2 * mdadm - Intel(R) Matrix Storage Manager Support
4 * Copyright (C) 2002-2008 Intel Corporation
6 * This program is free software; you can redistribute it and/or modify it
7 * under the terms and conditions of the GNU General Public License,
8 * version 2, as published by the Free Software Foundation.
10 * This program is distributed in the hope it will be useful, but WITHOUT
11 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
12 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for
15 * You should have received a copy of the GNU General Public License along with
16 * this program; if not, write to the Free Software Foundation, Inc.,
17 * 51 Franklin St - Fifth Floor, Boston, MA 02110-1301 USA.
20 #define HAVE_STDINT_H 1
24 #include "platform-intel.h"
30 /* MPB == Metadata Parameter Block */
31 #define MPB_SIGNATURE "Intel Raid ISM Cfg Sig. "
32 #define MPB_SIG_LEN (strlen(MPB_SIGNATURE))
33 #define MPB_VERSION_RAID0 "1.0.00"
34 #define MPB_VERSION_RAID1 "1.1.00"
35 #define MPB_VERSION_MANY_VOLUMES_PER_ARRAY "1.2.00"
36 #define MPB_VERSION_3OR4_DISK_ARRAY "1.2.01"
37 #define MPB_VERSION_RAID5 "1.2.02"
38 #define MPB_VERSION_5OR6_DISK_ARRAY "1.2.04"
39 #define MPB_VERSION_CNG "1.2.06"
40 #define MPB_VERSION_ATTRIBS "1.3.00"
41 #define MAX_SIGNATURE_LENGTH 32
42 #define MAX_RAID_SERIAL_LEN 16
44 #define MPB_ATTRIB_CHECKSUM_VERIFY __cpu_to_le32(0x80000000)
45 #define MPB_ATTRIB_PM __cpu_to_le32(0x40000000)
46 #define MPB_ATTRIB_2TB __cpu_to_le32(0x20000000)
47 #define MPB_ATTRIB_RAID0 __cpu_to_le32(0x00000001)
48 #define MPB_ATTRIB_RAID1 __cpu_to_le32(0x00000002)
49 #define MPB_ATTRIB_RAID10 __cpu_to_le32(0x00000004)
50 #define MPB_ATTRIB_RAID1E __cpu_to_le32(0x00000008)
51 #define MPB_ATTRIB_RAID5 __cpu_to_le32(0x00000010)
52 #define MPB_ATTRIB_RAIDCNG __cpu_to_le32(0x00000020)
54 #define MPB_SECTOR_CNT 418
55 #define IMSM_RESERVED_SECTORS 4096
57 /* Disk configuration info. */
58 #define IMSM_MAX_DEVICES 255
60 __u8 serial
[MAX_RAID_SERIAL_LEN
];/* 0xD8 - 0xE7 ascii serial number */
61 __u32 total_blocks
; /* 0xE8 - 0xEB total blocks */
62 __u32 scsi_id
; /* 0xEC - 0xEF scsi ID */
63 #define SPARE_DISK __cpu_to_le32(0x01) /* Spare */
64 #define CONFIGURED_DISK __cpu_to_le32(0x02) /* Member of some RaidDev */
65 #define FAILED_DISK __cpu_to_le32(0x04) /* Permanent failure */
66 #define USABLE_DISK __cpu_to_le32(0x08) /* Fully usable unless FAILED_DISK is set */
67 __u32 status
; /* 0xF0 - 0xF3 */
68 __u32 owner_cfg_num
; /* which config 0,1,2... owns this disk */
69 #define IMSM_DISK_FILLERS 4
70 __u32 filler
[IMSM_DISK_FILLERS
]; /* 0xF4 - 0x107 MPB_DISK_FILLERS for future expansion */
73 /* RAID map configuration infos. */
75 __u32 pba_of_lba0
; /* start address of partition */
76 __u32 blocks_per_member
;/* blocks per member */
77 __u32 num_data_stripes
; /* number of data stripes */
78 __u16 blocks_per_strip
;
79 __u8 map_state
; /* Normal, Uninitialized, Degraded, Failed */
80 #define IMSM_T_STATE_NORMAL 0
81 #define IMSM_T_STATE_UNINITIALIZED 1
82 #define IMSM_T_STATE_DEGRADED 2
83 #define IMSM_T_STATE_FAILED 3
85 #define IMSM_T_RAID0 0
86 #define IMSM_T_RAID1 1
87 #define IMSM_T_RAID5 5 /* since metadata version 1.2.02 ? */
88 __u8 num_members
; /* number of member disks */
89 __u8 num_domains
; /* number of parity domains */
90 __u8 failed_disk_num
; /* valid only when state is degraded */
92 __u32 filler
[7]; /* expansion area */
93 #define IMSM_ORD_REBUILD (1 << 24)
94 __u32 disk_ord_tbl
[1]; /* disk_ord_tbl[num_members],
95 * top byte contains some flags
97 } __attribute__ ((packed
));
100 __u32 curr_migr_unit
;
101 __u32 checkpoint_id
; /* id to access curr_migr_unit */
102 __u8 migr_state
; /* Normal or Migrating */
104 #define MIGR_REBUILD 1
105 #define MIGR_VERIFY 2 /* analagous to echo check > sync_action */
106 #define MIGR_GEN_MIGR 3
107 #define MIGR_STATE_CHANGE 4
108 __u8 migr_type
; /* Initializing, Rebuilding, ... */
110 __u8 fs_state
; /* fast-sync state for CnG (0xff == disabled) */
111 __u16 verify_errors
; /* number of mismatches */
112 __u16 bad_blocks
; /* number of bad blocks during verify */
114 struct imsm_map map
[1];
115 /* here comes another one if migr_state */
116 } __attribute__ ((packed
));
119 __u8 volume
[MAX_RAID_SERIAL_LEN
];
122 #define DEV_BOOTABLE __cpu_to_le32(0x01)
123 #define DEV_BOOT_DEVICE __cpu_to_le32(0x02)
124 #define DEV_READ_COALESCING __cpu_to_le32(0x04)
125 #define DEV_WRITE_COALESCING __cpu_to_le32(0x08)
126 #define DEV_LAST_SHUTDOWN_DIRTY __cpu_to_le32(0x10)
127 #define DEV_HIDDEN_AT_BOOT __cpu_to_le32(0x20)
128 #define DEV_CURRENTLY_HIDDEN __cpu_to_le32(0x40)
129 #define DEV_VERIFY_AND_FIX __cpu_to_le32(0x80)
130 #define DEV_MAP_STATE_UNINIT __cpu_to_le32(0x100)
131 #define DEV_NO_AUTO_RECOVERY __cpu_to_le32(0x200)
132 #define DEV_CLONE_N_GO __cpu_to_le32(0x400)
133 #define DEV_CLONE_MAN_SYNC __cpu_to_le32(0x800)
134 #define DEV_CNG_MASTER_DISK_NUM __cpu_to_le32(0x1000)
135 __u32 status
; /* Persistent RaidDev status */
136 __u32 reserved_blocks
; /* Reserved blocks at beginning of volume */
140 __u8 cng_master_disk
;
144 #define IMSM_DEV_FILLERS 10
145 __u32 filler
[IMSM_DEV_FILLERS
];
147 } __attribute__ ((packed
));
150 __u8 sig
[MAX_SIGNATURE_LENGTH
]; /* 0x00 - 0x1F */
151 __u32 check_sum
; /* 0x20 - 0x23 MPB Checksum */
152 __u32 mpb_size
; /* 0x24 - 0x27 Size of MPB */
153 __u32 family_num
; /* 0x28 - 0x2B Checksum from first time this config was written */
154 __u32 generation_num
; /* 0x2C - 0x2F Incremented each time this array's MPB is written */
155 __u32 error_log_size
; /* 0x30 - 0x33 in bytes */
156 __u32 attributes
; /* 0x34 - 0x37 */
157 __u8 num_disks
; /* 0x38 Number of configured disks */
158 __u8 num_raid_devs
; /* 0x39 Number of configured volumes */
159 __u8 error_log_pos
; /* 0x3A */
160 __u8 fill
[1]; /* 0x3B */
161 __u32 cache_size
; /* 0x3c - 0x40 in mb */
162 __u32 orig_family_num
; /* 0x40 - 0x43 original family num */
163 __u32 pwr_cycle_count
; /* 0x44 - 0x47 simulated power cycle count for array */
164 __u32 bbm_log_size
; /* 0x48 - 0x4B - size of bad Block Mgmt Log in bytes */
165 #define IMSM_FILLERS 35
166 __u32 filler
[IMSM_FILLERS
]; /* 0x4C - 0xD7 RAID_MPB_FILLERS */
167 struct imsm_disk disk
[1]; /* 0xD8 diskTbl[numDisks] */
168 /* here comes imsm_dev[num_raid_devs] */
169 /* here comes BBM logs */
170 } __attribute__ ((packed
));
172 #define BBM_LOG_MAX_ENTRIES 254
174 struct bbm_log_entry
{
175 __u64 defective_block_start
;
176 #define UNREADABLE 0xFFFFFFFF
177 __u32 spare_block_offset
;
178 __u16 remapped_marked_count
;
180 } __attribute__ ((__packed__
));
183 __u32 signature
; /* 0xABADB10C */
185 __u32 reserved_spare_block_count
; /* 0 */
186 __u32 reserved
; /* 0xFFFF */
187 __u64 first_spare_lba
;
188 struct bbm_log_entry mapped_block_entries
[BBM_LOG_MAX_ENTRIES
];
189 } __attribute__ ((__packed__
));
193 static char *map_state_str
[] = { "normal", "uninitialized", "degraded", "failed" };
196 static unsigned int sector_count(__u32 bytes
)
198 return ((bytes
+ (512-1)) & (~(512-1))) / 512;
201 static unsigned int mpb_sectors(struct imsm_super
*mpb
)
203 return sector_count(__le32_to_cpu(mpb
->mpb_size
));
207 struct imsm_dev
*dev
;
208 struct intel_dev
*next
;
212 /* internal representation of IMSM metadata */
215 void *buf
; /* O_DIRECT buffer for reading/writing metadata */
216 struct imsm_super
*anchor
; /* immovable parameters */
218 size_t len
; /* size of the 'buf' allocation */
219 void *next_buf
; /* for realloc'ing buf from the manager */
221 int updates_pending
; /* count of pending updates for mdmon */
222 int creating_imsm
; /* flag to indicate container creation */
223 int current_vol
; /* index of raid device undergoing creation */
224 __u32 create_offset
; /* common start for 'current_vol' */
225 struct intel_dev
*devlist
;
229 __u8 serial
[MAX_RAID_SERIAL_LEN
];
232 struct imsm_disk disk
;
235 struct extent
*e
; /* for determining freespace @ create */
236 int raiddisk
; /* slot to fill in autolayout */
238 struct dl
*add
; /* list of disks to add while mdmon active */
239 struct dl
*missing
; /* disks removed while we weren't looking */
240 struct bbm_log
*bbm_log
;
241 const char *hba
; /* device path of the raid controller for this metadata */
242 const struct imsm_orom
*orom
; /* platform firmware support */
246 unsigned long long start
, size
;
249 /* definition of messages passed to imsm_process_update */
250 enum imsm_update_type
{
251 update_activate_spare
,
256 struct imsm_update_activate_spare
{
257 enum imsm_update_type type
;
261 struct imsm_update_activate_spare
*next
;
265 __u8 serial
[MAX_RAID_SERIAL_LEN
];
268 struct imsm_update_create_array
{
269 enum imsm_update_type type
;
274 struct imsm_update_add_disk
{
275 enum imsm_update_type type
;
278 static struct supertype
*match_metadata_desc_imsm(char *arg
)
280 struct supertype
*st
;
282 if (strcmp(arg
, "imsm") != 0 &&
283 strcmp(arg
, "default") != 0
287 st
= malloc(sizeof(*st
));
288 memset(st
, 0, sizeof(*st
));
289 st
->ss
= &super_imsm
;
290 st
->max_devs
= IMSM_MAX_DEVICES
;
291 st
->minor_version
= 0;
297 static __u8
*get_imsm_version(struct imsm_super
*mpb
)
299 return &mpb
->sig
[MPB_SIG_LEN
];
303 /* retrieve a disk directly from the anchor when the anchor is known to be
304 * up-to-date, currently only at load time
306 static struct imsm_disk
*__get_imsm_disk(struct imsm_super
*mpb
, __u8 index
)
308 if (index
>= mpb
->num_disks
)
310 return &mpb
->disk
[index
];
314 /* retrieve a disk from the parsed metadata */
315 static struct imsm_disk
*get_imsm_disk(struct intel_super
*super
, __u8 index
)
319 for (d
= super
->disks
; d
; d
= d
->next
)
320 if (d
->index
== index
)
327 /* generate a checksum directly from the anchor when the anchor is known to be
328 * up-to-date, currently only at load or write_super after coalescing
330 static __u32
__gen_imsm_checksum(struct imsm_super
*mpb
)
332 __u32 end
= mpb
->mpb_size
/ sizeof(end
);
333 __u32
*p
= (__u32
*) mpb
;
337 sum
+= __le32_to_cpu(*p
);
341 return sum
- __le32_to_cpu(mpb
->check_sum
);
344 static size_t sizeof_imsm_map(struct imsm_map
*map
)
346 return sizeof(struct imsm_map
) + sizeof(__u32
) * (map
->num_members
- 1);
349 struct imsm_map
*get_imsm_map(struct imsm_dev
*dev
, int second_map
)
351 struct imsm_map
*map
= &dev
->vol
.map
[0];
353 if (second_map
&& !dev
->vol
.migr_state
)
355 else if (second_map
) {
358 return ptr
+ sizeof_imsm_map(map
);
364 /* return the size of the device.
