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 struct intel_dev
*devlist
;
254 __u8 serial
[MAX_RAID_SERIAL_LEN
];
257 struct imsm_disk disk
;
260 struct extent
*e
; /* for determining freespace @ create */
261 int raiddisk
; /* slot to fill in autolayout */
263 struct dl
*add
; /* list of disks to add while mdmon active */
264 struct dl
*missing
; /* disks removed while we weren't looking */
265 struct bbm_log
*bbm_log
;
266 const char *hba
; /* device path of the raid controller for this metadata */
267 const struct imsm_orom
*orom
; /* platform firmware support */
271 unsigned long long start
, size
;
274 /* definition of messages passed to imsm_process_update */
275 enum imsm_update_type
{
276 update_activate_spare
,
281 struct imsm_update_activate_spare
{
282 enum imsm_update_type type
;
286 struct imsm_update_activate_spare
*next
;
290 __u8 serial
[MAX_RAID_SERIAL_LEN
];
293 struct imsm_update_create_array
{
294 enum imsm_update_type type
;
299 struct imsm_update_add_disk
{
300 enum imsm_update_type type
;
303 static struct supertype
*match_metadata_desc_imsm(char *arg
)
305 struct supertype
*st
;
307 if (strcmp(arg
, "imsm") != 0 &&
308 strcmp(arg
, "default") != 0
312 st
= malloc(sizeof(*st
));
313 memset(st
, 0, sizeof(*st
));
314 st
->ss
= &super_imsm
;
315 st
->max_devs
= IMSM_MAX_DEVICES
;
316 st
->minor_version
= 0;
322 static __u8
*get_imsm_version(struct imsm_super
*mpb
)
324 return &mpb
->sig
[MPB_SIG_LEN
];
328 /* retrieve a disk directly from the anchor when the anchor is known to be
329 * up-to-date, currently only at load time
331 static struct imsm_disk
*__get_imsm_disk(struct imsm_super
*mpb
, __u8 index
)
333 if (index
>= mpb
->num_disks
)
335 return &mpb
->disk
[index
];
339 /* retrieve a disk from the parsed metadata */
340 static struct imsm_disk
*get_imsm_disk(struct intel_super
*super
, __u8 index
)
344 for (d
= super
->disks
; d
; d
= d
->next
)
345 if (d
->index
== index
)
352 /* generate a checksum directly from the anchor when the anchor is known to be
353 * up-to-date, currently only at load or write_super after coalescing
355 static __u32
__gen_imsm_checksum(struct imsm_super
*mpb
)
357 __u32 end
= mpb
->mpb_size
/ sizeof(end
);
358 __u32
*p
= (__u32
*) mpb
;
362 sum
+= __le32_to_cpu(*p
);
366 return sum
- __le32_to_cpu(mpb
->check_sum
);
369 static size_t sizeof_imsm_map(struct imsm_map
*map
)
371 return sizeof(struct imsm_map
) + sizeof(__u32
) * (map
->num_members
- 1);
374 struct imsm_map
*get_imsm_map(struct imsm_dev
*dev
, int second_map
)
376 struct imsm_map
*map
= &dev
->vol
.map
[0];
378 if (second_map
&& !dev
->vol
.migr_state
)
380 else if (second_map
) {
383 return ptr
+ sizeof_imsm_map(map
);
389 /* return the size of the device.
390 * migr_state increases the returned size if map[0] were to be duplicated
392 static size_t sizeof_imsm_dev(struct imsm_dev
*dev
, int migr_state
)
394 size_t size
= sizeof(*dev
) - sizeof(struct imsm_map
) +
395 sizeof_imsm_map(get_imsm_map(dev
, 0));
397 /* migrating means an additional map */
398 if (dev
->vol
.migr_state
)
399 size
+= sizeof_imsm_map(get_imsm_map(dev
, 1));
401 size
+= sizeof_imsm_map(get_imsm_map(dev
, 0));
407 /* retrieve disk serial number list from a metadata update */
408 static struct disk_info
*get_disk_info(struct imsm_update_create_array
*update
)
411 struct disk_info
*inf
;
413 inf
= u
+ sizeof(*update
) - sizeof(struct imsm_dev
) +
414 sizeof_imsm_dev(&update
->dev
, 0);
420 static struct imsm_dev
*__get_imsm_dev(struct imsm_super
*mpb
, __u8 index
)
426 if (index
>= mpb
->num_raid_devs
)
429 /* devices start after all disks */
430 offset
= ((void *) &mpb
->disk
[mpb
->num_disks
]) - _mpb
;
432 for (i
= 0; i
<= index
; i
++)
434 return _mpb
+ offset
;
436 offset
+= sizeof_imsm_dev(_mpb
+ offset
, 0);
441 static struct imsm_dev
*get_imsm_dev(struct intel_super
*super
, __u8 index
)
443 struct intel_dev
*dv
;
445 if (index
>= super
->anchor
->num_raid_devs
)
447 for (dv
= super
->devlist
; dv
; dv
= dv
->next
)
448 if (dv
->index
== index
)
453 static __u32
get_imsm_ord_tbl_ent(struct imsm_dev
*dev
, int slot
)
455 struct imsm_map
*map
;
457 if (dev
->vol
.migr_state
)
458 map
= get_imsm_map(dev
, 1);
460 map
= get_imsm_map(dev
, 0);
462 /* top byte identifies disk under rebuild */
463 return __le32_to_cpu(map
->disk_ord_tbl
[slot
]);
466 #define ord_to_idx(ord) (((ord) << 8) >> 8)
467 static __u32
get_imsm_disk_idx(struct imsm_dev
*dev
, int slot
)
469 __u32 ord
= get_imsm_ord_tbl_ent(dev
, slot
);
471 return ord_to_idx(ord
);
474 static void set_imsm_ord_tbl_ent(struct imsm_map
*map
, int slot
, __u32 ord
)
476 map
->disk_ord_tbl
[slot
] = __cpu_to_le32(ord
);
479 static int get_imsm_disk_slot(struct imsm_map
*map
, int idx
)
484 for (slot
= 0; slot
< map
->num_members
; slot
++) {
485 ord
= __le32_to_cpu(map
->disk_ord_tbl
[slot
]);
486 if (ord_to_idx(ord
) == idx
)
493 static int get_imsm_raid_level(struct imsm_map
*map
)
495 if (map
->raid_level
== 1) {
496 if (map
->num_members
== 2)
502 return map
->raid_level
;
505 static int cmp_extent(const void *av
, const void *bv
)
507 const struct extent
*a
= av
;
508 const struct extent
*b
= bv
;
509 if (a
->start
< b
->start
)
511 if (a
->start
> b
->start
)
516 static int count_memberships(struct dl
*dl
, struct intel_super
*super
)
521 for (i
= 0; i
< super
->anchor
->num_raid_devs
; i
++) {
522 struct imsm_dev
*dev
= get_imsm_dev(super
, i
);
523 struct imsm_map
*map
= get_imsm_map(dev
, 0);
525 if (get_imsm_disk_slot(map
, dl
->index
) >= 0)
532 static struct extent
*get_extents(struct intel_super
*super
, struct dl
*dl
)
534 /* find a list of used extents on the given physical device */
535 struct extent
*rv
, *e
;
537 int memberships
= count_memberships(dl
, super
);
538 __u32 reservation
= MPB_SECTOR_CNT
+ IMSM_RESERVED_SECTORS
;
540 rv
= malloc(sizeof(struct extent
) * (memberships
+ 1));
545 for (i
= 0; i
< super
->anchor
->num_raid_devs
; i
++) {
546 struct imsm_dev
*dev
= get_imsm_dev(super
, i
);
547 struct imsm_map
*map
= get_imsm_map(dev
, 0);
549 if (get_imsm_disk_slot(map
, dl
->index
) >= 0) {
550 e
->start
= __le32_to_cpu(map
->pba_of_lba0
);
551 e
->size
= __le32_to_cpu(map
->blocks_per_member
);
555 qsort(rv
, memberships
, sizeof(*rv
), cmp_extent
);
557 /* determine the start of the metadata
558 * when no raid devices are defined use the default
559 * ...otherwise allow the metadata to truncate the value
560 * as is the case with older versions of imsm
563 struct extent
*last
= &rv
[memberships
- 1];
566 remainder
= __le32_to_cpu(dl
->disk
.total_blocks
) -
567 (last
->start
+ last
->size
);
568 /* round down to 1k block to satisfy precision of the kernel
572 /* make sure remainder is still sane */
573 if (remainder
< ROUND_UP(super
->len
, 512) >> 9)
574 remainder
= ROUND_UP(super
->len
, 512) >> 9;
575 if (reservation
> remainder
)
576 reservation
= remainder
;
578 e
->start
= __le32_to_cpu(dl
->disk
.total_blocks
) - reservation
;
583 /* try to determine how much space is reserved for metadata from
584 * the last get_extents() entry, otherwise fallback to the
587 static __u32
imsm_reserved_sectors(struct intel_super
*super
, struct dl
*dl
)
593 /* for spares just return a minimal reservation which will grow
594 * once the spare is picked up by an array
597 return MPB_SECTOR_CNT
;
599 e
= get_extents(super
, dl
);
601 return MPB_SECTOR_CNT
+ IMSM_RESERVED_SECTORS
;
603 /* scroll to last entry */
604 for (i
= 0; e
[i
].size
; i
++)
607 rv
= __le32_to_cpu(dl
->disk
.total_blocks
) - e
[i
].start
;
615 static void print_imsm_dev(struct imsm_dev
*dev
, char *uuid
, int disk_idx
)
619 struct imsm_map
*map
= get_imsm_map(dev
, 0);
623 printf("[%.16s]:\n", dev
->volume
);
624 printf(" UUID : %s\n", uuid
);
625 printf(" RAID Level : %d\n", get_imsm_raid_level(map
));
626 printf(" Members : %d\n", map
->num_members
);
627 slot
= get_imsm_disk_slot(map
, disk_idx
);
629 ord
= get_imsm_ord_tbl_ent(dev
, slot
);
630 printf(" This Slot : %d%s\n", slot
,
631 ord
& IMSM_ORD_REBUILD
? " (out-of-sync)" : "");
633 printf(" This Slot : ?\n");
634 sz
= __le32_to_cpu(dev
->size_high
);
636 sz
+= __le32_to_cpu(dev
->size_low
);
637 printf(" Array Size : %llu%s\n", (unsigned long long)sz
,
638 human_size(sz
* 512));
639 sz
= __le32_to_cpu(map
->blocks_per_member
);
640 printf(" Per Dev Size : %llu%s\n", (unsigned long long)sz
,
641 human_size(sz
* 512));
642 printf(" Sector Offset : %u\n",
643 __le32_to_cpu(map
->pba_of_lba0
));
644 printf(" Num Stripes : %u\n",
645 __le32_to_cpu(map
->num_data_stripes
));
646 printf(" Chunk Size : %u KiB\n",
647 __le16_to_cpu(map
->blocks_per_strip
) / 2);
648 printf(" Reserved : %d\n", __le32_to_cpu(dev
->reserved_blocks
));
649 printf(" Migrate State : %s", dev
->vol
.migr_state
? "migrating" : "idle\n");
650 if (dev
->vol
.migr_state
) {
651 if (migr_type(dev
) == MIGR_INIT
)
652 printf(": initializing\n");
653 else if (migr_type(dev
) == MIGR_REBUILD
)
654 printf(": rebuilding\n");
655 else if (migr_type(dev
) == MIGR_VERIFY
)
657 else if (migr_type(dev
) == MIGR_GEN_MIGR
)
658 printf(": general migration\n");
659 else if (migr_type(dev
) == MIGR_STATE_CHANGE
)
660 printf(": state change\n");
661 else if (migr_type(dev
) == MIGR_REPAIR
)
662 printf(": repair\n");
664 printf(": <unknown:%d>\n", migr_type(dev
));
666 printf(" Map State : %s", map_state_str
[map
->map_state
]);
667 if (dev
->vol
.migr_state
) {
668 struct imsm_map
*map
= get_imsm_map(dev
, 1);
669 printf(" <-- %s", map_state_str
[map
->map_state
]);
672 printf(" Dirty State : %s\n", dev
->vol
.dirty
? "dirty" : "clean");
675 static void print_imsm_disk(struct imsm_super
*mpb
, int index
, __u32 reserved
)
677 struct imsm_disk
*disk
= __get_imsm_disk(mpb
, index
);
678 char str
[MAX_RAID_SERIAL_LEN
+ 1];
686 snprintf(str
, MAX_RAID_SERIAL_LEN
+ 1, "%s", disk
->serial
);
687 printf(" Disk%02d Serial : %s\n", index
, str
);
689 printf(" State :%s%s%s%s\n", s
&SPARE_DISK
? " spare" : "",
690 s
&CONFIGURED_DISK
? " active" : "",
691 s
&FAILED_DISK
? " failed" : "",
692 s
&USABLE_DISK
? " usable" : "");
693 printf(" Id : %08x\n", __le32_to_cpu(disk
->scsi_id
));
694 sz
= __le32_to_cpu(disk
->total_blocks
) - reserved
;
695 printf(" Usable Size : %llu%s\n", (unsigned long long)sz
,
696 human_size(sz
* 512));
699 static void getinfo_super_imsm(struct supertype
*st
, struct mdinfo
*info
);
701 static void examine_super_imsm(struct supertype
*st
, char *homehost
)
703 struct intel_super
*super
= st
->sb
;
704 struct imsm_super
*mpb
= super
->anchor
;
705 char str
[MAX_SIGNATURE_LENGTH
];
710 __u32 reserved
= imsm_reserved_sectors(super
, super
->disks
);
713 snprintf(str
, MPB_SIG_LEN
, "%s", mpb
->sig
);
714 printf(" Magic : %s\n", str
);
715 snprintf(str
, strlen(MPB_VERSION_RAID0
), "%s", get_imsm_version(mpb
));
716 printf(" Version : %s\n", get_imsm_version(mpb
));
717 printf(" Family : %08x\n", __le32_to_cpu(mpb
->family_num
));
718 printf(" Generation : %08x\n", __le32_to_cpu(mpb
->generation_num
));
719 getinfo_super_imsm(st
, &info
);
720 fname_from_uuid(st
, &info
, nbuf
, ':');
721 printf(" UUID : %s\n", nbuf
+ 5);
722 sum
= __le32_to_cpu(mpb
