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 /* round down to 1k block to satisfy precision of the kernel
496 /* make sure remainder is still sane */
497 if (remainder
< ROUND_UP(super
->len
, 512) >> 9)
498 remainder
= ROUND_UP(super
->len
, 512) >> 9;
499 if (reservation
> remainder
)
500 reservation
= remainder
;
502 e
->start
= __le32_to_cpu(dl
->disk
.total_blocks
) - reservation
;
507 /* try to determine how much space is reserved for metadata from
508 * the last get_extents() entry, otherwise fallback to the
511 static __u32
imsm_reserved_sectors(struct intel_super
*super
, struct dl
*dl
)
517 /* for spares just return a minimal reservation which will grow
518 * once the spare is picked up by an array
521 return MPB_SECTOR_CNT
;
523 e
= get_extents(super
, dl
);
525 return MPB_SECTOR_CNT
+ IMSM_RESERVED_SECTORS
;
527 /* scroll to last entry */
528 for (i
= 0; e
[i
].size
; i
++)
531 rv
= __le32_to_cpu(dl
->disk
.total_blocks
) - e
[i
].start
;
539 static void print_imsm_dev(struct imsm_dev
*dev
, char *uuid
, int disk_idx
)
543 struct imsm_map
*map
= get_imsm_map(dev
, 0);
547 printf("[%.16s]:\n", dev
->volume
);
548 printf(" UUID : %s\n", uuid
);
549 printf(" RAID Level : %d\n", get_imsm_raid_level(map
));
550 printf(" Members : %d\n", map
->num_members
);
551 for (slot
= 0; slot
< map
->num_members
; slot
++)
552 if (disk_idx
== get_imsm_disk_idx(dev
, slot
))
554 if (slot
< map
->num_members
) {
555 ord
= get_imsm_ord_tbl_ent(dev
, slot
);
556 printf(" This Slot : %d%s\n", slot
,
557 ord
& IMSM_ORD_REBUILD
? " (out-of-sync)" : "");
559 printf(" This Slot : ?\n");
560 sz
= __le32_to_cpu(dev
->size_high
);
562 sz
+= __le32_to_cpu(dev
->size_low
);
563 printf(" Array Size : %llu%s\n", (unsigned long long)sz
,
564 human_size(sz
* 512));
565 sz
= __le32_to_cpu(map
->blocks_per_member
);
566 printf(" Per Dev Size : %llu%s\n", (unsigned long long)sz
,
567 human_size(sz
* 512));
568 printf(" Sector Offset : %u\n",
569 __le32_to_cpu(map
->pba_of_lba0
));
570 printf(" Num Stripes : %u\n",
571 __le32_to_cpu(map
->num_data_stripes
));
572 printf(" Chunk Size : %u KiB\n",
573 __le16_to_cpu(map
->blocks_per_strip
) / 2);
574 printf(" Reserved : %d\n", __le32_to_cpu(dev
->reserved_blocks
));
575 printf(" Migrate State : %s", dev
->vol
.migr_state
? "migrating" : "idle");
576 if (dev
->vol
.migr_state
)
577 printf(": %s", dev
->vol
.migr_type
? "rebuilding" : "initializing");
579 printf(" Map State : %s", map_state_str
[map
->map_state
]);
580 if (dev
->vol
.migr_state
) {
581 struct imsm_map
*map
= get_imsm_map(dev
, 1);
582 printf(" <-- %s", map_state_str
[map
->map_state
]);
585 printf(" Dirty State : %s\n", dev
->vol
.dirty
? "dirty" : "clean");
588 static void print_imsm_disk(struct imsm_super
*mpb
, int index
, __u32 reserved
)
590 struct imsm_disk
*disk
= __get_imsm_disk(mpb
, index
);
591 char str
[MAX_RAID_SERIAL_LEN
+ 1];
599 snprintf(str
, MAX_RAID_SERIAL_LEN
+ 1, "%s", disk
->serial
);
600 printf(" Disk%02d Serial : %s\n", index
, str
);
602 printf(" State :%s%s%s%s\n", s
&SPARE_DISK
? " spare" : "",
603 s
&CONFIGURED_DISK
? " active" : "",
604 s
&FAILED_DISK
? " failed" : "",
605 s
&USABLE_DISK
? " usable" : "");
606 printf(" Id : %08x\n", __le32_to_cpu(disk
->scsi_id
));
607 sz
= __le32_to_cpu(disk
->total_blocks
) - reserved
;
608 printf(" Usable Size : %llu%s\n", (unsigned long long)sz
,
609 human_size(sz
* 512));
612 static void getinfo_super_imsm(struct supertype
*st
, struct mdinfo
*info
);
614 static void examine_super_imsm(struct supertype
*st
, char *homehost
)
616 struct intel_super
*super
= st
->sb
;
617 struct imsm_super
*mpb
= super
->anchor
;
618 char str
[MAX_SIGNATURE_LENGTH
];
623 __u32 reserved
= imsm_reserved_sectors(super
, super
->disks
);
626 snprintf(str
, MPB_SIG_LEN
, "%s", mpb
->sig
);
627 printf(" Magic : %s\n", str
);
628 snprintf(str
, strlen(MPB_VERSION_RAID0
), "%s", get_imsm_version(mpb
));
629 printf(" Version : %s\n", get_imsm_version(mpb
));
630 printf(" Family : %08x\n", __le32_to_cpu(mpb
->family_num
));
631 printf(" Generation : %08x\n", __le32_to_cpu(mpb
->generation_num
));
632 getinfo_super_imsm(st
, &info
);
633 fname_from_uuid(st
, &info
, nbuf
,'-');
634 printf(" UUID : %s\n", nbuf
+ 5);
635 sum
= __le32_to_cpu(mpb
->check_sum
);
636 printf(" Checksum : %08x %s\n", sum
,
637 __gen_imsm_checksum(mpb
) == sum
? "correct" : "incorrect");
638 printf(" MPB Sectors : %d\n", mpb_sectors(mpb
));
639 printf(" Disks : %d\n", mpb
->num_disks
);
640 printf(" RAID Devices : %d\n", mpb
->num_raid_devs
);
641 print_imsm_disk(mpb
, super
->disks
->index
, reserved
);
642 if (super
->bbm_log
) {
643 struct bbm_log
*log
= super
->bbm_log
;
646 printf("Bad Block Management Log:\n");
647 printf(" Log Size : %d\n", __le32_to_cpu(mpb
->bbm_log_size
));
648 printf(" Signature : %x\n", __le32_to_cpu(log
->signature
));
649 printf(" Entry Count : %d\n", __le32_to_cpu(log
->entry_count
));
650 printf(" Spare Blocks : %d\n", __le32_to_cpu(log
->reserved_spare_block_count
));
651 printf(" First Spare : %llx\n", __le64_to_cpu(log
->first_spare_lba
));
653 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
655 struct imsm_dev
*dev
= __get_imsm_dev(mpb
, i
);
657 super
->current_vol
= i
;
658 getinfo_super_imsm(st
, &info
);
659 fname_from_uuid(st
, &info
, nbuf
, '-');
660 print_imsm_dev(dev
, nbuf
+ 5, super
->disks
->index
);
662 for (i
= 0; i
< mpb
->num_disks
; i
++) {
663 if (i
== super
->disks
->index
)
665 print_imsm_disk(mpb
, i
, reserved
);
669 static void brief_examine_super_imsm(struct supertype
*st
)
671 /* We just write a generic IMSM ARRAY entry */
675 struct intel_super
*super
= st
->sb
;
678 if (!super
->anchor
->num_raid_devs
)
681 getinfo_super_imsm(st
, &info
);
682 fname_from_uuid(st
, &info
, nbuf
,'-');
683 printf("ARRAY metadata=imsm auto=md UUID=%s\n", nbuf
+ 5);
684 for (i
= 0; i
< super
->anchor
->num_raid_devs
; i
++) {
685 struct imsm_dev
*dev
= get_imsm_dev(super
, i
);
687 super
->current_vol
= i
;
688 getinfo_super_imsm(st
, &info
);
689 fname_from_uuid(st
, &info
, nbuf1
,'-');
690 printf("ARRAY /dev/md/%.16s container=%s\n"
691 " member=%d auto=mdp UUID=%s\n",
692 dev
->volume
, nbuf
+ 5, i
, nbuf1
+ 5);
696 static void detail_super_imsm(struct supertype
*st
, char *homehost
)
701 getinfo_super_imsm(st
, &info
);
702 fname_from_uuid(st
, &info
, nbuf
,'-');
703 printf("\n UUID : %s\n", nbuf
+ 5);
706 static void brief_detail_super_imsm(struct supertype
*st
)
710 getinfo_super_imsm(st
, &info
);
711 fname_from_uuid(st
, &info
, nbuf
,'-');
712 printf(" UUID=%s", nbuf
+ 5);
716 static int match_home_imsm(struct supertype
*st
, char *homehost
)
718 /* the imsm metadata format does not specify any host
719 * identification information. We return -1 since we can never
720 * confirm nor deny whether a given array is "meant" for this
721 * host. We rely on compare_super and the 'family_num' field to
722 * exclude member disks that do not belong, and we rely on
723 * mdadm.conf to specify the arrays that should be assembled.
724 * Auto-assembly may still pick up "foreign" arrays.
730 static void uuid_from_super_imsm(struct supertype
*st
, int uuid
[4])
732 /* The uuid returned here is used for:
733 * uuid to put into bitmap file (Create, Grow)
734 * uuid for backup header when saving critical section (Grow)
735 * comparing uuids when re-adding a device into an array
736 * In these cases the uuid required is that of the data-array,
737 * not the device-set.
738 * uuid to recognise same set when adding a missing device back
739 * to an array. This is a uuid for the device-set.
741 * For each of these we can make do with a truncated
742 * or hashed uuid rather than the original, as long as
744 * In each case the uuid required is that of the data-array,
745 * not the device-set.
747 /* imsm does not track uuid's so we synthesis one using sha1 on
748 * - The signature (Which is constant for all imsm array, but no matter)
749 * - the family_num of the container
750 * - the index number of the volume
751 * - the 'serial' number of the volume.
752 * Hopefully these are all constant.
