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 2210
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 __u32 status
; /* 0xF0 - 0xF3 */
68 __u32 owner_cfg_num
; /* which config 0,1,2... owns this disk */
69 #define IMSM_DISK_FILLERS 4
70 __u32 filler
[IMSM_DISK_FILLERS
]; /* 0xF4 - 0x107 MPB_DISK_FILLERS for future expansion */
73 /* RAID map configuration infos. */
75 __u32 pba_of_lba0
; /* start address of partition */
76 __u32 blocks_per_member
;/* blocks per member */
77 __u32 num_data_stripes
; /* number of data stripes */
78 __u16 blocks_per_strip
;
79 __u8 map_state
; /* Normal, Uninitialized, Degraded, Failed */
80 #define IMSM_T_STATE_NORMAL 0
81 #define IMSM_T_STATE_UNINITIALIZED 1
82 #define IMSM_T_STATE_DEGRADED 2
83 #define IMSM_T_STATE_FAILED 3
85 #define IMSM_T_RAID0 0
86 #define IMSM_T_RAID1 1
87 #define IMSM_T_RAID5 5 /* since metadata version 1.2.02 ? */
88 __u8 num_members
; /* number of member disks */
89 __u8 num_domains
; /* number of parity domains */
90 __u8 failed_disk_num
; /* valid only when state is degraded */
92 __u32 filler
[7]; /* expansion area */
93 #define IMSM_ORD_REBUILD (1 << 24)
94 __u32 disk_ord_tbl
[1]; /* disk_ord_tbl[num_members],
95 * top byte contains some flags
97 } __attribute__ ((packed
));
100 __u32 curr_migr_unit
;
101 __u32 checkpoint_id
; /* id to access curr_migr_unit */
102 __u8 migr_state
; /* Normal or Migrating */
104 #define MIGR_REBUILD 1
105 #define MIGR_VERIFY 2 /* analagous to echo check > sync_action */
106 #define MIGR_GEN_MIGR 3
107 #define MIGR_STATE_CHANGE 4
108 #define MIGR_REPAIR 5
109 __u8 migr_type
; /* Initializing, Rebuilding, ... */
111 __u8 fs_state
; /* fast-sync state for CnG (0xff == disabled) */
112 __u16 verify_errors
; /* number of mismatches */
113 __u16 bad_blocks
; /* number of bad blocks during verify */
115 struct imsm_map map
[1];
116 /* here comes another one if migr_state */
117 } __attribute__ ((packed
));
120 __u8 volume
[MAX_RAID_SERIAL_LEN
];
123 #define DEV_BOOTABLE __cpu_to_le32(0x01)
124 #define DEV_BOOT_DEVICE __cpu_to_le32(0x02)
125 #define DEV_READ_COALESCING __cpu_to_le32(0x04)
126 #define DEV_WRITE_COALESCING __cpu_to_le32(0x08)
127 #define DEV_LAST_SHUTDOWN_DIRTY __cpu_to_le32(0x10)
128 #define DEV_HIDDEN_AT_BOOT __cpu_to_le32(0x20)
129 #define DEV_CURRENTLY_HIDDEN __cpu_to_le32(0x40)
130 #define DEV_VERIFY_AND_FIX __cpu_to_le32(0x80)
131 #define DEV_MAP_STATE_UNINIT __cpu_to_le32(0x100)
132 #define DEV_NO_AUTO_RECOVERY __cpu_to_le32(0x200)
133 #define DEV_CLONE_N_GO __cpu_to_le32(0x400)
134 #define DEV_CLONE_MAN_SYNC __cpu_to_le32(0x800)
135 #define DEV_CNG_MASTER_DISK_NUM __cpu_to_le32(0x1000)
136 __u32 status
; /* Persistent RaidDev status */
137 __u32 reserved_blocks
; /* Reserved blocks at beginning of volume */
141 __u8 cng_master_disk
;
145 #define IMSM_DEV_FILLERS 10
146 __u32 filler
[IMSM_DEV_FILLERS
];
148 } __attribute__ ((packed
));
151 __u8 sig
[MAX_SIGNATURE_LENGTH
]; /* 0x00 - 0x1F */
152 __u32 check_sum
; /* 0x20 - 0x23 MPB Checksum */
153 __u32 mpb_size
; /* 0x24 - 0x27 Size of MPB */
154 __u32 family_num
; /* 0x28 - 0x2B Checksum from first time this config was written */
155 __u32 generation_num
; /* 0x2C - 0x2F Incremented each time this array's MPB is written */
156 __u32 error_log_size
; /* 0x30 - 0x33 in bytes */
157 __u32 attributes
; /* 0x34 - 0x37 */
158 __u8 num_disks
; /* 0x38 Number of configured disks */
159 __u8 num_raid_devs
; /* 0x39 Number of configured volumes */
160 __u8 error_log_pos
; /* 0x3A */
161 __u8 fill
[1]; /* 0x3B */
162 __u32 cache_size
; /* 0x3c - 0x40 in mb */
163 __u32 orig_family_num
; /* 0x40 - 0x43 original family num */
164 __u32 pwr_cycle_count
; /* 0x44 - 0x47 simulated power cycle count for array */
165 __u32 bbm_log_size
; /* 0x48 - 0x4B - size of bad Block Mgmt Log in bytes */
166 #define IMSM_FILLERS 35
167 __u32 filler
[IMSM_FILLERS
]; /* 0x4C - 0xD7 RAID_MPB_FILLERS */
168 struct imsm_disk disk
[1]; /* 0xD8 diskTbl[numDisks] */
169 /* here comes imsm_dev[num_raid_devs] */
170 /* here comes BBM logs */
171 } __attribute__ ((packed
));
173 #define BBM_LOG_MAX_ENTRIES 254
175 struct bbm_log_entry
{
176 __u64 defective_block_start
;
177 #define UNREADABLE 0xFFFFFFFF
178 __u32 spare_block_offset
;
179 __u16 remapped_marked_count
;
181 } __attribute__ ((__packed__
));
184 __u32 signature
; /* 0xABADB10C */
186 __u32 reserved_spare_block_count
; /* 0 */
187 __u32 reserved
; /* 0xFFFF */
188 __u64 first_spare_lba
;
189 struct bbm_log_entry mapped_block_entries
[BBM_LOG_MAX_ENTRIES
];
190 } __attribute__ ((__packed__
));
194 static char *map_state_str
[] = { "normal", "uninitialized", "degraded", "failed" };
197 #define RAID_DISK_RESERVED_BLOCKS_IMSM_HI 2209
199 #define GEN_MIGR_AREA_SIZE 2048 /* General Migration Copy Area size in blocks */
201 #define UNIT_SRC_NORMAL 0 /* Source data for curr_migr_unit must
202 * be recovered using srcMap */
203 #define UNIT_SRC_IN_CP_AREA 1 /* Source data for curr_migr_unit has
204 * already been migrated and must
205 * be recovered from checkpoint area */
207 __u32 rec_status
; /* Status used to determine how to restart
208 * migration in case it aborts
210 __u32 curr_migr_unit
; /* 0..numMigrUnits-1 */
211 __u32 family_num
; /* Family number of MPB
212 * containing the RaidDev
213 * that is migrating */
214 __u32 ascending_migr
; /* True if migrating in increasing
216 __u32 blocks_per_unit
; /* Num disk blocks per unit of operation */
217 __u32 dest_depth_per_unit
; /* Num member blocks each destMap
219 * advances per unit-of-operation */
220 __u32 ckpt_area_pba
; /* Pba of first block of ckpt copy area */
221 __u32 dest_1st_member_lba
; /* First member lba on first
222 * stripe of destination */
223 __u32 num_migr_units
; /* Total num migration units-of-op */
224 __u32 post_migr_vol_cap
; /* Size of volume after
225 * migration completes */
226 __u32 post_migr_vol_cap_hi
; /* Expansion space for LBA64 */
227 __u32 ckpt_read_disk_num
; /* Which member disk in destSubMap[0] the
228 * migration ckpt record was read from
229 * (for recovered migrations) */
230 } __attribute__ ((__packed__
));
232 static __u8
migr_type(struct imsm_dev
*dev
)
234 if (dev
->vol
.migr_type
== MIGR_VERIFY
&&
235 dev
->status
& DEV_VERIFY_AND_FIX
)
238 return dev
->vol
.migr_type
;
241 static void set_migr_type(struct imsm_dev
*dev
, __u8 migr_type
)
243 /* for compatibility with older oroms convert MIGR_REPAIR, into
244 * MIGR_VERIFY w/ DEV_VERIFY_AND_FIX status
246 if (migr_type
== MIGR_REPAIR
) {
247 dev
->vol
.migr_type
= MIGR_VERIFY
;
248 dev
->status
|= DEV_VERIFY_AND_FIX
;
250 dev
->vol
.migr_type
= migr_type
;
251 dev
->status
&= ~DEV_VERIFY_AND_FIX
;
255 static unsigned int sector_count(__u32 bytes
)
257 return ((bytes
+ (512-1)) & (~(512-1))) / 512;
260 static unsigned int mpb_sectors(struct imsm_super
*mpb
)
262 return sector_count(__le32_to_cpu(mpb
->mpb_size
));
266 struct imsm_dev
*dev
;
267 struct intel_dev
*next
;
272 enum sys_dev_type type
;
275 struct intel_hba
*next
;
282 /* internal representation of IMSM metadata */
285 void *buf
; /* O_DIRECT buffer for reading/writing metadata */
286 struct imsm_super
*anchor
; /* immovable parameters */
289 void *migr_rec_buf
; /* buffer for I/O operations */
290 struct migr_record
*migr_rec
; /* migration record */
292 size_t len
; /* size of the 'buf' allocation */
293 void *next_buf
; /* for realloc'ing buf from the manager */
295 int updates_pending
; /* count of pending updates for mdmon */
296 int current_vol
; /* index of raid device undergoing creation */
297 __u32 create_offset
; /* common start for 'current_vol' */
298 __u32 random
; /* random data for seeding new family numbers */
299 struct intel_dev
*devlist
;
303 __u8 serial
[MAX_RAID_SERIAL_LEN
];
306 struct imsm_disk disk
;
309 struct extent
*e
; /* for determining freespace @ create */
310 int raiddisk
; /* slot to fill in autolayout */
313 struct dl
*disk_mgmt_list
; /* list of disks to add/remove while mdmon
315 struct dl
*missing
; /* disks removed while we weren't looking */
316 struct bbm_log
*bbm_log
;
317 struct intel_hba
*hba
; /* device path of the raid controller for this metadata */
318 const struct imsm_orom
*orom
; /* platform firmware support */
319 struct intel_super
*next
; /* (temp) list for disambiguating family_num */
323 struct imsm_disk disk
;
324 #define IMSM_UNKNOWN_OWNER (-1)
326 struct intel_disk
*next
;
330 unsigned long long start
, size
;
333 /* definitions of reshape process types */
334 enum imsm_reshape_type
{
339 /* definition of messages passed to imsm_process_update */
340 enum imsm_update_type
{
341 update_activate_spare
,
345 update_add_remove_disk
,
346 update_reshape_container_disks
,
347 update_reshape_migration
,
349 update_general_migration_checkpoint
,
352 struct imsm_update_activate_spare
{
353 enum imsm_update_type type
;
357 struct imsm_update_activate_spare
*next
;
370 enum takeover_direction
{
374 struct imsm_update_takeover
{
375 enum imsm_update_type type
;
377 enum takeover_direction direction
;
380 struct imsm_update_reshape
{
381 enum imsm_update_type type
;
385 int new_disks
[1]; /* new_raid_disks - old_raid_disks makedev number */
388 struct imsm_update_reshape_migration
{
389 enum imsm_update_type type
;
392 /* fields for array migration changes
399 int new_disks
[1]; /* new_raid_disks - old_raid_disks makedev number */
402 struct imsm_update_general_migration_checkpoint
{
403 enum imsm_update_type type
;
404 __u32 curr_migr_unit
;
408 __u8 serial
[MAX_RAID_SERIAL_LEN
];
411 struct imsm_update_create_array
{
412 enum imsm_update_type type
;
417 struct imsm_update_kill_array
{
418 enum imsm_update_type type
;
422 struct imsm_update_rename_array
{
423 enum imsm_update_type type
;
424 __u8 name
[MAX_RAID_SERIAL_LEN
];
428 struct imsm_update_add_remove_disk
{
429 enum imsm_update_type type
;
433 static const char *_sys_dev_type
[] = {
434 [SYS_DEV_UNKNOWN
] = "Unknown",
435 [SYS_DEV_SAS
] = "SAS",
436 [SYS_DEV_SATA
] = "SATA"
439 const char *get_sys_dev_type(enum sys_dev_type type
)
441 if (type
>= SYS_DEV_MAX
)
442 type
= SYS_DEV_UNKNOWN
;
444 return _sys_dev_type
[type
];
447 static struct intel_hba
* alloc_intel_hba(struct sys_dev
*device
)
449 struct intel_hba
*result
= malloc(sizeof(*result
));
451 result
->type
= device
->type
;
452 result
->path
= strdup(device
->path
);
454 if (result
->path
&& (result
->pci_id
= strrchr(result
->path
, '/')) != NULL
)
460 static struct intel_hba
* find_intel_hba(struct intel_hba
*hba
, struct sys_dev
*device
)
462 struct intel_hba
*result
=NULL
;
463 for (result
= hba
; result
; result
= result
->next
) {
464 if (result
->type
== device
->type
&& strcmp(result
->path
, device
->path
) == 0)
470 static int attach_hba_to_super(struct intel_super
*super
, struct sys_dev
*device
)
472 struct intel_hba
*hba
;
474 /* check if disk attached to Intel HBA */
475 hba
= find_intel_hba(super
->hba
, device
);
478 /* Check if HBA is already attached to super */
479 if (super
->hba
== NULL
) {
480 super
->hba
= alloc_intel_hba(device
);
485 /* Intel metadata allows for all disks attached to the same type HBA.
486 * Do not sypport odf HBA types mixing
488 if (device
->type
!= hba
->type
)
494 hba
->next
= alloc_intel_hba(device
);
498 static struct sys_dev
* find_disk_attached_hba(int fd
, const char *devname
)
500 struct sys_dev
*list
, *elem
, *prev
;
503 if ((list
= find_intel_devices()) == NULL
)
507 disk_path
= (char *) devname
;
509 disk_path
= diskfd_to_devpath(fd
);
516 for (prev
= NULL
, elem
= list
; elem
; prev
= elem
, elem
= elem
->next
) {
517 if (path_attached_to_hba(disk_path
, elem
->path
)) {
521 prev
->next
= elem
->next
;
523 if (disk_path
!= devname
)
529 if (disk_path
!= devname
)
537 static int find_intel_hba_capability(int fd
, struct intel_super
*super
,
540 static struct supertype
*match_metadata_desc_imsm(char *arg
)
542 struct supertype
*st
;
544 if (strcmp(arg
, "imsm") != 0 &&
545 strcmp(arg
, "default") != 0
549 st
= malloc(sizeof(*st
));
552 memset(st
, 0, sizeof(*st
));
553 st
->container_dev
= NoMdDev
;
554 st
->ss
= &super_imsm
;
555 st
->max_devs
= IMSM_MAX_DEVICES
;
556 st
->minor_version
= 0;
562 static __u8
*get_imsm_version(struct imsm_super
*mpb
)
564 return &mpb
->sig
[MPB_SIG_LEN
];
568 /* retrieve a disk directly from the anchor when the anchor is known to be
569 * up-to-date, currently only at load time
571 static struct imsm_disk
*__get_imsm_disk(struct imsm_super
*mpb
, __u8 index
)
573 if (index
>= mpb
->num_disks
)
575 return &mpb
->disk
[index
];
578 /* retrieve the disk description based on a index of the disk
581 static struct dl
*get_imsm_dl_disk(struct intel_super
*super
, __u8 index
)
585 for (d
= super
->disks
; d
; d
= d
->next
)
586 if (d
->index
== index
)
591 /* retrieve a disk from the parsed metadata */
592 static struct imsm_disk
*get_imsm_disk(struct intel_super
*super
, __u8 index
)
596 dl
= get_imsm_dl_disk(super
, index
);
603 /* generate a checksum directly from the anchor when the anchor is known to be
604 * up-to-date, currently only at load or write_super after coalescing
606 static __u32
__gen_imsm_checksum(struct imsm_super
*mpb
)
608 __u32 end
= mpb
->mpb_size
/ sizeof(end
);
609 __u32
*p
= (__u32
*) mpb
;
613 sum
+= __le32_to_cpu(*p
);
617 return sum
- __le32_to_cpu(mpb
->check_sum
);
620 static size_t sizeof_imsm_map(struct imsm_map
*map
)
622 return sizeof(struct imsm_map
) + sizeof(__u32
) * (map
->num_members
- 1);
625 struct imsm_map
*get_imsm_map(struct imsm_dev
*dev
, int second_map
)
627 /* A device can have 2 maps if it is in the middle of a migration.
629 * 0 - we return the first map
630 * 1 - we return the second map if it exists, else NULL
631 * -1 - we return the second map if it exists, else the first
633 struct imsm_map
*map
= &dev
->vol
.map
[0];
635 if (second_map
== 1 && !dev
->vol
.migr_state
)
637 else if (second_map
== 1 ||
638 (second_map
< 0 && dev
->vol
.migr_state
)) {
641 return ptr
+ sizeof_imsm_map(map
);
647 /* return the size of the device.
648 * migr_state increases the returned size if map[0] were to be duplicated
650 static size_t sizeof_imsm_dev(struct imsm_dev
*dev
, int migr_state
)
652 size_t size
= sizeof(*dev
) - sizeof(struct imsm_map
) +
653 sizeof_imsm_map(get_imsm_map(dev
, 0));
655 /* migrating means an additional map */
656 if (dev
->vol
.migr_state
)
657 size
+= sizeof_imsm_map(get_imsm_map(dev
, 1));
659 size
+= sizeof_imsm_map(get_imsm_map(dev
, 0));
665 /* retrieve disk serial number list from a metadata update */
666 static struct disk_info
*get_disk_info(struct imsm_update_create_array
*update
)
669 struct disk_info
*inf
;
671 inf
= u
+ sizeof(*update
) - sizeof(struct imsm_dev
) +
672 sizeof_imsm_dev(&update
->dev
, 0);
678 static struct imsm_dev
*__get_imsm_dev(struct imsm_super
*mpb
, __u8 index
)
684 if (index
>= mpb
->num_raid_devs
)
687 /* devices start after all disks */
688 offset
= ((void *) &mpb
->disk
[mpb
->num_disks
]) - _mpb
;
690 for (i
= 0; i
<= index
; i
++)
692 return _mpb
+ offset
;
694 offset
+= sizeof_imsm_dev(_mpb
+ offset
, 0);
699 static struct imsm_dev
*get_imsm_dev(struct intel_super
*super
, __u8 index
)
701 struct intel_dev
*dv
;
703 if (index
>= super
->anchor
->num_raid_devs
)
705 for (dv
= super
->devlist
; dv
; dv
= dv
->next
)
706 if (dv
->index
== index
)
714 * == 1 get second map
715 * == -1 than get map according to the current migr_state
717 static __u32
get_imsm_ord_tbl_ent(struct imsm_dev
*dev
,
721 struct imsm_map
*map
;
723 map
= get_imsm_map(dev
, second_map
);
725 /* top byte identifies disk under rebuild */
726 return __le32_to_cpu(map
->disk_ord_tbl
[slot
]);
729 #define ord_to_idx(ord) (((ord) << 8) >> 8)
730 static __u32
get_imsm_disk_idx(struct imsm_dev
*dev
, int slot
, int second_map
)
732 __u32 ord
= get_imsm_ord_tbl_ent(dev
, slot
, second_map
);
734 return ord_to_idx(ord
);
737 static void set_imsm_ord_tbl_ent(struct imsm_map
*map
, int slot
, __u32 ord
)
739 map
->disk_ord_tbl
[slot
] = __cpu_to_le32(ord
);
742 static int get_imsm_disk_slot(struct imsm_map
*map
, unsigned idx
)
747 for (slot
= 0; slot
< map
->num_members
; slot
++) {
748 ord
= __le32_to_cpu(map
->disk_ord_tbl
[slot
]);
749 if (ord_to_idx(ord
) == idx
)
756 static int get_imsm_raid_level(struct imsm_map
*map
)
758 if (map
->raid_level
== 1) {
759 if (map
->num_members
== 2)
765 return map
->raid_level
;
768 static int cmp_extent(const void *av
, const void *bv
)
770 const struct extent
*a
= av
;
771 const struct extent
*b
= bv
;
772 if (a
->start
< b
->start
)
774 if (a
->start
> b
->start
)
779 static int count_memberships(struct dl
*dl
, struct intel_super
*super
)
784 for (i
= 0; i
< super
->anchor
->num_raid_devs
; i
++) {
785 struct imsm_dev
*dev
= get_imsm_dev(super
, i
);
786 struct imsm_map
*map
= get_imsm_map(dev
, 0);
788 if (get_imsm_disk_slot(map
, dl
->index
) >= 0)
795 static struct extent
*get_extents(struct intel_super
*super
, struct dl
*dl
)
797 /* find a list of used extents on the given physical device */
798 struct extent
*rv
, *e
;
800 int memberships
= count_memberships(dl
, super
);
801 __u32 reservation
= MPB_SECTOR_CNT
+ IMSM_RESERVED_SECTORS
;
803 rv
= malloc(sizeof(struct extent
) * (memberships
+ 1));
808 for (i
= 0; i
< super
->anchor
->num_raid_devs
; i
++) {
809 struct imsm_dev
*dev
= get_imsm_dev(super
, i
);
810 struct imsm_map
*map
= get_imsm_map(dev
, 0);
812 if (get_imsm_disk_slot(map
, dl
->index
) >= 0) {
813 e
->start
= __le32_to_cpu(map
->pba_of_lba0
);
814 e
->size
= __le32_to_cpu(map
->blocks_per_member
);
818 qsort(rv
, memberships
, sizeof(*rv
), cmp_extent
);
820 /* determine the start of the metadata
821 * when no raid devices are defined use the default
822 * ...otherwise allow the metadata to truncate the value
823 * as is the case with older versions of imsm
826 struct extent
*last
= &rv
[memberships
- 1];
829 remainder
= __le32_to_cpu(dl
->disk
.total_blocks
) -
830 (last
->start
+ last
->size
);
831 /* round down to 1k block to satisfy precision of the kernel
835 /* make sure remainder is still sane */
836 if (remainder
< (unsigned)ROUND_UP(super
->len
, 512) >> 9)
837 remainder
= ROUND_UP(super
->len
, 512) >> 9;
838 if (reservation
> remainder
)
839 reservation
= remainder
;
841 e
->start
= __le32_to_cpu(dl
->disk
.total_blocks
) - reservation
;
846 /* try to determine how much space is reserved for metadata from
847 * the last get_extents() entry, otherwise fallback to the
850 static __u32
imsm_reserved_sectors(struct intel_super
*super
, struct dl
*dl
)
856 /* for spares just return a minimal reservation which will grow
857 * once the spare is picked up by an array
860 return MPB_SECTOR_CNT
;
862 e
= get_extents(super
, dl
);
864 return MPB_SECTOR_CNT
+ IMSM_RESERVED_SECTORS
;
866 /* scroll to last entry */
867 for (i
= 0; e
[i
].size
; i
++)
870 rv
= __le32_to_cpu(dl
->disk
.total_blocks
) - e
[i
].start
;
877 static int is_spare(struct imsm_disk
*disk
)
879 return (disk
->status
& SPARE_DISK
) == SPARE_DISK
;
882 static int is_configured(struct imsm_disk
*disk
)
884 return (disk
->status
& CONFIGURED_DISK
) == CONFIGURED_DISK
;
887 static int is_failed(struct imsm_disk
*disk
)
889 return (disk
->status
& FAILED_DISK
) == FAILED_DISK
;
892 /* Return minimum size of a spare that can be used in this array*/
893 static unsigned long long min_acceptable_spare_size_imsm(struct supertype
*st
)
895 struct intel_super
*super
= st
->sb
;
899 unsigned long long rv
= 0;
903 /* find first active disk in array */
905 while (dl
&& (is_failed(&dl
->disk
) || dl
->index
== -1))
909 /* find last lba used by subarrays */
910 e
= get_extents(super
, dl
);
913 for (i
= 0; e
[i
].size
; i
++)
916 rv
= e
[i
-1].start
+ e
[i
-1].size
;
918 /* add the amount of space needed for metadata */
919 rv
= rv
+ MPB_SECTOR_CNT
+ IMSM_RESERVED_SECTORS
;
924 static __u64
blocks_per_migr_unit(struct intel_super
*super
,
925 struct imsm_dev
*dev
);
927 static void print_imsm_dev(struct intel_super
*super
,
928 struct imsm_dev
*dev
,
934 struct imsm_map
*map
= get_imsm_map(dev
, 0);
935 struct imsm_map
*map2
= get_imsm_map(dev
, 1);
939 printf("[%.16s]:\n", dev
->volume
);
940 printf(" UUID : %s\n", uuid
);
941 printf(" RAID Level : %d", get_imsm_raid_level(map
));
943 printf(" <-- %d", get_imsm_raid_level(map2
));
945 printf(" Members : %d", map
->num_members
);
947 printf(" <-- %d", map2
->num_members
);
949 printf(" Slots : [");
950 for (i
= 0; i
< map
->num_members
; i
++) {
951 ord
= get_imsm_ord_tbl_ent(dev
, i
, 0);
952 printf("%s", ord
& IMSM_ORD_REBUILD
? "_" : "U");
957 for (i
= 0; i
< map2
->num_members
; i
++) {
958 ord
= get_imsm_ord_tbl_ent(dev
, i
, 1);
959 printf("%s", ord
& IMSM_ORD_REBUILD
? "_" : "U");
964 printf(" Failed disk : ");
965 if (map
->failed_disk_num
== 0xff)
968 printf("%i", map
->failed_disk_num
);
970 slot
= get_imsm_disk_slot(map
, disk_idx
);
972 ord
= get_imsm_ord_tbl_ent(dev
, slot
, -1);
973 printf(" This Slot : %d%s\n", slot
,
974 ord
& IMSM_ORD_REBUILD
? " (out-of-sync)" : "");
976 printf(" This Slot : ?\n");
977 sz
= __le32_to_cpu(dev
->size_high
);
979 sz
+= __le32_to_cpu(dev
->size_low
);
980 printf(" Array Size : %llu%s\n", (unsigned long long)sz
,
981 human_size(sz
* 512));
982 sz
= __le32_to_cpu(map
->blocks_per_member
);
983 printf(" Per Dev Size : %llu%s\n", (unsigned long long)sz
,
984 human_size(sz
* 512));
985 printf(" Sector Offset : %u\n",
986 __le32_to_cpu(map
->pba_of_lba0
));
987 printf(" Num Stripes : %u\n",
988 __le32_to_cpu(map
->num_data_stripes
));
989 printf(" Chunk Size : %u KiB",
990 __le16_to_cpu(map
->blocks_per_strip
) / 2);
992 printf(" <-- %u KiB",
993 __le16_to_cpu(map2
->blocks_per_strip
) / 2);
995 printf(" Reserved : %d\n", __le32_to_cpu(dev
->reserved_blocks
));
996 printf(" Migrate State : ");
997 if (dev
->vol
.migr_state
) {
998 if (migr_type(dev
) == MIGR_INIT
)
999 printf("initialize\n");
1000 else if (migr_type(dev
) == MIGR_REBUILD
)
1001 printf("rebuild\n");
1002 else if (migr_type(dev
) == MIGR_VERIFY
)
1004 else if (migr_type(dev
) == MIGR_GEN_MIGR
)
1005 printf("general migration\n");
1006 else if (migr_type(dev
) == MIGR_STATE_CHANGE
)
1007 printf("state change\n");
1008 else if (migr_type(dev
) == MIGR_REPAIR
)
1011 printf("<unknown:%d>\n", migr_type(dev
));
1014 printf(" Map State : %s", map_state_str
[map
->map_state
]);
1015 if (dev
->vol
.migr_state
) {
1016 struct imsm_map
*map
= get_imsm_map(dev
, 1);
1018 printf(" <-- %s", map_state_str
[map
->map_state
]);
1019 printf("\n Checkpoint : %u (%llu)",
1020 __le32_to_cpu(dev
->vol
.curr_migr_unit
),
1021 (unsigned long long)blocks_per_migr_unit(super
, dev
));
1024 printf(" Dirty State : %s\n", dev
->vol
.dirty
? "dirty" : "clean");
1027 static void print_imsm_disk(struct imsm_super
*mpb
, int index
, __u32 reserved
)
1029 struct imsm_disk
*disk
= __get_imsm_disk(mpb
, index
);
1030 char str
[MAX_RAID_SERIAL_LEN
+ 1];
1033 if (index
< 0 || !disk
)
1037 snprintf(str
, MAX_RAID_SERIAL_LEN
+ 1, "%s", disk
->serial
);
1038 printf(" Disk%02d Serial : %s\n", index
, str
);
1039 printf(" State :%s%s%s\n", is_spare(disk
) ? " spare" : "",
1040 is_configured(disk
) ? " active" : "",
1041 is_failed(disk
) ? " failed" : "");
1042 printf(" Id : %08x\n", __le32_to_cpu(disk
->scsi_id
));
1043 sz
= __le32_to_cpu(disk
->total_blocks
) - reserved
;
1044 printf(" Usable Size : %llu%s\n", (unsigned long long)sz
,
1045 human_size(sz
* 512));
1048 static int is_gen_migration(struct imsm_dev
*dev
);
1050 void examine_migr_rec_imsm(struct intel_super
*super
)
1052 struct migr_record
*migr_rec
= super
->migr_rec
;
1053 struct imsm_super
*mpb
= super
->anchor
;
1056 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
1057 struct imsm_dev
*dev
= __get_imsm_dev(mpb
, i
);
1058 if (is_gen_migration(dev
) == 0)
1061 printf("\nMigration Record Information:");
1062 if (super
->disks
->index
> 1) {
1063 printf(" Empty\n ");
1064 printf("Examine one of first two disks in array\n");
1067 printf("\n Status : ");
1068 if (__le32_to_cpu(migr_rec
->rec_status
) == UNIT_SRC_NORMAL
)
1071 printf("Contains Data\n");
1072 printf(" Current Unit : %u\n",
1073 __le32_to_cpu(migr_rec
->curr_migr_unit
));
1074 printf(" Family : %u\n",
1075 __le32_to_cpu(migr_rec
->family_num
));
1076 printf(" Ascending : %u\n",
1077 __le32_to_cpu(migr_rec
->ascending_migr
));
1078 printf(" Blocks Per Unit : %u\n",
1079 __le32_to_cpu(migr_rec
->blocks_per_unit
));
1080 printf(" Dest. Depth Per Unit : %u\n",
1081 __le32_to_cpu(migr_rec
->dest_depth_per_unit
));
1082 printf(" Checkpoint Area pba : %u\n",
1083 __le32_to_cpu(migr_rec
->ckpt_area_pba
));
1084 printf(" First member lba : %u\n",
1085 __le32_to_cpu(migr_rec
->dest_1st_member_lba
));
1086 printf(" Total Number of Units : %u\n",
1087 __le32_to_cpu(migr_rec
->num_migr_units
));
1088 printf(" Size of volume : %u\n",
1089 __le32_to_cpu(migr_rec
->post_migr_vol_cap
));
1090 printf(" Expansion space for LBA64 : %u\n",
1091 __le32_to_cpu(migr_rec
->post_migr_vol_cap_hi
));
1092 printf(" Record was read from : %u\n",
1093 __le32_to_cpu(migr_rec
->ckpt_read_disk_num
));
1099 static void getinfo_super_imsm(struct supertype
*st
, struct mdinfo
*info
, char *map
);
1101 static void examine_super_imsm(struct supertype
*st
, char *homehost
)
1103 struct intel_super
*super
= st
->sb
;
1104 struct imsm_super
*mpb
= super
->anchor
;
1105 char str
[MAX_SIGNATURE_LENGTH
];
1110 __u32 reserved
= imsm_reserved_sectors(super
, super
->disks
);
1113 snprintf(str
, MPB_SIG_LEN
, "%s", mpb
->sig
);
1114 printf(" Magic : %s\n", str
);
1115 snprintf(str
, strlen(MPB_VERSION_RAID0
), "%s", get_imsm_version(mpb
));
1116 printf(" Version : %s\n", get_imsm_version(mpb
));
1117 printf(" Orig Family : %08x\n", __le32_to_cpu(mpb
->orig_family_num
));
1118 printf(" Family : %08x\n", __le32_to_cpu(mpb
->family_num
));
1119 printf(" Generation : %08x\n", __le32_to_cpu(mpb
