imsm(st
, ldsize
>> 9);
2760 static unsigned long long find_size(struct extent
*e
, int *idx
, int num_extents
)
2762 const unsigned long long base_start
= e
[*idx
].start
;
2763 unsigned long long end
= base_start
+ e
[*idx
].size
;
2766 if (base_start
== end
)
2770 for (i
= *idx
; i
< num_extents
; i
++) {
2771 /* extend overlapping extents */
2772 if (e
[i
].start
>= base_start
&&
2773 e
[i
].start
<= end
) {
2776 if (e
[i
].start
+ e
[i
].size
> end
)
2777 end
= e
[i
].start
+ e
[i
].size
;
2778 } else if (e
[i
].start
> end
) {
2784 return end
- base_start
;
2787 static unsigned long long merge_extents(struct intel_super
*super
, int sum_extents
)
2789 /* build a composite disk with all known extents and generate a new
2790 * 'maxsize' given the "all disks in an array must share a common start
2791 * offset" constraint
2793 struct extent
*e
= calloc(sum_extents
, sizeof(*e
));
2797 unsigned long long pos
;
2798 unsigned long long start
;
2799 unsigned long long maxsize
;
2800 unsigned long reserve
;
2803 return ~0ULL; /* error */
2805 /* coalesce and sort all extents. also, check to see if we need to
2806 * reserve space between member arrays
2809 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
2812 for (i
= 0; i
< dl
->extent_cnt
; i
++)
2815 qsort(e
, sum_extents
, sizeof(*e
), cmp_extent
);
2820 while (i
< sum_extents
) {
2821 e
[j
].start
= e
[i
].start
;
2822 e
[j
].size
= find_size(e
, &i
, sum_extents
);
2824 if (e
[j
-1].size
== 0)
2833 unsigned long long esize
;
2835 esize
= e
[i
].start
- pos
;
2836 if (esize
>= maxsize
) {
2841 pos
= e
[i
].start
+ e
[i
].size
;
2843 } while (e
[i
-1].size
);
2846 if (start_extent
> 0)
2847 reserve
= IMSM_RESERVED_SECTORS
; /* gap between raid regions */
2851 if (maxsize
< reserve
)
2854 super
->create_offset
= ~((__u32
) 0);
2855 if (start
+ reserve
> super
->create_offset
)
2856 return ~0ULL; /* start overflows create_offset */
2857 super
->create_offset
= start
+ reserve
;
2859 return maxsize
- reserve
;
2862 static int is_raid_level_supported(const struct imsm_orom
*orom
, int level
, int raiddisks
)
2864 if (level
< 0 || level
== 6 || level
== 4)
2867 /* if we have an orom prevent invalid raid levels */
2870 case 0: return imsm_orom_has_raid0(orom
);
2873 return imsm_orom_has_raid1e(orom
);
2874 return imsm_orom_has_raid1(orom
) && raiddisks
== 2;
2875 case 10: return imsm_orom_has_raid10(orom
) && raiddisks
== 4;
2876 case 5: return imsm_orom_has_raid5(orom
) && raiddisks
> 2;
2879 return 1; /* not on an Intel RAID platform so anything goes */
2884 #define pr_vrb(fmt, arg...) (void) (verbose && fprintf(stderr, Name fmt, ##arg))
2885 /* validate_geometry_imsm_volume - lifted from validate_geometry_ddf_bvd
2886 * FIX ME add ahci details
2888 static int validate_geometry_imsm_volume(struct supertype
*st
, int level
,
2889 int layout
, int raiddisks
, int chunk
,
2890 unsigned long long size
, char *dev
,
2891 unsigned long long *freesize
,
2895 struct intel_super
*super
= st
->sb
;
2896 struct imsm_super
*mpb
= super
->anchor
;
2898 unsigned long long pos
= 0;
2899 unsigned long long maxsize
;
2903 /* We must have the container info already read in. */
2907 if (!is_raid_level_supported(super
->orom
, level
, raiddisks
)) {
2908 pr_vrb(": platform does not support raid%d with %d disk%s\n",
2909 level
, raiddisks
, raiddisks
> 1 ? "s" : "");
2912 if (super
->orom
&& level
!= 1 &&
2913 !imsm_orom_has_chunk(super
->orom
, chunk
)) {
2914 pr_vrb(": platform does not support a chunk size of: %d\n", chunk
);
2917 if (layout
!= imsm_level_to_layout(level
)) {
2919 pr_vrb(": imsm raid 5 only supports the left-asymmetric layout\n");
2920 else if (level
== 10)
2921 pr_vrb(": imsm raid 10 only supports the n2 layout\n");
2923 pr_vrb(": imsm unknown layout %#x for this raid level %d\n",
2929 /* General test: make sure there is space for
2930 * 'raiddisks' device extents of size 'size' at a given
2933 unsigned long long minsize
= size
;
2934 unsigned long long start_offset
= ~0ULL;
2937 minsize
= MPB_SECTOR_CNT
+ IMSM_RESERVED_SECTORS
;
2938 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
2943 e
= get_extents(super
, dl
);
2946 unsigned long long esize
;
2947 esize
= e
[i
].start
- pos
;
2948 if (esize
>= minsize
)
2950 if (found
&& start_offset
== ~0ULL) {
2953 } else if (found
&& pos
!= start_offset
) {
2957 pos
= e
[i
].start
+ e
[i
].size
;
2959 } while (e
[i
-1].size
);
2964 if (dcnt
< raiddisks
) {
2966 fprintf(stderr
, Name
": imsm: Not enough "
2967 "devices with space for this array "
2975 /* This device must be a member of the set */
2976 if (stat(dev
, &stb
) < 0)
2978 if ((S_IFMT
& stb
.st_mode
) != S_IFBLK
)
2980 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
2981 if (dl
->major
== major(stb
.st_rdev
) &&
2982 dl
->minor
== minor(stb
.st_rdev
))
2987 fprintf(stderr
, Name
": %s is not in the "
2988 "same imsm set\n", dev
);
2990 } else if (super
->orom
&& dl
->index
< 0 && mpb
->num_raid_devs
) {
2991 /* If a volume is present then the current creation attempt
2992 * cannot incorporate new spares because the orom may not
2993 * understand this configuration (all member disks must be
2994 * members of each array in the container).
