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
{
410 /* definition of messages passed to imsm_process_update */
411 enum imsm_update_type
{
412 update_activate_spare
,
416 update_add_remove_disk
,
417 update_reshape_container_disks
,
418 update_reshape_migration
,
420 update_general_migration_checkpoint
,
423 struct imsm_update_activate_spare
{
424 enum imsm_update_type type
;
428 struct imsm_update_activate_spare
*next
;
441 enum takeover_direction
{
445 struct imsm_update_takeover
{
446 enum imsm_update_type type
;
448 enum takeover_direction direction
;
451 struct imsm_update_reshape
{
452 enum imsm_update_type type
;
456 int new_disks
[1]; /* new_raid_disks - old_raid_disks makedev number */
459 struct imsm_update_reshape_migration
{
460 enum imsm_update_type type
;
463 /* fields for array migration changes
470 int new_disks
[1]; /* new_raid_disks - old_raid_disks makedev number */
473 struct imsm_update_general_migration_checkpoint
{
474 enum imsm_update_type type
;
475 __u32 curr_migr_unit
;
479 __u8 serial
[MAX_RAID_SERIAL_LEN
];
482 struct imsm_update_create_array
{
483 enum imsm_update_type type
;
488 struct imsm_update_kill_array
{
489 enum imsm_update_type type
;
493 struct imsm_update_rename_array
{
494 enum imsm_update_type type
;
495 __u8 name
[MAX_RAID_SERIAL_LEN
];
499 struct imsm_update_add_remove_disk
{
500 enum imsm_update_type type
;
504 static const char *_sys_dev_type
[] = {
505 [SYS_DEV_UNKNOWN
] = "Unknown",
506 [SYS_DEV_SAS
] = "SAS",
507 [SYS_DEV_SATA
] = "SATA"
510 const char *get_sys_dev_type(enum sys_dev_type type
)
512 if (type
>= SYS_DEV_MAX
)
513 type
= SYS_DEV_UNKNOWN
;
515 return _sys_dev_type
[type
];
518 static struct intel_hba
* alloc_intel_hba(struct sys_dev
*device
)
520 struct intel_hba
*result
= malloc(sizeof(*result
));
522 result
->type
= device
->type
;
523 result
->path
= strdup(device
->path
);
525 if (result
->path
&& (result
->pci_id
= strrchr(result
->path
, '/')) != NULL
)
531 static struct intel_hba
* find_intel_hba(struct intel_hba
*hba
, struct sys_dev
*device
)
533 struct intel_hba
*result
=NULL
;
534 for (result
= hba
; result
; result
= result
->next
) {
535 if (result
->type
== device
->type
&& strcmp(result
->path
, device
->path
) == 0)
541 static int attach_hba_to_super(struct intel_super
*super
, struct sys_dev
*device
)
543 struct intel_hba
*hba
;
545 /* check if disk attached to Intel HBA */
546 hba
= find_intel_hba(super
->hba
, device
);
549 /* Check if HBA is already attached to super */
550 if (super
->hba
== NULL
) {
551 super
->hba
= alloc_intel_hba(device
);
556 /* Intel metadata allows for all disks attached to the same type HBA.