->check_sum
);
723 printf(" Checksum : %08x %s\n", sum
,
724 __gen_imsm_checksum(mpb
) == sum
? "correct" : "incorrect");
725 printf(" MPB Sectors : %d\n", mpb_sectors(mpb
));
726 printf(" Disks : %d\n", mpb
->num_disks
);
727 printf(" RAID Devices : %d\n", mpb
->num_raid_devs
);
728 print_imsm_disk(mpb
, super
->disks
->index
, reserved
);
729 if (super
->bbm_log
) {
730 struct bbm_log
*log
= super
->bbm_log
;
733 printf("Bad Block Management Log:\n");
734 printf(" Log Size : %d\n", __le32_to_cpu(mpb
->bbm_log_size
));
735 printf(" Signature : %x\n", __le32_to_cpu(log
->signature
));
736 printf(" Entry Count : %d\n", __le32_to_cpu(log
->entry_count
));
737 printf(" Spare Blocks : %d\n", __le32_to_cpu(log
->reserved_spare_block_count
));
738 printf(" First Spare : %llx\n",
739 (unsigned long long) __le64_to_cpu(log
->first_spare_lba
));
741 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
743 struct imsm_dev
*dev
= __get_imsm_dev(mpb
, i
);
745 super
->current_vol
= i
;
746 getinfo_super_imsm(st
, &info
);
747 fname_from_uuid(st
, &info
, nbuf
, ':');
748 print_imsm_dev(dev
, nbuf
+ 5, super
->disks
->index
);
750 for (i
= 0; i
< mpb
->num_disks
; i
++) {
751 if (i
== super
->disks
->index
)
753 print_imsm_disk(mpb
, i
, reserved
);
757 static void brief_examine_super_imsm(struct supertype
*st
, int verbose
)
759 /* We just write a generic IMSM ARRAY entry */
763 struct intel_super
*super
= st
->sb
;
766 if (!super
->anchor
->num_raid_devs
)
769 getinfo_super_imsm(st
, &info
);
770 fname_from_uuid(st
, &info
, nbuf
, ':');
771 printf("ARRAY metadata=imsm auto=md UUID=%s\n", nbuf
+ 5);
772 for (i
= 0; i
< super
->anchor
->num_raid_devs
; i
++) {
773 struct imsm_dev
*dev
= get_imsm_dev(super
, i
);
775 super
->current_vol
= i
;
776 getinfo_super_imsm(st
, &info
);
777 fname_from_uuid(st
, &info
, nbuf1
, ':');
778 printf("ARRAY /dev/md/%.16s container=%s\n"
779 " member=%d auto=mdp UUID=%s\n",
780 dev
->volume
, nbuf
+ 5, i
, nbuf1
+ 5);
784 static void export_examine_super_imsm(struct supertype
*st
)
786 struct intel_super
*super
= st
->sb
;
787 struct imsm_super
*mpb
= super
->anchor
;
791 getinfo_super_imsm(st
, &info
);
792 fname_from_uuid(st
, &info
, nbuf
, ':');
793 printf("MD_METADATA=imsm\n");
794 printf("MD_LEVEL=container\n");
795 printf("MD_UUID=%s\n", nbuf
+5);
796 printf("MD_DEVICES=%u\n", mpb
->num_disks
);
799 static void detail_super_imsm(struct supertype
*st
, char *homehost
)
804 getinfo_super_imsm(st
, &info
);
805 fname_from_uuid(st
, &info
, nbuf
, ':');
806 printf("\n UUID : %s\n", nbuf
+ 5);
809 static void brief_detail_super_imsm(struct supertype
*st
)
813 getinfo_super_imsm(st
, &info
);
814 fname_from_uuid(st
, &info
, nbuf
, ':');
815 printf(" UUID=%s", nbuf
+ 5);
818 static int imsm_read_serial(int fd
, char *devname
, __u8
*serial
);
819 static void fd2devname(int fd
, char *name
);
821 static int imsm_enumerate_ports(const char *hba_path
, int port_count
, int host_base
, int verbose
)
823 /* dump an unsorted list of devices attached to ahci, as well as
824 * non-connected ports
826 int hba_len
= strlen(hba_path
) + 1;
831 unsigned long port_mask
= (1 << port_count
) - 1;
833 if (port_count
> sizeof(port_mask
) * 8) {
835 fprintf(stderr
, Name
": port_count %d out of range\n", port_count
);
839 /* scroll through /sys/dev/block looking for devices attached to
842 dir
= opendir("/sys/dev/block");
843 for (ent
= dir
? readdir(dir
) : NULL
; ent
; ent
= readdir(dir
)) {
854 if (sscanf(ent
->d_name
, "%d:%d", &major
, &minor
) != 2)
856 path
= devt_to_devpath(makedev(major
, minor
));
859 if (!path_attached_to_hba(path
, hba_path
)) {
865 /* retrieve the scsi device type */
866 if (asprintf(&device
, "/sys/dev/block/%d:%d/device/xxxxxxx", major
, minor
) < 0) {
868 fprintf(stderr
, Name
": failed to allocate 'device'\n");
872 sprintf(device
, "/sys/dev/block/%d:%d/device/type", major
, minor
);
873 if (load_sys(device
, buf
) != 0) {
875 fprintf(stderr
, Name
": failed to read device type for %s\n",
881 type
= strtoul(buf
, NULL
, 10);
883 /* if it's not a disk print the vendor and model */
884 if (!(type
== 0 || type
== 7 || type
== 14)) {
887 sprintf(device
, "/sys/dev/block/%d:%d/device/vendor", major
, minor
);
888 if (load_sys(device
, buf
) == 0) {
889 strncpy(vendor
, buf
, sizeof(vendor
));
890 vendor
[sizeof(vendor
) - 1] = '\0';
891 c
= (char *) &vendor
[sizeof(vendor
) - 1];
892 while (isspace(*c
) || *c
== '\0')
896 sprintf(device
, "/sys/dev/block/%d:%d/device/model", major
, minor
);
897 if (load_sys(device
, buf
) == 0) {
898 strncpy(model
, buf
, sizeof(model
));
899 model
[sizeof(model
) - 1] = '\0';
900 c
= (char *) &model
[sizeof(model
) - 1];
901 while (isspace(*c
) || *c
== '\0')
905 if (vendor
[0] && model
[0])
906 sprintf(buf
, "%.64s %.64s", vendor
, model
);
908 switch (type
) { /* numbers from hald/linux/device.c */
909 case 1: sprintf(buf
, "tape"); break;
910 case 2: sprintf(buf
, "printer"); break;
911 case 3: sprintf(buf
, "processor"); break;
913 case 5: sprintf(buf
, "cdrom"); break;
914 case 6: sprintf(buf
, "scanner"); break;
915 case 8: sprintf(buf
, "media_changer"); break;
916 case 9: sprintf(buf
, "comm"); break;
917 case 12: sprintf(buf
, "raid"); break;
918 default: sprintf(buf
, "unknown");
924 /* chop device path to 'host%d' and calculate the port number */
925 c
= strchr(&path
[hba_len
], '/');
927 if (sscanf(&path
[hba_len
], "host%d", &port
) == 1)
931 *c
= '/'; /* repair the full string */
932 fprintf(stderr
, Name
": failed to determine port number for %s\n",
939 /* mark this port as used */
940 port_mask
&= ~(1 << port
);
942 /* print out the device information */
944 printf(" Port%d : - non-disk device (%s) -\n", port
, buf
);
948 fd
= dev_open(ent
->d_name
, O_RDONLY
);
950 printf(" Port%d : - disk info unavailable -\n", port
);
953 printf(" Port%d : %s", port
, buf
);
954 if (imsm_read_serial(fd
, NULL
, (__u8
*) buf
) == 0)
955 printf(" (%s)\n", buf
);
970 for (i
= 0; i
< port_count
; i
++)
971 if (port_mask
& (1 << i
))
972 printf(" Port%d : - no device attached -\n", i
);
978 static int detail_platform_imsm(int verbose
, int enumerate_only
)
980 /* There are two components to imsm platform support, the ahci SATA
981 * controller and the option-rom. To find the SATA controller we
982 * simply look in /sys/bus/pci/drivers/ahci to see if an ahci
983 * controller with the Intel vendor id is present. This approach
984 * allows mdadm to leverage the kernel's ahci detection logic, with the
985 * caveat that if ahci.ko is not loaded mdadm will not be able to
986 * detect platform raid capabilities. The option-rom resides in a
987 * platform "Adapter ROM". We scan for its signature to retrieve the
988 * platform capabilities. If raid support is disabled in the BIOS the
989 * option-rom capability structure will not be available.
991 const struct imsm_orom
*orom
;
992 struct sys_dev
*list
, *hba
;
995 const char *hba_path
;
999 if (enumerate_only
) {
1000 if (check_env("IMSM_NO_PLATFORM") || find_imsm_orom())
1005 list
= find_driver_devices("pci", "ahci");
1006 for (hba
= list
; hba
; hba
= hba
->next
)
1007 if (devpath_to_vendor(hba
->path
) == 0x8086)
1012 fprintf(stderr
, Name
": unable to find active ahci controller\n");
1013 free_sys_dev(&list
);
1016 fprintf(stderr
, Name
": found Intel SATA AHCI Controller\n");
1017 hba_path
= hba
->path
;
1019 free_sys_dev(&list
);
1021 orom
= find_imsm_orom();
1024 fprintf(stderr
, Name
": imsm option-rom not found\n");
1028 printf(" Platform : Intel(R) Matrix Storage Manager\n");
1029 printf(" Version : %d.%d.%d.%d\n", orom
->major_ver
, orom
->minor_ver
,
1030 orom
->hotfix_ver
, orom
->build
);
1031 printf(" RAID Levels :%s%s%s%s%s\n",
1032 imsm_orom_has_raid0(orom
) ? " raid0" : "",
1033 imsm_orom_has_raid1(orom
) ? " raid1" : "",
1034 imsm_orom_has_raid1e(orom
) ? " raid1e" : "",
1035 imsm_orom_has_raid10(orom
) ? " raid10" : "",
1036 imsm_orom_has_raid5(orom
) ? " raid5" : "");
1037 printf(" Chunk Sizes :%s%s%s%s%s%s%s%s%s%s%s%s%s%s%s%s\n",
1038 imsm_orom_has_chunk(orom
, 2) ? " 2k" : "",
1039 imsm_orom_has_chunk(orom
, 4) ? " 4k" : "",
1040 imsm_orom_has_chunk(orom
, 8) ? " 8k" : "",
1041 imsm_orom_has_chunk(orom
, 16) ? " 16k" : "",
1042 imsm_orom_has_chunk(orom
, 32) ? " 32k" : "",
1043 imsm_orom_has_chunk(orom
, 64) ? " 64k" : "",
1044 imsm_orom_has_chunk(orom
, 128) ? " 128k" : "",
1045 imsm_orom_has_chunk(orom
, 256) ? " 256k" : "",
1046 imsm_orom_has_chunk(orom
, 512) ? " 512k" : "",
1047 imsm_orom_has_chunk(orom
, 1024*1) ? " 1M" : "",
1048 imsm_orom_has_chunk(orom
, 1024*2) ? " 2M" : "",
1049 imsm_orom_has_chunk(orom
, 1024*4) ? " 4M" : "",
1050 imsm_orom_has_chunk(orom
, 1024*8) ? " 8M" : "",
1051 imsm_orom_has_chunk(orom
, 1024*16) ? " 16M" : "",
1052 imsm_orom_has_chunk(orom
, 1024*32) ? " 32M" : "",
1053 imsm_orom_has_chunk(orom
, 1024*64) ? " 64M" : "");
1054 printf(" Max Disks : %d\n", orom
->tds
);
1055 printf(" Max Volumes : %d\n", orom
->vpa
);
1056 printf(" I/O Controller : %s\n", hba_path
);
1058 /* find the smallest scsi host number to determine a port number base */
1059 dir
= opendir(hba_path
);
1060 for (ent
= dir
? readdir(dir
) : NULL
; ent
; ent
= readdir(dir
)) {
1063 if (sscanf(ent
->d_name
, "host%d", &host
) != 1)
1065 if (port_count
== 0)
1067 else if (host
< host_base
)
1070 if (host
+ 1 > port_count
+ host_base
)
1071 port_count
= host
+ 1 - host_base
;
1077 if (!port_count
|| imsm_enumerate_ports(hba_path
, port_count
,
1078 host_base
, verbose
) != 0) {
1080 fprintf(stderr
, Name
": failed to enumerate ports\n");
1088 static int match_home_imsm(struct supertype
*st
, char *homehost
)
1090 /* the imsm metadata format does not specify any host
1091 * identification information. We return -1 since we can never
1092 * confirm nor deny whether a given array is "meant" for this
1093 * host. We rely on compare_super and the 'family_num' field to
1094 * exclude member disks that do not belong, and we rely on
1095 * mdadm.conf to specify the arrays that should be assembled.
1096 * Auto-assembly may still pick up "foreign" arrays.
1102 static void uuid_from_super_imsm(struct supertype
*st
, int uuid
[4])
1104 /* The uuid returned here is used for:
1105 * uuid to put into bitmap file (Create, Grow)
1106 * uuid for backup header when saving critical section (Grow)
1107 * comparing uuids when re-adding a device into an array
1108 * In these cases the uuid required is that of the data-array,
1109 * not the device-set.
1110 * uuid to recognise same set when adding a missing device back
1111 * to an array. This is a uuid for the device-set.
1113 * For each of these we can make do with a truncated
1114 * or hashed uuid rather than the original, as long as
1116 * In each case the uuid required is that of the data-array,
1117 * not the device-set.
1119 /* imsm does not track uuid's so we synthesis one using sha1 on
1120 * - The signature (Which is constant for all imsm array, but no matter)
1121 * - the family_num of the container
1122 * - the index number of the volume
1123 * - the 'serial' number of the volume.
1124 * Hopefully these are all constant.