2996 fprintf(stderr
, Name
": %s is a spare and a volume"
2997 " is already defined for this container\n", dev
);
2998 fprintf(stderr
, Name
": The option-rom requires all member"
2999 " disks to be a member of all volumes\n");
3003 /* retrieve the largest free space block */
3004 e
= get_extents(super
, dl
);
3009 unsigned long long esize
;
3011 esize
= e
[i
].start
- pos
;
3012 if (esize
>= maxsize
)
3014 pos
= e
[i
].start
+ e
[i
].size
;
3016 } while (e
[i
-1].size
);
3021 fprintf(stderr
, Name
": unable to determine free space for: %s\n",
3025 if (maxsize
< size
) {
3027 fprintf(stderr
, Name
": %s not enough space (%llu < %llu)\n",
3028 dev
, maxsize
, size
);
3032 /* count total number of extents for merge */
3034 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
3036 i
+= dl
->extent_cnt
;
3038 maxsize
= merge_extents(super
, i
);
3039 if (maxsize
< size
) {
3041 fprintf(stderr
, Name
": not enough space after merge (%llu < %llu)\n",
3044 } else if (maxsize
== ~0ULL) {
3046 fprintf(stderr
, Name
": failed to merge %d extents\n", i
);
3050 *freesize
= maxsize
;
3055 static int validate_geometry_imsm(struct supertype
*st
, int level
, int layout
,
3056 int raiddisks
, int chunk
, unsigned long long size
,
3057 char *dev
, unsigned long long *freesize
,
3063 /* if given unused devices create a container
3064 * if given given devices in a container create a member volume
3066 if (level
== LEVEL_CONTAINER
) {
3067 /* Must be a fresh device to add to a container */
3068 return validate_geometry_imsm_container(st
, level
, layout
,
3069 raiddisks
, chunk
, size
,
3075 if (st
->sb
&& freesize
) {
3076 /* Should do auto-layout here */
3077 fprintf(stderr
, Name
": IMSM does not support auto-layout yet\n");
3083 /* creating in a given container */
3084 return validate_geometry_imsm_volume(st
, level
, layout
,
3085 raiddisks
, chunk
, size
,
3086 dev
, freesize
, verbose
);
3089 /* limit creation to the following levels */
3101 /* This device needs to be a device in an 'imsm' container */
3102 fd
= open(dev
, O_RDONLY
|O_EXCL
, 0);
3106 Name
": Cannot create this array on device %s\n",
3111 if (errno
!= EBUSY
|| (fd
= open(dev
, O_RDONLY
, 0)) < 0) {
3113 fprintf(stderr
, Name
": Cannot open %s: %s\n",
3114 dev
, strerror(errno
));
3117 /* Well, it is in use by someone, maybe an 'imsm' container. */
3118 cfd
= open_container(fd
);
3122 fprintf(stderr
, Name
": Cannot use %s: It is busy\n",
3126 sra
= sysfs_read(cfd
, 0, GET_VERSION
);
3128 if (sra
&& sra
->array
.major_version
== -1 &&
3129 strcmp(sra
->text_version
, "imsm") == 0) {
3130 /* This is a member of a imsm container. Load the container
3131 * and try to create a volume
3133 struct intel_super
*super
;
3135 if (load_super_imsm_all(st
, cfd
, (void **) &super
, NULL
, 1) == 0) {
3137 st
->container_dev
= fd2devnum(cfd
);
3139 return validate_geometry_imsm_volume(st
, level
, layout
,
3145 } else /* may belong to another container */
3150 #endif /* MDASSEMBLE */
3152 static struct mdinfo
*container_content_imsm(struct supertype
*st
)
3154 /* Given a container loaded by load_super_imsm_all,
3155 * extract information about all the arrays into
3158 * For each imsm_dev create an mdinfo, fill it in,
3159 * then look for matching devices in super->disks
3160 * and create appropriate device mdinfo.
