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
*entry
;
1953 for (entry
= orom_entries
; entry
; entry
= entry
->next
) {
1954 print_imsm_capability(&entry
->orom
);
1956 if (imsm_orom_is_nvme(&entry
->orom
)) {
1957 for (hba
= list
; hba
; hba
= hba
->next
) {
1958 if (hba
->type
== SYS_DEV_NVME
)
1959 printf(" NVMe Device : %s\n", hba
->path
);
1964 struct devid_list
*devid
;
1965 for (devid
= entry
->devid_list
; devid
; devid
= devid
->next
) {
1966 hba
= device_by_id(devid
->devid
);
1970 printf(" I/O Controller : %s (%s)\n",
1971 hba
->path
, get_sys_dev_type(hba
->type
));
1972 if (hba
->type
== SYS_DEV_SATA
) {
1973 host_base
= ahci_get_port_count(hba
->path
, &port_count
);
1974 if (ahci_enumerate_ports(hba
->path
, port_count
, host_base
, verbose
)) {
1976 pr_err("failed to enumerate ports on SATA controller at %s.\n", hba
->pci_id
);
1987 static int export_detail_platform_imsm(int verbose
, char *controller_path
)
1989 struct sys_dev
*list
, *hba
;
1992 list
= find_intel_devices();
1995 pr_err("IMSM_DETAIL_PLATFORM_ERROR=NO_INTEL_DEVICES\n");
2000 for (hba
= list
; hba
; hba
= hba
->next
) {
2001 if (controller_path
&& (compare_paths(hba
->path
,controller_path
) != 0))
2003 if (!find_imsm_capability(hba
) && verbose
> 0)
2004 pr_err("IMSM_DETAIL_PLATFORM_ERROR=NO_IMSM_CAPABLE_DEVICE_UNDER_%s\n", hba
->path
);
2009 const struct orom_entry
*entry
;
2011 for (entry
= orom_entries
; entry
; entry
= entry
->next
)
2012 print_imsm_capability_export(&entry
->orom
);
2019 static int match_home_imsm(struct supertype
*st
, char *homehost
)
2021 /* the imsm metadata format does not specify any host
2022 * identification information. We return -1 since we can never
2023 * confirm nor deny whether a given array is "meant" for this
2024 * host. We rely on compare_super and the 'family_num' fields to
2025 * exclude member disks that do not belong, and we rely on
2026 * mdadm.conf to specify the arrays that should be assembled.
2027 * Auto-assembly may still pick up "foreign" arrays.
2033 static void uuid_from_super_imsm(struct supertype
*st
, int uuid
[4])
2035 /* The uuid returned here is used for:
2036 * uuid to put into bitmap file (Create, Grow)
2037 * uuid for backup header when saving critical section (Grow)
2038 * comparing uuids when re-adding a device into an array
2039 * In these cases the uuid required is that of the data-array,
2040 * not the device-set.
2041 * uuid to recognise same set when adding a missing device back
2042 * to an array. This is a uuid for the device-set.
2044 * For each of these we can make do with a truncated
2045 * or hashed uuid rather than the original, as long as
2047 * In each case the uuid required is that of the data-array,
2048 * not the device-set.
2050 /* imsm does not track uuid's so we synthesis one using sha1 on
2051 * - The signature (Which is constant for all imsm array, but no matter)
2052 * - the orig_family_num of the container
2053 * - the index number of the volume
2054 * - the 'serial' number of the volume.
2055 * Hopefully these are all constant.
2057 struct intel_super
*super
= st
->sb
;
2060 struct sha1_ctx ctx
;
2061 struct imsm_dev
*dev
= NULL
;
2064 /* some mdadm versions failed to set ->orig_family_num, in which
2065 * case fall back to ->family_num. orig_family_num will be
2066 * fixed up with the first metadata update.
2068 family_num
= super
->anchor
->orig_family_num
;
2069 if (family_num
== 0)
2070 family_num
= super
->anchor
->family_num
;
2071 sha1_init_ctx(&ctx
);
2072 sha1_process_bytes(super
->anchor
->sig
, MPB_SIG_LEN
, &ctx
);
2073 sha1_process_bytes(&family_num
, sizeof(__u32
), &ctx
);
2074 if (super
->current_vol
>= 0)
2075 dev
= get_imsm_dev(super
, super
->current_vol
);
2077 __u32 vol
= super
->current_vol
;
2078 sha1_process_bytes(&vol
, sizeof(vol
), &ctx
);
2079 sha1_process_bytes(dev
->volume
, MAX_RAID_SERIAL_LEN
, &ctx
);
2081 sha1_finish_ctx(&ctx
, buf
);
2082 memcpy(uuid
, buf
, 4*4);
2087 get_imsm_numerical_version(struct imsm_super
*mpb
, int *m
, int *p
)
2089 __u8
*v
= get_imsm_version(mpb
);
2090 __u8
*end
= mpb
->sig
+ MAX_SIGNATURE_LENGTH
;
2091 char major
[] = { 0, 0, 0 };
2092 char minor
[] = { 0 ,0, 0 };
2093 char patch
[] = { 0, 0, 0 };
2094 char *ver_parse
[] = { major
, minor
, patch
};
2098 while (*v
!= '\0' && v
< end
) {
2099 if (*v
!= '.' && j
< 2)
2100 ver_parse
[i
][j
++] = *v
;
2108 *m
= strtol(minor
, NULL
, 0);
2109 *p
= strtol(patch
, NULL
, 0);
2113 static __u32
migr_strip_blocks_resync(struct imsm_dev
*dev
)
2115 /* migr_strip_size when repairing or initializing parity */
2116 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
2117 __u32 chunk
= __le32_to_cpu(map
->blocks_per_strip
);
2119 switch (get_imsm_raid_level(map
)) {
2124 return 128*1024 >> 9;
2128 static __u32
migr_strip_blocks_rebuild(struct imsm_dev
*dev
)
2130 /* migr_strip_size when rebuilding a degraded disk, no idea why
2131 * this is different than migr_strip_size_resync(), but it's good
2134 struct imsm_map
*map
= get_imsm_map(dev
, MAP_1
);
2135 __u32 chunk
= __le32_to_cpu(map
->blocks_per_strip
);
2137 switch (get_imsm_raid_level(map
)) {
2140 if (map
->num_members
% map
->num_domains
== 0)
2141 return 128*1024 >> 9;
2145 return max((__u32
) 64*1024 >> 9, chunk
);
2147 return 128*1024 >> 9;
2151 static __u32
num_stripes_per_unit_resync(struct imsm_dev
*dev
)
2153 struct imsm_map
*lo
= get_imsm_map(dev
, MAP_0
);
2154 struct imsm_map
*hi
= get_imsm_map(dev
, MAP_1
);
2155 __u32 lo_chunk
= __le32_to_cpu(lo
->blocks_per_strip
);
2156 __u32 hi_chunk
= __le32_to_cpu(hi
->blocks_per_strip
);
2158 return max((__u32
) 1, hi_chunk
/ lo_chunk
);
2161 static __u32
num_stripes_per_unit_rebuild(struct imsm_dev
*dev
)
2163 struct imsm_map
*lo
= get_imsm_map(dev
, MAP_0
);
2164 int level
= get_imsm_raid_level(lo
);
2166 if (level
== 1 || level
== 10) {
2167 struct imsm_map
*hi
= get_imsm_map(dev
, MAP_1
);
2169 return hi
->num_domains
;
2171 return num_stripes_per_unit_resync(dev
);
2174 static __u8
imsm_num_data_members(struct imsm_dev
*dev
, int second_map
)
2176 /* named 'imsm_' because raid0, raid1 and raid10
2177 * counter-intuitively have the same number of data disks
2179 struct imsm_map
*map
= get_imsm_map(dev
, second_map
);
2181 switch (get_imsm_raid_level(map
)) {
2183 return map
->num_members
;
2187 return map
->num_members
/2;
2189 return map
->num_members
- 1;
2191 dprintf("unsupported raid level\n");
2196 static __u32
parity_segment_depth(struct imsm_dev
*dev
)
2198 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
2199 __u32 chunk
= __le32_to_cpu(map
->blocks_per_strip
);
2201 switch(get_imsm_raid_level(map
)) {
2204 return chunk
* map
->num_domains
;
2206 return chunk
* map
->num_members
;
2212 static __u32
map_migr_block(struct imsm_dev
*dev
, __u32 block
)
2214 struct imsm_map
*map
= get_imsm_map(dev
, MAP_1
);
2215 __u32 chunk
= __le32_to_cpu(map
->blocks_per_strip
);
2216 __u32 strip
= block
/ chunk
;
2218 switch (get_imsm_raid_level(map
)) {
2221 __u32 vol_strip
= (strip
* map
->num_domains
) + 1;
2222 __u32 vol_stripe
= vol_strip
/ map
->num_members
;
2224 return vol_stripe
* chunk
+ block
% chunk
;
2226 __u32 stripe
= strip
/ (map
->num_members
- 1);
2228 return stripe
* chunk
+ block
% chunk
;
2235 static __u64
blocks_per_migr_unit(struct intel_super
*super
,
2236 struct imsm_dev
*dev
)
2238 /* calculate the conversion factor between per member 'blocks'
2239 * (md/{resync,rebuild}_start) and imsm migration units, return
2240 * 0 for the 'not migrating' and 'unsupported migration' cases
2242 if (!dev
->vol
.migr_state
)
2245 switch (migr_type(dev
)) {
2246 case MIGR_GEN_MIGR
: {
2247 struct migr_record
*migr_rec
= super
->migr_rec
;
2248 return __le32_to_cpu(migr_rec
->blocks_per_unit
);
2253 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
2254 __u32 stripes_per_unit
;
2255 __u32 blocks_per_unit
;
2264 /* yes, this is really the translation of migr_units to
2265 * per-member blocks in the 'resync' case
2267 stripes_per_unit
= num_stripes_per_unit_resync(dev
);
2268 migr_chunk
= migr_strip_blocks_resync(dev
);
2269 disks
= imsm_num_data_members(dev
, MAP_0
);
2270 blocks_per_unit
= stripes_per_unit
* migr_chunk
* disks
;
2271 stripe
= __le16_to_cpu(map
->blocks_per_strip
) * disks
;
2272 segment
= blocks_per_unit
/ stripe
;
2273 block_rel
= blocks_per_unit
- segment
* stripe
;
2274 parity_depth
= parity_segment_depth(dev
);
2275 block_map
= map_migr_block(dev
, block_rel
);
2276 return block_map
+ parity_depth
* segment
;
2278 case MIGR_REBUILD
: {
2279 __u32 stripes_per_unit
;
2282 stripes_per_unit
= num_stripes_per_unit_rebuild(dev
);
2283 migr_chunk
= migr_strip_blocks_rebuild(dev
);
2284 return migr_chunk
* stripes_per_unit
;
2286 case MIGR_STATE_CHANGE
:
2292 static int imsm_level_to_layout(int level
)
2300 return ALGORITHM_LEFT_ASYMMETRIC
;
2307 /*******************************************************************************
2308 * Function: read_imsm_migr_rec
2309 * Description: Function reads imsm migration record from last sector of disk
2311 * fd : disk descriptor
2312 * super : metadata info
2316 ******************************************************************************/
2317 static int read_imsm_migr_rec(int fd
, struct intel_super
*super
)
2320 unsigned long long dsize
;
2322 get_dev_size(fd
, NULL
, &dsize
);
2323 if (lseek64(fd
, dsize
- MIGR_REC_POSITION
, SEEK_SET
) < 0) {
2324 pr_err("Cannot seek to anchor block: %s\n",
2328 if (read(fd
, super
->migr_rec_buf
, MIGR_REC_BUF_SIZE
) !=
2329 MIGR_REC_BUF_SIZE
) {
2330 pr_err("Cannot read migr record block: %s\n",
2340 static struct imsm_dev
*imsm_get_device_during_migration(
2341 struct intel_super
*super
)
2344 struct intel_dev
*dv
;
2346 for (dv
= super
->devlist
; dv
; dv
= dv
->next
) {
2347 if (is_gen_migration(dv
->dev
))
2353 /*******************************************************************************
2354 * Function: load_imsm_migr_rec
2355 * Description: Function reads imsm migration record (it is stored at the last
2358 * super : imsm internal array info
2359 * info : general array info
2363 * -2 : no migration in progress
2364 ******************************************************************************/
2365 static int load_imsm_migr_rec(struct intel_super
*super
, struct mdinfo
*info
)
2368 struct dl
*dl
= NULL
;
2372 struct imsm_dev
*dev
;
2373 struct imsm_map
*map
= NULL
;
2376 /* find map under migration */
2377 dev
= imsm_get_device_during_migration(super
);
2378 /* nothing to load,no migration in progress?
2382 map
= get_imsm_map(dev
, MAP_0
);
2385 for (sd
= info
->devs
; sd
; sd
= sd
->next
) {
2386 /* skip spare and failed disks
2388 if (sd
->disk
.raid_disk
< 0)
2390 /* read only from one of the first two slots */
2392 slot
= get_imsm_disk_slot(map
,
2393 sd
->disk
.raid_disk
);
2394 if ((map
== NULL
) || (slot
> 1) || (slot
< 0))
2397 sprintf(nm
, "%d:%d", sd
->disk
.major
, sd
->disk
.minor
);
2398 fd
= dev_open(nm
, O_RDONLY
);
2404 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
2405 /* skip spare and failed disks
2409 /* read only from one of the first two slots */
2411 slot
= get_imsm_disk_slot(map
, dl
->index
);
2412 if ((map
== NULL
) || (slot
> 1) || (slot
< 0))
2414 sprintf(nm
, "%d:%d", dl
->major
, dl
->minor
);
2415 fd
= dev_open(nm
, O_RDONLY
);
2422 retval
= read_imsm_migr_rec(fd
, super
);
2431 /*******************************************************************************
2432 * function: imsm_create_metadata_checkpoint_update
2433 * Description: It creates update for checkpoint change.
2435 * super : imsm internal array info
2436 * u : pointer to prepared update
2439 * If length is equal to 0, input pointer u contains no update
2440 ******************************************************************************/
2441 static int imsm_create_metadata_checkpoint_update(
2442 struct intel_super
*super
,
2443 struct imsm_update_general_migration_checkpoint
**u
)
2446 int update_memory_size
= 0;
2448 dprintf("(enter)\n");
2454 /* size of all update data without anchor */
2455 update_memory_size
=
2456 sizeof(struct imsm_update_general_migration_checkpoint
);
2458 *u
= xcalloc(1, update_memory_size
);
2460 dprintf("error: cannot get memory\n");
2463 (*u
)->type
= update_general_migration_checkpoint
;
2464 (*u
)->curr_migr_unit
= __le32_to_cpu(super
->migr_rec
->curr_migr_unit
);
2465 dprintf("prepared for %u\n", (*u
)->curr_migr_unit
);
2467 return update_memory_size
;
2470 static void imsm_update_metadata_locally(struct supertype
*st
,
2471 void *buf
, int len
);
2473 /*******************************************************************************
2474 * Function: write_imsm_migr_rec
2475 * Description: Function writes imsm migration record
2476 * (at the last sector of disk)
2478 * super : imsm internal array info
2482 ******************************************************************************/
2483 static int write_imsm_migr_rec(struct supertype
*st
)
2485 struct intel_super
*super
= st
->sb
;
2486 unsigned long long dsize
;
2492 struct imsm_update_general_migration_checkpoint
*u
;
2493 struct imsm_dev
*dev
;
2494 struct imsm_map
*map
= NULL
;
2496 /* find map under migration */
2497 dev
= imsm_get_device_during_migration(super
);
2498 /* if no migration, write buffer anyway to clear migr_record
2499 * on disk based on first available device
2502 dev
= get_imsm_dev(super
, super
->current_vol
< 0 ? 0 :
2503 super
->current_vol
);
2505 map
= get_imsm_map(dev
, MAP_0
);
2507 for (sd
= super
->disks
; sd
; sd
= sd
->next
) {
2510 /* skip failed and spare devices */
2513 /* write to 2 first slots only */
2515 slot
= get_imsm_disk_slot(map
, sd
->index
);
2516 if ((map
== NULL
) || (slot
> 1) || (slot
< 0))
2519 sprintf(nm
, "%d:%d", sd
->major
, sd
->minor
);
2520 fd
= dev_open(nm
, O_RDWR
);
2523 get_dev_size(fd
, NULL
, &dsize
);
2524 if (lseek64(fd
, dsize
- MIGR_REC_POSITION
, SEEK_SET
) < 0) {
2525 pr_err("Cannot seek to anchor block: %s\n",
2529 if (write(fd
, super
->migr_rec_buf
, MIGR_REC_BUF_SIZE
) !=
2530 MIGR_REC_BUF_SIZE
) {
2531 pr_err("Cannot write migr record block: %s\n",
2538 /* update checkpoint information in metadata */
2539 len
= imsm_create_metadata_checkpoint_update(super
, &u
);
2542 dprintf("imsm: Cannot prepare update\n");
2545 /* update metadata locally */
2546 imsm_update_metadata_locally(st
, u
, len
);
2547 /* and possibly remotely */
2548 if (st
->update_tail
) {
2549 append_metadata_update(st
, u
, len
);
2550 /* during reshape we do all work inside metadata handler
2551 * manage_reshape(), so metadata update has to be triggered
2554 flush_metadata_updates(st
);
2555 st
->update_tail
= &st
->updates
;
2565 #endif /* MDASSEMBLE */
2567 /* spare/missing disks activations are not allowe when
2568 * array/container performs reshape operation, because
2569 * all arrays in container works on the same disks set
2571 int imsm_reshape_blocks_arrays_changes(struct intel_super
*super
)
2574 struct intel_dev
*i_dev
;
2575 struct imsm_dev
*dev
;
2577 /* check whole container
2579 for (i_dev
= super
->devlist
; i_dev
; i_dev
= i_dev
->next
) {
2581 if (is_gen_migration(dev
)) {
2582 /* No repair during any migration in container
2590 static unsigned long long imsm_component_size_aligment_check(int level
,
2592 unsigned long long component_size
)
2594 unsigned int component_size_alligment
;
2596 /* check component size aligment
2598 component_size_alligment
= component_size
% (chunk_size
/512);
2600 dprintf("(Level: %i, chunk_size = %i, component_size = %llu), component_size_alligment = %u\n",
2601 level
, chunk_size
, component_size
,
2602 component_size_alligment
);
2604 if (component_size_alligment
&& (level
!= 1) && (level
!= UnSet
)) {
2605 dprintf("imsm: reported component size alligned from %llu ",
2607 component_size
-= component_size_alligment
;
2608 dprintf_cont("to %llu (%i).\n",
2609 component_size
, component_size_alligment
);
2612 return component_size
;
2615 static void getinfo_super_imsm_volume(struct supertype
*st
, struct mdinfo
*info
, char *dmap
)
2617 struct intel_super
*super
= st
->sb
;
2618 struct migr_record
*migr_rec
= super
->migr_rec
;
2619 struct imsm_dev
*dev
= get_imsm_dev(super
, super
->current_vol
);
2620 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
2621 struct imsm_map
*prev_map
= get_imsm_map(dev
, MAP_1
);
2622 struct imsm_map
*map_to_analyse
= map
;
2624 int map_disks
= info
->array
.raid_disks
;
2626 memset(info
, 0, sizeof(*info
));
2628 map_to_analyse
= prev_map
;
2630 dl
= super
->current_disk
;
2632 info
->container_member
= super
->current_vol
;
2633 info
->array
.raid_disks
= map
->num_members
;
2634 info
->array
.level
= get_imsm_raid_level(map_to_analyse
);
2635 info
->array
.layout
= imsm_level_to_layout(info
->array
.level
);
2636 info
->array
.md_minor
= -1;
2637 info
->array
.ctime
= 0;
2638 info
->array
.utime
= 0;
2639 info
->array
.chunk_size
=
2640 __le16_to_cpu(map_to_analyse
->blocks_per_strip
) << 9;
2641 info
->array
.state
= !dev
->vol
.dirty
;
2642 info
->custom_array_size
= __le32_to_cpu(dev
->size_high
);
2643 info
->custom_array_size
<<= 32;
2644 info
->custom_array_size
|= __le32_to_cpu(dev
->size_low
);
2645 info
->recovery_blocked
= imsm_reshape_blocks_arrays_changes(st
->sb
);
2647 if (is_gen_migration(dev
)) {
2648 info
->reshape_active
= 1;
2649 info
->new_level
= get_imsm_raid_level(map
);
2650 info
->new_layout
= imsm_level_to_layout(info
->new_level
);
2651 info
->new_chunk
= __le16_to_cpu(map
->blocks_per_strip
) << 9;
2652 info
->delta_disks
= map
->num_members
- prev_map
->num_members
;
2653 if (info
->delta_disks
) {
2654 /* this needs to be applied to every array
2657 info
->reshape_active
= CONTAINER_RESHAPE
;
2659 /* We shape information that we give to md might have to be
2660 * modify to cope with md's requirement for reshaping arrays.
2661 * For example, when reshaping a RAID0, md requires it to be
2662 * presented as a degraded RAID4.
2663 * Also if a RAID0 is migrating to a RAID5 we need to specify
2664 * the array as already being RAID5, but the 'before' layout
2665 * is a RAID4-like layout.
2667 switch (info
->array
.level
) {
2669 switch(info
->new_level
) {
2671 /* conversion is happening as RAID4 */
2672 info
->array
.level
= 4;
2673 info
->array
.raid_disks
+= 1;
2676 /* conversion is happening as RAID5 */
2677 info
->array
.level
= 5;
2678 info
->array
.layout
= ALGORITHM_PARITY_N
;
2679 info
->delta_disks
-= 1;
2682 /* FIXME error message */
2683 info
->array
.level
= UnSet
;
2689 info
->new_level
= UnSet
;
2690 info
->new_layout
= UnSet
;
2691 info
->new_chunk
= info
->array
.chunk_size
;
2692 info
->delta_disks
= 0;
2696 info
->disk
.major
= dl
->major
;
2697 info
->disk
.minor
= dl
->minor
;
2698 info
->disk
.number
= dl
->index
;
2699 info
->disk
.raid_disk
= get_imsm_disk_slot(map_to_analyse
,
2703 info
->data_offset
= pba_of_lba0(map_to_analyse
);
2704 info
->component_size
= blocks_per_member(map_to_analyse
);
2706 info
->component_size
= imsm_component_size_aligment_check(
2708 info
->array
.chunk_size
,
2709 info
->component_size
);
2711 memset(info
->uuid
, 0, sizeof(info
->uuid
));
2712 info
->recovery_start
= MaxSector
;
2714 info
->reshape_progress
= 0;
2715 info
->resync_start
= MaxSector
;
2716 if ((map_to_analyse
->map_state
== IMSM_T_STATE_UNINITIALIZED
||
2718 imsm_reshape_blocks_arrays_changes(super
) == 0) {
2719 info
->resync_start
= 0;
2721 if (dev
->vol
.migr_state
) {
2722 switch (migr_type(dev
)) {
2725 __u64 blocks_per_unit
= blocks_per_migr_unit(super
,
2727 __u64 units
= __le32_to_cpu(dev
->vol
.curr_migr_unit
);
2729 info
->resync_start
= blocks_per_unit
* units
;
2732 case MIGR_GEN_MIGR
: {
2733 __u64 blocks_per_unit
= blocks_per_migr_unit(super
,
2735 __u64 units
= __le32_to_cpu(migr_rec
->curr_migr_unit
);
2736 unsigned long long array_blocks
;
2739 if (__le32_to_cpu(migr_rec
->ascending_migr
) &&
2741 (__le32_to_cpu(migr_rec
->num_migr_units
)-1)) &&
2742 (super
->migr_rec
->rec_status
==
2743 __cpu_to_le32(UNIT_SRC_IN_CP_AREA
)))
2746 info
->reshape_progress
= blocks_per_unit
* units
;
2748 dprintf("IMSM: General Migration checkpoint : %llu (%llu) -> read reshape progress : %llu\n",
2749 (unsigned long long)units
,
2750 (unsigned long long)blocks_per_unit
,
2751 info
->reshape_progress
);
2753 used_disks
= imsm_num_data_members(dev
, MAP_1
);
2754 if (used_disks
> 0) {
2755 array_blocks
= blocks_per_member(map
) *
2757 /* round array size down to closest MB
2759 info
->custom_array_size
= (array_blocks
2760 >> SECT_PER_MB_SHIFT
)
2761 << SECT_PER_MB_SHIFT
;
2765 /* we could emulate the checkpointing of
2766 * 'sync_action=check' migrations, but for now
2767 * we just immediately complete them
2770 /* this is handled by container_content_imsm() */
2771 case MIGR_STATE_CHANGE
:
2772 /* FIXME handle other migrations */
2774 /* we are not dirty, so... */
2775 info
->resync_start
= MaxSector
;
2779 strncpy(info
->name
, (char *) dev
->volume
, MAX_RAID_SERIAL_LEN
);
2780 info
->name
[MAX_RAID_SERIAL_LEN
] = 0;
2782 info
->array
.major_version
= -1;
2783 info
->array
.minor_version
= -2;
2784 sprintf(info
->text_version
, "/%s/%d", st
->container_devnm
, info
->container_member
);
2785 info
->safe_mode_delay
= 4000; /* 4 secs like the Matrix driver */
2786 uuid_from_super_imsm(st
, info
->uuid
);
2790 for (i
=0; i
<map_disks
; i
++) {
2792 if (i
< info
->array
.raid_disks
) {
2793 struct imsm_disk
*dsk
;
2794 j
= get_imsm_disk_idx(dev
, i
, MAP_X
);
2795 dsk
= get_imsm_disk(super
, j
);
2796 if (dsk
&& (dsk
->status
& CONFIGURED_DISK
))
2803 static __u8
imsm_check_degraded(struct intel_super
*super
, struct imsm_dev
*dev
,
2804 int failed
, int look_in_map
);
2806 static int imsm_count_failed(struct intel_super
*super
, struct imsm_dev
*dev
,
2810 static void manage_second_map(struct intel_super
*super
, struct imsm_dev
*dev
)
2812 if (is_gen_migration(dev
)) {
2815 struct imsm_map
*map2
= get_imsm_map(dev
, MAP_1
);
2817 failed
= imsm_count_failed(super
, dev
, MAP_1
);
2818 map_state
= imsm_check_degraded(super
, dev
, failed
, MAP_1
);
2819 if (map2
->map_state
!= map_state
) {
2820 map2
->map_state
= map_state
;
2821 super
->updates_pending
++;
2827 static struct imsm_disk
*get_imsm_missing(struct intel_super
*super
, __u8 index
)
2831 for (d
= super
->missing
; d
; d
= d
->next
)
2832 if (d
->index
== index
)
2837 static void getinfo_super_imsm(struct supertype
*st
, struct mdinfo
*info
, char *map
)
2839 struct intel_super
*super
= st
->sb
;
2840 struct imsm_disk
*disk
;
2841 int map_disks
= info
->array
.raid_disks
;
2842 int max_enough
= -1;
2844 struct imsm_super
*mpb
;
2846 if (super
->current_vol
>= 0) {
2847 getinfo_super_imsm_volume(st
, info
, map
);
2850 memset(info
, 0, sizeof(*info
));
2852 /* Set raid_disks to zero so that Assemble will always pull in valid
2855 info
->array
.raid_disks
= 0;
2856 info
->array
.level
= LEVEL_CONTAINER
;
2857 info
->array
.layout
= 0;
2858 info
->array
.md_minor
= -1;
2859 info
->array
.ctime
= 0; /* N/A for imsm */
2860 info
->array
.utime
= 0;
2861 info
->array
.chunk_size
= 0;
2863 info
->disk
.major
= 0;
2864 info
->disk
.minor
= 0;
2865 info
->disk
.raid_disk
= -1;
2866 info
->reshape_active
= 0;
2867 info
->array
.major_version
= -1;
2868 info
->array
.minor_version
= -2;
2869 strcpy(info
->text_version
, "imsm");
2870 info
->safe_mode_delay
= 0;
2871 info
->disk
.number
= -1;
2872 info
->disk
.state
= 0;
2874 info
->recovery_start
= MaxSector
;
2875 info
->recovery_blocked
= imsm_reshape_blocks_arrays_changes(st
->sb
);
2877 /* do we have the all the insync disks that we expect? */
2878 mpb
= super
->anchor
;
2880 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
2881 struct imsm_dev
*dev
= get_imsm_dev(super
, i
);
2882 int failed
, enough
, j
, missing
= 0;
2883 struct imsm_map
*map
;
2886 failed
= imsm_count_failed(super
, dev
, MAP_0
);
2887 state
= imsm_check_degraded(super
, dev
, failed
, MAP_0
);
2888 map
= get_imsm_map(dev
, MAP_0
);
2890 /* any newly missing disks?
2891 * (catches single-degraded vs double-degraded)
2893 for (j
= 0; j
< map
->num_members
; j
++) {
2894 __u32 ord
= get_imsm_ord_tbl_ent(dev
, j
, MAP_0
);
2895 __u32 idx
= ord_to_idx(ord
);
2897 if (!(ord
& IMSM_ORD_REBUILD
) &&
2898 get_imsm_missing(super
, idx
)) {
2904 if (state
== IMSM_T_STATE_FAILED
)
2906 else if (state
== IMSM_T_STATE_DEGRADED
&&
2907 (state
!= map
->map_state
|| missing
))
2909 else /* we're normal, or already degraded */
2911 if (is_gen_migration(dev
) && missing
) {
2912 /* during general migration we need all disks
2913 * that process is running on.
2914 * No new missing disk is allowed.
2918 /* no more checks necessary
2922 /* in the missing/failed disk case check to see
2923 * if at least one array is runnable
2925 max_enough
= max(max_enough
, enough
);
2927 dprintf("enough: %d\n", max_enough
);
2928 info
->container_enough
= max_enough
;
2931 __u32 reserved
= imsm_reserved_sectors(super
, super
->disks
);
2933 disk
= &super
->disks
->disk
;
2934 info
->data_offset
= total_blocks(&super
->disks
->disk
) - reserved
;
2935 info
->component_size
= reserved
;
2936 info
->disk
.state
= is_configured(disk
) ? (1 << MD_DISK_ACTIVE
) : 0;
2937 /* we don't change info->disk.raid_disk here because
2938 * this state will be finalized in mdmon after we have
2939 * found the 'most fresh' version of the metadata
2941 info
->disk
.state
|= is_failed(disk
) ? (1 << MD_DISK_FAULTY
) : 0;
2942 info
->disk
.state
|= is_spare(disk
) ? 0 : (1 << MD_DISK_SYNC
);
2945 /* only call uuid_from_super_imsm when this disk is part of a populated container,
2946 * ->compare_super may have updated the 'num_raid_devs' field for spares
2948 if (info
->disk
.state
& (1 << MD_DISK_SYNC
) || super
->anchor
->num_raid_devs
)
2949 uuid_from_super_imsm(st
, info
->uuid
);
2951 memcpy(info
->uuid
, uuid_zero
, sizeof(uuid_zero
));
2953 /* I don't know how to compute 'map' on imsm, so use safe default */
2956 for (i
= 0; i
< map_disks
; i
++)
2962 /* allocates memory and fills disk in mdinfo structure
2963 * for each disk in array */
2964 struct mdinfo
*getinfo_super_disks_imsm(struct supertype
*st
)
2966 struct mdinfo
*mddev
= NULL
;
2967 struct intel_super
*super
= st
->sb
;
2968 struct imsm_disk
*disk
;
2971 if (!super
|| !super
->disks
)
2974 mddev
= xcalloc(1, sizeof(*mddev
));
2978 tmp
= xcalloc(1, sizeof(*tmp
));
2980 tmp
->next
= mddev
->devs
;
2982 tmp
->disk
.number
= count
++;
2983 tmp
->disk
.major
= dl
->major
;
2984 tmp
->disk
.minor
= dl
->minor
;
2985 tmp
->disk
.state
= is_configured(disk
) ?