365 * migr_state increases the returned size if map[0] were to be duplicated
367 static size_t sizeof_imsm_dev(struct imsm_dev
*dev
, int migr_state
)
369 size_t size
= sizeof(*dev
) - sizeof(struct imsm_map
) +
370 sizeof_imsm_map(get_imsm_map(dev
, 0));
372 /* migrating means an additional map */
373 if (dev
->vol
.migr_state
)
374 size
+= sizeof_imsm_map(get_imsm_map(dev
, 1));
376 size
+= sizeof_imsm_map(get_imsm_map(dev
, 0));
382 /* retrieve disk serial number list from a metadata update */
383 static struct disk_info
*get_disk_info(struct imsm_update_create_array
*update
)
386 struct disk_info
*inf
;
388 inf
= u
+ sizeof(*update
) - sizeof(struct imsm_dev
) +
389 sizeof_imsm_dev(&update
->dev
, 0);
395 static struct imsm_dev
*__get_imsm_dev(struct imsm_super
*mpb
, __u8 index
)
401 if (index
>= mpb
->num_raid_devs
)
404 /* devices start after all disks */
405 offset
= ((void *) &mpb
->disk
[mpb
->num_disks
]) - _mpb
;
407 for (i
= 0; i
<= index
; i
++)
409 return _mpb
+ offset
;
411 offset
+= sizeof_imsm_dev(_mpb
+ offset
, 0);
416 static struct imsm_dev
*get_imsm_dev(struct intel_super
*super
, __u8 index
)
418 struct intel_dev
*dv
;
420 if (index
>= super
->anchor
->num_raid_devs
)
422 for (dv
= super
->devlist
; dv
; dv
= dv
->next
)
423 if (dv
->index
== index
)
428 static __u32
get_imsm_ord_tbl_ent(struct imsm_dev
*dev
, int slot
)
430 struct imsm_map
*map
;
432 if (dev
->vol
.migr_state
)
433 map
= get_imsm_map(dev
, 1);
435 map
= get_imsm_map(dev
, 0);
437 /* top byte identifies disk under rebuild */
438 return __le32_to_cpu(map
->disk_ord_tbl
[slot
]);
441 #define ord_to_idx(ord) (((ord) << 8) >> 8)
442 static __u32
get_imsm_disk_idx(struct imsm_dev
*dev
, int slot
)
444 __u32 ord
= get_imsm_ord_tbl_ent(dev
, slot
);
446 return ord_to_idx(ord
);
449 static void set_imsm_ord_tbl_ent(struct imsm_map
*map
, int slot
, __u32 ord
)
451 map
->disk_ord_tbl
[slot
] = __cpu_to_le32(ord
);
454 static int get_imsm_disk_slot(struct imsm_map
*map
, int idx
)
459 for (slot
= 0; slot
< map
->num_members
; slot
++) {
460 ord
= __le32_to_cpu(map
->disk_ord_tbl
[slot
]);
461 if (ord_to_idx(ord
) == idx
)
468 static int get_imsm_raid_level(struct imsm_map
*map
)
470 if (map
->raid_level
== 1) {
471 if (map
->num_members
== 2)
477 return map
->raid_level
;
480 static int cmp_extent(const void *av
, const void *bv
)
482 const struct extent
*a
= av
;
483 const struct extent
*b
= bv
;
484 if (a
->start
< b
->start
)
486 if (a
->start
> b
->start
)
491 static int count_memberships(struct dl
*dl
, struct intel_super
*super
)
496 for (i
= 0; i
< super
->anchor
->num_raid_devs
; i
++) {
497 struct imsm_dev
*dev
= get_imsm_dev(super
, i
);
498 struct imsm_map
*map
= get_imsm_map(dev
, 0);
500 if (get_imsm_disk_slot(map
, dl
->index
) >= 0)
507 static struct extent
*get_extents(struct intel_super
*super
, struct dl
*dl
)
509 /* find a list of used extents on the given physical device */
510 struct extent
*rv
, *e
;
512 int memberships
= count_memberships(dl
, super
);
513 __u32 reservation
= MPB_SECTOR_CNT
+ IMSM_RESERVED_SECTORS
;
515 rv
= malloc(sizeof(struct extent
) * (memberships
+ 1));
520 for (i
= 0; i
< super
->anchor
->num_raid_devs
; i
++) {
521 struct imsm_dev
*dev
= get_imsm_dev(super
, i
);
522 struct imsm_map
*map
= get_imsm_map(dev
, 0);
524 if (get_imsm_disk_slot(map
, dl
->index
) >= 0) {
525 e
->start
= __le32_to_cpu(map
->pba_of_lba0
);
526 e
->size
= __le32_to_cpu(map
->blocks_per_member
);
530 qsort(rv
, memberships
, sizeof(*rv
), cmp_extent
);
532 /* determine the start of the metadata
533 * when no raid devices are defined use the default
534 * ...otherwise allow the metadata to truncate the value
535 * as is the case with older versions of imsm
538 struct extent
*last
= &rv
[memberships
- 1];
541 remainder
= __le32_to_cpu(dl
->disk
.total_blocks
) -
542 (last
->start
+ last
->size
);
543 /* round down to 1k block to satisfy precision of the kernel
547 /* make sure remainder is still sane */
548 if (remainder
< ROUND_UP(super
->len
, 512) >> 9)
549 remainder
= ROUND_UP(super
->len
, 512) >> 9;
550 if (reservation
> remainder
)
551 reservation
= remainder
;
553 e
->start
= __le32_to_cpu(dl
->disk
.total_blocks
) - reservation
;
558 /* try to determine how much space is reserved for metadata from
559 * the last get_extents() entry, otherwise fallback to the
562 static __u32
imsm_reserved_sectors(struct intel_super
*super
, struct dl
*dl
)
568 /* for spares just return a minimal reservation which will grow
569 * once the spare is picked up by an array
572 return MPB_SECTOR_CNT
;
574 e
= get_extents(super
, dl
);
576 return MPB_SECTOR_CNT
+ IMSM_RESERVED_SECTORS
;
578 /* scroll to last entry */
579 for (i
= 0; e
[i
].size
; i
++)
582 rv
= __le32_to_cpu(dl
->disk
.total_blocks
) - e
[i
].start
;
590 static void print_imsm_dev(struct imsm_dev
*dev
, char *uuid
, int disk_idx
)
594 struct imsm_map
*map
= get_imsm_map(dev
, 0);
598 printf("[%.16s]:\n", dev
->volume
);
599 printf(" UUID : %s\n", uuid
);
600 printf(" RAID Level : %d\n", get_imsm_raid_level(map
));
601 printf(" Members : %d\n", map
->num_members
);
602 slot
= get_imsm_disk_slot(map
, disk_idx
);
604 ord
= get_imsm_ord_tbl_ent(dev
, slot
);
605 printf(" This Slot : %d%s\n", slot
,
606 ord
& IMSM_ORD_REBUILD
? " (out-of-sync)" : "");
608 printf(" This Slot : ?\n");
609 sz
= __le32_to_cpu(dev
->size_high
);
611 sz
+= __le32_to_cpu(dev
->size_low
);
612 printf(" Array Size : %llu%s\n", (unsigned long long)sz
,
613 human_size(sz
* 512));
614 sz
= __le32_to_cpu(map
->blocks_per_member
);
615 printf(" Per Dev Size : %llu%s\n", (unsigned long long)sz
,
616 human_size(sz
* 512));
617 printf(" Sector Offset : %u\n",
618 __le32_to_cpu(map
->pba_of_lba0
));
619 printf(" Num Stripes : %u\n",
620 __le32_to_cpu(map
->num_data_stripes
));
621 printf(" Chunk Size : %u KiB\n",
622 __le16_to_cpu(map
->blocks_per_strip
) / 2);
623 printf(" Reserved : %d\n", __le32_to_cpu(dev
->reserved_blocks
));
624 printf(" Migrate State : %s", dev
->vol
.migr_state
? "migrating" : "idle");
625 if (dev
->vol
.migr_state
)
626 printf(": %s", dev
->vol
.migr_type
? "rebuilding" : "initializing");
628 printf(" Map State : %s", map_state_str
[map
->map_state
]);
629 if (dev
->vol
.migr_state
) {
630 struct imsm_map
*map
= get_imsm_map(dev
, 1);
631 printf(" <-- %s", map_state_str
[map
->map_state
]);
634 printf(" Dirty State : %s\n", dev
->vol
.dirty
? "dirty" : "clean");
637 static void print_imsm_disk(struct imsm_super
*mpb
, int index
, __u32 reserved
)
639 struct imsm_disk
*disk
= __get_imsm_disk(mpb
, index
);
640 char str
[MAX_RAID_SERIAL_LEN
+ 1];
648 snprintf(str
, MAX_RAID_SERIAL_LEN
+ 1, "%s", disk
->serial
);
649 printf(" Disk%02d Serial : %s\n", index
, str
);
651 printf(" State :%s%s%s%s\n", s
&SPARE_DISK
? " spare" : "",
652 s
&CONFIGURED_DISK
? " active" : "",
653 s
&FAILED_DISK
? " failed" : "",
654 s
&USABLE_DISK
? " usable" : "");
655 printf(" Id : %08x\n", __le32_to_cpu(disk
->scsi_id
));
656 sz
= __le32_to_cpu(disk
->total_blocks
) - reserved
;
657 printf(" Usable Size : %llu%s\n", (unsigned long long)sz
,
658 human_size(sz
* 512));
661 static void getinfo_super_imsm(struct supertype
*st
, struct mdinfo
*info
);
663 static void examine_super_imsm(struct supertype
*st
, char *homehost
)
665 struct intel_super
*super
= st
->sb
;
666 struct imsm_super
*mpb
= super
->anchor
;
667 char str
[MAX_SIGNATURE_LENGTH
];
672 __u32 reserved
= imsm_reserved_sectors(super
, super
->disks
);
675 snprintf(str
, MPB_SIG_LEN
, "%s", mpb
->sig
);
676 printf(" Magic : %s\n", str
);
677 snprintf(str
, strlen(MPB_VERSION_RAID0
), "%s", get_imsm_version(mpb
));
678 printf(" Version : %s\n", get_imsm_version(mpb
));
679 printf(" Family : %08x\n", __le32_to_cpu(mpb
->family_num
));
680 printf(" Generation : %08x\n", __le32_to_cpu(mpb
->generation_num
));
681 getinfo_super_imsm(st
, &info
);
682 fname_from_uuid(st
, &info
, nbuf
,'-');
683 printf(" UUID : %s\n", nbuf
+ 5);
684 sum
= __le32_to_cpu(mpb
->check_sum
);
685 printf(" Checksum : %08x %s\n", sum
,
686 __gen_imsm_checksum(mpb
) == sum
? "correct" : "incorrect");
687 printf(" MPB Sectors : %d\n", mpb_sectors(mpb
));
688 printf(" Disks : %d\n", mpb
->num_disks
);
689 printf(" RAID Devices : %d\n", mpb
->num_raid_devs
);
690 print_imsm_disk(mpb
, super
->disks
->index
, reserved
);
691 if (super
->bbm_log
) {
692 struct bbm_log
*log
= super
->bbm_log
;
695 printf("Bad Block Management Log:\n");
696 printf(" Log Size : %d\n", __le32_to_cpu(mpb
->bbm_log_size
));
697 printf(" Signature : %x\n", __le32_to_cpu(log
->signature
));
698 printf(" Entry Count : %d\n", __le32_to_cpu(log
->entry_count
));
699 printf(" Spare Blocks : %d\n", __le32_to_cpu(log
->reserved_spare_block_count
));
700 printf(" First Spare : %llx\n", __le64_to_cpu(log
->first_spare_lba
));
702 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
704 struct imsm_dev
*dev
= __get_imsm_dev(mpb
, i
);
706 super
->current_vol
= i
;
707 getinfo_super_imsm(st
, &info
);
708 fname_from_uuid(st
, &info
, nbuf
, '-');
709 print_imsm_dev(dev
, nbuf
+ 5, super
->disks
->index
);
711 for (i
= 0; i
< mpb
->num_disks
; i
++) {
712 if (i
== super
->disks
->index
)
714 print_imsm_disk(mpb
, i
, reserved
);
718 static void brief_examine_super_imsm(struct supertype
*st
)
720 /* We just write a generic IMSM ARRAY entry */
724 struct intel_super
*super
= st
->sb
;
727 if (!super
->anchor
->num_raid_devs
)
730 getinfo_super_imsm(st
, &info
);
731 fname_from_uuid(st
, &info
, nbuf
,'-');
732 printf("ARRAY metadata=imsm auto=md UUID=%s\n", nbuf
+ 5);
733 for (i
= 0; i
< super
->anchor
->num_raid_devs
; i
++) {
734 struct imsm_dev
*dev
= get_imsm_dev(super
, i
);
736 super
->current_vol
= i
;
737 getinfo_super_imsm(st
, &info
);
738 fname_from_uuid(st
, &info
, nbuf1
,'-');
739 printf("ARRAY /dev/md/%.16s container=%s\n"
740 " member=%d auto=mdp UUID=%s\n",
741 dev
->volume
, nbuf
+ 5, i
, nbuf1
+ 5);
745 static void detail_super_imsm(struct supertype
*st
, char *homehost
)
750 getinfo_super_imsm(st
, &info
);
751 fname_from_uuid(st
, &info
, nbuf
,'-');
752 printf("\n UUID : %s\n", nbuf
+ 5);
755 static void brief_detail_super_imsm(struct supertype
*st
)
759 getinfo_super_imsm(st
, &info
);
760 fname_from_uuid(st
, &info
, nbuf
,'-');
761 printf(" UUID=%s", nbuf
+ 5);
764 static int imsm_read_serial(int fd
, char *devname
, __u8
*serial
);
765 static void fd2devname(int fd
, char *name
);
767 static int imsm_enumerate_ports(const char *hba_path
, int port_count
, int host_base
, int verbose
)
769 /* dump an unsorted list of devices attached to ahci, as well as
770 * non-connected ports
772 int hba_len
= strlen(hba_path
) + 1;
777 unsigned long port_mask
= (1 << port_count
) - 1;
779 if (port_count
> sizeof(port_mask
) * 8) {
781 fprintf(stderr
, Name
": port_count %d out of range\n", port_count
);
785 /* scroll through /sys/dev/block looking for devices attached to
788 dir
= opendir("/sys/dev/block");
789 for (ent
= dir
? readdir(dir
) : NULL
; ent
; ent
= readdir(dir
)) {
800 if (sscanf(ent
->d_name
, "%d:%d", &major
, &minor
) != 2)
802 path
= devt_to_devpath(makedev(major
, minor
));
805 if (!path_attached_to_hba(path
, hba_path
)) {
811 /* retrieve the scsi device type */
812 if (asprintf(&device
, "/sys/dev/block/%d:%d/device/xxxxxxx", major
, minor
) < 0) {
814 fprintf(stderr
, Name
": failed to allocate 'device'\n");
818 sprintf(device
, "/sys/dev/block/%d:%d/device/type", major
, minor
);
819 if (load_sys(device
, buf
) != 0) {
821 fprintf(stderr
, Name
": failed to read device type for %s\n",
827 type
= strtoul(buf
, NULL
, 10);
829 /* if it's not a disk print the vendor and model */
830 if (!(type
== 0 || type
== 7 || type
== 14)) {
833 sprintf(device
, "/sys/dev/block/%d:%d/device/vendor", major
, minor
);
834 if (load_sys(device
, buf
) == 0) {
835 strncpy(vendor
, buf
, sizeof(vendor
));
836 vendor
[sizeof(vendor
) - 1] = '\0';
837 c
= (char *) &vendor
[sizeof(vendor
) - 1];
838 while (isspace(*c
) || *c
== '\0')
842 sprintf(device
, "/sys/dev/block/%d:%d/device/model", major
, minor
);
843 if (load_sys(device
, buf
) == 0) {
844 strncpy(model
, buf
, sizeof(model
));
845 model
[sizeof(model
) - 1] = '\0';
846 c
= (char *) &model
[sizeof(model
) - 1];
847 while (isspace(*c
) || *c
== '\0')
851 if (vendor
[0] && model
[0])
852 sprintf(buf
, "%.64s %.64s", vendor
, model
);
854 switch (type
) { /* numbers from hald/linux/device.c */
855 case 1: sprintf(buf
, "tape"); break;
856 case 2: sprintf(buf
, "printer"); break;
857 case 3: sprintf(buf
, "processor"); break;
859 case 5: sprintf(buf
, "cdrom"); break;
860 case 6: sprintf(buf
, "scanner"); break;
861 case 8: sprintf(buf
, "media_changer"); break;
862 case 9: sprintf(buf
, "comm"); break;
863 case 12: sprintf(buf
, "raid"); break;
864 default: sprintf(buf
, "unknown");
870 /* chop device path to 'host%d' and calculate the port number */
871 c
= strchr(&path
[hba_len
], '/');
873 if (sscanf(&path
[hba_len
], "host%d", &port
) == 1)
877 *c
= '/'; /* repair the full string */
878 fprintf(stderr
, Name
": failed to determine port number for %s\n",
885 /* mark this port as used */
886 port_mask
&= ~(1 << port
);
888 /* print out the device information */
890 printf(" Port%d : - non-disk device (%s) -\n", port
, buf
);
894 fd
= dev_open(ent
->d_name
, O_RDONLY
);
896 printf(" Port%d : - disk info unavailable -\n", port
);
899 printf(" Port%d : %s", port
, buf
);
900 if (imsm_read_serial(fd
, NULL
, (__u8
*) buf
) == 0)
901 printf(" (%s)\n", buf
);
916 for (i
= 0; i
< port_count
; i
++)
917 if (port_mask
& (1 << i
))
918 printf(" Port%d : - no device attached -\n", i
);
924 static int detail_platform_imsm(int verbose
, int enumerate_only
)
926 /* There are two components to imsm platform support, the ahci SATA
927 * controller and the option-rom. To find the SATA controller we
928 * simply look in /sys/bus/pci/drivers/ahci to see if an ahci
929 * controller with the Intel vendor id is present. This approach
930 * allows mdadm to leverage the kernel's ahci detection logic, with the
931 * caveat that if ahci.ko is not loaded mdadm will not be able to
932 * detect platform raid capabilities. The option-rom resides in a
933 * platform "Adapter ROM". We scan for its signature to retrieve the
934 * platform capabilities. If raid support is disabled in the BIOS the
935 * option-rom capability structure will not be available.