3162 struct intel_super
*super
= st
->sb
;
3163 struct imsm_super
*mpb
= super
->anchor
;
3164 struct mdinfo
*rest
= NULL
;
3167 /* do not assemble arrays that might have bad blocks */
3168 if (imsm_bbm_log_size(super
->anchor
)) {
3169 fprintf(stderr
, Name
": BBM log found in metadata. "
3170 "Cannot activate array(s).\n");
3174 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
3175 struct imsm_dev
*dev
= get_imsm_dev(super
, i
);
3176 struct imsm_map
*map
= get_imsm_map(dev
, 0);
3177 struct mdinfo
*this;
3180 this = malloc(sizeof(*this));
3181 memset(this, 0, sizeof(*this));
3184 super
->current_vol
= i
;
3185 getinfo_super_imsm_volume(st
, this);
3186 for (slot
= 0 ; slot
< map
->num_members
; slot
++) {
3187 struct mdinfo
*info_d
;
3195 idx
= get_imsm_disk_idx(dev
, slot
);
3196 ord
= get_imsm_ord_tbl_ent(dev
, slot
);
3197 for (d
= super
->disks
; d
; d
= d
->next
)
3198 if (d
->index
== idx
)
3204 s
= d
? d
->disk
.status
: 0;
3205 if (s
& FAILED_DISK
)
3207 if (!(s
& USABLE_DISK
))
3209 if (ord
& IMSM_ORD_REBUILD
)
3213 * if we skip some disks the array will be assmebled degraded;
3214 * reset resync start to avoid a dirty-degraded situation
3216 * FIXME handle dirty degraded
3218 if (skip
&& !dev
->vol
.dirty
)
3219 this->resync_start
= ~0ULL;
3223 info_d
= malloc(sizeof(*info_d
));
3225 fprintf(stderr
, Name
": failed to allocate disk"
3226 " for volume %s\n", (char *) dev
->volume
);
3231 memset(info_d
, 0, sizeof(*info_d
));
3232 info_d
->next
= this->devs
;
3233 this->devs
= info_d
;
3235 info_d
->disk
.number
= d
->index
;
3236 info_d
->disk
.major
= d
->major
;
3237 info_d
->disk
.minor
= d
->minor
;
3238 info_d
->disk
.raid_disk
= slot
;
3240 this->array
.working_disks
++;
3242 info_d
->events
= __le32_to_cpu(mpb
->generation_num
);
3243 info_d
->data_offset
= __le32_to_cpu(map
->pba_of_lba0
);
3244 info_d
->component_size
= __le32_to_cpu(map
->blocks_per_member
);
3246 strcpy(info_d
->name
, d
->devname
);
3256 static int imsm_open_new(struct supertype
*c
, struct active_array
*a
,
3259 struct intel_super
*super
= c
->sb
;
3260 struct imsm_super
*mpb
= super
->anchor
;
3262 if (atoi(inst
) >= mpb
->num_raid_devs
) {
3263 fprintf(stderr
, "%s: subarry index %d, out of range\n",
3264 __func__
, atoi(inst
));
3268 dprintf("imsm: open_new %s\n", inst
);
3269 a
->info
.container_member
= atoi(inst
);
3273 static __u8
imsm_check_degraded(struct intel_super
*super
, struct imsm_dev
*dev
, int failed
)
3275 struct imsm_map
*map
= get_imsm_map(dev
, 0);
3278 return map
->map_state
== IMSM_T_STATE_UNINITIALIZED
?
3279 IMSM_T_STATE_UNINITIALIZED
: IMSM_T_STATE_NORMAL
;
3281 switch (get_imsm_raid_level(map
)) {
3283 return IMSM_T_STATE_FAILED
;
3286 if (failed
< map
->num_members
)
3287 return IMSM_T_STATE_DEGRADED
;
3289 return IMSM_T_STATE_FAILED
;
3294 * check to see if any mirrors have failed, otherwise we
3295 * are degraded. Even numbered slots are mirrored on
3299 /* gcc -Os complains that this is unused */
3300 int insync
= insync
;
3302 for (i
= 0; i
< map
->num_members
; i
++) {
3303 __u32 ord
= get_imsm_ord_tbl_ent(dev
, i
);
3304 int idx
= ord_to_idx(ord
);
3305 struct imsm_disk
*disk
;
3307 /* reset the potential in-sync count on even-numbered
3308 * slots. num_copies is always 2 for imsm raid10
3313 disk
= get_imsm_disk(super
, idx
);
3314 if (!disk
|| disk
->status
& FAILED_DISK
||
3315 ord
& IMSM_ORD_REBUILD
)
3318 /* no in-sync disks left in this mirror the
3322 return IMSM_T_STATE_FAILED
;
3325 return IMSM_T_STATE_DEGRADED
;
3329 return IMSM_T_STATE_DEGRADED
;
3331 return IMSM_T_STATE_FAILED
;
3337 return map
->map_state
;
3340 static int imsm_count_failed(struct intel_super
*super
, struct imsm_dev
*dev
)
3344 struct imsm_disk
*disk
;
3345 struct imsm_map
*map
= get_imsm_map(dev
, 0);
3346 struct imsm_map
*prev
= get_imsm_map(dev
, dev
->vol
.migr_state
);
3350 /* at the beginning of migration we set IMSM_ORD_REBUILD on
3351 * disks that are being rebuilt. New failures are recorded to
3352 * map[0]. So we look through all the disks we started with and
3353 * see if any failures are still present, or if any new ones
3356 * FIXME add support for online capacity expansion and
3357 * raid-level-migration
3359 for (i
= 0; i
< prev
->num_members
; i
++) {
3360 ord
= __le32_to_cpu(prev
->disk_ord_tbl
[i
]);
3361 ord
|= __le32_to_cpu(map
->disk_ord_tbl
[i
]);
3362 idx
= ord_to_idx(ord
);
3364 disk
= get_imsm_disk(super
, idx
);
3365 if (!disk
|| disk
->status
& FAILED_DISK
||
3366 ord
& IMSM_ORD_REBUILD
)
3373 static int is_resyncing(struct imsm_dev
*dev
)
3375 struct imsm_map
*migr_map
;
3377 if (!dev
->vol
.migr_state
)
3380 if (dev
->vol
.migr_type
== MIGR_INIT
)
3383 migr_map
= get_imsm_map(dev
, 1);
3385 if (migr_map
->map_state
== IMSM_T_STATE_NORMAL
)
3391 static int is_rebuilding(struct imsm_dev
*dev
)
3393 struct imsm_map
*migr_map
;
3395 if (!dev
->vol
.migr_state
)
3398 if (dev
->vol
.migr_type
!= MIGR_REBUILD
)
3401 migr_map
= get_imsm_map(dev
, 1);
3403 if (migr_map
->map_state
== IMSM_T_STATE_DEGRADED
)
3409 /* return true if we recorded new information */
3410 static int mark_failure(struct imsm_dev
*dev
, struct imsm_disk
*disk
, int idx
)
3414 struct imsm_map
*map
;
3416 /* new failures are always set in map[0] */
3417 map
= get_imsm_map(dev
, 0);
3419 slot
= get_imsm_disk_slot(map
, idx
);
3423 ord
= __le32_to_cpu(map
->disk_ord_tbl
[slot
]);
3424 if ((disk
->status
& FAILED_DISK
) && (ord
& IMSM_ORD_REBUILD
))
3427 disk
->status
|= FAILED_DISK
;
3428 set_imsm_ord_tbl_ent(map
, slot
, idx
| IMSM_ORD_REBUILD
);
3429 if (map
->failed_disk_num
== ~0)
3430 map
->failed_disk_num
= slot
;
3434 static void mark_missing(struct imsm_dev
*dev
, struct imsm_disk
*disk
, int idx
)
3436 mark_failure(dev
, disk
, idx
);
3438 if (disk
->scsi_id
== __cpu_to_le32(~(__u32
)0))
3441 disk
->scsi_id
= __cpu_to_le32(~(__u32
)0);
3442 memmove(&disk
->serial
[0], &disk
->serial
[1], MAX_RAID_SERIAL_LEN
- 1);
3445 /* Handle dirty -> clean transititions and resync. Degraded and rebuild
3446 * states are handled in imsm_set_disk() with one exception, when a
3447 * resync is stopped due to a new failure this routine will set the
3448 * 'degraded' state for the array.
