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
)) && ((slot
> 1) || (slot
< 0)))
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 struct imsm_map
*map
;
1176 if (is_gen_migration(dev
) == 0)
1179 printf("\nMigration Record Information:");
1181 /* first map under migration */
1182 map
= get_imsm_map(dev
, MAP_0
);
1184 slot
= get_imsm_disk_slot(map
, super
->disks
->index
);
1185 if ((map
== NULL
) || (slot
> 1) || (slot
< 0)) {
1186 printf(" Empty\n ");
1187 printf("Examine one of first two disks in array\n");
1190 printf("\n Status : ");
1191 if (__le32_to_cpu(migr_rec
->rec_status
) == UNIT_SRC_NORMAL
)
1194 printf("Contains Data\n");
1195 printf(" Current Unit : %u\n",
1196 __le32_to_cpu(migr_rec
->curr_migr_unit
));
1197 printf(" Family : %u\n",
1198 __le32_to_cpu(migr_rec
->family_num
));
1199 printf(" Ascending : %u\n",
1200 __le32_to_cpu(migr_rec
->ascending_migr
));
1201 printf(" Blocks Per Unit : %u\n",
1202 __le32_to_cpu(migr_rec
->blocks_per_unit
));
1203 printf(" Dest. Depth Per Unit : %u\n",
1204 __le32_to_cpu(migr_rec
->dest_depth_per_unit
));
1205 printf(" Checkpoint Area pba : %u\n",
1206 __le32_to_cpu(migr_rec
->ckpt_area_pba
));
1207 printf(" First member lba : %u\n",
1208 __le32_to_cpu(migr_rec
->dest_1st_member_lba
));
1209 printf(" Total Number of Units : %u\n",
1210 __le32_to_cpu(migr_rec
->num_migr_units
));
1211 printf(" Size of volume : %u\n",
1212 __le32_to_cpu(migr_rec
->post_migr_vol_cap
));
1213 printf(" Expansion space for LBA64 : %u\n",
1214 __le32_to_cpu(migr_rec
->post_migr_vol_cap_hi
));
1215 printf(" Record was read from : %u\n",
1216 __le32_to_cpu(migr_rec
->ckpt_read_disk_num
));
1221 #endif /* MDASSEMBLE */
1222 /*******************************************************************************
1223 * function: imsm_check_attributes
1224 * Description: Function checks if features represented by attributes flags
1225 * are supported by mdadm.
1227 * attributes - Attributes read from metadata
1229 * 0 - passed attributes contains unsupported features flags
1230 * 1 - all features are supported
1231 ******************************************************************************/
1232 static int imsm_check_attributes(__u32 attributes
)
1235 __u32 not_supported
= MPB_ATTRIB_SUPPORTED
^0xffffffff;
1237 not_supported
&= ~MPB_ATTRIB_IGNORED
;
1239 not_supported
&= attributes
;
1240 if (not_supported
) {
1241 fprintf(stderr
, Name
"(IMSM): Unsupported attributes : %x\n",
1242 (unsigned)__le32_to_cpu(not_supported
));
1243 if (not_supported
& MPB_ATTRIB_CHECKSUM_VERIFY
) {
1244 dprintf("\t\tMPB_ATTRIB_CHECKSUM_VERIFY \n");
1245 not_supported
^= MPB_ATTRIB_CHECKSUM_VERIFY
;
1247 if (not_supported
& MPB_ATTRIB_2TB
) {
1248 dprintf("\t\tMPB_ATTRIB_2TB\n");
1249 not_supported
^= MPB_ATTRIB_2TB
;
1251 if (not_supported
& MPB_ATTRIB_RAID0
) {
1252 dprintf("\t\tMPB_ATTRIB_RAID0\n");
1253 not_supported
^= MPB_ATTRIB_RAID0
;
1255 if (not_supported
& MPB_ATTRIB_RAID1
) {
1256 dprintf("\t\tMPB_ATTRIB_RAID1\n");
1257 not_supported
^= MPB_ATTRIB_RAID1
;
1259 if (not_supported
& MPB_ATTRIB_RAID10
) {
1260 dprintf("\t\tMPB_ATTRIB_RAID10\n");
1261 not_supported
^= MPB_ATTRIB_RAID10
;
1263 if (not_supported
& MPB_ATTRIB_RAID1E
) {
1264 dprintf("\t\tMPB_ATTRIB_RAID1E\n");
1265 not_supported
^= MPB_ATTRIB_RAID1E
;
1267 if (not_supported
& MPB_ATTRIB_RAID5
) {
1268 dprintf("\t\tMPB_ATTRIB_RAID5\n");
1269 not_supported
^= MPB_ATTRIB_RAID5
;
1271 if (not_supported
& MPB_ATTRIB_RAIDCNG
) {
1272 dprintf("\t\tMPB_ATTRIB_RAIDCNG\n");
1273 not_supported
^= MPB_ATTRIB_RAIDCNG
;
1275 if (not_supported
& MPB_ATTRIB_BBM
) {
1276 dprintf("\t\tMPB_ATTRIB_BBM\n");
1277 not_supported
^= MPB_ATTRIB_BBM
;
1279 if (not_supported
& MPB_ATTRIB_CHECKSUM_VERIFY
) {
1280 dprintf("\t\tMPB_ATTRIB_CHECKSUM_VERIFY (== MPB_ATTRIB_LEGACY)\n");
1281 not_supported
^= MPB_ATTRIB_CHECKSUM_VERIFY
;
1283 if (not_supported
& MPB_ATTRIB_EXP_STRIPE_SIZE
) {
1284 dprintf("\t\tMPB_ATTRIB_EXP_STRIP_SIZE\n");
1285 not_supported
^= MPB_ATTRIB_EXP_STRIPE_SIZE
;
1287 if (not_supported
& MPB_ATTRIB_2TB_DISK
) {
1288 dprintf("\t\tMPB_ATTRIB_2TB_DISK\n");
1289 not_supported
^= MPB_ATTRIB_2TB_DISK
;
1291 if (not_supported
& MPB_ATTRIB_NEVER_USE2
) {
1292 dprintf("\t\tMPB_ATTRIB_NEVER_USE2\n");
1293 not_supported
^= MPB_ATTRIB_NEVER_USE2
;
1295 if (not_supported
& MPB_ATTRIB_NEVER_USE
) {
1296 dprintf("\t\tMPB_ATTRIB_NEVER_USE\n");
1297 not_supported
^= MPB_ATTRIB_NEVER_USE
;
1301 dprintf(Name
"(IMSM): Unknown attributes : %x\n", not_supported
);
1310 static void getinfo_super_imsm(struct supertype
*st
, struct mdinfo
*info
, char *map
);
1312 static void examine_super_imsm(struct supertype
*st
, char *homehost
)
1314 struct intel_super
*super
= st
->sb
;
1315 struct imsm_super
*mpb
= super
->anchor
;
1316 char str
[MAX_SIGNATURE_LENGTH
];
1321 __u32 reserved
= imsm_reserved_sectors(super
, super
->disks
);
1324 snprintf(str
, MPB_SIG_LEN
, "%s", mpb
->sig
);
1325 printf(" Magic : %s\n", str
);
1326 snprintf(str
, strlen(MPB_VERSION_RAID0
), "%s", get_imsm_version(mpb
));
1327 printf(" Version : %s\n", get_imsm_version(mpb
));
1328 printf(" Orig Family : %08x\n", __le32_to_cpu(mpb
->orig_family_num
));
1329 printf(" Family : %08x\n", __le32_to_cpu(mpb
->family_num
));
1330 printf(" Generation : %08x\n", __le32_to_cpu(mpb
->generation_num
));
1331 printf(" Attributes : ");
1332 if (imsm_check_attributes(mpb
->attributes
))
1333 printf("All supported\n");
1335 printf("not supported\n");
1336 getinfo_super_imsm(st
, &info
, NULL
);
1337 fname_from_uuid(st
, &info
, nbuf
, ':');
1338 printf(" UUID : %s\n", nbuf
+ 5);
1339 sum
= __le32_to_cpu(mpb
->check_sum
);
1340 printf(" Checksum : %08x %s\n", sum
,
1341 __gen_imsm_checksum(mpb
) == sum
? "correct" : "incorrect");
1342 printf(" MPB Sectors : %d\n", mpb_sectors(mpb
));
1343 printf(" Disks : %d\n", mpb
->num_disks
);
1344 printf(" RAID Devices : %d\n", mpb
->num_raid_devs
);
1345 print_imsm_disk(__get_imsm_disk(mpb
, super
->disks
->index
), super
->disks
->index
, reserved
);
1346 if (super
->bbm_log
) {
1347 struct bbm_log
*log
= super
->bbm_log
;
1350 printf("Bad Block Management Log:\n");
1351 printf(" Log Size : %d\n", __le32_to_cpu(mpb
->bbm_log_size
));
1352 printf(" Signature : %x\n", __le32_to_cpu(log
->signature
));
1353 printf(" Entry Count : %d\n", __le32_to_cpu(log
->entry_count
));
1354 printf(" Spare Blocks : %d\n", __le32_to_cpu(log
->reserved_spare_block_count
));
1355 printf(" First Spare : %llx\n",
1356 (unsigned long long) __le64_to_cpu(log
->first_spare_lba
));
1358 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
1360 struct imsm_dev
*dev
= __get_imsm_dev(mpb
, i
);
1362 super
->current_vol
= i
;
1363 getinfo_super_imsm(st
, &info
, NULL
);
1364 fname_from_uuid(st
, &info
, nbuf
, ':');
1365 print_imsm_dev(super
, dev
, nbuf
+ 5, super
->disks
->index
);
1367 for (i
= 0; i
< mpb
->num_disks
; i
++) {
1368 if (i
== super
->disks
->index
)
1370 print_imsm_disk(__get_imsm_disk(mpb
, i
), i
, reserved
);
1373 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
1374 if (dl
->index
== -1)
1375 print_imsm_disk(&dl
->disk
, -1, reserved
);
1377 examine_migr_rec_imsm(super
);
1380 static void brief_examine_super_imsm(struct supertype
*st
, int verbose
)
1382 /* We just write a generic IMSM ARRAY entry */
1385 struct intel_super
*super
= st
->sb
;
1387 if (!super
->anchor
->num_raid_devs
) {
1388 printf("ARRAY metadata=imsm\n");
1392 getinfo_super_imsm(st
, &info
, NULL
);
1393 fname_from_uuid(st
, &info
, nbuf
, ':');
1394 printf("ARRAY metadata=imsm UUID=%s\n", nbuf
+ 5);
1397 static void brief_examine_subarrays_imsm(struct supertype
*st
, int verbose
)
1399 /* We just write a generic IMSM ARRAY entry */
1403 struct intel_super
*super
= st
->sb
;
1406 if (!super
->anchor
->num_raid_devs
)
1409 getinfo_super_imsm(st
, &info
, NULL
);
1410 fname_from_uuid(st
, &info
, nbuf
, ':');
1411 for (i
= 0; i
< super
->anchor
->num_raid_devs
; i
++) {
1412 struct imsm_dev
*dev
= get_imsm_dev(super
, i
);
1414 super
->current_vol
= i
;
1415 getinfo_super_imsm(st
, &info
, NULL
);
1416 fname_from_uuid(st
, &info
, nbuf1
, ':');
1417 printf("ARRAY /dev/md/%.16s container=%s member=%d UUID=%s\n",
1418 dev
->volume
, nbuf
+ 5, i
, nbuf1
+ 5);
1422 static void export_examine_super_imsm(struct supertype
*st
)
1424 struct intel_super
*super
= st
->sb
;
1425 struct imsm_super
*mpb
= super
->anchor
;
1429 getinfo_super_imsm(st
, &info
, NULL
);
1430 fname_from_uuid(st
, &info
, nbuf
, ':');
1431 printf("MD_METADATA=imsm\n");
1432 printf("MD_LEVEL=container\n");
1433 printf("MD_UUID=%s\n", nbuf
+5);
1434 printf("MD_DEVICES=%u\n", mpb
->num_disks
);
1437 static void detail_super_imsm(struct supertype
*st
, char *homehost
)
1442 getinfo_super_imsm(st
, &info
, NULL
);
1443 fname_from_uuid(st
, &info
, nbuf
, ':');
1444 printf("\n UUID : %s\n", nbuf
+ 5);
1447 static void brief_detail_super_imsm(struct supertype
*st
)
1451 getinfo_super_imsm(st
, &info
, NULL
);
1452 fname_from_uuid(st
, &info
, nbuf
, ':');
1453 printf(" UUID=%s", nbuf
+ 5);
1456 static int imsm_read_serial(int fd
, char *devname
, __u8
*serial
);
1457 static void fd2devname(int fd
, char *name
);
1459 static int ahci_enumerate_ports(const char *hba_path
, int port_count
, int host_base
, int verbose
)
1461 /* dump an unsorted list of devices attached to AHCI Intel storage
1462 * controller, as well as non-connected ports
1464 int hba_len
= strlen(hba_path
) + 1;
1469 unsigned long port_mask
= (1 << port_count
) - 1;
1471 if (port_count
> (int)sizeof(port_mask
) * 8) {
1473 fprintf(stderr
, Name
": port_count %d out of range\n", port_count
);
1477 /* scroll through /sys/dev/block looking for devices attached to
1480 dir
= opendir("/sys/dev/block");
1481 for (ent
= dir
? readdir(dir
) : NULL
; ent
; ent
= readdir(dir
)) {
1492 if (sscanf(ent
->d_name
, "%d:%d", &major
, &minor
) != 2)
1494 path
= devt_to_devpath(makedev(major
, minor
));
1497 if (!path_attached_to_hba(path
, hba_path
)) {
1503 /* retrieve the scsi device type */
1504 if (asprintf(&device
, "/sys/dev/block/%d:%d/device/xxxxxxx", major
, minor
) < 0) {
1506 fprintf(stderr
, Name
": failed to allocate 'device'\n");
1510 sprintf(device
, "/sys/dev/block/%d:%d/device/type", major
, minor
);
1511 if (load_sys(device
, buf
) != 0) {
1513 fprintf(stderr
, Name
": failed to read device type for %s\n",
1519 type
= strtoul(buf
, NULL
, 10);
1521 /* if it's not a disk print the vendor and model */
1522 if (!(type
== 0 || type
== 7 || type
== 14)) {
1525 sprintf(device
, "/sys/dev/block/%d:%d/device/vendor", major
, minor
);
1526 if (load_sys(device
, buf
) == 0) {
1527 strncpy(vendor
, buf
, sizeof(vendor
));
1528 vendor
[sizeof(vendor
) - 1] = '\0';
1529 c
= (char *) &vendor
[sizeof(vendor
) - 1];
1530 while (isspace(*c
) || *c
== '\0')
1534 sprintf(device
, "/sys/dev/block/%d:%d/device/model", major
, minor
);
1535 if (load_sys(device
, buf
) == 0) {
1536 strncpy(model
, buf
, sizeof(model
));
1537 model
[sizeof(model
) - 1] = '\0';
1538 c
= (char *) &model
[sizeof(model
) - 1];
1539 while (isspace(*c
) || *c
== '\0')
1543 if (vendor
[0] && model
[0])
1544 sprintf(buf
, "%.64s %.64s", vendor
, model
);
1546 switch (type
) { /* numbers from hald/linux/device.c */
1547 case 1: sprintf(buf
, "tape"); break;
1548 case 2: sprintf(buf
, "printer"); break;
1549 case 3: sprintf(buf
, "processor"); break;
1551 case 5: sprintf(buf
, "cdrom"); break;
1552 case 6: sprintf(buf
, "scanner"); break;
1553 case 8: sprintf(buf
, "media_changer"); break;
1554 case 9: sprintf(buf
, "comm"); break;
1555 case 12: sprintf(buf
, "raid"); break;
1556 default: sprintf(buf
, "unknown");
1562 /* chop device path to 'host%d' and calculate the port number */
1563 c
= strchr(&path
[hba_len
], '/');
1566 fprintf(stderr
, Name
": %s - invalid path name\n", path
+ hba_len
);
1571 if (sscanf(&path
[hba_len
], "host%d", &port
) == 1)
1575 *c
= '/'; /* repair the full string */
1576 fprintf(stderr
, Name
": failed to determine port number for %s\n",
1583 /* mark this port as used */
1584 port_mask
&= ~(1 << port
);
1586 /* print out the device information */
1588 printf(" Port%d : - non-disk device (%s) -\n", port
, buf
);
1592 fd
= dev_open(ent
->d_name
, O_RDONLY
);
1594 printf(" Port%d : - disk info unavailable -\n", port
);
1596 fd2devname(fd
, buf
);
1597 printf(" Port%d : %s", port
, buf
);
1598 if (imsm_read_serial(fd
, NULL
, (__u8
*) buf
) == 0)
1599 printf(" (%.*s)\n", MAX_RAID_SERIAL_LEN
, buf
);
1614 for (i
= 0; i
< port_count
; i
++)
1615 if (port_mask
& (1 << i
))
1616 printf(" Port%d : - no device attached -\n", i
);
1622 static void print_found_intel_controllers(struct sys_dev
*elem
)
1624 for (; elem
; elem
= elem
->next
) {
1625 fprintf(stderr
, Name
": found Intel(R) ");
1626 if (elem
->type
== SYS_DEV_SATA
)
1627 fprintf(stderr
, "SATA ");
1628 else if (elem
->type
== SYS_DEV_SAS
)
1629 fprintf(stderr
, "SAS ");
1630 fprintf(stderr
, "RAID controller");
1632 fprintf(stderr
, " at %s", elem
->pci_id
);
1633 fprintf(stderr
, ".\n");
1638 static int ahci_get_port_count(const char *hba_path
, int *port_count
)
1645 if ((dir
= opendir(hba_path
)) == NULL
)
1648 for (ent
= readdir(dir
); ent
; ent
= readdir(dir
)) {
1651 if (sscanf(ent
->d_name
, "host%d", &host
) != 1)
1653 if (*port_count
== 0)
1655 else if (host
< host_base
)
1658 if (host
+ 1 > *port_count
+ host_base
)
1659 *port_count
= host
+ 1 - host_base
;
1665 static void print_imsm_capability(const struct imsm_orom
*orom
)
1667 printf(" Platform : Intel(R) Matrix Storage Manager\n");
1668 printf(" Version : %d.%d.%d.%d\n", orom
->major_ver
, orom
->minor_ver
,
1669 orom
->hotfix_ver
, orom
->build
);
1670 printf(" RAID Levels :%s%s%s%s%s\n",
1671 imsm_orom_has_raid0(orom
) ? " raid0" : "",
1672 imsm_orom_has_raid1(orom
) ? " raid1" : "",
1673 imsm_orom_has_raid1e(orom
) ? " raid1e" : "",
1674 imsm_orom_has_raid10(orom
) ? " raid10" : "",
1675 imsm_orom_has_raid5(orom
) ? " raid5" : "");
1676 printf(" Chunk Sizes :%s%s%s%s%s%s%s%s%s%s%s%s%s%s%s%s\n",
1677 imsm_orom_has_chunk(orom
, 2) ? " 2k" : "",
1678 imsm_orom_has_chunk(orom
, 4) ? " 4k" : "",
1679 imsm_orom_has_chunk(orom
, 8) ? " 8k" : "",
1680 imsm_orom_has_chunk(orom
, 16) ? " 16k" : "",
1681 imsm_orom_has_chunk(orom
, 32) ? " 32k" : "",
1682 imsm_orom_has_chunk(orom
, 64) ? " 64k" : "",
1683 imsm_orom_has_chunk(orom
, 128) ? " 128k" : "",
1684 imsm_orom_has_chunk(orom
, 256) ? " 256k" : "",
1685 imsm_orom_has_chunk(orom
, 512) ? " 512k" : "",
1686 imsm_orom_has_chunk(orom
, 1024*1) ? " 1M" : "",
1687 imsm_orom_has_chunk(orom
, 1024*2) ? " 2M" : "",
1688 imsm_orom_has_chunk(orom
, 1024*4) ? " 4M" : "",
1689 imsm_orom_has_chunk(orom
, 1024*8) ? " 8M" : "",
1690 imsm_orom_has_chunk(orom
, 1024*16) ? " 16M" : "",
1691 imsm_orom_has_chunk(orom
, 1024*32) ? " 32M" : "",
1692 imsm_orom_has_chunk(orom
, 1024*64) ? " 64M" : "");
1693 printf(" Max Disks : %d\n", orom
->tds
);
1694 printf(" Max Volumes : %d\n", orom
->vpa
);
1698 static int detail_platform_imsm(int verbose
, int enumerate_only
)
1700 /* There are two components to imsm platform support, the ahci SATA
1701 * controller and the option-rom. To find the SATA controller we
1702 * simply look in /sys/bus/pci/drivers/ahci to see if an ahci
1703 * controller with the Intel vendor id is present. This approach
1704 * allows mdadm to leverage the kernel's ahci detection logic, with the
1705 * caveat that if ahci.ko is not loaded mdadm will not be able to
1706 * detect platform raid capabilities. The option-rom resides in a
1707 * platform "Adapter ROM". We scan for its signature to retrieve the
1708 * platform capabilities. If raid support is disabled in the BIOS the
1709 * option-rom capability structure will not be available.
1711 const struct imsm_orom
*orom
;
1712 struct sys_dev
*list
, *hba
;
1717 if (enumerate_only
) {
1718 if (check_env("IMSM_NO_PLATFORM"))
1720 list
= find_intel_devices();
1723 for (hba
= list
; hba
; hba
= hba
->next
) {
1724 orom
= find_imsm_capability(hba
->type
);
1730 free_sys_dev(&list
);
1734 list
= find_intel_devices();
1737 fprintf(stderr
, Name
": no active Intel(R) RAID "
1738 "controller found.\n");
1739 free_sys_dev(&list
);
1742 print_found_intel_controllers(list
);
1744 for (hba
= list
; hba
; hba
= hba
->next
) {
1745 orom
= find_imsm_capability(hba
->type
);
1747 fprintf(stderr
, Name
": imsm capabilities not found for controller: %s (type %s)\n",
1748 hba
->path
, get_sys_dev_type(hba
->type
));
1750 print_imsm_capability(orom
);
1753 for (hba
= list
; hba
; hba
= hba
->next
) {
1754 printf(" I/O Controller : %s (%s)\n",
1755 hba
->path
, get_sys_dev_type(hba
->type
));
1757 if (hba
->type
== SYS_DEV_SATA
) {
1758 host_base
= ahci_get_port_count(hba
->path
, &port_count
);
1759 if (ahci_enumerate_ports(hba
->path
, port_count
, host_base
, verbose
)) {
1761 fprintf(stderr
, Name
": failed to enumerate "
1762 "ports on SATA controller at %s.", hba
->pci_id
);
1768 free_sys_dev(&list
);
1773 static int match_home_imsm(struct supertype
*st
, char *homehost
)
1775 /* the imsm metadata format does not specify any host
1776 * identification information. We return -1 since we can never
1777 * confirm nor deny whether a given array is "meant" for this
1778 * host. We rely on compare_super and the 'family_num' fields to
1779 * exclude member disks that do not belong, and we rely on
1780 * mdadm.conf to specify the arrays that should be assembled.
1781 * Auto-assembly may still pick up "foreign" arrays.
1787 static void uuid_from_super_imsm(struct supertype
*st
, int uuid
[4])
1789 /* The uuid returned here is used for:
1790 * uuid to put into bitmap file (Create, Grow)
1791 * uuid for backup header when saving critical section (Grow)
1792 * comparing uuids when re-adding a device into an array
1793 * In these cases the uuid required is that of the data-array,
1794 * not the device-set.
1795 * uuid to recognise same set when adding a missing device back
1796 * to an array. This is a uuid for the device-set.
1798 * For each of these we can make do with a truncated
1799 * or hashed uuid rather than the original, as long as
1801 * In each case the uuid required is that of the data-array,
1802 * not the device-set.
1804 /* imsm does not track uuid's so we synthesis one using sha1 on
1805 * - The signature (Which is constant for all imsm array, but no matter)
1806 * - the orig_family_num of the container
1807 * - the index number of the volume
1808 * - the 'serial' number of the volume.
1809 * Hopefully these are all constant.
1811 struct intel_super
*super
= st
->sb
;
1814 struct sha1_ctx ctx
;
1815 struct imsm_dev
*dev
= NULL
;
1818 /* some mdadm versions failed to set ->orig_family_num, in which
1819 * case fall back to ->family_num. orig_family_num will be
1820 * fixed up with the first metadata update.
1822 family_num
= super
->anchor
->orig_family_num
;
1823 if (family_num
== 0)
1824 family_num
= super
->anchor
->family_num
;
1825 sha1_init_ctx(&ctx
);
1826 sha1_process_bytes(super
->anchor
->sig
, MPB_SIG_LEN
, &ctx
);
1827 sha1_process_bytes(&family_num
, sizeof(__u32
), &ctx
);
1828 if (super
->current_vol
>= 0)
1829 dev
= get_imsm_dev(super
, super
->current_vol
);
1831 __u32 vol
= super
->current_vol
;
1832 sha1_process_bytes(&vol
, sizeof(vol
), &ctx
);
1833 sha1_process_bytes(dev
->volume
, MAX_RAID_SERIAL_LEN
, &ctx
);
1835 sha1_finish_ctx(&ctx
, buf
);
1836 memcpy(uuid
, buf
, 4*4);
1841 get_imsm_numerical_version(struct imsm_super
*mpb
, int *m
, int *p
)
1843 __u8
*v
= get_imsm_version(mpb
);
1844 __u8
*end
= mpb
->sig
+ MAX_SIGNATURE_LENGTH
;
1845 char major
[] = { 0, 0, 0 };
1846 char minor
[] = { 0 ,0, 0 };
1847 char patch
[] = { 0, 0, 0 };
1848 char *ver_parse
[] = { major
, minor
, patch
};
1852 while (*v
!= '\0' && v
< end
) {
1853 if (*v
!= '.' && j
< 2)
1854 ver_parse
[i
][j
++] = *v
;
1862 *m
= strtol(minor
, NULL
, 0);
1863 *p
= strtol(patch
, NULL
, 0);
1867 static __u32
migr_strip_blocks_resync(struct imsm_dev
*dev
)
1869 /* migr_strip_size when repairing or initializing parity */
1870 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
1871 __u32 chunk
= __le32_to_cpu(map
->blocks_per_strip
);
1873 switch (get_imsm_raid_level(map
)) {
1878 return 128*1024 >> 9;
1882 static __u32
migr_strip_blocks_rebuild(struct imsm_dev
*dev
)
1884 /* migr_strip_size when rebuilding a degraded disk, no idea why
1885 * this is different than migr_strip_size_resync(), but it's good
1888 struct imsm_map
*map
= get_imsm_map(dev
, MAP_1
);
1889 __u32 chunk
= __le32_to_cpu(map
->blocks_per_strip
);
1891 switch (get_imsm_raid_level(map
)) {
1894 if (map
->num_members
% map
->num_domains
== 0)
1895 return 128*1024 >> 9;
1899 return max((__u32
) 64*1024 >> 9, chunk
);
1901 return 128*1024 >> 9;
1905 static __u32
num_stripes_per_unit_resync(struct imsm_dev
*dev
)
1907 struct imsm_map
*lo
= get_imsm_map(dev
, MAP_0
);
1908 struct imsm_map
*hi
= get_imsm_map(dev
, MAP_1
);
1909 __u32 lo_chunk
= __le32_to_cpu(lo
->blocks_per_strip
);
1910 __u32 hi_chunk
= __le32_to_cpu(hi
->blocks_per_strip
);
1912 return max((__u32
) 1, hi_chunk
/ lo_chunk
);
1915 static __u32
num_stripes_per_unit_rebuild(struct imsm_dev
*dev
)
1917 struct imsm_map
*lo
= get_imsm_map(dev
, MAP_0
);
1918 int level
= get_imsm_raid_level(lo
);
1920 if (level
== 1 || level
== 10) {
1921 struct imsm_map
*hi
= get_imsm_map(dev
, MAP_1
);
1923 return hi
->num_domains
;
1925 return num_stripes_per_unit_resync(dev
);
1928 static __u8
imsm_num_data_members(struct imsm_dev
*dev
, int second_map
)
1930 /* named 'imsm_' because raid0, raid1 and raid10
1931 * counter-intuitively have the same number of data disks
1933 struct imsm_map
*map
= get_imsm_map(dev
, second_map
);
1935 switch (get_imsm_raid_level(map
)) {
1939 return map
->num_members
;
1941 return map
->num_members
- 1;
1943 dprintf("%s: unsupported raid level\n", __func__
);
1948 static __u32
parity_segment_depth(struct imsm_dev
*dev
)
1950 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
1951 __u32 chunk
= __le32_to_cpu(map
->blocks_per_strip
);
1953 switch(get_imsm_raid_level(map
)) {
1956 return chunk
* map
->num_domains
;
1958 return chunk
* map
->num_members
;
1964 static __u32
map_migr_block(struct imsm_dev
*dev
, __u32 block
)
1966 struct imsm_map
*map
= get_imsm_map(dev
, MAP_1
);
1967 __u32 chunk
= __le32_to_cpu(map
->blocks_per_strip
);
1968 __u32 strip
= block
/ chunk
;
1970 switch (get_imsm_raid_level(map
)) {
1973 __u32 vol_strip
= (strip
* map
->num_domains
) + 1;
1974 __u32 vol_stripe
= vol_strip
/ map
->num_members
;
1976 return vol_stripe
* chunk
+ block
% chunk
;
1978 __u32 stripe
= strip
/ (map
->num_members
- 1);
1980 return stripe
* chunk
+ block
% chunk
;
1987 static __u64
blocks_per_migr_unit(struct intel_super
*super
,
1988 struct imsm_dev
*dev
)
1990 /* calculate the conversion factor between per member 'blocks'
1991 * (md/{resync,rebuild}_start) and imsm migration units, return
1992 * 0 for the 'not migrating' and 'unsupported migration' cases
1994 if (!dev
->vol
.migr_state
)
1997 switch (migr_type(dev
)) {
1998 case MIGR_GEN_MIGR
: {
1999 struct migr_record
*migr_rec
= super
->migr_rec
;
2000 return __le32_to_cpu(migr_rec
->blocks_per_unit
);
2005 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
2006 __u32 stripes_per_unit
;
2007 __u32 blocks_per_unit
;
2016 /* yes, this is really the translation of migr_units to
2017 * per-member blocks in the 'resync' case
2019 stripes_per_unit
= num_stripes_per_unit_resync(dev
);
2020 migr_chunk
= migr_strip_blocks_resync(dev
);
2021 disks
= imsm_num_data_members(dev
, MAP_0
);
2022 blocks_per_unit
= stripes_per_unit
* migr_chunk
* disks
;
2023 stripe
= __le16_to_cpu(map
->blocks_per_strip
) * disks
;
2024 segment
= blocks_per_unit
/ stripe
;
2025 block_rel
= blocks_per_unit
- segment
* stripe
;
2026 parity_depth
= parity_segment_depth(dev
);
2027 block_map
= map_migr_block(dev
, block_rel
);
2028 return block_map
+ parity_depth
* segment
;
2030 case MIGR_REBUILD
: {
2031 __u32 stripes_per_unit
;
2034 stripes_per_unit
= num_stripes_per_unit_rebuild(dev
);
2035 migr_chunk
= migr_strip_blocks_rebuild(dev
);
2036 return migr_chunk
* stripes_per_unit
;
2038 case MIGR_STATE_CHANGE
:
2044 static int imsm_level_to_layout(int level
)
2052 return ALGORITHM_LEFT_ASYMMETRIC
;
2059 /*******************************************************************************
2060 * Function: read_imsm_migr_rec
2061 * Description: Function reads imsm migration record from last sector of disk
2063 * fd : disk descriptor
2064 * super : metadata info
2068 ******************************************************************************/
2069 static int read_imsm_migr_rec(int fd
, struct intel_super
*super
)
2072 unsigned long long dsize
;
2074 get_dev_size(fd
, NULL
, &dsize
);
2075 if (lseek64(fd
, dsize
- 512, SEEK_SET
) < 0) {
2077 Name
": Cannot seek to anchor block: %s\n",
2081 if (read(fd
, super
->migr_rec_buf
, 512) != 512) {
2083 Name
": Cannot read migr record block: %s\n",
2093 static struct imsm_dev
*imsm_get_device_during_migration(
2094 struct intel_super
*super
)
2097 struct intel_dev
*dv
;
2099 for (dv
= super
->devlist
; dv
; dv
= dv
->next
) {
2100 if (is_gen_migration(dv
->dev
))
2106 /*******************************************************************************
2107 * Function: load_imsm_migr_rec
2108 * Description: Function reads imsm migration record (it is stored at the last
2111 * super : imsm internal array info
2112 * info : general array info
2116 ******************************************************************************/
2117 static int load_imsm_migr_rec(struct intel_super
*super
, struct mdinfo
*info
)
2120 struct dl
*dl
= NULL
;
2124 struct imsm_dev
*dev
;
2125 struct imsm_map
*map
= NULL
;
2128 /* find map under migration */
2129 dev
= imsm_get_device_during_migration(super
);
2130 /* nothing to load,no migration in progress?
