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 reserve_space(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
++;
6192 static int validate_geometry_imsm(struct supertype
*st
, int level
, int layout
,
6193 int raiddisks
, int *chunk
, unsigned long long size
,
6194 char *dev
, unsigned long long *freesize
,
6202 * if given unused devices create a container
6203 * if given given devices in a container create a member volume
6205 if (level
== LEVEL_CONTAINER
) {
6206 /* Must be a fresh device to add to a container */
6207 return validate_geometry_imsm_container(st
, level
, layout
,
6209 chunk
?*chunk
:0, size
,
6216 struct intel_super
*super
= st
->sb
;
6217 if (!validate_geometry_imsm_orom(st
->sb
, level
, layout
,
6218 raiddisks
, chunk
, size
,
6221 /* we are being asked to automatically layout a
6222 * new volume based on the current contents of
6223 * the container. If the the parameters can be
6224 * satisfied reserve_space will record the disks,
6225 * start offset, and size of the volume to be
6226 * created. add_to_super and getinfo_super
6227 * detect when autolayout is in progress.
6229 /* assuming that freesize is always given when array is
6231 if (super
->orom
&& freesize
) {
6233 count
= count_volumes(super
->hba
->path
,
6234 super
->orom
->dpa
, verbose
);
6235 if (super
->orom
->vphba
<= count
) {
6236 pr_vrb(": platform does not support more"
6237 " than %d raid volumes.\n",
6238 super
->orom
->vphba
);
6243 return reserve_space(st
, raiddisks
, size
,
6244 chunk
?*chunk
:0, freesize
);
6249 /* creating in a given container */
6250 return validate_geometry_imsm_volume(st
, level
, layout
,
6251 raiddisks
, chunk
, size
,
6252 dev
, freesize
, verbose
);
6255 /* This device needs to be a device in an 'imsm' container */
6256 fd
= open(dev
, O_RDONLY
|O_EXCL
, 0);
6260 Name
": Cannot create this array on device %s\n",
6265 if (errno
!= EBUSY
|| (fd
= open(dev
, O_RDONLY
, 0)) < 0) {
6267 fprintf(stderr
, Name
": Cannot open %s: %s\n",
6268 dev
, strerror(errno
));
6271 /* Well, it is in use by someone, maybe an 'imsm' container. */
6272 cfd
= open_container(fd
);
6276 fprintf(stderr
, Name
": Cannot use %s: It is busy\n",
6280 sra
= sysfs_read(cfd
, 0, GET_VERSION
);
6281 if (sra
&& sra
->array
.major_version
== -1 &&
6282 strcmp(sra
->text_version
, "imsm") == 0)
6286 /* This is a member of a imsm container. Load the container
6287 * and try to create a volume
6289 struct intel_super
*super
;
6291 if (load_super_imsm_all(st
, cfd
, (void **) &super
, NULL
, NULL
, 1) == 0) {
6293 st
->container_dev
= fd2devnum(cfd
);
6295 return validate_geometry_imsm_volume(st
, level
, layout
,
6304 fprintf(stderr
, Name
": failed container membership check\n");
6310 static void default_geometry_imsm(struct supertype
*st
, int *level
, int *layout
, int *chunk
)
6312 struct intel_super
*super
= st
->sb
;
6314 if (level
&& *level
== UnSet
)
6315 *level
= LEVEL_CONTAINER
;
6317 if (level
&& layout
&& *layout
== UnSet
)
6318 *layout
= imsm_level_to_layout(*level
);
6320 if (chunk
&& (*chunk
== UnSet
|| *chunk
== 0))
6321 *chunk
= imsm_default_chunk(super
->orom
);
6324 static void handle_missing(struct intel_super
*super
, struct imsm_dev
*dev
);
6326 static int kill_subarray_imsm(struct supertype
*st
)
6328 /* remove the subarray currently referenced by ->current_vol */
6330 struct intel_dev
**dp
;
6331 struct intel_super
*super
= st
->sb
;
6332 __u8 current_vol
= super
->current_vol
;
6333 struct imsm_super
*mpb
= super
->anchor
;
6335 if (super
->current_vol
< 0)
6337 super
->current_vol
= -1; /* invalidate subarray cursor */
6339 /* block deletions that would change the uuid of active subarrays
6341 * FIXME when immutable ids are available, but note that we'll
6342 * also need to fixup the invalidated/active subarray indexes in
6345 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
6348 if (i
< current_vol
)
6350 sprintf(subarray
, "%u", i
);
6351 if (is_subarray_active(subarray
, st
->devname
)) {
6353 Name
": deleting subarray-%d would change the UUID of active subarray-%d, aborting\n",
6360 if (st
->update_tail
) {
6361 struct imsm_update_kill_array
*u
= malloc(sizeof(*u
));
6365 u
->type
= update_kill_array
;
6366 u
->dev_idx
= current_vol
;
6367 append_metadata_update(st
, u
, sizeof(*u
));
6372 for (dp
= &super
->devlist
; *dp
;)
6373 if ((*dp
)->index
== current_vol
) {
6376 handle_missing(super
, (*dp
)->dev
);
6377 if ((*dp
)->index
> current_vol
)
6382 /* no more raid devices, all active components are now spares,
6383 * but of course failed are still failed
6385 if (--mpb
->num_raid_devs
== 0) {
6388 for (d
= super
->disks
; d
; d
= d
->next
)
6393 super
->updates_pending
++;
6398 static int update_subarray_imsm(struct supertype
*st
, char *subarray
,
6399 char *update
, struct mddev_ident
*ident
)
6401 /* update the subarray currently referenced by ->current_vol */
6402 struct intel_super
*super
= st
->sb
;
6403 struct imsm_super
*mpb
= super
->anchor
;
6405 if (strcmp(update
, "name") == 0) {
6406 char *name
= ident
->name
;
6410 if (is_subarray_active(subarray
, st
->devname
)) {
6412 Name
": Unable to update name of active subarray\n");
6416 if (!check_name(super
, name
, 0))
6419 vol
= strtoul(subarray
, &ep
, 10);
6420 if (*ep
!= '\0' || vol
>= super
->anchor
->num_raid_devs
)
6423 if (st
->update_tail
) {
6424 struct imsm_update_rename_array
*u
= malloc(sizeof(*u
));
6428 u
->type
= update_rename_array
;
6430 snprintf((char *) u
->name
, MAX_RAID_SERIAL_LEN
, "%s", name
);
6431 append_metadata_update(st
, u
, sizeof(*u
));
6433 struct imsm_dev
*dev
;
6436 dev
= get_imsm_dev(super
, vol
);
6437 snprintf((char *) dev
->volume
, MAX_RAID_SERIAL_LEN
, "%s", name
);
6438 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
6439 dev
= get_imsm_dev(super
, i
);
6440 handle_missing(super
, dev
);
6442 super
->updates_pending
++;
6449 #endif /* MDASSEMBLE */
6451 static int is_gen_migration(struct imsm_dev
*dev
)
6456 if (!dev
->vol
.migr_state
)
6459 if (migr_type(dev
) == MIGR_GEN_MIGR
)
6465 static int is_rebuilding(struct imsm_dev
*dev
)
6467 struct imsm_map
*migr_map
;
6469 if (!dev
->vol
.migr_state
)
6472 if (migr_type(dev
) != MIGR_REBUILD
)
6475 migr_map
= get_imsm_map(dev
, MAP_1
);
6477 if (migr_map
->map_state
== IMSM_T_STATE_DEGRADED
)
6484 static int is_initializing(struct imsm_dev
*dev
)
6486 struct imsm_map
*migr_map
;
6488 if (!dev
->vol
.migr_state
)
6491 if (migr_type(dev
) != MIGR_INIT
)
6494 migr_map
= get_imsm_map(dev
, MAP_1
);
6496 if (migr_map
->map_state
== IMSM_T_STATE_UNINITIALIZED
)
6503 static void update_recovery_start(struct intel_super
*super
,
6504 struct imsm_dev
*dev
,
6505 struct mdinfo
*array
)
6507 struct mdinfo
*rebuild
= NULL
;
6511 if (!is_rebuilding(dev
))
6514 /* Find the rebuild target, but punt on the dual rebuild case */
6515 for (d
= array
->devs
; d
; d
= d
->next
)
6516 if (d
->recovery_start
== 0) {
6523 /* (?) none of the disks are marked with
6524 * IMSM_ORD_REBUILD, so assume they are missing and the
6525 * disk_ord_tbl was not correctly updated
6527 dprintf("%s: failed to locate out-of-sync disk\n", __func__
);
6531 units
= __le32_to_cpu(dev
->vol
.curr_migr_unit
);
6532 rebuild
->recovery_start
= units
* blocks_per_migr_unit(super
, dev
);
6536 static int recover_backup_imsm(struct supertype
*st
, struct mdinfo
*info
);
6539 static struct mdinfo
*container_content_imsm(struct supertype
*st
, char *subarray
)
6541 /* Given a container loaded by load_super_imsm_all,
6542 * extract information about all the arrays into
6544 * If 'subarray' is given, just extract info about that array.
6546 * For each imsm_dev create an mdinfo, fill it in,
6547 * then look for matching devices in super->disks
6548 * and create appropriate device mdinfo.
6550 struct intel_super
*super
= st
->sb
;
6551 struct imsm_super
*mpb
= super
->anchor
;
6552 struct mdinfo
*rest
= NULL
;
6556 int spare_disks
= 0;
6558 /* do not assemble arrays when not all attributes are supported */
6559 if (imsm_check_attributes(mpb
->attributes
) == 0) {
6561 fprintf(stderr
, Name
": Unsupported attributes in IMSM metadata."
6562 "Arrays activation is blocked.\n");
6565 /* check for bad blocks */
6566 if (imsm_bbm_log_size(super
->anchor
)) {
6567 fprintf(stderr
, Name
": BBM log found in IMSM metadata."
6568 "Arrays activation is blocked.\n");
6573 /* count spare devices, not used in maps
6575 for (d
= super
->disks
; d
; d
= d
->next
)
6579 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
6580 struct imsm_dev
*dev
;
6581 struct imsm_map
*map
;
6582 struct imsm_map
*map2
;
6583 struct mdinfo
*this;
6591 (i
!= strtoul(subarray
, &ep
, 10) || *ep
!= '\0'))
6594 dev
= get_imsm_dev(super
, i
);
6595 map
= get_imsm_map(dev
, MAP_0
);
6596 map2
= get_imsm_map(dev
, MAP_1
);
6598 /* do not publish arrays that are in the middle of an
6599 * unsupported migration
6601 if (dev
->vol
.migr_state
&&
6602 (migr_type(dev
) == MIGR_STATE_CHANGE
)) {
6603 fprintf(stderr
, Name
": cannot assemble volume '%.16s':"
6604 " unsupported migration in progress\n",
6608 /* do not publish arrays that are not support by controller's
6612 this = malloc(sizeof(*this));
6614 fprintf(stderr
, Name
": failed to allocate %zu bytes\n",
6619 super
->current_vol
= i
;
6620 getinfo_super_imsm_volume(st
, this, NULL
);
6623 chunk
= __le16_to_cpu(map
->blocks_per_strip
) >> 1;
6624 /* mdadm does not support all metadata features- set the bit in all arrays state */
6625 if (!validate_geometry_imsm_orom(super
,
6626 get_imsm_raid_level(map
), /* RAID level */
6627 imsm_level_to_layout(get_imsm_raid_level(map
)),
6628 map
->num_members
, /* raid disks */
6629 &chunk
, join_u32(dev
->size_low
, dev
->size_high
),
6631 fprintf(stderr
, Name
": IMSM RAID geometry validation"
6632 " failed. Array %s activation is blocked.\n",
6634 this->array
.state
|=
6635 (1<<MD_SB_BLOCK_CONTAINER_RESHAPE
) |
6636 (1<<MD_SB_BLOCK_VOLUME
);
6640 /* if array has bad blocks, set suitable bit in all arrays state */
6642 this->array
.state
|=
6643 (1<<MD_SB_BLOCK_CONTAINER_RESHAPE
) |
6644 (1<<MD_SB_BLOCK_VOLUME
);
6646 for (slot
= 0 ; slot
< map
->num_members
; slot
++) {
6647 unsigned long long recovery_start
;
6648 struct mdinfo
*info_d
;
6655 idx
= get_imsm_disk_idx(dev
, slot
, MAP_0
);
6656 ord
= get_imsm_ord_tbl_ent(dev
, slot
, MAP_X
);
6657 for (d
= super
->disks
; d
; d
= d
->next
)
6658 if (d
->index
== idx
)
6661 recovery_start
= MaxSector
;
6664 if (d
&& is_failed(&d
->disk
))
6666 if (ord
& IMSM_ORD_REBUILD
)
6670 * if we skip some disks the array will be assmebled degraded;
6671 * reset resync start to avoid a dirty-degraded
6672 * situation when performing the intial sync
6674 * FIXME handle dirty degraded
6676 if ((skip
|| recovery_start
== 0) && !dev
->vol
.dirty
)
6677 this->resync_start
= MaxSector
;
6681 info_d
= calloc(1, sizeof(*info_d
));
6683 fprintf(stderr
, Name
": failed to allocate disk"
6684 " for volume %.16s\n", dev
->volume
);
6685 info_d
= this->devs
;
6687 struct mdinfo
*d
= info_d
->next
;
6696 info_d
->next
= this->devs
;
6697 this->devs
= info_d
;
6699 info_d
->disk
.number
= d
->index
;
6700 info_d
->disk
.major
= d
->major
;
6701 info_d
->disk
.minor
= d
->minor
;
6702 info_d
->disk
.raid_disk
= slot
;
6703 info_d
->recovery_start
= recovery_start
;
6705 if (slot
< map2
->num_members
)
6706 info_d
->disk
.state
= (1 << MD_DISK_ACTIVE
);
6708 this->array
.spare_disks
++;
6710 if (slot
< map
->num_members
)
6711 info_d
->disk
.state
= (1 << MD_DISK_ACTIVE
);
6713 this->array
.spare_disks
++;
6715 if (info_d
->recovery_start
== MaxSector
)
6716 this->array
.working_disks
++;
6718 info_d
->events
= __le32_to_cpu(mpb
->generation_num
);
6719 info_d
->data_offset
= pba_of_lba0(map
);
6720 info_d
->component_size
= blocks_per_member(map
);
6722 /* now that the disk list is up-to-date fixup recovery_start */
6723 update_recovery_start(super
, dev
, this);
6724 this->array
.spare_disks
+= spare_disks
;
6727 /* check for reshape */
6728 if (this->reshape_active
== 1)
6729 recover_backup_imsm(st
, this);
6738 static __u8
imsm_check_degraded(struct intel_super
*super
, struct imsm_dev
*dev
,
6739 int failed
, int look_in_map
)
6741 struct imsm_map
*map
;
6743 map
= get_imsm_map(dev
, look_in_map
);
6746 return map
->map_state
== IMSM_T_STATE_UNINITIALIZED
?