754 struct intel_super
*super
= st
->sb
;
758 struct imsm_dev
*dev
= NULL
;
761 sha1_process_bytes(super
->anchor
->sig
, MPB_SIG_LEN
, &ctx
);
762 sha1_process_bytes(&super
->anchor
->family_num
, sizeof(__u32
), &ctx
);
763 if (super
->current_vol
>= 0)
764 dev
= get_imsm_dev(super
, super
->current_vol
);
766 __u32 vol
= super
->current_vol
;
767 sha1_process_bytes(&vol
, sizeof(vol
), &ctx
);
768 sha1_process_bytes(dev
->volume
, MAX_RAID_SERIAL_LEN
, &ctx
);
770 sha1_finish_ctx(&ctx
, buf
);
771 memcpy(uuid
, buf
, 4*4);
776 get_imsm_numerical_version(struct imsm_super
*mpb
, int *m
, int *p
)
778 __u8
*v
= get_imsm_version(mpb
);
779 __u8
*end
= mpb
->sig
+ MAX_SIGNATURE_LENGTH
;
780 char major
[] = { 0, 0, 0 };
781 char minor
[] = { 0 ,0, 0 };
782 char patch
[] = { 0, 0, 0 };
783 char *ver_parse
[] = { major
, minor
, patch
};
787 while (*v
!= '\0' && v
< end
) {
788 if (*v
!= '.' && j
< 2)
789 ver_parse
[i
][j
++] = *v
;
797 *m
= strtol(minor
, NULL
, 0);
798 *p
= strtol(patch
, NULL
, 0);
802 static int imsm_level_to_layout(int level
)
810 return ALGORITHM_LEFT_ASYMMETRIC
;
817 static void getinfo_super_imsm_volume(struct supertype
*st
, struct mdinfo
*info
)
819 struct intel_super
*super
= st
->sb
;
820 struct imsm_dev
*dev
= get_imsm_dev(super
, super
->current_vol
);
821 struct imsm_map
*map
= get_imsm_map(dev
, 0);
823 info
->container_member
= super
->current_vol
;
824 info
->array
.raid_disks
= map
->num_members
;
825 info
->array
.level
= get_imsm_raid_level(map
);
826 info
->array
.layout
= imsm_level_to_layout(info
->array
.level
);
827 info
->array
.md_minor
= -1;
828 info
->array
.ctime
= 0;
829 info
->array
.utime
= 0;
830 info
->array
.chunk_size
= __le16_to_cpu(map
->blocks_per_strip
) << 9;
831 info
->array
.state
= !dev
->vol
.dirty
;
833 info
->disk
.major
= 0;
834 info
->disk
.minor
= 0;
836 info
->data_offset
= __le32_to_cpu(map
->pba_of_lba0
);
837 info
->component_size
= __le32_to_cpu(map
->blocks_per_member
);
838 memset(info
->uuid
, 0, sizeof(info
->uuid
));
840 if (map
->map_state
== IMSM_T_STATE_UNINITIALIZED
|| dev
->vol
.dirty
)
841 info
->resync_start
= 0;
842 else if (dev
->vol
.migr_state
)
843 info
->resync_start
= __le32_to_cpu(dev
->vol
.curr_migr_unit
);
845 info
->resync_start
= ~0ULL;
847 strncpy(info
->name
, (char *) dev
->volume
, MAX_RAID_SERIAL_LEN
);
848 info
->name
[MAX_RAID_SERIAL_LEN
] = 0;
850 info
->array
.major_version
= -1;
851 info
->array
.minor_version
= -2;
852 sprintf(info
->text_version
, "/%s/%d",
853 devnum2devname(st
->container_dev
),
854 info
->container_member
);
855 info
->safe_mode_delay
= 4000; /* 4 secs like the Matrix driver */
856 uuid_from_super_imsm(st
, info
->uuid
);
860 static void getinfo_super_imsm(struct supertype
*st
, struct mdinfo
*info
)
862 struct intel_super
*super
= st
->sb
;
863 struct imsm_disk
*disk
;
866 if (super
->current_vol
>= 0) {
867 getinfo_super_imsm_volume(st
, info
);
871 /* Set raid_disks to zero so that Assemble will always pull in valid
874 info
->array
.raid_disks
= 0;
875 info
->array
.level
= LEVEL_CONTAINER
;
876 info
->array
.layout
= 0;
877 info
->array
.md_minor
= -1;
878 info
->array
.ctime
= 0; /* N/A for imsm */
879 info
->array
.utime
= 0;
880 info
->array
.chunk_size
= 0;
882 info
->disk
.major
= 0;
883 info
->disk
.minor
= 0;
884 info
->disk
.raid_disk
= -1;
885 info
->reshape_active
= 0;
886 info
->array
.major_version
= -1;
887 info
->array
.minor_version
= -2;
888 strcpy(info
->text_version
, "imsm");
889 info
->safe_mode_delay
= 0;
890 info
->disk
.number
= -1;
891 info
->disk
.state
= 0;
895 __u32 reserved
= imsm_reserved_sectors(super
, super
->disks
);
897 disk
= &super
->disks
->disk
;
898 info
->data_offset
= __le32_to_cpu(disk
->total_blocks
) - reserved
;
899 info
->component_size
= reserved
;
901 info
->disk
.state
= s
& CONFIGURED_DISK
? (1 << MD_DISK_ACTIVE
) : 0;
902 info
->disk
.state
|= s
& FAILED_DISK
? (1 << MD_DISK_FAULTY
) : 0;
903 info
->disk
.state
|= s
& SPARE_DISK
? 0 : (1 << MD_DISK_SYNC
);
906 /* only call uuid_from_super_imsm when this disk is part of a populated container,
907 * ->compare_super may have updated the 'num_raid_devs' field for spares
909 if (info
->disk
.state
& (1 << MD_DISK_SYNC
) || super
->anchor
->num_raid_devs
)
910 uuid_from_super_imsm(st
, info
->uuid
);
912 memcpy(info
->uuid
, uuid_match_any
, sizeof(int[4]));
915 static int update_super_imsm(struct supertype
*st
, struct mdinfo
*info
,
916 char *update
, char *devname
, int verbose
,
917 int uuid_set
, char *homehost
)
921 /* For 'assemble' and 'force' we need to return non-zero if any
922 * change was made. For others, the return value is ignored.
923 * Update options are:
924 * force-one : This device looks a bit old but needs to be included,
925 * update age info appropriately.
926 * assemble: clear any 'faulty' flag to allow this device to
928 * force-array: Array is degraded but being forced, mark it clean
929 * if that will be needed to assemble it.
931 * newdev: not used ????
932 * grow: Array has gained a new device - this is currently for
934 * resync: mark as dirty so a resync will happen.
935 * name: update the name - preserving the homehost
937 * Following are not relevant for this imsm:
938 * sparc2.2 : update from old dodgey metadata
939 * super-minor: change the preferred_minor number
940 * summaries: update redundant counters.
941 * uuid: Change the uuid of the array to match watch is given
942 * homehost: update the recorded homehost
943 * _reshape_progress: record new reshape_progress position.
946 //struct intel_super *super = st->sb;
947 //struct imsm_super *mpb = super->mpb;
949 if (strcmp(update
, "grow") == 0) {
951 if (strcmp(update
, "resync") == 0) {
952 /* dev->vol.dirty = 1; */
955 /* IMSM has no concept of UUID or homehost */
960 static size_t disks_to_mpb_size(int disks
)
964 size
= sizeof(struct imsm_super
);
965 size
+= (disks
- 1) * sizeof(struct imsm_disk
);
966 size
+= 2 * sizeof(struct imsm_dev
);
967 /* up to 2 maps per raid device (-2 for imsm_maps in imsm_dev */
968 size
+= (4 - 2) * sizeof(struct imsm_map
);
969 /* 4 possible disk_ord_tbl's */
970 size
+= 4 * (disks
- 1) * sizeof(__u32
);
975 static __u64
avail_size_imsm(struct supertype
*st
, __u64 devsize
)
977 if (devsize
< (MPB_SECTOR_CNT
+ IMSM_RESERVED_SECTORS
))
980 return devsize
- (MPB_SECTOR_CNT
+ IMSM_RESERVED_SECTORS
);
983 static int compare_super_imsm(struct supertype
*st
, struct supertype
*tst
)
987 * 0 same, or first was empty, and second was copied
988 * 1 second had wrong number
992 struct intel_super
*first
= st
->sb
;
993 struct intel_super
*sec
= tst
->sb
;
1001 if (memcmp(first
->anchor
->sig
, sec
->anchor
->sig
, MAX_SIGNATURE_LENGTH
) != 0)
1004 /* if an anchor does not have num_raid_devs set then it is a free
1007 if (first
->anchor
->num_raid_devs
> 0 &&
1008 sec
->anchor
->num_raid_devs
> 0) {
1009 if (first
->anchor
->family_num
!= sec
->anchor
->family_num
)
1013 /* if 'first' is a spare promote it to a populated mpb with sec's
1016 if (first
->anchor
->num_raid_devs
== 0 &&
1017 sec
->anchor
->num_raid_devs
> 0) {
1020 /* we need to copy raid device info from sec if an allocation
1021 * fails here we don't associate the spare
1023 for (i
= 0; i
< sec
->anchor
->num_raid_devs
; i
++) {
1024 first
->dev_tbl
[i
] = malloc(sizeof(struct imsm_dev
));
1025 if (!first
->dev_tbl
) {
1027 free(first
->dev_tbl
[i
]);
1028 first
->dev_tbl
[i
] = NULL
;
1030 fprintf(stderr
, "imsm: failed to associate spare\n");
1033 *first
->dev_tbl
[i
] = *sec
->dev_tbl
[i
];
1036 first
->anchor
->num_raid_devs
= sec
->anchor
->num_raid_devs
;
1037 first
->anchor
->family_num
= sec
->anchor
->family_num
;
1043 static void fd2devname(int fd
, char *name
)
1052 if (fstat(fd
, &st
) != 0)
1054 sprintf(path
, "/sys/dev/block/%d:%d",
1055 major(st
.st_rdev
), minor(st
.st_rdev
));
1057 rv
= readlink(path
, dname
, sizeof(dname
));
1062 nm
= strrchr(dname
, '/');
1064 snprintf(name
, MAX_RAID_SERIAL_LEN
, "/dev/%s", nm
);
1068 extern int scsi_get_serial(int fd
, void *buf
, size_t buf_len
);
1070 static int imsm_read_serial(int fd
, char *devname
,
1071 __u8 serial
[MAX_RAID_SERIAL_LEN
])
1073 unsigned char scsi_serial
[255];
1079 memset(scsi_serial
, 0, sizeof(scsi_serial
));
1081 rv
= scsi_get_serial(fd
, scsi_serial
, sizeof(scsi_serial
));
1083 if (rv
&& check_env("IMSM_DEVNAME_AS_SERIAL")) {
1084 memset(serial
, 0, MAX_RAID_SERIAL_LEN
);
1085 fd2devname(fd
, (char *) serial
);
1092 Name
": Failed to retrieve serial for %s\n",
1097 /* trim leading whitespace */
1098 rsp_len
= scsi_serial
[3];
1099 rsp_buf
= (char *) &scsi_serial
[4];
1104 /* truncate len to the end of rsp_buf if necessary */
1105 if (c
+ MAX_RAID_SERIAL_LEN
> rsp_buf
+ rsp_len
)
1106 len
= rsp_len
- (c
- rsp_buf
);
1108 len
= MAX_RAID_SERIAL_LEN
;
1110 /* initialize the buffer and copy rsp_buf characters */
1111 memset(serial
, 0, MAX_RAID_SERIAL_LEN
);
1112 memcpy(serial
, c
, len
);
1114 /* trim trailing whitespace starting with the last character copied */
1115 c
= (char *) &serial
[len
- 1];
1116 while (isspace(*c
) || *c
== '\0')
1122 static int serialcmp(__u8
*s1
, __u8
*s2
)
1124 return strncmp((char *) s1
, (char *) s2
, MAX_RAID_SERIAL_LEN
);
1127 static void serialcpy(__u8
*dest
, __u8
*src
)
1129 strncpy((char *) dest
, (char *) src
, MAX_RAID_SERIAL_LEN
);
1133 load_imsm_disk(int fd
, struct intel_super
*super
, char *devname
, int keep_fd
)
1140 __u8 serial
[MAX_RAID_SERIAL_LEN
];
1142 rv
= imsm_read_serial(fd
, devname
, serial
);
1147 /* check if this is a disk we have seen before. it may be a spare in
1148 * super->disks while the current anchor believes it is a raid member,
1149 * check if we need to update dl->index
1151 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
1152 if (serialcmp(dl
->serial
, serial
) == 0)
1156 dl
= malloc(sizeof(*dl
));
1163 Name
": failed to allocate disk buffer for %s\n",
1170 dl
->major
= major(stb
.st_rdev
);
1171 dl
->minor
= minor(stb
.st_rdev
);
1172 dl
->next
= super
->disks
;
1173 dl
->fd
= keep_fd
? fd
: -1;
1174 dl
->devname
= devname
? strdup(devname
) : NULL
;
1175 serialcpy(dl
->serial
, serial
);
1177 } else if (keep_fd
) {
1182 /* look up this disk's index in the current anchor */
1183 for (i
= 0; i
< super
->anchor
->num_disks
; i
++) {
1184 struct imsm_disk
*disk_iter
;
1186 disk_iter
= __get_imsm_disk(super
->anchor
, i
);
1188 if (serialcmp(disk_iter
->serial
, dl
->serial
) == 0) {
1189 dl
->disk
= *disk_iter
;
1190 /* only set index on disks that are a member of a
1191 * populated contianer, i.e. one with raid_devs
1193 if (dl
->disk
.status
& FAILED_DISK
)
1195 else if (dl
->disk
.status
& SPARE_DISK
)
1204 /* no match, maybe a stale failed drive */
1205 if (i
== super
->anchor
->num_disks
&& dl
->index
>= 0) {
1206 dl
->disk
= *__get_imsm_disk(super
->anchor
, dl
->index
);
1207 if (dl
->disk
.status
& FAILED_DISK
)
1217 static void imsm_copy_dev(struct imsm_dev
*dest
, struct imsm_dev
*src
)
1219 memcpy(dest
, src
, sizeof_imsm_dev(src
, 0));
1223 /* When migrating map0 contains the 'destination' state while map1
1224 * contains the current state. When not migrating map0 contains the
1225 * current state. This routine assumes that map[0].map_state is set to
1226 * the current array state before being called.