2134 map
= get_imsm_map(dev
, MAP_0
);
2137 for (sd
= info
->devs
; sd
; sd
= sd
->next
) {
2138 /* skip spare and failed disks
2140 if (sd
->disk
.raid_disk
< 0)
2142 /* read only from one of the first two slots */
2144 slot
= get_imsm_disk_slot(map
,
2145 sd
->disk
.raid_disk
);
2146 if ((map
== NULL
) || (slot
> 1) || (slot
< 0))
2149 sprintf(nm
, "%d:%d", sd
->disk
.major
, sd
->disk
.minor
);
2150 fd
= dev_open(nm
, O_RDONLY
);
2156 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
2157 /* skip spare and failed disks
2161 /* read only from one of the first two slots */
2163 slot
= get_imsm_disk_slot(map
, dl
->index
);
2164 if ((map
== NULL
) || (slot
> 1) || (slot
< 0))
2166 sprintf(nm
, "%d:%d", dl
->major
, dl
->minor
);
2167 fd
= dev_open(nm
, O_RDONLY
);
2174 retval
= read_imsm_migr_rec(fd
, super
);
2183 /*******************************************************************************
2184 * function: imsm_create_metadata_checkpoint_update
2185 * Description: It creates update for checkpoint change.
2187 * super : imsm internal array info
2188 * u : pointer to prepared update
2191 * If length is equal to 0, input pointer u contains no update
2192 ******************************************************************************/
2193 static int imsm_create_metadata_checkpoint_update(
2194 struct intel_super
*super
,
2195 struct imsm_update_general_migration_checkpoint
**u
)
2198 int update_memory_size
= 0;
2200 dprintf("imsm_create_metadata_checkpoint_update(enter)\n");
2206 /* size of all update data without anchor */
2207 update_memory_size
=
2208 sizeof(struct imsm_update_general_migration_checkpoint
);
2210 *u
= calloc(1, update_memory_size
);
2212 dprintf("error: cannot get memory for "
2213 "imsm_create_metadata_checkpoint_update update\n");
2216 (*u
)->type
= update_general_migration_checkpoint
;
2217 (*u
)->curr_migr_unit
= __le32_to_cpu(super
->migr_rec
->curr_migr_unit
);
2218 dprintf("imsm_create_metadata_checkpoint_update: prepared for %u\n",
2219 (*u
)->curr_migr_unit
);
2221 return update_memory_size
;
2225 static void imsm_update_metadata_locally(struct supertype
*st
,
2226 void *buf
, int len
);
2228 /*******************************************************************************
2229 * Function: write_imsm_migr_rec
2230 * Description: Function writes imsm migration record
2231 * (at the last sector of disk)
2233 * super : imsm internal array info
2237 ******************************************************************************/
2238 static int write_imsm_migr_rec(struct supertype
*st
)
2240 struct intel_super
*super
= st
->sb
;
2241 unsigned long long dsize
;
2247 struct imsm_update_general_migration_checkpoint
*u
;
2248 struct imsm_dev
*dev
;
2249 struct imsm_map
*map
= NULL
;
2251 /* find map under migration */
2252 dev
= imsm_get_device_during_migration(super
);
2253 /* if no migration, write buffer anyway to clear migr_record
2254 * on disk based on first available device
2257 dev
= get_imsm_dev(super
, super
->current_vol
< 0 ? 0 :
2258 super
->current_vol
);
2260 map
= get_imsm_map(dev
, MAP_0
);
2262 for (sd
= super
->disks
; sd
; sd
= sd
->next
) {
2265 /* skip failed and spare devices */
2268 /* write to 2 first slots only */
2270 slot
= get_imsm_disk_slot(map
, sd
->index
);
2271 if ((map
== NULL
) || (slot
> 1) || (slot
< 0))
2274 sprintf(nm
, "%d:%d", sd
->major
, sd
->minor
);
2275 fd
= dev_open(nm
, O_RDWR
);
2278 get_dev_size(fd
, NULL
, &dsize
);
2279 if (lseek64(fd
, dsize
- 512, SEEK_SET
) < 0) {
2281 Name
": Cannot seek to anchor block: %s\n",
2285 if (write(fd
, super
->migr_rec_buf
, 512) != 512) {
2287 Name
": Cannot write migr record block: %s\n",
2294 /* update checkpoint information in metadata */
2295 len
= imsm_create_metadata_checkpoint_update(super
, &u
);
2298 dprintf("imsm: Cannot prepare update\n");
2301 /* update metadata locally */
2302 imsm_update_metadata_locally(st
, u
, len
);
2303 /* and possibly remotely */
2304 if (st
->update_tail
) {
2305 append_metadata_update(st
, u
, len
);
2306 /* during reshape we do all work inside metadata handler
2307 * manage_reshape(), so metadata update has to be triggered
2310 flush_metadata_updates(st
);
2311 st
->update_tail
= &st
->updates
;
2321 #endif /* MDASSEMBLE */
2323 /* spare/missing disks activations are not allowe when
2324 * array/container performs reshape operation, because
2325 * all arrays in container works on the same disks set
2327 int imsm_reshape_blocks_arrays_changes(struct intel_super
*super
)
2330 struct intel_dev
*i_dev
;
2331 struct imsm_dev
*dev
;
2333 /* check whole container
2335 for (i_dev
= super
->devlist
; i_dev
; i_dev
= i_dev
->next
) {
2337 if (is_gen_migration(dev
)) {
2338 /* No repair during any migration in container
2347 static void getinfo_super_imsm_volume(struct supertype
*st
, struct mdinfo
*info
, char *dmap
)
2349 struct intel_super
*super
= st
->sb
;
2350 struct migr_record
*migr_rec
= super
->migr_rec
;
2351 struct imsm_dev
*dev
= get_imsm_dev(super
, super
->current_vol
);
2352 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
2353 struct imsm_map
*prev_map
= get_imsm_map(dev
, MAP_1
);
2354 struct imsm_map
*map_to_analyse
= map
;
2357 unsigned int component_size_alligment
;
2358 int map_disks
= info
->array
.raid_disks
;
2360 memset(info
, 0, sizeof(*info
));
2362 map_to_analyse
= prev_map
;
2364 dl
= super
->current_disk
;
2366 info
->container_member
= super
->current_vol
;
2367 info
->array
.raid_disks
= map
->num_members
;
2368 info
->array
.level
= get_imsm_raid_level(map_to_analyse
);
2369 info
->array
.layout
= imsm_level_to_layout(info
->array
.level
);
2370 info
->array
.md_minor
= -1;
2371 info
->array
.ctime
= 0;
2372 info
->array
.utime
= 0;
2373 info
->array
.chunk_size
=
2374 __le16_to_cpu(map_to_analyse
->blocks_per_strip
) << 9;
2375 info
->array
.state
= !dev
->vol
.dirty
;
2376 info
->custom_array_size
= __le32_to_cpu(dev
->size_high
);
2377 info
->custom_array_size
<<= 32;
2378 info
->custom_array_size
|= __le32_to_cpu(dev
->size_low
);
2379 info
->recovery_blocked
= imsm_reshape_blocks_arrays_changes(st
->sb
);
2381 if (is_gen_migration(dev
)) {
2382 info
->reshape_active
= 1;
2383 info
->new_level
= get_imsm_raid_level(map
);
2384 info
->new_layout
= imsm_level_to_layout(info
->new_level
);
2385 info
->new_chunk
= __le16_to_cpu(map
->blocks_per_strip
) << 9;
2386 info
->delta_disks
= map
->num_members
- prev_map
->num_members
;
2387 if (info
->delta_disks
) {
2388 /* this needs to be applied to every array
2391 info
->reshape_active
= CONTAINER_RESHAPE
;
2393 /* We shape information that we give to md might have to be
2394 * modify to cope with md's requirement for reshaping arrays.
2395 * For example, when reshaping a RAID0, md requires it to be
2396 * presented as a degraded RAID4.
2397 * Also if a RAID0 is migrating to a RAID5 we need to specify
2398 * the array as already being RAID5, but the 'before' layout
2399 * is a RAID4-like layout.
2401 switch (info
->array
.level
) {
2403 switch(info
->new_level
) {
2405 /* conversion is happening as RAID4 */
2406 info
->array
.level
= 4;
2407 info
->array
.raid_disks
+= 1;
2410 /* conversion is happening as RAID5 */
2411 info
->array
.level
= 5;
2412 info
->array
.layout
= ALGORITHM_PARITY_N
;
2413 info
->delta_disks
-= 1;
2416 /* FIXME error message */
2417 info
->array
.level
= UnSet
;
2423 info
->new_level
= UnSet
;
2424 info
->new_layout
= UnSet
;
2425 info
->new_chunk
= info
->array
.chunk_size
;
2426 info
->delta_disks
= 0;
2430 info
->disk
.major
= dl
->major
;
2431 info
->disk
.minor
= dl
->minor
;
2432 info
->disk
.number
= dl
->index
;
2433 info
->disk
.raid_disk
= get_imsm_disk_slot(map_to_analyse
,
2437 info
->data_offset
= __le32_to_cpu(map_to_analyse
->pba_of_lba0
);
2438 info
->component_size
=
2439 __le32_to_cpu(map_to_analyse
->blocks_per_member
);
2441 /* check component size aligment
2443 component_size_alligment
=
2444 info
->component_size
% (info
->array
.chunk_size
/512);
2446 if (component_size_alligment
&&
2447 (info
->array
.level
!= 1) && (info
->array
.level
!= UnSet
)) {
2448 dprintf("imsm: reported component size alligned from %llu ",
2449 info
->component_size
);
2450 info
->component_size
-= component_size_alligment
;
2451 dprintf("to %llu (%i).\n",
2452 info
->component_size
, component_size_alligment
);
2455 memset(info
->uuid
, 0, sizeof(info
->uuid
));
2456 info
->recovery_start
= MaxSector
;
2458 info
->reshape_progress
= 0;
2459 info
->resync_start
= MaxSector
;
2460 if ((map_to_analyse
->map_state
== IMSM_T_STATE_UNINITIALIZED
||
2462 imsm_reshape_blocks_arrays_changes(super
) == 0) {
2463 info
->resync_start
= 0;
2465 if (dev
->vol
.migr_state
) {
2466 switch (migr_type(dev
)) {
2469 __u64 blocks_per_unit
= blocks_per_migr_unit(super
,
2471 __u64 units
= __le32_to_cpu(dev
->vol
.curr_migr_unit
);
2473 info
->resync_start
= blocks_per_unit
* units
;
2476 case MIGR_GEN_MIGR
: {
2477 __u64 blocks_per_unit
= blocks_per_migr_unit(super
,
2479 __u64 units
= __le32_to_cpu(migr_rec
->curr_migr_unit
);
2480 unsigned long long array_blocks
;
2483 if (__le32_to_cpu(migr_rec
->ascending_migr
) &&
2485 (__le32_to_cpu(migr_rec
->num_migr_units
)-1)) &&
2486 (super
->migr_rec
->rec_status
==
2487 __cpu_to_le32(UNIT_SRC_IN_CP_AREA
)))
2490 info
->reshape_progress
= blocks_per_unit
* units
;
2492 dprintf("IMSM: General Migration checkpoint : %llu "
2493 "(%llu) -> read reshape progress : %llu\n",
2494 (unsigned long long)units
,
2495 (unsigned long long)blocks_per_unit
,
2496 info
->reshape_progress
);
2498 used_disks
= imsm_num_data_members(dev
, MAP_1
);
2499 if (used_disks
> 0) {
2500 array_blocks
= map
->blocks_per_member
*
2502 /* round array size down to closest MB
2504 info
->custom_array_size
= (array_blocks
2505 >> SECT_PER_MB_SHIFT
)
2506 << SECT_PER_MB_SHIFT
;
2510 /* we could emulate the checkpointing of
2511 * 'sync_action=check' migrations, but for now
2512 * we just immediately complete them
2515 /* this is handled by container_content_imsm() */
2516 case MIGR_STATE_CHANGE
:
2517 /* FIXME handle other migrations */
2519 /* we are not dirty, so... */
2520 info
->resync_start
= MaxSector
;
2524 strncpy(info
->name
, (char *) dev
->volume
, MAX_RAID_SERIAL_LEN
);
2525 info
->name
[MAX_RAID_SERIAL_LEN
] = 0;
2527 info
->array
.major_version
= -1;
2528 info
->array
.minor_version
= -2;
2529 devname
= devnum2devname(st
->container_dev
);
2530 *info
->text_version
= '\0';
2532 sprintf(info
->text_version
, "/%s/%d", devname
, info
->container_member
);
2534 info
->safe_mode_delay
= 4000; /* 4 secs like the Matrix driver */
2535 uuid_from_super_imsm(st
, info
->uuid
);
2539 for (i
=0; i
<map_disks
; i
++) {
2541 if (i
< info
->array
.raid_disks
) {
2542 struct imsm_disk
*dsk
;
2543 j
= get_imsm_disk_idx(dev
, i
, MAP_X
);
2544 dsk
= get_imsm_disk(super
, j
);
2545 if (dsk
&& (dsk
->status
& CONFIGURED_DISK
))
2552 static __u8
imsm_check_degraded(struct intel_super
*super
, struct imsm_dev
*dev
,
2553 int failed
, int look_in_map
);
2555 static int imsm_count_failed(struct intel_super
*super
, struct imsm_dev
*dev
,
2558 static void manage_second_map(struct intel_super
*super
, struct imsm_dev
*dev
)
2560 if (is_gen_migration(dev
)) {
2563 struct imsm_map
*map2
= get_imsm_map(dev
, MAP_1
);
2565 failed
= imsm_count_failed(super
, dev
, MAP_1
);
2566 map_state
= imsm_check_degraded(super
, dev
, failed
, MAP_1
);
2567 if (map2
->map_state
!= map_state
) {
2568 map2
->map_state
= map_state
;
2569 super
->updates_pending
++;
2574 static struct imsm_disk
*get_imsm_missing(struct intel_super
*super
, __u8 index
)
2578 for (d
= super
->missing
; d
; d
= d
->next
)
2579 if (d
->index
== index
)
2584 static void getinfo_super_imsm(struct supertype
*st
, struct mdinfo
*info
, char *map
)
2586 struct intel_super
*super
= st
->sb
;
2587 struct imsm_disk
*disk
;
2588 int map_disks
= info
->array
.raid_disks
;
2589 int max_enough
= -1;
2591 struct imsm_super
*mpb
;
2593 if (super
->current_vol
>= 0) {
2594 getinfo_super_imsm_volume(st
, info
, map
);
2597 memset(info
, 0, sizeof(*info
));
2599 /* Set raid_disks to zero so that Assemble will always pull in valid
2602 info
->array
.raid_disks
= 0;
2603 info
->array
.level
= LEVEL_CONTAINER
;
2604 info
->array
.layout
= 0;
2605 info
->array
.md_minor
= -1;
2606 info
->array
.ctime
= 0; /* N/A for imsm */
2607 info
->array
.utime
= 0;
2608 info
->array
.chunk_size
= 0;
2610 info
->disk
.major
= 0;
2611 info
->disk
.minor
= 0;
2612 info
->disk
.raid_disk
= -1;
2613 info
->reshape_active
= 0;
2614 info
->array
.major_version
= -1;
2615 info
->array
.minor_version
= -2;
2616 strcpy(info
->text_version
, "imsm");
2617 info
->safe_mode_delay
= 0;
2618 info
->disk
.number
= -1;
2619 info
->disk
.state
= 0;
2621 info
->recovery_start
= MaxSector
;
2622 info
->recovery_blocked
= imsm_reshape_blocks_arrays_changes(st
->sb
);
2624 /* do we have the all the insync disks that we expect? */
2625 mpb
= super
->anchor
;
2627 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
2628 struct imsm_dev
*dev
= get_imsm_dev(super
, i
);
2629 int failed
, enough
, j
, missing
= 0;
2630 struct imsm_map
*map
;
2633 failed
= imsm_count_failed(super
, dev
, MAP_0
);
2634 state
= imsm_check_degraded(super
, dev
, failed
, MAP_0
);
2635 map
= get_imsm_map(dev
, MAP_0
);
2637 /* any newly missing disks?
2638 * (catches single-degraded vs double-degraded)
2640 for (j
= 0; j
< map
->num_members
; j
++) {
2641 __u32 ord
= get_imsm_ord_tbl_ent(dev
, j
, MAP_0
);
2642 __u32 idx
= ord_to_idx(ord
);
2644 if (!(ord
& IMSM_ORD_REBUILD
) &&
2645 get_imsm_missing(super
, idx
)) {
2651 if (state
== IMSM_T_STATE_FAILED
)
2653 else if (state
== IMSM_T_STATE_DEGRADED
&&
2654 (state
!= map
->map_state
|| missing
))
2656 else /* we're normal, or already degraded */
2659 /* in the missing/failed disk case check to see
2660 * if at least one array is runnable
2662 max_enough
= max(max_enough
, enough
);
2664 dprintf("%s: enough: %d\n", __func__
, max_enough
);
2665 info
->container_enough
= max_enough
;
2668 __u32 reserved
= imsm_reserved_sectors(super
, super
->disks
);
2670 disk
= &super
->disks
->disk
;
2671 info
->data_offset
= __le32_to_cpu(disk
->total_blocks
) - reserved
;
2672 info
->component_size
= reserved
;
2673 info
->disk
.state
= is_configured(disk
) ? (1 << MD_DISK_ACTIVE
) : 0;
2674 /* we don't change info->disk.raid_disk here because
2675 * this state will be finalized in mdmon after we have
2676 * found the 'most fresh' version of the metadata
2678 info
->disk
.state
|= is_failed(disk
) ? (1 << MD_DISK_FAULTY
) : 0;
2679 info
->disk
.state
|= is_spare(disk
) ? 0 : (1 << MD_DISK_SYNC
);
2682 /* only call uuid_from_super_imsm when this disk is part of a populated container,
2683 * ->compare_super may have updated the 'num_raid_devs' field for spares
2685 if (info
->disk
.state
& (1 << MD_DISK_SYNC
) || super
->anchor
->num_raid_devs
)
2686 uuid_from_super_imsm(st
, info
->uuid
);
2688 memcpy(info
->uuid
, uuid_zero
, sizeof(uuid_zero
));
2690 /* I don't know how to compute 'map' on imsm, so use safe default */
2693 for (i
= 0; i
< map_disks
; i
++)
2699 /* allocates memory and fills disk in mdinfo structure
2700 * for each disk in array */
2701 struct mdinfo
*getinfo_super_disks_imsm(struct supertype
*st
)
2703 struct mdinfo
*mddev
= NULL
;
2704 struct intel_super
*super
= st
->sb
;
2705 struct imsm_disk
*disk
;
2708 if (!super
|| !super
->disks
)
2711 mddev
= malloc(sizeof(*mddev
));
2713 fprintf(stderr
, Name
": Failed to allocate memory.\n");
2716 memset(mddev
, 0, sizeof(*mddev
));
2720 tmp
= malloc(sizeof(*tmp
));
2722 fprintf(stderr
, Name
": Failed to allocate memory.\n");
2727 memset(tmp
, 0, sizeof(*tmp
));
2729 tmp
->next
= mddev
->devs
;
2731 tmp
->disk
.number
= count
++;
2732 tmp
->disk
.major
= dl
->major
;
2733 tmp
->disk
.minor
= dl
->minor
;
2734 tmp
->disk
.state
= is_configured(disk
) ?
2735 (1 << MD_DISK_ACTIVE
) : 0;
2736 tmp
->disk
.state
|= is_failed(disk
) ? (1 << MD_DISK_FAULTY
) : 0;
2737 tmp
->disk
.state
|= is_spare(disk
) ? 0 : (1 << MD_DISK_SYNC
);
2738 tmp
->disk
.raid_disk
= -1;
2744 static int update_super_imsm(struct supertype
*st
, struct mdinfo
*info
,
2745 char *update
, char *devname
, int verbose
,
2746 int uuid_set
, char *homehost
)
2748 /* For 'assemble' and 'force' we need to return non-zero if any
2749 * change was made. For others, the return value is ignored.
2750 * Update options are:
2751 * force-one : This device looks a bit old but needs to be included,
2752 * update age info appropriately.
2753 * assemble: clear any 'faulty' flag to allow this device to
2755 * force-array: Array is degraded but being forced, mark it clean
2756 * if that will be needed to assemble it.
2758 * newdev: not used ????
2759 * grow: Array has gained a new device - this is currently for
2761 * resync: mark as dirty so a resync will happen.
2762 * name: update the name - preserving the homehost
2763 * uuid: Change the uuid of the array to match watch is given
2765 * Following are not relevant for this imsm:
2766 * sparc2.2 : update from old dodgey metadata
2767 * super-minor: change the preferred_minor number
2768 * summaries: update redundant counters.
2769 * homehost: update the recorded homehost
2770 * _reshape_progress: record new reshape_progress position.
2773 struct intel_super
*super
= st
->sb
;
2774 struct imsm_super
*mpb
;
2776 /* we can only update container info */
2777 if (!super
|| super
->current_vol
>= 0 || !super
->anchor
)
2780 mpb
= super
->anchor
;
2782 if (strcmp(update
, "uuid") == 0 && uuid_set
&& !info
->update_private
)
2784 else if (strcmp(update
, "uuid") == 0 && uuid_set
&& info
->update_private
) {
2785 mpb
->orig_family_num
= *((__u32
*) info
->update_private
);
2787 } else if (strcmp(update
, "uuid") == 0) {
2788 __u32
*new_family
= malloc(sizeof(*new_family
));
2790 /* update orig_family_number with the incoming random
2791 * data, report the new effective uuid, and store the
2792 * new orig_family_num for future updates.
2795 memcpy(&mpb
->orig_family_num
, info
->uuid
, sizeof(__u32
));
2796 uuid_from_super_imsm(st
, info
->uuid
);
2797 *new_family
= mpb
->orig_family_num
;
2798 info
->update_private
= new_family
;
2801 } else if (strcmp(update
, "assemble") == 0)
2806 /* successful update? recompute checksum */
2808 mpb
->check_sum
= __le32_to_cpu(__gen_imsm_checksum(mpb
));
2813 static size_t disks_to_mpb_size(int disks
)
2817 size
= sizeof(struct imsm_super
);
2818 size
+= (disks
- 1) * sizeof(struct imsm_disk
);
2819 size
+= 2 * sizeof(struct imsm_dev
);
2820 /* up to 2 maps per raid device (-2 for imsm_maps in imsm_dev */
2821 size
+= (4 - 2) * sizeof(struct imsm_map
);
2822 /* 4 possible disk_ord_tbl's */
2823 size
+= 4 * (disks
- 1) * sizeof(__u32
);
2828 static __u64
avail_size_imsm(struct supertype
*st
, __u64 devsize
)
2830 if (devsize
< (MPB_SECTOR_CNT
+ IMSM_RESERVED_SECTORS
))
2833 return devsize
- (MPB_SECTOR_CNT
+ IMSM_RESERVED_SECTORS
);
2836 static void free_devlist(struct intel_super
*super
)
2838 struct intel_dev
*dv
;
2840 while (super
->devlist
) {
2841 dv
= super
->devlist
->next
;
2842 free(super
->devlist
->dev
);
2843 free(super
->devlist
);
2844 super
->devlist
= dv
;
2848 static void imsm_copy_dev(struct imsm_dev
*dest
, struct imsm_dev
*src
)
2850 memcpy(dest
, src
, sizeof_imsm_dev(src
, 0));
2853 static int compare_super_imsm(struct supertype
*st
, struct supertype
*tst
)
2857 * 0 same, or first was empty, and second was copied
2858 * 1 second had wrong number
2860 * 3 wrong other info
2862 struct intel_super
*first
= st
->sb
;
2863 struct intel_super
*sec
= tst
->sb
;
2870 /* in platform dependent environment test if the disks
2871 * use the same Intel hba
2873 if (!check_env("IMSM_NO_PLATFORM")) {
2874 if (!first
->hba
|| !sec
->hba
||
2875 (first
->hba
->type
!= sec
->hba
->type
)) {
2877 "HBAs of devices does not match %s != %s\n",
2878 first
->hba
? get_sys_dev_type(first
->hba
->type
) : NULL
,
2879 sec
->hba
? get_sys_dev_type(sec
->hba
->type
) : NULL
);
2884 /* if an anchor does not have num_raid_devs set then it is a free
2887 if (first
->anchor
->num_raid_devs
> 0 &&
2888 sec
->anchor
->num_raid_devs
> 0) {
2889 /* Determine if these disks might ever have been
2890 * related. Further disambiguation can only take place
2891 * in load_super_imsm_all
2893 __u32 first_family
= first
->anchor
->orig_family_num
;
2894 __u32 sec_family
= sec
->anchor
->orig_family_num
;
2896 if (memcmp(first
->anchor
->sig
, sec
->anchor
->sig
,
2897 MAX_SIGNATURE_LENGTH
) != 0)
2900 if (first_family
== 0)
2901 first_family
= first
->anchor
->family_num
;
2902 if (sec_family
== 0)
2903 sec_family
= sec
->anchor
->family_num
;
2905 if (first_family
!= sec_family
)
2911 /* if 'first' is a spare promote it to a populated mpb with sec's
2914 if (first
->anchor
->num_raid_devs
== 0 &&
2915 sec
->anchor
->num_raid_devs
> 0) {
2917 struct intel_dev
*dv
;
2918 struct imsm_dev
*dev
;
2920 /* we need to copy raid device info from sec if an allocation
2921 * fails here we don't associate the spare
2923 for (i
= 0; i
< sec
->anchor
->num_raid_devs
; i
++) {
2924 dv
= malloc(sizeof(*dv
));
2927 dev
= malloc(sizeof_imsm_dev(get_imsm_dev(sec
, i
), 1));
2934 dv
->next
= first
->devlist
;
2935 first
->devlist
= dv
;
2937 if (i
< sec
->anchor
->num_raid_devs
) {
2938 /* allocation failure */
2939 free_devlist(first
);
2940 fprintf(stderr
, "imsm: failed to associate spare\n");
2943 first
->anchor
->num_raid_devs
= sec
->anchor
->num_raid_devs
;
2944 first
->anchor
->orig_family_num
= sec
->anchor
->orig_family_num
;
2945 first
->anchor
->family_num
= sec
->anchor
->family_num
;
2946 memcpy(first
->anchor
->sig
, sec
->anchor
->sig
, MAX_SIGNATURE_LENGTH
);
2947 for (i
= 0; i
< sec
->anchor
->num_raid_devs
; i
++)
2948 imsm_copy_dev(get_imsm_dev(first
, i
), get_imsm_dev(sec
, i
));
2954 static void fd2devname(int fd
, char *name
)
2958 char dname
[PATH_MAX
];
2963 if (fstat(fd
, &st
) != 0)
2965 sprintf(path
, "/sys/dev/block/%d:%d",
2966 major(st
.st_rdev
), minor(st
.st_rdev
));
2968 rv
= readlink(path
, dname
, sizeof(dname
)-1);
2973 nm
= strrchr(dname
, '/');
2976 snprintf(name
, MAX_RAID_SERIAL_LEN
, "/dev/%s", nm
);
2980 extern int scsi_get_serial(int fd
, void *buf
, size_t buf_len
);
2982 static int imsm_read_serial(int fd
, char *devname
,
2983 __u8 serial
[MAX_RAID_SERIAL_LEN
])
2985 unsigned char scsi_serial
[255];
2994 memset(scsi_serial
, 0, sizeof(scsi_serial
));
2996 rv
= scsi_get_serial(fd
, scsi_serial
, sizeof(scsi_serial
));
2998 if (rv
&& check_env("IMSM_DEVNAME_AS_SERIAL")) {
2999 memset(serial
, 0, MAX_RAID_SERIAL_LEN
);
3000 fd2devname(fd
, (char *) serial
);
3007 Name
": Failed to retrieve serial for %s\n",
3012 rsp_len
= scsi_serial
[3];
3016 Name
": Failed to retrieve serial for %s\n",
3020 rsp_buf
= (char *) &scsi_serial
[4];
3022 /* trim all whitespace and non-printable characters and convert
3025 for (i
= 0, dest
= rsp_buf
; i
< rsp_len
; i
++) {
3028 /* ':' is reserved for use in placeholder serial
3029 * numbers for missing disks
3037 len
= dest
- rsp_buf
;
3040 /* truncate leading characters */
3041 if (len
> MAX_RAID_SERIAL_LEN
) {
3042 dest
+= len
- MAX_RAID_SERIAL_LEN
;
3043 len
= MAX_RAID_SERIAL_LEN
;
3046 memset(serial
, 0, MAX_RAID_SERIAL_LEN
);
3047 memcpy(serial
, dest
, len
);
3052 static int serialcmp(__u8
*s1
, __u8
*s2
)
3054 return strncmp((char *) s1
, (char *) s2
, MAX_RAID_SERIAL_LEN
);
3057 static void serialcpy(__u8
*dest
, __u8
*src
)
3059 strncpy((char *) dest
, (char *) src
, MAX_RAID_SERIAL_LEN
);
3062 static struct dl
*serial_to_dl(__u8
*serial
, struct intel_super
*super
)
3066 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
3067 if (serialcmp(dl
->serial
, serial
) == 0)
3073 static struct imsm_disk
*
3074 __serial_to_disk(__u8
*serial
, struct imsm_super
*mpb
, int *idx
)
3078 for (i
= 0; i
< mpb
->num_disks
; i
++) {
3079 struct imsm_disk
*disk
= __get_imsm_disk(mpb
, i
);
3081 if (serialcmp(disk
->serial
, serial
) == 0) {
3092 load_imsm_disk(int fd
, struct intel_super
*super
, char *devname
, int keep_fd
)
3094 struct imsm_disk
*disk
;
3099 __u8 serial
[MAX_RAID_SERIAL_LEN
];
3101 rv
= imsm_read_serial(fd
, devname
, serial
);
3106 dl
= calloc(1, sizeof(*dl
));
3110 Name
": failed to allocate disk buffer for %s\n",
3116 dl
->major
= major(stb
.st_rdev
);
3117 dl
->minor
= minor(stb
.st_rdev
);
3118 dl
->next
= super
->disks
;
3119 dl
->fd
= keep_fd
? fd
: -1;
3120 assert(super
->disks
== NULL
);
3122 serialcpy(dl
->serial
, serial
);
3125 fd2devname(fd
, name
);
3127 dl
->devname
= strdup(devname
);
3129 dl
->devname
= strdup(name
);
3131 /* look up this disk's index in the current anchor */
3132 disk
= __serial_to_disk(dl
->serial
, super
->anchor
, &dl
->index
);
3135 /* only set index on disks that are a member of a
3136 * populated contianer, i.e. one with raid_devs
3138 if (is_failed(&dl
->disk
))
3140 else if (is_spare(&dl
->disk
))
3148 /* When migrating map0 contains the 'destination' state while map1
3149 * contains the current state. When not migrating map0 contains the
3150 * current state. This routine assumes that map[0].map_state is set to
3151 * the current array state before being called.
