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__
));
240 static char *map_state_str
[] = { "normal", "uninitialized", "degraded", "failed" };
243 #define RAID_DISK_RESERVED_BLOCKS_IMSM_HI 2209
245 #define GEN_MIGR_AREA_SIZE 2048 /* General Migration Copy Area size in blocks */
247 #define MIGR_REC_BUF_SIZE 512 /* size of migr_record i/o buffer */
248 #define MIGR_REC_POSITION 512 /* migr_record position offset on disk,
249 * MIGR_REC_BUF_SIZE <= MIGR_REC_POSITION
253 #define UNIT_SRC_NORMAL 0 /* Source data for curr_migr_unit must
254 * be recovered using srcMap */
255 #define UNIT_SRC_IN_CP_AREA 1 /* Source data for curr_migr_unit has
256 * already been migrated and must
257 * be recovered from checkpoint area */
259 __u32 rec_status
; /* Status used to determine how to restart
260 * migration in case it aborts
262 __u32 curr_migr_unit
; /* 0..numMigrUnits-1 */
263 __u32 family_num
; /* Family number of MPB
264 * containing the RaidDev
265 * that is migrating */
266 __u32 ascending_migr
; /* True if migrating in increasing
268 __u32 blocks_per_unit
; /* Num disk blocks per unit of operation */
269 __u32 dest_depth_per_unit
; /* Num member blocks each destMap
271 * advances per unit-of-operation */
272 __u32 ckpt_area_pba
; /* Pba of first block of ckpt copy area */
273 __u32 dest_1st_member_lba
; /* First member lba on first
274 * stripe of destination */
275 __u32 num_migr_units
; /* Total num migration units-of-op */
276 __u32 post_migr_vol_cap
; /* Size of volume after
277 * migration completes */
278 __u32 post_migr_vol_cap_hi
; /* Expansion space for LBA64 */
279 __u32 ckpt_read_disk_num
; /* Which member disk in destSubMap[0] the
280 * migration ckpt record was read from
281 * (for recovered migrations) */
282 } __attribute__ ((__packed__
));
287 * 2: metadata does not match
295 struct md_list
*next
;
298 #define pr_vrb(fmt, arg...) (void) (verbose && fprintf(stderr, Name fmt, ##arg))
300 static __u8
migr_type(struct imsm_dev
*dev
)
302 if (dev
->vol
.migr_type
== MIGR_VERIFY
&&
303 dev
->status
& DEV_VERIFY_AND_FIX
)
306 return dev
->vol
.migr_type
;
309 static void set_migr_type(struct imsm_dev
*dev
, __u8 migr_type
)
311 /* for compatibility with older oroms convert MIGR_REPAIR, into
312 * MIGR_VERIFY w/ DEV_VERIFY_AND_FIX status
314 if (migr_type
== MIGR_REPAIR
) {
315 dev
->vol
.migr_type
= MIGR_VERIFY
;
316 dev
->status
|= DEV_VERIFY_AND_FIX
;
318 dev
->vol
.migr_type
= migr_type
;
319 dev
->status
&= ~DEV_VERIFY_AND_FIX
;
323 static unsigned int sector_count(__u32 bytes
)
325 return ROUND_UP(bytes
, 512) / 512;
328 static unsigned int mpb_sectors(struct imsm_super
*mpb
)
330 return sector_count(__le32_to_cpu(mpb
->mpb_size
));
334 struct imsm_dev
*dev
;
335 struct intel_dev
*next
;
340 enum sys_dev_type type
;
343 struct intel_hba
*next
;
350 /* internal representation of IMSM metadata */
353 void *buf
; /* O_DIRECT buffer for reading/writing metadata */
354 struct imsm_super
*anchor
; /* immovable parameters */
357 void *migr_rec_buf
; /* buffer for I/O operations */
358 struct migr_record
*migr_rec
; /* migration record */
360 int clean_migration_record_by_mdmon
; /* when reshape is switched to next
361 array, it indicates that mdmon is allowed to clean migration
363 size_t len
; /* size of the 'buf' allocation */
364 void *next_buf
; /* for realloc'ing buf from the manager */
366 int updates_pending
; /* count of pending updates for mdmon */
367 int current_vol
; /* index of raid device undergoing creation */
368 unsigned long long create_offset
; /* common start for 'current_vol' */
369 __u32 random
; /* random data for seeding new family numbers */
370 struct intel_dev
*devlist
;
374 __u8 serial
[MAX_RAID_SERIAL_LEN
];
377 struct imsm_disk disk
;
380 struct extent
*e
; /* for determining freespace @ create */
381 int raiddisk
; /* slot to fill in autolayout */
383 } *disks
, *current_disk
;
384 struct dl
*disk_mgmt_list
; /* list of disks to add/remove while mdmon
386 struct dl
*missing
; /* disks removed while we weren't looking */
387 struct bbm_log
*bbm_log
;
388 struct intel_hba
*hba
; /* device path of the raid controller for this metadata */
389 const struct imsm_orom
*orom
; /* platform firmware support */
390 struct intel_super
*next
; /* (temp) list for disambiguating family_num */
394 struct imsm_disk disk
;
395 #define IMSM_UNKNOWN_OWNER (-1)
397 struct intel_disk
*next
;
401 unsigned long long start
, size
;
404 /* definitions of reshape process types */
405 enum imsm_reshape_type
{
411 /* definition of messages passed to imsm_process_update */
412 enum imsm_update_type
{
413 update_activate_spare
,
417 update_add_remove_disk
,
418 update_reshape_container_disks
,
419 update_reshape_migration
,
421 update_general_migration_checkpoint
,
425 struct imsm_update_activate_spare
{
426 enum imsm_update_type type
;
430 struct imsm_update_activate_spare
*next
;
443 enum takeover_direction
{
447 struct imsm_update_takeover
{
448 enum imsm_update_type type
;
450 enum takeover_direction direction
;
453 struct imsm_update_reshape
{
454 enum imsm_update_type type
;
458 int new_disks
[1]; /* new_raid_disks - old_raid_disks makedev number */
461 struct imsm_update_reshape_migration
{
462 enum imsm_update_type type
;
465 /* fields for array migration changes
472 int new_disks
[1]; /* new_raid_disks - old_raid_disks makedev number */
475 struct imsm_update_size_change
{
476 enum imsm_update_type type
;
481 struct imsm_update_general_migration_checkpoint
{
482 enum imsm_update_type type
;
483 __u32 curr_migr_unit
;
487 __u8 serial
[MAX_RAID_SERIAL_LEN
];
490 struct imsm_update_create_array
{
491 enum imsm_update_type type
;
496 struct imsm_update_kill_array
{
497 enum imsm_update_type type
;
501 struct imsm_update_rename_array
{
502 enum imsm_update_type type
;
503 __u8 name
[MAX_RAID_SERIAL_LEN
];
507 struct imsm_update_add_remove_disk
{
508 enum imsm_update_type type
;
512 static const char *_sys_dev_type
[] = {
513 [SYS_DEV_UNKNOWN
] = "Unknown",
514 [SYS_DEV_SAS
] = "SAS",
515 [SYS_DEV_SATA
] = "SATA"
518 const char *get_sys_dev_type(enum sys_dev_type type
)
520 if (type
>= SYS_DEV_MAX
)
521 type
= SYS_DEV_UNKNOWN
;
523 return _sys_dev_type
[type
];
526 static struct intel_hba
* alloc_intel_hba(struct sys_dev
*device
)
528 struct intel_hba
*result
= malloc(sizeof(*result
));
530 result
->type
= device
->type
;
531 result
->path
= strdup(device
->path
);
533 if (result
->path
&& (result
->pci_id
= strrchr(result
->path
, '/')) != NULL
)
539 static struct intel_hba
* find_intel_hba(struct intel_hba
*hba
, struct sys_dev
*device
)
541 struct intel_hba
*result
=NULL
;
542 for (result
= hba
; result
; result
= result
->next
) {
543 if (result
->type
== device
->type
&& strcmp(result
->path
, device
->path
) == 0)
549 static int attach_hba_to_super(struct intel_super
*super
, struct sys_dev
*device
)
551 struct intel_hba
*hba
;
553 /* check if disk attached to Intel HBA */
554 hba
= find_intel_hba(super
->hba
, device
);
557 /* Check if HBA is already attached to super */
558 if (super
->hba
== NULL
) {
559 super
->hba
= alloc_intel_hba(device
);
564 /* Intel metadata allows for all disks attached to the same type HBA.
565 * Do not sypport odf HBA types mixing
567 if (device
->type
!= hba
->type
)
573 hba
->next
= alloc_intel_hba(device
);
577 static struct sys_dev
* find_disk_attached_hba(int fd
, const char *devname
)
579 struct sys_dev
*list
, *elem
, *prev
;
582 if ((list
= find_intel_devices()) == NULL
)
586 disk_path
= (char *) devname
;
588 disk_path
= diskfd_to_devpath(fd
);
595 for (prev
= NULL
, elem
= list
; elem
; prev
= elem
, elem
= elem
->next
) {
596 if (path_attached_to_hba(disk_path
, elem
->path
)) {
600 prev
->next
= elem
->next
;
602 if (disk_path
!= devname
)
608 if (disk_path
!= devname
)
616 static int find_intel_hba_capability(int fd
, struct intel_super
*super
,
619 static struct supertype
*match_metadata_desc_imsm(char *arg
)
621 struct supertype
*st
;
623 if (strcmp(arg
, "imsm") != 0 &&
624 strcmp(arg
, "default") != 0
628 st
= malloc(sizeof(*st
));
631 memset(st
, 0, sizeof(*st
));
632 st
->container_dev
= NoMdDev
;
633 st
->ss
= &super_imsm
;
634 st
->max_devs
= IMSM_MAX_DEVICES
;
635 st
->minor_version
= 0;
641 static __u8
*get_imsm_version(struct imsm_super
*mpb
)
643 return &mpb
->sig
[MPB_SIG_LEN
];
647 /* retrieve a disk directly from the anchor when the anchor is known to be
648 * up-to-date, currently only at load time
650 static struct imsm_disk
*__get_imsm_disk(struct imsm_super
*mpb
, __u8 index
)
652 if (index
>= mpb
->num_disks
)
654 return &mpb
->disk
[index
];
657 /* retrieve the disk description based on a index of the disk
660 static struct dl
*get_imsm_dl_disk(struct intel_super
*super
, __u8 index
)
664 for (d
= super
->disks
; d
; d
= d
->next
)
665 if (d
->index
== index
)
670 /* retrieve a disk from the parsed metadata */
671 static struct imsm_disk
*get_imsm_disk(struct intel_super
*super
, __u8 index
)
675 dl
= get_imsm_dl_disk(super
, index
);
682 /* generate a checksum directly from the anchor when the anchor is known to be
683 * up-to-date, currently only at load or write_super after coalescing
685 static __u32
__gen_imsm_checksum(struct imsm_super
*mpb
)
687 __u32 end
= mpb
->mpb_size
/ sizeof(end
);
688 __u32
*p
= (__u32
*) mpb
;
692 sum
+= __le32_to_cpu(*p
);
696 return sum
- __le32_to_cpu(mpb
->check_sum
);
699 static size_t sizeof_imsm_map(struct imsm_map
*map
)
701 return sizeof(struct imsm_map
) + sizeof(__u32
) * (map
->num_members
- 1);
704 struct imsm_map
*get_imsm_map(struct imsm_dev
*dev
, int second_map
)
706 /* A device can have 2 maps if it is in the middle of a migration.
708 * MAP_0 - we return the first map
709 * MAP_1 - we return the second map if it exists, else NULL
710 * MAP_X - we return the second map if it exists, else the first
712 struct imsm_map
*map
= &dev
->vol
.map
[0];
713 struct imsm_map
*map2
= NULL
;
715 if (dev
->vol
.migr_state
)
716 map2
= (void *)map
+ sizeof_imsm_map(map
);
718 switch (second_map
) {
735 /* return the size of the device.
736 * migr_state increases the returned size if map[0] were to be duplicated
738 static size_t sizeof_imsm_dev(struct imsm_dev
*dev
, int migr_state
)
740 size_t size
= sizeof(*dev
) - sizeof(struct imsm_map
) +
741 sizeof_imsm_map(get_imsm_map(dev
, MAP_0
));
743 /* migrating means an additional map */
744 if (dev
->vol
.migr_state
)
745 size
+= sizeof_imsm_map(get_imsm_map(dev
, MAP_1
));
747 size
+= sizeof_imsm_map(get_imsm_map(dev
, MAP_0
));
753 /* retrieve disk serial number list from a metadata update */
754 static struct disk_info
*get_disk_info(struct imsm_update_create_array
*update
)
757 struct disk_info
*inf
;
759 inf
= u
+ sizeof(*update
) - sizeof(struct imsm_dev
) +
760 sizeof_imsm_dev(&update
->dev
, 0);
766 static struct imsm_dev
*__get_imsm_dev(struct imsm_super
*mpb
, __u8 index
)
772 if (index
>= mpb
->num_raid_devs
)
775 /* devices start after all disks */
776 offset
= ((void *) &mpb
->disk
[mpb
->num_disks
]) - _mpb
;
778 for (i
= 0; i
<= index
; i
++)
780 return _mpb
+ offset
;
782 offset
+= sizeof_imsm_dev(_mpb
+ offset
, 0);
787 static struct imsm_dev
*get_imsm_dev(struct intel_super
*super
, __u8 index
)
789 struct intel_dev
*dv
;
791 if (index
>= super
->anchor
->num_raid_devs
)
793 for (dv
= super
->devlist
; dv
; dv
= dv
->next
)
794 if (dv
->index
== index
)
801 * == MAP_0 get first map
802 * == MAP_1 get second map
803 * == MAP_X than get map according to the current migr_state
805 static __u32
get_imsm_ord_tbl_ent(struct imsm_dev
*dev
,
809 struct imsm_map
*map
;
811 map
= get_imsm_map(dev
, second_map
);
813 /* top byte identifies disk under rebuild */
814 return __le32_to_cpu(map
->disk_ord_tbl
[slot
]);
817 #define ord_to_idx(ord) (((ord) << 8) >> 8)
818 static __u32
get_imsm_disk_idx(struct imsm_dev
*dev
, int slot
, int second_map
)
820 __u32 ord
= get_imsm_ord_tbl_ent(dev
, slot
, second_map
);
822 return ord_to_idx(ord
);
825 static void set_imsm_ord_tbl_ent(struct imsm_map
*map
, int slot
, __u32 ord
)
827 map
->disk_ord_tbl
[slot
] = __cpu_to_le32(ord
);
830 static int get_imsm_disk_slot(struct imsm_map
*map
, unsigned idx
)
835 for (slot
= 0; slot
< map
->num_members
; slot
++) {
836 ord
= __le32_to_cpu(map
->disk_ord_tbl
[slot
]);
837 if (ord_to_idx(ord
) == idx
)
844 static int get_imsm_raid_level(struct imsm_map
*map
)
846 if (map
->raid_level
== 1) {
847 if (map
->num_members
== 2)
853 return map
->raid_level
;
856 static int cmp_extent(const void *av
, const void *bv
)
858 const struct extent
*a
= av
;
859 const struct extent
*b
= bv
;
860 if (a
->start
< b
->start
)
862 if (a
->start
> b
->start
)
867 static int count_memberships(struct dl
*dl
, struct intel_super
*super
)
872 for (i
= 0; i
< super
->anchor
->num_raid_devs
; i
++) {
873 struct imsm_dev
*dev
= get_imsm_dev(super
, i
);
874 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
876 if (get_imsm_disk_slot(map
, dl
->index
) >= 0)
883 static __u32
imsm_min_reserved_sectors(struct intel_super
*super
);
885 static int split_ull(unsigned long long n
, __u32
*lo
, __u32
*hi
)
887 if (lo
== 0 || hi
== 0)
889 *lo
= __le32_to_cpu((unsigned)n
);
890 *hi
= __le32_to_cpu((unsigned)(n
>> 32));
894 static unsigned long long join_u32(__u32 lo
, __u32 hi
)
896 return (unsigned long long)__le32_to_cpu(lo
) |
897 (((unsigned long long)__le32_to_cpu(hi
)) << 32);
900 static unsigned long long total_blocks(struct imsm_disk
*disk
)
904 return join_u32(disk
->total_blocks_lo
, disk
->total_blocks_hi
);
907 static unsigned long long pba_of_lba0(struct imsm_map
*map
)
911 return join_u32(map
->pba_of_lba0_lo
, map
->pba_of_lba0_hi
);
914 static unsigned long long blocks_per_member(struct imsm_map
*map
)
918 return join_u32(map
->blocks_per_member_lo
, map
->blocks_per_member_hi
);
921 static unsigned long long num_data_stripes(struct imsm_map
*map
)
925 return join_u32(map
->num_data_stripes_lo
, map
->num_data_stripes_hi
);
928 static void set_total_blocks(struct imsm_disk
*disk
, unsigned long long n
)
930 split_ull(n
, &disk
->total_blocks_lo
, &disk
->total_blocks_hi
);
933 static void set_pba_of_lba0(struct imsm_map
*map
, unsigned long long n
)
935 split_ull(n
, &map
->pba_of_lba0_lo
, &map
->pba_of_lba0_hi
);
938 static void set_blocks_per_member(struct imsm_map
*map
, unsigned long long n
)
940 split_ull(n
, &map
->blocks_per_member_lo
, &map
->blocks_per_member_hi
);
943 static void set_num_data_stripes(struct imsm_map
*map
, unsigned long long n
)
945 split_ull(n
, &map
->num_data_stripes_lo
, &map
->num_data_stripes_hi
);
948 static struct extent
*get_extents(struct intel_super
*super
, struct dl
*dl
)
950 /* find a list of used extents on the given physical device */
951 struct extent
*rv
, *e
;
953 int memberships
= count_memberships(dl
, super
);
956 /* trim the reserved area for spares, so they can join any array
957 * regardless of whether the OROM has assigned sectors from the
958 * IMSM_RESERVED_SECTORS region
961 reservation
= imsm_min_reserved_sectors(super
);
963 reservation
= MPB_SECTOR_CNT
+ IMSM_RESERVED_SECTORS
;
965 rv
= malloc(sizeof(struct extent
) * (memberships
+ 1));
970 for (i
= 0; i
< super
->anchor
->num_raid_devs
; i
++) {
971 struct imsm_dev
*dev
= get_imsm_dev(super
, i
);
972 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
974 if (get_imsm_disk_slot(map
, dl
->index
) >= 0) {
975 e
->start
= pba_of_lba0(map
);
976 e
->size
= blocks_per_member(map
);
980 qsort(rv
, memberships
, sizeof(*rv
), cmp_extent
);
982 /* determine the start of the metadata
983 * when no raid devices are defined use the default
984 * ...otherwise allow the metadata to truncate the value
985 * as is the case with older versions of imsm
988 struct extent
*last
= &rv
[memberships
- 1];
989 unsigned long long remainder
;
991 remainder
= total_blocks(&dl
->disk
) - (last
->start
+ last
->size
);
992 /* round down to 1k block to satisfy precision of the kernel
996 /* make sure remainder is still sane */
997 if (remainder
< (unsigned)ROUND_UP(super
->len
, 512) >> 9)
998 remainder
= ROUND_UP(super
->len
, 512) >> 9;
999 if (reservation
> remainder
)
1000 reservation
= remainder
;
1002 e
->start
= total_blocks(&dl
->disk
) - reservation
;
1007 /* try to determine how much space is reserved for metadata from
1008 * the last get_extents() entry, otherwise fallback to the
1011 static __u32
imsm_reserved_sectors(struct intel_super
*super
, struct dl
*dl
)
1017 /* for spares just return a minimal reservation which will grow
1018 * once the spare is picked up by an array
1020 if (dl
->index
== -1)
1021 return MPB_SECTOR_CNT
;
1023 e
= get_extents(super
, dl
);
1025 return MPB_SECTOR_CNT
+ IMSM_RESERVED_SECTORS
;
1027 /* scroll to last entry */
1028 for (i
= 0; e
[i
].size
; i
++)
1031 rv
= total_blocks(&dl
->disk
) - e
[i
].start
;
1038 static int is_spare(struct imsm_disk
*disk
)
1040 return (disk
->status
& SPARE_DISK
) == SPARE_DISK
;
1043 static int is_configured(struct imsm_disk
*disk
)
1045 return (disk
->status
& CONFIGURED_DISK
) == CONFIGURED_DISK
;
1048 static int is_failed(struct imsm_disk
*disk
)
1050 return (disk
->status
& FAILED_DISK
) == FAILED_DISK
;
1053 /* try to determine how much space is reserved for metadata from
1054 * the last get_extents() entry on the smallest active disk,
1055 * otherwise fallback to the default
1057 static __u32
imsm_min_reserved_sectors(struct intel_super
*super
)
1061 unsigned long long min_active
;
1063 __u32 rv
= MPB_SECTOR_CNT
+ IMSM_RESERVED_SECTORS
;
1064 struct dl
*dl
, *dl_min
= NULL
;
1070 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
1073 unsigned long long blocks
= total_blocks(&dl
->disk
);
1074 if (blocks
< min_active
|| min_active
== 0) {
1076 min_active
= blocks
;
1082 /* find last lba used by subarrays on the smallest active disk */
1083 e
= get_extents(super
, dl_min
);
1086 for (i
= 0; e
[i
].size
; i
++)
1089 remainder
= min_active
- e
[i
].start
;
1092 /* to give priority to recovery we should not require full
1093 IMSM_RESERVED_SECTORS from the spare */
1094 rv
= MPB_SECTOR_CNT
+ NUM_BLOCKS_DIRTY_STRIPE_REGION
;
1096 /* if real reservation is smaller use that value */
1097 return (remainder
< rv
) ? remainder
: rv
;
1100 /* Return minimum size of a spare that can be used in this array*/
1101 static unsigned long long min_acceptable_spare_size_imsm(struct supertype
*st
)
1103 struct intel_super
*super
= st
->sb
;
1107 unsigned long long rv
= 0;
1111 /* find first active disk in array */
1113 while (dl
&& (is_failed(&dl
->disk
) || dl
->index
== -1))
1117 /* find last lba used by subarrays */
1118 e
= get_extents(super
, dl
);
1121 for (i
= 0; e
[i
].size
; i
++)
1124 rv
= e
[i
-1].start
+ e
[i
-1].size
;
1127 /* add the amount of space needed for metadata */
1128 rv
= rv
+ imsm_min_reserved_sectors(super
);
1133 static int is_gen_migration(struct imsm_dev
*dev
);
1136 static __u64
blocks_per_migr_unit(struct intel_super
*super
,
1137 struct imsm_dev
*dev
);
1139 static void print_imsm_dev(struct intel_super
*super
,
1140 struct imsm_dev
*dev
,
1146 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
1147 struct imsm_map
*map2
= get_imsm_map(dev
, MAP_1
);
1151 printf("[%.16s]:\n", dev
->volume
);
1152 printf(" UUID : %s\n", uuid
);
1153 printf(" RAID Level : %d", get_imsm_raid_level(map
));
1155 printf(" <-- %d", get_imsm_raid_level(map2
));
1157 printf(" Members : %d", map
->num_members
);
1159 printf(" <-- %d", map2
->num_members
);
1161 printf(" Slots : [");
1162 for (i
= 0; i
< map
->num_members
; i
++) {
1163 ord
= get_imsm_ord_tbl_ent(dev
, i
, MAP_0
);
1164 printf("%s", ord
& IMSM_ORD_REBUILD
? "_" : "U");
1169 for (i
= 0; i
< map2
->num_members
; i
++) {
1170 ord
= get_imsm_ord_tbl_ent(dev
, i
, MAP_1
);
1171 printf("%s", ord
& IMSM_ORD_REBUILD
? "_" : "U");
1176 printf(" Failed disk : ");
1177 if (map
->failed_disk_num
== 0xff)
1180 printf("%i", map
->failed_disk_num
);
1182 slot
= get_imsm_disk_slot(map
, disk_idx
);
1184 ord
= get_imsm_ord_tbl_ent(dev
, slot
, MAP_X
);
1185 printf(" This Slot : %d%s\n", slot
,
1186 ord
& IMSM_ORD_REBUILD
? " (out-of-sync)" : "");
1188 printf(" This Slot : ?\n");
1189 sz
= __le32_to_cpu(dev
->size_high
);
1191 sz
+= __le32_to_cpu(dev
->size_low
);
1192 printf(" Array Size : %llu%s\n", (unsigned long long)sz
,
1193 human_size(sz
* 512));
1194 sz
= blocks_per_member(map
);
1195 printf(" Per Dev Size : %llu%s\n", (unsigned long long)sz
,
1196 human_size(sz
* 512));
1197 printf(" Sector Offset : %llu\n",
1199 printf(" Num Stripes : %llu\n",
1200 num_data_stripes(map
));
1201 printf(" Chunk Size : %u KiB",
1202 __le16_to_cpu(map
->blocks_per_strip
) / 2);
1204 printf(" <-- %u KiB",
1205 __le16_to_cpu(map2
->blocks_per_strip
) / 2);
1207 printf(" Reserved : %d\n", __le32_to_cpu(dev
->reserved_blocks
));
1208 printf(" Migrate State : ");
1209 if (dev
->vol
.migr_state
) {
1210 if (migr_type(dev
) == MIGR_INIT
)
1211 printf("initialize\n");
1212 else if (migr_type(dev
) == MIGR_REBUILD
)
1213 printf("rebuild\n");
1214 else if (migr_type(dev
) == MIGR_VERIFY
)
1216 else if (migr_type(dev
) == MIGR_GEN_MIGR
)
1217 printf("general migration\n");
1218 else if (migr_type(dev
) == MIGR_STATE_CHANGE
)
1219 printf("state change\n");
1220 else if (migr_type(dev
) == MIGR_REPAIR
)
1223 printf("<unknown:%d>\n", migr_type(dev
));
1226 printf(" Map State : %s", map_state_str
[map
->map_state
]);
1227 if (dev
->vol
.migr_state
) {
1228 struct imsm_map
*map
= get_imsm_map(dev
, MAP_1
);
1230 printf(" <-- %s", map_state_str
[map
->map_state
]);
1231 printf("\n Checkpoint : %u ",
1232 __le32_to_cpu(dev
->vol
.curr_migr_unit
));
1233 if ((is_gen_migration(dev
)) && ((slot
> 1) || (slot
< 0)))
1236 printf("(%llu)", (unsigned long long)
1237 blocks_per_migr_unit(super
, dev
));
1240 printf(" Dirty State : %s\n", dev
->vol
.dirty
? "dirty" : "clean");
1243 static void print_imsm_disk(struct imsm_disk
*disk
, int index
, __u32 reserved
)
1245 char str
[MAX_RAID_SERIAL_LEN
+ 1];
1248 if (index
< -1 || !disk
)
1252 snprintf(str
, MAX_RAID_SERIAL_LEN
+ 1, "%s", disk
->serial
);
1254 printf(" Disk%02d Serial : %s\n", index
, str
);
1256 printf(" Disk Serial : %s\n", str
);
1257 printf(" State :%s%s%s\n", is_spare(disk
) ? " spare" : "",
1258 is_configured(disk
) ? " active" : "",
1259 is_failed(disk
) ? " failed" : "");
1260 printf(" Id : %08x\n", __le32_to_cpu(disk
->scsi_id
));
1261 sz
= total_blocks(disk
) - reserved
;
1262 printf(" Usable Size : %llu%s\n", (unsigned long long)sz
,
1263 human_size(sz
* 512));
1266 void examine_migr_rec_imsm(struct intel_super
*super
)
1268 struct migr_record
*migr_rec
= super
->migr_rec
;
1269 struct imsm_super
*mpb
= super
->anchor
;
1272 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
1273 struct imsm_dev
*dev
= __get_imsm_dev(mpb
, i
);
1274 struct imsm_map
*map
;
1277 if (is_gen_migration(dev
) == 0)
1280 printf("\nMigration Record Information:");
1282 /* first map under migration */
1283 map
= get_imsm_map(dev
, MAP_0
);
1285 slot
= get_imsm_disk_slot(map
, super
->disks
->index
);
1286 if ((map
== NULL
) || (slot
> 1) || (slot
< 0)) {
1287 printf(" Empty\n ");
1288 printf("Examine one of first two disks in array\n");
1291 printf("\n Status : ");
1292 if (__le32_to_cpu(migr_rec
->rec_status
) == UNIT_SRC_NORMAL
)
1295 printf("Contains Data\n");
1296 printf(" Current Unit : %u\n",
1297 __le32_to_cpu(migr_rec
->curr_migr_unit
));
1298 printf(" Family : %u\n",
1299 __le32_to_cpu(migr_rec
->family_num
));
1300 printf(" Ascending : %u\n",
1301 __le32_to_cpu(migr_rec
->ascending_migr
));
1302 printf(" Blocks Per Unit : %u\n",
1303 __le32_to_cpu(migr_rec
->blocks_per_unit
));
1304 printf(" Dest. Depth Per Unit : %u\n",
1305 __le32_to_cpu(migr_rec
->dest_depth_per_unit
));
1306 printf(" Checkpoint Area pba : %u\n",
1307 __le32_to_cpu(migr_rec
->ckpt_area_pba
));
1308 printf(" First member lba : %u\n",
1309 __le32_to_cpu(migr_rec
->dest_1st_member_lba
));
1310 printf(" Total Number of Units : %u\n",
1311 __le32_to_cpu(migr_rec
->num_migr_units
));
1312 printf(" Size of volume : %u\n",
1313 __le32_to_cpu(migr_rec
->post_migr_vol_cap
));
1314 printf(" Expansion space for LBA64 : %u\n",
1315 __le32_to_cpu(migr_rec
->post_migr_vol_cap_hi
));
1316 printf(" Record was read from : %u\n",
1317 __le32_to_cpu(migr_rec
->ckpt_read_disk_num
));
1322 #endif /* MDASSEMBLE */
1323 /*******************************************************************************
1324 * function: imsm_check_attributes
1325 * Description: Function checks if features represented by attributes flags
1326 * are supported by mdadm.