6747 IMSM_T_STATE_UNINITIALIZED
: IMSM_T_STATE_NORMAL
;
6749 switch (get_imsm_raid_level(map
)) {
6751 return IMSM_T_STATE_FAILED
;
6754 if (failed
< map
->num_members
)
6755 return IMSM_T_STATE_DEGRADED
;
6757 return IMSM_T_STATE_FAILED
;
6762 * check to see if any mirrors have failed, otherwise we
6763 * are degraded. Even numbered slots are mirrored on
6767 /* gcc -Os complains that this is unused */
6768 int insync
= insync
;
6770 for (i
= 0; i
< map
->num_members
; i
++) {
6771 __u32 ord
= get_imsm_ord_tbl_ent(dev
, i
, MAP_X
);
6772 int idx
= ord_to_idx(ord
);
6773 struct imsm_disk
*disk
;
6775 /* reset the potential in-sync count on even-numbered
6776 * slots. num_copies is always 2 for imsm raid10
6781 disk
= get_imsm_disk(super
, idx
);
6782 if (!disk
|| is_failed(disk
) || ord
& IMSM_ORD_REBUILD
)
6785 /* no in-sync disks left in this mirror the
6789 return IMSM_T_STATE_FAILED
;
6792 return IMSM_T_STATE_DEGRADED
;
6796 return IMSM_T_STATE_DEGRADED
;
6798 return IMSM_T_STATE_FAILED
;
6804 return map
->map_state
;
6807 static int imsm_count_failed(struct intel_super
*super
, struct imsm_dev
*dev
,
6812 struct imsm_disk
*disk
;
6813 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
6814 struct imsm_map
*prev
= get_imsm_map(dev
, MAP_1
);
6815 struct imsm_map
*map_for_loop
;
6820 /* at the beginning of migration we set IMSM_ORD_REBUILD on
6821 * disks that are being rebuilt. New failures are recorded to
6822 * map[0]. So we look through all the disks we started with and
6823 * see if any failures are still present, or if any new ones
6827 if (prev
&& (map
->num_members
< prev
->num_members
))
6828 map_for_loop
= prev
;
6830 for (i
= 0; i
< map_for_loop
->num_members
; i
++) {
6832 /* when MAP_X is passed both maps failures are counted
6835 ((look_in_map
== MAP_1
) || (look_in_map
== MAP_X
)) &&
6836 (i
< prev
->num_members
)) {
6837 ord
= __le32_to_cpu(prev
->disk_ord_tbl
[i
]);
6838 idx_1
= ord_to_idx(ord
);
6840 disk
= get_imsm_disk(super
, idx_1
);
6841 if (!disk
|| is_failed(disk
) || ord
& IMSM_ORD_REBUILD
)
6844 if (((look_in_map
== MAP_0
) || (look_in_map
== MAP_X
)) &&
6845 (i
< map
->num_members
)) {
6846 ord
= __le32_to_cpu(map
->disk_ord_tbl
[i
]);
6847 idx
= ord_to_idx(ord
);
6850 disk
= get_imsm_disk(super
, idx
);
6851 if (!disk
|| is_failed(disk
) ||
6852 ord
& IMSM_ORD_REBUILD
)
6862 static int imsm_open_new(struct supertype
*c
, struct active_array
*a
,
6865 struct intel_super
*super
= c
->sb
;
6866 struct imsm_super
*mpb
= super
->anchor
;
6868 if (atoi(inst
) >= mpb
->num_raid_devs
) {
6869 fprintf(stderr
, "%s: subarry index %d, out of range\n",
6870 __func__
, atoi(inst
));
6874 dprintf("imsm: open_new %s\n", inst
);
6875 a
->info
.container_member
= atoi(inst
);
6879 static int is_resyncing(struct imsm_dev
*dev
)
6881 struct imsm_map
*migr_map
;
6883 if (!dev
->vol
.migr_state
)
6886 if (migr_type(dev
) == MIGR_INIT
||
6887 migr_type(dev
) == MIGR_REPAIR
)
6890 if (migr_type(dev
) == MIGR_GEN_MIGR
)
6893 migr_map
= get_imsm_map(dev
, MAP_1
);
6895 if ((migr_map
->map_state
== IMSM_T_STATE_NORMAL
) &&
6896 (dev
->vol
.migr_type
!= MIGR_GEN_MIGR
))
6902 /* return true if we recorded new information */
6903 static int mark_failure(struct imsm_dev
*dev
, struct imsm_disk
*disk
, int idx
)
6907 struct imsm_map
*map
;
6908 char buf
[MAX_RAID_SERIAL_LEN
+3];
6909 unsigned int len
, shift
= 0;
6911 /* new failures are always set in map[0] */
6912 map
= get_imsm_map(dev
, MAP_0
);
6914 slot
= get_imsm_disk_slot(map
, idx
);
6918 ord
= __le32_to_cpu(map
->disk_ord_tbl
[slot
]);
6919 if (is_failed(disk
) && (ord
& IMSM_ORD_REBUILD
))
6922 memcpy(buf
, disk
->serial
, MAX_RAID_SERIAL_LEN
);
6923 buf
[MAX_RAID_SERIAL_LEN
] = '\000';
6925 if ((len
= strlen(buf
)) >= MAX_RAID_SERIAL_LEN
)
6926 shift
= len
- MAX_RAID_SERIAL_LEN
+ 1;
6927 strncpy((char *)disk
->serial
, &buf
[shift
], MAX_RAID_SERIAL_LEN
);
6929 disk
->status
|= FAILED_DISK
;
6930 set_imsm_ord_tbl_ent(map
, slot
, idx
| IMSM_ORD_REBUILD
);
6931 /* mark failures in second map if second map exists and this disk
6933 * This is valid for migration, initialization and rebuild
6935 if (dev
->vol
.migr_state
) {
6936 struct imsm_map
*map2
= get_imsm_map(dev
, MAP_1
);
6937 int slot2
= get_imsm_disk_slot(map2
, idx
);
6939 if ((slot2
< map2
->num_members
) &&
6941 set_imsm_ord_tbl_ent(map2
, slot2
,
6942 idx
| IMSM_ORD_REBUILD
);
6944 if (map
->failed_disk_num
== 0xff)
6945 map
->failed_disk_num
= slot
;
6949 static void mark_missing(struct imsm_dev
*dev
, struct imsm_disk
*disk
, int idx
)
6951 mark_failure(dev
, disk
, idx
);
6953 if (disk
->scsi_id
== __cpu_to_le32(~(__u32
)0))
6956 disk
->scsi_id
= __cpu_to_le32(~(__u32
)0);
6957 memmove(&disk
->serial
[0], &disk
->serial
[1], MAX_RAID_SERIAL_LEN
- 1);
6960 static void handle_missing(struct intel_super
*super
, struct imsm_dev
*dev
)
6964 if (!super
->missing
)
6967 dprintf("imsm: mark missing\n");
6968 /* end process for initialization and rebuild only
6970 if (is_gen_migration(dev
) == 0) {
6974 failed
= imsm_count_failed(super
, dev
, MAP_0
);
6975 map_state
= imsm_check_degraded(super
, dev
, failed
, MAP_0
);
6977 end_migration(dev
, super
, map_state
);
6979 for (dl
= super
->missing
; dl
; dl
= dl
->next
)
6980 mark_missing(dev
, &dl
->disk
, dl
->index
);
6981 super
->updates_pending
++;
6984 static unsigned long long imsm_set_array_size(struct imsm_dev
*dev
,
6987 int used_disks
= imsm_num_data_members(dev
, MAP_0
);
6988 unsigned long long array_blocks
;
6989 struct imsm_map
*map
;
6991 if (used_disks
== 0) {
6992 /* when problems occures
6993 * return current array_blocks value
6995 array_blocks
= __le32_to_cpu(dev
->size_high
);
6996 array_blocks
= array_blocks
<< 32;
6997 array_blocks
+= __le32_to_cpu(dev
->size_low
);
6999 return array_blocks
;
7002 /* set array size in metadata
7004 if (new_size
<= 0) {
7005 /* OLCE size change is caused by added disks
7007 map
= get_imsm_map(dev
, MAP_0
);
7008 array_blocks
= blocks_per_member(map
) * used_disks
;
7010 /* Online Volume Size Change
7011 * Using available free space
7013 array_blocks
= new_size
;
7016 /* round array size down to closest MB
7018 array_blocks
= (array_blocks
>> SECT_PER_MB_SHIFT
) << SECT_PER_MB_SHIFT
;
7019 dev
->size_low
= __cpu_to_le32((__u32
)array_blocks
);
7020 dev
->size_high
= __cpu_to_le32((__u32
)(array_blocks
>> 32));
7022 return array_blocks
;
7025 static void imsm_set_disk(struct active_array
*a
, int n
, int state
);
7027 static void imsm_progress_container_reshape(struct intel_super
*super
)
7029 /* if no device has a migr_state, but some device has a
7030 * different number of members than the previous device, start
7031 * changing the number of devices in this device to match
7034 struct imsm_super
*mpb
= super
->anchor
;
7035 int prev_disks
= -1;
7039 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
7040 struct imsm_dev
*dev
= get_imsm_dev(super
, i
);
7041 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
7042 struct imsm_map
*map2
;
7043 int prev_num_members
;
7045 if (dev
->vol
.migr_state
)
7048 if (prev_disks
== -1)
7049 prev_disks
= map
->num_members
;
7050 if (prev_disks
== map
->num_members
)
7053 /* OK, this array needs to enter reshape mode.
7054 * i.e it needs a migr_state
7057 copy_map_size
= sizeof_imsm_map(map
);
7058 prev_num_members
= map
->num_members
;
7059 map
->num_members
= prev_disks
;
7060 dev
->vol
.migr_state
= 1;
7061 dev
->vol
.curr_migr_unit
= 0;
7062 set_migr_type(dev
, MIGR_GEN_MIGR
);
7063 for (i
= prev_num_members
;
7064 i
< map
->num_members
; i
++)
7065 set_imsm_ord_tbl_ent(map
, i
, i
);
7066 map2
= get_imsm_map(dev
, MAP_1
);
7067 /* Copy the current map */
7068 memcpy(map2
, map
, copy_map_size
);
7069 map2
->num_members
= prev_num_members
;
7071 imsm_set_array_size(dev
, -1);
7072 super
->clean_migration_record_by_mdmon
= 1;
7073 super
->updates_pending
++;
7077 /* Handle dirty -> clean transititions, resync and reshape. Degraded and rebuild
7078 * states are handled in imsm_set_disk() with one exception, when a
7079 * resync is stopped due to a new failure this routine will set the
7080 * 'degraded' state for the array.
7082 static int imsm_set_array_state(struct active_array
*a
, int consistent
)
7084 int inst
= a
->info
.container_member
;
7085 struct intel_super
*super
= a
->container
->sb
;
7086 struct imsm_dev
*dev
= get_imsm_dev(super
, inst
);
7087 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
7088 int failed
= imsm_count_failed(super
, dev
, MAP_0
);
7089 __u8 map_state
= imsm_check_degraded(super
, dev
, failed
, MAP_0
);
7090 __u32 blocks_per_unit
;
7092 if (dev
->vol
.migr_state
&&
7093 dev
->vol
.migr_type
== MIGR_GEN_MIGR
) {
7094 /* array state change is blocked due to reshape action
7096 * - abort the reshape (if last_checkpoint is 0 and action!= reshape)
7097 * - finish the reshape (if last_checkpoint is big and action != reshape)
7098 * - update curr_migr_unit
7100 if (a
->curr_action
== reshape
) {
7101 /* still reshaping, maybe update curr_migr_unit */
7102 goto mark_checkpoint
;
7104 if (a
->last_checkpoint
== 0 && a
->prev_action
== reshape
) {
7105 /* for some reason we aborted the reshape.
7107 * disable automatic metadata rollback
7108 * user action is required to recover process
7111 struct imsm_map
*map2
=
7112 get_imsm_map(dev
, MAP_1
);
7113 dev
->vol
.migr_state
= 0;
7114 set_migr_type(dev
, 0);
7115 dev
->vol
.curr_migr_unit
= 0;
7117 sizeof_imsm_map(map2
));
7118 super
->updates_pending
++;
7121 if (a
->last_checkpoint
>= a
->info
.component_size
) {
7122 unsigned long long array_blocks
;
7126 used_disks
= imsm_num_data_members(dev
, MAP_0
);
7127 if (used_disks
> 0) {
7129 blocks_per_member(map
) *
7131 /* round array size down to closest MB
7133 array_blocks
= (array_blocks
7134 >> SECT_PER_MB_SHIFT
)
7135 << SECT_PER_MB_SHIFT
;
7136 a
->info
.custom_array_size
= array_blocks
;
7137 /* encourage manager to update array
7141 a
->check_reshape
= 1;
7143 /* finalize online capacity expansion/reshape */
7144 for (mdi
= a
->info
.devs
; mdi
; mdi
= mdi
->next
)
7146 mdi
->disk
.raid_disk
,
7149 imsm_progress_container_reshape(super
);
7154 /* before we activate this array handle any missing disks */
7155 if (consistent
== 2)
7156 handle_missing(super
, dev
);
7158 if (consistent
== 2 &&
7159 (!is_resync_complete(&a
->info
) ||
7160 map_state
!= IMSM_T_STATE_NORMAL
||
7161 dev
->vol
.migr_state
))
7164 if (is_resync_complete(&a
->info
)) {
7165 /* complete intialization / resync,
7166 * recovery and interrupted recovery is completed in
7169 if (is_resyncing(dev
)) {
7170 dprintf("imsm: mark resync done\n");
7171 end_migration(dev
, super
, map_state
);
7172 super
->updates_pending
++;
7173 a
->last_checkpoint
= 0;
7175 } else if ((!is_resyncing(dev
) && !failed
) &&
7176 (imsm_reshape_blocks_arrays_changes(super
) == 0)) {
7177 /* mark the start of the init process if nothing is failed */
7178 dprintf("imsm: mark resync start\n");
7179 if (map
->map_state
== IMSM_T_STATE_UNINITIALIZED
)
7180 migrate(dev
, super
, IMSM_T_STATE_NORMAL
, MIGR_INIT
);
7182 migrate(dev
, super
, IMSM_T_STATE_NORMAL
, MIGR_REPAIR
);
7183 super
->updates_pending
++;
7187 /* skip checkpointing for general migration,
7188 * it is controlled in mdadm
7190 if (is_gen_migration(dev
))
7191 goto skip_mark_checkpoint
;
7193 /* check if we can update curr_migr_unit from resync_start, recovery_start */
7194 blocks_per_unit
= blocks_per_migr_unit(super
, dev
);
7195 if (blocks_per_unit
) {
7199 units
= a
->last_checkpoint
/ blocks_per_unit
;
7202 /* check that we did not overflow 32-bits, and that
7203 * curr_migr_unit needs updating
7205 if (units32
== units
&&
7207 __le32_to_cpu(dev
->vol
.curr_migr_unit
) != units32
) {
7208 dprintf("imsm: mark checkpoint (%u)\n", units32
);
7209 dev
->vol
.curr_migr_unit
= __cpu_to_le32(units32
);
7210 super