3153 * Migration is indicated by one of the following states
3154 * 1/ Idle (migr_state=0 map0state=normal||unitialized||degraded||failed)
3155 * 2/ Initialize (migr_state=1 migr_type=MIGR_INIT map0state=normal
3156 * map1state=unitialized)
3157 * 3/ Repair (Resync) (migr_state=1 migr_type=MIGR_REPAIR map0state=normal
3159 * 4/ Rebuild (migr_state=1 migr_type=MIGR_REBUILD map0state=normal
3160 * map1state=degraded)
3161 * 5/ Migration (mig_state=1 migr_type=MIGR_GEN_MIGR map0state=normal
3164 static void migrate(struct imsm_dev
*dev
, struct intel_super
*super
,
3165 __u8 to_state
, int migr_type
)
3167 struct imsm_map
*dest
;
3168 struct imsm_map
*src
= get_imsm_map(dev
, MAP_0
);
3170 dev
->vol
.migr_state
= 1;
3171 set_migr_type(dev
, migr_type
);
3172 dev
->vol
.curr_migr_unit
= 0;
3173 dest
= get_imsm_map(dev
, MAP_1
);
3175 /* duplicate and then set the target end state in map[0] */
3176 memcpy(dest
, src
, sizeof_imsm_map(src
));
3177 if ((migr_type
== MIGR_REBUILD
) ||
3178 (migr_type
== MIGR_GEN_MIGR
)) {
3182 for (i
= 0; i
< src
->num_members
; i
++) {
3183 ord
= __le32_to_cpu(src
->disk_ord_tbl
[i
]);
3184 set_imsm_ord_tbl_ent(src
, i
, ord_to_idx(ord
));
3188 if (migr_type
== MIGR_GEN_MIGR
)
3189 /* Clear migration record */
3190 memset(super
->migr_rec
, 0, sizeof(struct migr_record
));
3192 src
->map_state
= to_state
;
3195 static void end_migration(struct imsm_dev
*dev
, struct intel_super
*super
,
3198 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
3199 struct imsm_map
*prev
= get_imsm_map(dev
, dev
->vol
.migr_state
== 0 ?
3203 /* merge any IMSM_ORD_REBUILD bits that were not successfully
3204 * completed in the last migration.
3206 * FIXME add support for raid-level-migration
3208 if ((map_state
!= map
->map_state
) && (is_gen_migration(dev
) == 0) &&
3209 (prev
->map_state
!= IMSM_T_STATE_UNINITIALIZED
)) {
3210 /* when final map state is other than expected
3211 * merge maps (not for migration)
3215 for (i
= 0; i
< prev
->num_members
; i
++)
3216 for (j
= 0; j
< map
->num_members
; j
++)
3217 /* during online capacity expansion
3218 * disks position can be changed
3219 * if takeover is used
3221 if (ord_to_idx(map
->disk_ord_tbl
[j
]) ==
3222 ord_to_idx(prev
->disk_ord_tbl
[i
])) {
3223 map
->disk_ord_tbl
[j
] |=
3224 prev
->disk_ord_tbl
[i
];
3227 failed
= imsm_count_failed(super
, dev
, MAP_0
);
3228 map_state
= imsm_check_degraded(super
, dev
, failed
, MAP_0
);
3231 dev
->vol
.migr_state
= 0;
3232 set_migr_type(dev
, 0);
3233 dev
->vol
.curr_migr_unit
= 0;
3234 map
->map_state
= map_state
;
3238 static int parse_raid_devices(struct intel_super
*super
)
3241 struct imsm_dev
*dev_new
;
3242 size_t len
, len_migr
;
3244 size_t space_needed
= 0;
3245 struct imsm_super
*mpb
= super
->anchor
;
3247 for (i
= 0; i
< super
->anchor
->num_raid_devs
; i
++) {
3248 struct imsm_dev
*dev_iter
= __get_imsm_dev(super
->anchor
, i
);
3249 struct intel_dev
*dv
;
3251 len
= sizeof_imsm_dev(dev_iter
, 0);
3252 len_migr
= sizeof_imsm_dev(dev_iter
, 1);
3254 space_needed
+= len_migr
- len
;
3256 dv
= malloc(sizeof(*dv
));
3259 if (max_len
< len_migr
)
3261 if (max_len
> len_migr
)
3262 space_needed
+= max_len
- len_migr
;
3263 dev_new
= malloc(max_len
);
3268 imsm_copy_dev(dev_new
, dev_iter
);
3271 dv
->next
= super
->devlist
;
3272 super
->devlist
= dv
;
3275 /* ensure that super->buf is large enough when all raid devices
3278 if (__le32_to_cpu(mpb
->mpb_size
) + space_needed
> super
->len
) {
3281 len
= ROUND_UP(__le32_to_cpu(mpb
->mpb_size
) + space_needed
, 512);
3282 if (posix_memalign(&buf
, 512, len
) != 0)
3285 memcpy(buf
, super
->buf
, super
->len
);
3286 memset(buf
+ super
->len
, 0, len
- super
->len
);
3295 /* retrieve a pointer to the bbm log which starts after all raid devices */
3296 struct bbm_log
*__get_imsm_bbm_log(struct imsm_super
*mpb
)
3300 if (__le32_to_cpu(mpb
->bbm_log_size
)) {
3302 ptr
+= mpb
->mpb_size
- __le32_to_cpu(mpb
->bbm_log_size
);
3308 /*******************************************************************************
3309 * Function: check_mpb_migr_compatibility
3310 * Description: Function checks for unsupported migration features:
3311 * - migration optimization area (pba_of_lba0)
3312 * - descending reshape (ascending_migr)
3314 * super : imsm metadata information
3316 * 0 : migration is compatible
3317 * -1 : migration is not compatible
3318 ******************************************************************************/
3319 int check_mpb_migr_compatibility(struct intel_super
*super
)
3321 struct imsm_map
*map0
, *map1
;
3322 struct migr_record
*migr_rec
= super
->migr_rec
;
3325 for (i
= 0; i
< super
->anchor
->num_raid_devs
; i
++) {
3326 struct imsm_dev
*dev_iter
= __get_imsm_dev(super
->anchor
, i
);
3329 dev_iter
->vol
.migr_state
== 1 &&
3330 dev_iter
->vol
.migr_type
== MIGR_GEN_MIGR
) {
3331 /* This device is migrating */
3332 map0
= get_imsm_map(dev_iter
, MAP_0
);
3333 map1
= get_imsm_map(dev_iter
, MAP_1
);
3334 if (map0
->pba_of_lba0
!= map1
->pba_of_lba0
)
3335 /* migration optimization area was used */
3337 if (migr_rec
->ascending_migr
== 0
3338 && migr_rec
->dest_depth_per_unit
> 0)
3339 /* descending reshape not supported yet */
3346 static void __free_imsm(struct intel_super
*super
, int free_disks
);
3348 /* load_imsm_mpb - read matrix metadata
3349 * allocates super->mpb to be freed by free_imsm
3351 static int load_imsm_mpb(int fd
, struct intel_super
*super
, char *devname
)
3353 unsigned long long dsize
;
3354 unsigned long long sectors
;
3356 struct imsm_super
*anchor
;
3359 get_dev_size(fd
, NULL
, &dsize
);
3363 Name
": %s: device to small for imsm\n",
3368 if (lseek64(fd
, dsize
- (512 * 2), SEEK_SET
) < 0) {
3370 fprintf(stderr
, Name
3371 ": Cannot seek to anchor block on %s: %s\n",
3372 devname
, strerror(errno
));
3376 if (posix_memalign((void**)&anchor
, 512, 512) != 0) {
3379 Name
": Failed to allocate imsm anchor buffer"
3380 " on %s\n", devname
);
3383 if (read(fd
, anchor
, 512) != 512) {
3386 Name
": Cannot read anchor block on %s: %s\n",
3387 devname
, strerror(errno
));
3392 if (strncmp((char *) anchor
->sig
, MPB_SIGNATURE
, MPB_SIG_LEN
) != 0) {
3395 Name
": no IMSM anchor on %s\n", devname
);
3400 __free_imsm(super
, 0);
3401 /* reload capability and hba */
3403 /* capability and hba must be updated with new super allocation */
3404 find_intel_hba_capability(fd
, super
, devname
);
3405 super
->len
= ROUND_UP(anchor
->mpb_size
, 512);
3406 if (posix_memalign(&super
->buf
, 512, super
->len
) != 0) {
3409 Name
": unable to allocate %zu byte mpb buffer\n",
3414 memcpy(super
->buf
, anchor
, 512);
3416 sectors
= mpb_sectors(anchor
) - 1;
3419 if (posix_memalign(&super
->migr_rec_buf
, 512, 512) != 0) {
3420 fprintf(stderr
, Name
3421 ": %s could not allocate migr_rec buffer\n", __func__
);
3427 check_sum
= __gen_imsm_checksum(super
->anchor
);
3428 if (check_sum
!= __le32_to_cpu(super
->anchor
->check_sum
)) {
3431 Name
": IMSM checksum %x != %x on %s\n",
3433 __le32_to_cpu(super
->anchor
->check_sum
),
3441 /* read the extended mpb */
3442 if (lseek64(fd
, dsize
- (512 * (2 + sectors
)), SEEK_SET
) < 0) {
3445 Name
": Cannot seek to extended mpb on %s: %s\n",
3446 devname
, strerror(errno
));
3450 if ((unsigned)read(fd
, super
->buf
+ 512, super
->len
- 512) != super
->len
- 512) {
3453 Name
": Cannot read extended mpb on %s: %s\n",
3454 devname
, strerror(errno
));
3458 check_sum
= __gen_imsm_checksum(super
->anchor
);
3459 if (check_sum
!= __le32_to_cpu(super
->anchor
->check_sum
)) {
3462 Name
": IMSM checksum %x != %x on %s\n",
3463 check_sum
, __le32_to_cpu(super
->anchor
->check_sum
),
3468 /* FIXME the BBM log is disk specific so we cannot use this global
3469 * buffer for all disks. Ok for now since we only look at the global
3470 * bbm_log_size parameter to gate assembly
3472 super
->bbm_log
= __get_imsm_bbm_log(super
->anchor
);
3477 static int read_imsm_migr_rec(int fd
, struct intel_super
*super
);
3480 load_and_parse_mpb(int fd
, struct intel_super
*super
, char *devname
, int keep_fd
)
3484 err
= load_imsm_mpb(fd
, super
, devname
);
3487 err
= load_imsm_disk(fd
, super
, devname
, keep_fd
);
3490 err
= parse_raid_devices(super
);
3495 static void __free_imsm_disk(struct dl
*d
)
3507 static void free_imsm_disks(struct intel_super
*super
)
3511 while (super
->disks
) {
3513 super
->disks
= d
->next
;
3514 __free_imsm_disk(d
);
3516 while (super
->disk_mgmt_list
) {
3517 d
= super
->disk_mgmt_list
;
3518 super
->disk_mgmt_list
= d
->next
;
3519 __free_imsm_disk(d
);
3521 while (super
->missing
) {
3523 super
->missing
= d
->next
;
3524 __free_imsm_disk(d
);
3529 /* free all the pieces hanging off of a super pointer */
3530 static void __free_imsm(struct intel_super
*super
, int free_disks
)
3532 struct intel_hba
*elem
, *next
;
3538 /* unlink capability description */
3540 if (super
->migr_rec_buf
) {
3541 free(super
->migr_rec_buf
);
3542 super
->migr_rec_buf
= NULL
;
3545 free_imsm_disks(super
);
3546 free_devlist(super
);
3550 free((void *)elem
->path
);
3558 static void free_imsm(struct intel_super
*super
)
3560 __free_imsm(super
, 1);
3564 static void free_super_imsm(struct supertype
*st
)
3566 struct intel_super
*super
= st
->sb
;
3575 static struct intel_super
*alloc_super(void)
3577 struct intel_super
*super
= malloc(sizeof(*super
));
3580 memset(super
, 0, sizeof(*super
));
3581 super
->current_vol
= -1;
3582 super
->create_offset
= ~((__u32
) 0);
3588 * find and allocate hba and OROM/EFI based on valid fd of RAID component device
3590 static int find_intel_hba_capability(int fd
, struct intel_super
*super
, char *devname
)
3592 struct sys_dev
*hba_name
;
3595 if ((fd
< 0) || check_env("IMSM_NO_PLATFORM")) {
3600 hba_name
= find_disk_attached_hba(fd
, NULL
);
3604 Name
": %s is not attached to Intel(R) RAID controller.\n",
3608 rv
= attach_hba_to_super(super
, hba_name
);
3611 struct intel_hba
*hba
= super
->hba
;
3613 fprintf(stderr
, Name
": %s is attached to Intel(R) %s RAID "
3614 "controller (%s),\n"
3615 " but the container is assigned to Intel(R) "
3616 "%s RAID controller (",
3619 hba_name
->pci_id
? : "Err!",
3620 get_sys_dev_type(hba_name
->type
));
3623 fprintf(stderr
, "%s", hba
->pci_id
? : "Err!");
3625 fprintf(stderr
, ", ");
3629 fprintf(stderr
, ").\n"
3630 " Mixing devices attached to different controllers "
3631 "is not allowed.\n");
3633 free_sys_dev(&hba_name
);
3636 super
->orom
= find_imsm_capability(hba_name
->type
);
3637 free_sys_dev(&hba_name
);
3643 /* find_missing - helper routine for load_super_imsm_all that identifies
3644 * disks that have disappeared from the system. This routine relies on
3645 * the mpb being uptodate, which it is at load time.
3647 static int find_missing(struct intel_super
*super
)
3650 struct imsm_super
*mpb
= super
->anchor
;
3652 struct imsm_disk
*disk
;
3654 for (i
= 0; i
< mpb
->num_disks
; i
++) {
3655 disk
= __get_imsm_disk(mpb
, i
);
3656 dl
= serial_to_dl(disk
->serial
, super
);
3660 dl
= malloc(sizeof(*dl
));
3666 dl
->devname
= strdup("missing");
3668 serialcpy(dl
->serial
, disk
->serial
);
3671 dl
->next
= super
->missing
;
3672 super
->missing
= dl
;
3679 static struct intel_disk
*disk_list_get(__u8
*serial
, struct intel_disk
*disk_list
)
3681 struct intel_disk
*idisk
= disk_list
;
3684 if (serialcmp(idisk
->disk
.serial
, serial
) == 0)
3686 idisk
= idisk
->next
;
3692 static int __prep_thunderdome(struct intel_super
**table
, int tbl_size
,
3693 struct intel_super
*super
,
3694 struct intel_disk
**disk_list
)
3696 struct imsm_disk
*d
= &super
->disks
->disk
;
3697 struct imsm_super
*mpb
= super
->anchor
;
3700 for (i
= 0; i
< tbl_size
; i
++) {
3701 struct imsm_super
*tbl_mpb
= table
[i
]->anchor
;
3702 struct imsm_disk
*tbl_d
= &table
[i
]->disks
->disk
;
3704 if (tbl_mpb
->family_num
== mpb
->family_num
) {
3705 if (tbl_mpb
->check_sum
== mpb
->check_sum
) {
3706 dprintf("%s: mpb from %d:%d matches %d:%d\n",
3707 __func__
, super
->disks
->major
,
3708 super
->disks
->minor
,
3709 table
[i
]->disks
->major
,
3710 table
[i
]->disks
->minor
);
3714 if (((is_configured(d
) && !is_configured(tbl_d
)) ||
3715 is_configured(d
) == is_configured(tbl_d
)) &&
3716 tbl_mpb
->generation_num
< mpb
->generation_num
) {
3717 /* current version of the mpb is a
3718 * better candidate than the one in
3719 * super_table, but copy over "cross
3720 * generational" status
3722 struct intel_disk
*idisk
;
3724 dprintf("%s: mpb from %d:%d replaces %d:%d\n",
3725 __func__
, super
->disks
->major
,
3726 super
->disks
->minor
,
3727 table
[i
]->disks
->major
,
3728 table
[i
]->disks
->minor
);
3730 idisk
= disk_list_get(tbl_d
->serial
, *disk_list
);
3731 if (idisk
&& is_failed(&idisk
->disk
))
3732 tbl_d
->status
|= FAILED_DISK
;
3735 struct intel_disk
*idisk
;
3736 struct imsm_disk
*disk
;
3738 /* tbl_mpb is more up to date, but copy
3739 * over cross generational status before
3742 disk
= __serial_to_disk(d
->serial
, mpb
, NULL
);
3743 if (disk
&& is_failed(disk
))
3744 d
->status
|= FAILED_DISK
;
3746 idisk
= disk_list_get(d
->serial
, *disk_list
);
3749 if (disk
&& is_configured(disk
))
3750 idisk
->disk
.status
|= CONFIGURED_DISK
;
3753 dprintf("%s: mpb from %d:%d prefer %d:%d\n",
3754 __func__
, super
->disks
->major
,
3755 super
->disks
->minor
,
3756 table
[i
]->disks
->major
,
3757 table
[i
]->disks
->minor
);
3765 table
[tbl_size
++] = super
;
3769 /* update/extend the merged list of imsm_disk records */
3770 for (j
= 0; j
< mpb
->num_disks
; j
++) {
3771 struct imsm_disk
*disk
= __get_imsm_disk(mpb
, j
);
3772 struct intel_disk
*idisk
;
3774 idisk
= disk_list_get(disk
->serial
, *disk_list
);
3776 idisk
->disk
.status
|= disk
->status
;
3777 if (is_configured(&idisk
->disk
) ||
3778 is_failed(&idisk
->disk
))
3779 idisk
->disk
.status
&= ~(SPARE_DISK
);
3781 idisk
= calloc(1, sizeof(*idisk
));
3784 idisk
->owner
= IMSM_UNKNOWN_OWNER
;
3785 idisk
->disk
= *disk
;
3786 idisk
->next
= *disk_list
;
3790 if (serialcmp(idisk
->disk
.serial
, d
->serial
) == 0)
3797 static struct intel_super
*
3798 validate_members(struct intel_super
*super
, struct intel_disk
*disk_list
,
3801 struct imsm_super
*mpb
= super
->anchor
;
3805 for (i
= 0; i
< mpb
->num_disks
; i
++) {
3806 struct imsm_disk
*disk
= __get_imsm_disk(mpb
, i
);
3807 struct intel_disk
*idisk
;
3809 idisk
= disk_list_get(disk
->serial
, disk_list
);
3811 if (idisk
->owner
== owner
||
3812 idisk
->owner
== IMSM_UNKNOWN_OWNER
)
3815 dprintf("%s: '%.16s' owner %d != %d\n",
3816 __func__
, disk
->serial
, idisk
->owner
,
3819 dprintf("%s: unknown disk %x [%d]: %.16s\n",
3820 __func__
, __le32_to_cpu(mpb
->family_num
), i
,
3826 if (ok_count
== mpb
->num_disks
)
3831 static void show_conflicts(__u32 family_num
, struct intel_super
*super_list
)
3833 struct intel_super
*s
;
3835 for (s
= super_list
; s
; s
= s
->next
) {
3836 if (family_num
!= s
->anchor
->family_num
)
3838 fprintf(stderr
, "Conflict, offlining family %#x on '%s'\n",
3839 __le32_to_cpu(family_num
), s
->disks
->devname
);
3843 static struct intel_super
*
3844 imsm_thunderdome(struct intel_super
**super_list
, int len
)
3846 struct intel_super
*super_table
[len
];
3847 struct intel_disk
*disk_list
= NULL
;
3848 struct intel_super
*champion
, *spare
;
3849 struct intel_super
*s
, **del
;
3854 memset(super_table
, 0, sizeof(super_table
));
3855 for (s
= *super_list
; s
; s
= s
->next
)
3856 tbl_size
= __prep_thunderdome(super_table
, tbl_size
, s
, &disk_list
);
3858 for (i
= 0; i
< tbl_size
; i
++) {
3859 struct imsm_disk
*d
;
3860 struct intel_disk
*idisk
;
3861 struct imsm_super
*mpb
= super_table
[i
]->anchor
;
3864 d
= &s
->disks
->disk
;
3866 /* 'd' must appear in merged disk list for its
3867 * configuration to be valid
3869 idisk
= disk_list_get(d
->serial
, disk_list
);
3870 if (idisk
&& idisk
->owner
== i
)
3871 s
= validate_members(s
, disk_list
, i
);
3876 dprintf("%s: marking family: %#x from %d:%d offline\n",
3877 __func__
, mpb
->family_num
,
3878 super_table
[i
]->disks
->major
,
3879 super_table
[i
]->disks
->minor
);
3883 /* This is where the mdadm implementation differs from the Windows
3884 * driver which has no strict concept of a container. We can only
3885 * assemble one family from a container, so when returning a prodigal
3886 * array member to this system the code will not be able to disambiguate
3887 * the container contents that should be assembled ("foreign" versus
3888 * "local"). It requires user intervention to set the orig_family_num
3889 * to a new value to establish a new container. The Windows driver in
3890 * this situation fixes up the volume name in place and manages the
3891 * foreign array as an independent entity.
3896 for (i
= 0; i
< tbl_size
; i
++) {
3897 struct intel_super
*tbl_ent
= super_table
[i
];
3903 if (tbl_ent
->anchor
->num_raid_devs
== 0) {
3908 if (s
&& !is_spare
) {
3909 show_conflicts(tbl_ent
->anchor
->family_num
, *super_list
);
3911 } else if (!s
&& !is_spare
)
3924 fprintf(stderr
, "Chose family %#x on '%s', "
3925 "assemble conflicts to new container with '--update=uuid'\n",
3926 __le32_to_cpu(s
->anchor
->family_num
), s
->disks
->devname
);
3928 /* collect all dl's onto 'champion', and update them to
3929 * champion's version of the status
3931 for (s
= *super_list
; s
; s
= s
->next
) {
3932 struct imsm_super
*mpb
= champion
->anchor
;
3933 struct dl
*dl
= s
->disks
;
3938 for (i
= 0; i
< mpb
->num_disks
; i
++) {
3939 struct imsm_disk
*disk
;
3941 disk
= __serial_to_disk(dl
->serial
, mpb
, &dl
->index
);
3944 /* only set index on disks that are a member of
3945 * a populated contianer, i.e. one with
3948 if (is_failed(&dl
->disk
))
3950 else if (is_spare(&dl
->disk
))
3956 if (i
>= mpb
->num_disks
) {
3957 struct intel_disk
*idisk
;
3959 idisk
= disk_list_get(dl
->serial
, disk_list
);
3960 if (idisk
&& is_spare(&idisk
->disk
) &&
3961 !is_failed(&idisk
->disk
) && !is_configured(&idisk
->disk
))
3969 dl
->next
= champion
->disks
;
3970 champion
->disks
= dl
;
3974 /* delete 'champion' from super_list */
3975 for (del
= super_list
; *del
; ) {
3976 if (*del
== champion
) {
3977 *del
= (*del
)->next
;
3980 del
= &(*del
)->next
;
3982 champion
->next
= NULL
;
3986 struct intel_disk
*idisk
= disk_list
;
3988 disk_list
= disk_list
->next
;
3995 static int load_super_imsm_all(struct supertype
*st
, int fd
, void **sbp
,
3999 struct intel_super
*super_list
= NULL
;
4000 struct intel_super
*super
= NULL
;
4001 int devnum
= fd2devnum(fd
);
4007 /* check if 'fd' an opened container */
4008 sra
= sysfs_read(fd
, 0, GET_LEVEL
|GET_VERSION
|GET_DEVS
|GET_STATE
);
4012 if (sra
->array
.major_version
!= -1 ||
4013 sra
->array
.minor_version
!= -2 ||
4014 strcmp(sra
->text_version
, "imsm") != 0) {
4019 for (sd
= sra
->devs
, i
= 0; sd
; sd
= sd
->next
, i
++) {
4020 struct intel_super
*s
= alloc_super();
4028 s
->next
= super_list
;
4032 sprintf(nm
, "%d:%d", sd
->disk
.major
, sd
->disk
.minor
);
4033 dfd
= dev_open(nm
, O_RDWR
);
4037 rv
= find_intel_hba_capability(dfd
, s
, devname
);
4038 /* no orom/efi or non-intel hba of the disk */
4042 err
= load_and_parse_mpb(dfd
, s
, NULL
, 1);
4044 /* retry the load if we might have raced against mdmon */
4045 if (err
== 3 && mdmon_running(devnum
))
4046 for (retry
= 0; retry
< 3; retry
++) {
4048 err
= load_and_parse_mpb(dfd
, s
, NULL
, 1);
4056 /* all mpbs enter, maybe one leaves */
4057 super
= imsm_thunderdome(&super_list
, i
);
4063 if (find_missing(super
) != 0) {
4069 /* load migration record */
4070 err
= load_imsm_migr_rec(super
, NULL
);
4076 /* Check migration compatibility */
4077 if (check_mpb_migr_compatibility(super
) != 0) {
4078 fprintf(stderr
, Name
": Unsupported migration detected");
4080 fprintf(stderr
, " on %s\n", devname
);
4082 fprintf(stderr
, " (IMSM).\n");
4091 while (super_list
) {
4092 struct intel_super
*s
= super_list
;
4094 super_list
= super_list
->next
;
4103 st
->container_dev
= devnum
;
4104 if (err
== 0 && st
->ss
== NULL
) {
4105 st
->ss
= &super_imsm
;
4106 st
->minor_version
= 0;
4107 st
->max_devs
= IMSM_MAX_DEVICES
;
4112 static int load_container_imsm(struct supertype
*st
, int fd
, char *devname
)
4114 return load_super_imsm_all(st
, fd
, &st
->sb
, devname
);
4118 static int load_super_imsm(struct supertype
*st
, int fd
, char *devname
)
4120 struct intel_super
*super
;
4123 if (test_partition(fd
))
4124 /* IMSM not allowed on partitions */
4127 free_super_imsm(st
);
4129 super
= alloc_super();
4132 Name
": malloc of %zu failed.\n",
4136 /* Load hba and capabilities if they exist.
4137 * But do not preclude loading metadata in case capabilities or hba are
4138 * non-compliant and ignore_hw_compat is set.
4140 rv
= find_intel_hba_capability(fd
, super
, devname
);
4141 /* no orom/efi or non-intel hba of the disk */
4142 if ((rv
!= 0) && (st
->ignore_hw_compat
== 0)) {
4145 Name
": No OROM/EFI properties for %s\n", devname
);
4149 rv
= load_and_parse_mpb(fd
, super
, devname
, 0);
4154 Name
": Failed to load all information "
4155 "sections on %s\n", devname
);
4161 if (st
->ss
== NULL
) {
4162 st
->ss
= &super_imsm
;
4163 st
->minor_version
= 0;
4164 st
->max_devs
= IMSM_MAX_DEVICES
;
4167 /* load migration record */
4168 if (load_imsm_migr_rec(super
, NULL
) == 0) {
4169 /* Check for unsupported migration features */
4170 if (check_mpb_migr_compatibility(super
) != 0) {
4172 Name
": Unsupported migration detected");
4174 fprintf(stderr
, " on %s\n", devname
);
4176 fprintf(stderr
, " (IMSM).\n");
4184 static __u16
info_to_blocks_per_strip(mdu_array_info_t
*info
)
4186 if (info
->level
== 1)
4188 return info
->chunk_size
>> 9;
4191 static __u32
info_to_num_data_stripes(mdu_array_info_t
*info
, int num_domains
)
4195 num_stripes
= (info
->size
* 2) / info_to_blocks_per_strip(info
);
4196 num_stripes
/= num_domains
;
4201 static __u32
info_to_blocks_per_member(mdu_array_info_t
*info
)
4203 if (info
->level
== 1)
4204 return info
->size
* 2;
4206 return (info
->size
* 2) & ~(info_to_blocks_per_strip(info
) - 1);
4209 static void imsm_update_version_info(struct intel_super
*super
)
4211 /* update the version and attributes */
4212 struct imsm_super
*mpb
= super
->anchor
;
4214 struct imsm_dev
*dev
;
4215 struct imsm_map
*map
;
4218 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
4219 dev
= get_imsm_dev(super
, i
);
4220 map
= get_imsm_map(dev
, MAP_0
);
4221 if (__le32_to_cpu(dev
->size_high
) > 0)
4222 mpb
->attributes
|= MPB_ATTRIB_2TB
;
4224 /* FIXME detect when an array spans a port multiplier */
4226 mpb
->attributes
|= MPB_ATTRIB_PM
;
4229 if (mpb
->num_raid_devs
> 1 ||
4230 mpb
->attributes
!= MPB_ATTRIB_CHECKSUM_VERIFY
) {
4231 version
= MPB_VERSION_ATTRIBS
;
4232 switch (get_imsm_raid_level(map
)) {
4233 case 0: mpb
->attributes
|= MPB_ATTRIB_RAID0
; break;
4234 case 1: mpb
->attributes
|= MPB_ATTRIB_RAID1
; break;
4235 case 10: mpb
->attributes
|= MPB_ATTRIB_RAID10
; break;
4236 case 5: mpb
->attributes
|= MPB_ATTRIB_RAID5
; break;
4239 if (map
->num_members
>= 5)
4240 version
= MPB_VERSION_5OR6_DISK_ARRAY
;
4241 else if (dev
->status
== DEV_CLONE_N_GO
)
4242 version
= MPB_VERSION_CNG
;
4243 else if (get_imsm_raid_level(map
) == 5)
4244 version
= MPB_VERSION_RAID5
;
4245 else if (map
->num_members
>= 3)
4246 version
= MPB_VERSION_3OR4_DISK_ARRAY
;
4247 else if (get_imsm_raid_level(map
) == 1)
4248 version
= MPB_VERSION_RAID1
;
4250 version
= MPB_VERSION_RAID0
;
4252 strcpy(((char *) mpb
->sig
) + strlen(MPB_SIGNATURE
), version
);
4256 static int check_name(struct intel_super
*super
, char *name
, int quiet
)
4258 struct imsm_super
*mpb
= super
->anchor
;
4259 char *reason
= NULL
;
4262 if (strlen(name
) > MAX_RAID_SERIAL_LEN
)
4263 reason
= "must be 16 characters or less";
4265 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
4266 struct imsm_dev
*dev
= get_imsm_dev(super
, i
);
4268 if (strncmp((char *) dev
->volume
, name
, MAX_RAID_SERIAL_LEN
) == 0) {
4269 reason
= "already exists";
4274 if (reason
&& !quiet
)
4275 fprintf(stderr
, Name
": imsm volume name %s\n", reason
);
4280 static int init_super_imsm_volume(struct supertype
*st
, mdu_array_info_t
*info
,
4281 unsigned long long size
, char *name
,
4282 char *homehost
, int *uuid
)
4284 /* We are creating a volume inside a pre-existing container.
