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
45 #define MPB_ATTRIB_RAID0 __cpu_to_le32(0x00000001)
47 #define MPB_ATTRIB_RAID1 __cpu_to_le32(0x00000002)
49 #define MPB_ATTRIB_RAID10 __cpu_to_le32(0x00000004)
51 #define MPB_ATTRIB_RAID1E __cpu_to_le32(0x00000008)
53 #define MPB_ATTRIB_RAID5 __cpu_to_le32(0x00000010)
54 /* supports RAID CNG */
55 #define MPB_ATTRIB_RAIDCNG __cpu_to_le32(0x00000020)
56 /* supports expanded stripe sizes of 256K, 512K and 1MB */
57 #define MPB_ATTRIB_EXP_STRIPE_SIZE __cpu_to_le32(0x00000040)
59 /* The OROM Support RST Caching of Volumes */
60 #define MPB_ATTRIB_NVM __cpu_to_le32(0x02000000)
61 /* The OROM supports creating disks greater than 2TB */
62 #define MPB_ATTRIB_2TB_DISK __cpu_to_le32(0x04000000)
63 /* The OROM supports Bad Block Management */
64 #define MPB_ATTRIB_BBM __cpu_to_le32(0x08000000)
66 /* THe OROM Supports NVM Caching of Volumes */
67 #define MPB_ATTRIB_NEVER_USE2 __cpu_to_le32(0x10000000)
68 /* The OROM supports creating volumes greater than 2TB */
69 #define MPB_ATTRIB_2TB __cpu_to_le32(0x20000000)
70 /* originally for PMP, now it's wasted b/c. Never use this bit! */
71 #define MPB_ATTRIB_NEVER_USE __cpu_to_le32(0x40000000)
72 /* Verify MPB contents against checksum after reading MPB */
73 #define MPB_ATTRIB_CHECKSUM_VERIFY __cpu_to_le32(0x80000000)
75 /* Define all supported attributes that have to be accepted by mdadm
77 #define MPB_ATTRIB_SUPPORTED (MPB_ATTRIB_CHECKSUM_VERIFY | \
79 MPB_ATTRIB_2TB_DISK | \
84 MPB_ATTRIB_EXP_STRIPE_SIZE)
86 /* Define attributes that are unused but not harmful */
87 #define MPB_ATTRIB_IGNORED (MPB_ATTRIB_NEVER_USE)
89 #define MPB_SECTOR_CNT 2210
90 #define IMSM_RESERVED_SECTORS 4096
91 #define NUM_BLOCKS_DIRTY_STRIPE_REGION 2056
92 #define SECT_PER_MB_SHIFT 11
94 /* Disk configuration info. */
95 #define IMSM_MAX_DEVICES 255
97 __u8 serial
[MAX_RAID_SERIAL_LEN
];/* 0xD8 - 0xE7 ascii serial number */
98 __u32 total_blocks
; /* 0xE8 - 0xEB total blocks */
99 __u32 scsi_id
; /* 0xEC - 0xEF scsi ID */
100 #define SPARE_DISK __cpu_to_le32(0x01) /* Spare */
101 #define CONFIGURED_DISK __cpu_to_le32(0x02) /* Member of some RaidDev */
102 #define FAILED_DISK __cpu_to_le32(0x04) /* Permanent failure */
103 __u32 status
; /* 0xF0 - 0xF3 */
104 __u32 owner_cfg_num
; /* which config 0,1,2... owns this disk */
105 #define IMSM_DISK_FILLERS 4
106 __u32 filler
[IMSM_DISK_FILLERS
]; /* 0xF4 - 0x107 MPB_DISK_FILLERS for future expansion */
109 /* map selector for map managment
115 /* RAID map configuration infos. */
117 __u32 pba_of_lba0
; /* start address of partition */
118 __u32 blocks_per_member
;/* blocks per member */
119 __u32 num_data_stripes
; /* number of data stripes */
120 __u16 blocks_per_strip
;
121 __u8 map_state
; /* Normal, Uninitialized, Degraded, Failed */
122 #define IMSM_T_STATE_NORMAL 0
123 #define IMSM_T_STATE_UNINITIALIZED 1
124 #define IMSM_T_STATE_DEGRADED 2
125 #define IMSM_T_STATE_FAILED 3
127 #define IMSM_T_RAID0 0
128 #define IMSM_T_RAID1 1
129 #define IMSM_T_RAID5 5 /* since metadata version 1.2.02 ? */
130 __u8 num_members
; /* number of member disks */
131 __u8 num_domains
; /* number of parity domains */
132 __u8 failed_disk_num
; /* valid only when state is degraded */
134 __u32 filler
[7]; /* expansion area */
135 #define IMSM_ORD_REBUILD (1 << 24)
136 __u32 disk_ord_tbl
[1]; /* disk_ord_tbl[num_members],
137 * top byte contains some flags
139 } __attribute__ ((packed
));
142 __u32 curr_migr_unit
;
143 __u32 checkpoint_id
; /* id to access curr_migr_unit */
144 __u8 migr_state
; /* Normal or Migrating */
146 #define MIGR_REBUILD 1
147 #define MIGR_VERIFY 2 /* analagous to echo check > sync_action */
148 #define MIGR_GEN_MIGR 3
149 #define MIGR_STATE_CHANGE 4
150 #define MIGR_REPAIR 5
151 __u8 migr_type
; /* Initializing, Rebuilding, ... */
153 __u8 fs_state
; /* fast-sync state for CnG (0xff == disabled) */
154 __u16 verify_errors
; /* number of mismatches */
155 __u16 bad_blocks
; /* number of bad blocks during verify */
157 struct imsm_map map
[1];
158 /* here comes another one if migr_state */
159 } __attribute__ ((packed
));
162 __u8 volume
[MAX_RAID_SERIAL_LEN
];
165 #define DEV_BOOTABLE __cpu_to_le32(0x01)
166 #define DEV_BOOT_DEVICE __cpu_to_le32(0x02)
167 #define DEV_READ_COALESCING __cpu_to_le32(0x04)
168 #define DEV_WRITE_COALESCING __cpu_to_le32(0x08)
169 #define DEV_LAST_SHUTDOWN_DIRTY __cpu_to_le32(0x10)
170 #define DEV_HIDDEN_AT_BOOT __cpu_to_le32(0x20)
171 #define DEV_CURRENTLY_HIDDEN __cpu_to_le32(0x40)
172 #define DEV_VERIFY_AND_FIX __cpu_to_le32(0x80)
173 #define DEV_MAP_STATE_UNINIT __cpu_to_le32(0x100)
174 #define DEV_NO_AUTO_RECOVERY __cpu_to_le32(0x200)
175 #define DEV_CLONE_N_GO __cpu_to_le32(0x400)
176 #define DEV_CLONE_MAN_SYNC __cpu_to_le32(0x800)
177 #define DEV_CNG_MASTER_DISK_NUM __cpu_to_le32(0x1000)
178 __u32 status
; /* Persistent RaidDev status */
179 __u32 reserved_blocks
; /* Reserved blocks at beginning of volume */
183 __u8 cng_master_disk
;
187 #define IMSM_DEV_FILLERS 10
188 __u32 filler
[IMSM_DEV_FILLERS
];
190 } __attribute__ ((packed
));
193 __u8 sig
[MAX_SIGNATURE_LENGTH
]; /* 0x00 - 0x1F */
194 __u32 check_sum
; /* 0x20 - 0x23 MPB Checksum */
195 __u32 mpb_size
; /* 0x24 - 0x27 Size of MPB */
196 __u32 family_num
; /* 0x28 - 0x2B Checksum from first time this config was written */
197 __u32 generation_num
; /* 0x2C - 0x2F Incremented each time this array's MPB is written */
198 __u32 error_log_size
; /* 0x30 - 0x33 in bytes */
199 __u32 attributes
; /* 0x34 - 0x37 */
200 __u8 num_disks
; /* 0x38 Number of configured disks */
201 __u8 num_raid_devs
; /* 0x39 Number of configured volumes */
202 __u8 error_log_pos
; /* 0x3A */
203 __u8 fill
[1]; /* 0x3B */
204 __u32 cache_size
; /* 0x3c - 0x40 in mb */
205 __u32 orig_family_num
; /* 0x40 - 0x43 original family num */
206 __u32 pwr_cycle_count
; /* 0x44 - 0x47 simulated power cycle count for array */
207 __u32 bbm_log_size
; /* 0x48 - 0x4B - size of bad Block Mgmt Log in bytes */
208 #define IMSM_FILLERS 35
209 __u32 filler
[IMSM_FILLERS
]; /* 0x4C - 0xD7 RAID_MPB_FILLERS */
210 struct imsm_disk disk
[1]; /* 0xD8 diskTbl[numDisks] */
211 /* here comes imsm_dev[num_raid_devs] */
212 /* here comes BBM logs */
213 } __attribute__ ((packed
));
215 #define BBM_LOG_MAX_ENTRIES 254
217 struct bbm_log_entry
{
218 __u64 defective_block_start
;
219 #define UNREADABLE 0xFFFFFFFF
220 __u32 spare_block_offset
;
221 __u16 remapped_marked_count
;
223 } __attribute__ ((__packed__
));
226 __u32 signature
; /* 0xABADB10C */
228 __u32 reserved_spare_block_count
; /* 0 */
229 __u32 reserved
; /* 0xFFFF */
230 __u64 first_spare_lba
;
231 struct bbm_log_entry mapped_block_entries
[BBM_LOG_MAX_ENTRIES
];
232 } __attribute__ ((__packed__
));
236 static char *map_state_str
[] = { "normal", "uninitialized", "degraded", "failed" };
239 #define RAID_DISK_RESERVED_BLOCKS_IMSM_HI 2209
241 #define GEN_MIGR_AREA_SIZE 2048 /* General Migration Copy Area size in blocks */
243 #define UNIT_SRC_NORMAL 0 /* Source data for curr_migr_unit must
244 * be recovered using srcMap */
245 #define UNIT_SRC_IN_CP_AREA 1 /* Source data for curr_migr_unit has
246 * already been migrated and must
247 * be recovered from checkpoint area */
249 __u32 rec_status
; /* Status used to determine how to restart
250 * migration in case it aborts
252 __u32 curr_migr_unit
; /* 0..numMigrUnits-1 */
253 __u32 family_num
; /* Family number of MPB
254 * containing the RaidDev
255 * that is migrating */
256 __u32 ascending_migr
; /* True if migrating in increasing
258 __u32 blocks_per_unit
; /* Num disk blocks per unit of operation */
259 __u32 dest_depth_per_unit
; /* Num member blocks each destMap
261 * advances per unit-of-operation */
262 __u32 ckpt_area_pba
; /* Pba of first block of ckpt copy area */
263 __u32 dest_1st_member_lba
; /* First member lba on first
264 * stripe of destination */
265 __u32 num_migr_units
; /* Total num migration units-of-op */
266 __u32 post_migr_vol_cap
; /* Size of volume after
267 * migration completes */
268 __u32 post_migr_vol_cap_hi
; /* Expansion space for LBA64 */
269 __u32 ckpt_read_disk_num
; /* Which member disk in destSubMap[0] the
270 * migration ckpt record was read from
271 * (for recovered migrations) */
272 } __attribute__ ((__packed__
));
274 static __u8
migr_type(struct imsm_dev
*dev
)
276 if (dev
->vol
.migr_type
== MIGR_VERIFY
&&
277 dev
->status
& DEV_VERIFY_AND_FIX
)
280 return dev
->vol
.migr_type
;
283 static void set_migr_type(struct imsm_dev
*dev
, __u8 migr_type
)
285 /* for compatibility with older oroms convert MIGR_REPAIR, into
286 * MIGR_VERIFY w/ DEV_VERIFY_AND_FIX status
288 if (migr_type
== MIGR_REPAIR
) {
289 dev
->vol
.migr_type
= MIGR_VERIFY
;
290 dev
->status
|= DEV_VERIFY_AND_FIX
;
292 dev
->vol
.migr_type
= migr_type
;
293 dev
->status
&= ~DEV_VERIFY_AND_FIX
;
297 static unsigned int sector_count(__u32 bytes
)
299 return ((bytes
+ (512-1)) & (~(512-1))) / 512;
302 static unsigned int mpb_sectors(struct imsm_super
*mpb
)
304 return sector_count(__le32_to_cpu(mpb
->mpb_size
));
308 struct imsm_dev
*dev
;
309 struct intel_dev
*next
;
314 enum sys_dev_type type
;
317 struct intel_hba
*next
;
324 /* internal representation of IMSM metadata */
327 void *buf
; /* O_DIRECT buffer for reading/writing metadata */
328 struct imsm_super
*anchor
; /* immovable parameters */
331 void *migr_rec_buf
; /* buffer for I/O operations */
332 struct migr_record
*migr_rec
; /* migration record */
334 size_t len
; /* size of the 'buf' allocation */
335 void *next_buf
; /* for realloc'ing buf from the manager */
337 int updates_pending
; /* count of pending updates for mdmon */
338 int current_vol
; /* index of raid device undergoing creation */
339 __u32 create_offset
; /* common start for 'current_vol' */
340 __u32 random
; /* random data for seeding new family numbers */
341 struct intel_dev
*devlist
;
345 __u8 serial
[MAX_RAID_SERIAL_LEN
];
348 struct imsm_disk disk
;
351 struct extent
*e
; /* for determining freespace @ create */
352 int raiddisk
; /* slot to fill in autolayout */
354 } *disks
, *current_disk
;
355 struct dl
*disk_mgmt_list
; /* list of disks to add/remove while mdmon
357 struct dl
*missing
; /* disks removed while we weren't looking */
358 struct bbm_log
*bbm_log
;
359 struct intel_hba
*hba
; /* device path of the raid controller for this metadata */
360 const struct imsm_orom
*orom
; /* platform firmware support */
361 struct intel_super
*next
; /* (temp) list for disambiguating family_num */
365 struct imsm_disk disk
;
366 #define IMSM_UNKNOWN_OWNER (-1)
368 struct intel_disk
*next
;
372 unsigned long long start
, size
;
375 /* definitions of reshape process types */
376 enum imsm_reshape_type
{
381 /* definition of messages passed to imsm_process_update */
382 enum imsm_update_type
{
383 update_activate_spare
,
387 update_add_remove_disk
,
388 update_reshape_container_disks
,
389 update_reshape_migration
,
391 update_general_migration_checkpoint
,
394 struct imsm_update_activate_spare
{
395 enum imsm_update_type type
;
399 struct imsm_update_activate_spare
*next
;
412 enum takeover_direction
{
416 struct imsm_update_takeover
{
417 enum imsm_update_type type
;
419 enum takeover_direction direction
;
422 struct imsm_update_reshape
{
423 enum imsm_update_type type
;
427 int new_disks
[1]; /* new_raid_disks - old_raid_disks makedev number */
430 struct imsm_update_reshape_migration
{
431 enum imsm_update_type type
;
434 /* fields for array migration changes
441 int new_disks
[1]; /* new_raid_disks - old_raid_disks makedev number */
444 struct imsm_update_general_migration_checkpoint
{
445 enum imsm_update_type type
;
446 __u32 curr_migr_unit
;
450 __u8 serial
[MAX_RAID_SERIAL_LEN
];
453 struct imsm_update_create_array
{
454 enum imsm_update_type type
;
459 struct imsm_update_kill_array
{
460 enum imsm_update_type type
;
464 struct imsm_update_rename_array
{
465 enum imsm_update_type type
;
466 __u8 name
[MAX_RAID_SERIAL_LEN
];
470 struct imsm_update_add_remove_disk
{
471 enum imsm_update_type type
;
475 static const char *_sys_dev_type
[] = {
476 [SYS_DEV_UNKNOWN
] = "Unknown",
477 [SYS_DEV_SAS
] = "SAS",
478 [SYS_DEV_SATA
] = "SATA"
481 const char *get_sys_dev_type(enum sys_dev_type type
)
483 if (type
>= SYS_DEV_MAX
)
484 type
= SYS_DEV_UNKNOWN
;
486 return _sys_dev_type
[type
];
489 static struct intel_hba
* alloc_intel_hba(struct sys_dev
*device
)
491 struct intel_hba
*result
= malloc(sizeof(*result
));
493 result
->type
= device
->type
;
494 result
->path
= strdup(device
->path
);
496 if (result
->path
&& (result
->pci_id
= strrchr(result
->path
, '/')) != NULL
)
502 static struct intel_hba
* find_intel_hba(struct intel_hba
*hba
, struct sys_dev
*device
)
504 struct intel_hba
*result
=NULL
;
505 for (result
= hba
; result
; result
= result
->next
) {
506 if (result
->type
== device
->type
&& strcmp(result
->path
, device
->path
) == 0)
512 static int attach_hba_to_super(struct intel_super
*super
, struct sys_dev
*device
)
514 struct intel_hba
*hba
;
516 /* check if disk attached to Intel HBA */
517 hba
= find_intel_hba(super
->hba
, device
);
520 /* Check if HBA is already attached to super */
521 if (super
->hba
== NULL
) {
522 super
->hba
= alloc_intel_hba(device
);
527 /* Intel metadata allows for all disks attached to the same type HBA.
528 * Do not sypport odf HBA types mixing
530 if (device
->type
!= hba
->type
)
536 hba
->next
= alloc_intel_hba(device
);
540 static struct sys_dev
* find_disk_attached_hba(int fd
, const char *devname
)
542 struct sys_dev
*list
, *elem
, *prev
;
545 if ((list
= find_intel_devices()) == NULL
)
549 disk_path
= (char *) devname
;
551 disk_path
= diskfd_to_devpath(fd
);
558 for (prev
= NULL
, elem
= list
; elem
; prev
= elem
, elem
= elem
->next
) {
559 if (path_attached_to_hba(disk_path
, elem
->path
)) {
563 prev
->next
= elem
->next
;
565 if (disk_path
!= devname
)
571 if (disk_path
!= devname
)
579 static int find_intel_hba_capability(int fd
, struct intel_super
*super
,
582 static struct supertype
*match_metadata_desc_imsm(char *arg
)
584 struct supertype
*st
;
586 if (strcmp(arg
, "imsm") != 0 &&
587 strcmp(arg
, "default") != 0
591 st
= malloc(sizeof(*st
));
594 memset(st
, 0, sizeof(*st
));
595 st
->container_dev
= NoMdDev
;
596 st
->ss
= &super_imsm
;
597 st
->max_devs
= IMSM_MAX_DEVICES
;
598 st
->minor_version
= 0;
604 static __u8
*get_imsm_version(struct imsm_super
*mpb
)
606 return &mpb
->sig
[MPB_SIG_LEN
];
610 /* retrieve a disk directly from the anchor when the anchor is known to be
611 * up-to-date, currently only at load time
613 static struct imsm_disk
*__get_imsm_disk(struct imsm_super
*mpb
, __u8 index
)
615 if (index
>= mpb
->num_disks
)
617 return &mpb
->disk
[index
];
620 /* retrieve the disk description based on a index of the disk
623 static struct dl
*get_imsm_dl_disk(struct intel_super
*super
, __u8 index
)
627 for (d
= super
->disks
; d
; d
= d
->next
)
628 if (d
->index
== index
)
633 /* retrieve a disk from the parsed metadata */
634 static struct imsm_disk
*get_imsm_disk(struct intel_super
*super
, __u8 index
)
638 dl
= get_imsm_dl_disk(super
, index
);
645 /* generate a checksum directly from the anchor when the anchor is known to be
646 * up-to-date, currently only at load or write_super after coalescing
648 static __u32
__gen_imsm_checksum(struct imsm_super
*mpb
)
650 __u32 end
= mpb
->mpb_size
/ sizeof(end
);
651 __u32
*p
= (__u32
*) mpb
;
655 sum
+= __le32_to_cpu(*p
);
659 return sum
- __le32_to_cpu(mpb
->check_sum
);
662 static size_t sizeof_imsm_map(struct imsm_map
*map
)
664 return sizeof(struct imsm_map
) + sizeof(__u32
) * (map
->num_members
- 1);
667 struct imsm_map
*get_imsm_map(struct imsm_dev
*dev
, int second_map
)
669 /* A device can have 2 maps if it is in the middle of a migration.
671 * MAP_0 - we return the first map
672 * MAP_1 - we return the second map if it exists, else NULL
673 * MAP_X - we return the second map if it exists, else the first
675 struct imsm_map
*map
= &dev
->vol
.map
[0];
676 struct imsm_map
*map2
= NULL
;
678 if (dev
->vol
.migr_state
)
679 map2
= (void *)map
+ sizeof_imsm_map(map
);
681 switch (second_map
) {
698 /* return the size of the device.
699 * migr_state increases the returned size if map[0] were to be duplicated
701 static size_t sizeof_imsm_dev(struct imsm_dev
*dev
, int migr_state
)
703 size_t size
= sizeof(*dev
) - sizeof(struct imsm_map
) +
704 sizeof_imsm_map(get_imsm_map(dev
, MAP_0
));
706 /* migrating means an additional map */
707 if (dev
->vol
.migr_state
)
708 size
+= sizeof_imsm_map(get_imsm_map(dev
, MAP_1
));
710 size
+= sizeof_imsm_map(get_imsm_map(dev
, MAP_0
));
716 /* retrieve disk serial number list from a metadata update */
717 static struct disk_info
*get_disk_info(struct imsm_update_create_array
*update
)
720 struct disk_info
*inf
;
722 inf
= u
+ sizeof(*update
) - sizeof(struct imsm_dev
) +
723 sizeof_imsm_dev(&update
->dev
, 0);
729 static struct imsm_dev
*__get_imsm_dev(struct imsm_super
*mpb
, __u8 index
)
735 if (index
>= mpb
->num_raid_devs
)
738 /* devices start after all disks */
739 offset
= ((void *) &mpb
->disk
[mpb
->num_disks
]) - _mpb
;
741 for (i
= 0; i
<= index
; i
++)
743 return _mpb
+ offset
;
745 offset
+= sizeof_imsm_dev(_mpb
+ offset
, 0);
750 static struct imsm_dev
*get_imsm_dev(struct intel_super
*super
, __u8 index
)
752 struct intel_dev
*dv
;
754 if (index
>= super
->anchor
->num_raid_devs
)
756 for (dv
= super
->devlist
; dv
; dv
= dv
->next
)
757 if (dv
->index
== index
)
764 * == MAP_0 get first map
765 * == MAP_1 get second map
766 * == MAP_X than get map according to the current migr_state
768 static __u32
get_imsm_ord_tbl_ent(struct imsm_dev
*dev
,
772 struct imsm_map
*map
;
774 map
= get_imsm_map(dev
, second_map
);
776 /* top byte identifies disk under rebuild */
777 return __le32_to_cpu(map
->disk_ord_tbl
[slot
]);
780 #define ord_to_idx(ord) (((ord) << 8) >> 8)
781 static __u32
get_imsm_disk_idx(struct imsm_dev
*dev
, int slot
, int second_map
)
783 __u32 ord
= get_imsm_ord_tbl_ent(dev
, slot
, second_map
);
785 return ord_to_idx(ord
);
788 static void set_imsm_ord_tbl_ent(struct imsm_map
*map
, int slot
, __u32 ord
)
790 map
->disk_ord_tbl
[slot
] = __cpu_to_le32(ord
);
793 static int get_imsm_disk_slot(struct imsm_map
*map
, unsigned idx
)
798 for (slot
= 0; slot
< map
->num_members
; slot
++) {
799 ord
= __le32_to_cpu(map
->disk_ord_tbl
[slot
]);
800 if (ord_to_idx(ord
) == idx
)
807 static int get_imsm_raid_level(struct imsm_map
*map
)
809 if (map
->raid_level
== 1) {
810 if (map
->num_members
== 2)
816 return map
->raid_level
;
819 static int cmp_extent(const void *av
, const void *bv
)
821 const struct extent
*a
= av
;
822 const struct extent
*b
= bv
;
823 if (a
->start
< b
->start
)
825 if (a
->start
> b
->start
)
830 static int count_memberships(struct dl
*dl
, struct intel_super
*super
)
835 for (i
= 0; i
< super
->anchor
->num_raid_devs
; i
++) {
836 struct imsm_dev
*dev
= get_imsm_dev(super
, i
);
837 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
839 if (get_imsm_disk_slot(map
, dl
->index
) >= 0)
846 static __u32
imsm_min_reserved_sectors(struct intel_super
*super
);
848 static struct extent
*get_extents(struct intel_super
*super
, struct dl
*dl
)
850 /* find a list of used extents on the given physical device */
851 struct extent
*rv
, *e
;
853 int memberships
= count_memberships(dl
, super
);
856 /* trim the reserved area for spares, so they can join any array
857 * regardless of whether the OROM has assigned sectors from the
858 * IMSM_RESERVED_SECTORS region
861 reservation
= imsm_min_reserved_sectors(super
);
863 reservation
= MPB_SECTOR_CNT
+ IMSM_RESERVED_SECTORS
;
865 rv
= malloc(sizeof(struct extent
) * (memberships
+ 1));
870 for (i
= 0; i
< super
->anchor
->num_raid_devs
; i
++) {
871 struct imsm_dev
*dev
= get_imsm_dev(super
, i
);
872 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
874 if (get_imsm_disk_slot(map
, dl
->index
) >= 0) {
875 e
->start
= __le32_to_cpu(map
->pba_of_lba0
);
876 e
->size
= __le32_to_cpu(map
->blocks_per_member
);
880 qsort(rv
, memberships
, sizeof(*rv
), cmp_extent
);
882 /* determine the start of the metadata
883 * when no raid devices are defined use the default
884 * ...otherwise allow the metadata to truncate the value
885 * as is the case with older versions of imsm
888 struct extent
*last
= &rv
[memberships
- 1];
891 remainder
= __le32_to_cpu(dl
->disk
.total_blocks
) -
892 (last
->start
+ last
->size
);
893 /* round down to 1k block to satisfy precision of the kernel
897 /* make sure remainder is still sane */
898 if (remainder
< (unsigned)ROUND_UP(super
->len
, 512) >> 9)
899 remainder
= ROUND_UP(super
->len
, 512) >> 9;
900 if (reservation
> remainder
)
901 reservation
= remainder
;
903 e
->start
= __le32_to_cpu(dl
->disk
.total_blocks
) - reservation
;
908 /* try to determine how much space is reserved for metadata from
909 * the last get_extents() entry, otherwise fallback to the
912 static __u32
imsm_reserved_sectors(struct intel_super
*super
, struct dl
*dl
)
918 /* for spares just return a minimal reservation which will grow
919 * once the spare is picked up by an array
922 return MPB_SECTOR_CNT
;
924 e
= get_extents(super
, dl
);
926 return MPB_SECTOR_CNT
+ IMSM_RESERVED_SECTORS
;
928 /* scroll to last entry */
929 for (i
= 0; e
[i
].size
; i
++)
932 rv
= __le32_to_cpu(dl
->disk
.total_blocks
) - e
[i
].start
;
939 static int is_spare(struct imsm_disk
*disk
)
941 return (disk
->status
& SPARE_DISK
) == SPARE_DISK
;
944 static int is_configured(struct imsm_disk
*disk
)
946 return (disk
->status
& CONFIGURED_DISK
) == CONFIGURED_DISK
;
949 static int is_failed(struct imsm_disk
*disk
)
951 return (disk
->status
& FAILED_DISK
) == FAILED_DISK
;
954 /* try to determine how much space is reserved for metadata from
955 * the last get_extents() entry on the smallest active disk,
956 * otherwise fallback to the default
958 static __u32
imsm_min_reserved_sectors(struct intel_super
*super
)
962 __u32 min_active
, remainder
;
963 __u32 rv
= MPB_SECTOR_CNT
+ IMSM_RESERVED_SECTORS
;
964 struct dl
*dl
, *dl_min
= NULL
;
970 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
973 if (dl
->disk
.total_blocks
< min_active
|| min_active
== 0) {
975 min_active
= dl
->disk
.total_blocks
;
981 /* find last lba used by subarrays on the smallest active disk */
982 e
= get_extents(super
, dl_min
);
985 for (i
= 0; e
[i
].size
; i
++)
988 remainder
= min_active
- e
[i
].start
;
991 /* to give priority to recovery we should not require full
992 IMSM_RESERVED_SECTORS from the spare */
993 rv
= MPB_SECTOR_CNT
+ NUM_BLOCKS_DIRTY_STRIPE_REGION
;
995 /* if real reservation is smaller use that value */
996 return (remainder
< rv
) ? remainder
: rv
;
999 /* Return minimum size of a spare that can be used in this array*/
1000 static unsigned long long min_acceptable_spare_size_imsm(struct supertype
*st
)
1002 struct intel_super
*super
= st
->sb
;
1006 unsigned long long rv
= 0;
1010 /* find first active disk in array */
1012 while (dl
&& (is_failed(&dl
->disk
) || dl
->index
== -1))
1016 /* find last lba used by subarrays */
1017 e
= get_extents(super
, dl
);
1020 for (i
= 0; e
[i
].size
; i
++)
1023 rv
= e
[i
-1].start
+ e
[i
-1].size
;
1026 /* add the amount of space needed for metadata */
1027 rv
= rv
+ imsm_min_reserved_sectors(super
);
1032 static int is_gen_migration(struct imsm_dev
*dev
);
1035 static __u64
blocks_per_migr_unit(struct intel_super
*super
,
1036 struct imsm_dev
*dev
);
1038 static void print_imsm_dev(struct intel_super
*super
,
1039 struct imsm_dev
*dev
,
1045 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
1046 struct imsm_map
*map2
= get_imsm_map(dev
, MAP_1
);
1050 printf("[%.16s]:\n", dev
->volume
);
1051 printf(" UUID : %s\n", uuid
);
1052 printf(" RAID Level : %d", get_imsm_raid_level(map
));
1054 printf(" <-- %d", get_imsm_raid_level(map2
));
1056 printf(" Members : %d", map
->num_members
);
1058 printf(" <-- %d", map2
->num_members
);
1060 printf(" Slots : [");
1061 for (i
= 0; i
< map
->num_members
; i
++) {
1062 ord
= get_imsm_ord_tbl_ent(dev
, i
, MAP_0
);
1063 printf("%s", ord
& IMSM_ORD_REBUILD
? "_" : "U");
1068 for (i
= 0; i
< map2
->num_members
; i
++) {
1069 ord
= get_imsm_ord_tbl_ent(dev
, i
, MAP_1
);
1070 printf("%s", ord
& IMSM_ORD_REBUILD
? "_" : "U");
1075 printf(" Failed disk : ");
1076 if (map
->failed_disk_num
== 0xff)
1079 printf("%i", map
->failed_disk_num
);
1081 slot
= get_imsm_disk_slot(map
, disk_idx
);
1083 ord
= get_imsm_ord_tbl_ent(dev
, slot
, MAP_X
);
1084 printf(" This Slot : %d%s\n", slot
,
1085 ord
& IMSM_ORD_REBUILD
? " (out-of-sync)" : "");
1087 printf(" This Slot : ?\n");
1088 sz
= __le32_to_cpu(dev
->size_high
);
1090 sz
+= __le32_to_cpu(dev
->size_low
);
1091 printf(" Array Size : %llu%s\n", (unsigned long long)sz
,
1092 human_size(sz
* 512));
1093 sz
= __le32_to_cpu(map
->blocks_per_member
);
1094 printf(" Per Dev Size : %llu%s\n", (unsigned long long)sz
,
1095 human_size(sz
* 512));
1096 printf(" Sector Offset : %u\n",
1097 __le32_to_cpu(map
->pba_of_lba0
));
1098 printf(" Num Stripes : %u\n",
1099 __le32_to_cpu(map
->num_data_stripes
));
1100 printf(" Chunk Size : %u KiB",
1101 __le16_to_cpu(map
->blocks_per_strip
) / 2);
1103 printf(" <-- %u KiB",
1104 __le16_to_cpu(map2
->blocks_per_strip
) / 2);
1106 printf(" Reserved : %d\n", __le32_to_cpu(dev
->reserved_blocks
));
1107 printf(" Migrate State : ");
1108 if (dev
->vol
.migr_state
) {
1109 if (migr_type(dev
) == MIGR_INIT
)
1110 printf("initialize\n");
1111 else if (migr_type(dev
) == MIGR_REBUILD
)
1112 printf("rebuild\n");
1113 else if (migr_type(dev
) == MIGR_VERIFY
)
1115 else if (migr_type(dev
) == MIGR_GEN_MIGR
)
1116 printf("general migration\n");
1117 else if (migr_type(dev
) == MIGR_STATE_CHANGE
)
1118 printf("state change\n");
1119 else if (migr_type(dev
) == MIGR_REPAIR
)
1122 printf("<unknown:%d>\n", migr_type(dev
));
1125 printf(" Map State : %s", map_state_str
[map
->map_state
]);
1126 if (dev
->vol
.migr_state
) {
1127 struct imsm_map
*map
= get_imsm_map(dev
, MAP_1
);
1129 printf(" <-- %s", map_state_str
[map
->map_state
]);
1130 printf("\n Checkpoint : %u ",
1131 __le32_to_cpu(dev
->vol
.curr_migr_unit
));
1132 if ((is_gen_migration(dev
)) && (super
->disks
->index
> 1))
1135 printf("(%llu)", (unsigned long long)
1136 blocks_per_migr_unit(super
, dev
));
1139 printf(" Dirty State : %s\n", dev
->vol
.dirty
? "dirty" : "clean");
1142 static void print_imsm_disk(struct imsm_disk
*disk
, int index
, __u32 reserved
)
1144 char str
[MAX_RAID_SERIAL_LEN
+ 1];
1147 if (index
< -1 || !disk
)
1151 snprintf(str
, MAX_RAID_SERIAL_LEN
+ 1, "%s", disk
->serial
);
1153 printf(" Disk%02d Serial : %s\n", index
, str
);
1155 printf(" Disk Serial : %s\n", str
);
1156 printf(" State :%s%s%s\n", is_spare(disk
) ? " spare" : "",
1157 is_configured(disk
) ? " active" : "",
1158 is_failed(disk
) ? " failed" : "");
1159 printf(" Id : %08x\n", __le32_to_cpu(disk
->scsi_id
));
1160 sz
= __le32_to_cpu(disk
->total_blocks
) - reserved
;
1161 printf(" Usable Size : %llu%s\n", (unsigned long long)sz
,
1162 human_size(sz
* 512));
1165 void examine_migr_rec_imsm(struct intel_super
*super
)
1167 struct migr_record
*migr_rec
= super
->migr_rec
;
1168 struct imsm_super
*mpb
= super
->anchor
;
1171 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
1172 struct imsm_dev
*dev
= __get_imsm_dev(mpb
, i
);
1173 if (is_gen_migration(dev
) == 0)
1176 printf("\nMigration Record Information:");
1177 if (super
->disks
->index
> 1) {
1178 printf(" Empty\n ");
1179 printf("Examine one of first two disks in array\n");
1182 printf("\n Status : ");
1183 if (__le32_to_cpu(migr_rec
->rec_status
) == UNIT_SRC_NORMAL
)
1186 printf("Contains Data\n");
1187 printf(" Current Unit : %u\n",
1188 __le32_to_cpu(migr_rec
->curr_migr_unit
));
1189 printf(" Family : %u\n",
1190 __le32_to_cpu(migr_rec
->family_num
));
1191 printf(" Ascending : %u\n",
1192 __le32_to_cpu(migr_rec
->ascending_migr
));
1193 printf(" Blocks Per Unit : %u\n",
1194 __le32_to_cpu(migr_rec
->blocks_per_unit
));
1195 printf(" Dest. Depth Per Unit : %u\n",
1196 __le32_to_cpu(migr_rec
->dest_depth_per_unit
));
1197 printf(" Checkpoint Area pba : %u\n",
1198 __le32_to_cpu(migr_rec
->ckpt_area_pba
));
1199 printf(" First member lba : %u\n",
1200 __le32_to_cpu(migr_rec
->dest_1st_member_lba
));
1201 printf(" Total Number of Units : %u\n",
1202 __le32_to_cpu(migr_rec
->num_migr_units
));
1203 printf(" Size of volume : %u\n",
1204 __le32_to_cpu(migr_rec
->post_migr_vol_cap
));
1205 printf(" Expansion space for LBA64 : %u\n",
1206 __le32_to_cpu(migr_rec
->post_migr_vol_cap_hi
));
1207 printf(" Record was read from : %u\n",
1208 __le32_to_cpu(migr_rec
->ckpt_read_disk_num
));
1213 #endif /* MDASSEMBLE */
1214 /*******************************************************************************
1215 * function: imsm_check_attributes
1216 * Description: Function checks if features represented by attributes flags
1217 * are supported by mdadm.