1126 struct intel_super
*super
= st
->sb
;
1129 struct sha1_ctx ctx
;
1130 struct imsm_dev
*dev
= NULL
;
1132 sha1_init_ctx(&ctx
);
1133 sha1_process_bytes(super
->anchor
->sig
, MPB_SIG_LEN
, &ctx
);
1134 sha1_process_bytes(&super
->anchor
->family_num
, sizeof(__u32
), &ctx
);
1135 if (super
->current_vol
>= 0)
1136 dev
= get_imsm_dev(super
, super
->current_vol
);
1138 __u32 vol
= super
->current_vol
;
1139 sha1_process_bytes(&vol
, sizeof(vol
), &ctx
);
1140 sha1_process_bytes(dev
->volume
, MAX_RAID_SERIAL_LEN
, &ctx
);
1142 sha1_finish_ctx(&ctx
, buf
);
1143 memcpy(uuid
, buf
, 4*4);
1148 get_imsm_numerical_version(struct imsm_super
*mpb
, int *m
, int *p
)
1150 __u8
*v
= get_imsm_version(mpb
);
1151 __u8
*end
= mpb
->sig
+ MAX_SIGNATURE_LENGTH
;
1152 char major
[] = { 0, 0, 0 };
1153 char minor
[] = { 0 ,0, 0 };
1154 char patch
[] = { 0, 0, 0 };
1155 char *ver_parse
[] = { major
, minor
, patch
};
1159 while (*v
!= '\0' && v
< end
) {
1160 if (*v
!= '.' && j
< 2)
1161 ver_parse
[i
][j
++] = *v
;
1169 *m
= strtol(minor
, NULL
, 0);
1170 *p
= strtol(patch
, NULL
, 0);
1174 static int imsm_level_to_layout(int level
)
1182 return ALGORITHM_LEFT_ASYMMETRIC
;
1189 static void getinfo_super_imsm_volume(struct supertype
*st
, struct mdinfo
*info
)
1191 struct intel_super
*super
= st
->sb
;
1192 struct imsm_dev
*dev
= get_imsm_dev(super
, super
->current_vol
);
1193 struct imsm_map
*map
= get_imsm_map(dev
, 0);
1196 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
1197 if (dl
->raiddisk
== info
->disk
.raid_disk
)
1199 info
->container_member
= super
->current_vol
;
1200 info
->array
.raid_disks
= map
->num_members
;
1201 info
->array
.level
= get_imsm_raid_level(map
);
1202 info
->array
.layout
= imsm_level_to_layout(info
->array
.level
);
1203 info
->array
.md_minor
= -1;
1204 info
->array
.ctime
= 0;
1205 info
->array
.utime
= 0;
1206 info
->array
.chunk_size
= __le16_to_cpu(map
->blocks_per_strip
) << 9;
1207 info
->array
.state
= !dev
->vol
.dirty
;
1208 info
->custom_array_size
= __le32_to_cpu(dev
->size_high
);
1209 info
->custom_array_size
<<= 32;
1210 info
->custom_array_size
|= __le32_to_cpu(dev
->size_low
);
1212 info
->disk
.major
= 0;
1213 info
->disk
.minor
= 0;
1215 info
->disk
.major
= dl
->major
;
1216 info
->disk
.minor
= dl
->minor
;
1219 info
->data_offset
= __le32_to_cpu(map
->pba_of_lba0
);
1220 info
->component_size
= __le32_to_cpu(map
->blocks_per_member
);
1221 memset(info
->uuid
, 0, sizeof(info
->uuid
));
1223 if (map
->map_state
== IMSM_T_STATE_UNINITIALIZED
|| dev
->vol
.dirty
)
1224 info
->resync_start
= 0;
1225 else if (dev
->vol
.migr_state
)
1226 /* FIXME add curr_migr_unit to resync_start conversion */
1227 info
->resync_start
= 0;
1229 info
->resync_start
= ~0ULL;
1231 strncpy(info
->name
, (char *) dev
->volume
, MAX_RAID_SERIAL_LEN
);
1232 info
->name
[MAX_RAID_SERIAL_LEN
] = 0;
1234 info
->array
.major_version
= -1;
1235 info
->array
.minor_version
= -2;
1236 sprintf(info
->text_version
, "/%s/%d",
1237 devnum2devname(st
->container_dev
),
1238 info
->container_member
);
1239 info
->safe_mode_delay
= 4000; /* 4 secs like the Matrix driver */
1240 uuid_from_super_imsm(st
, info
->uuid
);
1243 /* check the config file to see if we can return a real uuid for this spare */
1244 static void fixup_container_spare_uuid(struct mdinfo
*inf
)
1246 struct mddev_ident_s
*array_list
;
1248 if (inf
->array
.level
!= LEVEL_CONTAINER
||
1249 memcmp(inf
->uuid
, uuid_match_any
, sizeof(int[4])) != 0)
1252 array_list
= conf_get_ident(NULL
);
1254 for (; array_list
; array_list
= array_list
->next
) {
1255 if (array_list
->uuid_set
) {
1256 struct supertype
*_sst
; /* spare supertype */
1257 struct supertype
*_cst
; /* container supertype */
1259 _cst
= array_list
->st
;
1260 _sst
= _cst
->ss
->match_metadata_desc(inf
->text_version
);
1262 memcpy(inf
->uuid
, array_list
->uuid
, sizeof(int[4]));
1270 static void getinfo_super_imsm(struct supertype
*st
, struct mdinfo
*info
)
1272 struct intel_super
*super
= st
->sb
;
1273 struct imsm_disk
*disk
;
1276 if (super
->current_vol
>= 0) {
1277 getinfo_super_imsm_volume(st
, info
);
1281 /* Set raid_disks to zero so that Assemble will always pull in valid
1284 info
->array
.raid_disks
= 0;
1285 info
->array
.level
= LEVEL_CONTAINER
;
1286 info
->array
.layout
= 0;
1287 info
->array
.md_minor
= -1;
1288 info
->array
.ctime
= 0; /* N/A for imsm */
1289 info
->array
.utime
= 0;
1290 info
->array
.chunk_size
= 0;
1292 info
->disk
.major
= 0;
1293 info
->disk
.minor
= 0;
1294 info
->disk
.raid_disk
= -1;
1295 info
->reshape_active
= 0;
1296 info
->array
.major_version
= -1;
1297 info
->array
.minor_version
= -2;
1298 strcpy(info
->text_version
, "imsm");
1299 info
->safe_mode_delay
= 0;
1300 info
->disk
.number
= -1;
1301 info
->disk
.state
= 0;
1305 __u32 reserved
= imsm_reserved_sectors(super
, super
->disks
);
1307 disk
= &super
->disks
->disk
;
1308 info
->data_offset
= __le32_to_cpu(disk
->total_blocks
) - reserved
;
1309 info
->component_size
= reserved
;
1311 info
->disk
.state
= s
& CONFIGURED_DISK
? (1 << MD_DISK_ACTIVE
) : 0;
1312 /* we don't change info->disk.raid_disk here because
1313 * this state will be finalized in mdmon after we have
1314 * found the 'most fresh' version of the metadata
1316 info
->disk
.state
|= s
& FAILED_DISK
? (1 << MD_DISK_FAULTY
) : 0;
1317 info
->disk
.state
|= s
& SPARE_DISK
? 0 : (1 << MD_DISK_SYNC
);
1320 /* only call uuid_from_super_imsm when this disk is part of a populated container,
1321 * ->compare_super may have updated the 'num_raid_devs' field for spares
1323 if (info
->disk
.state
& (1 << MD_DISK_SYNC
) || super
->anchor
->num_raid_devs
)
1324 uuid_from_super_imsm(st
, info
->uuid
);
1326 memcpy(info
->uuid
, uuid_match_any
, sizeof(int[4]));
1327 fixup_container_spare_uuid(info
);
1331 static int update_super_imsm(struct supertype
*st
, struct mdinfo
*info
,
1332 char *update
, char *devname
, int verbose
,
1333 int uuid_set
, char *homehost
)
1337 /* For 'assemble' and 'force' we need to return non-zero if any
1338 * change was made. For others, the return value is ignored.
1339 * Update options are:
1340 * force-one : This device looks a bit old but needs to be included,
1341 * update age info appropriately.
1342 * assemble: clear any 'faulty' flag to allow this device to
1344 * force-array: Array is degraded but being forced, mark it clean
1345 * if that will be needed to assemble it.
1347 * newdev: not used ????
1348 * grow: Array has gained a new device - this is currently for
1350 * resync: mark as dirty so a resync will happen.
1351 * name: update the name - preserving the homehost
1353 * Following are not relevant for this imsm:
1354 * sparc2.2 : update from old dodgey metadata
1355 * super-minor: change the preferred_minor number
1356 * summaries: update redundant counters.
1357 * uuid: Change the uuid of the array to match watch is given
1358 * homehost: update the recorded homehost
1359 * _reshape_progress: record new reshape_progress position.
1362 //struct intel_super *super = st->sb;
1363 //struct imsm_super *mpb = super->mpb;
1365 if (strcmp(update
, "grow") == 0) {
1367 if (strcmp(update
, "resync") == 0) {
1368 /* dev->vol.dirty = 1; */
1371 /* IMSM has no concept of UUID or homehost */
1376 static size_t disks_to_mpb_size(int disks
)
1380 size
= sizeof(struct imsm_super
);
1381 size
+= (disks
- 1) * sizeof(struct imsm_disk
);
1382 size
+= 2 * sizeof(struct imsm_dev
);
1383 /* up to 2 maps per raid device (-2 for imsm_maps in imsm_dev */
1384 size
+= (4 - 2) * sizeof(struct imsm_map
);
1385 /* 4 possible disk_ord_tbl's */
1386 size
+= 4 * (disks
- 1) * sizeof(__u32
);
1391 static __u64
avail_size_imsm(struct supertype
*st
, __u64 devsize
)
1393 if (devsize
< (MPB_SECTOR_CNT
+ IMSM_RESERVED_SECTORS
))
1396 return devsize
- (MPB_SECTOR_CNT
+ IMSM_RESERVED_SECTORS
);
1399 static void free_devlist(struct intel_super
*super
)
1401 struct intel_dev
*dv
;
1403 while (super
->devlist
) {
1404 dv
= super
->devlist
->next
;
1405 free(super
->devlist
->dev
);
1406 free(super
->devlist
);
1407 super
->devlist
= dv
;
1411 static void imsm_copy_dev(struct imsm_dev
*dest
, struct imsm_dev
*src
)
1413 memcpy(dest
, src
, sizeof_imsm_dev(src
, 0));
1416 static int compare_super_imsm(struct supertype
*st
, struct supertype
*tst
)
1420 * 0 same, or first was empty, and second was copied
1421 * 1 second had wrong number
1423 * 3 wrong other info
1425 struct intel_super
*first
= st
->sb
;
1426 struct intel_super
*sec
= tst
->sb
;
1434 if (memcmp(first
->anchor
->sig
, sec
->anchor
->sig
, MAX_SIGNATURE_LENGTH
) != 0)
1437 /* if an anchor does not have num_raid_devs set then it is a free
1440 if (first
->anchor
->num_raid_devs
> 0 &&
1441 sec
->anchor
->num_raid_devs
> 0) {
1442 if (first
->anchor
->family_num
!= sec
->anchor
->family_num
)
1446 /* if 'first' is a spare promote it to a populated mpb with sec's
1449 if (first
->anchor
->num_raid_devs
== 0 &&
1450 sec
->anchor
->num_raid_devs
> 0) {
1452 struct intel_dev
*dv
;
1453 struct imsm_dev
*dev
;
1455 /* we need to copy raid device info from sec if an allocation
1456 * fails here we don't associate the spare
1458 for (i
= 0; i
< sec
->anchor
->num_raid_devs
; i
++) {
1459 dv
= malloc(sizeof(*dv
));
1462 dev
= malloc(sizeof_imsm_dev(get_imsm_dev(sec
, i
), 1));
1469 dv
->next
= first
->devlist
;
1470 first
->devlist
= dv
;
1472 if (i
<= sec
->anchor
->num_raid_devs
) {
1473 /* allocation failure */
1474 free_devlist(first
);
1475 fprintf(stderr
, "imsm: failed to associate spare\n");
1478 for (i
= 0; i
< sec
->anchor
->num_raid_devs
; i
++)
1479 imsm_copy_dev(get_imsm_dev(first
, i
), get_imsm_dev(sec
, i
));
1481 first
->anchor
->num_raid_devs
= sec
->anchor
->num_raid_devs
;
1482 first
->anchor
->family_num
= sec
->anchor
->family_num
;
1488 static void fd2devname(int fd
, char *name
)
1497 if (fstat(fd
, &st
) != 0)
1499 sprintf(path
, "/sys/dev/block/%d:%d",
1500 major(st
.st_rdev
), minor(st
.st_rdev
));
1502 rv
= readlink(path
, dname
, sizeof(dname
));
1507 nm
= strrchr(dname
, '/');
1509 snprintf(name
, MAX_RAID_SERIAL_LEN
, "/dev/%s", nm
);
1513 extern int scsi_get_serial(int fd
, void *buf
, size_t buf_len
);
1515 static int imsm_read_serial(int fd
, char *devname
,
1516 __u8 serial
[MAX_RAID_SERIAL_LEN
])
1518 unsigned char scsi_serial
[255];
1527 memset(scsi_serial
, 0, sizeof(scsi_serial
));
1529 rv
= scsi_get_serial(fd
, scsi_serial
, sizeof(scsi_serial
));
1531 if (rv
&& check_env("IMSM_DEVNAME_AS_SERIAL")) {
1532 memset(serial
, 0, MAX_RAID_SERIAL_LEN
);
1533 fd2devname(fd
, (char *) serial
);
1540 Name
": Failed to retrieve serial for %s\n",
1545 rsp_len
= scsi_serial
[3];
1549 Name
": Failed to retrieve serial for %s\n",
1553 rsp_buf
= (char *) &scsi_serial
[4];
1555 /* trim all whitespace and non-printable characters and convert
1558 for (i
= 0, dest
= rsp_buf
; i
< rsp_len
; i
++) {
1561 /* ':' is reserved for use in placeholder serial
1562 * numbers for missing disks
1570 len
= dest
- rsp_buf
;
1573 /* truncate leading characters */
1574 if (len
> MAX_RAID_SERIAL_LEN
) {
1575 dest
+= len
- MAX_RAID_SERIAL_LEN
;
1576 len
= MAX_RAID_SERIAL_LEN
;
1579 memset(serial
, 0, MAX_RAID_SERIAL_LEN
);
1580 memcpy(serial
, dest
, len
);
1585 static int serialcmp(__u8
*s1
, __u8
*s2
)
1587 return strncmp((char *) s1
, (char *) s2
, MAX_RAID_SERIAL_LEN
);
1590 static void serialcpy(__u8
*dest
, __u8
*src
)
1592 strncpy((char *) dest
, (char *) src
, MAX_RAID_SERIAL_LEN
);
1595 static struct dl
*serial_to_dl(__u8
*serial
, struct intel_super
*super
)
1599 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
1600 if (serialcmp(dl
->serial
, serial
) == 0)
1607 load_imsm_disk(int fd
, struct intel_super
*super
, char *devname
, int keep_fd
)
1614 __u8 serial
[MAX_RAID_SERIAL_LEN
];
1616 rv
= imsm_read_serial(fd
, devname
, serial
);
1621 /* check if this is a disk we have seen before. it may be a spare in
1622 * super->disks while the current anchor believes it is a raid member,
1623 * check if we need to update dl->index
1625 dl
= serial_to_dl(serial
, super
);
1627 dl
= malloc(sizeof(*dl
));
1634 Name
": failed to allocate disk buffer for %s\n",
1641 dl
->major
= major(stb
.st_rdev
);
1642 dl
->minor
= minor(stb
.st_rdev
);
1643 dl
->next
= super
->disks
;
1644 dl
->fd
= keep_fd
? fd
: -1;
1645 dl
->devname
= devname
? strdup(devname
) : NULL
;
1646 serialcpy(dl
->serial
, serial
);
1649 } else if (keep_fd
) {
1654 /* look up this disk's index in the current anchor */
1655 for (i
= 0; i
< super
->anchor
->num_disks
; i
++) {
1656 struct imsm_disk
*disk_iter
;
1658 disk_iter
= __get_imsm_disk(super
->anchor
, i
);
1660 if (serialcmp(disk_iter
->serial
, dl
->serial
) == 0) {
1661 dl
->disk
= *disk_iter
;
1662 /* only set index on disks that are a member of a
1663 * populated contianer, i.e. one with raid_devs
1665 if (dl
->disk
.status
& FAILED_DISK
)
1667 else if (dl
->disk
.status
& SPARE_DISK
)
1676 /* no match, maybe a stale failed drive */
1677 if (i
== super
->anchor
->num_disks
&& dl
->index
>= 0) {
1678 dl
->disk
= *__get_imsm_disk(super
->anchor
, dl
->index
);
1679 if (dl
->disk
.status
& FAILED_DISK
)
1690 /* When migrating map0 contains the 'destination' state while map1
1691 * contains the current state. When not migrating map0 contains the
1692 * current state. This routine assumes that map[0].map_state is set to
1693 * the current array state before being called.
1695 * Migration is indicated by one of the following states
1696 * 1/ Idle (migr_state=0 map0state=normal||unitialized||degraded||failed)
1697 * 2/ Initialize (migr_state=1 migr_type=MIGR_INIT map0state=normal
1698 * map1state=unitialized)
1699 * 3/ Repair (Resync) (migr_state=1 migr_type=MIGR_REPAIR map0state=normal
1701 * 4/ Rebuild (migr_state=1 migr_type=MIGR_REBUILD map0state=normal
1702 * map1state=degraded)
1704 static void migrate(struct imsm_dev
*dev
, __u8 to_state
, int migr_type
)
1706 struct imsm_map
*dest
;
1707 struct imsm_map
*src
= get_imsm_map(dev
, 0);
1709 dev
->vol
.migr_state
= 1;
1710 set_migr_type(dev
, migr_type
);
1711 dev
->vol
.curr_migr_unit
= 0;
1712 dest
= get_imsm_map(dev
, 1);
1714 /* duplicate and then set the target end state in map[0] */
1715 memcpy(dest
, src
, sizeof_imsm_map(src
));
1716 if (migr_type
== MIGR_REBUILD
) {
1720 for (i
= 0; i
< src
->num_members
; i
++) {
1721 ord
= __le32_to_cpu(src
->disk_ord_tbl
[i
]);
1722 set_imsm_ord_tbl_ent(src
, i
, ord_to_idx(ord
));
1726 src
->map_state
= to_state
;
1729 static void end_migration(struct imsm_dev
*dev
, __u8 map_state
)
1731 struct imsm_map
*map
= get_imsm_map(dev
, 0);
1732 struct imsm_map
*prev
= get_imsm_map(dev
, dev
->vol
.migr_state
);
1735 /* merge any IMSM_ORD_REBUILD bits that were not successfully
1736 * completed in the last migration.
