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_lo
; /* 0xE8 - 0xEB total blocks lo */
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 __u32 total_blocks_hi
; /* 0xF4 - 0xF5 total blocks hi */
106 #define IMSM_DISK_FILLERS 3
107 __u32 filler
[IMSM_DISK_FILLERS
]; /* 0xF5 - 0x107 MPB_DISK_FILLERS for future expansion */
110 /* map selector for map managment
116 /* RAID map configuration infos. */
118 __u32 pba_of_lba0_lo
; /* start address of partition */
119 __u32 blocks_per_member_lo
;/* blocks per member */
120 __u32 num_data_stripes_lo
; /* number of data stripes */
121 __u16 blocks_per_strip
;
122 __u8 map_state
; /* Normal, Uninitialized, Degraded, Failed */
123 #define IMSM_T_STATE_NORMAL 0
124 #define IMSM_T_STATE_UNINITIALIZED 1
125 #define IMSM_T_STATE_DEGRADED 2
126 #define IMSM_T_STATE_FAILED 3
128 #define IMSM_T_RAID0 0
129 #define IMSM_T_RAID1 1
130 #define IMSM_T_RAID5 5 /* since metadata version 1.2.02 ? */
131 __u8 num_members
; /* number of member disks */
132 __u8 num_domains
; /* number of parity domains */
133 __u8 failed_disk_num
; /* valid only when state is degraded */
135 __u32 pba_of_lba0_hi
;
136 __u32 blocks_per_member_hi
;
137 __u32 num_data_stripes_hi
;
138 __u32 filler
[4]; /* expansion area */
139 #define IMSM_ORD_REBUILD (1 << 24)
140 __u32 disk_ord_tbl
[1]; /* disk_ord_tbl[num_members],
141 * top byte contains some flags
143 } __attribute__ ((packed
));
146 __u32 curr_migr_unit
;
147 __u32 checkpoint_id
; /* id to access curr_migr_unit */
148 __u8 migr_state
; /* Normal or Migrating */
150 #define MIGR_REBUILD 1
151 #define MIGR_VERIFY 2 /* analagous to echo check > sync_action */
152 #define MIGR_GEN_MIGR 3
153 #define MIGR_STATE_CHANGE 4
154 #define MIGR_REPAIR 5
155 __u8 migr_type
; /* Initializing, Rebuilding, ... */
157 __u8 fs_state
; /* fast-sync state for CnG (0xff == disabled) */
158 __u16 verify_errors
; /* number of mismatches */
159 __u16 bad_blocks
; /* number of bad blocks during verify */
161 struct imsm_map map
[1];
162 /* here comes another one if migr_state */
163 } __attribute__ ((packed
));
166 __u8 volume
[MAX_RAID_SERIAL_LEN
];
169 #define DEV_BOOTABLE __cpu_to_le32(0x01)
170 #define DEV_BOOT_DEVICE __cpu_to_le32(0x02)
171 #define DEV_READ_COALESCING __cpu_to_le32(0x04)
172 #define DEV_WRITE_COALESCING __cpu_to_le32(0x08)
173 #define DEV_LAST_SHUTDOWN_DIRTY __cpu_to_le32(0x10)
174 #define DEV_HIDDEN_AT_BOOT __cpu_to_le32(0x20)
175 #define DEV_CURRENTLY_HIDDEN __cpu_to_le32(0x40)
176 #define DEV_VERIFY_AND_FIX __cpu_to_le32(0x80)
177 #define DEV_MAP_STATE_UNINIT __cpu_to_le32(0x100)
178 #define DEV_NO_AUTO_RECOVERY __cpu_to_le32(0x200)
179 #define DEV_CLONE_N_GO __cpu_to_le32(0x400)
180 #define DEV_CLONE_MAN_SYNC __cpu_to_le32(0x800)
181 #define DEV_CNG_MASTER_DISK_NUM __cpu_to_le32(0x1000)
182 __u32 status
; /* Persistent RaidDev status */
183 __u32 reserved_blocks
; /* Reserved blocks at beginning of volume */
187 __u8 cng_master_disk
;
191 #define IMSM_DEV_FILLERS 10
192 __u32 filler
[IMSM_DEV_FILLERS
];
194 } __attribute__ ((packed
));
197 __u8 sig
[MAX_SIGNATURE_LENGTH
]; /* 0x00 - 0x1F */
198 __u32 check_sum
; /* 0x20 - 0x23 MPB Checksum */
199 __u32 mpb_size
; /* 0x24 - 0x27 Size of MPB */
200 __u32 family_num
; /* 0x28 - 0x2B Checksum from first time this config was written */
201 __u32 generation_num
; /* 0x2C - 0x2F Incremented each time this array's MPB is written */
202 __u32 error_log_size
; /* 0x30 - 0x33 in bytes */
203 __u32 attributes
; /* 0x34 - 0x37 */
204 __u8 num_disks
; /* 0x38 Number of configured disks */
205 __u8 num_raid_devs
; /* 0x39 Number of configured volumes */
206 __u8 error_log_pos
; /* 0x3A */
207 __u8 fill
[1]; /* 0x3B */
208 __u32 cache_size
; /* 0x3c - 0x40 in mb */
209 __u32 orig_family_num
; /* 0x40 - 0x43 original family num */
210 __u32 pwr_cycle_count
; /* 0x44 - 0x47 simulated power cycle count for array */
211 __u32 bbm_log_size
; /* 0x48 - 0x4B - size of bad Block Mgmt Log in bytes */
212 #define IMSM_FILLERS 35
213 __u32 filler
[IMSM_FILLERS
]; /* 0x4C - 0xD7 RAID_MPB_FILLERS */
214 struct imsm_disk disk
[1]; /* 0xD8 diskTbl[numDisks] */
215 /* here comes imsm_dev[num_raid_devs] */
216 /* here comes BBM logs */
217 } __attribute__ ((packed
));
219 #define BBM_LOG_MAX_ENTRIES 254
221 struct bbm_log_entry
{
222 __u64 defective_block_start
;
223 #define UNREADABLE 0xFFFFFFFF
224 __u32 spare_block_offset
;
225 __u16 remapped_marked_count
;
227 } __attribute__ ((__packed__
));
230 __u32 signature
; /* 0xABADB10C */
232 __u32 reserved_spare_block_count
; /* 0 */
233 __u32 reserved
; /* 0xFFFF */
234 __u64 first_spare_lba
;
235 struct bbm_log_entry mapped_block_entries
[BBM_LOG_MAX_ENTRIES
];
236 } __attribute__ ((__packed__
));
239 static char *map_state_str
[] = { "normal", "uninitialized", "degraded", "failed" };
242 #define RAID_DISK_RESERVED_BLOCKS_IMSM_HI 2209
244 #define GEN_MIGR_AREA_SIZE 2048 /* General Migration Copy Area size in blocks */
246 #define MIGR_REC_BUF_SIZE 512 /* size of migr_record i/o buffer */
247 #define MIGR_REC_POSITION 512 /* migr_record position offset on disk,
248 * MIGR_REC_BUF_SIZE <= MIGR_REC_POSITION
251 #define UNIT_SRC_NORMAL 0 /* Source data for curr_migr_unit must
252 * be recovered using srcMap */
253 #define UNIT_SRC_IN_CP_AREA 1 /* Source data for curr_migr_unit has
254 * already been migrated and must
255 * be recovered from checkpoint area */
257 __u32 rec_status
; /* Status used to determine how to restart
258 * migration in case it aborts
260 __u32 curr_migr_unit
; /* 0..numMigrUnits-1 */
261 __u32 family_num
; /* Family number of MPB
262 * containing the RaidDev
263 * that is migrating */
264 __u32 ascending_migr
; /* True if migrating in increasing
266 __u32 blocks_per_unit
; /* Num disk blocks per unit of operation */
267 __u32 dest_depth_per_unit
; /* Num member blocks each destMap
269 * advances per unit-of-operation */
270 __u32 ckpt_area_pba
; /* Pba of first block of ckpt copy area */
271 __u32 dest_1st_member_lba
; /* First member lba on first
272 * stripe of destination */
273 __u32 num_migr_units
; /* Total num migration units-of-op */
274 __u32 post_migr_vol_cap
; /* Size of volume after
275 * migration completes */
276 __u32 post_migr_vol_cap_hi
; /* Expansion space for LBA64 */
277 __u32 ckpt_read_disk_num
; /* Which member disk in destSubMap[0] the
278 * migration ckpt record was read from
279 * (for recovered migrations) */
280 } __attribute__ ((__packed__
));
285 * 2: metadata does not match
293 struct md_list
*next
;
296 #define pr_vrb(fmt, arg...) (void) (verbose && pr_err(fmt, ##arg))
298 static __u8
migr_type(struct imsm_dev
*dev
)
300 if (dev
->vol
.migr_type
== MIGR_VERIFY
&&
301 dev
->status
& DEV_VERIFY_AND_FIX
)
304 return dev
->vol
.migr_type
;
307 static void set_migr_type(struct imsm_dev
*dev
, __u8 migr_type
)
309 /* for compatibility with older oroms convert MIGR_REPAIR, into
310 * MIGR_VERIFY w/ DEV_VERIFY_AND_FIX status
312 if (migr_type
== MIGR_REPAIR
) {
313 dev
->vol
.migr_type
= MIGR_VERIFY
;
314 dev
->status
|= DEV_VERIFY_AND_FIX
;
316 dev
->vol
.migr_type
= migr_type
;
317 dev
->status
&= ~DEV_VERIFY_AND_FIX
;
321 static unsigned int sector_count(__u32 bytes
)
323 return ROUND_UP(bytes
, 512) / 512;
326 static unsigned int mpb_sectors(struct imsm_super
*mpb
)
328 return sector_count(__le32_to_cpu(mpb
->mpb_size
));
332 struct imsm_dev
*dev
;
333 struct intel_dev
*next
;
338 enum sys_dev_type type
;
341 struct intel_hba
*next
;
348 /* internal representation of IMSM metadata */
351 void *buf
; /* O_DIRECT buffer for reading/writing metadata */
352 struct imsm_super
*anchor
; /* immovable parameters */
355 void *migr_rec_buf
; /* buffer for I/O operations */
356 struct migr_record
*migr_rec
; /* migration record */
358 int clean_migration_record_by_mdmon
; /* when reshape is switched to next
359 array, it indicates that mdmon is allowed to clean migration
361 size_t len
; /* size of the 'buf' allocation */
362 void *next_buf
; /* for realloc'ing buf from the manager */
364 int updates_pending
; /* count of pending updates for mdmon */
365 int current_vol
; /* index of raid device undergoing creation */
366 unsigned long long create_offset
; /* common start for 'current_vol' */
367 __u32 random
; /* random data for seeding new family numbers */
368 struct intel_dev
*devlist
;
372 __u8 serial
[MAX_RAID_SERIAL_LEN
];
375 struct imsm_disk disk
;
378 struct extent
*e
; /* for determining freespace @ create */
379 int raiddisk
; /* slot to fill in autolayout */
381 } *disks
, *current_disk
;
382 struct dl
*disk_mgmt_list
; /* list of disks to add/remove while mdmon
384 struct dl
*missing
; /* disks removed while we weren't looking */
385 struct bbm_log
*bbm_log
;
386 struct intel_hba
*hba
; /* device path of the raid controller for this metadata */
387 const struct imsm_orom
*orom
; /* platform firmware support */
388 struct intel_super
*next
; /* (temp) list for disambiguating family_num */
392 struct imsm_disk disk
;
393 #define IMSM_UNKNOWN_OWNER (-1)
395 struct intel_disk
*next
;
399 unsigned long long start
, size
;
402 /* definitions of reshape process types */
403 enum imsm_reshape_type
{
409 /* definition of messages passed to imsm_process_update */
410 enum imsm_update_type
{
411 update_activate_spare
,
415 update_add_remove_disk
,
416 update_reshape_container_disks
,
417 update_reshape_migration
,
419 update_general_migration_checkpoint
,
423 struct imsm_update_activate_spare
{
424 enum imsm_update_type type
;
428 struct imsm_update_activate_spare
*next
;
434 unsigned long long size
;
441 enum takeover_direction
{
445 struct imsm_update_takeover
{
446 enum imsm_update_type type
;
448 enum takeover_direction direction
;
451 struct imsm_update_reshape
{
452 enum imsm_update_type type
;
456 int new_disks
[1]; /* new_raid_disks - old_raid_disks makedev number */
459 struct imsm_update_reshape_migration
{
460 enum imsm_update_type type
;
463 /* fields for array migration changes
470 int new_disks
[1]; /* new_raid_disks - old_raid_disks makedev number */
473 struct imsm_update_size_change
{
474 enum imsm_update_type type
;
479 struct imsm_update_general_migration_checkpoint
{
480 enum imsm_update_type type
;
481 __u32 curr_migr_unit
;
485 __u8 serial
[MAX_RAID_SERIAL_LEN
];
488 struct imsm_update_create_array
{
489 enum imsm_update_type type
;
494 struct imsm_update_kill_array
{
495 enum imsm_update_type type
;
499 struct imsm_update_rename_array
{
500 enum imsm_update_type type
;
501 __u8 name
[MAX_RAID_SERIAL_LEN
];
505 struct imsm_update_add_remove_disk
{
506 enum imsm_update_type type
;
509 static const char *_sys_dev_type
[] = {
510 [SYS_DEV_UNKNOWN
] = "Unknown",
511 [SYS_DEV_SAS
] = "SAS",
512 [SYS_DEV_SATA
] = "SATA",
513 [SYS_DEV_NVME
] = "NVMe"
516 const char *get_sys_dev_type(enum sys_dev_type type
)
518 if (type
>= SYS_DEV_MAX
)
519 type
= SYS_DEV_UNKNOWN
;
521 return _sys_dev_type
[type
];
524 static struct intel_hba
* alloc_intel_hba(struct sys_dev
*device
)
526 struct intel_hba
*result
= xmalloc(sizeof(*result
));
528 result
->type
= device
->type
;
529 result
->path
= xstrdup(device
->path
);
531 if (result
->path
&& (result
->pci_id
= strrchr(result
->path
, '/')) != NULL
)
537 static struct intel_hba
* find_intel_hba(struct intel_hba
*hba
, struct sys_dev
*device
)
539 struct intel_hba
*result
=NULL
;
540 for (result
= hba
; result
; result
= result
->next
) {
541 if (result
->type
== device
->type
&& strcmp(result
->path
, device
->path
) == 0)
547 static int attach_hba_to_super(struct intel_super
*super
, struct sys_dev
*device
)
549 struct intel_hba
*hba
;
551 /* check if disk attached to Intel HBA */
552 hba
= find_intel_hba(super
->hba
, device
);
555 /* Check if HBA is already attached to super */
556 if (super
->hba
== NULL
) {
557 super
->hba
= alloc_intel_hba(device
);
562 /* Intel metadata allows for all disks attached to the same type HBA.
563 * Do not support HBA types mixing
565 if (device
->type
!= hba
->type
)
568 /* Multiple same type HBAs can be used if they share the same OROM */
569 const struct imsm_orom
*device_orom
= get_orom_by_device_id(device
->dev_id
);
571 if (device_orom
!= super
->orom
)
577 hba
->next
= alloc_intel_hba(device
);
581 static struct sys_dev
* find_disk_attached_hba(int fd
, const char *devname
)
583 struct sys_dev
*list
, *elem
;
586 if ((list
= find_intel_devices()) == NULL
)
590 disk_path
= (char *) devname
;
592 disk_path
= diskfd_to_devpath(fd
);
597 for (elem
= list
; elem
; elem
= elem
->next
)
598 if (path_attached_to_hba(disk_path
, elem
->path
))
601 if (disk_path
!= devname
)
607 static int find_intel_hba_capability(int fd
, struct intel_super
*super
,
610 static struct supertype
*match_metadata_desc_imsm(char *arg
)
612 struct supertype
*st
;
614 if (strcmp(arg
, "imsm") != 0 &&
615 strcmp(arg
, "default") != 0
619 st
= xcalloc(1, sizeof(*st
));
620 st
->ss
= &super_imsm
;
621 st
->max_devs
= IMSM_MAX_DEVICES
;
622 st
->minor_version
= 0;
628 static __u8
*get_imsm_version(struct imsm_super
*mpb
)
630 return &mpb
->sig
[MPB_SIG_LEN
];
634 /* retrieve a disk directly from the anchor when the anchor is known to be
635 * up-to-date, currently only at load time
637 static struct imsm_disk
*__get_imsm_disk(struct imsm_super
*mpb
, __u8 index
)
639 if (index
>= mpb
->num_disks
)
641 return &mpb
->disk
[index
];
644 /* retrieve the disk description based on a index of the disk
647 static struct dl
*get_imsm_dl_disk(struct intel_super
*super
, __u8 index
)
651 for (d
= super
->disks
; d
; d
= d
->next
)
652 if (d
->index
== index
)
657 /* retrieve a disk from the parsed metadata */
658 static struct imsm_disk
*get_imsm_disk(struct intel_super
*super
, __u8 index
)
662 dl
= get_imsm_dl_disk(super
, index
);
669 /* generate a checksum directly from the anchor when the anchor is known to be
670 * up-to-date, currently only at load or write_super after coalescing
672 static __u32
__gen_imsm_checksum(struct imsm_super
*mpb
)
674 __u32 end
= mpb
->mpb_size
/ sizeof(end
);
675 __u32
*p
= (__u32
*) mpb
;
679 sum
+= __le32_to_cpu(*p
);
683 return sum
- __le32_to_cpu(mpb
->check_sum
);
686 static size_t sizeof_imsm_map(struct imsm_map
*map
)
688 return sizeof(struct imsm_map
) + sizeof(__u32
) * (map
->num_members
- 1);
691 struct imsm_map
*get_imsm_map(struct imsm_dev
*dev
, int second_map
)
693 /* A device can have 2 maps if it is in the middle of a migration.
695 * MAP_0 - we return the first map
696 * MAP_1 - we return the second map if it exists, else NULL
697 * MAP_X - we return the second map if it exists, else the first
699 struct imsm_map
*map
= &dev
->vol
.map
[0];
700 struct imsm_map
*map2
= NULL
;
702 if (dev
->vol
.migr_state
)
703 map2
= (void *)map
+ sizeof_imsm_map(map
);
705 switch (second_map
) {
722 /* return the size of the device.
723 * migr_state increases the returned size if map[0] were to be duplicated
725 static size_t sizeof_imsm_dev(struct imsm_dev
*dev
, int migr_state
)
727 size_t size
= sizeof(*dev
) - sizeof(struct imsm_map
) +
728 sizeof_imsm_map(get_imsm_map(dev
, MAP_0
));
730 /* migrating means an additional map */
731 if (dev
->vol
.migr_state
)
732 size
+= sizeof_imsm_map(get_imsm_map(dev
, MAP_1
));
734 size
+= sizeof_imsm_map(get_imsm_map(dev
, MAP_0
));
740 /* retrieve disk serial number list from a metadata update */
741 static struct disk_info
*get_disk_info(struct imsm_update_create_array
*update
)
744 struct disk_info
*inf
;
746 inf
= u
+ sizeof(*update
) - sizeof(struct imsm_dev
) +
747 sizeof_imsm_dev(&update
->dev
, 0);
753 static struct imsm_dev
*__get_imsm_dev(struct imsm_super
*mpb
, __u8 index
)
759 if (index
>= mpb
->num_raid_devs
)
762 /* devices start after all disks */
763 offset
= ((void *) &mpb
->disk
[mpb
->num_disks
]) - _mpb
;
765 for (i
= 0; i
<= index
; i
++)
767 return _mpb
+ offset
;
769 offset
+= sizeof_imsm_dev(_mpb
+ offset
, 0);
774 static struct imsm_dev
*get_imsm_dev(struct intel_super
*super
, __u8 index
)
776 struct intel_dev
*dv
;
778 if (index
>= super
->anchor
->num_raid_devs
)
780 for (dv
= super
->devlist
; dv
; dv
= dv
->next
)
781 if (dv
->index
== index
)
788 * == MAP_0 get first map
789 * == MAP_1 get second map
790 * == MAP_X than get map according to the current migr_state
792 static __u32
get_imsm_ord_tbl_ent(struct imsm_dev
*dev
,
796 struct imsm_map
*map
;
798 map
= get_imsm_map(dev
, second_map
);
800 /* top byte identifies disk under rebuild */
801 return __le32_to_cpu(map
->disk_ord_tbl
[slot
]);
804 #define ord_to_idx(ord) (((ord) << 8) >> 8)
805 static __u32
get_imsm_disk_idx(struct imsm_dev
*dev
, int slot
, int second_map
)
807 __u32 ord
= get_imsm_ord_tbl_ent(dev
, slot
, second_map
);
809 return ord_to_idx(ord
);
812 static void set_imsm_ord_tbl_ent(struct imsm_map
*map
, int slot
, __u32 ord
)
814 map
->disk_ord_tbl
[slot
] = __cpu_to_le32(ord
);
817 static int get_imsm_disk_slot(struct imsm_map
*map
, unsigned idx
)
822 for (slot
= 0; slot
< map
->num_members
; slot
++) {
823 ord
= __le32_to_cpu(map
->disk_ord_tbl
[slot
]);
824 if (ord_to_idx(ord
) == idx
)
831 static int get_imsm_raid_level(struct imsm_map
*map
)
833 if (map
->raid_level
== 1) {
834 if (map
->num_members
== 2)
840 return map
->raid_level
;
843 static int cmp_extent(const void *av
, const void *bv
)
845 const struct extent
*a
= av
;
846 const struct extent
*b
= bv
;
847 if (a
->start
< b
->start
)
849 if (a
->start
> b
->start
)
854 static int count_memberships(struct dl
*dl
, struct intel_super
*super
)
859 for (i
= 0; i
< super
->anchor
->num_raid_devs
; i
++) {
860 struct imsm_dev
*dev
= get_imsm_dev(super
, i
);
861 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
863 if (get_imsm_disk_slot(map
, dl
->index
) >= 0)
870 static __u32
imsm_min_reserved_sectors(struct intel_super
*super
);
872 static int split_ull(unsigned long long n
, __u32
*lo
, __u32
*hi
)
874 if (lo
== 0 || hi
== 0)
876 *lo
= __le32_to_cpu((unsigned)n
);
877 *hi
= __le32_to_cpu((unsigned)(n
>> 32));
881 static unsigned long long join_u32(__u32 lo
, __u32 hi
)
883 return (unsigned long long)__le32_to_cpu(lo
) |
884 (((unsigned long long)__le32_to_cpu(hi
)) << 32);
887 static unsigned long long total_blocks(struct imsm_disk
*disk
)
891 return join_u32(disk
->total_blocks_lo
, disk
->total_blocks_hi
);
894 static unsigned long long pba_of_lba0(struct imsm_map
*map
)
898 return join_u32(map
->pba_of_lba0_lo
, map
->pba_of_lba0_hi
);
901 static unsigned long long blocks_per_member(struct imsm_map
*map
)
905 return join_u32(map
->blocks_per_member_lo
, map
->blocks_per_member_hi
);
909 static unsigned long long num_data_stripes(struct imsm_map
*map
)
913 return join_u32(map
->num_data_stripes_lo
, map
->num_data_stripes_hi
);
916 static void set_total_blocks(struct imsm_disk
*disk
, unsigned long long n
)
918 split_ull(n
, &disk
->total_blocks_lo
, &disk
->total_blocks_hi
);
922 static void set_pba_of_lba0(struct imsm_map
*map
, unsigned long long n
)
924 split_ull(n
, &map
->pba_of_lba0_lo
, &map
->pba_of_lba0_hi
);
927 static void set_blocks_per_member(struct imsm_map
*map
, unsigned long long n
)
929 split_ull(n
, &map
->blocks_per_member_lo
, &map
->blocks_per_member_hi
);
932 static void set_num_data_stripes(struct imsm_map
*map
, unsigned long long n
)
934 split_ull(n
, &map
->num_data_stripes_lo
, &map
->num_data_stripes_hi
);
937 static struct extent
*get_extents(struct intel_super
*super
, struct dl
*dl
)
939 /* find a list of used extents on the given physical device */
940 struct extent
*rv
, *e
;
942 int memberships
= count_memberships(dl
, super
);
945 /* trim the reserved area for spares, so they can join any array
946 * regardless of whether the OROM has assigned sectors from the
947 * IMSM_RESERVED_SECTORS region
950 reservation
= imsm_min_reserved_sectors(super
);
952 reservation
= MPB_SECTOR_CNT
+ IMSM_RESERVED_SECTORS
;
954 rv
= xcalloc(sizeof(struct extent
), (memberships
+ 1));
957 for (i
= 0; i
< super
->anchor
->num_raid_devs
; i
++) {
958 struct imsm_dev
*dev
= get_imsm_dev(super
, i
);
959 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
961 if (get_imsm_disk_slot(map
, dl
->index
) >= 0) {
962 e
->start
= pba_of_lba0(map
);
963 e
->size
= blocks_per_member(map
);
967 qsort(rv
, memberships
, sizeof(*rv
), cmp_extent
);
969 /* determine the start of the metadata
970 * when no raid devices are defined use the default
971 * ...otherwise allow the metadata to truncate the value
972 * as is the case with older versions of imsm
975 struct extent
*last
= &rv
[memberships
- 1];
976 unsigned long long remainder
;
978 remainder
= total_blocks(&dl
->disk
) - (last
->start
+ last
->size
);
979 /* round down to 1k block to satisfy precision of the kernel
983 /* make sure remainder is still sane */
984 if (remainder
< (unsigned)ROUND_UP(super
->len
, 512) >> 9)
985 remainder
= ROUND_UP(super
->len
, 512) >> 9;
986 if (reservation
> remainder
)
987 reservation
= remainder
;
989 e
->start
= total_blocks(&dl
->disk
) - reservation
;
994 /* try to determine how much space is reserved for metadata from
995 * the last get_extents() entry, otherwise fallback to the
998 static __u32
imsm_reserved_sectors(struct intel_super
*super
, struct dl
*dl
)
1004 /* for spares just return a minimal reservation which will grow
1005 * once the spare is picked up by an array
1007 if (dl
->index
== -1)
1008 return MPB_SECTOR_CNT
;
1010 e
= get_extents(super
, dl
);
1012 return MPB_SECTOR_CNT
+ IMSM_RESERVED_SECTORS
;
1014 /* scroll to last entry */
1015 for (i
= 0; e
[i
].size
; i
++)
1018 rv
= total_blocks(&dl
->disk
) - e
[i
].start
;
1025 static int is_spare(struct imsm_disk
*disk
)
1027 return (disk
->status
& SPARE_DISK
) == SPARE_DISK
;
1030 static int is_configured(struct imsm_disk
*disk
)
1032 return (disk
->status
& CONFIGURED_DISK
) == CONFIGURED_DISK
;
1035 static int is_failed(struct imsm_disk
*disk
)
1037 return (disk
->status
& FAILED_DISK
) == FAILED_DISK
;
1040 /* try to determine how much space is reserved for metadata from
1041 * the last get_extents() entry on the smallest active disk,
1042 * otherwise fallback to the default
1044 static __u32
imsm_min_reserved_sectors(struct intel_super
*super
)
1048 unsigned long long min_active
;
1050 __u32 rv
= MPB_SECTOR_CNT
+ IMSM_RESERVED_SECTORS
;
1051 struct dl
*dl
, *dl_min
= NULL
;
1057 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
1060 unsigned long long blocks
= total_blocks(&dl
->disk
);
1061 if (blocks
< min_active
|| min_active
== 0) {
1063 min_active
= blocks
;
1069 /* find last lba used by subarrays on the smallest active disk */
1070 e
= get_extents(super
, dl_min
);
1073 for (i
= 0; e
[i
].size
; i
++)
1076 remainder
= min_active
- e
[i
].start
;
1079 /* to give priority to recovery we should not require full
1080 IMSM_RESERVED_SECTORS from the spare */
1081 rv
= MPB_SECTOR_CNT
+ NUM_BLOCKS_DIRTY_STRIPE_REGION
;
1083 /* if real reservation is smaller use that value */
1084 return (remainder
< rv
) ? remainder
: rv
;
1087 /* Return minimum size of a spare that can be used in this array*/
1088 static unsigned long long min_acceptable_spare_size_imsm(struct supertype
*st
)
1090 struct intel_super
*super
= st
->sb
;
1094 unsigned long long rv
= 0;
1098 /* find first active disk in array */
1100 while (dl
&& (is_failed(&dl
->disk
) || dl
->index
== -1))
1104 /* find last lba used by subarrays */
1105 e
= get_extents(super
, dl
);
1108 for (i
= 0; e
[i
].size
; i
++)
1111 rv
= e
[i
-1].start
+ e
[i
-1].size
;
1114 /* add the amount of space needed for metadata */
1115 rv
= rv
+ imsm_min_reserved_sectors(super
);
1120 static int is_gen_migration(struct imsm_dev
*dev
);
1123 static __u64
blocks_per_migr_unit(struct intel_super
*super
,
1124 struct imsm_dev
*dev
);
1126 static void print_imsm_dev(struct intel_super
*super
,
1127 struct imsm_dev
*dev
,
1133 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
1134 struct imsm_map
*map2
= get_imsm_map(dev
, MAP_1
);
1138 printf("[%.16s]:\n", dev
->volume
);
1139 printf(" UUID : %s\n", uuid
);
1140 printf(" RAID Level : %d", get_imsm_raid_level(map
));
1142 printf(" <-- %d", get_imsm_raid_level(map2
));
1144 printf(" Members : %d", map
->num_members
);
1146 printf(" <-- %d", map2
->num_members
);
1148 printf(" Slots : [");
1149 for (i
= 0; i
< map
->num_members
; i
++) {
1150 ord
= get_imsm_ord_tbl_ent(dev
, i
, MAP_0
);
1151 printf("%s", ord
& IMSM_ORD_REBUILD
? "_" : "U");
1156 for (i
= 0; i
< map2
->num_members
; i
++) {
1157 ord
= get_imsm_ord_tbl_ent(dev
, i
, MAP_1
);
1158 printf("%s", ord
& IMSM_ORD_REBUILD
? "_" : "U");
1163 printf(" Failed disk : ");
1164 if (map
->failed_disk_num
== 0xff)
1167 printf("%i", map
->failed_disk_num
);
1169 slot
= get_imsm_disk_slot(map
, disk_idx
);
1171 ord
= get_imsm_ord_tbl_ent(dev
, slot
, MAP_X
);
1172 printf(" This Slot : %d%s\n", slot
,
1173 ord
& IMSM_ORD_REBUILD
? " (out-of-sync)" : "");
1175 printf(" This Slot : ?\n");
1176 sz
= __le32_to_cpu(dev
->size_high
);
1178 sz
+= __le32_to_cpu(dev
->size_low
);
1179 printf(" Array Size : %llu%s\n", (unsigned long long)sz
,
1180 human_size(sz
* 512));
1181 sz
= blocks_per_member(map
);
1182 printf(" Per Dev Size : %llu%s\n", (unsigned long long)sz
,
1183 human_size(sz
* 512));
1184 printf(" Sector Offset : %llu\n",
1186 printf(" Num Stripes : %llu\n",
1187 num_data_stripes(map
));
1188 printf(" Chunk Size : %u KiB",
1189 __le16_to_cpu(map
->blocks_per_strip
) / 2);
1191 printf(" <-- %u KiB",
1192 __le16_to_cpu(map2
->blocks_per_strip
) / 2);
1194 printf(" Reserved : %d\n", __le32_to_cpu(dev
->reserved_blocks
));
1195 printf(" Migrate State : ");
1196 if (dev
->vol
.migr_state
) {
1197 if (migr_type(dev
) == MIGR_INIT
)
1198 printf("initialize\n");
1199 else if (migr_type(dev
) == MIGR_REBUILD
)
1200 printf("rebuild\n");
1201 else if (migr_type(dev
) == MIGR_VERIFY
)
1203 else if (migr_type(dev
) == MIGR_GEN_MIGR
)
1204 printf("general migration\n");
1205 else if (migr_type(dev
) == MIGR_STATE_CHANGE
)
1206 printf("state change\n");
1207 else if (migr_type(dev
) == MIGR_REPAIR
)
1210 printf("<unknown:%d>\n", migr_type(dev
));
1213 printf(" Map State : %s", map_state_str
[map
->map_state
]);
1214 if (dev
->vol
.migr_state
) {
1215 struct imsm_map
*map
= get_imsm_map(dev
, MAP_1
);
1217 printf(" <-- %s", map_state_str
[map
->map_state
]);
1218 printf("\n Checkpoint : %u ",
1219 __le32_to_cpu(dev
->vol
.curr_migr_unit
));
1220 if ((is_gen_migration(dev
)) && ((slot
> 1) || (slot
< 0)))
1223 printf("(%llu)", (unsigned long long)
1224 blocks_per_migr_unit(super
, dev
));
1227 printf(" Dirty State : %s\n", dev
->vol
.dirty
? "dirty" : "clean");
1230 static void print_imsm_disk(struct imsm_disk
*disk
, int index
, __u32 reserved
)
1232 char str
[MAX_RAID_SERIAL_LEN
+ 1];
1235 if (index
< -1 || !disk
)
1239 snprintf(str
, MAX_RAID_SERIAL_LEN
+ 1, "%s", disk
->serial
);
1241 printf(" Disk%02d Serial : %s\n", index
, str
);
1243 printf(" Disk Serial : %s\n", str
);
1244 printf(" State :%s%s%s\n", is_spare(disk
) ? " spare" : "",
1245 is_configured(disk
) ? " active" : "",
1246 is_failed(disk
) ? " failed" : "");
1247 printf(" Id : %08x\n", __le32_to_cpu(disk
->scsi_id
));
1248 sz
= total_blocks(disk
) - reserved
;
1249 printf(" Usable Size : %llu%s\n", (unsigned long long)sz
,
1250 human_size(sz
* 512));
1253 void examine_migr_rec_imsm(struct intel_super
*super
)
1255 struct migr_record
*migr_rec
= super
->migr_rec
;
1256 struct imsm_super
*mpb
= super
->anchor
;
1259 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
1260 struct imsm_dev
*dev
= __get_imsm_dev(mpb
, i
);
1261 struct imsm_map
*map
;
1264 if (is_gen_migration(dev
) == 0)
1267 printf("\nMigration Record Information:");
1269 /* first map under migration */
1270 map
= get_imsm_map(dev
, MAP_0
);
1272 slot
= get_imsm_disk_slot(map
, super
->disks
->index
);
1273 if ((map
== NULL
) || (slot
> 1) || (slot
< 0)) {
1274 printf(" Empty\n ");
1275 printf("Examine one of first two disks in array\n");
1278 printf("\n Status : ");
1279 if (__le32_to_cpu(migr_rec
->rec_status
) == UNIT_SRC_NORMAL
)
1282 printf("Contains Data\n");
1283 printf(" Current Unit : %u\n",
1284 __le32_to_cpu(migr_rec
->curr_migr_unit
));
1285 printf(" Family : %u\n",
1286 __le32_to_cpu(migr_rec
->family_num
));
1287 printf(" Ascending : %u\n",
1288 __le32_to_cpu(migr_rec
->ascending_migr
));
1289 printf(" Blocks Per Unit : %u\n",
1290 __le32_to_cpu(migr_rec
->blocks_per_unit
));
1291 printf(" Dest. Depth Per Unit : %u\n",
1292 __le32_to_cpu(migr_rec
->dest_depth_per_unit
));
1293 printf(" Checkpoint Area pba : %u\n",
1294 __le32_to_cpu(migr_rec
->ckpt_area_pba
));
1295 printf(" First member lba : %u\n",
1296 __le32_to_cpu(migr_rec
->dest_1st_member_lba
));
1297 printf(" Total Number of Units : %u\n",
1298 __le32_to_cpu(migr_rec
->num_migr_units
));
1299 printf(" Size of volume : %u\n",
1300 __le32_to_cpu(migr_rec
->post_migr_vol_cap
));
1301 printf(" Expansion space for LBA64 : %u\n",
1302 __le32_to_cpu(migr_rec
->post_migr_vol_cap_hi
));
1303 printf(" Record was read from : %u\n",
1304 __le32_to_cpu(migr_rec
->ckpt_read_disk_num
));
1309 #endif /* MDASSEMBLE */
1310 /*******************************************************************************
1311 * function: imsm_check_attributes
1312 * Description: Function checks if features represented by attributes flags
1313 * are supported by mdadm.