->updates_pending
++;
7214 skip_mark_checkpoint
:
7215 /* mark dirty / clean */
7216 if (dev
->vol
.dirty
!= !consistent
) {
7217 dprintf("imsm: mark '%s'\n", consistent
? "clean" : "dirty");
7222 super
->updates_pending
++;
7228 static void imsm_set_disk(struct active_array
*a
, int n
, int state
)
7230 int inst
= a
->info
.container_member
;
7231 struct intel_super
*super
= a
->container
->sb
;
7232 struct imsm_dev
*dev
= get_imsm_dev(super
, inst
);
7233 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
7234 struct imsm_disk
*disk
;
7239 if (n
> map
->num_members
)
7240 fprintf(stderr
, "imsm: set_disk %d out of range 0..%d\n",
7241 n
, map
->num_members
- 1);
7246 dprintf("imsm: set_disk %d:%x\n", n
, state
);
7248 ord
= get_imsm_ord_tbl_ent(dev
, n
, MAP_0
);
7249 disk
= get_imsm_disk(super
, ord_to_idx(ord
));
7251 /* check for new failures */
7252 if (state
& DS_FAULTY
) {
7253 if (mark_failure(dev
, disk
, ord_to_idx(ord
)))
7254 super
->updates_pending
++;
7257 /* check if in_sync */
7258 if (state
& DS_INSYNC
&& ord
& IMSM_ORD_REBUILD
&& is_rebuilding(dev
)) {
7259 struct imsm_map
*migr_map
= get_imsm_map(dev
, MAP_1
);
7261 set_imsm_ord_tbl_ent(migr_map
, n
, ord_to_idx(ord
));
7262 super
->updates_pending
++;
7265 failed
= imsm_count_failed(super
, dev
, MAP_0
);
7266 map_state
= imsm_check_degraded(super
, dev
, failed
, MAP_0
);
7268 /* check if recovery complete, newly degraded, or failed */
7269 dprintf("imsm: Detected transition to state ");
7270 switch (map_state
) {
7271 case IMSM_T_STATE_NORMAL
: /* transition to normal state */
7272 dprintf("normal: ");
7273 if (is_rebuilding(dev
)) {
7274 dprintf("while rebuilding");
7275 end_migration(dev
, super
, map_state
);
7276 map
= get_imsm_map(dev
, MAP_0
);
7277 map
->failed_disk_num
= ~0;
7278 super
->updates_pending
++;
7279 a
->last_checkpoint
= 0;
7282 if (is_gen_migration(dev
)) {
7283 dprintf("while general migration");
7284 if (a
->last_checkpoint
>= a
->info
.component_size
)
7285 end_migration(dev
, super
, map_state
);
7287 map
->map_state
= map_state
;
7288 map
= get_imsm_map(dev
, MAP_0
);
7289 map
->failed_disk_num
= ~0;
7290 super
->updates_pending
++;
7294 case IMSM_T_STATE_DEGRADED
: /* transition to degraded state */
7295 dprintf("degraded: ");
7296 if ((map
->map_state
!= map_state
) &&
7297 !dev
->vol
.migr_state
) {
7298 dprintf("mark degraded");
7299 map
->map_state
= map_state
;
7300 super
->updates_pending
++;
7301 a
->last_checkpoint
= 0;
7304 if (is_rebuilding(dev
)) {
7305 dprintf("while rebuilding.");
7306 if (map
->map_state
!= map_state
) {
7307 dprintf(" Map state change");
7308 end_migration(dev
, super
, map_state
);
7309 super
->updates_pending
++;
7313 if (is_gen_migration(dev
)) {
7314 dprintf("while general migration");
7315 if (a
->last_checkpoint
>= a
->info
.component_size
)
7316 end_migration(dev
, super
, map_state
);
7318 map
->map_state
= map_state
;
7319 manage_second_map(super
, dev
);
7321 super
->updates_pending
++;
7324 if (is_initializing(dev
)) {
7325 dprintf("while initialization.");
7326 map
->map_state
= map_state
;
7327 super
->updates_pending
++;
7331 case IMSM_T_STATE_FAILED
: /* transition to failed state */
7332 dprintf("failed: ");
7333 if (is_gen_migration(dev
)) {
7334 dprintf("while general migration");
7335 map
->map_state
= map_state
;
7336 super
->updates_pending
++;
7339 if (map
->map_state
!= map_state
) {
7340 dprintf("mark failed");
7341 end_migration(dev
, super
, map_state
);
7342 super
->updates_pending
++;
7343 a
->last_checkpoint
= 0;
7348 dprintf("state %i\n", map_state
);
7354 static int store_imsm_mpb(int fd
, struct imsm_super
*mpb
)
7357 __u32 mpb_size
= __le32_to_cpu(mpb
->mpb_size
);
7358 unsigned long long dsize
;
7359 unsigned long long sectors
;
7361 get_dev_size(fd
, NULL
, &dsize
);
7363 if (mpb_size
> 512) {
7364 /* -1 to account for anchor */
7365 sectors
= mpb_sectors(mpb
) - 1;
7367 /* write the extended mpb to the sectors preceeding the anchor */
7368 if (lseek64(fd
, dsize
- (512 * (2 + sectors
)), SEEK_SET
) < 0)
7371 if ((unsigned long long)write(fd
, buf
+ 512, 512 * sectors
)
7376 /* first block is stored on second to last sector of the disk */
7377 if (lseek64(fd
, dsize
- (512 * 2), SEEK_SET
) < 0)
7380 if (write(fd
, buf
, 512) != 512)
7386 static void imsm_sync_metadata(struct supertype
*container
)
7388 struct intel_super
*super
= container
->sb
;
7390 dprintf("sync metadata: %d\n", super
->updates_pending
);
7391 if (!super
->updates_pending
)
7394 write_super_imsm(container
, 0);
7396 super
->updates_pending
= 0;
7399 static struct dl
*imsm_readd(struct intel_super
*super
, int idx
, struct active_array
*a
)
7401 struct imsm_dev
*dev
= get_imsm_dev(super
, a
->info
.container_member
);
7402 int i
= get_imsm_disk_idx(dev
, idx
, MAP_X
);
7405 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
7409 if (dl
&& is_failed(&dl
->disk
))
7413 dprintf("%s: found %x:%x\n", __func__
, dl
->major
, dl
->minor
);
7418 static struct dl
*imsm_add_spare(struct intel_super
*super
, int slot
,
7419 struct active_array
*a
, int activate_new
,
7420 struct mdinfo
*additional_test_list
)
7422 struct imsm_dev
*dev
= get_imsm_dev(super
, a
->info
.container_member
);
7423 int idx
= get_imsm_disk_idx(dev
, slot
, MAP_X
);
7424 struct imsm_super
*mpb
= super
->anchor
;
7425 struct imsm_map
*map
;
7426 unsigned long long pos
;
7431 __u32 array_start
= 0;
7432 __u32 array_end
= 0;
7434 struct mdinfo
*test_list
;
7436 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
7437 /* If in this array, skip */
7438 for (d
= a
->info
.devs
; d
; d
= d
->next
)
7439 if (d
->state_fd
>= 0 &&
7440 d
->disk
.major
== dl
->major
&&
7441 d
->disk
.minor
== dl
->minor
) {
7442 dprintf("%x:%x already in array\n",
7443 dl
->major
, dl
->minor
);
7448 test_list
= additional_test_list
;
7450 if (test_list
->disk
.major
== dl
->major
&&
7451 test_list
->disk
.minor
== dl
->minor
) {
7452 dprintf("%x:%x already in additional test list\n",
7453 dl
->major
, dl
->minor
);
7456 test_list
= test_list
->next
;
7461 /* skip in use or failed drives */
7462 if (is_failed(&dl
->disk
) || idx
== dl
->index
||
7464 dprintf("%x:%x status (failed: %d index: %d)\n",
7465 dl
->major
, dl
->minor
, is_failed(&dl
->disk
), idx
);
7469 /* skip pure spares when we are looking for partially
7470 * assimilated drives
7472 if (dl
->index
== -1 && !activate_new
)
7475 /* Does this unused device have the requisite free space?
7476 * It needs to be able to cover all member volumes
7478 ex
= get_extents(super
, dl
);
7480 dprintf("cannot get extents\n");
7483 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
7484 dev
= get_imsm_dev(super
, i
);
7485 map
= get_imsm_map(dev
, MAP_0
);
7487 /* check if this disk is already a member of
7490 if (get_imsm_disk_slot(map
, dl
->index
) >= 0)
7496 array_start
= pba_of_lba0(map
);
7497 array_end
= array_start
+
7498 blocks_per_member(map
) - 1;
7501 /* check that we can start at pba_of_lba0 with
7502 * blocks_per_member of space
7504 if (array_start
>= pos
&& array_end
< ex
[j
].start
) {
7508 pos
= ex
[j
].start
+ ex
[j
].size
;
7510 } while (ex
[j
-1].size
);
7517 if (i
< mpb
->num_raid_devs
) {
7518 dprintf("%x:%x does not have %u to %u available\n",
7519 dl
->major
, dl
->minor
, array_start
, array_end
);
7530 static int imsm_rebuild_allowed(struct supertype
*cont
, int dev_idx
, int failed
)
7532 struct imsm_dev
*dev2
;
7533 struct imsm_map
*map
;
7539 dev2
= get_imsm_dev(cont
->sb
, dev_idx
);
7541 state
= imsm_check_degraded(cont
->sb
, dev2
, failed
, MAP_0
);
7542 if (state
== IMSM_T_STATE_FAILED
) {
7543 map
= get_imsm_map(dev2
, MAP_0
);
7546 for (slot
= 0; slot
< map
->num_members
; slot
++) {
7548 * Check if failed disks are deleted from intel
7549 * disk list or are marked to be deleted
7551 idx
= get_imsm_disk_idx(dev2
, slot
, MAP_X
);
7552 idisk
= get_imsm_dl_disk(cont
->sb
, idx
);
7554 * Do not rebuild the array if failed disks
7555 * from failed sub-array are not removed from
7559 is_failed(&idisk
->disk
) &&
7560 (idisk
->action
!= DISK_REMOVE
))
7568 static struct mdinfo
*imsm_activate_spare(struct active_array
*a
,
7569 struct metadata_update
**updates
)
7572 * Find a device with unused free space and use it to replace a
7573 * failed/vacant region in an array. We replace failed regions one a
7574 * array at a time. The result is that a new spare disk will be added
7575 * to the first failed array and after the monitor has finished
7576 * propagating failures the remainder will be consumed.
7578 * FIXME add a capability for mdmon to request spares from another
7582 struct intel_super
*super
= a
->container
->sb
;
7583 int inst
= a
->info
.container_member
;
7584 struct imsm_dev
*dev
= get_imsm_dev(super
, inst
);
7585 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
7586 int failed
= a
->info
.array
.raid_disks
;
7587 struct mdinfo
*rv
= NULL
;
7590 struct metadata_update
*mu
;
7592 struct imsm_update_activate_spare
*u
;
7597 for (d
= a
->info
.devs
; d
; d
= d
->next
) {
7598 if ((d
->curr_state
& DS_FAULTY
) &&
7600 /* wait for Removal to happen */
7602 if (d
->state_fd
>= 0)
7606 dprintf("imsm: activate spare: inst=%d failed=%d (%d) level=%d\n",
7607 inst
, failed
, a
->info
.array
.raid_disks
, a
->info
.array
.level
);
7609 if (imsm_reshape_blocks_arrays_changes(super
))
7612 /* Cannot activate another spare if rebuild is in progress already
7614 if (is_rebuilding(dev
)) {
7615 dprintf("imsm: No spare activation allowed. "
7616 "Rebuild in progress already.\n");
7620 if (a
->info
.array
.level
== 4)
7621 /* No repair for takeovered array
7622 * imsm doesn't support raid4
7626 if (imsm_check_degraded(super
, dev
, failed
, MAP_0
) !=
7627 IMSM_T_STATE_DEGRADED
)
7631 * If there are any failed disks check state of the other volume.
7632 * Block rebuild if the another one is failed until failed disks
7633 * are removed from container.
7636 dprintf("found failed disks in %.*s, check if there another"
7637 "failed sub-array.\n",
7638 MAX_RAID_SERIAL_LEN
, dev
->volume
);
7639 /* check if states of the other volumes allow for rebuild */
7640 for (i
= 0; i
< super
->anchor
->num_raid_devs
; i
++) {
7642 allowed
= imsm_rebuild_allowed(a
->container
,
7650 /* For each slot, if it is not working, find a spare */
7651 for (i
= 0; i
< a
->info
.array
.raid_disks
; i
++) {
7652 for (d
= a
->info
.devs
; d
; d
= d
->next
)
7653 if (d
->disk
.raid_disk
== i
)
7655 dprintf("found %d: %p %x\n", i
, d
, d
?d
->curr_state
:0);
7656 if (d
&& (d
->state_fd
>= 0))
7660 * OK, this device needs recovery. Try to re-add the
7661 * previous occupant of this slot, if this fails see if
7662 * we can continue the assimilation of a spare that was
7663 * partially assimilated, finally try to activate a new
7666 dl
= imsm_readd(super
, i
, a
);
7668 dl
= imsm_add_spare(super
, i
, a
, 0, rv
);
7670 dl
= imsm_add_spare(super
, i
, a
, 1, rv
);
7674 /* found a usable disk with enough space */
7675 di
= malloc(sizeof(*di
));
7678 memset(di
, 0, sizeof(*di
));
7680 /* dl->index will be -1 in the case we are activating a
7681 * pristine spare. imsm_process_update() will create a
7682 * new index in this case. Once a disk is found to be
7683 * failed in all member arrays it is kicked from the
7686 di
->disk
.number
= dl
->index
;
7688 /* (ab)use di->devs to store a pointer to the device
7691 di
->devs
= (struct mdinfo
*) dl
;
7693 di
->disk
.raid_disk
= i
;
7694 di
->disk
.major
= dl
->major
;
7695 di
->disk
.minor
= dl
->minor
;
7697 di
->recovery_start
= 0;
7698 di
->data_offset
= pba_of_lba0(map
);
7699 di
->component_size
= a
->info
.component_size
;
7700 di
->container_member
= inst
;
7701 super
->random
= random32();
7705 dprintf("%x:%x to be %d at %llu\n", dl
->major
, dl
->minor
,
7706 i
, di
->data_offset
);
7710 /* No spares found */
7712 /* Now 'rv' has a list of devices to return.
