2 * mdadm - Intel(R) Matrix Storage Manager Support
4 * Copyright (C) 2002-2008 Intel Corporation
6 * This program is free software; you can redistribute it and/or modify it
7 * under the terms and conditions of the GNU General Public License,
8 * version 2, as published by the Free Software Foundation.
10 * This program is distributed in the hope it will be useful, but WITHOUT
11 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
12 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for
15 * You should have received a copy of the GNU General Public License along with
16 * this program; if not, write to the Free Software Foundation, Inc.,
17 * 51 Franklin St - Fifth Floor, Boston, MA 02110-1301 USA.
20 #define HAVE_STDINT_H 1
24 #include "platform-intel.h"
30 /* MPB == Metadata Parameter Block */
31 #define MPB_SIGNATURE "Intel Raid ISM Cfg Sig. "
32 #define MPB_SIG_LEN (strlen(MPB_SIGNATURE))
33 #define MPB_VERSION_RAID0 "1.0.00"
34 #define MPB_VERSION_RAID1 "1.1.00"
35 #define MPB_VERSION_MANY_VOLUMES_PER_ARRAY "1.2.00"
36 #define MPB_VERSION_3OR4_DISK_ARRAY "1.2.01"
37 #define MPB_VERSION_RAID5 "1.2.02"
38 #define MPB_VERSION_5OR6_DISK_ARRAY "1.2.04"
39 #define MPB_VERSION_CNG "1.2.06"
40 #define MPB_VERSION_ATTRIBS "1.3.00"
41 #define MAX_SIGNATURE_LENGTH 32
42 #define MAX_RAID_SERIAL_LEN 16
44 #define MPB_ATTRIB_CHECKSUM_VERIFY __cpu_to_le32(0x80000000)
45 #define MPB_ATTRIB_PM __cpu_to_le32(0x40000000)
46 #define MPB_ATTRIB_2TB __cpu_to_le32(0x20000000)
47 #define MPB_ATTRIB_RAID0 __cpu_to_le32(0x00000001)
48 #define MPB_ATTRIB_RAID1 __cpu_to_le32(0x00000002)
49 #define MPB_ATTRIB_RAID10 __cpu_to_le32(0x00000004)
50 #define MPB_ATTRIB_RAID1E __cpu_to_le32(0x00000008)
51 #define MPB_ATTRIB_RAID5 __cpu_to_le32(0x00000010)
52 #define MPB_ATTRIB_RAIDCNG __cpu_to_le32(0x00000020)
54 #define MPB_SECTOR_CNT 418
55 #define IMSM_RESERVED_SECTORS 4096
56 #define SECT_PER_MB_SHIFT 11
58 /* Disk configuration info. */
59 #define IMSM_MAX_DEVICES 255
61 __u8 serial
[MAX_RAID_SERIAL_LEN
];/* 0xD8 - 0xE7 ascii serial number */
62 __u32 total_blocks
; /* 0xE8 - 0xEB total blocks */
63 __u32 scsi_id
; /* 0xEC - 0xEF scsi ID */
64 #define SPARE_DISK __cpu_to_le32(0x01) /* Spare */
65 #define CONFIGURED_DISK __cpu_to_le32(0x02) /* Member of some RaidDev */
66 #define FAILED_DISK __cpu_to_le32(0x04) /* Permanent failure */
67 #define USABLE_DISK __cpu_to_le32(0x08) /* Fully usable unless FAILED_DISK is set */
68 __u32 status
; /* 0xF0 - 0xF3 */
69 __u32 owner_cfg_num
; /* which config 0,1,2... owns this disk */
70 #define IMSM_DISK_FILLERS 4
71 __u32 filler
[IMSM_DISK_FILLERS
]; /* 0xF4 - 0x107 MPB_DISK_FILLERS for future expansion */
74 /* RAID map configuration infos. */
76 __u32 pba_of_lba0
; /* start address of partition */
77 __u32 blocks_per_member
;/* blocks per member */
78 __u32 num_data_stripes
; /* number of data stripes */
79 __u16 blocks_per_strip
;
80 __u8 map_state
; /* Normal, Uninitialized, Degraded, Failed */
81 #define IMSM_T_STATE_NORMAL 0
82 #define IMSM_T_STATE_UNINITIALIZED 1
83 #define IMSM_T_STATE_DEGRADED 2
84 #define IMSM_T_STATE_FAILED 3
86 #define IMSM_T_RAID0 0
87 #define IMSM_T_RAID1 1
88 #define IMSM_T_RAID5 5 /* since metadata version 1.2.02 ? */
89 __u8 num_members
; /* number of member disks */
90 __u8 num_domains
; /* number of parity domains */
91 __u8 failed_disk_num
; /* valid only when state is degraded */
93 __u32 filler
[7]; /* expansion area */
94 #define IMSM_ORD_REBUILD (1 << 24)
95 __u32 disk_ord_tbl
[1]; /* disk_ord_tbl[num_members],
96 * top byte contains some flags
98 } __attribute__ ((packed
));
101 __u32 curr_migr_unit
;
102 __u32 checkpoint_id
; /* id to access curr_migr_unit */
103 __u8 migr_state
; /* Normal or Migrating */
105 #define MIGR_REBUILD 1
106 #define MIGR_VERIFY 2 /* analagous to echo check > sync_action */
107 #define MIGR_GEN_MIGR 3
108 #define MIGR_STATE_CHANGE 4
109 #define MIGR_REPAIR 5
110 __u8 migr_type
; /* Initializing, Rebuilding, ... */
112 __u8 fs_state
; /* fast-sync state for CnG (0xff == disabled) */
113 __u16 verify_errors
; /* number of mismatches */
114 __u16 bad_blocks
; /* number of bad blocks during verify */
116 struct imsm_map map
[1];
117 /* here comes another one if migr_state */
118 } __attribute__ ((packed
));
121 __u8 volume
[MAX_RAID_SERIAL_LEN
];
124 #define DEV_BOOTABLE __cpu_to_le32(0x01)
125 #define DEV_BOOT_DEVICE __cpu_to_le32(0x02)
126 #define DEV_READ_COALESCING __cpu_to_le32(0x04)
127 #define DEV_WRITE_COALESCING __cpu_to_le32(0x08)
128 #define DEV_LAST_SHUTDOWN_DIRTY __cpu_to_le32(0x10)
129 #define DEV_HIDDEN_AT_BOOT __cpu_to_le32(0x20)
130 #define DEV_CURRENTLY_HIDDEN __cpu_to_le32(0x40)
131 #define DEV_VERIFY_AND_FIX __cpu_to_le32(0x80)
132 #define DEV_MAP_STATE_UNINIT __cpu_to_le32(0x100)
133 #define DEV_NO_AUTO_RECOVERY __cpu_to_le32(0x200)
134 #define DEV_CLONE_N_GO __cpu_to_le32(0x400)
135 #define DEV_CLONE_MAN_SYNC __cpu_to_le32(0x800)
136 #define DEV_CNG_MASTER_DISK_NUM __cpu_to_le32(0x1000)
137 __u32 status
; /* Persistent RaidDev status */
138 __u32 reserved_blocks
; /* Reserved blocks at beginning of volume */
142 __u8 cng_master_disk
;
146 #define IMSM_DEV_FILLERS 10
147 __u32 filler
[IMSM_DEV_FILLERS
];
149 } __attribute__ ((packed
));
152 __u8 sig
[MAX_SIGNATURE_LENGTH
]; /* 0x00 - 0x1F */
153 __u32 check_sum
; /* 0x20 - 0x23 MPB Checksum */
154 __u32 mpb_size
; /* 0x24 - 0x27 Size of MPB */
155 __u32 family_num
; /* 0x28 - 0x2B Checksum from first time this config was written */
156 __u32 generation_num
; /* 0x2C - 0x2F Incremented each time this array's MPB is written */
157 __u32 error_log_size
; /* 0x30 - 0x33 in bytes */
158 __u32 attributes
; /* 0x34 - 0x37 */
159 __u8 num_disks
; /* 0x38 Number of configured disks */
160 __u8 num_raid_devs
; /* 0x39 Number of configured volumes */
161 __u8 error_log_pos
; /* 0x3A */
162 __u8 fill
[1]; /* 0x3B */
163 __u32 cache_size
; /* 0x3c - 0x40 in mb */
164 __u32 orig_family_num
; /* 0x40 - 0x43 original family num */
165 __u32 pwr_cycle_count
; /* 0x44 - 0x47 simulated power cycle count for array */
166 __u32 bbm_log_size
; /* 0x48 - 0x4B - size of bad Block Mgmt Log in bytes */
167 #define IMSM_FILLERS 35
168 __u32 filler
[IMSM_FILLERS
]; /* 0x4C - 0xD7 RAID_MPB_FILLERS */
169 struct imsm_disk disk
[1]; /* 0xD8 diskTbl[numDisks] */
170 /* here comes imsm_dev[num_raid_devs] */
171 /* here comes BBM logs */
172 } __attribute__ ((packed
));
174 #define BBM_LOG_MAX_ENTRIES 254
176 struct bbm_log_entry
{
177 __u64 defective_block_start
;
178 #define UNREADABLE 0xFFFFFFFF
179 __u32 spare_block_offset
;
180 __u16 remapped_marked_count
;
182 } __attribute__ ((__packed__
));
185 __u32 signature
; /* 0xABADB10C */
187 __u32 reserved_spare_block_count
; /* 0 */
188 __u32 reserved
; /* 0xFFFF */
189 __u64 first_spare_lba
;
190 struct bbm_log_entry mapped_block_entries
[BBM_LOG_MAX_ENTRIES
];
191 } __attribute__ ((__packed__
));
195 static char *map_state_str
[] = { "normal", "uninitialized", "degraded", "failed" };
198 static __u8
migr_type(struct imsm_dev
*dev
)
200 if (dev
->vol
.migr_type
== MIGR_VERIFY
&&
201 dev
->status
& DEV_VERIFY_AND_FIX
)
204 return dev
->vol
.migr_type
;
207 static void set_migr_type(struct imsm_dev
*dev
, __u8 migr_type
)
209 /* for compatibility with older oroms convert MIGR_REPAIR, into
210 * MIGR_VERIFY w/ DEV_VERIFY_AND_FIX status
212 if (migr_type
== MIGR_REPAIR
) {
213 dev
->vol
.migr_type
= MIGR_VERIFY
;
214 dev
->status
|= DEV_VERIFY_AND_FIX
;
216 dev
->vol
.migr_type
= migr_type
;
217 dev
->status
&= ~DEV_VERIFY_AND_FIX
;
221 static unsigned int sector_count(__u32 bytes
)
223 return ((bytes
+ (512-1)) & (~(512-1))) / 512;
226 static unsigned int mpb_sectors(struct imsm_super
*mpb
)
228 return sector_count(__le32_to_cpu(mpb
->mpb_size
));
232 struct imsm_dev
*dev
;
233 struct intel_dev
*next
;
237 /* internal representation of IMSM metadata */
240 void *buf
; /* O_DIRECT buffer for reading/writing metadata */
241 struct imsm_super
*anchor
; /* immovable parameters */
243 size_t len
; /* size of the 'buf' allocation */
244 void *next_buf
; /* for realloc'ing buf from the manager */
246 int updates_pending
; /* count of pending updates for mdmon */
247 int creating_imsm
; /* flag to indicate container creation */
248 int current_vol
; /* index of raid device undergoing creation */
249 __u32 create_offset
; /* common start for 'current_vol' */
250 __u32 random
; /* random data for seeding new family numbers */
251 struct intel_dev
*devlist
;
255 __u8 serial
[MAX_RAID_SERIAL_LEN
];
258 struct imsm_disk disk
;
261 struct extent
*e
; /* for determining freespace @ create */
262 int raiddisk
; /* slot to fill in autolayout */
264 struct dl
*add
; /* list of disks to add while mdmon active */
265 struct dl
*missing
; /* disks removed while we weren't looking */
266 struct bbm_log
*bbm_log
;
267 const char *hba
; /* device path of the raid controller for this metadata */
268 const struct imsm_orom
*orom
; /* platform firmware support */
272 unsigned long long start
, size
;
275 /* definition of messages passed to imsm_process_update */
276 enum imsm_update_type
{
277 update_activate_spare
,
282 struct imsm_update_activate_spare
{
283 enum imsm_update_type type
;
287 struct imsm_update_activate_spare
*next
;
291 __u8 serial
[MAX_RAID_SERIAL_LEN
];
294 struct imsm_update_create_array
{
295 enum imsm_update_type type
;
300 struct imsm_update_add_disk
{
301 enum imsm_update_type type
;
304 static struct supertype
*match_metadata_desc_imsm(char *arg
)
306 struct supertype
*st
;
308 if (strcmp(arg
, "imsm") != 0 &&
309 strcmp(arg
, "default") != 0
313 st
= malloc(sizeof(*st
));
314 memset(st
, 0, sizeof(*st
));
315 st
->ss
= &super_imsm
;
316 st
->max_devs
= IMSM_MAX_DEVICES
;
317 st
->minor_version
= 0;
323 static __u8
*get_imsm_version(struct imsm_super
*mpb
)
325 return &mpb
->sig
[MPB_SIG_LEN
];
329 /* retrieve a disk directly from the anchor when the anchor is known to be
330 * up-to-date, currently only at load time
332 static struct imsm_disk
*__get_imsm_disk(struct imsm_super
*mpb
, __u8 index
)
334 if (index
>= mpb
->num_disks
)
336 return &mpb
->disk
[index
];
340 /* retrieve a disk from the parsed metadata */
341 static struct imsm_disk
*get_imsm_disk(struct intel_super
*super
, __u8 index
)
345 for (d
= super
->disks
; d
; d
= d
->next
)
346 if (d
->index
== index
)
353 /* generate a checksum directly from the anchor when the anchor is known to be
354 * up-to-date, currently only at load or write_super after coalescing
356 static __u32
__gen_imsm_checksum(struct imsm_super
*mpb
)
358 __u32 end
= mpb
->mpb_size
/ sizeof(end
);
359 __u32
*p
= (__u32
*) mpb
;
363 sum
+= __le32_to_cpu(*p
);
367 return sum
- __le32_to_cpu(mpb
->check_sum
);
370 static size_t sizeof_imsm_map(struct imsm_map
*map
)
372 return sizeof(struct imsm_map
) + sizeof(__u32
) * (map
->num_members
- 1);
375 struct imsm_map
*get_imsm_map(struct imsm_dev
*dev
, int second_map
)
377 struct imsm_map
*map
= &dev
->vol
.map
[0];
379 if (second_map
&& !dev
->vol
.migr_state
)
381 else if (second_map
) {
384 return ptr
+ sizeof_imsm_map(map
);
390 /* return the size of the device.