1315 * attributes - Attributes read from metadata
1317 * 0 - passed attributes contains unsupported features flags
1318 * 1 - all features are supported
1319 ******************************************************************************/
1320 static int imsm_check_attributes(__u32 attributes
)
1323 __u32 not_supported
= MPB_ATTRIB_SUPPORTED
^0xffffffff;
1325 not_supported
&= ~MPB_ATTRIB_IGNORED
;
1327 not_supported
&= attributes
;
1328 if (not_supported
) {
1329 pr_err("(IMSM): Unsupported attributes : %x\n",
1330 (unsigned)__le32_to_cpu(not_supported
));
1331 if (not_supported
& MPB_ATTRIB_CHECKSUM_VERIFY
) {
1332 dprintf("\t\tMPB_ATTRIB_CHECKSUM_VERIFY \n");
1333 not_supported
^= MPB_ATTRIB_CHECKSUM_VERIFY
;
1335 if (not_supported
& MPB_ATTRIB_2TB
) {
1336 dprintf("\t\tMPB_ATTRIB_2TB\n");
1337 not_supported
^= MPB_ATTRIB_2TB
;
1339 if (not_supported
& MPB_ATTRIB_RAID0
) {
1340 dprintf("\t\tMPB_ATTRIB_RAID0\n");
1341 not_supported
^= MPB_ATTRIB_RAID0
;
1343 if (not_supported
& MPB_ATTRIB_RAID1
) {
1344 dprintf("\t\tMPB_ATTRIB_RAID1\n");
1345 not_supported
^= MPB_ATTRIB_RAID1
;
1347 if (not_supported
& MPB_ATTRIB_RAID10
) {
1348 dprintf("\t\tMPB_ATTRIB_RAID10\n");
1349 not_supported
^= MPB_ATTRIB_RAID10
;
1351 if (not_supported
& MPB_ATTRIB_RAID1E
) {
1352 dprintf("\t\tMPB_ATTRIB_RAID1E\n");
1353 not_supported
^= MPB_ATTRIB_RAID1E
;
1355 if (not_supported
& MPB_ATTRIB_RAID5
) {
1356 dprintf("\t\tMPB_ATTRIB_RAID5\n");
1357 not_supported
^= MPB_ATTRIB_RAID5
;
1359 if (not_supported
& MPB_ATTRIB_RAIDCNG
) {
1360 dprintf("\t\tMPB_ATTRIB_RAIDCNG\n");
1361 not_supported
^= MPB_ATTRIB_RAIDCNG
;
1363 if (not_supported
& MPB_ATTRIB_BBM
) {
1364 dprintf("\t\tMPB_ATTRIB_BBM\n");
1365 not_supported
^= MPB_ATTRIB_BBM
;
1367 if (not_supported
& MPB_ATTRIB_CHECKSUM_VERIFY
) {
1368 dprintf("\t\tMPB_ATTRIB_CHECKSUM_VERIFY (== MPB_ATTRIB_LEGACY)\n");
1369 not_supported
^= MPB_ATTRIB_CHECKSUM_VERIFY
;
1371 if (not_supported
& MPB_ATTRIB_EXP_STRIPE_SIZE
) {
1372 dprintf("\t\tMPB_ATTRIB_EXP_STRIP_SIZE\n");
1373 not_supported
^= MPB_ATTRIB_EXP_STRIPE_SIZE
;
1375 if (not_supported
& MPB_ATTRIB_2TB_DISK
) {
1376 dprintf("\t\tMPB_ATTRIB_2TB_DISK\n");
1377 not_supported
^= MPB_ATTRIB_2TB_DISK
;
1379 if (not_supported
& MPB_ATTRIB_NEVER_USE2
) {
1380 dprintf("\t\tMPB_ATTRIB_NEVER_USE2\n");
1381 not_supported
^= MPB_ATTRIB_NEVER_USE2
;
1383 if (not_supported
& MPB_ATTRIB_NEVER_USE
) {
1384 dprintf("\t\tMPB_ATTRIB_NEVER_USE\n");
1385 not_supported
^= MPB_ATTRIB_NEVER_USE
;
1389 dprintf("(IMSM): Unknown attributes : %x\n", not_supported
);
1398 static void getinfo_super_imsm(struct supertype
*st
, struct mdinfo
*info
, char *map
);
1400 static void examine_super_imsm(struct supertype
*st
, char *homehost
)
1402 struct intel_super
*super
= st
->sb
;
1403 struct imsm_super
*mpb
= super
->anchor
;
1404 char str
[MAX_SIGNATURE_LENGTH
];
1409 __u32 reserved
= imsm_reserved_sectors(super
, super
->disks
);
1412 snprintf(str
, MPB_SIG_LEN
, "%s", mpb
->sig
);
1413 printf(" Magic : %s\n", str
);
1414 snprintf(str
, strlen(MPB_VERSION_RAID0
), "%s", get_imsm_version(mpb
));
1415 printf(" Version : %s\n", get_imsm_version(mpb
));
1416 printf(" Orig Family : %08x\n", __le32_to_cpu(mpb
->orig_family_num
));
1417 printf(" Family : %08x\n", __le32_to_cpu(mpb
->family_num
));
1418 printf(" Generation : %08x\n", __le32_to_cpu(mpb
->generation_num
));
1419 printf(" Attributes : ");
1420 if (imsm_check_attributes(mpb
->attributes
))
1421 printf("All supported\n");
1423 printf("not supported\n");
1424 getinfo_super_imsm(st
, &info
, NULL
);
1425 fname_from_uuid(st
, &info
, nbuf
, ':');
1426 printf(" UUID : %s\n", nbuf
+ 5);
1427 sum
= __le32_to_cpu(mpb
->check_sum
);
1428 printf(" Checksum : %08x %s\n", sum
,
1429 __gen_imsm_checksum(mpb
) == sum
? "correct" : "incorrect");
1430 printf(" MPB Sectors : %d\n", mpb_sectors(mpb
));
1431 printf(" Disks : %d\n", mpb
->num_disks
);
1432 printf(" RAID Devices : %d\n", mpb
->num_raid_devs
);
1433 print_imsm_disk(__get_imsm_disk(mpb
, super
->disks
->index
), super
->disks
->index
, reserved
);
1434 if (super
->bbm_log
) {
1435 struct bbm_log
*log
= super
->bbm_log
;
1438 printf("Bad Block Management Log:\n");
1439 printf(" Log Size : %d\n", __le32_to_cpu(mpb
->bbm_log_size
));
1440 printf(" Signature : %x\n", __le32_to_cpu(log
->signature
));
1441 printf(" Entry Count : %d\n", __le32_to_cpu(log
->entry_count
));
1442 printf(" Spare Blocks : %d\n", __le32_to_cpu(log
->reserved_spare_block_count
));
1443 printf(" First Spare : %llx\n",
1444 (unsigned long long) __le64_to_cpu(log
->first_spare_lba
));
1446 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
1448 struct imsm_dev
*dev
= __get_imsm_dev(mpb
, i
);
1450 super
->current_vol
= i
;
1451 getinfo_super_imsm(st
, &info
, NULL
);
1452 fname_from_uuid(st
, &info
, nbuf
, ':');
1453 print_imsm_dev(super
, dev
, nbuf
+ 5, super
->disks
->index
);
1455 for (i
= 0; i
< mpb
->num_disks
; i
++) {
1456 if (i
== super
->disks
->index
)
1458 print_imsm_disk(__get_imsm_disk(mpb
, i
), i
, reserved
);
1461 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
1462 if (dl
->index
== -1)
1463 print_imsm_disk(&dl
->disk
, -1, reserved
);
1465 examine_migr_rec_imsm(super
);
1468 static void brief_examine_super_imsm(struct supertype
*st
, int verbose
)
1470 /* We just write a generic IMSM ARRAY entry */
1473 struct intel_super
*super
= st
->sb
;
1475 if (!super
->anchor
->num_raid_devs
) {
1476 printf("ARRAY metadata=imsm\n");
1480 getinfo_super_imsm(st
, &info
, NULL
);
1481 fname_from_uuid(st
, &info
, nbuf
, ':');
1482 printf("ARRAY metadata=imsm UUID=%s\n", nbuf
+ 5);
1485 static void brief_examine_subarrays_imsm(struct supertype
*st
, int verbose
)
1487 /* We just write a generic IMSM ARRAY entry */
1491 struct intel_super
*super
= st
->sb
;
1494 if (!super
->anchor
->num_raid_devs
)
1497 getinfo_super_imsm(st
, &info
, NULL
);
1498 fname_from_uuid(st
, &info
, nbuf
, ':');
1499 for (i
= 0; i
< super
->anchor
->num_raid_devs
; i
++) {
1500 struct imsm_dev
*dev
= get_imsm_dev(super
, i
);
1502 super
->current_vol
= i
;
1503 getinfo_super_imsm(st
, &info
, NULL
);
1504 fname_from_uuid(st
, &info
, nbuf1
, ':');
1505 printf("ARRAY /dev/md/%.16s container=%s member=%d UUID=%s\n",
1506 dev
->volume
, nbuf
+ 5, i
, nbuf1
+ 5);
1510 static void export_examine_super_imsm(struct supertype
*st
)
1512 struct intel_super
*super
= st
->sb
;
1513 struct imsm_super
*mpb
= super
->anchor
;
1517 getinfo_super_imsm(st
, &info
, NULL
);
1518 fname_from_uuid(st
, &info
, nbuf
, ':');
1519 printf("MD_METADATA=imsm\n");
1520 printf("MD_LEVEL=container\n");
1521 printf("MD_UUID=%s\n", nbuf
+5);
1522 printf("MD_DEVICES=%u\n", mpb
->num_disks
);
1525 static int copy_metadata_imsm(struct supertype
*st
, int from
, int to
)
1527 /* The second last 512byte sector of the device contains
1528 * the "struct imsm_super" metadata.
1529 * This contains mpb_size which is the size in bytes of the
1530 * extended metadata. This is located immediately before
1532 * We want to read all that, plus the last sector which
1533 * may contain a migration record, and write it all
1537 unsigned long long dsize
, offset
;
1539 struct imsm_super
*sb
;
1542 if (posix_memalign(&buf
, 4096, 4096) != 0)
1545 if (!get_dev_size(from
, NULL
, &dsize
))
1548 if (lseek64(from
, dsize
-1024, 0) < 0)
1550 if (read(from
, buf
, 512) != 512)
1553 if (strncmp((char*)sb
->sig
, MPB_SIGNATURE
, MPB_SIG_LEN
) != 0)
1556 sectors
= mpb_sectors(sb
) + 2;
1557 offset
= dsize
- sectors
* 512;
1558 if (lseek64(from
, offset
, 0) < 0 ||
1559 lseek64(to
, offset
, 0) < 0)
1561 while (written
< sectors
* 512) {
1562 int n
= sectors
*512 - written
;
1565 if (read(from
, buf
, n
) != n
)
1567 if (write(to
, buf
, n
) != n
)
1578 static void detail_super_imsm(struct supertype
*st
, char *homehost
)
1583 getinfo_super_imsm(st
, &info
, NULL
);
1584 fname_from_uuid(st
, &info
, nbuf
, ':');
1585 printf("\n UUID : %s\n", nbuf
+ 5);
1588 static void brief_detail_super_imsm(struct supertype
*st
)
1592 getinfo_super_imsm(st
, &info
, NULL
);
1593 fname_from_uuid(st
, &info
, nbuf
, ':');
1594 printf(" UUID=%s", nbuf
+ 5);
1597 static int imsm_read_serial(int fd
, char *devname
, __u8
*serial
);
1598 static void fd2devname(int fd
, char *name
);
1600 static int ahci_enumerate_ports(const char *hba_path
, int port_count
, int host_base
, int verbose
)
1602 /* dump an unsorted list of devices attached to AHCI Intel storage
1603 * controller, as well as non-connected ports
1605 int hba_len
= strlen(hba_path
) + 1;
1610 unsigned long port_mask
= (1 << port_count
) - 1;
1612 if (port_count
> (int)sizeof(port_mask
) * 8) {
1614 pr_err("port_count %d out of range\n", port_count
);
1618 /* scroll through /sys/dev/block looking for devices attached to
1621 dir
= opendir("/sys/dev/block");
1622 for (ent
= dir
? readdir(dir
) : NULL
; ent
; ent
= readdir(dir
)) {
1633 if (sscanf(ent
->d_name
, "%d:%d", &major
, &minor
) != 2)
1635 path
= devt_to_devpath(makedev(major
, minor
));
1638 if (!path_attached_to_hba(path
, hba_path
)) {
1644 /* retrieve the scsi device type */
1645 if (asprintf(&device
, "/sys/dev/block/%d:%d/device/xxxxxxx", major
, minor
) < 0) {
1647 pr_err("failed to allocate 'device'\n");
1651 sprintf(device
, "/sys/dev/block/%d:%d/device/type", major
, minor
);
1652 if (load_sys(device
, buf
) != 0) {
1654 pr_err("failed to read device type for %s\n",
1660 type
= strtoul(buf
, NULL
, 10);
1662 /* if it's not a disk print the vendor and model */
1663 if (!(type
== 0 || type
== 7 || type
== 14)) {
1666 sprintf(device
, "/sys/dev/block/%d:%d/device/vendor", major
, minor
);
1667 if (load_sys(device
, buf
) == 0) {
1668 strncpy(vendor
, buf
, sizeof(vendor
));
1669 vendor
[sizeof(vendor
) - 1] = '\0';
1670 c
= (char *) &vendor
[sizeof(vendor
) - 1];
1671 while (isspace(*c
) || *c
== '\0')
1675 sprintf(device
, "/sys/dev/block/%d:%d/device/model", major
, minor
);
1676 if (load_sys(device
, buf
) == 0) {
1677 strncpy(model
, buf
, sizeof(model
));
1678 model
[sizeof(model
) - 1] = '\0';
1679 c
= (char *) &model
[sizeof(model
) - 1];
1680 while (isspace(*c
) || *c
== '\0')
1684 if (vendor
[0] && model
[0])
1685 sprintf(buf
, "%.64s %.64s", vendor
, model
);
1687 switch (type
) { /* numbers from hald/linux/device.c */
1688 case 1: sprintf(buf
, "tape"); break;
1689 case 2: sprintf(buf
, "printer"); break;
1690 case 3: sprintf(buf
, "processor"); break;
1692 case 5: sprintf(buf
, "cdrom"); break;
1693 case 6: sprintf(buf
, "scanner"); break;
1694 case 8: sprintf(buf
, "media_changer"); break;
1695 case 9: sprintf(buf
, "comm"); break;
1696 case 12: sprintf(buf
, "raid"); break;
1697 default: sprintf(buf
, "unknown");
1703 /* chop device path to 'host%d' and calculate the port number */
1704 c
= strchr(&path
[hba_len
], '/');
1707 pr_err("%s - invalid path name\n", path
+ hba_len
);
1712 if ((sscanf(&path
[hba_len
], "ata%d", &port
) == 1) ||
1713 ((sscanf(&path
[hba_len
], "host%d", &port
) == 1)))
1717 *c
= '/'; /* repair the full string */
1718 pr_err("failed to determine port number for %s\n",
1725 /* mark this port as used */
1726 port_mask
&= ~(1 << port
);
1728 /* print out the device information */
1730 printf(" Port%d : - non-disk device (%s) -\n", port
, buf
);
1734 fd
= dev_open(ent
->d_name
, O_RDONLY
);
1736 printf(" Port%d : - disk info unavailable -\n", port
);
1738 fd2devname(fd
, buf
);
1739 printf(" Port%d : %s", port
, buf
);
1740 if (imsm_read_serial(fd
, NULL
, (__u8
*) buf
) == 0)
1741 printf(" (%.*s)\n", MAX_RAID_SERIAL_LEN
, buf
);
1756 for (i
= 0; i
< port_count
; i
++)
1757 if (port_mask
& (1 << i
))
1758 printf(" Port%d : - no device attached -\n", i
);
1764 static void print_found_intel_controllers(struct sys_dev
*elem
)
1766 for (; elem
; elem
= elem
->next
) {
1767 pr_err("found Intel(R) ");
1768 if (elem
->type
== SYS_DEV_SATA
)
1769 fprintf(stderr
, "SATA ");
1770 else if (elem
->type
== SYS_DEV_SAS
)
1771 fprintf(stderr
, "SAS ");
1772 else if (elem
->type
== SYS_DEV_NVME
)
1773 fprintf(stderr
, "NVMe ");
1774 fprintf(stderr
, "RAID controller");
1776 fprintf(stderr
, " at %s", elem
->pci_id
);
1777 fprintf(stderr
, ".\n");
1782 static int ahci_get_port_count(const char *hba_path
, int *port_count
)
1789 if ((dir
= opendir(hba_path
)) == NULL
)
1792 for (ent
= readdir(dir
); ent
; ent
= readdir(dir
)) {
1795 if ((sscanf(ent
->d_name
, "ata%d", &host
) != 1) &&
1796 ((sscanf(ent
->d_name
, "host%d", &host
) != 1)))
1798 if (*port_count
== 0)
1800 else if (host
< host_base
)
1803 if (host
+ 1 > *port_count
+ host_base
)
1804 *port_count
= host
+ 1 - host_base
;
1810 static void print_imsm_capability(const struct imsm_orom
*orom
)
1812 printf(" Platform : Intel(R) ");
1813 if (orom
->capabilities
== 0 && orom
->driver_features
== 0)
1814 printf("Matrix Storage Manager\n");
1816 printf("Rapid Storage Technology%s\n",
1817 imsm_orom_is_enterprise(orom
) ? " enterprise" : "");
1818 if (orom
->major_ver
|| orom
->minor_ver
|| orom
->hotfix_ver
|| orom
->build
)
1819 printf(" Version : %d.%d.%d.%d\n", orom
->major_ver
,
1820 orom
->minor_ver
, orom
->hotfix_ver
, orom
->build
);
1821 printf(" RAID Levels :%s%s%s%s%s\n",
1822 imsm_orom_has_raid0(orom
) ? " raid0" : "",
1823 imsm_orom_has_raid1(orom
) ? " raid1" : "",
1824 imsm_orom_has_raid1e(orom
) ? " raid1e" : "",
1825 imsm_orom_has_raid10(orom
) ? " raid10" : "",
1826 imsm_orom_has_raid5(orom
) ? " raid5" : "");
1827 printf(" Chunk Sizes :%s%s%s%s%s%s%s%s%s%s%s%s%s%s%s%s\n",
1828 imsm_orom_has_chunk(orom
, 2) ? " 2k" : "",
1829 imsm_orom_has_chunk(orom
, 4) ? " 4k" : "",
1830 imsm_orom_has_chunk(orom
, 8) ? " 8k" : "",
1831 imsm_orom_has_chunk(orom
, 16) ? " 16k" : "",
1832 imsm_orom_has_chunk(orom
, 32) ? " 32k" : "",
1833 imsm_orom_has_chunk(orom
, 64) ? " 64k" : "",
1834 imsm_orom_has_chunk(orom
, 128) ? " 128k" : "",
1835 imsm_orom_has_chunk(orom
, 256) ? " 256k" : "",
1836 imsm_orom_has_chunk(orom
, 512) ? " 512k" : "",
1837 imsm_orom_has_chunk(orom
, 1024*1) ? " 1M" : "",
1838 imsm_orom_has_chunk(orom
, 1024*2) ? " 2M" : "",
1839 imsm_orom_has_chunk(orom
, 1024*4) ? " 4M" : "",
1840 imsm_orom_has_chunk(orom
, 1024*8) ? " 8M" : "",
1841 imsm_orom_has_chunk(orom
, 1024*16) ? " 16M" : "",
1842 imsm_orom_has_chunk(orom
, 1024*32) ? " 32M" : "",
1843 imsm_orom_has_chunk(orom
, 1024*64) ? " 64M" : "");
1844 printf(" 2TB volumes :%s supported\n",
1845 (orom
->attr
& IMSM_OROM_ATTR_2TB
)?"":" not");
1846 printf(" 2TB disks :%s supported\n",
1847 (orom
->attr
& IMSM_OROM_ATTR_2TB_DISK
)?"":" not");
1848 printf(" Max Disks : %d\n", orom
->tds
);
1849 printf(" Max Volumes : %d per array, %d per %s\n",
1850 orom
->vpa
, orom
->vphba
,
1851 imsm_orom_is_nvme(orom
) ? "platform" : "controller");
1855 static void print_imsm_capability_export(const struct imsm_orom
*orom
)
1857 printf("MD_FIRMWARE_TYPE=imsm\n");
1858 if (orom
->major_ver
|| orom
->minor_ver
|| orom
->hotfix_ver
|| orom
->build
)
1859 printf("IMSM_VERSION=%d.%d.%d.%d\n", orom
->major_ver
, orom
->minor_ver
,
1860 orom
->hotfix_ver
, orom
->build
);
1861 printf("IMSM_SUPPORTED_RAID_LEVELS=%s%s%s%s%s\n",
1862 imsm_orom_has_raid0(orom
) ? "raid0 " : "",
1863 imsm_orom_has_raid1(orom
) ? "raid1 " : "",
1864 imsm_orom_has_raid1e(orom
) ? "raid1e " : "",
1865 imsm_orom_has_raid5(orom
) ? "raid10 " : "",
1866 imsm_orom_has_raid10(orom
) ? "raid5 " : "");
1867 printf("IMSM_SUPPORTED_CHUNK_SIZES=%s%s%s%s%s%s%s%s%s%s%s%s%s%s%s%s\n",
1868 imsm_orom_has_chunk(orom
, 2) ? "2k " : "",
1869 imsm_orom_has_chunk(orom
, 4) ? "4k " : "",
1870 imsm_orom_has_chunk(orom
, 8) ? "8k " : "",
1871 imsm_orom_has_chunk(orom
, 16) ? "16k " : "",
1872 imsm_orom_has_chunk(orom
, 32) ? "32k " : "",
1873 imsm_orom_has_chunk(orom
, 64) ? "64k " : "",
1874 imsm_orom_has_chunk(orom
, 128) ? "128k " : "",
1875 imsm_orom_has_chunk(orom
, 256) ? "256k " : "",
1876 imsm_orom_has_chunk(orom
, 512) ? "512k " : "",
1877 imsm_orom_has_chunk(orom
, 1024*1) ? "1M " : "",
1878 imsm_orom_has_chunk(orom
, 1024*2) ? "2M " : "",
1879 imsm_orom_has_chunk(orom
, 1024*4) ? "4M " : "",
1880 imsm_orom_has_chunk(orom
, 1024*8) ? "8M " : "",
1881 imsm_orom_has_chunk(orom
, 1024*16) ? "16M " : "",
1882 imsm_orom_has_chunk(orom
, 1024*32) ? "32M " : "",
1883 imsm_orom_has_chunk(orom
, 1024*64) ? "64M " : "");
1884 printf("IMSM_2TB_VOLUMES=%s\n",(orom
->attr
& IMSM_OROM_ATTR_2TB
) ? "yes" : "no");
1885 printf("IMSM_2TB_DISKS=%s\n",(orom
->attr
& IMSM_OROM_ATTR_2TB_DISK
) ? "yes" : "no");
1886 printf("IMSM_MAX_DISKS=%d\n",orom
->tds
);
1887 printf("IMSM_MAX_VOLUMES_PER_ARRAY=%d\n",orom
->vpa
);
1888 printf("IMSM_MAX_VOLUMES_PER_CONTROLLER=%d\n",orom
->vphba
);
1891 static int detail_platform_imsm(int verbose
, int enumerate_only
, char *controller_path
)
1893 /* There are two components to imsm platform support, the ahci SATA
1894 * controller and the option-rom. To find the SATA controller we
1895 * simply look in /sys/bus/pci/drivers/ahci to see if an ahci
1896 * controller with the Intel vendor id is present. This approach
1897 * allows mdadm to leverage the kernel's ahci detection logic, with the
1898 * caveat that if ahci.ko is not loaded mdadm will not be able to
1899 * detect platform raid capabilities. The option-rom resides in a
1900 * platform "Adapter ROM". We scan for its signature to retrieve the
1901 * platform capabilities. If raid support is disabled in the BIOS the
1902 * option-rom capability structure will not be available.
1904 struct sys_dev
*list
, *hba
;
1909 if (enumerate_only
) {
1910 if (check_env("IMSM_NO_PLATFORM"))
1912 list
= find_intel_devices();
1915 for (hba
= list
; hba
; hba
= hba
->next
) {
1916 if (find_imsm_capability(hba
)) {
1926 list
= find_intel_devices();
1929 pr_err("no active Intel(R) RAID controller found.\n");
1931 } else if (verbose
> 0)
1932 print_found_intel_controllers(list
);
1934 for (hba
= list
; hba
; hba
= hba
->next
) {
1935 if (controller_path
&& (compare_paths(hba
->path
, controller_path
) != 0))
1937 if (!find_imsm_capability(hba
)) {
1938 pr_err("imsm capabilities not found for controller: %s (type %s)\n",
1939 hba
->path
, get_sys_dev_type(hba
->type
));
1945 if (controller_path
&& result
== 1) {
1946 pr_err("no active Intel(R) RAID controller found under %s\n",
1951 const struct orom_entry
*oroms
= get_oroms();
1954 for (i
= 0; i
< SYS_DEV_MAX
&& oroms
[i
].devid_list
; i
++) {
1955 print_imsm_capability(&oroms
[i
].orom
);
1957 if (imsm_orom_is_nvme(&oroms
[i
].orom
)) {
1958 for (hba
= list
; hba
; hba
= hba
->next
) {
1959 if (hba
->type
== SYS_DEV_NVME
)
1960 printf(" NVMe Device : %s\n", hba
->path
);
1965 struct devid_list
*devid
;
1966 for (devid
= oroms
[i
].devid_list
; devid
; devid
= devid
->next
) {
1967 hba
= device_by_id(devid
->devid
);
1971 printf(" I/O Controller : %s (%s)\n",
1972 hba
->path
, get_sys_dev_type(hba
->type
));
1973 if (hba
->type
== SYS_DEV_SATA
) {
1974 host_base
= ahci_get_port_count(hba
->path
, &port_count
);
1975 if (ahci_enumerate_ports(hba
->path
, port_count
, host_base
, verbose
)) {
1977 pr_err("failed to enumerate ports on SATA controller at %s.\n", hba
->pci_id
);
1988 static int export_detail_platform_imsm(int verbose
, char *controller_path
)
1990 struct sys_dev
*list
, *hba
;
1993 list
= find_intel_devices();
1996 pr_err("IMSM_DETAIL_PLATFORM_ERROR=NO_INTEL_DEVICES\n");
2001 for (hba
= list
; hba
; hba
= hba
->next
) {
2002 if (controller_path
&& (compare_paths(hba
->path
,controller_path
) != 0))
2004 if (!find_imsm_capability(hba
) && verbose
> 0)
2005 pr_err("IMSM_DETAIL_PLATFORM_ERROR=NO_IMSM_CAPABLE_DEVICE_UNDER_%s\n", hba
->path
);
2010 const struct orom_entry
*oroms
= get_oroms();
2013 for (i
= 0; i
< SYS_DEV_MAX
&& oroms
[i
].devid_list
; i
++)
2014 print_imsm_capability_export(&oroms
[i
].orom
);
2021 static int match_home_imsm(struct supertype
*st
, char *homehost
)
2023 /* the imsm metadata format does not specify any host
2024 * identification information. We return -1 since we can never
2025 * confirm nor deny whether a given array is "meant" for this
2026 * host. We rely on compare_super and the 'family_num' fields to
2027 * exclude member disks that do not belong, and we rely on
2028 * mdadm.conf to specify the arrays that should be assembled.
2029 * Auto-assembly may still pick up "foreign" arrays.
2035 static void uuid_from_super_imsm(struct supertype
*st
, int uuid
[4])
2037 /* The uuid returned here is used for:
2038 * uuid to put into bitmap file (Create, Grow)
2039 * uuid for backup header when saving critical section (Grow)
2040 * comparing uuids when re-adding a device into an array
2041 * In these cases the uuid required is that of the data-array,
2042 * not the device-set.
2043 * uuid to recognise same set when adding a missing device back
2044 * to an array. This is a uuid for the device-set.
2046 * For each of these we can make do with a truncated
2047 * or hashed uuid rather than the original, as long as
2049 * In each case the uuid required is that of the data-array,
2050 * not the device-set.
2052 /* imsm does not track uuid's so we synthesis one using sha1 on
2053 * - The signature (Which is constant for all imsm array, but no matter)
2054 * - the orig_family_num of the container
2055 * - the index number of the volume
2056 * - the 'serial' number of the volume.
2057 * Hopefully these are all constant.
2059 struct intel_super
*super
= st
->sb
;
2062 struct sha1_ctx ctx
;
2063 struct imsm_dev
*dev
= NULL
;
2066 /* some mdadm versions failed to set ->orig_family_num, in which
2067 * case fall back to ->family_num. orig_family_num will be
2068 * fixed up with the first metadata update.
2070 family_num
= super
->anchor
->orig_family_num
;
2071 if (family_num
== 0)
2072 family_num
= super
->anchor
->family_num
;
2073 sha1_init_ctx(&ctx
);
2074 sha1_process_bytes(super
->anchor
->sig
, MPB_SIG_LEN
, &ctx
);
2075 sha1_process_bytes(&family_num
, sizeof(__u32
), &ctx
);
2076 if (super
->current_vol
>= 0)
2077 dev
= get_imsm_dev(super
, super
->current_vol
);
2079 __u32 vol
= super
->current_vol
;
2080 sha1_process_bytes(&vol
, sizeof(vol
), &ctx
);
2081 sha1_process_bytes(dev
->volume
, MAX_RAID_SERIAL_LEN
, &ctx
);
2083 sha1_finish_ctx(&ctx
, buf
);
2084 memcpy(uuid
, buf
, 4*4);
2089 get_imsm_numerical_version(struct imsm_super
*mpb
, int *m
, int *p
)
2091 __u8
*v
= get_imsm_version(mpb
);
2092 __u8
*end
= mpb
->sig
+ MAX_SIGNATURE_LENGTH
;
2093 char major
[] = { 0, 0, 0 };
2094 char minor
[] = { 0 ,0, 0 };
2095 char patch
[] = { 0, 0, 0 };
2096 char *ver_parse
[] = { major
, minor
, patch
};
2100 while (*v
!= '\0' && v
< end
) {
2101 if (*v
!= '.' && j
< 2)
2102 ver_parse
[i
][j
++] = *v
;
2110 *m
= strtol(minor
, NULL
, 0);
2111 *p
= strtol(patch
, NULL
, 0);
2115 static __u32
migr_strip_blocks_resync(struct imsm_dev
*dev
)
2117 /* migr_strip_size when repairing or initializing parity */
2118 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
2119 __u32 chunk
= __le32_to_cpu(map
->blocks_per_strip
);
2121 switch (get_imsm_raid_level(map
)) {
2126 return 128*1024 >> 9;
2130 static __u32
migr_strip_blocks_rebuild(struct imsm_dev
*dev
)
2132 /* migr_strip_size when rebuilding a degraded disk, no idea why
2133 * this is different than migr_strip_size_resync(), but it's good
2136 struct imsm_map
*map
= get_imsm_map(dev
, MAP_1
);
2137 __u32 chunk
= __le32_to_cpu(map
->blocks_per_strip
);
2139 switch (get_imsm_raid_level(map
)) {
2142 if (map
->num_members
% map
->num_domains
== 0)
2143 return 128*1024 >> 9;
2147 return max((__u32
) 64*1024 >> 9, chunk
);
2149 return 128*1024 >> 9;
2153 static __u32
num_stripes_per_unit_resync(struct imsm_dev
*dev
)
2155 struct imsm_map
*lo
= get_imsm_map(dev
, MAP_0
);
2156 struct imsm_map
*hi
= get_imsm_map(dev
, MAP_1
);
2157 __u32 lo_chunk
= __le32_to_cpu(lo
->blocks_per_strip
);
2158 __u32 hi_chunk
= __le32_to_cpu(hi
->blocks_per_strip
);
2160 return max((__u32
) 1, hi_chunk
/ lo_chunk
);
2163 static __u32
num_stripes_per_unit_rebuild(struct imsm_dev
*dev
)
2165 struct imsm_map
*lo
= get_imsm_map(dev
, MAP_0
);
2166 int level
= get_imsm_raid_level(lo
);
2168 if (level
== 1 || level
== 10) {
2169 struct imsm_map
*hi
= get_imsm_map(dev
, MAP_1
);
2171 return hi
->num_domains
;
2173 return num_stripes_per_unit_resync(dev
);
2176 static __u8
imsm_num_data_members(struct imsm_dev
*dev
, int second_map
)
2178 /* named 'imsm_' because raid0, raid1 and raid10
2179 * counter-intuitively have the same number of data disks
2181 struct imsm_map
*map
= get_imsm_map(dev
, second_map
);
2183 switch (get_imsm_raid_level(map
)) {
2185 return map
->num_members
;
2189 return map
->num_members
/2;
2191 return map
->num_members
- 1;
2193 dprintf("unsupported raid level\n");
2198 static __u32
parity_segment_depth(struct imsm_dev
*dev
)
2200 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
2201 __u32 chunk
= __le32_to_cpu(map
->blocks_per_strip
);
2203 switch(get_imsm_raid_level(map
)) {
2206 return chunk
* map
->num_domains
;
2208 return chunk
* map
->num_members
;
2214 static __u32
map_migr_block(struct imsm_dev
*dev
, __u32 block
)
2216 struct imsm_map
*map
= get_imsm_map(dev
, MAP_1
);
2217 __u32 chunk
= __le32_to_cpu(map
->blocks_per_strip
);
2218 __u32 strip
= block
/ chunk
;
2220 switch (get_imsm_raid_level(map
)) {
2223 __u32 vol_strip
= (strip
* map
->num_domains
) + 1;
2224 __u32 vol_stripe
= vol_strip
/ map
->num_members
;
2226 return vol_stripe
* chunk
+ block
% chunk
;
2228 __u32 stripe
= strip
/ (map
->num_members
- 1);
2230 return stripe
* chunk
+ block
% chunk
;
2237 static __u64
blocks_per_migr_unit(struct intel_super
*super
,
2238 struct imsm_dev
*dev
)
2240 /* calculate the conversion factor between per member 'blocks'
2241 * (md/{resync,rebuild}_start) and imsm migration units, return
2242 * 0 for the 'not migrating' and 'unsupported migration' cases
2244 if (!dev
->vol
.migr_state
)
2247 switch (migr_type(dev
)) {
2248 case MIGR_GEN_MIGR
: {
2249 struct migr_record
*migr_rec
= super
->migr_rec
;
2250 return __le32_to_cpu(migr_rec
->blocks_per_unit
);
2255 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
2256 __u32 stripes_per_unit
;
2257 __u32 blocks_per_unit
;
2266 /* yes, this is really the translation of migr_units to
2267 * per-member blocks in the 'resync' case
2269 stripes_per_unit
= num_stripes_per_unit_resync(dev
);
2270 migr_chunk
= migr_strip_blocks_resync(dev
);
2271 disks
= imsm_num_data_members(dev
, MAP_0
);
2272 blocks_per_unit
= stripes_per_unit
* migr_chunk
* disks
;
2273 stripe
= __le16_to_cpu(map
->blocks_per_strip
) * disks
;
2274 segment
= blocks_per_unit
/ stripe
;
2275 block_rel
= blocks_per_unit
- segment
* stripe
;
2276 parity_depth
= parity_segment_depth(dev
);
2277 block_map
= map_migr_block(dev
, block_rel
);
2278 return block_map
+ parity_depth
* segment
;
2280 case MIGR_REBUILD
: {
2281 __u32 stripes_per_unit
;
2284 stripes_per_unit
= num_stripes_per_unit_rebuild(dev
);
2285 migr_chunk
= migr_strip_blocks_rebuild(dev
);
2286 return migr_chunk
* stripes_per_unit
;
2288 case MIGR_STATE_CHANGE
:
2294 static int imsm_level_to_layout(int level
)
2302 return ALGORITHM_LEFT_ASYMMETRIC
;
2309 /*******************************************************************************
2310 * Function: read_imsm_migr_rec
2311 * Description: Function reads imsm migration record from last sector of disk
2313 * fd : disk descriptor
2314 * super : metadata info
2318 ******************************************************************************/
2319 static int read_imsm_migr_rec(int fd
, struct intel_super
*super
)
2322 unsigned long long dsize
;
2324 get_dev_size(fd
, NULL
, &dsize
);
2325 if (lseek64(fd
, dsize
- MIGR_REC_POSITION
, SEEK_SET
) < 0) {
2326 pr_err("Cannot seek to anchor block: %s\n",
2330 if (read(fd
, super
->migr_rec_buf
, MIGR_REC_BUF_SIZE
) !=
2331 MIGR_REC_BUF_SIZE
) {
2332 pr_err("Cannot read migr record block: %s\n",
2342 static struct imsm_dev
*imsm_get_device_during_migration(
2343 struct intel_super
*super
)
2346 struct intel_dev
*dv
;
2348 for (dv
= super
->devlist
; dv
; dv
= dv
->next
) {
2349 if (is_gen_migration(dv
->dev
))
2355 /*******************************************************************************
2356 * Function: load_imsm_migr_rec
2357 * Description: Function reads imsm migration record (it is stored at the last
2360 * super : imsm internal array info
2361 * info : general array info
2365 * -2 : no migration in progress
2366 ******************************************************************************/
2367 static int load_imsm_migr_rec(struct intel_super
*super
, struct mdinfo
*info
)
2370 struct dl
*dl
= NULL
;
2374 struct imsm_dev
*dev
;
2375 struct imsm_map
*map
= NULL
;
2378 /* find map under migration */
2379 dev
= imsm_get_device_during_migration(super
);
2380 /* nothing to load,no migration in progress?
2384 map
= get_imsm_map(dev
, MAP_0
);
2387 for (sd
= info
->devs
; sd
; sd
= sd
->next
) {
2388 /* skip spare and failed disks
2390 if (sd
->disk
.raid_disk
< 0)
2392 /* read only from one of the first two slots */
2394 slot
= get_imsm_disk_slot(map
,
2395 sd
->disk
.raid_disk
);
2396 if ((map
== NULL
) || (slot
> 1) || (slot
< 0))
2399 sprintf(nm
, "%d:%d", sd
->disk
.major
, sd
->disk
.minor
);
2400 fd
= dev_open(nm
, O_RDONLY
);
2406 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
2407 /* skip spare and failed disks
2411 /* read only from one of the first two slots */
2413 slot
= get_imsm_disk_slot(map
, dl
->index
);
2414 if ((map
== NULL
) || (slot
> 1) || (slot
< 0))
2416 sprintf(nm
, "%d:%d", dl
->major
, dl
->minor
);
2417 fd
= dev_open(nm
, O_RDONLY
);
2424 retval
= read_imsm_migr_rec(fd
, super
);
2433 /*******************************************************************************
2434 * function: imsm_create_metadata_checkpoint_update
2435 * Description: It creates update for checkpoint change.
2437 * super : imsm internal array info
2438 * u : pointer to prepared update
2441 * If length is equal to 0, input pointer u contains no update
2442 ******************************************************************************/
2443 static int imsm_create_metadata_checkpoint_update(
2444 struct intel_super
*super
,
2445 struct imsm_update_general_migration_checkpoint
**u
)
2448 int update_memory_size
= 0;
2450 dprintf("(enter)\n");
2456 /* size of all update data without anchor */
2457 update_memory_size
=
2458 sizeof(struct imsm_update_general_migration_checkpoint
);
2460 *u
= xcalloc(1, update_memory_size
);
2462 dprintf("error: cannot get memory\n");
2465 (*u
)->type
= update_general_migration_checkpoint
;
2466 (*u
)->curr_migr_unit
= __le32_to_cpu(super
->migr_rec
->curr_migr_unit
);
2467 dprintf("prepared for %u\n", (*u
)->curr_migr_unit
);
2469 return update_memory_size
;
2472 static void imsm_update_metadata_locally(struct supertype
*st
,
2473 void *buf
, int len
);
2475 /*******************************************************************************
2476 * Function: write_imsm_migr_rec
2477 * Description: Function writes imsm migration record
2478 * (at the last sector of disk)
2480 * super : imsm internal array info
2484 ******************************************************************************/
2485 static int write_imsm_migr_rec(struct supertype
*st
)
2487 struct intel_super
*super
= st
->sb
;
2488 unsigned long long dsize
;
2494 struct imsm_update_general_migration_checkpoint
*u
;
2495 struct imsm_dev
*dev
;
2496 struct imsm_map
*map
= NULL
;
2498 /* find map under migration */
2499 dev
= imsm_get_device_during_migration(super
);
2500 /* if no migration, write buffer anyway to clear migr_record
2501 * on disk based on first available device
2504 dev
= get_imsm_dev(super
, super
->current_vol
< 0 ? 0 :
2505 super
->current_vol
);
2507 map
= get_imsm_map(dev
, MAP_0
);
2509 for (sd
= super
->disks
; sd
; sd
= sd
->next
) {
2512 /* skip failed and spare devices */
2515 /* write to 2 first slots only */
2517 slot
= get_imsm_disk_slot(map
, sd
->index
);
2518 if ((map
== NULL
) || (slot
> 1) || (slot
< 0))
2521 sprintf(nm
, "%d:%d", sd
->major
, sd
->minor
);
2522 fd
= dev_open(nm
, O_RDWR
);
2525 get_dev_size(fd
, NULL
, &dsize
);
2526 if (lseek64(fd
, dsize
- MIGR_REC_POSITION
, SEEK_SET
) < 0) {
2527 pr_err("Cannot seek to anchor block: %s\n",
2531 if (write(fd
, super
->migr_rec_buf
, MIGR_REC_BUF_SIZE
) !=
2532 MIGR_REC_BUF_SIZE
) {
2533 pr_err("Cannot write migr record block: %s\n",
2540 /* update checkpoint information in metadata */
2541 len
= imsm_create_metadata_checkpoint_update(super
, &u
);
2544 dprintf("imsm: Cannot prepare update\n");
2547 /* update metadata locally */
2548 imsm_update_metadata_locally(st
, u
, len
);
2549 /* and possibly remotely */
2550 if (st
->update_tail
) {
2551 append_metadata_update(st
, u
, len
);
2552 /* during reshape we do all work inside metadata handler
2553 * manage_reshape(), so metadata update has to be triggered
2556 flush_metadata_updates(st
);
2557 st
->update_tail
= &st
->updates
;
2567 #endif /* MDASSEMBLE */
2569 /* spare/missing disks activations are not allowe when
2570 * array/container performs reshape operation, because
2571 * all arrays in container works on the same disks set
2573 int imsm_reshape_blocks_arrays_changes(struct intel_super
*super
)
2576 struct intel_dev
*i_dev
;
2577 struct imsm_dev
*dev
;
2579 /* check whole container
2581 for (i_dev
= super
->devlist
; i_dev
; i_dev
= i_dev
->next
) {
2583 if (is_gen_migration(dev
)) {
2584 /* No repair during any migration in container
2592 static unsigned long long imsm_component_size_aligment_check(int level
,
2594 unsigned long long component_size
)
2596 unsigned int component_size_alligment
;
2598 /* check component size aligment
2600 component_size_alligment
= component_size
% (chunk_size
/512);
2602 dprintf("(Level: %i, chunk_size = %i, component_size = %llu), component_size_alligment = %u\n",
2603 level
, chunk_size
, component_size
,
2604 component_size_alligment
);
2606 if (component_size_alligment
&& (level
!= 1) && (level
!= UnSet
)) {
2607 dprintf("imsm: reported component size alligned from %llu ",
2609 component_size
-= component_size_alligment
;
2610 dprintf_cont("to %llu (%i).\n",
2611 component_size
, component_size_alligment
);
2614 return component_size
;
2617 static void getinfo_super_imsm_volume(struct supertype
*st
, struct mdinfo
*info
, char *dmap
)
2619 struct intel_super
*super
= st
->sb
;
2620 struct migr_record
*migr_rec
= super
->migr_rec
;
2621 struct imsm_dev
*dev
= get_imsm_dev(super
, super
->current_vol
);
2622 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
2623 struct imsm_map
*prev_map
= get_imsm_map(dev
, MAP_1
);
2624 struct imsm_map
*map_to_analyse
= map
;
2626 int map_disks
= info
->array
.raid_disks
;
2628 memset(info
, 0, sizeof(*info
));
2630 map_to_analyse
= prev_map
;
2632 dl
= super
->current_disk
;
2634 info
->container_member
= super
->current_vol
;
2635 info
->array
.raid_disks
= map
->num_members
;
2636 info
->array
.level
= get_imsm_raid_level(map_to_analyse
);
2637 info
->array
.layout
= imsm_level_to_layout(info
->array
.level
);
2638 info
->array
.md_minor
= -1;
2639 info
->array
.ctime
= 0;
2640 info
->array
.utime
= 0;
2641 info
->array
.chunk_size
=
2642 __le16_to_cpu(map_to_analyse
->blocks_per_strip
) << 9;
2643 info
->array
.state
= !dev
->vol
.dirty
;
2644 info
->custom_array_size
= __le32_to_cpu(dev
->size_high
);
2645 info
->custom_array_size
<<= 32;
2646 info
->custom_array_size
|= __le32_to_cpu(dev
->size_low
);
2647 info
->recovery_blocked
= imsm_reshape_blocks_arrays_changes(st
->sb
);
2649 if (is_gen_migration(dev
)) {
2650 info
->reshape_active
= 1;
2651 info
->new_level
= get_imsm_raid_level(map
);
2652 info
->new_layout
= imsm_level_to_layout(info
->new_level
);
2653 info
->new_chunk
= __le16_to_cpu(map
->blocks_per_strip
) << 9;
2654 info
->delta_disks
= map
->num_members
- prev_map
->num_members
;
2655 if (info
->delta_disks
) {
2656 /* this needs to be applied to every array
2659 info
->reshape_active
= CONTAINER_RESHAPE
;
2661 /* We shape information that we give to md might have to be
2662 * modify to cope with md's requirement for reshaping arrays.