7713 * Create a metadata_update record to update the
7714 * disk_ord_tbl for the array
7716 mu
= malloc(sizeof(*mu
));
7718 mu
->buf
= malloc(sizeof(struct imsm_update_activate_spare
) * num_spares
);
7719 if (mu
->buf
== NULL
) {
7726 struct mdinfo
*n
= rv
->next
;
7735 mu
->space_list
= NULL
;
7736 mu
->len
= sizeof(struct imsm_update_activate_spare
) * num_spares
;
7737 mu
->next
= *updates
;
7738 u
= (struct imsm_update_activate_spare
*) mu
->buf
;
7740 for (di
= rv
; di
; di
= di
->next
) {
7741 u
->type
= update_activate_spare
;
7742 u
->dl
= (struct dl
*) di
->devs
;
7744 u
->slot
= di
->disk
.raid_disk
;
7755 static int disks_overlap(struct intel_super
*super
, int idx
, struct imsm_update_create_array
*u
)
7757 struct imsm_dev
*dev
= get_imsm_dev(super
, idx
);
7758 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
7759 struct imsm_map
*new_map
= get_imsm_map(&u
->dev
, MAP_0
);
7760 struct disk_info
*inf
= get_disk_info(u
);
7761 struct imsm_disk
*disk
;
7765 for (i
= 0; i
< map
->num_members
; i
++) {
7766 disk
= get_imsm_disk(super
, get_imsm_disk_idx(dev
, i
, MAP_X
));
7767 for (j
= 0; j
< new_map
->num_members
; j
++)
7768 if (serialcmp(disk
->serial
, inf
[j
].serial
) == 0)
7776 static struct dl
*get_disk_super(struct intel_super
*super
, int major
, int minor
)
7778 struct dl
*dl
= NULL
;
7779 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
7780 if ((dl
->major
== major
) && (dl
->minor
== minor
))
7785 static int remove_disk_super(struct intel_super
*super
, int major
, int minor
)
7787 struct dl
*prev
= NULL
;
7791 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
7792 if ((dl
->major
== major
) && (dl
->minor
== minor
)) {
7795 prev
->next
= dl
->next
;
7797 super
->disks
= dl
->next
;
7799 __free_imsm_disk(dl
);
7800 dprintf("%s: removed %x:%x\n",
7801 __func__
, major
, minor
);
7809 static void imsm_delete(struct intel_super
*super
, struct dl
**dlp
, unsigned index
);
7811 static int add_remove_disk_update(struct intel_super
*super
)
7813 int check_degraded
= 0;
7814 struct dl
*disk
= NULL
;
7815 /* add/remove some spares to/from the metadata/contrainer */
7816 while (super
->disk_mgmt_list
) {
7817 struct dl
*disk_cfg
;
7819 disk_cfg
= super
->disk_mgmt_list
;
7820 super
->disk_mgmt_list
= disk_cfg
->next
;
7821 disk_cfg
->next
= NULL
;
7823 if (disk_cfg
->action
== DISK_ADD
) {
7824 disk_cfg
->next
= super
->disks
;
7825 super
->disks
= disk_cfg
;
7827 dprintf("%s: added %x:%x\n",
7828 __func__
, disk_cfg
->major
,
7830 } else if (disk_cfg
->action
== DISK_REMOVE
) {
7831 dprintf("Disk remove action processed: %x.%x\n",
7832 disk_cfg
->major
, disk_cfg
->minor
);
7833 disk
= get_disk_super(super
,
7837 /* store action status */
7838 disk
->action
= DISK_REMOVE
;
7839 /* remove spare disks only */
7840 if (disk
->index
== -1) {
7841 remove_disk_super(super
,
7846 /* release allocate disk structure */
7847 __free_imsm_disk(disk_cfg
);
7850 return check_degraded
;
7854 static int apply_reshape_migration_update(struct imsm_update_reshape_migration
*u
,
7855 struct intel_super
*super
,
7858 struct intel_dev
*id
;
7859 void **tofree
= NULL
;
7862 dprintf("apply_reshape_migration_update()\n");
7863 if ((u
->subdev
< 0) ||
7865 dprintf("imsm: Error: Wrong subdev: %i\n", u
->subdev
);
7868 if ((space_list
== NULL
) || (*space_list
== NULL
)) {
7869 dprintf("imsm: Error: Memory is not allocated\n");
7873 for (id
= super
->devlist
; id
; id
= id
->next
) {
7874 if (id
->index
== (unsigned)u
->subdev
) {
7875 struct imsm_dev
*dev
= get_imsm_dev(super
, u
->subdev
);
7876 struct imsm_map
*map
;
7877 struct imsm_dev
*new_dev
=
7878 (struct imsm_dev
*)*space_list
;
7879 struct imsm_map
*migr_map
= get_imsm_map(dev
, MAP_1
);
7881 struct dl
*new_disk
;
7883 if (new_dev
== NULL
)
7885 *space_list
= **space_list
;
7886 memcpy(new_dev
, dev
, sizeof_imsm_dev(dev
, 0));
7887 map
= get_imsm_map(new_dev
, MAP_0
);
7889 dprintf("imsm: Error: migration in progress");
7893 to_state
= map
->map_state
;
7894 if ((u
->new_level
== 5) && (map
->raid_level
== 0)) {
7896 /* this should not happen */
7897 if (u
->new_disks
[0] < 0) {
7898 map
->failed_disk_num
=
7899 map
->num_members
- 1;
7900 to_state
= IMSM_T_STATE_DEGRADED
;
7902 to_state
= IMSM_T_STATE_NORMAL
;
7904 migrate(new_dev
, super
, to_state
, MIGR_GEN_MIGR
);
7905 if (u
->new_level
> -1)
7906 map
->raid_level
= u
->new_level
;
7907 migr_map
= get_imsm_map(new_dev
, MAP_1
);
7908 if ((u
->new_level
== 5) &&
7909 (migr_map
->raid_level
== 0)) {
7910 int ord
= map
->num_members
- 1;
7911 migr_map
->num_members
--;
7912 if (u
->new_disks
[0] < 0)
7913 ord
|= IMSM_ORD_REBUILD
;
7914 set_imsm_ord_tbl_ent(map
,
7915 map
->num_members
- 1,
7919 tofree
= (void **)dev
;
7921 /* update chunk size
7923 if (u
->new_chunksize
> 0)
7924 map
->blocks_per_strip
=
7925 __cpu_to_le16(u
->new_chunksize
* 2);
7929 if ((u
->new_level
!= 5) ||
7930 (migr_map
->raid_level
!= 0) ||
7931 (migr_map
->raid_level
== map
->raid_level
))
7934 if (u
->new_disks
[0] >= 0) {
7937 new_disk
= get_disk_super(super
,
7938 major(u
->new_disks
[0]),
7939 minor(u
->new_disks
[0]));
7940 dprintf("imsm: new disk for reshape is: %i:%i "
7941 "(%p, index = %i)\n",
7942 major(u
->new_disks
[0]),
7943 minor(u
->new_disks
[0]),
7944 new_disk
, new_disk
->index
);
7945 if (new_disk
== NULL
)
7946 goto error_disk_add
;
7948 new_disk
->index
= map
->num_members
- 1;
7949 /* slot to fill in autolayout
7951 new_disk
->raiddisk
= new_disk
->index
;
7952 new_disk
->disk
.status
|= CONFIGURED_DISK
;
7953 new_disk
->disk
.status
&= ~SPARE_DISK
;
7955 goto error_disk_add
;
7958 *tofree
= *space_list
;
7959 /* calculate new size
7961 imsm_set_array_size(new_dev
, -1);
7968 *space_list
= tofree
;
7972 dprintf("Error: imsm: Cannot find disk.\n");
7976 static int apply_size_change_update(struct imsm_update_size_change
*u
,
7977 struct intel_super
*super
)
7979 struct intel_dev
*id
;
7982 dprintf("apply_size_change_update()\n");
7983 if ((u
->subdev
< 0) ||
7985 dprintf("imsm: Error: Wrong subdev: %i\n", u
->subdev
);
7989 for (id
= super
->devlist
; id
; id
= id
->next
) {
7990 if (id
->index
== (unsigned)u
->subdev
) {
7991 struct imsm_dev
*dev
= get_imsm_dev(super
, u
->subdev
);
7992 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
7993 int used_disks
= imsm_num_data_members(dev
, MAP_0
);
7994 unsigned long long blocks_per_member
;
7996 /* calculate new size
7998 blocks_per_member
= u
->new_size
/ used_disks
;
7999 dprintf("imsm: apply_size_change_update(size: %llu, "
8000 "blocks per member: %llu)\n",
8001 u
->new_size
, blocks_per_member
);
8002 set_blocks_per_member(map
, blocks_per_member
);
8003 imsm_set_array_size(dev
, u
->new_size
);
8014 static int apply_update_activate_spare(struct imsm_update_activate_spare
*u
,
8015 struct intel_super
*super
,
8016 struct active_array
*active_array
)
8018 struct imsm_super
*mpb
= super
->anchor
;
8019 struct imsm_dev
*dev
= get_imsm_dev(super
, u
->array
);
8020 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
8021 struct imsm_map
*migr_map
;
8022 struct active_array
*a
;
8023 struct imsm_disk
*disk
;
8030 int second_map_created
= 0;
8032 for (; u
; u
= u
->next
) {
8033 victim
= get_imsm_disk_idx(dev
, u
->slot
, MAP_X
);
8038 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
8043 fprintf(stderr
, "error: imsm_activate_spare passed "
8044 "an unknown disk (index: %d)\n",
8049 /* count failures (excluding rebuilds and the victim)
8050 * to determine map[0] state
8053 for (i
= 0; i
< map
->num_members
; i
++) {
8056 disk
= get_imsm_disk(super
,
8057 get_imsm_disk_idx(dev
, i
, MAP_X
));
8058 if (!disk
|| is_failed(disk
))
8062 /* adding a pristine spare, assign a new index */
8063 if (dl
->index
< 0) {
8064 dl
->index
= super
->anchor
->num_disks
;
8065 super
->anchor
->num_disks
++;
8068 disk
->status
|= CONFIGURED_DISK
;
8069 disk
->status
&= ~SPARE_DISK
;
8072 to_state
= imsm_check_degraded(super
, dev
, failed
, MAP_0
);
8073 if (!second_map_created
) {
8074 second_map_created
= 1;
8075 map
->map_state
= IMSM_T_STATE_DEGRADED
;
8076 migrate(dev
, super
, to_state
, MIGR_REBUILD
);
8078 map
->map_state
= to_state
;
8079 migr_map
= get_imsm_map(dev
, MAP_1
);
8080 set_imsm_ord_tbl_ent(map
, u
->slot
, dl
->index
);
8081 set_imsm_ord_tbl_ent(migr_map
, u
->slot
,
8082 dl
->index
| IMSM_ORD_REBUILD
);
8084 /* update the family_num to mark a new container
8085 * generation, being careful to record the existing
8086 * family_num in orig_family_num to clean up after
8087 * earlier mdadm versions that neglected to set it.
8089 if (mpb
->orig_family_num
== 0)
8090 mpb
->orig_family_num
= mpb
->family_num
;
8091 mpb
->family_num
+= super
->random
;
8093 /* count arrays using the victim in the metadata */
8095 for (a
= active_array
; a
; a
= a
->next
) {
8096 dev
= get_imsm_dev(super
, a
->info
.container_member
);
8097 map
= get_imsm_map(dev
, MAP_0
);
8099 if (get_imsm_disk_slot(map
, victim
) >= 0)
8103 /* delete the victim if it is no longer being
8109 /* We know that 'manager' isn't touching anything,
8110 * so it is safe to delete
8112 for (dlp
= &super
->disks
; *dlp
; dlp
= &(*dlp
)->next
)
8113 if ((*dlp
)->index
== victim
)
8116 /* victim may be on the missing list */
8118 for (dlp
= &super
->missing
; *dlp
;
8119 dlp
= &(*dlp
)->next
)
8120 if ((*dlp
)->index
== victim
)
8122 imsm_delete(super
, dlp
, victim
);
8129 static int apply_reshape_container_disks_update(struct imsm_update_reshape
*u
,
8130 struct intel_super
*super
,
8133 struct dl
*new_disk
;
8134 struct intel_dev
*id
;
8136 int delta_disks
= u
->new_raid_disks
- u
->old_raid_disks
;
8137 int disk_count
= u
->old_raid_disks
;
8138 void **tofree
= NULL
;
8139 int devices_to_reshape
= 1;
8140 struct imsm_super
*mpb
= super
->anchor
;
8142 unsigned int dev_id
;
8144 dprintf("imsm: apply_reshape_container_disks_update()\n");
8146 /* enable spares to use in array */
8147 for (i
= 0; i
< delta_disks
; i
++) {
8148 new_disk
= get_disk_super(super
,
8149 major(u
->new_disks
[i
]),
8150 minor(u
->new_disks
[i
]));
8151 dprintf("imsm: new disk for reshape is: %i:%i "
8152 "(%p, index = %i)\n",
8153 major(u
->new_disks
[i
]), minor(u
->new_disks
[i
]),
8154 new_disk
, new_disk
->index
);
8155 if ((new_disk
== NULL
) ||
8156 ((new_disk
->index
>= 0) &&
8157 (new_disk
->index
< u
->old_raid_disks
)))
8158 goto update_reshape_exit
;
8159 new_disk
->index
= disk_count
++;
8160 /* slot to fill in autolayout
8162 new_disk
->raiddisk
= new_disk
->index
;
8163 new_disk
->disk
.status
|=
8165 new_disk
->disk
.status
&= ~SPARE_DISK
;
8168 dprintf("imsm: volume set mpb->num_raid_devs = %i\n",
8169 mpb
->num_raid_devs
);
8170 /* manage changes in volume
8172 for (dev_id
= 0; dev_id
< mpb
->num_raid_devs
; dev_id
++) {
8173 void **sp
= *space_list
;
8174 struct imsm_dev
*newdev
;
8175 struct imsm_map
*newmap
, *oldmap
;
8177 for (id
= super
->devlist
; id
; id
= id
->next
) {
8178 if (id
->index
== dev_id
)
8187 /* Copy the dev, but not (all of) the map */
8188 memcpy(newdev
, id
->dev
, sizeof(*newdev
));
8189 oldmap
= get_imsm_map(id
->dev
, MAP_0
);
8190 newmap
= get_imsm_map(newdev
, MAP_0
);
8191 /* Copy the current map */
8192 memcpy(newmap
, oldmap
, sizeof_imsm_map(oldmap
));
8193 /* update one device only
8195 if (devices_to_reshape
) {
8196 dprintf("imsm: modifying subdev: %i\n",
8198 devices_to_reshape
--;
8199 newdev
->vol
.migr_state
= 1;
8200 newdev
->vol
.curr_migr_unit
= 0;
8201 set_migr_type(newdev
, MIGR_GEN_MIGR
);
8202 newmap
->num_members
= u
->new_raid_disks
;
8203 for (i
= 0; i
< delta_disks
; i
++) {
8204 set_imsm_ord_tbl_ent(newmap
,
8205 u
->old_raid_disks
+ i
,
8206 u
->old_raid_disks
+ i
);
8208 /* New map is correct, now need to save old map
8210 newmap
= get_imsm_map(newdev
, MAP_1
);
8211 memcpy(newmap
, oldmap
, sizeof_imsm_map(oldmap
));
8213 imsm_set_array_size(newdev
, -1);
8216 sp
= (void **)id
->dev
;
8221 /* Clear migration record */
8222 memset(super
->migr_rec
, 0, sizeof(struct migr_record
));
8225 *space_list
= tofree
;
8228 update_reshape_exit
:
8233 static int apply_takeover_update(struct imsm_update_takeover
*u
,
8234 struct intel_super
*super
,
8237 struct imsm_dev
*dev
= NULL
;
8238 struct intel_dev
*dv
;
8239 struct imsm_dev
*dev_new
;
8240 struct imsm_map
*map
;
8244 for (dv
= super
->devlist
; dv
; dv
= dv
->next
)
8245 if (dv
->index
== (unsigned int)u
->subarray
) {
8253 map
= get_imsm_map(dev
, MAP_0
);
8255 if (u
->direction
== R10_TO_R0
) {
8256 /* Number of failed disks must be half of initial disk number */
8257 if (imsm_count_failed(super
, dev
, MAP_0
) !=
8258 (map
->num_members
/ 2))
8261 /* iterate through devices to mark removed disks as spare */
8262 for (dm
= super
->disks
; dm
; dm
= dm
->next
) {
8263 if (dm
->disk
.status
& FAILED_DISK
) {
8264 int idx
= dm
->index
;
8265 /* update indexes on the disk list */
8266 /* FIXME this loop-with-the-loop looks wrong, I'm not convinced
8267 the index values will end up being correct.... NB */
8268 for (du
= super
->disks
; du
; du
= du
->next
)
8269 if (du
->index
> idx
)
8271 /* mark as spare disk */
8276 map
->num_members
= map
->num_members
/ 2;
8277 map
->map_state
= IMSM_T_STATE_NORMAL
;
8278 map
->num_domains
= 1;
8279 map
->raid_level
= 0;
8280 map
->failed_disk_num
= -1;
8283 if (u
->direction
== R0_TO_R10
) {
8285 /* update slots in current disk list */
8286 for (dm
= super
->disks
; dm
; dm
= dm
->next
) {
8290 /* create new *missing* disks */
8291 for (i
= 0; i
< map
->num_members
; i
++) {
8292 space
= *space_list
;
8295 *space_list
= *space
;
8297 memcpy(du
, super
->disks
, sizeof(*du
));
8301 du
->index
= (i
* 2) + 1;
8302 sprintf((char *)du
->disk
.serial
,
8303 " MISSING_%d", du
->index
);
8304 sprintf((char *)du
->serial
,
8305 "MISSING_%d", du
->index
);
8306 du
->next
= super
->missing
;
8307 super
->missing
= du
;
8309 /* create new dev and map */
8310 space
= *space_list
;
8313 *space_list
= *space
;
8314 dev_new
= (void *)space
;
8315 memcpy(dev_new
, dev
, sizeof(*dev
));
8316 /* update new map */
8317 map
= get_imsm_map(dev_new
, MAP_0
);
8318 map
->num_members
= map
->num_members
* 2;
8319 map
->map_state
= IMSM_T_STATE_DEGRADED
;
8320 map
->num_domains
= 2;
8321 map
->raid_level
= 1;
8322 /* replace dev<->dev_new */
8325 /* update disk order table */
8326 for (du
= super
->disks
; du
; du
= du
->next
)
8328 set_imsm_ord_tbl_ent(map
, du
->index
, du
->index
);
8329 for (du
= super
->missing
; du
; du
= du
->next
)
8330 if (du
->index
>= 0) {
8331 set_imsm_ord_tbl_ent(map
, du
->index
, du
->index
);
8332 mark_missing(dv
->dev
, &du
->disk
, du
->index
);
8338 static void imsm_process_update(struct supertype
*st
,
8339 struct metadata_update
*update
)
8342 * crack open the metadata_update envelope to find the update record
8343 * update can be one of:
8344 * update_reshape_container_disks - all the arrays in the container
8345 * are being reshaped to have more devices. We need to mark
8346 * the arrays for general migration and convert selected spares
8347 * into active devices.
8348 * update_activate_spare - a spare device has replaced a failed
8349 * device in an array, update the disk_ord_tbl. If this disk is
8350 * present in all member arrays then also clear the SPARE_DISK
8352 * update_create_array
8354 * update_rename_array
8355 * update_add_remove_disk
8357 struct intel_super
*super
= st
->sb
;
8358 struct imsm_super
*mpb
;
8359 enum imsm_update_type type
= *(enum imsm_update_type
*) update
->buf
;
8361 /* update requires a larger buf but the allocation failed */
8362 if (super
->next_len
&& !super
->next_buf
) {
8363 super
->next_len
= 0;
8367 if (super
->next_buf
) {
8368 memcpy(super
->next_buf
, super
->buf
, super
->len
);
8370 super
->len
= super
->next_len
;
8371 super
->buf
= super
->next_buf
;
8373 super
->next_len
= 0;
8374 super
->next_buf
= NULL
;
8377 mpb
= super
->anchor
;
8380 case update_general_migration_checkpoint
: {
8381 struct intel_dev
*id
;
8382 struct imsm_update_general_migration_checkpoint
*u
=
8383 (void *)update
->buf
;
8385 dprintf("imsm: process_update() "
8386 "for update_general_migration_checkpoint called\n");
8388 /* find device under general migration */
8389 for (id
= super
->devlist
; id
; id
= id
->next
) {
8390 if (is_gen_migration(id
->dev
)) {
8391 id
->dev
->vol
.curr_migr_unit
=
8392 __cpu_to_le32(u
->curr_migr_unit
);
8393 super
->updates_pending
++;
8398 case update_takeover
: {
8399 struct imsm_update_takeover
*u
= (void *)update
->buf
;
8400 if (apply_takeover_update(u
, super
, &update
->space_list
)) {
8401 imsm_update_version_info(super
);
8402 super
->updates_pending
++;
8407 case update_reshape_container_disks
: {
8408 struct imsm_update_reshape
*u
= (void *)update
->buf
;
8409 if (apply_reshape_container_disks_update(
8410 u
, super
, &update
->space_list
))
8411 super
->updates_pending
++;
8414 case update_reshape_migration
: {
8415 struct imsm_update_reshape_migration
*u
= (void *)update
->buf
;
8416 if (apply_reshape_migration_update(
8417 u
, super
, &update
->space_list
))
8418 super
->updates_pending
++;
8421 case update_size_change
: {
8422 struct imsm_update_size_change
*u
= (void *)update
->buf
;
8423 if (apply_size_change_update(u
, super
))
8424 super
->updates_pending
++;
8427 case update_activate_spare
: {
8428 struct imsm_update_activate_spare
*u
= (void *) update
->buf
;
8429 if (apply_update_activate_spare(u
, super
, st
->arrays
))
8430 super
->updates_pending
++;
8433 case update_create_array
: {
8434 /* someone wants to create a new array, we need to be aware of
8435 * a few races/collisions:
8436 * 1/ 'Create' called by two separate instances of mdadm
8437 * 2/ 'Create' versus 'activate_spare': mdadm has chosen
8438 * devices that have since been assimilated via
8440 * In the event this update can not be carried out mdadm will
8441 * (FIX ME) notice that its update did not take hold.