2986 (1 << MD_DISK_ACTIVE
) : 0;
2987 tmp
->disk
.state
|= is_failed(disk
) ? (1 << MD_DISK_FAULTY
) : 0;
2988 tmp
->disk
.state
|= is_spare(disk
) ? 0 : (1 << MD_DISK_SYNC
);
2989 tmp
->disk
.raid_disk
= -1;
2995 static int update_super_imsm(struct supertype
*st
, struct mdinfo
*info
,
2996 char *update
, char *devname
, int verbose
,
2997 int uuid_set
, char *homehost
)
2999 /* For 'assemble' and 'force' we need to return non-zero if any
3000 * change was made. For others, the return value is ignored.
3001 * Update options are:
3002 * force-one : This device looks a bit old but needs to be included,
3003 * update age info appropriately.
3004 * assemble: clear any 'faulty' flag to allow this device to
3006 * force-array: Array is degraded but being forced, mark it clean
3007 * if that will be needed to assemble it.
3009 * newdev: not used ????
3010 * grow: Array has gained a new device - this is currently for
3012 * resync: mark as dirty so a resync will happen.
3013 * name: update the name - preserving the homehost
3014 * uuid: Change the uuid of the array to match watch is given
3016 * Following are not relevant for this imsm:
3017 * sparc2.2 : update from old dodgey metadata
3018 * super-minor: change the preferred_minor number
3019 * summaries: update redundant counters.
3020 * homehost: update the recorded homehost
3021 * _reshape_progress: record new reshape_progress position.
3024 struct intel_super
*super
= st
->sb
;
3025 struct imsm_super
*mpb
;
3027 /* we can only update container info */
3028 if (!super
|| super
->current_vol
>= 0 || !super
->anchor
)
3031 mpb
= super
->anchor
;
3033 if (strcmp(update
, "uuid") == 0) {
3034 /* We take this to mean that the family_num should be updated.
3035 * However that is much smaller than the uuid so we cannot really
3036 * allow an explicit uuid to be given. And it is hard to reliably
3038 * So if !uuid_set we know the current uuid is random and just used
3039 * the first 'int' and copy it to the other 3 positions.
3040 * Otherwise we require the 4 'int's to be the same as would be the
3041 * case if we are using a random uuid. So an explicit uuid will be
3042 * accepted as long as all for ints are the same... which shouldn't hurt
3045 info
->uuid
[1] = info
->uuid
[2] = info
->uuid
[3] = info
->uuid
[0];
3048 if (info
->uuid
[0] != info
->uuid
[1] ||
3049 info
->uuid
[1] != info
->uuid
[2] ||
3050 info
->uuid
[2] != info
->uuid
[3])
3056 mpb
->orig_family_num
= info
->uuid
[0];
3057 } else if (strcmp(update
, "assemble") == 0)
3062 /* successful update? recompute checksum */
3064 mpb
->check_sum
= __le32_to_cpu(__gen_imsm_checksum(mpb
));
3069 static size_t disks_to_mpb_size(int disks
)
3073 size
= sizeof(struct imsm_super
);
3074 size
+= (disks
- 1) * sizeof(struct imsm_disk
);
3075 size
+= 2 * sizeof(struct imsm_dev
);
3076 /* up to 2 maps per raid device (-2 for imsm_maps in imsm_dev */
3077 size
+= (4 - 2) * sizeof(struct imsm_map
);
3078 /* 4 possible disk_ord_tbl's */
3079 size
+= 4 * (disks
- 1) * sizeof(__u32
);
3084 static __u64
avail_size_imsm(struct supertype
*st
, __u64 devsize
,
3085 unsigned long long data_offset
)
3087 if (devsize
< (MPB_SECTOR_CNT
+ IMSM_RESERVED_SECTORS
))
3090 return devsize
- (MPB_SECTOR_CNT
+ IMSM_RESERVED_SECTORS
);
3093 static void free_devlist(struct intel_super
*super
)
3095 struct intel_dev
*dv
;
3097 while (super
->devlist
) {
3098 dv
= super
->devlist
->next
;
3099 free(super
->devlist
->dev
);
3100 free(super
->devlist
);
3101 super
->devlist
= dv
;
3105 static void imsm_copy_dev(struct imsm_dev
*dest
, struct imsm_dev
*src
)
3107 memcpy(dest
, src
, sizeof_imsm_dev(src
, 0));
3110 static int compare_super_imsm(struct supertype
*st
, struct supertype
*tst
)
3114 * 0 same, or first was empty, and second was copied
3115 * 1 second had wrong number
3117 * 3 wrong other info
3119 struct intel_super
*first
= st
->sb
;
3120 struct intel_super
*sec
= tst
->sb
;
3127 /* in platform dependent environment test if the disks
3128 * use the same Intel hba
3129 * If not on Intel hba at all, allow anything.
3131 if (!check_env("IMSM_NO_PLATFORM") && first
->hba
&& sec
->hba
) {
3132 if (first
->hba
->type
!= sec
->hba
->type
) {
3134 "HBAs of devices do not match %s != %s\n",
3135 get_sys_dev_type(first
->hba
->type
),
3136 get_sys_dev_type(sec
->hba
->type
));
3139 if (first
->orom
!= sec
->orom
) {
3141 "HBAs of devices do not match %s != %s\n",
3142 first
->hba
->pci_id
, sec
->hba
->pci_id
);
3147 /* if an anchor does not have num_raid_devs set then it is a free
3150 if (first
->anchor
->num_raid_devs
> 0 &&
3151 sec
->anchor
->num_raid_devs
> 0) {
3152 /* Determine if these disks might ever have been
3153 * related. Further disambiguation can only take place
3154 * in load_super_imsm_all
3156 __u32 first_family
= first
->anchor
->orig_family_num
;
3157 __u32 sec_family
= sec
->anchor
->orig_family_num
;
3159 if (memcmp(first
->anchor
->sig
, sec
->anchor
->sig
,
3160 MAX_SIGNATURE_LENGTH
) != 0)
3163 if (first_family
== 0)
3164 first_family
= first
->anchor
->family_num
;
3165 if (sec_family
== 0)
3166 sec_family
= sec
->anchor
->family_num
;
3168 if (first_family
!= sec_family
)
3173 /* if 'first' is a spare promote it to a populated mpb with sec's
3176 if (first
->anchor
->num_raid_devs
== 0 &&
3177 sec
->anchor
->num_raid_devs
> 0) {
3179 struct intel_dev
*dv
;
3180 struct imsm_dev
*dev
;
3182 /* we need to copy raid device info from sec if an allocation
3183 * fails here we don't associate the spare
3185 for (i
= 0; i
< sec
->anchor
->num_raid_devs
; i
++) {
3186 dv
= xmalloc(sizeof(*dv
));
3187 dev
= xmalloc(sizeof_imsm_dev(get_imsm_dev(sec
, i
), 1));
3190 dv
->next
= first
->devlist
;
3191 first
->devlist
= dv
;
3193 if (i
< sec
->anchor
->num_raid_devs
) {
3194 /* allocation failure */
3195 free_devlist(first
);
3196 pr_err("imsm: failed to associate spare\n");
3199 first
->anchor
->num_raid_devs
= sec
->anchor
->num_raid_devs
;
3200 first
->anchor
->orig_family_num
= sec
->anchor
->orig_family_num
;
3201 first
->anchor
->family_num
= sec
->anchor
->family_num
;
3202 memcpy(first
->anchor
->sig
, sec
->anchor
->sig
, MAX_SIGNATURE_LENGTH
);
3203 for (i
= 0; i
< sec
->anchor
->num_raid_devs
; i
++)
3204 imsm_copy_dev(get_imsm_dev(first
, i
), get_imsm_dev(sec
, i
));
3210 static void fd2devname(int fd
, char *name
)
3214 char dname
[PATH_MAX
];
3219 if (fstat(fd
, &st
) != 0)
3221 sprintf(path
, "/sys/dev/block/%d:%d",
3222 major(st
.st_rdev
), minor(st
.st_rdev
));
3224 rv
= readlink(path
, dname
, sizeof(dname
)-1);
3229 nm
= strrchr(dname
, '/');
3232 snprintf(name
, MAX_RAID_SERIAL_LEN
, "/dev/%s", nm
);
3236 extern int scsi_get_serial(int fd
, void *buf
, size_t buf_len
);
3238 static int imsm_read_serial(int fd
, char *devname
,
3239 __u8 serial
[MAX_RAID_SERIAL_LEN
])
3241 unsigned char scsi_serial
[255];
3250 memset(scsi_serial
, 0, sizeof(scsi_serial
));
3252 rv
= scsi_get_serial(fd
, scsi_serial
, sizeof(scsi_serial
));
3254 if (rv
&& check_env("IMSM_DEVNAME_AS_SERIAL")) {
3255 memset(serial
, 0, MAX_RAID_SERIAL_LEN
);
3256 fd2devname(fd
, (char *) serial
);
3262 pr_err("Failed to retrieve serial for %s\n",
3267 rsp_len
= scsi_serial
[3];
3270 pr_err("Failed to retrieve serial for %s\n",
3274 rsp_buf
= (char *) &scsi_serial
[4];
3276 /* trim all whitespace and non-printable characters and convert
3279 for (i
= 0, dest
= rsp_buf
; i
< rsp_len
; i
++) {
3282 /* ':' is reserved for use in placeholder serial
3283 * numbers for missing disks
3291 len
= dest
- rsp_buf
;
3294 /* truncate leading characters */
3295 if (len
> MAX_RAID_SERIAL_LEN
) {
3296 dest
+= len
- MAX_RAID_SERIAL_LEN
;
3297 len
= MAX_RAID_SERIAL_LEN
;
3300 memset(serial
, 0, MAX_RAID_SERIAL_LEN
);
3301 memcpy(serial
, dest
, len
);
3306 static int serialcmp(__u8
*s1
, __u8
*s2
)
3308 return strncmp((char *) s1
, (char *) s2
, MAX_RAID_SERIAL_LEN
);
3311 static void serialcpy(__u8
*dest
, __u8
*src
)
3313 strncpy((char *) dest
, (char *) src
, MAX_RAID_SERIAL_LEN
);
3316 static struct dl
*serial_to_dl(__u8
*serial
, struct intel_super
*super
)
3320 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
3321 if (serialcmp(dl
->serial
, serial
) == 0)
3327 static struct imsm_disk
*
3328 __serial_to_disk(__u8
*serial
, struct imsm_super
*mpb
, int *idx
)
3332 for (i
= 0; i
< mpb
->num_disks
; i
++) {
3333 struct imsm_disk
*disk
= __get_imsm_disk(mpb
, i
);
3335 if (serialcmp(disk
->serial
, serial
) == 0) {
3346 load_imsm_disk(int fd
, struct intel_super
*super
, char *devname
, int keep_fd
)
3348 struct imsm_disk
*disk
;
3353 __u8 serial
[MAX_RAID_SERIAL_LEN
];
3355 rv
= imsm_read_serial(fd
, devname
, serial
);
3360 dl
= xcalloc(1, sizeof(*dl
));
3363 dl
->major
= major(stb
.st_rdev
);
3364 dl
->minor
= minor(stb
.st_rdev
);
3365 dl
->next
= super
->disks
;
3366 dl
->fd
= keep_fd
? fd
: -1;
3367 assert(super
->disks
== NULL
);
3369 serialcpy(dl
->serial
, serial
);
3372 fd2devname(fd
, name
);
3374 dl
->devname
= xstrdup(devname
);
3376 dl
->devname
= xstrdup(name
);
3378 /* look up this disk's index in the current anchor */
3379 disk
= __serial_to_disk(dl
->serial
, super
->anchor
, &dl
->index
);
3382 /* only set index on disks that are a member of a
3383 * populated contianer, i.e. one with raid_devs
3385 if (is_failed(&dl
->disk
))
3387 else if (is_spare(&dl
->disk
))
3395 /* When migrating map0 contains the 'destination' state while map1
3396 * contains the current state. When not migrating map0 contains the
3397 * current state. This routine assumes that map[0].map_state is set to
3398 * the current array state before being called.
3400 * Migration is indicated by one of the following states
3401 * 1/ Idle (migr_state=0 map0state=normal||unitialized||degraded||failed)
3402 * 2/ Initialize (migr_state=1 migr_type=MIGR_INIT map0state=normal
3403 * map1state=unitialized)
3404 * 3/ Repair (Resync) (migr_state=1 migr_type=MIGR_REPAIR map0state=normal
3406 * 4/ Rebuild (migr_state=1 migr_type=MIGR_REBUILD map0state=normal
3407 * map1state=degraded)
3408 * 5/ Migration (mig_state=1 migr_type=MIGR_GEN_MIGR map0state=normal
3411 static void migrate(struct imsm_dev
*dev
, struct intel_super
*super
,
3412 __u8 to_state
, int migr_type
)
3414 struct imsm_map
*dest
;
3415 struct imsm_map
*src
= get_imsm_map(dev
, MAP_0
);
3417 dev
->vol
.migr_state
= 1;
3418 set_migr_type(dev
, migr_type
);
3419 dev
->vol
.curr_migr_unit
= 0;
3420 dest
= get_imsm_map(dev
, MAP_1
);
3422 /* duplicate and then set the target end state in map[0] */
3423 memcpy(dest
, src
, sizeof_imsm_map(src
));
3424 if ((migr_type
== MIGR_REBUILD
) ||
3425 (migr_type
== MIGR_GEN_MIGR
)) {
3429 for (i
= 0; i
< src
->num_members
; i
++) {
3430 ord
= __le32_to_cpu(src
->disk_ord_tbl
[i
]);
3431 set_imsm_ord_tbl_ent(src
, i
, ord_to_idx(ord
));
3435 if (migr_type
== MIGR_GEN_MIGR
)
3436 /* Clear migration record */
3437 memset(super
->migr_rec
, 0, sizeof(struct migr_record
));
3439 src
->map_state
= to_state
;
3442 static void end_migration(struct imsm_dev
*dev
, struct intel_super
*super
,
3445 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
3446 struct imsm_map
*prev
= get_imsm_map(dev
, dev
->vol
.migr_state
== 0 ?
3450 /* merge any IMSM_ORD_REBUILD bits that were not successfully
3451 * completed in the last migration.
3453 * FIXME add support for raid-level-migration
3455 if ((map_state
!= map
->map_state
) && (is_gen_migration(dev
) == 0) &&
3456 (prev
->map_state
!= IMSM_T_STATE_UNINITIALIZED
)) {
3457 /* when final map state is other than expected
3458 * merge maps (not for migration)
3462 for (i
= 0; i
< prev
->num_members
; i
++)
3463 for (j
= 0; j
< map
->num_members
; j
++)
3464 /* during online capacity expansion
3465 * disks position can be changed
3466 * if takeover is used
3468 if (ord_to_idx(map
->disk_ord_tbl
[j
]) ==
3469 ord_to_idx(prev
->disk_ord_tbl
[i
])) {
3470 map
->disk_ord_tbl
[j
] |=
3471 prev
->disk_ord_tbl
[i
];
3474 failed
= imsm_count_failed(super
, dev
, MAP_0
);
3475 map_state
= imsm_check_degraded(super
, dev
, failed
, MAP_0
);
3478 dev
->vol
.migr_state
= 0;
3479 set_migr_type(dev
, 0);
3480 dev
->vol
.curr_migr_unit
= 0;
3481 map
->map_state
= map_state
;
3485 static int parse_raid_devices(struct intel_super
*super
)
3488 struct imsm_dev
*dev_new
;
3489 size_t len
, len_migr
;
3491 size_t space_needed
= 0;
3492 struct imsm_super
*mpb
= super
->anchor
;
3494 for (i
= 0; i
< super
->anchor
->num_raid_devs
; i
++) {
3495 struct imsm_dev
*dev_iter
= __get_imsm_dev(super
->anchor
, i
);
3496 struct intel_dev
*dv
;
3498 len
= sizeof_imsm_dev(dev_iter
, 0);
3499 len_migr
= sizeof_imsm_dev(dev_iter
, 1);
3501 space_needed
+= len_migr
- len
;
3503 dv
= xmalloc(sizeof(*dv
));
3504 if (max_len
< len_migr
)
3506 if (max_len
> len_migr
)
3507 space_needed
+= max_len
- len_migr
;
3508 dev_new
= xmalloc(max_len
);
3509 imsm_copy_dev(dev_new
, dev_iter
);
3512 dv
->next
= super
->devlist
;
3513 super
->devlist
= dv
;
3516 /* ensure that super->buf is large enough when all raid devices
3519 if (__le32_to_cpu(mpb
->mpb_size
) + space_needed
> super
->len
) {
3522 len
= ROUND_UP(__le32_to_cpu(mpb
->mpb_size
) + space_needed
, 512);
3523 if (posix_memalign(&buf
, 512, len
) != 0)
3526 memcpy(buf
, super
->buf
, super
->len
);
3527 memset(buf
+ super
->len
, 0, len
- super
->len
);
3536 /* retrieve a pointer to the bbm log which starts after all raid devices */
3537 struct bbm_log
*__get_imsm_bbm_log(struct imsm_super
*mpb
)
3541 if (__le32_to_cpu(mpb
->bbm_log_size
)) {
3543 ptr
+= mpb
->mpb_size
- __le32_to_cpu(mpb
->bbm_log_size
);
3549 /*******************************************************************************
3550 * Function: check_mpb_migr_compatibility
3551 * Description: Function checks for unsupported migration features:
3552 * - migration optimization area (pba_of_lba0)
3553 * - descending reshape (ascending_migr)
3555 * super : imsm metadata information
3557 * 0 : migration is compatible
3558 * -1 : migration is not compatible
3559 ******************************************************************************/
3560 int check_mpb_migr_compatibility(struct intel_super
*super
)
3562 struct imsm_map
*map0
, *map1
;
3563 struct migr_record
*migr_rec
= super
->migr_rec
;
3566 for (i
= 0; i
< super
->anchor
->num_raid_devs
; i
++) {
3567 struct imsm_dev
*dev_iter
= __get_imsm_dev(super
->anchor
, i
);
3570 dev_iter
->vol
.migr_state
== 1 &&
3571 dev_iter
->vol
.migr_type
== MIGR_GEN_MIGR
) {
3572 /* This device is migrating */
3573 map0
= get_imsm_map(dev_iter
, MAP_0
);
3574 map1
= get_imsm_map(dev_iter
, MAP_1
);
3575 if (pba_of_lba0(map0
) != pba_of_lba0(map1
))
3576 /* migration optimization area was used */
3578 if (migr_rec
->ascending_migr
== 0
3579 && migr_rec
->dest_depth_per_unit
> 0)
3580 /* descending reshape not supported yet */
3587 static void __free_imsm(struct intel_super
*super
, int free_disks
);
3589 /* load_imsm_mpb - read matrix metadata
3590 * allocates super->mpb to be freed by free_imsm
3592 static int load_imsm_mpb(int fd
, struct intel_super
*super
, char *devname
)
3594 unsigned long long dsize
;
3595 unsigned long long sectors
;
3597 struct imsm_super
*anchor
;
3600 get_dev_size(fd
, NULL
, &dsize
);
3603 pr_err("%s: device to small for imsm\n",
3608 if (lseek64(fd
, dsize
- (512 * 2), SEEK_SET
) < 0) {
3610 pr_err("Cannot seek to anchor block on %s: %s\n",
3611 devname
, strerror(errno
));
3615 if (posix_memalign((void**)&anchor
, 512, 512) != 0) {
3617 pr_err("Failed to allocate imsm anchor buffer on %s\n", devname
);
3620 if (read(fd
, anchor
, 512) != 512) {
3622 pr_err("Cannot read anchor block on %s: %s\n",
3623 devname
, strerror(errno
));
3628 if (strncmp((char *) anchor
->sig
, MPB_SIGNATURE
, MPB_SIG_LEN
) != 0) {
3630 pr_err("no IMSM anchor on %s\n", devname
);
3635 __free_imsm(super
, 0);
3636 /* reload capability and hba */
3638 /* capability and hba must be updated with new super allocation */
3639 find_intel_hba_capability(fd
, super
, devname
);
3640 super
->len
= ROUND_UP(anchor
->mpb_size
, 512);
3641 if (posix_memalign(&super
->buf
, 512, super
->len
) != 0) {
3643 pr_err("unable to allocate %zu byte mpb buffer\n",
3648 memcpy(super
->buf
, anchor
, 512);
3650 sectors
= mpb_sectors(anchor
) - 1;
3653 if (posix_memalign(&super
->migr_rec_buf
, 512, MIGR_REC_BUF_SIZE
) != 0) {
3654 pr_err("could not allocate migr_rec buffer\n");
3658 super
->clean_migration_record_by_mdmon
= 0;
3661 check_sum
= __gen_imsm_checksum(super
->anchor
);
3662 if (check_sum
!= __le32_to_cpu(super
->anchor
->check_sum
)) {
3664 pr_err("IMSM checksum %x != %x on %s\n",
3666 __le32_to_cpu(super
->anchor
->check_sum
),
3674 /* read the extended mpb */
3675 if (lseek64(fd
, dsize
- (512 * (2 + sectors
)), SEEK_SET
) < 0) {
3677 pr_err("Cannot seek to extended mpb on %s: %s\n",
3678 devname
, strerror(errno
));
3682 if ((unsigned)read(fd
, super
->buf
+ 512, super
->len
- 512) != super
->len
- 512) {
3684 pr_err("Cannot read extended mpb on %s: %s\n",
3685 devname
, strerror(errno
));
3689 check_sum
= __gen_imsm_checksum(super
->anchor
);
3690 if (check_sum
!= __le32_to_cpu(super
->anchor
->check_sum
)) {
3692 pr_err("IMSM checksum %x != %x on %s\n",
3693 check_sum
, __le32_to_cpu(super
->anchor
->check_sum
),
3698 /* FIXME the BBM log is disk specific so we cannot use this global
3699 * buffer for all disks. Ok for now since we only look at the global
3700 * bbm_log_size parameter to gate assembly
3702 super
->bbm_log
= __get_imsm_bbm_log(super
->anchor
);
3707 static int read_imsm_migr_rec(int fd
, struct intel_super
*super
);
3709 /* clears hi bits in metadata if MPB_ATTRIB_2TB_DISK not set */
3710 static void clear_hi(struct intel_super
*super
)
3712 struct imsm_super
*mpb
= super
->anchor
;
3714 if (mpb
->attributes
& MPB_ATTRIB_2TB_DISK
)
3716 for (i
= 0; i
< mpb
->num_disks
; ++i
) {
3717 struct imsm_disk
*disk
= &mpb
->disk
[i
];
3718 disk
->total_blocks_hi
= 0;
3720 for (i
= 0; i
< mpb
->num_raid_devs
; ++i
) {
3721 struct imsm_dev
*dev
= get_imsm_dev(super
, i
);
3724 for (n
= 0; n
< 2; ++n
) {
3725 struct imsm_map
*map
= get_imsm_map(dev
, n
);
3728 map
->pba_of_lba0_hi
= 0;
3729 map
->blocks_per_member_hi
= 0;
3730 map
->num_data_stripes_hi
= 0;
3736 load_and_parse_mpb(int fd
, struct intel_super
*super
, char *devname
, int keep_fd
)
3740 err
= load_imsm_mpb(fd
, super
, devname
);
3743 err
= load_imsm_disk(fd
, super
, devname
, keep_fd
);
3746 err
= parse_raid_devices(super
);
3751 static void __free_imsm_disk(struct dl
*d
)
3763 static void free_imsm_disks(struct intel_super
*super
)
3767 while (super
->disks
) {
3769 super
->disks
= d
->next
;
3770 __free_imsm_disk(d
);
3772 while (super
->disk_mgmt_list
) {
3773 d
= super
->disk_mgmt_list
;
3774 super
->disk_mgmt_list
= d
->next
;
3775 __free_imsm_disk(d
);
3777 while (super
->missing
) {
3779 super
->missing
= d
->next
;
3780 __free_imsm_disk(d
);
3785 /* free all the pieces hanging off of a super pointer */
3786 static void __free_imsm(struct intel_super
*super
, int free_disks
)
3788 struct intel_hba
*elem
, *next
;
3794 /* unlink capability description */
3796 if (super
->migr_rec_buf
) {
3797 free(super
->migr_rec_buf
);
3798 super
->migr_rec_buf
= NULL
;
3801 free_imsm_disks(super
);
3802 free_devlist(super
);
3806 free((void *)elem
->path
);
3814 static void free_imsm(struct intel_super
*super
)
3816 __free_imsm(super
, 1);
3820 static void free_super_imsm(struct supertype
*st
)
3822 struct intel_super
*super
= st
->sb
;
3831 static struct intel_super
*alloc_super(void)
3833 struct intel_super
*super
= xcalloc(1, sizeof(*super
));
3835 super
->current_vol
= -1;
3836 super
->create_offset
= ~((unsigned long long) 0);
3841 * find and allocate hba and OROM/EFI based on valid fd of RAID component device
3843 static int find_intel_hba_capability(int fd
, struct intel_super
*super
, char *devname
)
3845 struct sys_dev
*hba_name
;
3848 if ((fd
< 0) || check_env("IMSM_NO_PLATFORM")) {
3853 hba_name
= find_disk_attached_hba(fd
, NULL
);
3856 pr_err("%s is not attached to Intel(R) RAID controller.\n",
3860 rv
= attach_hba_to_super(super
, hba_name
);
3863 struct intel_hba
*hba
= super
->hba
;
3865 pr_err("%s is attached to Intel(R) %s RAID controller (%s),\n"
3866 " but the container is assigned to Intel(R) %s RAID controller (",
3868 get_sys_dev_type(hba_name
->type
),
3869 hba_name
->pci_id
? : "Err!",
3870 get_sys_dev_type(super
->hba
->type
));
3873 fprintf(stderr
, "%s", hba
->pci_id
? : "Err!");
3875 fprintf(stderr
, ", ");
3878 fprintf(stderr
, ").\n"
3879 " Mixing devices attached to different controllers is not allowed.\n");
3883 super
->orom
= find_imsm_capability(hba_name
);
3890 /* find_missing - helper routine for load_super_imsm_all that identifies
3891 * disks that have disappeared from the system. This routine relies on
3892 * the mpb being uptodate, which it is at load time.
3894 static int find_missing(struct intel_super
*super
)
3897 struct imsm_super
*mpb
= super
->anchor
;
3899 struct imsm_disk
*disk
;
3901 for (i
= 0; i
< mpb
->num_disks
; i
++) {
3902 disk
= __get_imsm_disk(mpb
, i
);
3903 dl
= serial_to_dl(disk
->serial
, super
);
3907 dl
= xmalloc(sizeof(*dl
));
3911 dl
->devname
= xstrdup("missing");
3913 serialcpy(dl
->serial
, disk
->serial
);
3916 dl
->next
= super
->missing
;
3917 super
->missing
= dl
;
3924 static struct intel_disk
*disk_list_get(__u8
*serial
, struct intel_disk
*disk_list
)
3926 struct intel_disk
*idisk
= disk_list
;
3929 if (serialcmp(idisk
->disk
.serial
, serial
) == 0)
3931 idisk
= idisk
->next
;
3937 static int __prep_thunderdome(struct intel_super
**table
, int tbl_size
,
3938 struct intel_super
*super
,
3939 struct intel_disk
**disk_list
)
3941 struct imsm_disk
*d
= &super
->disks
->disk
;
3942 struct imsm_super
*mpb
= super
->anchor
;
3945 for (i
= 0; i
< tbl_size
; i
++) {
3946 struct imsm_super
*tbl_mpb
= table
[i
]->anchor
;
3947 struct imsm_disk
*tbl_d
= &table
[i
]->disks
->disk
;
3949 if (tbl_mpb
->family_num
== mpb
->family_num
) {
3950 if (tbl_mpb
->check_sum
== mpb
->check_sum
) {
3951 dprintf("mpb from %d:%d matches %d:%d\n",
3952 super
->disks
->major
,
3953 super
->disks
->minor
,
3954 table
[i
]->disks
->major
,
3955 table
[i
]->disks
->minor
);
3959 if (((is_configured(d
) && !is_configured(tbl_d
)) ||
3960 is_configured(d
) == is_configured(tbl_d
)) &&
3961 tbl_mpb
->generation_num
< mpb
->generation_num
) {
3962 /* current version of the mpb is a
3963 * better candidate than the one in
3964 * super_table, but copy over "cross
3965 * generational" status
3967 struct intel_disk
*idisk
;
3969 dprintf("mpb from %d:%d replaces %d:%d\n",
3970 super
->disks
->major
,
3971 super
->disks
->minor
,
3972 table
[i
]->disks
->major
,
3973 table
[i
]->disks
->minor
);
3975 idisk
= disk_list_get(tbl_d
->serial
, *disk_list
);
3976 if (idisk
&& is_failed(&idisk
->disk
))
3977 tbl_d
->status
|= FAILED_DISK
;
3980 struct intel_disk
*idisk
;
3981 struct imsm_disk
*disk
;
3983 /* tbl_mpb is more up to date, but copy
3984 * over cross generational status before
3987 disk
= __serial_to_disk(d
->serial
, mpb
, NULL
);
3988 if (disk
&& is_failed(disk
))
3989 d
->status
|= FAILED_DISK
;
3991 idisk
= disk_list_get(d
->serial
, *disk_list
);
3994 if (disk
&& is_configured(disk
))
3995 idisk
->disk
.status
|= CONFIGURED_DISK
;
3998 dprintf("mpb from %d:%d prefer %d:%d\n",
3999 super
->disks
->major
,
4000 super
->disks
->minor
,
4001 table
[i
]->disks
->major
,
4002 table
[i
]->disks
->minor
);
4010 table
[tbl_size
++] = super
;
4014 /* update/extend the merged list of imsm_disk records */
4015 for (j
= 0; j
< mpb
->num_disks
; j
++) {
4016 struct imsm_disk
*disk
= __get_imsm_disk(mpb
, j
);
4017 struct intel_disk
*idisk
;
4019 idisk
= disk_list_get(disk
->serial
, *disk_list
);
4021 idisk
->disk
.status
|= disk
->status
;
4022 if (is_configured(&idisk
->disk
) ||
4023 is_failed(&idisk
->disk
))
4024 idisk
->disk
.status
&= ~(SPARE_DISK
);
4026 idisk
= xcalloc(1, sizeof(*idisk
));
4027 idisk
->owner
= IMSM_UNKNOWN_OWNER
;
4028 idisk
->disk
= *disk
;
4029 idisk
->next
= *disk_list
;
4033 if (serialcmp(idisk
->disk
.serial
, d
->serial
) == 0)
4040 static struct intel_super
*
4041 validate_members(struct intel_super
*super
, struct intel_disk
*disk_list
,
4044 struct imsm_super
*mpb
= super
->anchor
;
4048 for (i
= 0; i
< mpb
->num_disks
; i
++) {
4049 struct imsm_disk
*disk
= __get_imsm_disk(mpb
, i
);
4050 struct intel_disk
*idisk
;
4052 idisk
= disk_list_get(disk
->serial
, disk_list
);
4054 if (idisk
->owner
== owner
||
4055 idisk
->owner
== IMSM_UNKNOWN_OWNER
)
4058 dprintf("'%.16s' owner %d != %d\n",
4059 disk
->serial
, idisk
->owner
,
4062 dprintf("unknown disk %x [%d]: %.16s\n",
4063 __le32_to_cpu(mpb
->family_num
), i
,
4069 if (ok_count
== mpb
->num_disks
)
4074 static void show_conflicts(__u32 family_num
, struct intel_super
*super_list
)
4076 struct intel_super
*s
;
4078 for (s
= super_list
; s
; s
= s
->next
) {
4079 if (family_num
!= s
->anchor
->family_num
)
4081 pr_err("Conflict, offlining family %#x on '%s'\n",
4082 __le32_to_cpu(family_num
), s
->disks
->devname
);
4086 static struct intel_super
*
4087 imsm_thunderdome(struct intel_super
**super_list
, int len
)
4089 struct intel_super
*super_table
[len
];
4090 struct intel_disk
*disk_list
= NULL
;
4091 struct intel_super
*champion
, *spare
;
4092 struct intel_super
*s
, **del
;
4097 memset(super_table
, 0, sizeof(super_table
));
4098 for (s
= *super_list
; s
; s
= s
->next
)
4099 tbl_size
= __prep_thunderdome(super_table
, tbl_size
, s
, &disk_list
);
4101 for (i
= 0; i
< tbl_size
; i
++) {
4102 struct imsm_disk
*d
;
4103 struct intel_disk
*idisk
;
4104 struct imsm_super
*mpb
= super_table
[i
]->anchor
;
4107 d
= &s
->disks
->disk
;
4109 /* 'd' must appear in merged disk list for its
4110 * configuration to be valid
4112 idisk
= disk_list_get(d
->serial
, disk_list
);
4113 if (idisk
&& idisk
->owner
== i
)
4114 s
= validate_members(s
, disk_list
, i
);
4119 dprintf("marking family: %#x from %d:%d offline\n",
4121 super_table
[i
]->disks
->major
,
4122 super_table
[i
]->disks
->minor
);
4126 /* This is where the mdadm implementation differs from the Windows
4127 * driver which has no strict concept of a container. We can only
4128 * assemble one family from a container, so when returning a prodigal
4129 * array member to this system the code will not be able to disambiguate
4130 * the container contents that should be assembled ("foreign" versus
4131 * "local"). It requires user intervention to set the orig_family_num
4132 * to a new value to establish a new container. The Windows driver in
4133 * this situation fixes up the volume name in place and manages the
4134 * foreign array as an independent entity.
