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
);
922 static unsigned long long num_data_stripes(struct imsm_map
*map
)
926 return join_u32(map
->num_data_stripes_lo
, map
->num_data_stripes_hi
);
929 static void set_total_blocks(struct imsm_disk
*disk
, unsigned long long n
)
931 split_ull(n
, &disk
->total_blocks_lo
, &disk
->total_blocks_hi
);
935 static void set_pba_of_lba0(struct imsm_map
*map
, unsigned long long n
)
937 split_ull(n
, &map
->pba_of_lba0_lo
, &map
->pba_of_lba0_hi
);
940 static void set_blocks_per_member(struct imsm_map
*map
, unsigned long long n
)
942 split_ull(n
, &map
->blocks_per_member_lo
, &map
->blocks_per_member_hi
);
945 static void set_num_data_stripes(struct imsm_map
*map
, unsigned long long n
)
947 split_ull(n
, &map
->num_data_stripes_lo
, &map
->num_data_stripes_hi
);
950 static struct extent
*get_extents(struct intel_super
*super
, struct dl
*dl
)
952 /* find a list of used extents on the given physical device */
953 struct extent
*rv
, *e
;
955 int memberships
= count_memberships(dl
, super
);
958 /* trim the reserved area for spares, so they can join any array
959 * regardless of whether the OROM has assigned sectors from the
960 * IMSM_RESERVED_SECTORS region
963 reservation
= imsm_min_reserved_sectors(super
);
965 reservation
= MPB_SECTOR_CNT
+ IMSM_RESERVED_SECTORS
;
967 rv
= malloc(sizeof(struct extent
) * (memberships
+ 1));
972 for (i
= 0; i
< super
->anchor
->num_raid_devs
; i
++) {
973 struct imsm_dev
*dev
= get_imsm_dev(super
, i
);
974 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
976 if (get_imsm_disk_slot(map
, dl
->index
) >= 0) {
977 e
->start
= pba_of_lba0(map
);
978 e
->size
= blocks_per_member(map
);
982 qsort(rv
, memberships
, sizeof(*rv
), cmp_extent
);
984 /* determine the start of the metadata
985 * when no raid devices are defined use the default
986 * ...otherwise allow the metadata to truncate the value
987 * as is the case with older versions of imsm
990 struct extent
*last
= &rv
[memberships
- 1];
991 unsigned long long remainder
;
993 remainder
= total_blocks(&dl
->disk
) - (last
->start
+ last
->size
);
994 /* round down to 1k block to satisfy precision of the kernel
998 /* make sure remainder is still sane */
999 if (remainder
< (unsigned)ROUND_UP(super
->len
, 512) >> 9)
1000 remainder
= ROUND_UP(super
->len
, 512) >> 9;
1001 if (reservation
> remainder
)
1002 reservation
= remainder
;
1004 e
->start
= total_blocks(&dl
->disk
) - reservation
;
1009 /* try to determine how much space is reserved for metadata from
1010 * the last get_extents() entry, otherwise fallback to the
1013 static __u32
imsm_reserved_sectors(struct intel_super
*super
, struct dl
*dl
)
1019 /* for spares just return a minimal reservation which will grow
1020 * once the spare is picked up by an array
1022 if (dl
->index
== -1)
1023 return MPB_SECTOR_CNT
;
1025 e
= get_extents(super
, dl
);
1027 return MPB_SECTOR_CNT
+ IMSM_RESERVED_SECTORS
;
1029 /* scroll to last entry */
1030 for (i
= 0; e
[i
].size
; i
++)
1033 rv
= total_blocks(&dl
->disk
) - e
[i
].start
;
1040 static int is_spare(struct imsm_disk
*disk
)
1042 return (disk
->status
& SPARE_DISK
) == SPARE_DISK
;
1045 static int is_configured(struct imsm_disk
*disk
)
1047 return (disk
->status
& CONFIGURED_DISK
) == CONFIGURED_DISK
;
1050 static int is_failed(struct imsm_disk
*disk
)
1052 return (disk
->status
& FAILED_DISK
) == FAILED_DISK
;
1055 /* try to determine how much space is reserved for metadata from
1056 * the last get_extents() entry on the smallest active disk,
1057 * otherwise fallback to the default
1059 static __u32
imsm_min_reserved_sectors(struct intel_super
*super
)
1063 unsigned long long min_active
;
1065 __u32 rv
= MPB_SECTOR_CNT
+ IMSM_RESERVED_SECTORS
;
1066 struct dl
*dl
, *dl_min
= NULL
;
1072 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
1075 unsigned long long blocks
= total_blocks(&dl
->disk
);
1076 if (blocks
< min_active
|| min_active
== 0) {
1078 min_active
= blocks
;
1084 /* find last lba used by subarrays on the smallest active disk */
1085 e
= get_extents(super
, dl_min
);
1088 for (i
= 0; e
[i
].size
; i
++)
1091 remainder
= min_active
- e
[i
].start
;
1094 /* to give priority to recovery we should not require full
1095 IMSM_RESERVED_SECTORS from the spare */
1096 rv
= MPB_SECTOR_CNT
+ NUM_BLOCKS_DIRTY_STRIPE_REGION
;
1098 /* if real reservation is smaller use that value */
1099 return (remainder
< rv
) ? remainder
: rv
;
1102 /* Return minimum size of a spare that can be used in this array*/
1103 static unsigned long long min_acceptable_spare_size_imsm(struct supertype
*st
)
1105 struct intel_super
*super
= st
->sb
;
1109 unsigned long long rv
= 0;
1113 /* find first active disk in array */
1115 while (dl
&& (is_failed(&dl
->disk
) || dl
->index
== -1))
1119 /* find last lba used by subarrays */
1120 e
= get_extents(super
, dl
);
1123 for (i
= 0; e
[i
].size
; i
++)
1126 rv
= e
[i
-1].start
+ e
[i
-1].size
;
1129 /* add the amount of space needed for metadata */
1130 rv
= rv
+ imsm_min_reserved_sectors(super
);
1135 static int is_gen_migration(struct imsm_dev
*dev
);
1138 static __u64
blocks_per_migr_unit(struct intel_super
*super
,
1139 struct imsm_dev
*dev
);
1141 static void print_imsm_dev(struct intel_super
*super
,
1142 struct imsm_dev
*dev
,
1148 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
1149 struct imsm_map
*map2
= get_imsm_map(dev
, MAP_1
);
1153 printf("[%.16s]:\n", dev
->volume
);
1154 printf(" UUID : %s\n", uuid
);
1155 printf(" RAID Level : %d", get_imsm_raid_level(map
));
1157 printf(" <-- %d", get_imsm_raid_level(map2
));
1159 printf(" Members : %d", map
->num_members
);
1161 printf(" <-- %d", map2
->num_members
);
1163 printf(" Slots : [");
1164 for (i
= 0; i
< map
->num_members
; i
++) {
1165 ord
= get_imsm_ord_tbl_ent(dev
, i
, MAP_0
);
1166 printf("%s", ord
& IMSM_ORD_REBUILD
? "_" : "U");
1171 for (i
= 0; i
< map2
->num_members
; i
++) {
1172 ord
= get_imsm_ord_tbl_ent(dev
, i
, MAP_1
);
1173 printf("%s", ord
& IMSM_ORD_REBUILD
? "_" : "U");
1178 printf(" Failed disk : ");
1179 if (map
->failed_disk_num
== 0xff)
1182 printf("%i", map
->failed_disk_num
);
1184 slot
= get_imsm_disk_slot(map
, disk_idx
);
1186 ord
= get_imsm_ord_tbl_ent(dev
, slot
, MAP_X
);
1187 printf(" This Slot : %d%s\n", slot
,
1188 ord
& IMSM_ORD_REBUILD
? " (out-of-sync)" : "");
1190 printf(" This Slot : ?\n");
1191 sz
= __le32_to_cpu(dev
->size_high
);
1193 sz
+= __le32_to_cpu(dev
->size_low
);
1194 printf(" Array Size : %llu%s\n", (unsigned long long)sz
,
1195 human_size(sz
* 512));
1196 sz
= blocks_per_member(map
);
1197 printf(" Per Dev Size : %llu%s\n", (unsigned long long)sz
,
1198 human_size(sz
* 512));
1199 printf(" Sector Offset : %llu\n",
1201 printf(" Num Stripes : %llu\n",
1202 num_data_stripes(map
));
1203 printf(" Chunk Size : %u KiB",
1204 __le16_to_cpu(map
->blocks_per_strip
) / 2);
1206 printf(" <-- %u KiB",
1207 __le16_to_cpu(map2
->blocks_per_strip
) / 2);
1209 printf(" Reserved : %d\n", __le32_to_cpu(dev
->reserved_blocks
));
1210 printf(" Migrate State : ");
1211 if (dev
->vol
.migr_state
) {
1212 if (migr_type(dev
) == MIGR_INIT
)
1213 printf("initialize\n");
1214 else if (migr_type(dev
) == MIGR_REBUILD
)
1215 printf("rebuild\n");
1216 else if (migr_type(dev
) == MIGR_VERIFY
)
1218 else if (migr_type(dev
) == MIGR_GEN_MIGR
)
1219 printf("general migration\n");
1220 else if (migr_type(dev
) == MIGR_STATE_CHANGE
)
1221 printf("state change\n");
1222 else if (migr_type(dev
) == MIGR_REPAIR
)
1225 printf("<unknown:%d>\n", migr_type(dev
));
1228 printf(" Map State : %s", map_state_str
[map
->map_state
]);
1229 if (dev
->vol
.migr_state
) {
1230 struct imsm_map
*map
= get_imsm_map(dev
, MAP_1
);
1232 printf(" <-- %s", map_state_str
[map
->map_state
]);
1233 printf("\n Checkpoint : %u ",
1234 __le32_to_cpu(dev
->vol
.curr_migr_unit
));
1235 if ((is_gen_migration(dev
)) && ((slot
> 1) || (slot
< 0)))
1238 printf("(%llu)", (unsigned long long)
1239 blocks_per_migr_unit(super
, dev
));
1242 printf(" Dirty State : %s\n", dev
->vol
.dirty
? "dirty" : "clean");
1245 static void print_imsm_disk(struct imsm_disk
*disk
, int index
, __u32 reserved
)
1247 char str
[MAX_RAID_SERIAL_LEN
+ 1];
1250 if (index
< -1 || !disk
)
1254 snprintf(str
, MAX_RAID_SERIAL_LEN
+ 1, "%s", disk
->serial
);
1256 printf(" Disk%02d Serial : %s\n", index
, str
);
1258 printf(" Disk Serial : %s\n", str
);
1259 printf(" State :%s%s%s\n", is_spare(disk
) ? " spare" : "",
1260 is_configured(disk
) ? " active" : "",
1261 is_failed(disk
) ? " failed" : "");
1262 printf(" Id : %08x\n", __le32_to_cpu(disk
->scsi_id
));
1263 sz
= total_blocks(disk
) - reserved
;
1264 printf(" Usable Size : %llu%s\n", (unsigned long long)sz
,
1265 human_size(sz
* 512));
1268 void examine_migr_rec_imsm(struct intel_super
*super
)
1270 struct migr_record
*migr_rec
= super
->migr_rec
;
1271 struct imsm_super
*mpb
= super
->anchor
;
1274 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
1275 struct imsm_dev
*dev
= __get_imsm_dev(mpb
, i
);
1276 struct imsm_map
*map
;
1279 if (is_gen_migration(dev
) == 0)
1282 printf("\nMigration Record Information:");
1284 /* first map under migration */
1285 map
= get_imsm_map(dev
, MAP_0
);
1287 slot
= get_imsm_disk_slot(map
, super
->disks
->index
);
1288 if ((map
== NULL
) || (slot
> 1) || (slot
< 0)) {
1289 printf(" Empty\n ");
1290 printf("Examine one of first two disks in array\n");
1293 printf("\n Status : ");
1294 if (__le32_to_cpu(migr_rec
->rec_status
) == UNIT_SRC_NORMAL
)
1297 printf("Contains Data\n");
1298 printf(" Current Unit : %u\n",
1299 __le32_to_cpu(migr_rec
->curr_migr_unit
));
1300 printf(" Family : %u\n",
1301 __le32_to_cpu(migr_rec
->family_num
));
1302 printf(" Ascending : %u\n",
1303 __le32_to_cpu(migr_rec
->ascending_migr
));
1304 printf(" Blocks Per Unit : %u\n",
1305 __le32_to_cpu(migr_rec
->blocks_per_unit
));
1306 printf(" Dest. Depth Per Unit : %u\n",
1307 __le32_to_cpu(migr_rec
->dest_depth_per_unit
));
1308 printf(" Checkpoint Area pba : %u\n",
1309 __le32_to_cpu(migr_rec
->ckpt_area_pba
));
1310 printf(" First member lba : %u\n",
1311 __le32_to_cpu(migr_rec
->dest_1st_member_lba
));
1312 printf(" Total Number of Units : %u\n",
1313 __le32_to_cpu(migr_rec
->num_migr_units
));
1314 printf(" Size of volume : %u\n",
1315 __le32_to_cpu(migr_rec
->post_migr_vol_cap
));
1316 printf(" Expansion space for LBA64 : %u\n",
1317 __le32_to_cpu(migr_rec
->post_migr_vol_cap_hi
));
1318 printf(" Record was read from : %u\n",
1319 __le32_to_cpu(migr_rec
->ckpt_read_disk_num
));
1324 #endif /* MDASSEMBLE */
1325 /*******************************************************************************
1326 * function: imsm_check_attributes
1327 * Description: Function checks if features represented by attributes flags
1328 * are supported by mdadm.
1330 * attributes - Attributes read from metadata
1332 * 0 - passed attributes contains unsupported features flags
1333 * 1 - all features are supported
1334 ******************************************************************************/
1335 static int imsm_check_attributes(__u32 attributes
)
1338 __u32 not_supported
= MPB_ATTRIB_SUPPORTED
^0xffffffff;
1340 not_supported
&= ~MPB_ATTRIB_IGNORED
;
1342 not_supported
&= attributes
;
1343 if (not_supported
) {
1344 fprintf(stderr
, Name
"(IMSM): Unsupported attributes : %x\n",
1345 (unsigned)__le32_to_cpu(not_supported
));
1346 if (not_supported
& MPB_ATTRIB_CHECKSUM_VERIFY
) {
1347 dprintf("\t\tMPB_ATTRIB_CHECKSUM_VERIFY \n");
1348 not_supported
^= MPB_ATTRIB_CHECKSUM_VERIFY
;
1350 if (not_supported
& MPB_ATTRIB_2TB
) {
1351 dprintf("\t\tMPB_ATTRIB_2TB\n");
1352 not_supported
^= MPB_ATTRIB_2TB
;
1354 if (not_supported
& MPB_ATTRIB_RAID0
) {
1355 dprintf("\t\tMPB_ATTRIB_RAID0\n");
1356 not_supported
^= MPB_ATTRIB_RAID0
;
1358 if (not_supported
& MPB_ATTRIB_RAID1
) {
1359 dprintf("\t\tMPB_ATTRIB_RAID1\n");
1360 not_supported
^= MPB_ATTRIB_RAID1
;
1362 if (not_supported
& MPB_ATTRIB_RAID10
) {
1363 dprintf("\t\tMPB_ATTRIB_RAID10\n");
1364 not_supported
^= MPB_ATTRIB_RAID10
;
1366 if (not_supported
& MPB_ATTRIB_RAID1E
) {
1367 dprintf("\t\tMPB_ATTRIB_RAID1E\n");
1368 not_supported
^= MPB_ATTRIB_RAID1E
;
1370 if (not_supported
& MPB_ATTRIB_RAID5
) {
1371 dprintf("\t\tMPB_ATTRIB_RAID5\n");
1372 not_supported
^= MPB_ATTRIB_RAID5
;
1374 if (not_supported
& MPB_ATTRIB_RAIDCNG
) {
1375 dprintf("\t\tMPB_ATTRIB_RAIDCNG\n");
1376 not_supported
^= MPB_ATTRIB_RAIDCNG
;
1378 if (not_supported
& MPB_ATTRIB_BBM
) {
1379 dprintf("\t\tMPB_ATTRIB_BBM\n");
1380 not_supported
^= MPB_ATTRIB_BBM
;
1382 if (not_supported
& MPB_ATTRIB_CHECKSUM_VERIFY
) {
1383 dprintf("\t\tMPB_ATTRIB_CHECKSUM_VERIFY (== MPB_ATTRIB_LEGACY)\n");
1384 not_supported
^= MPB_ATTRIB_CHECKSUM_VERIFY
;
1386 if (not_supported
& MPB_ATTRIB_EXP_STRIPE_SIZE
) {
1387 dprintf("\t\tMPB_ATTRIB_EXP_STRIP_SIZE\n");
1388 not_supported
^= MPB_ATTRIB_EXP_STRIPE_SIZE
;
1390 if (not_supported
& MPB_ATTRIB_2TB_DISK
) {
1391 dprintf("\t\tMPB_ATTRIB_2TB_DISK\n");
1392 not_supported
^= MPB_ATTRIB_2TB_DISK
;
1394 if (not_supported
& MPB_ATTRIB_NEVER_USE2
) {
1395 dprintf("\t\tMPB_ATTRIB_NEVER_USE2\n");
1396 not_supported
^= MPB_ATTRIB_NEVER_USE2
;
1398 if (not_supported
& MPB_ATTRIB_NEVER_USE
) {
1399 dprintf("\t\tMPB_ATTRIB_NEVER_USE\n");
1400 not_supported
^= MPB_ATTRIB_NEVER_USE
;
1404 dprintf(Name
"(IMSM): Unknown attributes : %x\n", not_supported
);
1413 static void getinfo_super_imsm(struct supertype
*st
, struct mdinfo
*info
, char *map
);
1415 static void examine_super_imsm(struct supertype
*st
, char *homehost
)
1417 struct intel_super
*super
= st
->sb
;
1418 struct imsm_super
*mpb
= super
->anchor
;
1419 char str
[MAX_SIGNATURE_LENGTH
];
1424 __u32 reserved
= imsm_reserved_sectors(super
, super
->disks
);
1427 snprintf(str
, MPB_SIG_LEN
, "%s", mpb
->sig
);
1428 printf(" Magic : %s\n", str
);
1429 snprintf(str
, strlen(MPB_VERSION_RAID0
), "%s", get_imsm_version(mpb
));
1430 printf(" Version : %s\n", get_imsm_version(mpb
));
1431 printf(" Orig Family : %08x\n", __le32_to_cpu(mpb
->orig_family_num
));
1432 printf(" Family : %08x\n", __le32_to_cpu(mpb
->family_num
));
1433 printf(" Generation : %08x\n", __le32_to_cpu(mpb
->generation_num
));
1434 printf(" Attributes : ");
1435 if (imsm_check_attributes(mpb
->attributes
))
1436 printf("All supported\n");
1438 printf("not supported\n");
1439 getinfo_super_imsm(st
, &info
, NULL
);
1440 fname_from_uuid(st
, &info
, nbuf
, ':');
1441 printf(" UUID : %s\n", nbuf
+ 5);
1442 sum
= __le32_to_cpu(mpb
->check_sum
);
1443 printf(" Checksum : %08x %s\n", sum
,
1444 __gen_imsm_checksum(mpb
) == sum
? "correct" : "incorrect");
1445 printf(" MPB Sectors : %d\n", mpb_sectors(mpb
));
1446 printf(" Disks : %d\n", mpb
->num_disks
);
1447 printf(" RAID Devices : %d\n", mpb
->num_raid_devs
);
1448 print_imsm_disk(__get_imsm_disk(mpb
, super
->disks
->index
), super
->disks
->index
, reserved
);
1449 if (super
->bbm_log
) {
1450 struct bbm_log
*log
= super
->bbm_log
;
1453 printf("Bad Block Management Log:\n");
1454 printf(" Log Size : %d\n", __le32_to_cpu(mpb
->bbm_log_size
));
1455 printf(" Signature : %x\n", __le32_to_cpu(log
->signature
));
1456 printf(" Entry Count : %d\n", __le32_to_cpu(log
->entry_count
));
1457 printf(" Spare Blocks : %d\n", __le32_to_cpu(log
->reserved_spare_block_count
));
1458 printf(" First Spare : %llx\n",
1459 (unsigned long long) __le64_to_cpu(log
->first_spare_lba
));
1461 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
1463 struct imsm_dev
*dev
= __get_imsm_dev(mpb
, i
);
1465 super
->current_vol
= i
;
1466 getinfo_super_imsm(st
, &info
, NULL
);
1467 fname_from_uuid(st
, &info
, nbuf
, ':');
1468 print_imsm_dev(super
, dev
, nbuf
+ 5, super
->disks
->index
);
1470 for (i
= 0; i
< mpb
->num_disks
; i
++) {
1471 if (i
== super
->disks
->index
)
1473 print_imsm_disk(__get_imsm_disk(mpb
, i
), i
, reserved
);
1476 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
1477 if (dl
->index
== -1)
1478 print_imsm_disk(&dl
->disk
, -1, reserved
);
1480 examine_migr_rec_imsm(super
);
1483 static void brief_examine_super_imsm(struct supertype
*st
, int verbose
)
1485 /* We just write a generic IMSM ARRAY entry */
1488 struct intel_super
*super
= st
->sb
;
1490 if (!super
->anchor
->num_raid_devs
) {
1491 printf("ARRAY metadata=imsm\n");
1495 getinfo_super_imsm(st
, &info
, NULL
);
1496 fname_from_uuid(st
, &info
, nbuf
, ':');
1497 printf("ARRAY metadata=imsm UUID=%s\n", nbuf
+ 5);
1500 static void brief_examine_subarrays_imsm(struct supertype
*st
, int verbose
)
1502 /* We just write a generic IMSM ARRAY entry */
1506 struct intel_super
*super
= st
->sb
;
1509 if (!super
->anchor
->num_raid_devs
)
1512 getinfo_super_imsm(st
, &info
, NULL
);
1513 fname_from_uuid(st
, &info
, nbuf
, ':');
1514 for (i
= 0; i
< super
->anchor
->num_raid_devs
; i
++) {
1515 struct imsm_dev
*dev
= get_imsm_dev(super
, i
);
1517 super
->current_vol
= i
;
1518 getinfo_super_imsm(st
, &info
, NULL
);
1519 fname_from_uuid(st
, &info
, nbuf1
, ':');
1520 printf("ARRAY /dev/md/%.16s container=%s member=%d UUID=%s\n",
1521 dev
->volume
, nbuf
+ 5, i
, nbuf1
+ 5);
1525 static void export_examine_super_imsm(struct supertype
*st
)
1527 struct intel_super
*super
= st
->sb
;
1528 struct imsm_super
*mpb
= super
->anchor
;
1532 getinfo_super_imsm(st
, &info
, NULL
);
1533 fname_from_uuid(st
, &info
, nbuf
, ':');
1534 printf("MD_METADATA=imsm\n");
1535 printf("MD_LEVEL=container\n");
1536 printf("MD_UUID=%s\n", nbuf
+5);
1537 printf("MD_DEVICES=%u\n", mpb
->num_disks
);
1540 static void detail_super_imsm(struct supertype
*st
, char *homehost
)
1545 getinfo_super_imsm(st
, &info
, NULL
);
1546 fname_from_uuid(st
, &info
, nbuf
, ':');
1547 printf("\n UUID : %s\n", nbuf
+ 5);
1550 static void brief_detail_super_imsm(struct supertype
*st
)
1554 getinfo_super_imsm(st
, &info
, NULL
);
1555 fname_from_uuid(st
, &info
, nbuf
, ':');
1556 printf(" UUID=%s", nbuf
+ 5);
1559 static int imsm_read_serial(int fd
, char *devname
, __u8
*serial
);
1560 static void fd2devname(int fd
, char *name
);
1562 static int ahci_enumerate_ports(const char *hba_path
, int port_count
, int host_base
, int verbose
)
1564 /* dump an unsorted list of devices attached to AHCI Intel storage
1565 * controller, as well as non-connected ports
1567 int hba_len
= strlen(hba_path
) + 1;
1572 unsigned long port_mask
= (1 << port_count
) - 1;
1574 if (port_count
> (int)sizeof(port_mask
) * 8) {
1576 fprintf(stderr
, Name
": port_count %d out of range\n", port_count
);
1580 /* scroll through /sys/dev/block looking for devices attached to
1583 dir
= opendir("/sys/dev/block");
1584 for (ent
= dir
? readdir(dir
) : NULL
; ent
; ent
= readdir(dir
)) {
1595 if (sscanf(ent
->d_name
, "%d:%d", &major
, &minor
) != 2)
1597 path
= devt_to_devpath(makedev(major
, minor
));
1600 if (!path_attached_to_hba(path
, hba_path
)) {
1606 /* retrieve the scsi device type */
1607 if (asprintf(&device
, "/sys/dev/block/%d:%d/device/xxxxxxx", major
, minor
) < 0) {
1609 fprintf(stderr
, Name
": failed to allocate 'device'\n");
1613 sprintf(device
, "/sys/dev/block/%d:%d/device/type", major
, minor
);
1614 if (load_sys(device
, buf
) != 0) {
1616 fprintf(stderr
, Name
": failed to read device type for %s\n",
1622 type
= strtoul(buf
, NULL
, 10);
1624 /* if it's not a disk print the vendor and model */
1625 if (!(type
== 0 || type
== 7 || type
== 14)) {
1628 sprintf(device
, "/sys/dev/block/%d:%d/device/vendor", major
, minor
);
1629 if (load_sys(device
, buf
) == 0) {
1630 strncpy(vendor
, buf
, sizeof(vendor
));
1631 vendor
[sizeof(vendor
) - 1] = '\0';
1632 c
= (char *) &vendor
[sizeof(vendor
) - 1];
1633 while (isspace(*c
) || *c
== '\0')
1637 sprintf(device
, "/sys/dev/block/%d:%d/device/model", major
, minor
);
1638 if (load_sys(device
, buf
) == 0) {
1639 strncpy(model
, buf
, sizeof(model
));
1640 model
[sizeof(model
) - 1] = '\0';
1641 c
= (char *) &model
[sizeof(model
) - 1];
1642 while (isspace(*c
) || *c
== '\0')
1646 if (vendor
[0] && model
[0])
1647 sprintf(buf
, "%.64s %.64s", vendor
, model
);
1649 switch (type
) { /* numbers from hald/linux/device.c */
1650 case 1: sprintf(buf
, "tape"); break;
1651 case 2: sprintf(buf
, "printer"); break;
1652 case 3: sprintf(buf
, "processor"); break;
1654 case 5: sprintf(buf
, "cdrom"); break;
1655 case 6: sprintf(buf
, "scanner"); break;
1656 case 8: sprintf(buf
, "media_changer"); break;
1657 case 9: sprintf(buf
, "comm"); break;
1658 case 12: sprintf(buf
, "raid"); break;
1659 default: sprintf(buf
, "unknown");
1665 /* chop device path to 'host%d' and calculate the port number */
1666 c
= strchr(&path
[hba_len
], '/');
1669 fprintf(stderr
, Name
": %s - invalid path name\n", path
+ hba_len
);
1674 if (sscanf(&path
[hba_len
], "host%d", &port
) == 1)
1678 *c
= '/'; /* repair the full string */
1679 fprintf(stderr
, Name
": failed to determine port number for %s\n",
1686 /* mark this port as used */
1687 port_mask
&= ~(1 << port
);
1689 /* print out the device information */
1691 printf(" Port%d : - non-disk device (%s) -\n", port
, buf
);
1695 fd
= dev_open(ent
->d_name
, O_RDONLY
);
1697 printf(" Port%d : - disk info unavailable -\n", port
);
1699 fd2devname(fd
, buf
);
1700 printf(" Port%d : %s", port
, buf
);
1701 if (imsm_read_serial(fd
, NULL
, (__u8
*) buf
) == 0)
1702 printf(" (%.*s)\n", MAX_RAID_SERIAL_LEN
, buf
);
1717 for (i
= 0; i
< port_count
; i
++)
1718 if (port_mask
& (1 << i
))
1719 printf(" Port%d : - no device attached -\n", i
);
1725 static void print_found_intel_controllers(struct sys_dev
*elem
)
1727 for (; elem
; elem
= elem
->next
) {
1728 fprintf(stderr
, Name
": found Intel(R) ");
1729 if (elem
->type
== SYS_DEV_SATA
)
1730 fprintf(stderr
, "SATA ");
1731 else if (elem
->type
== SYS_DEV_SAS
)
1732 fprintf(stderr
, "SAS ");
1733 fprintf(stderr
, "RAID controller");
1735 fprintf(stderr
, " at %s", elem
->pci_id
);
1736 fprintf(stderr
, ".\n");
1741 static int ahci_get_port_count(const char *hba_path
, int *port_count
)
1748 if ((dir
= opendir(hba_path
)) == NULL
)
1751 for (ent
= readdir(dir
); ent
; ent
= readdir(dir
)) {
1754 if (sscanf(ent
->d_name
, "host%d", &host
) != 1)
1756 if (*port_count
== 0)
1758 else if (host
< host_base
)
1761 if (host
+ 1 > *port_count
+ host_base
)
1762 *port_count
= host
+ 1 - host_base
;
1768 static void print_imsm_capability(const struct imsm_orom
*orom
)
1770 printf(" Platform : Intel(R) Matrix Storage Manager\n");
1771 printf(" Version : %d.%d.%d.%d\n", orom
->major_ver
, orom
->minor_ver
,
1772 orom
->hotfix_ver
, orom
->build
);
1773 printf(" RAID Levels :%s%s%s%s%s\n",
1774 imsm_orom_has_raid0(orom
) ? " raid0" : "",
1775 imsm_orom_has_raid1(orom
) ? " raid1" : "",
1776 imsm_orom_has_raid1e(orom
) ? " raid1e" : "",
1777 imsm_orom_has_raid10(orom
) ? " raid10" : "",
1778 imsm_orom_has_raid5(orom
) ? " raid5" : "");
1779 printf(" Chunk Sizes :%s%s%s%s%s%s%s%s%s%s%s%s%s%s%s%s\n",
1780 imsm_orom_has_chunk(orom
, 2) ? " 2k" : "",
1781 imsm_orom_has_chunk(orom
, 4) ? " 4k" : "",
1782 imsm_orom_has_chunk(orom
, 8) ? " 8k" : "",
1783 imsm_orom_has_chunk(orom
, 16) ? " 16k" : "",
1784 imsm_orom_has_chunk(orom
, 32) ? " 32k" : "",
1785 imsm_orom_has_chunk(orom
, 64) ? " 64k" : "",
1786 imsm_orom_has_chunk(orom
, 128) ? " 128k" : "",
1787 imsm_orom_has_chunk(orom
, 256) ? " 256k" : "",
1788 imsm_orom_has_chunk(orom
, 512) ? " 512k" : "",
1789 imsm_orom_has_chunk(orom
, 1024*1) ? " 1M" : "",
1790 imsm_orom_has_chunk(orom
, 1024*2) ? " 2M" : "",
1791 imsm_orom_has_chunk(orom
, 1024*4) ? " 4M" : "",
1792 imsm_orom_has_chunk(orom
, 1024*8) ? " 8M" : "",
1793 imsm_orom_has_chunk(orom
, 1024*16) ? " 16M" : "",
1794 imsm_orom_has_chunk(orom
, 1024*32) ? " 32M" : "",
1795 imsm_orom_has_chunk(orom
, 1024*64) ? " 64M" : "");
1796 printf(" 2TB volumes :%s supported\n",
1797 (orom
->attr
& IMSM_OROM_ATTR_2TB
)?"":" not");
1798 printf(" 2TB disks :%s supported\n",
1799 (orom
->attr
& IMSM_OROM_ATTR_2TB_DISK
)?"":" not");
1800 printf(" Max Disks : %d\n", orom
->tds
);
1801 printf(" Max Volumes : %d per array, %d per controller\n",
1802 orom
->vpa
, orom
->vphba
);
1806 static int detail_platform_imsm(int verbose
, int enumerate_only
)
1808 /* There are two components to imsm platform support, the ahci SATA
1809 * controller and the option-rom. To find the SATA controller we
1810 * simply look in /sys/bus/pci/drivers/ahci to see if an ahci
1811 * controller with the Intel vendor id is present. This approach
1812 * allows mdadm to leverage the kernel's ahci detection logic, with the
1813 * caveat that if ahci.ko is not loaded mdadm will not be able to
1814 * detect platform raid capabilities. The option-rom resides in a
1815 * platform "Adapter ROM". We scan for its signature to retrieve the
1816 * platform capabilities. If raid support is disabled in the BIOS the
1817 * option-rom capability structure will not be available.
1819 const struct imsm_orom
*orom
;
1820 struct sys_dev
*list
, *hba
;
1825 if (enumerate_only
) {
1826 if (check_env("IMSM_NO_PLATFORM"))
1828 list
= find_intel_devices();
1831 for (hba
= list
; hba
; hba
= hba
->next
) {
1832 orom
= find_imsm_capability(hba
->type
);
1838 free_sys_dev(&list
);
1842 list
= find_intel_devices();
1845 fprintf(stderr
, Name
": no active Intel(R) RAID "
1846 "controller found.\n");
1847 free_sys_dev(&list
);
1850 print_found_intel_controllers(list
);
1852 for (hba
= list
; hba
; hba
= hba
->next
) {
1853 orom
= find_imsm_capability(hba
->type
);
1855 fprintf(stderr
, Name
": imsm capabilities not found for controller: %s (type %s)\n",
1856 hba
->path
, get_sys_dev_type(hba
->type
));
1858 print_imsm_capability(orom
);
1861 for (hba
= list
; hba
; hba
= hba
->next
) {
1862 printf(" I/O Controller : %s (%s)\n",
1863 hba
->path
, get_sys_dev_type(hba
->type
));
1865 if (hba
->type
== SYS_DEV_SATA
) {
1866 host_base
= ahci_get_port_count(hba
->path
, &port_count
);
1867 if (ahci_enumerate_ports(hba
->path
, port_count
, host_base
, verbose
)) {
1869 fprintf(stderr
, Name
": failed to enumerate "
1870 "ports on SATA controller at %s.", hba
->pci_id
);
1876 free_sys_dev(&list
);
1881 static int match_home_imsm(struct supertype
*st
, char *homehost
)
1883 /* the imsm metadata format does not specify any host
1884 * identification information. We return -1 since we can never
1885 * confirm nor deny whether a given array is "meant" for this
1886 * host. We rely on compare_super and the 'family_num' fields to
1887 * exclude member disks that do not belong, and we rely on
1888 * mdadm.conf to specify the arrays that should be assembled.
1889 * Auto-assembly may still pick up "foreign" arrays.
1895 static void uuid_from_super_imsm(struct supertype
*st
, int uuid
[4])
1897 /* The uuid returned here is used for:
1898 * uuid to put into bitmap file (Create, Grow)
1899 * uuid for backup header when saving critical section (Grow)
1900 * comparing uuids when re-adding a device into an array
1901 * In these cases the uuid required is that of the data-array,
1902 * not the device-set.
1903 * uuid to recognise same set when adding a missing device back
1904 * to an array. This is a uuid for the device-set.
1906 * For each of these we can make do with a truncated
1907 * or hashed uuid rather than the original, as long as
1909 * In each case the uuid required is that of the data-array,
1910 * not the device-set.
1912 /* imsm does not track uuid's so we synthesis one using sha1 on
1913 * - The signature (Which is constant for all imsm array, but no matter)
1914 * - the orig_family_num of the container
1915 * - the index number of the volume
1916 * - the 'serial' number of the volume.
1917 * Hopefully these are all constant.
1919 struct intel_super
*super
= st
->sb
;
1922 struct sha1_ctx ctx
;
1923 struct imsm_dev
*dev
= NULL
;
1926 /* some mdadm versions failed to set ->orig_family_num, in which
1927 * case fall back to ->family_num. orig_family_num will be
1928 * fixed up with the first metadata update.
