2 * mdadm - Intel(R) Matrix Storage Manager Support
4 * Copyright (C) 2002-2008 Intel Corporation
6 * This program is free software; you can redistribute it and/or modify it
7 * under the terms and conditions of the GNU General Public License,
8 * version 2, as published by the Free Software Foundation.
10 * This program is distributed in the hope it will be useful, but WITHOUT
11 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
12 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for
15 * You should have received a copy of the GNU General Public License along with
16 * this program; if not, write to the Free Software Foundation, Inc.,
17 * 51 Franklin St - Fifth Floor, Boston, MA 02110-1301 USA.
20 #define HAVE_STDINT_H 1
28 /* MPB == Metadata Parameter Block */
29 #define MPB_SIGNATURE "Intel Raid ISM Cfg Sig. "
30 #define MPB_SIG_LEN (strlen(MPB_SIGNATURE))
31 #define MPB_VERSION_RAID0 "1.0.00"
32 #define MPB_VERSION_RAID1 "1.1.00"
33 #define MPB_VERSION_MANY_VOLUMES_PER_ARRAY "1.2.00"
34 #define MPB_VERSION_3OR4_DISK_ARRAY "1.2.01"
35 #define MPB_VERSION_RAID5 "1.2.02"
36 #define MPB_VERSION_5OR6_DISK_ARRAY "1.2.04"
37 #define MPB_VERSION_CNG "1.2.06"
38 #define MPB_VERSION_ATTRIBS "1.3.00"
39 #define MAX_SIGNATURE_LENGTH 32
40 #define MAX_RAID_SERIAL_LEN 16
42 #define MPB_ATTRIB_CHECKSUM_VERIFY __cpu_to_le32(0x80000000)
43 #define MPB_ATTRIB_PM __cpu_to_le32(0x40000000)
44 #define MPB_ATTRIB_2TB __cpu_to_le32(0x20000000)
45 #define MPB_ATTRIB_RAID0 __cpu_to_le32(0x00000001)
46 #define MPB_ATTRIB_RAID1 __cpu_to_le32(0x00000002)
47 #define MPB_ATTRIB_RAID10 __cpu_to_le32(0x00000004)
48 #define MPB_ATTRIB_RAID1E __cpu_to_le32(0x00000008)
49 #define MPB_ATTRIB_RAID5 __cpu_to_le32(0x00000010)
50 #define MPB_ATTRIB_RAIDCNG __cpu_to_le32(0x00000020)
52 #define MPB_SECTOR_CNT 418
53 #define IMSM_RESERVED_SECTORS 4096
55 /* Disk configuration info. */
56 #define IMSM_MAX_DEVICES 255
58 __u8 serial
[MAX_RAID_SERIAL_LEN
];/* 0xD8 - 0xE7 ascii serial number */
59 __u32 total_blocks
; /* 0xE8 - 0xEB total blocks */
60 __u32 scsi_id
; /* 0xEC - 0xEF scsi ID */
61 #define SPARE_DISK __cpu_to_le32(0x01) /* Spare */
62 #define CONFIGURED_DISK __cpu_to_le32(0x02) /* Member of some RaidDev */
63 #define FAILED_DISK __cpu_to_le32(0x04) /* Permanent failure */
64 #define USABLE_DISK __cpu_to_le32(0x08) /* Fully usable unless FAILED_DISK is set */
65 __u32 status
; /* 0xF0 - 0xF3 */
66 __u32 owner_cfg_num
; /* which config 0,1,2... owns this disk */
67 #define IMSM_DISK_FILLERS 4
68 __u32 filler
[IMSM_DISK_FILLERS
]; /* 0xF4 - 0x107 MPB_DISK_FILLERS for future expansion */
71 /* RAID map configuration infos. */
73 __u32 pba_of_lba0
; /* start address of partition */
74 __u32 blocks_per_member
;/* blocks per member */
75 __u32 num_data_stripes
; /* number of data stripes */
76 __u16 blocks_per_strip
;
77 __u8 map_state
; /* Normal, Uninitialized, Degraded, Failed */
78 #define IMSM_T_STATE_NORMAL 0
79 #define IMSM_T_STATE_UNINITIALIZED 1
80 #define IMSM_T_STATE_DEGRADED 2
81 #define IMSM_T_STATE_FAILED 3
83 #define IMSM_T_RAID0 0
84 #define IMSM_T_RAID1 1
85 #define IMSM_T_RAID5 5 /* since metadata version 1.2.02 ? */
86 __u8 num_members
; /* number of member disks */
87 __u8 num_domains
; /* number of parity domains */
88 __u8 failed_disk_num
; /* valid only when state is degraded */
90 __u32 filler
[7]; /* expansion area */
91 #define IMSM_ORD_REBUILD (1 << 24)
92 __u32 disk_ord_tbl
[1]; /* disk_ord_tbl[num_members],
93 * top byte contains some flags
95 } __attribute__ ((packed
));
99 __u32 checkpoint_id
; /* id to access curr_migr_unit */
100 __u8 migr_state
; /* Normal or Migrating */
102 #define MIGR_REBUILD 1
103 #define MIGR_VERIFY 2 /* analagous to echo check > sync_action */
104 #define MIGR_GEN_MIGR 3
105 #define MIGR_STATE_CHANGE 4
106 __u8 migr_type
; /* Initializing, Rebuilding, ... */
108 __u8 fs_state
; /* fast-sync state for CnG (0xff == disabled) */
109 __u16 verify_errors
; /* number of mismatches */
110 __u16 bad_blocks
; /* number of bad blocks during verify */
112 struct imsm_map map
[1];
113 /* here comes another one if migr_state */
114 } __attribute__ ((packed
));
117 __u8 volume
[MAX_RAID_SERIAL_LEN
];
120 #define DEV_BOOTABLE __cpu_to_le32(0x01)
121 #define DEV_BOOT_DEVICE __cpu_to_le32(0x02)
122 #define DEV_READ_COALESCING __cpu_to_le32(0x04)
123 #define DEV_WRITE_COALESCING __cpu_to_le32(0x08)
124 #define DEV_LAST_SHUTDOWN_DIRTY __cpu_to_le32(0x10)
125 #define DEV_HIDDEN_AT_BOOT __cpu_to_le32(0x20)
126 #define DEV_CURRENTLY_HIDDEN __cpu_to_le32(0x40)
127 #define DEV_VERIFY_AND_FIX __cpu_to_le32(0x80)
128 #define DEV_MAP_STATE_UNINIT __cpu_to_le32(0x100)
129 #define DEV_NO_AUTO_RECOVERY __cpu_to_le32(0x200)
130 #define DEV_CLONE_N_GO __cpu_to_le32(0x400)
131 #define DEV_CLONE_MAN_SYNC __cpu_to_le32(0x800)
132 #define DEV_CNG_MASTER_DISK_NUM __cpu_to_le32(0x1000)
133 __u32 status
; /* Persistent RaidDev status */
134 __u32 reserved_blocks
; /* Reserved blocks at beginning of volume */
138 __u8 cng_master_disk
;
142 #define IMSM_DEV_FILLERS 10
143 __u32 filler
[IMSM_DEV_FILLERS
];
145 } __attribute__ ((packed
));
148 __u8 sig
[MAX_SIGNATURE_LENGTH
]; /* 0x00 - 0x1F */
149 __u32 check_sum
; /* 0x20 - 0x23 MPB Checksum */
150 __u32 mpb_size
; /* 0x24 - 0x27 Size of MPB */
151 __u32 family_num
; /* 0x28 - 0x2B Checksum from first time this config was written */
152 __u32 generation_num
; /* 0x2C - 0x2F Incremented each time this array's MPB is written */
153 __u32 error_log_size
; /* 0x30 - 0x33 in bytes */
154 __u32 attributes
; /* 0x34 - 0x37 */
155 __u8 num_disks
; /* 0x38 Number of configured disks */
156 __u8 num_raid_devs
; /* 0x39 Number of configured volumes */
157 __u8 error_log_pos
; /* 0x3A */
158 __u8 fill
[1]; /* 0x3B */
159 __u32 cache_size
; /* 0x3c - 0x40 in mb */
160 __u32 orig_family_num
; /* 0x40 - 0x43 original family num */
161 __u32 pwr_cycle_count
; /* 0x44 - 0x47 simulated power cycle count for array */
162 __u32 bbm_log_size
; /* 0x48 - 0x4B - size of bad Block Mgmt Log in bytes */
163 #define IMSM_FILLERS 35
164 __u32 filler
[IMSM_FILLERS
]; /* 0x4C - 0xD7 RAID_MPB_FILLERS */
165 struct imsm_disk disk
[1]; /* 0xD8 diskTbl[numDisks] */
166 /* here comes imsm_dev[num_raid_devs] */
167 /* here comes BBM logs */
168 } __attribute__ ((packed
));
170 #define BBM_LOG_MAX_ENTRIES 254
172 struct bbm_log_entry
{
173 __u64 defective_block_start
;
174 #define UNREADABLE 0xFFFFFFFF
175 __u32 spare_block_offset
;
176 __u16 remapped_marked_count
;
178 } __attribute__ ((__packed__
));
181 __u32 signature
; /* 0xABADB10C */
183 __u32 reserved_spare_block_count
; /* 0 */
184 __u32 reserved
; /* 0xFFFF */
185 __u64 first_spare_lba
;
186 struct bbm_log_entry mapped_block_entries
[BBM_LOG_MAX_ENTRIES
];
187 } __attribute__ ((__packed__
));
191 static char *map_state_str
[] = { "normal", "uninitialized", "degraded", "failed" };
194 static unsigned int sector_count(__u32 bytes
)
196 return ((bytes
+ (512-1)) & (~(512-1))) / 512;
199 static unsigned int mpb_sectors(struct imsm_super
*mpb
)
201 return sector_count(__le32_to_cpu(mpb
->mpb_size
));
204 /* internal representation of IMSM metadata */
207 void *buf
; /* O_DIRECT buffer for reading/writing metadata */
208 struct imsm_super
*anchor
; /* immovable parameters */
210 size_t len
; /* size of the 'buf' allocation */
211 void *next_buf
; /* for realloc'ing buf from the manager */
213 int updates_pending
; /* count of pending updates for mdmon */
214 int creating_imsm
; /* flag to indicate container creation */
215 int current_vol
; /* index of raid device undergoing creation */
216 #define IMSM_MAX_RAID_DEVS 2
217 struct imsm_dev
*dev_tbl
[IMSM_MAX_RAID_DEVS
];
221 __u8 serial
[MAX_RAID_SERIAL_LEN
];
224 struct imsm_disk disk
;
227 struct dl
*add
; /* list of disks to add while mdmon active */
228 struct dl
*missing
; /* disks removed while we weren't looking */
229 struct bbm_log
*bbm_log
;
233 unsigned long long start
, size
;
236 /* definition of messages passed to imsm_process_update */
237 enum imsm_update_type
{
238 update_activate_spare
,
243 struct imsm_update_activate_spare
{
244 enum imsm_update_type type
;
248 struct imsm_update_activate_spare
*next
;
251 struct imsm_update_create_array
{
252 enum imsm_update_type type
;
257 struct imsm_update_add_disk
{
258 enum imsm_update_type type
;
261 static struct supertype
*match_metadata_desc_imsm(char *arg
)
263 struct supertype
*st
;
265 if (strcmp(arg
, "imsm") != 0 &&
266 strcmp(arg
, "default") != 0
270 st
= malloc(sizeof(*st
));
271 memset(st
, 0, sizeof(*st
));
272 st
->ss
= &super_imsm
;
273 st
->max_devs
= IMSM_MAX_DEVICES
;
274 st
->minor_version
= 0;
280 static __u8
*get_imsm_version(struct imsm_super
*mpb
)
282 return &mpb
->sig
[MPB_SIG_LEN
];
286 /* retrieve a disk directly from the anchor when the anchor is known to be
287 * up-to-date, currently only at load time
289 static struct imsm_disk
*__get_imsm_disk(struct imsm_super
*mpb
, __u8 index
)
291 if (index
>= mpb
->num_disks
)
293 return &mpb
->disk
[index
];
297 /* retrieve a disk from the parsed metadata */
298 static struct imsm_disk
*get_imsm_disk(struct intel_super
*super
, __u8 index
)
302 for (d
= super
->disks
; d
; d
= d
->next
)
303 if (d
->index
== index
)
310 /* generate a checksum directly from the anchor when the anchor is known to be
311 * up-to-date, currently only at load or write_super after coalescing
313 static __u32
__gen_imsm_checksum(struct imsm_super
*mpb
)
315 __u32 end
= mpb
->mpb_size
/ sizeof(end
);
316 __u32
*p
= (__u32
*) mpb
;
320 sum
+= __le32_to_cpu(*p
);
324 return sum
- __le32_to_cpu(mpb
->check_sum
);
327 static size_t sizeof_imsm_map(struct imsm_map
*map
)
329 return sizeof(struct imsm_map
) + sizeof(__u32
) * (map
->num_members
- 1);
332 struct imsm_map
*get_imsm_map(struct imsm_dev
*dev
, int second_map
)
334 struct imsm_map
*map
= &dev
->vol
.map
[0];
336 if (second_map
&& !dev
->vol
.migr_state
)
338 else if (second_map
) {
341 return ptr
+ sizeof_imsm_map(map
);
347 /* return the size of the device.