4285 * so st->sb is already set.
4287 struct intel_super
*super
= st
->sb
;
4288 struct imsm_super
*mpb
= super
->anchor
;
4289 struct intel_dev
*dv
;
4290 struct imsm_dev
*dev
;
4291 struct imsm_vol
*vol
;
4292 struct imsm_map
*map
;
4293 int idx
= mpb
->num_raid_devs
;
4295 unsigned long long array_blocks
;
4296 size_t size_old
, size_new
;
4297 __u32 num_data_stripes
;
4299 if (super
->orom
&& mpb
->num_raid_devs
>= super
->orom
->vpa
) {
4300 fprintf(stderr
, Name
": This imsm-container already has the "
4301 "maximum of %d volumes\n", super
->orom
->vpa
);
4305 /* ensure the mpb is large enough for the new data */
4306 size_old
= __le32_to_cpu(mpb
->mpb_size
);
4307 size_new
= disks_to_mpb_size(info
->nr_disks
);
4308 if (size_new
> size_old
) {
4310 size_t size_round
= ROUND_UP(size_new
, 512);
4312 if (posix_memalign(&mpb_new
, 512, size_round
) != 0) {
4313 fprintf(stderr
, Name
": could not allocate new mpb\n");
4316 if (posix_memalign(&super
->migr_rec_buf
, 512, 512) != 0) {
4317 fprintf(stderr
, Name
4318 ": %s could not allocate migr_rec buffer\n",
4325 memcpy(mpb_new
, mpb
, size_old
);
4328 super
->anchor
= mpb_new
;
4329 mpb
->mpb_size
= __cpu_to_le32(size_new
);
4330 memset(mpb_new
+ size_old
, 0, size_round
- size_old
);
4332 super
->current_vol
= idx
;
4334 /* handle 'failed_disks' by either:
4335 * a) create dummy disk entries in the table if this the first
4336 * volume in the array. We add them here as this is the only
4337 * opportunity to add them. add_to_super_imsm_volume()
4338 * handles the non-failed disks and continues incrementing
4340 * b) validate that 'failed_disks' matches the current number
4341 * of missing disks if the container is populated
4343 if (super
->current_vol
== 0) {
4345 for (i
= 0; i
< info
->failed_disks
; i
++) {
4346 struct imsm_disk
*disk
;
4349 disk
= __get_imsm_disk(mpb
, i
);
4350 disk
->status
= CONFIGURED_DISK
| FAILED_DISK
;
4351 disk
->scsi_id
= __cpu_to_le32(~(__u32
)0);
4352 snprintf((char *) disk
->serial
, MAX_RAID_SERIAL_LEN
,
4355 find_missing(super
);
4360 for (d
= super
->missing
; d
; d
= d
->next
)
4362 if (info
->failed_disks
> missing
) {
4363 fprintf(stderr
, Name
": unable to add 'missing' disk to container\n");
4368 if (!check_name(super
, name
, 0))
4370 dv
= malloc(sizeof(*dv
));
4372 fprintf(stderr
, Name
": failed to allocate device list entry\n");
4375 dev
= calloc(1, sizeof(*dev
) + sizeof(__u32
) * (info
->raid_disks
- 1));
4378 fprintf(stderr
, Name
": could not allocate raid device\n");
4382 strncpy((char *) dev
->volume
, name
, MAX_RAID_SERIAL_LEN
);
4383 if (info
->level
== 1)
4384 array_blocks
= info_to_blocks_per_member(info
);
4386 array_blocks
= calc_array_size(info
->level
, info
->raid_disks
,
4387 info
->layout
, info
->chunk_size
,
4389 /* round array size down to closest MB */
4390 array_blocks
= (array_blocks
>> SECT_PER_MB_SHIFT
) << SECT_PER_MB_SHIFT
;
4392 dev
->size_low
= __cpu_to_le32((__u32
) array_blocks
);
4393 dev
->size_high
= __cpu_to_le32((__u32
) (array_blocks
>> 32));
4394 dev
->status
= (DEV_READ_COALESCING
| DEV_WRITE_COALESCING
);
4396 vol
->migr_state
= 0;
4397 set_migr_type(dev
, MIGR_INIT
);
4398 vol
->dirty
= !info
->state
;
4399 vol
->curr_migr_unit
= 0;
4400 map
= get_imsm_map(dev
, MAP_0
);
4401 map
->pba_of_lba0
= __cpu_to_le32(super
->create_offset
);
4402 map
->blocks_per_member
= __cpu_to_le32(info_to_blocks_per_member(info
));
4403 map
->blocks_per_strip
= __cpu_to_le16(info_to_blocks_per_strip(info
));
4404 map
->failed_disk_num
= ~0;
4405 if (info
->level
> 0)
4406 map
->map_state
= IMSM_T_STATE_UNINITIALIZED
;
4408 map
->map_state
= info
->failed_disks
? IMSM_T_STATE_FAILED
:
4409 IMSM_T_STATE_NORMAL
;
4412 if (info
->level
== 1 && info
->raid_disks
> 2) {
4415 fprintf(stderr
, Name
": imsm does not support more than 2 disks"
4416 "in a raid1 volume\n");
4420 map
->raid_level
= info
->level
;
4421 if (info
->level
== 10) {
4422 map
->raid_level
= 1;
4423 map
->num_domains
= info
->raid_disks
/ 2;
4424 } else if (info
->level
== 1)
4425 map
->num_domains
= info
->raid_disks
;
4427 map
->num_domains
= 1;
4429 num_data_stripes
= info_to_num_data_stripes(info
, map
->num_domains
);
4430 map
->num_data_stripes
= __cpu_to_le32(num_data_stripes
);
4432 map
->num_members
= info
->raid_disks
;
4433 for (i
= 0; i
< map
->num_members
; i
++) {
4434 /* initialized in add_to_super */
4435 set_imsm_ord_tbl_ent(map
, i
, IMSM_ORD_REBUILD
);
4437 mpb
->num_raid_devs
++;
4440 dv
->index
= super
->current_vol
;
4441 dv
->next
= super
->devlist
;
4442 super
->devlist
= dv
;
4444 imsm_update_version_info(super
);
4449 static int init_super_imsm(struct supertype
*st
, mdu_array_info_t
*info
,
4450 unsigned long long size
, char *name
,
4451 char *homehost
, int *uuid
)
4453 /* This is primarily called by Create when creating a new array.
4454 * We will then get add_to_super called for each component, and then
4455 * write_init_super called to write it out to each device.
4456 * For IMSM, Create can create on fresh devices or on a pre-existing
4458 * To create on a pre-existing array a different method will be called.
4459 * This one is just for fresh drives.
4461 struct intel_super
*super
;
4462 struct imsm_super
*mpb
;
4467 return init_super_imsm_volume(st
, info
, size
, name
, homehost
, uuid
);
4470 mpb_size
= disks_to_mpb_size(info
->nr_disks
);
4474 super
= alloc_super();
4475 if (super
&& posix_memalign(&super
->buf
, 512, mpb_size
) != 0) {
4480 fprintf(stderr
, Name
4481 ": %s could not allocate superblock\n", __func__
);
4484 if (posix_memalign(&super
->migr_rec_buf
, 512, 512) != 0) {
4485 fprintf(stderr
, Name
4486 ": %s could not allocate migr_rec buffer\n", __func__
);
4491 memset(super
->buf
, 0, mpb_size
);
4493 mpb
->mpb_size
= __cpu_to_le32(mpb_size
);
4497 /* zeroing superblock */
4501 mpb
->attributes
= MPB_ATTRIB_CHECKSUM_VERIFY
;
4503 version
= (char *) mpb
->sig
;
4504 strcpy(version
, MPB_SIGNATURE
);
4505 version
+= strlen(MPB_SIGNATURE
);
4506 strcpy(version
, MPB_VERSION_RAID0
);
4512 static int add_to_super_imsm_volume(struct supertype
*st
, mdu_disk_info_t
*dk
,
4513 int fd
, char *devname
)
4515 struct intel_super
*super
= st
->sb
;
4516 struct imsm_super
*mpb
= super
->anchor
;
4517 struct imsm_disk
*_disk
;
4518 struct imsm_dev
*dev
;
4519 struct imsm_map
*map
;
4523 dev
= get_imsm_dev(super
, super
->current_vol
);
4524 map
= get_imsm_map(dev
, MAP_0
);
4526 if (! (dk
->state
& (1<<MD_DISK_SYNC
))) {
4527 fprintf(stderr
, Name
": %s: Cannot add spare devices to IMSM volume\n",
4533 /* we're doing autolayout so grab the pre-marked (in
4534 * validate_geometry) raid_disk
4536 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
4537 if (dl
->raiddisk
== dk
->raid_disk
)
4540 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
4541 if (dl
->major
== dk
->major
&&
4542 dl
->minor
== dk
->minor
)
4547 fprintf(stderr
, Name
": %s is not a member of the same container\n", devname
);
4551 /* add a pristine spare to the metadata */
4552 if (dl
->index
< 0) {
4553 dl
->index
= super
->anchor
->num_disks
;
4554 super
->anchor
->num_disks
++;
4556 /* Check the device has not already been added */
4557 slot
= get_imsm_disk_slot(map
, dl
->index
);
4559 (get_imsm_ord_tbl_ent(dev
, slot
, MAP_X
) & IMSM_ORD_REBUILD
) == 0) {
4560 fprintf(stderr
, Name
": %s has been included in this array twice\n",
4564 set_imsm_ord_tbl_ent(map
, dk
->raid_disk
, dl
->index
);
4565 dl
->disk
.status
= CONFIGURED_DISK
;
4567 /* update size of 'missing' disks to be at least as large as the
4568 * largest acitve member (we only have dummy missing disks when
4569 * creating the first volume)
4571 if (super
->current_vol
== 0) {
4572 for (df
= super
->missing
; df
; df
= df
->next
) {
4573 if (dl
->disk
.total_blocks
> df
->disk
.total_blocks
)
4574 df
->disk
.total_blocks
= dl
->disk
.total_blocks
;
4575 _disk
= __get_imsm_disk(mpb
, df
->index
);
4580 /* refresh unset/failed slots to point to valid 'missing' entries */
4581 for (df
= super
->missing
; df
; df
= df
->next
)
4582 for (slot
= 0; slot
< mpb
->num_disks
; slot
++) {
4583 __u32 ord
= get_imsm_ord_tbl_ent(dev
, slot
, MAP_X
);
4585 if ((ord
& IMSM_ORD_REBUILD
) == 0)
4587 set_imsm_ord_tbl_ent(map
, slot
, df
->index
| IMSM_ORD_REBUILD
);
4588 if (is_gen_migration(dev
)) {
4589 struct imsm_map
*map2
= get_imsm_map(dev
,
4591 int slot2
= get_imsm_disk_slot(map2
, df
->index
);
4592 if ((slot2
< map2
->num_members
) &&
4594 __u32 ord2
= get_imsm_ord_tbl_ent(dev
,
4597 if ((unsigned)df
->index
==
4599 set_imsm_ord_tbl_ent(map2
,
4605 dprintf("set slot:%d to missing disk:%d\n", slot
, df
->index
);
4609 /* if we are creating the first raid device update the family number */
4610 if (super
->current_vol
== 0) {
4612 struct imsm_dev
*_dev
= __get_imsm_dev(mpb
, 0);
4614 _disk
= __get_imsm_disk(mpb
, dl
->index
);
4615 if (!_dev
|| !_disk
) {
4616 fprintf(stderr
, Name
": BUG mpb setup error\n");
4622 sum
+= __gen_imsm_checksum(mpb
);
4623 mpb
->family_num
= __cpu_to_le32(sum
);
4624 mpb
->orig_family_num
= mpb
->family_num
;
4626 super
->current_disk
= dl
;
4631 * Function marks disk as spare and restores disk serial
4632 * in case it was previously marked as failed by takeover operation
4634 * -1 : critical error
4635 * 0 : disk is marked as spare but serial is not set
4638 int mark_spare(struct dl
*disk
)
4640 __u8 serial
[MAX_RAID_SERIAL_LEN
];
4647 if (!imsm_read_serial(disk
->fd
, NULL
, serial
)) {
4648 /* Restore disk serial number, because takeover marks disk
4649 * as failed and adds to serial ':0' before it becomes
4652 serialcpy(disk
->serial
, serial
);
4653 serialcpy(disk
->disk
.serial
, serial
);
4656 disk
->disk
.status
= SPARE_DISK
;
4662 static int add_to_super_imsm(struct supertype
*st
, mdu_disk_info_t
*dk
,
4663 int fd
, char *devname
)
4665 struct intel_super
*super
= st
->sb
;
4667 unsigned long long size
;
4672 /* If we are on an RAID enabled platform check that the disk is
4673 * attached to the raid controller.
4674 * We do not need to test disks attachment for container based additions,
4675 * they shall be already tested when container was created/assembled.
4677 rv
= find_intel_hba_capability(fd
, super
, devname
);
4678 /* no orom/efi or non-intel hba of the disk */
4680 dprintf("capability: %p fd: %d ret: %d\n",
4681 super
->orom
, fd
, rv
);
4685 if (super
->current_vol
>= 0)
4686 return add_to_super_imsm_volume(st
, dk
, fd
, devname
);
4689 dd
= malloc(sizeof(*dd
));
4692 Name
": malloc failed %s:%d.\n", __func__
, __LINE__
);
4695 memset(dd
, 0, sizeof(*dd
));
4696 dd
->major
= major(stb
.st_rdev
);
4697 dd
->minor
= minor(stb
.st_rdev
);
4698 dd
->devname
= devname
? strdup(devname
) : NULL
;
4701 dd
->action
= DISK_ADD
;
4702 rv
= imsm_read_serial(fd
, devname
, dd
->serial
);
4705 Name
": failed to retrieve scsi serial, aborting\n");
4710 get_dev_size(fd
, NULL
, &size
);
4712 serialcpy(dd
->disk
.serial
, dd
->serial
);
4713 dd
->disk
.total_blocks
= __cpu_to_le32(size
);
4715 if (sysfs_disk_to_scsi_id(fd
, &id
) == 0)
4716 dd
->disk
.scsi_id
= __cpu_to_le32(id
);
4718 dd
->disk
.scsi_id
= __cpu_to_le32(0);
4720 if (st
->update_tail
) {
4721 dd
->next
= super
->disk_mgmt_list
;
4722 super
->disk_mgmt_list
= dd
;
4724 dd
->next
= super
->disks
;
4726 super
->updates_pending
++;
4733 static int remove_from_super_imsm(struct supertype
*st
, mdu_disk_info_t
*dk
)
4735 struct intel_super
*super
= st
->sb
;
4738 /* remove from super works only in mdmon - for communication
4739 * manager - monitor. Check if communication memory buffer
4742 if (!st
->update_tail
) {
4744 Name
": %s shall be used in mdmon context only"
4745 "(line %d).\n", __func__
, __LINE__
);
4748 dd
= malloc(sizeof(*dd
));
4751 Name
": malloc failed %s:%d.\n", __func__
, __LINE__
);
4754 memset(dd
, 0, sizeof(*dd
));
4755 dd
->major
= dk
->major
;
4756 dd
->minor
= dk
->minor
;
4759 dd
->action
= DISK_REMOVE
;
4761 dd
->next
= super
->disk_mgmt_list
;
4762 super
->disk_mgmt_list
= dd
;
4768 static int store_imsm_mpb(int fd
, struct imsm_super
*mpb
);
4772 struct imsm_super anchor
;
4773 } spare_record
__attribute__ ((aligned(512)));
4775 /* spare records have their own family number and do not have any defined raid
4778 static int write_super_imsm_spares(struct intel_super
*super
, int doclose
)
4780 struct imsm_super
*mpb
= super
->anchor
;
4781 struct imsm_super
*spare
= &spare_record
.anchor
;
4785 spare
->mpb_size
= __cpu_to_le32(sizeof(struct imsm_super
)),
4786 spare
->generation_num
= __cpu_to_le32(1UL),
4787 spare
->attributes
= MPB_ATTRIB_CHECKSUM_VERIFY
;
4788 spare
->num_disks
= 1,
4789 spare
->num_raid_devs
= 0,
4790 spare
->cache_size
= mpb
->cache_size
,
4791 spare
->pwr_cycle_count
= __cpu_to_le32(1),
4793 snprintf((char *) spare
->sig
, MAX_SIGNATURE_LENGTH
,
4794 MPB_SIGNATURE MPB_VERSION_RAID0
);
4796 for (d
= super
->disks
; d
; d
= d
->next
) {
4800 spare
->disk
[0] = d
->disk
;
4801 sum
= __gen_imsm_checksum(spare
);
4802 spare
->family_num
= __cpu_to_le32(sum
);
4803 spare
->orig_family_num
= 0;
4804 sum
= __gen_imsm_checksum(spare
);
4805 spare
->check_sum
= __cpu_to_le32(sum
);
4807 if (store_imsm_mpb(d
->fd
, spare
)) {
4808 fprintf(stderr
, "%s: failed for device %d:%d %s\n",
4809 __func__
, d
->major
, d
->minor
, strerror(errno
));
4821 static int write_super_imsm(struct supertype
*st
, int doclose
)
4823 struct intel_super
*super
= st
->sb
;
4824 struct imsm_super
*mpb
= super
->anchor
;
4830 __u32 mpb_size
= sizeof(struct imsm_super
) - sizeof(struct imsm_disk
);
4832 int clear_migration_record
= 1;
4834 /* 'generation' is incremented everytime the metadata is written */
4835 generation
= __le32_to_cpu(mpb
->generation_num
);
4837 mpb
->generation_num
= __cpu_to_le32(generation
);
4839 /* fix up cases where previous mdadm releases failed to set
4842 if (mpb
->orig_family_num
== 0)
4843 mpb
->orig_family_num
= mpb
->family_num
;
4845 for (d
= super
->disks
; d
; d
= d
->next
) {
4849 mpb
->disk
[d
->index
] = d
->disk
;
4853 for (d
= super
->missing
; d
; d
= d
->next
) {
4854 mpb
->disk
[d
->index
] = d
->disk
;
4857 mpb
->num_disks
= num_disks
;
4858 mpb_size
+= sizeof(struct imsm_disk
) * mpb
->num_disks
;
4860 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
4861 struct imsm_dev
*dev
= __get_imsm_dev(mpb
, i
);
4862 struct imsm_dev
*dev2
= get_imsm_dev(super
, i
);
4864 imsm_copy_dev(dev
, dev2
);
4865 mpb_size
+= sizeof_imsm_dev(dev
, 0);
4867 if (is_gen_migration(dev2
))
4868 clear_migration_record
= 0;
4870 mpb_size
+= __le32_to_cpu(mpb
->bbm_log_size
);
4871 mpb
->mpb_size
= __cpu_to_le32(mpb_size
);
4873 /* recalculate checksum */
4874 sum
= __gen_imsm_checksum(mpb
);
4875 mpb
->check_sum
= __cpu_to_le32(sum
);
4877 if (clear_migration_record
)
4878 memset(super
->migr_rec_buf
, 0, 512);
4880 /* write the mpb for disks that compose raid devices */
4881 for (d
= super
->disks
; d
; d
= d
->next
) {
4882 if (d
->index
< 0 || is_failed(&d
->disk
))
4884 if (store_imsm_mpb(d
->fd
, mpb
))
4885 fprintf(stderr
, "%s: failed for device %d:%d %s\n",
4886 __func__
, d
->major
, d
->minor
, strerror(errno
));
4887 if (clear_migration_record
) {
4888 unsigned long long dsize
;
4890 get_dev_size(d
->fd
, NULL
, &dsize
);
4891 if (lseek64(d
->fd
, dsize
- 512, SEEK_SET
) >= 0) {
4892 if (write(d
->fd
, super
->migr_rec_buf
, 512) != 512)
4893 perror("Write migr_rec failed");
4903 return write_super_imsm_spares(super
, doclose
);
4909 static int create_array(struct supertype
*st
, int dev_idx
)
4912 struct imsm_update_create_array
*u
;
4913 struct intel_super
*super
= st
->sb
;
4914 struct imsm_dev
*dev
= get_imsm_dev(super
, dev_idx
);
4915 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
4916 struct disk_info
*inf
;
4917 struct imsm_disk
*disk
;
4920 len
= sizeof(*u
) - sizeof(*dev
) + sizeof_imsm_dev(dev
, 0) +
4921 sizeof(*inf
) * map
->num_members
;
4924 fprintf(stderr
, "%s: failed to allocate update buffer\n",
4929 u
->type
= update_create_array
;
4930 u
->dev_idx
= dev_idx
;
4931 imsm_copy_dev(&u
->dev
, dev
);
4932 inf
= get_disk_info(u
);
4933 for (i
= 0; i
< map
->num_members
; i
++) {
4934 int idx
= get_imsm_disk_idx(dev
, i
, MAP_X
);
4936 disk
= get_imsm_disk(super
, idx
);
4937 serialcpy(inf
[i
].serial
, disk
->serial
);
4939 append_metadata_update(st
, u
, len
);
4944 static int mgmt_disk(struct supertype
*st
)
4946 struct intel_super
*super
= st
->sb
;
4948 struct imsm_update_add_remove_disk
*u
;
4950 if (!super
->disk_mgmt_list
)
4956 fprintf(stderr
, "%s: failed to allocate update buffer\n",
4961 u
->type
= update_add_remove_disk
;
4962 append_metadata_update(st
, u
, len
);
4967 static int write_init_super_imsm(struct supertype
*st
)
4969 struct intel_super
*super
= st
->sb
;
4970 int current_vol
= super
->current_vol
;
4972 /* we are done with current_vol reset it to point st at the container */
4973 super
->current_vol
= -1;
4975 if (st
->update_tail
) {
4976 /* queue the recently created array / added disk
4977 * as a metadata update */
4980 /* determine if we are creating a volume or adding a disk */
4981 if (current_vol
< 0) {
4982 /* in the mgmt (add/remove) disk case we are running
4983 * in mdmon context, so don't close fd's
4985 return mgmt_disk(st
);
4987 rv
= create_array(st
, current_vol
);
4992 for (d
= super
->disks
; d
; d
= d
->next
)
4993 Kill(d
->devname
, NULL
, 0, 1, 1);
4994 return write_super_imsm(st
, 1);
4999 static int store_super_imsm(struct supertype
*st
, int fd
)
5001 struct intel_super
*super
= st
->sb
;
5002 struct imsm_super
*mpb
= super
? super
->anchor
: NULL
;
5008 return store_imsm_mpb(fd
, mpb
);
5014 static int imsm_bbm_log_size(struct imsm_super
*mpb
)
5016 return __le32_to_cpu(mpb
->bbm_log_size
);
5020 static int validate_geometry_imsm_container(struct supertype
*st
, int level
,
5021 int layout
, int raiddisks
, int chunk
,
5022 unsigned long long size
, char *dev
,
5023 unsigned long long *freesize
,
5027 unsigned long long ldsize
;
5028 struct intel_super
*super
=NULL
;
5031 if (level
!= LEVEL_CONTAINER
)
5036 fd
= open(dev
, O_RDONLY
|O_EXCL
, 0);
5039 fprintf(stderr
, Name
": imsm: Cannot open %s: %s\n",
5040 dev
, strerror(errno
));
5043 if (!get_dev_size(fd
, dev
, &ldsize
)) {
5048 /* capabilities retrieve could be possible
5049 * note that there is no fd for the disks in array.
5051 super
= alloc_super();
5054 Name
": malloc of %zu failed.\n",
5060 rv
= find_intel_hba_capability(fd
, super
, verbose
? dev
: NULL
);
5064 fd2devname(fd
, str
);
5065 dprintf("validate_geometry_imsm_container: fd: %d %s orom: %p rv: %d raiddisk: %d\n",
5066 fd
, str
, super
->orom
, rv
, raiddisks
);
5068 /* no orom/efi or non-intel hba of the disk */
5074 if (super
->orom
&& raiddisks
> super
->orom
->tds
) {
5076 fprintf(stderr
, Name
": %d exceeds maximum number of"
5077 " platform supported disks: %d\n",
5078 raiddisks
, super
->orom
->tds
);
5084 *freesize
= avail_size_imsm(st
, ldsize
>> 9);
5090 static unsigned long long find_size(struct extent
*e
, int *idx
, int num_extents
)
5092 const unsigned long long base_start
= e
[*idx
].start
;
5093 unsigned long long end
= base_start
+ e
[*idx
].size
;
5096 if (base_start
== end
)
5100 for (i
= *idx
; i
< num_extents
; i
++) {
5101 /* extend overlapping extents */
5102 if (e
[i
].start
>= base_start
&&
5103 e
[i
].start
<= end
) {
5106 if (e
[i
].start
+ e
[i
].size
> end
)
5107 end
= e
[i
].start
+ e
[i
].size
;
5108 } else if (e
[i
].start
> end
) {
5114 return end
- base_start
;
5117 static unsigned long long merge_extents(struct intel_super
*super
, int sum_extents
)
5119 /* build a composite disk with all known extents and generate a new
5120 * 'maxsize' given the "all disks in an array must share a common start
5121 * offset" constraint
5123 struct extent
*e
= calloc(sum_extents
, sizeof(*e
));
5127 unsigned long long pos
;
5128 unsigned long long start
= 0;
5129 unsigned long long maxsize
;
5130 unsigned long reserve
;
5135 /* coalesce and sort all extents. also, check to see if we need to
5136 * reserve space between member arrays
5139 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
5142 for (i
= 0; i
< dl
->extent_cnt
; i
++)
5145 qsort(e
, sum_extents
, sizeof(*e
), cmp_extent
);
5150 while (i
< sum_extents
) {
5151 e
[j
].start
= e
[i
].start
;
5152 e
[j
].size
= find_size(e
, &i
, sum_extents
);
5154 if (e
[j
-1].size
== 0)
5163 unsigned long long esize
;
5165 esize
= e
[i
].start
- pos
;
5166 if (esize
>= maxsize
) {
5171 pos
= e
[i
].start
+ e
[i
].size
;
5173 } while (e
[i
-1].size
);
5179 /* FIXME assumes volume at offset 0 is the first volume in a
5182 if (start_extent
> 0)
5183 reserve
= IMSM_RESERVED_SECTORS
; /* gap between raid regions */
5187 if (maxsize
< reserve
)
5190 super
->create_offset
= ~((__u32
) 0);
5191 if (start
+ reserve
> super
->create_offset
)
5192 return 0; /* start overflows create_offset */
5193 super
->create_offset
= start
+ reserve
;
5195 return maxsize
- reserve
;
5198 static int is_raid_level_supported(const struct imsm_orom
*orom
, int level
, int raiddisks
)
5200 if (level
< 0 || level
== 6 || level
== 4)
5203 /* if we have an orom prevent invalid raid levels */
5206 case 0: return imsm_orom_has_raid0(orom
);
5209 return imsm_orom_has_raid1e(orom
);
5210 return imsm_orom_has_raid1(orom
) && raiddisks
== 2;
5211 case 10: return imsm_orom_has_raid10(orom
) && raiddisks
== 4;
5212 case 5: return imsm_orom_has_raid5(orom
) && raiddisks
> 2;
5215 return 1; /* not on an Intel RAID platform so anything goes */
5220 static int imsm_default_chunk(const struct imsm_orom
*orom
)
5222 /* up to 512 if the plaform supports it, otherwise the platform max.
5223 * 128 if no platform detected
5225 int fs
= max(7, orom
? fls(orom
->sss
) : 0);
5227 return min(512, (1 << fs
));
5230 #define pr_vrb(fmt, arg...) (void) (verbose && fprintf(stderr, Name fmt, ##arg))
5232 validate_geometry_imsm_orom(struct intel_super
*super
, int level
, int layout
,
5233 int raiddisks
, int *chunk
, int verbose
)
5235 /* check/set platform and metadata limits/defaults */
5236 if (super
->orom
&& raiddisks
> super
->orom
->dpa
) {
5237 pr_vrb(": platform supports a maximum of %d disks per array\n",
5242 /* capabilities of OROM tested - copied from validate_geometry_imsm_volume */
5243 if (!is_raid_level_supported(super
->orom
, level
, raiddisks
)) {
5244 pr_vrb(": platform does not support raid%d with %d disk%s\n",
5245 level
, raiddisks
, raiddisks
> 1 ? "s" : "");
5249 if (chunk
&& (*chunk
== 0 || *chunk
== UnSet
))
5250 *chunk
= imsm_default_chunk(super
->orom
);
5252 if (super
->orom
&& chunk
&& !imsm_orom_has_chunk(super
->orom
, *chunk
)) {
5253 pr_vrb(": platform does not support a chunk size of: "
5258 if (layout
!= imsm_level_to_layout(level
)) {
5260 pr_vrb(": imsm raid 5 only supports the left-asymmetric layout\n");
5261 else if (level
== 10)
5262 pr_vrb(": imsm raid 10 only supports the n2 layout\n");
5264 pr_vrb(": imsm unknown layout %#x for this raid level %d\n",
5271 /* validate_geometry_imsm_volume - lifted from validate_geometry_ddf_bvd
5272 * FIX ME add ahci details
5274 static int validate_geometry_imsm_volume(struct supertype
*st
, int level
,
5275 int layout
, int raiddisks
, int *chunk
,
5276 unsigned long long size
, char *dev
,
5277 unsigned long long *freesize
,
5281 struct intel_super
*super
= st
->sb
;
5282 struct imsm_super
*mpb
;
5284 unsigned long long pos
= 0;
5285 unsigned long long maxsize
;
5289 /* We must have the container info already read in. */
5293 mpb
= super
->anchor
;
5295 if (mpb
->num_raid_devs
> 0 && mpb
->num_disks
!= raiddisks
) {
5296 fprintf(stderr
, Name
": the option-rom requires all "
5297 "member disks to be a member of all volumes.\n");
5301 if (!validate_geometry_imsm_orom(super
, level
, layout
, raiddisks
, chunk
, verbose
)) {
5302 fprintf(stderr
, Name
": RAID gemetry validation failed. "
5303 "Cannot proceed with the action(s).\n");
5307 /* General test: make sure there is space for
5308 * 'raiddisks' device extents of size 'size' at a given
5311 unsigned long long minsize
= size
;
5312 unsigned long long start_offset
= MaxSector
;
5315 minsize
= MPB_SECTOR_CNT
+ IMSM_RESERVED_SECTORS
;
5316 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
5321 e
= get_extents(super
, dl
);
5324 unsigned long long esize
;
5325 esize
= e
[i
].start
- pos
;
5326 if (esize
>= minsize
)
5328 if (found
&& start_offset
== MaxSector
) {
5331 } else if (found
&& pos
!= start_offset
) {
5335 pos
= e
[i
].start
+ e
[i
].size
;
5337 } while (e
[i
-1].size
);
5342 if (dcnt
< raiddisks
) {
5344 fprintf(stderr
, Name
": imsm: Not enough "
5345 "devices with space for this array "
5353 /* This device must be a member of the set */
5354 if (stat(dev
, &stb
) < 0)
5356 if ((S_IFMT
& stb
.st_mode
) != S_IFBLK
)
5358 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
5359 if (dl
->major
== (int)major(stb
.st_rdev
) &&
5360 dl
->minor
== (int)minor(stb
.st_rdev
))
5365 fprintf(stderr
, Name
": %s is not in the "
5366 "same imsm set\n", dev
);
5368 } else if (super
->orom
&& dl
->index
< 0 && mpb
->num_raid_devs
) {
5369 /* If a volume is present then the current creation attempt
5370 * cannot incorporate new spares because the orom may not
5371 * understand this configuration (all member disks must be
5372 * members of each array in the container).