1219 * attributes - Attributes read from metadata
1221 * 0 - passed attributes contains unsupported features flags
1222 * 1 - all features are supported
1223 ******************************************************************************/
1224 static int imsm_check_attributes(__u32 attributes
)
1227 __u32 not_supported
= MPB_ATTRIB_SUPPORTED
^0xffffffff;
1229 not_supported
&= ~MPB_ATTRIB_IGNORED
;
1231 not_supported
&= attributes
;
1232 if (not_supported
) {
1233 fprintf(stderr
, Name
"(IMSM): Unsupported attributes : %x\n",
1234 (unsigned)__le32_to_cpu(not_supported
));
1235 if (not_supported
& MPB_ATTRIB_CHECKSUM_VERIFY
) {
1236 dprintf("\t\tMPB_ATTRIB_CHECKSUM_VERIFY \n");
1237 not_supported
^= MPB_ATTRIB_CHECKSUM_VERIFY
;
1239 if (not_supported
& MPB_ATTRIB_2TB
) {
1240 dprintf("\t\tMPB_ATTRIB_2TB\n");
1241 not_supported
^= MPB_ATTRIB_2TB
;
1243 if (not_supported
& MPB_ATTRIB_RAID0
) {
1244 dprintf("\t\tMPB_ATTRIB_RAID0\n");
1245 not_supported
^= MPB_ATTRIB_RAID0
;
1247 if (not_supported
& MPB_ATTRIB_RAID1
) {
1248 dprintf("\t\tMPB_ATTRIB_RAID1\n");
1249 not_supported
^= MPB_ATTRIB_RAID1
;
1251 if (not_supported
& MPB_ATTRIB_RAID10
) {
1252 dprintf("\t\tMPB_ATTRIB_RAID10\n");
1253 not_supported
^= MPB_ATTRIB_RAID10
;
1255 if (not_supported
& MPB_ATTRIB_RAID1E
) {
1256 dprintf("\t\tMPB_ATTRIB_RAID1E\n");
1257 not_supported
^= MPB_ATTRIB_RAID1E
;
1259 if (not_supported
& MPB_ATTRIB_RAID5
) {
1260 dprintf("\t\tMPB_ATTRIB_RAID5\n");
1261 not_supported
^= MPB_ATTRIB_RAID5
;
1263 if (not_supported
& MPB_ATTRIB_RAIDCNG
) {
1264 dprintf("\t\tMPB_ATTRIB_RAIDCNG\n");
1265 not_supported
^= MPB_ATTRIB_RAIDCNG
;
1267 if (not_supported
& MPB_ATTRIB_BBM
) {
1268 dprintf("\t\tMPB_ATTRIB_BBM\n");
1269 not_supported
^= MPB_ATTRIB_BBM
;
1271 if (not_supported
& MPB_ATTRIB_CHECKSUM_VERIFY
) {
1272 dprintf("\t\tMPB_ATTRIB_CHECKSUM_VERIFY (== MPB_ATTRIB_LEGACY)\n");
1273 not_supported
^= MPB_ATTRIB_CHECKSUM_VERIFY
;
1275 if (not_supported
& MPB_ATTRIB_EXP_STRIPE_SIZE
) {
1276 dprintf("\t\tMPB_ATTRIB_EXP_STRIP_SIZE\n");
1277 not_supported
^= MPB_ATTRIB_EXP_STRIPE_SIZE
;
1279 if (not_supported
& MPB_ATTRIB_2TB_DISK
) {
1280 dprintf("\t\tMPB_ATTRIB_2TB_DISK\n");
1281 not_supported
^= MPB_ATTRIB_2TB_DISK
;
1283 if (not_supported
& MPB_ATTRIB_NEVER_USE2
) {
1284 dprintf("\t\tMPB_ATTRIB_NEVER_USE2\n");
1285 not_supported
^= MPB_ATTRIB_NEVER_USE2
;
1287 if (not_supported
& MPB_ATTRIB_NEVER_USE
) {
1288 dprintf("\t\tMPB_ATTRIB_NEVER_USE\n");
1289 not_supported
^= MPB_ATTRIB_NEVER_USE
;
1293 dprintf(Name
"(IMSM): Unknown attributes : %x\n", not_supported
);
1302 static void getinfo_super_imsm(struct supertype
*st
, struct mdinfo
*info
, char *map
);
1304 static void examine_super_imsm(struct supertype
*st
, char *homehost
)
1306 struct intel_super
*super
= st
->sb
;
1307 struct imsm_super
*mpb
= super
->anchor
;
1308 char str
[MAX_SIGNATURE_LENGTH
];
1313 __u32 reserved
= imsm_reserved_sectors(super
, super
->disks
);
1316 snprintf(str
, MPB_SIG_LEN
, "%s", mpb
->sig
);
1317 printf(" Magic : %s\n", str
);
1318 snprintf(str
, strlen(MPB_VERSION_RAID0
), "%s", get_imsm_version(mpb
));
1319 printf(" Version : %s\n", get_imsm_version(mpb
));
1320 printf(" Orig Family : %08x\n", __le32_to_cpu(mpb
->orig_family_num
));
1321 printf(" Family : %08x\n", __le32_to_cpu(mpb
->family_num
));
1322 printf(" Generation : %08x\n", __le32_to_cpu(mpb
->generation_num
));
1323 printf(" Attributes : ");
1324 if (imsm_check_attributes(mpb
->attributes
))
1325 printf("All supported\n");
1327 printf("not supported\n");
1328 getinfo_super_imsm(st
, &info
, NULL
);
1329 fname_from_uuid(st
, &info
, nbuf
, ':');
1330 printf(" UUID : %s\n", nbuf
+ 5);
1331 sum
= __le32_to_cpu(mpb
->check_sum
);
1332 printf(" Checksum : %08x %s\n", sum
,
1333 __gen_imsm_checksum(mpb
) == sum
? "correct" : "incorrect");
1334 printf(" MPB Sectors : %d\n", mpb_sectors(mpb
));
1335 printf(" Disks : %d\n", mpb
->num_disks
);
1336 printf(" RAID Devices : %d\n", mpb
->num_raid_devs
);
1337 print_imsm_disk(__get_imsm_disk(mpb
, super
->disks
->index
), super
->disks
->index
, reserved
);
1338 if (super
->bbm_log
) {
1339 struct bbm_log
*log
= super
->bbm_log
;
1342 printf("Bad Block Management Log:\n");
1343 printf(" Log Size : %d\n", __le32_to_cpu(mpb
->bbm_log_size
));
1344 printf(" Signature : %x\n", __le32_to_cpu(log
->signature
));
1345 printf(" Entry Count : %d\n", __le32_to_cpu(log
->entry_count
));
1346 printf(" Spare Blocks : %d\n", __le32_to_cpu(log
->reserved_spare_block_count
));
1347 printf(" First Spare : %llx\n",
1348 (unsigned long long) __le64_to_cpu(log
->first_spare_lba
));
1350 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
1352 struct imsm_dev
*dev
= __get_imsm_dev(mpb
, i
);
1354 super
->current_vol
= i
;
1355 getinfo_super_imsm(st
, &info
, NULL
);
1356 fname_from_uuid(st
, &info
, nbuf
, ':');
1357 print_imsm_dev(super
, dev
, nbuf
+ 5, super
->disks
->index
);
1359 for (i
= 0; i
< mpb
->num_disks
; i
++) {
1360 if (i
== super
->disks
->index
)
1362 print_imsm_disk(__get_imsm_disk(mpb
, i
), i
, reserved
);
1365 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
1366 if (dl
->index
== -1)
1367 print_imsm_disk(&dl
->disk
, -1, reserved
);
1369 examine_migr_rec_imsm(super
);
1372 static void brief_examine_super_imsm(struct supertype
*st
, int verbose
)
1374 /* We just write a generic IMSM ARRAY entry */
1377 struct intel_super
*super
= st
->sb
;
1379 if (!super
->anchor
->num_raid_devs
) {
1380 printf("ARRAY metadata=imsm\n");
1384 getinfo_super_imsm(st
, &info
, NULL
);
1385 fname_from_uuid(st
, &info
, nbuf
, ':');
1386 printf("ARRAY metadata=imsm UUID=%s\n", nbuf
+ 5);
1389 static void brief_examine_subarrays_imsm(struct supertype
*st
, int verbose
)
1391 /* We just write a generic IMSM ARRAY entry */
1395 struct intel_super
*super
= st
->sb
;
1398 if (!super
->anchor
->num_raid_devs
)
1401 getinfo_super_imsm(st
, &info
, NULL
);
1402 fname_from_uuid(st
, &info
, nbuf
, ':');
1403 for (i
= 0; i
< super
->anchor
->num_raid_devs
; i
++) {
1404 struct imsm_dev
*dev
= get_imsm_dev(super
, i
);
1406 super
->current_vol
= i
;
1407 getinfo_super_imsm(st
, &info
, NULL
);
1408 fname_from_uuid(st
, &info
, nbuf1
, ':');
1409 printf("ARRAY /dev/md/%.16s container=%s member=%d UUID=%s\n",
1410 dev
->volume
, nbuf
+ 5, i
, nbuf1
+ 5);
1414 static void export_examine_super_imsm(struct supertype
*st
)
1416 struct intel_super
*super
= st
->sb
;
1417 struct imsm_super
*mpb
= super
->anchor
;
1421 getinfo_super_imsm(st
, &info
, NULL
);
1422 fname_from_uuid(st
, &info
, nbuf
, ':');
1423 printf("MD_METADATA=imsm\n");
1424 printf("MD_LEVEL=container\n");
1425 printf("MD_UUID=%s\n", nbuf
+5);
1426 printf("MD_DEVICES=%u\n", mpb
->num_disks
);
1429 static void detail_super_imsm(struct supertype
*st
, char *homehost
)
1434 getinfo_super_imsm(st
, &info
, NULL
);
1435 fname_from_uuid(st
, &info
, nbuf
, ':');
1436 printf("\n UUID : %s\n", nbuf
+ 5);
1439 static void brief_detail_super_imsm(struct supertype
*st
)
1443 getinfo_super_imsm(st
, &info
, NULL
);
1444 fname_from_uuid(st
, &info
, nbuf
, ':');
1445 printf(" UUID=%s", nbuf
+ 5);
1448 static int imsm_read_serial(int fd
, char *devname
, __u8
*serial
);
1449 static void fd2devname(int fd
, char *name
);
1451 static int ahci_enumerate_ports(const char *hba_path
, int port_count
, int host_base
, int verbose
)
1453 /* dump an unsorted list of devices attached to AHCI Intel storage
1454 * controller, as well as non-connected ports
1456 int hba_len
= strlen(hba_path
) + 1;
1461 unsigned long port_mask
= (1 << port_count
) - 1;
1463 if (port_count
> (int)sizeof(port_mask
) * 8) {
1465 fprintf(stderr
, Name
": port_count %d out of range\n", port_count
);
1469 /* scroll through /sys/dev/block looking for devices attached to
1472 dir
= opendir("/sys/dev/block");
1473 for (ent
= dir
? readdir(dir
) : NULL
; ent
; ent
= readdir(dir
)) {
1484 if (sscanf(ent
->d_name
, "%d:%d", &major
, &minor
) != 2)
1486 path
= devt_to_devpath(makedev(major
, minor
));
1489 if (!path_attached_to_hba(path
, hba_path
)) {
1495 /* retrieve the scsi device type */
1496 if (asprintf(&device
, "/sys/dev/block/%d:%d/device/xxxxxxx", major
, minor
) < 0) {
1498 fprintf(stderr
, Name
": failed to allocate 'device'\n");
1502 sprintf(device
, "/sys/dev/block/%d:%d/device/type", major
, minor
);
1503 if (load_sys(device
, buf
) != 0) {
1505 fprintf(stderr
, Name
": failed to read device type for %s\n",
1511 type
= strtoul(buf
, NULL
, 10);
1513 /* if it's not a disk print the vendor and model */
1514 if (!(type
== 0 || type
== 7 || type
== 14)) {
1517 sprintf(device
, "/sys/dev/block/%d:%d/device/vendor", major
, minor
);
1518 if (load_sys(device
, buf
) == 0) {
1519 strncpy(vendor
, buf
, sizeof(vendor
));
1520 vendor
[sizeof(vendor
) - 1] = '\0';
1521 c
= (char *) &vendor
[sizeof(vendor
) - 1];
1522 while (isspace(*c
) || *c
== '\0')
1526 sprintf(device
, "/sys/dev/block/%d:%d/device/model", major
, minor
);
1527 if (load_sys(device
, buf
) == 0) {
1528 strncpy(model
, buf
, sizeof(model
));
1529 model
[sizeof(model
) - 1] = '\0';
1530 c
= (char *) &model
[sizeof(model
) - 1];
1531 while (isspace(*c
) || *c
== '\0')
1535 if (vendor
[0] && model
[0])
1536 sprintf(buf
, "%.64s %.64s", vendor
, model
);
1538 switch (type
) { /* numbers from hald/linux/device.c */
1539 case 1: sprintf(buf
, "tape"); break;
1540 case 2: sprintf(buf
, "printer"); break;
1541 case 3: sprintf(buf
, "processor"); break;
1543 case 5: sprintf(buf
, "cdrom"); break;
1544 case 6: sprintf(buf
, "scanner"); break;
1545 case 8: sprintf(buf
, "media_changer"); break;
1546 case 9: sprintf(buf
, "comm"); break;
1547 case 12: sprintf(buf
, "raid"); break;
1548 default: sprintf(buf
, "unknown");
1554 /* chop device path to 'host%d' and calculate the port number */
1555 c
= strchr(&path
[hba_len
], '/');
1558 fprintf(stderr
, Name
": %s - invalid path name\n", path
+ hba_len
);
1563 if (sscanf(&path
[hba_len
], "host%d", &port
) == 1)
1567 *c
= '/'; /* repair the full string */
1568 fprintf(stderr
, Name
": failed to determine port number for %s\n",
1575 /* mark this port as used */
1576 port_mask
&= ~(1 << port
);
1578 /* print out the device information */
1580 printf(" Port%d : - non-disk device (%s) -\n", port
, buf
);
1584 fd
= dev_open(ent
->d_name
, O_RDONLY
);
1586 printf(" Port%d : - disk info unavailable -\n", port
);
1588 fd2devname(fd
, buf
);
1589 printf(" Port%d : %s", port
, buf
);
1590 if (imsm_read_serial(fd
, NULL
, (__u8
*) buf
) == 0)
1591 printf(" (%.*s)\n", MAX_RAID_SERIAL_LEN
, buf
);
1606 for (i
= 0; i
< port_count
; i
++)
1607 if (port_mask
& (1 << i
))
1608 printf(" Port%d : - no device attached -\n", i
);
1614 static void print_found_intel_controllers(struct sys_dev
*elem
)
1616 for (; elem
; elem
= elem
->next
) {
1617 fprintf(stderr
, Name
": found Intel(R) ");
1618 if (elem
->type
== SYS_DEV_SATA
)
1619 fprintf(stderr
, "SATA ");
1620 else if (elem
->type
== SYS_DEV_SAS
)
1621 fprintf(stderr
, "SAS ");
1622 fprintf(stderr
, "RAID controller");
1624 fprintf(stderr
, " at %s", elem
->pci_id
);
1625 fprintf(stderr
, ".\n");
1630 static int ahci_get_port_count(const char *hba_path
, int *port_count
)
1637 if ((dir
= opendir(hba_path
)) == NULL
)
1640 for (ent
= readdir(dir
); ent
; ent
= readdir(dir
)) {
1643 if (sscanf(ent
->d_name
, "host%d", &host
) != 1)
1645 if (*port_count
== 0)
1647 else if (host
< host_base
)
1650 if (host
+ 1 > *port_count
+ host_base
)
1651 *port_count
= host
+ 1 - host_base
;
1657 static void print_imsm_capability(const struct imsm_orom
*orom
)
1659 printf(" Platform : Intel(R) Matrix Storage Manager\n");
1660 printf(" Version : %d.%d.%d.%d\n", orom
->major_ver
, orom
->minor_ver
,
1661 orom
->hotfix_ver
, orom
->build
);
1662 printf(" RAID Levels :%s%s%s%s%s\n",
1663 imsm_orom_has_raid0(orom
) ? " raid0" : "",
1664 imsm_orom_has_raid1(orom
) ? " raid1" : "",
1665 imsm_orom_has_raid1e(orom
) ? " raid1e" : "",
1666 imsm_orom_has_raid10(orom
) ? " raid10" : "",
1667 imsm_orom_has_raid5(orom
) ? " raid5" : "");
1668 printf(" Chunk Sizes :%s%s%s%s%s%s%s%s%s%s%s%s%s%s%s%s\n",
1669 imsm_orom_has_chunk(orom
, 2) ? " 2k" : "",
1670 imsm_orom_has_chunk(orom
, 4) ? " 4k" : "",
1671 imsm_orom_has_chunk(orom
, 8) ? " 8k" : "",
1672 imsm_orom_has_chunk(orom
, 16) ? " 16k" : "",
1673 imsm_orom_has_chunk(orom
, 32) ? " 32k" : "",
1674 imsm_orom_has_chunk(orom
, 64) ? " 64k" : "",
1675 imsm_orom_has_chunk(orom
, 128) ? " 128k" : "",
1676 imsm_orom_has_chunk(orom
, 256) ? " 256k" : "",
1677 imsm_orom_has_chunk(orom
, 512) ? " 512k" : "",
1678 imsm_orom_has_chunk(orom
, 1024*1) ? " 1M" : "",
1679 imsm_orom_has_chunk(orom
, 1024*2) ? " 2M" : "",
1680 imsm_orom_has_chunk(orom
, 1024*4) ? " 4M" : "",
1681 imsm_orom_has_chunk(orom
, 1024*8) ? " 8M" : "",
1682 imsm_orom_has_chunk(orom
, 1024*16) ? " 16M" : "",
1683 imsm_orom_has_chunk(orom
, 1024*32) ? " 32M" : "",
1684 imsm_orom_has_chunk(orom
, 1024*64) ? " 64M" : "");
1685 printf(" Max Disks : %d\n", orom
->tds
);
1686 printf(" Max Volumes : %d\n", orom
->vpa
);
1690 static int detail_platform_imsm(int verbose
, int enumerate_only
)
1692 /* There are two components to imsm platform support, the ahci SATA
1693 * controller and the option-rom. To find the SATA controller we
1694 * simply look in /sys/bus/pci/drivers/ahci to see if an ahci
1695 * controller with the Intel vendor id is present. This approach
1696 * allows mdadm to leverage the kernel's ahci detection logic, with the
1697 * caveat that if ahci.ko is not loaded mdadm will not be able to
1698 * detect platform raid capabilities. The option-rom resides in a
1699 * platform "Adapter ROM". We scan for its signature to retrieve the
1700 * platform capabilities. If raid support is disabled in the BIOS the
1701 * option-rom capability structure will not be available.
1703 const struct imsm_orom
*orom
;
1704 struct sys_dev
*list
, *hba
;
1709 if (enumerate_only
) {
1710 if (check_env("IMSM_NO_PLATFORM"))
1712 list
= find_intel_devices();
1715 for (hba
= list
; hba
; hba
= hba
->next
) {
1716 orom
= find_imsm_capability(hba
->type
);
1722 free_sys_dev(&list
);
1726 list
= find_intel_devices();
1729 fprintf(stderr
, Name
": no active Intel(R) RAID "
1730 "controller found.\n");
1731 free_sys_dev(&list
);
1734 print_found_intel_controllers(list
);
1736 for (hba
= list
; hba
; hba
= hba
->next
) {
1737 orom
= find_imsm_capability(hba
->type
);
1739 fprintf(stderr
, Name
": imsm capabilities not found for controller: %s (type %s)\n",
1740 hba
->path
, get_sys_dev_type(hba
->type
));
1742 print_imsm_capability(orom
);
1745 for (hba
= list
; hba
; hba
= hba
->next
) {
1746 printf(" I/O Controller : %s (%s)\n",
1747 hba
->path
, get_sys_dev_type(hba
->type
));
1749 if (hba
->type
== SYS_DEV_SATA
) {
1750 host_base
= ahci_get_port_count(hba
->path
, &port_count
);
1751 if (ahci_enumerate_ports(hba
->path
, port_count
, host_base
, verbose
)) {
1753 fprintf(stderr
, Name
": failed to enumerate "
1754 "ports on SATA controller at %s.", hba
->pci_id
);
1760 free_sys_dev(&list
);
1765 static int match_home_imsm(struct supertype
*st
, char *homehost
)
1767 /* the imsm metadata format does not specify any host
1768 * identification information. We return -1 since we can never
1769 * confirm nor deny whether a given array is "meant" for this
1770 * host. We rely on compare_super and the 'family_num' fields to
1771 * exclude member disks that do not belong, and we rely on
1772 * mdadm.conf to specify the arrays that should be assembled.
1773 * Auto-assembly may still pick up "foreign" arrays.
1779 static void uuid_from_super_imsm(struct supertype
*st
, int uuid
[4])
1781 /* The uuid returned here is used for:
1782 * uuid to put into bitmap file (Create, Grow)
1783 * uuid for backup header when saving critical section (Grow)
1784 * comparing uuids when re-adding a device into an array
1785 * In these cases the uuid required is that of the data-array,
1786 * not the device-set.
1787 * uuid to recognise same set when adding a missing device back
1788 * to an array. This is a uuid for the device-set.
1790 * For each of these we can make do with a truncated
1791 * or hashed uuid rather than the original, as long as
1793 * In each case the uuid required is that of the data-array,
1794 * not the device-set.
1796 /* imsm does not track uuid's so we synthesis one using sha1 on
1797 * - The signature (Which is constant for all imsm array, but no matter)
1798 * - the orig_family_num of the container
1799 * - the index number of the volume
1800 * - the 'serial' number of the volume.
1801 * Hopefully these are all constant.
1803 struct intel_super
*super
= st
->sb
;
1806 struct sha1_ctx ctx
;
1807 struct imsm_dev
*dev
= NULL
;
1810 /* some mdadm versions failed to set ->orig_family_num, in which
1811 * case fall back to ->family_num. orig_family_num will be
1812 * fixed up with the first metadata update.
1814 family_num
= super
->anchor
->orig_family_num
;
1815 if (family_num
== 0)
1816 family_num
= super
->anchor
->family_num
;
1817 sha1_init_ctx(&ctx
);
1818 sha1_process_bytes(super
->anchor
->sig
, MPB_SIG_LEN
, &ctx
);
1819 sha1_process_bytes(&family_num
, sizeof(__u32
), &ctx
);
1820 if (super
->current_vol
>= 0)
1821 dev
= get_imsm_dev(super
, super
->current_vol
);
1823 __u32 vol
= super
->current_vol
;
1824 sha1_process_bytes(&vol
, sizeof(vol
), &ctx
);
1825 sha1_process_bytes(dev
->volume
, MAX_RAID_SERIAL_LEN
, &ctx
);
1827 sha1_finish_ctx(&ctx
, buf
);
1828 memcpy(uuid
, buf
, 4*4);
1833 get_imsm_numerical_version(struct imsm_super
*mpb
, int *m
, int *p
)
1835 __u8
*v
= get_imsm_version(mpb
);
1836 __u8
*end
= mpb
->sig
+ MAX_SIGNATURE_LENGTH
;
1837 char major
[] = { 0, 0, 0 };
1838 char minor
[] = { 0 ,0, 0 };
1839 char patch
[] = { 0, 0, 0 };
1840 char *ver_parse
[] = { major
, minor
, patch
};
1844 while (*v
!= '\0' && v
< end
) {
1845 if (*v
!= '.' && j
< 2)
1846 ver_parse
[i
][j
++] = *v
;
1854 *m
= strtol(minor
, NULL
, 0);
1855 *p
= strtol(patch
, NULL
, 0);
1859 static __u32
migr_strip_blocks_resync(struct imsm_dev
*dev
)
1861 /* migr_strip_size when repairing or initializing parity */
1862 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
1863 __u32 chunk
= __le32_to_cpu(map
->blocks_per_strip
);
1865 switch (get_imsm_raid_level(map
)) {
1870 return 128*1024 >> 9;
1874 static __u32
migr_strip_blocks_rebuild(struct imsm_dev
*dev
)
1876 /* migr_strip_size when rebuilding a degraded disk, no idea why
1877 * this is different than migr_strip_size_resync(), but it's good
1880 struct imsm_map
*map
= get_imsm_map(dev
, MAP_1
);
1881 __u32 chunk
= __le32_to_cpu(map
->blocks_per_strip
);
1883 switch (get_imsm_raid_level(map
)) {
1886 if (map
->num_members
% map
->num_domains
== 0)
1887 return 128*1024 >> 9;
1891 return max((__u32
) 64*1024 >> 9, chunk
);
1893 return 128*1024 >> 9;
1897 static __u32
num_stripes_per_unit_resync(struct imsm_dev
*dev
)
1899 struct imsm_map
*lo
= get_imsm_map(dev
, MAP_0
);
1900 struct imsm_map
*hi
= get_imsm_map(dev
, MAP_1
);
1901 __u32 lo_chunk
= __le32_to_cpu(lo
->blocks_per_strip
);
1902 __u32 hi_chunk
= __le32_to_cpu(hi
->blocks_per_strip
);
1904 return max((__u32
) 1, hi_chunk
/ lo_chunk
);
1907 static __u32
num_stripes_per_unit_rebuild(struct imsm_dev
*dev
)
1909 struct imsm_map
*lo
= get_imsm_map(dev
, MAP_0
);
1910 int level
= get_imsm_raid_level(lo
);
1912 if (level
== 1 || level
== 10) {
1913 struct imsm_map
*hi
= get_imsm_map(dev
, MAP_1
);
1915 return hi
->num_domains
;
1917 return num_stripes_per_unit_resync(dev
);
1920 static __u8
imsm_num_data_members(struct imsm_dev
*dev
, int second_map
)
1922 /* named 'imsm_' because raid0, raid1 and raid10
1923 * counter-intuitively have the same number of data disks
1925 struct imsm_map
*map
= get_imsm_map(dev
, second_map
);
1927 switch (get_imsm_raid_level(map
)) {
1931 return map
->num_members
;
1933 return map
->num_members
- 1;
1935 dprintf("%s: unsupported raid level\n", __func__
);
1940 static __u32
parity_segment_depth(struct imsm_dev
*dev
)
1942 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
1943 __u32 chunk
= __le32_to_cpu(map
->blocks_per_strip
);
1945 switch(get_imsm_raid_level(map
)) {
1948 return chunk
* map
->num_domains
;
1950 return chunk
* map
->num_members
;
1956 static __u32
map_migr_block(struct imsm_dev
*dev
, __u32 block
)
1958 struct imsm_map
*map
= get_imsm_map(dev
, MAP_1
);
1959 __u32 chunk
= __le32_to_cpu(map
->blocks_per_strip
);
1960 __u32 strip
= block
/ chunk
;
1962 switch (get_imsm_raid_level(map
)) {
1965 __u32 vol_strip
= (strip
* map
->num_domains
) + 1;
1966 __u32 vol_stripe
= vol_strip
/ map
->num_members
;
1968 return vol_stripe
* chunk
+ block
% chunk
;
1970 __u32 stripe
= strip
/ (map
->num_members
- 1);
1972 return stripe
* chunk
+ block
% chunk
;
1979 static __u64
blocks_per_migr_unit(struct intel_super
*super
,
1980 struct imsm_dev
*dev
)
1982 /* calculate the conversion factor between per member 'blocks'
1983 * (md/{resync,rebuild}_start) and imsm migration units, return
1984 * 0 for the 'not migrating' and 'unsupported migration' cases
1986 if (!dev
->vol
.migr_state
)
1989 switch (migr_type(dev
)) {
1990 case MIGR_GEN_MIGR
: {
1991 struct migr_record
*migr_rec
= super
->migr_rec
;
1992 return __le32_to_cpu(migr_rec
->blocks_per_unit
);
1997 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
1998 __u32 stripes_per_unit
;
1999 __u32 blocks_per_unit
;
2008 /* yes, this is really the translation of migr_units to
2009 * per-member blocks in the 'resync' case
2011 stripes_per_unit
= num_stripes_per_unit_resync(dev
);
2012 migr_chunk
= migr_strip_blocks_resync(dev
);
2013 disks
= imsm_num_data_members(dev
, MAP_0
);
2014 blocks_per_unit
= stripes_per_unit
* migr_chunk
* disks
;
2015 stripe
= __le16_to_cpu(map
->blocks_per_strip
) * disks
;
2016 segment
= blocks_per_unit
/ stripe
;
2017 block_rel
= blocks_per_unit
- segment
* stripe
;
2018 parity_depth
= parity_segment_depth(dev
);
2019 block_map
= map_migr_block(dev
, block_rel
);
2020 return block_map
+ parity_depth
* segment
;
2022 case MIGR_REBUILD
: {
2023 __u32 stripes_per_unit
;
2026 stripes_per_unit
= num_stripes_per_unit_rebuild(dev
);
2027 migr_chunk
= migr_strip_blocks_rebuild(dev
);
2028 return migr_chunk
* stripes_per_unit
;
2030 case MIGR_STATE_CHANGE
:
2036 static int imsm_level_to_layout(int level
)
2044 return ALGORITHM_LEFT_ASYMMETRIC
;
2051 /*******************************************************************************
2052 * Function: read_imsm_migr_rec
2053 * Description: Function reads imsm migration record from last sector of disk
2055 * fd : disk descriptor
2056 * super : metadata info
2060 ******************************************************************************/
2061 static int read_imsm_migr_rec(int fd
, struct intel_super
*super
)
2064 unsigned long long dsize
;
2066 get_dev_size(fd
, NULL
, &dsize
);
2067 if (lseek64(fd
, dsize
- 512, SEEK_SET
) < 0) {
2069 Name
": Cannot seek to anchor block: %s\n",
2073 if (read(fd
, super
->migr_rec_buf
, 512) != 512) {
2075 Name
": Cannot read migr record block: %s\n",
2085 /*******************************************************************************
2086 * Function: load_imsm_migr_rec
2087 * Description: Function reads imsm migration record (it is stored at the last
2090 * super : imsm internal array info
2091 * info : general array info
2095 ******************************************************************************/
2096 static int load_imsm_migr_rec(struct intel_super
*super
, struct mdinfo
*info
)
2099 struct dl
*dl
= NULL
;
2105 for (sd
= info
->devs
; sd
; sd
= sd
->next
) {
2106 /* read only from one of the first two slots */
2107 if ((sd
->disk
.raid_disk
> 1) ||
2108 (sd
->disk
.raid_disk
< 0))
2110 sprintf(nm
, "%d:%d", sd
->disk
.major
, sd
->disk
.minor
);
2111 fd
= dev_open(nm
, O_RDONLY
);
2117 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
2118 /* read only from one of the first two slots */
2121 sprintf(nm
, "%d:%d", dl
->major
, dl
->minor
);
2122 fd
= dev_open(nm
, O_RDONLY
);
2129 retval
= read_imsm_migr_rec(fd
, super
);
2138 /*******************************************************************************
2139 * function: imsm_create_metadata_checkpoint_update
2140 * Description: It creates update for checkpoint change.
2142 * super : imsm internal array info
2143 * u : pointer to prepared update
2146 * If length is equal to 0, input pointer u contains no update
2147 ******************************************************************************/
2148 static int imsm_create_metadata_checkpoint_update(
2149 struct intel_super
*super
,
2150 struct imsm_update_general_migration_checkpoint
**u
)
2153 int update_memory_size
= 0;
2155 dprintf("imsm_create_metadata_checkpoint_update(enter)\n");
2161 /* size of all update data without anchor */
2162 update_memory_size
=
2163 sizeof(struct imsm_update_general_migration_checkpoint
);
2165 *u
= calloc(1, update_memory_size
);
2167 dprintf("error: cannot get memory for "
2168 "imsm_create_metadata_checkpoint_update update\n");
2171 (*u
)->type
= update_general_migration_checkpoint
;
2172 (*u
)->curr_migr_unit
= __le32_to_cpu(super
->migr_rec
->curr_migr_unit
);
2173 dprintf("imsm_create_metadata_checkpoint_update: prepared for %u\n",
2174 (*u
)->curr_migr_unit
);
2176 return update_memory_size
;
2180 static void imsm_update_metadata_locally(struct supertype
*st
,
2181 void *buf
, int len
);
2183 /*******************************************************************************
2184 * Function: write_imsm_migr_rec
2185 * Description: Function writes imsm migration record
2186 * (at the last sector of disk)
2188 * super : imsm internal array info
2192 ******************************************************************************/
2193 static int write_imsm_migr_rec(struct supertype
*st
)
2195 struct intel_super
*super
= st
->sb
;
2196 unsigned long long dsize
;
2202 struct imsm_update_general_migration_checkpoint
*u
;
2204 for (sd
= super
->disks
; sd
; sd
= sd
->next
) {
2205 /* write to 2 first slots only */
2206 if ((sd
->index
< 0) || (sd
->index
> 1))
2208 sprintf(nm
, "%d:%d", sd
->major
, sd
->minor
);
2209 fd
= dev_open(nm
, O_RDWR
);
2212 get_dev_size(fd
, NULL
, &dsize
);
2213 if (lseek64(fd
, dsize
- 512, SEEK_SET
) < 0) {
2215 Name
": Cannot seek to anchor block: %s\n",
2219 if (write(fd
, super
->migr_rec_buf
, 512) != 512) {
2221 Name
": Cannot write migr record block: %s\n",
2228 /* update checkpoint information in metadata */
2229 len
= imsm_create_metadata_checkpoint_update(super
, &u
);
2232 dprintf("imsm: Cannot prepare update\n");
2235 /* update metadata locally */
2236 imsm_update_metadata_locally(st
, u
, len
);
2237 /* and possibly remotely */
2238 if (st
->update_tail
) {
2239 append_metadata_update(st
, u
, len
);
2240 /* during reshape we do all work inside metadata handler
2241 * manage_reshape(), so metadata update has to be triggered
2244 flush_metadata_updates(st
);
2245 st
->update_tail
= &st
->updates
;
2255 #endif /* MDASSEMBLE */
2257 /* spare/missing disks activations are not allowe when
2258 * array/container performs reshape operation, because
2259 * all arrays in container works on the same disks set
2261 int imsm_reshape_blocks_arrays_changes(struct intel_super
*super
)
2264 struct intel_dev
*i_dev
;
2265 struct imsm_dev
*dev
;
2267 /* check whole container
2269 for (i_dev
= super
->devlist
; i_dev
; i_dev
= i_dev
->next
) {
2271 if (is_gen_migration(dev
)) {
2272 /* No repair during any migration in container
2281 static void getinfo_super_imsm_volume(struct supertype
*st
, struct mdinfo
*info
, char *dmap
)
2283 struct intel_super
*super
= st
->sb
;
2284 struct migr_record
*migr_rec
= super
->migr_rec
;
2285 struct imsm_dev
*dev
= get_imsm_dev(super
, super
->current_vol
);
2286 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
2287 struct imsm_map
*prev_map
= get_imsm_map(dev
, MAP_1
);
2288 struct imsm_map
*map_to_analyse
= map
;
2291 unsigned int component_size_alligment
;
2292 int map_disks
= info
->array
.raid_disks
;
2294 memset(info
, 0, sizeof(*info
));
2296 map_to_analyse
= prev_map
;
2298 dl
= super
->current_disk
;
2300 info
->container_member
= super
->current_vol
;
2301 info
->array
.raid_disks
= map
->num_members
;
2302 info
->array
.level
= get_imsm_raid_level(map_to_analyse
);
2303 info
->array
.layout
= imsm_level_to_layout(info
->array
.level
);
2304 info
->array
.md_minor
= -1;
2305 info
->array
.ctime
= 0;
2306 info
->array
.utime
= 0;
2307 info
->array
.chunk_size
=
2308 __le16_to_cpu(map_to_analyse
->blocks_per_strip
) << 9;
2309 info
->array
.state
= !dev
->vol
.dirty
;
2310 info
->custom_array_size
= __le32_to_cpu(dev
->size_high
);
2311 info
->custom_array_size
<<= 32;
2312 info
->custom_array_size
|= __le32_to_cpu(dev
->size_low
);
2313 info
->recovery_blocked
= imsm_reshape_blocks_arrays_changes(st
->sb
);
2315 if (is_gen_migration(dev
)) {
2316 info
->reshape_active
= 1;
2317 info
->new_level
= get_imsm_raid_level(map
);
2318 info
->new_layout
= imsm_level_to_layout(info
->new_level
);
2319 info
->new_chunk
= __le16_to_cpu(map
->blocks_per_strip
) << 9;
2320 info
->delta_disks
= map
->num_members
- prev_map
->num_members
;
2321 if (info
->delta_disks
) {
2322 /* this needs to be applied to every array
2325 info
->reshape_active
= CONTAINER_RESHAPE
;
2327 /* We shape information that we give to md might have to be
2328 * modify to cope with md's requirement for reshaping arrays.