391 * migr_state increases the returned size if map[0] were to be duplicated
393 static size_t sizeof_imsm_dev(struct imsm_dev
*dev
, int migr_state
)
395 size_t size
= sizeof(*dev
) - sizeof(struct imsm_map
) +
396 sizeof_imsm_map(get_imsm_map(dev
, 0));
398 /* migrating means an additional map */
399 if (dev
->vol
.migr_state
)
400 size
+= sizeof_imsm_map(get_imsm_map(dev
, 1));
402 size
+= sizeof_imsm_map(get_imsm_map(dev
, 0));
408 /* retrieve disk serial number list from a metadata update */
409 static struct disk_info
*get_disk_info(struct imsm_update_create_array
*update
)
412 struct disk_info
*inf
;
414 inf
= u
+ sizeof(*update
) - sizeof(struct imsm_dev
) +
415 sizeof_imsm_dev(&update
->dev
, 0);
421 static struct imsm_dev
*__get_imsm_dev(struct imsm_super
*mpb
, __u8 index
)
427 if (index
>= mpb
->num_raid_devs
)
430 /* devices start after all disks */
431 offset
= ((void *) &mpb
->disk
[mpb
->num_disks
]) - _mpb
;
433 for (i
= 0; i
<= index
; i
++)
435 return _mpb
+ offset
;
437 offset
+= sizeof_imsm_dev(_mpb
+ offset
, 0);
442 static struct imsm_dev
*get_imsm_dev(struct intel_super
*super
, __u8 index
)
444 struct intel_dev
*dv
;
446 if (index
>= super
->anchor
->num_raid_devs
)
448 for (dv
= super
->devlist
; dv
; dv
= dv
->next
)
449 if (dv
->index
== index
)
454 static __u32
get_imsm_ord_tbl_ent(struct imsm_dev
*dev
, int slot
)
456 struct imsm_map
*map
;
458 if (dev
->vol
.migr_state
)
459 map
= get_imsm_map(dev
, 1);
461 map
= get_imsm_map(dev
, 0);
463 /* top byte identifies disk under rebuild */
464 return __le32_to_cpu(map
->disk_ord_tbl
[slot
]);
467 #define ord_to_idx(ord) (((ord) << 8) >> 8)
468 static __u32
get_imsm_disk_idx(struct imsm_dev
*dev
, int slot
)
470 __u32 ord
= get_imsm_ord_tbl_ent(dev
, slot
);
472 return ord_to_idx(ord
);
475 static void set_imsm_ord_tbl_ent(struct imsm_map
*map
, int slot
, __u32 ord
)
477 map
->disk_ord_tbl
[slot
] = __cpu_to_le32(ord
);
480 static int get_imsm_disk_slot(struct imsm_map
*map
, int idx
)
485 for (slot
= 0; slot
< map
->num_members
; slot
++) {
486 ord
= __le32_to_cpu(map
->disk_ord_tbl
[slot
]);
487 if (ord_to_idx(ord
) == idx
)
494 static int get_imsm_raid_level(struct imsm_map
*map
)
496 if (map
->raid_level
== 1) {
497 if (map
->num_members
== 2)
503 return map
->raid_level
;
506 static int cmp_extent(const void *av
, const void *bv
)
508 const struct extent
*a
= av
;
509 const struct extent
*b
= bv
;
510 if (a
->start
< b
->start
)
512 if (a
->start
> b
->start
)
517 static int count_memberships(struct dl
*dl
, struct intel_super
*super
)
522 for (i
= 0; i
< super
->anchor
->num_raid_devs
; i
++) {
523 struct imsm_dev
*dev
= get_imsm_dev(super
, i
);
524 struct imsm_map
*map
= get_imsm_map(dev
, 0);
526 if (get_imsm_disk_slot(map
, dl
->index
) >= 0)
533 static struct extent
*get_extents(struct intel_super
*super
, struct dl
*dl
)
535 /* find a list of used extents on the given physical device */
536 struct extent
*rv
, *e
;
538 int memberships
= count_memberships(dl
, super
);
539 __u32 reservation
= MPB_SECTOR_CNT
+ IMSM_RESERVED_SECTORS
;
541 rv
= malloc(sizeof(struct extent
) * (memberships
+ 1));
546 for (i
= 0; i
< super
->anchor
->num_raid_devs
; i
++) {
547 struct imsm_dev
*dev
= get_imsm_dev(super
, i
);
548 struct imsm_map
*map
= get_imsm_map(dev
, 0);
550 if (get_imsm_disk_slot(map
, dl
->index
) >= 0) {
551 e
->start
= __le32_to_cpu(map
->pba_of_lba0
);
552 e
->size
= __le32_to_cpu(map
->blocks_per_member
);
556 qsort(rv
, memberships
, sizeof(*rv
), cmp_extent
);
558 /* determine the start of the metadata
559 * when no raid devices are defined use the default
560 * ...otherwise allow the metadata to truncate the value
561 * as is the case with older versions of imsm
564 struct extent
*last
= &rv
[memberships
- 1];
567 remainder
= __le32_to_cpu(dl
->disk
.total_blocks
) -
568 (last
->start
+ last
->size
);
569 /* round down to 1k block to satisfy precision of the kernel
573 /* make sure remainder is still sane */
574 if (remainder
< ROUND_UP(super
->len
, 512) >> 9)
575 remainder
= ROUND_UP(super
->len
, 512) >> 9;
576 if (reservation
> remainder
)
577 reservation
= remainder
;
579 e
->start
= __le32_to_cpu(dl
->disk
.total_blocks
) - reservation
;
584 /* try to determine how much space is reserved for metadata from
585 * the last get_extents() entry, otherwise fallback to the
588 static __u32
imsm_reserved_sectors(struct intel_super
*super
, struct dl
*dl
)
594 /* for spares just return a minimal reservation which will grow
595 * once the spare is picked up by an array
598 return MPB_SECTOR_CNT
;
600 e
= get_extents(super
, dl
);
602 return MPB_SECTOR_CNT
+ IMSM_RESERVED_SECTORS
;
604 /* scroll to last entry */
605 for (i
= 0; e
[i
].size
; i
++)
608 rv
= __le32_to_cpu(dl
->disk
.total_blocks
) - e
[i
].start
;
616 static void print_imsm_dev(struct imsm_dev
*dev
, char *uuid
, int disk_idx
)
620 struct imsm_map
*map
= get_imsm_map(dev
, 0);
624 printf("[%.16s]:\n", dev
->volume
);
625 printf(" UUID : %s\n", uuid
);
626 printf(" RAID Level : %d\n", get_imsm_raid_level(map
));
627 printf(" Members : %d\n", map
->num_members
);
628 slot
= get_imsm_disk_slot(map
, disk_idx
);
630 ord
= get_imsm_ord_tbl_ent(dev
, slot
);
631 printf(" This Slot : %d%s\n", slot
,
632 ord
& IMSM_ORD_REBUILD
? " (out-of-sync)" : "");
634 printf(" This Slot : ?\n");
635 sz
= __le32_to_cpu(dev
->size_high
);
637 sz
+= __le32_to_cpu(dev
->size_low
);
638 printf(" Array Size : %llu%s\n", (unsigned long long)sz
,
639 human_size(sz
* 512));
640 sz
= __le32_to_cpu(map
->blocks_per_member
);
641 printf(" Per Dev Size : %llu%s\n", (unsigned long long)sz
,
642 human_size(sz
* 512));
643 printf(" Sector Offset : %u\n",
644 __le32_to_cpu(map
->pba_of_lba0
));
645 printf(" Num Stripes : %u\n",
646 __le32_to_cpu(map
->num_data_stripes
));
647 printf(" Chunk Size : %u KiB\n",
648 __le16_to_cpu(map
->blocks_per_strip
) / 2);
649 printf(" Reserved : %d\n", __le32_to_cpu(dev
->reserved_blocks
));
650 printf(" Migrate State : %s", dev
->vol
.migr_state
? "migrating" : "idle\n");
651 if (dev
->vol
.migr_state
) {
652 if (migr_type(dev
) == MIGR_INIT
)
653 printf(": initializing\n");
654 else if (migr_type(dev
) == MIGR_REBUILD
)
655 printf(": rebuilding\n");
656 else if (migr_type(dev
) == MIGR_VERIFY
)
658 else if (migr_type(dev
) == MIGR_GEN_MIGR
)
659 printf(": general migration\n");
660 else if (migr_type(dev
) == MIGR_STATE_CHANGE
)
661 printf(": state change\n");
662 else if (migr_type(dev
) == MIGR_REPAIR
)
663 printf(": repair\n");
665 printf(": <unknown:%d>\n", migr_type(dev
));
667 printf(" Map State : %s", map_state_str
[map
->map_state
]);
668 if (dev
->vol
.migr_state
) {
669 struct imsm_map
*map
= get_imsm_map(dev
, 1);
670 printf(" <-- %s", map_state_str
[map
->map_state
]);
673 printf(" Dirty State : %s\n", dev
->vol
.dirty
? "dirty" : "clean");
676 static void print_imsm_disk(struct imsm_super
*mpb
, int index
, __u32 reserved
)
678 struct imsm_disk
*disk
= __get_imsm_disk(mpb
, index
);
679 char str
[MAX_RAID_SERIAL_LEN
+ 1];
687 snprintf(str
, MAX_RAID_SERIAL_LEN
+ 1, "%s", disk
->serial
);
688 printf(" Disk%02d Serial : %s\n", index
, str
);
690 printf(" State :%s%s%s%s\n", s
&SPARE_DISK
? " spare" : "",
691 s
&CONFIGURED_DISK
? " active" : "",
692 s
&FAILED_DISK
? " failed" : "",
693 s
&USABLE_DISK
? " usable" : "");
694 printf(" Id : %08x\n", __le32_to_cpu(disk
->scsi_id
));
695 sz
= __le32_to_cpu(disk
->total_blocks
) - reserved
;
696 printf(" Usable Size : %llu%s\n", (unsigned long long)sz
,
697 human_size(sz
* 512));
700 static void getinfo_super_imsm(struct supertype
*st
, struct mdinfo
*info
);
702 static void examine_super_imsm(struct supertype
*st
, char *homehost
)
704 struct intel_super
*super
= st
->sb
;
705 struct imsm_super
*mpb
= super
->anchor
;
706 char str
[MAX_SIGNATURE_LENGTH
];
711 __u32 reserved
= imsm_reserved_sectors(super
, super
->disks
);
714 snprintf(str
, MPB_SIG_LEN
, "%s", mpb
->sig
);
715 printf(" Magic : %s\n", str
);
716 snprintf(str
, strlen(MPB_VERSION_RAID0
), "%s", get_imsm_version(mpb
));
717 printf(" Version : %s\n", get_imsm_version(mpb
));
718 printf(" Orig Family : %08x\n", __le32_to_cpu(mpb
->orig_family_num
));
719 printf(" Family : %08x\n", __le32_to_cpu(mpb
->family_num
));
720 printf(" Generation : %08x\n", __le32_to_cpu(mpb
->generation_num
));
721 getinfo_super_imsm(st
, &info
);
722 fname_from_uuid(st
, &info
, nbuf
, ':');
723 printf(" UUID : %s\n", nbuf
+ 5);
724 sum
= __le32_to_cpu(mpb
->check_sum
);
725 printf(" Checksum : %08x %s\n", sum
,
726 __gen_imsm_checksum(mpb
) == sum
? "correct" : "incorrect");
727 printf(" MPB Sectors : %d\n", mpb_sectors(mpb
));
728 printf(" Disks : %d\n", mpb
->num_disks
);
729 printf(" RAID Devices : %d\n", mpb
->num_raid_devs
);
730 print_imsm_disk(mpb
, super
->disks
->index
, reserved
);
731 if (super
->bbm_log
) {
732 struct bbm_log
*log
= super
->bbm_log
;
735 printf("Bad Block Management Log:\n");
736 printf(" Log Size : %d\n", __le32_to_cpu(mpb
->bbm_log_size
));
737 printf(" Signature : %x\n", __le32_to_cpu(log
->signature
));
738 printf(" Entry Count : %d\n", __le32_to_cpu(log
->entry_count
));
739 printf(" Spare Blocks : %d\n", __le32_to_cpu(log
->reserved_spare_block_count
));
740 printf(" First Spare : %llx\n",
741 (unsigned long long) __le64_to_cpu(log
->first_spare_lba
));
743 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
745 struct imsm_dev
*dev
= __get_imsm_dev(mpb
, i
);
747 super
->current_vol
= i
;
748 getinfo_super_imsm(st
, &info
);
749 fname_from_uuid(st
, &info
, nbuf
, ':');
750 print_imsm_dev(dev
, nbuf
+ 5, super
->disks
->index
);
752 for (i
= 0; i
< mpb
->num_disks
; i
++) {
753 if (i
== super
->disks
->index
)
755 print_imsm_disk(mpb
, i
, reserved
);
759 static void brief_examine_super_imsm(struct supertype
*st
, int verbose
)
761 /* We just write a generic IMSM ARRAY entry */
765 struct intel_super
*super
= st
->sb
;
768 if (!super
->anchor
->num_raid_devs
) {
769 printf("ARRAY metadata=imsm\n");
773 getinfo_super_imsm(st
, &info
);
774 fname_from_uuid(st
, &info
, nbuf
, ':');
775 for (i
= 0; i
< super
->anchor
->num_raid_devs
; i
++) {
776 struct imsm_dev
*dev
= get_imsm_dev(super
, i
);
778 super
->current_vol
= i
;
779 getinfo_super_imsm(st
, &info
);
780 fname_from_uuid(st
, &info
, nbuf1
, ':');
781 printf("ARRAY /dev/md/%.16s container=%s member=%d UUID=%s\n",
782 dev
->volume
, nbuf
+ 5, i
, nbuf1
+ 5);
784 printf("ARRAY metadata=imsm UUID=%s\n", nbuf
+ 5);
787 static void export_examine_super_imsm(struct supertype
*st
)
789 struct intel_super
*super
= st
->sb
;
790 struct imsm_super
*mpb
= super
->anchor
;
794 getinfo_super_imsm(st
, &info
);
795 fname_from_uuid(st
, &info
, nbuf
, ':');
796 printf("MD_METADATA=imsm\n");
797 printf("MD_LEVEL=container\n");
798 printf("MD_UUID=%s\n", nbuf
+5);
799 printf("MD_DEVICES=%u\n", mpb
->num_disks
);
802 static void detail_super_imsm(struct supertype
*st
, char *homehost
)
807 getinfo_super_imsm(st
, &info
);
808 fname_from_uuid(st
, &info
, nbuf
, ':');
809 printf("\n UUID : %s\n", nbuf
+ 5);
812 static void brief_detail_super_imsm(struct supertype
*st
)
816 getinfo_super_imsm(st
, &info
);
817 fname_from_uuid(st
, &info
, nbuf
, ':');
818 printf(" UUID=%s", nbuf
+ 5);
821 static int imsm_read_serial(int fd
, char *devname
, __u8
*serial
);
822 static void fd2devname(int fd
, char *name
);
824 static int imsm_enumerate_ports(const char *hba_path
, int port_count
, int host_base
, int verbose
)
826 /* dump an unsorted list of devices attached to ahci, as well as
827 * non-connected ports
829 int hba_len
= strlen(hba_path
) + 1;
834 unsigned long port_mask
= (1 << port_count
) - 1;
836 if (port_count
> sizeof(port_mask
) * 8) {
838 fprintf(stderr
, Name
": port_count %d out of range\n", port_count
);
842 /* scroll through /sys/dev/block looking for devices attached to
845 dir
= opendir("/sys/dev/block");
846 for (ent
= dir
? readdir(dir
) : NULL
; ent
; ent
= readdir(dir
)) {
857 if (sscanf(ent
->d_name
, "%d:%d", &major
, &minor
) != 2)
859 path
= devt_to_devpath(makedev(major
, minor
));
862 if (!path_attached_to_hba(path
, hba_path
)) {
868 /* retrieve the scsi device type */
869 if (asprintf(&device
, "/sys/dev/block/%d:%d/device/xxxxxxx", major
, minor
) < 0) {
871 fprintf(stderr
, Name
": failed to allocate 'device'\n");
875 sprintf(device
, "/sys/dev/block/%d:%d/device/type", major
, minor
);
876 if (load_sys(device
, buf
) != 0) {
878 fprintf(stderr
, Name
": failed to read device type for %s\n",
884 type
= strtoul(buf
, NULL
, 10);
886 /* if it's not a disk print the vendor and model */
887 if (!(type
== 0 || type
== 7 || type
== 14)) {
890 sprintf(device
, "/sys/dev/block/%d:%d/device/vendor", major
, minor
);
891 if (load_sys(device
, buf
) == 0) {
892 strncpy(vendor
, buf
, sizeof(vendor
));
893 vendor
[sizeof(vendor
) - 1] = '\0';
894 c
= (char *) &vendor
[sizeof(vendor
) - 1];
895 while (isspace(*c
) || *c
== '\0')
899 sprintf(device
, "/sys/dev/block/%d:%d/device/model", major
, minor
);
900 if (load_sys(device
, buf
) == 0) {
901 strncpy(model
, buf
, sizeof(model
));
902 model
[sizeof(model
) - 1] = '\0';
903 c
= (char *) &model
[sizeof(model
) - 1];
904 while (isspace(*c
) || *c
== '\0')
908 if (vendor
[0] && model
[0])
909 sprintf(buf
, "%.64s %.64s", vendor
, model
);
911 switch (type
) { /* numbers from hald/linux/device.c */
912 case 1: sprintf(buf
, "tape"); break;
913 case 2: sprintf(buf
, "printer"); break;
914 case 3: sprintf(buf
, "processor"); break;
916 case 5: sprintf(buf
, "cdrom"); break;
917 case 6: sprintf(buf
, "scanner"); break;
918 case 8: sprintf(buf
, "media_changer"); break;
919 case 9: sprintf(buf
, "comm"); break;
920 case 12: sprintf(buf
, "raid"); break;
921 default: sprintf(buf
, "unknown");
927 /* chop device path to 'host%d' and calculate the port number */
928 c
= strchr(&path
[hba_len
], '/');
930 if (sscanf(&path
[hba_len
], "host%d", &port
) == 1)
934 *c
= '/'; /* repair the full string */
935 fprintf(stderr
, Name
": failed to determine port number for %s\n",
942 /* mark this port as used */
943 port_mask
&= ~(1 << port
);
945 /* print out the device information */
947 printf(" Port%d : - non-disk device (%s) -\n", port
, buf
);
951 fd
= dev_open(ent
->d_name
, O_RDONLY
);
953 printf(" Port%d : - disk info unavailable -\n", port
);
956 printf(" Port%d : %s", port
, buf
);
957 if (imsm_read_serial(fd
, NULL
, (__u8
*) buf
) == 0)
958 printf(" (%s)\n", buf
);
973 for (i
= 0; i
< port_count
; i
++)
974 if (port_mask
& (1 << i
))
975 printf(" Port%d : - no device attached -\n", i
);
981 static int detail_platform_imsm(int verbose
, int enumerate_only
)
983 /* There are two components to imsm platform support, the ahci SATA
984 * controller and the option-rom. To find the SATA controller we
985 * simply look in /sys/bus/pci/drivers/ahci to see if an ahci
986 * controller with the Intel vendor id is present. This approach
987 * allows mdadm to leverage the kernel's ahci detection logic, with the
988 * caveat that if ahci.ko is not loaded mdadm will not be able to
989 * detect platform raid capabilities. The option-rom resides in a
990 * platform "Adapter ROM". We scan for its signature to retrieve the
991 * platform capabilities. If raid support is disabled in the BIOS the
992 * option-rom capability structure will not be available.