1930 family_num
= super
->anchor
->orig_family_num
;
1931 if (family_num
== 0)
1932 family_num
= super
->anchor
->family_num
;
1933 sha1_init_ctx(&ctx
);
1934 sha1_process_bytes(super
->anchor
->sig
, MPB_SIG_LEN
, &ctx
);
1935 sha1_process_bytes(&family_num
, sizeof(__u32
), &ctx
);
1936 if (super
->current_vol
>= 0)
1937 dev
= get_imsm_dev(super
, super
->current_vol
);
1939 __u32 vol
= super
->current_vol
;
1940 sha1_process_bytes(&vol
, sizeof(vol
), &ctx
);
1941 sha1_process_bytes(dev
->volume
, MAX_RAID_SERIAL_LEN
, &ctx
);
1943 sha1_finish_ctx(&ctx
, buf
);
1944 memcpy(uuid
, buf
, 4*4);
1949 get_imsm_numerical_version(struct imsm_super
*mpb
, int *m
, int *p
)
1951 __u8
*v
= get_imsm_version(mpb
);
1952 __u8
*end
= mpb
->sig
+ MAX_SIGNATURE_LENGTH
;
1953 char major
[] = { 0, 0, 0 };
1954 char minor
[] = { 0 ,0, 0 };
1955 char patch
[] = { 0, 0, 0 };
1956 char *ver_parse
[] = { major
, minor
, patch
};
1960 while (*v
!= '\0' && v
< end
) {
1961 if (*v
!= '.' && j
< 2)
1962 ver_parse
[i
][j
++] = *v
;
1970 *m
= strtol(minor
, NULL
, 0);
1971 *p
= strtol(patch
, NULL
, 0);
1975 static __u32
migr_strip_blocks_resync(struct imsm_dev
*dev
)
1977 /* migr_strip_size when repairing or initializing parity */
1978 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
1979 __u32 chunk
= __le32_to_cpu(map
->blocks_per_strip
);
1981 switch (get_imsm_raid_level(map
)) {
1986 return 128*1024 >> 9;
1990 static __u32
migr_strip_blocks_rebuild(struct imsm_dev
*dev
)
1992 /* migr_strip_size when rebuilding a degraded disk, no idea why
1993 * this is different than migr_strip_size_resync(), but it's good
1996 struct imsm_map
*map
= get_imsm_map(dev
, MAP_1
);
1997 __u32 chunk
= __le32_to_cpu(map
->blocks_per_strip
);
1999 switch (get_imsm_raid_level(map
)) {
2002 if (map
->num_members
% map
->num_domains
== 0)
2003 return 128*1024 >> 9;
2007 return max((__u32
) 64*1024 >> 9, chunk
);
2009 return 128*1024 >> 9;
2013 static __u32
num_stripes_per_unit_resync(struct imsm_dev
*dev
)
2015 struct imsm_map
*lo
= get_imsm_map(dev
, MAP_0
);
2016 struct imsm_map
*hi
= get_imsm_map(dev
, MAP_1
);
2017 __u32 lo_chunk
= __le32_to_cpu(lo
->blocks_per_strip
);
2018 __u32 hi_chunk
= __le32_to_cpu(hi
->blocks_per_strip
);
2020 return max((__u32
) 1, hi_chunk
/ lo_chunk
);
2023 static __u32
num_stripes_per_unit_rebuild(struct imsm_dev
*dev
)
2025 struct imsm_map
*lo
= get_imsm_map(dev
, MAP_0
);
2026 int level
= get_imsm_raid_level(lo
);
2028 if (level
== 1 || level
== 10) {
2029 struct imsm_map
*hi
= get_imsm_map(dev
, MAP_1
);
2031 return hi
->num_domains
;
2033 return num_stripes_per_unit_resync(dev
);
2036 static __u8
imsm_num_data_members(struct imsm_dev
*dev
, int second_map
)
2038 /* named 'imsm_' because raid0, raid1 and raid10
2039 * counter-intuitively have the same number of data disks
2041 struct imsm_map
*map
= get_imsm_map(dev
, second_map
);
2043 switch (get_imsm_raid_level(map
)) {
2045 return map
->num_members
;
2049 return map
->num_members
/2;
2051 return map
->num_members
- 1;
2053 dprintf("%s: unsupported raid level\n", __func__
);
2058 static __u32
parity_segment_depth(struct imsm_dev
*dev
)
2060 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
2061 __u32 chunk
= __le32_to_cpu(map
->blocks_per_strip
);
2063 switch(get_imsm_raid_level(map
)) {
2066 return chunk
* map
->num_domains
;
2068 return chunk
* map
->num_members
;
2074 static __u32
map_migr_block(struct imsm_dev
*dev
, __u32 block
)
2076 struct imsm_map
*map
= get_imsm_map(dev
, MAP_1
);
2077 __u32 chunk
= __le32_to_cpu(map
->blocks_per_strip
);
2078 __u32 strip
= block
/ chunk
;
2080 switch (get_imsm_raid_level(map
)) {
2083 __u32 vol_strip
= (strip
* map
->num_domains
) + 1;
2084 __u32 vol_stripe
= vol_strip
/ map
->num_members
;
2086 return vol_stripe
* chunk
+ block
% chunk
;
2088 __u32 stripe
= strip
/ (map
->num_members
- 1);
2090 return stripe
* chunk
+ block
% chunk
;
2097 static __u64
blocks_per_migr_unit(struct intel_super
*super
,
2098 struct imsm_dev
*dev
)
2100 /* calculate the conversion factor between per member 'blocks'
2101 * (md/{resync,rebuild}_start) and imsm migration units, return
2102 * 0 for the 'not migrating' and 'unsupported migration' cases
2104 if (!dev
->vol
.migr_state
)
2107 switch (migr_type(dev
)) {
2108 case MIGR_GEN_MIGR
: {
2109 struct migr_record
*migr_rec
= super
->migr_rec
;
2110 return __le32_to_cpu(migr_rec
->blocks_per_unit
);
2115 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
2116 __u32 stripes_per_unit
;
2117 __u32 blocks_per_unit
;
2126 /* yes, this is really the translation of migr_units to
2127 * per-member blocks in the 'resync' case
2129 stripes_per_unit
= num_stripes_per_unit_resync(dev
);
2130 migr_chunk
= migr_strip_blocks_resync(dev
);
2131 disks
= imsm_num_data_members(dev
, MAP_0
);
2132 blocks_per_unit
= stripes_per_unit
* migr_chunk
* disks
;
2133 stripe
= __le16_to_cpu(map
->blocks_per_strip
) * disks
;
2134 segment
= blocks_per_unit
/ stripe
;
2135 block_rel
= blocks_per_unit
- segment
* stripe
;
2136 parity_depth
= parity_segment_depth(dev
);
2137 block_map
= map_migr_block(dev
, block_rel
);
2138 return block_map
+ parity_depth
* segment
;
2140 case MIGR_REBUILD
: {
2141 __u32 stripes_per_unit
;
2144 stripes_per_unit
= num_stripes_per_unit_rebuild(dev
);
2145 migr_chunk
= migr_strip_blocks_rebuild(dev
);
2146 return migr_chunk
* stripes_per_unit
;
2148 case MIGR_STATE_CHANGE
:
2154 static int imsm_level_to_layout(int level
)
2162 return ALGORITHM_LEFT_ASYMMETRIC
;
2169 /*******************************************************************************
2170 * Function: read_imsm_migr_rec
2171 * Description: Function reads imsm migration record from last sector of disk
2173 * fd : disk descriptor
2174 * super : metadata info
2178 ******************************************************************************/
2179 static int read_imsm_migr_rec(int fd
, struct intel_super
*super
)
2182 unsigned long long dsize
;
2184 get_dev_size(fd
, NULL
, &dsize
);
2185 if (lseek64(fd
, dsize
- MIGR_REC_POSITION
, SEEK_SET
) < 0) {
2187 Name
": Cannot seek to anchor block: %s\n",
2191 if (read(fd
, super
->migr_rec_buf
, MIGR_REC_BUF_SIZE
) !=
2192 MIGR_REC_BUF_SIZE
) {
2194 Name
": Cannot read migr record block: %s\n",
2204 static struct imsm_dev
*imsm_get_device_during_migration(
2205 struct intel_super
*super
)
2208 struct intel_dev
*dv
;
2210 for (dv
= super
->devlist
; dv
; dv
= dv
->next
) {
2211 if (is_gen_migration(dv
->dev
))
2217 /*******************************************************************************
2218 * Function: load_imsm_migr_rec
2219 * Description: Function reads imsm migration record (it is stored at the last
2222 * super : imsm internal array info
2223 * info : general array info
2227 * -2 : no migration in progress
2228 ******************************************************************************/
2229 static int load_imsm_migr_rec(struct intel_super
*super
, struct mdinfo
*info
)
2232 struct dl
*dl
= NULL
;
2236 struct imsm_dev
*dev
;
2237 struct imsm_map
*map
= NULL
;
2240 /* find map under migration */
2241 dev
= imsm_get_device_during_migration(super
);
2242 /* nothing to load,no migration in progress?
2246 map
= get_imsm_map(dev
, MAP_0
);
2249 for (sd
= info
->devs
; sd
; sd
= sd
->next
) {
2250 /* skip spare and failed disks
2252 if (sd
->disk
.raid_disk
< 0)
2254 /* read only from one of the first two slots */
2256 slot
= get_imsm_disk_slot(map
,
2257 sd
->disk
.raid_disk
);
2258 if ((map
== NULL
) || (slot
> 1) || (slot
< 0))
2261 sprintf(nm
, "%d:%d", sd
->disk
.major
, sd
->disk
.minor
);
2262 fd
= dev_open(nm
, O_RDONLY
);
2268 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
2269 /* skip spare and failed disks
2273 /* read only from one of the first two slots */
2275 slot
= get_imsm_disk_slot(map
, dl
->index
);
2276 if ((map
== NULL
) || (slot
> 1) || (slot
< 0))
2278 sprintf(nm
, "%d:%d", dl
->major
, dl
->minor
);
2279 fd
= dev_open(nm
, O_RDONLY
);
2286 retval
= read_imsm_migr_rec(fd
, super
);
2295 /*******************************************************************************
2296 * function: imsm_create_metadata_checkpoint_update
2297 * Description: It creates update for checkpoint change.
2299 * super : imsm internal array info
2300 * u : pointer to prepared update
2303 * If length is equal to 0, input pointer u contains no update
2304 ******************************************************************************/
2305 static int imsm_create_metadata_checkpoint_update(
2306 struct intel_super
*super
,
2307 struct imsm_update_general_migration_checkpoint
**u
)
2310 int update_memory_size
= 0;
2312 dprintf("imsm_create_metadata_checkpoint_update(enter)\n");
2318 /* size of all update data without anchor */
2319 update_memory_size
=
2320 sizeof(struct imsm_update_general_migration_checkpoint
);
2322 *u
= calloc(1, update_memory_size
);
2324 dprintf("error: cannot get memory for "
2325 "imsm_create_metadata_checkpoint_update update\n");
2328 (*u
)->type
= update_general_migration_checkpoint
;
2329 (*u
)->curr_migr_unit
= __le32_to_cpu(super
->migr_rec
->curr_migr_unit
);
2330 dprintf("imsm_create_metadata_checkpoint_update: prepared for %u\n",
2331 (*u
)->curr_migr_unit
);
2333 return update_memory_size
;
2337 static void imsm_update_metadata_locally(struct supertype
*st
,
2338 void *buf
, int len
);
2340 /*******************************************************************************
2341 * Function: write_imsm_migr_rec
2342 * Description: Function writes imsm migration record
2343 * (at the last sector of disk)
2345 * super : imsm internal array info
2349 ******************************************************************************/
2350 static int write_imsm_migr_rec(struct supertype
*st
)
2352 struct intel_super
*super
= st
->sb
;
2353 unsigned long long dsize
;
2359 struct imsm_update_general_migration_checkpoint
*u
;
2360 struct imsm_dev
*dev
;
2361 struct imsm_map
*map
= NULL
;
2363 /* find map under migration */
2364 dev
= imsm_get_device_during_migration(super
);
2365 /* if no migration, write buffer anyway to clear migr_record
2366 * on disk based on first available device
2369 dev
= get_imsm_dev(super
, super
->current_vol
< 0 ? 0 :
2370 super
->current_vol
);
2372 map
= get_imsm_map(dev
, MAP_0
);
2374 for (sd
= super
->disks
; sd
; sd
= sd
->next
) {
2377 /* skip failed and spare devices */
2380 /* write to 2 first slots only */
2382 slot
= get_imsm_disk_slot(map
, sd
->index
);
2383 if ((map
== NULL
) || (slot
> 1) || (slot
< 0))
2386 sprintf(nm
, "%d:%d", sd
->major
, sd
->minor
);
2387 fd
= dev_open(nm
, O_RDWR
);
2390 get_dev_size(fd
, NULL
, &dsize
);
2391 if (lseek64(fd
, dsize
- MIGR_REC_POSITION
, SEEK_SET
) < 0) {
2393 Name
": Cannot seek to anchor block: %s\n",
2397 if (write(fd
, super
->migr_rec_buf
, MIGR_REC_BUF_SIZE
) !=
2398 MIGR_REC_BUF_SIZE
) {
2400 Name
": Cannot write migr record block: %s\n",
2407 /* update checkpoint information in metadata */
2408 len
= imsm_create_metadata_checkpoint_update(super
, &u
);
2411 dprintf("imsm: Cannot prepare update\n");
2414 /* update metadata locally */
2415 imsm_update_metadata_locally(st
, u
, len
);
2416 /* and possibly remotely */
2417 if (st
->update_tail
) {
2418 append_metadata_update(st
, u
, len
);
2419 /* during reshape we do all work inside metadata handler
2420 * manage_reshape(), so metadata update has to be triggered
2423 flush_metadata_updates(st
);
2424 st
->update_tail
= &st
->updates
;
2434 #endif /* MDASSEMBLE */
2436 /* spare/missing disks activations are not allowe when
2437 * array/container performs reshape operation, because
2438 * all arrays in container works on the same disks set
2440 int imsm_reshape_blocks_arrays_changes(struct intel_super
*super
)
2443 struct intel_dev
*i_dev
;
2444 struct imsm_dev
*dev
;
2446 /* check whole container
2448 for (i_dev
= super
->devlist
; i_dev
; i_dev
= i_dev
->next
) {
2450 if (is_gen_migration(dev
)) {
2451 /* No repair during any migration in container
2459 static unsigned long long imsm_component_size_aligment_check(int level
,
2461 unsigned long long component_size
)
2463 unsigned int component_size_alligment
;
2465 /* check component size aligment
2467 component_size_alligment
= component_size
% (chunk_size
/512);
2469 dprintf("imsm_component_size_aligment_check(Level: %i, "
2470 "chunk_size = %i, component_size = %llu), "
2471 "component_size_alligment = %u\n",
2472 level
, chunk_size
, component_size
,
2473 component_size_alligment
);
2475 if (component_size_alligment
&& (level
!= 1) && (level
!= UnSet
)) {
2476 dprintf("imsm: reported component size alligned from %llu ",
2478 component_size
-= component_size_alligment
;
2479 dprintf("to %llu (%i).\n",
2480 component_size
, component_size_alligment
);
2483 return component_size
;
2486 static void getinfo_super_imsm_volume(struct supertype
*st
, struct mdinfo
*info
, char *dmap
)
2488 struct intel_super
*super
= st
->sb
;
2489 struct migr_record
*migr_rec
= super
->migr_rec
;
2490 struct imsm_dev
*dev
= get_imsm_dev(super
, super
->current_vol
);
2491 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
2492 struct imsm_map
*prev_map
= get_imsm_map(dev
, MAP_1
);
2493 struct imsm_map
*map_to_analyse
= map
;
2496 int map_disks
= info
->array
.raid_disks
;
2498 memset(info
, 0, sizeof(*info
));
2500 map_to_analyse
= prev_map
;
2502 dl
= super
->current_disk
;
2504 info
->container_member
= super
->current_vol
;
2505 info
->array
.raid_disks
= map
->num_members
;
2506 info
->array
.level
= get_imsm_raid_level(map_to_analyse
);
2507 info
->array
.layout
= imsm_level_to_layout(info
->array
.level
);
2508 info
->array
.md_minor
= -1;
2509 info
->array
.ctime
= 0;
2510 info
->array
.utime
= 0;
2511 info
->array
.chunk_size
=
2512 __le16_to_cpu(map_to_analyse
->blocks_per_strip
) << 9;
2513 info
->array
.state
= !dev
->vol
.dirty
;
2514 info
->custom_array_size
= __le32_to_cpu(dev
->size_high
);
2515 info
->custom_array_size
<<= 32;
2516 info
->custom_array_size
|= __le32_to_cpu(dev
->size_low
);
2517 info
->recovery_blocked
= imsm_reshape_blocks_arrays_changes(st
->sb
);
2519 if (is_gen_migration(dev
)) {
2520 info
->reshape_active
= 1;
2521 info
->new_level
= get_imsm_raid_level(map
);
2522 info
->new_layout
= imsm_level_to_layout(info
->new_level
);
2523 info
->new_chunk
= __le16_to_cpu(map
->blocks_per_strip
) << 9;
2524 info
->delta_disks
= map
->num_members
- prev_map
->num_members
;
2525 if (info
->delta_disks
) {
2526 /* this needs to be applied to every array
2529 info
->reshape_active
= CONTAINER_RESHAPE
;
2531 /* We shape information that we give to md might have to be
2532 * modify to cope with md's requirement for reshaping arrays.
2533 * For example, when reshaping a RAID0, md requires it to be
2534 * presented as a degraded RAID4.
2535 * Also if a RAID0 is migrating to a RAID5 we need to specify
2536 * the array as already being RAID5, but the 'before' layout
2537 * is a RAID4-like layout.
2539 switch (info
->array
.level
) {
2541 switch(info
->new_level
) {
2543 /* conversion is happening as RAID4 */
2544 info
->array
.level
= 4;
2545 info
->array
.raid_disks
+= 1;
2548 /* conversion is happening as RAID5 */
2549 info
->array
.level
= 5;
2550 info
->array
.layout
= ALGORITHM_PARITY_N
;
2551 info
->delta_disks
-= 1;
2554 /* FIXME error message */
2555 info
->array
.level
= UnSet
;
2561 info
->new_level
= UnSet
;
2562 info
->new_layout
= UnSet
;
2563 info
->new_chunk
= info
->array
.chunk_size
;
2564 info
->delta_disks
= 0;
2568 info
->disk
.major
= dl
->major
;
2569 info
->disk
.minor
= dl
->minor
;
2570 info
->disk
.number
= dl
->index
;
2571 info
->disk
.raid_disk
= get_imsm_disk_slot(map_to_analyse
,
2575 info
->data_offset
= pba_of_lba0(map_to_analyse
);
2576 info
->component_size
= blocks_per_member(map_to_analyse
);
2578 info
->component_size
= imsm_component_size_aligment_check(
2580 info
->array
.chunk_size
,
2581 info
->component_size
);
2583 memset(info
->uuid
, 0, sizeof(info
->uuid
));
2584 info
->recovery_start
= MaxSector
;
2586 info
->reshape_progress
= 0;
2587 info
->resync_start
= MaxSector
;
2588 if ((map_to_analyse
->map_state
== IMSM_T_STATE_UNINITIALIZED
||
2590 imsm_reshape_blocks_arrays_changes(super
) == 0) {
2591 info
->resync_start
= 0;
2593 if (dev
->vol
.migr_state
) {
2594 switch (migr_type(dev
)) {
2597 __u64 blocks_per_unit
= blocks_per_migr_unit(super
,
2599 __u64 units
= __le32_to_cpu(dev
->vol
.curr_migr_unit
);
2601 info
->resync_start
= blocks_per_unit
* units
;
2604 case MIGR_GEN_MIGR
: {
2605 __u64 blocks_per_unit
= blocks_per_migr_unit(super
,
2607 __u64 units
= __le32_to_cpu(migr_rec
->curr_migr_unit
);
2608 unsigned long long array_blocks
;
2611 if (__le32_to_cpu(migr_rec
->ascending_migr
) &&
2613 (__le32_to_cpu(migr_rec
->num_migr_units
)-1)) &&
2614 (super
->migr_rec
->rec_status
==
2615 __cpu_to_le32(UNIT_SRC_IN_CP_AREA
)))
2618 info
->reshape_progress
= blocks_per_unit
* units
;
2620 dprintf("IMSM: General Migration checkpoint : %llu "
2621 "(%llu) -> read reshape progress : %llu\n",
2622 (unsigned long long)units
,
2623 (unsigned long long)blocks_per_unit
,
2624 info
->reshape_progress
);
2626 used_disks
= imsm_num_data_members(dev
, MAP_1
);
2627 if (used_disks
> 0) {
2628 array_blocks
= blocks_per_member(map
) *
2630 /* round array size down to closest MB
2632 info
->custom_array_size
= (array_blocks
2633 >> SECT_PER_MB_SHIFT
)
2634 << SECT_PER_MB_SHIFT
;
2638 /* we could emulate the checkpointing of
2639 * 'sync_action=check' migrations, but for now
2640 * we just immediately complete them
2643 /* this is handled by container_content_imsm() */
2644 case MIGR_STATE_CHANGE
:
2645 /* FIXME handle other migrations */
2647 /* we are not dirty, so... */
2648 info
->resync_start
= MaxSector
;
2652 strncpy(info
->name
, (char *) dev
->volume
, MAX_RAID_SERIAL_LEN
);
2653 info
->name
[MAX_RAID_SERIAL_LEN
] = 0;
2655 info
->array
.major_version
= -1;
2656 info
->array
.minor_version
= -2;
2657 devname
= devnum2devname(st
->container_dev
);
2658 *info
->text_version
= '\0';
2660 sprintf(info
->text_version
, "/%s/%d", devname
, info
->container_member
);
2662 info
->safe_mode_delay
= 4000; /* 4 secs like the Matrix driver */
2663 uuid_from_super_imsm(st
, info
->uuid
);
2667 for (i
=0; i
<map_disks
; i
++) {
2669 if (i
< info
->array
.raid_disks
) {
2670 struct imsm_disk
*dsk
;
2671 j
= get_imsm_disk_idx(dev
, i
, MAP_X
);
2672 dsk
= get_imsm_disk(super
, j
);
2673 if (dsk
&& (dsk
->status
& CONFIGURED_DISK
))
2680 static __u8
imsm_check_degraded(struct intel_super
*super
, struct imsm_dev
*dev
,
2681 int failed
, int look_in_map
);
2683 static int imsm_count_failed(struct intel_super
*super
, struct imsm_dev
*dev
,
2688 static void manage_second_map(struct intel_super
*super
, struct imsm_dev
*dev
)
2690 if (is_gen_migration(dev
)) {
2693 struct imsm_map
*map2
= get_imsm_map(dev
, MAP_1
);
2695 failed
= imsm_count_failed(super
, dev
, MAP_1
);
2696 map_state
= imsm_check_degraded(super
, dev
, failed
, MAP_1
);
2697 if (map2
->map_state
!= map_state
) {
2698 map2
->map_state
= map_state
;
2699 super
->updates_pending
++;
2705 static struct imsm_disk
*get_imsm_missing(struct intel_super
*super
, __u8 index
)
2709 for (d
= super
->missing
; d
; d
= d
->next
)
2710 if (d
->index
== index
)
2715 static void getinfo_super_imsm(struct supertype
*st
, struct mdinfo
*info
, char *map
)
2717 struct intel_super
*super
= st
->sb
;
2718 struct imsm_disk
*disk
;
2719 int map_disks
= info
->array
.raid_disks
;
2720 int max_enough
= -1;
2722 struct imsm_super
*mpb
;
2724 if (super
->current_vol
>= 0) {
2725 getinfo_super_imsm_volume(st
, info
, map
);
2728 memset(info
, 0, sizeof(*info
));
2730 /* Set raid_disks to zero so that Assemble will always pull in valid
2733 info
->array
.raid_disks
= 0;
2734 info
->array
.level
= LEVEL_CONTAINER
;
2735 info
->array
.layout
= 0;
2736 info
->array
.md_minor
= -1;
2737 info
->array
.ctime
= 0; /* N/A for imsm */
2738 info
->array
.utime
= 0;
2739 info
->array
.chunk_size
= 0;
2741 info
->disk
.major
= 0;
2742 info
->disk
.minor
= 0;
2743 info
->disk
.raid_disk
= -1;
2744 info
->reshape_active
= 0;
2745 info
->array
.major_version
= -1;
2746 info
->array
.minor_version
= -2;
2747 strcpy(info
->text_version
, "imsm");
2748 info
->safe_mode_delay
= 0;
2749 info
->disk
.number
= -1;
2750 info
->disk
.state
= 0;
2752 info
->recovery_start
= MaxSector
;
2753 info
->recovery_blocked
= imsm_reshape_blocks_arrays_changes(st
->sb
);
2755 /* do we have the all the insync disks that we expect? */
2756 mpb
= super
->anchor
;
2758 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
2759 struct imsm_dev
*dev
= get_imsm_dev(super
, i
);
2760 int failed
, enough
, j
, missing
= 0;
2761 struct imsm_map
*map
;
2764 failed
= imsm_count_failed(super
, dev
, MAP_0
);
2765 state
= imsm_check_degraded(super
, dev
, failed
, MAP_0
);
2766 map
= get_imsm_map(dev
, MAP_0
);
2768 /* any newly missing disks?
2769 * (catches single-degraded vs double-degraded)
2771 for (j
= 0; j
< map
->num_members
; j
++) {
2772 __u32 ord
= get_imsm_ord_tbl_ent(dev
, j
, MAP_0
);
2773 __u32 idx
= ord_to_idx(ord
);
2775 if (!(ord
& IMSM_ORD_REBUILD
) &&
2776 get_imsm_missing(super
, idx
)) {
2782 if (state
== IMSM_T_STATE_FAILED
)
2784 else if (state
== IMSM_T_STATE_DEGRADED
&&
2785 (state
!= map
->map_state
|| missing
))
2787 else /* we're normal, or already degraded */
2789 if (is_gen_migration(dev
) && missing
) {
2790 /* during general migration we need all disks
2791 * that process is running on.
2792 * No new missing disk is allowed.
2796 /* no more checks necessary
2800 /* in the missing/failed disk case check to see
2801 * if at least one array is runnable
2803 max_enough
= max(max_enough
, enough
);
2805 dprintf("%s: enough: %d\n", __func__
, max_enough
);
2806 info
->container_enough
= max_enough
;
2809 __u32 reserved
= imsm_reserved_sectors(super
, super
->disks
);
2811 disk
= &super
->disks
->disk
;
2812 info
->data_offset
= total_blocks(&super
->disks
->disk
) - reserved
;
2813 info
->component_size
= reserved
;
2814 info
->disk
.state
= is_configured(disk
) ? (1 << MD_DISK_ACTIVE
) : 0;
2815 /* we don't change info->disk.raid_disk here because
2816 * this state will be finalized in mdmon after we have
2817 * found the 'most fresh' version of the metadata
2819 info
->disk
.state
|= is_failed(disk
) ? (1 << MD_DISK_FAULTY
) : 0;
2820 info
->disk
.state
|= is_spare(disk
) ? 0 : (1 << MD_DISK_SYNC
);
2823 /* only call uuid_from_super_imsm when this disk is part of a populated container,
2824 * ->compare_super may have updated the 'num_raid_devs' field for spares
2826 if (info
->disk
.state
& (1 << MD_DISK_SYNC
) || super
->anchor
->num_raid_devs
)
2827 uuid_from_super_imsm(st
, info
->uuid
);
2829 memcpy(info
->uuid
, uuid_zero
, sizeof(uuid_zero
));
2831 /* I don't know how to compute 'map' on imsm, so use safe default */
2834 for (i
= 0; i
< map_disks
; i
++)
2840 /* allocates memory and fills disk in mdinfo structure
2841 * for each disk in array */
2842 struct mdinfo
*getinfo_super_disks_imsm(struct supertype
*st
)
2844 struct mdinfo
*mddev
= NULL
;
2845 struct intel_super
*super
= st
->sb
;
2846 struct imsm_disk
*disk
;
2849 if (!super
|| !super
->disks
)
2852 mddev
= malloc(sizeof(*mddev
));
2854 fprintf(stderr
, Name
": Failed to allocate memory.\n");
2857 memset(mddev
, 0, sizeof(*mddev
));
2861 tmp
= malloc(sizeof(*tmp
));
2863 fprintf(stderr
, Name
": Failed to allocate memory.\n");
2868 memset(tmp
, 0, sizeof(*tmp
));
2870 tmp
->next
= mddev
->devs
;
2872 tmp
->disk
.number
= count
++;
2873 tmp
->disk
.major
= dl
->major
;
2874 tmp
->disk
.minor
= dl
->minor
;
2875 tmp
->disk
.state
= is_configured(disk
) ?
2876 (1 << MD_DISK_ACTIVE
) : 0;
2877 tmp
->disk
.state
|= is_failed(disk
) ? (1 << MD_DISK_FAULTY
) : 0;
2878 tmp
->disk
.state
|= is_spare(disk
) ? 0 : (1 << MD_DISK_SYNC
);
2879 tmp
->disk
.raid_disk
= -1;
2885 static int update_super_imsm(struct supertype
*st
, struct mdinfo
*info
,
2886 char *update
, char *devname
, int verbose
,
2887 int uuid_set
, char *homehost
)
2889 /* For 'assemble' and 'force' we need to return non-zero if any
2890 * change was made. For others, the return value is ignored.
2891 * Update options are:
2892 * force-one : This device looks a bit old but needs to be included,
2893 * update age info appropriately.
2894 * assemble: clear any 'faulty' flag to allow this device to
2896 * force-array: Array is degraded but being forced, mark it clean
2897 * if that will be needed to assemble it.
2899 * newdev: not used ????
2900 * grow: Array has gained a new device - this is currently for
2902 * resync: mark as dirty so a resync will happen.
2903 * name: update the name - preserving the homehost
2904 * uuid: Change the uuid of the array to match watch is given
2906 * Following are not relevant for this imsm:
2907 * sparc2.2 : update from old dodgey metadata
2908 * super-minor: change the preferred_minor number
2909 * summaries: update redundant counters.
2910 * homehost: update the recorded homehost
2911 * _reshape_progress: record new reshape_progress position.
2914 struct intel_super
*super
= st
->sb
;
2915 struct imsm_super
*mpb
;
2917 /* we can only update container info */
2918 if (!super
|| super
->current_vol
>= 0 || !super
->anchor
)
2921 mpb
= super
->anchor
;
2923 if (strcmp(update
, "uuid") == 0) {
2924 /* We take this to mean that the family_num should be updated.
2925 * However that is much smaller than the uuid so we cannot really
2926 * allow an explicit uuid to be given. And it is hard to reliably
2928 * So if !uuid_set we know the current uuid is random and just used
2929 * the first 'int' and copy it to the other 3 positions.
2930 * Otherwise we require the 4 'int's to be the same as would be the
2931 * case if we are using a random uuid. So an explicit uuid will be
2932 * accepted as long as all for ints are the same... which shouldn't hurt
2935 info
->uuid
[1] = info
->uuid
[2] = info
->uuid
[3] = info
->uuid
[0];
2938 if (info
->uuid
[0] != info
->uuid
[1] ||
2939 info
->uuid
[1] != info
->uuid
[2] ||
2940 info
->uuid
[2] != info
->uuid
[3])
2946 mpb
->orig_family_num
= info
->uuid
[0];
2947 } else if (strcmp(update
, "assemble") == 0)
2952 /* successful update? recompute checksum */
2954 mpb
->check_sum
= __le32_to_cpu(__gen_imsm_checksum(mpb
));
2959 static size_t disks_to_mpb_size(int disks
)
2963 size
= sizeof(struct imsm_super
);
2964 size
+= (disks
- 1) * sizeof(struct imsm_disk
);
2965 size
+= 2 * sizeof(struct imsm_dev
);
2966 /* up to 2 maps per raid device (-2 for imsm_maps in imsm_dev */
2967 size
+= (4 - 2) * sizeof(struct imsm_map
);
2968 /* 4 possible disk_ord_tbl's */
2969 size
+= 4 * (disks
- 1) * sizeof(__u32
);
2974 static __u64
avail_size_imsm(struct supertype
*st
, __u64 devsize
)
2976 if (devsize
< (MPB_SECTOR_CNT
+ IMSM_RESERVED_SECTORS
))
2979 return devsize
- (MPB_SECTOR_CNT
+ IMSM_RESERVED_SECTORS
);
2982 static void free_devlist(struct intel_super
*super
)
2984 struct intel_dev
*dv
;
2986 while (super
->devlist
) {
2987 dv
= super
->devlist
->next
;
2988 free(super
->devlist
->dev
);
2989 free(super
->devlist
);
2990 super
->devlist
= dv
;
2994 static void imsm_copy_dev(struct imsm_dev
*dest
, struct imsm_dev
*src
)
2996 memcpy(dest
, src
, sizeof_imsm_dev(src
, 0));
2999 static int compare_super_imsm(struct supertype
*st
, struct supertype
*tst
)
3003 * 0 same, or first was empty, and second was copied
3004 * 1 second had wrong number
3006 * 3 wrong other info
3008 struct intel_super
*first
= st
->sb
;
3009 struct intel_super
*sec
= tst
->sb
;
3016 /* in platform dependent environment test if the disks
3017 * use the same Intel hba
3019 if (!check_env("IMSM_NO_PLATFORM")) {
3020 if (!first
->hba
|| !sec
->hba
||
3021 (first
->hba
->type
!= sec
->hba
->type
)) {
3023 "HBAs of devices does not match %s != %s\n",
3024 first
->hba
? get_sys_dev_type(first
->hba
->type
) : NULL
,
3025 sec
->hba
? get_sys_dev_type(sec
->hba
->type
) : NULL
);
3030 /* if an anchor does not have num_raid_devs set then it is a free
3033 if (first
->anchor
->num_raid_devs
> 0 &&
3034 sec
->anchor
->num_raid_devs
> 0) {
3035 /* Determine if these disks might ever have been
3036 * related. Further disambiguation can only take place
3037 * in load_super_imsm_all
3039 __u32 first_family
= first
->anchor
->orig_family_num
;
3040 __u32 sec_family
= sec
->anchor
->orig_family_num
;
3042 if (memcmp(first
->anchor
->sig
, sec
->anchor
->sig
,
3043 MAX_SIGNATURE_LENGTH
) != 0)
3046 if (first_family
== 0)
3047 first_family
= first
->anchor
->family_num
;
3048 if (sec_family
== 0)
3049 sec_family
= sec
->anchor
->family_num
;
3051 if (first_family
!= sec_family
)
3057 /* if 'first' is a spare promote it to a populated mpb with sec's
3060 if (first
->anchor
->num_raid_devs
== 0 &&
3061 sec
->anchor
->num_raid_devs
> 0) {
3063 struct intel_dev
*dv
;
3064 struct imsm_dev
*dev
;
3066 /* we need to copy raid device info from sec if an allocation
3067 * fails here we don't associate the spare
3069 for (i
= 0; i
< sec
->anchor
->num_raid_devs
; i
++) {
3070 dv
= malloc(sizeof(*dv
));
3073 dev
= malloc(sizeof_imsm_dev(get_imsm_dev(sec
, i
), 1));
3080 dv
->next
= first
->devlist
;
3081 first
->devlist
= dv
;
3083 if (i
< sec
->anchor
->num_raid_devs
) {
3084 /* allocation failure */
3085 free_devlist(first
);
3086 fprintf(stderr
, "imsm: failed to associate spare\n");
3089 first
->anchor
->num_raid_devs
= sec
->anchor
->num_raid_devs
;
3090 first
->anchor
->orig_family_num
= sec
->anchor
->orig_family_num
;
3091 first
->anchor
->family_num
= sec
->anchor
->family_num
;
3092 memcpy(first
->anchor
->sig
, sec
->anchor
->sig
, MAX_SIGNATURE_LENGTH
);
3093 for (i
= 0; i
< sec
->anchor
->num_raid_devs
; i
++)
3094 imsm_copy_dev(get_imsm_dev(first
, i
), get_imsm_dev(sec
, i
));
3100 static void fd2devname(int fd
, char *name
)
3104 char dname
[PATH_MAX
];
3109 if (fstat(fd
, &st
) != 0)
3111 sprintf(path
, "/sys/dev/block/%d:%d",
3112 major(st
.st_rdev
), minor(st
.st_rdev
));
3114 rv
= readlink(path
, dname
, sizeof(dname
)-1);
3119 nm
= strrchr(dname
, '/');
3122 snprintf(name
, MAX_RAID_SERIAL_LEN
, "/dev/%s", nm
);
3126 extern int scsi_get_serial(int fd
, void *buf
, size_t buf_len
);
3128 static int imsm_read_serial(int fd
, char *devname
,
3129 __u8 serial
[MAX_RAID_SERIAL_LEN
])
3131 unsigned char scsi_serial
[255];
3140 memset(scsi_serial
, 0, sizeof(scsi_serial
));
3142 rv
= scsi_get_serial(fd
, scsi_serial
, sizeof(scsi_serial
));
3144 if (rv
&& check_env("IMSM_DEVNAME_AS_SERIAL")) {
3145 memset(serial
, 0, MAX_RAID_SERIAL_LEN
);
3146 fd2devname(fd
, (char *) serial
);
3153 Name
": Failed to retrieve serial for %s\n",
3158 rsp_len
= scsi_serial
[3];
3162 Name
": Failed to retrieve serial for %s\n",
3166 rsp_buf
= (char *) &scsi_serial
[4];
3168 /* trim all whitespace and non-printable characters and convert
3171 for (i
= 0, dest
= rsp_buf
; i
< rsp_len
; i
++) {
3174 /* ':' is reserved for use in placeholder serial
3175 * numbers for missing disks
3183 len
= dest
- rsp_buf
;
3186 /* truncate leading characters */
3187 if (len
> MAX_RAID_SERIAL_LEN
) {
3188 dest
+= len
- MAX_RAID_SERIAL_LEN
;
3189 len
= MAX_RAID_SERIAL_LEN
;
3192 memset(serial
, 0, MAX_RAID_SERIAL_LEN
);
3193 memcpy(serial
, dest
, len
);
3198 static int serialcmp(__u8
*s1
, __u8
*s2
)
3200 return strncmp((char *) s1
, (char *) s2
, MAX_RAID_SERIAL_LEN
);
3203 static void serialcpy(__u8
*dest
, __u8
*src
)
3205 strncpy((char *) dest
, (char *) src
, MAX_RAID_SERIAL_LEN
);
3208 static struct dl
*serial_to_dl(__u8
*serial
, struct intel_super
*super
)
3212 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
3213 if (serialcmp(dl
->serial
, serial
) == 0)
3219 static struct imsm_disk
*
3220 __serial_to_disk(__u8
*serial
, struct imsm_super
*mpb
, int *idx
)
3224 for (i
= 0; i
< mpb
->num_disks
; i
++) {
3225 struct imsm_disk
*disk
= __get_imsm_disk(mpb
, i
);
3227 if (serialcmp(disk
->serial
, serial
) == 0) {
3238 load_imsm_disk(int fd
, struct intel_super
*super
, char *devname
, int keep_fd
)
3240 struct imsm_disk
*disk
;
3245 __u8 serial
[MAX_RAID_SERIAL_LEN
];
3247 rv
= imsm_read_serial(fd
, devname
, serial
);
3252 dl
= calloc(1, sizeof(*dl
));
3256 Name
": failed to allocate disk buffer for %s\n",
3262 dl
->major
= major(stb
.st_rdev
);
3263 dl
->minor
= minor(stb
.st_rdev
);
3264 dl
->next
= super
->disks
;
3265 dl
->fd
= keep_fd
? fd
: -1;
3266 assert(super
->disks
== NULL
);
3268 serialcpy(dl
->serial
, serial
);
3271 fd2devname(fd
, name
);
3273 dl
->devname
= strdup(devname
);
3275 dl
->devname
= strdup(name
);
3277 /* look up this disk's index in the current anchor */
3278 disk
= __serial_to_disk(dl
->serial
, super
->anchor
, &dl
->index
);
3281 /* only set index on disks that are a member of a
3282 * populated contianer, i.e. one with raid_devs
3284 if (is_failed(&dl
->disk
))
3286 else if (is_spare(&dl
->disk
))
3294 /* When migrating map0 contains the 'destination' state while map1
3295 * contains the current state. When not migrating map0 contains the
3296 * current state. This routine assumes that map[0].map_state is set to
3297 * the current array state before being called.