2329 * For example, when reshaping a RAID0, md requires it to be
2330 * presented as a degraded RAID4.
2331 * Also if a RAID0 is migrating to a RAID5 we need to specify
2332 * the array as already being RAID5, but the 'before' layout
2333 * is a RAID4-like layout.
2335 switch (info
->array
.level
) {
2337 switch(info
->new_level
) {
2339 /* conversion is happening as RAID4 */
2340 info
->array
.level
= 4;
2341 info
->array
.raid_disks
+= 1;
2344 /* conversion is happening as RAID5 */
2345 info
->array
.level
= 5;
2346 info
->array
.layout
= ALGORITHM_PARITY_N
;
2347 info
->delta_disks
-= 1;
2350 /* FIXME error message */
2351 info
->array
.level
= UnSet
;
2357 info
->new_level
= UnSet
;
2358 info
->new_layout
= UnSet
;
2359 info
->new_chunk
= info
->array
.chunk_size
;
2360 info
->delta_disks
= 0;
2364 info
->disk
.major
= dl
->major
;
2365 info
->disk
.minor
= dl
->minor
;
2366 info
->disk
.number
= dl
->index
;
2367 info
->disk
.raid_disk
= get_imsm_disk_slot(map_to_analyse
,
2371 info
->data_offset
= __le32_to_cpu(map_to_analyse
->pba_of_lba0
);
2372 info
->component_size
=
2373 __le32_to_cpu(map_to_analyse
->blocks_per_member
);
2375 /* check component size aligment
2377 component_size_alligment
=
2378 info
->component_size
% (info
->array
.chunk_size
/512);
2380 if (component_size_alligment
&&
2381 (info
->array
.level
!= 1) && (info
->array
.level
!= UnSet
)) {
2382 dprintf("imsm: reported component size alligned from %llu ",
2383 info
->component_size
);
2384 info
->component_size
-= component_size_alligment
;
2385 dprintf("to %llu (%i).\n",
2386 info
->component_size
, component_size_alligment
);
2389 memset(info
->uuid
, 0, sizeof(info
->uuid
));
2390 info
->recovery_start
= MaxSector
;
2392 info
->reshape_progress
= 0;
2393 info
->resync_start
= MaxSector
;
2394 if ((map_to_analyse
->map_state
== IMSM_T_STATE_UNINITIALIZED
||
2396 imsm_reshape_blocks_arrays_changes(super
) == 0) {
2397 info
->resync_start
= 0;
2399 if (dev
->vol
.migr_state
) {
2400 switch (migr_type(dev
)) {
2403 __u64 blocks_per_unit
= blocks_per_migr_unit(super
,
2405 __u64 units
= __le32_to_cpu(dev
->vol
.curr_migr_unit
);
2407 info
->resync_start
= blocks_per_unit
* units
;
2410 case MIGR_GEN_MIGR
: {
2411 __u64 blocks_per_unit
= blocks_per_migr_unit(super
,
2413 __u64 units
= __le32_to_cpu(migr_rec
->curr_migr_unit
);
2414 unsigned long long array_blocks
;
2417 if (__le32_to_cpu(migr_rec
->ascending_migr
) &&
2419 (__le32_to_cpu(migr_rec
->num_migr_units
)-1)) &&
2420 (super
->migr_rec
->rec_status
==
2421 __cpu_to_le32(UNIT_SRC_IN_CP_AREA
)))
2424 info
->reshape_progress
= blocks_per_unit
* units
;
2426 dprintf("IMSM: General Migration checkpoint : %llu "
2427 "(%llu) -> read reshape progress : %llu\n",
2428 (unsigned long long)units
,
2429 (unsigned long long)blocks_per_unit
,
2430 info
->reshape_progress
);
2432 used_disks
= imsm_num_data_members(dev
, MAP_1
);
2433 if (used_disks
> 0) {
2434 array_blocks
= map
->blocks_per_member
*
2436 /* round array size down to closest MB
2438 info
->custom_array_size
= (array_blocks
2439 >> SECT_PER_MB_SHIFT
)
2440 << SECT_PER_MB_SHIFT
;
2444 /* we could emulate the checkpointing of
2445 * 'sync_action=check' migrations, but for now
2446 * we just immediately complete them
2449 /* this is handled by container_content_imsm() */
2450 case MIGR_STATE_CHANGE
:
2451 /* FIXME handle other migrations */
2453 /* we are not dirty, so... */
2454 info
->resync_start
= MaxSector
;
2458 strncpy(info
->name
, (char *) dev
->volume
, MAX_RAID_SERIAL_LEN
);
2459 info
->name
[MAX_RAID_SERIAL_LEN
] = 0;
2461 info
->array
.major_version
= -1;
2462 info
->array
.minor_version
= -2;
2463 devname
= devnum2devname(st
->container_dev
);
2464 *info
->text_version
= '\0';
2466 sprintf(info
->text_version
, "/%s/%d", devname
, info
->container_member
);
2468 info
->safe_mode_delay
= 4000; /* 4 secs like the Matrix driver */
2469 uuid_from_super_imsm(st
, info
->uuid
);
2473 for (i
=0; i
<map_disks
; i
++) {
2475 if (i
< info
->array
.raid_disks
) {
2476 struct imsm_disk
*dsk
;
2477 j
= get_imsm_disk_idx(dev
, i
, MAP_X
);
2478 dsk
= get_imsm_disk(super
, j
);
2479 if (dsk
&& (dsk
->status
& CONFIGURED_DISK
))
2486 static __u8
imsm_check_degraded(struct intel_super
*super
, struct imsm_dev
*dev
,
2487 int failed
, int look_in_map
);
2489 static int imsm_count_failed(struct intel_super
*super
, struct imsm_dev
*dev
,
2492 static void manage_second_map(struct intel_super
*super
, struct imsm_dev
*dev
)
2494 if (is_gen_migration(dev
)) {
2497 struct imsm_map
*map2
= get_imsm_map(dev
, MAP_1
);
2499 failed
= imsm_count_failed(super
, dev
, MAP_1
);
2500 map_state
= imsm_check_degraded(super
, dev
, failed
, MAP_1
);
2501 if (map2
->map_state
!= map_state
) {
2502 map2
->map_state
= map_state
;
2503 super
->updates_pending
++;
2508 static struct imsm_disk
*get_imsm_missing(struct intel_super
*super
, __u8 index
)
2512 for (d
= super
->missing
; d
; d
= d
->next
)
2513 if (d
->index
== index
)
2518 static void getinfo_super_imsm(struct supertype
*st
, struct mdinfo
*info
, char *map
)
2520 struct intel_super
*super
= st
->sb
;
2521 struct imsm_disk
*disk
;
2522 int map_disks
= info
->array
.raid_disks
;
2523 int max_enough
= -1;
2525 struct imsm_super
*mpb
;
2527 if (super
->current_vol
>= 0) {
2528 getinfo_super_imsm_volume(st
, info
, map
);
2531 memset(info
, 0, sizeof(*info
));
2533 /* Set raid_disks to zero so that Assemble will always pull in valid
2536 info
->array
.raid_disks
= 0;
2537 info
->array
.level
= LEVEL_CONTAINER
;
2538 info
->array
.layout
= 0;
2539 info
->array
.md_minor
= -1;
2540 info
->array
.ctime
= 0; /* N/A for imsm */
2541 info
->array
.utime
= 0;
2542 info
->array
.chunk_size
= 0;
2544 info
->disk
.major
= 0;
2545 info
->disk
.minor
= 0;
2546 info
->disk
.raid_disk
= -1;
2547 info
->reshape_active
= 0;
2548 info
->array
.major_version
= -1;
2549 info
->array
.minor_version
= -2;
2550 strcpy(info
->text_version
, "imsm");
2551 info
->safe_mode_delay
= 0;
2552 info
->disk
.number
= -1;
2553 info
->disk
.state
= 0;
2555 info
->recovery_start
= MaxSector
;
2556 info
->recovery_blocked
= imsm_reshape_blocks_arrays_changes(st
->sb
);
2558 /* do we have the all the insync disks that we expect? */
2559 mpb
= super
->anchor
;
2561 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
2562 struct imsm_dev
*dev
= get_imsm_dev(super
, i
);
2563 int failed
, enough
, j
, missing
= 0;
2564 struct imsm_map
*map
;
2567 failed
= imsm_count_failed(super
, dev
, MAP_0
);
2568 state
= imsm_check_degraded(super
, dev
, failed
, MAP_0
);
2569 map
= get_imsm_map(dev
, MAP_0
);
2571 /* any newly missing disks?
2572 * (catches single-degraded vs double-degraded)
2574 for (j
= 0; j
< map
->num_members
; j
++) {
2575 __u32 ord
= get_imsm_ord_tbl_ent(dev
, j
, MAP_0
);
2576 __u32 idx
= ord_to_idx(ord
);
2578 if (!(ord
& IMSM_ORD_REBUILD
) &&
2579 get_imsm_missing(super
, idx
)) {
2585 if (state
== IMSM_T_STATE_FAILED
)
2587 else if (state
== IMSM_T_STATE_DEGRADED
&&
2588 (state
!= map
->map_state
|| missing
))
2590 else /* we're normal, or already degraded */
2593 /* in the missing/failed disk case check to see
2594 * if at least one array is runnable
2596 max_enough
= max(max_enough
, enough
);
2598 dprintf("%s: enough: %d\n", __func__
, max_enough
);
2599 info
->container_enough
= max_enough
;
2602 __u32 reserved
= imsm_reserved_sectors(super
, super
->disks
);
2604 disk
= &super
->disks
->disk
;
2605 info
->data_offset
= __le32_to_cpu(disk
->total_blocks
) - reserved
;
2606 info
->component_size
= reserved
;
2607 info
->disk
.state
= is_configured(disk
) ? (1 << MD_DISK_ACTIVE
) : 0;
2608 /* we don't change info->disk.raid_disk here because
2609 * this state will be finalized in mdmon after we have
2610 * found the 'most fresh' version of the metadata
2612 info
->disk
.state
|= is_failed(disk
) ? (1 << MD_DISK_FAULTY
) : 0;
2613 info
->disk
.state
|= is_spare(disk
) ? 0 : (1 << MD_DISK_SYNC
);
2616 /* only call uuid_from_super_imsm when this disk is part of a populated container,
2617 * ->compare_super may have updated the 'num_raid_devs' field for spares
2619 if (info
->disk
.state
& (1 << MD_DISK_SYNC
) || super
->anchor
->num_raid_devs
)
2620 uuid_from_super_imsm(st
, info
->uuid
);
2622 memcpy(info
->uuid
, uuid_zero
, sizeof(uuid_zero
));
2624 /* I don't know how to compute 'map' on imsm, so use safe default */
2627 for (i
= 0; i
< map_disks
; i
++)
2633 /* allocates memory and fills disk in mdinfo structure
2634 * for each disk in array */
2635 struct mdinfo
*getinfo_super_disks_imsm(struct supertype
*st
)
2637 struct mdinfo
*mddev
= NULL
;
2638 struct intel_super
*super
= st
->sb
;
2639 struct imsm_disk
*disk
;
2642 if (!super
|| !super
->disks
)
2645 mddev
= malloc(sizeof(*mddev
));
2647 fprintf(stderr
, Name
": Failed to allocate memory.\n");
2650 memset(mddev
, 0, sizeof(*mddev
));
2654 tmp
= malloc(sizeof(*tmp
));
2656 fprintf(stderr
, Name
": Failed to allocate memory.\n");
2661 memset(tmp
, 0, sizeof(*tmp
));
2663 tmp
->next
= mddev
->devs
;
2665 tmp
->disk
.number
= count
++;
2666 tmp
->disk
.major
= dl
->major
;
2667 tmp
->disk
.minor
= dl
->minor
;
2668 tmp
->disk
.state
= is_configured(disk
) ?
2669 (1 << MD_DISK_ACTIVE
) : 0;
2670 tmp
->disk
.state
|= is_failed(disk
) ? (1 << MD_DISK_FAULTY
) : 0;
2671 tmp
->disk
.state
|= is_spare(disk
) ? 0 : (1 << MD_DISK_SYNC
);
2672 tmp
->disk
.raid_disk
= -1;
2678 static int update_super_imsm(struct supertype
*st
, struct mdinfo
*info
,
2679 char *update
, char *devname
, int verbose
,
2680 int uuid_set
, char *homehost
)
2682 /* For 'assemble' and 'force' we need to return non-zero if any
2683 * change was made. For others, the return value is ignored.
2684 * Update options are:
2685 * force-one : This device looks a bit old but needs to be included,
2686 * update age info appropriately.
2687 * assemble: clear any 'faulty' flag to allow this device to
2689 * force-array: Array is degraded but being forced, mark it clean
2690 * if that will be needed to assemble it.
2692 * newdev: not used ????
2693 * grow: Array has gained a new device - this is currently for
2695 * resync: mark as dirty so a resync will happen.
2696 * name: update the name - preserving the homehost
2697 * uuid: Change the uuid of the array to match watch is given
2699 * Following are not relevant for this imsm:
2700 * sparc2.2 : update from old dodgey metadata
2701 * super-minor: change the preferred_minor number
2702 * summaries: update redundant counters.
2703 * homehost: update the recorded homehost
2704 * _reshape_progress: record new reshape_progress position.
2707 struct intel_super
*super
= st
->sb
;
2708 struct imsm_super
*mpb
;
2710 /* we can only update container info */
2711 if (!super
|| super
->current_vol
>= 0 || !super
->anchor
)
2714 mpb
= super
->anchor
;
2716 if (strcmp(update
, "uuid") == 0 && uuid_set
&& !info
->update_private
)
2718 else if (strcmp(update
, "uuid") == 0 && uuid_set
&& info
->update_private
) {
2719 mpb
->orig_family_num
= *((__u32
*) info
->update_private
);
2721 } else if (strcmp(update
, "uuid") == 0) {
2722 __u32
*new_family
= malloc(sizeof(*new_family
));
2724 /* update orig_family_number with the incoming random
2725 * data, report the new effective uuid, and store the
2726 * new orig_family_num for future updates.
2729 memcpy(&mpb
->orig_family_num
, info
->uuid
, sizeof(__u32
));
2730 uuid_from_super_imsm(st
, info
->uuid
);
2731 *new_family
= mpb
->orig_family_num
;
2732 info
->update_private
= new_family
;
2735 } else if (strcmp(update
, "assemble") == 0)
2740 /* successful update? recompute checksum */
2742 mpb
->check_sum
= __le32_to_cpu(__gen_imsm_checksum(mpb
));
2747 static size_t disks_to_mpb_size(int disks
)
2751 size
= sizeof(struct imsm_super
);
2752 size
+= (disks
- 1) * sizeof(struct imsm_disk
);
2753 size
+= 2 * sizeof(struct imsm_dev
);
2754 /* up to 2 maps per raid device (-2 for imsm_maps in imsm_dev */
2755 size
+= (4 - 2) * sizeof(struct imsm_map
);
2756 /* 4 possible disk_ord_tbl's */
2757 size
+= 4 * (disks
- 1) * sizeof(__u32
);
2762 static __u64
avail_size_imsm(struct supertype
*st
, __u64 devsize
)
2764 if (devsize
< (MPB_SECTOR_CNT
+ IMSM_RESERVED_SECTORS
))
2767 return devsize
- (MPB_SECTOR_CNT
+ IMSM_RESERVED_SECTORS
);
2770 static void free_devlist(struct intel_super
*super
)
2772 struct intel_dev
*dv
;
2774 while (super
->devlist
) {
2775 dv
= super
->devlist
->next
;
2776 free(super
->devlist
->dev
);
2777 free(super
->devlist
);
2778 super
->devlist
= dv
;
2782 static void imsm_copy_dev(struct imsm_dev
*dest
, struct imsm_dev
*src
)
2784 memcpy(dest
, src
, sizeof_imsm_dev(src
, 0));
2787 static int compare_super_imsm(struct supertype
*st
, struct supertype
*tst
)
2791 * 0 same, or first was empty, and second was copied
2792 * 1 second had wrong number
2794 * 3 wrong other info
2796 struct intel_super
*first
= st
->sb
;
2797 struct intel_super
*sec
= tst
->sb
;
2804 /* in platform dependent environment test if the disks
2805 * use the same Intel hba
2807 if (!check_env("IMSM_NO_PLATFORM")) {
2808 if (!first
->hba
|| !sec
->hba
||
2809 (first
->hba
->type
!= sec
->hba
->type
)) {
2811 "HBAs of devices does not match %s != %s\n",
2812 first
->hba
? get_sys_dev_type(first
->hba
->type
) : NULL
,
2813 sec
->hba
? get_sys_dev_type(sec
->hba
->type
) : NULL
);
2818 /* if an anchor does not have num_raid_devs set then it is a free
2821 if (first
->anchor
->num_raid_devs
> 0 &&
2822 sec
->anchor
->num_raid_devs
> 0) {
2823 /* Determine if these disks might ever have been
2824 * related. Further disambiguation can only take place
2825 * in load_super_imsm_all
2827 __u32 first_family
= first
->anchor
->orig_family_num
;
2828 __u32 sec_family
= sec
->anchor
->orig_family_num
;
2830 if (memcmp(first
->anchor
->sig
, sec
->anchor
->sig
,
2831 MAX_SIGNATURE_LENGTH
) != 0)
2834 if (first_family
== 0)
2835 first_family
= first
->anchor
->family_num
;
2836 if (sec_family
== 0)
2837 sec_family
= sec
->anchor
->family_num
;
2839 if (first_family
!= sec_family
)
2845 /* if 'first' is a spare promote it to a populated mpb with sec's
2848 if (first
->anchor
->num_raid_devs
== 0 &&
2849 sec
->anchor
->num_raid_devs
> 0) {
2851 struct intel_dev
*dv
;
2852 struct imsm_dev
*dev
;
2854 /* we need to copy raid device info from sec if an allocation
2855 * fails here we don't associate the spare
2857 for (i
= 0; i
< sec
->anchor
->num_raid_devs
; i
++) {
2858 dv
= malloc(sizeof(*dv
));
2861 dev
= malloc(sizeof_imsm_dev(get_imsm_dev(sec
, i
), 1));
2868 dv
->next
= first
->devlist
;
2869 first
->devlist
= dv
;
2871 if (i
< sec
->anchor
->num_raid_devs
) {
2872 /* allocation failure */
2873 free_devlist(first
);
2874 fprintf(stderr
, "imsm: failed to associate spare\n");
2877 first
->anchor
->num_raid_devs
= sec
->anchor
->num_raid_devs
;
2878 first
->anchor
->orig_family_num
= sec
->anchor
->orig_family_num
;
2879 first
->anchor
->family_num
= sec
->anchor
->family_num
;
2880 memcpy(first
->anchor
->sig
, sec
->anchor
->sig
, MAX_SIGNATURE_LENGTH
);
2881 for (i
= 0; i
< sec
->anchor
->num_raid_devs
; i
++)
2882 imsm_copy_dev(get_imsm_dev(first
, i
), get_imsm_dev(sec
, i
));
2888 static void fd2devname(int fd
, char *name
)
2892 char dname
[PATH_MAX
];
2897 if (fstat(fd
, &st
) != 0)
2899 sprintf(path
, "/sys/dev/block/%d:%d",
2900 major(st
.st_rdev
), minor(st
.st_rdev
));
2902 rv
= readlink(path
, dname
, sizeof(dname
)-1);
2907 nm
= strrchr(dname
, '/');
2910 snprintf(name
, MAX_RAID_SERIAL_LEN
, "/dev/%s", nm
);
2914 extern int scsi_get_serial(int fd
, void *buf
, size_t buf_len
);
2916 static int imsm_read_serial(int fd
, char *devname
,
2917 __u8 serial
[MAX_RAID_SERIAL_LEN
])
2919 unsigned char scsi_serial
[255];
2928 memset(scsi_serial
, 0, sizeof(scsi_serial
));
2930 rv
= scsi_get_serial(fd
, scsi_serial
, sizeof(scsi_serial
));
2932 if (rv
&& check_env("IMSM_DEVNAME_AS_SERIAL")) {
2933 memset(serial
, 0, MAX_RAID_SERIAL_LEN
);
2934 fd2devname(fd
, (char *) serial
);
2941 Name
": Failed to retrieve serial for %s\n",
2946 rsp_len
= scsi_serial
[3];
2950 Name
": Failed to retrieve serial for %s\n",
2954 rsp_buf
= (char *) &scsi_serial
[4];
2956 /* trim all whitespace and non-printable characters and convert
2959 for (i
= 0, dest
= rsp_buf
; i
< rsp_len
; i
++) {
2962 /* ':' is reserved for use in placeholder serial
2963 * numbers for missing disks
2971 len
= dest
- rsp_buf
;
2974 /* truncate leading characters */
2975 if (len
> MAX_RAID_SERIAL_LEN
) {
2976 dest
+= len
- MAX_RAID_SERIAL_LEN
;
2977 len
= MAX_RAID_SERIAL_LEN
;
2980 memset(serial
, 0, MAX_RAID_SERIAL_LEN
);
2981 memcpy(serial
, dest
, len
);
2986 static int serialcmp(__u8
*s1
, __u8
*s2
)
2988 return strncmp((char *) s1
, (char *) s2
, MAX_RAID_SERIAL_LEN
);
2991 static void serialcpy(__u8
*dest
, __u8
*src
)
2993 strncpy((char *) dest
, (char *) src
, MAX_RAID_SERIAL_LEN
);
2996 static struct dl
*serial_to_dl(__u8
*serial
, struct intel_super
*super
)
3000 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
3001 if (serialcmp(dl
->serial
, serial
) == 0)
3007 static struct imsm_disk
*
3008 __serial_to_disk(__u8
*serial
, struct imsm_super
*mpb
, int *idx
)
3012 for (i
= 0; i
< mpb
->num_disks
; i
++) {
3013 struct imsm_disk
*disk
= __get_imsm_disk(mpb
, i
);
3015 if (serialcmp(disk
->serial
, serial
) == 0) {
3026 load_imsm_disk(int fd
, struct intel_super
*super
, char *devname
, int keep_fd
)
3028 struct imsm_disk
*disk
;
3033 __u8 serial
[MAX_RAID_SERIAL_LEN
];
3035 rv
= imsm_read_serial(fd
, devname
, serial
);
3040 dl
= calloc(1, sizeof(*dl
));
3044 Name
": failed to allocate disk buffer for %s\n",
3050 dl
->major
= major(stb
.st_rdev
);
3051 dl
->minor
= minor(stb
.st_rdev
);
3052 dl
->next
= super
->disks
;
3053 dl
->fd
= keep_fd
? fd
: -1;
3054 assert(super
->disks
== NULL
);
3056 serialcpy(dl
->serial
, serial
);
3059 fd2devname(fd
, name
);
3061 dl
->devname
= strdup(devname
);
3063 dl
->devname
= strdup(name
);
3065 /* look up this disk's index in the current anchor */
3066 disk
= __serial_to_disk(dl
->serial
, super
->anchor
, &dl
->index
);
3069 /* only set index on disks that are a member of a
3070 * populated contianer, i.e. one with raid_devs
3072 if (is_failed(&dl
->disk
))
3074 else if (is_spare(&dl
->disk
))
3082 /* When migrating map0 contains the 'destination' state while map1
3083 * contains the current state. When not migrating map0 contains the
3084 * current state. This routine assumes that map[0].map_state is set to
3085 * the current array state before being called.
3087 * Migration is indicated by one of the following states
3088 * 1/ Idle (migr_state=0 map0state=normal||unitialized||degraded||failed)
3089 * 2/ Initialize (migr_state=1 migr_type=MIGR_INIT map0state=normal
3090 * map1state=unitialized)
3091 * 3/ Repair (Resync) (migr_state=1 migr_type=MIGR_REPAIR map0state=normal
3093 * 4/ Rebuild (migr_state=1 migr_type=MIGR_REBUILD map0state=normal
3094 * map1state=degraded)
3095 * 5/ Migration (mig_state=1 migr_type=MIGR_GEN_MIGR map0state=normal
3098 static void migrate(struct imsm_dev
*dev
, struct intel_super
*super
,
3099 __u8 to_state
, int migr_type
)
3101 struct imsm_map
*dest
;
3102 struct imsm_map
*src
= get_imsm_map(dev
, MAP_0
);
3104 dev
->vol
.migr_state
= 1;
3105 set_migr_type(dev
, migr_type
);
3106 dev
->vol
.curr_migr_unit
= 0;
3107 dest
= get_imsm_map(dev
, MAP_1
);
3109 /* duplicate and then set the target end state in map[0] */
3110 memcpy(dest
, src
, sizeof_imsm_map(src
));
3111 if ((migr_type
== MIGR_REBUILD
) ||
3112 (migr_type
== MIGR_GEN_MIGR
)) {
3116 for (i
= 0; i
< src
->num_members
; i
++) {
3117 ord
= __le32_to_cpu(src
->disk_ord_tbl
[i
]);
3118 set_imsm_ord_tbl_ent(src
, i
, ord_to_idx(ord
));
3122 if (migr_type
== MIGR_GEN_MIGR
)
3123 /* Clear migration record */
3124 memset(super
->migr_rec
, 0, sizeof(struct migr_record
));
3126 src
->map_state
= to_state
;
3129 static void end_migration(struct imsm_dev
*dev
, struct intel_super
*super
,
3132 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
3133 struct imsm_map
*prev
= get_imsm_map(dev
, dev
->vol
.migr_state
== 0 ?
3137 /* merge any IMSM_ORD_REBUILD bits that were not successfully
3138 * completed in the last migration.
3140 * FIXME add support for raid-level-migration
3142 if ((map_state
!= map
->map_state
) && (is_gen_migration(dev
) == 0) &&
3143 (prev
->map_state
!= IMSM_T_STATE_UNINITIALIZED
)) {
3144 /* when final map state is other than expected
3145 * merge maps (not for migration)
3149 for (i
= 0; i
< prev
->num_members
; i
++)
3150 for (j
= 0; j
< map
->num_members
; j
++)
3151 /* during online capacity expansion
3152 * disks position can be changed
3153 * if takeover is used
3155 if (ord_to_idx(map
->disk_ord_tbl
[j
]) ==
3156 ord_to_idx(prev
->disk_ord_tbl
[i
])) {
3157 map
->disk_ord_tbl
[j
] |=
3158 prev
->disk_ord_tbl
[i
];
3161 failed
= imsm_count_failed(super
, dev
, MAP_0
);
3162 map_state
= imsm_check_degraded(super
, dev
, failed
, MAP_0
);
3165 dev
->vol
.migr_state
= 0;
3166 set_migr_type(dev
, 0);
3167 dev
->vol
.curr_migr_unit
= 0;
3168 map
->map_state
= map_state
;
3172 static int parse_raid_devices(struct intel_super
*super
)
3175 struct imsm_dev
*dev_new
;
3176 size_t len
, len_migr
;
3178 size_t space_needed
= 0;
3179 struct imsm_super
*mpb
= super
->anchor
;
3181 for (i
= 0; i
< super
->anchor
->num_raid_devs
; i
++) {
3182 struct imsm_dev
*dev_iter
= __get_imsm_dev(super
->anchor
, i
);
3183 struct intel_dev
*dv
;
3185 len
= sizeof_imsm_dev(dev_iter
, 0);
3186 len_migr
= sizeof_imsm_dev(dev_iter
, 1);
3188 space_needed
+= len_migr
- len
;
3190 dv
= malloc(sizeof(*dv
));
3193 if (max_len
< len_migr
)
3195 if (max_len
> len_migr
)
3196 space_needed
+= max_len
- len_migr
;
3197 dev_new
= malloc(max_len
);
3202 imsm_copy_dev(dev_new
, dev_iter
);
3205 dv
->next
= super
->devlist
;
3206 super
->devlist
= dv
;
3209 /* ensure that super->buf is large enough when all raid devices
3212 if (__le32_to_cpu(mpb
->mpb_size
) + space_needed
> super
->len
) {
3215 len
= ROUND_UP(__le32_to_cpu(mpb
->mpb_size
) + space_needed
, 512);
3216 if (posix_memalign(&buf
, 512, len
) != 0)
3219 memcpy(buf
, super
->buf
, super
->len
);
3220 memset(buf
+ super
->len
, 0, len
- super
->len
);
3229 /* retrieve a pointer to the bbm log which starts after all raid devices */
3230 struct bbm_log
*__get_imsm_bbm_log(struct imsm_super
*mpb
)
3234 if (__le32_to_cpu(mpb
->bbm_log_size
)) {
3236 ptr
+= mpb
->mpb_size
- __le32_to_cpu(mpb
->bbm_log_size
);
3242 /*******************************************************************************
3243 * Function: check_mpb_migr_compatibility
3244 * Description: Function checks for unsupported migration features:
3245 * - migration optimization area (pba_of_lba0)
3246 * - descending reshape (ascending_migr)
3248 * super : imsm metadata information
3250 * 0 : migration is compatible
3251 * -1 : migration is not compatible
3252 ******************************************************************************/
3253 int check_mpb_migr_compatibility(struct intel_super
*super
)
3255 struct imsm_map
*map0
, *map1
;
3256 struct migr_record
*migr_rec
= super
->migr_rec
;
3259 for (i
= 0; i
< super
->anchor
->num_raid_devs
; i
++) {
3260 struct imsm_dev
*dev_iter
= __get_imsm_dev(super
->anchor
, i
);
3263 dev_iter
->vol
.migr_state
== 1 &&
3264 dev_iter
->vol
.migr_type
== MIGR_GEN_MIGR
) {
3265 /* This device is migrating */
3266 map0
= get_imsm_map(dev_iter
, MAP_0
);
3267 map1
= get_imsm_map(dev_iter
, MAP_1
);
3268 if (map0
->pba_of_lba0
!= map1
->pba_of_lba0
)
3269 /* migration optimization area was used */
3271 if (migr_rec
->ascending_migr
== 0
3272 && migr_rec
->dest_depth_per_unit
> 0)
3273 /* descending reshape not supported yet */
3280 static void __free_imsm(struct intel_super
*super
, int free_disks
);
3282 /* load_imsm_mpb - read matrix metadata
3283 * allocates super->mpb to be freed by free_imsm
3285 static int load_imsm_mpb(int fd
, struct intel_super
*super
, char *devname
)
3287 unsigned long long dsize
;
3288 unsigned long long sectors
;
3290 struct imsm_super
*anchor
;
3293 get_dev_size(fd
, NULL
, &dsize
);
3297 Name
": %s: device to small for imsm\n",
3302 if (lseek64(fd
, dsize
- (512 * 2), SEEK_SET
) < 0) {
3304 fprintf(stderr
, Name
3305 ": Cannot seek to anchor block on %s: %s\n",
3306 devname
, strerror(errno
));
3310 if (posix_memalign((void**)&anchor
, 512, 512) != 0) {
3313 Name
": Failed to allocate imsm anchor buffer"
3314 " on %s\n", devname
);
3317 if (read(fd
, anchor
, 512) != 512) {
3320 Name
": Cannot read anchor block on %s: %s\n",
3321 devname
, strerror(errno
));
3326 if (strncmp((char *) anchor
->sig
, MPB_SIGNATURE
, MPB_SIG_LEN
) != 0) {
3329 Name
": no IMSM anchor on %s\n", devname
);
3334 __free_imsm(super
, 0);
3335 /* reload capability and hba */
3337 /* capability and hba must be updated with new super allocation */
3338 find_intel_hba_capability(fd
, super
, devname
);
3339 super
->len
= ROUND_UP(anchor
->mpb_size
, 512);
3340 if (posix_memalign(&super
->buf
, 512, super
->len
) != 0) {
3343 Name
": unable to allocate %zu byte mpb buffer\n",
3348 memcpy(super
->buf
, anchor
, 512);
3350 sectors
= mpb_sectors(anchor
) - 1;
3353 if (posix_memalign(&super
->migr_rec_buf
, 512, 512) != 0) {
3354 fprintf(stderr
, Name
3355 ": %s could not allocate migr_rec buffer\n", __func__
);
3361 check_sum
= __gen_imsm_checksum(super
->anchor
);
3362 if (check_sum
!= __le32_to_cpu(super
->anchor
->check_sum
)) {
3365 Name
": IMSM checksum %x != %x on %s\n",
3367 __le32_to_cpu(super
->anchor
->check_sum
),
3375 /* read the extended mpb */
3376 if (lseek64(fd
, dsize
- (512 * (2 + sectors
)), SEEK_SET
) < 0) {
3379 Name
": Cannot seek to extended mpb on %s: %s\n",
3380 devname
, strerror(errno
));
3384 if ((unsigned)read(fd
, super
->buf
+ 512, super
->len
- 512) != super
->len
- 512) {
3387 Name
": Cannot read extended mpb on %s: %s\n",
3388 devname
, strerror(errno
));
3392 check_sum
= __gen_imsm_checksum(super
->anchor
);
3393 if (check_sum
!= __le32_to_cpu(super
->anchor
->check_sum
)) {
3396 Name
": IMSM checksum %x != %x on %s\n",
3397 check_sum
, __le32_to_cpu(super
->anchor
->check_sum
),
3402 /* FIXME the BBM log is disk specific so we cannot use this global
3403 * buffer for all disks. Ok for now since we only look at the global
3404 * bbm_log_size parameter to gate assembly
3406 super
->bbm_log
= __get_imsm_bbm_log(super
->anchor
);
3411 static int read_imsm_migr_rec(int fd
, struct intel_super
*super
);
3414 load_and_parse_mpb(int fd
, struct intel_super
*super
, char *devname
, int keep_fd
)
3418 err
= load_imsm_mpb(fd
, super
, devname
);
3421 err
= load_imsm_disk(fd
, super
, devname
, keep_fd
);
3424 err
= parse_raid_devices(super
);
3429 static void __free_imsm_disk(struct dl
*d
)
3441 static void free_imsm_disks(struct intel_super
*super
)
3445 while (super
->disks
) {
3447 super
->disks
= d
->next
;
3448 __free_imsm_disk(d
);
3450 while (super
->disk_mgmt_list
) {
3451 d
= super
->disk_mgmt_list
;
3452 super
->disk_mgmt_list
= d
->next
;
3453 __free_imsm_disk(d
);
3455 while (super
->missing
) {
3457 super
->missing
= d
->next
;
3458 __free_imsm_disk(d
);
3463 /* free all the pieces hanging off of a super pointer */
3464 static void __free_imsm(struct intel_super
*super
, int free_disks
)
3466 struct intel_hba
*elem
, *next
;
3472 /* unlink capability description */
3474 if (super
->migr_rec_buf
) {
3475 free(super
->migr_rec_buf
);
3476 super
->migr_rec_buf
= NULL
;
3479 free_imsm_disks(super
);
3480 free_devlist(super
);
3484 free((void *)elem
->path
);
3492 static void free_imsm(struct intel_super
*super
)
3494 __free_imsm(super
, 1);
3498 static void free_super_imsm(struct supertype
*st
)
3500 struct intel_super
*super
= st
->sb
;
3509 static struct intel_super
*alloc_super(void)
3511 struct intel_super
*super
= malloc(sizeof(*super
));
3514 memset(super
, 0, sizeof(*super
));
3515 super
->current_vol
= -1;
3516 super
->create_offset
= ~((__u32
) 0);
3522 * find and allocate hba and OROM/EFI based on valid fd of RAID component device
3524 static int find_intel_hba_capability(int fd
, struct intel_super
*super
, char *devname
)
3526 struct sys_dev
*hba_name
;
3529 if ((fd
< 0) || check_env("IMSM_NO_PLATFORM")) {
3534 hba_name
= find_disk_attached_hba(fd
, NULL
);
3538 Name
": %s is not attached to Intel(R) RAID controller.\n",
3542 rv
= attach_hba_to_super(super
, hba_name
);
3545 struct intel_hba
*hba
= super
->hba
;
3547 fprintf(stderr
, Name
": %s is attached to Intel(R) %s RAID "
3548 "controller (%s),\n"
3549 " but the container is assigned to Intel(R) "
3550 "%s RAID controller (",
3553 hba_name
->pci_id
? : "Err!",
3554 get_sys_dev_type(hba_name
->type
));
3557 fprintf(stderr
, "%s", hba
->pci_id
? : "Err!");
3559 fprintf(stderr
, ", ");
3563 fprintf(stderr
, ").\n"
3564 " Mixing devices attached to different controllers "
3565 "is not allowed.\n");
3567 free_sys_dev(&hba_name
);
3570 super
->orom
= find_imsm_capability(hba_name
->type
);
3571 free_sys_dev(&hba_name
);
3577 /* find_missing - helper routine for load_super_imsm_all that identifies
3578 * disks that have disappeared from the system. This routine relies on
3579 * the mpb being uptodate, which it is at load time.