4139 for (i
= 0; i
< tbl_size
; i
++) {
4140 struct intel_super
*tbl_ent
= super_table
[i
];
4146 if (tbl_ent
->anchor
->num_raid_devs
== 0) {
4151 if (s
&& !is_spare
) {
4152 show_conflicts(tbl_ent
->anchor
->family_num
, *super_list
);
4154 } else if (!s
&& !is_spare
)
4167 pr_err("Chose family %#x on '%s', assemble conflicts to new container with '--update=uuid'\n",
4168 __le32_to_cpu(s
->anchor
->family_num
), s
->disks
->devname
);
4170 /* collect all dl's onto 'champion', and update them to
4171 * champion's version of the status
4173 for (s
= *super_list
; s
; s
= s
->next
) {
4174 struct imsm_super
*mpb
= champion
->anchor
;
4175 struct dl
*dl
= s
->disks
;
4180 mpb
->attributes
|= s
->anchor
->attributes
& MPB_ATTRIB_2TB_DISK
;
4182 for (i
= 0; i
< mpb
->num_disks
; i
++) {
4183 struct imsm_disk
*disk
;
4185 disk
= __serial_to_disk(dl
->serial
, mpb
, &dl
->index
);
4188 /* only set index on disks that are a member of
4189 * a populated contianer, i.e. one with
4192 if (is_failed(&dl
->disk
))
4194 else if (is_spare(&dl
->disk
))
4200 if (i
>= mpb
->num_disks
) {
4201 struct intel_disk
*idisk
;
4203 idisk
= disk_list_get(dl
->serial
, disk_list
);
4204 if (idisk
&& is_spare(&idisk
->disk
) &&
4205 !is_failed(&idisk
->disk
) && !is_configured(&idisk
->disk
))
4213 dl
->next
= champion
->disks
;
4214 champion
->disks
= dl
;
4218 /* delete 'champion' from super_list */
4219 for (del
= super_list
; *del
; ) {
4220 if (*del
== champion
) {
4221 *del
= (*del
)->next
;
4224 del
= &(*del
)->next
;
4226 champion
->next
= NULL
;
4230 struct intel_disk
*idisk
= disk_list
;
4232 disk_list
= disk_list
->next
;
4240 get_sra_super_block(int fd
, struct intel_super
**super_list
, char *devname
, int *max
, int keep_fd
);
4241 static int get_super_block(struct intel_super
**super_list
, char *devnm
, char *devname
,
4242 int major
, int minor
, int keep_fd
);
4244 get_devlist_super_block(struct md_list
*devlist
, struct intel_super
**super_list
,
4245 int *max
, int keep_fd
);
4247 static int load_super_imsm_all(struct supertype
*st
, int fd
, void **sbp
,
4248 char *devname
, struct md_list
*devlist
,
4251 struct intel_super
*super_list
= NULL
;
4252 struct intel_super
*super
= NULL
;
4257 /* 'fd' is an opened container */
4258 err
= get_sra_super_block(fd
, &super_list
, devname
, &i
, keep_fd
);
4260 /* get super block from devlist devices */
4261 err
= get_devlist_super_block(devlist
, &super_list
, &i
, keep_fd
);
4264 /* all mpbs enter, maybe one leaves */
4265 super
= imsm_thunderdome(&super_list
, i
);
4271 if (find_missing(super
) != 0) {
4277 /* load migration record */
4278 err
= load_imsm_migr_rec(super
, NULL
);
4280 /* migration is in progress,
4281 * but migr_rec cannot be loaded,
4287 /* Check migration compatibility */
4288 if ((err
== 0) && (check_mpb_migr_compatibility(super
) != 0)) {
4289 pr_err("Unsupported migration detected");
4291 fprintf(stderr
, " on %s\n", devname
);
4293 fprintf(stderr
, " (IMSM).\n");
4302 while (super_list
) {
4303 struct intel_super
*s
= super_list
;
4305 super_list
= super_list
->next
;
4314 strcpy(st
->container_devnm
, fd2devnm(fd
));
4316 st
->container_devnm
[0] = 0;
4317 if (err
== 0 && st
->ss
== NULL
) {
4318 st
->ss
= &super_imsm
;
4319 st
->minor_version
= 0;
4320 st
->max_devs
= IMSM_MAX_DEVICES
;
4326 get_devlist_super_block(struct md_list
*devlist
, struct intel_super
**super_list
,
4327 int *max
, int keep_fd
)
4329 struct md_list
*tmpdev
;
4333 for (i
= 0, tmpdev
= devlist
; tmpdev
; tmpdev
= tmpdev
->next
) {
4334 if (tmpdev
->used
!= 1)
4336 if (tmpdev
->container
== 1) {
4338 int fd
= dev_open(tmpdev
->devname
, O_RDONLY
|O_EXCL
);
4340 pr_err("cannot open device %s: %s\n",
4341 tmpdev
->devname
, strerror(errno
));
4345 err
= get_sra_super_block(fd
, super_list
,
4346 tmpdev
->devname
, &lmax
,
4355 int major
= major(tmpdev
->st_rdev
);
4356 int minor
= minor(tmpdev
->st_rdev
);
4357 err
= get_super_block(super_list
,
4374 static int get_super_block(struct intel_super
**super_list
, char *devnm
, char *devname
,
4375 int major
, int minor
, int keep_fd
)
4377 struct intel_super
*s
= NULL
;
4389 sprintf(nm
, "%d:%d", major
, minor
);
4390 dfd
= dev_open(nm
, O_RDWR
);
4396 find_intel_hba_capability(dfd
, s
, devname
);
4397 err
= load_and_parse_mpb(dfd
, s
, NULL
, keep_fd
);
4399 /* retry the load if we might have raced against mdmon */
4400 if (err
== 3 && devnm
&& mdmon_running(devnm
))
4401 for (retry
= 0; retry
< 3; retry
++) {
4403 err
= load_and_parse_mpb(dfd
, s
, NULL
, keep_fd
);
4409 s
->next
= *super_list
;
4417 if ((dfd
>= 0) && (!keep_fd
))
4424 get_sra_super_block(int fd
, struct intel_super
**super_list
, char *devname
, int *max
, int keep_fd
)
4431 sra
= sysfs_read(fd
, NULL
, GET_LEVEL
|GET_VERSION
|GET_DEVS
|GET_STATE
);
4435 if (sra
->array
.major_version
!= -1 ||
4436 sra
->array
.minor_version
!= -2 ||
4437 strcmp(sra
->text_version
, "imsm") != 0) {
4442 devnm
= fd2devnm(fd
);
4443 for (sd
= sra
->devs
, i
= 0; sd
; sd
= sd
->next
, i
++) {
4444 if (get_super_block(super_list
, devnm
, devname
,
4445 sd
->disk
.major
, sd
->disk
.minor
, keep_fd
) != 0) {
4456 static int load_container_imsm(struct supertype
*st
, int fd
, char *devname
)
4458 return load_super_imsm_all(st
, fd
, &st
->sb
, devname
, NULL
, 1);
4462 static int load_super_imsm(struct supertype
*st
, int fd
, char *devname
)
4464 struct intel_super
*super
;
4468 if (test_partition(fd
))
4469 /* IMSM not allowed on partitions */
4472 free_super_imsm(st
);
4474 super
= alloc_super();
4475 /* Load hba and capabilities if they exist.
4476 * But do not preclude loading metadata in case capabilities or hba are
4477 * non-compliant and ignore_hw_compat is set.
4479 rv
= find_intel_hba_capability(fd
, super
, devname
);
4480 /* no orom/efi or non-intel hba of the disk */
4481 if ((rv
!= 0) && (st
->ignore_hw_compat
== 0)) {
4483 pr_err("No OROM/EFI properties for %s\n", devname
);
4487 rv
= load_and_parse_mpb(fd
, super
, devname
, 0);
4489 /* retry the load if we might have raced against mdmon */
4491 struct mdstat_ent
*mdstat
= mdstat_by_component(fd2devnm(fd
));
4493 if (mdstat
&& mdmon_running(mdstat
->devnm
) && getpid() != mdmon_pid(mdstat
->devnm
)) {
4494 for (retry
= 0; retry
< 3; retry
++) {
4496 rv
= load_and_parse_mpb(fd
, super
, devname
, 0);
4502 free_mdstat(mdstat
);
4507 pr_err("Failed to load all information sections on %s\n", devname
);
4513 if (st
->ss
== NULL
) {
4514 st
->ss
= &super_imsm
;
4515 st
->minor_version
= 0;
4516 st
->max_devs
= IMSM_MAX_DEVICES
;
4519 /* load migration record */
4520 if (load_imsm_migr_rec(super
, NULL
) == 0) {
4521 /* Check for unsupported migration features */
4522 if (check_mpb_migr_compatibility(super
) != 0) {
4523 pr_err("Unsupported migration detected");
4525 fprintf(stderr
, " on %s\n", devname
);
4527 fprintf(stderr
, " (IMSM).\n");
4535 static __u16
info_to_blocks_per_strip(mdu_array_info_t
*info
)
4537 if (info
->level
== 1)
4539 return info
->chunk_size
>> 9;
4542 static unsigned long long info_to_blocks_per_member(mdu_array_info_t
*info
,
4543 unsigned long long size
)
4545 if (info
->level
== 1)
4548 return (size
* 2) & ~(info_to_blocks_per_strip(info
) - 1);
4551 static void imsm_update_version_info(struct intel_super
*super
)
4553 /* update the version and attributes */
4554 struct imsm_super
*mpb
= super
->anchor
;
4556 struct imsm_dev
*dev
;
4557 struct imsm_map
*map
;
4560 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
4561 dev
= get_imsm_dev(super
, i
);
4562 map
= get_imsm_map(dev
, MAP_0
);
4563 if (__le32_to_cpu(dev
->size_high
) > 0)
4564 mpb
->attributes
|= MPB_ATTRIB_2TB
;
4566 /* FIXME detect when an array spans a port multiplier */
4568 mpb
->attributes
|= MPB_ATTRIB_PM
;
4571 if (mpb
->num_raid_devs
> 1 ||
4572 mpb
->attributes
!= MPB_ATTRIB_CHECKSUM_VERIFY
) {
4573 version
= MPB_VERSION_ATTRIBS
;
4574 switch (get_imsm_raid_level(map
)) {
4575 case 0: mpb
->attributes
|= MPB_ATTRIB_RAID0
; break;
4576 case 1: mpb
->attributes
|= MPB_ATTRIB_RAID1
; break;
4577 case 10: mpb
->attributes
|= MPB_ATTRIB_RAID10
; break;
4578 case 5: mpb
->attributes
|= MPB_ATTRIB_RAID5
; break;
4581 if (map
->num_members
>= 5)
4582 version
= MPB_VERSION_5OR6_DISK_ARRAY
;
4583 else if (dev
->status
== DEV_CLONE_N_GO
)
4584 version
= MPB_VERSION_CNG
;
4585 else if (get_imsm_raid_level(map
) == 5)
4586 version
= MPB_VERSION_RAID5
;
4587 else if (map
->num_members
>= 3)
4588 version
= MPB_VERSION_3OR4_DISK_ARRAY
;
4589 else if (get_imsm_raid_level(map
) == 1)
4590 version
= MPB_VERSION_RAID1
;
4592 version
= MPB_VERSION_RAID0
;
4594 strcpy(((char *) mpb
->sig
) + strlen(MPB_SIGNATURE
), version
);
4598 static int check_name(struct intel_super
*super
, char *name
, int quiet
)
4600 struct imsm_super
*mpb
= super
->anchor
;
4601 char *reason
= NULL
;
4604 if (strlen(name
) > MAX_RAID_SERIAL_LEN
)
4605 reason
= "must be 16 characters or less";
4607 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
4608 struct imsm_dev
*dev
= get_imsm_dev(super
, i
);
4610 if (strncmp((char *) dev
->volume
, name
, MAX_RAID_SERIAL_LEN
) == 0) {
4611 reason
= "already exists";
4616 if (reason
&& !quiet
)
4617 pr_err("imsm volume name %s\n", reason
);
4622 static int init_super_imsm_volume(struct supertype
*st
, mdu_array_info_t
*info
,
4623 unsigned long long size
, char *name
,
4624 char *homehost
, int *uuid
,
4625 long long data_offset
)
4627 /* We are creating a volume inside a pre-existing container.
4628 * so st->sb is already set.
4630 struct intel_super
*super
= st
->sb
;
4631 struct imsm_super
*mpb
= super
->anchor
;
4632 struct intel_dev
*dv
;
4633 struct imsm_dev
*dev
;
4634 struct imsm_vol
*vol
;
4635 struct imsm_map
*map
;
4636 int idx
= mpb
->num_raid_devs
;
4638 unsigned long long array_blocks
;
4639 size_t size_old
, size_new
;
4640 unsigned long long num_data_stripes
;
4642 if (super
->orom
&& mpb
->num_raid_devs
>= super
->orom
->vpa
) {
4643 pr_err("This imsm-container already has the maximum of %d volumes\n", super
->orom
->vpa
);
4647 /* ensure the mpb is large enough for the new data */
4648 size_old
= __le32_to_cpu(mpb
->mpb_size
);
4649 size_new
= disks_to_mpb_size(info
->nr_disks
);
4650 if (size_new
> size_old
) {
4652 size_t size_round
= ROUND_UP(size_new
, 512);
4654 if (posix_memalign(&mpb_new
, 512, size_round
) != 0) {
4655 pr_err("could not allocate new mpb\n");
4658 if (posix_memalign(&super
->migr_rec_buf
, 512,
4659 MIGR_REC_BUF_SIZE
) != 0) {
4660 pr_err("could not allocate migr_rec buffer\n");
4666 memcpy(mpb_new
, mpb
, size_old
);
4669 super
->anchor
= mpb_new
;
4670 mpb
->mpb_size
= __cpu_to_le32(size_new
);
4671 memset(mpb_new
+ size_old
, 0, size_round
- size_old
);
4673 super
->current_vol
= idx
;
4675 /* handle 'failed_disks' by either:
4676 * a) create dummy disk entries in the table if this the first
4677 * volume in the array. We add them here as this is the only
4678 * opportunity to add them. add_to_super_imsm_volume()
4679 * handles the non-failed disks and continues incrementing
4681 * b) validate that 'failed_disks' matches the current number
4682 * of missing disks if the container is populated
4684 if (super
->current_vol
== 0) {
4686 for (i
= 0; i
< info
->failed_disks
; i
++) {
4687 struct imsm_disk
*disk
;
4690 disk
= __get_imsm_disk(mpb
, i
);
4691 disk
->status
= CONFIGURED_DISK
| FAILED_DISK
;
4692 disk
->scsi_id
= __cpu_to_le32(~(__u32
)0);
4693 snprintf((char *) disk
->serial
, MAX_RAID_SERIAL_LEN
,
4696 find_missing(super
);
4701 for (d
= super
->missing
; d
; d
= d
->next
)
4703 if (info
->failed_disks
> missing
) {
4704 pr_err("unable to add 'missing' disk to container\n");
4709 if (!check_name(super
, name
, 0))
4711 dv
= xmalloc(sizeof(*dv
));
4712 dev
= xcalloc(1, sizeof(*dev
) + sizeof(__u32
) * (info
->raid_disks
- 1));
4713 strncpy((char *) dev
->volume
, name
, MAX_RAID_SERIAL_LEN
);
4714 array_blocks
= calc_array_size(info
->level
, info
->raid_disks
,
4715 info
->layout
, info
->chunk_size
,
4717 /* round array size down to closest MB */
4718 array_blocks
= (array_blocks
>> SECT_PER_MB_SHIFT
) << SECT_PER_MB_SHIFT
;
4720 dev
->size_low
= __cpu_to_le32((__u32
) array_blocks
);
4721 dev
->size_high
= __cpu_to_le32((__u32
) (array_blocks
>> 32));
4722 dev
->status
= (DEV_READ_COALESCING
| DEV_WRITE_COALESCING
);
4724 vol
->migr_state
= 0;
4725 set_migr_type(dev
, MIGR_INIT
);
4726 vol
->dirty
= !info
->state
;
4727 vol
->curr_migr_unit
= 0;
4728 map
= get_imsm_map(dev
, MAP_0
);
4729 set_pba_of_lba0(map
, super
->create_offset
);
4730 set_blocks_per_member(map
, info_to_blocks_per_member(info
, size
));
4731 map
->blocks_per_strip
= __cpu_to_le16(info_to_blocks_per_strip(info
));
4732 map
->failed_disk_num
= ~0;
4733 if (info
->level
> 0)
4734 map
->map_state
= (info
->state
? IMSM_T_STATE_NORMAL
4735 : IMSM_T_STATE_UNINITIALIZED
);
4737 map
->map_state
= info
->failed_disks
? IMSM_T_STATE_FAILED
:
4738 IMSM_T_STATE_NORMAL
;
4741 if (info
->level
== 1 && info
->raid_disks
> 2) {
4744 pr_err("imsm does not support more than 2 disksin a raid1 volume\n");
4748 map
->raid_level
= info
->level
;
4749 if (info
->level
== 10) {
4750 map
->raid_level
= 1;
4751 map
->num_domains
= info
->raid_disks
/ 2;
4752 } else if (info
->level
== 1)
4753 map
->num_domains
= info
->raid_disks
;
4755 map
->num_domains
= 1;
4757 /* info->size is only int so use the 'size' parameter instead */
4758 num_data_stripes
= (size
* 2) / info_to_blocks_per_strip(info
);
4759 num_data_stripes
/= map
->num_domains
;
4760 set_num_data_stripes(map
, num_data_stripes
);
4762 map
->num_members
= info
->raid_disks
;
4763 for (i
= 0; i
< map
->num_members
; i
++) {
4764 /* initialized in add_to_super */
4765 set_imsm_ord_tbl_ent(map
, i
, IMSM_ORD_REBUILD
);
4767 mpb
->num_raid_devs
++;
4770 dv
->index
= super
->current_vol
;
4771 dv
->next
= super
->devlist
;
4772 super
->devlist
= dv
;
4774 imsm_update_version_info(super
);
4779 static int init_super_imsm(struct supertype
*st
, mdu_array_info_t
*info
,
4780 unsigned long long size
, char *name
,
4781 char *homehost
, int *uuid
,
4782 unsigned long long data_offset
)
4784 /* This is primarily called by Create when creating a new array.
4785 * We will then get add_to_super called for each component, and then
4786 * write_init_super called to write it out to each device.
4787 * For IMSM, Create can create on fresh devices or on a pre-existing
4789 * To create on a pre-existing array a different method will be called.
4790 * This one is just for fresh drives.
4792 struct intel_super
*super
;
4793 struct imsm_super
*mpb
;
4797 if (data_offset
!= INVALID_SECTORS
) {
4798 pr_err("data-offset not supported by imsm\n");
4803 return init_super_imsm_volume(st
, info
, size
, name
, homehost
, uuid
,
4807 mpb_size
= disks_to_mpb_size(info
->nr_disks
);
4811 super
= alloc_super();
4812 if (super
&& posix_memalign(&super
->buf
, 512, mpb_size
) != 0) {
4817 pr_err("could not allocate superblock\n");
4820 if (posix_memalign(&super
->migr_rec_buf
, 512, MIGR_REC_BUF_SIZE
) != 0) {
4821 pr_err("could not allocate migr_rec buffer\n");
4826 memset(super
->buf
, 0, mpb_size
);
4828 mpb
->mpb_size
= __cpu_to_le32(mpb_size
);
4832 /* zeroing superblock */
4836 mpb
->attributes
= MPB_ATTRIB_CHECKSUM_VERIFY
;
4838 version
= (char *) mpb
->sig
;
4839 strcpy(version
, MPB_SIGNATURE
);
4840 version
+= strlen(MPB_SIGNATURE
);
4841 strcpy(version
, MPB_VERSION_RAID0
);
4847 static int add_to_super_imsm_volume(struct supertype
*st
, mdu_disk_info_t
*dk
,
4848 int fd
, char *devname
)
4850 struct intel_super
*super
= st
->sb
;
4851 struct imsm_super
*mpb
= super
->anchor
;
4852 struct imsm_disk
*_disk
;
4853 struct imsm_dev
*dev
;
4854 struct imsm_map
*map
;
4858 dev
= get_imsm_dev(super
, super
->current_vol
);
4859 map
= get_imsm_map(dev
, MAP_0
);
4861 if (! (dk
->state
& (1<<MD_DISK_SYNC
))) {
4862 pr_err("%s: Cannot add spare devices to IMSM volume\n",
4868 /* we're doing autolayout so grab the pre-marked (in
4869 * validate_geometry) raid_disk
4871 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
4872 if (dl
->raiddisk
== dk
->raid_disk
)
4875 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
4876 if (dl
->major
== dk
->major
&&
4877 dl
->minor
== dk
->minor
)
4882 pr_err("%s is not a member of the same container\n", devname
);
4886 /* add a pristine spare to the metadata */
4887 if (dl
->index
< 0) {
4888 dl
->index
= super
->anchor
->num_disks
;
4889 super
->anchor
->num_disks
++;
4891 /* Check the device has not already been added */
4892 slot
= get_imsm_disk_slot(map
, dl
->index
);
4894 (get_imsm_ord_tbl_ent(dev
, slot
, MAP_X
) & IMSM_ORD_REBUILD
) == 0) {
4895 pr_err("%s has been included in this array twice\n",
4899 set_imsm_ord_tbl_ent(map
, dk
->raid_disk
, dl
->index
);
4900 dl
->disk
.status
= CONFIGURED_DISK
;
4902 /* update size of 'missing' disks to be at least as large as the
4903 * largest acitve member (we only have dummy missing disks when
4904 * creating the first volume)
4906 if (super
->current_vol
== 0) {
4907 for (df
= super
->missing
; df
; df
= df
->next
) {
4908 if (total_blocks(&dl
->disk
) > total_blocks(&df
->disk
))
4909 set_total_blocks(&df
->disk
, total_blocks(&dl
->disk
));
4910 _disk
= __get_imsm_disk(mpb
, df
->index
);
4915 /* refresh unset/failed slots to point to valid 'missing' entries */
4916 for (df
= super
->missing
; df
; df
= df
->next
)
4917 for (slot
= 0; slot
< mpb
->num_disks
; slot
++) {
4918 __u32 ord
= get_imsm_ord_tbl_ent(dev
, slot
, MAP_X
);
4920 if ((ord
& IMSM_ORD_REBUILD
) == 0)
4922 set_imsm_ord_tbl_ent(map
, slot
, df
->index
| IMSM_ORD_REBUILD
);
4923 if (is_gen_migration(dev
)) {
4924 struct imsm_map
*map2
= get_imsm_map(dev
,
4926 int slot2
= get_imsm_disk_slot(map2
, df
->index
);
4927 if ((slot2
< map2
->num_members
) &&
4929 __u32 ord2
= get_imsm_ord_tbl_ent(dev
,
4932 if ((unsigned)df
->index
==
4934 set_imsm_ord_tbl_ent(map2
,
4940 dprintf("set slot:%d to missing disk:%d\n", slot
, df
->index
);
4944 /* if we are creating the first raid device update the family number */
4945 if (super
->current_vol
== 0) {
4947 struct imsm_dev
*_dev
= __get_imsm_dev(mpb
, 0);
4949 _disk
= __get_imsm_disk(mpb
, dl
->index
);
4950 if (!_dev
|| !_disk
) {
4951 pr_err("BUG mpb setup error\n");
4957 sum
+= __gen_imsm_checksum(mpb
);
4958 mpb
->family_num
= __cpu_to_le32(sum
);
4959 mpb
->orig_family_num
= mpb
->family_num
;
4961 super
->current_disk
= dl
;
4966 * Function marks disk as spare and restores disk serial
4967 * in case it was previously marked as failed by takeover operation
4969 * -1 : critical error
4970 * 0 : disk is marked as spare but serial is not set
4973 int mark_spare(struct dl
*disk
)
4975 __u8 serial
[MAX_RAID_SERIAL_LEN
];
4982 if (!imsm_read_serial(disk
->fd
, NULL
, serial
)) {
4983 /* Restore disk serial number, because takeover marks disk
4984 * as failed and adds to serial ':0' before it becomes
4987 serialcpy(disk
->serial
, serial
);
4988 serialcpy(disk
->disk
.serial
, serial
);
4991 disk
->disk
.status
= SPARE_DISK
;
4997 static int add_to_super_imsm(struct supertype
*st
, mdu_disk_info_t
*dk
,
4998 int fd
, char *devname
,
4999 unsigned long long data_offset
)
5001 struct intel_super
*super
= st
->sb
;
5003 unsigned long long size
;
5008 /* If we are on an RAID enabled platform check that the disk is
5009 * attached to the raid controller.
5010 * We do not need to test disks attachment for container based additions,
5011 * they shall be already tested when container was created/assembled.
5013 rv
= find_intel_hba_capability(fd
, super
, devname
);
5014 /* no orom/efi or non-intel hba of the disk */
5016 dprintf("capability: %p fd: %d ret: %d\n",
5017 super
->orom
, fd
, rv
);
5021 if (super
->current_vol
>= 0)
5022 return add_to_super_imsm_volume(st
, dk
, fd
, devname
);
5025 dd
= xcalloc(sizeof(*dd
), 1);
5026 dd
->major
= major(stb
.st_rdev
);
5027 dd
->minor
= minor(stb
.st_rdev
);
5028 dd
->devname
= devname
? xstrdup(devname
) : NULL
;
5031 dd
->action
= DISK_ADD
;
5032 rv
= imsm_read_serial(fd
, devname
, dd
->serial
);
5034 pr_err("failed to retrieve scsi serial, aborting\n");
5039 get_dev_size(fd
, NULL
, &size
);
5040 /* clear migr_rec when adding disk to container */
5041 memset(super
->migr_rec_buf
, 0, MIGR_REC_BUF_SIZE
);
5042 if (lseek64(fd
, size
- MIGR_REC_POSITION
, SEEK_SET
) >= 0) {
5043 if (write(fd
, super
->migr_rec_buf
,
5044 MIGR_REC_BUF_SIZE
) != MIGR_REC_BUF_SIZE
)
5045 perror("Write migr_rec failed");
5049 serialcpy(dd
->disk
.serial
, dd
->serial
);
5050 set_total_blocks(&dd
->disk
, size
);
5051 if (__le32_to_cpu(dd
->disk
.total_blocks_hi
) > 0) {
5052 struct imsm_super
*mpb
= super
->anchor
;
5053 mpb
->attributes
|= MPB_ATTRIB_2TB_DISK
;
5056 if (sysfs_disk_to_scsi_id(fd
, &id
) == 0)
5057 dd
->disk
.scsi_id
= __cpu_to_le32(id
);
5059 dd
->disk
.scsi_id
= __cpu_to_le32(0);
5061 if (st
->update_tail
) {
5062 dd
->next
= super
->disk_mgmt_list
;
5063 super
->disk_mgmt_list
= dd
;
5065 dd
->next
= super
->disks
;
5067 super
->updates_pending
++;
5073 static int remove_from_super_imsm(struct supertype
*st
, mdu_disk_info_t
*dk
)
5075 struct intel_super
*super
= st
->sb
;
5078 /* remove from super works only in mdmon - for communication
5079 * manager - monitor. Check if communication memory buffer
5082 if (!st
->update_tail
) {
5083 pr_err("shall be used in mdmon context only\n");
5086 dd
= xcalloc(1, sizeof(*dd
));
5087 dd
->major
= dk
->major
;
5088 dd
->minor
= dk
->minor
;
5091 dd
->action
= DISK_REMOVE
;
5093 dd
->next
= super
->disk_mgmt_list
;
5094 super
->disk_mgmt_list
= dd
;
5099 static int store_imsm_mpb(int fd
, struct imsm_super
*mpb
);
5103 struct imsm_super anchor
;
5104 } spare_record
__attribute__ ((aligned(512)));
5106 /* spare records have their own family number and do not have any defined raid
5109 static int write_super_imsm_spares(struct intel_super
*super
, int doclose
)
5111 struct imsm_super
*mpb
= super
->anchor
;
5112 struct imsm_super
*spare
= &spare_record
.anchor
;
5116 spare
->mpb_size
= __cpu_to_le32(sizeof(struct imsm_super
));
5117 spare
->generation_num
= __cpu_to_le32(1UL);
5118 spare
->attributes
= MPB_ATTRIB_CHECKSUM_VERIFY
;
5119 spare
->num_disks
= 1;
5120 spare
->num_raid_devs
= 0;
5121 spare
->cache_size
= mpb
->cache_size
;
5122 spare
->pwr_cycle_count
= __cpu_to_le32(1);
5124 snprintf((char *) spare
->sig
, MAX_SIGNATURE_LENGTH
,
5125 MPB_SIGNATURE MPB_VERSION_RAID0
);
5127 for (d
= super
->disks
; d
; d
= d
->next
) {
5131 spare
->disk
[0] = d
->disk
;
5132 if (__le32_to_cpu(d
->disk
.total_blocks_hi
) > 0)
5133 spare
->attributes
|= MPB_ATTRIB_2TB_DISK
;
5135 sum
= __gen_imsm_checksum(spare
);
5136 spare
->family_num
= __cpu_to_le32(sum
);
5137 spare
->orig_family_num
= 0;
5138 sum
= __gen_imsm_checksum(spare
);
5139 spare
->check_sum
= __cpu_to_le32(sum
);
5141 if (store_imsm_mpb(d
->fd
, spare
)) {
5142 pr_err("failed for device %d:%d %s\n",
5143 d
->major
, d
->minor
, strerror(errno
));
5155 static int write_super_imsm(struct supertype
*st
, int doclose
)
5157 struct intel_super
*super
= st
->sb
;
5158 struct imsm_super
*mpb
= super
->anchor
;
5164 __u32 mpb_size
= sizeof(struct imsm_super
) - sizeof(struct imsm_disk
);
5166 int clear_migration_record
= 1;
5168 /* 'generation' is incremented everytime the metadata is written */
5169 generation
= __le32_to_cpu(mpb
->generation_num
);
5171 mpb
->generation_num
= __cpu_to_le32(generation
);
5173 /* fix up cases where previous mdadm releases failed to set
5176 if (mpb
->orig_family_num
== 0)
5177 mpb
->orig_family_num
= mpb
->family_num
;
5179 for (d
= super
->disks
; d
; d
= d
->next
) {
5183 mpb
->disk
[d
->index
] = d
->disk
;
5187 for (d
= super
->missing
; d
; d
= d
->next
) {
5188 mpb
->disk
[d
->index
] = d
->disk
;
5191 mpb
->num_disks
= num_disks
;
5192 mpb_size
+= sizeof(struct imsm_disk
) * mpb
->num_disks
;
5194 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
5195 struct imsm_dev
*dev
= __get_imsm_dev(mpb
, i
);
5196 struct imsm_dev
*dev2
= get_imsm_dev(super
, i
);
5198 imsm_copy_dev(dev
, dev2
);
5199 mpb_size
+= sizeof_imsm_dev(dev
, 0);
5201 if (is_gen_migration(dev2
))
5202 clear_migration_record
= 0;
5204 mpb_size
+= __le32_to_cpu(mpb
->bbm_log_size
);
5205 mpb
->mpb_size
= __cpu_to_le32(mpb_size
);
5207 /* recalculate checksum */
5208 sum
= __gen_imsm_checksum(mpb
);
5209 mpb
->check_sum
= __cpu_to_le32(sum
);
5211 if (super
->clean_migration_record_by_mdmon
) {
5212 clear_migration_record
= 1;
5213 super
->clean_migration_record_by_mdmon
= 0;
5215 if (clear_migration_record
)
5216 memset(super
->migr_rec_buf
, 0, MIGR_REC_BUF_SIZE
);
5218 /* write the mpb for disks that compose raid devices */
5219 for (d
= super
->disks
; d
; d
= d
->next
) {
5220 if (d
->index
< 0 || is_failed(&d
->disk
))
5223 if (clear_migration_record
) {
5224 unsigned long long dsize
;
5226 get_dev_size(d
->fd
, NULL
, &dsize
);
5227 if (lseek64(d
->fd
, dsize
- 512, SEEK_SET
) >= 0) {
5228 if (write(d
->fd
, super
->migr_rec_buf
,
5229 MIGR_REC_BUF_SIZE
) != MIGR_REC_BUF_SIZE
)
5230 perror("Write migr_rec failed");
5234 if (store_imsm_mpb(d
->fd
, mpb
))
5236 "failed for device %d:%d (fd: %d)%s\n",
5238 d
->fd
, strerror(errno
));
5247 return write_super_imsm_spares(super
, doclose
);
5252 static int create_array(struct supertype
*st
, int dev_idx
)
5255 struct imsm_update_create_array
*u
;
5256 struct intel_super
*super
= st
->sb
;
5257 struct imsm_dev
*dev
= get_imsm_dev(super
, dev_idx
);
5258 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
5259 struct disk_info
*inf
;
5260 struct imsm_disk
*disk
;
5263 len
= sizeof(*u
) - sizeof(*dev
) + sizeof_imsm_dev(dev
, 0) +
5264 sizeof(*inf
) * map
->num_members
;
5266 u
->type
= update_create_array
;
5267 u
->dev_idx
= dev_idx
;
5268 imsm_copy_dev(&u
->dev
, dev
);
5269 inf
= get_disk_info(u
);
5270 for (i
= 0; i
< map
->num_members
; i
++) {
5271 int idx
= get_imsm_disk_idx(dev
, i
, MAP_X
);
5273 disk
= get_imsm_disk(super
, idx
);
5275 disk
= get_imsm_missing(super
, idx
);
5276 serialcpy(inf
[i
].serial
, disk
->serial
);
5278 append_metadata_update(st
, u
, len
);
5283 static int mgmt_disk(struct supertype
*st
)
5285 struct intel_super
*super
= st
->sb
;
5287 struct imsm_update_add_remove_disk
*u
;
5289 if (!super
->disk_mgmt_list
)
5294 u
->type
= update_add_remove_disk
;
5295 append_metadata_update(st
, u
, len
);
5300 static int write_init_super_imsm(struct supertype
*st
)
5302 struct intel_super
*super
= st
->sb
;
5303 int current_vol
= super
->current_vol
;
5305 /* we are done with current_vol reset it to point st at the container */
5306 super
->current_vol
= -1;
5308 if (st
->update_tail
) {
5309 /* queue the recently created array / added disk
5310 * as a metadata update */
5313 /* determine if we are creating a volume or adding a disk */
5314 if (current_vol
< 0) {
5315 /* in the mgmt (add/remove) disk case we are running
5316 * in mdmon context, so don't close fd's
5318 return mgmt_disk(st
);
5320 rv
= create_array(st
, current_vol
);
5325 for (d
= super
->disks
; d
; d
= d
->next
)
5326 Kill(d
->devname
, NULL
, 0, -1, 1);
5327 return write_super_imsm(st
, 1);
5332 static int store_super_imsm(struct supertype
*st
, int fd
)
5334 struct intel_super
*super
= st
->sb
;
5335 struct imsm_super
*mpb
= super
? super
->anchor
: NULL
;
5341 return store_imsm_mpb(fd
, mpb
);
5347 static int imsm_bbm_log_size(struct imsm_super
*mpb
)
5349 return __le32_to_cpu(mpb
->bbm_log_size
);
5353 static int validate_geometry_imsm_container(struct supertype
*st
, int level
,
5354 int layout
, int raiddisks
, int chunk
,
5355 unsigned long long size
,
5356 unsigned long long data_offset
,
5358 unsigned long long *freesize
,
5362 unsigned long long ldsize
;
5363 struct intel_super
*super
=NULL
;
5366 if (level
!= LEVEL_CONTAINER
)
5371 fd
= open(dev
, O_RDONLY
|O_EXCL
, 0);
5374 pr_err("imsm: Cannot open %s: %s\n",
5375 dev
, strerror(errno
));
5378 if (!get_dev_size(fd
, dev
, &ldsize
)) {
5383 /* capabilities retrieve could be possible
5384 * note that there is no fd for the disks in array.
