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
56 #define SECT_PER_MB_SHIFT 11
58 /* Disk configuration info. */
59 #define IMSM_MAX_DEVICES 255
61 __u8 serial
[MAX_RAID_SERIAL_LEN
];/* 0xD8 - 0xE7 ascii serial number */
62 __u32 total_blocks
; /* 0xE8 - 0xEB total blocks */
63 __u32 scsi_id
; /* 0xEC - 0xEF scsi ID */
64 #define SPARE_DISK __cpu_to_le32(0x01) /* Spare */
65 #define CONFIGURED_DISK __cpu_to_le32(0x02) /* Member of some RaidDev */
66 #define FAILED_DISK __cpu_to_le32(0x04) /* Permanent failure */
67 __u32 status
; /* 0xF0 - 0xF3 */
68 __u32 owner_cfg_num
; /* which config 0,1,2... owns this disk */
69 #define IMSM_DISK_FILLERS 4
70 __u32 filler
[IMSM_DISK_FILLERS
]; /* 0xF4 - 0x107 MPB_DISK_FILLERS for future expansion */
73 /* RAID map configuration infos. */
75 __u32 pba_of_lba0
; /* start address of partition */
76 __u32 blocks_per_member
;/* blocks per member */
77 __u32 num_data_stripes
; /* number of data stripes */
78 __u16 blocks_per_strip
;
79 __u8 map_state
; /* Normal, Uninitialized, Degraded, Failed */
80 #define IMSM_T_STATE_NORMAL 0
81 #define IMSM_T_STATE_UNINITIALIZED 1
82 #define IMSM_T_STATE_DEGRADED 2
83 #define IMSM_T_STATE_FAILED 3
85 #define IMSM_T_RAID0 0
86 #define IMSM_T_RAID1 1
87 #define IMSM_T_RAID5 5 /* since metadata version 1.2.02 ? */
88 __u8 num_members
; /* number of member disks */
89 __u8 num_domains
; /* number of parity domains */
90 __u8 failed_disk_num
; /* valid only when state is degraded */
92 __u32 filler
[7]; /* expansion area */
93 #define IMSM_ORD_REBUILD (1 << 24)
94 __u32 disk_ord_tbl
[1]; /* disk_ord_tbl[num_members],
95 * top byte contains some flags
97 } __attribute__ ((packed
));
100 __u32 curr_migr_unit
;
101 __u32 checkpoint_id
; /* id to access curr_migr_unit */
102 __u8 migr_state
; /* Normal or Migrating */
104 #define MIGR_REBUILD 1
105 #define MIGR_VERIFY 2 /* analagous to echo check > sync_action */
106 #define MIGR_GEN_MIGR 3
107 #define MIGR_STATE_CHANGE 4
108 #define MIGR_REPAIR 5
109 __u8 migr_type
; /* Initializing, Rebuilding, ... */
111 __u8 fs_state
; /* fast-sync state for CnG (0xff == disabled) */
112 __u16 verify_errors
; /* number of mismatches */
113 __u16 bad_blocks
; /* number of bad blocks during verify */
115 struct imsm_map map
[1];
116 /* here comes another one if migr_state */
117 } __attribute__ ((packed
));
120 __u8 volume
[MAX_RAID_SERIAL_LEN
];
123 #define DEV_BOOTABLE __cpu_to_le32(0x01)
124 #define DEV_BOOT_DEVICE __cpu_to_le32(0x02)
125 #define DEV_READ_COALESCING __cpu_to_le32(0x04)
126 #define DEV_WRITE_COALESCING __cpu_to_le32(0x08)
127 #define DEV_LAST_SHUTDOWN_DIRTY __cpu_to_le32(0x10)
128 #define DEV_HIDDEN_AT_BOOT __cpu_to_le32(0x20)
129 #define DEV_CURRENTLY_HIDDEN __cpu_to_le32(0x40)
130 #define DEV_VERIFY_AND_FIX __cpu_to_le32(0x80)
131 #define DEV_MAP_STATE_UNINIT __cpu_to_le32(0x100)
132 #define DEV_NO_AUTO_RECOVERY __cpu_to_le32(0x200)
133 #define DEV_CLONE_N_GO __cpu_to_le32(0x400)
134 #define DEV_CLONE_MAN_SYNC __cpu_to_le32(0x800)
135 #define DEV_CNG_MASTER_DISK_NUM __cpu_to_le32(0x1000)
136 __u32 status
; /* Persistent RaidDev status */
137 __u32 reserved_blocks
; /* Reserved blocks at beginning of volume */
141 __u8 cng_master_disk
;
145 #define IMSM_DEV_FILLERS 10
146 __u32 filler
[IMSM_DEV_FILLERS
];
148 } __attribute__ ((packed
));
151 __u8 sig
[MAX_SIGNATURE_LENGTH
]; /* 0x00 - 0x1F */
152 __u32 check_sum
; /* 0x20 - 0x23 MPB Checksum */
153 __u32 mpb_size
; /* 0x24 - 0x27 Size of MPB */
154 __u32 family_num
; /* 0x28 - 0x2B Checksum from first time this config was written */
155 __u32 generation_num
; /* 0x2C - 0x2F Incremented each time this array's MPB is written */
156 __u32 error_log_size
; /* 0x30 - 0x33 in bytes */
157 __u32 attributes
; /* 0x34 - 0x37 */
158 __u8 num_disks
; /* 0x38 Number of configured disks */
159 __u8 num_raid_devs
; /* 0x39 Number of configured volumes */
160 __u8 error_log_pos
; /* 0x3A */
161 __u8 fill
[1]; /* 0x3B */
162 __u32 cache_size
; /* 0x3c - 0x40 in mb */
163 __u32 orig_family_num
; /* 0x40 - 0x43 original family num */
164 __u32 pwr_cycle_count
; /* 0x44 - 0x47 simulated power cycle count for array */
165 __u32 bbm_log_size
; /* 0x48 - 0x4B - size of bad Block Mgmt Log in bytes */
166 #define IMSM_FILLERS 35
167 __u32 filler
[IMSM_FILLERS
]; /* 0x4C - 0xD7 RAID_MPB_FILLERS */
168 struct imsm_disk disk
[1]; /* 0xD8 diskTbl[numDisks] */
169 /* here comes imsm_dev[num_raid_devs] */
170 /* here comes BBM logs */
171 } __attribute__ ((packed
));
173 #define BBM_LOG_MAX_ENTRIES 254
175 struct bbm_log_entry
{
176 __u64 defective_block_start
;
177 #define UNREADABLE 0xFFFFFFFF
178 __u32 spare_block_offset
;
179 __u16 remapped_marked_count
;
181 } __attribute__ ((__packed__
));
184 __u32 signature
; /* 0xABADB10C */
186 __u32 reserved_spare_block_count
; /* 0 */
187 __u32 reserved
; /* 0xFFFF */
188 __u64 first_spare_lba
;
189 struct bbm_log_entry mapped_block_entries
[BBM_LOG_MAX_ENTRIES
];
190 } __attribute__ ((__packed__
));
194 static char *map_state_str
[] = { "normal", "uninitialized", "degraded", "failed" };
197 static __u8
migr_type(struct imsm_dev
*dev
)
199 if (dev
->vol
.migr_type
== MIGR_VERIFY
&&
200 dev
->status
& DEV_VERIFY_AND_FIX
)
203 return dev
->vol
.migr_type
;
206 static void set_migr_type(struct imsm_dev
*dev
, __u8 migr_type
)
208 /* for compatibility with older oroms convert MIGR_REPAIR, into
209 * MIGR_VERIFY w/ DEV_VERIFY_AND_FIX status
211 if (migr_type
== MIGR_REPAIR
) {
212 dev
->vol
.migr_type
= MIGR_VERIFY
;
213 dev
->status
|= DEV_VERIFY_AND_FIX
;
215 dev
->vol
.migr_type
= migr_type
;
216 dev
->status
&= ~DEV_VERIFY_AND_FIX
;
220 static unsigned int sector_count(__u32 bytes
)
222 return ((bytes
+ (512-1)) & (~(512-1))) / 512;
225 static unsigned int mpb_sectors(struct imsm_super
*mpb
)
227 return sector_count(__le32_to_cpu(mpb
->mpb_size
));
231 struct imsm_dev
*dev
;
232 struct intel_dev
*next
;
236 /* internal representation of IMSM metadata */
239 void *buf
; /* O_DIRECT buffer for reading/writing metadata */
240 struct imsm_super
*anchor
; /* immovable parameters */
242 size_t len
; /* size of the 'buf' allocation */
243 void *next_buf
; /* for realloc'ing buf from the manager */
245 int updates_pending
; /* count of pending updates for mdmon */
246 int creating_imsm
; /* flag to indicate container creation */
247 int current_vol
; /* index of raid device undergoing creation */
248 __u32 create_offset
; /* common start for 'current_vol' */
249 __u32 random
; /* random data for seeding new family numbers */
250 struct intel_dev
*devlist
;
254 __u8 serial
[MAX_RAID_SERIAL_LEN
];
257 struct imsm_disk disk
;
260 struct extent
*e
; /* for determining freespace @ create */
261 int raiddisk
; /* slot to fill in autolayout */
263 struct dl
*add
; /* list of disks to add while mdmon active */
264 struct dl
*missing
; /* disks removed while we weren't looking */
265 struct bbm_log
*bbm_log
;
266 const char *hba
; /* device path of the raid controller for this metadata */
267 const struct imsm_orom
*orom
; /* platform firmware support */
271 unsigned long long start
, size
;
274 /* definition of messages passed to imsm_process_update */
275 enum imsm_update_type
{
276 update_activate_spare
,
281 struct imsm_update_activate_spare
{
282 enum imsm_update_type type
;
286 struct imsm_update_activate_spare
*next
;
290 __u8 serial
[MAX_RAID_SERIAL_LEN
];
293 struct imsm_update_create_array
{
294 enum imsm_update_type type
;
299 struct imsm_update_add_disk
{
300 enum imsm_update_type type
;
303 static struct supertype
*match_metadata_desc_imsm(char *arg
)
305 struct supertype
*st
;
307 if (strcmp(arg
, "imsm") != 0 &&
308 strcmp(arg
, "default") != 0
312 st
= malloc(sizeof(*st
));
313 memset(st
, 0, sizeof(*st
));
314 st
->ss
= &super_imsm
;
315 st
->max_devs
= IMSM_MAX_DEVICES
;
316 st
->minor_version
= 0;
322 static __u8
*get_imsm_version(struct imsm_super
*mpb
)
324 return &mpb
->sig
[MPB_SIG_LEN
];
328 /* retrieve a disk directly from the anchor when the anchor is known to be
329 * up-to-date, currently only at load time
331 static struct imsm_disk
*__get_imsm_disk(struct imsm_super
*mpb
, __u8 index
)
333 if (index
>= mpb
->num_disks
)
335 return &mpb
->disk
[index
];
339 /* retrieve a disk from the parsed metadata */
340 static struct imsm_disk
*get_imsm_disk(struct intel_super
*super
, __u8 index
)
344 for (d
= super
->disks
; d
; d
= d
->next
)
345 if (d
->index
== index
)
352 /* generate a checksum directly from the anchor when the anchor is known to be
353 * up-to-date, currently only at load or write_super after coalescing
355 static __u32
__gen_imsm_checksum(struct imsm_super
*mpb
)
357 __u32 end
= mpb
->mpb_size
/ sizeof(end
);
358 __u32
*p
= (__u32
*) mpb
;
362 sum
+= __le32_to_cpu(*p
);
366 return sum
- __le32_to_cpu(mpb
->check_sum
);
369 static size_t sizeof_imsm_map(struct imsm_map
*map
)
371 return sizeof(struct imsm_map
) + sizeof(__u32
) * (map
->num_members
- 1);
374 struct imsm_map
*get_imsm_map(struct imsm_dev
*dev
, int second_map
)
376 struct imsm_map
*map
= &dev
->vol
.map
[0];
378 if (second_map
&& !dev
->vol
.migr_state
)
380 else if (second_map
) {
383 return ptr
+ sizeof_imsm_map(map
);
389 /* return the size of the device.