->generation_num
));
1120 getinfo_super_imsm(st
, &info
, NULL
);
1121 fname_from_uuid(st
, &info
, nbuf
, ':');
1122 printf(" UUID : %s\n", nbuf
+ 5);
1123 sum
= __le32_to_cpu(mpb
->check_sum
);
1124 printf(" Checksum : %08x %s\n", sum
,
1125 __gen_imsm_checksum(mpb
) == sum
? "correct" : "incorrect");
1126 printf(" MPB Sectors : %d\n", mpb_sectors(mpb
));
1127 printf(" Disks : %d\n", mpb
->num_disks
);
1128 printf(" RAID Devices : %d\n", mpb
->num_raid_devs
);
1129 print_imsm_disk(mpb
, super
->disks
->index
, reserved
);
1130 if (super
->bbm_log
) {
1131 struct bbm_log
*log
= super
->bbm_log
;
1134 printf("Bad Block Management Log:\n");
1135 printf(" Log Size : %d\n", __le32_to_cpu(mpb
->bbm_log_size
));
1136 printf(" Signature : %x\n", __le32_to_cpu(log
->signature
));
1137 printf(" Entry Count : %d\n", __le32_to_cpu(log
->entry_count
));
1138 printf(" Spare Blocks : %d\n", __le32_to_cpu(log
->reserved_spare_block_count
));
1139 printf(" First Spare : %llx\n",
1140 (unsigned long long) __le64_to_cpu(log
->first_spare_lba
));
1142 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
1144 struct imsm_dev
*dev
= __get_imsm_dev(mpb
, i
);
1146 super
->current_vol
= i
;
1147 getinfo_super_imsm(st
, &info
, NULL
);
1148 fname_from_uuid(st
, &info
, nbuf
, ':');
1149 print_imsm_dev(super
, dev
, nbuf
+ 5, super
->disks
->index
);
1151 for (i
= 0; i
< mpb
->num_disks
; i
++) {
1152 if (i
== super
->disks
->index
)
1154 print_imsm_disk(mpb
, i
, reserved
);
1156 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
1157 struct imsm_disk
*disk
;
1158 char str
[MAX_RAID_SERIAL_LEN
+ 1];
1166 snprintf(str
, MAX_RAID_SERIAL_LEN
+ 1, "%s", disk
->serial
);
1167 printf(" Disk Serial : %s\n", str
);
1168 printf(" State :%s%s%s\n", is_spare(disk
) ? " spare" : "",
1169 is_configured(disk
) ? " active" : "",
1170 is_failed(disk
) ? " failed" : "");
1171 printf(" Id : %08x\n", __le32_to_cpu(disk
->scsi_id
));
1172 sz
= __le32_to_cpu(disk
->total_blocks
) - reserved
;
1173 printf(" Usable Size : %llu%s\n", (unsigned long long)sz
,
1174 human_size(sz
* 512));
1177 examine_migr_rec_imsm(super
);
1180 static void brief_examine_super_imsm(struct supertype
*st
, int verbose
)
1182 /* We just write a generic IMSM ARRAY entry */
1185 struct intel_super
*super
= st
->sb
;
1187 if (!super
->anchor
->num_raid_devs
) {
1188 printf("ARRAY metadata=imsm\n");
1192 getinfo_super_imsm(st
, &info
, NULL
);
1193 fname_from_uuid(st
, &info
, nbuf
, ':');
1194 printf("ARRAY metadata=imsm UUID=%s\n", nbuf
+ 5);
1197 static void brief_examine_subarrays_imsm(struct supertype
*st
, int verbose
)
1199 /* We just write a generic IMSM ARRAY entry */
1203 struct intel_super
*super
= st
->sb
;
1206 if (!super
->anchor
->num_raid_devs
)
1209 getinfo_super_imsm(st
, &info
, NULL
);
1210 fname_from_uuid(st
, &info
, nbuf
, ':');
1211 for (i
= 0; i
< super
->anchor
->num_raid_devs
; i
++) {
1212 struct imsm_dev
*dev
= get_imsm_dev(super
, i
);
1214 super
->current_vol
= i
;
1215 getinfo_super_imsm(st
, &info
, NULL
);
1216 fname_from_uuid(st
, &info
, nbuf1
, ':');
1217 printf("ARRAY /dev/md/%.16s container=%s member=%d UUID=%s\n",
1218 dev
->volume
, nbuf
+ 5, i
, nbuf1
+ 5);
1222 static void export_examine_super_imsm(struct supertype
*st
)
1224 struct intel_super
*super
= st
->sb
;
1225 struct imsm_super
*mpb
= super
->anchor
;
1229 getinfo_super_imsm(st
, &info
, NULL
);
1230 fname_from_uuid(st
, &info
, nbuf
, ':');
1231 printf("MD_METADATA=imsm\n");
1232 printf("MD_LEVEL=container\n");
1233 printf("MD_UUID=%s\n", nbuf
+5);
1234 printf("MD_DEVICES=%u\n", mpb
->num_disks
);
1237 static void detail_super_imsm(struct supertype
*st
, char *homehost
)
1242 getinfo_super_imsm(st
, &info
, NULL
);
1243 fname_from_uuid(st
, &info
, nbuf
, ':');
1244 printf("\n UUID : %s\n", nbuf
+ 5);
1247 static void brief_detail_super_imsm(struct supertype
*st
)
1251 getinfo_super_imsm(st
, &info
, NULL
);
1252 fname_from_uuid(st
, &info
, nbuf
, ':');
1253 printf(" UUID=%s", nbuf
+ 5);
1256 static int imsm_read_serial(int fd
, char *devname
, __u8
*serial
);
1257 static void fd2devname(int fd
, char *name
);
1259 static int ahci_enumerate_ports(const char *hba_path
, int port_count
, int host_base
, int verbose
)
1261 /* dump an unsorted list of devices attached to AHCI Intel storage
1262 * controller, as well as non-connected ports
1264 int hba_len
= strlen(hba_path
) + 1;
1269 unsigned long port_mask
= (1 << port_count
) - 1;
1271 if (port_count
> (int)sizeof(port_mask
) * 8) {
1273 fprintf(stderr
, Name
": port_count %d out of range\n", port_count
);
1277 /* scroll through /sys/dev/block looking for devices attached to
1280 dir
= opendir("/sys/dev/block");
1281 for (ent
= dir
? readdir(dir
) : NULL
; ent
; ent
= readdir(dir
)) {
1292 if (sscanf(ent
->d_name
, "%d:%d", &major
, &minor
) != 2)
1294 path
= devt_to_devpath(makedev(major
, minor
));
1297 if (!path_attached_to_hba(path
, hba_path
)) {
1303 /* retrieve the scsi device type */
1304 if (asprintf(&device
, "/sys/dev/block/%d:%d/device/xxxxxxx", major
, minor
) < 0) {
1306 fprintf(stderr
, Name
": failed to allocate 'device'\n");
1310 sprintf(device
, "/sys/dev/block/%d:%d/device/type", major
, minor
);
1311 if (load_sys(device
, buf
) != 0) {
1313 fprintf(stderr
, Name
": failed to read device type for %s\n",
1319 type
= strtoul(buf
, NULL
, 10);
1321 /* if it's not a disk print the vendor and model */
1322 if (!(type
== 0 || type
== 7 || type
== 14)) {
1325 sprintf(device
, "/sys/dev/block/%d:%d/device/vendor", major
, minor
);
1326 if (load_sys(device
, buf
) == 0) {
1327 strncpy(vendor
, buf
, sizeof(vendor
));
1328 vendor
[sizeof(vendor
) - 1] = '\0';
1329 c
= (char *) &vendor
[sizeof(vendor
) - 1];
1330 while (isspace(*c
) || *c
== '\0')
1334 sprintf(device
, "/sys/dev/block/%d:%d/device/model", major
, minor
);
1335 if (load_sys(device
, buf
) == 0) {
1336 strncpy(model
, buf
, sizeof(model
));
1337 model
[sizeof(model
) - 1] = '\0';
1338 c
= (char *) &model
[sizeof(model
) - 1];
1339 while (isspace(*c
) || *c
== '\0')
1343 if (vendor
[0] && model
[0])
1344 sprintf(buf
, "%.64s %.64s", vendor
, model
);
1346 switch (type
) { /* numbers from hald/linux/device.c */
1347 case 1: sprintf(buf
, "tape"); break;
1348 case 2: sprintf(buf
, "printer"); break;
1349 case 3: sprintf(buf
, "processor"); break;
1351 case 5: sprintf(buf
, "cdrom"); break;
1352 case 6: sprintf(buf
, "scanner"); break;
1353 case 8: sprintf(buf
, "media_changer"); break;
1354 case 9: sprintf(buf
, "comm"); break;
1355 case 12: sprintf(buf
, "raid"); break;
1356 default: sprintf(buf
, "unknown");
1362 /* chop device path to 'host%d' and calculate the port number */
1363 c
= strchr(&path
[hba_len
], '/');
1366 fprintf(stderr
, Name
": %s - invalid path name\n", path
+ hba_len
);
1371 if (sscanf(&path
[hba_len
], "host%d", &port
) == 1)
1375 *c
= '/'; /* repair the full string */
1376 fprintf(stderr
, Name
": failed to determine port number for %s\n",
1383 /* mark this port as used */
1384 port_mask
&= ~(1 << port
);
1386 /* print out the device information */
1388 printf(" Port%d : - non-disk device (%s) -\n", port
, buf
);
1392 fd
= dev_open(ent
->d_name
, O_RDONLY
);
1394 printf(" Port%d : - disk info unavailable -\n", port
);
1396 fd2devname(fd
, buf
);
1397 printf(" Port%d : %s", port
, buf
);
1398 if (imsm_read_serial(fd
, NULL
, (__u8
*) buf
) == 0)
1399 printf(" (%s)\n", buf
);
1414 for (i
= 0; i
< port_count
; i
++)
1415 if (port_mask
& (1 << i
))
1416 printf(" Port%d : - no device attached -\n", i
);
1424 static void print_found_intel_controllers(struct sys_dev
*elem
)
1426 for (; elem
; elem
= elem
->next
) {
1427 fprintf(stderr
, Name
": found Intel(R) ");
1428 if (elem
->type
== SYS_DEV_SATA
)
1429 fprintf(stderr
, "SATA ");
1430 else if (elem
->type
== SYS_DEV_SAS
)
1431 fprintf(stderr
, "SAS ");
1432 fprintf(stderr
, "RAID controller");
1434 fprintf(stderr
, " at %s", elem
->pci_id
);
1435 fprintf(stderr
, ".\n");
1440 static int ahci_get_port_count(const char *hba_path
, int *port_count
)
1447 if ((dir
= opendir(hba_path
)) == NULL
)
1450 for (ent
= readdir(dir
); ent
; ent
= readdir(dir
)) {
1453 if (sscanf(ent
->d_name
, "host%d", &host
) != 1)
1455 if (*port_count
== 0)
1457 else if (host
< host_base
)
1460 if (host
+ 1 > *port_count
+ host_base
)
1461 *port_count
= host
+ 1 - host_base
;
1467 static void print_imsm_capability(const struct imsm_orom
*orom
)
1469 printf(" Platform : Intel(R) Matrix Storage Manager\n");
1470 printf(" Version : %d.%d.%d.%d\n", orom
->major_ver
, orom
->minor_ver
,
1471 orom
->hotfix_ver
, orom
->build
);
1472 printf(" RAID Levels :%s%s%s%s%s\n",
1473 imsm_orom_has_raid0(orom
) ? " raid0" : "",
1474 imsm_orom_has_raid1(orom
) ? " raid1" : "",
1475 imsm_orom_has_raid1e(orom
) ? " raid1e" : "",
1476 imsm_orom_has_raid10(orom
) ? " raid10" : "",
1477 imsm_orom_has_raid5(orom
) ? " raid5" : "");
1478 printf(" Chunk Sizes :%s%s%s%s%s%s%s%s%s%s%s%s%s%s%s%s\n",
1479 imsm_orom_has_chunk(orom
, 2) ? " 2k" : "",
1480 imsm_orom_has_chunk(orom
, 4) ? " 4k" : "",
1481 imsm_orom_has_chunk(orom
, 8) ? " 8k" : "",
1482 imsm_orom_has_chunk(orom
, 16) ? " 16k" : "",
1483 imsm_orom_has_chunk(orom
, 32) ? " 32k" : "",
1484 imsm_orom_has_chunk(orom
, 64) ? " 64k" : "",
1485 imsm_orom_has_chunk(orom
, 128) ? " 128k" : "",
1486 imsm_orom_has_chunk(orom
, 256) ? " 256k" : "",
1487 imsm_orom_has_chunk(orom
, 512) ? " 512k" : "",
1488 imsm_orom_has_chunk(orom
, 1024*1) ? " 1M" : "",
1489 imsm_orom_has_chunk(orom
, 1024*2) ? " 2M" : "",
1490 imsm_orom_has_chunk(orom
, 1024*4) ? " 4M" : "",
1491 imsm_orom_has_chunk(orom
, 1024*8) ? " 8M" : "",
1492 imsm_orom_has_chunk(orom
, 1024*16) ? " 16M" : "",
1493 imsm_orom_has_chunk(orom
, 1024*32) ? " 32M" : "",
1494 imsm_orom_has_chunk(orom
, 1024*64) ? " 64M" : "");
1495 printf(" Max Disks : %d\n", orom
->tds
);
1496 printf(" Max Volumes : %d\n", orom
->vpa
);
1500 static int detail_platform_imsm(int verbose
, int enumerate_only
)
1502 /* There are two components to imsm platform support, the ahci SATA
1503 * controller and the option-rom. To find the SATA controller we
1504 * simply look in /sys/bus/pci/drivers/ahci to see if an ahci
1505 * controller with the Intel vendor id is present. This approach
1506 * allows mdadm to leverage the kernel's ahci detection logic, with the
1507 * caveat that if ahci.ko is not loaded mdadm will not be able to
1508 * detect platform raid capabilities. The option-rom resides in a
1509 * platform "Adapter ROM". We scan for its signature to retrieve the
1510 * platform capabilities. If raid support is disabled in the BIOS the
1511 * option-rom capability structure will not be available.
1513 const struct imsm_orom
*orom
;
1514 struct sys_dev
*list
, *hba
;
1519 if (enumerate_only
) {
1520 if (check_env("IMSM_NO_PLATFORM"))
1522 list
= find_intel_devices();
1525 for (hba
= list
; hba
; hba
= hba
->next
) {
1526 orom
= find_imsm_capability(hba
->type
);
1532 free_sys_dev(&list
);
1536 list
= find_intel_devices();
1539 fprintf(stderr
, Name
": no active Intel(R) RAID "
1540 "controller found.\n");
1541 free_sys_dev(&list
);
1544 print_found_intel_controllers(list
);
1546 for (hba
= list
; hba
; hba
= hba
->next
) {
1547 orom
= find_imsm_capability(hba
->type
);
1549 fprintf(stderr
, Name
": imsm capabilities not found for controller: %s (type %s)\n",
1550 hba
->path
, get_sys_dev_type(hba
->type
));
1552 print_imsm_capability(orom
);
1555 for (hba
= list
; hba
; hba
= hba
->next
) {
1556 printf(" I/O Controller : %s (%s)\n",
1557 hba
->path
, get_sys_dev_type(hba
->type
));
1559 if (hba
->type
== SYS_DEV_SATA
) {
1560 host_base
= ahci_get_port_count(hba
->path
, &port_count
);
1561 if (ahci_enumerate_ports(hba
->path
, port_count
, host_base
, verbose
)) {
1563 fprintf(stderr
, Name
": failed to enumerate "
1564 "ports on SATA controller at %s.", hba
->pci_id
);
1570 free_sys_dev(&list
);
1575 static int match_home_imsm(struct supertype
*st
, char *homehost
)
1577 /* the imsm metadata format does not specify any host
1578 * identification information. We return -1 since we can never
1579 * confirm nor deny whether a given array is "meant" for this
1580 * host. We rely on compare_super and the 'family_num' fields to
1581 * exclude member disks that do not belong, and we rely on
1582 * mdadm.conf to specify the arrays that should be assembled.
1583 * Auto-assembly may still pick up "foreign" arrays.
1589 static void uuid_from_super_imsm(struct supertype
*st
, int uuid
[4])
1591 /* The uuid returned here is used for:
1592 * uuid to put into bitmap file (Create, Grow)
1593 * uuid for backup header when saving critical section (Grow)
1594 * comparing uuids when re-adding a device into an array
1595 * In these cases the uuid required is that of the data-array,
1596 * not the device-set.
1597 * uuid to recognise same set when adding a missing device back
1598 * to an array. This is a uuid for the device-set.
1600 * For each of these we can make do with a truncated
1601 * or hashed uuid rather than the original, as long as
1603 * In each case the uuid required is that of the data-array,
1604 * not the device-set.
1606 /* imsm does not track uuid's so we synthesis one using sha1 on
1607 * - The signature (Which is constant for all imsm array, but no matter)
1608 * - the orig_family_num of the container
1609 * - the index number of the volume
1610 * - the 'serial' number of the volume.
1611 * Hopefully these are all constant.
1613 struct intel_super
*super
= st
->sb
;
1616 struct sha1_ctx ctx
;
1617 struct imsm_dev
*dev
= NULL
;
1620 /* some mdadm versions failed to set ->orig_family_num, in which
1621 * case fall back to ->family_num. orig_family_num will be
1622 * fixed up with the first metadata update.
1624 family_num
= super
->anchor
->orig_family_num
;
1625 if (family_num
== 0)
1626 family_num
= super
->anchor
->family_num
;
1627 sha1_init_ctx(&ctx
);
1628 sha1_process_bytes(super
->anchor
->sig
, MPB_SIG_LEN
, &ctx
);
1629 sha1_process_bytes(&family_num
, sizeof(__u32
), &ctx
);
1630 if (super
->current_vol
>= 0)
1631 dev
= get_imsm_dev(super
, super
->current_vol
);
1633 __u32 vol
= super
->current_vol
;
1634 sha1_process_bytes(&vol
, sizeof(vol
), &ctx
);
1635 sha1_process_bytes(dev
->volume
, MAX_RAID_SERIAL_LEN
, &ctx
);
1637 sha1_finish_ctx(&ctx
, buf
);
1638 memcpy(uuid
, buf
, 4*4);
1643 get_imsm_numerical_version(struct imsm_super
*mpb
, int *m
, int *p
)
1645 __u8
*v
= get_imsm_version(mpb
);
1646 __u8
*end
= mpb
->sig
+ MAX_SIGNATURE_LENGTH
;
1647 char major
[] = { 0, 0, 0 };
1648 char minor
[] = { 0 ,0, 0 };
1649 char patch
[] = { 0, 0, 0 };
1650 char *ver_parse
[] = { major
, minor
, patch
};
1654 while (*v
!= '\0' && v
< end
) {
1655 if (*v
!= '.' && j
< 2)
1656 ver_parse
[i
][j
++] = *v
;
1664 *m
= strtol(minor
, NULL
, 0);
1665 *p
= strtol(patch
, NULL
, 0);
1669 static __u32
migr_strip_blocks_resync(struct imsm_dev
*dev
)
1671 /* migr_strip_size when repairing or initializing parity */
1672 struct imsm_map
*map
= get_imsm_map(dev
, 0);
1673 __u32 chunk
= __le32_to_cpu(map
->blocks_per_strip
);
1675 switch (get_imsm_raid_level(map
)) {
1680 return 128*1024 >> 9;
1684 static __u32
migr_strip_blocks_rebuild(struct imsm_dev
*dev
)
1686 /* migr_strip_size when rebuilding a degraded disk, no idea why
1687 * this is different than migr_strip_size_resync(), but it's good
1690 struct imsm_map
*map
= get_imsm_map(dev
, 1);
1691 __u32 chunk
= __le32_to_cpu(map
->blocks_per_strip
);
1693 switch (get_imsm_raid_level(map
)) {
1696 if (map
->num_members
% map
->num_domains
== 0)
1697 return 128*1024 >> 9;
1701 return max((__u32
) 64*1024 >> 9, chunk
);
1703 return 128*1024 >> 9;
1707 static __u32
num_stripes_per_unit_resync(struct imsm_dev
*dev
)
1709 struct imsm_map
*lo
= get_imsm_map(dev
, 0);
1710 struct imsm_map
*hi
= get_imsm_map(dev
, 1);
1711 __u32 lo_chunk
= __le32_to_cpu(lo
->blocks_per_strip
);
1712 __u32 hi_chunk
= __le32_to_cpu(hi
->blocks_per_strip
);
1714 return max((__u32
) 1, hi_chunk
/ lo_chunk
);
1717 static __u32
num_stripes_per_unit_rebuild(struct imsm_dev
*dev
)
1719 struct imsm_map
*lo
= get_imsm_map(dev
, 0);
1720 int level
= get_imsm_raid_level(lo
);
1722 if (level
== 1 || level
== 10) {
1723 struct imsm_map
*hi
= get_imsm_map(dev
, 1);
1725 return hi
->num_domains
;
1727 return num_stripes_per_unit_resync(dev
);
1730 static __u8
imsm_num_data_members(struct imsm_dev
*dev
, int second_map
)
1732 /* named 'imsm_' because raid0, raid1 and raid10
1733 * counter-intuitively have the same number of data disks
1735 struct imsm_map
*map
= get_imsm_map(dev
, second_map
);
1737 switch (get_imsm_raid_level(map
)) {
1741 return map
->num_members
;
1743 return map
->num_members
- 1;
1745 dprintf("%s: unsupported raid level\n", __func__
);
1750 static __u32
parity_segment_depth(struct imsm_dev
*dev
)
1752 struct imsm_map
*map
= get_imsm_map(dev
, 0);
1753 __u32 chunk
= __le32_to_cpu(map
->blocks_per_strip
);
1755 switch(get_imsm_raid_level(map
)) {
1758 return chunk
* map
->num_domains
;
1760 return chunk
* map
->num_members
;
1766 static __u32
map_migr_block(struct imsm_dev
*dev
, __u32 block
)
1768 struct imsm_map
*map
= get_imsm_map(dev
, 1);
1769 __u32 chunk
= __le32_to_cpu(map
->blocks_per_strip
);
1770 __u32 strip
= block
/ chunk
;
1772 switch (get_imsm_raid_level(map
)) {
1775 __u32 vol_strip
= (strip
* map
->num_domains
) + 1;
1776 __u32 vol_stripe
= vol_strip
/ map
->num_members
;
1778 return vol_stripe
* chunk
+ block
% chunk
;
1780 __u32 stripe
= strip
/ (map
->num_members
- 1);
1782 return stripe
* chunk
+ block
% chunk
;
1789 static __u64
blocks_per_migr_unit(struct intel_super
*super
,
1790 struct imsm_dev
*dev
)
1792 /* calculate the conversion factor between per member 'blocks'
1793 * (md/{resync,rebuild}_start) and imsm migration units, return
1794 * 0 for the 'not migrating' and 'unsupported migration' cases
1796 if (!dev
->vol
.migr_state
)
1799 switch (migr_type(dev
)) {
1800 case MIGR_GEN_MIGR
: {
1801 struct migr_record
*migr_rec
= super
->migr_rec
;
1802 return __le32_to_cpu(migr_rec
->blocks_per_unit
);
1807 struct imsm_map
*map
= get_imsm_map(dev
, 0);
1808 __u32 stripes_per_unit
;
1809 __u32 blocks_per_unit
;
1818 /* yes, this is really the translation of migr_units to
1819 * per-member blocks in the 'resync' case
1821 stripes_per_unit
= num_stripes_per_unit_resync(dev
);
1822 migr_chunk
= migr_strip_blocks_resync(dev
);
1823 disks
= imsm_num_data_members(dev
, 0);
1824 blocks_per_unit
= stripes_per_unit
* migr_chunk
* disks
;
1825 stripe
= __le32_to_cpu(map
->blocks_per_strip
) * disks
;
1826 segment
= blocks_per_unit
/ stripe
;
1827 block_rel
= blocks_per_unit
- segment
* stripe
;
1828 parity_depth
= parity_segment_depth(dev
);
1829 block_map
= map_migr_block(dev
, block_rel
);
1830 return block_map
+ parity_depth
* segment
;
1832 case MIGR_REBUILD
: {
1833 __u32 stripes_per_unit
;
1836 stripes_per_unit
= num_stripes_per_unit_rebuild(dev
);
1837 migr_chunk
= migr_strip_blocks_rebuild(dev
);
1838 return migr_chunk
* stripes_per_unit
;
1840 case MIGR_STATE_CHANGE
:
1846 static int imsm_level_to_layout(int level
)
1854 return ALGORITHM_LEFT_ASYMMETRIC
;
1861 /*******************************************************************************
1862 * Function: read_imsm_migr_rec
1863 * Description: Function reads imsm migration record from last sector of disk
1865 * fd : disk descriptor
1866 * super : metadata info
1870 ******************************************************************************/
1871 static int read_imsm_migr_rec(int fd
, struct intel_super
*super
)
1874 unsigned long long dsize
;
1876 get_dev_size(fd
, NULL
, &dsize
);
1877 if (lseek64(fd
, dsize
- 512, SEEK_SET
) < 0) {
1879 Name
": Cannot seek to anchor block: %s\n",
1883 if (read(fd
, super
->migr_rec_buf
, 512) != 512) {
1885 Name
": Cannot read migr record block: %s\n",
1895 /*******************************************************************************
1896 * Function: load_imsm_migr_rec
1897 * Description: Function reads imsm migration record (it is stored at the last
1900 * super : imsm internal array info
1901 * info : general array info
1905 ******************************************************************************/
1906 static int load_imsm_migr_rec(struct intel_super
*super
, struct mdinfo
*info
)
1909 struct dl
*dl
= NULL
;
1915 for (sd
= info
->devs
; sd
; sd
= sd
->next
) {
1916 /* read only from one of the first two slots */
1917 if ((sd
->disk
.raid_disk
> 1) ||
1918 (sd
->disk
.raid_disk
< 0))
1920 sprintf(nm
, "%d:%d", sd
->disk
.major
, sd
->disk
.minor
);
1921 fd
= dev_open(nm
, O_RDONLY
);
1927 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
1928 /* read only from one of the first two slots */
1931 sprintf(nm
, "%d:%d", dl
->major
, dl
->minor
);
1932 fd
= dev_open(nm
, O_RDONLY
);
1939 retval
= read_imsm_migr_rec(fd
, super
);
1947 /*******************************************************************************
1948 * function: imsm_create_metadata_checkpoint_update
1949 * Description: It creates update for checkpoint change.
1951 * super : imsm internal array info
1952 * u : pointer to prepared update
1955 * If length is equal to 0, input pointer u contains no update
1956 ******************************************************************************/
1957 static int imsm_create_metadata_checkpoint_update(
1958 struct intel_super
*super
,
1959 struct imsm_update_general_migration_checkpoint
**u
)
1962 int update_memory_size
= 0;
1964 dprintf("imsm_create_metadata_checkpoint_update(enter)\n");
1970 /* size of all update data without anchor */
1971 update_memory_size
=
1972 sizeof(struct imsm_update_general_migration_checkpoint
);
1974 *u
= calloc(1, update_memory_size
);
1976 dprintf("error: cannot get memory for "
1977 "imsm_create_metadata_checkpoint_update update\n");
1980 (*u
)->type
= update_general_migration_checkpoint
;
1981 (*u
)->curr_migr_unit
= __le32_to_cpu(super
->migr_rec
->curr_migr_unit
);
1982 dprintf("imsm_create_metadata_checkpoint_update: prepared for %u\n",
1983 (*u
)->curr_migr_unit
);
1985 return update_memory_size
;
1989 static void imsm_update_metadata_locally(struct supertype
*st
,
1990 void *buf
, int len
);
1992 /*******************************************************************************
1993 * Function: write_imsm_migr_rec
1994 * Description: Function writes imsm migration record
1995 * (at the last sector of disk)
1997 * super : imsm internal array info
2001 ******************************************************************************/
2002 static int write_imsm_migr_rec(struct supertype
*st
)
2004 struct intel_super
*super
= st
->sb
;
2005 unsigned long long dsize
;
2011 struct imsm_update_general_migration_checkpoint
*u
;
2013 for (sd
= super
->disks
; sd
; sd
= sd
->next
) {
2014 /* write to 2 first slots only */
2015 if ((sd
->index
< 0) || (sd
->index
> 1))
2017 sprintf(nm
, "%d:%d", sd
->major
, sd
->minor
);
2018 fd
= dev_open(nm
, O_RDWR
);
2021 get_dev_size(fd
, NULL
, &dsize
);
2022 if (lseek64(fd
, dsize
- 512, SEEK_SET
) < 0) {
2024 Name
": Cannot seek to anchor block: %s\n",
2028 if (write(fd
, super
->migr_rec_buf
, 512) != 512) {
2030 Name
": Cannot write migr record block: %s\n",
2037 /* update checkpoint information in metadata */
2038 len
= imsm_create_metadata_checkpoint_update(super
, &u
);
2041 dprintf("imsm: Cannot prepare update\n");
2044 /* update metadata locally */
2045 imsm_update_metadata_locally(st
, u
, len
);
2046 /* and possibly remotely */
2047 if (st
->update_tail
) {
2048 append_metadata_update(st
, u
, len
);
2049 /* during reshape we do all work inside metadata handler
2050 * manage_reshape(), so metadata update has to be triggered
2053 flush_metadata_updates(st
);
2054 st
->update_tail
= &st
->updates
;
2065 static void getinfo_super_imsm_volume(struct supertype
*st
, struct mdinfo
*info
, char *dmap
)
2067 struct intel_super
*super
= st
->sb
;
2068 struct migr_record
*migr_rec
= super
->migr_rec
;
2069 struct imsm_dev
*dev
= get_imsm_dev(super
, super
->current_vol
);
2070 struct imsm_map
*map
= get_imsm_map(dev
, 0);
2071 struct imsm_map
*prev_map
= get_imsm_map(dev
, 1);
2072 struct imsm_map
*map_to_analyse
= map
;
2075 unsigned int component_size_alligment
;
2076 int map_disks
= info
->array
.raid_disks
;
2078 memset(info
, 0, sizeof(*info
));
2080 map_to_analyse
= prev_map
;
2084 info
->container_member
= super
->current_vol
;
2085 info
->array
.raid_disks
= map
->num_members
;
2086 info
->array
.level
= get_imsm_raid_level(map_to_analyse
);
2087 info
->array
.layout
= imsm_level_to_layout(info
->array
.level
);
2088 info
->array
.md_minor
= -1;
2089 info
->array
.ctime
= 0;
2090 info
->array
.utime
= 0;
2091 info
->array
.chunk_size
=
2092 __le16_to_cpu(map_to_analyse
->blocks_per_strip
) << 9;
2093 info
->array
.state
= !dev
->vol
.dirty
;
2094 info
->custom_array_size
= __le32_to_cpu(dev
->size_high
);
2095 info
->custom_array_size
<<= 32;
2096 info
->custom_array_size
|= __le32_to_cpu(dev
->size_low
);
2097 if (prev_map
&& map
->map_state
== prev_map
->map_state
) {
2098 info
->reshape_active
= 1;
2099 info
->new_level
= get_imsm_raid_level(map
);
2100 info
->new_layout
= imsm_level_to_layout(info
->new_level
);
2101 info
->new_chunk
= __le16_to_cpu(map
->blocks_per_strip
) << 9;
2102 info
->delta_disks
= map
->num_members
- prev_map
->num_members
;
2103 if (info
->delta_disks
) {
2104 /* this needs to be applied to every array
2107 info
->reshape_active
= 2;
2109 /* We shape information that we give to md might have to be
2110 * modify to cope with md's requirement for reshaping arrays.
2111 * For example, when reshaping a RAID0, md requires it to be
2112 * presented as a degraded RAID4.
2113 * Also if a RAID0 is migrating to a RAID5 we need to specify
2114 * the array as already being RAID5, but the 'before' layout
2115 * is a RAID4-like layout.