3581 static int find_missing(struct intel_super
*super
)
3584 struct imsm_super
*mpb
= super
->anchor
;
3586 struct imsm_disk
*disk
;
3588 for (i
= 0; i
< mpb
->num_disks
; i
++) {
3589 disk
= __get_imsm_disk(mpb
, i
);
3590 dl
= serial_to_dl(disk
->serial
, super
);
3594 dl
= malloc(sizeof(*dl
));
3600 dl
->devname
= strdup("missing");
3602 serialcpy(dl
->serial
, disk
->serial
);
3605 dl
->next
= super
->missing
;
3606 super
->missing
= dl
;
3613 static struct intel_disk
*disk_list_get(__u8
*serial
, struct intel_disk
*disk_list
)
3615 struct intel_disk
*idisk
= disk_list
;
3618 if (serialcmp(idisk
->disk
.serial
, serial
) == 0)
3620 idisk
= idisk
->next
;
3626 static int __prep_thunderdome(struct intel_super
**table
, int tbl_size
,
3627 struct intel_super
*super
,
3628 struct intel_disk
**disk_list
)
3630 struct imsm_disk
*d
= &super
->disks
->disk
;
3631 struct imsm_super
*mpb
= super
->anchor
;
3634 for (i
= 0; i
< tbl_size
; i
++) {
3635 struct imsm_super
*tbl_mpb
= table
[i
]->anchor
;
3636 struct imsm_disk
*tbl_d
= &table
[i
]->disks
->disk
;
3638 if (tbl_mpb
->family_num
== mpb
->family_num
) {
3639 if (tbl_mpb
->check_sum
== mpb
->check_sum
) {
3640 dprintf("%s: mpb from %d:%d matches %d:%d\n",
3641 __func__
, super
->disks
->major
,
3642 super
->disks
->minor
,
3643 table
[i
]->disks
->major
,
3644 table
[i
]->disks
->minor
);
3648 if (((is_configured(d
) && !is_configured(tbl_d
)) ||
3649 is_configured(d
) == is_configured(tbl_d
)) &&
3650 tbl_mpb
->generation_num
< mpb
->generation_num
) {
3651 /* current version of the mpb is a
3652 * better candidate than the one in
3653 * super_table, but copy over "cross
3654 * generational" status
3656 struct intel_disk
*idisk
;
3658 dprintf("%s: mpb from %d:%d replaces %d:%d\n",
3659 __func__
, super
->disks
->major
,
3660 super
->disks
->minor
,
3661 table
[i
]->disks
->major
,
3662 table
[i
]->disks
->minor
);
3664 idisk
= disk_list_get(tbl_d
->serial
, *disk_list
);
3665 if (idisk
&& is_failed(&idisk
->disk
))
3666 tbl_d
->status
|= FAILED_DISK
;
3669 struct intel_disk
*idisk
;
3670 struct imsm_disk
*disk
;
3672 /* tbl_mpb is more up to date, but copy
3673 * over cross generational status before
3676 disk
= __serial_to_disk(d
->serial
, mpb
, NULL
);
3677 if (disk
&& is_failed(disk
))
3678 d
->status
|= FAILED_DISK
;
3680 idisk
= disk_list_get(d
->serial
, *disk_list
);
3683 if (disk
&& is_configured(disk
))
3684 idisk
->disk
.status
|= CONFIGURED_DISK
;
3687 dprintf("%s: mpb from %d:%d prefer %d:%d\n",
3688 __func__
, super
->disks
->major
,
3689 super
->disks
->minor
,
3690 table
[i
]->disks
->major
,
3691 table
[i
]->disks
->minor
);
3699 table
[tbl_size
++] = super
;
3703 /* update/extend the merged list of imsm_disk records */
3704 for (j
= 0; j
< mpb
->num_disks
; j
++) {
3705 struct imsm_disk
*disk
= __get_imsm_disk(mpb
, j
);
3706 struct intel_disk
*idisk
;
3708 idisk
= disk_list_get(disk
->serial
, *disk_list
);
3710 idisk
->disk
.status
|= disk
->status
;
3711 if (is_configured(&idisk
->disk
) ||
3712 is_failed(&idisk
->disk
))
3713 idisk
->disk
.status
&= ~(SPARE_DISK
);
3715 idisk
= calloc(1, sizeof(*idisk
));
3718 idisk
->owner
= IMSM_UNKNOWN_OWNER
;
3719 idisk
->disk
= *disk
;
3720 idisk
->next
= *disk_list
;
3724 if (serialcmp(idisk
->disk
.serial
, d
->serial
) == 0)
3731 static struct intel_super
*
3732 validate_members(struct intel_super
*super
, struct intel_disk
*disk_list
,
3735 struct imsm_super
*mpb
= super
->anchor
;
3739 for (i
= 0; i
< mpb
->num_disks
; i
++) {
3740 struct imsm_disk
*disk
= __get_imsm_disk(mpb
, i
);
3741 struct intel_disk
*idisk
;
3743 idisk
= disk_list_get(disk
->serial
, disk_list
);
3745 if (idisk
->owner
== owner
||
3746 idisk
->owner
== IMSM_UNKNOWN_OWNER
)
3749 dprintf("%s: '%.16s' owner %d != %d\n",
3750 __func__
, disk
->serial
, idisk
->owner
,
3753 dprintf("%s: unknown disk %x [%d]: %.16s\n",
3754 __func__
, __le32_to_cpu(mpb
->family_num
), i
,
3760 if (ok_count
== mpb
->num_disks
)
3765 static void show_conflicts(__u32 family_num
, struct intel_super
*super_list
)
3767 struct intel_super
*s
;
3769 for (s
= super_list
; s
; s
= s
->next
) {
3770 if (family_num
!= s
->anchor
->family_num
)
3772 fprintf(stderr
, "Conflict, offlining family %#x on '%s'\n",
3773 __le32_to_cpu(family_num
), s
->disks
->devname
);
3777 static struct intel_super
*
3778 imsm_thunderdome(struct intel_super
**super_list
, int len
)
3780 struct intel_super
*super_table
[len
];
3781 struct intel_disk
*disk_list
= NULL
;
3782 struct intel_super
*champion
, *spare
;
3783 struct intel_super
*s
, **del
;
3788 memset(super_table
, 0, sizeof(super_table
));
3789 for (s
= *super_list
; s
; s
= s
->next
)
3790 tbl_size
= __prep_thunderdome(super_table
, tbl_size
, s
, &disk_list
);
3792 for (i
= 0; i
< tbl_size
; i
++) {
3793 struct imsm_disk
*d
;
3794 struct intel_disk
*idisk
;
3795 struct imsm_super
*mpb
= super_table
[i
]->anchor
;
3798 d
= &s
->disks
->disk
;
3800 /* 'd' must appear in merged disk list for its
3801 * configuration to be valid
3803 idisk
= disk_list_get(d
->serial
, disk_list
);
3804 if (idisk
&& idisk
->owner
== i
)
3805 s
= validate_members(s
, disk_list
, i
);
3810 dprintf("%s: marking family: %#x from %d:%d offline\n",
3811 __func__
, mpb
->family_num
,
3812 super_table
[i
]->disks
->major
,
3813 super_table
[i
]->disks
->minor
);
3817 /* This is where the mdadm implementation differs from the Windows
3818 * driver which has no strict concept of a container. We can only
3819 * assemble one family from a container, so when returning a prodigal
3820 * array member to this system the code will not be able to disambiguate
3821 * the container contents that should be assembled ("foreign" versus
3822 * "local"). It requires user intervention to set the orig_family_num
3823 * to a new value to establish a new container. The Windows driver in
3824 * this situation fixes up the volume name in place and manages the
3825 * foreign array as an independent entity.
3830 for (i
= 0; i
< tbl_size
; i
++) {
3831 struct intel_super
*tbl_ent
= super_table
[i
];
3837 if (tbl_ent
->anchor
->num_raid_devs
== 0) {
3842 if (s
&& !is_spare
) {
3843 show_conflicts(tbl_ent
->anchor
->family_num
, *super_list
);
3845 } else if (!s
&& !is_spare
)
3858 fprintf(stderr
, "Chose family %#x on '%s', "
3859 "assemble conflicts to new container with '--update=uuid'\n",
3860 __le32_to_cpu(s
->anchor
->family_num
), s
->disks
->devname
);
3862 /* collect all dl's onto 'champion', and update them to
3863 * champion's version of the status
3865 for (s
= *super_list
; s
; s
= s
->next
) {
3866 struct imsm_super
*mpb
= champion
->anchor
;
3867 struct dl
*dl
= s
->disks
;
3872 for (i
= 0; i
< mpb
->num_disks
; i
++) {
3873 struct imsm_disk
*disk
;
3875 disk
= __serial_to_disk(dl
->serial
, mpb
, &dl
->index
);
3878 /* only set index on disks that are a member of
3879 * a populated contianer, i.e. one with
3882 if (is_failed(&dl
->disk
))
3884 else if (is_spare(&dl
->disk
))
3890 if (i
>= mpb
->num_disks
) {
3891 struct intel_disk
*idisk
;
3893 idisk
= disk_list_get(dl
->serial
, disk_list
);
3894 if (idisk
&& is_spare(&idisk
->disk
) &&
3895 !is_failed(&idisk
->disk
) && !is_configured(&idisk
->disk
))
3903 dl
->next
= champion
->disks
;
3904 champion
->disks
= dl
;
3908 /* delete 'champion' from super_list */
3909 for (del
= super_list
; *del
; ) {
3910 if (*del
== champion
) {
3911 *del
= (*del
)->next
;
3914 del
= &(*del
)->next
;
3916 champion
->next
= NULL
;
3920 struct intel_disk
*idisk
= disk_list
;
3922 disk_list
= disk_list
->next
;
3929 static int load_super_imsm_all(struct supertype
*st
, int fd
, void **sbp
,
3933 struct intel_super
*super_list
= NULL
;
3934 struct intel_super
*super
= NULL
;
3935 int devnum
= fd2devnum(fd
);
3941 /* check if 'fd' an opened container */
3942 sra
= sysfs_read(fd
, 0, GET_LEVEL
|GET_VERSION
|GET_DEVS
|GET_STATE
);
3946 if (sra
->array
.major_version
!= -1 ||
3947 sra
->array
.minor_version
!= -2 ||
3948 strcmp(sra
->text_version
, "imsm") != 0) {
3953 for (sd
= sra
->devs
, i
= 0; sd
; sd
= sd
->next
, i
++) {
3954 struct intel_super
*s
= alloc_super();
3962 s
->next
= super_list
;
3966 sprintf(nm
, "%d:%d", sd
->disk
.major
, sd
->disk
.minor
);
3967 dfd
= dev_open(nm
, O_RDWR
);
3971 rv
= find_intel_hba_capability(dfd
, s
, devname
);
3972 /* no orom/efi or non-intel hba of the disk */
3976 err
= load_and_parse_mpb(dfd
, s
, NULL
, 1);
3978 /* retry the load if we might have raced against mdmon */
3979 if (err
== 3 && mdmon_running(devnum
))
3980 for (retry
= 0; retry
< 3; retry
++) {
3982 err
= load_and_parse_mpb(dfd
, s
, NULL
, 1);
3990 /* all mpbs enter, maybe one leaves */
3991 super
= imsm_thunderdome(&super_list
, i
);
3997 if (find_missing(super
) != 0) {
4003 /* load migration record */
4004 err
= load_imsm_migr_rec(super
, NULL
);
4010 /* Check migration compatibility */
4011 if (check_mpb_migr_compatibility(super
) != 0) {
4012 fprintf(stderr
, Name
": Unsupported migration detected");
4014 fprintf(stderr
, " on %s\n", devname
);
4016 fprintf(stderr
, " (IMSM).\n");
4025 while (super_list
) {
4026 struct intel_super
*s
= super_list
;
4028 super_list
= super_list
->next
;
4037 st
->container_dev
= devnum
;
4038 if (err
== 0 && st
->ss
== NULL
) {
4039 st
->ss
= &super_imsm
;
4040 st
->minor_version
= 0;
4041 st
->max_devs
= IMSM_MAX_DEVICES
;
4046 static int load_container_imsm(struct supertype
*st
, int fd
, char *devname
)
4048 return load_super_imsm_all(st
, fd
, &st
->sb
, devname
);
4052 static int load_super_imsm(struct supertype
*st
, int fd
, char *devname
)
4054 struct intel_super
*super
;
4057 if (test_partition(fd
))
4058 /* IMSM not allowed on partitions */
4061 free_super_imsm(st
);
4063 super
= alloc_super();
4066 Name
": malloc of %zu failed.\n",
4070 /* Load hba and capabilities if they exist.
4071 * But do not preclude loading metadata in case capabilities or hba are
4072 * non-compliant and ignore_hw_compat is set.
4074 rv
= find_intel_hba_capability(fd
, super
, devname
);
4075 /* no orom/efi or non-intel hba of the disk */
4076 if ((rv
!= 0) && (st
->ignore_hw_compat
== 0)) {
4079 Name
": No OROM/EFI properties for %s\n", devname
);
4083 rv
= load_and_parse_mpb(fd
, super
, devname
, 0);
4088 Name
": Failed to load all information "
4089 "sections on %s\n", devname
);
4095 if (st
->ss
== NULL
) {
4096 st
->ss
= &super_imsm
;
4097 st
->minor_version
= 0;
4098 st
->max_devs
= IMSM_MAX_DEVICES
;
4101 /* load migration record */
4102 if (load_imsm_migr_rec(super
, NULL
) == 0) {
4103 /* Check for unsupported migration features */
4104 if (check_mpb_migr_compatibility(super
) != 0) {
4106 Name
": Unsupported migration detected");
4108 fprintf(stderr
, " on %s\n", devname
);
4110 fprintf(stderr
, " (IMSM).\n");
4118 static __u16
info_to_blocks_per_strip(mdu_array_info_t
*info
)
4120 if (info
->level
== 1)
4122 return info
->chunk_size
>> 9;
4125 static __u32
info_to_num_data_stripes(mdu_array_info_t
*info
, int num_domains
)
4129 num_stripes
= (info
->size
* 2) / info_to_blocks_per_strip(info
);
4130 num_stripes
/= num_domains
;
4135 static __u32
info_to_blocks_per_member(mdu_array_info_t
*info
)
4137 if (info
->level
== 1)
4138 return info
->size
* 2;
4140 return (info
->size
* 2) & ~(info_to_blocks_per_strip(info
) - 1);
4143 static void imsm_update_version_info(struct intel_super
*super
)
4145 /* update the version and attributes */
4146 struct imsm_super
*mpb
= super
->anchor
;
4148 struct imsm_dev
*dev
;
4149 struct imsm_map
*map
;
4152 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
4153 dev
= get_imsm_dev(super
, i
);
4154 map
= get_imsm_map(dev
, MAP_0
);
4155 if (__le32_to_cpu(dev
->size_high
) > 0)
4156 mpb
->attributes
|= MPB_ATTRIB_2TB
;
4158 /* FIXME detect when an array spans a port multiplier */
4160 mpb
->attributes
|= MPB_ATTRIB_PM
;
4163 if (mpb
->num_raid_devs
> 1 ||
4164 mpb
->attributes
!= MPB_ATTRIB_CHECKSUM_VERIFY
) {
4165 version
= MPB_VERSION_ATTRIBS
;
4166 switch (get_imsm_raid_level(map
)) {
4167 case 0: mpb
->attributes
|= MPB_ATTRIB_RAID0
; break;
4168 case 1: mpb
->attributes
|= MPB_ATTRIB_RAID1
; break;
4169 case 10: mpb
->attributes
|= MPB_ATTRIB_RAID10
; break;
4170 case 5: mpb
->attributes
|= MPB_ATTRIB_RAID5
; break;
4173 if (map
->num_members
>= 5)
4174 version
= MPB_VERSION_5OR6_DISK_ARRAY
;
4175 else if (dev
->status
== DEV_CLONE_N_GO
)
4176 version
= MPB_VERSION_CNG
;
4177 else if (get_imsm_raid_level(map
) == 5)
4178 version
= MPB_VERSION_RAID5
;
4179 else if (map
->num_members
>= 3)
4180 version
= MPB_VERSION_3OR4_DISK_ARRAY
;
4181 else if (get_imsm_raid_level(map
) == 1)
4182 version
= MPB_VERSION_RAID1
;
4184 version
= MPB_VERSION_RAID0
;
4186 strcpy(((char *) mpb
->sig
) + strlen(MPB_SIGNATURE
), version
);
4190 static int check_name(struct intel_super
*super
, char *name
, int quiet
)
4192 struct imsm_super
*mpb
= super
->anchor
;
4193 char *reason
= NULL
;
4196 if (strlen(name
) > MAX_RAID_SERIAL_LEN
)
4197 reason
= "must be 16 characters or less";
4199 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
4200 struct imsm_dev
*dev
= get_imsm_dev(super
, i
);
4202 if (strncmp((char *) dev
->volume
, name
, MAX_RAID_SERIAL_LEN
) == 0) {
4203 reason
= "already exists";
4208 if (reason
&& !quiet
)
4209 fprintf(stderr
, Name
": imsm volume name %s\n", reason
);
4214 static int init_super_imsm_volume(struct supertype
*st
, mdu_array_info_t
*info
,
4215 unsigned long long size
, char *name
,
4216 char *homehost
, int *uuid
)
4218 /* We are creating a volume inside a pre-existing container.
4219 * so st->sb is already set.
4221 struct intel_super
*super
= st
->sb
;
4222 struct imsm_super
*mpb
= super
->anchor
;
4223 struct intel_dev
*dv
;
4224 struct imsm_dev
*dev
;
4225 struct imsm_vol
*vol
;
4226 struct imsm_map
*map
;
4227 int idx
= mpb
->num_raid_devs
;
4229 unsigned long long array_blocks
;
4230 size_t size_old
, size_new
;
4231 __u32 num_data_stripes
;
4233 if (super
->orom
&& mpb
->num_raid_devs
>= super
->orom
->vpa
) {
4234 fprintf(stderr
, Name
": This imsm-container already has the "
4235 "maximum of %d volumes\n", super
->orom
->vpa
);
4239 /* ensure the mpb is large enough for the new data */
4240 size_old
= __le32_to_cpu(mpb
->mpb_size
);
4241 size_new
= disks_to_mpb_size(info
->nr_disks
);
4242 if (size_new
> size_old
) {
4244 size_t size_round
= ROUND_UP(size_new
, 512);
4246 if (posix_memalign(&mpb_new
, 512, size_round
) != 0) {
4247 fprintf(stderr
, Name
": could not allocate new mpb\n");
4250 if (posix_memalign(&super
->migr_rec_buf
, 512, 512) != 0) {
4251 fprintf(stderr
, Name
4252 ": %s could not allocate migr_rec buffer\n",
4259 memcpy(mpb_new
, mpb
, size_old
);
4262 super
->anchor
= mpb_new
;
4263 mpb
->mpb_size
= __cpu_to_le32(size_new
);
4264 memset(mpb_new
+ size_old
, 0, size_round
- size_old
);
4266 super
->current_vol
= idx
;
4268 /* handle 'failed_disks' by either:
4269 * a) create dummy disk entries in the table if this the first
4270 * volume in the array. We add them here as this is the only
4271 * opportunity to add them. add_to_super_imsm_volume()
4272 * handles the non-failed disks and continues incrementing
4274 * b) validate that 'failed_disks' matches the current number
4275 * of missing disks if the container is populated
4277 if (super
->current_vol
== 0) {
4279 for (i
= 0; i
< info
->failed_disks
; i
++) {
4280 struct imsm_disk
*disk
;
4283 disk
= __get_imsm_disk(mpb
, i
);
4284 disk
->status
= CONFIGURED_DISK
| FAILED_DISK
;
4285 disk
->scsi_id
= __cpu_to_le32(~(__u32
)0);
4286 snprintf((char *) disk
->serial
, MAX_RAID_SERIAL_LEN
,
4289 find_missing(super
);
4294 for (d
= super
->missing
; d
; d
= d
->next
)
4296 if (info
->failed_disks
> missing
) {
4297 fprintf(stderr
, Name
": unable to add 'missing' disk to container\n");
4302 if (!check_name(super
, name
, 0))
4304 dv
= malloc(sizeof(*dv
));
4306 fprintf(stderr
, Name
": failed to allocate device list entry\n");
4309 dev
= calloc(1, sizeof(*dev
) + sizeof(__u32
) * (info
->raid_disks
- 1));
4312 fprintf(stderr
, Name
": could not allocate raid device\n");
4316 strncpy((char *) dev
->volume
, name
, MAX_RAID_SERIAL_LEN
);
4317 if (info
->level
== 1)
4318 array_blocks
= info_to_blocks_per_member(info
);
4320 array_blocks
= calc_array_size(info
->level
, info
->raid_disks
,
4321 info
->layout
, info
->chunk_size
,
4323 /* round array size down to closest MB */
4324 array_blocks
= (array_blocks
>> SECT_PER_MB_SHIFT
) << SECT_PER_MB_SHIFT
;
4326 dev
->size_low
= __cpu_to_le32((__u32
) array_blocks
);
4327 dev
->size_high
= __cpu_to_le32((__u32
) (array_blocks
>> 32));
4328 dev
->status
= (DEV_READ_COALESCING
| DEV_WRITE_COALESCING
);
4330 vol
->migr_state
= 0;
4331 set_migr_type(dev
, MIGR_INIT
);
4332 vol
->dirty
= !info
->state
;
4333 vol
->curr_migr_unit
= 0;
4334 map
= get_imsm_map(dev
, MAP_0
);
4335 map
->pba_of_lba0
= __cpu_to_le32(super
->create_offset
);
4336 map
->blocks_per_member
= __cpu_to_le32(info_to_blocks_per_member(info
));
4337 map
->blocks_per_strip
= __cpu_to_le16(info_to_blocks_per_strip(info
));
4338 map
->failed_disk_num
= ~0;
4339 if (info
->level
> 0)
4340 map
->map_state
= IMSM_T_STATE_UNINITIALIZED
;
4342 map
->map_state
= info
->failed_disks
? IMSM_T_STATE_FAILED
:
4343 IMSM_T_STATE_NORMAL
;
4346 if (info
->level
== 1 && info
->raid_disks
> 2) {
4349 fprintf(stderr
, Name
": imsm does not support more than 2 disks"
4350 "in a raid1 volume\n");
4354 map
->raid_level
= info
->level
;
4355 if (info
->level
== 10) {
4356 map
->raid_level
= 1;
4357 map
->num_domains
= info
->raid_disks
/ 2;
4358 } else if (info
->level
== 1)
4359 map
->num_domains
= info
->raid_disks
;
4361 map
->num_domains
= 1;
4363 num_data_stripes
= info_to_num_data_stripes(info
, map
->num_domains
);
4364 map
->num_data_stripes
= __cpu_to_le32(num_data_stripes
);
4366 map
->num_members
= info
->raid_disks
;
4367 for (i
= 0; i
< map
->num_members
; i
++) {
4368 /* initialized in add_to_super */
4369 set_imsm_ord_tbl_ent(map
, i
, IMSM_ORD_REBUILD
);
4371 mpb
->num_raid_devs
++;
4374 dv
->index
= super
->current_vol
;
4375 dv
->next
= super
->devlist
;
4376 super
->devlist
= dv
;
4378 imsm_update_version_info(super
);
4383 static int init_super_imsm(struct supertype
*st
, mdu_array_info_t
*info
,
4384 unsigned long long size
, char *name
,
4385 char *homehost
, int *uuid
)
4387 /* This is primarily called by Create when creating a new array.
4388 * We will then get add_to_super called for each component, and then
4389 * write_init_super called to write it out to each device.
4390 * For IMSM, Create can create on fresh devices or on a pre-existing
4392 * To create on a pre-existing array a different method will be called.
4393 * This one is just for fresh drives.
4395 struct intel_super
*super
;
4396 struct imsm_super
*mpb
;
4401 return init_super_imsm_volume(st
, info
, size
, name
, homehost
, uuid
);
4404 mpb_size
= disks_to_mpb_size(info
->nr_disks
);
4408 super
= alloc_super();
4409 if (super
&& posix_memalign(&super
->buf
, 512, mpb_size
) != 0) {
4414 fprintf(stderr
, Name
4415 ": %s could not allocate superblock\n", __func__
);
4418 if (posix_memalign(&super
->migr_rec_buf
, 512, 512) != 0) {
4419 fprintf(stderr
, Name
4420 ": %s could not allocate migr_rec buffer\n", __func__
);
4425 memset(super
->buf
, 0, mpb_size
);
4427 mpb
->mpb_size
= __cpu_to_le32(mpb_size
);
4431 /* zeroing superblock */
4435 mpb
->attributes
= MPB_ATTRIB_CHECKSUM_VERIFY
;
4437 version
= (char *) mpb
->sig
;
4438 strcpy(version
, MPB_SIGNATURE
);
4439 version
+= strlen(MPB_SIGNATURE
);
4440 strcpy(version
, MPB_VERSION_RAID0
);
4446 static int add_to_super_imsm_volume(struct supertype
*st
, mdu_disk_info_t
*dk
,
4447 int fd
, char *devname
)
4449 struct intel_super
*super
= st
->sb
;
4450 struct imsm_super
*mpb
= super
->anchor
;
4451 struct imsm_disk
*_disk
;
4452 struct imsm_dev
*dev
;
4453 struct imsm_map
*map
;
4457 dev
= get_imsm_dev(super
, super
->current_vol
);
4458 map
= get_imsm_map(dev
, MAP_0
);
4460 if (! (dk
->state
& (1<<MD_DISK_SYNC
))) {
4461 fprintf(stderr
, Name
": %s: Cannot add spare devices to IMSM volume\n",
4467 /* we're doing autolayout so grab the pre-marked (in
4468 * validate_geometry) raid_disk
4470 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
4471 if (dl
->raiddisk
== dk
->raid_disk
)
4474 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
4475 if (dl
->major
== dk
->major
&&
4476 dl
->minor
== dk
->minor
)
4481 fprintf(stderr
, Name
": %s is not a member of the same container\n", devname
);
4485 /* add a pristine spare to the metadata */
4486 if (dl
->index
< 0) {
4487 dl
->index
= super
->anchor
->num_disks
;
4488 super
->anchor
->num_disks
++;
4490 /* Check the device has not already been added */
4491 slot
= get_imsm_disk_slot(map
, dl
->index
);
4493 (get_imsm_ord_tbl_ent(dev
, slot
, MAP_X
) & IMSM_ORD_REBUILD
) == 0) {
4494 fprintf(stderr
, Name
": %s has been included in this array twice\n",
4498 set_imsm_ord_tbl_ent(map
, dk
->raid_disk
, dl
->index
);
4499 dl
->disk
.status
= CONFIGURED_DISK
;
4501 /* update size of 'missing' disks to be at least as large as the
4502 * largest acitve member (we only have dummy missing disks when
4503 * creating the first volume)
4505 if (super
->current_vol
== 0) {
4506 for (df
= super
->missing
; df
; df
= df
->next
) {
4507 if (dl
->disk
.total_blocks
> df
->disk
.total_blocks
)
4508 df
->disk
.total_blocks
= dl
->disk
.total_blocks
;
4509 _disk
= __get_imsm_disk(mpb
, df
->index
);
4514 /* refresh unset/failed slots to point to valid 'missing' entries */
4515 for (df
= super
->missing
; df
; df
= df
->next
)
4516 for (slot
= 0; slot
< mpb
->num_disks
; slot
++) {
4517 __u32 ord
= get_imsm_ord_tbl_ent(dev
, slot
, MAP_X
);
4519 if ((ord
& IMSM_ORD_REBUILD
) == 0)
4521 set_imsm_ord_tbl_ent(map
, slot
, df
->index
| IMSM_ORD_REBUILD
);
4522 if (is_gen_migration(dev
)) {
4523 struct imsm_map
*map2
= get_imsm_map(dev
,
4525 int slot2
= get_imsm_disk_slot(map2
, df
->index
);
4526 if ((slot2
< map2
->num_members
) &&
4528 __u32 ord2
= get_imsm_ord_tbl_ent(dev
,
4531 if ((unsigned)df
->index
==
4533 set_imsm_ord_tbl_ent(map2
,
4539 dprintf("set slot:%d to missing disk:%d\n", slot
, df
->index
);
4543 /* if we are creating the first raid device update the family number */
4544 if (super
->current_vol
== 0) {
4546 struct imsm_dev
*_dev
= __get_imsm_dev(mpb
, 0);
4548 _disk
= __get_imsm_disk(mpb
, dl
->index
);
4549 if (!_dev
|| !_disk
) {
4550 fprintf(stderr
, Name
": BUG mpb setup error\n");
4556 sum
+= __gen_imsm_checksum(mpb
);
4557 mpb
->family_num
= __cpu_to_le32(sum
);
4558 mpb
->orig_family_num
= mpb
->family_num
;
4560 super
->current_disk
= dl
;
4565 * Function marks disk as spare and restores disk serial
4566 * in case it was previously marked as failed by takeover operation
4568 * -1 : critical error
4569 * 0 : disk is marked as spare but serial is not set
4572 int mark_spare(struct dl
*disk
)
4574 __u8 serial
[MAX_RAID_SERIAL_LEN
];
4581 if (!imsm_read_serial(disk
->fd
, NULL
, serial
)) {
4582 /* Restore disk serial number, because takeover marks disk
4583 * as failed and adds to serial ':0' before it becomes
4586 serialcpy(disk
->serial
, serial
);
4587 serialcpy(disk
->disk
.serial
, serial
);
4590 disk
->disk
.status
= SPARE_DISK
;
4596 static int add_to_super_imsm(struct supertype
*st
, mdu_disk_info_t
*dk
,
4597 int fd
, char *devname
)
4599 struct intel_super
*super
= st
->sb
;
4601 unsigned long long size
;
4606 /* If we are on an RAID enabled platform check that the disk is
4607 * attached to the raid controller.
4608 * We do not need to test disks attachment for container based additions,
4609 * they shall be already tested when container was created/assembled.