994 const struct imsm_orom
*orom
;
995 struct sys_dev
*list
, *hba
;
998 const char *hba_path
;
1002 if (enumerate_only
) {
1003 if (check_env("IMSM_NO_PLATFORM") || find_imsm_orom())
1008 list
= find_driver_devices("pci", "ahci");
1009 for (hba
= list
; hba
; hba
= hba
->next
)
1010 if (devpath_to_vendor(hba
->path
) == 0x8086)
1015 fprintf(stderr
, Name
": unable to find active ahci controller\n");
1016 free_sys_dev(&list
);
1019 fprintf(stderr
, Name
": found Intel SATA AHCI Controller\n");
1020 hba_path
= hba
->path
;
1022 free_sys_dev(&list
);
1024 orom
= find_imsm_orom();
1027 fprintf(stderr
, Name
": imsm option-rom not found\n");
1031 printf(" Platform : Intel(R) Matrix Storage Manager\n");
1032 printf(" Version : %d.%d.%d.%d\n", orom
->major_ver
, orom
->minor_ver
,
1033 orom
->hotfix_ver
, orom
->build
);
1034 printf(" RAID Levels :%s%s%s%s%s\n",
1035 imsm_orom_has_raid0(orom
) ? " raid0" : "",
1036 imsm_orom_has_raid1(orom
) ? " raid1" : "",
1037 imsm_orom_has_raid1e(orom
) ? " raid1e" : "",
1038 imsm_orom_has_raid10(orom
) ? " raid10" : "",
1039 imsm_orom_has_raid5(orom
) ? " raid5" : "");
1040 printf(" Chunk Sizes :%s%s%s%s%s%s%s%s%s%s%s%s%s%s%s%s\n",
1041 imsm_orom_has_chunk(orom
, 2) ? " 2k" : "",
1042 imsm_orom_has_chunk(orom
, 4) ? " 4k" : "",
1043 imsm_orom_has_chunk(orom
, 8) ? " 8k" : "",
1044 imsm_orom_has_chunk(orom
, 16) ? " 16k" : "",
1045 imsm_orom_has_chunk(orom
, 32) ? " 32k" : "",
1046 imsm_orom_has_chunk(orom
, 64) ? " 64k" : "",
1047 imsm_orom_has_chunk(orom
, 128) ? " 128k" : "",
1048 imsm_orom_has_chunk(orom
, 256) ? " 256k" : "",
1049 imsm_orom_has_chunk(orom
, 512) ? " 512k" : "",
1050 imsm_orom_has_chunk(orom
, 1024*1) ? " 1M" : "",
1051 imsm_orom_has_chunk(orom
, 1024*2) ? " 2M" : "",
1052 imsm_orom_has_chunk(orom
, 1024*4) ? " 4M" : "",
1053 imsm_orom_has_chunk(orom
, 1024*8) ? " 8M" : "",
1054 imsm_orom_has_chunk(orom
, 1024*16) ? " 16M" : "",
1055 imsm_orom_has_chunk(orom
, 1024*32) ? " 32M" : "",
1056 imsm_orom_has_chunk(orom
, 1024*64) ? " 64M" : "");
1057 printf(" Max Disks : %d\n", orom
->tds
);
1058 printf(" Max Volumes : %d\n", orom
->vpa
);
1059 printf(" I/O Controller : %s\n", hba_path
);
1061 /* find the smallest scsi host number to determine a port number base */
1062 dir
= opendir(hba_path
);
1063 for (ent
= dir
? readdir(dir
) : NULL
; ent
; ent
= readdir(dir
)) {
1066 if (sscanf(ent
->d_name
, "host%d", &host
) != 1)
1068 if (port_count
== 0)
1070 else if (host
< host_base
)
1073 if (host
+ 1 > port_count
+ host_base
)
1074 port_count
= host
+ 1 - host_base
;
1080 if (!port_count
|| imsm_enumerate_ports(hba_path
, port_count
,
1081 host_base
, verbose
) != 0) {
1083 fprintf(stderr
, Name
": failed to enumerate ports\n");
1091 static int match_home_imsm(struct supertype
*st
, char *homehost
)
1093 /* the imsm metadata format does not specify any host
1094 * identification information. We return -1 since we can never
1095 * confirm nor deny whether a given array is "meant" for this
1096 * host. We rely on compare_super and the 'family_num' fields to
1097 * exclude member disks that do not belong, and we rely on
1098 * mdadm.conf to specify the arrays that should be assembled.
1099 * Auto-assembly may still pick up "foreign" arrays.
1105 static void uuid_from_super_imsm(struct supertype
*st
, int uuid
[4])
1107 /* The uuid returned here is used for:
1108 * uuid to put into bitmap file (Create, Grow)
1109 * uuid for backup header when saving critical section (Grow)
1110 * comparing uuids when re-adding a device into an array
1111 * In these cases the uuid required is that of the data-array,
1112 * not the device-set.
1113 * uuid to recognise same set when adding a missing device back
1114 * to an array. This is a uuid for the device-set.
1116 * For each of these we can make do with a truncated
1117 * or hashed uuid rather than the original, as long as
1119 * In each case the uuid required is that of the data-array,
1120 * not the device-set.
1122 /* imsm does not track uuid's so we synthesis one using sha1 on
1123 * - The signature (Which is constant for all imsm array, but no matter)
1124 * - the orig_family_num of the container
1125 * - the index number of the volume
1126 * - the 'serial' number of the volume.
1127 * Hopefully these are all constant.
1129 struct intel_super
*super
= st
->sb
;
1132 struct sha1_ctx ctx
;
1133 struct imsm_dev
*dev
= NULL
;
1136 /* some mdadm versions failed to set ->orig_family_num, in which
1137 * case fall back to ->family_num. orig_family_num will be
1138 * fixed up with the first metadata update.
1140 family_num
= super
->anchor
->orig_family_num
;
1141 if (family_num
== 0)
1142 family_num
= super
->anchor
->family_num
;
1143 sha1_init_ctx(&ctx
);
1144 sha1_process_bytes(super
->anchor
->sig
, MPB_SIG_LEN
, &ctx
);
1145 sha1_process_bytes(&family_num
, sizeof(__u32
), &ctx
);
1146 if (super
->current_vol
>= 0)
1147 dev
= get_imsm_dev(super
, super
->current_vol
);
1149 __u32 vol
= super
->current_vol
;
1150 sha1_process_bytes(&vol
, sizeof(vol
), &ctx
);
1151 sha1_process_bytes(dev
->volume
, MAX_RAID_SERIAL_LEN
, &ctx
);
1153 sha1_finish_ctx(&ctx
, buf
);
1154 memcpy(uuid
, buf
, 4*4);
1159 get_imsm_numerical_version(struct imsm_super
*mpb
, int *m
, int *p
)
1161 __u8
*v
= get_imsm_version(mpb
);
1162 __u8
*end
= mpb
->sig
+ MAX_SIGNATURE_LENGTH
;
1163 char major
[] = { 0, 0, 0 };
1164 char minor
[] = { 0 ,0, 0 };
1165 char patch
[] = { 0, 0, 0 };
1166 char *ver_parse
[] = { major
, minor
, patch
};
1170 while (*v
!= '\0' && v
< end
) {
1171 if (*v
!= '.' && j
< 2)
1172 ver_parse
[i
][j
++] = *v
;
1180 *m
= strtol(minor
, NULL
, 0);
1181 *p
= strtol(patch
, NULL
, 0);
1185 static int imsm_level_to_layout(int level
)
1193 return ALGORITHM_LEFT_ASYMMETRIC
;
1200 static void getinfo_super_imsm_volume(struct supertype
*st
, struct mdinfo
*info
)
1202 struct intel_super
*super
= st
->sb
;
1203 struct imsm_dev
*dev
= get_imsm_dev(super
, super
->current_vol
);
1204 struct imsm_map
*map
= get_imsm_map(dev
, 0);
1207 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
1208 if (dl
->raiddisk
== info
->disk
.raid_disk
)
1210 info
->container_member
= super
->current_vol
;
1211 info
->array
.raid_disks
= map
->num_members
;
1212 info
->array
.level
= get_imsm_raid_level(map
);
1213 info
->array
.layout
= imsm_level_to_layout(info
->array
.level
);
1214 info
->array
.md_minor
= -1;
1215 info
->array
.ctime
= 0;
1216 info
->array
.utime
= 0;
1217 info
->array
.chunk_size
= __le16_to_cpu(map
->blocks_per_strip
) << 9;
1218 info
->array
.state
= !dev
->vol
.dirty
;
1219 info
->custom_array_size
= __le32_to_cpu(dev
->size_high
);
1220 info
->custom_array_size
<<= 32;
1221 info
->custom_array_size
|= __le32_to_cpu(dev
->size_low
);
1223 info
->disk
.major
= 0;
1224 info
->disk
.minor
= 0;
1226 info
->disk
.major
= dl
->major
;
1227 info
->disk
.minor
= dl
->minor
;
1230 info
->data_offset
= __le32_to_cpu(map
->pba_of_lba0
);
1231 info
->component_size
= __le32_to_cpu(map
->blocks_per_member
);
1232 memset(info
->uuid
, 0, sizeof(info
->uuid
));
1234 if (map
->map_state
== IMSM_T_STATE_UNINITIALIZED
|| dev
->vol
.dirty
)
1235 info
->resync_start
= 0;
1236 else if (dev
->vol
.migr_state
)
1237 /* FIXME add curr_migr_unit to resync_start conversion */
1238 info
->resync_start
= 0;
1240 info
->resync_start
= ~0ULL;
1242 strncpy(info
->name
, (char *) dev
->volume
, MAX_RAID_SERIAL_LEN
);
1243 info
->name
[MAX_RAID_SERIAL_LEN
] = 0;
1245 info
->array
.major_version
= -1;
1246 info
->array
.minor_version
= -2;
1247 sprintf(info
->text_version
, "/%s/%d",
1248 devnum2devname(st
->container_dev
),
1249 info
->container_member
);
1250 info
->safe_mode_delay
= 4000; /* 4 secs like the Matrix driver */
1251 uuid_from_super_imsm(st
, info
->uuid
);
1254 /* check the config file to see if we can return a real uuid for this spare */
1255 static void fixup_container_spare_uuid(struct mdinfo
*inf
)
1257 struct mddev_ident_s
*array_list
;
1259 if (inf
->array
.level
!= LEVEL_CONTAINER
||
1260 memcmp(inf
->uuid
, uuid_match_any
, sizeof(int[4])) != 0)
1263 array_list
= conf_get_ident(NULL
);
1265 for (; array_list
; array_list
= array_list
->next
) {
1266 if (array_list
->uuid_set
) {
1267 struct supertype
*_sst
; /* spare supertype */
1268 struct supertype
*_cst
; /* container supertype */
1270 _cst
= array_list
->st
;
1272 _sst
= _cst
->ss
->match_metadata_desc(inf
->text_version
);
1277 memcpy(inf
->uuid
, array_list
->uuid
, sizeof(int[4]));
1285 static void getinfo_super_imsm(struct supertype
*st
, struct mdinfo
*info
)
1287 struct intel_super
*super
= st
->sb
;
1288 struct imsm_disk
*disk
;
1291 if (super
->current_vol
>= 0) {
1292 getinfo_super_imsm_volume(st
, info
);
1296 /* Set raid_disks to zero so that Assemble will always pull in valid
1299 info
->array
.raid_disks
= 0;
1300 info
->array
.level
= LEVEL_CONTAINER
;
1301 info
->array
.layout
= 0;
1302 info
->array
.md_minor
= -1;
1303 info
->array
.ctime
= 0; /* N/A for imsm */
1304 info
->array
.utime
= 0;
1305 info
->array
.chunk_size
= 0;
1307 info
->disk
.major
= 0;
1308 info
->disk
.minor
= 0;
1309 info
->disk
.raid_disk
= -1;
1310 info
->reshape_active
= 0;
1311 info
->array
.major_version
= -1;
1312 info
->array
.minor_version
= -2;
1313 strcpy(info
->text_version
, "imsm");
1314 info
->safe_mode_delay
= 0;
1315 info
->disk
.number
= -1;
1316 info
->disk
.state
= 0;
1320 __u32 reserved
= imsm_reserved_sectors(super
, super
->disks
);
1322 disk
= &super
->disks
->disk
;
1323 info
->data_offset
= __le32_to_cpu(disk
->total_blocks
) - reserved
;
1324 info
->component_size
= reserved
;
1326 info
->disk
.state
= s
& CONFIGURED_DISK
? (1 << MD_DISK_ACTIVE
) : 0;
1327 /* we don't change info->disk.raid_disk here because
1328 * this state will be finalized in mdmon after we have
1329 * found the 'most fresh' version of the metadata
1331 info
->disk
.state
|= s
& FAILED_DISK
? (1 << MD_DISK_FAULTY
) : 0;
1332 info
->disk
.state
|= s
& SPARE_DISK
? 0 : (1 << MD_DISK_SYNC
);
1335 /* only call uuid_from_super_imsm when this disk is part of a populated container,
1336 * ->compare_super may have updated the 'num_raid_devs' field for spares
1338 if (info
->disk
.state
& (1 << MD_DISK_SYNC
) || super
->anchor
->num_raid_devs
)
1339 uuid_from_super_imsm(st
, info
->uuid
);
1341 memcpy(info
->uuid
, uuid_match_any
, sizeof(int[4]));
1342 fixup_container_spare_uuid(info
);
1346 static int update_super_imsm(struct supertype
*st
, struct mdinfo
*info
,
1347 char *update
, char *devname
, int verbose
,
1348 int uuid_set
, char *homehost
)
1352 /* For 'assemble' and 'force' we need to return non-zero if any
1353 * change was made. For others, the return value is ignored.
1354 * Update options are:
1355 * force-one : This device looks a bit old but needs to be included,
1356 * update age info appropriately.
1357 * assemble: clear any 'faulty' flag to allow this device to
1359 * force-array: Array is degraded but being forced, mark it clean
1360 * if that will be needed to assemble it.
1362 * newdev: not used ????
1363 * grow: Array has gained a new device - this is currently for
1365 * resync: mark as dirty so a resync will happen.
1366 * name: update the name - preserving the homehost
1368 * Following are not relevant for this imsm:
1369 * sparc2.2 : update from old dodgey metadata
1370 * super-minor: change the preferred_minor number
1371 * summaries: update redundant counters.
1372 * uuid: Change the uuid of the array to match watch is given
1373 * homehost: update the recorded homehost
1374 * _reshape_progress: record new reshape_progress position.