1738 * FIXME add support for online capacity expansion and
1739 * raid-level-migration
1741 for (i
= 0; i
< prev
->num_members
; i
++)
1742 map
->disk_ord_tbl
[i
] |= prev
->disk_ord_tbl
[i
];
1744 dev
->vol
.migr_state
= 0;
1745 dev
->vol
.curr_migr_unit
= 0;
1746 map
->map_state
= map_state
;
1750 static int parse_raid_devices(struct intel_super
*super
)
1753 struct imsm_dev
*dev_new
;
1754 size_t len
, len_migr
;
1755 size_t space_needed
= 0;
1756 struct imsm_super
*mpb
= super
->anchor
;
1758 for (i
= 0; i
< super
->anchor
->num_raid_devs
; i
++) {
1759 struct imsm_dev
*dev_iter
= __get_imsm_dev(super
->anchor
, i
);
1760 struct intel_dev
*dv
;
1762 len
= sizeof_imsm_dev(dev_iter
, 0);
1763 len_migr
= sizeof_imsm_dev(dev_iter
, 1);
1765 space_needed
+= len_migr
- len
;
1767 dv
= malloc(sizeof(*dv
));
1770 dev_new
= malloc(len_migr
);
1775 imsm_copy_dev(dev_new
, dev_iter
);
1778 dv
->next
= super
->devlist
;
1779 super
->devlist
= dv
;
1782 /* ensure that super->buf is large enough when all raid devices
1785 if (__le32_to_cpu(mpb
->mpb_size
) + space_needed
> super
->len
) {
1788 len
= ROUND_UP(__le32_to_cpu(mpb
->mpb_size
) + space_needed
, 512);
1789 if (posix_memalign(&buf
, 512, len
) != 0)
1792 memcpy(buf
, super
->buf
, super
->len
);
1793 memset(buf
+ super
->len
, 0, len
- super
->len
);
1802 /* retrieve a pointer to the bbm log which starts after all raid devices */
1803 struct bbm_log
*__get_imsm_bbm_log(struct imsm_super
*mpb
)
1807 if (__le32_to_cpu(mpb
->bbm_log_size
)) {
1809 ptr
+= mpb
->mpb_size
- __le32_to_cpu(mpb
->bbm_log_size
);
1815 static void __free_imsm(struct intel_super
*super
, int free_disks
);
1817 /* load_imsm_mpb - read matrix metadata
1818 * allocates super->mpb to be freed by free_super
1820 static int load_imsm_mpb(int fd
, struct intel_super
*super
, char *devname
)
1822 unsigned long long dsize
;
1823 unsigned long long sectors
;
1825 struct imsm_super
*anchor
;
1829 get_dev_size(fd
, NULL
, &dsize
);
1831 if (lseek64(fd
, dsize
- (512 * 2), SEEK_SET
) < 0) {
1834 Name
": Cannot seek to anchor block on %s: %s\n",
1835 devname
, strerror(errno
));
1839 if (posix_memalign((void**)&anchor
, 512, 512) != 0) {
1842 Name
": Failed to allocate imsm anchor buffer"
1843 " on %s\n", devname
);
1846 if (read(fd
, anchor
, 512) != 512) {
1849 Name
": Cannot read anchor block on %s: %s\n",
1850 devname
, strerror(errno
));
1855 if (strncmp((char *) anchor
->sig
, MPB_SIGNATURE
, MPB_SIG_LEN
) != 0) {
1858 Name
": no IMSM anchor on %s\n", devname
);
1863 __free_imsm(super
, 0);
1864 super
->len
= ROUND_UP(anchor
->mpb_size
, 512);
1865 if (posix_memalign(&super
->buf
, 512, super
->len
) != 0) {
1868 Name
": unable to allocate %zu byte mpb buffer\n",
1873 memcpy(super
->buf
, anchor
, 512);
1875 sectors
= mpb_sectors(anchor
) - 1;
1878 check_sum
= __gen_imsm_checksum(super
->anchor
);
1879 if (check_sum
!= __le32_to_cpu(super
->anchor
->check_sum
)) {
1882 Name
": IMSM checksum %x != %x on %s\n",
1884 __le32_to_cpu(super
->anchor
->check_sum
),
1889 rc
= load_imsm_disk(fd
, super
, devname
, 0);
1891 rc
= parse_raid_devices(super
);
1895 /* read the extended mpb */
1896 if (lseek64(fd
, dsize
- (512 * (2 + sectors
)), SEEK_SET
) < 0) {
1899 Name
": Cannot seek to extended mpb on %s: %s\n",
1900 devname
, strerror(errno
));
1904 if (read(fd
, super
->buf
+ 512, super
->len
- 512) != super
->len
- 512) {
1907 Name
": Cannot read extended mpb on %s: %s\n",
1908 devname
, strerror(errno
));
1912 check_sum
= __gen_imsm_checksum(super
->anchor
);
1913 if (check_sum
!= __le32_to_cpu(super
->anchor
->check_sum
)) {
1916 Name
": IMSM checksum %x != %x on %s\n",
1917 check_sum
, __le32_to_cpu(super
->anchor
->check_sum
),
1922 /* FIXME the BBM log is disk specific so we cannot use this global
1923 * buffer for all disks. Ok for now since we only look at the global
1924 * bbm_log_size parameter to gate assembly
1926 super
->bbm_log
= __get_imsm_bbm_log(super
->anchor
);
1928 rc
= load_imsm_disk(fd
, super
, devname
, 0);
1930 rc
= parse_raid_devices(super
);
1935 static void __free_imsm_disk(struct dl
*d
)
1946 static void free_imsm_disks(struct intel_super
*super
)
1950 while (super
->disks
) {
1952 super
->disks
= d
->next
;
1953 __free_imsm_disk(d
);
1955 while (super
->missing
) {
1957 super
->missing
= d
->next
;
1958 __free_imsm_disk(d
);
1963 /* free all the pieces hanging off of a super pointer */
1964 static void __free_imsm(struct intel_super
*super
, int free_disks
)
1971 free_imsm_disks(super
);
1972 free_devlist(super
);
1974 free((void *) super
->hba
);
1979 static void free_imsm(struct intel_super
*super
)
1981 __free_imsm(super
, 1);
1985 static void free_super_imsm(struct supertype
*st
)
1987 struct intel_super
*super
= st
->sb
;
1996 static struct intel_super
*alloc_super(int creating_imsm
)
1998 struct intel_super
*super
= malloc(sizeof(*super
));
2001 memset(super
, 0, sizeof(*super
));
2002 super
->creating_imsm
= creating_imsm
;
2003 super
->current_vol
= -1;
2004 super
->create_offset
= ~((__u32
) 0);
2005 if (!check_env("IMSM_NO_PLATFORM"))
2006 super
->orom
= find_imsm_orom();
2007 if (super
->orom
&& !check_env("IMSM_TEST_OROM")) {
2008 struct sys_dev
*list
, *ent
;
2010 /* find the first intel ahci controller */
2011 list
= find_driver_devices("pci", "ahci");
2012 for (ent
= list
; ent
; ent
= ent
->next
)
2013 if (devpath_to_vendor(ent
->path
) == 0x8086)
2016 super
->hba
= ent
->path
;
2019 free_sys_dev(&list
);
2027 /* find_missing - helper routine for load_super_imsm_all that identifies
2028 * disks that have disappeared from the system. This routine relies on
2029 * the mpb being uptodate, which it is at load time.
2031 static int find_missing(struct intel_super
*super
)
2034 struct imsm_super
*mpb
= super
->anchor
;
2036 struct imsm_disk
*disk
;
2038 for (i
= 0; i
< mpb
->num_disks
; i
++) {
2039 disk
= __get_imsm_disk(mpb
, i
);
2040 dl
= serial_to_dl(disk
->serial
, super
);
2044 dl
= malloc(sizeof(*dl
));
2050 dl
->devname
= strdup("missing");
2052 serialcpy(dl
->serial
, disk
->serial
);
2055 dl
->next
= super
->missing
;
2056 super
->missing
= dl
;
2062 static int load_super_imsm_all(struct supertype
*st
, int fd
, void **sbp
,
2063 char *devname
, int keep_fd
)
2066 struct intel_super
*super
;
2067 struct mdinfo
*sd
, *best
= NULL
;
2073 int devnum
= fd2devnum(fd
);
2075 enum sysfs_read_flags flags
;
2077 flags
= GET_LEVEL
|GET_VERSION
|GET_DEVS
|GET_STATE
;
2078 if (mdmon_running(devnum
))
2079 flags
|= SKIP_GONE_DEVS
;
2081 /* check if 'fd' an opened container */
2082 sra
= sysfs_read(fd
, 0, flags
);
2086 if (sra
->array
.major_version
!= -1 ||
2087 sra
->array
.minor_version
!= -2 ||
2088 strcmp(sra
->text_version
, "imsm") != 0)
2091 super
= alloc_super(0);
2095 /* find the most up to date disk in this array, skipping spares */
2096 for (sd
= sra
->devs
; sd
; sd
= sd
->next
) {
2097 sprintf(nm
, "%d:%d", sd
->disk
.major
, sd
->disk
.minor
);
2098 dfd
= dev_open(nm
, keep_fd
? O_RDWR
: O_RDONLY
);
2103 rv
= load_imsm_mpb(dfd
, super
, NULL
);
2105 /* retry the load if we might have raced against mdmon */
2106 if (rv
== 3 && mdmon_running(devnum
))
2107 for (retry
= 0; retry
< 3; retry
++) {
2109 rv
= load_imsm_mpb(dfd
, super
, NULL
);
2116 if (super
->anchor
->num_raid_devs
== 0)
2119 gen
= __le32_to_cpu(super
->anchor
->generation_num
);
2120 if (!best
|| gen
> bestgen
) {
2135 /* load the most up to date anchor */
2136 sprintf(nm
, "%d:%d", best
->disk
.major
, best
->disk
.minor
);
2137 dfd
= dev_open(nm
, O_RDONLY
);
2142 rv
= load_imsm_mpb(dfd
, super
, NULL
);
2149 /* re-parse the disk list with the current anchor */
2150 for (sd
= sra
->devs
; sd
; sd
= sd
->next
) {
2151 sprintf(nm
, "%d:%d", sd
->disk
.major
, sd
->disk
.minor
);
2152 dfd
= dev_open(nm
, keep_fd
? O_RDWR
: O_RDONLY
);
2157 load_imsm_disk(dfd
, super
, NULL
, keep_fd
);
2163 if (find_missing(super
) != 0) {
2168 if (st
->subarray
[0]) {
2169 if (atoi(st
->subarray
) <= super
->anchor
->num_raid_devs
)
2170 super
->current_vol
= atoi(st
->subarray
);
2176 st
->container_dev
= devnum
;
2177 if (st
->ss
== NULL
) {
2178 st
->ss
= &super_imsm
;
2179 st
->minor_version
= 0;
2180 st
->max_devs
= IMSM_MAX_DEVICES
;
2182 st
->loaded_container
= 1;
2188 static int load_super_imsm(struct supertype
*st
, int fd
, char *devname
)
2190 struct intel_super
*super
;
2194 if (load_super_imsm_all(st
, fd
, &st
->sb
, devname
, 1) == 0)
2197 if (st
->subarray
[0])
2198 return 1; /* FIXME */
2200 super
= alloc_super(0);
2203 Name
": malloc of %zu failed.\n",
2208 rv
= load_imsm_mpb(fd
, super
, devname
);
2213 Name
": Failed to load all information "
2214 "sections on %s\n", devname
);
2220 if (st
->ss
== NULL
) {
2221 st
->ss
= &super_imsm
;
2222 st
->minor_version
= 0;
2223 st
->max_devs
= IMSM_MAX_DEVICES
;
2225 st
->loaded_container
= 0;
2230 static __u16
info_to_blocks_per_strip(mdu_array_info_t
*info
)
2232 if (info
->level
== 1)
2234 return info
->chunk_size
>> 9;
2237 static __u32
info_to_num_data_stripes(mdu_array_info_t
*info
, int num_domains
)
2241 num_stripes
= (info
->size
* 2) / info_to_blocks_per_strip(info
);
2242 num_stripes
/= num_domains
;
2247 static __u32
info_to_blocks_per_member(mdu_array_info_t
*info
)
2249 if (info
->level
== 1)
2250 return info
->size
* 2;
2252 return (info
->size
* 2) & ~(info_to_blocks_per_strip(info
) - 1);
2255 static void imsm_update_version_info(struct intel_super
*super
)
2257 /* update the version and attributes */
2258 struct imsm_super
*mpb
= super
->anchor
;
2260 struct imsm_dev
*dev
;
2261 struct imsm_map
*map
;
2264 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
2265 dev
= get_imsm_dev(super
, i
);
2266 map
= get_imsm_map(dev
, 0);
2267 if (__le32_to_cpu(dev
->size_high
) > 0)
2268 mpb
->attributes
|= MPB_ATTRIB_2TB
;
2270 /* FIXME detect when an array spans a port multiplier */
2272 mpb
->attributes
|= MPB_ATTRIB_PM
;
2275 if (mpb
->num_raid_devs
> 1 ||
2276 mpb
->attributes
!= MPB_ATTRIB_CHECKSUM_VERIFY
) {
2277 version
= MPB_VERSION_ATTRIBS
;
2278 switch (get_imsm_raid_level(map
)) {
2279 case 0: mpb
->attributes
|= MPB_ATTRIB_RAID0
; break;
2280 case 1: mpb
->attributes
|= MPB_ATTRIB_RAID1
; break;
2281 case 10: mpb
->attributes
|= MPB_ATTRIB_RAID10
; break;
2282 case 5: mpb
->attributes
|= MPB_ATTRIB_RAID5
; break;
2285 if (map
->num_members
>= 5)
2286 version
= MPB_VERSION_5OR6_DISK_ARRAY
;
2287 else if (dev
->status
== DEV_CLONE_N_GO
)
2288 version
= MPB_VERSION_CNG
;
2289 else if (get_imsm_raid_level(map
) == 5)
2290 version
= MPB_VERSION_RAID5
;
2291 else if (map
->num_members
>= 3)
2292 version
= MPB_VERSION_3OR4_DISK_ARRAY
;
2293 else if (get_imsm_raid_level(map
) == 1)
2294 version
= MPB_VERSION_RAID1
;
2296 version
= MPB_VERSION_RAID0
;
2298 strcpy(((char *) mpb
->sig
) + strlen(MPB_SIGNATURE
), version
);
2302 static int init_super_imsm_volume(struct supertype
*st
, mdu_array_info_t
*info
,
2303 unsigned long long size
, char *name
,
2304 char *homehost
, int *uuid
)
2306 /* We are creating a volume inside a pre-existing container.
2307 * so st->sb is already set.