3299 * Migration is indicated by one of the following states
3300 * 1/ Idle (migr_state=0 map0state=normal||unitialized||degraded||failed)
3301 * 2/ Initialize (migr_state=1 migr_type=MIGR_INIT map0state=normal
3302 * map1state=unitialized)
3303 * 3/ Repair (Resync) (migr_state=1 migr_type=MIGR_REPAIR map0state=normal
3305 * 4/ Rebuild (migr_state=1 migr_type=MIGR_REBUILD map0state=normal
3306 * map1state=degraded)
3307 * 5/ Migration (mig_state=1 migr_type=MIGR_GEN_MIGR map0state=normal
3310 static void migrate(struct imsm_dev
*dev
, struct intel_super
*super
,
3311 __u8 to_state
, int migr_type
)
3313 struct imsm_map
*dest
;
3314 struct imsm_map
*src
= get_imsm_map(dev
, MAP_0
);
3316 dev
->vol
.migr_state
= 1;
3317 set_migr_type(dev
, migr_type
);
3318 dev
->vol
.curr_migr_unit
= 0;
3319 dest
= get_imsm_map(dev
, MAP_1
);
3321 /* duplicate and then set the target end state in map[0] */
3322 memcpy(dest
, src
, sizeof_imsm_map(src
));
3323 if ((migr_type
== MIGR_REBUILD
) ||
3324 (migr_type
== MIGR_GEN_MIGR
)) {
3328 for (i
= 0; i
< src
->num_members
; i
++) {
3329 ord
= __le32_to_cpu(src
->disk_ord_tbl
[i
]);
3330 set_imsm_ord_tbl_ent(src
, i
, ord_to_idx(ord
));
3334 if (migr_type
== MIGR_GEN_MIGR
)
3335 /* Clear migration record */
3336 memset(super
->migr_rec
, 0, sizeof(struct migr_record
));
3338 src
->map_state
= to_state
;
3341 static void end_migration(struct imsm_dev
*dev
, struct intel_super
*super
,
3344 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
3345 struct imsm_map
*prev
= get_imsm_map(dev
, dev
->vol
.migr_state
== 0 ?
3349 /* merge any IMSM_ORD_REBUILD bits that were not successfully
3350 * completed in the last migration.
3352 * FIXME add support for raid-level-migration
3354 if ((map_state
!= map
->map_state
) && (is_gen_migration(dev
) == 0) &&
3355 (prev
->map_state
!= IMSM_T_STATE_UNINITIALIZED
)) {
3356 /* when final map state is other than expected
3357 * merge maps (not for migration)
3361 for (i
= 0; i
< prev
->num_members
; i
++)
3362 for (j
= 0; j
< map
->num_members
; j
++)
3363 /* during online capacity expansion
3364 * disks position can be changed
3365 * if takeover is used
3367 if (ord_to_idx(map
->disk_ord_tbl
[j
]) ==
3368 ord_to_idx(prev
->disk_ord_tbl
[i
])) {
3369 map
->disk_ord_tbl
[j
] |=
3370 prev
->disk_ord_tbl
[i
];
3373 failed
= imsm_count_failed(super
, dev
, MAP_0
);
3374 map_state
= imsm_check_degraded(super
, dev
, failed
, MAP_0
);
3377 dev
->vol
.migr_state
= 0;
3378 set_migr_type(dev
, 0);
3379 dev
->vol
.curr_migr_unit
= 0;
3380 map
->map_state
= map_state
;
3384 static int parse_raid_devices(struct intel_super
*super
)
3387 struct imsm_dev
*dev_new
;
3388 size_t len
, len_migr
;
3390 size_t space_needed
= 0;
3391 struct imsm_super
*mpb
= super
->anchor
;
3393 for (i
= 0; i
< super
->anchor
->num_raid_devs
; i
++) {
3394 struct imsm_dev
*dev_iter
= __get_imsm_dev(super
->anchor
, i
);
3395 struct intel_dev
*dv
;
3397 len
= sizeof_imsm_dev(dev_iter
, 0);
3398 len_migr
= sizeof_imsm_dev(dev_iter
, 1);
3400 space_needed
+= len_migr
- len
;
3402 dv
= malloc(sizeof(*dv
));
3405 if (max_len
< len_migr
)
3407 if (max_len
> len_migr
)
3408 space_needed
+= max_len
- len_migr
;
3409 dev_new
= malloc(max_len
);
3414 imsm_copy_dev(dev_new
, dev_iter
);
3417 dv
->next
= super
->devlist
;
3418 super
->devlist
= dv
;
3421 /* ensure that super->buf is large enough when all raid devices
3424 if (__le32_to_cpu(mpb
->mpb_size
) + space_needed
> super
->len
) {
3427 len
= ROUND_UP(__le32_to_cpu(mpb
->mpb_size
) + space_needed
, 512);
3428 if (posix_memalign(&buf
, 512, len
) != 0)
3431 memcpy(buf
, super
->buf
, super
->len
);
3432 memset(buf
+ super
->len
, 0, len
- super
->len
);
3441 /* retrieve a pointer to the bbm log which starts after all raid devices */
3442 struct bbm_log
*__get_imsm_bbm_log(struct imsm_super
*mpb
)
3446 if (__le32_to_cpu(mpb
->bbm_log_size
)) {
3448 ptr
+= mpb
->mpb_size
- __le32_to_cpu(mpb
->bbm_log_size
);
3454 /*******************************************************************************
3455 * Function: check_mpb_migr_compatibility
3456 * Description: Function checks for unsupported migration features:
3457 * - migration optimization area (pba_of_lba0)
3458 * - descending reshape (ascending_migr)
3460 * super : imsm metadata information
3462 * 0 : migration is compatible
3463 * -1 : migration is not compatible
3464 ******************************************************************************/
3465 int check_mpb_migr_compatibility(struct intel_super
*super
)
3467 struct imsm_map
*map0
, *map1
;
3468 struct migr_record
*migr_rec
= super
->migr_rec
;
3471 for (i
= 0; i
< super
->anchor
->num_raid_devs
; i
++) {
3472 struct imsm_dev
*dev_iter
= __get_imsm_dev(super
->anchor
, i
);
3475 dev_iter
->vol
.migr_state
== 1 &&
3476 dev_iter
->vol
.migr_type
== MIGR_GEN_MIGR
) {
3477 /* This device is migrating */
3478 map0
= get_imsm_map(dev_iter
, MAP_0
);
3479 map1
= get_imsm_map(dev_iter
, MAP_1
);
3480 if (pba_of_lba0(map0
) != pba_of_lba0(map1
))
3481 /* migration optimization area was used */
3483 if (migr_rec
->ascending_migr
== 0
3484 && migr_rec
->dest_depth_per_unit
> 0)
3485 /* descending reshape not supported yet */
3492 static void __free_imsm(struct intel_super
*super
, int free_disks
);
3494 /* load_imsm_mpb - read matrix metadata
3495 * allocates super->mpb to be freed by free_imsm
3497 static int load_imsm_mpb(int fd
, struct intel_super
*super
, char *devname
)
3499 unsigned long long dsize
;
3500 unsigned long long sectors
;
3502 struct imsm_super
*anchor
;
3505 get_dev_size(fd
, NULL
, &dsize
);
3509 Name
": %s: device to small for imsm\n",
3514 if (lseek64(fd
, dsize
- (512 * 2), SEEK_SET
) < 0) {
3516 fprintf(stderr
, Name
3517 ": Cannot seek to anchor block on %s: %s\n",
3518 devname
, strerror(errno
));
3522 if (posix_memalign((void**)&anchor
, 512, 512) != 0) {
3525 Name
": Failed to allocate imsm anchor buffer"
3526 " on %s\n", devname
);
3529 if (read(fd
, anchor
, 512) != 512) {
3532 Name
": Cannot read anchor block on %s: %s\n",
3533 devname
, strerror(errno
));
3538 if (strncmp((char *) anchor
->sig
, MPB_SIGNATURE
, MPB_SIG_LEN
) != 0) {
3541 Name
": no IMSM anchor on %s\n", devname
);
3546 __free_imsm(super
, 0);
3547 /* reload capability and hba */
3549 /* capability and hba must be updated with new super allocation */
3550 find_intel_hba_capability(fd
, super
, devname
);
3551 super
->len
= ROUND_UP(anchor
->mpb_size
, 512);
3552 if (posix_memalign(&super
->buf
, 512, super
->len
) != 0) {
3555 Name
": unable to allocate %zu byte mpb buffer\n",
3560 memcpy(super
->buf
, anchor
, 512);
3562 sectors
= mpb_sectors(anchor
) - 1;
3565 if (posix_memalign(&super
->migr_rec_buf
, 512, MIGR_REC_BUF_SIZE
) != 0) {
3566 fprintf(stderr
, Name
3567 ": %s could not allocate migr_rec buffer\n", __func__
);
3571 super
->clean_migration_record_by_mdmon
= 0;
3574 check_sum
= __gen_imsm_checksum(super
->anchor
);
3575 if (check_sum
!= __le32_to_cpu(super
->anchor
->check_sum
)) {
3578 Name
": IMSM checksum %x != %x on %s\n",
3580 __le32_to_cpu(super
->anchor
->check_sum
),
3588 /* read the extended mpb */
3589 if (lseek64(fd
, dsize
- (512 * (2 + sectors
)), SEEK_SET
) < 0) {
3592 Name
": Cannot seek to extended mpb on %s: %s\n",
3593 devname
, strerror(errno
));
3597 if ((unsigned)read(fd
, super
->buf
+ 512, super
->len
- 512) != super
->len
- 512) {
3600 Name
": Cannot read extended mpb on %s: %s\n",
3601 devname
, strerror(errno
));
3605 check_sum
= __gen_imsm_checksum(super
->anchor
);
3606 if (check_sum
!= __le32_to_cpu(super
->anchor
->check_sum
)) {
3609 Name
": IMSM checksum %x != %x on %s\n",
3610 check_sum
, __le32_to_cpu(super
->anchor
->check_sum
),
3615 /* FIXME the BBM log is disk specific so we cannot use this global
3616 * buffer for all disks. Ok for now since we only look at the global
3617 * bbm_log_size parameter to gate assembly
3619 super
->bbm_log
= __get_imsm_bbm_log(super
->anchor
);
3624 static int read_imsm_migr_rec(int fd
, struct intel_super
*super
);
3626 /* clears hi bits in metadata if MPB_ATTRIB_2TB_DISK not set */
3627 static void clear_hi(struct intel_super
*super
)
3629 struct imsm_super
*mpb
= super
->anchor
;
3631 if (mpb
->attributes
& MPB_ATTRIB_2TB_DISK
)
3633 for (i
= 0; i
< mpb
->num_disks
; ++i
) {
3634 struct imsm_disk
*disk
= &mpb
->disk
[i
];
3635 disk
->total_blocks_hi
= 0;
3637 for (i
= 0; i
< mpb
->num_raid_devs
; ++i
) {
3638 struct imsm_dev
*dev
= get_imsm_dev(super
, i
);
3641 for (n
= 0; n
< 2; ++n
) {
3642 struct imsm_map
*map
= get_imsm_map(dev
, n
);
3645 map
->pba_of_lba0_hi
= 0;
3646 map
->blocks_per_member_hi
= 0;
3647 map
->num_data_stripes_hi
= 0;
3653 load_and_parse_mpb(int fd
, struct intel_super
*super
, char *devname
, int keep_fd
)
3657 err
= load_imsm_mpb(fd
, super
, devname
);
3660 err
= load_imsm_disk(fd
, super
, devname
, keep_fd
);
3663 err
= parse_raid_devices(super
);
3668 static void __free_imsm_disk(struct dl
*d
)
3680 static void free_imsm_disks(struct intel_super
*super
)
3684 while (super
->disks
) {
3686 super
->disks
= d
->next
;
3687 __free_imsm_disk(d
);
3689 while (super
->disk_mgmt_list
) {
3690 d
= super
->disk_mgmt_list
;
3691 super
->disk_mgmt_list
= d
->next
;
3692 __free_imsm_disk(d
);
3694 while (super
->missing
) {
3696 super
->missing
= d
->next
;
3697 __free_imsm_disk(d
);
3702 /* free all the pieces hanging off of a super pointer */
3703 static void __free_imsm(struct intel_super
*super
, int free_disks
)
3705 struct intel_hba
*elem
, *next
;
3711 /* unlink capability description */
3713 if (super
->migr_rec_buf
) {
3714 free(super
->migr_rec_buf
);
3715 super
->migr_rec_buf
= NULL
;
3718 free_imsm_disks(super
);
3719 free_devlist(super
);
3723 free((void *)elem
->path
);
3731 static void free_imsm(struct intel_super
*super
)
3733 __free_imsm(super
, 1);
3737 static void free_super_imsm(struct supertype
*st
)
3739 struct intel_super
*super
= st
->sb
;
3748 static struct intel_super
*alloc_super(void)
3750 struct intel_super
*super
= malloc(sizeof(*super
));
3753 memset(super
, 0, sizeof(*super
));
3754 super
->current_vol
= -1;
3755 super
->create_offset
= ~((unsigned long long) 0);
3761 * find and allocate hba and OROM/EFI based on valid fd of RAID component device
3763 static int find_intel_hba_capability(int fd
, struct intel_super
*super
, char *devname
)
3765 struct sys_dev
*hba_name
;
3768 if ((fd
< 0) || check_env("IMSM_NO_PLATFORM")) {
3773 hba_name
= find_disk_attached_hba(fd
, NULL
);
3777 Name
": %s is not attached to Intel(R) RAID controller.\n",
3781 rv
= attach_hba_to_super(super
, hba_name
);
3784 struct intel_hba
*hba
= super
->hba
;
3786 fprintf(stderr
, Name
": %s is attached to Intel(R) %s RAID "
3787 "controller (%s),\n"
3788 " but the container is assigned to Intel(R) "
3789 "%s RAID controller (",
3792 hba_name
->pci_id
? : "Err!",
3793 get_sys_dev_type(hba_name
->type
));
3796 fprintf(stderr
, "%s", hba
->pci_id
? : "Err!");
3798 fprintf(stderr
, ", ");
3802 fprintf(stderr
, ").\n"
3803 " Mixing devices attached to different controllers "
3804 "is not allowed.\n");
3806 free_sys_dev(&hba_name
);
3809 super
->orom
= find_imsm_capability(hba_name
->type
);
3810 free_sys_dev(&hba_name
);
3816 /* find_missing - helper routine for load_super_imsm_all that identifies
3817 * disks that have disappeared from the system. This routine relies on
3818 * the mpb being uptodate, which it is at load time.
3820 static int find_missing(struct intel_super
*super
)
3823 struct imsm_super
*mpb
= super
->anchor
;
3825 struct imsm_disk
*disk
;
3827 for (i
= 0; i
< mpb
->num_disks
; i
++) {
3828 disk
= __get_imsm_disk(mpb
, i
);
3829 dl
= serial_to_dl(disk
->serial
, super
);
3833 dl
= malloc(sizeof(*dl
));
3839 dl
->devname
= strdup("missing");
3841 serialcpy(dl
->serial
, disk
->serial
);
3844 dl
->next
= super
->missing
;
3845 super
->missing
= dl
;
3852 static struct intel_disk
*disk_list_get(__u8
*serial
, struct intel_disk
*disk_list
)
3854 struct intel_disk
*idisk
= disk_list
;
3857 if (serialcmp(idisk
->disk
.serial
, serial
) == 0)
3859 idisk
= idisk
->next
;
3865 static int __prep_thunderdome(struct intel_super
**table
, int tbl_size
,
3866 struct intel_super
*super
,
3867 struct intel_disk
**disk_list
)
3869 struct imsm_disk
*d
= &super
->disks
->disk
;
3870 struct imsm_super
*mpb
= super
->anchor
;
3873 for (i
= 0; i
< tbl_size
; i
++) {
3874 struct imsm_super
*tbl_mpb
= table
[i
]->anchor
;
3875 struct imsm_disk
*tbl_d
= &table
[i
]->disks
->disk
;
3877 if (tbl_mpb
->family_num
== mpb
->family_num
) {
3878 if (tbl_mpb
->check_sum
== mpb
->check_sum
) {
3879 dprintf("%s: mpb from %d:%d matches %d:%d\n",
3880 __func__
, super
->disks
->major
,
3881 super
->disks
->minor
,
3882 table
[i
]->disks
->major
,
3883 table
[i
]->disks
->minor
);
3887 if (((is_configured(d
) && !is_configured(tbl_d
)) ||
3888 is_configured(d
) == is_configured(tbl_d
)) &&
3889 tbl_mpb
->generation_num
< mpb
->generation_num
) {
3890 /* current version of the mpb is a
3891 * better candidate than the one in
3892 * super_table, but copy over "cross
3893 * generational" status
3895 struct intel_disk
*idisk
;
3897 dprintf("%s: mpb from %d:%d replaces %d:%d\n",
3898 __func__
, super
->disks
->major
,
3899 super
->disks
->minor
,
3900 table
[i
]->disks
->major
,
3901 table
[i
]->disks
->minor
);
3903 idisk
= disk_list_get(tbl_d
->serial
, *disk_list
);
3904 if (idisk
&& is_failed(&idisk
->disk
))
3905 tbl_d
->status
|= FAILED_DISK
;
3908 struct intel_disk
*idisk
;
3909 struct imsm_disk
*disk
;
3911 /* tbl_mpb is more up to date, but copy
3912 * over cross generational status before
3915 disk
= __serial_to_disk(d
->serial
, mpb
, NULL
);
3916 if (disk
&& is_failed(disk
))
3917 d
->status
|= FAILED_DISK
;
3919 idisk
= disk_list_get(d
->serial
, *disk_list
);
3922 if (disk
&& is_configured(disk
))
3923 idisk
->disk
.status
|= CONFIGURED_DISK
;
3926 dprintf("%s: mpb from %d:%d prefer %d:%d\n",
3927 __func__
, super
->disks
->major
,
3928 super
->disks
->minor
,
3929 table
[i
]->disks
->major
,
3930 table
[i
]->disks
->minor
);
3938 table
[tbl_size
++] = super
;
3942 /* update/extend the merged list of imsm_disk records */
3943 for (j
= 0; j
< mpb
->num_disks
; j
++) {
3944 struct imsm_disk
*disk
= __get_imsm_disk(mpb
, j
);
3945 struct intel_disk
*idisk
;
3947 idisk
= disk_list_get(disk
->serial
, *disk_list
);
3949 idisk
->disk
.status
|= disk
->status
;
3950 if (is_configured(&idisk
->disk
) ||
3951 is_failed(&idisk
->disk
))
3952 idisk
->disk
.status
&= ~(SPARE_DISK
);
3954 idisk
= calloc(1, sizeof(*idisk
));
3957 idisk
->owner
= IMSM_UNKNOWN_OWNER
;
3958 idisk
->disk
= *disk
;
3959 idisk
->next
= *disk_list
;
3963 if (serialcmp(idisk
->disk
.serial
, d
->serial
) == 0)
3970 static struct intel_super
*
3971 validate_members(struct intel_super
*super
, struct intel_disk
*disk_list
,
3974 struct imsm_super
*mpb
= super
->anchor
;
3978 for (i
= 0; i
< mpb
->num_disks
; i
++) {
3979 struct imsm_disk
*disk
= __get_imsm_disk(mpb
, i
);
3980 struct intel_disk
*idisk
;
3982 idisk
= disk_list_get(disk
->serial
, disk_list
);
3984 if (idisk
->owner
== owner
||
3985 idisk
->owner
== IMSM_UNKNOWN_OWNER
)
3988 dprintf("%s: '%.16s' owner %d != %d\n",
3989 __func__
, disk
->serial
, idisk
->owner
,
3992 dprintf("%s: unknown disk %x [%d]: %.16s\n",
3993 __func__
, __le32_to_cpu(mpb
->family_num
), i
,
3999 if (ok_count
== mpb
->num_disks
)
4004 static void show_conflicts(__u32 family_num
, struct intel_super
*super_list
)
4006 struct intel_super
*s
;
4008 for (s
= super_list
; s
; s
= s
->next
) {
4009 if (family_num
!= s
->anchor
->family_num
)
4011 fprintf(stderr
, "Conflict, offlining family %#x on '%s'\n",
4012 __le32_to_cpu(family_num
), s
->disks
->devname
);
4016 static struct intel_super
*
4017 imsm_thunderdome(struct intel_super
**super_list
, int len
)
4019 struct intel_super
*super_table
[len
];
4020 struct intel_disk
*disk_list
= NULL
;
4021 struct intel_super
*champion
, *spare
;
4022 struct intel_super
*s
, **del
;
4027 memset(super_table
, 0, sizeof(super_table
));
4028 for (s
= *super_list
; s
; s
= s
->next
)
4029 tbl_size
= __prep_thunderdome(super_table
, tbl_size
, s
, &disk_list
);
4031 for (i
= 0; i
< tbl_size
; i
++) {
4032 struct imsm_disk
*d
;
4033 struct intel_disk
*idisk
;
4034 struct imsm_super
*mpb
= super_table
[i
]->anchor
;
4037 d
= &s
->disks
->disk
;
4039 /* 'd' must appear in merged disk list for its
4040 * configuration to be valid
4042 idisk
= disk_list_get(d
->serial
, disk_list
);
4043 if (idisk
&& idisk
->owner
== i
)
4044 s
= validate_members(s
, disk_list
, i
);
4049 dprintf("%s: marking family: %#x from %d:%d offline\n",
4050 __func__
, mpb
->family_num
,
4051 super_table
[i
]->disks
->major
,
4052 super_table
[i
]->disks
->minor
);
4056 /* This is where the mdadm implementation differs from the Windows
4057 * driver which has no strict concept of a container. We can only
4058 * assemble one family from a container, so when returning a prodigal
4059 * array member to this system the code will not be able to disambiguate
4060 * the container contents that should be assembled ("foreign" versus
4061 * "local"). It requires user intervention to set the orig_family_num
4062 * to a new value to establish a new container. The Windows driver in
4063 * this situation fixes up the volume name in place and manages the
4064 * foreign array as an independent entity.
4069 for (i
= 0; i
< tbl_size
; i
++) {
4070 struct intel_super
*tbl_ent
= super_table
[i
];
4076 if (tbl_ent
->anchor
->num_raid_devs
== 0) {
4081 if (s
&& !is_spare
) {
4082 show_conflicts(tbl_ent
->anchor
->family_num
, *super_list
);
4084 } else if (!s
&& !is_spare
)
4097 fprintf(stderr
, "Chose family %#x on '%s', "
4098 "assemble conflicts to new container with '--update=uuid'\n",
4099 __le32_to_cpu(s
->anchor
->family_num
), s
->disks
->devname
);
4101 /* collect all dl's onto 'champion', and update them to
4102 * champion's version of the status
4104 for (s
= *super_list
; s
; s
= s
->next
) {
4105 struct imsm_super
*mpb
= champion
->anchor
;
4106 struct dl
*dl
= s
->disks
;
4111 mpb
->attributes
|= s
->anchor
->attributes
& MPB_ATTRIB_2TB_DISK
;
4113 for (i
= 0; i
< mpb
->num_disks
; i
++) {
4114 struct imsm_disk
*disk
;
4116 disk
= __serial_to_disk(dl
->serial
, mpb
, &dl
->index
);
4119 /* only set index on disks that are a member of
4120 * a populated contianer, i.e. one with
4123 if (is_failed(&dl
->disk
))
4125 else if (is_spare(&dl
->disk
))
4131 if (i
>= mpb
->num_disks
) {
4132 struct intel_disk
*idisk
;
4134 idisk
= disk_list_get(dl
->serial
, disk_list
);
4135 if (idisk
&& is_spare(&idisk
->disk
) &&
4136 !is_failed(&idisk
->disk
) && !is_configured(&idisk
->disk
))
4144 dl
->next
= champion
->disks
;
4145 champion
->disks
= dl
;
4149 /* delete 'champion' from super_list */
4150 for (del
= super_list
; *del
; ) {
4151 if (*del
== champion
) {
4152 *del
= (*del
)->next
;
4155 del
= &(*del
)->next
;
4157 champion
->next
= NULL
;
4161 struct intel_disk
*idisk
= disk_list
;
4163 disk_list
= disk_list
->next
;
4172 get_sra_super_block(int fd
, struct intel_super
**super_list
, char *devname
, int *max
, int keep_fd
);
4173 static int get_super_block(struct intel_super
**super_list
, int devnum
, char *devname
,
4174 int major
, int minor
, int keep_fd
);
4176 get_devlist_super_block(struct md_list
*devlist
, struct intel_super
**super_list
,
4177 int *max
, int keep_fd
);
4180 static int load_super_imsm_all(struct supertype
*st
, int fd
, void **sbp
,
4181 char *devname
, struct md_list
*devlist
,
4184 struct intel_super
*super_list
= NULL
;
4185 struct intel_super
*super
= NULL
;
4190 /* 'fd' is an opened container */
4191 err
= get_sra_super_block(fd
, &super_list
, devname
, &i
, keep_fd
);
4193 /* get super block from devlist devices */
4194 err
= get_devlist_super_block(devlist
, &super_list
, &i
, keep_fd
);
4197 /* all mpbs enter, maybe one leaves */
4198 super
= imsm_thunderdome(&super_list
, i
);
4204 if (find_missing(super
) != 0) {
4210 /* load migration record */
4211 err
= load_imsm_migr_rec(super
, NULL
);
4213 /* migration is in progress,
4214 * but migr_rec cannot be loaded,
4220 /* Check migration compatibility */
4221 if ((err
== 0) && (check_mpb_migr_compatibility(super
) != 0)) {
4222 fprintf(stderr
, Name
": Unsupported migration detected");
4224 fprintf(stderr
, " on %s\n", devname
);
4226 fprintf(stderr
, " (IMSM).\n");
4235 while (super_list
) {
4236 struct intel_super
*s
= super_list
;
4238 super_list
= super_list
->next
;
4248 st
->container_dev
= fd2devnum(fd
);
4250 st
->container_dev
= NoMdDev
;
4251 if (err
== 0 && st
->ss
== NULL
) {
4252 st
->ss
= &super_imsm
;
4253 st
->minor_version
= 0;
4254 st
->max_devs
= IMSM_MAX_DEVICES
;
4261 get_devlist_super_block(struct md_list
*devlist
, struct intel_super
**super_list
,
4262 int *max
, int keep_fd
)
4264 struct md_list
*tmpdev
;
4268 for (i
= 0, tmpdev
= devlist
; tmpdev
; tmpdev
= tmpdev
->next
) {
4269 if (tmpdev
->used
!= 1)
4271 if (tmpdev
->container
== 1) {
4273 int fd
= dev_open(tmpdev
->devname
, O_RDONLY
|O_EXCL
);
4275 fprintf(stderr
, Name
": cannot open device %s: %s\n",
4276 tmpdev
->devname
, strerror(errno
));
4280 err
= get_sra_super_block(fd
, super_list
,
4281 tmpdev
->devname
, &lmax
,
4290 int major
= major(tmpdev
->st_rdev
);
4291 int minor
= minor(tmpdev
->st_rdev
);
4292 err
= get_super_block(super_list
,
4309 static int get_super_block(struct intel_super
**super_list
, int devnum
, char *devname
,
4310 int major
, int minor
, int keep_fd
)
4312 struct intel_super
*s
= NULL
;
4325 sprintf(nm
, "%d:%d", major
, minor
);
4326 dfd
= dev_open(nm
, O_RDWR
);
4332 rv
= find_intel_hba_capability(dfd
, s
, devname
);
4333 /* no orom/efi or non-intel hba of the disk */
4339 err
= load_and_parse_mpb(dfd
, s
, NULL
, keep_fd
);
4341 /* retry the load if we might have raced against mdmon */
4342 if (err
== 3 && (devnum
!= -1) && mdmon_running(devnum
))
4343 for (retry
= 0; retry
< 3; retry
++) {
4345 err
= load_and_parse_mpb(dfd
, s
, NULL
, keep_fd
);
4351 s
->next
= *super_list
;
4359 if ((dfd
>= 0) && (!keep_fd
))
4366 get_sra_super_block(int fd
, struct intel_super
**super_list
, char *devname
, int *max
, int keep_fd
)
4373 sra
= sysfs_read(fd
, 0, GET_LEVEL
|GET_VERSION
|GET_DEVS
|GET_STATE
);
4377 if (sra
->array
.major_version
!= -1 ||
4378 sra
->array
.minor_version
!= -2 ||
4379 strcmp(sra
->text_version
, "imsm") != 0) {
4384 devnum
= fd2devnum(fd
);
4385 for (sd
= sra
->devs
, i
= 0; sd
; sd
= sd
->next
, i
++) {
4386 if (get_super_block(super_list
, devnum
, devname
,
4387 sd
->disk
.major
, sd
->disk
.minor
, keep_fd
) != 0) {
4398 static int load_container_imsm(struct supertype
*st
, int fd
, char *devname
)
4400 return load_super_imsm_all(st
, fd
, &st
->sb
, devname
, NULL
, 1);
4404 static int load_super_imsm(struct supertype
*st
, int fd
, char *devname
)
4406 struct intel_super
*super
;
4409 if (test_partition(fd
))
4410 /* IMSM not allowed on partitions */
4413 free_super_imsm(st
);
4415 super
= alloc_super();
4418 Name
": malloc of %zu failed.\n",
4422 /* Load hba and capabilities if they exist.
4423 * But do not preclude loading metadata in case capabilities or hba are
4424 * non-compliant and ignore_hw_compat is set.