2663 * For example, when reshaping a RAID0, md requires it to be
2664 * presented as a degraded RAID4.
2665 * Also if a RAID0 is migrating to a RAID5 we need to specify
2666 * the array as already being RAID5, but the 'before' layout
2667 * is a RAID4-like layout.
2669 switch (info
->array
.level
) {
2671 switch(info
->new_level
) {
2673 /* conversion is happening as RAID4 */
2674 info
->array
.level
= 4;
2675 info
->array
.raid_disks
+= 1;
2678 /* conversion is happening as RAID5 */
2679 info
->array
.level
= 5;
2680 info
->array
.layout
= ALGORITHM_PARITY_N
;
2681 info
->delta_disks
-= 1;
2684 /* FIXME error message */
2685 info
->array
.level
= UnSet
;
2691 info
->new_level
= UnSet
;
2692 info
->new_layout
= UnSet
;
2693 info
->new_chunk
= info
->array
.chunk_size
;
2694 info
->delta_disks
= 0;
2698 info
->disk
.major
= dl
->major
;
2699 info
->disk
.minor
= dl
->minor
;
2700 info
->disk
.number
= dl
->index
;
2701 info
->disk
.raid_disk
= get_imsm_disk_slot(map_to_analyse
,
2705 info
->data_offset
= pba_of_lba0(map_to_analyse
);
2706 info
->component_size
= blocks_per_member(map_to_analyse
);
2708 info
->component_size
= imsm_component_size_aligment_check(
2710 info
->array
.chunk_size
,
2711 info
->component_size
);
2713 memset(info
->uuid
, 0, sizeof(info
->uuid
));
2714 info
->recovery_start
= MaxSector
;
2716 info
->reshape_progress
= 0;
2717 info
->resync_start
= MaxSector
;
2718 if ((map_to_analyse
->map_state
== IMSM_T_STATE_UNINITIALIZED
||
2720 imsm_reshape_blocks_arrays_changes(super
) == 0) {
2721 info
->resync_start
= 0;
2723 if (dev
->vol
.migr_state
) {
2724 switch (migr_type(dev
)) {
2727 __u64 blocks_per_unit
= blocks_per_migr_unit(super
,
2729 __u64 units
= __le32_to_cpu(dev
->vol
.curr_migr_unit
);
2731 info
->resync_start
= blocks_per_unit
* units
;
2734 case MIGR_GEN_MIGR
: {
2735 __u64 blocks_per_unit
= blocks_per_migr_unit(super
,
2737 __u64 units
= __le32_to_cpu(migr_rec
->curr_migr_unit
);
2738 unsigned long long array_blocks
;
2741 if (__le32_to_cpu(migr_rec
->ascending_migr
) &&
2743 (__le32_to_cpu(migr_rec
->num_migr_units
)-1)) &&
2744 (super
->migr_rec
->rec_status
==
2745 __cpu_to_le32(UNIT_SRC_IN_CP_AREA
)))
2748 info
->reshape_progress
= blocks_per_unit
* units
;
2750 dprintf("IMSM: General Migration checkpoint : %llu (%llu) -> read reshape progress : %llu\n",
2751 (unsigned long long)units
,
2752 (unsigned long long)blocks_per_unit
,
2753 info
->reshape_progress
);
2755 used_disks
= imsm_num_data_members(dev
, MAP_1
);
2756 if (used_disks
> 0) {
2757 array_blocks
= blocks_per_member(map
) *
2759 /* round array size down to closest MB
2761 info
->custom_array_size
= (array_blocks
2762 >> SECT_PER_MB_SHIFT
)
2763 << SECT_PER_MB_SHIFT
;
2767 /* we could emulate the checkpointing of
2768 * 'sync_action=check' migrations, but for now
2769 * we just immediately complete them
2772 /* this is handled by container_content_imsm() */
2773 case MIGR_STATE_CHANGE
:
2774 /* FIXME handle other migrations */
2776 /* we are not dirty, so... */
2777 info
->resync_start
= MaxSector
;
2781 strncpy(info
->name
, (char *) dev
->volume
, MAX_RAID_SERIAL_LEN
);
2782 info
->name
[MAX_RAID_SERIAL_LEN
] = 0;
2784 info
->array
.major_version
= -1;
2785 info
->array
.minor_version
= -2;
2786 sprintf(info
->text_version
, "/%s/%d", st
->container_devnm
, info
->container_member
);
2787 info
->safe_mode_delay
= 4000; /* 4 secs like the Matrix driver */
2788 uuid_from_super_imsm(st
, info
->uuid
);
2792 for (i
=0; i
<map_disks
; i
++) {
2794 if (i
< info
->array
.raid_disks
) {
2795 struct imsm_disk
*dsk
;
2796 j
= get_imsm_disk_idx(dev
, i
, MAP_X
);
2797 dsk
= get_imsm_disk(super
, j
);
2798 if (dsk
&& (dsk
->status
& CONFIGURED_DISK
))
2805 static __u8
imsm_check_degraded(struct intel_super
*super
, struct imsm_dev
*dev
,
2806 int failed
, int look_in_map
);
2808 static int imsm_count_failed(struct intel_super
*super
, struct imsm_dev
*dev
,
2812 static void manage_second_map(struct intel_super
*super
, struct imsm_dev
*dev
)
2814 if (is_gen_migration(dev
)) {
2817 struct imsm_map
*map2
= get_imsm_map(dev
, MAP_1
);
2819 failed
= imsm_count_failed(super
, dev
, MAP_1
);
2820 map_state
= imsm_check_degraded(super
, dev
, failed
, MAP_1
);
2821 if (map2
->map_state
!= map_state
) {
2822 map2
->map_state
= map_state
;
2823 super
->updates_pending
++;
2829 static struct imsm_disk
*get_imsm_missing(struct intel_super
*super
, __u8 index
)
2833 for (d
= super
->missing
; d
; d
= d
->next
)
2834 if (d
->index
== index
)
2839 static void getinfo_super_imsm(struct supertype
*st
, struct mdinfo
*info
, char *map
)
2841 struct intel_super
*super
= st
->sb
;
2842 struct imsm_disk
*disk
;
2843 int map_disks
= info
->array
.raid_disks
;
2844 int max_enough
= -1;
2846 struct imsm_super
*mpb
;
2848 if (super
->current_vol
>= 0) {
2849 getinfo_super_imsm_volume(st
, info
, map
);
2852 memset(info
, 0, sizeof(*info
));
2854 /* Set raid_disks to zero so that Assemble will always pull in valid
2857 info
->array
.raid_disks
= 0;
2858 info
->array
.level
= LEVEL_CONTAINER
;
2859 info
->array
.layout
= 0;
2860 info
->array
.md_minor
= -1;
2861 info
->array
.ctime
= 0; /* N/A for imsm */
2862 info
->array
.utime
= 0;
2863 info
->array
.chunk_size
= 0;
2865 info
->disk
.major
= 0;
2866 info
->disk
.minor
= 0;
2867 info
->disk
.raid_disk
= -1;
2868 info
->reshape_active
= 0;
2869 info
->array
.major_version
= -1;
2870 info
->array
.minor_version
= -2;
2871 strcpy(info
->text_version
, "imsm");
2872 info
->safe_mode_delay
= 0;
2873 info
->disk
.number
= -1;
2874 info
->disk
.state
= 0;
2876 info
->recovery_start
= MaxSector
;
2877 info
->recovery_blocked
= imsm_reshape_blocks_arrays_changes(st
->sb
);
2879 /* do we have the all the insync disks that we expect? */
2880 mpb
= super
->anchor
;
2882 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
2883 struct imsm_dev
*dev
= get_imsm_dev(super
, i
);
2884 int failed
, enough
, j
, missing
= 0;
2885 struct imsm_map
*map
;
2888 failed
= imsm_count_failed(super
, dev
, MAP_0
);
2889 state
= imsm_check_degraded(super
, dev
, failed
, MAP_0
);
2890 map
= get_imsm_map(dev
, MAP_0
);
2892 /* any newly missing disks?
2893 * (catches single-degraded vs double-degraded)
2895 for (j
= 0; j
< map
->num_members
; j
++) {
2896 __u32 ord
= get_imsm_ord_tbl_ent(dev
, j
, MAP_0
);
2897 __u32 idx
= ord_to_idx(ord
);
2899 if (!(ord
& IMSM_ORD_REBUILD
) &&
2900 get_imsm_missing(super
, idx
)) {
2906 if (state
== IMSM_T_STATE_FAILED
)
2908 else if (state
== IMSM_T_STATE_DEGRADED
&&
2909 (state
!= map
->map_state
|| missing
))
2911 else /* we're normal, or already degraded */
2913 if (is_gen_migration(dev
) && missing
) {
2914 /* during general migration we need all disks
2915 * that process is running on.
2916 * No new missing disk is allowed.
2920 /* no more checks necessary
2924 /* in the missing/failed disk case check to see
2925 * if at least one array is runnable
2927 max_enough
= max(max_enough
, enough
);
2929 dprintf("enough: %d\n", max_enough
);
2930 info
->container_enough
= max_enough
;
2933 __u32 reserved
= imsm_reserved_sectors(super
, super
->disks
);
2935 disk
= &super
->disks
->disk
;
2936 info
->data_offset
= total_blocks(&super
->disks
->disk
) - reserved
;
2937 info
->component_size
= reserved
;
2938 info
->disk
.state
= is_configured(disk
) ? (1 << MD_DISK_ACTIVE
) : 0;
2939 /* we don't change info->disk.raid_disk here because
2940 * this state will be finalized in mdmon after we have
2941 * found the 'most fresh' version of the metadata
2943 info
->disk
.state
|= is_failed(disk
) ? (1 << MD_DISK_FAULTY
) : 0;
2944 info
->disk
.state
|= is_spare(disk
) ? 0 : (1 << MD_DISK_SYNC
);
2947 /* only call uuid_from_super_imsm when this disk is part of a populated container,
2948 * ->compare_super may have updated the 'num_raid_devs' field for spares
2950 if (info
->disk
.state
& (1 << MD_DISK_SYNC
) || super
->anchor
->num_raid_devs
)
2951 uuid_from_super_imsm(st
, info
->uuid
);
2953 memcpy(info
->uuid
, uuid_zero
, sizeof(uuid_zero
));
2955 /* I don't know how to compute 'map' on imsm, so use safe default */
2958 for (i
= 0; i
< map_disks
; i
++)
2964 /* allocates memory and fills disk in mdinfo structure
2965 * for each disk in array */
2966 struct mdinfo
*getinfo_super_disks_imsm(struct supertype
*st
)
2968 struct mdinfo
*mddev
= NULL
;
2969 struct intel_super
*super
= st
->sb
;
2970 struct imsm_disk
*disk
;
2973 if (!super
|| !super
->disks
)
2976 mddev
= xcalloc(1, sizeof(*mddev
));
2980 tmp
= xcalloc(1, sizeof(*tmp
));
2982 tmp
->next
= mddev
->devs
;
2984 tmp
->disk
.number
= count
++;
2985 tmp
->disk
.major
= dl
->major
;
2986 tmp
->disk
.minor
= dl
->minor
;
2987 tmp
->disk
.state
= is_configured(disk
) ?
2988 (1 << MD_DISK_ACTIVE
) : 0;
2989 tmp
->disk
.state
|= is_failed(disk
) ? (1 << MD_DISK_FAULTY
) : 0;
2990 tmp
->disk
.state
|= is_spare(disk
) ? 0 : (1 << MD_DISK_SYNC
);
2991 tmp
->disk
.raid_disk
= -1;
2997 static int update_super_imsm(struct supertype
*st
, struct mdinfo
*info
,
2998 char *update
, char *devname
, int verbose
,
2999 int uuid_set
, char *homehost
)
3001 /* For 'assemble' and 'force' we need to return non-zero if any
3002 * change was made. For others, the return value is ignored.
3003 * Update options are:
3004 * force-one : This device looks a bit old but needs to be included,
3005 * update age info appropriately.
3006 * assemble: clear any 'faulty' flag to allow this device to
3008 * force-array: Array is degraded but being forced, mark it clean
3009 * if that will be needed to assemble it.
3011 * newdev: not used ????
3012 * grow: Array has gained a new device - this is currently for
3014 * resync: mark as dirty so a resync will happen.
3015 * name: update the name - preserving the homehost
3016 * uuid: Change the uuid of the array to match watch is given
3018 * Following are not relevant for this imsm:
3019 * sparc2.2 : update from old dodgey metadata
3020 * super-minor: change the preferred_minor number
3021 * summaries: update redundant counters.
3022 * homehost: update the recorded homehost
3023 * _reshape_progress: record new reshape_progress position.
3026 struct intel_super
*super
= st
->sb
;
3027 struct imsm_super
*mpb
;
3029 /* we can only update container info */
3030 if (!super
|| super
->current_vol
>= 0 || !super
->anchor
)
3033 mpb
= super
->anchor
;
3035 if (strcmp(update
, "uuid") == 0) {
3036 /* We take this to mean that the family_num should be updated.
3037 * However that is much smaller than the uuid so we cannot really
3038 * allow an explicit uuid to be given. And it is hard to reliably
3040 * So if !uuid_set we know the current uuid is random and just used
3041 * the first 'int' and copy it to the other 3 positions.
3042 * Otherwise we require the 4 'int's to be the same as would be the
3043 * case if we are using a random uuid. So an explicit uuid will be
3044 * accepted as long as all for ints are the same... which shouldn't hurt
3047 info
->uuid
[1] = info
->uuid
[2] = info
->uuid
[3] = info
->uuid
[0];
3050 if (info
->uuid
[0] != info
->uuid
[1] ||
3051 info
->uuid
[1] != info
->uuid
[2] ||
3052 info
->uuid
[2] != info
->uuid
[3])
3058 mpb
->orig_family_num
= info
->uuid
[0];
3059 } else if (strcmp(update
, "assemble") == 0)
3064 /* successful update? recompute checksum */
3066 mpb
->check_sum
= __le32_to_cpu(__gen_imsm_checksum(mpb
));
3071 static size_t disks_to_mpb_size(int disks
)
3075 size
= sizeof(struct imsm_super
);
3076 size
+= (disks
- 1) * sizeof(struct imsm_disk
);
3077 size
+= 2 * sizeof(struct imsm_dev
);
3078 /* up to 2 maps per raid device (-2 for imsm_maps in imsm_dev */
3079 size
+= (4 - 2) * sizeof(struct imsm_map
);
3080 /* 4 possible disk_ord_tbl's */
3081 size
+= 4 * (disks
- 1) * sizeof(__u32
);
3086 static __u64
avail_size_imsm(struct supertype
*st
, __u64 devsize
,
3087 unsigned long long data_offset
)
3089 if (devsize
< (MPB_SECTOR_CNT
+ IMSM_RESERVED_SECTORS
))
3092 return devsize
- (MPB_SECTOR_CNT
+ IMSM_RESERVED_SECTORS
);
3095 static void free_devlist(struct intel_super
*super
)
3097 struct intel_dev
*dv
;
3099 while (super
->devlist
) {
3100 dv
= super
->devlist
->next
;
3101 free(super
->devlist
->dev
);
3102 free(super
->devlist
);
3103 super
->devlist
= dv
;
3107 static void imsm_copy_dev(struct imsm_dev
*dest
, struct imsm_dev
*src
)
3109 memcpy(dest
, src
, sizeof_imsm_dev(src
, 0));
3112 static int compare_super_imsm(struct supertype
*st
, struct supertype
*tst
)
3116 * 0 same, or first was empty, and second was copied
3117 * 1 second had wrong number
3119 * 3 wrong other info
3121 struct intel_super
*first
= st
->sb
;
3122 struct intel_super
*sec
= tst
->sb
;
3129 /* in platform dependent environment test if the disks
3130 * use the same Intel hba
3131 * If not on Intel hba at all, allow anything.
3133 if (!check_env("IMSM_NO_PLATFORM") && first
->hba
&& sec
->hba
) {
3134 if (first
->hba
->type
!= sec
->hba
->type
) {
3136 "HBAs of devices do not match %s != %s\n",
3137 get_sys_dev_type(first
->hba
->type
),
3138 get_sys_dev_type(sec
->hba
->type
));
3141 if (first
->orom
!= sec
->orom
) {
3143 "HBAs of devices do not match %s != %s\n",
3144 first
->hba
->pci_id
, sec
->hba
->pci_id
);
3149 /* if an anchor does not have num_raid_devs set then it is a free
3152 if (first
->anchor
->num_raid_devs
> 0 &&
3153 sec
->anchor
->num_raid_devs
> 0) {
3154 /* Determine if these disks might ever have been
3155 * related. Further disambiguation can only take place
3156 * in load_super_imsm_all
3158 __u32 first_family
= first
->anchor
->orig_family_num
;
3159 __u32 sec_family
= sec
->anchor
->orig_family_num
;
3161 if (memcmp(first
->anchor
->sig
, sec
->anchor
->sig
,
3162 MAX_SIGNATURE_LENGTH
) != 0)
3165 if (first_family
== 0)
3166 first_family
= first
->anchor
->family_num
;
3167 if (sec_family
== 0)
3168 sec_family
= sec
->anchor
->family_num
;
3170 if (first_family
!= sec_family
)
3175 /* if 'first' is a spare promote it to a populated mpb with sec's
3178 if (first
->anchor
->num_raid_devs
== 0 &&
3179 sec
->anchor
->num_raid_devs
> 0) {
3181 struct intel_dev
*dv
;
3182 struct imsm_dev
*dev
;
3184 /* we need to copy raid device info from sec if an allocation
3185 * fails here we don't associate the spare
3187 for (i
= 0; i
< sec
->anchor
->num_raid_devs
; i
++) {
3188 dv
= xmalloc(sizeof(*dv
));
3189 dev
= xmalloc(sizeof_imsm_dev(get_imsm_dev(sec
, i
), 1));
3192 dv
->next
= first
->devlist
;
3193 first
->devlist
= dv
;
3195 if (i
< sec
->anchor
->num_raid_devs
) {
3196 /* allocation failure */
3197 free_devlist(first
);
3198 pr_err("imsm: failed to associate spare\n");
3201 first
->anchor
->num_raid_devs
= sec
->anchor
->num_raid_devs
;
3202 first
->anchor
->orig_family_num
= sec
->anchor
->orig_family_num
;
3203 first
->anchor
->family_num
= sec
->anchor
->family_num
;
3204 memcpy(first
->anchor
->sig
, sec
->anchor
->sig
, MAX_SIGNATURE_LENGTH
);
3205 for (i
= 0; i
< sec
->anchor
->num_raid_devs
; i
++)
3206 imsm_copy_dev(get_imsm_dev(first
, i
), get_imsm_dev(sec
, i
));
3212 static void fd2devname(int fd
, char *name
)
3216 char dname
[PATH_MAX
];
3221 if (fstat(fd
, &st
) != 0)
3223 sprintf(path
, "/sys/dev/block/%d:%d",
3224 major(st
.st_rdev
), minor(st
.st_rdev
));
3226 rv
= readlink(path
, dname
, sizeof(dname
)-1);
3231 nm
= strrchr(dname
, '/');
3234 snprintf(name
, MAX_RAID_SERIAL_LEN
, "/dev/%s", nm
);
3238 extern int scsi_get_serial(int fd
, void *buf
, size_t buf_len
);
3240 static int imsm_read_serial(int fd
, char *devname
,
3241 __u8 serial
[MAX_RAID_SERIAL_LEN
])
3243 unsigned char scsi_serial
[255];
3252 memset(scsi_serial
, 0, sizeof(scsi_serial
));
3254 rv
= scsi_get_serial(fd
, scsi_serial
, sizeof(scsi_serial
));
3256 if (rv
&& check_env("IMSM_DEVNAME_AS_SERIAL")) {
3257 memset(serial
, 0, MAX_RAID_SERIAL_LEN
);
3258 fd2devname(fd
, (char *) serial
);
3264 pr_err("Failed to retrieve serial for %s\n",
3269 rsp_len
= scsi_serial
[3];
3272 pr_err("Failed to retrieve serial for %s\n",
3276 rsp_buf
= (char *) &scsi_serial
[4];
3278 /* trim all whitespace and non-printable characters and convert
3281 for (i
= 0, dest
= rsp_buf
; i
< rsp_len
; i
++) {
3284 /* ':' is reserved for use in placeholder serial
3285 * numbers for missing disks
3293 len
= dest
- rsp_buf
;
3296 /* truncate leading characters */
3297 if (len
> MAX_RAID_SERIAL_LEN
) {
3298 dest
+= len
- MAX_RAID_SERIAL_LEN
;
3299 len
= MAX_RAID_SERIAL_LEN
;
3302 memset(serial
, 0, MAX_RAID_SERIAL_LEN
);
3303 memcpy(serial
, dest
, len
);
3308 static int serialcmp(__u8
*s1
, __u8
*s2
)
3310 return strncmp((char *) s1
, (char *) s2
, MAX_RAID_SERIAL_LEN
);
3313 static void serialcpy(__u8
*dest
, __u8
*src
)
3315 strncpy((char *) dest
, (char *) src
, MAX_RAID_SERIAL_LEN
);
3318 static struct dl
*serial_to_dl(__u8
*serial
, struct intel_super
*super
)
3322 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
3323 if (serialcmp(dl
->serial
, serial
) == 0)
3329 static struct imsm_disk
*
3330 __serial_to_disk(__u8
*serial
, struct imsm_super
*mpb
, int *idx
)
3334 for (i
= 0; i
< mpb
->num_disks
; i
++) {
3335 struct imsm_disk
*disk
= __get_imsm_disk(mpb
, i
);
3337 if (serialcmp(disk
->serial
, serial
) == 0) {
3348 load_imsm_disk(int fd
, struct intel_super
*super
, char *devname
, int keep_fd
)
3350 struct imsm_disk
*disk
;
3355 __u8 serial
[MAX_RAID_SERIAL_LEN
];
3357 rv
= imsm_read_serial(fd
, devname
, serial
);
3362 dl
= xcalloc(1, sizeof(*dl
));
3365 dl
->major
= major(stb
.st_rdev
);
3366 dl
->minor
= minor(stb
.st_rdev
);
3367 dl
->next
= super
->disks
;
3368 dl
->fd
= keep_fd
? fd
: -1;
3369 assert(super
->disks
== NULL
);
3371 serialcpy(dl
->serial
, serial
);
3374 fd2devname(fd
, name
);
3376 dl
->devname
= xstrdup(devname
);
3378 dl
->devname
= xstrdup(name
);
3380 /* look up this disk's index in the current anchor */
3381 disk
= __serial_to_disk(dl
->serial
, super
->anchor
, &dl
->index
);
3384 /* only set index on disks that are a member of a
3385 * populated contianer, i.e. one with raid_devs
3387 if (is_failed(&dl
->disk
))
3389 else if (is_spare(&dl
->disk
))
3397 /* When migrating map0 contains the 'destination' state while map1
3398 * contains the current state. When not migrating map0 contains the
3399 * current state. This routine assumes that map[0].map_state is set to
3400 * the current array state before being called.
3402 * Migration is indicated by one of the following states
3403 * 1/ Idle (migr_state=0 map0state=normal||unitialized||degraded||failed)
3404 * 2/ Initialize (migr_state=1 migr_type=MIGR_INIT map0state=normal
3405 * map1state=unitialized)
3406 * 3/ Repair (Resync) (migr_state=1 migr_type=MIGR_REPAIR map0state=normal
3408 * 4/ Rebuild (migr_state=1 migr_type=MIGR_REBUILD map0state=normal
3409 * map1state=degraded)
3410 * 5/ Migration (mig_state=1 migr_type=MIGR_GEN_MIGR map0state=normal
3413 static void migrate(struct imsm_dev
*dev
, struct intel_super
*super
,
3414 __u8 to_state
, int migr_type
)
3416 struct imsm_map
*dest
;
3417 struct imsm_map
*src
= get_imsm_map(dev
, MAP_0
);
3419 dev
->vol
.migr_state
= 1;
3420 set_migr_type(dev
, migr_type
);
3421 dev
->vol
.curr_migr_unit
= 0;
3422 dest
= get_imsm_map(dev
, MAP_1
);
3424 /* duplicate and then set the target end state in map[0] */
3425 memcpy(dest
, src
, sizeof_imsm_map(src
));
3426 if ((migr_type
== MIGR_REBUILD
) ||
3427 (migr_type
== MIGR_GEN_MIGR
)) {
3431 for (i
= 0; i
< src
->num_members
; i
++) {
3432 ord
= __le32_to_cpu(src
->disk_ord_tbl
[i
]);
3433 set_imsm_ord_tbl_ent(src
, i
, ord_to_idx(ord
));
3437 if (migr_type
== MIGR_GEN_MIGR
)
3438 /* Clear migration record */
3439 memset(super
->migr_rec
, 0, sizeof(struct migr_record
));
3441 src
->map_state
= to_state
;
3444 static void end_migration(struct imsm_dev
*dev
, struct intel_super
*super
,
3447 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
3448 struct imsm_map
*prev
= get_imsm_map(dev
, dev
->vol
.migr_state
== 0 ?
3452 /* merge any IMSM_ORD_REBUILD bits that were not successfully
3453 * completed in the last migration.
3455 * FIXME add support for raid-level-migration
3457 if ((map_state
!= map
->map_state
) && (is_gen_migration(dev
) == 0) &&
3458 (prev
->map_state
!= IMSM_T_STATE_UNINITIALIZED
)) {
3459 /* when final map state is other than expected
3460 * merge maps (not for migration)
3464 for (i
= 0; i
< prev
->num_members
; i
++)
3465 for (j
= 0; j
< map
->num_members
; j
++)
3466 /* during online capacity expansion
3467 * disks position can be changed
3468 * if takeover is used
3470 if (ord_to_idx(map
->disk_ord_tbl
[j
]) ==
3471 ord_to_idx(prev
->disk_ord_tbl
[i
])) {
3472 map
->disk_ord_tbl
[j
] |=
3473 prev
->disk_ord_tbl
[i
];
3476 failed
= imsm_count_failed(super
, dev
, MAP_0
);
3477 map_state
= imsm_check_degraded(super
, dev
, failed
, MAP_0
);
3480 dev
->vol
.migr_state
= 0;
3481 set_migr_type(dev
, 0);
3482 dev
->vol
.curr_migr_unit
= 0;
3483 map
->map_state
= map_state
;
3487 static int parse_raid_devices(struct intel_super
*super
)
3490 struct imsm_dev
*dev_new
;
3491 size_t len
, len_migr
;
3493 size_t space_needed
= 0;
3494 struct imsm_super
*mpb
= super
->anchor
;
3496 for (i
= 0; i
< super
->anchor
->num_raid_devs
; i
++) {
3497 struct imsm_dev
*dev_iter
= __get_imsm_dev(super
->anchor
, i
);
3498 struct intel_dev
*dv
;
3500 len
= sizeof_imsm_dev(dev_iter
, 0);
3501 len_migr
= sizeof_imsm_dev(dev_iter
, 1);
3503 space_needed
+= len_migr
- len
;
3505 dv
= xmalloc(sizeof(*dv
));
3506 if (max_len
< len_migr
)
3508 if (max_len
> len_migr
)
3509 space_needed
+= max_len
- len_migr
;
3510 dev_new
= xmalloc(max_len
);
3511 imsm_copy_dev(dev_new
, dev_iter
);
3514 dv
->next
= super
->devlist
;
3515 super
->devlist
= dv
;
3518 /* ensure that super->buf is large enough when all raid devices
3521 if (__le32_to_cpu(mpb
->mpb_size
) + space_needed
> super
->len
) {
3524 len
= ROUND_UP(__le32_to_cpu(mpb
->mpb_size
) + space_needed
, 512);
3525 if (posix_memalign(&buf
, 512, len
) != 0)
3528 memcpy(buf
, super
->buf
, super
->len
);
3529 memset(buf
+ super
->len
, 0, len
- super
->len
);
3538 /* retrieve a pointer to the bbm log which starts after all raid devices */
3539 struct bbm_log
*__get_imsm_bbm_log(struct imsm_super
*mpb
)
3543 if (__le32_to_cpu(mpb
->bbm_log_size
)) {
3545 ptr
+= mpb
->mpb_size
- __le32_to_cpu(mpb
->bbm_log_size
);
3551 /*******************************************************************************
3552 * Function: check_mpb_migr_compatibility
3553 * Description: Function checks for unsupported migration features:
3554 * - migration optimization area (pba_of_lba0)
3555 * - descending reshape (ascending_migr)
3557 * super : imsm metadata information
3559 * 0 : migration is compatible
3560 * -1 : migration is not compatible
3561 ******************************************************************************/
3562 int check_mpb_migr_compatibility(struct intel_super
*super
)
3564 struct imsm_map
*map0
, *map1
;
3565 struct migr_record
*migr_rec
= super
->migr_rec
;
3568 for (i
= 0; i
< super
->anchor
->num_raid_devs
; i
++) {
3569 struct imsm_dev
*dev_iter
= __get_imsm_dev(super
->anchor
, i
);
3572 dev_iter
->vol
.migr_state
== 1 &&
3573 dev_iter
->vol
.migr_type
== MIGR_GEN_MIGR
) {
3574 /* This device is migrating */
3575 map0
= get_imsm_map(dev_iter
, MAP_0
);
3576 map1
= get_imsm_map(dev_iter
, MAP_1
);
3577 if (pba_of_lba0(map0
) != pba_of_lba0(map1
))
3578 /* migration optimization area was used */
3580 if (migr_rec
->ascending_migr
== 0
3581 && migr_rec
->dest_depth_per_unit
> 0)
3582 /* descending reshape not supported yet */
3589 static void __free_imsm(struct intel_super
*super
, int free_disks
);
3591 /* load_imsm_mpb - read matrix metadata
3592 * allocates super->mpb to be freed by free_imsm
3594 static int load_imsm_mpb(int fd
, struct intel_super
*super
, char *devname
)
3596 unsigned long long dsize
;
3597 unsigned long long sectors
;
3599 struct imsm_super
*anchor
;
3602 get_dev_size(fd
, NULL
, &dsize
);
3605 pr_err("%s: device to small for imsm\n",
3610 if (lseek64(fd
, dsize
- (512 * 2), SEEK_SET
) < 0) {
3612 pr_err("Cannot seek to anchor block on %s: %s\n",
3613 devname
, strerror(errno
));
3617 if (posix_memalign((void**)&anchor
, 512, 512) != 0) {
3619 pr_err("Failed to allocate imsm anchor buffer on %s\n", devname
);
3622 if (read(fd
, anchor
, 512) != 512) {
3624 pr_err("Cannot read anchor block on %s: %s\n",
3625 devname
, strerror(errno
));
3630 if (strncmp((char *) anchor
->sig
, MPB_SIGNATURE
, MPB_SIG_LEN
) != 0) {
3632 pr_err("no IMSM anchor on %s\n", devname
);
3637 __free_imsm(super
, 0);
3638 /* reload capability and hba */
3640 /* capability and hba must be updated with new super allocation */
3641 find_intel_hba_capability(fd
, super
, devname
);
3642 super
->len
= ROUND_UP(anchor
->mpb_size
, 512);
3643 if (posix_memalign(&super
->buf
, 512, super
->len
) != 0) {
3645 pr_err("unable to allocate %zu byte mpb buffer\n",
3650 memcpy(super
->buf
, anchor
, 512);
3652 sectors
= mpb_sectors(anchor
) - 1;
3655 if (posix_memalign(&super
->migr_rec_buf
, 512, MIGR_REC_BUF_SIZE
) != 0) {
3656 pr_err("could not allocate migr_rec buffer\n");
3660 super
->clean_migration_record_by_mdmon
= 0;
3663 check_sum
= __gen_imsm_checksum(super
->anchor
);
3664 if (check_sum
!= __le32_to_cpu(super
->anchor
->check_sum
)) {
3666 pr_err("IMSM checksum %x != %x on %s\n",
3668 __le32_to_cpu(super
->anchor
->check_sum
),
3676 /* read the extended mpb */
3677 if (lseek64(fd
, dsize
- (512 * (2 + sectors
)), SEEK_SET
) < 0) {
3679 pr_err("Cannot seek to extended mpb on %s: %s\n",
3680 devname
, strerror(errno
));
3684 if ((unsigned)read(fd
, super
->buf
+ 512, super
->len
- 512) != super
->len
- 512) {
3686 pr_err("Cannot read extended mpb on %s: %s\n",
3687 devname
, strerror(errno
));
3691 check_sum
= __gen_imsm_checksum(super
->anchor
);
3692 if (check_sum
!= __le32_to_cpu(super
->anchor
->check_sum
)) {
3694 pr_err("IMSM checksum %x != %x on %s\n",
3695 check_sum
, __le32_to_cpu(super
->anchor
->check_sum
),
3700 /* FIXME the BBM log is disk specific so we cannot use this global
3701 * buffer for all disks. Ok for now since we only look at the global
3702 * bbm_log_size parameter to gate assembly
3704 super
->bbm_log
= __get_imsm_bbm_log(super
->anchor
);
3709 static int read_imsm_migr_rec(int fd
, struct intel_super
*super
);
3711 /* clears hi bits in metadata if MPB_ATTRIB_2TB_DISK not set */
3712 static void clear_hi(struct intel_super
*super
)
3714 struct imsm_super
*mpb
= super
->anchor
;
3716 if (mpb
->attributes
& MPB_ATTRIB_2TB_DISK
)
3718 for (i
= 0; i
< mpb
->num_disks
; ++i
) {
3719 struct imsm_disk
*disk
= &mpb
->disk
[i
];
3720 disk
->total_blocks_hi
= 0;
3722 for (i
= 0; i
< mpb
->num_raid_devs
; ++i
) {
3723 struct imsm_dev
*dev
= get_imsm_dev(super
, i
);
3726 for (n
= 0; n
< 2; ++n
) {
3727 struct imsm_map
*map
= get_imsm_map(dev
, n
);
3730 map
->pba_of_lba0_hi
= 0;
3731 map
->blocks_per_member_hi
= 0;
3732 map
->num_data_stripes_hi
= 0;
3738 load_and_parse_mpb(int fd
, struct intel_super
*super
, char *devname
, int keep_fd
)
3742 err
= load_imsm_mpb(fd
, super
, devname
);
3745 err
= load_imsm_disk(fd
, super
, devname
, keep_fd
);
3748 err
= parse_raid_devices(super
);
3753 static void __free_imsm_disk(struct dl
*d
)
3765 static void free_imsm_disks(struct intel_super
*super
)
3769 while (super
->disks
) {
3771 super
->disks
= d
->next
;
3772 __free_imsm_disk(d
);
3774 while (super
->disk_mgmt_list
) {
3775 d
= super
->disk_mgmt_list
;
3776 super
->disk_mgmt_list
= d
->next
;
3777 __free_imsm_disk(d
);
3779 while (super
->missing
) {
3781 super
->missing
= d
->next
;
3782 __free_imsm_disk(d
);
3787 /* free all the pieces hanging off of a super pointer */
3788 static void __free_imsm(struct intel_super
*super
, int free_disks
)
3790 struct intel_hba
*elem
, *next
;
3796 /* unlink capability description */
3798 if (super
->migr_rec_buf
) {
3799 free(super
->migr_rec_buf
);
3800 super
->migr_rec_buf
= NULL
;
3803 free_imsm_disks(super
);
3804 free_devlist(super
);
3808 free((void *)elem
->path
);
3816 static void free_imsm(struct intel_super
*super
)
3818 __free_imsm(super
, 1);
3822 static void free_super_imsm(struct supertype
*st
)
3824 struct intel_super
*super
= st
->sb
;
3833 static struct intel_super
*alloc_super(void)
3835 struct intel_super
*super
= xcalloc(1, sizeof(*super
));
3837 super
->current_vol
= -1;
3838 super
->create_offset
= ~((unsigned long long) 0);
3843 * find and allocate hba and OROM/EFI based on valid fd of RAID component device
3845 static int find_intel_hba_capability(int fd
, struct intel_super
*super
, char *devname
)
3847 struct sys_dev
*hba_name
;
3850 if ((fd
< 0) || check_env("IMSM_NO_PLATFORM")) {
3855 hba_name
= find_disk_attached_hba(fd
, NULL
);
3858 pr_err("%s is not attached to Intel(R) RAID controller.\n",
3862 rv
= attach_hba_to_super(super
, hba_name
);
3865 struct intel_hba
*hba
= super
->hba
;
3867 pr_err("%s is attached to Intel(R) %s RAID controller (%s),\n"
3868 " but the container is assigned to Intel(R) %s RAID controller (",
3870 get_sys_dev_type(hba_name
->type
),
3871 hba_name
->pci_id
? : "Err!",
3872 get_sys_dev_type(super
->hba
->type
));
3875 fprintf(stderr
, "%s", hba
->pci_id
? : "Err!");
3877 fprintf(stderr
, ", ");
3880 fprintf(stderr
, ").\n"
3881 " Mixing devices attached to different controllers is not allowed.\n");
3885 super
->orom
= find_imsm_capability(hba_name
);
3892 /* find_missing - helper routine for load_super_imsm_all that identifies
3893 * disks that have disappeared from the system. This routine relies on
3894 * the mpb being uptodate, which it is at load time.