937 const struct imsm_orom
*orom
;
938 struct sys_dev
*list
, *hba
;
941 const char *hba_path
;
945 if (enumerate_only
) {
946 if (check_env("IMSM_NO_PLATFORM") || find_imsm_orom())
951 list
= find_driver_devices("pci", "ahci");
952 for (hba
= list
; hba
; hba
= hba
->next
)
953 if (devpath_to_vendor(hba
->path
) == 0x8086)
958 fprintf(stderr
, Name
": unable to find active ahci controller\n");
962 fprintf(stderr
, Name
": found Intel SATA AHCI Controller\n");
963 hba_path
= hba
->path
;
967 orom
= find_imsm_orom();
970 fprintf(stderr
, Name
": imsm option-rom not found\n");
974 printf(" Platform : Intel(R) Matrix Storage Manager\n");
975 printf(" Version : %d.%d.%d.%d\n", orom
->major_ver
, orom
->minor_ver
,
976 orom
->hotfix_ver
, orom
->build
);
977 printf(" RAID Levels :%s%s%s%s%s\n",
978 imsm_orom_has_raid0(orom
) ? " raid0" : "",
979 imsm_orom_has_raid1(orom
) ? " raid1" : "",
980 imsm_orom_has_raid1e(orom
) ? " raid1e" : "",
981 imsm_orom_has_raid10(orom
) ? " raid10" : "",
982 imsm_orom_has_raid5(orom
) ? " raid5" : "");
983 printf(" Chunk Sizes :%s%s%s%s%s%s%s%s%s%s%s%s%s%s%s%s\n",
984 imsm_orom_has_chunk(orom
, 2) ? " 2k" : "",
985 imsm_orom_has_chunk(orom
, 4) ? " 4k" : "",
986 imsm_orom_has_chunk(orom
, 8) ? " 8k" : "",
987 imsm_orom_has_chunk(orom
, 16) ? " 16k" : "",
988 imsm_orom_has_chunk(orom
, 32) ? " 32k" : "",
989 imsm_orom_has_chunk(orom
, 64) ? " 64k" : "",
990 imsm_orom_has_chunk(orom
, 128) ? " 128k" : "",
991 imsm_orom_has_chunk(orom
, 256) ? " 256k" : "",
992 imsm_orom_has_chunk(orom
, 512) ? " 512k" : "",
993 imsm_orom_has_chunk(orom
, 1024*1) ? " 1M" : "",
994 imsm_orom_has_chunk(orom
, 1024*2) ? " 2M" : "",
995 imsm_orom_has_chunk(orom
, 1024*4) ? " 4M" : "",
996 imsm_orom_has_chunk(orom
, 1024*8) ? " 8M" : "",
997 imsm_orom_has_chunk(orom
, 1024*16) ? " 16M" : "",
998 imsm_orom_has_chunk(orom
, 1024*32) ? " 32M" : "",
999 imsm_orom_has_chunk(orom
, 1024*64) ? " 64M" : "");
1000 printf(" Max Disks : %d\n", orom
->tds
);
1001 printf(" Max Volumes : %d\n", orom
->vpa
);
1002 printf(" I/O Controller : %s\n", hba_path
);
1004 /* find the smallest scsi host number to determine a port number base */
1005 dir
= opendir(hba_path
);
1006 for (ent
= dir
? readdir(dir
) : NULL
; ent
; ent
= readdir(dir
)) {
1009 if (sscanf(ent
->d_name
, "host%d", &host
) != 1)
1011 if (port_count
== 0)
1013 else if (host
< host_base
)
1016 if (host
+ 1 > port_count
+ host_base
)
1017 port_count
= host
+ 1 - host_base
;
1023 if (!port_count
|| imsm_enumerate_ports(hba_path
, port_count
,
1024 host_base
, verbose
) != 0) {
1026 fprintf(stderr
, Name
": failed to enumerate ports\n");
1034 static int match_home_imsm(struct supertype
*st
, char *homehost
)
1036 /* the imsm metadata format does not specify any host
1037 * identification information. We return -1 since we can never
1038 * confirm nor deny whether a given array is "meant" for this
1039 * host. We rely on compare_super and the 'family_num' field to
1040 * exclude member disks that do not belong, and we rely on
1041 * mdadm.conf to specify the arrays that should be assembled.
1042 * Auto-assembly may still pick up "foreign" arrays.
1048 static void uuid_from_super_imsm(struct supertype
*st
, int uuid
[4])
1050 /* The uuid returned here is used for:
1051 * uuid to put into bitmap file (Create, Grow)
1052 * uuid for backup header when saving critical section (Grow)
1053 * comparing uuids when re-adding a device into an array
1054 * In these cases the uuid required is that of the data-array,
1055 * not the device-set.
1056 * uuid to recognise same set when adding a missing device back
1057 * to an array. This is a uuid for the device-set.
1059 * For each of these we can make do with a truncated
1060 * or hashed uuid rather than the original, as long as
1062 * In each case the uuid required is that of the data-array,
1063 * not the device-set.
1065 /* imsm does not track uuid's so we synthesis one using sha1 on
1066 * - The signature (Which is constant for all imsm array, but no matter)
1067 * - the family_num of the container
1068 * - the index number of the volume
1069 * - the 'serial' number of the volume.
1070 * Hopefully these are all constant.
1072 struct intel_super
*super
= st
->sb
;
1075 struct sha1_ctx ctx
;
1076 struct imsm_dev
*dev
= NULL
;
1078 sha1_init_ctx(&ctx
);
1079 sha1_process_bytes(super
->anchor
->sig
, MPB_SIG_LEN
, &ctx
);
1080 sha1_process_bytes(&super
->anchor
->family_num
, sizeof(__u32
), &ctx
);
1081 if (super
->current_vol
>= 0)
1082 dev
= get_imsm_dev(super
, super
->current_vol
);
1084 __u32 vol
= super
->current_vol
;
1085 sha1_process_bytes(&vol
, sizeof(vol
), &ctx
);
1086 sha1_process_bytes(dev
->volume
, MAX_RAID_SERIAL_LEN
, &ctx
);
1088 sha1_finish_ctx(&ctx
, buf
);
1089 memcpy(uuid
, buf
, 4*4);
1094 get_imsm_numerical_version(struct imsm_super
*mpb
, int *m
, int *p
)
1096 __u8
*v
= get_imsm_version(mpb
);
1097 __u8
*end
= mpb
->sig
+ MAX_SIGNATURE_LENGTH
;
1098 char major
[] = { 0, 0, 0 };
1099 char minor
[] = { 0 ,0, 0 };
1100 char patch
[] = { 0, 0, 0 };
1101 char *ver_parse
[] = { major
, minor
, patch
};
1105 while (*v
!= '\0' && v
< end
) {
1106 if (*v
!= '.' && j
< 2)
1107 ver_parse
[i
][j
++] = *v
;
1115 *m
= strtol(minor
, NULL
, 0);
1116 *p
= strtol(patch
, NULL
, 0);
1120 static int imsm_level_to_layout(int level
)
1128 return ALGORITHM_LEFT_ASYMMETRIC
;
1135 static void getinfo_super_imsm_volume(struct supertype
*st
, struct mdinfo
*info
)
1137 struct intel_super
*super
= st
->sb
;
1138 struct imsm_dev
*dev
= get_imsm_dev(super
, super
->current_vol
);
1139 struct imsm_map
*map
= get_imsm_map(dev
, 0);
1142 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
1143 if (dl
->raiddisk
== info
->disk
.raid_disk
)
1145 info
->container_member
= super
->current_vol
;
1146 info
->array
.raid_disks
= map
->num_members
;
1147 info
->array
.level
= get_imsm_raid_level(map
);
1148 info
->array
.layout
= imsm_level_to_layout(info
->array
.level
);
1149 info
->array
.md_minor
= -1;
1150 info
->array
.ctime
= 0;
1151 info
->array
.utime
= 0;
1152 info
->array
.chunk_size
= __le16_to_cpu(map
->blocks_per_strip
) << 9;
1153 info
->array
.state
= !dev
->vol
.dirty
;
1155 info
->disk
.major
= 0;
1156 info
->disk
.minor
= 0;
1158 info
->disk
.major
= dl
->major
;
1159 info
->disk
.minor
= dl
->minor
;
1162 info
->data_offset
= __le32_to_cpu(map
->pba_of_lba0
);
1163 info
->component_size
= __le32_to_cpu(map
->blocks_per_member
);
1164 memset(info
->uuid
, 0, sizeof(info
->uuid
));
1166 if (map
->map_state
== IMSM_T_STATE_UNINITIALIZED
|| dev
->vol
.dirty
)
1167 info
->resync_start
= 0;
1168 else if (dev
->vol
.migr_state
)
1169 info
->resync_start
= __le32_to_cpu(dev
->vol
.curr_migr_unit
);
1171 info
->resync_start
= ~0ULL;
1173 strncpy(info
->name
, (char *) dev
->volume
, MAX_RAID_SERIAL_LEN
);
1174 info
->name
[MAX_RAID_SERIAL_LEN
] = 0;
1176 info
->array
.major_version
= -1;
1177 info
->array
.minor_version
= -2;
1178 sprintf(info
->text_version
, "/%s/%d",
1179 devnum2devname(st
->container_dev
),
1180 info
->container_member
);
1181 info
->safe_mode_delay
= 4000; /* 4 secs like the Matrix driver */
1182 uuid_from_super_imsm(st
, info
->uuid
);
1185 /* check the config file to see if we can return a real uuid for this spare */
1186 static void fixup_container_spare_uuid(struct mdinfo
*inf
)
1188 struct mddev_ident_s
*array_list
;
1190 if (inf
->array
.level
!= LEVEL_CONTAINER
||
1191 memcmp(inf
->uuid
, uuid_match_any
, sizeof(int[4])) != 0)
1194 array_list
= conf_get_ident(NULL
);
1196 for (; array_list
; array_list
= array_list
->next
) {
1197 if (array_list
->uuid_set
) {
1198 struct supertype
*_sst
; /* spare supertype */
1199 struct supertype
*_cst
; /* container supertype */
1201 _cst
= array_list
->st
;
1202 _sst
= _cst
->ss
->match_metadata_desc(inf
->text_version
);
1204 memcpy(inf
->uuid
, array_list
->uuid
, sizeof(int[4]));
1212 static void getinfo_super_imsm(struct supertype
*st
, struct mdinfo
*info
)
1214 struct intel_super
*super
= st
->sb
;
1215 struct imsm_disk
*disk
;
1218 if (super
->current_vol
>= 0) {
1219 getinfo_super_imsm_volume(st
, info
);
1223 /* Set raid_disks to zero so that Assemble will always pull in valid
1226 info
->array
.raid_disks
= 0;
1227 info
->array
.level
= LEVEL_CONTAINER
;
1228 info
->array
.layout
= 0;
1229 info
->array
.md_minor
= -1;
1230 info
->array
.ctime
= 0; /* N/A for imsm */
1231 info
->array
.utime
= 0;
1232 info
->array
.chunk_size
= 0;
1234 info
->disk
.major
= 0;
1235 info
->disk
.minor
= 0;
1236 info
->disk
.raid_disk
= -1;
1237 info
->reshape_active
= 0;
1238 info
->array
.major_version
= -1;
1239 info
->array
.minor_version
= -2;
1240 strcpy(info
->text_version
, "imsm");
1241 info
->safe_mode_delay
= 0;
1242 info
->disk
.number
= -1;
1243 info
->disk
.state
= 0;
1247 __u32 reserved
= imsm_reserved_sectors(super
, super
->disks
);
1249 disk
= &super
->disks
->disk
;
1250 info
->data_offset
= __le32_to_cpu(disk
->total_blocks
) - reserved
;
1251 info
->component_size
= reserved
;
1253 info
->disk
.state
= s
& CONFIGURED_DISK
? (1 << MD_DISK_ACTIVE
) : 0;
1254 /* we don't change info->disk.raid_disk here because
1255 * this state will be finalized in mdmon after we have
1256 * found the 'most fresh' version of the metadata
1258 info
->disk
.state
|= s
& FAILED_DISK
? (1 << MD_DISK_FAULTY
) : 0;
1259 info
->disk
.state
|= s
& SPARE_DISK
? 0 : (1 << MD_DISK_SYNC
);
1262 /* only call uuid_from_super_imsm when this disk is part of a populated container,
1263 * ->compare_super may have updated the 'num_raid_devs' field for spares
1265 if (info
->disk
.state
& (1 << MD_DISK_SYNC
) || super
->anchor
->num_raid_devs
)
1266 uuid_from_super_imsm(st
, info
->uuid
);
1268 memcpy(info
->uuid
, uuid_match_any
, sizeof(int[4]));
1269 fixup_container_spare_uuid(info
);
1273 static int update_super_imsm(struct supertype
*st
, struct mdinfo
*info
,
1274 char *update
, char *devname
, int verbose
,
1275 int uuid_set
, char *homehost
)
1279 /* For 'assemble' and 'force' we need to return non-zero if any
1280 * change was made. For others, the return value is ignored.
1281 * Update options are:
1282 * force-one : This device looks a bit old but needs to be included,
1283 * update age info appropriately.
1284 * assemble: clear any 'faulty' flag to allow this device to
1286 * force-array: Array is degraded but being forced, mark it clean
1287 * if that will be needed to assemble it.
1289 * newdev: not used ????
1290 * grow: Array has gained a new device - this is currently for
1292 * resync: mark as dirty so a resync will happen.
1293 * name: update the name - preserving the homehost
1295 * Following are not relevant for this imsm:
1296 * sparc2.2 : update from old dodgey metadata
1297 * super-minor: change the preferred_minor number
1298 * summaries: update redundant counters.
1299 * uuid: Change the uuid of the array to match watch is given
1300 * homehost: update the recorded homehost
1301 * _reshape_progress: record new reshape_progress position.
1304 //struct intel_super *super = st->sb;
1305 //struct imsm_super *mpb = super->mpb;
1307 if (strcmp(update
, "grow") == 0) {
1309 if (strcmp(update
, "resync") == 0) {
1310 /* dev->vol.dirty = 1; */
1313 /* IMSM has no concept of UUID or homehost */
1318 static size_t disks_to_mpb_size(int disks
)
1322 size
= sizeof(struct imsm_super
);
1323 size
+= (disks
- 1) * sizeof(struct imsm_disk
);
1324 size
+= 2 * sizeof(struct imsm_dev
);
1325 /* up to 2 maps per raid device (-2 for imsm_maps in imsm_dev */
1326 size
+= (4 - 2) * sizeof(struct imsm_map
);
1327 /* 4 possible disk_ord_tbl's */
1328 size
+= 4 * (disks
- 1) * sizeof(__u32
);
1333 static __u64
avail_size_imsm(struct supertype
*st
, __u64 devsize
)
1335 if (devsize
< (MPB_SECTOR_CNT
+ IMSM_RESERVED_SECTORS
))
1338 return devsize
- (MPB_SECTOR_CNT
+ IMSM_RESERVED_SECTORS
);
1341 static void free_devlist(struct intel_super
*super
)
1343 struct intel_dev
*dv
;
1345 while (super
->devlist
) {
1346 dv
= super
->devlist
->next
;
1347 free(super
->devlist
->dev
);
1348 free(super
->devlist
);
1349 super
->devlist
= dv
;
1353 static void imsm_copy_dev(struct imsm_dev
*dest
, struct imsm_dev
*src
)
1355 memcpy(dest
, src
, sizeof_imsm_dev(src
, 0));
1358 static int compare_super_imsm(struct supertype
*st
, struct supertype
*tst
)
1362 * 0 same, or first was empty, and second was copied
1363 * 1 second had wrong number
1365 * 3 wrong other info
1367 struct intel_super
*first
= st
->sb
;
1368 struct intel_super
*sec
= tst
->sb
;
1376 if (memcmp(first
->anchor
->sig
, sec
->anchor
->sig
, MAX_SIGNATURE_LENGTH
) != 0)
1379 /* if an anchor does not have num_raid_devs set then it is a free
1382 if (first
->anchor
->num_raid_devs
> 0 &&
1383 sec
->anchor
->num_raid_devs
> 0) {
1384 if (first
->anchor
->family_num
!= sec
->anchor
->family_num
)
1388 /* if 'first' is a spare promote it to a populated mpb with sec's
1391 if (first
->anchor
->num_raid_devs
== 0 &&
1392 sec
->anchor
->num_raid_devs
> 0) {
1394 struct intel_dev
*dv
;
1395 struct imsm_dev
*dev
;
1397 /* we need to copy raid device info from sec if an allocation
1398 * fails here we don't associate the spare
1400 for (i
= 0; i
< sec
->anchor
->num_raid_devs
; i
++) {
1401 dv
= malloc(sizeof(*dv
));
1404 dev
= malloc(sizeof_imsm_dev(get_imsm_dev(sec
, i
), 1));
1411 dv
->next
= first
->devlist
;
1412 first
->devlist
= dv
;
1414 if (i
<= sec
->anchor
->num_raid_devs
) {
1415 /* allocation failure */
1416 free_devlist(first
);
1417 fprintf(stderr
, "imsm: failed to associate spare\n");
1420 for (i
= 0; i
< sec
->anchor
->num_raid_devs
; i
++)
1421 imsm_copy_dev(get_imsm_dev(first
, i
), get_imsm_dev(sec
, i
));
1423 first
->anchor
->num_raid_devs
= sec
->anchor
->num_raid_devs
;
1424 first
->anchor
->family_num
= sec
->anchor
->family_num
;
1430 static void fd2devname(int fd
, char *name
)
1439 if (fstat(fd
, &st
) != 0)
1441 sprintf(path
, "/sys/dev/block/%d:%d",
1442 major(st
.st_rdev
), minor(st
.st_rdev
));
1444 rv
= readlink(path
, dname
, sizeof(dname
));
1449 nm
= strrchr(dname
, '/');
1451 snprintf(name
, MAX_RAID_SERIAL_LEN
, "/dev/%s", nm
);
1455 extern int scsi_get_serial(int fd
, void *buf
, size_t buf_len
);
1457 static int imsm_read_serial(int fd
, char *devname
,
1458 __u8 serial
[MAX_RAID_SERIAL_LEN
])
1460 unsigned char scsi_serial
[255];
1466 memset(scsi_serial
, 0, sizeof(scsi_serial
));
1468 rv
= scsi_get_serial(fd
, scsi_serial
, sizeof(scsi_serial
));
1470 if (rv
&& check_env("IMSM_DEVNAME_AS_SERIAL")) {
1471 memset(serial
, 0, MAX_RAID_SERIAL_LEN
);
1472 fd2devname(fd
, (char *) serial
);
1479 Name
": Failed to retrieve serial for %s\n",
1484 /* trim leading whitespace */
1485 rsp_len
= scsi_serial
[3];
1489 Name
": Failed to retrieve serial for %s\n",
1493 rsp_buf
= (char *) &scsi_serial
[4];
1498 /* truncate len to the end of rsp_buf if necessary */
1499 if (c
+ MAX_RAID_SERIAL_LEN
> rsp_buf
+ rsp_len
)
1500 len
= rsp_len
- (c
- rsp_buf
);
1502 len
= MAX_RAID_SERIAL_LEN
;
1504 /* initialize the buffer and copy rsp_buf characters */
1505 memset(serial
, 0, MAX_RAID_SERIAL_LEN
);
1506 memcpy(serial
, c
, len
);
1508 /* trim trailing whitespace starting with the last character copied */
1509 c
= (char *) &serial
[len
- 1];
1510 while (isspace(*c
) || *c
== '\0')
1516 static int serialcmp(__u8
*s1
, __u8
*s2
)
1518 return strncmp((char *) s1
, (char *) s2
, MAX_RAID_SERIAL_LEN
);
1521 static void serialcpy(__u8
*dest
, __u8
*src
)
1523 strncpy((char *) dest
, (char *) src
, MAX_RAID_SERIAL_LEN
);
1526 static struct dl
*serial_to_dl(__u8
*serial
, struct intel_super
*super
)
1530 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
1531 if (serialcmp(dl
->serial
, serial
) == 0)
1538 load_imsm_disk(int fd
, struct intel_super
*super
, char *devname
, int keep_fd
)
1545 __u8 serial
[MAX_RAID_SERIAL_LEN
];
1547 rv
= imsm_read_serial(fd
, devname
, serial
);
1552 /* check if this is a disk we have seen before. it may be a spare in
1553 * super->disks while the current anchor believes it is a raid member,
1554 * check if we need to update dl->index
1556 dl
= serial_to_dl(serial
, super
);
1558 dl
= malloc(sizeof(*dl
));
1565 Name
": failed to allocate disk buffer for %s\n",
1572 dl
->major
= major(stb
.st_rdev
);
1573 dl
->minor
= minor(stb
.st_rdev
);
1574 dl
->next
= super
->disks
;
1575 dl
->fd
= keep_fd
? fd
: -1;
1576 dl
->devname
= devname
? strdup(devname
) : NULL
;
1577 serialcpy(dl
->serial
, serial
);
1580 } else if (keep_fd
) {
1585 /* look up this disk's index in the current anchor */
1586 for (i
= 0; i
< super
->anchor
->num_disks
; i
++) {
1587 struct imsm_disk
*disk_iter
;
1589 disk_iter
= __get_imsm_disk(super
->anchor
, i
);
1591 if (serialcmp(disk_iter
->serial
, dl
->serial
) == 0) {
1592 dl
->disk
= *disk_iter
;
1593 /* only set index on disks that are a member of a
1594 * populated contianer, i.e. one with raid_devs
1596 if (dl
->disk
.status
& FAILED_DISK
)
1598 else if (dl
->disk
.status
& SPARE_DISK
)
1607 /* no match, maybe a stale failed drive */
1608 if (i
== super
->anchor
->num_disks
&& dl
->index
>= 0) {
1609 dl
->disk
= *__get_imsm_disk(super
->anchor
, dl
->index
);
1610 if (dl
->disk
.status
& FAILED_DISK
)
1621 /* When migrating map0 contains the 'destination' state while map1
1622 * contains the current state. When not migrating map0 contains the
1623 * current state. This routine assumes that map[0].map_state is set to
1624 * the current array state before being called.