2117 switch (info
->array
.level
) {
2119 switch(info
->new_level
) {
2121 /* conversion is happening as RAID4 */
2122 info
->array
.level
= 4;
2123 info
->array
.raid_disks
+= 1;
2126 /* conversion is happening as RAID5 */
2127 info
->array
.level
= 5;
2128 info
->array
.layout
= ALGORITHM_PARITY_N
;
2129 info
->array
.raid_disks
+= 1;
2130 info
->delta_disks
-= 1;
2133 /* FIXME error message */
2134 info
->array
.level
= UnSet
;
2140 info
->new_level
= UnSet
;
2141 info
->new_layout
= UnSet
;
2142 info
->new_chunk
= info
->array
.chunk_size
;
2143 info
->delta_disks
= 0;
2145 info
->disk
.major
= 0;
2146 info
->disk
.minor
= 0;
2148 info
->disk
.major
= dl
->major
;
2149 info
->disk
.minor
= dl
->minor
;
2152 info
->data_offset
= __le32_to_cpu(map_to_analyse
->pba_of_lba0
);
2153 info
->component_size
=
2154 __le32_to_cpu(map_to_analyse
->blocks_per_member
);
2156 /* check component size aligment
2158 component_size_alligment
=
2159 info
->component_size
% (info
->array
.chunk_size
/512);
2161 if (component_size_alligment
&&
2162 (info
->array
.level
!= 1) && (info
->array
.level
!= UnSet
)) {
2163 dprintf("imsm: reported component size alligned from %llu ",
2164 info
->component_size
);
2165 info
->component_size
-= component_size_alligment
;
2166 dprintf("to %llu (%i).\n",
2167 info
->component_size
, component_size_alligment
);
2170 memset(info
->uuid
, 0, sizeof(info
->uuid
));
2171 info
->recovery_start
= MaxSector
;
2173 info
->reshape_progress
= 0;
2174 info
->resync_start
= MaxSector
;
2175 if (map_to_analyse
->map_state
== IMSM_T_STATE_UNINITIALIZED
||
2177 info
->resync_start
= 0;
2179 if (dev
->vol
.migr_state
) {
2180 switch (migr_type(dev
)) {
2183 __u64 blocks_per_unit
= blocks_per_migr_unit(super
,
2185 __u64 units
= __le32_to_cpu(dev
->vol
.curr_migr_unit
);
2187 info
->resync_start
= blocks_per_unit
* units
;
2190 case MIGR_GEN_MIGR
: {
2191 __u64 blocks_per_unit
= blocks_per_migr_unit(super
,
2193 __u64 units
= __le32_to_cpu(migr_rec
->curr_migr_unit
);
2194 unsigned long long array_blocks
;
2197 info
->reshape_progress
= blocks_per_unit
* units
;
2199 dprintf("IMSM: General Migration checkpoint : %llu "
2200 "(%llu) -> read reshape progress : %llu\n",
2201 units
, blocks_per_unit
, info
->reshape_progress
);
2203 used_disks
= imsm_num_data_members(dev
, 1);
2204 if (used_disks
> 0) {
2205 array_blocks
= map
->blocks_per_member
*
2207 /* round array size down to closest MB
2209 info
->custom_array_size
= (array_blocks
2210 >> SECT_PER_MB_SHIFT
)
2211 << SECT_PER_MB_SHIFT
;
2215 /* we could emulate the checkpointing of
2216 * 'sync_action=check' migrations, but for now
2217 * we just immediately complete them
2220 /* this is handled by container_content_imsm() */
2221 case MIGR_STATE_CHANGE
:
2222 /* FIXME handle other migrations */
2224 /* we are not dirty, so... */
2225 info
->resync_start
= MaxSector
;
2229 strncpy(info
->name
, (char *) dev
->volume
, MAX_RAID_SERIAL_LEN
);
2230 info
->name
[MAX_RAID_SERIAL_LEN
] = 0;
2232 info
->array
.major_version
= -1;
2233 info
->array
.minor_version
= -2;
2234 devname
= devnum2devname(st
->container_dev
);
2235 *info
->text_version
= '\0';
2237 sprintf(info
->text_version
, "/%s/%d", devname
, info
->container_member
);
2239 info
->safe_mode_delay
= 4000; /* 4 secs like the Matrix driver */
2240 uuid_from_super_imsm(st
, info
->uuid
);
2244 for (i
=0; i
<map_disks
; i
++) {
2246 if (i
< info
->array
.raid_disks
) {
2247 struct imsm_disk
*dsk
;
2248 j
= get_imsm_disk_idx(dev
, i
, -1);
2249 dsk
= get_imsm_disk(super
, j
);
2250 if (dsk
&& (dsk
->status
& CONFIGURED_DISK
))
2257 static __u8
imsm_check_degraded(struct intel_super
*super
, struct imsm_dev
*dev
, int failed
);
2258 static int imsm_count_failed(struct intel_super
*super
, struct imsm_dev
*dev
);
2260 static struct imsm_disk
*get_imsm_missing(struct intel_super
*super
, __u8 index
)
2264 for (d
= super
->missing
; d
; d
= d
->next
)
2265 if (d
->index
== index
)
2270 static void getinfo_super_imsm(struct supertype
*st
, struct mdinfo
*info
, char *map
)
2272 struct intel_super
*super
= st
->sb
;
2273 struct imsm_disk
*disk
;
2274 int map_disks
= info
->array
.raid_disks
;
2275 int max_enough
= -1;
2277 struct imsm_super
*mpb
;
2279 if (super
->current_vol
>= 0) {
2280 getinfo_super_imsm_volume(st
, info
, map
);
2283 memset(info
, 0, sizeof(*info
));
2285 /* Set raid_disks to zero so that Assemble will always pull in valid
2288 info
->array
.raid_disks
= 0;
2289 info
->array
.level
= LEVEL_CONTAINER
;
2290 info
->array
.layout
= 0;
2291 info
->array
.md_minor
= -1;
2292 info
->array
.ctime
= 0; /* N/A for imsm */
2293 info
->array
.utime
= 0;
2294 info
->array
.chunk_size
= 0;
2296 info
->disk
.major
= 0;
2297 info
->disk
.minor
= 0;
2298 info
->disk
.raid_disk
= -1;
2299 info
->reshape_active
= 0;
2300 info
->array
.major_version
= -1;
2301 info
->array
.minor_version
= -2;
2302 strcpy(info
->text_version
, "imsm");
2303 info
->safe_mode_delay
= 0;
2304 info
->disk
.number
= -1;
2305 info
->disk
.state
= 0;
2307 info
->recovery_start
= MaxSector
;
2309 /* do we have the all the insync disks that we expect? */
2310 mpb
= super
->anchor
;
2312 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
2313 struct imsm_dev
*dev
= get_imsm_dev(super
, i
);
2314 int failed
, enough
, j
, missing
= 0;
2315 struct imsm_map
*map
;
2318 failed
= imsm_count_failed(super
, dev
);
2319 state
= imsm_check_degraded(super
, dev
, failed
);
2320 map
= get_imsm_map(dev
, dev
->vol
.migr_state
);
2322 /* any newly missing disks?
2323 * (catches single-degraded vs double-degraded)
2325 for (j
= 0; j
< map
->num_members
; j
++) {
2326 __u32 ord
= get_imsm_ord_tbl_ent(dev
, i
, -1);
2327 __u32 idx
= ord_to_idx(ord
);
2329 if (!(ord
& IMSM_ORD_REBUILD
) &&
2330 get_imsm_missing(super
, idx
)) {
2336 if (state
== IMSM_T_STATE_FAILED
)
2338 else if (state
== IMSM_T_STATE_DEGRADED
&&
2339 (state
!= map
->map_state
|| missing
))
2341 else /* we're normal, or already degraded */
2344 /* in the missing/failed disk case check to see
2345 * if at least one array is runnable
2347 max_enough
= max(max_enough
, enough
);
2349 dprintf("%s: enough: %d\n", __func__
, max_enough
);
2350 info
->container_enough
= max_enough
;
2353 __u32 reserved
= imsm_reserved_sectors(super
, super
->disks
);
2355 disk
= &super
->disks
->disk
;
2356 info
->data_offset
= __le32_to_cpu(disk
->total_blocks
) - reserved
;
2357 info
->component_size
= reserved
;
2358 info
->disk
.state
= is_configured(disk
) ? (1 << MD_DISK_ACTIVE
) : 0;
2359 /* we don't change info->disk.raid_disk here because
2360 * this state will be finalized in mdmon after we have
2361 * found the 'most fresh' version of the metadata
2363 info
->disk
.state
|= is_failed(disk
) ? (1 << MD_DISK_FAULTY
) : 0;
2364 info
->disk
.state
|= is_spare(disk
) ? 0 : (1 << MD_DISK_SYNC
);
2367 /* only call uuid_from_super_imsm when this disk is part of a populated container,
2368 * ->compare_super may have updated the 'num_raid_devs' field for spares
2370 if (info
->disk
.state
& (1 << MD_DISK_SYNC
) || super
->anchor
->num_raid_devs
)
2371 uuid_from_super_imsm(st
, info
->uuid
);
2373 memcpy(info
->uuid
, uuid_zero
, sizeof(uuid_zero
));
2375 /* I don't know how to compute 'map' on imsm, so use safe default */
2378 for (i
= 0; i
< map_disks
; i
++)
2384 /* allocates memory and fills disk in mdinfo structure
2385 * for each disk in array */
2386 struct mdinfo
*getinfo_super_disks_imsm(struct supertype
*st
)
2388 struct mdinfo
*mddev
= NULL
;
2389 struct intel_super
*super
= st
->sb
;
2390 struct imsm_disk
*disk
;
2393 if (!super
|| !super
->disks
)
2396 mddev
= malloc(sizeof(*mddev
));
2398 fprintf(stderr
, Name
": Failed to allocate memory.\n");
2401 memset(mddev
, 0, sizeof(*mddev
));
2405 tmp
= malloc(sizeof(*tmp
));
2407 fprintf(stderr
, Name
": Failed to allocate memory.\n");
2412 memset(tmp
, 0, sizeof(*tmp
));
2414 tmp
->next
= mddev
->devs
;
2416 tmp
->disk
.number
= count
++;
2417 tmp
->disk
.major
= dl
->major
;
2418 tmp
->disk
.minor
= dl
->minor
;
2419 tmp
->disk
.state
= is_configured(disk
) ?
2420 (1 << MD_DISK_ACTIVE
) : 0;
2421 tmp
->disk
.state
|= is_failed(disk
) ? (1 << MD_DISK_FAULTY
) : 0;
2422 tmp
->disk
.state
|= is_spare(disk
) ? 0 : (1 << MD_DISK_SYNC
);
2423 tmp
->disk
.raid_disk
= -1;
2429 static int update_super_imsm(struct supertype
*st
, struct mdinfo
*info
,
2430 char *update
, char *devname
, int verbose
,
2431 int uuid_set
, char *homehost
)
2433 /* For 'assemble' and 'force' we need to return non-zero if any
2434 * change was made. For others, the return value is ignored.
2435 * Update options are:
2436 * force-one : This device looks a bit old but needs to be included,
2437 * update age info appropriately.
2438 * assemble: clear any 'faulty' flag to allow this device to
2440 * force-array: Array is degraded but being forced, mark it clean
2441 * if that will be needed to assemble it.
2443 * newdev: not used ????
2444 * grow: Array has gained a new device - this is currently for
2446 * resync: mark as dirty so a resync will happen.
2447 * name: update the name - preserving the homehost
2448 * uuid: Change the uuid of the array to match watch is given
2450 * Following are not relevant for this imsm:
2451 * sparc2.2 : update from old dodgey metadata
2452 * super-minor: change the preferred_minor number
2453 * summaries: update redundant counters.
2454 * homehost: update the recorded homehost
2455 * _reshape_progress: record new reshape_progress position.
2458 struct intel_super
*super
= st
->sb
;
2459 struct imsm_super
*mpb
;
2461 /* we can only update container info */
2462 if (!super
|| super
->current_vol
>= 0 || !super
->anchor
)
2465 mpb
= super
->anchor
;
2467 if (strcmp(update
, "uuid") == 0 && uuid_set
&& !info
->update_private
)
2469 else if (strcmp(update
, "uuid") == 0 && uuid_set
&& info
->update_private
) {
2470 mpb
->orig_family_num
= *((__u32
*) info
->update_private
);
2472 } else if (strcmp(update
, "uuid") == 0) {
2473 __u32
*new_family
= malloc(sizeof(*new_family
));
2475 /* update orig_family_number with the incoming random
2476 * data, report the new effective uuid, and store the
2477 * new orig_family_num for future updates.
2480 memcpy(&mpb
->orig_family_num
, info
->uuid
, sizeof(__u32
));
2481 uuid_from_super_imsm(st
, info
->uuid
);
2482 *new_family
= mpb
->orig_family_num
;
2483 info
->update_private
= new_family
;
2486 } else if (strcmp(update
, "assemble") == 0)
2491 /* successful update? recompute checksum */
2493 mpb
->check_sum
= __le32_to_cpu(__gen_imsm_checksum(mpb
));
2498 static size_t disks_to_mpb_size(int disks
)
2502 size
= sizeof(struct imsm_super
);
2503 size
+= (disks
- 1) * sizeof(struct imsm_disk
);
2504 size
+= 2 * sizeof(struct imsm_dev
);
2505 /* up to 2 maps per raid device (-2 for imsm_maps in imsm_dev */
2506 size
+= (4 - 2) * sizeof(struct imsm_map
);
2507 /* 4 possible disk_ord_tbl's */
2508 size
+= 4 * (disks
- 1) * sizeof(__u32
);
2513 static __u64
avail_size_imsm(struct supertype
*st
, __u64 devsize
)
2515 if (devsize
< (MPB_SECTOR_CNT
+ IMSM_RESERVED_SECTORS
))
2518 return devsize
- (MPB_SECTOR_CNT
+ IMSM_RESERVED_SECTORS
);
2521 static void free_devlist(struct intel_super
*super
)
2523 struct intel_dev
*dv
;
2525 while (super
->devlist
) {
2526 dv
= super
->devlist
->next
;
2527 free(super
->devlist
->dev
);
2528 free(super
->devlist
);
2529 super
->devlist
= dv
;
2533 static void imsm_copy_dev(struct imsm_dev
*dest
, struct imsm_dev
*src
)
2535 memcpy(dest
, src
, sizeof_imsm_dev(src
, 0));
2538 static int compare_super_imsm(struct supertype
*st
, struct supertype
*tst
)
2542 * 0 same, or first was empty, and second was copied
2543 * 1 second had wrong number
2545 * 3 wrong other info
2547 struct intel_super
*first
= st
->sb
;
2548 struct intel_super
*sec
= tst
->sb
;
2555 /* in platform dependent environment test if the disks
2556 * use the same Intel hba
2558 if (!check_env("IMSM_NO_PLATFORM")) {
2559 if (!first
->hba
|| !sec
->hba
||
2560 (first
->hba
->type
!= sec
->hba
->type
)) {
2562 "HBAs of devices does not match %s != %s\n",
2563 first
->hba
? get_sys_dev_type(first
->hba
->type
) : NULL
,
2564 sec
->hba
? get_sys_dev_type(sec
->hba
->type
) : NULL
);
2569 /* if an anchor does not have num_raid_devs set then it is a free
2572 if (first
->anchor
->num_raid_devs
> 0 &&
2573 sec
->anchor
->num_raid_devs
> 0) {
2574 /* Determine if these disks might ever have been
2575 * related. Further disambiguation can only take place
2576 * in load_super_imsm_all
2578 __u32 first_family
= first
->anchor
->orig_family_num
;
2579 __u32 sec_family
= sec
->anchor
->orig_family_num
;
2581 if (memcmp(first
->anchor
->sig
, sec
->anchor
->sig
,
2582 MAX_SIGNATURE_LENGTH
) != 0)
2585 if (first_family
== 0)
2586 first_family
= first
->anchor
->family_num
;
2587 if (sec_family
== 0)
2588 sec_family
= sec
->anchor
->family_num
;
2590 if (first_family
!= sec_family
)
2596 /* if 'first' is a spare promote it to a populated mpb with sec's
2599 if (first
->anchor
->num_raid_devs
== 0 &&
2600 sec
->anchor
->num_raid_devs
> 0) {
2602 struct intel_dev
*dv
;
2603 struct imsm_dev
*dev
;
2605 /* we need to copy raid device info from sec if an allocation
2606 * fails here we don't associate the spare
2608 for (i
= 0; i
< sec
->anchor
->num_raid_devs
; i
++) {
2609 dv
= malloc(sizeof(*dv
));
2612 dev
= malloc(sizeof_imsm_dev(get_imsm_dev(sec
, i
), 1));
2619 dv
->next
= first
->devlist
;
2620 first
->devlist
= dv
;
2622 if (i
< sec
->anchor
->num_raid_devs
) {
2623 /* allocation failure */
2624 free_devlist(first
);
2625 fprintf(stderr
, "imsm: failed to associate spare\n");
2628 first
->anchor
->num_raid_devs
= sec
->anchor
->num_raid_devs
;
2629 first
->anchor
->orig_family_num
= sec
->anchor
->orig_family_num
;
2630 first
->anchor
->family_num
= sec
->anchor
->family_num
;
2631 memcpy(first
->anchor
->sig
, sec
->anchor
->sig
, MAX_SIGNATURE_LENGTH
);
2632 for (i
= 0; i
< sec
->anchor
->num_raid_devs
; i
++)
2633 imsm_copy_dev(get_imsm_dev(first
, i
), get_imsm_dev(sec
, i
));
2639 static void fd2devname(int fd
, char *name
)
2643 char dname
[PATH_MAX
];
2648 if (fstat(fd
, &st
) != 0)
2650 sprintf(path
, "/sys/dev/block/%d:%d",
2651 major(st
.st_rdev
), minor(st
.st_rdev
));
2653 rv
= readlink(path
, dname
, sizeof(dname
));
2658 nm
= strrchr(dname
, '/');
2660 snprintf(name
, MAX_RAID_SERIAL_LEN
, "/dev/%s", nm
);
2663 extern int scsi_get_serial(int fd
, void *buf
, size_t buf_len
);
2665 static int imsm_read_serial(int fd
, char *devname
,
2666 __u8 serial
[MAX_RAID_SERIAL_LEN
])
2668 unsigned char scsi_serial
[255];
2677 memset(scsi_serial
, 0, sizeof(scsi_serial
));
2679 rv
= scsi_get_serial(fd
, scsi_serial
, sizeof(scsi_serial
));
2681 if (rv
&& check_env("IMSM_DEVNAME_AS_SERIAL")) {
2682 memset(serial
, 0, MAX_RAID_SERIAL_LEN
);
2683 fd2devname(fd
, (char *) serial
);
2690 Name
": Failed to retrieve serial for %s\n",
2695 rsp_len
= scsi_serial
[3];
2699 Name
": Failed to retrieve serial for %s\n",
2703 rsp_buf
= (char *) &scsi_serial
[4];
2705 /* trim all whitespace and non-printable characters and convert
2708 for (i
= 0, dest
= rsp_buf
; i
< rsp_len
; i
++) {
2711 /* ':' is reserved for use in placeholder serial
2712 * numbers for missing disks
2720 len
= dest
- rsp_buf
;
2723 /* truncate leading characters */
2724 if (len
> MAX_RAID_SERIAL_LEN
) {
2725 dest
+= len
- MAX_RAID_SERIAL_LEN
;
2726 len
= MAX_RAID_SERIAL_LEN
;
2729 memset(serial
, 0, MAX_RAID_SERIAL_LEN
);
2730 memcpy(serial
, dest
, len
);
2735 static int serialcmp(__u8
*s1
, __u8
*s2
)
2737 return strncmp((char *) s1
, (char *) s2
, MAX_RAID_SERIAL_LEN
);
2740 static void serialcpy(__u8
*dest
, __u8
*src
)
2742 strncpy((char *) dest
, (char *) src
, MAX_RAID_SERIAL_LEN
);
2746 static struct dl
*serial_to_dl(__u8
*serial
, struct intel_super
*super
)
2750 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
2751 if (serialcmp(dl
->serial
, serial
) == 0)
2758 static struct imsm_disk
*
2759 __serial_to_disk(__u8
*serial
, struct imsm_super
*mpb
, int *idx
)
2763 for (i
= 0; i
< mpb
->num_disks
; i
++) {
2764 struct imsm_disk
*disk
= __get_imsm_disk(mpb
, i
);
2766 if (serialcmp(disk
->serial
, serial
) == 0) {
2777 load_imsm_disk(int fd
, struct intel_super
*super
, char *devname
, int keep_fd
)
2779 struct imsm_disk
*disk
;
2784 __u8 serial
[MAX_RAID_SERIAL_LEN
];
2786 rv
= imsm_read_serial(fd
, devname
, serial
);
2791 dl
= calloc(1, sizeof(*dl
));
2795 Name
": failed to allocate disk buffer for %s\n",
2801 dl
->major
= major(stb
.st_rdev
);
2802 dl
->minor
= minor(stb
.st_rdev
);
2803 dl
->next
= super
->disks
;
2804 dl
->fd
= keep_fd
? fd
: -1;
2805 assert(super
->disks
== NULL
);
2807 serialcpy(dl
->serial
, serial
);
2810 fd2devname(fd
, name
);
2812 dl
->devname
= strdup(devname
);
2814 dl
->devname
= strdup(name
);
2816 /* look up this disk's index in the current anchor */
2817 disk
= __serial_to_disk(dl
->serial
, super
->anchor
, &dl
->index
);
2820 /* only set index on disks that are a member of a
2821 * populated contianer, i.e. one with raid_devs
2823 if (is_failed(&dl
->disk
))
2825 else if (is_spare(&dl
->disk
))
2833 /* When migrating map0 contains the 'destination' state while map1
2834 * contains the current state. When not migrating map0 contains the
2835 * current state. This routine assumes that map[0].map_state is set to
2836 * the current array state before being called.
2838 * Migration is indicated by one of the following states
2839 * 1/ Idle (migr_state=0 map0state=normal||unitialized||degraded||failed)
2840 * 2/ Initialize (migr_state=1 migr_type=MIGR_INIT map0state=normal
2841 * map1state=unitialized)
2842 * 3/ Repair (Resync) (migr_state=1 migr_type=MIGR_REPAIR map0state=normal
2844 * 4/ Rebuild (migr_state=1 migr_type=MIGR_REBUILD map0state=normal
2845 * map1state=degraded)
2846 * 5/ Migration (mig_state=1 migr_type=MIGR_GEN_MIGR map0state=normal
2849 static void migrate(struct imsm_dev
*dev
, struct intel_super
*super
,
2850 __u8 to_state
, int migr_type
)
2852 struct imsm_map
*dest
;
2853 struct imsm_map
*src
= get_imsm_map(dev
, 0);
2855 dev
->vol
.migr_state
= 1;
2856 set_migr_type(dev
, migr_type
);
2857 dev
->vol
.curr_migr_unit
= 0;
2858 dest
= get_imsm_map(dev
, 1);
2860 /* duplicate and then set the target end state in map[0] */
2861 memcpy(dest
, src
, sizeof_imsm_map(src
));
2862 if ((migr_type
== MIGR_REBUILD
) ||
2863 (migr_type
== MIGR_GEN_MIGR
)) {
2867 for (i
= 0; i
< src
->num_members
; i
++) {
2868 ord
= __le32_to_cpu(src
->disk_ord_tbl
[i
]);
2869 set_imsm_ord_tbl_ent(src
, i
, ord_to_idx(ord
));
2873 if (migr_type
== MIGR_GEN_MIGR
)
2874 /* Clear migration record */
2875 memset(super
->migr_rec
, 0, sizeof(struct migr_record
));
2877 src
->map_state
= to_state
;
2880 static void end_migration(struct imsm_dev
*dev
, __u8 map_state
)
2882 struct imsm_map
*map
= get_imsm_map(dev
, 0);
2883 struct imsm_map
*prev
= get_imsm_map(dev
, dev
->vol
.migr_state
);
2886 /* merge any IMSM_ORD_REBUILD bits that were not successfully
2887 * completed in the last migration.
2889 * FIXME add support for raid-level-migration
2891 for (i
= 0; i
< prev
->num_members
; i
++)
2892 for (j
= 0; j
< map
->num_members
; j
++)
2893 /* during online capacity expansion
2894 * disks position can be changed if takeover is used
2896 if (ord_to_idx(map
->disk_ord_tbl
[j
]) ==
2897 ord_to_idx(prev
->disk_ord_tbl
[i
])) {
2898 map
->disk_ord_tbl
[j
] |= prev
->disk_ord_tbl
[i
];
2902 dev
->vol
.migr_state
= 0;
2903 dev
->vol
.migr_type
= 0;
2904 dev
->vol
.curr_migr_unit
= 0;
2905 map
->map_state
= map_state
;
2909 static int parse_raid_devices(struct intel_super
*super
)
2912 struct imsm_dev
*dev_new
;
2913 size_t len
, len_migr
;
2915 size_t space_needed
= 0;
2916 struct imsm_super
*mpb
= super
->anchor
;
2918 for (i
= 0; i
< super
->anchor
->num_raid_devs
; i
++) {
2919 struct imsm_dev
*dev_iter
= __get_imsm_dev(super
->anchor
, i
);
2920 struct intel_dev
*dv
;
2922 len
= sizeof_imsm_dev(dev_iter
, 0);
2923 len_migr
= sizeof_imsm_dev(dev_iter
, 1);
2925 space_needed
+= len_migr
- len
;
2927 dv
= malloc(sizeof(*dv
));
2930 if (max_len
< len_migr
)
2932 if (max_len
> len_migr
)
2933 space_needed
+= max_len
- len_migr
;
2934 dev_new
= malloc(max_len
);
2939 imsm_copy_dev(dev_new
, dev_iter
);
2942 dv
->next
= super
->devlist
;
2943 super
->devlist
= dv
;
2946 /* ensure that super->buf is large enough when all raid devices
2949 if (__le32_to_cpu(mpb
->mpb_size
) + space_needed
> super
->len
) {
2952 len
= ROUND_UP(__le32_to_cpu(mpb
->mpb_size
) + space_needed
, 512);
2953 if (posix_memalign(&buf
, 512, len
) != 0)
2956 memcpy(buf
, super
->buf
, super
->len
);
2957 memset(buf
+ super
->len
, 0, len
- super
->len
);
2966 /* retrieve a pointer to the bbm log which starts after all raid devices */
2967 struct bbm_log
*__get_imsm_bbm_log(struct imsm_super
*mpb
)
2971 if (__le32_to_cpu(mpb
->bbm_log_size
)) {
2973 ptr
+= mpb
->mpb_size
- __le32_to_cpu(mpb
->bbm_log_size
);
2979 /*******************************************************************************
2980 * Function: check_mpb_migr_compatibility
2981 * Description: Function checks for unsupported migration features:
2982 * - migration optimization area (pba_of_lba0)
2983 * - descending reshape (ascending_migr)
2985 * super : imsm metadata information
2987 * 0 : migration is compatible
2988 * -1 : migration is not compatible
2989 ******************************************************************************/
2990 int check_mpb_migr_compatibility(struct intel_super
*super
)
2992 struct imsm_map
*map0
, *map1
;
2993 struct migr_record
*migr_rec
= super
->migr_rec
;
2996 for (i
= 0; i
< super
->anchor
->num_raid_devs
; i
++) {
2997 struct imsm_dev
*dev_iter
= __get_imsm_dev(super
->anchor
, i
);
3000 dev_iter
->vol
.migr_state
== 1 &&
3001 dev_iter
->vol
.migr_type
== MIGR_GEN_MIGR
) {
3002 /* This device is migrating */
3003 map0
= get_imsm_map(dev_iter
, 0);
3004 map1
= get_imsm_map(dev_iter
, 1);
3005 if (map0
->pba_of_lba0
!= map1
->pba_of_lba0
)
3006 /* migration optimization area was used */
3008 if (migr_rec
->ascending_migr
== 0
3009 && migr_rec
->dest_depth_per_unit
> 0)
3010 /* descending reshape not supported yet */
3017 static void __free_imsm(struct intel_super
*super
, int free_disks
);
3019 /* load_imsm_mpb - read matrix metadata
3020 * allocates super->mpb to be freed by free_imsm
3022 static int load_imsm_mpb(int fd
, struct intel_super
*super
, char *devname
)
3024 unsigned long long dsize
;
3025 unsigned long long sectors
;
3027 struct imsm_super
*anchor
;
3030 get_dev_size(fd
, NULL
, &dsize
);
3034 Name
": %s: device to small for imsm\n",
3039 if (lseek64(fd
, dsize
- (512 * 2), SEEK_SET
) < 0) {
3041 fprintf(stderr
, Name
3042 ": Cannot seek to anchor block on %s: %s\n",
3043 devname
, strerror(errno
));
3047 if (posix_memalign((void**)&anchor
, 512, 512) != 0) {
3050 Name
": Failed to allocate imsm anchor buffer"
3051 " on %s\n", devname
);
3054 if (read(fd
, anchor
, 512) != 512) {
3057 Name
": Cannot read anchor block on %s: %s\n",
3058 devname
, strerror(errno
));
3063 if (strncmp((char *) anchor
->sig
, MPB_SIGNATURE
, MPB_SIG_LEN
) != 0) {
3066 Name
": no IMSM anchor on %s\n", devname
);
3071 __free_imsm(super
, 0);
3072 /* reload capability and hba */
3074 /* capability and hba must be updated with new super allocation */
3075 find_intel_hba_capability(fd
, super
, devname
);
3076 super
->len
= ROUND_UP(anchor
->mpb_size
, 512);
3077 if (posix_memalign(&super
->buf
, 512, super
->len
) != 0) {
3080 Name
": unable to allocate %zu byte mpb buffer\n",
3085 memcpy(super
->buf
, anchor
, 512);
3087 sectors
= mpb_sectors(anchor
) - 1;
3090 if (posix_memalign(&super
->migr_rec_buf
, 512, 512) != 0) {
3091 fprintf(stderr
, Name
3092 ": %s could not allocate migr_rec buffer\n", __func__
);
3098 check_sum
= __gen_imsm_checksum(super
->anchor
);
3099 if (check_sum
!= __le32_to_cpu(super
->anchor
->check_sum
)) {
3102 Name
": IMSM checksum %x != %x on %s\n",
3104 __le32_to_cpu(super
->anchor
->check_sum
),
3112 /* read the extended mpb */
3113 if (lseek64(fd
, dsize
- (512 * (2 + sectors
)), SEEK_SET
) < 0) {
3116 Name
": Cannot seek to extended mpb on %s: %s\n",
3117 devname
, strerror(errno
));
3121 if ((unsigned)read(fd
, super
->buf
+ 512, super
->len
- 512) != super
->len
- 512) {
3124 Name
": Cannot read extended mpb on %s: %s\n",
3125 devname
, strerror(errno
));
3129 check_sum
= __gen_imsm_checksum(super
->anchor
);
3130 if (check_sum
!= __le32_to_cpu(super
->anchor
->check_sum
)) {
3133 Name
": IMSM checksum %x != %x on %s\n",
3134 check_sum
, __le32_to_cpu(super
->anchor
->check_sum
),
3139 /* FIXME the BBM log is disk specific so we cannot use this global
3140 * buffer for all disks. Ok for now since we only look at the global
3141 * bbm_log_size parameter to gate assembly
3143 super
->bbm_log
= __get_imsm_bbm_log(super
->anchor
);
3148 static int read_imsm_migr_rec(int fd
, struct intel_super
*super
);
3151 load_and_parse_mpb(int fd
, struct intel_super
*super
, char *devname
, int keep_fd
)
3155 err
= load_imsm_mpb(fd
, super
, devname
);
3158 err
= load_imsm_disk(fd
, super
, devname
, keep_fd
);
3161 err
= parse_raid_devices(super
);
3166 static void __free_imsm_disk(struct dl
*d
)
3178 static void free_imsm_disks(struct intel_super
*super
)
3182 while (super
->disks
) {
3184 super
->disks
= d
->next
;
3185 __free_imsm_disk(d
);
3187 while (super
->disk_mgmt_list
) {
3188 d
= super
->disk_mgmt_list
;
3189 super
->disk_mgmt_list
= d
->next
;
3190 __free_imsm_disk(d
);
3192 while (super
->missing
) {
3194 super
->missing
= d
->next
;
3195 __free_imsm_disk(d
);
3200 /* free all the pieces hanging off of a super pointer */
3201 static void __free_imsm(struct intel_super
*super
, int free_disks
)
3203 struct intel_hba
*elem
, *next
;
3209 /* unlink capability description */
3211 if (super
->migr_rec_buf
) {
3212 free(super
->migr_rec_buf
);
3213 super
->migr_rec_buf
= NULL
;
3216 free_imsm_disks(super
);
3217 free_devlist(super
);
3221 free((void *)elem
->path
);
3229 static void free_imsm(struct intel_super
*super
)
3231 __free_imsm(super
, 1);
3235 static void free_super_imsm(struct supertype
*st
)
3237 struct intel_super
*super
= st
->sb
;
3246 static struct intel_super
*alloc_super(void)
3248 struct intel_super
*super
= malloc(sizeof(*super
));
3251 memset(super
, 0, sizeof(*super
));
3252 super
->current_vol
= -1;
3253 super
->create_offset
= ~((__u32
) 0);
3259 * find and allocate hba and OROM/EFI based on valid fd of RAID component device
3261 static int find_intel_hba_capability(int fd
, struct intel_super
*super
, char *devname
)
3263 struct sys_dev
*hba_name
;
3266 if ((fd
< 0) || check_env("IMSM_NO_PLATFORM")) {
3271 hba_name
= find_disk_attached_hba(fd
, NULL
);
3275 Name
": %s is not attached to Intel(R) RAID controller.\n",
3279 rv
= attach_hba_to_super(super
, hba_name
);
3282 struct intel_hba
*hba
= super
->hba
;
3284 fprintf(stderr
, Name
": %s is attached to Intel(R) %s RAID "
3285 "controller (%s),\n"
3286 " but the container is assigned to Intel(R) "
3287 "%s RAID controller (",
3290 hba_name
->pci_id
? : "Err!",
3291 get_sys_dev_type(hba_name
->type
));
3294 fprintf(stderr
, "%s", hba
->pci_id
? : "Err!");
3296 fprintf(stderr
, ", ");
3300 fprintf(stderr
, ").\n"
3301 " Mixing devices attached to different controllers "
3302 "is not allowed.\n");
3304 free_sys_dev(&hba_name
);
3307 super
->orom
= find_imsm_capability(hba_name
->type
);
3308 free_sys_dev(&hba_name
);
3315 /* find_missing - helper routine for load_super_imsm_all that identifies
3316 * disks that have disappeared from the system. This routine relies on
3317 * the mpb being uptodate, which it is at load time.