348 * migr_state increases the returned size if map[0] were to be duplicated
350 static size_t sizeof_imsm_dev(struct imsm_dev
*dev
, int migr_state
)
352 size_t size
= sizeof(*dev
) - sizeof(struct imsm_map
) +
353 sizeof_imsm_map(get_imsm_map(dev
, 0));
355 /* migrating means an additional map */
356 if (dev
->vol
.migr_state
)
357 size
+= sizeof_imsm_map(get_imsm_map(dev
, 1));
359 size
+= sizeof_imsm_map(get_imsm_map(dev
, 0));
364 static struct imsm_dev
*__get_imsm_dev(struct imsm_super
*mpb
, __u8 index
)
370 if (index
>= mpb
->num_raid_devs
)
373 /* devices start after all disks */
374 offset
= ((void *) &mpb
->disk
[mpb
->num_disks
]) - _mpb
;
376 for (i
= 0; i
<= index
; i
++)
378 return _mpb
+ offset
;
380 offset
+= sizeof_imsm_dev(_mpb
+ offset
, 0);
385 static struct imsm_dev
*get_imsm_dev(struct intel_super
*super
, __u8 index
)
387 if (index
>= super
->anchor
->num_raid_devs
)
389 return super
->dev_tbl
[index
];
392 static __u32
get_imsm_ord_tbl_ent(struct imsm_dev
*dev
, int slot
)
394 struct imsm_map
*map
;
396 if (dev
->vol
.migr_state
)
397 map
= get_imsm_map(dev
, 1);
399 map
= get_imsm_map(dev
, 0);
401 /* top byte identifies disk under rebuild */
402 return __le32_to_cpu(map
->disk_ord_tbl
[slot
]);
405 #define ord_to_idx(ord) (((ord) << 8) >> 8)
406 static __u32
get_imsm_disk_idx(struct imsm_dev
*dev
, int slot
)
408 __u32 ord
= get_imsm_ord_tbl_ent(dev
, slot
);
410 return ord_to_idx(ord
);
413 static void set_imsm_ord_tbl_ent(struct imsm_map
*map
, int slot
, __u32 ord
)
415 map
->disk_ord_tbl
[slot
] = __cpu_to_le32(ord
);
418 static int get_imsm_raid_level(struct imsm_map
*map
)
420 if (map
->raid_level
== 1) {
421 if (map
->num_members
== 2)
427 return map
->raid_level
;
430 static int cmp_extent(const void *av
, const void *bv
)
432 const struct extent
*a
= av
;
433 const struct extent
*b
= bv
;
434 if (a
->start
< b
->start
)
436 if (a
->start
> b
->start
)
441 static struct extent
*get_extents(struct intel_super
*super
, struct dl
*dl
)
443 /* find a list of used extents on the given physical device */
444 struct extent
*rv
, *e
;
447 __u32 reservation
= MPB_SECTOR_CNT
+ IMSM_RESERVED_SECTORS
;
449 for (i
= 0; i
< super
->anchor
->num_raid_devs
; i
++) {
450 struct imsm_dev
*dev
= get_imsm_dev(super
, i
);
451 struct imsm_map
*map
= get_imsm_map(dev
, 0);
453 for (j
= 0; j
< map
->num_members
; j
++) {
454 __u32 index
= get_imsm_disk_idx(dev
, j
);
456 if (index
== dl
->index
)
460 rv
= malloc(sizeof(struct extent
) * (memberships
+ 1));
465 for (i
= 0; i
< super
->anchor
->num_raid_devs
; i
++) {
466 struct imsm_dev
*dev
= get_imsm_dev(super
, i
);
467 struct imsm_map
*map
= get_imsm_map(dev
, 0);
469 for (j
= 0; j
< map
->num_members
; j
++) {
470 __u32 index
= get_imsm_disk_idx(dev
, j
);
472 if (index
== dl
->index
) {
473 e
->start
= __le32_to_cpu(map
->pba_of_lba0
);
474 e
->size
= __le32_to_cpu(map
->blocks_per_member
);
479 qsort(rv
, memberships
, sizeof(*rv
), cmp_extent
);
481 /* determine the start of the metadata
482 * when no raid devices are defined use the default
483 * ...otherwise allow the metadata to truncate the value
484 * as is the case with older versions of imsm
487 struct extent
*last
= &rv
[memberships
- 1];
490 remainder
= __le32_to_cpu(dl
->disk
.total_blocks
) -
491 (last
->start
+ last
->size
);
492 if (reservation
> remainder
)
493 reservation
= remainder
;
495 e
->start
= __le32_to_cpu(dl
->disk
.total_blocks
) - reservation
;
500 /* try to determine how much space is reserved for metadata from
501 * the last get_extents() entry, otherwise fallback to the
504 static __u32
imsm_reserved_sectors(struct intel_super
*super
, struct dl
*dl
)
510 /* for spares just return a minimal reservation which will grow
511 * once the spare is picked up by an array
514 return MPB_SECTOR_CNT
;
516 e
= get_extents(super
, dl
);
518 return MPB_SECTOR_CNT
+ IMSM_RESERVED_SECTORS
;
520 /* scroll to last entry */
521 for (i
= 0; e
[i
].size
; i
++)
524 rv
= __le32_to_cpu(dl
->disk
.total_blocks
) - e
[i
].start
;
532 static void print_imsm_dev(struct imsm_dev
*dev
, char *uuid
, int disk_idx
)
536 struct imsm_map
*map
= get_imsm_map(dev
, 0);
540 printf("[%.16s]:\n", dev
->volume
);
541 printf(" UUID : %s\n", uuid
);
542 printf(" RAID Level : %d\n", get_imsm_raid_level(map
));
543 printf(" Members : %d\n", map
->num_members
);
544 for (slot
= 0; slot
< map
->num_members
; slot
++)
545 if (disk_idx
== get_imsm_disk_idx(dev
, slot
))
547 if (slot
< map
->num_members
) {
548 ord
= get_imsm_ord_tbl_ent(dev
, slot
);
549 printf(" This Slot : %d%s\n", slot
,
550 ord
& IMSM_ORD_REBUILD
? " (out-of-sync)" : "");
552 printf(" This Slot : ?\n");
553 sz
= __le32_to_cpu(dev
->size_high
);
555 sz
+= __le32_to_cpu(dev
->size_low
);
556 printf(" Array Size : %llu%s\n", (unsigned long long)sz
,
557 human_size(sz
* 512));
558 sz
= __le32_to_cpu(map
->blocks_per_member
);
559 printf(" Per Dev Size : %llu%s\n", (unsigned long long)sz
,
560 human_size(sz
* 512));
561 printf(" Sector Offset : %u\n",
562 __le32_to_cpu(map
->pba_of_lba0
));
563 printf(" Num Stripes : %u\n",
564 __le32_to_cpu(map
->num_data_stripes
));
565 printf(" Chunk Size : %u KiB\n",
566 __le16_to_cpu(map
->blocks_per_strip
) / 2);
567 printf(" Reserved : %d\n", __le32_to_cpu(dev
->reserved_blocks
));
568 printf(" Migrate State : %s", dev
->vol
.migr_state
? "migrating" : "idle");
569 if (dev
->vol
.migr_state
)
570 printf(": %s", dev
->vol
.migr_type
? "rebuilding" : "initializing");
572 printf(" Map State : %s", map_state_str
[map
->map_state
]);
573 if (dev
->vol
.migr_state
) {
574 struct imsm_map
*map
= get_imsm_map(dev
, 1);
575 printf(" <-- %s", map_state_str
[map
->map_state
]);
578 printf(" Dirty State : %s\n", dev
->vol
.dirty
? "dirty" : "clean");
581 static void print_imsm_disk(struct imsm_super
*mpb
, int index
, __u32 reserved
)
583 struct imsm_disk
*disk
= __get_imsm_disk(mpb
, index
);
584 char str
[MAX_RAID_SERIAL_LEN
+ 1];
592 snprintf(str
, MAX_RAID_SERIAL_LEN
+ 1, "%s", disk
->serial
);
593 printf(" Disk%02d Serial : %s\n", index
, str
);
595 printf(" State :%s%s%s%s\n", s
&SPARE_DISK
? " spare" : "",
596 s
&CONFIGURED_DISK
? " active" : "",
597 s
&FAILED_DISK
? " failed" : "",
598 s
&USABLE_DISK
? " usable" : "");
599 printf(" Id : %08x\n", __le32_to_cpu(disk
->scsi_id
));
600 sz
= __le32_to_cpu(disk
->total_blocks
) - reserved
;
601 printf(" Usable Size : %llu%s\n", (unsigned long long)sz
,
602 human_size(sz
* 512));
605 static void getinfo_super_imsm(struct supertype
*st
, struct mdinfo
*info
);
607 static void examine_super_imsm(struct supertype
*st
, char *homehost
)
609 struct intel_super
*super
= st
->sb
;
610 struct imsm_super
*mpb
= super
->anchor
;
611 char str
[MAX_SIGNATURE_LENGTH
];
616 __u32 reserved
= imsm_reserved_sectors(super
, super
->disks
);
619 snprintf(str
, MPB_SIG_LEN
, "%s", mpb
->sig
);
620 printf(" Magic : %s\n", str
);
621 snprintf(str
, strlen(MPB_VERSION_RAID0
), "%s", get_imsm_version(mpb
));
622 printf(" Version : %s\n", get_imsm_version(mpb
));
623 printf(" Family : %08x\n", __le32_to_cpu(mpb
->family_num
));
624 printf(" Generation : %08x\n", __le32_to_cpu(mpb
->generation_num
));
625 getinfo_super_imsm(st
, &info
);
626 fname_from_uuid(st
, &info
, nbuf
,'-');
627 printf(" UUID : %s\n", nbuf
+ 5);
628 sum
= __le32_to_cpu(mpb
->check_sum
);
629 printf(" Checksum : %08x %s\n", sum
,
630 __gen_imsm_checksum(mpb
) == sum
? "correct" : "incorrect");
631 printf(" MPB Sectors : %d\n", mpb_sectors(mpb
));
632 printf(" Disks : %d\n", mpb
->num_disks
);
633 printf(" RAID Devices : %d\n", mpb
->num_raid_devs
);
634 print_imsm_disk(mpb
, super
->disks
->index
, reserved
);
635 if (super
->bbm_log
) {
636 struct bbm_log
*log
= super
->bbm_log
;
639 printf("Bad Block Management Log:\n");
640 printf(" Log Size : %d\n", __le32_to_cpu(mpb
->bbm_log_size
));
641 printf(" Signature : %x\n", __le32_to_cpu(log
->signature
));
642 printf(" Entry Count : %d\n", __le32_to_cpu(log
->entry_count
));
643 printf(" Spare Blocks : %d\n", __le32_to_cpu(log
->reserved_spare_block_count
));
644 printf(" First Spare : %llx\n", __le64_to_cpu(log
->first_spare_lba
));
646 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
648 struct imsm_dev
*dev
= __get_imsm_dev(mpb
, i
);
650 super
->current_vol
= i
;
651 getinfo_super_imsm(st
, &info
);
652 fname_from_uuid(st
, &info
, nbuf
, '-');
653 print_imsm_dev(dev
, nbuf
+ 5, super
->disks
->index
);
655 for (i
= 0; i
< mpb
->num_disks
; i
++) {
656 if (i
== super
->disks
->index
)
658 print_imsm_disk(mpb
, i
, reserved
);
662 static void brief_examine_super_imsm(struct supertype
*st
)
664 /* We just write a generic IMSM ARRAY entry */
668 struct intel_super
*super
= st
->sb
;
671 if (!super
->anchor
->num_raid_devs
)
674 getinfo_super_imsm(st
, &info
);
675 fname_from_uuid(st
, &info
, nbuf
,'-');
676 printf("ARRAY metadata=imsm auto=md UUID=%s\n", nbuf
+ 5);
677 for (i
= 0; i
< super
->anchor
->num_raid_devs
; i
++) {
678 struct imsm_dev
*dev
= get_imsm_dev(super
, i
);
680 super
->current_vol
= i
;
681 getinfo_super_imsm(st
, &info
);
682 fname_from_uuid(st
, &info
, nbuf1
,'-');
683 printf("ARRAY /dev/md/%.16s container=%s\n"
684 " member=%d auto=mdp UUID=%s\n",
685 dev
->volume
, nbuf
+ 5, i
, nbuf1
+ 5);
689 static void detail_super_imsm(struct supertype
*st
, char *homehost
)
694 getinfo_super_imsm(st
, &info
);
695 fname_from_uuid(st
, &info
, nbuf
,'-');
696 printf("\n UUID : %s\n", nbuf
+ 5);
699 static void brief_detail_super_imsm(struct supertype
*st
)
703 getinfo_super_imsm(st
, &info
);
704 fname_from_uuid(st
, &info
, nbuf
,'-');
705 printf(" UUID=%s", nbuf
+ 5);
709 static int match_home_imsm(struct supertype
*st
, char *homehost
)
711 /* the imsm metadata format does not specify any host
712 * identification information. We return -1 since we can never
713 * confirm nor deny whether a given array is "meant" for this
714 * host. We rely on compare_super and the 'family_num' field to
715 * exclude member disks that do not belong, and we rely on
716 * mdadm.conf to specify the arrays that should be assembled.
717 * Auto-assembly may still pick up "foreign" arrays.
723 static void uuid_from_super_imsm(struct supertype
*st
, int uuid
[4])
725 /* The uuid returned here is used for:
726 * uuid to put into bitmap file (Create, Grow)
727 * uuid for backup header when saving critical section (Grow)
728 * comparing uuids when re-adding a device into an array
729 * In these cases the uuid required is that of the data-array,
730 * not the device-set.