1626 * Migration is indicated by one of the following states
1627 * 1/ Idle (migr_state=0 map0state=normal||unitialized||degraded||failed)
1628 * 2/ Initialize (migr_state=1 migr_type=MIGR_INIT map0state=normal
1629 * map1state=unitialized)
1630 * 3/ Verify (Resync) (migr_state=1 migr_type=MIGR_REBUILD map0state=normal
1632 * 4/ Rebuild (migr_state=1 migr_type=MIGR_REBUILD map0state=normal
1633 * map1state=degraded)
1635 static void migrate(struct imsm_dev
*dev
, __u8 to_state
, int migr_type
)
1637 struct imsm_map
*dest
;
1638 struct imsm_map
*src
= get_imsm_map(dev
, 0);
1640 dev
->vol
.migr_state
= 1;
1641 dev
->vol
.migr_type
= migr_type
;
1642 dev
->vol
.curr_migr_unit
= 0;
1643 dest
= get_imsm_map(dev
, 1);
1645 /* duplicate and then set the target end state in map[0] */
1646 memcpy(dest
, src
, sizeof_imsm_map(src
));
1647 if (migr_type
== MIGR_REBUILD
) {
1651 for (i
= 0; i
< src
->num_members
; i
++) {
1652 ord
= __le32_to_cpu(src
->disk_ord_tbl
[i
]);
1653 set_imsm_ord_tbl_ent(src
, i
, ord_to_idx(ord
));
1657 src
->map_state
= to_state
;
1660 static void end_migration(struct imsm_dev
*dev
, __u8 map_state
)
1662 struct imsm_map
*map
= get_imsm_map(dev
, 0);
1663 struct imsm_map
*prev
= get_imsm_map(dev
, dev
->vol
.migr_state
);
1666 /* merge any IMSM_ORD_REBUILD bits that were not successfully
1667 * completed in the last migration.
1669 * FIXME add support for online capacity expansion and
1670 * raid-level-migration
1672 for (i
= 0; i
< prev
->num_members
; i
++)
1673 map
->disk_ord_tbl
[i
] |= prev
->disk_ord_tbl
[i
];
1675 dev
->vol
.migr_state
= 0;
1676 dev
->vol
.curr_migr_unit
= 0;
1677 map
->map_state
= map_state
;
1681 static int parse_raid_devices(struct intel_super
*super
)
1684 struct imsm_dev
*dev_new
;
1685 size_t len
, len_migr
;
1686 size_t space_needed
= 0;
1687 struct imsm_super
*mpb
= super
->anchor
;
1689 for (i
= 0; i
< super
->anchor
->num_raid_devs
; i
++) {
1690 struct imsm_dev
*dev_iter
= __get_imsm_dev(super
->anchor
, i
);
1691 struct intel_dev
*dv
;
1693 len
= sizeof_imsm_dev(dev_iter
, 0);
1694 len_migr
= sizeof_imsm_dev(dev_iter
, 1);
1696 space_needed
+= len_migr
- len
;
1698 dv
= malloc(sizeof(*dv
));
1701 dev_new
= malloc(len_migr
);
1706 imsm_copy_dev(dev_new
, dev_iter
);
1709 dv
->next
= super
->devlist
;
1710 super
->devlist
= dv
;
1713 /* ensure that super->buf is large enough when all raid devices
1716 if (__le32_to_cpu(mpb
->mpb_size
) + space_needed
> super
->len
) {
1719 len
= ROUND_UP(__le32_to_cpu(mpb
->mpb_size
) + space_needed
, 512);
1720 if (posix_memalign(&buf
, 512, len
) != 0)
1723 memcpy(buf
, super
->buf
, len
);
1732 /* retrieve a pointer to the bbm log which starts after all raid devices */
1733 struct bbm_log
*__get_imsm_bbm_log(struct imsm_super
*mpb
)
1737 if (__le32_to_cpu(mpb
->bbm_log_size
)) {
1739 ptr
+= mpb
->mpb_size
- __le32_to_cpu(mpb
->bbm_log_size
);
1745 static void __free_imsm(struct intel_super
*super
, int free_disks
);
1747 /* load_imsm_mpb - read matrix metadata
1748 * allocates super->mpb to be freed by free_super
1750 static int load_imsm_mpb(int fd
, struct intel_super
*super
, char *devname
)
1752 unsigned long long dsize
;
1753 unsigned long long sectors
;
1755 struct imsm_super
*anchor
;
1759 get_dev_size(fd
, NULL
, &dsize
);
1761 if (lseek64(fd
, dsize
- (512 * 2), SEEK_SET
) < 0) {
1764 Name
": Cannot seek to anchor block on %s: %s\n",
1765 devname
, strerror(errno
));
1769 if (posix_memalign((void**)&anchor
, 512, 512) != 0) {
1772 Name
": Failed to allocate imsm anchor buffer"
1773 " on %s\n", devname
);
1776 if (read(fd
, anchor
, 512) != 512) {
1779 Name
": Cannot read anchor block on %s: %s\n",
1780 devname
, strerror(errno
));
1785 if (strncmp((char *) anchor
->sig
, MPB_SIGNATURE
, MPB_SIG_LEN
) != 0) {
1788 Name
": no IMSM anchor on %s\n", devname
);
1793 __free_imsm(super
, 0);
1794 super
->len
= ROUND_UP(anchor
->mpb_size
, 512);
1795 if (posix_memalign(&super
->buf
, 512, super
->len
) != 0) {
1798 Name
": unable to allocate %zu byte mpb buffer\n",
1803 memcpy(super
->buf
, anchor
, 512);
1805 sectors
= mpb_sectors(anchor
) - 1;
1808 check_sum
= __gen_imsm_checksum(super
->anchor
);
1809 if (check_sum
!= __le32_to_cpu(super
->anchor
->check_sum
)) {
1812 Name
": IMSM checksum %x != %x on %s\n",
1814 __le32_to_cpu(super
->anchor
->check_sum
),
1819 rc
= load_imsm_disk(fd
, super
, devname
, 0);
1821 rc
= parse_raid_devices(super
);
1825 /* read the extended mpb */
1826 if (lseek64(fd
, dsize
- (512 * (2 + sectors
)), SEEK_SET
) < 0) {
1829 Name
": Cannot seek to extended mpb on %s: %s\n",
1830 devname
, strerror(errno
));
1834 if (read(fd
, super
->buf
+ 512, super
->len
- 512) != super
->len
- 512) {
1837 Name
": Cannot read extended mpb on %s: %s\n",
1838 devname
, strerror(errno
));
1842 check_sum
= __gen_imsm_checksum(super
->anchor
);
1843 if (check_sum
!= __le32_to_cpu(super
->anchor
->check_sum
)) {
1846 Name
": IMSM checksum %x != %x on %s\n",
1847 check_sum
, __le32_to_cpu(super
->anchor
->check_sum
),
1852 /* FIXME the BBM log is disk specific so we cannot use this global
1853 * buffer for all disks. Ok for now since we only look at the global
1854 * bbm_log_size parameter to gate assembly
1856 super
->bbm_log
= __get_imsm_bbm_log(super
->anchor
);
1858 rc
= load_imsm_disk(fd
, super
, devname
, 0);
1860 rc
= parse_raid_devices(super
);
1865 static void __free_imsm_disk(struct dl
*d
)
1876 static void free_imsm_disks(struct intel_super
*super
)
1880 while (super
->disks
) {
1882 super
->disks
= d
->next
;
1883 __free_imsm_disk(d
);
1885 while (super
->missing
) {
1887 super
->missing
= d
->next
;
1888 __free_imsm_disk(d
);
1893 /* free all the pieces hanging off of a super pointer */
1894 static void __free_imsm(struct intel_super
*super
, int free_disks
)
1901 free_imsm_disks(super
);
1902 free_devlist(super
);
1904 free((void *) super
->hba
);
1909 static void free_imsm(struct intel_super
*super
)
1911 __free_imsm(super
, 1);
1915 static void free_super_imsm(struct supertype
*st
)
1917 struct intel_super
*super
= st
->sb
;
1926 static struct intel_super
*alloc_super(int creating_imsm
)
1928 struct intel_super
*super
= malloc(sizeof(*super
));
1931 memset(super
, 0, sizeof(*super
));
1932 super
->creating_imsm
= creating_imsm
;
1933 super
->current_vol
= -1;
1934 super
->create_offset
= ~((__u32
) 0);
1935 if (!check_env("IMSM_NO_PLATFORM"))
1936 super
->orom
= find_imsm_orom();
1937 if (super
->orom
&& !check_env("IMSM_TEST_OROM")) {
1938 struct sys_dev
*list
, *ent
;
1940 /* find the first intel ahci controller */
1941 list
= find_driver_devices("pci", "ahci");
1942 for (ent
= list
; ent
; ent
= ent
->next
)
1943 if (devpath_to_vendor(ent
->path
) == 0x8086)
1946 super
->hba
= ent
->path
;
1949 free_sys_dev(&list
);
1957 /* find_missing - helper routine for load_super_imsm_all that identifies
1958 * disks that have disappeared from the system. This routine relies on
1959 * the mpb being uptodate, which it is at load time.
1961 static int find_missing(struct intel_super
*super
)
1964 struct imsm_super
*mpb
= super
->anchor
;
1966 struct imsm_disk
*disk
;
1968 for (i
= 0; i
< mpb
->num_disks
; i
++) {
1969 disk
= __get_imsm_disk(mpb
, i
);
1970 dl
= serial_to_dl(disk
->serial
, super
);
1974 dl
= malloc(sizeof(*dl
));
1980 dl
->devname
= strdup("missing");
1982 serialcpy(dl
->serial
, disk
->serial
);
1985 dl
->next
= super
->missing
;
1986 super
->missing
= dl
;
1992 static int load_super_imsm_all(struct supertype
*st
, int fd
, void **sbp
,
1993 char *devname
, int keep_fd
)
1996 struct intel_super
*super
;
1997 struct mdinfo
*sd
, *best
= NULL
;
2003 int devnum
= fd2devnum(fd
);
2005 enum sysfs_read_flags flags
;
2007 flags
= GET_LEVEL
|GET_VERSION
|GET_DEVS
|GET_STATE
;
2008 if (mdmon_running(devnum
))
2009 flags
|= SKIP_GONE_DEVS
;
2011 /* check if 'fd' an opened container */
2012 sra
= sysfs_read(fd
, 0, flags
);
2016 if (sra
->array
.major_version
!= -1 ||
2017 sra
->array
.minor_version
!= -2 ||
2018 strcmp(sra
->text_version
, "imsm") != 0)
2021 super
= alloc_super(0);
2025 /* find the most up to date disk in this array, skipping spares */
2026 for (sd
= sra
->devs
; sd
; sd
= sd
->next
) {
2027 sprintf(nm
, "%d:%d", sd
->disk
.major
, sd
->disk
.minor
);
2028 dfd
= dev_open(nm
, keep_fd
? O_RDWR
: O_RDONLY
);
2033 rv
= load_imsm_mpb(dfd
, super
, NULL
);
2035 /* retry the load if we might have raced against mdmon */
2036 if (rv
== 3 && mdmon_running(devnum
))
2037 for (retry
= 0; retry
< 3; retry
++) {
2039 rv
= load_imsm_mpb(dfd
, super
, NULL
);
2046 if (super
->anchor
->num_raid_devs
== 0)
2049 gen
= __le32_to_cpu(super
->anchor
->generation_num
);
2050 if (!best
|| gen
> bestgen
) {
2065 /* load the most up to date anchor */
2066 sprintf(nm
, "%d:%d", best
->disk
.major
, best
->disk
.minor
);
2067 dfd
= dev_open(nm
, O_RDONLY
);
2072 rv
= load_imsm_mpb(dfd
, super
, NULL
);
2079 /* re-parse the disk list with the current anchor */
2080 for (sd
= sra
->devs
; sd
; sd
= sd
->next
) {
2081 sprintf(nm
, "%d:%d", sd
->disk
.major
, sd
->disk
.minor
);
2082 dfd
= dev_open(nm
, keep_fd
? O_RDWR
: O_RDONLY
);
2087 load_imsm_disk(dfd
, super
, NULL
, keep_fd
);
2093 if (find_missing(super
) != 0) {
2098 if (st
->subarray
[0]) {
2099 if (atoi(st
->subarray
) <= super
->anchor
->num_raid_devs
)
2100 super
->current_vol
= atoi(st
->subarray
);
2106 st
->container_dev
= devnum
;
2107 if (st
->ss
== NULL
) {
2108 st
->ss
= &super_imsm
;
2109 st
->minor_version
= 0;
2110 st
->max_devs
= IMSM_MAX_DEVICES
;
2112 st
->loaded_container
= 1;
2118 static int load_super_imsm(struct supertype
*st
, int fd
, char *devname
)
2120 struct intel_super
*super
;
2124 if (load_super_imsm_all(st
, fd
, &st
->sb
, devname
, 1) == 0)
2127 if (st
->subarray
[0])
2128 return 1; /* FIXME */
2130 super
= alloc_super(0);
2133 Name
": malloc of %zu failed.\n",
2138 rv
= load_imsm_mpb(fd
, super
, devname
);
2143 Name
": Failed to load all information "
2144 "sections on %s\n", devname
);
2150 if (st
->ss
== NULL
) {
2151 st
->ss
= &super_imsm
;
2152 st
->minor_version
= 0;
2153 st
->max_devs
= IMSM_MAX_DEVICES
;
2155 st
->loaded_container
= 0;
2160 static __u16
info_to_blocks_per_strip(mdu_array_info_t
*info
)
2162 if (info
->level
== 1)
2164 return info
->chunk_size
>> 9;
2167 static __u32
info_to_num_data_stripes(mdu_array_info_t
*info
)
2171 num_stripes
= (info
->size
* 2) / info_to_blocks_per_strip(info
);
2172 if (info
->level
== 1)
2178 static __u32
info_to_blocks_per_member(mdu_array_info_t
*info
)
2180 if (info
->level
== 1)
2181 return info
->size
* 2;
2183 return (info
->size
* 2) & ~(info_to_blocks_per_strip(info
) - 1);
2186 static void imsm_update_version_info(struct intel_super
*super
)
2188 /* update the version and attributes */
2189 struct imsm_super
*mpb
= super
->anchor
;
2191 struct imsm_dev
*dev
;
2192 struct imsm_map
*map
;
2195 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
2196 dev
= get_imsm_dev(super
, i
);
2197 map
= get_imsm_map(dev
, 0);
2198 if (__le32_to_cpu(dev
->size_high
) > 0)
2199 mpb
->attributes
|= MPB_ATTRIB_2TB
;
2201 /* FIXME detect when an array spans a port multiplier */
2203 mpb
->attributes
|= MPB_ATTRIB_PM
;
2206 if (mpb
->num_raid_devs
> 1 ||
2207 mpb
->attributes
!= MPB_ATTRIB_CHECKSUM_VERIFY
) {
2208 version
= MPB_VERSION_ATTRIBS
;
2209 switch (get_imsm_raid_level(map
)) {
2210 case 0: mpb
->attributes
|= MPB_ATTRIB_RAID0
; break;
2211 case 1: mpb
->attributes
|= MPB_ATTRIB_RAID1
; break;
2212 case 10: mpb
->attributes
|= MPB_ATTRIB_RAID10
; break;
2213 case 5: mpb
->attributes
|= MPB_ATTRIB_RAID5
; break;
2216 if (map
->num_members
>= 5)
2217 version
= MPB_VERSION_5OR6_DISK_ARRAY
;
2218 else if (dev
->status
== DEV_CLONE_N_GO
)
2219 version
= MPB_VERSION_CNG
;
2220 else if (get_imsm_raid_level(map
) == 5)
2221 version
= MPB_VERSION_RAID5
;
2222 else if (map
->num_members
>= 3)
2223 version
= MPB_VERSION_3OR4_DISK_ARRAY
;
2224 else if (get_imsm_raid_level(map
) == 1)
2225 version
= MPB_VERSION_RAID1
;
2227 version
= MPB_VERSION_RAID0
;
2229 strcpy(((char *) mpb
->sig
) + strlen(MPB_SIGNATURE
), version
);
2233 static int init_super_imsm_volume(struct supertype
*st
, mdu_array_info_t
*info
,
2234 unsigned long long size
, char *name
,
2235 char *homehost
, int *uuid
)
2237 /* We are creating a volume inside a pre-existing container.