3896 static int find_missing(struct intel_super
*super
)
3899 struct imsm_super
*mpb
= super
->anchor
;
3901 struct imsm_disk
*disk
;
3903 for (i
= 0; i
< mpb
->num_disks
; i
++) {
3904 disk
= __get_imsm_disk(mpb
, i
);
3905 dl
= serial_to_dl(disk
->serial
, super
);
3909 dl
= xmalloc(sizeof(*dl
));
3913 dl
->devname
= xstrdup("missing");
3915 serialcpy(dl
->serial
, disk
->serial
);
3918 dl
->next
= super
->missing
;
3919 super
->missing
= dl
;
3926 static struct intel_disk
*disk_list_get(__u8
*serial
, struct intel_disk
*disk_list
)
3928 struct intel_disk
*idisk
= disk_list
;
3931 if (serialcmp(idisk
->disk
.serial
, serial
) == 0)
3933 idisk
= idisk
->next
;
3939 static int __prep_thunderdome(struct intel_super
**table
, int tbl_size
,
3940 struct intel_super
*super
,
3941 struct intel_disk
**disk_list
)
3943 struct imsm_disk
*d
= &super
->disks
->disk
;
3944 struct imsm_super
*mpb
= super
->anchor
;
3947 for (i
= 0; i
< tbl_size
; i
++) {
3948 struct imsm_super
*tbl_mpb
= table
[i
]->anchor
;
3949 struct imsm_disk
*tbl_d
= &table
[i
]->disks
->disk
;
3951 if (tbl_mpb
->family_num
== mpb
->family_num
) {
3952 if (tbl_mpb
->check_sum
== mpb
->check_sum
) {
3953 dprintf("mpb from %d:%d matches %d:%d\n",
3954 super
->disks
->major
,
3955 super
->disks
->minor
,
3956 table
[i
]->disks
->major
,
3957 table
[i
]->disks
->minor
);
3961 if (((is_configured(d
) && !is_configured(tbl_d
)) ||
3962 is_configured(d
) == is_configured(tbl_d
)) &&
3963 tbl_mpb
->generation_num
< mpb
->generation_num
) {
3964 /* current version of the mpb is a
3965 * better candidate than the one in
3966 * super_table, but copy over "cross
3967 * generational" status
3969 struct intel_disk
*idisk
;
3971 dprintf("mpb from %d:%d replaces %d:%d\n",
3972 super
->disks
->major
,
3973 super
->disks
->minor
,
3974 table
[i
]->disks
->major
,
3975 table
[i
]->disks
->minor
);
3977 idisk
= disk_list_get(tbl_d
->serial
, *disk_list
);
3978 if (idisk
&& is_failed(&idisk
->disk
))
3979 tbl_d
->status
|= FAILED_DISK
;
3982 struct intel_disk
*idisk
;
3983 struct imsm_disk
*disk
;
3985 /* tbl_mpb is more up to date, but copy
3986 * over cross generational status before
3989 disk
= __serial_to_disk(d
->serial
, mpb
, NULL
);
3990 if (disk
&& is_failed(disk
))
3991 d
->status
|= FAILED_DISK
;
3993 idisk
= disk_list_get(d
->serial
, *disk_list
);
3996 if (disk
&& is_configured(disk
))
3997 idisk
->disk
.status
|= CONFIGURED_DISK
;
4000 dprintf("mpb from %d:%d prefer %d:%d\n",
4001 super
->disks
->major
,
4002 super
->disks
->minor
,
4003 table
[i
]->disks
->major
,
4004 table
[i
]->disks
->minor
);
4012 table
[tbl_size
++] = super
;
4016 /* update/extend the merged list of imsm_disk records */
4017 for (j
= 0; j
< mpb
->num_disks
; j
++) {
4018 struct imsm_disk
*disk
= __get_imsm_disk(mpb
, j
);
4019 struct intel_disk
*idisk
;
4021 idisk
= disk_list_get(disk
->serial
, *disk_list
);
4023 idisk
->disk
.status
|= disk
->status
;
4024 if (is_configured(&idisk
->disk
) ||
4025 is_failed(&idisk
->disk
))
4026 idisk
->disk
.status
&= ~(SPARE_DISK
);
4028 idisk
= xcalloc(1, sizeof(*idisk
));
4029 idisk
->owner
= IMSM_UNKNOWN_OWNER
;
4030 idisk
->disk
= *disk
;
4031 idisk
->next
= *disk_list
;
4035 if (serialcmp(idisk
->disk
.serial
, d
->serial
) == 0)
4042 static struct intel_super
*
4043 validate_members(struct intel_super
*super
, struct intel_disk
*disk_list
,
4046 struct imsm_super
*mpb
= super
->anchor
;
4050 for (i
= 0; i
< mpb
->num_disks
; i
++) {
4051 struct imsm_disk
*disk
= __get_imsm_disk(mpb
, i
);
4052 struct intel_disk
*idisk
;
4054 idisk
= disk_list_get(disk
->serial
, disk_list
);
4056 if (idisk
->owner
== owner
||
4057 idisk
->owner
== IMSM_UNKNOWN_OWNER
)
4060 dprintf("'%.16s' owner %d != %d\n",
4061 disk
->serial
, idisk
->owner
,
4064 dprintf("unknown disk %x [%d]: %.16s\n",
4065 __le32_to_cpu(mpb
->family_num
), i
,
4071 if (ok_count
== mpb
->num_disks
)
4076 static void show_conflicts(__u32 family_num
, struct intel_super
*super_list
)
4078 struct intel_super
*s
;
4080 for (s
= super_list
; s
; s
= s
->next
) {
4081 if (family_num
!= s
->anchor
->family_num
)
4083 pr_err("Conflict, offlining family %#x on '%s'\n",
4084 __le32_to_cpu(family_num
), s
->disks
->devname
);
4088 static struct intel_super
*
4089 imsm_thunderdome(struct intel_super
**super_list
, int len
)
4091 struct intel_super
*super_table
[len
];
4092 struct intel_disk
*disk_list
= NULL
;
4093 struct intel_super
*champion
, *spare
;
4094 struct intel_super
*s
, **del
;
4099 memset(super_table
, 0, sizeof(super_table
));
4100 for (s
= *super_list
; s
; s
= s
->next
)
4101 tbl_size
= __prep_thunderdome(super_table
, tbl_size
, s
, &disk_list
);
4103 for (i
= 0; i
< tbl_size
; i
++) {
4104 struct imsm_disk
*d
;
4105 struct intel_disk
*idisk
;
4106 struct imsm_super
*mpb
= super_table
[i
]->anchor
;
4109 d
= &s
->disks
->disk
;
4111 /* 'd' must appear in merged disk list for its
4112 * configuration to be valid
4114 idisk
= disk_list_get(d
->serial
, disk_list
);
4115 if (idisk
&& idisk
->owner
== i
)
4116 s
= validate_members(s
, disk_list
, i
);
4121 dprintf("marking family: %#x from %d:%d offline\n",
4123 super_table
[i
]->disks
->major
,
4124 super_table
[i
]->disks
->minor
);
4128 /* This is where the mdadm implementation differs from the Windows
4129 * driver which has no strict concept of a container. We can only
4130 * assemble one family from a container, so when returning a prodigal
4131 * array member to this system the code will not be able to disambiguate
4132 * the container contents that should be assembled ("foreign" versus
4133 * "local"). It requires user intervention to set the orig_family_num
4134 * to a new value to establish a new container. The Windows driver in
4135 * this situation fixes up the volume name in place and manages the
4136 * foreign array as an independent entity.
4141 for (i
= 0; i
< tbl_size
; i
++) {
4142 struct intel_super
*tbl_ent
= super_table
[i
];
4148 if (tbl_ent
->anchor
->num_raid_devs
== 0) {
4153 if (s
&& !is_spare
) {
4154 show_conflicts(tbl_ent
->anchor
->family_num
, *super_list
);
4156 } else if (!s
&& !is_spare
)
4169 pr_err("Chose family %#x on '%s', assemble conflicts to new container with '--update=uuid'\n",
4170 __le32_to_cpu(s
->anchor
->family_num
), s
->disks
->devname
);
4172 /* collect all dl's onto 'champion', and update them to
4173 * champion's version of the status
4175 for (s
= *super_list
; s
; s
= s
->next
) {
4176 struct imsm_super
*mpb
= champion
->anchor
;
4177 struct dl
*dl
= s
->disks
;
4182 mpb
->attributes
|= s
->anchor
->attributes
& MPB_ATTRIB_2TB_DISK
;
4184 for (i
= 0; i
< mpb
->num_disks
; i
++) {
4185 struct imsm_disk
*disk
;
4187 disk
= __serial_to_disk(dl
->serial
, mpb
, &dl
->index
);
4190 /* only set index on disks that are a member of
4191 * a populated contianer, i.e. one with
4194 if (is_failed(&dl
->disk
))
4196 else if (is_spare(&dl
->disk
))
4202 if (i
>= mpb
->num_disks
) {
4203 struct intel_disk
*idisk
;
4205 idisk
= disk_list_get(dl
->serial
, disk_list
);
4206 if (idisk
&& is_spare(&idisk
->disk
) &&
4207 !is_failed(&idisk
->disk
) && !is_configured(&idisk
->disk
))
4215 dl
->next
= champion
->disks
;
4216 champion
->disks
= dl
;
4220 /* delete 'champion' from super_list */
4221 for (del
= super_list
; *del
; ) {
4222 if (*del
== champion
) {
4223 *del
= (*del
)->next
;
4226 del
= &(*del
)->next
;
4228 champion
->next
= NULL
;
4232 struct intel_disk
*idisk
= disk_list
;
4234 disk_list
= disk_list
->next
;
4242 get_sra_super_block(int fd
, struct intel_super
**super_list
, char *devname
, int *max
, int keep_fd
);
4243 static int get_super_block(struct intel_super
**super_list
, char *devnm
, char *devname
,
4244 int major
, int minor
, int keep_fd
);
4246 get_devlist_super_block(struct md_list
*devlist
, struct intel_super
**super_list
,
4247 int *max
, int keep_fd
);
4249 static int load_super_imsm_all(struct supertype
*st
, int fd
, void **sbp
,
4250 char *devname
, struct md_list
*devlist
,
4253 struct intel_super
*super_list
= NULL
;
4254 struct intel_super
*super
= NULL
;
4259 /* 'fd' is an opened container */
4260 err
= get_sra_super_block(fd
, &super_list
, devname
, &i
, keep_fd
);
4262 /* get super block from devlist devices */
4263 err
= get_devlist_super_block(devlist
, &super_list
, &i
, keep_fd
);
4266 /* all mpbs enter, maybe one leaves */
4267 super
= imsm_thunderdome(&super_list
, i
);
4273 if (find_missing(super
) != 0) {
4279 /* load migration record */
4280 err
= load_imsm_migr_rec(super
, NULL
);
4282 /* migration is in progress,
4283 * but migr_rec cannot be loaded,
4289 /* Check migration compatibility */
4290 if ((err
== 0) && (check_mpb_migr_compatibility(super
) != 0)) {
4291 pr_err("Unsupported migration detected");
4293 fprintf(stderr
, " on %s\n", devname
);
4295 fprintf(stderr
, " (IMSM).\n");
4304 while (super_list
) {
4305 struct intel_super
*s
= super_list
;
4307 super_list
= super_list
->next
;
4316 strcpy(st
->container_devnm
, fd2devnm(fd
));
4318 st
->container_devnm
[0] = 0;
4319 if (err
== 0 && st
->ss
== NULL
) {
4320 st
->ss
= &super_imsm
;
4321 st
->minor_version
= 0;
4322 st
->max_devs
= IMSM_MAX_DEVICES
;
4328 get_devlist_super_block(struct md_list
*devlist
, struct intel_super
**super_list
,
4329 int *max
, int keep_fd
)
4331 struct md_list
*tmpdev
;
4335 for (i
= 0, tmpdev
= devlist
; tmpdev
; tmpdev
= tmpdev
->next
) {
4336 if (tmpdev
->used
!= 1)
4338 if (tmpdev
->container
== 1) {
4340 int fd
= dev_open(tmpdev
->devname
, O_RDONLY
|O_EXCL
);
4342 pr_err("cannot open device %s: %s\n",
4343 tmpdev
->devname
, strerror(errno
));
4347 err
= get_sra_super_block(fd
, super_list
,
4348 tmpdev
->devname
, &lmax
,
4357 int major
= major(tmpdev
->st_rdev
);
4358 int minor
= minor(tmpdev
->st_rdev
);
4359 err
= get_super_block(super_list
,
4376 static int get_super_block(struct intel_super
**super_list
, char *devnm
, char *devname
,
4377 int major
, int minor
, int keep_fd
)
4379 struct intel_super
*s
= NULL
;
4391 sprintf(nm
, "%d:%d", major
, minor
);
4392 dfd
= dev_open(nm
, O_RDWR
);
4398 find_intel_hba_capability(dfd
, s
, devname
);
4399 err
= load_and_parse_mpb(dfd
, s
, NULL
, keep_fd
);
4401 /* retry the load if we might have raced against mdmon */
4402 if (err
== 3 && devnm
&& mdmon_running(devnm
))
4403 for (retry
= 0; retry
< 3; retry
++) {
4405 err
= load_and_parse_mpb(dfd
, s
, NULL
, keep_fd
);
4411 s
->next
= *super_list
;
4419 if ((dfd
>= 0) && (!keep_fd
))
4426 get_sra_super_block(int fd
, struct intel_super
**super_list
, char *devname
, int *max
, int keep_fd
)
4433 sra
= sysfs_read(fd
, NULL
, GET_LEVEL
|GET_VERSION
|GET_DEVS
|GET_STATE
);
4437 if (sra
->array
.major_version
!= -1 ||
4438 sra
->array
.minor_version
!= -2 ||
4439 strcmp(sra
->text_version
, "imsm") != 0) {
4444 devnm
= fd2devnm(fd
);
4445 for (sd
= sra
->devs
, i
= 0; sd
; sd
= sd
->next
, i
++) {
4446 if (get_super_block(super_list
, devnm
, devname
,
4447 sd
->disk
.major
, sd
->disk
.minor
, keep_fd
) != 0) {
4458 static int load_container_imsm(struct supertype
*st
, int fd
, char *devname
)
4460 return load_super_imsm_all(st
, fd
, &st
->sb
, devname
, NULL
, 1);
4464 static int load_super_imsm(struct supertype
*st
, int fd
, char *devname
)
4466 struct intel_super
*super
;
4470 if (test_partition(fd
))
4471 /* IMSM not allowed on partitions */
4474 free_super_imsm(st
);
4476 super
= alloc_super();
4477 /* Load hba and capabilities if they exist.
4478 * But do not preclude loading metadata in case capabilities or hba are
4479 * non-compliant and ignore_hw_compat is set.
4481 rv
= find_intel_hba_capability(fd
, super
, devname
);
4482 /* no orom/efi or non-intel hba of the disk */
4483 if ((rv
!= 0) && (st
->ignore_hw_compat
== 0)) {
4485 pr_err("No OROM/EFI properties for %s\n", devname
);
4489 rv
= load_and_parse_mpb(fd
, super
, devname
, 0);
4491 /* retry the load if we might have raced against mdmon */
4493 struct mdstat_ent
*mdstat
= mdstat_by_component(fd2devnm(fd
));
4495 if (mdstat
&& mdmon_running(mdstat
->devnm
) && getpid() != mdmon_pid(mdstat
->devnm
)) {
4496 for (retry
= 0; retry
< 3; retry
++) {
4498 rv
= load_and_parse_mpb(fd
, super
, devname
, 0);
4504 free_mdstat(mdstat
);
4509 pr_err("Failed to load all information sections on %s\n", devname
);
4515 if (st
->ss
== NULL
) {
4516 st
->ss
= &super_imsm
;
4517 st
->minor_version
= 0;
4518 st
->max_devs
= IMSM_MAX_DEVICES
;
4521 /* load migration record */
4522 if (load_imsm_migr_rec(super
, NULL
) == 0) {
4523 /* Check for unsupported migration features */
4524 if (check_mpb_migr_compatibility(super
) != 0) {
4525 pr_err("Unsupported migration detected");
4527 fprintf(stderr
, " on %s\n", devname
);
4529 fprintf(stderr
, " (IMSM).\n");
4537 static __u16
info_to_blocks_per_strip(mdu_array_info_t
*info
)
4539 if (info
->level
== 1)
4541 return info
->chunk_size
>> 9;
4544 static unsigned long long info_to_blocks_per_member(mdu_array_info_t
*info
,
4545 unsigned long long size
)
4547 if (info
->level
== 1)
4550 return (size
* 2) & ~(info_to_blocks_per_strip(info
) - 1);
4553 static void imsm_update_version_info(struct intel_super
*super
)
4555 /* update the version and attributes */
4556 struct imsm_super
*mpb
= super
->anchor
;
4558 struct imsm_dev
*dev
;
4559 struct imsm_map
*map
;
4562 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
4563 dev
= get_imsm_dev(super
, i
);
4564 map
= get_imsm_map(dev
, MAP_0
);
4565 if (__le32_to_cpu(dev
->size_high
) > 0)
4566 mpb
->attributes
|= MPB_ATTRIB_2TB
;
4568 /* FIXME detect when an array spans a port multiplier */
4570 mpb
->attributes
|= MPB_ATTRIB_PM
;
4573 if (mpb
->num_raid_devs
> 1 ||
4574 mpb
->attributes
!= MPB_ATTRIB_CHECKSUM_VERIFY
) {
4575 version
= MPB_VERSION_ATTRIBS
;
4576 switch (get_imsm_raid_level(map
)) {
4577 case 0: mpb
->attributes
|= MPB_ATTRIB_RAID0
; break;
4578 case 1: mpb
->attributes
|= MPB_ATTRIB_RAID1
; break;
4579 case 10: mpb
->attributes
|= MPB_ATTRIB_RAID10
; break;
4580 case 5: mpb
->attributes
|= MPB_ATTRIB_RAID5
; break;
4583 if (map
->num_members
>= 5)
4584 version
= MPB_VERSION_5OR6_DISK_ARRAY
;
4585 else if (dev
->status
== DEV_CLONE_N_GO
)
4586 version
= MPB_VERSION_CNG
;
4587 else if (get_imsm_raid_level(map
) == 5)
4588 version
= MPB_VERSION_RAID5
;
4589 else if (map
->num_members
>= 3)
4590 version
= MPB_VERSION_3OR4_DISK_ARRAY
;
4591 else if (get_imsm_raid_level(map
) == 1)
4592 version
= MPB_VERSION_RAID1
;
4594 version
= MPB_VERSION_RAID0
;
4596 strcpy(((char *) mpb
->sig
) + strlen(MPB_SIGNATURE
), version
);
4600 static int check_name(struct intel_super
*super
, char *name
, int quiet
)
4602 struct imsm_super
*mpb
= super
->anchor
;
4603 char *reason
= NULL
;
4606 if (strlen(name
) > MAX_RAID_SERIAL_LEN
)
4607 reason
= "must be 16 characters or less";
4609 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
4610 struct imsm_dev
*dev
= get_imsm_dev(super
, i
);
4612 if (strncmp((char *) dev
->volume
, name
, MAX_RAID_SERIAL_LEN
) == 0) {
4613 reason
= "already exists";
4618 if (reason
&& !quiet
)
4619 pr_err("imsm volume name %s\n", reason
);
4624 static int init_super_imsm_volume(struct supertype
*st
, mdu_array_info_t
*info
,
4625 unsigned long long size
, char *name
,
4626 char *homehost
, int *uuid
,
4627 long long data_offset
)
4629 /* We are creating a volume inside a pre-existing container.
4630 * so st->sb is already set.
4632 struct intel_super
*super
= st
->sb
;
4633 struct imsm_super
*mpb
= super
->anchor
;
4634 struct intel_dev
*dv
;
4635 struct imsm_dev
*dev
;
4636 struct imsm_vol
*vol
;
4637 struct imsm_map
*map
;
4638 int idx
= mpb
->num_raid_devs
;
4640 unsigned long long array_blocks
;
4641 size_t size_old
, size_new
;
4642 unsigned long long num_data_stripes
;
4644 if (super
->orom
&& mpb
->num_raid_devs
>= super
->orom
->vpa
) {
4645 pr_err("This imsm-container already has the maximum of %d volumes\n", super
->orom
->vpa
);
4649 /* ensure the mpb is large enough for the new data */
4650 size_old
= __le32_to_cpu(mpb
->mpb_size
);
4651 size_new
= disks_to_mpb_size(info
->nr_disks
);
4652 if (size_new
> size_old
) {
4654 size_t size_round
= ROUND_UP(size_new
, 512);
4656 if (posix_memalign(&mpb_new
, 512, size_round
) != 0) {
4657 pr_err("could not allocate new mpb\n");
4660 if (posix_memalign(&super
->migr_rec_buf
, 512,
4661 MIGR_REC_BUF_SIZE
) != 0) {
4662 pr_err("could not allocate migr_rec buffer\n");
4668 memcpy(mpb_new
, mpb
, size_old
);
4671 super
->anchor
= mpb_new
;
4672 mpb
->mpb_size
= __cpu_to_le32(size_new
);
4673 memset(mpb_new
+ size_old
, 0, size_round
- size_old
);
4675 super
->current_vol
= idx
;
4677 /* handle 'failed_disks' by either:
4678 * a) create dummy disk entries in the table if this the first
4679 * volume in the array. We add them here as this is the only
4680 * opportunity to add them. add_to_super_imsm_volume()
4681 * handles the non-failed disks and continues incrementing
4683 * b) validate that 'failed_disks' matches the current number
4684 * of missing disks if the container is populated
4686 if (super
->current_vol
== 0) {
4688 for (i
= 0; i
< info
->failed_disks
; i
++) {
4689 struct imsm_disk
*disk
;
4692 disk
= __get_imsm_disk(mpb
, i
);
4693 disk
->status
= CONFIGURED_DISK
| FAILED_DISK
;
4694 disk
->scsi_id
= __cpu_to_le32(~(__u32
)0);
4695 snprintf((char *) disk
->serial
, MAX_RAID_SERIAL_LEN
,
4698 find_missing(super
);
4703 for (d
= super
->missing
; d
; d
= d
->next
)
4705 if (info
->failed_disks
> missing
) {
4706 pr_err("unable to add 'missing' disk to container\n");
4711 if (!check_name(super
, name
, 0))
4713 dv
= xmalloc(sizeof(*dv
));
4714 dev
= xcalloc(1, sizeof(*dev
) + sizeof(__u32
) * (info
->raid_disks
- 1));
4715 strncpy((char *) dev
->volume
, name
, MAX_RAID_SERIAL_LEN
);
4716 array_blocks
= calc_array_size(info
->level
, info
->raid_disks
,
4717 info
->layout
, info
->chunk_size
,
4719 /* round array size down to closest MB */
4720 array_blocks
= (array_blocks
>> SECT_PER_MB_SHIFT
) << SECT_PER_MB_SHIFT
;
4722 dev
->size_low
= __cpu_to_le32((__u32
) array_blocks
);
4723 dev
->size_high
= __cpu_to_le32((__u32
) (array_blocks
>> 32));
4724 dev
->status
= (DEV_READ_COALESCING
| DEV_WRITE_COALESCING
);
4726 vol
->migr_state
= 0;
4727 set_migr_type(dev
, MIGR_INIT
);
4728 vol
->dirty
= !info
->state
;
4729 vol
->curr_migr_unit
= 0;
4730 map
= get_imsm_map(dev
, MAP_0
);
4731 set_pba_of_lba0(map
, super
->create_offset
);
4732 set_blocks_per_member(map
, info_to_blocks_per_member(info
, size
));
4733 map
->blocks_per_strip
= __cpu_to_le16(info_to_blocks_per_strip(info
));
4734 map
->failed_disk_num
= ~0;
4735 if (info
->level
> 0)
4736 map
->map_state
= (info
->state
? IMSM_T_STATE_NORMAL
4737 : IMSM_T_STATE_UNINITIALIZED
);
4739 map
->map_state
= info
->failed_disks
? IMSM_T_STATE_FAILED
:
4740 IMSM_T_STATE_NORMAL
;
4743 if (info
->level
== 1 && info
->raid_disks
> 2) {
4746 pr_err("imsm does not support more than 2 disksin a raid1 volume\n");
4750 map
->raid_level
= info
->level
;
4751 if (info
->level
== 10) {
4752 map
->raid_level
= 1;
4753 map
->num_domains
= info
->raid_disks
/ 2;
4754 } else if (info
->level
== 1)
4755 map
->num_domains
= info
->raid_disks
;
4757 map
->num_domains
= 1;
4759 /* info->size is only int so use the 'size' parameter instead */
4760 num_data_stripes
= (size
* 2) / info_to_blocks_per_strip(info
);
4761 num_data_stripes
/= map
->num_domains
;
4762 set_num_data_stripes(map
, num_data_stripes
);
4764 map
->num_members
= info
->raid_disks
;
4765 for (i
= 0; i
< map
->num_members
; i
++) {
4766 /* initialized in add_to_super */
4767 set_imsm_ord_tbl_ent(map
, i
, IMSM_ORD_REBUILD
);
4769 mpb
->num_raid_devs
++;
4772 dv
->index
= super
->current_vol
;
4773 dv
->next
= super
->devlist
;
4774 super
->devlist
= dv
;
4776 imsm_update_version_info(super
);
4781 static int init_super_imsm(struct supertype
*st
, mdu_array_info_t
*info
,
4782 unsigned long long size
, char *name
,
4783 char *homehost
, int *uuid
,
4784 unsigned long long data_offset
)
4786 /* This is primarily called by Create when creating a new array.
4787 * We will then get add_to_super called for each component, and then
4788 * write_init_super called to write it out to each device.
4789 * For IMSM, Create can create on fresh devices or on a pre-existing
4791 * To create on a pre-existing array a different method will be called.
4792 * This one is just for fresh drives.
4794 struct intel_super
*super
;
4795 struct imsm_super
*mpb
;
4799 if (data_offset
!= INVALID_SECTORS
) {
4800 pr_err("data-offset not supported by imsm\n");
4805 return init_super_imsm_volume(st
, info
, size
, name
, homehost
, uuid
,
4809 mpb_size
= disks_to_mpb_size(info
->nr_disks
);
4813 super
= alloc_super();
4814 if (super
&& posix_memalign(&super
->buf
, 512, mpb_size
) != 0) {
4819 pr_err("could not allocate superblock\n");
4822 if (posix_memalign(&super
->migr_rec_buf
, 512, MIGR_REC_BUF_SIZE
) != 0) {
4823 pr_err("could not allocate migr_rec buffer\n");
4828 memset(super
->buf
, 0, mpb_size
);
4830 mpb
->mpb_size
= __cpu_to_le32(mpb_size
);
4834 /* zeroing superblock */
4838 mpb
->attributes
= MPB_ATTRIB_CHECKSUM_VERIFY
;
4840 version
= (char *) mpb
->sig
;
4841 strcpy(version
, MPB_SIGNATURE
);
4842 version
+= strlen(MPB_SIGNATURE
);
4843 strcpy(version
, MPB_VERSION_RAID0
);
4849 static int add_to_super_imsm_volume(struct supertype
*st
, mdu_disk_info_t
*dk
,
4850 int fd
, char *devname
)
4852 struct intel_super
*super
= st
->sb
;
4853 struct imsm_super
*mpb
= super
->anchor
;
4854 struct imsm_disk
*_disk
;
4855 struct imsm_dev
*dev
;
4856 struct imsm_map
*map
;
4860 dev
= get_imsm_dev(super
, super
->current_vol
);
4861 map
= get_imsm_map(dev
, MAP_0
);
4863 if (! (dk
->state
& (1<<MD_DISK_SYNC
))) {
4864 pr_err("%s: Cannot add spare devices to IMSM volume\n",
4870 /* we're doing autolayout so grab the pre-marked (in
4871 * validate_geometry) raid_disk
4873 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
4874 if (dl
->raiddisk
== dk
->raid_disk
)
4877 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
4878 if (dl
->major
== dk
->major
&&
4879 dl
->minor
== dk
->minor
)
4884 pr_err("%s is not a member of the same container\n", devname
);
4888 /* add a pristine spare to the metadata */
4889 if (dl
->index
< 0) {
4890 dl
->index
= super
->anchor
->num_disks
;
4891 super
->anchor
->num_disks
++;
4893 /* Check the device has not already been added */
4894 slot
= get_imsm_disk_slot(map
, dl
->index
);
4896 (get_imsm_ord_tbl_ent(dev
, slot
, MAP_X
) & IMSM_ORD_REBUILD
) == 0) {
4897 pr_err("%s has been included in this array twice\n",
4901 set_imsm_ord_tbl_ent(map
, dk
->raid_disk
, dl
->index
);
4902 dl
->disk
.status
= CONFIGURED_DISK
;
4904 /* update size of 'missing' disks to be at least as large as the
4905 * largest acitve member (we only have dummy missing disks when
4906 * creating the first volume)
4908 if (super
->current_vol
== 0) {
4909 for (df
= super
->missing
; df
; df
= df
->next
) {
4910 if (total_blocks(&dl
->disk
) > total_blocks(&df
->disk
))
4911 set_total_blocks(&df
->disk
, total_blocks(&dl
->disk
));
4912 _disk
= __get_imsm_disk(mpb
, df
->index
);
4917 /* refresh unset/failed slots to point to valid 'missing' entries */
4918 for (df
= super
->missing
; df
; df
= df
->next
)
4919 for (slot
= 0; slot
< mpb
->num_disks
; slot
++) {
4920 __u32 ord
= get_imsm_ord_tbl_ent(dev
, slot
, MAP_X
);
4922 if ((ord
& IMSM_ORD_REBUILD
) == 0)
4924 set_imsm_ord_tbl_ent(map
, slot
, df
->index
| IMSM_ORD_REBUILD
);
4925 if (is_gen_migration(dev
)) {
4926 struct imsm_map
*map2
= get_imsm_map(dev
,
4928 int slot2
= get_imsm_disk_slot(map2
, df
->index
);
4929 if ((slot2
< map2
->num_members
) &&
4931 __u32 ord2
= get_imsm_ord_tbl_ent(dev
,
4934 if ((unsigned)df
->index
==
4936 set_imsm_ord_tbl_ent(map2
,
4942 dprintf("set slot:%d to missing disk:%d\n", slot
, df
->index
);
4946 /* if we are creating the first raid device update the family number */
4947 if (super
->current_vol
== 0) {
4949 struct imsm_dev
*_dev
= __get_imsm_dev(mpb
, 0);
4951 _disk
= __get_imsm_disk(mpb
, dl
->index
);
4952 if (!_dev
|| !_disk
) {
4953 pr_err("BUG mpb setup error\n");
4959 sum
+= __gen_imsm_checksum(mpb
);
4960 mpb
->family_num
= __cpu_to_le32(sum
);
4961 mpb
->orig_family_num
= mpb
->family_num
;
4963 super
->current_disk
= dl
;
4968 * Function marks disk as spare and restores disk serial
4969 * in case it was previously marked as failed by takeover operation
4971 * -1 : critical error
4972 * 0 : disk is marked as spare but serial is not set
4975 int mark_spare(struct dl
*disk
)
4977 __u8 serial
[MAX_RAID_SERIAL_LEN
];
4984 if (!imsm_read_serial(disk
->fd
, NULL
, serial
)) {
4985 /* Restore disk serial number, because takeover marks disk
4986 * as failed and adds to serial ':0' before it becomes
4989 serialcpy(disk
->serial
, serial
);
4990 serialcpy(disk
->disk
.serial
, serial
);
4993 disk
->disk
.status
= SPARE_DISK
;
4999 static int add_to_super_imsm(struct supertype
*st
, mdu_disk_info_t
*dk
,
5000 int fd
, char *devname
,
5001 unsigned long long data_offset
)
5003 struct intel_super
*super
= st
->sb
;
5005 unsigned long long size
;
5010 /* If we are on an RAID enabled platform check that the disk is
5011 * attached to the raid controller.
5012 * We do not need to test disks attachment for container based additions,
5013 * they shall be already tested when container was created/assembled.
5015 rv
= find_intel_hba_capability(fd
, super
, devname
);
5016 /* no orom/efi or non-intel hba of the disk */
5018 dprintf("capability: %p fd: %d ret: %d\n",
5019 super
->orom
, fd
, rv
);
5023 if (super
->current_vol
>= 0)
5024 return add_to_super_imsm_volume(st
, dk
, fd
, devname
);
5027 dd
= xcalloc(sizeof(*dd
), 1);
5028 dd
->major
= major(stb
.st_rdev
);
5029 dd
->minor
= minor(stb
.st_rdev
);
5030 dd
->devname
= devname
? xstrdup(devname
) : NULL
;
5033 dd
->action
= DISK_ADD
;
5034 rv
= imsm_read_serial(fd
, devname
, dd
->serial
);
5036 pr_err("failed to retrieve scsi serial, aborting\n");
5041 get_dev_size(fd
, NULL
, &size
);
5042 /* clear migr_rec when adding disk to container */
5043 memset(super
->migr_rec_buf
, 0, MIGR_REC_BUF_SIZE
);
5044 if (lseek64(fd
, size
- MIGR_REC_POSITION
, SEEK_SET
) >= 0) {
5045 if (write(fd
, super
->migr_rec_buf
,
5046 MIGR_REC_BUF_SIZE
) != MIGR_REC_BUF_SIZE
)
5047 perror("Write migr_rec failed");
5051 serialcpy(dd
->disk
.serial
, dd
->serial
);
5052 set_total_blocks(&dd
->disk
, size
);
5053 if (__le32_to_cpu(dd
->disk
.total_blocks_hi
) > 0) {
5054 struct imsm_super
*mpb
= super
->anchor
;
5055 mpb
->attributes
|= MPB_ATTRIB_2TB_DISK
;
5058 if (sysfs_disk_to_scsi_id(fd
, &id
) == 0)
5059 dd
->disk
.scsi_id
= __cpu_to_le32(id
);
5061 dd
->disk
.scsi_id
= __cpu_to_le32(0);
5063 if (st
->update_tail
) {
5064 dd
->next
= super
->disk_mgmt_list
;
5065 super
->disk_mgmt_list
= dd
;
5067 dd
->next
= super
->disks
;
5069 super
->updates_pending
++;
5075 static int remove_from_super_imsm(struct supertype
*st
, mdu_disk_info_t
*dk
)
5077 struct intel_super
*super
= st
->sb
;
5080 /* remove from super works only in mdmon - for communication
5081 * manager - monitor. Check if communication memory buffer
5084 if (!st
->update_tail
) {
5085 pr_err("shall be used in mdmon context only\n");
5088 dd
= xcalloc(1, sizeof(*dd
));
5089 dd
->major
= dk
->major
;
5090 dd
->minor
= dk
->minor
;
5093 dd
->action
= DISK_REMOVE
;
5095 dd
->next
= super
->disk_mgmt_list
;
5096 super
->disk_mgmt_list
= dd
;
5101 static int store_imsm_mpb(int fd
, struct imsm_super
*mpb
);
5105 struct imsm_super anchor
;
5106 } spare_record
__attribute__ ((aligned(512)));
5108 /* spare records have their own family number and do not have any defined raid
5111 static int write_super_imsm_spares(struct intel_super
*super
, int doclose
)
5113 struct imsm_super
*mpb
= super
->anchor
;
5114 struct imsm_super
*spare
= &spare_record
.anchor
;
5118 spare
->mpb_size
= __cpu_to_le32(sizeof(struct imsm_super
)),
5119 spare
->generation_num
= __cpu_to_le32(1UL),
5120 spare
->attributes
= MPB_ATTRIB_CHECKSUM_VERIFY
;
5121 spare
->num_disks
= 1,
5122 spare
->num_raid_devs
= 0,
5123 spare
->cache_size
= mpb
->cache_size
,
5124 spare
->pwr_cycle_count
= __cpu_to_le32(1),
5126 snprintf((char *) spare
->sig
, MAX_SIGNATURE_LENGTH
,
5127 MPB_SIGNATURE MPB_VERSION_RAID0
);
5129 for (d
= super
->disks
; d
; d
= d
->next
) {
5133 spare
->disk
[0] = d
->disk
;
5134 if (__le32_to_cpu(d
->disk
.total_blocks_hi
) > 0)
5135 spare
->attributes
|= MPB_ATTRIB_2TB_DISK
;
5137 sum
= __gen_imsm_checksum(spare
);
5138 spare
->family_num
= __cpu_to_le32(sum
);
5139 spare
->orig_family_num
= 0;
5140 sum
= __gen_imsm_checksum(spare
);
5141 spare
->check_sum
= __cpu_to_le32(sum
);
5143 if (store_imsm_mpb(d
->fd
, spare
)) {
5144 pr_err("failed for device %d:%d %s\n",
5145 d
->major
, d
->minor
, strerror(errno
));
5157 static int write_super_imsm(struct supertype
*st
, int doclose
)
5159 struct intel_super
*super
= st
->sb
;
5160 struct imsm_super
*mpb
= super
->anchor
;
5166 __u32 mpb_size
= sizeof(struct imsm_super
) - sizeof(struct imsm_disk
);
5168 int clear_migration_record
= 1;
5170 /* 'generation' is incremented everytime the metadata is written */
5171 generation
= __le32_to_cpu(mpb
->generation_num
);
5173 mpb
->generation_num
= __cpu_to_le32(generation
);
5175 /* fix up cases where previous mdadm releases failed to set
5178 if (mpb
->orig_family_num
== 0)
5179 mpb
->orig_family_num
= mpb
->family_num
;
5181 for (d
= super
->disks
; d
; d
= d
->next
) {
5185 mpb
->disk
[d
->index
] = d
->disk
;
5189 for (d
= super
->missing
; d
; d
= d
->next
) {
5190 mpb
->disk
[d
->index
] = d
->disk
;
5193 mpb
->num_disks
= num_disks
;
5194 mpb_size
+= sizeof(struct imsm_disk
) * mpb
->num_disks
;
5196 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
5197 struct imsm_dev
*dev
= __get_imsm_dev(mpb
, i
);
5198 struct imsm_dev
*dev2
= get_imsm_dev(super
, i
);
5200 imsm_copy_dev(dev
, dev2
);
5201 mpb_size
+= sizeof_imsm_dev(dev
, 0);
5203 if (is_gen_migration(dev2
))
5204 clear_migration_record
= 0;
5206 mpb_size
+= __le32_to_cpu(mpb
->bbm_log_size
);
5207 mpb
->mpb_size
= __cpu_to_le32(mpb_size
);
5209 /* recalculate checksum */
5210 sum
= __gen_imsm_checksum(mpb
);
5211 mpb
->check_sum
= __cpu_to_le32(sum
);
5213 if (super
->clean_migration_record_by_mdmon
) {
5214 clear_migration_record
= 1;
5215 super
->clean_migration_record_by_mdmon
= 0;
5217 if (clear_migration_record
)
5218 memset(super
->migr_rec_buf
, 0, MIGR_REC_BUF_SIZE
);
5220 /* write the mpb for disks that compose raid devices */
5221 for (d
= super
->disks
; d
; d
= d
->next
) {
5222 if (d
->index
< 0 || is_failed(&d
->disk
))
5225 if (clear_migration_record
) {
5226 unsigned long long dsize
;
5228 get_dev_size(d
->fd
, NULL
, &dsize
);
5229 if (lseek64(d
->fd
, dsize
- 512, SEEK_SET
) >= 0) {
5230 if (write(d
->fd
, super
->migr_rec_buf
,
5231 MIGR_REC_BUF_SIZE
) != MIGR_REC_BUF_SIZE
)
5232 perror("Write migr_rec failed");
5236 if (store_imsm_mpb(d
->fd
, mpb
))
5238 "failed for device %d:%d (fd: %d)%s\n",
5240 d
->fd
, strerror(errno
));
5249 return write_super_imsm_spares(super
, doclose
);
5254 static int create_array(struct supertype
*st
, int dev_idx
)
5257 struct imsm_update_create_array
*u
;
5258 struct intel_super
*super
= st
->sb
;
5259 struct imsm_dev
*dev
= get_imsm_dev(super
, dev_idx
);
5260 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
5261 struct disk_info
*inf
;
5262 struct imsm_disk
*disk
;
5265 len
= sizeof(*u
) - sizeof(*dev
) + sizeof_imsm_dev(dev
, 0) +
5266 sizeof(*inf
) * map
->num_members
;
5268 u
->type
= update_create_array
;
5269 u
->dev_idx
= dev_idx
;
5270 imsm_copy_dev(&u
->dev
, dev
);
5271 inf
= get_disk_info(u
);
5272 for (i
= 0; i
< map
->num_members
; i
++) {
5273 int idx
= get_imsm_disk_idx(dev
, i
, MAP_X
);
5275 disk
= get_imsm_disk(super
, idx
);
5277 disk
= get_imsm_missing(super
, idx
);
5278 serialcpy(inf
[i
].serial
, disk
->serial
);
5280 append_metadata_update(st
, u
, len
);
5285 static int mgmt_disk(struct supertype
*st
)
5287 struct intel_super
*super
= st
->sb
;
5289 struct imsm_update_add_remove_disk
*u
;
5291 if (!super
->disk_mgmt_list
)
5296 u
->type
= update_add_remove_disk
;
5297 append_metadata_update(st
, u
, len
);
5302 static int write_init_super_imsm(struct supertype
*st
)
5304 struct intel_super
*super
= st
->sb
;
5305 int current_vol
= super
->current_vol
;
5307 /* we are done with current_vol reset it to point st at the container */
5308 super
->current_vol
= -1;
5310 if (st
->update_tail
) {
5311 /* queue the recently created array / added disk
5312 * as a metadata update */
5315 /* determine if we are creating a volume or adding a disk */
5316 if (current_vol
< 0) {
5317 /* in the mgmt (add/remove) disk case we are running
5318 * in mdmon context, so don't close fd's
5320 return mgmt_disk(st
);
5322 rv
= create_array(st
, current_vol
);
5327 for (d
= super
->disks
; d
; d
= d
->next
)
5328 Kill(d
->devname
, NULL
, 0, -1, 1);
5329 return write_super_imsm(st
, 1);
5334 static int store_super_imsm(struct supertype
*st
, int fd
)
5336 struct intel_super
*super
= st
->sb
;
5337 struct imsm_super
*mpb
= super
? super
->anchor
: NULL
;
5343 return store_imsm_mpb(fd
, mpb
);
5349 static int imsm_bbm_log_size(struct imsm_super
*mpb
)
5351 return __le32_to_cpu(mpb
->bbm_log_size
);
5355 static int validate_geometry_imsm_container(struct supertype
*st
, int level
,
5356 int layout
, int raiddisks
, int chunk
,
5357 unsigned long long size
,
5358 unsigned long long data_offset
,
5360 unsigned long long *freesize
,
5364 unsigned long long ldsize
;
5365 struct intel_super
*super
=NULL
;
5368 if (level
!= LEVEL_CONTAINER
)
5373 fd
= open(dev
, O_RDONLY
|O_EXCL
, 0);
5376 pr_err("imsm: Cannot open %s: %s\n",
5377 dev
, strerror(errno
));
5380 if (!get_dev_size(fd
, dev
, &ldsize
)) {
5385 /* capabilities retrieve could be possible
5386 * note that there is no fd for the disks in array.