390 * migr_state increases the returned size if map[0] were to be duplicated
392 static size_t sizeof_imsm_dev(struct imsm_dev
*dev
, int migr_state
)
394 size_t size
= sizeof(*dev
) - sizeof(struct imsm_map
) +
395 sizeof_imsm_map(get_imsm_map(dev
, 0));
397 /* migrating means an additional map */
398 if (dev
->vol
.migr_state
)
399 size
+= sizeof_imsm_map(get_imsm_map(dev
, 1));
401 size
+= sizeof_imsm_map(get_imsm_map(dev
, 0));
407 /* retrieve disk serial number list from a metadata update */
408 static struct disk_info
*get_disk_info(struct imsm_update_create_array
*update
)
411 struct disk_info
*inf
;
413 inf
= u
+ sizeof(*update
) - sizeof(struct imsm_dev
) +
414 sizeof_imsm_dev(&update
->dev
, 0);
420 static struct imsm_dev
*__get_imsm_dev(struct imsm_super
*mpb
, __u8 index
)
426 if (index
>= mpb
->num_raid_devs
)
429 /* devices start after all disks */
430 offset
= ((void *) &mpb
->disk
[mpb
->num_disks
]) - _mpb
;
432 for (i
= 0; i
<= index
; i
++)
434 return _mpb
+ offset
;
436 offset
+= sizeof_imsm_dev(_mpb
+ offset
, 0);
441 static struct imsm_dev
*get_imsm_dev(struct intel_super
*super
, __u8 index
)
443 struct intel_dev
*dv
;
445 if (index
>= super
->anchor
->num_raid_devs
)
447 for (dv
= super
->devlist
; dv
; dv
= dv
->next
)
448 if (dv
->index
== index
)
453 static __u32
get_imsm_ord_tbl_ent(struct imsm_dev
*dev
, int slot
)
455 struct imsm_map
*map
;
457 if (dev
->vol
.migr_state
)
458 map
= get_imsm_map(dev
, 1);
460 map
= get_imsm_map(dev
, 0);
462 /* top byte identifies disk under rebuild */
463 return __le32_to_cpu(map
->disk_ord_tbl
[slot
]);
466 #define ord_to_idx(ord) (((ord) << 8) >> 8)
467 static __u32
get_imsm_disk_idx(struct imsm_dev
*dev
, int slot
)
469 __u32 ord
= get_imsm_ord_tbl_ent(dev
, slot
);
471 return ord_to_idx(ord
);
474 static void set_imsm_ord_tbl_ent(struct imsm_map
*map
, int slot
, __u32 ord
)
476 map
->disk_ord_tbl
[slot
] = __cpu_to_le32(ord
);
479 static int get_imsm_disk_slot(struct imsm_map
*map
, int idx
)
484 for (slot
= 0; slot
< map
->num_members
; slot
++) {
485 ord
= __le32_to_cpu(map
->disk_ord_tbl
[slot
]);
486 if (ord_to_idx(ord
) == idx
)
493 static int get_imsm_raid_level(struct imsm_map
*map
)
495 if (map
->raid_level
== 1) {
496 if (map
->num_members
== 2)
502 return map
->raid_level
;
505 static int cmp_extent(const void *av
, const void *bv
)
507 const struct extent
*a
= av
;
508 const struct extent
*b
= bv
;
509 if (a
->start
< b
->start
)
511 if (a
->start
> b
->start
)
516 static int count_memberships(struct dl
*dl
, struct intel_super
*super
)
521 for (i
= 0; i
< super
->anchor
->num_raid_devs
; i
++) {
522 struct imsm_dev
*dev
= get_imsm_dev(super
, i
);
523 struct imsm_map
*map
= get_imsm_map(dev
, 0);
525 if (get_imsm_disk_slot(map
, dl
->index
) >= 0)
532 static struct extent
*get_extents(struct intel_super
*super
, struct dl
*dl
)
534 /* find a list of used extents on the given physical device */
535 struct extent
*rv
, *e
;
537 int memberships
= count_memberships(dl
, super
);
538 __u32 reservation
= MPB_SECTOR_CNT
+ IMSM_RESERVED_SECTORS
;
540 rv
= malloc(sizeof(struct extent
) * (memberships
+ 1));
545 for (i
= 0; i
< super
->anchor
->num_raid_devs
; i
++) {
546 struct imsm_dev
*dev
= get_imsm_dev(super
, i
);
547 struct imsm_map
*map
= get_imsm_map(dev
, 0);
549 if (get_imsm_disk_slot(map
, dl
->index
) >= 0) {
550 e
->start
= __le32_to_cpu(map
->pba_of_lba0
);
551 e
->size
= __le32_to_cpu(map
->blocks_per_member
);
555 qsort(rv
, memberships
, sizeof(*rv
), cmp_extent
);
557 /* determine the start of the metadata
558 * when no raid devices are defined use the default
559 * ...otherwise allow the metadata to truncate the value
560 * as is the case with older versions of imsm
563 struct extent
*last
= &rv
[memberships
- 1];
566 remainder
= __le32_to_cpu(dl
->disk
.total_blocks
) -
567 (last
->start
+ last
->size
);
568 /* round down to 1k block to satisfy precision of the kernel
572 /* make sure remainder is still sane */
573 if (remainder
< ROUND_UP(super
->len
, 512) >> 9)
574 remainder
= ROUND_UP(super
->len
, 512) >> 9;
575 if (reservation
> remainder
)
576 reservation
= remainder
;
578 e
->start
= __le32_to_cpu(dl
->disk
.total_blocks
) - reservation
;
583 /* try to determine how much space is reserved for metadata from
584 * the last get_extents() entry, otherwise fallback to the
587 static __u32
imsm_reserved_sectors(struct intel_super
*super
, struct dl
*dl
)
593 /* for spares just return a minimal reservation which will grow
594 * once the spare is picked up by an array
597 return MPB_SECTOR_CNT
;
599 e
= get_extents(super
, dl
);
601 return MPB_SECTOR_CNT
+ IMSM_RESERVED_SECTORS
;
603 /* scroll to last entry */
604 for (i
= 0; e
[i
].size
; i
++)
607 rv
= __le32_to_cpu(dl
->disk
.total_blocks
) - e
[i
].start
;
615 static void print_imsm_dev(struct imsm_dev
*dev
, char *uuid
, int disk_idx
)
619 struct imsm_map
*map
= get_imsm_map(dev
, 0);
623 printf("[%.16s]:\n", dev
->volume
);
624 printf(" UUID : %s\n", uuid
);
625 printf(" RAID Level : %d\n", get_imsm_raid_level(map
));
626 printf(" Members : %d\n", map
->num_members
);
627 slot
= get_imsm_disk_slot(map
, disk_idx
);
629 ord
= get_imsm_ord_tbl_ent(dev
, slot
);
630 printf(" This Slot : %d%s\n", slot
,
631 ord
& IMSM_ORD_REBUILD
? " (out-of-sync)" : "");
633 printf(" This Slot : ?\n");
634 sz
= __le32_to_cpu(dev
->size_high
);
636 sz
+= __le32_to_cpu(dev
->size_low
);
637 printf(" Array Size : %llu%s\n", (unsigned long long)sz
,
638 human_size(sz
* 512));
639 sz
= __le32_to_cpu(map
->blocks_per_member
);
640 printf(" Per Dev Size : %llu%s\n", (unsigned long long)sz
,
641 human_size(sz
* 512));
642 printf(" Sector Offset : %u\n",
643 __le32_to_cpu(map
->pba_of_lba0
));
644 printf(" Num Stripes : %u\n",
645 __le32_to_cpu(map
->num_data_stripes
));
646 printf(" Chunk Size : %u KiB\n",
647 __le16_to_cpu(map
->blocks_per_strip
) / 2);
648 printf(" Reserved : %d\n", __le32_to_cpu(dev
->reserved_blocks
));
649 printf(" Migrate State : %s", dev
->vol
.migr_state
? "migrating" : "idle\n");
650 if (dev
->vol
.migr_state
) {
651 if (migr_type(dev
) == MIGR_INIT
)
652 printf(": initializing\n");
653 else if (migr_type(dev
) == MIGR_REBUILD
)
654 printf(": rebuilding\n");
655 else if (migr_type(dev
) == MIGR_VERIFY
)
657 else if (migr_type(dev
) == MIGR_GEN_MIGR
)
658 printf(": general migration\n");
659 else if (migr_type(dev
) == MIGR_STATE_CHANGE
)
660 printf(": state change\n");
661 else if (migr_type(dev
) == MIGR_REPAIR
)
662 printf(": repair\n");
664 printf(": <unknown:%d>\n", migr_type(dev
));
666 printf(" Map State : %s", map_state_str
[map
->map_state
]);
667 if (dev
->vol
.migr_state
) {
668 struct imsm_map
*map
= get_imsm_map(dev
, 1);
669 printf(" <-- %s", map_state_str
[map
->map_state
]);
672 printf(" Dirty State : %s\n", dev
->vol
.dirty
? "dirty" : "clean");
675 static void print_imsm_disk(struct imsm_super
*mpb
, int index
, __u32 reserved
)
677 struct imsm_disk
*disk
= __get_imsm_disk(mpb
, index
);
678 char str
[MAX_RAID_SERIAL_LEN
+ 1];
686 snprintf(str
, MAX_RAID_SERIAL_LEN
+ 1, "%s", disk
->serial
);
687 printf(" Disk%02d Serial : %s\n", index
, str
);
689 printf(" State :%s%s%s\n", s
&SPARE_DISK
? " spare" : "",
690 s
&CONFIGURED_DISK
? " active" : "",
691 s
&FAILED_DISK
? " failed" : "");
692 printf(" Id : %08x\n", __le32_to_cpu(disk
->scsi_id
));
693 sz
= __le32_to_cpu(disk
->total_blocks
) - reserved
;
694 printf(" Usable Size : %llu%s\n", (unsigned long long)sz
,
695 human_size(sz
* 512));
698 static void getinfo_super_imsm(struct supertype
*st
, struct mdinfo
*info
);
700 static void examine_super_imsm(struct supertype
*st
, char *homehost
)
702 struct intel_super
*super
= st
->sb
;
703 struct imsm_super
*mpb
= super
->anchor
;
704 char str
[MAX_SIGNATURE_LENGTH
];
709 __u32 reserved
= imsm_reserved_sectors(super
, super
->disks
);
712 snprintf(str
, MPB_SIG_LEN
, "%s", mpb
->sig
);
713 printf(" Magic : %s\n", str
);
714 snprintf(str
, strlen(MPB_VERSION_RAID0
), "%s", get_imsm_version(mpb
));
715 printf(" Version : %s\n", get_imsm_version(mpb
));
716 printf(" Orig Family : %08x\n", __le32_to_cpu(mpb
->orig_family_num
));
717 printf(" Family : %08x\n", __le32_to_cpu(mpb
->family_num
));
718 printf(" Generation : %08x\n", __le32_to_cpu(mpb
->generation_num
));
719 getinfo_super_imsm(st
, &info
);
720 fname_from_uuid(st
, &info
, nbuf
, ':');
721 printf(" UUID : %s\n", nbuf
+ 5);
722 sum
= __le32_to_cpu(mpb
->check_sum
);
723 printf(" Checksum : %08x %s\n", sum
,
724 __gen_imsm_checksum(mpb
) == sum
? "correct" : "incorrect");
725 printf(" MPB Sectors : %d\n", mpb_sectors(mpb
));
726 printf(" Disks : %d\n", mpb
->num_disks
);
727 printf(" RAID Devices : %d\n", mpb
->num_raid_devs
);
728 print_imsm_disk(mpb
, super
->disks
->index
, reserved
);
729 if (super
->bbm_log
) {
730 struct bbm_log
*log
= super
->bbm_log
;
733 printf("Bad Block Management Log:\n");
734 printf(" Log Size : %d\n", __le32_to_cpu(mpb
->bbm_log_size
));
735 printf(" Signature : %x\n", __le32_to_cpu(log
->signature
));
736 printf(" Entry Count : %d\n", __le32_to_cpu(log
->entry_count
));
737 printf(" Spare Blocks : %d\n", __le32_to_cpu(log
->reserved_spare_block_count
));
738 printf(" First Spare : %llx\n",
739 (unsigned long long) __le64_to_cpu(log
->first_spare_lba
));
741 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
743 struct imsm_dev
*dev
= __get_imsm_dev(mpb
, i
);
745 super
->current_vol
= i
;
746 getinfo_super_imsm(st
, &info
);
747 fname_from_uuid(st
, &info
, nbuf
, ':');
748 print_imsm_dev(dev
, nbuf
+ 5, super
->disks
->index
);
750 for (i
= 0; i
< mpb
->num_disks
; i
++) {
751 if (i
== super
->disks
->index
)
753 print_imsm_disk(mpb
, i
, reserved
);
757 static void brief_examine_super_imsm(struct supertype
*st
, int verbose
)
759 /* We just write a generic IMSM ARRAY entry */
762 struct intel_super
*super
= st
->sb
;
764 if (!super
->anchor
->num_raid_devs
) {
765 printf("ARRAY metadata=imsm\n");
769 getinfo_super_imsm(st
, &info
);
770 fname_from_uuid(st
, &info
, nbuf
, ':');
771 printf("ARRAY metadata=imsm UUID=%s\n", nbuf
+ 5);
774 static void brief_examine_subarrays_imsm(struct supertype
*st
, int verbose
)
776 /* We just write a generic IMSM ARRAY entry */
780 struct intel_super
*super
= st
->sb
;
783 if (!super
->anchor
->num_raid_devs
)
786 getinfo_super_imsm(st
, &info
);
787 fname_from_uuid(st
, &info
, nbuf
, ':');
788 for (i
= 0; i
< super
->anchor
->num_raid_devs
; i
++) {
789 struct imsm_dev
*dev
= get_imsm_dev(super
, i
);
791 super
->current_vol
= i
;
792 getinfo_super_imsm(st
, &info
);
793 fname_from_uuid(st
, &info
, nbuf1
, ':');
794 printf("ARRAY /dev/md/%.16s container=%s member=%d UUID=%s\n",
795 dev
->volume
, nbuf
+ 5, i
, nbuf1
+ 5);
799 static void export_examine_super_imsm(struct supertype
*st
)
801 struct intel_super
*super
= st
->sb
;
802 struct imsm_super
*mpb
= super
->anchor
;
806 getinfo_super_imsm(st
, &info
);
807 fname_from_uuid(st
, &info
, nbuf
, ':');
808 printf("MD_METADATA=imsm\n");
809 printf("MD_LEVEL=container\n");
810 printf("MD_UUID=%s\n", nbuf
+5);
811 printf("MD_DEVICES=%u\n", mpb
->num_disks
);
814 static void detail_super_imsm(struct supertype
*st
, char *homehost
)
819 getinfo_super_imsm(st
, &info
);
820 fname_from_uuid(st
, &info
, nbuf
, ':');
821 printf("\n UUID : %s\n", nbuf
+ 5);
824 static void brief_detail_super_imsm(struct supertype
*st
)
828 getinfo_super_imsm(st
, &info
);
829 fname_from_uuid(st
, &info
, nbuf
, ':');
830 printf(" UUID=%s", nbuf
+ 5);
833 static int imsm_read_serial(int fd
, char *devname
, __u8
*serial
);
834 static void fd2devname(int fd
, char *name
);
836 static int imsm_enumerate_ports(const char *hba_path
, int port_count
, int host_base
, int verbose
)
838 /* dump an unsorted list of devices attached to ahci, as well as
839 * non-connected ports
841 int hba_len
= strlen(hba_path
) + 1;
846 unsigned long port_mask
= (1 << port_count
) - 1;
848 if (port_count
> sizeof(port_mask
) * 8) {
850 fprintf(stderr
, Name
": port_count %d out of range\n", port_count
);
854 /* scroll through /sys/dev/block looking for devices attached to
857 dir
= opendir("/sys/dev/block");
858 for (ent
= dir
? readdir(dir
) : NULL
; ent
; ent
= readdir(dir
)) {
869 if (sscanf(ent
->d_name
, "%d:%d", &major
, &minor
) != 2)
871 path
= devt_to_devpath(makedev(major
, minor
));
874 if (!path_attached_to_hba(path
, hba_path
)) {
880 /* retrieve the scsi device type */
881 if (asprintf(&device
, "/sys/dev/block/%d:%d/device/xxxxxxx", major
, minor
) < 0) {
883 fprintf(stderr
, Name
": failed to allocate 'device'\n");
887 sprintf(device
, "/sys/dev/block/%d:%d/device/type", major
, minor
);
888 if (load_sys(device
, buf
) != 0) {
890 fprintf(stderr
, Name
": failed to read device type for %s\n",
896 type
= strtoul(buf
, NULL
, 10);
898 /* if it's not a disk print the vendor and model */
899 if (!(type
== 0 || type
== 7 || type
== 14)) {
902 sprintf(device
, "/sys/dev/block/%d:%d/device/vendor", major
, minor
);
903 if (load_sys(device
, buf
) == 0) {
904 strncpy(vendor
, buf
, sizeof(vendor
));
905 vendor
[sizeof(vendor
) - 1] = '\0';
906 c
= (char *) &vendor
[sizeof(vendor
) - 1];
907 while (isspace(*c
) || *c
== '\0')
911 sprintf(device
, "/sys/dev/block/%d:%d/device/model", major
, minor
);
912 if (load_sys(device
, buf
) == 0) {
913 strncpy(model
, buf
, sizeof(model
));
914 model
[sizeof(model
) - 1] = '\0';
915 c
= (char *) &model
[sizeof(model
) - 1];
916 while (isspace(*c
) || *c
== '\0')
920 if (vendor
[0] && model
[0])
921 sprintf(buf
, "%.64s %.64s", vendor
, model
);
923 switch (type
) { /* numbers from hald/linux/device.c */
924 case 1: sprintf(buf
, "tape"); break;
925 case 2: sprintf(buf
, "printer"); break;
926 case 3: sprintf(buf
, "processor"); break;
928 case 5: sprintf(buf
, "cdrom"); break;
929 case 6: sprintf(buf
, "scanner"); break;
930 case 8: sprintf(buf
, "media_changer"); break;
931 case 9: sprintf(buf
, "comm"); break;
932 case 12: sprintf(buf
, "raid"); break;
933 default: sprintf(buf
, "unknown");
939 /* chop device path to 'host%d' and calculate the port number */
940 c
= strchr(&path
[hba_len
], '/');
942 if (sscanf(&path
[hba_len
], "host%d", &port
) == 1)
946 *c
= '/'; /* repair the full string */
947 fprintf(stderr
, Name
": failed to determine port number for %s\n",
954 /* mark this port as used */
955 port_mask
&= ~(1 << port
);
957 /* print out the device information */
959 printf(" Port%d : - non-disk device (%s) -\n", port
, buf
);
963 fd
= dev_open(ent
->d_name
, O_RDONLY
);
965 printf(" Port%d : - disk info unavailable -\n", port
);
968 printf(" Port%d : %s", port
, buf
);
969 if (imsm_read_serial(fd
, NULL
, (__u8
*) buf
) == 0)
970 printf(" (%s)\n", buf
);
985 for (i
= 0; i
< port_count
; i
++)
986 if (port_mask
& (1 << i
))
987 printf(" Port%d : - no device attached -\n", i
);
993 static int detail_platform_imsm(int verbose
, int enumerate_only
)
995 /* There are two components to imsm platform support, the ahci SATA
996 * controller and the option-rom. To find the SATA controller we
997 * simply look in /sys/bus/pci/drivers/ahci to see if an ahci
998 * controller with the Intel vendor id is present. This approach
999 * allows mdadm to leverage the kernel's ahci detection logic, with the
1000 * caveat that if ahci.ko is not loaded mdadm will not be able to
1001 * detect platform raid capabilities. The option-rom resides in a
1002 * platform "Adapter ROM". We scan for its signature to retrieve the
1003 * platform capabilities. If raid support is disabled in the BIOS the
1004 * option-rom capability structure will not be available.
1006 const struct imsm_orom
*orom
;
1007 struct sys_dev
*list
, *hba
;
1010 const char *hba_path
;
1014 if (enumerate_only
) {
1015 if (check_env("IMSM_NO_PLATFORM") || find_imsm_orom())
1020 list
= find_driver_devices("pci", "ahci");
1021 for (hba
= list
; hba
; hba
= hba
->next
)
1022 if (devpath_to_vendor(hba
->path
) == 0x8086)
1027 fprintf(stderr
, Name
": unable to find active ahci controller\n");
1028 free_sys_dev(&list
);
1031 fprintf(stderr
, Name
": found Intel SATA AHCI Controller\n");
1032 hba_path
= hba
->path
;
1034 free_sys_dev(&list
);
1036 orom
= find_imsm_orom();
1039 fprintf(stderr
, Name
": imsm option-rom not found\n");
1043 printf(" Platform : Intel(R) Matrix Storage Manager\n");
1044 printf(" Version : %d.%d.%d.%d\n", orom
->major_ver
, orom
->minor_ver
,
1045 orom
->hotfix_ver
, orom
->build
);
1046 printf(" RAID Levels :%s%s%s%s%s\n",
1047 imsm_orom_has_raid0(orom
) ? " raid0" : "",
1048 imsm_orom_has_raid1(orom
) ? " raid1" : "",
1049 imsm_orom_has_raid1e(orom
) ? " raid1e" : "",
1050 imsm_orom_has_raid10(orom
) ? " raid10" : "",
1051 imsm_orom_has_raid5(orom
) ? " raid5" : "");
1052 printf(" Chunk Sizes :%s%s%s%s%s%s%s%s%s%s%s%s%s%s%s%s\n",
1053 imsm_orom_has_chunk(orom
, 2) ? " 2k" : "",
1054 imsm_orom_has_chunk(orom
, 4) ? " 4k" : "",
1055 imsm_orom_has_chunk(orom
, 8) ? " 8k" : "",
1056 imsm_orom_has_chunk(orom
, 16) ? " 16k" : "",
1057 imsm_orom_has_chunk(orom
, 32) ? " 32k" : "",
1058 imsm_orom_has_chunk(orom
, 64) ? " 64k" : "",
1059 imsm_orom_has_chunk(orom
, 128) ? " 128k" : "",
1060 imsm_orom_has_chunk(orom
, 256) ? " 256k" : "",
1061 imsm_orom_has_chunk(orom
, 512) ? " 512k" : "",
1062 imsm_orom_has_chunk(orom
, 1024*1) ? " 1M" : "",
1063 imsm_orom_has_chunk(orom
, 1024*2) ? " 2M" : "",
1064 imsm_orom_has_chunk(orom
, 1024*4) ? " 4M" : "",
1065 imsm_orom_has_chunk(orom
, 1024*8) ? " 8M" : "",
1066 imsm_orom_has_chunk(orom
, 1024*16) ? " 16M" : "",
1067 imsm_orom_has_chunk(orom
, 1024*32) ? " 32M" : "",
1068 imsm_orom_has_chunk(orom
, 1024*64) ? " 64M" : "");
1069 printf(" Max Disks : %d\n", orom
->tds
);
1070 printf(" Max Volumes : %d\n", orom
->vpa
);
1071 printf(" I/O Controller : %s\n", hba_path
);
1073 /* find the smallest scsi host number to determine a port number base */
1074 dir
= opendir(hba_path
);
1075 for (ent
= dir
? readdir(dir
) : NULL
; ent
; ent
= readdir(dir
)) {
1078 if (sscanf(ent
->d_name
, "host%d", &host
) != 1)
1080 if (port_count
== 0)
1082 else if (host
< host_base
)
1085 if (host
+ 1 > port_count
+ host_base
)
1086 port_count
= host
+ 1 - host_base
;
1092 if (!port_count
|| imsm_enumerate_ports(hba_path
, port_count
,
1093 host_base
, verbose
) != 0) {
1095 fprintf(stderr
, Name
": failed to enumerate ports\n");
1103 static int match_home_imsm(struct supertype
*st
, char *homehost
)
1105 /* the imsm metadata format does not specify any host
1106 * identification information. We return -1 since we can never
1107 * confirm nor deny whether a given array is "meant" for this
1108 * host. We rely on compare_super and the 'family_num' fields to
1109 * exclude member disks that do not belong, and we rely on
1110 * mdadm.conf to specify the arrays that should be assembled.