2238 * so st->sb is already set.
2240 struct intel_super
*super
= st
->sb
;
2241 struct imsm_super
*mpb
= super
->anchor
;
2242 struct intel_dev
*dv
;
2243 struct imsm_dev
*dev
;
2244 struct imsm_vol
*vol
;
2245 struct imsm_map
*map
;
2246 int idx
= mpb
->num_raid_devs
;
2248 unsigned long long array_blocks
;
2249 size_t size_old
, size_new
;
2251 if (super
->orom
&& mpb
->num_raid_devs
>= super
->orom
->vpa
) {
2252 fprintf(stderr
, Name
": This imsm-container already has the "
2253 "maximum of %d volumes\n", super
->orom
->vpa
);
2257 /* ensure the mpb is large enough for the new data */
2258 size_old
= __le32_to_cpu(mpb
->mpb_size
);
2259 size_new
= disks_to_mpb_size(info
->nr_disks
);
2260 if (size_new
> size_old
) {
2262 size_t size_round
= ROUND_UP(size_new
, 512);
2264 if (posix_memalign(&mpb_new
, 512, size_round
) != 0) {
2265 fprintf(stderr
, Name
": could not allocate new mpb\n");
2268 memcpy(mpb_new
, mpb
, size_old
);
2271 super
->anchor
= mpb_new
;
2272 mpb
->mpb_size
= __cpu_to_le32(size_new
);
2273 memset(mpb_new
+ size_old
, 0, size_round
- size_old
);
2275 super
->current_vol
= idx
;
2276 /* when creating the first raid device in this container set num_disks
2277 * to zero, i.e. delete this spare and add raid member devices in
2278 * add_to_super_imsm_volume()
2280 if (super
->current_vol
== 0)
2283 for (i
= 0; i
< super
->current_vol
; i
++) {
2284 dev
= get_imsm_dev(super
, i
);
2285 if (strncmp((char *) dev
->volume
, name
,
2286 MAX_RAID_SERIAL_LEN
) == 0) {
2287 fprintf(stderr
, Name
": '%s' is already defined for this container\n",
2293 sprintf(st
->subarray
, "%d", idx
);
2294 dv
= malloc(sizeof(*dv
));
2296 fprintf(stderr
, Name
": failed to allocate device list entry\n");
2299 dev
= malloc(sizeof(*dev
) + sizeof(__u32
) * (info
->raid_disks
- 1));
2302 fprintf(stderr
, Name
": could not allocate raid device\n");
2305 strncpy((char *) dev
->volume
, name
, MAX_RAID_SERIAL_LEN
);
2306 if (info
->level
== 1)
2307 array_blocks
= info_to_blocks_per_member(info
);
2309 array_blocks
= calc_array_size(info
->level
, info
->raid_disks
,
2310 info
->layout
, info
->chunk_size
,
2312 dev
->size_low
= __cpu_to_le32((__u32
) array_blocks
);
2313 dev
->size_high
= __cpu_to_le32((__u32
) (array_blocks
>> 32));
2314 dev
->status
= __cpu_to_le32(0);
2315 dev
->reserved_blocks
= __cpu_to_le32(0);
2317 vol
->migr_state
= 0;
2318 vol
->migr_type
= MIGR_INIT
;
2320 vol
->curr_migr_unit
= 0;
2321 map
= get_imsm_map(dev
, 0);
2322 map
->pba_of_lba0
= __cpu_to_le32(super
->create_offset
);
2323 map
->blocks_per_member
= __cpu_to_le32(info_to_blocks_per_member(info
));
2324 map
->blocks_per_strip
= __cpu_to_le16(info_to_blocks_per_strip(info
));
2325 map
->num_data_stripes
= __cpu_to_le32(info_to_num_data_stripes(info
));
2326 map
->failed_disk_num
= ~0;
2327 map
->map_state
= info
->level
? IMSM_T_STATE_UNINITIALIZED
:
2328 IMSM_T_STATE_NORMAL
;
2330 if (info
->level
== 1 && info
->raid_disks
> 2) {
2331 fprintf(stderr
, Name
": imsm does not support more than 2 disks"
2332 "in a raid1 volume\n");
2335 if (info
->level
== 10) {
2336 map
->raid_level
= 1;
2337 map
->num_domains
= info
->raid_disks
/ 2;
2339 map
->raid_level
= info
->level
;
2340 map
->num_domains
= !!map
->raid_level
;
2343 map
->num_members
= info
->raid_disks
;
2344 for (i
= 0; i
< map
->num_members
; i
++) {
2345 /* initialized in add_to_super */
2346 set_imsm_ord_tbl_ent(map
, i
, 0);
2348 mpb
->num_raid_devs
++;
2351 dv
->index
= super
->current_vol
;
2352 dv
->next
= super
->devlist
;
2353 super
->devlist
= dv
;
2355 imsm_update_version_info(super
);
2360 static int init_super_imsm(struct supertype
*st
, mdu_array_info_t
*info
,
2361 unsigned long long size
, char *name
,
2362 char *homehost
, int *uuid
)
2364 /* This is primarily called by Create when creating a new array.
2365 * We will then get add_to_super called for each component, and then
2366 * write_init_super called to write it out to each device.
2367 * For IMSM, Create can create on fresh devices or on a pre-existing
2369 * To create on a pre-existing array a different method will be called.
2370 * This one is just for fresh drives.
2372 struct intel_super
*super
;
2373 struct imsm_super
*mpb
;
2382 return init_super_imsm_volume(st
, info
, size
, name
, homehost
,
2385 super
= alloc_super(1);
2388 mpb_size
= disks_to_mpb_size(info
->nr_disks
);
2389 if (posix_memalign(&super
->buf
, 512, mpb_size
) != 0) {
2394 memset(mpb
, 0, mpb_size
);
2396 mpb
->attributes
= MPB_ATTRIB_CHECKSUM_VERIFY
;
2398 version
= (char *) mpb
->sig
;
2399 strcpy(version
, MPB_SIGNATURE
);
2400 version
+= strlen(MPB_SIGNATURE
);
2401 strcpy(version
, MPB_VERSION_RAID0
);
2402 mpb
->mpb_size
= mpb_size
;
2409 static int add_to_super_imsm_volume(struct supertype
*st
, mdu_disk_info_t
*dk
,
2410 int fd
, char *devname
)
2412 struct intel_super
*super
= st
->sb
;
2413 struct imsm_super
*mpb
= super
->anchor
;
2415 struct imsm_dev
*dev
;
2416 struct imsm_map
*map
;
2418 dev
= get_imsm_dev(super
, super
->current_vol
);
2419 map
= get_imsm_map(dev
, 0);
2421 if (! (dk
->state
& (1<<MD_DISK_SYNC
))) {
2422 fprintf(stderr
, Name
": %s: Cannot add spare devices to IMSM volume\n",
2428 /* we're doing autolayout so grab the pre-marked (in
2429 * validate_geometry) raid_disk
2431 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
2432 if (dl
->raiddisk
== dk
->raid_disk
)
2435 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
2436 if (dl
->major
== dk
->major
&&
2437 dl
->minor
== dk
->minor
)
2442 fprintf(stderr
, Name
": %s is not a member of the same container\n", devname
);
2446 /* add a pristine spare to the metadata */
2447 if (dl
->index
< 0) {
2448 dl
->index
= super
->anchor
->num_disks
;
2449 super
->anchor
->num_disks
++;
2451 set_imsm_ord_tbl_ent(map
, dk
->number
, dl
->index
);
2452 dl
->disk
.status
= CONFIGURED_DISK
| USABLE_DISK
;
2454 /* if we are creating the first raid device update the family number */
2455 if (super
->current_vol
== 0) {
2457 struct imsm_dev
*_dev
= __get_imsm_dev(mpb
, 0);
2458 struct imsm_disk
*_disk
= __get_imsm_disk(mpb
, dl
->index
);
2462 sum
= __gen_imsm_checksum(mpb
);
2463 mpb
->family_num
= __cpu_to_le32(sum
);
2469 static int add_to_super_imsm(struct supertype
*st
, mdu_disk_info_t
*dk
,
2470 int fd
, char *devname
)
2472 struct intel_super
*super
= st
->sb
;
2474 unsigned long long size
;
2479 /* if we are on an RAID enabled platform check that the disk is
2480 * attached to the raid controller
2482 if (super
->hba
&& !disk_attached_to_hba(fd
, super
->hba
)) {
2484 Name
": %s is not attached to the raid controller: %s\n",
2485 devname
? : "disk", super
->hba
);
2489 if (super
->current_vol
>= 0)
2490 return add_to_super_imsm_volume(st
, dk
, fd
, devname
);
2493 dd
= malloc(sizeof(*dd
));
2496 Name
": malloc failed %s:%d.\n", __func__
, __LINE__
);
2499 memset(dd
, 0, sizeof(*dd
));
2500 dd
->major
= major(stb
.st_rdev
);
2501 dd
->minor
= minor(stb
.st_rdev
);
2503 dd
->devname
= devname
? strdup(devname
) : NULL
;
2506 rv
= imsm_read_serial(fd
, devname
, dd
->serial
);
2509 Name
": failed to retrieve scsi serial, aborting\n");
2514 get_dev_size(fd
, NULL
, &size
);
2516 serialcpy(dd
->disk
.serial
, dd
->serial
);
2517 dd
->disk
.total_blocks
= __cpu_to_le32(size
);
2518 dd
->disk
.status
= USABLE_DISK
| SPARE_DISK
;
2519 if (sysfs_disk_to_scsi_id(fd
, &id
) == 0)
2520 dd
->disk
.scsi_id
= __cpu_to_le32(id
);
2522 dd
->disk
.scsi_id
= __cpu_to_le32(0);
2524 if (st
->update_tail
) {
2525 dd
->next
= super
->add
;
2528 dd
->next
= super
->disks
;
2535 static int store_imsm_mpb(int fd
, struct intel_super
*super
);
2537 /* spare records have their own family number and do not have any defined raid
2540 static int write_super_imsm_spares(struct intel_super
*super
, int doclose
)
2542 struct imsm_super mpb_save
;
2543 struct imsm_super
*mpb
= super
->anchor
;
2548 mpb
->num_raid_devs
= 0;
2550 mpb
->mpb_size
= sizeof(struct imsm_super
);
2551 mpb
->generation_num
= __cpu_to_le32(1UL);
2553 for (d
= super
->disks
; d
; d
= d
->next
) {
2557 mpb
->disk
[0] = d
->disk
;
2558 sum
= __gen_imsm_checksum(mpb
);
2559 mpb
->family_num
= __cpu_to_le32(sum
);
2560 sum
= __gen_imsm_checksum(mpb
);
2561 mpb
->check_sum
= __cpu_to_le32(sum
);
2563 if (store_imsm_mpb(d
->fd
, super
)) {
2564 fprintf(stderr
, "%s: failed for device %d:%d %s\n",
2565 __func__
, d
->major
, d
->minor
, strerror(errno
));
2579 static int write_super_imsm(struct intel_super
*super
, int doclose
)
2581 struct imsm_super
*mpb
= super
->anchor
;
2587 __u32 mpb_size
= sizeof(struct imsm_super
) - sizeof(struct imsm_disk
);
2589 /* 'generation' is incremented everytime the metadata is written */
2590 generation
= __le32_to_cpu(mpb
->generation_num
);
2592 mpb
->generation_num
= __cpu_to_le32(generation
);
2594 mpb_size
+= sizeof(struct imsm_disk
) * mpb
->num_disks
;
2595 for (d
= super
->disks
; d
; d
= d
->next
) {
2599 mpb
->disk
[d
->index
] = d
->disk
;
2601 for (d
= super
->missing
; d
; d
= d
->next
)
2602 mpb
->disk
[d
->index
] = d
->disk
;
2604 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
2605 struct imsm_dev
*dev
= __get_imsm_dev(mpb
, i
);
2607 imsm_copy_dev(dev
, get_imsm_dev(super
, i
));
2608 mpb_size
+= sizeof_imsm_dev(dev
, 0);
2610 mpb_size
+= __le32_to_cpu(mpb
->bbm_log_size
);
2611 mpb
->mpb_size
= __cpu_to_le32(mpb_size
);
2613 /* recalculate checksum */
2614 sum
= __gen_imsm_checksum(mpb
);
2615 mpb
->check_sum
= __cpu_to_le32(sum
);
2617 /* write the mpb for disks that compose raid devices */
2618 for (d
= super
->disks
; d
; d
= d
->next
) {
2621 if (store_imsm_mpb(d
->fd
, super
))
2622 fprintf(stderr
, "%s: failed for device %d:%d %s\n",
2623 __func__
, d
->major
, d
->minor
, strerror(errno
));
2631 return write_super_imsm_spares(super
, doclose
);
2637 static int create_array(struct supertype
*st
)
2640 struct imsm_update_create_array
*u
;
2641 struct intel_super
*super
= st
->sb
;
2642 struct imsm_dev
*dev
= get_imsm_dev(super
, super
->current_vol
);
2643 struct imsm_map
*map
= get_imsm_map(dev
, 0);
2644 struct disk_info
*inf
;
2645 struct imsm_disk
*disk
;
2649 len
= sizeof(*u
) - sizeof(*dev
) + sizeof_imsm_dev(dev
, 0) +
2650 sizeof(*inf
) * map
->num_members
;
2653 fprintf(stderr
, "%s: failed to allocate update buffer\n",
2658 u
->type
= update_create_array
;
2659 u
->dev_idx
= super
->current_vol
;
2660 imsm_copy_dev(&u
->dev
, dev
);
2661 inf
= get_disk_info(u
);
2662 for (i
= 0; i
< map
->num_members
; i
++) {
2663 idx
= get_imsm_disk_idx(dev
, i
);
2664 disk
= get_imsm_disk(super
, idx
);
2665 serialcpy(inf
[i
].serial
, disk
->serial
);
2667 append_metadata_update(st
, u
, len
);
2672 static int _add_disk(struct supertype
*st
)
2674 struct intel_super
*super
= st
->sb
;
2676 struct imsm_update_add_disk
*u
;
2684 fprintf(stderr
, "%s: failed to allocate update buffer\n",
2689 u
->type
= update_add_disk
;
2690 append_metadata_update(st
, u
, len
);
2695 static int write_init_super_imsm(struct supertype
*st
)
2697 if (st
->update_tail
) {
2698 /* queue the recently created array / added disk
2699 * as a metadata update */
2700 struct intel_super
*super
= st
->sb
;
2704 /* determine if we are creating a volume or adding a disk */
2705 if (super
->current_vol
< 0) {
2706 /* in the add disk case we are running in mdmon
2707 * context, so don't close fd's
2709 return _add_disk(st
);
2711 rv
= create_array(st
);
2713 for (d
= super
->disks
; d
; d
= d
->next
) {
2720 return write_super_imsm(st
->sb
, 1);
2724 static int store_zero_imsm(struct supertype
*st
, int fd
)
2726 unsigned long long dsize
;
2729 get_dev_size(fd
, NULL
, &dsize
);
2731 /* first block is stored on second to last sector of the disk */
2732 if (lseek64(fd
, dsize
- (512 * 2), SEEK_SET
) < 0)
2735 if (posix_memalign(&buf
, 512, 512) != 0)
2738 memset(buf
, 0, 512);
2739 if (write(fd
, buf
, 512) != 512)
2744 static int imsm_bbm_log_size(struct imsm_super
*mpb
)
2746 return __le32_to_cpu(mpb
->bbm_log_size
);
2750 static int validate_geometry_imsm_container(struct supertype
*st
, int level
,
2751 int layout
, int raiddisks
, int chunk
,
2752 unsigned long long size
, char *dev
,
2753 unsigned long long *freesize
,
2757 unsigned long long ldsize
;
2758 const struct imsm_orom
*orom
;
2760 if (level
!= LEVEL_CONTAINER
)
2765 if (check_env("IMSM_NO_PLATFORM"))
2768 orom
= find_imsm_orom();
2769 if (orom
&& raiddisks
> orom
->tds
) {
2771 fprintf(stderr
, Name
": %d exceeds maximum number of"
2772 " platform supported disks: %d\n",
2773 raiddisks
, orom
->tds
);
2777 fd
= open(dev
, O_RDONLY
|O_EXCL
, 0);
2780 fprintf(stderr
, Name