1228 * Migration is indicated by one of the following states
1229 * 1/ Idle (migr_state=0 map0state=normal||unitialized||degraded||failed)
1230 * 2/ Initialize (migr_state=1 migr_type=MIGR_INIT map0state=normal
1231 * map1state=unitialized)
1232 * 3/ Verify (Resync) (migr_state=1 migr_type=MIGR_REBUILD map0state=normal
1234 * 4/ Rebuild (migr_state=1 migr_type=MIGR_REBUILD map0state=normal
1235 * map1state=degraded)
1237 static void migrate(struct imsm_dev
*dev
, __u8 to_state
, int rebuild_resync
)
1239 struct imsm_map
*dest
;
1240 struct imsm_map
*src
= get_imsm_map(dev
, 0);
1242 dev
->vol
.migr_state
= 1;
1243 dev
->vol
.migr_type
= rebuild_resync
;
1244 dev
->vol
.curr_migr_unit
= 0;
1245 dest
= get_imsm_map(dev
, 1);
1247 memcpy(dest
, src
, sizeof_imsm_map(src
));
1248 src
->map_state
= to_state
;
1251 static void end_migration(struct imsm_dev
*dev
, __u8 map_state
)
1253 struct imsm_map
*map
= get_imsm_map(dev
, 0);
1255 dev
->vol
.migr_state
= 0;
1256 dev
->vol
.curr_migr_unit
= 0;
1257 map
->map_state
= map_state
;
1261 static int parse_raid_devices(struct intel_super
*super
)
1264 struct imsm_dev
*dev_new
;
1265 size_t len
, len_migr
;
1266 size_t space_needed
= 0;
1267 struct imsm_super
*mpb
= super
->anchor
;
1269 for (i
= 0; i
< super
->anchor
->num_raid_devs
; i
++) {
1270 struct imsm_dev
*dev_iter
= __get_imsm_dev(super
->anchor
, i
);
1272 len
= sizeof_imsm_dev(dev_iter
, 0);
1273 len_migr
= sizeof_imsm_dev(dev_iter
, 1);
1275 space_needed
+= len_migr
- len
;
1277 dev_new
= malloc(len_migr
);
1280 imsm_copy_dev(dev_new
, dev_iter
);
1281 super
->dev_tbl
[i
] = dev_new
;
1284 /* ensure that super->buf is large enough when all raid devices
1287 if (__le32_to_cpu(mpb
->mpb_size
) + space_needed
> super
->len
) {
1290 len
= ROUND_UP(__le32_to_cpu(mpb
->mpb_size
) + space_needed
, 512);
1291 if (posix_memalign(&buf
, 512, len
) != 0)
1294 memcpy(buf
, super
->buf
, len
);
1303 /* retrieve a pointer to the bbm log which starts after all raid devices */
1304 struct bbm_log
*__get_imsm_bbm_log(struct imsm_super
*mpb
)
1308 if (__le32_to_cpu(mpb
->bbm_log_size
)) {
1310 ptr
+= mpb
->mpb_size
- __le32_to_cpu(mpb
->bbm_log_size
);
1316 static void __free_imsm(struct intel_super
*super
, int free_disks
);
1318 /* load_imsm_mpb - read matrix metadata
1319 * allocates super->mpb to be freed by free_super
1321 static int load_imsm_mpb(int fd
, struct intel_super
*super
, char *devname
)
1323 unsigned long long dsize
;
1324 unsigned long long sectors
;
1326 struct imsm_super
*anchor
;
1330 get_dev_size(fd
, NULL
, &dsize
);
1332 if (lseek64(fd
, dsize
- (512 * 2), SEEK_SET
) < 0) {
1335 Name
": Cannot seek to anchor block on %s: %s\n",
1336 devname
, strerror(errno
));
1340 if (posix_memalign((void**)&anchor
, 512, 512) != 0) {
1343 Name
": Failed to allocate imsm anchor buffer"
1344 " on %s\n", devname
);
1347 if (read(fd
, anchor
, 512) != 512) {
1350 Name
": Cannot read anchor block on %s: %s\n",
1351 devname
, strerror(errno
));
1356 if (strncmp((char *) anchor
->sig
, MPB_SIGNATURE
, MPB_SIG_LEN
) != 0) {
1359 Name
": no IMSM anchor on %s\n", devname
);
1364 __free_imsm(super
, 0);
1365 super
->len
= ROUND_UP(anchor
->mpb_size
, 512);
1366 if (posix_memalign(&super
->buf
, 512, super
->len
) != 0) {
1369 Name
": unable to allocate %zu byte mpb buffer\n",
1374 memcpy(super
->buf
, anchor
, 512);
1376 sectors
= mpb_sectors(anchor
) - 1;
1379 rc
= load_imsm_disk(fd
, super
, devname
, 0);
1381 rc
= parse_raid_devices(super
);
1385 /* read the extended mpb */
1386 if (lseek64(fd
, dsize
- (512 * (2 + sectors
)), SEEK_SET
) < 0) {
1389 Name
": Cannot seek to extended mpb on %s: %s\n",
1390 devname
, strerror(errno
));
1394 if (read(fd
, super
->buf
+ 512, super
->len
- 512) != super
->len
- 512) {
1397 Name
": Cannot read extended mpb on %s: %s\n",
1398 devname
, strerror(errno
));
1402 check_sum
= __gen_imsm_checksum(super
->anchor
);
1403 if (check_sum
!= __le32_to_cpu(super
->anchor
->check_sum
)) {
1406 Name
": IMSM checksum %x != %x on %s\n",
1407 check_sum
, __le32_to_cpu(super
->anchor
->check_sum
),
1412 /* FIXME the BBM log is disk specific so we cannot use this global
1413 * buffer for all disks. Ok for now since we only look at the global
1414 * bbm_log_size parameter to gate assembly
1416 super
->bbm_log
= __get_imsm_bbm_log(super
->anchor
);
1418 rc
= load_imsm_disk(fd
, super
, devname
, 0);
1420 rc
= parse_raid_devices(super
);
1425 static void __free_imsm_disk(struct dl
*d
)
1434 static void free_imsm_disks(struct intel_super
*super
)
1438 while (super
->disks
) {
1440 super
->disks
= d
->next
;
1441 __free_imsm_disk(d
);
1443 while (super
->missing
) {
1445 super
->missing
= d
->next
;
1446 __free_imsm_disk(d
);
1451 /* free all the pieces hanging off of a super pointer */
1452 static void __free_imsm(struct intel_super
*super
, int free_disks
)
1461 free_imsm_disks(super
);
1462 for (i
= 0; i
< IMSM_MAX_RAID_DEVS
; i
++)
1463 if (super
->dev_tbl
[i
]) {
1464 free(super
->dev_tbl
[i
]);
1465 super
->dev_tbl
[i
] = NULL
;
1469 static void free_imsm(struct intel_super
*super
)
1471 __free_imsm(super
, 1);
1475 static void free_super_imsm(struct supertype
*st
)
1477 struct intel_super
*super
= st
->sb
;
1486 static struct intel_super
*alloc_super(int creating_imsm
)
1488 struct intel_super
*super
= malloc(sizeof(*super
));
1491 memset(super
, 0, sizeof(*super
));
1492 super
->creating_imsm
= creating_imsm
;
1493 super
->current_vol
= -1;
1500 /* find_missing - helper routine for load_super_imsm_all that identifies
1501 * disks that have disappeared from the system. This routine relies on
1502 * the mpb being uptodate, which it is at load time.