1111 * Auto-assembly may still pick up "foreign" arrays.
1117 static void uuid_from_super_imsm(struct supertype
*st
, int uuid
[4])
1119 /* The uuid returned here is used for:
1120 * uuid to put into bitmap file (Create, Grow)
1121 * uuid for backup header when saving critical section (Grow)
1122 * comparing uuids when re-adding a device into an array
1123 * In these cases the uuid required is that of the data-array,
1124 * not the device-set.
1125 * uuid to recognise same set when adding a missing device back
1126 * to an array. This is a uuid for the device-set.
1128 * For each of these we can make do with a truncated
1129 * or hashed uuid rather than the original, as long as
1131 * In each case the uuid required is that of the data-array,
1132 * not the device-set.
1134 /* imsm does not track uuid's so we synthesis one using sha1 on
1135 * - The signature (Which is constant for all imsm array, but no matter)
1136 * - the orig_family_num of the container
1137 * - the index number of the volume
1138 * - the 'serial' number of the volume.
1139 * Hopefully these are all constant.
1141 struct intel_super
*super
= st
->sb
;
1144 struct sha1_ctx ctx
;
1145 struct imsm_dev
*dev
= NULL
;
1148 /* some mdadm versions failed to set ->orig_family_num, in which
1149 * case fall back to ->family_num. orig_family_num will be
1150 * fixed up with the first metadata update.
1152 family_num
= super
->anchor
->orig_family_num
;
1153 if (family_num
== 0)
1154 family_num
= super
->anchor
->family_num
;
1155 sha1_init_ctx(&ctx
);
1156 sha1_process_bytes(super
->anchor
->sig
, MPB_SIG_LEN
, &ctx
);
1157 sha1_process_bytes(&family_num
, sizeof(__u32
), &ctx
);
1158 if (super
->current_vol
>= 0)
1159 dev
= get_imsm_dev(super
, super
->current_vol
);
1161 __u32 vol
= super
->current_vol
;
1162 sha1_process_bytes(&vol
, sizeof(vol
), &ctx
);
1163 sha1_process_bytes(dev
->volume
, MAX_RAID_SERIAL_LEN
, &ctx
);
1165 sha1_finish_ctx(&ctx
, buf
);
1166 memcpy(uuid
, buf
, 4*4);
1171 get_imsm_numerical_version(struct imsm_super
*mpb
, int *m
, int *p
)
1173 __u8
*v
= get_imsm_version(mpb
);
1174 __u8
*end
= mpb
->sig
+ MAX_SIGNATURE_LENGTH
;
1175 char major
[] = { 0, 0, 0 };
1176 char minor
[] = { 0 ,0, 0 };
1177 char patch
[] = { 0, 0, 0 };
1178 char *ver_parse
[] = { major
, minor
, patch
};
1182 while (*v
!= '\0' && v
< end
) {
1183 if (*v
!= '.' && j
< 2)
1184 ver_parse
[i
][j
++] = *v
;
1192 *m
= strtol(minor
, NULL
, 0);
1193 *p
= strtol(patch
, NULL
, 0);
1197 static int imsm_level_to_layout(int level
)
1205 return ALGORITHM_LEFT_ASYMMETRIC
;
1212 static void getinfo_super_imsm_volume(struct supertype
*st
, struct mdinfo
*info
)
1214 struct intel_super
*super
= st
->sb
;
1215 struct imsm_dev
*dev
= get_imsm_dev(super
, super
->current_vol
);
1216 struct imsm_map
*map
= get_imsm_map(dev
, 0);
1219 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
1220 if (dl
->raiddisk
== info
->disk
.raid_disk
)
1222 info
->container_member
= super
->current_vol
;
1223 info
->array
.raid_disks
= map
->num_members
;
1224 info
->array
.level
= get_imsm_raid_level(map
);
1225 info
->array
.layout
= imsm_level_to_layout(info
->array
.level
);
1226 info
->array
.md_minor
= -1;
1227 info
->array
.ctime
= 0;
1228 info
->array
.utime
= 0;
1229 info
->array
.chunk_size
= __le16_to_cpu(map
->blocks_per_strip
) << 9;
1230 info
->array
.state
= !dev
->vol
.dirty
;
1231 info
->custom_array_size
= __le32_to_cpu(dev
->size_high
);
1232 info
->custom_array_size
<<= 32;
1233 info
->custom_array_size
|= __le32_to_cpu(dev
->size_low
);
1235 info
->disk
.major
= 0;
1236 info
->disk
.minor
= 0;
1238 info
->disk
.major
= dl
->major
;
1239 info
->disk
.minor
= dl
->minor
;
1242 info
->data_offset
= __le32_to_cpu(map
->pba_of_lba0
);
1243 info
->component_size
= __le32_to_cpu(map
->blocks_per_member
);
1244 memset(info
->uuid
, 0, sizeof(info
->uuid
));
1246 if (map
->map_state
== IMSM_T_STATE_UNINITIALIZED
|| dev
->vol
.dirty
)
1247 info
->resync_start
= 0;
1248 else if (dev
->vol
.migr_state
)
1249 /* FIXME add curr_migr_unit to resync_start conversion */
1250 info
->resync_start
= 0;
1252 info
->resync_start
= ~0ULL;
1254 strncpy(info
->name
, (char *) dev
->volume
, MAX_RAID_SERIAL_LEN
);
1255 info
->name
[MAX_RAID_SERIAL_LEN
] = 0;
1257 info
->array
.major_version
= -1;
1258 info
->array
.minor_version
= -2;
1259 sprintf(info
->text_version
, "/%s/%d",
1260 devnum2devname(st
->container_dev
),
1261 info
->container_member
);
1262 info
->safe_mode_delay
= 4000; /* 4 secs like the Matrix driver */
1263 uuid_from_super_imsm(st
, info
->uuid
);
1266 /* check the config file to see if we can return a real uuid for this spare */
1267 static void fixup_container_spare_uuid(struct mdinfo
*inf
)
1269 struct mddev_ident_s
*array_list
;
1271 if (inf
->array
.level
!= LEVEL_CONTAINER
||
1272 memcmp(inf
->uuid
, uuid_match_any
, sizeof(int[4])) != 0)
1275 array_list
= conf_get_ident(NULL
);
1277 for (; array_list
; array_list
= array_list
->next
) {
1278 if (array_list
->uuid_set
) {
1279 struct supertype
*_sst
; /* spare supertype */
1280 struct supertype
*_cst
; /* container supertype */
1282 _cst
= array_list
->st
;
1284 _sst
= _cst
->ss
->match_metadata_desc(inf
->text_version
);
1289 memcpy(inf
->uuid
, array_list
->uuid
, sizeof(int[4]));
1297 static void getinfo_super_imsm(struct supertype
*st
, struct mdinfo
*info
)
1299 struct intel_super
*super
= st
->sb
;
1300 struct imsm_disk
*disk
;
1303 if (super
->current_vol
>= 0) {
1304 getinfo_super_imsm_volume(st
, info
);
1308 /* Set raid_disks to zero so that Assemble will always pull in valid
1311 info
->array
.raid_disks
= 0;
1312 info
->array
.level
= LEVEL_CONTAINER
;
1313 info
->array
.layout
= 0;
1314 info
->array
.md_minor
= -1;
1315 info
->array
.ctime
= 0; /* N/A for imsm */
1316 info
->array
.utime
= 0;
1317 info
->array
.chunk_size
= 0;
1319 info
->disk
.major
= 0;
1320 info
->disk
.minor
= 0;
1321 info
->disk
.raid_disk
= -1;
1322 info
->reshape_active
= 0;
1323 info
->array
.major_version
= -1;
1324 info
->array
.minor_version
= -2;
1325 strcpy(info
->text_version
, "imsm");
1326 info
->safe_mode_delay
= 0;
1327 info
->disk
.number
= -1;
1328 info
->disk
.state
= 0;
1332 __u32 reserved
= imsm_reserved_sectors(super
, super
->disks
);
1334 disk
= &super
->disks
->disk
;
1335 info
->data_offset
= __le32_to_cpu(disk
->total_blocks
) - reserved
;
1336 info
->component_size
= reserved
;
1338 info
->disk
.state
= s
& CONFIGURED_DISK
? (1 << MD_DISK_ACTIVE
) : 0;
1339 /* we don't change info->disk.raid_disk here because
1340 * this state will be finalized in mdmon after we have
1341 * found the 'most fresh' version of the metadata
1343 info
->disk
.state
|= s
& FAILED_DISK
? (1 << MD_DISK_FAULTY
) : 0;
1344 info
->disk
.state
|= s
& SPARE_DISK
? 0 : (1 << MD_DISK_SYNC
);
1347 /* only call uuid_from_super_imsm when this disk is part of a populated container,
1348 * ->compare_super may have updated the 'num_raid_devs' field for spares
1350 if (info
->disk
.state
& (1 << MD_DISK_SYNC
) || super
->anchor
->num_raid_devs
)
1351 uuid_from_super_imsm(st
, info
->uuid
);
1353 memcpy(info
->uuid
, uuid_match_any
, sizeof(int[4]));
1354 fixup_container_spare_uuid(info
);
1358 static int update_super_imsm(struct supertype
*st
, struct mdinfo
*info
,
1359 char *update
, char *devname
, int verbose
,
1360 int uuid_set
, char *homehost
)
1364 /* For 'assemble' and 'force' we need to return non-zero if any
1365 * change was made. For others, the return value is ignored.
1366 * Update options are:
1367 * force-one : This device looks a bit old but needs to be included,
1368 * update age info appropriately.
1369 * assemble: clear any 'faulty' flag to allow this device to
1371 * force-array: Array is degraded but being forced, mark it clean
1372 * if that will be needed to assemble it.
1374 * newdev: not used ????
1375 * grow: Array has gained a new device - this is currently for
1377 * resync: mark as dirty so a resync will happen.
1378 * name: update the name - preserving the homehost
1380 * Following are not relevant for this imsm:
1381 * sparc2.2 : update from old dodgey metadata
1382 * super-minor: change the preferred_minor number
1383 * summaries: update redundant counters.
1384 * uuid: Change the uuid of the array to match watch is given
1385 * homehost: update the recorded homehost
1386 * _reshape_progress: record new reshape_progress position.
1389 //struct intel_super *super = st->sb;
1390 //struct imsm_super *mpb = super->mpb;
1392 if (strcmp(update
, "grow") == 0) {
1394 if (strcmp(update
, "resync") == 0) {
1395 /* dev->vol.dirty = 1; */
1398 /* IMSM has no concept of UUID or homehost */
1403 static size_t disks_to_mpb_size(int disks
)
1407 size
= sizeof(struct imsm_super
);
1408 size
+= (disks
- 1) * sizeof(struct imsm_disk
);
1409 size
+= 2 * sizeof(struct imsm_dev
);
1410 /* up to 2 maps per raid device (-2 for imsm_maps in imsm_dev */
1411 size
+= (4 - 2) * sizeof(struct imsm_map
);
1412 /* 4 possible disk_ord_tbl's */
1413 size
+= 4 * (disks
- 1) * sizeof(__u32
);
1418 static __u64
avail_size_imsm(struct supertype
*st
, __u64 devsize
)
1420 if (devsize
< (MPB_SECTOR_CNT
+ IMSM_RESERVED_SECTORS
))
1423 return devsize
- (MPB_SECTOR_CNT
+ IMSM_RESERVED_SECTORS
);
1426 static void free_devlist(struct intel_super
*super
)
1428 struct intel_dev
*dv
;
1430 while (super
->devlist
) {
1431 dv
= super
->devlist
->next
;
1432 free(super
->devlist
->dev
);
1433 free(super
->devlist
);
1434 super
->devlist
= dv
;
1438 static void imsm_copy_dev(struct imsm_dev
*dest
, struct imsm_dev
*src
)
1440 memcpy(dest
, src
, sizeof_imsm_dev(src
, 0));
1443 static int compare_super_imsm(struct supertype
*st
, struct supertype
*tst
)
1447 * 0 same, or first was empty, and second was copied
1448 * 1 second had wrong number
1450 * 3 wrong other info
1452 struct intel_super
*first
= st
->sb
;
1453 struct intel_super
*sec
= tst
->sb
;
1461 if (memcmp(first
->anchor
->sig
, sec
->anchor
->sig
, MAX_SIGNATURE_LENGTH
) != 0)
1464 /* if an anchor does not have num_raid_devs set then it is a free
1467 if (first
->anchor
->num_raid_devs
> 0 &&
1468 sec
->anchor
->num_raid_devs
> 0) {
1469 if (first
->anchor
->orig_family_num
!= sec
->anchor
->orig_family_num
||
1470 first
->anchor
->family_num
!= sec
->anchor
->family_num
)
1474 /* if 'first' is a spare promote it to a populated mpb with sec's
1477 if (first
->anchor
->num_raid_devs
== 0 &&
1478 sec
->anchor
->num_raid_devs
> 0) {
1480 struct intel_dev
*dv
;
1481 struct imsm_dev
*dev
;
1483 /* we need to copy raid device info from sec if an allocation
1484 * fails here we don't associate the spare
1486 for (i
= 0; i
< sec
->anchor
->num_raid_devs
; i
++) {
1487 dv
= malloc(sizeof(*dv
));
1490 dev
= malloc(sizeof_imsm_dev(get_imsm_dev(sec
, i
), 1));
1497 dv
->next
= first
->devlist
;
1498 first
->devlist
= dv
;
1500 if (i
< sec
->anchor
->num_raid_devs
) {
1501 /* allocation failure */
1502 free_devlist(first
);
1503 fprintf(stderr
, "imsm: failed to associate spare\n");
1506 first
->anchor
->num_raid_devs
= sec
->anchor
->num_raid_devs
;
1507 first
->anchor
->orig_family_num
= sec
->anchor
->orig_family_num
;
1508 first
->anchor
->family_num
= sec
->anchor
->family_num
;
1509 for (i
= 0; i
< sec
->anchor
->num_raid_devs
; i
++)
1510 imsm_copy_dev(get_imsm_dev(first
, i
), get_imsm_dev(sec
, i
));
1516 static void fd2devname(int fd
, char *name
)
1525 if (fstat(fd
, &st
) != 0)
1527 sprintf(path
, "/sys/dev/block/%d:%d",
1528 major(st
.st_rdev
), minor(st
.st_rdev
));
1530 rv
= readlink(path
, dname
, sizeof(dname
));
1535 nm
= strrchr(dname
, '/');
1537 snprintf(name
, MAX_RAID_SERIAL_LEN
, "/dev/%s", nm
);
1541 extern int scsi_get_serial(int fd
, void *buf
, size_t buf_len
);
1543 static int imsm_read_serial(int fd
, char *devname
,
1544 __u8 serial
[MAX_RAID_SERIAL_LEN
])
1546 unsigned char scsi_serial
[255];
1555 memset(scsi_serial
, 0, sizeof(scsi_serial
));
1557 rv
= scsi_get_serial(fd
, scsi_serial
, sizeof(scsi_serial
));
1559 if (rv
&& check_env("IMSM_DEVNAME_AS_SERIAL")) {
1560 memset(serial
, 0, MAX_RAID_SERIAL_LEN
);
1561 fd2devname(fd
, (char *) serial
);
1568 Name
": Failed to retrieve serial for %s\n",
1573 rsp_len
= scsi_serial
[3];
1577 Name
": Failed to retrieve serial for %s\n",
1581 rsp_buf
= (char *) &scsi_serial
[4];
1583 /* trim all whitespace and non-printable characters and convert
1586 for (i
= 0, dest
= rsp_buf
; i
< rsp_len
; i
++) {
1589 /* ':' is reserved for use in placeholder serial
1590 * numbers for missing disks
1598 len
= dest
- rsp_buf
;
1601 /* truncate leading characters */
1602 if (len
> MAX_RAID_SERIAL_LEN
) {
1603 dest
+= len
- MAX_RAID_SERIAL_LEN
;
1604 len
= MAX_RAID_SERIAL_LEN
;
1607 memset(serial
, 0, MAX_RAID_SERIAL_LEN
);
1608 memcpy(serial
, dest
, len
);
1613 static int serialcmp(__u8
*s1
, __u8
*s2
)
1615 return strncmp((char *) s1
, (char *) s2
, MAX_RAID_SERIAL_LEN
);
1618 static void serialcpy(__u8
*dest
, __u8
*src
)
1620 strncpy((char *) dest
, (char *) src
, MAX_RAID_SERIAL_LEN
);
1623 static struct dl
*serial_to_dl(__u8
*serial
, struct intel_super
*super
)
1627 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
1628 if (serialcmp(dl
->serial
, serial
) == 0)
1635 load_imsm_disk(int fd
, struct intel_super
*super
, char *devname
, int keep_fd
)
1642 __u8 serial
[MAX_RAID_SERIAL_LEN
];
1644 rv
= imsm_read_serial(fd
, devname
, serial
);
1649 /* check if this is a disk we have seen before. it may be a spare in
1650 * super->disks while the current anchor believes it is a raid member,
1651 * check if we need to update dl->index
1653 dl
= serial_to_dl(serial
, super
);
1655 dl
= malloc(sizeof(*dl
));
1662 Name
": failed to allocate disk buffer for %s\n",
1669 dl
->major
= major(stb
.st_rdev
);
1670 dl
->minor
= minor(stb
.st_rdev
);
1671 dl
->next
= super
->disks
;
1672 dl
->fd
= keep_fd
? fd
: -1;
1673 dl
->devname
= devname
? strdup(devname
) : NULL
;
1674 serialcpy(dl
->serial
, serial
);
1677 } else if (keep_fd
) {
1682 /* look up this disk's index in the current anchor */
1683 for (i
= 0; i
< super
->anchor
->num_disks
; i
++) {
1684 struct imsm_disk
*disk_iter
;
1686 disk_iter
= __get_imsm_disk(super
->anchor
, i
);
1688 if (serialcmp(disk_iter
->serial
, dl
->serial
) == 0) {
1689 dl
->disk
= *disk_iter
;
1690 /* only set index on disks that are a member of a
1691 * populated contianer, i.e. one with raid_devs
1693 if (dl
->disk
.status
& FAILED_DISK
)
1695 else if (dl
->disk
.status
& SPARE_DISK
)
1704 /* no match, maybe a stale failed drive */
1705 if (i
== super
->anchor
->num_disks
&& dl
->index
>= 0) {
1706 dl
->disk
= *__get_imsm_disk(super
->anchor
, dl
->index
);
1707 if (dl
->disk
.status
& FAILED_DISK
)
1718 /* When migrating map0 contains the 'destination' state while map1
1719 * contains the current state. When not migrating map0 contains the
1720 * current state. This routine assumes that map[0].map_state is set to
1721 * the current array state before being called.