": imsm: Cannot open %s: %s\n",
2781 dev
, strerror(errno
));
2784 if (!get_dev_size(fd
, dev
, &ldsize
)) {
2790 *freesize
= avail_size_imsm(st
, ldsize
>> 9);
2795 static unsigned long long find_size(struct extent
*e
, int *idx
, int num_extents
)
2797 const unsigned long long base_start
= e
[*idx
].start
;
2798 unsigned long long end
= base_start
+ e
[*idx
].size
;
2801 if (base_start
== end
)
2805 for (i
= *idx
; i
< num_extents
; i
++) {
2806 /* extend overlapping extents */
2807 if (e
[i
].start
>= base_start
&&
2808 e
[i
].start
<= end
) {
2811 if (e
[i
].start
+ e
[i
].size
> end
)
2812 end
= e
[i
].start
+ e
[i
].size
;
2813 } else if (e
[i
].start
> end
) {
2819 return end
- base_start
;
2822 static unsigned long long merge_extents(struct intel_super
*super
, int sum_extents
)
2824 /* build a composite disk with all known extents and generate a new
2825 * 'maxsize' given the "all disks in an array must share a common start
2826 * offset" constraint
2828 struct extent
*e
= calloc(sum_extents
, sizeof(*e
));
2832 unsigned long long pos
;
2833 unsigned long long start
= 0;
2834 unsigned long long maxsize
;
2835 unsigned long reserve
;
2838 return ~0ULL; /* error */
2840 /* coalesce and sort all extents. also, check to see if we need to
2841 * reserve space between member arrays
2844 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
2847 for (i
= 0; i
< dl
->extent_cnt
; i
++)
2850 qsort(e
, sum_extents
, sizeof(*e
), cmp_extent
);
2855 while (i
< sum_extents
) {
2856 e
[j
].start
= e
[i
].start
;
2857 e
[j
].size
= find_size(e
, &i
, sum_extents
);
2859 if (e
[j
-1].size
== 0)
2868 unsigned long long esize
;
2870 esize
= e
[i
].start
- pos
;
2871 if (esize
>= maxsize
) {
2876 pos
= e
[i
].start
+ e
[i
].size
;
2878 } while (e
[i
-1].size
);
2881 if (start_extent
> 0)
2882 reserve
= IMSM_RESERVED_SECTORS
; /* gap between raid regions */
2886 if (maxsize
< reserve
)
2889 super
->create_offset
= ~((__u32
) 0);
2890 if (start
+ reserve
> super
->create_offset
)
2891 return ~0ULL; /* start overflows create_offset */
2892 super
->create_offset
= start
+ reserve
;
2894 return maxsize
- reserve
;
2897 static int is_raid_level_supported(const struct imsm_orom
*orom
, int level
, int raiddisks
)
2899 if (level
< 0 || level
== 6 || level
== 4)
2902 /* if we have an orom prevent invalid raid levels */
2905 case 0: return imsm_orom_has_raid0(orom
);
2908 return imsm_orom_has_raid1e(orom
);
2909 return imsm_orom_has_raid1(orom
) && raiddisks
== 2;
2910 case 10: return imsm_orom_has_raid10(orom
) && raiddisks
== 4;
2911 case 5: return imsm_orom_has_raid5(orom
) && raiddisks
> 2;
2914 return 1; /* not on an Intel RAID platform so anything goes */
2919 #define pr_vrb(fmt, arg...) (void) (verbose && fprintf(stderr, Name fmt, ##arg))
2920 /* validate_geometry_imsm_volume - lifted from validate_geometry_ddf_bvd
2921 * FIX ME add ahci details
2923 static int validate_geometry_imsm_volume(struct supertype
*st
, int level
,
2924 int layout
, int raiddisks
, int chunk
,
2925 unsigned long long size
, char *dev
,
2926 unsigned long long *freesize
,
2930 struct intel_super
*super
= st
->sb
;
2931 struct imsm_super
*mpb
= super
->anchor
;
2933 unsigned long long pos
= 0;
2934 unsigned long long maxsize
;
2938 /* We must have the container info already read in. */
2942 if (!is_raid_level_supported(super
->orom
, level
, raiddisks
)) {
2943 pr_vrb(": platform does not support raid%d with %d disk%s\n",
2944 level
, raiddisks
, raiddisks
> 1 ? "s" : "");
2947 if (super
->orom
&& level
!= 1 &&
2948 !imsm_orom_has_chunk(super
->orom
, chunk
)) {
2949 pr_vrb(": platform does not support a chunk size of: %d\n", chunk
);
2952 if (layout
!= imsm_level_to_layout(level
)) {
2954 pr_vrb(": imsm raid 5 only supports the left-asymmetric layout\n");
2955 else if (level
== 10)
2956 pr_vrb(": imsm raid 10 only supports the n2 layout\n");
2958 pr_vrb(": imsm unknown layout %#x for this raid level %d\n",
2964 /* General test: make sure there is space for
2965 * 'raiddisks' device extents of size 'size' at a given
2968 unsigned long long minsize
= size
;
2969 unsigned long long start_offset
= ~0ULL;
2972 minsize
= MPB_SECTOR_CNT
+ IMSM_RESERVED_SECTORS
;
2973 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
2978 e
= get_extents(super
, dl
);
2981 unsigned long long esize
;
2982 esize
= e
[i
].start
- pos
;
2983 if (esize
>= minsize
)
2985 if (found
&& start_offset
== ~0ULL) {
2988 } else if (found
&& pos
!= start_offset
) {
2992 pos
= e
[i
].start
+ e
[i
].size
;
2994 } while (e
[i
-1].size
);
2999 if (dcnt
< raiddisks
) {
3001 fprintf(stderr
, Name
": imsm: Not enough "
3002 "devices with space for this array "
3010 /* This device must be a member of the set */
3011 if (stat(dev
, &stb
) < 0)
3013 if ((S_IFMT
& stb
.st_mode
) != S_IFBLK
)
3015 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
3016 if (dl
->major
== major(stb
.st_rdev
) &&
3017 dl
->minor
== minor(stb
.st_rdev
))
3022 fprintf(stderr
, Name
": %s is not in the "
3023 "same imsm set\n", dev
);
3025 } else if (super
->orom
&& dl
->index
< 0 && mpb
->num_raid_devs
) {
3026 /* If a volume is present then the current creation attempt
3027 * cannot incorporate new spares because the orom may not
3028 * understand this configuration (all member disks must be
3029 * members of each array in the container).
3031 fprintf(stderr
, Name
": %s is a spare and a volume"
3032 " is already defined for this container\n", dev
);
3033 fprintf(stderr
, Name
": The option-rom requires all member"
3034 " disks to be a member of all volumes\n");
3038 /* retrieve the largest free space block */
3039 e
= get_extents(super
, dl
);
3044 unsigned long long esize
;
3046 esize
= e
[i
].start
- pos
;
3047 if (esize
>= maxsize
)
3049 pos
= e
[i
].start
+ e
[i
].size
;
3051 } while (e
[i
-1].size
);
3056 fprintf(stderr
, Name
": unable to determine free space for: %s\n",
3060 if (maxsize
< size
) {
3062 fprintf(stderr
, Name
": %s not enough space (%llu < %llu)\n",
3063 dev
, maxsize
, size
);
3067 /* count total number of extents for merge */
3069 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
3071 i
+= dl
->extent_cnt
;
3073 maxsize
= merge_extents(super
, i
);
3074 if (maxsize
< size
) {
3076 fprintf(stderr
, Name
": not enough space after merge (%llu < %llu)\n",
3079 } else if (maxsize
== ~0ULL) {
3081 fprintf(stderr
, Name
": failed to merge %d extents\n", i
);
3085 *freesize
= maxsize
;
3090 static int reserve_space(struct supertype
*st
, int raiddisks
,
3091 unsigned long long size
, int chunk
,
3092 unsigned long long *freesize
)
3094 struct intel_super
*super
= st
->sb
;
3095 struct imsm_super
*mpb
= super
->anchor
;
3100 unsigned long long maxsize
;
3101 unsigned long long minsize
;
3105 /* find the largest common start free region of the possible disks */
3109 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
3115 /* don't activate new spares if we are orom constrained
3116 * and there is already a volume active in the container
3118 if (super
->orom
&& dl
->index
< 0 && mpb
->num_raid_devs
)
3121 e
= get_extents(super
, dl
);
3124 for (i
= 1; e
[i
-1].size
; i
++)
3132 maxsize
= merge_extents(super
, extent_cnt
);
3137 if (cnt
< raiddisks
||
3138 (super
->orom
&& used
&& used
!= raiddisks
) ||
3139 maxsize
< minsize
) {
3140 fprintf(stderr
, Name
": not enough devices with space to create array.\n");
3141 return 0; /* No enough free spaces large enough */
3153 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
3155 dl
->raiddisk
= cnt
++;
3162 static int validate_geometry_imsm(struct supertype
*st
, int level
, int layout
,
3163 int raiddisks
, int chunk
, unsigned long long size
,
3164 char *dev
, unsigned long long *freesize
,
3170 /* if given unused devices create a container
3171 * if given given devices in a container create a member volume
3173 if (level
== LEVEL_CONTAINER
) {
3174 /* Must be a fresh device to add to a container */
3175 return validate_geometry_imsm_container(st
, level
, layout
,
3176 raiddisks
, chunk
, size
,
3182 if (st
->sb
&& freesize
) {
3183 /* we are being asked to automatically layout a
3184 * new volume based on the current contents of
3185 * the container. If the the parameters can be
3186 * satisfied reserve_space will record the disks,
3187 * start offset, and size of the volume to be
3188 * created. add_to_super and getinfo_super
3189 * detect when autolayout is in progress.
3191 return reserve_space(st
, raiddisks
, size
, chunk
, freesize
);
3196 /* creating in a given container */
3197 return validate_geometry_imsm_volume(st
, level
, layout
,
3198 raiddisks
, chunk
, size
,
3199 dev
, freesize
, verbose
);
3202 /* limit creation to the following levels */
3214 /* This device needs to be a device in an 'imsm' container */
3215 fd
= open(dev
, O_RDONLY
|O_EXCL
, 0);
3219 Name
": Cannot create this array on device %s\n",
3224 if (errno
!= EBUSY
|| (fd
= open(dev
, O_RDONLY
, 0)) < 0) {
3226 fprintf(stderr
, Name
": Cannot open %s: %s\n",
3227 dev
, strerror(errno
));
3230 /* Well, it is in use by someone, maybe an 'imsm' container. */
3231 cfd
= open_container(fd
);
3235 fprintf(stderr
, Name
": Cannot use %s: It is busy\n",
3239 sra
= sysfs_read(cfd
, 0, GET_VERSION
);
3241 if (sra
&& sra
->array
.major_version
== -1 &&
3242 strcmp(sra
->text_version
, "imsm") == 0) {
3243 /* This is a member of a imsm container. Load the container
3244 * and try to create a volume
3246 struct intel_super
*super
;
3248 if (load_super_imsm_all(st
, cfd
, (void **) &super
, NULL
, 1) == 0) {
3250 st
->container_dev
= fd2devnum(cfd
);
3252 return validate_geometry_imsm_volume(st
, level
, layout
,
3258 } else /* may belong to another container */
3263 #endif /* MDASSEMBLE */
3265 static struct mdinfo
*container_content_imsm(struct supertype
*st
)
3267 /* Given a container loaded by load_super_imsm_all,
3268 * extract information about all the arrays into
3271 * For each imsm_dev create an mdinfo, fill it in,
3272 * then look for matching devices in super->disks
3273 * and create appropriate device mdinfo.
3275 struct intel_super
*super
= st
->sb
;
3276 struct imsm_super
*mpb
= super
->anchor
;
3277 struct mdinfo
*rest
= NULL
;
3280 /* do not assemble arrays that might have bad blocks */
3281 if (imsm_bbm_log_size(super
->anchor
)) {
3282 fprintf(stderr
, Name
": BBM log found in metadata. "
3283 "Cannot activate array(s).\n");
3287 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
3288 struct imsm_dev
*dev
= get_imsm_dev(super
, i
);
3289 struct imsm_map
*map
= get_imsm_map(dev
, 0);
3290 struct mdinfo
*this;
3293 this = malloc(sizeof(*this));
3294 memset(this, 0, sizeof(*this));
3297 super
->current_vol
= i
;
3298 getinfo_super_imsm_volume(st
, this);
3299 for (slot
= 0 ; slot
< map
->num_members
; slot
++) {
3300 struct mdinfo
*info_d
;
3308 idx
= get_imsm_disk_idx(dev
, slot
);
3309 ord
= get_imsm_ord_tbl_ent(dev
, slot
);
3310 for (d
= super
->disks
; d
; d
= d
->next
)
3311 if (d
->index
== idx
)
3317 s
= d
? d
->disk
.status
: 0;
3318 if (s
& FAILED_DISK
)
3320 if (!(s
& USABLE_DISK
))
3322 if (ord
& IMSM_ORD_REBUILD
)
3326 * if we skip some disks the array will be assmebled degraded;
3327 * reset resync start to avoid a dirty-degraded situation
3329 * FIXME handle dirty degraded
3331 if (skip
&& !dev
->vol
.dirty
)
3332 this->resync_start
= ~0ULL;
3336 info_d
= malloc(sizeof(*info_d
));
3338 fprintf(stderr
, Name
": failed to allocate disk"
3339 " for volume %s\n", (char *) dev
->volume
);
3344 memset(info_d
, 0, sizeof(*info_d
));
3345 info_d
->next
= this->devs
;
3346 this->devs
= info_d
;
3348 info_d
->disk
.number
= d
->index
;
3349 info_d
->disk
.major
= d
->major
;
3350 info_d
->disk
.minor
= d
->minor
;
3351 info_d
->disk
.raid_disk
= slot
;
3353 this->array
.working_disks
++;
3355 info_d
->events
= __le32_to_cpu(mpb
->generation_num
);
3356 info_d
->data_offset
= __le32_to_cpu(map
->pba_of_lba0
);
3357 info_d
->component_size
= __le32_to_cpu(map
->blocks_per_member
);
3359 strcpy(info_d
->name
, d
->devname
);
3369 static int imsm_open_new(struct supertype
*c
, struct active_array
*a
,
3372 struct intel_super
*super
= c
->sb
;
3373 struct imsm_super
*mpb
= super
->anchor
;
3375 if (atoi(inst
) >= mpb
->num_raid_devs
) {
3376 fprintf(stderr
, "%s: subarry index %d, out of range\n",
3377 __func__
, atoi(inst
));
3381 dprintf("imsm: open_new %s\n", inst
);
3382 a
->info
.container_member
= atoi(inst
);
3386 static __u8
imsm_check_degraded(struct intel_super
*super
, struct imsm_dev
*dev
, int failed
)
3388 struct imsm_map
*map
= get_imsm_map(dev
, 0);
3391 return map
->map_state
== IMSM_T_STATE_UNINITIALIZED
?