5388 super
= alloc_super();
5389 rv
= find_intel_hba_capability(fd
, super
, verbose
> 0 ? dev
: NULL
);
5393 fd2devname(fd
, str
);
5394 dprintf("fd: %d %s orom: %p rv: %d raiddisk: %d\n",
5395 fd
, str
, super
->orom
, rv
, raiddisks
);
5397 /* no orom/efi or non-intel hba of the disk */
5404 if (raiddisks
> super
->orom
->tds
) {
5406 pr_err("%d exceeds maximum number of platform supported disks: %d\n",
5407 raiddisks
, super
->orom
->tds
);
5411 if ((super
->orom
->attr
& IMSM_OROM_ATTR_2TB_DISK
) == 0 &&
5412 (ldsize
>> 9) >> 32 > 0) {
5414 pr_err("%s exceeds maximum platform supported size\n", dev
);
5420 *freesize
= avail_size_imsm(st
, ldsize
>> 9, data_offset
);
5426 static unsigned long long find_size(struct extent
*e
, int *idx
, int num_extents
)
5428 const unsigned long long base_start
= e
[*idx
].start
;
5429 unsigned long long end
= base_start
+ e
[*idx
].size
;
5432 if (base_start
== end
)
5436 for (i
= *idx
; i
< num_extents
; i
++) {
5437 /* extend overlapping extents */
5438 if (e
[i
].start
>= base_start
&&
5439 e
[i
].start
<= end
) {
5442 if (e
[i
].start
+ e
[i
].size
> end
)
5443 end
= e
[i
].start
+ e
[i
].size
;
5444 } else if (e
[i
].start
> end
) {
5450 return end
- base_start
;
5453 static unsigned long long merge_extents(struct intel_super
*super
, int sum_extents
)
5455 /* build a composite disk with all known extents and generate a new
5456 * 'maxsize' given the "all disks in an array must share a common start
5457 * offset" constraint
5459 struct extent
*e
= xcalloc(sum_extents
, sizeof(*e
));
5463 unsigned long long pos
;
5464 unsigned long long start
= 0;
5465 unsigned long long maxsize
;
5466 unsigned long reserve
;
5468 /* coalesce and sort all extents. also, check to see if we need to
5469 * reserve space between member arrays
5472 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
5475 for (i
= 0; i
< dl
->extent_cnt
; i
++)
5478 qsort(e
, sum_extents
, sizeof(*e
), cmp_extent
);
5483 while (i
< sum_extents
) {
5484 e
[j
].start
= e
[i
].start
;
5485 e
[j
].size
= find_size(e
, &i
, sum_extents
);
5487 if (e
[j
-1].size
== 0)
5496 unsigned long long esize
;
5498 esize
= e
[i
].start
- pos
;
5499 if (esize
>= maxsize
) {
5504 pos
= e
[i
].start
+ e
[i
].size
;
5506 } while (e
[i
-1].size
);
5512 /* FIXME assumes volume at offset 0 is the first volume in a
5515 if (start_extent
> 0)
5516 reserve
= IMSM_RESERVED_SECTORS
; /* gap between raid regions */
5520 if (maxsize
< reserve
)
5523 super
->create_offset
= ~((unsigned long long) 0);
5524 if (start
+ reserve
> super
->create_offset
)
5525 return 0; /* start overflows create_offset */
5526 super
->create_offset
= start
+ reserve
;
5528 return maxsize
- reserve
;
5531 static int is_raid_level_supported(const struct imsm_orom
*orom
, int level
, int raiddisks
)
5533 if (level
< 0 || level
== 6 || level
== 4)
5536 /* if we have an orom prevent invalid raid levels */
5539 case 0: return imsm_orom_has_raid0(orom
);
5542 return imsm_orom_has_raid1e(orom
);
5543 return imsm_orom_has_raid1(orom
) && raiddisks
== 2;
5544 case 10: return imsm_orom_has_raid10(orom
) && raiddisks
== 4;
5545 case 5: return imsm_orom_has_raid5(orom
) && raiddisks
> 2;
5548 return 1; /* not on an Intel RAID platform so anything goes */
5554 active_arrays_by_format(char *name
, char* hba
, struct md_list
**devlist
,
5555 int dpa
, int verbose
)
5557 struct mdstat_ent
*mdstat
= mdstat_read(0, 0);
5558 struct mdstat_ent
*memb
= NULL
;
5561 struct md_list
*dv
= NULL
;
5564 for (memb
= mdstat
; memb
; memb
= memb
->next
) {
5565 if (memb
->metadata_version
&&
5566 (strncmp(memb
->metadata_version
, "external:", 9) == 0) &&
5567 (strcmp(&memb
->metadata_version
[9], name
) == 0) &&
5568 !is_subarray(memb
->metadata_version
+9) &&
5570 struct dev_member
*dev
= memb
->members
;
5572 while(dev
&& (fd
< 0)) {
5573 char *path
= xmalloc(strlen(dev
->name
) + strlen("/dev/") + 1);
5574 num
= sprintf(path
, "%s%s", "/dev/", dev
->name
);
5576 fd
= open(path
, O_RDONLY
, 0);
5577 if ((num
<= 0) || (fd
< 0)) {
5578 pr_vrb(": Cannot open %s: %s\n",
5579 dev
->name
, strerror(errno
));
5585 if ((fd
>= 0) && disk_attached_to_hba(fd
, hba
)) {
5586 struct mdstat_ent
*vol
;
5587 for (vol
= mdstat
; vol
; vol
= vol
->next
) {
5588 if ((vol
->active
> 0) &&
5589 vol
->metadata_version
&&
5590 is_container_member(vol
, memb
->dev
)) {
5595 if (*devlist
&& (found
< dpa
)) {
5596 dv
= xcalloc(1, sizeof(*dv
));
5597 dv
->devname
= xmalloc(strlen(memb
->dev
) + strlen("/dev/") + 1);
5598 sprintf(dv
->devname
, "%s%s", "/dev/", memb
->dev
);
5601 dv
->next
= *devlist
;
5609 free_mdstat(mdstat
);
5614 static struct md_list
*
5615 get_loop_devices(void)
5618 struct md_list
*devlist
= NULL
;
5619 struct md_list
*dv
= NULL
;
5621 for(i
= 0; i
< 12; i
++) {
5622 dv
= xcalloc(1, sizeof(*dv
));
5623 dv
->devname
= xmalloc(40);
5624 sprintf(dv
->devname
, "/dev/loop%d", i
);
5632 static struct md_list
*
5633 get_devices(const char *hba_path
)
5635 struct md_list
*devlist
= NULL
;
5636 struct md_list
*dv
= NULL
;
5642 devlist
= get_loop_devices();
5645 /* scroll through /sys/dev/block looking for devices attached to
5648 dir
= opendir("/sys/dev/block");
5649 for (ent
= dir
? readdir(dir
) : NULL
; ent
; ent
= readdir(dir
)) {
5654 if (sscanf(ent
->d_name
, "%d:%d", &major
, &minor
) != 2)
5656 path
= devt_to_devpath(makedev(major
, minor
));
5659 if (!path_attached_to_hba(path
, hba_path
)) {
5666 fd
= dev_open(ent
->d_name
, O_RDONLY
);
5668 fd2devname(fd
, buf
);
5671 pr_err("cannot open device: %s\n",
5676 dv
= xcalloc(1, sizeof(*dv
));
5677 dv
->devname
= xstrdup(buf
);
5684 devlist
= devlist
->next
;
5694 count_volumes_list(struct md_list
*devlist
, char *homehost
,
5695 int verbose
, int *found
)
5697 struct md_list
*tmpdev
;
5699 struct supertype
*st
= NULL
;
5701 /* first walk the list of devices to find a consistent set
5702 * that match the criterea, if that is possible.
5703 * We flag the ones we like with 'used'.
5706 st
= match_metadata_desc_imsm("imsm");
5708 pr_vrb(": cannot allocate memory for imsm supertype\n");
5712 for (tmpdev
= devlist
; tmpdev
; tmpdev
= tmpdev
->next
) {
5713 char *devname
= tmpdev
->devname
;
5715 struct supertype
*tst
;
5717 if (tmpdev
->used
> 1)
5719 tst
= dup_super(st
);
5721 pr_vrb(": cannot allocate memory for imsm supertype\n");
5724 tmpdev
->container
= 0;
5725 dfd
= dev_open(devname
, O_RDONLY
|O_EXCL
);
5727 dprintf("cannot open device %s: %s\n",
5728 devname
, strerror(errno
));
5730 } else if (fstat(dfd
, &stb
)< 0) {
5732 dprintf("fstat failed for %s: %s\n",
5733 devname
, strerror(errno
));
5735 } else if ((stb
.st_mode
& S_IFMT
) != S_IFBLK
) {
5736 dprintf("%s is not a block device.\n",
5739 } else if (must_be_container(dfd
)) {
5740 struct supertype
*cst
;
5741 cst
= super_by_fd(dfd
, NULL
);
5743 dprintf("cannot recognize container type %s\n",
5746 } else if (tst
->ss
!= st
->ss
) {
5747 dprintf("non-imsm container - ignore it: %s\n",
5750 } else if (!tst
->ss
->load_container
||
5751 tst
->ss
->load_container(tst
, dfd
, NULL
))
5754 tmpdev
->container
= 1;
5757 cst
->ss
->free_super(cst
);
5759 tmpdev
->st_rdev
= stb
.st_rdev
;
5760 if (tst
->ss
->load_super(tst
,dfd
, NULL
)) {
5761 dprintf("no RAID superblock on %s\n",
5764 } else if (tst
->ss
->compare_super
== NULL
) {
5765 dprintf("Cannot assemble %s metadata on %s\n",
5766 tst
->ss
->name
, devname
);
5772 if (tmpdev
->used
== 2 || tmpdev
->used
== 4) {
5773 /* Ignore unrecognised devices during auto-assembly */
5778 tst
->ss
->getinfo_super(tst
, &info
, NULL
);
5780 if (st
->minor_version
== -1)
5781 st
->minor_version
= tst
->minor_version
;
5783 if (memcmp(info
.uuid
, uuid_zero
,
5784 sizeof(int[4])) == 0) {
5785 /* this is a floating spare. It cannot define
5786 * an array unless there are no more arrays of
5787 * this type to be found. It can be included
5788 * in an array of this type though.
5794 if (st
->ss
!= tst
->ss
||
5795 st
->minor_version
!= tst
->minor_version
||
5796 st
->ss
->compare_super(st
, tst
) != 0) {
5797 /* Some mismatch. If exactly one array matches this host,
5798 * we can resolve on that one.
5799 * Or, if we are auto assembling, we just ignore the second
5802 dprintf("superblock on %s doesn't match others - assembly aborted\n",
5808 dprintf("found: devname: %s\n", devname
);
5812 tst
->ss
->free_super(tst
);
5816 if ((err
= load_super_imsm_all(st
, -1, &st
->sb
, NULL
, devlist
, 0)) == 0) {
5817 struct mdinfo
*iter
, *head
= st
->ss
->container_content(st
, NULL
);
5818 for (iter
= head
; iter
; iter
= iter
->next
) {
5819 dprintf("content->text_version: %s vol\n",
5820 iter
->text_version
);
5821 if (iter
->array
.state
& (1<<MD_SB_BLOCK_VOLUME
)) {
5822 /* do not assemble arrays with unsupported
5824 dprintf("Cannot activate member %s.\n",
5825 iter
->text_version
);
5832 dprintf("No valid super block on device list: err: %d %p\n",
5836 dprintf("no more devices to examine\n");
5839 for (tmpdev
= devlist
; tmpdev
; tmpdev
= tmpdev
->next
) {
5840 if ((tmpdev
->used
== 1) && (tmpdev
->found
)) {
5842 if (count
< tmpdev
->found
)
5845 count
-= tmpdev
->found
;
5848 if (tmpdev
->used
== 1)
5853 st
->ss
->free_super(st
);
5858 count_volumes(char *hba
, int dpa
, int verbose
)
5860 struct md_list
*devlist
= NULL
;
5864 devlist
= get_devices(hba
);
5865 /* if no intel devices return zero volumes */
5866 if (devlist
== NULL
)
5869 count
= active_arrays_by_format("imsm", hba
, &devlist
, dpa
, verbose
);
5870 dprintf("path: %s active arrays: %d\n", hba
, count
);
5871 if (devlist
== NULL
)
5875 count
+= count_volumes_list(devlist
,
5879 dprintf("found %d count: %d\n", found
, count
);
5882 dprintf("path: %s total number of volumes: %d\n", hba
, count
);
5885 struct md_list
*dv
= devlist
;
5886 devlist
= devlist
->next
;
5893 static int imsm_default_chunk(const struct imsm_orom
*orom
)
5895 /* up to 512 if the plaform supports it, otherwise the platform max.
5896 * 128 if no platform detected
5898 int fs
= max(7, orom
? fls(orom
->sss
) : 0);
5900 return min(512, (1 << fs
));
5904 validate_geometry_imsm_orom(struct intel_super
*super
, int level
, int layout
,
5905 int raiddisks
, int *chunk
, unsigned long long size
, int verbose
)
5907 /* check/set platform and metadata limits/defaults */
5908 if (super
->orom
&& raiddisks
> super
->orom
->dpa
) {
5909 pr_vrb(": platform supports a maximum of %d disks per array\n",
5914 /* capabilities of OROM tested - copied from validate_geometry_imsm_volume */
5915 if (!is_raid_level_supported(super
->orom
, level
, raiddisks
)) {
5916 pr_vrb(": platform does not support raid%d with %d disk%s\n",
5917 level
, raiddisks
, raiddisks
> 1 ? "s" : "");
5921 if (*chunk
== 0 || *chunk
== UnSet
)
5922 *chunk
= imsm_default_chunk(super
->orom
);
5924 if (super
->orom
&& !imsm_orom_has_chunk(super
->orom
, *chunk
)) {
5925 pr_vrb(": platform does not support a chunk size of: %d\n", *chunk
);
5929 if (layout
!= imsm_level_to_layout(level
)) {
5931 pr_vrb(": imsm raid 5 only supports the left-asymmetric layout\n");
5932 else if (level
== 10)
5933 pr_vrb(": imsm raid 10 only supports the n2 layout\n");
5935 pr_vrb(": imsm unknown layout %#x for this raid level %d\n",
5940 if (super
->orom
&& (super
->orom
->attr
& IMSM_OROM_ATTR_2TB
) == 0 &&
5941 (calc_array_size(level
, raiddisks
, layout
, *chunk
, size
) >> 32) > 0) {
5942 pr_vrb(": platform does not support a volume size over 2TB\n");
5949 /* validate_geometry_imsm_volume - lifted from validate_geometry_ddf_bvd
5950 * FIX ME add ahci details
5952 static int validate_geometry_imsm_volume(struct supertype
*st
, int level
,
5953 int layout
, int raiddisks
, int *chunk
,
5954 unsigned long long size
,
5955 unsigned long long data_offset
,
5957 unsigned long long *freesize
,
5961 struct intel_super
*super
= st
->sb
;
5962 struct imsm_super
*mpb
;
5964 unsigned long long pos
= 0;
5965 unsigned long long maxsize
;
5969 /* We must have the container info already read in. */
5973 mpb
= super
->anchor
;
5975 if (!validate_geometry_imsm_orom(super
, level
, layout
, raiddisks
, chunk
, size
, verbose
)) {
5976 pr_err("RAID gemetry validation failed. Cannot proceed with the action(s).\n");
5980 /* General test: make sure there is space for
5981 * 'raiddisks' device extents of size 'size' at a given
5984 unsigned long long minsize
= size
;
5985 unsigned long long start_offset
= MaxSector
;
5988 minsize
= MPB_SECTOR_CNT
+ IMSM_RESERVED_SECTORS
;
5989 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
5994 e
= get_extents(super
, dl
);
5997 unsigned long long esize
;
5998 esize
= e
[i
].start
- pos
;
5999 if (esize
>= minsize
)
6001 if (found
&& start_offset
== MaxSector
) {
6004 } else if (found
&& pos
!= start_offset
) {
6008 pos
= e
[i
].start
+ e
[i
].size
;
6010 } while (e
[i
-1].size
);
6015 if (dcnt
< raiddisks
) {
6017 pr_err("imsm: Not enough devices with space for this array (%d < %d)\n",
6024 /* This device must be a member of the set */
6025 if (stat(dev
, &stb
) < 0)
6027 if ((S_IFMT
& stb
.st_mode
) != S_IFBLK
)
6029 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
6030 if (dl
->major
== (int)major(stb
.st_rdev
) &&
6031 dl
->minor
== (int)minor(stb
.st_rdev
))
6036 pr_err("%s is not in the same imsm set\n", dev
);
6038 } else if (super
->orom
&& dl
->index
< 0 && mpb
->num_raid_devs
) {
6039 /* If a volume is present then the current creation attempt
6040 * cannot incorporate new spares because the orom may not
6041 * understand this configuration (all member disks must be
6042 * members of each array in the container).
6044 pr_err("%s is a spare and a volume is already defined for this container\n", dev
);
6045 pr_err("The option-rom requires all member disks to be a member of all volumes\n");
6047 } else if (super
->orom
&& mpb
->num_raid_devs
> 0 &&
6048 mpb
->num_disks
!= raiddisks
) {
6049 pr_err("The option-rom requires all member disks to be a member of all volumes\n");
6053 /* retrieve the largest free space block */
6054 e
= get_extents(super
, dl
);
6059 unsigned long long esize
;
6061 esize
= e
[i
].start
- pos
;
6062 if (esize
>= maxsize
)
6064 pos
= e
[i
].start
+ e
[i
].size
;
6066 } while (e
[i
-1].size
);
6071 pr_err("unable to determine free space for: %s\n",
6075 if (maxsize
< size
) {
6077 pr_err("%s not enough space (%llu < %llu)\n",
6078 dev
, maxsize
, size
);
6082 /* count total number of extents for merge */
6084 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
6086 i
+= dl
->extent_cnt
;
6088 maxsize
= merge_extents(super
, i
);
6090 if (!check_env("IMSM_NO_PLATFORM") &&
6091 mpb
->num_raid_devs
> 0 && size
&& size
!= maxsize
) {
6092 pr_err("attempting to create a second volume with size less then remaining space. Aborting...\n");
6096 if (maxsize
< size
|| maxsize
== 0) {
6099 pr_err("no free space left on device. Aborting...\n");
6101 pr_err("not enough space to create volume of given size (%llu < %llu). Aborting...\n",
6107 *freesize
= maxsize
;
6110 int count
= count_volumes(super
->hba
->path
,
6111 super
->orom
->dpa
, verbose
);
6112 if (super
->orom
->vphba
<= count
) {
6113 pr_vrb(": platform does not support more than %d raid volumes.\n",
6114 super
->orom
->vphba
);
6121 static int imsm_get_free_size(struct supertype
*st
, int raiddisks
,
6122 unsigned long long size
, int chunk
,
6123 unsigned long long *freesize
)
6125 struct intel_super
*super
= st
->sb
;
6126 struct imsm_super
*mpb
= super
->anchor
;
6131 unsigned long long maxsize
;
6132 unsigned long long minsize
;
6136 /* find the largest common start free region of the possible disks */
6140 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
6146 /* don't activate new spares if we are orom constrained
6147 * and there is already a volume active in the container
6149 if (super
->orom
&& dl
->index
< 0 && mpb
->num_raid_devs
)
6152 e
= get_extents(super
, dl
);
6155 for (i
= 1; e
[i
-1].size
; i
++)
6163 maxsize
= merge_extents(super
, extent_cnt
);
6167 minsize
= chunk
* 2;
6169 if (cnt
< raiddisks
||
6170 (super
->orom
&& used
&& used
!= raiddisks
) ||
6171 maxsize
< minsize
||
6173 pr_err("not enough devices with space to create array.\n");
6174 return 0; /* No enough free spaces large enough */
6185 if (!check_env("IMSM_NO_PLATFORM") &&
6186 mpb
->num_raid_devs
> 0 && size
&& size
!= maxsize
) {
6187 pr_err("attempting to create a second volume with size less then remaining space. Aborting...\n");
6191 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
6193 dl
->raiddisk
= cnt
++;
6197 dprintf("imsm: imsm_get_free_size() returns : %llu\n", size
);
6202 static int reserve_space(struct supertype
*st
, int raiddisks
,
6203 unsigned long long size
, int chunk
,
6204 unsigned long long *freesize
)
6206 struct intel_super
*super
= st
->sb
;
6211 rv
= imsm_get_free_size(st
, raiddisks
, size
, chunk
, freesize
);
6214 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
6216 dl
->raiddisk
= cnt
++;
6223 static int validate_geometry_imsm(struct supertype
*st
, int level
, int layout
,
6224 int raiddisks
, int *chunk
, unsigned long long size
,
6225 unsigned long long data_offset
,
6226 char *dev
, unsigned long long *freesize
,
6234 * if given unused devices create a container
6235 * if given given devices in a container create a member volume
6237 if (level
== LEVEL_CONTAINER
) {
6238 /* Must be a fresh device to add to a container */
6239 return validate_geometry_imsm_container(st
, level
, layout
,
6249 struct intel_super
*super
= st
->sb
;
6250 if (!validate_geometry_imsm_orom(st
->sb
, level
, layout
,
6251 raiddisks
, chunk
, size
,
6254 /* we are being asked to automatically layout a
6255 * new volume based on the current contents of
6256 * the container. If the the parameters can be
6257 * satisfied reserve_space will record the disks,
6258 * start offset, and size of the volume to be
6259 * created. add_to_super and getinfo_super
6260 * detect when autolayout is in progress.
6262 /* assuming that freesize is always given when array is
6264 if (super
->orom
&& freesize
) {
6266 count
= count_volumes(super
->hba
->path
,
6267 super
->orom
->dpa
, verbose
);
6268 if (super
->orom
->vphba
<= count
) {
6269 pr_vrb(": platform does not support more than %d raid volumes.\n",
6270 super
->orom
->vphba
);
6275 return reserve_space(st
, raiddisks
, size
,
6281 /* creating in a given container */
6282 return validate_geometry_imsm_volume(st
, level
, layout
,
6283 raiddisks
, chunk
, size
,
6285 dev
, freesize
, verbose
);
6288 /* This device needs to be a device in an 'imsm' container */
6289 fd
= open(dev
, O_RDONLY
|O_EXCL
, 0);
6292 pr_err("Cannot create this array on device %s\n",
6297 if (errno
!= EBUSY
|| (fd
= open(dev
, O_RDONLY
, 0)) < 0) {
6299 pr_err("Cannot open %s: %s\n",
6300 dev
, strerror(errno
));
6303 /* Well, it is in use by someone, maybe an 'imsm' container. */
6304 cfd
= open_container(fd
);
6308 pr_err("Cannot use %s: It is busy\n",
6312 sra
= sysfs_read(cfd
, NULL
, GET_VERSION
);
6313 if (sra
&& sra
->array
.major_version
== -1 &&
6314 strcmp(sra
->text_version
, "imsm") == 0)
6318 /* This is a member of a imsm container. Load the container
6319 * and try to create a volume
6321 struct intel_super
*super
;
6323 if (load_super_imsm_all(st
, cfd
, (void **) &super
, NULL
, NULL
, 1) == 0) {
6325 strcpy(st
->container_devnm
, fd2devnm(cfd
));
6327 return validate_geometry_imsm_volume(st
, level
, layout
,
6329 size
, data_offset
, dev
,
6336 pr_err("failed container membership check\n");
6342 static void default_geometry_imsm(struct supertype
*st
, int *level
, int *layout
, int *chunk
)
6344 struct intel_super
*super
= st
->sb
;
6346 if (level
&& *level
== UnSet
)
6347 *level
= LEVEL_CONTAINER
;
6349 if (level
&& layout
&& *layout
== UnSet
)
6350 *layout
= imsm_level_to_layout(*level
);
6352 if (chunk
&& (*chunk
== UnSet
|| *chunk
== 0))
6353 *chunk
= imsm_default_chunk(super
->orom
);
6356 static void handle_missing(struct intel_super
*super
, struct imsm_dev
*dev
);
6358 static int kill_subarray_imsm(struct supertype
*st
)
6360 /* remove the subarray currently referenced by ->current_vol */
6362 struct intel_dev
**dp
;
6363 struct intel_super
*super
= st
->sb
;
6364 __u8 current_vol
= super
->current_vol
;
6365 struct imsm_super
*mpb
= super
->anchor
;
6367 if (super
->current_vol
< 0)
6369 super
->current_vol
= -1; /* invalidate subarray cursor */
6371 /* block deletions that would change the uuid of active subarrays
6373 * FIXME when immutable ids are available, but note that we'll
6374 * also need to fixup the invalidated/active subarray indexes in
6377 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
6380 if (i
< current_vol
)
6382 sprintf(subarray
, "%u", i
);
6383 if (is_subarray_active(subarray
, st
->devnm
)) {
6384 pr_err("deleting subarray-%d would change the UUID of active subarray-%d, aborting\n",
6391 if (st
->update_tail
) {
6392 struct imsm_update_kill_array
*u
= xmalloc(sizeof(*u
));
6394 u
->type
= update_kill_array
;
6395 u
->dev_idx
= current_vol
;
6396 append_metadata_update(st
, u
, sizeof(*u
));
6401 for (dp
= &super
->devlist
; *dp
;)
6402 if ((*dp
)->index
== current_vol
) {
6405 handle_missing(super
, (*dp
)->dev
);
6406 if ((*dp
)->index
> current_vol
)
6411 /* no more raid devices, all active components are now spares,
6412 * but of course failed are still failed
6414 if (--mpb
->num_raid_devs
== 0) {
6417 for (d
= super
->disks
; d
; d
= d
->next
)
6422 super
->updates_pending
++;
6427 static int update_subarray_imsm(struct supertype
*st
, char *subarray
,
6428 char *update
, struct mddev_ident
*ident
)
6430 /* update the subarray currently referenced by ->current_vol */
6431 struct intel_super
*super
= st
->sb
;
6432 struct imsm_super
*mpb
= super
->anchor
;
6434 if (strcmp(update
, "name") == 0) {
6435 char *name
= ident
->name
;
6439 if (is_subarray_active(subarray
, st
->devnm
)) {
6440 pr_err("Unable to update name of active subarray\n");
6444 if (!check_name(super
, name
, 0))
6447 vol
= strtoul(subarray
, &ep
, 10);
6448 if (*ep
!= '\0' || vol
>= super
->anchor
->num_raid_devs
)
6451 if (st
->update_tail
) {
6452 struct imsm_update_rename_array
*u
= xmalloc(sizeof(*u
));
6454 u
->type
= update_rename_array
;
6456 snprintf((char *) u
->name
, MAX_RAID_SERIAL_LEN
, "%s", name
);
6457 append_metadata_update(st
, u
, sizeof(*u
));
6459 struct imsm_dev
*dev
;
6462 dev
= get_imsm_dev(super
, vol
);
6463 snprintf((char *) dev
->volume
, MAX_RAID_SERIAL_LEN
, "%s", name
);
6464 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
6465 dev
= get_imsm_dev(super
, i
);
6466 handle_missing(super
, dev
);
6468 super
->updates_pending
++;
6475 #endif /* MDASSEMBLE */
6477 static int is_gen_migration(struct imsm_dev
*dev
)
6482 if (!dev
->vol
.migr_state
)
6485 if (migr_type(dev
) == MIGR_GEN_MIGR
)
6491 static int is_rebuilding(struct imsm_dev
*dev
)
6493 struct imsm_map
*migr_map
;
6495 if (!dev
->vol
.migr_state
)
6498 if (migr_type(dev
) != MIGR_REBUILD
)
6501 migr_map
= get_imsm_map(dev
, MAP_1
);
6503 if (migr_map
->map_state
== IMSM_T_STATE_DEGRADED
)
6510 static int is_initializing(struct imsm_dev
*dev
)
6512 struct imsm_map
*migr_map
;
6514 if (!dev
->vol
.migr_state
)
6517 if (migr_type(dev
) != MIGR_INIT
)
6520 migr_map
= get_imsm_map(dev
, MAP_1
);
6522 if (migr_map
->map_state
== IMSM_T_STATE_UNINITIALIZED
)
6529 static void update_recovery_start(struct intel_super
*super
,
6530 struct imsm_dev
*dev
,
6531 struct mdinfo
*array
)
6533 struct mdinfo
*rebuild
= NULL
;
6537 if (!is_rebuilding(dev
))
6540 /* Find the rebuild target, but punt on the dual rebuild case */
6541 for (d
= array
->devs
; d
; d
= d
->next
)
6542 if (d
->recovery_start
== 0) {
6549 /* (?) none of the disks are marked with
6550 * IMSM_ORD_REBUILD, so assume they are missing and the
6551 * disk_ord_tbl was not correctly updated
6553 dprintf("failed to locate out-of-sync disk\n");
6557 units
= __le32_to_cpu(dev
->vol
.curr_migr_unit
);
6558 rebuild
->recovery_start
= units
* blocks_per_migr_unit(super
, dev
);
6562 static int recover_backup_imsm(struct supertype
*st
, struct mdinfo
*info
);
6565 static struct mdinfo
*container_content_imsm(struct supertype
*st
, char *subarray
)
6567 /* Given a container loaded by load_super_imsm_all,
6568 * extract information about all the arrays into
6570 * If 'subarray' is given, just extract info about that array.
6572 * For each imsm_dev create an mdinfo, fill it in,
6573 * then look for matching devices in super->disks
6574 * and create appropriate device mdinfo.
6576 struct intel_super
*super
= st
->sb
;
6577 struct imsm_super
*mpb
= super
->anchor
;
6578 struct mdinfo
*rest
= NULL
;
6582 int spare_disks
= 0;
6584 /* do not assemble arrays when not all attributes are supported */
6585 if (imsm_check_attributes(mpb
->attributes
) == 0) {
6587 pr_err("Unsupported attributes in IMSM metadata.Arrays activation is blocked.\n");
6590 /* check for bad blocks */
6591 if (imsm_bbm_log_size(super
->anchor
)) {
6592 pr_err("BBM log found in IMSM metadata.Arrays activation is blocked.\n");
6596 /* count spare devices, not used in maps
6598 for (d
= super
->disks
; d
; d
= d
->next
)
6602 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
6603 struct imsm_dev
*dev
;
6604 struct imsm_map
*map
;
6605 struct imsm_map
*map2
;
6606 struct mdinfo
*this;
6614 (i
!= strtoul(subarray
, &ep
, 10) || *ep
!= '\0'))
6617 dev
= get_imsm_dev(super
, i
);
6618 map
= get_imsm_map(dev
, MAP_0
);
6619 map2
= get_imsm_map(dev
, MAP_1
);
6621 /* do not publish arrays that are in the middle of an
6622 * unsupported migration
6624 if (dev
->vol
.migr_state
&&
6625 (migr_type(dev
) == MIGR_STATE_CHANGE
)) {
6626 pr_err("cannot assemble volume '%.16s': unsupported migration in progress\n",
6630 /* do not publish arrays that are not support by controller's
6634 this = xmalloc(sizeof(*this));
6636 super
->current_vol
= i
;
6637 getinfo_super_imsm_volume(st
, this, NULL
);
6640 chunk
= __le16_to_cpu(map
->blocks_per_strip
) >> 1;
6641 /* mdadm does not support all metadata features- set the bit in all arrays state */
6642 if (!validate_geometry_imsm_orom(super
,
6643 get_imsm_raid_level(map
), /* RAID level */
6644 imsm_level_to_layout(get_imsm_raid_level(map
)),
6645 map
->num_members
, /* raid disks */
6646 &chunk
, join_u32(dev
->size_low
, dev
->size_high
),
6648 pr_err("IMSM RAID geometry validation failed. Array %s activation is blocked.\n",
6650 this->array
.state
|=
6651 (1<<MD_SB_BLOCK_CONTAINER_RESHAPE
) |
6652 (1<<MD_SB_BLOCK_VOLUME
);
6656 /* if array has bad blocks, set suitable bit in all arrays state */
6658 this->array
.state
|=
6659 (1<<MD_SB_BLOCK_CONTAINER_RESHAPE
) |
6660 (1<<MD_SB_BLOCK_VOLUME
);
6662 for (slot
= 0 ; slot
< map
->num_members
; slot
++) {
6663 unsigned long long recovery_start
;
6664 struct mdinfo
*info_d
;
6671 idx
= get_imsm_disk_idx(dev
, slot
, MAP_0
);
6672 ord
= get_imsm_ord_tbl_ent(dev
, slot
, MAP_X
);
6673 for (d
= super
->disks
; d
; d
= d
->next
)
6674 if (d
->index
== idx
)
6677 recovery_start
= MaxSector
;
6680 if (d
&& is_failed(&d
->disk
))
6682 if (ord
& IMSM_ORD_REBUILD
)
6686 * if we skip some disks the array will be assmebled degraded;
6687 * reset resync start to avoid a dirty-degraded
6688 * situation when performing the intial sync
6690 * FIXME handle dirty degraded
6692 if ((skip
|| recovery_start
== 0) && !dev
->vol
.dirty
)
6693 this->resync_start
= MaxSector
;
6697 info_d
= xcalloc(1, sizeof(*info_d
));
6698 info_d
->next
= this->devs
;
6699 this->devs
= info_d
;
6701 info_d
->disk
.number
= d
->index
;
6702 info_d
->disk
.major
= d
->major
;
6703 info_d
->disk
.minor
= d
->minor
;
6704 info_d
->disk
.raid_disk
= slot
;
6705 info_d
->recovery_start
= recovery_start
;
6707 if (slot
< map2
->num_members
)
6708 info_d
->disk
.state
= (1 << MD_DISK_ACTIVE
);
6710 this->array
.spare_disks
++;
6712 if (slot
< map
->num_members
)
6713 info_d
->disk
.state
= (1 << MD_DISK_ACTIVE
);
6715 this->array
.spare_disks
++;
6717 if (info_d
->recovery_start
== MaxSector
)
6718 this->array
.working_disks
++;
6720 info_d
->events
= __le32_to_cpu(mpb
->generation_num
);
6721 info_d
->data_offset
= pba_of_lba0(map
);
6722 info_d
->component_size
= blocks_per_member(map
);
6724 /* now that the disk list is up-to-date fixup recovery_start */
6725 update_recovery_start(super
, dev
, this);
6726 this->array
.spare_disks
+= spare_disks
;
6729 /* check for reshape */
6730 if (this->reshape_active
== 1)
6731 recover_backup_imsm(st
, this);
6739 static __u8
imsm_check_degraded(struct intel_super
*super
, struct imsm_dev
*dev
,
6740 int failed
, int look_in_map
)
6742 struct imsm_map
*map
;
6744 map
= get_imsm_map(dev
, look_in_map
);
6747 return map
->map_state
== IMSM_T_STATE_UNINITIALIZED
?