5374 fprintf(stderr
, Name
": %s is a spare and a volume"
5375 " is already defined for this container\n", dev
);
5376 fprintf(stderr
, Name
": The option-rom requires all member"
5377 " disks to be a member of all volumes\n");
5381 /* retrieve the largest free space block */
5382 e
= get_extents(super
, dl
);
5387 unsigned long long esize
;
5389 esize
= e
[i
].start
- pos
;
5390 if (esize
>= maxsize
)
5392 pos
= e
[i
].start
+ e
[i
].size
;
5394 } while (e
[i
-1].size
);
5399 fprintf(stderr
, Name
": unable to determine free space for: %s\n",
5403 if (maxsize
< size
) {
5405 fprintf(stderr
, Name
": %s not enough space (%llu < %llu)\n",
5406 dev
, maxsize
, size
);
5410 /* count total number of extents for merge */
5412 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
5414 i
+= dl
->extent_cnt
;
5416 maxsize
= merge_extents(super
, i
);
5418 if (!check_env("IMSM_NO_PLATFORM") &&
5419 mpb
->num_raid_devs
> 0 && size
&& size
!= maxsize
) {
5420 fprintf(stderr
, Name
": attempting to create a second "
5421 "volume with size less then remaining space. "
5426 if (maxsize
< size
|| maxsize
== 0) {
5428 fprintf(stderr
, Name
": not enough space after merge (%llu < %llu)\n",
5433 *freesize
= maxsize
;
5438 static int reserve_space(struct supertype
*st
, int raiddisks
,
5439 unsigned long long size
, int chunk
,
5440 unsigned long long *freesize
)
5442 struct intel_super
*super
= st
->sb
;
5443 struct imsm_super
*mpb
= super
->anchor
;
5448 unsigned long long maxsize
;
5449 unsigned long long minsize
;
5453 /* find the largest common start free region of the possible disks */
5457 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
5463 /* don't activate new spares if we are orom constrained
5464 * and there is already a volume active in the container
5466 if (super
->orom
&& dl
->index
< 0 && mpb
->num_raid_devs
)
5469 e
= get_extents(super
, dl
);
5472 for (i
= 1; e
[i
-1].size
; i
++)
5480 maxsize
= merge_extents(super
, extent_cnt
);
5484 minsize
= chunk
* 2;
5486 if (cnt
< raiddisks
||
5487 (super
->orom
&& used
&& used
!= raiddisks
) ||
5488 maxsize
< minsize
||
5490 fprintf(stderr
, Name
": not enough devices with space to create array.\n");
5491 return 0; /* No enough free spaces large enough */
5503 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
5505 dl
->raiddisk
= cnt
++;
5512 static int validate_geometry_imsm(struct supertype
*st
, int level
, int layout
,
5513 int raiddisks
, int *chunk
, unsigned long long size
,
5514 char *dev
, unsigned long long *freesize
,
5522 * if given unused devices create a container
5523 * if given given devices in a container create a member volume
5525 if (level
== LEVEL_CONTAINER
) {
5526 /* Must be a fresh device to add to a container */
5527 return validate_geometry_imsm_container(st
, level
, layout
,
5529 chunk
?*chunk
:0, size
,
5536 if (!validate_geometry_imsm_orom(st
->sb
, level
, layout
,
5540 /* we are being asked to automatically layout a
5541 * new volume based on the current contents of
5542 * the container. If the the parameters can be
5543 * satisfied reserve_space will record the disks,
5544 * start offset, and size of the volume to be
5545 * created. add_to_super and getinfo_super
5546 * detect when autolayout is in progress.
5549 return reserve_space(st
, raiddisks
, size
,
5550 chunk
?*chunk
:0, freesize
);
5555 /* creating in a given container */
5556 return validate_geometry_imsm_volume(st
, level
, layout
,
5557 raiddisks
, chunk
, size
,
5558 dev
, freesize
, verbose
);
5561 /* This device needs to be a device in an 'imsm' container */
5562 fd
= open(dev
, O_RDONLY
|O_EXCL
, 0);
5566 Name
": Cannot create this array on device %s\n",
5571 if (errno
!= EBUSY
|| (fd
= open(dev
, O_RDONLY
, 0)) < 0) {
5573 fprintf(stderr
, Name
": Cannot open %s: %s\n",
5574 dev
, strerror(errno
));
5577 /* Well, it is in use by someone, maybe an 'imsm' container. */
5578 cfd
= open_container(fd
);
5582 fprintf(stderr
, Name
": Cannot use %s: It is busy\n",
5586 sra
= sysfs_read(cfd
, 0, GET_VERSION
);
5587 if (sra
&& sra
->array
.major_version
== -1 &&
5588 strcmp(sra
->text_version
, "imsm") == 0)
5592 /* This is a member of a imsm container. Load the container
5593 * and try to create a volume
5595 struct intel_super
*super
;
5597 if (load_super_imsm_all(st
, cfd
, (void **) &super
, NULL
) == 0) {
5599 st
->container_dev
= fd2devnum(cfd
);
5601 return validate_geometry_imsm_volume(st
, level
, layout
,
5610 fprintf(stderr
, Name
": failed container membership check\n");
5616 static void default_geometry_imsm(struct supertype
*st
, int *level
, int *layout
, int *chunk
)
5618 struct intel_super
*super
= st
->sb
;
5620 if (level
&& *level
== UnSet
)
5621 *level
= LEVEL_CONTAINER
;
5623 if (level
&& layout
&& *layout
== UnSet
)
5624 *layout
= imsm_level_to_layout(*level
);
5626 if (chunk
&& (*chunk
== UnSet
|| *chunk
== 0))
5627 *chunk
= imsm_default_chunk(super
->orom
);
5630 static void handle_missing(struct intel_super
*super
, struct imsm_dev
*dev
);
5632 static int kill_subarray_imsm(struct supertype
*st
)
5634 /* remove the subarray currently referenced by ->current_vol */
5636 struct intel_dev
**dp
;
5637 struct intel_super
*super
= st
->sb
;
5638 __u8 current_vol
= super
->current_vol
;
5639 struct imsm_super
*mpb
= super
->anchor
;
5641 if (super
->current_vol
< 0)
5643 super
->current_vol
= -1; /* invalidate subarray cursor */
5645 /* block deletions that would change the uuid of active subarrays
5647 * FIXME when immutable ids are available, but note that we'll
5648 * also need to fixup the invalidated/active subarray indexes in
5651 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
5654 if (i
< current_vol
)
5656 sprintf(subarray
, "%u", i
);
5657 if (is_subarray_active(subarray
, st
->devname
)) {
5659 Name
": deleting subarray-%d would change the UUID of active subarray-%d, aborting\n",
5666 if (st
->update_tail
) {
5667 struct imsm_update_kill_array
*u
= malloc(sizeof(*u
));
5671 u
->type
= update_kill_array
;
5672 u
->dev_idx
= current_vol
;
5673 append_metadata_update(st
, u
, sizeof(*u
));
5678 for (dp
= &super
->devlist
; *dp
;)
5679 if ((*dp
)->index
== current_vol
) {
5682 handle_missing(super
, (*dp
)->dev
);
5683 if ((*dp
)->index
> current_vol
)
5688 /* no more raid devices, all active components are now spares,
5689 * but of course failed are still failed
5691 if (--mpb
->num_raid_devs
== 0) {
5694 for (d
= super
->disks
; d
; d
= d
->next
)
5699 super
->updates_pending
++;
5704 static int update_subarray_imsm(struct supertype
*st
, char *subarray
,
5705 char *update
, struct mddev_ident
*ident
)
5707 /* update the subarray currently referenced by ->current_vol */
5708 struct intel_super
*super
= st
->sb
;
5709 struct imsm_super
*mpb
= super
->anchor
;
5711 if (strcmp(update
, "name") == 0) {
5712 char *name
= ident
->name
;
5716 if (is_subarray_active(subarray
, st
->devname
)) {
5718 Name
": Unable to update name of active subarray\n");
5722 if (!check_name(super
, name
, 0))
5725 vol
= strtoul(subarray
, &ep
, 10);
5726 if (*ep
!= '\0' || vol
>= super
->anchor
->num_raid_devs
)
5729 if (st
->update_tail
) {
5730 struct imsm_update_rename_array
*u
= malloc(sizeof(*u
));
5734 u
->type
= update_rename_array
;
5736 snprintf((char *) u
->name
, MAX_RAID_SERIAL_LEN
, "%s", name
);
5737 append_metadata_update(st
, u
, sizeof(*u
));
5739 struct imsm_dev
*dev
;
5742 dev
= get_imsm_dev(super
, vol
);
5743 snprintf((char *) dev
->volume
, MAX_RAID_SERIAL_LEN
, "%s", name
);
5744 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
5745 dev
= get_imsm_dev(super
, i
);
5746 handle_missing(super
, dev
);
5748 super
->updates_pending
++;
5755 #endif /* MDASSEMBLE */
5757 static int is_gen_migration(struct imsm_dev
*dev
)
5762 if (!dev
->vol
.migr_state
)
5765 if (migr_type(dev
) == MIGR_GEN_MIGR
)
5771 static int is_rebuilding(struct imsm_dev
*dev
)
5773 struct imsm_map
*migr_map
;
5775 if (!dev
->vol
.migr_state
)
5778 if (migr_type(dev
) != MIGR_REBUILD
)
5781 migr_map
= get_imsm_map(dev
, MAP_1
);
5783 if (migr_map
->map_state
== IMSM_T_STATE_DEGRADED
)
5789 static int is_initializing(struct imsm_dev
*dev
)
5791 struct imsm_map
*migr_map
;
5793 if (!dev
->vol
.migr_state
)
5796 if (migr_type(dev
) != MIGR_INIT
)
5799 migr_map
= get_imsm_map(dev
, MAP_1
);
5801 if (migr_map
->map_state
== IMSM_T_STATE_UNINITIALIZED
)
5808 static void update_recovery_start(struct intel_super
*super
,
5809 struct imsm_dev
*dev
,
5810 struct mdinfo
*array
)
5812 struct mdinfo
*rebuild
= NULL
;
5816 if (!is_rebuilding(dev
))
5819 /* Find the rebuild target, but punt on the dual rebuild case */
5820 for (d
= array
->devs
; d
; d
= d
->next
)
5821 if (d
->recovery_start
== 0) {
5828 /* (?) none of the disks are marked with
5829 * IMSM_ORD_REBUILD, so assume they are missing and the
5830 * disk_ord_tbl was not correctly updated
5832 dprintf("%s: failed to locate out-of-sync disk\n", __func__
);
5836 units
= __le32_to_cpu(dev
->vol
.curr_migr_unit
);
5837 rebuild
->recovery_start
= units
* blocks_per_migr_unit(super
, dev
);
5841 static int recover_backup_imsm(struct supertype
*st
, struct mdinfo
*info
);
5844 static struct mdinfo
*container_content_imsm(struct supertype
*st
, char *subarray
)
5846 /* Given a container loaded by load_super_imsm_all,
5847 * extract information about all the arrays into
5849 * If 'subarray' is given, just extract info about that array.
5851 * For each imsm_dev create an mdinfo, fill it in,
5852 * then look for matching devices in super->disks
5853 * and create appropriate device mdinfo.
5855 struct intel_super
*super
= st
->sb
;
5856 struct imsm_super
*mpb
= super
->anchor
;
5857 struct mdinfo
*rest
= NULL
;
5861 int spare_disks
= 0;
5863 /* do not assemble arrays when not all attributes are supported */
5864 if (imsm_check_attributes(mpb
->attributes
) == 0) {
5866 fprintf(stderr
, Name
": Unsupported attributes in IMSM metadata."
5867 "Arrays activation is blocked.\n");
5870 /* check for bad blocks */
5871 if (imsm_bbm_log_size(super
->anchor
)) {
5872 fprintf(stderr
, Name
": BBM log found in IMSM metadata."
5873 "Arrays activation is blocked.\n");
5878 /* count spare devices, not used in maps
5880 for (d
= super
->disks
; d
; d
= d
->next
)
5884 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
5885 struct imsm_dev
*dev
;
5886 struct imsm_map
*map
;
5887 struct imsm_map
*map2
;
5888 struct mdinfo
*this;
5893 (i
!= strtoul(subarray
, &ep
, 10) || *ep
!= '\0'))
5896 dev
= get_imsm_dev(super
, i
);
5897 map
= get_imsm_map(dev
, MAP_0
);
5898 map2
= get_imsm_map(dev
, MAP_1
);
5900 /* do not publish arrays that are in the middle of an
5901 * unsupported migration
5903 if (dev
->vol
.migr_state
&&
5904 (migr_type(dev
) == MIGR_STATE_CHANGE
)) {
5905 fprintf(stderr
, Name
": cannot assemble volume '%.16s':"
5906 " unsupported migration in progress\n",
5910 /* do not publish arrays that are not support by controller's
5914 chunk
= __le16_to_cpu(map
->blocks_per_strip
) >> 1;
5915 this = malloc(sizeof(*this));
5917 fprintf(stderr
, Name
": failed to allocate %zu bytes\n",
5922 super
->current_vol
= i
;
5923 getinfo_super_imsm_volume(st
, this, NULL
);
5926 /* mdadm does not support all metadata features- set the bit in all arrays state */
5927 if (!validate_geometry_imsm_orom(super
,
5928 get_imsm_raid_level(map
), /* RAID level */
5929 imsm_level_to_layout(get_imsm_raid_level(map
)),
5930 map
->num_members
, /* raid disks */
5933 fprintf(stderr
, Name
": IMSM RAID geometry validation"
5934 " failed. Array %s activation is blocked.\n",
5936 this->array
.state
|=
5937 (1<<MD_SB_BLOCK_CONTAINER_RESHAPE
) |
5938 (1<<MD_SB_BLOCK_VOLUME
);
5942 /* if array has bad blocks, set suitable bit in all arrays state */
5944 this->array
.state
|=
5945 (1<<MD_SB_BLOCK_CONTAINER_RESHAPE
) |
5946 (1<<MD_SB_BLOCK_VOLUME
);
5948 for (slot
= 0 ; slot
< map
->num_members
; slot
++) {
5949 unsigned long long recovery_start
;
5950 struct mdinfo
*info_d
;
5957 idx
= get_imsm_disk_idx(dev
, slot
, MAP_0
);
5958 ord
= get_imsm_ord_tbl_ent(dev
, slot
, MAP_X
);
5959 for (d
= super
->disks
; d
; d
= d
->next
)
5960 if (d
->index
== idx
)
5963 recovery_start
= MaxSector
;
5966 if (d
&& is_failed(&d
->disk
))
5968 if (ord
& IMSM_ORD_REBUILD
)
5972 * if we skip some disks the array will be assmebled degraded;
5973 * reset resync start to avoid a dirty-degraded
5974 * situation when performing the intial sync
5976 * FIXME handle dirty degraded
5978 if ((skip
|| recovery_start
== 0) && !dev
->vol
.dirty
)
5979 this->resync_start
= MaxSector
;
5983 info_d
= calloc(1, sizeof(*info_d
));
5985 fprintf(stderr
, Name
": failed to allocate disk"
5986 " for volume %.16s\n", dev
->volume
);
5987 info_d
= this->devs
;
5989 struct mdinfo
*d
= info_d
->next
;
5998 info_d
->next
= this->devs
;
5999 this->devs
= info_d
;
6001 info_d
->disk
.number
= d
->index
;
6002 info_d
->disk
.major
= d
->major
;
6003 info_d
->disk
.minor
= d
->minor
;
6004 info_d
->disk
.raid_disk
= slot
;
6005 info_d
->recovery_start
= recovery_start
;
6007 if (slot
< map2
->num_members
)
6008 info_d
->disk
.state
= (1 << MD_DISK_ACTIVE
);
6010 this->array
.spare_disks
++;
6012 if (slot
< map
->num_members
)
6013 info_d
->disk
.state
= (1 << MD_DISK_ACTIVE
);
6015 this->array
.spare_disks
++;
6017 if (info_d
->recovery_start
== MaxSector
)
6018 this->array
.working_disks
++;
6020 info_d
->events
= __le32_to_cpu(mpb
->generation_num
);
6021 info_d
->data_offset
= __le32_to_cpu(map
->pba_of_lba0
);
6022 info_d
->component_size
= __le32_to_cpu(map
->blocks_per_member
);
6024 /* now that the disk list is up-to-date fixup recovery_start */
6025 update_recovery_start(super
, dev
, this);
6026 this->array
.spare_disks
+= spare_disks
;
6029 /* check for reshape */
6030 if (this->reshape_active
== 1)
6031 recover_backup_imsm(st
, this);
6040 static __u8
imsm_check_degraded(struct intel_super
*super
, struct imsm_dev
*dev
,
6041 int failed
, int look_in_map
)
6043 struct imsm_map
*map
;
6045 map
= get_imsm_map(dev
, look_in_map
);
6048 return map
->map_state
== IMSM_T_STATE_UNINITIALIZED
?
6049 IMSM_T_STATE_UNINITIALIZED
: IMSM_T_STATE_NORMAL
;
6051 switch (get_imsm_raid_level(map
)) {
6053 return IMSM_T_STATE_FAILED
;
6056 if (failed
< map
->num_members
)
6057 return IMSM_T_STATE_DEGRADED
;
6059 return IMSM_T_STATE_FAILED
;
6064 * check to see if any mirrors have failed, otherwise we
6065 * are degraded. Even numbered slots are mirrored on
6069 /* gcc -Os complains that this is unused */
6070 int insync
= insync
;
6072 for (i
= 0; i
< map
->num_members
; i
++) {
6073 __u32 ord
= get_imsm_ord_tbl_ent(dev
, i
, MAP_X
);
6074 int idx
= ord_to_idx(ord
);
6075 struct imsm_disk
*disk
;
6077 /* reset the potential in-sync count on even-numbered
6078 * slots. num_copies is always 2 for imsm raid10
6083 disk
= get_imsm_disk(super
, idx
);
6084 if (!disk
|| is_failed(disk
) || ord
& IMSM_ORD_REBUILD
)
6087 /* no in-sync disks left in this mirror the
6091 return IMSM_T_STATE_FAILED
;
6094 return IMSM_T_STATE_DEGRADED
;
6098 return IMSM_T_STATE_DEGRADED
;
6100 return IMSM_T_STATE_FAILED
;
6106 return map
->map_state
;
6109 static int imsm_count_failed(struct intel_super
*super
, struct imsm_dev
*dev
,
6114 struct imsm_disk
*disk
;
6115 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
6116 struct imsm_map
*prev
= get_imsm_map(dev
, MAP_1
);
6117 struct imsm_map
*map_for_loop
;
6122 /* at the beginning of migration we set IMSM_ORD_REBUILD on
6123 * disks that are being rebuilt. New failures are recorded to
6124 * map[0]. So we look through all the disks we started with and
6125 * see if any failures are still present, or if any new ones
6129 if (prev
&& (map
->num_members
< prev
->num_members
))
6130 map_for_loop
= prev
;
6132 for (i
= 0; i
< map_for_loop
->num_members
; i
++) {
6134 /* when MAP_X is passed both maps failures are counted
6137 ((look_in_map
== MAP_1
) || (look_in_map
== MAP_X
)) &&
6138 (i
< prev
->num_members
)) {
6139 ord
= __le32_to_cpu(prev
->disk_ord_tbl
[i
]);
6140 idx_1
= ord_to_idx(ord
);
6142 disk
= get_imsm_disk(super
, idx_1
);
6143 if (!disk
|| is_failed(disk
) || ord
& IMSM_ORD_REBUILD
)
6146 if (((look_in_map
== MAP_0
) || (look_in_map
== MAP_X
)) &&
6147 (i
< map
->num_members
)) {
6148 ord
= __le32_to_cpu(map
->disk_ord_tbl
[i
]);
6149 idx
= ord_to_idx(ord
);
6152 disk
= get_imsm_disk(super
, idx
);
6153 if (!disk
|| is_failed(disk
) ||
6154 ord
& IMSM_ORD_REBUILD
)
6164 static int imsm_open_new(struct supertype
*c
, struct active_array
*a
,
6167 struct intel_super
*super
= c
->sb
;
6168 struct imsm_super
*mpb
= super
->anchor
;
6170 if (atoi(inst
) >= mpb
->num_raid_devs
) {
6171 fprintf(stderr
, "%s: subarry index %d, out of range\n",
6172 __func__
, atoi(inst
));
6176 dprintf("imsm: open_new %s\n", inst
);
6177 a
->info
.container_member
= atoi(inst
);
6181 static int is_resyncing(struct imsm_dev
*dev
)
6183 struct imsm_map
*migr_map
;
6185 if (!dev
->vol
.migr_state
)
6188 if (migr_type(dev
) == MIGR_INIT
||
6189 migr_type(dev
) == MIGR_REPAIR
)
6192 if (migr_type(dev
) == MIGR_GEN_MIGR
)
6195 migr_map
= get_imsm_map(dev
, MAP_1
);
6197 if ((migr_map
->map_state
== IMSM_T_STATE_NORMAL
) &&
6198 (dev
->vol
.migr_type
!= MIGR_GEN_MIGR
))
6204 /* return true if we recorded new information */
6205 static int mark_failure(struct imsm_dev
*dev
, struct imsm_disk
*disk
, int idx
)
6209 struct imsm_map
*map
;
6210 char buf
[MAX_RAID_SERIAL_LEN
+3];
6211 unsigned int len
, shift
= 0;
6213 /* new failures are always set in map[0] */
6214 map
= get_imsm_map(dev
, MAP_0
);
6216 slot
= get_imsm_disk_slot(map
, idx
);
6220 ord
= __le32_to_cpu(map
->disk_ord_tbl
[slot
]);
6221 if (is_failed(disk
) && (ord
& IMSM_ORD_REBUILD
))
6224 memcpy(buf
, disk
->serial
, MAX_RAID_SERIAL_LEN
);
6225 buf
[MAX_RAID_SERIAL_LEN
] = '\000';
6227 if ((len
= strlen(buf
)) >= MAX_RAID_SERIAL_LEN
)
6228 shift
= len
- MAX_RAID_SERIAL_LEN
+ 1;
6229 strncpy((char *)disk
->serial
, &buf
[shift
], MAX_RAID_SERIAL_LEN
);
6231 disk
->status
|= FAILED_DISK
;
6232 set_imsm_ord_tbl_ent(map
, slot
, idx
| IMSM_ORD_REBUILD
);
6233 /* mark failures in second map if second map exists and this disk
6235 * This is valid for migration, initialization and rebuild
6237 if (dev
->vol
.migr_state
) {
6238 struct imsm_map
*map2
= get_imsm_map(dev
, MAP_1
);
6239 int slot2
= get_imsm_disk_slot(map2
, idx
);
6241 if ((slot2
< map2
->num_members
) &&
6243 set_imsm_ord_tbl_ent(map2
, slot2
,
6244 idx
| IMSM_ORD_REBUILD
);
6246 if (map
->failed_disk_num
== 0xff)
6247 map
->failed_disk_num
= slot
;
6251 static void mark_missing(struct imsm_dev
*dev
, struct imsm_disk
*disk
, int idx
)
6253 mark_failure(dev
, disk
, idx
);
6255 if (disk
->scsi_id
== __cpu_to_le32(~(__u32
)0))
6258 disk
->scsi_id
= __cpu_to_le32(~(__u32
)0);
6259 memmove(&disk
->serial
[0], &disk
->serial
[1], MAX_RAID_SERIAL_LEN
- 1);
6262 static void handle_missing(struct intel_super
*super
, struct imsm_dev
*dev
)
6266 if (!super
->missing
)
6269 dprintf("imsm: mark missing\n");
6270 /* end process for initialization and rebuild only
6272 if (is_gen_migration(dev
) == 0) {
6276 failed
= imsm_count_failed(super
, dev
, MAP_0
);
6277 map_state
= imsm_check_degraded(super
, dev
, failed
, MAP_0
);
6279 end_migration(dev
, super
, map_state
);
6281 for (dl
= super
->missing
; dl
; dl
= dl
->next
)
6282 mark_missing(dev
, &dl
->disk
, dl
->index
);
6283 super
->updates_pending
++;
6286 static unsigned long long imsm_set_array_size(struct imsm_dev
*dev
)
6288 int used_disks
= imsm_num_data_members(dev
, MAP_0
);
6289 unsigned long long array_blocks
;
6290 struct imsm_map
*map
;
6292 if (used_disks
== 0) {
6293 /* when problems occures
6294 * return current array_blocks value
6296 array_blocks
= __le32_to_cpu(dev
->size_high
);
6297 array_blocks
= array_blocks
<< 32;
6298 array_blocks
+= __le32_to_cpu(dev
->size_low
);
6300 return array_blocks
;
6303 /* set array size in metadata
6305 map
= get_imsm_map(dev
, MAP_0
);
6306 array_blocks
= map
->blocks_per_member
* used_disks
;
6308 /* round array size down to closest MB
6310 array_blocks
= (array_blocks
>> SECT_PER_MB_SHIFT
) << SECT_PER_MB_SHIFT
;
6311 dev
->size_low
= __cpu_to_le32((__u32
)array_blocks
);
6312 dev
->size_high
= __cpu_to_le32((__u32
)(array_blocks
>> 32));
6314 return array_blocks
;
6317 static void imsm_set_disk(struct active_array
*a
, int n
, int state
);
6319 static void imsm_progress_container_reshape(struct intel_super
*super
)
6321 /* if no device has a migr_state, but some device has a
6322 * different number of members than the previous device, start
6323 * changing the number of devices in this device to match
6326 struct imsm_super
*mpb
= super
->anchor
;
6327 int prev_disks
= -1;
6331 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
6332 struct imsm_dev
*dev
= get_imsm_dev(super
, i
);
6333 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
6334 struct imsm_map
*map2
;
6335 int prev_num_members
;
6337 if (dev
->vol
.migr_state
)
6340 if (prev_disks
== -1)
6341 prev_disks
= map
->num_members
;
6342 if (prev_disks
== map
->num_members
)
6345 /* OK, this array needs to enter reshape mode.
6346 * i.e it needs a migr_state
6349 copy_map_size
= sizeof_imsm_map(map
);
6350 prev_num_members
= map
->num_members
;
6351 map
->num_members
= prev_disks
;
6352 dev
->vol
.migr_state
= 1;
6353 dev
->vol
.curr_migr_unit
= 0;
6354 set_migr_type(dev
, MIGR_GEN_MIGR
);
6355 for (i
= prev_num_members
;
6356 i
< map
->num_members
; i
++)
6357 set_imsm_ord_tbl_ent(map
, i
, i
);
6358 map2
= get_imsm_map(dev
, MAP_1
);
6359 /* Copy the current map */
6360 memcpy(map2
, map
, copy_map_size
);
6361 map2
->num_members
= prev_num_members
;
6363 imsm_set_array_size(dev
);
6364 super
->updates_pending
++;
6368 /* Handle dirty -> clean transititions, resync and reshape. Degraded and rebuild
6369 * states are handled in imsm_set_disk() with one exception, when a
6370 * resync is stopped due to a new failure this routine will set the
6371 * 'degraded' state for the array.
6373 static int imsm_set_array_state(struct active_array
*a
, int consistent
)
6375 int inst
= a
->info
.container_member
;
6376 struct intel_super
*super
= a
->container
->sb
;
6377 struct imsm_dev
*dev
= get_imsm_dev(super
, inst
);
6378 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
6379 int failed
= imsm_count_failed(super
, dev
, MAP_0
);
6380 __u8 map_state
= imsm_check_degraded(super
, dev
, failed
, MAP_0
);
6381 __u32 blocks_per_unit
;
6383 if (dev
->vol
.migr_state
&&
6384 dev
->vol
.migr_type
== MIGR_GEN_MIGR
) {
6385 /* array state change is blocked due to reshape action
6387 * - abort the reshape (if last_checkpoint is 0 and action!= reshape)
6388 * - finish the reshape (if last_checkpoint is big and action != reshape)
6389 * - update curr_migr_unit
6391 if (a
->curr_action
== reshape
) {
6392 /* still reshaping, maybe update curr_migr_unit */
6393 goto mark_checkpoint
;
6395 if (a
->last_checkpoint
== 0 && a
->prev_action
== reshape
) {
6396 /* for some reason we aborted the reshape.