5386 super
= alloc_super();
5387 rv
= find_intel_hba_capability(fd
, super
, verbose
> 0 ? dev
: NULL
);
5391 fd2devname(fd
, str
);
5392 dprintf("fd: %d %s orom: %p rv: %d raiddisk: %d\n",
5393 fd
, str
, super
->orom
, rv
, raiddisks
);
5395 /* no orom/efi or non-intel hba of the disk */
5402 if (raiddisks
> super
->orom
->tds
) {
5404 pr_err("%d exceeds maximum number of platform supported disks: %d\n",
5405 raiddisks
, super
->orom
->tds
);
5409 if ((super
->orom
->attr
& IMSM_OROM_ATTR_2TB_DISK
) == 0 &&
5410 (ldsize
>> 9) >> 32 > 0) {
5412 pr_err("%s exceeds maximum platform supported size\n", dev
);
5418 *freesize
= avail_size_imsm(st
, ldsize
>> 9, data_offset
);
5424 static unsigned long long find_size(struct extent
*e
, int *idx
, int num_extents
)
5426 const unsigned long long base_start
= e
[*idx
].start
;
5427 unsigned long long end
= base_start
+ e
[*idx
].size
;
5430 if (base_start
== end
)
5434 for (i
= *idx
; i
< num_extents
; i
++) {
5435 /* extend overlapping extents */
5436 if (e
[i
].start
>= base_start
&&
5437 e
[i
].start
<= end
) {
5440 if (e
[i
].start
+ e
[i
].size
> end
)
5441 end
= e
[i
].start
+ e
[i
].size
;
5442 } else if (e
[i
].start
> end
) {
5448 return end
- base_start
;
5451 static unsigned long long merge_extents(struct intel_super
*super
, int sum_extents
)
5453 /* build a composite disk with all known extents and generate a new
5454 * 'maxsize' given the "all disks in an array must share a common start
5455 * offset" constraint
5457 struct extent
*e
= xcalloc(sum_extents
, sizeof(*e
));
5461 unsigned long long pos
;
5462 unsigned long long start
= 0;
5463 unsigned long long maxsize
;
5464 unsigned long reserve
;
5466 /* coalesce and sort all extents. also, check to see if we need to
5467 * reserve space between member arrays
5470 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
5473 for (i
= 0; i
< dl
->extent_cnt
; i
++)
5476 qsort(e
, sum_extents
, sizeof(*e
), cmp_extent
);
5481 while (i
< sum_extents
) {
5482 e
[j
].start
= e
[i
].start
;
5483 e
[j
].size
= find_size(e
, &i
, sum_extents
);
5485 if (e
[j
-1].size
== 0)
5494 unsigned long long esize
;
5496 esize
= e
[i
].start
- pos
;
5497 if (esize
>= maxsize
) {
5502 pos
= e
[i
].start
+ e
[i
].size
;
5504 } while (e
[i
-1].size
);
5510 /* FIXME assumes volume at offset 0 is the first volume in a
5513 if (start_extent
> 0)
5514 reserve
= IMSM_RESERVED_SECTORS
; /* gap between raid regions */
5518 if (maxsize
< reserve
)
5521 super
->create_offset
= ~((unsigned long long) 0);
5522 if (start
+ reserve
> super
->create_offset
)
5523 return 0; /* start overflows create_offset */
5524 super
->create_offset
= start
+ reserve
;
5526 return maxsize
- reserve
;
5529 static int is_raid_level_supported(const struct imsm_orom
*orom
, int level
, int raiddisks
)
5531 if (level
< 0 || level
== 6 || level
== 4)
5534 /* if we have an orom prevent invalid raid levels */
5537 case 0: return imsm_orom_has_raid0(orom
);
5540 return imsm_orom_has_raid1e(orom
);
5541 return imsm_orom_has_raid1(orom
) && raiddisks
== 2;
5542 case 10: return imsm_orom_has_raid10(orom
) && raiddisks
== 4;
5543 case 5: return imsm_orom_has_raid5(orom
) && raiddisks
> 2;
5546 return 1; /* not on an Intel RAID platform so anything goes */
5552 active_arrays_by_format(char *name
, char* hba
, struct md_list
**devlist
,
5553 int dpa
, int verbose
)
5555 struct mdstat_ent
*mdstat
= mdstat_read(0, 0);
5556 struct mdstat_ent
*memb
= NULL
;
5559 struct md_list
*dv
= NULL
;
5562 for (memb
= mdstat
; memb
; memb
= memb
->next
) {
5563 if (memb
->metadata_version
&&
5564 (strncmp(memb
->metadata_version
, "external:", 9) == 0) &&
5565 (strcmp(&memb
->metadata_version
[9], name
) == 0) &&
5566 !is_subarray(memb
->metadata_version
+9) &&
5568 struct dev_member
*dev
= memb
->members
;
5570 while(dev
&& (fd
< 0)) {
5571 char *path
= xmalloc(strlen(dev
->name
) + strlen("/dev/") + 1);
5572 num
= sprintf(path
, "%s%s", "/dev/", dev
->name
);
5574 fd
= open(path
, O_RDONLY
, 0);
5575 if ((num
<= 0) || (fd
< 0)) {
5576 pr_vrb(": Cannot open %s: %s\n",
5577 dev
->name
, strerror(errno
));
5583 if ((fd
>= 0) && disk_attached_to_hba(fd
, hba
)) {
5584 struct mdstat_ent
*vol
;
5585 for (vol
= mdstat
; vol
; vol
= vol
->next
) {
5586 if ((vol
->active
> 0) &&
5587 vol
->metadata_version
&&
5588 is_container_member(vol
, memb
->dev
)) {
5593 if (*devlist
&& (found
< dpa
)) {
5594 dv
= xcalloc(1, sizeof(*dv
));
5595 dv
->devname
= xmalloc(strlen(memb
->dev
) + strlen("/dev/") + 1);
5596 sprintf(dv
->devname
, "%s%s", "/dev/", memb
->dev
);
5599 dv
->next
= *devlist
;
5607 free_mdstat(mdstat
);
5612 static struct md_list
*
5613 get_loop_devices(void)
5616 struct md_list
*devlist
= NULL
;
5617 struct md_list
*dv
= NULL
;
5619 for(i
= 0; i
< 12; i
++) {
5620 dv
= xcalloc(1, sizeof(*dv
));
5621 dv
->devname
= xmalloc(40);
5622 sprintf(dv
->devname
, "/dev/loop%d", i
);
5630 static struct md_list
*
5631 get_devices(const char *hba_path
)
5633 struct md_list
*devlist
= NULL
;
5634 struct md_list
*dv
= NULL
;
5640 devlist
= get_loop_devices();
5643 /* scroll through /sys/dev/block looking for devices attached to
5646 dir
= opendir("/sys/dev/block");
5647 for (ent
= dir
? readdir(dir
) : NULL
; ent
; ent
= readdir(dir
)) {
5652 if (sscanf(ent
->d_name
, "%d:%d", &major
, &minor
) != 2)
5654 path
= devt_to_devpath(makedev(major
, minor
));
5657 if (!path_attached_to_hba(path
, hba_path
)) {
5664 fd
= dev_open(ent
->d_name
, O_RDONLY
);
5666 fd2devname(fd
, buf
);
5669 pr_err("cannot open device: %s\n",
5674 dv
= xcalloc(1, sizeof(*dv
));
5675 dv
->devname
= xstrdup(buf
);
5682 devlist
= devlist
->next
;
5692 count_volumes_list(struct md_list
*devlist
, char *homehost
,
5693 int verbose
, int *found
)
5695 struct md_list
*tmpdev
;
5697 struct supertype
*st
= NULL
;
5699 /* first walk the list of devices to find a consistent set
5700 * that match the criterea, if that is possible.
5701 * We flag the ones we like with 'used'.
5704 st
= match_metadata_desc_imsm("imsm");
5706 pr_vrb(": cannot allocate memory for imsm supertype\n");
5710 for (tmpdev
= devlist
; tmpdev
; tmpdev
= tmpdev
->next
) {
5711 char *devname
= tmpdev
->devname
;
5713 struct supertype
*tst
;
5715 if (tmpdev
->used
> 1)
5717 tst
= dup_super(st
);
5719 pr_vrb(": cannot allocate memory for imsm supertype\n");
5722 tmpdev
->container
= 0;
5723 dfd
= dev_open(devname
, O_RDONLY
|O_EXCL
);
5725 dprintf("cannot open device %s: %s\n",
5726 devname
, strerror(errno
));
5728 } else if (fstat(dfd
, &stb
)< 0) {
5730 dprintf("fstat failed for %s: %s\n",
5731 devname
, strerror(errno
));
5733 } else if ((stb
.st_mode
& S_IFMT
) != S_IFBLK
) {
5734 dprintf("%s is not a block device.\n",
5737 } else if (must_be_container(dfd
)) {
5738 struct supertype
*cst
;
5739 cst
= super_by_fd(dfd
, NULL
);
5741 dprintf("cannot recognize container type %s\n",
5744 } else if (tst
->ss
!= st
->ss
) {
5745 dprintf("non-imsm container - ignore it: %s\n",
5748 } else if (!tst
->ss
->load_container
||
5749 tst
->ss
->load_container(tst
, dfd
, NULL
))
5752 tmpdev
->container
= 1;
5755 cst
->ss
->free_super(cst
);
5757 tmpdev
->st_rdev
= stb
.st_rdev
;
5758 if (tst
->ss
->load_super(tst
,dfd
, NULL
)) {
5759 dprintf("no RAID superblock on %s\n",
5762 } else if (tst
->ss
->compare_super
== NULL
) {
5763 dprintf("Cannot assemble %s metadata on %s\n",
5764 tst
->ss
->name
, devname
);
5770 if (tmpdev
->used
== 2 || tmpdev
->used
== 4) {
5771 /* Ignore unrecognised devices during auto-assembly */
5776 tst
->ss
->getinfo_super(tst
, &info
, NULL
);
5778 if (st
->minor_version
== -1)
5779 st
->minor_version
= tst
->minor_version
;
5781 if (memcmp(info
.uuid
, uuid_zero
,
5782 sizeof(int[4])) == 0) {
5783 /* this is a floating spare. It cannot define
5784 * an array unless there are no more arrays of
5785 * this type to be found. It can be included
5786 * in an array of this type though.
5792 if (st
->ss
!= tst
->ss
||
5793 st
->minor_version
!= tst
->minor_version
||
5794 st
->ss
->compare_super(st
, tst
) != 0) {
5795 /* Some mismatch. If exactly one array matches this host,
5796 * we can resolve on that one.
5797 * Or, if we are auto assembling, we just ignore the second
5800 dprintf("superblock on %s doesn't match others - assembly aborted\n",
5806 dprintf("found: devname: %s\n", devname
);
5810 tst
->ss
->free_super(tst
);
5814 if ((err
= load_super_imsm_all(st
, -1, &st
->sb
, NULL
, devlist
, 0)) == 0) {
5815 struct mdinfo
*iter
, *head
= st
->ss
->container_content(st
, NULL
);
5816 for (iter
= head
; iter
; iter
= iter
->next
) {
5817 dprintf("content->text_version: %s vol\n",
5818 iter
->text_version
);
5819 if (iter
->array
.state
& (1<<MD_SB_BLOCK_VOLUME
)) {
5820 /* do not assemble arrays with unsupported
5822 dprintf("Cannot activate member %s.\n",
5823 iter
->text_version
);
5830 dprintf("No valid super block on device list: err: %d %p\n",
5834 dprintf("no more devices to examine\n");
5837 for (tmpdev
= devlist
; tmpdev
; tmpdev
= tmpdev
->next
) {
5838 if ((tmpdev
->used
== 1) && (tmpdev
->found
)) {
5840 if (count
< tmpdev
->found
)
5843 count
-= tmpdev
->found
;
5846 if (tmpdev
->used
== 1)
5851 st
->ss
->free_super(st
);
5856 count_volumes(struct intel_hba
*hba
, int dpa
, int verbose
)
5858 struct sys_dev
*idev
, *intel_devices
= find_intel_devices();
5860 const struct orom_entry
*entry
;
5861 struct devid_list
*dv
, *devid_list
;
5863 if (!hba
|| !hba
->path
)
5866 for (idev
= intel_devices
; idev
; idev
= idev
->next
) {
5867 if (strstr(idev
->path
, hba
->path
))
5871 if (!idev
|| !idev
->dev_id
)
5874 entry
= get_orom_entry_by_device_id(idev
->dev_id
);
5876 if (!entry
|| !entry
->devid_list
)
5879 devid_list
= entry
->devid_list
;
5880 for (dv
= devid_list
; dv
; dv
= dv
->next
) {
5882 struct md_list
*devlist
= NULL
;
5883 struct sys_dev
*device
= device_by_id(dv
->devid
);
5888 hba_path
= device
->path
;
5892 devlist
= get_devices(hba_path
);
5893 /* if no intel devices return zero volumes */
5894 if (devlist
== NULL
)
5897 count
+= active_arrays_by_format("imsm", hba_path
, &devlist
, dpa
, verbose
);
5898 dprintf("path: %s active arrays: %d\n", hba_path
, count
);
5899 if (devlist
== NULL
)
5903 count
+= count_volumes_list(devlist
,
5907 dprintf("found %d count: %d\n", found
, count
);
5910 dprintf("path: %s total number of volumes: %d\n", hba_path
, count
);
5913 struct md_list
*dv
= devlist
;
5914 devlist
= devlist
->next
;
5922 static int imsm_default_chunk(const struct imsm_orom
*orom
)
5924 /* up to 512 if the plaform supports it, otherwise the platform max.
5925 * 128 if no platform detected
5927 int fs
= max(7, orom
? fls(orom
->sss
) : 0);
5929 return min(512, (1 << fs
));
5933 validate_geometry_imsm_orom(struct intel_super
*super
, int level
, int layout
,
5934 int raiddisks
, int *chunk
, unsigned long long size
, int verbose
)
5936 /* check/set platform and metadata limits/defaults */
5937 if (super
->orom
&& raiddisks
> super
->orom
->dpa
) {
5938 pr_vrb(": platform supports a maximum of %d disks per array\n",
5943 /* capabilities of OROM tested - copied from validate_geometry_imsm_volume */
5944 if (!is_raid_level_supported(super
->orom
, level
, raiddisks
)) {
5945 pr_vrb(": platform does not support raid%d with %d disk%s\n",
5946 level
, raiddisks
, raiddisks
> 1 ? "s" : "");
5950 if (*chunk
== 0 || *chunk
== UnSet
)
5951 *chunk
= imsm_default_chunk(super
->orom
);
5953 if (super
->orom
&& !imsm_orom_has_chunk(super
->orom
, *chunk
)) {
5954 pr_vrb(": platform does not support a chunk size of: %d\n", *chunk
);
5958 if (layout
!= imsm_level_to_layout(level
)) {
5960 pr_vrb(": imsm raid 5 only supports the left-asymmetric layout\n");
5961 else if (level
== 10)
5962 pr_vrb(": imsm raid 10 only supports the n2 layout\n");
5964 pr_vrb(": imsm unknown layout %#x for this raid level %d\n",
5969 if (super
->orom
&& (super
->orom
->attr
& IMSM_OROM_ATTR_2TB
) == 0 &&
5970 (calc_array_size(level
, raiddisks
, layout
, *chunk
, size
) >> 32) > 0) {
5971 pr_vrb(": platform does not support a volume size over 2TB\n");
5978 /* validate_geometry_imsm_volume - lifted from validate_geometry_ddf_bvd
5979 * FIX ME add ahci details
5981 static int validate_geometry_imsm_volume(struct supertype
*st
, int level
,
5982 int layout
, int raiddisks
, int *chunk
,
5983 unsigned long long size
,
5984 unsigned long long data_offset
,
5986 unsigned long long *freesize
,
5990 struct intel_super
*super
= st
->sb
;
5991 struct imsm_super
*mpb
;
5993 unsigned long long pos
= 0;
5994 unsigned long long maxsize
;
5998 /* We must have the container info already read in. */
6002 mpb
= super
->anchor
;
6004 if (!validate_geometry_imsm_orom(super
, level
, layout
, raiddisks
, chunk
, size
, verbose
)) {
6005 pr_err("RAID gemetry validation failed. Cannot proceed with the action(s).\n");
6009 /* General test: make sure there is space for
6010 * 'raiddisks' device extents of size 'size' at a given
6013 unsigned long long minsize
= size
;
6014 unsigned long long start_offset
= MaxSector
;
6017 minsize
= MPB_SECTOR_CNT
+ IMSM_RESERVED_SECTORS
;
6018 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
6023 e
= get_extents(super
, dl
);
6026 unsigned long long esize
;
6027 esize
= e
[i
].start
- pos
;
6028 if (esize
>= minsize
)
6030 if (found
&& start_offset
== MaxSector
) {
6033 } else if (found
&& pos
!= start_offset
) {
6037 pos
= e
[i
].start
+ e
[i
].size
;
6039 } while (e
[i
-1].size
);
6044 if (dcnt
< raiddisks
) {
6046 pr_err("imsm: Not enough devices with space for this array (%d < %d)\n",
6053 /* This device must be a member of the set */
6054 if (stat(dev
, &stb
) < 0)
6056 if ((S_IFMT
& stb
.st_mode
) != S_IFBLK
)
6058 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
6059 if (dl
->major
== (int)major(stb
.st_rdev
) &&
6060 dl
->minor
== (int)minor(stb
.st_rdev
))
6065 pr_err("%s is not in the same imsm set\n", dev
);
6067 } else if (super
->orom
&& dl
->index
< 0 && mpb
->num_raid_devs
) {
6068 /* If a volume is present then the current creation attempt
6069 * cannot incorporate new spares because the orom may not
6070 * understand this configuration (all member disks must be
6071 * members of each array in the container).
6073 pr_err("%s is a spare and a volume is already defined for this container\n", dev
);
6074 pr_err("The option-rom requires all member disks to be a member of all volumes\n");
6076 } else if (super
->orom
&& mpb
->num_raid_devs
> 0 &&
6077 mpb
->num_disks
!= raiddisks
) {
6078 pr_err("The option-rom requires all member disks to be a member of all volumes\n");
6082 /* retrieve the largest free space block */
6083 e
= get_extents(super
, dl
);
6088 unsigned long long esize
;
6090 esize
= e
[i
].start
- pos
;
6091 if (esize
>= maxsize
)
6093 pos
= e
[i
].start
+ e
[i
].size
;
6095 } while (e
[i
-1].size
);
6100 pr_err("unable to determine free space for: %s\n",
6104 if (maxsize
< size
) {
6106 pr_err("%s not enough space (%llu < %llu)\n",
6107 dev
, maxsize
, size
);
6111 /* count total number of extents for merge */
6113 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
6115 i
+= dl
->extent_cnt
;
6117 maxsize
= merge_extents(super
, i
);
6119 if (!check_env("IMSM_NO_PLATFORM") &&
6120 mpb
->num_raid_devs
> 0 && size
&& size
!= maxsize
) {
6121 pr_err("attempting to create a second volume with size less then remaining space. Aborting...\n");
6125 if (maxsize
< size
|| maxsize
== 0) {
6128 pr_err("no free space left on device. Aborting...\n");
6130 pr_err("not enough space to create volume of given size (%llu < %llu). Aborting...\n",
6136 *freesize
= maxsize
;
6139 int count
= count_volumes(super
->hba
,
6140 super
->orom
->dpa
, verbose
);
6141 if (super
->orom
->vphba
<= count
) {
6142 pr_vrb(": platform does not support more than %d raid volumes.\n",
6143 super
->orom
->vphba
);
6150 static int imsm_get_free_size(struct supertype
*st
, int raiddisks
,
6151 unsigned long long size
, int chunk
,
6152 unsigned long long *freesize
)
6154 struct intel_super
*super
= st
->sb
;
6155 struct imsm_super
*mpb
= super
->anchor
;
6160 unsigned long long maxsize
;
6161 unsigned long long minsize
;
6165 /* find the largest common start free region of the possible disks */
6169 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
6175 /* don't activate new spares if we are orom constrained
6176 * and there is already a volume active in the container
6178 if (super
->orom
&& dl
->index
< 0 && mpb
->num_raid_devs
)
6181 e
= get_extents(super
, dl
);
6184 for (i
= 1; e
[i
-1].size
; i
++)
6192 maxsize
= merge_extents(super
, extent_cnt
);
6196 minsize
= chunk
* 2;
6198 if (cnt
< raiddisks
||
6199 (super
->orom
&& used
&& used
!= raiddisks
) ||
6200 maxsize
< minsize
||
6202 pr_err("not enough devices with space to create array.\n");
6203 return 0; /* No enough free spaces large enough */
6214 if (!check_env("IMSM_NO_PLATFORM") &&
6215 mpb
->num_raid_devs
> 0 && size
&& size
!= maxsize
) {
6216 pr_err("attempting to create a second volume with size less then remaining space. Aborting...\n");
6220 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
6222 dl
->raiddisk
= cnt
++;
6226 dprintf("imsm: imsm_get_free_size() returns : %llu\n", size
);
6231 static int reserve_space(struct supertype
*st
, int raiddisks
,
6232 unsigned long long size
, int chunk
,
6233 unsigned long long *freesize
)
6235 struct intel_super
*super
= st
->sb
;
6240 rv
= imsm_get_free_size(st
, raiddisks
, size
, chunk
, freesize
);
6243 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
6245 dl
->raiddisk
= cnt
++;
6252 static int validate_geometry_imsm(struct supertype
*st
, int level
, int layout
,
6253 int raiddisks
, int *chunk
, unsigned long long size
,
6254 unsigned long long data_offset
,
6255 char *dev
, unsigned long long *freesize
,
6263 * if given unused devices create a container
6264 * if given given devices in a container create a member volume
6266 if (level
== LEVEL_CONTAINER
) {
6267 /* Must be a fresh device to add to a container */
6268 return validate_geometry_imsm_container(st
, level
, layout
,
6278 struct intel_super
*super
= st
->sb
;
6279 if (!validate_geometry_imsm_orom(st
->sb
, level
, layout
,
6280 raiddisks
, chunk
, size
,
6283 /* we are being asked to automatically layout a
6284 * new volume based on the current contents of
6285 * the container. If the the parameters can be
6286 * satisfied reserve_space will record the disks,
6287 * start offset, and size of the volume to be
6288 * created. add_to_super and getinfo_super
6289 * detect when autolayout is in progress.
6291 /* assuming that freesize is always given when array is
6293 if (super
->orom
&& freesize
) {
6295 count
= count_volumes(super
->hba
,
6296 super
->orom
->dpa
, verbose
);
6297 if (super
->orom
->vphba
<= count
) {
6298 pr_vrb(": platform does not support more than %d raid volumes.\n",
6299 super
->orom
->vphba
);
6304 return reserve_space(st
, raiddisks
, size
,
6310 /* creating in a given container */
6311 return validate_geometry_imsm_volume(st
, level
, layout
,
6312 raiddisks
, chunk
, size
,
6314 dev
, freesize
, verbose
);
6317 /* This device needs to be a device in an 'imsm' container */
6318 fd
= open(dev
, O_RDONLY
|O_EXCL
, 0);
6321 pr_err("Cannot create this array on device %s\n",
6326 if (errno
!= EBUSY
|| (fd
= open(dev
, O_RDONLY
, 0)) < 0) {
6328 pr_err("Cannot open %s: %s\n",
6329 dev
, strerror(errno
));
6332 /* Well, it is in use by someone, maybe an 'imsm' container. */
6333 cfd
= open_container(fd
);
6337 pr_err("Cannot use %s: It is busy\n",
6341 sra
= sysfs_read(cfd
, NULL
, GET_VERSION
);
6342 if (sra
&& sra
->array
.major_version
== -1 &&
6343 strcmp(sra
->text_version
, "imsm") == 0)
6347 /* This is a member of a imsm container. Load the container
6348 * and try to create a volume
6350 struct intel_super
*super
;
6352 if (load_super_imsm_all(st
, cfd
, (void **) &super
, NULL
, NULL
, 1) == 0) {
6354 strcpy(st
->container_devnm
, fd2devnm(cfd
));
6356 return validate_geometry_imsm_volume(st
, level
, layout
,
6358 size
, data_offset
, dev
,
6365 pr_err("failed container membership check\n");
6371 static void default_geometry_imsm(struct supertype
*st
, int *level
, int *layout
, int *chunk
)
6373 struct intel_super
*super
= st
->sb
;
6375 if (level
&& *level
== UnSet
)
6376 *level
= LEVEL_CONTAINER
;
6378 if (level
&& layout
&& *layout
== UnSet
)
6379 *layout
= imsm_level_to_layout(*level
);
6381 if (chunk
&& (*chunk
== UnSet
|| *chunk
== 0))
6382 *chunk
= imsm_default_chunk(super
->orom
);
6385 static void handle_missing(struct intel_super
*super
, struct imsm_dev
*dev
);
6387 static int kill_subarray_imsm(struct supertype
*st
)
6389 /* remove the subarray currently referenced by ->current_vol */
6391 struct intel_dev
**dp
;
6392 struct intel_super
*super
= st
->sb
;
6393 __u8 current_vol
= super
->current_vol
;
6394 struct imsm_super
*mpb
= super
->anchor
;
6396 if (super
->current_vol
< 0)
6398 super
->current_vol
= -1; /* invalidate subarray cursor */
6400 /* block deletions that would change the uuid of active subarrays
6402 * FIXME when immutable ids are available, but note that we'll
6403 * also need to fixup the invalidated/active subarray indexes in
6406 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
6409 if (i
< current_vol
)
6411 sprintf(subarray
, "%u", i
);
6412 if (is_subarray_active(subarray
, st
->devnm
)) {
6413 pr_err("deleting subarray-%d would change the UUID of active subarray-%d, aborting\n",
6420 if (st
->update_tail
) {
6421 struct imsm_update_kill_array
*u
= xmalloc(sizeof(*u
));
6423 u
->type
= update_kill_array
;
6424 u
->dev_idx
= current_vol
;
6425 append_metadata_update(st
, u
, sizeof(*u
));
6430 for (dp
= &super
->devlist
; *dp
;)
6431 if ((*dp
)->index
== current_vol
) {
6434 handle_missing(super
, (*dp
)->dev
);
6435 if ((*dp
)->index
> current_vol
)
6440 /* no more raid devices, all active components are now spares,
6441 * but of course failed are still failed
6443 if (--mpb
->num_raid_devs
== 0) {
6446 for (d
= super
->disks
; d
; d
= d
->next
)
6451 super
->updates_pending
++;
6456 static int update_subarray_imsm(struct supertype
*st
, char *subarray
,
6457 char *update
, struct mddev_ident
*ident
)
6459 /* update the subarray currently referenced by ->current_vol */
6460 struct intel_super
*super
= st
->sb
;
6461 struct imsm_super
*mpb
= super
->anchor
;
6463 if (strcmp(update
, "name") == 0) {
6464 char *name
= ident
->name
;
6468 if (is_subarray_active(subarray
, st
->devnm
)) {
6469 pr_err("Unable to update name of active subarray\n");
6473 if (!check_name(super
, name
, 0))
6476 vol
= strtoul(subarray
, &ep
, 10);
6477 if (*ep
!= '\0' || vol
>= super
->anchor
->num_raid_devs
)
6480 if (st
->update_tail
) {
6481 struct imsm_update_rename_array
*u
= xmalloc(sizeof(*u
));
6483 u
->type
= update_rename_array
;
6485 snprintf((char *) u
->name
, MAX_RAID_SERIAL_LEN
, "%s", name
);
6486 append_metadata_update(st
, u
, sizeof(*u
));
6488 struct imsm_dev
*dev
;
6491 dev
= get_imsm_dev(super
, vol
);
6492 snprintf((char *) dev
->volume
, MAX_RAID_SERIAL_LEN
, "%s", name
);
6493 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
6494 dev
= get_imsm_dev(super
, i
);
6495 handle_missing(super
, dev
);
6497 super
->updates_pending
++;
6504 #endif /* MDASSEMBLE */
6506 static int is_gen_migration(struct imsm_dev
*dev
)
6511 if (!dev
->vol
.migr_state
)
6514 if (migr_type(dev
) == MIGR_GEN_MIGR
)
6520 static int is_rebuilding(struct imsm_dev
*dev
)
6522 struct imsm_map
*migr_map
;
6524 if (!dev
->vol
.migr_state
)
6527 if (migr_type(dev
) != MIGR_REBUILD
)
6530 migr_map
= get_imsm_map(dev
, MAP_1
);
6532 if (migr_map
->map_state
== IMSM_T_STATE_DEGRADED
)
6539 static int is_initializing(struct imsm_dev
*dev
)
6541 struct imsm_map
*migr_map
;
6543 if (!dev
->vol
.migr_state
)
6546 if (migr_type(dev
) != MIGR_INIT
)
6549 migr_map
= get_imsm_map(dev
, MAP_1
);
6551 if (migr_map
->map_state
== IMSM_T_STATE_UNINITIALIZED
)
6558 static void update_recovery_start(struct intel_super
*super
,
6559 struct imsm_dev
*dev
,
6560 struct mdinfo
*array
)
6562 struct mdinfo
*rebuild
= NULL
;
6566 if (!is_rebuilding(dev
))
6569 /* Find the rebuild target, but punt on the dual rebuild case */
6570 for (d
= array
->devs
; d
; d
= d
->next
)
6571 if (d
->recovery_start
== 0) {
6578 /* (?) none of the disks are marked with
6579 * IMSM_ORD_REBUILD, so assume they are missing and the
6580 * disk_ord_tbl was not correctly updated
6582 dprintf("failed to locate out-of-sync disk\n");
6586 units
= __le32_to_cpu(dev
->vol
.curr_migr_unit
);
6587 rebuild
->recovery_start
= units
* blocks_per_migr_unit(super
, dev
);
6591 static int recover_backup_imsm(struct supertype
*st
, struct mdinfo
*info
);
6594 static struct mdinfo
*container_content_imsm(struct supertype
*st
, char *subarray
)
6596 /* Given a container loaded by load_super_imsm_all,
6597 * extract information about all the arrays into
6599 * If 'subarray' is given, just extract info about that array.
