2 * mdadm - Intel(R) Matrix Storage Manager Support
4 * Copyright (C) 2002-2008 Intel Corporation
6 * This program is free software; you can redistribute it and/or modify it
7 * under the terms and conditions of the GNU General Public License,
8 * version 2, as published by the Free Software Foundation.
10 * This program is distributed in the hope it will be useful, but WITHOUT
11 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
12 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for
15 * You should have received a copy of the GNU General Public License along with
16 * this program; if not, write to the Free Software Foundation, Inc.,
17 * 51 Franklin St - Fifth Floor, Boston, MA 02110-1301 USA.
20 #define HAVE_STDINT_H 1
24 #include "platform-intel.h"
30 /* MPB == Metadata Parameter Block */
31 #define MPB_SIGNATURE "Intel Raid ISM Cfg Sig. "
32 #define MPB_SIG_LEN (strlen(MPB_SIGNATURE))
33 #define MPB_VERSION_RAID0 "1.0.00"
34 #define MPB_VERSION_RAID1 "1.1.00"
35 #define MPB_VERSION_MANY_VOLUMES_PER_ARRAY "1.2.00"
36 #define MPB_VERSION_3OR4_DISK_ARRAY "1.2.01"
37 #define MPB_VERSION_RAID5 "1.2.02"
38 #define MPB_VERSION_5OR6_DISK_ARRAY "1.2.04"
39 #define MPB_VERSION_CNG "1.2.06"
40 #define MPB_VERSION_ATTRIBS "1.3.00"
41 #define MAX_SIGNATURE_LENGTH 32
42 #define MAX_RAID_SERIAL_LEN 16
45 #define MPB_ATTRIB_RAID0 __cpu_to_le32(0x00000001)
47 #define MPB_ATTRIB_RAID1 __cpu_to_le32(0x00000002)
49 #define MPB_ATTRIB_RAID10 __cpu_to_le32(0x00000004)
51 #define MPB_ATTRIB_RAID1E __cpu_to_le32(0x00000008)
53 #define MPB_ATTRIB_RAID5 __cpu_to_le32(0x00000010)
54 /* supports RAID CNG */
55 #define MPB_ATTRIB_RAIDCNG __cpu_to_le32(0x00000020)
56 /* supports expanded stripe sizes of 256K, 512K and 1MB */
57 #define MPB_ATTRIB_EXP_STRIPE_SIZE __cpu_to_le32(0x00000040)
59 /* The OROM Support RST Caching of Volumes */
60 #define MPB_ATTRIB_NVM __cpu_to_le32(0x02000000)
61 /* The OROM supports creating disks greater than 2TB */
62 #define MPB_ATTRIB_2TB_DISK __cpu_to_le32(0x04000000)
63 /* The OROM supports Bad Block Management */
64 #define MPB_ATTRIB_BBM __cpu_to_le32(0x08000000)
66 /* THe OROM Supports NVM Caching of Volumes */
67 #define MPB_ATTRIB_NEVER_USE2 __cpu_to_le32(0x10000000)
68 /* The OROM supports creating volumes greater than 2TB */
69 #define MPB_ATTRIB_2TB __cpu_to_le32(0x20000000)
70 /* originally for PMP, now it's wasted b/c. Never use this bit! */
71 #define MPB_ATTRIB_NEVER_USE __cpu_to_le32(0x40000000)
72 /* Verify MPB contents against checksum after reading MPB */
73 #define MPB_ATTRIB_CHECKSUM_VERIFY __cpu_to_le32(0x80000000)
75 /* Define all supported attributes that have to be accepted by mdadm
77 #define MPB_ATTRIB_SUPPORTED (MPB_ATTRIB_CHECKSUM_VERIFY | \
79 MPB_ATTRIB_2TB_DISK | \
84 MPB_ATTRIB_EXP_STRIPE_SIZE)
86 /* Define attributes that are unused but not harmful */
87 #define MPB_ATTRIB_IGNORED (MPB_ATTRIB_NEVER_USE)
89 #define MPB_SECTOR_CNT 2210
90 #define IMSM_RESERVED_SECTORS 4096
91 #define NUM_BLOCKS_DIRTY_STRIPE_REGION 2056
92 #define SECT_PER_MB_SHIFT 11
94 /* Disk configuration info. */
95 #define IMSM_MAX_DEVICES 255
97 __u8 serial
[MAX_RAID_SERIAL_LEN
];/* 0xD8 - 0xE7 ascii serial number */
98 __u32 total_blocks_lo
; /* 0xE8 - 0xEB total blocks lo */
99 __u32 scsi_id
; /* 0xEC - 0xEF scsi ID */
100 #define SPARE_DISK __cpu_to_le32(0x01) /* Spare */
101 #define CONFIGURED_DISK __cpu_to_le32(0x02) /* Member of some RaidDev */
102 #define FAILED_DISK __cpu_to_le32(0x04) /* Permanent failure */
103 __u32 status
; /* 0xF0 - 0xF3 */
104 __u32 owner_cfg_num
; /* which config 0,1,2... owns this disk */
105 __u32 total_blocks_hi
; /* 0xF4 - 0xF5 total blocks hi */
106 #define IMSM_DISK_FILLERS 3
107 __u32 filler
[IMSM_DISK_FILLERS
]; /* 0xF5 - 0x107 MPB_DISK_FILLERS for future expansion */
110 /* map selector for map managment
116 /* RAID map configuration infos. */
118 __u32 pba_of_lba0_lo
; /* start address of partition */
119 __u32 blocks_per_member_lo
;/* blocks per member */
120 __u32 num_data_stripes_lo
; /* number of data stripes */
121 __u16 blocks_per_strip
;
122 __u8 map_state
; /* Normal, Uninitialized, Degraded, Failed */
123 #define IMSM_T_STATE_NORMAL 0
124 #define IMSM_T_STATE_UNINITIALIZED 1
125 #define IMSM_T_STATE_DEGRADED 2
126 #define IMSM_T_STATE_FAILED 3
128 #define IMSM_T_RAID0 0
129 #define IMSM_T_RAID1 1
130 #define IMSM_T_RAID5 5 /* since metadata version 1.2.02 ? */
131 __u8 num_members
; /* number of member disks */
132 __u8 num_domains
; /* number of parity domains */
133 __u8 failed_disk_num
; /* valid only when state is degraded */
135 __u32 pba_of_lba0_hi
;
136 __u32 blocks_per_member_hi
;
137 __u32 num_data_stripes_hi
;
138 __u32 filler
[4]; /* expansion area */
139 #define IMSM_ORD_REBUILD (1 << 24)
140 __u32 disk_ord_tbl
[1]; /* disk_ord_tbl[num_members],
141 * top byte contains some flags
143 } __attribute__ ((packed
));
146 __u32 curr_migr_unit
;
147 __u32 checkpoint_id
; /* id to access curr_migr_unit */
148 __u8 migr_state
; /* Normal or Migrating */
150 #define MIGR_REBUILD 1
151 #define MIGR_VERIFY 2 /* analagous to echo check > sync_action */
152 #define MIGR_GEN_MIGR 3
153 #define MIGR_STATE_CHANGE 4
154 #define MIGR_REPAIR 5
155 __u8 migr_type
; /* Initializing, Rebuilding, ... */
157 __u8 fs_state
; /* fast-sync state for CnG (0xff == disabled) */
158 __u16 verify_errors
; /* number of mismatches */
159 __u16 bad_blocks
; /* number of bad blocks during verify */
161 struct imsm_map map
[1];
162 /* here comes another one if migr_state */
163 } __attribute__ ((packed
));
166 __u8 volume
[MAX_RAID_SERIAL_LEN
];
169 #define DEV_BOOTABLE __cpu_to_le32(0x01)
170 #define DEV_BOOT_DEVICE __cpu_to_le32(0x02)
171 #define DEV_READ_COALESCING __cpu_to_le32(0x04)
172 #define DEV_WRITE_COALESCING __cpu_to_le32(0x08)
173 #define DEV_LAST_SHUTDOWN_DIRTY __cpu_to_le32(0x10)
174 #define DEV_HIDDEN_AT_BOOT __cpu_to_le32(0x20)
175 #define DEV_CURRENTLY_HIDDEN __cpu_to_le32(0x40)
176 #define DEV_VERIFY_AND_FIX __cpu_to_le32(0x80)
177 #define DEV_MAP_STATE_UNINIT __cpu_to_le32(0x100)
178 #define DEV_NO_AUTO_RECOVERY __cpu_to_le32(0x200)
179 #define DEV_CLONE_N_GO __cpu_to_le32(0x400)
180 #define DEV_CLONE_MAN_SYNC __cpu_to_le32(0x800)
181 #define DEV_CNG_MASTER_DISK_NUM __cpu_to_le32(0x1000)
182 __u32 status
; /* Persistent RaidDev status */
183 __u32 reserved_blocks
; /* Reserved blocks at beginning of volume */
187 __u8 cng_master_disk
;
191 #define IMSM_DEV_FILLERS 10
192 __u32 filler
[IMSM_DEV_FILLERS
];
194 } __attribute__ ((packed
));
197 __u8 sig
[MAX_SIGNATURE_LENGTH
]; /* 0x00 - 0x1F */
198 __u32 check_sum
; /* 0x20 - 0x23 MPB Checksum */
199 __u32 mpb_size
; /* 0x24 - 0x27 Size of MPB */
200 __u32 family_num
; /* 0x28 - 0x2B Checksum from first time this config was written */
201 __u32 generation_num
; /* 0x2C - 0x2F Incremented each time this array's MPB is written */
202 __u32 error_log_size
; /* 0x30 - 0x33 in bytes */
203 __u32 attributes
; /* 0x34 - 0x37 */
204 __u8 num_disks
; /* 0x38 Number of configured disks */
205 __u8 num_raid_devs
; /* 0x39 Number of configured volumes */
206 __u8 error_log_pos
; /* 0x3A */
207 __u8 fill
[1]; /* 0x3B */
208 __u32 cache_size
; /* 0x3c - 0x40 in mb */
209 __u32 orig_family_num
; /* 0x40 - 0x43 original family num */
210 __u32 pwr_cycle_count
; /* 0x44 - 0x47 simulated power cycle count for array */
211 __u32 bbm_log_size
; /* 0x48 - 0x4B - size of bad Block Mgmt Log in bytes */
212 #define IMSM_FILLERS 35
213 __u32 filler
[IMSM_FILLERS
]; /* 0x4C - 0xD7 RAID_MPB_FILLERS */
214 struct imsm_disk disk
[1]; /* 0xD8 diskTbl[numDisks] */
215 /* here comes imsm_dev[num_raid_devs] */
216 /* here comes BBM logs */
217 } __attribute__ ((packed
));
219 #define BBM_LOG_MAX_ENTRIES 254
221 struct bbm_log_entry
{
222 __u64 defective_block_start
;
223 #define UNREADABLE 0xFFFFFFFF
224 __u32 spare_block_offset
;
225 __u16 remapped_marked_count
;
227 } __attribute__ ((__packed__
));
230 __u32 signature
; /* 0xABADB10C */
232 __u32 reserved_spare_block_count
; /* 0 */
233 __u32 reserved
; /* 0xFFFF */
234 __u64 first_spare_lba
;
235 struct bbm_log_entry mapped_block_entries
[BBM_LOG_MAX_ENTRIES
];
236 } __attribute__ ((__packed__
));
240 static char *map_state_str
[] = { "normal", "uninitialized", "degraded", "failed" };
243 #define RAID_DISK_RESERVED_BLOCKS_IMSM_HI 2209
245 #define GEN_MIGR_AREA_SIZE 2048 /* General Migration Copy Area size in blocks */
247 #define MIGR_REC_BUF_SIZE 512 /* size of migr_record i/o buffer */
248 #define MIGR_REC_POSITION 512 /* migr_record position offset on disk,
249 * MIGR_REC_BUF_SIZE <= MIGR_REC_POSITION
253 #define UNIT_SRC_NORMAL 0 /* Source data for curr_migr_unit must
254 * be recovered using srcMap */
255 #define UNIT_SRC_IN_CP_AREA 1 /* Source data for curr_migr_unit has
256 * already been migrated and must
257 * be recovered from checkpoint area */
259 __u32 rec_status
; /* Status used to determine how to restart
260 * migration in case it aborts
262 __u32 curr_migr_unit
; /* 0..numMigrUnits-1 */
263 __u32 family_num
; /* Family number of MPB
264 * containing the RaidDev
265 * that is migrating */
266 __u32 ascending_migr
; /* True if migrating in increasing
268 __u32 blocks_per_unit
; /* Num disk blocks per unit of operation */
269 __u32 dest_depth_per_unit
; /* Num member blocks each destMap
271 * advances per unit-of-operation */
272 __u32 ckpt_area_pba
; /* Pba of first block of ckpt copy area */
273 __u32 dest_1st_member_lba
; /* First member lba on first
274 * stripe of destination */
275 __u32 num_migr_units
; /* Total num migration units-of-op */
276 __u32 post_migr_vol_cap
; /* Size of volume after
277 * migration completes */
278 __u32 post_migr_vol_cap_hi
; /* Expansion space for LBA64 */
279 __u32 ckpt_read_disk_num
; /* Which member disk in destSubMap[0] the
280 * migration ckpt record was read from
281 * (for recovered migrations) */
282 } __attribute__ ((__packed__
));
287 * 2: metadata does not match
295 struct md_list
*next
;
298 #define pr_vrb(fmt, arg...) (void) (verbose && fprintf(stderr, Name fmt, ##arg))
300 static __u8
migr_type(struct imsm_dev
*dev
)
302 if (dev
->vol
.migr_type
== MIGR_VERIFY
&&
303 dev
->status
& DEV_VERIFY_AND_FIX
)
306 return dev
->vol
.migr_type
;
309 static void set_migr_type(struct imsm_dev
*dev
, __u8 migr_type
)
311 /* for compatibility with older oroms convert MIGR_REPAIR, into
312 * MIGR_VERIFY w/ DEV_VERIFY_AND_FIX status
314 if (migr_type
== MIGR_REPAIR
) {
315 dev
->vol
.migr_type
= MIGR_VERIFY
;
316 dev
->status
|= DEV_VERIFY_AND_FIX
;
318 dev
->vol
.migr_type
= migr_type
;
319 dev
->status
&= ~DEV_VERIFY_AND_FIX
;
323 static unsigned int sector_count(__u32 bytes
)
325 return ROUND_UP(bytes
, 512) / 512;
328 static unsigned int mpb_sectors(struct imsm_super
*mpb
)
330 return sector_count(__le32_to_cpu(mpb
->mpb_size
));
334 struct imsm_dev
*dev
;
335 struct intel_dev
*next
;
340 enum sys_dev_type type
;
343 struct intel_hba
*next
;
350 /* internal representation of IMSM metadata */
353 void *buf
; /* O_DIRECT buffer for reading/writing metadata */
354 struct imsm_super
*anchor
; /* immovable parameters */
357 void *migr_rec_buf
; /* buffer for I/O operations */
358 struct migr_record
*migr_rec
; /* migration record */
360 int clean_migration_record_by_mdmon
; /* when reshape is switched to next
361 array, it indicates that mdmon is allowed to clean migration
363 size_t len
; /* size of the 'buf' allocation */
364 void *next_buf
; /* for realloc'ing buf from the manager */
366 int updates_pending
; /* count of pending updates for mdmon */
367 int current_vol
; /* index of raid device undergoing creation */
368 unsigned long long create_offset
; /* common start for 'current_vol' */
369 __u32 random
; /* random data for seeding new family numbers */
370 struct intel_dev
*devlist
;
374 __u8 serial
[MAX_RAID_SERIAL_LEN
];
377 struct imsm_disk disk
;
380 struct extent
*e
; /* for determining freespace @ create */
381 int raiddisk
; /* slot to fill in autolayout */
383 } *disks
, *current_disk
;
384 struct dl
*disk_mgmt_list
; /* list of disks to add/remove while mdmon
386 struct dl
*missing
; /* disks removed while we weren't looking */
387 struct bbm_log
*bbm_log
;
388 struct intel_hba
*hba
; /* device path of the raid controller for this metadata */
389 const struct imsm_orom
*orom
; /* platform firmware support */
390 struct intel_super
*next
; /* (temp) list for disambiguating family_num */
394 struct imsm_disk disk
;
395 #define IMSM_UNKNOWN_OWNER (-1)
397 struct intel_disk
*next
;
401 unsigned long long start
, size
;
404 /* definitions of reshape process types */
405 enum imsm_reshape_type
{
411 /* definition of messages passed to imsm_process_update */
412 enum imsm_update_type
{
413 update_activate_spare
,
417 update_add_remove_disk
,
418 update_reshape_container_disks
,
419 update_reshape_migration
,
421 update_general_migration_checkpoint
,
425 struct imsm_update_activate_spare
{
426 enum imsm_update_type type
;
430 struct imsm_update_activate_spare
*next
;
443 enum takeover_direction
{
447 struct imsm_update_takeover
{
448 enum imsm_update_type type
;
450 enum takeover_direction direction
;
453 struct imsm_update_reshape
{
454 enum imsm_update_type type
;
458 int new_disks
[1]; /* new_raid_disks - old_raid_disks makedev number */
461 struct imsm_update_reshape_migration
{
462 enum imsm_update_type type
;
465 /* fields for array migration changes
472 int new_disks
[1]; /* new_raid_disks - old_raid_disks makedev number */
475 struct imsm_update_size_change
{
476 enum imsm_update_type type
;
481 struct imsm_update_general_migration_checkpoint
{
482 enum imsm_update_type type
;
483 __u32 curr_migr_unit
;
487 __u8 serial
[MAX_RAID_SERIAL_LEN
];
490 struct imsm_update_create_array
{
491 enum imsm_update_type type
;
496 struct imsm_update_kill_array
{
497 enum imsm_update_type type
;
501 struct imsm_update_rename_array
{
502 enum imsm_update_type type
;
503 __u8 name
[MAX_RAID_SERIAL_LEN
];
507 struct imsm_update_add_remove_disk
{
508 enum imsm_update_type type
;
512 static const char *_sys_dev_type
[] = {
513 [SYS_DEV_UNKNOWN
] = "Unknown",
514 [SYS_DEV_SAS
] = "SAS",
515 [SYS_DEV_SATA
] = "SATA"
518 const char *get_sys_dev_type(enum sys_dev_type type
)
520 if (type
>= SYS_DEV_MAX
)
521 type
= SYS_DEV_UNKNOWN
;
523 return _sys_dev_type
[type
];
526 static struct intel_hba
* alloc_intel_hba(struct sys_dev
*device
)
528 struct intel_hba
*result
= malloc(sizeof(*result
));
530 result
->type
= device
->type
;
531 result
->path
= strdup(device
->path
);
533 if (result
->path
&& (result
->pci_id
= strrchr(result
->path
, '/')) != NULL
)
539 static struct intel_hba
* find_intel_hba(struct intel_hba
*hba
, struct sys_dev
*device
)
541 struct intel_hba
*result
=NULL
;
542 for (result
= hba
; result
; result
= result
->next
) {
543 if (result
->type
== device
->type
&& strcmp(result
->path
, device
->path
) == 0)
549 static int attach_hba_to_super(struct intel_super
*super
, struct sys_dev
*device
)
551 struct intel_hba
*hba
;
553 /* check if disk attached to Intel HBA */
554 hba
= find_intel_hba(super
->hba
, device
);
557 /* Check if HBA is already attached to super */
558 if (super
->hba
== NULL
) {
559 super
->hba
= alloc_intel_hba(device
);
564 /* Intel metadata allows for all disks attached to the same type HBA.
565 * Do not sypport odf HBA types mixing
567 if (device
->type
!= hba
->type
)
573 hba
->next
= alloc_intel_hba(device
);
577 static struct sys_dev
* find_disk_attached_hba(int fd
, const char *devname
)
579 struct sys_dev
*list
, *elem
, *prev
;
582 if ((list
= find_intel_devices()) == NULL
)
586 disk_path
= (char *) devname
;
588 disk_path
= diskfd_to_devpath(fd
);
595 for (prev
= NULL
, elem
= list
; elem
; prev
= elem
, elem
= elem
->next
) {
596 if (path_attached_to_hba(disk_path
, elem
->path
)) {
600 prev
->next
= elem
->next
;
602 if (disk_path
!= devname
)
608 if (disk_path
!= devname
)
616 static int find_intel_hba_capability(int fd
, struct intel_super
*super
,
619 static struct supertype
*match_metadata_desc_imsm(char *arg
)
621 struct supertype
*st
;
623 if (strcmp(arg
, "imsm") != 0 &&
624 strcmp(arg
, "default") != 0
628 st
= malloc(sizeof(*st
));
631 memset(st
, 0, sizeof(*st
));
632 st
->container_dev
= NoMdDev
;
633 st
->ss
= &super_imsm
;
634 st
->max_devs
= IMSM_MAX_DEVICES
;
635 st
->minor_version
= 0;
641 static __u8
*get_imsm_version(struct imsm_super
*mpb
)
643 return &mpb
->sig
[MPB_SIG_LEN
];
647 /* retrieve a disk directly from the anchor when the anchor is known to be
648 * up-to-date, currently only at load time
650 static struct imsm_disk
*__get_imsm_disk(struct imsm_super
*mpb
, __u8 index
)
652 if (index
>= mpb
->num_disks
)
654 return &mpb
->disk
[index
];
657 /* retrieve the disk description based on a index of the disk
660 static struct dl
*get_imsm_dl_disk(struct intel_super
*super
, __u8 index
)
664 for (d
= super
->disks
; d
; d
= d
->next
)
665 if (d
->index
== index
)
670 /* retrieve a disk from the parsed metadata */
671 static struct imsm_disk
*get_imsm_disk(struct intel_super
*super
, __u8 index
)
675 dl
= get_imsm_dl_disk(super
, index
);
682 /* generate a checksum directly from the anchor when the anchor is known to be
683 * up-to-date, currently only at load or write_super after coalescing
685 static __u32
__gen_imsm_checksum(struct imsm_super
*mpb
)
687 __u32 end
= mpb
->mpb_size
/ sizeof(end
);
688 __u32
*p
= (__u32
*) mpb
;
692 sum
+= __le32_to_cpu(*p
);
696 return sum
- __le32_to_cpu(mpb
->check_sum
);
699 static size_t sizeof_imsm_map(struct imsm_map
*map
)
701 return sizeof(struct imsm_map
) + sizeof(__u32
) * (map
->num_members
- 1);
704 struct imsm_map
*get_imsm_map(struct imsm_dev
*dev
, int second_map
)
706 /* A device can have 2 maps if it is in the middle of a migration.
708 * MAP_0 - we return the first map
709 * MAP_1 - we return the second map if it exists, else NULL
710 * MAP_X - we return the second map if it exists, else the first
712 struct imsm_map
*map
= &dev
->vol
.map
[0];
713 struct imsm_map
*map2
= NULL
;
715 if (dev
->vol
.migr_state
)
716 map2
= (void *)map
+ sizeof_imsm_map(map
);
718 switch (second_map
) {
735 /* return the size of the device.
736 * migr_state increases the returned size if map[0] were to be duplicated
738 static size_t sizeof_imsm_dev(struct imsm_dev
*dev
, int migr_state
)
740 size_t size
= sizeof(*dev
) - sizeof(struct imsm_map
) +
741 sizeof_imsm_map(get_imsm_map(dev
, MAP_0
));
743 /* migrating means an additional map */
744 if (dev
->vol
.migr_state
)
745 size
+= sizeof_imsm_map(get_imsm_map(dev
, MAP_1
));
747 size
+= sizeof_imsm_map(get_imsm_map(dev
, MAP_0
));
753 /* retrieve disk serial number list from a metadata update */
754 static struct disk_info
*get_disk_info(struct imsm_update_create_array
*update
)
757 struct disk_info
*inf
;
759 inf
= u
+ sizeof(*update
) - sizeof(struct imsm_dev
) +
760 sizeof_imsm_dev(&update
->dev
, 0);
766 static struct imsm_dev
*__get_imsm_dev(struct imsm_super
*mpb
, __u8 index
)
772 if (index
>= mpb
->num_raid_devs
)
775 /* devices start after all disks */
776 offset
= ((void *) &mpb
->disk
[mpb
->num_disks
]) - _mpb
;
778 for (i
= 0; i
<= index
; i
++)
780 return _mpb
+ offset
;
782 offset
+= sizeof_imsm_dev(_mpb
+ offset
, 0);
787 static struct imsm_dev
*get_imsm_dev(struct intel_super
*super
, __u8 index
)
789 struct intel_dev
*dv
;
791 if (index
>= super
->anchor
->num_raid_devs
)
793 for (dv
= super
->devlist
; dv
; dv
= dv
->next
)
794 if (dv
->index
== index
)
801 * == MAP_0 get first map
802 * == MAP_1 get second map
803 * == MAP_X than get map according to the current migr_state
805 static __u32
get_imsm_ord_tbl_ent(struct imsm_dev
*dev
,
809 struct imsm_map
*map
;
811 map
= get_imsm_map(dev
, second_map
);
813 /* top byte identifies disk under rebuild */
814 return __le32_to_cpu(map
->disk_ord_tbl
[slot
]);
817 #define ord_to_idx(ord) (((ord) << 8) >> 8)
818 static __u32
get_imsm_disk_idx(struct imsm_dev
*dev
, int slot
, int second_map
)
820 __u32 ord
= get_imsm_ord_tbl_ent(dev
, slot
, second_map
);
822 return ord_to_idx(ord
);
825 static void set_imsm_ord_tbl_ent(struct imsm_map
*map
, int slot
, __u32 ord
)
827 map
->disk_ord_tbl
[slot
] = __cpu_to_le32(ord
);
830 static int get_imsm_disk_slot(struct imsm_map
*map
, unsigned idx
)
835 for (slot
= 0; slot
< map
->num_members
; slot
++) {
836 ord
= __le32_to_cpu(map
->disk_ord_tbl
[slot
]);
837 if (ord_to_idx(ord
) == idx
)
844 static int get_imsm_raid_level(struct imsm_map
*map
)
846 if (map
->raid_level
== 1) {
847 if (map
->num_members
== 2)
853 return map
->raid_level
;
856 static int cmp_extent(const void *av
, const void *bv
)
858 const struct extent
*a
= av
;
859 const struct extent
*b
= bv
;
860 if (a
->start
< b
->start
)
862 if (a
->start
> b
->start
)
867 static int count_memberships(struct dl
*dl
, struct intel_super
*super
)
872 for (i
= 0; i
< super
->anchor
->num_raid_devs
; i
++) {
873 struct imsm_dev
*dev
= get_imsm_dev(super
, i
);
874 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
876 if (get_imsm_disk_slot(map
, dl
->index
) >= 0)
883 static __u32
imsm_min_reserved_sectors(struct intel_super
*super
);
885 static int split_ull(unsigned long long n
, __u32
*lo
, __u32
*hi
)
887 if (lo
== 0 || hi
== 0)
889 *lo
= __le32_to_cpu((unsigned)n
);
890 *hi
= __le32_to_cpu((unsigned)(n
>> 32));
894 static unsigned long long join_u32(__u32 lo
, __u32 hi
)
896 return (unsigned long long)__le32_to_cpu(lo
) |
897 (((unsigned long long)__le32_to_cpu(hi
)) << 32);
900 static unsigned long long total_blocks(struct imsm_disk
*disk
)
904 return join_u32(disk
->total_blocks_lo
, disk
->total_blocks_hi
);
907 static unsigned long long pba_of_lba0(struct imsm_map
*map
)
911 return join_u32(map
->pba_of_lba0_lo
, map
->pba_of_lba0_hi
);
914 static unsigned long long blocks_per_member(struct imsm_map
*map
)
918 return join_u32(map
->blocks_per_member_lo
, map
->blocks_per_member_hi
);
921 static unsigned long long num_data_stripes(struct imsm_map
*map
)
925 return join_u32(map
->num_data_stripes_lo
, map
->num_data_stripes_hi
);
928 static void set_total_blocks(struct imsm_disk
*disk
, unsigned long long n
)
930 split_ull(n
, &disk
->total_blocks_lo
, &disk
->total_blocks_hi
);
933 static void set_pba_of_lba0(struct imsm_map
*map
, unsigned long long n
)
935 split_ull(n
, &map
->pba_of_lba0_lo
, &map
->pba_of_lba0_hi
);
938 static void set_blocks_per_member(struct imsm_map
*map
, unsigned long long n
)
940 split_ull(n
, &map
->blocks_per_member_lo
, &map
->blocks_per_member_hi
);
943 static void set_num_data_stripes(struct imsm_map
*map
, unsigned long long n
)
945 split_ull(n
, &map
->num_data_stripes_lo
, &map
->num_data_stripes_hi
);
948 static struct extent
*get_extents(struct intel_super
*super
, struct dl
*dl
)
950 /* find a list of used extents on the given physical device */
951 struct extent
*rv
, *e
;
953 int memberships
= count_memberships(dl
, super
);
956 /* trim the reserved area for spares, so they can join any array
957 * regardless of whether the OROM has assigned sectors from the
958 * IMSM_RESERVED_SECTORS region
961 reservation
= imsm_min_reserved_sectors(super
);
963 reservation
= MPB_SECTOR_CNT
+ IMSM_RESERVED_SECTORS
;
965 rv
= malloc(sizeof(struct extent
) * (memberships
+ 1));
970 for (i
= 0; i
< super
->anchor
->num_raid_devs
; i
++) {
971 struct imsm_dev
*dev
= get_imsm_dev(super
, i
);
972 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
974 if (get_imsm_disk_slot(map
, dl
->index
) >= 0) {
975 e
->start
= pba_of_lba0(map
);
976 e
->size
= blocks_per_member(map
);
980 qsort(rv
, memberships
, sizeof(*rv
), cmp_extent
);
982 /* determine the start of the metadata
983 * when no raid devices are defined use the default
984 * ...otherwise allow the metadata to truncate the value
985 * as is the case with older versions of imsm
988 struct extent
*last
= &rv
[memberships
- 1];
989 unsigned long long remainder
;
991 remainder
= total_blocks(&dl
->disk
) - (last
->start
+ last
->size
);
992 /* round down to 1k block to satisfy precision of the kernel
996 /* make sure remainder is still sane */
997 if (remainder
< (unsigned)ROUND_UP(super
->len
, 512) >> 9)
998 remainder
= ROUND_UP(super
->len
, 512) >> 9;
999 if (reservation
> remainder
)
1000 reservation
= remainder
;
1002 e
->start
= total_blocks(&dl
->disk
) - reservation
;
1007 /* try to determine how much space is reserved for metadata from
1008 * the last get_extents() entry, otherwise fallback to the
1011 static __u32
imsm_reserved_sectors(struct intel_super
*super
, struct dl
*dl
)
1017 /* for spares just return a minimal reservation which will grow
1018 * once the spare is picked up by an array
1020 if (dl
->index
== -1)
1021 return MPB_SECTOR_CNT
;
1023 e
= get_extents(super
, dl
);
1025 return MPB_SECTOR_CNT
+ IMSM_RESERVED_SECTORS
;
1027 /* scroll to last entry */
1028 for (i
= 0; e
[i
].size
; i
++)
1031 rv
= total_blocks(&dl
->disk
) - e
[i
].start
;
1038 static int is_spare(struct imsm_disk
*disk
)
1040 return (disk
->status
& SPARE_DISK
) == SPARE_DISK
;
1043 static int is_configured(struct imsm_disk
*disk
)
1045 return (disk
->status
& CONFIGURED_DISK
) == CONFIGURED_DISK
;
1048 static int is_failed(struct imsm_disk
*disk
)
1050 return (disk
->status
& FAILED_DISK
) == FAILED_DISK
;
1053 /* try to determine how much space is reserved for metadata from
1054 * the last get_extents() entry on the smallest active disk,
1055 * otherwise fallback to the default
1057 static __u32
imsm_min_reserved_sectors(struct intel_super
*super
)
1061 unsigned long long min_active
;
1063 __u32 rv
= MPB_SECTOR_CNT
+ IMSM_RESERVED_SECTORS
;
1064 struct dl
*dl
, *dl_min
= NULL
;
1070 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
1073 unsigned long long blocks
= total_blocks(&dl
->disk
);
1074 if (blocks
< min_active
|| min_active
== 0) {
1076 min_active
= blocks
;
1082 /* find last lba used by subarrays on the smallest active disk */
1083 e
= get_extents(super
, dl_min
);
1086 for (i
= 0; e
[i
].size
; i
++)
1089 remainder
= min_active
- e
[i
].start
;
1092 /* to give priority to recovery we should not require full
1093 IMSM_RESERVED_SECTORS from the spare */
1094 rv
= MPB_SECTOR_CNT
+ NUM_BLOCKS_DIRTY_STRIPE_REGION
;
1096 /* if real reservation is smaller use that value */
1097 return (remainder
< rv
) ? remainder
: rv
;
1100 /* Return minimum size of a spare that can be used in this array*/
1101 static unsigned long long min_acceptable_spare_size_imsm(struct supertype
*st
)
1103 struct intel_super
*super
= st
->sb
;
1107 unsigned long long rv
= 0;
1111 /* find first active disk in array */
1113 while (dl
&& (is_failed(&dl
->disk
) || dl
->index
== -1))
1117 /* find last lba used by subarrays */
1118 e
= get_extents(super
, dl
);
1121 for (i
= 0; e
[i
].size
; i
++)
1124 rv
= e
[i
-1].start
+ e
[i
-1].size
;
1127 /* add the amount of space needed for metadata */
1128 rv
= rv
+ imsm_min_reserved_sectors(super
);
1133 static int is_gen_migration(struct imsm_dev
*dev
);
1136 static __u64
blocks_per_migr_unit(struct intel_super
*super
,
1137 struct imsm_dev
*dev
);
1139 static void print_imsm_dev(struct intel_super
*super
,
1140 struct imsm_dev
*dev
,
1146 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
1147 struct imsm_map
*map2
= get_imsm_map(dev
, MAP_1
);
1151 printf("[%.16s]:\n", dev
->volume
);
1152 printf(" UUID : %s\n", uuid
);
1153 printf(" RAID Level : %d", get_imsm_raid_level(map
));
1155 printf(" <-- %d", get_imsm_raid_level(map2
));
1157 printf(" Members : %d", map
->num_members
);
1159 printf(" <-- %d", map2
->num_members
);
1161 printf(" Slots : [");
1162 for (i
= 0; i
< map
->num_members
; i
++) {
1163 ord
= get_imsm_ord_tbl_ent(dev
, i
, MAP_0
);
1164 printf("%s", ord
& IMSM_ORD_REBUILD
? "_" : "U");
1169 for (i
= 0; i
< map2
->num_members
; i
++) {
1170 ord
= get_imsm_ord_tbl_ent(dev
, i
, MAP_1
);
1171 printf("%s", ord
& IMSM_ORD_REBUILD
? "_" : "U");
1176 printf(" Failed disk : ");
1177 if (map
->failed_disk_num
== 0xff)
1180 printf("%i", map
->failed_disk_num
);
1182 slot
= get_imsm_disk_slot(map
, disk_idx
);
1184 ord
= get_imsm_ord_tbl_ent(dev
, slot
, MAP_X
);
1185 printf(" This Slot : %d%s\n", slot
,
1186 ord
& IMSM_ORD_REBUILD
? " (out-of-sync)" : "");
1188 printf(" This Slot : ?\n");
1189 sz
= __le32_to_cpu(dev
->size_high
);
1191 sz
+= __le32_to_cpu(dev
->size_low
);
1192 printf(" Array Size : %llu%s\n", (unsigned long long)sz
,
1193 human_size(sz
* 512));
1194 sz
= blocks_per_member(map
);
1195 printf(" Per Dev Size : %llu%s\n", (unsigned long long)sz
,
1196 human_size(sz
* 512));
1197 printf(" Sector Offset : %llu\n",
1199 printf(" Num Stripes : %llu\n",
1200 num_data_stripes(map
));
1201 printf(" Chunk Size : %u KiB",
1202 __le16_to_cpu(map
->blocks_per_strip
) / 2);
1204 printf(" <-- %u KiB",
1205 __le16_to_cpu(map2
->blocks_per_strip
) / 2);
1207 printf(" Reserved : %d\n", __le32_to_cpu(dev
->reserved_blocks
));
1208 printf(" Migrate State : ");
1209 if (dev
->vol
.migr_state
) {
1210 if (migr_type(dev
) == MIGR_INIT
)
1211 printf("initialize\n");
1212 else if (migr_type(dev
) == MIGR_REBUILD
)
1213 printf("rebuild\n");
1214 else if (migr_type(dev
) == MIGR_VERIFY
)
1216 else if (migr_type(dev
) == MIGR_GEN_MIGR
)
1217 printf("general migration\n");
1218 else if (migr_type(dev
) == MIGR_STATE_CHANGE
)
1219 printf("state change\n");
1220 else if (migr_type(dev
) == MIGR_REPAIR
)
1223 printf("<unknown:%d>\n", migr_type(dev
));
1226 printf(" Map State : %s", map_state_str
[map
->map_state
]);
1227 if (dev
->vol
.migr_state
) {
1228 struct imsm_map
*map
= get_imsm_map(dev
, MAP_1
);
1230 printf(" <-- %s", map_state_str
[map
->map_state
]);
1231 printf("\n Checkpoint : %u ",
1232 __le32_to_cpu(dev
->vol
.curr_migr_unit
));
1233 if ((is_gen_migration(dev
)) && ((slot
> 1) || (slot
< 0)))
1236 printf("(%llu)", (unsigned long long)
1237 blocks_per_migr_unit(super
, dev
));
1240 printf(" Dirty State : %s\n", dev
->vol
.dirty
? "dirty" : "clean");
1243 static void print_imsm_disk(struct imsm_disk
*disk
, int index
, __u32 reserved
)
1245 char str
[MAX_RAID_SERIAL_LEN
+ 1];
1248 if (index
< -1 || !disk
)
1252 snprintf(str
, MAX_RAID_SERIAL_LEN
+ 1, "%s", disk
->serial
);
1254 printf(" Disk%02d Serial : %s\n", index
, str
);
1256 printf(" Disk Serial : %s\n", str
);
1257 printf(" State :%s%s%s\n", is_spare(disk
) ? " spare" : "",
1258 is_configured(disk
) ? " active" : "",
1259 is_failed(disk
) ? " failed" : "");
1260 printf(" Id : %08x\n", __le32_to_cpu(disk
->scsi_id
));
1261 sz
= total_blocks(disk
) - reserved
;
1262 printf(" Usable Size : %llu%s\n", (unsigned long long)sz
,
1263 human_size(sz
* 512));
1266 void examine_migr_rec_imsm(struct intel_super
*super
)
1268 struct migr_record
*migr_rec
= super
->migr_rec
;
1269 struct imsm_super
*mpb
= super
->anchor
;
1272 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
1273 struct imsm_dev
*dev
= __get_imsm_dev(mpb
, i
);
1274 struct imsm_map
*map
;
1277 if (is_gen_migration(dev
) == 0)
1280 printf("\nMigration Record Information:");
1282 /* first map under migration */
1283 map
= get_imsm_map(dev
, MAP_0
);
1285 slot
= get_imsm_disk_slot(map
, super
->disks
->index
);
1286 if ((map
== NULL
) || (slot
> 1) || (slot
< 0)) {
1287 printf(" Empty\n ");
1288 printf("Examine one of first two disks in array\n");
1291 printf("\n Status : ");
1292 if (__le32_to_cpu(migr_rec
->rec_status
) == UNIT_SRC_NORMAL
)
1295 printf("Contains Data\n");
1296 printf(" Current Unit : %u\n",
1297 __le32_to_cpu(migr_rec
->curr_migr_unit
));
1298 printf(" Family : %u\n",
1299 __le32_to_cpu(migr_rec
->family_num
));
1300 printf(" Ascending : %u\n",
1301 __le32_to_cpu(migr_rec
->ascending_migr
));
1302 printf(" Blocks Per Unit : %u\n",
1303 __le32_to_cpu(migr_rec
->blocks_per_unit
));
1304 printf(" Dest. Depth Per Unit : %u\n",
1305 __le32_to_cpu(migr_rec
->dest_depth_per_unit
));
1306 printf(" Checkpoint Area pba : %u\n",
1307 __le32_to_cpu(migr_rec
->ckpt_area_pba
));
1308 printf(" First member lba : %u\n",
1309 __le32_to_cpu(migr_rec
->dest_1st_member_lba
));
1310 printf(" Total Number of Units : %u\n",
1311 __le32_to_cpu(migr_rec
->num_migr_units
));
1312 printf(" Size of volume : %u\n",
1313 __le32_to_cpu(migr_rec
->post_migr_vol_cap
));
1314 printf(" Expansion space for LBA64 : %u\n",
1315 __le32_to_cpu(migr_rec
->post_migr_vol_cap_hi
));
1316 printf(" Record was read from : %u\n",
1317 __le32_to_cpu(migr_rec
->ckpt_read_disk_num
));
1322 #endif /* MDASSEMBLE */
1323 /*******************************************************************************
1324 * function: imsm_check_attributes
1325 * Description: Function checks if features represented by attributes flags
1326 * are supported by mdadm.