1723 * Migration is indicated by one of the following states
1724 * 1/ Idle (migr_state=0 map0state=normal||unitialized||degraded||failed)
1725 * 2/ Initialize (migr_state=1 migr_type=MIGR_INIT map0state=normal
1726 * map1state=unitialized)
1727 * 3/ Repair (Resync) (migr_state=1 migr_type=MIGR_REPAIR map0state=normal
1729 * 4/ Rebuild (migr_state=1 migr_type=MIGR_REBUILD map0state=normal
1730 * map1state=degraded)
1732 static void migrate(struct imsm_dev
*dev
, __u8 to_state
, int migr_type
)
1734 struct imsm_map
*dest
;
1735 struct imsm_map
*src
= get_imsm_map(dev
, 0);
1737 dev
->vol
.migr_state
= 1;
1738 set_migr_type(dev
, migr_type
);
1739 dev
->vol
.curr_migr_unit
= 0;
1740 dest
= get_imsm_map(dev
, 1);
1742 /* duplicate and then set the target end state in map[0] */
1743 memcpy(dest
, src
, sizeof_imsm_map(src
));
1744 if (migr_type
== MIGR_REBUILD
) {
1748 for (i
= 0; i
< src
->num_members
; i
++) {
1749 ord
= __le32_to_cpu(src
->disk_ord_tbl
[i
]);
1750 set_imsm_ord_tbl_ent(src
, i
, ord_to_idx(ord
));
1754 src
->map_state
= to_state
;
1757 static void end_migration(struct imsm_dev
*dev
, __u8 map_state
)
1759 struct imsm_map
*map
= get_imsm_map(dev
, 0);
1760 struct imsm_map
*prev
= get_imsm_map(dev
, dev
->vol
.migr_state
);
1763 /* merge any IMSM_ORD_REBUILD bits that were not successfully
1764 * completed in the last migration.
1766 * FIXME add support for online capacity expansion and
1767 * raid-level-migration
1769 for (i
= 0; i
< prev
->num_members
; i
++)
1770 map
->disk_ord_tbl
[i
] |= prev
->disk_ord_tbl
[i
];
1772 dev
->vol
.migr_state
= 0;
1773 dev
->vol
.curr_migr_unit
= 0;
1774 map
->map_state
= map_state
;
1778 static int parse_raid_devices(struct intel_super
*super
)
1781 struct imsm_dev
*dev_new
;
1782 size_t len
, len_migr
;
1783 size_t space_needed
= 0;
1784 struct imsm_super
*mpb
= super
->anchor
;
1786 for (i
= 0; i
< super
->anchor
->num_raid_devs
; i
++) {
1787 struct imsm_dev
*dev_iter
= __get_imsm_dev(super
->anchor
, i
);
1788 struct intel_dev
*dv
;
1790 len
= sizeof_imsm_dev(dev_iter
, 0);
1791 len_migr
= sizeof_imsm_dev(dev_iter
, 1);
1793 space_needed
+= len_migr
- len
;
1795 dv
= malloc(sizeof(*dv
));
1798 dev_new
= malloc(len_migr
);
1803 imsm_copy_dev(dev_new
, dev_iter
);
1806 dv
->next
= super
->devlist
;
1807 super
->devlist
= dv
;
1810 /* ensure that super->buf is large enough when all raid devices
1813 if (__le32_to_cpu(mpb
->mpb_size
) + space_needed
> super
->len
) {
1816 len
= ROUND_UP(__le32_to_cpu(mpb
->mpb_size
) + space_needed
, 512);
1817 if (posix_memalign(&buf
, 512, len
) != 0)
1820 memcpy(buf
, super
->buf
, super
->len
);
1821 memset(buf
+ super
->len
, 0, len
- super
->len
);
1830 /* retrieve a pointer to the bbm log which starts after all raid devices */
1831 struct bbm_log
*__get_imsm_bbm_log(struct imsm_super
*mpb
)
1835 if (__le32_to_cpu(mpb
->bbm_log_size
)) {
1837 ptr
+= mpb
->mpb_size
- __le32_to_cpu(mpb
->bbm_log_size
);
1843 static void __free_imsm(struct intel_super
*super
, int free_disks
);
1845 /* load_imsm_mpb - read matrix metadata
1846 * allocates super->mpb to be freed by free_super
1848 static int load_imsm_mpb(int fd
, struct intel_super
*super
, char *devname
)
1850 unsigned long long dsize
;
1851 unsigned long long sectors
;
1853 struct imsm_super
*anchor
;
1857 get_dev_size(fd
, NULL
, &dsize
);
1859 if (lseek64(fd
, dsize
- (512 * 2), SEEK_SET
) < 0) {
1862 Name
": Cannot seek to anchor block on %s: %s\n",
1863 devname
, strerror(errno
));
1867 if (posix_memalign((void**)&anchor
, 512, 512) != 0) {
1870 Name
": Failed to allocate imsm anchor buffer"
1871 " on %s\n", devname
);
1874 if (read(fd
, anchor
, 512) != 512) {
1877 Name
": Cannot read anchor block on %s: %s\n",
1878 devname
, strerror(errno
));
1883 if (strncmp((char *) anchor
->sig
, MPB_SIGNATURE
, MPB_SIG_LEN
) != 0) {
1886 Name
": no IMSM anchor on %s\n", devname
);
1891 __free_imsm(super
, 0);
1892 super
->len
= ROUND_UP(anchor
->mpb_size
, 512);
1893 if (posix_memalign(&super
->buf
, 512, super
->len
) != 0) {
1896 Name
": unable to allocate %zu byte mpb buffer\n",
1901 memcpy(super
->buf
, anchor
, 512);
1903 sectors
= mpb_sectors(anchor
) - 1;
1906 check_sum
= __gen_imsm_checksum(super
->anchor
);
1907 if (check_sum
!= __le32_to_cpu(super
->anchor
->check_sum
)) {
1910 Name
": IMSM checksum %x != %x on %s\n",
1912 __le32_to_cpu(super
->anchor
->check_sum
),
1917 rc
= load_imsm_disk(fd
, super
, devname
, 0);
1919 rc
= parse_raid_devices(super
);
1923 /* read the extended mpb */
1924 if (lseek64(fd
, dsize
- (512 * (2 + sectors
)), SEEK_SET
) < 0) {
1927 Name
": Cannot seek to extended mpb on %s: %s\n",
1928 devname
, strerror(errno
));
1932 if (read(fd
, super
->buf
+ 512, super
->len
- 512) != super
->len
- 512) {
1935 Name
": Cannot read extended mpb on %s: %s\n",
1936 devname
, strerror(errno
));
1940 check_sum
= __gen_imsm_checksum(super
->anchor
);
1941 if (check_sum
!= __le32_to_cpu(super
->anchor
->check_sum
)) {
1944 Name
": IMSM checksum %x != %x on %s\n",
1945 check_sum
, __le32_to_cpu(super
->anchor
->check_sum
),
1950 /* FIXME the BBM log is disk specific so we cannot use this global
1951 * buffer for all disks. Ok for now since we only look at the global
1952 * bbm_log_size parameter to gate assembly
1954 super
->bbm_log
= __get_imsm_bbm_log(super
->anchor
);
1956 rc
= load_imsm_disk(fd
, super
, devname
, 0);
1958 rc
= parse_raid_devices(super
);
1963 static void __free_imsm_disk(struct dl
*d
)
1974 static void free_imsm_disks(struct intel_super
*super
)
1978 while (super
->disks
) {
1980 super
->disks
= d
->next
;
1981 __free_imsm_disk(d
);
1983 while (super
->missing
) {
1985 super
->missing
= d
->next
;
1986 __free_imsm_disk(d
);
1991 /* free all the pieces hanging off of a super pointer */
1992 static void __free_imsm(struct intel_super
*super
, int free_disks
)
1999 free_imsm_disks(super
);
2000 free_devlist(super
);
2002 free((void *) super
->hba
);
2007 static void free_imsm(struct intel_super
*super
)
2009 __free_imsm(super
, 1);
2013 static void free_super_imsm(struct supertype
*st
)
2015 struct intel_super
*super
= st
->sb
;
2024 static struct intel_super
*alloc_super(int creating_imsm
)
2026 struct intel_super
*super
= malloc(sizeof(*super
));
2029 memset(super
, 0, sizeof(*super
));
2030 super
->creating_imsm
= creating_imsm
;
2031 super
->current_vol
= -1;
2032 super
->create_offset
= ~((__u32
) 0);
2033 if (!check_env("IMSM_NO_PLATFORM"))
2034 super
->orom
= find_imsm_orom();
2035 if (super
->orom
&& !check_env("IMSM_TEST_OROM")) {
2036 struct sys_dev
*list
, *ent
;
2038 /* find the first intel ahci controller */
2039 list
= find_driver_devices("pci", "ahci");
2040 for (ent
= list
; ent
; ent
= ent
->next
)
2041 if (devpath_to_vendor(ent
->path
) == 0x8086)
2044 super
->hba
= ent
->path
;
2047 free_sys_dev(&list
);
2055 /* find_missing - helper routine for load_super_imsm_all that identifies
2056 * disks that have disappeared from the system. This routine relies on
2057 * the mpb being uptodate, which it is at load time.
2059 static int find_missing(struct intel_super
*super
)
2062 struct imsm_super
*mpb
= super
->anchor
;
2064 struct imsm_disk
*disk
;
2066 for (i
= 0; i
< mpb
->num_disks
; i
++) {
2067 disk
= __get_imsm_disk(mpb
, i
);
2068 dl
= serial_to_dl(disk
->serial
, super
);
2072 dl
= malloc(sizeof(*dl
));
2078 dl
->devname
= strdup("missing");
2080 serialcpy(dl
->serial
, disk
->serial
);
2083 dl
->next
= super
->missing
;
2084 super
->missing
= dl
;
2090 static int load_super_imsm_all(struct supertype
*st
, int fd
, void **sbp
,
2091 char *devname
, int keep_fd
)
2094 struct intel_super
*super
;
2095 struct mdinfo
*sd
, *best
= NULL
;
2101 int devnum
= fd2devnum(fd
);
2103 enum sysfs_read_flags flags
;
2105 flags
= GET_LEVEL
|GET_VERSION
|GET_DEVS
|GET_STATE
;
2106 if (mdmon_running(devnum
))
2107 flags
|= SKIP_GONE_DEVS
;
2109 /* check if 'fd' an opened container */
2110 sra
= sysfs_read(fd
, 0, flags
);
2114 if (sra
->array
.major_version
!= -1 ||
2115 sra
->array
.minor_version
!= -2 ||
2116 strcmp(sra
->text_version
, "imsm") != 0)
2119 super
= alloc_super(0);
2123 /* find the most up to date disk in this array, skipping spares */
2124 for (sd
= sra
->devs
; sd
; sd
= sd
->next
) {
2125 sprintf(nm
, "%d:%d", sd
->disk
.major
, sd
->disk
.minor
);
2126 dfd
= dev_open(nm
, keep_fd
? O_RDWR
: O_RDONLY
);
2131 rv
= load_imsm_mpb(dfd
, super
, NULL
);
2133 /* retry the load if we might have raced against mdmon */
2134 if (rv
== 3 && mdmon_running(devnum
))
2135 for (retry
= 0; retry
< 3; retry
++) {
2137 rv
= load_imsm_mpb(dfd
, super
, NULL
);
2144 if (super
->anchor
->num_raid_devs
== 0)
2147 gen
= __le32_to_cpu(super
->anchor
->generation_num
);
2148 if (!best
|| gen
> bestgen
) {
2163 /* load the most up to date anchor */
2164 sprintf(nm
, "%d:%d", best
->disk
.major
, best
->disk
.minor
);
2165 dfd
= dev_open(nm
, O_RDONLY
);
2170 rv
= load_imsm_mpb(dfd
, super
, NULL
);
2177 /* re-parse the disk list with the current anchor */
2178 for (sd
= sra
->devs
; sd
; sd
= sd
->next
) {
2179 sprintf(nm
, "%d:%d", sd
->disk
.major
, sd
->disk
.minor
);
2180 dfd
= dev_open(nm
, keep_fd
? O_RDWR
: O_RDONLY
);
2185 load_imsm_disk(dfd
, super
, NULL
, keep_fd
);
2191 if (find_missing(super
) != 0) {
2196 if (st
->subarray
[0]) {
2197 if (atoi(st
->subarray
) <= super
->anchor
->num_raid_devs
)
2198 super
->current_vol
= atoi(st
->subarray
);
2206 st
->container_dev
= devnum
;
2207 if (st
->ss
== NULL
) {
2208 st
->ss
= &super_imsm
;
2209 st
->minor_version
= 0;
2210 st
->max_devs
= IMSM_MAX_DEVICES
;
2212 st
->loaded_container
= 1;
2218 static int load_super_imsm(struct supertype
*st
, int fd
, char *devname
)
2220 struct intel_super
*super
;
2224 if (load_super_imsm_all(st
, fd
, &st
->sb
, devname
, 1) == 0)
2228 free_super_imsm(st
);
2230 super
= alloc_super(0);
2233 Name
": malloc of %zu failed.\n",
2238 rv
= load_imsm_mpb(fd
, super
, devname
);
2243 Name
": Failed to load all information "
2244 "sections on %s\n", devname
);
2249 if (st
->subarray
[0]) {
2250 if (atoi(st
->subarray
) <= super
->anchor
->num_raid_devs
)
2251 super
->current_vol
= atoi(st
->subarray
);
2259 if (st
->ss
== NULL
) {
2260 st
->ss
= &super_imsm
;
2261 st
->minor_version
= 0;
2262 st
->max_devs
= IMSM_MAX_DEVICES
;
2264 st
->loaded_container
= 0;
2269 static __u16
info_to_blocks_per_strip(mdu_array_info_t
*info
)
2271 if (info
->level
== 1)
2273 return info
->chunk_size
>> 9;
2276 static __u32
info_to_num_data_stripes(mdu_array_info_t
*info
, int num_domains
)
2280 num_stripes
= (info
->size
* 2) / info_to_blocks_per_strip(info
);
2281 num_stripes
/= num_domains
;
2286 static __u32
info_to_blocks_per_member(mdu_array_info_t
*info
)
2288 if (info
->level
== 1)
2289 return info
->size
* 2;
2291 return (info
->size
* 2) & ~(info_to_blocks_per_strip(info
) - 1);
2294 static void imsm_update_version_info(struct intel_super
*super
)
2296 /* update the version and attributes */
2297 struct imsm_super
*mpb
= super
->anchor
;
2299 struct imsm_dev
*dev
;
2300 struct imsm_map
*map
;
2303 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
2304 dev
= get_imsm_dev(super
, i
);
2305 map
= get_imsm_map(dev
, 0);
2306 if (__le32_to_cpu(dev
->size_high
) > 0)
2307 mpb
->attributes
|= MPB_ATTRIB_2TB
;
2309 /* FIXME detect when an array spans a port multiplier */
2311 mpb
->attributes
|= MPB_ATTRIB_PM
;
2314 if (mpb
->num_raid_devs
> 1 ||
2315 mpb
->attributes
!= MPB_ATTRIB_CHECKSUM_VERIFY
) {
2316 version
= MPB_VERSION_ATTRIBS
;
2317 switch (get_imsm_raid_level(map
)) {
2318 case 0: mpb
->attributes
|= MPB_ATTRIB_RAID0
; break;
2319 case 1: mpb
->attributes
|= MPB_ATTRIB_RAID1
; break;
2320 case 10: mpb
->attributes
|= MPB_ATTRIB_RAID10
; break;
2321 case 5: mpb
->attributes
|= MPB_ATTRIB_RAID5
; break;
2324 if (map
->num_members
>= 5)
2325 version
= MPB_VERSION_5OR6_DISK_ARRAY
;
2326 else if (dev
->status
== DEV_CLONE_N_GO
)
2327 version
= MPB_VERSION_CNG
;
2328 else if (get_imsm_raid_level(map
) == 5)
2329 version
= MPB_VERSION_RAID5
;
2330 else if (map
->num_members
>= 3)
2331 version
= MPB_VERSION_3OR4_DISK_ARRAY
;
2332 else if (get_imsm_raid_level(map
) == 1)
2333 version
= MPB_VERSION_RAID1
;
2335 version
= MPB_VERSION_RAID0
;
2337 strcpy(((char *) mpb
->sig
) + strlen(MPB_SIGNATURE
), version
);
2341 static int init_super_imsm_volume(struct supertype
*st
, mdu_array_info_t
*info
,
2342 unsigned long long size
, char *name
,
2343 char *homehost
, int *uuid
)
2345 /* We are creating a volume inside a pre-existing container.