6601 * For each imsm_dev create an mdinfo, fill it in,
6602 * then look for matching devices in super->disks
6603 * and create appropriate device mdinfo.
6605 struct intel_super
*super
= st
->sb
;
6606 struct imsm_super
*mpb
= super
->anchor
;
6607 struct mdinfo
*rest
= NULL
;
6611 int spare_disks
= 0;
6613 /* do not assemble arrays when not all attributes are supported */
6614 if (imsm_check_attributes(mpb
->attributes
) == 0) {
6616 pr_err("Unsupported attributes in IMSM metadata.Arrays activation is blocked.\n");
6619 /* check for bad blocks */
6620 if (imsm_bbm_log_size(super
->anchor
)) {
6621 pr_err("BBM log found in IMSM metadata.Arrays activation is blocked.\n");
6625 /* count spare devices, not used in maps
6627 for (d
= super
->disks
; d
; d
= d
->next
)
6631 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
6632 struct imsm_dev
*dev
;
6633 struct imsm_map
*map
;
6634 struct imsm_map
*map2
;
6635 struct mdinfo
*this;
6643 (i
!= strtoul(subarray
, &ep
, 10) || *ep
!= '\0'))
6646 dev
= get_imsm_dev(super
, i
);
6647 map
= get_imsm_map(dev
, MAP_0
);
6648 map2
= get_imsm_map(dev
, MAP_1
);
6650 /* do not publish arrays that are in the middle of an
6651 * unsupported migration
6653 if (dev
->vol
.migr_state
&&
6654 (migr_type(dev
) == MIGR_STATE_CHANGE
)) {
6655 pr_err("cannot assemble volume '%.16s': unsupported migration in progress\n",
6659 /* do not publish arrays that are not support by controller's
6663 this = xmalloc(sizeof(*this));
6665 super
->current_vol
= i
;
6666 getinfo_super_imsm_volume(st
, this, NULL
);
6669 chunk
= __le16_to_cpu(map
->blocks_per_strip
) >> 1;
6670 /* mdadm does not support all metadata features- set the bit in all arrays state */
6671 if (!validate_geometry_imsm_orom(super
,
6672 get_imsm_raid_level(map
), /* RAID level */
6673 imsm_level_to_layout(get_imsm_raid_level(map
)),
6674 map
->num_members
, /* raid disks */
6675 &chunk
, join_u32(dev
->size_low
, dev
->size_high
),
6677 pr_err("IMSM RAID geometry validation failed. Array %s activation is blocked.\n",
6679 this->array
.state
|=
6680 (1<<MD_SB_BLOCK_CONTAINER_RESHAPE
) |
6681 (1<<MD_SB_BLOCK_VOLUME
);
6685 /* if array has bad blocks, set suitable bit in all arrays state */
6687 this->array
.state
|=
6688 (1<<MD_SB_BLOCK_CONTAINER_RESHAPE
) |
6689 (1<<MD_SB_BLOCK_VOLUME
);
6691 for (slot
= 0 ; slot
< map
->num_members
; slot
++) {
6692 unsigned long long recovery_start
;
6693 struct mdinfo
*info_d
;
6700 idx
= get_imsm_disk_idx(dev
, slot
, MAP_0
);
6701 ord
= get_imsm_ord_tbl_ent(dev
, slot
, MAP_X
);
6702 for (d
= super
->disks
; d
; d
= d
->next
)
6703 if (d
->index
== idx
)
6706 recovery_start
= MaxSector
;
6709 if (d
&& is_failed(&d
->disk
))
6711 if (ord
& IMSM_ORD_REBUILD
)
6715 * if we skip some disks the array will be assmebled degraded;
6716 * reset resync start to avoid a dirty-degraded
6717 * situation when performing the intial sync
6719 * FIXME handle dirty degraded
6721 if ((skip
|| recovery_start
== 0) && !dev
->vol
.dirty
)
6722 this->resync_start
= MaxSector
;
6726 info_d
= xcalloc(1, sizeof(*info_d
));
6727 info_d
->next
= this->devs
;
6728 this->devs
= info_d
;
6730 info_d
->disk
.number
= d
->index
;
6731 info_d
->disk
.major
= d
->major
;
6732 info_d
->disk
.minor
= d
->minor
;
6733 info_d
->disk
.raid_disk
= slot
;
6734 info_d
->recovery_start
= recovery_start
;
6736 if (slot
< map2
->num_members
)
6737 info_d
->disk
.state
= (1 << MD_DISK_ACTIVE
);
6739 this->array
.spare_disks
++;
6741 if (slot
< map
->num_members
)
6742 info_d
->disk
.state
= (1 << MD_DISK_ACTIVE
);
6744 this->array
.spare_disks
++;
6746 if (info_d
->recovery_start
== MaxSector
)
6747 this->array
.working_disks
++;
6749 info_d
->events
= __le32_to_cpu(mpb
->generation_num
);
6750 info_d
->data_offset
= pba_of_lba0(map
);
6751 info_d
->component_size
= blocks_per_member(map
);
6753 /* now that the disk list is up-to-date fixup recovery_start */
6754 update_recovery_start(super
, dev
, this);
6755 this->array
.spare_disks
+= spare_disks
;
6758 /* check for reshape */
6759 if (this->reshape_active
== 1)
6760 recover_backup_imsm(st
, this);
6768 static __u8
imsm_check_degraded(struct intel_super
*super
, struct imsm_dev
*dev
,
6769 int failed
, int look_in_map
)
6771 struct imsm_map
*map
;
6773 map
= get_imsm_map(dev
, look_in_map
);
6776 return map
->map_state
== IMSM_T_STATE_UNINITIALIZED
?
6777 IMSM_T_STATE_UNINITIALIZED
: IMSM_T_STATE_NORMAL
;
6779 switch (get_imsm_raid_level(map
)) {
6781 return IMSM_T_STATE_FAILED
;
6784 if (failed
< map
->num_members
)
6785 return IMSM_T_STATE_DEGRADED
;
6787 return IMSM_T_STATE_FAILED
;
6792 * check to see if any mirrors have failed, otherwise we
6793 * are degraded. Even numbered slots are mirrored on
6797 /* gcc -Os complains that this is unused */
6798 int insync
= insync
;
6800 for (i
= 0; i
< map
->num_members
; i
++) {
6801 __u32 ord
= get_imsm_ord_tbl_ent(dev
, i
, MAP_X
);
6802 int idx
= ord_to_idx(ord
);
6803 struct imsm_disk
*disk
;
6805 /* reset the potential in-sync count on even-numbered
6806 * slots. num_copies is always 2 for imsm raid10
6811 disk
= get_imsm_disk(super
, idx
);
6812 if (!disk
|| is_failed(disk
) || ord
& IMSM_ORD_REBUILD
)
6815 /* no in-sync disks left in this mirror the
6819 return IMSM_T_STATE_FAILED
;
6822 return IMSM_T_STATE_DEGRADED
;
6826 return IMSM_T_STATE_DEGRADED
;
6828 return IMSM_T_STATE_FAILED
;
6834 return map
->map_state
;
6837 static int imsm_count_failed(struct intel_super
*super
, struct imsm_dev
*dev
,
6842 struct imsm_disk
*disk
;
6843 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
6844 struct imsm_map
*prev
= get_imsm_map(dev
, MAP_1
);
6845 struct imsm_map
*map_for_loop
;
6850 /* at the beginning of migration we set IMSM_ORD_REBUILD on
6851 * disks that are being rebuilt. New failures are recorded to
6852 * map[0]. So we look through all the disks we started with and
6853 * see if any failures are still present, or if any new ones
6857 if (prev
&& (map
->num_members
< prev
->num_members
))
6858 map_for_loop
= prev
;
6860 for (i
= 0; i
< map_for_loop
->num_members
; i
++) {
6862 /* when MAP_X is passed both maps failures are counted
6865 ((look_in_map
== MAP_1
) || (look_in_map
== MAP_X
)) &&
6866 (i
< prev
->num_members
)) {
6867 ord
= __le32_to_cpu(prev
->disk_ord_tbl
[i
]);
6868 idx_1
= ord_to_idx(ord
);
6870 disk
= get_imsm_disk(super
, idx_1
);
6871 if (!disk
|| is_failed(disk
) || ord
& IMSM_ORD_REBUILD
)
6874 if (((look_in_map
== MAP_0
) || (look_in_map
== MAP_X
)) &&
6875 (i
< map
->num_members
)) {
6876 ord
= __le32_to_cpu(map
->disk_ord_tbl
[i
]);
6877 idx
= ord_to_idx(ord
);
6880 disk
= get_imsm_disk(super
, idx
);
6881 if (!disk
|| is_failed(disk
) ||
6882 ord
& IMSM_ORD_REBUILD
)
6892 static int imsm_open_new(struct supertype
*c
, struct active_array
*a
,
6895 struct intel_super
*super
= c
->sb
;
6896 struct imsm_super
*mpb
= super
->anchor
;
6898 if (atoi(inst
) >= mpb
->num_raid_devs
) {
6899 pr_err("subarry index %d, out of range\n", atoi(inst
));
6903 dprintf("imsm: open_new %s\n", inst
);
6904 a
->info
.container_member
= atoi(inst
);
6908 static int is_resyncing(struct imsm_dev
*dev
)
6910 struct imsm_map
*migr_map
;
6912 if (!dev
->vol
.migr_state
)
6915 if (migr_type(dev
) == MIGR_INIT
||
6916 migr_type(dev
) == MIGR_REPAIR
)
6919 if (migr_type(dev
) == MIGR_GEN_MIGR
)
6922 migr_map
= get_imsm_map(dev
, MAP_1
);
6924 if ((migr_map
->map_state
== IMSM_T_STATE_NORMAL
) &&
6925 (dev
->vol
.migr_type
!= MIGR_GEN_MIGR
))
6931 /* return true if we recorded new information */
6932 static int mark_failure(struct imsm_dev
*dev
, struct imsm_disk
*disk
, int idx
)
6936 struct imsm_map
*map
;
6937 char buf
[MAX_RAID_SERIAL_LEN
+3];
6938 unsigned int len
, shift
= 0;
6940 /* new failures are always set in map[0] */
6941 map
= get_imsm_map(dev
, MAP_0
);
6943 slot
= get_imsm_disk_slot(map
, idx
);
6947 ord
= __le32_to_cpu(map
->disk_ord_tbl
[slot
]);
6948 if (is_failed(disk
) && (ord
& IMSM_ORD_REBUILD
))
6951 memcpy(buf
, disk
->serial
, MAX_RAID_SERIAL_LEN
);
6952 buf
[MAX_RAID_SERIAL_LEN
] = '\000';
6954 if ((len
= strlen(buf
)) >= MAX_RAID_SERIAL_LEN
)
6955 shift
= len
- MAX_RAID_SERIAL_LEN
+ 1;
6956 strncpy((char *)disk
->serial
, &buf
[shift
], MAX_RAID_SERIAL_LEN
);
6958 disk
->status
|= FAILED_DISK
;
6959 set_imsm_ord_tbl_ent(map
, slot
, idx
| IMSM_ORD_REBUILD
);
6960 /* mark failures in second map if second map exists and this disk
6962 * This is valid for migration, initialization and rebuild
6964 if (dev
->vol
.migr_state
) {
6965 struct imsm_map
*map2
= get_imsm_map(dev
, MAP_1
);
6966 int slot2
= get_imsm_disk_slot(map2
, idx
);
6968 if ((slot2
< map2
->num_members
) &&
6970 set_imsm_ord_tbl_ent(map2
, slot2
,
6971 idx
| IMSM_ORD_REBUILD
);
6973 if (map
->failed_disk_num
== 0xff)
6974 map
->failed_disk_num
= slot
;
6978 static void mark_missing(struct imsm_dev
*dev
, struct imsm_disk
*disk
, int idx
)
6980 mark_failure(dev
, disk
, idx
);
6982 if (disk
->scsi_id
== __cpu_to_le32(~(__u32
)0))
6985 disk
->scsi_id
= __cpu_to_le32(~(__u32
)0);
6986 memmove(&disk
->serial
[0], &disk
->serial
[1], MAX_RAID_SERIAL_LEN
- 1);
6989 static void handle_missing(struct intel_super
*super
, struct imsm_dev
*dev
)
6993 if (!super
->missing
)
6996 /* When orom adds replacement for missing disk it does
6997 * not remove entry of missing disk, but just updates map with
6998 * new added disk. So it is not enough just to test if there is
6999 * any missing disk, we have to look if there are any failed disks
7000 * in map to stop migration */
7002 dprintf("imsm: mark missing\n");
7003 /* end process for initialization and rebuild only
7005 if (is_gen_migration(dev
) == 0) {
7009 failed
= imsm_count_failed(super
, dev
, MAP_0
);
7010 map_state
= imsm_check_degraded(super
, dev
, failed
, MAP_0
);
7013 end_migration(dev
, super
, map_state
);
7015 for (dl
= super
->missing
; dl
; dl
= dl
->next
)
7016 mark_missing(dev
, &dl
->disk
, dl
->index
);
7017 super
->updates_pending
++;
7020 static unsigned long long imsm_set_array_size(struct imsm_dev
*dev
,
7023 int used_disks
= imsm_num_data_members(dev
, MAP_0
);
7024 unsigned long long array_blocks
;
7025 struct imsm_map
*map
;
7027 if (used_disks
== 0) {
7028 /* when problems occures
7029 * return current array_blocks value
7031 array_blocks
= __le32_to_cpu(dev
->size_high
);
7032 array_blocks
= array_blocks
<< 32;
7033 array_blocks
+= __le32_to_cpu(dev
->size_low
);
7035 return array_blocks
;
7038 /* set array size in metadata
7040 if (new_size
<= 0) {
7041 /* OLCE size change is caused by added disks
7043 map
= get_imsm_map(dev
, MAP_0
);
7044 array_blocks
= blocks_per_member(map
) * used_disks
;
7046 /* Online Volume Size Change
7047 * Using available free space
7049 array_blocks
= new_size
;
7052 /* round array size down to closest MB
7054 array_blocks
= (array_blocks
>> SECT_PER_MB_SHIFT
) << SECT_PER_MB_SHIFT
;
7055 dev
->size_low
= __cpu_to_le32((__u32
)array_blocks
);
7056 dev
->size_high
= __cpu_to_le32((__u32
)(array_blocks
>> 32));
7058 return array_blocks
;
7061 static void imsm_set_disk(struct active_array
*a
, int n
, int state
);
7063 static void imsm_progress_container_reshape(struct intel_super
*super
)
7065 /* if no device has a migr_state, but some device has a
7066 * different number of members than the previous device, start
7067 * changing the number of devices in this device to match
7070 struct imsm_super
*mpb
= super
->anchor
;
7071 int prev_disks
= -1;
7075 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
7076 struct imsm_dev
*dev
= get_imsm_dev(super
, i
);
7077 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
7078 struct imsm_map
*map2
;
7079 int prev_num_members
;
7081 if (dev
->vol
.migr_state
)
7084 if (prev_disks
== -1)
7085 prev_disks
= map
->num_members
;
7086 if (prev_disks
== map
->num_members
)
7089 /* OK, this array needs to enter reshape mode.
7090 * i.e it needs a migr_state
7093 copy_map_size
= sizeof_imsm_map(map
);
7094 prev_num_members
= map
->num_members
;
7095 map
->num_members
= prev_disks
;
7096 dev
->vol
.migr_state
= 1;
7097 dev
->vol
.curr_migr_unit
= 0;
7098 set_migr_type(dev
, MIGR_GEN_MIGR
);
7099 for (i
= prev_num_members
;
7100 i
< map
->num_members
; i
++)
7101 set_imsm_ord_tbl_ent(map
, i
, i
);
7102 map2
= get_imsm_map(dev
, MAP_1
);
7103 /* Copy the current map */
7104 memcpy(map2
, map
, copy_map_size
);
7105 map2
->num_members
= prev_num_members
;
7107 imsm_set_array_size(dev
, -1);
7108 super
->clean_migration_record_by_mdmon
= 1;
7109 super
->updates_pending
++;
7113 /* Handle dirty -> clean transititions, resync and reshape. Degraded and rebuild
7114 * states are handled in imsm_set_disk() with one exception, when a
7115 * resync is stopped due to a new failure this routine will set the
7116 * 'degraded' state for the array.
7118 static int imsm_set_array_state(struct active_array
*a
, int consistent
)
7120 int inst
= a
->info
.container_member
;
7121 struct intel_super
*super
= a
->container
->sb
;
7122 struct imsm_dev
*dev
= get_imsm_dev(super
, inst
);
7123 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
7124 int failed
= imsm_count_failed(super
, dev
, MAP_0
);
7125 __u8 map_state
= imsm_check_degraded(super
, dev
, failed
, MAP_0
);
7126 __u32 blocks_per_unit
;
7128 if (dev
->vol
.migr_state
&&
7129 dev
->vol
.migr_type
== MIGR_GEN_MIGR
) {
7130 /* array state change is blocked due to reshape action
7132 * - abort the reshape (if last_checkpoint is 0 and action!= reshape)
7133 * - finish the reshape (if last_checkpoint is big and action != reshape)
7134 * - update curr_migr_unit
7136 if (a
->curr_action
== reshape
) {
7137 /* still reshaping, maybe update curr_migr_unit */
7138 goto mark_checkpoint
;
7140 if (a
->last_checkpoint
== 0 && a
->prev_action
== reshape
) {
7141 /* for some reason we aborted the reshape.
7143 * disable automatic metadata rollback
7144 * user action is required to recover process
7147 struct imsm_map
*map2
=
7148 get_imsm_map(dev
, MAP_1
);
7149 dev
->vol
.migr_state
= 0;
7150 set_migr_type(dev
, 0);
7151 dev
->vol
.curr_migr_unit
= 0;
7153 sizeof_imsm_map(map2
));
7154 super
->updates_pending
++;
7157 if (a
->last_checkpoint
>= a
->info
.component_size
) {
7158 unsigned long long array_blocks
;
7162 used_disks
= imsm_num_data_members(dev
, MAP_0
);
7163 if (used_disks
> 0) {
7165 blocks_per_member(map
) *
7167 /* round array size down to closest MB
7169 array_blocks
= (array_blocks
7170 >> SECT_PER_MB_SHIFT
)
7171 << SECT_PER_MB_SHIFT
;
7172 a
->info
.custom_array_size
= array_blocks
;
7173 /* encourage manager to update array
7177 a
->check_reshape
= 1;
7179 /* finalize online capacity expansion/reshape */
7180 for (mdi
= a
->info
.devs
; mdi
; mdi
= mdi
->next
)
7182 mdi
->disk
.raid_disk
,
7185 imsm_progress_container_reshape(super
);
7190 /* before we activate this array handle any missing disks */
7191 if (consistent
== 2)
7192 handle_missing(super
, dev
);
7194 if (consistent
== 2 &&
7195 (!is_resync_complete(&a
->info
) ||
7196 map_state
!= IMSM_T_STATE_NORMAL
||
7197 dev
->vol
.migr_state
))
7200 if (is_resync_complete(&a
->info
)) {
7201 /* complete intialization / resync,
7202 * recovery and interrupted recovery is completed in
7205 if (is_resyncing(dev
)) {
7206 dprintf("imsm: mark resync done\n");
7207 end_migration(dev
, super
, map_state
);
7208 super
->updates_pending
++;
7209 a
->last_checkpoint
= 0;
7211 } else if ((!is_resyncing(dev
) && !failed
) &&
7212 (imsm_reshape_blocks_arrays_changes(super
) == 0)) {
7213 /* mark the start of the init process if nothing is failed */
7214 dprintf("imsm: mark resync start\n");
7215 if (map
->map_state
== IMSM_T_STATE_UNINITIALIZED
)
7216 migrate(dev
, super
, IMSM_T_STATE_NORMAL
, MIGR_INIT
);
7218 migrate(dev
, super
, IMSM_T_STATE_NORMAL
, MIGR_REPAIR
);
7219 super
->updates_pending
++;
7223 /* skip checkpointing for general migration,
7224 * it is controlled in mdadm
7226 if (is_gen_migration(dev
))
7227 goto skip_mark_checkpoint
;
7229 /* check if we can update curr_migr_unit from resync_start, recovery_start */
7230 blocks_per_unit
= blocks_per_migr_unit(super
, dev
);
7231 if (blocks_per_unit
) {
7235 units
= a
->last_checkpoint
/ blocks_per_unit
;
7238 /* check that we did not overflow 32-bits, and that
7239 * curr_migr_unit needs updating
7241 if (units32
== units
&&
7243 __le32_to_cpu(dev
->vol
.curr_migr_unit
) != units32
) {
7244 dprintf("imsm: mark checkpoint (%u)\n", units32
);
7245 dev
->vol
.curr_migr_unit
= __cpu_to_le32(units32
);
7246 super
->updates_pending
++;
7250 skip_mark_checkpoint
:
7251 /* mark dirty / clean */
7252 if (dev
->vol
.dirty
!= !consistent
) {
7253 dprintf("imsm: mark '%s'\n", consistent
? "clean" : "dirty");
7258 super
->updates_pending
++;
7264 static void imsm_set_disk(struct active_array
*a
, int n
, int state
)
7266 int inst
= a
->info
.container_member
;
7267 struct intel_super
*super
= a
->container
->sb
;
7268 struct imsm_dev
*dev
= get_imsm_dev(super
, inst
);
7269 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
7270 struct imsm_disk
*disk
;
7272 int recovery_not_finished
= 0;
7277 if (n
> map
->num_members
)
7278 pr_err("imsm: set_disk %d out of range 0..%d\n",
7279 n
, map
->num_members
- 1);
7284 dprintf("imsm: set_disk %d:%x\n", n
, state
);
7286 ord
= get_imsm_ord_tbl_ent(dev
, n
, MAP_0
);
7287 disk
= get_imsm_disk(super
, ord_to_idx(ord
));
7289 /* check for new failures */
7290 if (state
& DS_FAULTY
) {
7291 if (mark_failure(dev
, disk
, ord_to_idx(ord
)))
7292 super
->updates_pending
++;
7295 /* check if in_sync */
7296 if (state
& DS_INSYNC
&& ord
& IMSM_ORD_REBUILD
&& is_rebuilding(dev
)) {
7297 struct imsm_map
*migr_map
= get_imsm_map(dev
, MAP_1
);
7299 set_imsm_ord_tbl_ent(migr_map
, n
, ord_to_idx(ord
));
7300 super
->updates_pending
++;
7303 failed
= imsm_count_failed(super
, dev
, MAP_0
);
7304 map_state
= imsm_check_degraded(super
, dev
, failed
, MAP_0
);
7306 /* check if recovery complete, newly degraded, or failed */
7307 dprintf("imsm: Detected transition to state ");
7308 switch (map_state
) {
7309 case IMSM_T_STATE_NORMAL
: /* transition to normal state */
7310 dprintf("normal: ");
7311 if (is_rebuilding(dev
)) {
7312 dprintf_cont("while rebuilding");
7313 /* check if recovery is really finished */
7314 for (mdi
= a
->info
.devs
; mdi
; mdi
= mdi
->next
)
7315 if (mdi
->recovery_start
!= MaxSector
) {
7316 recovery_not_finished
= 1;
7319 if (recovery_not_finished
) {
7321 dprintf("Rebuild has not finished yet, state not changed");
7322 if (a
->last_checkpoint
< mdi
->recovery_start
) {
7323 a
->last_checkpoint
= mdi
->recovery_start
;
7324 super
->updates_pending
++;
7328 end_migration(dev
, super
, map_state
);
7329 map
= get_imsm_map(dev
, MAP_0
);
7330 map
->failed_disk_num
= ~0;
7331 super
->updates_pending
++;
7332 a
->last_checkpoint
= 0;
7335 if (is_gen_migration(dev
)) {
7336 dprintf_cont("while general migration");
7337 if (a
->last_checkpoint
>= a
->info
.component_size
)
7338 end_migration(dev
, super
, map_state
);
7340 map
->map_state
= map_state
;
7341 map
= get_imsm_map(dev
, MAP_0
);
7342 map
->failed_disk_num
= ~0;
7343 super
->updates_pending
++;
7347 case IMSM_T_STATE_DEGRADED
: /* transition to degraded state */
7348 dprintf_cont("degraded: ");
7349 if ((map
->map_state
!= map_state
) &&
7350 !dev
->vol
.migr_state
) {
7351 dprintf_cont("mark degraded");
7352 map
->map_state
= map_state
;
7353 super
->updates_pending
++;
7354 a
->last_checkpoint
= 0;
7357 if (is_rebuilding(dev
)) {
7358 dprintf_cont("while rebuilding.");
7359 if (map
->map_state
!= map_state
) {
7360 dprintf_cont(" Map state change");
7361 end_migration(dev
, super
, map_state
);
7362 super
->updates_pending
++;
7366 if (is_gen_migration(dev
)) {
7367 dprintf_cont("while general migration");
7368 if (a
->last_checkpoint
>= a
->info
.component_size
)
7369 end_migration(dev
, super
, map_state
);
7371 map
->map_state
= map_state
;
7372 manage_second_map(super
, dev
);
7374 super
->updates_pending
++;
7377 if (is_initializing(dev
)) {
7378 dprintf_cont("while initialization.");
7379 map
->map_state
= map_state
;
7380 super
->updates_pending
++;
7384 case IMSM_T_STATE_FAILED
: /* transition to failed state */
7385 dprintf_cont("failed: ");
7386 if (is_gen_migration(dev
)) {
7387 dprintf_cont("while general migration");
7388 map
->map_state
= map_state
;
7389 super
->updates_pending
++;
7392 if (map
->map_state
!= map_state
) {
7393 dprintf_cont("mark failed");
7394 end_migration(dev
, super
, map_state
);
7395 super
->updates_pending
++;
7396 a
->last_checkpoint
= 0;
7401 dprintf_cont("state %i\n", map_state
);
7406 static int store_imsm_mpb(int fd
, struct imsm_super
*mpb
)
7409 __u32 mpb_size
= __le32_to_cpu(mpb
->mpb_size
);
7410 unsigned long long dsize
;
7411 unsigned long long sectors
;
7413 get_dev_size(fd
, NULL
, &dsize
);
7415 if (mpb_size
> 512) {
7416 /* -1 to account for anchor */
7417 sectors
= mpb_sectors(mpb
) - 1;
7419 /* write the extended mpb to the sectors preceeding the anchor */
7420 if (lseek64(fd
, dsize
- (512 * (2 + sectors
)), SEEK_SET
) < 0)
7423 if ((unsigned long long)write(fd
, buf
+ 512, 512 * sectors
)
7428 /* first block is stored on second to last sector of the disk */
7429 if (lseek64(fd
, dsize
- (512 * 2), SEEK_SET
) < 0)
7432 if (write(fd
, buf
, 512) != 512)
7438 static void imsm_sync_metadata(struct supertype
*container
)
7440 struct intel_super
*super
= container
->sb
;
7442 dprintf("sync metadata: %d\n", super
->updates_pending
);
7443 if (!super
->updates_pending
)
7446 write_super_imsm(container
, 0);
7448 super
->updates_pending
= 0;
7451 static struct dl
*imsm_readd(struct intel_super
*super
, int idx
, struct active_array
*a
)
7453 struct imsm_dev
*dev
= get_imsm_dev(super
, a
->info
.container_member
);
7454 int i
= get_imsm_disk_idx(dev
, idx
, MAP_X
);
7457 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
7461 if (dl
&& is_failed(&dl
->disk
))
7465 dprintf("found %x:%x\n", dl
->major
, dl
->minor
);
7470 static struct dl
*imsm_add_spare(struct intel_super
*super
, int slot
,
7471 struct active_array
*a
, int activate_new
,
7472 struct mdinfo
*additional_test_list
)
7474 struct imsm_dev
*dev
= get_imsm_dev(super
, a
->info
.container_member
);
7475 int idx
= get_imsm_disk_idx(dev
, slot
, MAP_X
);
7476 struct imsm_super
*mpb
= super
->anchor
;
7477 struct imsm_map
*map
;
7478 unsigned long long pos
;
7483 __u32 array_start
= 0;
7484 __u32 array_end
= 0;
7486 struct mdinfo
*test_list
;
7488 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
7489 /* If in this array, skip */
7490 for (d
= a
->info
.devs
; d
; d
= d
->next
)
7491 if (d
->state_fd
>= 0 &&
7492 d
->disk
.major
== dl
->major
&&
7493 d
->disk
.minor
== dl
->minor
) {
7494 dprintf("%x:%x already in array\n",
7495 dl
->major
, dl
->minor
);
7500 test_list
= additional_test_list
;
7502 if (test_list
->disk
.major
== dl
->major
&&
7503 test_list
->disk
.minor
== dl
->minor
) {
7504 dprintf("%x:%x already in additional test list\n",
7505 dl
->major
, dl
->minor
);
7508 test_list
= test_list
->next
;
7513 /* skip in use or failed drives */
7514 if (is_failed(&dl
->disk
) || idx
== dl
->index
||
7516 dprintf("%x:%x status (failed: %d index: %d)\n",
7517 dl
->major
, dl
->minor
, is_failed(&dl
->disk
), idx
);
7521 /* skip pure spares when we are looking for partially
7522 * assimilated drives
7524 if (dl
->index
== -1 && !activate_new
)
7527 /* Does this unused device have the requisite free space?