6748 IMSM_T_STATE_UNINITIALIZED
: IMSM_T_STATE_NORMAL
;
6750 switch (get_imsm_raid_level(map
)) {
6752 return IMSM_T_STATE_FAILED
;
6755 if (failed
< map
->num_members
)
6756 return IMSM_T_STATE_DEGRADED
;
6758 return IMSM_T_STATE_FAILED
;
6763 * check to see if any mirrors have failed, otherwise we
6764 * are degraded. Even numbered slots are mirrored on
6768 /* gcc -Os complains that this is unused */
6769 int insync
= insync
;
6771 for (i
= 0; i
< map
->num_members
; i
++) {
6772 __u32 ord
= get_imsm_ord_tbl_ent(dev
, i
, MAP_X
);
6773 int idx
= ord_to_idx(ord
);
6774 struct imsm_disk
*disk
;
6776 /* reset the potential in-sync count on even-numbered
6777 * slots. num_copies is always 2 for imsm raid10
6782 disk
= get_imsm_disk(super
, idx
);
6783 if (!disk
|| is_failed(disk
) || ord
& IMSM_ORD_REBUILD
)
6786 /* no in-sync disks left in this mirror the
6790 return IMSM_T_STATE_FAILED
;
6793 return IMSM_T_STATE_DEGRADED
;
6797 return IMSM_T_STATE_DEGRADED
;
6799 return IMSM_T_STATE_FAILED
;
6805 return map
->map_state
;
6808 static int imsm_count_failed(struct intel_super
*super
, struct imsm_dev
*dev
,
6813 struct imsm_disk
*disk
;
6814 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
6815 struct imsm_map
*prev
= get_imsm_map(dev
, MAP_1
);
6816 struct imsm_map
*map_for_loop
;
6821 /* at the beginning of migration we set IMSM_ORD_REBUILD on
6822 * disks that are being rebuilt. New failures are recorded to
6823 * map[0]. So we look through all the disks we started with and
6824 * see if any failures are still present, or if any new ones
6828 if (prev
&& (map
->num_members
< prev
->num_members
))
6829 map_for_loop
= prev
;
6831 for (i
= 0; i
< map_for_loop
->num_members
; i
++) {
6833 /* when MAP_X is passed both maps failures are counted
6836 ((look_in_map
== MAP_1
) || (look_in_map
== MAP_X
)) &&
6837 (i
< prev
->num_members
)) {
6838 ord
= __le32_to_cpu(prev
->disk_ord_tbl
[i
]);
6839 idx_1
= ord_to_idx(ord
);
6841 disk
= get_imsm_disk(super
, idx_1
);
6842 if (!disk
|| is_failed(disk
) || ord
& IMSM_ORD_REBUILD
)
6845 if (((look_in_map
== MAP_0
) || (look_in_map
== MAP_X
)) &&
6846 (i
< map
->num_members
)) {
6847 ord
= __le32_to_cpu(map
->disk_ord_tbl
[i
]);
6848 idx
= ord_to_idx(ord
);
6851 disk
= get_imsm_disk(super
, idx
);
6852 if (!disk
|| is_failed(disk
) ||
6853 ord
& IMSM_ORD_REBUILD
)
6863 static int imsm_open_new(struct supertype
*c
, struct active_array
*a
,
6866 struct intel_super
*super
= c
->sb
;
6867 struct imsm_super
*mpb
= super
->anchor
;
6869 if (atoi(inst
) >= mpb
->num_raid_devs
) {
6870 pr_err("subarry index %d, out of range\n", atoi(inst
));
6874 dprintf("imsm: open_new %s\n", inst
);
6875 a
->info
.container_member
= atoi(inst
);
6879 static int is_resyncing(struct imsm_dev
*dev
)
6881 struct imsm_map
*migr_map
;
6883 if (!dev
->vol
.migr_state
)
6886 if (migr_type(dev
) == MIGR_INIT
||
6887 migr_type(dev
) == MIGR_REPAIR
)
6890 if (migr_type(dev
) == MIGR_GEN_MIGR
)
6893 migr_map
= get_imsm_map(dev
, MAP_1
);
6895 if ((migr_map
->map_state
== IMSM_T_STATE_NORMAL
) &&
6896 (dev
->vol
.migr_type
!= MIGR_GEN_MIGR
))
6902 /* return true if we recorded new information */
6903 static int mark_failure(struct imsm_dev
*dev
, struct imsm_disk
*disk
, int idx
)
6907 struct imsm_map
*map
;
6908 char buf
[MAX_RAID_SERIAL_LEN
+3];
6909 unsigned int len
, shift
= 0;
6911 /* new failures are always set in map[0] */
6912 map
= get_imsm_map(dev
, MAP_0
);
6914 slot
= get_imsm_disk_slot(map
, idx
);
6918 ord
= __le32_to_cpu(map
->disk_ord_tbl
[slot
]);
6919 if (is_failed(disk
) && (ord
& IMSM_ORD_REBUILD
))
6922 memcpy(buf
, disk
->serial
, MAX_RAID_SERIAL_LEN
);
6923 buf
[MAX_RAID_SERIAL_LEN
] = '\000';
6925 if ((len
= strlen(buf
)) >= MAX_RAID_SERIAL_LEN
)
6926 shift
= len
- MAX_RAID_SERIAL_LEN
+ 1;
6927 strncpy((char *)disk
->serial
, &buf
[shift
], MAX_RAID_SERIAL_LEN
);
6929 disk
->status
|= FAILED_DISK
;
6930 set_imsm_ord_tbl_ent(map
, slot
, idx
| IMSM_ORD_REBUILD
);
6931 /* mark failures in second map if second map exists and this disk
6933 * This is valid for migration, initialization and rebuild
6935 if (dev
->vol
.migr_state
) {
6936 struct imsm_map
*map2
= get_imsm_map(dev
, MAP_1
);
6937 int slot2
= get_imsm_disk_slot(map2
, idx
);
6939 if ((slot2
< map2
->num_members
) &&
6941 set_imsm_ord_tbl_ent(map2
, slot2
,
6942 idx
| IMSM_ORD_REBUILD
);
6944 if (map
->failed_disk_num
== 0xff)
6945 map
->failed_disk_num
= slot
;
6949 static void mark_missing(struct imsm_dev
*dev
, struct imsm_disk
*disk
, int idx
)
6951 mark_failure(dev
, disk
, idx
);
6953 if (disk
->scsi_id
== __cpu_to_le32(~(__u32
)0))
6956 disk
->scsi_id
= __cpu_to_le32(~(__u32
)0);
6957 memmove(&disk
->serial
[0], &disk
->serial
[1], MAX_RAID_SERIAL_LEN
- 1);
6960 static void handle_missing(struct intel_super
*super
, struct imsm_dev
*dev
)
6964 if (!super
->missing
)
6967 /* When orom adds replacement for missing disk it does
6968 * not remove entry of missing disk, but just updates map with
6969 * new added disk. So it is not enough just to test if there is
6970 * any missing disk, we have to look if there are any failed disks
6971 * in map to stop migration */
6973 dprintf("imsm: mark missing\n");
6974 /* end process for initialization and rebuild only
6976 if (is_gen_migration(dev
) == 0) {
6980 failed
= imsm_count_failed(super
, dev
, MAP_0
);
6981 map_state
= imsm_check_degraded(super
, dev
, failed
, MAP_0
);
6984 end_migration(dev
, super
, map_state
);
6986 for (dl
= super
->missing
; dl
; dl
= dl
->next
)
6987 mark_missing(dev
, &dl
->disk
, dl
->index
);
6988 super
->updates_pending
++;
6991 static unsigned long long imsm_set_array_size(struct imsm_dev
*dev
,
6994 int used_disks
= imsm_num_data_members(dev
, MAP_0
);
6995 unsigned long long array_blocks
;
6996 struct imsm_map
*map
;
6998 if (used_disks
== 0) {
6999 /* when problems occures
7000 * return current array_blocks value
7002 array_blocks
= __le32_to_cpu(dev
->size_high
);
7003 array_blocks
= array_blocks
<< 32;
7004 array_blocks
+= __le32_to_cpu(dev
->size_low
);
7006 return array_blocks
;
7009 /* set array size in metadata
7011 if (new_size
<= 0) {
7012 /* OLCE size change is caused by added disks
7014 map
= get_imsm_map(dev
, MAP_0
);
7015 array_blocks
= blocks_per_member(map
) * used_disks
;
7017 /* Online Volume Size Change
7018 * Using available free space
7020 array_blocks
= new_size
;
7023 /* round array size down to closest MB
7025 array_blocks
= (array_blocks
>> SECT_PER_MB_SHIFT
) << SECT_PER_MB_SHIFT
;
7026 dev
->size_low
= __cpu_to_le32((__u32
)array_blocks
);
7027 dev
->size_high
= __cpu_to_le32((__u32
)(array_blocks
>> 32));
7029 return array_blocks
;
7032 static void imsm_set_disk(struct active_array
*a
, int n
, int state
);
7034 static void imsm_progress_container_reshape(struct intel_super
*super
)
7036 /* if no device has a migr_state, but some device has a
7037 * different number of members than the previous device, start
7038 * changing the number of devices in this device to match
7041 struct imsm_super
*mpb
= super
->anchor
;
7042 int prev_disks
= -1;
7046 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
7047 struct imsm_dev
*dev
= get_imsm_dev(super
, i
);
7048 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
7049 struct imsm_map
*map2
;
7050 int prev_num_members
;
7052 if (dev
->vol
.migr_state
)
7055 if (prev_disks
== -1)
7056 prev_disks
= map
->num_members
;
7057 if (prev_disks
== map
->num_members
)
7060 /* OK, this array needs to enter reshape mode.
7061 * i.e it needs a migr_state
7064 copy_map_size
= sizeof_imsm_map(map
);
7065 prev_num_members
= map
->num_members
;
7066 map
->num_members
= prev_disks
;
7067 dev
->vol
.migr_state
= 1;
7068 dev
->vol
.curr_migr_unit
= 0;
7069 set_migr_type(dev
, MIGR_GEN_MIGR
);
7070 for (i
= prev_num_members
;
7071 i
< map
->num_members
; i
++)
7072 set_imsm_ord_tbl_ent(map
, i
, i
);
7073 map2
= get_imsm_map(dev
, MAP_1
);
7074 /* Copy the current map */
7075 memcpy(map2
, map
, copy_map_size
);
7076 map2
->num_members
= prev_num_members
;
7078 imsm_set_array_size(dev
, -1);
7079 super
->clean_migration_record_by_mdmon
= 1;
7080 super
->updates_pending
++;
7084 /* Handle dirty -> clean transititions, resync and reshape. Degraded and rebuild
7085 * states are handled in imsm_set_disk() with one exception, when a
7086 * resync is stopped due to a new failure this routine will set the
7087 * 'degraded' state for the array.
7089 static int imsm_set_array_state(struct active_array
*a
, int consistent
)
7091 int inst
= a
->info
.container_member
;
7092 struct intel_super
*super
= a
->container
->sb
;
7093 struct imsm_dev
*dev
= get_imsm_dev(super
, inst
);
7094 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
7095 int failed
= imsm_count_failed(super
, dev
, MAP_0
);
7096 __u8 map_state
= imsm_check_degraded(super
, dev
, failed
, MAP_0
);
7097 __u32 blocks_per_unit
;
7099 if (dev
->vol
.migr_state
&&
7100 dev
->vol
.migr_type
== MIGR_GEN_MIGR
) {
7101 /* array state change is blocked due to reshape action
7103 * - abort the reshape (if last_checkpoint is 0 and action!= reshape)
7104 * - finish the reshape (if last_checkpoint is big and action != reshape)
7105 * - update curr_migr_unit
7107 if (a
->curr_action
== reshape
) {
7108 /* still reshaping, maybe update curr_migr_unit */
7109 goto mark_checkpoint
;
7111 if (a
->last_checkpoint
== 0 && a
->prev_action
== reshape
) {
7112 /* for some reason we aborted the reshape.
7114 * disable automatic metadata rollback
7115 * user action is required to recover process
7118 struct imsm_map
*map2
=
7119 get_imsm_map(dev
, MAP_1
);
7120 dev
->vol
.migr_state
= 0;
7121 set_migr_type(dev
, 0);
7122 dev
->vol
.curr_migr_unit
= 0;
7124 sizeof_imsm_map(map2
));
7125 super
->updates_pending
++;
7128 if (a
->last_checkpoint
>= a
->info
.component_size
) {
7129 unsigned long long array_blocks
;
7133 used_disks
= imsm_num_data_members(dev
, MAP_0
);
7134 if (used_disks
> 0) {
7136 blocks_per_member(map
) *
7138 /* round array size down to closest MB
7140 array_blocks
= (array_blocks
7141 >> SECT_PER_MB_SHIFT
)
7142 << SECT_PER_MB_SHIFT
;
7143 a
->info
.custom_array_size
= array_blocks
;
7144 /* encourage manager to update array
7148 a
->check_reshape
= 1;
7150 /* finalize online capacity expansion/reshape */
7151 for (mdi
= a
->info
.devs
; mdi
; mdi
= mdi
->next
)
7153 mdi
->disk
.raid_disk
,
7156 imsm_progress_container_reshape(super
);
7161 /* before we activate this array handle any missing disks */
7162 if (consistent
== 2)
7163 handle_missing(super
, dev
);
7165 if (consistent
== 2 &&
7166 (!is_resync_complete(&a
->info
) ||
7167 map_state
!= IMSM_T_STATE_NORMAL
||
7168 dev
->vol
.migr_state
))
7171 if (is_resync_complete(&a
->info
)) {
7172 /* complete intialization / resync,
7173 * recovery and interrupted recovery is completed in
7176 if (is_resyncing(dev
)) {
7177 dprintf("imsm: mark resync done\n");
7178 end_migration(dev
, super
, map_state
);
7179 super
->updates_pending
++;
7180 a
->last_checkpoint
= 0;
7182 } else if ((!is_resyncing(dev
) && !failed
) &&
7183 (imsm_reshape_blocks_arrays_changes(super
) == 0)) {
7184 /* mark the start of the init process if nothing is failed */
7185 dprintf("imsm: mark resync start\n");
7186 if (map
->map_state
== IMSM_T_STATE_UNINITIALIZED
)
7187 migrate(dev
, super
, IMSM_T_STATE_NORMAL
, MIGR_INIT
);
7189 migrate(dev
, super
, IMSM_T_STATE_NORMAL
, MIGR_REPAIR
);
7190 super
->updates_pending
++;
7194 /* skip checkpointing for general migration,
7195 * it is controlled in mdadm
7197 if (is_gen_migration(dev
))
7198 goto skip_mark_checkpoint
;
7200 /* check if we can update curr_migr_unit from resync_start, recovery_start */
7201 blocks_per_unit
= blocks_per_migr_unit(super
, dev
);
7202 if (blocks_per_unit
) {
7206 units
= a
->last_checkpoint
/ blocks_per_unit
;
7209 /* check that we did not overflow 32-bits, and that
7210 * curr_migr_unit needs updating
7212 if (units32
== units
&&
7214 __le32_to_cpu(dev
->vol
.curr_migr_unit
) != units32
) {
7215 dprintf("imsm: mark checkpoint (%u)\n", units32
);
7216 dev
->vol
.curr_migr_unit
= __cpu_to_le32(units32
);
7217 super
->updates_pending
++;
7221 skip_mark_checkpoint
:
7222 /* mark dirty / clean */
7223 if (dev
->vol
.dirty
!= !consistent
) {
7224 dprintf("imsm: mark '%s'\n", consistent
? "clean" : "dirty");
7229 super
->updates_pending
++;
7235 static void imsm_set_disk(struct active_array
*a
, int n
, int state
)
7237 int inst
= a
->info
.container_member
;
7238 struct intel_super
*super
= a
->container
->sb
;
7239 struct imsm_dev
*dev
= get_imsm_dev(super
, inst
);
7240 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
7241 struct imsm_disk
*disk
;
7243 int recovery_not_finished
= 0;
7248 if (n
> map
->num_members
)
7249 pr_err("imsm: set_disk %d out of range 0..%d\n",
7250 n
, map
->num_members
- 1);
7255 dprintf("imsm: set_disk %d:%x\n", n
, state
);
7257 ord
= get_imsm_ord_tbl_ent(dev
, n
, MAP_0
);
7258 disk
= get_imsm_disk(super
, ord_to_idx(ord
));
7260 /* check for new failures */
7261 if (state
& DS_FAULTY
) {
7262 if (mark_failure(dev
, disk
, ord_to_idx(ord
)))
7263 super
->updates_pending
++;
7266 /* check if in_sync */
7267 if (state
& DS_INSYNC
&& ord
& IMSM_ORD_REBUILD
&& is_rebuilding(dev
)) {
7268 struct imsm_map
*migr_map
= get_imsm_map(dev
, MAP_1
);
7270 set_imsm_ord_tbl_ent(migr_map
, n
, ord_to_idx(ord
));
7271 super
->updates_pending
++;
7274 failed
= imsm_count_failed(super
, dev
, MAP_0
);
7275 map_state
= imsm_check_degraded(super
, dev
, failed
, MAP_0
);
7277 /* check if recovery complete, newly degraded, or failed */
7278 dprintf("imsm: Detected transition to state ");
7279 switch (map_state
) {
7280 case IMSM_T_STATE_NORMAL
: /* transition to normal state */
7281 dprintf("normal: ");
7282 if (is_rebuilding(dev
)) {
7283 dprintf_cont("while rebuilding");
7284 /* check if recovery is really finished */
7285 for (mdi
= a
->info
.devs
; mdi
; mdi
= mdi
->next
)
7286 if (mdi
->recovery_start
!= MaxSector
) {
7287 recovery_not_finished
= 1;
7290 if (recovery_not_finished
) {
7292 dprintf("Rebuild has not finished yet, state not changed");
7293 if (a
->last_checkpoint
< mdi
->recovery_start
) {
7294 a
->last_checkpoint
= mdi
->recovery_start
;
7295 super
->updates_pending
++;
7299 end_migration(dev
, super
, map_state
);
7300 map
= get_imsm_map(dev
, MAP_0
);
7301 map
->failed_disk_num
= ~0;
7302 super
->updates_pending
++;
7303 a
->last_checkpoint
= 0;
7306 if (is_gen_migration(dev
)) {
7307 dprintf_cont("while general migration");
7308 if (a
->last_checkpoint
>= a
->info
.component_size
)
7309 end_migration(dev
, super
, map_state
);
7311 map
->map_state
= map_state
;
7312 map
= get_imsm_map(dev
, MAP_0
);
7313 map
->failed_disk_num
= ~0;
7314 super
->updates_pending
++;
7318 case IMSM_T_STATE_DEGRADED
: /* transition to degraded state */
7319 dprintf_cont("degraded: ");
7320 if ((map
->map_state
!= map_state
) &&
7321 !dev
->vol
.migr_state
) {
7322 dprintf_cont("mark degraded");
7323 map
->map_state
= map_state
;
7324 super
->updates_pending
++;
7325 a
->last_checkpoint
= 0;
7328 if (is_rebuilding(dev
)) {
7329 dprintf_cont("while rebuilding.");
7330 if (map
->map_state
!= map_state
) {
7331 dprintf_cont(" Map state change");
7332 end_migration(dev
, super
, map_state
);
7333 super
->updates_pending
++;
7337 if (is_gen_migration(dev
)) {
7338 dprintf_cont("while general migration");
7339 if (a
->last_checkpoint
>= a
->info
.component_size
)
7340 end_migration(dev
, super
, map_state
);
7342 map
->map_state
= map_state
;
7343 manage_second_map(super
, dev
);
7345 super
->updates_pending
++;
7348 if (is_initializing(dev
)) {
7349 dprintf_cont("while initialization.");
7350 map
->map_state
= map_state
;
7351 super
->updates_pending
++;
7355 case IMSM_T_STATE_FAILED
: /* transition to failed state */
7356 dprintf_cont("failed: ");
7357 if (is_gen_migration(dev
)) {
7358 dprintf_cont("while general migration");
7359 map
->map_state
= map_state
;
7360 super
->updates_pending
++;
7363 if (map
->map_state
!= map_state
) {
7364 dprintf_cont("mark failed");
7365 end_migration(dev
, super
, map_state
);
7366 super
->updates_pending
++;
7367 a
->last_checkpoint
= 0;
7372 dprintf_cont("state %i\n", map_state
);
7377 static int store_imsm_mpb(int fd
, struct imsm_super
*mpb
)
7380 __u32 mpb_size
= __le32_to_cpu(mpb
->mpb_size
);
7381 unsigned long long dsize
;
7382 unsigned long long sectors
;
7384 get_dev_size(fd
, NULL
, &dsize
);
7386 if (mpb_size
> 512) {
7387 /* -1 to account for anchor */
7388 sectors
= mpb_sectors(mpb
) - 1;
7390 /* write the extended mpb to the sectors preceeding the anchor */
7391 if (lseek64(fd
, dsize
- (512 * (2 + sectors
)), SEEK_SET
) < 0)
7394 if ((unsigned long long)write(fd
, buf
+ 512, 512 * sectors
)
7399 /* first block is stored on second to last sector of the disk */
7400 if (lseek64(fd
, dsize
- (512 * 2), SEEK_SET
) < 0)
7403 if (write(fd
, buf
, 512) != 512)
7409 static void imsm_sync_metadata(struct supertype
*container
)
7411 struct intel_super
*super
= container
->sb
;
7413 dprintf("sync metadata: %d\n", super
->updates_pending
);
7414 if (!super
->updates_pending
)
7417 write_super_imsm(container
, 0);
7419 super
->updates_pending
= 0;
7422 static struct dl
*imsm_readd(struct intel_super
*super
, int idx
, struct active_array
*a
)
7424 struct imsm_dev
*dev
= get_imsm_dev(super
, a
->info
.container_member
);
7425 int i
= get_imsm_disk_idx(dev
, idx
, MAP_X
);
7428 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
7432 if (dl
&& is_failed(&dl
->disk
))
7436 dprintf("found %x:%x\n", dl
->major
, dl
->minor
);
7441 static struct dl
*imsm_add_spare(struct intel_super
*super
, int slot
,
7442 struct active_array
*a
, int activate_new
,
7443 struct mdinfo
*additional_test_list
)
7445 struct imsm_dev
*dev
= get_imsm_dev(super
, a
->info
.container_member
);
7446 int idx
= get_imsm_disk_idx(dev
, slot
, MAP_X
);
7447 struct imsm_super
*mpb
= super
->anchor
;
7448 struct imsm_map
*map
;
7449 unsigned long long pos
;
7454 __u32 array_start
= 0;
7455 __u32 array_end
= 0;
7457 struct mdinfo
*test_list
;
7459 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
7460 /* If in this array, skip */
7461 for (d
= a
->info
.devs
; d
; d
= d
->next
)
7462 if (d
->state_fd
>= 0 &&
7463 d
->disk
.major
== dl
->major
&&
7464 d
->disk
.minor
== dl
->minor
) {
7465 dprintf("%x:%x already in array\n",
7466 dl
->major
, dl
->minor
);
7471 test_list
= additional_test_list
;
7473 if (test_list
->disk
.major
== dl
->major
&&
7474 test_list
->disk
.minor
== dl
->minor
) {
7475 dprintf("%x:%x already in additional test list\n",
7476 dl
->major
, dl
->minor
);
7479 test_list
= test_list
->next
;
7484 /* skip in use or failed drives */
7485 if (is_failed(&dl
->disk
) || idx
== dl
->index
||
7487 dprintf("%x:%x status (failed: %d index: %d)\n",
7488 dl
->major
, dl
->minor
, is_failed(&dl
->disk
), idx
);
7492 /* skip pure spares when we are looking for partially
7493 * assimilated drives
7495 if (dl
->index
== -1 && !activate_new
)
7498 /* Does this unused device have the requisite free space?
7499 * It needs to be able to cover all member volumes
7501 ex
= get_extents(super
, dl
);
7503 dprintf("cannot get extents\n");
7506 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
7507 dev
= get_imsm_dev(super
, i
);
7508 map
= get_imsm_map(dev
, MAP_0
);
7510 /* check if this disk is already a member of
7513 if (get_imsm_disk_slot(map
, dl
->index
) >= 0)
7519 array_start
= pba_of_lba0(map
);
7520 array_end
= array_start
+
7521 blocks_per_member(map
) - 1;
7524 /* check that we can start at pba_of_lba0 with
7525 * blocks_per_member of space
7527 if (array_start
>= pos
&& array_end
< ex
[j
].start
) {
7531 pos
= ex
[j
].start
+ ex
[j
].size
;
7533 } while (ex
[j
-1].size
);
7540 if (i
< mpb
->num_raid_devs
) {
7541 dprintf("%x:%x does not have %u to %u available\n",
7542 dl
->major
, dl
->minor
, array_start
, array_end
);
7552 static int imsm_rebuild_allowed(struct supertype
*cont
, int dev_idx
, int failed
)
7554 struct imsm_dev
*dev2
;
7555 struct imsm_map
*map
;
7561 dev2
= get_imsm_dev(cont
->sb
, dev_idx
);
7563 state
= imsm_check_degraded(cont
->sb
, dev2
, failed
, MAP_0
);
7564 if (state
== IMSM_T_STATE_FAILED
) {
7565 map
= get_imsm_map(dev2
, MAP_0
);
7568 for (slot
= 0; slot
< map
->num_members
; slot
++) {
7570 * Check if failed disks are deleted from intel
7571 * disk list or are marked to be deleted
7573 idx
= get_imsm_disk_idx(dev2
, slot
, MAP_X
);
7574 idisk
= get_imsm_dl_disk(cont
->sb
, idx
);
7576 * Do not rebuild the array if failed disks
7577 * from failed sub-array are not removed from
7581 is_failed(&idisk
->disk
) &&
7582 (idisk
->action
!= DISK_REMOVE
))
7590 static struct mdinfo
*imsm_activate_spare(struct active_array
*a
,
7591 struct metadata_update
**updates
)
7594 * Find a device with unused free space and use it to replace a
7595 * failed/vacant region in an array. We replace failed regions one a
7596 * array at a time. The result is that a new spare disk will be added
7597 * to the first failed array and after the monitor has finished
7598 * propagating failures the remainder will be consumed.
7600 * FIXME add a capability for mdmon to request spares from another
7604 struct intel_super
*super
= a
->container
->sb
;
7605 int inst
= a
->info
.container_member
;
7606 struct imsm_dev
*dev
= get_imsm_dev(super
, inst
);
7607 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
7608 int failed
= a
->info
.array
.raid_disks
;
7609 struct mdinfo
*rv
= NULL
;
7612 struct metadata_update
*mu
;
7614 struct imsm_update_activate_spare
*u
;
7619 for (d
= a
->info
.devs
; d
; d
= d
->next
) {
7620 if ((d
->curr_state
& DS_FAULTY
) &&
7622 /* wait for Removal to happen */
7624 if (d
->state_fd
>= 0)
7628 dprintf("imsm: activate spare: inst=%d failed=%d (%d) level=%d\n",
7629 inst
, failed
, a
->info
.array
.raid_disks
, a
->info
.array
.level
);
7631 if (imsm_reshape_blocks_arrays_changes(super
))
7634 /* Cannot activate another spare if rebuild is in progress already
7636 if (is_rebuilding(dev
)) {
7637 dprintf("imsm: No spare activation allowed. Rebuild in progress already.\n");
7641 if (a
->info
.array
.level
== 4)
7642 /* No repair for takeovered array
7643 * imsm doesn't support raid4
7647 if (imsm_check_degraded(super
, dev
, failed
, MAP_0
) !=
7648 IMSM_T_STATE_DEGRADED
)
7652 * If there are any failed disks check state of the other volume.
7653 * Block rebuild if the another one is failed until failed disks
7654 * are removed from container.
7657 dprintf("found failed disks in %.*s, check if there anotherfailed sub-array.\n",
7658 MAX_RAID_SERIAL_LEN
, dev
->volume
);
7659 /* check if states of the other volumes allow for rebuild */
7660 for (i
= 0; i
< super
->anchor
->num_raid_devs
; i
++) {
7662 allowed
= imsm_rebuild_allowed(a
->container
,
7670 /* For each slot, if it is not working, find a spare */
7671 for (i
= 0; i
< a
->info
.array
.raid_disks
; i
++) {
7672 for (d
= a
->info
.devs
; d
; d
= d
->next
)
7673 if (d
->disk
.raid_disk
== i
)
7675 dprintf("found %d: %p %x\n", i
, d
, d
?d
->curr_state
:0);
7676 if (d
&& (d
->state_fd
>= 0))
7680 * OK, this device needs recovery. Try to re-add the
7681 * previous occupant of this slot, if this fails see if
7682 * we can continue the assimilation of a spare that was
7683 * partially assimilated, finally try to activate a new
7686 dl
= imsm_readd(super
, i
, a
);
7688 dl
= imsm_add_spare(super
, i
, a
, 0, rv
);
7690 dl
= imsm_add_spare(super
, i
, a
, 1, rv
);
7694 /* found a usable disk with enough space */
7695 di
= xcalloc(1, sizeof(*di
));
7697 /* dl->index will be -1 in the case we are activating a
7698 * pristine spare. imsm_process_update() will create a
7699 * new index in this case. Once a disk is found to be
7700 * failed in all member arrays it is kicked from the
7703 di
->disk
.number
= dl
->index
;
7705 /* (ab)use di->devs to store a pointer to the device
7708 di
->devs
= (struct mdinfo
*) dl
;
7710 di
->disk
.raid_disk
= i
;
7711 di
->disk
.major
= dl
->major
;
7712 di
->disk
.minor
= dl
->minor
;
7714 di
->recovery_start
= 0;
7715 di
->data_offset
= pba_of_lba0(map
);
7716 di
->component_size
= a
->info
.component_size
;
7717 di
->container_member
= inst
;
7718 super
->random
= random32();
7722 dprintf("%x:%x to be %d at %llu\n", dl
->major
, dl
->minor
,
7723 i
, di
->data_offset
);
7727 /* No spares found */
7729 /* Now 'rv' has a list of devices to return.
7730 * Create a metadata_update record to update the
7731 * disk_ord_tbl for the array
7733 mu
= xmalloc(sizeof(*mu
));
7734 mu
->buf
= xcalloc(num_spares
,
7735 sizeof(struct imsm_update_activate_spare
));
7737 mu
->space_list
= NULL
;
7738 mu
->len
= sizeof(struct imsm_update_activate_spare
) * num_spares
;
7739 mu
->next
= *updates
;
7740 u
= (struct imsm_update_activate_spare
*) mu
->buf
;
7742 for (di
= rv
; di
; di
= di
->next
) {
7743 u
->type
= update_activate_spare
;
7744 u
->dl
= (struct dl
*) di
->devs
;
7746 u
->slot
= di
->disk
.raid_disk
;
7757 static int disks_overlap(struct intel_super
*super
, int idx
, struct imsm_update_create_array
*u
)
7759 struct imsm_dev
*dev
= get_imsm_dev(super
, idx
);
7760 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
7761 struct imsm_map
*new_map
= get_imsm_map(&u
->dev
, MAP_0
);
7762 struct disk_info
*inf
= get_disk_info(u
);
7763 struct imsm_disk
*disk
;
7767 for (i
= 0; i
< map
->num_members
; i
++) {
7768 disk
= get_imsm_disk(super
, get_imsm_disk_idx(dev
, i
, MAP_X
));
7769 for (j
= 0; j
< new_map
->num_members
; j
++)
7770 if (serialcmp(disk
->serial
, inf
[j
].serial
) == 0)
7777 static struct dl
*get_disk_super(struct intel_super
*super
, int major
, int minor
)
7779 struct dl
*dl
= NULL
;
7780 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
7781 if ((dl
->major
== major
) && (dl
->minor
== minor
))
7786 static int remove_disk_super(struct intel_super
*super
, int major
, int minor
)
7788 struct dl
*prev
= NULL
;
7792 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
7793 if ((dl
->major
== major
) && (dl
->minor
== minor
)) {
7796 prev
->next
= dl
->next
;
7798 super
->disks
= dl
->next
;
7800 __free_imsm_disk(dl
);
7801 dprintf("removed %x:%x\n", major
, minor
);
7809 static void imsm_delete(struct intel_super
*super
, struct dl
**dlp
, unsigned index
);
7811 static int add_remove_disk_update(struct intel_super
*super
)
7813 int check_degraded
= 0;
7814 struct dl
*disk
= NULL
;
7815 /* add/remove some spares to/from the metadata/contrainer */
7816 while (super
->disk_mgmt_list
) {
7817 struct dl
*disk_cfg
;
7819 disk_cfg
= super
->disk_mgmt_list
;
7820 super
->disk_mgmt_list
= disk_cfg
->next
;
7821 disk_cfg
->next
= NULL
;
7823 if (disk_cfg
->action
== DISK_ADD
) {
7824 disk_cfg
->next
= super
->disks
;
7825 super
->disks
= disk_cfg
;
7827 dprintf("added %x:%x\n",
7828 disk_cfg
->major
, disk_cfg
->minor
);
7829 } else if (disk_cfg
->action
== DISK_REMOVE
) {
7830 dprintf("Disk remove action processed: %x.%x\n",
7831 disk_cfg
->major
, disk_cfg
->minor
);
7832 disk
= get_disk_super(super
,
7836 /* store action status */
7837 disk
->action
= DISK_REMOVE
;
7838 /* remove spare disks only */
7839 if (disk
->index
== -1) {
7840 remove_disk_super(super
,
7845 /* release allocate disk structure */
7846 __free_imsm_disk(disk_cfg
);
7849 return check_degraded
;
7852 static int apply_reshape_migration_update(struct imsm_update_reshape_migration
*u
,
7853 struct intel_super
*super
,
7856 struct intel_dev
*id
;
7857 void **tofree
= NULL
;
7860 dprintf("(enter)\n");
7861 if ((u
->subdev
< 0) ||
7863 dprintf("imsm: Error: Wrong subdev: %i\n", u
->subdev
);
7866 if ((space_list
== NULL
) || (*space_list
== NULL
)) {
7867 dprintf("imsm: Error: Memory is not allocated\n");
7871 for (id
= super
->devlist
; id
; id
= id
->next
) {
7872 if (id
->index
== (unsigned)u
->subdev
) {
7873 struct imsm_dev
*dev
= get_imsm_dev(super
, u
->subdev
);
7874 struct imsm_map
*map
;
7875 struct imsm_dev
*new_dev
=
7876 (struct imsm_dev
*)*space_list
;
7877 struct imsm_map
*migr_map
= get_imsm_map(dev
, MAP_1
);
7879 struct dl
*new_disk
;
7881 if (new_dev
== NULL
)
7883 *space_list
= **space_list
;
7884 memcpy(new_dev
, dev
, sizeof_imsm_dev(dev
, 0));
7885 map
= get_imsm_map(new_dev
, MAP_0
);
7887 dprintf("imsm: Error: migration in progress");
7891 to_state
= map
->map_state
;
7892 if ((u
->new_level
== 5) && (map
->raid_level
== 0)) {
7894 /* this should not happen */
7895 if (u
->new_disks
[0] < 0) {
7896 map
->failed_disk_num
=
7897 map
->num_members
- 1;
7898 to_state
= IMSM_T_STATE_DEGRADED
;
7900 to_state
= IMSM_T_STATE_NORMAL
;
7902 migrate(new_dev
, super
, to_state
, MIGR_GEN_MIGR
);
7903 if (u
->new_level
> -1)
7904 map
->raid_level
= u
->new_level
;
7905 migr_map
= get_imsm_map(new_dev
, MAP_1
);
7906 if ((u
->new_level
== 5) &&
7907 (migr_map
->raid_level
== 0)) {
7908 int ord
= map
->num_members
- 1;
7909 migr_map
->num_members
--;
7910 if (u
->new_disks
[0] < 0)
7911 ord
|= IMSM_ORD_REBUILD
;
7912 set_imsm_ord_tbl_ent(map
,
7913 map
->num_members
- 1,
7917 tofree
= (void **)dev
;
7919 /* update chunk size
7921 if (u
->new_chunksize
> 0)
7922 map
->blocks_per_strip
=
7923 __cpu_to_le16(u
->new_chunksize
* 2);
7927 if ((u
->new_level
!= 5) ||
7928 (migr_map
->raid_level
!= 0) ||
7929 (migr_map
->raid_level
== map
->raid_level
))
7932 if (u
->new_disks
[0] >= 0) {
7935 new_disk
= get_disk_super(super
,
7936 major(u
->new_disks
[0]),
7937 minor(u
->new_disks
[0]));
7938 dprintf("imsm: new disk for reshape is: %i:%i (%p, index = %i)\n",
7939 major(u
->new_disks
[0]),
7940 minor(u
->new_disks
[0]),
7941 new_disk
, new_disk
->index
);
7942 if (new_disk
== NULL
)
7943 goto error_disk_add
;
7945 new_disk
->index
= map
->num_members
- 1;
7946 /* slot to fill in autolayout
7948 new_disk
->raiddisk
= new_disk
->index
;
7949 new_disk
->disk
.status
|= CONFIGURED_DISK
;
7950 new_disk
->disk
.status
&= ~SPARE_DISK
;
7952 goto error_disk_add
;
7955 *tofree
= *space_list
;
7956 /* calculate new size
7958 imsm_set_array_size(new_dev
, -1);
7965 *space_list
= tofree
;
7969 dprintf("Error: imsm: Cannot find disk.\n");
7973 static int apply_size_change_update(struct imsm_update_size_change
*u
,
7974 struct intel_super
*super
)
7976 struct intel_dev
*id
;
7979 dprintf("(enter)\n");
7980 if ((u
->subdev
< 0) ||
7982 dprintf("imsm: Error: Wrong subdev: %i\n", u
->subdev
);
7986 for (id
= super
->devlist
; id
; id
= id
->next
) {
7987 if (id
->index
== (unsigned)u
->subdev
) {
7988 struct imsm_dev
*dev
= get_imsm_dev(super
, u
->subdev
);
7989 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
7990 int used_disks
= imsm_num_data_members(dev
, MAP_0
);
7991 unsigned long long blocks_per_member
;
7993 /* calculate new size
7995 blocks_per_member
= u
->new_size
/ used_disks
;
7996 dprintf("(size: %llu, blocks per member: %llu)\n",
7997 u
->new_size
, blocks_per_member
);
7998 set_blocks_per_member(map
, blocks_per_member
);
7999 imsm_set_array_size(dev
, u
->new_size
);
8009 static int apply_update_activate_spare(struct imsm_update_activate_spare
*u
,
8010 struct intel_super
*super
,
8011 struct active_array
*active_array
)
8013 struct imsm_super
*mpb
= super
->anchor
;
8014 struct imsm_dev
*dev
= get_imsm_dev(super
, u
->array
);
8015 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
8016 struct imsm_map
*migr_map
;
8017 struct active_array
*a
;
8018 struct imsm_disk
*disk
;
8025 int second_map_created
= 0;
8027 for (; u
; u
= u
->next
) {
8028 victim
= get_imsm_disk_idx(dev
, u
->slot
, MAP_X
);
8033 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
8038 pr_err("error: imsm_activate_spare passed an unknown disk (index: %d)\n",
8043 /* count failures (excluding rebuilds and the victim)
8044 * to determine map[0] state
8047 for (i
= 0; i
< map
->num_members
; i
++) {
8050 disk
= get_imsm_disk(super
,
8051 get_imsm_disk_idx(dev
, i
, MAP_X
));
8052 if (!disk
|| is_failed(disk
))
8056 /* adding a pristine spare, assign a new index */
8057 if (dl
->index
< 0) {
8058 dl
->index
= super
->anchor
->num_disks
;
8059 super
->anchor
->num_disks
++;
8062 disk
->status
|= CONFIGURED_DISK
;
8063 disk
->status
&= ~SPARE_DISK
;
8066 to_state
= imsm_check_degraded(super
, dev
, failed
, MAP_0
);
8067 if (!second_map_created
) {
8068 second_map_created
= 1;
8069 map
->map_state
= IMSM_T_STATE_DEGRADED
;
8070 migrate(dev
, super
, to_state
, MIGR_REBUILD
);
8072 map
->map_state
= to_state
;
8073 migr_map
= get_imsm_map(dev
, MAP_1
);
8074 set_imsm_ord_tbl_ent(map
, u
->slot
, dl
->index
);
8075 set_imsm_ord_tbl_ent(migr_map
, u
->slot
,
8076 dl
->index
| IMSM_ORD_REBUILD
);
8078 /* update the family_num to mark a new container
8079 * generation, being careful to record the existing
8080 * family_num in orig_family_num to clean up after
8081 * earlier mdadm versions that neglected to set it.