1504 static int find_missing(struct intel_super
*super
)
1507 struct imsm_super
*mpb
= super
->anchor
;
1509 struct imsm_disk
*disk
;
1511 for (i
= 0; i
< mpb
->num_disks
; i
++) {
1512 disk
= __get_imsm_disk(mpb
, i
);
1513 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
1514 if (serialcmp(dl
->disk
.serial
, disk
->serial
) == 0)
1518 /* ok we have a 'disk' without a live entry in
1521 if (disk
->status
& FAILED_DISK
|| !(disk
->status
& USABLE_DISK
))
1522 continue; /* never mind, already marked */
1524 dl
= malloc(sizeof(*dl
));
1530 dl
->devname
= strdup("missing");
1532 serialcpy(dl
->serial
, disk
->serial
);
1534 dl
->next
= super
->missing
;
1535 super
->missing
= dl
;
1541 static int load_super_imsm_all(struct supertype
*st
, int fd
, void **sbp
,
1542 char *devname
, int keep_fd
)
1545 struct intel_super
*super
;
1546 struct mdinfo
*sd
, *best
= NULL
;
1553 /* check if this disk is a member of an active array */
1554 sra
= sysfs_read(fd
, 0, GET_LEVEL
|GET_VERSION
|GET_DEVS
|GET_STATE
);
1558 if (sra
->array
.major_version
!= -1 ||
1559 sra
->array
.minor_version
!= -2 ||
1560 strcmp(sra
->text_version
, "imsm") != 0)
1563 super
= alloc_super(0);
1567 /* find the most up to date disk in this array, skipping spares */
1568 for (sd
= sra
->devs
; sd
; sd
= sd
->next
) {
1569 sprintf(nm
, "%d:%d", sd
->disk
.major
, sd
->disk
.minor
);
1570 dfd
= dev_open(nm
, keep_fd
? O_RDWR
: O_RDONLY
);
1575 rv
= load_imsm_mpb(dfd
, super
, NULL
);
1579 if (super
->anchor
->num_raid_devs
== 0)
1582 gen
= __le32_to_cpu(super
->anchor
->generation_num
);
1583 if (!best
|| gen
> bestgen
) {
1598 /* load the most up to date anchor */
1599 sprintf(nm
, "%d:%d", best
->disk
.major
, best
->disk
.minor
);
1600 dfd
= dev_open(nm
, O_RDONLY
);
1605 rv
= load_imsm_mpb(dfd
, super
, NULL
);
1612 /* re-parse the disk list with the current anchor */
1613 for (sd
= sra
->devs
; sd
; sd
= sd
->next
) {
1614 sprintf(nm
, "%d:%d", sd
->disk
.major
, sd
->disk
.minor
);
1615 dfd
= dev_open(nm
, keep_fd
? O_RDWR
: O_RDONLY
);
1620 load_imsm_disk(dfd
, super
, NULL
, keep_fd
);
1626 if (find_missing(super
) != 0) {
1631 if (st
->subarray
[0]) {
1632 if (atoi(st
->subarray
) <= super
->anchor
->num_raid_devs
)
1633 super
->current_vol
= atoi(st
->subarray
);
1639 st
->container_dev
= fd2devnum(fd
);
1640 if (st
->ss
== NULL
) {
1641 st
->ss
= &super_imsm
;
1642 st
->minor_version
= 0;
1643 st
->max_devs
= IMSM_MAX_DEVICES
;
1645 st
->loaded_container
= 1;
1651 static int load_super_imsm(struct supertype
*st
, int fd
, char *devname
)
1653 struct intel_super
*super
;
1657 if (load_super_imsm_all(st
, fd
, &st
->sb
, devname
, 1) == 0)
1660 if (st
->subarray
[0])
1661 return 1; /* FIXME */
1663 super
= alloc_super(0);
1666 Name
": malloc of %zu failed.\n",
1671 rv
= load_imsm_mpb(fd
, super
, devname
);
1676 Name
": Failed to load all information "
1677 "sections on %s\n", devname
);
1683 if (st
->ss
== NULL
) {
1684 st
->ss
= &super_imsm
;
1685 st
->minor_version
= 0;
1686 st
->max_devs
= IMSM_MAX_DEVICES
;
1688 st
->loaded_container
= 0;
1693 static __u16
info_to_blocks_per_strip(mdu_array_info_t
*info
)
1695 if (info
->level
== 1)
1697 return info
->chunk_size
>> 9;
1700 static __u32
info_to_num_data_stripes(mdu_array_info_t
*info
)
1704 num_stripes
= (info
->size
* 2) / info_to_blocks_per_strip(info
);
1705 if (info
->level
== 1)
1711 static __u32
info_to_blocks_per_member(mdu_array_info_t
*info
)
1713 return (info
->size
* 2) & ~(info_to_blocks_per_strip(info
) - 1);
1716 static void imsm_update_version_info(struct intel_super
*super
)
1718 /* update the version and attributes */
1719 struct imsm_super
*mpb
= super
->anchor
;
1721 struct imsm_dev
*dev
;
1722 struct imsm_map
*map
;
1725 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
1726 dev
= get_imsm_dev(super
, i
);
1727 map
= get_imsm_map(dev
, 0);
1728 if (__le32_to_cpu(dev
->size_high
) > 0)
1729 mpb
->attributes
|= MPB_ATTRIB_2TB
;
1731 /* FIXME detect when an array spans a port multiplier */
1733 mpb
->attributes
|= MPB_ATTRIB_PM
;
1736 if (mpb
->num_raid_devs
> 1 ||
1737 mpb
->attributes
!= MPB_ATTRIB_CHECKSUM_VERIFY
) {
1738 version
= MPB_VERSION_ATTRIBS
;
1739 switch (get_imsm_raid_level(map
)) {
1740 case 0: mpb
->attributes
|= MPB_ATTRIB_RAID0
; break;
1741 case 1: mpb
->attributes
|= MPB_ATTRIB_RAID1
; break;
1742 case 10: mpb
->attributes
|= MPB_ATTRIB_RAID10
; break;
1743 case 5: mpb
->attributes
|= MPB_ATTRIB_RAID5
; break;
1746 if (map
->num_members
>= 5)
1747 version
= MPB_VERSION_5OR6_DISK_ARRAY
;
1748 else if (dev
->status
== DEV_CLONE_N_GO
)
1749 version
= MPB_VERSION_CNG
;
1750 else if (get_imsm_raid_level(map
) == 5)
1751 version
= MPB_VERSION_RAID5
;
1752 else if (map
->num_members
>= 3)
1753 version
= MPB_VERSION_3OR4_DISK_ARRAY
;
1754 else if (get_imsm_raid_level(map
) == 1)
1755 version
= MPB_VERSION_RAID1
;
1757 version
= MPB_VERSION_RAID0
;
1759 strcpy(((char *) mpb
->sig
) + strlen(MPB_SIGNATURE
), version
);
1763 static int init_super_imsm_volume(struct supertype
*st
, mdu_array_info_t
*info
,
1764 unsigned long long size
, char *name
,
1765 char *homehost
, int *uuid
)
1767 /* We are creating a volume inside a pre-existing container.
1768 * so st->sb is already set.
1770 struct intel_super
*super
= st
->sb
;
1771 struct imsm_super
*mpb
= super
->anchor
;
1772 struct imsm_dev
*dev
;
1773 struct imsm_vol
*vol
;
1774 struct imsm_map
*map
;
1775 int idx
= mpb
->num_raid_devs
;
1777 unsigned long long array_blocks
;
1779 size_t size_old
, size_new
;
1781 if (mpb
->num_raid_devs
>= 2) {
1782 fprintf(stderr
, Name
": This imsm-container already has the "
1783 "maximum of 2 volumes\n");
1787 /* ensure the mpb is large enough for the new data */
1788 size_old
= __le32_to_cpu(mpb
->mpb_size
);
1789 size_new
= disks_to_mpb_size(info
->nr_disks
);
1790 if (size_new
> size_old
) {
1792 size_t size_round
= ROUND_UP(size_new
, 512);
1794 if (posix_memalign(&mpb_new
, 512, size_round
) != 0) {
1795 fprintf(stderr
, Name
": could not allocate new mpb\n");
1798 memcpy(mpb_new
, mpb
, size_old
);
1801 super
->anchor
= mpb_new
;
1802 mpb
->mpb_size
= __cpu_to_le32(size_new
);
1803 memset(mpb_new
+ size_old
, 0, size_round
- size_old
);
1805 super
->current_vol
= idx
;
1806 /* when creating the first raid device in this container set num_disks
1807 * to zero, i.e. delete this spare and add raid member devices in
1808 * add_to_super_imsm_volume()
1810 if (super
->current_vol
== 0)
1812 sprintf(st
->subarray
, "%d", idx
);
1813 dev
= malloc(sizeof(*dev
) + sizeof(__u32
) * (info
->raid_disks
- 1));
1815 fprintf(stderr
, Name
": could not allocate raid device\n");
1818 strncpy((char *) dev
->volume
, name
, MAX_RAID_SERIAL_LEN
);
1819 array_blocks
= calc_array_size(info
->level
, info
->raid_disks
,
1820 info
->layout
, info
->chunk_size
,
1822 dev
->size_low
= __cpu_to_le32((__u32
) array_blocks
);
1823 dev
->size_high
= __cpu_to_le32((__u32
) (array_blocks
>> 32));
1824 dev
->status
= __cpu_to_le32(0);
1825 dev
->reserved_blocks
= __cpu_to_le32(0);
1827 vol
->migr_state
= 0;
1828 vol
->migr_type
= MIGR_INIT
;
1830 vol
->curr_migr_unit
= 0;
1831 for (i
= 0; i
< idx
; i
++) {
1832 struct imsm_dev
*prev
= get_imsm_dev(super
, i
);
1833 struct imsm_map
*pmap
= get_imsm_map(prev
, 0);
1835 offset
+= __le32_to_cpu(pmap
->blocks_per_member
);
1836 offset
+= IMSM_RESERVED_SECTORS
;
1838 map
= get_imsm_map(dev
, 0);
1839 map
->pba_of_lba0
= __cpu_to_le32(offset
);
1840 map
->blocks_per_member
= __cpu_to_le32(info_to_blocks_per_member(info
));
1841 map
->blocks_per_strip
= __cpu_to_le16(info_to_blocks_per_strip(info
));
1842 map
->num_data_stripes
= __cpu_to_le32(info_to_num_data_stripes(info
));
1843 map
->map_state
= info
->level
? IMSM_T_STATE_UNINITIALIZED
:
1844 IMSM_T_STATE_NORMAL
;
1846 if (info
->level
== 1 && info
->raid_disks
> 2) {
1847 fprintf(stderr
, Name
": imsm does not support more than 2 disks"
1848 "in a raid1 volume\n");
1851 if (info
->level
== 10) {
1852 map
->raid_level
= 1;
1853 map
->num_domains
= info
->raid_disks
/ 2;
1855 map
->raid_level
= info
->level
;
1856 map
->num_domains
= !!map
->raid_level
;
1859 map
->num_members
= info
->raid_disks
;
1860 for (i
= 0; i
< map
->num_members
; i
++) {
1861 /* initialized in add_to_super */
1862 set_imsm_ord_tbl_ent(map
, i
, 0);
1864 mpb
->num_raid_devs
++;
1865 super
->dev_tbl
[super
->current_vol
] = dev
;
1867 imsm_update_version_info(super
);
1872 static int init_super_imsm(struct supertype
*st
, mdu_array_info_t
*info
,
1873 unsigned long long size
, char *name
,
1874 char *homehost
, int *uuid
)
1876 /* This is primarily called by Create when creating a new array.
1877 * We will then get add_to_super called for each component, and then
1878 * write_init_super called to write it out to each device.
1879 * For IMSM, Create can create on fresh devices or on a pre-existing
1881 * To create on a pre-existing array a different method will be called.
1882 * This one is just for fresh drives.