1328 * attributes - Attributes read from metadata
1330 * 0 - passed attributes contains unsupported features flags
1331 * 1 - all features are supported
1332 ******************************************************************************/
1333 static int imsm_check_attributes(__u32 attributes
)
1336 __u32 not_supported
= MPB_ATTRIB_SUPPORTED
^0xffffffff;
1338 not_supported
&= ~MPB_ATTRIB_IGNORED
;
1340 not_supported
&= attributes
;
1341 if (not_supported
) {
1342 fprintf(stderr
, Name
"(IMSM): Unsupported attributes : %x\n",
1343 (unsigned)__le32_to_cpu(not_supported
));
1344 if (not_supported
& MPB_ATTRIB_CHECKSUM_VERIFY
) {
1345 dprintf("\t\tMPB_ATTRIB_CHECKSUM_VERIFY \n");
1346 not_supported
^= MPB_ATTRIB_CHECKSUM_VERIFY
;
1348 if (not_supported
& MPB_ATTRIB_2TB
) {
1349 dprintf("\t\tMPB_ATTRIB_2TB\n");
1350 not_supported
^= MPB_ATTRIB_2TB
;
1352 if (not_supported
& MPB_ATTRIB_RAID0
) {
1353 dprintf("\t\tMPB_ATTRIB_RAID0\n");
1354 not_supported
^= MPB_ATTRIB_RAID0
;
1356 if (not_supported
& MPB_ATTRIB_RAID1
) {
1357 dprintf("\t\tMPB_ATTRIB_RAID1\n");
1358 not_supported
^= MPB_ATTRIB_RAID1
;
1360 if (not_supported
& MPB_ATTRIB_RAID10
) {
1361 dprintf("\t\tMPB_ATTRIB_RAID10\n");
1362 not_supported
^= MPB_ATTRIB_RAID10
;
1364 if (not_supported
& MPB_ATTRIB_RAID1E
) {
1365 dprintf("\t\tMPB_ATTRIB_RAID1E\n");
1366 not_supported
^= MPB_ATTRIB_RAID1E
;
1368 if (not_supported
& MPB_ATTRIB_RAID5
) {
1369 dprintf("\t\tMPB_ATTRIB_RAID5\n");
1370 not_supported
^= MPB_ATTRIB_RAID5
;
1372 if (not_supported
& MPB_ATTRIB_RAIDCNG
) {
1373 dprintf("\t\tMPB_ATTRIB_RAIDCNG\n");
1374 not_supported
^= MPB_ATTRIB_RAIDCNG
;
1376 if (not_supported
& MPB_ATTRIB_BBM
) {
1377 dprintf("\t\tMPB_ATTRIB_BBM\n");
1378 not_supported
^= MPB_ATTRIB_BBM
;
1380 if (not_supported
& MPB_ATTRIB_CHECKSUM_VERIFY
) {
1381 dprintf("\t\tMPB_ATTRIB_CHECKSUM_VERIFY (== MPB_ATTRIB_LEGACY)\n");
1382 not_supported
^= MPB_ATTRIB_CHECKSUM_VERIFY
;
1384 if (not_supported
& MPB_ATTRIB_EXP_STRIPE_SIZE
) {
1385 dprintf("\t\tMPB_ATTRIB_EXP_STRIP_SIZE\n");
1386 not_supported
^= MPB_ATTRIB_EXP_STRIPE_SIZE
;
1388 if (not_supported
& MPB_ATTRIB_2TB_DISK
) {
1389 dprintf("\t\tMPB_ATTRIB_2TB_DISK\n");
1390 not_supported
^= MPB_ATTRIB_2TB_DISK
;
1392 if (not_supported
& MPB_ATTRIB_NEVER_USE2
) {
1393 dprintf("\t\tMPB_ATTRIB_NEVER_USE2\n");
1394 not_supported
^= MPB_ATTRIB_NEVER_USE2
;
1396 if (not_supported
& MPB_ATTRIB_NEVER_USE
) {
1397 dprintf("\t\tMPB_ATTRIB_NEVER_USE\n");
1398 not_supported
^= MPB_ATTRIB_NEVER_USE
;
1402 dprintf(Name
"(IMSM): Unknown attributes : %x\n", not_supported
);
1411 static void getinfo_super_imsm(struct supertype
*st
, struct mdinfo
*info
, char *map
);
1413 static void examine_super_imsm(struct supertype
*st
, char *homehost
)
1415 struct intel_super
*super
= st
->sb
;
1416 struct imsm_super
*mpb
= super
->anchor
;
1417 char str
[MAX_SIGNATURE_LENGTH
];
1422 __u32 reserved
= imsm_reserved_sectors(super
, super
->disks
);
1425 snprintf(str
, MPB_SIG_LEN
, "%s", mpb
->sig
);
1426 printf(" Magic : %s\n", str
);
1427 snprintf(str
, strlen(MPB_VERSION_RAID0
), "%s", get_imsm_version(mpb
));
1428 printf(" Version : %s\n", get_imsm_version(mpb
));
1429 printf(" Orig Family : %08x\n", __le32_to_cpu(mpb
->orig_family_num
));
1430 printf(" Family : %08x\n", __le32_to_cpu(mpb
->family_num
));
1431 printf(" Generation : %08x\n", __le32_to_cpu(mpb
->generation_num
));
1432 printf(" Attributes : ");
1433 if (imsm_check_attributes(mpb
->attributes
))
1434 printf("All supported\n");
1436 printf("not supported\n");
1437 getinfo_super_imsm(st
, &info
, NULL
);
1438 fname_from_uuid(st
, &info
, nbuf
, ':');
1439 printf(" UUID : %s\n", nbuf
+ 5);
1440 sum
= __le32_to_cpu(mpb
->check_sum
);
1441 printf(" Checksum : %08x %s\n", sum
,
1442 __gen_imsm_checksum(mpb
) == sum
? "correct" : "incorrect");
1443 printf(" MPB Sectors : %d\n", mpb_sectors(mpb
));
1444 printf(" Disks : %d\n", mpb
->num_disks
);
1445 printf(" RAID Devices : %d\n", mpb
->num_raid_devs
);
1446 print_imsm_disk(__get_imsm_disk(mpb
, super
->disks
->index
), super
->disks
->index
, reserved
);
1447 if (super
->bbm_log
) {
1448 struct bbm_log
*log
= super
->bbm_log
;
1451 printf("Bad Block Management Log:\n");
1452 printf(" Log Size : %d\n", __le32_to_cpu(mpb
->bbm_log_size
));
1453 printf(" Signature : %x\n", __le32_to_cpu(log
->signature
));
1454 printf(" Entry Count : %d\n", __le32_to_cpu(log
->entry_count
));
1455 printf(" Spare Blocks : %d\n", __le32_to_cpu(log
->reserved_spare_block_count
));
1456 printf(" First Spare : %llx\n",
1457 (unsigned long long) __le64_to_cpu(log
->first_spare_lba
));
1459 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
1461 struct imsm_dev
*dev
= __get_imsm_dev(mpb
, i
);
1463 super
->current_vol
= i
;
1464 getinfo_super_imsm(st
, &info
, NULL
);
1465 fname_from_uuid(st
, &info
, nbuf
, ':');
1466 print_imsm_dev(super
, dev
, nbuf
+ 5, super
->disks
->index
);
1468 for (i
= 0; i
< mpb
->num_disks
; i
++) {
1469 if (i
== super
->disks
->index
)
1471 print_imsm_disk(__get_imsm_disk(mpb
, i
), i
, reserved
);
1474 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
1475 if (dl
->index
== -1)
1476 print_imsm_disk(&dl
->disk
, -1, reserved
);
1478 examine_migr_rec_imsm(super
);
1481 static void brief_examine_super_imsm(struct supertype
*st
, int verbose
)
1483 /* We just write a generic IMSM ARRAY entry */
1486 struct intel_super
*super
= st
->sb
;
1488 if (!super
->anchor
->num_raid_devs
) {
1489 printf("ARRAY metadata=imsm\n");
1493 getinfo_super_imsm(st
, &info
, NULL
);
1494 fname_from_uuid(st
, &info
, nbuf
, ':');
1495 printf("ARRAY metadata=imsm UUID=%s\n", nbuf
+ 5);
1498 static void brief_examine_subarrays_imsm(struct supertype
*st
, int verbose
)
1500 /* We just write a generic IMSM ARRAY entry */
1504 struct intel_super
*super
= st
->sb
;
1507 if (!super
->anchor
->num_raid_devs
)
1510 getinfo_super_imsm(st
, &info
, NULL
);
1511 fname_from_uuid(st
, &info
, nbuf
, ':');
1512 for (i
= 0; i
< super
->anchor
->num_raid_devs
; i
++) {
1513 struct imsm_dev
*dev
= get_imsm_dev(super
, i
);
1515 super
->current_vol
= i
;
1516 getinfo_super_imsm(st
, &info
, NULL
);
1517 fname_from_uuid(st
, &info
, nbuf1
, ':');
1518 printf("ARRAY /dev/md/%.16s container=%s member=%d UUID=%s\n",
1519 dev
->volume
, nbuf
+ 5, i
, nbuf1
+ 5);
1523 static void export_examine_super_imsm(struct supertype
*st
)
1525 struct intel_super
*super
= st
->sb
;
1526 struct imsm_super
*mpb
= super
->anchor
;
1530 getinfo_super_imsm(st
, &info
, NULL
);
1531 fname_from_uuid(st
, &info
, nbuf
, ':');
1532 printf("MD_METADATA=imsm\n");
1533 printf("MD_LEVEL=container\n");
1534 printf("MD_UUID=%s\n", nbuf
+5);
1535 printf("MD_DEVICES=%u\n", mpb
->num_disks
);
1538 static void detail_super_imsm(struct supertype
*st
, char *homehost
)
1543 getinfo_super_imsm(st
, &info
, NULL
);
1544 fname_from_uuid(st
, &info
, nbuf
, ':');
1545 printf("\n UUID : %s\n", nbuf
+ 5);
1548 static void brief_detail_super_imsm(struct supertype
*st
)
1552 getinfo_super_imsm(st
, &info
, NULL
);
1553 fname_from_uuid(st
, &info
, nbuf
, ':');
1554 printf(" UUID=%s", nbuf
+ 5);
1557 static int imsm_read_serial(int fd
, char *devname
, __u8
*serial
);
1558 static void fd2devname(int fd
, char *name
);
1560 static int ahci_enumerate_ports(const char *hba_path
, int port_count
, int host_base
, int verbose
)
1562 /* dump an unsorted list of devices attached to AHCI Intel storage
1563 * controller, as well as non-connected ports
1565 int hba_len
= strlen(hba_path
) + 1;
1570 unsigned long port_mask
= (1 << port_count
) - 1;
1572 if (port_count
> (int)sizeof(port_mask
) * 8) {
1574 fprintf(stderr
, Name
": port_count %d out of range\n", port_count
);
1578 /* scroll through /sys/dev/block looking for devices attached to
1581 dir
= opendir("/sys/dev/block");
1582 for (ent
= dir
? readdir(dir
) : NULL
; ent
; ent
= readdir(dir
)) {
1593 if (sscanf(ent
->d_name
, "%d:%d", &major
, &minor
) != 2)
1595 path
= devt_to_devpath(makedev(major
, minor
));
1598 if (!path_attached_to_hba(path
, hba_path
)) {
1604 /* retrieve the scsi device type */
1605 if (asprintf(&device
, "/sys/dev/block/%d:%d/device/xxxxxxx", major
, minor
) < 0) {
1607 fprintf(stderr
, Name
": failed to allocate 'device'\n");
1611 sprintf(device
, "/sys/dev/block/%d:%d/device/type", major
, minor
);
1612 if (load_sys(device
, buf
) != 0) {
1614 fprintf(stderr
, Name
": failed to read device type for %s\n",
1620 type
= strtoul(buf
, NULL
, 10);
1622 /* if it's not a disk print the vendor and model */
1623 if (!(type
== 0 || type
== 7 || type
== 14)) {
1626 sprintf(device
, "/sys/dev/block/%d:%d/device/vendor", major
, minor
);
1627 if (load_sys(device
, buf
) == 0) {
1628 strncpy(vendor
, buf
, sizeof(vendor
));
1629 vendor
[sizeof(vendor
) - 1] = '\0';
1630 c
= (char *) &vendor
[sizeof(vendor
) - 1];
1631 while (isspace(*c
) || *c
== '\0')
1635 sprintf(device
, "/sys/dev/block/%d:%d/device/model", major
, minor
);
1636 if (load_sys(device
, buf
) == 0) {
1637 strncpy(model
, buf
, sizeof(model
));
1638 model
[sizeof(model
) - 1] = '\0';
1639 c
= (char *) &model
[sizeof(model
) - 1];
1640 while (isspace(*c
) || *c
== '\0')
1644 if (vendor
[0] && model
[0])
1645 sprintf(buf
, "%.64s %.64s", vendor
, model
);
1647 switch (type
) { /* numbers from hald/linux/device.c */
1648 case 1: sprintf(buf
, "tape"); break;
1649 case 2: sprintf(buf
, "printer"); break;
1650 case 3: sprintf(buf
, "processor"); break;
1652 case 5: sprintf(buf
, "cdrom"); break;
1653 case 6: sprintf(buf
, "scanner"); break;
1654 case 8: sprintf(buf
, "media_changer"); break;
1655 case 9: sprintf(buf
, "comm"); break;
1656 case 12: sprintf(buf
, "raid"); break;
1657 default: sprintf(buf
, "unknown");
1663 /* chop device path to 'host%d' and calculate the port number */
1664 c
= strchr(&path
[hba_len
], '/');
1667 fprintf(stderr
, Name
": %s - invalid path name\n", path
+ hba_len
);
1672 if (sscanf(&path
[hba_len
], "host%d", &port
) == 1)
1676 *c
= '/'; /* repair the full string */
1677 fprintf(stderr
, Name
": failed to determine port number for %s\n",
1684 /* mark this port as used */
1685 port_mask
&= ~(1 << port
);
1687 /* print out the device information */
1689 printf(" Port%d : - non-disk device (%s) -\n", port
, buf
);
1693 fd
= dev_open(ent
->d_name
, O_RDONLY
);
1695 printf(" Port%d : - disk info unavailable -\n", port
);
1697 fd2devname(fd
, buf
);
1698 printf(" Port%d : %s", port
, buf
);
1699 if (imsm_read_serial(fd
, NULL
, (__u8
*) buf
) == 0)
1700 printf(" (%.*s)\n", MAX_RAID_SERIAL_LEN
, buf
);
1715 for (i
= 0; i
< port_count
; i
++)
1716 if (port_mask
& (1 << i
))
1717 printf(" Port%d : - no device attached -\n", i
);
1723 static void print_found_intel_controllers(struct sys_dev
*elem
)
1725 for (; elem
; elem
= elem
->next
) {
1726 fprintf(stderr
, Name
": found Intel(R) ");
1727 if (elem
->type
== SYS_DEV_SATA
)
1728 fprintf(stderr
, "SATA ");
1729 else if (elem
->type
== SYS_DEV_SAS
)
1730 fprintf(stderr
, "SAS ");
1731 fprintf(stderr
, "RAID controller");
1733 fprintf(stderr
, " at %s", elem
->pci_id
);
1734 fprintf(stderr
, ".\n");
1739 static int ahci_get_port_count(const char *hba_path
, int *port_count
)
1746 if ((dir
= opendir(hba_path
)) == NULL
)
1749 for (ent
= readdir(dir
); ent
; ent
= readdir(dir
)) {
1752 if (sscanf(ent
->d_name
, "host%d", &host
) != 1)
1754 if (*port_count
== 0)
1756 else if (host
< host_base
)
1759 if (host
+ 1 > *port_count
+ host_base
)
1760 *port_count
= host
+ 1 - host_base
;
1766 static void print_imsm_capability(const struct imsm_orom
*orom
)
1768 printf(" Platform : Intel(R) Matrix Storage Manager\n");
1769 printf(" Version : %d.%d.%d.%d\n", orom
->major_ver
, orom
->minor_ver
,
1770 orom
->hotfix_ver
, orom
->build
);
1771 printf(" RAID Levels :%s%s%s%s%s\n",
1772 imsm_orom_has_raid0(orom
) ? " raid0" : "",
1773 imsm_orom_has_raid1(orom
) ? " raid1" : "",
1774 imsm_orom_has_raid1e(orom
) ? " raid1e" : "",
1775 imsm_orom_has_raid10(orom
) ? " raid10" : "",
1776 imsm_orom_has_raid5(orom
) ? " raid5" : "");
1777 printf(" Chunk Sizes :%s%s%s%s%s%s%s%s%s%s%s%s%s%s%s%s\n",
1778 imsm_orom_has_chunk(orom
, 2) ? " 2k" : "",
1779 imsm_orom_has_chunk(orom
, 4) ? " 4k" : "",
1780 imsm_orom_has_chunk(orom
, 8) ? " 8k" : "",
1781 imsm_orom_has_chunk(orom
, 16) ? " 16k" : "",
1782 imsm_orom_has_chunk(orom
, 32) ? " 32k" : "",
1783 imsm_orom_has_chunk(orom
, 64) ? " 64k" : "",
1784 imsm_orom_has_chunk(orom
, 128) ? " 128k" : "",
1785 imsm_orom_has_chunk(orom
, 256) ? " 256k" : "",
1786 imsm_orom_has_chunk(orom
, 512) ? " 512k" : "",
1787 imsm_orom_has_chunk(orom
, 1024*1) ? " 1M" : "",
1788 imsm_orom_has_chunk(orom
, 1024*2) ? " 2M" : "",
1789 imsm_orom_has_chunk(orom
, 1024*4) ? " 4M" : "",
1790 imsm_orom_has_chunk(orom
, 1024*8) ? " 8M" : "",
1791 imsm_orom_has_chunk(orom
, 1024*16) ? " 16M" : "",
1792 imsm_orom_has_chunk(orom
, 1024*32) ? " 32M" : "",
1793 imsm_orom_has_chunk(orom
, 1024*64) ? " 64M" : "");
1794 printf(" 2TB volumes :%s supported\n",
1795 (orom
->attr
& IMSM_OROM_ATTR_2TB
)?"":" not");
1796 printf(" 2TB disks :%s supported\n",
1797 (orom
->attr
& IMSM_OROM_ATTR_2TB_DISK
)?"":" not");
1798 printf(" Max Disks : %d\n", orom
->tds
);
1799 printf(" Max Volumes : %d per array, %d per controller\n",
1800 orom
->vpa
, orom
->vphba
);
1804 static int detail_platform_imsm(int verbose
, int enumerate_only
)
1806 /* There are two components to imsm platform support, the ahci SATA
1807 * controller and the option-rom. To find the SATA controller we
1808 * simply look in /sys/bus/pci/drivers/ahci to see if an ahci
1809 * controller with the Intel vendor id is present. This approach
1810 * allows mdadm to leverage the kernel's ahci detection logic, with the
1811 * caveat that if ahci.ko is not loaded mdadm will not be able to
1812 * detect platform raid capabilities. The option-rom resides in a
1813 * platform "Adapter ROM". We scan for its signature to retrieve the
1814 * platform capabilities. If raid support is disabled in the BIOS the
1815 * option-rom capability structure will not be available.
1817 const struct imsm_orom
*orom
;
1818 struct sys_dev
*list
, *hba
;
1823 if (enumerate_only
) {
1824 if (check_env("IMSM_NO_PLATFORM"))
1826 list
= find_intel_devices();
1829 for (hba
= list
; hba
; hba
= hba
->next
) {
1830 orom
= find_imsm_capability(hba
->type
);
1836 free_sys_dev(&list
);
1840 list
= find_intel_devices();
1843 fprintf(stderr
, Name
": no active Intel(R) RAID "
1844 "controller found.\n");
1845 free_sys_dev(&list
);
1848 print_found_intel_controllers(list
);
1850 for (hba
= list
; hba
; hba
= hba
->next
) {
1851 orom
= find_imsm_capability(hba
->type
);
1853 fprintf(stderr
, Name
": imsm capabilities not found for controller: %s (type %s)\n",
1854 hba
->path
, get_sys_dev_type(hba
->type
));
1856 print_imsm_capability(orom
);
1859 for (hba
= list
; hba
; hba
= hba
->next
) {
1860 printf(" I/O Controller : %s (%s)\n",
1861 hba
->path
, get_sys_dev_type(hba
->type
));
1863 if (hba
->type
== SYS_DEV_SATA
) {
1864 host_base
= ahci_get_port_count(hba
->path
, &port_count
);
1865 if (ahci_enumerate_ports(hba
->path
, port_count
, host_base
, verbose
)) {
1867 fprintf(stderr
, Name
": failed to enumerate "
1868 "ports on SATA controller at %s.", hba
->pci_id
);
1874 free_sys_dev(&list
);
1879 static int match_home_imsm(struct supertype
*st
, char *homehost
)
1881 /* the imsm metadata format does not specify any host
1882 * identification information. We return -1 since we can never
1883 * confirm nor deny whether a given array is "meant" for this
1884 * host. We rely on compare_super and the 'family_num' fields to
1885 * exclude member disks that do not belong, and we rely on
1886 * mdadm.conf to specify the arrays that should be assembled.
1887 * Auto-assembly may still pick up "foreign" arrays.
1893 static void uuid_from_super_imsm(struct supertype
*st
, int uuid
[4])
1895 /* The uuid returned here is used for:
1896 * uuid to put into bitmap file (Create, Grow)
1897 * uuid for backup header when saving critical section (Grow)
1898 * comparing uuids when re-adding a device into an array
1899 * In these cases the uuid required is that of the data-array,
1900 * not the device-set.
1901 * uuid to recognise same set when adding a missing device back
1902 * to an array. This is a uuid for the device-set.
1904 * For each of these we can make do with a truncated
1905 * or hashed uuid rather than the original, as long as
1907 * In each case the uuid required is that of the data-array,
1908 * not the device-set.
1910 /* imsm does not track uuid's so we synthesis one using sha1 on
1911 * - The signature (Which is constant for all imsm array, but no matter)
1912 * - the orig_family_num of the container
1913 * - the index number of the volume
1914 * - the 'serial' number of the volume.
1915 * Hopefully these are all constant.
1917 struct intel_super
*super
= st
->sb
;
1920 struct sha1_ctx ctx
;
1921 struct imsm_dev
*dev
= NULL
;
1924 /* some mdadm versions failed to set ->orig_family_num, in which
1925 * case fall back to ->family_num. orig_family_num will be
1926 * fixed up with the first metadata update.
1928 family_num
= super
->anchor
->orig_family_num
;
1929 if (family_num
== 0)
1930 family_num
= super
->anchor
->family_num
;
1931 sha1_init_ctx(&ctx
);
1932 sha1_process_bytes(super
->anchor
->sig
, MPB_SIG_LEN
, &ctx
);
1933 sha1_process_bytes(&family_num
, sizeof(__u32
), &ctx
);
1934 if (super
->current_vol
>= 0)
1935 dev
= get_imsm_dev(super
, super
->current_vol
);
1937 __u32 vol
= super
->current_vol
;
1938 sha1_process_bytes(&vol
, sizeof(vol
), &ctx
);
1939 sha1_process_bytes(dev
->volume
, MAX_RAID_SERIAL_LEN
, &ctx
);
1941 sha1_finish_ctx(&ctx
, buf
);
1942 memcpy(uuid
, buf
, 4*4);
1947 get_imsm_numerical_version(struct imsm_super
*mpb
, int *m
, int *p
)
1949 __u8
*v
= get_imsm_version(mpb
);
1950 __u8
*end
= mpb
->sig
+ MAX_SIGNATURE_LENGTH
;
1951 char major
[] = { 0, 0, 0 };
1952 char minor
[] = { 0 ,0, 0 };
1953 char patch
[] = { 0, 0, 0 };
1954 char *ver_parse
[] = { major
, minor
, patch
};
1958 while (*v
!= '\0' && v
< end
) {
1959 if (*v
!= '.' && j
< 2)
1960 ver_parse
[i
][j
++] = *v
;
1968 *m
= strtol(minor
, NULL
, 0);
1969 *p
= strtol(patch
, NULL
, 0);
1973 static __u32
migr_strip_blocks_resync(struct imsm_dev
*dev
)
1975 /* migr_strip_size when repairing or initializing parity */
1976 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
1977 __u32 chunk
= __le32_to_cpu(map
->blocks_per_strip
);
1979 switch (get_imsm_raid_level(map
)) {
1984 return 128*1024 >> 9;
1988 static __u32
migr_strip_blocks_rebuild(struct imsm_dev
*dev
)
1990 /* migr_strip_size when rebuilding a degraded disk, no idea why
1991 * this is different than migr_strip_size_resync(), but it's good
1994 struct imsm_map
*map
= get_imsm_map(dev
, MAP_1
);
1995 __u32 chunk
= __le32_to_cpu(map
->blocks_per_strip
);
1997 switch (get_imsm_raid_level(map
)) {
2000 if (map
->num_members
% map
->num_domains
== 0)
2001 return 128*1024 >> 9;
2005 return max((__u32
) 64*1024 >> 9, chunk
);
2007 return 128*1024 >> 9;
2011 static __u32
num_stripes_per_unit_resync(struct imsm_dev
*dev
)
2013 struct imsm_map
*lo
= get_imsm_map(dev
, MAP_0
);
2014 struct imsm_map
*hi
= get_imsm_map(dev
, MAP_1
);
2015 __u32 lo_chunk
= __le32_to_cpu(lo
->blocks_per_strip
);
2016 __u32 hi_chunk
= __le32_to_cpu(hi
->blocks_per_strip
);
2018 return max((__u32
) 1, hi_chunk
/ lo_chunk
);
2021 static __u32
num_stripes_per_unit_rebuild(struct imsm_dev
*dev
)
2023 struct imsm_map
*lo
= get_imsm_map(dev
, MAP_0
);
2024 int level
= get_imsm_raid_level(lo
);
2026 if (level
== 1 || level
== 10) {
2027 struct imsm_map
*hi
= get_imsm_map(dev
, MAP_1
);
2029 return hi
->num_domains
;
2031 return num_stripes_per_unit_resync(dev
);
2034 static __u8
imsm_num_data_members(struct imsm_dev
*dev
, int second_map
)
2036 /* named 'imsm_' because raid0, raid1 and raid10
2037 * counter-intuitively have the same number of data disks
2039 struct imsm_map
*map
= get_imsm_map(dev
, second_map
);
2041 switch (get_imsm_raid_level(map
)) {
2043 return map
->num_members
;
2047 return map
->num_members
/2;
2049 return map
->num_members
- 1;
2051 dprintf("%s: unsupported raid level\n", __func__
);
2056 static __u32
parity_segment_depth(struct imsm_dev
*dev
)
2058 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
2059 __u32 chunk
= __le32_to_cpu(map
->blocks_per_strip
);
2061 switch(get_imsm_raid_level(map
)) {
2064 return chunk
* map
->num_domains
;
2066 return chunk
* map
->num_members
;
2072 static __u32
map_migr_block(struct imsm_dev
*dev
, __u32 block
)
2074 struct imsm_map
*map
= get_imsm_map(dev
, MAP_1
);
2075 __u32 chunk
= __le32_to_cpu(map
->blocks_per_strip
);
2076 __u32 strip
= block
/ chunk
;
2078 switch (get_imsm_raid_level(map
)) {
2081 __u32 vol_strip
= (strip
* map
->num_domains
) + 1;
2082 __u32 vol_stripe
= vol_strip
/ map
->num_members
;
2084 return vol_stripe
* chunk
+ block
% chunk
;
2086 __u32 stripe
= strip
/ (map
->num_members
- 1);
2088 return stripe
* chunk
+ block
% chunk
;
2095 static __u64
blocks_per_migr_unit(struct intel_super
*super
,
2096 struct imsm_dev
*dev
)
2098 /* calculate the conversion factor between per member 'blocks'
2099 * (md/{resync,rebuild}_start) and imsm migration units, return
2100 * 0 for the 'not migrating' and 'unsupported migration' cases
2102 if (!dev
->vol
.migr_state
)
2105 switch (migr_type(dev
)) {
2106 case MIGR_GEN_MIGR
: {
2107 struct migr_record
*migr_rec
= super
->migr_rec
;
2108 return __le32_to_cpu(migr_rec
->blocks_per_unit
);
2113 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
2114 __u32 stripes_per_unit
;
2115 __u32 blocks_per_unit
;
2124 /* yes, this is really the translation of migr_units to
2125 * per-member blocks in the 'resync' case
2127 stripes_per_unit
= num_stripes_per_unit_resync(dev
);
2128 migr_chunk
= migr_strip_blocks_resync(dev
);
2129 disks
= imsm_num_data_members(dev
, MAP_0
);
2130 blocks_per_unit
= stripes_per_unit
* migr_chunk
* disks
;
2131 stripe
= __le16_to_cpu(map
->blocks_per_strip
) * disks
;
2132 segment
= blocks_per_unit
/ stripe
;
2133 block_rel
= blocks_per_unit
- segment
* stripe
;
2134 parity_depth
= parity_segment_depth(dev
);
2135 block_map
= map_migr_block(dev
, block_rel
);
2136 return block_map
+ parity_depth
* segment
;
2138 case MIGR_REBUILD
: {
2139 __u32 stripes_per_unit
;
2142 stripes_per_unit
= num_stripes_per_unit_rebuild(dev
);
2143 migr_chunk
= migr_strip_blocks_rebuild(dev
);
2144 return migr_chunk
* stripes_per_unit
;
2146 case MIGR_STATE_CHANGE
:
2152 static int imsm_level_to_layout(int level
)
2160 return ALGORITHM_LEFT_ASYMMETRIC
;
2167 /*******************************************************************************
2168 * Function: read_imsm_migr_rec
2169 * Description: Function reads imsm migration record from last sector of disk
2171 * fd : disk descriptor
2172 * super : metadata info
2176 ******************************************************************************/
2177 static int read_imsm_migr_rec(int fd
, struct intel_super
*super
)
2180 unsigned long long dsize
;
2182 get_dev_size(fd
, NULL
, &dsize
);
2183 if (lseek64(fd
, dsize
- MIGR_REC_POSITION
, SEEK_SET
) < 0) {
2185 Name
": Cannot seek to anchor block: %s\n",
2189 if (read(fd
, super
->migr_rec_buf
, MIGR_REC_BUF_SIZE
) !=
2190 MIGR_REC_BUF_SIZE
) {
2192 Name
": Cannot read migr record block: %s\n",
2202 static struct imsm_dev
*imsm_get_device_during_migration(
2203 struct intel_super
*super
)
2206 struct intel_dev
*dv
;
2208 for (dv
= super
->devlist
; dv
; dv
= dv
->next
) {
2209 if (is_gen_migration(dv
->dev
))
2215 /*******************************************************************************
2216 * Function: load_imsm_migr_rec
2217 * Description: Function reads imsm migration record (it is stored at the last
2220 * super : imsm internal array info
2221 * info : general array info
2225 * -2 : no migration in progress
2226 ******************************************************************************/
2227 static int load_imsm_migr_rec(struct intel_super
*super
, struct mdinfo
*info
)
2230 struct dl
*dl
= NULL
;
2234 struct imsm_dev
*dev
;
2235 struct imsm_map
*map
= NULL
;
2238 /* find map under migration */
2239 dev
= imsm_get_device_during_migration(super
);
2240 /* nothing to load,no migration in progress?
2244 map
= get_imsm_map(dev
, MAP_0
);
2247 for (sd
= info
->devs
; sd
; sd
= sd
->next
) {
2248 /* skip spare and failed disks
2250 if (sd
->disk
.raid_disk
< 0)
2252 /* read only from one of the first two slots */
2254 slot
= get_imsm_disk_slot(map
,
2255 sd
->disk
.raid_disk
);
2256 if ((map
== NULL
) || (slot
> 1) || (slot
< 0))
2259 sprintf(nm
, "%d:%d", sd
->disk
.major
, sd
->disk
.minor
);
2260 fd
= dev_open(nm
, O_RDONLY
);
2266 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
2267 /* skip spare and failed disks
2271 /* read only from one of the first two slots */
2273 slot
= get_imsm_disk_slot(map
, dl
->index
);
2274 if ((map
== NULL
) || (slot
> 1) || (slot
< 0))
2276 sprintf(nm
, "%d:%d", dl
->major
, dl
->minor
);
2277 fd
= dev_open(nm
, O_RDONLY
);
2284 retval
= read_imsm_migr_rec(fd
, super
);
2293 /*******************************************************************************
2294 * function: imsm_create_metadata_checkpoint_update
2295 * Description: It creates update for checkpoint change.
2297 * super : imsm internal array info
2298 * u : pointer to prepared update
2301 * If length is equal to 0, input pointer u contains no update
2302 ******************************************************************************/
2303 static int imsm_create_metadata_checkpoint_update(
2304 struct intel_super
*super
,
2305 struct imsm_update_general_migration_checkpoint
**u
)
2308 int update_memory_size
= 0;
2310 dprintf("imsm_create_metadata_checkpoint_update(enter)\n");
2316 /* size of all update data without anchor */
2317 update_memory_size
=
2318 sizeof(struct imsm_update_general_migration_checkpoint
);
2320 *u
= calloc(1, update_memory_size
);
2322 dprintf("error: cannot get memory for "
2323 "imsm_create_metadata_checkpoint_update update\n");
2326 (*u
)->type
= update_general_migration_checkpoint
;
2327 (*u
)->curr_migr_unit
= __le32_to_cpu(super
->migr_rec
->curr_migr_unit
);
2328 dprintf("imsm_create_metadata_checkpoint_update: prepared for %u\n",
2329 (*u
)->curr_migr_unit
);
2331 return update_memory_size
;
2335 static void imsm_update_metadata_locally(struct supertype
*st
,
2336 void *buf
, int len
);
2338 /*******************************************************************************
2339 * Function: write_imsm_migr_rec
2340 * Description: Function writes imsm migration record
2341 * (at the last sector of disk)
2343 * super : imsm internal array info
2347 ******************************************************************************/
2348 static int write_imsm_migr_rec(struct supertype
*st
)
2350 struct intel_super
*super
= st
->sb
;
2351 unsigned long long dsize
;
2357 struct imsm_update_general_migration_checkpoint
*u
;
2358 struct imsm_dev
*dev
;
2359 struct imsm_map
*map
= NULL
;
2361 /* find map under migration */
2362 dev
= imsm_get_device_during_migration(super
);
2363 /* if no migration, write buffer anyway to clear migr_record
2364 * on disk based on first available device
2367 dev
= get_imsm_dev(super
, super
->current_vol
< 0 ? 0 :
2368 super
->current_vol
);
2370 map
= get_imsm_map(dev
, MAP_0
);
2372 for (sd
= super
->disks
; sd
; sd
= sd
->next
) {
2375 /* skip failed and spare devices */
2378 /* write to 2 first slots only */
2380 slot
= get_imsm_disk_slot(map
, sd
->index
);
2381 if ((map
== NULL
) || (slot
> 1) || (slot
< 0))
2384 sprintf(nm
, "%d:%d", sd
->major
, sd
->minor
);
2385 fd
= dev_open(nm
, O_RDWR
);
2388 get_dev_size(fd
, NULL
, &dsize
);
2389 if (lseek64(fd
, dsize
- MIGR_REC_POSITION
, SEEK_SET
) < 0) {
2391 Name
": Cannot seek to anchor block: %s\n",
2395 if (write(fd
, super
->migr_rec_buf
, MIGR_REC_BUF_SIZE
) !=
2396 MIGR_REC_BUF_SIZE
) {
2398 Name
": Cannot write migr record block: %s\n",
2405 /* update checkpoint information in metadata */
2406 len
= imsm_create_metadata_checkpoint_update(super
, &u
);
2409 dprintf("imsm: Cannot prepare update\n");
2412 /* update metadata locally */
2413 imsm_update_metadata_locally(st
, u
, len
);
2414 /* and possibly remotely */
2415 if (st
->update_tail
) {
2416 append_metadata_update(st
, u
, len
);
2417 /* during reshape we do all work inside metadata handler
2418 * manage_reshape(), so metadata update has to be triggered
2421 flush_metadata_updates(st
);
2422 st
->update_tail
= &st
->updates
;
2432 #endif /* MDASSEMBLE */
2434 /* spare/missing disks activations are not allowe when
2435 * array/container performs reshape operation, because
2436 * all arrays in container works on the same disks set
2438 int imsm_reshape_blocks_arrays_changes(struct intel_super
*super
)
2441 struct intel_dev
*i_dev
;
2442 struct imsm_dev
*dev
;
2444 /* check whole container
2446 for (i_dev
= super
->devlist
; i_dev
; i_dev
= i_dev
->next
) {
2448 if (is_gen_migration(dev
)) {
2449 /* No repair during any migration in container
2458 static void getinfo_super_imsm_volume(struct supertype
*st
, struct mdinfo
*info
, char *dmap
)
2460 struct intel_super
*super
= st
->sb
;
2461 struct migr_record
*migr_rec
= super
->migr_rec
;
2462 struct imsm_dev
*dev
= get_imsm_dev(super
, super
->current_vol
);
2463 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
2464 struct imsm_map
*prev_map
= get_imsm_map(dev
, MAP_1
);
2465 struct imsm_map
*map_to_analyse
= map
;
2468 unsigned int component_size_alligment
;
2469 int map_disks
= info
->array
.raid_disks
;
2471 memset(info
, 0, sizeof(*info
));
2473 map_to_analyse
= prev_map
;
2475 dl
= super
->current_disk
;
2477 info
->container_member
= super
->current_vol
;
2478 info
->array
.raid_disks
= map
->num_members
;
2479 info
->array
.level
= get_imsm_raid_level(map_to_analyse
);
2480 info
->array
.layout
= imsm_level_to_layout(info
->array
.level
);
2481 info
->array
.md_minor
= -1;
2482 info
->array
.ctime
= 0;
2483 info
->array
.utime
= 0;
2484 info
->array
.chunk_size
=
2485 __le16_to_cpu(map_to_analyse
->blocks_per_strip
) << 9;
2486 info
->array
.state
= !dev
->vol
.dirty
;
2487 info
->custom_array_size
= __le32_to_cpu(dev
->size_high
);
2488 info
->custom_array_size
<<= 32;
2489 info
->custom_array_size
|= __le32_to_cpu(dev
->size_low
);
2490 info
->recovery_blocked
= imsm_reshape_blocks_arrays_changes(st
->sb
);
2492 if (is_gen_migration(dev
)) {
2493 info
->reshape_active
= 1;
2494 info
->new_level
= get_imsm_raid_level(map
);
2495 info
->new_layout
= imsm_level_to_layout(info
->new_level
);
2496 info
->new_chunk
= __le16_to_cpu(map
->blocks_per_strip
) << 9;
2497 info
->delta_disks
= map
->num_members
- prev_map
->num_members
;
2498 if (info
->delta_disks
) {
2499 /* this needs to be applied to every array
2502 info
->reshape_active
= CONTAINER_RESHAPE
;
2504 /* We shape information that we give to md might have to be
2505 * modify to cope with md's requirement for reshaping arrays.