4611 rv
= find_intel_hba_capability(fd
, super
, devname
);
4612 /* no orom/efi or non-intel hba of the disk */
4614 dprintf("capability: %p fd: %d ret: %d\n",
4615 super
->orom
, fd
, rv
);
4619 if (super
->current_vol
>= 0)
4620 return add_to_super_imsm_volume(st
, dk
, fd
, devname
);
4623 dd
= malloc(sizeof(*dd
));
4626 Name
": malloc failed %s:%d.\n", __func__
, __LINE__
);
4629 memset(dd
, 0, sizeof(*dd
));
4630 dd
->major
= major(stb
.st_rdev
);
4631 dd
->minor
= minor(stb
.st_rdev
);
4632 dd
->devname
= devname
? strdup(devname
) : NULL
;
4635 dd
->action
= DISK_ADD
;
4636 rv
= imsm_read_serial(fd
, devname
, dd
->serial
);
4639 Name
": failed to retrieve scsi serial, aborting\n");
4644 get_dev_size(fd
, NULL
, &size
);
4646 serialcpy(dd
->disk
.serial
, dd
->serial
);
4647 dd
->disk
.total_blocks
= __cpu_to_le32(size
);
4649 if (sysfs_disk_to_scsi_id(fd
, &id
) == 0)
4650 dd
->disk
.scsi_id
= __cpu_to_le32(id
);
4652 dd
->disk
.scsi_id
= __cpu_to_le32(0);
4654 if (st
->update_tail
) {
4655 dd
->next
= super
->disk_mgmt_list
;
4656 super
->disk_mgmt_list
= dd
;
4658 dd
->next
= super
->disks
;
4660 super
->updates_pending
++;
4667 static int remove_from_super_imsm(struct supertype
*st
, mdu_disk_info_t
*dk
)
4669 struct intel_super
*super
= st
->sb
;
4672 /* remove from super works only in mdmon - for communication
4673 * manager - monitor. Check if communication memory buffer
4676 if (!st
->update_tail
) {
4678 Name
": %s shall be used in mdmon context only"
4679 "(line %d).\n", __func__
, __LINE__
);
4682 dd
= malloc(sizeof(*dd
));
4685 Name
": malloc failed %s:%d.\n", __func__
, __LINE__
);
4688 memset(dd
, 0, sizeof(*dd
));
4689 dd
->major
= dk
->major
;
4690 dd
->minor
= dk
->minor
;
4693 dd
->action
= DISK_REMOVE
;
4695 dd
->next
= super
->disk_mgmt_list
;
4696 super
->disk_mgmt_list
= dd
;
4702 static int store_imsm_mpb(int fd
, struct imsm_super
*mpb
);
4706 struct imsm_super anchor
;
4707 } spare_record
__attribute__ ((aligned(512)));
4709 /* spare records have their own family number and do not have any defined raid
4712 static int write_super_imsm_spares(struct intel_super
*super
, int doclose
)
4714 struct imsm_super
*mpb
= super
->anchor
;
4715 struct imsm_super
*spare
= &spare_record
.anchor
;
4719 spare
->mpb_size
= __cpu_to_le32(sizeof(struct imsm_super
)),
4720 spare
->generation_num
= __cpu_to_le32(1UL),
4721 spare
->attributes
= MPB_ATTRIB_CHECKSUM_VERIFY
;
4722 spare
->num_disks
= 1,
4723 spare
->num_raid_devs
= 0,
4724 spare
->cache_size
= mpb
->cache_size
,
4725 spare
->pwr_cycle_count
= __cpu_to_le32(1),
4727 snprintf((char *) spare
->sig
, MAX_SIGNATURE_LENGTH
,
4728 MPB_SIGNATURE MPB_VERSION_RAID0
);
4730 for (d
= super
->disks
; d
; d
= d
->next
) {
4734 spare
->disk
[0] = d
->disk
;
4735 sum
= __gen_imsm_checksum(spare
);
4736 spare
->family_num
= __cpu_to_le32(sum
);
4737 spare
->orig_family_num
= 0;
4738 sum
= __gen_imsm_checksum(spare
);
4739 spare
->check_sum
= __cpu_to_le32(sum
);
4741 if (store_imsm_mpb(d
->fd
, spare
)) {
4742 fprintf(stderr
, "%s: failed for device %d:%d %s\n",
4743 __func__
, d
->major
, d
->minor
, strerror(errno
));
4755 static int write_super_imsm(struct supertype
*st
, int doclose
)
4757 struct intel_super
*super
= st
->sb
;
4758 struct imsm_super
*mpb
= super
->anchor
;
4764 __u32 mpb_size
= sizeof(struct imsm_super
) - sizeof(struct imsm_disk
);
4766 int clear_migration_record
= 1;
4768 /* 'generation' is incremented everytime the metadata is written */
4769 generation
= __le32_to_cpu(mpb
->generation_num
);
4771 mpb
->generation_num
= __cpu_to_le32(generation
);
4773 /* fix up cases where previous mdadm releases failed to set
4776 if (mpb
->orig_family_num
== 0)
4777 mpb
->orig_family_num
= mpb
->family_num
;
4779 for (d
= super
->disks
; d
; d
= d
->next
) {
4783 mpb
->disk
[d
->index
] = d
->disk
;
4787 for (d
= super
->missing
; d
; d
= d
->next
) {
4788 mpb
->disk
[d
->index
] = d
->disk
;
4791 mpb
->num_disks
= num_disks
;
4792 mpb_size
+= sizeof(struct imsm_disk
) * mpb
->num_disks
;
4794 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
4795 struct imsm_dev
*dev
= __get_imsm_dev(mpb
, i
);
4796 struct imsm_dev
*dev2
= get_imsm_dev(super
, i
);
4798 imsm_copy_dev(dev
, dev2
);
4799 mpb_size
+= sizeof_imsm_dev(dev
, 0);
4801 if (is_gen_migration(dev2
))
4802 clear_migration_record
= 0;
4804 mpb_size
+= __le32_to_cpu(mpb
->bbm_log_size
);
4805 mpb
->mpb_size
= __cpu_to_le32(mpb_size
);
4807 /* recalculate checksum */
4808 sum
= __gen_imsm_checksum(mpb
);
4809 mpb
->check_sum
= __cpu_to_le32(sum
);
4811 if (clear_migration_record
)
4812 memset(super
->migr_rec_buf
, 0, 512);
4814 /* write the mpb for disks that compose raid devices */
4815 for (d
= super
->disks
; d
; d
= d
->next
) {
4816 if (d
->index
< 0 || is_failed(&d
->disk
))
4818 if (store_imsm_mpb(d
->fd
, mpb
))
4819 fprintf(stderr
, "%s: failed for device %d:%d %s\n",
4820 __func__
, d
->major
, d
->minor
, strerror(errno
));
4821 if (clear_migration_record
) {
4822 unsigned long long dsize
;
4824 get_dev_size(d
->fd
, NULL
, &dsize
);
4825 if (lseek64(d
->fd
, dsize
- 512, SEEK_SET
) >= 0) {
4826 if (write(d
->fd
, super
->migr_rec_buf
, 512) != 512)
4827 perror("Write migr_rec failed");
4837 return write_super_imsm_spares(super
, doclose
);
4843 static int create_array(struct supertype
*st
, int dev_idx
)
4846 struct imsm_update_create_array
*u
;
4847 struct intel_super
*super
= st
->sb
;
4848 struct imsm_dev
*dev
= get_imsm_dev(super
, dev_idx
);
4849 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
4850 struct disk_info
*inf
;
4851 struct imsm_disk
*disk
;
4854 len
= sizeof(*u
) - sizeof(*dev
) + sizeof_imsm_dev(dev
, 0) +
4855 sizeof(*inf
) * map
->num_members
;
4858 fprintf(stderr
, "%s: failed to allocate update buffer\n",
4863 u
->type
= update_create_array
;
4864 u
->dev_idx
= dev_idx
;
4865 imsm_copy_dev(&u
->dev
, dev
);
4866 inf
= get_disk_info(u
);
4867 for (i
= 0; i
< map
->num_members
; i
++) {
4868 int idx
= get_imsm_disk_idx(dev
, i
, MAP_X
);
4870 disk
= get_imsm_disk(super
, idx
);
4871 serialcpy(inf
[i
].serial
, disk
->serial
);
4873 append_metadata_update(st
, u
, len
);
4878 static int mgmt_disk(struct supertype
*st
)
4880 struct intel_super
*super
= st
->sb
;
4882 struct imsm_update_add_remove_disk
*u
;
4884 if (!super
->disk_mgmt_list
)
4890 fprintf(stderr
, "%s: failed to allocate update buffer\n",
4895 u
->type
= update_add_remove_disk
;
4896 append_metadata_update(st
, u
, len
);
4901 static int write_init_super_imsm(struct supertype
*st
)
4903 struct intel_super
*super
= st
->sb
;
4904 int current_vol
= super
->current_vol
;
4906 /* we are done with current_vol reset it to point st at the container */
4907 super
->current_vol
= -1;
4909 if (st
->update_tail
) {
4910 /* queue the recently created array / added disk
4911 * as a metadata update */
4914 /* determine if we are creating a volume or adding a disk */
4915 if (current_vol
< 0) {
4916 /* in the mgmt (add/remove) disk case we are running
4917 * in mdmon context, so don't close fd's
4919 return mgmt_disk(st
);
4921 rv
= create_array(st
, current_vol
);
4926 for (d
= super
->disks
; d
; d
= d
->next
)
4927 Kill(d
->devname
, NULL
, 0, 1, 1);
4928 return write_super_imsm(st
, 1);
4933 static int store_super_imsm(struct supertype
*st
, int fd
)
4935 struct intel_super
*super
= st
->sb
;
4936 struct imsm_super
*mpb
= super
? super
->anchor
: NULL
;
4942 return store_imsm_mpb(fd
, mpb
);
4948 static int imsm_bbm_log_size(struct imsm_super
*mpb
)
4950 return __le32_to_cpu(mpb
->bbm_log_size
);
4954 static int validate_geometry_imsm_container(struct supertype
*st
, int level
,
4955 int layout
, int raiddisks
, int chunk
,
4956 unsigned long long size
, char *dev
,
4957 unsigned long long *freesize
,
4961 unsigned long long ldsize
;
4962 struct intel_super
*super
=NULL
;
4965 if (level
!= LEVEL_CONTAINER
)
4970 fd
= open(dev
, O_RDONLY
|O_EXCL
, 0);
4973 fprintf(stderr
, Name
": imsm: Cannot open %s: %s\n",
4974 dev
, strerror(errno
));
4977 if (!get_dev_size(fd
, dev
, &ldsize
)) {
4982 /* capabilities retrieve could be possible
4983 * note that there is no fd for the disks in array.
4985 super
= alloc_super();
4988 Name
": malloc of %zu failed.\n",
4994 rv
= find_intel_hba_capability(fd
, super
, verbose
? dev
: NULL
);
4998 fd2devname(fd
, str
);
4999 dprintf("validate_geometry_imsm_container: fd: %d %s orom: %p rv: %d raiddisk: %d\n",
5000 fd
, str
, super
->orom
, rv
, raiddisks
);
5002 /* no orom/efi or non-intel hba of the disk */
5008 if (super
->orom
&& raiddisks
> super
->orom
->tds
) {
5010 fprintf(stderr
, Name
": %d exceeds maximum number of"
5011 " platform supported disks: %d\n",
5012 raiddisks
, super
->orom
->tds
);
5018 *freesize
= avail_size_imsm(st
, ldsize
>> 9);
5024 static unsigned long long find_size(struct extent
*e
, int *idx
, int num_extents
)
5026 const unsigned long long base_start
= e
[*idx
].start
;
5027 unsigned long long end
= base_start
+ e
[*idx
].size
;
5030 if (base_start
== end
)
5034 for (i
= *idx
; i
< num_extents
; i
++) {
5035 /* extend overlapping extents */
5036 if (e
[i
].start
>= base_start
&&
5037 e
[i
].start
<= end
) {
5040 if (e
[i
].start
+ e
[i
].size
> end
)
5041 end
= e
[i
].start
+ e
[i
].size
;
5042 } else if (e
[i
].start
> end
) {
5048 return end
- base_start
;
5051 static unsigned long long merge_extents(struct intel_super
*super
, int sum_extents
)
5053 /* build a composite disk with all known extents and generate a new
5054 * 'maxsize' given the "all disks in an array must share a common start
5055 * offset" constraint
5057 struct extent
*e
= calloc(sum_extents
, sizeof(*e
));
5061 unsigned long long pos
;
5062 unsigned long long start
= 0;
5063 unsigned long long maxsize
;
5064 unsigned long reserve
;
5069 /* coalesce and sort all extents. also, check to see if we need to
5070 * reserve space between member arrays
5073 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
5076 for (i
= 0; i
< dl
->extent_cnt
; i
++)
5079 qsort(e
, sum_extents
, sizeof(*e
), cmp_extent
);
5084 while (i
< sum_extents
) {
5085 e
[j
].start
= e
[i
].start
;
5086 e
[j
].size
= find_size(e
, &i
, sum_extents
);
5088 if (e
[j
-1].size
== 0)
5097 unsigned long long esize
;
5099 esize
= e
[i
].start
- pos
;
5100 if (esize
>= maxsize
) {
5105 pos
= e
[i
].start
+ e
[i
].size
;
5107 } while (e
[i
-1].size
);
5113 /* FIXME assumes volume at offset 0 is the first volume in a
5116 if (start_extent
> 0)
5117 reserve
= IMSM_RESERVED_SECTORS
; /* gap between raid regions */
5121 if (maxsize
< reserve
)
5124 super
->create_offset
= ~((__u32
) 0);
5125 if (start
+ reserve
> super
->create_offset
)
5126 return 0; /* start overflows create_offset */
5127 super
->create_offset
= start
+ reserve
;
5129 return maxsize
- reserve
;
5132 static int is_raid_level_supported(const struct imsm_orom
*orom
, int level
, int raiddisks
)
5134 if (level
< 0 || level
== 6 || level
== 4)
5137 /* if we have an orom prevent invalid raid levels */
5140 case 0: return imsm_orom_has_raid0(orom
);
5143 return imsm_orom_has_raid1e(orom
);
5144 return imsm_orom_has_raid1(orom
) && raiddisks
== 2;
5145 case 10: return imsm_orom_has_raid10(orom
) && raiddisks
== 4;
5146 case 5: return imsm_orom_has_raid5(orom
) && raiddisks
> 2;
5149 return 1; /* not on an Intel RAID platform so anything goes */
5154 static int imsm_default_chunk(const struct imsm_orom
*orom
)
5156 /* up to 512 if the plaform supports it, otherwise the platform max.
5157 * 128 if no platform detected
5159 int fs
= max(7, orom
? fls(orom
->sss
) : 0);
5161 return min(512, (1 << fs
));
5164 #define pr_vrb(fmt, arg...) (void) (verbose && fprintf(stderr, Name fmt, ##arg))
5166 validate_geometry_imsm_orom(struct intel_super
*super
, int level
, int layout
,
5167 int raiddisks
, int *chunk
, int verbose
)
5169 /* check/set platform and metadata limits/defaults */
5170 if (super
->orom
&& raiddisks
> super
->orom
->dpa
) {
5171 pr_vrb(": platform supports a maximum of %d disks per array\n",
5176 /* capabilities of OROM tested - copied from validate_geometry_imsm_volume */
5177 if (!is_raid_level_supported(super
->orom
, level
, raiddisks
)) {
5178 pr_vrb(": platform does not support raid%d with %d disk%s\n",
5179 level
, raiddisks
, raiddisks
> 1 ? "s" : "");
5183 if (chunk
&& (*chunk
== 0 || *chunk
== UnSet
))
5184 *chunk
= imsm_default_chunk(super
->orom
);
5186 if (super
->orom
&& chunk
&& !imsm_orom_has_chunk(super
->orom
, *chunk
)) {
5187 pr_vrb(": platform does not support a chunk size of: "
5192 if (layout
!= imsm_level_to_layout(level
)) {
5194 pr_vrb(": imsm raid 5 only supports the left-asymmetric layout\n");
5195 else if (level
== 10)
5196 pr_vrb(": imsm raid 10 only supports the n2 layout\n");
5198 pr_vrb(": imsm unknown layout %#x for this raid level %d\n",
5205 /* validate_geometry_imsm_volume - lifted from validate_geometry_ddf_bvd
5206 * FIX ME add ahci details
5208 static int validate_geometry_imsm_volume(struct supertype
*st
, int level
,
5209 int layout
, int raiddisks
, int *chunk
,
5210 unsigned long long size
, char *dev
,
5211 unsigned long long *freesize
,
5215 struct intel_super
*super
= st
->sb
;
5216 struct imsm_super
*mpb
;
5218 unsigned long long pos
= 0;
5219 unsigned long long maxsize
;
5223 /* We must have the container info already read in. */
5227 mpb
= super
->anchor
;
5229 if (mpb
->num_raid_devs
> 0 && mpb
->num_disks
!= raiddisks
) {
5230 fprintf(stderr
, Name
": the option-rom requires all "
5231 "member disks to be a member of all volumes.\n");
5235 if (!validate_geometry_imsm_orom(super
, level
, layout
, raiddisks
, chunk
, verbose
)) {
5236 fprintf(stderr
, Name
": RAID gemetry validation failed. "
5237 "Cannot proceed with the action(s).\n");
5241 /* General test: make sure there is space for
5242 * 'raiddisks' device extents of size 'size' at a given
5245 unsigned long long minsize
= size
;
5246 unsigned long long start_offset
= MaxSector
;
5249 minsize
= MPB_SECTOR_CNT
+ IMSM_RESERVED_SECTORS
;
5250 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
5255 e
= get_extents(super
, dl
);
5258 unsigned long long esize
;
5259 esize
= e
[i
].start
- pos
;
5260 if (esize
>= minsize
)
5262 if (found
&& start_offset
== MaxSector
) {
5265 } else if (found
&& pos
!= start_offset
) {
5269 pos
= e
[i
].start
+ e
[i
].size
;
5271 } while (e
[i
-1].size
);
5276 if (dcnt
< raiddisks
) {
5278 fprintf(stderr
, Name
": imsm: Not enough "
5279 "devices with space for this array "
5287 /* This device must be a member of the set */
5288 if (stat(dev
, &stb
) < 0)
5290 if ((S_IFMT
& stb
.st_mode
) != S_IFBLK
)
5292 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
5293 if (dl
->major
== (int)major(stb
.st_rdev
) &&
5294 dl
->minor
== (int)minor(stb
.st_rdev
))
5299 fprintf(stderr
, Name
": %s is not in the "
5300 "same imsm set\n", dev
);
5302 } else if (super
->orom
&& dl
->index
< 0 && mpb
->num_raid_devs
) {
5303 /* If a volume is present then the current creation attempt
5304 * cannot incorporate new spares because the orom may not
5305 * understand this configuration (all member disks must be
5306 * members of each array in the container).
5308 fprintf(stderr
, Name
": %s is a spare and a volume"
5309 " is already defined for this container\n", dev
);
5310 fprintf(stderr
, Name
": The option-rom requires all member"
5311 " disks to be a member of all volumes\n");
5315 /* retrieve the largest free space block */
5316 e
= get_extents(super
, dl
);
5321 unsigned long long esize
;
5323 esize
= e
[i
].start
- pos
;
5324 if (esize
>= maxsize
)
5326 pos
= e
[i
].start
+ e
[i
].size
;
5328 } while (e
[i
-1].size
);
5333 fprintf(stderr
, Name
": unable to determine free space for: %s\n",
5337 if (maxsize
< size
) {
5339 fprintf(stderr
, Name
": %s not enough space (%llu < %llu)\n",
5340 dev
, maxsize
, size
);
5344 /* count total number of extents for merge */
5346 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
5348 i
+= dl
->extent_cnt
;
5350 maxsize
= merge_extents(super
, i
);
5352 if (!check_env("IMSM_NO_PLATFORM") &&
5353 mpb
->num_raid_devs
> 0 && size
&& size
!= maxsize
) {
5354 fprintf(stderr
, Name
": attempting to create a second "
5355 "volume with size less then remaining space. "
5360 if (maxsize
< size
|| maxsize
== 0) {
5362 fprintf(stderr
, Name
": not enough space after merge (%llu < %llu)\n",
5367 *freesize
= maxsize
;
5372 static int reserve_space(struct supertype
*st
, int raiddisks
,
5373 unsigned long long size
, int chunk
,
5374 unsigned long long *freesize
)
5376 struct intel_super
*super
= st
->sb
;
5377 struct imsm_super
*mpb
= super
->anchor
;
5382 unsigned long long maxsize
;
5383 unsigned long long minsize
;
5387 /* find the largest common start free region of the possible disks */
5391 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
5397 /* don't activate new spares if we are orom constrained
5398 * and there is already a volume active in the container
5400 if (super
->orom
&& dl
->index
< 0 && mpb
->num_raid_devs
)
5403 e
= get_extents(super
, dl
);
5406 for (i
= 1; e
[i
-1].size
; i
++)
5414 maxsize
= merge_extents(super
, extent_cnt
);
5418 minsize
= chunk
* 2;
5420 if (cnt
< raiddisks
||
5421 (super
->orom
&& used
&& used
!= raiddisks
) ||
5422 maxsize
< minsize
||
5424 fprintf(stderr
, Name
": not enough devices with space to create array.\n");
5425 return 0; /* No enough free spaces large enough */
5437 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
5439 dl
->raiddisk
= cnt
++;
5446 static int validate_geometry_imsm(struct supertype
*st
, int level
, int layout
,
5447 int raiddisks
, int *chunk
, unsigned long long size
,
5448 char *dev
, unsigned long long *freesize
,
5456 * if given unused devices create a container
5457 * if given given devices in a container create a member volume
5459 if (level
== LEVEL_CONTAINER
) {
5460 /* Must be a fresh device to add to a container */
5461 return validate_geometry_imsm_container(st
, level
, layout
,
5463 chunk
?*chunk
:0, size
,
5470 if (!validate_geometry_imsm_orom(st
->sb
, level
, layout
,
5474 /* we are being asked to automatically layout a
5475 * new volume based on the current contents of
5476 * the container. If the the parameters can be
5477 * satisfied reserve_space will record the disks,
5478 * start offset, and size of the volume to be
5479 * created. add_to_super and getinfo_super
5480 * detect when autolayout is in progress.
5483 return reserve_space(st
, raiddisks
, size
,
5484 chunk
?*chunk
:0, freesize
);
5489 /* creating in a given container */
5490 return validate_geometry_imsm_volume(st
, level
, layout
,
5491 raiddisks
, chunk
, size
,
5492 dev
, freesize
, verbose
);
5495 /* This device needs to be a device in an 'imsm' container */
5496 fd
= open(dev
, O_RDONLY
|O_EXCL
, 0);
5500 Name
": Cannot create this array on device %s\n",
5505 if (errno
!= EBUSY
|| (fd
= open(dev
, O_RDONLY
, 0)) < 0) {
5507 fprintf(stderr
, Name
": Cannot open %s: %s\n",
5508 dev
, strerror(errno
));
5511 /* Well, it is in use by someone, maybe an 'imsm' container. */
5512 cfd
= open_container(fd
);
5516 fprintf(stderr
, Name
": Cannot use %s: It is busy\n",
5520 sra
= sysfs_read(cfd
, 0, GET_VERSION
);
5521 if (sra
&& sra
->array
.major_version
== -1 &&
5522 strcmp(sra
->text_version
, "imsm") == 0)
5526 /* This is a member of a imsm container. Load the container
5527 * and try to create a volume
5529 struct intel_super
*super
;
5531 if (load_super_imsm_all(st
, cfd
, (void **) &super
, NULL
) == 0) {
5533 st
->container_dev
= fd2devnum(cfd
);
5535 return validate_geometry_imsm_volume(st
, level
, layout
,
5544 fprintf(stderr
, Name
": failed container membership check\n");
5550 static void default_geometry_imsm(struct supertype
*st
, int *level
, int *layout
, int *chunk
)
5552 struct intel_super
*super
= st
->sb
;
5554 if (level
&& *level
== UnSet
)
5555 *level
= LEVEL_CONTAINER
;
5557 if (level
&& layout
&& *layout
== UnSet
)
5558 *layout
= imsm_level_to_layout(*level
);
5560 if (chunk
&& (*chunk
== UnSet
|| *chunk
== 0))
5561 *chunk
= imsm_default_chunk(super
->orom
);
5564 static void handle_missing(struct intel_super
*super
, struct imsm_dev
*dev
);
5566 static int kill_subarray_imsm(struct supertype
*st
)
5568 /* remove the subarray currently referenced by ->current_vol */
5570 struct intel_dev
**dp
;
5571 struct intel_super
*super
= st
->sb
;
5572 __u8 current_vol
= super
->current_vol
;
5573 struct imsm_super
*mpb
= super
->anchor
;
5575 if (super
->current_vol
< 0)
5577 super
->current_vol
= -1; /* invalidate subarray cursor */
5579 /* block deletions that would change the uuid of active subarrays
5581 * FIXME when immutable ids are available, but note that we'll
5582 * also need to fixup the invalidated/active subarray indexes in
5585 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
5588 if (i
< current_vol
)
5590 sprintf(subarray
, "%u", i
);
5591 if (is_subarray_active(subarray
, st
->devname
)) {
5593 Name
": deleting subarray-%d would change the UUID of active subarray-%d, aborting\n",
5600 if (st
->update_tail
) {
5601 struct imsm_update_kill_array
*u
= malloc(sizeof(*u
));
5605 u
->type
= update_kill_array
;
5606 u
->dev_idx
= current_vol
;
5607 append_metadata_update(st
, u
, sizeof(*u
));
5612 for (dp
= &super
->devlist
; *dp
;)
5613 if ((*dp
)->index
== current_vol
) {
5616 handle_missing(super
, (*dp
)->dev
);
5617 if ((*dp
)->index
> current_vol
)
5622 /* no more raid devices, all active components are now spares,
5623 * but of course failed are still failed
5625 if (--mpb
->num_raid_devs
== 0) {
5628 for (d
= super
->disks
; d
; d
= d
->next
)
5633 super
->updates_pending
++;
5638 static int update_subarray_imsm(struct supertype
*st
, char *subarray
,
5639 char *update
, struct mddev_ident
*ident
)
5641 /* update the subarray currently referenced by ->current_vol */
5642 struct intel_super
*super
= st
->sb
;
5643 struct imsm_super
*mpb
= super
->anchor
;
5645 if (strcmp(update
, "name") == 0) {
5646 char *name
= ident
->name
;
5650 if (is_subarray_active(subarray
, st
->devname
)) {
5652 Name
": Unable to update name of active subarray\n");
5656 if (!check_name(super
, name
, 0))
5659 vol
= strtoul(subarray
, &ep
, 10);
5660 if (*ep
!= '\0' || vol
>= super
->anchor
->num_raid_devs
)
5663 if (st
->update_tail
) {
5664 struct imsm_update_rename_array
*u
= malloc(sizeof(*u
));
5668 u
->type
= update_rename_array
;
5670 snprintf((char *) u
->name
, MAX_RAID_SERIAL_LEN
, "%s", name
);
5671 append_metadata_update(st
, u
, sizeof(*u
));
5673 struct imsm_dev
*dev
;
5676 dev
= get_imsm_dev(super
, vol
);
5677 snprintf((char *) dev
->volume
, MAX_RAID_SERIAL_LEN
, "%s", name
);
5678 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
5679 dev
= get_imsm_dev(super
, i
);
5680 handle_missing(super
, dev
);
5682 super
->updates_pending
++;
5689 #endif /* MDASSEMBLE */
5691 static int is_gen_migration(struct imsm_dev
*dev
)
5696 if (!dev
->vol
.migr_state
)
5699 if (migr_type(dev
) == MIGR_GEN_MIGR
)
5705 static int is_rebuilding(struct imsm_dev
*dev
)
5707 struct imsm_map
*migr_map
;
5709 if (!dev
->vol
.migr_state
)
5712 if (migr_type(dev
) != MIGR_REBUILD
)
5715 migr_map
= get_imsm_map(dev
, MAP_1
);
5717 if (migr_map
->map_state
== IMSM_T_STATE_DEGRADED
)
5723 static int is_initializing(struct imsm_dev
*dev
)
5725 struct imsm_map
*migr_map
;
5727 if (!dev
->vol
.migr_state
)
5730 if (migr_type(dev
) != MIGR_INIT
)
5733 migr_map
= get_imsm_map(dev
, MAP_1
);
5735 if (migr_map
->map_state
== IMSM_T_STATE_UNINITIALIZED
)
5742 static void update_recovery_start(struct intel_super
*super
,
5743 struct imsm_dev
*dev
,
5744 struct mdinfo
*array
)
5746 struct mdinfo
*rebuild
= NULL
;
5750 if (!is_rebuilding(dev
))
5753 /* Find the rebuild target, but punt on the dual rebuild case */
5754 for (d
= array
->devs
; d
; d
= d
->next
)
5755 if (d
->recovery_start
== 0) {
5762 /* (?) none of the disks are marked with
5763 * IMSM_ORD_REBUILD, so assume they are missing and the
5764 * disk_ord_tbl was not correctly updated
5766 dprintf("%s: failed to locate out-of-sync disk\n", __func__
);
5770 units
= __le32_to_cpu(dev
->vol
.curr_migr_unit
);
5771 rebuild
->recovery_start
= units
* blocks_per_migr_unit(super
, dev
);
5775 static int recover_backup_imsm(struct supertype
*st
, struct mdinfo
*info
);
5778 static struct mdinfo
*container_content_imsm(struct supertype
*st
, char *subarray
)
5780 /* Given a container loaded by load_super_imsm_all,
5781 * extract information about all the arrays into
5783 * If 'subarray' is given, just extract info about that array.
5785 * For each imsm_dev create an mdinfo, fill it in,
5786 * then look for matching devices in super->disks
5787 * and create appropriate device mdinfo.
5789 struct intel_super
*super
= st
->sb
;
5790 struct imsm_super
*mpb
= super
->anchor
;
5791 struct mdinfo
*rest
= NULL
;
5795 int spare_disks
= 0;
5797 /* do not assemble arrays when not all attributes are supported */
5798 if (imsm_check_attributes(mpb
->attributes
) == 0) {
5800 fprintf(stderr
, Name
": Unsupported attributes in IMSM metadata."
5801 "Arrays activation is blocked.\n");
5804 /* check for bad blocks */
5805 if (imsm_bbm_log_size(super
->anchor
)) {
5806 fprintf(stderr
, Name
": BBM log found in IMSM metadata."
5807 "Arrays activation is blocked.\n");
5812 /* count spare devices, not used in maps
5814 for (d
= super
->disks
; d
; d
= d
->next
)
5818 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
5819 struct imsm_dev
*dev
;
5820 struct imsm_map
*map
;
5821 struct imsm_map
*map2
;
5822 struct mdinfo
*this;
5827 (i
!= strtoul(subarray
, &ep
, 10) || *ep
!= '\0'))
5830 dev
= get_imsm_dev(super
, i
);
5831 map
= get_imsm_map(dev
, MAP_0
);
5832 map2
= get_imsm_map(dev
, MAP_1
);
5834 /* do not publish arrays that are in the middle of an
5835 * unsupported migration
5837 if (dev
->vol
.migr_state
&&
5838 (migr_type(dev
) == MIGR_STATE_CHANGE
)) {
5839 fprintf(stderr
, Name
": cannot assemble volume '%.16s':"
5840 " unsupported migration in progress\n",
5844 /* do not publish arrays that are not support by controller's
5848 chunk
= __le16_to_cpu(map
->blocks_per_strip
) >> 1;
5849 this = malloc(sizeof(*this));
5851 fprintf(stderr
, Name
": failed to allocate %zu bytes\n",
5856 super
->current_vol
= i
;
5857 getinfo_super_imsm_volume(st
, this, NULL
);
5860 /* mdadm does not support all metadata features- set the bit in all arrays state */
5861 if (!validate_geometry_imsm_orom(super
,
5862 get_imsm_raid_level(map
), /* RAID level */
5863 imsm_level_to_layout(get_imsm_raid_level(map
)),
5864 map
->num_members
, /* raid disks */
5867 fprintf(stderr
, Name
": IMSM RAID geometry validation"
5868 " failed. Array %s activation is blocked.\n",
5870 this->array
.state
|=
5871 (1<<MD_SB_BLOCK_CONTAINER_RESHAPE
) |
5872 (1<<MD_SB_BLOCK_VOLUME
);
5876 /* if array has bad blocks, set suitable bit in all arrays state */
5878 this->array
.state
|=
5879 (1<<MD_SB_BLOCK_CONTAINER_RESHAPE
) |
5880 (1<<MD_SB_BLOCK_VOLUME
);
5882 for (slot
= 0 ; slot
< map
->num_members
; slot
++) {
5883 unsigned long long recovery_start
;
5884 struct mdinfo
*info_d
;
5891 idx
= get_imsm_disk_idx(dev
, slot
, MAP_0
);
5892 ord
= get_imsm_ord_tbl_ent(dev
, slot
, MAP_X
);
5893 for (d
= super
->disks
; d
; d
= d
->next
)
5894 if (d
->index
== idx
)
5897 recovery_start
= MaxSector
;
5900 if (d
&& is_failed(&d
->disk
))
5902 if (ord
& IMSM_ORD_REBUILD
)
5906 * if we skip some disks the array will be assmebled degraded;
5907 * reset resync start to avoid a dirty-degraded
5908 * situation when performing the intial sync
5910 * FIXME handle dirty degraded
5912 if ((skip
|| recovery_start
== 0) && !dev
->vol
.dirty
)
5913 this->resync_start
= MaxSector
;
5917 info_d
= calloc(1, sizeof(*info_d
));
5919 fprintf(stderr
, Name
": failed to allocate disk"
5920 " for volume %.16s\n", dev
->volume
);
5921 info_d
= this->devs
;
5923 struct mdinfo
*d
= info_d
->next
;
5932 info_d
->next
= this->devs
;
5933 this->devs
= info_d
;
5935 info_d
->disk
.number
= d
->index
;
5936 info_d
->disk
.major
= d
->major
;
5937 info_d
->disk
.minor
= d
->minor
;
5938 info_d
->disk
.raid_disk
= slot
;
5939 info_d
->recovery_start
= recovery_start
;
5941 if (slot
< map2
->num_members
)
5942 info_d
->disk
.state
= (1 << MD_DISK_ACTIVE
);
5944 this->array
.spare_disks
++;
5946 if (slot
< map
->num_members
)
5947 info_d
->disk
.state
= (1 << MD_DISK_ACTIVE
);
5949 this->array
.spare_disks
++;
5951 if (info_d
->recovery_start
== MaxSector
)
5952 this->array
.working_disks
++;
5954 info_d
->events
= __le32_to_cpu(mpb
->generation_num
);
5955 info_d
->data_offset
= __le32_to_cpu(map
->pba_of_lba0
);
5956 info_d
->component_size
= __le32_to_cpu(map
->blocks_per_member
);
5958 /* now that the disk list is up-to-date fixup recovery_start */
5959 update_recovery_start(super
, dev
, this);
5960 this->array
.spare_disks
+= spare_disks
;
5963 /* check for reshape */
5964 if (this->reshape_active
== 1)
5965 recover_backup_imsm(st
, this);
5974 static __u8
imsm_check_degraded(struct intel_super
*super
, struct imsm_dev
*dev
,
5975 int failed
, int look_in_map
)
5977 struct imsm_map
*map
;
5979 map
= get_imsm_map(dev
, look_in_map
);
5982 return map
->map_state
== IMSM_T_STATE_UNINITIALIZED
?