1884 struct intel_super
*super
;
1885 struct imsm_super
*mpb
;
1894 return init_super_imsm_volume(st
, info
, size
, name
, homehost
,
1897 super
= alloc_super(1);
1900 mpb_size
= disks_to_mpb_size(info
->nr_disks
);
1901 if (posix_memalign(&super
->buf
, 512, mpb_size
) != 0) {
1906 memset(mpb
, 0, mpb_size
);
1908 mpb
->attributes
= MPB_ATTRIB_CHECKSUM_VERIFY
;
1910 version
= (char *) mpb
->sig
;
1911 strcpy(version
, MPB_SIGNATURE
);
1912 version
+= strlen(MPB_SIGNATURE
);
1913 strcpy(version
, MPB_VERSION_RAID0
);
1914 mpb
->mpb_size
= mpb_size
;
1921 static int add_to_super_imsm_volume(struct supertype
*st
, mdu_disk_info_t
*dk
,
1922 int fd
, char *devname
)
1924 struct intel_super
*super
= st
->sb
;
1925 struct imsm_super
*mpb
= super
->anchor
;
1927 struct imsm_dev
*dev
;
1928 struct imsm_map
*map
;
1930 dev
= get_imsm_dev(super
, super
->current_vol
);
1931 map
= get_imsm_map(dev
, 0);
1933 if (! (dk
->state
& (1<<MD_DISK_SYNC
))) {
1934 fprintf(stderr
, Name
": %s: Cannot add spare devices to IMSM volume\n",
1939 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
1940 if (dl
->major
== dk
->major
&&
1941 dl
->minor
== dk
->minor
)
1945 fprintf(stderr
, Name
": %s is not a member of the same container\n", devname
);
1949 /* add a pristine spare to the metadata */
1950 if (dl
->index
< 0) {
1951 dl
->index
= super
->anchor
->num_disks
;
1952 super
->anchor
->num_disks
++;
1954 set_imsm_ord_tbl_ent(map
, dk
->number
, dl
->index
);
1955 dl
->disk
.status
= CONFIGURED_DISK
| USABLE_DISK
;
1957 /* if we are creating the first raid device update the family number */
1958 if (super
->current_vol
== 0) {
1960 struct imsm_dev
*_dev
= __get_imsm_dev(mpb
, 0);
1961 struct imsm_disk
*_disk
= __get_imsm_disk(mpb
, dl
->index
);
1965 sum
= __gen_imsm_checksum(mpb
);
1966 mpb
->family_num
= __cpu_to_le32(sum
);
1972 static int add_to_super_imsm(struct supertype
*st
, mdu_disk_info_t
*dk
,
1973 int fd
, char *devname
)
1975 struct intel_super
*super
= st
->sb
;
1977 unsigned long long size
;
1982 if (super
->current_vol
>= 0)
1983 return add_to_super_imsm_volume(st
, dk
, fd
, devname
);
1986 dd
= malloc(sizeof(*dd
));
1989 Name
": malloc failed %s:%d.\n", __func__
, __LINE__
);
1992 memset(dd
, 0, sizeof(*dd
));
1993 dd
->major
= major(stb
.st_rdev
);
1994 dd
->minor
= minor(stb
.st_rdev
);
1996 dd
->devname
= devname
? strdup(devname
) : NULL
;
1998 rv
= imsm_read_serial(fd
, devname
, dd
->serial
);
2001 Name
": failed to retrieve scsi serial, aborting\n");
2006 get_dev_size(fd
, NULL
, &size
);
2008 serialcpy(dd
->disk
.serial
, dd
->serial
);
2009 dd
->disk
.total_blocks
= __cpu_to_le32(size
);
2010 dd
->disk
.status
= USABLE_DISK
| SPARE_DISK
;
2011 if (sysfs_disk_to_scsi_id(fd
, &id
) == 0)
2012 dd
->disk
.scsi_id
= __cpu_to_le32(id
);
2014 dd
->disk
.scsi_id
= __cpu_to_le32(0);
2016 if (st
->update_tail
) {
2017 dd
->next
= super
->add
;
2020 dd
->next
= super
->disks
;
2027 static int store_imsm_mpb(int fd
, struct intel_super
*super
);
2029 /* spare records have their own family number and do not have any defined raid
2032 static int write_super_imsm_spares(struct intel_super
*super
, int doclose
)
2034 struct imsm_super mpb_save
;
2035 struct imsm_super
*mpb
= super
->anchor
;
2040 mpb
->num_raid_devs
= 0;
2042 mpb
->mpb_size
= sizeof(struct imsm_super
);
2043 mpb
->generation_num
= __cpu_to_le32(1UL);
2045 for (d
= super
->disks
; d
; d
= d
->next
) {
2049 mpb
->disk
[0] = d
->disk
;
2050 sum
= __gen_imsm_checksum(mpb
);
2051 mpb
->family_num
= __cpu_to_le32(sum
);
2052 sum
= __gen_imsm_checksum(mpb
);
2053 mpb
->check_sum
= __cpu_to_le32(sum
);
2055 if (store_imsm_mpb(d
->fd
, super
)) {
2056 fprintf(stderr
, "%s: failed for device %d:%d %s\n",
2057 __func__
, d
->major
, d
->minor
, strerror(errno
));
2071 static int write_super_imsm(struct intel_super
*super
, int doclose
)
2073 struct imsm_super
*mpb
= super
->anchor
;
2079 __u32 mpb_size
= sizeof(struct imsm_super
) - sizeof(struct imsm_disk
);
2081 /* 'generation' is incremented everytime the metadata is written */
2082 generation
= __le32_to_cpu(mpb
->generation_num
);
2084 mpb
->generation_num
= __cpu_to_le32(generation
);
2086 mpb_size
+= sizeof(struct imsm_disk
) * mpb
->num_disks
;
2087 for (d
= super
->disks
; d
; d
= d
->next
) {
2091 mpb
->disk
[d
->index
] = d
->disk
;
2093 for (d
= super
->missing
; d
; d
= d
->next
)
2094 mpb
->disk
[d
->index
] = d
->disk
;
2096 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
2097 struct imsm_dev
*dev
= __get_imsm_dev(mpb
, i
);
2099 imsm_copy_dev(dev
, super
->dev_tbl
[i
]);
2100 mpb_size
+= sizeof_imsm_dev(dev
, 0);
2102 mpb_size
+= __le32_to_cpu(mpb
->bbm_log_size
);
2103 mpb
->mpb_size
= __cpu_to_le32(mpb_size
);
2105 /* recalculate checksum */
2106 sum
= __gen_imsm_checksum(mpb
);
2107 mpb
->check_sum
= __cpu_to_le32(sum
);
2109 /* write the mpb for disks that compose raid devices */
2110 for (d
= super
->disks
; d
; d
= d
->next
) {
2113 if (store_imsm_mpb(d
->fd
, super
))
2114 fprintf(stderr
, "%s: failed for device %d:%d %s\n",
2115 __func__
, d
->major
, d
->minor
, strerror(errno
));
2123 return write_super_imsm_spares(super
, doclose
);
2129 static int create_array(struct supertype
*st
)
2132 struct imsm_update_create_array
*u
;
2133 struct intel_super
*super
= st
->sb
;
2134 struct imsm_dev
*dev
= get_imsm_dev(super
, super
->current_vol
);
2136 len
= sizeof(*u
) - sizeof(*dev
) + sizeof_imsm_dev(dev
, 0);
2139 fprintf(stderr
, "%s: failed to allocate update buffer\n",
2144 u
->type
= update_create_array
;
2145 u
->dev_idx
= super
->current_vol
;
2146 imsm_copy_dev(&u
->dev
, dev
);
2147 append_metadata_update(st
, u
, len
);
2152 static int _add_disk(struct supertype
*st
)
2154 struct intel_super
*super
= st
->sb
;
2156 struct imsm_update_add_disk
*u
;
2164 fprintf(stderr
, "%s: failed to allocate update buffer\n",
2169 u
->type
= update_add_disk
;
2170 append_metadata_update(st
, u
, len
);
2175 static int write_init_super_imsm(struct supertype
*st
)
2177 if (st
->update_tail
) {
2178 /* queue the recently created array / added disk
2179 * as a metadata update */
2180 struct intel_super
*super
= st
->sb
;
2184 /* determine if we are creating a volume or adding a disk */
2185 if (super
->current_vol
< 0) {
2186 /* in the add disk case we are running in mdmon
2187 * context, so don't close fd's
2189 return _add_disk(st
);
2191 rv
= create_array(st
);
2193 for (d
= super
->disks
; d
; d
= d
->next
) {
2200 return write_super_imsm(st
->sb
, 1);
2204 static int store_zero_imsm(struct supertype
*st
, int fd
)
2206 unsigned long long dsize
;
2209 get_dev_size(fd
, NULL
, &dsize
);
2211 /* first block is stored on second to last sector of the disk */
2212 if (lseek64(fd
, dsize
- (512 * 2), SEEK_SET
) < 0)
2215 if (posix_memalign(&buf
, 512, 512) != 0)
2218 memset(buf
, 0, 512);
2219 if (write(fd
, buf
, 512) != 512)
2224 static int imsm_bbm_log_size(struct imsm_super
*mpb
)
2226 return __le32_to_cpu(mpb
->bbm_log_size
);
2230 static int validate_geometry_imsm_container(struct supertype
*st
, int level
,
2231 int layout
, int raiddisks
, int chunk
,
2232 unsigned long long size
, char *dev
,
2233 unsigned long long *freesize
,
2237 unsigned long long ldsize
;
2239 if (level
!= LEVEL_CONTAINER
)
2244 fd
= open(dev
, O_RDONLY
|O_EXCL
, 0);
2247 fprintf(stderr
, Name
": imsm: Cannot open %s: %s\n",
2248 dev
, strerror(errno
));
2251 if (!get_dev_size(fd
, dev
, &ldsize
)) {
2257 *freesize
= avail_size_imsm(st
, ldsize
>> 9);
2262 /* validate_geometry_imsm_volume - lifted from validate_geometry_ddf_bvd
2263 * FIX ME add ahci details
2265 static int validate_geometry_imsm_volume(struct supertype
*st
, int level
,
2266 int layout
, int raiddisks
, int chunk
,
2267 unsigned long long size
, char *dev
,
2268 unsigned long long *freesize
,
2272 struct intel_super
*super
= st
->sb
;
2274 unsigned long long pos
= 0;
2275 unsigned long long maxsize
;
2279 if (level
== LEVEL_CONTAINER
)
2282 if (level
== 1 && raiddisks
> 2) {
2284 fprintf(stderr
, Name
": imsm does not support more "
2285 "than 2 in a raid1 configuration\n");
2289 /* We must have the container info already read in. */
2294 /* General test: make sure there is space for
2295 * 'raiddisks' device extents of size 'size' at a given
2298 unsigned long long minsize
= size
;
2299 unsigned long long start_offset
= ~0ULL;
2302 minsize
= MPB_SECTOR_CNT
+ IMSM_RESERVED_SECTORS
;
2303 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
2308 e
= get_extents(super
, dl
);
2311 unsigned long long esize
;
2312 esize
= e
[i
].start
- pos
;
2313 if (esize
>= minsize
)
2315 if (found
&& start_offset
== ~0ULL) {
2318 } else if (found
&& pos
!= start_offset
) {
2322 pos
= e
[i
].start
+ e
[i
].size
;
2324 } while (e
[i
-1].size
);
2329 if (dcnt
< raiddisks
) {
2331 fprintf(stderr
, Name
": imsm: Not enough "
2332 "devices with space for this array "
2339 /* This device must be a member of the set */
2340 if (stat(dev
, &stb
) < 0)
2342 if ((S_IFMT
& stb
.st_mode
) != S_IFBLK
)
2344 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
2345 if (dl
->major
== major(stb
.st_rdev
) &&
2346 dl
->minor
== minor(stb
.st_rdev
))
2351 fprintf(stderr
, Name
": %s is not in the "
2352 "same imsm set\n", dev
);
2355 e
= get_extents(super
, dl
);
2359 unsigned long long esize
;
2360 esize
= e
[i
].start
- pos
;
2361 if (esize
>= maxsize
)
2363 pos
= e
[i
].start
+ e
[i
].size
;
2365 } while (e
[i
-1].size
);
2366 *freesize
= maxsize
;
2371 static int validate_geometry_imsm(struct supertype
*st
, int level
, int layout
,
2372 int raiddisks
, int chunk
, unsigned long long size
,
2373 char *dev
, unsigned long long *freesize
,
2379 /* if given unused devices create a container
2380 * if given given devices in a container create a member volume
2382 if (level
== LEVEL_CONTAINER
) {
2383 /* Must be a fresh device to add to a container */
2384 return validate_geometry_imsm_container(st
, level
, layout
,
2385 raiddisks
, chunk
, size
,
2391 if (st
->sb
&& freesize
) {
2392 /* Should do auto-layout here */
2393 fprintf(stderr
, Name
": IMSM does not support auto-layout yet\n");
2399 /* creating in a given container */
2400 return validate_geometry_imsm_volume(st
, level
, layout
,
2401 raiddisks
, chunk
, size
,
2402 dev
, freesize
, verbose
);
2405 /* limit creation to the following levels */
2417 /* This device needs to be a device in an 'imsm' container */
2418 fd
= open(dev
, O_RDONLY
|O_EXCL
, 0);
2422 Name
": Cannot create this array on device %s\n",
2427 if (errno
!= EBUSY
|| (fd
= open(dev
, O_RDONLY
, 0)) < 0) {
2429 fprintf(stderr
, Name
": Cannot open %s: %s\n",
2430 dev
, strerror(errno
));
2433 /* Well, it is in use by someone, maybe an 'imsm' container. */
2434 cfd
= open_container(fd
);
2438 fprintf(stderr
, Name
": Cannot use %s: It is busy\n",
2442 sra
= sysfs_read(cfd
, 0, GET_VERSION
);
2444 if (sra
&& sra
->array
.major_version
== -1 &&
2445 strcmp(sra
->text_version
, "imsm") == 0) {
2446 /* This is a member of a imsm container. Load the container
2447 * and try to create a volume
2449 struct intel_super
*super
;
2451 if (load_super_imsm_all(st
, cfd
, (void **) &super
, NULL
, 1) == 0) {
2453 st
->container_dev
= fd2devnum(cfd
);
2455 return validate_geometry_imsm_volume(st
, level
, layout
,
2461 } else /* may belong to another container */
2466 #endif /* MDASSEMBLE */
2468 static struct mdinfo
*container_content_imsm(struct supertype
*st
)
2470 /* Given a container loaded by load_super_imsm_all,
2471 * extract information about all the arrays into
2474 * For each imsm_dev create an mdinfo, fill it in,
2475 * then look for matching devices in super->disks
2476 * and create appropriate device mdinfo.