1328 * attributes - Attributes read from metadata
1330 * 0 - passed attributes contains unsupported features flags
1331 * 1 - all features are supported
1332 ******************************************************************************/
1333 static int imsm_check_attributes(__u32 attributes
)
1336 __u32 not_supported
= MPB_ATTRIB_SUPPORTED
^0xffffffff;
1338 not_supported
&= ~MPB_ATTRIB_IGNORED
;
1340 not_supported
&= attributes
;
1341 if (not_supported
) {
1342 fprintf(stderr
, Name
"(IMSM): Unsupported attributes : %x\n",
1343 (unsigned)__le32_to_cpu(not_supported
));
1344 if (not_supported
& MPB_ATTRIB_CHECKSUM_VERIFY
) {
1345 dprintf("\t\tMPB_ATTRIB_CHECKSUM_VERIFY \n");
1346 not_supported
^= MPB_ATTRIB_CHECKSUM_VERIFY
;
1348 if (not_supported
& MPB_ATTRIB_2TB
) {
1349 dprintf("\t\tMPB_ATTRIB_2TB\n");
1350 not_supported
^= MPB_ATTRIB_2TB
;
1352 if (not_supported
& MPB_ATTRIB_RAID0
) {
1353 dprintf("\t\tMPB_ATTRIB_RAID0\n");
1354 not_supported
^= MPB_ATTRIB_RAID0
;
1356 if (not_supported
& MPB_ATTRIB_RAID1
) {
1357 dprintf("\t\tMPB_ATTRIB_RAID1\n");
1358 not_supported
^= MPB_ATTRIB_RAID1
;
1360 if (not_supported
& MPB_ATTRIB_RAID10
) {
1361 dprintf("\t\tMPB_ATTRIB_RAID10\n");
1362 not_supported
^= MPB_ATTRIB_RAID10
;
1364 if (not_supported
& MPB_ATTRIB_RAID1E
) {
1365 dprintf("\t\tMPB_ATTRIB_RAID1E\n");
1366 not_supported
^= MPB_ATTRIB_RAID1E
;
1368 if (not_supported
& MPB_ATTRIB_RAID5
) {
1369 dprintf("\t\tMPB_ATTRIB_RAID5\n");
1370 not_supported
^= MPB_ATTRIB_RAID5
;
1372 if (not_supported
& MPB_ATTRIB_RAIDCNG
) {
1373 dprintf("\t\tMPB_ATTRIB_RAIDCNG\n");
1374 not_supported
^= MPB_ATTRIB_RAIDCNG
;
1376 if (not_supported
& MPB_ATTRIB_BBM
) {
1377 dprintf("\t\tMPB_ATTRIB_BBM\n");
1378 not_supported
^= MPB_ATTRIB_BBM
;
1380 if (not_supported
& MPB_ATTRIB_CHECKSUM_VERIFY
) {
1381 dprintf("\t\tMPB_ATTRIB_CHECKSUM_VERIFY (== MPB_ATTRIB_LEGACY)\n");
1382 not_supported
^= MPB_ATTRIB_CHECKSUM_VERIFY
;
1384 if (not_supported
& MPB_ATTRIB_EXP_STRIPE_SIZE
) {
1385 dprintf("\t\tMPB_ATTRIB_EXP_STRIP_SIZE\n");
1386 not_supported
^= MPB_ATTRIB_EXP_STRIPE_SIZE
;
1388 if (not_supported
& MPB_ATTRIB_2TB_DISK
) {
1389 dprintf("\t\tMPB_ATTRIB_2TB_DISK\n");
1390 not_supported
^= MPB_ATTRIB_2TB_DISK
;
1392 if (not_supported
& MPB_ATTRIB_NEVER_USE2
) {
1393 dprintf("\t\tMPB_ATTRIB_NEVER_USE2\n");
1394 not_supported
^= MPB_ATTRIB_NEVER_USE2
;
1396 if (not_supported
& MPB_ATTRIB_NEVER_USE
) {
1397 dprintf("\t\tMPB_ATTRIB_NEVER_USE\n");
1398 not_supported
^= MPB_ATTRIB_NEVER_USE
;
1402 dprintf(Name
"(IMSM): Unknown attributes : %x\n", not_supported
);
1411 static void getinfo_super_imsm(struct supertype
*st
, struct mdinfo
*info
, char *map
);
1413 static void examine_super_imsm(struct supertype
*st
, char *homehost
)
1415 struct intel_super
*super
= st
->sb
;
1416 struct imsm_super
*mpb
= super
->anchor
;
1417 char str
[MAX_SIGNATURE_LENGTH
];
1422 __u32 reserved
= imsm_reserved_sectors(super
, super
->disks
);
1425 snprintf(str
, MPB_SIG_LEN
, "%s", mpb
->sig
);
1426 printf(" Magic : %s\n", str
);
1427 snprintf(str
, strlen(MPB_VERSION_RAID0
), "%s", get_imsm_version(mpb
));
1428 printf(" Version : %s\n", get_imsm_version(mpb
));
1429 printf(" Orig Family : %08x\n", __le32_to_cpu(mpb
->orig_family_num
));
1430 printf(" Family : %08x\n", __le32_to_cpu(mpb
->family_num
));
1431 printf(" Generation : %08x\n", __le32_to_cpu(mpb
->generation_num
));
1432 printf(" Attributes : ");
1433 if (imsm_check_attributes(mpb
->attributes
))
1434 printf("All supported\n");
1436 printf("not supported\n");
1437 getinfo_super_imsm(st
, &info
, NULL
);
1438 fname_from_uuid(st
, &info
, nbuf
, ':');
1439 printf(" UUID : %s\n", nbuf
+ 5);
1440 sum
= __le32_to_cpu(mpb
->check_sum
);
1441 printf(" Checksum : %08x %s\n", sum
,
1442 __gen_imsm_checksum(mpb
) == sum
? "correct" : "incorrect");
1443 printf(" MPB Sectors : %d\n", mpb_sectors(mpb
));
1444 printf(" Disks : %d\n", mpb
->num_disks
);
1445 printf(" RAID Devices : %d\n", mpb
->num_raid_devs
);
1446 print_imsm_disk(__get_imsm_disk(mpb
, super
->disks
->index
), super
->disks
->index
, reserved
);
1447 if (super
->bbm_log
) {
1448 struct bbm_log
*log
= super
->bbm_log
;
1451 printf("Bad Block Management Log:\n");
1452 printf(" Log Size : %d\n", __le32_to_cpu(mpb
->bbm_log_size
));
1453 printf(" Signature : %x\n", __le32_to_cpu(log
->signature
));
1454 printf(" Entry Count : %d\n", __le32_to_cpu(log
->entry_count
));
1455 printf(" Spare Blocks : %d\n", __le32_to_cpu(log
->reserved_spare_block_count
));
1456 printf(" First Spare : %llx\n",
1457 (unsigned long long) __le64_to_cpu(log
->first_spare_lba
));
1459 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
1461 struct imsm_dev
*dev
= __get_imsm_dev(mpb
, i
);
1463 super
->current_vol
= i
;
1464 getinfo_super_imsm(st
, &info
, NULL
);
1465 fname_from_uuid(st
, &info
, nbuf
, ':');
1466 print_imsm_dev(super
, dev
, nbuf
+ 5, super
->disks
->index
);
1468 for (i
= 0; i
< mpb
->num_disks
; i
++) {
1469 if (i
== super
->disks
->index
)
1471 print_imsm_disk(__get_imsm_disk(mpb
, i
), i
, reserved
);
1474 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
1475 if (dl
->index
== -1)
1476 print_imsm_disk(&dl
->disk
, -1, reserved
);
1478 examine_migr_rec_imsm(super
);
1481 static void brief_examine_super_imsm(struct supertype
*st
, int verbose
)
1483 /* We just write a generic IMSM ARRAY entry */
1486 struct intel_super
*super
= st
->sb
;
1488 if (!super
->anchor
->num_raid_devs
) {
1489 printf("ARRAY metadata=imsm\n");
1493 getinfo_super_imsm(st
, &info
, NULL
);
1494 fname_from_uuid(st
, &info
, nbuf
, ':');
1495 printf("ARRAY metadata=imsm UUID=%s\n", nbuf
+ 5);
1498 static void brief_examine_subarrays_imsm(struct supertype
*st
, int verbose
)
1500 /* We just write a generic IMSM ARRAY entry */
1504 struct intel_super
*super
= st
->sb
;
1507 if (!super
->anchor
->num_raid_devs
)
1510 getinfo_super_imsm(st
, &info
, NULL
);
1511 fname_from_uuid(st
, &info
, nbuf
, ':');
1512 for (i
= 0; i
< super
->anchor
->num_raid_devs
; i
++) {
1513 struct imsm_dev
*dev
= get_imsm_dev(super
, i
);
1515 super
->current_vol
= i
;
1516 getinfo_super_imsm(st
, &info
, NULL
);
1517 fname_from_uuid(st
, &info
, nbuf1
, ':');
1518 printf("ARRAY /dev/md/%.16s container=%s member=%d UUID=%s\n",
1519 dev
->volume
, nbuf
+ 5, i
, nbuf1
+ 5);
1523 static void export_examine_super_imsm(struct supertype
*st
)
1525 struct intel_super
*super
= st
->sb
;
1526 struct imsm_super
*mpb
= super
->anchor
;
1530 getinfo_super_imsm(st
, &info
, NULL
);
1531 fname_from_uuid(st
, &info
, nbuf
, ':');
1532 printf("MD_METADATA=imsm\n");
1533 printf("MD_LEVEL=container\n");
1534 printf("MD_UUID=%s\n", nbuf
+5);
1535 printf("MD_DEVICES=%u\n", mpb
->num_disks
);
1538 static void detail_super_imsm(struct supertype
*st
, char *homehost
)
1543 getinfo_super_imsm(st
, &info
, NULL
);
1544 fname_from_uuid(st
, &info
, nbuf
, ':');
1545 printf("\n UUID : %s\n", nbuf
+ 5);
1548 static void brief_detail_super_imsm(struct supertype
*st
)
1552 getinfo_super_imsm(st
, &info
, NULL
);
1553 fname_from_uuid(st
, &info
, nbuf
, ':');
1554 printf(" UUID=%s", nbuf
+ 5);
1557 static int imsm_read_serial(int fd
, char *devname
, __u8
*serial
);
1558 static void fd2devname(int fd
, char *name
);
1560 static int ahci_enumerate_ports(const char *hba_path
, int port_count
, int host_base
, int verbose
)
1562 /* dump an unsorted list of devices attached to AHCI Intel storage
1563 * controller, as well as non-connected ports
1565 int hba_len
= strlen(hba_path
) + 1;
1570 unsigned long port_mask
= (1 << port_count
) - 1;
1572 if (port_count
> (int)sizeof(port_mask
) * 8) {
1574 fprintf(stderr
, Name
": port_count %d out of range\n", port_count
);
1578 /* scroll through /sys/dev/block looking for devices attached to
1581 dir
= opendir("/sys/dev/block");
1582 for (ent
= dir
? readdir(dir
) : NULL
; ent
; ent
= readdir(dir
)) {
1593 if (sscanf(ent
->d_name
, "%d:%d", &major
, &minor
) != 2)
1595 path
= devt_to_devpath(makedev(major
, minor
));
1598 if (!path_attached_to_hba(path
, hba_path
)) {
1604 /* retrieve the scsi device type */
1605 if (asprintf(&device
, "/sys/dev/block/%d:%d/device/xxxxxxx", major
, minor
) < 0) {
1607 fprintf(stderr
, Name
": failed to allocate 'device'\n");
1611 sprintf(device
, "/sys/dev/block/%d:%d/device/type", major
, minor
);
1612 if (load_sys(device
, buf
) != 0) {
1614 fprintf(stderr
, Name
": failed to read device type for %s\n",
1620 type
= strtoul(buf
, NULL
, 10);
1622 /* if it's not a disk print the vendor and model */
1623 if (!(type
== 0 || type
== 7 || type
== 14)) {
1626 sprintf(device
, "/sys/dev/block/%d:%d/device/vendor", major
, minor
);
1627 if (load_sys(device
, buf
) == 0) {
1628 strncpy(vendor
, buf
, sizeof(vendor
));
1629 vendor
[sizeof(vendor
) - 1] = '\0';
1630 c
= (char *) &vendor
[sizeof(vendor
) - 1];
1631 while (isspace(*c
) || *c
== '\0')
1635 sprintf(device
, "/sys/dev/block/%d:%d/device/model", major
, minor
);
1636 if (load_sys(device
, buf
) == 0) {
1637 strncpy(model
, buf
, sizeof(model
));
1638 model
[sizeof(model
) - 1] = '\0';
1639 c
= (char *) &model
[sizeof(model
) - 1];
1640 while (isspace(*c
) || *c
== '\0')
1644 if (vendor
[0] && model
[0])
1645 sprintf(buf
, "%.64s %.64s", vendor
, model
);
1647 switch (type
) { /* numbers from hald/linux/device.c */
1648 case 1: sprintf(buf
, "tape"); break;
1649 case 2: sprintf(buf
, "printer"); break;
1650 case 3: sprintf(buf
, "processor"); break;
1652 case 5: sprintf(buf
, "cdrom"); break;
1653 case 6: sprintf(buf
, "scanner"); break;
1654 case 8: sprintf(buf
, "media_changer"); break;
1655 case 9: sprintf(buf
, "comm"); break;
1656 case 12: sprintf(buf
, "raid"); break;
1657 default: sprintf(buf
, "unknown");
1663 /* chop device path to 'host%d' and calculate the port number */
1664 c
= strchr(&path
[hba_len
], '/');
1667 fprintf(stderr
, Name
": %s - invalid path name\n", path
+ hba_len
);
1672 if (sscanf(&path
[hba_len
], "host%d", &port
) == 1)
1676 *c
= '/'; /* repair the full string */
1677 fprintf(stderr
, Name
": failed to determine port number for %s\n",
1684 /* mark this port as used */
1685 port_mask
&= ~(1 << port
);
1687 /* print out the device information */
1689 printf(" Port%d : - non-disk device (%s) -\n", port
, buf
);
1693 fd
= dev_open(ent
->d_name
, O_RDONLY
);
1695 printf(" Port%d : - disk info unavailable -\n", port
);
1697 fd2devname(fd
, buf
);
1698 printf(" Port%d : %s", port
, buf
);
1699 if (imsm_read_serial(fd
, NULL
, (__u8
*) buf
) == 0)
1700 printf(" (%.*s)\n", MAX_RAID_SERIAL_LEN
, buf
);
1715 for (i
= 0; i
< port_count
; i
++)
1716 if (port_mask
& (1 << i
))
1717 printf(" Port%d : - no device attached -\n", i
);
1723 static void print_found_intel_controllers(struct sys_dev
*elem
)
1725 for (; elem
; elem
= elem
->next
) {
1726 fprintf(stderr
, Name
": found Intel(R) ");
1727 if (elem
->type
== SYS_DEV_SATA
)
1728 fprintf(stderr
, "SATA ");
1729 else if (elem
->type
== SYS_DEV_SAS
)
1730 fprintf(stderr
, "SAS ");
1731 fprintf(stderr
, "RAID controller");
1733 fprintf(stderr
, " at %s", elem
->pci_id
);
1734 fprintf(stderr
, ".\n");
1739 static int ahci_get_port_count(const char *hba_path
, int *port_count
)
1746 if ((dir
= opendir(hba_path
)) == NULL
)
1749 for (ent
= readdir(dir
); ent
; ent
= readdir(dir
)) {
1752 if (sscanf(ent
->d_name
, "host%d", &host
) != 1)
1754 if (*port_count
== 0)
1756 else if (host
< host_base
)
1759 if (host
+ 1 > *port_count
+ host_base
)
1760 *port_count
= host
+ 1 - host_base
;
1766 static void print_imsm_capability(const struct imsm_orom
*orom
)
1768 printf(" Platform : Intel(R) Matrix Storage Manager\n");
1769 printf(" Version : %d.%d.%d.%d\n", orom
->major_ver
, orom
->minor_ver
,
1770 orom
->hotfix_ver
, orom
->build
);
1771 printf(" RAID Levels :%s%s%s%s%s\n",
1772 imsm_orom_has_raid0(orom
) ? " raid0" : "",
1773 imsm_orom_has_raid1(orom
) ? " raid1" : "",
1774 imsm_orom_has_raid1e(orom
) ? " raid1e" : "",
1775 imsm_orom_has_raid10(orom
) ? " raid10" : "",
1776 imsm_orom_has_raid5(orom
) ? " raid5" : "");
1777 printf(" Chunk Sizes :%s%s%s%s%s%s%s%s%s%s%s%s%s%s%s%s\n",
1778 imsm_orom_has_chunk(orom
, 2) ? " 2k" : "",
1779 imsm_orom_has_chunk(orom
, 4) ? " 4k" : "",
1780 imsm_orom_has_chunk(orom
, 8) ? " 8k" : "",
1781 imsm_orom_has_chunk(orom
, 16) ? " 16k" : "",
1782 imsm_orom_has_chunk(orom
, 32) ? " 32k" : "",
1783 imsm_orom_has_chunk(orom
, 64) ? " 64k" : "",
1784 imsm_orom_has_chunk(orom
, 128) ? " 128k" : "",
1785 imsm_orom_has_chunk(orom
, 256) ? " 256k" : "",
1786 imsm_orom_has_chunk(orom
, 512) ? " 512k" : "",
1787 imsm_orom_has_chunk(orom
, 1024*1) ? " 1M" : "",
1788 imsm_orom_has_chunk(orom
, 1024*2) ? " 2M" : "",
1789 imsm_orom_has_chunk(orom
, 1024*4) ? " 4M" : "",
1790 imsm_orom_has_chunk(orom
, 1024*8) ? " 8M" : "",
1791 imsm_orom_has_chunk(orom
, 1024*16) ? " 16M" : "",
1792 imsm_orom_has_chunk(orom
, 1024*32) ? " 32M" : "",
1793 imsm_orom_has_chunk(orom
, 1024*64) ? " 64M" : "");
1794 printf(" 2TB volumes :%s supported\n",
1795 (orom
->attr
& IMSM_OROM_ATTR_2TB
)?"":" not");
1796 printf(" 2TB disks :%s supported\n",
1797 (orom
->attr
& IMSM_OROM_ATTR_2TB_DISK
)?"":" not");
1798 printf(" Max Disks : %d\n", orom
->tds
);
1799 printf(" Max Volumes : %d per array, %d per controller\n",
1800 orom
->vpa
, orom
->vphba
);
1804 static int detail_platform_imsm(int verbose
, int enumerate_only
)
1806 /* There are two components to imsm platform support, the ahci SATA
1807 * controller and the option-rom. To find the SATA controller we
1808 * simply look in /sys/bus/pci/drivers/ahci to see if an ahci
1809 * controller with the Intel vendor id is present. This approach
1810 * allows mdadm to leverage the kernel's ahci detection logic, with the
1811 * caveat that if ahci.ko is not loaded mdadm will not be able to
1812 * detect platform raid capabilities. The option-rom resides in a
1813 * platform "Adapter ROM". We scan for its signature to retrieve the
1814 * platform capabilities. If raid support is disabled in the BIOS the
1815 * option-rom capability structure will not be available.
1817 const struct imsm_orom
*orom
;
1818 struct sys_dev
*list
, *hba
;
1823 if (enumerate_only
) {
1824 if (check_env("IMSM_NO_PLATFORM"))
1826 list
= find_intel_devices();
1829 for (hba
= list
; hba
; hba
= hba
->next
) {
1830 orom
= find_imsm_capability(hba
->type
);
1836 free_sys_dev(&list
);
1840 list
= find_intel_devices();
1843 fprintf(stderr
, Name
": no active Intel(R) RAID "
1844 "controller found.\n");
1845 free_sys_dev(&list
);
1848 print_found_intel_controllers(list
);
1850 for (hba
= list
; hba
; hba
= hba
->next
) {
1851 orom
= find_imsm_capability(hba
->type
);
1853 fprintf(stderr
, Name
": imsm capabilities not found for controller: %s (type %s)\n",
1854 hba
->path
, get_sys_dev_type(hba
->type
));
1856 print_imsm_capability(orom
);
1859 for (hba
= list
; hba
; hba
= hba
->next
) {
1860 printf(" I/O Controller : %s (%s)\n",
1861 hba
->path
, get_sys_dev_type(hba
->type
));
1863 if (hba
->type
== SYS_DEV_SATA
) {
1864 host_base
= ahci_get_port_count(hba
->path
, &port_count
);
1865 if (ahci_enumerate_ports(hba
->path
, port_count
, host_base
, verbose
)) {
1867 fprintf(stderr
, Name
": failed to enumerate "
1868 "ports on SATA controller at %s.", hba
->pci_id
);
1874 free_sys_dev(&list
);
1879 static int match_home_imsm(struct supertype
*st
, char *homehost
)
1881 /* the imsm metadata format does not specify any host
1882 * identification information. We return -1 since we can never
1883 * confirm nor deny whether a given array is "meant" for this
1884 * host. We rely on compare_super and the 'family_num' fields to
1885 * exclude member disks that do not belong, and we rely on
1886 * mdadm.conf to specify the arrays that should be assembled.
1887 * Auto-assembly may still pick up "foreign" arrays.
1893 static void uuid_from_super_imsm(struct supertype
*st
, int uuid
[4])
1895 /* The uuid returned here is used for:
1896 * uuid to put into bitmap file (Create, Grow)
1897 * uuid for backup header when saving critical section (Grow)
1898 * comparing uuids when re-adding a device into an array
1899 * In these cases the uuid required is that of the data-array,
1900 * not the device-set.
1901 * uuid to recognise same set when adding a missing device back
1902 * to an array. This is a uuid for the device-set.
1904 * For each of these we can make do with a truncated
1905 * or hashed uuid rather than the original, as long as
1907 * In each case the uuid required is that of the data-array,
1908 * not the device-set.
1910 /* imsm does not track uuid's so we synthesis one using sha1 on
1911 * - The signature (Which is constant for all imsm array, but no matter)
1912 * - the orig_family_num of the container
1913 * - the index number of the volume
1914 * - the 'serial' number of the volume.
1915 * Hopefully these are all constant.
1917 struct intel_super
*super
= st
->sb
;
1920 struct sha1_ctx ctx
;
1921 struct imsm_dev
*dev
= NULL
;
1924 /* some mdadm versions failed to set ->orig_family_num, in which
1925 * case fall back to ->family_num. orig_family_num will be
1926 * fixed up with the first metadata update.
1928 family_num
= super
->anchor
->orig_family_num
;
1929 if (family_num
== 0)
1930 family_num
= super
->anchor
->family_num
;
1931 sha1_init_ctx(&ctx
);
1932 sha1_process_bytes(super
->anchor
->sig
, MPB_SIG_LEN
, &ctx
);
1933 sha1_process_bytes(&family_num
, sizeof(__u32
), &ctx
);
1934 if (super
->current_vol
>= 0)
1935 dev
= get_imsm_dev(super
, super
->current_vol
);
1937 __u32 vol
= super
->current_vol
;
1938 sha1_process_bytes(&vol
, sizeof(vol
), &ctx
);
1939 sha1_process_bytes(dev
->volume
, MAX_RAID_SERIAL_LEN
, &ctx
);
1941 sha1_finish_ctx(&ctx
, buf
);
1942 memcpy(uuid
, buf
, 4*4);
1947 get_imsm_numerical_version(struct imsm_super
*mpb
, int *m
, int *p
)
1949 __u8
*v
= get_imsm_version(mpb
);
1950 __u8
*end
= mpb
->sig
+ MAX_SIGNATURE_LENGTH
;
1951 char major
[] = { 0, 0, 0 };
1952 char minor
[] = { 0 ,0, 0 };
1953 char patch
[] = { 0, 0, 0 };
1954 char *ver_parse
[] = { major
, minor
, patch
};
1958 while (*v
!= '\0' && v
< end
) {
1959 if (*v
!= '.' && j
< 2)
1960 ver_parse
[i
][j
++] = *v
;
1968 *m
= strtol(minor
, NULL
, 0);
1969 *p
= strtol(patch
, NULL
, 0);
1973 static __u32
migr_strip_blocks_resync(struct imsm_dev
*dev
)
1975 /* migr_strip_size when repairing or initializing parity */
1976 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
1977 __u32 chunk
= __le32_to_cpu(map
->blocks_per_strip
);
1979 switch (get_imsm_raid_level(map
)) {
1984 return 128*1024 >> 9;
1988 static __u32
migr_strip_blocks_rebuild(struct imsm_dev
*dev
)
1990 /* migr_strip_size when rebuilding a degraded disk, no idea why
1991 * this is different than migr_strip_size_resync(), but it's good
1994 struct imsm_map
*map
= get_imsm_map(dev
, MAP_1
);
1995 __u32 chunk
= __le32_to_cpu(map
->blocks_per_strip
);
1997 switch (get_imsm_raid_level(map
)) {
2000 if (map
->num_members
% map
->num_domains
== 0)
2001 return 128*1024 >> 9;
2005 return max((__u32
) 64*1024 >> 9, chunk
);
2007 return 128*1024 >> 9;
2011 static __u32
num_stripes_per_unit_resync(struct imsm_dev
*dev
)
2013 struct imsm_map
*lo
= get_imsm_map(dev
, MAP_0
);
2014 struct imsm_map
*hi
= get_imsm_map(dev
, MAP_1
);
2015 __u32 lo_chunk
= __le32_to_cpu(lo
->blocks_per_strip
);
2016 __u32 hi_chunk
= __le32_to_cpu(hi
->blocks_per_strip
);
2018 return max((__u32
) 1, hi_chunk
/ lo_chunk
);
2021 static __u32
num_stripes_per_unit_rebuild(struct imsm_dev
*dev
)
2023 struct imsm_map
*lo
= get_imsm_map(dev
, MAP_0
);
2024 int level
= get_imsm_raid_level(lo
);
2026 if (level
== 1 || level
== 10) {
2027 struct imsm_map
*hi
= get_imsm_map(dev
, MAP_1
);
2029 return hi
->num_domains
;
2031 return num_stripes_per_unit_resync(dev
);
2034 static __u8
imsm_num_data_members(struct imsm_dev
*dev
, int second_map
)
2036 /* named 'imsm_' because raid0, raid1 and raid10
2037 * counter-intuitively have the same number of data disks
2039 struct imsm_map
*map
= get_imsm_map(dev
, second_map
);
2041 switch (get_imsm_raid_level(map
)) {
2043 return map
->num_members
;
2047 return map
->num_members
/2;
2049 return map
->num_members
- 1;
2051 dprintf("%s: unsupported raid level\n", __func__
);
2056 static __u32
parity_segment_depth(struct imsm_dev
*dev
)
2058 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
2059 __u32 chunk
= __le32_to_cpu(map
->blocks_per_strip
);
2061 switch(get_imsm_raid_level(map
)) {
2064 return chunk
* map
->num_domains
;
2066 return chunk
* map
->num_members
;
2072 static __u32
map_migr_block(struct imsm_dev
*dev
, __u32 block
)
2074 struct imsm_map
*map
= get_imsm_map(dev
, MAP_1
);
2075 __u32 chunk
= __le32_to_cpu(map
->blocks_per_strip
);
2076 __u32 strip
= block
/ chunk
;
2078 switch (get_imsm_raid_level(map
)) {
2081 __u32 vol_strip
= (strip
* map
->num_domains
) + 1;
2082 __u32 vol_stripe
= vol_strip
/ map
->num_members
;
2084 return vol_stripe
* chunk
+ block
% chunk
;
2086 __u32 stripe
= strip
/ (map
->num_members
- 1);
2088 return stripe
* chunk
+ block
% chunk
;
2095 static __u64
blocks_per_migr_unit(struct intel_super
*super
,
2096 struct imsm_dev
*dev
)
2098 /* calculate the conversion factor between per member 'blocks'
2099 * (md/{resync,rebuild}_start) and imsm migration units, return
2100 * 0 for the 'not migrating' and 'unsupported migration' cases
2102 if (!dev
->vol
.migr_state
)
2105 switch (migr_type(dev
)) {
2106 case MIGR_GEN_MIGR
: {
2107 struct migr_record
*migr_rec
= super
->migr_rec
;
2108 return __le32_to_cpu(migr_rec
->blocks_per_unit
);
2113 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
2114 __u32 stripes_per_unit
;
2115 __u32 blocks_per_unit
;
2124 /* yes, this is really the translation of migr_units to
2125 * per-member blocks in the 'resync' case
2127 stripes_per_unit
= num_stripes_per_unit_resync(dev
);
2128 migr_chunk
= migr_strip_blocks_resync(dev
);
2129 disks
= imsm_num_data_members(dev
, MAP_0
);
2130 blocks_per_unit
= stripes_per_unit
* migr_chunk
* disks
;
2131 stripe
= __le16_to_cpu(map
->blocks_per_strip
) * disks
;
2132 segment
= blocks_per_unit
/ stripe
;
2133 block_rel
= blocks_per_unit
- segment
* stripe
;
2134 parity_depth
= parity_segment_depth(dev
);
2135 block_map
= map_migr_block(dev
, block_rel
);
2136 return block_map
+ parity_depth
* segment
;
2138 case MIGR_REBUILD
: {
2139 __u32 stripes_per_unit
;
2142 stripes_per_unit
= num_stripes_per_unit_rebuild(dev
);
2143 migr_chunk
= migr_strip_blocks_rebuild(dev
);
2144 return migr_chunk
* stripes_per_unit
;
2146 case MIGR_STATE_CHANGE
:
2152 static int imsm_level_to_layout(int level
)
2160 return ALGORITHM_LEFT_ASYMMETRIC
;
2167 /*******************************************************************************
2168 * Function: read_imsm_migr_rec
2169 * Description: Function reads imsm migration record from last sector of disk
2171 * fd : disk descriptor
2172 * super : metadata info
2176 ******************************************************************************/
2177 static int read_imsm_migr_rec(int fd
, struct intel_super
*super
)
2180 unsigned long long dsize
;
2182 get_dev_size(fd
, NULL
, &dsize
);
2183 if (lseek64(fd
, dsize
- MIGR_REC_POSITION
, SEEK_SET
) < 0) {
2185 Name
": Cannot seek to anchor block: %s\n",
2189 if (read(fd
, super
->migr_rec_buf
, MIGR_REC_BUF_SIZE
) !=
2190 MIGR_REC_BUF_SIZE
) {
2192 Name
": Cannot read migr record block: %s\n",
2202 static struct imsm_dev
*imsm_get_device_during_migration(
2203 struct intel_super
*super
)
2206 struct intel_dev
*dv
;
2208 for (dv
= super
->devlist
; dv
; dv
= dv
->next
) {
2209 if (is_gen_migration(dv
->dev
))
2215 /*******************************************************************************
2216 * Function: load_imsm_migr_rec
2217 * Description: Function reads imsm migration record (it is stored at the last
2220 * super : imsm internal array info
2221 * info : general array info
2225 * -2 : no migration in progress
2226 ******************************************************************************/
2227 static int load_imsm_migr_rec(struct intel_super
*super
, struct mdinfo
*info
)
2230 struct dl
*dl
= NULL
;
2234 struct imsm_dev
*dev
;
2235 struct imsm_map
*map
= NULL
;
2238 /* find map under migration */
2239 dev
= imsm_get_device_during_migration(super
);
2240 /* nothing to load,no migration in progress?
2244 map
= get_imsm_map(dev
, MAP_0
);
2247 for (sd
= info
->devs
; sd
; sd
= sd
->next
) {
2248 /* skip spare and failed disks
2250 if (sd
->disk
.raid_disk
< 0)
2252 /* read only from one of the first two slots */
2254 slot
= get_imsm_disk_slot(map
,
2255 sd
->disk
.raid_disk
);
2256 if ((map
== NULL
) || (slot
> 1) || (slot
< 0))
2259 sprintf(nm
, "%d:%d", sd
->disk
.major
, sd
->disk
.minor
);
2260 fd
= dev_open(nm
, O_RDONLY
);
2266 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
2267 /* skip spare and failed disks
2271 /* read only from one of the first two slots */
2273 slot
= get_imsm_disk_slot(map
, dl
->index
);
2274 if ((map
== NULL
) || (slot
> 1) || (slot
< 0))
2276 sprintf(nm
, "%d:%d", dl
->major
, dl
->minor
);
2277 fd
= dev_open(nm
, O_RDONLY
);
2284 retval
= read_imsm_migr_rec(fd
, super
);
2293 /*******************************************************************************
2294 * function: imsm_create_metadata_checkpoint_update
2295 * Description: It creates update for checkpoint change.
2297 * super : imsm internal array info
2298 * u : pointer to prepared update
2301 * If length is equal to 0, input pointer u contains no update
2302 ******************************************************************************/
2303 static int imsm_create_metadata_checkpoint_update(
2304 struct intel_super
*super
,
2305 struct imsm_update_general_migration_checkpoint
**u
)
2308 int update_memory_size
= 0;
2310 dprintf("imsm_create_metadata_checkpoint_update(enter)\n");
2316 /* size of all update data without anchor */
2317 update_memory_size
=
2318 sizeof(struct imsm_update_general_migration_checkpoint
);
2320 *u
= calloc(1, update_memory_size
);
2322 dprintf("error: cannot get memory for "
2323 "imsm_create_metadata_checkpoint_update update\n");
2326 (*u
)->type
= update_general_migration_checkpoint
;
2327 (*u
)->curr_migr_unit
= __le32_to_cpu(super
->migr_rec
->curr_migr_unit
);
2328 dprintf("imsm_create_metadata_checkpoint_update: prepared for %u\n",
2329 (*u
)->curr_migr_unit
);
2331 return update_memory_size
;
2335 static void imsm_update_metadata_locally(struct supertype
*st
,
2336 void *buf
, int len
);
2338 /*******************************************************************************
2339 * Function: write_imsm_migr_rec
2340 * Description: Function writes imsm migration record
2341 * (at the last sector of disk)
2343 * super : imsm internal array info
2347 ******************************************************************************/
2348 static int write_imsm_migr_rec(struct supertype
*st
)
2350 struct intel_super
*super
= st
->sb
;
2351 unsigned long long dsize
;
2357 struct imsm_update_general_migration_checkpoint
*u
;
2358 struct imsm_dev
*dev
;
2359 struct imsm_map
*map
= NULL
;
2361 /* find map under migration */
2362 dev
= imsm_get_device_during_migration(super
);
2363 /* if no migration, write buffer anyway to clear migr_record
2364 * on disk based on first available device
2367 dev
= get_imsm_dev(super
, super
->current_vol
< 0 ? 0 :
2368 super
->current_vol
);
2370 map
= get_imsm_map(dev
, MAP_0
);
2372 for (sd
= super
->disks
; sd
; sd
= sd
->next
) {
2375 /* skip failed and spare devices */
2378 /* write to 2 first slots only */
2380 slot
= get_imsm_disk_slot(map
, sd
->index
);
2381 if ((map
== NULL
) || (slot
> 1) || (slot
< 0))
2384 sprintf(nm
, "%d:%d", sd
->major
, sd
->minor
);
2385 fd
= dev_open(nm
, O_RDWR
);
2388 get_dev_size(fd
, NULL
, &dsize
);
2389 if (lseek64(fd
, dsize
- MIGR_REC_POSITION
, SEEK_SET
) < 0) {
2391 Name
": Cannot seek to anchor block: %s\n",
2395 if (write(fd
, super
->migr_rec_buf
, MIGR_REC_BUF_SIZE
) !=
2396 MIGR_REC_BUF_SIZE
) {
2398 Name
": Cannot write migr record block: %s\n",
2405 /* update checkpoint information in metadata */
2406 len
= imsm_create_metadata_checkpoint_update(super
, &u
);
2409 dprintf("imsm: Cannot prepare update\n");
2412 /* update metadata locally */
2413 imsm_update_metadata_locally(st
, u
, len
);
2414 /* and possibly remotely */
2415 if (st
->update_tail
) {
2416 append_metadata_update(st
, u
, len
);
2417 /* during reshape we do all work inside metadata handler
2418 * manage_reshape(), so metadata update has to be triggered
2421 flush_metadata_updates(st
);
2422 st
->update_tail
= &st
->updates
;
2432 #endif /* MDASSEMBLE */
2434 /* spare/missing disks activations are not allowe when
2435 * array/container performs reshape operation, because
2436 * all arrays in container works on the same disks set
2438 int imsm_reshape_blocks_arrays_changes(struct intel_super
*super
)
2441 struct intel_dev
*i_dev
;
2442 struct imsm_dev
*dev
;
2444 /* check whole container
2446 for (i_dev
= super
->devlist
; i_dev
; i_dev
= i_dev
->next
) {
2448 if (is_gen_migration(dev
)) {
2449 /* No repair during any migration in container
2457 static unsigned long long imsm_component_size_aligment_check(int level
,
2459 unsigned long long component_size
)
2461 unsigned int component_size_alligment
;
2463 /* check component size aligment
2465 component_size_alligment
= component_size
% (chunk_size
/512);
2467 dprintf("imsm_component_size_aligment_check(Level: %i, "
2468 "chunk_size = %i, component_size = %llu), "
2469 "component_size_alligment = %u\n",
2470 level
, chunk_size
, component_size
,
2471 component_size_alligment
);
2473 if (component_size_alligment
&& (level
!= 1) && (level
!= UnSet
)) {
2474 dprintf("imsm: reported component size alligned from %llu ",
2476 component_size
-= component_size_alligment
;
2477 dprintf("to %llu (%i).\n",
2478 component_size
, component_size_alligment
);
2481 return component_size
;
2484 static void getinfo_super_imsm_volume(struct supertype
*st
, struct mdinfo
*info
, char *dmap
)
2486 struct intel_super
*super
= st
->sb
;
2487 struct migr_record
*migr_rec
= super
->migr_rec
;
2488 struct imsm_dev
*dev
= get_imsm_dev(super
, super
->current_vol
);
2489 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
2490 struct imsm_map
*prev_map
= get_imsm_map(dev
, MAP_1
);
2491 struct imsm_map
*map_to_analyse
= map
;
2494 int map_disks
= info
->array
.raid_disks
;
2496 memset(info
, 0, sizeof(*info
));
2498 map_to_analyse
= prev_map
;
2500 dl
= super
->current_disk
;
2502 info
->container_member
= super
->current_vol
;
2503 info
->array
.raid_disks
= map
->num_members
;
2504 info
->array
.level
= get_imsm_raid_level(map_to_analyse
);
2505 info
->array
.layout
= imsm_level_to_layout(info
->array
.level
);
2506 info
->array
.md_minor
= -1;
2507 info
->array
.ctime
= 0;
2508 info
->array
.utime
= 0;
2509 info
->array
.chunk_size
=
2510 __le16_to_cpu(map_to_analyse
->blocks_per_strip
) << 9;
2511 info
->array
.state
= !dev
->vol
.dirty
;
2512 info
->custom_array_size
= __le32_to_cpu(dev
->size_high
);
2513 info
->custom_array_size
<<= 32;
2514 info
->custom_array_size
|= __le32_to_cpu(dev
->size_low
);
2515 info
->recovery_blocked
= imsm_reshape_blocks_arrays_changes(st
->sb
);
2517 if (is_gen_migration(dev
)) {
2518 info
->reshape_active
= 1;
2519 info
->new_level
= get_imsm_raid_level(map
);
2520 info
->new_layout
= imsm_level_to_layout(info
->new_level
);
2521 info
->new_chunk
= __le16_to_cpu(map
->blocks_per_strip
) << 9;
2522 info
->delta_disks
= map
->num_members
- prev_map
->num_members
;
2523 if (info
->delta_disks
) {
2524 /* this needs to be applied to every array
2527 info
->reshape_active
= CONTAINER_RESHAPE
;
2529 /* We shape information that we give to md might have to be
2530 * modify to cope with md's requirement for reshaping arrays.
2531 * For example, when reshaping a RAID0, md requires it to be
2532 * presented as a degraded RAID4.
2533 * Also if a RAID0 is migrating to a RAID5 we need to specify
2534 * the array as already being RAID5, but the 'before' layout
2535 * is a RAID4-like layout.
2537 switch (info
->array
.level
) {
2539 switch(info
->new_level
) {
2541 /* conversion is happening as RAID4 */
2542 info
->array
.level
= 4;
2543 info
->array
.raid_disks
+= 1;
2546 /* conversion is happening as RAID5 */
2547 info
->array
.level
= 5;
2548 info
->array
.layout
= ALGORITHM_PARITY_N
;
2549 info
->delta_disks
-= 1;
2552 /* FIXME error message */
2553 info
->array
.level
= UnSet
;
2559 info
->new_level
= UnSet
;
2560 info
->new_layout
= UnSet
;
2561 info
->new_chunk
= info
->array
.chunk_size
;
2562 info
->delta_disks
= 0;
2566 info
->disk
.major
= dl
->major
;
2567 info
->disk
.minor
= dl
->minor
;
2568 info
->disk
.number
= dl
->index
;
2569 info
->disk
.raid_disk
= get_imsm_disk_slot(map_to_analyse
,
2573 info
->data_offset
= pba_of_lba0(map_to_analyse
);
2574 info
->component_size
= blocks_per_member(map_to_analyse
);
2576 info
->component_size
= imsm_component_size_aligment_check(
2578 info
->array
.chunk_size
,
2579 info
->component_size
);
2581 memset(info
->uuid
, 0, sizeof(info
->uuid
));
2582 info
->recovery_start
= MaxSector
;
2584 info
->reshape_progress
= 0;
2585 info
->resync_start
= MaxSector
;
2586 if ((map_to_analyse
->map_state
== IMSM_T_STATE_UNINITIALIZED
||
2588 imsm_reshape_blocks_arrays_changes(super
) == 0) {
2589 info
->resync_start
= 0;
2591 if (dev
->vol
.migr_state
) {
2592 switch (migr_type(dev
)) {
2595 __u64 blocks_per_unit
= blocks_per_migr_unit(super
,
2597 __u64 units
= __le32_to_cpu(dev
->vol
.curr_migr_unit
);
2599 info
->resync_start
= blocks_per_unit
* units
;
2602 case MIGR_GEN_MIGR
: {
2603 __u64 blocks_per_unit
= blocks_per_migr_unit(super
,
2605 __u64 units
= __le32_to_cpu(migr_rec
->curr_migr_unit
);
2606 unsigned long long array_blocks
;
2609 if (__le32_to_cpu(migr_rec
->ascending_migr
) &&
2611 (__le32_to_cpu(migr_rec
->num_migr_units
)-1)) &&
2612 (super
->migr_rec
->rec_status
==
2613 __cpu_to_le32(UNIT_SRC_IN_CP_AREA
)))
2616 info
->reshape_progress
= blocks_per_unit
* units
;
2618 dprintf("IMSM: General Migration checkpoint : %llu "
2619 "(%llu) -> read reshape progress : %llu\n",
2620 (unsigned long long)units
,
2621 (unsigned long long)blocks_per_unit
,
2622 info
->reshape_progress
);
2624 used_disks
= imsm_num_data_members(dev
, MAP_1
);
2625 if (used_disks
> 0) {
2626 array_blocks
= blocks_per_member(map
) *
2628 /* round array size down to closest MB
2630 info
->custom_array_size
= (array_blocks
2631 >> SECT_PER_MB_SHIFT
)
2632 << SECT_PER_MB_SHIFT
;
2636 /* we could emulate the checkpointing of
2637 * 'sync_action=check' migrations, but for now
2638 * we just immediately complete them
2641 /* this is handled by container_content_imsm() */
2642 case MIGR_STATE_CHANGE
:
2643 /* FIXME handle other migrations */
2645 /* we are not dirty, so... */
2646 info
->resync_start
= MaxSector
;
2650 strncpy(info
->name
, (char *) dev
->volume
, MAX_RAID_SERIAL_LEN
);
2651 info
->name
[MAX_RAID_SERIAL_LEN
] = 0;
2653 info
->array
.major_version
= -1;
2654 info
->array
.minor_version
= -2;
2655 devname
= devnum2devname(st
->container_dev
);
2656 *info
->text_version
= '\0';
2658 sprintf(info
->text_version
, "/%s/%d", devname
, info
->container_member
);
2660 info
->safe_mode_delay
= 4000; /* 4 secs like the Matrix driver */
2661 uuid_from_super_imsm(st
, info
->uuid
);
2665 for (i
=0; i
<map_disks
; i
++) {
2667 if (i
< info
->array
.raid_disks
) {
2668 struct imsm_disk
*dsk
;
2669 j
= get_imsm_disk_idx(dev
, i
, MAP_X
);
2670 dsk
= get_imsm_disk(super
, j
);
2671 if (dsk
&& (dsk
->status
& CONFIGURED_DISK
))
2678 static __u8
imsm_check_degraded(struct intel_super
*super
, struct imsm_dev
*dev
,
2679 int failed
, int look_in_map
);
2681 static int imsm_count_failed(struct intel_super
*super
, struct imsm_dev
*dev
,
2686 static void manage_second_map(struct intel_super
*super
, struct imsm_dev
*dev
)
2688 if (is_gen_migration(dev
)) {
2691 struct imsm_map
*map2
= get_imsm_map(dev
, MAP_1
);
2693 failed
= imsm_count_failed(super
, dev
, MAP_1
);
2694 map_state
= imsm_check_degraded(super
, dev
, failed
, MAP_1
);
2695 if (map2
->map_state
!= map_state
) {
2696 map2
->map_state
= map_state
;
2697 super
->updates_pending
++;
2703 static struct imsm_disk
*get_imsm_missing(struct intel_super
*super
, __u8 index
)
2707 for (d
= super
->missing
; d
; d
= d
->next
)
2708 if (d
->index
== index
)
2713 static void getinfo_super_imsm(struct supertype
*st
, struct mdinfo
*info
, char *map
)
2715 struct intel_super
*super
= st
->sb
;
2716 struct imsm_disk
*disk
;
2717 int map_disks
= info
->array
.raid_disks
;
2718 int max_enough
= -1;
2720 struct imsm_super
*mpb
;
2722 if (super
->current_vol
>= 0) {
2723 getinfo_super_imsm_volume(st
, info
, map
);
2726 memset(info
, 0, sizeof(*info
));
2728 /* Set raid_disks to zero so that Assemble will always pull in valid
2731 info
->array
.raid_disks
= 0;
2732 info
->array
.level
= LEVEL_CONTAINER
;
2733 info
->array
.layout
= 0;
2734 info
->array
.md_minor
= -1;
2735 info
->array
.ctime
= 0; /* N/A for imsm */
2736 info
->array
.utime
= 0;
2737 info
->array
.chunk_size
= 0;
2739 info
->disk
.major
= 0;
2740 info
->disk
.minor
= 0;
2741 info
->disk
.raid_disk
= -1;
2742 info
->reshape_active
= 0;
2743 info
->array
.major_version
= -1;
2744 info
->array
.minor_version
= -2;
2745 strcpy(info
->text_version
, "imsm");
2746 info
->safe_mode_delay
= 0;
2747 info
->disk
.number
= -1;
2748 info
->disk
.state
= 0;
2750 info
->recovery_start
= MaxSector
;
2751 info
->recovery_blocked
= imsm_reshape_blocks_arrays_changes(st
->sb
);
2753 /* do we have the all the insync disks that we expect? */
2754 mpb
= super
->anchor
;
2756 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
2757 struct imsm_dev
*dev
= get_imsm_dev(super
, i
);
2758 int failed
, enough
, j
, missing
= 0;
2759 struct imsm_map
*map
;
2762 failed
= imsm_count_failed(super
, dev
, MAP_0
);
2763 state
= imsm_check_degraded(super
, dev
, failed
, MAP_0
);
2764 map
= get_imsm_map(dev
, MAP_0
);
2766 /* any newly missing disks?