3319 static int find_missing(struct intel_super
*super
)
3322 struct imsm_super
*mpb
= super
->anchor
;
3324 struct imsm_disk
*disk
;
3326 for (i
= 0; i
< mpb
->num_disks
; i
++) {
3327 disk
= __get_imsm_disk(mpb
, i
);
3328 dl
= serial_to_dl(disk
->serial
, super
);
3332 dl
= malloc(sizeof(*dl
));
3338 dl
->devname
= strdup("missing");
3340 serialcpy(dl
->serial
, disk
->serial
);
3343 dl
->next
= super
->missing
;
3344 super
->missing
= dl
;
3350 static struct intel_disk
*disk_list_get(__u8
*serial
, struct intel_disk
*disk_list
)
3352 struct intel_disk
*idisk
= disk_list
;
3355 if (serialcmp(idisk
->disk
.serial
, serial
) == 0)
3357 idisk
= idisk
->next
;
3363 static int __prep_thunderdome(struct intel_super
**table
, int tbl_size
,
3364 struct intel_super
*super
,
3365 struct intel_disk
**disk_list
)
3367 struct imsm_disk
*d
= &super
->disks
->disk
;
3368 struct imsm_super
*mpb
= super
->anchor
;
3371 for (i
= 0; i
< tbl_size
; i
++) {
3372 struct imsm_super
*tbl_mpb
= table
[i
]->anchor
;
3373 struct imsm_disk
*tbl_d
= &table
[i
]->disks
->disk
;
3375 if (tbl_mpb
->family_num
== mpb
->family_num
) {
3376 if (tbl_mpb
->check_sum
== mpb
->check_sum
) {
3377 dprintf("%s: mpb from %d:%d matches %d:%d\n",
3378 __func__
, super
->disks
->major
,
3379 super
->disks
->minor
,
3380 table
[i
]->disks
->major
,
3381 table
[i
]->disks
->minor
);
3385 if (((is_configured(d
) && !is_configured(tbl_d
)) ||
3386 is_configured(d
) == is_configured(tbl_d
)) &&
3387 tbl_mpb
->generation_num
< mpb
->generation_num
) {
3388 /* current version of the mpb is a
3389 * better candidate than the one in
3390 * super_table, but copy over "cross
3391 * generational" status
3393 struct intel_disk
*idisk
;
3395 dprintf("%s: mpb from %d:%d replaces %d:%d\n",
3396 __func__
, super
->disks
->major
,
3397 super
->disks
->minor
,
3398 table
[i
]->disks
->major
,
3399 table
[i
]->disks
->minor
);
3401 idisk
= disk_list_get(tbl_d
->serial
, *disk_list
);
3402 if (idisk
&& is_failed(&idisk
->disk
))
3403 tbl_d
->status
|= FAILED_DISK
;
3406 struct intel_disk
*idisk
;
3407 struct imsm_disk
*disk
;
3409 /* tbl_mpb is more up to date, but copy
3410 * over cross generational status before
3413 disk
= __serial_to_disk(d
->serial
, mpb
, NULL
);
3414 if (disk
&& is_failed(disk
))
3415 d
->status
|= FAILED_DISK
;
3417 idisk
= disk_list_get(d
->serial
, *disk_list
);
3420 if (disk
&& is_configured(disk
))
3421 idisk
->disk
.status
|= CONFIGURED_DISK
;
3424 dprintf("%s: mpb from %d:%d prefer %d:%d\n",
3425 __func__
, super
->disks
->major
,
3426 super
->disks
->minor
,
3427 table
[i
]->disks
->major
,
3428 table
[i
]->disks
->minor
);
3436 table
[tbl_size
++] = super
;
3440 /* update/extend the merged list of imsm_disk records */
3441 for (j
= 0; j
< mpb
->num_disks
; j
++) {
3442 struct imsm_disk
*disk
= __get_imsm_disk(mpb
, j
);
3443 struct intel_disk
*idisk
;
3445 idisk
= disk_list_get(disk
->serial
, *disk_list
);
3447 idisk
->disk
.status
|= disk
->status
;
3448 if (is_configured(&idisk
->disk
) ||
3449 is_failed(&idisk
->disk
))
3450 idisk
->disk
.status
&= ~(SPARE_DISK
);
3452 idisk
= calloc(1, sizeof(*idisk
));
3455 idisk
->owner
= IMSM_UNKNOWN_OWNER
;
3456 idisk
->disk
= *disk
;
3457 idisk
->next
= *disk_list
;
3461 if (serialcmp(idisk
->disk
.serial
, d
->serial
) == 0)
3468 static struct intel_super
*
3469 validate_members(struct intel_super
*super
, struct intel_disk
*disk_list
,
3472 struct imsm_super
*mpb
= super
->anchor
;
3476 for (i
= 0; i
< mpb
->num_disks
; i
++) {
3477 struct imsm_disk
*disk
= __get_imsm_disk(mpb
, i
);
3478 struct intel_disk
*idisk
;
3480 idisk
= disk_list_get(disk
->serial
, disk_list
);
3482 if (idisk
->owner
== owner
||
3483 idisk
->owner
== IMSM_UNKNOWN_OWNER
)
3486 dprintf("%s: '%.16s' owner %d != %d\n",
3487 __func__
, disk
->serial
, idisk
->owner
,
3490 dprintf("%s: unknown disk %x [%d]: %.16s\n",
3491 __func__
, __le32_to_cpu(mpb
->family_num
), i
,
3497 if (ok_count
== mpb
->num_disks
)
3502 static void show_conflicts(__u32 family_num
, struct intel_super
*super_list
)
3504 struct intel_super
*s
;
3506 for (s
= super_list
; s
; s
= s
->next
) {
3507 if (family_num
!= s
->anchor
->family_num
)
3509 fprintf(stderr
, "Conflict, offlining family %#x on '%s'\n",
3510 __le32_to_cpu(family_num
), s
->disks
->devname
);
3514 static struct intel_super
*
3515 imsm_thunderdome(struct intel_super
**super_list
, int len
)
3517 struct intel_super
*super_table
[len
];
3518 struct intel_disk
*disk_list
= NULL
;
3519 struct intel_super
*champion
, *spare
;
3520 struct intel_super
*s
, **del
;
3525 memset(super_table
, 0, sizeof(super_table
));
3526 for (s
= *super_list
; s
; s
= s
->next
)
3527 tbl_size
= __prep_thunderdome(super_table
, tbl_size
, s
, &disk_list
);
3529 for (i
= 0; i
< tbl_size
; i
++) {
3530 struct imsm_disk
*d
;
3531 struct intel_disk
*idisk
;
3532 struct imsm_super
*mpb
= super_table
[i
]->anchor
;
3535 d
= &s
->disks
->disk
;
3537 /* 'd' must appear in merged disk list for its
3538 * configuration to be valid
3540 idisk
= disk_list_get(d
->serial
, disk_list
);
3541 if (idisk
&& idisk
->owner
== i
)
3542 s
= validate_members(s
, disk_list
, i
);
3547 dprintf("%s: marking family: %#x from %d:%d offline\n",
3548 __func__
, mpb
->family_num
,
3549 super_table
[i
]->disks
->major
,
3550 super_table
[i
]->disks
->minor
);
3554 /* This is where the mdadm implementation differs from the Windows
3555 * driver which has no strict concept of a container. We can only
3556 * assemble one family from a container, so when returning a prodigal
3557 * array member to this system the code will not be able to disambiguate
3558 * the container contents that should be assembled ("foreign" versus
3559 * "local"). It requires user intervention to set the orig_family_num
3560 * to a new value to establish a new container. The Windows driver in
3561 * this situation fixes up the volume name in place and manages the
3562 * foreign array as an independent entity.
3567 for (i
= 0; i
< tbl_size
; i
++) {
3568 struct intel_super
*tbl_ent
= super_table
[i
];
3574 if (tbl_ent
->anchor
->num_raid_devs
== 0) {
3579 if (s
&& !is_spare
) {
3580 show_conflicts(tbl_ent
->anchor
->family_num
, *super_list
);
3582 } else if (!s
&& !is_spare
)
3595 fprintf(stderr
, "Chose family %#x on '%s', "
3596 "assemble conflicts to new container with '--update=uuid'\n",
3597 __le32_to_cpu(s
->anchor
->family_num
), s
->disks
->devname
);
3599 /* collect all dl's onto 'champion', and update them to
3600 * champion's version of the status
3602 for (s
= *super_list
; s
; s
= s
->next
) {
3603 struct imsm_super
*mpb
= champion
->anchor
;
3604 struct dl
*dl
= s
->disks
;
3609 for (i
= 0; i
< mpb
->num_disks
; i
++) {
3610 struct imsm_disk
*disk
;
3612 disk
= __serial_to_disk(dl
->serial
, mpb
, &dl
->index
);
3615 /* only set index on disks that are a member of
3616 * a populated contianer, i.e. one with
3619 if (is_failed(&dl
->disk
))
3621 else if (is_spare(&dl
->disk
))
3627 if (i
>= mpb
->num_disks
) {
3628 struct intel_disk
*idisk
;
3630 idisk
= disk_list_get(dl
->serial
, disk_list
);
3631 if (idisk
&& is_spare(&idisk
->disk
) &&
3632 !is_failed(&idisk
->disk
) && !is_configured(&idisk
->disk
))
3640 dl
->next
= champion
->disks
;
3641 champion
->disks
= dl
;
3645 /* delete 'champion' from super_list */
3646 for (del
= super_list
; *del
; ) {
3647 if (*del
== champion
) {
3648 *del
= (*del
)->next
;
3651 del
= &(*del
)->next
;
3653 champion
->next
= NULL
;
3657 struct intel_disk
*idisk
= disk_list
;
3659 disk_list
= disk_list
->next
;
3666 static int load_super_imsm_all(struct supertype
*st
, int fd
, void **sbp
,
3670 struct intel_super
*super_list
= NULL
;
3671 struct intel_super
*super
= NULL
;
3672 int devnum
= fd2devnum(fd
);
3678 /* check if 'fd' an opened container */
3679 sra
= sysfs_read(fd
, 0, GET_LEVEL
|GET_VERSION
|GET_DEVS
|GET_STATE
);
3683 if (sra
->array
.major_version
!= -1 ||
3684 sra
->array
.minor_version
!= -2 ||
3685 strcmp(sra
->text_version
, "imsm") != 0) {
3690 for (sd
= sra
->devs
, i
= 0; sd
; sd
= sd
->next
, i
++) {
3691 struct intel_super
*s
= alloc_super();
3699 s
->next
= super_list
;
3703 sprintf(nm
, "%d:%d", sd
->disk
.major
, sd
->disk
.minor
);
3704 dfd
= dev_open(nm
, O_RDWR
);
3708 rv
= find_intel_hba_capability(dfd
, s
, devname
);
3709 /* no orom/efi or non-intel hba of the disk */
3713 err
= load_and_parse_mpb(dfd
, s
, NULL
, 1);
3715 /* retry the load if we might have raced against mdmon */
3716 if (err
== 3 && mdmon_running(devnum
))
3717 for (retry
= 0; retry
< 3; retry
++) {
3719 err
= load_and_parse_mpb(dfd
, s
, NULL
, 1);
3727 /* all mpbs enter, maybe one leaves */
3728 super
= imsm_thunderdome(&super_list
, i
);
3734 if (find_missing(super
) != 0) {
3740 /* load migration record */
3741 err
= load_imsm_migr_rec(super
, NULL
);
3747 /* Check migration compatibility */
3748 if (check_mpb_migr_compatibility(super
) != 0) {
3749 fprintf(stderr
, Name
": Unsupported migration detected");
3751 fprintf(stderr
, " on %s\n", devname
);
3753 fprintf(stderr
, " (IMSM).\n");
3762 while (super_list
) {
3763 struct intel_super
*s
= super_list
;
3765 super_list
= super_list
->next
;
3774 st
->container_dev
= devnum
;
3775 if (err
== 0 && st
->ss
== NULL
) {
3776 st
->ss
= &super_imsm
;
3777 st
->minor_version
= 0;
3778 st
->max_devs
= IMSM_MAX_DEVICES
;
3783 static int load_container_imsm(struct supertype
*st
, int fd
, char *devname
)
3785 return load_super_imsm_all(st
, fd
, &st
->sb
, devname
);
3789 static int load_super_imsm(struct supertype
*st
, int fd
, char *devname
)
3791 struct intel_super
*super
;
3794 if (test_partition(fd
))
3795 /* IMSM not allowed on partitions */
3798 free_super_imsm(st
);
3800 super
= alloc_super();
3803 Name
": malloc of %zu failed.\n",
3807 /* Load hba and capabilities if they exist.
3808 * But do not preclude loading metadata in case capabilities or hba are
3809 * non-compliant and ignore_hw_compat is set.
3811 rv
= find_intel_hba_capability(fd
, super
, devname
);
3812 /* no orom/efi or non-intel hba of the disk */
3813 if ((rv
!= 0) && (st
->ignore_hw_compat
== 0)) {
3816 Name
": No OROM/EFI properties for %s\n", devname
);
3820 rv
= load_and_parse_mpb(fd
, super
, devname
, 0);
3825 Name
": Failed to load all information "
3826 "sections on %s\n", devname
);
3832 if (st
->ss
== NULL
) {
3833 st
->ss
= &super_imsm
;
3834 st
->minor_version
= 0;
3835 st
->max_devs
= IMSM_MAX_DEVICES
;
3838 /* load migration record */
3839 if (load_imsm_migr_rec(super
, NULL
) == 0) {
3840 /* Check for unsupported migration features */
3841 if (check_mpb_migr_compatibility(super
) != 0) {
3843 Name
": Unsupported migration detected");
3845 fprintf(stderr
, " on %s\n", devname
);
3847 fprintf(stderr
, " (IMSM).\n");
3855 static __u16
info_to_blocks_per_strip(mdu_array_info_t
*info
)
3857 if (info
->level
== 1)
3859 return info
->chunk_size
>> 9;
3862 static __u32
info_to_num_data_stripes(mdu_array_info_t
*info
, int num_domains
)
3866 num_stripes
= (info
->size
* 2) / info_to_blocks_per_strip(info
);
3867 num_stripes
/= num_domains
;
3872 static __u32
info_to_blocks_per_member(mdu_array_info_t
*info
)
3874 if (info
->level
== 1)
3875 return info
->size
* 2;
3877 return (info
->size
* 2) & ~(info_to_blocks_per_strip(info
) - 1);
3880 static void imsm_update_version_info(struct intel_super
*super
)
3882 /* update the version and attributes */
3883 struct imsm_super
*mpb
= super
->anchor
;
3885 struct imsm_dev
*dev
;
3886 struct imsm_map
*map
;
3889 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
3890 dev
= get_imsm_dev(super
, i
);
3891 map
= get_imsm_map(dev
, 0);
3892 if (__le32_to_cpu(dev
->size_high
) > 0)
3893 mpb
->attributes
|= MPB_ATTRIB_2TB
;
3895 /* FIXME detect when an array spans a port multiplier */
3897 mpb
->attributes
|= MPB_ATTRIB_PM
;
3900 if (mpb
->num_raid_devs
> 1 ||
3901 mpb
->attributes
!= MPB_ATTRIB_CHECKSUM_VERIFY
) {
3902 version
= MPB_VERSION_ATTRIBS
;
3903 switch (get_imsm_raid_level(map
)) {
3904 case 0: mpb
->attributes
|= MPB_ATTRIB_RAID0
; break;
3905 case 1: mpb
->attributes
|= MPB_ATTRIB_RAID1
; break;
3906 case 10: mpb
->attributes
|= MPB_ATTRIB_RAID10
; break;
3907 case 5: mpb
->attributes
|= MPB_ATTRIB_RAID5
; break;
3910 if (map
->num_members
>= 5)
3911 version
= MPB_VERSION_5OR6_DISK_ARRAY
;
3912 else if (dev
->status
== DEV_CLONE_N_GO
)
3913 version
= MPB_VERSION_CNG
;
3914 else if (get_imsm_raid_level(map
) == 5)
3915 version
= MPB_VERSION_RAID5
;
3916 else if (map
->num_members
>= 3)
3917 version
= MPB_VERSION_3OR4_DISK_ARRAY
;
3918 else if (get_imsm_raid_level(map
) == 1)
3919 version
= MPB_VERSION_RAID1
;
3921 version
= MPB_VERSION_RAID0
;
3923 strcpy(((char *) mpb
->sig
) + strlen(MPB_SIGNATURE
), version
);
3927 static int check_name(struct intel_super
*super
, char *name
, int quiet
)
3929 struct imsm_super
*mpb
= super
->anchor
;
3930 char *reason
= NULL
;
3933 if (strlen(name
) > MAX_RAID_SERIAL_LEN
)
3934 reason
= "must be 16 characters or less";
3936 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
3937 struct imsm_dev
*dev
= get_imsm_dev(super
, i
);
3939 if (strncmp((char *) dev
->volume
, name
, MAX_RAID_SERIAL_LEN
) == 0) {
3940 reason
= "already exists";
3945 if (reason
&& !quiet
)
3946 fprintf(stderr
, Name
": imsm volume name %s\n", reason
);
3951 static int init_super_imsm_volume(struct supertype
*st
, mdu_array_info_t
*info
,
3952 unsigned long long size
, char *name
,
3953 char *homehost
, int *uuid
)
3955 /* We are creating a volume inside a pre-existing container.
3956 * so st->sb is already set.
3958 struct intel_super
*super
= st
->sb
;
3959 struct imsm_super
*mpb
= super
->anchor
;
3960 struct intel_dev
*dv
;
3961 struct imsm_dev
*dev
;
3962 struct imsm_vol
*vol
;
3963 struct imsm_map
*map
;
3964 int idx
= mpb
->num_raid_devs
;
3966 unsigned long long array_blocks
;
3967 size_t size_old
, size_new
;
3968 __u32 num_data_stripes
;
3970 if (super
->orom
&& mpb
->num_raid_devs
>= super
->orom
->vpa
) {
3971 fprintf(stderr
, Name
": This imsm-container already has the "
3972 "maximum of %d volumes\n", super
->orom
->vpa
);
3976 /* ensure the mpb is large enough for the new data */
3977 size_old
= __le32_to_cpu(mpb
->mpb_size
);
3978 size_new
= disks_to_mpb_size(info
->nr_disks
);
3979 if (size_new
> size_old
) {
3981 size_t size_round
= ROUND_UP(size_new
, 512);
3983 if (posix_memalign(&mpb_new
, 512, size_round
) != 0) {
3984 fprintf(stderr
, Name
": could not allocate new mpb\n");
3987 if (posix_memalign(&super
->migr_rec_buf
, 512, 512) != 0) {
3988 fprintf(stderr
, Name
3989 ": %s could not allocate migr_rec buffer\n",
3995 memcpy(mpb_new
, mpb
, size_old
);
3998 super
->anchor
= mpb_new
;
3999 mpb
->mpb_size
= __cpu_to_le32(size_new
);
4000 memset(mpb_new
+ size_old
, 0, size_round
- size_old
);
4002 super
->current_vol
= idx
;
4003 /* when creating the first raid device in this container set num_disks
4004 * to zero, i.e. delete this spare and add raid member devices in
4005 * add_to_super_imsm_volume()
4007 if (super
->current_vol
== 0)
4010 if (!check_name(super
, name
, 0))
4012 dv
= malloc(sizeof(*dv
));
4014 fprintf(stderr
, Name
": failed to allocate device list entry\n");
4017 dev
= calloc(1, sizeof(*dev
) + sizeof(__u32
) * (info
->raid_disks
- 1));
4020 fprintf(stderr
, Name
": could not allocate raid device\n");
4024 strncpy((char *) dev
->volume
, name
, MAX_RAID_SERIAL_LEN
);
4025 if (info
->level
== 1)
4026 array_blocks
= info_to_blocks_per_member(info
);
4028 array_blocks
= calc_array_size(info
->level
, info
->raid_disks
,
4029 info
->layout
, info
->chunk_size
,
4031 /* round array size down to closest MB */
4032 array_blocks
= (array_blocks
>> SECT_PER_MB_SHIFT
) << SECT_PER_MB_SHIFT
;
4034 dev
->size_low
= __cpu_to_le32((__u32
) array_blocks
);
4035 dev
->size_high
= __cpu_to_le32((__u32
) (array_blocks
>> 32));
4036 dev
->status
= (DEV_READ_COALESCING
| DEV_WRITE_COALESCING
);
4038 vol
->migr_state
= 0;
4039 set_migr_type(dev
, MIGR_INIT
);
4041 vol
->curr_migr_unit
= 0;
4042 map
= get_imsm_map(dev
, 0);
4043 map
->pba_of_lba0
= __cpu_to_le32(super
->create_offset
);
4044 map
->blocks_per_member
= __cpu_to_le32(info_to_blocks_per_member(info
));
4045 map
->blocks_per_strip
= __cpu_to_le16(info_to_blocks_per_strip(info
));
4046 map
->failed_disk_num
= ~0;
4047 map
->map_state
= info
->level
? IMSM_T_STATE_UNINITIALIZED
:
4048 IMSM_T_STATE_NORMAL
;
4051 if (info
->level
== 1 && info
->raid_disks
> 2) {
4054 fprintf(stderr
, Name
": imsm does not support more than 2 disks"
4055 "in a raid1 volume\n");
4059 map
->raid_level
= info
->level
;
4060 if (info
->level
== 10) {
4061 map
->raid_level
= 1;
4062 map
->num_domains
= info
->raid_disks
/ 2;
4063 } else if (info
->level
== 1)
4064 map
->num_domains
= info
->raid_disks
;
4066 map
->num_domains
= 1;
4068 num_data_stripes
= info_to_num_data_stripes(info
, map
->num_domains
);
4069 map
->num_data_stripes
= __cpu_to_le32(num_data_stripes
);
4071 map
->num_members
= info
->raid_disks
;
4072 for (i
= 0; i
< map
->num_members
; i
++) {
4073 /* initialized in add_to_super */
4074 set_imsm_ord_tbl_ent(map
, i
, IMSM_ORD_REBUILD
);
4076 mpb
->num_raid_devs
++;
4079 dv
->index
= super
->current_vol
;
4080 dv
->next
= super
->devlist
;
4081 super
->devlist
= dv
;
4083 imsm_update_version_info(super
);
4088 static int init_super_imsm(struct supertype
*st
, mdu_array_info_t
*info
,
4089 unsigned long long size
, char *name
,
4090 char *homehost
, int *uuid
)
4092 /* This is primarily called by Create when creating a new array.
4093 * We will then get add_to_super called for each component, and then
4094 * write_init_super called to write it out to each device.
4095 * For IMSM, Create can create on fresh devices or on a pre-existing
4097 * To create on a pre-existing array a different method will be called.
4098 * This one is just for fresh drives.
4100 struct intel_super
*super
;
4101 struct imsm_super
*mpb
;
4106 return init_super_imsm_volume(st
, info
, size
, name
, homehost
, uuid
);
4109 mpb_size
= disks_to_mpb_size(info
->nr_disks
);
4113 super
= alloc_super();
4114 if (super
&& posix_memalign(&super
->buf
, 512, mpb_size
) != 0) {
4119 fprintf(stderr
, Name
4120 ": %s could not allocate superblock\n", __func__
);
4123 if (posix_memalign(&super
->migr_rec_buf
, 512, 512) != 0) {
4124 fprintf(stderr
, Name
4125 ": %s could not allocate migr_rec buffer\n", __func__
);
4130 memset(super
->buf
, 0, mpb_size
);
4132 mpb
->mpb_size
= __cpu_to_le32(mpb_size
);
4136 /* zeroing superblock */
4140 mpb
->attributes
= MPB_ATTRIB_CHECKSUM_VERIFY
;
4142 version
= (char *) mpb
->sig
;
4143 strcpy(version
, MPB_SIGNATURE
);
4144 version
+= strlen(MPB_SIGNATURE
);
4145 strcpy(version
, MPB_VERSION_RAID0
);
4151 static int add_to_super_imsm_volume(struct supertype
*st
, mdu_disk_info_t
*dk
,
4152 int fd
, char *devname
)
4154 struct intel_super
*super
= st
->sb
;
4155 struct imsm_super
*mpb
= super
->anchor
;
4157 struct imsm_dev
*dev
;
4158 struct imsm_map
*map
;
4161 dev
= get_imsm_dev(super
, super
->current_vol
);
4162 map
= get_imsm_map(dev
, 0);
4164 if (! (dk
->state
& (1<<MD_DISK_SYNC
))) {
4165 fprintf(stderr
, Name
": %s: Cannot add spare devices to IMSM volume\n",
4171 /* we're doing autolayout so grab the pre-marked (in
4172 * validate_geometry) raid_disk
4174 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
4175 if (dl
->raiddisk
== dk
->raid_disk
)
4178 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
4179 if (dl
->major
== dk
->major
&&
4180 dl
->minor
== dk
->minor
)
4185 fprintf(stderr
, Name
": %s is not a member of the same container\n", devname
);
4189 /* add a pristine spare to the metadata */
4190 if (dl
->index
< 0) {
4191 dl
->index
= super
->anchor
->num_disks
;
4192 super
->anchor
->num_disks
++;
4194 /* Check the device has not already been added */
4195 slot
= get_imsm_disk_slot(map
, dl
->index
);
4197 (get_imsm_ord_tbl_ent(dev
, slot
, -1) & IMSM_ORD_REBUILD
) == 0) {
4198 fprintf(stderr
, Name
": %s has been included in this array twice\n",
4202 set_imsm_ord_tbl_ent(map
, dk
->number
, dl
->index
);
4203 dl
->disk
.status
= CONFIGURED_DISK
;
4205 /* if we are creating the first raid device update the family number */
4206 if (super
->current_vol
== 0) {
4208 struct imsm_dev
*_dev
= __get_imsm_dev(mpb
, 0);
4209 struct imsm_disk
*_disk
= __get_imsm_disk(mpb
, dl
->index
);
4211 if (!_dev
|| !_disk
) {
4212 fprintf(stderr
, Name
": BUG mpb setup error\n");
4218 sum
+= __gen_imsm_checksum(mpb
);
4219 mpb
->family_num
= __cpu_to_le32(sum
);
4220 mpb
->orig_family_num
= mpb
->family_num
;
4227 static int add_to_super_imsm(struct supertype
*st
, mdu_disk_info_t
*dk
,
4228 int fd
, char *devname
)
4230 struct intel_super
*super
= st
->sb
;
4232 unsigned long long size
;
4237 /* If we are on an RAID enabled platform check that the disk is
4238 * attached to the raid controller.
4239 * We do not need to test disks attachment for container based additions,
4240 * they shall be already tested when container was created/assembled.