1377 //struct intel_super *super = st->sb;
1378 //struct imsm_super *mpb = super->mpb;
1380 if (strcmp(update
, "grow") == 0) {
1382 if (strcmp(update
, "resync") == 0) {
1383 /* dev->vol.dirty = 1; */
1386 /* IMSM has no concept of UUID or homehost */
1391 static size_t disks_to_mpb_size(int disks
)
1395 size
= sizeof(struct imsm_super
);
1396 size
+= (disks
- 1) * sizeof(struct imsm_disk
);
1397 size
+= 2 * sizeof(struct imsm_dev
);
1398 /* up to 2 maps per raid device (-2 for imsm_maps in imsm_dev */
1399 size
+= (4 - 2) * sizeof(struct imsm_map
);
1400 /* 4 possible disk_ord_tbl's */
1401 size
+= 4 * (disks
- 1) * sizeof(__u32
);
1406 static __u64
avail_size_imsm(struct supertype
*st
, __u64 devsize
)
1408 if (devsize
< (MPB_SECTOR_CNT
+ IMSM_RESERVED_SECTORS
))
1411 return devsize
- (MPB_SECTOR_CNT
+ IMSM_RESERVED_SECTORS
);
1414 static void free_devlist(struct intel_super
*super
)
1416 struct intel_dev
*dv
;
1418 while (super
->devlist
) {
1419 dv
= super
->devlist
->next
;
1420 free(super
->devlist
->dev
);
1421 free(super
->devlist
);
1422 super
->devlist
= dv
;
1426 static void imsm_copy_dev(struct imsm_dev
*dest
, struct imsm_dev
*src
)
1428 memcpy(dest
, src
, sizeof_imsm_dev(src
, 0));
1431 static int compare_super_imsm(struct supertype
*st
, struct supertype
*tst
)
1435 * 0 same, or first was empty, and second was copied
1436 * 1 second had wrong number
1438 * 3 wrong other info
1440 struct intel_super
*first
= st
->sb
;
1441 struct intel_super
*sec
= tst
->sb
;
1449 if (memcmp(first
->anchor
->sig
, sec
->anchor
->sig
, MAX_SIGNATURE_LENGTH
) != 0)
1452 /* if an anchor does not have num_raid_devs set then it is a free
1455 if (first
->anchor
->num_raid_devs
> 0 &&
1456 sec
->anchor
->num_raid_devs
> 0) {
1457 if (first
->anchor
->orig_family_num
!= sec
->anchor
->orig_family_num
||
1458 first
->anchor
->family_num
!= sec
->anchor
->family_num
)
1462 /* if 'first' is a spare promote it to a populated mpb with sec's
1465 if (first
->anchor
->num_raid_devs
== 0 &&
1466 sec
->anchor
->num_raid_devs
> 0) {
1468 struct intel_dev
*dv
;
1469 struct imsm_dev
*dev
;
1471 /* we need to copy raid device info from sec if an allocation
1472 * fails here we don't associate the spare
1474 for (i
= 0; i
< sec
->anchor
->num_raid_devs
; i
++) {
1475 dv
= malloc(sizeof(*dv
));
1478 dev
= malloc(sizeof_imsm_dev(get_imsm_dev(sec
, i
), 1));
1485 dv
->next
= first
->devlist
;
1486 first
->devlist
= dv
;
1488 if (i
<= sec
->anchor
->num_raid_devs
) {
1489 /* allocation failure */
1490 free_devlist(first
);
1491 fprintf(stderr
, "imsm: failed to associate spare\n");
1494 for (i
= 0; i
< sec
->anchor
->num_raid_devs
; i
++)
1495 imsm_copy_dev(get_imsm_dev(first
, i
), get_imsm_dev(sec
, i
));
1497 first
->anchor
->num_raid_devs
= sec
->anchor
->num_raid_devs
;
1498 first
->anchor
->orig_family_num
= sec
->anchor
->orig_family_num
;
1499 first
->anchor
->family_num
= sec
->anchor
->family_num
;
1505 static void fd2devname(int fd
, char *name
)
1514 if (fstat(fd
, &st
) != 0)
1516 sprintf(path
, "/sys/dev/block/%d:%d",
1517 major(st
.st_rdev
), minor(st
.st_rdev
));
1519 rv
= readlink(path
, dname
, sizeof(dname
));
1524 nm
= strrchr(dname
, '/');
1526 snprintf(name
, MAX_RAID_SERIAL_LEN
, "/dev/%s", nm
);
1530 extern int scsi_get_serial(int fd
, void *buf
, size_t buf_len
);
1532 static int imsm_read_serial(int fd
, char *devname
,
1533 __u8 serial
[MAX_RAID_SERIAL_LEN
])
1535 unsigned char scsi_serial
[255];
1544 memset(scsi_serial
, 0, sizeof(scsi_serial
));
1546 rv
= scsi_get_serial(fd
, scsi_serial
, sizeof(scsi_serial
));
1548 if (rv
&& check_env("IMSM_DEVNAME_AS_SERIAL")) {
1549 memset(serial
, 0, MAX_RAID_SERIAL_LEN
);
1550 fd2devname(fd
, (char *) serial
);
1557 Name
": Failed to retrieve serial for %s\n",
1562 rsp_len
= scsi_serial
[3];
1566 Name
": Failed to retrieve serial for %s\n",
1570 rsp_buf
= (char *) &scsi_serial
[4];
1572 /* trim all whitespace and non-printable characters and convert
1575 for (i
= 0, dest
= rsp_buf
; i
< rsp_len
; i
++) {
1578 /* ':' is reserved for use in placeholder serial
1579 * numbers for missing disks
1587 len
= dest
- rsp_buf
;
1590 /* truncate leading characters */
1591 if (len
> MAX_RAID_SERIAL_LEN
) {
1592 dest
+= len
- MAX_RAID_SERIAL_LEN
;
1593 len
= MAX_RAID_SERIAL_LEN
;
1596 memset(serial
, 0, MAX_RAID_SERIAL_LEN
);
1597 memcpy(serial
, dest
, len
);
1602 static int serialcmp(__u8
*s1
, __u8
*s2
)
1604 return strncmp((char *) s1
, (char *) s2
, MAX_RAID_SERIAL_LEN
);
1607 static void serialcpy(__u8
*dest
, __u8
*src
)
1609 strncpy((char *) dest
, (char *) src
, MAX_RAID_SERIAL_LEN
);
1612 static struct dl
*serial_to_dl(__u8
*serial
, struct intel_super
*super
)
1616 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
1617 if (serialcmp(dl
->serial
, serial
) == 0)
1624 load_imsm_disk(int fd
, struct intel_super
*super
, char *devname
, int keep_fd
)
1631 __u8 serial
[MAX_RAID_SERIAL_LEN
];
1633 rv
= imsm_read_serial(fd
, devname
, serial
);
1638 /* check if this is a disk we have seen before. it may be a spare in
1639 * super->disks while the current anchor believes it is a raid member,
1640 * check if we need to update dl->index
1642 dl
= serial_to_dl(serial
, super
);
1644 dl
= malloc(sizeof(*dl
));
1651 Name
": failed to allocate disk buffer for %s\n",
1658 dl
->major
= major(stb
.st_rdev
);
1659 dl
->minor
= minor(stb
.st_rdev
);
1660 dl
->next
= super
->disks
;
1661 dl
->fd
= keep_fd
? fd
: -1;
1662 dl
->devname
= devname
? strdup(devname
) : NULL
;
1663 serialcpy(dl
->serial
, serial
);
1666 } else if (keep_fd
) {
1671 /* look up this disk's index in the current anchor */
1672 for (i
= 0; i
< super
->anchor
->num_disks
; i
++) {
1673 struct imsm_disk
*disk_iter
;
1675 disk_iter
= __get_imsm_disk(super
->anchor
, i
);
1677 if (serialcmp(disk_iter
->serial
, dl
->serial
) == 0) {
1678 dl
->disk
= *disk_iter
;
1679 /* only set index on disks that are a member of a
1680 * populated contianer, i.e. one with raid_devs
1682 if (dl
->disk
.status
& FAILED_DISK
)
1684 else if (dl
->disk
.status
& SPARE_DISK
)
1693 /* no match, maybe a stale failed drive */
1694 if (i
== super
->anchor
->num_disks
&& dl
->index
>= 0) {
1695 dl
->disk
= *__get_imsm_disk(super
->anchor
, dl
->index
);
1696 if (dl
->disk
.status
& FAILED_DISK
)
1707 /* When migrating map0 contains the 'destination' state while map1
1708 * contains the current state. When not migrating map0 contains the
1709 * current state. This routine assumes that map[0].map_state is set to
1710 * the current array state before being called.
1712 * Migration is indicated by one of the following states
1713 * 1/ Idle (migr_state=0 map0state=normal||unitialized||degraded||failed)
1714 * 2/ Initialize (migr_state=1 migr_type=MIGR_INIT map0state=normal
1715 * map1state=unitialized)
1716 * 3/ Repair (Resync) (migr_state=1 migr_type=MIGR_REPAIR map0state=normal
1718 * 4/ Rebuild (migr_state=1 migr_type=MIGR_REBUILD map0state=normal
1719 * map1state=degraded)
1721 static void migrate(struct imsm_dev
*dev
, __u8 to_state
, int migr_type
)
1723 struct imsm_map
*dest
;
1724 struct imsm_map
*src
= get_imsm_map(dev
, 0);
1726 dev
->vol
.migr_state
= 1;
1727 set_migr_type(dev
, migr_type
);
1728 dev
->vol
.curr_migr_unit
= 0;
1729 dest
= get_imsm_map(dev
, 1);
1731 /* duplicate and then set the target end state in map[0] */
1732 memcpy(dest
, src
, sizeof_imsm_map(src
));
1733 if (migr_type
== MIGR_REBUILD
) {
1737 for (i
= 0; i
< src
->num_members
; i
++) {
1738 ord
= __le32_to_cpu(src
->disk_ord_tbl
[i
]);
1739 set_imsm_ord_tbl_ent(src
, i
, ord_to_idx(ord
));
1743 src
->map_state
= to_state
;
1746 static void end_migration(struct imsm_dev
*dev
, __u8 map_state
)
1748 struct imsm_map
*map
= get_imsm_map(dev
, 0);
1749 struct imsm_map
*prev
= get_imsm_map(dev
, dev
->vol
.migr_state
);
1752 /* merge any IMSM_ORD_REBUILD bits that were not successfully
1753 * completed in the last migration.
1755 * FIXME add support for online capacity expansion and
1756 * raid-level-migration
1758 for (i
= 0; i
< prev
->num_members
; i
++)
1759 map
->disk_ord_tbl
[i
] |= prev
->disk_ord_tbl
[i
];
1761 dev
->vol
.migr_state
= 0;
1762 dev
->vol
.curr_migr_unit
= 0;
1763 map
->map_state
= map_state
;
1767 static int parse_raid_devices(struct intel_super
*super
)
1770 struct imsm_dev
*dev_new
;
1771 size_t len
, len_migr
;
1772 size_t space_needed
= 0;
1773 struct imsm_super
*mpb
= super
->anchor
;
1775 for (i
= 0; i
< super
->anchor
->num_raid_devs
; i
++) {
1776 struct imsm_dev
*dev_iter
= __get_imsm_dev(super
->anchor
, i
);
1777 struct intel_dev
*dv
;
1779 len
= sizeof_imsm_dev(dev_iter
, 0);
1780 len_migr
= sizeof_imsm_dev(dev_iter
, 1);
1782 space_needed
+= len_migr
- len
;
1784 dv
= malloc(sizeof(*dv
));
1787 dev_new
= malloc(len_migr
);
1792 imsm_copy_dev(dev_new
, dev_iter
);
1795 dv
->next
= super
->devlist
;
1796 super
->devlist
= dv
;
1799 /* ensure that super->buf is large enough when all raid devices
1802 if (__le32_to_cpu(mpb
->mpb_size
) + space_needed
> super
->len
) {
1805 len
= ROUND_UP(__le32_to_cpu(mpb
->mpb_size
) + space_needed
, 512);
1806 if (posix_memalign(&buf
, 512, len
) != 0)
1809 memcpy(buf
, super
->buf
, super
->len
);
1810 memset(buf
+ super
->len
, 0, len
- super
->len
);
1819 /* retrieve a pointer to the bbm log which starts after all raid devices */
1820 struct bbm_log
*__get_imsm_bbm_log(struct imsm_super
*mpb
)
1824 if (__le32_to_cpu(mpb
->bbm_log_size
)) {
1826 ptr
+= mpb
->mpb_size
- __le32_to_cpu(mpb
->bbm_log_size
);
1832 static void __free_imsm(struct intel_super
*super
, int free_disks
);
1834 /* load_imsm_mpb - read matrix metadata
1835 * allocates super->mpb to be freed by free_super
1837 static int load_imsm_mpb(int fd
, struct intel_super
*super
, char *devname
)
1839 unsigned long long dsize
;
1840 unsigned long long sectors
;
1842 struct imsm_super
*anchor
;
1846 get_dev_size(fd
, NULL
, &dsize
);
1848 if (lseek64(fd
, dsize
- (512 * 2), SEEK_SET
) < 0) {
1851 Name
": Cannot seek to anchor block on %s: %s\n",
1852 devname
, strerror(errno
));
1856 if (posix_memalign((void**)&anchor
, 512, 512) != 0) {
1859 Name
": Failed to allocate imsm anchor buffer"
1860 " on %s\n", devname
);
1863 if (read(fd
, anchor
, 512) != 512) {
1866 Name
": Cannot read anchor block on %s: %s\n",
1867 devname
, strerror(errno
));
1872 if (strncmp((char *) anchor
->sig
, MPB_SIGNATURE
, MPB_SIG_LEN
) != 0) {
1875 Name
": no IMSM anchor on %s\n", devname
);
1880 __free_imsm(super
, 0);
1881 super
->len
= ROUND_UP(anchor
->mpb_size
, 512);
1882 if (posix_memalign(&super
->buf
, 512, super
->len
) != 0) {
1885 Name
": unable to allocate %zu byte mpb buffer\n",
1890 memcpy(super
->buf
, anchor
, 512);
1892 sectors
= mpb_sectors(anchor
) - 1;
1895 check_sum
= __gen_imsm_checksum(super
->anchor
);
1896 if (check_sum
!= __le32_to_cpu(super
->anchor
->check_sum
)) {
1899 Name
": IMSM checksum %x != %x on %s\n",
1901 __le32_to_cpu(super
->anchor
->check_sum
),
1906 rc
= load_imsm_disk(fd
, super
, devname
, 0);
1908 rc
= parse_raid_devices(super
);
1912 /* read the extended mpb */
1913 if (lseek64(fd
, dsize
- (512 * (2 + sectors
)), SEEK_SET
) < 0) {
1916 Name
": Cannot seek to extended mpb on %s: %s\n",
1917 devname
, strerror(errno
));
1921 if (read(fd
, super
->buf
+ 512, super
->len
- 512) != super
->len
- 512) {
1924 Name
": Cannot read extended mpb on %s: %s\n",
1925 devname
, strerror(errno
));
1929 check_sum
= __gen_imsm_checksum(super
->anchor
);
1930 if (check_sum
!= __le32_to_cpu(super
->anchor
->check_sum
)) {
1933 Name
": IMSM checksum %x != %x on %s\n",
1934 check_sum
, __le32_to_cpu(super
->anchor
->check_sum
),
1939 /* FIXME the BBM log is disk specific so we cannot use this global
1940 * buffer for all disks. Ok for now since we only look at the global
1941 * bbm_log_size parameter to gate assembly
1943 super
->bbm_log
= __get_imsm_bbm_log(super
->anchor
);
1945 rc
= load_imsm_disk(fd
, super
, devname
, 0);
1947 rc
= parse_raid_devices(super
);
1952 static void __free_imsm_disk(struct dl
*d
)
1963 static void free_imsm_disks(struct intel_super
*super
)
1967 while (super
->disks
) {
1969 super
->disks
= d
->next
;
1970 __free_imsm_disk(d
);
1972 while (super
->missing
) {
1974 super
->missing
= d
->next
;
1975 __free_imsm_disk(d
);
1980 /* free all the pieces hanging off of a super pointer */
1981 static void __free_imsm(struct intel_super
*super
, int free_disks
)
1988 free_imsm_disks(super
);
1989 free_devlist(super
);
1991 free((void *) super
->hba
);
1996 static void free_imsm(struct intel_super
*super
)
1998 __free_imsm(super
, 1);
2002 static void free_super_imsm(struct supertype
*st
)
2004 struct intel_super
*super
= st
->sb
;
2013 static struct intel_super
*alloc_super(int creating_imsm
)
2015 struct intel_super
*super
= malloc(sizeof(*super
));
2018 memset(super
, 0, sizeof(*super
));
2019 super
->creating_imsm
= creating_imsm
;
2020 super
->current_vol
= -1;
2021 super
->create_offset
= ~((__u32
) 0);
2022 if (!check_env("IMSM_NO_PLATFORM"))
2023 super
->orom
= find_imsm_orom();
2024 if (super
->orom
&& !check_env("IMSM_TEST_OROM")) {
2025 struct sys_dev
*list
, *ent
;
2027 /* find the first intel ahci controller */
2028 list
= find_driver_devices("pci", "ahci");
2029 for (ent
= list
; ent
; ent
= ent
->next
)
2030 if (devpath_to_vendor(ent
->path
) == 0x8086)
2033 super
->hba
= ent
->path
;
2036 free_sys_dev(&list
);
2044 /* find_missing - helper routine for load_super_imsm_all that identifies
2045 * disks that have disappeared from the system. This routine relies on
2046 * the mpb being uptodate, which it is at load time.