2767 * (catches single-degraded vs double-degraded)
2769 for (j
= 0; j
< map
->num_members
; j
++) {
2770 __u32 ord
= get_imsm_ord_tbl_ent(dev
, j
, MAP_0
);
2771 __u32 idx
= ord_to_idx(ord
);
2773 if (!(ord
& IMSM_ORD_REBUILD
) &&
2774 get_imsm_missing(super
, idx
)) {
2780 if (state
== IMSM_T_STATE_FAILED
)
2782 else if (state
== IMSM_T_STATE_DEGRADED
&&
2783 (state
!= map
->map_state
|| missing
))
2785 else /* we're normal, or already degraded */
2787 if (is_gen_migration(dev
) && missing
) {
2788 /* during general migration we need all disks
2789 * that process is running on.
2790 * No new missing disk is allowed.
2794 /* no more checks necessary
2798 /* in the missing/failed disk case check to see
2799 * if at least one array is runnable
2801 max_enough
= max(max_enough
, enough
);
2803 dprintf("%s: enough: %d\n", __func__
, max_enough
);
2804 info
->container_enough
= max_enough
;
2807 __u32 reserved
= imsm_reserved_sectors(super
, super
->disks
);
2809 disk
= &super
->disks
->disk
;
2810 info
->data_offset
= total_blocks(&super
->disks
->disk
) - reserved
;
2811 info
->component_size
= reserved
;
2812 info
->disk
.state
= is_configured(disk
) ? (1 << MD_DISK_ACTIVE
) : 0;
2813 /* we don't change info->disk.raid_disk here because
2814 * this state will be finalized in mdmon after we have
2815 * found the 'most fresh' version of the metadata
2817 info
->disk
.state
|= is_failed(disk
) ? (1 << MD_DISK_FAULTY
) : 0;
2818 info
->disk
.state
|= is_spare(disk
) ? 0 : (1 << MD_DISK_SYNC
);
2821 /* only call uuid_from_super_imsm when this disk is part of a populated container,
2822 * ->compare_super may have updated the 'num_raid_devs' field for spares
2824 if (info
->disk
.state
& (1 << MD_DISK_SYNC
) || super
->anchor
->num_raid_devs
)
2825 uuid_from_super_imsm(st
, info
->uuid
);
2827 memcpy(info
->uuid
, uuid_zero
, sizeof(uuid_zero
));
2829 /* I don't know how to compute 'map' on imsm, so use safe default */
2832 for (i
= 0; i
< map_disks
; i
++)
2838 /* allocates memory and fills disk in mdinfo structure
2839 * for each disk in array */
2840 struct mdinfo
*getinfo_super_disks_imsm(struct supertype
*st
)
2842 struct mdinfo
*mddev
= NULL
;
2843 struct intel_super
*super
= st
->sb
;
2844 struct imsm_disk
*disk
;
2847 if (!super
|| !super
->disks
)
2850 mddev
= malloc(sizeof(*mddev
));
2852 fprintf(stderr
, Name
": Failed to allocate memory.\n");
2855 memset(mddev
, 0, sizeof(*mddev
));
2859 tmp
= malloc(sizeof(*tmp
));
2861 fprintf(stderr
, Name
": Failed to allocate memory.\n");
2866 memset(tmp
, 0, sizeof(*tmp
));
2868 tmp
->next
= mddev
->devs
;
2870 tmp
->disk
.number
= count
++;
2871 tmp
->disk
.major
= dl
->major
;
2872 tmp
->disk
.minor
= dl
->minor
;
2873 tmp
->disk
.state
= is_configured(disk
) ?
2874 (1 << MD_DISK_ACTIVE
) : 0;
2875 tmp
->disk
.state
|= is_failed(disk
) ? (1 << MD_DISK_FAULTY
) : 0;
2876 tmp
->disk
.state
|= is_spare(disk
) ? 0 : (1 << MD_DISK_SYNC
);
2877 tmp
->disk
.raid_disk
= -1;
2883 static int update_super_imsm(struct supertype
*st
, struct mdinfo
*info
,
2884 char *update
, char *devname
, int verbose
,
2885 int uuid_set
, char *homehost
)
2887 /* For 'assemble' and 'force' we need to return non-zero if any
2888 * change was made. For others, the return value is ignored.
2889 * Update options are:
2890 * force-one : This device looks a bit old but needs to be included,
2891 * update age info appropriately.
2892 * assemble: clear any 'faulty' flag to allow this device to
2894 * force-array: Array is degraded but being forced, mark it clean
2895 * if that will be needed to assemble it.
2897 * newdev: not used ????
2898 * grow: Array has gained a new device - this is currently for
2900 * resync: mark as dirty so a resync will happen.
2901 * name: update the name - preserving the homehost
2902 * uuid: Change the uuid of the array to match watch is given
2904 * Following are not relevant for this imsm:
2905 * sparc2.2 : update from old dodgey metadata
2906 * super-minor: change the preferred_minor number
2907 * summaries: update redundant counters.
2908 * homehost: update the recorded homehost
2909 * _reshape_progress: record new reshape_progress position.
2912 struct intel_super
*super
= st
->sb
;
2913 struct imsm_super
*mpb
;
2915 /* we can only update container info */
2916 if (!super
|| super
->current_vol
>= 0 || !super
->anchor
)
2919 mpb
= super
->anchor
;
2921 if (strcmp(update
, "uuid") == 0) {
2922 /* We take this to mean that the family_num should be updated.
2923 * However that is much smaller than the uuid so we cannot really
2924 * allow an explicit uuid to be given. And it is hard to reliably
2926 * So if !uuid_set we know the current uuid is random and just used
2927 * the first 'int' and copy it to the other 3 positions.
2928 * Otherwise we require the 4 'int's to be the same as would be the
2929 * case if we are using a random uuid. So an explicit uuid will be
2930 * accepted as long as all for ints are the same... which shouldn't hurt
2933 info
->uuid
[1] = info
->uuid
[2] = info
->uuid
[3] = info
->uuid
[0];
2936 if (info
->uuid
[0] != info
->uuid
[1] ||
2937 info
->uuid
[1] != info
->uuid
[2] ||
2938 info
->uuid
[2] != info
->uuid
[3])
2944 mpb
->orig_family_num
= info
->uuid
[0];
2945 } else if (strcmp(update
, "assemble") == 0)
2950 /* successful update? recompute checksum */
2952 mpb
->check_sum
= __le32_to_cpu(__gen_imsm_checksum(mpb
));
2957 static size_t disks_to_mpb_size(int disks
)
2961 size
= sizeof(struct imsm_super
);
2962 size
+= (disks
- 1) * sizeof(struct imsm_disk
);
2963 size
+= 2 * sizeof(struct imsm_dev
);
2964 /* up to 2 maps per raid device (-2 for imsm_maps in imsm_dev */
2965 size
+= (4 - 2) * sizeof(struct imsm_map
);
2966 /* 4 possible disk_ord_tbl's */
2967 size
+= 4 * (disks
- 1) * sizeof(__u32
);
2972 static __u64
avail_size_imsm(struct supertype
*st
, __u64 devsize
)
2974 if (devsize
< (MPB_SECTOR_CNT
+ IMSM_RESERVED_SECTORS
))
2977 return devsize
- (MPB_SECTOR_CNT
+ IMSM_RESERVED_SECTORS
);
2980 static void free_devlist(struct intel_super
*super
)
2982 struct intel_dev
*dv
;
2984 while (super
->devlist
) {
2985 dv
= super
->devlist
->next
;
2986 free(super
->devlist
->dev
);
2987 free(super
->devlist
);
2988 super
->devlist
= dv
;
2992 static void imsm_copy_dev(struct imsm_dev
*dest
, struct imsm_dev
*src
)
2994 memcpy(dest
, src
, sizeof_imsm_dev(src
, 0));
2997 static int compare_super_imsm(struct supertype
*st
, struct supertype
*tst
)
3001 * 0 same, or first was empty, and second was copied
3002 * 1 second had wrong number
3004 * 3 wrong other info
3006 struct intel_super
*first
= st
->sb
;
3007 struct intel_super
*sec
= tst
->sb
;
3014 /* in platform dependent environment test if the disks
3015 * use the same Intel hba
3017 if (!check_env("IMSM_NO_PLATFORM")) {
3018 if (!first
->hba
|| !sec
->hba
||
3019 (first
->hba
->type
!= sec
->hba
->type
)) {
3021 "HBAs of devices does not match %s != %s\n",
3022 first
->hba
? get_sys_dev_type(first
->hba
->type
) : NULL
,
3023 sec
->hba
? get_sys_dev_type(sec
->hba
->type
) : NULL
);
3028 /* if an anchor does not have num_raid_devs set then it is a free
3031 if (first
->anchor
->num_raid_devs
> 0 &&
3032 sec
->anchor
->num_raid_devs
> 0) {
3033 /* Determine if these disks might ever have been
3034 * related. Further disambiguation can only take place
3035 * in load_super_imsm_all
3037 __u32 first_family
= first
->anchor
->orig_family_num
;
3038 __u32 sec_family
= sec
->anchor
->orig_family_num
;
3040 if (memcmp(first
->anchor
->sig
, sec
->anchor
->sig
,
3041 MAX_SIGNATURE_LENGTH
) != 0)
3044 if (first_family
== 0)
3045 first_family
= first
->anchor
->family_num
;
3046 if (sec_family
== 0)
3047 sec_family
= sec
->anchor
->family_num
;
3049 if (first_family
!= sec_family
)
3055 /* if 'first' is a spare promote it to a populated mpb with sec's
3058 if (first
->anchor
->num_raid_devs
== 0 &&
3059 sec
->anchor
->num_raid_devs
> 0) {
3061 struct intel_dev
*dv
;
3062 struct imsm_dev
*dev
;
3064 /* we need to copy raid device info from sec if an allocation
3065 * fails here we don't associate the spare
3067 for (i
= 0; i
< sec
->anchor
->num_raid_devs
; i
++) {
3068 dv
= malloc(sizeof(*dv
));
3071 dev
= malloc(sizeof_imsm_dev(get_imsm_dev(sec
, i
), 1));
3078 dv
->next
= first
->devlist
;
3079 first
->devlist
= dv
;
3081 if (i
< sec
->anchor
->num_raid_devs
) {
3082 /* allocation failure */
3083 free_devlist(first
);
3084 fprintf(stderr
, "imsm: failed to associate spare\n");
3087 first
->anchor
->num_raid_devs
= sec
->anchor
->num_raid_devs
;
3088 first
->anchor
->orig_family_num
= sec
->anchor
->orig_family_num
;
3089 first
->anchor
->family_num
= sec
->anchor
->family_num
;
3090 memcpy(first
->anchor
->sig
, sec
->anchor
->sig
, MAX_SIGNATURE_LENGTH
);
3091 for (i
= 0; i
< sec
->anchor
->num_raid_devs
; i
++)
3092 imsm_copy_dev(get_imsm_dev(first
, i
), get_imsm_dev(sec
, i
));
3098 static void fd2devname(int fd
, char *name
)
3102 char dname
[PATH_MAX
];
3107 if (fstat(fd
, &st
) != 0)
3109 sprintf(path
, "/sys/dev/block/%d:%d",
3110 major(st
.st_rdev
), minor(st
.st_rdev
));
3112 rv
= readlink(path
, dname
, sizeof(dname
)-1);
3117 nm
= strrchr(dname
, '/');
3120 snprintf(name
, MAX_RAID_SERIAL_LEN
, "/dev/%s", nm
);
3124 extern int scsi_get_serial(int fd
, void *buf
, size_t buf_len
);
3126 static int imsm_read_serial(int fd
, char *devname
,
3127 __u8 serial
[MAX_RAID_SERIAL_LEN
])
3129 unsigned char scsi_serial
[255];
3138 memset(scsi_serial
, 0, sizeof(scsi_serial
));
3140 rv
= scsi_get_serial(fd
, scsi_serial
, sizeof(scsi_serial
));
3142 if (rv
&& check_env("IMSM_DEVNAME_AS_SERIAL")) {
3143 memset(serial
, 0, MAX_RAID_SERIAL_LEN
);
3144 fd2devname(fd
, (char *) serial
);
3151 Name
": Failed to retrieve serial for %s\n",
3156 rsp_len
= scsi_serial
[3];
3160 Name
": Failed to retrieve serial for %s\n",
3164 rsp_buf
= (char *) &scsi_serial
[4];
3166 /* trim all whitespace and non-printable characters and convert
3169 for (i
= 0, dest
= rsp_buf
; i
< rsp_len
; i
++) {
3172 /* ':' is reserved for use in placeholder serial
3173 * numbers for missing disks
3181 len
= dest
- rsp_buf
;
3184 /* truncate leading characters */
3185 if (len
> MAX_RAID_SERIAL_LEN
) {
3186 dest
+= len
- MAX_RAID_SERIAL_LEN
;
3187 len
= MAX_RAID_SERIAL_LEN
;
3190 memset(serial
, 0, MAX_RAID_SERIAL_LEN
);
3191 memcpy(serial
, dest
, len
);
3196 static int serialcmp(__u8
*s1
, __u8
*s2
)
3198 return strncmp((char *) s1
, (char *) s2
, MAX_RAID_SERIAL_LEN
);
3201 static void serialcpy(__u8
*dest
, __u8
*src
)
3203 strncpy((char *) dest
, (char *) src
, MAX_RAID_SERIAL_LEN
);
3206 static struct dl
*serial_to_dl(__u8
*serial
, struct intel_super
*super
)
3210 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
3211 if (serialcmp(dl
->serial
, serial
) == 0)
3217 static struct imsm_disk
*
3218 __serial_to_disk(__u8
*serial
, struct imsm_super
*mpb
, int *idx
)
3222 for (i
= 0; i
< mpb
->num_disks
; i
++) {
3223 struct imsm_disk
*disk
= __get_imsm_disk(mpb
, i
);
3225 if (serialcmp(disk
->serial
, serial
) == 0) {
3236 load_imsm_disk(int fd
, struct intel_super
*super
, char *devname
, int keep_fd
)
3238 struct imsm_disk
*disk
;
3243 __u8 serial
[MAX_RAID_SERIAL_LEN
];
3245 rv
= imsm_read_serial(fd
, devname
, serial
);
3250 dl
= calloc(1, sizeof(*dl
));
3254 Name
": failed to allocate disk buffer for %s\n",
3260 dl
->major
= major(stb
.st_rdev
);
3261 dl
->minor
= minor(stb
.st_rdev
);
3262 dl
->next
= super
->disks
;
3263 dl
->fd
= keep_fd
? fd
: -1;
3264 assert(super
->disks
== NULL
);
3266 serialcpy(dl
->serial
, serial
);
3269 fd2devname(fd
, name
);
3271 dl
->devname
= strdup(devname
);
3273 dl
->devname
= strdup(name
);
3275 /* look up this disk's index in the current anchor */
3276 disk
= __serial_to_disk(dl
->serial
, super
->anchor
, &dl
->index
);
3279 /* only set index on disks that are a member of a
3280 * populated contianer, i.e. one with raid_devs
3282 if (is_failed(&dl
->disk
))
3284 else if (is_spare(&dl
->disk
))
3292 /* When migrating map0 contains the 'destination' state while map1
3293 * contains the current state. When not migrating map0 contains the
3294 * current state. This routine assumes that map[0].map_state is set to
3295 * the current array state before being called.
3297 * Migration is indicated by one of the following states
3298 * 1/ Idle (migr_state=0 map0state=normal||unitialized||degraded||failed)
3299 * 2/ Initialize (migr_state=1 migr_type=MIGR_INIT map0state=normal
3300 * map1state=unitialized)
3301 * 3/ Repair (Resync) (migr_state=1 migr_type=MIGR_REPAIR map0state=normal
3303 * 4/ Rebuild (migr_state=1 migr_type=MIGR_REBUILD map0state=normal
3304 * map1state=degraded)
3305 * 5/ Migration (mig_state=1 migr_type=MIGR_GEN_MIGR map0state=normal
3308 static void migrate(struct imsm_dev
*dev
, struct intel_super
*super
,
3309 __u8 to_state
, int migr_type
)
3311 struct imsm_map
*dest
;
3312 struct imsm_map
*src
= get_imsm_map(dev
, MAP_0
);
3314 dev
->vol
.migr_state
= 1;
3315 set_migr_type(dev
, migr_type
);
3316 dev
->vol
.curr_migr_unit
= 0;
3317 dest
= get_imsm_map(dev
, MAP_1
);
3319 /* duplicate and then set the target end state in map[0] */
3320 memcpy(dest
, src
, sizeof_imsm_map(src
));
3321 if ((migr_type
== MIGR_REBUILD
) ||
3322 (migr_type
== MIGR_GEN_MIGR
)) {
3326 for (i
= 0; i
< src
->num_members
; i
++) {
3327 ord
= __le32_to_cpu(src
->disk_ord_tbl
[i
]);
3328 set_imsm_ord_tbl_ent(src
, i
, ord_to_idx(ord
));
3332 if (migr_type
== MIGR_GEN_MIGR
)
3333 /* Clear migration record */
3334 memset(super
->migr_rec
, 0, sizeof(struct migr_record
));
3336 src
->map_state
= to_state
;
3339 static void end_migration(struct imsm_dev
*dev
, struct intel_super
*super
,
3342 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
3343 struct imsm_map
*prev
= get_imsm_map(dev
, dev
->vol
.migr_state
== 0 ?
3347 /* merge any IMSM_ORD_REBUILD bits that were not successfully
3348 * completed in the last migration.
3350 * FIXME add support for raid-level-migration
3352 if ((map_state
!= map
->map_state
) && (is_gen_migration(dev
) == 0) &&
3353 (prev
->map_state
!= IMSM_T_STATE_UNINITIALIZED
)) {
3354 /* when final map state is other than expected
3355 * merge maps (not for migration)
3359 for (i
= 0; i
< prev
->num_members
; i
++)
3360 for (j
= 0; j
< map
->num_members
; j
++)
3361 /* during online capacity expansion
3362 * disks position can be changed
3363 * if takeover is used
3365 if (ord_to_idx(map
->disk_ord_tbl
[j
]) ==
3366 ord_to_idx(prev
->disk_ord_tbl
[i
])) {
3367 map
->disk_ord_tbl
[j
] |=
3368 prev
->disk_ord_tbl
[i
];
3371 failed
= imsm_count_failed(super
, dev
, MAP_0
);
3372 map_state
= imsm_check_degraded(super
, dev
, failed
, MAP_0
);
3375 dev
->vol
.migr_state
= 0;
3376 set_migr_type(dev
, 0);
3377 dev
->vol
.curr_migr_unit
= 0;
3378 map
->map_state
= map_state
;
3382 static int parse_raid_devices(struct intel_super
*super
)
3385 struct imsm_dev
*dev_new
;
3386 size_t len
, len_migr
;
3388 size_t space_needed
= 0;
3389 struct imsm_super
*mpb
= super
->anchor
;
3391 for (i
= 0; i
< super
->anchor
->num_raid_devs
; i
++) {
3392 struct imsm_dev
*dev_iter
= __get_imsm_dev(super
->anchor
, i
);
3393 struct intel_dev
*dv
;
3395 len
= sizeof_imsm_dev(dev_iter
, 0);
3396 len_migr
= sizeof_imsm_dev(dev_iter
, 1);
3398 space_needed
+= len_migr
- len
;
3400 dv
= malloc(sizeof(*dv
));
3403 if (max_len
< len_migr
)
3405 if (max_len
> len_migr
)
3406 space_needed
+= max_len
- len_migr
;
3407 dev_new
= malloc(max_len
);
3412 imsm_copy_dev(dev_new
, dev_iter
);
3415 dv
->next
= super
->devlist
;
3416 super
->devlist
= dv
;
3419 /* ensure that super->buf is large enough when all raid devices
3422 if (__le32_to_cpu(mpb
->mpb_size
) + space_needed
> super
->len
) {
3425 len
= ROUND_UP(__le32_to_cpu(mpb
->mpb_size
) + space_needed
, 512);
3426 if (posix_memalign(&buf
, 512, len
) != 0)
3429 memcpy(buf
, super
->buf
, super
->len
);
3430 memset(buf
+ super
->len
, 0, len
- super
->len
);
3439 /* retrieve a pointer to the bbm log which starts after all raid devices */
3440 struct bbm_log
*__get_imsm_bbm_log(struct imsm_super
*mpb
)
3444 if (__le32_to_cpu(mpb
->bbm_log_size
)) {
3446 ptr
+= mpb
->mpb_size
- __le32_to_cpu(mpb
->bbm_log_size
);
3452 /*******************************************************************************
3453 * Function: check_mpb_migr_compatibility
3454 * Description: Function checks for unsupported migration features:
3455 * - migration optimization area (pba_of_lba0)
3456 * - descending reshape (ascending_migr)
3458 * super : imsm metadata information
3460 * 0 : migration is compatible
3461 * -1 : migration is not compatible
3462 ******************************************************************************/
3463 int check_mpb_migr_compatibility(struct intel_super
*super
)
3465 struct imsm_map
*map0
, *map1
;
3466 struct migr_record
*migr_rec
= super
->migr_rec
;
3469 for (i
= 0; i
< super
->anchor
->num_raid_devs
; i
++) {
3470 struct imsm_dev
*dev_iter
= __get_imsm_dev(super
->anchor
, i
);
3473 dev_iter
->vol
.migr_state
== 1 &&
3474 dev_iter
->vol
.migr_type
== MIGR_GEN_MIGR
) {
3475 /* This device is migrating */
3476 map0
= get_imsm_map(dev_iter
, MAP_0
);
3477 map1
= get_imsm_map(dev_iter
, MAP_1
);
3478 if (pba_of_lba0(map0
) != pba_of_lba0(map1
))
3479 /* migration optimization area was used */
3481 if (migr_rec
->ascending_migr
== 0
3482 && migr_rec
->dest_depth_per_unit
> 0)
3483 /* descending reshape not supported yet */
3490 static void __free_imsm(struct intel_super
*super
, int free_disks
);
3492 /* load_imsm_mpb - read matrix metadata
3493 * allocates super->mpb to be freed by free_imsm
3495 static int load_imsm_mpb(int fd
, struct intel_super
*super
, char *devname
)
3497 unsigned long long dsize
;
3498 unsigned long long sectors
;
3500 struct imsm_super
*anchor
;
3503 get_dev_size(fd
, NULL
, &dsize
);
3507 Name
": %s: device to small for imsm\n",
3512 if (lseek64(fd
, dsize
- (512 * 2), SEEK_SET
) < 0) {
3514 fprintf(stderr
, Name
3515 ": Cannot seek to anchor block on %s: %s\n",
3516 devname
, strerror(errno
));
3520 if (posix_memalign((void**)&anchor
, 512, 512) != 0) {
3523 Name
": Failed to allocate imsm anchor buffer"
3524 " on %s\n", devname
);
3527 if (read(fd
, anchor
, 512) != 512) {
3530 Name
": Cannot read anchor block on %s: %s\n",
3531 devname
, strerror(errno
));
3536 if (strncmp((char *) anchor
->sig
, MPB_SIGNATURE
, MPB_SIG_LEN
) != 0) {
3539 Name
": no IMSM anchor on %s\n", devname
);
3544 __free_imsm(super
, 0);
3545 /* reload capability and hba */
3547 /* capability and hba must be updated with new super allocation */
3548 find_intel_hba_capability(fd
, super
, devname
);
3549 super
->len
= ROUND_UP(anchor
->mpb_size
, 512);
3550 if (posix_memalign(&super
->buf
, 512, super
->len
) != 0) {
3553 Name
": unable to allocate %zu byte mpb buffer\n",
3558 memcpy(super
->buf
, anchor
, 512);
3560 sectors
= mpb_sectors(anchor
) - 1;
3563 if (posix_memalign(&super
->migr_rec_buf
, 512, MIGR_REC_BUF_SIZE
) != 0) {
3564 fprintf(stderr
, Name
3565 ": %s could not allocate migr_rec buffer\n", __func__
);
3569 super
->clean_migration_record_by_mdmon
= 0;
3572 check_sum
= __gen_imsm_checksum(super
->anchor
);
3573 if (check_sum
!= __le32_to_cpu(super
->anchor
->check_sum
)) {
3576 Name
": IMSM checksum %x != %x on %s\n",
3578 __le32_to_cpu(super
->anchor
->check_sum
),
3586 /* read the extended mpb */
3587 if (lseek64(fd
, dsize
- (512 * (2 + sectors
)), SEEK_SET
) < 0) {
3590 Name
": Cannot seek to extended mpb on %s: %s\n",
3591 devname
, strerror(errno
));
3595 if ((unsigned)read(fd
, super
->buf
+ 512, super
->len
- 512) != super
->len
- 512) {
3598 Name
": Cannot read extended mpb on %s: %s\n",
3599 devname
, strerror(errno
));
3603 check_sum
= __gen_imsm_checksum(super
->anchor
);
3604 if (check_sum
!= __le32_to_cpu(super
->anchor
->check_sum
)) {
3607 Name
": IMSM checksum %x != %x on %s\n",
3608 check_sum
, __le32_to_cpu(super
->anchor
->check_sum
),
3613 /* FIXME the BBM log is disk specific so we cannot use this global
3614 * buffer for all disks. Ok for now since we only look at the global
3615 * bbm_log_size parameter to gate assembly
3617 super
->bbm_log
= __get_imsm_bbm_log(super
->anchor
);
3622 static int read_imsm_migr_rec(int fd
, struct intel_super
*super
);
3624 /* clears hi bits in metadata if MPB_ATTRIB_2TB_DISK not set */
3625 static void clear_hi(struct intel_super
*super
)
3627 struct imsm_super
*mpb
= super
->anchor
;
3629 if (mpb
->attributes
& MPB_ATTRIB_2TB_DISK
)
3631 for (i
= 0; i
< mpb
->num_disks
; ++i
) {
3632 struct imsm_disk
*disk
= &mpb
->disk
[i
];
3633 disk
->total_blocks_hi
= 0;
3635 for (i
= 0; i
< mpb
->num_raid_devs
; ++i
) {
3636 struct imsm_dev
*dev
= get_imsm_dev(super
, i
);
3639 for (n
= 0; n
< 2; ++n
) {
3640 struct imsm_map
*map
= get_imsm_map(dev
, n
);
3643 map
->pba_of_lba0_hi
= 0;
3644 map
->blocks_per_member_hi
= 0;
3645 map
->num_data_stripes_hi
= 0;
3651 load_and_parse_mpb(int fd
, struct intel_super
*super
, char *devname
, int keep_fd
)
3655 err
= load_imsm_mpb(fd
, super
, devname
);
3658 err
= load_imsm_disk(fd
, super
, devname
, keep_fd
);
3661 err
= parse_raid_devices(super
);
3666 static void __free_imsm_disk(struct dl
*d
)
3678 static void free_imsm_disks(struct intel_super
*super
)
3682 while (super
->disks
) {
3684 super
->disks
= d
->next
;
3685 __free_imsm_disk(d
);
3687 while (super
->disk_mgmt_list
) {
3688 d
= super
->disk_mgmt_list
;
3689 super
->disk_mgmt_list
= d
->next
;
3690 __free_imsm_disk(d
);
3692 while (super
->missing
) {
3694 super
->missing
= d
->next
;
3695 __free_imsm_disk(d
);
3700 /* free all the pieces hanging off of a super pointer */
3701 static void __free_imsm(struct intel_super
*super
, int free_disks
)
3703 struct intel_hba
*elem
, *next
;
3709 /* unlink capability description */
3711 if (super
->migr_rec_buf
) {
3712 free(super
->migr_rec_buf
);
3713 super
->migr_rec_buf
= NULL
;
3716 free_imsm_disks(super
);
3717 free_devlist(super
);
3721 free((void *)elem
->path
);
3729 static void free_imsm(struct intel_super
*super
)
3731 __free_imsm(super
, 1);
3735 static void free_super_imsm(struct supertype
*st
)
3737 struct intel_super
*super
= st
->sb
;
3746 static struct intel_super
*alloc_super(void)
3748 struct intel_super
*super
= malloc(sizeof(*super
));
3751 memset(super
, 0, sizeof(*super
));
3752 super
->current_vol
= -1;
3753 super
->create_offset
= ~((unsigned long long) 0);
3759 * find and allocate hba and OROM/EFI based on valid fd of RAID component device
3761 static int find_intel_hba_capability(int fd
, struct intel_super
*super
, char *devname
)
3763 struct sys_dev
*hba_name
;
3766 if ((fd
< 0) || check_env("IMSM_NO_PLATFORM")) {
3771 hba_name
= find_disk_attached_hba(fd
, NULL
);
3775 Name
": %s is not attached to Intel(R) RAID controller.\n",
3779 rv
= attach_hba_to_super(super
, hba_name
);
3782 struct intel_hba
*hba
= super
->hba
;
3784 fprintf(stderr
, Name
": %s is attached to Intel(R) %s RAID "
3785 "controller (%s),\n"
3786 " but the container is assigned to Intel(R) "
3787 "%s RAID controller (",
3790 hba_name
->pci_id
? : "Err!",
3791 get_sys_dev_type(hba_name
->type
));
3794 fprintf(stderr
, "%s", hba
->pci_id
? : "Err!");
3796 fprintf(stderr
, ", ");
3800 fprintf(stderr
, ").\n"
3801 " Mixing devices attached to different controllers "
3802 "is not allowed.\n");
3804 free_sys_dev(&hba_name
);
3807 super
->orom
= find_imsm_capability(hba_name
->type
);
3808 free_sys_dev(&hba_name
);
3814 /* find_missing - helper routine for load_super_imsm_all that identifies
3815 * disks that have disappeared from the system. This routine relies on
3816 * the mpb being uptodate, which it is at load time.
3818 static int find_missing(struct intel_super
*super
)
3821 struct imsm_super
*mpb
= super
->anchor
;
3823 struct imsm_disk
*disk
;
3825 for (i
= 0; i
< mpb
->num_disks
; i
++) {
3826 disk
= __get_imsm_disk(mpb
, i
);
3827 dl
= serial_to_dl(disk
->serial
, super
);
3831 dl
= malloc(sizeof(*dl
));
3837 dl
->devname
= strdup("missing");
3839 serialcpy(dl
->serial
, disk
->serial
);
3842 dl
->next
= super
->missing
;
3843 super
->missing
= dl
;
3850 static struct intel_disk
*disk_list_get(__u8
*serial
, struct intel_disk
*disk_list
)
3852 struct intel_disk
*idisk
= disk_list
;
3855 if (serialcmp(idisk
->disk
.serial
, serial
) == 0)
3857 idisk
= idisk
->next
;
3863 static int __prep_thunderdome(struct intel_super
**table
, int tbl_size
,
3864 struct intel_super
*super
,
3865 struct intel_disk
**disk_list
)
3867 struct imsm_disk
*d
= &super
->disks
->disk
;
3868 struct imsm_super
*mpb
= super
->anchor
;
3871 for (i
= 0; i
< tbl_size
; i
++) {
3872 struct imsm_super
*tbl_mpb
= table
[i
]->anchor
;
3873 struct imsm_disk
*tbl_d
= &table
[i
]->disks
->disk
;
3875 if (tbl_mpb
->family_num
== mpb
->family_num
) {
3876 if (tbl_mpb
->check_sum
== mpb
->check_sum
) {
3877 dprintf("%s: mpb from %d:%d matches %d:%d\n",
3878 __func__
, super
->disks
->major
,
3879 super
->disks
->minor
,
3880 table
[i
]->disks
->major
,
3881 table
[i
]->disks
->minor
);
3885 if (((is_configured(d
) && !is_configured(tbl_d
)) ||
3886 is_configured(d
) == is_configured(tbl_d
)) &&
3887 tbl_mpb
->generation_num
< mpb
->generation_num
) {
3888 /* current version of the mpb is a
3889 * better candidate than the one in
3890 * super_table, but copy over "cross
3891 * generational" status
3893 struct intel_disk
*idisk
;
3895 dprintf("%s: mpb from %d:%d replaces %d:%d\n",
3896 __func__
, super
->disks
->major
,
3897 super
->disks
->minor
,
3898 table
[i
]->disks
->major
,
3899 table
[i
]->disks
->minor
);
3901 idisk
= disk_list_get(tbl_d
->serial
, *disk_list
);
3902 if (idisk
&& is_failed(&idisk
->disk
))
3903 tbl_d
->status
|= FAILED_DISK
;
3906 struct intel_disk
*idisk
;
3907 struct imsm_disk
*disk
;
3909 /* tbl_mpb is more up to date, but copy
3910 * over cross generational status before
3913 disk
= __serial_to_disk(d
->serial
, mpb
, NULL
);
3914 if (disk
&& is_failed(disk
))
3915 d
->status
|= FAILED_DISK
;
3917 idisk
= disk_list_get(d
->serial
, *disk_list
);
3920 if (disk
&& is_configured(disk
))
3921 idisk
->disk
.status
|= CONFIGURED_DISK
;
3924 dprintf("%s: mpb from %d:%d prefer %d:%d\n",
3925 __func__
, super
->disks
->major
,
3926 super
->disks
->minor
,
3927 table
[i
]->disks
->major
,
3928 table
[i
]->disks
->minor
);
3936 table
[tbl_size
++] = super
;
3940 /* update/extend the merged list of imsm_disk records */
3941 for (j
= 0; j
< mpb
->num_disks
; j
++) {
3942 struct imsm_disk
*disk
= __get_imsm_disk(mpb
, j
);
3943 struct intel_disk
*idisk
;
3945 idisk
= disk_list_get(disk
->serial
, *disk_list
);
3947 idisk
->disk
.status
|= disk
->status
;
3948 if (is_configured(&idisk
->disk
) ||
3949 is_failed(&idisk
->disk
))
3950 idisk
->disk
.status
&= ~(SPARE_DISK
);
3952 idisk
= calloc(1, sizeof(*idisk
));
3955 idisk
->owner
= IMSM_UNKNOWN_OWNER
;
3956 idisk
->disk
= *disk
;
3957 idisk
->next
= *disk_list
;
3961 if (serialcmp(idisk
->disk
.serial
, d
->serial
) == 0)
3968 static struct intel_super
*
3969 validate_members(struct intel_super
*super
, struct intel_disk
*disk_list
,
3972 struct imsm_super
*mpb
= super
->anchor
;
3976 for (i
= 0; i
< mpb
->num_disks
; i
++) {
3977 struct imsm_disk
*disk
= __get_imsm_disk(mpb
, i
);
3978 struct intel_disk
*idisk
;
3980 idisk
= disk_list_get(disk
->serial
, disk_list
);
3982 if (idisk
->owner
== owner
||
3983 idisk
->owner
== IMSM_UNKNOWN_OWNER
)
3986 dprintf("%s: '%.16s' owner %d != %d\n",
3987 __func__
, disk
->serial
, idisk
->owner
,
3990 dprintf("%s: unknown disk %x [%d]: %.16s\n",
3991 __func__
, __le32_to_cpu(mpb
->family_num
), i
,
3997 if (ok_count
== mpb
->num_disks
)
4002 static void show_conflicts(__u32 family_num
, struct intel_super
*super_list
)
4004 struct intel_super
*s
;
4006 for (s
= super_list
; s
; s
= s
->next
) {
4007 if (family_num
!= s
->anchor
->family_num
)
4009 fprintf(stderr
, "Conflict, offlining family %#x on '%s'\n",
4010 __le32_to_cpu(family_num
), s
->disks
->devname
);
4014 static struct intel_super
*
4015 imsm_thunderdome(struct intel_super
**super_list
, int len
)
4017 struct intel_super
*super_table
[len
];
4018 struct intel_disk
*disk_list
= NULL
;
4019 struct intel_super
*champion
, *spare
;
4020 struct intel_super
*s
, **del
;
4025 memset(super_table
, 0, sizeof(super_table
));
4026 for (s
= *super_list
; s
; s
= s
->next
)
4027 tbl_size
= __prep_thunderdome(super_table
, tbl_size
, s
, &disk_list
);
4029 for (i
= 0; i
< tbl_size
; i
++) {
4030 struct imsm_disk
*d
;
4031 struct intel_disk
*idisk
;
4032 struct imsm_super
*mpb
= super_table
[i
]->anchor
;
4035 d
= &s
->disks
->disk
;
4037 /* 'd' must appear in merged disk list for its
4038 * configuration to be valid
4040 idisk
= disk_list_get(d
->serial
, disk_list
);
4041 if (idisk
&& idisk
->owner
== i
)
4042 s
= validate_members(s
, disk_list
, i
);
4047 dprintf("%s: marking family: %#x from %d:%d offline\n",
4048 __func__
, mpb
->family_num
,
4049 super_table
[i
]->disks
->major
,
4050 super_table
[i
]->disks
->minor
);
4054 /* This is where the mdadm implementation differs from the Windows
4055 * driver which has no strict concept of a container. We can only
4056 * assemble one family from a container, so when returning a prodigal
4057 * array member to this system the code will not be able to disambiguate
4058 * the container contents that should be assembled ("foreign" versus
4059 * "local"). It requires user intervention to set the orig_family_num
4060 * to a new value to establish a new container. The Windows driver in
4061 * this situation fixes up the volume name in place and manages the
4062 * foreign array as an independent entity.