6398 * disable automatic metadata rollback
6399 * user action is required to recover process
6402 struct imsm_map
*map2
=
6403 get_imsm_map(dev
, MAP_1
);
6404 dev
->vol
.migr_state
= 0;
6405 set_migr_type(dev
, 0);
6406 dev
->vol
.curr_migr_unit
= 0;
6408 sizeof_imsm_map(map2
));
6409 super
->updates_pending
++;
6412 if (a
->last_checkpoint
>= a
->info
.component_size
) {
6413 unsigned long long array_blocks
;
6417 used_disks
= imsm_num_data_members(dev
, MAP_0
);
6418 if (used_disks
> 0) {
6420 map
->blocks_per_member
*
6422 /* round array size down to closest MB
6424 array_blocks
= (array_blocks
6425 >> SECT_PER_MB_SHIFT
)
6426 << SECT_PER_MB_SHIFT
;
6427 a
->info
.custom_array_size
= array_blocks
;
6428 /* encourage manager to update array
6432 a
->check_reshape
= 1;
6434 /* finalize online capacity expansion/reshape */
6435 for (mdi
= a
->info
.devs
; mdi
; mdi
= mdi
->next
)
6437 mdi
->disk
.raid_disk
,
6440 imsm_progress_container_reshape(super
);
6445 /* before we activate this array handle any missing disks */
6446 if (consistent
== 2)
6447 handle_missing(super
, dev
);
6449 if (consistent
== 2 &&
6450 (!is_resync_complete(&a
->info
) ||
6451 map_state
!= IMSM_T_STATE_NORMAL
||
6452 dev
->vol
.migr_state
))
6455 if (is_resync_complete(&a
->info
)) {
6456 /* complete intialization / resync,
6457 * recovery and interrupted recovery is completed in
6460 if (is_resyncing(dev
)) {
6461 dprintf("imsm: mark resync done\n");
6462 end_migration(dev
, super
, map_state
);
6463 super
->updates_pending
++;
6464 a
->last_checkpoint
= 0;
6466 } else if ((!is_resyncing(dev
) && !failed
) &&
6467 (imsm_reshape_blocks_arrays_changes(super
) == 0)) {
6468 /* mark the start of the init process if nothing is failed */
6469 dprintf("imsm: mark resync start\n");
6470 if (map
->map_state
== IMSM_T_STATE_UNINITIALIZED
)
6471 migrate(dev
, super
, IMSM_T_STATE_NORMAL
, MIGR_INIT
);
6473 migrate(dev
, super
, IMSM_T_STATE_NORMAL
, MIGR_REPAIR
);
6474 super
->updates_pending
++;
6478 /* skip checkpointing for general migration,
6479 * it is controlled in mdadm
6481 if (is_gen_migration(dev
))
6482 goto skip_mark_checkpoint
;
6484 /* check if we can update curr_migr_unit from resync_start, recovery_start */
6485 blocks_per_unit
= blocks_per_migr_unit(super
, dev
);
6486 if (blocks_per_unit
) {
6490 units
= a
->last_checkpoint
/ blocks_per_unit
;
6493 /* check that we did not overflow 32-bits, and that
6494 * curr_migr_unit needs updating
6496 if (units32
== units
&&
6498 __le32_to_cpu(dev
->vol
.curr_migr_unit
) != units32
) {
6499 dprintf("imsm: mark checkpoint (%u)\n", units32
);
6500 dev
->vol
.curr_migr_unit
= __cpu_to_le32(units32
);
6501 super
->updates_pending
++;
6505 skip_mark_checkpoint
:
6506 /* mark dirty / clean */
6507 if (dev
->vol
.dirty
!= !consistent
) {
6508 dprintf("imsm: mark '%s'\n", consistent
? "clean" : "dirty");
6513 super
->updates_pending
++;
6519 static void imsm_set_disk(struct active_array
*a
, int n
, int state
)
6521 int inst
= a
->info
.container_member
;
6522 struct intel_super
*super
= a
->container
->sb
;
6523 struct imsm_dev
*dev
= get_imsm_dev(super
, inst
);
6524 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
6525 struct imsm_disk
*disk
;
6530 if (n
> map
->num_members
)
6531 fprintf(stderr
, "imsm: set_disk %d out of range 0..%d\n",
6532 n
, map
->num_members
- 1);
6537 dprintf("imsm: set_disk %d:%x\n", n
, state
);
6539 ord
= get_imsm_ord_tbl_ent(dev
, n
, MAP_0
);
6540 disk
= get_imsm_disk(super
, ord_to_idx(ord
));
6542 /* check for new failures */
6543 if (state
& DS_FAULTY
) {
6544 if (mark_failure(dev
, disk
, ord_to_idx(ord
)))
6545 super
->updates_pending
++;
6548 /* check if in_sync */
6549 if (state
& DS_INSYNC
&& ord
& IMSM_ORD_REBUILD
&& is_rebuilding(dev
)) {
6550 struct imsm_map
*migr_map
= get_imsm_map(dev
, MAP_1
);
6552 set_imsm_ord_tbl_ent(migr_map
, n
, ord_to_idx(ord
));
6553 super
->updates_pending
++;
6556 failed
= imsm_count_failed(super
, dev
, MAP_0
);
6557 map_state
= imsm_check_degraded(super
, dev
, failed
, MAP_0
);
6559 /* check if recovery complete, newly degraded, or failed */
6560 dprintf("imsm: Detected transition to state ");
6561 switch (map_state
) {
6562 case IMSM_T_STATE_NORMAL
: /* transition to normal state */
6563 dprintf("normal: ");
6564 if (is_rebuilding(dev
)) {
6565 dprintf("while rebuilding");
6566 end_migration(dev
, super
, map_state
);
6567 map
= get_imsm_map(dev
, MAP_0
);
6568 map
->failed_disk_num
= ~0;
6569 super
->updates_pending
++;
6570 a
->last_checkpoint
= 0;
6573 if (is_gen_migration(dev
)) {
6574 dprintf("while general migration");
6575 if (a
->last_checkpoint
>= a
->info
.component_size
)
6576 end_migration(dev
, super
, map_state
);
6578 map
->map_state
= map_state
;
6579 map
= get_imsm_map(dev
, MAP_0
);
6580 map
->failed_disk_num
= ~0;
6581 super
->updates_pending
++;
6585 case IMSM_T_STATE_DEGRADED
: /* transition to degraded state */
6586 dprintf("degraded: ");
6587 if ((map
->map_state
!= map_state
) &&
6588 !dev
->vol
.migr_state
) {
6589 dprintf("mark degraded");
6590 map
->map_state
= map_state
;
6591 super
->updates_pending
++;
6592 a
->last_checkpoint
= 0;
6595 if (is_rebuilding(dev
)) {
6596 dprintf("while rebuilding.");
6597 if (map
->map_state
!= map_state
) {
6598 dprintf(" Map state change");
6599 end_migration(dev
, super
, map_state
);
6600 super
->updates_pending
++;
6604 if (is_gen_migration(dev
)) {
6605 dprintf("while general migration");
6606 if (a
->last_checkpoint
>= a
->info
.component_size
)
6607 end_migration(dev
, super
, map_state
);
6609 map
->map_state
= map_state
;
6610 manage_second_map(super
, dev
);
6612 super
->updates_pending
++;
6615 if (is_initializing(dev
)) {
6616 dprintf("while initialization.");
6617 map
->map_state
= map_state
;
6618 super
->updates_pending
++;
6622 case IMSM_T_STATE_FAILED
: /* transition to failed state */
6623 dprintf("failed: ");
6624 if (is_gen_migration(dev
)) {
6625 dprintf("while general migration");
6626 map
->map_state
= map_state
;
6627 super
->updates_pending
++;
6630 if (map
->map_state
!= map_state
) {
6631 dprintf("mark failed");
6632 end_migration(dev
, super
, map_state
);
6633 super
->updates_pending
++;
6634 a
->last_checkpoint
= 0;
6639 dprintf("state %i\n", map_state
);
6645 static int store_imsm_mpb(int fd
, struct imsm_super
*mpb
)
6648 __u32 mpb_size
= __le32_to_cpu(mpb
->mpb_size
);
6649 unsigned long long dsize
;
6650 unsigned long long sectors
;
6652 get_dev_size(fd
, NULL
, &dsize
);
6654 if (mpb_size
> 512) {
6655 /* -1 to account for anchor */
6656 sectors
= mpb_sectors(mpb
) - 1;
6658 /* write the extended mpb to the sectors preceeding the anchor */
6659 if (lseek64(fd
, dsize
- (512 * (2 + sectors
)), SEEK_SET
) < 0)
6662 if ((unsigned long long)write(fd
, buf
+ 512, 512 * sectors
)
6667 /* first block is stored on second to last sector of the disk */
6668 if (lseek64(fd
, dsize
- (512 * 2), SEEK_SET
) < 0)
6671 if (write(fd
, buf
, 512) != 512)
6677 static void imsm_sync_metadata(struct supertype
*container
)
6679 struct intel_super
*super
= container
->sb
;
6681 dprintf("sync metadata: %d\n", super
->updates_pending
);
6682 if (!super
->updates_pending
)
6685 write_super_imsm(container
, 0);
6687 super
->updates_pending
= 0;
6690 static struct dl
*imsm_readd(struct intel_super
*super
, int idx
, struct active_array
*a
)
6692 struct imsm_dev
*dev
= get_imsm_dev(super
, a
->info
.container_member
);
6693 int i
= get_imsm_disk_idx(dev
, idx
, MAP_X
);
6696 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
6700 if (dl
&& is_failed(&dl
->disk
))
6704 dprintf("%s: found %x:%x\n", __func__
, dl
->major
, dl
->minor
);
6709 static struct dl
*imsm_add_spare(struct intel_super
*super
, int slot
,
6710 struct active_array
*a
, int activate_new
,
6711 struct mdinfo
*additional_test_list
)
6713 struct imsm_dev
*dev
= get_imsm_dev(super
, a
->info
.container_member
);
6714 int idx
= get_imsm_disk_idx(dev
, slot
, MAP_X
);
6715 struct imsm_super
*mpb
= super
->anchor
;
6716 struct imsm_map
*map
;
6717 unsigned long long pos
;
6722 __u32 array_start
= 0;
6723 __u32 array_end
= 0;
6725 struct mdinfo
*test_list
;
6727 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
6728 /* If in this array, skip */
6729 for (d
= a
->info
.devs
; d
; d
= d
->next
)
6730 if (d
->state_fd
>= 0 &&
6731 d
->disk
.major
== dl
->major
&&
6732 d
->disk
.minor
== dl
->minor
) {
6733 dprintf("%x:%x already in array\n",
6734 dl
->major
, dl
->minor
);
6739 test_list
= additional_test_list
;
6741 if (test_list
->disk
.major
== dl
->major
&&
6742 test_list
->disk
.minor
== dl
->minor
) {
6743 dprintf("%x:%x already in additional test list\n",
6744 dl
->major
, dl
->minor
);
6747 test_list
= test_list
->next
;
6752 /* skip in use or failed drives */
6753 if (is_failed(&dl
->disk
) || idx
== dl
->index
||
6755 dprintf("%x:%x status (failed: %d index: %d)\n",
6756 dl
->major
, dl
->minor
, is_failed(&dl
->disk
), idx
);
6760 /* skip pure spares when we are looking for partially
6761 * assimilated drives
6763 if (dl
->index
== -1 && !activate_new
)
6766 /* Does this unused device have the requisite free space?
6767 * It needs to be able to cover all member volumes
6769 ex
= get_extents(super
, dl
);
6771 dprintf("cannot get extents\n");
6774 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
6775 dev
= get_imsm_dev(super
, i
);
6776 map
= get_imsm_map(dev
, MAP_0
);
6778 /* check if this disk is already a member of
6781 if (get_imsm_disk_slot(map
, dl
->index
) >= 0)
6787 array_start
= __le32_to_cpu(map
->pba_of_lba0
);
6788 array_end
= array_start
+
6789 __le32_to_cpu(map
->blocks_per_member
) - 1;
6792 /* check that we can start at pba_of_lba0 with
6793 * blocks_per_member of space
6795 if (array_start
>= pos
&& array_end
< ex
[j
].start
) {
6799 pos
= ex
[j
].start
+ ex
[j
].size
;
6801 } while (ex
[j
-1].size
);
6808 if (i
< mpb
->num_raid_devs
) {
6809 dprintf("%x:%x does not have %u to %u available\n",
6810 dl
->major
, dl
->minor
, array_start
, array_end
);
6821 static int imsm_rebuild_allowed(struct supertype
*cont
, int dev_idx
, int failed
)
6823 struct imsm_dev
*dev2
;
6824 struct imsm_map
*map
;
6830 dev2
= get_imsm_dev(cont
->sb
, dev_idx
);
6832 state
= imsm_check_degraded(cont
->sb
, dev2
, failed
, MAP_0
);
6833 if (state
== IMSM_T_STATE_FAILED
) {
6834 map
= get_imsm_map(dev2
, MAP_0
);
6837 for (slot
= 0; slot
< map
->num_members
; slot
++) {
6839 * Check if failed disks are deleted from intel
6840 * disk list or are marked to be deleted
6842 idx
= get_imsm_disk_idx(dev2
, slot
, MAP_X
);
6843 idisk
= get_imsm_dl_disk(cont
->sb
, idx
);
6845 * Do not rebuild the array if failed disks
6846 * from failed sub-array are not removed from
6850 is_failed(&idisk
->disk
) &&
6851 (idisk
->action
!= DISK_REMOVE
))
6859 static struct mdinfo
*imsm_activate_spare(struct active_array
*a
,
6860 struct metadata_update
**updates
)
6863 * Find a device with unused free space and use it to replace a
6864 * failed/vacant region in an array. We replace failed regions one a
6865 * array at a time. The result is that a new spare disk will be added
6866 * to the first failed array and after the monitor has finished
6867 * propagating failures the remainder will be consumed.
6869 * FIXME add a capability for mdmon to request spares from another
6873 struct intel_super
*super
= a
->container
->sb
;
6874 int inst
= a
->info
.container_member
;
6875 struct imsm_dev
*dev
= get_imsm_dev(super
, inst
);
6876 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
6877 int failed
= a
->info
.array
.raid_disks
;
6878 struct mdinfo
*rv
= NULL
;
6881 struct metadata_update
*mu
;
6883 struct imsm_update_activate_spare
*u
;
6888 for (d
= a
->info
.devs
; d
; d
= d
->next
) {
6889 if ((d
->curr_state
& DS_FAULTY
) &&
6891 /* wait for Removal to happen */
6893 if (d
->state_fd
>= 0)
6897 dprintf("imsm: activate spare: inst=%d failed=%d (%d) level=%d\n",
6898 inst
, failed
, a
->info
.array
.raid_disks
, a
->info
.array
.level
);
6900 if (imsm_reshape_blocks_arrays_changes(super
))
6903 /* Cannot activate another spare if rebuild is in progress already
6905 if (is_rebuilding(dev
)) {
6906 dprintf("imsm: No spare activation allowed. "
6907 "Rebuild in progress already.\n");
6911 if (a
->info
.array
.level
== 4)
6912 /* No repair for takeovered array
6913 * imsm doesn't support raid4
6917 if (imsm_check_degraded(super
, dev
, failed
, MAP_0
) !=
6918 IMSM_T_STATE_DEGRADED
)
6922 * If there are any failed disks check state of the other volume.
6923 * Block rebuild if the another one is failed until failed disks
6924 * are removed from container.
6927 dprintf("found failed disks in %.*s, check if there another"
6928 "failed sub-array.\n",
6929 MAX_RAID_SERIAL_LEN
, dev
->volume
);
6930 /* check if states of the other volumes allow for rebuild */
6931 for (i
= 0; i
< super
->anchor
->num_raid_devs
; i
++) {
6933 allowed
= imsm_rebuild_allowed(a
->container
,
6941 /* For each slot, if it is not working, find a spare */
6942 for (i
= 0; i
< a
->info
.array
.raid_disks
; i
++) {
6943 for (d
= a
->info
.devs
; d
; d
= d
->next
)
6944 if (d
->disk
.raid_disk
== i
)
6946 dprintf("found %d: %p %x\n", i
, d
, d
?d
->curr_state
:0);
6947 if (d
&& (d
->state_fd
>= 0))
6951 * OK, this device needs recovery. Try to re-add the
6952 * previous occupant of this slot, if this fails see if
6953 * we can continue the assimilation of a spare that was
6954 * partially assimilated, finally try to activate a new
6957 dl
= imsm_readd(super
, i
, a
);
6959 dl
= imsm_add_spare(super
, i
, a
, 0, rv
);
6961 dl
= imsm_add_spare(super
, i
, a
, 1, rv
);
6965 /* found a usable disk with enough space */
6966 di
= malloc(sizeof(*di
));
6969 memset(di
, 0, sizeof(*di
));
6971 /* dl->index will be -1 in the case we are activating a
6972 * pristine spare. imsm_process_update() will create a
6973 * new index in this case. Once a disk is found to be
6974 * failed in all member arrays it is kicked from the
6977 di
->disk
.number
= dl
->index
;
6979 /* (ab)use di->devs to store a pointer to the device
6982 di
->devs
= (struct mdinfo
*) dl
;
6984 di
->disk
.raid_disk
= i
;
6985 di
->disk
.major
= dl
->major
;
6986 di
->disk
.minor
= dl
->minor
;
6988 di
->recovery_start
= 0;
6989 di
->data_offset
= __le32_to_cpu(map
->pba_of_lba0
);
6990 di
->component_size
= a
->info
.component_size
;
6991 di
->container_member
= inst
;
6992 super
->random
= random32();
6996 dprintf("%x:%x to be %d at %llu\n", dl
->major
, dl
->minor
,
6997 i
, di
->data_offset
);
7001 /* No spares found */
7003 /* Now 'rv' has a list of devices to return.
7004 * Create a metadata_update record to update the
7005 * disk_ord_tbl for the array
7007 mu
= malloc(sizeof(*mu
));
7009 mu
->buf
= malloc(sizeof(struct imsm_update_activate_spare
) * num_spares
);
7010 if (mu
->buf
== NULL
) {
7017 struct mdinfo
*n
= rv
->next
;
7026 mu
->space_list
= NULL
;
7027 mu
->len
= sizeof(struct imsm_update_activate_spare
) * num_spares
;
7028 mu
->next
= *updates
;
7029 u
= (struct imsm_update_activate_spare
*) mu
->buf
;
7031 for (di
= rv
; di
; di
= di
->next
) {
7032 u
->type
= update_activate_spare
;
7033 u
->dl
= (struct dl
*) di
->devs
;
7035 u
->slot
= di
->disk
.raid_disk
;
7046 static int disks_overlap(struct intel_super
*super
, int idx
, struct imsm_update_create_array
*u
)
7048 struct imsm_dev
*dev
= get_imsm_dev(super
, idx
);
7049 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
7050 struct imsm_map
*new_map
= get_imsm_map(&u
->dev
, MAP_0
);
7051 struct disk_info
*inf
= get_disk_info(u
);
7052 struct imsm_disk
*disk
;
7056 for (i
= 0; i
< map
->num_members
; i
++) {
7057 disk
= get_imsm_disk(super
, get_imsm_disk_idx(dev
, i
, MAP_X
));
7058 for (j
= 0; j
< new_map
->num_members
; j
++)
7059 if (serialcmp(disk
->serial
, inf
[j
].serial
) == 0)
7067 static struct dl
*get_disk_super(struct intel_super
*super
, int major
, int minor
)
7069 struct dl
*dl
= NULL
;
7070 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
7071 if ((dl
->major
== major
) && (dl
->minor
== minor
))
7076 static int remove_disk_super(struct intel_super
*super
, int major
, int minor
)
7078 struct dl
*prev
= NULL
;
7082 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
7083 if ((dl
->major
== major
) && (dl
->minor
== minor
)) {
7086 prev
->next
= dl
->next
;
7088 super
->disks
= dl
->next
;
7090 __free_imsm_disk(dl
);
7091 dprintf("%s: removed %x:%x\n",
7092 __func__
, major
, minor
);
7100 static void imsm_delete(struct intel_super
*super
, struct dl
**dlp
, unsigned index
);
7102 static int add_remove_disk_update(struct intel_super
*super
)
7104 int check_degraded
= 0;
7105 struct dl
*disk
= NULL
;
7106 /* add/remove some spares to/from the metadata/contrainer */
7107 while (super
->disk_mgmt_list
) {
7108 struct dl
*disk_cfg
;
7110 disk_cfg
= super
->disk_mgmt_list
;
7111 super
->disk_mgmt_list
= disk_cfg
->next
;
7112 disk_cfg
->next
= NULL
;
7114 if (disk_cfg
->action
== DISK_ADD
) {
7115 disk_cfg
->next
= super
->disks
;
7116 super
->disks
= disk_cfg
;
7118 dprintf("%s: added %x:%x\n",
7119 __func__
, disk_cfg
->major
,
7121 } else if (disk_cfg
->action
== DISK_REMOVE
) {
7122 dprintf("Disk remove action processed: %x.%x\n",
7123 disk_cfg
->major
, disk_cfg
->minor
);
7124 disk
= get_disk_super(super
,
7128 /* store action status */
7129 disk
->action
= DISK_REMOVE
;
7130 /* remove spare disks only */
7131 if (disk
->index
== -1) {
7132 remove_disk_super(super
,
7137 /* release allocate disk structure */
7138 __free_imsm_disk(disk_cfg
);
7141 return check_degraded
;
7145 static int apply_reshape_migration_update(struct imsm_update_reshape_migration
*u
,
7146 struct intel_super
*super
,
7149 struct intel_dev
*id
;
7150 void **tofree
= NULL
;
7153 dprintf("apply_reshape_migration_update()\n");
7154 if ((u
->subdev
< 0) ||
7156 dprintf("imsm: Error: Wrong subdev: %i\n", u
->subdev
);
7159 if ((space_list
== NULL
) || (*space_list
== NULL
)) {
7160 dprintf("imsm: Error: Memory is not allocated\n");
7164 for (id
= super
->devlist
; id
; id
= id
->next
) {
7165 if (id
->index
== (unsigned)u
->subdev
) {
7166 struct imsm_dev
*dev
= get_imsm_dev(super
, u
->subdev
);
7167 struct imsm_map
*map
;
7168 struct imsm_dev
*new_dev
=
7169 (struct imsm_dev
*)*space_list
;
7170 struct imsm_map
*migr_map
= get_imsm_map(dev
, MAP_1
);
7172 struct dl
*new_disk
;
7174 if (new_dev
== NULL
)
7176 *space_list
= **space_list
;
7177 memcpy(new_dev
, dev
, sizeof_imsm_dev(dev
, 0));
7178 map
= get_imsm_map(new_dev
, MAP_0
);
7180 dprintf("imsm: Error: migration in progress");
7184 to_state
= map
->map_state
;
7185 if ((u
->new_level
== 5) && (map
->raid_level
== 0)) {
7187 /* this should not happen */
7188 if (u
->new_disks
[0] < 0) {
7189 map
->failed_disk_num
=
7190 map
->num_members
- 1;
7191 to_state
= IMSM_T_STATE_DEGRADED
;
7193 to_state
= IMSM_T_STATE_NORMAL
;
7195 migrate(new_dev
, super
, to_state
, MIGR_GEN_MIGR
);
7196 if (u
->new_level
> -1)
7197 map
->raid_level
= u
->new_level
;
7198 migr_map
= get_imsm_map(new_dev
, MAP_1
);
7199 if ((u
->new_level
== 5) &&
7200 (migr_map
->raid_level
== 0)) {
7201 int ord
= map
->num_members
- 1;
7202 migr_map
->num_members
--;
7203 if (u
->new_disks
[0] < 0)
7204 ord
|= IMSM_ORD_REBUILD
;
7205 set_imsm_ord_tbl_ent(map
,
7206 map
->num_members
- 1,
7210 tofree
= (void **)dev
;
7212 /* update chunk size
7214 if (u
->new_chunksize
> 0)
7215 map
->blocks_per_strip
=
7216 __cpu_to_le16(u
->new_chunksize
* 2);
7220 if ((u
->new_level
!= 5) ||
7221 (migr_map
->raid_level
!= 0) ||
7222 (migr_map
->raid_level
== map
->raid_level
))
7225 if (u
->new_disks
[0] >= 0) {
7228 new_disk
= get_disk_super(super
,
7229 major(u
->new_disks
[0]),
7230 minor(u
->new_disks
[0]));
7231 dprintf("imsm: new disk for reshape is: %i:%i "
7232 "(%p, index = %i)\n",
7233 major(u
->new_disks
[0]),
7234 minor(u
->new_disks
[0]),
7235 new_disk
, new_disk
->index
);
7236 if (new_disk
== NULL
)
7237 goto error_disk_add
;
7239 new_disk
->index
= map
->num_members
- 1;
7240 /* slot to fill in autolayout
7242 new_disk
->raiddisk
= new_disk
->index
;
7243 new_disk
->disk
.status
|= CONFIGURED_DISK
;
7244 new_disk
->disk
.status
&= ~SPARE_DISK
;
7246 goto error_disk_add
;
7249 *tofree
= *space_list
;
7250 /* calculate new size
7252 imsm_set_array_size(new_dev
);
7259 *space_list
= tofree
;
7263 dprintf("Error: imsm: Cannot find disk.\n");
7267 static int apply_update_activate_spare(struct imsm_update_activate_spare
*u
,
7268 struct intel_super
*super
,
7269 struct active_array
*active_array
)
7271 struct imsm_super
*mpb
= super
->anchor
;
7272 struct imsm_dev
*dev
= get_imsm_dev(super
, u
->array
);
7273 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
7274 struct imsm_map
*migr_map
;
7275 struct active_array
*a
;
7276 struct imsm_disk
*disk
;
7283 int second_map_created
= 0;
7285 for (; u
; u
= u
->next
) {
7286 victim
= get_imsm_disk_idx(dev
, u
->slot
, MAP_X
);
7291 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
7296 fprintf(stderr
, "error: imsm_activate_spare passed "
7297 "an unknown disk (index: %d)\n",
7302 /* count failures (excluding rebuilds and the victim)
7303 * to determine map[0] state
7306 for (i
= 0; i
< map
->num_members
; i
++) {
7309 disk
= get_imsm_disk(super
,
7310 get_imsm_disk_idx(dev
, i
, MAP_X
));
7311 if (!disk
|| is_failed(disk
))
7315 /* adding a pristine spare, assign a new index */
7316 if (dl
->index
< 0) {
7317 dl
->index
= super
->anchor
->num_disks
;
7318 super
->anchor
->num_disks
++;
7321 disk
->status
|= CONFIGURED_DISK
;
7322 disk
->status
&= ~SPARE_DISK
;
7325 to_state
= imsm_check_degraded(super
, dev
, failed
, MAP_0
);
7326 if (!second_map_created
) {
7327 second_map_created
= 1;
7328 map
->map_state
= IMSM_T_STATE_DEGRADED
;
7329 migrate(dev
, super
, to_state
, MIGR_REBUILD
);
7331 map
->map_state
= to_state
;
7332 migr_map
= get_imsm_map(dev
, MAP_1
);
7333 set_imsm_ord_tbl_ent(map
, u
->slot
, dl
->index
);
7334 set_imsm_ord_tbl_ent(migr_map
, u
->slot
,
7335 dl
->index
| IMSM_ORD_REBUILD
);
7337 /* update the family_num to mark a new container
7338 * generation, being careful to record the existing
7339 * family_num in orig_family_num to clean up after
7340 * earlier mdadm versions that neglected to set it.
7342 if (mpb
->orig_family_num
== 0)
7343 mpb
->orig_family_num
= mpb
->family_num
;
7344 mpb
->family_num
+= super
->random
;
7346 /* count arrays using the victim in the metadata */
7348 for (a
= active_array
; a
; a
= a
->next
) {
7349 dev
= get_imsm_dev(super
, a
->info
.container_member
);
7350 map
= get_imsm_map(dev
, MAP_0
);
7352 if (get_imsm_disk_slot(map
, victim
) >= 0)
7356 /* delete the victim if it is no longer being
7362 /* We know that 'manager' isn't touching anything,
7363 * so it is safe to delete
7365 for (dlp
= &super
->disks
; *dlp
; dlp
= &(*dlp
)->next
)
7366 if ((*dlp
)->index
== victim
)
7369 /* victim may be on the missing list */
7371 for (dlp
= &super
->missing
; *dlp
;
7372 dlp
= &(*dlp
)->next
)
7373 if ((*dlp
)->index
== victim
)
7375 imsm_delete(super
, dlp
, victim
);
7382 static int apply_reshape_container_disks_update(struct imsm_update_reshape
*u
,
7383 struct intel_super
*super
,
7386 struct dl
*new_disk
;
7387 struct intel_dev
*id
;
7389 int delta_disks
= u
->new_raid_disks
- u
->old_raid_disks
;
7390 int disk_count
= u
->old_raid_disks
;
7391 void **tofree
= NULL
;
7392 int devices_to_reshape
= 1;
7393 struct imsm_super
*mpb
= super
->anchor
;
7395 unsigned int dev_id
;
7397 dprintf("imsm: apply_reshape_container_disks_update()\n");
7399 /* enable spares to use in array */
7400 for (i
= 0; i
< delta_disks
; i
++) {
7401 new_disk
= get_disk_super(super
,
7402 major(u
->new_disks
[i
]),
7403 minor(u
->new_disks
[i
]));
7404 dprintf("imsm: new disk for reshape is: %i:%i "
7405 "(%p, index = %i)\n",
7406 major(u
->new_disks
[i
]), minor(u
->new_disks
[i
]),
7407 new_disk
, new_disk
->index
);
7408 if ((new_disk
== NULL
) ||
7409 ((new_disk
->index
>= 0) &&
7410 (new_disk
->index
< u
->old_raid_disks
)))
7411 goto update_reshape_exit
;
7412 new_disk
->index
= disk_count
++;
7413 /* slot to fill in autolayout
7415 new_disk
->raiddisk
= new_disk
->index
;
7416 new_disk
->disk
.status
|=
7418 new_disk
->disk
.status
&= ~SPARE_DISK
;
7421 dprintf("imsm: volume set mpb->num_raid_devs = %i\n",
7422 mpb
->num_raid_devs
);
7423 /* manage changes in volume
7425 for (dev_id
= 0; dev_id
< mpb
->num_raid_devs
; dev_id
++) {
7426 void **sp
= *space_list
;
7427 struct imsm_dev
*newdev
;
7428 struct imsm_map
*newmap
, *oldmap
;
7430 for (id
= super
->devlist
; id
; id
= id
->next
) {
7431 if (id
->index
== dev_id
)
7440 /* Copy the dev, but not (all of) the map */
7441 memcpy(newdev
, id
->dev
, sizeof(*newdev
));
7442 oldmap
= get_imsm_map(id
->dev
, MAP_0
);
7443 newmap
= get_imsm_map(newdev
, MAP_0
);
7444 /* Copy the current map */
7445 memcpy(newmap
, oldmap
, sizeof_imsm_map(oldmap
));
7446 /* update one device only
7448 if (devices_to_reshape
) {
7449 dprintf("imsm: modifying subdev: %i\n",
7451 devices_to_reshape
--;
7452 newdev
->vol
.migr_state
= 1;
7453 newdev
->vol
.curr_migr_unit
= 0;
7454 set_migr_type(newdev
, MIGR_GEN_MIGR
);
7455 newmap
->num_members
= u
->new_raid_disks
;
7456 for (i
= 0; i
< delta_disks
; i
++) {
7457 set_imsm_ord_tbl_ent(newmap
,
7458 u
->old_raid_disks
+ i
,
7459 u
->old_raid_disks
+ i
);
7461 /* New map is correct, now need to save old map
7463 newmap
= get_imsm_map(newdev
, MAP_1
);
7464 memcpy(newmap
, oldmap
, sizeof_imsm_map(oldmap
));
7466 imsm_set_array_size(newdev
);
7469 sp
= (void **)id
->dev
;
7474 /* Clear migration record */
7475 memset(super
->migr_rec
, 0, sizeof(struct migr_record
));
7478 *space_list
= tofree
;
7481 update_reshape_exit
:
7486 static int apply_takeover_update(struct imsm_update_takeover
*u
,
7487 struct intel_super
*super
,
7490 struct imsm_dev
*dev
= NULL
;
7491 struct intel_dev
*dv
;
7492 struct imsm_dev
*dev_new
;
7493 struct imsm_map
*map
;
7497 for (dv
= super
->devlist
; dv
; dv
= dv
->next
)
7498 if (dv
->index
== (unsigned int)u
->subarray
) {
7506 map
= get_imsm_map(dev
, MAP_0
);
7508 if (u
->direction
== R10_TO_R0
) {
7509 /* Number of failed disks must be half of initial disk number */
7510 if (imsm_count_failed(super
, dev
, MAP_0
) !=
7511 (map
->num_members
/ 2))
7514 /* iterate through devices to mark removed disks as spare */
7515 for (dm
= super
->disks
; dm
; dm
= dm
->next
) {
7516 if (dm
->disk
.status
& FAILED_DISK
) {
7517 int idx
= dm
->index
;
7518 /* update indexes on the disk list */
7519 /* FIXME this loop-with-the-loop looks wrong, I'm not convinced
7520 the index values will end up being correct.... NB */
7521 for (du
= super
->disks
; du
; du
= du
->next
)
7522 if (du
->index
> idx
)
7524 /* mark as spare disk */
7529 map
->num_members
= map
->num_members
/ 2;
7530 map
->map_state
= IMSM_T_STATE_NORMAL
;
7531 map
->num_domains
= 1;
7532 map
->raid_level
= 0;
7533 map
->failed_disk_num
= -1;
7536 if (u
->direction
== R0_TO_R10
) {
7538 /* update slots in current disk list */
7539 for (dm
= super
->disks
; dm
; dm
= dm
->next
) {
7543 /* create new *missing* disks */
7544 for (i
= 0; i
< map
->num_members
; i
++) {
7545 space
= *space_list
;
7548 *space_list
= *space
;
7550 memcpy(du
, super
->disks
, sizeof(*du
));
7554 du
->index
= (i
* 2) + 1;
7555 sprintf((char *)du
->disk
.serial
,
7556 " MISSING_%d", du
->index
);
7557 sprintf((char *)du
->serial
,
7558 "MISSING_%d", du
->index
);
7559 du
->next
= super
->missing
;
7560 super
->missing
= du
;
7562 /* create new dev and map */
7563 space
= *space_list
;
7566 *space_list
= *space
;
7567 dev_new
= (void *)space
;
7568 memcpy(dev_new
, dev
, sizeof(*dev
));
7569 /* update new map */
7570 map
= get_imsm_map(dev_new
, MAP_0
);
7571 map
->num_members
= map
->num_members
* 2;
7572 map
->map_state
= IMSM_T_STATE_DEGRADED
;
7573 map
->num_domains
= 2;
7574 map
->raid_level
= 1;
7575 /* replace dev<->dev_new */
7578 /* update disk order table */
7579 for (du
= super
->disks
; du
; du
= du
->next
)
7581 set_imsm_ord_tbl_ent(map
, du
->index
, du
->index
);
7582 for (du
= super
->missing
; du
; du
= du
->next
)
7583 if (du
->index
>= 0) {
7584 set_imsm_ord_tbl_ent(map
, du
->index
, du
->index
);
7585 mark_missing(dv
->dev
, &du
->disk
, du
->index
);
7591 static void imsm_process_update(struct supertype
*st
,
7592 struct metadata_update
*update
)
7595 * crack open the metadata_update envelope to find the update record
7596 * update can be one of:
7597 * update_reshape_container_disks - all the arrays in the container
7598 * are being reshaped to have more devices. We need to mark
7599 * the arrays for general migration and convert selected spares
7600 * into active devices.