557 * Do not sypport odf HBA types mixing
559 if (device
->type
!= hba
->type
)
565 hba
->next
= alloc_intel_hba(device
);
569 static struct sys_dev
* find_disk_attached_hba(int fd
, const char *devname
)
571 struct sys_dev
*list
, *elem
, *prev
;
574 if ((list
= find_intel_devices()) == NULL
)
578 disk_path
= (char *) devname
;
580 disk_path
= diskfd_to_devpath(fd
);
587 for (prev
= NULL
, elem
= list
; elem
; prev
= elem
, elem
= elem
->next
) {
588 if (path_attached_to_hba(disk_path
, elem
->path
)) {
592 prev
->next
= elem
->next
;
594 if (disk_path
!= devname
)
600 if (disk_path
!= devname
)
608 static int find_intel_hba_capability(int fd
, struct intel_super
*super
,
611 static struct supertype
*match_metadata_desc_imsm(char *arg
)
613 struct supertype
*st
;
615 if (strcmp(arg
, "imsm") != 0 &&
616 strcmp(arg
, "default") != 0
620 st
= malloc(sizeof(*st
));
623 memset(st
, 0, sizeof(*st
));
624 st
->container_dev
= NoMdDev
;
625 st
->ss
= &super_imsm
;
626 st
->max_devs
= IMSM_MAX_DEVICES
;
627 st
->minor_version
= 0;
633 static __u8
*get_imsm_version(struct imsm_super
*mpb
)
635 return &mpb
->sig
[MPB_SIG_LEN
];
639 /* retrieve a disk directly from the anchor when the anchor is known to be
640 * up-to-date, currently only at load time
642 static struct imsm_disk
*__get_imsm_disk(struct imsm_super
*mpb
, __u8 index
)
644 if (index
>= mpb
->num_disks
)
646 return &mpb
->disk
[index
];
649 /* retrieve the disk description based on a index of the disk
652 static struct dl
*get_imsm_dl_disk(struct intel_super
*super
, __u8 index
)
656 for (d
= super
->disks
; d
; d
= d
->next
)
657 if (d
->index
== index
)
662 /* retrieve a disk from the parsed metadata */
663 static struct imsm_disk
*get_imsm_disk(struct intel_super
*super
, __u8 index
)
667 dl
= get_imsm_dl_disk(super
, index
);
674 /* generate a checksum directly from the anchor when the anchor is known to be
675 * up-to-date, currently only at load or write_super after coalescing
677 static __u32
__gen_imsm_checksum(struct imsm_super
*mpb
)
679 __u32 end
= mpb
->mpb_size
/ sizeof(end
);
680 __u32
*p
= (__u32
*) mpb
;
684 sum
+= __le32_to_cpu(*p
);
688 return sum
- __le32_to_cpu(mpb
->check_sum
);
691 static size_t sizeof_imsm_map(struct imsm_map
*map
)
693 return sizeof(struct imsm_map
) + sizeof(__u32
) * (map
->num_members
- 1);
696 struct imsm_map
*get_imsm_map(struct imsm_dev
*dev
, int second_map
)
698 /* A device can have 2 maps if it is in the middle of a migration.
700 * MAP_0 - we return the first map
701 * MAP_1 - we return the second map if it exists, else NULL
702 * MAP_X - we return the second map if it exists, else the first
704 struct imsm_map
*map
= &dev
->vol
.map
[0];
705 struct imsm_map
*map2
= NULL
;
707 if (dev
->vol
.migr_state
)
708 map2
= (void *)map
+ sizeof_imsm_map(map
);
710 switch (second_map
) {
727 /* return the size of the device.
728 * migr_state increases the returned size if map[0] were to be duplicated
730 static size_t sizeof_imsm_dev(struct imsm_dev
*dev
, int migr_state
)
732 size_t size
= sizeof(*dev
) - sizeof(struct imsm_map
) +
733 sizeof_imsm_map(get_imsm_map(dev
, MAP_0
));
735 /* migrating means an additional map */
736 if (dev
->vol
.migr_state
)
737 size
+= sizeof_imsm_map(get_imsm_map(dev
, MAP_1
));
739 size
+= sizeof_imsm_map(get_imsm_map(dev
, MAP_0
));
745 /* retrieve disk serial number list from a metadata update */
746 static struct disk_info
*get_disk_info(struct imsm_update_create_array
*update
)
749 struct disk_info
*inf
;
751 inf
= u
+ sizeof(*update
) - sizeof(struct imsm_dev
) +
752 sizeof_imsm_dev(&update
->dev
, 0);
758 static struct imsm_dev
*__get_imsm_dev(struct imsm_super
*mpb
, __u8 index
)
764 if (index
>= mpb
->num_raid_devs
)
767 /* devices start after all disks */