8443 struct imsm_update_create_array
*u
= (void *) update
->buf
;
8444 struct intel_dev
*dv
;
8445 struct imsm_dev
*dev
;
8446 struct imsm_map
*map
, *new_map
;
8447 unsigned long long start
, end
;
8448 unsigned long long new_start
, new_end
;
8450 struct disk_info
*inf
;
8453 /* handle racing creates: first come first serve */
8454 if (u
->dev_idx
< mpb
->num_raid_devs
) {
8455 dprintf("%s: subarray %d already defined\n",
8456 __func__
, u
->dev_idx
);
8460 /* check update is next in sequence */
8461 if (u
->dev_idx
!= mpb
->num_raid_devs
) {
8462 dprintf("%s: can not create array %d expected index %d\n",
8463 __func__
, u
->dev_idx
, mpb
->num_raid_devs
);
8467 new_map
= get_imsm_map(&u
->dev
, MAP_0
);
8468 new_start
= pba_of_lba0(new_map
);
8469 new_end
= new_start
+ blocks_per_member(new_map
);
8470 inf
= get_disk_info(u
);
8472 /* handle activate_spare versus create race:
8473 * check to make sure that overlapping arrays do not include
8476 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
8477 dev
= get_imsm_dev(super
, i
);
8478 map
= get_imsm_map(dev
, MAP_0
);
8479 start
= pba_of_lba0(map
);
8480 end
= start
+ blocks_per_member(map
);
8481 if ((new_start
>= start
&& new_start
<= end
) ||
8482 (start
>= new_start
&& start
<= new_end
))
8487 if (disks_overlap(super
, i
, u
)) {
8488 dprintf("%s: arrays overlap\n", __func__
);
8493 /* check that prepare update was successful */
8494 if (!update
->space
) {
8495 dprintf("%s: prepare update failed\n", __func__
);
8499 /* check that all disks are still active before committing
8500 * changes. FIXME: could we instead handle this by creating a
8501 * degraded array? That's probably not what the user expects,
8502 * so better to drop this update on the floor.
8504 for (i
= 0; i
< new_map
->num_members
; i
++) {
8505 dl
= serial_to_dl(inf
[i
].serial
, super
);
8507 dprintf("%s: disk disappeared\n", __func__
);
8512 super
->updates_pending
++;
8514 /* convert spares to members and fixup ord_tbl */
8515 for (i
= 0; i
< new_map
->num_members
; i
++) {
8516 dl
= serial_to_dl(inf
[i
].serial
, super
);
8517 if (dl
->index
== -1) {
8518 dl
->index
= mpb
->num_disks
;
8520 dl
->disk
.status
|= CONFIGURED_DISK
;
8521 dl
->disk
.status
&= ~SPARE_DISK
;
8523 set_imsm_ord_tbl_ent(new_map
, i
, dl
->index
);
8528 update
->space
= NULL
;
8529 imsm_copy_dev(dev
, &u
->dev
);
8530 dv
->index
= u
->dev_idx
;
8531 dv
->next
= super
->devlist
;
8532 super
->devlist
= dv
;
8533 mpb
->num_raid_devs
++;
8535 imsm_update_version_info(super
);
8538 /* mdmon knows how to release update->space, but not
8539 * ((struct intel_dev *) update->space)->dev
8541 if (update
->space
) {
8547 case update_kill_array
: {
8548 struct imsm_update_kill_array
*u
= (void *) update
->buf
;
8549 int victim
= u
->dev_idx
;
8550 struct active_array
*a
;
8551 struct intel_dev
**dp
;
8552 struct imsm_dev
*dev
;
8554 /* sanity check that we are not affecting the uuid of
8555 * active arrays, or deleting an active array
8557 * FIXME when immutable ids are available, but note that
8558 * we'll also need to fixup the invalidated/active
8559 * subarray indexes in mdstat
8561 for (a
= st
->arrays
; a
; a
= a
->next
)
8562 if (a
->info
.container_member
>= victim
)
8564 /* by definition if mdmon is running at least one array
8565 * is active in the container, so checking
8566 * mpb->num_raid_devs is just extra paranoia
8568 dev
= get_imsm_dev(super
, victim
);
8569 if (a
|| !dev
|| mpb
->num_raid_devs
== 1) {
8570 dprintf("failed to delete subarray-%d\n", victim
);
8574 for (dp
= &super
->devlist
; *dp
;)
8575 if ((*dp
)->index
== (unsigned)super
->current_vol
) {
8578 if ((*dp
)->index
> (unsigned)victim
)
8582 mpb
->num_raid_devs
--;
8583 super
->updates_pending
++;
8586 case update_rename_array
: {
8587 struct imsm_update_rename_array
*u
= (void *) update
->buf
;
8588 char name
[MAX_RAID_SERIAL_LEN
+1];
8589 int target
= u
->dev_idx
;
8590 struct active_array
*a
;
8591 struct imsm_dev
*dev
;
8593 /* sanity check that we are not affecting the uuid of
8596 snprintf(name
, MAX_RAID_SERIAL_LEN
, "%s", (char *) u
->name
);
8597 name
[MAX_RAID_SERIAL_LEN
] = '\0';
8598 for (a
= st
->arrays
; a
; a
= a
->next
)
8599 if (a
->info
.container_member
== target
)
8601 dev
= get_imsm_dev(super
, u
->dev_idx
);
8602 if (a
|| !dev
|| !check_name(super
, name
, 1)) {
8603 dprintf("failed to rename subarray-%d\n", target
);
8607 snprintf((char *) dev
->volume
, MAX_RAID_SERIAL_LEN
, "%s", name
);
8608 super
->updates_pending
++;
8611 case update_add_remove_disk
: {
8612 /* we may be able to repair some arrays if disks are
8613 * being added, check teh status of add_remove_disk
8614 * if discs has been added.
8616 if (add_remove_disk_update(super
)) {
8617 struct active_array
*a
;
8619 super
->updates_pending
++;
8620 for (a
= st
->arrays
; a
; a
= a
->next
)
8621 a
->check_degraded
= 1;
8626 fprintf(stderr
, "error: unsuported process update type:"
8627 "(type: %d)\n", type
);
8631 static struct mdinfo
*get_spares_for_grow(struct supertype
*st
);
8633 static void imsm_prepare_update(struct supertype
*st
,
8634 struct metadata_update
*update
)
8637 * Allocate space to hold new disk entries, raid-device entries or a new
8638 * mpb if necessary. The manager synchronously waits for updates to
8639 * complete in the monitor, so new mpb buffers allocated here can be
8640 * integrated by the monitor thread without worrying about live pointers
8641 * in the manager thread.
8643 enum imsm_update_type type
= *(enum imsm_update_type
*) update
->buf
;
8644 struct intel_super
*super
= st
->sb
;
8645 struct imsm_super
*mpb
= super
->anchor
;
8650 case update_general_migration_checkpoint
:
8651 dprintf("imsm: prepare_update() "
8652 "for update_general_migration_checkpoint called\n");
8654 case update_takeover
: {
8655 struct imsm_update_takeover
*u
= (void *)update
->buf
;
8656 if (u
->direction
== R0_TO_R10
) {
8657 void **tail
= (void **)&update
->space_list
;
8658 struct imsm_dev
*dev
= get_imsm_dev(super
, u
->subarray
);
8659 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
8660 int num_members
= map
->num_members
;
8664 /* allocate memory for added disks */
8665 for (i
= 0; i
< num_members
; i
++) {
8666 size
= sizeof(struct dl
);
8667 space
= malloc(size
);
8676 /* allocate memory for new device */
8677 size
= sizeof_imsm_dev(super
->devlist
->dev
, 0) +
8678 (num_members
* sizeof(__u32
));
8679 space
= malloc(size
);
8688 len
= disks_to_mpb_size(num_members
* 2);
8690 /* if allocation didn't success, free buffer */
8691 while (update
->space_list
) {
8692 void **sp
= update
->space_list
;
8693 update
->space_list
= *sp
;
8701 case update_reshape_container_disks
: {
8702 /* Every raid device in the container is about to
8703 * gain some more devices, and we will enter a
8705 * So each 'imsm_map' will be bigger, and the imsm_vol
8706 * will now hold 2 of them.
8707 * Thus we need new 'struct imsm_dev' allocations sized
8708 * as sizeof_imsm_dev but with more devices in both maps.
8710 struct imsm_update_reshape
*u
= (void *)update
->buf
;
8711 struct intel_dev
*dl
;
8712 void **space_tail
= (void**)&update
->space_list
;
8714 dprintf("imsm: imsm_prepare_update() for update_reshape\n");
8716 for (dl
= super
->devlist
; dl
; dl
= dl
->next
) {
8717 int size
= sizeof_imsm_dev(dl
->dev
, 1);
8719 if (u
->new_raid_disks
> u
->old_raid_disks
)
8720 size
+= sizeof(__u32
)*2*
8721 (u
->new_raid_disks
- u
->old_raid_disks
);
8730 len
= disks_to_mpb_size(u
->new_raid_disks
);
8731 dprintf("New anchor length is %llu\n", (unsigned long long)len
);
8734 case update_reshape_migration
: {
8735 /* for migration level 0->5 we need to add disks
8736 * so the same as for container operation we will copy
8737 * device to the bigger location.
8738 * in memory prepared device and new disk area are prepared
8739 * for usage in process update
8741 struct imsm_update_reshape_migration
*u
= (void *)update
->buf
;
8742 struct intel_dev
*id
;
8743 void **space_tail
= (void **)&update
->space_list
;
8746 int current_level
= -1;
8748 dprintf("imsm: imsm_prepare_update() for update_reshape\n");
8750 /* add space for bigger array in update
8752 for (id
= super
->devlist
; id
; id
= id
->next
) {
8753 if (id
->index
== (unsigned)u
->subdev
) {
8754 size
= sizeof_imsm_dev(id
->dev
, 1);
8755 if (u
->new_raid_disks
> u
->old_raid_disks
)
8756 size
+= sizeof(__u32
)*2*
8757 (u
->new_raid_disks
- u
->old_raid_disks
);
8767 if (update
->space_list
== NULL
)
8770 /* add space for disk in update
8772 size
= sizeof(struct dl
);
8775 free(update
->space_list
);
8776 update
->space_list
= NULL
;
8783 /* add spare device to update
8785 for (id
= super
->devlist
; id
; id
= id
->next
)
8786 if (id
->index
== (unsigned)u
->subdev
) {
8787 struct imsm_dev
*dev
;
8788 struct imsm_map
*map
;
8790 dev
= get_imsm_dev(super
, u
->subdev
);
8791 map
= get_imsm_map(dev
, MAP_0
);
8792 current_level
= map
->raid_level
;
8795 if ((u
->new_level
== 5) && (u
->new_level
!= current_level
)) {
8796 struct mdinfo
*spares
;
8798 spares
= get_spares_for_grow(st
);
8806 makedev(dev
->disk
.major
,
8808 dl
= get_disk_super(super
,
8811 dl
->index
= u
->old_raid_disks
;
8817 len
= disks_to_mpb_size(u
->new_raid_disks
);
8818 dprintf("New anchor length is %llu\n", (unsigned long long)len
);
8821 case update_size_change
: {
8824 case update_create_array
: {
8825 struct imsm_update_create_array
*u
= (void *) update
->buf
;
8826 struct intel_dev
*dv
;
8827 struct imsm_dev
*dev
= &u
->dev
;
8828 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
8830 struct disk_info
*inf
;
8834 inf
= get_disk_info(u
);
8835 len
= sizeof_imsm_dev(dev
, 1);
8836 /* allocate a new super->devlist entry */
8837 dv
= malloc(sizeof(*dv
));
8839 dv
->dev
= malloc(len
);
8844 update
->space
= NULL
;
8848 /* count how many spares will be converted to members */
8849 for (i
= 0; i
< map
->num_members
; i
++) {
8850 dl
= serial_to_dl(inf
[i
].serial
, super
);
8852 /* hmm maybe it failed?, nothing we can do about
8857 if (count_memberships(dl
, super
) == 0)
8860 len
+= activate
* sizeof(struct imsm_disk
);
8867 /* check if we need a larger metadata buffer */
8868 if (super
->next_buf
)
8869 buf_len
= super
->next_len
;
8871 buf_len
= super
->len
;
8873 if (__le32_to_cpu(mpb
->mpb_size
) + len
> buf_len
) {
8874 /* ok we need a larger buf than what is currently allocated
8875 * if this allocation fails process_update will notice that
8876 * ->next_len is set and ->next_buf is NULL
8878 buf_len
= ROUND_UP(__le32_to_cpu(mpb
->mpb_size
) + len
, 512);
8879 if (super
->next_buf
)
8880 free(super
->next_buf
);
8882 super
->next_len
= buf_len
;
8883 if (posix_memalign(&super
->next_buf
, 512, buf_len
) == 0)
8884 memset(super
->next_buf
, 0, buf_len
);
8886 super
->next_buf
= NULL
;
8890 /* must be called while manager is quiesced */
8891 static void imsm_delete(struct intel_super
*super
, struct dl
**dlp
, unsigned index
)
8893 struct imsm_super
*mpb
= super
->anchor
;
8895 struct imsm_dev
*dev
;
8896 struct imsm_map
*map
;
8897 int i
, j
, num_members
;
8900 dprintf("%s: deleting device[%d] from imsm_super\n",
8903 /* shift all indexes down one */
8904 for (iter
= super
->disks
; iter
; iter
= iter
->next
)
8905 if (iter
->index
> (int)index
)
8907 for (iter
= super
->missing
; iter
; iter
= iter
->next
)
8908 if (iter
->index
> (int)index
)
8911 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
8912 dev
= get_imsm_dev(super
, i
);
8913 map
= get_imsm_map(dev
, MAP_0
);
8914 num_members
= map
->num_members
;
8915 for (j
= 0; j
< num_members
; j
++) {
8916 /* update ord entries being careful not to propagate
8917 * ord-flags to the first map
8919 ord
= get_imsm_ord_tbl_ent(dev
, j
, MAP_X
);
8921 if (ord_to_idx(ord
) <= index
)
8924 map
= get_imsm_map(dev
, MAP_0
);
8925 set_imsm_ord_tbl_ent(map
, j
, ord_to_idx(ord
- 1));
8926 map
= get_imsm_map(dev
, MAP_1
);
8928 set_imsm_ord_tbl_ent(map
, j
, ord
- 1);
8933 super
->updates_pending
++;
8935 struct dl
*dl
= *dlp
;
8937 *dlp
= (*dlp
)->next
;
8938 __free_imsm_disk(dl
);
8941 #endif /* MDASSEMBLE */
8943 static void close_targets(int *targets
, int new_disks
)
8950 for (i
= 0; i
< new_disks
; i
++) {
8951 if (targets
[i
] >= 0) {
8958 static int imsm_get_allowed_degradation(int level
, int raid_disks
,
8959 struct intel_super
*super
,
8960 struct imsm_dev
*dev
)
8966 struct imsm_map
*map
;
8969 ret_val
= raid_disks
/2;
8970 /* check map if all disks pairs not failed
8973 map
= get_imsm_map(dev
, MAP_0
);
8974 for (i
= 0; i
< ret_val
; i
++) {
8975 int degradation
= 0;
8976 if (get_imsm_disk(super
, i
) == NULL
)
8978 if (get_imsm_disk(super
, i
+ 1) == NULL
)
8980 if (degradation
== 2)
8983 map
= get_imsm_map(dev
, MAP_1
);
8984 /* if there is no second map
8985 * result can be returned
8989 /* check degradation in second map
8991 for (i
= 0; i
< ret_val
; i
++) {
8992 int degradation
= 0;
8993 if (get_imsm_disk(super
, i
) == NULL
)
8995 if (get_imsm_disk(super
, i
+ 1) == NULL
)
8997 if (degradation
== 2)
9012 /*******************************************************************************
9013 * Function: open_backup_targets
9014 * Description: Function opens file descriptors for all devices given in
9017 * info : general array info
9018 * raid_disks : number of disks
9019 * raid_fds : table of device's file descriptors
9020 * super : intel super for raid10 degradation check
9021 * dev : intel device for raid10 degradation check
9025 ******************************************************************************/
9026 int open_backup_targets(struct mdinfo