3392 IMSM_T_STATE_UNINITIALIZED
: IMSM_T_STATE_NORMAL
;
3394 switch (get_imsm_raid_level(map
)) {
3396 return IMSM_T_STATE_FAILED
;
3399 if (failed
< map
->num_members
)
3400 return IMSM_T_STATE_DEGRADED
;
3402 return IMSM_T_STATE_FAILED
;
3407 * check to see if any mirrors have failed, otherwise we
3408 * are degraded. Even numbered slots are mirrored on
3412 /* gcc -Os complains that this is unused */
3413 int insync
= insync
;
3415 for (i
= 0; i
< map
->num_members
; i
++) {
3416 __u32 ord
= get_imsm_ord_tbl_ent(dev
, i
);
3417 int idx
= ord_to_idx(ord
);
3418 struct imsm_disk
*disk
;
3420 /* reset the potential in-sync count on even-numbered
3421 * slots. num_copies is always 2 for imsm raid10
3426 disk
= get_imsm_disk(super
, idx
);
3427 if (!disk
|| disk
->status
& FAILED_DISK
||
3428 ord
& IMSM_ORD_REBUILD
)
3431 /* no in-sync disks left in this mirror the
3435 return IMSM_T_STATE_FAILED
;
3438 return IMSM_T_STATE_DEGRADED
;
3442 return IMSM_T_STATE_DEGRADED
;
3444 return IMSM_T_STATE_FAILED
;
3450 return map
->map_state
;
3453 static int imsm_count_failed(struct intel_super
*super
, struct imsm_dev
*dev
)
3457 struct imsm_disk
*disk
;
3458 struct imsm_map
*map
= get_imsm_map(dev
, 0);
3459 struct imsm_map
*prev
= get_imsm_map(dev
, dev
->vol
.migr_state
);
3463 /* at the beginning of migration we set IMSM_ORD_REBUILD on
3464 * disks that are being rebuilt. New failures are recorded to
3465 * map[0]. So we look through all the disks we started with and
3466 * see if any failures are still present, or if any new ones
3469 * FIXME add support for online capacity expansion and
3470 * raid-level-migration
3472 for (i
= 0; i
< prev
->num_members
; i
++) {
3473 ord
= __le32_to_cpu(prev
->disk_ord_tbl
[i
]);
3474 ord
|= __le32_to_cpu(map
->disk_ord_tbl
[i
]);
3475 idx
= ord_to_idx(ord
);
3477 disk
= get_imsm_disk(super
, idx
);
3478 if (!disk
|| disk
->status
& FAILED_DISK
||
3479 ord
& IMSM_ORD_REBUILD
)
3486 static int is_resyncing(struct imsm_dev
*dev
)
3488 struct imsm_map
*migr_map
;
3490 if (!dev
->vol
.migr_state
)
3493 if (dev
->vol
.migr_type
== MIGR_INIT
)
3496 migr_map
= get_imsm_map(dev
, 1);
3498 if (migr_map
->map_state
== IMSM_T_STATE_NORMAL
)
3504 static int is_rebuilding(struct imsm_dev
*dev
)
3506 struct imsm_map
*migr_map
;
3508 if (!dev
->vol
.migr_state
)
3511 if (dev
->vol
.migr_type
!= MIGR_REBUILD
)
3514 migr_map
= get_imsm_map(dev
, 1);
3516 if (migr_map
->map_state
== IMSM_T_STATE_DEGRADED
)
3522 /* return true if we recorded new information */
3523 static int mark_failure(struct imsm_dev
*dev
, struct imsm_disk
*disk
, int idx
)
3527 struct imsm_map
*map
;
3529 /* new failures are always set in map[0] */
3530 map
= get_imsm_map(dev
, 0);
3532 slot
= get_imsm_disk_slot(map
, idx
);
3536 ord
= __le32_to_cpu(map
->disk_ord_tbl
[slot
]);
3537 if ((disk
->status
& FAILED_DISK
) && (ord
& IMSM_ORD_REBUILD
))
3540 disk
->status
|= FAILED_DISK
;
3541 set_imsm_ord_tbl_ent(map
, slot
, idx
| IMSM_ORD_REBUILD
);
3542 if (map
->failed_disk_num
== ~0)
3543 map
->failed_disk_num
= slot
;
3547 static void mark_missing(struct imsm_dev
*dev
, struct imsm_disk
*disk
, int idx
)
3549 mark_failure(dev
, disk
, idx
);
3551 if (disk
->scsi_id
== __cpu_to_le32(~(__u32
)0))
3554 disk
->scsi_id
= __cpu_to_le32(~(__u32
)0);
3555 memmove(&disk
->serial
[0], &disk
->serial
[1], MAX_RAID_SERIAL_LEN
- 1);
3558 /* Handle dirty -> clean transititions and resync. Degraded and rebuild
3559 * states are handled in imsm_set_disk() with one exception, when a
3560 * resync is stopped due to a new failure this routine will set the
3561 * 'degraded' state for the array.
3563 static int imsm_set_array_state(struct active_array
*a
, int consistent
)
3565 int inst
= a
->info
.container_member
;
3566 struct intel_super
*super
= a
->container
->sb
;
3567 struct imsm_dev
*dev
= get_imsm_dev(super
, inst
);
3568 struct imsm_map
*map
= get_imsm_map(dev
, 0);
3569 int failed
= imsm_count_failed(super
, dev
);
3570 __u8 map_state
= imsm_check_degraded(super
, dev
, failed
);
3572 /* before we activate this array handle any missing disks */
3573 if (consistent
== 2 && super
->missing
) {
3576 dprintf("imsm: mark missing\n");
3577 end_migration(dev
, map_state
);
3578 for (dl
= super
->missing
; dl
; dl
= dl
->next
)
3579 mark_missing(dev
, &dl
->disk
, dl
->index
);
3580 super
->updates_pending
++;
3583 if (consistent
== 2 &&
3584 (!is_resync_complete(a
) ||
3585 map_state
!= IMSM_T_STATE_NORMAL
||
3586 dev
->vol
.migr_state
))
3589 if (is_resync_complete(a
)) {
3590 /* complete intialization / resync,
3591 * recovery and interrupted recovery is completed in
3594 if (is_resyncing(dev
)) {
3595 dprintf("imsm: mark resync done\n");
3596 end_migration(dev
, map_state
);
3597 super
->updates_pending
++;
3599 } else if (!is_resyncing(dev
) && !failed
) {
3600 /* mark the start of the init process if nothing is failed */
3601 dprintf("imsm: mark resync start (%llu)\n", a
->resync_start
);
3602 if (map
->map_state
== IMSM_T_STATE_NORMAL
)
3603 migrate(dev
, IMSM_T_STATE_NORMAL
, MIGR_REBUILD
);
3605 migrate(dev
, IMSM_T_STATE_NORMAL
, MIGR_INIT
);
3606 super
->updates_pending
++;
3609 /* check if we can update the migration checkpoint */
3610 if (dev
->vol
.migr_state
&&
3611 __le32_to_cpu(dev
->vol
.curr_migr_unit
) != a
->resync_start
) {
3612 dprintf("imsm: checkpoint migration (%llu)\n", a
->resync_start
);
3613 dev
->vol
.curr_migr_unit
= __cpu_to_le32(a
->resync_start
);
3614 super
->updates_pending
++;
3617 /* mark dirty / clean */
3618 if (dev
->vol
.dirty
!= !consistent
) {
3619 dprintf("imsm: mark '%s' (%llu)\n",
3620 consistent
? "clean" : "dirty", a
->resync_start
);
3625 super
->updates_pending
++;
3630 static void imsm_set_disk(struct active_array
*a
, int n
, int state
)
3632 int inst
= a
->info
.container_member
;
3633 struct intel_super
*super
= a
->container
->sb
;
3634 struct imsm_dev
*dev
= get_imsm_dev(super
, inst
);
3635 struct imsm_map
*map
= get_imsm_map(dev
, 0);
3636 struct imsm_disk
*disk
;
3641 if (n
> map
->num_members
)
3642 fprintf(stderr
, "imsm: set_disk %d out of range 0..%d\n",
3643 n
, map
->num_members
- 1);
3648 dprintf("imsm: set_disk %d:%x\n", n
, state
);
3650 ord
= get_imsm_ord_tbl_ent(dev
, n
);
3651 disk
= get_imsm_disk(super
, ord_to_idx(ord
));
3653 /* check for new failures */
3654 if (state
& DS_FAULTY
) {
3655 if (mark_failure(dev
, disk
, ord_to_idx(ord
)))
3656 super
->updates_pending
++;
3659 /* check if in_sync */
3660 if (state
& DS_INSYNC
&& ord
& IMSM_ORD_REBUILD
&& is_rebuilding(dev
)) {
3661 struct imsm_map
*migr_map
= get_imsm_map(dev
, 1);
3663 set_imsm_ord_tbl_ent(migr_map
, n
, ord_to_idx(ord
));
3664 super
->updates_pending
++;
3667 failed
= imsm_count_failed(super
, dev
);
3668 map_state
= imsm_check_degraded(super
, dev
, failed
);
3670 /* check if recovery complete, newly degraded, or failed */
3671 if (map_state
== IMSM_T_STATE_NORMAL
&& is_rebuilding(dev
)) {
3672 end_migration(dev
, map_state
);
3673 map
= get_imsm_map(dev
, 0);
3674 map
->failed_disk_num
= ~0;
3675 super
->updates_pending
++;
3676 } else if (map_state
== IMSM_T_STATE_DEGRADED
&&
3677 map
->map_state
!= map_state
&&
3678 !dev
->vol
.migr_state
) {
3679 dprintf("imsm: mark degraded\n");
3680 map
->map_state
= map_state
;
3681 super
->updates_pending
++;
3682 } else if (map_state
== IMSM_T_STATE_FAILED
&&
3683 map
->map_state
!= map_state
) {
3684 dprintf("imsm: mark failed\n");
3685 end_migration(dev
, map_state
);
3686 super
->updates_pending
++;
3690 static int store_imsm_mpb(int fd
, struct intel_super
*super
)
3692 struct imsm_super
*mpb
= super
->anchor
;
3693 __u32 mpb_size
= __le32_to_cpu(mpb
->mpb_size
);
3694 unsigned long long dsize
;
3695 unsigned long long sectors
;
3697 get_dev_size(fd
, NULL
, &dsize
);
3699 if (mpb_size
> 512) {
3700 /* -1 to account for anchor */
3701 sectors
= mpb_sectors(mpb
) - 1;
3703 /* write the extended mpb to the sectors preceeding the anchor */
3704 if (lseek64(fd
, dsize
- (512 * (2 + sectors
)), SEEK_SET
) < 0)
3707 if (write(fd
, super
->buf
+ 512, 512 * sectors
) != 512 * sectors
)
3711 /* first block is stored on second to last sector of the disk */
3712 if (lseek64(fd
, dsize
- (512 * 2), SEEK_SET
) < 0)
3715 if (write(fd
, super
->buf
, 512) != 512)
3721 static void imsm_sync_metadata(struct supertype
*container
)
3723 struct intel_super
*super
= container
->sb
;
3725 if (!super
->updates_pending
)
3728 write_super_imsm(super
, 0);
3730 super
->updates_pending
= 0;
3733 static struct dl
*imsm_readd(struct intel_super
*super
, int idx
, struct active_array
*a
)
3735 struct imsm_dev
*dev
= get_imsm_dev(super
, a
->info
.container_member
);
3736 int i
= get_imsm_disk_idx(dev
, idx
);
3739 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
3743 if (dl
&& dl
->disk
.status
& FAILED_DISK
)
3747 dprintf("%s: found %x:%x\n", __func__
, dl
->major
, dl
->minor
);
3752 static struct dl
*imsm_add_spare(struct intel_super
*super
, int slot
,
3753 struct active_array
*a
, int activate_new
)
3755 struct imsm_dev
*dev
= get_imsm_dev(super
, a
->info
.container_member
);
3756 int idx
= get_imsm_disk_idx(dev
, slot
);
3757 struct imsm_super
*mpb
= super
->anchor
;
3758 struct imsm_map
*map
;
3759 unsigned long long esize
;
3760 unsigned long long pos
;
3769 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
3770 /* If in this array, skip */
3771 for (d
= a
->info
.devs
; d
; d
= d
->next
)
3772 if (d
->state_fd
>= 0 &&
3773 d
->disk
.major
== dl
->major
&&
3774 d
->disk
.minor
== dl
->minor
) {
3775 dprintf("%x:%x already in array\n", dl
->major
, dl
->minor
);
3781 /* skip in use or failed drives */
3782 if (dl
->disk
.status
& FAILED_DISK
|| idx
== dl
->index
||
3784 dprintf("%x:%x status (failed: %d index: %d)\n",
3785 dl
->major
, dl
->minor
,
3786 (dl
->disk
.status
& FAILED_DISK
) == FAILED_DISK
, idx
);
3790 /* skip pure spares when we are looking for partially
3791 * assimilated drives
3793 if (dl
->index
== -1 && !activate_new
)
3796 /* Does this unused device have the requisite free space?
3797 * It needs to be able to cover all member volumes
3799 ex
= get_extents(super
, dl
);
3801 dprintf("cannot get extents\n");
3804 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
3805 dev
= get_imsm_dev(super
, i
);
3806 map
= get_imsm_map(dev
, 0);
3808 /* check if this disk is already a member of
3811 if (get_imsm_disk_slot(map
, dl
->index
) >= 0)
3817 array_start
= __le32_to_cpu(map
->pba_of_lba0
);
3818 blocks
= __le32_to_cpu(map
->blocks_per_member
);
3821 /* check that we can start at pba_of_lba0 with
3822 * blocks_per_member of space
3824 esize
= ex
[j
].start
- pos
;
3825 if (array_start
>= pos
&&
3826 array_start
+ blocks
< ex
[j
].start
) {
3830 pos
= ex
[j
].start
+ ex
[j
].size
;
3832 } while (ex
[j
-1].size
);
3839 if (i
< mpb
->num_raid_devs
) {
3840 dprintf("%x:%x does not have %u at %u\n",
3841 dl
->major
, dl
->minor
,
3842 blocks
, array_start
);
3852 static struct mdinfo
*imsm_activate_spare(struct active_array
*a
,
3853 struct metadata_update
**updates
)
3856 * Find a device with unused free space and use it to replace a
3857 * failed/vacant region in an array. We replace failed regions one a
3858 * array at a time. The result is that a new spare disk will be added
3859 * to the first failed array and after the monitor has finished
3860 * propagating failures the remainder will be consumed.