2048 static int find_missing(struct intel_super
*super
)
2051 struct imsm_super
*mpb
= super
->anchor
;
2053 struct imsm_disk
*disk
;
2055 for (i
= 0; i
< mpb
->num_disks
; i
++) {
2056 disk
= __get_imsm_disk(mpb
, i
);
2057 dl
= serial_to_dl(disk
->serial
, super
);
2061 dl
= malloc(sizeof(*dl
));
2067 dl
->devname
= strdup("missing");
2069 serialcpy(dl
->serial
, disk
->serial
);
2072 dl
->next
= super
->missing
;
2073 super
->missing
= dl
;
2079 static int load_super_imsm_all(struct supertype
*st
, int fd
, void **sbp
,
2080 char *devname
, int keep_fd
)
2083 struct intel_super
*super
;
2084 struct mdinfo
*sd
, *best
= NULL
;
2090 int devnum
= fd2devnum(fd
);
2092 enum sysfs_read_flags flags
;
2094 flags
= GET_LEVEL
|GET_VERSION
|GET_DEVS
|GET_STATE
;
2095 if (mdmon_running(devnum
))
2096 flags
|= SKIP_GONE_DEVS
;
2098 /* check if 'fd' an opened container */
2099 sra
= sysfs_read(fd
, 0, flags
);
2103 if (sra
->array
.major_version
!= -1 ||
2104 sra
->array
.minor_version
!= -2 ||
2105 strcmp(sra
->text_version
, "imsm") != 0)
2108 super
= alloc_super(0);
2112 /* find the most up to date disk in this array, skipping spares */
2113 for (sd
= sra
->devs
; sd
; sd
= sd
->next
) {
2114 sprintf(nm
, "%d:%d", sd
->disk
.major
, sd
->disk
.minor
);
2115 dfd
= dev_open(nm
, keep_fd
? O_RDWR
: O_RDONLY
);
2120 rv
= load_imsm_mpb(dfd
, super
, NULL
);
2122 /* retry the load if we might have raced against mdmon */
2123 if (rv
== 3 && mdmon_running(devnum
))
2124 for (retry
= 0; retry
< 3; retry
++) {
2126 rv
= load_imsm_mpb(dfd
, super
, NULL
);
2133 if (super
->anchor
->num_raid_devs
== 0)
2136 gen
= __le32_to_cpu(super
->anchor
->generation_num
);
2137 if (!best
|| gen
> bestgen
) {
2152 /* load the most up to date anchor */
2153 sprintf(nm
, "%d:%d", best
->disk
.major
, best
->disk
.minor
);
2154 dfd
= dev_open(nm
, O_RDONLY
);
2159 rv
= load_imsm_mpb(dfd
, super
, NULL
);
2166 /* re-parse the disk list with the current anchor */
2167 for (sd
= sra
->devs
; sd
; sd
= sd
->next
) {
2168 sprintf(nm
, "%d:%d", sd
->disk
.major
, sd
->disk
.minor
);
2169 dfd
= dev_open(nm
, keep_fd
? O_RDWR
: O_RDONLY
);
2174 load_imsm_disk(dfd
, super
, NULL
, keep_fd
);
2180 if (find_missing(super
) != 0) {
2185 if (st
->subarray
[0]) {
2186 if (atoi(st
->subarray
) <= super
->anchor
->num_raid_devs
)
2187 super
->current_vol
= atoi(st
->subarray
);
2195 st
->container_dev
= devnum
;
2196 if (st
->ss
== NULL
) {
2197 st
->ss
= &super_imsm
;
2198 st
->minor_version
= 0;
2199 st
->max_devs
= IMSM_MAX_DEVICES
;
2201 st
->loaded_container
= 1;
2207 static int load_super_imsm(struct supertype
*st
, int fd
, char *devname
)
2209 struct intel_super
*super
;
2213 if (load_super_imsm_all(st
, fd
, &st
->sb
, devname
, 1) == 0)
2217 free_super_imsm(st
);
2219 super
= alloc_super(0);
2222 Name
": malloc of %zu failed.\n",
2227 rv
= load_imsm_mpb(fd
, super
, devname
);
2232 Name
": Failed to load all information "
2233 "sections on %s\n", devname
);
2238 if (st
->subarray
[0]) {
2239 if (atoi(st
->subarray
) <= super
->anchor
->num_raid_devs
)
2240 super
->current_vol
= atoi(st
->subarray
);
2248 if (st
->ss
== NULL
) {
2249 st
->ss
= &super_imsm
;
2250 st
->minor_version
= 0;
2251 st
->max_devs
= IMSM_MAX_DEVICES
;
2253 st
->loaded_container
= 0;
2258 static __u16
info_to_blocks_per_strip(mdu_array_info_t
*info
)
2260 if (info
->level
== 1)
2262 return info
->chunk_size
>> 9;
2265 static __u32
info_to_num_data_stripes(mdu_array_info_t
*info
, int num_domains
)
2269 num_stripes
= (info
->size
* 2) / info_to_blocks_per_strip(info
);
2270 num_stripes
/= num_domains
;
2275 static __u32
info_to_blocks_per_member(mdu_array_info_t
*info
)
2277 if (info
->level
== 1)
2278 return info
->size
* 2;
2280 return (info
->size
* 2) & ~(info_to_blocks_per_strip(info
) - 1);
2283 static void imsm_update_version_info(struct intel_super
*super
)
2285 /* update the version and attributes */
2286 struct imsm_super
*mpb
= super
->anchor
;
2288 struct imsm_dev
*dev
;
2289 struct imsm_map
*map
;
2292 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
2293 dev
= get_imsm_dev(super
, i
);
2294 map
= get_imsm_map(dev
, 0);
2295 if (__le32_to_cpu(dev
->size_high
) > 0)
2296 mpb
->attributes
|= MPB_ATTRIB_2TB
;
2298 /* FIXME detect when an array spans a port multiplier */
2300 mpb
->attributes
|= MPB_ATTRIB_PM
;
2303 if (mpb
->num_raid_devs
> 1 ||
2304 mpb
->attributes
!= MPB_ATTRIB_CHECKSUM_VERIFY
) {
2305 version
= MPB_VERSION_ATTRIBS
;
2306 switch (get_imsm_raid_level(map
)) {
2307 case 0: mpb
->attributes
|= MPB_ATTRIB_RAID0
; break;
2308 case 1: mpb
->attributes
|= MPB_ATTRIB_RAID1
; break;
2309 case 10: mpb
->attributes
|= MPB_ATTRIB_RAID10
; break;
2310 case 5: mpb
->attributes
|= MPB_ATTRIB_RAID5
; break;
2313 if (map
->num_members
>= 5)
2314 version
= MPB_VERSION_5OR6_DISK_ARRAY
;
2315 else if (dev
->status
== DEV_CLONE_N_GO
)
2316 version
= MPB_VERSION_CNG
;
2317 else if (get_imsm_raid_level(map
) == 5)
2318 version
= MPB_VERSION_RAID5
;
2319 else if (map
->num_members
>= 3)
2320 version
= MPB_VERSION_3OR4_DISK_ARRAY
;
2321 else if (get_imsm_raid_level(map
) == 1)
2322 version
= MPB_VERSION_RAID1
;
2324 version
= MPB_VERSION_RAID0
;
2326 strcpy(((char *) mpb
->sig
) + strlen(MPB_SIGNATURE
), version
);
2330 static int init_super_imsm_volume(struct supertype
*st
, mdu_array_info_t
*info
,
2331 unsigned long long size
, char *name
,
2332 char *homehost
, int *uuid
)
2334 /* We are creating a volume inside a pre-existing container.
2335 * so st->sb is already set.
2337 struct intel_super
*super
= st
->sb
;
2338 struct imsm_super
*mpb
= super
->anchor
;
2339 struct intel_dev
*dv
;
2340 struct imsm_dev
*dev
;
2341 struct imsm_vol
*vol
;
2342 struct imsm_map
*map
;
2343 int idx
= mpb
->num_raid_devs
;
2345 unsigned long long array_blocks
;
2346 size_t size_old
, size_new
;
2347 __u32 num_data_stripes
;
2349 if (super
->orom
&& mpb
->num_raid_devs
>= super
->orom
->vpa
) {
2350 fprintf(stderr
, Name
": This imsm-container already has the "
2351 "maximum of %d volumes\n", super
->orom
->vpa
);
2355 /* ensure the mpb is large enough for the new data */
2356 size_old
= __le32_to_cpu(mpb
->mpb_size
);
2357 size_new
= disks_to_mpb_size(info
->nr_disks
);
2358 if (size_new
> size_old
) {
2360 size_t size_round
= ROUND_UP(size_new
, 512);
2362 if (posix_memalign(&mpb_new
, 512, size_round
) != 0) {
2363 fprintf(stderr
, Name
": could not allocate new mpb\n");
2366 memcpy(mpb_new
, mpb
, size_old
);
2369 super
->anchor
= mpb_new
;
2370 mpb
->mpb_size
= __cpu_to_le32(size_new
);
2371 memset(mpb_new
+ size_old
, 0, size_round
- size_old
);
2373 super
->current_vol
= idx
;
2374 /* when creating the first raid device in this container set num_disks
2375 * to zero, i.e. delete this spare and add raid member devices in
2376 * add_to_super_imsm_volume()
2378 if (super
->current_vol
== 0)
2381 for (i
= 0; i
< super
->current_vol
; i
++) {
2382 dev
= get_imsm_dev(super
, i
);
2383 if (strncmp((char *) dev
->volume
, name
,
2384 MAX_RAID_SERIAL_LEN
) == 0) {
2385 fprintf(stderr
, Name
": '%s' is already defined for this container\n",
2391 sprintf(st
->subarray
, "%d", idx
);
2392 dv
= malloc(sizeof(*dv
));
2394 fprintf(stderr
, Name
": failed to allocate device list entry\n");
2397 dev
= malloc(sizeof(*dev
) + sizeof(__u32
) * (info
->raid_disks
- 1));
2400 fprintf(stderr
, Name
": could not allocate raid device\n");
2403 strncpy((char *) dev
->volume
, name
, MAX_RAID_SERIAL_LEN
);
2404 if (info
->level
== 1)
2405 array_blocks
= info_to_blocks_per_member(info
);
2407 array_blocks
= calc_array_size(info
->level
, info
->raid_disks
,
2408 info
->layout
, info
->chunk_size
,
2410 /* round array size down to closest MB */
2411 array_blocks
= (array_blocks
>> SECT_PER_MB_SHIFT
) << SECT_PER_MB_SHIFT
;
2413 dev
->size_low
= __cpu_to_le32((__u32
) array_blocks
);
2414 dev
->size_high
= __cpu_to_le32((__u32
) (array_blocks
>> 32));
2415 dev
->status
= __cpu_to_le32(0);
2416 dev
->reserved_blocks
= __cpu_to_le32(0);
2418 vol
->migr_state
= 0;
2419 set_migr_type(dev
, MIGR_INIT
);
2421 vol
->curr_migr_unit
= 0;
2422 map
= get_imsm_map(dev
, 0);
2423 map
->pba_of_lba0
= __cpu_to_le32(super
->create_offset
);
2424 map
->blocks_per_member
= __cpu_to_le32(info_to_blocks_per_member(info
));
2425 map
->blocks_per_strip
= __cpu_to_le16(info_to_blocks_per_strip(info
));
2426 map
->failed_disk_num
= ~0;
2427 map
->map_state
= info
->level
? IMSM_T_STATE_UNINITIALIZED
:
2428 IMSM_T_STATE_NORMAL
;
2431 if (info
->level
== 1 && info
->raid_disks
> 2) {
2432 fprintf(stderr
, Name
": imsm does not support more than 2 disks"
2433 "in a raid1 volume\n");
2437 map
->raid_level
= info
->level
;
2438 if (info
->level
== 10) {
2439 map
->raid_level
= 1;
2440 map
->num_domains
= info
->raid_disks
/ 2;
2441 } else if (info
->level
== 1)
2442 map
->num_domains
= info
->raid_disks
;
2444 map
->num_domains
= 1;
2446 num_data_stripes
= info_to_num_data_stripes(info
, map
->num_domains
);
2447 map
->num_data_stripes
= __cpu_to_le32(num_data_stripes
);
2449 map
->num_members
= info
->raid_disks
;
2450 for (i
= 0; i
< map
->num_members
; i
++) {
2451 /* initialized in add_to_super */
2452 set_imsm_ord_tbl_ent(map
, i
, 0);
2454 mpb
->num_raid_devs
++;
2457 dv
->index
= super
->current_vol
;
2458 dv
->next
= super
->devlist
;
2459 super
->devlist
= dv
;
2461 imsm_update_version_info(super
);
2466 static int init_super_imsm(struct supertype
*st
, mdu_array_info_t
*info
,
2467 unsigned long long size
, char *name
,
2468 char *homehost
, int *uuid
)
2470 /* This is primarily called by Create when creating a new array.
2471 * We will then get add_to_super called for each component, and then
2472 * write_init_super called to write it out to each device.
2473 * For IMSM, Create can create on fresh devices or on a pre-existing
2475 * To create on a pre-existing array a different method will be called.
2476 * This one is just for fresh drives.
2478 struct intel_super
*super
;
2479 struct imsm_super
*mpb
;
2488 return init_super_imsm_volume(st
, info
, size
, name
, homehost
,
2491 super
= alloc_super(1);
2494 mpb_size
= disks_to_mpb_size(info
->nr_disks
);
2495 if (posix_memalign(&super
->buf
, 512, mpb_size
) != 0) {
2500 memset(mpb
, 0, mpb_size
);
2502 mpb
->attributes
= MPB_ATTRIB_CHECKSUM_VERIFY
;
2504 version
= (char *) mpb
->sig
;
2505 strcpy(version
, MPB_SIGNATURE
);
2506 version
+= strlen(MPB_SIGNATURE
);
2507 strcpy(version
, MPB_VERSION_RAID0
);
2508 mpb
->mpb_size
= mpb_size
;
2515 static int add_to_super_imsm_volume(struct supertype
*st
, mdu_disk_info_t
*dk
,
2516 int fd
, char *devname
)
2518 struct intel_super
*super
= st
->sb
;
2519 struct imsm_super
*mpb
= super
->anchor
;
2521 struct imsm_dev
*dev
;
2522 struct imsm_map
*map
;
2524 dev
= get_imsm_dev(super
, super
->current_vol
);
2525 map
= get_imsm_map(dev
, 0);
2527 if (! (dk
->state
& (1<<MD_DISK_SYNC
))) {
2528 fprintf(stderr
, Name
": %s: Cannot add spare devices to IMSM volume\n",
2534 /* we're doing autolayout so grab the pre-marked (in
2535 * validate_geometry) raid_disk
2537 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
2538 if (dl
->raiddisk
== dk
->raid_disk
)
2541 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
2542 if (dl
->major
== dk
->major
&&
2543 dl
->minor
== dk
->minor
)
2548 fprintf(stderr
, Name
": %s is not a member of the same container\n", devname
);
2552 /* add a pristine spare to the metadata */
2553 if (dl
->index
< 0) {
2554 dl
->index
= super
->anchor
->num_disks
;
2555 super
->anchor
->num_disks
++;
2557 set_imsm_ord_tbl_ent(map
, dk
->number
, dl
->index
);
2558 dl
->disk
.status
= CONFIGURED_DISK
| USABLE_DISK
;
2560 /* if we are creating the first raid device update the family number */
2561 if (super
->current_vol
== 0) {
2563 struct imsm_dev
*_dev
= __get_imsm_dev(mpb
, 0);
2564 struct imsm_disk
*_disk
= __get_imsm_disk(mpb
, dl
->index
);
2569 sum
+= __gen_imsm_checksum(mpb
);
2570 mpb
->family_num
= __cpu_to_le32(sum
);
2571 mpb
->orig_family_num
= mpb
->family_num
;
2577 static int add_to_super_imsm(struct supertype
*st
, mdu_disk_info_t
*dk
,
2578 int fd
, char *devname
)
2580 struct intel_super
*super
= st
->sb
;
2582 unsigned long long size
;
2587 /* if we are on an RAID enabled platform check that the disk is
2588 * attached to the raid controller
2590 if (super
->hba
&& !disk_attached_to_hba(fd
, super
->hba
)) {
2592 Name
": %s is not attached to the raid controller: %s\n",
2593 devname
? : "disk", super
->hba
);
2597 if (super
->current_vol
>= 0)
2598 return add_to_super_imsm_volume(st
, dk
, fd
, devname
);
2601 dd
= malloc(sizeof(*dd
));
2604 Name
": malloc failed %s:%d.\n", __func__
, __LINE__
);
2607 memset(dd
, 0, sizeof(*dd
));
2608 dd
->major
= major(stb
.st_rdev
);
2609 dd
->minor
= minor(stb
.st_rdev
);
2611 dd
->devname
= devname
? strdup(devname
) : NULL
;
2614 rv
= imsm_read_serial(fd
, devname
, dd
->serial
);
2617 Name
": failed to retrieve scsi serial, aborting\n");
2622 get_dev_size(fd
, NULL
, &size
);
2624 serialcpy(dd
->disk
.serial
, dd
->serial
);
2625 dd
->disk
.total_blocks
= __cpu_to_le32(size
);
2626 dd
->disk
.status
= USABLE_DISK
| SPARE_DISK
;
2627 if (sysfs_disk_to_scsi_id(fd
, &id
) == 0)
2628 dd
->disk
.scsi_id
= __cpu_to_le32(id
);
2630 dd
->disk
.scsi_id
= __cpu_to_le32(0);
2632 if (st
->update_tail
) {
2633 dd
->next
= super
->add
;
2636 dd
->next
= super
->disks
;
2643 static int store_imsm_mpb(int fd
, struct intel_super
*super
);
2645 /* spare records have their own family number and do not have any defined raid
2648 static int write_super_imsm_spares(struct intel_super
*super
, int doclose
)
2650 struct imsm_super mpb_save
;
2651 struct imsm_super
*mpb
= super
->anchor
;
2656 mpb
->num_raid_devs
= 0;
2658 mpb
->mpb_size
= sizeof(struct imsm_super
);
2659 mpb
->generation_num
= __cpu_to_le32(1UL);
2661 for (d
= super
->disks
; d
; d
= d
->next
) {
2665 mpb
->disk
[0] = d
->disk
;
2666 sum
= __gen_imsm_checksum(mpb
);
2667 mpb
->family_num
= __cpu_to_le32(sum
);
2668 mpb
->orig_family_num
= 0;
2669 sum
= __gen_imsm_checksum(mpb
);
2670 mpb
->check_sum
= __cpu_to_le32(sum
);
2672 if (store_imsm_mpb(d
->fd
, super
)) {
2673 fprintf(stderr
, "%s: failed for device %d:%d %s\n",
2674 __func__
, d
->major
, d
->minor
, strerror(errno
));
2688 static int write_super_imsm(struct intel_super
*super
, int doclose
)
2690 struct imsm_super
*mpb
= super
->anchor
;
2696 __u32 mpb_size