2309 struct intel_super
*super
= st
->sb
;
2310 struct imsm_super
*mpb
= super
->anchor
;
2311 struct intel_dev
*dv
;
2312 struct imsm_dev
*dev
;
2313 struct imsm_vol
*vol
;
2314 struct imsm_map
*map
;
2315 int idx
= mpb
->num_raid_devs
;
2317 unsigned long long array_blocks
;
2318 size_t size_old
, size_new
;
2319 __u32 num_data_stripes
;
2321 if (super
->orom
&& mpb
->num_raid_devs
>= super
->orom
->vpa
) {
2322 fprintf(stderr
, Name
": This imsm-container already has the "
2323 "maximum of %d volumes\n", super
->orom
->vpa
);
2327 /* ensure the mpb is large enough for the new data */
2328 size_old
= __le32_to_cpu(mpb
->mpb_size
);
2329 size_new
= disks_to_mpb_size(info
->nr_disks
);
2330 if (size_new
> size_old
) {
2332 size_t size_round
= ROUND_UP(size_new
, 512);
2334 if (posix_memalign(&mpb_new
, 512, size_round
) != 0) {
2335 fprintf(stderr
, Name
": could not allocate new mpb\n");
2338 memcpy(mpb_new
, mpb
, size_old
);
2341 super
->anchor
= mpb_new
;
2342 mpb
->mpb_size
= __cpu_to_le32(size_new
);
2343 memset(mpb_new
+ size_old
, 0, size_round
- size_old
);
2345 super
->current_vol
= idx
;
2346 /* when creating the first raid device in this container set num_disks
2347 * to zero, i.e. delete this spare and add raid member devices in
2348 * add_to_super_imsm_volume()
2350 if (super
->current_vol
== 0)
2353 for (i
= 0; i
< super
->current_vol
; i
++) {
2354 dev
= get_imsm_dev(super
, i
);
2355 if (strncmp((char *) dev
->volume
, name
,
2356 MAX_RAID_SERIAL_LEN
) == 0) {
2357 fprintf(stderr
, Name
": '%s' is already defined for this container\n",
2363 sprintf(st
->subarray
, "%d", idx
);
2364 dv
= malloc(sizeof(*dv
));
2366 fprintf(stderr
, Name
": failed to allocate device list entry\n");
2369 dev
= malloc(sizeof(*dev
) + sizeof(__u32
) * (info
->raid_disks
- 1));
2372 fprintf(stderr
, Name
": could not allocate raid device\n");
2375 strncpy((char *) dev
->volume
, name
, MAX_RAID_SERIAL_LEN
);
2376 if (info
->level
== 1)
2377 array_blocks
= info_to_blocks_per_member(info
);
2379 array_blocks
= calc_array_size(info
->level
, info
->raid_disks
,
2380 info
->layout
, info
->chunk_size
,
2382 /* round array size down to closest MB */
2383 array_blocks
= (array_blocks
>> SECT_PER_MB_SHIFT
) << SECT_PER_MB_SHIFT
;
2385 dev
->size_low
= __cpu_to_le32((__u32
) array_blocks
);
2386 dev
->size_high
= __cpu_to_le32((__u32
) (array_blocks
>> 32));
2387 dev
->status
= __cpu_to_le32(0);
2388 dev
->reserved_blocks
= __cpu_to_le32(0);
2390 vol
->migr_state
= 0;
2391 set_migr_type(dev
, MIGR_INIT
);
2393 vol
->curr_migr_unit
= 0;
2394 map
= get_imsm_map(dev
, 0);
2395 map
->pba_of_lba0
= __cpu_to_le32(super
->create_offset
);
2396 map
->blocks_per_member
= __cpu_to_le32(info_to_blocks_per_member(info
));
2397 map
->blocks_per_strip
= __cpu_to_le16(info_to_blocks_per_strip(info
));
2398 map
->failed_disk_num
= ~0;
2399 map
->map_state
= info
->level
? IMSM_T_STATE_UNINITIALIZED
:
2400 IMSM_T_STATE_NORMAL
;
2403 if (info
->level
== 1 && info
->raid_disks
> 2) {
2404 fprintf(stderr
, Name
": imsm does not support more than 2 disks"
2405 "in a raid1 volume\n");
2408 if (info
->level
== 10) {
2409 map
->raid_level
= 1;
2410 map
->num_domains
= info
->raid_disks
/ 2;
2412 map
->raid_level
= info
->level
;
2413 map
->num_domains
= 1;
2415 num_data_stripes
= info_to_num_data_stripes(info
, map
->num_domains
);
2416 map
->num_data_stripes
= __cpu_to_le32(num_data_stripes
);
2418 map
->num_members
= info
->raid_disks
;
2419 for (i
= 0; i
< map
->num_members
; i
++) {
2420 /* initialized in add_to_super */
2421 set_imsm_ord_tbl_ent(map
, i
, 0);
2423 mpb
->num_raid_devs
++;
2426 dv
->index
= super
->current_vol
;
2427 dv
->next
= super
->devlist
;
2428 super
->devlist
= dv
;
2430 imsm_update_version_info(super
);
2435 static int init_super_imsm(struct supertype
*st
, mdu_array_info_t
*info
,
2436 unsigned long long size
, char *name
,
2437 char *homehost
, int *uuid
)
2439 /* This is primarily called by Create when creating a new array.
2440 * We will then get add_to_super called for each component, and then
2441 * write_init_super called to write it out to each device.
2442 * For IMSM, Create can create on fresh devices or on a pre-existing
2444 * To create on a pre-existing array a different method will be called.
2445 * This one is just for fresh drives.
2447 struct intel_super
*super
;
2448 struct imsm_super
*mpb
;
2457 return init_super_imsm_volume(st
, info
, size
, name
, homehost
,
2460 super
= alloc_super(1);
2463 mpb_size
= disks_to_mpb_size(info
->nr_disks
);
2464 if (posix_memalign(&super
->buf
, 512, mpb_size
) != 0) {
2469 memset(mpb
, 0, mpb_size
);
2471 mpb
->attributes
= MPB_ATTRIB_CHECKSUM_VERIFY
;
2473 version
= (char *) mpb
->sig
;
2474 strcpy(version
, MPB_SIGNATURE
);
2475 version
+= strlen(MPB_SIGNATURE
);
2476 strcpy(version
, MPB_VERSION_RAID0
);
2477 mpb
->mpb_size
= mpb_size
;
2484 static int add_to_super_imsm_volume(struct supertype
*st
, mdu_disk_info_t
*dk
,
2485 int fd
, char *devname
)
2487 struct intel_super
*super
= st
->sb
;
2488 struct imsm_super
*mpb
= super
->anchor
;
2490 struct imsm_dev
*dev
;
2491 struct imsm_map
*map
;
2493 dev
= get_imsm_dev(super
, super
->current_vol
);
2494 map
= get_imsm_map(dev
, 0);
2496 if (! (dk
->state
& (1<<MD_DISK_SYNC
))) {
2497 fprintf(stderr
, Name
": %s: Cannot add spare devices to IMSM volume\n",
2503 /* we're doing autolayout so grab the pre-marked (in
2504 * validate_geometry) raid_disk
2506 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
2507 if (dl
->raiddisk
== dk
->raid_disk
)
2510 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
2511 if (dl
->major
== dk
->major
&&
2512 dl
->minor
== dk
->minor
)
2517 fprintf(stderr
, Name
": %s is not a member of the same container\n", devname
);
2521 /* add a pristine spare to the metadata */
2522 if (dl
->index
< 0) {
2523 dl
->index
= super
->anchor
->num_disks
;
2524 super
->anchor
->num_disks
++;
2526 set_imsm_ord_tbl_ent(map
, dk
->number
, dl
->index
);
2527 dl
->disk
.status
= CONFIGURED_DISK
| USABLE_DISK
;
2529 /* if we are creating the first raid device update the family number */
2530 if (super
->current_vol
== 0) {
2532 struct imsm_dev
*_dev
= __get_imsm_dev(mpb
, 0);
2533 struct imsm_disk
*_disk
= __get_imsm_disk(mpb
, dl
->index
);
2537 sum
= __gen_imsm_checksum(mpb
);
2538 mpb
->family_num
= __cpu_to_le32(sum
);
2544 static int add_to_super_imsm(struct supertype
*st
, mdu_disk_info_t
*dk
,
2545 int fd
, char *devname
)
2547 struct intel_super
*super
= st
->sb
;
2549 unsigned long long size
;
2554 /* if we are on an RAID enabled platform check that the disk is
2555 * attached to the raid controller
2557 if (super
->hba
&& !disk_attached_to_hba(fd
, super
->hba
)) {
2559 Name
": %s is not attached to the raid controller: %s\n",
2560 devname
? : "disk", super
->hba
);
2564 if (super
->current_vol
>= 0)
2565 return add_to_super_imsm_volume(st
, dk
, fd
, devname
);
2568 dd
= malloc(sizeof(*dd
));
2571 Name
": malloc failed %s:%d.\n", __func__
, __LINE__
);
2574 memset(dd
, 0, sizeof(*dd
));
2575 dd
->major
= major(stb
.st_rdev
);
2576 dd
->minor
= minor(stb
.st_rdev
);
2578 dd
->devname
= devname
? strdup(devname
) : NULL
;
2581 rv
= imsm_read_serial(fd
, devname
, dd
->serial
);
2584 Name
": failed to retrieve scsi serial, aborting\n");
2589 get_dev_size(fd
, NULL
, &size
);
2591 serialcpy(dd
->disk
.serial
, dd
->serial
);
2592 dd
->disk
.total_blocks
= __cpu_to_le32(size
);
2593 dd
->disk
.status
= USABLE_DISK
| SPARE_DISK
;
2594 if (sysfs_disk_to_scsi_id(fd
, &id
) == 0)
2595 dd
->disk
.scsi_id
= __cpu_to_le32(id
);
2597 dd
->disk
.scsi_id
= __cpu_to_le32(0);
2599 if (st
->update_tail
) {
2600 dd
->next
= super
->add
;
2603 dd
->next
= super
->disks
;
2610 static int store_imsm_mpb(int fd
, struct intel_super
*super
);
2612 /* spare records have their own family number and do not have any defined raid
2615 static int write_super_imsm_spares(struct intel_super
*super
, int doclose
)
2617 struct imsm_super mpb_save
;
2618 struct imsm_super
*mpb
= super
->anchor
;
2623 mpb
->num_raid_devs
= 0;
2625 mpb
->mpb_size
= sizeof(struct imsm_super
);
2626 mpb
->generation_num
= __cpu_to_le32(1UL);
2628 for (d
= super
->disks
; d
; d
= d
->next
) {
2632 mpb
->disk
[0] = d
->disk
;
2633 sum
= __gen_imsm_checksum(mpb
);
2634 mpb
->family_num
= __cpu_to_le32(sum
);
2635 sum
= __gen_imsm_checksum(mpb
);
2636 mpb
->check_sum
= __cpu_to_le32(sum
);
2638 if (store_imsm_mpb(d
->fd
, super
)) {
2639 fprintf(stderr
, "%s: failed for device %d:%d %s\n",
2640 __func__
, d
->major
, d
->minor
, strerror(errno
));
2654 static int write_super_imsm(struct intel_super
*super
, int doclose
)
2656 struct imsm_super
*mpb
= super
->anchor
;
2662 __u32 mpb_size
= sizeof(struct imsm_super
) - sizeof(struct imsm_disk
);
2664 /* 'generation' is incremented everytime the metadata is written */
2665 generation
= __le32_to_cpu(mpb
->generation_num
);
2667 mpb
->generation_num
= __cpu_to_le32(generation
);
2669 mpb_size
+= sizeof(struct imsm_disk
) * mpb
->num_disks
;
2670 for (d
= super
->disks
; d
; d
= d
->next
) {
2674 mpb
->disk
[d
->index
] = d
->disk
;
2676 for (d
= super
->missing
; d
; d
= d
->next
)
2677 mpb
->disk
[d
->index
] = d
->disk
;
2679 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
2680 struct imsm_dev
*dev
= __get_imsm_dev(mpb
, i
);
2682 imsm_copy_dev(dev
, get_imsm_dev(super
, i
));
2683 mpb_size
+= sizeof_imsm_dev(dev
, 0);
2685 mpb_size
+= __le32_to_cpu(mpb
->bbm_log_size
);
2686 mpb
->mpb_size
= __cpu_to_le32(mpb_size
);
2688 /* recalculate checksum */
2689 sum
= __gen_imsm_checksum(mpb
);
2690 mpb
->check_sum
= __cpu_to_le32(sum
);
2692 /* write the mpb for disks that compose raid devices */
2693 for (d
= super
->disks
; d
; d
= d
->next
) {
2696 if (store_imsm_mpb(d
->fd
, super
))
2697 fprintf(stderr
, "%s: failed for device %d:%d %s\n",
2698 __func__
, d
->major
, d
->minor
, strerror(errno
));
2706 return write_super_imsm_spares(super
, doclose
);
2712 static int create_array(struct supertype
*st
)
2715 struct imsm_update_create_array
*u
;
2716 struct intel_super
*super
= st
->sb
;
2717 struct imsm_dev
*dev
= get_imsm_dev(super
, super
->current_vol
);
2718 struct imsm_map
*map
= get_imsm_map(dev
, 0);
2719 struct disk_info
*inf
;
2720 struct imsm_disk
*disk
;
2724 len
= sizeof(*u
) - sizeof(*dev
) + sizeof_imsm_dev(dev
, 0) +
2725 sizeof(*inf
) * map
->num_members
;
2728 fprintf(stderr
, "%s: failed to allocate update buffer\n",
2733 u
->type
= update_create_array
;
2734 u
->dev_idx
= super
->current_vol
;
2735 imsm_copy_dev(&u
->dev
, dev
);
2736 inf
= get_disk_info(u
);
2737 for (i
= 0; i
< map
->num_members
; i
++) {
2738 idx
= get_imsm_disk_idx(dev
, i
);
2739 disk
= get_imsm_disk(super
, idx
);
2740 serialcpy(inf
[i
].serial
, disk
->serial
);
2742 append_metadata_update(st
, u
, len
);
2747 static int _add_disk(struct supertype
*st
)
2749 struct intel_super
*super
= st
->sb
;
2751 struct imsm_update_add_disk
*u
;
2759 fprintf(stderr
, "%s: failed to allocate update buffer\n",
2764 u
->type
= update_add_disk
;
2765 append_metadata_update(st
, u
, len
);
2770 static int write_init_super_imsm(struct supertype
*st
)
2772 if (st
->update_tail
) {
2773 /* queue the recently created array / added disk
2774 * as a metadata update */
2775 struct intel_super
*super
= st
->sb
;
2779 /* determine if we are creating a volume or adding a disk */
2780 if (super
->current_vol
< 0) {
2781 /* in the add disk case we are running in mdmon
2782 * context, so don't close fd's
2784 return _add_disk(st
);
2786 rv
= create_array(st
);
2788 for (d
= super
->disks
; d
; d
= d
->next
) {
2795 return write_super_imsm(st
->sb
, 1);
2799 static int store_zero_imsm(struct supertype
*st
, int fd
)
2801 unsigned long long dsize
;
2804 get_dev_size(fd
, NULL
, &dsize
);
2806 /* first block is stored on second to last sector of the disk */
2807 if (lseek64(fd
, dsize
- (512 * 2), SEEK_SET
) < 0)
2810 if (posix_memalign(&buf
, 512, 512) != 0)
2813 memset(buf
, 0, 512);
2814 if (write(fd
, buf
, 512) != 512)
2819 static int imsm_bbm_log_size(struct imsm_super
*mpb
)
2821 return __le32_to_cpu(mpb
->bbm_log_size
);
2825 static int validate_geometry_imsm_container(struct supertype
*st
, int level
,
2826 int layout
, int raiddisks
, int chunk
,
2827 unsigned long long size
, char *dev
,
2828 unsigned long long *freesize
,
2832 unsigned long long ldsize
;
2833 const struct imsm_orom
*orom
;
2835 if (level
!= LEVEL_CONTAINER
)
2840 if (check_env("IMSM_NO_PLATFORM"))
2843 orom
= find_imsm_orom();
2844 if (orom
&& raiddisks
> orom
->tds
) {
2846 fprintf(stderr
, Name
": %d exceeds maximum number of"
2847 " platform supported disks: %d\n",
2848 raiddisks
, orom
->tds
);
2852 fd
= open(dev
, O_RDONLY
|O_EXCL
, 0);
2855 fprintf(stderr
, Name
": imsm: Cannot open %s: %s\n",
2856 dev
, strerror(errno
));
2859 if (!get_dev_size(fd