2506 * For example, when reshaping a RAID0, md requires it to be
2507 * presented as a degraded RAID4.
2508 * Also if a RAID0 is migrating to a RAID5 we need to specify
2509 * the array as already being RAID5, but the 'before' layout
2510 * is a RAID4-like layout.
2512 switch (info
->array
.level
) {
2514 switch(info
->new_level
) {
2516 /* conversion is happening as RAID4 */
2517 info
->array
.level
= 4;
2518 info
->array
.raid_disks
+= 1;
2521 /* conversion is happening as RAID5 */
2522 info
->array
.level
= 5;
2523 info
->array
.layout
= ALGORITHM_PARITY_N
;
2524 info
->delta_disks
-= 1;
2527 /* FIXME error message */
2528 info
->array
.level
= UnSet
;
2534 info
->new_level
= UnSet
;
2535 info
->new_layout
= UnSet
;
2536 info
->new_chunk
= info
->array
.chunk_size
;
2537 info
->delta_disks
= 0;
2541 info
->disk
.major
= dl
->major
;
2542 info
->disk
.minor
= dl
->minor
;
2543 info
->disk
.number
= dl
->index
;
2544 info
->disk
.raid_disk
= get_imsm_disk_slot(map_to_analyse
,
2548 info
->data_offset
= pba_of_lba0(map_to_analyse
);
2549 info
->component_size
= blocks_per_member(map_to_analyse
);
2551 /* check component size aligment
2553 component_size_alligment
=
2554 info
->component_size
% (info
->array
.chunk_size
/512);
2556 if (component_size_alligment
&&
2557 (info
->array
.level
!= 1) && (info
->array
.level
!= UnSet
)) {
2558 dprintf("imsm: reported component size alligned from %llu ",
2559 info
->component_size
);
2560 info
->component_size
-= component_size_alligment
;
2561 dprintf("to %llu (%i).\n",
2562 info
->component_size
, component_size_alligment
);
2565 memset(info
->uuid
, 0, sizeof(info
->uuid
));
2566 info
->recovery_start
= MaxSector
;
2568 info
->reshape_progress
= 0;
2569 info
->resync_start
= MaxSector
;
2570 if ((map_to_analyse
->map_state
== IMSM_T_STATE_UNINITIALIZED
||
2572 imsm_reshape_blocks_arrays_changes(super
) == 0) {
2573 info
->resync_start
= 0;
2575 if (dev
->vol
.migr_state
) {
2576 switch (migr_type(dev
)) {
2579 __u64 blocks_per_unit
= blocks_per_migr_unit(super
,
2581 __u64 units
= __le32_to_cpu(dev
->vol
.curr_migr_unit
);
2583 info
->resync_start
= blocks_per_unit
* units
;
2586 case MIGR_GEN_MIGR
: {
2587 __u64 blocks_per_unit
= blocks_per_migr_unit(super
,
2589 __u64 units
= __le32_to_cpu(migr_rec
->curr_migr_unit
);
2590 unsigned long long array_blocks
;
2593 if (__le32_to_cpu(migr_rec
->ascending_migr
) &&
2595 (__le32_to_cpu(migr_rec
->num_migr_units
)-1)) &&
2596 (super
->migr_rec
->rec_status
==
2597 __cpu_to_le32(UNIT_SRC_IN_CP_AREA
)))
2600 info
->reshape_progress
= blocks_per_unit
* units
;
2602 dprintf("IMSM: General Migration checkpoint : %llu "
2603 "(%llu) -> read reshape progress : %llu\n",
2604 (unsigned long long)units
,
2605 (unsigned long long)blocks_per_unit
,
2606 info
->reshape_progress
);
2608 used_disks
= imsm_num_data_members(dev
, MAP_1
);
2609 if (used_disks
> 0) {
2610 array_blocks
= blocks_per_member(map
) *
2612 /* round array size down to closest MB
2614 info
->custom_array_size
= (array_blocks
2615 >> SECT_PER_MB_SHIFT
)
2616 << SECT_PER_MB_SHIFT
;
2620 /* we could emulate the checkpointing of
2621 * 'sync_action=check' migrations, but for now
2622 * we just immediately complete them
2625 /* this is handled by container_content_imsm() */
2626 case MIGR_STATE_CHANGE
:
2627 /* FIXME handle other migrations */
2629 /* we are not dirty, so... */
2630 info
->resync_start
= MaxSector
;
2634 strncpy(info
->name
, (char *) dev
->volume
, MAX_RAID_SERIAL_LEN
);
2635 info
->name
[MAX_RAID_SERIAL_LEN
] = 0;
2637 info
->array
.major_version
= -1;
2638 info
->array
.minor_version
= -2;
2639 devname
= devnum2devname(st
->container_dev
);
2640 *info
->text_version
= '\0';
2642 sprintf(info
->text_version
, "/%s/%d", devname
, info
->container_member
);
2644 info
->safe_mode_delay
= 4000; /* 4 secs like the Matrix driver */
2645 uuid_from_super_imsm(st
, info
->uuid
);
2649 for (i
=0; i
<map_disks
; i
++) {
2651 if (i
< info
->array
.raid_disks
) {
2652 struct imsm_disk
*dsk
;
2653 j
= get_imsm_disk_idx(dev
, i
, MAP_X
);
2654 dsk
= get_imsm_disk(super
, j
);
2655 if (dsk
&& (dsk
->status
& CONFIGURED_DISK
))
2662 static __u8
imsm_check_degraded(struct intel_super
*super
, struct imsm_dev
*dev
,
2663 int failed
, int look_in_map
);
2665 static int imsm_count_failed(struct intel_super
*super
, struct imsm_dev
*dev
,
2670 static void manage_second_map(struct intel_super
*super
, struct imsm_dev
*dev
)
2672 if (is_gen_migration(dev
)) {
2675 struct imsm_map
*map2
= get_imsm_map(dev
, MAP_1
);
2677 failed
= imsm_count_failed(super
, dev
, MAP_1
);
2678 map_state
= imsm_check_degraded(super
, dev
, failed
, MAP_1
);
2679 if (map2
->map_state
!= map_state
) {
2680 map2
->map_state
= map_state
;
2681 super
->updates_pending
++;
2687 static struct imsm_disk
*get_imsm_missing(struct intel_super
*super
, __u8 index
)
2691 for (d
= super
->missing
; d
; d
= d
->next
)
2692 if (d
->index
== index
)
2697 static void getinfo_super_imsm(struct supertype
*st
, struct mdinfo
*info
, char *map
)
2699 struct intel_super
*super
= st
->sb
;
2700 struct imsm_disk
*disk
;
2701 int map_disks
= info
->array
.raid_disks
;
2702 int max_enough
= -1;
2704 struct imsm_super
*mpb
;
2706 if (super
->current_vol
>= 0) {
2707 getinfo_super_imsm_volume(st
, info
, map
);
2710 memset(info
, 0, sizeof(*info
));
2712 /* Set raid_disks to zero so that Assemble will always pull in valid
2715 info
->array
.raid_disks
= 0;
2716 info
->array
.level
= LEVEL_CONTAINER
;
2717 info
->array
.layout
= 0;
2718 info
->array
.md_minor
= -1;
2719 info
->array
.ctime
= 0; /* N/A for imsm */
2720 info
->array
.utime
= 0;
2721 info
->array
.chunk_size
= 0;
2723 info
->disk
.major
= 0;
2724 info
->disk
.minor
= 0;
2725 info
->disk
.raid_disk
= -1;
2726 info
->reshape_active
= 0;
2727 info
->array
.major_version
= -1;
2728 info
->array
.minor_version
= -2;
2729 strcpy(info
->text_version
, "imsm");
2730 info
->safe_mode_delay
= 0;
2731 info
->disk
.number
= -1;
2732 info
->disk
.state
= 0;
2734 info
->recovery_start
= MaxSector
;
2735 info
->recovery_blocked
= imsm_reshape_blocks_arrays_changes(st
->sb
);
2737 /* do we have the all the insync disks that we expect? */
2738 mpb
= super
->anchor
;
2740 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
2741 struct imsm_dev
*dev
= get_imsm_dev(super
, i
);
2742 int failed
, enough
, j
, missing
= 0;
2743 struct imsm_map
*map
;
2746 failed
= imsm_count_failed(super
, dev
, MAP_0
);
2747 state
= imsm_check_degraded(super
, dev
, failed
, MAP_0
);
2748 map
= get_imsm_map(dev
, MAP_0
);
2750 /* any newly missing disks?
2751 * (catches single-degraded vs double-degraded)
2753 for (j
= 0; j
< map
->num_members
; j
++) {
2754 __u32 ord
= get_imsm_ord_tbl_ent(dev
, j
, MAP_0
);
2755 __u32 idx
= ord_to_idx(ord
);
2757 if (!(ord
& IMSM_ORD_REBUILD
) &&
2758 get_imsm_missing(super
, idx
)) {
2764 if (state
== IMSM_T_STATE_FAILED
)
2766 else if (state
== IMSM_T_STATE_DEGRADED
&&
2767 (state
!= map
->map_state
|| missing
))
2769 else /* we're normal, or already degraded */
2771 if (is_gen_migration(dev
) && missing
) {
2772 /* during general migration we need all disks
2773 * that process is running on.
2774 * No new missing disk is allowed.
2778 /* no more checks necessary
2782 /* in the missing/failed disk case check to see
2783 * if at least one array is runnable
2785 max_enough
= max(max_enough
, enough
);
2787 dprintf("%s: enough: %d\n", __func__
, max_enough
);
2788 info
->container_enough
= max_enough
;
2791 __u32 reserved
= imsm_reserved_sectors(super
, super
->disks
);
2793 disk
= &super
->disks
->disk
;
2794 info
->data_offset
= total_blocks(&super
->disks
->disk
) - reserved
;
2795 info
->component_size
= reserved
;
2796 info
->disk
.state
= is_configured(disk
) ? (1 << MD_DISK_ACTIVE
) : 0;
2797 /* we don't change info->disk.raid_disk here because
2798 * this state will be finalized in mdmon after we have
2799 * found the 'most fresh' version of the metadata
2801 info
->disk
.state
|= is_failed(disk
) ? (1 << MD_DISK_FAULTY
) : 0;
2802 info
->disk
.state
|= is_spare(disk
) ? 0 : (1 << MD_DISK_SYNC
);
2805 /* only call uuid_from_super_imsm when this disk is part of a populated container,
2806 * ->compare_super may have updated the 'num_raid_devs' field for spares
2808 if (info
->disk
.state
& (1 << MD_DISK_SYNC
) || super
->anchor
->num_raid_devs
)
2809 uuid_from_super_imsm(st
, info
->uuid
);
2811 memcpy(info
->uuid
, uuid_zero
, sizeof(uuid_zero
));
2813 /* I don't know how to compute 'map' on imsm, so use safe default */
2816 for (i
= 0; i
< map_disks
; i
++)
2822 /* allocates memory and fills disk in mdinfo structure
2823 * for each disk in array */
2824 struct mdinfo
*getinfo_super_disks_imsm(struct supertype
*st
)
2826 struct mdinfo
*mddev
= NULL
;
2827 struct intel_super
*super
= st
->sb
;
2828 struct imsm_disk
*disk
;
2831 if (!super
|| !super
->disks
)
2834 mddev
= malloc(sizeof(*mddev
));
2836 fprintf(stderr
, Name
": Failed to allocate memory.\n");
2839 memset(mddev
, 0, sizeof(*mddev
));
2843 tmp
= malloc(sizeof(*tmp
));
2845 fprintf(stderr
, Name
": Failed to allocate memory.\n");
2850 memset(tmp
, 0, sizeof(*tmp
));
2852 tmp
->next
= mddev
->devs
;
2854 tmp
->disk
.number
= count
++;
2855 tmp
->disk
.major
= dl
->major
;
2856 tmp
->disk
.minor
= dl
->minor
;
2857 tmp
->disk
.state
= is_configured(disk
) ?
2858 (1 << MD_DISK_ACTIVE
) : 0;
2859 tmp
->disk
.state
|= is_failed(disk
) ? (1 << MD_DISK_FAULTY
) : 0;
2860 tmp
->disk
.state
|= is_spare(disk
) ? 0 : (1 << MD_DISK_SYNC
);
2861 tmp
->disk
.raid_disk
= -1;
2867 static int update_super_imsm(struct supertype
*st
, struct mdinfo
*info
,
2868 char *update
, char *devname
, int verbose
,
2869 int uuid_set
, char *homehost
)
2871 /* For 'assemble' and 'force' we need to return non-zero if any
2872 * change was made. For others, the return value is ignored.
2873 * Update options are:
2874 * force-one : This device looks a bit old but needs to be included,
2875 * update age info appropriately.
2876 * assemble: clear any 'faulty' flag to allow this device to
2878 * force-array: Array is degraded but being forced, mark it clean
2879 * if that will be needed to assemble it.
2881 * newdev: not used ????
2882 * grow: Array has gained a new device - this is currently for
2884 * resync: mark as dirty so a resync will happen.
2885 * name: update the name - preserving the homehost
2886 * uuid: Change the uuid of the array to match watch is given
2888 * Following are not relevant for this imsm:
2889 * sparc2.2 : update from old dodgey metadata
2890 * super-minor: change the preferred_minor number
2891 * summaries: update redundant counters.
2892 * homehost: update the recorded homehost
2893 * _reshape_progress: record new reshape_progress position.
2896 struct intel_super
*super
= st
->sb
;
2897 struct imsm_super
*mpb
;
2899 /* we can only update container info */
2900 if (!super
|| super
->current_vol
>= 0 || !super
->anchor
)
2903 mpb
= super
->anchor
;
2905 if (strcmp(update
, "uuid") == 0) {
2906 /* We take this to mean that the family_num should be updated.
2907 * However that is much smaller than the uuid so we cannot really
2908 * allow an explicit uuid to be given. And it is hard to reliably
2910 * So if !uuid_set we know the current uuid is random and just used
2911 * the first 'int' and copy it to the other 3 positions.
2912 * Otherwise we require the 4 'int's to be the same as would be the
2913 * case if we are using a random uuid. So an explicit uuid will be
2914 * accepted as long as all for ints are the same... which shouldn't hurt
2917 info
->uuid
[1] = info
->uuid
[2] = info
->uuid
[3] = info
->uuid
[0];
2920 if (info
->uuid
[0] != info
->uuid
[1] ||
2921 info
->uuid
[1] != info
->uuid
[2] ||
2922 info
->uuid
[2] != info
->uuid
[3])
2928 mpb
->orig_family_num
= info
->uuid
[0];
2929 } else if (strcmp(update
, "assemble") == 0)
2934 /* successful update? recompute checksum */
2936 mpb
->check_sum
= __le32_to_cpu(__gen_imsm_checksum(mpb
));
2941 static size_t disks_to_mpb_size(int disks
)
2945 size
= sizeof(struct imsm_super
);
2946 size
+= (disks
- 1) * sizeof(struct imsm_disk
);
2947 size
+= 2 * sizeof(struct imsm_dev
);
2948 /* up to 2 maps per raid device (-2 for imsm_maps in imsm_dev */
2949 size
+= (4 - 2) * sizeof(struct imsm_map
);
2950 /* 4 possible disk_ord_tbl's */
2951 size
+= 4 * (disks
- 1) * sizeof(__u32
);
2956 static __u64
avail_size_imsm(struct supertype
*st
, __u64 devsize
)
2958 if (devsize
< (MPB_SECTOR_CNT
+ IMSM_RESERVED_SECTORS
))
2961 return devsize
- (MPB_SECTOR_CNT
+ IMSM_RESERVED_SECTORS
);
2964 static void free_devlist(struct intel_super
*super
)
2966 struct intel_dev
*dv
;
2968 while (super
->devlist
) {
2969 dv
= super
->devlist
->next
;
2970 free(super
->devlist
->dev
);
2971 free(super
->devlist
);
2972 super
->devlist
= dv
;
2976 static void imsm_copy_dev(struct imsm_dev
*dest
, struct imsm_dev
*src
)
2978 memcpy(dest
, src
, sizeof_imsm_dev(src
, 0));
2981 static int compare_super_imsm(struct supertype
*st
, struct supertype
*tst
)
2985 * 0 same, or first was empty, and second was copied
2986 * 1 second had wrong number
2988 * 3 wrong other info
2990 struct intel_super
*first
= st
->sb
;
2991 struct intel_super
*sec
= tst
->sb
;
2998 /* in platform dependent environment test if the disks
2999 * use the same Intel hba
3001 if (!check_env("IMSM_NO_PLATFORM")) {
3002 if (!first
->hba
|| !sec
->hba
||
3003 (first
->hba
->type
!= sec
->hba
->type
)) {
3005 "HBAs of devices does not match %s != %s\n",
3006 first
->hba
? get_sys_dev_type(first
->hba
->type
) : NULL
,
3007 sec
->hba
? get_sys_dev_type(sec
->hba
->type
) : NULL
);
3012 /* if an anchor does not have num_raid_devs set then it is a free
3015 if (first
->anchor
->num_raid_devs
> 0 &&
3016 sec
->anchor
->num_raid_devs
> 0) {
3017 /* Determine if these disks might ever have been
3018 * related. Further disambiguation can only take place
3019 * in load_super_imsm_all
3021 __u32 first_family
= first
->anchor
->orig_family_num
;
3022 __u32 sec_family
= sec
->anchor
->orig_family_num
;
3024 if (memcmp(first
->anchor
->sig
, sec
->anchor
->sig
,
3025 MAX_SIGNATURE_LENGTH
) != 0)
3028 if (first_family
== 0)
3029 first_family
= first
->anchor
->family_num
;
3030 if (sec_family
== 0)
3031 sec_family
= sec
->anchor
->family_num
;
3033 if (first_family
!= sec_family
)
3039 /* if 'first' is a spare promote it to a populated mpb with sec's
3042 if (first
->anchor
->num_raid_devs
== 0 &&
3043 sec
->anchor
->num_raid_devs
> 0) {
3045 struct intel_dev
*dv
;
3046 struct imsm_dev
*dev
;
3048 /* we need to copy raid device info from sec if an allocation
3049 * fails here we don't associate the spare
3051 for (i
= 0; i
< sec
->anchor
->num_raid_devs
; i
++) {
3052 dv
= malloc(sizeof(*dv
));
3055 dev
= malloc(sizeof_imsm_dev(get_imsm_dev(sec
, i
), 1));
3062 dv
->next
= first
->devlist
;
3063 first
->devlist
= dv
;
3065 if (i
< sec
->anchor
->num_raid_devs
) {
3066 /* allocation failure */
3067 free_devlist(first
);
3068 fprintf(stderr
, "imsm: failed to associate spare\n");
3071 first
->anchor
->num_raid_devs
= sec
->anchor
->num_raid_devs
;
3072 first
->anchor
->orig_family_num
= sec
->anchor
->orig_family_num
;
3073 first
->anchor
->family_num
= sec
->anchor
->family_num
;
3074 memcpy(first
->anchor
->sig
, sec
->anchor
->sig
, MAX_SIGNATURE_LENGTH
);
3075 for (i
= 0; i
< sec
->anchor
->num_raid_devs
; i
++)
3076 imsm_copy_dev(get_imsm_dev(first
, i
), get_imsm_dev(sec
, i
));
3082 static void fd2devname(int fd
, char *name
)
3086 char dname
[PATH_MAX
];
3091 if (fstat(fd
, &st
) != 0)
3093 sprintf(path
, "/sys/dev/block/%d:%d",
3094 major(st
.st_rdev
), minor(st
.st_rdev
));
3096 rv
= readlink(path
, dname
, sizeof(dname
)-1);
3101 nm
= strrchr(dname
, '/');
3104 snprintf(name
, MAX_RAID_SERIAL_LEN
, "/dev/%s", nm
);
3108 extern int scsi_get_serial(int fd
, void *buf
, size_t buf_len
);
3110 static int imsm_read_serial(int fd
, char *devname
,
3111 __u8 serial
[MAX_RAID_SERIAL_LEN
])
3113 unsigned char scsi_serial
[255];
3122 memset(scsi_serial
, 0, sizeof(scsi_serial
));
3124 rv
= scsi_get_serial(fd
, scsi_serial
, sizeof(scsi_serial
));
3126 if (rv
&& check_env("IMSM_DEVNAME_AS_SERIAL")) {
3127 memset(serial
, 0, MAX_RAID_SERIAL_LEN
);
3128 fd2devname(fd
, (char *) serial
);
3135 Name
": Failed to retrieve serial for %s\n",
3140 rsp_len
= scsi_serial
[3];
3144 Name
": Failed to retrieve serial for %s\n",
3148 rsp_buf
= (char *) &scsi_serial
[4];
3150 /* trim all whitespace and non-printable characters and convert
3153 for (i
= 0, dest
= rsp_buf
; i
< rsp_len
; i
++) {
3156 /* ':' is reserved for use in placeholder serial
3157 * numbers for missing disks
3165 len
= dest
- rsp_buf
;
3168 /* truncate leading characters */
3169 if (len
> MAX_RAID_SERIAL_LEN
) {
3170 dest
+= len
- MAX_RAID_SERIAL_LEN
;
3171 len
= MAX_RAID_SERIAL_LEN
;
3174 memset(serial
, 0, MAX_RAID_SERIAL_LEN
);
3175 memcpy(serial
, dest
, len
);
3180 static int serialcmp(__u8
*s1
, __u8
*s2
)
3182 return strncmp((char *) s1
, (char *) s2
, MAX_RAID_SERIAL_LEN
);
3185 static void serialcpy(__u8
*dest
, __u8
*src
)
3187 strncpy((char *) dest
, (char *) src
, MAX_RAID_SERIAL_LEN
);
3190 static struct dl
*serial_to_dl(__u8
*serial
, struct intel_super
*super
)
3194 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
3195 if (serialcmp(dl
->serial
, serial
) == 0)
3201 static struct imsm_disk
*
3202 __serial_to_disk(__u8
*serial
, struct imsm_super
*mpb
, int *idx
)
3206 for (i
= 0; i
< mpb
->num_disks
; i
++) {
3207 struct imsm_disk
*disk
= __get_imsm_disk(mpb
, i
);
3209 if (serialcmp(disk
->serial
, serial
) == 0) {
3220 load_imsm_disk(int fd
, struct intel_super
*super
, char *devname
, int keep_fd
)
3222 struct imsm_disk
*disk
;
3227 __u8 serial
[MAX_RAID_SERIAL_LEN
];
3229 rv
= imsm_read_serial(fd
, devname
, serial
);
3234 dl
= calloc(1, sizeof(*dl
));
3238 Name
": failed to allocate disk buffer for %s\n",
3244 dl
->major
= major(stb
.st_rdev
);
3245 dl
->minor
= minor(stb
.st_rdev
);
3246 dl
->next
= super
->disks
;
3247 dl
->fd
= keep_fd
? fd
: -1;
3248 assert(super
->disks
== NULL
);
3250 serialcpy(dl
->serial
, serial
);
3253 fd2devname(fd
, name
);
3255 dl
->devname
= strdup(devname
);
3257 dl
->devname
= strdup(name
);
3259 /* look up this disk's index in the current anchor */
3260 disk
= __serial_to_disk(dl
->serial
, super
->anchor
, &dl
->index
);
3263 /* only set index on disks that are a member of a
3264 * populated contianer, i.e. one with raid_devs
3266 if (is_failed(&dl
->disk
))
3268 else if (is_spare(&dl
->disk
))
3276 /* When migrating map0 contains the 'destination' state while map1
3277 * contains the current state. When not migrating map0 contains the
3278 * current state. This routine assumes that map[0].map_state is set to
3279 * the current array state before being called.
3281 * Migration is indicated by one of the following states
3282 * 1/ Idle (migr_state=0 map0state=normal||unitialized||degraded||failed)
3283 * 2/ Initialize (migr_state=1 migr_type=MIGR_INIT map0state=normal
3284 * map1state=unitialized)
3285 * 3/ Repair (Resync) (migr_state=1 migr_type=MIGR_REPAIR map0state=normal
3287 * 4/ Rebuild (migr_state=1 migr_type=MIGR_REBUILD map0state=normal
3288 * map1state=degraded)
3289 * 5/ Migration (mig_state=1 migr_type=MIGR_GEN_MIGR map0state=normal
3292 static void migrate(struct imsm_dev
*dev
, struct intel_super
*super
,
3293 __u8 to_state
, int migr_type
)
3295 struct imsm_map
*dest
;
3296 struct imsm_map
*src
= get_imsm_map(dev
, MAP_0
);
3298 dev
->vol
.migr_state
= 1;
3299 set_migr_type(dev
, migr_type
);
3300 dev
->vol
.curr_migr_unit
= 0;
3301 dest
= get_imsm_map(dev
, MAP_1
);
3303 /* duplicate and then set the target end state in map[0] */
3304 memcpy(dest
, src
, sizeof_imsm_map(src
));
3305 if ((migr_type
== MIGR_REBUILD
) ||
3306 (migr_type
== MIGR_GEN_MIGR
)) {
3310 for (i
= 0; i
< src
->num_members
; i
++) {
3311 ord
= __le32_to_cpu(src
->disk_ord_tbl
[i
]);
3312 set_imsm_ord_tbl_ent(src
, i
, ord_to_idx(ord
));
3316 if (migr_type
== MIGR_GEN_MIGR
)
3317 /* Clear migration record */
3318 memset(super
->migr_rec
, 0, sizeof(struct migr_record
));
3320 src
->map_state
= to_state
;
3323 static void end_migration(struct imsm_dev
*dev
, struct intel_super
*super
,
3326 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
3327 struct imsm_map
*prev
= get_imsm_map(dev
, dev
->vol
.migr_state
== 0 ?
3331 /* merge any IMSM_ORD_REBUILD bits that were not successfully
3332 * completed in the last migration.
3334 * FIXME add support for raid-level-migration
3336 if ((map_state
!= map
->map_state
) && (is_gen_migration(dev
) == 0) &&
3337 (prev
->map_state
!= IMSM_T_STATE_UNINITIALIZED
)) {
3338 /* when final map state is other than expected
3339 * merge maps (not for migration)
3343 for (i
= 0; i
< prev
->num_members
; i
++)
3344 for (j
= 0; j
< map
->num_members
; j
++)
3345 /* during online capacity expansion
3346 * disks position can be changed
3347 * if takeover is used
3349 if (ord_to_idx(map
->disk_ord_tbl
[j
]) ==
3350 ord_to_idx(prev
->disk_ord_tbl
[i
])) {
3351 map
->disk_ord_tbl
[j
] |=
3352 prev
->disk_ord_tbl
[i
];
3355 failed
= imsm_count_failed(super
, dev
, MAP_0
);
3356 map_state
= imsm_check_degraded(super
, dev
, failed
, MAP_0
);
3359 dev
->vol
.migr_state
= 0;
3360 set_migr_type(dev
, 0);
3361 dev
->vol
.curr_migr_unit
= 0;
3362 map
->map_state
= map_state
;
3366 static int parse_raid_devices(struct intel_super
*super
)
3369 struct imsm_dev
*dev_new
;
3370 size_t len
, len_migr
;
3372 size_t space_needed
= 0;
3373 struct imsm_super
*mpb
= super
->anchor
;
3375 for (i
= 0; i
< super
->anchor
->num_raid_devs
; i
++) {
3376 struct imsm_dev
*dev_iter
= __get_imsm_dev(super
->anchor
, i
);
3377 struct intel_dev
*dv
;
3379 len
= sizeof_imsm_dev(dev_iter
, 0);
3380 len_migr
= sizeof_imsm_dev(dev_iter
, 1);
3382 space_needed
+= len_migr
- len
;
3384 dv
= malloc(sizeof(*dv
));
3387 if (max_len
< len_migr
)
3389 if (max_len
> len_migr
)
3390 space_needed
+= max_len
- len_migr
;
3391 dev_new
= malloc(max_len
);
3396 imsm_copy_dev(dev_new
, dev_iter
);
3399 dv
->next
= super
->devlist
;
3400 super
->devlist
= dv
;
3403 /* ensure that super->buf is large enough when all raid devices
3406 if (__le32_to_cpu(mpb
->mpb_size
) + space_needed
> super
->len
) {
3409 len
= ROUND_UP(__le32_to_cpu(mpb
->mpb_size
) + space_needed
, 512);
3410 if (posix_memalign(&buf
, 512, len
) != 0)
3413 memcpy(buf
, super
->buf
, super
->len
);
3414 memset(buf
+ super
->len
, 0, len
- super
->len
);
3423 /* retrieve a pointer to the bbm log which starts after all raid devices */
3424 struct bbm_log
*__get_imsm_bbm_log(struct imsm_super
*mpb
)
3428 if (__le32_to_cpu(mpb
->bbm_log_size
)) {
3430 ptr
+= mpb
->mpb_size
- __le32_to_cpu(mpb
->bbm_log_size
);
3436 /*******************************************************************************
3437 * Function: check_mpb_migr_compatibility
3438 * Description: Function checks for unsupported migration features:
3439 * - migration optimization area (pba_of_lba0)
3440 * - descending reshape (ascending_migr)
3442 * super : imsm metadata information
3444 * 0 : migration is compatible
3445 * -1 : migration is not compatible
3446 ******************************************************************************/
3447 int check_mpb_migr_compatibility(struct intel_super
*super
)
3449 struct imsm_map
*map0
, *map1
;
3450 struct migr_record
*migr_rec
= super
->migr_rec
;
3453 for (i
= 0; i
< super
->anchor
->num_raid_devs
; i
++) {
3454 struct imsm_dev
*dev_iter
= __get_imsm_dev(super
->anchor
, i
);
3457 dev_iter
->vol
.migr_state
== 1 &&
3458 dev_iter
->vol
.migr_type
== MIGR_GEN_MIGR
) {
3459 /* This device is migrating */
3460 map0
= get_imsm_map(dev_iter
, MAP_0
);
3461 map1
= get_imsm_map(dev_iter
, MAP_1
);
3462 if (pba_of_lba0(map0
) != pba_of_lba0(map1
))
3463 /* migration optimization area was used */
3465 if (migr_rec
->ascending_migr
== 0
3466 && migr_rec
->dest_depth_per_unit
> 0)
3467 /* descending reshape not supported yet */
3474 static void __free_imsm(struct intel_super
*super
, int free_disks
);
3476 /* load_imsm_mpb - read matrix metadata
3477 * allocates super->mpb to be freed by free_imsm
3479 static int load_imsm_mpb(int fd
, struct intel_super
*super
, char *devname
)
3481 unsigned long long dsize
;
3482 unsigned long long sectors
;
3484 struct imsm_super
*anchor
;
3487 get_dev_size(fd
, NULL
, &dsize
);
3491 Name
": %s: device to small for imsm\n",
3496 if (lseek64(fd
, dsize
- (512 * 2), SEEK_SET
) < 0) {
3498 fprintf(stderr
, Name
3499 ": Cannot seek to anchor block on %s: %s\n",
3500 devname
, strerror(errno
));
3504 if (posix_memalign((void**)&anchor
, 512, 512) != 0) {
3507 Name
": Failed to allocate imsm anchor buffer"
3508 " on %s\n", devname
);
3511 if (read(fd
, anchor
, 512) != 512) {
3514 Name
": Cannot read anchor block on %s: %s\n",
3515 devname
, strerror(errno
));
3520 if (strncmp((char *) anchor
->sig
, MPB_SIGNATURE
, MPB_SIG_LEN
) != 0) {
3523 Name
": no IMSM anchor on %s\n", devname
);
3528 __free_imsm(super
, 0);
3529 /* reload capability and hba */
3531 /* capability and hba must be updated with new super allocation */
3532 find_intel_hba_capability(fd
, super
, devname
);
3533 super
->len
= ROUND_UP(anchor
->mpb_size
, 512);
3534 if (posix_memalign(&super
->buf
, 512, super
->len
) != 0) {
3537 Name
": unable to allocate %zu byte mpb buffer\n",
3542 memcpy(super
->buf
, anchor
, 512);
3544 sectors
= mpb_sectors(anchor
) - 1;
3547 if (posix_memalign(&super
->migr_rec_buf
, 512, MIGR_REC_BUF_SIZE
) != 0) {
3548 fprintf(stderr
, Name
3549 ": %s could not allocate migr_rec buffer\n", __func__
);
3553 super
->clean_migration_record_by_mdmon
= 0;
3556 check_sum
= __gen_imsm_checksum(super
->anchor
);
3557 if (check_sum
!= __le32_to_cpu(super
->anchor
->check_sum
)) {
3560 Name
": IMSM checksum %x != %x on %s\n",
3562 __le32_to_cpu(super
->anchor
->check_sum
),
3570 /* read the extended mpb */
3571 if (lseek64(fd
, dsize
- (512 * (2 + sectors
)), SEEK_SET
) < 0) {
3574 Name
": Cannot seek to extended mpb on %s: %s\n",
3575 devname
, strerror(errno
));
3579 if ((unsigned)read(fd
, super
->buf
+ 512, super
->len
- 512) != super
->len
- 512) {
3582 Name
": Cannot read extended mpb on %s: %s\n",
3583 devname
, strerror(errno
));
3587 check_sum
= __gen_imsm_checksum(super
->anchor
);
3588 if (check_sum
!= __le32_to_cpu(super
->anchor
->check_sum
)) {
3591 Name
": IMSM checksum %x != %x on %s\n",
3592 check_sum
, __le32_to_cpu(super
->anchor
->check_sum
),
3597 /* FIXME the BBM log is disk specific so we cannot use this global
3598 * buffer for all disks. Ok for now since we only look at the global
3599 * bbm_log_size parameter to gate assembly
3601 super
->bbm_log
= __get_imsm_bbm_log(super
->anchor
);
3606 static int read_imsm_migr_rec(int fd
, struct intel_super
*super
);
3608 /* clears hi bits in metadata if MPB_ATTRIB_2TB_DISK not set */
3609 static void clear_hi(struct intel_super
*super
)
3611 struct imsm_super
*mpb
= super
->anchor
;
3613 if (mpb
->attributes
& MPB_ATTRIB_2TB_DISK
)
3615 for (i
= 0; i
< mpb
->num_disks
; ++i
) {
3616 struct imsm_disk
*disk
= &mpb
->disk
[i
];
3617 disk
->total_blocks_hi
= 0;
3619 for (i
= 0; i
< mpb
->num_raid_devs
; ++i
) {
3620 struct imsm_dev
*dev
= get_imsm_dev(super
, i
);
3623 for (n
= 0; n
< 2; ++n
) {
3624 struct imsm_map
*map
= get_imsm_map(dev
, n
);
3627 map
->pba_of_lba0_hi
= 0;
3628 map
->blocks_per_member_hi
= 0;
3629 map
->num_data_stripes_hi
= 0;
3635 load_and_parse_mpb(int fd
, struct intel_super
*super
, char *devname
, int keep_fd
)
3639 err
= load_imsm_mpb(fd
, super
, devname
);
3642 err
= load_imsm_disk(fd
, super
, devname
, keep_fd
);
3645 err
= parse_raid_devices(super
);
3650 static void __free_imsm_disk(struct dl
*d
)
3662 static void free_imsm_disks(struct intel_super
*super
)
3666 while (super
->disks
) {
3668 super
->disks
= d
->next
;
3669 __free_imsm_disk(d
);
3671 while (super
->disk_mgmt_list
) {
3672 d
= super
->disk_mgmt_list
;
3673 super
->disk_mgmt_list
= d
->next
;
3674 __free_imsm_disk(d
);
3676 while (super
->missing
) {
3678 super
->missing
= d
->next
;
3679 __free_imsm_disk(d
);
3684 /* free all the pieces hanging off of a super pointer */
3685 static void __free_imsm(struct intel_super
*super
, int free_disks
)
3687 struct intel_hba
*elem
, *next
;
3693 /* unlink capability description */
3695 if (super
->migr_rec_buf
) {
3696 free(super
->migr_rec_buf
);
3697 super
->migr_rec_buf
= NULL
;
3700 free_imsm_disks(super
);
3701 free_devlist(super
);
3705 free((void *)elem
->path
);
3713 static void free_imsm(struct intel_super
*super
)
3715 __free_imsm(super
, 1);
3719 static void free_super_imsm(struct supertype
*st
)
3721 struct intel_super
*super
= st
->sb
;
3730 static struct intel_super
*alloc_super(void)
3732 struct intel_super
*super
= malloc(sizeof(*super
));
3735 memset(super
, 0, sizeof(*super
));
3736 super
->current_vol
= -1;
3737 super
->create_offset
= ~((unsigned long long) 0);
3743 * find and allocate hba and OROM/EFI based on valid fd of RAID component device
3745 static int find_intel_hba_capability(int fd
, struct intel_super
*super
, char *devname
)
3747 struct sys_dev
*hba_name
;
3750 if ((fd
< 0) || check_env("IMSM_NO_PLATFORM")) {
3755 hba_name
= find_disk_attached_hba(fd
, NULL
);
3759 Name
": %s is not attached to Intel(R) RAID controller.\n",
3763 rv
= attach_hba_to_super(super
, hba_name
);
3766 struct intel_hba
*hba
= super
->hba
;
3768 fprintf(stderr
, Name
": %s is attached to Intel(R) %s RAID "
3769 "controller (%s),\n"
3770 " but the container is assigned to Intel(R) "
3771 "%s RAID controller (",
3774 hba_name
->pci_id
? : "Err!",
3775 get_sys_dev_type(hba_name
->type
));
3778 fprintf(stderr
, "%s", hba
->pci_id
? : "Err!");
3780 fprintf(stderr
, ", ");
3784 fprintf(stderr
, ").\n"
3785 " Mixing devices attached to different controllers "
3786 "is not allowed.\n");
3788 free_sys_dev(&hba_name
);
3791 super
->orom
= find_imsm_capability(hba_name
->type
);
3792 free_sys_dev(&hba_name
);
3798 /* find_missing - helper routine for load_super_imsm_all that identifies
3799 * disks that have disappeared from the system. This routine relies on
3800 * the mpb being uptodate, which it is at load time.