*info
, int raid_disks
, int *raid_fds
,
9027 struct intel_super
*super
, struct imsm_dev
*dev
)
9033 for (i
= 0; i
< raid_disks
; i
++)
9036 for (sd
= info
->devs
; sd
; sd
= sd
->next
) {
9039 if (sd
->disk
.state
& (1<<MD_DISK_FAULTY
)) {
9040 dprintf("disk is faulty!!\n");
9044 if ((sd
->disk
.raid_disk
>= raid_disks
) ||
9045 (sd
->disk
.raid_disk
< 0))
9048 dn
= map_dev(sd
->disk
.major
,
9050 raid_fds
[sd
->disk
.raid_disk
] = dev_open(dn
, O_RDWR
);
9051 if (raid_fds
[sd
->disk
.raid_disk
] < 0) {
9052 fprintf(stderr
, "cannot open component\n");
9057 /* check if maximum array degradation level is not exceeded
9059 if ((raid_disks
- opened
) >
9060 imsm_get_allowed_degradation(info
->new_level
,
9063 fprintf(stderr
, "Not enough disks can be opened.\n");
9064 close_targets(raid_fds
, raid_disks
);
9071 /*******************************************************************************
9072 * Function: init_migr_record_imsm
9073 * Description: Function inits imsm migration record
9075 * super : imsm internal array info
9076 * dev : device under migration
9077 * info : general array info to find the smallest device
9080 ******************************************************************************/
9081 void init_migr_record_imsm(struct supertype
*st
, struct imsm_dev
*dev
,
9082 struct mdinfo
*info
)
9084 struct intel_super
*super
= st
->sb
;
9085 struct migr_record
*migr_rec
= super
->migr_rec
;
9087 unsigned long long dsize
, dev_sectors
;
9088 long long unsigned min_dev_sectors
= -1LLU;
9092 struct imsm_map
*map_dest
= get_imsm_map(dev
, MAP_0
);
9093 struct imsm_map
*map_src
= get_imsm_map(dev
, MAP_1
);
9094 unsigned long long num_migr_units
;
9095 unsigned long long array_blocks
;
9097 memset(migr_rec
, 0, sizeof(struct migr_record
));
9098 migr_rec
->family_num
= __cpu_to_le32(super
->anchor
->family_num
);
9100 /* only ascending reshape supported now */
9101 migr_rec
->ascending_migr
= __cpu_to_le32(1);
9103 migr_rec
->dest_depth_per_unit
= GEN_MIGR_AREA_SIZE
/
9104 max(map_dest
->blocks_per_strip
, map_src
->blocks_per_strip
);
9105 migr_rec
->dest_depth_per_unit
*=
9106 max(map_dest
->blocks_per_strip
, map_src
->blocks_per_strip
);
9107 new_data_disks
= imsm_num_data_members(dev
, MAP_0
);
9108 migr_rec
->blocks_per_unit
=
9109 __cpu_to_le32(migr_rec
->dest_depth_per_unit
* new_data_disks
);
9110 migr_rec
->dest_depth_per_unit
=
9111 __cpu_to_le32(migr_rec
->dest_depth_per_unit
);
9112 array_blocks
= info
->component_size
* new_data_disks
;
9114 array_blocks
/ __le32_to_cpu(migr_rec
->blocks_per_unit
);
9116 if (array_blocks
% __le32_to_cpu(migr_rec
->blocks_per_unit
))
9118 migr_rec
->num_migr_units
= __cpu_to_le32(num_migr_units
);
9120 migr_rec
->post_migr_vol_cap
= dev
->size_low
;
9121 migr_rec
->post_migr_vol_cap_hi
= dev
->size_high
;
9124 /* Find the smallest dev */
9125 for (sd
= info
->devs
; sd
; sd
= sd
->next
) {
9126 sprintf(nm
, "%d:%d", sd
->disk
.major
, sd
->disk
.minor
);
9127 fd
= dev_open(nm
, O_RDONLY
);
9130 get_dev_size(fd
, NULL
, &dsize
);
9131 dev_sectors
= dsize
/ 512;
9132 if (dev_sectors
< min_dev_sectors
)
9133 min_dev_sectors
= dev_sectors
;
9136 migr_rec
->ckpt_area_pba
= __cpu_to_le32(min_dev_sectors
-
9137 RAID_DISK_RESERVED_BLOCKS_IMSM_HI
);
9139 write_imsm_migr_rec(st
);
9144 /*******************************************************************************
9145 * Function: save_backup_imsm
9146 * Description: Function saves critical data stripes to Migration Copy Area
9147 * and updates the current migration unit status.
9148 * Use restore_stripes() to form a destination stripe,
9149 * and to write it to the Copy Area.
9151 * st : supertype information
9152 * dev : imsm device that backup is saved for
9153 * info : general array info
9154 * buf : input buffer
9155 * length : length of data to backup (blocks_per_unit)
9159 ******************************************************************************/
9160 int save_backup_imsm(struct supertype
*st
,
9161 struct imsm_dev
*dev
,
9162 struct mdinfo
*info
,
9167 struct intel_super
*super
= st
->sb
;
9168 unsigned long long *target_offsets
= NULL
;
9169 int *targets
= NULL
;
9171 struct imsm_map
*map_dest
= get_imsm_map(dev
, MAP_0
);
9172 int new_disks
= map_dest
->num_members
;
9173 int dest_layout
= 0;
9175 unsigned long long start
;
9176 int data_disks
= imsm_num_data_members(dev
, MAP_0
);
9178 targets
= malloc(new_disks
* sizeof(int));
9182 for (i
= 0; i
< new_disks
; i
++)
9185 target_offsets
= malloc(new_disks
* sizeof(unsigned long long));
9186 if (!target_offsets
)
9189 start
= info
->reshape_progress
* 512;
9190 for (i
= 0; i
< new_disks
; i
++) {
9191 target_offsets
[i
] = (unsigned long long)
9192 __le32_to_cpu(super
->migr_rec
->ckpt_area_pba
) * 512;
9193 /* move back copy area adderss, it will be moved forward
9194 * in restore_stripes() using start input variable
9196 target_offsets
[i
] -= start
/data_disks
;
9199 if (open_backup_targets(info
, new_disks
, targets
,
9203 dest_layout
= imsm_level_to_layout(map_dest
->raid_level
);
9204 dest_chunk
= __le16_to_cpu(map_dest
->blocks_per_strip
) * 512;
9206 if (restore_stripes(targets
, /* list of dest devices */
9207 target_offsets
, /* migration record offsets */
9210 map_dest
->raid_level
,
9212 -1, /* source backup file descriptor */
9213 0, /* input buf offset
9214 * always 0 buf is already offseted */
9218 fprintf(stderr
, Name
": Error restoring stripes\n");
9226 close_targets(targets
, new_disks
);
9229 free(target_offsets
);
9234 /*******************************************************************************
9235 * Function: save_checkpoint_imsm
9236 * Description: Function called for current unit status update
9237 * in the migration record. It writes it to disk.
9239 * super : imsm internal array info
9240 * info : general array info
9244 * 2: failure, means no valid migration record
9245 * / no general migration in progress /
9246 ******************************************************************************/
9247 int save_checkpoint_imsm(struct supertype
*st
, struct mdinfo
*info
, int state
)
9249 struct intel_super
*super
= st
->sb
;
9250 unsigned long long blocks_per_unit
;
9251 unsigned long long curr_migr_unit
;
9253 if (load_imsm_migr_rec(super
, info
) != 0) {
9254 dprintf("imsm: ERROR: Cannot read migration record "
9255 "for checkpoint save.\n");
9259 blocks_per_unit
= __le32_to_cpu(super
->migr_rec
->blocks_per_unit
);
9260 if (blocks_per_unit
== 0) {
9261 dprintf("imsm: no migration in progress.\n");
9264 curr_migr_unit
= info
->reshape_progress
/ blocks_per_unit
;
9265 /* check if array is alligned to copy area
9266 * if it is not alligned, add one to current migration unit value
9267 * this can happend on array reshape finish only
9269 if (info
->reshape_progress
% blocks_per_unit
)
9272 super
->migr_rec
->curr_migr_unit
=
9273 __cpu_to_le32(curr_migr_unit
);
9274 super
->migr_rec
->rec_status
= __cpu_to_le32(state
);
9275 super
->migr_rec
->dest_1st_member_lba
=
9276 __cpu_to_le32(curr_migr_unit
*
9277 __le32_to_cpu(super
->migr_rec
->dest_depth_per_unit
));
9278 if (write_imsm_migr_rec(st
) < 0) {
9279 dprintf("imsm: Cannot write migration record "
9280 "outside backup area\n");
9287 /*******************************************************************************
9288 * Function: recover_backup_imsm
9289 * Description: Function recovers critical data from the Migration Copy Area
9290 * while assembling an array.
9292 * super : imsm internal array info
9293 * info : general array info
9295 * 0 : success (or there is no data to recover)
9297 ******************************************************************************/
9298 int recover_backup_imsm(struct supertype
*st
, struct mdinfo
*info
)
9300 struct intel_super
*super
= st
->sb
;
9301 struct migr_record
*migr_rec
= super
->migr_rec
;
9302 struct imsm_map
*map_dest
= NULL
;
9303 struct intel_dev
*id
= NULL
;
9304 unsigned long long read_offset
;
9305 unsigned long long write_offset
;
9307 int *targets
= NULL
;
9308 int new_disks
, i
, err
;
9311 unsigned long curr_migr_unit
= __le32_to_cpu(migr_rec
->curr_migr_unit
);
9312 unsigned long num_migr_units
= __le32_to_cpu(migr_rec
->num_migr_units
);
9314 int skipped_disks
= 0;
9316 err
= sysfs_get_str(info
, NULL
, "array_state", (char *)buffer
, 20);
9320 /* recover data only during assemblation */
9321 if (strncmp(buffer
, "inactive", 8) != 0)
9323 /* no data to recover */
9324 if (__le32_to_cpu(migr_rec
->rec_status
) == UNIT_SRC_NORMAL
)
9326 if (curr_migr_unit
>= num_migr_units
)
9329 /* find device during reshape */
9330 for (id
= super
->devlist
; id
; id
= id
->next
)
9331 if (is_gen_migration(id
->dev
))
9336 map_dest
= get_imsm_map(id
->dev
, MAP_0
);
9337 new_disks
= map_dest
->num_members
;
9339 read_offset
= (unsigned long long)
9340 __le32_to_cpu(migr_rec
->ckpt_area_pba
) * 512;
9342 write_offset
= ((unsigned long long)
9343 __le32_to_cpu(migr_rec
->dest_1st_member_lba
) +
9344 pba_of_lba0(map_dest
)) * 512;
9346 unit_len
= __le32_to_cpu(migr_rec
->dest_depth_per_unit
) * 512;
9347 if (posix_memalign((void **)&buf
, 512, unit_len
) != 0)
9349 targets
= malloc(new_disks
* sizeof(int));
9353 if (open_backup_targets(info
, new_disks
, targets
, super
, id
->dev
)) {
9355 Name
": Cannot open some devices belonging to array.\n");
9359 for (i
= 0; i
< new_disks
; i
++) {
9360 if (targets
[i
] < 0) {
9364 if (lseek64(targets
[i
], read_offset
, SEEK_SET
) < 0) {
9366 Name
": Cannot seek to block: %s\n",
9371 if ((unsigned)read(targets
[i
], buf
, unit_len
) != unit_len
) {
9373 Name
": Cannot read copy area block: %s\n",
9378 if (lseek64(targets
[i
], write_offset
, SEEK_SET
) < 0) {
9380 Name
": Cannot seek to block: %s\n",
9385 if ((unsigned)write(targets
[i
], buf
, unit_len
) != unit_len
) {
9387 Name
": Cannot restore block: %s\n",
9394 if (skipped_disks
> imsm_get_allowed_degradation(info
->new_level
,
9399 Name
": Cannot restore data from backup."
9400 " Too many failed disks\n");
9404 if (save_checkpoint_imsm(st
, info
, UNIT_SRC_NORMAL
)) {
9405 /* ignore error == 2, this can mean end of reshape here
9407 dprintf("imsm: Cannot write checkpoint to "
9408 "migration record (UNIT_SRC_NORMAL) during restart\n");
9414 for (i
= 0; i
< new_disks
; i
++)
9423 static char disk_by_path
[] = "/dev/disk/by-path/";
9425 static const char *imsm_get_disk_controller_domain(const char *path
)
9427 char disk_path
[PATH_MAX
];
9431 strncpy(disk_path
, disk_by_path
, PATH_MAX
- 1);
9432 strncat(disk_path
, path
, PATH_MAX
- strlen(disk_path
) - 1);
9433 if (stat(disk_path
, &st
) == 0) {
9434 struct sys_dev
* hba
;
9437 path
= devt_to_devpath(st
.st_rdev
);
9440 hba
= find_disk_attached_hba(-1, path
);
9441 if (hba
&& hba
->type
== SYS_DEV_SAS
)
9443 else if (hba
&& hba
->type
== SYS_DEV_SATA
)
9447 dprintf("path: %s hba: %s attached: %s\n",
9448 path
, (hba
) ? hba
->path
: "NULL", drv
);
9456 static int imsm_find_array_minor_by_subdev(int subdev
, int container
, int *minor
)
9458 char subdev_name
[20];
9459 struct mdstat_ent
*mdstat
;
9461 sprintf(subdev_name
, "%d", subdev
);
9462 mdstat
= mdstat_by_subdev(subdev_name
, container
);
9466 *minor
= mdstat
->devnum
;
9467 free_mdstat(mdstat
);
9471 static int imsm_reshape_is_allowed_on_container(struct supertype
*st
,
9472 struct geo_params
*geo
,
9473 int *old_raid_disks
,
9476 /* currently we only support increasing the number of devices
9477 * for a container. This increases the number of device for each
9478 * member array. They must all be RAID0 or RAID5.
9481 struct mdinfo
*info
, *member
;
9482 int devices_that_can_grow
= 0;
9484 dprintf("imsm: imsm_reshape_is_allowed_on_container(ENTER): "
9485 "st->devnum = (%i)\n",
9488 if (geo
->size
!= -1 ||
9489 geo
->level
!= UnSet
||
9490 geo
->layout
!= UnSet
||
9491 geo
->chunksize
!= 0 ||
9492 geo
->raid_disks
== UnSet
) {
9493 dprintf("imsm: Container operation is allowed for "
9494 "raid disks number change only.\n");
9498 if (direction
== ROLLBACK_METADATA_CHANGES
) {
9499 dprintf("imsm: Metadata changes rollback is not supported for "
9500 "container operation.\n");
9504 info
= container_content_imsm(st
, NULL
);
9505 for (member
= info
; member
; member
= member
->next
) {
9509 dprintf("imsm: checking device_num: %i\n",
9510 member
->container_member
);
9512 if (geo
->raid_disks
<= member
->array
.raid_disks
) {
9513 /* we work on container for Online Capacity Expansion
9514 * only so raid_disks has to grow
9516 dprintf("imsm: for container operation raid disks "
9517 "increase is required\n");
9521 if ((info
->array
.level
!= 0) &&
9522 (info
->array
.level
!= 5)) {
9523 /* we cannot use this container with other raid level
9525 dprintf("imsm: for container operation wrong"
9526 " raid level (%i) detected\n",
9530 /* check for platform support
9531 * for this raid level configuration
9533 struct intel_super
*super
= st
->sb
;
9534 if (!is_raid_level_supported(super
->orom
,
9535 member
->array
.level
,
9537 dprintf("platform does not support raid%d with"
9541 geo
->raid_disks
> 1 ? "s" : "");
9544 /* check if component size is aligned to chunk size
9546 if (info
->component_size
%
9547 (info
->array
.chunk_size
/512)) {
9548 dprintf("Component size is not aligned to "
9554 if (*old_raid_disks
&&
9555 info
->array
.raid_disks
!= *old_raid_disks
)
9557 *old_raid_disks
= info
->array
.raid_disks
;
9559 /* All raid5 and raid0 volumes in container
9560 * have to be ready for Online Capacity Expansion
9561 * so they need to be assembled. We have already
9562 * checked that no recovery etc is happening.