731 * uuid to recognise same set when adding a missing device back
732 * to an array. This is a uuid for the device-set.
734 * For each of these we can make do with a truncated
735 * or hashed uuid rather than the original, as long as
737 * In each case the uuid required is that of the data-array,
738 * not the device-set.
740 /* imsm does not track uuid's so we synthesis one using sha1 on
741 * - The signature (Which is constant for all imsm array, but no matter)
742 * - the family_num of the container
743 * - the index number of the volume
744 * - the 'serial' number of the volume.
745 * Hopefully these are all constant.
747 struct intel_super
*super
= st
->sb
;
751 struct imsm_dev
*dev
= NULL
;
754 sha1_process_bytes(super
->anchor
->sig
, MPB_SIG_LEN
, &ctx
);
755 sha1_process_bytes(&super
->anchor
->family_num
, sizeof(__u32
), &ctx
);
756 if (super
->current_vol
>= 0)
757 dev
= get_imsm_dev(super
, super
->current_vol
);
759 __u32 vol
= super
->current_vol
;
760 sha1_process_bytes(&vol
, sizeof(vol
), &ctx
);
761 sha1_process_bytes(dev
->volume
, MAX_RAID_SERIAL_LEN
, &ctx
);
763 sha1_finish_ctx(&ctx
, buf
);
764 memcpy(uuid
, buf
, 4*4);
769 get_imsm_numerical_version(struct imsm_super
*mpb
, int *m
, int *p
)
771 __u8
*v
= get_imsm_version(mpb
);
772 __u8
*end
= mpb
->sig
+ MAX_SIGNATURE_LENGTH
;
773 char major
[] = { 0, 0, 0 };
774 char minor
[] = { 0 ,0, 0 };
775 char patch
[] = { 0, 0, 0 };
776 char *ver_parse
[] = { major
, minor
, patch
};
780 while (*v
!= '\0' && v
< end
) {
781 if (*v
!= '.' && j
< 2)
782 ver_parse
[i
][j
++] = *v
;
790 *m
= strtol(minor
, NULL
, 0);
791 *p
= strtol(patch
, NULL
, 0);
795 static int imsm_level_to_layout(int level
)
803 return ALGORITHM_LEFT_ASYMMETRIC
;
810 static void getinfo_super_imsm_volume(struct supertype
*st
, struct mdinfo
*info
)
812 struct intel_super
*super
= st
->sb
;
813 struct imsm_dev
*dev
= get_imsm_dev(super
, super
->current_vol
);
814 struct imsm_map
*map
= get_imsm_map(dev
, 0);
816 info
->container_member
= super
->current_vol
;
817 info
->array
.raid_disks
= map
->num_members
;
818 info
->array
.level
= get_imsm_raid_level(map
);
819 info
->array
.layout
= imsm_level_to_layout(info
->array
.level
);
820 info
->array
.md_minor
= -1;
821 info
->array
.ctime
= 0;
822 info
->array
.utime
= 0;
823 info
->array
.chunk_size
= __le16_to_cpu(map
->blocks_per_strip
) << 9;
824 info
->array
.state
= !dev
->vol
.dirty
;
826 info
->disk
.major
= 0;
827 info
->disk
.minor
= 0;
829 info
->data_offset
= __le32_to_cpu(map
->pba_of_lba0
);
830 info
->component_size
= __le32_to_cpu(map
->blocks_per_member
);
831 memset(info
->uuid
, 0, sizeof(info
->uuid
));
833 if (map
->map_state
== IMSM_T_STATE_UNINITIALIZED
|| dev
->vol
.dirty
)
834 info
->resync_start
= 0;
835 else if (dev
->vol
.migr_state
)
836 info
->resync_start
= __le32_to_cpu(dev
->vol
.curr_migr_unit
);
838 info
->resync_start
= ~0ULL;
840 strncpy(info
->name
, (char *) dev
->volume
, MAX_RAID_SERIAL_LEN
);
841 info
->name
[MAX_RAID_SERIAL_LEN
] = 0;
843 info
->array
.major_version
= -1;
844 info
->array
.minor_version
= -2;
845 sprintf(info
->text_version
, "/%s/%d",
846 devnum2devname(st
->container_dev
),
847 info
->container_member
);
848 info
->safe_mode_delay
= 4000; /* 4 secs like the Matrix driver */
849 uuid_from_super_imsm(st
, info
->uuid
);
853 static void getinfo_super_imsm(struct supertype
*st
, struct mdinfo
*info
)
855 struct intel_super
*super
= st
->sb
;
856 struct imsm_disk
*disk
;
859 if (super
->current_vol
>= 0) {
860 getinfo_super_imsm_volume(st
, info
);
864 /* Set raid_disks to zero so that Assemble will always pull in valid
867 info
->array
.raid_disks
= 0;
868 info
->array
.level
= LEVEL_CONTAINER
;
869 info
->array
.layout
= 0;
870 info
->array
.md_minor
= -1;
871 info
->array
.ctime
= 0; /* N/A for imsm */
872 info
->array
.utime
= 0;
873 info
->array
.chunk_size
= 0;
875 info
->disk
.major
= 0;
876 info
->disk
.minor
= 0;
877 info
->disk
.raid_disk
= -1;
878 info
->reshape_active
= 0;
879 info
->array
.major_version
= -1;
880 info
->array
.minor_version
= -2;
881 strcpy(info
->text_version
, "imsm");
882 info
->safe_mode_delay
= 0;
883 info
->disk
.number
= -1;
884 info
->disk
.state
= 0;
888 __u32 reserved
= imsm_reserved_sectors(super
, super
->disks
);
890 disk
= &super
->disks
->disk
;
891 info
->data_offset
= __le32_to_cpu(disk
->total_blocks
) - reserved
;
892 info
->component_size
= reserved
;
894 info
->disk
.state
= s
& CONFIGURED_DISK
? (1 << MD_DISK_ACTIVE
) : 0;
895 info
->disk
.state
|= s
& FAILED_DISK
? (1 << MD_DISK_FAULTY
) : 0;
896 info
->disk
.state
|= s
& SPARE_DISK
? 0 : (1 << MD_DISK_SYNC
);
899 /* only call uuid_from_super_imsm when this disk is part of a populated container,
900 * ->compare_super may have updated the 'num_raid_devs' field for spares
902 if (info
->disk
.state
& (1 << MD_DISK_SYNC
) || super
->anchor
->num_raid_devs
)
903 uuid_from_super_imsm(st
, info
->uuid
);
905 memcpy(info
->uuid
, uuid_match_any
, sizeof(int[4]));
908 static int update_super_imsm(struct supertype
*st
, struct mdinfo
*info
,
909 char *update
, char *devname
, int verbose
,
910 int uuid_set
, char *homehost
)
914 /* For 'assemble' and 'force' we need to return non-zero if any
915 * change was made. For others, the return value is ignored.
916 * Update options are:
917 * force-one : This device looks a bit old but needs to be included,
918 * update age info appropriately.
919 * assemble: clear any 'faulty' flag to allow this device to
921 * force-array: Array is degraded but being forced, mark it clean
922 * if that will be needed to assemble it.
924 * newdev: not used ????
925 * grow: Array has gained a new device - this is currently for
927 * resync: mark as dirty so a resync will happen.
928 * name: update the name - preserving the homehost
930 * Following are not relevant for this imsm:
931 * sparc2.2 : update from old dodgey metadata
932 * super-minor: change the preferred_minor number
933 * summaries: update redundant counters.
934 * uuid: Change the uuid of the array to match watch is given
935 * homehost: update the recorded homehost
936 * _reshape_progress: record new reshape_progress position.
939 //struct intel_super *super = st->sb;
940 //struct imsm_super *mpb = super->mpb;
942 if (strcmp(update
, "grow") == 0) {
944 if (strcmp(update
, "resync") == 0) {
945 /* dev->vol.dirty = 1; */
948 /* IMSM has no concept of UUID or homehost */
953 static size_t disks_to_mpb_size(int disks
)
957 size
= sizeof(struct imsm_super
);
958 size
+= (disks
- 1) * sizeof(struct imsm_disk
);
959 size
+= 2 * sizeof(struct imsm_dev
);
960 /* up to 2 maps per raid device (-2 for imsm_maps in imsm_dev */
961 size
+= (4 - 2) * sizeof(struct imsm_map
);
962 /* 4 possible disk_ord_tbl's */
963 size
+= 4 * (disks
- 1) * sizeof(__u32
);
968 static __u64
avail_size_imsm(struct supertype
*st
, __u64 devsize
)
970 if (devsize
< (MPB_SECTOR_CNT
+ IMSM_RESERVED_SECTORS
))
973 return devsize
- (MPB_SECTOR_CNT
+ IMSM_RESERVED_SECTORS
);
976 static int compare_super_imsm(struct supertype
*st
, struct supertype
*tst
)
980 * 0 same, or first was empty, and second was copied
981 * 1 second had wrong number
985 struct intel_super
*first
= st
->sb
;
986 struct intel_super
*sec
= tst
->sb
;
994 if (memcmp(first
->anchor
->sig
, sec
->anchor
->sig
, MAX_SIGNATURE_LENGTH
) != 0)
997 /* if an anchor does not have num_raid_devs set then it is a free
1000 if (first
->anchor
->num_raid_devs
> 0 &&
1001 sec
->anchor
->num_raid_devs
> 0) {
1002 if (first
->anchor
->family_num
!= sec
->anchor
->family_num
)
1006 /* if 'first' is a spare promote it to a populated mpb with sec's
1009 if (first
->anchor
->num_raid_devs
== 0 &&
1010 sec
->anchor
->num_raid_devs
> 0) {
1013 /* we need to copy raid device info from sec if an allocation
1014 * fails here we don't associate the spare
1016 for (i
= 0; i
< sec
->anchor
->num_raid_devs
; i
++) {
1017 first
->dev_tbl
[i
] = malloc(sizeof(struct imsm_dev
));
1018 if (!first
->dev_tbl
) {
1020 free(first
->dev_tbl
[i
]);
1021 first
->dev_tbl
[i
] = NULL
;
1023 fprintf(stderr
, "imsm: failed to associate spare\n");
1026 *first
->dev_tbl
[i
] = *sec
->dev_tbl
[i
];
1029 first
->anchor
->num_raid_devs
= sec
->anchor
->num_raid_devs
;
1030 first
->anchor
->family_num
= sec
->anchor
->family_num
;
1036 static void fd2devname(int fd
, char *name
)
1045 if (fstat(fd
, &st
) != 0)
1047 sprintf(path
, "/sys/dev/block/%d:%d",
1048 major(st
.st_rdev
), minor(st
.st_rdev
));
1050 rv
= readlink(path
, dname
, sizeof(dname
));
1055 nm
= strrchr(dname
, '/');
1057 snprintf(name
, MAX_RAID_SERIAL_LEN
, "/dev/%s", nm
);
1061 extern int scsi_get_serial(int fd
, void *buf
, size_t buf_len
);
1063 static int imsm_read_serial(int fd
, char *devname
,
1064 __u8 serial
[MAX_RAID_SERIAL_LEN
])
1066 unsigned char scsi_serial
[255];
1072 memset(scsi_serial
, 0, sizeof(scsi_serial
));
1074 rv
= scsi_get_serial(fd
, scsi_serial
, sizeof(scsi_serial
));
1076 if (rv
&& check_env("IMSM_DEVNAME_AS_SERIAL")) {
1077 memset(serial
, 0, MAX_RAID_SERIAL_LEN
);
1078 fd2devname(fd
, (char *) serial
);
1085 Name
": Failed to retrieve serial for %s\n",
1090 /* trim leading whitespace */
1091 rsp_len
= scsi_serial
[3];
1092 rsp_buf
= (char *) &scsi_serial
[4];
1097 /* truncate len to the end of rsp_buf if necessary */
1098 if (c
+ MAX_RAID_SERIAL_LEN
> rsp_buf
+ rsp_len
)
1099 len
= rsp_len
- (c
- rsp_buf
);
1101 len
= MAX_RAID_SERIAL_LEN
;
1103 /* initialize the buffer and copy rsp_buf characters */
1104 memset(serial
, 0, MAX_RAID_SERIAL_LEN
);
1105 memcpy(serial
, c
, len
);
1107 /* trim trailing whitespace starting with the last character copied */
1108 c
= (char *) &serial
[len
- 1];
1109 while (isspace(*c
) || *c
== '\0')
1115 static int serialcmp(__u8
*s1
, __u8
*s2
)
1117 return strncmp((char *) s1
, (char *) s2
, MAX_RAID_SERIAL_LEN
);
1120 static void serialcpy(__u8
*dest
, __u8
*src
)
1122 strncpy((char *) dest
, (char *) src
, MAX_RAID_SERIAL_LEN
);
1126 load_imsm_disk(int fd
, struct intel_super
*super
, char *devname
, int keep_fd
)
1133 __u8 serial
[MAX_RAID_SERIAL_LEN
];
1135 rv
= imsm_read_serial(fd
, devname
, serial
);
1140 /* check if this is a disk we have seen before. it may be a spare in
1141 * super->disks while the current anchor believes it is a raid member,
1142 * check if we need to update dl->index
1144 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
1145 if (serialcmp(dl
->serial
, serial
) == 0)
1149 dl
= malloc(sizeof(*dl
));
1156 Name
": failed to allocate disk buffer for %s\n",
1163 dl
->major
= major(stb
.st_rdev
);
1164 dl
->minor
= minor(stb
.st_rdev
);
1165 dl
->next
= super
->disks
;
1166 dl
->fd
= keep_fd
? fd
: -1;
1167 dl
->devname
= devname
? strdup(devname
) : NULL
;
1168 serialcpy(dl
->serial
, serial
);
1170 } else if (keep_fd
) {
1175 /* look up this disk's index in the current anchor */
1176 for (i
= 0; i
< super
->anchor
->num_disks
; i
++) {
1177 struct imsm_disk
*disk_iter
;
1179 disk_iter
= __get_imsm_disk(super
->anchor
, i
);
1181 if (serialcmp(disk_iter
->serial
, dl
->serial
) == 0) {
1182 dl
->disk
= *disk_iter
;
1183 /* only set index on disks that are a member of a
1184 * populated contianer, i.e. one with raid_devs
1186 if (dl
->disk
.status
& FAILED_DISK
)
1188 else if (dl
->disk
.status
& SPARE_DISK
)
1197 /* no match, maybe a stale failed drive */
1198 if (i
== super
->anchor
->num_disks
&& dl
->index
>= 0) {
1199 dl
->disk
= *__get_imsm_disk(super
->anchor
, dl
->index
);
1200 if (dl
->disk
.status
& FAILED_DISK
)
1210 static void imsm_copy_dev(struct imsm_dev
*dest
, struct imsm_dev
*src
)
1212 memcpy(dest
, src
, sizeof_imsm_dev(src
, 0));
1216 /* When migrating map0 contains the 'destination' state while map1
1217 * contains the current state. When not migrating map0 contains the
1218 * current state. This routine assumes that map[0].map_state is set to
1219 * the current array state before being called.