4426 rv
= find_intel_hba_capability(fd
, super
, devname
);
4427 /* no orom/efi or non-intel hba of the disk */
4428 if ((rv
!= 0) && (st
->ignore_hw_compat
== 0)) {
4431 Name
": No OROM/EFI properties for %s\n", devname
);
4435 rv
= load_and_parse_mpb(fd
, super
, devname
, 0);
4440 Name
": Failed to load all information "
4441 "sections on %s\n", devname
);
4447 if (st
->ss
== NULL
) {
4448 st
->ss
= &super_imsm
;
4449 st
->minor_version
= 0;
4450 st
->max_devs
= IMSM_MAX_DEVICES
;
4453 /* load migration record */
4454 if (load_imsm_migr_rec(super
, NULL
) == 0) {
4455 /* Check for unsupported migration features */
4456 if (check_mpb_migr_compatibility(super
) != 0) {
4458 Name
": Unsupported migration detected");
4460 fprintf(stderr
, " on %s\n", devname
);
4462 fprintf(stderr
, " (IMSM).\n");
4470 static __u16
info_to_blocks_per_strip(mdu_array_info_t
*info
)
4472 if (info
->level
== 1)
4474 return info
->chunk_size
>> 9;
4477 static unsigned long long info_to_blocks_per_member(mdu_array_info_t
*info
,
4478 unsigned long long size
)
4480 if (info
->level
== 1)
4483 return (size
* 2) & ~(info_to_blocks_per_strip(info
) - 1);
4486 static void imsm_update_version_info(struct intel_super
*super
)
4488 /* update the version and attributes */
4489 struct imsm_super
*mpb
= super
->anchor
;
4491 struct imsm_dev
*dev
;
4492 struct imsm_map
*map
;
4495 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
4496 dev
= get_imsm_dev(super
, i
);
4497 map
= get_imsm_map(dev
, MAP_0
);
4498 if (__le32_to_cpu(dev
->size_high
) > 0)
4499 mpb
->attributes
|= MPB_ATTRIB_2TB
;
4501 /* FIXME detect when an array spans a port multiplier */
4503 mpb
->attributes
|= MPB_ATTRIB_PM
;
4506 if (mpb
->num_raid_devs
> 1 ||
4507 mpb
->attributes
!= MPB_ATTRIB_CHECKSUM_VERIFY
) {
4508 version
= MPB_VERSION_ATTRIBS
;
4509 switch (get_imsm_raid_level(map
)) {
4510 case 0: mpb
->attributes
|= MPB_ATTRIB_RAID0
; break;
4511 case 1: mpb
->attributes
|= MPB_ATTRIB_RAID1
; break;
4512 case 10: mpb
->attributes
|= MPB_ATTRIB_RAID10
; break;
4513 case 5: mpb
->attributes
|= MPB_ATTRIB_RAID5
; break;
4516 if (map
->num_members
>= 5)
4517 version
= MPB_VERSION_5OR6_DISK_ARRAY
;
4518 else if (dev
->status
== DEV_CLONE_N_GO
)
4519 version
= MPB_VERSION_CNG
;
4520 else if (get_imsm_raid_level(map
) == 5)
4521 version
= MPB_VERSION_RAID5
;
4522 else if (map
->num_members
>= 3)
4523 version
= MPB_VERSION_3OR4_DISK_ARRAY
;
4524 else if (get_imsm_raid_level(map
) == 1)
4525 version
= MPB_VERSION_RAID1
;
4527 version
= MPB_VERSION_RAID0
;
4529 strcpy(((char *) mpb
->sig
) + strlen(MPB_SIGNATURE
), version
);
4533 static int check_name(struct intel_super
*super
, char *name
, int quiet
)
4535 struct imsm_super
*mpb
= super
->anchor
;
4536 char *reason
= NULL
;
4539 if (strlen(name
) > MAX_RAID_SERIAL_LEN
)
4540 reason
= "must be 16 characters or less";
4542 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
4543 struct imsm_dev
*dev
= get_imsm_dev(super
, i
);
4545 if (strncmp((char *) dev
->volume
, name
, MAX_RAID_SERIAL_LEN
) == 0) {
4546 reason
= "already exists";
4551 if (reason
&& !quiet
)
4552 fprintf(stderr
, Name
": imsm volume name %s\n", reason
);
4557 static int init_super_imsm_volume(struct supertype
*st
, mdu_array_info_t
*info
,
4558 unsigned long long size
, char *name
,
4559 char *homehost
, int *uuid
)
4561 /* We are creating a volume inside a pre-existing container.
4562 * so st->sb is already set.
4564 struct intel_super
*super
= st
->sb
;
4565 struct imsm_super
*mpb
= super
->anchor
;
4566 struct intel_dev
*dv
;
4567 struct imsm_dev
*dev
;
4568 struct imsm_vol
*vol
;
4569 struct imsm_map
*map
;
4570 int idx
= mpb
->num_raid_devs
;
4572 unsigned long long array_blocks
;
4573 size_t size_old
, size_new
;
4574 unsigned long long num_data_stripes
;
4576 if (super
->orom
&& mpb
->num_raid_devs
>= super
->orom
->vpa
) {
4577 fprintf(stderr
, Name
": This imsm-container already has the "
4578 "maximum of %d volumes\n", super
->orom
->vpa
);
4582 /* ensure the mpb is large enough for the new data */
4583 size_old
= __le32_to_cpu(mpb
->mpb_size
);
4584 size_new
= disks_to_mpb_size(info
->nr_disks
);
4585 if (size_new
> size_old
) {
4587 size_t size_round
= ROUND_UP(size_new
, 512);
4589 if (posix_memalign(&mpb_new
, 512, size_round
) != 0) {
4590 fprintf(stderr
, Name
": could not allocate new mpb\n");
4593 if (posix_memalign(&super
->migr_rec_buf
, 512,
4594 MIGR_REC_BUF_SIZE
) != 0) {
4595 fprintf(stderr
, Name
4596 ": %s could not allocate migr_rec buffer\n",
4603 memcpy(mpb_new
, mpb
, size_old
);
4606 super
->anchor
= mpb_new
;
4607 mpb
->mpb_size
= __cpu_to_le32(size_new
);
4608 memset(mpb_new
+ size_old
, 0, size_round
- size_old
);
4610 super
->current_vol
= idx
;
4612 /* handle 'failed_disks' by either:
4613 * a) create dummy disk entries in the table if this the first
4614 * volume in the array. We add them here as this is the only
4615 * opportunity to add them. add_to_super_imsm_volume()
4616 * handles the non-failed disks and continues incrementing
4618 * b) validate that 'failed_disks' matches the current number
4619 * of missing disks if the container is populated
4621 if (super
->current_vol
== 0) {
4623 for (i
= 0; i
< info
->failed_disks
; i
++) {
4624 struct imsm_disk
*disk
;
4627 disk
= __get_imsm_disk(mpb
, i
);
4628 disk
->status
= CONFIGURED_DISK
| FAILED_DISK
;
4629 disk
->scsi_id
= __cpu_to_le32(~(__u32
)0);
4630 snprintf((char *) disk
->serial
, MAX_RAID_SERIAL_LEN
,
4633 find_missing(super
);
4638 for (d
= super
->missing
; d
; d
= d
->next
)
4640 if (info
->failed_disks
> missing
) {
4641 fprintf(stderr
, Name
": unable to add 'missing' disk to container\n");
4646 if (!check_name(super
, name
, 0))
4648 dv
= malloc(sizeof(*dv
));
4650 fprintf(stderr
, Name
": failed to allocate device list entry\n");
4653 dev
= calloc(1, sizeof(*dev
) + sizeof(__u32
) * (info
->raid_disks
- 1));
4656 fprintf(stderr
, Name
": could not allocate raid device\n");
4660 strncpy((char *) dev
->volume
, name
, MAX_RAID_SERIAL_LEN
);
4661 array_blocks
= calc_array_size(info
->level
, info
->raid_disks
,
4662 info
->layout
, info
->chunk_size
,
4664 /* round array size down to closest MB */
4665 array_blocks
= (array_blocks
>> SECT_PER_MB_SHIFT
) << SECT_PER_MB_SHIFT
;
4667 dev
->size_low
= __cpu_to_le32((__u32
) array_blocks
);
4668 dev
->size_high
= __cpu_to_le32((__u32
) (array_blocks
>> 32));
4669 dev
->status
= (DEV_READ_COALESCING
| DEV_WRITE_COALESCING
);
4671 vol
->migr_state
= 0;
4672 set_migr_type(dev
, MIGR_INIT
);
4673 vol
->dirty
= !info
->state
;
4674 vol
->curr_migr_unit
= 0;
4675 map
= get_imsm_map(dev
, MAP_0
);
4676 set_pba_of_lba0(map
, super
->create_offset
);
4677 set_blocks_per_member(map
, info_to_blocks_per_member(info
, size
));
4678 map
->blocks_per_strip
= __cpu_to_le16(info_to_blocks_per_strip(info
));
4679 map
->failed_disk_num
= ~0;
4680 if (info
->level
> 0)
4681 map
->map_state
= IMSM_T_STATE_UNINITIALIZED
;
4683 map
->map_state
= info
->failed_disks
? IMSM_T_STATE_FAILED
:
4684 IMSM_T_STATE_NORMAL
;
4687 if (info
->level
== 1 && info
->raid_disks
> 2) {
4690 fprintf(stderr
, Name
": imsm does not support more than 2 disks"
4691 "in a raid1 volume\n");
4695 map
->raid_level
= info
->level
;
4696 if (info
->level
== 10) {
4697 map
->raid_level
= 1;
4698 map
->num_domains
= info
->raid_disks
/ 2;
4699 } else if (info
->level
== 1)
4700 map
->num_domains
= info
->raid_disks
;
4702 map
->num_domains
= 1;
4704 /* info->size is only int so use the 'size' parameter instead */
4705 num_data_stripes
= (size
* 2) / info_to_blocks_per_strip(info
);
4706 num_data_stripes
/= map
->num_domains
;
4707 set_num_data_stripes(map
, num_data_stripes
);
4709 map
->num_members
= info
->raid_disks
;
4710 for (i
= 0; i
< map
->num_members
; i
++) {
4711 /* initialized in add_to_super */
4712 set_imsm_ord_tbl_ent(map
, i
, IMSM_ORD_REBUILD
);
4714 mpb
->num_raid_devs
++;
4717 dv
->index
= super
->current_vol
;
4718 dv
->next
= super
->devlist
;
4719 super
->devlist
= dv
;
4721 imsm_update_version_info(super
);
4726 static int init_super_imsm(struct supertype
*st
, mdu_array_info_t
*info
,
4727 unsigned long long size
, char *name
,
4728 char *homehost
, int *uuid
)
4730 /* This is primarily called by Create when creating a new array.
4731 * We will then get add_to_super called for each component, and then
4732 * write_init_super called to write it out to each device.
4733 * For IMSM, Create can create on fresh devices or on a pre-existing
4735 * To create on a pre-existing array a different method will be called.
4736 * This one is just for fresh drives.
4738 struct intel_super
*super
;
4739 struct imsm_super
*mpb
;
4744 return init_super_imsm_volume(st
, info
, size
, name
, homehost
, uuid
);
4747 mpb_size
= disks_to_mpb_size(info
->nr_disks
);
4751 super
= alloc_super();
4752 if (super
&& posix_memalign(&super
->buf
, 512, mpb_size
) != 0) {
4757 fprintf(stderr
, Name
4758 ": %s could not allocate superblock\n", __func__
);
4761 if (posix_memalign(&super
->migr_rec_buf
, 512, MIGR_REC_BUF_SIZE
) != 0) {
4762 fprintf(stderr
, Name
4763 ": %s could not allocate migr_rec buffer\n", __func__
);
4768 memset(super
->buf
, 0, mpb_size
);
4770 mpb
->mpb_size
= __cpu_to_le32(mpb_size
);
4774 /* zeroing superblock */
4778 mpb
->attributes
= MPB_ATTRIB_CHECKSUM_VERIFY
;
4780 version
= (char *) mpb
->sig
;
4781 strcpy(version
, MPB_SIGNATURE
);
4782 version
+= strlen(MPB_SIGNATURE
);
4783 strcpy(version
, MPB_VERSION_RAID0
);
4789 static int add_to_super_imsm_volume(struct supertype
*st
, mdu_disk_info_t
*dk
,
4790 int fd
, char *devname
)
4792 struct intel_super
*super
= st
->sb
;
4793 struct imsm_super
*mpb
= super
->anchor
;
4794 struct imsm_disk
*_disk
;
4795 struct imsm_dev
*dev
;
4796 struct imsm_map
*map
;
4800 dev
= get_imsm_dev(super
, super
->current_vol
);
4801 map
= get_imsm_map(dev
, MAP_0
);
4803 if (! (dk
->state
& (1<<MD_DISK_SYNC
))) {
4804 fprintf(stderr
, Name
": %s: Cannot add spare devices to IMSM volume\n",
4810 /* we're doing autolayout so grab the pre-marked (in
4811 * validate_geometry) raid_disk
4813 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
4814 if (dl
->raiddisk
== dk
->raid_disk
)
4817 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
4818 if (dl
->major
== dk
->major
&&
4819 dl
->minor
== dk
->minor
)
4824 fprintf(stderr
, Name
": %s is not a member of the same container\n", devname
);
4828 /* add a pristine spare to the metadata */
4829 if (dl
->index
< 0) {
4830 dl
->index
= super
->anchor
->num_disks
;
4831 super
->anchor
->num_disks
++;
4833 /* Check the device has not already been added */
4834 slot
= get_imsm_disk_slot(map
, dl
->index
);
4836 (get_imsm_ord_tbl_ent(dev
, slot
, MAP_X
) & IMSM_ORD_REBUILD
) == 0) {
4837 fprintf(stderr
, Name
": %s has been included in this array twice\n",
4841 set_imsm_ord_tbl_ent(map
, dk
->raid_disk
, dl
->index
);
4842 dl
->disk
.status
= CONFIGURED_DISK
;
4844 /* update size of 'missing' disks to be at least as large as the
4845 * largest acitve member (we only have dummy missing disks when
4846 * creating the first volume)
4848 if (super
->current_vol
== 0) {
4849 for (df
= super
->missing
; df
; df
= df
->next
) {
4850 if (total_blocks(&dl
->disk
) > total_blocks(&df
->disk
))
4851 set_total_blocks(&df
->disk
, total_blocks(&dl
->disk
));
4852 _disk
= __get_imsm_disk(mpb
, df
->index
);
4857 /* refresh unset/failed slots to point to valid 'missing' entries */
4858 for (df
= super
->missing
; df
; df
= df
->next
)
4859 for (slot
= 0; slot
< mpb
->num_disks
; slot
++) {
4860 __u32 ord
= get_imsm_ord_tbl_ent(dev
, slot
, MAP_X
);
4862 if ((ord
& IMSM_ORD_REBUILD
) == 0)
4864 set_imsm_ord_tbl_ent(map
, slot
, df
->index
| IMSM_ORD_REBUILD
);
4865 if (is_gen_migration(dev
)) {
4866 struct imsm_map
*map2
= get_imsm_map(dev
,
4868 int slot2
= get_imsm_disk_slot(map2
, df
->index
);
4869 if ((slot2
< map2
->num_members
) &&
4871 __u32 ord2
= get_imsm_ord_tbl_ent(dev
,
4874 if ((unsigned)df
->index
==
4876 set_imsm_ord_tbl_ent(map2
,
4882 dprintf("set slot:%d to missing disk:%d\n", slot
, df
->index
);
4886 /* if we are creating the first raid device update the family number */
4887 if (super
->current_vol
== 0) {
4889 struct imsm_dev
*_dev
= __get_imsm_dev(mpb
, 0);
4891 _disk
= __get_imsm_disk(mpb
, dl
->index
);
4892 if (!_dev
|| !_disk
) {
4893 fprintf(stderr
, Name
": BUG mpb setup error\n");
4899 sum
+= __gen_imsm_checksum(mpb
);
4900 mpb
->family_num
= __cpu_to_le32(sum
);
4901 mpb
->orig_family_num
= mpb
->family_num
;
4903 super
->current_disk
= dl
;
4908 * Function marks disk as spare and restores disk serial
4909 * in case it was previously marked as failed by takeover operation
4911 * -1 : critical error
4912 * 0 : disk is marked as spare but serial is not set
4915 int mark_spare(struct dl
*disk
)
4917 __u8 serial
[MAX_RAID_SERIAL_LEN
];
4924 if (!imsm_read_serial(disk
->fd
, NULL
, serial
)) {
4925 /* Restore disk serial number, because takeover marks disk
4926 * as failed and adds to serial ':0' before it becomes
4929 serialcpy(disk
->serial
, serial
);
4930 serialcpy(disk
->disk
.serial
, serial
);
4933 disk
->disk
.status
= SPARE_DISK
;
4939 static int add_to_super_imsm(struct supertype
*st
, mdu_disk_info_t
*dk
,
4940 int fd
, char *devname
)
4942 struct intel_super
*super
= st
->sb
;
4944 unsigned long long size
;
4949 /* If we are on an RAID enabled platform check that the disk is
4950 * attached to the raid controller.
4951 * We do not need to test disks attachment for container based additions,
4952 * they shall be already tested when container was created/assembled.
4954 rv
= find_intel_hba_capability(fd
, super
, devname
);
4955 /* no orom/efi or non-intel hba of the disk */
4957 dprintf("capability: %p fd: %d ret: %d\n",
4958 super
->orom
, fd
, rv
);
4962 if (super
->current_vol
>= 0)
4963 return add_to_super_imsm_volume(st
, dk
, fd
, devname
);
4966 dd
= malloc(sizeof(*dd
));
4969 Name
": malloc failed %s:%d.\n", __func__
, __LINE__
);
4972 memset(dd
, 0, sizeof(*dd
));
4973 dd
->major
= major(stb
.st_rdev
);
4974 dd
->minor
= minor(stb
.st_rdev
);
4975 dd
->devname
= devname
? strdup(devname
) : NULL
;
4978 dd
->action
= DISK_ADD
;
4979 rv
= imsm_read_serial(fd
, devname
, dd
->serial
);
4982 Name
": failed to retrieve scsi serial, aborting\n");
4987 get_dev_size(fd
, NULL
, &size
);
4989 serialcpy(dd
->disk
.serial
, dd
->serial
);
4990 set_total_blocks(&dd
->disk
, size
);
4991 if (__le32_to_cpu(dd
->disk
.total_blocks_hi
) > 0) {
4992 struct imsm_super
*mpb
= super
->anchor
;
4993 mpb
->attributes
|= MPB_ATTRIB_2TB_DISK
;
4996 if (sysfs_disk_to_scsi_id(fd
, &id
) == 0)
4997 dd
->disk
.scsi_id
= __cpu_to_le32(id
);
4999 dd
->disk
.scsi_id
= __cpu_to_le32(0);
5001 if (st
->update_tail
) {
5002 dd
->next
= super
->disk_mgmt_list
;
5003 super
->disk_mgmt_list
= dd
;
5005 dd
->next
= super
->disks
;
5007 super
->updates_pending
++;
5014 static int remove_from_super_imsm(struct supertype
*st
, mdu_disk_info_t
*dk
)
5016 struct intel_super
*super
= st
->sb
;
5019 /* remove from super works only in mdmon - for communication
5020 * manager - monitor. Check if communication memory buffer
5023 if (!st
->update_tail
) {
5025 Name
": %s shall be used in mdmon context only"
5026 "(line %d).\n", __func__
, __LINE__
);
5029 dd
= malloc(sizeof(*dd
));
5032 Name
": malloc failed %s:%d.\n", __func__
, __LINE__
);
5035 memset(dd
, 0, sizeof(*dd
));
5036 dd
->major
= dk
->major
;
5037 dd
->minor
= dk
->minor
;
5040 dd
->action
= DISK_REMOVE
;
5042 dd
->next
= super
->disk_mgmt_list
;
5043 super
->disk_mgmt_list
= dd
;
5049 static int store_imsm_mpb(int fd
, struct imsm_super
*mpb
);
5053 struct imsm_super anchor
;
5054 } spare_record
__attribute__ ((aligned(512)));
5056 /* spare records have their own family number and do not have any defined raid
5059 static int write_super_imsm_spares(struct intel_super
*super
, int doclose
)
5061 struct imsm_super
*mpb
= super
->anchor
;
5062 struct imsm_super
*spare
= &spare_record
.anchor
;
5066 spare
->mpb_size
= __cpu_to_le32(sizeof(struct imsm_super
)),
5067 spare
->generation_num
= __cpu_to_le32(1UL),
5068 spare
->attributes
= MPB_ATTRIB_CHECKSUM_VERIFY
;
5069 spare
->num_disks
= 1,
5070 spare
->num_raid_devs
= 0,
5071 spare
->cache_size
= mpb
->cache_size
,
5072 spare
->pwr_cycle_count
= __cpu_to_le32(1),
5074 snprintf((char *) spare
->sig
, MAX_SIGNATURE_LENGTH
,
5075 MPB_SIGNATURE MPB_VERSION_RAID0
);
5077 for (d
= super
->disks
; d
; d
= d
->next
) {
5081 spare
->disk
[0] = d
->disk
;
5082 if (__le32_to_cpu(d
->disk
.total_blocks_hi
) > 0)
5083 spare
->attributes
|= MPB_ATTRIB_2TB_DISK
;
5085 sum
= __gen_imsm_checksum(spare
);
5086 spare
->family_num
= __cpu_to_le32(sum
);
5087 spare
->orig_family_num
= 0;
5088 sum
= __gen_imsm_checksum(spare
);
5089 spare
->check_sum
= __cpu_to_le32(sum
);
5091 if (store_imsm_mpb(d
->fd
, spare
)) {
5092 fprintf(stderr
, "%s: failed for device %d:%d %s\n",
5093 __func__
, d
->major
, d
->minor
, strerror(errno
));
5105 static int write_super_imsm(struct supertype
*st
, int doclose
)
5107 struct intel_super
*super
= st
->sb
;
5108 struct imsm_super
*mpb
= super
->anchor
;
5114 __u32 mpb_size
= sizeof(struct imsm_super
) - sizeof(struct imsm_disk
);
5116 int clear_migration_record
= 1;
5118 /* 'generation' is incremented everytime the metadata is written */
5119 generation
= __le32_to_cpu(mpb
->generation_num
);
5121 mpb
->generation_num
= __cpu_to_le32(generation
);
5123 /* fix up cases where previous mdadm releases failed to set
5126 if (mpb
->orig_family_num
== 0)
5127 mpb
->orig_family_num
= mpb
->family_num
;
5129 for (d
= super
->disks
; d
; d
= d
->next
) {
5133 mpb
->disk
[d
->index
] = d
->disk
;
5137 for (d
= super
->missing
; d
; d
= d
->next
) {
5138 mpb
->disk
[d
->index
] = d
->disk
;
5141 mpb
->num_disks
= num_disks
;
5142 mpb_size
+= sizeof(struct imsm_disk
) * mpb
->num_disks
;
5144 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
5145 struct imsm_dev
*dev
= __get_imsm_dev(mpb
, i
);
5146 struct imsm_dev
*dev2
= get_imsm_dev(super
, i
);
5148 imsm_copy_dev(dev
, dev2
);
5149 mpb_size
+= sizeof_imsm_dev(dev
, 0);
5151 if (is_gen_migration(dev2
))
5152 clear_migration_record
= 0;
5154 mpb_size
+= __le32_to_cpu(mpb
->bbm_log_size
);
5155 mpb
->mpb_size
= __cpu_to_le32(mpb_size
);
5157 /* recalculate checksum */
5158 sum
= __gen_imsm_checksum(mpb
);
5159 mpb
->check_sum
= __cpu_to_le32(sum
);
5161 if (super
->clean_migration_record_by_mdmon
) {
5162 clear_migration_record
= 1;
5163 super
->clean_migration_record_by_mdmon
= 0;
5165 if (clear_migration_record
)
5166 memset(super
->migr_rec_buf
, 0, MIGR_REC_BUF_SIZE
);
5168 /* write the mpb for disks that compose raid devices */
5169 for (d
= super
->disks
; d
; d
= d
->next
) {
5170 if (d
->index
< 0 || is_failed(&d
->disk
))
5173 if (clear_migration_record
) {
5174 unsigned long long dsize
;
5176 get_dev_size(d
->fd
, NULL
, &dsize
);
5177 if (lseek64(d
->fd
, dsize
- 512, SEEK_SET
) >= 0) {
5178 if (write(d
->fd
, super
->migr_rec_buf
,
5179 MIGR_REC_BUF_SIZE
) != MIGR_REC_BUF_SIZE
)
5180 perror("Write migr_rec failed");
5184 if (store_imsm_mpb(d
->fd
, mpb
))
5186 "%s: failed for device %d:%d (fd: %d)%s\n",
5187 __func__
, d
->major
, d
->minor
,
5188 d
->fd
, strerror(errno
));
5197 return write_super_imsm_spares(super
, doclose
);
5203 static int create_array(struct supertype
*st
, int dev_idx
)
5206 struct imsm_update_create_array
*u
;
5207 struct intel_super
*super
= st
->sb
;
5208 struct imsm_dev
*dev
= get_imsm_dev(super
, dev_idx
);
5209 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
5210 struct disk_info
*inf
;
5211 struct imsm_disk
*disk
;
5214 len
= sizeof(*u
) - sizeof(*dev
) + sizeof_imsm_dev(dev
, 0) +
5215 sizeof(*inf
) * map
->num_members
;
5218 fprintf(stderr
, "%s: failed to allocate update buffer\n",
5223 u
->type
= update_create_array
;
5224 u
->dev_idx
= dev_idx
;
5225 imsm_copy_dev(&u
->dev
, dev
);
5226 inf
= get_disk_info(u
);
5227 for (i
= 0; i
< map
->num_members
; i
++) {
5228 int idx
= get_imsm_disk_idx(dev
, i
, MAP_X
);
5230 disk
= get_imsm_disk(super
, idx
);
5231 serialcpy(inf
[i
].serial
, disk
->serial
);
5233 append_metadata_update(st
, u
, len
);
5238 static int mgmt_disk(struct supertype
*st
)
5240 struct intel_super
*super
= st
->sb
;
5242 struct imsm_update_add_remove_disk
*u
;
5244 if (!super
->disk_mgmt_list
)
5250 fprintf(stderr
, "%s: failed to allocate update buffer\n",
5255 u
->type
= update_add_remove_disk
;
5256 append_metadata_update(st
, u
, len
);
5261 static int write_init_super_imsm(struct supertype
*st
)
5263 struct intel_super
*super
= st
->sb
;
5264 int current_vol
= super
->current_vol
;
5266 /* we are done with current_vol reset it to point st at the container */
5267 super
->current_vol
= -1;
5269 if (st
->update_tail
) {
5270 /* queue the recently created array / added disk
5271 * as a metadata update */
5274 /* determine if we are creating a volume or adding a disk */
5275 if (current_vol
< 0) {
5276 /* in the mgmt (add/remove) disk case we are running
5277 * in mdmon context, so don't close fd's
5279 return mgmt_disk(st
);
5281 rv
= create_array(st
, current_vol
);
5286 for (d
= super
->disks
; d
; d
= d
->next
)
5287 Kill(d
->devname
, NULL
, 0, 1, 1);
5288 return write_super_imsm(st
, 1);
5293 static int store_super_imsm(struct supertype
*st
, int fd
)
5295 struct intel_super
*super
= st
->sb
;
5296 struct imsm_super
*mpb
= super
? super
->anchor
: NULL
;
5302 return store_imsm_mpb(fd
, mpb
);
5308 static int imsm_bbm_log_size(struct imsm_super
*mpb
)
5310 return __le32_to_cpu(mpb
->bbm_log_size
);
5314 static int validate_geometry_imsm_container(struct supertype
*st
, int level
,
5315 int layout
, int raiddisks
, int chunk
,
5316 unsigned long long size
, char *dev
,
5317 unsigned long long *freesize
,
5321 unsigned long long ldsize
;
5322 struct intel_super
*super
=NULL
;
5325 if (level
!= LEVEL_CONTAINER
)
5330 fd
= open(dev
, O_RDONLY
|O_EXCL
, 0);
5333 fprintf(stderr
, Name
": imsm: Cannot open %s: %s\n",
5334 dev
, strerror(errno
));
5337 if (!get_dev_size(fd
, dev
, &ldsize
)) {
5342 /* capabilities retrieve could be possible
5343 * note that there is no fd for the disks in array.
5345 super
= alloc_super();
5348 Name
": malloc of %zu failed.\n",
5354 rv
= find_intel_hba_capability(fd
, super
, verbose
? dev
: NULL
);
5358 fd2devname(fd
, str
);
5359 dprintf("validate_geometry_imsm_container: fd: %d %s orom: %p rv: %d raiddisk: %d\n",
5360 fd
, str
, super
->orom
, rv
, raiddisks
);
5362 /* no orom/efi or non-intel hba of the disk */
5369 if (raiddisks
> super
->orom
->tds
) {
5371 fprintf(stderr
, Name
": %d exceeds maximum number of"
5372 " platform supported disks: %d\n",
5373 raiddisks
, super
->orom
->tds
);
5377 if ((super
->orom
->attr
& IMSM_OROM_ATTR_2TB_DISK
) == 0 &&
5378 (ldsize
>> 9) >> 32 > 0) {
5380 fprintf(stderr
, Name
": %s exceeds maximum platform supported size\n", dev
);
5386 *freesize
= avail_size_imsm(st
, ldsize
>> 9);
5392 static unsigned long long find_size(struct extent
*e
, int *idx
, int num_extents
)
5394 const unsigned long long base_start
= e
[*idx
].start
;
5395 unsigned long long end
= base_start
+ e
[*idx
].size
;
5398 if (base_start
== end
)
5402 for (i
= *idx
; i
< num_extents
; i
++) {
5403 /* extend overlapping extents */
5404 if (e
[i
].start
>= base_start
&&
5405 e
[i
].start
<= end
) {
5408 if (e
[i
].start
+ e
[i
].size
> end
)
5409 end
= e
[i
].start
+ e
[i
].size
;
5410 } else if (e
[i
].start
> end
) {
5416 return end
- base_start
;
5419 static unsigned long long merge_extents(struct intel_super
*super
, int sum_extents
)
5421 /* build a composite disk with all known extents and generate a new
5422 * 'maxsize' given the "all disks in an array must share a common start
5423 * offset" constraint
5425 struct extent
*e
= calloc(sum_extents
, sizeof(*e
));
5429 unsigned long long pos
;
5430 unsigned long long start
= 0;
5431 unsigned long long maxsize
;
5432 unsigned long reserve
;
5437 /* coalesce and sort all extents. also, check to see if we need to
5438 * reserve space between member arrays
5441 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
5444 for (i
= 0; i
< dl
->extent_cnt
; i
++)
5447 qsort(e
, sum_extents
, sizeof(*e
), cmp_extent
);
5452 while (i
< sum_extents
) {
5453 e
[j
].start
= e
[i
].start
;
5454 e
[j
].size
= find_size(e
, &i
, sum_extents
);
5456 if (e
[j
-1].size
== 0)
5465 unsigned long long esize
;
5467 esize
= e
[i
].start
- pos
;
5468 if (esize
>= maxsize
) {
5473 pos
= e
[i
].start
+ e
[i
].size
;
5475 } while (e
[i
-1].size
);
5481 /* FIXME assumes volume at offset 0 is the first volume in a
5484 if (start_extent
> 0)
5485 reserve
= IMSM_RESERVED_SECTORS
; /* gap between raid regions */
5489 if (maxsize
< reserve
)
5492 super
->create_offset
= ~((unsigned long long) 0);
5493 if (start
+ reserve
> super
->create_offset
)
5494 return 0; /* start overflows create_offset */
5495 super
->create_offset
= start
+ reserve
;
5497 return maxsize
- reserve
;
5500 static int is_raid_level_supported(const struct imsm_orom
*orom
, int level
, int raiddisks
)
5502 if (level
< 0 || level
== 6 || level
== 4)
5505 /* if we have an orom prevent invalid raid levels */
5508 case 0: return imsm_orom_has_raid0(orom
);
5511 return imsm_orom_has_raid1e(orom
);
5512 return imsm_orom_has_raid1(orom
) && raiddisks
== 2;
5513 case 10: return imsm_orom_has_raid10(orom
) && raiddisks
== 4;
5514 case 5: return imsm_orom_has_raid5(orom
) && raiddisks
> 2;
5517 return 1; /* not on an Intel RAID platform so anything goes */
5524 active_arrays_by_format(char *name
, char* hba
, struct md_list
**devlist
,
5525 int dpa
, int verbose
)
5527 struct mdstat_ent
*mdstat
= mdstat_read(0, 0);
5528 struct mdstat_ent
*memb
= NULL
;
5531 struct md_list
*dv
= NULL
;
5534 for (memb
= mdstat
; memb
; memb
= memb
->next
) {
5535 if (memb
->metadata_version
&&
5536 (strncmp(memb
->metadata_version
, "external:", 9) == 0) &&
5537 (strcmp(&memb
->metadata_version
[9], name
) == 0) &&
5538 !is_subarray(memb
->metadata_version
+9) &&
5540 struct dev_member
*dev
= memb
->members
;
5542 while(dev
&& (fd
< 0)) {
5543 char *path
= malloc(strlen(dev
->name
) + strlen("/dev/") + 1);
5545 num
= sprintf(path
, "%s%s", "/dev/", dev
->name
);
5547 fd
= open(path
, O_RDONLY
, 0);
5548 if ((num
<= 0) || (fd
< 0)) {
5549 pr_vrb(": Cannot open %s: %s\n",
5550 dev
->name
, strerror(errno
));
5557 if ((fd
>= 0) && disk_attached_to_hba(fd
, hba
)) {
5558 struct mdstat_ent
*vol
;
5559 for (vol
= mdstat
; vol
; vol
= vol
->next
) {
5560 if ((vol
->active
> 0) &&
5561 vol
->metadata_version
&&
5562 is_container_member(vol
, memb
->dev
)) {
5567 if (*devlist
&& (found
< dpa
)) {
5568 dv
= calloc(1, sizeof(*dv
));
5570 fprintf(stderr
, Name
": calloc failed\n");
5572 dv
->devname
= malloc(strlen(memb
->dev
) + strlen("/dev/") + 1);
5573 if (dv
->devname
!= NULL
) {
5574 sprintf(dv
->devname
, "%s%s", "/dev/", memb
->dev
);
5577 dv
->next
= *devlist
;
5588 free_mdstat(mdstat
);
5593 static struct md_list
*
5594 get_loop_devices(void)
5597 struct md_list
*devlist
= NULL
;
5598 struct md_list
*dv
= NULL
;
5600 for(i
= 0; i
< 12; i
++) {
5601 dv
= calloc(1, sizeof(*dv
));
5603 fprintf(stderr
, Name
": calloc failed\n");
5606 dv
->devname
= malloc(40);
5607 if (dv
->devname
== NULL
) {
5608 fprintf(stderr
, Name
": malloc failed\n");
5612 sprintf(dv
->devname
, "/dev/loop%d", i
);
5620 static struct md_list
*
5621 get_devices(const char *hba_path
)
5623 struct md_list
*devlist
= NULL
;
5624 struct md_list
*dv
= NULL
;
5630 devlist
= get_loop_devices();
5633 /* scroll through /sys/dev/block looking for devices attached to
5636 dir
= opendir("/sys/dev/block");
5637 for (ent
= dir
? readdir(dir
) : NULL
; ent
; ent
= readdir(dir
)) {
5642 if (sscanf(ent
->d_name
, "%d:%d", &major
, &minor
) != 2)
5644 path
= devt_to_devpath(makedev(major
, minor
));
5647 if (!path_attached_to_hba(path
, hba_path
)) {
5654 fd
= dev_open(ent
->d_name
, O_RDONLY
);
5656 fd2devname(fd
, buf
);
5659 fprintf(stderr
, Name
": cannot open device: %s\n",
5665 dv
= calloc(1, sizeof(*dv
));
5667 fprintf(stderr
, Name
": malloc failed\n");
5671 dv
->devname
= strdup(buf
);
5672 if (dv
->devname
== NULL
) {
5673 fprintf(stderr
, Name
": malloc failed\n");
5684 devlist
= devlist
->next
;
5694 count_volumes_list(struct md_list
*devlist
, char *homehost
,
5695 int verbose
, int *found
)
5697 struct md_list
*tmpdev
;
5699 struct supertype
*st
= NULL
;
5701 /* first walk the list of devices to find a consistent set
5702 * that match the criterea, if that is possible.
5703 * We flag the ones we like with 'used'.
5706 st
= match_metadata_desc_imsm("imsm");
5708 pr_vrb(": cannot allocate memory for imsm supertype\n");
5712 for (tmpdev
= devlist
; tmpdev
; tmpdev
= tmpdev
->next
) {
5713 char *devname
= tmpdev
->devname
;
5715 struct supertype
*tst
;
5717 if (tmpdev
->used
> 1)
5719 tst
= dup_super(st
);
5721 pr_vrb(": cannot allocate memory for imsm supertype\n");
5724 tmpdev
->container
= 0;
5725 dfd
= dev_open(devname
, O_RDONLY
|O_EXCL
);
5727 dprintf(": cannot open device %s: %s\n",
5728 devname
, strerror(errno
));
5730 } else if (fstat(dfd
, &stb
)< 0) {
5732 dprintf(": fstat failed for %s: %s\n",
5733 devname
, strerror(errno
));
5735 } else if ((stb
.st_mode
& S_IFMT
) != S_IFBLK
) {
5736 dprintf(": %s is not a block device.\n",
5739 } else if (must_be_container(dfd
)) {
5740 struct supertype
*cst
;
5741 cst
= super_by_fd(dfd
, NULL
);
5743 dprintf(": cannot recognize container type %s\n",
5746 } else if (tst
->ss
!= st
->ss
) {
5747 dprintf(": non-imsm container - ignore it: %s\n",
5750 } else if (!tst
->ss
->load_container
||
5751 tst
->ss
->load_container(tst
, dfd
, NULL
))
5754 tmpdev
->container
= 1;
5757 cst
->ss
->free_super(cst
);
5759 tmpdev
->st_rdev
= stb
.st_rdev
;
5760 if (tst
->ss
->load_super(tst
,dfd
, NULL
)) {
5761 dprintf(": no RAID superblock on %s\n",
5764 } else if (tst
->ss
->compare_super
== NULL
) {
5765 dprintf(": Cannot assemble %s metadata on %s\n",
5766 tst
->ss
->name
, devname
);
5772 if (tmpdev
->used
== 2 || tmpdev
->used
== 4) {
5773 /* Ignore unrecognised devices during auto-assembly */
5778 tst
->ss
->getinfo_super(tst
, &info
, NULL
);
5780 if (st
->minor_version
== -1)
5781 st
->minor_version
= tst
->minor_version
;
5783 if (memcmp(info
.uuid
, uuid_zero
,
5784 sizeof(int[4])) == 0) {
5785 /* this is a floating spare. It cannot define
5786 * an array unless there are no more arrays of
5787 * this type to be found. It can be included
5788 * in an array of this type though.