9564 result
= imsm_find_array_minor_by_subdev(member
->container_member
,
9568 dprintf("imsm: cannot find array\n");
9571 devices_that_can_grow
++;
9574 if (!member
&& devices_that_can_grow
)
9578 dprintf("\tContainer operation allowed\n");
9580 dprintf("\tError: %i\n", ret_val
);
9585 /* Function: get_spares_for_grow
9586 * Description: Allocates memory and creates list of spare devices
9587 * avaliable in container. Checks if spare drive size is acceptable.
9588 * Parameters: Pointer to the supertype structure
9589 * Returns: Pointer to the list of spare devices (mdinfo structure) on success,
9592 static struct mdinfo
*get_spares_for_grow(struct supertype
*st
)
9594 unsigned long long min_size
= min_acceptable_spare_size_imsm(st
);
9595 return container_choose_spares(st
, min_size
, NULL
, NULL
, NULL
, 0);
9598 /******************************************************************************
9599 * function: imsm_create_metadata_update_for_reshape
9600 * Function creates update for whole IMSM container.
9602 ******************************************************************************/
9603 static int imsm_create_metadata_update_for_reshape(
9604 struct supertype
*st
,
9605 struct geo_params
*geo
,
9607 struct imsm_update_reshape
**updatep
)
9609 struct intel_super
*super
= st
->sb
;
9610 struct imsm_super
*mpb
= super
->anchor
;
9611 int update_memory_size
= 0;
9612 struct imsm_update_reshape
*u
= NULL
;
9613 struct mdinfo
*spares
= NULL
;
9615 int delta_disks
= 0;
9618 dprintf("imsm_update_metadata_for_reshape(enter) raid_disks = %i\n",
9621 delta_disks
= geo
->raid_disks
- old_raid_disks
;
9623 /* size of all update data without anchor */
9624 update_memory_size
= sizeof(struct imsm_update_reshape
);
9626 /* now add space for spare disks that we need to add. */
9627 update_memory_size
+= sizeof(u
->new_disks
[0]) * (delta_disks
- 1);
9629 u
= calloc(1, update_memory_size
);
9632 "cannot get memory for imsm_update_reshape update\n");
9635 u
->type
= update_reshape_container_disks
;
9636 u
->old_raid_disks
= old_raid_disks
;
9637 u
->new_raid_disks
= geo
->raid_disks
;
9639 /* now get spare disks list
9641 spares
= get_spares_for_grow(st
);
9644 || delta_disks
> spares
->array
.spare_disks
) {
9645 fprintf(stderr
, Name
": imsm: ERROR: Cannot get spare devices "
9646 "for %s.\n", geo
->dev_name
);
9651 /* we have got spares
9652 * update disk list in imsm_disk list table in anchor
9654 dprintf("imsm: %i spares are available.\n\n",
9655 spares
->array
.spare_disks
);
9658 for (i
= 0; i
< delta_disks
; i
++) {
9663 u
->new_disks
[i
] = makedev(dev
->disk
.major
,
9665 dl
= get_disk_super(super
, dev
->disk
.major
, dev
->disk
.minor
);
9666 dl
->index
= mpb
->num_disks
;
9676 dprintf("imsm: reshape update preparation :");
9677 if (i
== delta_disks
) {
9680 return update_memory_size
;
9683 dprintf(" Error\n");
9689 /******************************************************************************
9690 * function: imsm_create_metadata_update_for_size_change()
9691 * Creates update for IMSM array for array size change.
9693 ******************************************************************************/
9694 static int imsm_create_metadata_update_for_size_change(
9695 struct supertype
*st
,
9696 struct geo_params
*geo
,
9697 struct imsm_update_size_change
**updatep
)
9699 struct intel_super
*super
= st
->sb
;
9700 int update_memory_size
= 0;
9701 struct imsm_update_size_change
*u
= NULL
;
9703 dprintf("imsm_create_metadata_update_for_size_change(enter)"
9704 " New size = %llu\n", geo
->size
);
9706 /* size of all update data without anchor */
9707 update_memory_size
= sizeof(struct imsm_update_size_change
);
9709 u
= calloc(1, update_memory_size
);
9711 dprintf("error: cannot get memory for "
9712 "imsm_create_metadata_update_for_size_change\n");
9715 u
->type
= update_size_change
;
9716 u
->subdev
= super
->current_vol
;
9717 u
->new_size
= geo
->size
;
9719 dprintf("imsm: reshape update preparation : OK\n");
9722 return update_memory_size
;
9725 /******************************************************************************
9726 * function: imsm_create_metadata_update_for_migration()
9727 * Creates update for IMSM array.
9729 ******************************************************************************/
9730 static int imsm_create_metadata_update_for_migration(
9731 struct supertype
*st
,
9732 struct geo_params
*geo
,
9733 struct imsm_update_reshape_migration
**updatep
)
9735 struct intel_super
*super
= st
->sb
;
9736 int update_memory_size
= 0;
9737 struct imsm_update_reshape_migration
*u
= NULL
;
9738 struct imsm_dev
*dev
;
9739 int previous_level
= -1;
9741 dprintf("imsm_create_metadata_update_for_migration(enter)"
9742 " New Level = %i\n", geo
->level
);
9744 /* size of all update data without anchor */
9745 update_memory_size
= sizeof(struct imsm_update_reshape_migration
);
9747 u
= calloc(1, update_memory_size
);
9749 dprintf("error: cannot get memory for "
9750 "imsm_create_metadata_update_for_migration\n");
9753 u
->type
= update_reshape_migration
;
9754 u
->subdev
= super
->current_vol
;
9755 u
->new_level
= geo
->level
;
9756 u
->new_layout
= geo
->layout
;
9757 u
->new_raid_disks
= u
->old_raid_disks
= geo
->raid_disks
;
9758 u
->new_disks
[0] = -1;
9759 u
->new_chunksize
= -1;
9761 dev
= get_imsm_dev(super
, u
->subdev
);
9763 struct imsm_map
*map
;
9765 map
= get_imsm_map(dev
, MAP_0
);
9767 int current_chunk_size
=
9768 __le16_to_cpu(map
->blocks_per_strip
) / 2;
9770 if (geo
->chunksize
!= current_chunk_size
) {
9771 u
->new_chunksize
= geo
->chunksize
/ 1024;
9773 "chunk size change from %i to %i\n",
9774 current_chunk_size
, u
->new_chunksize
);
9776 previous_level
= map
->raid_level
;
9779 if ((geo
->level
== 5) && (previous_level
== 0)) {
9780 struct mdinfo
*spares
= NULL
;
9782 u
->new_raid_disks
++;
9783 spares
= get_spares_for_grow(st
);
9784 if ((spares
== NULL
) || (spares
->array
.spare_disks
< 1)) {
9787 update_memory_size
= 0;
9788 dprintf("error: cannot get spare device "
9789 "for requested migration");
9794 dprintf("imsm: reshape update preparation : OK\n");
9797 return update_memory_size
;
9800 static void imsm_update_metadata_locally(struct supertype
*st
,
9803 struct metadata_update mu
;
9808 mu
.space_list
= NULL
;
9810 imsm_prepare_update(st
, &mu
);
9811 imsm_process_update(st
, &mu
);
9813 while (mu
.space_list
) {
9814 void **space
= mu
.space_list
;
9815 mu
.space_list
= *space
;
9820 /***************************************************************************
9821 * Function: imsm_analyze_change
9822 * Description: Function analyze change for single volume
9823 * and validate if transition is supported
9824 * Parameters: Geometry parameters, supertype structure,
9825 * metadata change direction (apply/rollback)
9826 * Returns: Operation type code on success, -1 if fail
9827 ****************************************************************************/
9828 enum imsm_reshape_type
imsm_analyze_change(struct supertype
*st
,
9829 struct geo_params
*geo
,
9836 /* number of added/removed disks in operation result */
9837 int devNumChange
= 0;
9838 /* imsm compatible layout value for array geometry verification */
9839 int imsm_layout
= -1;
9841 struct imsm_dev
*dev
;
9842 struct intel_super
*super
;
9843 long long current_size
;
9845 getinfo_super_imsm_volume(st
, &info
, NULL
);
9846 if ((geo
->level
!= info
.array
.level
) &&
9847 (geo
->level
>= 0) &&
9848 (geo
->level
!= UnSet
)) {
9849 switch (info
.array
.level
) {
9851 if (geo
->level
== 5) {
9852 change
= CH_MIGRATION
;
9853 if (geo
->layout
!= ALGORITHM_LEFT_ASYMMETRIC
) {
9855 Name
" Error. Requested Layout "
9856 "not supported (left-asymmetric layout "
9857 "is supported only)!\n");
9859 goto analyse_change_exit
;
9861 imsm_layout
= geo
->layout
;
9863 devNumChange
= 1; /* parity disk added */
9864 } else if (geo
->level
== 10) {
9865 change
= CH_TAKEOVER
;
9867 devNumChange
= 2; /* two mirrors added */
9868 imsm_layout
= 0x102; /* imsm supported layout */
9873 if (geo
->level
== 0) {
9874 change
= CH_TAKEOVER
;
9876 devNumChange
= -(geo
->raid_disks
/2);
9877 imsm_layout
= 0; /* imsm raid0 layout */
9883 Name
" Error. Level Migration from %d to %d "
9885 info
.array
.level
, geo
->level
);
9886 goto analyse_change_exit
;
9889 geo
->level
= info
.array
.level
;
9891 if ((geo
->layout
!= info
.array
.layout
)
9892 && ((geo
->layout
!= UnSet
) && (geo
->layout
!= -1))) {
9893 change
= CH_MIGRATION
;
9894 if ((info
.array
.layout
== 0)
9895 && (info
.array
.level
== 5)
9896 && (geo
->layout
== 5)) {
9897 /* reshape 5 -> 4 */
9898 } else if ((info
.array
.layout
== 5)
9899 && (info
.array
.level
== 5)
9900 && (geo
->layout
== 0)) {
9901 /* reshape 4 -> 5 */
9906 Name
" Error. Layout Migration from %d to %d "
9908 info
.array
.layout
, geo
->layout
);
9910 goto analyse_change_exit
;
9913 geo
->layout
= info
.array
.layout
;
9914 if (imsm_layout
== -1)
9915 imsm_layout
= info
.array
.layout
;
9918 if ((geo
->chunksize
> 0) && (geo
->chunksize
!= UnSet
)
9919 && (geo
->chunksize
!= info
.array
.chunk_size
))
9920 change
= CH_MIGRATION
;
9922 geo
->chunksize
= info
.array
.chunk_size
;
9924 chunk
= geo
->chunksize
/ 1024;
9927 dev
= get_imsm_dev(super
, super
->current_vol
);
9928 data_disks
= imsm_num_data_members(dev
, MAP_0
);
9929 /* compute current size in K per disk member
9931 current_size
= info
.custom_array_size
/ 2 / data_disks
;
9933 if ((current_size
!= geo
->size
) && (geo
->size
> 0)) {
9936 Name
" Error. Size change should be the only "
9937 "one at a time.\n");
9939 goto analyse_change_exit
;
9941 if ((super
->current_vol
+ 1) != super
->anchor
->num_raid_devs
) {
9943 Name
" Error. The last volume in container "
9944 "can be expanded only (%i/%i).\n",
9945 super
->current_vol
, st
->devnum
);
9946 goto analyse_change_exit
;
9949 if ((direction
== ROLLBACK_METADATA_CHANGES
)) {
9950 /* accept size for rollback only
9953 /* round size due to metadata compatibility
9955 geo
->size
= (geo
->size
>> SECT_PER_MB_SHIFT
)
9956 << SECT_PER_MB_SHIFT
;
9957 dprintf("Prepare update for size change to %llu\n",
9959 if (current_size
>= geo
->size
) {
9961 Name
" Error. Size expanssion is "
9962 "supported only (current size is %llu, "
9963 "requested size /rounded/ is %llu).\n",
9964 current_size
, geo
->size
);
9965 goto analyse_change_exit
;
9968 geo
->size
*= data_disks
;
9969 geo
->raid_disks
= dev
->vol
.map
->num_members
;
9970 change
= CH_ARRAY_SIZE
;
9972 if (!validate_geometry_imsm(st
,
9975 geo
->raid_disks
+ devNumChange
,
9982 struct intel_super
*super
= st
->sb
;
9983 struct imsm_super
*mpb
= super
->anchor
;
9985 if (mpb
->num_raid_devs
> 1) {
9987 Name
" Error. Cannot perform operation on %s"
9988 "- for this operation it MUST be single "
9989 "array in container\n",
9995 analyse_change_exit
:
9996 if ((direction
== ROLLBACK_METADATA_CHANGES
) &&
9997 ((change
== CH_MIGRATION
) || (change
== CH_TAKEOVER
))) {
9998 dprintf("imsm: Metadata changes rollback is not supported for "
9999 "migration and takeover operations.\n");
10005 int imsm_takeover(struct supertype
*st
, struct geo_params
*geo
)
10007 struct intel_super
*super
= st
->sb
;
10008 struct imsm_update_takeover
*u
;
10010 u
= malloc(sizeof(struct imsm_update_takeover
));
10014 u
->type
= update_takeover
;
10015 u
->subarray
= super
->current_vol
;
10017 /* 10->0 transition */
10018 if (geo
->level
== 0)
10019 u
->direction
= R10_TO_R0
;
10021 /* 0->10 transition */
10022 if (geo
->level
== 10)
10023 u
->direction
= R0_TO_R10
;
10025 /* update metadata locally */
10026 imsm_update_metadata_locally(st
, u
,
10027 sizeof(struct imsm_update_takeover
));
10028 /* and possibly remotely */
10029 if (st
->update_tail
)
10030 append_metadata_update(st
, u
,
10031 sizeof(struct imsm_update_takeover
));
10038 static int imsm_reshape_super(struct supertype
*st
, long long size
, int level
,
10039 int layout
, int chunksize
, int raid_disks
,
10040 int delta_disks
, char *backup
, char *dev
,
10041 int direction
, int verbose
)
10044 struct geo_params geo
;
10046 dprintf("imsm: reshape_super called.\n");
10048 memset(&geo
, 0, sizeof(struct geo_params
));
10050 geo
.dev_name
= dev
;
10051 geo
.dev_id
= st
->devnum
;
10054 geo
.layout
= layout
;
10055 geo
.chunksize
= chunksize
;
10056 geo
.raid_disks
= raid_disks
;
10057 if (delta_disks
!= UnSet
)
10058 geo
.raid_disks
+= delta_disks
;
10060 dprintf("\tfor level : %i\n", geo
.level
);
10061 dprintf("\tfor raid_disks : %i\n", geo
.raid_disks
);
10063 if (experimental() == 0)
10066 if (st
->container_dev
== st
->devnum
) {
10067 /* On container level we can only increase number of devices. */
10068 dprintf("imsm: info: Container operation\n");
10069 int old_raid_disks
= 0;
10071 if (imsm_reshape_is_allowed_on_container(
10072 st
, &geo
, &old_raid_disks
, direction
)) {
10073 struct imsm_update_reshape
*u
= NULL
;
10076 len
= imsm_create_metadata_update_for_reshape(
10077 st
, &geo
, old_raid_disks
, &u
);
10080 dprintf("imsm: Cannot prepare update\n");
10081 goto exit_imsm_reshape_super
;
10085 /* update metadata locally */
10086 imsm_update_metadata_locally(st
, u
, len
);
10087 /* and possibly remotely */
10088 if (st
->update_tail
)
10089 append_metadata_update(st
, u
, len
);
10094 fprintf(stderr
, Name
": (imsm) Operation "
10095 "is not allowed on this container\n");
10098 /* On volume level we support following operations
10099 * - takeover: raid10 -> raid0; raid0 -> raid10
10100 * - chunk size migration
10101 * - migration: raid5 -> raid0; raid0 -> raid5
10103 struct intel_super
*super
= st
->sb
;
10104 struct intel_dev
*dev
= super
->devlist
;
10105 int change
, devnum
;
10106 dprintf("imsm: info: Volume operation\n");
10107 /* find requested device */
10109 if (imsm_find_array_minor_by_subdev(
10110 dev
->index
, st
->container_dev
, &devnum
) == 0
10111 && devnum
== geo
.dev_id
)
10116 fprintf(stderr
, Name
" Cannot find %s (%i) subarray\n",
10117 geo
.dev_name
, geo
.dev_id
);
10118 goto exit_imsm_reshape_super
;
10120 super
->current_vol
= dev
->index
;
10121 change
= imsm_analyze_change(st
, &geo
, direction
);
10124 ret_val
= imsm_takeover(st
, &geo
);
10126 case CH_MIGRATION
: {
10127 struct imsm_update_reshape_migration
*u
= NULL
;
10129 imsm_create_metadata_update_for_migration(
10133 "Cannot prepare update\n");
10137 /* update metadata locally */
10138 imsm_update_metadata_locally(st
, u
, len
);
10139 /* and possibly remotely */
10140 if (st
->update_tail
)
10141 append_metadata_update(st
, u
, len
);
10146 case CH_ARRAY_SIZE
: {
10147 struct imsm_update_size_change
*u
= NULL
;
10149 imsm_create_metadata_update_for_size_change(
10153 "Cannot prepare update\n");
10157 /* update metadata locally */
10158 imsm_update_metadata_locally(st
, u
, len
);
10159 /* and possibly remotely */
10160 if (st
->update_tail
)
10161 append_metadata_update(st
, u
, len
);
10171 exit_imsm_reshape_super
:
10172 dprintf("imsm: reshape_super Exit code = %i\n", ret_val
);
10176 /*******************************************************************************
10177 * Function: wait_for_reshape_imsm
10178 * Description: Function writes new sync_max value and waits until
10179 * reshape process reach new position
10181 * sra : general array info
10182 * ndata : number of disks in new array's layout
10185 * 1 : there is no reshape in progress,
10187 ******************************************************************************/
10188 int wait_for_reshape_imsm(struct mdinfo
*sra
, int ndata
)
10190 int fd
= sysfs_get_fd(sra
, NULL
, "reshape_position");
10191 unsigned long long completed
;
10192 /* to_complete : new sync_max position */
10193 unsigned long long to_complete
= sra
->reshape_progress
;
10194 unsigned long long position_to_set
= to_complete
/ ndata
;
10197 dprintf("imsm: wait_for_reshape_imsm() "
10198 "cannot open reshape_position\n");
10202 if (sysfs_fd_get_ll(fd
, &completed
) < 0) {
10203 dprintf("imsm: wait_for_reshape_imsm() "
10204 "cannot read reshape_position (no reshape in progres)\n");
10209 if (completed
> to_complete
) {
10210 dprintf("imsm: wait_for_reshape_imsm() "
10211 "wrong next position to set %llu (%llu)\n",
10212 to_complete
, completed
);
10216 dprintf("Position set: %llu\n", position_to_set
);
10217 if (sysfs_set_num(sra
, NULL
, "sync_max",
10218 position_to_set
) != 0) {
10219 dprintf("imsm: wait_for_reshape_imsm() "
10220 "cannot set reshape position to %llu\n",
10231 select(fd
+1, &rfds
, NULL
, NULL
, NULL
);
10232 if (sysfs_get_str(sra
, NULL
, "sync_action",
10234 strncmp(action
, "reshape", 7) != 0)
10236 if (sysfs_fd_get_ll(fd
, &completed
) < 0) {
10237 dprintf("imsm: wait_for_reshape_imsm() "
10238 "cannot read reshape_position (in loop)\n");
10242 } while (completed
< to_complete
);
10248 /*******************************************************************************
10249 * Function: check_degradation_change
10250 * Description: Check that array hasn't become failed.