5983 IMSM_T_STATE_UNINITIALIZED
: IMSM_T_STATE_NORMAL
;
5985 switch (get_imsm_raid_level(map
)) {
5987 return IMSM_T_STATE_FAILED
;
5990 if (failed
< map
->num_members
)
5991 return IMSM_T_STATE_DEGRADED
;
5993 return IMSM_T_STATE_FAILED
;
5998 * check to see if any mirrors have failed, otherwise we
5999 * are degraded. Even numbered slots are mirrored on
6003 /* gcc -Os complains that this is unused */
6004 int insync
= insync
;
6006 for (i
= 0; i
< map
->num_members
; i
++) {
6007 __u32 ord
= get_imsm_ord_tbl_ent(dev
, i
, MAP_X
);
6008 int idx
= ord_to_idx(ord
);
6009 struct imsm_disk
*disk
;
6011 /* reset the potential in-sync count on even-numbered
6012 * slots. num_copies is always 2 for imsm raid10
6017 disk
= get_imsm_disk(super
, idx
);
6018 if (!disk
|| is_failed(disk
) || ord
& IMSM_ORD_REBUILD
)
6021 /* no in-sync disks left in this mirror the
6025 return IMSM_T_STATE_FAILED
;
6028 return IMSM_T_STATE_DEGRADED
;
6032 return IMSM_T_STATE_DEGRADED
;
6034 return IMSM_T_STATE_FAILED
;
6040 return map
->map_state
;
6043 static int imsm_count_failed(struct intel_super
*super
, struct imsm_dev
*dev
,
6048 struct imsm_disk
*disk
;
6049 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
6050 struct imsm_map
*prev
= get_imsm_map(dev
, MAP_1
);
6051 struct imsm_map
*map_for_loop
;
6056 /* at the beginning of migration we set IMSM_ORD_REBUILD on
6057 * disks that are being rebuilt. New failures are recorded to
6058 * map[0]. So we look through all the disks we started with and
6059 * see if any failures are still present, or if any new ones
6063 if (prev
&& (map
->num_members
< prev
->num_members
))
6064 map_for_loop
= prev
;
6066 for (i
= 0; i
< map_for_loop
->num_members
; i
++) {
6068 /* when MAP_X is passed both maps failures are counted
6071 ((look_in_map
== MAP_1
) || (look_in_map
== MAP_X
)) &&
6072 (i
< prev
->num_members
)) {
6073 ord
= __le32_to_cpu(prev
->disk_ord_tbl
[i
]);
6074 idx_1
= ord_to_idx(ord
);
6076 disk
= get_imsm_disk(super
, idx_1
);
6077 if (!disk
|| is_failed(disk
) || ord
& IMSM_ORD_REBUILD
)
6080 if (((look_in_map
== MAP_0
) || (look_in_map
== MAP_X
)) &&
6081 (i
< map
->num_members
)) {
6082 ord
= __le32_to_cpu(map
->disk_ord_tbl
[i
]);
6083 idx
= ord_to_idx(ord
);
6086 disk
= get_imsm_disk(super
, idx
);
6087 if (!disk
|| is_failed(disk
) ||
6088 ord
& IMSM_ORD_REBUILD
)
6098 static int imsm_open_new(struct supertype
*c
, struct active_array
*a
,
6101 struct intel_super
*super
= c
->sb
;
6102 struct imsm_super
*mpb
= super
->anchor
;
6104 if (atoi(inst
) >= mpb
->num_raid_devs
) {
6105 fprintf(stderr
, "%s: subarry index %d, out of range\n",
6106 __func__
, atoi(inst
));
6110 dprintf("imsm: open_new %s\n", inst
);
6111 a
->info
.container_member
= atoi(inst
);
6115 static int is_resyncing(struct imsm_dev
*dev
)
6117 struct imsm_map
*migr_map
;
6119 if (!dev
->vol
.migr_state
)
6122 if (migr_type(dev
) == MIGR_INIT
||
6123 migr_type(dev
) == MIGR_REPAIR
)
6126 if (migr_type(dev
) == MIGR_GEN_MIGR
)
6129 migr_map
= get_imsm_map(dev
, MAP_1
);
6131 if ((migr_map
->map_state
== IMSM_T_STATE_NORMAL
) &&
6132 (dev
->vol
.migr_type
!= MIGR_GEN_MIGR
))
6138 /* return true if we recorded new information */
6139 static int mark_failure(struct imsm_dev
*dev
, struct imsm_disk
*disk
, int idx
)
6143 struct imsm_map
*map
;
6144 char buf
[MAX_RAID_SERIAL_LEN
+3];
6145 unsigned int len
, shift
= 0;
6147 /* new failures are always set in map[0] */
6148 map
= get_imsm_map(dev
, MAP_0
);
6150 slot
= get_imsm_disk_slot(map
, idx
);
6154 ord
= __le32_to_cpu(map
->disk_ord_tbl
[slot
]);
6155 if (is_failed(disk
) && (ord
& IMSM_ORD_REBUILD
))
6158 memcpy(buf
, disk
->serial
, MAX_RAID_SERIAL_LEN
);
6159 buf
[MAX_RAID_SERIAL_LEN
] = '\000';
6161 if ((len
= strlen(buf
)) >= MAX_RAID_SERIAL_LEN
)
6162 shift
= len
- MAX_RAID_SERIAL_LEN
+ 1;
6163 strncpy((char *)disk
->serial
, &buf
[shift
], MAX_RAID_SERIAL_LEN
);
6165 disk
->status
|= FAILED_DISK
;
6166 set_imsm_ord_tbl_ent(map
, slot
, idx
| IMSM_ORD_REBUILD
);
6167 /* mark failures in second map if second map exists and this disk
6169 * This is valid for migration, initialization and rebuild
6171 if (dev
->vol
.migr_state
) {
6172 struct imsm_map
*map2
= get_imsm_map(dev
, MAP_1
);
6173 int slot2
= get_imsm_disk_slot(map2
, idx
);
6175 if ((slot2
< map2
->num_members
) &&
6177 set_imsm_ord_tbl_ent(map2
, slot2
,
6178 idx
| IMSM_ORD_REBUILD
);
6180 if (map
->failed_disk_num
== 0xff)
6181 map
->failed_disk_num
= slot
;
6185 static void mark_missing(struct imsm_dev
*dev
, struct imsm_disk
*disk
, int idx
)
6187 mark_failure(dev
, disk
, idx
);
6189 if (disk
->scsi_id
== __cpu_to_le32(~(__u32
)0))
6192 disk
->scsi_id
= __cpu_to_le32(~(__u32
)0);
6193 memmove(&disk
->serial
[0], &disk
->serial
[1], MAX_RAID_SERIAL_LEN
- 1);
6196 static void handle_missing(struct intel_super
*super
, struct imsm_dev
*dev
)
6200 if (!super
->missing
)
6203 dprintf("imsm: mark missing\n");
6204 /* end process for initialization and rebuild only
6206 if (is_gen_migration(dev
) == 0) {
6210 failed
= imsm_count_failed(super
, dev
, MAP_0
);
6211 map_state
= imsm_check_degraded(super
, dev
, failed
, MAP_0
);
6213 end_migration(dev
, super
, map_state
);
6215 for (dl
= super
->missing
; dl
; dl
= dl
->next
)
6216 mark_missing(dev
, &dl
->disk
, dl
->index
);
6217 super
->updates_pending
++;
6220 static unsigned long long imsm_set_array_size(struct imsm_dev
*dev
)
6222 int used_disks
= imsm_num_data_members(dev
, MAP_0
);
6223 unsigned long long array_blocks
;
6224 struct imsm_map
*map
;
6226 if (used_disks
== 0) {
6227 /* when problems occures
6228 * return current array_blocks value
6230 array_blocks
= __le32_to_cpu(dev
->size_high
);
6231 array_blocks
= array_blocks
<< 32;
6232 array_blocks
+= __le32_to_cpu(dev
->size_low
);
6234 return array_blocks
;
6237 /* set array size in metadata
6239 map
= get_imsm_map(dev
, MAP_0
);
6240 array_blocks
= map
->blocks_per_member
* used_disks
;
6242 /* round array size down to closest MB
6244 array_blocks
= (array_blocks
>> SECT_PER_MB_SHIFT
) << SECT_PER_MB_SHIFT
;
6245 dev
->size_low
= __cpu_to_le32((__u32
)array_blocks
);
6246 dev
->size_high
= __cpu_to_le32((__u32
)(array_blocks
>> 32));
6248 return array_blocks
;
6251 static void imsm_set_disk(struct active_array
*a
, int n
, int state
);
6253 static void imsm_progress_container_reshape(struct intel_super
*super
)
6255 /* if no device has a migr_state, but some device has a
6256 * different number of members than the previous device, start
6257 * changing the number of devices in this device to match
6260 struct imsm_super
*mpb
= super
->anchor
;
6261 int prev_disks
= -1;
6265 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
6266 struct imsm_dev
*dev
= get_imsm_dev(super
, i
);
6267 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
6268 struct imsm_map
*map2
;
6269 int prev_num_members
;
6271 if (dev
->vol
.migr_state
)
6274 if (prev_disks
== -1)
6275 prev_disks
= map
->num_members
;
6276 if (prev_disks
== map
->num_members
)
6279 /* OK, this array needs to enter reshape mode.
6280 * i.e it needs a migr_state
6283 copy_map_size
= sizeof_imsm_map(map
);
6284 prev_num_members
= map
->num_members
;
6285 map
->num_members
= prev_disks
;
6286 dev
->vol
.migr_state
= 1;
6287 dev
->vol
.curr_migr_unit
= 0;
6288 set_migr_type(dev
, MIGR_GEN_MIGR
);
6289 for (i
= prev_num_members
;
6290 i
< map
->num_members
; i
++)
6291 set_imsm_ord_tbl_ent(map
, i
, i
);
6292 map2
= get_imsm_map(dev
, MAP_1
);
6293 /* Copy the current map */
6294 memcpy(map2
, map
, copy_map_size
);
6295 map2
->num_members
= prev_num_members
;
6297 imsm_set_array_size(dev
);
6298 super
->updates_pending
++;
6302 /* Handle dirty -> clean transititions, resync and reshape. Degraded and rebuild
6303 * states are handled in imsm_set_disk() with one exception, when a
6304 * resync is stopped due to a new failure this routine will set the
6305 * 'degraded' state for the array.
6307 static int imsm_set_array_state(struct active_array
*a
, int consistent
)
6309 int inst
= a
->info
.container_member
;
6310 struct intel_super
*super
= a
->container
->sb
;
6311 struct imsm_dev
*dev
= get_imsm_dev(super
, inst
);
6312 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
6313 int failed
= imsm_count_failed(super
, dev
, MAP_0
);
6314 __u8 map_state
= imsm_check_degraded(super
, dev
, failed
, MAP_0
);
6315 __u32 blocks_per_unit
;
6317 if (dev
->vol
.migr_state
&&
6318 dev
->vol
.migr_type
== MIGR_GEN_MIGR
) {
6319 /* array state change is blocked due to reshape action
6321 * - abort the reshape (if last_checkpoint is 0 and action!= reshape)
6322 * - finish the reshape (if last_checkpoint is big and action != reshape)
6323 * - update curr_migr_unit
6325 if (a
->curr_action
== reshape
) {
6326 /* still reshaping, maybe update curr_migr_unit */
6327 goto mark_checkpoint
;
6329 if (a
->last_checkpoint
== 0 && a
->prev_action
== reshape
) {
6330 /* for some reason we aborted the reshape.
6332 * disable automatic metadata rollback
6333 * user action is required to recover process
6336 struct imsm_map
*map2
=
6337 get_imsm_map(dev
, MAP_1
);
6338 dev
->vol
.migr_state
= 0;
6339 set_migr_type(dev
, 0);
6340 dev
->vol
.curr_migr_unit
= 0;
6342 sizeof_imsm_map(map2
));
6343 super
->updates_pending
++;
6346 if (a
->last_checkpoint
>= a
->info
.component_size
) {
6347 unsigned long long array_blocks
;
6351 used_disks
= imsm_num_data_members(dev
, MAP_0
);
6352 if (used_disks
> 0) {
6354 map
->blocks_per_member
*
6356 /* round array size down to closest MB
6358 array_blocks
= (array_blocks
6359 >> SECT_PER_MB_SHIFT
)
6360 << SECT_PER_MB_SHIFT
;
6361 a
->info
.custom_array_size
= array_blocks
;
6362 /* encourage manager to update array
6366 a
->check_reshape
= 1;
6368 /* finalize online capacity expansion/reshape */
6369 for (mdi
= a
->info
.devs
; mdi
; mdi
= mdi
->next
)
6371 mdi
->disk
.raid_disk
,
6374 imsm_progress_container_reshape(super
);
6379 /* before we activate this array handle any missing disks */
6380 if (consistent
== 2)
6381 handle_missing(super
, dev
);
6383 if (consistent
== 2 &&
6384 (!is_resync_complete(&a
->info
) ||
6385 map_state
!= IMSM_T_STATE_NORMAL
||
6386 dev
->vol
.migr_state
))
6389 if (is_resync_complete(&a
->info
)) {
6390 /* complete intialization / resync,
6391 * recovery and interrupted recovery is completed in
6394 if (is_resyncing(dev
)) {
6395 dprintf("imsm: mark resync done\n");
6396 end_migration(dev
, super
, map_state
);
6397 super
->updates_pending
++;
6398 a
->last_checkpoint
= 0;
6400 } else if ((!is_resyncing(dev
) && !failed
) &&
6401 (imsm_reshape_blocks_arrays_changes(super
) == 0)) {
6402 /* mark the start of the init process if nothing is failed */
6403 dprintf("imsm: mark resync start\n");
6404 if (map
->map_state
== IMSM_T_STATE_UNINITIALIZED
)
6405 migrate(dev
, super
, IMSM_T_STATE_NORMAL
, MIGR_INIT
);
6407 migrate(dev
, super
, IMSM_T_STATE_NORMAL
, MIGR_REPAIR
);
6408 super
->updates_pending
++;
6412 /* skip checkpointing for general migration,
6413 * it is controlled in mdadm
6415 if (is_gen_migration(dev
))
6416 goto skip_mark_checkpoint
;
6418 /* check if we can update curr_migr_unit from resync_start, recovery_start */
6419 blocks_per_unit
= blocks_per_migr_unit(super
, dev
);
6420 if (blocks_per_unit
) {
6424 units
= a
->last_checkpoint
/ blocks_per_unit
;
6427 /* check that we did not overflow 32-bits, and that
6428 * curr_migr_unit needs updating
6430 if (units32
== units
&&
6432 __le32_to_cpu(dev
->vol
.curr_migr_unit
) != units32
) {
6433 dprintf("imsm: mark checkpoint (%u)\n", units32
);
6434 dev
->vol
.curr_migr_unit
= __cpu_to_le32(units32
);
6435 super
->updates_pending
++;
6439 skip_mark_checkpoint
:
6440 /* mark dirty / clean */
6441 if (dev
->vol
.dirty
!= !consistent
) {
6442 dprintf("imsm: mark '%s'\n", consistent
? "clean" : "dirty");
6447 super
->updates_pending
++;
6453 static void imsm_set_disk(struct active_array
*a
, int n
, int state
)
6455 int inst
= a
->info
.container_member
;
6456 struct intel_super
*super
= a
->container
->sb
;
6457 struct imsm_dev
*dev
= get_imsm_dev(super
, inst
);
6458 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
6459 struct imsm_disk
*disk
;
6464 if (n
> map
->num_members
)
6465 fprintf(stderr
, "imsm: set_disk %d out of range 0..%d\n",
6466 n
, map
->num_members
- 1);
6471 dprintf("imsm: set_disk %d:%x\n", n
, state
);
6473 ord
= get_imsm_ord_tbl_ent(dev
, n
, MAP_0
);
6474 disk
= get_imsm_disk(super
, ord_to_idx(ord
));
6476 /* check for new failures */
6477 if (state
& DS_FAULTY
) {
6478 if (mark_failure(dev
, disk
, ord_to_idx(ord
)))
6479 super
->updates_pending
++;
6482 /* check if in_sync */
6483 if (state
& DS_INSYNC
&& ord
& IMSM_ORD_REBUILD
&& is_rebuilding(dev
)) {
6484 struct imsm_map
*migr_map
= get_imsm_map(dev
, MAP_1
);
6486 set_imsm_ord_tbl_ent(migr_map
, n
, ord_to_idx(ord
));
6487 super
->updates_pending
++;
6490 failed
= imsm_count_failed(super
, dev
, MAP_0
);
6491 map_state
= imsm_check_degraded(super
, dev
, failed
, MAP_0
);
6493 /* check if recovery complete, newly degraded, or failed */
6494 dprintf("imsm: Detected transition to state ");
6495 switch (map_state
) {
6496 case IMSM_T_STATE_NORMAL
: /* transition to normal state */
6497 dprintf("normal: ");
6498 if (is_rebuilding(dev
)) {
6499 dprintf("while rebuilding");
6500 end_migration(dev
, super
, map_state
);
6501 map
= get_imsm_map(dev
, MAP_0
);
6502 map
->failed_disk_num
= ~0;
6503 super
->updates_pending
++;
6504 a
->last_checkpoint
= 0;
6507 if (is_gen_migration(dev
)) {
6508 dprintf("while general migration");
6509 if (a
->last_checkpoint
>= a
->info
.component_size
)
6510 end_migration(dev
, super
, map_state
);
6512 map
->map_state
= map_state
;
6513 map
= get_imsm_map(dev
, MAP_0
);
6514 map
->failed_disk_num
= ~0;
6515 super
->updates_pending
++;
6519 case IMSM_T_STATE_DEGRADED
: /* transition to degraded state */
6520 dprintf("degraded: ");
6521 if ((map
->map_state
!= map_state
) &&
6522 !dev
->vol
.migr_state
) {
6523 dprintf("mark degraded");
6524 map
->map_state
= map_state
;
6525 super
->updates_pending
++;
6526 a
->last_checkpoint
= 0;
6529 if (is_rebuilding(dev
)) {
6530 dprintf("while rebuilding.");
6531 if (map
->map_state
!= map_state
) {
6532 dprintf(" Map state change");
6533 end_migration(dev
, super
, map_state
);
6534 super
->updates_pending
++;
6538 if (is_gen_migration(dev
)) {
6539 dprintf("while general migration");
6540 if (a
->last_checkpoint
>= a
->info
.component_size
)
6541 end_migration(dev
, super
, map_state
);
6543 map
->map_state
= map_state
;
6544 manage_second_map(super
, dev
);
6546 super
->updates_pending
++;
6549 if (is_initializing(dev
)) {
6550 dprintf("while initialization.");
6551 map
->map_state
= map_state
;
6552 super
->updates_pending
++;
6556 case IMSM_T_STATE_FAILED
: /* transition to failed state */
6557 dprintf("failed: ");
6558 if (is_gen_migration(dev
)) {
6559 dprintf("while general migration");
6560 map
->map_state
= map_state
;
6561 super
->updates_pending
++;
6564 if (map
->map_state
!= map_state
) {
6565 dprintf("mark failed");
6566 end_migration(dev
, super
, map_state
);
6567 super
->updates_pending
++;
6568 a
->last_checkpoint
= 0;
6573 dprintf("state %i\n", map_state
);
6579 static int store_imsm_mpb(int fd
, struct imsm_super
*mpb
)
6582 __u32 mpb_size
= __le32_to_cpu(mpb
->mpb_size
);
6583 unsigned long long dsize
;
6584 unsigned long long sectors
;
6586 get_dev_size(fd
, NULL
, &dsize
);
6588 if (mpb_size
> 512) {
6589 /* -1 to account for anchor */
6590 sectors
= mpb_sectors(mpb
) - 1;
6592 /* write the extended mpb to the sectors preceeding the anchor */
6593 if (lseek64(fd
, dsize
- (512 * (2 + sectors
)), SEEK_SET
) < 0)
6596 if ((unsigned long long)write(fd
, buf
+ 512, 512 * sectors
)
6601 /* first block is stored on second to last sector of the disk */
6602 if (lseek64(fd
, dsize
- (512 * 2), SEEK_SET
) < 0)
6605 if (write(fd
, buf
, 512) != 512)
6611 static void imsm_sync_metadata(struct supertype
*container
)
6613 struct intel_super
*super
= container
->sb
;
6615 dprintf("sync metadata: %d\n", super
->updates_pending
);
6616 if (!super
->updates_pending
)
6619 write_super_imsm(container
, 0);
6621 super
->updates_pending
= 0;
6624 static struct dl
*imsm_readd(struct intel_super
*super
, int idx
, struct active_array
*a
)
6626 struct imsm_dev
*dev
= get_imsm_dev(super
, a
->info
.container_member
);
6627 int i
= get_imsm_disk_idx(dev
, idx
, MAP_X
);
6630 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
6634 if (dl
&& is_failed(&dl
->disk
))
6638 dprintf("%s: found %x:%x\n", __func__
, dl
->major
, dl
->minor
);
6643 static struct dl
*imsm_add_spare(struct intel_super
*super
, int slot
,
6644 struct active_array
*a
, int activate_new
,
6645 struct mdinfo
*additional_test_list
)
6647 struct imsm_dev
*dev
= get_imsm_dev(super
, a
->info
.container_member
);
6648 int idx
= get_imsm_disk_idx(dev
, slot
, MAP_X
);
6649 struct imsm_super
*mpb
= super
->anchor
;
6650 struct imsm_map
*map
;
6651 unsigned long long pos
;
6656 __u32 array_start
= 0;
6657 __u32 array_end
= 0;
6659 struct mdinfo
*test_list
;
6661 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
6662 /* If in this array, skip */
6663 for (d
= a
->info
.devs
; d
; d
= d
->next
)
6664 if (d
->state_fd
>= 0 &&
6665 d
->disk
.major
== dl
->major
&&
6666 d
->disk
.minor
== dl
->minor
) {
6667 dprintf("%x:%x already in array\n",
6668 dl
->major
, dl
->minor
);
6673 test_list
= additional_test_list
;
6675 if (test_list
->disk
.major
== dl
->major
&&
6676 test_list
->disk
.minor
== dl
->minor
) {
6677 dprintf("%x:%x already in additional test list\n",
6678 dl
->major
, dl
->minor
);
6681 test_list
= test_list
->next
;
6686 /* skip in use or failed drives */
6687 if (is_failed(&dl
->disk
) || idx
== dl
->index
||
6689 dprintf("%x:%x status (failed: %d index: %d)\n",
6690 dl
->major
, dl
->minor
, is_failed(&dl
->disk
), idx
);
6694 /* skip pure spares when we are looking for partially
6695 * assimilated drives
6697 if (dl
->index
== -1 && !activate_new
)
6700 /* Does this unused device have the requisite free space?
6701 * It needs to be able to cover all member volumes
6703 ex
= get_extents(super
, dl
);
6705 dprintf("cannot get extents\n");
6708 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
6709 dev
= get_imsm_dev(super
, i
);
6710 map
= get_imsm_map(dev
, MAP_0
);
6712 /* check if this disk is already a member of
6715 if (get_imsm_disk_slot(map
, dl
->index
) >= 0)
6721 array_start
= __le32_to_cpu(map
->pba_of_lba0
);
6722 array_end
= array_start
+
6723 __le32_to_cpu(map
->blocks_per_member
) - 1;
6726 /* check that we can start at pba_of_lba0 with
6727 * blocks_per_member of space
6729 if (array_start
>= pos
&& array_end
< ex
[j
].start
) {
6733 pos
= ex
[j
].start
+ ex
[j
].size
;
6735 } while (ex
[j
-1].size
);
6742 if (i
< mpb
->num_raid_devs
) {
6743 dprintf("%x:%x does not have %u to %u available\n",
6744 dl
->major
, dl
->minor
, array_start
, array_end
);
6755 static int imsm_rebuild_allowed(struct supertype
*cont
, int dev_idx
, int failed
)
6757 struct imsm_dev
*dev2
;
6758 struct imsm_map
*map
;
6764 dev2
= get_imsm_dev(cont
->sb
, dev_idx
);
6766 state
= imsm_check_degraded(cont
->sb
, dev2
, failed
, MAP_0
);
6767 if (state
== IMSM_T_STATE_FAILED
) {
6768 map
= get_imsm_map(dev2
, MAP_0
);
6771 for (slot
= 0; slot
< map
->num_members
; slot
++) {
6773 * Check if failed disks are deleted from intel
6774 * disk list or are marked to be deleted
6776 idx
= get_imsm_disk_idx(dev2
, slot
, MAP_X
);
6777 idisk
= get_imsm_dl_disk(cont
->sb
, idx
);
6779 * Do not rebuild the array if failed disks
6780 * from failed sub-array are not removed from
6784 is_failed(&idisk
->disk
) &&
6785 (idisk
->action
!= DISK_REMOVE
))
6793 static struct mdinfo
*imsm_activate_spare(struct active_array
*a
,
6794 struct metadata_update
**updates
)
6797 * Find a device with unused free space and use it to replace a
6798 * failed/vacant region in an array. We replace failed regions one a
6799 * array at a time. The result is that a new spare disk will be added
6800 * to the first failed array and after the monitor has finished
6801 * propagating failures the remainder will be consumed.
6803 * FIXME add a capability for mdmon to request spares from another
6807 struct intel_super
*super
= a
->container
->sb
;
6808 int inst
= a
->info
.container_member
;
6809 struct imsm_dev
*dev
= get_imsm_dev(super
, inst
);
6810 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
6811 int failed
= a
->info
.array
.raid_disks
;
6812 struct mdinfo
*rv
= NULL
;
6815 struct metadata_update
*mu
;
6817 struct imsm_update_activate_spare
*u
;
6822 for (d
= a
->info
.devs
; d
; d
= d
->next
) {
6823 if ((d
->curr_state
& DS_FAULTY
) &&
6825 /* wait for Removal to happen */
6827 if (d
->state_fd
>= 0)
6831 dprintf("imsm: activate spare: inst=%d failed=%d (%d) level=%d\n",
6832 inst
, failed
, a
->info
.array
.raid_disks
, a
->info
.array
.level
);
6834 if (imsm_reshape_blocks_arrays_changes(super
))
6837 /* Cannot activate another spare if rebuild is in progress already
6839 if (is_rebuilding(dev
)) {
6840 dprintf("imsm: No spare activation allowed. "
6841 "Rebuild in progress already.\n");
6845 if (a
->info
.array
.level
== 4)
6846 /* No repair for takeovered array
6847 * imsm doesn't support raid4
6851 if (imsm_check_degraded(super
, dev
, failed
, MAP_0
) !=
6852 IMSM_T_STATE_DEGRADED
)
6856 * If there are any failed disks check state of the other volume.
6857 * Block rebuild if the another one is failed until failed disks
6858 * are removed from container.
6861 dprintf("found failed disks in %.*s, check if there another"
6862 "failed sub-array.\n",
6863 MAX_RAID_SERIAL_LEN
, dev
->volume
);
6864 /* check if states of the other volumes allow for rebuild */
6865 for (i
= 0; i
< super
->anchor
->num_raid_devs
; i
++) {
6867 allowed
= imsm_rebuild_allowed(a
->container
,
6875 /* For each slot, if it is not working, find a spare */
6876 for (i
= 0; i
< a
->info
.array
.raid_disks
; i
++) {
6877 for (d
= a
->info
.devs
; d
; d
= d
->next
)
6878 if (d
->disk
.raid_disk
== i
)
6880 dprintf("found %d: %p %x\n", i
, d
, d
?d
->curr_state
:0);
6881 if (d
&& (d
->state_fd
>= 0))
6885 * OK, this device needs recovery. Try to re-add the
6886 * previous occupant of this slot, if this fails see if
6887 * we can continue the assimilation of a spare that was
6888 * partially assimilated, finally try to activate a new
6891 dl
= imsm_readd(super
, i
, a
);
6893 dl
= imsm_add_spare(super
, i
, a
, 0, rv
);
6895 dl
= imsm_add_spare(super
, i
, a
, 1, rv
);
6899 /* found a usable disk with enough space */
6900 di
= malloc(sizeof(*di
));
6903 memset(di
, 0, sizeof(*di
));
6905 /* dl->index will be -1 in the case we are activating a
6906 * pristine spare. imsm_process_update() will create a
6907 * new index in this case. Once a disk is found to be
6908 * failed in all member arrays it is kicked from the
6911 di
->disk
.number
= dl
->index
;
6913 /* (ab)use di->devs to store a pointer to the device
6916 di
->devs
= (struct mdinfo
*) dl
;
6918 di
->disk
.raid_disk
= i
;
6919 di
->disk
.major
= dl
->major
;
6920 di
->disk
.minor
= dl
->minor
;
6922 di
->recovery_start
= 0;
6923 di
->data_offset
= __le32_to_cpu(map
->pba_of_lba0
);
6924 di
->component_size
= a
->info
.component_size
;
6925 di
->container_member
= inst
;
6926 super
->random
= random32();
6930 dprintf("%x:%x to be %d at %llu\n", dl
->major
, dl
->minor
,
6931 i
, di
->data_offset
);
6935 /* No spares found */
6937 /* Now 'rv' has a list of devices to return.
6938 * Create a metadata_update record to update the
6939 * disk_ord_tbl for the array
6941 mu
= malloc(sizeof(*mu
));
6943 mu
->buf
= malloc(sizeof(struct imsm_update_activate_spare
) * num_spares
);
6944 if (mu
->buf
== NULL
) {
6951 struct mdinfo
*n
= rv
->next
;
6960 mu
->space_list
= NULL
;
6961 mu
->len
= sizeof(struct imsm_update_activate_spare
) * num_spares
;
6962 mu
->next
= *updates
;
6963 u
= (struct imsm_update_activate_spare
*) mu
->buf
;
6965 for (di
= rv
; di
; di
= di
->next
) {
6966 u
->type
= update_activate_spare
;
6967 u
->dl
= (struct dl
*) di
->devs
;
6969 u
->slot
= di
->disk
.raid_disk
;
6980 static int disks_overlap(struct intel_super
*super
, int idx
, struct imsm_update_create_array
*u
)
6982 struct imsm_dev
*dev
= get_imsm_dev(super
, idx
);
6983 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
6984 struct imsm_map
*new_map
= get_imsm_map(&u
->dev
, MAP_0
);
6985 struct disk_info
*inf
= get_disk_info(u
);
6986 struct imsm_disk
*disk
;
6990 for (i
= 0; i
< map
->num_members
; i
++) {
6991 disk
= get_imsm_disk(super
, get_imsm_disk_idx(dev
, i
, MAP_X
));
6992 for (j
= 0; j
< new_map
->num_members
; j
++)
6993 if (serialcmp(disk
->serial
, inf
[j
].serial
) == 0)
7001 static struct dl
*get_disk_super(struct intel_super
*super
, int major
, int minor
)
7003 struct dl
*dl
= NULL
;
7004 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
7005 if ((dl
->major
== major
) && (dl
->minor
== minor
))
7010 static int remove_disk_super(struct intel_super
*super
, int major
, int minor
)
7012 struct dl
*prev
= NULL
;
7016 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
7017 if ((dl
->major
== major
) && (dl
->minor
== minor
)) {
7020 prev
->next
= dl
->next
;
7022 super
->disks
= dl
->next
;
7024 __free_imsm_disk(dl
);
7025 dprintf("%s: removed %x:%x\n",
7026 __func__
, major
, minor
);
7034 static void imsm_delete(struct intel_super
*super
, struct dl
**dlp
, unsigned index
);
7036 static int add_remove_disk_update(struct intel_super
*super
)
7038 int check_degraded
= 0;
7039 struct dl
*disk
= NULL
;
7040 /* add/remove some spares to/from the metadata/contrainer */
7041 while (super
->disk_mgmt_list
) {
7042 struct dl
*disk_cfg
;
7044 disk_cfg
= super
->disk_mgmt_list
;
7045 super
->disk_mgmt_list
= disk_cfg
->next
;
7046 disk_cfg
->next
= NULL
;
7048 if (disk_cfg
->action
== DISK_ADD
) {
7049 disk_cfg
->next
= super
->disks
;
7050 super
->disks
= disk_cfg
;
7052 dprintf("%s: added %x:%x\n",
7053 __func__
, disk_cfg
->major
,
7055 } else if (disk_cfg
->action
== DISK_REMOVE
) {
7056 dprintf("Disk remove action processed: %x.%x\n",
7057 disk_cfg
->major
, disk_cfg
->minor
);
7058 disk
= get_disk_super(super
,
7062 /* store action status */
7063 disk
->action
= DISK_REMOVE
;
7064 /* remove spare disks only */
7065 if (disk
->index
== -1) {
7066 remove_disk_super(super
,
7071 /* release allocate disk structure */
7072 __free_imsm_disk(disk_cfg
);
7075 return check_degraded
;
7079 static int apply_reshape_migration_update(struct imsm_update_reshape_migration
*u
,
7080 struct intel_super
*super
,
7083 struct intel_dev
*id
;
7084 void **tofree
= NULL
;
7087 dprintf("apply_reshape_migration_update()\n");
7088 if ((u
->subdev
< 0) ||
7090 dprintf("imsm: Error: Wrong subdev: %i\n", u
->subdev
);
7093 if ((space_list
== NULL
) || (*space_list
== NULL
)) {
7094 dprintf("imsm: Error: Memory is not allocated\n");
7098 for (id
= super
->devlist
; id
; id
= id
->next
) {
7099 if (id
->index
== (unsigned)u
->subdev
) {
7100 struct imsm_dev
*dev
= get_imsm_dev(super
, u
->subdev
);
7101 struct imsm_map
*map
;
7102 struct imsm_dev
*new_dev
=
7103 (struct imsm_dev
*)*space_list
;
7104 struct imsm_map
*migr_map
= get_imsm_map(dev
, MAP_1
);
7106 struct dl
*new_disk
;
7108 if (new_dev
== NULL
)
7110 *space_list
= **space_list
;
7111 memcpy(new_dev
, dev
, sizeof_imsm_dev(dev
, 0));
7112 map
= get_imsm_map(new_dev
, MAP_0
);
7114 dprintf("imsm: Error: migration in progress");
7118 to_state
= map
->map_state
;
7119 if ((u
->new_level
== 5) && (map
->raid_level
== 0)) {
7121 /* this should not happen */
7122 if (u
->new_disks
[0] < 0) {
7123 map
->failed_disk_num
=
7124 map
->num_members
- 1;
7125 to_state
= IMSM_T_STATE_DEGRADED
;
7127 to_state
= IMSM_T_STATE_NORMAL
;
7129 migrate(new_dev
, super
, to_state
, MIGR_GEN_MIGR
);
7130 if (u
->new_level
> -1)
7131 map
->raid_level
= u
->new_level
;
7132 migr_map
= get_imsm_map(new_dev
, MAP_1
);
7133 if ((u
->new_level
== 5) &&
7134 (migr_map
->raid_level
== 0)) {
7135 int ord
= map
->num_members
- 1;
7136 migr_map
->num_members
--;
7137 if (u
->new_disks
[0] < 0)
7138 ord
|= IMSM_ORD_REBUILD
;
7139 set_imsm_ord_tbl_ent(map
,
7140 map
->num_members
- 1,
7144 tofree
= (void **)dev
;
7146 /* update chunk size
7148 if (u
->new_chunksize
> 0)
7149 map
->blocks_per_strip
=
7150 __cpu_to_le16(u
->new_chunksize
* 2);
7154 if ((u
->new_level
!= 5) ||
7155 (migr_map
->raid_level
!= 0) ||
7156 (migr_map
->raid_level
== map
->raid_level
))
7159 if (u
->new_disks
[0] >= 0) {
7162 new_disk
= get_disk_super(super
,
7163 major(u
->new_disks
[0]),
7164 minor(u
->new_disks
[0]));
7165 dprintf("imsm: new disk for reshape is: %i:%i "
7166 "(%p, index = %i)\n",
7167 major(u
->new_disks
[0]),
7168 minor(u
->new_disks
[0]),
7169 new_disk
, new_disk
->index
);
7170 if (new_disk
== NULL
)
7171 goto error_disk_add
;
7173 new_disk
->index
= map
->num_members
- 1;
7174 /* slot to fill in autolayout
7176 new_disk
->raiddisk
= new_disk
->index
;
7177 new_disk
->disk
.status
|= CONFIGURED_DISK
;
7178 new_disk
->disk
.status
&= ~SPARE_DISK
;
7180 goto error_disk_add
;
7183 *tofree
= *space_list
;
7184 /* calculate new size
7186 imsm_set_array_size(new_dev
);
7193 *space_list
= tofree
;
7197 dprintf("Error: imsm: Cannot find disk.\n");
7201 static int apply_update_activate_spare(struct imsm_update_activate_spare
*u
,
7202 struct intel_super
*super
,
7203 struct active_array
*active_array
)
7205 struct imsm_super
*mpb
= super
->anchor
;
7206 struct imsm_dev
*dev
= get_imsm_dev(super
, u
->array
);
7207 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
7208 struct imsm_map
*migr_map
;
7209 struct active_array
*a
;
7210 struct imsm_disk
*disk
;
7217 int second_map_created
= 0;
7219 for (; u
; u
= u
->next
) {
7220 victim
= get_imsm_disk_idx(dev
, u
->slot
, MAP_X
);
7225 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
7230 fprintf(stderr
, "error: imsm_activate_spare passed "
7231 "an unknown disk (index: %d)\n",
7236 /* count failures (excluding rebuilds and the victim)
7237 * to determine map[0] state
7240 for (i
= 0; i
< map
->num_members
; i
++) {
7243 disk
= get_imsm_disk(super
,
7244 get_imsm_disk_idx(dev
, i
, MAP_X
));
7245 if (!disk
|| is_failed(disk
))
7249 /* adding a pristine spare, assign a new index */
7250 if (dl
->index
< 0) {
7251 dl
->index
= super
->anchor
->num_disks
;
7252 super
->anchor
->num_disks
++;
7255 disk
->status
|= CONFIGURED_DISK
;
7256 disk
->status
&= ~SPARE_DISK
;
7259 to_state
= imsm_check_degraded(super
, dev
, failed
, MAP_0
);
7260 if (!second_map_created
) {
7261 second_map_created
= 1;
7262 map
->map_state
= IMSM_T_STATE_DEGRADED
;
7263 migrate(dev
, super
, to_state
, MIGR_REBUILD
);
7265 map
->map_state
= to_state
;
7266 migr_map
= get_imsm_map(dev
, MAP_1
);
7267 set_imsm_ord_tbl_ent(map
, u
->slot
, dl
->index
);
7268 set_imsm_ord_tbl_ent(migr_map
, u
->slot
,
7269 dl
->index
| IMSM_ORD_REBUILD
);
7271 /* update the family_num to mark a new container
7272 * generation, being careful to record the existing
7273 * family_num in orig_family_num to clean up after
7274 * earlier mdadm versions that neglected to set it.