7528 * It needs to be able to cover all member volumes
7530 ex
= get_extents(super
, dl
);
7532 dprintf("cannot get extents\n");
7535 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
7536 dev
= get_imsm_dev(super
, i
);
7537 map
= get_imsm_map(dev
, MAP_0
);
7539 /* check if this disk is already a member of
7542 if (get_imsm_disk_slot(map
, dl
->index
) >= 0)
7548 array_start
= pba_of_lba0(map
);
7549 array_end
= array_start
+
7550 blocks_per_member(map
) - 1;
7553 /* check that we can start at pba_of_lba0 with
7554 * blocks_per_member of space
7556 if (array_start
>= pos
&& array_end
< ex
[j
].start
) {
7560 pos
= ex
[j
].start
+ ex
[j
].size
;
7562 } while (ex
[j
-1].size
);
7569 if (i
< mpb
->num_raid_devs
) {
7570 dprintf("%x:%x does not have %u to %u available\n",
7571 dl
->major
, dl
->minor
, array_start
, array_end
);
7581 static int imsm_rebuild_allowed(struct supertype
*cont
, int dev_idx
, int failed
)
7583 struct imsm_dev
*dev2
;
7584 struct imsm_map
*map
;
7590 dev2
= get_imsm_dev(cont
->sb
, dev_idx
);
7592 state
= imsm_check_degraded(cont
->sb
, dev2
, failed
, MAP_0
);
7593 if (state
== IMSM_T_STATE_FAILED
) {
7594 map
= get_imsm_map(dev2
, MAP_0
);
7597 for (slot
= 0; slot
< map
->num_members
; slot
++) {
7599 * Check if failed disks are deleted from intel
7600 * disk list or are marked to be deleted
7602 idx
= get_imsm_disk_idx(dev2
, slot
, MAP_X
);
7603 idisk
= get_imsm_dl_disk(cont
->sb
, idx
);
7605 * Do not rebuild the array if failed disks
7606 * from failed sub-array are not removed from
7610 is_failed(&idisk
->disk
) &&
7611 (idisk
->action
!= DISK_REMOVE
))
7619 static struct mdinfo
*imsm_activate_spare(struct active_array
*a
,
7620 struct metadata_update
**updates
)
7623 * Find a device with unused free space and use it to replace a
7624 * failed/vacant region in an array. We replace failed regions one a
7625 * array at a time. The result is that a new spare disk will be added
7626 * to the first failed array and after the monitor has finished
7627 * propagating failures the remainder will be consumed.
7629 * FIXME add a capability for mdmon to request spares from another
7633 struct intel_super
*super
= a
->container
->sb
;
7634 int inst
= a
->info
.container_member
;
7635 struct imsm_dev
*dev
= get_imsm_dev(super
, inst
);
7636 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
7637 int failed
= a
->info
.array
.raid_disks
;
7638 struct mdinfo
*rv
= NULL
;
7641 struct metadata_update
*mu
;
7643 struct imsm_update_activate_spare
*u
;
7648 for (d
= a
->info
.devs
; d
; d
= d
->next
) {
7649 if ((d
->curr_state
& DS_FAULTY
) &&
7651 /* wait for Removal to happen */
7653 if (d
->state_fd
>= 0)
7657 dprintf("imsm: activate spare: inst=%d failed=%d (%d) level=%d\n",
7658 inst
, failed
, a
->info
.array
.raid_disks
, a
->info
.array
.level
);
7660 if (imsm_reshape_blocks_arrays_changes(super
))
7663 /* Cannot activate another spare if rebuild is in progress already
7665 if (is_rebuilding(dev
)) {
7666 dprintf("imsm: No spare activation allowed. Rebuild in progress already.\n");
7670 if (a
->info
.array
.level
== 4)
7671 /* No repair for takeovered array
7672 * imsm doesn't support raid4
7676 if (imsm_check_degraded(super
, dev
, failed
, MAP_0
) !=
7677 IMSM_T_STATE_DEGRADED
)
7681 * If there are any failed disks check state of the other volume.
7682 * Block rebuild if the another one is failed until failed disks
7683 * are removed from container.
7686 dprintf("found failed disks in %.*s, check if there anotherfailed sub-array.\n",
7687 MAX_RAID_SERIAL_LEN
, dev
->volume
);
7688 /* check if states of the other volumes allow for rebuild */
7689 for (i
= 0; i
< super
->anchor
->num_raid_devs
; i
++) {
7691 allowed
= imsm_rebuild_allowed(a
->container
,
7699 /* For each slot, if it is not working, find a spare */
7700 for (i
= 0; i
< a
->info
.array
.raid_disks
; i
++) {
7701 for (d
= a
->info
.devs
; d
; d
= d
->next
)
7702 if (d
->disk
.raid_disk
== i
)
7704 dprintf("found %d: %p %x\n", i
, d
, d
?d
->curr_state
:0);
7705 if (d
&& (d
->state_fd
>= 0))
7709 * OK, this device needs recovery. Try to re-add the
7710 * previous occupant of this slot, if this fails see if
7711 * we can continue the assimilation of a spare that was
7712 * partially assimilated, finally try to activate a new
7715 dl
= imsm_readd(super
, i
, a
);
7717 dl
= imsm_add_spare(super
, i
, a
, 0, rv
);
7719 dl
= imsm_add_spare(super
, i
, a
, 1, rv
);
7723 /* found a usable disk with enough space */
7724 di
= xcalloc(1, sizeof(*di
));
7726 /* dl->index will be -1 in the case we are activating a
7727 * pristine spare. imsm_process_update() will create a
7728 * new index in this case. Once a disk is found to be
7729 * failed in all member arrays it is kicked from the
7732 di
->disk
.number
= dl
->index
;
7734 /* (ab)use di->devs to store a pointer to the device
7737 di
->devs
= (struct mdinfo
*) dl
;
7739 di
->disk
.raid_disk
= i
;
7740 di
->disk
.major
= dl
->major
;
7741 di
->disk
.minor
= dl
->minor
;
7743 di
->recovery_start
= 0;
7744 di
->data_offset
= pba_of_lba0(map
);
7745 di
->component_size
= a
->info
.component_size
;
7746 di
->container_member
= inst
;
7747 super
->random
= random32();
7751 dprintf("%x:%x to be %d at %llu\n", dl
->major
, dl
->minor
,
7752 i
, di
->data_offset
);
7756 /* No spares found */
7758 /* Now 'rv' has a list of devices to return.
7759 * Create a metadata_update record to update the
7760 * disk_ord_tbl for the array
7762 mu
= xmalloc(sizeof(*mu
));
7763 mu
->buf
= xcalloc(num_spares
,
7764 sizeof(struct imsm_update_activate_spare
));
7766 mu
->space_list
= NULL
;
7767 mu
->len
= sizeof(struct imsm_update_activate_spare
) * num_spares
;
7768 mu
->next
= *updates
;
7769 u
= (struct imsm_update_activate_spare
*) mu
->buf
;
7771 for (di
= rv
; di
; di
= di
->next
) {
7772 u
->type
= update_activate_spare
;
7773 u
->dl
= (struct dl
*) di
->devs
;
7775 u
->slot
= di
->disk
.raid_disk
;
7786 static int disks_overlap(struct intel_super
*super
, int idx
, struct imsm_update_create_array
*u
)
7788 struct imsm_dev
*dev
= get_imsm_dev(super
, idx
);
7789 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
7790 struct imsm_map
*new_map
= get_imsm_map(&u
->dev
, MAP_0
);
7791 struct disk_info
*inf
= get_disk_info(u
);
7792 struct imsm_disk
*disk
;
7796 for (i
= 0; i
< map
->num_members
; i
++) {
7797 disk
= get_imsm_disk(super
, get_imsm_disk_idx(dev
, i
, MAP_X
));
7798 for (j
= 0; j
< new_map
->num_members
; j
++)
7799 if (serialcmp(disk
->serial
, inf
[j
].serial
) == 0)
7806 static struct dl
*get_disk_super(struct intel_super
*super
, int major
, int minor
)
7808 struct dl
*dl
= NULL
;
7809 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
7810 if ((dl
->major
== major
) && (dl
->minor
== minor
))
7815 static int remove_disk_super(struct intel_super
*super
, int major
, int minor
)
7817 struct dl
*prev
= NULL
;
7821 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
7822 if ((dl
->major
== major
) && (dl
->minor
== minor
)) {
7825 prev
->next
= dl
->next
;
7827 super
->disks
= dl
->next
;
7829 __free_imsm_disk(dl
);
7830 dprintf("removed %x:%x\n", major
, minor
);
7838 static void imsm_delete(struct intel_super
*super
, struct dl
**dlp
, unsigned index
);
7840 static int add_remove_disk_update(struct intel_super
*super
)
7842 int check_degraded
= 0;
7843 struct dl
*disk
= NULL
;
7844 /* add/remove some spares to/from the metadata/contrainer */
7845 while (super
->disk_mgmt_list
) {
7846 struct dl
*disk_cfg
;
7848 disk_cfg
= super
->disk_mgmt_list
;
7849 super
->disk_mgmt_list
= disk_cfg
->next
;
7850 disk_cfg
->next
= NULL
;
7852 if (disk_cfg
->action
== DISK_ADD
) {
7853 disk_cfg
->next
= super
->disks
;
7854 super
->disks
= disk_cfg
;
7856 dprintf("added %x:%x\n",
7857 disk_cfg
->major
, disk_cfg
->minor
);
7858 } else if (disk_cfg
->action
== DISK_REMOVE
) {
7859 dprintf("Disk remove action processed: %x.%x\n",
7860 disk_cfg
->major
, disk_cfg
->minor
);
7861 disk
= get_disk_super(super
,
7865 /* store action status */
7866 disk
->action
= DISK_REMOVE
;
7867 /* remove spare disks only */
7868 if (disk
->index
== -1) {
7869 remove_disk_super(super
,
7874 /* release allocate disk structure */
7875 __free_imsm_disk(disk_cfg
);
7878 return check_degraded
;
7881 static int apply_reshape_migration_update(struct imsm_update_reshape_migration
*u
,
7882 struct intel_super
*super
,
7885 struct intel_dev
*id
;
7886 void **tofree
= NULL
;
7889 dprintf("(enter)\n");
7890 if ((u
->subdev
< 0) ||
7892 dprintf("imsm: Error: Wrong subdev: %i\n", u
->subdev
);
7895 if ((space_list
== NULL
) || (*space_list
== NULL
)) {
7896 dprintf("imsm: Error: Memory is not allocated\n");
7900 for (id
= super
->devlist
; id
; id
= id
->next
) {
7901 if (id
->index
== (unsigned)u
->subdev
) {
7902 struct imsm_dev
*dev
= get_imsm_dev(super
, u
->subdev
);
7903 struct imsm_map
*map
;
7904 struct imsm_dev
*new_dev
=
7905 (struct imsm_dev
*)*space_list
;
7906 struct imsm_map
*migr_map
= get_imsm_map(dev
, MAP_1
);
7908 struct dl
*new_disk
;
7910 if (new_dev
== NULL
)
7912 *space_list
= **space_list
;
7913 memcpy(new_dev
, dev
, sizeof_imsm_dev(dev
, 0));
7914 map
= get_imsm_map(new_dev
, MAP_0
);
7916 dprintf("imsm: Error: migration in progress");
7920 to_state
= map
->map_state
;
7921 if ((u
->new_level
== 5) && (map
->raid_level
== 0)) {
7923 /* this should not happen */
7924 if (u
->new_disks
[0] < 0) {
7925 map
->failed_disk_num
=
7926 map
->num_members
- 1;
7927 to_state
= IMSM_T_STATE_DEGRADED
;
7929 to_state
= IMSM_T_STATE_NORMAL
;
7931 migrate(new_dev
, super
, to_state
, MIGR_GEN_MIGR
);
7932 if (u
->new_level
> -1)
7933 map
->raid_level
= u
->new_level
;
7934 migr_map
= get_imsm_map(new_dev
, MAP_1
);
7935 if ((u
->new_level
== 5) &&
7936 (migr_map
->raid_level
== 0)) {
7937 int ord
= map
->num_members
- 1;
7938 migr_map
->num_members
--;
7939 if (u
->new_disks
[0] < 0)
7940 ord
|= IMSM_ORD_REBUILD
;
7941 set_imsm_ord_tbl_ent(map
,
7942 map
->num_members
- 1,
7946 tofree
= (void **)dev
;
7948 /* update chunk size
7950 if (u
->new_chunksize
> 0)
7951 map
->blocks_per_strip
=
7952 __cpu_to_le16(u
->new_chunksize
* 2);
7956 if ((u
->new_level
!= 5) ||
7957 (migr_map
->raid_level
!= 0) ||
7958 (migr_map
->raid_level
== map
->raid_level
))
7961 if (u
->new_disks
[0] >= 0) {
7964 new_disk
= get_disk_super(super
,
7965 major(u
->new_disks
[0]),
7966 minor(u
->new_disks
[0]));
7967 dprintf("imsm: new disk for reshape is: %i:%i (%p, index = %i)\n",
7968 major(u
->new_disks
[0]),
7969 minor(u
->new_disks
[0]),
7970 new_disk
, new_disk
->index
);
7971 if (new_disk
== NULL
)
7972 goto error_disk_add
;
7974 new_disk
->index
= map
->num_members
- 1;
7975 /* slot to fill in autolayout
7977 new_disk
->raiddisk
= new_disk
->index
;
7978 new_disk
->disk
.status
|= CONFIGURED_DISK
;
7979 new_disk
->disk
.status
&= ~SPARE_DISK
;
7981 goto error_disk_add
;
7984 *tofree
= *space_list
;
7985 /* calculate new size
7987 imsm_set_array_size(new_dev
, -1);
7994 *space_list
= tofree
;
7998 dprintf("Error: imsm: Cannot find disk.\n");
8002 static int apply_size_change_update(struct imsm_update_size_change
*u
,
8003 struct intel_super
*super
)
8005 struct intel_dev
*id
;
8008 dprintf("(enter)\n");
8009 if ((u
->subdev
< 0) ||
8011 dprintf("imsm: Error: Wrong subdev: %i\n", u
->subdev
);
8015 for (id
= super
->devlist
; id
; id
= id
->next
) {
8016 if (id
->index
== (unsigned)u
->subdev
) {
8017 struct imsm_dev
*dev
= get_imsm_dev(super
, u
->subdev
);
8018 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
8019 int used_disks
= imsm_num_data_members(dev
, MAP_0
);
8020 unsigned long long blocks_per_member
;
8022 /* calculate new size
8024 blocks_per_member
= u
->new_size
/ used_disks
;
8025 dprintf("(size: %llu, blocks per member: %llu)\n",
8026 u
->new_size
, blocks_per_member
);
8027 set_blocks_per_member(map
, blocks_per_member
);
8028 imsm_set_array_size(dev
, u
->new_size
);
8038 static int apply_update_activate_spare(struct imsm_update_activate_spare
*u
,
8039 struct intel_super
*super
,
8040 struct active_array
*active_array
)
8042 struct imsm_super
*mpb
= super
->anchor
;
8043 struct imsm_dev
*dev
= get_imsm_dev(super
, u
->array
);
8044 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
8045 struct imsm_map
*migr_map
;
8046 struct active_array
*a
;
8047 struct imsm_disk
*disk
;
8054 int second_map_created
= 0;
8056 for (; u
; u
= u
->next
) {
8057 victim
= get_imsm_disk_idx(dev
, u
->slot
, MAP_X
);
8062 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
8067 pr_err("error: imsm_activate_spare passed an unknown disk (index: %d)\n",
8072 /* count failures (excluding rebuilds and the victim)
8073 * to determine map[0] state
8076 for (i
= 0; i
< map
->num_members
; i
++) {
8079 disk
= get_imsm_disk(super
,
8080 get_imsm_disk_idx(dev
, i
, MAP_X
));
8081 if (!disk
|| is_failed(disk
))
8085 /* adding a pristine spare, assign a new index */
8086 if (dl
->index
< 0) {
8087 dl
->index
= super
->anchor
->num_disks
;
8088 super
->anchor
->num_disks
++;
8091 disk
->status
|= CONFIGURED_DISK
;
8092 disk
->status
&= ~SPARE_DISK
;
8095 to_state
= imsm_check_degraded(super
, dev
, failed
, MAP_0
);
8096 if (!second_map_created
) {
8097 second_map_created
= 1;
8098 map
->map_state
= IMSM_T_STATE_DEGRADED
;
8099 migrate(dev
, super
, to_state
, MIGR_REBUILD
);
8101 map
->map_state
= to_state
;
8102 migr_map
= get_imsm_map(dev
, MAP_1
);
8103 set_imsm_ord_tbl_ent(map
, u
->slot
, dl
->index
);
8104 set_imsm_ord_tbl_ent(migr_map
, u
->slot
,
8105 dl
->index
| IMSM_ORD_REBUILD
);
8107 /* update the family_num to mark a new container
8108 * generation, being careful to record the existing
8109 * family_num in orig_family_num to clean up after
8110 * earlier mdadm versions that neglected to set it.
8112 if (mpb
->orig_family_num
== 0)
8113 mpb
->orig_family_num
= mpb
->family_num
;
8114 mpb
->family_num
+= super
->random
;
8116 /* count arrays using the victim in the metadata */
8118 for (a
= active_array
; a
; a
= a
->next
) {
8119 dev
= get_imsm_dev(super
, a
->info
.container_member
);
8120 map
= get_imsm_map(dev
, MAP_0
);
8122 if (get_imsm_disk_slot(map
, victim
) >= 0)
8126 /* delete the victim if it is no longer being
8132 /* We know that 'manager' isn't touching anything,
8133 * so it is safe to delete
8135 for (dlp
= &super
->disks
; *dlp
; dlp
= &(*dlp
)->next
)
8136 if ((*dlp
)->index
== victim
)
8139 /* victim may be on the missing list */
8141 for (dlp
= &super
->missing
; *dlp
;
8142 dlp
= &(*dlp
)->next
)
8143 if ((*dlp
)->index
== victim
)
8145 imsm_delete(super
, dlp
, victim
);
8152 static int apply_reshape_container_disks_update(struct imsm_update_reshape
*u
,
8153 struct intel_super
*super
,
8156 struct dl
*new_disk
;
8157 struct intel_dev
*id
;
8159 int delta_disks
= u
->new_raid_disks
- u
->old_raid_disks
;
8160 int disk_count
= u
->old_raid_disks
;
8161 void **tofree
= NULL
;
8162 int devices_to_reshape
= 1;
8163 struct imsm_super
*mpb
= super
->anchor
;
8165 unsigned int dev_id
;
8167 dprintf("(enter)\n");
8169 /* enable spares to use in array */
8170 for (i
= 0; i
< delta_disks
; i
++) {
8171 new_disk
= get_disk_super(super
,
8172 major(u
->new_disks
[i
]),
8173 minor(u
->new_disks
[i
]));
8174 dprintf("imsm: new disk for reshape is: %i:%i (%p, index = %i)\n",
8175 major(u
->new_disks
[i
]), minor(u
->new_disks
[i
]),
8176 new_disk
, new_disk
->index
);
8177 if ((new_disk
== NULL
) ||
8178 ((new_disk
->index
>= 0) &&
8179 (new_disk
->index
< u
->old_raid_disks
)))
8180 goto update_reshape_exit
;
8181 new_disk
->index
= disk_count
++;
8182 /* slot to fill in autolayout
8184 new_disk
->raiddisk
= new_disk
->index
;
8185 new_disk
->disk
.status
|=
8187 new_disk
->disk
.status
&= ~SPARE_DISK
;
8190 dprintf("imsm: volume set mpb->num_raid_devs = %i\n",
8191 mpb
->num_raid_devs
);
8192 /* manage changes in volume
8194 for (dev_id
= 0; dev_id
< mpb
->num_raid_devs
; dev_id
++) {
8195 void **sp
= *space_list
;
8196 struct imsm_dev
*newdev
;
8197 struct imsm_map
*newmap
, *oldmap
;
8199 for (id
= super
->devlist
; id
; id
= id
->next
) {
8200 if (id
->index
== dev_id
)
8209 /* Copy the dev, but not (all of) the map */
8210 memcpy(newdev
, id
->dev
, sizeof(*newdev
));
8211 oldmap
= get_imsm_map(id
->dev
, MAP_0
);
8212 newmap
= get_imsm_map(newdev
, MAP_0
);
8213 /* Copy the current map */
8214 memcpy(newmap
, oldmap
, sizeof_imsm_map(oldmap
));
8215 /* update one device only
8217 if (devices_to_reshape
) {
8218 dprintf("imsm: modifying subdev: %i\n",
8220 devices_to_reshape
--;
8221 newdev
->vol
.migr_state
= 1;
8222 newdev
->vol
.curr_migr_unit
= 0;
8223 set_migr_type(newdev
, MIGR_GEN_MIGR
);
8224 newmap
->num_members
= u
->new_raid_disks
;
8225 for (i
= 0; i
< delta_disks
; i
++) {
8226 set_imsm_ord_tbl_ent(newmap
,
8227 u
->old_raid_disks
+ i
,
8228 u
->old_raid_disks
+ i
);
8230 /* New map is correct, now need to save old map
8232 newmap
= get_imsm_map(newdev
, MAP_1
);
8233 memcpy(newmap
, oldmap
, sizeof_imsm_map(oldmap
));
8235 imsm_set_array_size(newdev
, -1);
8238 sp
= (void **)id
->dev
;
8243 /* Clear migration record */
8244 memset(super
->migr_rec
, 0, sizeof(struct migr_record
));
8247 *space_list
= tofree
;
8250 update_reshape_exit
:
8255 static int apply_takeover_update(struct imsm_update_takeover
*u
,
8256 struct intel_super
*super
,
8259 struct imsm_dev
*dev
= NULL
;
8260 struct intel_dev
*dv
;
8261 struct imsm_dev
*dev_new
;
8262 struct imsm_map
*map
;
8266 for (dv
= super
->devlist
; dv
; dv
= dv
->next
)
8267 if (dv
->index
== (unsigned int)u
->subarray
) {
8275 map
= get_imsm_map(dev
, MAP_0
);
8277 if (u
->direction
== R10_TO_R0
) {
8278 /* Number of failed disks must be half of initial disk number */
8279 if (imsm_count_failed(super
, dev
, MAP_0
) !=
8280 (map
->num_members
/ 2))
8283 /* iterate through devices to mark removed disks as spare */
8284 for (dm
= super
->disks
; dm
; dm
= dm
->next
) {
8285 if (dm
->disk
.status
& FAILED_DISK
) {
8286 int idx
= dm
->index
;
8287 /* update indexes on the disk list */
8288 /* FIXME this loop-with-the-loop looks wrong, I'm not convinced
8289 the index values will end up being correct.... NB */
8290 for (du
= super
->disks
; du
; du
= du
->next
)
8291 if (du
->index
> idx
)
8293 /* mark as spare disk */
8298 map
->num_members
= map
->num_members
/ 2;
8299 map
->map_state
= IMSM_T_STATE_NORMAL
;
8300 map
->num_domains
= 1;
8301 map
->raid_level
= 0;
8302 map
->failed_disk_num
= -1;
8305 if (u
->direction
== R0_TO_R10
) {
8307 /* update slots in current disk list */
8308 for (dm
= super
->disks
; dm
; dm
= dm
->next
) {
8312 /* create new *missing* disks */
8313 for (i
= 0; i
< map
->num_members
; i
++) {
8314 space
= *space_list
;
8317 *space_list
= *space
;
8319 memcpy(du
, super
->disks
, sizeof(*du
));
8323 du
->index
= (i
* 2) + 1;
8324 sprintf((char *)du
->disk
.serial
,
8325 " MISSING_%d", du
->index
);
8326 sprintf((char *)du
->serial
,
8327 "MISSING_%d", du
->index
);
8328 du
->next
= super
->missing
;
8329 super
->missing
= du
;
8331 /* create new dev and map */
8332 space
= *space_list
;
8335 *space_list
= *space
;
8336 dev_new
= (void *)space
;
8337 memcpy(dev_new
, dev
, sizeof(*dev
));
8338 /* update new map */
8339 map
= get_imsm_map(dev_new
, MAP_0
);
8340 map
->num_members
= map
->num_members
* 2;
8341 map
->map_state
= IMSM_T_STATE_DEGRADED
;
8342 map
->num_domains
= 2;
8343 map
->raid_level
= 1;
8344 /* replace dev<->dev_new */
8347 /* update disk order table */
8348 for (du
= super
->disks
; du
; du
= du
->next
)
8350 set_imsm_ord_tbl_ent(map
, du
->index
, du
->index
);
8351 for (du
= super
->missing
; du
; du
= du
->next
)
8352 if (du
->index
>= 0) {
8353 set_imsm_ord_tbl_ent(map
, du
->index
, du
->index
);
8354 mark_missing(dv
->dev
, &du
->disk
, du
->index
);
8360 static void imsm_process_update(struct supertype
*st
,
8361 struct metadata_update
*update
)
8364 * crack open the metadata_update envelope to find the update record
8365 * update can be one of:
8366 * update_reshape_container_disks - all the arrays in the container
8367 * are being reshaped to have more devices. We need to mark
8368 * the arrays for general migration and convert selected spares
8369 * into active devices.
8370 * update_activate_spare - a spare device has replaced a failed
8371 * device in an array, update the disk_ord_tbl. If this disk is
8372 * present in all member arrays then also clear the SPARE_DISK
8374 * update_create_array
8376 * update_rename_array
8377 * update_add_remove_disk
8379 struct intel_super
*super
= st
->sb
;
8380 struct imsm_super
*mpb
;
8381 enum imsm_update_type type
= *(enum imsm_update_type
*) update
->buf
;
8383 /* update requires a larger buf but the allocation failed */
8384 if (super
->next_len
&& !super
->next_buf
) {
8385 super
->next_len
= 0;
8389 if (super
->next_buf
) {
8390 memcpy(super
->next_buf
, super
->buf
, super
->len
);
8392 super
->len
= super
->next_len
;
8393 super
->buf
= super
->next_buf
;
8395 super
->next_len
= 0;
8396 super
->next_buf
= NULL
;
8399 mpb
= super
->anchor
;
8402 case update_general_migration_checkpoint
: {
8403 struct intel_dev
*id
;
8404 struct imsm_update_general_migration_checkpoint
*u
=
8405 (void *)update
->buf
;
8407 dprintf("called for update_general_migration_checkpoint\n");
8409 /* find device under general migration */
8410 for (id
= super
->devlist
; id
; id
= id
->next
) {
8411 if (is_gen_migration(id
->dev
)) {
8412 id
->dev
->vol
.curr_migr_unit
=
8413 __cpu_to_le32(u
->curr_migr_unit
);
8414 super
->updates_pending
++;
8419 case update_takeover
: {
8420 struct imsm_update_takeover
*u
= (void *)update
->buf
;
8421 if (apply_takeover_update(u
, super
, &update
->space_list
)) {
8422 imsm_update_version_info(super
);
8423 super
->updates_pending
++;
8428 case update_reshape_container_disks
: {
8429 struct imsm_update_reshape
*u
= (void *)update
->buf
;
8430 if (apply_reshape_container_disks_update(
8431 u
, super
, &update
->space_list
))
8432 super
->updates_pending
++;
8435 case update_reshape_migration
: {
8436 struct imsm_update_reshape_migration
*u
= (void *)update
->buf
;
8437 if (apply_reshape_migration_update(
8438 u
, super
, &update
->space_list
))
8439 super
->updates_pending
++;
8442 case update_size_change
: {
8443 struct imsm_update_size_change
*u
= (void *)update
->buf
;
8444 if (apply_size_change_update(u
, super
))
8445 super
->updates_pending
++;
8448 case update_activate_spare
: {
8449 struct imsm_update_activate_spare
*u
= (void *) update
->buf
;
8450 if (apply_update_activate_spare(u
, super
, st
->arrays
))
8451 super
->updates_pending
++;
8454 case update_create_array
: {
8455 /* someone wants to create a new array, we need to be aware of
8456 * a few races/collisions:
8457 * 1/ 'Create' called by two separate instances of mdadm
8458 * 2/ 'Create' versus 'activate_spare': mdadm has chosen
8459 * devices that have since been assimilated via
8461 * In the event this update can not be carried out mdadm will
8462 * (FIX ME) notice that its update did not take hold.
8464 struct imsm_update_create_array
*u
= (void *) update
->buf
;
8465 struct intel_dev
*dv
;
8466 struct imsm_dev
*dev
;
8467 struct imsm_map
*map
, *new_map
;
8468 unsigned long long start
, end
;
8469 unsigned long long new_start
, new_end
;
8471 struct disk_info
*inf
;
8474 /* handle racing creates: first come first serve */
8475 if (u
->dev_idx
< mpb
->num_raid_devs
) {
8476 dprintf("subarray %d already defined\n", u
->dev_idx
);
8480 /* check update is next in sequence */
8481 if (u
->dev_idx
!= mpb
->num_raid_devs
) {
8482 dprintf("can not create array %d expected index %d\n",
8483 u
->dev_idx
, mpb
->num_raid_devs
);
8487 new_map
= get_imsm_map(&u
->dev
, MAP_0
);
8488 new_start
= pba_of_lba0(new_map
);
8489 new_end
= new_start
+ blocks_per_member(new_map
);
8490 inf
= get_disk_info(u
);
8492 /* handle activate_spare versus create race:
8493 * check to make sure that overlapping arrays do not include
8496 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
8497 dev
= get_imsm_dev(super
, i
);
8498 map
= get_imsm_map(dev
, MAP_0
);
8499 start
= pba_of_lba0(map
);
8500 end
= start
+ blocks_per_member(map
);
8501 if ((new_start
>= start
&& new_start
<= end
) ||
8502 (start
>= new_start
&& start
<= new_end
))
8507 if (disks_overlap(super
, i
, u
)) {
8508 dprintf("arrays overlap\n");
8513 /* check that prepare update was successful */
8514 if (!update
->space
) {
8515 dprintf("prepare update failed\n");
8519 /* check that all disks are still active before committing
8520 * changes. FIXME: could we instead handle this by creating a
8521 * degraded array? That's probably not what the user expects,
8522 * so better to drop this update on the floor.