4067 for (i
= 0; i
< tbl_size
; i
++) {
4068 struct intel_super
*tbl_ent
= super_table
[i
];
4074 if (tbl_ent
->anchor
->num_raid_devs
== 0) {
4079 if (s
&& !is_spare
) {
4080 show_conflicts(tbl_ent
->anchor
->family_num
, *super_list
);
4082 } else if (!s
&& !is_spare
)
4095 fprintf(stderr
, "Chose family %#x on '%s', "
4096 "assemble conflicts to new container with '--update=uuid'\n",
4097 __le32_to_cpu(s
->anchor
->family_num
), s
->disks
->devname
);
4099 /* collect all dl's onto 'champion', and update them to
4100 * champion's version of the status
4102 for (s
= *super_list
; s
; s
= s
->next
) {
4103 struct imsm_super
*mpb
= champion
->anchor
;
4104 struct dl
*dl
= s
->disks
;
4109 for (i
= 0; i
< mpb
->num_disks
; i
++) {
4110 struct imsm_disk
*disk
;
4112 disk
= __serial_to_disk(dl
->serial
, mpb
, &dl
->index
);
4115 /* only set index on disks that are a member of
4116 * a populated contianer, i.e. one with
4119 if (is_failed(&dl
->disk
))
4121 else if (is_spare(&dl
->disk
))
4127 if (i
>= mpb
->num_disks
) {
4128 struct intel_disk
*idisk
;
4130 idisk
= disk_list_get(dl
->serial
, disk_list
);
4131 if (idisk
&& is_spare(&idisk
->disk
) &&
4132 !is_failed(&idisk
->disk
) && !is_configured(&idisk
->disk
))
4140 dl
->next
= champion
->disks
;
4141 champion
->disks
= dl
;
4145 /* delete 'champion' from super_list */
4146 for (del
= super_list
; *del
; ) {
4147 if (*del
== champion
) {
4148 *del
= (*del
)->next
;
4151 del
= &(*del
)->next
;
4153 champion
->next
= NULL
;
4157 struct intel_disk
*idisk
= disk_list
;
4159 disk_list
= disk_list
->next
;
4168 get_sra_super_block(int fd
, struct intel_super
**super_list
, char *devname
, int *max
, int keep_fd
);
4169 static int get_super_block(struct intel_super
**super_list
, int devnum
, char *devname
,
4170 int major
, int minor
, int keep_fd
);
4172 get_devlist_super_block(struct md_list
*devlist
, struct intel_super
**super_list
,
4173 int *max
, int keep_fd
);
4176 static int load_super_imsm_all(struct supertype
*st
, int fd
, void **sbp
,
4177 char *devname
, struct md_list
*devlist
,
4180 struct intel_super
*super_list
= NULL
;
4181 struct intel_super
*super
= NULL
;
4186 /* 'fd' is an opened container */
4187 err
= get_sra_super_block(fd
, &super_list
, devname
, &i
, keep_fd
);
4189 /* get super block from devlist devices */
4190 err
= get_devlist_super_block(devlist
, &super_list
, &i
, keep_fd
);
4193 /* all mpbs enter, maybe one leaves */
4194 super
= imsm_thunderdome(&super_list
, i
);
4200 if (find_missing(super
) != 0) {
4206 /* load migration record */
4207 err
= load_imsm_migr_rec(super
, NULL
);
4209 /* migration is in progress,
4210 * but migr_rec cannot be loaded,
4216 /* Check migration compatibility */
4217 if ((err
== 0) && (check_mpb_migr_compatibility(super
) != 0)) {
4218 fprintf(stderr
, Name
": Unsupported migration detected");
4220 fprintf(stderr
, " on %s\n", devname
);
4222 fprintf(stderr
, " (IMSM).\n");
4231 while (super_list
) {
4232 struct intel_super
*s
= super_list
;
4234 super_list
= super_list
->next
;
4244 st
->container_dev
= fd2devnum(fd
);
4246 st
->container_dev
= NoMdDev
;
4247 if (err
== 0 && st
->ss
== NULL
) {
4248 st
->ss
= &super_imsm
;
4249 st
->minor_version
= 0;
4250 st
->max_devs
= IMSM_MAX_DEVICES
;
4257 get_devlist_super_block(struct md_list
*devlist
, struct intel_super
**super_list
,
4258 int *max
, int keep_fd
)
4260 struct md_list
*tmpdev
;
4264 for (i
= 0, tmpdev
= devlist
; tmpdev
; tmpdev
= tmpdev
->next
) {
4265 if (tmpdev
->used
!= 1)
4267 if (tmpdev
->container
== 1) {
4269 int fd
= dev_open(tmpdev
->devname
, O_RDONLY
|O_EXCL
);
4271 fprintf(stderr
, Name
": cannot open device %s: %s\n",
4272 tmpdev
->devname
, strerror(errno
));
4276 err
= get_sra_super_block(fd
, super_list
,
4277 tmpdev
->devname
, &lmax
,
4286 int major
= major(tmpdev
->st_rdev
);
4287 int minor
= minor(tmpdev
->st_rdev
);
4288 err
= get_super_block(super_list
,
4305 static int get_super_block(struct intel_super
**super_list
, int devnum
, char *devname
,
4306 int major
, int minor
, int keep_fd
)
4308 struct intel_super
*s
= NULL
;
4321 sprintf(nm
, "%d:%d", major
, minor
);
4322 dfd
= dev_open(nm
, O_RDWR
);
4328 rv
= find_intel_hba_capability(dfd
, s
, devname
);
4329 /* no orom/efi or non-intel hba of the disk */
4335 err
= load_and_parse_mpb(dfd
, s
, NULL
, keep_fd
);
4337 /* retry the load if we might have raced against mdmon */
4338 if (err
== 3 && (devnum
!= -1) && mdmon_running(devnum
))
4339 for (retry
= 0; retry
< 3; retry
++) {
4341 err
= load_and_parse_mpb(dfd
, s
, NULL
, keep_fd
);
4347 s
->next
= *super_list
;
4355 if ((dfd
>= 0) && (!keep_fd
))
4362 get_sra_super_block(int fd
, struct intel_super
**super_list
, char *devname
, int *max
, int keep_fd
)
4369 sra
= sysfs_read(fd
, 0, GET_LEVEL
|GET_VERSION
|GET_DEVS
|GET_STATE
);
4373 if (sra
->array
.major_version
!= -1 ||
4374 sra
->array
.minor_version
!= -2 ||
4375 strcmp(sra
->text_version
, "imsm") != 0) {
4380 devnum
= fd2devnum(fd
);
4381 for (sd
= sra
->devs
, i
= 0; sd
; sd
= sd
->next
, i
++) {
4382 if (get_super_block(super_list
, devnum
, devname
,
4383 sd
->disk
.major
, sd
->disk
.minor
, keep_fd
) != 0) {
4394 static int load_container_imsm(struct supertype
*st
, int fd
, char *devname
)
4396 return load_super_imsm_all(st
, fd
, &st
->sb
, devname
, NULL
, 1);
4400 static int load_super_imsm(struct supertype
*st
, int fd
, char *devname
)
4402 struct intel_super
*super
;
4405 if (test_partition(fd
))
4406 /* IMSM not allowed on partitions */
4409 free_super_imsm(st
);
4411 super
= alloc_super();
4414 Name
": malloc of %zu failed.\n",
4418 /* Load hba and capabilities if they exist.
4419 * But do not preclude loading metadata in case capabilities or hba are
4420 * non-compliant and ignore_hw_compat is set.
4422 rv
= find_intel_hba_capability(fd
, super
, devname
);
4423 /* no orom/efi or non-intel hba of the disk */
4424 if ((rv
!= 0) && (st
->ignore_hw_compat
== 0)) {
4427 Name
": No OROM/EFI properties for %s\n", devname
);
4431 rv
= load_and_parse_mpb(fd
, super
, devname
, 0);
4436 Name
": Failed to load all information "
4437 "sections on %s\n", devname
);
4443 if (st
->ss
== NULL
) {
4444 st
->ss
= &super_imsm
;
4445 st
->minor_version
= 0;
4446 st
->max_devs
= IMSM_MAX_DEVICES
;
4449 /* load migration record */
4450 if (load_imsm_migr_rec(super
, NULL
) == 0) {
4451 /* Check for unsupported migration features */
4452 if (check_mpb_migr_compatibility(super
) != 0) {
4454 Name
": Unsupported migration detected");
4456 fprintf(stderr
, " on %s\n", devname
);
4458 fprintf(stderr
, " (IMSM).\n");
4466 static __u16
info_to_blocks_per_strip(mdu_array_info_t
*info
)
4468 if (info
->level
== 1)
4470 return info
->chunk_size
>> 9;
4473 static unsigned long long info_to_blocks_per_member(mdu_array_info_t
*info
,
4474 unsigned long long size
)
4476 if (info
->level
== 1)
4479 return (size
* 2) & ~(info_to_blocks_per_strip(info
) - 1);
4482 static void imsm_update_version_info(struct intel_super
*super
)
4484 /* update the version and attributes */
4485 struct imsm_super
*mpb
= super
->anchor
;
4487 struct imsm_dev
*dev
;
4488 struct imsm_map
*map
;
4491 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
4492 dev
= get_imsm_dev(super
, i
);
4493 map
= get_imsm_map(dev
, MAP_0
);
4494 if (__le32_to_cpu(dev
->size_high
) > 0)
4495 mpb
->attributes
|= MPB_ATTRIB_2TB
;
4497 /* FIXME detect when an array spans a port multiplier */
4499 mpb
->attributes
|= MPB_ATTRIB_PM
;
4502 if (mpb
->num_raid_devs
> 1 ||
4503 mpb
->attributes
!= MPB_ATTRIB_CHECKSUM_VERIFY
) {
4504 version
= MPB_VERSION_ATTRIBS
;
4505 switch (get_imsm_raid_level(map
)) {
4506 case 0: mpb
->attributes
|= MPB_ATTRIB_RAID0
; break;
4507 case 1: mpb
->attributes
|= MPB_ATTRIB_RAID1
; break;
4508 case 10: mpb
->attributes
|= MPB_ATTRIB_RAID10
; break;
4509 case 5: mpb
->attributes
|= MPB_ATTRIB_RAID5
; break;
4512 if (map
->num_members
>= 5)
4513 version
= MPB_VERSION_5OR6_DISK_ARRAY
;
4514 else if (dev
->status
== DEV_CLONE_N_GO
)
4515 version
= MPB_VERSION_CNG
;
4516 else if (get_imsm_raid_level(map
) == 5)
4517 version
= MPB_VERSION_RAID5
;
4518 else if (map
->num_members
>= 3)
4519 version
= MPB_VERSION_3OR4_DISK_ARRAY
;
4520 else if (get_imsm_raid_level(map
) == 1)
4521 version
= MPB_VERSION_RAID1
;
4523 version
= MPB_VERSION_RAID0
;
4525 strcpy(((char *) mpb
->sig
) + strlen(MPB_SIGNATURE
), version
);
4529 static int check_name(struct intel_super
*super
, char *name
, int quiet
)
4531 struct imsm_super
*mpb
= super
->anchor
;
4532 char *reason
= NULL
;
4535 if (strlen(name
) > MAX_RAID_SERIAL_LEN
)
4536 reason
= "must be 16 characters or less";
4538 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
4539 struct imsm_dev
*dev
= get_imsm_dev(super
, i
);
4541 if (strncmp((char *) dev
->volume
, name
, MAX_RAID_SERIAL_LEN
) == 0) {
4542 reason
= "already exists";
4547 if (reason
&& !quiet
)
4548 fprintf(stderr
, Name
": imsm volume name %s\n", reason
);
4553 static int init_super_imsm_volume(struct supertype
*st
, mdu_array_info_t
*info
,
4554 unsigned long long size
, char *name
,
4555 char *homehost
, int *uuid
)
4557 /* We are creating a volume inside a pre-existing container.
4558 * so st->sb is already set.
4560 struct intel_super
*super
= st
->sb
;
4561 struct imsm_super
*mpb
= super
->anchor
;
4562 struct intel_dev
*dv
;
4563 struct imsm_dev
*dev
;
4564 struct imsm_vol
*vol
;
4565 struct imsm_map
*map
;
4566 int idx
= mpb
->num_raid_devs
;
4568 unsigned long long array_blocks
;
4569 size_t size_old
, size_new
;
4570 unsigned long long num_data_stripes
;
4572 if (super
->orom
&& mpb
->num_raid_devs
>= super
->orom
->vpa
) {
4573 fprintf(stderr
, Name
": This imsm-container already has the "
4574 "maximum of %d volumes\n", super
->orom
->vpa
);
4578 /* ensure the mpb is large enough for the new data */
4579 size_old
= __le32_to_cpu(mpb
->mpb_size
);
4580 size_new
= disks_to_mpb_size(info
->nr_disks
);
4581 if (size_new
> size_old
) {
4583 size_t size_round
= ROUND_UP(size_new
, 512);
4585 if (posix_memalign(&mpb_new
, 512, size_round
) != 0) {
4586 fprintf(stderr
, Name
": could not allocate new mpb\n");
4589 if (posix_memalign(&super
->migr_rec_buf
, 512,
4590 MIGR_REC_BUF_SIZE
) != 0) {
4591 fprintf(stderr
, Name
4592 ": %s could not allocate migr_rec buffer\n",
4599 memcpy(mpb_new
, mpb
, size_old
);
4602 super
->anchor
= mpb_new
;
4603 mpb
->mpb_size
= __cpu_to_le32(size_new
);
4604 memset(mpb_new
+ size_old
, 0, size_round
- size_old
);
4606 super
->current_vol
= idx
;
4608 /* handle 'failed_disks' by either:
4609 * a) create dummy disk entries in the table if this the first
4610 * volume in the array. We add them here as this is the only
4611 * opportunity to add them. add_to_super_imsm_volume()
4612 * handles the non-failed disks and continues incrementing
4614 * b) validate that 'failed_disks' matches the current number
4615 * of missing disks if the container is populated
4617 if (super
->current_vol
== 0) {
4619 for (i
= 0; i
< info
->failed_disks
; i
++) {
4620 struct imsm_disk
*disk
;
4623 disk
= __get_imsm_disk(mpb
, i
);
4624 disk
->status
= CONFIGURED_DISK
| FAILED_DISK
;
4625 disk
->scsi_id
= __cpu_to_le32(~(__u32
)0);
4626 snprintf((char *) disk
->serial
, MAX_RAID_SERIAL_LEN
,
4629 find_missing(super
);
4634 for (d
= super
->missing
; d
; d
= d
->next
)
4636 if (info
->failed_disks
> missing
) {
4637 fprintf(stderr
, Name
": unable to add 'missing' disk to container\n");
4642 if (!check_name(super
, name
, 0))
4644 dv
= malloc(sizeof(*dv
));
4646 fprintf(stderr
, Name
": failed to allocate device list entry\n");
4649 dev
= calloc(1, sizeof(*dev
) + sizeof(__u32
) * (info
->raid_disks
- 1));
4652 fprintf(stderr
, Name
": could not allocate raid device\n");
4656 strncpy((char *) dev
->volume
, name
, MAX_RAID_SERIAL_LEN
);
4657 array_blocks
= calc_array_size(info
->level
, info
->raid_disks
,
4658 info
->layout
, info
->chunk_size
,
4660 /* round array size down to closest MB */
4661 array_blocks
= (array_blocks
>> SECT_PER_MB_SHIFT
) << SECT_PER_MB_SHIFT
;
4663 dev
->size_low
= __cpu_to_le32((__u32
) array_blocks
);
4664 dev
->size_high
= __cpu_to_le32((__u32
) (array_blocks
>> 32));
4665 dev
->status
= (DEV_READ_COALESCING
| DEV_WRITE_COALESCING
);
4667 vol
->migr_state
= 0;
4668 set_migr_type(dev
, MIGR_INIT
);
4669 vol
->dirty
= !info
->state
;
4670 vol
->curr_migr_unit
= 0;
4671 map
= get_imsm_map(dev
, MAP_0
);
4672 set_pba_of_lba0(map
, super
->create_offset
);
4673 set_blocks_per_member(map
, info_to_blocks_per_member(info
, size
));
4674 map
->blocks_per_strip
= __cpu_to_le16(info_to_blocks_per_strip(info
));
4675 map
->failed_disk_num
= ~0;
4676 if (info
->level
> 0)
4677 map
->map_state
= IMSM_T_STATE_UNINITIALIZED
;
4679 map
->map_state
= info
->failed_disks
? IMSM_T_STATE_FAILED
:
4680 IMSM_T_STATE_NORMAL
;
4683 if (info
->level
== 1 && info
->raid_disks
> 2) {
4686 fprintf(stderr
, Name
": imsm does not support more than 2 disks"
4687 "in a raid1 volume\n");
4691 map
->raid_level
= info
->level
;
4692 if (info
->level
== 10) {
4693 map
->raid_level
= 1;
4694 map
->num_domains
= info
->raid_disks
/ 2;
4695 } else if (info
->level
== 1)
4696 map
->num_domains
= info
->raid_disks
;
4698 map
->num_domains
= 1;
4700 /* info->size is only int so use the 'size' parameter instead */
4701 num_data_stripes
= (size
* 2) / info_to_blocks_per_strip(info
);
4702 num_data_stripes
/= map
->num_domains
;
4703 set_num_data_stripes(map
, num_data_stripes
);
4705 map
->num_members
= info
->raid_disks
;
4706 for (i
= 0; i
< map
->num_members
; i
++) {
4707 /* initialized in add_to_super */
4708 set_imsm_ord_tbl_ent(map
, i
, IMSM_ORD_REBUILD
);
4710 mpb
->num_raid_devs
++;
4713 dv
->index
= super
->current_vol
;
4714 dv
->next
= super
->devlist
;
4715 super
->devlist
= dv
;
4717 imsm_update_version_info(super
);
4722 static int init_super_imsm(struct supertype
*st
, mdu_array_info_t
*info
,
4723 unsigned long long size
, char *name
,
4724 char *homehost
, int *uuid
)
4726 /* This is primarily called by Create when creating a new array.
4727 * We will then get add_to_super called for each component, and then
4728 * write_init_super called to write it out to each device.
4729 * For IMSM, Create can create on fresh devices or on a pre-existing
4731 * To create on a pre-existing array a different method will be called.
4732 * This one is just for fresh drives.
4734 struct intel_super
*super
;
4735 struct imsm_super
*mpb
;
4740 return init_super_imsm_volume(st
, info
, size
, name
, homehost
, uuid
);
4743 mpb_size
= disks_to_mpb_size(info
->nr_disks
);
4747 super
= alloc_super();
4748 if (super
&& posix_memalign(&super
->buf
, 512, mpb_size
) != 0) {
4753 fprintf(stderr
, Name
4754 ": %s could not allocate superblock\n", __func__
);
4757 if (posix_memalign(&super
->migr_rec_buf
, 512, MIGR_REC_BUF_SIZE
) != 0) {
4758 fprintf(stderr
, Name
4759 ": %s could not allocate migr_rec buffer\n", __func__
);
4764 memset(super
->buf
, 0, mpb_size
);
4766 mpb
->mpb_size
= __cpu_to_le32(mpb_size
);
4770 /* zeroing superblock */
4774 mpb
->attributes
= MPB_ATTRIB_CHECKSUM_VERIFY
;
4776 version
= (char *) mpb
->sig
;
4777 strcpy(version
, MPB_SIGNATURE
);
4778 version
+= strlen(MPB_SIGNATURE
);
4779 strcpy(version
, MPB_VERSION_RAID0
);
4785 static int add_to_super_imsm_volume(struct supertype
*st
, mdu_disk_info_t
*dk
,
4786 int fd
, char *devname
)
4788 struct intel_super
*super
= st
->sb
;
4789 struct imsm_super
*mpb
= super
->anchor
;
4790 struct imsm_disk
*_disk
;
4791 struct imsm_dev
*dev
;
4792 struct imsm_map
*map
;
4796 dev
= get_imsm_dev(super
, super
->current_vol
);
4797 map
= get_imsm_map(dev
, MAP_0
);
4799 if (! (dk
->state
& (1<<MD_DISK_SYNC
))) {
4800 fprintf(stderr
, Name
": %s: Cannot add spare devices to IMSM volume\n",
4806 /* we're doing autolayout so grab the pre-marked (in
4807 * validate_geometry) raid_disk
4809 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
4810 if (dl
->raiddisk
== dk
->raid_disk
)
4813 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
4814 if (dl
->major
== dk
->major
&&
4815 dl
->minor
== dk
->minor
)
4820 fprintf(stderr
, Name
": %s is not a member of the same container\n", devname
);
4824 /* add a pristine spare to the metadata */
4825 if (dl
->index
< 0) {
4826 dl
->index
= super
->anchor
->num_disks
;
4827 super
->anchor
->num_disks
++;
4829 /* Check the device has not already been added */
4830 slot
= get_imsm_disk_slot(map
, dl
->index
);
4832 (get_imsm_ord_tbl_ent(dev
, slot
, MAP_X
) & IMSM_ORD_REBUILD
) == 0) {
4833 fprintf(stderr
, Name
": %s has been included in this array twice\n",
4837 set_imsm_ord_tbl_ent(map
, dk
->raid_disk
, dl
->index
);
4838 dl
->disk
.status
= CONFIGURED_DISK
;
4840 /* update size of 'missing' disks to be at least as large as the
4841 * largest acitve member (we only have dummy missing disks when
4842 * creating the first volume)
4844 if (super
->current_vol
== 0) {
4845 for (df
= super
->missing
; df
; df
= df
->next
) {
4846 if (total_blocks(&dl
->disk
) > total_blocks(&df
->disk
))
4847 set_total_blocks(&df
->disk
, total_blocks(&dl
->disk
));
4848 _disk
= __get_imsm_disk(mpb
, df
->index
);
4853 /* refresh unset/failed slots to point to valid 'missing' entries */
4854 for (df
= super
->missing
; df
; df
= df
->next
)
4855 for (slot
= 0; slot
< mpb
->num_disks
; slot
++) {
4856 __u32 ord
= get_imsm_ord_tbl_ent(dev
, slot
, MAP_X
);
4858 if ((ord
& IMSM_ORD_REBUILD
) == 0)
4860 set_imsm_ord_tbl_ent(map
, slot
, df
->index
| IMSM_ORD_REBUILD
);
4861 if (is_gen_migration(dev
)) {
4862 struct imsm_map
*map2
= get_imsm_map(dev
,
4864 int slot2
= get_imsm_disk_slot(map2
, df
->index
);
4865 if ((slot2
< map2
->num_members
) &&
4867 __u32 ord2
= get_imsm_ord_tbl_ent(dev
,
4870 if ((unsigned)df
->index
==
4872 set_imsm_ord_tbl_ent(map2
,
4878 dprintf("set slot:%d to missing disk:%d\n", slot
, df
->index
);
4882 /* if we are creating the first raid device update the family number */
4883 if (super
->current_vol
== 0) {
4885 struct imsm_dev
*_dev
= __get_imsm_dev(mpb
, 0);
4887 _disk
= __get_imsm_disk(mpb
, dl
->index
);
4888 if (!_dev
|| !_disk
) {
4889 fprintf(stderr
, Name
": BUG mpb setup error\n");
4895 sum
+= __gen_imsm_checksum(mpb
);
4896 mpb
->family_num
= __cpu_to_le32(sum
);
4897 mpb
->orig_family_num
= mpb
->family_num
;
4899 super
->current_disk
= dl
;
4904 * Function marks disk as spare and restores disk serial
4905 * in case it was previously marked as failed by takeover operation
4907 * -1 : critical error
4908 * 0 : disk is marked as spare but serial is not set
4911 int mark_spare(struct dl
*disk
)
4913 __u8 serial
[MAX_RAID_SERIAL_LEN
];
4920 if (!imsm_read_serial(disk
->fd
, NULL
, serial
)) {
4921 /* Restore disk serial number, because takeover marks disk
4922 * as failed and adds to serial ':0' before it becomes
4925 serialcpy(disk
->serial
, serial
);
4926 serialcpy(disk
->disk
.serial
, serial
);
4929 disk
->disk
.status
= SPARE_DISK
;
4935 static int add_to_super_imsm(struct supertype
*st
, mdu_disk_info_t
*dk
,
4936 int fd
, char *devname
)
4938 struct intel_super
*super
= st
->sb
;
4940 unsigned long long size
;
4945 /* If we are on an RAID enabled platform check that the disk is
4946 * attached to the raid controller.
4947 * We do not need to test disks attachment for container based additions,
4948 * they shall be already tested when container was created/assembled.
4950 rv
= find_intel_hba_capability(fd
, super
, devname
);
4951 /* no orom/efi or non-intel hba of the disk */
4953 dprintf("capability: %p fd: %d ret: %d\n",
4954 super
->orom
, fd
, rv
);
4958 if (super
->current_vol
>= 0)
4959 return add_to_super_imsm_volume(st
, dk
, fd
, devname
);
4962 dd
= malloc(sizeof(*dd
));
4965 Name
": malloc failed %s:%d.\n", __func__
, __LINE__
);
4968 memset(dd
, 0, sizeof(*dd
));
4969 dd
->major
= major(stb
.st_rdev
);
4970 dd
->minor
= minor(stb
.st_rdev
);
4971 dd
->devname
= devname
? strdup(devname
) : NULL
;
4974 dd
->action
= DISK_ADD
;
4975 rv
= imsm_read_serial(fd
, devname
, dd
->serial
);
4978 Name
": failed to retrieve scsi serial, aborting\n");
4983 get_dev_size(fd
, NULL
, &size
);
4985 serialcpy(dd
->disk
.serial
, dd
->serial
);
4986 set_total_blocks(&dd
->disk
, size
);
4987 if (__le32_to_cpu(dd
->disk
.total_blocks_hi
) > 0) {
4988 struct imsm_super
*mpb
= super
->anchor
;
4989 mpb
->attributes
|= MPB_ATTRIB_2TB_DISK
;
4992 if (sysfs_disk_to_scsi_id(fd
, &id
) == 0)
4993 dd
->disk
.scsi_id
= __cpu_to_le32(id
);
4995 dd
->disk
.scsi_id
= __cpu_to_le32(0);
4997 if (st
->update_tail
) {
4998 dd
->next
= super
->disk_mgmt_list
;
4999 super
->disk_mgmt_list
= dd
;
5001 dd
->next
= super
->disks
;
5003 super
->updates_pending
++;
5010 static int remove_from_super_imsm(struct supertype
*st
, mdu_disk_info_t
*dk
)
5012 struct intel_super
*super
= st
->sb
;
5015 /* remove from super works only in mdmon - for communication
5016 * manager - monitor. Check if communication memory buffer
5019 if (!st
->update_tail
) {
5021 Name
": %s shall be used in mdmon context only"
5022 "(line %d).\n", __func__
, __LINE__
);
5025 dd
= malloc(sizeof(*dd
));
5028 Name
": malloc failed %s:%d.\n", __func__
, __LINE__
);
5031 memset(dd
, 0, sizeof(*dd
));
5032 dd
->major
= dk
->major
;
5033 dd
->minor
= dk
->minor
;
5036 dd
->action
= DISK_REMOVE
;
5038 dd
->next
= super
->disk_mgmt_list
;
5039 super
->disk_mgmt_list
= dd
;
5045 static int store_imsm_mpb(int fd
, struct imsm_super
*mpb
);
5049 struct imsm_super anchor
;
5050 } spare_record
__attribute__ ((aligned(512)));
5052 /* spare records have their own family number and do not have any defined raid
5055 static int write_super_imsm_spares(struct intel_super
*super
, int doclose
)
5057 struct imsm_super
*mpb
= super
->anchor
;
5058 struct imsm_super
*spare
= &spare_record
.anchor
;
5062 spare
->mpb_size
= __cpu_to_le32(sizeof(struct imsm_super
)),
5063 spare
->generation_num
= __cpu_to_le32(1UL),
5064 spare
->attributes
= MPB_ATTRIB_CHECKSUM_VERIFY
;
5065 spare
->num_disks
= 1,
5066 spare
->num_raid_devs
= 0,
5067 spare
->cache_size
= mpb
->cache_size
,
5068 spare
->pwr_cycle_count
= __cpu_to_le32(1),
5070 snprintf((char *) spare
->sig
, MAX_SIGNATURE_LENGTH
,
5071 MPB_SIGNATURE MPB_VERSION_RAID0
);
5073 for (d
= super
->disks
; d
; d
= d
->next
) {
5077 spare
->disk
[0] = d
->disk
;
5078 if (__le32_to_cpu(d
->disk
.total_blocks_hi
) > 0)
5079 spare
->attributes
|= MPB_ATTRIB_2TB_DISK
;
5081 sum
= __gen_imsm_checksum(spare
);
5082 spare
->family_num
= __cpu_to_le32(sum
);
5083 spare
->orig_family_num
= 0;
5084 sum
= __gen_imsm_checksum(spare
);
5085 spare
->check_sum
= __cpu_to_le32(sum
);
5087 if (store_imsm_mpb(d
->fd
, spare
)) {
5088 fprintf(stderr
, "%s: failed for device %d:%d %s\n",
5089 __func__
, d
->major
, d
->minor
, strerror(errno
));
5101 static int write_super_imsm(struct supertype
*st
, int doclose
)
5103 struct intel_super
*super
= st
->sb
;
5104 struct imsm_super
*mpb
= super
->anchor
;
5110 __u32 mpb_size
= sizeof(struct imsm_super
) - sizeof(struct imsm_disk
);
5112 int clear_migration_record
= 1;
5114 /* 'generation' is incremented everytime the metadata is written */
5115 generation
= __le32_to_cpu(mpb
->generation_num
);
5117 mpb
->generation_num
= __cpu_to_le32(generation
);
5119 /* fix up cases where previous mdadm releases failed to set
5122 if (mpb
->orig_family_num
== 0)
5123 mpb
->orig_family_num
= mpb
->family_num
;
5125 for (d
= super
->disks
; d
; d
= d
->next
) {
5129 mpb
->disk
[d
->index
] = d
->disk
;
5133 for (d
= super
->missing
; d
; d
= d
->next
) {
5134 mpb
->disk
[d
->index
] = d
->disk
;
5137 mpb
->num_disks
= num_disks
;
5138 mpb_size
+= sizeof(struct imsm_disk
) * mpb
->num_disks
;
5140 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
5141 struct imsm_dev
*dev
= __get_imsm_dev(mpb
, i
);
5142 struct imsm_dev
*dev2
= get_imsm_dev(super
, i
);
5144 imsm_copy_dev(dev
, dev2
);
5145 mpb_size
+= sizeof_imsm_dev(dev
, 0);
5147 if (is_gen_migration(dev2
))
5148 clear_migration_record
= 0;
5150 mpb_size
+= __le32_to_cpu(mpb
->bbm_log_size
);
5151 mpb
->mpb_size
= __cpu_to_le32(mpb_size
);
5153 /* recalculate checksum */
5154 sum
= __gen_imsm_checksum(mpb
);
5155 mpb
->check_sum
= __cpu_to_le32(sum
);
5157 if (super
->clean_migration_record_by_mdmon
) {
5158 clear_migration_record
= 1;
5159 super
->clean_migration_record_by_mdmon
= 0;
5161 if (clear_migration_record
)
5162 memset(super
->migr_rec_buf
, 0, MIGR_REC_BUF_SIZE
);
5164 /* write the mpb for disks that compose raid devices */
5165 for (d
= super
->disks
; d
; d
= d
->next
) {
5166 if (d
->index
< 0 || is_failed(&d
->disk
))
5169 if (clear_migration_record
) {
5170 unsigned long long dsize
;
5172 get_dev_size(d
->fd
, NULL
, &dsize
);
5173 if (lseek64(d
->fd
, dsize
- 512, SEEK_SET
) >= 0) {
5174 if (write(d
->fd
, super
->migr_rec_buf
,
5175 MIGR_REC_BUF_SIZE
) != MIGR_REC_BUF_SIZE
)
5176 perror("Write migr_rec failed");
5180 if (store_imsm_mpb(d
->fd
, mpb
))
5182 "%s: failed for device %d:%d (fd: %d)%s\n",
5183 __func__
, d
->major
, d
->minor
,
5184 d
->fd
, strerror(errno
));
5193 return write_super_imsm_spares(super
, doclose
);
5199 static int create_array(struct supertype
*st
, int dev_idx
)
5202 struct imsm_update_create_array
*u
;
5203 struct intel_super
*super
= st
->sb
;
5204 struct imsm_dev
*dev
= get_imsm_dev(super
, dev_idx
);
5205 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
5206 struct disk_info
*inf
;
5207 struct imsm_disk
*disk
;
5210 len
= sizeof(*u
) - sizeof(*dev
) + sizeof_imsm_dev(dev
, 0) +
5211 sizeof(*inf
) * map
->num_members
;
5214 fprintf(stderr
, "%s: failed to allocate update buffer\n",
5219 u
->type
= update_create_array
;
5220 u
->dev_idx
= dev_idx
;
5221 imsm_copy_dev(&u
->dev
, dev
);
5222 inf
= get_disk_info(u
);
5223 for (i
= 0; i
< map
->num_members
; i
++) {
5224 int idx
= get_imsm_disk_idx(dev
, i
, MAP_X
);
5226 disk
= get_imsm_disk(super
, idx
);
5227 serialcpy(inf
[i
].serial
, disk
->serial
);
5229 append_metadata_update(st
, u
, len
);
5234 static int mgmt_disk(struct supertype
*st
)
5236 struct intel_super
*super
= st
->sb
;
5238 struct imsm_update_add_remove_disk
*u
;
5240 if (!super
->disk_mgmt_list
)
5246 fprintf(stderr
, "%s: failed to allocate update buffer\n",
5251 u
->type
= update_add_remove_disk
;
5252 append_metadata_update(st
, u
, len
);
5257 static int write_init_super_imsm(struct supertype
*st
)
5259 struct intel_super
*super
= st
->sb
;
5260 int current_vol
= super
->current_vol
;
5262 /* we are done with current_vol reset it to point st at the container */
5263 super
->current_vol
= -1;
5265 if (st
->update_tail
) {
5266 /* queue the recently created array / added disk
5267 * as a metadata update */
5270 /* determine if we are creating a volume or adding a disk */
5271 if (current_vol
< 0) {
5272 /* in the mgmt (add/remove) disk case we are running
5273 * in mdmon context, so don't close fd's
5275 return mgmt_disk(st
);
5277 rv
= create_array(st
, current_vol
);
5282 for (d
= super
->disks
; d
; d
= d
->next
)
5283 Kill(d
->devname
, NULL
, 0, 1, 1);
5284 return write_super_imsm(st
, 1);
5289 static int store_super_imsm(struct supertype
*st
, int fd
)
5291 struct intel_super
*super
= st
->sb
;
5292 struct imsm_super
*mpb
= super
? super
->anchor
: NULL
;
5298 return store_imsm_mpb(fd
, mpb
);
5304 static int imsm_bbm_log_size(struct imsm_super
*mpb
)
5306 return __le32_to_cpu(mpb
->bbm_log_size
);
5310 static int validate_geometry_imsm_container(struct supertype
*st
, int level
,
5311 int layout
, int raiddisks
, int chunk
,
5312 unsigned long long size
, char *dev
,
5313 unsigned long long *freesize
,
5317 unsigned long long ldsize
;
5318 struct intel_super
*super
=NULL
;
5321 if (level
!= LEVEL_CONTAINER
)
5326 fd
= open(dev
, O_RDONLY
|O_EXCL
, 0);
5329 fprintf(stderr
, Name
": imsm: Cannot open %s: %s\n",
5330 dev
, strerror(errno
));
5333 if (!get_dev_size(fd
, dev
, &ldsize
)) {
5338 /* capabilities retrieve could be possible
5339 * note that there is no fd for the disks in array.