2346 * so st->sb is already set.
2348 struct intel_super
*super
= st
->sb
;
2349 struct imsm_super
*mpb
= super
->anchor
;
2350 struct intel_dev
*dv
;
2351 struct imsm_dev
*dev
;
2352 struct imsm_vol
*vol
;
2353 struct imsm_map
*map
;
2354 int idx
= mpb
->num_raid_devs
;
2356 unsigned long long array_blocks
;
2357 size_t size_old
, size_new
;
2358 __u32 num_data_stripes
;
2360 if (super
->orom
&& mpb
->num_raid_devs
>= super
->orom
->vpa
) {
2361 fprintf(stderr
, Name
": This imsm-container already has the "
2362 "maximum of %d volumes\n", super
->orom
->vpa
);
2366 /* ensure the mpb is large enough for the new data */
2367 size_old
= __le32_to_cpu(mpb
->mpb_size
);
2368 size_new
= disks_to_mpb_size(info
->nr_disks
);
2369 if (size_new
> size_old
) {
2371 size_t size_round
= ROUND_UP(size_new
, 512);
2373 if (posix_memalign(&mpb_new
, 512, size_round
) != 0) {
2374 fprintf(stderr
, Name
": could not allocate new mpb\n");
2377 memcpy(mpb_new
, mpb
, size_old
);
2380 super
->anchor
= mpb_new
;
2381 mpb
->mpb_size
= __cpu_to_le32(size_new
);
2382 memset(mpb_new
+ size_old
, 0, size_round
- size_old
);
2384 super
->current_vol
= idx
;
2385 /* when creating the first raid device in this container set num_disks
2386 * to zero, i.e. delete this spare and add raid member devices in
2387 * add_to_super_imsm_volume()
2389 if (super
->current_vol
== 0)
2392 for (i
= 0; i
< super
->current_vol
; i
++) {
2393 dev
= get_imsm_dev(super
, i
);
2394 if (strncmp((char *) dev
->volume
, name
,
2395 MAX_RAID_SERIAL_LEN
) == 0) {
2396 fprintf(stderr
, Name
": '%s' is already defined for this container\n",
2402 sprintf(st
->subarray
, "%d", idx
);
2403 dv
= malloc(sizeof(*dv
));
2405 fprintf(stderr
, Name
": failed to allocate device list entry\n");
2408 dev
= malloc(sizeof(*dev
) + sizeof(__u32
) * (info
->raid_disks
- 1));
2411 fprintf(stderr
, Name
": could not allocate raid device\n");
2414 strncpy((char *) dev
->volume
, name
, MAX_RAID_SERIAL_LEN
);
2415 if (info
->level
== 1)
2416 array_blocks
= info_to_blocks_per_member(info
);
2418 array_blocks
= calc_array_size(info
->level
, info
->raid_disks
,
2419 info
->layout
, info
->chunk_size
,
2421 /* round array size down to closest MB */
2422 array_blocks
= (array_blocks
>> SECT_PER_MB_SHIFT
) << SECT_PER_MB_SHIFT
;
2424 dev
->size_low
= __cpu_to_le32((__u32
) array_blocks
);
2425 dev
->size_high
= __cpu_to_le32((__u32
) (array_blocks
>> 32));
2426 dev
->status
= __cpu_to_le32(0);
2427 dev
->reserved_blocks
= __cpu_to_le32(0);
2429 vol
->migr_state
= 0;
2430 set_migr_type(dev
, MIGR_INIT
);
2432 vol
->curr_migr_unit
= 0;
2433 map
= get_imsm_map(dev
, 0);
2434 map
->pba_of_lba0
= __cpu_to_le32(super
->create_offset
);
2435 map
->blocks_per_member
= __cpu_to_le32(info_to_blocks_per_member(info
));
2436 map
->blocks_per_strip
= __cpu_to_le16(info_to_blocks_per_strip(info
));
2437 map
->failed_disk_num
= ~0;
2438 map
->map_state
= info
->level
? IMSM_T_STATE_UNINITIALIZED
:
2439 IMSM_T_STATE_NORMAL
;
2442 if (info
->level
== 1 && info
->raid_disks
> 2) {
2443 fprintf(stderr
, Name
": imsm does not support more than 2 disks"
2444 "in a raid1 volume\n");
2448 map
->raid_level
= info
->level
;
2449 if (info
->level
== 10) {
2450 map
->raid_level
= 1;
2451 map
->num_domains
= info
->raid_disks
/ 2;
2452 } else if (info
->level
== 1)
2453 map
->num_domains
= info
->raid_disks
;
2455 map
->num_domains
= 1;
2457 num_data_stripes
= info_to_num_data_stripes(info
, map
->num_domains
);
2458 map
->num_data_stripes
= __cpu_to_le32(num_data_stripes
);
2460 map
->num_members
= info
->raid_disks
;
2461 for (i
= 0; i
< map
->num_members
; i
++) {
2462 /* initialized in add_to_super */
2463 set_imsm_ord_tbl_ent(map
, i
, 0);
2465 mpb
->num_raid_devs
++;
2468 dv
->index
= super
->current_vol
;
2469 dv
->next
= super
->devlist
;
2470 super
->devlist
= dv
;
2472 imsm_update_version_info(super
);
2477 static int init_super_imsm(struct supertype
*st
, mdu_array_info_t
*info
,
2478 unsigned long long size
, char *name
,
2479 char *homehost
, int *uuid
)
2481 /* This is primarily called by Create when creating a new array.
2482 * We will then get add_to_super called for each component, and then
2483 * write_init_super called to write it out to each device.
2484 * For IMSM, Create can create on fresh devices or on a pre-existing
2486 * To create on a pre-existing array a different method will be called.
2487 * This one is just for fresh drives.
2489 struct intel_super
*super
;
2490 struct imsm_super
*mpb
;
2499 return init_super_imsm_volume(st
, info
, size
, name
, homehost
,
2502 super
= alloc_super(1);
2505 mpb_size
= disks_to_mpb_size(info
->nr_disks
);
2506 if (posix_memalign(&super
->buf
, 512, mpb_size
) != 0) {
2511 memset(mpb
, 0, mpb_size
);
2513 mpb
->attributes
= MPB_ATTRIB_CHECKSUM_VERIFY
;
2515 version
= (char *) mpb
->sig
;
2516 strcpy(version
, MPB_SIGNATURE
);
2517 version
+= strlen(MPB_SIGNATURE
);
2518 strcpy(version
, MPB_VERSION_RAID0
);
2519 mpb
->mpb_size
= mpb_size
;
2526 static int add_to_super_imsm_volume(struct supertype
*st
, mdu_disk_info_t
*dk
,
2527 int fd
, char *devname
)
2529 struct intel_super
*super
= st
->sb
;
2530 struct imsm_super
*mpb
= super
->anchor
;
2532 struct imsm_dev
*dev
;
2533 struct imsm_map
*map
;
2535 dev
= get_imsm_dev(super
, super
->current_vol
);
2536 map
= get_imsm_map(dev
, 0);
2538 if (! (dk
->state
& (1<<MD_DISK_SYNC
))) {
2539 fprintf(stderr
, Name
": %s: Cannot add spare devices to IMSM volume\n",
2545 /* we're doing autolayout so grab the pre-marked (in
2546 * validate_geometry) raid_disk
2548 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
2549 if (dl
->raiddisk
== dk
->raid_disk
)
2552 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
2553 if (dl
->major
== dk
->major
&&
2554 dl
->minor
== dk
->minor
)
2559 fprintf(stderr
, Name
": %s is not a member of the same container\n", devname
);
2563 /* add a pristine spare to the metadata */
2564 if (dl
->index
< 0) {
2565 dl
->index
= super
->anchor
->num_disks
;
2566 super
->anchor
->num_disks
++;
2568 set_imsm_ord_tbl_ent(map
, dk
->number
, dl
->index
);
2569 dl
->disk
.status
= CONFIGURED_DISK
;
2571 /* if we are creating the first raid device update the family number */
2572 if (super
->current_vol
== 0) {
2574 struct imsm_dev
*_dev
= __get_imsm_dev(mpb
, 0);
2575 struct imsm_disk
*_disk
= __get_imsm_disk(mpb
, dl
->index
);
2580 sum
+= __gen_imsm_checksum(mpb
);
2581 mpb
->family_num
= __cpu_to_le32(sum
);
2582 mpb
->orig_family_num
= mpb
->family_num
;
2588 static int add_to_super_imsm(struct supertype
*st
, mdu_disk_info_t
*dk
,
2589 int fd
, char *devname
)
2591 struct intel_super
*super
= st
->sb
;
2593 unsigned long long size
;
2598 /* if we are on an RAID enabled platform check that the disk is
2599 * attached to the raid controller
2601 if (super
->hba
&& !disk_attached_to_hba(fd
, super
->hba
)) {
2603 Name
": %s is not attached to the raid controller: %s\n",
2604 devname
? : "disk", super
->hba
);
2608 if (super
->current_vol
>= 0)
2609 return add_to_super_imsm_volume(st
, dk
, fd
, devname
);
2612 dd
= malloc(sizeof(*dd
));
2615 Name
": malloc failed %s:%d.\n", __func__
, __LINE__
);
2618 memset(dd
, 0, sizeof(*dd
));
2619 dd
->major
= major(stb
.st_rdev
);
2620 dd
->minor
= minor(stb
.st_rdev
);
2622 dd
->devname
= devname
? strdup(devname
) : NULL
;
2625 rv
= imsm_read_serial(fd
, devname
, dd
->serial
);
2628 Name
": failed to retrieve scsi serial, aborting\n");
2633 get_dev_size(fd
, NULL
, &size
);
2635 serialcpy(dd
->disk
.serial
, dd
->serial
);
2636 dd
->disk
.total_blocks
= __cpu_to_le32(size
);
2637 dd
->disk
.status
= SPARE_DISK
;
2638 if (sysfs_disk_to_scsi_id(fd
, &id
) == 0)
2639 dd
->disk
.scsi_id
= __cpu_to_le32(id
);
2641 dd
->disk
.scsi_id
= __cpu_to_le32(0);
2643 if (st
->update_tail
) {
2644 dd
->next
= super
->add
;
2647 dd
->next
= super
->disks
;
2654 static int store_imsm_mpb(int fd
, struct intel_super
*super
);
2656 /* spare records have their own family number and do not have any defined raid
2659 static int write_super_imsm_spares(struct intel_super
*super
, int doclose
)
2661 struct imsm_super mpb_save
;
2662 struct imsm_super
*mpb
= super
->anchor
;
2667 mpb
->num_raid_devs
= 0;
2669 mpb
->mpb_size
= sizeof(struct imsm_super
);
2670 mpb
->generation_num
= __cpu_to_le32(1UL);
2672 for (d
= super
->disks
; d
; d
= d
->next
) {
2676 mpb
->disk
[0] = d
->disk
;
2677 sum
= __gen_imsm_checksum(mpb
);
2678 mpb
->family_num
= __cpu_to_le32(sum
);
2679 mpb
->orig_family_num
= 0;
2680 sum
= __gen_imsm_checksum(mpb
);
2681 mpb
->check_sum
= __cpu_to_le32(sum
);
2683 if (store_imsm_mpb(d
->fd
, super
)) {
2684 fprintf(stderr
, "%s: failed for device %d:%d %s\n",
2685 __func__
, d
->major
, d
->minor
, strerror(errno
));
2699 static int write_super_imsm(struct intel_super
*super
, int doclose
)
2701 struct imsm_super
*mpb
= super
->anchor
;
2707 __u32 mpb_size
= sizeof(struct imsm_super
) - sizeof(struct imsm_disk
);
2709 /* 'generation' is incremented everytime the metadata is written */
2710 generation
= __le32_to_cpu(mpb
->generation_num
);
2712 mpb
->generation_num
= __cpu_to_le32(generation
);
2714 /* fix up cases where previous mdadm releases failed to set
2717 if (mpb
->orig_family_num
== 0)
2718 mpb
->orig_family_num
= mpb
->family_num
;
2720 mpb_size
+= sizeof(struct imsm_disk
) * mpb
->num_disks
;
2721 for (d
= super
->disks
; d
; d
= d
->next
) {
2725 mpb
->disk
[d
->index
] = d
->disk
;
2727 for (d
= super
->missing
; d
; d
= d
->next
)
2728 mpb
->disk
[d
->index
] = d
->disk
;
2730 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
2731 struct imsm_dev
*dev
= __get_imsm_dev(mpb
, i
);
2733 imsm_copy_dev(dev
, get_imsm_dev(super
, i
));
2734 mpb_size
+= sizeof_imsm_dev(dev
, 0);
2736 mpb_size
+= __le32_to_cpu(mpb
->bbm_log_size
);
2737 mpb
->mpb_size
= __cpu_to_le32(mpb_size
);
2739 /* recalculate checksum */
2740 sum
= __gen_imsm_checksum(mpb
);
2741 mpb
->check_sum
= __cpu_to_le32(sum
);
2743 /* write the mpb for disks that compose raid devices */
2744 for (d
= super
->disks
; d
; d
= d
->next
) {
2747 if (store_imsm_mpb(d
->fd
, super
))
2748 fprintf(stderr
, "%s: failed for device %d:%d %s\n",
2749 __func__
, d
->major
, d
->minor
, strerror(errno
));
2757 return write_super_imsm_spares(super
, doclose
);
2763 static int create_array(struct supertype
*st
, int dev_idx
)
2766 struct imsm_update_create_array
*u
;
2767 struct intel_super
*super
= st
->sb
;
2768 struct imsm_dev
*dev
= get_imsm_dev(super
, dev_idx
);
2769 struct imsm_map
*map
= get_imsm_map(dev
, 0);
2770 struct disk_info
*inf
;
2771 struct imsm_disk
*disk
;
2774 len
= sizeof(*u
) - sizeof(*dev
) + sizeof_imsm_dev(dev
, 0) +
2775 sizeof(*inf
) * map
->num_members
;
2778 fprintf(stderr
, "%s: failed to allocate update buffer\n",
2783 u
->type
= update_create_array
;
2784 u
->dev_idx
= dev_idx
;
2785 imsm_copy_dev(&u
->dev
, dev
);
2786 inf
= get_disk_info(u
);
2787 for (i
= 0; i
< map
->num_members
; i
++) {
2788 int idx
= get_imsm_disk_idx(dev
, i
);
2790 disk
= get_imsm_disk(super
, idx
);
2791 serialcpy(inf
[i
].serial
, disk
->serial
);
2793 append_metadata_update(st
, u
, len
);
2798 static int _add_disk(struct supertype
*st
)
2800 struct intel_super
*super
= st
->sb
;
2802 struct imsm_update_add_disk
*u
;
2810 fprintf(stderr
, "%s: failed to allocate update buffer\n",
2815 u
->type
= update_add_disk
;
2816 append_metadata_update(st
, u
, len
);
2821 static int write_init_super_imsm(struct supertype
*st
)
2823 struct intel_super
*super
= st
->sb
;
2824 int current_vol
= super
->current_vol
;
2826 /* we are done with current_vol reset it to point st at the container */
2827 super
->current_vol
= -1;
2829 if (st
->update_tail
) {
2830 /* queue the recently created array / added disk
2831 * as a metadata update */
2835 /* determine if we are creating a volume or adding a disk */
2836 if (current_vol
< 0) {
2837 /* in the add disk case we are running in mdmon
2838 * context, so don't close fd's
2840 return _add_disk(st
);
2842 rv
= create_array(st
, current_vol
);
2844 for (d
= super
->disks
; d
; d
= d
->next
) {
2851 return write_super_imsm(st
->sb
, 1);
2855 static int store_zero_imsm(struct supertype
*st
, int fd
)
2857 unsigned long long dsize
;
2860 get_dev_size(fd
, NULL
, &dsize
);
2862 /* first block is stored on second to last sector of the disk */
2863 if (lseek64(fd
, dsize
- (512 * 2), SEEK_SET
) < 0)
2866 if (posix_memalign(&buf
, 512, 512) != 0)
2869 memset(buf
, 0, 512);
2870 if (write(fd
, buf
, 512) != 512)
2875 static int imsm_bbm_log_size(struct imsm_super
*mpb
)
2877 return __le32_to_cpu(mpb
->bbm_log_size
);
2881 static int validate_geometry_imsm_container(struct supertype
*st
, int level
,
2882 int layout
, int raiddisks
, int chunk
,
2883 unsigned long long size
, char *dev
,
2884 unsigned long long *freesize
,
2888 unsigned long long ldsize
;
2889 const struct imsm_orom
*orom
;
2891 if (level
!= LEVEL_CONTAINER
)
2896 if (check_env("IMSM_NO_PLATFORM"))
2899 orom
= find_imsm_orom();
2900 if (orom
&& raiddisks
> orom
->tds
) {
2902 fprintf(stderr
, Name