8524 for (i
= 0; i
< new_map
->num_members
; i
++) {
8525 dl
= serial_to_dl(inf
[i
].serial
, super
);
8527 dprintf("disk disappeared\n");
8532 super
->updates_pending
++;
8534 /* convert spares to members and fixup ord_tbl */
8535 for (i
= 0; i
< new_map
->num_members
; i
++) {
8536 dl
= serial_to_dl(inf
[i
].serial
, super
);
8537 if (dl
->index
== -1) {
8538 dl
->index
= mpb
->num_disks
;
8540 dl
->disk
.status
|= CONFIGURED_DISK
;
8541 dl
->disk
.status
&= ~SPARE_DISK
;
8543 set_imsm_ord_tbl_ent(new_map
, i
, dl
->index
);
8548 update
->space
= NULL
;
8549 imsm_copy_dev(dev
, &u
->dev
);
8550 dv
->index
= u
->dev_idx
;
8551 dv
->next
= super
->devlist
;
8552 super
->devlist
= dv
;
8553 mpb
->num_raid_devs
++;
8555 imsm_update_version_info(super
);
8558 /* mdmon knows how to release update->space, but not
8559 * ((struct intel_dev *) update->space)->dev
8561 if (update
->space
) {
8567 case update_kill_array
: {
8568 struct imsm_update_kill_array
*u
= (void *) update
->buf
;
8569 int victim
= u
->dev_idx
;
8570 struct active_array
*a
;
8571 struct intel_dev
**dp
;
8572 struct imsm_dev
*dev
;
8574 /* sanity check that we are not affecting the uuid of
8575 * active arrays, or deleting an active array
8577 * FIXME when immutable ids are available, but note that
8578 * we'll also need to fixup the invalidated/active
8579 * subarray indexes in mdstat
8581 for (a
= st
->arrays
; a
; a
= a
->next
)
8582 if (a
->info
.container_member
>= victim
)
8584 /* by definition if mdmon is running at least one array
8585 * is active in the container, so checking
8586 * mpb->num_raid_devs is just extra paranoia
8588 dev
= get_imsm_dev(super
, victim
);
8589 if (a
|| !dev
|| mpb
->num_raid_devs
== 1) {
8590 dprintf("failed to delete subarray-%d\n", victim
);
8594 for (dp
= &super
->devlist
; *dp
;)
8595 if ((*dp
)->index
== (unsigned)super
->current_vol
) {
8598 if ((*dp
)->index
> (unsigned)victim
)
8602 mpb
->num_raid_devs
--;
8603 super
->updates_pending
++;
8606 case update_rename_array
: {
8607 struct imsm_update_rename_array
*u
= (void *) update
->buf
;
8608 char name
[MAX_RAID_SERIAL_LEN
+1];
8609 int target
= u
->dev_idx
;
8610 struct active_array
*a
;
8611 struct imsm_dev
*dev
;
8613 /* sanity check that we are not affecting the uuid of
8616 snprintf(name
, MAX_RAID_SERIAL_LEN
, "%s", (char *) u
->name
);
8617 name
[MAX_RAID_SERIAL_LEN
] = '\0';
8618 for (a
= st
->arrays
; a
; a
= a
->next
)
8619 if (a
->info
.container_member
== target
)
8621 dev
= get_imsm_dev(super
, u
->dev_idx
);
8622 if (a
|| !dev
|| !check_name(super
, name
, 1)) {
8623 dprintf("failed to rename subarray-%d\n", target
);
8627 snprintf((char *) dev
->volume
, MAX_RAID_SERIAL_LEN
, "%s", name
);
8628 super
->updates_pending
++;
8631 case update_add_remove_disk
: {
8632 /* we may be able to repair some arrays if disks are
8633 * being added, check the status of add_remove_disk
8634 * if discs has been added.
8636 if (add_remove_disk_update(super
)) {
8637 struct active_array
*a
;
8639 super
->updates_pending
++;
8640 for (a
= st
->arrays
; a
; a
= a
->next
)
8641 a
->check_degraded
= 1;
8646 pr_err("error: unsuported process update type:(type: %d)\n", type
);
8650 static struct mdinfo
*get_spares_for_grow(struct supertype
*st
);
8652 static int imsm_prepare_update(struct supertype
*st
,
8653 struct metadata_update
*update
)
8656 * Allocate space to hold new disk entries, raid-device entries or a new
8657 * mpb if necessary. The manager synchronously waits for updates to
8658 * complete in the monitor, so new mpb buffers allocated here can be
8659 * integrated by the monitor thread without worrying about live pointers
8660 * in the manager thread.
8662 enum imsm_update_type type
;
8663 struct intel_super
*super
= st
->sb
;
8664 struct imsm_super
*mpb
= super
->anchor
;
8668 if (update
->len
< (int)sizeof(type
))
8671 type
= *(enum imsm_update_type
*) update
->buf
;
8674 case update_general_migration_checkpoint
:
8675 if (update
->len
< (int)sizeof(struct imsm_update_general_migration_checkpoint
))
8677 dprintf("called for update_general_migration_checkpoint\n");
8679 case update_takeover
: {
8680 struct imsm_update_takeover
*u
= (void *)update
->buf
;
8681 if (update
->len
< (int)sizeof(*u
))
8683 if (u
->direction
== R0_TO_R10
) {
8684 void **tail
= (void **)&update
->space_list
;
8685 struct imsm_dev
*dev
= get_imsm_dev(super
, u
->subarray
);
8686 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
8687 int num_members
= map
->num_members
;
8690 /* allocate memory for added disks */
8691 for (i
= 0; i
< num_members
; i
++) {
8692 size
= sizeof(struct dl
);
8693 space
= xmalloc(size
);
8698 /* allocate memory for new device */
8699 size
= sizeof_imsm_dev(super
->devlist
->dev
, 0) +
8700 (num_members
* sizeof(__u32
));
8701 space
= xmalloc(size
);
8705 len
= disks_to_mpb_size(num_members
* 2);
8710 case update_reshape_container_disks
: {
8711 /* Every raid device in the container is about to
8712 * gain some more devices, and we will enter a
8714 * So each 'imsm_map' will be bigger, and the imsm_vol
8715 * will now hold 2 of them.
8716 * Thus we need new 'struct imsm_dev' allocations sized
8717 * as sizeof_imsm_dev but with more devices in both maps.
8719 struct imsm_update_reshape
*u
= (void *)update
->buf
;
8720 struct intel_dev
*dl
;
8721 void **space_tail
= (void**)&update
->space_list
;
8723 if (update
->len
< (int)sizeof(*u
))
8726 dprintf("for update_reshape\n");
8728 for (dl
= super
->devlist
; dl
; dl
= dl
->next
) {
8729 int size
= sizeof_imsm_dev(dl
->dev
, 1);
8731 if (u
->new_raid_disks
> u
->old_raid_disks
)
8732 size
+= sizeof(__u32
)*2*
8733 (u
->new_raid_disks
- u
->old_raid_disks
);
8740 len
= disks_to_mpb_size(u
->new_raid_disks
);
8741 dprintf("New anchor length is %llu\n", (unsigned long long)len
);
8744 case update_reshape_migration
: {
8745 /* for migration level 0->5 we need to add disks
8746 * so the same as for container operation we will copy
8747 * device to the bigger location.
8748 * in memory prepared device and new disk area are prepared
8749 * for usage in process update
8751 struct imsm_update_reshape_migration
*u
= (void *)update
->buf
;
8752 struct intel_dev
*id
;
8753 void **space_tail
= (void **)&update
->space_list
;
8756 int current_level
= -1;
8758 if (update
->len
< (int)sizeof(*u
))
8761 dprintf("for update_reshape\n");
8763 /* add space for bigger array in update
8765 for (id
= super
->devlist
; id
; id
= id
->next
) {
8766 if (id
->index
== (unsigned)u
->subdev
) {
8767 size
= sizeof_imsm_dev(id
->dev
, 1);
8768 if (u
->new_raid_disks
> u
->old_raid_disks
)
8769 size
+= sizeof(__u32
)*2*
8770 (u
->new_raid_disks
- u
->old_raid_disks
);
8778 if (update
->space_list
== NULL
)
8781 /* add space for disk in update
8783 size
= sizeof(struct dl
);
8789 /* add spare device to update
8791 for (id
= super
->devlist
; id
; id
= id
->next
)
8792 if (id
->index
== (unsigned)u
->subdev
) {
8793 struct imsm_dev
*dev
;
8794 struct imsm_map
*map
;
8796 dev
= get_imsm_dev(super
, u
->subdev
);
8797 map
= get_imsm_map(dev
, MAP_0
);
8798 current_level
= map
->raid_level
;
8801 if ((u
->new_level
== 5) && (u
->new_level
!= current_level
)) {
8802 struct mdinfo
*spares
;
8804 spares
= get_spares_for_grow(st
);
8812 makedev(dev
->disk
.major
,
8814 dl
= get_disk_super(super
,
8817 dl
->index
= u
->old_raid_disks
;
8823 len
= disks_to_mpb_size(u
->new_raid_disks
);
8824 dprintf("New anchor length is %llu\n", (unsigned long long)len
);
8827 case update_size_change
: {
8828 if (update
->len
< (int)sizeof(struct imsm_update_size_change
))
8832 case update_activate_spare
: {
8833 if (update
->len
< (int)sizeof(struct imsm_update_activate_spare
))
8837 case update_create_array
: {
8838 struct imsm_update_create_array
*u
= (void *) update
->buf
;
8839 struct intel_dev
*dv
;
8840 struct imsm_dev
*dev
= &u
->dev
;
8841 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
8843 struct disk_info
*inf
;
8847 if (update
->len
< (int)sizeof(*u
))
8850 inf
= get_disk_info(u
);
8851 len
= sizeof_imsm_dev(dev
, 1);
8852 /* allocate a new super->devlist entry */
8853 dv
= xmalloc(sizeof(*dv
));
8854 dv
->dev
= xmalloc(len
);
8857 /* count how many spares will be converted to members */
8858 for (i
= 0; i
< map
->num_members
; i
++) {
8859 dl
= serial_to_dl(inf
[i
].serial
, super
);
8861 /* hmm maybe it failed?, nothing we can do about
8866 if (count_memberships(dl
, super
) == 0)
8869 len
+= activate
* sizeof(struct imsm_disk
);
8872 case update_kill_array
: {
8873 if (update
->len
< (int)sizeof(struct imsm_update_kill_array
))
8877 case update_rename_array
: {
8878 if (update
->len
< (int)sizeof(struct imsm_update_rename_array
))
8882 case update_add_remove_disk
:
8883 /* no update->len needed */
8889 /* check if we need a larger metadata buffer */
8890 if (super
->next_buf
)
8891 buf_len
= super
->next_len
;
8893 buf_len
= super
->len
;
8895 if (__le32_to_cpu(mpb
->mpb_size
) + len
> buf_len
) {
8896 /* ok we need a larger buf than what is currently allocated
8897 * if this allocation fails process_update will notice that
8898 * ->next_len is set and ->next_buf is NULL
8900 buf_len
= ROUND_UP(__le32_to_cpu(mpb
->mpb_size
) + len
, 512);
8901 if (super
->next_buf
)
8902 free(super
->next_buf
);
8904 super
->next_len
= buf_len
;
8905 if (posix_memalign(&super
->next_buf
, 512, buf_len
) == 0)
8906 memset(super
->next_buf
, 0, buf_len
);
8908 super
->next_buf
= NULL
;
8913 /* must be called while manager is quiesced */
8914 static void imsm_delete(struct intel_super
*super
, struct dl
**dlp
, unsigned index
)
8916 struct imsm_super
*mpb
= super
->anchor
;
8918 struct imsm_dev
*dev
;
8919 struct imsm_map
*map
;
8920 int i
, j
, num_members
;
8923 dprintf("deleting device[%d] from imsm_super\n", index
);
8925 /* shift all indexes down one */
8926 for (iter
= super
->disks
; iter
; iter
= iter
->next
)
8927 if (iter
->index
> (int)index
)
8929 for (iter
= super
->missing
; iter
; iter
= iter
->next
)
8930 if (iter
->index
> (int)index
)
8933 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
8934 dev
= get_imsm_dev(super
, i
);
8935 map
= get_imsm_map(dev
, MAP_0
);
8936 num_members
= map
->num_members
;
8937 for (j
= 0; j
< num_members
; j
++) {
8938 /* update ord entries being careful not to propagate
8939 * ord-flags to the first map
8941 ord
= get_imsm_ord_tbl_ent(dev
, j
, MAP_X
);
8943 if (ord_to_idx(ord
) <= index
)
8946 map
= get_imsm_map(dev
, MAP_0
);
8947 set_imsm_ord_tbl_ent(map
, j
, ord_to_idx(ord
- 1));
8948 map
= get_imsm_map(dev
, MAP_1
);
8950 set_imsm_ord_tbl_ent(map
, j
, ord
- 1);
8955 super
->updates_pending
++;
8957 struct dl
*dl
= *dlp
;
8959 *dlp
= (*dlp
)->next
;
8960 __free_imsm_disk(dl
);
8963 #endif /* MDASSEMBLE */
8965 static void close_targets(int *targets
, int new_disks
)
8972 for (i
= 0; i
< new_disks
; i
++) {
8973 if (targets
[i
] >= 0) {
8980 static int imsm_get_allowed_degradation(int level
, int raid_disks
,
8981 struct intel_super
*super
,
8982 struct imsm_dev
*dev
)
8988 struct imsm_map
*map
;
8991 ret_val
= raid_disks
/2;
8992 /* check map if all disks pairs not failed
8995 map
= get_imsm_map(dev
, MAP_0
);
8996 for (i
= 0; i
< ret_val
; i
++) {
8997 int degradation
= 0;
8998 if (get_imsm_disk(super
, i
) == NULL
)
9000 if (get_imsm_disk(super
, i
+ 1) == NULL
)
9002 if (degradation
== 2)
9005 map
= get_imsm_map(dev
, MAP_1
);
9006 /* if there is no second map
9007 * result can be returned
9011 /* check degradation in second map
9013 for (i
= 0; i
< ret_val
; i
++) {
9014 int degradation
= 0;
9015 if (get_imsm_disk(super
, i
) == NULL
)
9017 if (get_imsm_disk(super
, i
+ 1) == NULL
)
9019 if (degradation
== 2)
9033 /*******************************************************************************
9034 * Function: open_backup_targets
9035 * Description: Function opens file descriptors for all devices given in
9038 * info : general array info
9039 * raid_disks : number of disks
9040 * raid_fds : table of device's file descriptors
9041 * super : intel super for raid10 degradation check
9042 * dev : intel device for raid10 degradation check
9046 ******************************************************************************/
9047 int open_backup_targets(struct mdinfo
*info
, int raid_disks
, int *raid_fds
,
9048 struct intel_super
*super
, struct imsm_dev
*dev
)
9054 for (i
= 0; i
< raid_disks
; i
++)
9057 for (sd
= info
->devs
; sd
; sd
= sd
->next
) {
9060 if (sd
->disk
.state
& (1<<MD_DISK_FAULTY
)) {
9061 dprintf("disk is faulty!!\n");
9065 if ((sd
->disk
.raid_disk
>= raid_disks
) ||
9066 (sd
->disk
.raid_disk
< 0))
9069 dn
= map_dev(sd
->disk
.major
,
9071 raid_fds
[sd
->disk
.raid_disk
] = dev_open(dn
, O_RDWR
);
9072 if (raid_fds
[sd
->disk
.raid_disk
] < 0) {
9073 pr_err("cannot open component\n");
9078 /* check if maximum array degradation level is not exceeded
9080 if ((raid_disks
- opened
) >
9081 imsm_get_allowed_degradation(info
->new_level
,
9084 pr_err("Not enough disks can be opened.\n");
9085 close_targets(raid_fds
, raid_disks
);
9091 /*******************************************************************************
9092 * Function: validate_container_imsm
9093 * Description: This routine validates container after assemble,
9094 * eg. if devices in container are under the same controller.
9097 * info : linked list with info about devices used in array
9101 ******************************************************************************/
9102 int validate_container_imsm(struct mdinfo
*info
)
9104 if (check_env("IMSM_NO_PLATFORM"))
9107 struct sys_dev
*idev
;
9108 struct sys_dev
*hba
= NULL
;
9109 struct sys_dev
*intel_devices
= find_intel_devices();
9110 char *dev_path
= devt_to_devpath(makedev(info
->disk
.major
,
9113 for (idev
= intel_devices
; idev
; idev
= idev
->next
) {
9114 if (dev_path
&& strstr(dev_path
, idev
->path
)) {
9123 pr_err("WARNING - Cannot detect HBA for device %s!\n",
9124 devid2kname(makedev(info
->disk
.major
, info
->disk
.minor
)));
9128 const struct imsm_orom
*orom
= get_orom_by_device_id(hba
->dev_id
);
9131 for (dev
= info
->next
; dev
; dev
= dev
->next
) {
9132 dev_path
= devt_to_devpath(makedev(dev
->disk
.major
, dev
->disk
.minor
));
9134 struct sys_dev
*hba2
= NULL
;
9135 for (idev
= intel_devices
; idev
; idev
= idev
->next
) {
9136 if (dev_path
&& strstr(dev_path
, idev
->path
)) {
9144 const struct imsm_orom
*orom2
= hba2
== NULL
? NULL
:
9145 get_orom_by_device_id(hba2
->dev_id
);
9147 if (hba2
&& hba
->type
!= hba2
->type
) {
9148 pr_err("WARNING - HBAs of devices do not match %s != %s\n",
9149 get_sys_dev_type(hba
->type
), get_sys_dev_type(hba2
->type
));
9153 if (orom
!= orom2
) {
9154 pr_err("WARNING - IMSM container assembled with disks under different HBAs!\n"
9155 " This operation is not supported and can lead to data loss.\n");
9160 pr_err("WARNING - IMSM container assembled with disks under HBAs without IMSM platform support!\n"
9161 " This operation is not supported and can lead to data loss.\n");
9169 /*******************************************************************************
9170 * Function: init_migr_record_imsm
9171 * Description: Function inits imsm migration record
9173 * super : imsm internal array info
9174 * dev : device under migration
9175 * info : general array info to find the smallest device
9178 ******************************************************************************/
9179 void init_migr_record_imsm(struct supertype
*st
, struct imsm_dev
*dev
,
9180 struct mdinfo
*info
)
9182 struct intel_super
*super
= st
->sb
;
9183 struct migr_record
*migr_rec
= super
->migr_rec
;
9185 unsigned long long dsize
, dev_sectors
;
9186 long long unsigned min_dev_sectors
= -1LLU;
9190 struct imsm_map
*map_dest
= get_imsm_map(dev
, MAP_0
);
9191 struct imsm_map
*map_src
= get_imsm_map(dev
, MAP_1
);
9192 unsigned long long num_migr_units
;
9193 unsigned long long array_blocks
;
9195 memset(migr_rec
, 0, sizeof(struct migr_record
));
9196 migr_rec
->family_num
= __cpu_to_le32(super
->anchor
->family_num
);
9198 /* only ascending reshape supported now */
9199 migr_rec
->ascending_migr
= __cpu_to_le32(1);
9201 migr_rec
->dest_depth_per_unit
= GEN_MIGR_AREA_SIZE
/
9202 max(map_dest
->blocks_per_strip
, map_src
->blocks_per_strip
);
9203 migr_rec
->dest_depth_per_unit
*=
9204 max(map_dest
->blocks_per_strip
, map_src
->blocks_per_strip
);
9205 new_data_disks
= imsm_num_data_members(dev
, MAP_0
);
9206 migr_rec
->blocks_per_unit
=
9207 __cpu_to_le32(migr_rec
->dest_depth_per_unit
* new_data_disks
);
9208 migr_rec
->dest_depth_per_unit
=
9209 __cpu_to_le32(migr_rec
->dest_depth_per_unit
);
9210 array_blocks
= info
->component_size
* new_data_disks
;
9212 array_blocks
/ __le32_to_cpu(migr_rec
->blocks_per_unit
);
9214 if (array_blocks
% __le32_to_cpu(migr_rec
->blocks_per_unit
))
9216 migr_rec
->num_migr_units
= __cpu_to_le32(num_migr_units
);
9218 migr_rec
->post_migr_vol_cap
= dev
->size_low
;
9219 migr_rec
->post_migr_vol_cap_hi
= dev
->size_high
;
9221 /* Find the smallest dev */
9222 for (sd
= info
->devs
; sd
; sd
= sd
->next
) {
9223 sprintf(nm
, "%d:%d", sd
->disk
.major
, sd
->disk
.minor
);
9224 fd
= dev_open(nm
, O_RDONLY
);
9227 get_dev_size(fd
, NULL
, &dsize
);
9228 dev_sectors
= dsize
/ 512;
9229 if (dev_sectors
< min_dev_sectors
)
9230 min_dev_sectors
= dev_sectors
;
9233 migr_rec
->ckpt_area_pba
= __cpu_to_le32(min_dev_sectors
-
9234 RAID_DISK_RESERVED_BLOCKS_IMSM_HI
);
9236 write_imsm_migr_rec(st
);
9241 /*******************************************************************************
9242 * Function: save_backup_imsm
9243 * Description: Function saves critical data stripes to Migration Copy Area
9244 * and updates the current migration unit status.
9245 * Use restore_stripes() to form a destination stripe,
9246 * and to write it to the Copy Area.
9248 * st : supertype information
9249 * dev : imsm device that backup is saved for
9250 * info : general array info
9251 * buf : input buffer
9252 * length : length of data to backup (blocks_per_unit)
9256 ******************************************************************************/
9257 int save_backup_imsm(struct supertype
*st
,
9258 struct imsm_dev
*dev
,
9259 struct mdinfo
*info
,
9264 struct intel_super
*super
= st
->sb
;
9265 unsigned long long *target_offsets
= NULL
;
9266 int *targets
= NULL
;
9268 struct imsm_map
*map_dest
= get_imsm_map(dev
, MAP_0
);
9269 int new_disks
= map_dest
->num_members
;
9270 int dest_layout
= 0;
9272 unsigned long long start
;
9273 int data_disks
= imsm_num_data_members(dev
, MAP_0
);
9275 targets
= xmalloc(new_disks
* sizeof(int));
9277 for (i
= 0; i
< new_disks
; i
++)
9280 target_offsets
= xcalloc(new_disks
, sizeof(unsigned long long));
9282 start
= info
->reshape_progress
* 512;
9283 for (i
= 0; i
< new_disks
; i
++) {
9284 target_offsets
[i
] = (unsigned long long)
9285 __le32_to_cpu(super
->migr_rec
->ckpt_area_pba
) * 512;
9286 /* move back copy area adderss, it will be moved forward
9287 * in restore_stripes() using start input variable
9289 target_offsets
[i
] -= start
/data_disks
;
9292 if (open_backup_targets(info
, new_disks
, targets
,
9296 dest_layout
= imsm_level_to_layout(map_dest
->raid_level
);
9297 dest_chunk
= __le16_to_cpu(map_dest
->blocks_per_strip
) * 512;
9299 if (restore_stripes(targets
, /* list of dest devices */
9300 target_offsets
, /* migration record offsets */
9303 map_dest
->raid_level
,
9305 -1, /* source backup file descriptor */
9306 0, /* input buf offset
9307 * always 0 buf is already offseted */
9311 pr_err("Error restoring stripes\n");
9319 close_targets(targets
, new_disks
);
9322 free(target_offsets
);
9327 /*******************************************************************************
9328 * Function: save_checkpoint_imsm
9329 * Description: Function called for current unit status update
9330 * in the migration record. It writes it to disk.
9332 * super : imsm internal array info
9333 * info : general array info
9337 * 2: failure, means no valid migration record
9338 * / no general migration in progress /
9339 ******************************************************************************/
9340 int save_checkpoint_imsm(struct supertype
*st
, struct mdinfo
*info
, int state
)
9342 struct intel_super
*super
= st
->sb
;
9343 unsigned long long blocks_per_unit
;
9344 unsigned long long curr_migr_unit
;
9346 if (load_imsm_migr_rec(super
, info
) != 0) {
9347 dprintf("imsm: ERROR: Cannot read migration record for checkpoint save.\n");
9351 blocks_per_unit
= __le32_to_cpu(super
->migr_rec
->blocks_per_unit
);
9352 if (blocks_per_unit
== 0) {
9353 dprintf("imsm: no migration in progress.\n");
9356 curr_migr_unit
= info
->reshape_progress
/ blocks_per_unit
;
9357 /* check if array is alligned to copy area
9358 * if it is not alligned, add one to current migration unit value
9359 * this can happend on array reshape finish only
9361 if (info
->reshape_progress
% blocks_per_unit
)
9364 super
->migr_rec
->curr_migr_unit
=
9365 __cpu_to_le32(curr_migr_unit
);
9366 super
->migr_rec
->rec_status
= __cpu_to_le32(state
);
9367 super
->migr_rec
->dest_1st_member_lba
=
9368 __cpu_to_le32(curr_migr_unit
*
9369 __le32_to_cpu(super
->migr_rec
->dest_depth_per_unit
));
9370 if (write_imsm_migr_rec(st
) < 0) {
9371 dprintf("imsm: Cannot write migration record outside backup area\n");
9378 /*******************************************************************************
9379 * Function: recover_backup_imsm
9380 * Description: Function recovers critical data from the Migration Copy Area
9381 * while assembling an array.
9383 * super : imsm internal array info
9384 * info : general array info
9386 * 0 : success (or there is no data to recover)
9388 ******************************************************************************/
9389 int recover_backup_imsm(struct supertype
*st
, struct mdinfo
*info
)
9391 struct intel_super
*super
= st
->sb
;
9392 struct migr_record
*migr_rec
= super
->migr_rec
;
9393 struct imsm_map
*map_dest
= NULL
;
9394 struct intel_dev
*id
= NULL
;
9395 unsigned long long read_offset
;
9396 unsigned long long write_offset
;
9398 int *targets
= NULL
;
9399 int new_disks
, i
, err
;
9402 unsigned long curr_migr_unit
= __le32_to_cpu(migr_rec
->curr_migr_unit
);
9403 unsigned long num_migr_units
= __le32_to_cpu(migr_rec
->num_migr_units
);
9405 int skipped_disks
= 0;
9407 err
= sysfs_get_str(info
, NULL
, "array_state", (char *)buffer
, 20);
9411 /* recover data only during assemblation */
9412 if (strncmp(buffer
, "inactive", 8) != 0)
9414 /* no data to recover */
9415 if (__le32_to_cpu(migr_rec
->rec_status
) == UNIT_SRC_NORMAL
)
9417 if (curr_migr_unit
>= num_migr_units
)
9420 /* find device during reshape */
9421 for (id
= super
->devlist
; id
; id
= id
->next
)
9422 if (is_gen_migration(id
->dev
))
9427 map_dest
= get_imsm_map(id
->dev
, MAP_0
);
9428 new_disks
= map_dest
->num_members
;
9430 read_offset
= (unsigned long long)
9431 __le32_to_cpu(migr_rec
->ckpt_area_pba
) * 512;
9433 write_offset
= ((unsigned long long)
9434 __le32_to_cpu(migr_rec
->dest_1st_member_lba
) +
9435 pba_of_lba0(map_dest
)) * 512;
9437 unit_len
= __le32_to_cpu(migr_rec
->dest_depth_per_unit
) * 512;
9438 if (posix_memalign((void **)&buf
, 512, unit_len
) != 0)
9440 targets
= xcalloc(new_disks
, sizeof(int));
9442 if (open_backup_targets(info
, new_disks
, targets
, super
, id
->dev
)) {
9443 pr_err("Cannot open some devices belonging to array.\n");
9447 for (i
= 0; i
< new_disks
; i
++) {
9448 if (targets
[i
] < 0) {
9452 if (lseek64(targets
[i
], read_offset
, SEEK_SET
) < 0) {
9453 pr_err("Cannot seek to block: %s\n",
9458 if ((unsigned)read(targets
[i
], buf
, unit_len
) != unit_len
) {
9459 pr_err("Cannot read copy area block: %s\n",
9464 if (lseek64(targets
[i
], write_offset
, SEEK_SET
) < 0) {
9465 pr_err("Cannot seek to block: %s\n",
9470 if ((unsigned)write(targets
[i
], buf
, unit_len
) != unit_len
) {
9471 pr_err("Cannot restore block: %s\n",
9478 if (skipped_disks
> imsm_get_allowed_degradation(info
->new_level
,
9482 pr_err("Cannot restore data from backup. Too many failed disks\n");
9486 if (save_checkpoint_imsm(st
, info
, UNIT_SRC_NORMAL
)) {
9487 /* ignore error == 2, this can mean end of reshape here
9489 dprintf("imsm: Cannot write checkpoint to migration record (UNIT_SRC_NORMAL) during restart\n");
9495 for (i
= 0; i
< new_disks
; i
++)
9504 static char disk_by_path
[] = "/dev/disk/by-path/";
9506 static const char *imsm_get_disk_controller_domain(const char *path
)
9508 char disk_path
[PATH_MAX
];
9512 strcpy(disk_path
, disk_by_path
);
9513 strncat(disk_path
, path
, PATH_MAX
- strlen(disk_path
) - 1);
9514 if (stat(disk_path
, &st
) == 0) {
9515 struct sys_dev
* hba
;
9518 path
= devt_to_devpath(st
.st_rdev
);
9521 hba
= find_disk_attached_hba(-1, path
);
9522 if (hba
&& hba
->type
== SYS_DEV_SAS
)
9524 else if (hba
&& hba
->type
== SYS_DEV_SATA
)
9528 dprintf("path: %s hba: %s attached: %s\n",
9529 path
, (hba
) ? hba
->path
: "NULL", drv
);
9535 static char *imsm_find_array_devnm_by_subdev(int subdev
, char *container
)
9537 static char devnm
[32];
9538 char subdev_name
[20];
9539 struct mdstat_ent
*mdstat
;
9541 sprintf(subdev_name
, "%d", subdev
);
9542 mdstat
= mdstat_by_subdev(subdev_name
, container
);
9546 strcpy(devnm
, mdstat
->devnm
);
9547 free_mdstat(mdstat
);
9551 static int imsm_reshape_is_allowed_on_container(struct supertype
*st
,
9552 struct geo_params
*geo
,
9553 int *old_raid_disks
,
9556 /* currently we only support increasing the number of devices
9557 * for a container. This increases the number of device for each
9558 * member array. They must all be RAID0 or RAID5.
9561 struct mdinfo
*info
, *member
;
9562 int devices_that_can_grow
= 0;
9564 dprintf("imsm: imsm_reshape_is_allowed_on_container(ENTER): st->devnm = (%s)\n", st
->devnm
);
9566 if (geo
->size
> 0 ||
9567 geo
->level
!= UnSet
||
9568 geo
->layout
!= UnSet
||
9569 geo
->chunksize
!= 0 ||
9570 geo
->raid_disks
== UnSet
) {
9571 dprintf("imsm: Container operation is allowed for raid disks number change only.\n");
9575 if (direction
== ROLLBACK_METADATA_CHANGES
) {
9576 dprintf("imsm: Metadata changes rollback is not supported for container operation.\n");
9580 info
= container_content_imsm(st
, NULL
);
9581 for (member
= info
; member
; member
= member
->next
) {
9584 dprintf("imsm: checking device_num: %i\n",
9585 member
->container_member
);
9587 if (geo
->raid_disks
<= member
->array
.raid_disks
) {
9588 /* we work on container for Online Capacity Expansion
9589 * only so raid_disks has to grow
9591 dprintf("imsm: for container operation raid disks increase is required\n");
9595 if ((info
->array
.level
!= 0) &&
9596 (info
->array
.level
!= 5)) {
9597 /* we cannot use this container with other raid level
9599 dprintf("imsm: for container operation wrong raid level (%i) detected\n",
9603 /* check for platform support
9604 * for this raid level configuration
9606 struct intel_super
*super
= st
->sb
;
9607 if (!is_raid_level_supported(super
->orom
,
9608 member
->array
.level
,
9610 dprintf("platform does not support raid%d with %d disk%s\n",
9613 geo
->raid_disks
> 1 ? "s" : "");
9616 /* check if component size is aligned to chunk size
9618 if (info
->component_size
%
9619 (info
->array
.chunk_size
/512)) {
9620 dprintf("Component size is not aligned to chunk size\n");
9625 if (*old_raid_disks
&&
9626 info
->array
.raid_disks
!= *old_raid_disks
)
9628 *old_raid_disks
= info
->array
.raid_disks
;
9630 /* All raid5 and raid0 volumes in container
9631 * have to be ready for Online Capacity Expansion
9632 * so they need to be assembled. We have already
9633 * checked that no recovery etc is happening.
9635 result
= imsm_find_array_devnm_by_subdev(member
->container_member
,
9636 st
->container_devnm
);
9637 if (result
== NULL
) {
9638 dprintf("imsm: cannot find array\n");
9641 devices_that_can_grow
++;
9644 if (!member
&& devices_that_can_grow
)
9648 dprintf("Container operation allowed\n");
9650 dprintf("Error: %i\n", ret_val
);
9655 /* Function: get_spares_for_grow
9656 * Description: Allocates memory and creates list of spare devices
9657 * avaliable in container. Checks if spare drive size is acceptable.
9658 * Parameters: Pointer to the supertype structure
9659 * Returns: Pointer to the list of spare devices (mdinfo structure) on success,
9662 static struct mdinfo
*get_spares_for_grow(struct supertype
*st
)
9664 unsigned long long min_size
= min_acceptable_spare_size_imsm(st
);
9665 return container_choose_spares(st
, min_size
, NULL
, NULL
, NULL
, 0);
9668 /******************************************************************************
9669 * function: imsm_create_metadata_update_for_reshape
9670 * Function creates update for whole IMSM container.