5794 if (st
->ss
!= tst
->ss
||
5795 st
->minor_version
!= tst
->minor_version
||
5796 st
->ss
->compare_super(st
, tst
) != 0) {
5797 /* Some mismatch. If exactly one array matches this host,
5798 * we can resolve on that one.
5799 * Or, if we are auto assembling, we just ignore the second
5802 dprintf(": superblock on %s doesn't match others - assembly aborted\n",
5808 dprintf("found: devname: %s\n", devname
);
5812 tst
->ss
->free_super(tst
);
5816 if ((err
= load_super_imsm_all(st
, -1, &st
->sb
, NULL
, devlist
, 0)) == 0) {
5817 struct mdinfo
*iter
, *head
= st
->ss
->container_content(st
, NULL
);
5818 for (iter
= head
; iter
; iter
= iter
->next
) {
5819 dprintf("content->text_version: %s vol\n",
5820 iter
->text_version
);
5821 if (iter
->array
.state
& (1<<MD_SB_BLOCK_VOLUME
)) {
5822 /* do not assemble arrays with unsupported
5824 dprintf(": Cannot activate member %s.\n",
5825 iter
->text_version
);
5832 dprintf(" no valid super block on device list: err: %d %p\n",
5836 dprintf(" no more devices to examin\n");
5839 for (tmpdev
= devlist
; tmpdev
; tmpdev
= tmpdev
->next
) {
5840 if ((tmpdev
->used
== 1) && (tmpdev
->found
)) {
5842 if (count
< tmpdev
->found
)
5845 count
-= tmpdev
->found
;
5848 if (tmpdev
->used
== 1)
5853 st
->ss
->free_super(st
);
5859 count_volumes(char *hba
, int dpa
, int verbose
)
5861 struct md_list
*devlist
= NULL
;
5865 devlist
= get_devices(hba
);
5866 /* if no intel devices return zero volumes */
5867 if (devlist
== NULL
)
5870 count
= active_arrays_by_format("imsm", hba
, &devlist
, dpa
, verbose
);
5871 dprintf(" path: %s active arrays: %d\n", hba
, count
);
5872 if (devlist
== NULL
)
5876 count
+= count_volumes_list(devlist
,
5880 dprintf("found %d count: %d\n", found
, count
);
5883 dprintf("path: %s total number of volumes: %d\n", hba
, count
);
5886 struct md_list
*dv
= devlist
;
5887 devlist
= devlist
->next
;
5894 static int imsm_default_chunk(const struct imsm_orom
*orom
)
5896 /* up to 512 if the plaform supports it, otherwise the platform max.
5897 * 128 if no platform detected
5899 int fs
= max(7, orom
? fls(orom
->sss
) : 0);
5901 return min(512, (1 << fs
));
5905 validate_geometry_imsm_orom(struct intel_super
*super
, int level
, int layout
,
5906 int raiddisks
, int *chunk
, unsigned long long size
, int verbose
)
5908 /* check/set platform and metadata limits/defaults */
5909 if (super
->orom
&& raiddisks
> super
->orom
->dpa
) {
5910 pr_vrb(": platform supports a maximum of %d disks per array\n",
5915 /* capabilities of OROM tested - copied from validate_geometry_imsm_volume */
5916 if (!is_raid_level_supported(super
->orom
, level
, raiddisks
)) {
5917 pr_vrb(": platform does not support raid%d with %d disk%s\n",
5918 level
, raiddisks
, raiddisks
> 1 ? "s" : "");
5922 if (chunk
&& (*chunk
== 0 || *chunk
== UnSet
))
5923 *chunk
= imsm_default_chunk(super
->orom
);
5925 if (super
->orom
&& chunk
&& !imsm_orom_has_chunk(super
->orom
, *chunk
)) {
5926 pr_vrb(": platform does not support a chunk size of: "
5931 if (layout
!= imsm_level_to_layout(level
)) {
5933 pr_vrb(": imsm raid 5 only supports the left-asymmetric layout\n");
5934 else if (level
== 10)
5935 pr_vrb(": imsm raid 10 only supports the n2 layout\n");
5937 pr_vrb(": imsm unknown layout %#x for this raid level %d\n",
5942 if (super
->orom
&& (super
->orom
->attr
& IMSM_OROM_ATTR_2TB
) == 0 && chunk
&&
5943 (calc_array_size(level
, raiddisks
, layout
, *chunk
, size
) >> 32) > 0) {
5944 pr_vrb(": platform does not support a volume size over 2TB\n");
5950 /* validate_geometry_imsm_volume - lifted from validate_geometry_ddf_bvd
5951 * FIX ME add ahci details
5953 static int validate_geometry_imsm_volume(struct supertype
*st
, int level
,
5954 int layout
, int raiddisks
, int *chunk
,
5955 unsigned long long size
, char *dev
,
5956 unsigned long long *freesize
,
5960 struct intel_super
*super
= st
->sb
;
5961 struct imsm_super
*mpb
;
5963 unsigned long long pos
= 0;
5964 unsigned long long maxsize
;
5968 /* We must have the container info already read in. */
5972 mpb
= super
->anchor
;
5974 if (!validate_geometry_imsm_orom(super
, level
, layout
, raiddisks
, chunk
, size
, verbose
)) {
5975 fprintf(stderr
, Name
": RAID gemetry validation failed. "
5976 "Cannot proceed with the action(s).\n");
5980 /* General test: make sure there is space for
5981 * 'raiddisks' device extents of size 'size' at a given
5984 unsigned long long minsize
= size
;
5985 unsigned long long start_offset
= MaxSector
;
5988 minsize
= MPB_SECTOR_CNT
+ IMSM_RESERVED_SECTORS
;
5989 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
5994 e
= get_extents(super
, dl
);
5997 unsigned long long esize
;
5998 esize
= e
[i
].start
- pos
;
5999 if (esize
>= minsize
)
6001 if (found
&& start_offset
== MaxSector
) {
6004 } else if (found
&& pos
!= start_offset
) {
6008 pos
= e
[i
].start
+ e
[i
].size
;
6010 } while (e
[i
-1].size
);
6015 if (dcnt
< raiddisks
) {
6017 fprintf(stderr
, Name
": imsm: Not enough "
6018 "devices with space for this array "
6026 /* This device must be a member of the set */
6027 if (stat(dev
, &stb
) < 0)
6029 if ((S_IFMT
& stb
.st_mode
) != S_IFBLK
)
6031 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
6032 if (dl
->major
== (int)major(stb
.st_rdev
) &&
6033 dl
->minor
== (int)minor(stb
.st_rdev
))
6038 fprintf(stderr
, Name
": %s is not in the "
6039 "same imsm set\n", dev
);
6041 } else if (super
->orom
&& dl
->index
< 0 && mpb
->num_raid_devs
) {
6042 /* If a volume is present then the current creation attempt
6043 * cannot incorporate new spares because the orom may not
6044 * understand this configuration (all member disks must be
6045 * members of each array in the container).
6047 fprintf(stderr
, Name
": %s is a spare and a volume"
6048 " is already defined for this container\n", dev
);
6049 fprintf(stderr
, Name
": The option-rom requires all member"
6050 " disks to be a member of all volumes\n");
6052 } else if (super
->orom
&& mpb
->num_raid_devs
> 0 &&
6053 mpb
->num_disks
!= raiddisks
) {
6054 fprintf(stderr
, Name
": The option-rom requires all member"
6055 " disks to be a member of all volumes\n");
6059 /* retrieve the largest free space block */
6060 e
= get_extents(super
, dl
);
6065 unsigned long long esize
;
6067 esize
= e
[i
].start
- pos
;
6068 if (esize
>= maxsize
)
6070 pos
= e
[i
].start
+ e
[i
].size
;
6072 } while (e
[i
-1].size
);
6077 fprintf(stderr
, Name
": unable to determine free space for: %s\n",
6081 if (maxsize
< size
) {
6083 fprintf(stderr
, Name
": %s not enough space (%llu < %llu)\n",
6084 dev
, maxsize
, size
);
6088 /* count total number of extents for merge */
6090 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
6092 i
+= dl
->extent_cnt
;
6094 maxsize
= merge_extents(super
, i
);
6096 if (!check_env("IMSM_NO_PLATFORM") &&
6097 mpb
->num_raid_devs
> 0 && size
&& size
!= maxsize
) {
6098 fprintf(stderr
, Name
": attempting to create a second "
6099 "volume with size less then remaining space. "
6104 if (maxsize
< size
|| maxsize
== 0) {
6107 fprintf(stderr
, Name
": no free space"
6108 " left on device. Aborting...\n");
6110 fprintf(stderr
, Name
": not enough space"
6111 " to create volume of given size"
6112 " (%llu < %llu). Aborting...\n",
6118 *freesize
= maxsize
;
6121 int count
= count_volumes(super
->hba
->path
,
6122 super
->orom
->dpa
, verbose
);
6123 if (super
->orom
->vphba
<= count
) {
6124 pr_vrb(": platform does not support more than %d raid volumes.\n",
6125 super
->orom
->vphba
);
6132 static int imsm_get_free_size(struct supertype
*st
, int raiddisks
,
6133 unsigned long long size
, int chunk
,
6134 unsigned long long *freesize
)
6136 struct intel_super
*super
= st
->sb
;
6137 struct imsm_super
*mpb
= super
->anchor
;
6142 unsigned long long maxsize
;
6143 unsigned long long minsize
;
6147 /* find the largest common start free region of the possible disks */
6151 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
6157 /* don't activate new spares if we are orom constrained
6158 * and there is already a volume active in the container
6160 if (super
->orom
&& dl
->index
< 0 && mpb
->num_raid_devs
)
6163 e
= get_extents(super
, dl
);
6166 for (i
= 1; e
[i
-1].size
; i
++)
6174 maxsize
= merge_extents(super
, extent_cnt
);
6178 minsize
= chunk
* 2;
6180 if (cnt
< raiddisks
||
6181 (super
->orom
&& used
&& used
!= raiddisks
) ||
6182 maxsize
< minsize
||
6184 fprintf(stderr
, Name
": not enough devices with space to create array.\n");
6185 return 0; /* No enough free spaces large enough */
6196 if (!check_env("IMSM_NO_PLATFORM") &&
6197 mpb
->num_raid_devs
> 0 && size
&& size
!= maxsize
) {
6198 fprintf(stderr
, Name
": attempting to create a second "
6199 "volume with size less then remaining space. "
6204 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
6206 dl
->raiddisk
= cnt
++;
6210 dprintf("imsm: imsm_get_free_size() returns : %llu\n", size
);
6215 static int reserve_space(struct supertype
*st
, int raiddisks
,
6216 unsigned long long size
, int chunk
,
6217 unsigned long long *freesize
)
6219 struct intel_super
*super
= st
->sb
;
6224 rv
= imsm_get_free_size(st
, raiddisks
, size
, chunk
, freesize
);
6227 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
6229 dl
->raiddisk
= cnt
++;
6236 static int validate_geometry_imsm(struct supertype
*st
, int level
, int layout
,
6237 int raiddisks
, int *chunk
, unsigned long long size
,
6238 char *dev
, unsigned long long *freesize
,
6246 * if given unused devices create a container
6247 * if given given devices in a container create a member volume
6249 if (level
== LEVEL_CONTAINER
) {
6250 /* Must be a fresh device to add to a container */
6251 return validate_geometry_imsm_container(st
, level
, layout
,
6253 chunk
?*chunk
:0, size
,
6260 struct intel_super
*super
= st
->sb
;
6261 if (!validate_geometry_imsm_orom(st
->sb
, level
, layout
,
6262 raiddisks
, chunk
, size
,
6265 /* we are being asked to automatically layout a
6266 * new volume based on the current contents of
6267 * the container. If the the parameters can be
6268 * satisfied reserve_space will record the disks,
6269 * start offset, and size of the volume to be
6270 * created. add_to_super and getinfo_super
6271 * detect when autolayout is in progress.
6273 /* assuming that freesize is always given when array is
6275 if (super
->orom
&& freesize
) {
6277 count
= count_volumes(super
->hba
->path
,
6278 super
->orom
->dpa
, verbose
);
6279 if (super
->orom
->vphba
<= count
) {
6280 pr_vrb(": platform does not support more"
6281 " than %d raid volumes.\n",
6282 super
->orom
->vphba
);
6287 return reserve_space(st
, raiddisks
, size
,
6288 chunk
?*chunk
:0, freesize
);
6293 /* creating in a given container */
6294 return validate_geometry_imsm_volume(st
, level
, layout
,
6295 raiddisks
, chunk
, size
,
6296 dev
, freesize
, verbose
);
6299 /* This device needs to be a device in an 'imsm' container */
6300 fd
= open(dev
, O_RDONLY
|O_EXCL
, 0);
6304 Name
": Cannot create this array on device %s\n",
6309 if (errno
!= EBUSY
|| (fd
= open(dev
, O_RDONLY
, 0)) < 0) {
6311 fprintf(stderr
, Name
": Cannot open %s: %s\n",
6312 dev
, strerror(errno
));
6315 /* Well, it is in use by someone, maybe an 'imsm' container. */
6316 cfd
= open_container(fd
);
6320 fprintf(stderr
, Name
": Cannot use %s: It is busy\n",
6324 sra
= sysfs_read(cfd
, 0, GET_VERSION
);
6325 if (sra
&& sra
->array
.major_version
== -1 &&
6326 strcmp(sra
->text_version
, "imsm") == 0)
6330 /* This is a member of a imsm container. Load the container
6331 * and try to create a volume
6333 struct intel_super
*super
;
6335 if (load_super_imsm_all(st
, cfd
, (void **) &super
, NULL
, NULL
, 1) == 0) {
6337 st
->container_dev
= fd2devnum(cfd
);
6339 return validate_geometry_imsm_volume(st
, level
, layout
,
6348 fprintf(stderr
, Name
": failed container membership check\n");
6354 static void default_geometry_imsm(struct supertype
*st
, int *level
, int *layout
, int *chunk
)
6356 struct intel_super
*super
= st
->sb
;
6358 if (level
&& *level
== UnSet
)
6359 *level
= LEVEL_CONTAINER
;
6361 if (level
&& layout
&& *layout
== UnSet
)
6362 *layout
= imsm_level_to_layout(*level
);
6364 if (chunk
&& (*chunk
== UnSet
|| *chunk
== 0))
6365 *chunk
= imsm_default_chunk(super
->orom
);
6368 static void handle_missing(struct intel_super
*super
, struct imsm_dev
*dev
);
6370 static int kill_subarray_imsm(struct supertype
*st
)
6372 /* remove the subarray currently referenced by ->current_vol */
6374 struct intel_dev
**dp
;
6375 struct intel_super
*super
= st
->sb
;
6376 __u8 current_vol
= super
->current_vol
;
6377 struct imsm_super
*mpb
= super
->anchor
;
6379 if (super
->current_vol
< 0)
6381 super
->current_vol
= -1; /* invalidate subarray cursor */
6383 /* block deletions that would change the uuid of active subarrays
6385 * FIXME when immutable ids are available, but note that we'll
6386 * also need to fixup the invalidated/active subarray indexes in
6389 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
6392 if (i
< current_vol
)
6394 sprintf(subarray
, "%u", i
);
6395 if (is_subarray_active(subarray
, st
->devname
)) {
6397 Name
": deleting subarray-%d would change the UUID of active subarray-%d, aborting\n",
6404 if (st
->update_tail
) {
6405 struct imsm_update_kill_array
*u
= malloc(sizeof(*u
));
6409 u
->type
= update_kill_array
;
6410 u
->dev_idx
= current_vol
;
6411 append_metadata_update(st
, u
, sizeof(*u
));
6416 for (dp
= &super
->devlist
; *dp
;)
6417 if ((*dp
)->index
== current_vol
) {
6420 handle_missing(super
, (*dp
)->dev
);
6421 if ((*dp
)->index
> current_vol
)
6426 /* no more raid devices, all active components are now spares,
6427 * but of course failed are still failed
6429 if (--mpb
->num_raid_devs
== 0) {
6432 for (d
= super
->disks
; d
; d
= d
->next
)
6437 super
->updates_pending
++;
6442 static int update_subarray_imsm(struct supertype
*st
, char *subarray
,
6443 char *update
, struct mddev_ident
*ident
)
6445 /* update the subarray currently referenced by ->current_vol */
6446 struct intel_super
*super
= st
->sb
;
6447 struct imsm_super
*mpb
= super
->anchor
;
6449 if (strcmp(update
, "name") == 0) {
6450 char *name
= ident
->name
;
6454 if (is_subarray_active(subarray
, st
->devname
)) {
6456 Name
": Unable to update name of active subarray\n");
6460 if (!check_name(super
, name
, 0))
6463 vol
= strtoul(subarray
, &ep
, 10);
6464 if (*ep
!= '\0' || vol
>= super
->anchor
->num_raid_devs
)
6467 if (st
->update_tail
) {
6468 struct imsm_update_rename_array
*u
= malloc(sizeof(*u
));
6472 u
->type
= update_rename_array
;
6474 snprintf((char *) u
->name
, MAX_RAID_SERIAL_LEN
, "%s", name
);
6475 append_metadata_update(st
, u
, sizeof(*u
));
6477 struct imsm_dev
*dev
;
6480 dev
= get_imsm_dev(super
, vol
);
6481 snprintf((char *) dev
->volume
, MAX_RAID_SERIAL_LEN
, "%s", name
);
6482 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
6483 dev
= get_imsm_dev(super
, i
);
6484 handle_missing(super
, dev
);
6486 super
->updates_pending
++;
6493 #endif /* MDASSEMBLE */
6495 static int is_gen_migration(struct imsm_dev
*dev
)
6500 if (!dev
->vol
.migr_state
)
6503 if (migr_type(dev
) == MIGR_GEN_MIGR
)
6509 static int is_rebuilding(struct imsm_dev
*dev
)
6511 struct imsm_map
*migr_map
;
6513 if (!dev
->vol
.migr_state
)
6516 if (migr_type(dev
) != MIGR_REBUILD
)
6519 migr_map
= get_imsm_map(dev
, MAP_1
);
6521 if (migr_map
->map_state
== IMSM_T_STATE_DEGRADED
)
6528 static int is_initializing(struct imsm_dev
*dev
)
6530 struct imsm_map
*migr_map
;
6532 if (!dev
->vol
.migr_state
)
6535 if (migr_type(dev
) != MIGR_INIT
)
6538 migr_map
= get_imsm_map(dev
, MAP_1
);
6540 if (migr_map
->map_state
== IMSM_T_STATE_UNINITIALIZED
)
6547 static void update_recovery_start(struct intel_super
*super
,
6548 struct imsm_dev
*dev
,
6549 struct mdinfo
*array
)
6551 struct mdinfo
*rebuild
= NULL
;
6555 if (!is_rebuilding(dev
))
6558 /* Find the rebuild target, but punt on the dual rebuild case */
6559 for (d
= array
->devs
; d
; d
= d
->next
)
6560 if (d
->recovery_start
== 0) {
6567 /* (?) none of the disks are marked with
6568 * IMSM_ORD_REBUILD, so assume they are missing and the
6569 * disk_ord_tbl was not correctly updated
6571 dprintf("%s: failed to locate out-of-sync disk\n", __func__
);
6575 units
= __le32_to_cpu(dev
->vol
.curr_migr_unit
);
6576 rebuild
->recovery_start
= units
* blocks_per_migr_unit(super
, dev
);
6580 static int recover_backup_imsm(struct supertype
*st
, struct mdinfo
*info
);
6583 static struct mdinfo
*container_content_imsm(struct supertype
*st
, char *subarray
)
6585 /* Given a container loaded by load_super_imsm_all,
6586 * extract information about all the arrays into
6588 * If 'subarray' is given, just extract info about that array.
6590 * For each imsm_dev create an mdinfo, fill it in,
6591 * then look for matching devices in super->disks
6592 * and create appropriate device mdinfo.
6594 struct intel_super
*super
= st
->sb
;
6595 struct imsm_super
*mpb
= super
->anchor
;
6596 struct mdinfo
*rest
= NULL
;
6600 int spare_disks
= 0;
6602 /* do not assemble arrays when not all attributes are supported */
6603 if (imsm_check_attributes(mpb
->attributes
) == 0) {
6605 fprintf(stderr
, Name
": Unsupported attributes in IMSM metadata."
6606 "Arrays activation is blocked.\n");
6609 /* check for bad blocks */
6610 if (imsm_bbm_log_size(super
->anchor
)) {
6611 fprintf(stderr
, Name
": BBM log found in IMSM metadata."
6612 "Arrays activation is blocked.\n");
6617 /* count spare devices, not used in maps
6619 for (d
= super
->disks
; d
; d
= d
->next
)
6623 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
6624 struct imsm_dev
*dev
;
6625 struct imsm_map
*map
;
6626 struct imsm_map
*map2
;
6627 struct mdinfo
*this;
6635 (i
!= strtoul(subarray
, &ep
, 10) || *ep
!= '\0'))
6638 dev
= get_imsm_dev(super
, i
);
6639 map
= get_imsm_map(dev
, MAP_0
);
6640 map2
= get_imsm_map(dev
, MAP_1
);
6642 /* do not publish arrays that are in the middle of an
6643 * unsupported migration
6645 if (dev
->vol
.migr_state
&&
6646 (migr_type(dev
) == MIGR_STATE_CHANGE
)) {
6647 fprintf(stderr
, Name
": cannot assemble volume '%.16s':"
6648 " unsupported migration in progress\n",
6652 /* do not publish arrays that are not support by controller's
6656 this = malloc(sizeof(*this));
6658 fprintf(stderr
, Name
": failed to allocate %zu bytes\n",
6663 super
->current_vol
= i
;
6664 getinfo_super_imsm_volume(st
, this, NULL
);
6667 chunk
= __le16_to_cpu(map
->blocks_per_strip
) >> 1;
6668 /* mdadm does not support all metadata features- set the bit in all arrays state */
6669 if (!validate_geometry_imsm_orom(super
,
6670 get_imsm_raid_level(map
), /* RAID level */
6671 imsm_level_to_layout(get_imsm_raid_level(map
)),
6672 map
->num_members
, /* raid disks */
6673 &chunk
, join_u32(dev
->size_low
, dev
->size_high
),
6675 fprintf(stderr
, Name
": IMSM RAID geometry validation"
6676 " failed. Array %s activation is blocked.\n",
6678 this->array
.state
|=
6679 (1<<MD_SB_BLOCK_CONTAINER_RESHAPE
) |
6680 (1<<MD_SB_BLOCK_VOLUME
);
6684 /* if array has bad blocks, set suitable bit in all arrays state */
6686 this->array
.state
|=
6687 (1<<MD_SB_BLOCK_CONTAINER_RESHAPE
) |
6688 (1<<MD_SB_BLOCK_VOLUME
);
6690 for (slot
= 0 ; slot
< map
->num_members
; slot
++) {
6691 unsigned long long recovery_start
;
6692 struct mdinfo
*info_d
;
6699 idx
= get_imsm_disk_idx(dev
, slot
, MAP_0
);
6700 ord
= get_imsm_ord_tbl_ent(dev
, slot
, MAP_X
);
6701 for (d
= super
->disks
; d
; d
= d
->next
)
6702 if (d
->index
== idx
)
6705 recovery_start
= MaxSector
;
6708 if (d
&& is_failed(&d
->disk
))
6710 if (ord
& IMSM_ORD_REBUILD
)
6714 * if we skip some disks the array will be assmebled degraded;
6715 * reset resync start to avoid a dirty-degraded
6716 * situation when performing the intial sync
6718 * FIXME handle dirty degraded
6720 if ((skip
|| recovery_start
== 0) && !dev
->vol
.dirty
)
6721 this->resync_start
= MaxSector
;
6725 info_d
= calloc(1, sizeof(*info_d
));
6727 fprintf(stderr
, Name
": failed to allocate disk"
6728 " for volume %.16s\n", dev
->volume
);
6729 info_d
= this->devs
;
6731 struct mdinfo
*d
= info_d
->next
;
6740 info_d
->next
= this->devs
;
6741 this->devs
= info_d
;
6743 info_d
->disk
.number
= d
->index
;
6744 info_d
->disk
.major
= d
->major
;
6745 info_d
->disk
.minor
= d
->minor
;
6746 info_d
->disk
.raid_disk
= slot
;
6747 info_d
->recovery_start
= recovery_start
;
6749 if (slot
< map2
->num_members
)
6750 info_d
->disk
.state
= (1 << MD_DISK_ACTIVE
);
6752 this->array
.spare_disks
++;
6754 if (slot
< map
->num_members
)
6755 info_d
->disk
.state
= (1 << MD_DISK_ACTIVE
);
6757 this->array
.spare_disks
++;
6759 if (info_d
->recovery_start
== MaxSector
)
6760 this->array
.working_disks
++;
6762 info_d
->events
= __le32_to_cpu(mpb
->generation_num
);
6763 info_d
->data_offset
= pba_of_lba0(map
);
6764 info_d
->component_size
= blocks_per_member(map
);
6766 /* now that the disk list is up-to-date fixup recovery_start */
6767 update_recovery_start(super
, dev
, this);
6768 this->array
.spare_disks
+= spare_disks
;
6771 /* check for reshape */
6772 if (this->reshape_active
== 1)
6773 recover_backup_imsm(st
, this);
6782 static __u8
imsm_check_degraded(struct intel_super
*super
, struct imsm_dev
*dev
,
6783 int failed
, int look_in_map
)
6785 struct imsm_map
*map
;
6787 map
= get_imsm_map(dev
, look_in_map
);
6790 return map
->map_state
== IMSM_T_STATE_UNINITIALIZED
?
6791 IMSM_T_STATE_UNINITIALIZED
: IMSM_T_STATE_NORMAL
;
6793 switch (get_imsm_raid_level(map
)) {
6795 return IMSM_T_STATE_FAILED
;
6798 if (failed
< map
->num_members
)
6799 return IMSM_T_STATE_DEGRADED
;
6801 return IMSM_T_STATE_FAILED
;
6806 * check to see if any mirrors have failed, otherwise we
6807 * are degraded. Even numbered slots are mirrored on
6811 /* gcc -Os complains that this is unused */
6812 int insync
= insync
;
6814 for (i
= 0; i
< map
->num_members
; i
++) {
6815 __u32 ord
= get_imsm_ord_tbl_ent(dev
, i
, MAP_X
);
6816 int idx
= ord_to_idx(ord
);
6817 struct imsm_disk
*disk
;
6819 /* reset the potential in-sync count on even-numbered
6820 * slots. num_copies is always 2 for imsm raid10
6825 disk
= get_imsm_disk(super
, idx
);
6826 if (!disk
|| is_failed(disk
) || ord
& IMSM_ORD_REBUILD
)
6829 /* no in-sync disks left in this mirror the
6833 return IMSM_T_STATE_FAILED
;
6836 return IMSM_T_STATE_DEGRADED
;
6840 return IMSM_T_STATE_DEGRADED
;
6842 return IMSM_T_STATE_FAILED
;
6848 return map
->map_state
;
6851 static int imsm_count_failed(struct intel_super
*super
, struct imsm_dev
*dev
,
6856 struct imsm_disk
*disk
;
6857 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
6858 struct imsm_map
*prev
= get_imsm_map(dev
, MAP_1
);
6859 struct imsm_map
*map_for_loop
;
6864 /* at the beginning of migration we set IMSM_ORD_REBUILD on
6865 * disks that are being rebuilt. New failures are recorded to
6866 * map[0]. So we look through all the disks we started with and
6867 * see if any failures are still present, or if any new ones
6871 if (prev
&& (map
->num_members
< prev
->num_members
))
6872 map_for_loop
= prev
;
6874 for (i
= 0; i
< map_for_loop
->num_members
; i
++) {
6876 /* when MAP_X is passed both maps failures are counted
6879 ((look_in_map
== MAP_1
) || (look_in_map
== MAP_X
)) &&
6880 (i
< prev
->num_members
)) {
6881 ord
= __le32_to_cpu(prev
->disk_ord_tbl
[i
]);
6882 idx_1
= ord_to_idx(ord
);
6884 disk
= get_imsm_disk(super
, idx_1
);
6885 if (!disk
|| is_failed(disk
) || ord
& IMSM_ORD_REBUILD
)
6888 if (((look_in_map
== MAP_0
) || (look_in_map
== MAP_X
)) &&
6889 (i
< map
->num_members
)) {
6890 ord
= __le32_to_cpu(map
->disk_ord_tbl
[i
]);
6891 idx
= ord_to_idx(ord
);
6894 disk
= get_imsm_disk(super
, idx
);
6895 if (!disk
|| is_failed(disk
) ||
6896 ord
& IMSM_ORD_REBUILD
)
6906 static int imsm_open_new(struct supertype
*c
, struct active_array
*a
,
6909 struct intel_super
*super
= c
->sb
;
6910 struct imsm_super
*mpb
= super
->anchor
;
6912 if (atoi(inst
) >= mpb
->num_raid_devs
) {
6913 fprintf(stderr
, "%s: subarry index %d, out of range\n",
6914 __func__
, atoi(inst
));
6918 dprintf("imsm: open_new %s\n", inst
);
6919 a
->info
.container_member
= atoi(inst
);
6923 static int is_resyncing(struct imsm_dev
*dev
)
6925 struct imsm_map
*migr_map
;
6927 if (!dev
->vol
.migr_state
)
6930 if (migr_type(dev
) == MIGR_INIT
||
6931 migr_type(dev
) == MIGR_REPAIR
)
6934 if (migr_type(dev
) == MIGR_GEN_MIGR
)
6937 migr_map
= get_imsm_map(dev
, MAP_1
);
6939 if ((migr_map
->map_state
== IMSM_T_STATE_NORMAL
) &&
6940 (dev
->vol
.migr_type
!= MIGR_GEN_MIGR
))
6946 /* return true if we recorded new information */
6947 static int mark_failure(struct imsm_dev
*dev
, struct imsm_disk
*disk
, int idx
)
6951 struct imsm_map
*map
;
6952 char buf
[MAX_RAID_SERIAL_LEN
+3];
6953 unsigned int len
, shift
= 0;
6955 /* new failures are always set in map[0] */
6956 map
= get_imsm_map(dev
, MAP_0
);
6958 slot
= get_imsm_disk_slot(map
, idx
);
6962 ord
= __le32_to_cpu(map
->disk_ord_tbl
[slot
]);
6963 if (is_failed(disk
) && (ord
& IMSM_ORD_REBUILD
))
6966 memcpy(buf
, disk
->serial
, MAX_RAID_SERIAL_LEN
);
6967 buf
[MAX_RAID_SERIAL_LEN
] = '\000';
6969 if ((len
= strlen(buf
)) >= MAX_RAID_SERIAL_LEN
)
6970 shift
= len
- MAX_RAID_SERIAL_LEN
+ 1;
6971 strncpy((char *)disk
->serial
, &buf
[shift
], MAX_RAID_SERIAL_LEN
);
6973 disk
->status
|= FAILED_DISK
;
6974 set_imsm_ord_tbl_ent(map
, slot
, idx
| IMSM_ORD_REBUILD
);
6975 /* mark failures in second map if second map exists and this disk
6977 * This is valid for migration, initialization and rebuild
6979 if (dev
->vol
.migr_state
) {
6980 struct imsm_map
*map2
= get_imsm_map(dev
, MAP_1
);
6981 int slot2
= get_imsm_disk_slot(map2
, idx
);
6983 if ((slot2
< map2
->num_members
) &&
6985 set_imsm_ord_tbl_ent(map2
, slot2
,
6986 idx
| IMSM_ORD_REBUILD
);
6988 if (map
->failed_disk_num
== 0xff)
6989 map
->failed_disk_num
= slot
;
6993 static void mark_missing(struct imsm_dev
*dev
, struct imsm_disk
*disk
, int idx
)
6995 mark_failure(dev
, disk
, idx
);
6997 if (disk
->scsi_id
== __cpu_to_le32(~(__u32
)0))
7000 disk
->scsi_id
= __cpu_to_le32(~(__u32
)0);
7001 memmove(&disk
->serial
[0], &disk
->serial
[1], MAX_RAID_SERIAL_LEN
- 1);
7004 static void handle_missing(struct intel_super
*super
, struct imsm_dev
*dev
)
7008 if (!super
->missing
)
7011 dprintf("imsm: mark missing\n");
7012 /* end process for initialization and rebuild only
7014 if (is_gen_migration(dev
) == 0) {
7018 failed
= imsm_count_failed(super
, dev
, MAP_0
);
7019 map_state
= imsm_check_degraded(super
, dev
, failed
, MAP_0
);
7021 end_migration(dev
, super
, map_state
);
7023 for (dl
= super
->missing
; dl
; dl
= dl
->next
)
7024 mark_missing(dev
, &dl
->disk
, dl
->index
);
7025 super
->updates_pending
++;
7028 static unsigned long long imsm_set_array_size(struct imsm_dev
*dev
,
7031 int used_disks
= imsm_num_data_members(dev
, MAP_0
);
7032 unsigned long long array_blocks
;
7033 struct imsm_map
*map
;
7035 if (used_disks
== 0) {
7036 /* when problems occures
7037 * return current array_blocks value
7039 array_blocks
= __le32_to_cpu(dev
->size_high
);
7040 array_blocks
= array_blocks
<< 32;
7041 array_blocks
+= __le32_to_cpu(dev
->size_low
);
7043 return array_blocks
;
7046 /* set array size in metadata
7048 if (new_size
<= 0) {
7049 /* OLCE size change is caused by added disks
7051 map
= get_imsm_map(dev
, MAP_0
);
7052 array_blocks
= blocks_per_member(map
) * used_disks
;
7054 /* Online Volume Size Change
7055 * Using available free space
7057 array_blocks
= new_size
;
7060 /* round array size down to closest MB
7062 array_blocks
= (array_blocks
>> SECT_PER_MB_SHIFT
) << SECT_PER_MB_SHIFT
;
7063 dev
->size_low
= __cpu_to_le32((__u32
)array_blocks
);
7064 dev
->size_high
= __cpu_to_le32((__u32
)(array_blocks
>> 32));
7066 return array_blocks
;
7069 static void imsm_set_disk(struct active_array
*a
, int n
, int state
);
7071 static void imsm_progress_container_reshape(struct intel_super
*super
)
7073 /* if no device has a migr_state, but some device has a
7074 * different number of members than the previous device, start
7075 * changing the number of devices in this device to match
7078 struct imsm_super
*mpb
= super
->anchor
;
7079 int prev_disks
= -1;
7083 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
7084 struct imsm_dev
*dev
= get_imsm_dev(super
, i
);
7085 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
7086 struct imsm_map
*map2
;
7087 int prev_num_members
;
7089 if (dev
->vol
.migr_state
)
7092 if (prev_disks
== -1)
7093 prev_disks
= map
->num_members
;
7094 if (prev_disks
== map
->num_members
)
7097 /* OK, this array needs to enter reshape mode.