3862 * FIXME add a capability for mdmon to request spares from another
3866 struct intel_super
*super
= a
->container
->sb
;
3867 int inst
= a
->info
.container_member
;
3868 struct imsm_dev
*dev
= get_imsm_dev(super
, inst
);
3869 struct imsm_map
*map
= get_imsm_map(dev
, 0);
3870 int failed
= a
->info
.array
.raid_disks
;
3871 struct mdinfo
*rv
= NULL
;
3874 struct metadata_update
*mu
;
3876 struct imsm_update_activate_spare
*u
;
3880 for (d
= a
->info
.devs
; d
; d
= d
->next
) {
3881 if ((d
->curr_state
& DS_FAULTY
) &&
3883 /* wait for Removal to happen */
3885 if (d
->state_fd
>= 0)
3889 dprintf("imsm: activate spare: inst=%d failed=%d (%d) level=%d\n",
3890 inst
, failed
, a
->info
.array
.raid_disks
, a
->info
.array
.level
);
3891 if (imsm_check_degraded(super
, dev
, failed
) != IMSM_T_STATE_DEGRADED
)
3894 /* For each slot, if it is not working, find a spare */
3895 for (i
= 0; i
< a
->info
.array
.raid_disks
; i
++) {
3896 for (d
= a
->info
.devs
; d
; d
= d
->next
)
3897 if (d
->disk
.raid_disk
== i
)
3899 dprintf("found %d: %p %x\n", i
, d
, d
?d
->curr_state
:0);
3900 if (d
&& (d
->state_fd
>= 0))
3904 * OK, this device needs recovery. Try to re-add the
3905 * previous occupant of this slot, if this fails see if
3906 * we can continue the assimilation of a spare that was
3907 * partially assimilated, finally try to activate a new
3910 dl
= imsm_readd(super
, i
, a
);
3912 dl
= imsm_add_spare(super
, i
, a
, 0);
3914 dl
= imsm_add_spare(super
, i
, a
, 1);
3918 /* found a usable disk with enough space */
3919 di
= malloc(sizeof(*di
));
3922 memset(di
, 0, sizeof(*di
));
3924 /* dl->index will be -1 in the case we are activating a
3925 * pristine spare. imsm_process_update() will create a
3926 * new index in this case. Once a disk is found to be
3927 * failed in all member arrays it is kicked from the
3930 di
->disk
.number
= dl
->index
;
3932 /* (ab)use di->devs to store a pointer to the device
3935 di
->devs
= (struct mdinfo
*) dl
;
3937 di
->disk
.raid_disk
= i
;
3938 di
->disk
.major
= dl
->major
;
3939 di
->disk
.minor
= dl
->minor
;
3941 di
->data_offset
= __le32_to_cpu(map
->pba_of_lba0
);
3942 di
->component_size
= a
->info
.component_size
;
3943 di
->container_member
= inst
;
3947 dprintf("%x:%x to be %d at %llu\n", dl
->major
, dl
->minor
,
3948 i
, di
->data_offset
);
3954 /* No spares found */
3956 /* Now 'rv' has a list of devices to return.
3957 * Create a metadata_update record to update the
3958 * disk_ord_tbl for the array
3960 mu
= malloc(sizeof(*mu
));
3962 mu
->buf
= malloc(sizeof(struct imsm_update_activate_spare
) * num_spares
);
3963 if (mu
->buf
== NULL
) {
3970 struct mdinfo
*n
= rv
->next
;
3979 mu
->len
= sizeof(struct imsm_update_activate_spare
) * num_spares
;
3980 mu
->next
= *updates
;
3981 u
= (struct imsm_update_activate_spare
*) mu
->buf
;
3983 for (di
= rv
; di
; di
= di
->next
) {
3984 u
->type
= update_activate_spare
;
3985 u
->dl
= (struct dl
*) di
->devs
;
3987 u
->slot
= di
->disk
.raid_disk
;
3998 static int disks_overlap(struct intel_super
*super
, int idx
, struct imsm_update_create_array
*u
)
4000 struct imsm_dev
*dev
= get_imsm_dev(super
, idx
);
4001 struct imsm_map
*map
= get_imsm_map(dev
, 0);
4002 struct imsm_map
*new_map
= get_imsm_map(&u
->dev
, 0);
4003 struct disk_info
*inf
= get_disk_info(u
);
4004 struct imsm_disk
*disk
;
4008 for (i
= 0; i
< map
->num_members
; i
++) {
4009 disk
= get_imsm_disk(super
, get_imsm_disk_idx(dev
, i
));
4010 for (j
= 0; j
< new_map
->num_members
; j
++)
4011 if (serialcmp(disk
->serial
, inf
[j
].serial
) == 0)
4018 static void imsm_delete(struct intel_super
*super
, struct dl
**dlp
, int index
);
4020 static void imsm_process_update(struct supertype
*st
,
4021 struct metadata_update
*update
)
4024 * crack open the metadata_update envelope to find the update record
4025 * update can be one of:
4026 * update_activate_spare - a spare device has replaced a failed
4027 * device in an array, update the disk_ord_tbl. If this disk is
4028 * present in all member arrays then also clear the SPARE_DISK
4031 struct intel_super
*super
= st
->sb
;
4032 struct imsm_super
*mpb
;
4033 enum imsm_update_type type
= *(enum imsm_update_type
*) update
->buf
;
4035 /* update requires a larger buf but the allocation failed */
4036 if (super
->next_len
&& !super
->next_buf
) {
4037 super
->next_len
= 0;
4041 if (super
->next_buf
) {
4042 memcpy(super
->next_buf
, super
->buf
, super
->len
);
4044 super
->len
= super
->next_len
;
4045 super
->buf
= super
->next_buf
;
4047 super
->next_len
= 0;
4048 super
->next_buf
= NULL
;
4051 mpb
= super
->anchor
;
4054 case update_activate_spare
: {
4055 struct imsm_update_activate_spare
*u
= (void *) update
->buf
;
4056 struct imsm_dev
*dev
= get_imsm_dev(super
, u
->array
);
4057 struct imsm_map
*map
= get_imsm_map(dev
, 0);
4058 struct imsm_map
*migr_map
;
4059 struct active_array
*a
;
4060 struct imsm_disk
*disk
;
4065 int victim
= get_imsm_disk_idx(dev
, u
->slot
);
4068 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
4073 fprintf(stderr
, "error: imsm_activate_spare passed "
4074 "an unknown disk (index: %d)\n",
4079 super
->updates_pending
++;
4081 /* count failures (excluding rebuilds and the victim)
4082 * to determine map[0] state
4085 for (i
= 0; i
< map
->num_members
; i
++) {
4088 disk
= get_imsm_disk(super
, get_imsm_disk_idx(dev
, i
));
4089 if (!disk
|| disk
->status
& FAILED_DISK
)
4093 /* adding a pristine spare, assign a new index */
4094 if (dl
->index
< 0) {
4095 dl
->index
= super
->anchor
->num_disks
;
4096 super
->anchor
->num_disks
++;
4099 disk
->status
|= CONFIGURED_DISK
;
4100 disk
->status
&= ~SPARE_DISK
;
4103 to_state
= imsm_check_degraded(super
, dev
, failed
);
4104 map
->map_state
= IMSM_T_STATE_DEGRADED
;
4105 migrate(dev
, to_state
, MIGR_REBUILD
);
4106 migr_map
= get_imsm_map(dev
, 1);
4107 set_imsm_ord_tbl_ent(map
, u
->slot
, dl
->index
);
4108 set_imsm_ord_tbl_ent(migr_map
, u
->slot
, dl
->index
| IMSM_ORD_REBUILD
);
4110 /* count arrays using the victim in the metadata */
4112 for (a
= st
->arrays
; a
; a
= a
->next
) {
4113 dev
= get_imsm_dev(super
, a
->info
.container_member
);
4114 map
= get_imsm_map(dev
, 0);
4116 if (get_imsm_disk_slot(map
, victim
) >= 0)
4120 /* delete the victim if it is no longer being
4126 /* We know that 'manager' isn't touching anything,
4127 * so it is safe to delete
4129 for (dlp
= &super
->disks
; *dlp
; dlp
= &(*dlp
)->next
)
4130 if ((*dlp
)->index
== victim
)
4133 /* victim may be on the missing list */
4135 for (dlp
= &super
->missing
; *dlp
; dlp
= &(*dlp
)->next
)
4136 if ((*dlp
)->index
== victim
)
4138 imsm_delete(super
, dlp
, victim
);
4142 case update_create_array
: {
4143 /* someone wants to create a new array, we need to be aware of
4144 * a few races/collisions:
4145 * 1/ 'Create' called by two separate instances of mdadm
4146 * 2/ 'Create' versus 'activate_spare': mdadm has chosen
4147 * devices that have since been assimilated via
4149 * In the event this update can not be carried out mdadm will
4150 * (FIX ME) notice that its update did not take hold.
4152 struct imsm_update_create_array
*u
= (void *) update
->buf
;
4153 struct intel_dev
*dv
;
4154 struct imsm_dev
*dev
;
4155 struct imsm_map
*map
, *new_map
;
4156 unsigned long long start
, end
;
4157 unsigned long long new_start
, new_end
;
4159 struct disk_info
*inf
;
4162 /* handle racing creates: first come first serve */
4163 if (u
->dev_idx
< mpb
->num_raid_devs
) {
4164 dprintf("%s: subarray %d already defined\n",
4165 __func__
, u
->dev_idx
);
4169 /* check update is next in sequence */
4170 if (u
->dev_idx
!= mpb
->num_raid_devs
) {
4171 dprintf("%s: can not create array %d expected index %d\n",
4172 __func__
, u
->dev_idx
, mpb
->num_raid_devs
);
4176 new_map
= get_imsm_map(&u
->dev
, 0);
4177 new_start
= __le32_to_cpu(new_map
->pba_of_lba0
);
4178 new_end
= new_start
+ __le32_to_cpu(new_map
->blocks_per_member
);
4179 inf
= get_disk_info(u
);
4181 /* handle activate_spare versus create race:
4182 * check to make sure that overlapping arrays do not include
4185 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
4186 dev
= get_imsm_dev(super
, i
);
4187 map
= get_imsm_map(dev
, 0);
4188 start
= __le32_to_cpu(map
->pba_of_lba0
);
4189 end
= start
+ __le32_to_cpu(map
->blocks_per_member
);
4190 if ((new_start
>= start
&& new_start
<= end
) ||
4191 (start
>= new_start
&& start
<= new_end
))
4196 if (disks_overlap(super
, i
, u
)) {
4197 dprintf("%s: arrays overlap\n", __func__
);
4202 /* check that prepare update was successful */
4203 if (!update
->space
) {
4204 dprintf("%s: prepare update failed\n", __func__
);
4208 /* check that all disks are still active before committing
4209 * changes. FIXME: could we instead handle this by creating a
4210 * degraded array? That's probably not what the user expects,
4211 * so better to drop this update on the floor.
4213 for (i
= 0; i
< new_map
->num_members
; i
++) {
4214 dl
= serial_to_dl(inf
[i
].serial
, super
);
4216 dprintf("%s: disk disappeared\n", __func__
);
4221 super
->updates_pending
++;
4223 /* convert spares to members and fixup ord_tbl */
4224 for (i
= 0; i
< new_map
->num_members
; i
++) {
4225 dl
= serial_to_dl(inf
[i
].serial
, super
);
4226 if (dl
->index
== -1) {
4227 dl
->index
= mpb
->num_disks
;
4229 dl
->disk
.status
|= CONFIGURED_DISK
;
4230 dl
->disk
.status
&= ~SPARE_DISK
;
4232 set_imsm_ord_tbl_ent(new_map
, i
, dl
->index
);
4237 update
->space
= NULL
;
4238 imsm_copy_dev(dev
, &u
->dev
);
4239 dv
->index
= u
->dev_idx
;
4240 dv
->next
= super
->devlist
;
4241 super
->devlist
= dv
;
4242 mpb
->num_raid_devs
++;
4244 imsm_update_version_info(super
);
4247 /* mdmon knows how to release update->space, but not
4248 * ((struct intel_dev *) update->space)->dev
4250 if (update
->space
) {
4256 case update_add_disk
:
4258 /* we may be able to repair some arrays if disks are
4261 struct active_array
*a
;
4263 super
->updates_pending
++;
4264 for (a
= st
->arrays
; a
; a
= a
->next
)
4265 a
->check_degraded
= 1;
4267 /* add some spares to the metadata */
4268 while (super
->add
) {
4272 super
->add
= al
->next
;
4273 al
->next
= super
->disks
;
4275 dprintf("%s: added %x:%x\n",
4276 __func__
, al
->major
, al
->minor
);
4283 static void imsm_prepare_update(struct supertype
*st
,
4284 struct metadata_update
*update
)
4287 * Allocate space to hold new disk entries, raid-device entries or a new
4288 * mpb if necessary. The manager synchronously waits for updates to
4289 * complete in the monitor, so new mpb buffers allocated here can be
4290 * integrated by the monitor thread without worrying about live pointers
4291 * in the manager thread.
4293 enum imsm_update_type type
= *(enum imsm_update_type
*) update
->buf
;
4294 struct intel_super
*super
= st
->sb
;
4295 struct imsm_super
*mpb
= super
->anchor
;
4300 case update_create_array
: {
4301 struct imsm_update_create_array
*u
= (void *) update
->buf
;
4302 struct intel_dev
*dv
;
4303 struct imsm_dev
*dev
= &u
->dev
;
4304 struct imsm_map
*map
= get_imsm_map(dev
, 0);
4306 struct disk_info
*inf
;
4310 inf
= get_disk_info(u
);
4311 len
= sizeof_imsm_dev(dev
, 1);
4312 /* allocate a new super->devlist entry */
4313 dv
= malloc(sizeof(*dv
));
4315 dv
->dev
= malloc(len
);
4320 update
->space
= NULL
;
4324 /* count how many spares will be converted to members */
4325 for (i
= 0; i
< map
->num_members
; i
++) {
4326 dl
= serial_to_dl(inf
[i
].serial
, super
);
4328 /* hmm maybe it failed?, nothing we can do about
4333 if (count_memberships(dl
, super
) == 0)
4336 len
+= activate
* sizeof(struct imsm_disk
);
4343 /* check if we need a larger metadata buffer */
4344 if (super
->next_buf
)
4345 buf_len
= super
->next_len
;
4347 buf_len
= super
->len
;
4349 if (__le32_to_cpu(mpb
->mpb_size
) + len
> buf_len
) {
4350 /* ok we need a larger buf than what is currently allocated
4351 * if this allocation fails process_update will notice that
4352 * ->next_len is set and ->next_buf is NULL
4354 buf_len
= ROUND_UP(__le32_to_cpu(mpb
->mpb_size
) + len
, 512);
4355 if (super
->next_buf
)
4356 free(super
->next_buf
);
4358 super
->next_len
= buf_len
;
4359 if (posix_memalign(&super
->next_buf
, 512, buf_len
) != 0)
4360 super
->next_buf
= NULL
;
4364 /* must be called while manager is quiesced */
4365 static void imsm_delete(struct intel_super
*super
, struct dl
**dlp
, int index
)
4367 struct imsm_super
*mpb
= super
->anchor
;
4369 struct imsm_dev
*dev
;
4370 struct imsm_map
*map
;
4371 int i
, j
, num_members
;
4374 dprintf("%s: deleting device[%d] from imsm_super\n",
4377 /* shift all indexes down one */
4378 for (iter
= super
->disks
; iter
; iter
= iter
->next
)
4379 if (iter
->index
> index
)
4381 for (iter
= super
->missing
; iter
; iter
= iter
->next
)
4382 if (iter
->index
> index
)
4385 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
4386 dev
= get_imsm_dev(super
, i
);
4387 map
= get_imsm_map(dev
, 0);
4388 num_members
= map
->num_members
;
4389 for (j
= 0; j
< num_members
; j
++) {
4390 /* update ord entries being careful not to propagate
4391 * ord-flags to the first map
4393 ord
= get_imsm_ord_tbl_ent(dev
, j
);
4395 if (ord_to_idx(ord
) <= index
)
4398 map
= get_imsm_map(dev
, 0);
4399 set_imsm_ord_tbl_ent(map
, j
, ord_to_idx(ord
- 1));
4400 map
= get_imsm_map(dev
, 1);
4402 set_imsm_ord_tbl_ent(map
, j
, ord
- 1);
4407 super
->updates_pending
++;
4409 struct dl
*dl
= *dlp
;
4411 *dlp
= (*dlp
)->next
;
4412 __free_imsm_disk(dl
);
4415 #endif /* MDASSEMBLE */
4417 struct superswitch super_imsm
= {
4419 .examine_super
= examine_super_imsm
,
4420 .brief_examine_super
= brief_examine_super_imsm
,
4421 .detail_super
= detail_super_imsm
,
4422 .brief_detail_super
= brief_detail_super_imsm
,
4423 .write_init_super
= write_init_super_imsm
,
4424 .validate_geometry
= validate_geometry_imsm
,
4425 .add_to_super
= add_to_super_imsm
,
4426 .detail_platform
= detail_platform_imsm
,
4428 .match_home
= match_home_imsm
,
4429 .uuid_from_super
= uuid_from_super_imsm
,
4430 .getinfo_super
= getinfo_super_imsm
,
4431 .update_super
= update_super_imsm
,
4433 .avail_size
= avail_size_imsm
,
4435 .compare_super
= compare_super_imsm
,
4437 .load_super
= load_super_imsm
,
4438 .init_super
= init_super_imsm
,
4439 .store_super
= store_zero_imsm
,
4440 .free_super
= free_super_imsm
,
4441 .match_metadata_desc
= match_metadata_desc_imsm
,
4442 .container_content
= container_content_imsm
,
4443 .default_layout
= imsm_level_to_layout
,
4450 .open_new
= imsm_open_new
,
4451 .load_super
= load_super_imsm
,
4452 .set_array_state
= imsm_set_array_state
,
4453 .set_disk
= imsm_set_disk
,
4454 .sync_metadata
= imsm_sync_metadata
,
4455 .activate_spare
= imsm_activate_spare
,
4456 .process_update
= imsm_process_update
,
4457 .prepare_update
= imsm_prepare_update
,
4458 #endif /* MDASSEMBLE */