= sizeof(struct imsm_super
) - sizeof(struct imsm_disk
);
2698 /* 'generation' is incremented everytime the metadata is written */
2699 generation
= __le32_to_cpu(mpb
->generation_num
);
2701 mpb
->generation_num
= __cpu_to_le32(generation
);
2703 /* fix up cases where previous mdadm releases failed to set
2706 if (mpb
->orig_family_num
== 0)
2707 mpb
->orig_family_num
= mpb
->family_num
;
2709 mpb_size
+= sizeof(struct imsm_disk
) * mpb
->num_disks
;
2710 for (d
= super
->disks
; d
; d
= d
->next
) {
2714 mpb
->disk
[d
->index
] = d
->disk
;
2716 for (d
= super
->missing
; d
; d
= d
->next
)
2717 mpb
->disk
[d
->index
] = d
->disk
;
2719 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
2720 struct imsm_dev
*dev
= __get_imsm_dev(mpb
, i
);
2722 imsm_copy_dev(dev
, get_imsm_dev(super
, i
));
2723 mpb_size
+= sizeof_imsm_dev(dev
, 0);
2725 mpb_size
+= __le32_to_cpu(mpb
->bbm_log_size
);
2726 mpb
->mpb_size
= __cpu_to_le32(mpb_size
);
2728 /* recalculate checksum */
2729 sum
= __gen_imsm_checksum(mpb
);
2730 mpb
->check_sum
= __cpu_to_le32(sum
);
2732 /* write the mpb for disks that compose raid devices */
2733 for (d
= super
->disks
; d
; d
= d
->next
) {
2736 if (store_imsm_mpb(d
->fd
, super
))
2737 fprintf(stderr
, "%s: failed for device %d:%d %s\n",
2738 __func__
, d
->major
, d
->minor
, strerror(errno
));
2746 return write_super_imsm_spares(super
, doclose
);
2752 static int create_array(struct supertype
*st
, int dev_idx
)
2755 struct imsm_update_create_array
*u
;
2756 struct intel_super
*super
= st
->sb
;
2757 struct imsm_dev
*dev
= get_imsm_dev(super
, dev_idx
);
2758 struct imsm_map
*map
= get_imsm_map(dev
, 0);
2759 struct disk_info
*inf
;
2760 struct imsm_disk
*disk
;
2763 len
= sizeof(*u
) - sizeof(*dev
) + sizeof_imsm_dev(dev
, 0) +
2764 sizeof(*inf
) * map
->num_members
;
2767 fprintf(stderr
, "%s: failed to allocate update buffer\n",
2772 u
->type
= update_create_array
;
2773 u
->dev_idx
= dev_idx
;
2774 imsm_copy_dev(&u
->dev
, dev
);
2775 inf
= get_disk_info(u
);
2776 for (i
= 0; i
< map
->num_members
; i
++) {
2777 int idx
= get_imsm_disk_idx(dev
, i
);
2779 disk
= get_imsm_disk(super
, idx
);
2780 serialcpy(inf
[i
].serial
, disk
->serial
);
2782 append_metadata_update(st
, u
, len
);
2787 static int _add_disk(struct supertype
*st
)
2789 struct intel_super
*super
= st
->sb
;
2791 struct imsm_update_add_disk
*u
;
2799 fprintf(stderr
, "%s: failed to allocate update buffer\n",
2804 u
->type
= update_add_disk
;
2805 append_metadata_update(st
, u
, len
);
2810 static int write_init_super_imsm(struct supertype
*st
)
2812 struct intel_super
*super
= st
->sb
;
2813 int current_vol
= super
->current_vol
;
2815 /* we are done with current_vol reset it to point st at the container */
2816 super
->current_vol
= -1;
2818 if (st
->update_tail
) {
2819 /* queue the recently created array / added disk
2820 * as a metadata update */
2824 /* determine if we are creating a volume or adding a disk */
2825 if (current_vol
< 0) {
2826 /* in the add disk case we are running in mdmon
2827 * context, so don't close fd's
2829 return _add_disk(st
);
2831 rv
= create_array(st
, current_vol
);
2833 for (d
= super
->disks
; d
; d
= d
->next
) {
2840 return write_super_imsm(st
->sb
, 1);
2844 static int store_zero_imsm(struct supertype
*st
, int fd
)
2846 unsigned long long dsize
;
2849 get_dev_size(fd
, NULL
, &dsize
);
2851 /* first block is stored on second to last sector of the disk */
2852 if (lseek64(fd
, dsize
- (512 * 2), SEEK_SET
) < 0)
2855 if (posix_memalign(&buf
, 512, 512) != 0)
2858 memset(buf
, 0, 512);
2859 if (write(fd
, buf
, 512) != 512)
2864 static int imsm_bbm_log_size(struct imsm_super
*mpb
)
2866 return __le32_to_cpu(mpb
->bbm_log_size
);
2870 static int validate_geometry_imsm_container(struct supertype
*st
, int level
,
2871 int layout
, int raiddisks
, int chunk
,
2872 unsigned long long size
, char *dev
,
2873 unsigned long long *freesize
,
2877 unsigned long long ldsize
;
2878 const struct imsm_orom
*orom
;
2880 if (level
!= LEVEL_CONTAINER
)
2885 if (check_env("IMSM_NO_PLATFORM"))
2888 orom
= find_imsm_orom();
2889 if (orom
&& raiddisks
> orom
->tds
) {
2891 fprintf(stderr
, Name
": %d exceeds maximum number of"
2892 " platform supported disks: %d\n",
2893 raiddisks
, orom
->tds
);
2897 fd
= open(dev
, O_RDONLY
|O_EXCL
, 0);
2900 fprintf(stderr
, Name
": imsm: Cannot open %s: %s\n",
2901 dev
, strerror(errno
));
2904 if (!get_dev_size(fd
, dev
, &ldsize
)) {
2910 *freesize
= avail_size_imsm(st
, ldsize
>> 9);
2915 static unsigned long long find_size(struct extent
*e
, int *idx
, int num_extents
)
2917 const unsigned long long base_start
= e
[*idx
].start
;
2918 unsigned long long end
= base_start
+ e
[*idx
].size
;
2921 if (base_start
== end
)
2925 for (i
= *idx
; i
< num_extents
; i
++) {
2926 /* extend overlapping extents */
2927 if (e
[i
].start
>= base_start
&&
2928 e
[i
].start
<= end
) {
2931 if (e
[i
].start
+ e
[i
].size
> end
)
2932 end
= e
[i
].start
+ e
[i
].size
;
2933 } else if (e
[i
].start
> end
) {
2939 return end
- base_start
;
2942 static unsigned long long merge_extents(struct intel_super
*super
, int sum_extents
)
2944 /* build a composite disk with all known extents and generate a new
2945 * 'maxsize' given the "all disks in an array must share a common start
2946 * offset" constraint
2948 struct extent
*e
= calloc(sum_extents
, sizeof(*e
));
2952 unsigned long long pos
;
2953 unsigned long long start
= 0;
2954 unsigned long long maxsize
;
2955 unsigned long reserve
;
2958 return ~0ULL; /* error */
2960 /* coalesce and sort all extents. also, check to see if we need to
2961 * reserve space between member arrays
2964 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
2967 for (i
= 0; i
< dl
->extent_cnt
; i
++)
2970 qsort(e
, sum_extents
, sizeof(*e
), cmp_extent
);
2975 while (i
< sum_extents
) {
2976 e
[j
].start
= e
[i
].start
;
2977 e
[j
].size
= find_size(e
, &i
, sum_extents
);
2979 if (e
[j
-1].size
== 0)
2988 unsigned long long esize
;
2990 esize
= e
[i
].start
- pos
;
2991 if (esize
>= maxsize
) {
2996 pos
= e
[i
].start
+ e
[i
].size
;
2998 } while (e
[i
-1].size
);
3001 if (start_extent
> 0)
3002 reserve
= IMSM_RESERVED_SECTORS
; /* gap between raid regions */
3006 if (maxsize
< reserve
)
3009 super
->create_offset
= ~((__u32
) 0);
3010 if (start
+ reserve
> super
->create_offset
)
3011 return ~0ULL; /* start overflows create_offset */
3012 super
->create_offset
= start
+ reserve
;
3014 return maxsize
- reserve
;
3017 static int is_raid_level_supported(const struct imsm_orom
*orom
, int level
, int raiddisks
)
3019 if (level
< 0 || level
== 6 || level
== 4)
3022 /* if we have an orom prevent invalid raid levels */
3025 case 0: return imsm_orom_has_raid0(orom
);
3028 return imsm_orom_has_raid1e(orom
);
3029 return imsm_orom_has_raid1(orom
) && raiddisks
== 2;
3030 case 10: return imsm_orom_has_raid10(orom
) && raiddisks
== 4;
3031 case 5: return imsm_orom_has_raid5(orom
) && raiddisks
> 2;
3034 return 1; /* not on an Intel RAID platform so anything goes */
3039 #define pr_vrb(fmt, arg...) (void) (verbose && fprintf(stderr, Name fmt, ##arg))
3040 /* validate_geometry_imsm_volume - lifted from validate_geometry_ddf_bvd
3041 * FIX ME add ahci details
3043 static int validate_geometry_imsm_volume(struct supertype
*st
, int level
,
3044 int layout
, int raiddisks
, int chunk
,
3045 unsigned long long size
, char *dev
,
3046 unsigned long long *freesize
,
3050 struct intel_super
*super
= st
->sb
;
3051 struct imsm_super
*mpb
= super
->anchor
;
3053 unsigned long long pos
= 0;
3054 unsigned long long maxsize
;
3058 /* We must have the container info already read in. */
3062 if (!is_raid_level_supported(super
->orom
, level
, raiddisks
)) {
3063 pr_vrb(": platform does not support raid%d with %d disk%s\n",
3064 level
, raiddisks
, raiddisks
> 1 ? "s" : "");
3067 if (super
->orom
&& level
!= 1 &&
3068 !imsm_orom_has_chunk(super
->orom
, chunk
)) {
3069 pr_vrb(": platform does not support a chunk size of: %d\n", chunk
);
3072 if (layout
!= imsm_level_to_layout(level
)) {
3074 pr_vrb(": imsm raid 5 only supports the left-asymmetric layout\n");
3075 else if (level
== 10)
3076 pr_vrb(": imsm raid 10 only supports the n2 layout\n");
3078 pr_vrb(": imsm unknown layout %#x for this raid level %d\n",
3084 /* General test: make sure there is space for
3085 * 'raiddisks' device extents of size 'size' at a given
3088 unsigned long long minsize
= size
;
3089 unsigned long long start_offset
= ~0ULL;
3092 minsize
= MPB_SECTOR_CNT
+ IMSM_RESERVED_SECTORS
;
3093 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
3098 e
= get_extents(super
, dl
);
3101 unsigned long long esize
;
3102 esize
= e
[i
].start
- pos
;
3103 if (esize
>= minsize
)
3105 if (found
&& start_offset
== ~0ULL) {
3108 } else if (found
&& pos
!= start_offset
) {
3112 pos
= e
[i
].start
+ e
[i
].size
;
3114 } while (e
[i
-1].size
);
3119 if (dcnt
< raiddisks
) {
3121 fprintf(stderr
, Name
": imsm: Not enough "
3122 "devices with space for this array "
3130 /* This device must be a member of the set */
3131 if (stat(dev
, &stb
) < 0)
3133 if ((S_IFMT
& stb
.st_mode
) != S_IFBLK
)
3135 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
3136 if (dl
->major
== major(stb
.st_rdev
) &&
3137 dl
->minor
== minor(stb
.st_rdev
))
3142 fprintf(stderr
, Name
": %s is not in the "
3143 "same imsm set\n", dev
);
3145 } else if (super
->orom
&& dl
->index
< 0 && mpb
->num_raid_devs
) {
3146 /* If a volume is present then the current creation attempt
3147 * cannot incorporate new spares because the orom may not
3148 * understand this configuration (all member disks must be
3149 * members of each array in the container).
3151 fprintf(stderr
, Name
": %s is a spare and a volume"
3152 " is already defined for this container\n", dev
);
3153 fprintf(stderr
, Name
": The option-rom requires all member"
3154 " disks to be a member of all volumes\n");
3158 /* retrieve the largest free space block */
3159 e
= get_extents(super
, dl
);
3164 unsigned long long esize
;
3166 esize
= e
[i
].start
- pos
;
3167 if (esize
>= maxsize
)
3169 pos
= e
[i
].start
+ e
[i
].size
;
3171 } while (e
[i
-1].size
);
3176 fprintf(stderr
, Name
": unable to determine free space for: %s\n",
3180 if (maxsize
< size
) {
3182 fprintf(stderr
, Name
": %s not enough space (%llu < %llu)\n",
3183 dev
, maxsize
, size
);
3187 /* count total number of extents for merge */
3189 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
3191 i
+= dl
->extent_cnt
;
3193 maxsize
= merge_extents(super
, i
);
3194 if (maxsize
< size
) {
3196 fprintf(stderr
, Name
": not enough space after merge (%llu < %llu)\n",
3199 } else if (maxsize
== ~0ULL) {
3201 fprintf(stderr
, Name
": failed to merge %d extents\n", i
);
3205 *freesize
= maxsize
;
3210 static int reserve_space(struct supertype
*st
, int raiddisks
,
3211 unsigned long long size
, int chunk
,
3212 unsigned long long *freesize
)
3214 struct intel_super
*super
= st
->sb
;
3215 struct imsm_super
*mpb
= super
->anchor
;
3220 unsigned long long maxsize
;
3221 unsigned long long minsize
;
3225 /* find the largest common start free region of the possible disks */
3229 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
3235 /* don't activate new spares if we are orom constrained
3236 * and there is already a volume active in the container
3238 if (super
->orom
&& dl
->index
< 0 && mpb
->num_raid_devs
)
3241 e
= get_extents(super
, dl
);
3244 for (i
= 1; e
[i
-1].size
; i
++)
3252 maxsize
= merge_extents(super
, extent_cnt
);
3257 if (cnt
< raiddisks
||
3258 (super
->orom
&& used
&& used
!= raiddisks
) ||
3259 maxsize
< minsize
) {
3260 fprintf(stderr
, Name
": not enough devices with space to create array.\n");
3261 return 0; /* No enough free spaces large enough */
3273 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
3275 dl
->raiddisk
= cnt
++;
3282 static int validate_geometry_imsm(struct supertype
*st
, int level
, int layout
,
3283 int raiddisks
, int chunk
, unsigned long long size
,
3284 char *dev
, unsigned long long *freesize
,
3290 /* if given unused devices create a container
3291 * if given given devices in a container create a member volume
3293 if (level
== LEVEL_CONTAINER
) {
3294 /* Must be a fresh device to add to a container */
3295 return validate_geometry_imsm_container(st
, level
, layout
,
3296 raiddisks
, chunk
, size
,
3302 if (st
->sb
&& freesize
) {
3303 /* we are being asked to automatically layout a
3304 * new volume based on the current contents of
3305 * the container. If the the parameters can be
3306 * satisfied reserve_space will record the disks,
3307 * start offset, and size of the volume to be
3308 * created. add_to_super and getinfo_super
3309 * detect when autolayout is in progress.
3311 return reserve_space(st
, raiddisks
, size
, chunk
, freesize
);
3316 /* creating in a given container */
3317 return validate_geometry_imsm_volume(st
, level
, layout
,
3318 raiddisks
, chunk
, size
,
3319 dev
, freesize
, verbose
);
3322 /* limit creation to the following levels */
3334 /* This device needs to be a device in an 'imsm' container */
3335 fd
= open(dev
, O_RDONLY
|O_EXCL
, 0);
3339 Name
": Cannot create this array on device %s\n",
3344 if (errno
!= EBUSY
|| (fd
= open(dev
, O_RDONLY
, 0)) < 0) {
3346 fprintf(stderr
, Name
": Cannot open %s: %s\n",
3347 dev
, strerror(errno
));
3350 /* Well, it is in use by someone, maybe an 'imsm' container. */
3351 cfd
= open_container(fd
);
3355 fprintf(stderr
, Name
": Cannot use %s: It is busy\n",
3359 sra
= sysfs_read(cfd
, 0, GET_VERSION
);
3361 if (sra
&& sra
->array
.major_version
== -1 &&
3362 strcmp(sra
->text_version
, "imsm") == 0) {
3363 /* This is a member of a imsm container. Load the container
3364 * and try to create a volume
3366 struct intel_super
*super
;
3368 if (load_super_imsm_all(st
, cfd
, (void **) &super
, NULL
, 1) == 0) {
3370 st
->container_dev
= fd2devnum(cfd
);
3372 return validate_geometry_imsm_volume(st
, level
, layout
,
3378 } else /* may belong to another container */
3383 #endif /* MDASSEMBLE */
3385 static struct mdinfo
*container_content_imsm(struct supertype
*st
)
3387 /* Given a container loaded by load_super_imsm_all,
3388 * extract information about all the arrays into
3391 * For each imsm_dev create an mdinfo, fill it in,
3392 * then look for matching devices in super->disks
3393 * and create appropriate device mdinfo.