7098 * i.e it needs a migr_state
7101 copy_map_size
= sizeof_imsm_map(map
);
7102 prev_num_members
= map
->num_members
;
7103 map
->num_members
= prev_disks
;
7104 dev
->vol
.migr_state
= 1;
7105 dev
->vol
.curr_migr_unit
= 0;
7106 set_migr_type(dev
, MIGR_GEN_MIGR
);
7107 for (i
= prev_num_members
;
7108 i
< map
->num_members
; i
++)
7109 set_imsm_ord_tbl_ent(map
, i
, i
);
7110 map2
= get_imsm_map(dev
, MAP_1
);
7111 /* Copy the current map */
7112 memcpy(map2
, map
, copy_map_size
);
7113 map2
->num_members
= prev_num_members
;
7115 imsm_set_array_size(dev
, -1);
7116 super
->clean_migration_record_by_mdmon
= 1;
7117 super
->updates_pending
++;
7121 /* Handle dirty -> clean transititions, resync and reshape. Degraded and rebuild
7122 * states are handled in imsm_set_disk() with one exception, when a
7123 * resync is stopped due to a new failure this routine will set the
7124 * 'degraded' state for the array.
7126 static int imsm_set_array_state(struct active_array
*a
, int consistent
)
7128 int inst
= a
->info
.container_member
;
7129 struct intel_super
*super
= a
->container
->sb
;
7130 struct imsm_dev
*dev
= get_imsm_dev(super
, inst
);
7131 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
7132 int failed
= imsm_count_failed(super
, dev
, MAP_0
);
7133 __u8 map_state
= imsm_check_degraded(super
, dev
, failed
, MAP_0
);
7134 __u32 blocks_per_unit
;
7136 if (dev
->vol
.migr_state
&&
7137 dev
->vol
.migr_type
== MIGR_GEN_MIGR
) {
7138 /* array state change is blocked due to reshape action
7140 * - abort the reshape (if last_checkpoint is 0 and action!= reshape)
7141 * - finish the reshape (if last_checkpoint is big and action != reshape)
7142 * - update curr_migr_unit
7144 if (a
->curr_action
== reshape
) {
7145 /* still reshaping, maybe update curr_migr_unit */
7146 goto mark_checkpoint
;
7148 if (a
->last_checkpoint
== 0 && a
->prev_action
== reshape
) {
7149 /* for some reason we aborted the reshape.
7151 * disable automatic metadata rollback
7152 * user action is required to recover process
7155 struct imsm_map
*map2
=
7156 get_imsm_map(dev
, MAP_1
);
7157 dev
->vol
.migr_state
= 0;
7158 set_migr_type(dev
, 0);
7159 dev
->vol
.curr_migr_unit
= 0;
7161 sizeof_imsm_map(map2
));
7162 super
->updates_pending
++;
7165 if (a
->last_checkpoint
>= a
->info
.component_size
) {
7166 unsigned long long array_blocks
;
7170 used_disks
= imsm_num_data_members(dev
, MAP_0
);
7171 if (used_disks
> 0) {
7173 blocks_per_member(map
) *
7175 /* round array size down to closest MB
7177 array_blocks
= (array_blocks
7178 >> SECT_PER_MB_SHIFT
)
7179 << SECT_PER_MB_SHIFT
;
7180 a
->info
.custom_array_size
= array_blocks
;
7181 /* encourage manager to update array
7185 a
->check_reshape
= 1;
7187 /* finalize online capacity expansion/reshape */
7188 for (mdi
= a
->info
.devs
; mdi
; mdi
= mdi
->next
)
7190 mdi
->disk
.raid_disk
,
7193 imsm_progress_container_reshape(super
);
7198 /* before we activate this array handle any missing disks */
7199 if (consistent
== 2)
7200 handle_missing(super
, dev
);
7202 if (consistent
== 2 &&
7203 (!is_resync_complete(&a
->info
) ||
7204 map_state
!= IMSM_T_STATE_NORMAL
||
7205 dev
->vol
.migr_state
))
7208 if (is_resync_complete(&a
->info
)) {
7209 /* complete intialization / resync,
7210 * recovery and interrupted recovery is completed in
7213 if (is_resyncing(dev
)) {
7214 dprintf("imsm: mark resync done\n");
7215 end_migration(dev
, super
, map_state
);
7216 super
->updates_pending
++;
7217 a
->last_checkpoint
= 0;
7219 } else if ((!is_resyncing(dev
) && !failed
) &&
7220 (imsm_reshape_blocks_arrays_changes(super
) == 0)) {
7221 /* mark the start of the init process if nothing is failed */
7222 dprintf("imsm: mark resync start\n");
7223 if (map
->map_state
== IMSM_T_STATE_UNINITIALIZED
)
7224 migrate(dev
, super
, IMSM_T_STATE_NORMAL
, MIGR_INIT
);
7226 migrate(dev
, super
, IMSM_T_STATE_NORMAL
, MIGR_REPAIR
);
7227 super
->updates_pending
++;
7231 /* skip checkpointing for general migration,
7232 * it is controlled in mdadm
7234 if (is_gen_migration(dev
))
7235 goto skip_mark_checkpoint
;
7237 /* check if we can update curr_migr_unit from resync_start, recovery_start */
7238 blocks_per_unit
= blocks_per_migr_unit(super
, dev
);
7239 if (blocks_per_unit
) {
7243 units
= a
->last_checkpoint
/ blocks_per_unit
;
7246 /* check that we did not overflow 32-bits, and that
7247 * curr_migr_unit needs updating
7249 if (units32
== units
&&
7251 __le32_to_cpu(dev
->vol
.curr_migr_unit
) != units32
) {
7252 dprintf("imsm: mark checkpoint (%u)\n", units32
);
7253 dev
->vol
.curr_migr_unit
= __cpu_to_le32(units32
);
7254 super
->updates_pending
++;
7258 skip_mark_checkpoint
:
7259 /* mark dirty / clean */
7260 if (dev
->vol
.dirty
!= !consistent
) {
7261 dprintf("imsm: mark '%s'\n", consistent
? "clean" : "dirty");
7266 super
->updates_pending
++;
7272 static void imsm_set_disk(struct active_array
*a
, int n
, int state
)
7274 int inst
= a
->info
.container_member
;
7275 struct intel_super
*super
= a
->container
->sb
;
7276 struct imsm_dev
*dev
= get_imsm_dev(super
, inst
);
7277 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
7278 struct imsm_disk
*disk
;
7280 int recovery_not_finished
= 0;
7285 if (n
> map
->num_members
)
7286 fprintf(stderr
, "imsm: set_disk %d out of range 0..%d\n",
7287 n
, map
->num_members
- 1);
7292 dprintf("imsm: set_disk %d:%x\n", n
, state
);
7294 ord
= get_imsm_ord_tbl_ent(dev
, n
, MAP_0
);
7295 disk
= get_imsm_disk(super
, ord_to_idx(ord
));
7297 /* check for new failures */
7298 if (state
& DS_FAULTY
) {
7299 if (mark_failure(dev
, disk
, ord_to_idx(ord
)))
7300 super
->updates_pending
++;
7303 /* check if in_sync */
7304 if (state
& DS_INSYNC
&& ord
& IMSM_ORD_REBUILD
&& is_rebuilding(dev
)) {
7305 struct imsm_map
*migr_map
= get_imsm_map(dev
, MAP_1
);
7307 set_imsm_ord_tbl_ent(migr_map
, n
, ord_to_idx(ord
));
7308 super
->updates_pending
++;
7311 failed
= imsm_count_failed(super
, dev
, MAP_0
);
7312 map_state
= imsm_check_degraded(super
, dev
, failed
, MAP_0
);
7314 /* check if recovery complete, newly degraded, or failed */
7315 dprintf("imsm: Detected transition to state ");
7316 switch (map_state
) {
7317 case IMSM_T_STATE_NORMAL
: /* transition to normal state */
7318 dprintf("normal: ");
7319 if (is_rebuilding(dev
)) {
7320 dprintf("while rebuilding");
7321 /* check if recovery is really finished */
7322 for (mdi
= a
->info
.devs
; mdi
; mdi
= mdi
->next
)
7323 if (mdi
->recovery_start
!= MaxSector
) {
7324 recovery_not_finished
= 1;
7327 if (recovery_not_finished
) {
7328 dprintf("\nimsm: Rebuild has not finished yet, "
7329 "state not changed");
7330 if (a
->last_checkpoint
< mdi
->recovery_start
) {
7331 a
->last_checkpoint
= mdi
->recovery_start
;
7332 super
->updates_pending
++;
7336 end_migration(dev
, super
, map_state
);
7337 map
= get_imsm_map(dev
, MAP_0
);
7338 map
->failed_disk_num
= ~0;
7339 super
->updates_pending
++;
7340 a
->last_checkpoint
= 0;
7343 if (is_gen_migration(dev
)) {
7344 dprintf("while general migration");
7345 if (a
->last_checkpoint
>= a
->info
.component_size
)
7346 end_migration(dev
, super
, map_state
);
7348 map
->map_state
= map_state
;
7349 map
= get_imsm_map(dev
, MAP_0
);
7350 map
->failed_disk_num
= ~0;
7351 super
->updates_pending
++;
7355 case IMSM_T_STATE_DEGRADED
: /* transition to degraded state */
7356 dprintf("degraded: ");
7357 if ((map
->map_state
!= map_state
) &&
7358 !dev
->vol
.migr_state
) {
7359 dprintf("mark degraded");
7360 map
->map_state
= map_state
;
7361 super
->updates_pending
++;
7362 a
->last_checkpoint
= 0;
7365 if (is_rebuilding(dev
)) {
7366 dprintf("while rebuilding.");
7367 if (map
->map_state
!= map_state
) {
7368 dprintf(" Map state change");
7369 end_migration(dev
, super
, map_state
);
7370 super
->updates_pending
++;
7374 if (is_gen_migration(dev
)) {
7375 dprintf("while general migration");
7376 if (a
->last_checkpoint
>= a
->info
.component_size
)
7377 end_migration(dev
, super
, map_state
);
7379 map
->map_state
= map_state
;
7380 manage_second_map(super
, dev
);
7382 super
->updates_pending
++;
7385 if (is_initializing(dev
)) {
7386 dprintf("while initialization.");
7387 map
->map_state
= map_state
;
7388 super
->updates_pending
++;
7392 case IMSM_T_STATE_FAILED
: /* transition to failed state */
7393 dprintf("failed: ");
7394 if (is_gen_migration(dev
)) {
7395 dprintf("while general migration");
7396 map
->map_state
= map_state
;
7397 super
->updates_pending
++;
7400 if (map
->map_state
!= map_state
) {
7401 dprintf("mark failed");
7402 end_migration(dev
, super
, map_state
);
7403 super
->updates_pending
++;
7404 a
->last_checkpoint
= 0;
7409 dprintf("state %i\n", map_state
);
7415 static int store_imsm_mpb(int fd
, struct imsm_super
*mpb
)
7418 __u32 mpb_size
= __le32_to_cpu(mpb
->mpb_size
);
7419 unsigned long long dsize
;
7420 unsigned long long sectors
;
7422 get_dev_size(fd
, NULL
, &dsize
);
7424 if (mpb_size
> 512) {
7425 /* -1 to account for anchor */
7426 sectors
= mpb_sectors(mpb
) - 1;
7428 /* write the extended mpb to the sectors preceeding the anchor */
7429 if (lseek64(fd
, dsize
- (512 * (2 + sectors
)), SEEK_SET
) < 0)
7432 if ((unsigned long long)write(fd
, buf
+ 512, 512 * sectors
)
7437 /* first block is stored on second to last sector of the disk */
7438 if (lseek64(fd
, dsize
- (512 * 2), SEEK_SET
) < 0)
7441 if (write(fd
, buf
, 512) != 512)
7447 static void imsm_sync_metadata(struct supertype
*container
)
7449 struct intel_super
*super
= container
->sb
;
7451 dprintf("sync metadata: %d\n", super
->updates_pending
);
7452 if (!super
->updates_pending
)
7455 write_super_imsm(container
, 0);
7457 super
->updates_pending
= 0;
7460 static struct dl
*imsm_readd(struct intel_super
*super
, int idx
, struct active_array
*a
)
7462 struct imsm_dev
*dev
= get_imsm_dev(super
, a
->info
.container_member
);
7463 int i
= get_imsm_disk_idx(dev
, idx
, MAP_X
);
7466 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
7470 if (dl
&& is_failed(&dl
->disk
))
7474 dprintf("%s: found %x:%x\n", __func__
, dl
->major
, dl
->minor
);
7479 static struct dl
*imsm_add_spare(struct intel_super
*super
, int slot
,
7480 struct active_array
*a
, int activate_new
,
7481 struct mdinfo
*additional_test_list
)
7483 struct imsm_dev
*dev
= get_imsm_dev(super
, a
->info
.container_member
);
7484 int idx
= get_imsm_disk_idx(dev
, slot
, MAP_X
);
7485 struct imsm_super
*mpb
= super
->anchor
;
7486 struct imsm_map
*map
;
7487 unsigned long long pos
;
7492 __u32 array_start
= 0;
7493 __u32 array_end
= 0;
7495 struct mdinfo
*test_list
;
7497 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
7498 /* If in this array, skip */
7499 for (d
= a
->info
.devs
; d
; d
= d
->next
)
7500 if (d
->state_fd
>= 0 &&
7501 d
->disk
.major
== dl
->major
&&
7502 d
->disk
.minor
== dl
->minor
) {
7503 dprintf("%x:%x already in array\n",
7504 dl
->major
, dl
->minor
);
7509 test_list
= additional_test_list
;
7511 if (test_list
->disk
.major
== dl
->major
&&
7512 test_list
->disk
.minor
== dl
->minor
) {
7513 dprintf("%x:%x already in additional test list\n",
7514 dl
->major
, dl
->minor
);
7517 test_list
= test_list
->next
;
7522 /* skip in use or failed drives */
7523 if (is_failed(&dl
->disk
) || idx
== dl
->index
||
7525 dprintf("%x:%x status (failed: %d index: %d)\n",
7526 dl
->major
, dl
->minor
, is_failed(&dl
->disk
), idx
);
7530 /* skip pure spares when we are looking for partially
7531 * assimilated drives
7533 if (dl
->index
== -1 && !activate_new
)
7536 /* Does this unused device have the requisite free space?
7537 * It needs to be able to cover all member volumes
7539 ex
= get_extents(super
, dl
);
7541 dprintf("cannot get extents\n");
7544 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
7545 dev
= get_imsm_dev(super
, i
);
7546 map
= get_imsm_map(dev
, MAP_0
);
7548 /* check if this disk is already a member of
7551 if (get_imsm_disk_slot(map
, dl
->index
) >= 0)
7557 array_start
= pba_of_lba0(map
);
7558 array_end
= array_start
+
7559 blocks_per_member(map
) - 1;
7562 /* check that we can start at pba_of_lba0 with
7563 * blocks_per_member of space
7565 if (array_start
>= pos
&& array_end
< ex
[j
].start
) {
7569 pos
= ex
[j
].start
+ ex
[j
].size
;
7571 } while (ex
[j
-1].size
);
7578 if (i
< mpb
->num_raid_devs
) {
7579 dprintf("%x:%x does not have %u to %u available\n",
7580 dl
->major
, dl
->minor
, array_start
, array_end
);
7591 static int imsm_rebuild_allowed(struct supertype
*cont
, int dev_idx
, int failed
)
7593 struct imsm_dev
*dev2
;
7594 struct imsm_map
*map
;
7600 dev2
= get_imsm_dev(cont
->sb
, dev_idx
);
7602 state
= imsm_check_degraded(cont
->sb
, dev2
, failed
, MAP_0
);
7603 if (state
== IMSM_T_STATE_FAILED
) {
7604 map
= get_imsm_map(dev2
, MAP_0
);
7607 for (slot
= 0; slot
< map
->num_members
; slot
++) {
7609 * Check if failed disks are deleted from intel
7610 * disk list or are marked to be deleted
7612 idx
= get_imsm_disk_idx(dev2
, slot
, MAP_X
);
7613 idisk
= get_imsm_dl_disk(cont
->sb
, idx
);
7615 * Do not rebuild the array if failed disks
7616 * from failed sub-array are not removed from
7620 is_failed(&idisk
->disk
) &&
7621 (idisk
->action
!= DISK_REMOVE
))
7629 static struct mdinfo
*imsm_activate_spare(struct active_array
*a
,
7630 struct metadata_update
**updates
)
7633 * Find a device with unused free space and use it to replace a
7634 * failed/vacant region in an array. We replace failed regions one a
7635 * array at a time. The result is that a new spare disk will be added
7636 * to the first failed array and after the monitor has finished
7637 * propagating failures the remainder will be consumed.
7639 * FIXME add a capability for mdmon to request spares from another
7643 struct intel_super
*super
= a
->container
->sb
;
7644 int inst
= a
->info
.container_member
;
7645 struct imsm_dev
*dev
= get_imsm_dev(super
, inst
);
7646 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
7647 int failed
= a
->info
.array
.raid_disks
;
7648 struct mdinfo
*rv
= NULL
;
7651 struct metadata_update
*mu
;
7653 struct imsm_update_activate_spare
*u
;
7658 for (d
= a
->info
.devs
; d
; d
= d
->next
) {
7659 if ((d
->curr_state
& DS_FAULTY
) &&
7661 /* wait for Removal to happen */
7663 if (d
->state_fd
>= 0)
7667 dprintf("imsm: activate spare: inst=%d failed=%d (%d) level=%d\n",
7668 inst
, failed
, a
->info
.array
.raid_disks
, a
->info
.array
.level
);
7670 if (imsm_reshape_blocks_arrays_changes(super
))
7673 /* Cannot activate another spare if rebuild is in progress already
7675 if (is_rebuilding(dev
)) {
7676 dprintf("imsm: No spare activation allowed. "
7677 "Rebuild in progress already.\n");
7681 if (a
->info
.array
.level
== 4)
7682 /* No repair for takeovered array
7683 * imsm doesn't support raid4
7687 if (imsm_check_degraded(super
, dev
, failed
, MAP_0
) !=
7688 IMSM_T_STATE_DEGRADED
)
7692 * If there are any failed disks check state of the other volume.
7693 * Block rebuild if the another one is failed until failed disks
7694 * are removed from container.
7697 dprintf("found failed disks in %.*s, check if there another"
7698 "failed sub-array.\n",
7699 MAX_RAID_SERIAL_LEN
, dev
->volume
);
7700 /* check if states of the other volumes allow for rebuild */
7701 for (i
= 0; i
< super
->anchor
->num_raid_devs
; i
++) {
7703 allowed
= imsm_rebuild_allowed(a
->container
,
7711 /* For each slot, if it is not working, find a spare */
7712 for (i
= 0; i
< a
->info
.array
.raid_disks
; i
++) {
7713 for (d
= a
->info
.devs
; d
; d
= d
->next
)
7714 if (d
->disk
.raid_disk
== i
)
7716 dprintf("found %d: %p %x\n", i
, d
, d
?d
->curr_state
:0);
7717 if (d
&& (d
->state_fd
>= 0))
7721 * OK, this device needs recovery. Try to re-add the
7722 * previous occupant of this slot, if this fails see if
7723 * we can continue the assimilation of a spare that was
7724 * partially assimilated, finally try to activate a new
7727 dl
= imsm_readd(super
, i
, a
);
7729 dl
= imsm_add_spare(super
, i
, a
, 0, rv
);
7731 dl
= imsm_add_spare(super
, i
, a
, 1, rv
);
7735 /* found a usable disk with enough space */
7736 di
= malloc(sizeof(*di
));
7739 memset(di
, 0, sizeof(*di
));
7741 /* dl->index will be -1 in the case we are activating a
7742 * pristine spare. imsm_process_update() will create a
7743 * new index in this case. Once a disk is found to be
7744 * failed in all member arrays it is kicked from the
7747 di
->disk
.number
= dl
->index
;
7749 /* (ab)use di->devs to store a pointer to the device
7752 di
->devs
= (struct mdinfo
*) dl
;
7754 di
->disk
.raid_disk
= i
;
7755 di
->disk
.major
= dl
->major
;
7756 di
->disk
.minor
= dl
->minor
;
7758 di
->recovery_start
= 0;
7759 di
->data_offset
= pba_of_lba0(map
);
7760 di
->component_size
= a
->info
.component_size
;
7761 di
->container_member
= inst
;
7762 super
->random
= random32();
7766 dprintf("%x:%x to be %d at %llu\n", dl
->major
, dl
->minor
,
7767 i
, di
->data_offset
);
7771 /* No spares found */
7773 /* Now 'rv' has a list of devices to return.
7774 * Create a metadata_update record to update the
7775 * disk_ord_tbl for the array
7777 mu
= malloc(sizeof(*mu
));
7779 mu
->buf
= malloc(sizeof(struct imsm_update_activate_spare
) * num_spares
);
7780 if (mu
->buf
== NULL
) {
7787 struct mdinfo
*n
= rv
->next
;
7796 mu
->space_list
= NULL
;
7797 mu
->len
= sizeof(struct imsm_update_activate_spare
) * num_spares
;
7798 mu
->next
= *updates
;
7799 u
= (struct imsm_update_activate_spare
*) mu
->buf
;
7801 for (di
= rv
; di
; di
= di
->next
) {
7802 u
->type
= update_activate_spare
;
7803 u
->dl
= (struct dl
*) di
->devs
;
7805 u
->slot
= di
->disk
.raid_disk
;
7816 static int disks_overlap(struct intel_super
*super
, int idx
, struct imsm_update_create_array
*u
)
7818 struct imsm_dev
*dev
= get_imsm_dev(super
, idx
);
7819 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
7820 struct imsm_map
*new_map
= get_imsm_map(&u
->dev
, MAP_0
);
7821 struct disk_info
*inf
= get_disk_info(u
);
7822 struct imsm_disk
*disk
;
7826 for (i
= 0; i
< map
->num_members
; i
++) {
7827 disk
= get_imsm_disk(super
, get_imsm_disk_idx(dev
, i
, MAP_X
));
7828 for (j
= 0; j
< new_map
->num_members
; j
++)
7829 if (serialcmp(disk
->serial
, inf
[j
].serial
) == 0)
7837 static struct dl
*get_disk_super(struct intel_super
*super
, int major
, int minor
)
7839 struct dl
*dl
= NULL
;
7840 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
7841 if ((dl
->major
== major
) && (dl
->minor
== minor
))
7846 static int remove_disk_super(struct intel_super
*super
, int major
, int minor
)
7848 struct dl
*prev
= NULL
;
7852 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
7853 if ((dl
->major
== major
) && (dl
->minor
== minor
)) {
7856 prev
->next
= dl
->next
;
7858 super
->disks
= dl
->next
;
7860 __free_imsm_disk(dl
);
7861 dprintf("%s: removed %x:%x\n",
7862 __func__
, major
, minor
);
7870 static void imsm_delete(struct intel_super
*super
, struct dl
**dlp
, unsigned index
);
7872 static int add_remove_disk_update(struct intel_super
*super
)
7874 int check_degraded
= 0;
7875 struct dl
*disk
= NULL
;
7876 /* add/remove some spares to/from the metadata/contrainer */
7877 while (super
->disk_mgmt_list
) {
7878 struct dl
*disk_cfg
;
7880 disk_cfg
= super
->disk_mgmt_list
;
7881 super
->disk_mgmt_list
= disk_cfg
->next
;
7882 disk_cfg
->next
= NULL
;
7884 if (disk_cfg
->action
== DISK_ADD
) {
7885 disk_cfg
->next
= super
->disks
;
7886 super
->disks
= disk_cfg
;
7888 dprintf("%s: added %x:%x\n",
7889 __func__
, disk_cfg
->major
,
7891 } else if (disk_cfg
->action
== DISK_REMOVE
) {
7892 dprintf("Disk remove action processed: %x.%x\n",
7893 disk_cfg
->major
, disk_cfg
->minor
);
7894 disk
= get_disk_super(super
,
7898 /* store action status */
7899 disk
->action
= DISK_REMOVE
;
7900 /* remove spare disks only */
7901 if (disk
->index
== -1) {
7902 remove_disk_super(super
,
7907 /* release allocate disk structure */
7908 __free_imsm_disk(disk_cfg
);
7911 return check_degraded
;
7915 static int apply_reshape_migration_update(struct imsm_update_reshape_migration
*u
,
7916 struct intel_super
*super
,
7919 struct intel_dev
*id
;
7920 void **tofree
= NULL
;
7923 dprintf("apply_reshape_migration_update()\n");
7924 if ((u
->subdev
< 0) ||
7926 dprintf("imsm: Error: Wrong subdev: %i\n", u
->subdev
);
7929 if ((space_list
== NULL
) || (*space_list
== NULL
)) {
7930 dprintf("imsm: Error: Memory is not allocated\n");
7934 for (id
= super
->devlist
; id
; id
= id
->next
) {
7935 if (id
->index
== (unsigned)u
->subdev
) {
7936 struct imsm_dev
*dev
= get_imsm_dev(super
, u
->subdev
);
7937 struct imsm_map
*map
;
7938 struct imsm_dev
*new_dev
=
7939 (struct imsm_dev
*)*space_list
;
7940 struct imsm_map
*migr_map
= get_imsm_map(dev
, MAP_1
);
7942 struct dl
*new_disk
;
7944 if (new_dev
== NULL
)
7946 *space_list
= **space_list
;
7947 memcpy(new_dev
, dev
, sizeof_imsm_dev(dev
, 0));
7948 map
= get_imsm_map(new_dev
, MAP_0
);
7950 dprintf("imsm: Error: migration in progress");
7954 to_state
= map
->map_state
;
7955 if ((u
->new_level
== 5) && (map
->raid_level
== 0)) {
7957 /* this should not happen */
7958 if (u
->new_disks
[0] < 0) {
7959 map
->failed_disk_num
=
7960 map
->num_members
- 1;
7961 to_state
= IMSM_T_STATE_DEGRADED
;
7963 to_state
= IMSM_T_STATE_NORMAL
;
7965 migrate(new_dev
, super
, to_state
, MIGR_GEN_MIGR
);
7966 if (u
->new_level
> -1)
7967 map
->raid_level
= u
->new_level
;
7968 migr_map
= get_imsm_map(new_dev
, MAP_1
);
7969 if ((u
->new_level
== 5) &&
7970 (migr_map
->raid_level
== 0)) {
7971 int ord
= map
->num_members
- 1;
7972 migr_map
->num_members
--;
7973 if (u
->new_disks
[0] < 0)
7974 ord
|= IMSM_ORD_REBUILD
;
7975 set_imsm_ord_tbl_ent(map
,
7976 map
->num_members
- 1,
7980 tofree
= (void **)dev
;
7982 /* update chunk size
7984 if (u
->new_chunksize
> 0)
7985 map
->blocks_per_strip
=
7986 __cpu_to_le16(u
->new_chunksize
* 2);
7990 if ((u
->new_level
!= 5) ||
7991 (migr_map
->raid_level
!= 0) ||
7992 (migr_map
->raid_level
== map
->raid_level
))
7995 if (u
->new_disks
[0] >= 0) {
7998 new_disk
= get_disk_super(super
,
7999 major(u
->new_disks
[0]),
8000 minor(u
->new_disks
[0]));
8001 dprintf("imsm: new disk for reshape is: %i:%i "
8002 "(%p, index = %i)\n",
8003 major(u
->new_disks
[0]),
8004 minor(u
->new_disks
[0]),
8005 new_disk
, new_disk
->index
);
8006 if (new_disk
== NULL
)
8007 goto error_disk_add
;
8009 new_disk
->index
= map
->num_members
- 1;
8010 /* slot to fill in autolayout
8012 new_disk
->raiddisk
= new_disk
->index
;
8013 new_disk
->disk
.status
|= CONFIGURED_DISK
;
8014 new_disk
->disk
.status
&= ~SPARE_DISK
;
8016 goto error_disk_add
;
8019 *tofree
= *space_list
;
8020 /* calculate new size
8022 imsm_set_array_size(new_dev
, -1);
8029 *space_list
= tofree
;
8033 dprintf("Error: imsm: Cannot find disk.\n");
8037 static int apply_size_change_update(struct imsm_update_size_change
*u
,
8038 struct intel_super
*super
)
8040 struct intel_dev
*id
;
8043 dprintf("apply_size_change_update()\n");
8044 if ((u
->subdev
< 0) ||
8046 dprintf("imsm: Error: Wrong subdev: %i\n", u
->subdev
);
8050 for (id
= super
->devlist
; id
; id
= id
->next
) {
8051 if (id
->index
== (unsigned)u
->subdev
) {
8052 struct imsm_dev
*dev
= get_imsm_dev(super
, u
->subdev
);
8053 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
8054 int used_disks
= imsm_num_data_members(dev
, MAP_0
);
8055 unsigned long long blocks_per_member
;
8057 /* calculate new size
8059 blocks_per_member
= u
->new_size
/ used_disks
;
8060 dprintf("imsm: apply_size_change_update(size: %llu, "
8061 "blocks per member: %llu)\n",
8062 u
->new_size
, blocks_per_member
);
8063 set_blocks_per_member(map
, blocks_per_member
);
8064 imsm_set_array_size(dev
, u
->new_size
);
8075 static int apply_update_activate_spare(struct imsm_update_activate_spare
*u
,
8076 struct intel_super
*super
,
8077 struct active_array
*active_array
)
8079 struct imsm_super
*mpb
= super
->anchor
;
8080 struct imsm_dev
*dev
= get_imsm_dev(super
, u
->array
);
8081 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
8082 struct imsm_map
*migr_map
;
8083 struct active_array
*a
;
8084 struct imsm_disk
*disk
;
8091 int second_map_created
= 0;
8093 for (; u
; u
= u
->next
) {
8094 victim
= get_imsm_disk_idx(dev
, u
->slot
, MAP_X
);
8099 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
8104 fprintf(stderr
, "error: imsm_activate_spare passed "
8105 "an unknown disk (index: %d)\n",
8110 /* count failures (excluding rebuilds and the victim)
8111 * to determine map[0] state
8114 for (i
= 0; i
< map
->num_members
; i
++) {
8117 disk
= get_imsm_disk(super
,
8118 get_imsm_disk_idx(dev
, i
, MAP_X
));
8119 if (!disk
|| is_failed(disk
))
8123 /* adding a pristine spare, assign a new index */
8124 if (dl
->index
< 0) {
8125 dl
->index
= super
->anchor
->num_disks
;
8126 super
->anchor
->num_disks
++;
8129 disk
->status
|= CONFIGURED_DISK
;
8130 disk
->status
&= ~SPARE_DISK
;
8133 to_state
= imsm_check_degraded(super
, dev
, failed
, MAP_0
);
8134 if (!second_map_created
) {
8135 second_map_created
= 1;
8136 map
->map_state
= IMSM_T_STATE_DEGRADED
;
8137 migrate(dev
, super
, to_state
, MIGR_REBUILD
);
8139 map
->map_state
= to_state
;
8140 migr_map
= get_imsm_map(dev
, MAP_1
);
8141 set_imsm_ord_tbl_ent(map
, u
->slot
, dl
->index
);
8142 set_imsm_ord_tbl_ent(migr_map
, u
->slot
,
8143 dl
->index
| IMSM_ORD_REBUILD
);
8145 /* update the family_num to mark a new container
8146 * generation, being careful to record the existing
8147 * family_num in orig_family_num to clean up after
8148 * earlier mdadm versions that neglected to set it.
8150 if (mpb
->orig_family_num
== 0)
8151 mpb
->orig_family_num
= mpb
->family_num
;
8152 mpb
->family_num
+= super
->random
;
8154 /* count arrays using the victim in the metadata */
8156 for (a
= active_array
; a
; a
= a
->next
) {
8157 dev
= get_imsm_dev(super
, a
->info
.container_member
);
8158 map
= get_imsm_map(dev
, MAP_0
);
8160 if (get_imsm_disk_slot(map
, victim
) >= 0)
8164 /* delete the victim if it is no longer being
8170 /* We know that 'manager' isn't touching anything,
8171 * so it is safe to delete
8173 for (dlp
= &super
->disks
; *dlp
; dlp
= &(*dlp
)->next
)
8174 if ((*dlp
)->index
== victim
)
8177 /* victim may be on the missing list */
8179 for (dlp
= &super
->missing
; *dlp
;
8180 dlp
= &(*dlp
)->next
)
8181 if ((*dlp
)->index
== victim
)
8183 imsm_delete(super
, dlp
, victim
);
8190 static int apply_reshape_container_disks_update(struct imsm_update_reshape
*u
,
8191 struct intel_super
*super
,
8194 struct dl
*new_disk
;
8195 struct intel_dev
*id
;
8197 int delta_disks
= u
->new_raid_disks
- u
->old_raid_disks
;
8198 int disk_count
= u
->old_raid_disks
;
8199 void **tofree
= NULL
;
8200 int devices_to_reshape
= 1;
8201 struct imsm_super
*mpb
= super
->anchor
;
8203 unsigned int dev_id
;
8205 dprintf("imsm: apply_reshape_container_disks_update()\n");
8207 /* enable spares to use in array */
8208 for (i
= 0; i
< delta_disks
; i
++) {
8209 new_disk
= get_disk_super(super
,
8210 major(u
->new_disks
[i
]),
8211 minor(u
->new_disks
[i
]));
8212 dprintf("imsm: new disk for reshape is: %i:%i "
8213 "(%p, index = %i)\n",
8214 major(u
->new_disks
[i
]), minor(u
->new_disks
[i
]),
8215 new_disk
, new_disk
->index
);
8216 if ((new_disk
== NULL
) ||
8217 ((new_disk
->index
>= 0) &&
8218 (new_disk
->index
< u
->old_raid_disks
)))
8219 goto update_reshape_exit
;
8220 new_disk
->index
= disk_count
++;
8221 /* slot to fill in autolayout
8223 new_disk
->raiddisk
= new_disk
->index
;
8224 new_disk
->disk
.status
|=
8226 new_disk
->disk
.status
&= ~SPARE_DISK
;
8229 dprintf("imsm: volume set mpb->num_raid_devs = %i\n",
8230 mpb
->num_raid_devs
);
8231 /* manage changes in volume
8233 for (dev_id
= 0; dev_id
< mpb
->num_raid_devs
; dev_id
++) {
8234 void **sp
= *space_list
;
8235 struct imsm_dev
*newdev
;
8236 struct imsm_map
*newmap
, *oldmap
;
8238 for (id
= super
->devlist
; id
; id
= id
->next
) {
8239 if (id
->index
== dev_id
)
8248 /* Copy the dev, but not (all of) the map */
8249 memcpy(newdev
, id
->dev
, sizeof(*newdev
));
8250 oldmap
= get_imsm_map(id
->dev
, MAP_0
);
8251 newmap
= get_imsm_map(newdev
, MAP_0
);
8252 /* Copy the current map */
8253 memcpy(newmap
, oldmap
, sizeof_imsm_map(oldmap
));
8254 /* update one device only
8256 if (devices_to_reshape
) {
8257 dprintf("imsm: modifying subdev: %i\n",
8259 devices_to_reshape
--;
8260 newdev
->vol
.migr_state
= 1;
8261 newdev
->vol
.curr_migr_unit
= 0;
8262 set_migr_type(newdev
, MIGR_GEN_MIGR
);
8263 newmap
->num_members
= u
->new_raid_disks
;
8264 for (i
= 0; i
< delta_disks
; i
++) {
8265 set_imsm_ord_tbl_ent(newmap
,
8266 u
->old_raid_disks
+ i
,
8267 u
->old_raid_disks
+ i
);
8269 /* New map is correct, now need to save old map
8271 newmap
= get_imsm_map(newdev
, MAP_1
);
8272 memcpy(newmap
, oldmap
, sizeof_imsm_map(oldmap
));
8274 imsm_set_array_size(newdev
, -1);
8277 sp
= (void **)id
->dev
;
8282 /* Clear migration record */
8283 memset(super
->migr_rec
, 0, sizeof(struct migr_record
));
8286 *space_list
= tofree
;
8289 update_reshape_exit
:
8294 static int apply_takeover_update(struct imsm_update_takeover
*u
,
8295 struct intel_super
*super
,
8298 struct imsm_dev
*dev
= NULL
;
8299 struct intel_dev
*dv
;
8300 struct imsm_dev
*dev_new
;
8301 struct imsm_map
*map
;
8305 for (dv
= super
->devlist
; dv
; dv
= dv
->next
)
8306 if (dv
->index
== (unsigned int)u
->subarray
) {
8314 map
= get_imsm_map(dev
, MAP_0
);
8316 if (u
->direction
== R10_TO_R0
) {
8317 /* Number of failed disks must be half of initial disk number */
8318 if (imsm_count_failed(super
, dev
, MAP_0
) !=
8319 (map
->num_members
/ 2))
8322 /* iterate through devices to mark removed disks as spare */
8323 for (dm
= super
->disks
; dm
; dm
= dm
->next
) {
8324 if (dm
->disk
.status
& FAILED_DISK
) {
8325 int idx
= dm
->index
;
8326 /* update indexes on the disk list */
8327 /* FIXME this loop-with-the-loop looks wrong, I'm not convinced
8328 the index values will end up being correct.... NB */
8329 for (du
= super
->disks
; du
; du
= du
->next
)
8330 if (du
->index
> idx
)
8332 /* mark as spare disk */
8337 map
->num_members
= map
->num_members
/ 2;
8338 map
->map_state
= IMSM_T_STATE_NORMAL
;
8339 map
->num_domains
= 1;
8340 map
->raid_level
= 0;
8341 map
->failed_disk_num
= -1;
8344 if (u
->direction
== R0_TO_R10
) {
8346 /* update slots in current disk list */
8347 for (dm
= super
->disks
; dm
; dm
= dm
->next
) {
8351 /* create new *missing* disks */
8352 for (i
= 0; i
< map
->num_members
; i
++) {
8353 space
= *space_list
;
8356 *space_list
= *space
;
8358 memcpy(du
, super
->disks
, sizeof(*du
));
8362 du
->index
= (i
* 2) + 1;
8363 sprintf((char *)du
->disk
.serial
,
8364 " MISSING_%d", du
->index
);
8365 sprintf((char *)du
->serial
,
8366 "MISSING_%d", du
->index
);
8367 du
->next
= super
->missing
;
8368 super
->missing
= du
;
8370 /* create new dev and map */
8371 space
= *space_list
;
8374 *space_list
= *space
;
8375 dev_new
= (void *)space
;
8376 memcpy(dev_new
, dev
, sizeof(*dev
));
8377 /* update new map */
8378 map
= get_imsm_map(dev_new
, MAP_0
);
8379 map
->num_members
= map
->num_members
* 2;
8380 map
->map_state
= IMSM_T_STATE_DEGRADED
;
8381 map
->num_domains
= 2;
8382 map
->raid_level
= 1;
8383 /* replace dev<->dev_new */
8386 /* update disk order table */
8387 for (du
= super
->disks
; du
; du
= du
->next
)
8389 set_imsm_ord_tbl_ent(map
, du
->index
, du
->index
);
8390 for (du
= super
->missing
; du
; du
= du
->next
)
8391 if (du
->index
>= 0) {
8392 set_imsm_ord_tbl_ent(map
, du
->index
, du
->index
);
8393 mark_missing(dv
->dev
, &du
->disk
, du
->index
);
8399 static void imsm_process_update(struct supertype
*st
,
8400 struct metadata_update
*update
)
8403 * crack open the metadata_update envelope to find the update record
8404 * update can be one of:
8405 * update_reshape_container_disks - all the arrays in the container
8406 * are being reshaped to have more devices. We need to mark
8407 * the arrays for general migration and convert selected spares
8408 * into active devices.