, dev
, &ldsize
)) {
2865 *freesize
= avail_size_imsm(st
, ldsize
>> 9);
2870 static unsigned long long find_size(struct extent
*e
, int *idx
, int num_extents
)
2872 const unsigned long long base_start
= e
[*idx
].start
;
2873 unsigned long long end
= base_start
+ e
[*idx
].size
;
2876 if (base_start
== end
)
2880 for (i
= *idx
; i
< num_extents
; i
++) {
2881 /* extend overlapping extents */
2882 if (e
[i
].start
>= base_start
&&
2883 e
[i
].start
<= end
) {
2886 if (e
[i
].start
+ e
[i
].size
> end
)
2887 end
= e
[i
].start
+ e
[i
].size
;
2888 } else if (e
[i
].start
> end
) {
2894 return end
- base_start
;
2897 static unsigned long long merge_extents(struct intel_super
*super
, int sum_extents
)
2899 /* build a composite disk with all known extents and generate a new
2900 * 'maxsize' given the "all disks in an array must share a common start
2901 * offset" constraint
2903 struct extent
*e
= calloc(sum_extents
, sizeof(*e
));
2907 unsigned long long pos
;
2908 unsigned long long start
= 0;
2909 unsigned long long maxsize
;
2910 unsigned long reserve
;
2913 return ~0ULL; /* error */
2915 /* coalesce and sort all extents. also, check to see if we need to
2916 * reserve space between member arrays
2919 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
2922 for (i
= 0; i
< dl
->extent_cnt
; i
++)
2925 qsort(e
, sum_extents
, sizeof(*e
), cmp_extent
);
2930 while (i
< sum_extents
) {
2931 e
[j
].start
= e
[i
].start
;
2932 e
[j
].size
= find_size(e
, &i
, sum_extents
);
2934 if (e
[j
-1].size
== 0)
2943 unsigned long long esize
;
2945 esize
= e
[i
].start
- pos
;
2946 if (esize
>= maxsize
) {
2951 pos
= e
[i
].start
+ e
[i
].size
;
2953 } while (e
[i
-1].size
);
2956 if (start_extent
> 0)
2957 reserve
= IMSM_RESERVED_SECTORS
; /* gap between raid regions */
2961 if (maxsize
< reserve
)
2964 super
->create_offset
= ~((__u32
) 0);
2965 if (start
+ reserve
> super
->create_offset
)
2966 return ~0ULL; /* start overflows create_offset */
2967 super
->create_offset
= start
+ reserve
;
2969 return maxsize
- reserve
;
2972 static int is_raid_level_supported(const struct imsm_orom
*orom
, int level
, int raiddisks
)
2974 if (level
< 0 || level
== 6 || level
== 4)
2977 /* if we have an orom prevent invalid raid levels */
2980 case 0: return imsm_orom_has_raid0(orom
);
2983 return imsm_orom_has_raid1e(orom
);
2984 return imsm_orom_has_raid1(orom
) && raiddisks
== 2;
2985 case 10: return imsm_orom_has_raid10(orom
) && raiddisks
== 4;
2986 case 5: return imsm_orom_has_raid5(orom
) && raiddisks
> 2;
2989 return 1; /* not on an Intel RAID platform so anything goes */
2994 #define pr_vrb(fmt, arg...) (void) (verbose && fprintf(stderr, Name fmt, ##arg))
2995 /* validate_geometry_imsm_volume - lifted from validate_geometry_ddf_bvd
2996 * FIX ME add ahci details
2998 static int validate_geometry_imsm_volume(struct supertype
*st
, int level
,
2999 int layout
, int raiddisks
, int chunk
,
3000 unsigned long long size
, char *dev
,
3001 unsigned long long *freesize
,
3005 struct intel_super
*super
= st
->sb
;
3006 struct imsm_super
*mpb
= super
->anchor
;
3008 unsigned long long pos
= 0;
3009 unsigned long long maxsize
;
3013 /* We must have the container info already read in. */
3017 if (!is_raid_level_supported(super
->orom
, level
, raiddisks
)) {
3018 pr_vrb(": platform does not support raid%d with %d disk%s\n",
3019 level
, raiddisks
, raiddisks
> 1 ? "s" : "");
3022 if (super
->orom
&& level
!= 1 &&
3023 !imsm_orom_has_chunk(super
->orom
, chunk
)) {
3024 pr_vrb(": platform does not support a chunk size of: %d\n", chunk
);
3027 if (layout
!= imsm_level_to_layout(level
)) {
3029 pr_vrb(": imsm raid 5 only supports the left-asymmetric layout\n");
3030 else if (level
== 10)
3031 pr_vrb(": imsm raid 10 only supports the n2 layout\n");
3033 pr_vrb(": imsm unknown layout %#x for this raid level %d\n",
3039 /* General test: make sure there is space for
3040 * 'raiddisks' device extents of size 'size' at a given
3043 unsigned long long minsize
= size
;
3044 unsigned long long start_offset
= ~0ULL;
3047 minsize
= MPB_SECTOR_CNT
+ IMSM_RESERVED_SECTORS
;
3048 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
3053 e
= get_extents(super
, dl
);
3056 unsigned long long esize
;
3057 esize
= e
[i
].start
- pos
;
3058 if (esize
>= minsize
)
3060 if (found
&& start_offset
== ~0ULL) {
3063 } else if (found
&& pos
!= start_offset
) {
3067 pos
= e
[i
].start
+ e
[i
].size
;
3069 } while (e
[i
-1].size
);
3074 if (dcnt
< raiddisks
) {
3076 fprintf(stderr
, Name
": imsm: Not enough "
3077 "devices with space for this array "
3085 /* This device must be a member of the set */
3086 if (stat(dev
, &stb
) < 0)
3088 if ((S_IFMT
& stb
.st_mode
) != S_IFBLK
)
3090 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
3091 if (dl
->major
== major(stb
.st_rdev
) &&
3092 dl
->minor
== minor(stb
.st_rdev
))
3097 fprintf(stderr
, Name
": %s is not in the "
3098 "same imsm set\n", dev
);
3100 } else if (super
->orom
&& dl
->index
< 0 && mpb
->num_raid_devs
) {
3101 /* If a volume is present then the current creation attempt
3102 * cannot incorporate new spares because the orom may not
3103 * understand this configuration (all member disks must be
3104 * members of each array in the container).
3106 fprintf(stderr
, Name
": %s is a spare and a volume"
3107 " is already defined for this container\n", dev
);
3108 fprintf(stderr
, Name
": The option-rom requires all member"
3109 " disks to be a member of all volumes\n");
3113 /* retrieve the largest free space block */
3114 e
= get_extents(super
, dl
);
3119 unsigned long long esize
;
3121 esize
= e
[i
].start
- pos
;
3122 if (esize
>= maxsize
)
3124 pos
= e
[i
].start
+ e
[i
].size
;
3126 } while (e
[i
-1].size
);
3131 fprintf(stderr
, Name
": unable to determine free space for: %s\n",
3135 if (maxsize
< size
) {
3137 fprintf(stderr
, Name
": %s not enough space (%llu < %llu)\n",
3138 dev
, maxsize
, size
);
3142 /* count total number of extents for merge */
3144 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
3146 i
+= dl
->extent_cnt
;
3148 maxsize
= merge_extents(super
, i
);
3149 if (maxsize
< size
) {
3151 fprintf(stderr
, Name
": not enough space after merge (%llu < %llu)\n",
3154 } else if (maxsize
== ~0ULL) {
3156 fprintf(stderr
, Name
": failed to merge %d extents\n", i
);
3160 *freesize
= maxsize
;
3165 static int reserve_space(struct supertype
*st
, int raiddisks
,
3166 unsigned long long size
, int chunk
,
3167 unsigned long long *freesize
)
3169 struct intel_super
*super
= st
->sb
;
3170 struct imsm_super
*mpb
= super
->anchor
;
3175 unsigned long long maxsize
;
3176 unsigned long long minsize
;
3180 /* find the largest common start free region of the possible disks */
3184 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
3190 /* don't activate new spares if we are orom constrained
3191 * and there is already a volume active in the container
3193 if (super
->orom
&& dl
->index
< 0 && mpb
->num_raid_devs
)
3196 e
= get_extents(super
, dl
);
3199 for (i
= 1; e
[i
-1].size
; i
++)
3207 maxsize
= merge_extents(super
, extent_cnt
);
3212 if (cnt
< raiddisks
||
3213 (super
->orom
&& used
&& used
!= raiddisks
) ||
3214 maxsize
< minsize
) {
3215 fprintf(stderr
, Name
": not enough devices with space to create array.\n");
3216 return 0; /* No enough free spaces large enough */
3228 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
3230 dl
->raiddisk
= cnt
++;
3237 static int validate_geometry_imsm(struct supertype
*st
, int level
, int layout
,
3238 int raiddisks
, int chunk
, unsigned long long size
,
3239 char *dev
, unsigned long long *freesize
,
3245 /* if given unused devices create a container
3246 * if given given devices in a container create a member volume
3248 if (level
== LEVEL_CONTAINER
) {
3249 /* Must be a fresh device to add to a container */
3250 return validate_geometry_imsm_container(st
, level
, layout
,
3251 raiddisks
, chunk
, size
,
3257 if (st
->sb
&& freesize
) {
3258 /* we are being asked to automatically layout a
3259 * new volume based on the current contents of
3260 * the container. If the the parameters can be
3261 * satisfied reserve_space will record the disks,
3262 * start offset, and size of the volume to be
3263 * created. add_to_super and getinfo_super
3264 * detect when autolayout is in progress.
3266 return reserve_space(st
, raiddisks
, size
, chunk
, freesize
);
3271 /* creating in a given container */
3272 return validate_geometry_imsm_volume(st
, level
, layout
,
3273 raiddisks
, chunk
, size
,
3274 dev
, freesize
, verbose
);
3277 /* limit creation to the following levels */
3289 /* This device needs to be a device in an 'imsm' container */
3290 fd
= open(dev
, O_RDONLY
|O_EXCL
, 0);
3294 Name
": Cannot create this array on device %s\n",
3299 if (errno
!= EBUSY
|| (fd
= open(dev
, O_RDONLY
, 0)) < 0) {
3301 fprintf(stderr
, Name
": Cannot open %s: %s\n",
3302 dev
, strerror(errno
));
3305 /* Well, it is in use by someone, maybe an 'imsm' container. */
3306 cfd
= open_container(fd
);
3310 fprintf(stderr
, Name
": Cannot use %s: It is busy\n",
3314 sra
= sysfs_read(cfd
, 0, GET_VERSION
);
3316 if (sra
&& sra
->array
.major_version
== -1 &&
3317 strcmp(sra
->text_version
, "imsm") == 0) {
3318 /* This is a member of a imsm container. Load the container
3319 * and try to create a volume
3321 struct intel_super
*super
;
3323 if (load_super_imsm_all(st
, cfd
, (void **) &super
, NULL
, 1) == 0) {
3325 st
->container_dev
= fd2devnum(cfd
);
3327 return validate_geometry_imsm_volume(st
, level
, layout
,
3333 } else /* may belong to another container */
3338 #endif /* MDASSEMBLE */
3340 static struct mdinfo
*container_content_imsm(struct supertype
*st
)
3342 /* Given a container loaded by load_super_imsm_all,
3343 * extract information about all the arrays into
3346 * For each imsm_dev create an mdinfo, fill it in,
3347 * then look for matching devices in super->disks
3348 * and create appropriate device mdinfo.
3350 struct intel_super
*super
= st
->sb
;
3351 struct imsm_super
*mpb
= super
->anchor
;
3352 struct mdinfo
*rest
= NULL
;
3355 /* do not assemble arrays that might have bad blocks */
3356 if (imsm_bbm_log_size(super
->anchor
)) {
3357 fprintf(stderr
, Name
": BBM log found in metadata. "
3358 "Cannot activate array(s).\n");
3362 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
3363 struct imsm_dev
*dev
= get_imsm_dev(super
, i
);
3364 struct imsm_map
*map
= get_imsm_map(dev
, 0);
3365 struct mdinfo
*this;
3368 /* do not publish arrays that are in the middle of an
3369 * unsupported migration
3371 if (dev
->vol
.migr_state
&&
3372 (migr_type(dev
) == MIGR_GEN_MIGR
||
3373 migr_type(dev
) == MIGR_STATE_CHANGE
)) {
3374 fprintf(stderr
, Name
": cannot assemble volume '%.16s':"
3375 " unsupported migration in progress\n",
3380 this = malloc(sizeof(*this));
3381 memset(this, 0, sizeof(*this));
3384 super
->current_vol
= i
;
3385 getinfo_super_imsm_volume(st
, this);
3386 for (slot
= 0 ; slot
< map
->num_members
; slot
++) {
3387 struct mdinfo
*info_d
;
3395 idx
= get_imsm_disk_idx(dev
, slot
);
3396 ord
= get_imsm_ord_tbl_ent(dev
, slot
);
3397 for (d
= super
->disks
; d
; d
= d
->next
)
3398 if (d
->index
== idx
)
3404 s
= d
? d
->disk
.status
: 0;
3405 if (s
& FAILED_DISK
)
3407 if (!(s
& USABLE_DISK
))
3409 if (ord
& IMSM_ORD_REBUILD
)
3413 * if we skip some disks the array will be assmebled degraded;
3414 * reset resync start to avoid a dirty-degraded situation
3416 * FIXME handle dirty degraded
3418 if (skip
&& !dev
->vol
.dirty
)
3419 this->resync_start
= ~0ULL;
3423 info_d
= malloc(sizeof(*info_d
));
3425 fprintf(stderr
, Name
": failed to allocate disk"
3426 " for volume %.16s\n", dev
->volume
);
3431 memset(info_d
, 0, sizeof(*info_d
));
3432 info_d
->next
= this->devs
;
3433 this->devs
= info_d
;
3435 info_d
->disk
.number
= d
->index
;
3436 info_d
->disk
.major
= d
->major
;
3437 info_d
->disk
.minor
= d
->minor
;
3438 info_d
->disk
.raid_disk
= slot
;
3440 this->array
.working_disks
++;
3442 info_d
->events
= __le32_to_cpu(mpb
->generation_num
);
3443 info_d
->data_offset
= __le32_to_cpu(map
->pba_of_lba0
);
3444 info_d
->component_size
= __le32_to_cpu(map
->blocks_per_member
);
3446 strcpy(info_d
->name
, d
->devname
);
3456 static int imsm_open_new(struct supertype
*c
, struct active_array
*a
,
3459 struct intel_super
*super
= c
->sb
;
3460 struct imsm_super
*mpb
= super
->anchor
;
3462 if (atoi(inst
) >= mpb
->num_raid_devs
) {
3463 fprintf(stderr
, "%s: subarry index %d, out of range\n",
3464 __func__
, atoi(inst
));
3468 dprintf("imsm: open_new %s\n", inst
);
3469 a
->info
.container_member
= atoi(inst
);
3473 static __u8
imsm_check_degraded(struct intel_super
*super
, struct imsm_dev
*dev
, int failed
)
3475 struct imsm_map
*map
= get_imsm_map(dev
, 0);
3478 return map
->map_state
== IMSM_T_STATE_UNINITIALIZED
?