7276 if (mpb
->orig_family_num
== 0)
7277 mpb
->orig_family_num
= mpb
->family_num
;
7278 mpb
->family_num
+= super
->random
;
7280 /* count arrays using the victim in the metadata */
7282 for (a
= active_array
; a
; a
= a
->next
) {
7283 dev
= get_imsm_dev(super
, a
->info
.container_member
);
7284 map
= get_imsm_map(dev
, MAP_0
);
7286 if (get_imsm_disk_slot(map
, victim
) >= 0)
7290 /* delete the victim if it is no longer being
7296 /* We know that 'manager' isn't touching anything,
7297 * so it is safe to delete
7299 for (dlp
= &super
->disks
; *dlp
; dlp
= &(*dlp
)->next
)
7300 if ((*dlp
)->index
== victim
)
7303 /* victim may be on the missing list */
7305 for (dlp
= &super
->missing
; *dlp
;
7306 dlp
= &(*dlp
)->next
)
7307 if ((*dlp
)->index
== victim
)
7309 imsm_delete(super
, dlp
, victim
);
7316 static int apply_reshape_container_disks_update(struct imsm_update_reshape
*u
,
7317 struct intel_super
*super
,
7320 struct dl
*new_disk
;
7321 struct intel_dev
*id
;
7323 int delta_disks
= u
->new_raid_disks
- u
->old_raid_disks
;
7324 int disk_count
= u
->old_raid_disks
;
7325 void **tofree
= NULL
;
7326 int devices_to_reshape
= 1;
7327 struct imsm_super
*mpb
= super
->anchor
;
7329 unsigned int dev_id
;
7331 dprintf("imsm: apply_reshape_container_disks_update()\n");
7333 /* enable spares to use in array */
7334 for (i
= 0; i
< delta_disks
; i
++) {
7335 new_disk
= get_disk_super(super
,
7336 major(u
->new_disks
[i
]),
7337 minor(u
->new_disks
[i
]));
7338 dprintf("imsm: new disk for reshape is: %i:%i "
7339 "(%p, index = %i)\n",
7340 major(u
->new_disks
[i
]), minor(u
->new_disks
[i
]),
7341 new_disk
, new_disk
->index
);
7342 if ((new_disk
== NULL
) ||
7343 ((new_disk
->index
>= 0) &&
7344 (new_disk
->index
< u
->old_raid_disks
)))
7345 goto update_reshape_exit
;
7346 new_disk
->index
= disk_count
++;
7347 /* slot to fill in autolayout
7349 new_disk
->raiddisk
= new_disk
->index
;
7350 new_disk
->disk
.status
|=
7352 new_disk
->disk
.status
&= ~SPARE_DISK
;
7355 dprintf("imsm: volume set mpb->num_raid_devs = %i\n",
7356 mpb
->num_raid_devs
);
7357 /* manage changes in volume
7359 for (dev_id
= 0; dev_id
< mpb
->num_raid_devs
; dev_id
++) {
7360 void **sp
= *space_list
;
7361 struct imsm_dev
*newdev
;
7362 struct imsm_map
*newmap
, *oldmap
;
7364 for (id
= super
->devlist
; id
; id
= id
->next
) {
7365 if (id
->index
== dev_id
)
7374 /* Copy the dev, but not (all of) the map */
7375 memcpy(newdev
, id
->dev
, sizeof(*newdev
));
7376 oldmap
= get_imsm_map(id
->dev
, MAP_0
);
7377 newmap
= get_imsm_map(newdev
, MAP_0
);
7378 /* Copy the current map */
7379 memcpy(newmap
, oldmap
, sizeof_imsm_map(oldmap
));
7380 /* update one device only
7382 if (devices_to_reshape
) {
7383 dprintf("imsm: modifying subdev: %i\n",
7385 devices_to_reshape
--;
7386 newdev
->vol
.migr_state
= 1;
7387 newdev
->vol
.curr_migr_unit
= 0;
7388 set_migr_type(newdev
, MIGR_GEN_MIGR
);
7389 newmap
->num_members
= u
->new_raid_disks
;
7390 for (i
= 0; i
< delta_disks
; i
++) {
7391 set_imsm_ord_tbl_ent(newmap
,
7392 u
->old_raid_disks
+ i
,
7393 u
->old_raid_disks
+ i
);
7395 /* New map is correct, now need to save old map
7397 newmap
= get_imsm_map(newdev
, MAP_1
);
7398 memcpy(newmap
, oldmap
, sizeof_imsm_map(oldmap
));
7400 imsm_set_array_size(newdev
);
7403 sp
= (void **)id
->dev
;
7408 /* Clear migration record */
7409 memset(super
->migr_rec
, 0, sizeof(struct migr_record
));
7412 *space_list
= tofree
;
7415 update_reshape_exit
:
7420 static int apply_takeover_update(struct imsm_update_takeover
*u
,
7421 struct intel_super
*super
,
7424 struct imsm_dev
*dev
= NULL
;
7425 struct intel_dev
*dv
;
7426 struct imsm_dev
*dev_new
;
7427 struct imsm_map
*map
;
7431 for (dv
= super
->devlist
; dv
; dv
= dv
->next
)
7432 if (dv
->index
== (unsigned int)u
->subarray
) {
7440 map
= get_imsm_map(dev
, MAP_0
);
7442 if (u
->direction
== R10_TO_R0
) {
7443 /* Number of failed disks must be half of initial disk number */
7444 if (imsm_count_failed(super
, dev
, MAP_0
) !=
7445 (map
->num_members
/ 2))
7448 /* iterate through devices to mark removed disks as spare */
7449 for (dm
= super
->disks
; dm
; dm
= dm
->next
) {
7450 if (dm
->disk
.status
& FAILED_DISK
) {
7451 int idx
= dm
->index
;
7452 /* update indexes on the disk list */
7453 /* FIXME this loop-with-the-loop looks wrong, I'm not convinced
7454 the index values will end up being correct.... NB */
7455 for (du
= super
->disks
; du
; du
= du
->next
)
7456 if (du
->index
> idx
)
7458 /* mark as spare disk */
7463 map
->num_members
= map
->num_members
/ 2;
7464 map
->map_state
= IMSM_T_STATE_NORMAL
;
7465 map
->num_domains
= 1;
7466 map
->raid_level
= 0;
7467 map
->failed_disk_num
= -1;
7470 if (u
->direction
== R0_TO_R10
) {
7472 /* update slots in current disk list */
7473 for (dm
= super
->disks
; dm
; dm
= dm
->next
) {
7477 /* create new *missing* disks */
7478 for (i
= 0; i
< map
->num_members
; i
++) {
7479 space
= *space_list
;
7482 *space_list
= *space
;
7484 memcpy(du
, super
->disks
, sizeof(*du
));
7488 du
->index
= (i
* 2) + 1;
7489 sprintf((char *)du
->disk
.serial
,
7490 " MISSING_%d", du
->index
);
7491 sprintf((char *)du
->serial
,
7492 "MISSING_%d", du
->index
);
7493 du
->next
= super
->missing
;
7494 super
->missing
= du
;
7496 /* create new dev and map */
7497 space
= *space_list
;
7500 *space_list
= *space
;
7501 dev_new
= (void *)space
;
7502 memcpy(dev_new
, dev
, sizeof(*dev
));
7503 /* update new map */
7504 map
= get_imsm_map(dev_new
, MAP_0
);
7505 map
->num_members
= map
->num_members
* 2;
7506 map
->map_state
= IMSM_T_STATE_DEGRADED
;
7507 map
->num_domains
= 2;
7508 map
->raid_level
= 1;
7509 /* replace dev<->dev_new */
7512 /* update disk order table */
7513 for (du
= super
->disks
; du
; du
= du
->next
)
7515 set_imsm_ord_tbl_ent(map
, du
->index
, du
->index
);
7516 for (du
= super
->missing
; du
; du
= du
->next
)
7517 if (du
->index
>= 0) {
7518 set_imsm_ord_tbl_ent(map
, du
->index
, du
->index
);
7519 mark_missing(dv
->dev
, &du
->disk
, du
->index
);
7525 static void imsm_process_update(struct supertype
*st
,
7526 struct metadata_update
*update
)
7529 * crack open the metadata_update envelope to find the update record
7530 * update can be one of:
7531 * update_reshape_container_disks - all the arrays in the container
7532 * are being reshaped to have more devices. We need to mark
7533 * the arrays for general migration and convert selected spares
7534 * into active devices.
7535 * update_activate_spare - a spare device has replaced a failed
7536 * device in an array, update the disk_ord_tbl. If this disk is
7537 * present in all member arrays then also clear the SPARE_DISK
7539 * update_create_array
7541 * update_rename_array
7542 * update_add_remove_disk
7544 struct intel_super
*super
= st
->sb
;
7545 struct imsm_super
*mpb
;
7546 enum imsm_update_type type
= *(enum imsm_update_type
*) update
->buf
;
7548 /* update requires a larger buf but the allocation failed */
7549 if (super
->next_len
&& !super
->next_buf
) {
7550 super
->next_len
= 0;
7554 if (super
->next_buf
) {
7555 memcpy(super
->next_buf
, super
->buf
, super
->len
);
7557 super
->len
= super
->next_len
;
7558 super
->buf
= super
->next_buf
;
7560 super
->next_len
= 0;
7561 super
->next_buf
= NULL
;
7564 mpb
= super
->anchor
;
7567 case update_general_migration_checkpoint
: {
7568 struct intel_dev
*id
;
7569 struct imsm_update_general_migration_checkpoint
*u
=
7570 (void *)update
->buf
;
7572 dprintf("imsm: process_update() "
7573 "for update_general_migration_checkpoint called\n");
7575 /* find device under general migration */
7576 for (id
= super
->devlist
; id
; id
= id
->next
) {
7577 if (is_gen_migration(id
->dev
)) {
7578 id
->dev
->vol
.curr_migr_unit
=
7579 __cpu_to_le32(u
->curr_migr_unit
);
7580 super
->updates_pending
++;
7585 case update_takeover
: {
7586 struct imsm_update_takeover
*u
= (void *)update
->buf
;
7587 if (apply_takeover_update(u
, super
, &update
->space_list
)) {
7588 imsm_update_version_info(super
);
7589 super
->updates_pending
++;
7594 case update_reshape_container_disks
: {
7595 struct imsm_update_reshape
*u
= (void *)update
->buf
;
7596 if (apply_reshape_container_disks_update(
7597 u
, super
, &update
->space_list
))
7598 super
->updates_pending
++;
7601 case update_reshape_migration
: {
7602 struct imsm_update_reshape_migration
*u
= (void *)update
->buf
;
7603 if (apply_reshape_migration_update(
7604 u
, super
, &update
->space_list
))
7605 super
->updates_pending
++;
7608 case update_activate_spare
: {
7609 struct imsm_update_activate_spare
*u
= (void *) update
->buf
;
7610 if (apply_update_activate_spare(u
, super
, st
->arrays
))
7611 super
->updates_pending
++;
7614 case update_create_array
: {
7615 /* someone wants to create a new array, we need to be aware of
7616 * a few races/collisions:
7617 * 1/ 'Create' called by two separate instances of mdadm
7618 * 2/ 'Create' versus 'activate_spare': mdadm has chosen
7619 * devices that have since been assimilated via
7621 * In the event this update can not be carried out mdadm will
7622 * (FIX ME) notice that its update did not take hold.
7624 struct imsm_update_create_array
*u
= (void *) update
->buf
;
7625 struct intel_dev
*dv
;
7626 struct imsm_dev
*dev
;
7627 struct imsm_map
*map
, *new_map
;
7628 unsigned long long start
, end
;
7629 unsigned long long new_start
, new_end
;
7631 struct disk_info
*inf
;
7634 /* handle racing creates: first come first serve */
7635 if (u
->dev_idx
< mpb
->num_raid_devs
) {
7636 dprintf("%s: subarray %d already defined\n",
7637 __func__
, u
->dev_idx
);
7641 /* check update is next in sequence */
7642 if (u
->dev_idx
!= mpb
->num_raid_devs
) {
7643 dprintf("%s: can not create array %d expected index %d\n",
7644 __func__
, u
->dev_idx
, mpb
->num_raid_devs
);
7648 new_map
= get_imsm_map(&u
->dev
, MAP_0
);
7649 new_start
= __le32_to_cpu(new_map
->pba_of_lba0
);
7650 new_end
= new_start
+ __le32_to_cpu(new_map
->blocks_per_member
);
7651 inf
= get_disk_info(u
);
7653 /* handle activate_spare versus create race:
7654 * check to make sure that overlapping arrays do not include
7657 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
7658 dev
= get_imsm_dev(super
, i
);
7659 map
= get_imsm_map(dev
, MAP_0
);
7660 start
= __le32_to_cpu(map
->pba_of_lba0
);
7661 end
= start
+ __le32_to_cpu(map
->blocks_per_member
);
7662 if ((new_start
>= start
&& new_start
<= end
) ||
7663 (start
>= new_start
&& start
<= new_end
))
7668 if (disks_overlap(super
, i
, u
)) {
7669 dprintf("%s: arrays overlap\n", __func__
);
7674 /* check that prepare update was successful */
7675 if (!update
->space
) {
7676 dprintf("%s: prepare update failed\n", __func__
);
7680 /* check that all disks are still active before committing
7681 * changes. FIXME: could we instead handle this by creating a
7682 * degraded array? That's probably not what the user expects,
7683 * so better to drop this update on the floor.
7685 for (i
= 0; i
< new_map
->num_members
; i
++) {
7686 dl
= serial_to_dl(inf
[i
].serial
, super
);
7688 dprintf("%s: disk disappeared\n", __func__
);
7693 super
->updates_pending
++;
7695 /* convert spares to members and fixup ord_tbl */
7696 for (i
= 0; i
< new_map
->num_members
; i
++) {
7697 dl
= serial_to_dl(inf
[i
].serial
, super
);
7698 if (dl
->index
== -1) {
7699 dl
->index
= mpb
->num_disks
;
7701 dl
->disk
.status
|= CONFIGURED_DISK
;
7702 dl
->disk
.status
&= ~SPARE_DISK
;
7704 set_imsm_ord_tbl_ent(new_map
, i
, dl
->index
);
7709 update
->space
= NULL
;
7710 imsm_copy_dev(dev
, &u
->dev
);
7711 dv
->index
= u
->dev_idx
;
7712 dv
->next
= super
->devlist
;
7713 super
->devlist
= dv
;
7714 mpb
->num_raid_devs
++;
7716 imsm_update_version_info(super
);
7719 /* mdmon knows how to release update->space, but not
7720 * ((struct intel_dev *) update->space)->dev
7722 if (update
->space
) {
7728 case update_kill_array
: {
7729 struct imsm_update_kill_array
*u
= (void *) update
->buf
;
7730 int victim
= u
->dev_idx
;
7731 struct active_array
*a
;
7732 struct intel_dev
**dp
;
7733 struct imsm_dev
*dev
;
7735 /* sanity check that we are not affecting the uuid of
7736 * active arrays, or deleting an active array
7738 * FIXME when immutable ids are available, but note that
7739 * we'll also need to fixup the invalidated/active
7740 * subarray indexes in mdstat
7742 for (a
= st
->arrays
; a
; a
= a
->next
)
7743 if (a
->info
.container_member
>= victim
)
7745 /* by definition if mdmon is running at least one array
7746 * is active in the container, so checking
7747 * mpb->num_raid_devs is just extra paranoia
7749 dev
= get_imsm_dev(super
, victim
);
7750 if (a
|| !dev
|| mpb
->num_raid_devs
== 1) {
7751 dprintf("failed to delete subarray-%d\n", victim
);
7755 for (dp
= &super
->devlist
; *dp
;)
7756 if ((*dp
)->index
== (unsigned)super
->current_vol
) {
7759 if ((*dp
)->index
> (unsigned)victim
)
7763 mpb
->num_raid_devs
--;
7764 super
->updates_pending
++;
7767 case update_rename_array
: {
7768 struct imsm_update_rename_array
*u
= (void *) update
->buf
;
7769 char name
[MAX_RAID_SERIAL_LEN
+1];
7770 int target
= u
->dev_idx
;
7771 struct active_array
*a
;
7772 struct imsm_dev
*dev
;
7774 /* sanity check that we are not affecting the uuid of
7777 snprintf(name
, MAX_RAID_SERIAL_LEN
, "%s", (char *) u
->name
);
7778 name
[MAX_RAID_SERIAL_LEN
] = '\0';
7779 for (a
= st
->arrays
; a
; a
= a
->next
)
7780 if (a
->info
.container_member
== target
)
7782 dev
= get_imsm_dev(super
, u
->dev_idx
);
7783 if (a
|| !dev
|| !check_name(super
, name
, 1)) {
7784 dprintf("failed to rename subarray-%d\n", target
);
7788 snprintf((char *) dev
->volume
, MAX_RAID_SERIAL_LEN
, "%s", name
);
7789 super
->updates_pending
++;
7792 case update_add_remove_disk
: {
7793 /* we may be able to repair some arrays if disks are
7794 * being added, check teh status of add_remove_disk
7795 * if discs has been added.
7797 if (add_remove_disk_update(super
)) {
7798 struct active_array
*a
;
7800 super
->updates_pending
++;
7801 for (a
= st
->arrays
; a
; a
= a
->next
)
7802 a
->check_degraded
= 1;
7807 fprintf(stderr
, "error: unsuported process update type:"
7808 "(type: %d)\n", type
);
7812 static struct mdinfo
*get_spares_for_grow(struct supertype
*st
);
7814 static void imsm_prepare_update(struct supertype
*st
,
7815 struct metadata_update
*update
)
7818 * Allocate space to hold new disk entries, raid-device entries or a new
7819 * mpb if necessary. The manager synchronously waits for updates to
7820 * complete in the monitor, so new mpb buffers allocated here can be
7821 * integrated by the monitor thread without worrying about live pointers
7822 * in the manager thread.
7824 enum imsm_update_type type
= *(enum imsm_update_type
*) update
->buf
;
7825 struct intel_super
*super
= st
->sb
;
7826 struct imsm_super
*mpb
= super
->anchor
;
7831 case update_general_migration_checkpoint
:
7832 dprintf("imsm: prepare_update() "
7833 "for update_general_migration_checkpoint called\n");
7835 case update_takeover
: {
7836 struct imsm_update_takeover
*u
= (void *)update
->buf
;
7837 if (u
->direction
== R0_TO_R10
) {
7838 void **tail
= (void **)&update
->space_list
;
7839 struct imsm_dev
*dev
= get_imsm_dev(super
, u
->subarray
);
7840 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
7841 int num_members
= map
->num_members
;
7845 /* allocate memory for added disks */
7846 for (i
= 0; i
< num_members
; i
++) {
7847 size
= sizeof(struct dl
);
7848 space
= malloc(size
);
7857 /* allocate memory for new device */
7858 size
= sizeof_imsm_dev(super
->devlist
->dev
, 0) +
7859 (num_members
* sizeof(__u32
));
7860 space
= malloc(size
);
7869 len
= disks_to_mpb_size(num_members
* 2);
7871 /* if allocation didn't success, free buffer */
7872 while (update
->space_list
) {
7873 void **sp
= update
->space_list
;
7874 update
->space_list
= *sp
;
7882 case update_reshape_container_disks
: {
7883 /* Every raid device in the container is about to
7884 * gain some more devices, and we will enter a
7886 * So each 'imsm_map' will be bigger, and the imsm_vol
7887 * will now hold 2 of them.
7888 * Thus we need new 'struct imsm_dev' allocations sized
7889 * as sizeof_imsm_dev but with more devices in both maps.
7891 struct imsm_update_reshape
*u
= (void *)update
->buf
;
7892 struct intel_dev
*dl
;
7893 void **space_tail
= (void**)&update
->space_list
;
7895 dprintf("imsm: imsm_prepare_update() for update_reshape\n");
7897 for (dl
= super
->devlist
; dl
; dl
= dl
->next
) {
7898 int size
= sizeof_imsm_dev(dl
->dev
, 1);
7900 if (u
->new_raid_disks
> u
->old_raid_disks
)
7901 size
+= sizeof(__u32
)*2*
7902 (u
->new_raid_disks
- u
->old_raid_disks
);
7911 len
= disks_to_mpb_size(u
->new_raid_disks
);
7912 dprintf("New anchor length is %llu\n", (unsigned long long)len
);
7915 case update_reshape_migration
: {
7916 /* for migration level 0->5 we need to add disks
7917 * so the same as for container operation we will copy
7918 * device to the bigger location.
7919 * in memory prepared device and new disk area are prepared
7920 * for usage in process update
7922 struct imsm_update_reshape_migration
*u
= (void *)update
->buf
;
7923 struct intel_dev
*id
;
7924 void **space_tail
= (void **)&update
->space_list
;
7927 int current_level
= -1;
7929 dprintf("imsm: imsm_prepare_update() for update_reshape\n");
7931 /* add space for bigger array in update
7933 for (id
= super
->devlist
; id
; id
= id
->next
) {
7934 if (id
->index
== (unsigned)u
->subdev
) {
7935 size
= sizeof_imsm_dev(id
->dev
, 1);
7936 if (u
->new_raid_disks
> u
->old_raid_disks
)
7937 size
+= sizeof(__u32
)*2*
7938 (u
->new_raid_disks
- u
->old_raid_disks
);
7948 if (update
->space_list
== NULL
)
7951 /* add space for disk in update
7953 size
= sizeof(struct dl
);
7956 free(update
->space_list
);
7957 update
->space_list
= NULL
;
7964 /* add spare device to update
7966 for (id
= super
->devlist
; id
; id
= id
->next
)
7967 if (id
->index
== (unsigned)u
->subdev
) {
7968 struct imsm_dev
*dev
;
7969 struct imsm_map
*map
;
7971 dev
= get_imsm_dev(super
, u
->subdev
);
7972 map
= get_imsm_map(dev
, MAP_0
);
7973 current_level
= map
->raid_level
;
7976 if ((u
->new_level
== 5) && (u
->new_level
!= current_level
)) {
7977 struct mdinfo
*spares
;
7979 spares
= get_spares_for_grow(st
);
7987 makedev(dev
->disk
.major
,
7989 dl
= get_disk_super(super
,
7992 dl
->index
= u
->old_raid_disks
;
7998 len
= disks_to_mpb_size(u
->new_raid_disks
);
7999 dprintf("New anchor length is %llu\n", (unsigned long long)len
);
8002 case update_create_array
: {
8003 struct imsm_update_create_array
*u
= (void *) update
->buf
;
8004 struct intel_dev
*dv
;
8005 struct imsm_dev
*dev
= &u
->dev
;
8006 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
8008 struct disk_info
*inf
;
8012 inf
= get_disk_info(u
);
8013 len
= sizeof_imsm_dev(dev
, 1);
8014 /* allocate a new super->devlist entry */
8015 dv
= malloc(sizeof(*dv
));
8017 dv
->dev
= malloc(len
);
8022 update
->space
= NULL
;
8026 /* count how many spares will be converted to members */
8027 for (i
= 0; i
< map
->num_members
; i
++) {
8028 dl
= serial_to_dl(inf
[i
].serial
, super
);
8030 /* hmm maybe it failed?, nothing we can do about
8035 if (count_memberships(dl
, super
) == 0)
8038 len
+= activate
* sizeof(struct imsm_disk
);
8045 /* check if we need a larger metadata buffer */
8046 if (super
->next_buf
)
8047 buf_len
= super
->next_len
;
8049 buf_len
= super
->len
;
8051 if (__le32_to_cpu(mpb
->mpb_size
) + len
> buf_len
) {
8052 /* ok we need a larger buf than what is currently allocated
8053 * if this allocation fails process_update will notice that
8054 * ->next_len is set and ->next_buf is NULL
8056 buf_len
= ROUND_UP(__le32_to_cpu(mpb
->mpb_size
) + len
, 512);
8057 if (super
->next_buf
)
8058 free(super
->next_buf
);
8060 super
->next_len
= buf_len
;
8061 if (posix_memalign(&super
->next_buf
, 512, buf_len
) == 0)
8062 memset(super
->next_buf
, 0, buf_len
);
8064 super
->next_buf
= NULL
;
8068 /* must be called while manager is quiesced */
8069 static void imsm_delete(struct intel_super
*super
, struct dl
**dlp
, unsigned index
)
8071 struct imsm_super
*mpb
= super
->anchor
;
8073 struct imsm_dev
*dev
;
8074 struct imsm_map
*map
;
8075 int i
, j
, num_members
;
8078 dprintf("%s: deleting device[%d] from imsm_super\n",
8081 /* shift all indexes down one */
8082 for (iter
= super
->disks
; iter
; iter
= iter
->next
)
8083 if (iter
->index
> (int)index
)
8085 for (iter
= super
->missing
; iter
; iter
= iter
->next
)
8086 if (iter
->index
> (int)index
)
8089 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
8090 dev
= get_imsm_dev(super
, i
);
8091 map
= get_imsm_map(dev
, MAP_0
);
8092 num_members
= map
->num_members
;
8093 for (j
= 0; j
< num_members
; j
++) {
8094 /* update ord entries being careful not to propagate
8095 * ord-flags to the first map
8097 ord
= get_imsm_ord_tbl_ent(dev
, j
, MAP_X
);
8099 if (ord_to_idx(ord
) <= index
)
8102 map
= get_imsm_map(dev
, MAP_0
);
8103 set_imsm_ord_tbl_ent(map
, j
, ord_to_idx(ord
- 1));
8104 map
= get_imsm_map(dev
, MAP_1
);
8106 set_imsm_ord_tbl_ent(map
, j
, ord
- 1);
8111 super
->updates_pending
++;
8113 struct dl
*dl
= *dlp
;
8115 *dlp
= (*dlp
)->next
;
8116 __free_imsm_disk(dl
);
8119 #endif /* MDASSEMBLE */
8121 static void close_targets(int *targets
, int new_disks
)
8128 for (i
= 0; i
< new_disks
; i
++) {
8129 if (targets
[i
] >= 0) {
8136 static int imsm_get_allowed_degradation(int level
, int raid_disks
,
8137 struct intel_super
*super
,
8138 struct imsm_dev
*dev
)
8143 struct imsm_map
*map
;
8146 ret_val
= raid_disks
/2;
8147 /* check map if all disks pairs not failed
8150 map
= get_imsm_map(dev
, MAP_0
);
8151 for (i
= 0; i
< ret_val
; i
++) {
8152 int degradation
= 0;
8153 if (get_imsm_disk(super
, i
) == NULL
)
8155 if (get_imsm_disk(super
, i
+ 1) == NULL
)
8157 if (degradation
== 2)
8160 map
= get_imsm_map(dev
, MAP_1
);
8161 /* if there is no second map
8162 * result can be returned
8166 /* check degradation in second map
8168 for (i
= 0; i
< ret_val
; i
++) {
8169 int degradation
= 0;
8170 if (get_imsm_disk(super
, i
) == NULL
)
8172 if (get_imsm_disk(super
, i
+ 1) == NULL
)
8174 if (degradation
== 2)
8189 /*******************************************************************************
8190 * Function: open_backup_targets
8191 * Description: Function opens file descriptors for all devices given in
8194 * info : general array info
8195 * raid_disks : number of disks
8196 * raid_fds : table of device's file descriptors
8197 * super : intel super for raid10 degradation check
8198 * dev : intel device for raid10 degradation check
8202 ******************************************************************************/
8203 int open_backup_targets(struct mdinfo
*info
, int raid_disks
, int *raid_fds
,
8204 struct intel_super
*super
, struct imsm_dev
*dev
)
8210 for (i
= 0; i
< raid_disks
; i
++)
8213 for (sd
= info
->devs
; sd
; sd
= sd
->next
) {
8216 if (sd
->disk
.state
& (1<<MD_DISK_FAULTY
)) {
8217 dprintf("disk is faulty!!\n");
8221 if ((sd
->disk
.raid_disk
>= raid_disks
) ||
8222 (sd
->disk
.raid_disk
< 0))
8225 dn
= map_dev(sd
->disk
.major
,
8227 raid_fds
[sd
->disk
.raid_disk
] = dev_open(dn
, O_RDWR
);
8228 if (raid_fds
[sd
->disk
.raid_disk
] < 0) {
8229 fprintf(stderr
, "cannot open component\n");
8234 /* check if maximum array degradation level is not exceeded
8236 if ((raid_disks
- opened
) >
8237 imsm_get_allowed_degradation(info
->new_level
,
8240 fprintf(stderr
, "Not enough disks can be opened.\n");
8241 close_targets(raid_fds
, raid_disks
);
8248 /*******************************************************************************
8249 * Function: init_migr_record_imsm
8250 * Description: Function inits imsm migration record
8252 * super : imsm internal array info
8253 * dev : device under migration
8254 * info : general array info to find the smallest device
8257 ******************************************************************************/
8258 void init_migr_record_imsm(struct supertype
*st
, struct imsm_dev
*dev
,
8259 struct mdinfo
*info
)
8261 struct intel_super
*super
= st
->sb
;
8262 struct migr_record
*migr_rec
= super
->migr_rec
;
8264 unsigned long long dsize
, dev_sectors
;
8265 long long unsigned min_dev_sectors
= -1LLU;
8269 struct imsm_map
*map_dest
= get_imsm_map(dev
, MAP_0
);
8270 struct imsm_map
*map_src
= get_imsm_map(dev
, MAP_1
);
8271 unsigned long long num_migr_units
;
8272 unsigned long long array_blocks
;
8274 memset(migr_rec
, 0, sizeof(struct migr_record
));
8275 migr_rec
->family_num
= __cpu_to_le32(super
->anchor
->family_num
);
8277 /* only ascending reshape supported now */
8278 migr_rec
->ascending_migr
= __cpu_to_le32(1);
8280 migr_rec
->dest_depth_per_unit
= GEN_MIGR_AREA_SIZE
/
8281 max(map_dest
->blocks_per_strip
, map_src
->blocks_per_strip
);
8282 migr_rec
->dest_depth_per_unit
*= map_dest
->blocks_per_strip
;
8283 new_data_disks
= imsm_num_data_members(dev
, MAP_0
);
8284 migr_rec
->blocks_per_unit
=
8285 __cpu_to_le32(migr_rec
->dest_depth_per_unit
* new_data_disks
);
8286 migr_rec
->dest_depth_per_unit
=
8287 __cpu_to_le32(migr_rec
->dest_depth_per_unit
);
8288 array_blocks
= info
->component_size
* new_data_disks
;
8290 array_blocks
/ __le32_to_cpu(migr_rec
->blocks_per_unit
);
8292 if (array_blocks
% __le32_to_cpu(migr_rec
->blocks_per_unit
))
8294 migr_rec
->num_migr_units
= __cpu_to_le32(num_migr_units
);
8296 migr_rec
->post_migr_vol_cap
= dev
->size_low
;
8297 migr_rec
->post_migr_vol_cap_hi
= dev
->size_high
;
8300 /* Find the smallest dev */
8301 for (sd
= info
->devs
; sd
; sd
= sd
->next
) {
8302 sprintf(nm
, "%d:%d", sd
->disk
.major
, sd
->disk
.minor
);
8303 fd
= dev_open(nm
, O_RDONLY
);
8306 get_dev_size(fd
, NULL
, &dsize
);
8307 dev_sectors
= dsize
/ 512;
8308 if (dev_sectors
< min_dev_sectors
)
8309 min_dev_sectors
= dev_sectors
;
8312 migr_rec
->ckpt_area_pba
= __cpu_to_le32(min_dev_sectors
-
8313 RAID_DISK_RESERVED_BLOCKS_IMSM_HI
);
8315 write_imsm_migr_rec(st
);
8320 /*******************************************************************************
8321 * Function: save_backup_imsm
8322 * Description: Function saves critical data stripes to Migration Copy Area
8323 * and updates the current migration unit status.
8324 * Use restore_stripes() to form a destination stripe,
8325 * and to write it to the Copy Area.
8327 * st : supertype information
8328 * dev : imsm device that backup is saved for
8329 * info : general array info
8330 * buf : input buffer
8331 * length : length of data to backup (blocks_per_unit)
8335 ******************************************************************************/
8336 int save_backup_imsm(struct supertype
*st
,
8337 struct imsm_dev
*dev
,
8338 struct mdinfo
*info
,
8343 struct intel_super
*super
= st
->sb
;
8344 unsigned long long *target_offsets
= NULL
;
8345 int *targets
= NULL
;
8347 struct imsm_map
*map_dest
= get_imsm_map(dev
, MAP_0
);
8348 int new_disks
= map_dest
->num_members
;
8349 int dest_layout
= 0;
8351 unsigned long long start
;
8352 int data_disks
= imsm_num_data_members(dev
, MAP_0
);
8354 targets
= malloc(new_disks
* sizeof(int));
8358 for (i
= 0; i
< new_disks
; i
++)
8361 target_offsets
= malloc(new_disks
* sizeof(unsigned long long));
8362 if (!target_offsets
)
8365 start
= info
->reshape_progress
* 512;
8366 for (i
= 0; i
< new_disks
; i
++) {
8367 target_offsets
[i
] = (unsigned long long)
8368 __le32_to_cpu(super
->migr_rec
->ckpt_area_pba
) * 512;
8369 /* move back copy area adderss, it will be moved forward
8370 * in restore_stripes() using start input variable
8372 target_offsets
[i
] -= start
/data_disks
;
8375 if (open_backup_targets(info
, new_disks
, targets
,
8379 dest_layout
= imsm_level_to_layout(map_dest
->raid_level
);
8380 dest_chunk
= __le16_to_cpu(map_dest
->blocks_per_strip
) * 512;
8382 if (restore_stripes(targets
, /* list of dest devices */
8383 target_offsets
, /* migration record offsets */
8386 map_dest
->raid_level
,
8388 -1, /* source backup file descriptor */
8389 0, /* input buf offset
8390 * always 0 buf is already offseted */
8394 fprintf(stderr
, Name
": Error restoring stripes\n");
8402 close_targets(targets
, new_disks
);
8405 free(target_offsets
);
8410 /*******************************************************************************
8411 * Function: save_checkpoint_imsm
8412 * Description: Function called for current unit status update
8413 * in the migration record. It writes it to disk.