1221 * Migration is indicated by one of the following states
1222 * 1/ Idle (migr_state=0 map0state=normal||unitialized||degraded||failed)
1223 * 2/ Initialize (migr_state=1 migr_type=MIGR_INIT map0state=normal
1224 * map1state=unitialized)
1225 * 3/ Verify (Resync) (migr_state=1 migr_type=MIGR_REBUILD map0state=normal
1227 * 4/ Rebuild (migr_state=1 migr_type=MIGR_REBUILD map0state=normal
1228 * map1state=degraded)
1230 static void migrate(struct imsm_dev
*dev
, __u8 to_state
, int rebuild_resync
)
1232 struct imsm_map
*dest
;
1233 struct imsm_map
*src
= get_imsm_map(dev
, 0);
1235 dev
->vol
.migr_state
= 1;
1236 dev
->vol
.migr_type
= rebuild_resync
;
1237 dev
->vol
.curr_migr_unit
= 0;
1238 dest
= get_imsm_map(dev
, 1);
1240 memcpy(dest
, src
, sizeof_imsm_map(src
));
1241 src
->map_state
= to_state
;
1244 static void end_migration(struct imsm_dev
*dev
, __u8 map_state
)
1246 struct imsm_map
*map
= get_imsm_map(dev
, 0);
1248 dev
->vol
.migr_state
= 0;
1249 dev
->vol
.curr_migr_unit
= 0;
1250 map
->map_state
= map_state
;
1254 static int parse_raid_devices(struct intel_super
*super
)
1257 struct imsm_dev
*dev_new
;
1258 size_t len
, len_migr
;
1259 size_t space_needed
= 0;
1260 struct imsm_super
*mpb
= super
->anchor
;
1262 for (i
= 0; i
< super
->anchor
->num_raid_devs
; i
++) {
1263 struct imsm_dev
*dev_iter
= __get_imsm_dev(super
->anchor
, i
);
1265 len
= sizeof_imsm_dev(dev_iter
, 0);
1266 len_migr
= sizeof_imsm_dev(dev_iter
, 1);
1268 space_needed
+= len_migr
- len
;
1270 dev_new
= malloc(len_migr
);
1273 imsm_copy_dev(dev_new
, dev_iter
);
1274 super
->dev_tbl
[i
] = dev_new
;
1277 /* ensure that super->buf is large enough when all raid devices
1280 if (__le32_to_cpu(mpb
->mpb_size
) + space_needed
> super
->len
) {
1283 len
= ROUND_UP(__le32_to_cpu(mpb
->mpb_size
) + space_needed
, 512);
1284 if (posix_memalign(&buf
, 512, len
) != 0)
1287 memcpy(buf
, super
->buf
, len
);
1296 /* retrieve a pointer to the bbm log which starts after all raid devices */
1297 struct bbm_log
*__get_imsm_bbm_log(struct imsm_super
*mpb
)
1301 if (__le32_to_cpu(mpb
->bbm_log_size
)) {
1303 ptr
+= mpb
->mpb_size
- __le32_to_cpu(mpb
->bbm_log_size
);
1309 static void __free_imsm(struct intel_super
*super
, int free_disks
);
1311 /* load_imsm_mpb - read matrix metadata
1312 * allocates super->mpb to be freed by free_super
1314 static int load_imsm_mpb(int fd
, struct intel_super
*super
, char *devname
)
1316 unsigned long long dsize
;
1317 unsigned long long sectors
;
1319 struct imsm_super
*anchor
;
1323 get_dev_size(fd
, NULL
, &dsize
);
1325 if (lseek64(fd
, dsize
- (512 * 2), SEEK_SET
) < 0) {
1328 Name
": Cannot seek to anchor block on %s: %s\n",
1329 devname
, strerror(errno
));
1333 if (posix_memalign((void**)&anchor
, 512, 512) != 0) {
1336 Name
": Failed to allocate imsm anchor buffer"
1337 " on %s\n", devname
);
1340 if (read(fd
, anchor
, 512) != 512) {
1343 Name
": Cannot read anchor block on %s: %s\n",
1344 devname
, strerror(errno
));
1349 if (strncmp((char *) anchor
->sig
, MPB_SIGNATURE
, MPB_SIG_LEN
) != 0) {
1352 Name
": no IMSM anchor on %s\n", devname
);
1357 __free_imsm(super
, 0);
1358 super
->len
= ROUND_UP(anchor
->mpb_size
, 512);
1359 if (posix_memalign(&super
->buf
, 512, super
->len
) != 0) {
1362 Name
": unable to allocate %zu byte mpb buffer\n",
1367 memcpy(super
->buf
, anchor
, 512);
1369 sectors
= mpb_sectors(anchor
) - 1;
1372 rc
= load_imsm_disk(fd
, super
, devname
, 0);
1374 rc
= parse_raid_devices(super
);
1378 /* read the extended mpb */
1379 if (lseek64(fd
, dsize
- (512 * (2 + sectors
)), SEEK_SET
) < 0) {
1382 Name
": Cannot seek to extended mpb on %s: %s\n",
1383 devname
, strerror(errno
));
1387 if (read(fd
, super
->buf
+ 512, super
->len
- 512) != super
->len
- 512) {
1390 Name
": Cannot read extended mpb on %s: %s\n",
1391 devname
, strerror(errno
));
1395 check_sum
= __gen_imsm_checksum(super
->anchor
);
1396 if (check_sum
!= __le32_to_cpu(super
->anchor
->check_sum
)) {
1399 Name
": IMSM checksum %x != %x on %s\n",
1400 check_sum
, __le32_to_cpu(super
->anchor
->check_sum
),
1405 /* FIXME the BBM log is disk specific so we cannot use this global
1406 * buffer for all disks. Ok for now since we only look at the global
1407 * bbm_log_size parameter to gate assembly
1409 super
->bbm_log
= __get_imsm_bbm_log(super
->anchor
);
1411 rc
= load_imsm_disk(fd
, super
, devname
, 0);
1413 rc
= parse_raid_devices(super
);
1418 static void __free_imsm_disk(struct dl
*d
)
1427 static void free_imsm_disks(struct intel_super
*super
)
1431 while (super
->disks
) {
1433 super
->disks
= d
->next
;
1434 __free_imsm_disk(d
);
1436 while (super
->missing
) {
1438 super
->missing
= d
->next
;
1439 __free_imsm_disk(d
);
1444 /* free all the pieces hanging off of a super pointer */
1445 static void __free_imsm(struct intel_super
*super
, int free_disks
)
1454 free_imsm_disks(super
);
1455 for (i
= 0; i
< IMSM_MAX_RAID_DEVS
; i
++)
1456 if (super
->dev_tbl
[i
]) {
1457 free(super
->dev_tbl
[i
]);
1458 super
->dev_tbl
[i
] = NULL
;
1462 static void free_imsm(struct intel_super
*super
)
1464 __free_imsm(super
, 1);
1468 static void free_super_imsm(struct supertype
*st
)
1470 struct intel_super
*super
= st
->sb
;
1479 static struct intel_super
*alloc_super(int creating_imsm
)
1481 struct intel_super
*super
= malloc(sizeof(*super
));
1484 memset(super
, 0, sizeof(*super
));
1485 super
->creating_imsm
= creating_imsm
;
1486 super
->current_vol
= -1;
1493 /* find_missing - helper routine for load_super_imsm_all that identifies
1494 * disks that have disappeared from the system. This routine relies on
1495 * the mpb being uptodate, which it is at load time.
1497 static int find_missing(struct intel_super
*super
)
1500 struct imsm_super
*mpb
= super
->anchor
;
1502 struct imsm_disk
*disk
;
1504 for (i
= 0; i
< mpb
->num_disks
; i
++) {
1505 disk
= __get_imsm_disk(mpb
, i
);
1506 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
1507 if (serialcmp(dl
->disk
.serial
, disk
->serial
) == 0)
1511 /* ok we have a 'disk' without a live entry in
1514 if (disk
->status
& FAILED_DISK
|| !(disk
->status
& USABLE_DISK
))
1515 continue; /* never mind, already marked */
1517 dl
= malloc(sizeof(*dl
));
1523 dl
->devname
= strdup("missing");
1525 serialcpy(dl
->serial
, disk
->serial
);
1527 dl
->next
= super
->missing
;
1528 super
->missing
= dl
;
1534 static int load_super_imsm_all(struct supertype
*st
, int fd
, void **sbp
,
1535 char *devname
, int keep_fd
)
1538 struct intel_super
*super
;
1539 struct mdinfo
*sd
, *best
= NULL
;
1546 /* check if this disk is a member of an active array */
1547 sra
= sysfs_read(fd
, 0, GET_LEVEL
|GET_VERSION
|GET_DEVS
|GET_STATE
);
1551 if (sra
->array
.major_version
!= -1 ||
1552 sra
->array
.minor_version
!= -2 ||
1553 strcmp(sra
->text_version
, "imsm") != 0)
1556 super
= alloc_super(0);
1560 /* find the most up to date disk in this array, skipping spares */
1561 for (sd
= sra
->devs
; sd
; sd
= sd
->next
) {
1562 sprintf(nm
, "%d:%d", sd
->disk
.major
, sd
->disk
.minor
);
1563 dfd
= dev_open(nm
, keep_fd
? O_RDWR
: O_RDONLY
);
1568 rv
= load_imsm_mpb(dfd
, super
, NULL
);
1572 if (super
->anchor
->num_raid_devs
== 0)
1575 gen
= __le32_to_cpu(super
->anchor
->generation_num
);
1576 if (!best
|| gen
> bestgen
) {
1591 /* load the most up to date anchor */
1592 sprintf(nm
, "%d:%d", best
->disk
.major
, best
->disk
.minor
);
1593 dfd
= dev_open(nm
, O_RDONLY
);
1598 rv
= load_imsm_mpb(dfd
, super
, NULL
);
1605 /* re-parse the disk list with the current anchor */
1606 for (sd
= sra
->devs
; sd
; sd
= sd
->next
) {
1607 sprintf(nm
, "%d:%d", sd
->disk
.major
, sd
->disk
.minor
);
1608 dfd
= dev_open(nm
, keep_fd
? O_RDWR
: O_RDONLY
);
1613 load_imsm_disk(dfd
, super
, NULL
, keep_fd
);
1619 if (find_missing(super
) != 0) {
1624 if (st
->subarray
[0]) {
1625 if (atoi(st
->subarray
) <= super
->anchor
->num_raid_devs
)
1626 super
->current_vol
= atoi(st
->subarray
);
1632 st
->container_dev
= fd2devnum(fd
);
1633 if (st
->ss
== NULL
) {
1634 st
->ss
= &super_imsm
;
1635 st
->minor_version
= 0;
1636 st
->max_devs
= IMSM_MAX_DEVICES
;
1638 st
->loaded_container
= 1;
1644 static int load_super_imsm(struct supertype
*st
, int fd
, char *devname
)
1646 struct intel_super
*super
;
1650 if (load_super_imsm_all(st
, fd
, &st
->sb
, devname
, 1) == 0)
1653 if (st
->subarray
[0])
1654 return 1; /* FIXME */
1656 super
= alloc_super(0);
1659 Name
": malloc of %zu failed.\n",
1664 rv
= load_imsm_mpb(fd
, super
, devname
);
1669 Name
": Failed to load all information "
1670 "sections on %s\n", devname
);
1676 if (st
->ss
== NULL
) {
1677 st
->ss
= &super_imsm
;
1678 st
->minor_version
= 0;
1679 st
->max_devs
= IMSM_MAX_DEVICES
;
1681 st
->loaded_container
= 0;
1686 static __u16
info_to_blocks_per_strip(mdu_array_info_t
*info
)
1688 if (info
->level
== 1)
1690 return info
->chunk_size
>> 9;
1693 static __u32
info_to_num_data_stripes(mdu_array_info_t
*info
)
1697 num_stripes
= (info
->size
* 2) / info_to_blocks_per_strip(info
);
1698 if (info
->level
== 1)
1704 static __u32
info_to_blocks_per_member(mdu_array_info_t
*info
)
1706 return (info
->size
* 2) & ~(info_to_blocks_per_strip(info
) - 1);
1709 static void imsm_update_version_info(struct intel_super
*super
)
1711 /* update the version and attributes */
1712 struct imsm_super
*mpb
= super
->anchor
;
1714 struct imsm_dev
*dev
;
1715 struct imsm_map
*map
;
1718 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
1719 dev
= get_imsm_dev(super
, i
);
1720 map
= get_imsm_map(dev
, 0);
1721 if (__le32_to_cpu(dev
->size_high
) > 0)
1722 mpb
->attributes
|= MPB_ATTRIB_2TB
;
1724 /* FIXME detect when an array spans a port multiplier */
1726 mpb
->attributes
|= MPB_ATTRIB_PM
;
1729 if (mpb
->num_raid_devs
> 1 ||
1730 mpb
->attributes
!= MPB_ATTRIB_CHECKSUM_VERIFY
) {
1731 version
= MPB_VERSION_ATTRIBS
;
1732 switch (get_imsm_raid_level(map
)) {
1733 case 0: mpb
->attributes
|= MPB_ATTRIB_RAID0
; break;
1734 case 1: mpb
->attributes
|= MPB_ATTRIB_RAID1
; break;
1735 case 10: mpb
->attributes
|= MPB_ATTRIB_RAID10
; break;
1736 case 5: mpb
->attributes
|= MPB_ATTRIB_RAID5
; break;
1739 if (map
->num_members
>= 5)
1740 version
= MPB_VERSION_5OR6_DISK_ARRAY
;
1741 else if (dev
->status
== DEV_CLONE_N_GO
)
1742 version
= MPB_VERSION_CNG
;
1743 else if (get_imsm_raid_level(map
) == 5)
1744 version
= MPB_VERSION_RAID5
;
1745 else if (map
->num_members
>= 3)
1746 version
= MPB_VERSION_3OR4_DISK_ARRAY
;
1747 else if (get_imsm_raid_level(map
) == 1)
1748 version
= MPB_VERSION_RAID1
;
1750 version
= MPB_VERSION_RAID0
;
1752 strcpy(((char *) mpb
->sig
) + strlen(MPB_SIGNATURE
), version
);
1756 static int init_super_imsm_volume(struct supertype
*st
, mdu_array_info_t
*info
,
1757 unsigned long long size
, char *name
,
1758 char *homehost
, int *uuid
)
1760 /* We are creating a volume inside a pre-existing container.