": %d exceeds maximum number of"
2903 " platform supported disks: %d\n",
2904 raiddisks
, orom
->tds
);
2908 fd
= open(dev
, O_RDONLY
|O_EXCL
, 0);
2911 fprintf(stderr
, Name
": imsm: Cannot open %s: %s\n",
2912 dev
, strerror(errno
));
2915 if (!get_dev_size(fd
, dev
, &ldsize
)) {
2921 *freesize
= avail_size_imsm(st
, ldsize
>> 9);
2926 static unsigned long long find_size(struct extent
*e
, int *idx
, int num_extents
)
2928 const unsigned long long base_start
= e
[*idx
].start
;
2929 unsigned long long end
= base_start
+ e
[*idx
].size
;
2932 if (base_start
== end
)
2936 for (i
= *idx
; i
< num_extents
; i
++) {
2937 /* extend overlapping extents */
2938 if (e
[i
].start
>= base_start
&&
2939 e
[i
].start
<= end
) {
2942 if (e
[i
].start
+ e
[i
].size
> end
)
2943 end
= e
[i
].start
+ e
[i
].size
;
2944 } else if (e
[i
].start
> end
) {
2950 return end
- base_start
;
2953 static unsigned long long merge_extents(struct intel_super
*super
, int sum_extents
)
2955 /* build a composite disk with all known extents and generate a new
2956 * 'maxsize' given the "all disks in an array must share a common start
2957 * offset" constraint
2959 struct extent
*e
= calloc(sum_extents
, sizeof(*e
));
2963 unsigned long long pos
;
2964 unsigned long long start
= 0;
2965 unsigned long long maxsize
;
2966 unsigned long reserve
;
2969 return ~0ULL; /* error */
2971 /* coalesce and sort all extents. also, check to see if we need to
2972 * reserve space between member arrays
2975 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
2978 for (i
= 0; i
< dl
->extent_cnt
; i
++)
2981 qsort(e
, sum_extents
, sizeof(*e
), cmp_extent
);
2986 while (i
< sum_extents
) {
2987 e
[j
].start
= e
[i
].start
;
2988 e
[j
].size
= find_size(e
, &i
, sum_extents
);
2990 if (e
[j
-1].size
== 0)
2999 unsigned long long esize
;
3001 esize
= e
[i
].start
- pos
;
3002 if (esize
>= maxsize
) {
3007 pos
= e
[i
].start
+ e
[i
].size
;
3009 } while (e
[i
-1].size
);
3012 if (start_extent
> 0)
3013 reserve
= IMSM_RESERVED_SECTORS
; /* gap between raid regions */
3017 if (maxsize
< reserve
)
3020 super
->create_offset
= ~((__u32
) 0);
3021 if (start
+ reserve
> super
->create_offset
)
3022 return ~0ULL; /* start overflows create_offset */
3023 super
->create_offset
= start
+ reserve
;
3025 return maxsize
- reserve
;
3028 static int is_raid_level_supported(const struct imsm_orom
*orom
, int level
, int raiddisks
)
3030 if (level
< 0 || level
== 6 || level
== 4)
3033 /* if we have an orom prevent invalid raid levels */
3036 case 0: return imsm_orom_has_raid0(orom
);
3039 return imsm_orom_has_raid1e(orom
);
3040 return imsm_orom_has_raid1(orom
) && raiddisks
== 2;
3041 case 10: return imsm_orom_has_raid10(orom
) && raiddisks
== 4;
3042 case 5: return imsm_orom_has_raid5(orom
) && raiddisks
> 2;
3045 return 1; /* not on an Intel RAID platform so anything goes */
3050 #define pr_vrb(fmt, arg...) (void) (verbose && fprintf(stderr, Name fmt, ##arg))
3051 /* validate_geometry_imsm_volume - lifted from validate_geometry_ddf_bvd
3052 * FIX ME add ahci details
3054 static int validate_geometry_imsm_volume(struct supertype
*st
, int level
,
3055 int layout
, int raiddisks
, int chunk
,
3056 unsigned long long size
, char *dev
,
3057 unsigned long long *freesize
,
3061 struct intel_super
*super
= st
->sb
;
3062 struct imsm_super
*mpb
= super
->anchor
;
3064 unsigned long long pos
= 0;
3065 unsigned long long maxsize
;
3069 /* We must have the container info already read in. */
3073 if (!is_raid_level_supported(super
->orom
, level
, raiddisks
)) {
3074 pr_vrb(": platform does not support raid%d with %d disk%s\n",
3075 level
, raiddisks
, raiddisks
> 1 ? "s" : "");
3078 if (super
->orom
&& level
!= 1 &&
3079 !imsm_orom_has_chunk(super
->orom
, chunk
)) {
3080 pr_vrb(": platform does not support a chunk size of: %d\n", chunk
);
3083 if (layout
!= imsm_level_to_layout(level
)) {
3085 pr_vrb(": imsm raid 5 only supports the left-asymmetric layout\n");
3086 else if (level
== 10)
3087 pr_vrb(": imsm raid 10 only supports the n2 layout\n");
3089 pr_vrb(": imsm unknown layout %#x for this raid level %d\n",
3095 /* General test: make sure there is space for
3096 * 'raiddisks' device extents of size 'size' at a given
3099 unsigned long long minsize
= size
;
3100 unsigned long long start_offset
= ~0ULL;
3103 minsize
= MPB_SECTOR_CNT
+ IMSM_RESERVED_SECTORS
;
3104 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
3109 e
= get_extents(super
, dl
);
3112 unsigned long long esize
;
3113 esize
= e
[i
].start
- pos
;
3114 if (esize
>= minsize
)
3116 if (found
&& start_offset
== ~0ULL) {
3119 } else if (found
&& pos
!= start_offset
) {
3123 pos
= e
[i
].start
+ e
[i
].size
;
3125 } while (e
[i
-1].size
);
3130 if (dcnt
< raiddisks
) {
3132 fprintf(stderr
, Name
": imsm: Not enough "
3133 "devices with space for this array "
3141 /* This device must be a member of the set */
3142 if (stat(dev
, &stb
) < 0)
3144 if ((S_IFMT
& stb
.st_mode
) != S_IFBLK
)
3146 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
3147 if (dl
->major
== major(stb
.st_rdev
) &&
3148 dl
->minor
== minor(stb
.st_rdev
))
3153 fprintf(stderr
, Name
": %s is not in the "
3154 "same imsm set\n", dev
);
3156 } else if (super
->orom
&& dl
->index
< 0 && mpb
->num_raid_devs
) {
3157 /* If a volume is present then the current creation attempt
3158 * cannot incorporate new spares because the orom may not
3159 * understand this configuration (all member disks must be
3160 * members of each array in the container).
3162 fprintf(stderr
, Name
": %s is a spare and a volume"
3163 " is already defined for this container\n", dev
);
3164 fprintf(stderr
, Name
": The option-rom requires all member"
3165 " disks to be a member of all volumes\n");
3169 /* retrieve the largest free space block */
3170 e
= get_extents(super
, dl
);
3175 unsigned long long esize
;
3177 esize
= e
[i
].start
- pos
;
3178 if (esize
>= maxsize
)
3180 pos
= e
[i
].start
+ e
[i
].size
;
3182 } while (e
[i
-1].size
);
3187 fprintf(stderr
, Name
": unable to determine free space for: %s\n",
3191 if (maxsize
< size
) {
3193 fprintf(stderr
, Name
": %s not enough space (%llu < %llu)\n",
3194 dev
, maxsize
, size
);
3198 /* count total number of extents for merge */
3200 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
3202 i
+= dl
->extent_cnt
;
3204 maxsize
= merge_extents(super
, i
);
3205 if (maxsize
< size
) {
3207 fprintf(stderr
, Name
": not enough space after merge (%llu < %llu)\n",
3210 } else if (maxsize
== ~0ULL) {
3212 fprintf(stderr
, Name
": failed to merge %d extents\n", i
);
3216 *freesize
= maxsize
;
3221 static int reserve_space(struct supertype
*st
, int raiddisks
,
3222 unsigned long long size
, int chunk
,
3223 unsigned long long *freesize
)
3225 struct intel_super
*super
= st
->sb
;
3226 struct imsm_super
*mpb
= super
->anchor
;
3231 unsigned long long maxsize
;
3232 unsigned long long minsize
;
3236 /* find the largest common start free region of the possible disks */
3240 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
3246 /* don't activate new spares if we are orom constrained
3247 * and there is already a volume active in the container
3249 if (super
->orom
&& dl
->index
< 0 && mpb
->num_raid_devs
)
3252 e
= get_extents(super
, dl
);
3255 for (i
= 1; e
[i
-1].size
; i
++)
3263 maxsize
= merge_extents(super
, extent_cnt
);
3268 if (cnt
< raiddisks
||
3269 (super
->orom
&& used
&& used
!= raiddisks
) ||
3270 maxsize
< minsize
) {
3271 fprintf(stderr
, Name
": not enough devices with space to create array.\n");
3272 return 0; /* No enough free spaces large enough */
3284 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
3286 dl
->raiddisk
= cnt
++;
3293 static int validate_geometry_imsm(struct supertype
*st
, int level
, int layout
,
3294 int raiddisks
, int chunk
, unsigned long long size
,
3295 char *dev
, unsigned long long *freesize
,
3301 /* if given unused devices create a container
3302 * if given given devices in a container create a member volume
3304 if (level
== LEVEL_CONTAINER
) {
3305 /* Must be a fresh device to add to a container */
3306 return validate_geometry_imsm_container(st
, level
, layout
,
3307 raiddisks
, chunk
, size
,
3313 if (st
->sb
&& freesize
) {
3314 /* we are being asked to automatically layout a
3315 * new volume based on the current contents of
3316 * the container. If the the parameters can be
3317 * satisfied reserve_space will record the disks,
3318 * start offset, and size of the volume to be
3319 * created. add_to_super and getinfo_super
3320 * detect when autolayout is in progress.
3322 return reserve_space(st
, raiddisks
, size
, chunk
, freesize
);
3327 /* creating in a given container */
3328 return validate_geometry_imsm_volume(st
, level
, layout
,
3329 raiddisks
, chunk
, size
,
3330 dev
, freesize
, verbose
);
3333 /* limit creation to the following levels */
3345 /* This device needs to be a device in an 'imsm' container */
3346 fd
= open(dev
, O_RDONLY
|O_EXCL
, 0);
3350 Name
": Cannot create this array on device %s\n",
3355 if (errno
!= EBUSY
|| (fd
= open(dev
, O_RDONLY
, 0)) < 0) {
3357 fprintf(stderr
, Name
": Cannot open %s: %s\n",
3358 dev
, strerror(errno
));
3361 /* Well, it is in use by someone, maybe an 'imsm' container. */
3362 cfd
= open_container(fd
);
3366 fprintf(stderr
, Name
": Cannot use %s: It is busy\n",
3370 sra
= sysfs_read(cfd
, 0, GET_VERSION
);
3372 if (sra
&& sra
->array
.major_version
== -1 &&
3373 strcmp(sra
->text_version
, "imsm") == 0) {
3374 /* This is a member of a imsm container. Load the container
3375 * and try to create a volume
3377 struct intel_super
*super
;
3379 if (load_super_imsm_all(st
, cfd
, (void **) &super
, NULL
, 1) == 0) {
3381 st
->container_dev
= fd2devnum(cfd
);
3383 return validate_geometry_imsm_volume(st
, level
, layout
,
3389 } else /* may belong to another container */
3394 #endif /* MDASSEMBLE */
3396 static struct mdinfo
*container_content_imsm(struct supertype
*st
)
3398 /* Given a container loaded by load_super_imsm_all,
3399 * extract information about all the arrays into
3402 * For each imsm_dev create an mdinfo, fill it in,
3403 * then look for matching devices in super->disks
3404 * and create appropriate device mdinfo.
3406 struct intel_super
*super
= st
->sb
;
3407 struct imsm_super
*mpb
= super
->anchor
;
3408 struct mdinfo
*rest
= NULL
;
3411 /* do not assemble arrays that might have bad blocks */
3412 if (imsm_bbm_log_size(super
->anchor
)) {
3413 fprintf(stderr
, Name
": BBM log found in metadata. "
3414 "Cannot activate array(s).\n");
3418 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
3419 struct imsm_dev
*dev
= get_imsm_dev(super
, i
);
3420 struct imsm_map
*map
= get_imsm_map(dev
, 0);
3421 struct mdinfo
*this;
3424 /* do not publish arrays that are in the middle of an
3425 * unsupported migration
3427 if (dev
->vol
.migr_state
&&
3428 (migr_type(dev
) == MIGR_GEN_MIGR
||
3429 migr_type(dev
) == MIGR_STATE_CHANGE
)) {
3430 fprintf(stderr
, Name
": cannot assemble volume '%.16s':"
3431 " unsupported migration in progress\n",
3436 this = malloc(sizeof(*this));
3437 memset(this, 0, sizeof(*this));
3440 super
->current_vol
= i
;
3441 getinfo_super_imsm_volume(st
, this);
3442 for (slot
= 0 ; slot
< map
->num_members
; slot
++) {
3443 struct mdinfo
*info_d
;
3451 idx
= get_imsm_disk_idx(dev
, slot
);
3452 ord
= get_imsm_ord_tbl_ent(dev
, slot
);
3453 for (d
= super
->disks
; d
; d
= d
->next
)
3454 if (d
->index
== idx
)
3460 s
= d
? d
->disk
.status
: 0;
3461 if (s
& FAILED_DISK
)
3463 if (ord
& IMSM_ORD_REBUILD
)
3467 * if we skip some disks the array will be assmebled degraded;
3468 * reset resync start to avoid a dirty-degraded situation
3470 * FIXME handle dirty degraded
3472 if (skip
&& !dev
->vol
.dirty
)
3473 this->resync_start
= ~0ULL;
3477 info_d
= malloc(sizeof(*info_d
));
3479 fprintf(stderr
, Name
": failed to allocate disk"
3480 " for volume %.16s\n", dev
->volume
);
3485 memset(info_d
, 0, sizeof(*info_d
));
3486 info_d
->next
= this->devs
;
3487 this->devs
= info_d
;
3489 info_d
->disk
.number
= d
->index
;
3490 info_d
->disk
.major
= d
->major
;
3491 info_d
->disk
.minor
= d
->minor
;
3492 info_d
->disk
.raid_disk
= slot
;
3494 this->array
.working_disks
++;
3496 info_d
->events
= __le32_to_cpu(mpb
->generation_num
);
3497 info_d
->data_offset
= __le32_to_cpu(map
->pba_of_lba0
);
3498 info_d
->component_size
= __le32_to_cpu(map
->blocks_per_member
);
3500 strcpy(info_d
->name
, d
->devname
);
3510 static int imsm_open_new(struct supertype
*c
, struct active_array
*a
,
3513 struct intel_super
*super
= c
->sb
;
3514 struct imsm_super
*mpb
= super
->anchor
;
3516 if (atoi(inst
) >= mpb
->num_raid_devs
) {
3517 fprintf(stderr
, "%s: subarry index %d, out of range\n",
3518 __func__
, atoi(inst
));
3522 dprintf("imsm: open_new %s\n", inst
);
3523 a
->info
.container_member
= atoi(inst
);
3527 static __u8
imsm_check_degraded(struct intel_super
*super
, struct imsm_dev
*dev
, int failed
)
3529 struct imsm_map
*map
= get_imsm_map(dev
, 0);
3532 return map
->map_state
== IMSM_T_STATE_UNINITIALIZED
?