3395 struct intel_super
*super
= st
->sb
;
3396 struct imsm_super
*mpb
= super
->anchor
;
3397 struct mdinfo
*rest
= NULL
;
3400 /* do not assemble arrays that might have bad blocks */
3401 if (imsm_bbm_log_size(super
->anchor
)) {
3402 fprintf(stderr
, Name
": BBM log found in metadata. "
3403 "Cannot activate array(s).\n");
3407 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
3408 struct imsm_dev
*dev
= get_imsm_dev(super
, i
);
3409 struct imsm_map
*map
= get_imsm_map(dev
, 0);
3410 struct mdinfo
*this;
3413 /* do not publish arrays that are in the middle of an
3414 * unsupported migration
3416 if (dev
->vol
.migr_state
&&
3417 (migr_type(dev
) == MIGR_GEN_MIGR
||
3418 migr_type(dev
) == MIGR_STATE_CHANGE
)) {
3419 fprintf(stderr
, Name
": cannot assemble volume '%.16s':"
3420 " unsupported migration in progress\n",
3425 this = malloc(sizeof(*this));
3426 memset(this, 0, sizeof(*this));
3429 super
->current_vol
= i
;
3430 getinfo_super_imsm_volume(st
, this);
3431 for (slot
= 0 ; slot
< map
->num_members
; slot
++) {
3432 struct mdinfo
*info_d
;
3440 idx
= get_imsm_disk_idx(dev
, slot
);
3441 ord
= get_imsm_ord_tbl_ent(dev
, slot
);
3442 for (d
= super
->disks
; d
; d
= d
->next
)
3443 if (d
->index
== idx
)
3449 s
= d
? d
->disk
.status
: 0;
3450 if (s
& FAILED_DISK
)
3452 if (!(s
& USABLE_DISK
))
3454 if (ord
& IMSM_ORD_REBUILD
)
3458 * if we skip some disks the array will be assmebled degraded;
3459 * reset resync start to avoid a dirty-degraded situation
3461 * FIXME handle dirty degraded
3463 if (skip
&& !dev
->vol
.dirty
)
3464 this->resync_start
= ~0ULL;
3468 info_d
= malloc(sizeof(*info_d
));
3470 fprintf(stderr
, Name
": failed to allocate disk"
3471 " for volume %.16s\n", dev
->volume
);
3476 memset(info_d
, 0, sizeof(*info_d
));
3477 info_d
->next
= this->devs
;
3478 this->devs
= info_d
;
3480 info_d
->disk
.number
= d
->index
;
3481 info_d
->disk
.major
= d
->major
;
3482 info_d
->disk
.minor
= d
->minor
;
3483 info_d
->disk
.raid_disk
= slot
;
3485 this->array
.working_disks
++;
3487 info_d
->events
= __le32_to_cpu(mpb
->generation_num
);
3488 info_d
->data_offset
= __le32_to_cpu(map
->pba_of_lba0
);
3489 info_d
->component_size
= __le32_to_cpu(map
->blocks_per_member
);
3491 strcpy(info_d
->name
, d
->devname
);
3501 static int imsm_open_new(struct supertype
*c
, struct active_array
*a
,
3504 struct intel_super
*super
= c
->sb
;
3505 struct imsm_super
*mpb
= super
->anchor
;
3507 if (atoi(inst
) >= mpb
->num_raid_devs
) {
3508 fprintf(stderr
, "%s: subarry index %d, out of range\n",
3509 __func__
, atoi(inst
));
3513 dprintf("imsm: open_new %s\n", inst
);
3514 a
->info
.container_member
= atoi(inst
);
3518 static __u8
imsm_check_degraded(struct intel_super
*super
, struct imsm_dev
*dev
, int failed
)
3520 struct imsm_map
*map
= get_imsm_map(dev
, 0);
3523 return map
->map_state
== IMSM_T_STATE_UNINITIALIZED
?
3524 IMSM_T_STATE_UNINITIALIZED
: IMSM_T_STATE_NORMAL
;
3526 switch (get_imsm_raid_level(map
)) {
3528 return IMSM_T_STATE_FAILED
;
3531 if (failed
< map
->num_members
)
3532 return IMSM_T_STATE_DEGRADED
;
3534 return IMSM_T_STATE_FAILED
;
3539 * check to see if any mirrors have failed, otherwise we
3540 * are degraded. Even numbered slots are mirrored on
3544 /* gcc -Os complains that this is unused */
3545 int insync
= insync
;
3547 for (i
= 0; i
< map
->num_members
; i
++) {
3548 __u32 ord
= get_imsm_ord_tbl_ent(dev
, i
);
3549 int idx
= ord_to_idx(ord
);
3550 struct imsm_disk
*disk
;
3552 /* reset the potential in-sync count on even-numbered
3553 * slots. num_copies is always 2 for imsm raid10
3558 disk
= get_imsm_disk(super
, idx
);
3559 if (!disk
|| disk
->status
& FAILED_DISK
||
3560 ord
& IMSM_ORD_REBUILD
)
3563 /* no in-sync disks left in this mirror the
3567 return IMSM_T_STATE_FAILED
;
3570 return IMSM_T_STATE_DEGRADED
;
3574 return IMSM_T_STATE_DEGRADED
;
3576 return IMSM_T_STATE_FAILED
;
3582 return map
->map_state
;
3585 static int imsm_count_failed(struct intel_super
*super
, struct imsm_dev
*dev
)
3589 struct imsm_disk
*disk
;
3590 struct imsm_map
*map
= get_imsm_map(dev
, 0);
3591 struct imsm_map
*prev
= get_imsm_map(dev
, dev
->vol
.migr_state
);
3595 /* at the beginning of migration we set IMSM_ORD_REBUILD on
3596 * disks that are being rebuilt. New failures are recorded to
3597 * map[0]. So we look through all the disks we started with and
3598 * see if any failures are still present, or if any new ones
3601 * FIXME add support for online capacity expansion and
3602 * raid-level-migration
3604 for (i
= 0; i
< prev
->num_members
; i
++) {
3605 ord
= __le32_to_cpu(prev
->disk_ord_tbl
[i
]);
3606 ord
|= __le32_to_cpu(map
->disk_ord_tbl
[i
]);
3607 idx
= ord_to_idx(ord
);
3609 disk
= get_imsm_disk(super
, idx
);
3610 if (!disk
|| disk
->status
& FAILED_DISK
||
3611 ord
& IMSM_ORD_REBUILD
)
3618 static int is_resyncing(struct imsm_dev
*dev
)
3620 struct imsm_map
*migr_map
;
3622 if (!dev
->vol
.migr_state
)
3625 if (migr_type(dev
) == MIGR_INIT
||
3626 migr_type(dev
) == MIGR_REPAIR
)
3629 migr_map
= get_imsm_map(dev
, 1);
3631 if (migr_map
->map_state
== IMSM_T_STATE_NORMAL
)
3637 static int is_rebuilding(struct imsm_dev
*dev
)
3639 struct imsm_map
*migr_map
;
3641 if (!dev
->vol
.migr_state
)
3644 if (migr_type(dev
) != MIGR_REBUILD
)
3647 migr_map
= get_imsm_map(dev
, 1);
3649 if (migr_map
->map_state
== IMSM_T_STATE_DEGRADED
)
3655 /* return true if we recorded new information */
3656 static int mark_failure(struct imsm_dev
*dev
, struct imsm_disk
*disk
, int idx
)
3660 struct imsm_map
*map
;
3662 /* new failures are always set in map[0] */
3663 map
= get_imsm_map(dev
, 0);
3665 slot
= get_imsm_disk_slot(map
, idx
);
3669 ord
= __le32_to_cpu(map
->disk_ord_tbl
[slot
]);
3670 if ((disk
->status
& FAILED_DISK
) && (ord
& IMSM_ORD_REBUILD
))
3673 disk
->status
|= FAILED_DISK
;
3674 set_imsm_ord_tbl_ent(map
, slot
, idx
| IMSM_ORD_REBUILD
);
3675 if (~map
->failed_disk_num
== 0)
3676 map
->failed_disk_num
= slot
;
3680 static void mark_missing(struct imsm_dev
*dev
, struct imsm_disk
*disk
, int idx
)
3682 mark_failure(dev
, disk
, idx
);
3684 if (disk
->scsi_id
== __cpu_to_le32(~(__u32
)0))
3687 disk
->scsi_id
= __cpu_to_le32(~(__u32
)0);
3688 memmove(&disk
->serial
[0], &disk
->serial
[1], MAX_RAID_SERIAL_LEN
- 1);
3691 /* Handle dirty -> clean transititions and resync. Degraded and rebuild
3692 * states are handled in imsm_set_disk() with one exception, when a
3693 * resync is stopped due to a new failure this routine will set the
3694 * 'degraded' state for the array.
3696 static int imsm_set_array_state(struct active_array
*a
, int consistent
)
3698 int inst
= a
->info
.container_member
;
3699 struct intel_super
*super
= a
->container
->sb
;
3700 struct imsm_dev
*dev
= get_imsm_dev(super
, inst
);
3701 struct imsm_map
*map
= get_imsm_map(dev
, 0);
3702 int failed
= imsm_count_failed(super
, dev
);
3703 __u8 map_state
= imsm_check_degraded(super
, dev
, failed
);
3705 /* before we activate this array handle any missing disks */
3706 if (consistent
== 2 && super
->missing
) {
3709 dprintf("imsm: mark missing\n");
3710 end_migration(dev
, map_state
);
3711 for (dl
= super
->missing
; dl
; dl
= dl
->next
)
3712 mark_missing(dev
, &dl
->disk
, dl
->index
);
3713 super
->updates_pending
++;
3716 if (consistent
== 2 &&
3717 (!is_resync_complete(a
) ||
3718 map_state
!= IMSM_T_STATE_NORMAL
||
3719 dev
->vol
.migr_state
))
3722 if (is_resync_complete(a
)) {
3723 /* complete intialization / resync,
3724 * recovery and interrupted recovery is completed in
3727 if (is_resyncing(dev
)) {
3728 dprintf("imsm: mark resync done\n");
3729 end_migration(dev
, map_state
);
3730 super
->updates_pending
++;
3732 } else if (!is_resyncing(dev
) && !failed
) {
3733 /* mark the start of the init process if nothing is failed */
3734 dprintf("imsm: mark resync start (%llu)\n", a
->resync_start
);
3735 if (map
->map_state
== IMSM_T_STATE_UNINITIALIZED
)
3736 migrate(dev
, IMSM_T_STATE_NORMAL
, MIGR_INIT
);
3738 migrate(dev
, IMSM_T_STATE_NORMAL
, MIGR_REPAIR
);
3739 super
->updates_pending
++;
3742 /* FIXME check if we can update curr_migr_unit from resync_start */
3744 /* mark dirty / clean */
3745 if (dev
->vol
.dirty
!= !consistent
) {
3746 dprintf("imsm: mark '%s' (%llu)\n",
3747 consistent
? "clean" : "dirty", a
->resync_start
);
3752 super
->updates_pending
++;
3757 static void imsm_set_disk(struct active_array
*a
, int n
, int state
)
3759 int inst
= a
->info
.container_member
;
3760 struct intel_super
*super
= a
->container
->sb
;
3761 struct imsm_dev
*dev
= get_imsm_dev(super
, inst
);
3762 struct imsm_map
*map
= get_imsm_map(dev
, 0);
3763 struct imsm_disk
*disk
;
3768 if (n
> map
->num_members
)
3769 fprintf(stderr
, "imsm: set_disk %d out of range 0..%d\n",
3770 n
, map
->num_members
- 1);
3775 dprintf("imsm: set_disk %d:%x\n", n
, state
);
3777 ord
= get_imsm_ord_tbl_ent(dev
, n
);
3778 disk
= get_imsm_disk(super
, ord_to_idx(ord
));
3780 /* check for new failures */
3781 if (state
& DS_FAULTY
) {
3782 if (mark_failure(dev
, disk
, ord_to_idx(ord
)))
3783 super
->updates_pending
++;
3786 /* check if in_sync */
3787 if (state
& DS_INSYNC
&& ord
& IMSM_ORD_REBUILD
&& is_rebuilding(dev
)) {
3788 struct imsm_map
*migr_map
= get_imsm_map(dev
, 1);
3790 set_imsm_ord_tbl_ent(migr_map
, n
, ord_to_idx(ord
));
3791 super
->updates_pending
++;
3794 failed
= imsm_count_failed(super
, dev
);
3795 map_state
= imsm_check_degraded(super
, dev
, failed
);
3797 /* check if recovery complete, newly degraded, or failed */
3798 if (map_state
== IMSM_T_STATE_NORMAL
&& is_rebuilding(dev
)) {
3799 end_migration(dev
, map_state
);
3800 map
= get_imsm_map(dev
, 0);
3801 map
->failed_disk_num
= ~0;
3802 super
->updates_pending
++;
3803 } else if (map_state
== IMSM_T_STATE_DEGRADED
&&
3804 map
->map_state
!= map_state
&&
3805 !dev
->vol
.migr_state
) {
3806 dprintf("imsm: mark degraded\n");
3807 map
->map_state
= map_state
;
3808 super
->updates_pending
++;
3809 } else if (map_state
== IMSM_T_STATE_FAILED
&&
3810 map
->map_state
!= map_state
) {
3811 dprintf("imsm: mark failed\n");
3812 end_migration(dev
, map_state
);
3813 super
->updates_pending
++;
3817 static int store_imsm_mpb(int fd
, struct intel_super
*super
)
3819 struct imsm_super
*mpb
= super
->anchor
;
3820 __u32 mpb_size
= __le32_to_cpu(mpb
->mpb_size
);
3821 unsigned long long dsize
;
3822 unsigned long long sectors
;
3824 get_dev_size(fd
, NULL
, &dsize
);
3826 if (mpb_size
> 512) {
3827 /* -1 to account for anchor */
3828 sectors
= mpb_sectors(mpb
) - 1;
3830 /* write the extended mpb to the sectors preceeding the anchor */
3831 if (lseek64(fd
, dsize
- (512 * (2 + sectors
)), SEEK_SET
) < 0)
3834 if (write(fd
, super
->buf
+ 512, 512 * sectors
) != 512 * sectors
)
3838 /* first block is stored on second to last sector of the disk */
3839 if (lseek64(fd
, dsize
- (512 * 2), SEEK_SET
) < 0)
3842 if (write(fd
, super
->buf
, 512) != 512)
3848 static void imsm_sync_metadata(struct supertype
*container
)
3850 struct intel_super
*super
= container
->sb
;
3852 if (!super
->updates_pending
)
3855 write_super_imsm(super
, 0);
3857 super
->updates_pending
= 0;
3860 static struct dl
*imsm_readd(struct intel_super
*super
, int idx
, struct active_array
*a
)
3862 struct imsm_dev
*dev
= get_imsm_dev(super
, a
->info
.container_member
);
3863 int i
= get_imsm_disk_idx(dev
, idx
);
3866 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
3870 if (dl
&& dl
->disk
.status
& FAILED_DISK
)
3874 dprintf("%s: found %x:%x\n", __func__
, dl
->major
, dl
->minor
);
3879 static struct dl
*imsm_add_spare(struct intel_super
*super
, int slot
,
3880 struct active_array
*a
, int activate_new
)
3882 struct imsm_dev
*dev
= get_imsm_dev(super
, a
->info
.container_member
);
3883 int idx
= get_imsm_disk_idx(dev
, slot
);
3884 struct imsm_super
*mpb
= super
->anchor
;
3885 struct imsm_map
*map
;
3886 unsigned long long pos
;
3895 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
3896 /* If in this array, skip */
3897 for (d
= a
->info
.devs
; d
; d
= d
->next
)
3898 if (d
->state_fd
>= 0 &&
3899 d
->disk
.major
== dl
->major
&&
3900 d
->disk
.minor
== dl
->minor
) {
3901 dprintf("%x:%x already in array\n", dl
->major
, dl
->minor
);
3907 /* skip in use or failed drives */
3908 if (dl
->disk
.status
& FAILED_DISK
|| idx
== dl
->index
||
3910 dprintf("%x:%x status (failed: %d index: %d)\n",
3911 dl
->major
, dl
->minor
,
3912 (dl
->disk
.status
& FAILED_DISK
) == FAILED_DISK
, idx
);
3916 /* skip pure spares when we are looking for partially
3917 * assimilated drives
3919 if (dl
->index
== -1 && !activate_new
)
3922 /* Does this unused device have the requisite free space?
3923 * It needs to be able to cover all member volumes
3925 ex
= get_extents(super
, dl
);
3927 dprintf("cannot get extents\n");
3930 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
3931 dev
= get_imsm_dev(super
, i
);
3932 map
= get_imsm_map(dev
, 0);
3934 /* check if this disk is already a member of
3937 if (get_imsm_disk_slot(map
, dl
->index
) >= 0)
3943 array_start
= __le32_to_cpu(map
->pba_of_lba0
);
3944 array_end
= array_start
+
3945 __le32_to_cpu(map
->blocks_per_member
) - 1;
3948 /* check that we can start at pba_of_lba0 with
3949 * blocks_per_member of space
3951 if (array_start
>= pos
&& array_end
< ex
[j
].start
) {
3955 pos
= ex
[j
].start
+ ex
[j
].size
;
3957 } while (ex
[j
-1].size
);
3964 if (i
< mpb
->num_raid_devs
) {
3965 dprintf("%x:%x does not have %u to %u available\n",
3966 dl
->major
, dl
->minor
, array_start
, array_end
);
3976 static struct mdinfo
*imsm_activate_spare(struct active_array
*a
,
3977 struct metadata_update
**updates
)
3980 * Find a device with unused free space and use it to replace a
3981 * failed/vacant region in an array. We replace failed regions one a
3982 * array at a time. The result is that a new spare disk will be added
3983 * to the first failed array and after the monitor has finished
3984 * propagating failures the remainder will be consumed.