3802 static int find_missing(struct intel_super
*super
)
3805 struct imsm_super
*mpb
= super
->anchor
;
3807 struct imsm_disk
*disk
;
3809 for (i
= 0; i
< mpb
->num_disks
; i
++) {
3810 disk
= __get_imsm_disk(mpb
, i
);
3811 dl
= serial_to_dl(disk
->serial
, super
);
3815 dl
= malloc(sizeof(*dl
));
3821 dl
->devname
= strdup("missing");
3823 serialcpy(dl
->serial
, disk
->serial
);
3826 dl
->next
= super
->missing
;
3827 super
->missing
= dl
;
3834 static struct intel_disk
*disk_list_get(__u8
*serial
, struct intel_disk
*disk_list
)
3836 struct intel_disk
*idisk
= disk_list
;
3839 if (serialcmp(idisk
->disk
.serial
, serial
) == 0)
3841 idisk
= idisk
->next
;
3847 static int __prep_thunderdome(struct intel_super
**table
, int tbl_size
,
3848 struct intel_super
*super
,
3849 struct intel_disk
**disk_list
)
3851 struct imsm_disk
*d
= &super
->disks
->disk
;
3852 struct imsm_super
*mpb
= super
->anchor
;
3855 for (i
= 0; i
< tbl_size
; i
++) {
3856 struct imsm_super
*tbl_mpb
= table
[i
]->anchor
;
3857 struct imsm_disk
*tbl_d
= &table
[i
]->disks
->disk
;
3859 if (tbl_mpb
->family_num
== mpb
->family_num
) {
3860 if (tbl_mpb
->check_sum
== mpb
->check_sum
) {
3861 dprintf("%s: mpb from %d:%d matches %d:%d\n",
3862 __func__
, super
->disks
->major
,
3863 super
->disks
->minor
,
3864 table
[i
]->disks
->major
,
3865 table
[i
]->disks
->minor
);
3869 if (((is_configured(d
) && !is_configured(tbl_d
)) ||
3870 is_configured(d
) == is_configured(tbl_d
)) &&
3871 tbl_mpb
->generation_num
< mpb
->generation_num
) {
3872 /* current version of the mpb is a
3873 * better candidate than the one in
3874 * super_table, but copy over "cross
3875 * generational" status
3877 struct intel_disk
*idisk
;
3879 dprintf("%s: mpb from %d:%d replaces %d:%d\n",
3880 __func__
, super
->disks
->major
,
3881 super
->disks
->minor
,
3882 table
[i
]->disks
->major
,
3883 table
[i
]->disks
->minor
);
3885 idisk
= disk_list_get(tbl_d
->serial
, *disk_list
);
3886 if (idisk
&& is_failed(&idisk
->disk
))
3887 tbl_d
->status
|= FAILED_DISK
;
3890 struct intel_disk
*idisk
;
3891 struct imsm_disk
*disk
;
3893 /* tbl_mpb is more up to date, but copy
3894 * over cross generational status before
3897 disk
= __serial_to_disk(d
->serial
, mpb
, NULL
);
3898 if (disk
&& is_failed(disk
))
3899 d
->status
|= FAILED_DISK
;
3901 idisk
= disk_list_get(d
->serial
, *disk_list
);
3904 if (disk
&& is_configured(disk
))
3905 idisk
->disk
.status
|= CONFIGURED_DISK
;
3908 dprintf("%s: mpb from %d:%d prefer %d:%d\n",
3909 __func__
, super
->disks
->major
,
3910 super
->disks
->minor
,
3911 table
[i
]->disks
->major
,
3912 table
[i
]->disks
->minor
);
3920 table
[tbl_size
++] = super
;
3924 /* update/extend the merged list of imsm_disk records */
3925 for (j
= 0; j
< mpb
->num_disks
; j
++) {
3926 struct imsm_disk
*disk
= __get_imsm_disk(mpb
, j
);
3927 struct intel_disk
*idisk
;
3929 idisk
= disk_list_get(disk
->serial
, *disk_list
);
3931 idisk
->disk
.status
|= disk
->status
;
3932 if (is_configured(&idisk
->disk
) ||
3933 is_failed(&idisk
->disk
))
3934 idisk
->disk
.status
&= ~(SPARE_DISK
);
3936 idisk
= calloc(1, sizeof(*idisk
));
3939 idisk
->owner
= IMSM_UNKNOWN_OWNER
;
3940 idisk
->disk
= *disk
;
3941 idisk
->next
= *disk_list
;
3945 if (serialcmp(idisk
->disk
.serial
, d
->serial
) == 0)
3952 static struct intel_super
*
3953 validate_members(struct intel_super
*super
, struct intel_disk
*disk_list
,
3956 struct imsm_super
*mpb
= super
->anchor
;
3960 for (i
= 0; i
< mpb
->num_disks
; i
++) {
3961 struct imsm_disk
*disk
= __get_imsm_disk(mpb
, i
);
3962 struct intel_disk
*idisk
;
3964 idisk
= disk_list_get(disk
->serial
, disk_list
);
3966 if (idisk
->owner
== owner
||
3967 idisk
->owner
== IMSM_UNKNOWN_OWNER
)
3970 dprintf("%s: '%.16s' owner %d != %d\n",
3971 __func__
, disk
->serial
, idisk
->owner
,
3974 dprintf("%s: unknown disk %x [%d]: %.16s\n",
3975 __func__
, __le32_to_cpu(mpb
->family_num
), i
,
3981 if (ok_count
== mpb
->num_disks
)
3986 static void show_conflicts(__u32 family_num
, struct intel_super
*super_list
)
3988 struct intel_super
*s
;
3990 for (s
= super_list
; s
; s
= s
->next
) {
3991 if (family_num
!= s
->anchor
->family_num
)
3993 fprintf(stderr
, "Conflict, offlining family %#x on '%s'\n",
3994 __le32_to_cpu(family_num
), s
->disks
->devname
);
3998 static struct intel_super
*
3999 imsm_thunderdome(struct intel_super
**super_list
, int len
)
4001 struct intel_super
*super_table
[len
];
4002 struct intel_disk
*disk_list
= NULL
;
4003 struct intel_super
*champion
, *spare
;
4004 struct intel_super
*s
, **del
;
4009 memset(super_table
, 0, sizeof(super_table
));
4010 for (s
= *super_list
; s
; s
= s
->next
)
4011 tbl_size
= __prep_thunderdome(super_table
, tbl_size
, s
, &disk_list
);
4013 for (i
= 0; i
< tbl_size
; i
++) {
4014 struct imsm_disk
*d
;
4015 struct intel_disk
*idisk
;
4016 struct imsm_super
*mpb
= super_table
[i
]->anchor
;
4019 d
= &s
->disks
->disk
;
4021 /* 'd' must appear in merged disk list for its
4022 * configuration to be valid
4024 idisk
= disk_list_get(d
->serial
, disk_list
);
4025 if (idisk
&& idisk
->owner
== i
)
4026 s
= validate_members(s
, disk_list
, i
);
4031 dprintf("%s: marking family: %#x from %d:%d offline\n",
4032 __func__
, mpb
->family_num
,
4033 super_table
[i
]->disks
->major
,
4034 super_table
[i
]->disks
->minor
);
4038 /* This is where the mdadm implementation differs from the Windows
4039 * driver which has no strict concept of a container. We can only
4040 * assemble one family from a container, so when returning a prodigal
4041 * array member to this system the code will not be able to disambiguate
4042 * the container contents that should be assembled ("foreign" versus
4043 * "local"). It requires user intervention to set the orig_family_num
4044 * to a new value to establish a new container. The Windows driver in
4045 * this situation fixes up the volume name in place and manages the
4046 * foreign array as an independent entity.
4051 for (i
= 0; i
< tbl_size
; i
++) {
4052 struct intel_super
*tbl_ent
= super_table
[i
];
4058 if (tbl_ent
->anchor
->num_raid_devs
== 0) {
4063 if (s
&& !is_spare
) {
4064 show_conflicts(tbl_ent
->anchor
->family_num
, *super_list
);
4066 } else if (!s
&& !is_spare
)
4079 fprintf(stderr
, "Chose family %#x on '%s', "
4080 "assemble conflicts to new container with '--update=uuid'\n",
4081 __le32_to_cpu(s
->anchor
->family_num
), s
->disks
->devname
);
4083 /* collect all dl's onto 'champion', and update them to
4084 * champion's version of the status
4086 for (s
= *super_list
; s
; s
= s
->next
) {
4087 struct imsm_super
*mpb
= champion
->anchor
;
4088 struct dl
*dl
= s
->disks
;
4093 for (i
= 0; i
< mpb
->num_disks
; i
++) {
4094 struct imsm_disk
*disk
;
4096 disk
= __serial_to_disk(dl
->serial
, mpb
, &dl
->index
);
4099 /* only set index on disks that are a member of
4100 * a populated contianer, i.e. one with
4103 if (is_failed(&dl
->disk
))
4105 else if (is_spare(&dl
->disk
))
4111 if (i
>= mpb
->num_disks
) {
4112 struct intel_disk
*idisk
;
4114 idisk
= disk_list_get(dl
->serial
, disk_list
);
4115 if (idisk
&& is_spare(&idisk
->disk
) &&
4116 !is_failed(&idisk
->disk
) && !is_configured(&idisk
->disk
))
4124 dl
->next
= champion
->disks
;
4125 champion
->disks
= dl
;
4129 /* delete 'champion' from super_list */
4130 for (del
= super_list
; *del
; ) {
4131 if (*del
== champion
) {
4132 *del
= (*del
)->next
;
4135 del
= &(*del
)->next
;
4137 champion
->next
= NULL
;
4141 struct intel_disk
*idisk
= disk_list
;
4143 disk_list
= disk_list
->next
;
4152 get_sra_super_block(int fd
, struct intel_super
**super_list
, char *devname
, int *max
, int keep_fd
);
4153 static int get_super_block(struct intel_super
**super_list
, int devnum
, char *devname
,
4154 int major
, int minor
, int keep_fd
);
4156 get_devlist_super_block(struct md_list
*devlist
, struct intel_super
**super_list
,
4157 int *max
, int keep_fd
);
4160 static int load_super_imsm_all(struct supertype
*st
, int fd
, void **sbp
,
4161 char *devname
, struct md_list
*devlist
,
4164 struct intel_super
*super_list
= NULL
;
4165 struct intel_super
*super
= NULL
;
4170 /* 'fd' is an opened container */
4171 err
= get_sra_super_block(fd
, &super_list
, devname
, &i
, keep_fd
);
4173 /* get super block from devlist devices */
4174 err
= get_devlist_super_block(devlist
, &super_list
, &i
, keep_fd
);
4177 /* all mpbs enter, maybe one leaves */
4178 super
= imsm_thunderdome(&super_list
, i
);
4184 if (find_missing(super
) != 0) {
4190 /* load migration record */
4191 err
= load_imsm_migr_rec(super
, NULL
);
4193 /* migration is in progress,
4194 * but migr_rec cannot be loaded,
4200 /* Check migration compatibility */
4201 if ((err
== 0) && (check_mpb_migr_compatibility(super
) != 0)) {
4202 fprintf(stderr
, Name
": Unsupported migration detected");
4204 fprintf(stderr
, " on %s\n", devname
);
4206 fprintf(stderr
, " (IMSM).\n");
4215 while (super_list
) {
4216 struct intel_super
*s
= super_list
;
4218 super_list
= super_list
->next
;
4228 st
->container_dev
= fd2devnum(fd
);
4230 st
->container_dev
= NoMdDev
;
4231 if (err
== 0 && st
->ss
== NULL
) {
4232 st
->ss
= &super_imsm
;
4233 st
->minor_version
= 0;
4234 st
->max_devs
= IMSM_MAX_DEVICES
;
4241 get_devlist_super_block(struct md_list
*devlist
, struct intel_super
**super_list
,
4242 int *max
, int keep_fd
)
4244 struct md_list
*tmpdev
;
4248 for (i
= 0, tmpdev
= devlist
; tmpdev
; tmpdev
= tmpdev
->next
) {
4249 if (tmpdev
->used
!= 1)
4251 if (tmpdev
->container
== 1) {
4253 int fd
= dev_open(tmpdev
->devname
, O_RDONLY
|O_EXCL
);
4255 fprintf(stderr
, Name
": cannot open device %s: %s\n",
4256 tmpdev
->devname
, strerror(errno
));
4260 err
= get_sra_super_block(fd
, super_list
,
4261 tmpdev
->devname
, &lmax
,
4270 int major
= major(tmpdev
->st_rdev
);
4271 int minor
= minor(tmpdev
->st_rdev
);
4272 err
= get_super_block(super_list
,
4289 static int get_super_block(struct intel_super
**super_list
, int devnum
, char *devname
,
4290 int major
, int minor
, int keep_fd
)
4292 struct intel_super
*s
= NULL
;
4305 sprintf(nm
, "%d:%d", major
, minor
);
4306 dfd
= dev_open(nm
, O_RDWR
);
4312 rv
= find_intel_hba_capability(dfd
, s
, devname
);
4313 /* no orom/efi or non-intel hba of the disk */
4319 err
= load_and_parse_mpb(dfd
, s
, NULL
, keep_fd
);
4321 /* retry the load if we might have raced against mdmon */
4322 if (err
== 3 && (devnum
!= -1) && mdmon_running(devnum
))
4323 for (retry
= 0; retry
< 3; retry
++) {
4325 err
= load_and_parse_mpb(dfd
, s
, NULL
, keep_fd
);
4331 s
->next
= *super_list
;
4339 if ((dfd
>= 0) && (!keep_fd
))
4346 get_sra_super_block(int fd
, struct intel_super
**super_list
, char *devname
, int *max
, int keep_fd
)
4353 sra
= sysfs_read(fd
, 0, GET_LEVEL
|GET_VERSION
|GET_DEVS
|GET_STATE
);
4357 if (sra
->array
.major_version
!= -1 ||
4358 sra
->array
.minor_version
!= -2 ||
4359 strcmp(sra
->text_version
, "imsm") != 0) {
4364 devnum
= fd2devnum(fd
);
4365 for (sd
= sra
->devs
, i
= 0; sd
; sd
= sd
->next
, i
++) {
4366 if (get_super_block(super_list
, devnum
, devname
,
4367 sd
->disk
.major
, sd
->disk
.minor
, keep_fd
) != 0) {
4378 static int load_container_imsm(struct supertype
*st
, int fd
, char *devname
)
4380 return load_super_imsm_all(st
, fd
, &st
->sb
, devname
, NULL
, 1);
4384 static int load_super_imsm(struct supertype
*st
, int fd
, char *devname
)
4386 struct intel_super
*super
;
4389 if (test_partition(fd
))
4390 /* IMSM not allowed on partitions */
4393 free_super_imsm(st
);
4395 super
= alloc_super();
4398 Name
": malloc of %zu failed.\n",
4402 /* Load hba and capabilities if they exist.
4403 * But do not preclude loading metadata in case capabilities or hba are
4404 * non-compliant and ignore_hw_compat is set.
4406 rv
= find_intel_hba_capability(fd
, super
, devname
);
4407 /* no orom/efi or non-intel hba of the disk */
4408 if ((rv
!= 0) && (st
->ignore_hw_compat
== 0)) {
4411 Name
": No OROM/EFI properties for %s\n", devname
);
4415 rv
= load_and_parse_mpb(fd
, super
, devname
, 0);
4420 Name
": Failed to load all information "
4421 "sections on %s\n", devname
);
4427 if (st
->ss
== NULL
) {
4428 st
->ss
= &super_imsm
;
4429 st
->minor_version
= 0;
4430 st
->max_devs
= IMSM_MAX_DEVICES
;
4433 /* load migration record */
4434 if (load_imsm_migr_rec(super
, NULL
) == 0) {
4435 /* Check for unsupported migration features */
4436 if (check_mpb_migr_compatibility(super
) != 0) {
4438 Name
": Unsupported migration detected");
4440 fprintf(stderr
, " on %s\n", devname
);
4442 fprintf(stderr
, " (IMSM).\n");
4450 static __u16
info_to_blocks_per_strip(mdu_array_info_t
*info
)
4452 if (info
->level
== 1)
4454 return info
->chunk_size
>> 9;
4457 static unsigned long long info_to_blocks_per_member(mdu_array_info_t
*info
,
4458 unsigned long long size
)
4460 if (info
->level
== 1)
4463 return (size
* 2) & ~(info_to_blocks_per_strip(info
) - 1);
4466 static void imsm_update_version_info(struct intel_super
*super
)
4468 /* update the version and attributes */
4469 struct imsm_super
*mpb
= super
->anchor
;
4471 struct imsm_dev
*dev
;
4472 struct imsm_map
*map
;
4475 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
4476 dev
= get_imsm_dev(super
, i
);
4477 map
= get_imsm_map(dev
, MAP_0
);
4478 if (__le32_to_cpu(dev
->size_high
) > 0)
4479 mpb
->attributes
|= MPB_ATTRIB_2TB
;
4481 /* FIXME detect when an array spans a port multiplier */
4483 mpb
->attributes
|= MPB_ATTRIB_PM
;
4486 if (mpb
->num_raid_devs
> 1 ||
4487 mpb
->attributes
!= MPB_ATTRIB_CHECKSUM_VERIFY
) {
4488 version
= MPB_VERSION_ATTRIBS
;
4489 switch (get_imsm_raid_level(map
)) {
4490 case 0: mpb
->attributes
|= MPB_ATTRIB_RAID0
; break;
4491 case 1: mpb
->attributes
|= MPB_ATTRIB_RAID1
; break;
4492 case 10: mpb
->attributes
|= MPB_ATTRIB_RAID10
; break;
4493 case 5: mpb
->attributes
|= MPB_ATTRIB_RAID5
; break;
4496 if (map
->num_members
>= 5)
4497 version
= MPB_VERSION_5OR6_DISK_ARRAY
;
4498 else if (dev
->status
== DEV_CLONE_N_GO
)
4499 version
= MPB_VERSION_CNG
;
4500 else if (get_imsm_raid_level(map
) == 5)
4501 version
= MPB_VERSION_RAID5
;
4502 else if (map
->num_members
>= 3)
4503 version
= MPB_VERSION_3OR4_DISK_ARRAY
;
4504 else if (get_imsm_raid_level(map
) == 1)
4505 version
= MPB_VERSION_RAID1
;
4507 version
= MPB_VERSION_RAID0
;
4509 strcpy(((char *) mpb
->sig
) + strlen(MPB_SIGNATURE
), version
);
4513 static int check_name(struct intel_super
*super
, char *name
, int quiet
)
4515 struct imsm_super
*mpb
= super
->anchor
;
4516 char *reason
= NULL
;
4519 if (strlen(name
) > MAX_RAID_SERIAL_LEN
)
4520 reason
= "must be 16 characters or less";
4522 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
4523 struct imsm_dev
*dev
= get_imsm_dev(super
, i
);
4525 if (strncmp((char *) dev
->volume
, name
, MAX_RAID_SERIAL_LEN
) == 0) {
4526 reason
= "already exists";
4531 if (reason
&& !quiet
)
4532 fprintf(stderr
, Name
": imsm volume name %s\n", reason
);
4537 static int init_super_imsm_volume(struct supertype
*st
, mdu_array_info_t
*info
,
4538 unsigned long long size
, char *name
,
4539 char *homehost
, int *uuid
)
4541 /* We are creating a volume inside a pre-existing container.
4542 * so st->sb is already set.
4544 struct intel_super
*super
= st
->sb
;
4545 struct imsm_super
*mpb
= super
->anchor
;
4546 struct intel_dev
*dv
;
4547 struct imsm_dev
*dev
;
4548 struct imsm_vol
*vol
;
4549 struct imsm_map
*map
;
4550 int idx
= mpb
->num_raid_devs
;
4552 unsigned long long array_blocks
;
4553 size_t size_old
, size_new
;
4554 unsigned long long num_data_stripes
;
4556 if (super
->orom
&& mpb
->num_raid_devs
>= super
->orom
->vpa
) {
4557 fprintf(stderr
, Name
": This imsm-container already has the "
4558 "maximum of %d volumes\n", super
->orom
->vpa
);
4562 /* ensure the mpb is large enough for the new data */
4563 size_old
= __le32_to_cpu(mpb
->mpb_size
);
4564 size_new
= disks_to_mpb_size(info
->nr_disks
);
4565 if (size_new
> size_old
) {
4567 size_t size_round
= ROUND_UP(size_new
, 512);
4569 if (posix_memalign(&mpb_new
, 512, size_round
) != 0) {
4570 fprintf(stderr
, Name
": could not allocate new mpb\n");
4573 if (posix_memalign(&super
->migr_rec_buf
, 512,
4574 MIGR_REC_BUF_SIZE
) != 0) {
4575 fprintf(stderr
, Name
4576 ": %s could not allocate migr_rec buffer\n",
4583 memcpy(mpb_new
, mpb
, size_old
);
4586 super
->anchor
= mpb_new
;
4587 mpb
->mpb_size
= __cpu_to_le32(size_new
);
4588 memset(mpb_new
+ size_old
, 0, size_round
- size_old
);
4590 super
->current_vol
= idx
;
4592 /* handle 'failed_disks' by either:
4593 * a) create dummy disk entries in the table if this the first
4594 * volume in the array. We add them here as this is the only
4595 * opportunity to add them. add_to_super_imsm_volume()
4596 * handles the non-failed disks and continues incrementing
4598 * b) validate that 'failed_disks' matches the current number
4599 * of missing disks if the container is populated
4601 if (super
->current_vol
== 0) {
4603 for (i
= 0; i
< info
->failed_disks
; i
++) {
4604 struct imsm_disk
*disk
;
4607 disk
= __get_imsm_disk(mpb
, i
);
4608 disk
->status
= CONFIGURED_DISK
| FAILED_DISK
;
4609 disk
->scsi_id
= __cpu_to_le32(~(__u32
)0);
4610 snprintf((char *) disk
->serial
, MAX_RAID_SERIAL_LEN
,
4613 find_missing(super
);
4618 for (d
= super
->missing
; d
; d
= d
->next
)
4620 if (info
->failed_disks
> missing
) {
4621 fprintf(stderr
, Name
": unable to add 'missing' disk to container\n");
4626 if (!check_name(super
, name
, 0))
4628 dv
= malloc(sizeof(*dv
));
4630 fprintf(stderr
, Name
": failed to allocate device list entry\n");
4633 dev
= calloc(1, sizeof(*dev
) + sizeof(__u32
) * (info
->raid_disks
- 1));
4636 fprintf(stderr
, Name
": could not allocate raid device\n");
4640 strncpy((char *) dev
->volume
, name
, MAX_RAID_SERIAL_LEN
);
4641 array_blocks
= calc_array_size(info
->level
, info
->raid_disks
,
4642 info
->layout
, info
->chunk_size
,
4644 /* round array size down to closest MB */
4645 array_blocks
= (array_blocks
>> SECT_PER_MB_SHIFT
) << SECT_PER_MB_SHIFT
;
4647 dev
->size_low
= __cpu_to_le32((__u32
) array_blocks
);
4648 dev
->size_high
= __cpu_to_le32((__u32
) (array_blocks
>> 32));
4649 dev
->status
= (DEV_READ_COALESCING
| DEV_WRITE_COALESCING
);
4651 vol
->migr_state
= 0;
4652 set_migr_type(dev
, MIGR_INIT
);
4653 vol
->dirty
= !info
->state
;
4654 vol
->curr_migr_unit
= 0;
4655 map
= get_imsm_map(dev
, MAP_0
);
4656 set_pba_of_lba0(map
, super
->create_offset
);
4657 set_blocks_per_member(map
, info_to_blocks_per_member(info
, size
));
4658 map
->blocks_per_strip
= __cpu_to_le16(info_to_blocks_per_strip(info
));
4659 map
->failed_disk_num
= ~0;
4660 if (info
->level
> 0)
4661 map
->map_state
= IMSM_T_STATE_UNINITIALIZED
;
4663 map
->map_state
= info
->failed_disks
? IMSM_T_STATE_FAILED
:
4664 IMSM_T_STATE_NORMAL
;
4667 if (info
->level
== 1 && info
->raid_disks
> 2) {
4670 fprintf(stderr
, Name
": imsm does not support more than 2 disks"
4671 "in a raid1 volume\n");
4675 map
->raid_level
= info
->level
;
4676 if (info
->level
== 10) {
4677 map
->raid_level
= 1;
4678 map
->num_domains
= info
->raid_disks
/ 2;
4679 } else if (info
->level
== 1)
4680 map
->num_domains
= info
->raid_disks
;
4682 map
->num_domains
= 1;
4684 /* info->size is only int so use the 'size' parameter instead */
4685 num_data_stripes
= (size
* 2) / info_to_blocks_per_strip(info
);
4686 num_data_stripes
/= map
->num_domains
;
4687 set_num_data_stripes(map
, num_data_stripes
);
4689 map
->num_members
= info
->raid_disks
;
4690 for (i
= 0; i
< map
->num_members
; i
++) {
4691 /* initialized in add_to_super */
4692 set_imsm_ord_tbl_ent(map
, i
, IMSM_ORD_REBUILD
);
4694 mpb
->num_raid_devs
++;
4697 dv
->index
= super
->current_vol
;
4698 dv
->next
= super
->devlist
;
4699 super
->devlist
= dv
;
4701 imsm_update_version_info(super
);
4706 static int init_super_imsm(struct supertype
*st
, mdu_array_info_t
*info
,
4707 unsigned long long size
, char *name
,
4708 char *homehost
, int *uuid
)
4710 /* This is primarily called by Create when creating a new array.
4711 * We will then get add_to_super called for each component, and then
4712 * write_init_super called to write it out to each device.
4713 * For IMSM, Create can create on fresh devices or on a pre-existing
4715 * To create on a pre-existing array a different method will be called.
4716 * This one is just for fresh drives.
4718 struct intel_super
*super
;
4719 struct imsm_super
*mpb
;
4724 return init_super_imsm_volume(st
, info
, size
, name
, homehost
, uuid
);
4727 mpb_size
= disks_to_mpb_size(info
->nr_disks
);
4731 super
= alloc_super();
4732 if (super
&& posix_memalign(&super
->buf
, 512, mpb_size
) != 0) {
4737 fprintf(stderr
, Name
4738 ": %s could not allocate superblock\n", __func__
);
4741 if (posix_memalign(&super
->migr_rec_buf
, 512, MIGR_REC_BUF_SIZE
) != 0) {
4742 fprintf(stderr
, Name
4743 ": %s could not allocate migr_rec buffer\n", __func__
);
4748 memset(super
->buf
, 0, mpb_size
);
4750 mpb
->mpb_size
= __cpu_to_le32(mpb_size
);
4754 /* zeroing superblock */
4758 mpb
->attributes
= MPB_ATTRIB_CHECKSUM_VERIFY
;
4760 version
= (char *) mpb
->sig
;
4761 strcpy(version
, MPB_SIGNATURE
);
4762 version
+= strlen(MPB_SIGNATURE
);
4763 strcpy(version
, MPB_VERSION_RAID0
);
4769 static int add_to_super_imsm_volume(struct supertype
*st
, mdu_disk_info_t
*dk
,
4770 int fd
, char *devname
)
4772 struct intel_super
*super
= st
->sb
;
4773 struct imsm_super
*mpb
= super
->anchor
;
4774 struct imsm_disk
*_disk
;
4775 struct imsm_dev
*dev
;
4776 struct imsm_map
*map
;
4780 dev
= get_imsm_dev(super
, super
->current_vol
);
4781 map
= get_imsm_map(dev
, MAP_0
);
4783 if (! (dk
->state
& (1<<MD_DISK_SYNC
))) {
4784 fprintf(stderr
, Name
": %s: Cannot add spare devices to IMSM volume\n",
4790 /* we're doing autolayout so grab the pre-marked (in
4791 * validate_geometry) raid_disk
4793 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
4794 if (dl
->raiddisk
== dk
->raid_disk
)
4797 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
4798 if (dl
->major
== dk
->major
&&
4799 dl
->minor
== dk
->minor
)
4804 fprintf(stderr
, Name
": %s is not a member of the same container\n", devname
);
4808 /* add a pristine spare to the metadata */
4809 if (dl
->index
< 0) {
4810 dl
->index
= super
->anchor
->num_disks
;
4811 super
->anchor
->num_disks
++;
4813 /* Check the device has not already been added */
4814 slot
= get_imsm_disk_slot(map
, dl
->index
);
4816 (get_imsm_ord_tbl_ent(dev
, slot
, MAP_X
) & IMSM_ORD_REBUILD
) == 0) {
4817 fprintf(stderr
, Name
": %s has been included in this array twice\n",
4821 set_imsm_ord_tbl_ent(map
, dk
->raid_disk
, dl
->index
);
4822 dl
->disk
.status
= CONFIGURED_DISK
;
4824 /* update size of 'missing' disks to be at least as large as the
4825 * largest acitve member (we only have dummy missing disks when
4826 * creating the first volume)
4828 if (super
->current_vol
== 0) {
4829 for (df
= super
->missing
; df
; df
= df
->next
) {
4830 if (total_blocks(&dl
->disk
) > total_blocks(&df
->disk
))
4831 set_total_blocks(&df
->disk
, total_blocks(&dl
->disk
));
4832 _disk
= __get_imsm_disk(mpb
, df
->index
);
4837 /* refresh unset/failed slots to point to valid 'missing' entries */
4838 for (df
= super
->missing
; df
; df
= df
->next
)
4839 for (slot
= 0; slot
< mpb
->num_disks
; slot
++) {
4840 __u32 ord
= get_imsm_ord_tbl_ent(dev
, slot
, MAP_X
);
4842 if ((ord
& IMSM_ORD_REBUILD
) == 0)
4844 set_imsm_ord_tbl_ent(map
, slot
, df
->index
| IMSM_ORD_REBUILD
);
4845 if (is_gen_migration(dev
)) {
4846 struct imsm_map
*map2
= get_imsm_map(dev
,
4848 int slot2
= get_imsm_disk_slot(map2
, df
->index
);
4849 if ((slot2
< map2
->num_members
) &&
4851 __u32 ord2
= get_imsm_ord_tbl_ent(dev
,
4854 if ((unsigned)df
->index
==
4856 set_imsm_ord_tbl_ent(map2
,
4862 dprintf("set slot:%d to missing disk:%d\n", slot
, df
->index
);
4866 /* if we are creating the first raid device update the family number */
4867 if (super
->current_vol
== 0) {
4869 struct imsm_dev
*_dev
= __get_imsm_dev(mpb
, 0);
4871 _disk
= __get_imsm_disk(mpb
, dl
->index
);
4872 if (!_dev
|| !_disk
) {
4873 fprintf(stderr
, Name
": BUG mpb setup error\n");
4879 sum
+= __gen_imsm_checksum(mpb
);
4880 mpb
->family_num
= __cpu_to_le32(sum
);
4881 mpb
->orig_family_num
= mpb
->family_num
;
4883 super
->current_disk
= dl
;
4888 * Function marks disk as spare and restores disk serial
4889 * in case it was previously marked as failed by takeover operation
4891 * -1 : critical error
4892 * 0 : disk is marked as spare but serial is not set
4895 int mark_spare(struct dl
*disk
)
4897 __u8 serial
[MAX_RAID_SERIAL_LEN
];
4904 if (!imsm_read_serial(disk
->fd
, NULL
, serial
)) {
4905 /* Restore disk serial number, because takeover marks disk
4906 * as failed and adds to serial ':0' before it becomes
4909 serialcpy(disk
->serial
, serial
);
4910 serialcpy(disk
->disk
.serial
, serial
);
4913 disk
->disk
.status
= SPARE_DISK
;
4919 static int add_to_super_imsm(struct supertype
*st
, mdu_disk_info_t
*dk
,
4920 int fd
, char *devname
)
4922 struct intel_super
*super
= st
->sb
;
4924 unsigned long long size
;
4929 /* If we are on an RAID enabled platform check that the disk is
4930 * attached to the raid controller.
4931 * We do not need to test disks attachment for container based additions,
4932 * they shall be already tested when container was created/assembled.
4934 rv
= find_intel_hba_capability(fd
, super
, devname
);
4935 /* no orom/efi or non-intel hba of the disk */
4937 dprintf("capability: %p fd: %d ret: %d\n",
4938 super
->orom
, fd
, rv
);
4942 if (super
->current_vol
>= 0)
4943 return add_to_super_imsm_volume(st
, dk
, fd
, devname
);
4946 dd
= malloc(sizeof(*dd
));
4949 Name
": malloc failed %s:%d.\n", __func__
, __LINE__
);
4952 memset(dd
, 0, sizeof(*dd
));
4953 dd
->major
= major(stb
.st_rdev
);
4954 dd
->minor
= minor(stb
.st_rdev
);
4955 dd
->devname
= devname
? strdup(devname
) : NULL
;
4958 dd
->action
= DISK_ADD
;
4959 rv
= imsm_read_serial(fd
, devname
, dd
->serial
);
4962 Name
": failed to retrieve scsi serial, aborting\n");
4967 get_dev_size(fd
, NULL
, &size
);
4969 serialcpy(dd
->disk
.serial
, dd
->serial
);
4970 set_total_blocks(&dd
->disk
, size
);
4971 if (__le32_to_cpu(dd
->disk
.total_blocks_hi
) > 0) {
4972 struct imsm_super
*mpb
= super
->anchor
;
4973 mpb
->attributes
|= MPB_ATTRIB_2TB_DISK
;
4976 if (sysfs_disk_to_scsi_id(fd
, &id
) == 0)
4977 dd
->disk
.scsi_id
= __cpu_to_le32(id
);
4979 dd
->disk
.scsi_id
= __cpu_to_le32(0);
4981 if (st
->update_tail
) {
4982 dd
->next
= super
->disk_mgmt_list
;
4983 super
->disk_mgmt_list
= dd
;
4985 dd
->next
= super
->disks
;
4987 super
->updates_pending
++;
4994 static int remove_from_super_imsm(struct supertype
*st
, mdu_disk_info_t
*dk
)
4996 struct intel_super
*super
= st
->sb
;
4999 /* remove from super works only in mdmon - for communication
5000 * manager - monitor. Check if communication memory buffer
5003 if (!st
->update_tail
) {
5005 Name
": %s shall be used in mdmon context only"
5006 "(line %d).\n", __func__
, __LINE__
);
5009 dd
= malloc(sizeof(*dd
));
5012 Name
": malloc failed %s:%d.\n", __func__
, __LINE__
);
5015 memset(dd
, 0, sizeof(*dd
));
5016 dd
->major
= dk
->major
;
5017 dd
->minor
= dk
->minor
;
5020 dd
->action
= DISK_REMOVE
;
5022 dd
->next
= super
->disk_mgmt_list
;
5023 super
->disk_mgmt_list
= dd
;
5029 static int store_imsm_mpb(int fd
, struct imsm_super
*mpb
);
5033 struct imsm_super anchor
;
5034 } spare_record
__attribute__ ((aligned(512)));
5036 /* spare records have their own family number and do not have any defined raid
5039 static int write_super_imsm_spares(struct intel_super
*super
, int doclose
)
5041 struct imsm_super
*mpb
= super
->anchor
;
5042 struct imsm_super
*spare
= &spare_record
.anchor
;
5046 spare
->mpb_size
= __cpu_to_le32(sizeof(struct imsm_super
)),
5047 spare
->generation_num
= __cpu_to_le32(1UL),
5048 spare
->attributes
= MPB_ATTRIB_CHECKSUM_VERIFY
;
5049 spare
->num_disks
= 1,
5050 spare
->num_raid_devs
= 0,
5051 spare
->cache_size
= mpb
->cache_size
,
5052 spare
->pwr_cycle_count
= __cpu_to_le32(1),
5054 snprintf((char *) spare
->sig
, MAX_SIGNATURE_LENGTH
,
5055 MPB_SIGNATURE MPB_VERSION_RAID0
);
5057 for (d
= super
->disks
; d
; d
= d
->next
) {
5061 spare
->disk
[0] = d
->disk
;
5062 if (__le32_to_cpu(d
->disk
.total_blocks_hi
) > 0)
5063 spare
->attributes
|= MPB_ATTRIB_2TB_DISK
;
5065 sum
= __gen_imsm_checksum(spare
);
5066 spare
->family_num
= __cpu_to_le32(sum
);
5067 spare
->orig_family_num
= 0;
5068 sum
= __gen_imsm_checksum(spare
);
5069 spare
->check_sum
= __cpu_to_le32(sum
);
5071 if (store_imsm_mpb(d
->fd
, spare
)) {
5072 fprintf(stderr
, "%s: failed for device %d:%d %s\n",
5073 __func__
, d
->major
, d
->minor
, strerror(errno
));
5085 static int write_super_imsm(struct supertype
*st
, int doclose
)
5087 struct intel_super
*super
= st
->sb
;
5088 struct imsm_super
*mpb
= super
->anchor
;
5094 __u32 mpb_size
= sizeof(struct imsm_super
) - sizeof(struct imsm_disk
);
5096 int clear_migration_record
= 1;
5098 /* 'generation' is incremented everytime the metadata is written */
5099 generation
= __le32_to_cpu(mpb
->generation_num
);
5101 mpb
->generation_num
= __cpu_to_le32(generation
);
5103 /* fix up cases where previous mdadm releases failed to set
5106 if (mpb
->orig_family_num
== 0)
5107 mpb
->orig_family_num
= mpb
->family_num
;
5109 for (d
= super
->disks
; d
; d
= d
->next
) {
5113 mpb
->disk
[d
->index
] = d
->disk
;
5117 for (d
= super
->missing
; d
; d
= d
->next
) {
5118 mpb
->disk
[d
->index
] = d
->disk
;
5121 mpb
->num_disks
= num_disks
;
5122 mpb_size
+= sizeof(struct imsm_disk
) * mpb
->num_disks
;
5124 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
5125 struct imsm_dev
*dev
= __get_imsm_dev(mpb
, i
);
5126 struct imsm_dev
*dev2
= get_imsm_dev(super
, i
);
5128 imsm_copy_dev(dev
, dev2
);
5129 mpb_size
+= sizeof_imsm_dev(dev
, 0);
5131 if (is_gen_migration(dev2
))
5132 clear_migration_record
= 0;
5134 mpb_size
+= __le32_to_cpu(mpb
->bbm_log_size
);
5135 mpb
->mpb_size
= __cpu_to_le32(mpb_size
);
5137 /* recalculate checksum */
5138 sum
= __gen_imsm_checksum(mpb
);
5139 mpb
->check_sum
= __cpu_to_le32(sum
);
5141 if (super
->clean_migration_record_by_mdmon
) {
5142 clear_migration_record
= 1;
5143 super
->clean_migration_record_by_mdmon
= 0;
5145 if (clear_migration_record
)
5146 memset(super
->migr_rec_buf
, 0, MIGR_REC_BUF_SIZE
);
5148 /* write the mpb for disks that compose raid devices */
5149 for (d
= super
->disks
; d
; d
= d
->next
) {
5150 if (d
->index
< 0 || is_failed(&d
->disk
))
5153 if (clear_migration_record
) {
5154 unsigned long long dsize
;
5156 get_dev_size(d
->fd
, NULL
, &dsize
);
5157 if (lseek64(d
->fd
, dsize
- 512, SEEK_SET
) >= 0) {
5158 if (write(d
->fd
, super
->migr_rec_buf
,
5159 MIGR_REC_BUF_SIZE
) != MIGR_REC_BUF_SIZE
)
5160 perror("Write migr_rec failed");
5164 if (store_imsm_mpb(d
->fd
, mpb
))
5166 "%s: failed for device %d:%d (fd: %d)%s\n",
5167 __func__
, d
->major
, d
->minor
,
5168 d
->fd
, strerror(errno
));
5177 return write_super_imsm_spares(super
, doclose
);
5183 static int create_array(struct supertype
*st
, int dev_idx
)
5186 struct imsm_update_create_array
*u
;
5187 struct intel_super
*super
= st
->sb
;
5188 struct imsm_dev
*dev
= get_imsm_dev(super
, dev_idx
);
5189 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
5190 struct disk_info
*inf
;
5191 struct imsm_disk
*disk
;
5194 len
= sizeof(*u
) - sizeof(*dev
) + sizeof_imsm_dev(dev
, 0) +
5195 sizeof(*inf
) * map
->num_members
;
5198 fprintf(stderr
, "%s: failed to allocate update buffer\n",
5203 u
->type
= update_create_array
;
5204 u
->dev_idx
= dev_idx
;
5205 imsm_copy_dev(&u
->dev
, dev
);
5206 inf
= get_disk_info(u
);
5207 for (i
= 0; i
< map
->num_members
; i
++) {
5208 int idx
= get_imsm_disk_idx(dev
, i
, MAP_X
);
5210 disk
= get_imsm_disk(super
, idx
);
5211 serialcpy(inf
[i
].serial
, disk
->serial
);
5213 append_metadata_update(st
, u
, len
);
5218 static int mgmt_disk(struct supertype
*st
)
5220 struct intel_super
*super
= st
->sb
;
5222 struct imsm_update_add_remove_disk
*u
;
5224 if (!super
->disk_mgmt_list
)
5230 fprintf(stderr
, "%s: failed to allocate update buffer\n",
5235 u
->type
= update_add_remove_disk
;
5236 append_metadata_update(st
, u
, len
);
5241 static int write_init_super_imsm(struct supertype
*st
)
5243 struct intel_super
*super
= st
->sb
;
5244 int current_vol
= super
->current_vol
;
5246 /* we are done with current_vol reset it to point st at the container */
5247 super
->current_vol
= -1;
5249 if (st
->update_tail
) {
5250 /* queue the recently created array / added disk
5251 * as a metadata update */
5254 /* determine if we are creating a volume or adding a disk */
5255 if (current_vol
< 0) {
5256 /* in the mgmt (add/remove) disk case we are running
5257 * in mdmon context, so don't close fd's
5259 return mgmt_disk(st
);
5261 rv
= create_array(st
, current_vol
);
5266 for (d
= super
->disks
; d
; d
= d
->next
)
5267 Kill(d
->devname
, NULL
, 0, 1, 1);
5268 return write_super_imsm(st
, 1);
5273 static int store_super_imsm(struct supertype
*st
, int fd
)
5275 struct intel_super
*super
= st
->sb
;
5276 struct imsm_super
*mpb
= super
? super
->anchor
: NULL
;
5282 return store_imsm_mpb(fd
, mpb
);
5288 static int imsm_bbm_log_size(struct imsm_super
*mpb
)
5290 return __le32_to_cpu(mpb
->bbm_log_size
);
5294 static int validate_geometry_imsm_container(struct supertype
*st
, int level
,
5295 int layout
, int raiddisks
, int chunk
,
5296 unsigned long long size
, char *dev
,
5297 unsigned long long *freesize
,
5301 unsigned long long ldsize
;
5302 struct intel_super
*super
=NULL
;
5305 if (level
!= LEVEL_CONTAINER
)
5310 fd
= open(dev
, O_RDONLY
|O_EXCL
, 0);
5313 fprintf(stderr
, Name
": imsm: Cannot open %s: %s\n",
5314 dev
, strerror(errno
));
5317 if (!get_dev_size(fd
, dev
, &ldsize
)) {
5322 /* capabilities retrieve could be possible
5323 * note that there is no fd for the disks in array.