7601 * update_activate_spare - a spare device has replaced a failed
7602 * device in an array, update the disk_ord_tbl. If this disk is
7603 * present in all member arrays then also clear the SPARE_DISK
7605 * update_create_array
7607 * update_rename_array
7608 * update_add_remove_disk
7610 struct intel_super
*super
= st
->sb
;
7611 struct imsm_super
*mpb
;
7612 enum imsm_update_type type
= *(enum imsm_update_type
*) update
->buf
;
7614 /* update requires a larger buf but the allocation failed */
7615 if (super
->next_len
&& !super
->next_buf
) {
7616 super
->next_len
= 0;
7620 if (super
->next_buf
) {
7621 memcpy(super
->next_buf
, super
->buf
, super
->len
);
7623 super
->len
= super
->next_len
;
7624 super
->buf
= super
->next_buf
;
7626 super
->next_len
= 0;
7627 super
->next_buf
= NULL
;
7630 mpb
= super
->anchor
;
7633 case update_general_migration_checkpoint
: {
7634 struct intel_dev
*id
;
7635 struct imsm_update_general_migration_checkpoint
*u
=
7636 (void *)update
->buf
;
7638 dprintf("imsm: process_update() "
7639 "for update_general_migration_checkpoint called\n");
7641 /* find device under general migration */
7642 for (id
= super
->devlist
; id
; id
= id
->next
) {
7643 if (is_gen_migration(id
->dev
)) {
7644 id
->dev
->vol
.curr_migr_unit
=
7645 __cpu_to_le32(u
->curr_migr_unit
);
7646 super
->updates_pending
++;
7651 case update_takeover
: {
7652 struct imsm_update_takeover
*u
= (void *)update
->buf
;
7653 if (apply_takeover_update(u
, super
, &update
->space_list
)) {
7654 imsm_update_version_info(super
);
7655 super
->updates_pending
++;
7660 case update_reshape_container_disks
: {
7661 struct imsm_update_reshape
*u
= (void *)update
->buf
;
7662 if (apply_reshape_container_disks_update(
7663 u
, super
, &update
->space_list
))
7664 super
->updates_pending
++;
7667 case update_reshape_migration
: {
7668 struct imsm_update_reshape_migration
*u
= (void *)update
->buf
;
7669 if (apply_reshape_migration_update(
7670 u
, super
, &update
->space_list
))
7671 super
->updates_pending
++;
7674 case update_activate_spare
: {
7675 struct imsm_update_activate_spare
*u
= (void *) update
->buf
;
7676 if (apply_update_activate_spare(u
, super
, st
->arrays
))
7677 super
->updates_pending
++;
7680 case update_create_array
: {
7681 /* someone wants to create a new array, we need to be aware of
7682 * a few races/collisions:
7683 * 1/ 'Create' called by two separate instances of mdadm
7684 * 2/ 'Create' versus 'activate_spare': mdadm has chosen
7685 * devices that have since been assimilated via
7687 * In the event this update can not be carried out mdadm will
7688 * (FIX ME) notice that its update did not take hold.
7690 struct imsm_update_create_array
*u
= (void *) update
->buf
;
7691 struct intel_dev
*dv
;
7692 struct imsm_dev
*dev
;
7693 struct imsm_map
*map
, *new_map
;
7694 unsigned long long start
, end
;
7695 unsigned long long new_start
, new_end
;
7697 struct disk_info
*inf
;
7700 /* handle racing creates: first come first serve */
7701 if (u
->dev_idx
< mpb
->num_raid_devs
) {
7702 dprintf("%s: subarray %d already defined\n",
7703 __func__
, u
->dev_idx
);
7707 /* check update is next in sequence */
7708 if (u
->dev_idx
!= mpb
->num_raid_devs
) {
7709 dprintf("%s: can not create array %d expected index %d\n",
7710 __func__
, u
->dev_idx
, mpb
->num_raid_devs
);
7714 new_map
= get_imsm_map(&u
->dev
, MAP_0
);
7715 new_start
= __le32_to_cpu(new_map
->pba_of_lba0
);
7716 new_end
= new_start
+ __le32_to_cpu(new_map
->blocks_per_member
);
7717 inf
= get_disk_info(u
);
7719 /* handle activate_spare versus create race:
7720 * check to make sure that overlapping arrays do not include
7723 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
7724 dev
= get_imsm_dev(super
, i
);
7725 map
= get_imsm_map(dev
, MAP_0
);
7726 start
= __le32_to_cpu(map
->pba_of_lba0
);
7727 end
= start
+ __le32_to_cpu(map
->blocks_per_member
);
7728 if ((new_start
>= start
&& new_start
<= end
) ||
7729 (start
>= new_start
&& start
<= new_end
))
7734 if (disks_overlap(super
, i
, u
)) {
7735 dprintf("%s: arrays overlap\n", __func__
);
7740 /* check that prepare update was successful */
7741 if (!update
->space
) {
7742 dprintf("%s: prepare update failed\n", __func__
);
7746 /* check that all disks are still active before committing
7747 * changes. FIXME: could we instead handle this by creating a
7748 * degraded array? That's probably not what the user expects,
7749 * so better to drop this update on the floor.
7751 for (i
= 0; i
< new_map
->num_members
; i
++) {
7752 dl
= serial_to_dl(inf
[i
].serial
, super
);
7754 dprintf("%s: disk disappeared\n", __func__
);
7759 super
->updates_pending
++;
7761 /* convert spares to members and fixup ord_tbl */
7762 for (i
= 0; i
< new_map
->num_members
; i
++) {
7763 dl
= serial_to_dl(inf
[i
].serial
, super
);
7764 if (dl
->index
== -1) {
7765 dl
->index
= mpb
->num_disks
;
7767 dl
->disk
.status
|= CONFIGURED_DISK
;
7768 dl
->disk
.status
&= ~SPARE_DISK
;
7770 set_imsm_ord_tbl_ent(new_map
, i
, dl
->index
);
7775 update
->space
= NULL
;
7776 imsm_copy_dev(dev
, &u
->dev
);
7777 dv
->index
= u
->dev_idx
;
7778 dv
->next
= super
->devlist
;
7779 super
->devlist
= dv
;
7780 mpb
->num_raid_devs
++;
7782 imsm_update_version_info(super
);
7785 /* mdmon knows how to release update->space, but not
7786 * ((struct intel_dev *) update->space)->dev
7788 if (update
->space
) {
7794 case update_kill_array
: {
7795 struct imsm_update_kill_array
*u
= (void *) update
->buf
;
7796 int victim
= u
->dev_idx
;
7797 struct active_array
*a
;
7798 struct intel_dev
**dp
;
7799 struct imsm_dev
*dev
;
7801 /* sanity check that we are not affecting the uuid of
7802 * active arrays, or deleting an active array
7804 * FIXME when immutable ids are available, but note that
7805 * we'll also need to fixup the invalidated/active
7806 * subarray indexes in mdstat
7808 for (a
= st
->arrays
; a
; a
= a
->next
)
7809 if (a
->info
.container_member
>= victim
)
7811 /* by definition if mdmon is running at least one array
7812 * is active in the container, so checking
7813 * mpb->num_raid_devs is just extra paranoia
7815 dev
= get_imsm_dev(super
, victim
);
7816 if (a
|| !dev
|| mpb
->num_raid_devs
== 1) {
7817 dprintf("failed to delete subarray-%d\n", victim
);
7821 for (dp
= &super
->devlist
; *dp
;)
7822 if ((*dp
)->index
== (unsigned)super
->current_vol
) {
7825 if ((*dp
)->index
> (unsigned)victim
)
7829 mpb
->num_raid_devs
--;
7830 super
->updates_pending
++;
7833 case update_rename_array
: {
7834 struct imsm_update_rename_array
*u
= (void *) update
->buf
;
7835 char name
[MAX_RAID_SERIAL_LEN
+1];
7836 int target
= u
->dev_idx
;
7837 struct active_array
*a
;
7838 struct imsm_dev
*dev
;
7840 /* sanity check that we are not affecting the uuid of
7843 snprintf(name
, MAX_RAID_SERIAL_LEN
, "%s", (char *) u
->name
);
7844 name
[MAX_RAID_SERIAL_LEN
] = '\0';
7845 for (a
= st
->arrays
; a
; a
= a
->next
)
7846 if (a
->info
.container_member
== target
)
7848 dev
= get_imsm_dev(super
, u
->dev_idx
);
7849 if (a
|| !dev
|| !check_name(super
, name
, 1)) {
7850 dprintf("failed to rename subarray-%d\n", target
);
7854 snprintf((char *) dev
->volume
, MAX_RAID_SERIAL_LEN
, "%s", name
);
7855 super
->updates_pending
++;
7858 case update_add_remove_disk
: {
7859 /* we may be able to repair some arrays if disks are
7860 * being added, check teh status of add_remove_disk
7861 * if discs has been added.
7863 if (add_remove_disk_update(super
)) {
7864 struct active_array
*a
;
7866 super
->updates_pending
++;
7867 for (a
= st
->arrays
; a
; a
= a
->next
)
7868 a
->check_degraded
= 1;
7873 fprintf(stderr
, "error: unsuported process update type:"
7874 "(type: %d)\n", type
);
7878 static struct mdinfo
*get_spares_for_grow(struct supertype
*st
);
7880 static void imsm_prepare_update(struct supertype
*st
,
7881 struct metadata_update
*update
)
7884 * Allocate space to hold new disk entries, raid-device entries or a new
7885 * mpb if necessary. The manager synchronously waits for updates to
7886 * complete in the monitor, so new mpb buffers allocated here can be
7887 * integrated by the monitor thread without worrying about live pointers
7888 * in the manager thread.
7890 enum imsm_update_type type
= *(enum imsm_update_type
*) update
->buf
;
7891 struct intel_super
*super
= st
->sb
;
7892 struct imsm_super
*mpb
= super
->anchor
;
7897 case update_general_migration_checkpoint
:
7898 dprintf("imsm: prepare_update() "
7899 "for update_general_migration_checkpoint called\n");
7901 case update_takeover
: {
7902 struct imsm_update_takeover
*u
= (void *)update
->buf
;
7903 if (u
->direction
== R0_TO_R10
) {
7904 void **tail
= (void **)&update
->space_list
;
7905 struct imsm_dev
*dev
= get_imsm_dev(super
, u
->subarray
);
7906 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
7907 int num_members
= map
->num_members
;
7911 /* allocate memory for added disks */
7912 for (i
= 0; i
< num_members
; i
++) {
7913 size
= sizeof(struct dl
);
7914 space
= malloc(size
);
7923 /* allocate memory for new device */
7924 size
= sizeof_imsm_dev(super
->devlist
->dev
, 0) +
7925 (num_members
* sizeof(__u32
));
7926 space
= malloc(size
);
7935 len
= disks_to_mpb_size(num_members
* 2);
7937 /* if allocation didn't success, free buffer */
7938 while (update
->space_list
) {
7939 void **sp
= update
->space_list
;
7940 update
->space_list
= *sp
;
7948 case update_reshape_container_disks
: {
7949 /* Every raid device in the container is about to
7950 * gain some more devices, and we will enter a
7952 * So each 'imsm_map' will be bigger, and the imsm_vol
7953 * will now hold 2 of them.
7954 * Thus we need new 'struct imsm_dev' allocations sized
7955 * as sizeof_imsm_dev but with more devices in both maps.
7957 struct imsm_update_reshape
*u
= (void *)update
->buf
;
7958 struct intel_dev
*dl
;
7959 void **space_tail
= (void**)&update
->space_list
;
7961 dprintf("imsm: imsm_prepare_update() for update_reshape\n");
7963 for (dl
= super
->devlist
; dl
; dl
= dl
->next
) {
7964 int size
= sizeof_imsm_dev(dl
->dev
, 1);
7966 if (u
->new_raid_disks
> u
->old_raid_disks
)
7967 size
+= sizeof(__u32
)*2*
7968 (u
->new_raid_disks
- u
->old_raid_disks
);
7977 len
= disks_to_mpb_size(u
->new_raid_disks
);
7978 dprintf("New anchor length is %llu\n", (unsigned long long)len
);
7981 case update_reshape_migration
: {
7982 /* for migration level 0->5 we need to add disks
7983 * so the same as for container operation we will copy
7984 * device to the bigger location.
7985 * in memory prepared device and new disk area are prepared
7986 * for usage in process update
7988 struct imsm_update_reshape_migration
*u
= (void *)update
->buf
;
7989 struct intel_dev
*id
;
7990 void **space_tail
= (void **)&update
->space_list
;
7993 int current_level
= -1;
7995 dprintf("imsm: imsm_prepare_update() for update_reshape\n");
7997 /* add space for bigger array in update
7999 for (id
= super
->devlist
; id
; id
= id
->next
) {
8000 if (id
->index
== (unsigned)u
->subdev
) {
8001 size
= sizeof_imsm_dev(id
->dev
, 1);
8002 if (u
->new_raid_disks
> u
->old_raid_disks
)
8003 size
+= sizeof(__u32
)*2*
8004 (u
->new_raid_disks
- u
->old_raid_disks
);
8014 if (update
->space_list
== NULL
)
8017 /* add space for disk in update
8019 size
= sizeof(struct dl
);
8022 free(update
->space_list
);
8023 update
->space_list
= NULL
;
8030 /* add spare device to update
8032 for (id
= super
->devlist
; id
; id
= id
->next
)
8033 if (id
->index
== (unsigned)u
->subdev
) {
8034 struct imsm_dev
*dev
;
8035 struct imsm_map
*map
;
8037 dev
= get_imsm_dev(super
, u
->subdev
);
8038 map
= get_imsm_map(dev
, MAP_0
);
8039 current_level
= map
->raid_level
;
8042 if ((u
->new_level
== 5) && (u
->new_level
!= current_level
)) {
8043 struct mdinfo
*spares
;
8045 spares
= get_spares_for_grow(st
);
8053 makedev(dev
->disk
.major
,
8055 dl
= get_disk_super(super
,
8058 dl
->index
= u
->old_raid_disks
;
8064 len
= disks_to_mpb_size(u
->new_raid_disks
);
8065 dprintf("New anchor length is %llu\n", (unsigned long long)len
);
8068 case update_create_array
: {
8069 struct imsm_update_create_array
*u
= (void *) update
->buf
;
8070 struct intel_dev
*dv
;
8071 struct imsm_dev
*dev
= &u
->dev
;
8072 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
8074 struct disk_info
*inf
;
8078 inf
= get_disk_info(u
);
8079 len
= sizeof_imsm_dev(dev
, 1);
8080 /* allocate a new super->devlist entry */
8081 dv
= malloc(sizeof(*dv
));
8083 dv
->dev
= malloc(len
);
8088 update
->space
= NULL
;
8092 /* count how many spares will be converted to members */
8093 for (i
= 0; i
< map
->num_members
; i
++) {
8094 dl
= serial_to_dl(inf
[i
].serial
, super
);
8096 /* hmm maybe it failed?, nothing we can do about
8101 if (count_memberships(dl
, super
) == 0)
8104 len
+= activate
* sizeof(struct imsm_disk
);
8111 /* check if we need a larger metadata buffer */
8112 if (super
->next_buf
)
8113 buf_len
= super
->next_len
;
8115 buf_len
= super
->len
;
8117 if (__le32_to_cpu(mpb
->mpb_size
) + len
> buf_len
) {
8118 /* ok we need a larger buf than what is currently allocated
8119 * if this allocation fails process_update will notice that
8120 * ->next_len is set and ->next_buf is NULL
8122 buf_len
= ROUND_UP(__le32_to_cpu(mpb
->mpb_size
) + len
, 512);
8123 if (super
->next_buf
)
8124 free(super
->next_buf
);
8126 super
->next_len
= buf_len
;
8127 if (posix_memalign(&super
->next_buf
, 512, buf_len
) == 0)
8128 memset(super
->next_buf
, 0, buf_len
);
8130 super
->next_buf
= NULL
;
8134 /* must be called while manager is quiesced */
8135 static void imsm_delete(struct intel_super
*super
, struct dl
**dlp
, unsigned index
)
8137 struct imsm_super
*mpb
= super
->anchor
;
8139 struct imsm_dev
*dev
;
8140 struct imsm_map
*map
;
8141 int i
, j
, num_members
;
8144 dprintf("%s: deleting device[%d] from imsm_super\n",
8147 /* shift all indexes down one */
8148 for (iter
= super
->disks
; iter
; iter
= iter
->next
)
8149 if (iter
->index
> (int)index
)
8151 for (iter
= super
->missing
; iter
; iter
= iter
->next
)
8152 if (iter
->index
> (int)index
)
8155 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
8156 dev
= get_imsm_dev(super
, i
);
8157 map
= get_imsm_map(dev
, MAP_0
);
8158 num_members
= map
->num_members
;
8159 for (j
= 0; j
< num_members
; j
++) {
8160 /* update ord entries being careful not to propagate
8161 * ord-flags to the first map
8163 ord
= get_imsm_ord_tbl_ent(dev
, j
, MAP_X
);
8165 if (ord_to_idx(ord
) <= index
)
8168 map
= get_imsm_map(dev
, MAP_0
);
8169 set_imsm_ord_tbl_ent(map
, j
, ord_to_idx(ord
- 1));
8170 map
= get_imsm_map(dev
, MAP_1
);
8172 set_imsm_ord_tbl_ent(map
, j
, ord
- 1);
8177 super
->updates_pending
++;
8179 struct dl
*dl
= *dlp
;
8181 *dlp
= (*dlp
)->next
;
8182 __free_imsm_disk(dl
);
8185 #endif /* MDASSEMBLE */
8187 static void close_targets(int *targets
, int new_disks
)
8194 for (i
= 0; i
< new_disks
; i
++) {
8195 if (targets
[i
] >= 0) {
8202 static int imsm_get_allowed_degradation(int level
, int raid_disks
,
8203 struct intel_super
*super
,
8204 struct imsm_dev
*dev
)
8209 struct imsm_map
*map
;
8212 ret_val
= raid_disks
/2;
8213 /* check map if all disks pairs not failed
8216 map
= get_imsm_map(dev
, MAP_0
);
8217 for (i
= 0; i
< ret_val
; i
++) {
8218 int degradation
= 0;
8219 if (get_imsm_disk(super
, i
) == NULL
)
8221 if (get_imsm_disk(super
, i
+ 1) == NULL
)
8223 if (degradation
== 2)
8226 map
= get_imsm_map(dev
, MAP_1
);
8227 /* if there is no second map
8228 * result can be returned
8232 /* check degradation in second map
8234 for (i
= 0; i
< ret_val
; i
++) {
8235 int degradation
= 0;
8236 if (get_imsm_disk(super
, i
) == NULL
)
8238 if (get_imsm_disk(super
, i
+ 1) == NULL
)
8240 if (degradation
== 2)
8255 /*******************************************************************************
8256 * Function: open_backup_targets
8257 * Description: Function opens file descriptors for all devices given in
8260 * info : general array info
8261 * raid_disks : number of disks
8262 * raid_fds : table of device's file descriptors
8263 * super : intel super for raid10 degradation check
8264 * dev : intel device for raid10 degradation check
8268 ******************************************************************************/
8269 int open_backup_targets(struct mdinfo
*info
, int raid_disks
, int *raid_fds
,
8270 struct intel_super
*super
, struct imsm_dev
*dev
)
8276 for (i
= 0; i
< raid_disks
; i
++)
8279 for (sd
= info
->devs
; sd
; sd
= sd
->next
) {
8282 if (sd
->disk
.state
& (1<<MD_DISK_FAULTY
)) {
8283 dprintf("disk is faulty!!\n");
8287 if ((sd
->disk
.raid_disk
>= raid_disks
) ||
8288 (sd
->disk
.raid_disk
< 0))
8291 dn
= map_dev(sd
->disk
.major
,
8293 raid_fds
[sd
->disk
.raid_disk
] = dev_open(dn
, O_RDWR
);
8294 if (raid_fds
[sd
->disk
.raid_disk
] < 0) {
8295 fprintf(stderr
, "cannot open component\n");
8300 /* check if maximum array degradation level is not exceeded
8302 if ((raid_disks
- opened
) >
8303 imsm_get_allowed_degradation(info
->new_level
,
8306 fprintf(stderr
, "Not enough disks can be opened.\n");
8307 close_targets(raid_fds
, raid_disks
);
8314 /*******************************************************************************
8315 * Function: init_migr_record_imsm
8316 * Description: Function inits imsm migration record
8318 * super : imsm internal array info
8319 * dev : device under migration
8320 * info : general array info to find the smallest device
8323 ******************************************************************************/
8324 void init_migr_record_imsm(struct supertype
*st
, struct imsm_dev
*dev
,
8325 struct mdinfo
*info
)
8327 struct intel_super
*super
= st
->sb
;
8328 struct migr_record
*migr_rec
= super
->migr_rec
;
8330 unsigned long long dsize
, dev_sectors
;
8331 long long unsigned min_dev_sectors
= -1LLU;
8335 struct imsm_map
*map_dest
= get_imsm_map(dev
, MAP_0
);
8336 struct imsm_map
*map_src
= get_imsm_map(dev
, MAP_1
);
8337 unsigned long long num_migr_units
;
8338 unsigned long long array_blocks
;
8340 memset(migr_rec
, 0, sizeof(struct migr_record
));
8341 migr_rec
->family_num
= __cpu_to_le32(super
->anchor
->family_num
);
8343 /* only ascending reshape supported now */
8344 migr_rec
->ascending_migr
= __cpu_to_le32(1);
8346 migr_rec
->dest_depth_per_unit
= GEN_MIGR_AREA_SIZE
/
8347 max(map_dest
->blocks_per_strip
, map_src
->blocks_per_strip
);
8348 migr_rec
->dest_depth_per_unit
*= map_dest
->blocks_per_strip
;
8349 new_data_disks
= imsm_num_data_members(dev
, MAP_0
);
8350 migr_rec
->blocks_per_unit
=
8351 __cpu_to_le32(migr_rec
->dest_depth_per_unit
* new_data_disks
);
8352 migr_rec
->dest_depth_per_unit
=
8353 __cpu_to_le32(migr_rec
->dest_depth_per_unit
);
8354 array_blocks
= info
->component_size
* new_data_disks
;
8356 array_blocks
/ __le32_to_cpu(migr_rec
->blocks_per_unit
);
8358 if (array_blocks
% __le32_to_cpu(migr_rec
->blocks_per_unit
))
8360 migr_rec
->num_migr_units
= __cpu_to_le32(num_migr_units
);
8362 migr_rec
->post_migr_vol_cap
= dev
->size_low
;
8363 migr_rec
->post_migr_vol_cap_hi
= dev
->size_high
;
8366 /* Find the smallest dev */
8367 for (sd
= info
->devs
; sd
; sd
= sd
->next
) {
8368 sprintf(nm
, "%d:%d", sd
->disk
.major
, sd
->disk
.minor
);
8369 fd
= dev_open(nm
, O_RDONLY
);
8372 get_dev_size(fd
, NULL
, &dsize
);
8373 dev_sectors
= dsize
/ 512;
8374 if (dev_sectors
< min_dev_sectors
)
8375 min_dev_sectors
= dev_sectors
;
8378 migr_rec
->ckpt_area_pba
= __cpu_to_le32(min_dev_sectors
-
8379 RAID_DISK_RESERVED_BLOCKS_IMSM_HI
);
8381 write_imsm_migr_rec(st
);
8386 /*******************************************************************************
8387 * Function: save_backup_imsm
8388 * Description: Function saves critical data stripes to Migration Copy Area
8389 * and updates the current migration unit status.
8390 * Use restore_stripes() to form a destination stripe,
8391 * and to write it to the Copy Area.
8393 * st : supertype information
8394 * dev : imsm device that backup is saved for
8395 * info : general array info
8396 * buf : input buffer
8397 * length : length of data to backup (blocks_per_unit)
8401 ******************************************************************************/
8402 int save_backup_imsm(struct supertype
*st
,
8403 struct imsm_dev
*dev
,
8404 struct mdinfo
*info
,
8409 struct intel_super
*super
= st
->sb
;
8410 unsigned long long *target_offsets
= NULL
;
8411 int *targets
= NULL
;
8413 struct imsm_map
*map_dest
= get_imsm_map(dev
, MAP_0
);
8414 int new_disks
= map_dest
->num_members
;
8415 int dest_layout
= 0;
8417 unsigned long long start
;
8418 int data_disks
= imsm_num_data_members(dev
, MAP_0
);
8420 targets
= malloc(new_disks
* sizeof(int));
8424 for (i
= 0; i
< new_disks
; i
++)
8427 target_offsets
= malloc(new_disks
* sizeof(unsigned long long));
8428 if (!target_offsets
)
8431 start
= info
->reshape_progress
* 512;
8432 for (i
= 0; i
< new_disks
; i
++) {
8433 target_offsets
[i
] = (unsigned long long)
8434 __le32_to_cpu(super
->migr_rec
->ckpt_area_pba
) * 512;
8435 /* move back copy area adderss, it will be moved forward
8436 * in restore_stripes() using start input variable
8438 target_offsets
[i
] -= start
/data_disks
;
8441 if (open_backup_targets(info
, new_disks
, targets
,
8445 dest_layout
= imsm_level_to_layout(map_dest
->raid_level
);
8446 dest_chunk
= __le16_to_cpu(map_dest
->blocks_per_strip
) * 512;
8448 if (restore_stripes(targets
, /* list of dest devices */
8449 target_offsets
, /* migration record offsets */
8452 map_dest
->raid_level
,
8454 -1, /* source backup file descriptor */
8455 0, /* input buf offset
8456 * always 0 buf is already offseted */
8460 fprintf(stderr
, Name
": Error restoring stripes\n");
8468 close_targets(targets
, new_disks
);
8471 free(target_offsets
);
8476 /*******************************************************************************
8477 * Function: save_checkpoint_imsm
8478 * Description: Function called for current unit status update
8479 * in the migration record. It writes it to disk.
8481 * super : imsm internal array info
8482 * info : general array info
8486 * 2: failure, means no valid migration record
8487 * / no general migration in progress /
8488 ******************************************************************************/
8489 int save_checkpoint_imsm(struct supertype
*st
, struct mdinfo
*info
, int state
)
8491 struct intel_super
*super
= st
->sb
;
8492 unsigned long long blocks_per_unit
;
8493 unsigned long long curr_migr_unit
;
8495 if (load_imsm_migr_rec(super
, info
) != 0) {
8496 dprintf("imsm: ERROR: Cannot read migration record "
8497 "for checkpoint save.\n");
8501 blocks_per_unit
= __le32_to_cpu(super
->migr_rec
->blocks_per_unit
);
8502 if (blocks_per_unit
== 0) {
8503 dprintf("imsm: no migration in progress.\n");
8506 curr_migr_unit
= info
->reshape_progress
/ blocks_per_unit
;
8507 /* check if array is alligned to copy area
8508 * if it is not alligned, add one to current migration unit value
8509 * this can happend on array reshape finish only
8511 if (info
->reshape_progress
% blocks_per_unit
)
8514 super
->migr_rec
->curr_migr_unit
=
8515 __cpu_to_le32(curr_migr_unit
);
8516 super
->migr_rec
->rec_status
= __cpu_to_le32(state
);
8517 super
->migr_rec
->dest_1st_member_lba
=
8518 __cpu_to_le32(curr_migr_unit
*
8519 __le32_to_cpu(super
->migr_rec
->dest_depth_per_unit
));
8520 if (write_imsm_migr_rec(st
) < 0) {
8521 dprintf("imsm: Cannot write migration record "
8522 "outside backup area\n");
8529 /*******************************************************************************
8530 * Function: recover_backup_imsm
8531 * Description: Function recovers critical data from the Migration Copy Area
8532 * while assembling an array.
8534 * super : imsm internal array info
8535 * info : general array info
8537 * 0 : success (or there is no data to recover)
8539 ******************************************************************************/
8540 int recover_backup_imsm(struct supertype
*st
, struct mdinfo
*info
)
8542 struct intel_super
*super
= st
->sb
;
8543 struct migr_record
*migr_rec
= super
->migr_rec
;
8544 struct imsm_map
*map_dest
= NULL
;
8545 struct intel_dev
*id
= NULL
;
8546 unsigned long long read_offset
;
8547 unsigned long long write_offset
;
8549 int *targets
= NULL
;
8550 int new_disks
, i
, err
;
8553 unsigned long curr_migr_unit
= __le32_to_cpu(migr_rec
->curr_migr_unit
);
8554 unsigned long num_migr_units
= __le32_to_cpu(migr_rec
->num_migr_units
);
8556 int skipped_disks
= 0;
8558 err
= sysfs_get_str(info
, NULL
, "array_state", (char *)buffer
, 20);
8562 /* recover data only during assemblation */
8563 if (strncmp(buffer
, "inactive", 8) != 0)
8565 /* no data to recover */
8566 if (__le32_to_cpu(migr_rec
->rec_status
) == UNIT_SRC_NORMAL
)
8568 if (curr_migr_unit
>= num_migr_units
)
8571 /* find device during reshape */
8572 for (id
= super
->devlist
; id
; id
= id
->next
)
8573 if (is_gen_migration(id
->dev
))
8578 map_dest
= get_imsm_map(id
->dev
, MAP_0
);
8579 new_disks
= map_dest
->num_members
;
8581 read_offset
= (unsigned long long)
8582 __le32_to_cpu(migr_rec
->ckpt_area_pba
) * 512;
8584 write_offset
= ((unsigned long long)
8585 __le32_to_cpu(migr_rec
->dest_1st_member_lba
) +
8586 __le32_to_cpu(map_dest
->pba_of_lba0
)) * 512;
8588 unit_len
= __le32_to_cpu(migr_rec
->dest_depth_per_unit
) * 512;
8589 if (posix_memalign((void **)&buf
, 512, unit_len
) != 0)
8591 targets
= malloc(new_disks
* sizeof(int));
8595 if (open_backup_targets(info
, new_disks
, targets
, super
, id
->dev
)) {
8597 Name
": Cannot open some devices belonging to array.\n");
8601 for (i
= 0; i
< new_disks
; i
++) {
8602 if (targets
[i
] < 0) {
8606 if (lseek64(targets
[i
], read_offset
, SEEK_SET
) < 0) {
8608 Name
": Cannot seek to block: %s\n",
8613 if ((unsigned)read(targets
[i
], buf
, unit_len
) != unit_len
) {
8615 Name
": Cannot read copy area block: %s\n",
8620 if (lseek64(targets
[i
], write_offset
, SEEK_SET
) < 0) {
8622 Name
": Cannot seek to block: %s\n",
8627 if ((unsigned)write(targets
[i
], buf
, unit_len
) != unit_len
) {
8629 Name
": Cannot restore block: %s\n",
8636 if (skipped_disks
> imsm_get_allowed_degradation(info
->new_level
,
8641 Name
": Cannot restore data from backup."