3986 * FIXME add a capability for mdmon to request spares from another
3990 struct intel_super
*super
= a
->container
->sb
;
3991 int inst
= a
->info
.container_member
;
3992 struct imsm_dev
*dev
= get_imsm_dev(super
, inst
);
3993 struct imsm_map
*map
= get_imsm_map(dev
, 0);
3994 int failed
= a
->info
.array
.raid_disks
;
3995 struct mdinfo
*rv
= NULL
;
3998 struct metadata_update
*mu
;
4000 struct imsm_update_activate_spare
*u
;
4004 for (d
= a
->info
.devs
; d
; d
= d
->next
) {
4005 if ((d
->curr_state
& DS_FAULTY
) &&
4007 /* wait for Removal to happen */
4009 if (d
->state_fd
>= 0)
4013 dprintf("imsm: activate spare: inst=%d failed=%d (%d) level=%d\n",
4014 inst
, failed
, a
->info
.array
.raid_disks
, a
->info
.array
.level
);
4015 if (imsm_check_degraded(super
, dev
, failed
) != IMSM_T_STATE_DEGRADED
)
4018 /* For each slot, if it is not working, find a spare */
4019 for (i
= 0; i
< a
->info
.array
.raid_disks
; i
++) {
4020 for (d
= a
->info
.devs
; d
; d
= d
->next
)
4021 if (d
->disk
.raid_disk
== i
)
4023 dprintf("found %d: %p %x\n", i
, d
, d
?d
->curr_state
:0);
4024 if (d
&& (d
->state_fd
>= 0))
4028 * OK, this device needs recovery. Try to re-add the
4029 * previous occupant of this slot, if this fails see if
4030 * we can continue the assimilation of a spare that was
4031 * partially assimilated, finally try to activate a new
4034 dl
= imsm_readd(super
, i
, a
);
4036 dl
= imsm_add_spare(super
, i
, a
, 0);
4038 dl
= imsm_add_spare(super
, i
, a
, 1);
4042 /* found a usable disk with enough space */
4043 di
= malloc(sizeof(*di
));
4046 memset(di
, 0, sizeof(*di
));
4048 /* dl->index will be -1 in the case we are activating a
4049 * pristine spare. imsm_process_update() will create a
4050 * new index in this case. Once a disk is found to be
4051 * failed in all member arrays it is kicked from the
4054 di
->disk
.number
= dl
->index
;
4056 /* (ab)use di->devs to store a pointer to the device
4059 di
->devs
= (struct mdinfo
*) dl
;
4061 di
->disk
.raid_disk
= i
;
4062 di
->disk
.major
= dl
->major
;
4063 di
->disk
.minor
= dl
->minor
;
4065 di
->data_offset
= __le32_to_cpu(map
->pba_of_lba0
);
4066 di
->component_size
= a
->info
.component_size
;
4067 di
->container_member
= inst
;
4068 super
->random
= random32();
4072 dprintf("%x:%x to be %d at %llu\n", dl
->major
, dl
->minor
,
4073 i
, di
->data_offset
);
4079 /* No spares found */
4081 /* Now 'rv' has a list of devices to return.
4082 * Create a metadata_update record to update the
4083 * disk_ord_tbl for the array
4085 mu
= malloc(sizeof(*mu
));
4087 mu
->buf
= malloc(sizeof(struct imsm_update_activate_spare
) * num_spares
);
4088 if (mu
->buf
== NULL
) {
4095 struct mdinfo
*n
= rv
->next
;
4104 mu
->len
= sizeof(struct imsm_update_activate_spare
) * num_spares
;
4105 mu
->next
= *updates
;
4106 u
= (struct imsm_update_activate_spare
*) mu
->buf
;
4108 for (di
= rv
; di
; di
= di
->next
) {
4109 u
->type
= update_activate_spare
;
4110 u
->dl
= (struct dl
*) di
->devs
;
4112 u
->slot
= di
->disk
.raid_disk
;
4123 static int disks_overlap(struct intel_super
*super
, int idx
, struct imsm_update_create_array
*u
)
4125 struct imsm_dev
*dev
= get_imsm_dev(super
, idx
);
4126 struct imsm_map
*map
= get_imsm_map(dev
, 0);
4127 struct imsm_map
*new_map
= get_imsm_map(&u
->dev
, 0);
4128 struct disk_info
*inf
= get_disk_info(u
);
4129 struct imsm_disk
*disk
;
4133 for (i
= 0; i
< map
->num_members
; i
++) {
4134 disk
= get_imsm_disk(super
, get_imsm_disk_idx(dev
, i
));
4135 for (j
= 0; j
< new_map
->num_members
; j
++)
4136 if (serialcmp(disk
->serial
, inf
[j
].serial
) == 0)
4143 static void imsm_delete(struct intel_super
*super
, struct dl
**dlp
, int index
);
4145 static void imsm_process_update(struct supertype
*st
,
4146 struct metadata_update
*update
)
4149 * crack open the metadata_update envelope to find the update record
4150 * update can be one of:
4151 * update_activate_spare - a spare device has replaced a failed
4152 * device in an array, update the disk_ord_tbl. If this disk is
4153 * present in all member arrays then also clear the SPARE_DISK
4156 struct intel_super
*super
= st
->sb
;
4157 struct imsm_super
*mpb
;
4158 enum imsm_update_type type
= *(enum imsm_update_type
*) update
->buf
;
4160 /* update requires a larger buf but the allocation failed */
4161 if (super
->next_len
&& !super
->next_buf
) {
4162 super
->next_len
= 0;
4166 if (super
->next_buf
) {
4167 memcpy(super
->next_buf
, super
->buf
, super
->len
);
4169 super
->len
= super
->next_len
;
4170 super
->buf
= super
->next_buf
;
4172 super
->next_len
= 0;
4173 super
->next_buf
= NULL
;
4176 mpb
= super
->anchor
;
4179 case update_activate_spare
: {
4180 struct imsm_update_activate_spare
*u
= (void *) update
->buf
;
4181 struct imsm_dev
*dev
= get_imsm_dev(super
, u
->array
);
4182 struct imsm_map
*map
= get_imsm_map(dev
, 0);
4183 struct imsm_map
*migr_map
;
4184 struct active_array
*a
;
4185 struct imsm_disk
*disk
;
4190 int victim
= get_imsm_disk_idx(dev
, u
->slot
);
4193 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
4198 fprintf(stderr
, "error: imsm_activate_spare passed "
4199 "an unknown disk (index: %d)\n",
4204 super
->updates_pending
++;
4206 /* count failures (excluding rebuilds and the victim)
4207 * to determine map[0] state
4210 for (i
= 0; i
< map
->num_members
; i
++) {
4213 disk
= get_imsm_disk(super
, get_imsm_disk_idx(dev
, i
));
4214 if (!disk
|| disk
->status
& FAILED_DISK
)
4218 /* adding a pristine spare, assign a new index */
4219 if (dl
->index
< 0) {
4220 dl
->index
= super
->anchor
->num_disks
;
4221 super
->anchor
->num_disks
++;
4224 disk
->status
|= CONFIGURED_DISK
;
4225 disk
->status
&= ~SPARE_DISK
;
4228 to_state
= imsm_check_degraded(super
, dev
, failed
);
4229 map
->map_state
= IMSM_T_STATE_DEGRADED
;
4230 migrate(dev
, to_state
, MIGR_REBUILD
);
4231 migr_map
= get_imsm_map(dev
, 1);
4232 set_imsm_ord_tbl_ent(map
, u
->slot
, dl
->index
);
4233 set_imsm_ord_tbl_ent(migr_map
, u
->slot
, dl
->index
| IMSM_ORD_REBUILD
);
4235 /* update the family_num to mark a new container
4236 * generation, being careful to record the existing
4237 * family_num in orig_family_num to clean up after
4238 * earlier mdadm versions that neglected to set it.
4240 if (mpb
->orig_family_num
== 0)
4241 mpb
->orig_family_num
= mpb
->family_num
;
4242 mpb
->family_num
+= super
->random
;
4244 /* count arrays using the victim in the metadata */
4246 for (a
= st
->arrays
; a
; a
= a
->next
) {
4247 dev
= get_imsm_dev(super
, a
->info
.container_member
);
4248 map
= get_imsm_map(dev
, 0);
4250 if (get_imsm_disk_slot(map
, victim
) >= 0)
4254 /* delete the victim if it is no longer being
4260 /* We know that 'manager' isn't touching anything,
4261 * so it is safe to delete
4263 for (dlp
= &super
->disks
; *dlp
; dlp
= &(*dlp
)->next
)
4264 if ((*dlp
)->index
== victim
)
4267 /* victim may be on the missing list */
4269 for (dlp
= &super
->missing
; *dlp
; dlp
= &(*dlp
)->next
)
4270 if ((*dlp
)->index
== victim
)
4272 imsm_delete(super
, dlp
, victim
);
4276 case update_create_array
: {
4277 /* someone wants to create a new array, we need to be aware of
4278 * a few races/collisions:
4279 * 1/ 'Create' called by two separate instances of mdadm
4280 * 2/ 'Create' versus 'activate_spare': mdadm has chosen
4281 * devices that have since been assimilated via
4283 * In the event this update can not be carried out mdadm will
4284 * (FIX ME) notice that its update did not take hold.
4286 struct imsm_update_create_array
*u
= (void *) update
->buf
;
4287 struct intel_dev
*dv
;
4288 struct imsm_dev
*dev
;
4289 struct imsm_map
*map
, *new_map
;
4290 unsigned long long start
, end
;
4291 unsigned long long new_start
, new_end
;
4293 struct disk_info
*inf
;
4296 /* handle racing creates: first come first serve */
4297 if (u
->dev_idx
< mpb
->num_raid_devs
) {
4298 dprintf("%s: subarray %d already defined\n",
4299 __func__
, u
->dev_idx
);
4303 /* check update is next in sequence */
4304 if (u
->dev_idx
!= mpb
->num_raid_devs
) {
4305 dprintf("%s: can not create array %d expected index %d\n",
4306 __func__
, u
->dev_idx
, mpb
->num_raid_devs
);
4310 new_map
= get_imsm_map(&u
->dev
, 0);
4311 new_start
= __le32_to_cpu(new_map
->pba_of_lba0
);
4312 new_end
= new_start
+ __le32_to_cpu(new_map
->blocks_per_member
);
4313 inf
= get_disk_info(u
);
4315 /* handle activate_spare versus create race:
4316 * check to make sure that overlapping arrays do not include
4319 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
4320 dev
= get_imsm_dev(super
, i
);
4321 map
= get_imsm_map(dev
, 0);
4322 start
= __le32_to_cpu(map
->pba_of_lba0
);
4323 end
= start
+ __le32_to_cpu(map
->blocks_per_member
);
4324 if ((new_start
>= start
&& new_start
<= end
) ||
4325 (start
>= new_start
&& start
<= new_end
))
4330 if (disks_overlap(super
, i
, u
)) {
4331 dprintf("%s: arrays overlap\n", __func__
);
4336 /* check that prepare update was successful */
4337 if (!update
->space
) {
4338 dprintf("%s: prepare update failed\n", __func__
);
4342 /* check that all disks are still active before committing
4343 * changes. FIXME: could we instead handle this by creating a
4344 * degraded array? That's probably not what the user expects,
4345 * so better to drop this update on the floor.
4347 for (i
= 0; i
< new_map
->num_members
; i
++) {
4348 dl
= serial_to_dl(inf
[i
].serial
, super
);
4350 dprintf("%s: disk disappeared\n", __func__
);
4355 super
->updates_pending
++;
4357 /* convert spares to members and fixup ord_tbl */
4358 for (i
= 0; i
< new_map
->num_members
; i
++) {
4359 dl
= serial_to_dl(inf
[i
].serial
, super
);
4360 if (dl
->index
== -1) {
4361 dl
->index
= mpb
->num_disks
;
4363 dl
->disk
.status
|= CONFIGURED_DISK
;
4364 dl
->disk
.status
&= ~SPARE_DISK
;
4366 set_imsm_ord_tbl_ent(new_map
, i
, dl
->index
);
4371 update
->space
= NULL
;
4372 imsm_copy_dev(dev
, &u
->dev
);
4373 dv
->index
= u
->dev_idx
;
4374 dv
->next
= super
->devlist
;
4375 super
->devlist
= dv
;
4376 mpb
->num_raid_devs
++;
4378 imsm_update_version_info(super
);
4381 /* mdmon knows how to release update->space, but not
4382 * ((struct intel_dev *) update->space)->dev
4384 if (update
->space
) {
4390 case update_add_disk
:
4392 /* we may be able to repair some arrays if disks are
4395 struct active_array
*a
;
4397 super
->updates_pending
++;
4398 for (a
= st
->arrays
; a
; a
= a
->next
)
4399 a
->check_degraded
= 1;
4401 /* add some spares to the metadata */
4402 while (super
->add
) {
4406 super
->add
= al
->next
;
4407 al
->next
= super
->disks
;
4409 dprintf("%s: added %x:%x\n",
4410 __func__
, al
->major
, al
->minor
);
4417 static void imsm_prepare_update(struct supertype
*st
,
4418 struct metadata_update
*update
)
4421 * Allocate space to hold new disk entries, raid-device entries or a new
4422 * mpb if necessary. The manager synchronously waits for updates to
4423 * complete in the monitor, so new mpb buffers allocated here can be
4424 * integrated by the monitor thread without worrying about live pointers
4425 * in the manager thread.
4427 enum imsm_update_type type
= *(enum imsm_update_type
*) update
->buf
;
4428 struct intel_super
*super
= st
->sb
;
4429 struct imsm_super
*mpb
= super
->anchor
;
4434 case update_create_array
: {
4435 struct imsm_update_create_array
*u
= (void *) update
->buf
;
4436 struct intel_dev
*dv
;
4437 struct imsm_dev
*dev
= &u
->dev
;
4438 struct imsm_map
*map
= get_imsm_map(dev
, 0);
4440 struct disk_info
*inf
;
4444 inf
= get_disk_info(u
);
4445 len
= sizeof_imsm_dev(dev
, 1);
4446 /* allocate a new super->devlist entry */
4447 dv
= malloc(sizeof(*dv
));
4449 dv
->dev
= malloc(len
);
4454 update
->space
= NULL
;
4458 /* count how many spares will be converted to members */
4459 for (i
= 0; i
< map
->num_members
; i
++) {
4460 dl
= serial_to_dl(inf
[i
].serial
, super
);
4462 /* hmm maybe it failed?, nothing we can do about
4467 if (count_memberships(dl
, super
) == 0)
4470 len
+= activate
* sizeof(struct imsm_disk
);
4477 /* check if we need a larger metadata buffer */
4478 if (super
->next_buf
)
4479 buf_len
= super
->next_len
;
4481 buf_len
= super
->len
;
4483 if (__le32_to_cpu(mpb
->mpb_size
) + len
> buf_len
) {
4484 /* ok we need a larger buf than what is currently allocated
4485 * if this allocation fails process_update will notice that
4486 * ->next_len is set and ->next_buf is NULL
4488 buf_len
= ROUND_UP(__le32_to_cpu(mpb
->mpb_size
) + len
, 512);
4489 if (super
->next_buf
)
4490 free(super
->next_buf
);
4492 super
->next_len
= buf_len
;
4493 if (posix_memalign(&super
->next_buf
, 512, buf_len
) == 0)
4494 memset(super
->next_buf
, 0, buf_len
);
4496 super
->next_buf
= NULL
;
4500 /* must be called while manager is quiesced */
4501 static void imsm_delete(struct intel_super
*super
, struct dl
**dlp
, int index
)
4503 struct imsm_super
*mpb
= super
->anchor
;
4505 struct imsm_dev
*dev
;
4506 struct imsm_map
*map
;
4507 int i
, j
, num_members
;
4510 dprintf("%s: deleting device[%d] from imsm_super\n",
4513 /* shift all indexes down one */
4514 for (iter
= super
->disks
; iter
; iter
= iter
->next
)
4515 if (iter
->index
> index
)
4517 for (iter
= super
->missing
; iter
; iter
= iter
->next
)
4518 if (iter
->index
> index
)
4521 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
4522 dev
= get_imsm_dev(super
, i
);
4523 map
= get_imsm_map(dev
, 0);
4524 num_members
= map
->num_members
;
4525 for (j
= 0; j
< num_members
; j
++) {
4526 /* update ord entries being careful not to propagate
4527 * ord-flags to the first map
4529 ord
= get_imsm_ord_tbl_ent(dev
, j
);
4531 if (ord_to_idx(ord
) <= index
)
4534 map
= get_imsm_map(dev
, 0);
4535 set_imsm_ord_tbl_ent(map
, j
, ord_to_idx(ord
- 1));
4536 map
= get_imsm_map(dev
, 1);
4538 set_imsm_ord_tbl_ent(map
, j
, ord
- 1);
4543 super
->updates_pending
++;
4545 struct dl
*dl
= *dlp
;
4547 *dlp
= (*dlp
)->next
;
4548 __free_imsm_disk(dl
);
4551 #endif /* MDASSEMBLE */
4553 struct superswitch super_imsm
= {
4555 .examine_super
= examine_super_imsm
,
4556 .brief_examine_super
= brief_examine_super_imsm
,
4557 .export_examine_super
= export_examine_super_imsm
,
4558 .detail_super
= detail_super_imsm
,
4559 .brief_detail_super
= brief_detail_super_imsm
,
4560 .write_init_super
= write_init_super_imsm
,
4561 .validate_geometry
= validate_geometry_imsm
,
4562 .add_to_super
= add_to_super_imsm
,
4563 .detail_platform
= detail_platform_imsm
,
4565 .match_home
= match_home_imsm
,
4566 .uuid_from_super
= uuid_from_super_imsm
,
4567 .getinfo_super
= getinfo_super_imsm
,
4568 .update_super
= update_super_imsm
,
4570 .avail_size
= avail_size_imsm
,
4572 .compare_super
= compare_super_imsm
,
4574 .load_super
= load_super_imsm
,
4575 .init_super
= init_super_imsm
,
4576 .store_super
= store_zero_imsm
,
4577 .free_super
= free_super_imsm
,
4578 .match_metadata_desc
= match_metadata_desc_imsm
,
4579 .container_content
= container_content_imsm
,
4580 .default_layout
= imsm_level_to_layout
,
4587 .open_new
= imsm_open_new
,
4588 .load_super
= load_super_imsm
,
4589 .set_array_state
= imsm_set_array_state
,
4590 .set_disk
= imsm_set_disk
,
4591 .sync_metadata
= imsm_sync_metadata
,
4592 .activate_spare
= imsm_activate_spare
,
4593 .process_update
= imsm_process_update
,
4594 .prepare_update
= imsm_prepare_update
,
4595 #endif /* MDASSEMBLE */