5341 super
= alloc_super();
5344 Name
": malloc of %zu failed.\n",
5350 rv
= find_intel_hba_capability(fd
, super
, verbose
? dev
: NULL
);
5354 fd2devname(fd
, str
);
5355 dprintf("validate_geometry_imsm_container: fd: %d %s orom: %p rv: %d raiddisk: %d\n",
5356 fd
, str
, super
->orom
, rv
, raiddisks
);
5358 /* no orom/efi or non-intel hba of the disk */
5365 if (raiddisks
> super
->orom
->tds
) {
5367 fprintf(stderr
, Name
": %d exceeds maximum number of"
5368 " platform supported disks: %d\n",
5369 raiddisks
, super
->orom
->tds
);
5373 if ((super
->orom
->attr
& IMSM_OROM_ATTR_2TB_DISK
) == 0 &&
5374 (ldsize
>> 9) >> 32 > 0) {
5376 fprintf(stderr
, Name
": %s exceeds maximum platform supported size\n", dev
);
5382 *freesize
= avail_size_imsm(st
, ldsize
>> 9);
5388 static unsigned long long find_size(struct extent
*e
, int *idx
, int num_extents
)
5390 const unsigned long long base_start
= e
[*idx
].start
;
5391 unsigned long long end
= base_start
+ e
[*idx
].size
;
5394 if (base_start
== end
)
5398 for (i
= *idx
; i
< num_extents
; i
++) {
5399 /* extend overlapping extents */
5400 if (e
[i
].start
>= base_start
&&
5401 e
[i
].start
<= end
) {
5404 if (e
[i
].start
+ e
[i
].size
> end
)
5405 end
= e
[i
].start
+ e
[i
].size
;
5406 } else if (e
[i
].start
> end
) {
5412 return end
- base_start
;
5415 static unsigned long long merge_extents(struct intel_super
*super
, int sum_extents
)
5417 /* build a composite disk with all known extents and generate a new
5418 * 'maxsize' given the "all disks in an array must share a common start
5419 * offset" constraint
5421 struct extent
*e
= calloc(sum_extents
, sizeof(*e
));
5425 unsigned long long pos
;
5426 unsigned long long start
= 0;
5427 unsigned long long maxsize
;
5428 unsigned long reserve
;
5433 /* coalesce and sort all extents. also, check to see if we need to
5434 * reserve space between member arrays
5437 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
5440 for (i
= 0; i
< dl
->extent_cnt
; i
++)
5443 qsort(e
, sum_extents
, sizeof(*e
), cmp_extent
);
5448 while (i
< sum_extents
) {
5449 e
[j
].start
= e
[i
].start
;
5450 e
[j
].size
= find_size(e
, &i
, sum_extents
);
5452 if (e
[j
-1].size
== 0)
5461 unsigned long long esize
;
5463 esize
= e
[i
].start
- pos
;
5464 if (esize
>= maxsize
) {
5469 pos
= e
[i
].start
+ e
[i
].size
;
5471 } while (e
[i
-1].size
);
5477 /* FIXME assumes volume at offset 0 is the first volume in a
5480 if (start_extent
> 0)
5481 reserve
= IMSM_RESERVED_SECTORS
; /* gap between raid regions */
5485 if (maxsize
< reserve
)
5488 super
->create_offset
= ~((unsigned long long) 0);
5489 if (start
+ reserve
> super
->create_offset
)
5490 return 0; /* start overflows create_offset */
5491 super
->create_offset
= start
+ reserve
;
5493 return maxsize
- reserve
;
5496 static int is_raid_level_supported(const struct imsm_orom
*orom
, int level
, int raiddisks
)
5498 if (level
< 0 || level
== 6 || level
== 4)
5501 /* if we have an orom prevent invalid raid levels */
5504 case 0: return imsm_orom_has_raid0(orom
);
5507 return imsm_orom_has_raid1e(orom
);
5508 return imsm_orom_has_raid1(orom
) && raiddisks
== 2;
5509 case 10: return imsm_orom_has_raid10(orom
) && raiddisks
== 4;
5510 case 5: return imsm_orom_has_raid5(orom
) && raiddisks
> 2;
5513 return 1; /* not on an Intel RAID platform so anything goes */
5520 active_arrays_by_format(char *name
, char* hba
, struct md_list
**devlist
,
5521 int dpa
, int verbose
)
5523 struct mdstat_ent
*mdstat
= mdstat_read(0, 0);
5524 struct mdstat_ent
*memb
= NULL
;
5527 struct md_list
*dv
= NULL
;
5530 for (memb
= mdstat
; memb
; memb
= memb
->next
) {
5531 if (memb
->metadata_version
&&
5532 (strncmp(memb
->metadata_version
, "external:", 9) == 0) &&
5533 (strcmp(&memb
->metadata_version
[9], name
) == 0) &&
5534 !is_subarray(memb
->metadata_version
+9) &&
5536 struct dev_member
*dev
= memb
->members
;
5538 while(dev
&& (fd
< 0)) {
5539 char *path
= malloc(strlen(dev
->name
) + strlen("/dev/") + 1);
5541 num
= sprintf(path
, "%s%s", "/dev/", dev
->name
);
5543 fd
= open(path
, O_RDONLY
, 0);
5544 if ((num
<= 0) || (fd
< 0)) {
5545 pr_vrb(": Cannot open %s: %s\n",
5546 dev
->name
, strerror(errno
));
5553 if ((fd
>= 0) && disk_attached_to_hba(fd
, hba
)) {
5554 struct mdstat_ent
*vol
;
5555 for (vol
= mdstat
; vol
; vol
= vol
->next
) {
5556 if ((vol
->active
> 0) &&
5557 vol
->metadata_version
&&
5558 is_container_member(vol
, memb
->dev
)) {
5563 if (*devlist
&& (found
< dpa
)) {
5564 dv
= calloc(1, sizeof(*dv
));
5566 fprintf(stderr
, Name
": calloc failed\n");
5568 dv
->devname
= malloc(strlen(memb
->dev
) + strlen("/dev/") + 1);
5569 if (dv
->devname
!= NULL
) {
5570 sprintf(dv
->devname
, "%s%s", "/dev/", memb
->dev
);
5573 dv
->next
= *devlist
;
5584 free_mdstat(mdstat
);
5589 static struct md_list
*
5590 get_loop_devices(void)
5593 struct md_list
*devlist
= NULL
;
5594 struct md_list
*dv
= NULL
;
5596 for(i
= 0; i
< 12; i
++) {
5597 dv
= calloc(1, sizeof(*dv
));
5599 fprintf(stderr
, Name
": calloc failed\n");
5602 dv
->devname
= malloc(40);
5603 if (dv
->devname
== NULL
) {
5604 fprintf(stderr
, Name
": malloc failed\n");
5608 sprintf(dv
->devname
, "/dev/loop%d", i
);
5616 static struct md_list
*
5617 get_devices(const char *hba_path
)
5619 struct md_list
*devlist
= NULL
;
5620 struct md_list
*dv
= NULL
;
5626 devlist
= get_loop_devices();
5629 /* scroll through /sys/dev/block looking for devices attached to
5632 dir
= opendir("/sys/dev/block");
5633 for (ent
= dir
? readdir(dir
) : NULL
; ent
; ent
= readdir(dir
)) {
5638 if (sscanf(ent
->d_name
, "%d:%d", &major
, &minor
) != 2)
5640 path
= devt_to_devpath(makedev(major
, minor
));
5643 if (!path_attached_to_hba(path
, hba_path
)) {
5650 fd
= dev_open(ent
->d_name
, O_RDONLY
);
5652 fd2devname(fd
, buf
);
5655 fprintf(stderr
, Name
": cannot open device: %s\n",
5661 dv
= calloc(1, sizeof(*dv
));
5663 fprintf(stderr
, Name
": malloc failed\n");
5667 dv
->devname
= strdup(buf
);
5668 if (dv
->devname
== NULL
) {
5669 fprintf(stderr
, Name
": malloc failed\n");
5680 devlist
= devlist
->next
;
5689 count_volumes_list(struct md_list
*devlist
, char *homehost
,
5690 int verbose
, int *found
)
5692 struct md_list
*tmpdev
;
5694 struct supertype
*st
= NULL
;
5696 /* first walk the list of devices to find a consistent set
5697 * that match the criterea, if that is possible.
5698 * We flag the ones we like with 'used'.
5701 st
= match_metadata_desc_imsm("imsm");
5703 pr_vrb(": cannot allocate memory for imsm supertype\n");
5707 for (tmpdev
= devlist
; tmpdev
; tmpdev
= tmpdev
->next
) {
5708 char *devname
= tmpdev
->devname
;
5710 struct supertype
*tst
;
5712 if (tmpdev
->used
> 1)
5714 tst
= dup_super(st
);
5716 pr_vrb(": cannot allocate memory for imsm supertype\n");
5719 tmpdev
->container
= 0;
5720 dfd
= dev_open(devname
, O_RDONLY
|O_EXCL
);
5722 dprintf(": cannot open device %s: %s\n",
5723 devname
, strerror(errno
));
5725 } else if (fstat(dfd
, &stb
)< 0) {
5727 dprintf(": fstat failed for %s: %s\n",
5728 devname
, strerror(errno
));
5730 } else if ((stb
.st_mode
& S_IFMT
) != S_IFBLK
) {
5731 dprintf(": %s is not a block device.\n",
5734 } else if (must_be_container(dfd
)) {
5735 struct supertype
*cst
;
5736 cst
= super_by_fd(dfd
, NULL
);
5738 dprintf(": cannot recognize container type %s\n",
5741 } else if (tst
->ss
!= st
->ss
) {
5742 dprintf(": non-imsm container - ignore it: %s\n",
5745 } else if (!tst
->ss
->load_container
||
5746 tst
->ss
->load_container(tst
, dfd
, NULL
))
5749 tmpdev
->container
= 1;
5752 cst
->ss
->free_super(cst
);
5754 tmpdev
->st_rdev
= stb
.st_rdev
;
5755 if (tst
->ss
->load_super(tst
,dfd
, NULL
)) {
5756 dprintf(": no RAID superblock on %s\n",
5759 } else if (tst
->ss
->compare_super
== NULL
) {
5760 dprintf(": Cannot assemble %s metadata on %s\n",
5761 tst
->ss
->name
, devname
);
5767 if (tmpdev
->used
== 2 || tmpdev
->used
== 4) {
5768 /* Ignore unrecognised devices during auto-assembly */
5773 tst
->ss
->getinfo_super(tst
, &info
, NULL
);
5775 if (st
->minor_version
== -1)
5776 st
->minor_version
= tst
->minor_version
;
5778 if (memcmp(info
.uuid
, uuid_zero
,
5779 sizeof(int[4])) == 0) {
5780 /* this is a floating spare. It cannot define
5781 * an array unless there are no more arrays of
5782 * this type to be found. It can be included
5783 * in an array of this type though.
5789 if (st
->ss
!= tst
->ss
||
5790 st
->minor_version
!= tst
->minor_version
||
5791 st
->ss
->compare_super(st
, tst
) != 0) {
5792 /* Some mismatch. If exactly one array matches this host,
5793 * we can resolve on that one.
5794 * Or, if we are auto assembling, we just ignore the second
5797 dprintf(": superblock on %s doesn't match others - assembly aborted\n",
5803 dprintf("found: devname: %s\n", devname
);
5807 tst
->ss
->free_super(tst
);
5811 if ((err
= load_super_imsm_all(st
, -1, &st
->sb
, NULL
, devlist
, 0)) == 0) {
5812 struct mdinfo
*iter
, *head
= st
->ss
->container_content(st
, NULL
);
5813 for (iter
= head
; iter
; iter
= iter
->next
) {
5814 dprintf("content->text_version: %s vol\n",
5815 iter
->text_version
);
5816 if (iter
->array
.state
& (1<<MD_SB_BLOCK_VOLUME
)) {
5817 /* do not assemble arrays with unsupported
5819 dprintf(": Cannot activate member %s.\n",
5820 iter
->text_version
);
5827 dprintf(" no valid super block on device list: err: %d %p\n",
5831 dprintf(" no more devices to examin\n");
5834 for (tmpdev
= devlist
; tmpdev
; tmpdev
= tmpdev
->next
) {
5835 if ((tmpdev
->used
== 1) && (tmpdev
->found
)) {
5837 if (count
< tmpdev
->found
)
5840 count
-= tmpdev
->found
;
5843 if (tmpdev
->used
== 1)
5848 st
->ss
->free_super(st
);
5854 count_volumes(char *hba
, int dpa
, int verbose
)
5856 struct md_list
*devlist
= NULL
;
5860 devlist
= get_devices(hba
);
5861 /* if no intel devices return zero volumes */
5862 if (devlist
== NULL
)
5865 count
= active_arrays_by_format("imsm", hba
, &devlist
, dpa
, verbose
);
5866 dprintf(" path: %s active arrays: %d\n", hba
, count
);
5867 if (devlist
== NULL
)
5871 count
+= count_volumes_list(devlist
,
5875 dprintf("found %d count: %d\n", found
, count
);
5878 dprintf("path: %s total number of volumes: %d\n", hba
, count
);
5881 struct md_list
*dv
= devlist
;
5882 devlist
= devlist
->next
;
5889 static int imsm_default_chunk(const struct imsm_orom
*orom
)
5891 /* up to 512 if the plaform supports it, otherwise the platform max.
5892 * 128 if no platform detected
5894 int fs
= max(7, orom
? fls(orom
->sss
) : 0);
5896 return min(512, (1 << fs
));
5900 validate_geometry_imsm_orom(struct intel_super
*super
, int level
, int layout
,
5901 int raiddisks
, int *chunk
, unsigned long long size
, int verbose
)
5903 /* check/set platform and metadata limits/defaults */
5904 if (super
->orom
&& raiddisks
> super
->orom
->dpa
) {
5905 pr_vrb(": platform supports a maximum of %d disks per array\n",
5910 /* capabilities of OROM tested - copied from validate_geometry_imsm_volume */
5911 if (!is_raid_level_supported(super
->orom
, level
, raiddisks
)) {
5912 pr_vrb(": platform does not support raid%d with %d disk%s\n",
5913 level
, raiddisks
, raiddisks
> 1 ? "s" : "");
5917 if (chunk
&& (*chunk
== 0 || *chunk
== UnSet
))
5918 *chunk
= imsm_default_chunk(super
->orom
);
5920 if (super
->orom
&& chunk
&& !imsm_orom_has_chunk(super
->orom
, *chunk
)) {
5921 pr_vrb(": platform does not support a chunk size of: "
5926 if (layout
!= imsm_level_to_layout(level
)) {
5928 pr_vrb(": imsm raid 5 only supports the left-asymmetric layout\n");
5929 else if (level
== 10)
5930 pr_vrb(": imsm raid 10 only supports the n2 layout\n");
5932 pr_vrb(": imsm unknown layout %#x for this raid level %d\n",
5937 if (super
->orom
&& (super
->orom
->attr
& IMSM_OROM_ATTR_2TB
) == 0 && chunk
&&
5938 (calc_array_size(level
, raiddisks
, layout
, *chunk
, size
) >> 32) > 0) {
5939 pr_vrb(": platform does not support a volume size over 2TB\n");
5945 /* validate_geometry_imsm_volume - lifted from validate_geometry_ddf_bvd
5946 * FIX ME add ahci details
5948 static int validate_geometry_imsm_volume(struct supertype
*st
, int level
,
5949 int layout
, int raiddisks
, int *chunk
,
5950 unsigned long long size
, char *dev
,
5951 unsigned long long *freesize
,
5955 struct intel_super
*super
= st
->sb
;
5956 struct imsm_super
*mpb
;
5958 unsigned long long pos
= 0;
5959 unsigned long long maxsize
;
5963 /* We must have the container info already read in. */
5967 mpb
= super
->anchor
;
5969 if (!validate_geometry_imsm_orom(super
, level
, layout
, raiddisks
, chunk
, size
, verbose
)) {
5970 fprintf(stderr
, Name
": RAID gemetry validation failed. "
5971 "Cannot proceed with the action(s).\n");
5975 /* General test: make sure there is space for
5976 * 'raiddisks' device extents of size 'size' at a given
5979 unsigned long long minsize
= size
;
5980 unsigned long long start_offset
= MaxSector
;
5983 minsize
= MPB_SECTOR_CNT
+ IMSM_RESERVED_SECTORS
;
5984 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
5989 e
= get_extents(super
, dl
);
5992 unsigned long long esize
;
5993 esize
= e
[i
].start
- pos
;
5994 if (esize
>= minsize
)
5996 if (found
&& start_offset
== MaxSector
) {
5999 } else if (found
&& pos
!= start_offset
) {
6003 pos
= e
[i
].start
+ e
[i
].size
;
6005 } while (e
[i
-1].size
);
6010 if (dcnt
< raiddisks
) {
6012 fprintf(stderr
, Name
": imsm: Not enough "
6013 "devices with space for this array "
6021 /* This device must be a member of the set */
6022 if (stat(dev
, &stb
) < 0)
6024 if ((S_IFMT
& stb
.st_mode
) != S_IFBLK
)
6026 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
6027 if (dl
->major
== (int)major(stb
.st_rdev
) &&
6028 dl
->minor
== (int)minor(stb
.st_rdev
))
6033 fprintf(stderr
, Name
": %s is not in the "
6034 "same imsm set\n", dev
);
6036 } else if (super
->orom
&& dl
->index
< 0 && mpb
->num_raid_devs
) {
6037 /* If a volume is present then the current creation attempt
6038 * cannot incorporate new spares because the orom may not
6039 * understand this configuration (all member disks must be
6040 * members of each array in the container).
6042 fprintf(stderr
, Name
": %s is a spare and a volume"
6043 " is already defined for this container\n", dev
);
6044 fprintf(stderr
, Name
": The option-rom requires all member"
6045 " disks to be a member of all volumes\n");
6047 } else if (super
->orom
&& mpb
->num_raid_devs
> 0 &&
6048 mpb
->num_disks
!= raiddisks
) {
6049 fprintf(stderr
, Name
": The option-rom requires all member"
6050 " disks to be a member of all volumes\n");
6054 /* retrieve the largest free space block */
6055 e
= get_extents(super
, dl
);
6060 unsigned long long esize
;
6062 esize
= e
[i
].start
- pos
;
6063 if (esize
>= maxsize
)
6065 pos
= e
[i
].start
+ e
[i
].size
;
6067 } while (e
[i
-1].size
);
6072 fprintf(stderr
, Name
": unable to determine free space for: %s\n",
6076 if (maxsize
< size
) {
6078 fprintf(stderr
, Name
": %s not enough space (%llu < %llu)\n",
6079 dev
, maxsize
, size
);
6083 /* count total number of extents for merge */
6085 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
6087 i
+= dl
->extent_cnt
;
6089 maxsize
= merge_extents(super
, i
);
6091 if (!check_env("IMSM_NO_PLATFORM") &&
6092 mpb
->num_raid_devs
> 0 && size
&& size
!= maxsize
) {
6093 fprintf(stderr
, Name
": attempting to create a second "
6094 "volume with size less then remaining space. "
6099 if (maxsize
< size
|| maxsize
== 0) {
6102 fprintf(stderr
, Name
": no free space"
6103 " left on device. Aborting...\n");
6105 fprintf(stderr
, Name
": not enough space"
6106 " to create volume of given size"
6107 " (%llu < %llu). Aborting...\n",
6113 *freesize
= maxsize
;
6116 int count
= count_volumes(super
->hba
->path
,
6117 super
->orom
->dpa
, verbose
);
6118 if (super
->orom
->vphba
<= count
) {
6119 pr_vrb(": platform does not support more than %d raid volumes.\n",
6120 super
->orom
->vphba
);
6127 static int imsm_get_free_size(struct supertype
*st
, int raiddisks
,
6128 unsigned long long size
, int chunk
,
6129 unsigned long long *freesize
)
6131 struct intel_super
*super
= st
->sb
;
6132 struct imsm_super
*mpb
= super
->anchor
;
6137 unsigned long long maxsize
;
6138 unsigned long long minsize
;
6142 /* find the largest common start free region of the possible disks */
6146 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
6152 /* don't activate new spares if we are orom constrained
6153 * and there is already a volume active in the container
6155 if (super
->orom
&& dl
->index
< 0 && mpb
->num_raid_devs
)
6158 e
= get_extents(super
, dl
);
6161 for (i
= 1; e
[i
-1].size
; i
++)
6169 maxsize
= merge_extents(super
, extent_cnt
);
6173 minsize
= chunk
* 2;
6175 if (cnt
< raiddisks
||
6176 (super
->orom
&& used
&& used
!= raiddisks
) ||
6177 maxsize
< minsize
||
6179 fprintf(stderr
, Name
": not enough devices with space to create array.\n");
6180 return 0; /* No enough free spaces large enough */
6191 if (!check_env("IMSM_NO_PLATFORM") &&
6192 mpb
->num_raid_devs
> 0 && size
&& size
!= maxsize
) {
6193 fprintf(stderr
, Name
": attempting to create a second "
6194 "volume with size less then remaining space. "
6199 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
6201 dl
->raiddisk
= cnt
++;
6205 dprintf("imsm: imsm_get_free_size() returns : %llu\n", size
);
6210 static int reserve_space(struct supertype
*st
, int raiddisks
,
6211 unsigned long long size
, int chunk
,
6212 unsigned long long *freesize
)
6214 struct intel_super
*super
= st
->sb
;
6219 rv
= imsm_get_free_size(st
, raiddisks
, size
, chunk
, freesize
);
6222 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
6224 dl
->raiddisk
= cnt
++;
6231 static int validate_geometry_imsm(struct supertype
*st
, int level
, int layout
,
6232 int raiddisks
, int *chunk
, unsigned long long size
,
6233 char *dev
, unsigned long long *freesize
,
6241 * if given unused devices create a container
6242 * if given given devices in a container create a member volume
6244 if (level
== LEVEL_CONTAINER
) {
6245 /* Must be a fresh device to add to a container */
6246 return validate_geometry_imsm_container(st
, level
, layout
,
6248 chunk
?*chunk
:0, size
,
6255 struct intel_super
*super
= st
->sb
;
6256 if (!validate_geometry_imsm_orom(st
->sb
, level
, layout
,
6257 raiddisks
, chunk
, size
,
6260 /* we are being asked to automatically layout a
6261 * new volume based on the current contents of
6262 * the container. If the the parameters can be
6263 * satisfied reserve_space will record the disks,
6264 * start offset, and size of the volume to be
6265 * created. add_to_super and getinfo_super
6266 * detect when autolayout is in progress.
6268 /* assuming that freesize is always given when array is
6270 if (super
->orom
&& freesize
) {
6272 count
= count_volumes(super
->hba
->path
,
6273 super
->orom
->dpa
, verbose
);
6274 if (super
->orom
->vphba
<= count
) {
6275 pr_vrb(": platform does not support more"
6276 " than %d raid volumes.\n",
6277 super
->orom
->vphba
);
6282 return reserve_space(st
, raiddisks
, size
,
6283 chunk
?*chunk
:0, freesize
);
6288 /* creating in a given container */
6289 return validate_geometry_imsm_volume(st
, level
, layout
,
6290 raiddisks
, chunk
, size
,
6291 dev
, freesize
, verbose
);
6294 /* This device needs to be a device in an 'imsm' container */
6295 fd
= open(dev
, O_RDONLY
|O_EXCL
, 0);
6299 Name
": Cannot create this array on device %s\n",
6304 if (errno
!= EBUSY
|| (fd
= open(dev
, O_RDONLY
, 0)) < 0) {
6306 fprintf(stderr
, Name
": Cannot open %s: %s\n",
6307 dev
, strerror(errno
));
6310 /* Well, it is in use by someone, maybe an 'imsm' container. */
6311 cfd
= open_container(fd
);
6315 fprintf(stderr
, Name
": Cannot use %s: It is busy\n",
6319 sra
= sysfs_read(cfd
, 0, GET_VERSION
);
6320 if (sra
&& sra
->array
.major_version
== -1 &&
6321 strcmp(sra
->text_version
, "imsm") == 0)
6325 /* This is a member of a imsm container. Load the container
6326 * and try to create a volume
6328 struct intel_super
*super
;
6330 if (load_super_imsm_all(st
, cfd
, (void **) &super
, NULL
, NULL
, 1) == 0) {
6332 st
->container_dev
= fd2devnum(cfd
);
6334 return validate_geometry_imsm_volume(st
, level
, layout
,
6343 fprintf(stderr
, Name
": failed container membership check\n");
6349 static void default_geometry_imsm(struct supertype
*st
, int *level
, int *layout
, int *chunk
)
6351 struct intel_super
*super
= st
->sb
;
6353 if (level
&& *level
== UnSet
)
6354 *level
= LEVEL_CONTAINER
;
6356 if (level
&& layout
&& *layout
== UnSet
)
6357 *layout
= imsm_level_to_layout(*level
);
6359 if (chunk
&& (*chunk
== UnSet
|| *chunk
== 0))
6360 *chunk
= imsm_default_chunk(super
->orom
);
6363 static void handle_missing(struct intel_super
*super
, struct imsm_dev
*dev
);
6365 static int kill_subarray_imsm(struct supertype
*st
)
6367 /* remove the subarray currently referenced by ->current_vol */
6369 struct intel_dev
**dp
;
6370 struct intel_super
*super
= st
->sb
;
6371 __u8 current_vol
= super
->current_vol
;
6372 struct imsm_super
*mpb
= super
->anchor
;
6374 if (super
->current_vol
< 0)
6376 super
->current_vol
= -1; /* invalidate subarray cursor */
6378 /* block deletions that would change the uuid of active subarrays
6380 * FIXME when immutable ids are available, but note that we'll
6381 * also need to fixup the invalidated/active subarray indexes in
6384 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
6387 if (i
< current_vol
)
6389 sprintf(subarray
, "%u", i
);
6390 if (is_subarray_active(subarray
, st
->devname
)) {
6392 Name
": deleting subarray-%d would change the UUID of active subarray-%d, aborting\n",
6399 if (st
->update_tail
) {
6400 struct imsm_update_kill_array
*u
= malloc(sizeof(*u
));
6404 u
->type
= update_kill_array
;
6405 u
->dev_idx
= current_vol
;
6406 append_metadata_update(st
, u
, sizeof(*u
));
6411 for (dp
= &super
->devlist
; *dp
;)
6412 if ((*dp
)->index
== current_vol
) {
6415 handle_missing(super
, (*dp
)->dev
);
6416 if ((*dp
)->index
> current_vol
)
6421 /* no more raid devices, all active components are now spares,
6422 * but of course failed are still failed
6424 if (--mpb
->num_raid_devs
== 0) {
6427 for (d
= super
->disks
; d
; d
= d
->next
)
6432 super
->updates_pending
++;
6437 static int update_subarray_imsm(struct supertype
*st
, char *subarray
,
6438 char *update
, struct mddev_ident
*ident
)
6440 /* update the subarray currently referenced by ->current_vol */
6441 struct intel_super
*super
= st
->sb
;
6442 struct imsm_super
*mpb
= super
->anchor
;
6444 if (strcmp(update
, "name") == 0) {
6445 char *name
= ident
->name
;
6449 if (is_subarray_active(subarray
, st
->devname
)) {
6451 Name
": Unable to update name of active subarray\n");
6455 if (!check_name(super
, name
, 0))
6458 vol
= strtoul(subarray
, &ep
, 10);
6459 if (*ep
!= '\0' || vol
>= super
->anchor
->num_raid_devs
)
6462 if (st
->update_tail
) {
6463 struct imsm_update_rename_array
*u
= malloc(sizeof(*u
));
6467 u
->type
= update_rename_array
;
6469 snprintf((char *) u
->name
, MAX_RAID_SERIAL_LEN
, "%s", name
);
6470 append_metadata_update(st
, u
, sizeof(*u
));
6472 struct imsm_dev
*dev
;
6475 dev
= get_imsm_dev(super
, vol
);
6476 snprintf((char *) dev
->volume
, MAX_RAID_SERIAL_LEN
, "%s", name
);
6477 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
6478 dev
= get_imsm_dev(super
, i
);
6479 handle_missing(super
, dev
);
6481 super
->updates_pending
++;
6488 #endif /* MDASSEMBLE */
6490 static int is_gen_migration(struct imsm_dev
*dev
)
6495 if (!dev
->vol
.migr_state
)
6498 if (migr_type(dev
) == MIGR_GEN_MIGR
)
6504 static int is_rebuilding(struct imsm_dev
*dev
)
6506 struct imsm_map
*migr_map
;
6508 if (!dev
->vol
.migr_state
)
6511 if (migr_type(dev
) != MIGR_REBUILD
)
6514 migr_map
= get_imsm_map(dev
, MAP_1
);
6516 if (migr_map
->map_state
== IMSM_T_STATE_DEGRADED
)
6523 static int is_initializing(struct imsm_dev
*dev
)
6525 struct imsm_map
*migr_map
;
6527 if (!dev
->vol
.migr_state
)
6530 if (migr_type(dev
) != MIGR_INIT
)
6533 migr_map
= get_imsm_map(dev
, MAP_1
);
6535 if (migr_map
->map_state
== IMSM_T_STATE_UNINITIALIZED
)
6542 static void update_recovery_start(struct intel_super
*super
,
6543 struct imsm_dev
*dev
,
6544 struct mdinfo
*array
)
6546 struct mdinfo
*rebuild
= NULL
;
6550 if (!is_rebuilding(dev
))
6553 /* Find the rebuild target, but punt on the dual rebuild case */
6554 for (d
= array
->devs
; d
; d
= d
->next
)
6555 if (d
->recovery_start
== 0) {
6562 /* (?) none of the disks are marked with
6563 * IMSM_ORD_REBUILD, so assume they are missing and the
6564 * disk_ord_tbl was not correctly updated
6566 dprintf("%s: failed to locate out-of-sync disk\n", __func__
);
6570 units
= __le32_to_cpu(dev
->vol
.curr_migr_unit
);
6571 rebuild
->recovery_start
= units
* blocks_per_migr_unit(super
, dev
);
6575 static int recover_backup_imsm(struct supertype
*st
, struct mdinfo
*info
);
6578 static struct mdinfo
*container_content_imsm(struct supertype
*st
, char *subarray
)
6580 /* Given a container loaded by load_super_imsm_all,
6581 * extract information about all the arrays into
6583 * If 'subarray' is given, just extract info about that array.
6585 * For each imsm_dev create an mdinfo, fill it in,
6586 * then look for matching devices in super->disks
6587 * and create appropriate device mdinfo.
6589 struct intel_super
*super
= st
->sb
;
6590 struct imsm_super
*mpb
= super
->anchor
;
6591 struct mdinfo
*rest
= NULL
;
6595 int spare_disks
= 0;
6597 /* do not assemble arrays when not all attributes are supported */
6598 if (imsm_check_attributes(mpb
->attributes
) == 0) {
6600 fprintf(stderr
, Name
": Unsupported attributes in IMSM metadata."
6601 "Arrays activation is blocked.\n");
6604 /* check for bad blocks */
6605 if (imsm_bbm_log_size(super
->anchor
)) {
6606 fprintf(stderr
, Name
": BBM log found in IMSM metadata."
6607 "Arrays activation is blocked.\n");
6612 /* count spare devices, not used in maps
6614 for (d
= super
->disks
; d
; d
= d
->next
)
6618 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
6619 struct imsm_dev
*dev
;
6620 struct imsm_map
*map
;
6621 struct imsm_map
*map2
;
6622 struct mdinfo
*this;
6630 (i
!= strtoul(subarray
, &ep
, 10) || *ep
!= '\0'))
6633 dev
= get_imsm_dev(super
, i
);
6634 map
= get_imsm_map(dev
, MAP_0
);
6635 map2
= get_imsm_map(dev
, MAP_1
);
6637 /* do not publish arrays that are in the middle of an
6638 * unsupported migration
6640 if (dev
->vol
.migr_state
&&
6641 (migr_type(dev
) == MIGR_STATE_CHANGE
)) {
6642 fprintf(stderr
, Name
": cannot assemble volume '%.16s':"
6643 " unsupported migration in progress\n",
6647 /* do not publish arrays that are not support by controller's
6651 this = malloc(sizeof(*this));
6653 fprintf(stderr
, Name
": failed to allocate %zu bytes\n",
6658 super
->current_vol
= i
;
6659 getinfo_super_imsm_volume(st
, this, NULL
);
6662 chunk
= __le16_to_cpu(map
->blocks_per_strip
) >> 1;
6663 /* mdadm does not support all metadata features- set the bit in all arrays state */
6664 if (!validate_geometry_imsm_orom(super
,
6665 get_imsm_raid_level(map
), /* RAID level */
6666 imsm_level_to_layout(get_imsm_raid_level(map
)),
6667 map
->num_members
, /* raid disks */
6668 &chunk
, join_u32(dev
->size_low
, dev
->size_high
),
6670 fprintf(stderr
, Name
": IMSM RAID geometry validation"
6671 " failed. Array %s activation is blocked.\n",
6673 this->array
.state
|=
6674 (1<<MD_SB_BLOCK_CONTAINER_RESHAPE
) |
6675 (1<<MD_SB_BLOCK_VOLUME
);
6679 /* if array has bad blocks, set suitable bit in all arrays state */
6681 this->array
.state
|=
6682 (1<<MD_SB_BLOCK_CONTAINER_RESHAPE
) |
6683 (1<<MD_SB_BLOCK_VOLUME
);
6685 for (slot
= 0 ; slot
< map
->num_members
; slot
++) {
6686 unsigned long long recovery_start
;
6687 struct mdinfo
*info_d
;
6694 idx
= get_imsm_disk_idx(dev
, slot
, MAP_0
);
6695 ord
= get_imsm_ord_tbl_ent(dev
, slot
, MAP_X
);
6696 for (d
= super
->disks
; d
; d
= d
->next
)
6697 if (d
->index
== idx
)
6700 recovery_start
= MaxSector
;
6703 if (d
&& is_failed(&d
->disk
))
6705 if (ord
& IMSM_ORD_REBUILD
)
6709 * if we skip some disks the array will be assmebled degraded;
6710 * reset resync start to avoid a dirty-degraded
6711 * situation when performing the intial sync
6713 * FIXME handle dirty degraded
6715 if ((skip
|| recovery_start
== 0) && !dev
->vol
.dirty
)
6716 this->resync_start
= MaxSector
;
6720 info_d
= calloc(1, sizeof(*info_d
));
6722 fprintf(stderr
, Name
": failed to allocate disk"
6723 " for volume %.16s\n", dev
->volume
);
6724 info_d
= this->devs
;
6726 struct mdinfo
*d
= info_d
->next
;
6735 info_d
->next
= this->devs
;
6736 this->devs
= info_d
;
6738 info_d
->disk
.number
= d
->index
;
6739 info_d
->disk
.major
= d
->major
;
6740 info_d
->disk
.minor
= d
->minor
;
6741 info_d
->disk
.raid_disk
= slot
;
6742 info_d
->recovery_start
= recovery_start
;
6744 if (slot
< map2
->num_members
)
6745 info_d
->disk
.state
= (1 << MD_DISK_ACTIVE
);
6747 this->array
.spare_disks
++;
6749 if (slot
< map
->num_members
)
6750 info_d
->disk
.state
= (1 << MD_DISK_ACTIVE
);
6752 this->array
.spare_disks
++;
6754 if (info_d
->recovery_start
== MaxSector
)
6755 this->array
.working_disks
++;
6757 info_d
->events
= __le32_to_cpu(mpb
->generation_num
);
6758 info_d
->data_offset
= pba_of_lba0(map
);
6759 info_d
->component_size
= blocks_per_member(map
);
6761 /* now that the disk list is up-to-date fixup recovery_start */
6762 update_recovery_start(super
, dev
, this);
6763 this->array
.spare_disks
+= spare_disks
;
6766 /* check for reshape */
6767 if (this->reshape_active
== 1)
6768 recover_backup_imsm(st
, this);
6777 static __u8
imsm_check_degraded(struct intel_super
*super
, struct imsm_dev
*dev
,
6778 int failed
, int look_in_map
)
6780 struct imsm_map
*map
;
6782 map
= get_imsm_map(dev
, look_in_map
);
6785 return map
->map_state
== IMSM_T_STATE_UNINITIALIZED
?