4242 rv
= find_intel_hba_capability(fd
, super
, devname
);
4243 /* no orom/efi or non-intel hba of the disk */
4245 dprintf("capability: %p fd: %d ret: %d\n",
4246 super
->orom
, fd
, rv
);
4250 if (super
->current_vol
>= 0)
4251 return add_to_super_imsm_volume(st
, dk
, fd
, devname
);
4254 dd
= malloc(sizeof(*dd
));
4257 Name
": malloc failed %s:%d.\n", __func__
, __LINE__
);
4260 memset(dd
, 0, sizeof(*dd
));
4261 dd
->major
= major(stb
.st_rdev
);
4262 dd
->minor
= minor(stb
.st_rdev
);
4264 dd
->devname
= devname
? strdup(devname
) : NULL
;
4267 dd
->action
= DISK_ADD
;
4268 rv
= imsm_read_serial(fd
, devname
, dd
->serial
);
4271 Name
": failed to retrieve scsi serial, aborting\n");
4276 get_dev_size(fd
, NULL
, &size
);
4278 serialcpy(dd
->disk
.serial
, dd
->serial
);
4279 dd
->disk
.total_blocks
= __cpu_to_le32(size
);
4280 dd
->disk
.status
= SPARE_DISK
;
4281 if (sysfs_disk_to_scsi_id(fd
, &id
) == 0)
4282 dd
->disk
.scsi_id
= __cpu_to_le32(id
);
4284 dd
->disk
.scsi_id
= __cpu_to_le32(0);
4286 if (st
->update_tail
) {
4287 dd
->next
= super
->disk_mgmt_list
;
4288 super
->disk_mgmt_list
= dd
;
4290 dd
->next
= super
->disks
;
4292 super
->updates_pending
++;
4299 static int remove_from_super_imsm(struct supertype
*st
, mdu_disk_info_t
*dk
)
4301 struct intel_super
*super
= st
->sb
;
4304 /* remove from super works only in mdmon - for communication
4305 * manager - monitor. Check if communication memory buffer
4308 if (!st
->update_tail
) {
4310 Name
": %s shall be used in mdmon context only"
4311 "(line %d).\n", __func__
, __LINE__
);
4314 dd
= malloc(sizeof(*dd
));
4317 Name
": malloc failed %s:%d.\n", __func__
, __LINE__
);
4320 memset(dd
, 0, sizeof(*dd
));
4321 dd
->major
= dk
->major
;
4322 dd
->minor
= dk
->minor
;
4325 dd
->disk
.status
= SPARE_DISK
;
4326 dd
->action
= DISK_REMOVE
;
4328 dd
->next
= super
->disk_mgmt_list
;
4329 super
->disk_mgmt_list
= dd
;
4335 static int store_imsm_mpb(int fd
, struct imsm_super
*mpb
);
4339 struct imsm_super anchor
;
4340 } spare_record
__attribute__ ((aligned(512)));
4342 /* spare records have their own family number and do not have any defined raid
4345 static int write_super_imsm_spares(struct intel_super
*super
, int doclose
)
4347 struct imsm_super
*mpb
= super
->anchor
;
4348 struct imsm_super
*spare
= &spare_record
.anchor
;
4352 spare
->mpb_size
= __cpu_to_le32(sizeof(struct imsm_super
)),
4353 spare
->generation_num
= __cpu_to_le32(1UL),
4354 spare
->attributes
= MPB_ATTRIB_CHECKSUM_VERIFY
;
4355 spare
->num_disks
= 1,
4356 spare
->num_raid_devs
= 0,
4357 spare
->cache_size
= mpb
->cache_size
,
4358 spare
->pwr_cycle_count
= __cpu_to_le32(1),
4360 snprintf((char *) spare
->sig
, MAX_SIGNATURE_LENGTH
,
4361 MPB_SIGNATURE MPB_VERSION_RAID0
);
4363 for (d
= super
->disks
; d
; d
= d
->next
) {
4367 spare
->disk
[0] = d
->disk
;
4368 sum
= __gen_imsm_checksum(spare
);
4369 spare
->family_num
= __cpu_to_le32(sum
);
4370 spare
->orig_family_num
= 0;
4371 sum
= __gen_imsm_checksum(spare
);
4372 spare
->check_sum
= __cpu_to_le32(sum
);
4374 if (store_imsm_mpb(d
->fd
, spare
)) {
4375 fprintf(stderr
, "%s: failed for device %d:%d %s\n",
4376 __func__
, d
->major
, d
->minor
, strerror(errno
));
4388 static int is_gen_migration(struct imsm_dev
*dev
);
4390 static int write_super_imsm(struct supertype
*st
, int doclose
)
4392 struct intel_super
*super
= st
->sb
;
4393 struct imsm_super
*mpb
= super
->anchor
;
4399 __u32 mpb_size
= sizeof(struct imsm_super
) - sizeof(struct imsm_disk
);
4401 int clear_migration_record
= 1;
4403 /* 'generation' is incremented everytime the metadata is written */
4404 generation
= __le32_to_cpu(mpb
->generation_num
);
4406 mpb
->generation_num
= __cpu_to_le32(generation
);
4408 /* fix up cases where previous mdadm releases failed to set
4411 if (mpb
->orig_family_num
== 0)
4412 mpb
->orig_family_num
= mpb
->family_num
;
4414 for (d
= super
->disks
; d
; d
= d
->next
) {
4418 mpb
->disk
[d
->index
] = d
->disk
;
4422 for (d
= super
->missing
; d
; d
= d
->next
) {
4423 mpb
->disk
[d
->index
] = d
->disk
;
4426 mpb
->num_disks
= num_disks
;
4427 mpb_size
+= sizeof(struct imsm_disk
) * mpb
->num_disks
;
4429 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
4430 struct imsm_dev
*dev
= __get_imsm_dev(mpb
, i
);
4431 struct imsm_dev
*dev2
= get_imsm_dev(super
, i
);
4433 imsm_copy_dev(dev
, dev2
);
4434 mpb_size
+= sizeof_imsm_dev(dev
, 0);
4436 if (is_gen_migration(dev2
))
4437 clear_migration_record
= 0;
4439 mpb_size
+= __le32_to_cpu(mpb
->bbm_log_size
);
4440 mpb
->mpb_size
= __cpu_to_le32(mpb_size
);
4442 /* recalculate checksum */
4443 sum
= __gen_imsm_checksum(mpb
);
4444 mpb
->check_sum
= __cpu_to_le32(sum
);
4446 if (clear_migration_record
)
4447 memset(super
->migr_rec_buf
, 0, 512);
4449 /* write the mpb for disks that compose raid devices */
4450 for (d
= super
->disks
; d
; d
= d
->next
) {
4453 if (store_imsm_mpb(d
->fd
, mpb
))
4454 fprintf(stderr
, "%s: failed for device %d:%d %s\n",
4455 __func__
, d
->major
, d
->minor
, strerror(errno
));
4456 if (clear_migration_record
) {
4457 unsigned long long dsize
;
4459 get_dev_size(d
->fd
, NULL
, &dsize
);
4460 if (lseek64(d
->fd
, dsize
- 512, SEEK_SET
) >= 0) {
4461 write(d
->fd
, super
->migr_rec_buf
, 512);
4471 return write_super_imsm_spares(super
, doclose
);
4477 static int create_array(struct supertype
*st
, int dev_idx
)
4480 struct imsm_update_create_array
*u
;
4481 struct intel_super
*super
= st
->sb
;
4482 struct imsm_dev
*dev
= get_imsm_dev(super
, dev_idx
);
4483 struct imsm_map
*map
= get_imsm_map(dev
, 0);
4484 struct disk_info
*inf
;
4485 struct imsm_disk
*disk
;
4488 len
= sizeof(*u
) - sizeof(*dev
) + sizeof_imsm_dev(dev
, 0) +
4489 sizeof(*inf
) * map
->num_members
;
4492 fprintf(stderr
, "%s: failed to allocate update buffer\n",
4497 u
->type
= update_create_array
;
4498 u
->dev_idx
= dev_idx
;
4499 imsm_copy_dev(&u
->dev
, dev
);
4500 inf
= get_disk_info(u
);
4501 for (i
= 0; i
< map
->num_members
; i
++) {
4502 int idx
= get_imsm_disk_idx(dev
, i
, -1);
4504 disk
= get_imsm_disk(super
, idx
);
4505 serialcpy(inf
[i
].serial
, disk
->serial
);
4507 append_metadata_update(st
, u
, len
);
4512 static int mgmt_disk(struct supertype
*st
)
4514 struct intel_super
*super
= st
->sb
;
4516 struct imsm_update_add_remove_disk
*u
;
4518 if (!super
->disk_mgmt_list
)
4524 fprintf(stderr
, "%s: failed to allocate update buffer\n",
4529 u
->type
= update_add_remove_disk
;
4530 append_metadata_update(st
, u
, len
);
4535 static int write_init_super_imsm(struct supertype
*st
)
4537 struct intel_super
*super
= st
->sb
;
4538 int current_vol
= super
->current_vol
;
4540 /* we are done with current_vol reset it to point st at the container */
4541 super
->current_vol
= -1;
4543 if (st
->update_tail
) {
4544 /* queue the recently created array / added disk
4545 * as a metadata update */
4548 /* determine if we are creating a volume or adding a disk */
4549 if (current_vol
< 0) {
4550 /* in the mgmt (add/remove) disk case we are running
4551 * in mdmon context, so don't close fd's
4553 return mgmt_disk(st
);
4555 rv
= create_array(st
, current_vol
);
4560 for (d
= super
->disks
; d
; d
= d
->next
)
4561 Kill(d
->devname
, NULL
, 0, 1, 1);
4562 return write_super_imsm(st
, 1);
4567 static int store_super_imsm(struct supertype
*st
, int fd
)
4569 struct intel_super
*super
= st
->sb
;
4570 struct imsm_super
*mpb
= super
? super
->anchor
: NULL
;
4576 return store_imsm_mpb(fd
, mpb
);
4582 static int imsm_bbm_log_size(struct imsm_super
*mpb
)
4584 return __le32_to_cpu(mpb
->bbm_log_size
);
4588 static int validate_geometry_imsm_container(struct supertype
*st
, int level
,
4589 int layout
, int raiddisks
, int chunk
,
4590 unsigned long long size
, char *dev
,
4591 unsigned long long *freesize
,
4595 unsigned long long ldsize
;
4596 struct intel_super
*super
=NULL
;
4599 if (level
!= LEVEL_CONTAINER
)
4604 fd
= open(dev
, O_RDONLY
|O_EXCL
, 0);
4607 fprintf(stderr
, Name
": imsm: Cannot open %s: %s\n",
4608 dev
, strerror(errno
));
4611 if (!get_dev_size(fd
, dev
, &ldsize
)) {
4616 /* capabilities retrieve could be possible
4617 * note that there is no fd for the disks in array.
4619 super
= alloc_super();
4622 Name
": malloc of %zu failed.\n",
4628 rv
= find_intel_hba_capability(fd
, super
, verbose
? dev
: NULL
);
4632 fd2devname(fd
, str
);
4633 dprintf("validate_geometry_imsm_container: fd: %d %s orom: %p rv: %d raiddisk: %d\n",
4634 fd
, str
, super
->orom
, rv
, raiddisks
);
4636 /* no orom/efi or non-intel hba of the disk */
4642 if (super
->orom
&& raiddisks
> super
->orom
->tds
) {
4644 fprintf(stderr
, Name
": %d exceeds maximum number of"
4645 " platform supported disks: %d\n",
4646 raiddisks
, super
->orom
->tds
);
4652 *freesize
= avail_size_imsm(st
, ldsize
>> 9);
4658 static unsigned long long find_size(struct extent
*e
, int *idx
, int num_extents
)
4660 const unsigned long long base_start
= e
[*idx
].start
;
4661 unsigned long long end
= base_start
+ e
[*idx
].size
;
4664 if (base_start
== end
)
4668 for (i
= *idx
; i
< num_extents
; i
++) {
4669 /* extend overlapping extents */
4670 if (e
[i
].start
>= base_start
&&
4671 e
[i
].start
<= end
) {
4674 if (e
[i
].start
+ e
[i
].size
> end
)
4675 end
= e
[i
].start
+ e
[i
].size
;
4676 } else if (e
[i
].start
> end
) {
4682 return end
- base_start
;
4685 static unsigned long long merge_extents(struct intel_super
*super
, int sum_extents
)
4687 /* build a composite disk with all known extents and generate a new
4688 * 'maxsize' given the "all disks in an array must share a common start
4689 * offset" constraint
4691 struct extent
*e
= calloc(sum_extents
, sizeof(*e
));
4695 unsigned long long pos
;
4696 unsigned long long start
= 0;
4697 unsigned long long maxsize
;
4698 unsigned long reserve
;
4703 /* coalesce and sort all extents. also, check to see if we need to
4704 * reserve space between member arrays
4707 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
4710 for (i
= 0; i
< dl
->extent_cnt
; i
++)
4713 qsort(e
, sum_extents
, sizeof(*e
), cmp_extent
);
4718 while (i
< sum_extents
) {
4719 e
[j
].start
= e
[i
].start
;
4720 e
[j
].size
= find_size(e
, &i
, sum_extents
);
4722 if (e
[j
-1].size
== 0)
4731 unsigned long long esize
;
4733 esize
= e
[i
].start
- pos
;
4734 if (esize
>= maxsize
) {
4739 pos
= e
[i
].start
+ e
[i
].size
;
4741 } while (e
[i
-1].size
);
4747 /* FIXME assumes volume at offset 0 is the first volume in a
4750 if (start_extent
> 0)
4751 reserve
= IMSM_RESERVED_SECTORS
; /* gap between raid regions */
4755 if (maxsize
< reserve
)
4758 super
->create_offset
= ~((__u32
) 0);
4759 if (start
+ reserve
> super
->create_offset
)
4760 return 0; /* start overflows create_offset */
4761 super
->create_offset
= start
+ reserve
;
4763 return maxsize
- reserve
;
4766 static int is_raid_level_supported(const struct imsm_orom
*orom
, int level
, int raiddisks
)
4768 if (level
< 0 || level
== 6 || level
== 4)
4771 /* if we have an orom prevent invalid raid levels */
4774 case 0: return imsm_orom_has_raid0(orom
);
4777 return imsm_orom_has_raid1e(orom
);
4778 return imsm_orom_has_raid1(orom
) && raiddisks
== 2;
4779 case 10: return imsm_orom_has_raid10(orom
) && raiddisks
== 4;
4780 case 5: return imsm_orom_has_raid5(orom
) && raiddisks
> 2;
4783 return 1; /* not on an Intel RAID platform so anything goes */
4789 #define pr_vrb(fmt, arg...) (void) (verbose && fprintf(stderr, Name fmt, ##arg))
4791 * validate volume parameters with OROM/EFI capabilities
4794 validate_geometry_imsm_orom(struct intel_super
*super
, int level
, int layout
,
4795 int raiddisks
, int *chunk
, int verbose
)
4800 /* validate container capabilities */
4801 if (super
->orom
&& raiddisks
> super
->orom
->tds
) {
4803 fprintf(stderr
, Name
": %d exceeds maximum number of"
4804 " platform supported disks: %d\n",
4805 raiddisks
, super
->orom
->tds
);
4809 /* capabilities of OROM tested - copied from validate_geometry_imsm_volume */
4810 if (super
->orom
&& (!is_raid_level_supported(super
->orom
, level
,
4812 pr_vrb(": platform does not support raid%d with %d disk%s\n",
4813 level
, raiddisks
, raiddisks
> 1 ? "s" : "");
4816 if (super
->orom
&& level
!= 1) {
4817 if (chunk
&& (*chunk
== 0 || *chunk
== UnSet
))
4818 *chunk
= imsm_orom_default_chunk(super
->orom
);
4819 else if (chunk
&& !imsm_orom_has_chunk(super
->orom
, *chunk
)) {
4820 pr_vrb(": platform does not support a chunk size of: "
4825 if (layout
!= imsm_level_to_layout(level
)) {
4827 pr_vrb(": imsm raid 5 only supports the left-asymmetric layout\n");
4828 else if (level
== 10)
4829 pr_vrb(": imsm raid 10 only supports the n2 layout\n");
4831 pr_vrb(": imsm unknown layout %#x for this raid level %d\n",
4838 /* validate_geometry_imsm_volume - lifted from validate_geometry_ddf_bvd
4839 * FIX ME add ahci details
4841 static int validate_geometry_imsm_volume(struct supertype
*st
, int level
,
4842 int layout
, int raiddisks
, int *chunk
,
4843 unsigned long long size
, char *dev
,
4844 unsigned long long *freesize
,
4848 struct intel_super
*super
= st
->sb
;
4849 struct imsm_super
*mpb
= super
->anchor
;
4851 unsigned long long pos
= 0;
4852 unsigned long long maxsize
;
4856 /* We must have the container info already read in. */
4860 if (!validate_geometry_imsm_orom(super
, level
, layout
, raiddisks
, chunk
, verbose
)) {
4861 fprintf(stderr
, Name
": RAID gemetry validation failed. "
4862 "Cannot proceed with the action(s).\n");
4866 /* General test: make sure there is space for
4867 * 'raiddisks' device extents of size 'size' at a given
4870 unsigned long long minsize
= size
;
4871 unsigned long long start_offset
= MaxSector
;
4874 minsize
= MPB_SECTOR_CNT
+ IMSM_RESERVED_SECTORS
;
4875 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
4880 e
= get_extents(super
, dl
);
4883 unsigned long long esize
;
4884 esize
= e
[i
].start
- pos
;
4885 if (esize
>= minsize
)
4887 if (found
&& start_offset
== MaxSector
) {
4890 } else if (found
&& pos
!= start_offset
) {
4894 pos
= e
[i
].start
+ e
[i
].size
;
4896 } while (e
[i
-1].size
);
4901 if (dcnt
< raiddisks
) {
4903 fprintf(stderr
, Name
": imsm: Not enough "
4904 "devices with space for this array "
4912 /* This device must be a member of the set */
4913 if (stat(dev
, &stb
) < 0)
4915 if ((S_IFMT
& stb
.st_mode
) != S_IFBLK
)
4917 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
4918 if (dl
->major
== (int)major(stb
.st_rdev
) &&
4919 dl
->minor
== (int)minor(stb
.st_rdev
))
4924 fprintf(stderr
, Name
": %s is not in the "
4925 "same imsm set\n", dev
);
4927 } else if (super
->orom
&& dl
->index
< 0 && mpb
->num_raid_devs
) {
4928 /* If a volume is present then the current creation attempt
4929 * cannot incorporate new spares because the orom may not
4930 * understand this configuration (all member disks must be
4931 * members of each array in the container).
4933 fprintf(stderr
, Name
": %s is a spare and a volume"
4934 " is already defined for this container\n", dev
);
4935 fprintf(stderr
, Name
": The option-rom requires all member"
4936 " disks to be a member of all volumes\n");
4940 /* retrieve the largest free space block */
4941 e
= get_extents(super
, dl
);
4946 unsigned long long esize
;
4948 esize
= e
[i
].start
- pos
;
4949 if (esize
>= maxsize
)
4951 pos
= e
[i
].start
+ e
[i
].size
;
4953 } while (e
[i
-1].size
);
4958 fprintf(stderr
, Name
": unable to determine free space for: %s\n",
4962 if (maxsize
< size
) {
4964 fprintf(stderr
, Name
": %s not enough space (%llu < %llu)\n",
4965 dev
, maxsize
, size
);
4969 /* count total number of extents for merge */
4971 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
4973 i
+= dl
->extent_cnt
;
4975 maxsize
= merge_extents(super
, i
);
4976 if (maxsize
< size
|| maxsize
== 0) {
4978 fprintf(stderr
, Name
": not enough space after merge (%llu < %llu)\n",
4983 *freesize
= maxsize
;
4988 static int reserve_space(struct supertype
*st
, int raiddisks
,
4989 unsigned long long size
, int chunk
,
4990 unsigned long long *freesize
)
4992 struct intel_super
*super
= st
->sb
;
4993 struct imsm_super
*mpb
= super
->anchor
;
4998 unsigned long long maxsize
;
4999 unsigned long long minsize
;
5003 /* find the largest common start free region of the possible disks */
5007 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
5013 /* don't activate new spares if we are orom constrained
5014 * and there is already a volume active in the container
5016 if (super
->orom
&& dl
->index
< 0 && mpb
->num_raid_devs
)
5019 e
= get_extents(super
, dl
);
5022 for (i
= 1; e
[i
-1].size
; i
++)
5030 maxsize
= merge_extents(super
, extent_cnt
);
5034 minsize
= chunk
* 2;
5036 if (cnt
< raiddisks
||
5037 (super
->orom
&& used
&& used
!= raiddisks
) ||
5038 maxsize
< minsize
||
5040 fprintf(stderr
, Name
": not enough devices with space to create array.\n");
5041 return 0; /* No enough free spaces large enough */
5053 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
5055 dl
->raiddisk
= cnt
++;
5062 static int validate_geometry_imsm(struct supertype
*st
, int level
, int layout
,
5063 int raiddisks
, int *chunk
, unsigned long long size
,
5064 char *dev
, unsigned long long *freesize
,
5072 * if given unused devices create a container
5073 * if given given devices in a container create a member volume
5075 if (level
== LEVEL_CONTAINER
) {
5076 /* Must be a fresh device to add to a container */
5077 return validate_geometry_imsm_container(st
, level
, layout
,
5079 chunk
?*chunk
:0, size
,
5085 if (st
->sb
&& freesize
) {
5086 /* we are being asked to automatically layout a
5087 * new volume based on the current contents of
5088 * the container. If the the parameters can be
5089 * satisfied reserve_space will record the disks,
5090 * start offset, and size of the volume to be
5091 * created. add_to_super and getinfo_super
5092 * detect when autolayout is in progress.
5094 if (!validate_geometry_imsm_orom(st
->sb
, level
, layout
,
5098 return reserve_space(st
, raiddisks
, size
,
5099 chunk
?*chunk
:0, freesize
);
5104 /* creating in a given container */
5105 return validate_geometry_imsm_volume(st
, level
, layout
,
5106 raiddisks
, chunk
, size
,
5107 dev
, freesize
, verbose
);
5110 /* This device needs to be a device in an 'imsm' container */
5111 fd
= open(dev
, O_RDONLY
|O_EXCL
, 0);
5115 Name
": Cannot create this array on device %s\n",
5120 if (errno
!= EBUSY
|| (fd
= open(dev
, O_RDONLY
, 0)) < 0) {
5122 fprintf(stderr
, Name
": Cannot open %s: %s\n",
5123 dev
, strerror(errno
));
5126 /* Well, it is in use by someone, maybe an 'imsm' container. */
5127 cfd
= open_container(fd
);
5131 fprintf(stderr
, Name
": Cannot use %s: It is busy\n",
5135 sra
= sysfs_read(cfd
, 0, GET_VERSION
);
5136 if (sra
&& sra
->array
.major_version
== -1 &&
5137 strcmp(sra
->text_version
, "imsm") == 0)
5141 /* This is a member of a imsm container. Load the container
5142 * and try to create a volume
5144 struct intel_super
*super
;
5146 if (load_super_imsm_all(st
, cfd
, (void **) &super
, NULL
) == 0) {
5148 st
->container_dev
= fd2devnum(cfd
);
5150 return validate_geometry_imsm_volume(st
, level
, layout
,
5158 fprintf(stderr
, Name
": failed container membership check\n");
5164 static void default_geometry_imsm(struct supertype
*st
, int *level
, int *layout
, int *chunk
)
5166 struct intel_super
*super
= st
->sb
;
5168 if (level
&& *level
== UnSet
)
5169 *level
= LEVEL_CONTAINER
;
5171 if (level
&& layout
&& *layout
== UnSet
)
5172 *layout
= imsm_level_to_layout(*level
);
5174 if (chunk
&& (*chunk
== UnSet
|| *chunk
== 0) &&
5175 super
&& super
->orom
)
5176 *chunk
= imsm_orom_default_chunk(super
->orom
);
5179 static void handle_missing(struct intel_super
*super
, struct imsm_dev
*dev
);
5181 static int kill_subarray_imsm(struct supertype
*st
)
5183 /* remove the subarray currently referenced by ->current_vol */
5185 struct intel_dev
**dp
;
5186 struct intel_super
*super
= st
->sb
;
5187 __u8 current_vol
= super
->current_vol
;
5188 struct imsm_super
*mpb
= super
->anchor
;
5190 if (super
->current_vol
< 0)
5192 super
->current_vol
= -1; /* invalidate subarray cursor */
5194 /* block deletions that would change the uuid of active subarrays
5196 * FIXME when immutable ids are available, but note that we'll
5197 * also need to fixup the invalidated/active subarray indexes in
5200 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
5203 if (i
< current_vol
)
5205 sprintf(subarray
, "%u", i
);
5206 if (is_subarray_active(subarray
, st
->devname
)) {
5208 Name
": deleting subarray-%d would change the UUID of active subarray-%d, aborting\n",
5215 if (st
->update_tail
) {
5216 struct imsm_update_kill_array
*u
= malloc(sizeof(*u
));
5220 u
->type
= update_kill_array
;
5221 u
->dev_idx
= current_vol
;
5222 append_metadata_update(st
, u
, sizeof(*u
));
5227 for (dp
= &super
->devlist
; *dp
;)
5228 if ((*dp
)->index
== current_vol
) {
5231 handle_missing(super
, (*dp
)->dev
);
5232 if ((*dp
)->index
> current_vol
)
5237 /* no more raid devices, all active components are now spares,
5238 * but of course failed are still failed
5240 if (--mpb
->num_raid_devs
== 0) {
5243 for (d
= super
->disks
; d
; d
= d
->next
)
5244 if (d
->index
> -2) {
5246 d
->disk
.status
= SPARE_DISK
;
5250 super
->updates_pending
++;
5255 static int update_subarray_imsm(struct supertype
*st
, char *subarray
,
5256 char *update
, struct mddev_ident
*ident
)
5258 /* update the subarray currently referenced by ->current_vol */
5259 struct intel_super
*super
= st
->sb
;
5260 struct imsm_super
*mpb
= super
->anchor
;
5262 if (strcmp(update
, "name") == 0) {
5263 char *name
= ident
->name
;
5267 if (is_subarray_active(subarray
, st
->devname
)) {
5269 Name
": Unable to update name of active subarray\n");
5273 if (!check_name(super
, name
, 0))
5276 vol
= strtoul(subarray
, &ep
, 10);
5277 if (*ep
!= '\0' || vol
>= super
->anchor
->num_raid_devs
)
5280 if (st
->update_tail
) {
5281 struct imsm_update_rename_array
*u
= malloc(sizeof(*u
));
5285 u
->type
= update_rename_array
;
5287 snprintf((char *) u
->name
, MAX_RAID_SERIAL_LEN
, "%s", name
);
5288 append_metadata_update(st
, u
, sizeof(*u
));
5290 struct imsm_dev
*dev
;
5293 dev
= get_imsm_dev(super
, vol
);
5294 snprintf((char *) dev
->volume
, MAX_RAID_SERIAL_LEN
, "%s", name
);
5295 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
5296 dev
= get_imsm_dev(super
, i
);
5297 handle_missing(super
, dev
);
5299 super
->updates_pending
++;
5307 static int is_gen_migration(struct imsm_dev
*dev
)
5309 if (!dev
->vol
.migr_state
)
5312 if (migr_type(dev
) == MIGR_GEN_MIGR
)
5317 #endif /* MDASSEMBLE */
5319 static int is_rebuilding(struct imsm_dev
*dev
)
5321 struct imsm_map
*migr_map
;
5323 if (!dev
->vol
.migr_state
)
5326 if (migr_type(dev
) != MIGR_REBUILD
)
5329 migr_map
= get_imsm_map(dev
, 1);
5331 if (migr_map
->map_state
== IMSM_T_STATE_DEGRADED
)
5337 static void update_recovery_start(struct intel_super
*super
,
5338 struct imsm_dev
*dev
,
5339 struct mdinfo
*array
)
5341 struct mdinfo
*rebuild
= NULL
;
5345 if (!is_rebuilding(dev
))
5348 /* Find the rebuild target, but punt on the dual rebuild case */
5349 for (d
= array
->devs
; d
; d
= d
->next
)
5350 if (d
->recovery_start
== 0) {
5357 /* (?) none of the disks are marked with
5358 * IMSM_ORD_REBUILD, so assume they are missing and the
5359 * disk_ord_tbl was not correctly updated
5361 dprintf("%s: failed to locate out-of-sync disk\n", __func__
);
5365 units
= __le32_to_cpu(dev
->vol
.curr_migr_unit
);
5366 rebuild
->recovery_start
= units
* blocks_per_migr_unit(super
, dev
);
5369 static int recover_backup_imsm(struct supertype
*st
, struct mdinfo
*info
);
5371 static struct mdinfo
*container_content_imsm(struct supertype
*st
, char *subarray
)
5373 /* Given a container loaded by load_super_imsm_all,
5374 * extract information about all the arrays into
5376 * If 'subarray' is given, just extract info about that array.
5378 * For each imsm_dev create an mdinfo, fill it in,
5379 * then look for matching devices in super->disks
5380 * and create appropriate device mdinfo.
5382 struct intel_super
*super
= st
->sb
;
5383 struct imsm_super
*mpb
= super
->anchor
;
5384 struct mdinfo
*rest
= NULL
;
5388 int spare_disks
= 0;
5390 /* check for bad blocks */
5391 if (imsm_bbm_log_size(super
->anchor
))
5394 /* count spare devices, not used in maps
5396 for (d
= super
->disks
; d
; d
= d
->next
)
5400 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
5401 struct imsm_dev
*dev
;
5402 struct imsm_map
*map
;
5403 struct imsm_map
*map2
;
5404 struct mdinfo
*this;
5409 (i
!= strtoul(subarray
, &ep
, 10) || *ep
!= '\0'))
5412 dev
= get_imsm_dev(super
, i
);
5413 map
= get_imsm_map(dev
, 0);
5414 map2
= get_imsm_map(dev
, 1);
5416 /* do not publish arrays that are in the middle of an
5417 * unsupported migration
5419 if (dev
->vol
.migr_state
&&
5420 (migr_type(dev
) == MIGR_STATE_CHANGE
)) {
5421 fprintf(stderr
, Name
": cannot assemble volume '%.16s':"
5422 " unsupported migration in progress\n",
5426 /* do not publish arrays that are not support by controller's
5430 chunk
= __le16_to_cpu(map
->blocks_per_strip
) >> 1;
5432 if (!validate_geometry_imsm_orom(super
,
5433 get_imsm_raid_level(map
), /* RAID level */
5434 imsm_level_to_layout(get_imsm_raid_level(map
)),
5435 map
->num_members
, /* raid disks */
5438 fprintf(stderr
, Name
": RAID gemetry validation failed. "
5439 "Cannot proceed with the action(s).\n");
5442 #endif /* MDASSEMBLE */
5443 this = malloc(sizeof(*this));
5445 fprintf(stderr
, Name
": failed to allocate %zu bytes\n",
5450 super
->current_vol
= i
;
5451 getinfo_super_imsm_volume(st
, this, NULL
);
5453 for (slot
= 0 ; slot
< map
->num_members
; slot
++) {
5454 unsigned long long recovery_start
;
5455 struct mdinfo
*info_d
;
5462 idx
= get_imsm_disk_idx(dev
, slot
, 0);
5463 ord
= get_imsm_ord_tbl_ent(dev
, slot
, -1);
5464 for (d
= super
->disks
; d
; d
= d
->next
)
5465 if (d
->index
== idx
)
5468 recovery_start
= MaxSector
;
5471 if (d
&& is_failed(&d
->disk
))
5473 if (ord
& IMSM_ORD_REBUILD
)
5477 * if we skip some disks the array will be assmebled degraded;
5478 * reset resync start to avoid a dirty-degraded
5479 * situation when performing the intial sync
5481 * FIXME handle dirty degraded
5483 if ((skip
|| recovery_start
== 0) && !dev
->vol
.dirty
)
5484 this->resync_start
= MaxSector
;
5488 info_d
= calloc(1, sizeof(*info_d
));
5490 fprintf(stderr
, Name
": failed to allocate disk"
5491 " for volume %.16s\n", dev
->volume
);
5492 info_d
= this->devs
;
5494 struct mdinfo
*d
= info_d
->next
;
5503 info_d
->next
= this->devs
;
5504 this->devs
= info_d
;
5506 info_d
->disk
.number
= d
->index
;
5507 info_d
->disk
.major
= d
->major
;
5508 info_d
->disk
.minor
= d
->minor
;
5509 info_d
->disk
.raid_disk
= slot
;
5510 info_d
->recovery_start
= recovery_start
;
5512 if (slot
< map2
->num_members
)
5513 info_d
->disk
.state
= (1 << MD_DISK_ACTIVE
);
5515 this->array
.spare_disks
++;
5517 if (slot
< map
->num_members
)
5518 info_d
->disk
.state
= (1 << MD_DISK_ACTIVE
);
5520 this->array
.spare_disks
++;
5522 if (info_d
->recovery_start
== MaxSector
)
5523 this->array
.working_disks
++;
5525 info_d
->events
= __le32_to_cpu(mpb
->generation_num
);
5526 info_d
->data_offset
= __le32_to_cpu(map
->pba_of_lba0
);
5527 info_d
->component_size
= __le32_to_cpu(map
->blocks_per_member
);
5529 /* now that the disk list is up-to-date fixup recovery_start */
5530 update_recovery_start(super
, dev
, this);
5531 this->array
.spare_disks
+= spare_disks
;
5533 /* check for reshape */
5534 if (this->reshape_active
== 1)
5535 recover_backup_imsm(st
, this);
5540 /* if array has bad blocks, set suitable bit in array status */
5542 rest
->array
.state
|= (1<<MD_SB_BBM_ERRORS
);
5548 static __u8
imsm_check_degraded(struct intel_super
*super
, struct imsm_dev
*dev
, int failed
)
5550 struct imsm_map
*map
= get_imsm_map(dev
, 0);
5553 return map
->map_state
== IMSM_T_STATE_UNINITIALIZED
?
5554 IMSM_T_STATE_UNINITIALIZED
: IMSM_T_STATE_NORMAL
;
5556 switch (get_imsm_raid_level(map
)) {
5558 return IMSM_T_STATE_FAILED
;
5561 if (failed
< map
->num_members
)
5562 return IMSM_T_STATE_DEGRADED
;
5564 return IMSM_T_STATE_FAILED
;
5569 * check to see if any mirrors have failed, otherwise we
5570 * are degraded. Even numbered slots are mirrored on
5574 /* gcc -Os complains that this is unused */
5575 int insync
= insync
;
5577 for (i
= 0; i
< map
->num_members
; i
++) {
5578 __u32 ord
= get_imsm_ord_tbl_ent(dev
, i
, -1);
5579 int idx
= ord_to_idx(ord
);
5580 struct imsm_disk
*disk
;
5582 /* reset the potential in-sync count on even-numbered
5583 * slots. num_copies is always 2 for imsm raid10
5588 disk
= get_imsm_disk(super
, idx
);
5589 if (!disk
|| is_failed(disk
) || ord
& IMSM_ORD_REBUILD
)
5592 /* no in-sync disks left in this mirror the
5596 return IMSM_T_STATE_FAILED
;
5599 return IMSM_T_STATE_DEGRADED
;
5603 return IMSM_T_STATE_DEGRADED
;
5605 return IMSM_T_STATE_FAILED
;
5611 return map
->map_state
;
5614 static int imsm_count_failed(struct intel_super
*super
, struct imsm_dev
*dev
)
5618 struct imsm_disk
*disk
;
5619 struct imsm_map
*map
= get_imsm_map(dev
, 0);
5620 struct imsm_map
*prev
= get_imsm_map(dev
, dev
->vol
.migr_state
);
5624 /* at the beginning of migration we set IMSM_ORD_REBUILD on
5625 * disks that are being rebuilt. New failures are recorded to
5626 * map[0]. So we look through all the disks we started with and
5627 * see if any failures are still present, or if any new ones
5630 * FIXME add support for online capacity expansion and
5631 * raid-level-migration
5633 for (i
= 0; i
< prev
->num_members
; i
++) {
5634 ord
= __le32_to_cpu(prev
->disk_ord_tbl
[i
]);
5635 ord
|= __le32_to_cpu(map
->disk_ord_tbl
[i
]);
5636 idx
= ord_to_idx(ord
);
5638 disk
= get_imsm_disk(super
, idx
);
5639 if (!disk
|| is_failed(disk
) || ord
& IMSM_ORD_REBUILD
)
5647 static int imsm_open_new(struct supertype
*c
, struct active_array
*a
,
5650 struct intel_super
*super
= c
->sb
;
5651 struct imsm_super
*mpb
= super
->anchor
;
5653 if (atoi(inst
) >= mpb
->num_raid_devs
) {
5654 fprintf(stderr
, "%s: subarry index %d, out of range\n",
5655 __func__
, atoi(inst
));
5659 dprintf("imsm: open_new %s\n", inst
);
5660 a
->info
.container_member
= atoi(inst
);
5664 static int is_resyncing(struct imsm_dev
*dev
)
5666 struct imsm_map
*migr_map
;
5668 if (!dev
->vol
.migr_state
)
5671 if (migr_type(dev
) == MIGR_INIT
||
5672 migr_type(dev
) == MIGR_REPAIR
)
5675 if (migr_type(dev
) == MIGR_GEN_MIGR
)
5678 migr_map
= get_imsm_map(dev
, 1);
5680 if ((migr_map
->map_state
== IMSM_T_STATE_NORMAL
) &&
5681 (dev
->vol
.migr_type
!= MIGR_GEN_MIGR
))
5687 /* return true if we recorded new information */
5688 static int mark_failure(struct imsm_dev
*dev
, struct imsm_disk
*disk
, int idx
)
5692 struct imsm_map
*map
;
5694 /* new failures are always set in map[0] */
5695 map
= get_imsm_map(dev
, 0);
5697 slot
= get_imsm_disk_slot(map
, idx
);
5701 ord
= __le32_to_cpu(map
->disk_ord_tbl
[slot
]);
5702 if (is_failed(disk
) && (ord
& IMSM_ORD_REBUILD
))
5705 disk
->status
|= FAILED_DISK
;
5706 set_imsm_ord_tbl_ent(map
, slot
, idx
| IMSM_ORD_REBUILD
);
5707 if (map
->failed_disk_num
== 0xff)
5708 map
->failed_disk_num
= slot
;
5712 static void mark_missing(struct imsm_dev
*dev
, struct imsm_disk
*disk
, int idx
)
5714 mark_failure(dev
, disk
, idx
);
5716 if (disk
->scsi_id
== __cpu_to_le32(~(__u32
)0))
5719 disk
->scsi_id
= __cpu_to_le32(~(__u32
)0);
5720 memmove(&disk
->serial
[0], &disk
->serial
[1], MAX_RAID_SERIAL_LEN
- 1);
5723 static void handle_missing(struct intel_super
*super
, struct imsm_dev
*dev
)
5729 if (!super
->missing
)
5731 failed
= imsm_count_failed(super
, dev
);
5732 map_state
= imsm_check_degraded(super
, dev
, failed
);
5734 dprintf("imsm: mark missing\n");
5735 end_migration(dev
, map_state
);
5736 for (dl
= super
->missing
; dl
; dl
= dl
->next
)
5737 mark_missing(dev
, &dl
->disk
, dl
->index
);
5738 super
->updates_pending
++;
5741 static unsigned long long imsm_set_array_size(struct imsm_dev
*dev
)
5743 int used_disks
= imsm_num_data_members(dev
, 0);
5744 unsigned long long array_blocks
;
5745 struct imsm_map
*map
;
5747 if (used_disks
== 0) {
5748 /* when problems occures
5749 * return current array_blocks value
5751 array_blocks
= __le32_to_cpu(dev
->size_high
);
5752 array_blocks
= array_blocks
<< 32;
5753 array_blocks
+= __le32_to_cpu(dev
->size_low
);
5755 return array_blocks
;
5758 /* set array size in metadata
5760 map
= get_imsm_map(dev
, 0);
5761 array_blocks
= map
->blocks_per_member
* used_disks
;
5763 /* round array size down to closest MB
5765 array_blocks
= (array_blocks
>> SECT_PER_MB_SHIFT
) << SECT_PER_MB_SHIFT
;
5766 dev
->size_low
= __cpu_to_le32((__u32
)array_blocks
);
5767 dev
->size_high
= __cpu_to_le32((__u32
)(array_blocks
>> 32));
5769 return array_blocks
;
5772 static void imsm_set_disk(struct active_array
*a
, int n
, int state
);
5774 static void imsm_progress_container_reshape(struct intel_super
*super
)
5776 /* if no device has a migr_state, but some device has a
5777 * different number of members than the previous device, start
5778 * changing the number of devices in this device to match
5781 struct imsm_super
*mpb
= super
->anchor
;
5782 int prev_disks
= -1;
5786 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
5787 struct imsm_dev
*dev
= get_imsm_dev(super
, i
);
5788 struct imsm_map
*map
= get_imsm_map(dev
, 0);
5789 struct imsm_map
*map2
;
5790 int prev_num_members
;
5792 if (dev
->vol
.migr_state
)
5795 if (prev_disks
== -1)
5796 prev_disks
= map
->num_members
;
5797 if (prev_disks
== map
->num_members
)
5800 /* OK, this array needs to enter reshape mode.