3450 static int imsm_set_array_state(struct active_array
*a
, int consistent
)
3452 int inst
= a
->info
.container_member
;
3453 struct intel_super
*super
= a
->container
->sb
;
3454 struct imsm_dev
*dev
= get_imsm_dev(super
, inst
);
3455 struct imsm_map
*map
= get_imsm_map(dev
, 0);
3456 int failed
= imsm_count_failed(super
, dev
);
3457 __u8 map_state
= imsm_check_degraded(super
, dev
, failed
);
3459 /* before we activate this array handle any missing disks */
3460 if (consistent
== 2 && super
->missing
) {
3463 dprintf("imsm: mark missing\n");
3464 end_migration(dev
, map_state
);
3465 for (dl
= super
->missing
; dl
; dl
= dl
->next
)
3466 mark_missing(dev
, &dl
->disk
, dl
->index
);
3467 super
->updates_pending
++;
3470 if (consistent
== 2 &&
3471 (!is_resync_complete(a
) ||
3472 map_state
!= IMSM_T_STATE_NORMAL
||
3473 dev
->vol
.migr_state
))
3476 if (is_resync_complete(a
)) {
3477 /* complete intialization / resync,
3478 * recovery and interrupted recovery is completed in
3481 if (is_resyncing(dev
)) {
3482 dprintf("imsm: mark resync done\n");
3483 end_migration(dev
, map_state
);
3484 super
->updates_pending
++;
3486 } else if (!is_resyncing(dev
) && !failed
) {
3487 /* mark the start of the init process if nothing is failed */
3488 dprintf("imsm: mark resync start (%llu)\n", a
->resync_start
);
3489 if (map
->map_state
== IMSM_T_STATE_NORMAL
)
3490 migrate(dev
, IMSM_T_STATE_NORMAL
, MIGR_REBUILD
);
3492 migrate(dev
, IMSM_T_STATE_NORMAL
, MIGR_INIT
);
3493 super
->updates_pending
++;
3496 /* check if we can update the migration checkpoint */
3497 if (dev
->vol
.migr_state
&&
3498 __le32_to_cpu(dev
->vol
.curr_migr_unit
) != a
->resync_start
) {
3499 dprintf("imsm: checkpoint migration (%llu)\n", a
->resync_start
);
3500 dev
->vol
.curr_migr_unit
= __cpu_to_le32(a
->resync_start
);
3501 super
->updates_pending
++;
3504 /* mark dirty / clean */
3505 if (dev
->vol
.dirty
!= !consistent
) {
3506 dprintf("imsm: mark '%s' (%llu)\n",
3507 consistent
? "clean" : "dirty", a
->resync_start
);
3512 super
->updates_pending
++;
3517 static void imsm_set_disk(struct active_array
*a
, int n
, int state
)
3519 int inst
= a
->info
.container_member
;
3520 struct intel_super
*super
= a
->container
->sb
;
3521 struct imsm_dev
*dev
= get_imsm_dev(super
, inst
);
3522 struct imsm_map
*map
= get_imsm_map(dev
, 0);
3523 struct imsm_disk
*disk
;
3528 if (n
> map
->num_members
)
3529 fprintf(stderr
, "imsm: set_disk %d out of range 0..%d\n",
3530 n
, map
->num_members
- 1);
3535 dprintf("imsm: set_disk %d:%x\n", n
, state
);
3537 ord
= get_imsm_ord_tbl_ent(dev
, n
);
3538 disk
= get_imsm_disk(super
, ord_to_idx(ord
));
3540 /* check for new failures */
3541 if (state
& DS_FAULTY
) {
3542 if (mark_failure(dev
, disk
, ord_to_idx(ord
)))
3543 super
->updates_pending
++;
3546 /* check if in_sync */
3547 if (state
& DS_INSYNC
&& ord
& IMSM_ORD_REBUILD
&& is_rebuilding(dev
)) {
3548 struct imsm_map
*migr_map
= get_imsm_map(dev
, 1);
3550 set_imsm_ord_tbl_ent(migr_map
, n
, ord_to_idx(ord
));
3551 super
->updates_pending
++;
3554 failed
= imsm_count_failed(super
, dev
);
3555 map_state
= imsm_check_degraded(super
, dev
, failed
);
3557 /* check if recovery complete, newly degraded, or failed */
3558 if (map_state
== IMSM_T_STATE_NORMAL
&& is_rebuilding(dev
)) {
3559 end_migration(dev
, map_state
);
3560 map
= get_imsm_map(dev
, 0);
3561 map
->failed_disk_num
= ~0;
3562 super
->updates_pending
++;
3563 } else if (map_state
== IMSM_T_STATE_DEGRADED
&&
3564 map
->map_state
!= map_state
&&
3565 !dev
->vol
.migr_state
) {
3566 dprintf("imsm: mark degraded\n");
3567 map
->map_state
= map_state
;
3568 super
->updates_pending
++;
3569 } else if (map_state
== IMSM_T_STATE_FAILED
&&
3570 map
->map_state
!= map_state
) {
3571 dprintf("imsm: mark failed\n");
3572 end_migration(dev
, map_state
);
3573 super
->updates_pending
++;