1761 * so st->sb is already set.
1763 struct intel_super
*super
= st
->sb
;
1764 struct imsm_super
*mpb
= super
->anchor
;
1765 struct imsm_dev
*dev
;
1766 struct imsm_vol
*vol
;
1767 struct imsm_map
*map
;
1768 int idx
= mpb
->num_raid_devs
;
1770 unsigned long long array_blocks
;
1772 size_t size_old
, size_new
;
1774 if (mpb
->num_raid_devs
>= 2) {
1775 fprintf(stderr
, Name
": This imsm-container already has the "
1776 "maximum of 2 volumes\n");
1780 /* ensure the mpb is large enough for the new data */
1781 size_old
= __le32_to_cpu(mpb
->mpb_size
);
1782 size_new
= disks_to_mpb_size(info
->nr_disks
);
1783 if (size_new
> size_old
) {
1785 size_t size_round
= ROUND_UP(size_new
, 512);
1787 if (posix_memalign(&mpb_new
, 512, size_round
) != 0) {
1788 fprintf(stderr
, Name
": could not allocate new mpb\n");
1791 memcpy(mpb_new
, mpb
, size_old
);
1794 super
->anchor
= mpb_new
;
1795 mpb
->mpb_size
= __cpu_to_le32(size_new
);
1796 memset(mpb_new
+ size_old
, 0, size_round
- size_old
);
1798 super
->current_vol
= idx
;
1799 /* when creating the first raid device in this container set num_disks
1800 * to zero, i.e. delete this spare and add raid member devices in
1801 * add_to_super_imsm_volume()
1803 if (super
->current_vol
== 0)
1805 sprintf(st
->subarray
, "%d", idx
);
1806 dev
= malloc(sizeof(*dev
) + sizeof(__u32
) * (info
->raid_disks
- 1));
1808 fprintf(stderr
, Name
": could not allocate raid device\n");
1811 strncpy((char *) dev
->volume
, name
, MAX_RAID_SERIAL_LEN
);
1812 array_blocks
= calc_array_size(info
->level
, info
->raid_disks
,
1813 info
->layout
, info
->chunk_size
,
1815 dev
->size_low
= __cpu_to_le32((__u32
) array_blocks
);
1816 dev
->size_high
= __cpu_to_le32((__u32
) (array_blocks
>> 32));
1817 dev
->status
= __cpu_to_le32(0);
1818 dev
->reserved_blocks
= __cpu_to_le32(0);
1820 vol
->migr_state
= 0;
1821 vol
->migr_type
= MIGR_INIT
;
1823 vol
->curr_migr_unit
= 0;
1824 for (i
= 0; i
< idx
; i
++) {
1825 struct imsm_dev
*prev
= get_imsm_dev(super
, i
);
1826 struct imsm_map
*pmap
= get_imsm_map(prev
, 0);
1828 offset
+= __le32_to_cpu(pmap
->blocks_per_member
);
1829 offset
+= IMSM_RESERVED_SECTORS
;
1831 map
= get_imsm_map(dev
, 0);
1832 map
->pba_of_lba0
= __cpu_to_le32(offset
);
1833 map
->blocks_per_member
= __cpu_to_le32(info_to_blocks_per_member(info
));
1834 map
->blocks_per_strip
= __cpu_to_le16(info_to_blocks_per_strip(info
));
1835 map
->num_data_stripes
= __cpu_to_le32(info_to_num_data_stripes(info
));
1836 map
->map_state
= info
->level
? IMSM_T_STATE_UNINITIALIZED
:
1837 IMSM_T_STATE_NORMAL
;
1839 if (info
->level
== 1 && info
->raid_disks
> 2) {
1840 fprintf(stderr
, Name
": imsm does not support more than 2 disks"
1841 "in a raid1 volume\n");
1844 if (info
->level
== 10) {
1845 map
->raid_level
= 1;
1846 map
->num_domains
= info
->raid_disks
/ 2;
1848 map
->raid_level
= info
->level
;
1849 map
->num_domains
= !!map
->raid_level
;
1852 map
->num_members
= info
->raid_disks
;
1853 for (i
= 0; i
< map
->num_members
; i
++) {
1854 /* initialized in add_to_super */
1855 set_imsm_ord_tbl_ent(map
, i
, 0);
1857 mpb
->num_raid_devs
++;
1858 super
->dev_tbl
[super
->current_vol
] = dev
;
1860 imsm_update_version_info(super
);
1865 static int init_super_imsm(struct supertype
*st
, mdu_array_info_t
*info
,
1866 unsigned long long size
, char *name
,
1867 char *homehost
, int *uuid
)
1869 /* This is primarily called by Create when creating a new array.
1870 * We will then get add_to_super called for each component, and then
1871 * write_init_super called to write it out to each device.
1872 * For IMSM, Create can create on fresh devices or on a pre-existing
1874 * To create on a pre-existing array a different method will be called.
1875 * This one is just for fresh drives.
1877 struct intel_super
*super
;
1878 struct imsm_super
*mpb
;
1887 return init_super_imsm_volume(st
, info
, size
, name
, homehost
,
1890 super
= alloc_super(1);
1893 mpb_size
= disks_to_mpb_size(info
->nr_disks
);
1894 if (posix_memalign(&super
->buf
, 512, mpb_size
) != 0) {
1899 memset(mpb
, 0, mpb_size
);
1901 mpb
->attributes
= MPB_ATTRIB_CHECKSUM_VERIFY
;
1903 version
= (char *) mpb
->sig
;
1904 strcpy(version
, MPB_SIGNATURE
);
1905 version
+= strlen(MPB_SIGNATURE
);
1906 strcpy(version
, MPB_VERSION_RAID0
);
1907 mpb
->mpb_size
= mpb_size
;
1914 static void add_to_super_imsm_volume(struct supertype
*st
, mdu_disk_info_t
*dk
,
1915 int fd
, char *devname
)
1917 struct intel_super
*super
= st
->sb
;
1918 struct imsm_super
*mpb
= super
->anchor
;
1920 struct imsm_dev
*dev
;
1921 struct imsm_map
*map
;
1923 dev
= get_imsm_dev(super
, super
->current_vol
);
1924 map
= get_imsm_map(dev
, 0);
1926 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
1927 if (dl
->major
== dk
->major
&&
1928 dl
->minor
== dk
->minor
)
1931 if (!dl
|| ! (dk
->state
& (1<<MD_DISK_SYNC
)))
1934 /* add a pristine spare to the metadata */
1935 if (dl
->index
< 0) {
1936 dl
->index
= super
->anchor
->num_disks
;
1937 super
->anchor
->num_disks
++;
1939 set_imsm_ord_tbl_ent(map
, dk
->number
, dl
->index
);
1940 dl
->disk
.status
= CONFIGURED_DISK
| USABLE_DISK
;
1942 /* if we are creating the first raid device update the family number */
1943 if (super
->current_vol
== 0) {
1945 struct imsm_dev
*_dev
= __get_imsm_dev(mpb
, 0);
1946 struct imsm_disk
*_disk
= __get_imsm_disk(mpb
, dl
->index
);
1950 sum
= __gen_imsm_checksum(mpb
);
1951 mpb
->family_num
= __cpu_to_le32(sum
);
1955 static void add_to_super_imsm(struct supertype
*st
, mdu_disk_info_t
*dk
,
1956 int fd
, char *devname
)
1958 struct intel_super
*super
= st
->sb
;
1960 unsigned long long size
;
1965 if (super
->current_vol
>= 0) {
1966 add_to_super_imsm_volume(st
, dk
, fd
, devname
);
1971 dd
= malloc(sizeof(*dd
));
1974 Name
": malloc failed %s:%d.\n", __func__
, __LINE__
);
1977 memset(dd
, 0, sizeof(*dd
));
1978 dd
->major
= major(stb
.st_rdev
);
1979 dd
->minor
= minor(stb
.st_rdev
);
1981 dd
->devname
= devname
? strdup(devname
) : NULL
;
1983 rv
= imsm_read_serial(fd
, devname
, dd
->serial
);
1986 Name
": failed to retrieve scsi serial, aborting\n");
1991 get_dev_size(fd
, NULL
, &size
);
1993 serialcpy(dd
->disk
.serial
, dd
->serial
);
1994 dd
->disk
.total_blocks
= __cpu_to_le32(size
);
1995 dd
->disk
.status
= USABLE_DISK
| SPARE_DISK
;
1996 if (sysfs_disk_to_scsi_id(fd
, &id
) == 0)
1997 dd
->disk
.scsi_id
= __cpu_to_le32(id
);
1999 dd
->disk
.scsi_id
= __cpu_to_le32(0);
2001 if (st
->update_tail
) {
2002 dd
->next
= super
->add
;
2005 dd
->next
= super
->disks
;
2010 static int store_imsm_mpb(int fd
, struct intel_super
*super
);
2012 /* spare records have their own family number and do not have any defined raid
2015 static int write_super_imsm_spares(struct intel_super
*super
, int doclose
)
2017 struct imsm_super mpb_save
;
2018 struct imsm_super
*mpb
= super
->anchor
;
2023 mpb
->num_raid_devs
= 0;
2025 mpb
->mpb_size
= sizeof(struct imsm_super
);
2026 mpb
->generation_num
= __cpu_to_le32(1UL);
2028 for (d
= super
->disks
; d
; d
= d
->next
) {
2032 mpb
->disk
[0] = d
->disk
;
2033 sum
= __gen_imsm_checksum(mpb
);
2034 mpb
->family_num
= __cpu_to_le32(sum
);
2035 sum
= __gen_imsm_checksum(mpb
);
2036 mpb
->check_sum
= __cpu_to_le32(sum
);
2038 if (store_imsm_mpb(d
->fd
, super
)) {
2039 fprintf(stderr
, "%s: failed for device %d:%d %s\n",
2040 __func__
, d
->major
, d
->minor
, strerror(errno
));
2054 static int write_super_imsm(struct intel_super
*super
, int doclose
)
2056 struct imsm_super
*mpb
= super
->anchor
;
2062 __u32 mpb_size
= sizeof(struct imsm_super
) - sizeof(struct imsm_disk
);
2064 /* 'generation' is incremented everytime the metadata is written */
2065 generation
= __le32_to_cpu(mpb
->generation_num
);
2067 mpb
->generation_num
= __cpu_to_le32(generation
);
2069 mpb_size
+= sizeof(struct imsm_disk
) * mpb
->num_disks
;
2070 for (d
= super
->disks
; d
; d
= d
->next
) {
2074 mpb
->disk
[d
->index
] = d
->disk
;
2076 for (d
= super
->missing
; d
; d
= d
->next
)
2077 mpb
->disk
[d
->index
] = d
->disk
;
2079 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
2080 struct imsm_dev
*dev
= __get_imsm_dev(mpb
, i
);
2082 imsm_copy_dev(dev
, super
->dev_tbl
[i
]);
2083 mpb_size
+= sizeof_imsm_dev(dev
, 0);
2085 mpb_size
+= __le32_to_cpu(mpb
->bbm_log_size
);
2086 mpb
->mpb_size
= __cpu_to_le32(mpb_size
);
2088 /* recalculate checksum */
2089 sum
= __gen_imsm_checksum(mpb
);
2090 mpb
->check_sum
= __cpu_to_le32(sum
);
2092 /* write the mpb for disks that compose raid devices */
2093 for (d
= super
->disks
; d
; d
= d
->next
) {
2096 if (store_imsm_mpb(d
->fd
, super
))
2097 fprintf(stderr
, "%s: failed for device %d:%d %s\n",
2098 __func__
, d
->major
, d
->minor
, strerror(errno
));
2106 return write_super_imsm_spares(super
, doclose
);
2112 static int create_array(struct supertype
*st
)
2115 struct imsm_update_create_array
*u
;
2116 struct intel_super
*super
= st
->sb
;
2117 struct imsm_dev
*dev
= get_imsm_dev(super
, super
->current_vol
);
2119 len
= sizeof(*u
) - sizeof(*dev
) + sizeof_imsm_dev(dev
, 0);
2122 fprintf(stderr
, "%s: failed to allocate update buffer\n",
2127 u
->type
= update_create_array
;
2128 u
->dev_idx
= super
->current_vol
;
2129 imsm_copy_dev(&u
->dev
, dev
);
2130 append_metadata_update(st
, u
, len
);
2135 static int _add_disk(struct supertype
*st
)
2137 struct intel_super
*super
= st
->sb
;
2139 struct imsm_update_add_disk
*u
;
2147 fprintf(stderr
, "%s: failed to allocate update buffer\n",
2152 u
->type
= update_add_disk
;
2153 append_metadata_update(st
, u
, len
);
2158 static int write_init_super_imsm(struct supertype
*st
)
2160 if (st
->update_tail
) {
2161 /* queue the recently created array / added disk
2162 * as a metadata update */
2163 struct intel_super
*super
= st
->sb
;
2167 /* determine if we are creating a volume or adding a disk */
2168 if (super
->current_vol
< 0) {
2169 /* in the add disk case we are running in mdmon
2170 * context, so don't close fd's
2172 return _add_disk(st
);
2174 rv
= create_array(st
);
2176 for (d
= super
->disks
; d
; d
= d
->next
) {
2183 return write_super_imsm(st
->sb
, 1);