8409 * update_activate_spare - a spare device has replaced a failed
8410 * device in an array, update the disk_ord_tbl. If this disk is
8411 * present in all member arrays then also clear the SPARE_DISK
8413 * update_create_array
8415 * update_rename_array
8416 * update_add_remove_disk
8418 struct intel_super
*super
= st
->sb
;
8419 struct imsm_super
*mpb
;
8420 enum imsm_update_type type
= *(enum imsm_update_type
*) update
->buf
;
8422 /* update requires a larger buf but the allocation failed */
8423 if (super
->next_len
&& !super
->next_buf
) {
8424 super
->next_len
= 0;
8428 if (super
->next_buf
) {
8429 memcpy(super
->next_buf
, super
->buf
, super
->len
);
8431 super
->len
= super
->next_len
;
8432 super
->buf
= super
->next_buf
;
8434 super
->next_len
= 0;
8435 super
->next_buf
= NULL
;
8438 mpb
= super
->anchor
;
8441 case update_general_migration_checkpoint
: {
8442 struct intel_dev
*id
;
8443 struct imsm_update_general_migration_checkpoint
*u
=
8444 (void *)update
->buf
;
8446 dprintf("imsm: process_update() "
8447 "for update_general_migration_checkpoint called\n");
8449 /* find device under general migration */
8450 for (id
= super
->devlist
; id
; id
= id
->next
) {
8451 if (is_gen_migration(id
->dev
)) {
8452 id
->dev
->vol
.curr_migr_unit
=
8453 __cpu_to_le32(u
->curr_migr_unit
);
8454 super
->updates_pending
++;
8459 case update_takeover
: {
8460 struct imsm_update_takeover
*u
= (void *)update
->buf
;
8461 if (apply_takeover_update(u
, super
, &update
->space_list
)) {
8462 imsm_update_version_info(super
);
8463 super
->updates_pending
++;
8468 case update_reshape_container_disks
: {
8469 struct imsm_update_reshape
*u
= (void *)update
->buf
;
8470 if (apply_reshape_container_disks_update(
8471 u
, super
, &update
->space_list
))
8472 super
->updates_pending
++;
8475 case update_reshape_migration
: {
8476 struct imsm_update_reshape_migration
*u
= (void *)update
->buf
;
8477 if (apply_reshape_migration_update(
8478 u
, super
, &update
->space_list
))
8479 super
->updates_pending
++;
8482 case update_size_change
: {
8483 struct imsm_update_size_change
*u
= (void *)update
->buf
;
8484 if (apply_size_change_update(u
, super
))
8485 super
->updates_pending
++;
8488 case update_activate_spare
: {
8489 struct imsm_update_activate_spare
*u
= (void *) update
->buf
;
8490 if (apply_update_activate_spare(u
, super
, st
->arrays
))
8491 super
->updates_pending
++;
8494 case update_create_array
: {
8495 /* someone wants to create a new array, we need to be aware of
8496 * a few races/collisions:
8497 * 1/ 'Create' called by two separate instances of mdadm
8498 * 2/ 'Create' versus 'activate_spare': mdadm has chosen
8499 * devices that have since been assimilated via
8501 * In the event this update can not be carried out mdadm will
8502 * (FIX ME) notice that its update did not take hold.
8504 struct imsm_update_create_array
*u
= (void *) update
->buf
;
8505 struct intel_dev
*dv
;
8506 struct imsm_dev
*dev
;
8507 struct imsm_map
*map
, *new_map
;
8508 unsigned long long start
, end
;
8509 unsigned long long new_start
, new_end
;
8511 struct disk_info
*inf
;
8514 /* handle racing creates: first come first serve */
8515 if (u
->dev_idx
< mpb
->num_raid_devs
) {
8516 dprintf("%s: subarray %d already defined\n",
8517 __func__
, u
->dev_idx
);
8521 /* check update is next in sequence */
8522 if (u
->dev_idx
!= mpb
->num_raid_devs
) {
8523 dprintf("%s: can not create array %d expected index %d\n",
8524 __func__
, u
->dev_idx
, mpb
->num_raid_devs
);
8528 new_map
= get_imsm_map(&u
->dev
, MAP_0
);
8529 new_start
= pba_of_lba0(new_map
);
8530 new_end
= new_start
+ blocks_per_member(new_map
);
8531 inf
= get_disk_info(u
);
8533 /* handle activate_spare versus create race:
8534 * check to make sure that overlapping arrays do not include
8537 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
8538 dev
= get_imsm_dev(super
, i
);
8539 map
= get_imsm_map(dev
, MAP_0
);
8540 start
= pba_of_lba0(map
);
8541 end
= start
+ blocks_per_member(map
);
8542 if ((new_start
>= start
&& new_start
<= end
) ||
8543 (start
>= new_start
&& start
<= new_end
))
8548 if (disks_overlap(super
, i
, u
)) {
8549 dprintf("%s: arrays overlap\n", __func__
);
8554 /* check that prepare update was successful */
8555 if (!update
->space
) {
8556 dprintf("%s: prepare update failed\n", __func__
);
8560 /* check that all disks are still active before committing
8561 * changes. FIXME: could we instead handle this by creating a
8562 * degraded array? That's probably not what the user expects,
8563 * so better to drop this update on the floor.
8565 for (i
= 0; i
< new_map
->num_members
; i
++) {
8566 dl
= serial_to_dl(inf
[i
].serial
, super
);
8568 dprintf("%s: disk disappeared\n", __func__
);
8573 super
->updates_pending
++;
8575 /* convert spares to members and fixup ord_tbl */
8576 for (i
= 0; i
< new_map
->num_members
; i
++) {
8577 dl
= serial_to_dl(inf
[i
].serial
, super
);
8578 if (dl
->index
== -1) {
8579 dl
->index
= mpb
->num_disks
;
8581 dl
->disk
.status
|= CONFIGURED_DISK
;
8582 dl
->disk
.status
&= ~SPARE_DISK
;
8584 set_imsm_ord_tbl_ent(new_map
, i
, dl
->index
);
8589 update
->space
= NULL
;
8590 imsm_copy_dev(dev
, &u
->dev
);
8591 dv
->index
= u
->dev_idx
;
8592 dv
->next
= super
->devlist
;
8593 super
->devlist
= dv
;
8594 mpb
->num_raid_devs
++;
8596 imsm_update_version_info(super
);
8599 /* mdmon knows how to release update->space, but not
8600 * ((struct intel_dev *) update->space)->dev
8602 if (update
->space
) {
8608 case update_kill_array
: {
8609 struct imsm_update_kill_array
*u
= (void *) update
->buf
;
8610 int victim
= u
->dev_idx
;
8611 struct active_array
*a
;
8612 struct intel_dev
**dp
;
8613 struct imsm_dev
*dev
;
8615 /* sanity check that we are not affecting the uuid of
8616 * active arrays, or deleting an active array
8618 * FIXME when immutable ids are available, but note that
8619 * we'll also need to fixup the invalidated/active
8620 * subarray indexes in mdstat
8622 for (a
= st
->arrays
; a
; a
= a
->next
)
8623 if (a
->info
.container_member
>= victim
)
8625 /* by definition if mdmon is running at least one array
8626 * is active in the container, so checking
8627 * mpb->num_raid_devs is just extra paranoia
8629 dev
= get_imsm_dev(super
, victim
);
8630 if (a
|| !dev
|| mpb
->num_raid_devs
== 1) {
8631 dprintf("failed to delete subarray-%d\n", victim
);
8635 for (dp
= &super
->devlist
; *dp
;)
8636 if ((*dp
)->index
== (unsigned)super
->current_vol
) {
8639 if ((*dp
)->index
> (unsigned)victim
)
8643 mpb
->num_raid_devs
--;
8644 super
->updates_pending
++;
8647 case update_rename_array
: {
8648 struct imsm_update_rename_array
*u
= (void *) update
->buf
;
8649 char name
[MAX_RAID_SERIAL_LEN
+1];
8650 int target
= u
->dev_idx
;
8651 struct active_array
*a
;
8652 struct imsm_dev
*dev
;
8654 /* sanity check that we are not affecting the uuid of
8657 snprintf(name
, MAX_RAID_SERIAL_LEN
, "%s", (char *) u
->name
);
8658 name
[MAX_RAID_SERIAL_LEN
] = '\0';
8659 for (a
= st
->arrays
; a
; a
= a
->next
)
8660 if (a
->info
.container_member
== target
)
8662 dev
= get_imsm_dev(super
, u
->dev_idx
);
8663 if (a
|| !dev
|| !check_name(super
, name
, 1)) {
8664 dprintf("failed to rename subarray-%d\n", target
);
8668 snprintf((char *) dev
->volume
, MAX_RAID_SERIAL_LEN
, "%s", name
);
8669 super
->updates_pending
++;
8672 case update_add_remove_disk
: {
8673 /* we may be able to repair some arrays if disks are
8674 * being added, check teh status of add_remove_disk
8675 * if discs has been added.
8677 if (add_remove_disk_update(super
)) {
8678 struct active_array
*a
;
8680 super
->updates_pending
++;
8681 for (a
= st
->arrays
; a
; a
= a
->next
)
8682 a
->check_degraded
= 1;
8687 fprintf(stderr
, "error: unsuported process update type:"
8688 "(type: %d)\n", type
);
8692 static struct mdinfo
*get_spares_for_grow(struct supertype
*st
);
8694 static void imsm_prepare_update(struct supertype
*st
,
8695 struct metadata_update
*update
)
8698 * Allocate space to hold new disk entries, raid-device entries or a new
8699 * mpb if necessary. The manager synchronously waits for updates to
8700 * complete in the monitor, so new mpb buffers allocated here can be
8701 * integrated by the monitor thread without worrying about live pointers
8702 * in the manager thread.
8704 enum imsm_update_type type
= *(enum imsm_update_type
*) update
->buf
;
8705 struct intel_super
*super
= st
->sb
;
8706 struct imsm_super
*mpb
= super
->anchor
;
8711 case update_general_migration_checkpoint
:
8712 dprintf("imsm: prepare_update() "
8713 "for update_general_migration_checkpoint called\n");
8715 case update_takeover
: {
8716 struct imsm_update_takeover
*u
= (void *)update
->buf
;
8717 if (u
->direction
== R0_TO_R10
) {
8718 void **tail
= (void **)&update
->space_list
;
8719 struct imsm_dev
*dev
= get_imsm_dev(super
, u
->subarray
);
8720 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
8721 int num_members
= map
->num_members
;
8725 /* allocate memory for added disks */
8726 for (i
= 0; i
< num_members
; i
++) {
8727 size
= sizeof(struct dl
);
8728 space
= malloc(size
);
8737 /* allocate memory for new device */
8738 size
= sizeof_imsm_dev(super
->devlist
->dev
, 0) +
8739 (num_members
* sizeof(__u32
));
8740 space
= malloc(size
);
8749 len
= disks_to_mpb_size(num_members
* 2);
8751 /* if allocation didn't success, free buffer */
8752 while (update
->space_list
) {
8753 void **sp
= update
->space_list
;
8754 update
->space_list
= *sp
;
8762 case update_reshape_container_disks
: {
8763 /* Every raid device in the container is about to
8764 * gain some more devices, and we will enter a
8766 * So each 'imsm_map' will be bigger, and the imsm_vol
8767 * will now hold 2 of them.
8768 * Thus we need new 'struct imsm_dev' allocations sized
8769 * as sizeof_imsm_dev but with more devices in both maps.
8771 struct imsm_update_reshape
*u
= (void *)update
->buf
;
8772 struct intel_dev
*dl
;
8773 void **space_tail
= (void**)&update
->space_list
;
8775 dprintf("imsm: imsm_prepare_update() for update_reshape\n");
8777 for (dl
= super
->devlist
; dl
; dl
= dl
->next
) {
8778 int size
= sizeof_imsm_dev(dl
->dev
, 1);
8780 if (u
->new_raid_disks
> u
->old_raid_disks
)
8781 size
+= sizeof(__u32
)*2*
8782 (u
->new_raid_disks
- u
->old_raid_disks
);
8791 len
= disks_to_mpb_size(u
->new_raid_disks
);
8792 dprintf("New anchor length is %llu\n", (unsigned long long)len
);
8795 case update_reshape_migration
: {
8796 /* for migration level 0->5 we need to add disks
8797 * so the same as for container operation we will copy
8798 * device to the bigger location.
8799 * in memory prepared device and new disk area are prepared
8800 * for usage in process update
8802 struct imsm_update_reshape_migration
*u
= (void *)update
->buf
;
8803 struct intel_dev
*id
;
8804 void **space_tail
= (void **)&update
->space_list
;
8807 int current_level
= -1;
8809 dprintf("imsm: imsm_prepare_update() for update_reshape\n");
8811 /* add space for bigger array in update
8813 for (id
= super
->devlist
; id
; id
= id
->next
) {
8814 if (id
->index
== (unsigned)u
->subdev
) {
8815 size
= sizeof_imsm_dev(id
->dev
, 1);
8816 if (u
->new_raid_disks
> u
->old_raid_disks
)
8817 size
+= sizeof(__u32
)*2*
8818 (u
->new_raid_disks
- u
->old_raid_disks
);
8828 if (update
->space_list
== NULL
)
8831 /* add space for disk in update
8833 size
= sizeof(struct dl
);
8836 free(update
->space_list
);
8837 update
->space_list
= NULL
;
8844 /* add spare device to update
8846 for (id
= super
->devlist
; id
; id
= id
->next
)
8847 if (id
->index
== (unsigned)u
->subdev
) {
8848 struct imsm_dev
*dev
;
8849 struct imsm_map
*map
;
8851 dev
= get_imsm_dev(super
, u
->subdev
);
8852 map
= get_imsm_map(dev
, MAP_0
);
8853 current_level
= map
->raid_level
;
8856 if ((u
->new_level
== 5) && (u
->new_level
!= current_level
)) {
8857 struct mdinfo
*spares
;
8859 spares
= get_spares_for_grow(st
);
8867 makedev(dev
->disk
.major
,
8869 dl
= get_disk_super(super
,
8872 dl
->index
= u
->old_raid_disks
;
8878 len
= disks_to_mpb_size(u
->new_raid_disks
);
8879 dprintf("New anchor length is %llu\n", (unsigned long long)len
);
8882 case update_size_change
: {
8885 case update_create_array
: {
8886 struct imsm_update_create_array
*u
= (void *) update
->buf
;
8887 struct intel_dev
*dv
;
8888 struct imsm_dev
*dev
= &u
->dev
;
8889 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
8891 struct disk_info
*inf
;
8895 inf
= get_disk_info(u
);
8896 len
= sizeof_imsm_dev(dev
, 1);
8897 /* allocate a new super->devlist entry */
8898 dv
= malloc(sizeof(*dv
));
8900 dv
->dev
= malloc(len
);
8905 update
->space
= NULL
;
8909 /* count how many spares will be converted to members */
8910 for (i
= 0; i
< map
->num_members
; i
++) {
8911 dl
= serial_to_dl(inf
[i
].serial
, super
);
8913 /* hmm maybe it failed?, nothing we can do about
8918 if (count_memberships(dl
, super
) == 0)
8921 len
+= activate
* sizeof(struct imsm_disk
);
8928 /* check if we need a larger metadata buffer */
8929 if (super
->next_buf
)
8930 buf_len
= super
->next_len
;
8932 buf_len
= super
->len
;
8934 if (__le32_to_cpu(mpb
->mpb_size
) + len
> buf_len
) {
8935 /* ok we need a larger buf than what is currently allocated
8936 * if this allocation fails process_update will notice that
8937 * ->next_len is set and ->next_buf is NULL
8939 buf_len
= ROUND_UP(__le32_to_cpu(mpb
->mpb_size
) + len
, 512);
8940 if (super
->next_buf
)
8941 free(super
->next_buf
);
8943 super
->next_len
= buf_len
;
8944 if (posix_memalign(&super
->next_buf
, 512, buf_len
) == 0)
8945 memset(super
->next_buf
, 0, buf_len
);
8947 super
->next_buf
= NULL
;
8951 /* must be called while manager is quiesced */
8952 static void imsm_delete(struct intel_super
*super
, struct dl
**dlp
, unsigned index
)
8954 struct imsm_super
*mpb
= super
->anchor
;
8956 struct imsm_dev
*dev
;
8957 struct imsm_map
*map
;
8958 int i
, j
, num_members
;
8961 dprintf("%s: deleting device[%d] from imsm_super\n",
8964 /* shift all indexes down one */
8965 for (iter
= super
->disks
; iter
; iter
= iter
->next
)
8966 if (iter
->index
> (int)index
)
8968 for (iter
= super
->missing
; iter
; iter
= iter
->next
)
8969 if (iter
->index
> (int)index
)
8972 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
8973 dev
= get_imsm_dev(super
, i
);
8974 map
= get_imsm_map(dev
, MAP_0
);
8975 num_members
= map
->num_members
;
8976 for (j
= 0; j
< num_members
; j
++) {
8977 /* update ord entries being careful not to propagate
8978 * ord-flags to the first map
8980 ord
= get_imsm_ord_tbl_ent(dev
, j
, MAP_X
);
8982 if (ord_to_idx(ord
) <= index
)
8985 map
= get_imsm_map(dev
, MAP_0
);
8986 set_imsm_ord_tbl_ent(map
, j
, ord_to_idx(ord
- 1));
8987 map
= get_imsm_map(dev
, MAP_1
);
8989 set_imsm_ord_tbl_ent(map
, j
, ord
- 1);
8994 super
->updates_pending
++;
8996 struct dl
*dl
= *dlp
;
8998 *dlp
= (*dlp
)->next
;
8999 __free_imsm_disk(dl
);
9002 #endif /* MDASSEMBLE */
9004 static void close_targets(int *targets
, int new_disks
)
9011 for (i
= 0; i
< new_disks
; i
++) {
9012 if (targets
[i
] >= 0) {
9019 static int imsm_get_allowed_degradation(int level
, int raid_disks
,
9020 struct intel_super
*super
,
9021 struct imsm_dev
*dev
)
9027 struct imsm_map
*map
;
9030 ret_val
= raid_disks
/2;
9031 /* check map if all disks pairs not failed
9034 map
= get_imsm_map(dev
, MAP_0
);
9035 for (i
= 0; i
< ret_val
; i
++) {
9036 int degradation
= 0;
9037 if (get_imsm_disk(super
, i
) == NULL
)
9039 if (get_imsm_disk(super
, i
+ 1) == NULL
)
9041 if (degradation
== 2)
9044 map
= get_imsm_map(dev
, MAP_1
);
9045 /* if there is no second map
9046 * result can be returned
9050 /* check degradation in second map
9052 for (i
= 0; i
< ret_val
; i
++) {
9053 int degradation
= 0;
9054 if (get_imsm_disk(super
, i
) == NULL
)
9056 if (get_imsm_disk(super
, i
+ 1) == NULL
)
9058 if (degradation
== 2)
9073 /*******************************************************************************
9074 * Function: open_backup_targets
9075 * Description: Function opens file descriptors for all devices given in
9078 * info : general array info
9079 * raid_disks : number of disks
9080 * raid_fds : table of device's file descriptors
9081 * super : intel super for raid10 degradation check
9082 * dev : intel device for raid10 degradation check
9086 ******************************************************************************/
9087 int open_backup_targets(struct mdinfo
*info
, int raid_disks
, int *raid_fds
,
9088 struct intel_super
*super
, struct imsm_dev
*dev
)
9094 for (i
= 0; i
< raid_disks
; i
++)
9097 for (sd
= info
->devs
; sd
; sd
= sd
->next
) {
9100 if (sd
->disk
.state
& (1<<MD_DISK_FAULTY
)) {
9101 dprintf("disk is faulty!!\n");
9105 if ((sd
->disk
.raid_disk
>= raid_disks
) ||
9106 (sd
->disk
.raid_disk
< 0))
9109 dn
= map_dev(sd
->disk
.major
,
9111 raid_fds
[sd
->disk
.raid_disk
] = dev_open(dn
, O_RDWR
);
9112 if (raid_fds
[sd
->disk
.raid_disk
] < 0) {
9113 fprintf(stderr
, "cannot open component\n");
9118 /* check if maximum array degradation level is not exceeded
9120 if ((raid_disks
- opened
) >
9121 imsm_get_allowed_degradation(info
->new_level
,
9124 fprintf(stderr
, "Not enough disks can be opened.\n");
9125 close_targets(raid_fds
, raid_disks
);
9132 /*******************************************************************************
9133 * Function: init_migr_record_imsm
9134 * Description: Function inits imsm migration record
9136 * super : imsm internal array info
9137 * dev : device under migration
9138 * info : general array info to find the smallest device
9141 ******************************************************************************/
9142 void init_migr_record_imsm(struct supertype
*st
, struct imsm_dev
*dev
,
9143 struct mdinfo
*info
)
9145 struct intel_super
*super
= st
->sb
;
9146 struct migr_record
*migr_rec
= super
->migr_rec
;
9148 unsigned long long dsize
, dev_sectors
;
9149 long long unsigned min_dev_sectors
= -1LLU;
9153 struct imsm_map
*map_dest
= get_imsm_map(dev
, MAP_0
);
9154 struct imsm_map
*map_src
= get_imsm_map(dev
, MAP_1
);
9155 unsigned long long num_migr_units
;
9156 unsigned long long array_blocks
;
9158 memset(migr_rec
, 0, sizeof(struct migr_record
));
9159 migr_rec
->family_num
= __cpu_to_le32(super
->anchor
->family_num
);
9161 /* only ascending reshape supported now */
9162 migr_rec
->ascending_migr
= __cpu_to_le32(1);
9164 migr_rec
->dest_depth_per_unit
= GEN_MIGR_AREA_SIZE
/
9165 max(map_dest
->blocks_per_strip
, map_src
->blocks_per_strip
);
9166 migr_rec
->dest_depth_per_unit
*=
9167 max(map_dest
->blocks_per_strip
, map_src
->blocks_per_strip
);
9168 new_data_disks
= imsm_num_data_members(dev
, MAP_0
);
9169 migr_rec
->blocks_per_unit
=
9170 __cpu_to_le32(migr_rec
->dest_depth_per_unit
* new_data_disks
);
9171 migr_rec
->dest_depth_per_unit
=
9172 __cpu_to_le32(migr_rec
->dest_depth_per_unit
);
9173 array_blocks
= info
->component_size
* new_data_disks
;
9175 array_blocks
/ __le32_to_cpu(migr_rec
->blocks_per_unit
);
9177 if (array_blocks
% __le32_to_cpu(migr_rec
->blocks_per_unit
))
9179 migr_rec
->num_migr_units
= __cpu_to_le32(num_migr_units
);
9181 migr_rec
->post_migr_vol_cap
= dev
->size_low
;
9182 migr_rec
->post_migr_vol_cap_hi
= dev
->size_high
;
9185 /* Find the smallest dev */
9186 for (sd
= info
->devs
; sd
; sd
= sd
->next
) {
9187 sprintf(nm
, "%d:%d", sd
->disk
.major
, sd
->disk
.minor
);
9188 fd
= dev_open(nm
, O_RDONLY
);
9191 get_dev_size(fd
, NULL
, &dsize
);
9192 dev_sectors
= dsize
/ 512;
9193 if (dev_sectors
< min_dev_sectors
)
9194 min_dev_sectors
= dev_sectors
;
9197 migr_rec
->ckpt_area_pba
= __cpu_to_le32(min_dev_sectors
-
9198 RAID_DISK_RESERVED_BLOCKS_IMSM_HI
);
9200 write_imsm_migr_rec(st
);
9205 /*******************************************************************************
9206 * Function: save_backup_imsm
9207 * Description: Function saves critical data stripes to Migration Copy Area
9208 * and updates the current migration unit status.
9209 * Use restore_stripes() to form a destination stripe,
9210 * and to write it to the Copy Area.
9212 * st : supertype information
9213 * dev : imsm device that backup is saved for
9214 * info : general array info
9215 * buf : input buffer
9216 * length : length of data to backup (blocks_per_unit)
9220 ******************************************************************************/
9221 int save_backup_imsm(struct supertype
*st
,
9222 struct imsm_dev
*dev
,
9223 struct mdinfo
*info
,
9228 struct intel_super
*super
= st
->sb
;
9229 unsigned long long *target_offsets
= NULL
;
9230 int *targets
= NULL
;
9232 struct imsm_map
*map_dest
= get_imsm_map(dev
, MAP_0
);
9233 int new_disks
= map_dest
->num_members
;
9234 int dest_layout
= 0;
9236 unsigned long long start
;
9237 int data_disks
= imsm_num_data_members(dev
, MAP_0
);
9239 targets
= malloc(new_disks
* sizeof(int));
9243 for (i
= 0; i
< new_disks
; i
++)
9246 target_offsets
= malloc(new_disks
* sizeof(unsigned long long));
9247 if (!target_offsets
)
9250 start
= info
->reshape_progress
* 512;
9251 for (i
= 0; i
< new_disks
; i
++) {
9252 target_offsets
[i
] = (unsigned long long)
9253 __le32_to_cpu(super
->migr_rec
->ckpt_area_pba
) * 512;
9254 /* move back copy area adderss, it will be moved forward
9255 * in restore_stripes() using start input variable
9257 target_offsets
[i
] -= start
/data_disks
;
9260 if (open_backup_targets(info
, new_disks
, targets
,
9264 dest_layout
= imsm_level_to_layout(map_dest
->raid_level
);
9265 dest_chunk
= __le16_to_cpu(map_dest
->blocks_per_strip
) * 512;
9267 if (restore_stripes(targets
, /* list of dest devices */
9268 target_offsets
, /* migration record offsets */
9271 map_dest
->raid_level
,
9273 -1, /* source backup file descriptor */
9274 0, /* input buf offset
9275 * always 0 buf is already offseted */
9279 fprintf(stderr
, Name
": Error restoring stripes\n");
9287 close_targets(targets
, new_disks
);
9290 free(target_offsets
);
9295 /*******************************************************************************
9296 * Function: save_checkpoint_imsm
9297 * Description: Function called for current unit status update
9298 * in the migration record. It writes it to disk.
9300 * super : imsm internal array info
9301 * info : general array info
9305 * 2: failure, means no valid migration record
9306 * / no general migration in progress /
9307 ******************************************************************************/
9308 int save_checkpoint_imsm(struct supertype
*st
, struct mdinfo
*info
, int state
)
9310 struct intel_super
*super
= st
->sb
;
9311 unsigned long long blocks_per_unit
;
9312 unsigned long long curr_migr_unit
;
9314 if (load_imsm_migr_rec(super
, info
) != 0) {
9315 dprintf("imsm: ERROR: Cannot read migration record "
9316 "for checkpoint save.\n");
9320 blocks_per_unit
= __le32_to_cpu(super
->migr_rec
->blocks_per_unit
);
9321 if (blocks_per_unit
== 0) {
9322 dprintf("imsm: no migration in progress.\n");
9325 curr_migr_unit
= info
->reshape_progress
/ blocks_per_unit
;
9326 /* check if array is alligned to copy area
9327 * if it is not alligned, add one to current migration unit value
9328 * this can happend on array reshape finish only
9330 if (info
->reshape_progress
% blocks_per_unit
)
9333 super
->migr_rec
->curr_migr_unit
=
9334 __cpu_to_le32(curr_migr_unit
);
9335 super
->migr_rec
->rec_status
= __cpu_to_le32(state
);
9336 super
->migr_rec
->dest_1st_member_lba
=
9337 __cpu_to_le32(curr_migr_unit
*
9338 __le32_to_cpu(super
->migr_rec
->dest_depth_per_unit
));
9339 if (write_imsm_migr_rec(st
) < 0) {
9340 dprintf("imsm: Cannot write migration record "
9341 "outside backup area\n");
9348 /*******************************************************************************
9349 * Function: recover_backup_imsm
9350 * Description: Function recovers critical data from the Migration Copy Area
9351 * while assembling an array.
9353 * super : imsm internal array info
9354 * info : general array info
9356 * 0 : success (or there is no data to recover)
9358 ******************************************************************************/
9359 int recover_backup_imsm(struct supertype
*st
, struct mdinfo
*info
)
9361 struct intel_super
*super
= st
->sb
;
9362 struct migr_record
*migr_rec
= super
->migr_rec
;
9363 struct imsm_map
*map_dest
= NULL
;
9364 struct intel_dev
*id
= NULL
;
9365 unsigned long long read_offset
;
9366 unsigned long long write_offset
;
9368 int *targets
= NULL
;
9369 int new_disks
, i
, err
;
9372 unsigned long curr_migr_unit
= __le32_to_cpu(migr_rec
->curr_migr_unit
);
9373 unsigned long num_migr_units
= __le32_to_cpu(migr_rec
->num_migr_units
);
9375 int skipped_disks
= 0;
9377 err
= sysfs_get_str(info
, NULL
, "array_state", (char *)buffer
, 20);
9381 /* recover data only during assemblation */
9382 if (strncmp(buffer
, "inactive", 8) != 0)
9384 /* no data to recover */
9385 if (__le32_to_cpu(migr_rec
->rec_status
) == UNIT_SRC_NORMAL
)
9387 if (curr_migr_unit
>= num_migr_units
)
9390 /* find device during reshape */
9391 for (id
= super
->devlist
; id
; id
= id
->next
)
9392 if (is_gen_migration(id
->dev
))
9397 map_dest
= get_imsm_map(id
->dev
, MAP_0
);
9398 new_disks
= map_dest
->num_members
;
9400 read_offset
= (unsigned long long)
9401 __le32_to_cpu(migr_rec
->ckpt_area_pba
) * 512;
9403 write_offset
= ((unsigned long long)
9404 __le32_to_cpu(migr_rec
->dest_1st_member_lba
) +
9405 pba_of_lba0(map_dest
)) * 512;
9407 unit_len
= __le32_to_cpu(migr_rec
->dest_depth_per_unit
) * 512;
9408 if (posix_memalign((void **)&buf
, 512, unit_len
) != 0)
9410 targets
= malloc(new_disks
* sizeof(int));
9414 if (open_backup_targets(info
, new_disks
, targets
, super
, id
->dev
)) {
9416 Name
": Cannot open some devices belonging to array.\n");
9420 for (i
= 0; i
< new_disks
; i
++) {
9421 if (targets
[i
] < 0) {
9425 if (lseek64(targets
[i
], read_offset
, SEEK_SET
) < 0) {
9427 Name
": Cannot seek to block: %s\n",
9432 if ((unsigned)read(targets
[i
], buf
, unit_len
) != unit_len
) {
9434 Name
": Cannot read copy area block: %s\n",
9439 if (lseek64(targets
[i
], write_offset
, SEEK_SET
) < 0) {
9441 Name
": Cannot seek to block: %s\n",
9446 if ((unsigned)write(targets
[i
], buf
, unit_len
) != unit_len
) {
9448 Name
": Cannot restore block: %s\n",
9455 if (skipped_disks
> imsm_get_allowed_degradation(info
->new_level
,
9460 Name
": Cannot restore data from backup."
9461 " Too many failed disks\n");
9465 if (save_checkpoint_imsm(st
, info
, UNIT_SRC_NORMAL
)) {
9466 /* ignore error == 2, this can mean end of reshape here
9468 dprintf("imsm: Cannot write checkpoint to "
9469 "migration record (UNIT_SRC_NORMAL) during restart\n");
9475 for (i
= 0; i
< new_disks
; i
++)
9484 static char disk_by_path
[] = "/dev/disk/by-path/";
9486 static const char *imsm_get_disk_controller_domain(const char *path
)
9488 char disk_path
[PATH_MAX
];
9492 strncpy(disk_path
, disk_by_path
, PATH_MAX
- 1);
9493 strncat(disk_path
, path
, PATH_MAX
- strlen(disk_path
) - 1);
9494 if (stat(disk_path
, &st
) == 0) {
9495 struct sys_dev
* hba
;
9498 path
= devt_to_devpath(st
.st_rdev
);
9501 hba
= find_disk_attached_hba(-1, path
);
9502 if (hba
&& hba
->type
== SYS_DEV_SAS
)
9504 else if (hba
&& hba
->type
== SYS_DEV_SATA
)
9508 dprintf("path: %s hba: %s attached: %s\n",
9509 path
, (hba
) ? hba
->path
: "NULL", drv
);
9517 static int imsm_find_array_minor_by_subdev(int subdev
, int container
, int *minor
)
9519 char subdev_name
[20];
9520 struct mdstat_ent
*mdstat
;
9522 sprintf(subdev_name
, "%d", subdev
);
9523 mdstat
= mdstat_by_subdev(subdev_name
, container
);
9527 *minor
= mdstat
->devnum
;
9528 free_mdstat(mdstat
);
9532 static int imsm_reshape_is_allowed_on_container(struct supertype
*st
,
9533 struct geo_params
*geo
,
9534 int *old_raid_disks
,
9537 /* currently we only support increasing the number of devices
9538 * for a container. This increases the number of device for each
9539 * member array. They must all be RAID0 or RAID5.