5801 * i.e it needs a migr_state
5804 copy_map_size
= sizeof_imsm_map(map
);
5805 prev_num_members
= map
->num_members
;
5806 map
->num_members
= prev_disks
;
5807 dev
->vol
.migr_state
= 1;
5808 dev
->vol
.curr_migr_unit
= 0;
5809 dev
->vol
.migr_type
= MIGR_GEN_MIGR
;
5810 for (i
= prev_num_members
;
5811 i
< map
->num_members
; i
++)
5812 set_imsm_ord_tbl_ent(map
, i
, i
);
5813 map2
= get_imsm_map(dev
, 1);
5814 /* Copy the current map */
5815 memcpy(map2
, map
, copy_map_size
);
5816 map2
->num_members
= prev_num_members
;
5818 imsm_set_array_size(dev
);
5819 super
->updates_pending
++;
5823 /* Handle dirty -> clean transititions, resync and reshape. Degraded and rebuild
5824 * states are handled in imsm_set_disk() with one exception, when a
5825 * resync is stopped due to a new failure this routine will set the
5826 * 'degraded' state for the array.
5828 static int imsm_set_array_state(struct active_array
*a
, int consistent
)
5830 int inst
= a
->info
.container_member
;
5831 struct intel_super
*super
= a
->container
->sb
;
5832 struct imsm_dev
*dev
= get_imsm_dev(super
, inst
);
5833 struct imsm_map
*map
= get_imsm_map(dev
, 0);
5834 int failed
= imsm_count_failed(super
, dev
);
5835 __u8 map_state
= imsm_check_degraded(super
, dev
, failed
);
5836 __u32 blocks_per_unit
;
5838 if (dev
->vol
.migr_state
&&
5839 dev
->vol
.migr_type
== MIGR_GEN_MIGR
) {
5840 /* array state change is blocked due to reshape action
5842 * - abort the reshape (if last_checkpoint is 0 and action!= reshape)
5843 * - finish the reshape (if last_checkpoint is big and action != reshape)
5844 * - update curr_migr_unit
5846 if (a
->curr_action
== reshape
) {
5847 /* still reshaping, maybe update curr_migr_unit */
5848 goto mark_checkpoint
;
5850 if (a
->last_checkpoint
== 0 && a
->prev_action
== reshape
) {
5851 /* for some reason we aborted the reshape.
5854 struct imsm_map
*map2
= get_imsm_map(dev
, 1);
5855 dev
->vol
.migr_state
= 0;
5856 dev
->vol
.migr_type
= 0;
5857 dev
->vol
.curr_migr_unit
= 0;
5858 memcpy(map
, map2
, sizeof_imsm_map(map2
));
5859 super
->updates_pending
++;
5861 if (a
->last_checkpoint
>= a
->info
.component_size
) {
5862 unsigned long long array_blocks
;
5866 used_disks
= imsm_num_data_members(dev
, 0);
5867 if (used_disks
> 0) {
5869 map
->blocks_per_member
*
5871 /* round array size down to closest MB
5873 array_blocks
= (array_blocks
5874 >> SECT_PER_MB_SHIFT
)
5875 << SECT_PER_MB_SHIFT
;
5876 a
->info
.custom_array_size
= array_blocks
;
5877 /* encourage manager to update array
5881 a
->check_reshape
= 1;
5883 /* finalize online capacity expansion/reshape */
5884 for (mdi
= a
->info
.devs
; mdi
; mdi
= mdi
->next
)
5886 mdi
->disk
.raid_disk
,
5889 imsm_progress_container_reshape(super
);
5894 /* before we activate this array handle any missing disks */
5895 if (consistent
== 2)
5896 handle_missing(super
, dev
);
5898 if (consistent
== 2 &&
5899 (!is_resync_complete(&a
->info
) ||
5900 map_state
!= IMSM_T_STATE_NORMAL
||
5901 dev
->vol
.migr_state
))
5904 if (is_resync_complete(&a
->info
)) {
5905 /* complete intialization / resync,
5906 * recovery and interrupted recovery is completed in
5909 if (is_resyncing(dev
)) {
5910 dprintf("imsm: mark resync done\n");
5911 end_migration(dev
, map_state
);
5912 super
->updates_pending
++;
5913 a
->last_checkpoint
= 0;
5915 } else if (!is_resyncing(dev
) && !failed
) {
5916 /* mark the start of the init process if nothing is failed */
5917 dprintf("imsm: mark resync start\n");
5918 if (map
->map_state
== IMSM_T_STATE_UNINITIALIZED
)
5919 migrate(dev
, super
, IMSM_T_STATE_NORMAL
, MIGR_INIT
);
5921 migrate(dev
, super
, IMSM_T_STATE_NORMAL
, MIGR_REPAIR
);
5922 super
->updates_pending
++;
5926 /* skip checkpointing for general migration,
5927 * it is controlled in mdadm
5929 if (is_gen_migration(dev
))
5930 goto skip_mark_checkpoint
;
5932 /* check if we can update curr_migr_unit from resync_start, recovery_start */
5933 blocks_per_unit
= blocks_per_migr_unit(super
, dev
);
5934 if (blocks_per_unit
) {
5938 units
= a
->last_checkpoint
/ blocks_per_unit
;
5941 /* check that we did not overflow 32-bits, and that
5942 * curr_migr_unit needs updating
5944 if (units32
== units
&&
5946 __le32_to_cpu(dev
->vol
.curr_migr_unit
) != units32
) {
5947 dprintf("imsm: mark checkpoint (%u)\n", units32
);
5948 dev
->vol
.curr_migr_unit
= __cpu_to_le32(units32
);
5949 super
->updates_pending
++;
5953 skip_mark_checkpoint
:
5954 /* mark dirty / clean */
5955 if (dev
->vol
.dirty
!= !consistent
) {
5956 dprintf("imsm: mark '%s'\n", consistent
? "clean" : "dirty");
5961 super
->updates_pending
++;
5967 static void imsm_set_disk(struct active_array
*a
, int n
, int state
)
5969 int inst
= a
->info
.container_member
;
5970 struct intel_super
*super
= a
->container
->sb
;
5971 struct imsm_dev
*dev
= get_imsm_dev(super
, inst
);
5972 struct imsm_map
*map
= get_imsm_map(dev
, 0);
5973 struct imsm_disk
*disk
;
5978 if (n
> map
->num_members
)
5979 fprintf(stderr
, "imsm: set_disk %d out of range 0..%d\n",
5980 n
, map
->num_members
- 1);
5985 dprintf("imsm: set_disk %d:%x\n", n
, state
);
5987 ord
= get_imsm_ord_tbl_ent(dev
, n
, -1);
5988 disk
= get_imsm_disk(super
, ord_to_idx(ord
));
5990 /* check for new failures */
5991 if (state
& DS_FAULTY
) {
5992 if (mark_failure(dev
, disk
, ord_to_idx(ord
)))
5993 super
->updates_pending
++;
5996 /* check if in_sync */
5997 if (state
& DS_INSYNC
&& ord
& IMSM_ORD_REBUILD
&& is_rebuilding(dev
)) {
5998 struct imsm_map
*migr_map
= get_imsm_map(dev
, 1);
6000 set_imsm_ord_tbl_ent(migr_map
, n
, ord_to_idx(ord
));
6001 super
->updates_pending
++;
6004 failed
= imsm_count_failed(super
, dev
);
6005 map_state
= imsm_check_degraded(super
, dev
, failed
);
6007 /* check if recovery complete, newly degraded, or failed */
6008 if (map_state
== IMSM_T_STATE_NORMAL
&& is_rebuilding(dev
)) {
6009 end_migration(dev
, map_state
);
6010 map
= get_imsm_map(dev
, 0);
6011 map
->failed_disk_num
= ~0;
6012 super
->updates_pending
++;
6013 a
->last_checkpoint
= 0;
6014 } else if (map_state
== IMSM_T_STATE_DEGRADED
&&
6015 map
->map_state
!= map_state
&&
6016 !dev
->vol
.migr_state
) {
6017 dprintf("imsm: mark degraded\n");
6018 map
->map_state
= map_state
;
6019 super
->updates_pending
++;
6020 a
->last_checkpoint
= 0;
6021 } else if (map_state
== IMSM_T_STATE_FAILED
&&
6022 map
->map_state
!= map_state
) {
6023 dprintf("imsm: mark failed\n");
6024 end_migration(dev
, map_state
);
6025 super
->updates_pending
++;
6026 a
->last_checkpoint
= 0;
6027 } else if (is_gen_migration(dev
)) {
6028 dprintf("imsm: Detected General Migration in state: ");
6029 if (map_state
== IMSM_T_STATE_NORMAL
) {
6030 end_migration(dev
, map_state
);
6031 map
= get_imsm_map(dev
, 0);
6032 map
->failed_disk_num
= ~0;
6033 dprintf("normal\n");
6035 if (map_state
== IMSM_T_STATE_DEGRADED
) {
6036 printf("degraded\n");
6037 end_migration(dev
, map_state
);
6039 dprintf("failed\n");
6041 map
->map_state
= map_state
;
6043 super
->updates_pending
++;
6047 static int store_imsm_mpb(int fd
, struct imsm_super
*mpb
)
6050 __u32 mpb_size
= __le32_to_cpu(mpb
->mpb_size
);
6051 unsigned long long dsize
;
6052 unsigned long long sectors
;
6054 get_dev_size(fd
, NULL
, &dsize
);
6056 if (mpb_size
> 512) {
6057 /* -1 to account for anchor */
6058 sectors
= mpb_sectors(mpb
) - 1;
6060 /* write the extended mpb to the sectors preceeding the anchor */
6061 if (lseek64(fd
, dsize
- (512 * (2 + sectors
)), SEEK_SET
) < 0)
6064 if ((unsigned long long)write(fd
, buf
+ 512, 512 * sectors
)
6069 /* first block is stored on second to last sector of the disk */
6070 if (lseek64(fd
, dsize
- (512 * 2), SEEK_SET
) < 0)
6073 if (write(fd
, buf
, 512) != 512)
6079 static void imsm_sync_metadata(struct supertype
*container
)
6081 struct intel_super
*super
= container
->sb
;
6083 dprintf("sync metadata: %d\n", super
->updates_pending
);
6084 if (!super
->updates_pending
)
6087 write_super_imsm(container
, 0);
6089 super
->updates_pending
= 0;
6092 static struct dl
*imsm_readd(struct intel_super
*super
, int idx
, struct active_array
*a
)
6094 struct imsm_dev
*dev
= get_imsm_dev(super
, a
->info
.container_member
);
6095 int i
= get_imsm_disk_idx(dev
, idx
, -1);
6098 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
6102 if (dl
&& is_failed(&dl
->disk
))
6106 dprintf("%s: found %x:%x\n", __func__
, dl
->major
, dl
->minor
);
6111 static struct dl
*imsm_add_spare(struct intel_super
*super
, int slot
,
6112 struct active_array
*a
, int activate_new
,
6113 struct mdinfo
*additional_test_list
)
6115 struct imsm_dev
*dev
= get_imsm_dev(super
, a
->info
.container_member
);
6116 int idx
= get_imsm_disk_idx(dev
, slot
, -1);
6117 struct imsm_super
*mpb
= super
->anchor
;
6118 struct imsm_map
*map
;
6119 unsigned long long pos
;
6124 __u32 array_start
= 0;
6125 __u32 array_end
= 0;
6127 struct mdinfo
*test_list
;
6129 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
6130 /* If in this array, skip */
6131 for (d
= a
->info
.devs
; d
; d
= d
->next
)
6132 if (d
->state_fd
>= 0 &&
6133 d
->disk
.major
== dl
->major
&&
6134 d
->disk
.minor
== dl
->minor
) {
6135 dprintf("%x:%x already in array\n",
6136 dl
->major
, dl
->minor
);
6141 test_list
= additional_test_list
;
6143 if (test_list
->disk
.major
== dl
->major
&&
6144 test_list
->disk
.minor
== dl
->minor
) {
6145 dprintf("%x:%x already in additional test list\n",
6146 dl
->major
, dl
->minor
);
6149 test_list
= test_list
->next
;
6154 /* skip in use or failed drives */
6155 if (is_failed(&dl
->disk
) || idx
== dl
->index
||
6157 dprintf("%x:%x status (failed: %d index: %d)\n",
6158 dl
->major
, dl
->minor
, is_failed(&dl
->disk
), idx
);
6162 /* skip pure spares when we are looking for partially
6163 * assimilated drives
6165 if (dl
->index
== -1 && !activate_new
)
6168 /* Does this unused device have the requisite free space?
6169 * It needs to be able to cover all member volumes
6171 ex
= get_extents(super
, dl
);
6173 dprintf("cannot get extents\n");
6176 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
6177 dev
= get_imsm_dev(super
, i
);
6178 map
= get_imsm_map(dev
, 0);
6180 /* check if this disk is already a member of
6183 if (get_imsm_disk_slot(map
, dl
->index
) >= 0)
6189 array_start
= __le32_to_cpu(map
->pba_of_lba0
);
6190 array_end
= array_start
+
6191 __le32_to_cpu(map
->blocks_per_member
) - 1;
6194 /* check that we can start at pba_of_lba0 with
6195 * blocks_per_member of space
6197 if (array_start
>= pos
&& array_end
< ex
[j
].start
) {
6201 pos
= ex
[j
].start
+ ex
[j
].size
;
6203 } while (ex
[j
-1].size
);
6210 if (i
< mpb
->num_raid_devs
) {
6211 dprintf("%x:%x does not have %u to %u available\n",
6212 dl
->major
, dl
->minor
, array_start
, array_end
);
6223 static int imsm_rebuild_allowed(struct supertype
*cont
, int dev_idx
, int failed
)
6225 struct imsm_dev
*dev2
;
6226 struct imsm_map
*map
;
6232 dev2
= get_imsm_dev(cont
->sb
, dev_idx
);
6234 state
= imsm_check_degraded(cont
->sb
, dev2
, failed
);
6235 if (state
== IMSM_T_STATE_FAILED
) {
6236 map
= get_imsm_map(dev2
, 0);
6239 for (slot
= 0; slot
< map
->num_members
; slot
++) {
6241 * Check if failed disks are deleted from intel
6242 * disk list or are marked to be deleted
6244 idx
= get_imsm_disk_idx(dev2
, slot
, -1);
6245 idisk
= get_imsm_dl_disk(cont
->sb
, idx
);
6247 * Do not rebuild the array if failed disks
6248 * from failed sub-array are not removed from
6252 is_failed(&idisk
->disk
) &&
6253 (idisk
->action
!= DISK_REMOVE
))
6261 static struct mdinfo
*imsm_activate_spare(struct active_array
*a
,
6262 struct metadata_update
**updates
)
6265 * Find a device with unused free space and use it to replace a
6266 * failed/vacant region in an array. We replace failed regions one a
6267 * array at a time. The result is that a new spare disk will be added
6268 * to the first failed array and after the monitor has finished
6269 * propagating failures the remainder will be consumed.
6271 * FIXME add a capability for mdmon to request spares from another
6275 struct intel_super
*super
= a
->container
->sb
;
6276 int inst
= a
->info
.container_member
;
6277 struct imsm_dev
*dev
= get_imsm_dev(super
, inst
);
6278 struct imsm_map
*map
= get_imsm_map(dev
, 0);
6279 int failed
= a
->info
.array
.raid_disks
;
6280 struct mdinfo
*rv
= NULL
;
6283 struct metadata_update
*mu
;
6285 struct imsm_update_activate_spare
*u
;
6290 for (d
= a
->info
.devs
; d
; d
= d
->next
) {
6291 if ((d
->curr_state
& DS_FAULTY
) &&
6293 /* wait for Removal to happen */
6295 if (d
->state_fd
>= 0)
6299 dprintf("imsm: activate spare: inst=%d failed=%d (%d) level=%d\n",
6300 inst
, failed
, a
->info
.array
.raid_disks
, a
->info
.array
.level
);
6302 if (dev
->vol
.migr_state
&&
6303 dev
->vol
.migr_type
== MIGR_GEN_MIGR
)
6304 /* No repair during migration */
6307 if (a
->info
.array
.level
== 4)
6308 /* No repair for takeovered array
6309 * imsm doesn't support raid4
6313 if (imsm_check_degraded(super
, dev
, failed
) != IMSM_T_STATE_DEGRADED
)
6317 * If there are any failed disks check state of the other volume.
6318 * Block rebuild if the another one is failed until failed disks
6319 * are removed from container.
6322 dprintf("found failed disks in %s, check if there another"
6323 "failed sub-array.\n",
6325 /* check if states of the other volumes allow for rebuild */
6326 for (i
= 0; i
< super
->anchor
->num_raid_devs
; i
++) {
6328 allowed
= imsm_rebuild_allowed(a
->container
,
6336 /* For each slot, if it is not working, find a spare */
6337 for (i
= 0; i
< a
->info
.array
.raid_disks
; i
++) {
6338 for (d
= a
->info
.devs
; d
; d
= d
->next
)
6339 if (d
->disk
.raid_disk
== i
)
6341 dprintf("found %d: %p %x\n", i
, d
, d
?d
->curr_state
:0);
6342 if (d
&& (d
->state_fd
>= 0))
6346 * OK, this device needs recovery. Try to re-add the
6347 * previous occupant of this slot, if this fails see if
6348 * we can continue the assimilation of a spare that was
6349 * partially assimilated, finally try to activate a new
6352 dl
= imsm_readd(super
, i
, a
);
6354 dl
= imsm_add_spare(super
, i
, a
, 0, NULL
);
6356 dl
= imsm_add_spare(super
, i
, a
, 1, NULL
);
6360 /* found a usable disk with enough space */
6361 di
= malloc(sizeof(*di
));
6364 memset(di
, 0, sizeof(*di
));
6366 /* dl->index will be -1 in the case we are activating a
6367 * pristine spare. imsm_process_update() will create a
6368 * new index in this case. Once a disk is found to be
6369 * failed in all member arrays it is kicked from the
6372 di
->disk
.number
= dl
->index
;
6374 /* (ab)use di->devs to store a pointer to the device
6377 di
->devs
= (struct mdinfo
*) dl
;
6379 di
->disk
.raid_disk
= i
;
6380 di
->disk
.major
= dl
->major
;
6381 di
->disk
.minor
= dl
->minor
;
6383 di
->recovery_start
= 0;
6384 di
->data_offset
= __le32_to_cpu(map
->pba_of_lba0
);
6385 di
->component_size
= a
->info
.component_size
;
6386 di
->container_member
= inst
;
6387 super
->random
= random32();
6391 dprintf("%x:%x to be %d at %llu\n", dl
->major
, dl
->minor
,
6392 i
, di
->data_offset
);
6398 /* No spares found */
6400 /* Now 'rv' has a list of devices to return.
6401 * Create a metadata_update record to update the
6402 * disk_ord_tbl for the array
6404 mu
= malloc(sizeof(*mu
));
6406 mu
->buf
= malloc(sizeof(struct imsm_update_activate_spare
) * num_spares
);
6407 if (mu
->buf
== NULL
) {
6414 struct mdinfo
*n
= rv
->next
;
6423 mu
->space_list
= NULL
;
6424 mu
->len
= sizeof(struct imsm_update_activate_spare
) * num_spares
;
6425 mu
->next
= *updates
;
6426 u
= (struct imsm_update_activate_spare
*) mu
->buf
;
6428 for (di
= rv
; di
; di
= di
->next
) {
6429 u
->type
= update_activate_spare
;
6430 u
->dl
= (struct dl
*) di
->devs
;
6432 u
->slot
= di
->disk
.raid_disk
;
6443 static int disks_overlap(struct intel_super
*super
, int idx
, struct imsm_update_create_array
*u
)
6445 struct imsm_dev
*dev
= get_imsm_dev(super
, idx
);
6446 struct imsm_map
*map
= get_imsm_map(dev
, 0);
6447 struct imsm_map
*new_map
= get_imsm_map(&u
->dev
, 0);
6448 struct disk_info
*inf
= get_disk_info(u
);
6449 struct imsm_disk
*disk
;
6453 for (i
= 0; i
< map
->num_members
; i
++) {
6454 disk
= get_imsm_disk(super
, get_imsm_disk_idx(dev
, i
, -1));
6455 for (j
= 0; j
< new_map
->num_members
; j
++)
6456 if (serialcmp(disk
->serial
, inf
[j
].serial
) == 0)
6464 static struct dl
*get_disk_super(struct intel_super
*super
, int major
, int minor
)
6466 struct dl
*dl
= NULL
;
6467 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
6468 if ((dl
->major
== major
) && (dl
->minor
== minor
))
6473 static int remove_disk_super(struct intel_super
*super
, int major
, int minor
)
6475 struct dl
*prev
= NULL
;
6479 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
6480 if ((dl
->major
== major
) && (dl
->minor
== minor
)) {
6483 prev
->next
= dl
->next
;
6485 super
->disks
= dl
->next
;
6487 __free_imsm_disk(dl
);
6488 dprintf("%s: removed %x:%x\n",
6489 __func__
, major
, minor
);
6497 static void imsm_delete(struct intel_super
*super
, struct dl
**dlp
, unsigned index
);
6499 static int add_remove_disk_update(struct intel_super
*super
)
6501 int check_degraded
= 0;
6502 struct dl
*disk
= NULL
;
6503 /* add/remove some spares to/from the metadata/contrainer */
6504 while (super
->disk_mgmt_list
) {
6505 struct dl
*disk_cfg
;
6507 disk_cfg
= super
->disk_mgmt_list
;
6508 super
->disk_mgmt_list
= disk_cfg
->next
;
6509 disk_cfg
->next
= NULL
;
6511 if (disk_cfg
->action
== DISK_ADD
) {
6512 disk_cfg
->next
= super
->disks
;
6513 super
->disks
= disk_cfg
;
6515 dprintf("%s: added %x:%x\n",
6516 __func__
, disk_cfg
->major
,
6518 } else if (disk_cfg
->action
== DISK_REMOVE
) {
6519 dprintf("Disk remove action processed: %x.%x\n",
6520 disk_cfg
->major
, disk_cfg
->minor
);
6521 disk
= get_disk_super(super
,
6525 /* store action status */
6526 disk
->action
= DISK_REMOVE
;
6527 /* remove spare disks only */
6528 if (disk
->index
== -1) {
6529 remove_disk_super(super
,
6534 /* release allocate disk structure */
6535 __free_imsm_disk(disk_cfg
);
6538 return check_degraded
;
6542 static int apply_reshape_migration_update(struct imsm_update_reshape_migration
*u
,
6543 struct intel_super
*super
,
6546 struct intel_dev
*id
;
6547 void **tofree
= NULL
;
6550 dprintf("apply_reshape_migration_update()\n");
6551 if ((u
->subdev
< 0) ||
6553 dprintf("imsm: Error: Wrong subdev: %i\n", u
->subdev
);
6556 if ((space_list
== NULL
) || (*space_list
== NULL
)) {
6557 dprintf("imsm: Error: Memory is not allocated\n");
6561 for (id
= super
->devlist
; id
; id
= id
->next
) {
6562 if (id
->index
== (unsigned)u
->subdev
) {
6563 struct imsm_dev
*dev
= get_imsm_dev(super
, u
->subdev
);
6564 struct imsm_map
*map
;
6565 struct imsm_dev
*new_dev
=
6566 (struct imsm_dev
*)*space_list
;
6567 struct imsm_map
*migr_map
= get_imsm_map(dev
, 1);
6569 struct dl
*new_disk
;
6571 if (new_dev
== NULL
)
6573 *space_list
= **space_list
;
6574 memcpy(new_dev
, dev
, sizeof_imsm_dev(dev
, 0));
6575 map
= get_imsm_map(new_dev
, 0);
6577 dprintf("imsm: Error: migration in progress");
6581 to_state
= map
->map_state
;
6582 if ((u
->new_level
== 5) && (map
->raid_level
== 0)) {
6584 /* this should not happen */
6585 if (u
->new_disks
[0] < 0) {
6586 map
->failed_disk_num
=
6587 map
->num_members
- 1;
6588 to_state
= IMSM_T_STATE_DEGRADED
;
6590 to_state
= IMSM_T_STATE_NORMAL
;
6592 migrate(new_dev
, super
, to_state
, MIGR_GEN_MIGR
);
6593 if (u
->new_level
> -1)
6594 map
->raid_level
= u
->new_level
;
6595 migr_map
= get_imsm_map(new_dev
, 1);
6596 if ((u
->new_level
== 5) &&
6597 (migr_map
->raid_level
== 0)) {
6598 int ord
= map
->num_members
- 1;
6599 migr_map
->num_members
--;
6600 if (u
->new_disks
[0] < 0)
6601 ord
|= IMSM_ORD_REBUILD
;
6602 set_imsm_ord_tbl_ent(map
,
6603 map
->num_members
- 1,
6607 tofree
= (void **)dev
;
6609 /* update chunk size
6611 if (u
->new_chunksize
> 0)
6612 map
->blocks_per_strip
=
6613 __cpu_to_le16(u
->new_chunksize
* 2);
6617 if ((u
->new_level
!= 5) ||
6618 (migr_map
->raid_level
!= 0) ||
6619 (migr_map
->raid_level
== map
->raid_level
))
6622 if (u
->new_disks
[0] >= 0) {
6625 new_disk
= get_disk_super(super
,
6626 major(u
->new_disks
[0]),
6627 minor(u
->new_disks
[0]));
6628 dprintf("imsm: new disk for reshape is: %i:%i "
6629 "(%p, index = %i)\n",
6630 major(u
->new_disks
[0]),
6631 minor(u
->new_disks
[0]),
6632 new_disk
, new_disk
->index
);
6633 if (new_disk
== NULL
)
6634 goto error_disk_add
;
6636 new_disk
->index
= map
->num_members
- 1;
6637 /* slot to fill in autolayout
6639 new_disk
->raiddisk
= new_disk
->index
;
6640 new_disk
->disk
.status
|= CONFIGURED_DISK
;
6641 new_disk
->disk
.status
&= ~SPARE_DISK
;
6643 goto error_disk_add
;
6646 *tofree
= *space_list
;
6647 /* calculate new size
6649 imsm_set_array_size(new_dev
);
6656 *space_list
= tofree
;
6660 dprintf("Error: imsm: Cannot find disk.\n");
6665 static int apply_reshape_container_disks_update(struct imsm_update_reshape
*u
,
6666 struct intel_super
*super
,
6669 struct dl
*new_disk
;
6670 struct intel_dev
*id
;
6672 int delta_disks
= u
->new_raid_disks
- u
->old_raid_disks
;
6673 int disk_count
= u
->old_raid_disks
;
6674 void **tofree
= NULL
;
6675 int devices_to_reshape
= 1;
6676 struct imsm_super
*mpb
= super
->anchor
;
6678 unsigned int dev_id
;
6680 dprintf("imsm: apply_reshape_container_disks_update()\n");
6682 /* enable spares to use in array */
6683 for (i
= 0; i
< delta_disks
; i
++) {
6684 new_disk
= get_disk_super(super
,
6685 major(u
->new_disks
[i
]),
6686 minor(u
->new_disks
[i
]));
6687 dprintf("imsm: new disk for reshape is: %i:%i "
6688 "(%p, index = %i)\n",
6689 major(u
->new_disks
[i
]), minor(u
->new_disks
[i
]),
6690 new_disk
, new_disk
->index
);
6691 if ((new_disk
== NULL
) ||
6692 ((new_disk
->index
>= 0) &&
6693 (new_disk
->index
< u
->old_raid_disks
)))
6694 goto update_reshape_exit
;
6695 new_disk
->index
= disk_count
++;
6696 /* slot to fill in autolayout
6698 new_disk
->raiddisk
= new_disk
->index
;
6699 new_disk
->disk
.status
|=
6701 new_disk
->disk
.status
&= ~SPARE_DISK
;
6704 dprintf("imsm: volume set mpb->num_raid_devs = %i\n",
6705 mpb
->num_raid_devs
);
6706 /* manage changes in volume
6708 for (dev_id
= 0; dev_id
< mpb
->num_raid_devs
; dev_id
++) {
6709 void **sp
= *space_list
;
6710 struct imsm_dev
*newdev
;
6711 struct imsm_map
*newmap
, *oldmap
;
6713 for (id
= super
->devlist
; id
; id
= id
->next
) {
6714 if (id
->index
== dev_id
)
6723 /* Copy the dev, but not (all of) the map */
6724 memcpy(newdev
, id
->dev
, sizeof(*newdev
));
6725 oldmap
= get_imsm_map(id
->dev
, 0);
6726 newmap
= get_imsm_map(newdev
, 0);
6727 /* Copy the current map */
6728 memcpy(newmap
, oldmap
, sizeof_imsm_map(oldmap
));
6729 /* update one device only
6731 if (devices_to_reshape
) {
6732 dprintf("imsm: modifying subdev: %i\n",
6734 devices_to_reshape
--;
6735 newdev
->vol
.migr_state
= 1;
6736 newdev
->vol
.curr_migr_unit
= 0;
6737 newdev
->vol
.migr_type
= MIGR_GEN_MIGR
;
6738 newmap
->num_members
= u
->new_raid_disks
;
6739 for (i
= 0; i
< delta_disks
; i
++) {
6740 set_imsm_ord_tbl_ent(newmap
,
6741 u
->old_raid_disks
+ i
,
6742 u
->old_raid_disks
+ i
);
6744 /* New map is correct, now need to save old map
6746 newmap
= get_imsm_map(newdev
, 1);
6747 memcpy(newmap
, oldmap
, sizeof_imsm_map(oldmap
));
6749 imsm_set_array_size(newdev
);
6752 sp
= (void **)id
->dev
;
6757 /* Clear migration record */
6758 memset(super
->migr_rec
, 0, sizeof(struct migr_record
));
6761 *space_list
= tofree
;
6764 update_reshape_exit
:
6769 static int apply_takeover_update(struct imsm_update_takeover
*u
,
6770 struct intel_super
*super
,
6773 struct imsm_dev
*dev
= NULL
;
6774 struct intel_dev
*dv
;
6775 struct imsm_dev
*dev_new
;
6776 struct imsm_map
*map
;
6780 for (dv
= super
->devlist
; dv
; dv
= dv
->next
)
6781 if (dv
->index
== (unsigned int)u
->subarray
) {
6789 map
= get_imsm_map(dev
, 0);
6791 if (u
->direction
== R10_TO_R0
) {
6792 /* Number of failed disks must be half of initial disk number */
6793 if (imsm_count_failed(super
, dev
) != (map
->num_members
/ 2))
6796 /* iterate through devices to mark removed disks as spare */
6797 for (dm
= super
->disks
; dm
; dm
= dm
->next
) {
6798 if (dm
->disk
.status
& FAILED_DISK
) {
6799 int idx
= dm
->index
;
6800 /* update indexes on the disk list */
6801 /* FIXME this loop-with-the-loop looks wrong, I'm not convinced
6802 the index values will end up being correct.... NB */
6803 for (du
= super
->disks
; du
; du
= du
->next
)
6804 if (du
->index
> idx
)
6806 /* mark as spare disk */
6807 dm
->disk
.status
= SPARE_DISK
;
6812 map
->num_members
= map
->num_members
/ 2;
6813 map
->map_state
= IMSM_T_STATE_NORMAL
;
6814 map
->num_domains
= 1;
6815 map
->raid_level
= 0;
6816 map
->failed_disk_num
= -1;
6819 if (u
->direction
== R0_TO_R10
) {
6821 /* update slots in current disk list */
6822 for (dm
= super
->disks
; dm
; dm
= dm
->next
) {
6826 /* create new *missing* disks */
6827 for (i
= 0; i
< map
->num_members
; i
++) {
6828 space
= *space_list
;
6831 *space_list
= *space
;
6833 memcpy(du
, super
->disks
, sizeof(*du
));
6837 du
->index
= (i
* 2) + 1;
6838 sprintf((char *)du
->disk
.serial
,
6839 " MISSING_%d", du
->index
);
6840 sprintf((char *)du
->serial
,
6841 "MISSING_%d", du
->index
);
6842 du
->next
= super
->missing
;
6843 super
->missing
= du
;
6845 /* create new dev and map */
6846 space
= *space_list
;
6849 *space_list
= *space
;
6850 dev_new
= (void *)space
;
6851 memcpy(dev_new
, dev
, sizeof(*dev
));
6852 /* update new map */
6853 map
= get_imsm_map(dev_new
, 0);
6854 map
->num_members
= map
->num_members
* 2;
6855 map
->map_state
= IMSM_T_STATE_DEGRADED
;
6856 map
->num_domains
= 2;
6857 map
->raid_level
= 1;
6858 /* replace dev<->dev_new */
6861 /* update disk order table */
6862 for (du
= super
->disks
; du
; du
= du
->next
)
6864 set_imsm_ord_tbl_ent(map
, du
->index
, du
->index
);
6865 for (du
= super
->missing
; du
; du
= du
->next
)
6866 if (du
->index
>= 0) {
6867 set_imsm_ord_tbl_ent(map
, du
->index
, du
->index
);
6868 mark_missing(dev_new
, &du
->disk
, du
->index
);
6874 static void imsm_process_update(struct supertype
*st
,
6875 struct metadata_update
*update
)
6878 * crack open the metadata_update envelope to find the update record
6879 * update can be one of:
6880 * update_reshape_container_disks - all the arrays in the container
6881 * are being reshaped to have more devices. We need to mark
6882 * the arrays for general migration and convert selected spares
6883 * into active devices.