768 offset
= ((void *) &mpb
->disk
[mpb
->num_disks
]) - _mpb
;
770 for (i
= 0; i
<= index
; i
++)
772 return _mpb
+ offset
;
774 offset
+= sizeof_imsm_dev(_mpb
+ offset
, 0);
779 static struct imsm_dev
*get_imsm_dev(struct intel_super
*super
, __u8 index
)
781 struct intel_dev
*dv
;
783 if (index
>= super
->anchor
->num_raid_devs
)
785 for (dv
= super
->devlist
; dv
; dv
= dv
->next
)
786 if (dv
->index
== index
)
793 * == MAP_0 get first map
794 * == MAP_1 get second map
795 * == MAP_X than get map according to the current migr_state
797 static __u32
get_imsm_ord_tbl_ent(struct imsm_dev
*dev
,
801 struct imsm_map
*map
;
803 map
= get_imsm_map(dev
, second_map
);
805 /* top byte identifies disk under rebuild */
806 return __le32_to_cpu(map
->disk_ord_tbl
[slot
]);
809 #define ord_to_idx(ord) (((ord) << 8) >> 8)
810 static __u32
get_imsm_disk_idx(struct imsm_dev
*dev
, int slot
, int second_map
)
812 __u32 ord
= get_imsm_ord_tbl_ent(dev
, slot
, second_map
);
814 return ord_to_idx(ord
);
817 static void set_imsm_ord_tbl_ent(struct imsm_map
*map
, int slot
, __u32 ord
)
819 map
->disk_ord_tbl
[slot
] = __cpu_to_le32(ord
);
822 static int get_imsm_disk_slot(struct imsm_map
*map
, unsigned idx
)
827 for (slot
= 0; slot
< map
->num_members
; slot
++) {
828 ord
= __le32_to_cpu(map
->disk_ord_tbl
[slot
]);
829 if (ord_to_idx(ord
) == idx
)
836 static int get_imsm_raid_level(struct imsm_map
*map
)
838 if (map
->raid_level
== 1) {
839 if (map
->num_members
== 2)
845 return map
->raid_level
;
848 static int cmp_extent(const void *av
, const void *bv
)
850 const struct extent
*a
= av
;
851 const struct extent
*b
= bv
;
852 if (a
->start
< b
->start
)
854 if (a
->start
> b
->start
)
859 static int count_memberships(struct dl
*dl
, struct intel_super
*super
)
864 for (i
= 0; i
< super
->anchor
->num_raid_devs
; i
++) {
865 struct imsm_dev
*dev
= get_imsm_dev(super
, i
);
866 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
868 if (get_imsm_disk_slot(map
, dl
->index
) >= 0)
875 static __u32
imsm_min_reserved_sectors(struct intel_super
*super
);
877 static int split_ull(unsigned long long n
, __u32
*lo
, __u32
*hi
)
879 if (lo
== 0 || hi
== 0)
881 *lo
= __le32_to_cpu((unsigned)n
);
882 *hi
= __le32_to_cpu((unsigned)(n
>> 32));
886 static unsigned long long join_u32(__u32 lo
, __u32 hi
)
888 return (unsigned long long)__le32_to_cpu(lo
) |
889 (((unsigned long long)__le32_to_cpu(hi
)) << 32);
892 static unsigned long long total_blocks(struct imsm_disk
*disk
)
896 return join_u32(disk
->total_blocks_lo
, disk
->total_blocks_hi
);
899 static unsigned long long pba_of_lba0(struct imsm_map
*map
)
903 return join_u32(map
->pba_of_lba0_lo
, map
->pba_of_lba0_hi
);
906 static unsigned long long blocks_per_member(struct imsm_map
*map
)
910 return join_u32(map
->blocks_per_member_lo
, map
->blocks_per_member_hi
);
913 static unsigned long long num_data_stripes(struct imsm_map
*map
)
917 return join_u32(map
->num_data_stripes_lo
, map
->num_data_stripes_hi
);
920 static void set_total_blocks(struct imsm_disk
*disk
, unsigned long long n
)
922 split_ull(n
, &disk
->total_blocks_lo
, &disk
->total_blocks_hi
);
925 static void set_pba_of_lba0(struct imsm_map
*map
, unsigned long long n
)
927 split_ull(n
, &map
->pba_of_lba0_lo
, &map
->pba_of_lba0_hi
);
930 static void set_blocks_per_member(struct imsm_map
*map
, unsigned long long n
)
932 split_ull(n
, &map
->blocks_per_member_lo
, &map
->blocks_per_member_hi
);
935 static void set_num_data_stripes(struct imsm_map
*map
, unsigned long long n
)
937 split_ull(n
, &map
->num_data_stripes_lo
, &map
->num_data_stripes_hi
);
940 static struct extent
*get_extents(struct intel_super
*super
, struct dl
*dl
)
942 /* find a list of used extents on the given physical device */