3533 IMSM_T_STATE_UNINITIALIZED
: IMSM_T_STATE_NORMAL
;
3535 switch (get_imsm_raid_level(map
)) {
3537 return IMSM_T_STATE_FAILED
;
3540 if (failed
< map
->num_members
)
3541 return IMSM_T_STATE_DEGRADED
;
3543 return IMSM_T_STATE_FAILED
;
3548 * check to see if any mirrors have failed, otherwise we
3549 * are degraded. Even numbered slots are mirrored on
3553 /* gcc -Os complains that this is unused */
3554 int insync
= insync
;
3556 for (i
= 0; i
< map
->num_members
; i
++) {
3557 __u32 ord
= get_imsm_ord_tbl_ent(dev
, i
);
3558 int idx
= ord_to_idx(ord
);
3559 struct imsm_disk
*disk
;
3561 /* reset the potential in-sync count on even-numbered
3562 * slots. num_copies is always 2 for imsm raid10
3567 disk
= get_imsm_disk(super
, idx
);
3568 if (!disk
|| disk
->status
& FAILED_DISK
||
3569 ord
& IMSM_ORD_REBUILD
)
3572 /* no in-sync disks left in this mirror the
3576 return IMSM_T_STATE_FAILED
;
3579 return IMSM_T_STATE_DEGRADED
;
3583 return IMSM_T_STATE_DEGRADED
;
3585 return IMSM_T_STATE_FAILED
;
3591 return map
->map_state
;
3594 static int imsm_count_failed(struct intel_super
*super
, struct imsm_dev
*dev
)
3598 struct imsm_disk
*disk
;
3599 struct imsm_map
*map
= get_imsm_map(dev
, 0);
3600 struct imsm_map
*prev
= get_imsm_map(dev
, dev
->vol
.migr_state
);
3604 /* at the beginning of migration we set IMSM_ORD_REBUILD on
3605 * disks that are being rebuilt. New failures are recorded to
3606 * map[0]. So we look through all the disks we started with and
3607 * see if any failures are still present, or if any new ones
3610 * FIXME add support for online capacity expansion and
3611 * raid-level-migration
3613 for (i
= 0; i
< prev
->num_members
; i
++) {
3614 ord
= __le32_to_cpu(prev
->disk_ord_tbl
[i
]);
3615 ord
|= __le32_to_cpu(map
->disk_ord_tbl
[i
]);
3616 idx
= ord_to_idx(ord
);
3618 disk
= get_imsm_disk(super
, idx
);
3619 if (!disk
|| disk
->status
& FAILED_DISK
||
3620 ord
& IMSM_ORD_REBUILD
)
3627 static int is_resyncing(struct imsm_dev
*dev
)
3629 struct imsm_map
*migr_map
;
3631 if (!dev
->vol
.migr_state
)
3634 if (migr_type(dev
) == MIGR_INIT
||
3635 migr_type(dev
) == MIGR_REPAIR
)
3638 migr_map
= get_imsm_map(dev
, 1);
3640 if (migr_map
->map_state
== IMSM_T_STATE_NORMAL
)
3646 static int is_rebuilding(struct imsm_dev
*dev
)
3648 struct imsm_map
*migr_map
;
3650 if (!dev
->vol
.migr_state
)
3653 if (migr_type(dev
) != MIGR_REBUILD
)
3656 migr_map
= get_imsm_map(dev
, 1);
3658 if (migr_map
->map_state
== IMSM_T_STATE_DEGRADED
)
3664 /* return true if we recorded new information */
3665 static int mark_failure(struct imsm_dev
*dev
, struct imsm_disk
*disk
, int idx
)
3669 struct imsm_map
*map
;
3671 /* new failures are always set in map[0] */
3672 map
= get_imsm_map(dev
, 0);
3674 slot
= get_imsm_disk_slot(map
, idx
);
3678 ord
= __le32_to_cpu(map
->disk_ord_tbl
[slot
]);
3679 if ((disk
->status
& FAILED_DISK
) && (ord
& IMSM_ORD_REBUILD
))
3682 disk
->status
|= FAILED_DISK
;
3683 disk
->status
&= ~CONFIGURED_DISK
;
3684 set_imsm_ord_tbl_ent(map
, slot
, idx
| IMSM_ORD_REBUILD
);
3685 if (~map
->failed_disk_num
== 0)
3686 map
->failed_disk_num
= slot
;
3690 static void mark_missing(struct imsm_dev
*dev
, struct imsm_disk
*disk
, int idx
)
3692 mark_failure(dev
, disk
, idx
);
3694 if (disk
->scsi_id
== __cpu_to_le32(~(__u32
)0))
3697 disk
->scsi_id
= __cpu_to_le32(~(__u32
)0);
3698 memmove(&disk
->serial
[0], &disk
->serial
[1], MAX_RAID_SERIAL_LEN
- 1);
3701 /* Handle dirty -> clean transititions and resync. Degraded and rebuild
3702 * states are handled in imsm_set_disk() with one exception, when a
3703 * resync is stopped due to a new failure this routine will set the
3704 * 'degraded' state for the array.
3706 static int imsm_set_array_state(struct active_array
*a
, int consistent
)
3708 int inst
= a
->info
.container_member
;
3709 struct intel_super
*super
= a
->container
->sb
;
3710 struct imsm_dev
*dev
= get_imsm_dev(super
, inst
);
3711 struct imsm_map
*map
= get_imsm_map(dev
, 0);
3712 int failed
= imsm_count_failed(super
, dev
);
3713 __u8 map_state
= imsm_check_degraded(super
, dev
, failed
);
3715 /* before we activate this array handle any missing disks */
3716 if (consistent
== 2 && super
->missing
) {
3719 dprintf("imsm: mark missing\n");
3720 end_migration(dev
, map_state
);
3721 for (dl
= super
->missing
; dl
; dl
= dl
->next
)
3722 mark_missing(dev
, &dl
->disk
, dl
->index
);
3723 super
->updates_pending
++;
3726 if (consistent
== 2 &&
3727 (!is_resync_complete(a
) ||
3728 map_state
!= IMSM_T_STATE_NORMAL
||
3729 dev
->vol
.migr_state
))
3732 if (is_resync_complete(a
)) {
3733 /* complete intialization / resync,
3734 * recovery and interrupted recovery is completed in
3737 if (is_resyncing(dev
)) {
3738 dprintf("imsm: mark resync done\n");
3739 end_migration(dev
, map_state
);
3740 super
->updates_pending
++;
3742 } else if (!is_resyncing(dev
) && !failed
) {
3743 /* mark the start of the init process if nothing is failed */
3744 dprintf("imsm: mark resync start (%llu)\n", a
->resync_start
);
3745 if (map
->map_state
== IMSM_T_STATE_UNINITIALIZED
)
3746 migrate(dev
, IMSM_T_STATE_NORMAL
, MIGR_INIT
);
3748 migrate(dev
, IMSM_T_STATE_NORMAL
, MIGR_REPAIR
);
3749 super
->updates_pending
++;
3752 /* FIXME check if we can update curr_migr_unit from resync_start */
3754 /* mark dirty / clean */
3755 if (dev
->vol
.dirty
!= !consistent
) {
3756 dprintf("imsm: mark '%s' (%llu)\n",
3757 consistent
? "clean" : "dirty", a
->resync_start
);
3762 super
->updates_pending
++;
3767 static void imsm_set_disk(struct active_array
*a
, int n
, int state
)
3769 int inst
= a
->info
.container_member
;
3770 struct intel_super
*super
= a
->container
->sb
;
3771 struct imsm_dev
*dev
= get_imsm_dev(super
, inst
);
3772 struct imsm_map
*map
= get_imsm_map(dev
, 0);
3773 struct imsm_disk
*disk
;
3778 if (n
> map
->num_members
)
3779 fprintf(stderr
, "imsm: set_disk %d out of range 0..%d\n",
3780 n
, map
->num_members
- 1);
3785 dprintf("imsm: set_disk %d:%x\n", n
, state
);
3787 ord
= get_imsm_ord_tbl_ent(dev
, n
);
3788 disk
= get_imsm_disk(super
, ord_to_idx(ord
));
3790 /* check for new failures */
3791 if (state
& DS_FAULTY
) {
3792 if (mark_failure(dev
, disk
, ord_to_idx(ord
)))
3793 super
->updates_pending
++;
3796 /* check if in_sync */
3797 if (state
& DS_INSYNC
&& ord
& IMSM_ORD_REBUILD
&& is_rebuilding(dev
)) {
3798 struct imsm_map
*migr_map
= get_imsm_map(dev
, 1);
3800 set_imsm_ord_tbl_ent(migr_map
, n
, ord_to_idx(ord
));
3801 super
->updates_pending
++;
3804 failed
= imsm_count_failed(super
, dev
);
3805 map_state
= imsm_check_degraded(super
, dev
, failed
);
3807 /* check if recovery complete, newly degraded, or failed */
3808 if (map_state
== IMSM_T_STATE_NORMAL
&& is_rebuilding(dev
)) {
3809 end_migration(dev
, map_state
);
3810 map
= get_imsm_map(dev
, 0);
3811 map
->failed_disk_num
= ~0;
3812 super
->updates_pending
++;
3813 } else if (map_state
== IMSM_T_STATE_DEGRADED
&&
3814 map
->map_state
!= map_state
&&
3815 !dev
->vol
.migr_state
) {
3816 dprintf("imsm: mark degraded\n");
3817 map
->map_state
= map_state
;
3818 super
->updates_pending
++;
3819 } else if (map_state
== IMSM_T_STATE_FAILED
&&
3820 map
->map_state
!= map_state
) {
3821 dprintf("imsm: mark failed\n");
3822 end_migration(dev
, map_state
);
3823 super
->updates_pending
++;
3827 static int store_imsm_mpb(int fd
, struct intel_super
*super
)
3829 struct imsm_super
*mpb
= super
->anchor
;
3830 __u32 mpb_size
= __le32_to_cpu(mpb
->mpb_size
);
3831 unsigned long long dsize
;
3832 unsigned long long sectors
;
3834 get_dev_size(fd
, NULL
, &dsize
);
3836 if (mpb_size
> 512) {
3837 /* -1 to account for anchor */
3838 sectors
= mpb_sectors(mpb
) - 1;
3840 /* write the extended mpb to the sectors preceeding the anchor */
3841 if (lseek64(fd
, dsize
- (512 * (2 + sectors
)), SEEK_SET
) < 0)
3844 if (write(fd
, super
->buf
+ 512, 512 * sectors
) != 512 * sectors
)
3848 /* first block is stored on second to last sector of the disk */
3849 if (lseek64(fd
, dsize
- (512 * 2), SEEK_SET
) < 0)
3852 if (write(fd
, super
->buf
, 512) != 512)
3858 static void imsm_sync_metadata(struct supertype
*container
)
3860 struct intel_super
*super
= container
->sb
;
3862 if (!super
->updates_pending
)
3865 write_super_imsm(super
, 0);
3867 super
->updates_pending
= 0;
3870 static struct dl
*imsm_readd(struct intel_super
*super
, int idx
, struct active_array
*a
)
3872 struct imsm_dev
*dev
= get_imsm_dev(super
, a
->info
.container_member
);
3873 int i
= get_imsm_disk_idx(dev
, idx
);
3876 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
3880 if (dl
&& dl
->disk
.status
& FAILED_DISK
)
3884 dprintf("%s: found %x:%x\n", __func__
, dl
->major
, dl
->minor
);
3889 static struct dl
*imsm_add_spare(struct intel_super
*super
, int slot
,
3890 struct active_array
*a
, int activate_new
)
3892 struct imsm_dev
*dev
= get_imsm_dev(super
, a
->info
.container_member
);
3893 int idx
= get_imsm_disk_idx(dev
, slot
);
3894 struct imsm_super
*mpb
= super
->anchor
;
3895 struct imsm_map
*map
;
3896 unsigned long long pos
;
3905 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
3906 /* If in this array, skip */
3907 for (d
= a
->info
.devs
; d
; d
= d
->next
)
3908 if (d
->state_fd
>= 0 &&
3909 d
->disk
.major
== dl
->major
&&
3910 d
->disk
.minor
== dl
->minor
) {
3911 dprintf("%x:%x already in array\n", dl
->major
, dl
->minor
);
3917 /* skip in use or failed drives */
3918 if (dl
->disk
.status
& FAILED_DISK
|| idx
== dl
->index
||
3920 dprintf("%x:%x status (failed: %d index: %d)\n",
3921 dl
->major
, dl
->minor
,
3922 (dl
->disk
.status
& FAILED_DISK
) == FAILED_DISK
, idx
);
3926 /* skip pure spares when we are looking for partially
3927 * assimilated drives
3929 if (dl
->index
== -1 && !activate_new
)
3932 /* Does this unused device have the requisite free space?
3933 * It needs to be able to cover all member volumes
3935 ex
= get_extents(super
, dl
);
3937 dprintf("cannot get extents\n");
3940 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
3941 dev
= get_imsm_dev(super
, i
);
3942 map
= get_imsm_map(dev
, 0);
3944 /* check if this disk is already a member of
3947 if (get_imsm_disk_slot(map
, dl
->index
) >= 0)
3953 array_start
= __le32_to_cpu(map
->pba_of_lba0
);
3954 array_end
= array_start
+
3955 __le32_to_cpu(map
->blocks_per_member
) - 1;
3958 /* check that we can start at pba_of_lba0 with
3959 * blocks_per_member of space
3961 if (array_start
>= pos
&& array_end
< ex
[j
].start
) {
3965 pos
= ex
[j
].start
+ ex
[j
].size
;
3967 } while (ex
[j
-1].size
);
3974 if (i
< mpb
->num_raid_devs
) {
3975 dprintf("%x:%x does not have %u to %u available\n",
3976 dl
->major
, dl
->minor
, array_start
, array_end
);
3986 static struct mdinfo
*imsm_activate_spare(struct active_array
*a
,
3987 struct metadata_update
**updates
)
3990 * Find a device with unused free space and use it to replace a
3991 * failed/vacant region in an array. We replace failed regions one a
3992 * array at a time. The result is that a new spare disk will be added
3993 * to the first failed array and after the monitor has finished
3994 * propagating failures the remainder will be consumed.
3996 * FIXME add a capability for mdmon to request spares from another
4000 struct intel_super
*super
= a
->container
->sb
;
4001 int inst
= a
->info
.container_member
;
4002 struct imsm_dev
*dev
= get_imsm_dev(super
, inst
);
4003 struct imsm_map
*map
= get_imsm_map(dev
, 0);
4004 int failed
= a
->info
.array
.raid_disks
;
4005 struct mdinfo
*rv
= NULL
;
4008 struct metadata_update
*mu
;
4010 struct imsm_update_activate_spare
*u
;
4014 for (d
= a
->info
.devs
; d
; d
= d
->next
) {
4015 if ((d
->curr_state
& DS_FAULTY
) &&
4017 /* wait for Removal to happen */
4019 if (d
->state_fd
>= 0)
4023 dprintf("imsm: activate spare: inst=%d failed=%d (%d) level=%d\n",
4024 inst
, failed
, a
->info
.array
.raid_disks
, a
->info
.array
.level
);
4025 if (imsm_check_degraded(super
, dev
, failed
) != IMSM_T_STATE_DEGRADED
)
4028 /* For each slot, if it is not working, find a spare */
4029 for (i
= 0; i
< a
->info
.array
.raid_disks
; i
++) {
4030 for (d
= a
->info
.devs
; d
; d
= d
->next
)
4031 if (d
->disk
.raid_disk
== i
)
4033 dprintf("found %d: %p %x\n", i
, d
, d
?d
->curr_state
:0);
4034 if (d
&& (d
->state_fd
>= 0))
4038 * OK, this device needs recovery. Try to re-add the
4039 * previous occupant of this slot, if this fails see if
4040 * we can continue the assimilation of a spare that was
4041 * partially assimilated, finally try to activate a new
4044 dl
= imsm_readd(super
, i
, a
);
4046 dl
= imsm_add_spare(super
, i
, a
, 0);
4048 dl
= imsm_add_spare(super
, i
, a
, 1);
4052 /* found a usable disk with enough space */
4053 di
= malloc(sizeof(*di
));
4056 memset(di
, 0, sizeof(*di
));
4058 /* dl->index will be -1 in the case we are activating a
4059 * pristine spare. imsm_process_update() will create a
4060 * new index in this case. Once a disk is found to be
4061 * failed in all member arrays it is kicked from the
4064 di
->disk
.number
= dl
->index
;
4066 /* (ab)use di->devs to store a pointer to the device
4069 di
->devs
= (struct mdinfo
*) dl
;
4071 di
->disk
.raid_disk
= i
;
4072 di
->disk
.major
= dl
->major
;
4073 di
->disk
.minor
= dl
->minor
;
4075 di
->data_offset
= __le32_to_cpu(map
->pba_of_lba0
);
4076 di
->component_size
= a
->info
.component_size
;
4077 di
->container_member
= inst
;
4078 super
->random
= random32();
4082 dprintf("%x:%x to be %d at %llu\n", dl
->major
, dl
->minor
,
4083 i
, di
->data_offset
);
4089 /* No spares found */
4091 /* Now 'rv' has a list of devices to return.