6884 * update_activate_spare - a spare device has replaced a failed
6885 * device in an array, update the disk_ord_tbl. If this disk is
6886 * present in all member arrays then also clear the SPARE_DISK
6888 * update_create_array
6890 * update_rename_array
6891 * update_add_remove_disk
6893 struct intel_super
*super
= st
->sb
;
6894 struct imsm_super
*mpb
;
6895 enum imsm_update_type type
= *(enum imsm_update_type
*) update
->buf
;
6897 /* update requires a larger buf but the allocation failed */
6898 if (super
->next_len
&& !super
->next_buf
) {
6899 super
->next_len
= 0;
6903 if (super
->next_buf
) {
6904 memcpy(super
->next_buf
, super
->buf
, super
->len
);
6906 super
->len
= super
->next_len
;
6907 super
->buf
= super
->next_buf
;
6909 super
->next_len
= 0;
6910 super
->next_buf
= NULL
;
6913 mpb
= super
->anchor
;
6916 case update_general_migration_checkpoint
: {
6917 struct intel_dev
*id
;
6918 struct imsm_update_general_migration_checkpoint
*u
=
6919 (void *)update
->buf
;
6921 dprintf("imsm: process_update() "
6922 "for update_general_migration_checkpoint called\n");
6924 /* find device under general migration */
6925 for (id
= super
->devlist
; id
; id
= id
->next
) {
6926 if (is_gen_migration(id
->dev
)) {
6927 id
->dev
->vol
.curr_migr_unit
=
6928 __cpu_to_le32(u
->curr_migr_unit
);
6929 super
->updates_pending
++;
6934 case update_takeover
: {
6935 struct imsm_update_takeover
*u
= (void *)update
->buf
;
6936 if (apply_takeover_update(u
, super
, &update
->space_list
)) {
6937 imsm_update_version_info(super
);
6938 super
->updates_pending
++;
6943 case update_reshape_container_disks
: {
6944 struct imsm_update_reshape
*u
= (void *)update
->buf
;
6945 if (apply_reshape_container_disks_update(
6946 u
, super
, &update
->space_list
))
6947 super
->updates_pending
++;
6950 case update_reshape_migration
: {
6951 struct imsm_update_reshape_migration
*u
= (void *)update
->buf
;
6952 if (apply_reshape_migration_update(
6953 u
, super
, &update
->space_list
))
6954 super
->updates_pending
++;
6957 case update_activate_spare
: {
6958 struct imsm_update_activate_spare
*u
= (void *) update
->buf
;
6959 struct imsm_dev
*dev
= get_imsm_dev(super
, u
->array
);
6960 struct imsm_map
*map
= get_imsm_map(dev
, 0);
6961 struct imsm_map
*migr_map
;
6962 struct active_array
*a
;
6963 struct imsm_disk
*disk
;
6968 int victim
= get_imsm_disk_idx(dev
, u
->slot
, -1);
6971 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
6976 fprintf(stderr
, "error: imsm_activate_spare passed "
6977 "an unknown disk (index: %d)\n",
6982 super
->updates_pending
++;
6983 /* count failures (excluding rebuilds and the victim)
6984 * to determine map[0] state
6987 for (i
= 0; i
< map
->num_members
; i
++) {
6990 disk
= get_imsm_disk(super
,
6991 get_imsm_disk_idx(dev
, i
, -1));
6992 if (!disk
|| is_failed(disk
))
6996 /* adding a pristine spare, assign a new index */
6997 if (dl
->index
< 0) {
6998 dl
->index
= super
->anchor
->num_disks
;
6999 super
->anchor
->num_disks
++;
7002 disk
->status
|= CONFIGURED_DISK
;
7003 disk
->status
&= ~SPARE_DISK
;
7006 to_state
= imsm_check_degraded(super
, dev
, failed
);
7007 map
->map_state
= IMSM_T_STATE_DEGRADED
;
7008 migrate(dev
, super
, to_state
, MIGR_REBUILD
);
7009 migr_map
= get_imsm_map(dev
, 1);
7010 set_imsm_ord_tbl_ent(map
, u
->slot
, dl
->index
);
7011 set_imsm_ord_tbl_ent(migr_map
, u
->slot
, dl
->index
| IMSM_ORD_REBUILD
);
7013 /* update the family_num to mark a new container
7014 * generation, being careful to record the existing
7015 * family_num in orig_family_num to clean up after
7016 * earlier mdadm versions that neglected to set it.
7018 if (mpb
->orig_family_num
== 0)
7019 mpb
->orig_family_num
= mpb
->family_num
;
7020 mpb
->family_num
+= super
->random
;
7022 /* count arrays using the victim in the metadata */
7024 for (a
= st
->arrays
; a
; a
= a
->next
) {
7025 dev
= get_imsm_dev(super
, a
->info
.container_member
);
7026 map
= get_imsm_map(dev
, 0);
7028 if (get_imsm_disk_slot(map
, victim
) >= 0)
7032 /* delete the victim if it is no longer being
7038 /* We know that 'manager' isn't touching anything,
7039 * so it is safe to delete
7041 for (dlp
= &super
->disks
; *dlp
; dlp
= &(*dlp
)->next
)
7042 if ((*dlp
)->index
== victim
)
7045 /* victim may be on the missing list */
7047 for (dlp
= &super
->missing
; *dlp
; dlp
= &(*dlp
)->next
)
7048 if ((*dlp
)->index
== victim
)
7050 imsm_delete(super
, dlp
, victim
);
7054 case update_create_array
: {
7055 /* someone wants to create a new array, we need to be aware of
7056 * a few races/collisions:
7057 * 1/ 'Create' called by two separate instances of mdadm
7058 * 2/ 'Create' versus 'activate_spare': mdadm has chosen
7059 * devices that have since been assimilated via
7061 * In the event this update can not be carried out mdadm will
7062 * (FIX ME) notice that its update did not take hold.
7064 struct imsm_update_create_array
*u
= (void *) update
->buf
;
7065 struct intel_dev
*dv
;
7066 struct imsm_dev
*dev
;
7067 struct imsm_map
*map
, *new_map
;
7068 unsigned long long start
, end
;
7069 unsigned long long new_start
, new_end
;
7071 struct disk_info
*inf
;
7074 /* handle racing creates: first come first serve */
7075 if (u
->dev_idx
< mpb
->num_raid_devs
) {
7076 dprintf("%s: subarray %d already defined\n",
7077 __func__
, u
->dev_idx
);
7081 /* check update is next in sequence */
7082 if (u
->dev_idx
!= mpb
->num_raid_devs
) {
7083 dprintf("%s: can not create array %d expected index %d\n",
7084 __func__
, u
->dev_idx
, mpb
->num_raid_devs
);
7088 new_map
= get_imsm_map(&u
->dev
, 0);
7089 new_start
= __le32_to_cpu(new_map
->pba_of_lba0
);
7090 new_end
= new_start
+ __le32_to_cpu(new_map
->blocks_per_member
);
7091 inf
= get_disk_info(u
);
7093 /* handle activate_spare versus create race:
7094 * check to make sure that overlapping arrays do not include
7097 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
7098 dev
= get_imsm_dev(super
, i
);
7099 map
= get_imsm_map(dev
, 0);
7100 start
= __le32_to_cpu(map
->pba_of_lba0
);
7101 end
= start
+ __le32_to_cpu(map
->blocks_per_member
);
7102 if ((new_start
>= start
&& new_start
<= end
) ||
7103 (start
>= new_start
&& start
<= new_end
))
7108 if (disks_overlap(super
, i
, u
)) {
7109 dprintf("%s: arrays overlap\n", __func__
);
7114 /* check that prepare update was successful */
7115 if (!update
->space
) {
7116 dprintf("%s: prepare update failed\n", __func__
);
7120 /* check that all disks are still active before committing
7121 * changes. FIXME: could we instead handle this by creating a
7122 * degraded array? That's probably not what the user expects,
7123 * so better to drop this update on the floor.
7125 for (i
= 0; i
< new_map
->num_members
; i
++) {
7126 dl
= serial_to_dl(inf
[i
].serial
, super
);
7128 dprintf("%s: disk disappeared\n", __func__
);
7133 super
->updates_pending
++;
7135 /* convert spares to members and fixup ord_tbl */
7136 for (i
= 0; i
< new_map
->num_members
; i
++) {
7137 dl
= serial_to_dl(inf
[i
].serial
, super
);
7138 if (dl
->index
== -1) {
7139 dl
->index
= mpb
->num_disks
;
7141 dl
->disk
.status
|= CONFIGURED_DISK
;
7142 dl
->disk
.status
&= ~SPARE_DISK
;
7144 set_imsm_ord_tbl_ent(new_map
, i
, dl
->index
);
7149 update
->space
= NULL
;
7150 imsm_copy_dev(dev
, &u
->dev
);
7151 dv
->index
= u
->dev_idx
;
7152 dv
->next
= super
->devlist
;
7153 super
->devlist
= dv
;
7154 mpb
->num_raid_devs
++;
7156 imsm_update_version_info(super
);
7159 /* mdmon knows how to release update->space, but not
7160 * ((struct intel_dev *) update->space)->dev
7162 if (update
->space
) {
7168 case update_kill_array
: {
7169 struct imsm_update_kill_array
*u
= (void *) update
->buf
;
7170 int victim
= u
->dev_idx
;
7171 struct active_array
*a
;
7172 struct intel_dev
**dp
;
7173 struct imsm_dev
*dev
;
7175 /* sanity check that we are not affecting the uuid of
7176 * active arrays, or deleting an active array
7178 * FIXME when immutable ids are available, but note that
7179 * we'll also need to fixup the invalidated/active
7180 * subarray indexes in mdstat
7182 for (a
= st
->arrays
; a
; a
= a
->next
)
7183 if (a
->info
.container_member
>= victim
)
7185 /* by definition if mdmon is running at least one array
7186 * is active in the container, so checking
7187 * mpb->num_raid_devs is just extra paranoia
7189 dev
= get_imsm_dev(super
, victim
);
7190 if (a
|| !dev
|| mpb
->num_raid_devs
== 1) {
7191 dprintf("failed to delete subarray-%d\n", victim
);
7195 for (dp
= &super
->devlist
; *dp
;)
7196 if ((*dp
)->index
== (unsigned)super
->current_vol
) {
7199 if ((*dp
)->index
> (unsigned)victim
)
7203 mpb
->num_raid_devs
--;
7204 super
->updates_pending
++;
7207 case update_rename_array
: {
7208 struct imsm_update_rename_array
*u
= (void *) update
->buf
;
7209 char name
[MAX_RAID_SERIAL_LEN
+1];
7210 int target
= u
->dev_idx
;
7211 struct active_array
*a
;
7212 struct imsm_dev
*dev
;
7214 /* sanity check that we are not affecting the uuid of
7217 snprintf(name
, MAX_RAID_SERIAL_LEN
, "%s", (char *) u
->name
);
7218 name
[MAX_RAID_SERIAL_LEN
] = '\0';
7219 for (a
= st
->arrays
; a
; a
= a
->next
)
7220 if (a
->info
.container_member
== target
)
7222 dev
= get_imsm_dev(super
, u
->dev_idx
);
7223 if (a
|| !dev
|| !check_name(super
, name
, 1)) {
7224 dprintf("failed to rename subarray-%d\n", target
);
7228 snprintf((char *) dev
->volume
, MAX_RAID_SERIAL_LEN
, "%s", name
);
7229 super
->updates_pending
++;
7232 case update_add_remove_disk
: {
7233 /* we may be able to repair some arrays if disks are
7234 * being added, check teh status of add_remove_disk
7235 * if discs has been added.
7237 if (add_remove_disk_update(super
)) {
7238 struct active_array
*a
;
7240 super
->updates_pending
++;
7241 for (a
= st
->arrays
; a
; a
= a
->next
)
7242 a
->check_degraded
= 1;
7247 fprintf(stderr
, "error: unsuported process update type:"
7248 "(type: %d)\n", type
);
7252 static struct mdinfo
*get_spares_for_grow(struct supertype
*st
);
7254 static void imsm_prepare_update(struct supertype
*st
,
7255 struct metadata_update
*update
)
7258 * Allocate space to hold new disk entries, raid-device entries or a new
7259 * mpb if necessary. The manager synchronously waits for updates to
7260 * complete in the monitor, so new mpb buffers allocated here can be
7261 * integrated by the monitor thread without worrying about live pointers
7262 * in the manager thread.
7264 enum imsm_update_type type
= *(enum imsm_update_type
*) update
->buf
;
7265 struct intel_super
*super
= st
->sb
;
7266 struct imsm_super
*mpb
= super
->anchor
;
7271 case update_general_migration_checkpoint
:
7272 dprintf("imsm: prepare_update() "
7273 "for update_general_migration_checkpoint called\n");
7275 case update_takeover
: {
7276 struct imsm_update_takeover
*u
= (void *)update
->buf
;
7277 if (u
->direction
== R0_TO_R10
) {
7278 void **tail
= (void **)&update
->space_list
;
7279 struct imsm_dev
*dev
= get_imsm_dev(super
, u
->subarray
);
7280 struct imsm_map
*map
= get_imsm_map(dev
, 0);
7281 int num_members
= map
->num_members
;
7285 /* allocate memory for added disks */
7286 for (i
= 0; i
< num_members
; i
++) {
7287 size
= sizeof(struct dl
);
7288 space
= malloc(size
);
7297 /* allocate memory for new device */
7298 size
= sizeof_imsm_dev(super
->devlist
->dev
, 0) +
7299 (num_members
* sizeof(__u32
));
7300 space
= malloc(size
);
7309 len
= disks_to_mpb_size(num_members
* 2);
7311 /* if allocation didn't success, free buffer */
7312 while (update
->space_list
) {
7313 void **sp
= update
->space_list
;
7314 update
->space_list
= *sp
;
7322 case update_reshape_container_disks
: {
7323 /* Every raid device in the container is about to
7324 * gain some more devices, and we will enter a
7326 * So each 'imsm_map' will be bigger, and the imsm_vol
7327 * will now hold 2 of them.
7328 * Thus we need new 'struct imsm_dev' allocations sized
7329 * as sizeof_imsm_dev but with more devices in both maps.
7331 struct imsm_update_reshape
*u
= (void *)update
->buf
;
7332 struct intel_dev
*dl
;
7333 void **space_tail
= (void**)&update
->space_list
;
7335 dprintf("imsm: imsm_prepare_update() for update_reshape\n");
7337 for (dl
= super
->devlist
; dl
; dl
= dl
->next
) {
7338 int size
= sizeof_imsm_dev(dl
->dev
, 1);
7340 if (u
->new_raid_disks
> u
->old_raid_disks
)
7341 size
+= sizeof(__u32
)*2*
7342 (u
->new_raid_disks
- u
->old_raid_disks
);
7351 len
= disks_to_mpb_size(u
->new_raid_disks
);
7352 dprintf("New anchor length is %llu\n", (unsigned long long)len
);
7355 case update_reshape_migration
: {
7356 /* for migration level 0->5 we need to add disks
7357 * so the same as for container operation we will copy
7358 * device to the bigger location.
7359 * in memory prepared device and new disk area are prepared
7360 * for usage in process update
7362 struct imsm_update_reshape_migration
*u
= (void *)update
->buf
;
7363 struct intel_dev
*id
;
7364 void **space_tail
= (void **)&update
->space_list
;
7367 int current_level
= -1;
7369 dprintf("imsm: imsm_prepare_update() for update_reshape\n");
7371 /* add space for bigger array in update
7373 for (id
= super
->devlist
; id
; id
= id
->next
) {
7374 if (id
->index
== (unsigned)u
->subdev
) {
7375 size
= sizeof_imsm_dev(id
->dev
, 1);
7376 if (u
->new_raid_disks
> u
->old_raid_disks
)
7377 size
+= sizeof(__u32
)*2*
7378 (u
->new_raid_disks
- u
->old_raid_disks
);
7388 if (update
->space_list
== NULL
)
7391 /* add space for disk in update
7393 size
= sizeof(struct dl
);
7396 free(update
->space_list
);
7397 update
->space_list
= NULL
;
7404 /* add spare device to update
7406 for (id
= super
->devlist
; id
; id
= id
->next
)
7407 if (id
->index
== (unsigned)u
->subdev
) {
7408 struct imsm_dev
*dev
;
7409 struct imsm_map
*map
;
7411 dev
= get_imsm_dev(super
, u
->subdev
);
7412 map
= get_imsm_map(dev
, 0);
7413 current_level
= map
->raid_level
;
7416 if ((u
->new_level
== 5) && (u
->new_level
!= current_level
)) {
7417 struct mdinfo
*spares
;
7419 spares
= get_spares_for_grow(st
);
7427 makedev(dev
->disk
.major
,
7429 dl
= get_disk_super(super
,
7432 dl
->index
= u
->old_raid_disks
;
7438 len
= disks_to_mpb_size(u
->new_raid_disks
);
7439 dprintf("New anchor length is %llu\n", (unsigned long long)len
);
7442 case update_create_array
: {
7443 struct imsm_update_create_array
*u
= (void *) update
->buf
;
7444 struct intel_dev
*dv
;
7445 struct imsm_dev
*dev
= &u
->dev
;
7446 struct imsm_map
*map
= get_imsm_map(dev
, 0);
7448 struct disk_info
*inf
;
7452 inf
= get_disk_info(u
);
7453 len
= sizeof_imsm_dev(dev
, 1);
7454 /* allocate a new super->devlist entry */
7455 dv
= malloc(sizeof(*dv
));
7457 dv
->dev
= malloc(len
);
7462 update
->space
= NULL
;
7466 /* count how many spares will be converted to members */
7467 for (i
= 0; i
< map
->num_members
; i
++) {
7468 dl
= serial_to_dl(inf
[i
].serial
, super
);
7470 /* hmm maybe it failed?, nothing we can do about
7475 if (count_memberships(dl
, super
) == 0)
7478 len
+= activate
* sizeof(struct imsm_disk
);
7485 /* check if we need a larger metadata buffer */
7486 if (super
->next_buf
)
7487 buf_len
= super
->next_len
;
7489 buf_len
= super
->len
;
7491 if (__le32_to_cpu(mpb
->mpb_size
) + len
> buf_len
) {
7492 /* ok we need a larger buf than what is currently allocated
7493 * if this allocation fails process_update will notice that
7494 * ->next_len is set and ->next_buf is NULL
7496 buf_len
= ROUND_UP(__le32_to_cpu(mpb
->mpb_size
) + len
, 512);
7497 if (super
->next_buf
)
7498 free(super
->next_buf
);
7500 super
->next_len
= buf_len
;
7501 if (posix_memalign(&super
->next_buf
, 512, buf_len
) == 0)
7502 memset(super
->next_buf
, 0, buf_len
);
7504 super
->next_buf
= NULL
;
7508 /* must be called while manager is quiesced */
7509 static void imsm_delete(struct intel_super
*super
, struct dl
**dlp
, unsigned index
)
7511 struct imsm_super
*mpb
= super
->anchor
;
7513 struct imsm_dev
*dev
;
7514 struct imsm_map
*map
;
7515 int i
, j
, num_members
;
7518 dprintf("%s: deleting device[%d] from imsm_super\n",
7521 /* shift all indexes down one */
7522 for (iter
= super
->disks
; iter
; iter
= iter
->next
)
7523 if (iter
->index
> (int)index
)
7525 for (iter
= super
->missing
; iter
; iter
= iter
->next
)
7526 if (iter
->index
> (int)index
)
7529 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
7530 dev
= get_imsm_dev(super
, i
);
7531 map
= get_imsm_map(dev
, 0);
7532 num_members
= map
->num_members
;
7533 for (j
= 0; j
< num_members
; j
++) {
7534 /* update ord entries being careful not to propagate
7535 * ord-flags to the first map
7537 ord
= get_imsm_ord_tbl_ent(dev
, j
, -1);
7539 if (ord_to_idx(ord
) <= index
)
7542 map
= get_imsm_map(dev
, 0);
7543 set_imsm_ord_tbl_ent(map
, j
, ord_to_idx(ord
- 1));
7544 map
= get_imsm_map(dev
, 1);
7546 set_imsm_ord_tbl_ent(map
, j
, ord
- 1);
7551 super
->updates_pending
++;
7553 struct dl
*dl
= *dlp
;
7555 *dlp
= (*dlp
)->next
;
7556 __free_imsm_disk(dl
);
7560 /*******************************************************************************
7561 * Function: open_backup_targets
7562 * Description: Function opens file descriptors for all devices given in
7565 * info : general array info
7566 * raid_disks : number of disks
7567 * raid_fds : table of device's file descriptors
7571 ******************************************************************************/
7572 int open_backup_targets(struct mdinfo
*info
, int raid_disks
, int *raid_fds
)
7576 for (sd
= info
->devs
; sd
; sd
= sd
->next
) {
7579 if (sd
->disk
.state
& (1<<MD_DISK_FAULTY
)) {
7580 dprintf("disk is faulty!!\n");
7584 if ((sd
->disk
.raid_disk
>= raid_disks
) ||
7585 (sd
->disk
.raid_disk
< 0))
7588 dn
= map_dev(sd
->disk
.major
,
7590 raid_fds
[sd
->disk
.raid_disk
] = dev_open(dn
, O_RDWR
);
7591 if (raid_fds
[sd
->disk
.raid_disk
] < 0) {
7592 fprintf(stderr
, "cannot open component\n");
7599 /*******************************************************************************
7600 * Function: init_migr_record_imsm
7601 * Description: Function inits imsm migration record
7603 * super : imsm internal array info
7604 * dev : device under migration
7605 * info : general array info to find the smallest device
7608 ******************************************************************************/
7609 void init_migr_record_imsm(struct supertype
*st
, struct imsm_dev
*dev
,
7610 struct mdinfo
*info
)
7612 struct intel_super
*super
= st
->sb
;
7613 struct migr_record
*migr_rec
= super
->migr_rec
;
7615 unsigned long long dsize
, dev_sectors
;
7616 long long unsigned min_dev_sectors
= -1LLU;
7620 struct imsm_map
*map_dest
= get_imsm_map(dev
, 0);
7621 struct imsm_map
*map_src
= get_imsm_map(dev
, 1);
7622 unsigned long long num_migr_units
;
7624 unsigned long long array_blocks
=
7625 (((unsigned long long)__le32_to_cpu(dev
->size_high
)) << 32) +
7626 __le32_to_cpu(dev
->size_low
);
7628 memset(migr_rec
, 0, sizeof(struct migr_record
));
7629 migr_rec
->family_num
= __cpu_to_le32(super
->anchor
->family_num
);
7631 /* only ascending reshape supported now */
7632 migr_rec
->ascending_migr
= __cpu_to_le32(1);
7634 migr_rec
->dest_depth_per_unit
= GEN_MIGR_AREA_SIZE
/
7635 max(map_dest
->blocks_per_strip
, map_src
->blocks_per_strip
);
7636 migr_rec
->dest_depth_per_unit
*= map_dest
->blocks_per_strip
;
7637 new_data_disks
= imsm_num_data_members(dev
, 0);
7638 migr_rec
->blocks_per_unit
=
7639 __cpu_to_le32(migr_rec
->dest_depth_per_unit
* new_data_disks
);
7640 migr_rec
->dest_depth_per_unit
=
7641 __cpu_to_le32(migr_rec
->dest_depth_per_unit
);
7644 array_blocks
/ __le32_to_cpu(migr_rec
->blocks_per_unit
);
7646 if (array_blocks
% __le32_to_cpu(migr_rec
->blocks_per_unit
))
7648 migr_rec
->num_migr_units
= __cpu_to_le32(num_migr_units
);
7650 migr_rec
->post_migr_vol_cap
= dev
->size_low
;
7651 migr_rec
->post_migr_vol_cap_hi
= dev
->size_high
;
7654 /* Find the smallest dev */
7655 for (sd
= info
->devs
; sd
; sd
= sd
->next
) {
7656 sprintf(nm
, "%d:%d", sd
->disk
.major
, sd
->disk
.minor
);
7657 fd
= dev_open(nm
, O_RDONLY
);
7660 get_dev_size(fd
, NULL
, &dsize
);
7661 dev_sectors
= dsize
/ 512;
7662 if (dev_sectors
< min_dev_sectors
)
7663 min_dev_sectors
= dev_sectors
;
7666 migr_rec
->ckpt_area_pba
= __cpu_to_le32(min_dev_sectors
-
7667 RAID_DISK_RESERVED_BLOCKS_IMSM_HI
);
7669 write_imsm_migr_rec(st
);
7674 /*******************************************************************************
7675 * Function: save_backup_imsm
7676 * Description: Function saves critical data stripes to Migration Copy Area
7677 * and updates the current migration unit status.
7678 * Use restore_stripes() to form a destination stripe,
7679 * and to write it to the Copy Area.
7681 * st : supertype information
7682 * info : general array info
7683 * buf : input buffer
7684 * write_offset : address of data to backup
7685 * length : length of data to backup (blocks_per_unit)
7689 ******************************************************************************/
7690 int save_backup_imsm(struct supertype
*st
,
7691 struct imsm_dev
*dev
,
7692 struct mdinfo
*info
,
7698 struct intel_super
*super
= st
->sb
;
7699 unsigned long long *target_offsets
= NULL
;
7700 int *targets
= NULL
;
7702 struct imsm_map
*map_dest
= get_imsm_map(dev
, 0);
7703 int new_disks
= map_dest
->num_members
;
7705 targets
= malloc(new_disks
* sizeof(int));
7709 target_offsets
= malloc(new_disks
* sizeof(unsigned long long));
7710 if (!target_offsets
)
7713 for (i
= 0; i
< new_disks
; i
++) {
7715 target_offsets
[i
] = (unsigned long long)
7716 __le32_to_cpu(super
->migr_rec
->ckpt_area_pba
) * 512;
7719 if (open_backup_targets(info
, new_disks
, targets
))
7722 if (restore_stripes(targets
, /* list of dest devices */
7723 target_offsets
, /* migration record offsets */
7728 -1, /* source backup file descriptor */
7729 0, /* input buf offset
7730 * always 0 buf is already offset */
7734 fprintf(stderr
, Name
": Error restoring stripes\n");
7742 for (i
= 0; i
< new_disks
; i
++)
7743 if (targets
[i
] >= 0)
7747 free(target_offsets
);
7752 /*******************************************************************************
7753 * Function: save_checkpoint_imsm
7754 * Description: Function called for current unit status update
7755 * in the migration record. It writes it to disk.
7757 * super : imsm internal array info
7758 * info : general array info
7762 * 2: failure, means no valid migration record
7763 * / no general migration in progress /
7764 ******************************************************************************/
7765 int save_checkpoint_imsm(struct supertype
*st
, struct mdinfo
*info
, int state
)
7767 struct intel_super
*super
= st
->sb
;
7768 if (load_imsm_migr_rec(super
, info
) != 0) {
7769 dprintf("imsm: ERROR: Cannot read migration record "
7770 "for checkpoint save.\n");
7774 if (__le32_to_cpu(super
->migr_rec
->blocks_per_unit
) == 0) {
7775 dprintf("imsm: no migration in progress.\n");
7779 super
->migr_rec
->curr_migr_unit
=
7780 __cpu_to_le32(info
->reshape_progress
/
7781 __le32_to_cpu(super
->migr_rec
->blocks_per_unit
));
7782 super
->migr_rec
->rec_status
= __cpu_to_le32(state
);
7783 super
->migr_rec
->dest_1st_member_lba
=
7784 __cpu_to_le32((__le32_to_cpu(super
->migr_rec
->curr_migr_unit
))
7785 * __le32_to_cpu(super
->migr_rec
->dest_depth_per_unit
));
7786 if (write_imsm_migr_rec(st
) < 0) {
7787 dprintf("imsm: Cannot write migration record "
7788 "outside backup area\n");
7795 static __u64
blocks_per_migr_unit(struct intel_super
*super
,
7796 struct imsm_dev
*dev
);
7798 /*******************************************************************************
7799 * Function: recover_backup_imsm
7800 * Description: Function recovers critical data from the Migration Copy Area
7801 * while assembling an array.