2187 static int store_zero_imsm(struct supertype
*st
, int fd
)
2189 unsigned long long dsize
;
2192 get_dev_size(fd
, NULL
, &dsize
);
2194 /* first block is stored on second to last sector of the disk */
2195 if (lseek64(fd
, dsize
- (512 * 2), SEEK_SET
) < 0)
2198 if (posix_memalign(&buf
, 512, 512) != 0)
2201 memset(buf
, 0, 512);
2202 if (write(fd
, buf
, 512) != 512)
2207 static int imsm_bbm_log_size(struct imsm_super
*mpb
)
2209 return __le32_to_cpu(mpb
->bbm_log_size
);
2213 static int validate_geometry_imsm_container(struct supertype
*st
, int level
,
2214 int layout
, int raiddisks
, int chunk
,
2215 unsigned long long size
, char *dev
,
2216 unsigned long long *freesize
,
2220 unsigned long long ldsize
;
2222 if (level
!= LEVEL_CONTAINER
)
2227 fd
= open(dev
, O_RDONLY
|O_EXCL
, 0);
2230 fprintf(stderr
, Name
": imsm: Cannot open %s: %s\n",
2231 dev
, strerror(errno
));
2234 if (!get_dev_size(fd
, dev
, &ldsize
)) {
2240 *freesize
= avail_size_imsm(st
, ldsize
>> 9);
2245 /* validate_geometry_imsm_volume - lifted from validate_geometry_ddf_bvd
2246 * FIX ME add ahci details
2248 static int validate_geometry_imsm_volume(struct supertype
*st
, int level
,
2249 int layout
, int raiddisks
, int chunk
,
2250 unsigned long long size
, char *dev
,
2251 unsigned long long *freesize
,
2255 struct intel_super
*super
= st
->sb
;
2257 unsigned long long pos
= 0;
2258 unsigned long long maxsize
;
2262 if (level
== LEVEL_CONTAINER
)
2265 if (level
== 1 && raiddisks
> 2) {
2267 fprintf(stderr
, Name
": imsm does not support more "
2268 "than 2 in a raid1 configuration\n");
2272 /* We must have the container info already read in. */
2277 /* General test: make sure there is space for
2278 * 'raiddisks' device extents of size 'size' at a given
2281 unsigned long long minsize
= size
*2 /* convert to blocks */;
2282 unsigned long long start_offset
= ~0ULL;
2285 minsize
= MPB_SECTOR_CNT
+ IMSM_RESERVED_SECTORS
;
2286 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
2291 e
= get_extents(super
, dl
);
2294 unsigned long long esize
;
2295 esize
= e
[i
].start
- pos
;
2296 if (esize
>= minsize
)
2298 if (found
&& start_offset
== ~0ULL) {
2301 } else if (found
&& pos
!= start_offset
) {
2305 pos
= e
[i
].start
+ e
[i
].size
;
2307 } while (e
[i
-1].size
);
2312 if (dcnt
< raiddisks
) {
2314 fprintf(stderr
, Name
": imsm: Not enough "
2315 "devices with space for this array "
2322 /* This device must be a member of the set */
2323 if (stat(dev
, &stb
) < 0)
2325 if ((S_IFMT
& stb
.st_mode
) != S_IFBLK
)
2327 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
2328 if (dl
->major
== major(stb
.st_rdev
) &&
2329 dl
->minor
== minor(stb
.st_rdev
))
2334 fprintf(stderr
, Name
": %s is not in the "
2335 "same imsm set\n", dev
);
2338 e
= get_extents(super
, dl
);
2342 unsigned long long esize
;
2343 esize
= e
[i
].start
- pos
;
2344 if (esize
>= maxsize
)
2346 pos
= e
[i
].start
+ e
[i
].size
;
2348 } while (e
[i
-1].size
);
2349 *freesize
= maxsize
;
2354 static int validate_geometry_imsm(struct supertype
*st
, int level
, int layout
,
2355 int raiddisks
, int chunk
, unsigned long long size
,
2356 char *dev
, unsigned long long *freesize
,
2362 /* if given unused devices create a container
2363 * if given given devices in a container create a member volume
2365 if (level
== LEVEL_CONTAINER
) {
2366 /* Must be a fresh device to add to a container */
2367 return validate_geometry_imsm_container(st
, level
, layout
,
2368 raiddisks
, chunk
, size
,
2374 /* creating in a given container */
2375 return validate_geometry_imsm_volume(st
, level
, layout
,
2376 raiddisks
, chunk
, size
,
2377 dev
, freesize
, verbose
);
2380 /* limit creation to the following levels */
2392 /* This device needs to be a device in an 'imsm' container */
2393 fd
= open(dev
, O_RDONLY
|O_EXCL
, 0);
2397 Name
": Cannot create this array on device %s\n",
2402 if (errno
!= EBUSY
|| (fd
= open(dev
, O_RDONLY
, 0)) < 0) {
2404 fprintf(stderr
, Name
": Cannot open %s: %s\n",
2405 dev
, strerror(errno
));
2408 /* Well, it is in use by someone, maybe an 'imsm' container. */
2409 cfd
= open_container(fd
);
2413 fprintf(stderr
, Name
": Cannot use %s: It is busy\n",
2417 sra
= sysfs_read(cfd
, 0, GET_VERSION
);
2419 if (sra
&& sra
->array
.major_version
== -1 &&
2420 strcmp(sra
->text_version
, "imsm") == 0) {
2421 /* This is a member of a imsm container. Load the container
2422 * and try to create a volume
2424 struct intel_super
*super
;
2426 if (load_super_imsm_all(st
, cfd
, (void **) &super
, NULL
, 1) == 0) {
2428 st
->container_dev
= fd2devnum(cfd
);
2430 return validate_geometry_imsm_volume(st
, level
, layout
,
2436 } else /* may belong to another container */
2441 #endif /* MDASSEMBLE */
2443 static struct mdinfo
*container_content_imsm(struct supertype
*st
)
2445 /* Given a container loaded by load_super_imsm_all,
2446 * extract information about all the arrays into
2449 * For each imsm_dev create an mdinfo, fill it in,
2450 * then look for matching devices in super->disks
2451 * and create appropriate device mdinfo.
2453 struct intel_super
*super
= st
->sb
;
2454 struct imsm_super
*mpb
= super
->anchor
;
2455 struct mdinfo
*rest
= NULL
;
2458 /* do not assemble arrays that might have bad blocks */
2459 if (imsm_bbm_log_size(super
->anchor
)) {
2460 fprintf(stderr
, Name
": BBM log found in metadata. "
2461 "Cannot activate array(s).\n");
2465 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
2466 struct imsm_dev
*dev
= get_imsm_dev(super
, i
);
2467 struct imsm_map
*map
= get_imsm_map(dev
, 0);
2468 struct mdinfo
*this;
2471 this = malloc(sizeof(*this));
2472 memset(this, 0, sizeof(*this));
2475 super
->current_vol
= i
;
2476 getinfo_super_imsm_volume(st
, this);
2477 for (slot
= 0 ; slot
< map
->num_members
; slot
++) {
2478 struct mdinfo
*info_d
;
2486 idx
= get_imsm_disk_idx(dev
, slot
);
2487 ord
= get_imsm_ord_tbl_ent(dev
, slot
);
2488 for (d
= super
->disks
; d
; d
= d
->next
)
2489 if (d
->index
== idx
)
2495 s
= d
? d
->disk
.status
: 0;
2496 if (s
& FAILED_DISK
)
2498 if (!(s
& USABLE_DISK
))
2500 if (ord
& IMSM_ORD_REBUILD
)
2504 * if we skip some disks the array will be assmebled degraded;
2505 * reset resync start to avoid a dirty-degraded situation
2507 * FIXME handle dirty degraded
2509 if (skip
&& !dev
->vol
.dirty
)
2510 this->resync_start
= ~0ULL;
2514 info_d
= malloc(sizeof(*info_d
));
2516 fprintf(stderr
, Name
": failed to allocate disk"
2517 " for volume %s\n", (char *) dev
->volume
);
2522 memset(info_d
, 0, sizeof(*info_d
));
2523 info_d
->next
= this->devs
;
2524 this->devs
= info_d
;
2526 info_d
->disk
.number
= d
->index
;
2527 info_d
->disk
.major
= d
->major
;
2528 info_d
->disk
.minor
= d
->minor
;
2529 info_d
->disk
.raid_disk
= slot
;
2531 this->array
.working_disks
++;
2533 info_d
->events
= __le32_to_cpu(mpb
->generation_num
);
2534 info_d
->data_offset
= __le32_to_cpu(map
->pba_of_lba0
);
2535 info_d
->component_size
= __le32_to_cpu(map
->blocks_per_member
);
2537 strcpy(info_d
->name
, d
->devname
);
2547 static int imsm_open_new(struct supertype
*c
, struct active_array
*a
,
2550 struct intel_super
*super
= c
->sb
;
2551 struct imsm_super
*mpb
= super
->anchor
;
2553 if (atoi(inst
) >= mpb
->num_raid_devs
) {
2554 fprintf(stderr
, "%s: subarry index %d, out of range\n",
2555 __func__
, atoi(inst
));
2559 dprintf("imsm: open_new %s\n", inst
);
2560 a
->info
.container_member
= atoi(inst
);
2564 static __u8
imsm_check_degraded(struct intel_super
*super
, struct imsm_dev
*dev
, int failed
)
2566 struct imsm_map
*map
= get_imsm_map(dev
, 0);
2569 return map
->map_state
== IMSM_T_STATE_UNINITIALIZED
?
2570 IMSM_T_STATE_UNINITIALIZED
: IMSM_T_STATE_NORMAL
;
2572 switch (get_imsm_raid_level(map
)) {
2574 return IMSM_T_STATE_FAILED
;
2577 if (failed
< map
->num_members
)
2578 return IMSM_T_STATE_DEGRADED
;
2580 return IMSM_T_STATE_FAILED
;
2585 * check to see if any mirrors have failed, otherwise we
2586 * are degraded. Even numbered slots are mirrored on
2590 /* gcc -Os complains that this is unused */
2591 int insync
= insync
;
2593 for (i
= 0; i
< map
->num_members
; i
++) {
2594 __u32 ord
= get_imsm_ord_tbl_ent(dev
, i
);
2595 int idx
= ord_to_idx(ord
);
2596 struct imsm_disk
*disk
;
2598 /* reset the potential in-sync count on even-numbered
2599 * slots. num_copies is always 2 for imsm raid10
2604 disk
= get_imsm_disk(super
, idx
);
2605 if (!disk
|| disk
->status
& FAILED_DISK
||
2606 ord
& IMSM_ORD_REBUILD
)
2609 /* no in-sync disks left in this mirror the
2613 return IMSM_T_STATE_FAILED
;
2616 return IMSM_T_STATE_DEGRADED
;
2620 return IMSM_T_STATE_DEGRADED
;
2622 return IMSM_T_STATE_FAILED
;
2628 return map
->map_state
;
2631 static int imsm_count_failed(struct intel_super
*super
, struct imsm_dev
*dev
)
2635 struct imsm_disk
*disk
;
2636 struct imsm_map
*map
= get_imsm_map(dev
, 0);
2638 for (i
= 0; i
< map
->num_members
; i
++) {
2639 __u32 ord
= get_imsm_ord_tbl_ent(dev
, i
);
2640 int idx
= ord_to_idx(ord
);
2642 disk
= get_imsm_disk(super
, idx
);
2643 if (!disk
|| disk
->status
& FAILED_DISK
||
2644 ord
& IMSM_ORD_REBUILD
)
2651 static int is_resyncing(struct imsm_dev
*dev
)
2653 struct imsm_map
*migr_map
;
2655 if (!dev
->vol
.migr_state
)
2658 if (dev
->vol
.migr_type
== MIGR_INIT
)
2661 migr_map
= get_imsm_map(dev
, 1);
2663 if (migr_map
->map_state
== IMSM_T_STATE_NORMAL
)
2669 static int is_rebuilding(struct imsm_dev
*dev
)
2671 struct imsm_map
*migr_map
;
2673 if (!dev
->vol
.migr_state
)
2676 if (dev
->vol
.migr_type
!= MIGR_REBUILD
)
2679 migr_map
= get_imsm_map(dev
, 1);
2681 if (migr_map
->map_state
== IMSM_T_STATE_DEGRADED
)
2687 static void mark_failure(struct imsm_disk
*disk
)
2689 if (disk
->status
& FAILED_DISK
)
2691 disk
->status
|= FAILED_DISK
;
2692 disk
->scsi_id
= __cpu_to_le32(~(__u32
)0);
2693 memmove(&disk
->serial
[0], &disk
->serial
[1], MAX_RAID_SERIAL_LEN
- 1);
2696 /* Handle dirty -> clean transititions and resync. Degraded and rebuild
2697 * states are handled in imsm_set_disk() with one exception, when a
2698 * resync is stopped due to a new failure this routine will set the
2699 * 'degraded' state for the array.