6786 IMSM_T_STATE_UNINITIALIZED
: IMSM_T_STATE_NORMAL
;
6788 switch (get_imsm_raid_level(map
)) {
6790 return IMSM_T_STATE_FAILED
;
6793 if (failed
< map
->num_members
)
6794 return IMSM_T_STATE_DEGRADED
;
6796 return IMSM_T_STATE_FAILED
;
6801 * check to see if any mirrors have failed, otherwise we
6802 * are degraded. Even numbered slots are mirrored on
6806 /* gcc -Os complains that this is unused */
6807 int insync
= insync
;
6809 for (i
= 0; i
< map
->num_members
; i
++) {
6810 __u32 ord
= get_imsm_ord_tbl_ent(dev
, i
, MAP_X
);
6811 int idx
= ord_to_idx(ord
);
6812 struct imsm_disk
*disk
;
6814 /* reset the potential in-sync count on even-numbered
6815 * slots. num_copies is always 2 for imsm raid10
6820 disk
= get_imsm_disk(super
, idx
);
6821 if (!disk
|| is_failed(disk
) || ord
& IMSM_ORD_REBUILD
)
6824 /* no in-sync disks left in this mirror the
6828 return IMSM_T_STATE_FAILED
;
6831 return IMSM_T_STATE_DEGRADED
;
6835 return IMSM_T_STATE_DEGRADED
;
6837 return IMSM_T_STATE_FAILED
;
6843 return map
->map_state
;
6846 static int imsm_count_failed(struct intel_super
*super
, struct imsm_dev
*dev
,
6851 struct imsm_disk
*disk
;
6852 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
6853 struct imsm_map
*prev
= get_imsm_map(dev
, MAP_1
);
6854 struct imsm_map
*map_for_loop
;
6859 /* at the beginning of migration we set IMSM_ORD_REBUILD on
6860 * disks that are being rebuilt. New failures are recorded to
6861 * map[0]. So we look through all the disks we started with and
6862 * see if any failures are still present, or if any new ones
6866 if (prev
&& (map
->num_members
< prev
->num_members
))
6867 map_for_loop
= prev
;
6869 for (i
= 0; i
< map_for_loop
->num_members
; i
++) {
6871 /* when MAP_X is passed both maps failures are counted
6874 ((look_in_map
== MAP_1
) || (look_in_map
== MAP_X
)) &&
6875 (i
< prev
->num_members
)) {
6876 ord
= __le32_to_cpu(prev
->disk_ord_tbl
[i
]);
6877 idx_1
= ord_to_idx(ord
);
6879 disk
= get_imsm_disk(super
, idx_1
);
6880 if (!disk
|| is_failed(disk
) || ord
& IMSM_ORD_REBUILD
)
6883 if (((look_in_map
== MAP_0
) || (look_in_map
== MAP_X
)) &&
6884 (i
< map
->num_members
)) {
6885 ord
= __le32_to_cpu(map
->disk_ord_tbl
[i
]);
6886 idx
= ord_to_idx(ord
);
6889 disk
= get_imsm_disk(super
, idx
);
6890 if (!disk
|| is_failed(disk
) ||
6891 ord
& IMSM_ORD_REBUILD
)
6901 static int imsm_open_new(struct supertype
*c
, struct active_array
*a
,
6904 struct intel_super
*super
= c
->sb
;
6905 struct imsm_super
*mpb
= super
->anchor
;
6907 if (atoi(inst
) >= mpb
->num_raid_devs
) {
6908 fprintf(stderr
, "%s: subarry index %d, out of range\n",
6909 __func__
, atoi(inst
));
6913 dprintf("imsm: open_new %s\n", inst
);
6914 a
->info
.container_member
= atoi(inst
);
6918 static int is_resyncing(struct imsm_dev
*dev
)
6920 struct imsm_map
*migr_map
;
6922 if (!dev
->vol
.migr_state
)
6925 if (migr_type(dev
) == MIGR_INIT
||
6926 migr_type(dev
) == MIGR_REPAIR
)
6929 if (migr_type(dev
) == MIGR_GEN_MIGR
)
6932 migr_map
= get_imsm_map(dev
, MAP_1
);
6934 if ((migr_map
->map_state
== IMSM_T_STATE_NORMAL
) &&
6935 (dev
->vol
.migr_type
!= MIGR_GEN_MIGR
))
6941 /* return true if we recorded new information */
6942 static int mark_failure(struct imsm_dev
*dev
, struct imsm_disk
*disk
, int idx
)
6946 struct imsm_map
*map
;
6947 char buf
[MAX_RAID_SERIAL_LEN
+3];
6948 unsigned int len
, shift
= 0;
6950 /* new failures are always set in map[0] */
6951 map
= get_imsm_map(dev
, MAP_0
);
6953 slot
= get_imsm_disk_slot(map
, idx
);
6957 ord
= __le32_to_cpu(map
->disk_ord_tbl
[slot
]);
6958 if (is_failed(disk
) && (ord
& IMSM_ORD_REBUILD
))
6961 memcpy(buf
, disk
->serial
, MAX_RAID_SERIAL_LEN
);
6962 buf
[MAX_RAID_SERIAL_LEN
] = '\000';
6964 if ((len
= strlen(buf
)) >= MAX_RAID_SERIAL_LEN
)
6965 shift
= len
- MAX_RAID_SERIAL_LEN
+ 1;
6966 strncpy((char *)disk
->serial
, &buf
[shift
], MAX_RAID_SERIAL_LEN
);
6968 disk
->status
|= FAILED_DISK
;
6969 set_imsm_ord_tbl_ent(map
, slot
, idx
| IMSM_ORD_REBUILD
);
6970 /* mark failures in second map if second map exists and this disk
6972 * This is valid for migration, initialization and rebuild
6974 if (dev
->vol
.migr_state
) {
6975 struct imsm_map
*map2
= get_imsm_map(dev
, MAP_1
);
6976 int slot2
= get_imsm_disk_slot(map2
, idx
);
6978 if ((slot2
< map2
->num_members
) &&
6980 set_imsm_ord_tbl_ent(map2
, slot2
,
6981 idx
| IMSM_ORD_REBUILD
);
6983 if (map
->failed_disk_num
== 0xff)
6984 map
->failed_disk_num
= slot
;
6988 static void mark_missing(struct imsm_dev
*dev
, struct imsm_disk
*disk
, int idx
)
6990 mark_failure(dev
, disk
, idx
);
6992 if (disk
->scsi_id
== __cpu_to_le32(~(__u32
)0))
6995 disk
->scsi_id
= __cpu_to_le32(~(__u32
)0);
6996 memmove(&disk
->serial
[0], &disk
->serial
[1], MAX_RAID_SERIAL_LEN
- 1);
6999 static void handle_missing(struct intel_super
*super
, struct imsm_dev
*dev
)
7003 if (!super
->missing
)
7006 dprintf("imsm: mark missing\n");
7007 /* end process for initialization and rebuild only
7009 if (is_gen_migration(dev
) == 0) {
7013 failed
= imsm_count_failed(super
, dev
, MAP_0
);
7014 map_state
= imsm_check_degraded(super
, dev
, failed
, MAP_0
);
7016 end_migration(dev
, super
, map_state
);
7018 for (dl
= super
->missing
; dl
; dl
= dl
->next
)
7019 mark_missing(dev
, &dl
->disk
, dl
->index
);
7020 super
->updates_pending
++;
7023 static unsigned long long imsm_set_array_size(struct imsm_dev
*dev
,
7026 int used_disks
= imsm_num_data_members(dev
, MAP_0
);
7027 unsigned long long array_blocks
;
7028 struct imsm_map
*map
;
7030 if (used_disks
== 0) {
7031 /* when problems occures
7032 * return current array_blocks value
7034 array_blocks
= __le32_to_cpu(dev
->size_high
);
7035 array_blocks
= array_blocks
<< 32;
7036 array_blocks
+= __le32_to_cpu(dev
->size_low
);
7038 return array_blocks
;
7041 /* set array size in metadata
7043 if (new_size
<= 0) {
7044 /* OLCE size change is caused by added disks
7046 map
= get_imsm_map(dev
, MAP_0
);
7047 array_blocks
= blocks_per_member(map
) * used_disks
;
7049 /* Online Volume Size Change
7050 * Using available free space
7052 array_blocks
= new_size
;
7055 /* round array size down to closest MB
7057 array_blocks
= (array_blocks
>> SECT_PER_MB_SHIFT
) << SECT_PER_MB_SHIFT
;
7058 dev
->size_low
= __cpu_to_le32((__u32
)array_blocks
);
7059 dev
->size_high
= __cpu_to_le32((__u32
)(array_blocks
>> 32));
7061 return array_blocks
;
7064 static void imsm_set_disk(struct active_array
*a
, int n
, int state
);
7066 static void imsm_progress_container_reshape(struct intel_super
*super
)
7068 /* if no device has a migr_state, but some device has a
7069 * different number of members than the previous device, start
7070 * changing the number of devices in this device to match
7073 struct imsm_super
*mpb
= super
->anchor
;
7074 int prev_disks
= -1;
7078 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
7079 struct imsm_dev
*dev
= get_imsm_dev(super
, i
);
7080 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
7081 struct imsm_map
*map2
;
7082 int prev_num_members
;
7084 if (dev
->vol
.migr_state
)
7087 if (prev_disks
== -1)
7088 prev_disks
= map
->num_members
;
7089 if (prev_disks
== map
->num_members
)
7092 /* OK, this array needs to enter reshape mode.
7093 * i.e it needs a migr_state
7096 copy_map_size
= sizeof_imsm_map(map
);
7097 prev_num_members
= map
->num_members
;
7098 map
->num_members
= prev_disks
;
7099 dev
->vol
.migr_state
= 1;
7100 dev
->vol
.curr_migr_unit
= 0;
7101 set_migr_type(dev
, MIGR_GEN_MIGR
);
7102 for (i
= prev_num_members
;
7103 i
< map
->num_members
; i
++)
7104 set_imsm_ord_tbl_ent(map
, i
, i
);
7105 map2
= get_imsm_map(dev
, MAP_1
);
7106 /* Copy the current map */
7107 memcpy(map2
, map
, copy_map_size
);
7108 map2
->num_members
= prev_num_members
;
7110 imsm_set_array_size(dev
, -1);
7111 super
->clean_migration_record_by_mdmon
= 1;
7112 super
->updates_pending
++;
7116 /* Handle dirty -> clean transititions, resync and reshape. Degraded and rebuild
7117 * states are handled in imsm_set_disk() with one exception, when a
7118 * resync is stopped due to a new failure this routine will set the
7119 * 'degraded' state for the array.
7121 static int imsm_set_array_state(struct active_array
*a
, int consistent
)
7123 int inst
= a
->info
.container_member
;
7124 struct intel_super
*super
= a
->container
->sb
;
7125 struct imsm_dev
*dev
= get_imsm_dev(super
, inst
);
7126 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
7127 int failed
= imsm_count_failed(super
, dev
, MAP_0
);
7128 __u8 map_state
= imsm_check_degraded(super
, dev
, failed
, MAP_0
);
7129 __u32 blocks_per_unit
;
7131 if (dev
->vol
.migr_state
&&
7132 dev
->vol
.migr_type
== MIGR_GEN_MIGR
) {
7133 /* array state change is blocked due to reshape action
7135 * - abort the reshape (if last_checkpoint is 0 and action!= reshape)
7136 * - finish the reshape (if last_checkpoint is big and action != reshape)
7137 * - update curr_migr_unit
7139 if (a
->curr_action
== reshape
) {
7140 /* still reshaping, maybe update curr_migr_unit */
7141 goto mark_checkpoint
;
7143 if (a
->last_checkpoint
== 0 && a
->prev_action
== reshape
) {
7144 /* for some reason we aborted the reshape.
7146 * disable automatic metadata rollback
7147 * user action is required to recover process
7150 struct imsm_map
*map2
=
7151 get_imsm_map(dev
, MAP_1
);
7152 dev
->vol
.migr_state
= 0;
7153 set_migr_type(dev
, 0);
7154 dev
->vol
.curr_migr_unit
= 0;
7156 sizeof_imsm_map(map2
));
7157 super
->updates_pending
++;
7160 if (a
->last_checkpoint
>= a
->info
.component_size
) {
7161 unsigned long long array_blocks
;
7165 used_disks
= imsm_num_data_members(dev
, MAP_0
);
7166 if (used_disks
> 0) {
7168 blocks_per_member(map
) *
7170 /* round array size down to closest MB
7172 array_blocks
= (array_blocks
7173 >> SECT_PER_MB_SHIFT
)
7174 << SECT_PER_MB_SHIFT
;
7175 a
->info
.custom_array_size
= array_blocks
;
7176 /* encourage manager to update array
7180 a
->check_reshape
= 1;
7182 /* finalize online capacity expansion/reshape */
7183 for (mdi
= a
->info
.devs
; mdi
; mdi
= mdi
->next
)
7185 mdi
->disk
.raid_disk
,
7188 imsm_progress_container_reshape(super
);
7193 /* before we activate this array handle any missing disks */
7194 if (consistent
== 2)
7195 handle_missing(super
, dev
);
7197 if (consistent
== 2 &&
7198 (!is_resync_complete(&a
->info
) ||
7199 map_state
!= IMSM_T_STATE_NORMAL
||
7200 dev
->vol
.migr_state
))
7203 if (is_resync_complete(&a
->info
)) {
7204 /* complete intialization / resync,
7205 * recovery and interrupted recovery is completed in
7208 if (is_resyncing(dev
)) {
7209 dprintf("imsm: mark resync done\n");
7210 end_migration(dev
, super
, map_state
);
7211 super
->updates_pending
++;
7212 a
->last_checkpoint
= 0;
7214 } else if ((!is_resyncing(dev
) && !failed
) &&
7215 (imsm_reshape_blocks_arrays_changes(super
) == 0)) {
7216 /* mark the start of the init process if nothing is failed */
7217 dprintf("imsm: mark resync start\n");
7218 if (map
->map_state
== IMSM_T_STATE_UNINITIALIZED
)
7219 migrate(dev
, super
, IMSM_T_STATE_NORMAL
, MIGR_INIT
);
7221 migrate(dev
, super
, IMSM_T_STATE_NORMAL
, MIGR_REPAIR
);
7222 super
->updates_pending
++;
7226 /* skip checkpointing for general migration,
7227 * it is controlled in mdadm
7229 if (is_gen_migration(dev
))
7230 goto skip_mark_checkpoint
;
7232 /* check if we can update curr_migr_unit from resync_start, recovery_start */
7233 blocks_per_unit
= blocks_per_migr_unit(super
, dev
);
7234 if (blocks_per_unit
) {
7238 units
= a
->last_checkpoint
/ blocks_per_unit
;
7241 /* check that we did not overflow 32-bits, and that
7242 * curr_migr_unit needs updating
7244 if (units32
== units
&&
7246 __le32_to_cpu(dev
->vol
.curr_migr_unit
) != units32
) {
7247 dprintf("imsm: mark checkpoint (%u)\n", units32
);
7248 dev
->vol
.curr_migr_unit
= __cpu_to_le32(units32
);
7249 super
->updates_pending
++;
7253 skip_mark_checkpoint
:
7254 /* mark dirty / clean */
7255 if (dev
->vol
.dirty
!= !consistent
) {
7256 dprintf("imsm: mark '%s'\n", consistent
? "clean" : "dirty");
7261 super
->updates_pending
++;
7267 static void imsm_set_disk(struct active_array
*a
, int n
, int state
)
7269 int inst
= a
->info
.container_member
;
7270 struct intel_super
*super
= a
->container
->sb
;
7271 struct imsm_dev
*dev
= get_imsm_dev(super
, inst
);
7272 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
7273 struct imsm_disk
*disk
;
7278 if (n
> map
->num_members
)
7279 fprintf(stderr
, "imsm: set_disk %d out of range 0..%d\n",
7280 n
, map
->num_members
- 1);
7285 dprintf("imsm: set_disk %d:%x\n", n
, state
);
7287 ord
= get_imsm_ord_tbl_ent(dev
, n
, MAP_0
);
7288 disk
= get_imsm_disk(super
, ord_to_idx(ord
));
7290 /* check for new failures */
7291 if (state
& DS_FAULTY
) {
7292 if (mark_failure(dev
, disk
, ord_to_idx(ord
)))
7293 super
->updates_pending
++;
7296 /* check if in_sync */
7297 if (state
& DS_INSYNC
&& ord
& IMSM_ORD_REBUILD
&& is_rebuilding(dev
)) {
7298 struct imsm_map
*migr_map
= get_imsm_map(dev
, MAP_1
);
7300 set_imsm_ord_tbl_ent(migr_map
, n
, ord_to_idx(ord
));
7301 super
->updates_pending
++;
7304 failed
= imsm_count_failed(super
, dev
, MAP_0
);
7305 map_state
= imsm_check_degraded(super
, dev
, failed
, MAP_0
);
7307 /* check if recovery complete, newly degraded, or failed */
7308 dprintf("imsm: Detected transition to state ");
7309 switch (map_state
) {
7310 case IMSM_T_STATE_NORMAL
: /* transition to normal state */
7311 dprintf("normal: ");
7312 if (is_rebuilding(dev
)) {
7313 dprintf("while rebuilding");
7314 end_migration(dev
, super
, map_state
);
7315 map
= get_imsm_map(dev
, MAP_0
);
7316 map
->failed_disk_num
= ~0;
7317 super
->updates_pending
++;
7318 a
->last_checkpoint
= 0;
7321 if (is_gen_migration(dev
)) {
7322 dprintf("while general migration");
7323 if (a
->last_checkpoint
>= a
->info
.component_size
)
7324 end_migration(dev
, super
, map_state
);
7326 map
->map_state
= map_state
;
7327 map
= get_imsm_map(dev
, MAP_0
);
7328 map
->failed_disk_num
= ~0;
7329 super
->updates_pending
++;
7333 case IMSM_T_STATE_DEGRADED
: /* transition to degraded state */
7334 dprintf("degraded: ");
7335 if ((map
->map_state
!= map_state
) &&
7336 !dev
->vol
.migr_state
) {
7337 dprintf("mark degraded");
7338 map
->map_state
= map_state
;
7339 super
->updates_pending
++;
7340 a
->last_checkpoint
= 0;
7343 if (is_rebuilding(dev
)) {
7344 dprintf("while rebuilding.");
7345 if (map
->map_state
!= map_state
) {
7346 dprintf(" Map state change");
7347 end_migration(dev
, super
, map_state
);
7348 super
->updates_pending
++;
7352 if (is_gen_migration(dev
)) {
7353 dprintf("while general migration");
7354 if (a
->last_checkpoint
>= a
->info
.component_size
)
7355 end_migration(dev
, super
, map_state
);
7357 map
->map_state
= map_state
;
7358 manage_second_map(super
, dev
);
7360 super
->updates_pending
++;
7363 if (is_initializing(dev
)) {
7364 dprintf("while initialization.");
7365 map
->map_state
= map_state
;
7366 super
->updates_pending
++;
7370 case IMSM_T_STATE_FAILED
: /* transition to failed state */
7371 dprintf("failed: ");
7372 if (is_gen_migration(dev
)) {
7373 dprintf("while general migration");
7374 map
->map_state
= map_state
;
7375 super
->updates_pending
++;
7378 if (map
->map_state
!= map_state
) {
7379 dprintf("mark failed");
7380 end_migration(dev
, super
, map_state
);
7381 super
->updates_pending
++;
7382 a
->last_checkpoint
= 0;
7387 dprintf("state %i\n", map_state
);
7393 static int store_imsm_mpb(int fd
, struct imsm_super
*mpb
)
7396 __u32 mpb_size
= __le32_to_cpu(mpb
->mpb_size
);
7397 unsigned long long dsize
;
7398 unsigned long long sectors
;
7400 get_dev_size(fd
, NULL
, &dsize
);
7402 if (mpb_size
> 512) {
7403 /* -1 to account for anchor */
7404 sectors
= mpb_sectors(mpb
) - 1;
7406 /* write the extended mpb to the sectors preceeding the anchor */
7407 if (lseek64(fd
, dsize
- (512 * (2 + sectors
)), SEEK_SET
) < 0)
7410 if ((unsigned long long)write(fd
, buf
+ 512, 512 * sectors
)
7415 /* first block is stored on second to last sector of the disk */
7416 if (lseek64(fd
, dsize
- (512 * 2), SEEK_SET
) < 0)
7419 if (write(fd
, buf
, 512) != 512)
7425 static void imsm_sync_metadata(struct supertype
*container
)
7427 struct intel_super
*super
= container
->sb
;
7429 dprintf("sync metadata: %d\n", super
->updates_pending
);
7430 if (!super
->updates_pending
)
7433 write_super_imsm(container
, 0);
7435 super
->updates_pending
= 0;
7438 static struct dl
*imsm_readd(struct intel_super
*super
, int idx
, struct active_array
*a
)
7440 struct imsm_dev
*dev
= get_imsm_dev(super
, a
->info
.container_member
);
7441 int i
= get_imsm_disk_idx(dev
, idx
, MAP_X
);
7444 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
7448 if (dl
&& is_failed(&dl
->disk
))
7452 dprintf("%s: found %x:%x\n", __func__
, dl
->major
, dl
->minor
);
7457 static struct dl
*imsm_add_spare(struct intel_super
*super
, int slot
,
7458 struct active_array
*a
, int activate_new
,
7459 struct mdinfo
*additional_test_list
)
7461 struct imsm_dev
*dev
= get_imsm_dev(super
, a
->info
.container_member
);
7462 int idx
= get_imsm_disk_idx(dev
, slot
, MAP_X
);
7463 struct imsm_super
*mpb
= super
->anchor
;
7464 struct imsm_map
*map
;
7465 unsigned long long pos
;
7470 __u32 array_start
= 0;
7471 __u32 array_end
= 0;
7473 struct mdinfo
*test_list
;
7475 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
7476 /* If in this array, skip */
7477 for (d
= a
->info
.devs
; d
; d
= d
->next
)
7478 if (d
->state_fd
>= 0 &&
7479 d
->disk
.major
== dl
->major
&&
7480 d
->disk
.minor
== dl
->minor
) {
7481 dprintf("%x:%x already in array\n",
7482 dl
->major
, dl
->minor
);
7487 test_list
= additional_test_list
;
7489 if (test_list
->disk
.major
== dl
->major
&&
7490 test_list
->disk
.minor
== dl
->minor
) {
7491 dprintf("%x:%x already in additional test list\n",
7492 dl
->major
, dl
->minor
);
7495 test_list
= test_list
->next
;
7500 /* skip in use or failed drives */
7501 if (is_failed(&dl
->disk
) || idx
== dl
->index
||
7503 dprintf("%x:%x status (failed: %d index: %d)\n",
7504 dl
->major
, dl
->minor
, is_failed(&dl
->disk
), idx
);
7508 /* skip pure spares when we are looking for partially
7509 * assimilated drives
7511 if (dl
->index
== -1 && !activate_new
)
7514 /* Does this unused device have the requisite free space?
7515 * It needs to be able to cover all member volumes
7517 ex
= get_extents(super
, dl
);
7519 dprintf("cannot get extents\n");
7522 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
7523 dev
= get_imsm_dev(super
, i
);
7524 map
= get_imsm_map(dev
, MAP_0
);
7526 /* check if this disk is already a member of
7529 if (get_imsm_disk_slot(map
, dl
->index
) >= 0)
7535 array_start
= pba_of_lba0(map
);
7536 array_end
= array_start
+
7537 blocks_per_member(map
) - 1;
7540 /* check that we can start at pba_of_lba0 with
7541 * blocks_per_member of space
7543 if (array_start
>= pos
&& array_end
< ex
[j
].start
) {
7547 pos
= ex
[j
].start
+ ex
[j
].size
;
7549 } while (ex
[j
-1].size
);
7556 if (i
< mpb
->num_raid_devs
) {
7557 dprintf("%x:%x does not have %u to %u available\n",
7558 dl
->major
, dl
->minor
, array_start
, array_end
);
7569 static int imsm_rebuild_allowed(struct supertype
*cont
, int dev_idx
, int failed
)
7571 struct imsm_dev
*dev2
;
7572 struct imsm_map
*map
;
7578 dev2
= get_imsm_dev(cont
->sb
, dev_idx
);
7580 state
= imsm_check_degraded(cont
->sb
, dev2
, failed
, MAP_0
);
7581 if (state
== IMSM_T_STATE_FAILED
) {
7582 map
= get_imsm_map(dev2
, MAP_0
);
7585 for (slot
= 0; slot
< map
->num_members
; slot
++) {
7587 * Check if failed disks are deleted from intel
7588 * disk list or are marked to be deleted
7590 idx
= get_imsm_disk_idx(dev2
, slot
, MAP_X
);
7591 idisk
= get_imsm_dl_disk(cont
->sb
, idx
);
7593 * Do not rebuild the array if failed disks
7594 * from failed sub-array are not removed from
7598 is_failed(&idisk
->disk
) &&
7599 (idisk
->action
!= DISK_REMOVE
))
7607 static struct mdinfo
*imsm_activate_spare(struct active_array
*a
,
7608 struct metadata_update
**updates
)
7611 * Find a device with unused free space and use it to replace a
7612 * failed/vacant region in an array. We replace failed regions one a
7613 * array at a time. The result is that a new spare disk will be added
7614 * to the first failed array and after the monitor has finished
7615 * propagating failures the remainder will be consumed.
7617 * FIXME add a capability for mdmon to request spares from another
7621 struct intel_super
*super
= a
->container
->sb
;
7622 int inst
= a
->info
.container_member
;
7623 struct imsm_dev
*dev
= get_imsm_dev(super
, inst
);
7624 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
7625 int failed
= a
->info
.array
.raid_disks
;
7626 struct mdinfo
*rv
= NULL
;
7629 struct metadata_update
*mu
;
7631 struct imsm_update_activate_spare
*u
;
7636 for (d
= a
->info
.devs
; d
; d
= d
->next
) {
7637 if ((d
->curr_state
& DS_FAULTY
) &&
7639 /* wait for Removal to happen */
7641 if (d
->state_fd
>= 0)
7645 dprintf("imsm: activate spare: inst=%d failed=%d (%d) level=%d\n",
7646 inst
, failed
, a
->info
.array
.raid_disks
, a
->info
.array
.level
);
7648 if (imsm_reshape_blocks_arrays_changes(super
))
7651 /* Cannot activate another spare if rebuild is in progress already
7653 if (is_rebuilding(dev
)) {
7654 dprintf("imsm: No spare activation allowed. "
7655 "Rebuild in progress already.\n");
7659 if (a
->info
.array
.level
== 4)
7660 /* No repair for takeovered array
7661 * imsm doesn't support raid4
7665 if (imsm_check_degraded(super
, dev
, failed
, MAP_0
) !=
7666 IMSM_T_STATE_DEGRADED
)
7670 * If there are any failed disks check state of the other volume.
7671 * Block rebuild if the another one is failed until failed disks
7672 * are removed from container.
7675 dprintf("found failed disks in %.*s, check if there another"
7676 "failed sub-array.\n",
7677 MAX_RAID_SERIAL_LEN
, dev
->volume
);
7678 /* check if states of the other volumes allow for rebuild */
7679 for (i
= 0; i
< super
->anchor
->num_raid_devs
; i
++) {
7681 allowed
= imsm_rebuild_allowed(a
->container
,
7689 /* For each slot, if it is not working, find a spare */
7690 for (i
= 0; i
< a
->info
.array
.raid_disks
; i
++) {
7691 for (d
= a
->info
.devs
; d
; d
= d
->next
)
7692 if (d
->disk
.raid_disk
== i
)
7694 dprintf("found %d: %p %x\n", i
, d
, d
?d
->curr_state
:0);
7695 if (d
&& (d
->state_fd
>= 0))
7699 * OK, this device needs recovery. Try to re-add the
7700 * previous occupant of this slot, if this fails see if
7701 * we can continue the assimilation of a spare that was
7702 * partially assimilated, finally try to activate a new
7705 dl
= imsm_readd(super
, i
, a
);
7707 dl
= imsm_add_spare(super
, i
, a
, 0, rv
);
7709 dl
= imsm_add_spare(super
, i
, a
, 1, rv
);
7713 /* found a usable disk with enough space */
7714 di
= malloc(sizeof(*di
));
7717 memset(di
, 0, sizeof(*di
));
7719 /* dl->index will be -1 in the case we are activating a
7720 * pristine spare. imsm_process_update() will create a
7721 * new index in this case. Once a disk is found to be
7722 * failed in all member arrays it is kicked from the
7725 di
->disk
.number
= dl
->index
;
7727 /* (ab)use di->devs to store a pointer to the device
7730 di
->devs
= (struct mdinfo
*) dl
;
7732 di
->disk
.raid_disk
= i
;
7733 di
->disk
.major
= dl
->major
;
7734 di
->disk
.minor
= dl
->minor
;
7736 di
->recovery_start
= 0;
7737 di
->data_offset
= pba_of_lba0(map
);
7738 di
->component_size
= a
->info
.component_size
;
7739 di
->container_member
= inst
;
7740 super
->random
= random32();
7744 dprintf("%x:%x to be %d at %llu\n", dl
->major
, dl
->minor
,
7745 i
, di
->data_offset
);
7749 /* No spares found */
7751 /* Now 'rv' has a list of devices to return.
7752 * Create a metadata_update record to update the
7753 * disk_ord_tbl for the array
7755 mu
= malloc(sizeof(*mu
));
7757 mu
->buf
= malloc(sizeof(struct imsm_update_activate_spare
) * num_spares
);
7758 if (mu
->buf
== NULL
) {
7765 struct mdinfo
*n
= rv
->next
;
7774 mu
->space_list
= NULL
;
7775 mu
->len
= sizeof(struct imsm_update_activate_spare
) * num_spares
;
7776 mu
->next
= *updates
;
7777 u
= (struct imsm_update_activate_spare
*) mu
->buf
;
7779 for (di
= rv
; di
; di
= di
->next
) {
7780 u
->type
= update_activate_spare
;
7781 u
->dl
= (struct dl
*) di
->devs
;
7783 u
->slot
= di
->disk
.raid_disk
;
7794 static int disks_overlap(struct intel_super
*super
, int idx
, struct imsm_update_create_array
*u
)
7796 struct imsm_dev
*dev
= get_imsm_dev(super
, idx
);
7797 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
7798 struct imsm_map
*new_map
= get_imsm_map(&u
->dev
, MAP_0
);
7799 struct disk_info
*inf
= get_disk_info(u
);
7800 struct imsm_disk
*disk
;
7804 for (i
= 0; i
< map
->num_members
; i
++) {
7805 disk
= get_imsm_disk(super
, get_imsm_disk_idx(dev
, i
, MAP_X
));
7806 for (j
= 0; j
< new_map
->num_members
; j
++)
7807 if (serialcmp(disk
->serial
, inf
[j
].serial
) == 0)
7815 static struct dl
*get_disk_super(struct intel_super
*super
, int major
, int minor
)
7817 struct dl
*dl
= NULL
;
7818 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
7819 if ((dl
->major
== major
) && (dl
->minor
== minor
))
7824 static int remove_disk_super(struct intel_super
*super
, int major
, int minor
)
7826 struct dl
*prev
= NULL
;
7830 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
7831 if ((dl
->major
== major
) && (dl
->minor
== minor
)) {
7834 prev
->next
= dl
->next
;
7836 super
->disks
= dl
->next
;
7838 __free_imsm_disk(dl
);
7839 dprintf("%s: removed %x:%x\n",
7840 __func__
, major
, minor
);
7848 static void imsm_delete(struct intel_super
*super
, struct dl
**dlp
, unsigned index
);
7850 static int add_remove_disk_update(struct intel_super
*super
)
7852 int check_degraded
= 0;
7853 struct dl
*disk
= NULL
;
7854 /* add/remove some spares to/from the metadata/contrainer */
7855 while (super
->disk_mgmt_list
) {
7856 struct dl
*disk_cfg
;
7858 disk_cfg
= super
->disk_mgmt_list
;
7859 super
->disk_mgmt_list
= disk_cfg
->next
;
7860 disk_cfg
->next
= NULL
;
7862 if (disk_cfg
->action
== DISK_ADD
) {
7863 disk_cfg
->next
= super
->disks
;
7864 super
->disks
= disk_cfg
;
7866 dprintf("%s: added %x:%x\n",
7867 __func__
, disk_cfg
->major
,
7869 } else if (disk_cfg
->action
== DISK_REMOVE
) {
7870 dprintf("Disk remove action processed: %x.%x\n",
7871 disk_cfg
->major
, disk_cfg
->minor
);
7872 disk
= get_disk_super(super
,
7876 /* store action status */
7877 disk
->action
= DISK_REMOVE
;
7878 /* remove spare disks only */
7879 if (disk
->index
== -1) {
7880 remove_disk_super(super
,
7885 /* release allocate disk structure */
7886 __free_imsm_disk(disk_cfg
);
7889 return check_degraded
;
7893 static int apply_reshape_migration_update(struct imsm_update_reshape_migration
*u
,
7894 struct intel_super
*super
,
7897 struct intel_dev
*id
;
7898 void **tofree
= NULL
;
7901 dprintf("apply_reshape_migration_update()\n");
7902 if ((u
->subdev
< 0) ||
7904 dprintf("imsm: Error: Wrong subdev: %i\n", u
->subdev
);
7907 if ((space_list
== NULL
) || (*space_list
== NULL
)) {
7908 dprintf("imsm: Error: Memory is not allocated\n");
7912 for (id
= super
->devlist
; id
; id
= id
->next
) {
7913 if (id
->index
== (unsigned)u
->subdev
) {
7914 struct imsm_dev
*dev
= get_imsm_dev(super
, u
->subdev
);
7915 struct imsm_map
*map
;
7916 struct imsm_dev
*new_dev
=
7917 (struct imsm_dev
*)*space_list
;
7918 struct imsm_map
*migr_map
= get_imsm_map(dev
, MAP_1
);
7920 struct dl
*new_disk
;
7922 if (new_dev
== NULL
)
7924 *space_list
= **space_list
;
7925 memcpy(new_dev
, dev
, sizeof_imsm_dev(dev
, 0));
7926 map
= get_imsm_map(new_dev
, MAP_0
);
7928 dprintf("imsm: Error: migration in progress");
7932 to_state
= map
->map_state
;
7933 if ((u
->new_level
== 5) && (map
->raid_level
== 0)) {
7935 /* this should not happen */
7936 if (u
->new_disks
[0] < 0) {
7937 map
->failed_disk_num
=
7938 map
->num_members
- 1;
7939 to_state
= IMSM_T_STATE_DEGRADED
;
7941 to_state
= IMSM_T_STATE_NORMAL
;
7943 migrate(new_dev
, super
, to_state
, MIGR_GEN_MIGR
);
7944 if (u
->new_level
> -1)
7945 map
->raid_level
= u
->new_level
;
7946 migr_map
= get_imsm_map(new_dev
, MAP_1
);
7947 if ((u
->new_level
== 5) &&
7948 (migr_map
->raid_level
== 0)) {
7949 int ord
= map
->num_members
- 1;
7950 migr_map
->num_members
--;
7951 if (u
->new_disks
[0] < 0)
7952 ord
|= IMSM_ORD_REBUILD
;
7953 set_imsm_ord_tbl_ent(map
,
7954 map
->num_members
- 1,
7958 tofree
= (void **)dev
;
7960 /* update chunk size
7962 if (u
->new_chunksize
> 0)
7963 map
->blocks_per_strip
=
7964 __cpu_to_le16(u
->new_chunksize
* 2);
7968 if ((u
->new_level
!= 5) ||
7969 (migr_map
->raid_level
!= 0) ||
7970 (migr_map
->raid_level
== map
->raid_level
))
7973 if (u
->new_disks
[0] >= 0) {
7976 new_disk
= get_disk_super(super
,
7977 major(u
->new_disks
[0]),
7978 minor(u
->new_disks
[0]));
7979 dprintf("imsm: new disk for reshape is: %i:%i "
7980 "(%p, index = %i)\n",
7981 major(u
->new_disks
[0]),
7982 minor(u
->new_disks
[0]),
7983 new_disk
, new_disk
->index
);
7984 if (new_disk
== NULL
)
7985 goto error_disk_add
;
7987 new_disk
->index
= map
->num_members
- 1;
7988 /* slot to fill in autolayout
7990 new_disk
->raiddisk
= new_disk
->index
;
7991 new_disk
->disk
.status
|= CONFIGURED_DISK
;
7992 new_disk
->disk
.status
&= ~SPARE_DISK
;
7994 goto error_disk_add
;
7997 *tofree
= *space_list
;
7998 /* calculate new size
8000 imsm_set_array_size(new_dev
, -1);
8007 *space_list
= tofree
;
8011 dprintf("Error: imsm: Cannot find disk.\n");
8015 static int apply_size_change_update(struct imsm_update_size_change
*u
,
8016 struct intel_super
*super
)
8018 struct intel_dev
*id
;
8021 dprintf("apply_size_change_update()\n");
8022 if ((u
->subdev
< 0) ||
8024 dprintf("imsm: Error: Wrong subdev: %i\n", u
->subdev
);
8028 for (id
= super
->devlist
; id
; id
= id
->next
) {
8029 if (id
->index
== (unsigned)u
->subdev
) {
8030 struct imsm_dev
*dev
= get_imsm_dev(super
, u
->subdev
);
8031 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
8032 int used_disks
= imsm_num_data_members(dev
, MAP_0
);
8033 unsigned long long blocks_per_member
;
8035 /* calculate new size
8037 blocks_per_member
= u
->new_size
/ used_disks
;
8038 dprintf("imsm: apply_size_change_update(size: %llu, "
8039 "blocks per member: %llu)\n",
8040 u
->new_size
, blocks_per_member
);
8041 set_blocks_per_member(map
, blocks_per_member
);
8042 imsm_set_array_size(dev
, u
->new_size
);
8053 static int apply_update_activate_spare(struct imsm_update_activate_spare
*u
,
8054 struct intel_super
*super
,
8055 struct active_array
*active_array
)
8057 struct imsm_super
*mpb
= super
->anchor
;
8058 struct imsm_dev
*dev
= get_imsm_dev(super
, u
->array
);
8059 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
8060 struct imsm_map
*migr_map
;
8061 struct active_array
*a
;
8062 struct imsm_disk
*disk
;
8069 int second_map_created
= 0;
8071 for (; u
; u
= u
->next
) {
8072 victim
= get_imsm_disk_idx(dev
, u
->slot
, MAP_X
);
8077 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
8082 fprintf(stderr
, "error: imsm_activate_spare passed "
8083 "an unknown disk (index: %d)\n",
8088 /* count failures (excluding rebuilds and the victim)
8089 * to determine map[0] state
8092 for (i
= 0; i
< map
->num_members
; i
++) {
8095 disk
= get_imsm_disk(super
,
8096 get_imsm_disk_idx(dev
, i
, MAP_X
));
8097 if (!disk
|| is_failed(disk
))
8101 /* adding a pristine spare, assign a new index */
8102 if (dl
->index
< 0) {
8103 dl
->index
= super
->anchor
->num_disks
;
8104 super
->anchor
->num_disks
++;
8107 disk
->status
|= CONFIGURED_DISK
;
8108 disk
->status
&= ~SPARE_DISK
;
8111 to_state
= imsm_check_degraded(super
, dev
, failed
, MAP_0
);
8112 if (!second_map_created
) {
8113 second_map_created
= 1;
8114 map
->map_state
= IMSM_T_STATE_DEGRADED
;
8115 migrate(dev
, super
, to_state
, MIGR_REBUILD
);
8117 map
->map_state
= to_state
;
8118 migr_map
= get_imsm_map(dev
, MAP_1
);
8119 set_imsm_ord_tbl_ent(map
, u
->slot
, dl
->index
);
8120 set_imsm_ord_tbl_ent(migr_map
, u
->slot
,
8121 dl
->index
| IMSM_ORD_REBUILD
);
8123 /* update the family_num to mark a new container
8124 * generation, being careful to record the existing
8125 * family_num in orig_family_num to clean up after
8126 * earlier mdadm versions that neglected to set it.