2478 struct intel_super
*super
= st
->sb
;
2479 struct imsm_super
*mpb
= super
->anchor
;
2480 struct mdinfo
*rest
= NULL
;
2483 /* do not assemble arrays that might have bad blocks */
2484 if (imsm_bbm_log_size(super
->anchor
)) {
2485 fprintf(stderr
, Name
": BBM log found in metadata. "
2486 "Cannot activate array(s).\n");
2490 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
2491 struct imsm_dev
*dev
= get_imsm_dev(super
, i
);
2492 struct imsm_map
*map
= get_imsm_map(dev
, 0);
2493 struct mdinfo
*this;
2496 this = malloc(sizeof(*this));
2497 memset(this, 0, sizeof(*this));
2500 super
->current_vol
= i
;
2501 getinfo_super_imsm_volume(st
, this);
2502 for (slot
= 0 ; slot
< map
->num_members
; slot
++) {
2503 struct mdinfo
*info_d
;
2511 idx
= get_imsm_disk_idx(dev
, slot
);
2512 ord
= get_imsm_ord_tbl_ent(dev
, slot
);
2513 for (d
= super
->disks
; d
; d
= d
->next
)
2514 if (d
->index
== idx
)
2520 s
= d
? d
->disk
.status
: 0;
2521 if (s
& FAILED_DISK
)
2523 if (!(s
& USABLE_DISK
))
2525 if (ord
& IMSM_ORD_REBUILD
)
2529 * if we skip some disks the array will be assmebled degraded;
2530 * reset resync start to avoid a dirty-degraded situation
2532 * FIXME handle dirty degraded
2534 if (skip
&& !dev
->vol
.dirty
)
2535 this->resync_start
= ~0ULL;
2539 info_d
= malloc(sizeof(*info_d
));
2541 fprintf(stderr
, Name
": failed to allocate disk"
2542 " for volume %s\n", (char *) dev
->volume
);
2547 memset(info_d
, 0, sizeof(*info_d
));
2548 info_d
->next
= this->devs
;
2549 this->devs
= info_d
;
2551 info_d
->disk
.number
= d
->index
;
2552 info_d
->disk
.major
= d
->major
;
2553 info_d
->disk
.minor
= d
->minor
;
2554 info_d
->disk
.raid_disk
= slot
;
2556 this->array
.working_disks
++;
2558 info_d
->events
= __le32_to_cpu(mpb
->generation_num
);
2559 info_d
->data_offset
= __le32_to_cpu(map
->pba_of_lba0
);
2560 info_d
->component_size
= __le32_to_cpu(map
->blocks_per_member
);
2562 strcpy(info_d
->name
, d
->devname
);
2572 static int imsm_open_new(struct supertype
*c
, struct active_array
*a
,
2575 struct intel_super
*super
= c
->sb
;
2576 struct imsm_super
*mpb
= super
->anchor
;
2578 if (atoi(inst
) >= mpb
->num_raid_devs
) {
2579 fprintf(stderr
, "%s: subarry index %d, out of range\n",
2580 __func__
, atoi(inst
));
2584 dprintf("imsm: open_new %s\n", inst
);
2585 a
->info
.container_member
= atoi(inst
);
2589 static __u8
imsm_check_degraded(struct intel_super
*super
, struct imsm_dev
*dev
, int failed
)
2591 struct imsm_map
*map
= get_imsm_map(dev
, 0);
2594 return map
->map_state
== IMSM_T_STATE_UNINITIALIZED
?
2595 IMSM_T_STATE_UNINITIALIZED
: IMSM_T_STATE_NORMAL
;
2597 switch (get_imsm_raid_level(map
)) {
2599 return IMSM_T_STATE_FAILED
;
2602 if (failed
< map
->num_members
)
2603 return IMSM_T_STATE_DEGRADED
;
2605 return IMSM_T_STATE_FAILED
;
2610 * check to see if any mirrors have failed, otherwise we
2611 * are degraded. Even numbered slots are mirrored on
2615 /* gcc -Os complains that this is unused */
2616 int insync
= insync
;
2618 for (i
= 0; i
< map
->num_members
; i
++) {
2619 __u32 ord
= get_imsm_ord_tbl_ent(dev
, i
);
2620 int idx
= ord_to_idx(ord
);
2621 struct imsm_disk
*disk
;
2623 /* reset the potential in-sync count on even-numbered
2624 * slots. num_copies is always 2 for imsm raid10
2629 disk
= get_imsm_disk(super
, idx
);
2630 if (!disk
|| disk
->status
& FAILED_DISK
||
2631 ord
& IMSM_ORD_REBUILD
)
2634 /* no in-sync disks left in this mirror the
2638 return IMSM_T_STATE_FAILED
;
2641 return IMSM_T_STATE_DEGRADED
;
2645 return IMSM_T_STATE_DEGRADED
;
2647 return IMSM_T_STATE_FAILED
;
2653 return map
->map_state
;
2656 static int imsm_count_failed(struct intel_super
*super
, struct imsm_dev
*dev
)
2660 struct imsm_disk
*disk
;
2661 struct imsm_map
*map
= get_imsm_map(dev
, 0);
2663 for (i
= 0; i
< map
->num_members
; i
++) {
2664 __u32 ord
= get_imsm_ord_tbl_ent(dev
, i
);
2665 int idx
= ord_to_idx(ord
);
2667 disk
= get_imsm_disk(super
, idx
);
2668 if (!disk
|| disk
->status
& FAILED_DISK
||
2669 ord
& IMSM_ORD_REBUILD
)
2676 static int is_resyncing(struct imsm_dev
*dev
)
2678 struct imsm_map
*migr_map
;
2680 if (!dev
->vol
.migr_state
)
2683 if (dev
->vol
.migr_type
== MIGR_INIT
)
2686 migr_map
= get_imsm_map(dev
, 1);
2688 if (migr_map
->map_state
== IMSM_T_STATE_NORMAL
)
2694 static int is_rebuilding(struct imsm_dev
*dev
)
2696 struct imsm_map
*migr_map
;
2698 if (!dev
->vol
.migr_state
)
2701 if (dev
->vol
.migr_type
!= MIGR_REBUILD
)
2704 migr_map
= get_imsm_map(dev
, 1);
2706 if (migr_map
->map_state
== IMSM_T_STATE_DEGRADED
)
2712 static void mark_failure(struct imsm_disk
*disk
)
2714 if (disk
->status
& FAILED_DISK
)
2716 disk
->status
|= FAILED_DISK
;
2717 disk
->scsi_id
= __cpu_to_le32(~(__u32
)0);
2718 memmove(&disk
->serial
[0], &disk
->serial
[1], MAX_RAID_SERIAL_LEN
- 1);
2721 /* Handle dirty -> clean transititions and resync. Degraded and rebuild
2722 * states are handled in imsm_set_disk() with one exception, when a
2723 * resync is stopped due to a new failure this routine will set the
2724 * 'degraded' state for the array.