5325 super
= alloc_super();
5328 Name
": malloc of %zu failed.\n",
5334 rv
= find_intel_hba_capability(fd
, super
, verbose
? dev
: NULL
);
5338 fd2devname(fd
, str
);
5339 dprintf("validate_geometry_imsm_container: fd: %d %s orom: %p rv: %d raiddisk: %d\n",
5340 fd
, str
, super
->orom
, rv
, raiddisks
);
5342 /* no orom/efi or non-intel hba of the disk */
5349 if (raiddisks
> super
->orom
->tds
) {
5351 fprintf(stderr
, Name
": %d exceeds maximum number of"
5352 " platform supported disks: %d\n",
5353 raiddisks
, super
->orom
->tds
);
5357 if ((super
->orom
->attr
& IMSM_OROM_ATTR_2TB_DISK
) == 0 &&
5358 (ldsize
>> 9) >> 32 > 0) {
5360 fprintf(stderr
, Name
": %s exceeds maximum platform supported size\n", dev
);
5366 *freesize
= avail_size_imsm(st
, ldsize
>> 9);
5372 static unsigned long long find_size(struct extent
*e
, int *idx
, int num_extents
)
5374 const unsigned long long base_start
= e
[*idx
].start
;
5375 unsigned long long end
= base_start
+ e
[*idx
].size
;
5378 if (base_start
== end
)
5382 for (i
= *idx
; i
< num_extents
; i
++) {
5383 /* extend overlapping extents */
5384 if (e
[i
].start
>= base_start
&&
5385 e
[i
].start
<= end
) {
5388 if (e
[i
].start
+ e
[i
].size
> end
)
5389 end
= e
[i
].start
+ e
[i
].size
;
5390 } else if (e
[i
].start
> end
) {
5396 return end
- base_start
;
5399 static unsigned long long merge_extents(struct intel_super
*super
, int sum_extents
)
5401 /* build a composite disk with all known extents and generate a new
5402 * 'maxsize' given the "all disks in an array must share a common start
5403 * offset" constraint
5405 struct extent
*e
= calloc(sum_extents
, sizeof(*e
));
5409 unsigned long long pos
;
5410 unsigned long long start
= 0;
5411 unsigned long long maxsize
;
5412 unsigned long reserve
;
5417 /* coalesce and sort all extents. also, check to see if we need to
5418 * reserve space between member arrays
5421 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
5424 for (i
= 0; i
< dl
->extent_cnt
; i
++)
5427 qsort(e
, sum_extents
, sizeof(*e
), cmp_extent
);
5432 while (i
< sum_extents
) {
5433 e
[j
].start
= e
[i
].start
;
5434 e
[j
].size
= find_size(e
, &i
, sum_extents
);
5436 if (e
[j
-1].size
== 0)
5445 unsigned long long esize
;
5447 esize
= e
[i
].start
- pos
;
5448 if (esize
>= maxsize
) {
5453 pos
= e
[i
].start
+ e
[i
].size
;
5455 } while (e
[i
-1].size
);
5461 /* FIXME assumes volume at offset 0 is the first volume in a
5464 if (start_extent
> 0)
5465 reserve
= IMSM_RESERVED_SECTORS
; /* gap between raid regions */
5469 if (maxsize
< reserve
)
5472 super
->create_offset
= ~((unsigned long long) 0);
5473 if (start
+ reserve
> super
->create_offset
)
5474 return 0; /* start overflows create_offset */
5475 super
->create_offset
= start
+ reserve
;
5477 return maxsize
- reserve
;
5480 static int is_raid_level_supported(const struct imsm_orom
*orom
, int level
, int raiddisks
)
5482 if (level
< 0 || level
== 6 || level
== 4)
5485 /* if we have an orom prevent invalid raid levels */
5488 case 0: return imsm_orom_has_raid0(orom
);
5491 return imsm_orom_has_raid1e(orom
);
5492 return imsm_orom_has_raid1(orom
) && raiddisks
== 2;
5493 case 10: return imsm_orom_has_raid10(orom
) && raiddisks
== 4;
5494 case 5: return imsm_orom_has_raid5(orom
) && raiddisks
> 2;
5497 return 1; /* not on an Intel RAID platform so anything goes */
5504 active_arrays_by_format(char *name
, char* hba
, struct md_list
**devlist
,
5505 int dpa
, int verbose
)
5507 struct mdstat_ent
*mdstat
= mdstat_read(0, 0);
5508 struct mdstat_ent
*memb
= NULL
;
5511 struct md_list
*dv
= NULL
;
5514 for (memb
= mdstat
; memb
; memb
= memb
->next
) {
5515 if (memb
->metadata_version
&&
5516 (strncmp(memb
->metadata_version
, "external:", 9) == 0) &&
5517 (strcmp(&memb
->metadata_version
[9], name
) == 0) &&
5518 !is_subarray(memb
->metadata_version
+9) &&
5520 struct dev_member
*dev
= memb
->members
;
5522 while(dev
&& (fd
< 0)) {
5523 char *path
= malloc(strlen(dev
->name
) + strlen("/dev/") + 1);
5525 num
= sprintf(path
, "%s%s", "/dev/", dev
->name
);
5527 fd
= open(path
, O_RDONLY
, 0);
5528 if ((num
<= 0) || (fd
< 0)) {
5529 pr_vrb(": Cannot open %s: %s\n",
5530 dev
->name
, strerror(errno
));
5537 if ((fd
>= 0) && disk_attached_to_hba(fd
, hba
)) {
5538 struct mdstat_ent
*vol
;
5539 for (vol
= mdstat
; vol
; vol
= vol
->next
) {
5540 if ((vol
->active
> 0) &&
5541 vol
->metadata_version
&&
5542 is_container_member(vol
, memb
->dev
)) {
5547 if (*devlist
&& (found
< dpa
)) {
5548 dv
= calloc(1, sizeof(*dv
));
5550 fprintf(stderr
, Name
": calloc failed\n");
5552 dv
->devname
= malloc(strlen(memb
->dev
) + strlen("/dev/") + 1);
5553 if (dv
->devname
!= NULL
) {
5554 sprintf(dv
->devname
, "%s%s", "/dev/", memb
->dev
);
5557 dv
->next
= *devlist
;
5568 free_mdstat(mdstat
);
5573 static struct md_list
*
5574 get_loop_devices(void)
5577 struct md_list
*devlist
= NULL
;
5578 struct md_list
*dv
= NULL
;
5580 for(i
= 0; i
< 12; i
++) {
5581 dv
= calloc(1, sizeof(*dv
));
5583 fprintf(stderr
, Name
": calloc failed\n");
5586 dv
->devname
= malloc(40);
5587 if (dv
->devname
== NULL
) {
5588 fprintf(stderr
, Name
": malloc failed\n");
5592 sprintf(dv
->devname
, "/dev/loop%d", i
);
5600 static struct md_list
*
5601 get_devices(const char *hba_path
)
5603 struct md_list
*devlist
= NULL
;
5604 struct md_list
*dv
= NULL
;
5610 devlist
= get_loop_devices();
5613 /* scroll through /sys/dev/block looking for devices attached to
5616 dir
= opendir("/sys/dev/block");
5617 for (ent
= dir
? readdir(dir
) : NULL
; ent
; ent
= readdir(dir
)) {
5622 if (sscanf(ent
->d_name
, "%d:%d", &major
, &minor
) != 2)
5624 path
= devt_to_devpath(makedev(major
, minor
));
5627 if (!path_attached_to_hba(path
, hba_path
)) {
5634 fd
= dev_open(ent
->d_name
, O_RDONLY
);
5636 fd2devname(fd
, buf
);
5639 fprintf(stderr
, Name
": cannot open device: %s\n",
5645 dv
= calloc(1, sizeof(*dv
));
5647 fprintf(stderr
, Name
": malloc failed\n");
5651 dv
->devname
= strdup(buf
);
5652 if (dv
->devname
== NULL
) {
5653 fprintf(stderr
, Name
": malloc failed\n");
5664 devlist
= devlist
->next
;
5673 count_volumes_list(struct md_list
*devlist
, char *homehost
,
5674 int verbose
, int *found
)
5676 struct md_list
*tmpdev
;
5678 struct supertype
*st
= NULL
;
5680 /* first walk the list of devices to find a consistent set
5681 * that match the criterea, if that is possible.
5682 * We flag the ones we like with 'used'.
5685 st
= match_metadata_desc_imsm("imsm");
5687 pr_vrb(": cannot allocate memory for imsm supertype\n");
5691 for (tmpdev
= devlist
; tmpdev
; tmpdev
= tmpdev
->next
) {
5692 char *devname
= tmpdev
->devname
;
5694 struct supertype
*tst
;
5696 if (tmpdev
->used
> 1)
5698 tst
= dup_super(st
);
5700 pr_vrb(": cannot allocate memory for imsm supertype\n");
5703 tmpdev
->container
= 0;
5704 dfd
= dev_open(devname
, O_RDONLY
|O_EXCL
);
5706 dprintf(": cannot open device %s: %s\n",
5707 devname
, strerror(errno
));
5709 } else if (fstat(dfd
, &stb
)< 0) {
5711 dprintf(": fstat failed for %s: %s\n",
5712 devname
, strerror(errno
));
5714 } else if ((stb
.st_mode
& S_IFMT
) != S_IFBLK
) {
5715 dprintf(": %s is not a block device.\n",
5718 } else if (must_be_container(dfd
)) {
5719 struct supertype
*cst
;
5720 cst
= super_by_fd(dfd
, NULL
);
5722 dprintf(": cannot recognize container type %s\n",
5725 } else if (tst
->ss
!= st
->ss
) {
5726 dprintf(": non-imsm container - ignore it: %s\n",
5729 } else if (!tst
->ss
->load_container
||
5730 tst
->ss
->load_container(tst
, dfd
, NULL
))
5733 tmpdev
->container
= 1;
5736 cst
->ss
->free_super(cst
);
5738 tmpdev
->st_rdev
= stb
.st_rdev
;
5739 if (tst
->ss
->load_super(tst
,dfd
, NULL
)) {
5740 dprintf(": no RAID superblock on %s\n",
5743 } else if (tst
->ss
->compare_super
== NULL
) {
5744 dprintf(": Cannot assemble %s metadata on %s\n",
5745 tst
->ss
->name
, devname
);
5751 if (tmpdev
->used
== 2 || tmpdev
->used
== 4) {
5752 /* Ignore unrecognised devices during auto-assembly */
5757 tst
->ss
->getinfo_super(tst
, &info
, NULL
);
5759 if (st
->minor_version
== -1)
5760 st
->minor_version
= tst
->minor_version
;
5762 if (memcmp(info
.uuid
, uuid_zero
,
5763 sizeof(int[4])) == 0) {
5764 /* this is a floating spare. It cannot define
5765 * an array unless there are no more arrays of
5766 * this type to be found. It can be included
5767 * in an array of this type though.
5773 if (st
->ss
!= tst
->ss
||
5774 st
->minor_version
!= tst
->minor_version
||
5775 st
->ss
->compare_super(st
, tst
) != 0) {
5776 /* Some mismatch. If exactly one array matches this host,
5777 * we can resolve on that one.
5778 * Or, if we are auto assembling, we just ignore the second
5781 dprintf(": superblock on %s doesn't match others - assembly aborted\n",
5787 dprintf("found: devname: %s\n", devname
);
5791 tst
->ss
->free_super(tst
);
5795 if ((err
= load_super_imsm_all(st
, -1, &st
->sb
, NULL
, devlist
, 0)) == 0) {
5796 struct mdinfo
*iter
, *head
= st
->ss
->container_content(st
, NULL
);
5797 for (iter
= head
; iter
; iter
= iter
->next
) {
5798 dprintf("content->text_version: %s vol\n",
5799 iter
->text_version
);
5800 if (iter
->array
.state
& (1<<MD_SB_BLOCK_VOLUME
)) {
5801 /* do not assemble arrays with unsupported
5803 dprintf(": Cannot activate member %s.\n",
5804 iter
->text_version
);
5811 dprintf(" no valid super block on device list: err: %d %p\n",
5815 dprintf(" no more devices to examin\n");
5818 for (tmpdev
= devlist
; tmpdev
; tmpdev
= tmpdev
->next
) {
5819 if ((tmpdev
->used
== 1) && (tmpdev
->found
)) {
5821 if (count
< tmpdev
->found
)
5824 count
-= tmpdev
->found
;
5827 if (tmpdev
->used
== 1)
5832 st
->ss
->free_super(st
);
5838 count_volumes(char *hba
, int dpa
, int verbose
)
5840 struct md_list
*devlist
= NULL
;
5844 devlist
= get_devices(hba
);
5845 /* if no intel devices return zero volumes */
5846 if (devlist
== NULL
)
5849 count
= active_arrays_by_format("imsm", hba
, &devlist
, dpa
, verbose
);
5850 dprintf(" path: %s active arrays: %d\n", hba
, count
);
5851 if (devlist
== NULL
)
5855 count
+= count_volumes_list(devlist
,
5859 dprintf("found %d count: %d\n", found
, count
);
5862 dprintf("path: %s total number of volumes: %d\n", hba
, count
);
5865 struct md_list
*dv
= devlist
;
5866 devlist
= devlist
->next
;
5873 static int imsm_default_chunk(const struct imsm_orom
*orom
)
5875 /* up to 512 if the plaform supports it, otherwise the platform max.
5876 * 128 if no platform detected
5878 int fs
= max(7, orom
? fls(orom
->sss
) : 0);
5880 return min(512, (1 << fs
));
5884 validate_geometry_imsm_orom(struct intel_super
*super
, int level
, int layout
,
5885 int raiddisks
, int *chunk
, unsigned long long size
, int verbose
)
5887 /* check/set platform and metadata limits/defaults */
5888 if (super
->orom
&& raiddisks
> super
->orom
->dpa
) {
5889 pr_vrb(": platform supports a maximum of %d disks per array\n",
5894 /* capabilities of OROM tested - copied from validate_geometry_imsm_volume */
5895 if (!is_raid_level_supported(super
->orom
, level
, raiddisks
)) {
5896 pr_vrb(": platform does not support raid%d with %d disk%s\n",
5897 level
, raiddisks
, raiddisks
> 1 ? "s" : "");
5901 if (chunk
&& (*chunk
== 0 || *chunk
== UnSet
))
5902 *chunk
= imsm_default_chunk(super
->orom
);
5904 if (super
->orom
&& chunk
&& !imsm_orom_has_chunk(super
->orom
, *chunk
)) {
5905 pr_vrb(": platform does not support a chunk size of: "
5910 if (layout
!= imsm_level_to_layout(level
)) {
5912 pr_vrb(": imsm raid 5 only supports the left-asymmetric layout\n");
5913 else if (level
== 10)
5914 pr_vrb(": imsm raid 10 only supports the n2 layout\n");
5916 pr_vrb(": imsm unknown layout %#x for this raid level %d\n",
5921 if (super
->orom
&& (super
->orom
->attr
& IMSM_OROM_ATTR_2TB
) == 0 && chunk
&&
5922 (calc_array_size(level
, raiddisks
, layout
, *chunk
, size
) >> 32) > 0) {
5923 pr_vrb(": platform does not support a volume size over 2TB\n");
5929 /* validate_geometry_imsm_volume - lifted from validate_geometry_ddf_bvd
5930 * FIX ME add ahci details
5932 static int validate_geometry_imsm_volume(struct supertype
*st
, int level
,
5933 int layout
, int raiddisks
, int *chunk
,
5934 unsigned long long size
, char *dev
,
5935 unsigned long long *freesize
,
5939 struct intel_super
*super
= st
->sb
;
5940 struct imsm_super
*mpb
;
5942 unsigned long long pos
= 0;
5943 unsigned long long maxsize
;
5947 /* We must have the container info already read in. */
5951 mpb
= super
->anchor
;
5953 if (!validate_geometry_imsm_orom(super
, level
, layout
, raiddisks
, chunk
, size
, verbose
)) {
5954 fprintf(stderr
, Name
": RAID gemetry validation failed. "
5955 "Cannot proceed with the action(s).\n");
5959 /* General test: make sure there is space for
5960 * 'raiddisks' device extents of size 'size' at a given
5963 unsigned long long minsize
= size
;
5964 unsigned long long start_offset
= MaxSector
;
5967 minsize
= MPB_SECTOR_CNT
+ IMSM_RESERVED_SECTORS
;
5968 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
5973 e
= get_extents(super
, dl
);
5976 unsigned long long esize
;
5977 esize
= e
[i
].start
- pos
;
5978 if (esize
>= minsize
)
5980 if (found
&& start_offset
== MaxSector
) {
5983 } else if (found
&& pos
!= start_offset
) {
5987 pos
= e
[i
].start
+ e
[i
].size
;
5989 } while (e
[i
-1].size
);
5994 if (dcnt
< raiddisks
) {
5996 fprintf(stderr
, Name
": imsm: Not enough "
5997 "devices with space for this array "
6005 /* This device must be a member of the set */
6006 if (stat(dev
, &stb
) < 0)
6008 if ((S_IFMT
& stb
.st_mode
) != S_IFBLK
)
6010 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
6011 if (dl
->major
== (int)major(stb
.st_rdev
) &&
6012 dl
->minor
== (int)minor(stb
.st_rdev
))
6017 fprintf(stderr
, Name
": %s is not in the "
6018 "same imsm set\n", dev
);
6020 } else if (super
->orom
&& dl
->index
< 0 && mpb
->num_raid_devs
) {
6021 /* If a volume is present then the current creation attempt
6022 * cannot incorporate new spares because the orom may not
6023 * understand this configuration (all member disks must be
6024 * members of each array in the container).
6026 fprintf(stderr
, Name
": %s is a spare and a volume"
6027 " is already defined for this container\n", dev
);
6028 fprintf(stderr
, Name
": The option-rom requires all member"
6029 " disks to be a member of all volumes\n");
6031 } else if (super
->orom
&& mpb
->num_raid_devs
> 0 &&
6032 mpb
->num_disks
!= raiddisks
) {
6033 fprintf(stderr
, Name
": The option-rom requires all member"
6034 " disks to be a member of all volumes\n");
6038 /* retrieve the largest free space block */
6039 e
= get_extents(super
, dl
);
6044 unsigned long long esize
;
6046 esize
= e
[i
].start
- pos
;
6047 if (esize
>= maxsize
)
6049 pos
= e
[i
].start
+ e
[i
].size
;
6051 } while (e
[i
-1].size
);
6056 fprintf(stderr
, Name
": unable to determine free space for: %s\n",
6060 if (maxsize
< size
) {
6062 fprintf(stderr
, Name
": %s not enough space (%llu < %llu)\n",
6063 dev
, maxsize
, size
);
6067 /* count total number of extents for merge */
6069 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
6071 i
+= dl
->extent_cnt
;
6073 maxsize
= merge_extents(super
, i
);
6075 if (!check_env("IMSM_NO_PLATFORM") &&
6076 mpb
->num_raid_devs
> 0 && size
&& size
!= maxsize
) {
6077 fprintf(stderr
, Name
": attempting to create a second "
6078 "volume with size less then remaining space. "
6083 if (maxsize
< size
|| maxsize
== 0) {
6086 fprintf(stderr
, Name
": no free space"
6087 " left on device. Aborting...\n");
6089 fprintf(stderr
, Name
": not enough space"
6090 " to create volume of given size"
6091 " (%llu < %llu). Aborting...\n",
6097 *freesize
= maxsize
;
6100 int count
= count_volumes(super
->hba
->path
,
6101 super
->orom
->dpa
, verbose
);
6102 if (super
->orom
->vphba
<= count
) {
6103 pr_vrb(": platform does not support more than %d raid volumes.\n",
6104 super
->orom
->vphba
);
6111 static int imsm_get_free_size(struct supertype
*st
, int raiddisks
,
6112 unsigned long long size
, int chunk
,
6113 unsigned long long *freesize
)
6115 struct intel_super
*super
= st
->sb
;
6116 struct imsm_super
*mpb
= super
->anchor
;
6121 unsigned long long maxsize
;
6122 unsigned long long minsize
;
6126 /* find the largest common start free region of the possible disks */
6130 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
6136 /* don't activate new spares if we are orom constrained
6137 * and there is already a volume active in the container
6139 if (super
->orom
&& dl
->index
< 0 && mpb
->num_raid_devs
)
6142 e
= get_extents(super
, dl
);
6145 for (i
= 1; e
[i
-1].size
; i
++)
6153 maxsize
= merge_extents(super
, extent_cnt
);
6157 minsize
= chunk
* 2;
6159 if (cnt
< raiddisks
||
6160 (super
->orom
&& used
&& used
!= raiddisks
) ||
6161 maxsize
< minsize
||
6163 fprintf(stderr
, Name
": not enough devices with space to create array.\n");
6164 return 0; /* No enough free spaces large enough */
6175 if (!check_env("IMSM_NO_PLATFORM") &&
6176 mpb
->num_raid_devs
> 0 && size
&& size
!= maxsize
) {
6177 fprintf(stderr
, Name
": attempting to create a second "
6178 "volume with size less then remaining space. "
6183 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
6185 dl
->raiddisk
= cnt
++;
6189 dprintf("imsm: imsm_get_free_size() returns : %llu\n", size
);
6194 static int reserve_space(struct supertype
*st
, int raiddisks
,
6195 unsigned long long size
, int chunk
,
6196 unsigned long long *freesize
)
6198 struct intel_super
*super
= st
->sb
;
6203 rv
= imsm_get_free_size(st
, raiddisks
, size
, chunk
, freesize
);
6206 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
6208 dl
->raiddisk
= cnt
++;
6215 static int validate_geometry_imsm(struct supertype
*st
, int level
, int layout
,
6216 int raiddisks
, int *chunk
, unsigned long long size
,
6217 char *dev
, unsigned long long *freesize
,
6225 * if given unused devices create a container
6226 * if given given devices in a container create a member volume
6228 if (level
== LEVEL_CONTAINER
) {
6229 /* Must be a fresh device to add to a container */
6230 return validate_geometry_imsm_container(st
, level
, layout
,
6232 chunk
?*chunk
:0, size
,
6239 struct intel_super
*super
= st
->sb
;
6240 if (!validate_geometry_imsm_orom(st
->sb
, level
, layout
,
6241 raiddisks
, chunk
, size
,
6244 /* we are being asked to automatically layout a
6245 * new volume based on the current contents of
6246 * the container. If the the parameters can be
6247 * satisfied reserve_space will record the disks,
6248 * start offset, and size of the volume to be
6249 * created. add_to_super and getinfo_super
6250 * detect when autolayout is in progress.
6252 /* assuming that freesize is always given when array is
6254 if (super
->orom
&& freesize
) {
6256 count
= count_volumes(super
->hba
->path
,
6257 super
->orom
->dpa
, verbose
);
6258 if (super
->orom
->vphba
<= count
) {
6259 pr_vrb(": platform does not support more"
6260 " than %d raid volumes.\n",
6261 super
->orom
->vphba
);
6266 return reserve_space(st
, raiddisks
, size
,
6267 chunk
?*chunk
:0, freesize
);
6272 /* creating in a given container */
6273 return validate_geometry_imsm_volume(st
, level
, layout
,
6274 raiddisks
, chunk
, size
,
6275 dev
, freesize
, verbose
);
6278 /* This device needs to be a device in an 'imsm' container */
6279 fd
= open(dev
, O_RDONLY
|O_EXCL
, 0);
6283 Name
": Cannot create this array on device %s\n",
6288 if (errno
!= EBUSY
|| (fd
= open(dev
, O_RDONLY
, 0)) < 0) {
6290 fprintf(stderr
, Name
": Cannot open %s: %s\n",
6291 dev
, strerror(errno
));
6294 /* Well, it is in use by someone, maybe an 'imsm' container. */
6295 cfd
= open_container(fd
);
6299 fprintf(stderr
, Name
": Cannot use %s: It is busy\n",
6303 sra
= sysfs_read(cfd
, 0, GET_VERSION
);
6304 if (sra
&& sra
->array
.major_version
== -1 &&
6305 strcmp(sra
->text_version
, "imsm") == 0)
6309 /* This is a member of a imsm container. Load the container
6310 * and try to create a volume
6312 struct intel_super
*super
;
6314 if (load_super_imsm_all(st
, cfd
, (void **) &super
, NULL
, NULL
, 1) == 0) {
6316 st
->container_dev
= fd2devnum(cfd
);
6318 return validate_geometry_imsm_volume(st
, level
, layout
,
6327 fprintf(stderr
, Name
": failed container membership check\n");
6333 static void default_geometry_imsm(struct supertype
*st
, int *level
, int *layout
, int *chunk
)
6335 struct intel_super
*super
= st
->sb
;
6337 if (level
&& *level
== UnSet
)
6338 *level
= LEVEL_CONTAINER
;
6340 if (level
&& layout
&& *layout
== UnSet
)
6341 *layout
= imsm_level_to_layout(*level
);
6343 if (chunk
&& (*chunk
== UnSet
|| *chunk
== 0))
6344 *chunk
= imsm_default_chunk(super
->orom
);
6347 static void handle_missing(struct intel_super
*super
, struct imsm_dev
*dev
);
6349 static int kill_subarray_imsm(struct supertype
*st
)
6351 /* remove the subarray currently referenced by ->current_vol */
6353 struct intel_dev
**dp
;
6354 struct intel_super
*super
= st
->sb
;
6355 __u8 current_vol
= super
->current_vol
;
6356 struct imsm_super
*mpb
= super
->anchor
;
6358 if (super
->current_vol
< 0)
6360 super
->current_vol
= -1; /* invalidate subarray cursor */
6362 /* block deletions that would change the uuid of active subarrays
6364 * FIXME when immutable ids are available, but note that we'll
6365 * also need to fixup the invalidated/active subarray indexes in
6368 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
6371 if (i
< current_vol
)
6373 sprintf(subarray
, "%u", i
);
6374 if (is_subarray_active(subarray
, st
->devname
)) {
6376 Name
": deleting subarray-%d would change the UUID of active subarray-%d, aborting\n",
6383 if (st
->update_tail
) {
6384 struct imsm_update_kill_array
*u
= malloc(sizeof(*u
));
6388 u
->type
= update_kill_array
;
6389 u
->dev_idx
= current_vol
;
6390 append_metadata_update(st
, u
, sizeof(*u
));
6395 for (dp
= &super
->devlist
; *dp
;)
6396 if ((*dp
)->index
== current_vol
) {
6399 handle_missing(super
, (*dp
)->dev
);
6400 if ((*dp
)->index
> current_vol
)
6405 /* no more raid devices, all active components are now spares,
6406 * but of course failed are still failed
6408 if (--mpb
->num_raid_devs
== 0) {
6411 for (d
= super
->disks
; d
; d
= d
->next
)
6416 super
->updates_pending
++;
6421 static int update_subarray_imsm(struct supertype
*st
, char *subarray
,
6422 char *update
, struct mddev_ident
*ident
)
6424 /* update the subarray currently referenced by ->current_vol */
6425 struct intel_super
*super
= st
->sb
;
6426 struct imsm_super
*mpb
= super
->anchor
;
6428 if (strcmp(update
, "name") == 0) {
6429 char *name
= ident
->name
;
6433 if (is_subarray_active(subarray
, st
->devname
)) {
6435 Name
": Unable to update name of active subarray\n");
6439 if (!check_name(super
, name
, 0))
6442 vol
= strtoul(subarray
, &ep
, 10);
6443 if (*ep
!= '\0' || vol
>= super
->anchor
->num_raid_devs
)
6446 if (st
->update_tail
) {
6447 struct imsm_update_rename_array
*u
= malloc(sizeof(*u
));
6451 u
->type
= update_rename_array
;
6453 snprintf((char *) u
->name
, MAX_RAID_SERIAL_LEN
, "%s", name
);
6454 append_metadata_update(st
, u
, sizeof(*u
));
6456 struct imsm_dev
*dev
;
6459 dev
= get_imsm_dev(super
, vol
);
6460 snprintf((char *) dev
->volume
, MAX_RAID_SERIAL_LEN
, "%s", name
);
6461 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
6462 dev
= get_imsm_dev(super
, i
);
6463 handle_missing(super
, dev
);
6465 super
->updates_pending
++;
6472 #endif /* MDASSEMBLE */
6474 static int is_gen_migration(struct imsm_dev
*dev
)
6479 if (!dev
->vol
.migr_state
)
6482 if (migr_type(dev
) == MIGR_GEN_MIGR
)
6488 static int is_rebuilding(struct imsm_dev
*dev
)
6490 struct imsm_map
*migr_map
;
6492 if (!dev
->vol
.migr_state
)
6495 if (migr_type(dev
) != MIGR_REBUILD
)
6498 migr_map
= get_imsm_map(dev
, MAP_1
);
6500 if (migr_map
->map_state
== IMSM_T_STATE_DEGRADED
)
6507 static int is_initializing(struct imsm_dev
*dev
)
6509 struct imsm_map
*migr_map
;
6511 if (!dev
->vol
.migr_state
)
6514 if (migr_type(dev
) != MIGR_INIT
)
6517 migr_map
= get_imsm_map(dev
, MAP_1
);
6519 if (migr_map
->map_state
== IMSM_T_STATE_UNINITIALIZED
)
6526 static void update_recovery_start(struct intel_super
*super
,
6527 struct imsm_dev
*dev
,
6528 struct mdinfo
*array
)
6530 struct mdinfo
*rebuild
= NULL
;
6534 if (!is_rebuilding(dev
))
6537 /* Find the rebuild target, but punt on the dual rebuild case */
6538 for (d
= array
->devs
; d
; d
= d
->next
)
6539 if (d
->recovery_start
== 0) {
6546 /* (?) none of the disks are marked with
6547 * IMSM_ORD_REBUILD, so assume they are missing and the
6548 * disk_ord_tbl was not correctly updated
6550 dprintf("%s: failed to locate out-of-sync disk\n", __func__
);
6554 units
= __le32_to_cpu(dev
->vol
.curr_migr_unit
);
6555 rebuild
->recovery_start
= units
* blocks_per_migr_unit(super
, dev
);
6559 static int recover_backup_imsm(struct supertype
*st
, struct mdinfo
*info
);
6562 static struct mdinfo
*container_content_imsm(struct supertype
*st
, char *subarray
)
6564 /* Given a container loaded by load_super_imsm_all,
6565 * extract information about all the arrays into
6567 * If 'subarray' is given, just extract info about that array.
6569 * For each imsm_dev create an mdinfo, fill it in,
6570 * then look for matching devices in super->disks
6571 * and create appropriate device mdinfo.
6573 struct intel_super
*super
= st
->sb
;
6574 struct imsm_super
*mpb
= super
->anchor
;
6575 struct mdinfo
*rest
= NULL
;
6579 int spare_disks
= 0;
6581 /* do not assemble arrays when not all attributes are supported */
6582 if (imsm_check_attributes(mpb
->attributes
) == 0) {
6584 fprintf(stderr
, Name
": Unsupported attributes in IMSM metadata."
6585 "Arrays activation is blocked.\n");
6588 /* check for bad blocks */
6589 if (imsm_bbm_log_size(super
->anchor
)) {
6590 fprintf(stderr
, Name
": BBM log found in IMSM metadata."
6591 "Arrays activation is blocked.\n");
6596 /* count spare devices, not used in maps
6598 for (d
= super
->disks
; d
; d
= d
->next
)
6602 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
6603 struct imsm_dev
*dev
;
6604 struct imsm_map
*map
;
6605 struct imsm_map
*map2
;
6606 struct mdinfo
*this;
6614 (i
!= strtoul(subarray
, &ep
, 10) || *ep
!= '\0'))
6617 dev
= get_imsm_dev(super
, i
);
6618 map
= get_imsm_map(dev
, MAP_0
);
6619 map2
= get_imsm_map(dev
, MAP_1
);
6621 /* do not publish arrays that are in the middle of an
6622 * unsupported migration
6624 if (dev
->vol
.migr_state
&&
6625 (migr_type(dev
) == MIGR_STATE_CHANGE
)) {
6626 fprintf(stderr
, Name
": cannot assemble volume '%.16s':"
6627 " unsupported migration in progress\n",
6631 /* do not publish arrays that are not support by controller's
6635 this = malloc(sizeof(*this));
6637 fprintf(stderr
, Name
": failed to allocate %zu bytes\n",
6642 super
->current_vol
= i
;
6643 getinfo_super_imsm_volume(st
, this, NULL
);
6646 chunk
= __le16_to_cpu(map
->blocks_per_strip
) >> 1;
6647 /* mdadm does not support all metadata features- set the bit in all arrays state */
6648 if (!validate_geometry_imsm_orom(super
,
6649 get_imsm_raid_level(map
), /* RAID level */
6650 imsm_level_to_layout(get_imsm_raid_level(map
)),
6651 map
->num_members
, /* raid disks */
6652 &chunk
, join_u32(dev
->size_low
, dev
->size_high
),
6654 fprintf(stderr
, Name
": IMSM RAID geometry validation"
6655 " failed. Array %s activation is blocked.\n",
6657 this->array
.state
|=
6658 (1<<MD_SB_BLOCK_CONTAINER_RESHAPE
) |
6659 (1<<MD_SB_BLOCK_VOLUME
);
6663 /* if array has bad blocks, set suitable bit in all arrays state */
6665 this->array
.state
|=
6666 (1<<MD_SB_BLOCK_CONTAINER_RESHAPE
) |
6667 (1<<MD_SB_BLOCK_VOLUME
);
6669 for (slot
= 0 ; slot
< map
->num_members
; slot
++) {
6670 unsigned long long recovery_start
;
6671 struct mdinfo
*info_d
;
6678 idx
= get_imsm_disk_idx(dev
, slot
, MAP_0
);
6679 ord
= get_imsm_ord_tbl_ent(dev
, slot
, MAP_X
);
6680 for (d
= super
->disks
; d
; d
= d
->next
)
6681 if (d
->index
== idx
)
6684 recovery_start
= MaxSector
;
6687 if (d
&& is_failed(&d
->disk
))
6689 if (ord
& IMSM_ORD_REBUILD
)
6693 * if we skip some disks the array will be assmebled degraded;
6694 * reset resync start to avoid a dirty-degraded
6695 * situation when performing the intial sync
6697 * FIXME handle dirty degraded
6699 if ((skip
|| recovery_start
== 0) && !dev
->vol
.dirty
)
6700 this->resync_start
= MaxSector
;
6704 info_d
= calloc(1, sizeof(*info_d
));
6706 fprintf(stderr
, Name
": failed to allocate disk"
6707 " for volume %.16s\n", dev
->volume
);
6708 info_d
= this->devs
;
6710 struct mdinfo
*d
= info_d
->next
;
6719 info_d
->next
= this->devs
;
6720 this->devs
= info_d
;
6722 info_d
->disk
.number
= d
->index
;
6723 info_d
->disk
.major
= d
->major
;
6724 info_d
->disk
.minor
= d
->minor
;
6725 info_d
->disk
.raid_disk
= slot
;
6726 info_d
->recovery_start
= recovery_start
;
6728 if (slot
< map2
->num_members
)
6729 info_d
->disk
.state
= (1 << MD_DISK_ACTIVE
);
6731 this->array
.spare_disks
++;
6733 if (slot
< map
->num_members
)
6734 info_d
->disk
.state
= (1 << MD_DISK_ACTIVE
);
6736 this->array
.spare_disks
++;
6738 if (info_d
->recovery_start
== MaxSector
)
6739 this->array
.working_disks
++;
6741 info_d
->events
= __le32_to_cpu(mpb
->generation_num
);
6742 info_d
->data_offset
= pba_of_lba0(map
);
6743 info_d
->component_size
= blocks_per_member(map
);
6745 /* now that the disk list is up-to-date fixup recovery_start */
6746 update_recovery_start(super
, dev
, this);
6747 this->array
.spare_disks
+= spare_disks
;
6750 /* check for reshape */
6751 if (this->reshape_active
== 1)
6752 recover_backup_imsm(st
, this);
6761 static __u8
imsm_check_degraded(struct intel_super
*super
, struct imsm_dev
*dev
,
6762 int failed
, int look_in_map
)
6764 struct imsm_map
*map
;
6766 map
= get_imsm_map(dev
, look_in_map
);
6769 return map
->map_state
== IMSM_T_STATE_UNINITIALIZED
?