7803 * super : imsm internal array info
7804 * info : general array info
7806 * 0 : success (or there is no data to recover)
7808 ******************************************************************************/
7809 int recover_backup_imsm(struct supertype
*st
, struct mdinfo
*info
)
7811 struct intel_super
*super
= st
->sb
;
7812 struct migr_record
*migr_rec
= super
->migr_rec
;
7813 struct imsm_map
*map_dest
= NULL
;
7814 struct intel_dev
*id
= NULL
;
7815 unsigned long long read_offset
;
7816 unsigned long long write_offset
;
7818 int *targets
= NULL
;
7819 int new_disks
, i
, err
;
7822 unsigned long curr_migr_unit
= __le32_to_cpu(migr_rec
->curr_migr_unit
);
7823 unsigned long num_migr_units
= __le32_to_cpu(migr_rec
->num_migr_units
);
7824 int ascending
= __le32_to_cpu(migr_rec
->ascending_migr
);
7827 err
= sysfs_get_str(info
, NULL
, "array_state", (char *)buffer
, 20);
7831 /* recover data only during assemblation */
7832 if (strncmp(buffer
, "inactive", 8) != 0)
7834 /* no data to recover */
7835 if (__le32_to_cpu(migr_rec
->rec_status
) == UNIT_SRC_NORMAL
)
7837 if (curr_migr_unit
>= num_migr_units
)
7840 /* find device during reshape */
7841 for (id
= super
->devlist
; id
; id
= id
->next
)
7842 if (is_gen_migration(id
->dev
))
7847 map_dest
= get_imsm_map(id
->dev
, 0);
7848 new_disks
= map_dest
->num_members
;
7850 read_offset
= (unsigned long long)
7851 __le32_to_cpu(migr_rec
->ckpt_area_pba
) * 512;
7853 write_offset
= ((unsigned long long)
7854 __le32_to_cpu(migr_rec
->dest_1st_member_lba
) +
7855 info
->data_offset
) * 512;
7857 unit_len
= __le32_to_cpu(migr_rec
->dest_depth_per_unit
) * 512;
7858 if (posix_memalign((void **)&buf
, 512, unit_len
) != 0)
7860 targets
= malloc(new_disks
* sizeof(int));
7864 open_backup_targets(info
, new_disks
, targets
);
7866 for (i
= 0; i
< new_disks
; i
++) {
7867 if (lseek64(targets
[i
], read_offset
, SEEK_SET
) < 0) {
7869 Name
": Cannot seek to block: %s\n",
7873 if (read(targets
[i
], buf
, unit_len
) != unit_len
) {
7875 Name
": Cannot read copy area block: %s\n",
7879 if (lseek64(targets
[i
], write_offset
, SEEK_SET
) < 0) {
7881 Name
": Cannot seek to block: %s\n",
7885 if (write(targets
[i
], buf
, unit_len
) != unit_len
) {
7887 Name
": Cannot restore block: %s\n",
7893 if (ascending
&& curr_migr_unit
< (num_migr_units
-1))
7896 migr_rec
->curr_migr_unit
= __le32_to_cpu(curr_migr_unit
);
7897 super
->migr_rec
->rec_status
= __cpu_to_le32(UNIT_SRC_NORMAL
);
7898 if (write_imsm_migr_rec(st
) == 0) {
7899 __u64 blocks_per_unit
= blocks_per_migr_unit(super
, id
->dev
);
7900 info
->reshape_progress
= curr_migr_unit
* blocks_per_unit
;
7906 for (i
= 0; i
< new_disks
; i
++)
7915 static char disk_by_path
[] = "/dev/disk/by-path/";
7917 static const char *imsm_get_disk_controller_domain(const char *path
)
7919 char disk_path
[PATH_MAX
];
7923 strncpy(disk_path
, disk_by_path
, PATH_MAX
- 1);
7924 strncat(disk_path
, path
, PATH_MAX
- strlen(disk_path
) - 1);
7925 if (stat(disk_path
, &st
) == 0) {
7926 struct sys_dev
* hba
;
7929 path
= devt_to_devpath(st
.st_rdev
);
7932 hba
= find_disk_attached_hba(-1, path
);
7933 if (hba
&& hba
->type
== SYS_DEV_SAS
)
7935 else if (hba
&& hba
->type
== SYS_DEV_SATA
)
7939 dprintf("path: %s hba: %s attached: %s\n",
7940 path
, (hba
) ? hba
->path
: "NULL", drv
);
7948 static int imsm_find_array_minor_by_subdev(int subdev
, int container
, int *minor
)
7950 char subdev_name
[20];
7951 struct mdstat_ent
*mdstat
;
7953 sprintf(subdev_name
, "%d", subdev
);
7954 mdstat
= mdstat_by_subdev(subdev_name
, container
);
7958 *minor
= mdstat
->devnum
;
7959 free_mdstat(mdstat
);
7963 static int imsm_reshape_is_allowed_on_container(struct supertype
*st
,
7964 struct geo_params
*geo
,
7965 int *old_raid_disks
)
7967 /* currently we only support increasing the number of devices
7968 * for a container. This increases the number of device for each
7969 * member array. They must all be RAID0 or RAID5.
7972 struct mdinfo
*info
, *member
;
7973 int devices_that_can_grow
= 0;
7975 dprintf("imsm: imsm_reshape_is_allowed_on_container(ENTER): "
7976 "st->devnum = (%i)\n",
7979 if (geo
->size
!= -1 ||
7980 geo
->level
!= UnSet
||
7981 geo
->layout
!= UnSet
||
7982 geo
->chunksize
!= 0 ||
7983 geo
->raid_disks
== UnSet
) {
7984 dprintf("imsm: Container operation is allowed for "
7985 "raid disks number change only.\n");
7989 info
= container_content_imsm(st
, NULL
);
7990 for (member
= info
; member
; member
= member
->next
) {
7994 dprintf("imsm: checking device_num: %i\n",
7995 member
->container_member
);
7997 if (geo
->raid_disks
<= member
->array
.raid_disks
) {
7998 /* we work on container for Online Capacity Expansion
7999 * only so raid_disks has to grow
8001 dprintf("imsm: for container operation raid disks "
8002 "increase is required\n");
8006 if ((info
->array
.level
!= 0) &&
8007 (info
->array
.level
!= 5)) {
8008 /* we cannot use this container with other raid level
8010 dprintf("imsm: for container operation wrong"
8011 " raid level (%i) detected\n",
8015 /* check for platform support
8016 * for this raid level configuration
8018 struct intel_super
*super
= st
->sb
;
8019 if (!is_raid_level_supported(super
->orom
,
8020 member
->array
.level
,
8022 dprintf("platform does not support raid%d with"
8026 geo
->raid_disks
> 1 ? "s" : "");
8029 /* check if component size is aligned to chunk size
8031 if (info
->component_size
%
8032 (info
->array
.chunk_size
/512)) {
8033 dprintf("Component size is not aligned to "
8039 if (*old_raid_disks
&&
8040 info
->array
.raid_disks
!= *old_raid_disks
)
8042 *old_raid_disks
= info
->array
.raid_disks
;
8044 /* All raid5 and raid0 volumes in container
8045 * have to be ready for Online Capacity Expansion
8046 * so they need to be assembled. We have already
8047 * checked that no recovery etc is happening.
8049 result
= imsm_find_array_minor_by_subdev(member
->container_member
,
8053 dprintf("imsm: cannot find array\n");
8056 devices_that_can_grow
++;
8059 if (!member
&& devices_that_can_grow
)
8063 dprintf("\tContainer operation allowed\n");
8065 dprintf("\tError: %i\n", ret_val
);
8070 /* Function: get_spares_for_grow
8071 * Description: Allocates memory and creates list of spare devices
8072 * avaliable in container. Checks if spare drive size is acceptable.
8073 * Parameters: Pointer to the supertype structure
8074 * Returns: Pointer to the list of spare devices (mdinfo structure) on success,
8077 static struct mdinfo
*get_spares_for_grow(struct supertype
*st
)
8079 unsigned long long min_size
= min_acceptable_spare_size_imsm(st
);
8080 return container_choose_spares(st
, min_size
, NULL
, NULL
, NULL
, 0);
8083 /******************************************************************************
8084 * function: imsm_create_metadata_update_for_reshape
8085 * Function creates update for whole IMSM container.
8087 ******************************************************************************/
8088 static int imsm_create_metadata_update_for_reshape(
8089 struct supertype
*st
,
8090 struct geo_params
*geo
,
8092 struct imsm_update_reshape
**updatep
)
8094 struct intel_super
*super
= st
->sb
;
8095 struct imsm_super
*mpb
= super
->anchor
;
8096 int update_memory_size
= 0;
8097 struct imsm_update_reshape
*u
= NULL
;
8098 struct mdinfo
*spares
= NULL
;
8100 int delta_disks
= 0;
8103 dprintf("imsm_update_metadata_for_reshape(enter) raid_disks = %i\n",
8106 delta_disks
= geo
->raid_disks
- old_raid_disks
;
8108 /* size of all update data without anchor */
8109 update_memory_size
= sizeof(struct imsm_update_reshape
);
8111 /* now add space for spare disks that we need to add. */
8112 update_memory_size
+= sizeof(u
->new_disks
[0]) * (delta_disks
- 1);
8114 u
= calloc(1, update_memory_size
);
8117 "cannot get memory for imsm_update_reshape update\n");
8120 u
->type
= update_reshape_container_disks
;
8121 u
->old_raid_disks
= old_raid_disks
;
8122 u
->new_raid_disks
= geo
->raid_disks
;
8124 /* now get spare disks list
8126 spares
= get_spares_for_grow(st
);
8129 || delta_disks
> spares
->array
.spare_disks
) {
8130 fprintf(stderr
, Name
": imsm: ERROR: Cannot get spare devices "
8131 "for %s.\n", geo
->dev_name
);
8135 /* we have got spares
8136 * update disk list in imsm_disk list table in anchor
8138 dprintf("imsm: %i spares are available.\n\n",
8139 spares
->array
.spare_disks
);
8142 for (i
= 0; i
< delta_disks
; i
++) {
8147 u
->new_disks
[i
] = makedev(dev
->disk
.major
,
8149 dl
= get_disk_super(super
, dev
->disk
.major
, dev
->disk
.minor
);
8150 dl
->index
= mpb
->num_disks
;
8160 dprintf("imsm: reshape update preparation :");
8161 if (i
== delta_disks
) {
8164 return update_memory_size
;
8167 dprintf(" Error\n");
8172 /******************************************************************************
8173 * function: imsm_create_metadata_update_for_migration()
8174 * Creates update for IMSM array.
8176 ******************************************************************************/
8177 static int imsm_create_metadata_update_for_migration(
8178 struct supertype
*st
,
8179 struct geo_params
*geo
,
8180 struct imsm_update_reshape_migration
**updatep
)
8182 struct intel_super
*super
= st
->sb
;
8183 int update_memory_size
= 0;
8184 struct imsm_update_reshape_migration
*u
= NULL
;
8185 struct imsm_dev
*dev
;
8186 int previous_level
= -1;
8188 dprintf("imsm_create_metadata_update_for_migration(enter)"
8189 " New Level = %i\n", geo
->level
);
8191 /* size of all update data without anchor */
8192 update_memory_size
= sizeof(struct imsm_update_reshape_migration
);
8194 u
= calloc(1, update_memory_size
);
8196 dprintf("error: cannot get memory for "
8197 "imsm_create_metadata_update_for_migration\n");
8200 u
->type
= update_reshape_migration
;
8201 u
->subdev
= super
->current_vol
;
8202 u
->new_level
= geo
->level
;
8203 u
->new_layout
= geo
->layout
;
8204 u
->new_raid_disks
= u
->old_raid_disks
= geo
->raid_disks
;
8205 u
->new_disks
[0] = -1;
8206 u
->new_chunksize
= -1;
8208 dev
= get_imsm_dev(super
, u
->subdev
);
8210 struct imsm_map
*map
;
8212 map
= get_imsm_map(dev
, 0);
8214 int current_chunk_size
=
8215 __le16_to_cpu(map
->blocks_per_strip
) / 2;
8217 if (geo
->chunksize
!= current_chunk_size
) {
8218 u
->new_chunksize
= geo
->chunksize
/ 1024;
8220 "chunk size change from %i to %i\n",
8221 current_chunk_size
, u
->new_chunksize
);
8223 previous_level
= map
->raid_level
;
8226 if ((geo
->level
== 5) && (previous_level
== 0)) {
8227 struct mdinfo
*spares
= NULL
;
8229 u
->new_raid_disks
++;
8230 spares
= get_spares_for_grow(st
);
8231 if ((spares
== NULL
) || (spares
->array
.spare_disks
< 1)) {
8234 update_memory_size
= 0;
8235 dprintf("error: cannot get spare device "
8236 "for requested migration");
8241 dprintf("imsm: reshape update preparation : OK\n");
8244 return update_memory_size
;
8247 static void imsm_update_metadata_locally(struct supertype
*st
,
8250 struct metadata_update mu
;
8255 mu
.space_list
= NULL
;
8257 imsm_prepare_update(st
, &mu
);
8258 imsm_process_update(st
, &mu
);
8260 while (mu
.space_list
) {
8261 void **space
= mu
.space_list
;
8262 mu
.space_list
= *space
;
8267 /***************************************************************************
8268 * Function: imsm_analyze_change
8269 * Description: Function analyze change for single volume
8270 * and validate if transition is supported
8271 * Parameters: Geometry parameters, supertype structure
8272 * Returns: Operation type code on success, -1 if fail
8273 ****************************************************************************/
8274 enum imsm_reshape_type
imsm_analyze_change(struct supertype
*st
,
8275 struct geo_params
*geo
)
8282 getinfo_super_imsm_volume(st
, &info
, NULL
);
8284 if ((geo
->level
!= info
.array
.level
) &&
8285 (geo
->level
>= 0) &&
8286 (geo
->level
!= UnSet
)) {
8287 switch (info
.array
.level
) {
8289 if (geo
->level
== 5) {
8290 change
= CH_MIGRATION
;
8293 if (geo
->level
== 10) {
8294 change
= CH_TAKEOVER
;
8299 if (geo
->level
== 0) {
8300 change
= CH_TAKEOVER
;
8305 if (geo
->level
== 0) {
8306 change
= CH_TAKEOVER
;
8313 Name
" Error. Level Migration from %d to %d "
8315 info
.array
.level
, geo
->level
);
8316 goto analyse_change_exit
;
8319 geo
->level
= info
.array
.level
;
8321 if ((geo
->layout
!= info
.array
.layout
)
8322 && ((geo
->layout
!= UnSet
) && (geo
->layout
!= -1))) {
8323 change
= CH_MIGRATION
;
8324 if ((info
.array
.layout
== 0)
8325 && (info
.array
.level
== 5)
8326 && (geo
->layout
== 5)) {
8327 /* reshape 5 -> 4 */
8328 } else if ((info
.array
.layout
== 5)
8329 && (info
.array
.level
== 5)
8330 && (geo
->layout
== 0)) {
8331 /* reshape 4 -> 5 */
8336 Name
" Error. Layout Migration from %d to %d "
8338 info
.array
.layout
, geo
->layout
);
8340 goto analyse_change_exit
;
8343 geo
->layout
= info
.array
.layout
;
8345 if ((geo
->chunksize
> 0) && (geo
->chunksize
!= UnSet
)
8346 && (geo
->chunksize
!= info
.array
.chunk_size
))
8347 change
= CH_MIGRATION
;
8349 geo
->chunksize
= info
.array
.chunk_size
;
8351 chunk
= geo
->chunksize
/ 1024;
8352 if (!validate_geometry_imsm(st
,
8362 struct intel_super
*super
= st
->sb
;
8363 struct imsm_super
*mpb
= super
->anchor
;
8365 if (mpb
->num_raid_devs
> 1) {
8367 Name
" Error. Cannot perform operation on %s"
8368 "- for this operation it MUST be single "
8369 "array in container\n",
8375 analyse_change_exit
:
8380 int imsm_takeover(struct supertype
*st
, struct geo_params
*geo
)
8382 struct intel_super
*super
= st
->sb
;
8383 struct imsm_update_takeover
*u
;
8385 u
= malloc(sizeof(struct imsm_update_takeover
));
8389 u
->type
= update_takeover
;
8390 u
->subarray
= super
->current_vol
;
8392 /* 10->0 transition */
8393 if (geo
->level
== 0)
8394 u
->direction
= R10_TO_R0
;
8396 /* 0->10 transition */
8397 if (geo
->level
== 10)
8398 u
->direction
= R0_TO_R10
;
8400 /* update metadata locally */
8401 imsm_update_metadata_locally(st
, u
,
8402 sizeof(struct imsm_update_takeover
));
8403 /* and possibly remotely */
8404 if (st
->update_tail
)
8405 append_metadata_update(st
, u
,
8406 sizeof(struct imsm_update_takeover
));
8413 static int imsm_reshape_super(struct supertype
*st
, long long size
, int level
,
8414 int layout
, int chunksize
, int raid_disks
,
8415 int delta_disks
, char *backup
, char *dev
,
8419 struct geo_params geo
;
8421 dprintf("imsm: reshape_super called.\n");
8423 memset(&geo
, 0, sizeof(struct geo_params
));
8426 geo
.dev_id
= st
->devnum
;
8429 geo
.layout
= layout
;
8430 geo
.chunksize
= chunksize
;
8431 geo
.raid_disks
= raid_disks
;
8432 if (delta_disks
!= UnSet
)
8433 geo
.raid_disks
+= delta_disks
;
8435 dprintf("\tfor level : %i\n", geo
.level
);
8436 dprintf("\tfor raid_disks : %i\n", geo
.raid_disks
);
8438 if (experimental() == 0)
8441 if (st
->container_dev
== st
->devnum
) {
8442 /* On container level we can only increase number of devices. */
8443 dprintf("imsm: info: Container operation\n");
8444 int old_raid_disks
= 0;
8446 if (imsm_reshape_is_allowed_on_container(
8447 st
, &geo
, &old_raid_disks
)) {
8448 struct imsm_update_reshape
*u
= NULL
;
8451 len
= imsm_create_metadata_update_for_reshape(
8452 st
, &geo
, old_raid_disks
, &u
);
8455 dprintf("imsm: Cannot prepare update\n");
8456 goto exit_imsm_reshape_super
;
8460 /* update metadata locally */
8461 imsm_update_metadata_locally(st
, u
, len
);
8462 /* and possibly remotely */
8463 if (st
->update_tail
)
8464 append_metadata_update(st
, u
, len
);
8469 fprintf(stderr
, Name
": (imsm) Operation "
8470 "is not allowed on this container\n");
8473 /* On volume level we support following operations
8474 * - takeover: raid10 -> raid0; raid0 -> raid10
8475 * - chunk size migration
8476 * - migration: raid5 -> raid0; raid0 -> raid5
8478 struct intel_super
*super
= st
->sb
;
8479 struct intel_dev
*dev
= super
->devlist
;
8481 dprintf("imsm: info: Volume operation\n");
8482 /* find requested device */
8484 imsm_find_array_minor_by_subdev(dev
->index
, st
->container_dev
, &devnum
);
8485 if (devnum
== geo
.dev_id
)
8490 fprintf(stderr
, Name
" Cannot find %s (%i) subarray\n",
8491 geo
.dev_name
, geo
.dev_id
);
8492 goto exit_imsm_reshape_super
;
8494 super
->current_vol
= dev
->index
;
8495 change
= imsm_analyze_change(st
, &geo
);
8498 ret_val
= imsm_takeover(st
, &geo
);
8500 case CH_MIGRATION
: {
8501 struct imsm_update_reshape_migration
*u
= NULL
;
8503 imsm_create_metadata_update_for_migration(
8507 "Cannot prepare update\n");
8511 /* update metadata locally */
8512 imsm_update_metadata_locally(st
, u
, len
);
8513 /* and possibly remotely */
8514 if (st
->update_tail
)
8515 append_metadata_update(st
, u
, len
);
8525 exit_imsm_reshape_super
:
8526 dprintf("imsm: reshape_super Exit code = %i\n", ret_val
);
8530 /*******************************************************************************
8531 * Function: wait_for_reshape_imsm
8532 * Description: Function writes new sync_max value and waits until
8533 * reshape process reach new position
8535 * sra : general array info
8536 * to_complete : new sync_max position
8537 * ndata : number of disks in new array's layout
8540 * 1 : there is no reshape in progress,
8542 ******************************************************************************/
8543 int wait_for_reshape_imsm(struct mdinfo
*sra
, unsigned long long to_complete
,
8546 int fd
= sysfs_get_fd(sra
, NULL
, "reshape_position");
8547 unsigned long long completed
;
8549 struct timeval timeout
;
8554 sysfs_fd_get_ll(fd
, &completed
);
8556 if (to_complete
== 0) {/* reshape till the end of array */
8557 sysfs_set_str(sra
, NULL
, "sync_max", "max");
8558 to_complete
= MaxSector
;
8560 if (completed
> to_complete
) {
8564 if (sysfs_set_num(sra
, NULL
, "sync_max",
8565 to_complete
/ ndata
) != 0) {
8571 /* FIXME should not need a timeout at all */
8572 timeout
.tv_sec
= 30;
8573 timeout
.tv_usec
= 0;
8579 select(fd
+1, NULL
, NULL
, &rfds
, &timeout
);
8580 if (sysfs_fd_get_ll(fd
, &completed
) < 0) {
8584 if (sysfs_get_str(sra
, NULL
, "sync_action",
8586 strncmp(action
, "reshape", 7) != 0)
8588 } while (completed
< to_complete
);
8594 /*******************************************************************************
8595 * Function: check_degradation_change
8596 * Description: Check that array hasn't become failed.
8598 * info : for sysfs access
8599 * sources : source disks descriptors
8600 * degraded: previous degradation level
8603 ******************************************************************************/
8604 int check_degradation_change(struct mdinfo
*info
,
8608 unsigned long long new_degraded
;
8609 sysfs_get_ll(info
, NULL
, "degraded", &new_degraded
);
8610 if (new_degraded
!= (unsigned long long)degraded
) {
8611 /* check each device to ensure it is still working */
8614 for (sd
= info
->devs
; sd
; sd
= sd
->next
) {
8615 if (sd
->disk
.state
& (1<<MD_DISK_FAULTY
))
8617 if (sd
->disk
.state
& (1<<MD_DISK_SYNC
)) {
8619 if (sysfs_get_str(info
,
8620 sd
, "state", sbuf
, 20) < 0 ||
8621 strstr(sbuf
, "faulty") ||
8622 strstr(sbuf
, "in_sync") == NULL
) {
8623 /* this device is dead */
8624 sd
->disk
.state
= (1<<MD_DISK_FAULTY
);
8625 if (sd
->disk
.raid_disk
>= 0 &&
8626 sources
[sd
->disk
.raid_disk
] >= 0) {
8628 sd
->disk
.raid_disk
]);
8629 sources
[sd
->disk
.raid_disk
] =
8638 return new_degraded
;
8641 /*******************************************************************************
8642 * Function: imsm_manage_reshape
8643 * Description: Function finds array under reshape and it manages reshape
8644 * process. It creates stripes backups (if required) and sets
8647 * afd : Backup handle (nattive) - not used
8648 * sra : general array info
8649 * reshape : reshape parameters - not used
8650 * st : supertype structure
8651 * blocks : size of critical section [blocks]
8652 * fds : table of source device descriptor
8653 * offsets : start of array (offest per devices)
8655 * destfd : table of destination device descriptor
8656 * destoffsets : table of destination offsets (per device)
8658 * 1 : success, reshape is done
8660 ******************************************************************************/
8661 static int imsm_manage_reshape(
8662 int afd
, struct mdinfo
*sra
, struct reshape
*reshape
,
8663 struct supertype
*st
, unsigned long backup_blocks
,
8664 int *fds
, unsigned long long *offsets
,
8665 int dests
, int *destfd
, unsigned long long *destoffsets
)
8668 struct intel_super
*super
= st
->sb
;
8669 struct intel_dev
*dv
= NULL
;
8670 struct imsm_dev
*dev
= NULL
;
8671 struct imsm_map
*map_src
, *map_dest
;
8672 int migr_vol_qan
= 0;
8673 int ndata
, odata
; /* [bytes] */
8674 int chunk
; /* [bytes] */
8675 struct migr_record
*migr_rec
;
8677 unsigned int buf_size
; /* [bytes] */
8678 unsigned long long max_position
; /* array size [bytes] */
8679 unsigned long long next_step
; /* [blocks]/[bytes] */
8680 unsigned long long old_data_stripe_length
;
8681 unsigned long long new_data_stripe_length
;
8682 unsigned long long start_src
; /* [bytes] */
8683 unsigned long long start
; /* [bytes] */
8684 unsigned long long start_buf_shift
; /* [bytes] */
8687 if (!fds
|| !offsets
|| !destfd
|| !destoffsets
|| !sra
)
8690 /* Find volume during the reshape */
8691 for (dv
= super
->devlist
; dv
; dv
= dv
->next
) {
8692 if (dv
->dev
->vol
.migr_type
== MIGR_GEN_MIGR
8693 && dv
->dev
->vol
.migr_state
== 1) {
8698 /* Only one volume can migrate at the same time */
8699 if (migr_vol_qan
!= 1) {
8700 fprintf(stderr
, Name
" : %s", migr_vol_qan
?
8701 "Number of migrating volumes greater than 1\n" :
8702 "There is no volume during migrationg\n");
8706 map_src
= get_imsm_map(dev
, 1);
8707 if (map_src
== NULL
)
8709 map_dest
= get_imsm_map(dev
, 0);
8711 ndata
= imsm_num_data_members(dev
, 0);
8712 odata
= imsm_num_data_members(dev
, 1);
8714 chunk
= map_src
->blocks_per_strip
* 512;
8715 old_data_stripe_length
= odata
* chunk
;
8717 migr_rec
= super
->migr_rec
;
8720 sra
->new_chunk
= __le16_to_cpu(map_dest
->blocks_per_strip
) * 512;
8721 sra
->new_level
= map_dest
->raid_level
;
8722 new_data_stripe_length
= sra
->new_chunk
* ndata
;
8724 /* initialize migration record for start condition */
8725 if (sra
->reshape_progress
== 0)
8726 init_migr_record_imsm(st
, dev
, sra
);
8729 buf_size
= __le32_to_cpu(migr_rec
->blocks_per_unit
) * 512;
8730 /* extend buffer size for parity disk */
8731 buf_size
+= __le32_to_cpu(migr_rec
->dest_depth_per_unit
) * 512;
8732 /* add space for stripe aligment */
8733 buf_size
+= old_data_stripe_length
;
8734 if (posix_memalign((void **)&buf
, 4096, buf_size
)) {
8735 dprintf("imsm: Cannot allocate checpoint buffer\n");
8740 __le32_to_cpu(migr_rec
->post_migr_vol_cap
) +
8741 ((unsigned long long)__le32_to_cpu(
8742 migr_rec
->post_migr_vol_cap_hi
) << 32);
8744 while (__le32_to_cpu(migr_rec
->curr_migr_unit
) <
8745 __le32_to_cpu(migr_rec
->num_migr_units
)) {
8746 /* current reshape position [blocks] */
8747 unsigned long long current_position
=
8748 __le32_to_cpu(migr_rec
->blocks_per_unit
)
8749 * __le32_to_cpu(migr_rec
->curr_migr_unit
);
8750 unsigned long long border
;
8752 /* Check that array hasn't become failed.
8754 degraded
= check_degradation_change(sra
, fds
, degraded
);
8756 dprintf("imsm: Abort reshape due to degradation"
8757 " level (%i)\n", degraded
);
8761 next_step
= __le32_to_cpu(migr_rec
->blocks_per_unit
);
8763 if ((current_position
+ next_step
) > max_position
)
8764 next_step
= max_position
- current_position
;
8766 start
= (map_src
->pba_of_lba0
+ dev
->reserved_blocks
+
8767 current_position
) * 512;
8769 /* allign reading start to old geometry */
8770 start_buf_shift
= start
% old_data_stripe_length
;
8771 start_src
= start
- start_buf_shift
;
8773 border
= (start_src
/ odata
) - (start
/ ndata
);
8775 if (border
<= __le32_to_cpu(migr_rec
->dest_depth_per_unit
)) {
8776 /* save critical stripes to buf
8777 * start - start address of current unit
8779 * start_src - start address of current unit
8780 * to backup alligned to source array
8783 unsigned long long next_step_filler
= 0;
8784 unsigned long long copy_length
= next_step
* 512;
8786 /* allign copy area length to stripe in old geometry */
8787 next_step_filler
= ((copy_length
+ start_buf_shift
)
8788 % old_data_stripe_length
);
8789 if (next_step_filler
)
8790 next_step_filler
= (old_data_stripe_length
8791 - next_step_filler
);
8792 dprintf("save_stripes() parameters: start = %llu,"
8793 "\tstart_src = %llu,\tnext_step*512 = %llu,"
8794 "\tstart_in_buf_shift = %llu,"
8795 "\tnext_step_filler = %llu\n",
8796 start
, start_src
, copy_length
,
8797 start_buf_shift
, next_step_filler
);
8799 if (save_stripes(fds
, offsets
, map_src
->num_members
,
8800 chunk
, sra
->array
.level
,
8801 sra
->array
.layout
, 0, NULL
, start_src
,
8803 next_step_filler
+ start_buf_shift
,
8805 dprintf("imsm: Cannot save stripes"
8809 /* Convert data to destination format and store it
8810 * in backup general migration area
8812 if (save_backup_imsm(st
, dev
, sra
,
8813 buf
+ start_buf_shift
,
8814 ndata
, copy_length
)) {
8815 dprintf("imsm: Cannot save stripes to "
8816 "target devices\n");
8819 if (save_checkpoint_imsm(st
, sra
,
8820 UNIT_SRC_IN_CP_AREA
)) {
8821 dprintf("imsm: Cannot write checkpoint to "
8822 "migration record (UNIT_SRC_IN_CP_AREA)\n");
8825 /* decrease backup_blocks */
8826 if (backup_blocks
> (unsigned long)next_step
)
8827 backup_blocks
-= next_step
;
8831 /* When data backed up, checkpoint stored,
8832 * kick the kernel to reshape unit of data
8834 next_step
= next_step
+ sra
->reshape_progress
;
8835 sysfs_set_num(sra
, NULL
, "suspend_lo", sra
->reshape_progress
);
8836 sysfs_set_num(sra
, NULL
, "suspend_hi", next_step
);
8838 /* wait until reshape finish */
8839 if (wait_for_reshape_imsm(sra
, next_step
, ndata
) < 0) {
8840 dprintf("wait_for_reshape_imsm returned error!\n");
8844 sra
->reshape_progress
= next_step
;
8846 if (save_checkpoint_imsm(st
, sra
, UNIT_SRC_NORMAL
) == 1) {
8847 /* ignore error == 2, this can mean end of reshape here
8849 dprintf("imsm: Cannot write checkpoint to "
8850 "migration record (UNIT_SRC_NORMAL)\n");
8856 /* return '1' if done */
8864 #endif /* MDASSEMBLE */
8866 struct superswitch super_imsm
= {
8868 .examine_super
= examine_super_imsm
,
8869 .brief_examine_super
= brief_examine_super_imsm
,
8870 .brief_examine_subarrays
= brief_examine_subarrays_imsm
,
8871 .export_examine_super
= export_examine_super_imsm
,
8872 .detail_super
= detail_super_imsm
,
8873 .brief_detail_super
= brief_detail_super_imsm
,
8874 .write_init_super
= write_init_super_imsm
,
8875 .validate_geometry
= validate_geometry_imsm
,
8876 .add_to_super
= add_to_super_imsm
,
8877 .remove_from_super
= remove_from_super_imsm
,
8878 .detail_platform
= detail_platform_imsm
,
8879 .kill_subarray
= kill_subarray_imsm
,
8880 .update_subarray
= update_subarray_imsm
,
8881 .load_container
= load_container_imsm
,
8882 .default_geometry
= default_geometry_imsm
,
8883 .get_disk_controller_domain
= imsm_get_disk_controller_domain
,
8884 .reshape_super
= imsm_reshape_super
,
8885 .manage_reshape
= imsm_manage_reshape
,
8887 .match_home
= match_home_imsm
,
8888 .uuid_from_super
= uuid_from_super_imsm
,
8889 .getinfo_super
= getinfo_super_imsm
,
8890 .getinfo_super_disks
= getinfo_super_disks_imsm
,
8891 .update_super
= update_super_imsm
,
8893 .avail_size
= avail_size_imsm
,
8894 .min_acceptable_spare_size
= min_acceptable_spare_size_imsm
,
8896 .compare_super
= compare_super_imsm
,
8898 .load_super
= load_super_imsm
,
8899 .init_super
= init_super_imsm
,
8900 .store_super
= store_super_imsm
,
8901 .free_super
= free_super_imsm
,
8902 .match_metadata_desc
= match_metadata_desc_imsm
,
8903 .container_content
= container_content_imsm
,
8905 .recover_backup
= recover_backup_imsm
,
8912 .open_new
= imsm_open_new
,
8913 .set_array_state
= imsm_set_array_state
,
8914 .set_disk
= imsm_set_disk
,
8915 .sync_metadata
= imsm_sync_metadata
,
8916 .activate_spare
= imsm_activate_spare
,
8917 .process_update
= imsm_process_update
,
8918 .prepare_update
= imsm_prepare_update
,
8919 #endif /* MDASSEMBLE */