2726 static int imsm_set_array_state(struct active_array
*a
, int consistent
)
2728 int inst
= a
->info
.container_member
;
2729 struct intel_super
*super
= a
->container
->sb
;
2730 struct imsm_dev
*dev
= get_imsm_dev(super
, inst
);
2731 struct imsm_map
*map
= get_imsm_map(dev
, 0);
2732 int failed
= imsm_count_failed(super
, dev
);
2733 __u8 map_state
= imsm_check_degraded(super
, dev
, failed
);
2735 /* before we activate this array handle any missing disks */
2736 if (consistent
== 2 && super
->missing
) {
2739 dprintf("imsm: mark missing\n");
2740 end_migration(dev
, map_state
);
2741 for (dl
= super
->missing
; dl
; dl
= dl
->next
)
2742 mark_failure(&dl
->disk
);
2743 super
->updates_pending
++;
2746 if (consistent
== 2 &&
2747 (!is_resync_complete(a
) ||
2748 map_state
!= IMSM_T_STATE_NORMAL
||
2749 dev
->vol
.migr_state
))
2752 if (is_resync_complete(a
)) {
2753 /* complete intialization / resync,
2754 * recovery is completed in ->set_disk
2756 if (is_resyncing(dev
)) {
2757 dprintf("imsm: mark resync done\n");
2758 end_migration(dev
, map_state
);
2759 super
->updates_pending
++;
2761 } else if (!is_resyncing(dev
) && !failed
) {
2762 /* mark the start of the init process if nothing is failed */
2763 dprintf("imsm: mark resync start (%llu)\n", a
->resync_start
);
2764 if (map
->map_state
== IMSM_T_STATE_NORMAL
)
2765 migrate(dev
, IMSM_T_STATE_NORMAL
, MIGR_REBUILD
);
2767 migrate(dev
, IMSM_T_STATE_NORMAL
, MIGR_INIT
);
2768 super
->updates_pending
++;
2771 /* check if we can update the migration checkpoint */
2772 if (dev
->vol
.migr_state
&&
2773 __le32_to_cpu(dev
->vol
.curr_migr_unit
) != a
->resync_start
) {
2774 dprintf("imsm: checkpoint migration (%llu)\n", a
->resync_start
);
2775 dev
->vol
.curr_migr_unit
= __cpu_to_le32(a
->resync_start
);
2776 super
->updates_pending
++;
2779 /* mark dirty / clean */
2780 if (dev
->vol
.dirty
!= !consistent
) {
2781 dprintf("imsm: mark '%s' (%llu)\n",
2782 consistent
? "clean" : "dirty", a
->resync_start
);
2787 super
->updates_pending
++;
2792 static void imsm_set_disk(struct active_array
*a
, int n
, int state
)
2794 int inst
= a
->info
.container_member
;
2795 struct intel_super
*super
= a
->container
->sb
;
2796 struct imsm_dev
*dev
= get_imsm_dev(super
, inst
);
2797 struct imsm_map
*map
= get_imsm_map(dev
, 0);
2798 struct imsm_disk
*disk
;
2803 if (n
> map
->num_members
)
2804 fprintf(stderr
, "imsm: set_disk %d out of range 0..%d\n",
2805 n
, map
->num_members
- 1);
2810 dprintf("imsm: set_disk %d:%x\n", n
, state
);
2812 ord
= get_imsm_ord_tbl_ent(dev
, n
);
2813 disk
= get_imsm_disk(super
, ord_to_idx(ord
));
2815 /* check for new failures */
2816 if ((state
& DS_FAULTY
) && !(disk
->status
& FAILED_DISK
)) {
2818 super
->updates_pending
++;
2821 /* check if in_sync */
2822 if (state
& DS_INSYNC
&& ord
& IMSM_ORD_REBUILD
) {
2823 struct imsm_map
*migr_map
= get_imsm_map(dev
, 1);
2825 set_imsm_ord_tbl_ent(migr_map
, n
, ord_to_idx(ord
));
2826 super
->updates_pending
++;
2829 failed
= imsm_count_failed(super
, dev
);
2830 map_state
= imsm_check_degraded(super
, dev
, failed
);
2832 /* check if recovery complete, newly degraded, or failed */
2833 if (map_state
== IMSM_T_STATE_NORMAL
&& is_rebuilding(dev
)) {
2834 end_migration(dev
, map_state
);
2835 super
->updates_pending
++;
2836 } else if (map_state
== IMSM_T_STATE_DEGRADED
&&
2837 map
->map_state
!= map_state
&&
2838 !dev
->vol
.migr_state
) {
2839 dprintf("imsm: mark degraded\n");
2840 map
->map_state
= map_state
;
2841 super
->updates_pending
++;
2842 } else if (map_state
== IMSM_T_STATE_FAILED
&&
2843 map
->map_state
!= map_state
) {
2844 dprintf("imsm: mark failed\n");
2845 end_migration(dev
, map_state
);
2846 super
->updates_pending
++;
2850 static int store_imsm_mpb(int fd
, struct intel_super
*super
)
2852 struct imsm_super
*mpb
= super
->anchor
;
2853 __u32 mpb_size
= __le32_to_cpu(mpb
->mpb_size
);
2854 unsigned long long dsize
;
2855 unsigned long long sectors
;
2857 get_dev_size(fd
, NULL
, &dsize
);
2859 if (mpb_size
> 512) {
2860 /* -1 to account for anchor */
2861 sectors
= mpb_sectors(mpb
) - 1;
2863 /* write the extended mpb to the sectors preceeding the anchor */
2864 if (lseek64(fd
, dsize
- (512 * (2 + sectors
)), SEEK_SET
) < 0)
2867 if (write(fd
, super
->buf
+ 512, 512 * sectors
) != 512 * sectors
)
2871 /* first block is stored on second to last sector of the disk */
2872 if (lseek64(fd
, dsize
- (512 * 2), SEEK_SET
) < 0)
2875 if (write(fd
, super
->buf
, 512) != 512)
2881 static void imsm_sync_metadata(struct supertype
*container
)
2883 struct intel_super
*super
= container
->sb
;
2885 if (!super
->updates_pending
)
2888 write_super_imsm(super
, 0);
2890 super
->updates_pending
= 0;
2893 static struct dl
*imsm_readd(struct intel_super
*super
, int idx
, struct active_array
*a
)
2895 struct imsm_dev
*dev
= get_imsm_dev(super
, a
->info
.container_member
);
2896 int i
= get_imsm_disk_idx(dev
, idx
);
2899 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
2903 if (dl
&& dl
->disk
.status
& FAILED_DISK
)
2907 dprintf("%s: found %x:%x\n", __func__
, dl
->major
, dl
->minor
);
2912 static struct dl
*imsm_add_spare(struct intel_super
*super
, int slot
, struct active_array
*a
)
2914 struct imsm_dev
*dev
= get_imsm_dev(super
, a
->info
.container_member
);
2915 int idx
= get_imsm_disk_idx(dev
, slot
);
2916 struct imsm_map
*map
= get_imsm_map(dev
, 0);
2917 unsigned long long esize
;
2918 unsigned long long pos
;
2926 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
2927 /* If in this array, skip */
2928 for (d
= a
->info
.devs
; d
; d
= d
->next
)
2929 if (d
->state_fd
>= 0 &&
2930 d
->disk
.major
== dl
->major
&&
2931 d
->disk
.minor
== dl
->minor
) {
2932 dprintf("%x:%x already in array\n", dl
->major
, dl
->minor
);
2938 /* skip in use or failed drives */
2939 if (dl
->disk
.status
& FAILED_DISK
|| idx
== dl
->index
) {
2940 dprintf("%x:%x status ( %s%s)\n",
2941 dl
->major
, dl
->minor
,
2942 dl
->disk
.status
& FAILED_DISK
? "failed " : "",
2943 idx
== dl
->index
? "in use " : "");
2947 /* Does this unused device have the requisite free space?
2948 * We need a->info.component_size sectors
2950 ex
= get_extents(super
, dl
);
2952 dprintf("cannot get extents\n");
2958 array_start
= __le32_to_cpu(map
->pba_of_lba0
);
2961 /* check that we can start at pba_of_lba0 with
2962 * a->info.component_size of space
2964 esize
= ex
[j
].start
- pos
;
2965 if (array_start
>= pos
&&
2966 array_start
+ a
->info
.component_size
< ex
[j
].start
) {
2970 pos
= ex
[j
].start
+ ex
[j
].size
;
2973 } while (ex
[j
-1].size
);
2977 dprintf("%x:%x does not have %llu at %d\n",
2978 dl
->major
, dl
->minor
,
2979 a
->info
.component_size
,
2980 __le32_to_cpu(map
->pba_of_lba0
));
2990 static struct mdinfo
*imsm_activate_spare(struct active_array
*a
,
2991 struct metadata_update
**updates
)
2994 * Find a device with unused free space and use it to replace a
2995 * failed/vacant region in an array. We replace failed regions one a
2996 * array at a time. The result is that a new spare disk will be added
2997 * to the first failed array and after the monitor has finished
2998 * propagating failures the remainder will be consumed.
3000 * FIXME add a capability for mdmon to request spares from another
3004 struct intel_super
*super
= a
->container
->sb
;
3005 int inst
= a
->info
.container_member
;
3006 struct imsm_dev
*dev
= get_imsm_dev(super
, inst
);
3007 struct imsm_map
*map
= get_imsm_map(dev
, 0);
3008 int failed
= a
->info
.array
.raid_disks
;
3009 struct mdinfo
*rv
= NULL
;
3012 struct metadata_update
*mu
;
3014 struct imsm_update_activate_spare
*u
;
3018 for (d
= a
->info
.devs
; d
; d
= d
->next
) {
3019 if ((d
->curr_state
& DS_FAULTY
) &&
3021 /* wait for Removal to happen */
3023 if (d
->state_fd
>= 0)
3027 dprintf("imsm: activate spare: inst=%d failed=%d (%d) level=%d\n",
3028 inst
, failed
, a
->info
.array
.raid_disks
, a
->info
.array
.level
);
3029 if (imsm_check_degraded(super
, dev
, failed
) != IMSM_T_STATE_DEGRADED
)
3032 /* For each slot, if it is not working, find a spare */
3033 for (i
= 0; i
< a
->info
.array
.raid_disks
; i
++) {
3034 for (d
= a
->info
.devs
; d
; d
= d
->next
)
3035 if (d
->disk
.raid_disk
== i
)
3037 dprintf("found %d: %p %x\n", i
, d
, d
?d
->curr_state
:0);
3038 if (d
&& (d
->state_fd
>= 0))
3042 * OK, this device needs recovery. Try to re-add the previous
3043 * occupant of this slot, if this fails add a new spare
3045 dl
= imsm_readd(super
, i
, a
);
3047 dl
= imsm_add_spare(super
, i
, a
);
3051 /* found a usable disk with enough space */
3052 di
= malloc(sizeof(*di
));
3055 memset(di
, 0, sizeof(*di
));
3057 /* dl->index will be -1 in the case we are activating a
3058 * pristine spare. imsm_process_update() will create a
3059 * new index in this case. Once a disk is found to be
3060 * failed in all member arrays it is kicked from the
3063 di
->disk
.number
= dl
->index
;
3065 /* (ab)use di->devs to store a pointer to the device
3068 di
->devs
= (struct mdinfo
*) dl
;
3070 di
->disk
.raid_disk
= i
;
3071 di
->disk
.major
= dl
->major
;
3072 di
->disk
.minor
= dl
->minor
;
3074 di
->data_offset
= __le32_to_cpu(map
->pba_of_lba0
);
3075 di
->component_size
= a
->info
.component_size
;
3076 di
->container_member
= inst
;
3080 dprintf("%x:%x to be %d at %llu\n", dl
->major
, dl
->minor
,
3081 i
, di
->data_offset
);
3087 /* No spares found */
3089 /* Now 'rv' has a list of devices to return.