6770 IMSM_T_STATE_UNINITIALIZED
: IMSM_T_STATE_NORMAL
;
6772 switch (get_imsm_raid_level(map
)) {
6774 return IMSM_T_STATE_FAILED
;
6777 if (failed
< map
->num_members
)
6778 return IMSM_T_STATE_DEGRADED
;
6780 return IMSM_T_STATE_FAILED
;
6785 * check to see if any mirrors have failed, otherwise we
6786 * are degraded. Even numbered slots are mirrored on
6790 /* gcc -Os complains that this is unused */
6791 int insync
= insync
;
6793 for (i
= 0; i
< map
->num_members
; i
++) {
6794 __u32 ord
= get_imsm_ord_tbl_ent(dev
, i
, MAP_X
);
6795 int idx
= ord_to_idx(ord
);
6796 struct imsm_disk
*disk
;
6798 /* reset the potential in-sync count on even-numbered
6799 * slots. num_copies is always 2 for imsm raid10
6804 disk
= get_imsm_disk(super
, idx
);
6805 if (!disk
|| is_failed(disk
) || ord
& IMSM_ORD_REBUILD
)
6808 /* no in-sync disks left in this mirror the
6812 return IMSM_T_STATE_FAILED
;
6815 return IMSM_T_STATE_DEGRADED
;
6819 return IMSM_T_STATE_DEGRADED
;
6821 return IMSM_T_STATE_FAILED
;
6827 return map
->map_state
;
6830 static int imsm_count_failed(struct intel_super
*super
, struct imsm_dev
*dev
,
6835 struct imsm_disk
*disk
;
6836 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
6837 struct imsm_map
*prev
= get_imsm_map(dev
, MAP_1
);
6838 struct imsm_map
*map_for_loop
;
6843 /* at the beginning of migration we set IMSM_ORD_REBUILD on
6844 * disks that are being rebuilt. New failures are recorded to
6845 * map[0]. So we look through all the disks we started with and
6846 * see if any failures are still present, or if any new ones
6850 if (prev
&& (map
->num_members
< prev
->num_members
))
6851 map_for_loop
= prev
;
6853 for (i
= 0; i
< map_for_loop
->num_members
; i
++) {
6855 /* when MAP_X is passed both maps failures are counted
6858 ((look_in_map
== MAP_1
) || (look_in_map
== MAP_X
)) &&
6859 (i
< prev
->num_members
)) {
6860 ord
= __le32_to_cpu(prev
->disk_ord_tbl
[i
]);
6861 idx_1
= ord_to_idx(ord
);
6863 disk
= get_imsm_disk(super
, idx_1
);
6864 if (!disk
|| is_failed(disk
) || ord
& IMSM_ORD_REBUILD
)
6867 if (((look_in_map
== MAP_0
) || (look_in_map
== MAP_X
)) &&
6868 (i
< map
->num_members
)) {
6869 ord
= __le32_to_cpu(map
->disk_ord_tbl
[i
]);
6870 idx
= ord_to_idx(ord
);
6873 disk
= get_imsm_disk(super
, idx
);
6874 if (!disk
|| is_failed(disk
) ||
6875 ord
& IMSM_ORD_REBUILD
)
6885 static int imsm_open_new(struct supertype
*c
, struct active_array
*a
,
6888 struct intel_super
*super
= c
->sb
;
6889 struct imsm_super
*mpb
= super
->anchor
;
6891 if (atoi(inst
) >= mpb
->num_raid_devs
) {
6892 fprintf(stderr
, "%s: subarry index %d, out of range\n",
6893 __func__
, atoi(inst
));
6897 dprintf("imsm: open_new %s\n", inst
);
6898 a
->info
.container_member
= atoi(inst
);
6902 static int is_resyncing(struct imsm_dev
*dev
)
6904 struct imsm_map
*migr_map
;
6906 if (!dev
->vol
.migr_state
)
6909 if (migr_type(dev
) == MIGR_INIT
||
6910 migr_type(dev
) == MIGR_REPAIR
)
6913 if (migr_type(dev
) == MIGR_GEN_MIGR
)
6916 migr_map
= get_imsm_map(dev
, MAP_1
);
6918 if ((migr_map
->map_state
== IMSM_T_STATE_NORMAL
) &&
6919 (dev
->vol
.migr_type
!= MIGR_GEN_MIGR
))
6925 /* return true if we recorded new information */
6926 static int mark_failure(struct imsm_dev
*dev
, struct imsm_disk
*disk
, int idx
)
6930 struct imsm_map
*map
;
6931 char buf
[MAX_RAID_SERIAL_LEN
+3];
6932 unsigned int len
, shift
= 0;
6934 /* new failures are always set in map[0] */
6935 map
= get_imsm_map(dev
, MAP_0
);
6937 slot
= get_imsm_disk_slot(map
, idx
);
6941 ord
= __le32_to_cpu(map
->disk_ord_tbl
[slot
]);
6942 if (is_failed(disk
) && (ord
& IMSM_ORD_REBUILD
))
6945 memcpy(buf
, disk
->serial
, MAX_RAID_SERIAL_LEN
);
6946 buf
[MAX_RAID_SERIAL_LEN
] = '\000';
6948 if ((len
= strlen(buf
)) >= MAX_RAID_SERIAL_LEN
)
6949 shift
= len
- MAX_RAID_SERIAL_LEN
+ 1;
6950 strncpy((char *)disk
->serial
, &buf
[shift
], MAX_RAID_SERIAL_LEN
);
6952 disk
->status
|= FAILED_DISK
;
6953 set_imsm_ord_tbl_ent(map
, slot
, idx
| IMSM_ORD_REBUILD
);
6954 /* mark failures in second map if second map exists and this disk
6956 * This is valid for migration, initialization and rebuild
6958 if (dev
->vol
.migr_state
) {
6959 struct imsm_map
*map2
= get_imsm_map(dev
, MAP_1
);
6960 int slot2
= get_imsm_disk_slot(map2
, idx
);
6962 if ((slot2
< map2
->num_members
) &&
6964 set_imsm_ord_tbl_ent(map2
, slot2
,
6965 idx
| IMSM_ORD_REBUILD
);
6967 if (map
->failed_disk_num
== 0xff)
6968 map
->failed_disk_num
= slot
;
6972 static void mark_missing(struct imsm_dev
*dev
, struct imsm_disk
*disk
, int idx
)
6974 mark_failure(dev
, disk
, idx
);
6976 if (disk
->scsi_id
== __cpu_to_le32(~(__u32
)0))
6979 disk
->scsi_id
= __cpu_to_le32(~(__u32
)0);
6980 memmove(&disk
->serial
[0], &disk
->serial
[1], MAX_RAID_SERIAL_LEN
- 1);
6983 static void handle_missing(struct intel_super
*super
, struct imsm_dev
*dev
)
6987 if (!super
->missing
)
6990 dprintf("imsm: mark missing\n");
6991 /* end process for initialization and rebuild only
6993 if (is_gen_migration(dev
) == 0) {
6997 failed
= imsm_count_failed(super
, dev
, MAP_0
);
6998 map_state
= imsm_check_degraded(super
, dev
, failed
, MAP_0
);
7000 end_migration(dev
, super
, map_state
);
7002 for (dl
= super
->missing
; dl
; dl
= dl
->next
)
7003 mark_missing(dev
, &dl
->disk
, dl
->index
);
7004 super
->updates_pending
++;
7007 static unsigned long long imsm_set_array_size(struct imsm_dev
*dev
,
7010 int used_disks
= imsm_num_data_members(dev
, MAP_0
);
7011 unsigned long long array_blocks
;
7012 struct imsm_map
*map
;
7014 if (used_disks
== 0) {
7015 /* when problems occures
7016 * return current array_blocks value
7018 array_blocks
= __le32_to_cpu(dev
->size_high
);
7019 array_blocks
= array_blocks
<< 32;
7020 array_blocks
+= __le32_to_cpu(dev
->size_low
);
7022 return array_blocks
;
7025 /* set array size in metadata
7027 if (new_size
<= 0) {
7028 /* OLCE size change is caused by added disks
7030 map
= get_imsm_map(dev
, MAP_0
);
7031 array_blocks
= blocks_per_member(map
) * used_disks
;
7033 /* Online Volume Size Change
7034 * Using available free space
7036 array_blocks
= new_size
;
7039 /* round array size down to closest MB
7041 array_blocks
= (array_blocks
>> SECT_PER_MB_SHIFT
) << SECT_PER_MB_SHIFT
;
7042 dev
->size_low
= __cpu_to_le32((__u32
)array_blocks
);
7043 dev
->size_high
= __cpu_to_le32((__u32
)(array_blocks
>> 32));
7045 return array_blocks
;
7048 static void imsm_set_disk(struct active_array
*a
, int n
, int state
);
7050 static void imsm_progress_container_reshape(struct intel_super
*super
)
7052 /* if no device has a migr_state, but some device has a
7053 * different number of members than the previous device, start
7054 * changing the number of devices in this device to match
7057 struct imsm_super
*mpb
= super
->anchor
;
7058 int prev_disks
= -1;
7062 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
7063 struct imsm_dev
*dev
= get_imsm_dev(super
, i
);
7064 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
7065 struct imsm_map
*map2
;
7066 int prev_num_members
;
7068 if (dev
->vol
.migr_state
)
7071 if (prev_disks
== -1)
7072 prev_disks
= map
->num_members
;
7073 if (prev_disks
== map
->num_members
)
7076 /* OK, this array needs to enter reshape mode.
7077 * i.e it needs a migr_state
7080 copy_map_size
= sizeof_imsm_map(map
);
7081 prev_num_members
= map
->num_members
;
7082 map
->num_members
= prev_disks
;
7083 dev
->vol
.migr_state
= 1;
7084 dev
->vol
.curr_migr_unit
= 0;
7085 set_migr_type(dev
, MIGR_GEN_MIGR
);
7086 for (i
= prev_num_members
;
7087 i
< map
->num_members
; i
++)
7088 set_imsm_ord_tbl_ent(map
, i
, i
);
7089 map2
= get_imsm_map(dev
, MAP_1
);
7090 /* Copy the current map */
7091 memcpy(map2
, map
, copy_map_size
);
7092 map2
->num_members
= prev_num_members
;
7094 imsm_set_array_size(dev
, -1);
7095 super
->clean_migration_record_by_mdmon
= 1;
7096 super
->updates_pending
++;
7100 /* Handle dirty -> clean transititions, resync and reshape. Degraded and rebuild
7101 * states are handled in imsm_set_disk() with one exception, when a
7102 * resync is stopped due to a new failure this routine will set the
7103 * 'degraded' state for the array.
7105 static int imsm_set_array_state(struct active_array
*a
, int consistent
)
7107 int inst
= a
->info
.container_member
;
7108 struct intel_super
*super
= a
->container
->sb
;
7109 struct imsm_dev
*dev
= get_imsm_dev(super
, inst
);
7110 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
7111 int failed
= imsm_count_failed(super
, dev
, MAP_0
);
7112 __u8 map_state
= imsm_check_degraded(super
, dev
, failed
, MAP_0
);
7113 __u32 blocks_per_unit
;
7115 if (dev
->vol
.migr_state
&&
7116 dev
->vol
.migr_type
== MIGR_GEN_MIGR
) {
7117 /* array state change is blocked due to reshape action
7119 * - abort the reshape (if last_checkpoint is 0 and action!= reshape)
7120 * - finish the reshape (if last_checkpoint is big and action != reshape)
7121 * - update curr_migr_unit
7123 if (a
->curr_action
== reshape
) {
7124 /* still reshaping, maybe update curr_migr_unit */
7125 goto mark_checkpoint
;
7127 if (a
->last_checkpoint
== 0 && a
->prev_action
== reshape
) {
7128 /* for some reason we aborted the reshape.
7130 * disable automatic metadata rollback
7131 * user action is required to recover process
7134 struct imsm_map
*map2
=
7135 get_imsm_map(dev
, MAP_1
);
7136 dev
->vol
.migr_state
= 0;
7137 set_migr_type(dev
, 0);
7138 dev
->vol
.curr_migr_unit
= 0;
7140 sizeof_imsm_map(map2
));
7141 super
->updates_pending
++;
7144 if (a
->last_checkpoint
>= a
->info
.component_size
) {
7145 unsigned long long array_blocks
;
7149 used_disks
= imsm_num_data_members(dev
, MAP_0
);
7150 if (used_disks
> 0) {
7152 blocks_per_member(map
) *
7154 /* round array size down to closest MB
7156 array_blocks
= (array_blocks
7157 >> SECT_PER_MB_SHIFT
)
7158 << SECT_PER_MB_SHIFT
;
7159 a
->info
.custom_array_size
= array_blocks
;
7160 /* encourage manager to update array
7164 a
->check_reshape
= 1;
7166 /* finalize online capacity expansion/reshape */
7167 for (mdi
= a
->info
.devs
; mdi
; mdi
= mdi
->next
)
7169 mdi
->disk
.raid_disk
,
7172 imsm_progress_container_reshape(super
);
7177 /* before we activate this array handle any missing disks */
7178 if (consistent
== 2)
7179 handle_missing(super
, dev
);
7181 if (consistent
== 2 &&
7182 (!is_resync_complete(&a
->info
) ||
7183 map_state
!= IMSM_T_STATE_NORMAL
||
7184 dev
->vol
.migr_state
))
7187 if (is_resync_complete(&a
->info
)) {
7188 /* complete intialization / resync,
7189 * recovery and interrupted recovery is completed in
7192 if (is_resyncing(dev
)) {
7193 dprintf("imsm: mark resync done\n");
7194 end_migration(dev
, super
, map_state
);
7195 super
->updates_pending
++;
7196 a
->last_checkpoint
= 0;
7198 } else if ((!is_resyncing(dev
) && !failed
) &&
7199 (imsm_reshape_blocks_arrays_changes(super
) == 0)) {
7200 /* mark the start of the init process if nothing is failed */
7201 dprintf("imsm: mark resync start\n");
7202 if (map
->map_state
== IMSM_T_STATE_UNINITIALIZED
)
7203 migrate(dev
, super
, IMSM_T_STATE_NORMAL
, MIGR_INIT
);
7205 migrate(dev
, super
, IMSM_T_STATE_NORMAL
, MIGR_REPAIR
);
7206 super
->updates_pending
++;
7210 /* skip checkpointing for general migration,
7211 * it is controlled in mdadm
7213 if (is_gen_migration(dev
))
7214 goto skip_mark_checkpoint
;
7216 /* check if we can update curr_migr_unit from resync_start, recovery_start */
7217 blocks_per_unit
= blocks_per_migr_unit(super
, dev
);
7218 if (blocks_per_unit
) {
7222 units
= a
->last_checkpoint
/ blocks_per_unit
;
7225 /* check that we did not overflow 32-bits, and that
7226 * curr_migr_unit needs updating
7228 if (units32
== units
&&
7230 __le32_to_cpu(dev
->vol
.curr_migr_unit
) != units32
) {
7231 dprintf("imsm: mark checkpoint (%u)\n", units32
);
7232 dev
->vol
.curr_migr_unit
= __cpu_to_le32(units32
);
7233 super
->updates_pending
++;
7237 skip_mark_checkpoint
:
7238 /* mark dirty / clean */
7239 if (dev
->vol
.dirty
!= !consistent
) {
7240 dprintf("imsm: mark '%s'\n", consistent
? "clean" : "dirty");
7245 super
->updates_pending
++;
7251 static void imsm_set_disk(struct active_array
*a
, int n
, int state
)
7253 int inst
= a
->info
.container_member
;
7254 struct intel_super
*super
= a
->container
->sb
;
7255 struct imsm_dev
*dev
= get_imsm_dev(super
, inst
);
7256 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
7257 struct imsm_disk
*disk
;
7262 if (n
> map
->num_members
)
7263 fprintf(stderr
, "imsm: set_disk %d out of range 0..%d\n",
7264 n
, map
->num_members
- 1);
7269 dprintf("imsm: set_disk %d:%x\n", n
, state
);
7271 ord
= get_imsm_ord_tbl_ent(dev
, n
, MAP_0
);
7272 disk
= get_imsm_disk(super
, ord_to_idx(ord
));
7274 /* check for new failures */
7275 if (state
& DS_FAULTY
) {
7276 if (mark_failure(dev
, disk
, ord_to_idx(ord
)))
7277 super
->updates_pending
++;
7280 /* check if in_sync */
7281 if (state
& DS_INSYNC
&& ord
& IMSM_ORD_REBUILD
&& is_rebuilding(dev
)) {
7282 struct imsm_map
*migr_map
= get_imsm_map(dev
, MAP_1
);
7284 set_imsm_ord_tbl_ent(migr_map
, n
, ord_to_idx(ord
));
7285 super
->updates_pending
++;
7288 failed
= imsm_count_failed(super
, dev
, MAP_0
);
7289 map_state
= imsm_check_degraded(super
, dev
, failed
, MAP_0
);
7291 /* check if recovery complete, newly degraded, or failed */
7292 dprintf("imsm: Detected transition to state ");
7293 switch (map_state
) {
7294 case IMSM_T_STATE_NORMAL
: /* transition to normal state */
7295 dprintf("normal: ");
7296 if (is_rebuilding(dev
)) {
7297 dprintf("while rebuilding");
7298 end_migration(dev
, super
, map_state
);
7299 map
= get_imsm_map(dev
, MAP_0
);
7300 map
->failed_disk_num
= ~0;
7301 super
->updates_pending
++;
7302 a
->last_checkpoint
= 0;
7305 if (is_gen_migration(dev
)) {
7306 dprintf("while general migration");
7307 if (a
->last_checkpoint
>= a
->info
.component_size
)
7308 end_migration(dev
, super
, map_state
);
7310 map
->map_state
= map_state
;
7311 map
= get_imsm_map(dev
, MAP_0
);
7312 map
->failed_disk_num
= ~0;
7313 super
->updates_pending
++;
7317 case IMSM_T_STATE_DEGRADED
: /* transition to degraded state */
7318 dprintf("degraded: ");
7319 if ((map
->map_state
!= map_state
) &&
7320 !dev
->vol
.migr_state
) {
7321 dprintf("mark degraded");
7322 map
->map_state
= map_state
;
7323 super
->updates_pending
++;
7324 a
->last_checkpoint
= 0;
7327 if (is_rebuilding(dev
)) {
7328 dprintf("while rebuilding.");
7329 if (map
->map_state
!= map_state
) {
7330 dprintf(" Map state change");
7331 end_migration(dev
, super
, map_state
);
7332 super
->updates_pending
++;
7336 if (is_gen_migration(dev
)) {
7337 dprintf("while general migration");
7338 if (a
->last_checkpoint
>= a
->info
.component_size
)
7339 end_migration(dev
, super
, map_state
);
7341 map
->map_state
= map_state
;
7342 manage_second_map(super
, dev
);
7344 super
->updates_pending
++;
7347 if (is_initializing(dev
)) {
7348 dprintf("while initialization.");
7349 map
->map_state
= map_state
;
7350 super
->updates_pending
++;
7354 case IMSM_T_STATE_FAILED
: /* transition to failed state */
7355 dprintf("failed: ");
7356 if (is_gen_migration(dev
)) {
7357 dprintf("while general migration");
7358 map
->map_state
= map_state
;
7359 super
->updates_pending
++;
7362 if (map
->map_state
!= map_state
) {
7363 dprintf("mark failed");
7364 end_migration(dev
, super
, map_state
);
7365 super
->updates_pending
++;
7366 a
->last_checkpoint
= 0;
7371 dprintf("state %i\n", map_state
);
7377 static int store_imsm_mpb(int fd
, struct imsm_super
*mpb
)
7380 __u32 mpb_size
= __le32_to_cpu(mpb
->mpb_size
);
7381 unsigned long long dsize
;
7382 unsigned long long sectors
;
7384 get_dev_size(fd
, NULL
, &dsize
);
7386 if (mpb_size
> 512) {
7387 /* -1 to account for anchor */
7388 sectors
= mpb_sectors(mpb
) - 1;
7390 /* write the extended mpb to the sectors preceeding the anchor */
7391 if (lseek64(fd
, dsize
- (512 * (2 + sectors
)), SEEK_SET
) < 0)
7394 if ((unsigned long long)write(fd
, buf
+ 512, 512 * sectors
)
7399 /* first block is stored on second to last sector of the disk */
7400 if (lseek64(fd
, dsize
- (512 * 2), SEEK_SET
) < 0)
7403 if (write(fd
, buf
, 512) != 512)
7409 static void imsm_sync_metadata(struct supertype
*container
)
7411 struct intel_super
*super
= container
->sb
;
7413 dprintf("sync metadata: %d\n", super
->updates_pending
);
7414 if (!super
->updates_pending
)
7417 write_super_imsm(container
, 0);
7419 super
->updates_pending
= 0;
7422 static struct dl
*imsm_readd(struct intel_super
*super
, int idx
, struct active_array
*a
)
7424 struct imsm_dev
*dev
= get_imsm_dev(super
, a
->info
.container_member
);
7425 int i
= get_imsm_disk_idx(dev
, idx
, MAP_X
);
7428 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
7432 if (dl
&& is_failed(&dl
->disk
))
7436 dprintf("%s: found %x:%x\n", __func__
, dl
->major
, dl
->minor
);
7441 static struct dl
*imsm_add_spare(struct intel_super
*super
, int slot
,
7442 struct active_array
*a
, int activate_new
,
7443 struct mdinfo
*additional_test_list
)
7445 struct imsm_dev
*dev
= get_imsm_dev(super
, a
->info
.container_member
);
7446 int idx
= get_imsm_disk_idx(dev
, slot
, MAP_X
);
7447 struct imsm_super
*mpb
= super
->anchor
;
7448 struct imsm_map
*map
;
7449 unsigned long long pos
;
7454 __u32 array_start
= 0;
7455 __u32 array_end
= 0;
7457 struct mdinfo
*test_list
;
7459 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
7460 /* If in this array, skip */
7461 for (d
= a
->info
.devs
; d
; d
= d
->next
)
7462 if (d
->state_fd
>= 0 &&
7463 d
->disk
.major
== dl
->major
&&
7464 d
->disk
.minor
== dl
->minor
) {
7465 dprintf("%x:%x already in array\n",
7466 dl
->major
, dl
->minor
);
7471 test_list
= additional_test_list
;
7473 if (test_list
->disk
.major
== dl
->major
&&
7474 test_list
->disk
.minor
== dl
->minor
) {
7475 dprintf("%x:%x already in additional test list\n",
7476 dl
->major
, dl
->minor
);
7479 test_list
= test_list
->next
;
7484 /* skip in use or failed drives */
7485 if (is_failed(&dl
->disk
) || idx
== dl
->index
||
7487 dprintf("%x:%x status (failed: %d index: %d)\n",
7488 dl
->major
, dl
->minor
, is_failed(&dl
->disk
), idx
);
7492 /* skip pure spares when we are looking for partially
7493 * assimilated drives
7495 if (dl
->index
== -1 && !activate_new
)
7498 /* Does this unused device have the requisite free space?
7499 * It needs to be able to cover all member volumes
7501 ex
= get_extents(super
, dl
);
7503 dprintf("cannot get extents\n");
7506 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
7507 dev
= get_imsm_dev(super
, i
);
7508 map
= get_imsm_map(dev
, MAP_0
);
7510 /* check if this disk is already a member of
7513 if (get_imsm_disk_slot(map
, dl
->index
) >= 0)
7519 array_start
= pba_of_lba0(map
);
7520 array_end
= array_start
+
7521 blocks_per_member(map
) - 1;
7524 /* check that we can start at pba_of_lba0 with
7525 * blocks_per_member of space
7527 if (array_start
>= pos
&& array_end
< ex
[j
].start
) {
7531 pos
= ex
[j
].start
+ ex
[j
].size
;
7533 } while (ex
[j
-1].size
);
7540 if (i
< mpb
->num_raid_devs
) {
7541 dprintf("%x:%x does not have %u to %u available\n",
7542 dl
->major
, dl
->minor
, array_start
, array_end
);
7553 static int imsm_rebuild_allowed(struct supertype
*cont
, int dev_idx
, int failed
)
7555 struct imsm_dev
*dev2
;
7556 struct imsm_map
*map
;
7562 dev2
= get_imsm_dev(cont
->sb
, dev_idx
);
7564 state
= imsm_check_degraded(cont
->sb
, dev2
, failed
, MAP_0
);
7565 if (state
== IMSM_T_STATE_FAILED
) {
7566 map
= get_imsm_map(dev2
, MAP_0
);
7569 for (slot
= 0; slot
< map
->num_members
; slot
++) {
7571 * Check if failed disks are deleted from intel
7572 * disk list or are marked to be deleted
7574 idx
= get_imsm_disk_idx(dev2
, slot
, MAP_X
);
7575 idisk
= get_imsm_dl_disk(cont
->sb
, idx
);
7577 * Do not rebuild the array if failed disks
7578 * from failed sub-array are not removed from
7582 is_failed(&idisk
->disk
) &&
7583 (idisk
->action
!= DISK_REMOVE
))
7591 static struct mdinfo
*imsm_activate_spare(struct active_array
*a
,
7592 struct metadata_update
**updates
)
7595 * Find a device with unused free space and use it to replace a
7596 * failed/vacant region in an array. We replace failed regions one a
7597 * array at a time. The result is that a new spare disk will be added
7598 * to the first failed array and after the monitor has finished
7599 * propagating failures the remainder will be consumed.
7601 * FIXME add a capability for mdmon to request spares from another
7605 struct intel_super
*super
= a
->container
->sb
;
7606 int inst
= a
->info
.container_member
;
7607 struct imsm_dev
*dev
= get_imsm_dev(super
, inst
);
7608 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
7609 int failed
= a
->info
.array
.raid_disks
;
7610 struct mdinfo
*rv
= NULL
;
7613 struct metadata_update
*mu
;
7615 struct imsm_update_activate_spare
*u
;
7620 for (d
= a
->info
.devs
; d
; d
= d
->next
) {
7621 if ((d
->curr_state
& DS_FAULTY
) &&
7623 /* wait for Removal to happen */
7625 if (d
->state_fd
>= 0)
7629 dprintf("imsm: activate spare: inst=%d failed=%d (%d) level=%d\n",
7630 inst
, failed
, a
->info
.array
.raid_disks
, a
->info
.array
.level
);
7632 if (imsm_reshape_blocks_arrays_changes(super
))
7635 /* Cannot activate another spare if rebuild is in progress already
7637 if (is_rebuilding(dev
)) {
7638 dprintf("imsm: No spare activation allowed. "
7639 "Rebuild in progress already.\n");
7643 if (a
->info
.array
.level
== 4)
7644 /* No repair for takeovered array
7645 * imsm doesn't support raid4
7649 if (imsm_check_degraded(super
, dev
, failed
, MAP_0
) !=
7650 IMSM_T_STATE_DEGRADED
)
7654 * If there are any failed disks check state of the other volume.
7655 * Block rebuild if the another one is failed until failed disks
7656 * are removed from container.
7659 dprintf("found failed disks in %.*s, check if there another"
7660 "failed sub-array.\n",
7661 MAX_RAID_SERIAL_LEN
, dev
->volume
);
7662 /* check if states of the other volumes allow for rebuild */
7663 for (i
= 0; i
< super
->anchor
->num_raid_devs
; i
++) {
7665 allowed
= imsm_rebuild_allowed(a
->container
,
7673 /* For each slot, if it is not working, find a spare */
7674 for (i
= 0; i
< a
->info
.array
.raid_disks
; i
++) {
7675 for (d
= a
->info
.devs
; d
; d
= d
->next
)
7676 if (d
->disk
.raid_disk
== i
)
7678 dprintf("found %d: %p %x\n", i
, d
, d
?d
->curr_state
:0);
7679 if (d
&& (d
->state_fd
>= 0))
7683 * OK, this device needs recovery. Try to re-add the
7684 * previous occupant of this slot, if this fails see if
7685 * we can continue the assimilation of a spare that was
7686 * partially assimilated, finally try to activate a new
7689 dl
= imsm_readd(super
, i
, a
);
7691 dl
= imsm_add_spare(super
, i
, a
, 0, rv
);
7693 dl
= imsm_add_spare(super
, i
, a
, 1, rv
);
7697 /* found a usable disk with enough space */
7698 di
= malloc(sizeof(*di
));
7701 memset(di
, 0, sizeof(*di
));
7703 /* dl->index will be -1 in the case we are activating a
7704 * pristine spare. imsm_process_update() will create a
7705 * new index in this case. Once a disk is found to be
7706 * failed in all member arrays it is kicked from the
7709 di
->disk
.number
= dl
->index
;
7711 /* (ab)use di->devs to store a pointer to the device
7714 di
->devs
= (struct mdinfo
*) dl
;
7716 di
->disk
.raid_disk
= i
;
7717 di
->disk
.major
= dl
->major
;
7718 di
->disk
.minor
= dl
->minor
;
7720 di
->recovery_start
= 0;
7721 di
->data_offset
= pba_of_lba0(map
);
7722 di
->component_size
= a
->info
.component_size
;
7723 di
->container_member
= inst
;
7724 super
->random
= random32();
7728 dprintf("%x:%x to be %d at %llu\n", dl
->major
, dl
->minor
,
7729 i
, di
->data_offset
);
7733 /* No spares found */
7735 /* Now 'rv' has a list of devices to return.
7736 * Create a metadata_update record to update the
7737 * disk_ord_tbl for the array
7739 mu
= malloc(sizeof(*mu
));
7741 mu
->buf
= malloc(sizeof(struct imsm_update_activate_spare
) * num_spares
);
7742 if (mu
->buf
== NULL
) {
7749 struct mdinfo
*n
= rv
->next
;
7758 mu
->space_list
= NULL
;
7759 mu
->len
= sizeof(struct imsm_update_activate_spare
) * num_spares
;
7760 mu
->next
= *updates
;
7761 u
= (struct imsm_update_activate_spare
*) mu
->buf
;
7763 for (di
= rv
; di
; di
= di
->next
) {
7764 u
->type
= update_activate_spare
;
7765 u
->dl
= (struct dl
*) di
->devs
;
7767 u
->slot
= di
->disk
.raid_disk
;
7778 static int disks_overlap(struct intel_super
*super
, int idx
, struct imsm_update_create_array
*u
)
7780 struct imsm_dev
*dev
= get_imsm_dev(super
, idx
);
7781 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
7782 struct imsm_map
*new_map
= get_imsm_map(&u
->dev
, MAP_0
);
7783 struct disk_info
*inf
= get_disk_info(u
);
7784 struct imsm_disk
*disk
;
7788 for (i
= 0; i
< map
->num_members
; i
++) {
7789 disk
= get_imsm_disk(super
, get_imsm_disk_idx(dev
, i
, MAP_X
));
7790 for (j
= 0; j
< new_map
->num_members
; j
++)
7791 if (serialcmp(disk
->serial
, inf
[j
].serial
) == 0)
7799 static struct dl
*get_disk_super(struct intel_super
*super
, int major
, int minor
)
7801 struct dl
*dl
= NULL
;
7802 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
7803 if ((dl
->major
== major
) && (dl
->minor
== minor
))
7808 static int remove_disk_super(struct intel_super
*super
, int major
, int minor
)
7810 struct dl
*prev
= NULL
;
7814 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
7815 if ((dl
->major
== major
) && (dl
->minor
== minor
)) {
7818 prev
->next
= dl
->next
;
7820 super
->disks
= dl
->next
;
7822 __free_imsm_disk(dl
);
7823 dprintf("%s: removed %x:%x\n",
7824 __func__
, major
, minor
);
7832 static void imsm_delete(struct intel_super
*super
, struct dl
**dlp
, unsigned index
);
7834 static int add_remove_disk_update(struct intel_super
*super
)
7836 int check_degraded
= 0;
7837 struct dl
*disk
= NULL
;
7838 /* add/remove some spares to/from the metadata/contrainer */
7839 while (super
->disk_mgmt_list
) {
7840 struct dl
*disk_cfg
;
7842 disk_cfg
= super
->disk_mgmt_list
;
7843 super
->disk_mgmt_list
= disk_cfg
->next
;
7844 disk_cfg
->next
= NULL
;
7846 if (disk_cfg
->action
== DISK_ADD
) {
7847 disk_cfg
->next
= super
->disks
;
7848 super
->disks
= disk_cfg
;
7850 dprintf("%s: added %x:%x\n",
7851 __func__
, disk_cfg
->major
,
7853 } else if (disk_cfg
->action
== DISK_REMOVE
) {
7854 dprintf("Disk remove action processed: %x.%x\n",
7855 disk_cfg
->major
, disk_cfg
->minor
);
7856 disk
= get_disk_super(super
,
7860 /* store action status */
7861 disk
->action
= DISK_REMOVE
;
7862 /* remove spare disks only */
7863 if (disk
->index
== -1) {
7864 remove_disk_super(super
,
7869 /* release allocate disk structure */
7870 __free_imsm_disk(disk_cfg
);
7873 return check_degraded
;
7877 static int apply_reshape_migration_update(struct imsm_update_reshape_migration
*u
,
7878 struct intel_super
*super
,
7881 struct intel_dev
*id
;
7882 void **tofree
= NULL
;
7885 dprintf("apply_reshape_migration_update()\n");
7886 if ((u
->subdev
< 0) ||
7888 dprintf("imsm: Error: Wrong subdev: %i\n", u
->subdev
);
7891 if ((space_list
== NULL
) || (*space_list
== NULL
)) {
7892 dprintf("imsm: Error: Memory is not allocated\n");
7896 for (id
= super
->devlist
; id
; id
= id
->next
) {
7897 if (id
->index
== (unsigned)u
->subdev
) {
7898 struct imsm_dev
*dev
= get_imsm_dev(super
, u
->subdev
);
7899 struct imsm_map
*map
;
7900 struct imsm_dev
*new_dev
=
7901 (struct imsm_dev
*)*space_list
;
7902 struct imsm_map
*migr_map
= get_imsm_map(dev
, MAP_1
);
7904 struct dl
*new_disk
;
7906 if (new_dev
== NULL
)
7908 *space_list
= **space_list
;
7909 memcpy(new_dev
, dev
, sizeof_imsm_dev(dev
, 0));
7910 map
= get_imsm_map(new_dev
, MAP_0
);
7912 dprintf("imsm: Error: migration in progress");
7916 to_state
= map
->map_state
;
7917 if ((u
->new_level
== 5) && (map
->raid_level
== 0)) {
7919 /* this should not happen */
7920 if (u
->new_disks
[0] < 0) {
7921 map
->failed_disk_num
=
7922 map
->num_members
- 1;
7923 to_state
= IMSM_T_STATE_DEGRADED
;
7925 to_state
= IMSM_T_STATE_NORMAL
;
7927 migrate(new_dev
, super
, to_state
, MIGR_GEN_MIGR
);
7928 if (u
->new_level
> -1)
7929 map
->raid_level
= u
->new_level
;
7930 migr_map
= get_imsm_map(new_dev
, MAP_1
);
7931 if ((u
->new_level
== 5) &&
7932 (migr_map
->raid_level
== 0)) {
7933 int ord
= map
->num_members
- 1;
7934 migr_map
->num_members
--;
7935 if (u
->new_disks
[0] < 0)
7936 ord
|= IMSM_ORD_REBUILD
;
7937 set_imsm_ord_tbl_ent(map
,
7938 map
->num_members
- 1,
7942 tofree
= (void **)dev
;
7944 /* update chunk size
7946 if (u
->new_chunksize
> 0)
7947 map
->blocks_per_strip
=
7948 __cpu_to_le16(u
->new_chunksize
* 2);
7952 if ((u
->new_level
!= 5) ||
7953 (migr_map
->raid_level
!= 0) ||
7954 (migr_map
->raid_level
== map
->raid_level
))
7957 if (u
->new_disks
[0] >= 0) {
7960 new_disk
= get_disk_super(super
,
7961 major(u
->new_disks
[0]),
7962 minor(u
->new_disks
[0]));
7963 dprintf("imsm: new disk for reshape is: %i:%i "
7964 "(%p, index = %i)\n",
7965 major(u
->new_disks
[0]),
7966 minor(u
->new_disks
[0]),
7967 new_disk
, new_disk
->index
);
7968 if (new_disk
== NULL
)
7969 goto error_disk_add
;
7971 new_disk
->index
= map
->num_members
- 1;
7972 /* slot to fill in autolayout
7974 new_disk
->raiddisk
= new_disk
->index
;
7975 new_disk
->disk
.status
|= CONFIGURED_DISK
;
7976 new_disk
->disk
.status
&= ~SPARE_DISK
;
7978 goto error_disk_add
;
7981 *tofree
= *space_list
;
7982 /* calculate new size
7984 imsm_set_array_size(new_dev
, -1);
7991 *space_list
= tofree
;
7995 dprintf("Error: imsm: Cannot find disk.\n");
7999 static int apply_size_change_update(struct imsm_update_size_change
*u
,
8000 struct intel_super
*super
)
8002 struct intel_dev
*id
;
8005 dprintf("apply_size_change_update()\n");
8006 if ((u
->subdev
< 0) ||
8008 dprintf("imsm: Error: Wrong subdev: %i\n", u
->subdev
);
8012 for (id
= super
->devlist
; id
; id
= id
->next
) {
8013 if (id
->index
== (unsigned)u
->subdev
) {
8014 struct imsm_dev
*dev
= get_imsm_dev(super
, u
->subdev
);
8015 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
8016 int used_disks
= imsm_num_data_members(dev
, MAP_0
);
8017 unsigned long long blocks_per_member
;
8019 /* calculate new size
8021 blocks_per_member
= u
->new_size
/ used_disks
;
8022 dprintf("imsm: apply_size_change_update(size: %llu, "
8023 "blocks per member: %llu)\n",
8024 u
->new_size
, blocks_per_member
);
8025 set_blocks_per_member(map
, blocks_per_member
);
8026 imsm_set_array_size(dev
, u
->new_size
);
8037 static int apply_update_activate_spare(struct imsm_update_activate_spare
*u
,
8038 struct intel_super
*super
,
8039 struct active_array
*active_array
)
8041 struct imsm_super
*mpb
= super
->anchor
;
8042 struct imsm_dev
*dev
= get_imsm_dev(super
, u
->array
);
8043 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
8044 struct imsm_map
*migr_map
;
8045 struct active_array
*a
;
8046 struct imsm_disk
*disk
;
8053 int second_map_created
= 0;
8055 for (; u
; u
= u
->next
) {
8056 victim
= get_imsm_disk_idx(dev
, u
->slot
, MAP_X
);
8061 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
8066 fprintf(stderr
, "error: imsm_activate_spare passed "
8067 "an unknown disk (index: %d)\n",
8072 /* count failures (excluding rebuilds and the victim)
8073 * to determine map[0] state
8076 for (i
= 0; i
< map
->num_members
; i
++) {
8079 disk
= get_imsm_disk(super
,
8080 get_imsm_disk_idx(dev
, i
, MAP_X
));
8081 if (!disk
|| is_failed(disk
))
8085 /* adding a pristine spare, assign a new index */
8086 if (dl
->index
< 0) {
8087 dl
->index
= super
->anchor
->num_disks
;
8088 super
->anchor
->num_disks
++;
8091 disk
->status
|= CONFIGURED_DISK
;
8092 disk
->status
&= ~SPARE_DISK
;
8095 to_state
= imsm_check_degraded(super
, dev
, failed
, MAP_0
);
8096 if (!second_map_created
) {
8097 second_map_created
= 1;
8098 map
->map_state
= IMSM_T_STATE_DEGRADED
;
8099 migrate(dev
, super
, to_state
, MIGR_REBUILD
);
8101 map
->map_state
= to_state
;
8102 migr_map
= get_imsm_map(dev
, MAP_1
);
8103 set_imsm_ord_tbl_ent(map
, u
->slot
, dl
->index
);
8104 set_imsm_ord_tbl_ent(migr_map
, u
->slot
,
8105 dl
->index
| IMSM_ORD_REBUILD
);
8107 /* update the family_num to mark a new container
8108 * generation, being careful to record the existing
8109 * family_num in orig_family_num to clean up after
8110 * earlier mdadm versions that neglected to set it.