8128 if (mpb
->orig_family_num
== 0)
8129 mpb
->orig_family_num
= mpb
->family_num
;
8130 mpb
->family_num
+= super
->random
;
8132 /* count arrays using the victim in the metadata */
8134 for (a
= active_array
; a
; a
= a
->next
) {
8135 dev
= get_imsm_dev(super
, a
->info
.container_member
);
8136 map
= get_imsm_map(dev
, MAP_0
);
8138 if (get_imsm_disk_slot(map
, victim
) >= 0)
8142 /* delete the victim if it is no longer being
8148 /* We know that 'manager' isn't touching anything,
8149 * so it is safe to delete
8151 for (dlp
= &super
->disks
; *dlp
; dlp
= &(*dlp
)->next
)
8152 if ((*dlp
)->index
== victim
)
8155 /* victim may be on the missing list */
8157 for (dlp
= &super
->missing
; *dlp
;
8158 dlp
= &(*dlp
)->next
)
8159 if ((*dlp
)->index
== victim
)
8161 imsm_delete(super
, dlp
, victim
);
8168 static int apply_reshape_container_disks_update(struct imsm_update_reshape
*u
,
8169 struct intel_super
*super
,
8172 struct dl
*new_disk
;
8173 struct intel_dev
*id
;
8175 int delta_disks
= u
->new_raid_disks
- u
->old_raid_disks
;
8176 int disk_count
= u
->old_raid_disks
;
8177 void **tofree
= NULL
;
8178 int devices_to_reshape
= 1;
8179 struct imsm_super
*mpb
= super
->anchor
;
8181 unsigned int dev_id
;
8183 dprintf("imsm: apply_reshape_container_disks_update()\n");
8185 /* enable spares to use in array */
8186 for (i
= 0; i
< delta_disks
; i
++) {
8187 new_disk
= get_disk_super(super
,
8188 major(u
->new_disks
[i
]),
8189 minor(u
->new_disks
[i
]));
8190 dprintf("imsm: new disk for reshape is: %i:%i "
8191 "(%p, index = %i)\n",
8192 major(u
->new_disks
[i
]), minor(u
->new_disks
[i
]),
8193 new_disk
, new_disk
->index
);
8194 if ((new_disk
== NULL
) ||
8195 ((new_disk
->index
>= 0) &&
8196 (new_disk
->index
< u
->old_raid_disks
)))
8197 goto update_reshape_exit
;
8198 new_disk
->index
= disk_count
++;
8199 /* slot to fill in autolayout
8201 new_disk
->raiddisk
= new_disk
->index
;
8202 new_disk
->disk
.status
|=
8204 new_disk
->disk
.status
&= ~SPARE_DISK
;
8207 dprintf("imsm: volume set mpb->num_raid_devs = %i\n",
8208 mpb
->num_raid_devs
);
8209 /* manage changes in volume
8211 for (dev_id
= 0; dev_id
< mpb
->num_raid_devs
; dev_id
++) {
8212 void **sp
= *space_list
;
8213 struct imsm_dev
*newdev
;
8214 struct imsm_map
*newmap
, *oldmap
;
8216 for (id
= super
->devlist
; id
; id
= id
->next
) {
8217 if (id
->index
== dev_id
)
8226 /* Copy the dev, but not (all of) the map */
8227 memcpy(newdev
, id
->dev
, sizeof(*newdev
));
8228 oldmap
= get_imsm_map(id
->dev
, MAP_0
);
8229 newmap
= get_imsm_map(newdev
, MAP_0
);
8230 /* Copy the current map */
8231 memcpy(newmap
, oldmap
, sizeof_imsm_map(oldmap
));
8232 /* update one device only
8234 if (devices_to_reshape
) {
8235 dprintf("imsm: modifying subdev: %i\n",
8237 devices_to_reshape
--;
8238 newdev
->vol
.migr_state
= 1;
8239 newdev
->vol
.curr_migr_unit
= 0;
8240 set_migr_type(newdev
, MIGR_GEN_MIGR
);
8241 newmap
->num_members
= u
->new_raid_disks
;
8242 for (i
= 0; i
< delta_disks
; i
++) {
8243 set_imsm_ord_tbl_ent(newmap
,
8244 u
->old_raid_disks
+ i
,
8245 u
->old_raid_disks
+ i
);
8247 /* New map is correct, now need to save old map
8249 newmap
= get_imsm_map(newdev
, MAP_1
);
8250 memcpy(newmap
, oldmap
, sizeof_imsm_map(oldmap
));
8252 imsm_set_array_size(newdev
, -1);
8255 sp
= (void **)id
->dev
;
8260 /* Clear migration record */
8261 memset(super
->migr_rec
, 0, sizeof(struct migr_record
));
8264 *space_list
= tofree
;
8267 update_reshape_exit
:
8272 static int apply_takeover_update(struct imsm_update_takeover
*u
,
8273 struct intel_super
*super
,
8276 struct imsm_dev
*dev
= NULL
;
8277 struct intel_dev
*dv
;
8278 struct imsm_dev
*dev_new
;
8279 struct imsm_map
*map
;
8283 for (dv
= super
->devlist
; dv
; dv
= dv
->next
)
8284 if (dv
->index
== (unsigned int)u
->subarray
) {
8292 map
= get_imsm_map(dev
, MAP_0
);
8294 if (u
->direction
== R10_TO_R0
) {
8295 /* Number of failed disks must be half of initial disk number */
8296 if (imsm_count_failed(super
, dev
, MAP_0
) !=
8297 (map
->num_members
/ 2))
8300 /* iterate through devices to mark removed disks as spare */
8301 for (dm
= super
->disks
; dm
; dm
= dm
->next
) {
8302 if (dm
->disk
.status
& FAILED_DISK
) {
8303 int idx
= dm
->index
;
8304 /* update indexes on the disk list */
8305 /* FIXME this loop-with-the-loop looks wrong, I'm not convinced
8306 the index values will end up being correct.... NB */
8307 for (du
= super
->disks
; du
; du
= du
->next
)
8308 if (du
->index
> idx
)
8310 /* mark as spare disk */
8315 map
->num_members
= map
->num_members
/ 2;
8316 map
->map_state
= IMSM_T_STATE_NORMAL
;
8317 map
->num_domains
= 1;
8318 map
->raid_level
= 0;
8319 map
->failed_disk_num
= -1;
8322 if (u
->direction
== R0_TO_R10
) {
8324 /* update slots in current disk list */
8325 for (dm
= super
->disks
; dm
; dm
= dm
->next
) {
8329 /* create new *missing* disks */
8330 for (i
= 0; i
< map
->num_members
; i
++) {
8331 space
= *space_list
;
8334 *space_list
= *space
;
8336 memcpy(du
, super
->disks
, sizeof(*du
));
8340 du
->index
= (i
* 2) + 1;
8341 sprintf((char *)du
->disk
.serial
,
8342 " MISSING_%d", du
->index
);
8343 sprintf((char *)du
->serial
,
8344 "MISSING_%d", du
->index
);
8345 du
->next
= super
->missing
;
8346 super
->missing
= du
;
8348 /* create new dev and map */
8349 space
= *space_list
;
8352 *space_list
= *space
;
8353 dev_new
= (void *)space
;
8354 memcpy(dev_new
, dev
, sizeof(*dev
));
8355 /* update new map */
8356 map
= get_imsm_map(dev_new
, MAP_0
);
8357 map
->num_members
= map
->num_members
* 2;
8358 map
->map_state
= IMSM_T_STATE_DEGRADED
;
8359 map
->num_domains
= 2;
8360 map
->raid_level
= 1;
8361 /* replace dev<->dev_new */
8364 /* update disk order table */
8365 for (du
= super
->disks
; du
; du
= du
->next
)
8367 set_imsm_ord_tbl_ent(map
, du
->index
, du
->index
);
8368 for (du
= super
->missing
; du
; du
= du
->next
)
8369 if (du
->index
>= 0) {
8370 set_imsm_ord_tbl_ent(map
, du
->index
, du
->index
);
8371 mark_missing(dv
->dev
, &du
->disk
, du
->index
);
8377 static void imsm_process_update(struct supertype
*st
,
8378 struct metadata_update
*update
)
8381 * crack open the metadata_update envelope to find the update record
8382 * update can be one of:
8383 * update_reshape_container_disks - all the arrays in the container
8384 * are being reshaped to have more devices. We need to mark
8385 * the arrays for general migration and convert selected spares
8386 * into active devices.
8387 * update_activate_spare - a spare device has replaced a failed
8388 * device in an array, update the disk_ord_tbl. If this disk is
8389 * present in all member arrays then also clear the SPARE_DISK
8391 * update_create_array
8393 * update_rename_array
8394 * update_add_remove_disk
8396 struct intel_super
*super
= st
->sb
;
8397 struct imsm_super
*mpb
;
8398 enum imsm_update_type type
= *(enum imsm_update_type
*) update
->buf
;
8400 /* update requires a larger buf but the allocation failed */
8401 if (super
->next_len
&& !super
->next_buf
) {
8402 super
->next_len
= 0;
8406 if (super
->next_buf
) {
8407 memcpy(super
->next_buf
, super
->buf
, super
->len
);
8409 super
->len
= super
->next_len
;
8410 super
->buf
= super
->next_buf
;
8412 super
->next_len
= 0;
8413 super
->next_buf
= NULL
;
8416 mpb
= super
->anchor
;
8419 case update_general_migration_checkpoint
: {
8420 struct intel_dev
*id
;
8421 struct imsm_update_general_migration_checkpoint
*u
=
8422 (void *)update
->buf
;
8424 dprintf("imsm: process_update() "
8425 "for update_general_migration_checkpoint called\n");
8427 /* find device under general migration */
8428 for (id
= super
->devlist
; id
; id
= id
->next
) {
8429 if (is_gen_migration(id
->dev
)) {
8430 id
->dev
->vol
.curr_migr_unit
=
8431 __cpu_to_le32(u
->curr_migr_unit
);
8432 super
->updates_pending
++;
8437 case update_takeover
: {
8438 struct imsm_update_takeover
*u
= (void *)update
->buf
;
8439 if (apply_takeover_update(u
, super
, &update
->space_list
)) {
8440 imsm_update_version_info(super
);
8441 super
->updates_pending
++;
8446 case update_reshape_container_disks
: {
8447 struct imsm_update_reshape
*u
= (void *)update
->buf
;
8448 if (apply_reshape_container_disks_update(
8449 u
, super
, &update
->space_list
))
8450 super
->updates_pending
++;
8453 case update_reshape_migration
: {
8454 struct imsm_update_reshape_migration
*u
= (void *)update
->buf
;
8455 if (apply_reshape_migration_update(
8456 u
, super
, &update
->space_list
))
8457 super
->updates_pending
++;
8460 case update_size_change
: {
8461 struct imsm_update_size_change
*u
= (void *)update
->buf
;
8462 if (apply_size_change_update(u
, super
))
8463 super
->updates_pending
++;
8466 case update_activate_spare
: {
8467 struct imsm_update_activate_spare
*u
= (void *) update
->buf
;
8468 if (apply_update_activate_spare(u
, super
, st
->arrays
))
8469 super
->updates_pending
++;
8472 case update_create_array
: {
8473 /* someone wants to create a new array, we need to be aware of
8474 * a few races/collisions:
8475 * 1/ 'Create' called by two separate instances of mdadm
8476 * 2/ 'Create' versus 'activate_spare': mdadm has chosen
8477 * devices that have since been assimilated via
8479 * In the event this update can not be carried out mdadm will
8480 * (FIX ME) notice that its update did not take hold.
8482 struct imsm_update_create_array
*u
= (void *) update
->buf
;
8483 struct intel_dev
*dv
;
8484 struct imsm_dev
*dev
;
8485 struct imsm_map
*map
, *new_map
;
8486 unsigned long long start
, end
;
8487 unsigned long long new_start
, new_end
;
8489 struct disk_info
*inf
;
8492 /* handle racing creates: first come first serve */
8493 if (u
->dev_idx
< mpb
->num_raid_devs
) {
8494 dprintf("%s: subarray %d already defined\n",
8495 __func__
, u
->dev_idx
);
8499 /* check update is next in sequence */
8500 if (u
->dev_idx
!= mpb
->num_raid_devs
) {
8501 dprintf("%s: can not create array %d expected index %d\n",
8502 __func__
, u
->dev_idx
, mpb
->num_raid_devs
);
8506 new_map
= get_imsm_map(&u
->dev
, MAP_0
);
8507 new_start
= pba_of_lba0(new_map
);
8508 new_end
= new_start
+ blocks_per_member(new_map
);
8509 inf
= get_disk_info(u
);
8511 /* handle activate_spare versus create race:
8512 * check to make sure that overlapping arrays do not include
8515 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
8516 dev
= get_imsm_dev(super
, i
);
8517 map
= get_imsm_map(dev
, MAP_0
);
8518 start
= pba_of_lba0(map
);
8519 end
= start
+ blocks_per_member(map
);
8520 if ((new_start
>= start
&& new_start
<= end
) ||
8521 (start
>= new_start
&& start
<= new_end
))
8526 if (disks_overlap(super
, i
, u
)) {
8527 dprintf("%s: arrays overlap\n", __func__
);
8532 /* check that prepare update was successful */
8533 if (!update
->space
) {
8534 dprintf("%s: prepare update failed\n", __func__
);
8538 /* check that all disks are still active before committing
8539 * changes. FIXME: could we instead handle this by creating a
8540 * degraded array? That's probably not what the user expects,
8541 * so better to drop this update on the floor.
8543 for (i
= 0; i
< new_map
->num_members
; i
++) {
8544 dl
= serial_to_dl(inf
[i
].serial
, super
);
8546 dprintf("%s: disk disappeared\n", __func__
);
8551 super
->updates_pending
++;
8553 /* convert spares to members and fixup ord_tbl */
8554 for (i
= 0; i
< new_map
->num_members
; i
++) {
8555 dl
= serial_to_dl(inf
[i
].serial
, super
);
8556 if (dl
->index
== -1) {
8557 dl
->index
= mpb
->num_disks
;
8559 dl
->disk
.status
|= CONFIGURED_DISK
;
8560 dl
->disk
.status
&= ~SPARE_DISK
;
8562 set_imsm_ord_tbl_ent(new_map
, i
, dl
->index
);
8567 update
->space
= NULL
;
8568 imsm_copy_dev(dev
, &u
->dev
);
8569 dv
->index
= u
->dev_idx
;
8570 dv
->next
= super
->devlist
;
8571 super
->devlist
= dv
;
8572 mpb
->num_raid_devs
++;
8574 imsm_update_version_info(super
);
8577 /* mdmon knows how to release update->space, but not
8578 * ((struct intel_dev *) update->space)->dev
8580 if (update
->space
) {
8586 case update_kill_array
: {
8587 struct imsm_update_kill_array
*u
= (void *) update
->buf
;
8588 int victim
= u
->dev_idx
;
8589 struct active_array
*a
;
8590 struct intel_dev
**dp
;
8591 struct imsm_dev
*dev
;
8593 /* sanity check that we are not affecting the uuid of
8594 * active arrays, or deleting an active array
8596 * FIXME when immutable ids are available, but note that
8597 * we'll also need to fixup the invalidated/active
8598 * subarray indexes in mdstat
8600 for (a
= st
->arrays
; a
; a
= a
->next
)
8601 if (a
->info
.container_member
>= victim
)
8603 /* by definition if mdmon is running at least one array
8604 * is active in the container, so checking
8605 * mpb->num_raid_devs is just extra paranoia
8607 dev
= get_imsm_dev(super
, victim
);
8608 if (a
|| !dev
|| mpb
->num_raid_devs
== 1) {
8609 dprintf("failed to delete subarray-%d\n", victim
);
8613 for (dp
= &super
->devlist
; *dp
;)
8614 if ((*dp
)->index
== (unsigned)super
->current_vol
) {
8617 if ((*dp
)->index
> (unsigned)victim
)
8621 mpb
->num_raid_devs
--;
8622 super
->updates_pending
++;
8625 case update_rename_array
: {
8626 struct imsm_update_rename_array
*u
= (void *) update
->buf
;
8627 char name
[MAX_RAID_SERIAL_LEN
+1];
8628 int target
= u
->dev_idx
;
8629 struct active_array
*a
;
8630 struct imsm_dev
*dev
;
8632 /* sanity check that we are not affecting the uuid of
8635 snprintf(name
, MAX_RAID_SERIAL_LEN
, "%s", (char *) u
->name
);
8636 name
[MAX_RAID_SERIAL_LEN
] = '\0';
8637 for (a
= st
->arrays
; a
; a
= a
->next
)
8638 if (a
->info
.container_member
== target
)
8640 dev
= get_imsm_dev(super
, u
->dev_idx
);
8641 if (a
|| !dev
|| !check_name(super
, name
, 1)) {
8642 dprintf("failed to rename subarray-%d\n", target
);
8646 snprintf((char *) dev
->volume
, MAX_RAID_SERIAL_LEN
, "%s", name
);
8647 super
->updates_pending
++;
8650 case update_add_remove_disk
: {
8651 /* we may be able to repair some arrays if disks are
8652 * being added, check teh status of add_remove_disk
8653 * if discs has been added.
8655 if (add_remove_disk_update(super
)) {
8656 struct active_array
*a
;
8658 super
->updates_pending
++;
8659 for (a
= st
->arrays
; a
; a
= a
->next
)
8660 a
->check_degraded
= 1;
8665 fprintf(stderr
, "error: unsuported process update type:"
8666 "(type: %d)\n", type
);
8670 static struct mdinfo
*get_spares_for_grow(struct supertype
*st
);
8672 static void imsm_prepare_update(struct supertype
*st
,
8673 struct metadata_update
*update
)
8676 * Allocate space to hold new disk entries, raid-device entries or a new
8677 * mpb if necessary. The manager synchronously waits for updates to
8678 * complete in the monitor, so new mpb buffers allocated here can be
8679 * integrated by the monitor thread without worrying about live pointers
8680 * in the manager thread.
8682 enum imsm_update_type type
= *(enum imsm_update_type
*) update
->buf
;
8683 struct intel_super
*super
= st
->sb
;
8684 struct imsm_super
*mpb
= super
->anchor
;
8689 case update_general_migration_checkpoint
:
8690 dprintf("imsm: prepare_update() "
8691 "for update_general_migration_checkpoint called\n");
8693 case update_takeover
: {
8694 struct imsm_update_takeover
*u
= (void *)update
->buf
;
8695 if (u
->direction
== R0_TO_R10
) {
8696 void **tail
= (void **)&update
->space_list
;
8697 struct imsm_dev
*dev
= get_imsm_dev(super
, u
->subarray
);
8698 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
8699 int num_members
= map
->num_members
;
8703 /* allocate memory for added disks */
8704 for (i
= 0; i
< num_members
; i
++) {
8705 size
= sizeof(struct dl
);
8706 space
= malloc(size
);
8715 /* allocate memory for new device */
8716 size
= sizeof_imsm_dev(super
->devlist
->dev
, 0) +
8717 (num_members
* sizeof(__u32
));
8718 space
= malloc(size
);
8727 len
= disks_to_mpb_size(num_members
* 2);
8729 /* if allocation didn't success, free buffer */
8730 while (update
->space_list
) {
8731 void **sp
= update
->space_list
;
8732 update
->space_list
= *sp
;
8740 case update_reshape_container_disks
: {
8741 /* Every raid device in the container is about to
8742 * gain some more devices, and we will enter a
8744 * So each 'imsm_map' will be bigger, and the imsm_vol
8745 * will now hold 2 of them.
8746 * Thus we need new 'struct imsm_dev' allocations sized
8747 * as sizeof_imsm_dev but with more devices in both maps.
8749 struct imsm_update_reshape
*u
= (void *)update
->buf
;
8750 struct intel_dev
*dl
;
8751 void **space_tail
= (void**)&update
->space_list
;
8753 dprintf("imsm: imsm_prepare_update() for update_reshape\n");
8755 for (dl
= super
->devlist
; dl
; dl
= dl
->next
) {
8756 int size
= sizeof_imsm_dev(dl
->dev
, 1);
8758 if (u
->new_raid_disks
> u
->old_raid_disks
)
8759 size
+= sizeof(__u32
)*2*
8760 (u
->new_raid_disks
- u
->old_raid_disks
);
8769 len
= disks_to_mpb_size(u
->new_raid_disks
);
8770 dprintf("New anchor length is %llu\n", (unsigned long long)len
);
8773 case update_reshape_migration
: {
8774 /* for migration level 0->5 we need to add disks
8775 * so the same as for container operation we will copy
8776 * device to the bigger location.
8777 * in memory prepared device and new disk area are prepared
8778 * for usage in process update
8780 struct imsm_update_reshape_migration
*u
= (void *)update
->buf
;
8781 struct intel_dev
*id
;
8782 void **space_tail
= (void **)&update
->space_list
;
8785 int current_level
= -1;
8787 dprintf("imsm: imsm_prepare_update() for update_reshape\n");
8789 /* add space for bigger array in update
8791 for (id
= super
->devlist
; id
; id
= id
->next
) {
8792 if (id
->index
== (unsigned)u
->subdev
) {
8793 size
= sizeof_imsm_dev(id
->dev
, 1);
8794 if (u
->new_raid_disks
> u
->old_raid_disks
)
8795 size
+= sizeof(__u32
)*2*
8796 (u
->new_raid_disks
- u
->old_raid_disks
);
8806 if (update
->space_list
== NULL
)
8809 /* add space for disk in update
8811 size
= sizeof(struct dl
);
8814 free(update
->space_list
);
8815 update
->space_list
= NULL
;
8822 /* add spare device to update
8824 for (id
= super
->devlist
; id
; id
= id
->next
)
8825 if (id
->index
== (unsigned)u
->subdev
) {
8826 struct imsm_dev
*dev
;
8827 struct imsm_map
*map
;
8829 dev
= get_imsm_dev(super
, u
->subdev
);
8830 map
= get_imsm_map(dev
, MAP_0
);
8831 current_level
= map
->raid_level
;
8834 if ((u
->new_level
== 5) && (u
->new_level
!= current_level
)) {
8835 struct mdinfo
*spares
;
8837 spares
= get_spares_for_grow(st
);
8845 makedev(dev
->disk
.major
,
8847 dl
= get_disk_super(super
,
8850 dl
->index
= u
->old_raid_disks
;
8856 len
= disks_to_mpb_size(u
->new_raid_disks
);
8857 dprintf("New anchor length is %llu\n", (unsigned long long)len
);
8860 case update_size_change
: {
8863 case update_create_array
: {
8864 struct imsm_update_create_array
*u
= (void *) update
->buf
;
8865 struct intel_dev
*dv
;
8866 struct imsm_dev
*dev
= &u
->dev
;
8867 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
8869 struct disk_info
*inf
;
8873 inf
= get_disk_info(u
);
8874 len
= sizeof_imsm_dev(dev
, 1);
8875 /* allocate a new super->devlist entry */
8876 dv
= malloc(sizeof(*dv
));
8878 dv
->dev
= malloc(len
);
8883 update
->space
= NULL
;
8887 /* count how many spares will be converted to members */
8888 for (i
= 0; i
< map
->num_members
; i
++) {
8889 dl
= serial_to_dl(inf
[i
].serial
, super
);
8891 /* hmm maybe it failed?, nothing we can do about
8896 if (count_memberships(dl
, super
) == 0)
8899 len
+= activate
* sizeof(struct imsm_disk
);
8906 /* check if we need a larger metadata buffer */
8907 if (super
->next_buf
)
8908 buf_len
= super
->next_len
;
8910 buf_len
= super
->len
;
8912 if (__le32_to_cpu(mpb
->mpb_size
) + len
> buf_len
) {
8913 /* ok we need a larger buf than what is currently allocated
8914 * if this allocation fails process_update will notice that
8915 * ->next_len is set and ->next_buf is NULL
8917 buf_len
= ROUND_UP(__le32_to_cpu(mpb
->mpb_size
) + len
, 512);
8918 if (super
->next_buf
)
8919 free(super
->next_buf
);
8921 super
->next_len
= buf_len
;
8922 if (posix_memalign(&super
->next_buf
, 512, buf_len
) == 0)
8923 memset(super
->next_buf
, 0, buf_len
);
8925 super
->next_buf
= NULL
;
8929 /* must be called while manager is quiesced */
8930 static void imsm_delete(struct intel_super
*super
, struct dl
**dlp
, unsigned index
)
8932 struct imsm_super
*mpb
= super
->anchor
;
8934 struct imsm_dev
*dev
;
8935 struct imsm_map
*map
;
8936 int i
, j
, num_members
;
8939 dprintf("%s: deleting device[%d] from imsm_super\n",
8942 /* shift all indexes down one */
8943 for (iter
= super
->disks
; iter
; iter
= iter
->next
)
8944 if (iter
->index
> (int)index
)
8946 for (iter
= super
->missing
; iter
; iter
= iter
->next
)
8947 if (iter
->index
> (int)index
)
8950 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
8951 dev
= get_imsm_dev(super
, i
);
8952 map
= get_imsm_map(dev
, MAP_0
);
8953 num_members
= map
->num_members
;
8954 for (j
= 0; j
< num_members
; j
++) {
8955 /* update ord entries being careful not to propagate
8956 * ord-flags to the first map
8958 ord
= get_imsm_ord_tbl_ent(dev
, j
, MAP_X
);
8960 if (ord_to_idx(ord
) <= index
)
8963 map
= get_imsm_map(dev
, MAP_0
);
8964 set_imsm_ord_tbl_ent(map
, j
, ord_to_idx(ord
- 1));
8965 map
= get_imsm_map(dev
, MAP_1
);
8967 set_imsm_ord_tbl_ent(map
, j
, ord
- 1);
8972 super
->updates_pending
++;
8974 struct dl
*dl
= *dlp
;
8976 *dlp
= (*dlp
)->next
;
8977 __free_imsm_disk(dl
);
8980 #endif /* MDASSEMBLE */
8982 static void close_targets(int *targets
, int new_disks
)
8989 for (i
= 0; i
< new_disks
; i
++) {
8990 if (targets
[i
] >= 0) {
8997 static int imsm_get_allowed_degradation(int level
, int raid_disks
,
8998 struct intel_super
*super
,
8999 struct imsm_dev
*dev
)
9005 struct imsm_map
*map
;
9008 ret_val
= raid_disks
/2;
9009 /* check map if all disks pairs not failed
9012 map
= get_imsm_map(dev
, MAP_0
);
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)
9022 map
= get_imsm_map(dev
, MAP_1
);
9023 /* if there is no second map
9024 * result can be returned
9028 /* check degradation in second map
9030 for (i
= 0; i
< ret_val
; i
++) {
9031 int degradation
= 0;
9032 if (get_imsm_disk(super
, i
) == NULL
)
9034 if (get_imsm_disk(super
, i
+ 1) == NULL
)
9036 if (degradation
== 2)
9051 /*******************************************************************************
9052 * Function: open_backup_targets
9053 * Description: Function opens file descriptors for all devices given in
9056 * info : general array info
9057 * raid_disks : number of disks
9058 * raid_fds : table of device's file descriptors
9059 * super : intel super for raid10 degradation check
9060 * dev : intel device for raid10 degradation check
9064 ******************************************************************************/
9065 int open_backup_targets(struct mdinfo
*info
, int raid_disks
, int *raid_fds
,
9066 struct intel_super
*super
, struct imsm_dev
*dev
)
9072 for (i
= 0; i
< raid_disks
; i
++)
9075 for (sd
= info
->devs
; sd
; sd
= sd
->next
) {
9078 if (sd
->disk
.state
& (1<<MD_DISK_FAULTY
)) {
9079 dprintf("disk is faulty!!\n");
9083 if ((sd
->disk
.raid_disk
>= raid_disks
) ||
9084 (sd
->disk
.raid_disk
< 0))
9087 dn
= map_dev(sd
->disk
.major
,
9089 raid_fds
[sd
->disk
.raid_disk
] = dev_open(dn
, O_RDWR
);
9090 if (raid_fds
[sd
->disk
.raid_disk
] < 0) {
9091 fprintf(stderr
, "cannot open component\n");
9096 /* check if maximum array degradation level is not exceeded
9098 if ((raid_disks
- opened
) >
9099 imsm_get_allowed_degradation(info
->new_level
,
9102 fprintf(stderr
, "Not enough disks can be opened.\n");
9103 close_targets(raid_fds
, raid_disks
);
9110 /*******************************************************************************
9111 * Function: init_migr_record_imsm
9112 * Description: Function inits imsm migration record
9114 * super : imsm internal array info
9115 * dev : device under migration
9116 * info : general array info to find the smallest device
9119 ******************************************************************************/
9120 void init_migr_record_imsm(struct supertype
*st
, struct imsm_dev
*dev
,
9121 struct mdinfo
*info
)
9123 struct intel_super
*super
= st
->sb
;
9124 struct migr_record
*migr_rec
= super
->migr_rec
;
9126 unsigned long long dsize
, dev_sectors
;
9127 long long unsigned min_dev_sectors
= -1LLU;
9131 struct imsm_map
*map_dest
= get_imsm_map(dev
, MAP_0
);
9132 struct imsm_map
*map_src
= get_imsm_map(dev
, MAP_1
);
9133 unsigned long long num_migr_units
;
9134 unsigned long long array_blocks
;
9136 memset(migr_rec
, 0, sizeof(struct migr_record
));
9137 migr_rec
->family_num
= __cpu_to_le32(super
->anchor
->family_num
);
9139 /* only ascending reshape supported now */
9140 migr_rec
->ascending_migr
= __cpu_to_le32(1);
9142 migr_rec
->dest_depth_per_unit
= GEN_MIGR_AREA_SIZE
/
9143 max(map_dest
->blocks_per_strip
, map_src
->blocks_per_strip
);
9144 migr_rec
->dest_depth_per_unit
*=
9145 max(map_dest
->blocks_per_strip
, map_src
->blocks_per_strip
);
9146 new_data_disks
= imsm_num_data_members(dev
, MAP_0
);
9147 migr_rec
->blocks_per_unit
=
9148 __cpu_to_le32(migr_rec
->dest_depth_per_unit
* new_data_disks
);
9149 migr_rec
->dest_depth_per_unit
=
9150 __cpu_to_le32(migr_rec
->dest_depth_per_unit
);
9151 array_blocks
= info
->component_size
* new_data_disks
;
9153 array_blocks
/ __le32_to_cpu(migr_rec
->blocks_per_unit
);
9155 if (array_blocks
% __le32_to_cpu(migr_rec
->blocks_per_unit
))
9157 migr_rec
->num_migr_units
= __cpu_to_le32(num_migr_units
);
9159 migr_rec
->post_migr_vol_cap
= dev
->size_low
;
9160 migr_rec
->post_migr_vol_cap_hi
= dev
->size_high
;
9163 /* Find the smallest dev */
9164 for (sd
= info
->devs
; sd
; sd
= sd
->next
) {
9165 sprintf(nm
, "%d:%d", sd
->disk
.major
, sd
->disk
.minor
);
9166 fd
= dev_open(nm
, O_RDONLY
);
9169 get_dev_size(fd
, NULL
, &dsize
);
9170 dev_sectors
= dsize
/ 512;
9171 if (dev_sectors
< min_dev_sectors
)
9172 min_dev_sectors
= dev_sectors
;
9175 migr_rec
->ckpt_area_pba
= __cpu_to_le32(min_dev_sectors
-
9176 RAID_DISK_RESERVED_BLOCKS_IMSM_HI
);
9178 write_imsm_migr_rec(st
);
9183 /*******************************************************************************
9184 * Function: save_backup_imsm
9185 * Description: Function saves critical data stripes to Migration Copy Area
9186 * and updates the current migration unit status.
9187 * Use restore_stripes() to form a destination stripe,
9188 * and to write it to the Copy Area.
9190 * st : supertype information
9191 * dev : imsm device that backup is saved for
9192 * info : general array info
9193 * buf : input buffer
9194 * length : length of data to backup (blocks_per_unit)
9198 ******************************************************************************/
9199 int save_backup_imsm(struct supertype
*st
,
9200 struct imsm_dev
*dev
,
9201 struct mdinfo
*info
,
9206 struct intel_super
*super
= st
->sb
;
9207 unsigned long long *target_offsets
= NULL
;
9208 int *targets
= NULL
;
9210 struct imsm_map
*map_dest
= get_imsm_map(dev
, MAP_0
);
9211 int new_disks
= map_dest
->num_members
;
9212 int dest_layout
= 0;
9214 unsigned long long start
;
9215 int data_disks
= imsm_num_data_members(dev
, MAP_0
);
9217 targets
= malloc(new_disks
* sizeof(int));
9221 for (i
= 0; i
< new_disks
; i
++)
9224 target_offsets
= malloc(new_disks
* sizeof(unsigned long long));
9225 if (!target_offsets
)
9228 start
= info
->reshape_progress
* 512;
9229 for (i
= 0; i
< new_disks
; i
++) {
9230 target_offsets
[i
] = (unsigned long long)
9231 __le32_to_cpu(super
->migr_rec
->ckpt_area_pba
) * 512;
9232 /* move back copy area adderss, it will be moved forward
9233 * in restore_stripes() using start input variable
9235 target_offsets
[i
] -= start
/data_disks
;
9238 if (open_backup_targets(info
, new_disks
, targets
,
9242 dest_layout
= imsm_level_to_layout(map_dest
->raid_level
);
9243 dest_chunk
= __le16_to_cpu(map_dest
->blocks_per_strip
) * 512;
9245 if (restore_stripes(targets
, /* list of dest devices */
9246 target_offsets
, /* migration record offsets */
9249 map_dest
->raid_level
,
9251 -1, /* source backup file descriptor */
9252 0, /* input buf offset
9253 * always 0 buf is already offseted */
9257 fprintf(stderr
, Name
": Error restoring stripes\n");
9265 close_targets(targets
, new_disks
);
9268 free(target_offsets
);
9273 /*******************************************************************************
9274 * Function: save_checkpoint_imsm
9275 * Description: Function called for current unit status update
9276 * in the migration record. It writes it to disk.
9278 * super : imsm internal array info
9279 * info : general array info
9283 * 2: failure, means no valid migration record
9284 * / no general migration in progress /
9285 ******************************************************************************/
9286 int save_checkpoint_imsm(struct supertype
*st
, struct mdinfo
*info
, int state
)
9288 struct intel_super
*super
= st
->sb
;
9289 unsigned long long blocks_per_unit
;
9290 unsigned long long curr_migr_unit
;
9292 if (load_imsm_migr_rec(super
, info
) != 0) {
9293 dprintf("imsm: ERROR: Cannot read migration record "
9294 "for checkpoint save.\n");
9298 blocks_per_unit
= __le32_to_cpu(super
->migr_rec
->blocks_per_unit
);
9299 if (blocks_per_unit
== 0) {
9300 dprintf("imsm: no migration in progress.\n");
9303 curr_migr_unit
= info
->reshape_progress
/ blocks_per_unit
;
9304 /* check if array is alligned to copy area
9305 * if it is not alligned, add one to current migration unit value
9306 * this can happend on array reshape finish only
9308 if (info
->reshape_progress
% blocks_per_unit
)
9311 super
->migr_rec
->curr_migr_unit
=
9312 __cpu_to_le32(curr_migr_unit
);
9313 super
->migr_rec
->rec_status
= __cpu_to_le32(state
);
9314 super
->migr_rec
->dest_1st_member_lba
=
9315 __cpu_to_le32(curr_migr_unit
*
9316 __le32_to_cpu(super
->migr_rec
->dest_depth_per_unit
));
9317 if (write_imsm_migr_rec(st
) < 0) {
9318 dprintf("imsm: Cannot write migration record "
9319 "outside backup area\n");
9326 /*******************************************************************************
9327 * Function: recover_backup_imsm
9328 * Description: Function recovers critical data from the Migration Copy Area
9329 * while assembling an array.
9331 * super : imsm internal array info
9332 * info : general array info
9334 * 0 : success (or there is no data to recover)
9336 ******************************************************************************/
9337 int recover_backup_imsm(struct supertype
*st
, struct mdinfo
*info
)
9339 struct intel_super
*super
= st
->sb
;
9340 struct migr_record
*migr_rec
= super
->migr_rec
;
9341 struct imsm_map
*map_dest
= NULL
;
9342 struct intel_dev
*id
= NULL
;
9343 unsigned long long read_offset
;
9344 unsigned long long write_offset
;
9346 int *targets
= NULL
;
9347 int new_disks
, i
, err
;
9350 unsigned long curr_migr_unit
= __le32_to_cpu(migr_rec
->curr_migr_unit
);
9351 unsigned long num_migr_units
= __le32_to_cpu(migr_rec
->num_migr_units
);
9353 int skipped_disks
= 0;
9355 err
= sysfs_get_str(info
, NULL
, "array_state", (char *)buffer
, 20);
9359 /* recover data only during assemblation */
9360 if (strncmp(buffer
, "inactive", 8) != 0)
9362 /* no data to recover */
9363 if (__le32_to_cpu(migr_rec
->rec_status
) == UNIT_SRC_NORMAL
)
9365 if (curr_migr_unit
>= num_migr_units
)
9368 /* find device during reshape */
9369 for (id
= super
->devlist
; id
; id
= id
->next
)
9370 if (is_gen_migration(id
->dev
))
9375 map_dest
= get_imsm_map(id
->dev
, MAP_0
);
9376 new_disks
= map_dest
->num_members
;
9378 read_offset
= (unsigned long long)
9379 __le32_to_cpu(migr_rec
->ckpt_area_pba
) * 512;
9381 write_offset
= ((unsigned long long)
9382 __le32_to_cpu(migr_rec
->dest_1st_member_lba
) +
9383 pba_of_lba0(map_dest
)) * 512;
9385 unit_len
= __le32_to_cpu(migr_rec
->dest_depth_per_unit
) * 512;
9386 if (posix_memalign((void **)&buf
, 512, unit_len
) != 0)
9388 targets
= malloc(new_disks
* sizeof(int));
9392 if (open_backup_targets(info
, new_disks
, targets
, super
, id
->dev
)) {
9394 Name
": Cannot open some devices belonging to array.\n");
9398 for (i
= 0; i
< new_disks
; i
++) {
9399 if (targets
[i
] < 0) {
9403 if (lseek64(targets
[i
], read_offset
, SEEK_SET
) < 0) {
9405 Name
": Cannot seek to block: %s\n",
9410 if ((unsigned)read(targets
[i
], buf
, unit_len
) != unit_len
) {
9412 Name
": Cannot read copy area block: %s\n",
9417 if (lseek64(targets
[i
], write_offset
, SEEK_SET
) < 0) {
9419 Name
": Cannot seek to block: %s\n",
9424 if ((unsigned)write(targets
[i
], buf
, unit_len
) != unit_len
) {
9426 Name
": Cannot restore block: %s\n",
9433 if (skipped_disks
> imsm_get_allowed_degradation(info
->new_level
,
9438 Name
": Cannot restore data from backup."