9672 ******************************************************************************/
9673 static int imsm_create_metadata_update_for_reshape(
9674 struct supertype
*st
,
9675 struct geo_params
*geo
,
9677 struct imsm_update_reshape
**updatep
)
9679 struct intel_super
*super
= st
->sb
;
9680 struct imsm_super
*mpb
= super
->anchor
;
9681 int update_memory_size
= 0;
9682 struct imsm_update_reshape
*u
= NULL
;
9683 struct mdinfo
*spares
= NULL
;
9685 int delta_disks
= 0;
9688 dprintf("(enter) raid_disks = %i\n", geo
->raid_disks
);
9690 delta_disks
= geo
->raid_disks
- old_raid_disks
;
9692 /* size of all update data without anchor */
9693 update_memory_size
= sizeof(struct imsm_update_reshape
);
9695 /* now add space for spare disks that we need to add. */
9696 update_memory_size
+= sizeof(u
->new_disks
[0]) * (delta_disks
- 1);
9698 u
= xcalloc(1, update_memory_size
);
9699 u
->type
= update_reshape_container_disks
;
9700 u
->old_raid_disks
= old_raid_disks
;
9701 u
->new_raid_disks
= geo
->raid_disks
;
9703 /* now get spare disks list
9705 spares
= get_spares_for_grow(st
);
9708 || delta_disks
> spares
->array
.spare_disks
) {
9709 pr_err("imsm: ERROR: Cannot get spare devices for %s.\n", geo
->dev_name
);
9714 /* we have got spares
9715 * update disk list in imsm_disk list table in anchor
9717 dprintf("imsm: %i spares are available.\n\n",
9718 spares
->array
.spare_disks
);
9721 for (i
= 0; i
< delta_disks
; i
++) {
9726 u
->new_disks
[i
] = makedev(dev
->disk
.major
,
9728 dl
= get_disk_super(super
, dev
->disk
.major
, dev
->disk
.minor
);
9729 dl
->index
= mpb
->num_disks
;
9739 dprintf("imsm: reshape update preparation :");
9740 if (i
== delta_disks
) {
9741 dprintf_cont(" OK\n");
9743 return update_memory_size
;
9746 dprintf_cont(" Error\n");
9751 /******************************************************************************
9752 * function: imsm_create_metadata_update_for_size_change()
9753 * Creates update for IMSM array for array size change.
9755 ******************************************************************************/
9756 static int imsm_create_metadata_update_for_size_change(
9757 struct supertype
*st
,
9758 struct geo_params
*geo
,
9759 struct imsm_update_size_change
**updatep
)
9761 struct intel_super
*super
= st
->sb
;
9762 int update_memory_size
= 0;
9763 struct imsm_update_size_change
*u
= NULL
;
9765 dprintf("(enter) New size = %llu\n", geo
->size
);
9767 /* size of all update data without anchor */
9768 update_memory_size
= sizeof(struct imsm_update_size_change
);
9770 u
= xcalloc(1, update_memory_size
);
9771 u
->type
= update_size_change
;
9772 u
->subdev
= super
->current_vol
;
9773 u
->new_size
= geo
->size
;
9775 dprintf("imsm: reshape update preparation : OK\n");
9778 return update_memory_size
;
9781 /******************************************************************************
9782 * function: imsm_create_metadata_update_for_migration()
9783 * Creates update for IMSM array.
9785 ******************************************************************************/
9786 static int imsm_create_metadata_update_for_migration(
9787 struct supertype
*st
,
9788 struct geo_params
*geo
,
9789 struct imsm_update_reshape_migration
**updatep
)
9791 struct intel_super
*super
= st
->sb
;
9792 int update_memory_size
= 0;
9793 struct imsm_update_reshape_migration
*u
= NULL
;
9794 struct imsm_dev
*dev
;
9795 int previous_level
= -1;
9797 dprintf("(enter) New Level = %i\n", geo
->level
);
9799 /* size of all update data without anchor */
9800 update_memory_size
= sizeof(struct imsm_update_reshape_migration
);
9802 u
= xcalloc(1, update_memory_size
);
9803 u
->type
= update_reshape_migration
;
9804 u
->subdev
= super
->current_vol
;
9805 u
->new_level
= geo
->level
;
9806 u
->new_layout
= geo
->layout
;
9807 u
->new_raid_disks
= u
->old_raid_disks
= geo
->raid_disks
;
9808 u
->new_disks
[0] = -1;
9809 u
->new_chunksize
= -1;
9811 dev
= get_imsm_dev(super
, u
->subdev
);
9813 struct imsm_map
*map
;
9815 map
= get_imsm_map(dev
, MAP_0
);
9817 int current_chunk_size
=
9818 __le16_to_cpu(map
->blocks_per_strip
) / 2;
9820 if (geo
->chunksize
!= current_chunk_size
) {
9821 u
->new_chunksize
= geo
->chunksize
/ 1024;
9822 dprintf("imsm: chunk size change from %i to %i\n",
9823 current_chunk_size
, u
->new_chunksize
);
9825 previous_level
= map
->raid_level
;
9828 if ((geo
->level
== 5) && (previous_level
== 0)) {
9829 struct mdinfo
*spares
= NULL
;
9831 u
->new_raid_disks
++;
9832 spares
= get_spares_for_grow(st
);
9833 if ((spares
== NULL
) || (spares
->array
.spare_disks
< 1)) {
9836 update_memory_size
= 0;
9837 dprintf("error: cannot get spare device for requested migration");
9842 dprintf("imsm: reshape update preparation : OK\n");
9845 return update_memory_size
;
9848 static void imsm_update_metadata_locally(struct supertype
*st
,
9851 struct metadata_update mu
;
9856 mu
.space_list
= NULL
;
9858 if (imsm_prepare_update(st
, &mu
))
9859 imsm_process_update(st
, &mu
);
9861 while (mu
.space_list
) {
9862 void **space
= mu
.space_list
;
9863 mu
.space_list
= *space
;
9868 /***************************************************************************
9869 * Function: imsm_analyze_change
9870 * Description: Function analyze change for single volume
9871 * and validate if transition is supported
9872 * Parameters: Geometry parameters, supertype structure,
9873 * metadata change direction (apply/rollback)
9874 * Returns: Operation type code on success, -1 if fail
9875 ****************************************************************************/
9876 enum imsm_reshape_type
imsm_analyze_change(struct supertype
*st
,
9877 struct geo_params
*geo
,
9884 /* number of added/removed disks in operation result */
9885 int devNumChange
= 0;
9886 /* imsm compatible layout value for array geometry verification */
9887 int imsm_layout
= -1;
9889 struct imsm_dev
*dev
;
9890 struct intel_super
*super
;
9891 unsigned long long current_size
;
9892 unsigned long long free_size
;
9893 unsigned long long max_size
;
9896 getinfo_super_imsm_volume(st
, &info
, NULL
);
9897 if ((geo
->level
!= info
.array
.level
) &&
9898 (geo
->level
>= 0) &&
9899 (geo
->level
!= UnSet
)) {
9900 switch (info
.array
.level
) {
9902 if (geo
->level
== 5) {
9903 change
= CH_MIGRATION
;
9904 if (geo
->layout
!= ALGORITHM_LEFT_ASYMMETRIC
) {
9905 pr_err("Error. Requested Layout not supported (left-asymmetric layout is supported only)!\n");
9907 goto analyse_change_exit
;
9909 imsm_layout
= geo
->layout
;
9911 devNumChange
= 1; /* parity disk added */
9912 } else if (geo
->level
== 10) {
9913 change
= CH_TAKEOVER
;
9915 devNumChange
= 2; /* two mirrors added */
9916 imsm_layout
= 0x102; /* imsm supported layout */
9921 if (geo
->level
== 0) {
9922 change
= CH_TAKEOVER
;
9924 devNumChange
= -(geo
->raid_disks
/2);
9925 imsm_layout
= 0; /* imsm raid0 layout */
9930 pr_err("Error. Level Migration from %d to %d not supported!\n",
9931 info
.array
.level
, geo
->level
);
9932 goto analyse_change_exit
;
9935 geo
->level
= info
.array
.level
;
9937 if ((geo
->layout
!= info
.array
.layout
)
9938 && ((geo
->layout
!= UnSet
) && (geo
->layout
!= -1))) {
9939 change
= CH_MIGRATION
;
9940 if ((info
.array
.layout
== 0)
9941 && (info
.array
.level
== 5)
9942 && (geo
->layout
== 5)) {
9943 /* reshape 5 -> 4 */
9944 } else if ((info
.array
.layout
== 5)
9945 && (info
.array
.level
== 5)
9946 && (geo
->layout
== 0)) {
9947 /* reshape 4 -> 5 */
9951 pr_err("Error. Layout Migration from %d to %d not supported!\n",
9952 info
.array
.layout
, geo
->layout
);
9954 goto analyse_change_exit
;
9957 geo
->layout
= info
.array
.layout
;
9958 if (imsm_layout
== -1)
9959 imsm_layout
= info
.array
.layout
;
9962 if ((geo
->chunksize
> 0) && (geo
->chunksize
!= UnSet
)
9963 && (geo
->chunksize
!= info
.array
.chunk_size
))
9964 change
= CH_MIGRATION
;
9966 geo
->chunksize
= info
.array
.chunk_size
;
9968 chunk
= geo
->chunksize
/ 1024;
9971 dev
= get_imsm_dev(super
, super
->current_vol
);
9972 data_disks
= imsm_num_data_members(dev
, MAP_0
);
9973 /* compute current size per disk member
9975 current_size
= info
.custom_array_size
/ data_disks
;
9977 if ((geo
->size
> 0) && (geo
->size
!= MAX_SIZE
)) {
9978 /* align component size
9980 geo
->size
= imsm_component_size_aligment_check(
9981 get_imsm_raid_level(dev
->vol
.map
),
9984 if (geo
->size
== 0) {
9985 pr_err("Error. Size expansion is supported only (current size is %llu, requested size /rounded/ is 0).\n",
9987 goto analyse_change_exit
;
9991 if ((current_size
!= geo
->size
) && (geo
->size
> 0)) {
9993 pr_err("Error. Size change should be the only one at a time.\n");
9995 goto analyse_change_exit
;
9997 if ((super
->current_vol
+ 1) != super
->anchor
->num_raid_devs
) {
9998 pr_err("Error. The last volume in container can be expanded only (%i/%s).\n",
9999 super
->current_vol
, st
->devnm
);
10000 goto analyse_change_exit
;
10002 /* check the maximum available size
10004 rv
= imsm_get_free_size(st
, dev
->vol
.map
->num_members
,
10005 0, chunk
, &free_size
);
10007 /* Cannot find maximum available space
10011 max_size
= free_size
+ current_size
;
10012 /* align component size
10014 max_size
= imsm_component_size_aligment_check(
10015 get_imsm_raid_level(dev
->vol
.map
),
10019 if (geo
->size
== MAX_SIZE
) {
10020 /* requested size change to the maximum available size
10022 if (max_size
== 0) {
10023 pr_err("Error. Cannot find maximum available space.\n");
10025 goto analyse_change_exit
;
10027 geo
->size
= max_size
;
10030 if ((direction
== ROLLBACK_METADATA_CHANGES
)) {
10031 /* accept size for rollback only
10034 /* round size due to metadata compatibility
10036 geo
->size
= (geo
->size
>> SECT_PER_MB_SHIFT
)
10037 << SECT_PER_MB_SHIFT
;
10038 dprintf("Prepare update for size change to %llu\n",
10040 if (current_size
>= geo
->size
) {
10041 pr_err("Error. Size expansion is supported only (current size is %llu, requested size /rounded/ is %llu).\n",
10042 current_size
, geo
->size
);
10043 goto analyse_change_exit
;
10045 if (max_size
&& geo
->size
> max_size
) {
10046 pr_err("Error. Requested size is larger than maximum available size (maximum available size is %llu, requested size /rounded/ is %llu).\n",
10047 max_size
, geo
->size
);
10048 goto analyse_change_exit
;
10051 geo
->size
*= data_disks
;
10052 geo
->raid_disks
= dev
->vol
.map
->num_members
;
10053 change
= CH_ARRAY_SIZE
;
10055 if (!validate_geometry_imsm(st
,
10058 geo
->raid_disks
+ devNumChange
,
10060 geo
->size
, INVALID_SECTORS
,
10065 struct intel_super
*super
= st
->sb
;
10066 struct imsm_super
*mpb
= super
->anchor
;
10068 if (mpb
->num_raid_devs
> 1) {
10069 pr_err("Error. Cannot perform operation on %s- for this operation it MUST be single array in container\n",
10075 analyse_change_exit
:
10076 if ((direction
== ROLLBACK_METADATA_CHANGES
) &&
10077 ((change
== CH_MIGRATION
) || (change
== CH_TAKEOVER
))) {
10078 dprintf("imsm: Metadata changes rollback is not supported for migration and takeover operations.\n");
10084 int imsm_takeover(struct supertype
*st
, struct geo_params
*geo
)
10086 struct intel_super
*super
= st
->sb
;
10087 struct imsm_update_takeover
*u
;
10089 u
= xmalloc(sizeof(struct imsm_update_takeover
));
10091 u
->type
= update_takeover
;
10092 u
->subarray
= super
->current_vol
;
10094 /* 10->0 transition */
10095 if (geo
->level
== 0)
10096 u
->direction
= R10_TO_R0
;
10098 /* 0->10 transition */
10099 if (geo
->level
== 10)
10100 u
->direction
= R0_TO_R10
;
10102 /* update metadata locally */
10103 imsm_update_metadata_locally(st
, u
,
10104 sizeof(struct imsm_update_takeover
));
10105 /* and possibly remotely */
10106 if (st
->update_tail
)
10107 append_metadata_update(st
, u
,
10108 sizeof(struct imsm_update_takeover
));
10115 static int imsm_reshape_super(struct supertype
*st
, unsigned long long size
,
10117 int layout
, int chunksize
, int raid_disks
,
10118 int delta_disks
, char *backup
, char *dev
,
10119 int direction
, int verbose
)
10122 struct geo_params geo
;
10124 dprintf("(enter)\n");
10126 memset(&geo
, 0, sizeof(struct geo_params
));
10128 geo
.dev_name
= dev
;
10129 strcpy(geo
.devnm
, st
->devnm
);
10132 geo
.layout
= layout
;
10133 geo
.chunksize
= chunksize
;
10134 geo
.raid_disks
= raid_disks
;
10135 if (delta_disks
!= UnSet
)
10136 geo
.raid_disks
+= delta_disks
;
10138 dprintf("for level : %i\n", geo
.level
);
10139 dprintf("for raid_disks : %i\n", geo
.raid_disks
);
10141 if (experimental() == 0)
10144 if (strcmp(st
->container_devnm
, st
->devnm
) == 0) {
10145 /* On container level we can only increase number of devices. */
10146 dprintf("imsm: info: Container operation\n");
10147 int old_raid_disks
= 0;
10149 if (imsm_reshape_is_allowed_on_container(
10150 st
, &geo
, &old_raid_disks
, direction
)) {
10151 struct imsm_update_reshape
*u
= NULL
;
10154 len
= imsm_create_metadata_update_for_reshape(
10155 st
, &geo
, old_raid_disks
, &u
);
10158 dprintf("imsm: Cannot prepare update\n");
10159 goto exit_imsm_reshape_super
;
10163 /* update metadata locally */
10164 imsm_update_metadata_locally(st
, u
, len
);
10165 /* and possibly remotely */
10166 if (st
->update_tail
)
10167 append_metadata_update(st
, u
, len
);
10172 pr_err("(imsm) Operation is not allowed on this container\n");
10175 /* On volume level we support following operations
10176 * - takeover: raid10 -> raid0; raid0 -> raid10
10177 * - chunk size migration
10178 * - migration: raid5 -> raid0; raid0 -> raid5
10180 struct intel_super
*super
= st
->sb
;
10181 struct intel_dev
*dev
= super
->devlist
;
10183 dprintf("imsm: info: Volume operation\n");
10184 /* find requested device */
10187 imsm_find_array_devnm_by_subdev(
10188 dev
->index
, st
->container_devnm
);
10189 if (devnm
&& strcmp(devnm
, geo
.devnm
) == 0)
10194 pr_err("Cannot find %s (%s) subarray\n",
10195 geo
.dev_name
, geo
.devnm
);
10196 goto exit_imsm_reshape_super
;
10198 super
->current_vol
= dev
->index
;
10199 change
= imsm_analyze_change(st
, &geo
, direction
);
10202 ret_val
= imsm_takeover(st
, &geo
);
10204 case CH_MIGRATION
: {
10205 struct imsm_update_reshape_migration
*u
= NULL
;
10207 imsm_create_metadata_update_for_migration(
10210 dprintf("imsm: Cannot prepare update\n");
10214 /* update metadata locally */
10215 imsm_update_metadata_locally(st
, u
, len
);
10216 /* and possibly remotely */
10217 if (st
->update_tail
)
10218 append_metadata_update(st
, u
, len
);
10223 case CH_ARRAY_SIZE
: {
10224 struct imsm_update_size_change
*u
= NULL
;
10226 imsm_create_metadata_update_for_size_change(
10229 dprintf("imsm: Cannot prepare update\n");
10233 /* update metadata locally */
10234 imsm_update_metadata_locally(st
, u
, len
);
10235 /* and possibly remotely */
10236 if (st
->update_tail
)
10237 append_metadata_update(st
, u
, len
);
10247 exit_imsm_reshape_super
:
10248 dprintf("imsm: reshape_super Exit code = %i\n", ret_val
);
10252 /*******************************************************************************
10253 * Function: wait_for_reshape_imsm
10254 * Description: Function writes new sync_max value and waits until
10255 * reshape process reach new position
10257 * sra : general array info
10258 * ndata : number of disks in new array's layout
10261 * 1 : there is no reshape in progress,
10263 ******************************************************************************/
10264 int wait_for_reshape_imsm(struct mdinfo
*sra
, int ndata
)
10266 int fd
= sysfs_get_fd(sra
, NULL
, "sync_completed");
10267 unsigned long long completed
;
10268 /* to_complete : new sync_max position */
10269 unsigned long long to_complete
= sra
->reshape_progress
;
10270 unsigned long long position_to_set
= to_complete
/ ndata
;
10273 dprintf("cannot open reshape_position\n");
10277 if (sysfs_fd_get_ll(fd
, &completed
) < 0) {
10278 dprintf("cannot read reshape_position (no reshape in progres)\n");
10283 if (completed
> position_to_set
) {
10284 dprintf("wrong next position to set %llu (%llu)\n",
10285 to_complete
, position_to_set
);
10289 dprintf("Position set: %llu\n", position_to_set
);
10290 if (sysfs_set_num(sra
, NULL
, "sync_max",
10291 position_to_set
) != 0) {
10292 dprintf("cannot set reshape position to %llu\n",
10300 sysfs_wait(fd
, NULL
);
10301 if (sysfs_get_str(sra
, NULL
, "sync_action",
10303 strncmp(action
, "reshape", 7) != 0)
10305 if (sysfs_fd_get_ll(fd
, &completed
) < 0) {
10306 dprintf("cannot read reshape_position (in loop)\n");
10310 } while (completed
< position_to_set
);
10316 /*******************************************************************************
10317 * Function: check_degradation_change
10318 * Description: Check that array hasn't become failed.
10320 * info : for sysfs access
10321 * sources : source disks descriptors
10322 * degraded: previous degradation level
10324 * degradation level
10325 ******************************************************************************/
10326 int check_degradation_change(struct mdinfo
*info
,
10330 unsigned long long new_degraded
;
10333 rv
= sysfs_get_ll(info
, NULL
, "degraded", &new_degraded
);
10334 if ((rv
== -1) || (new_degraded
!= (unsigned long long)degraded
)) {
10335 /* check each device to ensure it is still working */
10338 for (sd
= info
->devs
; sd
; sd
= sd
->next
) {
10339 if (sd
->disk
.state
& (1<<MD_DISK_FAULTY
))
10341 if (sd
->disk
.state
& (1<<MD_DISK_SYNC
)) {
10343 if (sysfs_get_str(info
,
10344 sd
, "state", sbuf
, 20) < 0 ||
10345 strstr(sbuf
, "faulty") ||
10346 strstr(sbuf
, "in_sync") == NULL
) {
10347 /* this device is dead */
10348 sd
->disk
.state
= (1<<MD_DISK_FAULTY
);
10349 if (sd
->disk
.raid_disk
>= 0 &&
10350 sources
[sd
->disk
.raid_disk
] >= 0) {
10352 sd
->disk
.raid_disk
]);
10353 sources
[sd
->disk
.raid_disk
] =
10362 return new_degraded
;
10365 /*******************************************************************************
10366 * Function: imsm_manage_reshape
10367 * Description: Function finds array under reshape and it manages reshape
10368 * process. It creates stripes backups (if required) and sets
10371 * afd : Backup handle (nattive) - not used
10372 * sra : general array info
10373 * reshape : reshape parameters - not used
10374 * st : supertype structure
10375 * blocks : size of critical section [blocks]
10376 * fds : table of source device descriptor
10377 * offsets : start of array (offest per devices)
10379 * destfd : table of destination device descriptor
10380 * destoffsets : table of destination offsets (per device)
10382 * 1 : success, reshape is done
10384 ******************************************************************************/
10385 static int imsm_manage_reshape(
10386 int afd
, struct mdinfo
*sra
, struct reshape
*reshape
,
10387 struct supertype
*st
, unsigned long backup_blocks
,
10388 int *fds
, unsigned long long *offsets
,
10389 int dests
, int *destfd
, unsigned long long *destoffsets
)
10392 struct intel_super
*super
= st
->sb
;
10393 struct intel_dev
*dv
= NULL
;
10394 struct imsm_dev
*dev
= NULL
;
10395 struct imsm_map
*map_src
;
10396 int migr_vol_qan
= 0;
10397 int ndata
, odata
; /* [bytes] */
10398 int chunk
; /* [bytes] */
10399 struct migr_record
*migr_rec
;
10401 unsigned int buf_size
; /* [bytes] */
10402 unsigned long long max_position
; /* array size [bytes] */
10403 unsigned long long next_step
; /* [blocks]/[bytes] */
10404 unsigned long long old_data_stripe_length
;
10405 unsigned long long start_src
; /* [bytes] */
10406 unsigned long long start
; /* [bytes] */
10407 unsigned long long start_buf_shift
; /* [bytes] */
10409 int source_layout
= 0;
10411 if (!fds
|| !offsets
|| !sra
)
10414 /* Find volume during the reshape */
10415 for (dv
= super
->devlist
; dv
; dv
= dv
->next
) {
10416 if (dv
->dev
->vol
.migr_type
== MIGR_GEN_MIGR
10417 && dv
->dev
->vol
.migr_state
== 1) {
10422 /* Only one volume can migrate at the same time */
10423 if (migr_vol_qan
!= 1) {
10424 pr_err(": %s", migr_vol_qan
?
10425 "Number of migrating volumes greater than 1\n" :
10426 "There is no volume during migrationg\n");
10430 map_src
= get_imsm_map(dev
, MAP_1
);
10431 if (map_src
== NULL
)
10434 ndata
= imsm_num_data_members(dev
, MAP_0
);
10435 odata
= imsm_num_data_members(dev
, MAP_1
);
10437 chunk
= __le16_to_cpu(map_src
->blocks_per_strip
) * 512;
10438 old_data_stripe_length
= odata
* chunk
;
10440 migr_rec
= super
->migr_rec
;
10442 /* initialize migration record for start condition */
10443 if (sra
->reshape_progress
== 0)
10444 init_migr_record_imsm(st
, dev
, sra
);
10446 if (__le32_to_cpu(migr_rec
->rec_status
) != UNIT_SRC_NORMAL
) {
10447 dprintf("imsm: cannot restart migration when data are present in copy area.\n");
10450 /* Save checkpoint to update migration record for current
10451 * reshape position (in md). It can be farther than current
10452 * reshape position in metadata.
10454 if (save_checkpoint_imsm(st
, sra
, UNIT_SRC_NORMAL
) == 1) {
10455 /* ignore error == 2, this can mean end of reshape here
10457 dprintf("imsm: Cannot write checkpoint to migration record (UNIT_SRC_NORMAL, initial save)\n");
10462 /* size for data */
10463 buf_size
= __le32_to_cpu(migr_rec
->blocks_per_unit
) * 512;
10464 /* extend buffer size for parity disk */
10465 buf_size
+= __le32_to_cpu(migr_rec
->dest_depth_per_unit
) * 512;
10466 /* add space for stripe aligment */
10467 buf_size
+= old_data_stripe_length
;
10468 if (posix_memalign((void **)&buf
, 4096, buf_size
)) {
10469 dprintf("imsm: Cannot allocate checpoint buffer\n");
10473 max_position
= sra
->component_size
* ndata
;
10474 source_layout
= imsm_level_to_layout(map_src
->raid_level
);
10476 while (__le32_to_cpu(migr_rec
->curr_migr_unit
) <
10477 __le32_to_cpu(migr_rec
->num_migr_units
)) {
10478 /* current reshape position [blocks] */
10479 unsigned long long current_position
=
10480 __le32_to_cpu(migr_rec
->blocks_per_unit
)
10481 * __le32_to_cpu(migr_rec
->curr_migr_unit
);
10482 unsigned long long border
;
10484 /* Check that array hasn't become failed.
10486 degraded
= check_degradation_change(sra
, fds
, degraded
);
10487 if (degraded
> 1) {
10488 dprintf("imsm: Abort reshape due to degradation level (%i)\n", degraded
);
10492 next_step
= __le32_to_cpu(migr_rec
->blocks_per_unit
);
10494 if ((current_position
+ next_step
) > max_position
)
10495 next_step
= max_position
- current_position
;
10497 start
= current_position
* 512;
10499 /* allign reading start to old geometry */
10500 start_buf_shift
= start
% old_data_stripe_length
;
10501 start_src
= start
- start_buf_shift
;
10503 border
= (start_src
/ odata
) - (start
/ ndata
);
10505 if (border
<= __le32_to_cpu(migr_rec
->dest_depth_per_unit
)) {
10506 /* save critical stripes to buf
10507 * start - start address of current unit
10508 * to backup [bytes]
10509 * start_src - start address of current unit
10510 * to backup alligned to source array
10513 unsigned long long next_step_filler
= 0;
10514 unsigned long long copy_length
= next_step
* 512;
10516 /* allign copy area length to stripe in old geometry */
10517 next_step_filler
= ((copy_length
+ start_buf_shift
)
10518 % old_data_stripe_length
);
10519 if (next_step_filler
)
10520 next_step_filler
= (old_data_stripe_length
10521 - next_step_filler
);
10522 dprintf("save_stripes() parameters: start = %llu,\tstart_src = %llu,\tnext_step*512 = %llu,\tstart_in_buf_shift = %llu,\tnext_step_filler = %llu\n",
10523 start
, start_src
, copy_length
,
10524 start_buf_shift
, next_step_filler
);
10526 if (save_stripes(fds
, offsets
, map_src
->num_members
,
10527 chunk
, map_src
->raid_level
,
10528 source_layout
, 0, NULL
, start_src
,
10530 next_step_filler
+ start_buf_shift
,
10532 dprintf("imsm: Cannot save stripes to buffer\n");
10535 /* Convert data to destination format and store it
10536 * in backup general migration area
10538 if (save_backup_imsm(st
, dev
, sra
,
10539 buf
+ start_buf_shift
, copy_length
)) {
10540 dprintf("imsm: Cannot save stripes to target devices\n");
10543 if (save_checkpoint_imsm(st
, sra
,
10544 UNIT_SRC_IN_CP_AREA
)) {
10545 dprintf("imsm: Cannot write checkpoint to migration record (UNIT_SRC_IN_CP_AREA)\n");
10549 /* set next step to use whole border area */
10550 border
/= next_step
;
10552 next_step
*= border
;
10554 /* When data backed up, checkpoint stored,
10555 * kick the kernel to reshape unit of data
10557 next_step
= next_step
+ sra
->reshape_progress
;
10558 /* limit next step to array max position */
10559 if (next_step
> max_position
)
10560 next_step
= max_position
;
10561 sysfs_set_num(sra
, NULL
, "suspend_lo", sra
->reshape_progress
);
10562 sysfs_set_num(sra
, NULL
, "suspend_hi", next_step
);
10563 sra
->reshape_progress
= next_step
;
10565 /* wait until reshape finish */
10566 if (wait_for_reshape_imsm(sra
, ndata
) < 0) {
10567 dprintf("wait_for_reshape_imsm returned error!\n");
10573 if (save_checkpoint_imsm(st
, sra
, UNIT_SRC_NORMAL
) == 1) {
10574 /* ignore error == 2, this can mean end of reshape here
10576 dprintf("imsm: Cannot write checkpoint to migration record (UNIT_SRC_NORMAL)\n");
10582 /* clear migr_rec on disks after successful migration */
10585 memset(super
->migr_rec_buf
, 0, MIGR_REC_BUF_SIZE
);
10586 for (d
= super
->disks
; d
; d
= d
->next
) {
10587 if (d
->index
< 0 || is_failed(&d
->disk
))
10589 unsigned long long dsize
;
10591 get_dev_size(d
->fd
, NULL
, &dsize
);
10592 if (lseek64(d
->fd
, dsize
- MIGR_REC_POSITION
,
10594 if (write(d
->fd
, super
->migr_rec_buf
,
10595 MIGR_REC_BUF_SIZE
) != MIGR_REC_BUF_SIZE
)
10596 perror("Write migr_rec failed");
10600 /* return '1' if done */
10604 abort_reshape(sra
);
10609 #endif /* MDASSEMBLE */
10611 struct superswitch super_imsm
= {
10613 .examine_super
= examine_super_imsm
,
10614 .brief_examine_super
= brief_examine_super_imsm
,
10615 .brief_examine_subarrays
= brief_examine_subarrays_imsm
,
10616 .export_examine_super
= export_examine_super_imsm
,
10617 .detail_super
= detail_super_imsm
,
10618 .brief_detail_super
= brief_detail_super_imsm
,
10619 .write_init_super
= write_init_super_imsm
,
10620 .validate_geometry
= validate_geometry_imsm
,
10621 .add_to_super
= add_to_super_imsm
,
10622 .remove_from_super
= remove_from_super_imsm
,
10623 .detail_platform
= detail_platform_imsm
,
10624 .export_detail_platform
= export_detail_platform_imsm
,
10625 .kill_subarray
= kill_subarray_imsm
,
10626 .update_subarray
= update_subarray_imsm
,
10627 .load_container
= load_container_imsm
,
10628 .default_geometry
= default_geometry_imsm
,
10629 .get_disk_controller_domain
= imsm_get_disk_controller_domain
,
10630 .reshape_super
= imsm_reshape_super
,
10631 .manage_reshape
= imsm_manage_reshape
,
10632 .recover_backup
= recover_backup_imsm
,
10633 .copy_metadata
= copy_metadata_imsm
,
10635 .match_home
= match_home_imsm
,
10636 .uuid_from_super
= uuid_from_super_imsm
,
10637 .getinfo_super
= getinfo_super_imsm
,
10638 .getinfo_super_disks
= getinfo_super_disks_imsm
,
10639 .update_super
= update_super_imsm
,
10641 .avail_size
= avail_size_imsm
,
10642 .min_acceptable_spare_size
= min_acceptable_spare_size_imsm
,
10644 .compare_super
= compare_super_imsm
,
10646 .load_super
= load_super_imsm
,
10647 .init_super
= init_super_imsm
,
10648 .store_super
= store_super_imsm
,
10649 .free_super
= free_super_imsm
,
10650 .match_metadata_desc
= match_metadata_desc_imsm
,
10651 .container_content
= container_content_imsm
,
10652 .validate_container
= validate_container_imsm
,
10659 .open_new
= imsm_open_new
,
10660 .set_array_state
= imsm_set_array_state
,
10661 .set_disk
= imsm_set_disk
,
10662 .sync_metadata
= imsm_sync_metadata
,
10663 .activate_spare
= imsm_activate_spare
,
10664 .process_update
= imsm_process_update
,
10665 .prepare_update
= imsm_prepare_update
,
10666 #endif /* MDASSEMBLE */