8112 if (mpb
->orig_family_num
== 0)
8113 mpb
->orig_family_num
= mpb
->family_num
;
8114 mpb
->family_num
+= super
->random
;
8116 /* count arrays using the victim in the metadata */
8118 for (a
= active_array
; a
; a
= a
->next
) {
8119 dev
= get_imsm_dev(super
, a
->info
.container_member
);
8120 map
= get_imsm_map(dev
, MAP_0
);
8122 if (get_imsm_disk_slot(map
, victim
) >= 0)
8126 /* delete the victim if it is no longer being
8132 /* We know that 'manager' isn't touching anything,
8133 * so it is safe to delete
8135 for (dlp
= &super
->disks
; *dlp
; dlp
= &(*dlp
)->next
)
8136 if ((*dlp
)->index
== victim
)
8139 /* victim may be on the missing list */
8141 for (dlp
= &super
->missing
; *dlp
;
8142 dlp
= &(*dlp
)->next
)
8143 if ((*dlp
)->index
== victim
)
8145 imsm_delete(super
, dlp
, victim
);
8152 static int apply_reshape_container_disks_update(struct imsm_update_reshape
*u
,
8153 struct intel_super
*super
,
8156 struct dl
*new_disk
;
8157 struct intel_dev
*id
;
8159 int delta_disks
= u
->new_raid_disks
- u
->old_raid_disks
;
8160 int disk_count
= u
->old_raid_disks
;
8161 void **tofree
= NULL
;
8162 int devices_to_reshape
= 1;
8163 struct imsm_super
*mpb
= super
->anchor
;
8165 unsigned int dev_id
;
8167 dprintf("imsm: apply_reshape_container_disks_update()\n");
8169 /* enable spares to use in array */
8170 for (i
= 0; i
< delta_disks
; i
++) {
8171 new_disk
= get_disk_super(super
,
8172 major(u
->new_disks
[i
]),
8173 minor(u
->new_disks
[i
]));
8174 dprintf("imsm: new disk for reshape is: %i:%i "
8175 "(%p, index = %i)\n",
8176 major(u
->new_disks
[i
]), minor(u
->new_disks
[i
]),
8177 new_disk
, new_disk
->index
);
8178 if ((new_disk
== NULL
) ||
8179 ((new_disk
->index
>= 0) &&
8180 (new_disk
->index
< u
->old_raid_disks
)))
8181 goto update_reshape_exit
;
8182 new_disk
->index
= disk_count
++;
8183 /* slot to fill in autolayout
8185 new_disk
->raiddisk
= new_disk
->index
;
8186 new_disk
->disk
.status
|=
8188 new_disk
->disk
.status
&= ~SPARE_DISK
;
8191 dprintf("imsm: volume set mpb->num_raid_devs = %i\n",
8192 mpb
->num_raid_devs
);
8193 /* manage changes in volume
8195 for (dev_id
= 0; dev_id
< mpb
->num_raid_devs
; dev_id
++) {
8196 void **sp
= *space_list
;
8197 struct imsm_dev
*newdev
;
8198 struct imsm_map
*newmap
, *oldmap
;
8200 for (id
= super
->devlist
; id
; id
= id
->next
) {
8201 if (id
->index
== dev_id
)
8210 /* Copy the dev, but not (all of) the map */
8211 memcpy(newdev
, id
->dev
, sizeof(*newdev
));
8212 oldmap
= get_imsm_map(id
->dev
, MAP_0
);
8213 newmap
= get_imsm_map(newdev
, MAP_0
);
8214 /* Copy the current map */
8215 memcpy(newmap
, oldmap
, sizeof_imsm_map(oldmap
));
8216 /* update one device only
8218 if (devices_to_reshape
) {
8219 dprintf("imsm: modifying subdev: %i\n",
8221 devices_to_reshape
--;
8222 newdev
->vol
.migr_state
= 1;
8223 newdev
->vol
.curr_migr_unit
= 0;
8224 set_migr_type(newdev
, MIGR_GEN_MIGR
);
8225 newmap
->num_members
= u
->new_raid_disks
;
8226 for (i
= 0; i
< delta_disks
; i
++) {
8227 set_imsm_ord_tbl_ent(newmap
,
8228 u
->old_raid_disks
+ i
,
8229 u
->old_raid_disks
+ i
);
8231 /* New map is correct, now need to save old map
8233 newmap
= get_imsm_map(newdev
, MAP_1
);
8234 memcpy(newmap
, oldmap
, sizeof_imsm_map(oldmap
));
8236 imsm_set_array_size(newdev
, -1);
8239 sp
= (void **)id
->dev
;
8244 /* Clear migration record */
8245 memset(super
->migr_rec
, 0, sizeof(struct migr_record
));
8248 *space_list
= tofree
;
8251 update_reshape_exit
:
8256 static int apply_takeover_update(struct imsm_update_takeover
*u
,
8257 struct intel_super
*super
,
8260 struct imsm_dev
*dev
= NULL
;
8261 struct intel_dev
*dv
;
8262 struct imsm_dev
*dev_new
;
8263 struct imsm_map
*map
;
8267 for (dv
= super
->devlist
; dv
; dv
= dv
->next
)
8268 if (dv
->index
== (unsigned int)u
->subarray
) {
8276 map
= get_imsm_map(dev
, MAP_0
);
8278 if (u
->direction
== R10_TO_R0
) {
8279 /* Number of failed disks must be half of initial disk number */
8280 if (imsm_count_failed(super
, dev
, MAP_0
) !=
8281 (map
->num_members
/ 2))
8284 /* iterate through devices to mark removed disks as spare */
8285 for (dm
= super
->disks
; dm
; dm
= dm
->next
) {
8286 if (dm
->disk
.status
& FAILED_DISK
) {
8287 int idx
= dm
->index
;
8288 /* update indexes on the disk list */
8289 /* FIXME this loop-with-the-loop looks wrong, I'm not convinced
8290 the index values will end up being correct.... NB */
8291 for (du
= super
->disks
; du
; du
= du
->next
)
8292 if (du
->index
> idx
)
8294 /* mark as spare disk */
8299 map
->num_members
= map
->num_members
/ 2;
8300 map
->map_state
= IMSM_T_STATE_NORMAL
;
8301 map
->num_domains
= 1;
8302 map
->raid_level
= 0;
8303 map
->failed_disk_num
= -1;
8306 if (u
->direction
== R0_TO_R10
) {
8308 /* update slots in current disk list */
8309 for (dm
= super
->disks
; dm
; dm
= dm
->next
) {
8313 /* create new *missing* disks */
8314 for (i
= 0; i
< map
->num_members
; i
++) {
8315 space
= *space_list
;
8318 *space_list
= *space
;
8320 memcpy(du
, super
->disks
, sizeof(*du
));
8324 du
->index
= (i
* 2) + 1;
8325 sprintf((char *)du
->disk
.serial
,
8326 " MISSING_%d", du
->index
);
8327 sprintf((char *)du
->serial
,
8328 "MISSING_%d", du
->index
);
8329 du
->next
= super
->missing
;
8330 super
->missing
= du
;
8332 /* create new dev and map */
8333 space
= *space_list
;
8336 *space_list
= *space
;
8337 dev_new
= (void *)space
;
8338 memcpy(dev_new
, dev
, sizeof(*dev
));
8339 /* update new map */
8340 map
= get_imsm_map(dev_new
, MAP_0
);
8341 map
->num_members
= map
->num_members
* 2;
8342 map
->map_state
= IMSM_T_STATE_DEGRADED
;
8343 map
->num_domains
= 2;
8344 map
->raid_level
= 1;
8345 /* replace dev<->dev_new */
8348 /* update disk order table */
8349 for (du
= super
->disks
; du
; du
= du
->next
)
8351 set_imsm_ord_tbl_ent(map
, du
->index
, du
->index
);
8352 for (du
= super
->missing
; du
; du
= du
->next
)
8353 if (du
->index
>= 0) {
8354 set_imsm_ord_tbl_ent(map
, du
->index
, du
->index
);
8355 mark_missing(dv
->dev
, &du
->disk
, du
->index
);
8361 static void imsm_process_update(struct supertype
*st
,
8362 struct metadata_update
*update
)
8365 * crack open the metadata_update envelope to find the update record
8366 * update can be one of:
8367 * update_reshape_container_disks - all the arrays in the container
8368 * are being reshaped to have more devices. We need to mark
8369 * the arrays for general migration and convert selected spares
8370 * into active devices.
8371 * update_activate_spare - a spare device has replaced a failed
8372 * device in an array, update the disk_ord_tbl. If this disk is
8373 * present in all member arrays then also clear the SPARE_DISK
8375 * update_create_array
8377 * update_rename_array
8378 * update_add_remove_disk
8380 struct intel_super
*super
= st
->sb
;
8381 struct imsm_super
*mpb
;
8382 enum imsm_update_type type
= *(enum imsm_update_type
*) update
->buf
;
8384 /* update requires a larger buf but the allocation failed */
8385 if (super
->next_len
&& !super
->next_buf
) {
8386 super
->next_len
= 0;
8390 if (super
->next_buf
) {
8391 memcpy(super
->next_buf
, super
->buf
, super
->len
);
8393 super
->len
= super
->next_len
;
8394 super
->buf
= super
->next_buf
;
8396 super
->next_len
= 0;
8397 super
->next_buf
= NULL
;
8400 mpb
= super
->anchor
;
8403 case update_general_migration_checkpoint
: {
8404 struct intel_dev
*id
;
8405 struct imsm_update_general_migration_checkpoint
*u
=
8406 (void *)update
->buf
;
8408 dprintf("imsm: process_update() "
8409 "for update_general_migration_checkpoint called\n");
8411 /* find device under general migration */
8412 for (id
= super
->devlist
; id
; id
= id
->next
) {
8413 if (is_gen_migration(id
->dev
)) {
8414 id
->dev
->vol
.curr_migr_unit
=
8415 __cpu_to_le32(u
->curr_migr_unit
);
8416 super
->updates_pending
++;
8421 case update_takeover
: {
8422 struct imsm_update_takeover
*u
= (void *)update
->buf
;
8423 if (apply_takeover_update(u
, super
, &update
->space_list
)) {
8424 imsm_update_version_info(super
);
8425 super
->updates_pending
++;
8430 case update_reshape_container_disks
: {
8431 struct imsm_update_reshape
*u
= (void *)update
->buf
;
8432 if (apply_reshape_container_disks_update(
8433 u
, super
, &update
->space_list
))
8434 super
->updates_pending
++;
8437 case update_reshape_migration
: {
8438 struct imsm_update_reshape_migration
*u
= (void *)update
->buf
;
8439 if (apply_reshape_migration_update(
8440 u
, super
, &update
->space_list
))
8441 super
->updates_pending
++;
8444 case update_size_change
: {
8445 struct imsm_update_size_change
*u
= (void *)update
->buf
;
8446 if (apply_size_change_update(u
, super
))
8447 super
->updates_pending
++;
8450 case update_activate_spare
: {
8451 struct imsm_update_activate_spare
*u
= (void *) update
->buf
;
8452 if (apply_update_activate_spare(u
, super
, st
->arrays
))
8453 super
->updates_pending
++;
8456 case update_create_array
: {
8457 /* someone wants to create a new array, we need to be aware of
8458 * a few races/collisions:
8459 * 1/ 'Create' called by two separate instances of mdadm
8460 * 2/ 'Create' versus 'activate_spare': mdadm has chosen
8461 * devices that have since been assimilated via
8463 * In the event this update can not be carried out mdadm will
8464 * (FIX ME) notice that its update did not take hold.
8466 struct imsm_update_create_array
*u
= (void *) update
->buf
;
8467 struct intel_dev
*dv
;
8468 struct imsm_dev
*dev
;
8469 struct imsm_map
*map
, *new_map
;
8470 unsigned long long start
, end
;
8471 unsigned long long new_start
, new_end
;
8473 struct disk_info
*inf
;
8476 /* handle racing creates: first come first serve */
8477 if (u
->dev_idx
< mpb
->num_raid_devs
) {
8478 dprintf("%s: subarray %d already defined\n",
8479 __func__
, u
->dev_idx
);
8483 /* check update is next in sequence */
8484 if (u
->dev_idx
!= mpb
->num_raid_devs
) {
8485 dprintf("%s: can not create array %d expected index %d\n",
8486 __func__
, u
->dev_idx
, mpb
->num_raid_devs
);
8490 new_map
= get_imsm_map(&u
->dev
, MAP_0
);
8491 new_start
= pba_of_lba0(new_map
);
8492 new_end
= new_start
+ blocks_per_member(new_map
);
8493 inf
= get_disk_info(u
);
8495 /* handle activate_spare versus create race:
8496 * check to make sure that overlapping arrays do not include
8499 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
8500 dev
= get_imsm_dev(super
, i
);
8501 map
= get_imsm_map(dev
, MAP_0
);
8502 start
= pba_of_lba0(map
);
8503 end
= start
+ blocks_per_member(map
);
8504 if ((new_start
>= start
&& new_start
<= end
) ||
8505 (start
>= new_start
&& start
<= new_end
))
8510 if (disks_overlap(super
, i
, u
)) {
8511 dprintf("%s: arrays overlap\n", __func__
);
8516 /* check that prepare update was successful */
8517 if (!update
->space
) {
8518 dprintf("%s: prepare update failed\n", __func__
);
8522 /* check that all disks are still active before committing
8523 * changes. FIXME: could we instead handle this by creating a
8524 * degraded array? That's probably not what the user expects,
8525 * so better to drop this update on the floor.
8527 for (i
= 0; i
< new_map
->num_members
; i
++) {
8528 dl
= serial_to_dl(inf
[i
].serial
, super
);
8530 dprintf("%s: disk disappeared\n", __func__
);
8535 super
->updates_pending
++;
8537 /* convert spares to members and fixup ord_tbl */
8538 for (i
= 0; i
< new_map
->num_members
; i
++) {
8539 dl
= serial_to_dl(inf
[i
].serial
, super
);
8540 if (dl
->index
== -1) {
8541 dl
->index
= mpb
->num_disks
;
8543 dl
->disk
.status
|= CONFIGURED_DISK
;
8544 dl
->disk
.status
&= ~SPARE_DISK
;
8546 set_imsm_ord_tbl_ent(new_map
, i
, dl
->index
);
8551 update
->space
= NULL
;
8552 imsm_copy_dev(dev
, &u
->dev
);
8553 dv
->index
= u
->dev_idx
;
8554 dv
->next
= super
->devlist
;
8555 super
->devlist
= dv
;
8556 mpb
->num_raid_devs
++;
8558 imsm_update_version_info(super
);
8561 /* mdmon knows how to release update->space, but not
8562 * ((struct intel_dev *) update->space)->dev
8564 if (update
->space
) {
8570 case update_kill_array
: {
8571 struct imsm_update_kill_array
*u
= (void *) update
->buf
;
8572 int victim
= u
->dev_idx
;
8573 struct active_array
*a
;
8574 struct intel_dev
**dp
;
8575 struct imsm_dev
*dev
;
8577 /* sanity check that we are not affecting the uuid of
8578 * active arrays, or deleting an active array
8580 * FIXME when immutable ids are available, but note that
8581 * we'll also need to fixup the invalidated/active
8582 * subarray indexes in mdstat
8584 for (a
= st
->arrays
; a
; a
= a
->next
)
8585 if (a
->info
.container_member
>= victim
)
8587 /* by definition if mdmon is running at least one array
8588 * is active in the container, so checking
8589 * mpb->num_raid_devs is just extra paranoia
8591 dev
= get_imsm_dev(super
, victim
);
8592 if (a
|| !dev
|| mpb
->num_raid_devs
== 1) {
8593 dprintf("failed to delete subarray-%d\n", victim
);
8597 for (dp
= &super
->devlist
; *dp
;)
8598 if ((*dp
)->index
== (unsigned)super
->current_vol
) {
8601 if ((*dp
)->index
> (unsigned)victim
)
8605 mpb
->num_raid_devs
--;
8606 super
->updates_pending
++;
8609 case update_rename_array
: {
8610 struct imsm_update_rename_array
*u
= (void *) update
->buf
;
8611 char name
[MAX_RAID_SERIAL_LEN
+1];
8612 int target
= u
->dev_idx
;
8613 struct active_array
*a
;
8614 struct imsm_dev
*dev
;
8616 /* sanity check that we are not affecting the uuid of
8619 snprintf(name
, MAX_RAID_SERIAL_LEN
, "%s", (char *) u
->name
);
8620 name
[MAX_RAID_SERIAL_LEN
] = '\0';
8621 for (a
= st
->arrays
; a
; a
= a
->next
)
8622 if (a
->info
.container_member
== target
)
8624 dev
= get_imsm_dev(super
, u
->dev_idx
);
8625 if (a
|| !dev
|| !check_name(super
, name
, 1)) {
8626 dprintf("failed to rename subarray-%d\n", target
);
8630 snprintf((char *) dev
->volume
, MAX_RAID_SERIAL_LEN
, "%s", name
);
8631 super
->updates_pending
++;
8634 case update_add_remove_disk
: {
8635 /* we may be able to repair some arrays if disks are
8636 * being added, check teh status of add_remove_disk
8637 * if discs has been added.
8639 if (add_remove_disk_update(super
)) {
8640 struct active_array
*a
;
8642 super
->updates_pending
++;
8643 for (a
= st
->arrays
; a
; a
= a
->next
)
8644 a
->check_degraded
= 1;
8649 fprintf(stderr
, "error: unsuported process update type:"
8650 "(type: %d)\n", type
);
8654 static struct mdinfo
*get_spares_for_grow(struct supertype
*st
);
8656 static void imsm_prepare_update(struct supertype
*st
,
8657 struct metadata_update
*update
)
8660 * Allocate space to hold new disk entries, raid-device entries or a new
8661 * mpb if necessary. The manager synchronously waits for updates to
8662 * complete in the monitor, so new mpb buffers allocated here can be
8663 * integrated by the monitor thread without worrying about live pointers
8664 * in the manager thread.
8666 enum imsm_update_type type
= *(enum imsm_update_type
*) update
->buf
;
8667 struct intel_super
*super
= st
->sb
;
8668 struct imsm_super
*mpb
= super
->anchor
;
8673 case update_general_migration_checkpoint
:
8674 dprintf("imsm: prepare_update() "
8675 "for update_general_migration_checkpoint called\n");
8677 case update_takeover
: {
8678 struct imsm_update_takeover
*u
= (void *)update
->buf
;
8679 if (u
->direction
== R0_TO_R10
) {
8680 void **tail
= (void **)&update
->space_list
;
8681 struct imsm_dev
*dev
= get_imsm_dev(super
, u
->subarray
);
8682 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
8683 int num_members
= map
->num_members
;
8687 /* allocate memory for added disks */
8688 for (i
= 0; i
< num_members
; i
++) {
8689 size
= sizeof(struct dl
);
8690 space
= malloc(size
);
8699 /* allocate memory for new device */
8700 size
= sizeof_imsm_dev(super
->devlist
->dev
, 0) +
8701 (num_members
* sizeof(__u32
));
8702 space
= malloc(size
);
8711 len
= disks_to_mpb_size(num_members
* 2);
8713 /* if allocation didn't success, free buffer */
8714 while (update
->space_list
) {
8715 void **sp
= update
->space_list
;
8716 update
->space_list
= *sp
;
8724 case update_reshape_container_disks
: {
8725 /* Every raid device in the container is about to
8726 * gain some more devices, and we will enter a
8728 * So each 'imsm_map' will be bigger, and the imsm_vol
8729 * will now hold 2 of them.
8730 * Thus we need new 'struct imsm_dev' allocations sized
8731 * as sizeof_imsm_dev but with more devices in both maps.
8733 struct imsm_update_reshape
*u
= (void *)update
->buf
;
8734 struct intel_dev
*dl
;
8735 void **space_tail
= (void**)&update
->space_list
;
8737 dprintf("imsm: imsm_prepare_update() for update_reshape\n");
8739 for (dl
= super
->devlist
; dl
; dl
= dl
->next
) {
8740 int size
= sizeof_imsm_dev(dl
->dev
, 1);
8742 if (u
->new_raid_disks
> u
->old_raid_disks
)
8743 size
+= sizeof(__u32
)*2*
8744 (u
->new_raid_disks
- u
->old_raid_disks
);
8753 len
= disks_to_mpb_size(u
->new_raid_disks
);
8754 dprintf("New anchor length is %llu\n", (unsigned long long)len
);
8757 case update_reshape_migration
: {
8758 /* for migration level 0->5 we need to add disks
8759 * so the same as for container operation we will copy
8760 * device to the bigger location.
8761 * in memory prepared device and new disk area are prepared
8762 * for usage in process update
8764 struct imsm_update_reshape_migration
*u
= (void *)update
->buf
;
8765 struct intel_dev
*id
;
8766 void **space_tail
= (void **)&update
->space_list
;
8769 int current_level
= -1;
8771 dprintf("imsm: imsm_prepare_update() for update_reshape\n");
8773 /* add space for bigger array in update
8775 for (id
= super
->devlist
; id
; id
= id
->next
) {
8776 if (id
->index
== (unsigned)u
->subdev
) {
8777 size
= sizeof_imsm_dev(id
->dev
, 1);
8778 if (u
->new_raid_disks
> u
->old_raid_disks
)
8779 size
+= sizeof(__u32
)*2*
8780 (u
->new_raid_disks
- u
->old_raid_disks
);
8790 if (update
->space_list
== NULL
)
8793 /* add space for disk in update
8795 size
= sizeof(struct dl
);
8798 free(update
->space_list
);
8799 update
->space_list
= NULL
;
8806 /* add spare device to update
8808 for (id
= super
->devlist
; id
; id
= id
->next
)
8809 if (id
->index
== (unsigned)u
->subdev
) {
8810 struct imsm_dev
*dev
;
8811 struct imsm_map
*map
;
8813 dev
= get_imsm_dev(super
, u
->subdev
);
8814 map
= get_imsm_map(dev
, MAP_0
);
8815 current_level
= map
->raid_level
;
8818 if ((u
->new_level
== 5) && (u
->new_level
!= current_level
)) {
8819 struct mdinfo
*spares
;
8821 spares
= get_spares_for_grow(st
);
8829 makedev(dev
->disk
.major
,
8831 dl
= get_disk_super(super
,
8834 dl
->index
= u
->old_raid_disks
;
8840 len
= disks_to_mpb_size(u
->new_raid_disks
);
8841 dprintf("New anchor length is %llu\n", (unsigned long long)len
);
8844 case update_size_change
: {
8847 case update_create_array
: {
8848 struct imsm_update_create_array
*u
= (void *) update
->buf
;
8849 struct intel_dev
*dv
;
8850 struct imsm_dev
*dev
= &u
->dev
;
8851 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
8853 struct disk_info
*inf
;
8857 inf
= get_disk_info(u
);
8858 len
= sizeof_imsm_dev(dev
, 1);
8859 /* allocate a new super->devlist entry */
8860 dv
= malloc(sizeof(*dv
));
8862 dv
->dev
= malloc(len
);
8867 update
->space
= NULL
;
8871 /* count how many spares will be converted to members */
8872 for (i
= 0; i
< map
->num_members
; i
++) {
8873 dl
= serial_to_dl(inf
[i
].serial
, super
);
8875 /* hmm maybe it failed?, nothing we can do about
8880 if (count_memberships(dl
, super
) == 0)
8883 len
+= activate
* sizeof(struct imsm_disk
);
8890 /* check if we need a larger metadata buffer */
8891 if (super
->next_buf
)
8892 buf_len
= super
->next_len
;
8894 buf_len
= super
->len
;
8896 if (__le32_to_cpu(mpb
->mpb_size
) + len
> buf_len
) {
8897 /* ok we need a larger buf than what is currently allocated
8898 * if this allocation fails process_update will notice that
8899 * ->next_len is set and ->next_buf is NULL
8901 buf_len
= ROUND_UP(__le32_to_cpu(mpb
->mpb_size
) + len
, 512);
8902 if (super
->next_buf
)
8903 free(super
->next_buf
);
8905 super
->next_len
= buf_len
;
8906 if (posix_memalign(&super
->next_buf
, 512, buf_len
) == 0)
8907 memset(super
->next_buf
, 0, buf_len
);
8909 super
->next_buf
= NULL
;
8913 /* must be called while manager is quiesced */
8914 static void imsm_delete(struct intel_super
*super
, struct dl
**dlp
, unsigned index
)
8916 struct imsm_super
*mpb
= super
->anchor
;
8918 struct imsm_dev
*dev
;
8919 struct imsm_map
*map
;
8920 int i
, j
, num_members
;
8923 dprintf("%s: deleting device[%d] from imsm_super\n",
8926 /* shift all indexes down one */
8927 for (iter
= super
->disks
; iter
; iter
= iter
->next
)
8928 if (iter
->index
> (int)index
)
8930 for (iter
= super
->missing
; iter
; iter
= iter
->next
)
8931 if (iter
->index
> (int)index
)
8934 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
8935 dev
= get_imsm_dev(super
, i
);
8936 map
= get_imsm_map(dev
, MAP_0
);
8937 num_members
= map
->num_members
;
8938 for (j
= 0; j
< num_members
; j
++) {
8939 /* update ord entries being careful not to propagate
8940 * ord-flags to the first map
8942 ord
= get_imsm_ord_tbl_ent(dev
, j
, MAP_X
);
8944 if (ord_to_idx(ord
) <= index
)
8947 map
= get_imsm_map(dev
, MAP_0
);
8948 set_imsm_ord_tbl_ent(map
, j
, ord_to_idx(ord
- 1));
8949 map
= get_imsm_map(dev
, MAP_1
);
8951 set_imsm_ord_tbl_ent(map
, j
, ord
- 1);
8956 super
->updates_pending
++;
8958 struct dl
*dl
= *dlp
;
8960 *dlp
= (*dlp
)->next
;
8961 __free_imsm_disk(dl
);
8964 #endif /* MDASSEMBLE */
8966 static void close_targets(int *targets
, int new_disks
)
8973 for (i
= 0; i
< new_disks
; i
++) {
8974 if (targets
[i
] >= 0) {
8981 static int imsm_get_allowed_degradation(int level
, int raid_disks
,
8982 struct intel_super
*super
,
8983 struct imsm_dev
*dev
)
8989 struct imsm_map
*map
;
8992 ret_val
= raid_disks
/2;
8993 /* check map if all disks pairs not failed
8996 map
= get_imsm_map(dev
, MAP_0
);
8997 for (i
= 0; i
< ret_val
; i
++) {
8998 int degradation
= 0;
8999 if (get_imsm_disk(super
, i
) == NULL
)
9001 if (get_imsm_disk(super
, i
+ 1) == NULL
)
9003 if (degradation
== 2)
9006 map
= get_imsm_map(dev
, MAP_1
);
9007 /* if there is no second map
9008 * result can be returned
9012 /* check degradation in second map
9014 for (i
= 0; i
< ret_val
; i
++) {
9015 int degradation
= 0;
9016 if (get_imsm_disk(super
, i
) == NULL
)
9018 if (get_imsm_disk(super
, i
+ 1) == NULL
)
9020 if (degradation
== 2)
9035 /*******************************************************************************
9036 * Function: open_backup_targets
9037 * Description: Function opens file descriptors for all devices given in
9040 * info : general array info
9041 * raid_disks : number of disks
9042 * raid_fds : table of device's file descriptors
9043 * super : intel super for raid10 degradation check
9044 * dev : intel device for raid10 degradation check
9048 ******************************************************************************/
9049 int open_backup_targets(struct mdinfo
*info
, int raid_disks
, int *raid_fds
,
9050 struct intel_super
*super
, struct imsm_dev
*dev
)
9056 for (i
= 0; i
< raid_disks
; i
++)
9059 for (sd
= info
->devs
; sd
; sd
= sd
->next
) {
9062 if (sd
->disk
.state
& (1<<MD_DISK_FAULTY
)) {
9063 dprintf("disk is faulty!!\n");
9067 if ((sd
->disk
.raid_disk
>= raid_disks
) ||
9068 (sd
->disk
.raid_disk
< 0))
9071 dn
= map_dev(sd
->disk
.major
,
9073 raid_fds
[sd
->disk
.raid_disk
] = dev_open(dn
, O_RDWR
);
9074 if (raid_fds
[sd
->disk
.raid_disk
] < 0) {
9075 fprintf(stderr
, "cannot open component\n");
9080 /* check if maximum array degradation level is not exceeded
9082 if ((raid_disks
- opened
) >
9083 imsm_get_allowed_degradation(info
->new_level
,
9086 fprintf(stderr
, "Not enough disks can be opened.\n");
9087 close_targets(raid_fds
, raid_disks
);
9094 /*******************************************************************************
9095 * Function: init_migr_record_imsm
9096 * Description: Function inits imsm migration record
9098 * super : imsm internal array info
9099 * dev : device under migration
9100 * info : general array info to find the smallest device
9103 ******************************************************************************/
9104 void init_migr_record_imsm(struct supertype
*st
, struct imsm_dev
*dev
,
9105 struct mdinfo
*info
)
9107 struct intel_super
*super
= st
->sb
;
9108 struct migr_record
*migr_rec
= super
->migr_rec
;
9110 unsigned long long dsize
, dev_sectors
;
9111 long long unsigned min_dev_sectors
= -1LLU;
9115 struct imsm_map
*map_dest
= get_imsm_map(dev
, MAP_0
);
9116 struct imsm_map
*map_src
= get_imsm_map(dev
, MAP_1
);
9117 unsigned long long num_migr_units
;
9118 unsigned long long array_blocks
;
9120 memset(migr_rec
, 0, sizeof(struct migr_record
));
9121 migr_rec
->family_num
= __cpu_to_le32(super
->anchor
->family_num
);
9123 /* only ascending reshape supported now */
9124 migr_rec
->ascending_migr
= __cpu_to_le32(1);
9126 migr_rec
->dest_depth_per_unit
= GEN_MIGR_AREA_SIZE
/
9127 max(map_dest
->blocks_per_strip
, map_src
->blocks_per_strip
);
9128 migr_rec
->dest_depth_per_unit
*=
9129 max(map_dest
->blocks_per_strip
, map_src
->blocks_per_strip
);
9130 new_data_disks
= imsm_num_data_members(dev
, MAP_0
);
9131 migr_rec
->blocks_per_unit
=
9132 __cpu_to_le32(migr_rec
->dest_depth_per_unit
* new_data_disks
);
9133 migr_rec
->dest_depth_per_unit
=
9134 __cpu_to_le32(migr_rec
->dest_depth_per_unit
);
9135 array_blocks
= info
->component_size
* new_data_disks
;
9137 array_blocks
/ __le32_to_cpu(migr_rec
->blocks_per_unit
);
9139 if (array_blocks
% __le32_to_cpu(migr_rec
->blocks_per_unit
))
9141 migr_rec
->num_migr_units
= __cpu_to_le32(num_migr_units
);
9143 migr_rec
->post_migr_vol_cap
= dev
->size_low
;
9144 migr_rec
->post_migr_vol_cap_hi
= dev
->size_high
;
9147 /* Find the smallest dev */
9148 for (sd
= info
->devs
; sd
; sd
= sd
->next
) {
9149 sprintf(nm
, "%d:%d", sd
->disk
.major
, sd
->disk
.minor
);
9150 fd
= dev_open(nm
, O_RDONLY
);
9153 get_dev_size(fd
, NULL
, &dsize
);
9154 dev_sectors
= dsize
/ 512;
9155 if (dev_sectors
< min_dev_sectors
)
9156 min_dev_sectors
= dev_sectors
;
9159 migr_rec
->ckpt_area_pba
= __cpu_to_le32(min_dev_sectors
-
9160 RAID_DISK_RESERVED_BLOCKS_IMSM_HI
);
9162 write_imsm_migr_rec(st
);
9167 /*******************************************************************************
9168 * Function: save_backup_imsm
9169 * Description: Function saves critical data stripes to Migration Copy Area
9170 * and updates the current migration unit status.
9171 * Use restore_stripes() to form a destination stripe,
9172 * and to write it to the Copy Area.
9174 * st : supertype information
9175 * dev : imsm device that backup is saved for
9176 * info : general array info
9177 * buf : input buffer
9178 * length : length of data to backup (blocks_per_unit)
9182 ******************************************************************************/
9183 int save_backup_imsm(struct supertype
*st
,
9184 struct imsm_dev
*dev
,
9185 struct mdinfo
*info
,
9190 struct intel_super
*super
= st
->sb
;
9191 unsigned long long *target_offsets
= NULL
;
9192 int *targets
= NULL
;
9194 struct imsm_map
*map_dest
= get_imsm_map(dev
, MAP_0
);
9195 int new_disks
= map_dest
->num_members
;
9196 int dest_layout
= 0;
9198 unsigned long long start
;
9199 int data_disks
= imsm_num_data_members(dev
, MAP_0
);
9201 targets
= malloc(new_disks
* sizeof(int));
9205 for (i
= 0; i
< new_disks
; i
++)
9208 target_offsets
= malloc(new_disks
* sizeof(unsigned long long));
9209 if (!target_offsets
)
9212 start
= info
->reshape_progress
* 512;
9213 for (i
= 0; i
< new_disks
; i
++) {
9214 target_offsets
[i
] = (unsigned long long)
9215 __le32_to_cpu(super
->migr_rec
->ckpt_area_pba
) * 512;
9216 /* move back copy area adderss, it will be moved forward
9217 * in restore_stripes() using start input variable
9219 target_offsets
[i
] -= start
/data_disks
;
9222 if (open_backup_targets(info
, new_disks
, targets
,
9226 dest_layout
= imsm_level_to_layout(map_dest
->raid_level
);
9227 dest_chunk
= __le16_to_cpu(map_dest
->blocks_per_strip
) * 512;
9229 if (restore_stripes(targets
, /* list of dest devices */
9230 target_offsets
, /* migration record offsets */
9233 map_dest
->raid_level
,
9235 -1, /* source backup file descriptor */
9236 0, /* input buf offset
9237 * always 0 buf is already offseted */
9241 fprintf(stderr
, Name
": Error restoring stripes\n");
9249 close_targets(targets
, new_disks
);
9252 free(target_offsets
);
9257 /*******************************************************************************
9258 * Function: save_checkpoint_imsm
9259 * Description: Function called for current unit status update
9260 * in the migration record. It writes it to disk.
9262 * super : imsm internal array info
9263 * info : general array info
9267 * 2: failure, means no valid migration record
9268 * / no general migration in progress /
9269 ******************************************************************************/
9270 int save_checkpoint_imsm(struct supertype
*st
, struct mdinfo
*info
, int state
)
9272 struct intel_super
*super
= st
->sb
;
9273 unsigned long long blocks_per_unit
;
9274 unsigned long long curr_migr_unit
;
9276 if (load_imsm_migr_rec(super
, info
) != 0) {
9277 dprintf("imsm: ERROR: Cannot read migration record "
9278 "for checkpoint save.\n");
9282 blocks_per_unit
= __le32_to_cpu(super
->migr_rec
->blocks_per_unit
);
9283 if (blocks_per_unit
== 0) {
9284 dprintf("imsm: no migration in progress.\n");
9287 curr_migr_unit
= info
->reshape_progress
/ blocks_per_unit
;
9288 /* check if array is alligned to copy area
9289 * if it is not alligned, add one to current migration unit value
9290 * this can happend on array reshape finish only
9292 if (info
->reshape_progress
% blocks_per_unit
)
9295 super
->migr_rec
->curr_migr_unit
=
9296 __cpu_to_le32(curr_migr_unit
);
9297 super
->migr_rec
->rec_status
= __cpu_to_le32(state
);
9298 super
->migr_rec
->dest_1st_member_lba
=
9299 __cpu_to_le32(curr_migr_unit
*
9300 __le32_to_cpu(super
->migr_rec
->dest_depth_per_unit
));
9301 if (write_imsm_migr_rec(st
) < 0) {
9302 dprintf("imsm: Cannot write migration record "
9303 "outside backup area\n");
9310 /*******************************************************************************
9311 * Function: recover_backup_imsm
9312 * Description: Function recovers critical data from the Migration Copy Area
9313 * while assembling an array.