9439 " Too many failed disks\n");
9443 if (save_checkpoint_imsm(st
, info
, UNIT_SRC_NORMAL
)) {
9444 /* ignore error == 2, this can mean end of reshape here
9446 dprintf("imsm: Cannot write checkpoint to "
9447 "migration record (UNIT_SRC_NORMAL) during restart\n");
9453 for (i
= 0; i
< new_disks
; i
++)
9462 static char disk_by_path
[] = "/dev/disk/by-path/";
9464 static const char *imsm_get_disk_controller_domain(const char *path
)
9466 char disk_path
[PATH_MAX
];
9470 strncpy(disk_path
, disk_by_path
, PATH_MAX
- 1);
9471 strncat(disk_path
, path
, PATH_MAX
- strlen(disk_path
) - 1);
9472 if (stat(disk_path
, &st
) == 0) {
9473 struct sys_dev
* hba
;
9476 path
= devt_to_devpath(st
.st_rdev
);
9479 hba
= find_disk_attached_hba(-1, path
);
9480 if (hba
&& hba
->type
== SYS_DEV_SAS
)
9482 else if (hba
&& hba
->type
== SYS_DEV_SATA
)
9486 dprintf("path: %s hba: %s attached: %s\n",
9487 path
, (hba
) ? hba
->path
: "NULL", drv
);
9495 static int imsm_find_array_minor_by_subdev(int subdev
, int container
, int *minor
)
9497 char subdev_name
[20];
9498 struct mdstat_ent
*mdstat
;
9500 sprintf(subdev_name
, "%d", subdev
);
9501 mdstat
= mdstat_by_subdev(subdev_name
, container
);
9505 *minor
= mdstat
->devnum
;
9506 free_mdstat(mdstat
);
9510 static int imsm_reshape_is_allowed_on_container(struct supertype
*st
,
9511 struct geo_params
*geo
,
9512 int *old_raid_disks
,
9515 /* currently we only support increasing the number of devices
9516 * for a container. This increases the number of device for each
9517 * member array. They must all be RAID0 or RAID5.
9520 struct mdinfo
*info
, *member
;
9521 int devices_that_can_grow
= 0;
9523 dprintf("imsm: imsm_reshape_is_allowed_on_container(ENTER): "
9524 "st->devnum = (%i)\n",
9527 if (geo
->size
!= -1 ||
9528 geo
->level
!= UnSet
||
9529 geo
->layout
!= UnSet
||
9530 geo
->chunksize
!= 0 ||
9531 geo
->raid_disks
== UnSet
) {
9532 dprintf("imsm: Container operation is allowed for "
9533 "raid disks number change only.\n");
9537 if (direction
== ROLLBACK_METADATA_CHANGES
) {
9538 dprintf("imsm: Metadata changes rollback is not supported for "
9539 "container operation.\n");
9543 info
= container_content_imsm(st
, NULL
);
9544 for (member
= info
; member
; member
= member
->next
) {
9548 dprintf("imsm: checking device_num: %i\n",
9549 member
->container_member
);
9551 if (geo
->raid_disks
<= member
->array
.raid_disks
) {
9552 /* we work on container for Online Capacity Expansion
9553 * only so raid_disks has to grow
9555 dprintf("imsm: for container operation raid disks "
9556 "increase is required\n");
9560 if ((info
->array
.level
!= 0) &&
9561 (info
->array
.level
!= 5)) {
9562 /* we cannot use this container with other raid level
9564 dprintf("imsm: for container operation wrong"
9565 " raid level (%i) detected\n",
9569 /* check for platform support
9570 * for this raid level configuration
9572 struct intel_super
*super
= st
->sb
;
9573 if (!is_raid_level_supported(super
->orom
,
9574 member
->array
.level
,
9576 dprintf("platform does not support raid%d with"
9580 geo
->raid_disks
> 1 ? "s" : "");
9583 /* check if component size is aligned to chunk size
9585 if (info
->component_size
%
9586 (info
->array
.chunk_size
/512)) {
9587 dprintf("Component size is not aligned to "
9593 if (*old_raid_disks
&&
9594 info
->array
.raid_disks
!= *old_raid_disks
)
9596 *old_raid_disks
= info
->array
.raid_disks
;
9598 /* All raid5 and raid0 volumes in container
9599 * have to be ready for Online Capacity Expansion
9600 * so they need to be assembled. We have already
9601 * checked that no recovery etc is happening.
9603 result
= imsm_find_array_minor_by_subdev(member
->container_member
,
9607 dprintf("imsm: cannot find array\n");
9610 devices_that_can_grow
++;
9613 if (!member
&& devices_that_can_grow
)
9617 dprintf("\tContainer operation allowed\n");
9619 dprintf("\tError: %i\n", ret_val
);
9624 /* Function: get_spares_for_grow
9625 * Description: Allocates memory and creates list of spare devices
9626 * avaliable in container. Checks if spare drive size is acceptable.
9627 * Parameters: Pointer to the supertype structure
9628 * Returns: Pointer to the list of spare devices (mdinfo structure) on success,
9631 static struct mdinfo
*get_spares_for_grow(struct supertype
*st
)
9633 unsigned long long min_size
= min_acceptable_spare_size_imsm(st
);
9634 return container_choose_spares(st
, min_size
, NULL
, NULL
, NULL
, 0);
9637 /******************************************************************************
9638 * function: imsm_create_metadata_update_for_reshape
9639 * Function creates update for whole IMSM container.
9641 ******************************************************************************/
9642 static int imsm_create_metadata_update_for_reshape(
9643 struct supertype
*st
,
9644 struct geo_params
*geo
,
9646 struct imsm_update_reshape
**updatep
)
9648 struct intel_super
*super
= st
->sb
;
9649 struct imsm_super
*mpb
= super
->anchor
;
9650 int update_memory_size
= 0;
9651 struct imsm_update_reshape
*u
= NULL
;
9652 struct mdinfo
*spares
= NULL
;
9654 int delta_disks
= 0;
9657 dprintf("imsm_update_metadata_for_reshape(enter) raid_disks = %i\n",
9660 delta_disks
= geo
->raid_disks
- old_raid_disks
;
9662 /* size of all update data without anchor */
9663 update_memory_size
= sizeof(struct imsm_update_reshape
);
9665 /* now add space for spare disks that we need to add. */
9666 update_memory_size
+= sizeof(u
->new_disks
[0]) * (delta_disks
- 1);
9668 u
= calloc(1, update_memory_size
);
9671 "cannot get memory for imsm_update_reshape update\n");
9674 u
->type
= update_reshape_container_disks
;
9675 u
->old_raid_disks
= old_raid_disks
;
9676 u
->new_raid_disks
= geo
->raid_disks
;
9678 /* now get spare disks list
9680 spares
= get_spares_for_grow(st
);
9683 || delta_disks
> spares
->array
.spare_disks
) {
9684 fprintf(stderr
, Name
": imsm: ERROR: Cannot get spare devices "
9685 "for %s.\n", geo
->dev_name
);
9690 /* we have got spares
9691 * update disk list in imsm_disk list table in anchor
9693 dprintf("imsm: %i spares are available.\n\n",
9694 spares
->array
.spare_disks
);
9697 for (i
= 0; i
< delta_disks
; i
++) {
9702 u
->new_disks
[i
] = makedev(dev
->disk
.major
,
9704 dl
= get_disk_super(super
, dev
->disk
.major
, dev
->disk
.minor
);
9705 dl
->index
= mpb
->num_disks
;
9715 dprintf("imsm: reshape update preparation :");
9716 if (i
== delta_disks
) {
9719 return update_memory_size
;
9722 dprintf(" Error\n");
9728 /******************************************************************************
9729 * function: imsm_create_metadata_update_for_size_change()
9730 * Creates update for IMSM array for array size change.
9732 ******************************************************************************/
9733 static int imsm_create_metadata_update_for_size_change(
9734 struct supertype
*st
,
9735 struct geo_params
*geo
,
9736 struct imsm_update_size_change
**updatep
)
9738 struct intel_super
*super
= st
->sb
;
9739 int update_memory_size
= 0;
9740 struct imsm_update_size_change
*u
= NULL
;
9742 dprintf("imsm_create_metadata_update_for_size_change(enter)"
9743 " New size = %llu\n", geo
->size
);
9745 /* size of all update data without anchor */
9746 update_memory_size
= sizeof(struct imsm_update_size_change
);
9748 u
= calloc(1, update_memory_size
);
9750 dprintf("error: cannot get memory for "
9751 "imsm_create_metadata_update_for_size_change\n");
9754 u
->type
= update_size_change
;
9755 u
->subdev
= super
->current_vol
;
9756 u
->new_size
= geo
->size
;
9758 dprintf("imsm: reshape update preparation : OK\n");
9761 return update_memory_size
;
9764 /******************************************************************************
9765 * function: imsm_create_metadata_update_for_migration()
9766 * Creates update for IMSM array.
9768 ******************************************************************************/
9769 static int imsm_create_metadata_update_for_migration(
9770 struct supertype
*st
,
9771 struct geo_params
*geo
,
9772 struct imsm_update_reshape_migration
**updatep
)
9774 struct intel_super
*super
= st
->sb
;
9775 int update_memory_size
= 0;
9776 struct imsm_update_reshape_migration
*u
= NULL
;
9777 struct imsm_dev
*dev
;
9778 int previous_level
= -1;
9780 dprintf("imsm_create_metadata_update_for_migration(enter)"
9781 " New Level = %i\n", geo
->level
);
9783 /* size of all update data without anchor */
9784 update_memory_size
= sizeof(struct imsm_update_reshape_migration
);
9786 u
= calloc(1, update_memory_size
);
9788 dprintf("error: cannot get memory for "
9789 "imsm_create_metadata_update_for_migration\n");
9792 u
->type
= update_reshape_migration
;
9793 u
->subdev
= super
->current_vol
;
9794 u
->new_level
= geo
->level
;
9795 u
->new_layout
= geo
->layout
;
9796 u
->new_raid_disks
= u
->old_raid_disks
= geo
->raid_disks
;
9797 u
->new_disks
[0] = -1;
9798 u
->new_chunksize
= -1;
9800 dev
= get_imsm_dev(super
, u
->subdev
);
9802 struct imsm_map
*map
;
9804 map
= get_imsm_map(dev
, MAP_0
);
9806 int current_chunk_size
=
9807 __le16_to_cpu(map
->blocks_per_strip
) / 2;
9809 if (geo
->chunksize
!= current_chunk_size
) {
9810 u
->new_chunksize
= geo
->chunksize
/ 1024;
9812 "chunk size change from %i to %i\n",
9813 current_chunk_size
, u
->new_chunksize
);
9815 previous_level
= map
->raid_level
;
9818 if ((geo
->level
== 5) && (previous_level
== 0)) {
9819 struct mdinfo
*spares
= NULL
;
9821 u
->new_raid_disks
++;
9822 spares
= get_spares_for_grow(st
);
9823 if ((spares
== NULL
) || (spares
->array
.spare_disks
< 1)) {
9826 update_memory_size
= 0;
9827 dprintf("error: cannot get spare device "
9828 "for requested migration");
9833 dprintf("imsm: reshape update preparation : OK\n");
9836 return update_memory_size
;
9839 static void imsm_update_metadata_locally(struct supertype
*st
,
9842 struct metadata_update mu
;
9847 mu
.space_list
= NULL
;
9849 imsm_prepare_update(st
, &mu
);
9850 imsm_process_update(st
, &mu
);
9852 while (mu
.space_list
) {
9853 void **space
= mu
.space_list
;
9854 mu
.space_list
= *space
;
9859 /***************************************************************************
9860 * Function: imsm_analyze_change
9861 * Description: Function analyze change for single volume
9862 * and validate if transition is supported
9863 * Parameters: Geometry parameters, supertype structure,
9864 * metadata change direction (apply/rollback)
9865 * Returns: Operation type code on success, -1 if fail
9866 ****************************************************************************/
9867 enum imsm_reshape_type
imsm_analyze_change(struct supertype
*st
,
9868 struct geo_params
*geo
,
9875 /* number of added/removed disks in operation result */
9876 int devNumChange
= 0;
9877 /* imsm compatible layout value for array geometry verification */
9878 int imsm_layout
= -1;
9880 struct imsm_dev
*dev
;
9881 struct intel_super
*super
;
9882 long long current_size
;
9884 getinfo_super_imsm_volume(st
, &info
, NULL
);
9885 if ((geo
->level
!= info
.array
.level
) &&
9886 (geo
->level
>= 0) &&
9887 (geo
->level
!= UnSet
)) {
9888 switch (info
.array
.level
) {
9890 if (geo
->level
== 5) {
9891 change
= CH_MIGRATION
;
9892 if (geo
->layout
!= ALGORITHM_LEFT_ASYMMETRIC
) {
9894 Name
" Error. Requested Layout "
9895 "not supported (left-asymmetric layout "
9896 "is supported only)!\n");
9898 goto analyse_change_exit
;
9900 imsm_layout
= geo
->layout
;
9902 devNumChange
= 1; /* parity disk added */
9903 } else if (geo
->level
== 10) {
9904 change
= CH_TAKEOVER
;
9906 devNumChange
= 2; /* two mirrors added */
9907 imsm_layout
= 0x102; /* imsm supported layout */
9912 if (geo
->level
== 0) {
9913 change
= CH_TAKEOVER
;
9915 devNumChange
= -(geo
->raid_disks
/2);
9916 imsm_layout
= 0; /* imsm raid0 layout */
9922 Name
" Error. Level Migration from %d to %d "
9924 info
.array
.level
, geo
->level
);
9925 goto analyse_change_exit
;
9928 geo
->level
= info
.array
.level
;
9930 if ((geo
->layout
!= info
.array
.layout
)
9931 && ((geo
->layout
!= UnSet
) && (geo
->layout
!= -1))) {
9932 change
= CH_MIGRATION
;
9933 if ((info
.array
.layout
== 0)
9934 && (info
.array
.level
== 5)
9935 && (geo
->layout
== 5)) {
9936 /* reshape 5 -> 4 */
9937 } else if ((info
.array
.layout
== 5)
9938 && (info
.array
.level
== 5)
9939 && (geo
->layout
== 0)) {
9940 /* reshape 4 -> 5 */
9945 Name
" Error. Layout Migration from %d to %d "
9947 info
.array
.layout
, geo
->layout
);
9949 goto analyse_change_exit
;
9952 geo
->layout
= info
.array
.layout
;
9953 if (imsm_layout
== -1)
9954 imsm_layout
= info
.array
.layout
;
9957 if ((geo
->chunksize
> 0) && (geo
->chunksize
!= UnSet
)
9958 && (geo
->chunksize
!= info
.array
.chunk_size
))
9959 change
= CH_MIGRATION
;
9961 geo
->chunksize
= info
.array
.chunk_size
;
9963 chunk
= geo
->chunksize
/ 1024;
9966 dev
= get_imsm_dev(super
, super
->current_vol
);
9967 data_disks
= imsm_num_data_members(dev
, MAP_0
);
9968 /* compute current size per disk member
9970 current_size
= info
.custom_array_size
/ data_disks
;
9972 if (geo
->size
> 0) {
9973 /* align component size
9975 geo
->size
= imsm_component_size_aligment_check(
9976 get_imsm_raid_level(dev
->vol
.map
),
9981 if ((current_size
!= geo
->size
) && (geo
->size
>= 0)) {
9984 Name
" Error. Size change should be the only "
9985 "one at a time.\n");
9987 goto analyse_change_exit
;
9989 if ((super
->current_vol
+ 1) != super
->anchor
->num_raid_devs
) {
9991 Name
" Error. The last volume in container "
9992 "can be expanded only (%i/%i).\n",
9993 super
->current_vol
, st
->devnum
);
9994 goto analyse_change_exit
;
9996 if (geo
->size
== 0) {
9997 /* requested size change to the maximum available size
9999 unsigned long long freesize
;
10002 rv
= imsm_get_free_size(st
, dev
->vol
.map
->num_members
,
10003 0, chunk
, &freesize
);
10005 fprintf(stderr
, Name
" Error. Cannot find "
10006 "maximum available space.\n");
10008 goto analyse_change_exit
;
10010 geo
->size
= freesize
+ current_size
;
10012 /* align component size
10014 geo
->size
= imsm_component_size_aligment_check(
10015 get_imsm_raid_level(dev
->vol
.map
),
10020 if ((direction
== ROLLBACK_METADATA_CHANGES
)) {
10021 /* accept size for rollback only
10024 /* round size due to metadata compatibility
10026 geo
->size
= (geo
->size
>> SECT_PER_MB_SHIFT
)
10027 << SECT_PER_MB_SHIFT
;
10028 dprintf("Prepare update for size change to %llu\n",
10030 if (current_size
>= geo
->size
) {
10032 Name
" Error. Size expanssion is "
10033 "supported only (current size is %llu, "
10034 "requested size /rounded/ is %llu).\n",
10035 current_size
, geo
->size
);
10036 goto analyse_change_exit
;
10039 geo
->size
*= data_disks
;
10040 geo
->raid_disks
= dev
->vol
.map
->num_members
;
10041 change
= CH_ARRAY_SIZE
;
10043 if (!validate_geometry_imsm(st
,
10046 geo
->raid_disks
+ devNumChange
,
10053 struct intel_super
*super
= st
->sb
;
10054 struct imsm_super
*mpb
= super
->anchor
;
10056 if (mpb
->num_raid_devs
> 1) {
10058 Name
" Error. Cannot perform operation on %s"
10059 "- for this operation it MUST be single "
10060 "array in container\n",
10066 analyse_change_exit
:
10067 if ((direction
== ROLLBACK_METADATA_CHANGES
) &&
10068 ((change
== CH_MIGRATION
) || (change
== CH_TAKEOVER
))) {
10069 dprintf("imsm: Metadata changes rollback is not supported for "
10070 "migration and takeover operations.\n");
10076 int imsm_takeover(struct supertype
*st
, struct geo_params
*geo
)
10078 struct intel_super
*super
= st
->sb
;
10079 struct imsm_update_takeover
*u
;
10081 u
= malloc(sizeof(struct imsm_update_takeover
));
10085 u
->type
= update_takeover
;
10086 u
->subarray
= super
->current_vol
;
10088 /* 10->0 transition */
10089 if (geo
->level
== 0)
10090 u
->direction
= R10_TO_R0
;
10092 /* 0->10 transition */
10093 if (geo
->level
== 10)
10094 u
->direction
= R0_TO_R10
;
10096 /* update metadata locally */
10097 imsm_update_metadata_locally(st
, u
,
10098 sizeof(struct imsm_update_takeover
));
10099 /* and possibly remotely */
10100 if (st
->update_tail
)
10101 append_metadata_update(st
, u
,
10102 sizeof(struct imsm_update_takeover
));
10109 static int imsm_reshape_super(struct supertype
*st
, long long size
, int level
,
10110 int layout
, int chunksize
, int raid_disks
,
10111 int delta_disks
, char *backup
, char *dev
,
10112 int direction
, int verbose
)
10115 struct geo_params geo
;
10117 dprintf("imsm: reshape_super called.\n");
10119 memset(&geo
, 0, sizeof(struct geo_params
));
10121 geo
.dev_name
= dev
;
10122 geo
.dev_id
= st
->devnum
;
10125 geo
.layout
= layout
;
10126 geo
.chunksize
= chunksize
;
10127 geo
.raid_disks
= raid_disks
;
10128 if (delta_disks
!= UnSet
)
10129 geo
.raid_disks
+= delta_disks
;
10131 dprintf("\tfor level : %i\n", geo
.level
);
10132 dprintf("\tfor raid_disks : %i\n", geo
.raid_disks
);
10134 if (experimental() == 0)
10137 if (st
->container_dev
== st
->devnum
) {
10138 /* On container level we can only increase number of devices. */
10139 dprintf("imsm: info: Container operation\n");
10140 int old_raid_disks
= 0;
10142 if (imsm_reshape_is_allowed_on_container(
10143 st
, &geo
, &old_raid_disks
, direction
)) {
10144 struct imsm_update_reshape
*u
= NULL
;
10147 len
= imsm_create_metadata_update_for_reshape(
10148 st
, &geo
, old_raid_disks
, &u
);
10151 dprintf("imsm: Cannot prepare update\n");
10152 goto exit_imsm_reshape_super
;
10156 /* update metadata locally */
10157 imsm_update_metadata_locally(st
, u
, len
);
10158 /* and possibly remotely */
10159 if (st
->update_tail
)
10160 append_metadata_update(st
, u
, len
);
10165 fprintf(stderr
, Name
": (imsm) Operation "
10166 "is not allowed on this container\n");
10169 /* On volume level we support following operations
10170 * - takeover: raid10 -> raid0; raid0 -> raid10
10171 * - chunk size migration
10172 * - migration: raid5 -> raid0; raid0 -> raid5
10174 struct intel_super
*super
= st
->sb
;
10175 struct intel_dev
*dev
= super
->devlist
;
10176 int change
, devnum
;
10177 dprintf("imsm: info: Volume operation\n");
10178 /* find requested device */
10180 if (imsm_find_array_minor_by_subdev(
10181 dev
->index
, st
->container_dev
, &devnum
) == 0
10182 && devnum
== geo
.dev_id
)
10187 fprintf(stderr
, Name
" Cannot find %s (%i) subarray\n",
10188 geo
.dev_name
, geo
.dev_id
);
10189 goto exit_imsm_reshape_super
;
10191 super
->current_vol
= dev
->index
;
10192 change
= imsm_analyze_change(st
, &geo
, direction
);
10195 ret_val
= imsm_takeover(st
, &geo
);
10197 case CH_MIGRATION
: {
10198 struct imsm_update_reshape_migration
*u
= NULL
;
10200 imsm_create_metadata_update_for_migration(
10204 "Cannot prepare update\n");
10208 /* update metadata locally */
10209 imsm_update_metadata_locally(st
, u
, len
);
10210 /* and possibly remotely */
10211 if (st
->update_tail
)
10212 append_metadata_update(st
, u
, len
);
10217 case CH_ARRAY_SIZE
: {
10218 struct imsm_update_size_change
*u
= NULL
;
10220 imsm_create_metadata_update_for_size_change(
10224 "Cannot prepare update\n");
10228 /* update metadata locally */
10229 imsm_update_metadata_locally(st
, u
, len
);
10230 /* and possibly remotely */
10231 if (st
->update_tail
)
10232 append_metadata_update(st
, u
, len
);
10242 exit_imsm_reshape_super
:
10243 dprintf("imsm: reshape_super Exit code = %i\n", ret_val
);
10247 /*******************************************************************************
10248 * Function: wait_for_reshape_imsm
10249 * Description: Function writes new sync_max value and waits until
10250 * reshape process reach new position
10252 * sra : general array info
10253 * ndata : number of disks in new array's layout
10256 * 1 : there is no reshape in progress,
10258 ******************************************************************************/
10259 int wait_for_reshape_imsm(struct mdinfo
*sra
, int ndata
)
10261 int fd
= sysfs_get_fd(sra
, NULL
, "reshape_position");
10262 unsigned long long completed
;
10263 /* to_complete : new sync_max position */
10264 unsigned long long to_complete
= sra
->reshape_progress
;
10265 unsigned long long position_to_set
= to_complete
/ ndata
;
10268 dprintf("imsm: wait_for_reshape_imsm() "
10269 "cannot open reshape_position\n");
10273 if (sysfs_fd_get_ll(fd
, &completed
) < 0) {
10274 dprintf("imsm: wait_for_reshape_imsm() "
10275 "cannot read reshape_position (no reshape in progres)\n");
10280 if (completed
> to_complete
) {
10281 dprintf("imsm: wait_for_reshape_imsm() "
10282 "wrong next position to set %llu (%llu)\n",
10283 to_complete
, completed
);
10287 dprintf("Position set: %llu\n", position_to_set
);
10288 if (sysfs_set_num(sra
, NULL
, "sync_max",
10289 position_to_set
) != 0) {
10290 dprintf("imsm: wait_for_reshape_imsm() "
10291 "cannot set reshape position to %llu\n",
10302 select(fd
+1, &rfds
, NULL
, NULL
, NULL
);
10303 if (sysfs_get_str(sra
, NULL
, "sync_action",
10305 strncmp(action
, "reshape", 7) != 0)
10307 if (sysfs_fd_get_ll(fd
, &completed
) < 0) {
10308 dprintf("imsm: wait_for_reshape_imsm() "
10309 "cannot read reshape_position (in loop)\n");
10313 } while (completed
< to_complete
);
10319 /*******************************************************************************
10320 * Function: check_degradation_change
10321 * Description: Check that array hasn't become failed.
10323 * info : for sysfs access
10324 * sources : source disks descriptors
10325 * degraded: previous degradation level
10327 * degradation level
10328 ******************************************************************************/
10329 int check_degradation_change(struct mdinfo
*info
,
10333 unsigned long long new_degraded
;
10334 sysfs_get_ll(info
, NULL
, "degraded", &new_degraded
);
10335 if (new_degraded
!= (unsigned long long)degraded
) {
10336 /* check each device to ensure it is still working */
10339 for (sd
= info
->devs
; sd
; sd
= sd
->next
) {
10340 if (sd
->disk
.state
& (1<<MD_DISK_FAULTY
))
10342 if (sd
->disk
.state
& (1<<MD_DISK_SYNC
)) {
10344 if (sysfs_get_str(info
,
10345 sd
, "state", sbuf
, 20) < 0 ||
10346 strstr(sbuf
, "faulty") ||
10347 strstr(sbuf
, "in_sync") == NULL
) {
10348 /* this device is dead */
10349 sd
->disk
.state
= (1<<MD_DISK_FAULTY
);
10350 if (sd
->disk
.raid_disk
>= 0 &&
10351 sources
[sd
->disk
.raid_disk
] >= 0) {
10353 sd
->disk
.raid_disk
]);
10354 sources
[sd
->disk
.raid_disk
] =
10363 return new_degraded
;
10366 /*******************************************************************************
10367 * Function: imsm_manage_reshape
10368 * Description: Function finds array under reshape and it manages reshape
10369 * process. It creates stripes backups (if required) and sets
10372 * afd : Backup handle (nattive) - not used
10373 * sra : general array info
10374 * reshape : reshape parameters - not used
10375 * st : supertype structure
10376 * blocks : size of critical section [blocks]
10377 * fds : table of source device descriptor
10378 * offsets : start of array (offest per devices)
10380 * destfd : table of destination device descriptor
10381 * destoffsets : table of destination offsets (per device)
10383 * 1 : success, reshape is done
10385 ******************************************************************************/
10386 static int imsm_manage_reshape(
10387 int afd
, struct mdinfo
*sra
, struct reshape
*reshape
,
10388 struct supertype
*st
, unsigned long backup_blocks
,
10389 int *fds
, unsigned long long *offsets
,
10390 int dests
, int *destfd
, unsigned long long *destoffsets
)
10393 struct intel_super
*super
= st
->sb
;
10394 struct intel_dev
*dv
= NULL
;
10395 struct imsm_dev
*dev
= NULL
;
10396 struct imsm_map
*map_src
;
10397 int migr_vol_qan
= 0;
10398 int ndata
, odata
; /* [bytes] */
10399 int chunk
; /* [bytes] */
10400 struct migr_record
*migr_rec
;
10402 unsigned int buf_size
; /* [bytes] */
10403 unsigned long long max_position
; /* array size [bytes] */
10404 unsigned long long next_step
; /* [blocks]/[bytes] */
10405 unsigned long long old_data_stripe_length
;
10406 unsigned long long start_src
; /* [bytes] */
10407 unsigned long long start
; /* [bytes] */
10408 unsigned long long start_buf_shift
; /* [bytes] */
10410 int source_layout
= 0;
10412 if (!fds
|| !offsets
|| !sra
)
10415 /* Find volume during the reshape */
10416 for (dv
= super
->devlist
; dv
; dv
= dv
->next
) {
10417 if (dv
->dev
->vol
.migr_type
== MIGR_GEN_MIGR
10418 && dv
->dev
->vol
.migr_state
== 1) {
10423 /* Only one volume can migrate at the same time */
10424 if (migr_vol_qan
!= 1) {
10425 fprintf(stderr
, Name
" : %s", migr_vol_qan
?
10426 "Number of migrating volumes greater than 1\n" :
10427 "There is no volume during migrationg\n");
10431 map_src
= get_imsm_map(dev
, MAP_1
);
10432 if (map_src
== NULL
)
10435 ndata
= imsm_num_data_members(dev
, MAP_0
);
10436 odata
= imsm_num_data_members(dev
, MAP_1
);
10438 chunk
= __le16_to_cpu(map_src
->blocks_per_strip
) * 512;
10439 old_data_stripe_length
= odata
* chunk
;
10441 migr_rec
= super
->migr_rec
;
10443 /* initialize migration record for start condition */
10444 if (sra
->reshape_progress
== 0)
10445 init_migr_record_imsm(st
, dev
, sra
);
10447 if (__le32_to_cpu(migr_rec
->rec_status
) != UNIT_SRC_NORMAL
) {
10448 dprintf("imsm: cannot restart migration when data "
10449 "are present in copy area.\n");
10452 /* Save checkpoint to update migration record for current
10453 * reshape position (in md). It can be farther than current
10454 * reshape position in metadata.
10456 if (save_checkpoint_imsm(st
, sra
, UNIT_SRC_NORMAL
) == 1) {
10457 /* ignore error == 2, this can mean end of reshape here
10459 dprintf("imsm: Cannot write checkpoint to "
10460 "migration record (UNIT_SRC_NORMAL, "
10461 "initial save)\n");
10466 /* size for data */
10467 buf_size
= __le32_to_cpu(migr_rec
->blocks_per_unit
) * 512;
10468 /* extend buffer size for parity disk */
10469 buf_size
+= __le32_to_cpu(migr_rec
->dest_depth_per_unit
) * 512;
10470 /* add space for stripe aligment */
10471 buf_size
+= old_data_stripe_length
;
10472 if (posix_memalign((void **)&buf
, 4096, buf_size
)) {
10473 dprintf("imsm: Cannot allocate checpoint buffer\n");
10477 max_position
= sra
->component_size
* ndata
;
10478 source_layout
= imsm_level_to_layout(map_src
->raid_level
);
10480 while (__le32_to_cpu(migr_rec
->curr_migr_unit
) <
10481 __le32_to_cpu(migr_rec
->num_migr_units
)) {
10482 /* current reshape position [blocks] */
10483 unsigned long long current_position
=
10484 __le32_to_cpu(migr_rec
->blocks_per_unit
)
10485 * __le32_to_cpu(migr_rec
->curr_migr_unit
);
10486 unsigned long long border
;
10488 /* Check that array hasn't become failed.
10490 degraded
= check_degradation_change(sra
, fds
, degraded
);
10491 if (degraded
> 1) {
10492 dprintf("imsm: Abort reshape due to degradation"
10493 " level (%i)\n", degraded
);
10497 next_step
= __le32_to_cpu(migr_rec
->blocks_per_unit
);
10499 if ((current_position
+ next_step
) > max_position
)
10500 next_step
= max_position
- current_position
;
10502 start
= current_position
* 512;
10504 /* allign reading start to old geometry */
10505 start_buf_shift
= start
% old_data_stripe_length
;
10506 start_src
= start
- start_buf_shift
;
10508 border
= (start_src
/ odata
) - (start
/ ndata
);
10510 if (border
<= __le32_to_cpu(migr_rec
->dest_depth_per_unit
)) {
10511 /* save critical stripes to buf
10512 * start - start address of current unit
10513 * to backup [bytes]
10514 * start_src - start address of current unit
10515 * to backup alligned to source array
10518 unsigned long long next_step_filler
= 0;
10519 unsigned long long copy_length
= next_step
* 512;
10521 /* allign copy area length to stripe in old geometry */
10522 next_step_filler
= ((copy_length
+ start_buf_shift
)
10523 % old_data_stripe_length
);
10524 if (next_step_filler
)
10525 next_step_filler
= (old_data_stripe_length
10526 - next_step_filler
);
10527 dprintf("save_stripes() parameters: start = %llu,"
10528 "\tstart_src = %llu,\tnext_step*512 = %llu,"
10529 "\tstart_in_buf_shift = %llu,"
10530 "\tnext_step_filler = %llu\n",
10531 start
, start_src
, copy_length
,
10532 start_buf_shift
, next_step_filler
);
10534 if (save_stripes(fds
, offsets
, map_src
->num_members
,
10535 chunk
, map_src
->raid_level
,
10536 source_layout
, 0, NULL
, start_src
,
10538 next_step_filler
+ start_buf_shift
,
10540 dprintf("imsm: Cannot save stripes"
10544 /* Convert data to destination format and store it
10545 * in backup general migration area
10547 if (save_backup_imsm(st
, dev
, sra
,
10548 buf
+ start_buf_shift
, copy_length
)) {
10549 dprintf("imsm: Cannot save stripes to "
10550 "target devices\n");
10553 if (save_checkpoint_imsm(st
, sra
,
10554 UNIT_SRC_IN_CP_AREA
)) {
10555 dprintf("imsm: Cannot write checkpoint to "
10556 "migration record (UNIT_SRC_IN_CP_AREA)\n");
10560 /* set next step to use whole border area */
10561 border
/= next_step
;
10563 next_step
*= border
;
10565 /* When data backed up, checkpoint stored,
10566 * kick the kernel to reshape unit of data
10568 next_step
= next_step
+ sra
->reshape_progress
;
10569 /* limit next step to array max position */
10570 if (next_step
> max_position
)
10571 next_step
= max_position
;
10572 sysfs_set_num(sra
, NULL
, "suspend_lo", sra
->reshape_progress
);
10573 sysfs_set_num(sra
, NULL
, "suspend_hi", next_step
);
10574 sra
->reshape_progress
= next_step
;
10576 /* wait until reshape finish */
10577 if (wait_for_reshape_imsm(sra
, ndata
) < 0) {
10578 dprintf("wait_for_reshape_imsm returned error!\n");
10582 if (save_checkpoint_imsm(st
, sra
, UNIT_SRC_NORMAL
) == 1) {
10583 /* ignore error == 2, this can mean end of reshape here
10585 dprintf("imsm: Cannot write checkpoint to "
10586 "migration record (UNIT_SRC_NORMAL)\n");
10592 /* return '1' if done */
10596 abort_reshape(sra
);
10600 #endif /* MDASSEMBLE */
10602 struct superswitch super_imsm
= {
10604 .examine_super
= examine_super_imsm
,
10605 .brief_examine_super
= brief_examine_super_imsm
,
10606 .brief_examine_subarrays
= brief_examine_subarrays_imsm
,
10607 .export_examine_super
= export_examine_super_imsm
,
10608 .detail_super
= detail_super_imsm
,
10609 .brief_detail_super
= brief_detail_super_imsm
,
10610 .write_init_super
= write_init_super_imsm
,
10611 .validate_geometry
= validate_geometry_imsm
,
10612 .add_to_super
= add_to_super_imsm
,
10613 .remove_from_super
= remove_from_super_imsm
,
10614 .detail_platform
= detail_platform_imsm
,
10615 .kill_subarray
= kill_subarray_imsm
,
10616 .update_subarray
= update_subarray_imsm
,
10617 .load_container
= load_container_imsm
,
10618 .default_geometry
= default_geometry_imsm
,
10619 .get_disk_controller_domain
= imsm_get_disk_controller_domain
,
10620 .reshape_super
= imsm_reshape_super
,
10621 .manage_reshape
= imsm_manage_reshape
,
10622 .recover_backup
= recover_backup_imsm
,
10624 .match_home
= match_home_imsm
,
10625 .uuid_from_super
= uuid_from_super_imsm
,
10626 .getinfo_super
= getinfo_super_imsm
,
10627 .getinfo_super_disks
= getinfo_super_disks_imsm
,
10628 .update_super
= update_super_imsm
,
10630 .avail_size
= avail_size_imsm
,
10631 .min_acceptable_spare_size
= min_acceptable_spare_size_imsm
,
10633 .compare_super
= compare_super_imsm
,
10635 .load_super
= load_super_imsm
,
10636 .init_super
= init_super_imsm
,
10637 .store_super
= store_super_imsm
,
10638 .free_super
= free_super_imsm
,
10639 .match_metadata_desc
= match_metadata_desc_imsm
,
10640 .container_content
= container_content_imsm
,
10648 .open_new
= imsm_open_new
,
10649 .set_array_state
= imsm_set_array_state
,
10650 .set_disk
= imsm_set_disk
,
10651 .sync_metadata
= imsm_sync_metadata
,
10652 .activate_spare
= imsm_activate_spare
,
10653 .process_update
= imsm_process_update
,
10654 .prepare_update
= imsm_prepare_update
,
10655 #endif /* MDASSEMBLE */