3577 static int store_imsm_mpb(int fd
, struct intel_super
*super
)
3579 struct imsm_super
*mpb
= super
->anchor
;
3580 __u32 mpb_size
= __le32_to_cpu(mpb
->mpb_size
);
3581 unsigned long long dsize
;
3582 unsigned long long sectors
;
3584 get_dev_size(fd
, NULL
, &dsize
);
3586 if (mpb_size
> 512) {
3587 /* -1 to account for anchor */
3588 sectors
= mpb_sectors(mpb
) - 1;
3590 /* write the extended mpb to the sectors preceeding the anchor */
3591 if (lseek64(fd
, dsize
- (512 * (2 + sectors
)), SEEK_SET
) < 0)
3594 if (write(fd
, super
->buf
+ 512, 512 * sectors
) != 512 * sectors
)
3598 /* first block is stored on second to last sector of the disk */
3599 if (lseek64(fd
, dsize
- (512 * 2), SEEK_SET
) < 0)
3602 if (write(fd
, super
->buf
, 512) != 512)
3608 static void imsm_sync_metadata(struct supertype
*container
)
3610 struct intel_super
*super
= container
->sb
;
3612 if (!super
->updates_pending
)
3615 write_super_imsm(super
, 0);
3617 super
->updates_pending
= 0;
3620 static struct dl
*imsm_readd(struct intel_super
*super
, int idx
, struct active_array
*a
)
3622 struct imsm_dev
*dev
= get_imsm_dev(super
, a
->info
.container_member
);
3623 int i
= get_imsm_disk_idx(dev
, idx
);
3626 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
3630 if (dl
&& dl
->disk
.status
& FAILED_DISK
)
3634 dprintf("%s: found %x:%x\n", __func__
, dl
->major
, dl
->minor
);
3639 static struct dl
*imsm_add_spare(struct intel_super
*super
, int slot
,
3640 struct active_array
*a
, int activate_new
)
3642 struct imsm_dev
*dev
= get_imsm_dev(super
, a
->info
.container_member
);
3643 int idx
= get_imsm_disk_idx(dev
, slot
);
3644 struct imsm_super
*mpb
= super
->anchor
;
3645 struct imsm_map
*map
;
3646 unsigned long long esize
;
3647 unsigned long long pos
;
3656 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
3657 /* If in this array, skip */
3658 for (d
= a
->info
.devs
; d
; d
= d
->next
)
3659 if (d
->state_fd
>= 0 &&
3660 d
->disk
.major
== dl
->major
&&
3661 d
->disk
.minor
== dl
->minor
) {
3662 dprintf("%x:%x already in array\n", dl
->major
, dl
->minor
);
3668 /* skip in use or failed drives */
3669 if (dl
->disk
.status
& FAILED_DISK
|| idx
== dl
->index
||
3671 dprintf("%x:%x status (failed: %d index: %d)\n",
3672 dl
->major
, dl
->minor
,
3673 (dl
->disk
.status
& FAILED_DISK
) == FAILED_DISK
, idx
);
3677 /* skip pure spares when we are looking for partially
3678 * assimilated drives
3680 if (dl
->index
== -1 && !activate_new
)
3683 /* Does this unused device have the requisite free space?
3684 * It needs to be able to cover all member volumes
3686 ex
= get_extents(super
, dl
);
3688 dprintf("cannot get extents\n");
3691 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
3692 dev
= get_imsm_dev(super
, i
);
3693 map
= get_imsm_map(dev
, 0);
3695 /* check if this disk is already a member of
3698 if (get_imsm_disk_slot(map
, dl
->index
) >= 0)
3704 array_start
= __le32_to_cpu(map
->pba_of_lba0
);
3705 blocks
= __le32_to_cpu(map
->blocks_per_member
);
3708 /* check that we can start at pba_of_lba0 with
3709 * blocks_per_member of space
3711 esize
= ex
[j
].start
- pos
;
3712 if (array_start
>= pos
&&
3713 array_start
+ blocks
< ex
[j
].start
) {
3717 pos
= ex
[j
].start
+ ex
[j
].size
;
3719 } while (ex
[j
-1].size
);
3726 if (i
< mpb
->num_raid_devs
) {
3727 dprintf("%x:%x does not have %u at %u\n",
3728 dl
->major
, dl
->minor
,
3729 blocks
, array_start
);
3739 static struct mdinfo
*imsm_activate_spare(struct active_array
*a
,
3740 struct metadata_update
**updates
)
3743 * Find a device with unused free space and use it to replace a
3744 * failed/vacant region in an array. We replace failed regions one a
3745 * array at a time. The result is that a new spare disk will be added
3746 * to the first failed array and after the monitor has finished
3747 * propagating failures the remainder will be consumed.