3479 IMSM_T_STATE_UNINITIALIZED
: IMSM_T_STATE_NORMAL
;
3481 switch (get_imsm_raid_level(map
)) {
3483 return IMSM_T_STATE_FAILED
;
3486 if (failed
< map
->num_members
)
3487 return IMSM_T_STATE_DEGRADED
;
3489 return IMSM_T_STATE_FAILED
;
3494 * check to see if any mirrors have failed, otherwise we
3495 * are degraded. Even numbered slots are mirrored on
3499 /* gcc -Os complains that this is unused */
3500 int insync
= insync
;
3502 for (i
= 0; i
< map
->num_members
; i
++) {
3503 __u32 ord
= get_imsm_ord_tbl_ent(dev
, i
);
3504 int idx
= ord_to_idx(ord
);
3505 struct imsm_disk
*disk
;
3507 /* reset the potential in-sync count on even-numbered
3508 * slots. num_copies is always 2 for imsm raid10
3513 disk
= get_imsm_disk(super
, idx
);
3514 if (!disk
|| disk
->status
& FAILED_DISK
||
3515 ord
& IMSM_ORD_REBUILD
)
3518 /* no in-sync disks left in this mirror the
3522 return IMSM_T_STATE_FAILED
;
3525 return IMSM_T_STATE_DEGRADED
;
3529 return IMSM_T_STATE_DEGRADED
;
3531 return IMSM_T_STATE_FAILED
;
3537 return map
->map_state
;
3540 static int imsm_count_failed(struct intel_super
*super
, struct imsm_dev
*dev
)
3544 struct imsm_disk
*disk
;
3545 struct imsm_map
*map
= get_imsm_map(dev
, 0);
3546 struct imsm_map
*prev
= get_imsm_map(dev
, dev
->vol
.migr_state
);
3550 /* at the beginning of migration we set IMSM_ORD_REBUILD on
3551 * disks that are being rebuilt. New failures are recorded to
3552 * map[0]. So we look through all the disks we started with and
3553 * see if any failures are still present, or if any new ones
3556 * FIXME add support for online capacity expansion and
3557 * raid-level-migration
3559 for (i
= 0; i
< prev
->num_members
; i
++) {
3560 ord
= __le32_to_cpu(prev
->disk_ord_tbl
[i
]);
3561 ord
|= __le32_to_cpu(map
->disk_ord_tbl
[i
]);
3562 idx
= ord_to_idx(ord
);
3564 disk
= get_imsm_disk(super
, idx
);
3565 if (!disk
|| disk
->status
& FAILED_DISK
||
3566 ord
& IMSM_ORD_REBUILD
)
3573 static int is_resyncing(struct imsm_dev
*dev
)
3575 struct imsm_map
*migr_map
;
3577 if (!dev
->vol
.migr_state
)
3580 if (migr_type(dev
) == MIGR_INIT
||
3581 migr_type(dev
) == MIGR_REPAIR
)
3584 migr_map
= get_imsm_map(dev
, 1);
3586 if (migr_map
->map_state
== IMSM_T_STATE_NORMAL
)
3592 static int is_rebuilding(struct imsm_dev
*dev
)
3594 struct imsm_map
*migr_map
;
3596 if (!dev
->vol
.migr_state
)
3599 if (migr_type(dev
) != MIGR_REBUILD
)
3602 migr_map
= get_imsm_map(dev
, 1);
3604 if (migr_map
->map_state
== IMSM_T_STATE_DEGRADED
)
3610 /* return true if we recorded new information */
3611 static int mark_failure(struct imsm_dev
*dev
, struct imsm_disk
*disk
, int idx
)
3615 struct imsm_map
*map
;
3617 /* new failures are always set in map[0] */
3618 map
= get_imsm_map(dev
, 0);
3620 slot
= get_imsm_disk_slot(map
, idx
);
3624 ord
= __le32_to_cpu(map
->disk_ord_tbl
[slot
]);
3625 if ((disk
->status
& FAILED_DISK
) && (ord
& IMSM_ORD_REBUILD
))
3628 disk
->status
|= FAILED_DISK
;
3629 set_imsm_ord_tbl_ent(map
, slot
, idx
| IMSM_ORD_REBUILD
);
3630 if (map
->failed_disk_num
== ~0)
3631 map
->failed_disk_num
= slot
;
3635 static void mark_missing(struct imsm_dev
*dev
, struct imsm_disk
*disk
, int idx
)
3637 mark_failure(dev
, disk
, idx
);
3639 if (disk
->scsi_id
== __cpu_to_le32(~(__u32
)0))
3642 disk
->scsi_id
= __cpu_to_le32(~(__u32
)0);
3643 memmove(&disk
->serial
[0], &disk
->serial
[1], MAX_RAID_SERIAL_LEN
- 1);
3646 /* Handle dirty -> clean transititions and resync. Degraded and rebuild
3647 * states are handled in imsm_set_disk() with one exception, when a
3648 * resync is stopped due to a new failure this routine will set the
3649 * 'degraded' state for the array.
3651 static int imsm_set_array_state(struct active_array
*a
, int consistent
)
3653 int inst
= a
->info
.container_member
;
3654 struct intel_super
*super
= a
->container
->sb
;
3655 struct imsm_dev
*dev
= get_imsm_dev(super
, inst
);
3656 struct imsm_map
*map
= get_imsm_map(dev
, 0);
3657 int failed
= imsm_count_failed(super
, dev
);
3658 __u8 map_state
= imsm_check_degraded(super
, dev
, failed
);
3660 /* before we activate this array handle any missing disks */
3661 if (consistent
== 2 && super
->missing
) {
3664 dprintf("imsm: mark missing\n");
3665 end_migration(dev
, map_state
);
3666 for (dl
= super
->missing
; dl
; dl
= dl
->next
)
3667 mark_missing(dev
, &dl
->disk
, dl
->index
);
3668 super
->updates_pending
++;
3671 if (consistent
== 2 &&
3672 (!is_resync_complete(a
) ||
3673 map_state
!= IMSM_T_STATE_NORMAL
||
3674 dev
->vol
.migr_state
))
3677 if (is_resync_complete(a
)) {
3678 /* complete intialization / resync,
3679 * recovery and interrupted recovery is completed in
3682 if (is_resyncing(dev
)) {
3683 dprintf("imsm: mark resync done\n");
3684 end_migration(dev
, map_state
);
3685 super
->updates_pending
++;
3687 } else if (!is_resyncing(dev
) && !failed
) {
3688 /* mark the start of the init process if nothing is failed */
3689 dprintf("imsm: mark resync start (%llu)\n", a
->resync_start
);
3690 if (map
->map_state
== IMSM_T_STATE_UNINITIALIZED
)
3691 migrate(dev
, IMSM_T_STATE_NORMAL
, MIGR_INIT
);
3693 migrate(dev
, IMSM_T_STATE_NORMAL
, MIGR_REPAIR
);
3694 super
->updates_pending
++;
3697 /* FIXME check if we can update curr_migr_unit from resync_start */
3699 /* mark dirty / clean */
3700 if (dev
->vol
.dirty
!= !consistent
) {
3701 dprintf("imsm: mark '%s' (%llu)\n",
3702 consistent
? "clean" : "dirty", a
->resync_start
);
3707 super
->updates_pending
++;
3712 static void imsm_set_disk(struct active_array
*a
, int n
, int state
)
3714 int inst
= a
->info
.container_member
;
3715 struct intel_super
*super
= a
->container
->sb
;
3716 struct imsm_dev
*dev
= get_imsm_dev(super
, inst
);
3717 struct imsm_map
*map
= get_imsm_map(dev
, 0);
3718 struct imsm_disk
*disk
;
3723 if (n
> map
->num_members
)
3724 fprintf(stderr
, "imsm: set_disk %d out of range 0..%d\n",
3725 n
, map
->num_members
- 1);
3730 dprintf("imsm: set_disk %d:%x\n", n
, state
);
3732 ord
= get_imsm_ord_tbl_ent(dev
, n
);
3733 disk
= get_imsm_disk(super
, ord_to_idx(ord
));
3735 /* check for new failures */
3736 if (state
& DS_FAULTY
) {
3737 if (mark_failure(dev
, disk
, ord_to_idx(ord
)))
3738 super
->updates_pending
++;
3741 /* check if in_sync */
3742 if (state
& DS_INSYNC
&& ord
& IMSM_ORD_REBUILD
&& is_rebuilding(dev
)) {
3743 struct imsm_map
*migr_map
= get_imsm_map(dev
, 1);
3745 set_imsm_ord_tbl_ent(migr_map
, n
, ord_to_idx(ord
));
3746 super
->updates_pending
++;
3749 failed
= imsm_count_failed(super
, dev
);
3750 map_state
= imsm_check_degraded(super
, dev
, failed
);
3752 /* check if recovery complete, newly degraded, or failed */
3753 if (map_state
== IMSM_T_STATE_NORMAL
&& is_rebuilding(dev
)) {
3754 end_migration(dev
, map_state
);
3755 map
= get_imsm_map(dev
, 0);
3756 map
->failed_disk_num
= ~0;
3757 super
->updates_pending
++;
3758 } else if (map_state
== IMSM_T_STATE_DEGRADED
&&
3759 map
->map_state
!= map_state
&&
3760 !dev
->vol
.migr_state
) {
3761 dprintf("imsm: mark degraded\n");
3762 map
->map_state
= map_state
;
3763 super
->updates_pending
++;
3764 } else if (map_state
== IMSM_T_STATE_FAILED
&&
3765 map
->map_state
!= map_state
) {
3766 dprintf("imsm: mark failed\n");
3767 end_migration(dev
, map_state
);
3768 super
->updates_pending
++;
3772 static int store_imsm_mpb(int fd
, struct intel_super
*super
)
3774 struct imsm_super
*mpb
= super
->anchor
;
3775 __u32 mpb_size
= __le32_to_cpu(mpb
->mpb_size
);
3776 unsigned long long dsize
;
3777 unsigned long long sectors
;
3779 get_dev_size(fd
, NULL
, &dsize
);
3781 if (mpb_size
> 512) {
3782 /* -1 to account for anchor */
3783 sectors
= mpb_sectors(mpb
) - 1;
3785 /* write the extended mpb to the sectors preceeding the anchor */
3786 if (lseek64(fd
, dsize
- (512 * (2 + sectors
)), SEEK_SET
) < 0)
3789 if (write(fd
, super
->buf
+ 512, 512 * sectors
) != 512 * sectors
)
3793 /* first block is stored on second to last sector of the disk */
3794 if (lseek64(fd
, dsize
- (512 * 2), SEEK_SET
) < 0)
3797 if (write(fd
, super
->buf
, 512) != 512)
3803 static void imsm_sync_metadata(struct supertype
*container
)
3805 struct intel_super
*super
= container
->sb
;
3807 if (!super
->updates_pending
)
3810 write_super_imsm(super
, 0);
3812 super
->updates_pending
= 0;
3815 static struct dl
*imsm_readd(struct intel_super
*super
, int idx
, struct active_array
*a
)
3817 struct imsm_dev
*dev
= get_imsm_dev(super
, a
->info
.container_member
);
3818 int i
= get_imsm_disk_idx(dev
, idx
);
3821 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
3825 if (dl
&& dl
->disk
.status
& FAILED_DISK
)
3829 dprintf("%s: found %x:%x\n", __func__
, dl
->major
, dl
->minor
);
3834 static struct dl
*imsm_add_spare(struct intel_super
*super
, int slot
,
3835 struct active_array
*a
, int activate_new
)
3837 struct imsm_dev
*dev
= get_imsm_dev(super
, a
->info
.container_member
);
3838 int idx
= get_imsm_disk_idx(dev
, slot
);
3839 struct imsm_super
*mpb
= super
->anchor
;
3840 struct imsm_map
*map
;
3841 unsigned long long esize
;
3842 unsigned long long pos
;
3851 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
3852 /* If in this array, skip */
3853 for (d
= a
->info
.devs
; d
; d
= d
->next
)
3854 if (d
->state_fd
>= 0 &&
3855 d
->disk
.major
== dl
->major
&&
3856 d
->disk
.minor
== dl
->minor
) {
3857 dprintf("%x:%x already in array\n", dl
->major
, dl
->minor
);
3863 /* skip in use or failed drives */
3864 if (dl
->disk
.status
& FAILED_DISK
|| idx
== dl
->index
||
3866 dprintf("%x:%x status (failed: %d index: %d)\n",
3867 dl
->major
, dl
->minor
,
3868 (dl
->disk
.status
& FAILED_DISK
) == FAILED_DISK
, idx
);
3872 /* skip pure spares when we are looking for partially
3873 * assimilated drives
3875 if (dl
->index
== -1 && !activate_new
)
3878 /* Does this unused device have the requisite free space?
3879 * It needs to be able to cover all member volumes
3881 ex
= get_extents(super
, dl
);
3883 dprintf("cannot get extents\n");
3886 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
3887 dev
= get_imsm_dev(super
, i
);
3888 map
= get_imsm_map(dev
, 0);
3890 /* check if this disk is already a member of
3893 if (get_imsm_disk_slot(map
, dl
->index
) >= 0)
3899 array_start
= __le32_to_cpu(map
->pba_of_lba0
);
3900 blocks
= __le32_to_cpu(map
->blocks_per_member
);
3903 /* check that we can start at pba_of_lba0 with
3904 * blocks_per_member of space
3906 esize
= ex
[j
].start
- pos
;
3907 if (array_start
>= pos
&&
3908 array_start
+ blocks
< ex
[j
].start
) {
3912 pos
= ex
[j
].start
+ ex
[j
].size
;
3914 } while (ex
[j
-1].size
);
3921 if (i
< mpb
->num_raid_devs
) {
3922 dprintf("%x:%x does not have %u at %u\n",
3923 dl
->major
, dl
->minor
,
3924 blocks
, array_start
);
3934 static struct mdinfo
*imsm_activate_spare(struct active_array
*a
,
3935 struct metadata_update
**updates
)
3938 * Find a device with unused free space and use it to replace a
3939 * failed/vacant region in an array. We replace failed regions one a
3940 * array at a time. The result is that a new spare disk will be added
3941 * to the first failed array and after the monitor has finished
3942 * propagating failures the remainder will be consumed.
3944 * FIXME add a capability for mdmon to request spares from another
3948 struct intel_super
*super
= a
->container
->sb
;
3949 int inst
= a
->info
.container_member
;
3950 struct imsm_dev
*dev
= get_imsm_dev(super
, inst
);
3951 struct imsm_map
*map
= get_imsm_map(dev
, 0);
3952 int failed
= a
->info
.array
.raid_disks
;
3953 struct mdinfo
*rv
= NULL
;
3956 struct metadata_update
*mu
;
3958 struct imsm_update_activate_spare
*u
;
3962 for (d
= a
->info
.devs
; d
; d
= d
->next
) {
3963 if ((d
->curr_state
& DS_FAULTY
) &&
3965 /* wait for Removal to happen */
3967 if (d
->state_fd
>= 0)
3971 dprintf("imsm: activate spare: inst=%d failed=%d (%d) level=%d\n",
3972 inst
, failed
, a
->info
.array
.raid_disks
, a
->info
.array
.level
);
3973 if (imsm_check_degraded(super
, dev
, failed
) != IMSM_T_STATE_DEGRADED
)
3976 /* For each slot, if it is not working, find a spare */
3977 for (i
= 0; i
< a
->info
.array
.raid_disks
; i
++) {
3978 for (d
= a
->info
.devs
; d
; d
= d
->next
)
3979 if (d
->disk
.raid_disk
== i
)
3981 dprintf("found %d: %p %x\n", i
, d
, d
?d
->curr_state
:0);
3982 if (d
&& (d
->state_fd
>= 0))
3986 * OK, this device needs recovery. Try to re-add the
3987 * previous occupant of this slot, if this fails see if
3988 * we can continue the assimilation of a spare that was
3989 * partially assimilated, finally try to activate a new
3992 dl
= imsm_readd(super
, i
, a
);
3994 dl
= imsm_add_spare(super
, i
, a
, 0);
3996 dl
= imsm_add_spare(super
, i
, a
, 1);
4000 /* found a usable disk with enough space */
4001 di
= malloc(sizeof(*di
));
4004 memset(di
, 0, sizeof(*di
));
4006 /* dl->index will be -1 in the case we are activating a
4007 * pristine spare. imsm_process_update() will create a
4008 * new index in this case. Once a disk is found to be
4009 * failed in all member arrays it is kicked from the
4012 di
->disk
.number
= dl
->index
;
4014 /* (ab)use di->devs to store a pointer to the device
4017 di
->devs
= (struct mdinfo
*) dl
;
4019 di
->disk
.raid_disk
= i
;
4020 di
->disk
.major
= dl
->major
;
4021 di
->disk
.minor
= dl
->minor
;
4023 di
->data_offset
= __le32_to_cpu(map
->pba_of_lba0
);
4024 di
->component_size
= a
->info
.component_size
;
4025 di
->container_member
= inst
;
4029 dprintf("%x:%x to be %d at %llu\n", dl
->major
, dl
->minor
,
4030 i
, di
->data_offset
);
4036 /* No spares found */
4038 /* Now 'rv' has a list of devices to return.