8083 if (mpb
->orig_family_num
== 0)
8084 mpb
->orig_family_num
= mpb
->family_num
;
8085 mpb
->family_num
+= super
->random
;
8087 /* count arrays using the victim in the metadata */
8089 for (a
= active_array
; a
; a
= a
->next
) {
8090 dev
= get_imsm_dev(super
, a
->info
.container_member
);
8091 map
= get_imsm_map(dev
, MAP_0
);
8093 if (get_imsm_disk_slot(map
, victim
) >= 0)
8097 /* delete the victim if it is no longer being
8103 /* We know that 'manager' isn't touching anything,
8104 * so it is safe to delete
8106 for (dlp
= &super
->disks
; *dlp
; dlp
= &(*dlp
)->next
)
8107 if ((*dlp
)->index
== victim
)
8110 /* victim may be on the missing list */
8112 for (dlp
= &super
->missing
; *dlp
;
8113 dlp
= &(*dlp
)->next
)
8114 if ((*dlp
)->index
== victim
)
8116 imsm_delete(super
, dlp
, victim
);
8123 static int apply_reshape_container_disks_update(struct imsm_update_reshape
*u
,
8124 struct intel_super
*super
,
8127 struct dl
*new_disk
;
8128 struct intel_dev
*id
;
8130 int delta_disks
= u
->new_raid_disks
- u
->old_raid_disks
;
8131 int disk_count
= u
->old_raid_disks
;
8132 void **tofree
= NULL
;
8133 int devices_to_reshape
= 1;
8134 struct imsm_super
*mpb
= super
->anchor
;
8136 unsigned int dev_id
;
8138 dprintf("(enter)\n");
8140 /* enable spares to use in array */
8141 for (i
= 0; i
< delta_disks
; i
++) {
8142 new_disk
= get_disk_super(super
,
8143 major(u
->new_disks
[i
]),
8144 minor(u
->new_disks
[i
]));
8145 dprintf("imsm: new disk for reshape is: %i:%i (%p, index = %i)\n",
8146 major(u
->new_disks
[i
]), minor(u
->new_disks
[i
]),
8147 new_disk
, new_disk
->index
);
8148 if ((new_disk
== NULL
) ||
8149 ((new_disk
->index
>= 0) &&
8150 (new_disk
->index
< u
->old_raid_disks
)))
8151 goto update_reshape_exit
;
8152 new_disk
->index
= disk_count
++;
8153 /* slot to fill in autolayout
8155 new_disk
->raiddisk
= new_disk
->index
;
8156 new_disk
->disk
.status
|=
8158 new_disk
->disk
.status
&= ~SPARE_DISK
;
8161 dprintf("imsm: volume set mpb->num_raid_devs = %i\n",
8162 mpb
->num_raid_devs
);
8163 /* manage changes in volume
8165 for (dev_id
= 0; dev_id
< mpb
->num_raid_devs
; dev_id
++) {
8166 void **sp
= *space_list
;
8167 struct imsm_dev
*newdev
;
8168 struct imsm_map
*newmap
, *oldmap
;
8170 for (id
= super
->devlist
; id
; id
= id
->next
) {
8171 if (id
->index
== dev_id
)
8180 /* Copy the dev, but not (all of) the map */
8181 memcpy(newdev
, id
->dev
, sizeof(*newdev
));
8182 oldmap
= get_imsm_map(id
->dev
, MAP_0
);
8183 newmap
= get_imsm_map(newdev
, MAP_0
);
8184 /* Copy the current map */
8185 memcpy(newmap
, oldmap
, sizeof_imsm_map(oldmap
));
8186 /* update one device only
8188 if (devices_to_reshape
) {
8189 dprintf("imsm: modifying subdev: %i\n",
8191 devices_to_reshape
--;
8192 newdev
->vol
.migr_state
= 1;
8193 newdev
->vol
.curr_migr_unit
= 0;
8194 set_migr_type(newdev
, MIGR_GEN_MIGR
);
8195 newmap
->num_members
= u
->new_raid_disks
;
8196 for (i
= 0; i
< delta_disks
; i
++) {
8197 set_imsm_ord_tbl_ent(newmap
,
8198 u
->old_raid_disks
+ i
,
8199 u
->old_raid_disks
+ i
);
8201 /* New map is correct, now need to save old map
8203 newmap
= get_imsm_map(newdev
, MAP_1
);
8204 memcpy(newmap
, oldmap
, sizeof_imsm_map(oldmap
));
8206 imsm_set_array_size(newdev
, -1);
8209 sp
= (void **)id
->dev
;
8214 /* Clear migration record */
8215 memset(super
->migr_rec
, 0, sizeof(struct migr_record
));
8218 *space_list
= tofree
;
8221 update_reshape_exit
:
8226 static int apply_takeover_update(struct imsm_update_takeover
*u
,
8227 struct intel_super
*super
,
8230 struct imsm_dev
*dev
= NULL
;
8231 struct intel_dev
*dv
;
8232 struct imsm_dev
*dev_new
;
8233 struct imsm_map
*map
;
8237 for (dv
= super
->devlist
; dv
; dv
= dv
->next
)
8238 if (dv
->index
== (unsigned int)u
->subarray
) {
8246 map
= get_imsm_map(dev
, MAP_0
);
8248 if (u
->direction
== R10_TO_R0
) {
8249 /* Number of failed disks must be half of initial disk number */
8250 if (imsm_count_failed(super
, dev
, MAP_0
) !=
8251 (map
->num_members
/ 2))
8254 /* iterate through devices to mark removed disks as spare */
8255 for (dm
= super
->disks
; dm
; dm
= dm
->next
) {
8256 if (dm
->disk
.status
& FAILED_DISK
) {
8257 int idx
= dm
->index
;
8258 /* update indexes on the disk list */
8259 /* FIXME this loop-with-the-loop looks wrong, I'm not convinced
8260 the index values will end up being correct.... NB */
8261 for (du
= super
->disks
; du
; du
= du
->next
)
8262 if (du
->index
> idx
)
8264 /* mark as spare disk */
8269 map
->num_members
= map
->num_members
/ 2;
8270 map
->map_state
= IMSM_T_STATE_NORMAL
;
8271 map
->num_domains
= 1;
8272 map
->raid_level
= 0;
8273 map
->failed_disk_num
= -1;
8276 if (u
->direction
== R0_TO_R10
) {
8278 /* update slots in current disk list */
8279 for (dm
= super
->disks
; dm
; dm
= dm
->next
) {
8283 /* create new *missing* disks */
8284 for (i
= 0; i
< map
->num_members
; i
++) {
8285 space
= *space_list
;
8288 *space_list
= *space
;
8290 memcpy(du
, super
->disks
, sizeof(*du
));
8294 du
->index
= (i
* 2) + 1;
8295 sprintf((char *)du
->disk
.serial
,
8296 " MISSING_%d", du
->index
);
8297 sprintf((char *)du
->serial
,
8298 "MISSING_%d", du
->index
);
8299 du
->next
= super
->missing
;
8300 super
->missing
= du
;
8302 /* create new dev and map */
8303 space
= *space_list
;
8306 *space_list
= *space
;
8307 dev_new
= (void *)space
;
8308 memcpy(dev_new
, dev
, sizeof(*dev
));
8309 /* update new map */
8310 map
= get_imsm_map(dev_new
, MAP_0
);
8311 map
->num_members
= map
->num_members
* 2;
8312 map
->map_state
= IMSM_T_STATE_DEGRADED
;
8313 map
->num_domains
= 2;
8314 map
->raid_level
= 1;
8315 /* replace dev<->dev_new */
8318 /* update disk order table */
8319 for (du
= super
->disks
; du
; du
= du
->next
)
8321 set_imsm_ord_tbl_ent(map
, du
->index
, du
->index
);
8322 for (du
= super
->missing
; du
; du
= du
->next
)
8323 if (du
->index
>= 0) {
8324 set_imsm_ord_tbl_ent(map
, du
->index
, du
->index
);
8325 mark_missing(dv
->dev
, &du
->disk
, du
->index
);
8331 static void imsm_process_update(struct supertype
*st
,
8332 struct metadata_update
*update
)
8335 * crack open the metadata_update envelope to find the update record
8336 * update can be one of:
8337 * update_reshape_container_disks - all the arrays in the container
8338 * are being reshaped to have more devices. We need to mark
8339 * the arrays for general migration and convert selected spares
8340 * into active devices.
8341 * update_activate_spare - a spare device has replaced a failed
8342 * device in an array, update the disk_ord_tbl. If this disk is
8343 * present in all member arrays then also clear the SPARE_DISK
8345 * update_create_array
8347 * update_rename_array
8348 * update_add_remove_disk
8350 struct intel_super
*super
= st
->sb
;
8351 struct imsm_super
*mpb
;
8352 enum imsm_update_type type
= *(enum imsm_update_type
*) update
->buf
;
8354 /* update requires a larger buf but the allocation failed */
8355 if (super
->next_len
&& !super
->next_buf
) {
8356 super
->next_len
= 0;
8360 if (super
->next_buf
) {
8361 memcpy(super
->next_buf
, super
->buf
, super
->len
);
8363 super
->len
= super
->next_len
;
8364 super
->buf
= super
->next_buf
;
8366 super
->next_len
= 0;
8367 super
->next_buf
= NULL
;
8370 mpb
= super
->anchor
;
8373 case update_general_migration_checkpoint
: {
8374 struct intel_dev
*id
;
8375 struct imsm_update_general_migration_checkpoint
*u
=
8376 (void *)update
->buf
;
8378 dprintf("called for update_general_migration_checkpoint\n");
8380 /* find device under general migration */
8381 for (id
= super
->devlist
; id
; id
= id
->next
) {
8382 if (is_gen_migration(id
->dev
)) {
8383 id
->dev
->vol
.curr_migr_unit
=
8384 __cpu_to_le32(u
->curr_migr_unit
);
8385 super
->updates_pending
++;
8390 case update_takeover
: {
8391 struct imsm_update_takeover
*u
= (void *)update
->buf
;
8392 if (apply_takeover_update(u
, super
, &update
->space_list
)) {
8393 imsm_update_version_info(super
);
8394 super
->updates_pending
++;
8399 case update_reshape_container_disks
: {
8400 struct imsm_update_reshape
*u
= (void *)update
->buf
;
8401 if (apply_reshape_container_disks_update(
8402 u
, super
, &update
->space_list
))
8403 super
->updates_pending
++;
8406 case update_reshape_migration
: {
8407 struct imsm_update_reshape_migration
*u
= (void *)update
->buf
;
8408 if (apply_reshape_migration_update(
8409 u
, super
, &update
->space_list
))
8410 super
->updates_pending
++;
8413 case update_size_change
: {
8414 struct imsm_update_size_change
*u
= (void *)update
->buf
;
8415 if (apply_size_change_update(u
, super
))
8416 super
->updates_pending
++;
8419 case update_activate_spare
: {
8420 struct imsm_update_activate_spare
*u
= (void *) update
->buf
;
8421 if (apply_update_activate_spare(u
, super
, st
->arrays
))
8422 super
->updates_pending
++;
8425 case update_create_array
: {
8426 /* someone wants to create a new array, we need to be aware of
8427 * a few races/collisions:
8428 * 1/ 'Create' called by two separate instances of mdadm
8429 * 2/ 'Create' versus 'activate_spare': mdadm has chosen
8430 * devices that have since been assimilated via
8432 * In the event this update can not be carried out mdadm will
8433 * (FIX ME) notice that its update did not take hold.
8435 struct imsm_update_create_array
*u
= (void *) update
->buf
;
8436 struct intel_dev
*dv
;
8437 struct imsm_dev
*dev
;
8438 struct imsm_map
*map
, *new_map
;
8439 unsigned long long start
, end
;
8440 unsigned long long new_start
, new_end
;
8442 struct disk_info
*inf
;
8445 /* handle racing creates: first come first serve */
8446 if (u
->dev_idx
< mpb
->num_raid_devs
) {
8447 dprintf("subarray %d already defined\n", u
->dev_idx
);
8451 /* check update is next in sequence */
8452 if (u
->dev_idx
!= mpb
->num_raid_devs
) {
8453 dprintf("can not create array %d expected index %d\n",
8454 u
->dev_idx
, mpb
->num_raid_devs
);
8458 new_map
= get_imsm_map(&u
->dev
, MAP_0
);
8459 new_start
= pba_of_lba0(new_map
);
8460 new_end
= new_start
+ blocks_per_member(new_map
);
8461 inf
= get_disk_info(u
);
8463 /* handle activate_spare versus create race:
8464 * check to make sure that overlapping arrays do not include
8467 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
8468 dev
= get_imsm_dev(super
, i
);
8469 map
= get_imsm_map(dev
, MAP_0
);
8470 start
= pba_of_lba0(map
);
8471 end
= start
+ blocks_per_member(map
);
8472 if ((new_start
>= start
&& new_start
<= end
) ||
8473 (start
>= new_start
&& start
<= new_end
))
8478 if (disks_overlap(super
, i
, u
)) {
8479 dprintf("arrays overlap\n");
8484 /* check that prepare update was successful */
8485 if (!update
->space
) {
8486 dprintf("prepare update failed\n");
8490 /* check that all disks are still active before committing
8491 * changes. FIXME: could we instead handle this by creating a
8492 * degraded array? That's probably not what the user expects,
8493 * so better to drop this update on the floor.
8495 for (i
= 0; i
< new_map
->num_members
; i
++) {
8496 dl
= serial_to_dl(inf
[i
].serial
, super
);
8498 dprintf("disk disappeared\n");
8503 super
->updates_pending
++;
8505 /* convert spares to members and fixup ord_tbl */
8506 for (i
= 0; i
< new_map
->num_members
; i
++) {
8507 dl
= serial_to_dl(inf
[i
].serial
, super
);
8508 if (dl
->index
== -1) {
8509 dl
->index
= mpb
->num_disks
;
8511 dl
->disk
.status
|= CONFIGURED_DISK
;
8512 dl
->disk
.status
&= ~SPARE_DISK
;
8514 set_imsm_ord_tbl_ent(new_map
, i
, dl
->index
);
8519 update
->space
= NULL
;
8520 imsm_copy_dev(dev
, &u
->dev
);
8521 dv
->index
= u
->dev_idx
;
8522 dv
->next
= super
->devlist
;
8523 super
->devlist
= dv
;
8524 mpb
->num_raid_devs
++;
8526 imsm_update_version_info(super
);
8529 /* mdmon knows how to release update->space, but not
8530 * ((struct intel_dev *) update->space)->dev
8532 if (update
->space
) {
8538 case update_kill_array
: {
8539 struct imsm_update_kill_array
*u
= (void *) update
->buf
;
8540 int victim
= u
->dev_idx
;
8541 struct active_array
*a
;
8542 struct intel_dev
**dp
;
8543 struct imsm_dev
*dev
;
8545 /* sanity check that we are not affecting the uuid of
8546 * active arrays, or deleting an active array
8548 * FIXME when immutable ids are available, but note that
8549 * we'll also need to fixup the invalidated/active
8550 * subarray indexes in mdstat
8552 for (a
= st
->arrays
; a
; a
= a
->next
)
8553 if (a
->info
.container_member
>= victim
)
8555 /* by definition if mdmon is running at least one array
8556 * is active in the container, so checking
8557 * mpb->num_raid_devs is just extra paranoia
8559 dev
= get_imsm_dev(super
, victim
);
8560 if (a
|| !dev
|| mpb
->num_raid_devs
== 1) {
8561 dprintf("failed to delete subarray-%d\n", victim
);
8565 for (dp
= &super
->devlist
; *dp
;)
8566 if ((*dp
)->index
== (unsigned)super
->current_vol
) {
8569 if ((*dp
)->index
> (unsigned)victim
)
8573 mpb
->num_raid_devs
--;
8574 super
->updates_pending
++;
8577 case update_rename_array
: {
8578 struct imsm_update_rename_array
*u
= (void *) update
->buf
;
8579 char name
[MAX_RAID_SERIAL_LEN
+1];
8580 int target
= u
->dev_idx
;
8581 struct active_array
*a
;
8582 struct imsm_dev
*dev
;
8584 /* sanity check that we are not affecting the uuid of
8587 snprintf(name
, MAX_RAID_SERIAL_LEN
, "%s", (char *) u
->name
);
8588 name
[MAX_RAID_SERIAL_LEN
] = '\0';
8589 for (a
= st
->arrays
; a
; a
= a
->next
)
8590 if (a
->info
.container_member
== target
)
8592 dev
= get_imsm_dev(super
, u
->dev_idx
);
8593 if (a
|| !dev
|| !check_name(super
, name
, 1)) {
8594 dprintf("failed to rename subarray-%d\n", target
);
8598 snprintf((char *) dev
->volume
, MAX_RAID_SERIAL_LEN
, "%s", name
);
8599 super
->updates_pending
++;
8602 case update_add_remove_disk
: {
8603 /* we may be able to repair some arrays if disks are
8604 * being added, check the status of add_remove_disk
8605 * if discs has been added.
8607 if (add_remove_disk_update(super
)) {
8608 struct active_array
*a
;
8610 super
->updates_pending
++;
8611 for (a
= st
->arrays
; a
; a
= a
->next
)
8612 a
->check_degraded
= 1;
8617 pr_err("error: unsuported process update type:(type: %d)\n", type
);
8621 static struct mdinfo
*get_spares_for_grow(struct supertype
*st
);
8623 static int imsm_prepare_update(struct supertype
*st
,
8624 struct metadata_update
*update
)
8627 * Allocate space to hold new disk entries, raid-device entries or a new
8628 * mpb if necessary. The manager synchronously waits for updates to
8629 * complete in the monitor, so new mpb buffers allocated here can be
8630 * integrated by the monitor thread without worrying about live pointers
8631 * in the manager thread.
8633 enum imsm_update_type type
;
8634 struct intel_super
*super
= st
->sb
;
8635 struct imsm_super
*mpb
= super
->anchor
;
8639 if (update
->len
< (int)sizeof(type
))
8642 type
= *(enum imsm_update_type
*) update
->buf
;
8645 case update_general_migration_checkpoint
:
8646 if (update
->len
< (int)sizeof(struct imsm_update_general_migration_checkpoint
))
8648 dprintf("called for update_general_migration_checkpoint\n");
8650 case update_takeover
: {
8651 struct imsm_update_takeover
*u
= (void *)update
->buf
;
8652 if (update
->len
< (int)sizeof(*u
))
8654 if (u
->direction
== R0_TO_R10
) {
8655 void **tail
= (void **)&update
->space_list
;
8656 struct imsm_dev
*dev
= get_imsm_dev(super
, u
->subarray
);
8657 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
8658 int num_members
= map
->num_members
;
8661 /* allocate memory for added disks */
8662 for (i
= 0; i
< num_members
; i
++) {
8663 size
= sizeof(struct dl
);
8664 space
= xmalloc(size
);
8669 /* allocate memory for new device */
8670 size
= sizeof_imsm_dev(super
->devlist
->dev
, 0) +
8671 (num_members
* sizeof(__u32
));
8672 space
= xmalloc(size
);
8676 len
= disks_to_mpb_size(num_members
* 2);
8681 case update_reshape_container_disks
: {
8682 /* Every raid device in the container is about to
8683 * gain some more devices, and we will enter a
8685 * So each 'imsm_map' will be bigger, and the imsm_vol
8686 * will now hold 2 of them.
8687 * Thus we need new 'struct imsm_dev' allocations sized
8688 * as sizeof_imsm_dev but with more devices in both maps.
8690 struct imsm_update_reshape
*u
= (void *)update
->buf
;
8691 struct intel_dev
*dl
;
8692 void **space_tail
= (void**)&update
->space_list
;
8694 if (update
->len
< (int)sizeof(*u
))
8697 dprintf("for update_reshape\n");
8699 for (dl
= super
->devlist
; dl
; dl
= dl
->next
) {
8700 int size
= sizeof_imsm_dev(dl
->dev
, 1);
8702 if (u
->new_raid_disks
> u
->old_raid_disks
)
8703 size
+= sizeof(__u32
)*2*
8704 (u
->new_raid_disks
- u
->old_raid_disks
);
8711 len
= disks_to_mpb_size(u
->new_raid_disks
);
8712 dprintf("New anchor length is %llu\n", (unsigned long long)len
);
8715 case update_reshape_migration
: {
8716 /* for migration level 0->5 we need to add disks
8717 * so the same as for container operation we will copy
8718 * device to the bigger location.
8719 * in memory prepared device and new disk area are prepared
8720 * for usage in process update
8722 struct imsm_update_reshape_migration
*u
= (void *)update
->buf
;
8723 struct intel_dev
*id
;
8724 void **space_tail
= (void **)&update
->space_list
;
8727 int current_level
= -1;
8729 if (update
->len
< (int)sizeof(*u
))
8732 dprintf("for update_reshape\n");
8734 /* add space for bigger array in update
8736 for (id
= super
->devlist
; id
; id
= id
->next
) {
8737 if (id
->index
== (unsigned)u
->subdev
) {
8738 size
= sizeof_imsm_dev(id
->dev
, 1);
8739 if (u
->new_raid_disks
> u
->old_raid_disks
)
8740 size
+= sizeof(__u32
)*2*
8741 (u
->new_raid_disks
- u
->old_raid_disks
);
8749 if (update
->space_list
== NULL
)
8752 /* add space for disk in update
8754 size
= sizeof(struct dl
);
8760 /* add spare device to update
8762 for (id
= super
->devlist
; id
; id
= id
->next
)
8763 if (id
->index
== (unsigned)u
->subdev
) {
8764 struct imsm_dev
*dev
;
8765 struct imsm_map
*map
;
8767 dev
= get_imsm_dev(super
, u
->subdev
);
8768 map
= get_imsm_map(dev
, MAP_0
);
8769 current_level
= map
->raid_level
;
8772 if ((u
->new_level
== 5) && (u
->new_level
!= current_level
)) {
8773 struct mdinfo
*spares
;
8775 spares
= get_spares_for_grow(st
);
8783 makedev(dev
->disk
.major
,
8785 dl
= get_disk_super(super
,
8788 dl
->index
= u
->old_raid_disks
;
8794 len
= disks_to_mpb_size(u
->new_raid_disks
);
8795 dprintf("New anchor length is %llu\n", (unsigned long long)len
);
8798 case update_size_change
: {
8799 if (update
->len
< (int)sizeof(struct imsm_update_size_change
))
8803 case update_activate_spare
: {
8804 if (update
->len
< (int)sizeof(struct imsm_update_activate_spare
))
8808 case update_create_array
: {
8809 struct imsm_update_create_array
*u
= (void *) update
->buf
;
8810 struct intel_dev
*dv
;
8811 struct imsm_dev
*dev
= &u
->dev
;
8812 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
8814 struct disk_info
*inf
;
8818 if (update
->len
< (int)sizeof(*u
))
8821 inf
= get_disk_info(u
);
8822 len
= sizeof_imsm_dev(dev
, 1);
8823 /* allocate a new super->devlist entry */
8824 dv
= xmalloc(sizeof(*dv
));
8825 dv
->dev
= xmalloc(len
);
8828 /* count how many spares will be converted to members */
8829 for (i
= 0; i
< map
->num_members
; i
++) {
8830 dl
= serial_to_dl(inf
[i
].serial
, super
);
8832 /* hmm maybe it failed?, nothing we can do about
8837 if (count_memberships(dl
, super
) == 0)
8840 len
+= activate
* sizeof(struct imsm_disk
);
8843 case update_kill_array
: {
8844 if (update
->len
< (int)sizeof(struct imsm_update_kill_array
))
8848 case update_rename_array
: {
8849 if (update
->len
< (int)sizeof(struct imsm_update_rename_array
))
8853 case update_add_remove_disk
:
8854 /* no update->len needed */
8860 /* check if we need a larger metadata buffer */
8861 if (super
->next_buf
)
8862 buf_len
= super
->next_len
;
8864 buf_len
= super
->len
;
8866 if (__le32_to_cpu(mpb
->mpb_size
) + len
> buf_len
) {
8867 /* ok we need a larger buf than what is currently allocated
8868 * if this allocation fails process_update will notice that
8869 * ->next_len is set and ->next_buf is NULL
8871 buf_len
= ROUND_UP(__le32_to_cpu(mpb
->mpb_size
) + len
, 512);
8872 if (super
->next_buf
)
8873 free(super
->next_buf
);
8875 super
->next_len
= buf_len
;
8876 if (posix_memalign(&super
->next_buf
, 512, buf_len
) == 0)
8877 memset(super
->next_buf
, 0, buf_len
);
8879 super
->next_buf
= NULL
;
8884 /* must be called while manager is quiesced */
8885 static void imsm_delete(struct intel_super
*super
, struct dl
**dlp
, unsigned index
)
8887 struct imsm_super
*mpb
= super
->anchor
;
8889 struct imsm_dev
*dev
;
8890 struct imsm_map
*map
;
8891 int i
, j
, num_members
;
8894 dprintf("deleting device[%d] from imsm_super\n", index
);
8896 /* shift all indexes down one */
8897 for (iter
= super
->disks
; iter
; iter
= iter
->next
)
8898 if (iter
->index
> (int)index
)
8900 for (iter
= super
->missing
; iter
; iter
= iter
->next
)
8901 if (iter
->index
> (int)index
)
8904 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
8905 dev
= get_imsm_dev(super
, i
);
8906 map
= get_imsm_map(dev
, MAP_0
);
8907 num_members
= map
->num_members
;
8908 for (j
= 0; j
< num_members
; j
++) {
8909 /* update ord entries being careful not to propagate
8910 * ord-flags to the first map
8912 ord
= get_imsm_ord_tbl_ent(dev
, j
, MAP_X
);
8914 if (ord_to_idx(ord
) <= index
)
8917 map
= get_imsm_map(dev
, MAP_0
);
8918 set_imsm_ord_tbl_ent(map
, j
, ord_to_idx(ord
- 1));
8919 map
= get_imsm_map(dev
, MAP_1
);
8921 set_imsm_ord_tbl_ent(map
, j
, ord
- 1);
8926 super
->updates_pending
++;
8928 struct dl
*dl
= *dlp
;
8930 *dlp
= (*dlp
)->next
;
8931 __free_imsm_disk(dl
);
8934 #endif /* MDASSEMBLE */
8936 static void close_targets(int *targets
, int new_disks
)
8943 for (i
= 0; i
< new_disks
; i
++) {
8944 if (targets
[i
] >= 0) {
8951 static int imsm_get_allowed_degradation(int level
, int raid_disks
,
8952 struct intel_super
*super
,
8953 struct imsm_dev
*dev
)
8959 struct imsm_map
*map
;
8962 ret_val
= raid_disks
/2;
8963 /* check map if all disks pairs not failed
8966 map
= get_imsm_map(dev
, MAP_0
);
8967 for (i
= 0; i
< ret_val
; i
++) {
8968 int degradation
= 0;
8969 if (get_imsm_disk(super
, i
) == NULL
)
8971 if (get_imsm_disk(super
, i
+ 1) == NULL
)
8973 if (degradation
== 2)
8976 map
= get_imsm_map(dev
, MAP_1
);
8977 /* if there is no second map
8978 * result can be returned
8982 /* check degradation in second map
8984 for (i
= 0; i
< ret_val
; i
++) {
8985 int degradation
= 0;
8986 if (get_imsm_disk(super
, i
) == NULL
)
8988 if (get_imsm_disk(super
, i
+ 1) == NULL
)
8990 if (degradation
== 2)
9004 /*******************************************************************************
9005 * Function: open_backup_targets
9006 * Description: Function opens file descriptors for all devices given in
9009 * info : general array info
9010 * raid_disks : number of disks
9011 * raid_fds : table of device's file descriptors
9012 * super : intel super for raid10 degradation check
9013 * dev : intel device for raid10 degradation check
9017 ******************************************************************************/
9018 int open_backup_targets(struct mdinfo
*info
, int raid_disks
, int *raid_fds
,
9019 struct intel_super
*super
, struct imsm_dev
*dev
)
9025 for (i
= 0; i
< raid_disks
; i
++)
9028 for (sd
= info
->devs
; sd
; sd
= sd
->next
) {
9031 if (sd
->disk
.state
& (1<<MD_DISK_FAULTY
)) {
9032 dprintf("disk is faulty!!\n");
9036 if ((sd
->disk
.raid_disk
>= raid_disks
) ||
9037 (sd
->disk
.raid_disk
< 0))
9040 dn
= map_dev(sd
->disk
.major
,
9042 raid_fds
[sd
->disk
.raid_disk
] = dev_open(dn
, O_RDWR
);
9043 if (raid_fds
[sd
->disk
.raid_disk
] < 0) {
9044 pr_err("cannot open component\n");
9049 /* check if maximum array degradation level is not exceeded
9051 if ((raid_disks
- opened
) >
9052 imsm_get_allowed_degradation(info
->new_level
,
9055 pr_err("Not enough disks can be opened.\n");
9056 close_targets(raid_fds
, raid_disks
);
9062 /*******************************************************************************
9063 * Function: validate_container_imsm
9064 * Description: This routine validates container after assemble,
9065 * eg. if devices in container are under the same controller.
9068 * info : linked list with info about devices used in array
9072 ******************************************************************************/
9073 int validate_container_imsm(struct mdinfo
*info
)
9075 if (check_env("IMSM_NO_PLATFORM"))
9078 struct sys_dev
*idev
;
9079 struct sys_dev
*hba
= NULL
;
9080 struct sys_dev
*intel_devices
= find_intel_devices();
9081 char *dev_path
= devt_to_devpath(makedev(info
->disk
.major
,
9084 for (idev
= intel_devices
; idev
; idev
= idev
->next
) {
9085 if (dev_path
&& strstr(dev_path
, idev
->path
)) {
9094 pr_err("WARNING - Cannot detect HBA for device %s!\n",
9095 devid2kname(makedev(info
->disk
.major
, info
->disk
.minor
)));
9099 const struct imsm_orom
*orom
= get_orom_by_device_id(hba
->dev_id
);
9102 for (dev
= info
->next
; dev
; dev
= dev
->next
) {
9103 dev_path
= devt_to_devpath(makedev(dev
->disk
.major
, dev
->disk
.minor
));
9105 struct sys_dev
*hba2
= NULL
;
9106 for (idev
= intel_devices
; idev
; idev
= idev
->next
) {
9107 if (dev_path
&& strstr(dev_path
, idev
->path
)) {
9115 const struct imsm_orom
*orom2
= hba2
== NULL
? NULL
:
9116 get_orom_by_device_id(hba2
->dev_id
);
9118 if (hba2
&& hba
->type
!= hba2
->type
) {
9119 pr_err("WARNING - HBAs of devices do not match %s != %s\n",
9120 get_sys_dev_type(hba
->type
), get_sys_dev_type(hba2
->type
));
9124 if (orom
!= orom2
) {
9125 pr_err("WARNING - IMSM container assembled with disks under different HBAs!\n"
9126 " This operation is not supported and can lead to data loss.\n");
9131 pr_err("WARNING - IMSM container assembled with disks under HBAs without IMSM platform support!\n"
9132 " This operation is not supported and can lead to data loss.\n");
9140 /*******************************************************************************
9141 * Function: init_migr_record_imsm
9142 * Description: Function inits imsm migration record
9144 * super : imsm internal array info
9145 * dev : device under migration
9146 * info : general array info to find the smallest device
9149 ******************************************************************************/
9150 void init_migr_record_imsm(struct supertype
*st
, struct imsm_dev
*dev
,
9151 struct mdinfo
*info
)
9153 struct intel_super
*super
= st
->sb
;
9154 struct migr_record
*migr_rec
= super
->migr_rec
;
9156 unsigned long long dsize
, dev_sectors
;
9157 long long unsigned min_dev_sectors
= -1LLU;
9161 struct imsm_map
*map_dest
= get_imsm_map(dev
, MAP_0
);
9162 struct imsm_map
*map_src
= get_imsm_map(dev
, MAP_1
);
9163 unsigned long long num_migr_units
;
9164 unsigned long long array_blocks
;
9166 memset(migr_rec
, 0, sizeof(struct migr_record
));
9167 migr_rec
->family_num
= __cpu_to_le32(super
->anchor
->family_num
);
9169 /* only ascending reshape supported now */
9170 migr_rec
->ascending_migr
= __cpu_to_le32(1);
9172 migr_rec
->dest_depth_per_unit
= GEN_MIGR_AREA_SIZE
/
9173 max(map_dest
->blocks_per_strip
, map_src
->blocks_per_strip
);
9174 migr_rec
->dest_depth_per_unit
*=
9175 max(map_dest
->blocks_per_strip
, map_src
->blocks_per_strip
);
9176 new_data_disks
= imsm_num_data_members(dev
, MAP_0
);
9177 migr_rec
->blocks_per_unit
=
9178 __cpu_to_le32(migr_rec
->dest_depth_per_unit
* new_data_disks
);
9179 migr_rec
->dest_depth_per_unit
=
9180 __cpu_to_le32(migr_rec
->dest_depth_per_unit
);
9181 array_blocks
= info
->component_size
* new_data_disks
;
9183 array_blocks
/ __le32_to_cpu(migr_rec
->blocks_per_unit
);
9185 if (array_blocks
% __le32_to_cpu(migr_rec
->blocks_per_unit
))
9187 migr_rec
->num_migr_units
= __cpu_to_le32(num_migr_units
);
9189 migr_rec
->post_migr_vol_cap
= dev
->size_low
;
9190 migr_rec
->post_migr_vol_cap_hi
= dev
->size_high
;
9192 /* Find the smallest dev */
9193 for (sd
= info
->devs
; sd
; sd
= sd
->next
) {
9194 sprintf(nm
, "%d:%d", sd
->disk
.major
, sd
->disk
.minor
);
9195 fd
= dev_open(nm
, O_RDONLY
);
9198 get_dev_size(fd
, NULL
, &dsize
);
9199 dev_sectors
= dsize
/ 512;
9200 if (dev_sectors
< min_dev_sectors
)
9201 min_dev_sectors
= dev_sectors
;
9204 migr_rec
->ckpt_area_pba
= __cpu_to_le32(min_dev_sectors
-
9205 RAID_DISK_RESERVED_BLOCKS_IMSM_HI
);
9207 write_imsm_migr_rec(st
);
9212 /*******************************************************************************
9213 * Function: save_backup_imsm
9214 * Description: Function saves critical data stripes to Migration Copy Area
9215 * and updates the current migration unit status.
9216 * Use restore_stripes() to form a destination stripe,
9217 * and to write it to the Copy Area.
9219 * st : supertype information
9220 * dev : imsm device that backup is saved for
9221 * info : general array info
9222 * buf : input buffer
9223 * length : length of data to backup (blocks_per_unit)
9227 ******************************************************************************/
9228 int save_backup_imsm(struct supertype
*st
,
9229 struct imsm_dev
*dev
,
9230 struct mdinfo
*info
,
9235 struct intel_super
*super
= st
->sb
;
9236 unsigned long long *target_offsets
= NULL
;
9237 int *targets
= NULL
;
9239 struct imsm_map
*map_dest
= get_imsm_map(dev
, MAP_0
);
9240 int new_disks
= map_dest
->num_members
;
9241 int dest_layout
= 0;
9243 unsigned long long start
;
9244 int data_disks
= imsm_num_data_members(dev
, MAP_0
);
9246 targets
= xmalloc(new_disks
* sizeof(int));
9248 for (i
= 0; i
< new_disks
; i
++)
9251 target_offsets
= xcalloc(new_disks
, sizeof(unsigned long long));
9253 start
= info
->reshape_progress
* 512;
9254 for (i
= 0; i
< new_disks
; i
++) {
9255 target_offsets
[i
] = (unsigned long long)
9256 __le32_to_cpu(super
->migr_rec
->ckpt_area_pba
) * 512;
9257 /* move back copy area adderss, it will be moved forward
9258 * in restore_stripes() using start input variable
9260 target_offsets
[i
] -= start
/data_disks
;
9263 if (open_backup_targets(info
, new_disks
, targets
,
9267 dest_layout
= imsm_level_to_layout(map_dest
->raid_level
);
9268 dest_chunk
= __le16_to_cpu(map_dest
->blocks_per_strip
) * 512;
9270 if (restore_stripes(targets
, /* list of dest devices */
9271 target_offsets
, /* migration record offsets */
9274 map_dest
->raid_level
,
9276 -1, /* source backup file descriptor */
9277 0, /* input buf offset
9278 * always 0 buf is already offseted */
9282 pr_err("Error restoring stripes\n");
9290 close_targets(targets
, new_disks
);
9293 free(target_offsets
);
9298 /*******************************************************************************
9299 * Function: save_checkpoint_imsm
9300 * Description: Function called for current unit status update
9301 * in the migration record. It writes it to disk.
9303 * super : imsm internal array info
9304 * info : general array info
9308 * 2: failure, means no valid migration record
9309 * / no general migration in progress /
9310 ******************************************************************************/
9311 int save_checkpoint_imsm(struct supertype
*st
, struct mdinfo
*info
, int state
)
9313 struct intel_super
*super
= st
->sb
;
9314 unsigned long long blocks_per_unit
;
9315 unsigned long long curr_migr_unit
;
9317 if (load_imsm_migr_rec(super
, info
) != 0) {
9318 dprintf("imsm: ERROR: Cannot read migration record for checkpoint save.\n");
9322 blocks_per_unit
= __le32_to_cpu(super
->migr_rec
->blocks_per_unit
);
9323 if (blocks_per_unit
== 0) {
9324 dprintf("imsm: no migration in progress.\n");
9327 curr_migr_unit
= info
->reshape_progress
/ blocks_per_unit
;
9328 /* check if array is alligned to copy area
9329 * if it is not alligned, add one to current migration unit value
9330 * this can happend on array reshape finish only
9332 if (info
->reshape_progress
% blocks_per_unit
)
9335 super
->migr_rec
->curr_migr_unit
=
9336 __cpu_to_le32(curr_migr_unit
);
9337 super
->migr_rec
->rec_status
= __cpu_to_le32(state
);
9338 super
->migr_rec
->dest_1st_member_lba
=
9339 __cpu_to_le32(curr_migr_unit
*
9340 __le32_to_cpu(super
->migr_rec
->dest_depth_per_unit
));
9341 if (write_imsm_migr_rec(st
) < 0) {
9342 dprintf("imsm: Cannot write migration record outside backup area\n");
9349 /*******************************************************************************
9350 * Function: recover_backup_imsm
9351 * Description: Function recovers critical data from the Migration Copy Area
9352 * while assembling an array.