9315 * super : imsm internal array info
9316 * info : general array info
9318 * 0 : success (or there is no data to recover)
9320 ******************************************************************************/
9321 int recover_backup_imsm(struct supertype
*st
, struct mdinfo
*info
)
9323 struct intel_super
*super
= st
->sb
;
9324 struct migr_record
*migr_rec
= super
->migr_rec
;
9325 struct imsm_map
*map_dest
= NULL
;
9326 struct intel_dev
*id
= NULL
;
9327 unsigned long long read_offset
;
9328 unsigned long long write_offset
;
9330 int *targets
= NULL
;
9331 int new_disks
, i
, err
;
9334 unsigned long curr_migr_unit
= __le32_to_cpu(migr_rec
->curr_migr_unit
);
9335 unsigned long num_migr_units
= __le32_to_cpu(migr_rec
->num_migr_units
);
9337 int skipped_disks
= 0;
9339 err
= sysfs_get_str(info
, NULL
, "array_state", (char *)buffer
, 20);
9343 /* recover data only during assemblation */
9344 if (strncmp(buffer
, "inactive", 8) != 0)
9346 /* no data to recover */
9347 if (__le32_to_cpu(migr_rec
->rec_status
) == UNIT_SRC_NORMAL
)
9349 if (curr_migr_unit
>= num_migr_units
)
9352 /* find device during reshape */
9353 for (id
= super
->devlist
; id
; id
= id
->next
)
9354 if (is_gen_migration(id
->dev
))
9359 map_dest
= get_imsm_map(id
->dev
, MAP_0
);
9360 new_disks
= map_dest
->num_members
;
9362 read_offset
= (unsigned long long)
9363 __le32_to_cpu(migr_rec
->ckpt_area_pba
) * 512;
9365 write_offset
= ((unsigned long long)
9366 __le32_to_cpu(migr_rec
->dest_1st_member_lba
) +
9367 pba_of_lba0(map_dest
)) * 512;
9369 unit_len
= __le32_to_cpu(migr_rec
->dest_depth_per_unit
) * 512;
9370 if (posix_memalign((void **)&buf
, 512, unit_len
) != 0)
9372 targets
= malloc(new_disks
* sizeof(int));
9376 if (open_backup_targets(info
, new_disks
, targets
, super
, id
->dev
)) {
9378 Name
": Cannot open some devices belonging to array.\n");
9382 for (i
= 0; i
< new_disks
; i
++) {
9383 if (targets
[i
] < 0) {
9387 if (lseek64(targets
[i
], read_offset
, SEEK_SET
) < 0) {
9389 Name
": Cannot seek to block: %s\n",
9394 if ((unsigned)read(targets
[i
], buf
, unit_len
) != unit_len
) {
9396 Name
": Cannot read copy area block: %s\n",
9401 if (lseek64(targets
[i
], write_offset
, SEEK_SET
) < 0) {
9403 Name
": Cannot seek to block: %s\n",
9408 if ((unsigned)write(targets
[i
], buf
, unit_len
) != unit_len
) {
9410 Name
": Cannot restore block: %s\n",
9417 if (skipped_disks
> imsm_get_allowed_degradation(info
->new_level
,
9422 Name
": Cannot restore data from backup."
9423 " Too many failed disks\n");
9427 if (save_checkpoint_imsm(st
, info
, UNIT_SRC_NORMAL
)) {
9428 /* ignore error == 2, this can mean end of reshape here
9430 dprintf("imsm: Cannot write checkpoint to "
9431 "migration record (UNIT_SRC_NORMAL) during restart\n");
9437 for (i
= 0; i
< new_disks
; i
++)
9446 static char disk_by_path
[] = "/dev/disk/by-path/";
9448 static const char *imsm_get_disk_controller_domain(const char *path
)
9450 char disk_path
[PATH_MAX
];
9454 strncpy(disk_path
, disk_by_path
, PATH_MAX
- 1);
9455 strncat(disk_path
, path
, PATH_MAX
- strlen(disk_path
) - 1);
9456 if (stat(disk_path
, &st
) == 0) {
9457 struct sys_dev
* hba
;
9460 path
= devt_to_devpath(st
.st_rdev
);
9463 hba
= find_disk_attached_hba(-1, path
);
9464 if (hba
&& hba
->type
== SYS_DEV_SAS
)
9466 else if (hba
&& hba
->type
== SYS_DEV_SATA
)
9470 dprintf("path: %s hba: %s attached: %s\n",
9471 path
, (hba
) ? hba
->path
: "NULL", drv
);
9479 static int imsm_find_array_minor_by_subdev(int subdev
, int container
, int *minor
)
9481 char subdev_name
[20];
9482 struct mdstat_ent
*mdstat
;
9484 sprintf(subdev_name
, "%d", subdev
);
9485 mdstat
= mdstat_by_subdev(subdev_name
, container
);
9489 *minor
= mdstat
->devnum
;
9490 free_mdstat(mdstat
);
9494 static int imsm_reshape_is_allowed_on_container(struct supertype
*st
,
9495 struct geo_params
*geo
,
9496 int *old_raid_disks
,
9499 /* currently we only support increasing the number of devices
9500 * for a container. This increases the number of device for each
9501 * member array. They must all be RAID0 or RAID5.
9504 struct mdinfo
*info
, *member
;
9505 int devices_that_can_grow
= 0;
9507 dprintf("imsm: imsm_reshape_is_allowed_on_container(ENTER): "
9508 "st->devnum = (%i)\n",
9511 if (geo
->size
!= -1 ||
9512 geo
->level
!= UnSet
||
9513 geo
->layout
!= UnSet
||
9514 geo
->chunksize
!= 0 ||
9515 geo
->raid_disks
== UnSet
) {
9516 dprintf("imsm: Container operation is allowed for "
9517 "raid disks number change only.\n");
9521 if (direction
== ROLLBACK_METADATA_CHANGES
) {
9522 dprintf("imsm: Metadata changes rollback is not supported for "
9523 "container operation.\n");
9527 info
= container_content_imsm(st
, NULL
);
9528 for (member
= info
; member
; member
= member
->next
) {
9532 dprintf("imsm: checking device_num: %i\n",
9533 member
->container_member
);
9535 if (geo
->raid_disks
<= member
->array
.raid_disks
) {
9536 /* we work on container for Online Capacity Expansion
9537 * only so raid_disks has to grow
9539 dprintf("imsm: for container operation raid disks "
9540 "increase is required\n");
9544 if ((info
->array
.level
!= 0) &&
9545 (info
->array
.level
!= 5)) {
9546 /* we cannot use this container with other raid level
9548 dprintf("imsm: for container operation wrong"
9549 " raid level (%i) detected\n",
9553 /* check for platform support
9554 * for this raid level configuration
9556 struct intel_super
*super
= st
->sb
;
9557 if (!is_raid_level_supported(super
->orom
,
9558 member
->array
.level
,
9560 dprintf("platform does not support raid%d with"
9564 geo
->raid_disks
> 1 ? "s" : "");
9567 /* check if component size is aligned to chunk size
9569 if (info
->component_size
%
9570 (info
->array
.chunk_size
/512)) {
9571 dprintf("Component size is not aligned to "
9577 if (*old_raid_disks
&&
9578 info
->array
.raid_disks
!= *old_raid_disks
)
9580 *old_raid_disks
= info
->array
.raid_disks
;
9582 /* All raid5 and raid0 volumes in container
9583 * have to be ready for Online Capacity Expansion
9584 * so they need to be assembled. We have already
9585 * checked that no recovery etc is happening.
9587 result
= imsm_find_array_minor_by_subdev(member
->container_member
,
9591 dprintf("imsm: cannot find array\n");
9594 devices_that_can_grow
++;
9597 if (!member
&& devices_that_can_grow
)
9601 dprintf("\tContainer operation allowed\n");
9603 dprintf("\tError: %i\n", ret_val
);
9608 /* Function: get_spares_for_grow
9609 * Description: Allocates memory and creates list of spare devices
9610 * avaliable in container. Checks if spare drive size is acceptable.
9611 * Parameters: Pointer to the supertype structure
9612 * Returns: Pointer to the list of spare devices (mdinfo structure) on success,
9615 static struct mdinfo
*get_spares_for_grow(struct supertype
*st
)
9617 unsigned long long min_size
= min_acceptable_spare_size_imsm(st
);
9618 return container_choose_spares(st
, min_size
, NULL
, NULL
, NULL
, 0);
9621 /******************************************************************************
9622 * function: imsm_create_metadata_update_for_reshape
9623 * Function creates update for whole IMSM container.
9625 ******************************************************************************/
9626 static int imsm_create_metadata_update_for_reshape(
9627 struct supertype
*st
,
9628 struct geo_params
*geo
,
9630 struct imsm_update_reshape
**updatep
)
9632 struct intel_super
*super
= st
->sb
;
9633 struct imsm_super
*mpb
= super
->anchor
;
9634 int update_memory_size
= 0;
9635 struct imsm_update_reshape
*u
= NULL
;
9636 struct mdinfo
*spares
= NULL
;
9638 int delta_disks
= 0;
9641 dprintf("imsm_update_metadata_for_reshape(enter) raid_disks = %i\n",
9644 delta_disks
= geo
->raid_disks
- old_raid_disks
;
9646 /* size of all update data without anchor */
9647 update_memory_size
= sizeof(struct imsm_update_reshape
);
9649 /* now add space for spare disks that we need to add. */
9650 update_memory_size
+= sizeof(u
->new_disks
[0]) * (delta_disks
- 1);
9652 u
= calloc(1, update_memory_size
);
9655 "cannot get memory for imsm_update_reshape update\n");
9658 u
->type
= update_reshape_container_disks
;
9659 u
->old_raid_disks
= old_raid_disks
;
9660 u
->new_raid_disks
= geo
->raid_disks
;
9662 /* now get spare disks list
9664 spares
= get_spares_for_grow(st
);
9667 || delta_disks
> spares
->array
.spare_disks
) {
9668 fprintf(stderr
, Name
": imsm: ERROR: Cannot get spare devices "
9669 "for %s.\n", geo
->dev_name
);
9674 /* we have got spares
9675 * update disk list in imsm_disk list table in anchor
9677 dprintf("imsm: %i spares are available.\n\n",
9678 spares
->array
.spare_disks
);
9681 for (i
= 0; i
< delta_disks
; i
++) {
9686 u
->new_disks
[i
] = makedev(dev
->disk
.major
,
9688 dl
= get_disk_super(super
, dev
->disk
.major
, dev
->disk
.minor
);
9689 dl
->index
= mpb
->num_disks
;
9699 dprintf("imsm: reshape update preparation :");
9700 if (i
== delta_disks
) {
9703 return update_memory_size
;
9706 dprintf(" Error\n");
9712 /******************************************************************************
9713 * function: imsm_create_metadata_update_for_size_change()
9714 * Creates update for IMSM array for array size change.
9716 ******************************************************************************/
9717 static int imsm_create_metadata_update_for_size_change(
9718 struct supertype
*st
,
9719 struct geo_params
*geo
,
9720 struct imsm_update_size_change
**updatep
)
9722 struct intel_super
*super
= st
->sb
;
9723 int update_memory_size
= 0;
9724 struct imsm_update_size_change
*u
= NULL
;
9726 dprintf("imsm_create_metadata_update_for_size_change(enter)"
9727 " New size = %llu\n", geo
->size
);
9729 /* size of all update data without anchor */
9730 update_memory_size
= sizeof(struct imsm_update_size_change
);
9732 u
= calloc(1, update_memory_size
);
9734 dprintf("error: cannot get memory for "
9735 "imsm_create_metadata_update_for_size_change\n");
9738 u
->type
= update_size_change
;
9739 u
->subdev
= super
->current_vol
;
9740 u
->new_size
= geo
->size
;
9742 dprintf("imsm: reshape update preparation : OK\n");
9745 return update_memory_size
;
9748 /******************************************************************************
9749 * function: imsm_create_metadata_update_for_migration()
9750 * Creates update for IMSM array.
9752 ******************************************************************************/
9753 static int imsm_create_metadata_update_for_migration(
9754 struct supertype
*st
,
9755 struct geo_params
*geo
,
9756 struct imsm_update_reshape_migration
**updatep
)
9758 struct intel_super
*super
= st
->sb
;
9759 int update_memory_size
= 0;
9760 struct imsm_update_reshape_migration
*u
= NULL
;
9761 struct imsm_dev
*dev
;
9762 int previous_level
= -1;
9764 dprintf("imsm_create_metadata_update_for_migration(enter)"
9765 " New Level = %i\n", geo
->level
);
9767 /* size of all update data without anchor */
9768 update_memory_size
= sizeof(struct imsm_update_reshape_migration
);
9770 u
= calloc(1, update_memory_size
);
9772 dprintf("error: cannot get memory for "
9773 "imsm_create_metadata_update_for_migration\n");
9776 u
->type
= update_reshape_migration
;
9777 u
->subdev
= super
->current_vol
;
9778 u
->new_level
= geo
->level
;
9779 u
->new_layout
= geo
->layout
;
9780 u
->new_raid_disks
= u
->old_raid_disks
= geo
->raid_disks
;
9781 u
->new_disks
[0] = -1;
9782 u
->new_chunksize
= -1;
9784 dev
= get_imsm_dev(super
, u
->subdev
);
9786 struct imsm_map
*map
;
9788 map
= get_imsm_map(dev
, MAP_0
);
9790 int current_chunk_size
=
9791 __le16_to_cpu(map
->blocks_per_strip
) / 2;
9793 if (geo
->chunksize
!= current_chunk_size
) {
9794 u
->new_chunksize
= geo
->chunksize
/ 1024;
9796 "chunk size change from %i to %i\n",
9797 current_chunk_size
, u
->new_chunksize
);
9799 previous_level
= map
->raid_level
;
9802 if ((geo
->level
== 5) && (previous_level
== 0)) {
9803 struct mdinfo
*spares
= NULL
;
9805 u
->new_raid_disks
++;
9806 spares
= get_spares_for_grow(st
);
9807 if ((spares
== NULL
) || (spares
->array
.spare_disks
< 1)) {
9810 update_memory_size
= 0;
9811 dprintf("error: cannot get spare device "
9812 "for requested migration");
9817 dprintf("imsm: reshape update preparation : OK\n");
9820 return update_memory_size
;
9823 static void imsm_update_metadata_locally(struct supertype
*st
,
9826 struct metadata_update mu
;
9831 mu
.space_list
= NULL
;
9833 imsm_prepare_update(st
, &mu
);
9834 imsm_process_update(st
, &mu
);
9836 while (mu
.space_list
) {
9837 void **space
= mu
.space_list
;
9838 mu
.space_list
= *space
;
9843 /***************************************************************************
9844 * Function: imsm_analyze_change
9845 * Description: Function analyze change for single volume
9846 * and validate if transition is supported
9847 * Parameters: Geometry parameters, supertype structure,
9848 * metadata change direction (apply/rollback)
9849 * Returns: Operation type code on success, -1 if fail
9850 ****************************************************************************/
9851 enum imsm_reshape_type
imsm_analyze_change(struct supertype
*st
,
9852 struct geo_params
*geo
,
9859 /* number of added/removed disks in operation result */
9860 int devNumChange
= 0;
9861 /* imsm compatible layout value for array geometry verification */
9862 int imsm_layout
= -1;
9864 struct imsm_dev
*dev
;
9865 struct intel_super
*super
;
9866 long long current_size
;
9868 getinfo_super_imsm_volume(st
, &info
, NULL
);
9869 if ((geo
->level
!= info
.array
.level
) &&
9870 (geo
->level
>= 0) &&
9871 (geo
->level
!= UnSet
)) {
9872 switch (info
.array
.level
) {
9874 if (geo
->level
== 5) {
9875 change
= CH_MIGRATION
;
9876 if (geo
->layout
!= ALGORITHM_LEFT_ASYMMETRIC
) {
9878 Name
" Error. Requested Layout "
9879 "not supported (left-asymmetric layout "
9880 "is supported only)!\n");
9882 goto analyse_change_exit
;
9884 imsm_layout
= geo
->layout
;
9886 devNumChange
= 1; /* parity disk added */
9887 } else if (geo
->level
== 10) {
9888 change
= CH_TAKEOVER
;
9890 devNumChange
= 2; /* two mirrors added */
9891 imsm_layout
= 0x102; /* imsm supported layout */
9896 if (geo
->level
== 0) {
9897 change
= CH_TAKEOVER
;
9899 devNumChange
= -(geo
->raid_disks
/2);
9900 imsm_layout
= 0; /* imsm raid0 layout */
9906 Name
" Error. Level Migration from %d to %d "
9908 info
.array
.level
, geo
->level
);
9909 goto analyse_change_exit
;
9912 geo
->level
= info
.array
.level
;
9914 if ((geo
->layout
!= info
.array
.layout
)
9915 && ((geo
->layout
!= UnSet
) && (geo
->layout
!= -1))) {
9916 change
= CH_MIGRATION
;
9917 if ((info
.array
.layout
== 0)
9918 && (info
.array
.level
== 5)
9919 && (geo
->layout
== 5)) {
9920 /* reshape 5 -> 4 */
9921 } else if ((info
.array
.layout
== 5)
9922 && (info
.array
.level
== 5)
9923 && (geo
->layout
== 0)) {
9924 /* reshape 4 -> 5 */
9929 Name
" Error. Layout Migration from %d to %d "
9931 info
.array
.layout
, geo
->layout
);
9933 goto analyse_change_exit
;
9936 geo
->layout
= info
.array
.layout
;
9937 if (imsm_layout
== -1)
9938 imsm_layout
= info
.array
.layout
;
9941 if ((geo
->chunksize
> 0) && (geo
->chunksize
!= UnSet
)
9942 && (geo
->chunksize
!= info
.array
.chunk_size
))
9943 change
= CH_MIGRATION
;
9945 geo
->chunksize
= info
.array
.chunk_size
;
9947 chunk
= geo
->chunksize
/ 1024;
9950 dev
= get_imsm_dev(super
, super
->current_vol
);
9951 data_disks
= imsm_num_data_members(dev
, MAP_0
);
9952 /* compute current size in K per disk member
9954 current_size
= info
.custom_array_size
/ 2 / data_disks
;
9956 if ((current_size
!= geo
->size
) && (geo
->size
> 0)) {
9959 Name
" Error. Size change should be the only "
9960 "one at a time.\n");
9962 goto analyse_change_exit
;
9964 if ((super
->current_vol
+ 1) != super
->anchor
->num_raid_devs
) {
9966 Name
" Error. The last volume in container "
9967 "can be expanded only (%i/%i).\n",
9968 super
->current_vol
, st
->devnum
);
9969 goto analyse_change_exit
;
9972 if ((direction
== ROLLBACK_METADATA_CHANGES
)) {
9973 /* accept size for rollback only
9976 /* round size due to metadata compatibility
9978 geo
->size
= (geo
->size
>> SECT_PER_MB_SHIFT
)
9979 << SECT_PER_MB_SHIFT
;
9980 dprintf("Prepare update for size change to %llu\n",
9982 if (current_size
>= geo
->size
) {
9984 Name
" Error. Size expanssion is "
9985 "supported only (current size is %llu, "
9986 "requested size /rounded/ is %llu).\n",
9987 current_size
, geo
->size
);
9988 goto analyse_change_exit
;
9991 geo
->size
*= data_disks
;
9992 geo
->raid_disks
= dev
->vol
.map
->num_members
;
9993 change
= CH_ARRAY_SIZE
;
9995 if (!validate_geometry_imsm(st
,
9998 geo
->raid_disks
+ devNumChange
,
10005 struct intel_super
*super
= st
->sb
;
10006 struct imsm_super
*mpb
= super
->anchor
;
10008 if (mpb
->num_raid_devs
> 1) {
10010 Name
" Error. Cannot perform operation on %s"
10011 "- for this operation it MUST be single "
10012 "array in container\n",
10018 analyse_change_exit
:
10019 if ((direction
== ROLLBACK_METADATA_CHANGES
) &&
10020 ((change
== CH_MIGRATION
) || (change
== CH_TAKEOVER
))) {
10021 dprintf("imsm: Metadata changes rollback is not supported for "
10022 "migration and takeover operations.\n");
10028 int imsm_takeover(struct supertype
*st
, struct geo_params
*geo
)
10030 struct intel_super
*super
= st
->sb
;
10031 struct imsm_update_takeover
*u
;
10033 u
= malloc(sizeof(struct imsm_update_takeover
));
10037 u
->type
= update_takeover
;
10038 u
->subarray
= super
->current_vol
;
10040 /* 10->0 transition */
10041 if (geo
->level
== 0)
10042 u
->direction
= R10_TO_R0
;
10044 /* 0->10 transition */
10045 if (geo
->level
== 10)
10046 u
->direction
= R0_TO_R10
;
10048 /* update metadata locally */
10049 imsm_update_metadata_locally(st
, u
,
10050 sizeof(struct imsm_update_takeover
));
10051 /* and possibly remotely */
10052 if (st
->update_tail
)
10053 append_metadata_update(st
, u
,
10054 sizeof(struct imsm_update_takeover
));
10061 static int imsm_reshape_super(struct supertype
*st
, long long size
, int level
,
10062 int layout
, int chunksize
, int raid_disks
,
10063 int delta_disks
, char *backup
, char *dev
,
10064 int direction
, int verbose
)
10067 struct geo_params geo
;
10069 dprintf("imsm: reshape_super called.\n");
10071 memset(&geo
, 0, sizeof(struct geo_params
));
10073 geo
.dev_name
= dev
;
10074 geo
.dev_id
= st
->devnum
;
10077 geo
.layout
= layout
;
10078 geo
.chunksize
= chunksize
;
10079 geo
.raid_disks
= raid_disks
;
10080 if (delta_disks
!= UnSet
)
10081 geo
.raid_disks
+= delta_disks
;
10083 dprintf("\tfor level : %i\n", geo
.level
);
10084 dprintf("\tfor raid_disks : %i\n", geo
.raid_disks
);
10086 if (experimental() == 0)
10089 if (st
->container_dev
== st
->devnum
) {
10090 /* On container level we can only increase number of devices. */
10091 dprintf("imsm: info: Container operation\n");
10092 int old_raid_disks
= 0;
10094 if (imsm_reshape_is_allowed_on_container(
10095 st
, &geo
, &old_raid_disks
, direction
)) {
10096 struct imsm_update_reshape
*u
= NULL
;
10099 len
= imsm_create_metadata_update_for_reshape(
10100 st
, &geo
, old_raid_disks
, &u
);
10103 dprintf("imsm: Cannot prepare update\n");
10104 goto exit_imsm_reshape_super
;
10108 /* update metadata locally */
10109 imsm_update_metadata_locally(st
, u
, len
);
10110 /* and possibly remotely */
10111 if (st
->update_tail
)
10112 append_metadata_update(st
, u
, len
);
10117 fprintf(stderr
, Name
": (imsm) Operation "
10118 "is not allowed on this container\n");
10121 /* On volume level we support following operations
10122 * - takeover: raid10 -> raid0; raid0 -> raid10
10123 * - chunk size migration
10124 * - migration: raid5 -> raid0; raid0 -> raid5
10126 struct intel_super
*super
= st
->sb
;
10127 struct intel_dev
*dev
= super
->devlist
;
10128 int change
, devnum
;
10129 dprintf("imsm: info: Volume operation\n");
10130 /* find requested device */
10132 if (imsm_find_array_minor_by_subdev(
10133 dev
->index
, st
->container_dev
, &devnum
) == 0
10134 && devnum
== geo
.dev_id
)
10139 fprintf(stderr
, Name
" Cannot find %s (%i) subarray\n",
10140 geo
.dev_name
, geo
.dev_id
);
10141 goto exit_imsm_reshape_super
;
10143 super
->current_vol
= dev
->index
;
10144 change
= imsm_analyze_change(st
, &geo
, direction
);
10147 ret_val
= imsm_takeover(st
, &geo
);
10149 case CH_MIGRATION
: {
10150 struct imsm_update_reshape_migration
*u
= NULL
;
10152 imsm_create_metadata_update_for_migration(
10156 "Cannot prepare update\n");
10160 /* update metadata locally */
10161 imsm_update_metadata_locally(st
, u
, len
);
10162 /* and possibly remotely */
10163 if (st
->update_tail
)
10164 append_metadata_update(st
, u
, len
);
10169 case CH_ARRAY_SIZE
: {
10170 struct imsm_update_size_change
*u
= NULL
;
10172 imsm_create_metadata_update_for_size_change(
10176 "Cannot prepare update\n");
10180 /* update metadata locally */
10181 imsm_update_metadata_locally(st
, u
, len
);
10182 /* and possibly remotely */
10183 if (st
->update_tail
)
10184 append_metadata_update(st
, u
, len
);
10194 exit_imsm_reshape_super
:
10195 dprintf("imsm: reshape_super Exit code = %i\n", ret_val
);
10199 /*******************************************************************************
10200 * Function: wait_for_reshape_imsm
10201 * Description: Function writes new sync_max value and waits until
10202 * reshape process reach new position
10204 * sra : general array info
10205 * ndata : number of disks in new array's layout
10208 * 1 : there is no reshape in progress,
10210 ******************************************************************************/
10211 int wait_for_reshape_imsm(struct mdinfo
*sra
, int ndata
)
10213 int fd
= sysfs_get_fd(sra
, NULL
, "reshape_position");
10214 unsigned long long completed
;
10215 /* to_complete : new sync_max position */
10216 unsigned long long to_complete
= sra
->reshape_progress
;
10217 unsigned long long position_to_set
= to_complete
/ ndata
;
10220 dprintf("imsm: wait_for_reshape_imsm() "
10221 "cannot open reshape_position\n");
10225 if (sysfs_fd_get_ll(fd
, &completed
) < 0) {
10226 dprintf("imsm: wait_for_reshape_imsm() "
10227 "cannot read reshape_position (no reshape in progres)\n");
10232 if (completed
> to_complete
) {
10233 dprintf("imsm: wait_for_reshape_imsm() "
10234 "wrong next position to set %llu (%llu)\n",
10235 to_complete
, completed
);
10239 dprintf("Position set: %llu\n", position_to_set
);
10240 if (sysfs_set_num(sra
, NULL
, "sync_max",
10241 position_to_set
) != 0) {
10242 dprintf("imsm: wait_for_reshape_imsm() "
10243 "cannot set reshape position to %llu\n",
10254 select(fd
+1, &rfds
, NULL
, NULL
, NULL
);
10255 if (sysfs_get_str(sra
, NULL
, "sync_action",
10257 strncmp(action
, "reshape", 7) != 0)
10259 if (sysfs_fd_get_ll(fd
, &completed
) < 0) {
10260 dprintf("imsm: wait_for_reshape_imsm() "
10261 "cannot read reshape_position (in loop)\n");
10265 } while (completed
< to_complete
);
10271 /*******************************************************************************
10272 * Function: check_degradation_change
10273 * Description: Check that array hasn't become failed.
10275 * info : for sysfs access
10276 * sources : source disks descriptors
10277 * degraded: previous degradation level
10279 * degradation level
10280 ******************************************************************************/
10281 int check_degradation_change(struct mdinfo
*info
,
10285 unsigned long long new_degraded
;
10286 sysfs_get_ll(info
, NULL
, "degraded", &new_degraded
);
10287 if (new_degraded
!= (unsigned long long)degraded
) {
10288 /* check each device to ensure it is still working */
10291 for (sd
= info
->devs
; sd
; sd
= sd
->next
) {
10292 if (sd
->disk
.state
& (1<<MD_DISK_FAULTY
))
10294 if (sd
->disk
.state
& (1<<MD_DISK_SYNC
)) {
10296 if (sysfs_get_str(info
,
10297 sd
, "state", sbuf
, 20) < 0 ||
10298 strstr(sbuf
, "faulty") ||
10299 strstr(sbuf
, "in_sync") == NULL
) {
10300 /* this device is dead */
10301 sd
->disk
.state
= (1<<MD_DISK_FAULTY
);
10302 if (sd
->disk
.raid_disk
>= 0 &&
10303 sources
[sd
->disk
.raid_disk
] >= 0) {
10305 sd
->disk
.raid_disk
]);
10306 sources
[sd
->disk
.raid_disk
] =
10315 return new_degraded
;
10318 /*******************************************************************************
10319 * Function: imsm_manage_reshape
10320 * Description: Function finds array under reshape and it manages reshape
10321 * process. It creates stripes backups (if required) and sets
10324 * afd : Backup handle (nattive) - not used
10325 * sra : general array info
10326 * reshape : reshape parameters - not used
10327 * st : supertype structure
10328 * blocks : size of critical section [blocks]
10329 * fds : table of source device descriptor
10330 * offsets : start of array (offest per devices)
10332 * destfd : table of destination device descriptor
10333 * destoffsets : table of destination offsets (per device)
10335 * 1 : success, reshape is done
10337 ******************************************************************************/
10338 static int imsm_manage_reshape(
10339 int afd
, struct mdinfo
*sra
, struct reshape
*reshape
,
10340 struct supertype
*st
, unsigned long backup_blocks
,
10341 int *fds
, unsigned long long *offsets
,
10342 int dests
, int *destfd
, unsigned long long *destoffsets
)
10345 struct intel_super
*super
= st
->sb
;
10346 struct intel_dev
*dv
= NULL
;
10347 struct imsm_dev
*dev
= NULL
;
10348 struct imsm_map
*map_src
;
10349 int migr_vol_qan
= 0;
10350 int ndata
, odata
; /* [bytes] */
10351 int chunk
; /* [bytes] */
10352 struct migr_record
*migr_rec
;
10354 unsigned int buf_size
; /* [bytes] */
10355 unsigned long long max_position
; /* array size [bytes] */
10356 unsigned long long next_step
; /* [blocks]/[bytes] */
10357 unsigned long long old_data_stripe_length
;
10358 unsigned long long start_src
; /* [bytes] */
10359 unsigned long long start
; /* [bytes] */
10360 unsigned long long start_buf_shift
; /* [bytes] */
10362 int source_layout
= 0;
10364 if (!fds
|| !offsets
|| !sra
)
10367 /* Find volume during the reshape */
10368 for (dv
= super
->devlist
; dv
; dv
= dv
->next
) {
10369 if (dv
->dev
->vol
.migr_type
== MIGR_GEN_MIGR
10370 && dv
->dev
->vol
.migr_state
== 1) {
10375 /* Only one volume can migrate at the same time */
10376 if (migr_vol_qan
!= 1) {
10377 fprintf(stderr
, Name
" : %s", migr_vol_qan
?
10378 "Number of migrating volumes greater than 1\n" :
10379 "There is no volume during migrationg\n");
10383 map_src
= get_imsm_map(dev
, MAP_1
);
10384 if (map_src
== NULL
)
10387 ndata
= imsm_num_data_members(dev
, MAP_0
);
10388 odata
= imsm_num_data_members(dev
, MAP_1
);
10390 chunk
= __le16_to_cpu(map_src
->blocks_per_strip
) * 512;
10391 old_data_stripe_length
= odata
* chunk
;
10393 migr_rec
= super
->migr_rec
;
10395 /* initialize migration record for start condition */
10396 if (sra
->reshape_progress
== 0)
10397 init_migr_record_imsm(st
, dev
, sra
);
10399 if (__le32_to_cpu(migr_rec
->rec_status
) != UNIT_SRC_NORMAL
) {
10400 dprintf("imsm: cannot restart migration when data "
10401 "are present in copy area.\n");
10404 /* Save checkpoint to update migration record for current
10405 * reshape position (in md). It can be farther than current
10406 * reshape position in metadata.
10408 if (save_checkpoint_imsm(st
, sra
, UNIT_SRC_NORMAL
) == 1) {
10409 /* ignore error == 2, this can mean end of reshape here
10411 dprintf("imsm: Cannot write checkpoint to "
10412 "migration record (UNIT_SRC_NORMAL, "
10413 "initial save)\n");
10418 /* size for data */
10419 buf_size
= __le32_to_cpu(migr_rec
->blocks_per_unit
) * 512;
10420 /* extend buffer size for parity disk */
10421 buf_size
+= __le32_to_cpu(migr_rec
->dest_depth_per_unit
) * 512;
10422 /* add space for stripe aligment */
10423 buf_size
+= old_data_stripe_length
;
10424 if (posix_memalign((void **)&buf
, 4096, buf_size
)) {
10425 dprintf("imsm: Cannot allocate checpoint buffer\n");
10429 max_position
= sra
->component_size
* ndata
;
10430 source_layout
= imsm_level_to_layout(map_src
->raid_level
);
10432 while (__le32_to_cpu(migr_rec
->curr_migr_unit
) <
10433 __le32_to_cpu(migr_rec
->num_migr_units
)) {
10434 /* current reshape position [blocks] */
10435 unsigned long long current_position
=
10436 __le32_to_cpu(migr_rec
->blocks_per_unit
)
10437 * __le32_to_cpu(migr_rec
->curr_migr_unit
);
10438 unsigned long long border
;
10440 /* Check that array hasn't become failed.
10442 degraded
= check_degradation_change(sra
, fds
, degraded
);
10443 if (degraded
> 1) {
10444 dprintf("imsm: Abort reshape due to degradation"
10445 " level (%i)\n", degraded
);
10449 next_step
= __le32_to_cpu(migr_rec
->blocks_per_unit
);
10451 if ((current_position
+ next_step
) > max_position
)
10452 next_step
= max_position
- current_position
;
10454 start
= current_position
* 512;
10456 /* allign reading start to old geometry */
10457 start_buf_shift
= start
% old_data_stripe_length
;
10458 start_src
= start
- start_buf_shift
;
10460 border
= (start_src
/ odata
) - (start
/ ndata
);
10462 if (border
<= __le32_to_cpu(migr_rec
->dest_depth_per_unit
)) {
10463 /* save critical stripes to buf
10464 * start - start address of current unit
10465 * to backup [bytes]
10466 * start_src - start address of current unit
10467 * to backup alligned to source array
10470 unsigned long long next_step_filler
= 0;
10471 unsigned long long copy_length
= next_step
* 512;
10473 /* allign copy area length to stripe in old geometry */
10474 next_step_filler
= ((copy_length
+ start_buf_shift
)
10475 % old_data_stripe_length
);
10476 if (next_step_filler
)
10477 next_step_filler
= (old_data_stripe_length
10478 - next_step_filler
);
10479 dprintf("save_stripes() parameters: start = %llu,"
10480 "\tstart_src = %llu,\tnext_step*512 = %llu,"
10481 "\tstart_in_buf_shift = %llu,"
10482 "\tnext_step_filler = %llu\n",
10483 start
, start_src
, copy_length
,
10484 start_buf_shift
, next_step_filler
);
10486 if (save_stripes(fds
, offsets
, map_src
->num_members
,
10487 chunk
, map_src
->raid_level
,
10488 source_layout
, 0, NULL
, start_src
,
10490 next_step_filler
+ start_buf_shift
,
10492 dprintf("imsm: Cannot save stripes"
10496 /* Convert data to destination format and store it
10497 * in backup general migration area
10499 if (save_backup_imsm(st
, dev
, sra
,
10500 buf
+ start_buf_shift
, copy_length
)) {
10501 dprintf("imsm: Cannot save stripes to "
10502 "target devices\n");
10505 if (save_checkpoint_imsm(st
, sra
,
10506 UNIT_SRC_IN_CP_AREA
)) {
10507 dprintf("imsm: Cannot write checkpoint to "
10508 "migration record (UNIT_SRC_IN_CP_AREA)\n");
10512 /* set next step to use whole border area */
10513 border
/= next_step
;
10515 next_step
*= border
;
10517 /* When data backed up, checkpoint stored,
10518 * kick the kernel to reshape unit of data
10520 next_step
= next_step
+ sra
->reshape_progress
;
10521 /* limit next step to array max position */
10522 if (next_step
> max_position
)
10523 next_step
= max_position
;
10524 sysfs_set_num(sra
, NULL
, "suspend_lo", sra
->reshape_progress
);
10525 sysfs_set_num(sra
, NULL
, "suspend_hi", next_step
);
10526 sra
->reshape_progress
= next_step
;
10528 /* wait until reshape finish */
10529 if (wait_for_reshape_imsm(sra
, ndata
) < 0) {
10530 dprintf("wait_for_reshape_imsm returned error!\n");
10534 if (save_checkpoint_imsm(st
, sra
, UNIT_SRC_NORMAL
) == 1) {
10535 /* ignore error == 2, this can mean end of reshape here
10537 dprintf("imsm: Cannot write checkpoint to "
10538 "migration record (UNIT_SRC_NORMAL)\n");
10544 /* return '1' if done */
10548 abort_reshape(sra
);
10552 #endif /* MDASSEMBLE */
10554 struct superswitch super_imsm
= {
10556 .examine_super
= examine_super_imsm
,
10557 .brief_examine_super
= brief_examine_super_imsm
,
10558 .brief_examine_subarrays
= brief_examine_subarrays_imsm
,
10559 .export_examine_super
= export_examine_super_imsm
,
10560 .detail_super
= detail_super_imsm
,
10561 .brief_detail_super
= brief_detail_super_imsm
,
10562 .write_init_super
= write_init_super_imsm
,
10563 .validate_geometry
= validate_geometry_imsm
,
10564 .add_to_super
= add_to_super_imsm
,
10565 .remove_from_super
= remove_from_super_imsm
,
10566 .detail_platform
= detail_platform_imsm
,
10567 .kill_subarray
= kill_subarray_imsm
,
10568 .update_subarray
= update_subarray_imsm
,
10569 .load_container
= load_container_imsm
,
10570 .default_geometry
= default_geometry_imsm
,
10571 .get_disk_controller_domain
= imsm_get_disk_controller_domain
,
10572 .reshape_super
= imsm_reshape_super
,
10573 .manage_reshape
= imsm_manage_reshape
,
10574 .recover_backup
= recover_backup_imsm
,
10576 .match_home
= match_home_imsm
,
10577 .uuid_from_super
= uuid_from_super_imsm
,
10578 .getinfo_super
= getinfo_super_imsm
,
10579 .getinfo_super_disks
= getinfo_super_disks_imsm
,
10580 .update_super
= update_super_imsm
,
10582 .avail_size
= avail_size_imsm
,
10583 .min_acceptable_spare_size
= min_acceptable_spare_size_imsm
,
10585 .compare_super
= compare_super_imsm
,
10587 .load_super
= load_super_imsm
,
10588 .init_super
= init_super_imsm
,
10589 .store_super
= store_super_imsm
,
10590 .free_super
= free_super_imsm
,
10591 .match_metadata_desc
= match_metadata_desc_imsm
,
10592 .container_content
= container_content_imsm
,
10600 .open_new
= imsm_open_new
,
10601 .set_array_state
= imsm_set_array_state
,
10602 .set_disk
= imsm_set_disk
,
10603 .sync_metadata
= imsm_sync_metadata
,
10604 .activate_spare
= imsm_activate_spare
,
10605 .process_update
= imsm_process_update
,
10606 .prepare_update
= imsm_prepare_update
,
10607 #endif /* MDASSEMBLE */