3090 * Create a metadata_update record to update the
3091 * disk_ord_tbl for the array
3093 mu
= malloc(sizeof(*mu
));
3095 mu
->buf
= malloc(sizeof(struct imsm_update_activate_spare
) * num_spares
);
3096 if (mu
->buf
== NULL
) {
3103 struct mdinfo
*n
= rv
->next
;
3112 mu
->len
= sizeof(struct imsm_update_activate_spare
) * num_spares
;
3113 mu
->next
= *updates
;
3114 u
= (struct imsm_update_activate_spare
*) mu
->buf
;
3116 for (di
= rv
; di
; di
= di
->next
) {
3117 u
->type
= update_activate_spare
;
3118 u
->dl
= (struct dl
*) di
->devs
;
3120 u
->slot
= di
->disk
.raid_disk
;
3131 static int disks_overlap(struct imsm_dev
*d1
, struct imsm_dev
*d2
)
3133 struct imsm_map
*m1
= get_imsm_map(d1
, 0);
3134 struct imsm_map
*m2
= get_imsm_map(d2
, 0);
3139 for (i
= 0; i
< m1
->num_members
; i
++) {
3140 idx
= get_imsm_disk_idx(d1
, i
);
3141 for (j
= 0; j
< m2
->num_members
; j
++)
3142 if (idx
== get_imsm_disk_idx(d2
, j
))
3149 static void imsm_delete(struct intel_super
*super
, struct dl
**dlp
, int index
);
3151 static void imsm_process_update(struct supertype
*st
,
3152 struct metadata_update
*update
)
3155 * crack open the metadata_update envelope to find the update record
3156 * update can be one of:
3157 * update_activate_spare - a spare device has replaced a failed
3158 * device in an array, update the disk_ord_tbl. If this disk is
3159 * present in all member arrays then also clear the SPARE_DISK
3162 struct intel_super
*super
= st
->sb
;
3163 struct imsm_super
*mpb
;
3164 enum imsm_update_type type
= *(enum imsm_update_type
*) update
->buf
;
3166 /* update requires a larger buf but the allocation failed */
3167 if (super
->next_len
&& !super
->next_buf
) {
3168 super
->next_len
= 0;
3172 if (super
->next_buf
) {
3173 memcpy(super
->next_buf
, super
->buf
, super
->len
);
3175 super
->len
= super
->next_len
;
3176 super
->buf
= super
->next_buf
;
3178 super
->next_len
= 0;
3179 super
->next_buf
= NULL
;
3182 mpb
= super
->anchor
;
3185 case update_activate_spare
: {
3186 struct imsm_update_activate_spare
*u
= (void *) update
->buf
;
3187 struct imsm_dev
*dev
= get_imsm_dev(super
, u
->array
);
3188 struct imsm_map
*map
= get_imsm_map(dev
, 0);
3189 struct imsm_map
*migr_map
;
3190 struct active_array
*a
;
3191 struct imsm_disk
*disk
;
3196 int victim
= get_imsm_disk_idx(dev
, u
->slot
);
3199 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
3204 fprintf(stderr
, "error: imsm_activate_spare passed "
3205 "an unknown disk (index: %d)\n",
3210 super
->updates_pending
++;
3212 /* count failures (excluding rebuilds and the victim)
3213 * to determine map[0] state
3216 for (i
= 0; i
< map
->num_members
; i
++) {
3219 disk
= get_imsm_disk(super
, get_imsm_disk_idx(dev
, i
));
3220 if (!disk
|| disk
->status
& FAILED_DISK
)
3224 /* adding a pristine spare, assign a new index */
3225 if (dl
->index
< 0) {
3226 dl
->index
= super
->anchor
->num_disks
;
3227 super
->anchor
->num_disks
++;
3230 disk
->status
|= CONFIGURED_DISK
;
3231 disk
->status
&= ~SPARE_DISK
;
3234 to_state
= imsm_check_degraded(super
, dev
, failed
);
3235 map
->map_state
= IMSM_T_STATE_DEGRADED
;
3236 migrate(dev
, to_state
, MIGR_REBUILD
);
3237 migr_map
= get_imsm_map(dev
, 1);
3238 set_imsm_ord_tbl_ent(map
, u
->slot
, dl
->index
);
3239 set_imsm_ord_tbl_ent(migr_map
, u
->slot
, dl
->index
| IMSM_ORD_REBUILD
);
3241 /* count arrays using the victim in the metadata */
3243 for (a
= st
->arrays
; a
; a
= a
->next
) {
3244 dev
= get_imsm_dev(super
, a
->info
.container_member
);
3245 for (i
= 0; i
< map
->num_members
; i
++)
3246 if (victim
== get_imsm_disk_idx(dev
, i
))
3250 /* delete the victim if it is no longer being
3256 /* We know that 'manager' isn't touching anything,
3257 * so it is safe to delete
3259 for (dlp
= &super
->disks
; *dlp
; dlp
= &(*dlp
)->next
)
3260 if ((*dlp
)->index
== victim
)
3263 /* victim may be on the missing list */
3265 for (dlp
= &super
->missing
; *dlp
; dlp
= &(*dlp
)->next
)
3266 if ((*dlp
)->index
== victim
)
3268 imsm_delete(super
, dlp
, victim
);
3272 case update_create_array
: {
3273 /* someone wants to create a new array, we need to be aware of
3274 * a few races/collisions:
3275 * 1/ 'Create' called by two separate instances of mdadm
3276 * 2/ 'Create' versus 'activate_spare': mdadm has chosen
3277 * devices that have since been assimilated via
3279 * In the event this update can not be carried out mdadm will
3280 * (FIX ME) notice that its update did not take hold.
3282 struct imsm_update_create_array
*u
= (void *) update
->buf
;
3283 struct imsm_dev
*dev
;
3284 struct imsm_map
*map
, *new_map
;
3285 unsigned long long start
, end
;
3286 unsigned long long new_start
, new_end
;
3290 /* handle racing creates: first come first serve */
3291 if (u
->dev_idx
< mpb
->num_raid_devs
) {
3292 dprintf("%s: subarray %d already defined\n",
3293 __func__
, u
->dev_idx
);
3297 /* check update is next in sequence */
3298 if (u
->dev_idx
!= mpb
->num_raid_devs
) {
3299 dprintf("%s: can not create array %d expected index %d\n",
3300 __func__
, u
->dev_idx
, mpb
->num_raid_devs
);
3304 new_map
= get_imsm_map(&u
->dev
, 0);
3305 new_start
= __le32_to_cpu(new_map
->pba_of_lba0
);
3306 new_end
= new_start
+ __le32_to_cpu(new_map
->blocks_per_member
);
3308 /* handle activate_spare versus create race:
3309 * check to make sure that overlapping arrays do not include
3312 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
3313 dev
= get_imsm_dev(super
, i
);
3314 map
= get_imsm_map(dev
, 0);
3315 start
= __le32_to_cpu(map
->pba_of_lba0
);
3316 end
= start
+ __le32_to_cpu(map
->blocks_per_member
);
3317 if ((new_start
>= start
&& new_start
<= end
) ||
3318 (start
>= new_start
&& start
<= new_end
))
3320 if (overlap
&& disks_overlap(dev
, &u
->dev
)) {
3321 dprintf("%s: arrays overlap\n", __func__
);
3325 /* check num_members sanity */
3326 if (new_map
->num_members
> mpb
->num_disks
) {
3327 dprintf("%s: num_disks out of range\n", __func__
);
3331 /* check that prepare update was successful */
3332 if (!update
->space
) {
3333 dprintf("%s: prepare update failed\n", __func__
);
3337 super
->updates_pending
++;
3338 dev
= update
->space
;
3339 map
= get_imsm_map(dev
, 0);
3340 update
->space
= NULL
;
3341 imsm_copy_dev(dev
, &u
->dev
);
3342 map
= get_imsm_map(dev
, 0);
3343 super
->dev_tbl
[u
->dev_idx
] = dev
;
3344 mpb
->num_raid_devs
++;
3347 for (i
= 0; i
< map
->num_members
; i
++) {
3348 struct imsm_disk
*disk
;
3350 disk
= get_imsm_disk(super
, get_imsm_disk_idx(dev
, i
));
3351 disk
->status
|= CONFIGURED_DISK
;
3352 disk
->status
&= ~SPARE_DISK
;
3355 imsm_update_version_info(super
);
3359 case update_add_disk
:
3361 /* we may be able to repair some arrays if disks are
3364 struct active_array
*a
;
3366 super
->updates_pending
++;
3367 for (a
= st
->arrays
; a
; a
= a
->next
)
3368 a
->check_degraded
= 1;
3370 /* add some spares to the metadata */
3371 while (super
->add
) {
3375 super
->add
= al
->next
;
3376 al
->next
= super
->disks
;
3378 dprintf("%s: added %x:%x\n",
3379 __func__
, al
->major
, al
->minor
);
3386 static void imsm_prepare_update(struct supertype
*st
,
3387 struct metadata_update
*update
)
3390 * Allocate space to hold new disk entries, raid-device entries or a new
3391 * mpb if necessary. The manager synchronously waits for updates to
3392 * complete in the monitor, so new mpb buffers allocated here can be
3393 * integrated by the monitor thread without worrying about live pointers
3394 * in the manager thread.
3396 enum imsm_update_type type
= *(enum imsm_update_type
*) update
->buf
;
3397 struct intel_super
*super
= st
->sb
;
3398 struct imsm_super
*mpb
= super
->anchor
;
3403 case update_create_array
: {
3404 struct imsm_update_create_array
*u
= (void *) update
->buf
;
3406 len
= sizeof_imsm_dev(&u
->dev
, 1);
3407 update
->space
= malloc(len
);
3414 /* check if we need a larger metadata buffer */
3415 if (super
->next_buf
)
3416 buf_len
= super
->next_len
;
3418 buf_len
= super
->len
;
3420 if (__le32_to_cpu(mpb
->mpb_size
) + len
> buf_len
) {
3421 /* ok we need a larger buf than what is currently allocated
3422 * if this allocation fails process_update will notice that
3423 * ->next_len is set and ->next_buf is NULL
3425 buf_len
= ROUND_UP(__le32_to_cpu(mpb
->mpb_size
) + len
, 512);
3426 if (super
->next_buf
)
3427 free(super
->next_buf
);
3429 super
->next_len
= buf_len
;
3430 if (posix_memalign(&super
->next_buf
, buf_len
, 512) != 0)
3431 super
->next_buf
= NULL
;
3435 /* must be called while manager is quiesced */
3436 static void imsm_delete(struct intel_super
*super
, struct dl
**dlp
, int index
)
3438 struct imsm_super
*mpb
= super
->anchor
;
3440 struct imsm_dev
*dev
;
3441 struct imsm_map
*map
;
3442 int i
, j
, num_members
;
3445 dprintf("%s: deleting device[%d] from imsm_super\n",
3448 /* shift all indexes down one */
3449 for (iter
= super
->disks
; iter
; iter
= iter
->next
)
3450 if (iter
->index
> index
)
3452 for (iter
= super
->missing
; iter
; iter
= iter
->next
)
3453 if (iter
->index
> index
)
3456 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
3457 dev
= get_imsm_dev(super
, i
);
3458 map
= get_imsm_map(dev
, 0);
3459 num_members
= map
->num_members
;
3460 for (j
= 0; j
< num_members
; j
++) {
3461 /* update ord entries being careful not to propagate
3462 * ord-flags to the first map
3464 ord
= get_imsm_ord_tbl_ent(dev
, j
);
3466 if (ord_to_idx(ord
) <= index
)
3469 map
= get_imsm_map(dev
, 0);
3470 set_imsm_ord_tbl_ent(map
, j
, ord_to_idx(ord
- 1));
3471 map
= get_imsm_map(dev
, 1);
3473 set_imsm_ord_tbl_ent(map
, j
, ord
- 1);
3478 super
->updates_pending
++;
3480 struct dl
*dl
= *dlp
;
3482 *dlp
= (*dlp
)->next
;
3483 __free_imsm_disk(dl
);
3486 #endif /* MDASSEMBLE */
3488 struct superswitch super_imsm
= {
3490 .examine_super
= examine_super_imsm
,
3491 .brief_examine_super
= brief_examine_super_imsm
,
3492 .detail_super
= detail_super_imsm
,
3493 .brief_detail_super
= brief_detail_super_imsm
,
3494 .write_init_super
= write_init_super_imsm
,
3495 .validate_geometry
= validate_geometry_imsm
,
3496 .add_to_super
= add_to_super_imsm
,
3498 .match_home
= match_home_imsm
,
3499 .uuid_from_super
= uuid_from_super_imsm
,
3500 .getinfo_super
= getinfo_super_imsm
,
3501 .update_super
= update_super_imsm
,
3503 .avail_size
= avail_size_imsm
,
3505 .compare_super
= compare_super_imsm
,
3507 .load_super
= load_super_imsm
,
3508 .init_super
= init_super_imsm
,
3509 .store_super
= store_zero_imsm
,
3510 .free_super
= free_super_imsm
,
3511 .match_metadata_desc
= match_metadata_desc_imsm
,
3512 .container_content
= container_content_imsm
,
3518 .open_new
= imsm_open_new
,
3519 .load_super
= load_super_imsm
,
3520 .set_array_state
= imsm_set_array_state
,
3521 .set_disk
= imsm_set_disk
,
3522 .sync_metadata
= imsm_sync_metadata
,
3523 .activate_spare
= imsm_activate_spare
,
3524 .process_update
= imsm_process_update
,
3525 .prepare_update
= imsm_prepare_update
,
3526 #endif /* MDASSEMBLE */