2701 static int imsm_set_array_state(struct active_array
*a
, int consistent
)
2703 int inst
= a
->info
.container_member
;
2704 struct intel_super
*super
= a
->container
->sb
;
2705 struct imsm_dev
*dev
= get_imsm_dev(super
, inst
);
2706 struct imsm_map
*map
= get_imsm_map(dev
, 0);
2707 int failed
= imsm_count_failed(super
, dev
);
2708 __u8 map_state
= imsm_check_degraded(super
, dev
, failed
);
2710 /* before we activate this array handle any missing disks */
2711 if (consistent
== 2 && super
->missing
) {
2714 dprintf("imsm: mark missing\n");
2715 end_migration(dev
, map_state
);
2716 for (dl
= super
->missing
; dl
; dl
= dl
->next
)
2717 mark_failure(&dl
->disk
);
2718 super
->updates_pending
++;
2721 if (consistent
== 2 &&
2722 (!is_resync_complete(a
) ||
2723 map_state
!= IMSM_T_STATE_NORMAL
||
2724 dev
->vol
.migr_state
))
2727 if (is_resync_complete(a
)) {
2728 /* complete intialization / resync,
2729 * recovery is completed in ->set_disk
2731 if (is_resyncing(dev
)) {
2732 dprintf("imsm: mark resync done\n");
2733 end_migration(dev
, map_state
);
2734 super
->updates_pending
++;
2736 } else if (!is_resyncing(dev
) && !failed
) {
2737 /* mark the start of the init process if nothing is failed */
2738 dprintf("imsm: mark resync start (%llu)\n", a
->resync_start
);
2739 if (map
->map_state
== IMSM_T_STATE_NORMAL
)
2740 migrate(dev
, IMSM_T_STATE_NORMAL
, MIGR_REBUILD
);
2742 migrate(dev
, IMSM_T_STATE_NORMAL
, MIGR_INIT
);
2743 super
->updates_pending
++;
2746 /* check if we can update the migration checkpoint */
2747 if (dev
->vol
.migr_state
&&
2748 __le32_to_cpu(dev
->vol
.curr_migr_unit
) != a
->resync_start
) {
2749 dprintf("imsm: checkpoint migration (%llu)\n", a
->resync_start
);
2750 dev
->vol
.curr_migr_unit
= __cpu_to_le32(a
->resync_start
);
2751 super
->updates_pending
++;
2754 /* mark dirty / clean */
2755 if (dev
->vol
.dirty
!= !consistent
) {
2756 dprintf("imsm: mark '%s' (%llu)\n",
2757 consistent
? "clean" : "dirty", a
->resync_start
);
2762 super
->updates_pending
++;
2767 static void imsm_set_disk(struct active_array
*a
, int n
, int state
)
2769 int inst
= a
->info
.container_member
;
2770 struct intel_super
*super
= a
->container
->sb
;
2771 struct imsm_dev
*dev
= get_imsm_dev(super
, inst
);
2772 struct imsm_map
*map
= get_imsm_map(dev
, 0);
2773 struct imsm_disk
*disk
;
2778 if (n
> map
->num_members
)
2779 fprintf(stderr
, "imsm: set_disk %d out of range 0..%d\n",
2780 n
, map
->num_members
- 1);
2785 dprintf("imsm: set_disk %d:%x\n", n
, state
);
2787 ord
= get_imsm_ord_tbl_ent(dev
, n
);
2788 disk
= get_imsm_disk(super
, ord_to_idx(ord
));
2790 /* check for new failures */
2791 if ((state
& DS_FAULTY
) && !(disk
->status
& FAILED_DISK
)) {
2793 super
->updates_pending
++;
2796 /* check if in_sync */
2797 if (state
& DS_INSYNC
&& ord
& IMSM_ORD_REBUILD
) {
2798 struct imsm_map
*migr_map
= get_imsm_map(dev
, 1);
2800 set_imsm_ord_tbl_ent(migr_map
, n
, ord_to_idx(ord
));
2801 super
->updates_pending
++;
2804 failed
= imsm_count_failed(super
, dev
);
2805 map_state
= imsm_check_degraded(super
, dev
, failed
);
2807 /* check if recovery complete, newly degraded, or failed */
2808 if (map_state
== IMSM_T_STATE_NORMAL
&& is_rebuilding(dev
)) {
2809 end_migration(dev
, map_state
);
2810 super
->updates_pending
++;
2811 } else if (map_state
== IMSM_T_STATE_DEGRADED
&&
2812 map
->map_state
!= map_state
&&
2813 !dev
->vol
.migr_state
) {
2814 dprintf("imsm: mark degraded\n");
2815 map
->map_state
= map_state
;
2816 super
->updates_pending
++;
2817 } else if (map_state
== IMSM_T_STATE_FAILED
&&
2818 map
->map_state
!= map_state
) {
2819 dprintf("imsm: mark failed\n");
2820 end_migration(dev
, map_state
);
2821 super
->updates_pending
++;
2825 static int store_imsm_mpb(int fd
, struct intel_super
*super
)
2827 struct imsm_super
*mpb
= super
->anchor
;
2828 __u32 mpb_size
= __le32_to_cpu(mpb
->mpb_size
);
2829 unsigned long long dsize
;
2830 unsigned long long sectors
;
2832 get_dev_size(fd
, NULL
, &dsize
);
2834 if (mpb_size
> 512) {
2835 /* -1 to account for anchor */
2836 sectors
= mpb_sectors(mpb
) - 1;
2838 /* write the extended mpb to the sectors preceeding the anchor */
2839 if (lseek64(fd
, dsize
- (512 * (2 + sectors
)), SEEK_SET
) < 0)
2842 if (write(fd
, super
->buf
+ 512, 512 * sectors
) != 512 * sectors
)
2846 /* first block is stored on second to last sector of the disk */
2847 if (lseek64(fd
, dsize
- (512 * 2), SEEK_SET
) < 0)
2850 if (write(fd
, super
->buf
, 512) != 512)
2856 static void imsm_sync_metadata(struct supertype
*container
)
2858 struct intel_super
*super
= container
->sb
;
2860 if (!super
->updates_pending
)
2863 write_super_imsm(super
, 0);
2865 super
->updates_pending
= 0;
2868 static struct dl
*imsm_readd(struct intel_super
*super
, int idx
, struct active_array
*a
)
2870 struct imsm_dev
*dev
= get_imsm_dev(super
, a
->info
.container_member
);
2871 int i
= get_imsm_disk_idx(dev
, idx
);
2874 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
2878 if (dl
&& dl
->disk
.status
& FAILED_DISK
)
2882 dprintf("%s: found %x:%x\n", __func__
, dl
->major
, dl
->minor
);
2887 static struct dl
*imsm_add_spare(struct intel_super
*super
, int slot
, struct active_array
*a
)
2889 struct imsm_dev
*dev
= get_imsm_dev(super
, a
->info
.container_member
);
2890 int idx
= get_imsm_disk_idx(dev
, slot
);
2891 struct imsm_map
*map
= get_imsm_map(dev
, 0);
2892 unsigned long long esize
;
2893 unsigned long long pos
;
2901 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
2902 /* If in this array, skip */
2903 for (d
= a
->info
.devs
; d
; d
= d
->next
)
2904 if (d
->state_fd
>= 0 &&
2905 d
->disk
.major
== dl
->major
&&
2906 d
->disk
.minor
== dl
->minor
) {
2907 dprintf("%x:%x already in array\n", dl
->major
, dl
->minor
);
2913 /* skip in use or failed drives */
2914 if (dl
->disk
.status
& FAILED_DISK
|| idx
== dl
->index
) {
2915 dprintf("%x:%x status ( %s%s)\n",
2916 dl
->major
, dl
->minor
,
2917 dl
->disk
.status
& FAILED_DISK
? "failed " : "",
2918 idx
== dl
->index
? "in use " : "");
2922 /* Does this unused device have the requisite free space?
2923 * We need a->info.component_size sectors
2925 ex
= get_extents(super
, dl
);
2927 dprintf("cannot get extents\n");
2933 array_start
= __le32_to_cpu(map
->pba_of_lba0
);
2936 /* check that we can start at pba_of_lba0 with
2937 * a->info.component_size of space
2939 esize
= ex
[j
].start
- pos
;
2940 if (array_start
>= pos
&&
2941 array_start
+ a
->info
.component_size
< ex
[j
].start
) {
2945 pos
= ex
[j
].start
+ ex
[j
].size
;
2948 } while (ex
[j
-1].size
);
2952 dprintf("%x:%x does not have %llu at %d\n",
2953 dl
->major
, dl
->minor
,
2954 a
->info
.component_size
,
2955 __le32_to_cpu(map
->pba_of_lba0
));
2965 static struct mdinfo
*imsm_activate_spare(struct active_array
*a
,
2966 struct metadata_update
**updates
)
2969 * Find a device with unused free space and use it to replace a
2970 * failed/vacant region in an array. We replace failed regions one a
2971 * array at a time. The result is that a new spare disk will be added
2972 * to the first failed array and after the monitor has finished
2973 * propagating failures the remainder will be consumed.
2975 * FIXME add a capability for mdmon to request spares from another
2979 struct intel_super
*super
= a
->container
->sb
;
2980 int inst
= a
->info
.container_member
;
2981 struct imsm_dev
*dev
= get_imsm_dev(super
, inst
);
2982 struct imsm_map
*map
= get_imsm_map(dev
, 0);
2983 int failed
= a
->info
.array
.raid_disks
;
2984 struct mdinfo
*rv
= NULL
;
2987 struct metadata_update
*mu
;
2989 struct imsm_update_activate_spare
*u
;
2993 for (d
= a
->info
.devs
; d
; d
= d
->next
) {
2994 if ((d
->curr_state
& DS_FAULTY
) &&
2996 /* wait for Removal to happen */
2998 if (d
->state_fd
>= 0)
3002 dprintf("imsm: activate spare: inst=%d failed=%d (%d) level=%d\n",
3003 inst
, failed
, a
->info
.array
.raid_disks
, a
->info
.array
.level
);
3004 if (imsm_check_degraded(super
, dev
, failed
) != IMSM_T_STATE_DEGRADED
)
3007 /* For each slot, if it is not working, find a spare */
3008 for (i
= 0; i
< a
->info
.array
.raid_disks
; i
++) {
3009 for (d
= a
->info
.devs
; d
; d
= d
->next
)
3010 if (d
->disk
.raid_disk
== i
)
3012 dprintf("found %d: %p %x\n", i
, d
, d
?d
->curr_state
:0);
3013 if (d
&& (d
->state_fd
>= 0))
3017 * OK, this device needs recovery. Try to re-add the previous
3018 * occupant of this slot, if this fails add a new spare
3020 dl
= imsm_readd(super
, i
, a
);
3022 dl
= imsm_add_spare(super
, i
, a
);
3026 /* found a usable disk with enough space */
3027 di
= malloc(sizeof(*di
));
3030 memset(di
, 0, sizeof(*di
));
3032 /* dl->index will be -1 in the case we are activating a
3033 * pristine spare. imsm_process_update() will create a
3034 * new index in this case. Once a disk is found to be
3035 * failed in all member arrays it is kicked from the
3038 di
->disk
.number
= dl
->index
;
3040 /* (ab)use di->devs to store a pointer to the device
3043 di
->devs
= (struct mdinfo
*) dl
;
3045 di
->disk
.raid_disk
= i
;
3046 di
->disk
.major
= dl
->major
;
3047 di
->disk
.minor
= dl
->minor
;
3049 di
->data_offset
= __le32_to_cpu(map
->pba_of_lba0
);
3050 di
->component_size
= a
->info
.component_size
;
3051 di
->container_member
= inst
;
3055 dprintf("%x:%x to be %d at %llu\n", dl
->major
, dl
->minor
,
3056 i
, di
->data_offset
);
3062 /* No spares found */
3064 /* Now 'rv' has a list of devices to return.