3749 * FIXME add a capability for mdmon to request spares from another
3753 struct intel_super
*super
= a
->container
->sb
;
3754 int inst
= a
->info
.container_member
;
3755 struct imsm_dev
*dev
= get_imsm_dev(super
, inst
);
3756 struct imsm_map
*map
= get_imsm_map(dev
, 0);
3757 int failed
= a
->info
.array
.raid_disks
;
3758 struct mdinfo
*rv
= NULL
;
3761 struct metadata_update
*mu
;
3763 struct imsm_update_activate_spare
*u
;
3767 for (d
= a
->info
.devs
; d
; d
= d
->next
) {
3768 if ((d
->curr_state
& DS_FAULTY
) &&
3770 /* wait for Removal to happen */
3772 if (d
->state_fd
>= 0)
3776 dprintf("imsm: activate spare: inst=%d failed=%d (%d) level=%d\n",
3777 inst
, failed
, a
->info
.array
.raid_disks
, a
->info
.array
.level
);
3778 if (imsm_check_degraded(super
, dev
, failed
) != IMSM_T_STATE_DEGRADED
)
3781 /* For each slot, if it is not working, find a spare */
3782 for (i
= 0; i
< a
->info
.array
.raid_disks
; i
++) {
3783 for (d
= a
->info
.devs
; d
; d
= d
->next
)
3784 if (d
->disk
.raid_disk
== i
)
3786 dprintf("found %d: %p %x\n", i
, d
, d
?d
->curr_state
:0);
3787 if (d
&& (d
->state_fd
>= 0))
3791 * OK, this device needs recovery. Try to re-add the
3792 * previous occupant of this slot, if this fails see if
3793 * we can continue the assimilation of a spare that was
3794 * partially assimilated, finally try to activate a new
3797 dl
= imsm_readd(super
, i
, a
);
3799 dl
= imsm_add_spare(super
, i
, a
, 0);
3801 dl
= imsm_add_spare(super
, i
, a
, 1);
3805 /* found a usable disk with enough space */
3806 di
= malloc(sizeof(*di
));
3809 memset(di
, 0, sizeof(*di
));
3811 /* dl->index will be -1 in the case we are activating a
3812 * pristine spare. imsm_process_update() will create a
3813 * new index in this case. Once a disk is found to be
3814 * failed in all member arrays it is kicked from the
3817 di
->disk
.number
= dl
->index
;
3819 /* (ab)use di->devs to store a pointer to the device
3822 di
->devs
= (struct mdinfo
*) dl
;
3824 di
->disk
.raid_disk
= i
;
3825 di
->disk
.major
= dl
->major
;
3826 di
->disk
.minor
= dl
->minor
;
3828 di
->data_offset
= __le32_to_cpu(map
->pba_of_lba0
);
3829 di
->component_size
= a
->info
.component_size
;
3830 di
->container_member
= inst
;
3834 dprintf("%x:%x to be %d at %llu\n", dl
->major
, dl
->minor
,
3835 i
, di
->data_offset
);
3841 /* No spares found */
3843 /* Now 'rv' has a list of devices to return.
3844 * Create a metadata_update record to update the
3845 * disk_ord_tbl for the array
3847 mu
= malloc(sizeof(*mu
));
3849 mu
->buf
= malloc(sizeof(struct imsm_update_activate_spare
) * num_spares
);
3850 if (mu
->buf
== NULL
) {
3857 struct mdinfo
*n
= rv
->next
;
3866 mu
->len
= sizeof(struct imsm_update_activate_spare
) * num_spares
;
3867 mu
->next
= *updates
;
3868 u
= (struct imsm_update_activate_spare
*) mu
->buf
;
3870 for (di
= rv
; di
; di
= di
->next
) {
3871 u
->type
= update_activate_spare
;
3872 u
->dl
= (struct dl
*) di
->devs
;
3874 u
->slot
= di
->disk
.raid_disk
;
3885 static int disks_overlap(struct intel_super
*super
, int idx
, struct imsm_update_create_array
*u
)
3887 struct imsm_dev
*dev
= get_imsm_dev(super
, idx
);
3888 struct imsm_map
*map
= get_imsm_map(dev
, 0);
3889 struct imsm_map
*new_map
= get_imsm_map(&u
->dev
, 0);
3890 struct disk_info
*inf
= get_disk_info(u
);
3891 struct imsm_disk
*disk
;
3895 for (i
= 0; i
< map
->num_members
; i
++) {
3896 disk
= get_imsm_disk(super
, get_imsm_disk_idx(dev
, i
));
3897 for (j
= 0; j
< new_map
->num_members
; j
++)
3898 if (serialcmp(disk
->serial
, inf
[j
].serial
) == 0)
3905 static void imsm_delete(struct intel_super
*super
, struct dl
**dlp
, int index
);
3907 static void imsm_process_update(struct supertype
*st
,
3908 struct metadata_update
*update
)
3911 * crack open the metadata_update envelope to find the update record
3912 * update can be one of:
3913 * update_activate_spare - a spare device has replaced a failed
3914 * device in an array, update the disk_ord_tbl. If this disk is
3915 * present in all member arrays then also clear the SPARE_DISK
3918 struct intel_super
*super
= st
->sb
;
3919 struct imsm_super
*mpb
;
3920 enum imsm_update_type type
= *(enum imsm_update_type
*) update
->buf
;
3922 /* update requires a larger buf but the allocation failed */
3923 if (super
->next_len
&& !super
->next_buf
) {
3924 super
->next_len
= 0;
3928 if (super
->next_buf
) {
3929 memcpy(super
->next_buf
, super
->buf
, super
->len
);
3931 super
->len
= super
->next_len
;
3932 super
->buf
= super
->next_buf
;
3934 super
->next_len
= 0;
3935 super
->next_buf
= NULL
;
3938 mpb
= super
->anchor
;
3941 case update_activate_spare
: {
3942 struct imsm_update_activate_spare
*u
= (void *) update
->buf
;
3943 struct imsm_dev
*dev
= get_imsm_dev(super
, u
->array
);
3944 struct imsm_map
*map
= get_imsm_map(dev
, 0);
3945 struct imsm_map
*migr_map
;
3946 struct active_array
*a
;
3947 struct imsm_disk
*disk
;
3952 int victim
= get_imsm_disk_idx(dev
, u
->slot
);
3955 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
3960 fprintf(stderr
, "error: imsm_activate_spare passed "
3961 "an unknown disk (index: %d)\n",
3966 super
->updates_pending
++;
3968 /* count failures (excluding rebuilds and the victim)
3969 * to determine map[0] state
3972 for (i
= 0; i
< map
->num_members
; i
++) {
3975 disk
= get_imsm_disk(super