8642 " Too many failed disks\n");
8646 if (save_checkpoint_imsm(st
, info
, UNIT_SRC_NORMAL
)) {
8647 /* ignore error == 2, this can mean end of reshape here
8649 dprintf("imsm: Cannot write checkpoint to "
8650 "migration record (UNIT_SRC_NORMAL) during restart\n");
8656 for (i
= 0; i
< new_disks
; i
++)
8665 static char disk_by_path
[] = "/dev/disk/by-path/";
8667 static const char *imsm_get_disk_controller_domain(const char *path
)
8669 char disk_path
[PATH_MAX
];
8673 strncpy(disk_path
, disk_by_path
, PATH_MAX
- 1);
8674 strncat(disk_path
, path
, PATH_MAX
- strlen(disk_path
) - 1);
8675 if (stat(disk_path
, &st
) == 0) {
8676 struct sys_dev
* hba
;
8679 path
= devt_to_devpath(st
.st_rdev
);
8682 hba
= find_disk_attached_hba(-1, path
);
8683 if (hba
&& hba
->type
== SYS_DEV_SAS
)
8685 else if (hba
&& hba
->type
== SYS_DEV_SATA
)
8689 dprintf("path: %s hba: %s attached: %s\n",
8690 path
, (hba
) ? hba
->path
: "NULL", drv
);
8698 static int imsm_find_array_minor_by_subdev(int subdev
, int container
, int *minor
)
8700 char subdev_name
[20];
8701 struct mdstat_ent
*mdstat
;
8703 sprintf(subdev_name
, "%d", subdev
);
8704 mdstat
= mdstat_by_subdev(subdev_name
, container
);
8708 *minor
= mdstat
->devnum
;
8709 free_mdstat(mdstat
);
8713 static int imsm_reshape_is_allowed_on_container(struct supertype
*st
,
8714 struct geo_params
*geo
,
8715 int *old_raid_disks
)
8717 /* currently we only support increasing the number of devices
8718 * for a container. This increases the number of device for each
8719 * member array. They must all be RAID0 or RAID5.
8722 struct mdinfo
*info
, *member
;
8723 int devices_that_can_grow
= 0;
8725 dprintf("imsm: imsm_reshape_is_allowed_on_container(ENTER): "
8726 "st->devnum = (%i)\n",
8729 if (geo
->size
!= -1 ||
8730 geo
->level
!= UnSet
||
8731 geo
->layout
!= UnSet
||
8732 geo
->chunksize
!= 0 ||
8733 geo
->raid_disks
== UnSet
) {
8734 dprintf("imsm: Container operation is allowed for "
8735 "raid disks number change only.\n");
8739 info
= container_content_imsm(st
, NULL
);
8740 for (member
= info
; member
; member
= member
->next
) {
8744 dprintf("imsm: checking device_num: %i\n",
8745 member
->container_member
);
8747 if (geo
->raid_disks
<= member
->array
.raid_disks
) {
8748 /* we work on container for Online Capacity Expansion
8749 * only so raid_disks has to grow
8751 dprintf("imsm: for container operation raid disks "
8752 "increase is required\n");
8756 if ((info
->array
.level
!= 0) &&
8757 (info
->array
.level
!= 5)) {
8758 /* we cannot use this container with other raid level
8760 dprintf("imsm: for container operation wrong"
8761 " raid level (%i) detected\n",
8765 /* check for platform support
8766 * for this raid level configuration
8768 struct intel_super
*super
= st
->sb
;
8769 if (!is_raid_level_supported(super
->orom
,
8770 member
->array
.level
,
8772 dprintf("platform does not support raid%d with"
8776 geo
->raid_disks
> 1 ? "s" : "");
8779 /* check if component size is aligned to chunk size
8781 if (info
->component_size
%
8782 (info
->array
.chunk_size
/512)) {
8783 dprintf("Component size is not aligned to "
8789 if (*old_raid_disks
&&
8790 info
->array
.raid_disks
!= *old_raid_disks
)
8792 *old_raid_disks
= info
->array
.raid_disks
;
8794 /* All raid5 and raid0 volumes in container
8795 * have to be ready for Online Capacity Expansion
8796 * so they need to be assembled. We have already
8797 * checked that no recovery etc is happening.
8799 result
= imsm_find_array_minor_by_subdev(member
->container_member
,
8803 dprintf("imsm: cannot find array\n");
8806 devices_that_can_grow
++;
8809 if (!member
&& devices_that_can_grow
)
8813 dprintf("\tContainer operation allowed\n");
8815 dprintf("\tError: %i\n", ret_val
);
8820 /* Function: get_spares_for_grow
8821 * Description: Allocates memory and creates list of spare devices
8822 * avaliable in container. Checks if spare drive size is acceptable.
8823 * Parameters: Pointer to the supertype structure
8824 * Returns: Pointer to the list of spare devices (mdinfo structure) on success,
8827 static struct mdinfo
*get_spares_for_grow(struct supertype
*st
)
8829 unsigned long long min_size
= min_acceptable_spare_size_imsm(st
);
8830 return container_choose_spares(st
, min_size
, NULL
, NULL
, NULL
, 0);
8833 /******************************************************************************
8834 * function: imsm_create_metadata_update_for_reshape
8835 * Function creates update for whole IMSM container.
8837 ******************************************************************************/
8838 static int imsm_create_metadata_update_for_reshape(
8839 struct supertype
*st
,
8840 struct geo_params
*geo
,
8842 struct imsm_update_reshape
**updatep
)
8844 struct intel_super
*super
= st
->sb
;
8845 struct imsm_super
*mpb
= super
->anchor
;
8846 int update_memory_size
= 0;
8847 struct imsm_update_reshape
*u
= NULL
;
8848 struct mdinfo
*spares
= NULL
;
8850 int delta_disks
= 0;
8853 dprintf("imsm_update_metadata_for_reshape(enter) raid_disks = %i\n",
8856 delta_disks
= geo
->raid_disks
- old_raid_disks
;
8858 /* size of all update data without anchor */
8859 update_memory_size
= sizeof(struct imsm_update_reshape
);
8861 /* now add space for spare disks that we need to add. */
8862 update_memory_size
+= sizeof(u
->new_disks
[0]) * (delta_disks
- 1);
8864 u
= calloc(1, update_memory_size
);
8867 "cannot get memory for imsm_update_reshape update\n");
8870 u
->type
= update_reshape_container_disks
;
8871 u
->old_raid_disks
= old_raid_disks
;
8872 u
->new_raid_disks
= geo
->raid_disks
;
8874 /* now get spare disks list
8876 spares
= get_spares_for_grow(st
);
8879 || delta_disks
> spares
->array
.spare_disks
) {
8880 fprintf(stderr
, Name
": imsm: ERROR: Cannot get spare devices "
8881 "for %s.\n", geo
->dev_name
);
8886 /* we have got spares
8887 * update disk list in imsm_disk list table in anchor
8889 dprintf("imsm: %i spares are available.\n\n",
8890 spares
->array
.spare_disks
);
8893 for (i
= 0; i
< delta_disks
; i
++) {
8898 u
->new_disks
[i
] = makedev(dev
->disk
.major
,
8900 dl
= get_disk_super(super
, dev
->disk
.major
, dev
->disk
.minor
);
8901 dl
->index
= mpb
->num_disks
;
8911 dprintf("imsm: reshape update preparation :");
8912 if (i
== delta_disks
) {
8915 return update_memory_size
;
8918 dprintf(" Error\n");
8923 /******************************************************************************
8924 * function: imsm_create_metadata_update_for_migration()
8925 * Creates update for IMSM array.
8927 ******************************************************************************/
8928 static int imsm_create_metadata_update_for_migration(
8929 struct supertype
*st
,
8930 struct geo_params
*geo
,
8931 struct imsm_update_reshape_migration
**updatep
)
8933 struct intel_super
*super
= st
->sb
;
8934 int update_memory_size
= 0;
8935 struct imsm_update_reshape_migration
*u
= NULL
;
8936 struct imsm_dev
*dev
;
8937 int previous_level
= -1;
8939 dprintf("imsm_create_metadata_update_for_migration(enter)"
8940 " New Level = %i\n", geo
->level
);
8942 /* size of all update data without anchor */
8943 update_memory_size
= sizeof(struct imsm_update_reshape_migration
);
8945 u
= calloc(1, update_memory_size
);
8947 dprintf("error: cannot get memory for "
8948 "imsm_create_metadata_update_for_migration\n");
8951 u
->type
= update_reshape_migration
;
8952 u
->subdev
= super
->current_vol
;
8953 u
->new_level
= geo
->level
;
8954 u
->new_layout
= geo
->layout
;
8955 u
->new_raid_disks
= u
->old_raid_disks
= geo
->raid_disks
;
8956 u
->new_disks
[0] = -1;
8957 u
->new_chunksize
= -1;
8959 dev
= get_imsm_dev(super
, u
->subdev
);
8961 struct imsm_map
*map
;
8963 map
= get_imsm_map(dev
, MAP_0
);
8965 int current_chunk_size
=
8966 __le16_to_cpu(map
->blocks_per_strip
) / 2;
8968 if (geo
->chunksize
!= current_chunk_size
) {
8969 u
->new_chunksize
= geo
->chunksize
/ 1024;
8971 "chunk size change from %i to %i\n",
8972 current_chunk_size
, u
->new_chunksize
);
8974 previous_level
= map
->raid_level
;
8977 if ((geo
->level
== 5) && (previous_level
== 0)) {
8978 struct mdinfo
*spares
= NULL
;
8980 u
->new_raid_disks
++;
8981 spares
= get_spares_for_grow(st
);
8982 if ((spares
== NULL
) || (spares
->array
.spare_disks
< 1)) {
8985 update_memory_size
= 0;
8986 dprintf("error: cannot get spare device "
8987 "for requested migration");
8992 dprintf("imsm: reshape update preparation : OK\n");
8995 return update_memory_size
;
8998 static void imsm_update_metadata_locally(struct supertype
*st
,
9001 struct metadata_update mu
;
9006 mu
.space_list
= NULL
;
9008 imsm_prepare_update(st
, &mu
);
9009 imsm_process_update(st
, &mu
);
9011 while (mu
.space_list
) {
9012 void **space
= mu
.space_list
;
9013 mu
.space_list
= *space
;
9018 /***************************************************************************
9019 * Function: imsm_analyze_change
9020 * Description: Function analyze change for single volume
9021 * and validate if transition is supported
9022 * Parameters: Geometry parameters, supertype structure
9023 * Returns: Operation type code on success, -1 if fail
9024 ****************************************************************************/
9025 enum imsm_reshape_type
imsm_analyze_change(struct supertype
*st
,
9026 struct geo_params
*geo
)
9035 getinfo_super_imsm_volume(st
, &info
, NULL
);
9036 if ((geo
->level
!= info
.array
.level
) &&
9037 (geo
->level
>= 0) &&
9038 (geo
->level
!= UnSet
)) {
9039 switch (info
.array
.level
) {
9041 if (geo
->level
== 5) {
9042 change
= CH_MIGRATION
;
9043 if (geo
->layout
!= ALGORITHM_LEFT_ASYMMETRIC
) {
9045 Name
" Error. Requested Layout "
9046 "not supported (left-asymmetric layout "
9047 "is supported only)!\n");
9049 goto analyse_change_exit
;
9051 layout
= geo
->layout
;
9053 devNumChange
= 1; /* parity disk added */
9054 } else if (geo
->level
== 10) {
9055 change
= CH_TAKEOVER
;
9057 devNumChange
= 2; /* two mirrors added */
9058 layout
= 0x102; /* imsm supported layout */
9063 if (geo
->level
== 0) {
9064 change
= CH_TAKEOVER
;
9066 devNumChange
= -(geo
->raid_disks
/2);
9067 layout
= 0; /* imsm raid0 layout */
9073 Name
" Error. Level Migration from %d to %d "
9075 info
.array
.level
, geo
->level
);
9076 goto analyse_change_exit
;
9079 geo
->level
= info
.array
.level
;
9081 if ((geo
->layout
!= info
.array
.layout
)
9082 && ((geo
->layout
!= UnSet
) && (geo
->layout
!= -1))) {
9083 change
= CH_MIGRATION
;
9084 if ((info
.array
.layout
== 0)
9085 && (info
.array
.level
== 5)
9086 && (geo
->layout
== 5)) {
9087 /* reshape 5 -> 4 */
9088 } else if ((info
.array
.layout
== 5)
9089 && (info
.array
.level
== 5)
9090 && (geo
->layout
== 0)) {
9091 /* reshape 4 -> 5 */
9096 Name
" Error. Layout Migration from %d to %d "
9098 info
.array
.layout
, geo
->layout
);
9100 goto analyse_change_exit
;
9103 geo
->layout
= info
.array
.layout
;
9105 if ((geo
->chunksize
> 0) && (geo
->chunksize
!= UnSet
)
9106 && (geo
->chunksize
!= info
.array
.chunk_size
))
9107 change
= CH_MIGRATION
;
9109 geo
->chunksize
= info
.array
.chunk_size
;
9111 chunk
= geo
->chunksize
/ 1024;
9112 if (!validate_geometry_imsm(st
,
9115 geo
->raid_disks
+ devNumChange
,
9122 struct intel_super
*super
= st
->sb
;
9123 struct imsm_super
*mpb
= super
->anchor
;
9125 if (mpb
->num_raid_devs
> 1) {
9127 Name
" Error. Cannot perform operation on %s"
9128 "- for this operation it MUST be single "
9129 "array in container\n",
9135 analyse_change_exit
:
9140 int imsm_takeover(struct supertype
*st
, struct geo_params
*geo
)
9142 struct intel_super
*super
= st
->sb
;
9143 struct imsm_update_takeover
*u
;
9145 u
= malloc(sizeof(struct imsm_update_takeover
));
9149 u
->type
= update_takeover
;
9150 u
->subarray
= super
->current_vol
;
9152 /* 10->0 transition */
9153 if (geo
->level
== 0)
9154 u
->direction
= R10_TO_R0
;
9156 /* 0->10 transition */
9157 if (geo
->level
== 10)
9158 u
->direction
= R0_TO_R10
;
9160 /* update metadata locally */
9161 imsm_update_metadata_locally(st
, u
,
9162 sizeof(struct imsm_update_takeover
));
9163 /* and possibly remotely */
9164 if (st
->update_tail
)
9165 append_metadata_update(st
, u
,
9166 sizeof(struct imsm_update_takeover
));
9173 static int imsm_reshape_super(struct supertype
*st
, long long size
, int level
,
9174 int layout
, int chunksize
, int raid_disks
,
9175 int delta_disks
, char *backup
, char *dev
,
9179 struct geo_params geo
;
9181 dprintf("imsm: reshape_super called.\n");
9183 memset(&geo
, 0, sizeof(struct geo_params
));
9186 geo
.dev_id
= st
->devnum
;
9189 geo
.layout
= layout
;
9190 geo
.chunksize
= chunksize
;
9191 geo
.raid_disks
= raid_disks
;
9192 if (delta_disks
!= UnSet
)
9193 geo
.raid_disks
+= delta_disks
;
9195 dprintf("\tfor level : %i\n", geo
.level
);
9196 dprintf("\tfor raid_disks : %i\n", geo
.raid_disks
);
9198 if (experimental() == 0)
9201 if (st
->container_dev
== st
->devnum
) {
9202 /* On container level we can only increase number of devices. */
9203 dprintf("imsm: info: Container operation\n");
9204 int old_raid_disks
= 0;
9206 if (imsm_reshape_is_allowed_on_container(
9207 st
, &geo
, &old_raid_disks
)) {
9208 struct imsm_update_reshape
*u
= NULL
;
9211 len
= imsm_create_metadata_update_for_reshape(
9212 st
, &geo
, old_raid_disks
, &u
);
9215 dprintf("imsm: Cannot prepare update\n");
9216 goto exit_imsm_reshape_super
;
9220 /* update metadata locally */
9221 imsm_update_metadata_locally(st
, u
, len
);
9222 /* and possibly remotely */
9223 if (st
->update_tail
)
9224 append_metadata_update(st
, u
, len
);
9229 fprintf(stderr
, Name
": (imsm) Operation "
9230 "is not allowed on this container\n");
9233 /* On volume level we support following operations
9234 * - takeover: raid10 -> raid0; raid0 -> raid10
9235 * - chunk size migration
9236 * - migration: raid5 -> raid0; raid0 -> raid5
9238 struct intel_super
*super
= st
->sb
;
9239 struct intel_dev
*dev
= super
->devlist
;
9241 dprintf("imsm: info: Volume operation\n");
9242 /* find requested device */
9244 if (imsm_find_array_minor_by_subdev(
9245 dev
->index
, st
->container_dev
, &devnum
) == 0
9246 && devnum
== geo
.dev_id
)
9251 fprintf(stderr
, Name
" Cannot find %s (%i) subarray\n",
9252 geo
.dev_name
, geo
.dev_id
);
9253 goto exit_imsm_reshape_super
;
9255 super
->current_vol
= dev
->index
;
9256 change
= imsm_analyze_change(st
, &geo
);
9259 ret_val
= imsm_takeover(st
, &geo
);
9261 case CH_MIGRATION
: {
9262 struct imsm_update_reshape_migration
*u
= NULL
;
9264 imsm_create_metadata_update_for_migration(
9268 "Cannot prepare update\n");
9272 /* update metadata locally */
9273 imsm_update_metadata_locally(st
, u
, len
);
9274 /* and possibly remotely */
9275 if (st
->update_tail
)
9276 append_metadata_update(st
, u
, len
);
9286 exit_imsm_reshape_super
:
9287 dprintf("imsm: reshape_super Exit code = %i\n", ret_val
);
9291 /*******************************************************************************
9292 * Function: wait_for_reshape_imsm
9293 * Description: Function writes new sync_max value and waits until
9294 * reshape process reach new position
9296 * sra : general array info
9297 * ndata : number of disks in new array's layout
9300 * 1 : there is no reshape in progress,
9302 ******************************************************************************/
9303 int wait_for_reshape_imsm(struct mdinfo
*sra
, int ndata
)
9305 int fd
= sysfs_get_fd(sra
, NULL
, "reshape_position");
9306 unsigned long long completed
;
9307 /* to_complete : new sync_max position */
9308 unsigned long long to_complete
= sra
->reshape_progress
;
9309 unsigned long long position_to_set
= to_complete
/ ndata
;
9312 dprintf("imsm: wait_for_reshape_imsm() "
9313 "cannot open reshape_position\n");
9317 if (sysfs_fd_get_ll(fd
, &completed
) < 0) {
9318 dprintf("imsm: wait_for_reshape_imsm() "
9319 "cannot read reshape_position (no reshape in progres)\n");
9324 if (completed
> to_complete
) {
9325 dprintf("imsm: wait_for_reshape_imsm() "
9326 "wrong next position to set %llu (%llu)\n",
9327 to_complete
, completed
);
9331 dprintf("Position set: %llu\n", position_to_set
);
9332 if (sysfs_set_num(sra
, NULL
, "sync_max",
9333 position_to_set
) != 0) {
9334 dprintf("imsm: wait_for_reshape_imsm() "
9335 "cannot set reshape position to %llu\n",
9346 select(fd
+1, &rfds
, NULL
, NULL
, NULL
);
9347 if (sysfs_get_str(sra
, NULL
, "sync_action",
9349 strncmp(action
, "reshape", 7) != 0)
9351 if (sysfs_fd_get_ll(fd
, &completed
) < 0) {
9352 dprintf("imsm: wait_for_reshape_imsm() "
9353 "cannot read reshape_position (in loop)\n");
9357 } while (completed
< to_complete
);
9363 /*******************************************************************************
9364 * Function: check_degradation_change
9365 * Description: Check that array hasn't become failed.
9367 * info : for sysfs access
9368 * sources : source disks descriptors
9369 * degraded: previous degradation level
9372 ******************************************************************************/
9373 int check_degradation_change(struct mdinfo
*info
,
9377 unsigned long long new_degraded
;
9378 sysfs_get_ll(info
, NULL
, "degraded", &new_degraded
);
9379 if (new_degraded
!= (unsigned long long)degraded
) {
9380 /* check each device to ensure it is still working */
9383 for (sd
= info
->devs
; sd
; sd
= sd
->next
) {
9384 if (sd
->disk
.state
& (1<<MD_DISK_FAULTY
))
9386 if (sd
->disk
.state
& (1<<MD_DISK_SYNC
)) {
9388 if (sysfs_get_str(info
,
9389 sd
, "state", sbuf
, 20) < 0 ||
9390 strstr(sbuf
, "faulty") ||
9391 strstr(sbuf
, "in_sync") == NULL
) {
9392 /* this device is dead */
9393 sd
->disk
.state
= (1<<MD_DISK_FAULTY
);
9394 if (sd
->disk
.raid_disk
>= 0 &&
9395 sources
[sd
->disk
.raid_disk
] >= 0) {
9397 sd
->disk
.raid_disk
]);
9398 sources
[sd
->disk
.raid_disk
] =
9407 return new_degraded
;
9410 /*******************************************************************************
9411 * Function: imsm_manage_reshape
9412 * Description: Function finds array under reshape and it manages reshape
9413 * process. It creates stripes backups (if required) and sets
9416 * afd : Backup handle (nattive) - not used
9417 * sra : general array info
9418 * reshape : reshape parameters - not used
9419 * st : supertype structure
9420 * blocks : size of critical section [blocks]
9421 * fds : table of source device descriptor
9422 * offsets : start of array (offest per devices)
9424 * destfd : table of destination device descriptor
9425 * destoffsets : table of destination offsets (per device)
9427 * 1 : success, reshape is done
9429 ******************************************************************************/
9430 static int imsm_manage_reshape(
9431 int afd
, struct mdinfo
*sra
, struct reshape
*reshape
,
9432 struct supertype
*st
, unsigned long backup_blocks
,
9433 int *fds
, unsigned long long *offsets
,
9434 int dests
, int *destfd
, unsigned long long *destoffsets
)
9437 struct intel_super
*super
= st
->sb
;
9438 struct intel_dev
*dv
= NULL
;
9439 struct imsm_dev
*dev
= NULL
;
9440 struct imsm_map
*map_src
;
9441 int migr_vol_qan
= 0;
9442 int ndata
, odata
; /* [bytes] */
9443 int chunk
; /* [bytes] */
9444 struct migr_record
*migr_rec
;
9446 unsigned int buf_size
; /* [bytes] */
9447 unsigned long long max_position
; /* array size [bytes] */
9448 unsigned long long next_step
; /* [blocks]/[bytes] */
9449 unsigned long long old_data_stripe_length
;
9450 unsigned long long start_src
; /* [bytes] */
9451 unsigned long long start
; /* [bytes] */
9452 unsigned long long start_buf_shift
; /* [bytes] */
9454 int source_layout
= 0;
9456 if (!fds
|| !offsets
|| !sra
)
9459 /* Find volume during the reshape */
9460 for (dv
= super
->devlist
; dv
; dv
= dv
->next
) {
9461 if (dv
->dev
->vol
.migr_type
== MIGR_GEN_MIGR
9462 && dv
->dev
->vol
.migr_state
== 1) {
9467 /* Only one volume can migrate at the same time */
9468 if (migr_vol_qan
!= 1) {
9469 fprintf(stderr
, Name
" : %s", migr_vol_qan
?
9470 "Number of migrating volumes greater than 1\n" :
9471 "There is no volume during migrationg\n");
9475 map_src
= get_imsm_map(dev
, MAP_1
);
9476 if (map_src
== NULL
)
9479 ndata
= imsm_num_data_members(dev
, MAP_0
);
9480 odata
= imsm_num_data_members(dev
, MAP_1
);
9482 chunk
= __le16_to_cpu(map_src
->blocks_per_strip
) * 512;
9483 old_data_stripe_length
= odata
* chunk
;
9485 migr_rec
= super
->migr_rec
;
9487 /* initialize migration record for start condition */
9488 if (sra
->reshape_progress
== 0)
9489 init_migr_record_imsm(st
, dev
, sra
);
9491 if (__le32_to_cpu(migr_rec
->rec_status
) != UNIT_SRC_NORMAL
) {
9492 dprintf("imsm: cannot restart migration when data "
9493 "are present in copy area.\n");
9499 buf_size
= __le32_to_cpu(migr_rec
->blocks_per_unit
) * 512;
9500 /* extend buffer size for parity disk */
9501 buf_size
+= __le32_to_cpu(migr_rec
->dest_depth_per_unit
) * 512;
9502 /* add space for stripe aligment */
9503 buf_size
+= old_data_stripe_length
;
9504 if (posix_memalign((void **)&buf
, 4096, buf_size
)) {
9505 dprintf("imsm: Cannot allocate checpoint buffer\n");
9509 max_position
= sra
->component_size
* ndata
;
9510 source_layout
= imsm_level_to_layout(map_src
->raid_level
);
9512 while (__le32_to_cpu(migr_rec
->curr_migr_unit
) <
9513 __le32_to_cpu(migr_rec
->num_migr_units
)) {
9514 /* current reshape position [blocks] */
9515 unsigned long long current_position
=
9516 __le32_to_cpu(migr_rec
->blocks_per_unit
)
9517 * __le32_to_cpu(migr_rec
->curr_migr_unit
);
9518 unsigned long long border
;
9520 /* Check that array hasn't become failed.
9522 degraded
= check_degradation_change(sra
, fds
, degraded
);
9524 dprintf("imsm: Abort reshape due to degradation"
9525 " level (%i)\n", degraded
);
9529 next_step
= __le32_to_cpu(migr_rec
->blocks_per_unit
);
9531 if ((current_position
+ next_step
) > max_position
)
9532 next_step
= max_position
- current_position
;
9534 start
= current_position
* 512;
9536 /* allign reading start to old geometry */
9537 start_buf_shift
= start
% old_data_stripe_length
;
9538 start_src
= start
- start_buf_shift
;
9540 border
= (start_src
/ odata
) - (start
/ ndata
);
9542 if (border
<= __le32_to_cpu(migr_rec
->dest_depth_per_unit
)) {
9543 /* save critical stripes to buf
9544 * start - start address of current unit
9546 * start_src - start address of current unit
9547 * to backup alligned to source array
9550 unsigned long long next_step_filler
= 0;
9551 unsigned long long copy_length
= next_step
* 512;
9553 /* allign copy area length to stripe in old geometry */
9554 next_step_filler
= ((copy_length
+ start_buf_shift
)
9555 % old_data_stripe_length
);
9556 if (next_step_filler
)
9557 next_step_filler
= (old_data_stripe_length
9558 - next_step_filler
);
9559 dprintf("save_stripes() parameters: start = %llu,"
9560 "\tstart_src = %llu,\tnext_step*512 = %llu,"
9561 "\tstart_in_buf_shift = %llu,"
9562 "\tnext_step_filler = %llu\n",
9563 start
, start_src
, copy_length
,
9564 start_buf_shift
, next_step_filler
);
9566 if (save_stripes(fds
, offsets
, map_src
->num_members
,
9567 chunk
, map_src
->raid_level
,
9568 source_layout
, 0, NULL
, start_src
,
9570 next_step_filler
+ start_buf_shift
,
9572 dprintf("imsm: Cannot save stripes"
9576 /* Convert data to destination format and store it
9577 * in backup general migration area
9579 if (save_backup_imsm(st
, dev
, sra
,
9580 buf
+ start_buf_shift
, copy_length
)) {
9581 dprintf("imsm: Cannot save stripes to "
9582 "target devices\n");
9585 if (save_checkpoint_imsm(st
, sra
,
9586 UNIT_SRC_IN_CP_AREA
)) {
9587 dprintf("imsm: Cannot write checkpoint to "
9588 "migration record (UNIT_SRC_IN_CP_AREA)\n");
9592 /* set next step to use whole border area */
9593 border
/= next_step
;
9595 next_step
*= border
;
9597 /* When data backed up, checkpoint stored,
9598 * kick the kernel to reshape unit of data
9600 next_step
= next_step
+ sra
->reshape_progress
;
9601 /* limit next step to array max position */
9602 if (next_step
> max_position
)
9603 next_step
= max_position
;
9604 sysfs_set_num(sra
, NULL
, "suspend_lo", sra
->reshape_progress
);
9605 sysfs_set_num(sra
, NULL
, "suspend_hi", next_step
);
9606 sra
->reshape_progress
= next_step
;
9608 /* wait until reshape finish */
9609 if (wait_for_reshape_imsm(sra
, ndata
) < 0) {
9610 dprintf("wait_for_reshape_imsm returned error!\n");
9614 if (save_checkpoint_imsm(st
, sra
, UNIT_SRC_NORMAL
) == 1) {
9615 /* ignore error == 2, this can mean end of reshape here
9617 dprintf("imsm: Cannot write checkpoint to "
9618 "migration record (UNIT_SRC_NORMAL)\n");
9624 /* return '1' if done */
9632 #endif /* MDASSEMBLE */
9634 struct superswitch super_imsm
= {
9636 .examine_super
= examine_super_imsm
,
9637 .brief_examine_super
= brief_examine_super_imsm
,
9638 .brief_examine_subarrays
= brief_examine_subarrays_imsm
,
9639 .export_examine_super
= export_examine_super_imsm
,
9640 .detail_super
= detail_super_imsm
,
9641 .brief_detail_super
= brief_detail_super_imsm
,
9642 .write_init_super
= write_init_super_imsm
,
9643 .validate_geometry
= validate_geometry_imsm
,
9644 .add_to_super
= add_to_super_imsm
,
9645 .remove_from_super
= remove_from_super_imsm
,
9646 .detail_platform
= detail_platform_imsm
,
9647 .kill_subarray
= kill_subarray_imsm
,
9648 .update_subarray
= update_subarray_imsm
,
9649 .load_container
= load_container_imsm
,
9650 .default_geometry
= default_geometry_imsm
,
9651 .get_disk_controller_domain
= imsm_get_disk_controller_domain
,
9652 .reshape_super
= imsm_reshape_super
,
9653 .manage_reshape
= imsm_manage_reshape
,
9654 .recover_backup
= recover_backup_imsm
,
9656 .match_home
= match_home_imsm
,
9657 .uuid_from_super
= uuid_from_super_imsm
,
9658 .getinfo_super
= getinfo_super_imsm
,
9659 .getinfo_super_disks
= getinfo_super_disks_imsm
,
9660 .update_super
= update_super_imsm
,
9662 .avail_size
= avail_size_imsm
,
9663 .min_acceptable_spare_size
= min_acceptable_spare_size_imsm
,
9665 .compare_super
= compare_super_imsm
,
9667 .load_super
= load_super_imsm
,
9668 .init_super
= init_super_imsm
,
9669 .store_super
= store_super_imsm
,
9670 .free_super
= free_super_imsm
,
9671 .match_metadata_desc
= match_metadata_desc_imsm
,
9672 .container_content
= container_content_imsm
,
9680 .open_new
= imsm_open_new
,
9681 .set_array_state
= imsm_set_array_state
,
9682 .set_disk
= imsm_set_disk
,
9683 .sync_metadata
= imsm_sync_metadata
,
9684 .activate_spare
= imsm_activate_spare
,
9685 .process_update
= imsm_process_update
,
9686 .prepare_update
= imsm_prepare_update
,
9687 #endif /* MDASSEMBLE */