3065 * Create a metadata_update record to update the
3066 * disk_ord_tbl for the array
3068 mu
= malloc(sizeof(*mu
));
3070 mu
->buf
= malloc(sizeof(struct imsm_update_activate_spare
) * num_spares
);
3071 if (mu
->buf
== NULL
) {
3078 struct mdinfo
*n
= rv
->next
;
3087 mu
->len
= sizeof(struct imsm_update_activate_spare
) * num_spares
;
3088 mu
->next
= *updates
;
3089 u
= (struct imsm_update_activate_spare
*) mu
->buf
;
3091 for (di
= rv
; di
; di
= di
->next
) {
3092 u
->type
= update_activate_spare
;
3093 u
->dl
= (struct dl
*) di
->devs
;
3095 u
->slot
= di
->disk
.raid_disk
;
3106 static int disks_overlap(struct imsm_dev
*d1
, struct imsm_dev
*d2
)
3108 struct imsm_map
*m1
= get_imsm_map(d1
, 0);
3109 struct imsm_map
*m2
= get_imsm_map(d2
, 0);
3114 for (i
= 0; i
< m1
->num_members
; i
++) {
3115 idx
= get_imsm_disk_idx(d1
, i
);
3116 for (j
= 0; j
< m2
->num_members
; j
++)
3117 if (idx
== get_imsm_disk_idx(d2
, j
))
3124 static void imsm_delete(struct intel_super
*super
, struct dl
**dlp
, int index
);
3126 static void imsm_process_update(struct supertype
*st
,
3127 struct metadata_update
*update
)
3130 * crack open the metadata_update envelope to find the update record
3131 * update can be one of:
3132 * update_activate_spare - a spare device has replaced a failed
3133 * device in an array, update the disk_ord_tbl. If this disk is
3134 * present in all member arrays then also clear the SPARE_DISK
3137 struct intel_super
*super
= st
->sb
;
3138 struct imsm_super
*mpb
;
3139 enum imsm_update_type type
= *(enum imsm_update_type
*) update
->buf
;
3141 /* update requires a larger buf but the allocation failed */
3142 if (super
->next_len
&& !super
->next_buf
) {
3143 super
->next_len
= 0;
3147 if (super
->next_buf
) {
3148 memcpy(super
->next_buf
, super
->buf
, super
->len
);
3150 super
->len
= super
->next_len
;
3151 super
->buf
= super
->next_buf
;
3153 super
->next_len
= 0;
3154 super
->next_buf
= NULL
;
3157 mpb
= super
->anchor
;
3160 case update_activate_spare
: {
3161 struct imsm_update_activate_spare
*u
= (void *) update
->buf
;
3162 struct imsm_dev
*dev
= get_imsm_dev(super
, u
->array
);
3163 struct imsm_map
*map
= get_imsm_map(dev
, 0);
3164 struct imsm_map
*migr_map
;
3165 struct active_array
*a
;
3166 struct imsm_disk
*disk
;
3171 int victim
= get_imsm_disk_idx(dev
, u
->slot
);
3174 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
3179 fprintf(stderr
, "error: imsm_activate_spare passed "
3180 "an unknown disk (index: %d)\n",
3185 super
->updates_pending
++;
3187 /* count failures (excluding rebuilds and the victim)
3188 * to determine map[0] state
3191 for (i
= 0; i
< map
->num_members
; i
++) {
3194 disk
= get_imsm_disk(super
, get_imsm_disk_idx(dev
, i
));
3195 if (!disk
|| disk
->status
& FAILED_DISK
)
3199 /* adding a pristine spare, assign a new index */
3200 if (dl
->index
< 0) {
3201 dl
->index
= super
->anchor
->num_disks
;
3202 super
->anchor
->num_disks
++;
3205 disk
->status
|= CONFIGURED_DISK
;
3206 disk
->status
&= ~SPARE_DISK
;
3209 to_state
= imsm_check_degraded(super
, dev
, failed
);
3210 map
->map_state
= IMSM_T_STATE_DEGRADED
;
3211 migrate(dev
, to_state
, MIGR_REBUILD
);
3212 migr_map
= get_imsm_map(dev
, 1);
3213 set_imsm_ord_tbl_ent(map
, u
->slot
, dl
->index
);
3214 set_imsm_ord_tbl_ent(migr_map
, u
->slot
, dl
->index
| IMSM_ORD_REBUILD
);
3216 /* count arrays using the victim in the metadata */
3218 for (a
= st
->arrays
; a
; a
= a
->next
) {
3219 dev
= get_imsm_dev(super
, a
->info
.container_member
);
3220 for (i
= 0; i
< map
->num_members
; i
++)
3221 if (victim
== get_imsm_disk_idx(dev
, i
))
3225 /* delete the victim if it is no longer being
3231 /* We know that 'manager' isn't touching anything,
3232 * so it is safe to delete
3234 for (dlp
= &super
->disks
; *dlp
; dlp
= &(*dlp
)->next
)
3235 if ((*dlp
)->index
== victim
)
3238 /* victim may be on the missing list */
3240 for (dlp
= &super
->missing
; *dlp
; dlp
= &(*dlp
)->next
)
3241 if ((*dlp
)->index
== victim
)
3243 imsm_delete(super
, dlp
, victim
);
3247 case update_create_array
: {
3248 /* someone wants to create a new array, we need to be aware of
3249 * a few races/collisions:
3250 * 1/ 'Create' called by two separate instances of mdadm
3251 * 2/ 'Create' versus 'activate_spare': mdadm has chosen
3252 * devices that have since been assimilated via
3254 * In the event this update can not be carried out mdadm will
3255 * (FIX ME) notice that its update did not take hold.
3257 struct imsm_update_create_array
*u
= (void *) update
->buf
;
3258 struct imsm_dev
*dev
;
3259 struct imsm_map
*map
, *new_map
;
3260 unsigned long long start
, end
;
3261 unsigned long long new_start
, new_end
;
3265 /* handle racing creates: first come first serve */
3266 if (u
->dev_idx
< mpb
->num_raid_devs
) {
3267 dprintf("%s: subarray %d already defined\n",
3268 __func__
, u
->dev_idx
);
3272 /* check update is next in sequence */
3273 if (u
->dev_idx
!= mpb
->num_raid_devs
) {
3274 dprintf("%s: can not create array %d expected index %d\n",
3275 __func__
, u
->dev_idx
, mpb
->num_raid_devs
);
3279 new_map
= get_imsm_map(&u
->dev
, 0);
3280 new_start
= __le32_to_cpu(new_map
->pba_of_lba0
);
3281 new_end
= new_start
+ __le32_to_cpu(new_map
->blocks_per_member
);
3283 /* handle activate_spare versus create race:
3284 * check to make sure that overlapping arrays do not include
3287 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
3288 dev
= get_imsm_dev(super
, i
);
3289 map
= get_imsm_map(dev
, 0);
3290 start
= __le32_to_cpu(map
->pba_of_lba0
);
3291 end
= start
+ __le32_to_cpu(map
->blocks_per_member
);
3292 if ((new_start
>= start
&& new_start
<= end
) ||
3293 (start
>= new_start
&& start
<= new_end
))
3295 if (overlap
&& disks_overlap(dev
, &u
->dev
)) {
3296 dprintf("%s: arrays overlap\n", __func__
);
3300 /* check num_members sanity */
3301 if (new_map
->num_members
> mpb
->num_disks
) {
3302 dprintf("%s: num_disks out of range\n", __func__
);
3306 /* check that prepare update was successful */
3307 if (!update
->space
) {
3308 dprintf("%s: prepare update failed\n", __func__
);
3312 super
->updates_pending
++;
3313 dev
= update
->space
;
3314 map
= get_imsm_map(dev
, 0);
3315 update
->space
= NULL
;
3316 imsm_copy_dev(dev
, &u
->dev
);
3317 map
= get_imsm_map(dev
, 0);
3318 super
->dev_tbl
[u
->dev_idx
] = dev
;
3319 mpb
->num_raid_devs
++;
3322 for (i
= 0; i
< map
->num_members
; i
++) {
3323 struct imsm_disk
*disk
;
3325 disk
= get_imsm_disk(super
, get_imsm_disk_idx(dev
, i
));
3326 disk
->status
|= CONFIGURED_DISK
;
3327 disk
->status
&= ~SPARE_DISK
;
3330 imsm_update_version_info(super
);
3334 case update_add_disk
:
3336 /* we may be able to repair some arrays if disks are
3339 struct active_array
*a
;
3341 super
->updates_pending
++;
3342 for (a
= st
->arrays
; a
; a
= a
->next
)
3343 a
->check_degraded
= 1;
3345 /* add some spares to the metadata */
3346 while (super
->add
) {
3350 super
->add
= al
->next
;
3351 al
->next
= super
->disks
;
3353 dprintf("%s: added %x:%x\n",
3354 __func__
, al
->major
, al
->minor
);
3361 static void imsm_prepare_update(struct supertype
*st
,
3362 struct metadata_update
*update
)
3365 * Allocate space to hold new disk entries, raid-device entries or a new
3366 * mpb if necessary. The manager synchronously waits for updates to
3367 * complete in the monitor, so new mpb buffers allocated here can be
3368 * integrated by the monitor thread without worrying about live pointers
3369 * in the manager thread.
3371 enum imsm_update_type type
= *(enum imsm_update_type
*) update
->buf
;
3372 struct intel_super
*super
= st
->sb
;
3373 struct imsm_super
*mpb
= super
->anchor
;
3378 case update_create_array
: {
3379 struct imsm_update_create_array
*u
= (void *) update
->buf
;
3381 len
= sizeof_imsm_dev(&u
->dev
, 1);
3382 update
->space
= malloc(len
);
3389 /* check if we need a larger metadata buffer */
3390 if (super
->next_buf
)
3391 buf_len
= super
->next_len
;
3393 buf_len
= super
->len
;
3395 if (__le32_to_cpu(mpb
->mpb_size
) + len
> buf_len
) {
3396 /* ok we need a larger buf than what is currently allocated
3397 * if this allocation fails process_update will notice that
3398 * ->next_len is set and ->next_buf is NULL
3400 buf_len
= ROUND_UP(__le32_to_cpu(mpb
->mpb_size
) + len
, 512);
3401 if (super
->next_buf
)
3402 free(super
->next_buf
);
3404 super
->next_len
= buf_len
;
3405 if (posix_memalign(&super
->next_buf
, buf_len
, 512) != 0)
3406 super
->next_buf
= NULL
;
3410 /* must be called while manager is quiesced */
3411 static void imsm_delete(struct intel_super
*super
, struct dl
**dlp
, int index
)
3413 struct imsm_super
*mpb
= super
->anchor
;
3415 struct imsm_dev
*dev
;
3416 struct imsm_map
*map
;
3417 int i
, j
, num_members
;
3420 dprintf("%s: deleting device[%d] from imsm_super\n",
3423 /* shift all indexes down one */
3424 for (iter
= super
->disks
; iter
; iter
= iter
->next
)
3425 if (iter
->index
> index
)
3427 for (iter
= super
->missing
; iter
; iter
= iter
->next
)
3428 if (iter
->index
> index
)
3431 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
3432 dev
= get_imsm_dev(super
, i
);
3433 map
= get_imsm_map(dev
, 0);
3434 num_members
= map
->num_members
;
3435 for (j
= 0; j
< num_members
; j
++) {
3436 /* update ord entries being careful not to propagate
3437 * ord-flags to the first map
3439 ord
= get_imsm_ord_tbl_ent(dev
, j
);
3441 if (ord_to_idx(ord
) <= index
)
3444 map
= get_imsm_map(dev
, 0);
3445 set_imsm_ord_tbl_ent(map
, j
, ord_to_idx(ord
- 1));
3446 map
= get_imsm_map(dev
, 1);
3448 set_imsm_ord_tbl_ent(map
, j
, ord
- 1);
3453 super
->updates_pending
++;
3455 struct dl
*dl
= *dlp
;
3457 *dlp
= (*dlp
)->next
;
3458 __free_imsm_disk(dl
);
3461 #endif /* MDASSEMBLE */
3463 struct superswitch super_imsm
= {
3465 .examine_super
= examine_super_imsm
,
3466 .brief_examine_super
= brief_examine_super_imsm
,
3467 .detail_super
= detail_super_imsm
,
3468 .brief_detail_super
= brief_detail_super_imsm
,
3469 .write_init_super
= write_init_super_imsm
,
3470 .validate_geometry
= validate_geometry_imsm
,
3471 .add_to_super
= add_to_super_imsm
,
3473 .match_home
= match_home_imsm
,
3474 .uuid_from_super
= uuid_from_super_imsm
,
3475 .getinfo_super
= getinfo_super_imsm
,
3476 .update_super
= update_super_imsm
,
3478 .avail_size
= avail_size_imsm
,
3480 .compare_super
= compare_super_imsm
,
3482 .load_super
= load_super_imsm
,
3483 .init_super
= init_super_imsm
,
3484 .store_super
= store_zero_imsm
,
3485 .free_super
= free_super_imsm
,
3486 .match_metadata_desc
= match_metadata_desc_imsm
,
3487 .container_content
= container_content_imsm
,
3493 .open_new
= imsm_open_new
,
3494 .load_super
= load_super_imsm
,
3495 .set_array_state
= imsm_set_array_state
,
3496 .set_disk
= imsm_set_disk
,
3497 .sync_metadata
= imsm_sync_metadata
,
3498 .activate_spare
= imsm_activate_spare
,
3499 .process_update
= imsm_process_update
,
3500 .prepare_update
= imsm_prepare_update
,
3501 #endif /* MDASSEMBLE */