, get_imsm_disk_idx(dev
, i
));
3976 if (!disk
|| disk
->status
& FAILED_DISK
)
3980 /* adding a pristine spare, assign a new index */
3981 if (dl
->index
< 0) {
3982 dl
->index
= super
->anchor
->num_disks
;
3983 super
->anchor
->num_disks
++;
3986 disk
->status
|= CONFIGURED_DISK
;
3987 disk
->status
&= ~SPARE_DISK
;
3990 to_state
= imsm_check_degraded(super
, dev
, failed
);
3991 map
->map_state
= IMSM_T_STATE_DEGRADED
;
3992 migrate(dev
, to_state
, MIGR_REBUILD
);
3993 migr_map
= get_imsm_map(dev
, 1);
3994 set_imsm_ord_tbl_ent(map
, u
->slot
, dl
->index
);
3995 set_imsm_ord_tbl_ent(migr_map
, u
->slot
, dl
->index
| IMSM_ORD_REBUILD
);
3997 /* count arrays using the victim in the metadata */
3999 for (a
= st
->arrays
; a
; a
= a
->next
) {
4000 dev
= get_imsm_dev(super
, a
->info
.container_member
);
4001 map
= get_imsm_map(dev
, 0);
4003 if (get_imsm_disk_slot(map
, victim
) >= 0)
4007 /* delete the victim if it is no longer being
4013 /* We know that 'manager' isn't touching anything,
4014 * so it is safe to delete
4016 for (dlp
= &super
->disks
; *dlp
; dlp
= &(*dlp
)->next
)
4017 if ((*dlp
)->index
== victim
)
4020 /* victim may be on the missing list */
4022 for (dlp
= &super
->missing
; *dlp
; dlp
= &(*dlp
)->next
)
4023 if ((*dlp
)->index
== victim
)
4025 imsm_delete(super
, dlp
, victim
);
4029 case update_create_array
: {
4030 /* someone wants to create a new array, we need to be aware of
4031 * a few races/collisions:
4032 * 1/ 'Create' called by two separate instances of mdadm
4033 * 2/ 'Create' versus 'activate_spare': mdadm has chosen
4034 * devices that have since been assimilated via
4036 * In the event this update can not be carried out mdadm will
4037 * (FIX ME) notice that its update did not take hold.
4039 struct imsm_update_create_array
*u
= (void *) update
->buf
;
4040 struct intel_dev
*dv
;
4041 struct imsm_dev
*dev
;
4042 struct imsm_map
*map
, *new_map
;
4043 unsigned long long start
, end
;
4044 unsigned long long new_start
, new_end
;
4046 struct disk_info
*inf
;
4049 /* handle racing creates: first come first serve */
4050 if (u
->dev_idx
< mpb
->num_raid_devs
) {
4051 dprintf("%s: subarray %d already defined\n",
4052 __func__
, u
->dev_idx
);
4056 /* check update is next in sequence */
4057 if (u
->dev_idx
!= mpb
->num_raid_devs
) {
4058 dprintf("%s: can not create array %d expected index %d\n",
4059 __func__
, u
->dev_idx
, mpb
->num_raid_devs
);
4063 new_map
= get_imsm_map(&u
->dev
, 0);
4064 new_start
= __le32_to_cpu(new_map
->pba_of_lba0
);
4065 new_end
= new_start
+ __le32_to_cpu(new_map
->blocks_per_member
);
4066 inf
= get_disk_info(u
);
4068 /* handle activate_spare versus create race:
4069 * check to make sure that overlapping arrays do not include
4072 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
4073 dev
= get_imsm_dev(super
, i
);
4074 map
= get_imsm_map(dev
, 0);
4075 start
= __le32_to_cpu(map
->pba_of_lba0
);
4076 end
= start
+ __le32_to_cpu(map
->blocks_per_member
);
4077 if ((new_start
>= start
&& new_start
<= end
) ||
4078 (start
>= new_start
&& start
<= new_end
))
4083 if (disks_overlap(super
, i
, u
)) {
4084 dprintf("%s: arrays overlap\n", __func__
);
4089 /* check that prepare update was successful */
4090 if (!update
->space
) {
4091 dprintf("%s: prepare update failed\n", __func__
);
4095 /* check that all disks are still active before committing
4096 * changes. FIXME: could we instead handle this by creating a
4097 * degraded array? That's probably not what the user expects,
4098 * so better to drop this update on the floor.
4100 for (i
= 0; i
< new_map
->num_members
; i
++) {
4101 dl
= serial_to_dl(inf
[i
].serial
, super
);
4103 dprintf("%s: disk disappeared\n", __func__
);
4108 super
->updates_pending
++;
4110 /* convert spares to members and fixup ord_tbl */
4111 for (i
= 0; i
< new_map
->num_members
; i
++) {
4112 dl
= serial_to_dl(inf
[i
].serial
, super
);
4113 if (dl
->index
== -1) {
4114 dl
->index
= mpb
->num_disks
;
4116 dl
->disk
.status
|= CONFIGURED_DISK
;
4117 dl
->disk
.status
&= ~SPARE_DISK
;
4119 set_imsm_ord_tbl_ent(new_map
, i
, dl
->index
);
4124 update
->space
= NULL
;
4125 imsm_copy_dev(dev
, &u
->dev
);
4126 dv
->index
= u
->dev_idx
;
4127 dv
->next
= super
->devlist
;
4128 super
->devlist
= dv
;
4129 mpb
->num_raid_devs
++;
4131 imsm_update_version_info(super
);
4134 /* mdmon knows how to release update->space, but not
4135 * ((struct intel_dev *) update->space)->dev
4137 if (update
->space
) {
4143 case update_add_disk
:
4145 /* we may be able to repair some arrays if disks are
4148 struct active_array
*a
;
4150 super
->updates_pending
++;
4151 for (a
= st
->arrays
; a
; a
= a
->next
)
4152 a
->check_degraded
= 1;
4154 /* add some spares to the metadata */
4155 while (super
->add
) {
4159 super
->add
= al
->next
;
4160 al
->next
= super
->disks
;
4162 dprintf("%s: added %x:%x\n",
4163 __func__
, al
->major
, al
->minor
);
4170 static void imsm_prepare_update(struct supertype
*st
,
4171 struct metadata_update
*update
)
4174 * Allocate space to hold new disk entries, raid-device entries or a new
4175 * mpb if necessary. The manager synchronously waits for updates to
4176 * complete in the monitor, so new mpb buffers allocated here can be
4177 * integrated by the monitor thread without worrying about live pointers
4178 * in the manager thread.