4039 * Create a metadata_update record to update the
4040 * disk_ord_tbl for the array
4042 mu
= malloc(sizeof(*mu
));
4044 mu
->buf
= malloc(sizeof(struct imsm_update_activate_spare
) * num_spares
);
4045 if (mu
->buf
== NULL
) {
4052 struct mdinfo
*n
= rv
->next
;
4061 mu
->len
= sizeof(struct imsm_update_activate_spare
) * num_spares
;
4062 mu
->next
= *updates
;
4063 u
= (struct imsm_update_activate_spare
*) mu
->buf
;
4065 for (di
= rv
; di
; di
= di
->next
) {
4066 u
->type
= update_activate_spare
;
4067 u
->dl
= (struct dl
*) di
->devs
;
4069 u
->slot
= di
->disk
.raid_disk
;
4080 static int disks_overlap(struct intel_super
*super
, int idx
, struct imsm_update_create_array
*u
)
4082 struct imsm_dev
*dev
= get_imsm_dev(super
, idx
);
4083 struct imsm_map
*map
= get_imsm_map(dev
, 0);
4084 struct imsm_map
*new_map
= get_imsm_map(&u
->dev
, 0);
4085 struct disk_info
*inf
= get_disk_info(u
);
4086 struct imsm_disk
*disk
;
4090 for (i
= 0; i
< map
->num_members
; i
++) {
4091 disk
= get_imsm_disk(super
, get_imsm_disk_idx(dev
, i
));
4092 for (j
= 0; j
< new_map
->num_members
; j
++)
4093 if (serialcmp(disk
->serial
, inf
[j
].serial
) == 0)
4100 static void imsm_delete(struct intel_super
*super
, struct dl
**dlp
, int index
);
4102 static void imsm_process_update(struct supertype
*st
,
4103 struct metadata_update
*update
)
4106 * crack open the metadata_update envelope to find the update record
4107 * update can be one of:
4108 * update_activate_spare - a spare device has replaced a failed
4109 * device in an array, update the disk_ord_tbl. If this disk is
4110 * present in all member arrays then also clear the SPARE_DISK
4113 struct intel_super
*super
= st
->sb
;
4114 struct imsm_super
*mpb
;
4115 enum imsm_update_type type
= *(enum imsm_update_type
*) update
->buf
;
4117 /* update requires a larger buf but the allocation failed */
4118 if (super
->next_len
&& !super
->next_buf
) {
4119 super
->next_len
= 0;
4123 if (super
->next_buf
) {
4124 memcpy(super
->next_buf
, super
->buf
, super
->len
);
4126 super
->len
= super
->next_len
;
4127 super
->buf
= super
->next_buf
;
4129 super
->next_len
= 0;
4130 super
->next_buf
= NULL
;
4133 mpb
= super
->anchor
;
4136 case update_activate_spare
: {
4137 struct imsm_update_activate_spare
*u
= (void *) update
->buf
;
4138 struct imsm_dev
*dev
= get_imsm_dev(super
, u
->array
);
4139 struct imsm_map
*map
= get_imsm_map(dev
, 0);
4140 struct imsm_map
*migr_map
;
4141 struct active_array
*a
;
4142 struct imsm_disk
*disk
;
4147 int victim
= get_imsm_disk_idx(dev
, u
->slot
);
4150 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
4155 fprintf(stderr
, "error: imsm_activate_spare passed "
4156 "an unknown disk (index: %d)\n",
4161 super
->updates_pending
++;
4163 /* count failures (excluding rebuilds and the victim)
4164 * to determine map[0] state
4167 for (i
= 0; i
< map
->num_members
; i
++) {
4170 disk
= get_imsm_disk(super
, get_imsm_disk_idx(dev
, i
));
4171 if (!disk
|| disk
->status
& FAILED_DISK
)
4175 /* adding a pristine spare, assign a new index */
4176 if (dl
->index
< 0) {
4177 dl
->index
= super
->anchor
->num_disks
;
4178 super
->anchor
->num_disks
++;
4181 disk
->status
|= CONFIGURED_DISK
;
4182 disk
->status
&= ~SPARE_DISK
;
4185 to_state
= imsm_check_degraded(super
, dev
, failed
);
4186 map
->map_state
= IMSM_T_STATE_DEGRADED
;
4187 migrate(dev
, to_state
, MIGR_REBUILD
);
4188 migr_map
= get_imsm_map(dev
, 1);
4189 set_imsm_ord_tbl_ent(map
, u
->slot
, dl
->index
);
4190 set_imsm_ord_tbl_ent(migr_map
, u
->slot
, dl
->index
| IMSM_ORD_REBUILD
);
4192 /* count arrays using the victim in the metadata */
4194 for (a
= st
->arrays
; a
; a
= a
->next
) {
4195 dev
= get_imsm_dev(super
, a
->info
.container_member
);
4196 map
= get_imsm_map(dev
, 0);
4198 if (get_imsm_disk_slot(map
, victim
) >= 0)
4202 /* delete the victim if it is no longer being
4208 /* We know that 'manager' isn't touching anything,
4209 * so it is safe to delete
4211 for (dlp
= &super
->disks
; *dlp
; dlp
= &(*dlp
)->next
)
4212 if ((*dlp
)->index
== victim
)
4215 /* victim may be on the missing list */
4217 for (dlp
= &super
->missing
; *dlp
; dlp
= &(*dlp
)->next
)
4218 if ((*dlp
)->index
== victim
)
4220 imsm_delete(super
, dlp
, victim
);
4224 case update_create_array
: {
4225 /* someone wants to create a new array, we need to be aware of
4226 * a few races/collisions:
4227 * 1/ 'Create' called by two separate instances of mdadm
4228 * 2/ 'Create' versus 'activate_spare': mdadm has chosen
4229 * devices that have since been assimilated via
4231 * In the event this update can not be carried out mdadm will
4232 * (FIX ME) notice that its update did not take hold.
4234 struct imsm_update_create_array
*u
= (void *) update
->buf
;
4235 struct intel_dev
*dv
;
4236 struct imsm_dev
*dev
;
4237 struct imsm_map
*map
, *new_map
;
4238 unsigned long long start
, end
;
4239 unsigned long long new_start
, new_end
;
4241 struct disk_info
*inf
;
4244 /* handle racing creates: first come first serve */
4245 if (u
->dev_idx
< mpb
->num_raid_devs
) {
4246 dprintf("%s: subarray %d already defined\n",
4247 __func__
, u
->dev_idx
);
4251 /* check update is next in sequence */
4252 if (u
->dev_idx
!= mpb
->num_raid_devs
) {
4253 dprintf("%s: can not create array %d expected index %d\n",
4254 __func__
, u
->dev_idx
, mpb
->num_raid_devs
);
4258 new_map
= get_imsm_map(&u
->dev
, 0);
4259 new_start
= __le32_to_cpu(new_map
->pba_of_lba0
);
4260 new_end
= new_start
+ __le32_to_cpu(new_map
->blocks_per_member
);
4261 inf
= get_disk_info(u
);
4263 /* handle activate_spare versus create race:
4264 * check to make sure that overlapping arrays do not include
4267 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
4268 dev
= get_imsm_dev(super
, i
);
4269 map
= get_imsm_map(dev
, 0);
4270 start
= __le32_to_cpu(map
->pba_of_lba0
);
4271 end
= start
+ __le32_to_cpu(map
->blocks_per_member
);
4272 if ((new_start
>= start
&& new_start
<= end
) ||
4273 (start
>= new_start
&& start
<= new_end
))
4278 if (disks_overlap(super
, i
, u
)) {
4279 dprintf("%s: arrays overlap\n", __func__
);
4284 /* check that prepare update was successful */
4285 if (!update
->space
) {
4286 dprintf("%s: prepare update failed\n", __func__
);
4290 /* check that all disks are still active before committing
4291 * changes. FIXME: could we instead handle this by creating a
4292 * degraded array? That's probably not what the user expects,
4293 * so better to drop this update on the floor.
4295 for (i
= 0; i
< new_map
->num_members
; i
++) {
4296 dl
= serial_to_dl(inf
[i
].serial
, super
);
4298 dprintf("%s: disk disappeared\n", __func__
);
4303 super
->updates_pending
++;
4305 /* convert spares to members and fixup ord_tbl */
4306 for (i
= 0; i
< new_map
->num_members
; i
++) {
4307 dl
= serial_to_dl(inf
[i
].serial
, super
);
4308 if (dl
->index
== -1) {
4309 dl
->index
= mpb
->num_disks
;
4311 dl
->disk
.status
|= CONFIGURED_DISK
;
4312 dl
->disk
.status
&= ~SPARE_DISK
;
4314 set_imsm_ord_tbl_ent(new_map
, i
, dl
->index
);
4319 update
->space
= NULL
;
4320 imsm_copy_dev(dev
, &u
->dev
);
4321 dv
->index
= u
->dev_idx
;
4322 dv
->next
= super
->devlist
;
4323 super
->devlist
= dv
;
4324 mpb
->num_raid_devs
++;
4326 imsm_update_version_info(super
);
4329 /* mdmon knows how to release update->space, but not
4330 * ((struct intel_dev *) update->space)->dev
4332 if (update
->space
) {
4338 case update_add_disk
:
4340 /* we may be able to repair some arrays if disks are
4343 struct active_array
*a
;
4345 super
->updates_pending
++;
4346 for (a
= st
->arrays
; a
; a
= a
->next
)
4347 a
->check_degraded
= 1;
4349 /* add some spares to the metadata */
4350 while (super
->add
) {
4354 super
->add
= al
->next
;
4355 al
->next
= super
->disks
;
4357 dprintf("%s: added %x:%x\n",
4358 __func__
, al
->major
, al
->minor
);
4365 static void imsm_prepare_update(struct supertype
*st
,
4366 struct metadata_update
*update
)
4369 * Allocate space to hold new disk entries, raid-device entries or a new
4370 * mpb if necessary. The manager synchronously waits for updates to
4371 * complete in the monitor, so new mpb buffers allocated here can be
4372 * integrated by the monitor thread without worrying about live pointers
4373 * in the manager thread.
4375 enum imsm_update_type type
= *(enum imsm_update_type
*) update
->buf
;
4376 struct intel_super
*super
= st
->sb
;
4377 struct imsm_super
*mpb
= super
->anchor
;
4382 case update_create_array
: {
4383 struct imsm_update_create_array
*u
= (void *) update
->buf
;
4384 struct intel_dev
*dv
;
4385 struct imsm_dev
*dev
= &u
->dev
;
4386 struct imsm_map
*map
= get_imsm_map(dev
, 0);
4388 struct disk_info
*inf
;
4392 inf
= get_disk_info(u
);
4393 len
= sizeof_imsm_dev(dev
, 1);
4394 /* allocate a new super->devlist entry */
4395 dv
= malloc(sizeof(*dv
));
4397 dv
->dev
= malloc(len
);
4402 update
->space
= NULL
;
4406 /* count how many spares will be converted to members */
4407 for (i
= 0; i
< map
->num_members
; i
++) {
4408 dl
= serial_to_dl(inf
[i
].serial
, super
);
4410 /* hmm maybe it failed?, nothing we can do about
4415 if (count_memberships(dl
, super
) == 0)
4418 len
+= activate
* sizeof(struct imsm_disk
);
4425 /* check if we need a larger metadata buffer */
4426 if (super
->next_buf
)
4427 buf_len
= super
->next_len
;
4429 buf_len
= super
->len
;
4431 if (__le32_to_cpu(mpb
->mpb_size
) + len
> buf_len
) {
4432 /* ok we need a larger buf than what is currently allocated
4433 * if this allocation fails process_update will notice that
4434 * ->next_len is set and ->next_buf is NULL
4436 buf_len
= ROUND_UP(__le32_to_cpu(mpb
->mpb_size
) + len
, 512);
4437 if (super
->next_buf
)
4438 free(super
->next_buf
);
4440 super
->next_len
= buf_len
;
4441 if (posix_memalign(&super
->next_buf
, 512, buf_len
) == 0)
4442 memset(super
->next_buf
, 0, buf_len
);
4444 super
->next_buf
= NULL
;
4448 /* must be called while manager is quiesced */
4449 static void imsm_delete(struct intel_super
*super
, struct dl
**dlp
, int index
)
4451 struct imsm_super
*mpb
= super
->anchor
;
4453 struct imsm_dev
*dev
;
4454 struct imsm_map
*map
;
4455 int i
, j
, num_members
;
4458 dprintf("%s: deleting device[%d] from imsm_super\n",
4461 /* shift all indexes down one */
4462 for (iter
= super
->disks
; iter
; iter
= iter
->next
)
4463 if (iter
->index
> index
)
4465 for (iter
= super
->missing
; iter
; iter
= iter
->next
)
4466 if (iter
->index
> index
)
4469 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
4470 dev
= get_imsm_dev(super
, i
);
4471 map
= get_imsm_map(dev
, 0);
4472 num_members
= map
->num_members
;
4473 for (j
= 0; j
< num_members
; j
++) {
4474 /* update ord entries being careful not to propagate
4475 * ord-flags to the first map
4477 ord
= get_imsm_ord_tbl_ent(dev
, j
);
4479 if (ord_to_idx(ord
) <= index
)
4482 map
= get_imsm_map(dev
, 0);
4483 set_imsm_ord_tbl_ent(map
, j
, ord_to_idx(ord
- 1));
4484 map
= get_imsm_map(dev
, 1);
4486 set_imsm_ord_tbl_ent(map
, j
, ord
- 1);
4491 super
->updates_pending
++;
4493 struct dl
*dl
= *dlp
;
4495 *dlp
= (*dlp
)->next
;
4496 __free_imsm_disk(dl
);
4499 #endif /* MDASSEMBLE */
4501 struct superswitch super_imsm
= {
4503 .examine_super
= examine_super_imsm
,
4504 .brief_examine_super
= brief_examine_super_imsm
,
4505 .export_examine_super
= export_examine_super_imsm
,
4506 .detail_super
= detail_super_imsm
,
4507 .brief_detail_super
= brief_detail_super_imsm
,
4508 .write_init_super
= write_init_super_imsm
,
4509 .validate_geometry
= validate_geometry_imsm
,
4510 .add_to_super
= add_to_super_imsm
,
4511 .detail_platform
= detail_platform_imsm
,
4513 .match_home
= match_home_imsm
,
4514 .uuid_from_super
= uuid_from_super_imsm
,
4515 .getinfo_super
= getinfo_super_imsm
,
4516 .update_super
= update_super_imsm
,
4518 .avail_size
= avail_size_imsm
,
4520 .compare_super
= compare_super_imsm
,
4522 .load_super
= load_super_imsm
,
4523 .init_super
= init_super_imsm
,
4524 .store_super
= store_zero_imsm
,
4525 .free_super
= free_super_imsm
,
4526 .match_metadata_desc
= match_metadata_desc_imsm
,
4527 .container_content
= container_content_imsm
,
4528 .default_layout
= imsm_level_to_layout
,
4535 .open_new
= imsm_open_new
,
4536 .load_super
= load_super_imsm
,
4537 .set_array_state
= imsm_set_array_state
,
4538 .set_disk
= imsm_set_disk
,
4539 .sync_metadata
= imsm_sync_metadata
,
4540 .activate_spare
= imsm_activate_spare
,
4541 .process_update
= imsm_process_update
,
4542 .prepare_update
= imsm_prepare_update
,
4543 #endif /* MDASSEMBLE */