10252 * info : for sysfs access
10253 * sources : source disks descriptors
10254 * degraded: previous degradation level
10256 * degradation level
10257 ******************************************************************************/
10258 int check_degradation_change(struct mdinfo
*info
,
10262 unsigned long long new_degraded
;
10263 sysfs_get_ll(info
, NULL
, "degraded", &new_degraded
);
10264 if (new_degraded
!= (unsigned long long)degraded
) {
10265 /* check each device to ensure it is still working */
10268 for (sd
= info
->devs
; sd
; sd
= sd
->next
) {
10269 if (sd
->disk
.state
& (1<<MD_DISK_FAULTY
))
10271 if (sd
->disk
.state
& (1<<MD_DISK_SYNC
)) {
10273 if (sysfs_get_str(info
,
10274 sd
, "state", sbuf
, 20) < 0 ||
10275 strstr(sbuf
, "faulty") ||
10276 strstr(sbuf
, "in_sync") == NULL
) {
10277 /* this device is dead */
10278 sd
->disk
.state
= (1<<MD_DISK_FAULTY
);
10279 if (sd
->disk
.raid_disk
>= 0 &&
10280 sources
[sd
->disk
.raid_disk
] >= 0) {
10282 sd
->disk
.raid_disk
]);
10283 sources
[sd
->disk
.raid_disk
] =
10292 return new_degraded
;
10295 /*******************************************************************************
10296 * Function: imsm_manage_reshape
10297 * Description: Function finds array under reshape and it manages reshape
10298 * process. It creates stripes backups (if required) and sets
10301 * afd : Backup handle (nattive) - not used
10302 * sra : general array info
10303 * reshape : reshape parameters - not used
10304 * st : supertype structure
10305 * blocks : size of critical section [blocks]
10306 * fds : table of source device descriptor
10307 * offsets : start of array (offest per devices)
10309 * destfd : table of destination device descriptor
10310 * destoffsets : table of destination offsets (per device)
10312 * 1 : success, reshape is done
10314 ******************************************************************************/
10315 static int imsm_manage_reshape(
10316 int afd
, struct mdinfo
*sra
, struct reshape
*reshape
,
10317 struct supertype
*st
, unsigned long backup_blocks
,
10318 int *fds
, unsigned long long *offsets
,
10319 int dests
, int *destfd
, unsigned long long *destoffsets
)
10322 struct intel_super
*super
= st
->sb
;
10323 struct intel_dev
*dv
= NULL
;
10324 struct imsm_dev
*dev
= NULL
;
10325 struct imsm_map
*map_src
;
10326 int migr_vol_qan
= 0;
10327 int ndata
, odata
; /* [bytes] */
10328 int chunk
; /* [bytes] */
10329 struct migr_record
*migr_rec
;
10331 unsigned int buf_size
; /* [bytes] */
10332 unsigned long long max_position
; /* array size [bytes] */
10333 unsigned long long next_step
; /* [blocks]/[bytes] */
10334 unsigned long long old_data_stripe_length
;
10335 unsigned long long start_src
; /* [bytes] */
10336 unsigned long long start
; /* [bytes] */
10337 unsigned long long start_buf_shift
; /* [bytes] */
10339 int source_layout
= 0;
10341 if (!fds
|| !offsets
|| !sra
)
10344 /* Find volume during the reshape */
10345 for (dv
= super
->devlist
; dv
; dv
= dv
->next
) {
10346 if (dv
->dev
->vol
.migr_type
== MIGR_GEN_MIGR
10347 && dv
->dev
->vol
.migr_state
== 1) {
10352 /* Only one volume can migrate at the same time */
10353 if (migr_vol_qan
!= 1) {
10354 fprintf(stderr
, Name
" : %s", migr_vol_qan
?
10355 "Number of migrating volumes greater than 1\n" :
10356 "There is no volume during migrationg\n");
10360 map_src
= get_imsm_map(dev
, MAP_1
);
10361 if (map_src
== NULL
)
10364 ndata
= imsm_num_data_members(dev
, MAP_0
);
10365 odata
= imsm_num_data_members(dev
, MAP_1
);
10367 chunk
= __le16_to_cpu(map_src
->blocks_per_strip
) * 512;
10368 old_data_stripe_length
= odata
* chunk
;
10370 migr_rec
= super
->migr_rec
;
10372 /* initialize migration record for start condition */
10373 if (sra
->reshape_progress
== 0)
10374 init_migr_record_imsm(st
, dev
, sra
);
10376 if (__le32_to_cpu(migr_rec
->rec_status
) != UNIT_SRC_NORMAL
) {
10377 dprintf("imsm: cannot restart migration when data "
10378 "are present in copy area.\n");
10381 /* Save checkpoint to update migration record for current
10382 * reshape position (in md). It can be farther than current
10383 * reshape position in metadata.
10385 if (save_checkpoint_imsm(st
, sra
, UNIT_SRC_NORMAL
) == 1) {
10386 /* ignore error == 2, this can mean end of reshape here
10388 dprintf("imsm: Cannot write checkpoint to "
10389 "migration record (UNIT_SRC_NORMAL, "
10390 "initial save)\n");
10395 /* size for data */
10396 buf_size
= __le32_to_cpu(migr_rec
->blocks_per_unit
) * 512;
10397 /* extend buffer size for parity disk */
10398 buf_size
+= __le32_to_cpu(migr_rec
->dest_depth_per_unit
) * 512;
10399 /* add space for stripe aligment */
10400 buf_size
+= old_data_stripe_length
;
10401 if (posix_memalign((void **)&buf
, 4096, buf_size
)) {
10402 dprintf("imsm: Cannot allocate checpoint buffer\n");
10406 max_position
= sra
->component_size
* ndata
;
10407 source_layout
= imsm_level_to_layout(map_src
->raid_level
);
10409 while (__le32_to_cpu(migr_rec
->curr_migr_unit
) <
10410 __le32_to_cpu(migr_rec
->num_migr_units
)) {
10411 /* current reshape position [blocks] */
10412 unsigned long long current_position
=
10413 __le32_to_cpu(migr_rec
->blocks_per_unit
)
10414 * __le32_to_cpu(migr_rec
->curr_migr_unit
);
10415 unsigned long long border
;
10417 /* Check that array hasn't become failed.
10419 degraded
= check_degradation_change(sra
, fds
, degraded
);
10420 if (degraded
> 1) {
10421 dprintf("imsm: Abort reshape due to degradation"
10422 " level (%i)\n", degraded
);
10426 next_step
= __le32_to_cpu(migr_rec
->blocks_per_unit
);
10428 if ((current_position
+ next_step
) > max_position
)
10429 next_step
= max_position
- current_position
;
10431 start
= current_position
* 512;
10433 /* allign reading start to old geometry */
10434 start_buf_shift
= start
% old_data_stripe_length
;
10435 start_src
= start
- start_buf_shift
;
10437 border
= (start_src
/ odata
) - (start
/ ndata
);
10439 if (border
<= __le32_to_cpu(migr_rec
->dest_depth_per_unit
)) {
10440 /* save critical stripes to buf
10441 * start - start address of current unit
10442 * to backup [bytes]
10443 * start_src - start address of current unit
10444 * to backup alligned to source array
10447 unsigned long long next_step_filler
= 0;
10448 unsigned long long copy_length
= next_step
* 512;
10450 /* allign copy area length to stripe in old geometry */
10451 next_step_filler
= ((copy_length
+ start_buf_shift
)
10452 % old_data_stripe_length
);
10453 if (next_step_filler
)
10454 next_step_filler
= (old_data_stripe_length
10455 - next_step_filler
);
10456 dprintf("save_stripes() parameters: start = %llu,"
10457 "\tstart_src = %llu,\tnext_step*512 = %llu,"
10458 "\tstart_in_buf_shift = %llu,"
10459 "\tnext_step_filler = %llu\n",
10460 start
, start_src
, copy_length
,
10461 start_buf_shift
, next_step_filler
);
10463 if (save_stripes(fds
, offsets
, map_src
->num_members
,
10464 chunk
, map_src
->raid_level
,
10465 source_layout
, 0, NULL
, start_src
,
10467 next_step_filler
+ start_buf_shift
,
10469 dprintf("imsm: Cannot save stripes"
10473 /* Convert data to destination format and store it
10474 * in backup general migration area
10476 if (save_backup_imsm(st
, dev
, sra
,
10477 buf
+ start_buf_shift
, copy_length
)) {
10478 dprintf("imsm: Cannot save stripes to "
10479 "target devices\n");
10482 if (save_checkpoint_imsm(st
, sra
,
10483 UNIT_SRC_IN_CP_AREA
)) {
10484 dprintf("imsm: Cannot write checkpoint to "
10485 "migration record (UNIT_SRC_IN_CP_AREA)\n");
10489 /* set next step to use whole border area */
10490 border
/= next_step
;
10492 next_step
*= border
;
10494 /* When data backed up, checkpoint stored,
10495 * kick the kernel to reshape unit of data
10497 next_step
= next_step
+ sra
->reshape_progress
;
10498 /* limit next step to array max position */
10499 if (next_step
> max_position
)
10500 next_step
= max_position
;
10501 sysfs_set_num(sra
, NULL
, "suspend_lo", sra
->reshape_progress
);
10502 sysfs_set_num(sra
, NULL
, "suspend_hi", next_step
);
10503 sra
->reshape_progress
= next_step
;
10505 /* wait until reshape finish */
10506 if (wait_for_reshape_imsm(sra
, ndata
) < 0) {
10507 dprintf("wait_for_reshape_imsm returned error!\n");
10511 if (save_checkpoint_imsm(st
, sra
, UNIT_SRC_NORMAL
) == 1) {
10512 /* ignore error == 2, this can mean end of reshape here
10514 dprintf("imsm: Cannot write checkpoint to "
10515 "migration record (UNIT_SRC_NORMAL)\n");
10521 /* return '1' if done */
10525 abort_reshape(sra
);
10529 #endif /* MDASSEMBLE */
10531 struct superswitch super_imsm
= {
10533 .examine_super
= examine_super_imsm
,
10534 .brief_examine_super
= brief_examine_super_imsm
,
10535 .brief_examine_subarrays
= brief_examine_subarrays_imsm
,
10536 .export_examine_super
= export_examine_super_imsm
,
10537 .detail_super
= detail_super_imsm
,
10538 .brief_detail_super
= brief_detail_super_imsm
,
10539 .write_init_super
= write_init_super_imsm
,
10540 .validate_geometry
= validate_geometry_imsm
,
10541 .add_to_super
= add_to_super_imsm
,
10542 .remove_from_super
= remove_from_super_imsm
,
10543 .detail_platform
= detail_platform_imsm
,
10544 .kill_subarray
= kill_subarray_imsm
,
10545 .update_subarray
= update_subarray_imsm
,
10546 .load_container
= load_container_imsm
,
10547 .default_geometry
= default_geometry_imsm
,
10548 .get_disk_controller_domain
= imsm_get_disk_controller_domain
,
10549 .reshape_super
= imsm_reshape_super
,
10550 .manage_reshape
= imsm_manage_reshape
,
10551 .recover_backup
= recover_backup_imsm
,
10553 .match_home
= match_home_imsm
,
10554 .uuid_from_super
= uuid_from_super_imsm
,
10555 .getinfo_super
= getinfo_super_imsm
,
10556 .getinfo_super_disks
= getinfo_super_disks_imsm
,
10557 .update_super
= update_super_imsm
,
10559 .avail_size
= avail_size_imsm
,
10560 .min_acceptable_spare_size
= min_acceptable_spare_size_imsm
,
10562 .compare_super
= compare_super_imsm
,
10564 .load_super
= load_super_imsm
,
10565 .init_super
= init_super_imsm
,
10566 .store_super
= store_super_imsm
,
10567 .free_super
= free_super_imsm
,
10568 .match_metadata_desc
= match_metadata_desc_imsm
,
10569 .container_content
= container_content_imsm
,
10577 .open_new
= imsm_open_new
,
10578 .set_array_state
= imsm_set_array_state
,
10579 .set_disk
= imsm_set_disk
,
10580 .sync_metadata
= imsm_sync_metadata
,
10581 .activate_spare
= imsm_activate_spare
,
10582 .process_update
= imsm_process_update
,
10583 .prepare_update
= imsm_prepare_update
,
10584 #endif /* MDASSEMBLE */