4180 enum imsm_update_type type
= *(enum imsm_update_type
*) update
->buf
;
4181 struct intel_super
*super
= st
->sb
;
4182 struct imsm_super
*mpb
= super
->anchor
;
4187 case update_create_array
: {
4188 struct imsm_update_create_array
*u
= (void *) update
->buf
;
4189 struct intel_dev
*dv
;
4190 struct imsm_dev
*dev
= &u
->dev
;
4191 struct imsm_map
*map
= get_imsm_map(dev
, 0);
4193 struct disk_info
*inf
;
4197 inf
= get_disk_info(u
);
4198 len
= sizeof_imsm_dev(dev
, 1);
4199 /* allocate a new super->devlist entry */
4200 dv
= malloc(sizeof(*dv
));
4202 dv
->dev
= malloc(len
);
4207 update
->space
= NULL
;
4211 /* count how many spares will be converted to members */
4212 for (i
= 0; i
< map
->num_members
; i
++) {
4213 dl
= serial_to_dl(inf
[i
].serial
, super
);
4215 /* hmm maybe it failed?, nothing we can do about
4220 if (count_memberships(dl
, super
) == 0)
4223 len
+= activate
* sizeof(struct imsm_disk
);
4230 /* check if we need a larger metadata buffer */
4231 if (super
->next_buf
)
4232 buf_len
= super
->next_len
;
4234 buf_len
= super
->len
;
4236 if (__le32_to_cpu(mpb
->mpb_size
) + len
> buf_len
) {
4237 /* ok we need a larger buf than what is currently allocated
4238 * if this allocation fails process_update will notice that
4239 * ->next_len is set and ->next_buf is NULL
4241 buf_len
= ROUND_UP(__le32_to_cpu(mpb
->mpb_size
) + len
, 512);
4242 if (super
->next_buf
)
4243 free(super
->next_buf
);
4245 super
->next_len
= buf_len
;
4246 if (posix_memalign(&super
->next_buf
, 512, buf_len
) != 0)
4247 super
->next_buf
= NULL
;
4251 /* must be called while manager is quiesced */
4252 static void imsm_delete(struct intel_super
*super
, struct dl
**dlp
, int index
)
4254 struct imsm_super
*mpb
= super
->anchor
;
4256 struct imsm_dev
*dev
;
4257 struct imsm_map
*map
;
4258 int i
, j
, num_members
;
4261 dprintf("%s: deleting device[%d] from imsm_super\n",
4264 /* shift all indexes down one */
4265 for (iter
= super
->disks
; iter
; iter
= iter
->next
)
4266 if (iter
->index
> index
)
4268 for (iter
= super
->missing
; iter
; iter
= iter
->next
)
4269 if (iter
->index
> index
)
4272 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
4273 dev
= get_imsm_dev(super
, i
);
4274 map
= get_imsm_map(dev
, 0);
4275 num_members
= map
->num_members
;
4276 for (j
= 0; j
< num_members
; j
++) {
4277 /* update ord entries being careful not to propagate
4278 * ord-flags to the first map
4280 ord
= get_imsm_ord_tbl_ent(dev
, j
);
4282 if (ord_to_idx(ord
) <= index
)
4285 map
= get_imsm_map(dev
, 0);
4286 set_imsm_ord_tbl_ent(map
, j
, ord_to_idx(ord
- 1));
4287 map
= get_imsm_map(dev
, 1);
4289 set_imsm_ord_tbl_ent(map
, j
, ord
- 1);
4294 super
->updates_pending
++;
4296 struct dl
*dl
= *dlp
;
4298 *dlp
= (*dlp
)->next
;
4299 __free_imsm_disk(dl
);
4302 #endif /* MDASSEMBLE */
4304 struct superswitch super_imsm
= {
4306 .examine_super
= examine_super_imsm
,
4307 .brief_examine_super
= brief_examine_super_imsm
,
4308 .detail_super
= detail_super_imsm
,
4309 .brief_detail_super
= brief_detail_super_imsm
,
4310 .write_init_super
= write_init_super_imsm
,
4311 .validate_geometry
= validate_geometry_imsm
,
4312 .add_to_super
= add_to_super_imsm
,
4313 .detail_platform
= detail_platform_imsm
,
4315 .match_home
= match_home_imsm
,
4316 .uuid_from_super
= uuid_from_super_imsm
,
4317 .getinfo_super
= getinfo_super_imsm
,
4318 .update_super
= update_super_imsm
,
4320 .avail_size
= avail_size_imsm
,
4322 .compare_super
= compare_super_imsm
,
4324 .load_super
= load_super_imsm
,
4325 .init_super
= init_super_imsm
,
4326 .store_super
= store_zero_imsm
,
4327 .free_super
= free_super_imsm
,
4328 .match_metadata_desc
= match_metadata_desc_imsm
,
4329 .container_content
= container_content_imsm
,
4330 .default_layout
= imsm_level_to_layout
,
4337 .open_new
= imsm_open_new
,
4338 .load_super
= load_super_imsm
,
4339 .set_array_state
= imsm_set_array_state
,
4340 .set_disk
= imsm_set_disk
,
4341 .sync_metadata
= imsm_sync_metadata
,
4342 .activate_spare
= imsm_activate_spare
,
4343 .process_update
= imsm_process_update
,
4344 .prepare_update
= imsm_prepare_update
,
4345 #endif /* MDASSEMBLE */