8415 * super : imsm internal array info
8416 * info : general array info
8420 * 2: failure, means no valid migration record
8421 * / no general migration in progress /
8422 ******************************************************************************/
8423 int save_checkpoint_imsm(struct supertype
*st
, struct mdinfo
*info
, int state
)
8425 struct intel_super
*super
= st
->sb
;
8426 unsigned long long blocks_per_unit
;
8427 unsigned long long curr_migr_unit
;
8429 if (load_imsm_migr_rec(super
, info
) != 0) {
8430 dprintf("imsm: ERROR: Cannot read migration record "
8431 "for checkpoint save.\n");
8435 blocks_per_unit
= __le32_to_cpu(super
->migr_rec
->blocks_per_unit
);
8436 if (blocks_per_unit
== 0) {
8437 dprintf("imsm: no migration in progress.\n");
8440 curr_migr_unit
= info
->reshape_progress
/ blocks_per_unit
;
8441 /* check if array is alligned to copy area
8442 * if it is not alligned, add one to current migration unit value
8443 * this can happend on array reshape finish only
8445 if (info
->reshape_progress
% blocks_per_unit
)
8448 super
->migr_rec
->curr_migr_unit
=
8449 __cpu_to_le32(curr_migr_unit
);
8450 super
->migr_rec
->rec_status
= __cpu_to_le32(state
);
8451 super
->migr_rec
->dest_1st_member_lba
=
8452 __cpu_to_le32(curr_migr_unit
*
8453 __le32_to_cpu(super
->migr_rec
->dest_depth_per_unit
));
8454 if (write_imsm_migr_rec(st
) < 0) {
8455 dprintf("imsm: Cannot write migration record "
8456 "outside backup area\n");
8463 /*******************************************************************************
8464 * Function: recover_backup_imsm
8465 * Description: Function recovers critical data from the Migration Copy Area
8466 * while assembling an array.
8468 * super : imsm internal array info
8469 * info : general array info
8471 * 0 : success (or there is no data to recover)
8473 ******************************************************************************/
8474 int recover_backup_imsm(struct supertype
*st
, struct mdinfo
*info
)
8476 struct intel_super
*super
= st
->sb
;
8477 struct migr_record
*migr_rec
= super
->migr_rec
;
8478 struct imsm_map
*map_dest
= NULL
;
8479 struct intel_dev
*id
= NULL
;
8480 unsigned long long read_offset
;
8481 unsigned long long write_offset
;
8483 int *targets
= NULL
;
8484 int new_disks
, i
, err
;
8487 unsigned long curr_migr_unit
= __le32_to_cpu(migr_rec
->curr_migr_unit
);
8488 unsigned long num_migr_units
= __le32_to_cpu(migr_rec
->num_migr_units
);
8490 int skipped_disks
= 0;
8492 err
= sysfs_get_str(info
, NULL
, "array_state", (char *)buffer
, 20);
8496 /* recover data only during assemblation */
8497 if (strncmp(buffer
, "inactive", 8) != 0)
8499 /* no data to recover */
8500 if (__le32_to_cpu(migr_rec
->rec_status
) == UNIT_SRC_NORMAL
)
8502 if (curr_migr_unit
>= num_migr_units
)
8505 /* find device during reshape */
8506 for (id
= super
->devlist
; id
; id
= id
->next
)
8507 if (is_gen_migration(id
->dev
))
8512 map_dest
= get_imsm_map(id
->dev
, MAP_0
);
8513 new_disks
= map_dest
->num_members
;
8515 read_offset
= (unsigned long long)
8516 __le32_to_cpu(migr_rec
->ckpt_area_pba
) * 512;
8518 write_offset
= ((unsigned long long)
8519 __le32_to_cpu(migr_rec
->dest_1st_member_lba
) +
8520 __le32_to_cpu(map_dest
->pba_of_lba0
)) * 512;
8522 unit_len
= __le32_to_cpu(migr_rec
->dest_depth_per_unit
) * 512;
8523 if (posix_memalign((void **)&buf
, 512, unit_len
) != 0)
8525 targets
= malloc(new_disks
* sizeof(int));
8529 if (open_backup_targets(info
, new_disks
, targets
, super
, id
->dev
)) {
8531 Name
": Cannot open some devices belonging to array.\n");
8535 for (i
= 0; i
< new_disks
; i
++) {
8536 if (targets
[i
] < 0) {
8540 if (lseek64(targets
[i
], read_offset
, SEEK_SET
) < 0) {
8542 Name
": Cannot seek to block: %s\n",
8547 if ((unsigned)read(targets
[i
], buf
, unit_len
) != unit_len
) {
8549 Name
": Cannot read copy area block: %s\n",
8554 if (lseek64(targets
[i
], write_offset
, SEEK_SET
) < 0) {
8556 Name
": Cannot seek to block: %s\n",
8561 if ((unsigned)write(targets
[i
], buf
, unit_len
) != unit_len
) {
8563 Name
": Cannot restore block: %s\n",
8570 if (skipped_disks
> imsm_get_allowed_degradation(info
->new_level
,
8575 Name
": Cannot restore data from backup."
8576 " Too many failed disks\n");
8580 if (save_checkpoint_imsm(st
, info
, UNIT_SRC_NORMAL
)) {
8581 /* ignore error == 2, this can mean end of reshape here
8583 dprintf("imsm: Cannot write checkpoint to "
8584 "migration record (UNIT_SRC_NORMAL) during restart\n");
8590 for (i
= 0; i
< new_disks
; i
++)
8599 static char disk_by_path
[] = "/dev/disk/by-path/";
8601 static const char *imsm_get_disk_controller_domain(const char *path
)
8603 char disk_path
[PATH_MAX
];
8607 strncpy(disk_path
, disk_by_path
, PATH_MAX
- 1);
8608 strncat(disk_path
, path
, PATH_MAX
- strlen(disk_path
) - 1);
8609 if (stat(disk_path
, &st
) == 0) {
8610 struct sys_dev
* hba
;
8613 path
= devt_to_devpath(st
.st_rdev
);
8616 hba
= find_disk_attached_hba(-1, path
);
8617 if (hba
&& hba
->type
== SYS_DEV_SAS
)
8619 else if (hba
&& hba
->type
== SYS_DEV_SATA
)
8623 dprintf("path: %s hba: %s attached: %s\n",
8624 path
, (hba
) ? hba
->path
: "NULL", drv
);
8632 static int imsm_find_array_minor_by_subdev(int subdev
, int container
, int *minor
)
8634 char subdev_name
[20];
8635 struct mdstat_ent
*mdstat
;
8637 sprintf(subdev_name
, "%d", subdev
);
8638 mdstat
= mdstat_by_subdev(subdev_name
, container
);
8642 *minor
= mdstat
->devnum
;
8643 free_mdstat(mdstat
);
8647 static int imsm_reshape_is_allowed_on_container(struct supertype
*st
,
8648 struct geo_params
*geo
,
8649 int *old_raid_disks
)
8651 /* currently we only support increasing the number of devices
8652 * for a container. This increases the number of device for each
8653 * member array. They must all be RAID0 or RAID5.
8656 struct mdinfo
*info
, *member
;
8657 int devices_that_can_grow
= 0;
8659 dprintf("imsm: imsm_reshape_is_allowed_on_container(ENTER): "
8660 "st->devnum = (%i)\n",
8663 if (geo
->size
!= -1 ||
8664 geo
->level
!= UnSet
||
8665 geo
->layout
!= UnSet
||
8666 geo
->chunksize
!= 0 ||
8667 geo
->raid_disks
== UnSet
) {
8668 dprintf("imsm: Container operation is allowed for "
8669 "raid disks number change only.\n");
8673 info
= container_content_imsm(st
, NULL
);
8674 for (member
= info
; member
; member
= member
->next
) {
8678 dprintf("imsm: checking device_num: %i\n",
8679 member
->container_member
);
8681 if (geo
->raid_disks
<= member
->array
.raid_disks
) {
8682 /* we work on container for Online Capacity Expansion
8683 * only so raid_disks has to grow
8685 dprintf("imsm: for container operation raid disks "
8686 "increase is required\n");
8690 if ((info
->array
.level
!= 0) &&
8691 (info
->array
.level
!= 5)) {
8692 /* we cannot use this container with other raid level
8694 dprintf("imsm: for container operation wrong"
8695 " raid level (%i) detected\n",
8699 /* check for platform support
8700 * for this raid level configuration
8702 struct intel_super
*super
= st
->sb
;
8703 if (!is_raid_level_supported(super
->orom
,
8704 member
->array
.level
,
8706 dprintf("platform does not support raid%d with"
8710 geo
->raid_disks
> 1 ? "s" : "");
8713 /* check if component size is aligned to chunk size
8715 if (info
->component_size
%
8716 (info
->array
.chunk_size
/512)) {
8717 dprintf("Component size is not aligned to "
8723 if (*old_raid_disks
&&
8724 info
->array
.raid_disks
!= *old_raid_disks
)
8726 *old_raid_disks
= info
->array
.raid_disks
;
8728 /* All raid5 and raid0 volumes in container
8729 * have to be ready for Online Capacity Expansion
8730 * so they need to be assembled. We have already
8731 * checked that no recovery etc is happening.
8733 result
= imsm_find_array_minor_by_subdev(member
->container_member
,
8737 dprintf("imsm: cannot find array\n");
8740 devices_that_can_grow
++;
8743 if (!member
&& devices_that_can_grow
)
8747 dprintf("\tContainer operation allowed\n");
8749 dprintf("\tError: %i\n", ret_val
);
8754 /* Function: get_spares_for_grow
8755 * Description: Allocates memory and creates list of spare devices
8756 * avaliable in container. Checks if spare drive size is acceptable.
8757 * Parameters: Pointer to the supertype structure
8758 * Returns: Pointer to the list of spare devices (mdinfo structure) on success,
8761 static struct mdinfo
*get_spares_for_grow(struct supertype
*st
)
8763 unsigned long long min_size
= min_acceptable_spare_size_imsm(st
);
8764 return container_choose_spares(st
, min_size
, NULL
, NULL
, NULL
, 0);
8767 /******************************************************************************
8768 * function: imsm_create_metadata_update_for_reshape
8769 * Function creates update for whole IMSM container.
8771 ******************************************************************************/
8772 static int imsm_create_metadata_update_for_reshape(
8773 struct supertype
*st
,
8774 struct geo_params
*geo
,
8776 struct imsm_update_reshape
**updatep
)
8778 struct intel_super
*super
= st
->sb
;
8779 struct imsm_super
*mpb
= super
->anchor
;
8780 int update_memory_size
= 0;
8781 struct imsm_update_reshape
*u
= NULL
;
8782 struct mdinfo
*spares
= NULL
;
8784 int delta_disks
= 0;
8787 dprintf("imsm_update_metadata_for_reshape(enter) raid_disks = %i\n",
8790 delta_disks
= geo
->raid_disks
- old_raid_disks
;
8792 /* size of all update data without anchor */
8793 update_memory_size
= sizeof(struct imsm_update_reshape
);
8795 /* now add space for spare disks that we need to add. */
8796 update_memory_size
+= sizeof(u
->new_disks
[0]) * (delta_disks
- 1);
8798 u
= calloc(1, update_memory_size
);
8801 "cannot get memory for imsm_update_reshape update\n");
8804 u
->type
= update_reshape_container_disks
;
8805 u
->old_raid_disks
= old_raid_disks
;
8806 u
->new_raid_disks
= geo
->raid_disks
;
8808 /* now get spare disks list
8810 spares
= get_spares_for_grow(st
);
8813 || delta_disks
> spares
->array
.spare_disks
) {
8814 fprintf(stderr
, Name
": imsm: ERROR: Cannot get spare devices "
8815 "for %s.\n", geo
->dev_name
);
8820 /* we have got spares
8821 * update disk list in imsm_disk list table in anchor
8823 dprintf("imsm: %i spares are available.\n\n",
8824 spares
->array
.spare_disks
);
8827 for (i
= 0; i
< delta_disks
; i
++) {
8832 u
->new_disks
[i
] = makedev(dev
->disk
.major
,
8834 dl
= get_disk_super(super
, dev
->disk
.major
, dev
->disk
.minor
);
8835 dl
->index
= mpb
->num_disks
;
8845 dprintf("imsm: reshape update preparation :");
8846 if (i
== delta_disks
) {
8849 return update_memory_size
;
8852 dprintf(" Error\n");
8857 /******************************************************************************
8858 * function: imsm_create_metadata_update_for_migration()
8859 * Creates update for IMSM array.
8861 ******************************************************************************/
8862 static int imsm_create_metadata_update_for_migration(
8863 struct supertype
*st
,
8864 struct geo_params
*geo
,
8865 struct imsm_update_reshape_migration
**updatep
)
8867 struct intel_super
*super
= st
->sb
;
8868 int update_memory_size
= 0;
8869 struct imsm_update_reshape_migration
*u
= NULL
;
8870 struct imsm_dev
*dev
;
8871 int previous_level
= -1;
8873 dprintf("imsm_create_metadata_update_for_migration(enter)"
8874 " New Level = %i\n", geo
->level
);
8876 /* size of all update data without anchor */
8877 update_memory_size
= sizeof(struct imsm_update_reshape_migration
);
8879 u
= calloc(1, update_memory_size
);
8881 dprintf("error: cannot get memory for "
8882 "imsm_create_metadata_update_for_migration\n");
8885 u
->type
= update_reshape_migration
;
8886 u
->subdev
= super
->current_vol
;
8887 u
->new_level
= geo
->level
;
8888 u
->new_layout
= geo
->layout
;
8889 u
->new_raid_disks
= u
->old_raid_disks
= geo
->raid_disks
;
8890 u
->new_disks
[0] = -1;
8891 u
->new_chunksize
= -1;
8893 dev
= get_imsm_dev(super
, u
->subdev
);
8895 struct imsm_map
*map
;
8897 map
= get_imsm_map(dev
, MAP_0
);
8899 int current_chunk_size
=
8900 __le16_to_cpu(map
->blocks_per_strip
) / 2;
8902 if (geo
->chunksize
!= current_chunk_size
) {
8903 u
->new_chunksize
= geo
->chunksize
/ 1024;
8905 "chunk size change from %i to %i\n",
8906 current_chunk_size
, u
->new_chunksize
);
8908 previous_level
= map
->raid_level
;
8911 if ((geo
->level
== 5) && (previous_level
== 0)) {
8912 struct mdinfo
*spares
= NULL
;
8914 u
->new_raid_disks
++;
8915 spares
= get_spares_for_grow(st
);
8916 if ((spares
== NULL
) || (spares
->array
.spare_disks
< 1)) {
8919 update_memory_size
= 0;
8920 dprintf("error: cannot get spare device "
8921 "for requested migration");
8926 dprintf("imsm: reshape update preparation : OK\n");
8929 return update_memory_size
;
8932 static void imsm_update_metadata_locally(struct supertype
*st
,
8935 struct metadata_update mu
;
8940 mu
.space_list
= NULL
;
8942 imsm_prepare_update(st
, &mu
);
8943 imsm_process_update(st
, &mu
);
8945 while (mu
.space_list
) {
8946 void **space
= mu
.space_list
;
8947 mu
.space_list
= *space
;
8952 /***************************************************************************
8953 * Function: imsm_analyze_change
8954 * Description: Function analyze change for single volume
8955 * and validate if transition is supported
8956 * Parameters: Geometry parameters, supertype structure
8957 * Returns: Operation type code on success, -1 if fail
8958 ****************************************************************************/
8959 enum imsm_reshape_type
imsm_analyze_change(struct supertype
*st
,
8960 struct geo_params
*geo
)
8969 getinfo_super_imsm_volume(st
, &info
, NULL
);
8970 if ((geo
->level
!= info
.array
.level
) &&
8971 (geo
->level
>= 0) &&
8972 (geo
->level
!= UnSet
)) {
8973 switch (info
.array
.level
) {
8975 if (geo
->level
== 5) {
8976 change
= CH_MIGRATION
;
8977 if (geo
->layout
!= ALGORITHM_LEFT_ASYMMETRIC
) {
8979 Name
" Error. Requested Layout "
8980 "not supported (left-asymmetric layout "
8981 "is supported only)!\n");
8983 goto analyse_change_exit
;
8985 layout
= geo
->layout
;
8987 devNumChange
= 1; /* parity disk added */
8988 } else if (geo
->level
== 10) {
8989 change
= CH_TAKEOVER
;
8991 devNumChange
= 2; /* two mirrors added */
8992 layout
= 0x102; /* imsm supported layout */
8997 if (geo
->level
== 0) {
8998 change
= CH_TAKEOVER
;
9000 devNumChange
= -(geo
->raid_disks
/2);
9001 layout
= 0; /* imsm raid0 layout */
9007 Name
" Error. Level Migration from %d to %d "
9009 info
.array
.level
, geo
->level
);
9010 goto analyse_change_exit
;
9013 geo
->level
= info
.array
.level
;
9015 if ((geo
->layout
!= info
.array
.layout
)
9016 && ((geo
->layout
!= UnSet
) && (geo
->layout
!= -1))) {
9017 change
= CH_MIGRATION
;
9018 if ((info
.array
.layout
== 0)
9019 && (info
.array
.level
== 5)
9020 && (geo
->layout
== 5)) {
9021 /* reshape 5 -> 4 */
9022 } else if ((info
.array
.layout
== 5)
9023 && (info
.array
.level
== 5)
9024 && (geo
->layout
== 0)) {
9025 /* reshape 4 -> 5 */
9030 Name
" Error. Layout Migration from %d to %d "
9032 info
.array
.layout
, geo
->layout
);
9034 goto analyse_change_exit
;
9037 geo
->layout
= info
.array
.layout
;
9039 if ((geo
->chunksize
> 0) && (geo
->chunksize
!= UnSet
)
9040 && (geo
->chunksize
!= info
.array
.chunk_size
))
9041 change
= CH_MIGRATION
;
9043 geo
->chunksize
= info
.array
.chunk_size
;
9045 chunk
= geo
->chunksize
/ 1024;
9046 if (!validate_geometry_imsm(st
,
9049 geo
->raid_disks
+ devNumChange
,
9056 struct intel_super
*super
= st
->sb
;
9057 struct imsm_super
*mpb
= super
->anchor
;
9059 if (mpb
->num_raid_devs
> 1) {
9061 Name
" Error. Cannot perform operation on %s"
9062 "- for this operation it MUST be single "
9063 "array in container\n",
9069 analyse_change_exit
:
9074 int imsm_takeover(struct supertype
*st
, struct geo_params
*geo
)
9076 struct intel_super
*super
= st
->sb
;
9077 struct imsm_update_takeover
*u
;
9079 u
= malloc(sizeof(struct imsm_update_takeover
));
9083 u
->type
= update_takeover
;
9084 u
->subarray
= super
->current_vol
;
9086 /* 10->0 transition */
9087 if (geo
->level
== 0)
9088 u
->direction
= R10_TO_R0
;
9090 /* 0->10 transition */
9091 if (geo
->level
== 10)
9092 u
->direction
= R0_TO_R10
;
9094 /* update metadata locally */
9095 imsm_update_metadata_locally(st
, u
,
9096 sizeof(struct imsm_update_takeover
));
9097 /* and possibly remotely */
9098 if (st
->update_tail
)
9099 append_metadata_update(st
, u
,
9100 sizeof(struct imsm_update_takeover
));
9107 static int imsm_reshape_super(struct supertype
*st
, long long size
, int level
,
9108 int layout
, int chunksize
, int raid_disks
,
9109 int delta_disks
, char *backup
, char *dev
,
9113 struct geo_params geo
;
9115 dprintf("imsm: reshape_super called.\n");
9117 memset(&geo
, 0, sizeof(struct geo_params
));
9120 geo
.dev_id
= st
->devnum
;
9123 geo
.layout
= layout
;
9124 geo
.chunksize
= chunksize
;
9125 geo
.raid_disks
= raid_disks
;
9126 if (delta_disks
!= UnSet
)
9127 geo
.raid_disks
+= delta_disks
;
9129 dprintf("\tfor level : %i\n", geo
.level
);
9130 dprintf("\tfor raid_disks : %i\n", geo
.raid_disks
);
9132 if (experimental() == 0)
9135 if (st
->container_dev
== st
->devnum
) {
9136 /* On container level we can only increase number of devices. */
9137 dprintf("imsm: info: Container operation\n");
9138 int old_raid_disks
= 0;
9140 if (imsm_reshape_is_allowed_on_container(
9141 st
, &geo
, &old_raid_disks
)) {
9142 struct imsm_update_reshape
*u
= NULL
;
9145 len
= imsm_create_metadata_update_for_reshape(
9146 st
, &geo
, old_raid_disks
, &u
);
9149 dprintf("imsm: Cannot prepare update\n");
9150 goto exit_imsm_reshape_super
;
9154 /* update metadata locally */
9155 imsm_update_metadata_locally(st
, u
, len
);
9156 /* and possibly remotely */
9157 if (st
->update_tail
)
9158 append_metadata_update(st
, u
, len
);
9163 fprintf(stderr
, Name
": (imsm) Operation "
9164 "is not allowed on this container\n");
9167 /* On volume level we support following operations
9168 * - takeover: raid10 -> raid0; raid0 -> raid10
9169 * - chunk size migration
9170 * - migration: raid5 -> raid0; raid0 -> raid5
9172 struct intel_super
*super
= st
->sb
;
9173 struct intel_dev
*dev
= super
->devlist
;
9175 dprintf("imsm: info: Volume operation\n");
9176 /* find requested device */
9178 if (imsm_find_array_minor_by_subdev(
9179 dev
->index
, st
->container_dev
, &devnum
) == 0
9180 && devnum
== geo
.dev_id
)
9185 fprintf(stderr
, Name
" Cannot find %s (%i) subarray\n",
9186 geo
.dev_name
, geo
.dev_id
);
9187 goto exit_imsm_reshape_super
;
9189 super
->current_vol
= dev
->index
;
9190 change
= imsm_analyze_change(st
, &geo
);
9193 ret_val
= imsm_takeover(st
, &geo
);
9195 case CH_MIGRATION
: {
9196 struct imsm_update_reshape_migration
*u
= NULL
;
9198 imsm_create_metadata_update_for_migration(
9202 "Cannot prepare update\n");
9206 /* update metadata locally */
9207 imsm_update_metadata_locally(st
, u
, len
);
9208 /* and possibly remotely */
9209 if (st
->update_tail
)
9210 append_metadata_update(st
, u
, len
);
9220 exit_imsm_reshape_super
:
9221 dprintf("imsm: reshape_super Exit code = %i\n", ret_val
);
9225 /*******************************************************************************
9226 * Function: wait_for_reshape_imsm
9227 * Description: Function writes new sync_max value and waits until
9228 * reshape process reach new position
9230 * sra : general array info
9231 * ndata : number of disks in new array's layout
9234 * 1 : there is no reshape in progress,
9236 ******************************************************************************/
9237 int wait_for_reshape_imsm(struct mdinfo
*sra
, int ndata
)
9239 int fd
= sysfs_get_fd(sra
, NULL
, "reshape_position");
9240 unsigned long long completed
;
9241 /* to_complete : new sync_max position */
9242 unsigned long long to_complete
= sra
->reshape_progress
;
9243 unsigned long long position_to_set
= to_complete
/ ndata
;
9246 dprintf("imsm: wait_for_reshape_imsm() "
9247 "cannot open reshape_position\n");
9251 if (sysfs_fd_get_ll(fd
, &completed
) < 0) {
9252 dprintf("imsm: wait_for_reshape_imsm() "
9253 "cannot read reshape_position (no reshape in progres)\n");
9258 if (completed
> to_complete
) {
9259 dprintf("imsm: wait_for_reshape_imsm() "
9260 "wrong next position to set %llu (%llu)\n",
9261 to_complete
, completed
);
9265 dprintf("Position set: %llu\n", position_to_set
);
9266 if (sysfs_set_num(sra
, NULL
, "sync_max",
9267 position_to_set
) != 0) {
9268 dprintf("imsm: wait_for_reshape_imsm() "
9269 "cannot set reshape position to %llu\n",
9280 select(fd
+1, &rfds
, NULL
, NULL
, NULL
);
9281 if (sysfs_get_str(sra
, NULL
, "sync_action",
9283 strncmp(action
, "reshape", 7) != 0)
9285 if (sysfs_fd_get_ll(fd
, &completed
) < 0) {
9286 dprintf("imsm: wait_for_reshape_imsm() "
9287 "cannot read reshape_position (in loop)\n");
9291 } while (completed
< to_complete
);
9297 /*******************************************************************************
9298 * Function: check_degradation_change
9299 * Description: Check that array hasn't become failed.
9301 * info : for sysfs access
9302 * sources : source disks descriptors
9303 * degraded: previous degradation level
9306 ******************************************************************************/
9307 int check_degradation_change(struct mdinfo
*info
,
9311 unsigned long long new_degraded
;
9312 sysfs_get_ll(info
, NULL
, "degraded", &new_degraded
);
9313 if (new_degraded
!= (unsigned long long)degraded
) {
9314 /* check each device to ensure it is still working */
9317 for (sd
= info
->devs
; sd
; sd
= sd
->next
) {
9318 if (sd
->disk
.state
& (1<<MD_DISK_FAULTY
))
9320 if (sd
->disk
.state
& (1<<MD_DISK_SYNC
)) {
9322 if (sysfs_get_str(info
,
9323 sd
, "state", sbuf
, 20) < 0 ||
9324 strstr(sbuf
, "faulty") ||
9325 strstr(sbuf
, "in_sync") == NULL
) {
9326 /* this device is dead */
9327 sd
->disk
.state
= (1<<MD_DISK_FAULTY
);
9328 if (sd
->disk
.raid_disk
>= 0 &&
9329 sources
[sd
->disk
.raid_disk
] >= 0) {
9331 sd
->disk
.raid_disk
]);
9332 sources
[sd
->disk
.raid_disk
] =
9341 return new_degraded
;
9344 /*******************************************************************************
9345 * Function: imsm_manage_reshape
9346 * Description: Function finds array under reshape and it manages reshape
9347 * process. It creates stripes backups (if required) and sets
9350 * afd : Backup handle (nattive) - not used
9351 * sra : general array info
9352 * reshape : reshape parameters - not used
9353 * st : supertype structure
9354 * blocks : size of critical section [blocks]
9355 * fds : table of source device descriptor
9356 * offsets : start of array (offest per devices)
9358 * destfd : table of destination device descriptor
9359 * destoffsets : table of destination offsets (per device)
9361 * 1 : success, reshape is done
9363 ******************************************************************************/
9364 static int imsm_manage_reshape(
9365 int afd
, struct mdinfo
*sra
, struct reshape
*reshape
,
9366 struct supertype
*st
, unsigned long backup_blocks
,
9367 int *fds
, unsigned long long *offsets
,
9368 int dests
, int *destfd
, unsigned long long *destoffsets
)
9371 struct intel_super
*super
= st
->sb
;
9372 struct intel_dev
*dv
= NULL
;
9373 struct imsm_dev
*dev
= NULL
;
9374 struct imsm_map
*map_src
;
9375 int migr_vol_qan
= 0;
9376 int ndata
, odata
; /* [bytes] */
9377 int chunk
; /* [bytes] */
9378 struct migr_record
*migr_rec
;
9380 unsigned int buf_size
; /* [bytes] */
9381 unsigned long long max_position
; /* array size [bytes] */
9382 unsigned long long next_step
; /* [blocks]/[bytes] */
9383 unsigned long long old_data_stripe_length
;
9384 unsigned long long start_src
; /* [bytes] */
9385 unsigned long long start
; /* [bytes] */
9386 unsigned long long start_buf_shift
; /* [bytes] */
9388 int source_layout
= 0;
9390 if (!fds
|| !offsets
|| !sra
)
9393 /* Find volume during the reshape */
9394 for (dv
= super
->devlist
; dv
; dv
= dv
->next
) {
9395 if (dv
->dev
->vol
.migr_type
== MIGR_GEN_MIGR
9396 && dv
->dev
->vol
.migr_state
== 1) {
9401 /* Only one volume can migrate at the same time */
9402 if (migr_vol_qan
!= 1) {
9403 fprintf(stderr
, Name
" : %s", migr_vol_qan
?
9404 "Number of migrating volumes greater than 1\n" :
9405 "There is no volume during migrationg\n");
9409 map_src
= get_imsm_map(dev
, MAP_1
);
9410 if (map_src
== NULL
)
9413 ndata
= imsm_num_data_members(dev
, MAP_0
);
9414 odata
= imsm_num_data_members(dev
, MAP_1
);
9416 chunk
= __le16_to_cpu(map_src
->blocks_per_strip
) * 512;
9417 old_data_stripe_length
= odata
* chunk
;
9419 migr_rec
= super
->migr_rec
;
9421 /* initialize migration record for start condition */
9422 if (sra
->reshape_progress
== 0)
9423 init_migr_record_imsm(st
, dev
, sra
);
9425 if (__le32_to_cpu(migr_rec
->rec_status
) != UNIT_SRC_NORMAL
) {
9426 dprintf("imsm: cannot restart migration when data "
9427 "are present in copy area.\n");
9433 buf_size
= __le32_to_cpu(migr_rec
->blocks_per_unit
) * 512;
9434 /* extend buffer size for parity disk */
9435 buf_size
+= __le32_to_cpu(migr_rec
->dest_depth_per_unit
) * 512;
9436 /* add space for stripe aligment */
9437 buf_size
+= old_data_stripe_length
;
9438 if (posix_memalign((void **)&buf
, 4096, buf_size
)) {
9439 dprintf("imsm: Cannot allocate checpoint buffer\n");
9443 max_position
= sra
->component_size
* ndata
;
9444 source_layout
= imsm_level_to_layout(map_src
->raid_level
);
9446 while (__le32_to_cpu(migr_rec
->curr_migr_unit
) <
9447 __le32_to_cpu(migr_rec
->num_migr_units
)) {
9448 /* current reshape position [blocks] */
9449 unsigned long long current_position
=
9450 __le32_to_cpu(migr_rec
->blocks_per_unit
)
9451 * __le32_to_cpu(migr_rec
->curr_migr_unit
);
9452 unsigned long long border
;
9454 /* Check that array hasn't become failed.
9456 degraded
= check_degradation_change(sra
, fds
, degraded
);
9458 dprintf("imsm: Abort reshape due to degradation"
9459 " level (%i)\n", degraded
);
9463 next_step
= __le32_to_cpu(migr_rec
->blocks_per_unit
);
9465 if ((current_position
+ next_step
) > max_position
)
9466 next_step
= max_position
- current_position
;
9468 start
= current_position
* 512;
9470 /* allign reading start to old geometry */
9471 start_buf_shift
= start
% old_data_stripe_length
;
9472 start_src
= start
- start_buf_shift
;
9474 border
= (start_src
/ odata
) - (start
/ ndata
);
9476 if (border
<= __le32_to_cpu(migr_rec
->dest_depth_per_unit
)) {
9477 /* save critical stripes to buf
9478 * start - start address of current unit
9480 * start_src - start address of current unit
9481 * to backup alligned to source array
9484 unsigned long long next_step_filler
= 0;
9485 unsigned long long copy_length
= next_step
* 512;
9487 /* allign copy area length to stripe in old geometry */
9488 next_step_filler
= ((copy_length
+ start_buf_shift
)
9489 % old_data_stripe_length
);
9490 if (next_step_filler
)
9491 next_step_filler
= (old_data_stripe_length
9492 - next_step_filler
);
9493 dprintf("save_stripes() parameters: start = %llu,"
9494 "\tstart_src = %llu,\tnext_step*512 = %llu,"
9495 "\tstart_in_buf_shift = %llu,"
9496 "\tnext_step_filler = %llu\n",
9497 start
, start_src
, copy_length
,
9498 start_buf_shift
, next_step_filler
);
9500 if (save_stripes(fds
, offsets
, map_src
->num_members
,
9501 chunk
, map_src
->raid_level
,
9502 source_layout
, 0, NULL
, start_src
,
9504 next_step_filler
+ start_buf_shift
,
9506 dprintf("imsm: Cannot save stripes"
9510 /* Convert data to destination format and store it
9511 * in backup general migration area
9513 if (save_backup_imsm(st
, dev
, sra
,
9514 buf
+ start_buf_shift
, copy_length
)) {
9515 dprintf("imsm: Cannot save stripes to "
9516 "target devices\n");
9519 if (save_checkpoint_imsm(st
, sra
,
9520 UNIT_SRC_IN_CP_AREA
)) {
9521 dprintf("imsm: Cannot write checkpoint to "
9522 "migration record (UNIT_SRC_IN_CP_AREA)\n");
9526 /* set next step to use whole border area */
9527 border
/= next_step
;
9529 next_step
*= border
;
9531 /* When data backed up, checkpoint stored,
9532 * kick the kernel to reshape unit of data
9534 next_step
= next_step
+ sra
->reshape_progress
;
9535 /* limit next step to array max position */
9536 if (next_step
> max_position
)
9537 next_step
= max_position
;
9538 sysfs_set_num(sra
, NULL
, "suspend_lo", sra
->reshape_progress
);
9539 sysfs_set_num(sra
, NULL
, "suspend_hi", next_step
);
9540 sra
->reshape_progress
= next_step
;
9542 /* wait until reshape finish */
9543 if (wait_for_reshape_imsm(sra
, ndata
) < 0) {
9544 dprintf("wait_for_reshape_imsm returned error!\n");
9548 if (save_checkpoint_imsm(st
, sra
, UNIT_SRC_NORMAL
) == 1) {
9549 /* ignore error == 2, this can mean end of reshape here
9551 dprintf("imsm: Cannot write checkpoint to "
9552 "migration record (UNIT_SRC_NORMAL)\n");
9558 /* return '1' if done */
9566 #endif /* MDASSEMBLE */
9568 struct superswitch super_imsm
= {
9570 .examine_super
= examine_super_imsm
,
9571 .brief_examine_super
= brief_examine_super_imsm
,
9572 .brief_examine_subarrays
= brief_examine_subarrays_imsm
,
9573 .export_examine_super
= export_examine_super_imsm
,
9574 .detail_super
= detail_super_imsm
,
9575 .brief_detail_super
= brief_detail_super_imsm
,
9576 .write_init_super
= write_init_super_imsm
,
9577 .validate_geometry
= validate_geometry_imsm
,
9578 .add_to_super
= add_to_super_imsm
,
9579 .remove_from_super
= remove_from_super_imsm
,
9580 .detail_platform
= detail_platform_imsm
,
9581 .kill_subarray
= kill_subarray_imsm
,
9582 .update_subarray
= update_subarray_imsm
,
9583 .load_container
= load_container_imsm
,
9584 .default_geometry
= default_geometry_imsm
,
9585 .get_disk_controller_domain
= imsm_get_disk_controller_domain
,
9586 .reshape_super
= imsm_reshape_super
,
9587 .manage_reshape
= imsm_manage_reshape
,
9588 .recover_backup
= recover_backup_imsm
,
9590 .match_home
= match_home_imsm
,
9591 .uuid_from_super
= uuid_from_super_imsm
,
9592 .getinfo_super
= getinfo_super_imsm
,
9593 .getinfo_super_disks
= getinfo_super_disks_imsm
,
9594 .update_super
= update_super_imsm
,
9596 .avail_size
= avail_size_imsm
,
9597 .min_acceptable_spare_size
= min_acceptable_spare_size_imsm
,
9599 .compare_super
= compare_super_imsm
,
9601 .load_super
= load_super_imsm
,
9602 .init_super
= init_super_imsm
,
9603 .store_super
= store_super_imsm
,
9604 .free_super
= free_super_imsm
,
9605 .match_metadata_desc
= match_metadata_desc_imsm
,
9606 .container_content
= container_content_imsm
,
9614 .open_new
= imsm_open_new
,
9615 .set_array_state
= imsm_set_array_state
,
9616 .set_disk
= imsm_set_disk
,
9617 .sync_metadata
= imsm_sync_metadata
,
9618 .activate_spare
= imsm_activate_spare
,
9619 .process_update
= imsm_process_update
,
9620 .prepare_update
= imsm_prepare_update
,
9621 #endif /* MDASSEMBLE */