9542 struct mdinfo
*info
, *member
;
9543 int devices_that_can_grow
= 0;
9545 dprintf("imsm: imsm_reshape_is_allowed_on_container(ENTER): "
9546 "st->devnum = (%i)\n",
9549 if (geo
->size
!= -1 ||
9550 geo
->level
!= UnSet
||
9551 geo
->layout
!= UnSet
||
9552 geo
->chunksize
!= 0 ||
9553 geo
->raid_disks
== UnSet
) {
9554 dprintf("imsm: Container operation is allowed for "
9555 "raid disks number change only.\n");
9559 if (direction
== ROLLBACK_METADATA_CHANGES
) {
9560 dprintf("imsm: Metadata changes rollback is not supported for "
9561 "container operation.\n");
9565 info
= container_content_imsm(st
, NULL
);
9566 for (member
= info
; member
; member
= member
->next
) {
9570 dprintf("imsm: checking device_num: %i\n",
9571 member
->container_member
);
9573 if (geo
->raid_disks
<= member
->array
.raid_disks
) {
9574 /* we work on container for Online Capacity Expansion
9575 * only so raid_disks has to grow
9577 dprintf("imsm: for container operation raid disks "
9578 "increase is required\n");
9582 if ((info
->array
.level
!= 0) &&
9583 (info
->array
.level
!= 5)) {
9584 /* we cannot use this container with other raid level
9586 dprintf("imsm: for container operation wrong"
9587 " raid level (%i) detected\n",
9591 /* check for platform support
9592 * for this raid level configuration
9594 struct intel_super
*super
= st
->sb
;
9595 if (!is_raid_level_supported(super
->orom
,
9596 member
->array
.level
,
9598 dprintf("platform does not support raid%d with"
9602 geo
->raid_disks
> 1 ? "s" : "");
9605 /* check if component size is aligned to chunk size
9607 if (info
->component_size
%
9608 (info
->array
.chunk_size
/512)) {
9609 dprintf("Component size is not aligned to "
9615 if (*old_raid_disks
&&
9616 info
->array
.raid_disks
!= *old_raid_disks
)
9618 *old_raid_disks
= info
->array
.raid_disks
;
9620 /* All raid5 and raid0 volumes in container
9621 * have to be ready for Online Capacity Expansion
9622 * so they need to be assembled. We have already
9623 * checked that no recovery etc is happening.
9625 result
= imsm_find_array_minor_by_subdev(member
->container_member
,
9629 dprintf("imsm: cannot find array\n");
9632 devices_that_can_grow
++;
9635 if (!member
&& devices_that_can_grow
)
9639 dprintf("\tContainer operation allowed\n");
9641 dprintf("\tError: %i\n", ret_val
);
9646 /* Function: get_spares_for_grow
9647 * Description: Allocates memory and creates list of spare devices
9648 * avaliable in container. Checks if spare drive size is acceptable.
9649 * Parameters: Pointer to the supertype structure
9650 * Returns: Pointer to the list of spare devices (mdinfo structure) on success,
9653 static struct mdinfo
*get_spares_for_grow(struct supertype
*st
)
9655 unsigned long long min_size
= min_acceptable_spare_size_imsm(st
);
9656 return container_choose_spares(st
, min_size
, NULL
, NULL
, NULL
, 0);
9659 /******************************************************************************
9660 * function: imsm_create_metadata_update_for_reshape
9661 * Function creates update for whole IMSM container.
9663 ******************************************************************************/
9664 static int imsm_create_metadata_update_for_reshape(
9665 struct supertype
*st
,
9666 struct geo_params
*geo
,
9668 struct imsm_update_reshape
**updatep
)
9670 struct intel_super
*super
= st
->sb
;
9671 struct imsm_super
*mpb
= super
->anchor
;
9672 int update_memory_size
= 0;
9673 struct imsm_update_reshape
*u
= NULL
;
9674 struct mdinfo
*spares
= NULL
;
9676 int delta_disks
= 0;
9679 dprintf("imsm_update_metadata_for_reshape(enter) raid_disks = %i\n",
9682 delta_disks
= geo
->raid_disks
- old_raid_disks
;
9684 /* size of all update data without anchor */
9685 update_memory_size
= sizeof(struct imsm_update_reshape
);
9687 /* now add space for spare disks that we need to add. */
9688 update_memory_size
+= sizeof(u
->new_disks
[0]) * (delta_disks
- 1);
9690 u
= calloc(1, update_memory_size
);
9693 "cannot get memory for imsm_update_reshape update\n");
9696 u
->type
= update_reshape_container_disks
;
9697 u
->old_raid_disks
= old_raid_disks
;
9698 u
->new_raid_disks
= geo
->raid_disks
;
9700 /* now get spare disks list
9702 spares
= get_spares_for_grow(st
);
9705 || delta_disks
> spares
->array
.spare_disks
) {
9706 fprintf(stderr
, Name
": imsm: ERROR: Cannot get spare devices "
9707 "for %s.\n", geo
->dev_name
);
9712 /* we have got spares
9713 * update disk list in imsm_disk list table in anchor
9715 dprintf("imsm: %i spares are available.\n\n",
9716 spares
->array
.spare_disks
);
9719 for (i
= 0; i
< delta_disks
; i
++) {
9724 u
->new_disks
[i
] = makedev(dev
->disk
.major
,
9726 dl
= get_disk_super(super
, dev
->disk
.major
, dev
->disk
.minor
);
9727 dl
->index
= mpb
->num_disks
;
9737 dprintf("imsm: reshape update preparation :");
9738 if (i
== delta_disks
) {
9741 return update_memory_size
;
9744 dprintf(" Error\n");
9750 /******************************************************************************
9751 * function: imsm_create_metadata_update_for_size_change()
9752 * Creates update for IMSM array for array size change.
9754 ******************************************************************************/
9755 static int imsm_create_metadata_update_for_size_change(
9756 struct supertype
*st
,
9757 struct geo_params
*geo
,
9758 struct imsm_update_size_change
**updatep
)
9760 struct intel_super
*super
= st
->sb
;
9761 int update_memory_size
= 0;
9762 struct imsm_update_size_change
*u
= NULL
;
9764 dprintf("imsm_create_metadata_update_for_size_change(enter)"
9765 " New size = %llu\n", geo
->size
);
9767 /* size of all update data without anchor */
9768 update_memory_size
= sizeof(struct imsm_update_size_change
);
9770 u
= calloc(1, update_memory_size
);
9772 dprintf("error: cannot get memory for "
9773 "imsm_create_metadata_update_for_size_change\n");
9776 u
->type
= update_size_change
;
9777 u
->subdev
= super
->current_vol
;
9778 u
->new_size
= geo
->size
;
9780 dprintf("imsm: reshape update preparation : OK\n");
9783 return update_memory_size
;
9786 /******************************************************************************
9787 * function: imsm_create_metadata_update_for_migration()
9788 * Creates update for IMSM array.
9790 ******************************************************************************/
9791 static int imsm_create_metadata_update_for_migration(
9792 struct supertype
*st
,
9793 struct geo_params
*geo
,
9794 struct imsm_update_reshape_migration
**updatep
)
9796 struct intel_super
*super
= st
->sb
;
9797 int update_memory_size
= 0;
9798 struct imsm_update_reshape_migration
*u
= NULL
;
9799 struct imsm_dev
*dev
;
9800 int previous_level
= -1;
9802 dprintf("imsm_create_metadata_update_for_migration(enter)"
9803 " New Level = %i\n", geo
->level
);
9805 /* size of all update data without anchor */
9806 update_memory_size
= sizeof(struct imsm_update_reshape_migration
);
9808 u
= calloc(1, update_memory_size
);
9810 dprintf("error: cannot get memory for "
9811 "imsm_create_metadata_update_for_migration\n");
9814 u
->type
= update_reshape_migration
;
9815 u
->subdev
= super
->current_vol
;
9816 u
->new_level
= geo
->level
;
9817 u
->new_layout
= geo
->layout
;
9818 u
->new_raid_disks
= u
->old_raid_disks
= geo
->raid_disks
;
9819 u
->new_disks
[0] = -1;
9820 u
->new_chunksize
= -1;
9822 dev
= get_imsm_dev(super
, u
->subdev
);
9824 struct imsm_map
*map
;
9826 map
= get_imsm_map(dev
, MAP_0
);
9828 int current_chunk_size
=
9829 __le16_to_cpu(map
->blocks_per_strip
) / 2;
9831 if (geo
->chunksize
!= current_chunk_size
) {
9832 u
->new_chunksize
= geo
->chunksize
/ 1024;
9834 "chunk size change from %i to %i\n",
9835 current_chunk_size
, u
->new_chunksize
);
9837 previous_level
= map
->raid_level
;
9840 if ((geo
->level
== 5) && (previous_level
== 0)) {
9841 struct mdinfo
*spares
= NULL
;
9843 u
->new_raid_disks
++;
9844 spares
= get_spares_for_grow(st
);
9845 if ((spares
== NULL
) || (spares
->array
.spare_disks
< 1)) {
9848 update_memory_size
= 0;
9849 dprintf("error: cannot get spare device "
9850 "for requested migration");
9855 dprintf("imsm: reshape update preparation : OK\n");
9858 return update_memory_size
;
9861 static void imsm_update_metadata_locally(struct supertype
*st
,
9864 struct metadata_update mu
;
9869 mu
.space_list
= NULL
;
9871 imsm_prepare_update(st
, &mu
);
9872 imsm_process_update(st
, &mu
);
9874 while (mu
.space_list
) {
9875 void **space
= mu
.space_list
;
9876 mu
.space_list
= *space
;
9881 /***************************************************************************
9882 * Function: imsm_analyze_change
9883 * Description: Function analyze change for single volume
9884 * and validate if transition is supported
9885 * Parameters: Geometry parameters, supertype structure,
9886 * metadata change direction (apply/rollback)
9887 * Returns: Operation type code on success, -1 if fail
9888 ****************************************************************************/
9889 enum imsm_reshape_type
imsm_analyze_change(struct supertype
*st
,
9890 struct geo_params
*geo
,
9897 /* number of added/removed disks in operation result */
9898 int devNumChange
= 0;
9899 /* imsm compatible layout value for array geometry verification */
9900 int imsm_layout
= -1;
9902 struct imsm_dev
*dev
;
9903 struct intel_super
*super
;
9904 long long current_size
;
9905 unsigned long long free_size
;
9909 getinfo_super_imsm_volume(st
, &info
, NULL
);
9910 if ((geo
->level
!= info
.array
.level
) &&
9911 (geo
->level
>= 0) &&
9912 (geo
->level
!= UnSet
)) {
9913 switch (info
.array
.level
) {
9915 if (geo
->level
== 5) {
9916 change
= CH_MIGRATION
;
9917 if (geo
->layout
!= ALGORITHM_LEFT_ASYMMETRIC
) {
9919 Name
" Error. Requested Layout "
9920 "not supported (left-asymmetric layout "
9921 "is supported only)!\n");
9923 goto analyse_change_exit
;
9925 imsm_layout
= geo
->layout
;
9927 devNumChange
= 1; /* parity disk added */
9928 } else if (geo
->level
== 10) {
9929 change
= CH_TAKEOVER
;
9931 devNumChange
= 2; /* two mirrors added */
9932 imsm_layout
= 0x102; /* imsm supported layout */
9937 if (geo
->level
== 0) {
9938 change
= CH_TAKEOVER
;
9940 devNumChange
= -(geo
->raid_disks
/2);
9941 imsm_layout
= 0; /* imsm raid0 layout */
9947 Name
" Error. Level Migration from %d to %d "
9949 info
.array
.level
, geo
->level
);
9950 goto analyse_change_exit
;
9953 geo
->level
= info
.array
.level
;
9955 if ((geo
->layout
!= info
.array
.layout
)
9956 && ((geo
->layout
!= UnSet
) && (geo
->layout
!= -1))) {
9957 change
= CH_MIGRATION
;
9958 if ((info
.array
.layout
== 0)
9959 && (info
.array
.level
== 5)
9960 && (geo
->layout
== 5)) {
9961 /* reshape 5 -> 4 */
9962 } else if ((info
.array
.layout
== 5)
9963 && (info
.array
.level
== 5)
9964 && (geo
->layout
== 0)) {
9965 /* reshape 4 -> 5 */
9970 Name
" Error. Layout Migration from %d to %d "
9972 info
.array
.layout
, geo
->layout
);
9974 goto analyse_change_exit
;
9977 geo
->layout
= info
.array
.layout
;
9978 if (imsm_layout
== -1)
9979 imsm_layout
= info
.array
.layout
;
9982 if ((geo
->chunksize
> 0) && (geo
->chunksize
!= UnSet
)
9983 && (geo
->chunksize
!= info
.array
.chunk_size
))
9984 change
= CH_MIGRATION
;
9986 geo
->chunksize
= info
.array
.chunk_size
;
9988 chunk
= geo
->chunksize
/ 1024;
9991 dev
= get_imsm_dev(super
, super
->current_vol
);
9992 data_disks
= imsm_num_data_members(dev
, MAP_0
);
9993 /* compute current size per disk member
9995 current_size
= info
.custom_array_size
/ data_disks
;
9997 if (geo
->size
> 0) {
9998 /* align component size
10000 geo
->size
= imsm_component_size_aligment_check(
10001 get_imsm_raid_level(dev
->vol
.map
),
10006 if ((current_size
!= geo
->size
) && (geo
->size
>= 0)) {
10007 if (change
!= -1) {
10009 Name
" Error. Size change should be the only "
10010 "one at a time.\n");
10012 goto analyse_change_exit
;
10014 if ((super
->current_vol
+ 1) != super
->anchor
->num_raid_devs
) {
10016 Name
" Error. The last volume in container "
10017 "can be expanded only (%i/%i).\n",
10018 super
->current_vol
, st
->devnum
);
10019 goto analyse_change_exit
;
10021 /* check the maximum available size
10023 rv
= imsm_get_free_size(st
, dev
->vol
.map
->num_members
,
10024 0, chunk
, &free_size
);
10026 /* Cannot find maximum available space
10030 max_size
= free_size
+ current_size
;
10031 /* align component size
10033 max_size
= imsm_component_size_aligment_check(
10034 get_imsm_raid_level(dev
->vol
.map
),
10038 if (geo
->size
== 0) {
10039 /* requested size change to the maximum available size
10041 if (max_size
== 0) {
10042 fprintf(stderr
, Name
" Error. Cannot find "
10043 "maximum available space.\n");
10045 goto analyse_change_exit
;
10047 geo
->size
= max_size
;
10050 if ((direction
== ROLLBACK_METADATA_CHANGES
)) {
10051 /* accept size for rollback only
10054 /* round size due to metadata compatibility
10056 geo
->size
= (geo
->size
>> SECT_PER_MB_SHIFT
)
10057 << SECT_PER_MB_SHIFT
;
10058 dprintf("Prepare update for size change to %llu\n",
10060 if (current_size
>= geo
->size
) {
10062 Name
" Error. Size expansion is "
10063 "supported only (current size is %llu, "
10064 "requested size /rounded/ is %llu).\n",
10065 current_size
, geo
->size
);
10066 goto analyse_change_exit
;
10068 if (max_size
&& geo
->size
> max_size
) {
10070 Name
" Error. Requested size is larger "
10071 "than maximum available size (maximum "
10072 "available size is %llu, "
10073 "requested size /rounded/ is %llu).\n",
10074 max_size
, geo
->size
);
10075 goto analyse_change_exit
;
10078 geo
->size
*= data_disks
;
10079 geo
->raid_disks
= dev
->vol
.map
->num_members
;
10080 change
= CH_ARRAY_SIZE
;
10082 if (!validate_geometry_imsm(st
,
10085 geo
->raid_disks
+ devNumChange
,
10092 struct intel_super
*super
= st
->sb
;
10093 struct imsm_super
*mpb
= super
->anchor
;
10095 if (mpb
->num_raid_devs
> 1) {
10097 Name
" Error. Cannot perform operation on %s"
10098 "- for this operation it MUST be single "
10099 "array in container\n",
10105 analyse_change_exit
:
10106 if ((direction
== ROLLBACK_METADATA_CHANGES
) &&
10107 ((change
== CH_MIGRATION
) || (change
== CH_TAKEOVER
))) {
10108 dprintf("imsm: Metadata changes rollback is not supported for "
10109 "migration and takeover operations.\n");
10115 int imsm_takeover(struct supertype
*st
, struct geo_params
*geo
)
10117 struct intel_super
*super
= st
->sb
;
10118 struct imsm_update_takeover
*u
;
10120 u
= malloc(sizeof(struct imsm_update_takeover
));
10124 u
->type
= update_takeover
;
10125 u
->subarray
= super
->current_vol
;
10127 /* 10->0 transition */
10128 if (geo
->level
== 0)
10129 u
->direction
= R10_TO_R0
;
10131 /* 0->10 transition */
10132 if (geo
->level
== 10)
10133 u
->direction
= R0_TO_R10
;
10135 /* update metadata locally */
10136 imsm_update_metadata_locally(st
, u
,
10137 sizeof(struct imsm_update_takeover
));
10138 /* and possibly remotely */
10139 if (st
->update_tail
)
10140 append_metadata_update(st
, u
,
10141 sizeof(struct imsm_update_takeover
));
10148 static int imsm_reshape_super(struct supertype
*st
, long long size
, int level
,
10149 int layout
, int chunksize
, int raid_disks
,
10150 int delta_disks
, char *backup
, char *dev
,
10151 int direction
, int verbose
)
10154 struct geo_params geo
;
10156 dprintf("imsm: reshape_super called.\n");
10158 memset(&geo
, 0, sizeof(struct geo_params
));
10160 geo
.dev_name
= dev
;
10161 geo
.dev_id
= st
->devnum
;
10164 geo
.layout
= layout
;
10165 geo
.chunksize
= chunksize
;
10166 geo
.raid_disks
= raid_disks
;
10167 if (delta_disks
!= UnSet
)
10168 geo
.raid_disks
+= delta_disks
;
10170 dprintf("\tfor level : %i\n", geo
.level
);
10171 dprintf("\tfor raid_disks : %i\n", geo
.raid_disks
);
10173 if (experimental() == 0)
10176 if (st
->container_dev
== st
->devnum
) {
10177 /* On container level we can only increase number of devices. */
10178 dprintf("imsm: info: Container operation\n");
10179 int old_raid_disks
= 0;
10181 if (imsm_reshape_is_allowed_on_container(
10182 st
, &geo
, &old_raid_disks
, direction
)) {
10183 struct imsm_update_reshape
*u
= NULL
;
10186 len
= imsm_create_metadata_update_for_reshape(
10187 st
, &geo
, old_raid_disks
, &u
);
10190 dprintf("imsm: Cannot prepare update\n");
10191 goto exit_imsm_reshape_super
;
10195 /* update metadata locally */
10196 imsm_update_metadata_locally(st
, u
, len
);
10197 /* and possibly remotely */
10198 if (st
->update_tail
)
10199 append_metadata_update(st
, u
, len
);
10204 fprintf(stderr
, Name
": (imsm) Operation "
10205 "is not allowed on this container\n");
10208 /* On volume level we support following operations
10209 * - takeover: raid10 -> raid0; raid0 -> raid10
10210 * - chunk size migration
10211 * - migration: raid5 -> raid0; raid0 -> raid5
10213 struct intel_super
*super
= st
->sb
;
10214 struct intel_dev
*dev
= super
->devlist
;
10215 int change
, devnum
;
10216 dprintf("imsm: info: Volume operation\n");
10217 /* find requested device */
10219 if (imsm_find_array_minor_by_subdev(
10220 dev
->index
, st
->container_dev
, &devnum
) == 0
10221 && devnum
== geo
.dev_id
)
10226 fprintf(stderr
, Name
" Cannot find %s (%i) subarray\n",
10227 geo
.dev_name
, geo
.dev_id
);
10228 goto exit_imsm_reshape_super
;
10230 super
->current_vol
= dev
->index
;
10231 change
= imsm_analyze_change(st
, &geo
, direction
);
10234 ret_val
= imsm_takeover(st
, &geo
);
10236 case CH_MIGRATION
: {
10237 struct imsm_update_reshape_migration
*u
= NULL
;
10239 imsm_create_metadata_update_for_migration(
10243 "Cannot prepare update\n");
10247 /* update metadata locally */
10248 imsm_update_metadata_locally(st
, u
, len
);
10249 /* and possibly remotely */
10250 if (st
->update_tail
)
10251 append_metadata_update(st
, u
, len
);
10256 case CH_ARRAY_SIZE
: {
10257 struct imsm_update_size_change
*u
= NULL
;
10259 imsm_create_metadata_update_for_size_change(
10263 "Cannot prepare update\n");
10267 /* update metadata locally */
10268 imsm_update_metadata_locally(st
, u
, len
);
10269 /* and possibly remotely */
10270 if (st
->update_tail
)
10271 append_metadata_update(st
, u
, len
);
10281 exit_imsm_reshape_super
:
10282 dprintf("imsm: reshape_super Exit code = %i\n", ret_val
);
10286 /*******************************************************************************
10287 * Function: wait_for_reshape_imsm
10288 * Description: Function writes new sync_max value and waits until
10289 * reshape process reach new position
10291 * sra : general array info
10292 * ndata : number of disks in new array's layout
10295 * 1 : there is no reshape in progress,
10297 ******************************************************************************/
10298 int wait_for_reshape_imsm(struct mdinfo
*sra
, int ndata
)
10300 int fd
= sysfs_get_fd(sra
, NULL
, "reshape_position");
10301 unsigned long long completed
;
10302 /* to_complete : new sync_max position */
10303 unsigned long long to_complete
= sra
->reshape_progress
;
10304 unsigned long long position_to_set
= to_complete
/ ndata
;
10307 dprintf("imsm: wait_for_reshape_imsm() "
10308 "cannot open reshape_position\n");
10312 if (sysfs_fd_get_ll(fd
, &completed
) < 0) {
10313 dprintf("imsm: wait_for_reshape_imsm() "
10314 "cannot read reshape_position (no reshape in progres)\n");
10319 if (completed
> to_complete
) {
10320 dprintf("imsm: wait_for_reshape_imsm() "
10321 "wrong next position to set %llu (%llu)\n",
10322 to_complete
, completed
);
10326 dprintf("Position set: %llu\n", position_to_set
);
10327 if (sysfs_set_num(sra
, NULL
, "sync_max",
10328 position_to_set
) != 0) {
10329 dprintf("imsm: wait_for_reshape_imsm() "
10330 "cannot set reshape position to %llu\n",
10341 select(fd
+1, &rfds
, NULL
, NULL
, NULL
);
10342 if (sysfs_get_str(sra
, NULL
, "sync_action",
10344 strncmp(action
, "reshape", 7) != 0)
10346 if (sysfs_fd_get_ll(fd
, &completed
) < 0) {
10347 dprintf("imsm: wait_for_reshape_imsm() "
10348 "cannot read reshape_position (in loop)\n");
10352 } while (completed
< to_complete
);
10358 /*******************************************************************************
10359 * Function: check_degradation_change
10360 * Description: Check that array hasn't become failed.
10362 * info : for sysfs access
10363 * sources : source disks descriptors
10364 * degraded: previous degradation level
10366 * degradation level
10367 ******************************************************************************/
10368 int check_degradation_change(struct mdinfo
*info
,
10372 unsigned long long new_degraded
;
10375 rv
= sysfs_get_ll(info
, NULL
, "degraded", &new_degraded
);
10376 if ((rv
== -1) || (new_degraded
!= (unsigned long long)degraded
)) {
10377 /* check each device to ensure it is still working */
10380 for (sd
= info
->devs
; sd
; sd
= sd
->next
) {
10381 if (sd
->disk
.state
& (1<<MD_DISK_FAULTY
))
10383 if (sd
->disk
.state
& (1<<MD_DISK_SYNC
)) {
10385 if (sysfs_get_str(info
,
10386 sd
, "state", sbuf
, 20) < 0 ||
10387 strstr(sbuf
, "faulty") ||
10388 strstr(sbuf
, "in_sync") == NULL
) {
10389 /* this device is dead */
10390 sd
->disk
.state
= (1<<MD_DISK_FAULTY
);
10391 if (sd
->disk
.raid_disk
>= 0 &&
10392 sources
[sd
->disk
.raid_disk
] >= 0) {
10394 sd
->disk
.raid_disk
]);
10395 sources
[sd
->disk
.raid_disk
] =
10404 return new_degraded
;
10407 /*******************************************************************************
10408 * Function: imsm_manage_reshape
10409 * Description: Function finds array under reshape and it manages reshape
10410 * process. It creates stripes backups (if required) and sets
10413 * afd : Backup handle (nattive) - not used
10414 * sra : general array info
10415 * reshape : reshape parameters - not used
10416 * st : supertype structure
10417 * blocks : size of critical section [blocks]
10418 * fds : table of source device descriptor
10419 * offsets : start of array (offest per devices)
10421 * destfd : table of destination device descriptor
10422 * destoffsets : table of destination offsets (per device)
10424 * 1 : success, reshape is done
10426 ******************************************************************************/
10427 static int imsm_manage_reshape(
10428 int afd
, struct mdinfo
*sra
, struct reshape
*reshape
,
10429 struct supertype
*st
, unsigned long backup_blocks
,
10430 int *fds
, unsigned long long *offsets
,
10431 int dests
, int *destfd
, unsigned long long *destoffsets
)
10434 struct intel_super
*super
= st
->sb
;
10435 struct intel_dev
*dv
= NULL
;
10436 struct imsm_dev
*dev
= NULL
;
10437 struct imsm_map
*map_src
;
10438 int migr_vol_qan
= 0;
10439 int ndata
, odata
; /* [bytes] */
10440 int chunk
; /* [bytes] */
10441 struct migr_record
*migr_rec
;
10443 unsigned int buf_size
; /* [bytes] */
10444 unsigned long long max_position
; /* array size [bytes] */
10445 unsigned long long next_step
; /* [blocks]/[bytes] */
10446 unsigned long long old_data_stripe_length
;
10447 unsigned long long start_src
; /* [bytes] */
10448 unsigned long long start
; /* [bytes] */
10449 unsigned long long start_buf_shift
; /* [bytes] */
10451 int source_layout
= 0;
10453 if (!fds
|| !offsets
|| !sra
)
10456 /* Find volume during the reshape */
10457 for (dv
= super
->devlist
; dv
; dv
= dv
->next
) {
10458 if (dv
->dev
->vol
.migr_type
== MIGR_GEN_MIGR
10459 && dv
->dev
->vol
.migr_state
== 1) {
10464 /* Only one volume can migrate at the same time */
10465 if (migr_vol_qan
!= 1) {
10466 fprintf(stderr
, Name
" : %s", migr_vol_qan
?
10467 "Number of migrating volumes greater than 1\n" :
10468 "There is no volume during migrationg\n");
10472 map_src
= get_imsm_map(dev
, MAP_1
);
10473 if (map_src
== NULL
)
10476 ndata
= imsm_num_data_members(dev
, MAP_0
);
10477 odata
= imsm_num_data_members(dev
, MAP_1
);
10479 chunk
= __le16_to_cpu(map_src
->blocks_per_strip
) * 512;
10480 old_data_stripe_length
= odata
* chunk
;
10482 migr_rec
= super
->migr_rec
;
10484 /* initialize migration record for start condition */
10485 if (sra
->reshape_progress
== 0)
10486 init_migr_record_imsm(st
, dev
, sra
);
10488 if (__le32_to_cpu(migr_rec
->rec_status
) != UNIT_SRC_NORMAL
) {
10489 dprintf("imsm: cannot restart migration when data "
10490 "are present in copy area.\n");
10493 /* Save checkpoint to update migration record for current
10494 * reshape position (in md). It can be farther than current
10495 * reshape position in metadata.
10497 if (save_checkpoint_imsm(st
, sra
, UNIT_SRC_NORMAL
) == 1) {
10498 /* ignore error == 2, this can mean end of reshape here
10500 dprintf("imsm: Cannot write checkpoint to "
10501 "migration record (UNIT_SRC_NORMAL, "
10502 "initial save)\n");
10507 /* size for data */
10508 buf_size
= __le32_to_cpu(migr_rec
->blocks_per_unit
) * 512;
10509 /* extend buffer size for parity disk */
10510 buf_size
+= __le32_to_cpu(migr_rec
->dest_depth_per_unit
) * 512;
10511 /* add space for stripe aligment */
10512 buf_size
+= old_data_stripe_length
;
10513 if (posix_memalign((void **)&buf
, 4096, buf_size
)) {
10514 dprintf("imsm: Cannot allocate checpoint buffer\n");
10518 max_position
= sra
->component_size
* ndata
;
10519 source_layout
= imsm_level_to_layout(map_src
->raid_level
);
10521 while (__le32_to_cpu(migr_rec
->curr_migr_unit
) <
10522 __le32_to_cpu(migr_rec
->num_migr_units
)) {
10523 /* current reshape position [blocks] */
10524 unsigned long long current_position
=
10525 __le32_to_cpu(migr_rec
->blocks_per_unit
)
10526 * __le32_to_cpu(migr_rec
->curr_migr_unit
);
10527 unsigned long long border
;
10529 /* Check that array hasn't become failed.
10531 degraded
= check_degradation_change(sra
, fds
, degraded
);
10532 if (degraded
> 1) {
10533 dprintf("imsm: Abort reshape due to degradation"
10534 " level (%i)\n", degraded
);
10538 next_step
= __le32_to_cpu(migr_rec
->blocks_per_unit
);
10540 if ((current_position
+ next_step
) > max_position
)
10541 next_step
= max_position
- current_position
;
10543 start
= current_position
* 512;
10545 /* allign reading start to old geometry */
10546 start_buf_shift
= start
% old_data_stripe_length
;
10547 start_src
= start
- start_buf_shift
;
10549 border
= (start_src
/ odata
) - (start
/ ndata
);
10551 if (border
<= __le32_to_cpu(migr_rec
->dest_depth_per_unit
)) {
10552 /* save critical stripes to buf
10553 * start - start address of current unit
10554 * to backup [bytes]
10555 * start_src - start address of current unit
10556 * to backup alligned to source array
10559 unsigned long long next_step_filler
= 0;
10560 unsigned long long copy_length
= next_step
* 512;
10562 /* allign copy area length to stripe in old geometry */
10563 next_step_filler
= ((copy_length
+ start_buf_shift
)
10564 % old_data_stripe_length
);
10565 if (next_step_filler
)
10566 next_step_filler
= (old_data_stripe_length
10567 - next_step_filler
);
10568 dprintf("save_stripes() parameters: start = %llu,"
10569 "\tstart_src = %llu,\tnext_step*512 = %llu,"
10570 "\tstart_in_buf_shift = %llu,"
10571 "\tnext_step_filler = %llu\n",
10572 start
, start_src
, copy_length
,
10573 start_buf_shift
, next_step_filler
);
10575 if (save_stripes(fds
, offsets
, map_src
->num_members
,
10576 chunk
, map_src
->raid_level
,
10577 source_layout
, 0, NULL
, start_src
,
10579 next_step_filler
+ start_buf_shift
,
10581 dprintf("imsm: Cannot save stripes"
10585 /* Convert data to destination format and store it
10586 * in backup general migration area
10588 if (save_backup_imsm(st
, dev
, sra
,
10589 buf
+ start_buf_shift
, copy_length
)) {
10590 dprintf("imsm: Cannot save stripes to "
10591 "target devices\n");
10594 if (save_checkpoint_imsm(st
, sra
,
10595 UNIT_SRC_IN_CP_AREA
)) {
10596 dprintf("imsm: Cannot write checkpoint to "
10597 "migration record (UNIT_SRC_IN_CP_AREA)\n");
10601 /* set next step to use whole border area */
10602 border
/= next_step
;
10604 next_step
*= border
;
10606 /* When data backed up, checkpoint stored,
10607 * kick the kernel to reshape unit of data
10609 next_step
= next_step
+ sra
->reshape_progress
;
10610 /* limit next step to array max position */
10611 if (next_step
> max_position
)
10612 next_step
= max_position
;
10613 sysfs_set_num(sra
, NULL
, "suspend_lo", sra
->reshape_progress
);
10614 sysfs_set_num(sra
, NULL
, "suspend_hi", next_step
);
10615 sra
->reshape_progress
= next_step
;
10617 /* wait until reshape finish */
10618 if (wait_for_reshape_imsm(sra
, ndata
) < 0) {
10619 dprintf("wait_for_reshape_imsm returned error!\n");
10623 if (save_checkpoint_imsm(st
, sra
, UNIT_SRC_NORMAL
) == 1) {
10624 /* ignore error == 2, this can mean end of reshape here
10626 dprintf("imsm: Cannot write checkpoint to "
10627 "migration record (UNIT_SRC_NORMAL)\n");
10633 /* return '1' if done */
10637 abort_reshape(sra
);
10641 #endif /* MDASSEMBLE */
10643 struct superswitch super_imsm
= {
10645 .examine_super
= examine_super_imsm
,
10646 .brief_examine_super
= brief_examine_super_imsm
,
10647 .brief_examine_subarrays
= brief_examine_subarrays_imsm
,
10648 .export_examine_super
= export_examine_super_imsm
,
10649 .detail_super
= detail_super_imsm
,
10650 .brief_detail_super
= brief_detail_super_imsm
,
10651 .write_init_super
= write_init_super_imsm
,
10652 .validate_geometry
= validate_geometry_imsm
,
10653 .add_to_super
= add_to_super_imsm
,
10654 .remove_from_super
= remove_from_super_imsm
,
10655 .detail_platform
= detail_platform_imsm
,
10656 .kill_subarray
= kill_subarray_imsm
,
10657 .update_subarray
= update_subarray_imsm
,
10658 .load_container
= load_container_imsm
,
10659 .default_geometry
= default_geometry_imsm
,
10660 .get_disk_controller_domain
= imsm_get_disk_controller_domain
,
10661 .reshape_super
= imsm_reshape_super
,
10662 .manage_reshape
= imsm_manage_reshape
,
10663 .recover_backup
= recover_backup_imsm
,
10665 .match_home
= match_home_imsm
,
10666 .uuid_from_super
= uuid_from_super_imsm
,
10667 .getinfo_super
= getinfo_super_imsm
,
10668 .getinfo_super_disks
= getinfo_super_disks_imsm
,
10669 .update_super
= update_super_imsm
,
10671 .avail_size
= avail_size_imsm
,
10672 .min_acceptable_spare_size
= min_acceptable_spare_size_imsm
,
10674 .compare_super
= compare_super_imsm
,
10676 .load_super
= load_super_imsm
,
10677 .init_super
= init_super_imsm
,
10678 .store_super
= store_super_imsm
,
10679 .free_super
= free_super_imsm
,
10680 .match_metadata_desc
= match_metadata_desc_imsm
,
10681 .container_content
= container_content_imsm
,
10689 .open_new
= imsm_open_new
,
10690 .set_array_state
= imsm_set_array_state
,
10691 .set_disk
= imsm_set_disk
,
10692 .sync_metadata
= imsm_sync_metadata
,
10693 .activate_spare
= imsm_activate_spare
,
10694 .process_update
= imsm_process_update
,
10695 .prepare_update
= imsm_prepare_update
,
10696 #endif /* MDASSEMBLE */