9354 * super : imsm internal array info
9355 * info : general array info
9357 * 0 : success (or there is no data to recover)
9359 ******************************************************************************/
9360 int recover_backup_imsm(struct supertype
*st
, struct mdinfo
*info
)
9362 struct intel_super
*super
= st
->sb
;
9363 struct migr_record
*migr_rec
= super
->migr_rec
;
9364 struct imsm_map
*map_dest
= NULL
;
9365 struct intel_dev
*id
= NULL
;
9366 unsigned long long read_offset
;
9367 unsigned long long write_offset
;
9369 int *targets
= NULL
;
9370 int new_disks
, i
, err
;
9373 unsigned long curr_migr_unit
= __le32_to_cpu(migr_rec
->curr_migr_unit
);
9374 unsigned long num_migr_units
= __le32_to_cpu(migr_rec
->num_migr_units
);
9376 int skipped_disks
= 0;
9378 err
= sysfs_get_str(info
, NULL
, "array_state", (char *)buffer
, 20);
9382 /* recover data only during assemblation */
9383 if (strncmp(buffer
, "inactive", 8) != 0)
9385 /* no data to recover */
9386 if (__le32_to_cpu(migr_rec
->rec_status
) == UNIT_SRC_NORMAL
)
9388 if (curr_migr_unit
>= num_migr_units
)
9391 /* find device during reshape */
9392 for (id
= super
->devlist
; id
; id
= id
->next
)
9393 if (is_gen_migration(id
->dev
))
9398 map_dest
= get_imsm_map(id
->dev
, MAP_0
);
9399 new_disks
= map_dest
->num_members
;
9401 read_offset
= (unsigned long long)
9402 __le32_to_cpu(migr_rec
->ckpt_area_pba
) * 512;
9404 write_offset
= ((unsigned long long)
9405 __le32_to_cpu(migr_rec
->dest_1st_member_lba
) +
9406 pba_of_lba0(map_dest
)) * 512;
9408 unit_len
= __le32_to_cpu(migr_rec
->dest_depth_per_unit
) * 512;
9409 if (posix_memalign((void **)&buf
, 512, unit_len
) != 0)
9411 targets
= xcalloc(new_disks
, sizeof(int));
9413 if (open_backup_targets(info
, new_disks
, targets
, super
, id
->dev
)) {
9414 pr_err("Cannot open some devices belonging to array.\n");
9418 for (i
= 0; i
< new_disks
; i
++) {
9419 if (targets
[i
] < 0) {
9423 if (lseek64(targets
[i
], read_offset
, SEEK_SET
) < 0) {
9424 pr_err("Cannot seek to block: %s\n",
9429 if ((unsigned)read(targets
[i
], buf
, unit_len
) != unit_len
) {
9430 pr_err("Cannot read copy area block: %s\n",
9435 if (lseek64(targets
[i
], write_offset
, SEEK_SET
) < 0) {
9436 pr_err("Cannot seek to block: %s\n",
9441 if ((unsigned)write(targets
[i
], buf
, unit_len
) != unit_len
) {
9442 pr_err("Cannot restore block: %s\n",
9449 if (skipped_disks
> imsm_get_allowed_degradation(info
->new_level
,
9453 pr_err("Cannot restore data from backup. Too many failed disks\n");
9457 if (save_checkpoint_imsm(st
, info
, UNIT_SRC_NORMAL
)) {
9458 /* ignore error == 2, this can mean end of reshape here
9460 dprintf("imsm: Cannot write checkpoint to migration record (UNIT_SRC_NORMAL) during restart\n");
9466 for (i
= 0; i
< new_disks
; i
++)
9475 static char disk_by_path
[] = "/dev/disk/by-path/";
9477 static const char *imsm_get_disk_controller_domain(const char *path
)
9479 char disk_path
[PATH_MAX
];
9483 strcpy(disk_path
, disk_by_path
);
9484 strncat(disk_path
, path
, PATH_MAX
- strlen(disk_path
) - 1);
9485 if (stat(disk_path
, &st
) == 0) {
9486 struct sys_dev
* hba
;
9489 path
= devt_to_devpath(st
.st_rdev
);
9492 hba
= find_disk_attached_hba(-1, path
);
9493 if (hba
&& hba
->type
== SYS_DEV_SAS
)
9495 else if (hba
&& hba
->type
== SYS_DEV_SATA
)
9499 dprintf("path: %s hba: %s attached: %s\n",
9500 path
, (hba
) ? hba
->path
: "NULL", drv
);
9506 static char *imsm_find_array_devnm_by_subdev(int subdev
, char *container
)
9508 static char devnm
[32];
9509 char subdev_name
[20];
9510 struct mdstat_ent
*mdstat
;
9512 sprintf(subdev_name
, "%d", subdev
);
9513 mdstat
= mdstat_by_subdev(subdev_name
, container
);
9517 strcpy(devnm
, mdstat
->devnm
);
9518 free_mdstat(mdstat
);
9522 static int imsm_reshape_is_allowed_on_container(struct supertype
*st
,
9523 struct geo_params
*geo
,
9524 int *old_raid_disks
,
9527 /* currently we only support increasing the number of devices
9528 * for a container. This increases the number of device for each
9529 * member array. They must all be RAID0 or RAID5.
9532 struct mdinfo
*info
, *member
;
9533 int devices_that_can_grow
= 0;
9535 dprintf("imsm: imsm_reshape_is_allowed_on_container(ENTER): st->devnm = (%s)\n", st
->devnm
);
9537 if (geo
->size
> 0 ||
9538 geo
->level
!= UnSet
||
9539 geo
->layout
!= UnSet
||
9540 geo
->chunksize
!= 0 ||
9541 geo
->raid_disks
== UnSet
) {
9542 dprintf("imsm: Container operation is allowed for raid disks number change only.\n");
9546 if (direction
== ROLLBACK_METADATA_CHANGES
) {
9547 dprintf("imsm: Metadata changes rollback is not supported for container operation.\n");
9551 info
= container_content_imsm(st
, NULL
);
9552 for (member
= info
; member
; member
= member
->next
) {
9555 dprintf("imsm: checking device_num: %i\n",
9556 member
->container_member
);
9558 if (geo
->raid_disks
<= member
->array
.raid_disks
) {
9559 /* we work on container for Online Capacity Expansion
9560 * only so raid_disks has to grow
9562 dprintf("imsm: for container operation raid disks increase is required\n");
9566 if ((info
->array
.level
!= 0) &&
9567 (info
->array
.level
!= 5)) {
9568 /* we cannot use this container with other raid level
9570 dprintf("imsm: for container operation wrong raid level (%i) detected\n",
9574 /* check for platform support
9575 * for this raid level configuration
9577 struct intel_super
*super
= st
->sb
;
9578 if (!is_raid_level_supported(super
->orom
,
9579 member
->array
.level
,
9581 dprintf("platform does not support raid%d with %d disk%s\n",
9584 geo
->raid_disks
> 1 ? "s" : "");
9587 /* check if component size is aligned to chunk size
9589 if (info
->component_size
%
9590 (info
->array
.chunk_size
/512)) {
9591 dprintf("Component size is not aligned to chunk size\n");
9596 if (*old_raid_disks
&&
9597 info
->array
.raid_disks
!= *old_raid_disks
)
9599 *old_raid_disks
= info
->array
.raid_disks
;
9601 /* All raid5 and raid0 volumes in container
9602 * have to be ready for Online Capacity Expansion
9603 * so they need to be assembled. We have already
9604 * checked that no recovery etc is happening.
9606 result
= imsm_find_array_devnm_by_subdev(member
->container_member
,
9607 st
->container_devnm
);
9608 if (result
== NULL
) {
9609 dprintf("imsm: cannot find array\n");
9612 devices_that_can_grow
++;
9615 if (!member
&& devices_that_can_grow
)
9619 dprintf("Container operation allowed\n");
9621 dprintf("Error: %i\n", ret_val
);
9626 /* Function: get_spares_for_grow
9627 * Description: Allocates memory and creates list of spare devices
9628 * avaliable in container. Checks if spare drive size is acceptable.
9629 * Parameters: Pointer to the supertype structure
9630 * Returns: Pointer to the list of spare devices (mdinfo structure) on success,
9633 static struct mdinfo
*get_spares_for_grow(struct supertype
*st
)
9635 unsigned long long min_size
= min_acceptable_spare_size_imsm(st
);
9636 return container_choose_spares(st
, min_size
, NULL
, NULL
, NULL
, 0);
9639 /******************************************************************************
9640 * function: imsm_create_metadata_update_for_reshape
9641 * Function creates update for whole IMSM container.
9643 ******************************************************************************/
9644 static int imsm_create_metadata_update_for_reshape(
9645 struct supertype
*st
,
9646 struct geo_params
*geo
,
9648 struct imsm_update_reshape
**updatep
)
9650 struct intel_super
*super
= st
->sb
;
9651 struct imsm_super
*mpb
= super
->anchor
;
9652 int update_memory_size
= 0;
9653 struct imsm_update_reshape
*u
= NULL
;
9654 struct mdinfo
*spares
= NULL
;
9656 int delta_disks
= 0;
9659 dprintf("(enter) raid_disks = %i\n", geo
->raid_disks
);
9661 delta_disks
= geo
->raid_disks
- old_raid_disks
;
9663 /* size of all update data without anchor */
9664 update_memory_size
= sizeof(struct imsm_update_reshape
);
9666 /* now add space for spare disks that we need to add. */
9667 update_memory_size
+= sizeof(u
->new_disks
[0]) * (delta_disks
- 1);
9669 u
= xcalloc(1, update_memory_size
);
9670 u
->type
= update_reshape_container_disks
;
9671 u
->old_raid_disks
= old_raid_disks
;
9672 u
->new_raid_disks
= geo
->raid_disks
;
9674 /* now get spare disks list
9676 spares
= get_spares_for_grow(st
);
9679 || delta_disks
> spares
->array
.spare_disks
) {
9680 pr_err("imsm: ERROR: Cannot get spare devices for %s.\n", geo
->dev_name
);
9685 /* we have got spares
9686 * update disk list in imsm_disk list table in anchor
9688 dprintf("imsm: %i spares are available.\n\n",
9689 spares
->array
.spare_disks
);
9692 for (i
= 0; i
< delta_disks
; i
++) {
9697 u
->new_disks
[i
] = makedev(dev
->disk
.major
,
9699 dl
= get_disk_super(super
, dev
->disk
.major
, dev
->disk
.minor
);
9700 dl
->index
= mpb
->num_disks
;
9710 dprintf("imsm: reshape update preparation :");
9711 if (i
== delta_disks
) {
9712 dprintf_cont(" OK\n");
9714 return update_memory_size
;
9717 dprintf_cont(" Error\n");
9722 /******************************************************************************
9723 * function: imsm_create_metadata_update_for_size_change()
9724 * Creates update for IMSM array for array size change.
9726 ******************************************************************************/
9727 static int imsm_create_metadata_update_for_size_change(
9728 struct supertype
*st
,
9729 struct geo_params
*geo
,
9730 struct imsm_update_size_change
**updatep
)
9732 struct intel_super
*super
= st
->sb
;
9733 int update_memory_size
= 0;
9734 struct imsm_update_size_change
*u
= NULL
;
9736 dprintf("(enter) New size = %llu\n", geo
->size
);
9738 /* size of all update data without anchor */
9739 update_memory_size
= sizeof(struct imsm_update_size_change
);
9741 u
= xcalloc(1, update_memory_size
);
9742 u
->type
= update_size_change
;
9743 u
->subdev
= super
->current_vol
;
9744 u
->new_size
= geo
->size
;
9746 dprintf("imsm: reshape update preparation : OK\n");
9749 return update_memory_size
;
9752 /******************************************************************************
9753 * function: imsm_create_metadata_update_for_migration()
9754 * Creates update for IMSM array.
9756 ******************************************************************************/
9757 static int imsm_create_metadata_update_for_migration(
9758 struct supertype
*st
,
9759 struct geo_params
*geo
,
9760 struct imsm_update_reshape_migration
**updatep
)
9762 struct intel_super
*super
= st
->sb
;
9763 int update_memory_size
= 0;
9764 struct imsm_update_reshape_migration
*u
= NULL
;
9765 struct imsm_dev
*dev
;
9766 int previous_level
= -1;
9768 dprintf("(enter) New Level = %i\n", geo
->level
);
9770 /* size of all update data without anchor */
9771 update_memory_size
= sizeof(struct imsm_update_reshape_migration
);
9773 u
= xcalloc(1, update_memory_size
);
9774 u
->type
= update_reshape_migration
;
9775 u
->subdev
= super
->current_vol
;
9776 u
->new_level
= geo
->level
;
9777 u
->new_layout
= geo
->layout
;
9778 u
->new_raid_disks
= u
->old_raid_disks
= geo
->raid_disks
;
9779 u
->new_disks
[0] = -1;
9780 u
->new_chunksize
= -1;
9782 dev
= get_imsm_dev(super
, u
->subdev
);
9784 struct imsm_map
*map
;
9786 map
= get_imsm_map(dev
, MAP_0
);
9788 int current_chunk_size
=
9789 __le16_to_cpu(map
->blocks_per_strip
) / 2;
9791 if (geo
->chunksize
!= current_chunk_size
) {
9792 u
->new_chunksize
= geo
->chunksize
/ 1024;
9793 dprintf("imsm: chunk size change from %i to %i\n",
9794 current_chunk_size
, u
->new_chunksize
);
9796 previous_level
= map
->raid_level
;
9799 if ((geo
->level
== 5) && (previous_level
== 0)) {
9800 struct mdinfo
*spares
= NULL
;
9802 u
->new_raid_disks
++;
9803 spares
= get_spares_for_grow(st
);
9804 if ((spares
== NULL
) || (spares
->array
.spare_disks
< 1)) {
9807 update_memory_size
= 0;
9808 dprintf("error: cannot get spare device for requested migration");
9813 dprintf("imsm: reshape update preparation : OK\n");
9816 return update_memory_size
;
9819 static void imsm_update_metadata_locally(struct supertype
*st
,
9822 struct metadata_update mu
;
9827 mu
.space_list
= NULL
;
9829 if (imsm_prepare_update(st
, &mu
))
9830 imsm_process_update(st
, &mu
);
9832 while (mu
.space_list
) {
9833 void **space
= mu
.space_list
;
9834 mu
.space_list
= *space
;
9839 /***************************************************************************
9840 * Function: imsm_analyze_change
9841 * Description: Function analyze change for single volume
9842 * and validate if transition is supported
9843 * Parameters: Geometry parameters, supertype structure,
9844 * metadata change direction (apply/rollback)
9845 * Returns: Operation type code on success, -1 if fail
9846 ****************************************************************************/
9847 enum imsm_reshape_type
imsm_analyze_change(struct supertype
*st
,
9848 struct geo_params
*geo
,
9855 /* number of added/removed disks in operation result */
9856 int devNumChange
= 0;
9857 /* imsm compatible layout value for array geometry verification */
9858 int imsm_layout
= -1;
9860 struct imsm_dev
*dev
;
9861 struct intel_super
*super
;
9862 unsigned long long current_size
;
9863 unsigned long long free_size
;
9864 unsigned long long max_size
;
9867 getinfo_super_imsm_volume(st
, &info
, NULL
);
9868 if ((geo
->level
!= info
.array
.level
) &&
9869 (geo
->level
>= 0) &&
9870 (geo
->level
!= UnSet
)) {
9871 switch (info
.array
.level
) {
9873 if (geo
->level
== 5) {
9874 change
= CH_MIGRATION
;
9875 if (geo
->layout
!= ALGORITHM_LEFT_ASYMMETRIC
) {
9876 pr_err("Error. Requested Layout not supported (left-asymmetric layout is supported only)!\n");
9878 goto analyse_change_exit
;
9880 imsm_layout
= geo
->layout
;
9882 devNumChange
= 1; /* parity disk added */
9883 } else if (geo
->level
== 10) {
9884 change
= CH_TAKEOVER
;
9886 devNumChange
= 2; /* two mirrors added */
9887 imsm_layout
= 0x102; /* imsm supported layout */
9892 if (geo
->level
== 0) {
9893 change
= CH_TAKEOVER
;
9895 devNumChange
= -(geo
->raid_disks
/2);
9896 imsm_layout
= 0; /* imsm raid0 layout */
9901 pr_err("Error. Level Migration from %d to %d not supported!\n",
9902 info
.array
.level
, geo
->level
);
9903 goto analyse_change_exit
;
9906 geo
->level
= info
.array
.level
;
9908 if ((geo
->layout
!= info
.array
.layout
)
9909 && ((geo
->layout
!= UnSet
) && (geo
->layout
!= -1))) {
9910 change
= CH_MIGRATION
;
9911 if ((info
.array
.layout
== 0)
9912 && (info
.array
.level
== 5)
9913 && (geo
->layout
== 5)) {
9914 /* reshape 5 -> 4 */
9915 } else if ((info
.array
.layout
== 5)
9916 && (info
.array
.level
== 5)
9917 && (geo
->layout
== 0)) {
9918 /* reshape 4 -> 5 */
9922 pr_err("Error. Layout Migration from %d to %d not supported!\n",
9923 info
.array
.layout
, geo
->layout
);
9925 goto analyse_change_exit
;
9928 geo
->layout
= info
.array
.layout
;
9929 if (imsm_layout
== -1)
9930 imsm_layout
= info
.array
.layout
;
9933 if ((geo
->chunksize
> 0) && (geo
->chunksize
!= UnSet
)
9934 && (geo
->chunksize
!= info
.array
.chunk_size
))
9935 change
= CH_MIGRATION
;
9937 geo
->chunksize
= info
.array
.chunk_size
;
9939 chunk
= geo
->chunksize
/ 1024;
9942 dev
= get_imsm_dev(super
, super
->current_vol
);
9943 data_disks
= imsm_num_data_members(dev
, MAP_0
);
9944 /* compute current size per disk member
9946 current_size
= info
.custom_array_size
/ data_disks
;
9948 if ((geo
->size
> 0) && (geo
->size
!= MAX_SIZE
)) {
9949 /* align component size
9951 geo
->size
= imsm_component_size_aligment_check(
9952 get_imsm_raid_level(dev
->vol
.map
),
9955 if (geo
->size
== 0) {
9956 pr_err("Error. Size expansion is supported only (current size is %llu, requested size /rounded/ is 0).\n",
9958 goto analyse_change_exit
;
9962 if ((current_size
!= geo
->size
) && (geo
->size
> 0)) {
9964 pr_err("Error. Size change should be the only one at a time.\n");
9966 goto analyse_change_exit
;
9968 if ((super
->current_vol
+ 1) != super
->anchor
->num_raid_devs
) {
9969 pr_err("Error. The last volume in container can be expanded only (%i/%s).\n",
9970 super
->current_vol
, st
->devnm
);
9971 goto analyse_change_exit
;
9973 /* check the maximum available size
9975 rv
= imsm_get_free_size(st
, dev
->vol
.map
->num_members
,
9976 0, chunk
, &free_size
);
9978 /* Cannot find maximum available space
9982 max_size
= free_size
+ current_size
;
9983 /* align component size
9985 max_size
= imsm_component_size_aligment_check(
9986 get_imsm_raid_level(dev
->vol
.map
),
9990 if (geo
->size
== MAX_SIZE
) {
9991 /* requested size change to the maximum available size
9993 if (max_size
== 0) {
9994 pr_err("Error. Cannot find maximum available space.\n");
9996 goto analyse_change_exit
;
9998 geo
->size
= max_size
;
10001 if ((direction
== ROLLBACK_METADATA_CHANGES
)) {
10002 /* accept size for rollback only
10005 /* round size due to metadata compatibility
10007 geo
->size
= (geo
->size
>> SECT_PER_MB_SHIFT
)
10008 << SECT_PER_MB_SHIFT
;
10009 dprintf("Prepare update for size change to %llu\n",
10011 if (current_size
>= geo
->size
) {
10012 pr_err("Error. Size expansion is supported only (current size is %llu, requested size /rounded/ is %llu).\n",
10013 current_size
, geo
->size
);
10014 goto analyse_change_exit
;
10016 if (max_size
&& geo
->size
> max_size
) {
10017 pr_err("Error. Requested size is larger than maximum available size (maximum available size is %llu, requested size /rounded/ is %llu).\n",
10018 max_size
, geo
->size
);
10019 goto analyse_change_exit
;
10022 geo
->size
*= data_disks
;
10023 geo
->raid_disks
= dev
->vol
.map
->num_members
;
10024 change
= CH_ARRAY_SIZE
;
10026 if (!validate_geometry_imsm(st
,
10029 geo
->raid_disks
+ devNumChange
,
10031 geo
->size
, INVALID_SECTORS
,
10036 struct intel_super
*super
= st
->sb
;
10037 struct imsm_super
*mpb
= super
->anchor
;
10039 if (mpb
->num_raid_devs
> 1) {
10040 pr_err("Error. Cannot perform operation on %s- for this operation it MUST be single array in container\n",
10046 analyse_change_exit
:
10047 if ((direction
== ROLLBACK_METADATA_CHANGES
) &&
10048 ((change
== CH_MIGRATION
) || (change
== CH_TAKEOVER
))) {
10049 dprintf("imsm: Metadata changes rollback is not supported for migration and takeover operations.\n");
10055 int imsm_takeover(struct supertype
*st
, struct geo_params
*geo
)
10057 struct intel_super
*super
= st
->sb
;
10058 struct imsm_update_takeover
*u
;
10060 u
= xmalloc(sizeof(struct imsm_update_takeover
));
10062 u
->type
= update_takeover
;
10063 u
->subarray
= super
->current_vol
;
10065 /* 10->0 transition */
10066 if (geo
->level
== 0)
10067 u
->direction
= R10_TO_R0
;
10069 /* 0->10 transition */
10070 if (geo
->level
== 10)
10071 u
->direction
= R0_TO_R10
;
10073 /* update metadata locally */
10074 imsm_update_metadata_locally(st
, u
,
10075 sizeof(struct imsm_update_takeover
));
10076 /* and possibly remotely */
10077 if (st
->update_tail
)
10078 append_metadata_update(st
, u
,
10079 sizeof(struct imsm_update_takeover
));
10086 static int imsm_reshape_super(struct supertype
*st
, unsigned long long size
,
10088 int layout
, int chunksize
, int raid_disks
,
10089 int delta_disks
, char *backup
, char *dev
,
10090 int direction
, int verbose
)
10093 struct geo_params geo
;
10095 dprintf("(enter)\n");
10097 memset(&geo
, 0, sizeof(struct geo_params
));
10099 geo
.dev_name
= dev
;
10100 strcpy(geo
.devnm
, st
->devnm
);
10103 geo
.layout
= layout
;
10104 geo
.chunksize
= chunksize
;
10105 geo
.raid_disks
= raid_disks
;
10106 if (delta_disks
!= UnSet
)
10107 geo
.raid_disks
+= delta_disks
;
10109 dprintf("for level : %i\n", geo
.level
);
10110 dprintf("for raid_disks : %i\n", geo
.raid_disks
);
10112 if (experimental() == 0)
10115 if (strcmp(st
->container_devnm
, st
->devnm
) == 0) {
10116 /* On container level we can only increase number of devices. */
10117 dprintf("imsm: info: Container operation\n");
10118 int old_raid_disks
= 0;
10120 if (imsm_reshape_is_allowed_on_container(
10121 st
, &geo
, &old_raid_disks
, direction
)) {
10122 struct imsm_update_reshape
*u
= NULL
;
10125 len
= imsm_create_metadata_update_for_reshape(
10126 st
, &geo
, old_raid_disks
, &u
);
10129 dprintf("imsm: Cannot prepare update\n");
10130 goto exit_imsm_reshape_super
;
10134 /* update metadata locally */
10135 imsm_update_metadata_locally(st
, u
, len
);
10136 /* and possibly remotely */
10137 if (st
->update_tail
)
10138 append_metadata_update(st
, u
, len
);
10143 pr_err("(imsm) Operation is not allowed on this container\n");
10146 /* On volume level we support following operations
10147 * - takeover: raid10 -> raid0; raid0 -> raid10
10148 * - chunk size migration
10149 * - migration: raid5 -> raid0; raid0 -> raid5
10151 struct intel_super
*super
= st
->sb
;
10152 struct intel_dev
*dev
= super
->devlist
;
10154 dprintf("imsm: info: Volume operation\n");
10155 /* find requested device */
10158 imsm_find_array_devnm_by_subdev(
10159 dev
->index
, st
->container_devnm
);
10160 if (devnm
&& strcmp(devnm
, geo
.devnm
) == 0)
10165 pr_err("Cannot find %s (%s) subarray\n",
10166 geo
.dev_name
, geo
.devnm
);
10167 goto exit_imsm_reshape_super
;
10169 super
->current_vol
= dev
->index
;
10170 change
= imsm_analyze_change(st
, &geo
, direction
);
10173 ret_val
= imsm_takeover(st
, &geo
);
10175 case CH_MIGRATION
: {
10176 struct imsm_update_reshape_migration
*u
= NULL
;
10178 imsm_create_metadata_update_for_migration(
10181 dprintf("imsm: Cannot prepare update\n");
10185 /* update metadata locally */
10186 imsm_update_metadata_locally(st
, u
, len
);
10187 /* and possibly remotely */
10188 if (st
->update_tail
)
10189 append_metadata_update(st
, u
, len
);
10194 case CH_ARRAY_SIZE
: {
10195 struct imsm_update_size_change
*u
= NULL
;
10197 imsm_create_metadata_update_for_size_change(
10200 dprintf("imsm: Cannot prepare update\n");
10204 /* update metadata locally */
10205 imsm_update_metadata_locally(st
, u
, len
);
10206 /* and possibly remotely */
10207 if (st
->update_tail
)
10208 append_metadata_update(st
, u
, len
);
10218 exit_imsm_reshape_super
:
10219 dprintf("imsm: reshape_super Exit code = %i\n", ret_val
);
10223 /*******************************************************************************
10224 * Function: wait_for_reshape_imsm
10225 * Description: Function writes new sync_max value and waits until
10226 * reshape process reach new position
10228 * sra : general array info
10229 * ndata : number of disks in new array's layout
10232 * 1 : there is no reshape in progress,
10234 ******************************************************************************/
10235 int wait_for_reshape_imsm(struct mdinfo
*sra
, int ndata
)
10237 int fd
= sysfs_get_fd(sra
, NULL
, "sync_completed");
10238 unsigned long long completed
;
10239 /* to_complete : new sync_max position */
10240 unsigned long long to_complete
= sra
->reshape_progress
;
10241 unsigned long long position_to_set
= to_complete
/ ndata
;
10244 dprintf("cannot open reshape_position\n");
10248 if (sysfs_fd_get_ll(fd
, &completed
) < 0) {
10249 dprintf("cannot read reshape_position (no reshape in progres)\n");
10254 if (completed
> position_to_set
) {
10255 dprintf("wrong next position to set %llu (%llu)\n",
10256 to_complete
, position_to_set
);
10260 dprintf("Position set: %llu\n", position_to_set
);
10261 if (sysfs_set_num(sra
, NULL
, "sync_max",
10262 position_to_set
) != 0) {
10263 dprintf("cannot set reshape position to %llu\n",
10271 sysfs_wait(fd
, NULL
);
10272 if (sysfs_get_str(sra
, NULL
, "sync_action",
10274 strncmp(action
, "reshape", 7) != 0)
10276 if (sysfs_fd_get_ll(fd
, &completed
) < 0) {
10277 dprintf("cannot read reshape_position (in loop)\n");
10281 } while (completed
< position_to_set
);
10287 /*******************************************************************************
10288 * Function: check_degradation_change
10289 * Description: Check that array hasn't become failed.
10291 * info : for sysfs access
10292 * sources : source disks descriptors
10293 * degraded: previous degradation level
10295 * degradation level
10296 ******************************************************************************/
10297 int check_degradation_change(struct mdinfo
*info
,
10301 unsigned long long new_degraded
;
10304 rv
= sysfs_get_ll(info
, NULL
, "degraded", &new_degraded
);
10305 if ((rv
== -1) || (new_degraded
!= (unsigned long long)degraded
)) {
10306 /* check each device to ensure it is still working */
10309 for (sd
= info
->devs
; sd
; sd
= sd
->next
) {
10310 if (sd
->disk
.state
& (1<<MD_DISK_FAULTY
))
10312 if (sd
->disk
.state
& (1<<MD_DISK_SYNC
)) {
10314 if (sysfs_get_str(info
,
10315 sd
, "state", sbuf
, 20) < 0 ||
10316 strstr(sbuf
, "faulty") ||
10317 strstr(sbuf
, "in_sync") == NULL
) {
10318 /* this device is dead */
10319 sd
->disk
.state
= (1<<MD_DISK_FAULTY
);
10320 if (sd
->disk
.raid_disk
>= 0 &&
10321 sources
[sd
->disk
.raid_disk
] >= 0) {
10323 sd
->disk
.raid_disk
]);
10324 sources
[sd
->disk
.raid_disk
] =
10333 return new_degraded
;
10336 /*******************************************************************************
10337 * Function: imsm_manage_reshape
10338 * Description: Function finds array under reshape and it manages reshape
10339 * process. It creates stripes backups (if required) and sets
10342 * afd : Backup handle (nattive) - not used
10343 * sra : general array info
10344 * reshape : reshape parameters - not used
10345 * st : supertype structure
10346 * blocks : size of critical section [blocks]
10347 * fds : table of source device descriptor
10348 * offsets : start of array (offest per devices)
10350 * destfd : table of destination device descriptor
10351 * destoffsets : table of destination offsets (per device)
10353 * 1 : success, reshape is done
10355 ******************************************************************************/
10356 static int imsm_manage_reshape(
10357 int afd
, struct mdinfo
*sra
, struct reshape
*reshape
,
10358 struct supertype
*st
, unsigned long backup_blocks
,
10359 int *fds
, unsigned long long *offsets
,
10360 int dests
, int *destfd
, unsigned long long *destoffsets
)
10363 struct intel_super
*super
= st
->sb
;
10364 struct intel_dev
*dv
= NULL
;
10365 struct imsm_dev
*dev
= NULL
;
10366 struct imsm_map
*map_src
;
10367 int migr_vol_qan
= 0;
10368 int ndata
, odata
; /* [bytes] */
10369 int chunk
; /* [bytes] */
10370 struct migr_record
*migr_rec
;
10372 unsigned int buf_size
; /* [bytes] */
10373 unsigned long long max_position
; /* array size [bytes] */
10374 unsigned long long next_step
; /* [blocks]/[bytes] */
10375 unsigned long long old_data_stripe_length
;
10376 unsigned long long start_src
; /* [bytes] */
10377 unsigned long long start
; /* [bytes] */
10378 unsigned long long start_buf_shift
; /* [bytes] */
10380 int source_layout
= 0;
10382 if (!fds
|| !offsets
|| !sra
)
10385 /* Find volume during the reshape */
10386 for (dv
= super
->devlist
; dv
; dv
= dv
->next
) {
10387 if (dv
->dev
->vol
.migr_type
== MIGR_GEN_MIGR
10388 && dv
->dev
->vol
.migr_state
== 1) {
10393 /* Only one volume can migrate at the same time */
10394 if (migr_vol_qan
!= 1) {
10395 pr_err(": %s", migr_vol_qan
?
10396 "Number of migrating volumes greater than 1\n" :
10397 "There is no volume during migrationg\n");
10401 map_src
= get_imsm_map(dev
, MAP_1
);
10402 if (map_src
== NULL
)
10405 ndata
= imsm_num_data_members(dev
, MAP_0
);
10406 odata
= imsm_num_data_members(dev
, MAP_1
);
10408 chunk
= __le16_to_cpu(map_src
->blocks_per_strip
) * 512;
10409 old_data_stripe_length
= odata
* chunk
;
10411 migr_rec
= super
->migr_rec
;
10413 /* initialize migration record for start condition */
10414 if (sra
->reshape_progress
== 0)
10415 init_migr_record_imsm(st
, dev
, sra
);
10417 if (__le32_to_cpu(migr_rec
->rec_status
) != UNIT_SRC_NORMAL
) {
10418 dprintf("imsm: cannot restart migration when data are present in copy area.\n");
10421 /* Save checkpoint to update migration record for current
10422 * reshape position (in md). It can be farther than current
10423 * reshape position in metadata.
10425 if (save_checkpoint_imsm(st
, sra
, UNIT_SRC_NORMAL
) == 1) {
10426 /* ignore error == 2, this can mean end of reshape here
10428 dprintf("imsm: Cannot write checkpoint to migration record (UNIT_SRC_NORMAL, initial save)\n");
10433 /* size for data */
10434 buf_size
= __le32_to_cpu(migr_rec
->blocks_per_unit
) * 512;
10435 /* extend buffer size for parity disk */
10436 buf_size
+= __le32_to_cpu(migr_rec
->dest_depth_per_unit
) * 512;
10437 /* add space for stripe aligment */
10438 buf_size
+= old_data_stripe_length
;
10439 if (posix_memalign((void **)&buf
, 4096, buf_size
)) {
10440 dprintf("imsm: Cannot allocate checpoint buffer\n");
10444 max_position
= sra
->component_size
* ndata
;
10445 source_layout
= imsm_level_to_layout(map_src
->raid_level
);
10447 while (__le32_to_cpu(migr_rec
->curr_migr_unit
) <
10448 __le32_to_cpu(migr_rec
->num_migr_units
)) {
10449 /* current reshape position [blocks] */
10450 unsigned long long current_position
=
10451 __le32_to_cpu(migr_rec
->blocks_per_unit
)
10452 * __le32_to_cpu(migr_rec
->curr_migr_unit
);
10453 unsigned long long border
;
10455 /* Check that array hasn't become failed.
10457 degraded
= check_degradation_change(sra
, fds
, degraded
);
10458 if (degraded
> 1) {
10459 dprintf("imsm: Abort reshape due to degradation level (%i)\n", degraded
);
10463 next_step
= __le32_to_cpu(migr_rec
->blocks_per_unit
);
10465 if ((current_position
+ next_step
) > max_position
)
10466 next_step
= max_position
- current_position
;
10468 start
= current_position
* 512;
10470 /* allign reading start to old geometry */
10471 start_buf_shift
= start
% old_data_stripe_length
;
10472 start_src
= start
- start_buf_shift
;
10474 border
= (start_src
/ odata
) - (start
/ ndata
);
10476 if (border
<= __le32_to_cpu(migr_rec
->dest_depth_per_unit
)) {
10477 /* save critical stripes to buf
10478 * start - start address of current unit
10479 * to backup [bytes]
10480 * start_src - start address of current unit
10481 * to backup alligned to source array
10484 unsigned long long next_step_filler
= 0;
10485 unsigned long long copy_length
= next_step
* 512;
10487 /* allign copy area length to stripe in old geometry */
10488 next_step_filler
= ((copy_length
+ start_buf_shift
)
10489 % old_data_stripe_length
);
10490 if (next_step_filler
)
10491 next_step_filler
= (old_data_stripe_length
10492 - next_step_filler
);
10493 dprintf("save_stripes() parameters: start = %llu,\tstart_src = %llu,\tnext_step*512 = %llu,\tstart_in_buf_shift = %llu,\tnext_step_filler = %llu\n",
10494 start
, start_src
, copy_length
,
10495 start_buf_shift
, next_step_filler
);
10497 if (save_stripes(fds
, offsets
, map_src
->num_members
,
10498 chunk
, map_src
->raid_level
,
10499 source_layout
, 0, NULL
, start_src
,
10501 next_step_filler
+ start_buf_shift
,
10503 dprintf("imsm: Cannot save stripes to buffer\n");
10506 /* Convert data to destination format and store it
10507 * in backup general migration area
10509 if (save_backup_imsm(st
, dev
, sra
,
10510 buf
+ start_buf_shift
, copy_length
)) {
10511 dprintf("imsm: Cannot save stripes to target devices\n");
10514 if (save_checkpoint_imsm(st
, sra
,
10515 UNIT_SRC_IN_CP_AREA
)) {
10516 dprintf("imsm: Cannot write checkpoint to migration record (UNIT_SRC_IN_CP_AREA)\n");
10520 /* set next step to use whole border area */
10521 border
/= next_step
;
10523 next_step
*= border
;
10525 /* When data backed up, checkpoint stored,
10526 * kick the kernel to reshape unit of data
10528 next_step
= next_step
+ sra
->reshape_progress
;
10529 /* limit next step to array max position */
10530 if (next_step
> max_position
)
10531 next_step
= max_position
;
10532 sysfs_set_num(sra
, NULL
, "suspend_lo", sra
->reshape_progress
);
10533 sysfs_set_num(sra
, NULL
, "suspend_hi", next_step
);
10534 sra
->reshape_progress
= next_step
;
10536 /* wait until reshape finish */
10537 if (wait_for_reshape_imsm(sra
, ndata
) < 0) {
10538 dprintf("wait_for_reshape_imsm returned error!\n");
10544 if (save_checkpoint_imsm(st
, sra
, UNIT_SRC_NORMAL
) == 1) {
10545 /* ignore error == 2, this can mean end of reshape here
10547 dprintf("imsm: Cannot write checkpoint to migration record (UNIT_SRC_NORMAL)\n");
10553 /* clear migr_rec on disks after successful migration */
10556 memset(super
->migr_rec_buf
, 0, MIGR_REC_BUF_SIZE
);
10557 for (d
= super
->disks
; d
; d
= d
->next
) {
10558 if (d
->index
< 0 || is_failed(&d
->disk
))
10560 unsigned long long dsize
;
10562 get_dev_size(d
->fd
, NULL
, &dsize
);
10563 if (lseek64(d
->fd
, dsize
- MIGR_REC_POSITION
,
10565 if (write(d
->fd
, super
->migr_rec_buf
,
10566 MIGR_REC_BUF_SIZE
) != MIGR_REC_BUF_SIZE
)
10567 perror("Write migr_rec failed");
10571 /* return '1' if done */
10575 abort_reshape(sra
);
10580 #endif /* MDASSEMBLE */
10582 struct superswitch super_imsm
= {
10584 .examine_super
= examine_super_imsm
,
10585 .brief_examine_super
= brief_examine_super_imsm
,
10586 .brief_examine_subarrays
= brief_examine_subarrays_imsm
,
10587 .export_examine_super
= export_examine_super_imsm
,
10588 .detail_super
= detail_super_imsm
,
10589 .brief_detail_super
= brief_detail_super_imsm
,
10590 .write_init_super
= write_init_super_imsm
,
10591 .validate_geometry
= validate_geometry_imsm
,
10592 .add_to_super
= add_to_super_imsm
,
10593 .remove_from_super
= remove_from_super_imsm
,
10594 .detail_platform
= detail_platform_imsm
,
10595 .export_detail_platform
= export_detail_platform_imsm
,
10596 .kill_subarray
= kill_subarray_imsm
,
10597 .update_subarray
= update_subarray_imsm
,
10598 .load_container
= load_container_imsm
,
10599 .default_geometry
= default_geometry_imsm
,
10600 .get_disk_controller_domain
= imsm_get_disk_controller_domain
,
10601 .reshape_super
= imsm_reshape_super
,
10602 .manage_reshape
= imsm_manage_reshape
,
10603 .recover_backup
= recover_backup_imsm
,
10604 .copy_metadata
= copy_metadata_imsm
,
10606 .match_home
= match_home_imsm
,
10607 .uuid_from_super
= uuid_from_super_imsm
,
10608 .getinfo_super
= getinfo_super_imsm
,
10609 .getinfo_super_disks
= getinfo_super_disks_imsm
,
10610 .update_super
= update_super_imsm
,
10612 .avail_size
= avail_size_imsm
,
10613 .min_acceptable_spare_size
= min_acceptable_spare_size_imsm
,
10615 .compare_super
= compare_super_imsm
,
10617 .load_super
= load_super_imsm
,
10618 .init_super
= init_super_imsm
,
10619 .store_super
= store_super_imsm
,
10620 .free_super
= free_super_imsm
,
10621 .match_metadata_desc
= match_metadata_desc_imsm
,
10622 .container_content
= container_content_imsm
,
10623 .validate_container
= validate_container_imsm
,
10630 .open_new
= imsm_open_new
,
10631 .set_array_state
= imsm_set_array_state
,
10632 .set_disk
= imsm_set_disk
,
10633 .sync_metadata
= imsm_sync_metadata
,
10634 .activate_spare
= imsm_activate_spare
,
10635 .process_update
= imsm_process_update
,
10636 .prepare_update
= imsm_prepare_update
,
10637 #endif /* MDASSEMBLE */