4092 * Create a metadata_update record to update the
4093 * disk_ord_tbl for the array
4095 mu
= malloc(sizeof(*mu
));
4097 mu
->buf
= malloc(sizeof(struct imsm_update_activate_spare
) * num_spares
);
4098 if (mu
->buf
== NULL
) {
4105 struct mdinfo
*n
= rv
->next
;
4114 mu
->len
= sizeof(struct imsm_update_activate_spare
) * num_spares
;
4115 mu
->next
= *updates
;
4116 u
= (struct imsm_update_activate_spare
*) mu
->buf
;
4118 for (di
= rv
; di
; di
= di
->next
) {
4119 u
->type
= update_activate_spare
;
4120 u
->dl
= (struct dl
*) di
->devs
;
4122 u
->slot
= di
->disk
.raid_disk
;
4133 static int disks_overlap(struct intel_super
*super
, int idx
, struct imsm_update_create_array
*u
)
4135 struct imsm_dev
*dev
= get_imsm_dev(super
, idx
);
4136 struct imsm_map
*map
= get_imsm_map(dev
, 0);
4137 struct imsm_map
*new_map
= get_imsm_map(&u
->dev
, 0);
4138 struct disk_info
*inf
= get_disk_info(u
);
4139 struct imsm_disk
*disk
;
4143 for (i
= 0; i
< map
->num_members
; i
++) {
4144 disk
= get_imsm_disk(super
, get_imsm_disk_idx(dev
, i
));
4145 for (j
= 0; j
< new_map
->num_members
; j
++)
4146 if (serialcmp(disk
->serial
, inf
[j
].serial
) == 0)
4153 static void imsm_delete(struct intel_super
*super
, struct dl
**dlp
, int index
);
4155 static void imsm_process_update(struct supertype
*st
,
4156 struct metadata_update
*update
)
4159 * crack open the metadata_update envelope to find the update record
4160 * update can be one of:
4161 * update_activate_spare - a spare device has replaced a failed
4162 * device in an array, update the disk_ord_tbl. If this disk is
4163 * present in all member arrays then also clear the SPARE_DISK
4166 struct intel_super
*super
= st
->sb
;
4167 struct imsm_super
*mpb
;
4168 enum imsm_update_type type
= *(enum imsm_update_type
*) update
->buf
;
4170 /* update requires a larger buf but the allocation failed */
4171 if (super
->next_len
&& !super
->next_buf
) {
4172 super
->next_len
= 0;
4176 if (super
->next_buf
) {
4177 memcpy(super
->next_buf
, super
->buf
, super
->len
);
4179 super
->len
= super
->next_len
;
4180 super
->buf
= super
->next_buf
;
4182 super
->next_len
= 0;
4183 super
->next_buf
= NULL
;
4186 mpb
= super
->anchor
;
4189 case update_activate_spare
: {
4190 struct imsm_update_activate_spare
*u
= (void *) update
->buf
;
4191 struct imsm_dev
*dev
= get_imsm_dev(super
, u
->array
);
4192 struct imsm_map
*map
= get_imsm_map(dev
, 0);
4193 struct imsm_map
*migr_map
;
4194 struct active_array
*a
;
4195 struct imsm_disk
*disk
;
4200 int victim
= get_imsm_disk_idx(dev
, u
->slot
);
4203 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
4208 fprintf(stderr
, "error: imsm_activate_spare passed "
4209 "an unknown disk (index: %d)\n",
4214 super
->updates_pending
++;
4216 /* count failures (excluding rebuilds and the victim)
4217 * to determine map[0] state
4220 for (i
= 0; i
< map
->num_members
; i
++) {
4223 disk
= get_imsm_disk(super
, get_imsm_disk_idx(dev
, i
));
4224 if (!disk
|| disk
->status
& FAILED_DISK
)
4228 /* adding a pristine spare, assign a new index */
4229 if (dl
->index
< 0) {
4230 dl
->index
= super
->anchor
->num_disks
;
4231 super
->anchor
->num_disks
++;
4234 disk
->status
|= CONFIGURED_DISK
;
4235 disk
->status
&= ~SPARE_DISK
;
4238 to_state
= imsm_check_degraded(super
, dev
, failed
);
4239 map
->map_state
= IMSM_T_STATE_DEGRADED
;
4240 migrate(dev
, to_state
, MIGR_REBUILD
);
4241 migr_map
= get_imsm_map(dev
, 1);
4242 set_imsm_ord_tbl_ent(map
, u
->slot
, dl
->index
);
4243 set_imsm_ord_tbl_ent(migr_map
, u
->slot
, dl
->index
| IMSM_ORD_REBUILD
);
4245 /* update the family_num to mark a new container
4246 * generation, being careful to record the existing
4247 * family_num in orig_family_num to clean up after
4248 * earlier mdadm versions that neglected to set it.
4250 if (mpb
->orig_family_num
== 0)
4251 mpb
->orig_family_num
= mpb
->family_num
;
4252 mpb
->family_num
+= super
->random
;
4254 /* count arrays using the victim in the metadata */
4256 for (a
= st
->arrays
; a
; a
= a
->next
) {
4257 dev
= get_imsm_dev(super
, a
->info
.container_member
);
4258 map
= get_imsm_map(dev
, 0);
4260 if (get_imsm_disk_slot(map
, victim
) >= 0)
4264 /* delete the victim if it is no longer being
4270 /* We know that 'manager' isn't touching anything,
4271 * so it is safe to delete
4273 for (dlp
= &super
->disks
; *dlp
; dlp
= &(*dlp
)->next
)
4274 if ((*dlp
)->index
== victim
)
4277 /* victim may be on the missing list */
4279 for (dlp
= &super
->missing
; *dlp
; dlp
= &(*dlp
)->next
)
4280 if ((*dlp
)->index
== victim
)
4282 imsm_delete(super
, dlp
, victim
);
4286 case update_create_array
: {
4287 /* someone wants to create a new array, we need to be aware of
4288 * a few races/collisions:
4289 * 1/ 'Create' called by two separate instances of mdadm
4290 * 2/ 'Create' versus 'activate_spare': mdadm has chosen
4291 * devices that have since been assimilated via
4293 * In the event this update can not be carried out mdadm will
4294 * (FIX ME) notice that its update did not take hold.
4296 struct imsm_update_create_array
*u
= (void *) update
->buf
;
4297 struct intel_dev
*dv
;
4298 struct imsm_dev
*dev
;
4299 struct imsm_map
*map
, *new_map
;
4300 unsigned long long start
, end
;
4301 unsigned long long new_start
, new_end
;
4303 struct disk_info
*inf
;
4306 /* handle racing creates: first come first serve */
4307 if (u
->dev_idx
< mpb
->num_raid_devs
) {
4308 dprintf("%s: subarray %d already defined\n",
4309 __func__
, u
->dev_idx
);
4313 /* check update is next in sequence */
4314 if (u
->dev_idx
!= mpb
->num_raid_devs
) {
4315 dprintf("%s: can not create array %d expected index %d\n",
4316 __func__
, u
->dev_idx
, mpb
->num_raid_devs
);
4320 new_map
= get_imsm_map(&u
->dev
, 0);
4321 new_start
= __le32_to_cpu(new_map
->pba_of_lba0
);
4322 new_end
= new_start
+ __le32_to_cpu(new_map
->blocks_per_member
);
4323 inf
= get_disk_info(u
);
4325 /* handle activate_spare versus create race:
4326 * check to make sure that overlapping arrays do not include
4329 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
4330 dev
= get_imsm_dev(super
, i
);
4331 map
= get_imsm_map(dev
, 0);
4332 start
= __le32_to_cpu(map
->pba_of_lba0
);
4333 end
= start
+ __le32_to_cpu(map
->blocks_per_member
);
4334 if ((new_start
>= start
&& new_start
<= end
) ||
4335 (start
>= new_start
&& start
<= new_end
))
4340 if (disks_overlap(super
, i
, u
)) {
4341 dprintf("%s: arrays overlap\n", __func__
);
4346 /* check that prepare update was successful */
4347 if (!update
->space
) {
4348 dprintf("%s: prepare update failed\n", __func__
);
4352 /* check that all disks are still active before committing
4353 * changes. FIXME: could we instead handle this by creating a
4354 * degraded array? That's probably not what the user expects,
4355 * so better to drop this update on the floor.
4357 for (i
= 0; i
< new_map
->num_members
; i
++) {
4358 dl
= serial_to_dl(inf
[i
].serial
, super
);
4360 dprintf("%s: disk disappeared\n", __func__
);
4365 super
->updates_pending
++;
4367 /* convert spares to members and fixup ord_tbl */
4368 for (i
= 0; i
< new_map
->num_members
; i
++) {
4369 dl
= serial_to_dl(inf
[i
].serial
, super
);
4370 if (dl
->index
== -1) {
4371 dl
->index
= mpb
->num_disks
;
4373 dl
->disk
.status
|= CONFIGURED_DISK
;
4374 dl
->disk
.status
&= ~SPARE_DISK
;
4376 set_imsm_ord_tbl_ent(new_map
, i
, dl
->index
);
4381 update
->space
= NULL
;
4382 imsm_copy_dev(dev
, &u
->dev
);
4383 dv
->index
= u
->dev_idx
;
4384 dv
->next
= super
->devlist
;
4385 super
->devlist
= dv
;
4386 mpb
->num_raid_devs
++;
4388 imsm_update_version_info(super
);
4391 /* mdmon knows how to release update->space, but not
4392 * ((struct intel_dev *) update->space)->dev
4394 if (update
->space
) {
4400 case update_add_disk
:
4402 /* we may be able to repair some arrays if disks are
4405 struct active_array
*a
;
4407 super
->updates_pending
++;
4408 for (a
= st
->arrays
; a
; a
= a
->next
)
4409 a
->check_degraded
= 1;
4411 /* add some spares to the metadata */
4412 while (super
->add
) {
4416 super
->add
= al
->next
;
4417 al
->next
= super
->disks
;
4419 dprintf("%s: added %x:%x\n",
4420 __func__
, al
->major
, al
->minor
);
4427 static void imsm_prepare_update(struct supertype
*st
,
4428 struct metadata_update
*update
)
4431 * Allocate space to hold new disk entries, raid-device entries or a new
4432 * mpb if necessary. The manager synchronously waits for updates to
4433 * complete in the monitor, so new mpb buffers allocated here can be
4434 * integrated by the monitor thread without worrying about live pointers
4435 * in the manager thread.
4437 enum imsm_update_type type
= *(enum imsm_update_type
*) update
->buf
;
4438 struct intel_super
*super
= st
->sb
;
4439 struct imsm_super
*mpb
= super
->anchor
;
4444 case update_create_array
: {
4445 struct imsm_update_create_array
*u
= (void *) update
->buf
;
4446 struct intel_dev
*dv
;
4447 struct imsm_dev
*dev
= &u
->dev
;
4448 struct imsm_map
*map
= get_imsm_map(dev
, 0);
4450 struct disk_info
*inf
;
4454 inf
= get_disk_info(u
);
4455 len
= sizeof_imsm_dev(dev
, 1);
4456 /* allocate a new super->devlist entry */
4457 dv
= malloc(sizeof(*dv
));
4459 dv
->dev
= malloc(len
);
4464 update
->space
= NULL
;
4468 /* count how many spares will be converted to members */
4469 for (i
= 0; i
< map
->num_members
; i
++) {
4470 dl
= serial_to_dl(inf
[i
].serial
, super
);
4472 /* hmm maybe it failed?, nothing we can do about
4477 if (count_memberships(dl
, super
) == 0)
4480 len
+= activate
* sizeof(struct imsm_disk
);
4487 /* check if we need a larger metadata buffer */
4488 if (super
->next_buf
)
4489 buf_len
= super
->next_len
;
4491 buf_len
= super
->len
;
4493 if (__le32_to_cpu(mpb
->mpb_size
) + len
> buf_len
) {
4494 /* ok we need a larger buf than what is currently allocated
4495 * if this allocation fails process_update will notice that
4496 * ->next_len is set and ->next_buf is NULL
4498 buf_len
= ROUND_UP(__le32_to_cpu(mpb
->mpb_size
) + len
, 512);
4499 if (super
->next_buf
)
4500 free(super
->next_buf
);
4502 super
->next_len
= buf_len
;
4503 if (posix_memalign(&super
->next_buf
, 512, buf_len
) == 0)
4504 memset(super
->next_buf
, 0, buf_len
);
4506 super
->next_buf
= NULL
;
4510 /* must be called while manager is quiesced */
4511 static void imsm_delete(struct intel_super
*super
, struct dl
**dlp
, int index
)
4513 struct imsm_super
*mpb
= super
->anchor
;
4515 struct imsm_dev
*dev
;
4516 struct imsm_map
*map
;
4517 int i
, j
, num_members
;
4520 dprintf("%s: deleting device[%d] from imsm_super\n",
4523 /* shift all indexes down one */
4524 for (iter
= super
->disks
; iter
; iter
= iter
->next
)
4525 if (iter
->index
> index
)
4527 for (iter
= super
->missing
; iter
; iter
= iter
->next
)
4528 if (iter
->index
> index
)
4531 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
4532 dev
= get_imsm_dev(super
, i
);
4533 map
= get_imsm_map(dev
, 0);
4534 num_members
= map
->num_members
;
4535 for (j
= 0; j
< num_members
; j
++) {
4536 /* update ord entries being careful not to propagate
4537 * ord-flags to the first map
4539 ord
= get_imsm_ord_tbl_ent(dev
, j
);
4541 if (ord_to_idx(ord
) <= index
)
4544 map
= get_imsm_map(dev
, 0);
4545 set_imsm_ord_tbl_ent(map
, j
, ord_to_idx(ord
- 1));
4546 map
= get_imsm_map(dev
, 1);
4548 set_imsm_ord_tbl_ent(map
, j
, ord
- 1);
4553 super
->updates_pending
++;
4555 struct dl
*dl
= *dlp
;
4557 *dlp
= (*dlp
)->next
;
4558 __free_imsm_disk(dl
);
4561 #endif /* MDASSEMBLE */
4563 struct superswitch super_imsm
= {
4565 .examine_super
= examine_super_imsm
,
4566 .brief_examine_super
= brief_examine_super_imsm
,
4567 .brief_examine_subarrays
= brief_examine_subarrays_imsm
,
4568 .export_examine_super
= export_examine_super_imsm
,
4569 .detail_super
= detail_super_imsm
,
4570 .brief_detail_super
= brief_detail_super_imsm
,
4571 .write_init_super
= write_init_super_imsm
,
4572 .validate_geometry
= validate_geometry_imsm
,
4573 .add_to_super
= add_to_super_imsm
,
4574 .detail_platform
= detail_platform_imsm
,
4576 .match_home
= match_home_imsm
,
4577 .uuid_from_super
= uuid_from_super_imsm
,
4578 .getinfo_super
= getinfo_super_imsm
,
4579 .update_super
= update_super_imsm
,
4581 .avail_size
= avail_size_imsm
,
4583 .compare_super
= compare_super_imsm
,
4585 .load_super
= load_super_imsm
,
4586 .init_super
= init_super_imsm
,
4587 .store_super
= store_zero_imsm
,
4588 .free_super
= free_super_imsm
,
4589 .match_metadata_desc
= match_metadata_desc_imsm
,
4590 .container_content
= container_content_imsm
,
4591 .default_layout
= imsm_level_to_layout
,
4598 .open_new
= imsm_open_new
,
4599 .load_super
= load_super_imsm
,
4600 .set_array_state
= imsm_set_array_state
,
4601 .set_disk
= imsm_set_disk
,
4602 .sync_metadata
= imsm_sync_metadata
,
4603 .activate_spare
= imsm_activate_spare
,
4604 .process_update
= imsm_process_update
,
4605 .prepare_update
= imsm_prepare_update
,
4606 #endif /* MDASSEMBLE */