943 struct extent
*rv
, *e
;
945 int memberships
= count_memberships(dl
, super
);
948 /* trim the reserved area for spares, so they can join any array
949 * regardless of whether the OROM has assigned sectors from the
950 * IMSM_RESERVED_SECTORS region
953 reservation
= imsm_min_reserved_sectors(super
);
955 reservation
= MPB_SECTOR_CNT
+ IMSM_RESERVED_SECTORS
;
957 rv
= malloc(sizeof(struct extent
) * (memberships
+ 1));
962 for (i
= 0; i
< super
->anchor
->num_raid_devs
; i
++) {
963 struct imsm_dev
*dev
= get_imsm_dev(super
, i
);
964 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
966 if (get_imsm_disk_slot(map
, dl
->index
) >= 0) {
967 e
->start
= pba_of_lba0(map
);
968 e
->size
= blocks_per_member(map
);
972 qsort(rv
, memberships
, sizeof(*rv
), cmp_extent
);
974 /* determine the start of the metadata
975 * when no raid devices are defined use the default
976 * ...otherwise allow the metadata to truncate the value
977 * as is the case with older versions of imsm
980 struct extent
*last
= &rv
[memberships
- 1];
981 unsigned long long remainder
;
983 remainder
= total_blocks(&dl
->disk
) - (last
->start
+ last
->size
);
984 /* round down to 1k block to satisfy precision of the kernel
988 /* make sure remainder is still sane */
989 if (remainder
< (unsigned)ROUND_UP(super
->len
, 512) >> 9)
990 remainder
= ROUND_UP(super
->len
, 512) >> 9;
991 if (reservation
> remainder
)
992 reservation
= remainder
;
994 e
->start
= total_blocks(&dl
->disk
) - reservation
;
999 /* try to determine how much space is reserved for metadata from
1000 * the last get_extents() entry, otherwise fallback to the
1003 static __u32
imsm_reserved_sectors(struct intel_super
*super
, struct dl
*dl
)
1009 /* for spares just return a minimal reservation which will grow
1010 * once the spare is picked up by an array
1012 if (dl
->index
== -1)
1013 return MPB_SECTOR_CNT
;
1015 e
= get_extents(super
, dl
);
1017 return MPB_SECTOR_CNT
+ IMSM_RESERVED_SECTORS
;
1019 /* scroll to last entry */
1020 for (i
= 0; e
[i
].size
; i
++)
1023 rv
= total_blocks(&dl
->disk
) - e
[i
].start
;
1030 static int is_spare(struct imsm_disk
*disk
)
1032 return (disk
->status
& SPARE_DISK
) == SPARE_DISK
;
1035 static int is_configured(struct imsm_disk
*disk
)
1037 return (disk
->status
& CONFIGURED_DISK
) == CONFIGURED_DISK
;
1040 static int is_failed(struct imsm_disk
*disk
)
1042 return (disk
->status
& FAILED_DISK
) == FAILED_DISK
;
1045 /* try to determine how much space is reserved for metadata from
1046 * the last get_extents() entry on the smallest active disk,
1047 * otherwise fallback to the default
1049 static __u32
imsm_min_reserved_sectors(struct intel_super
*super
)
1053 unsigned long long min_active
;
1055 __u32 rv
= MPB_SECTOR_CNT
+ IMSM_RESERVED_SECTORS
;
1056 struct dl
*dl
, *dl_min
= NULL
;
1062 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
1065 unsigned long long blocks
= total_blocks(&dl
->disk
);
1066 if (blocks
< min_active
|| min_active
== 0) {
1068 min_active
= blocks
;
1074 /* find last lba used by subarrays on the smallest active disk */
1075 e
= get_extents(super
, dl_min
);
1078 for (i
= 0; e
[i
].size
; i
++)
1081 remainder
= min_active
- e
[i
].start
;
1084 /* to give priority to recovery we should not require full
1085 IMSM_RESERVED_SECTORS from the spare */
1086 rv
= MPB_SECTOR_CNT
+ NUM_BLOCKS_DIRTY_STRIPE_REGION
;
1088 /* if real reservation is smaller use that value */
1089 return (remainder
< rv
) ? remainder
: rv
;
1092 /* Return minimum size of a spare that can be used in this array*/
1093 static unsigned long long min_acceptable_spare_size_imsm(struct supertype
*st
)
1095 struct intel_super
*super
= st
->sb
;
1099 unsigned long long rv
= 0;
1103 /* find first active disk in array */
1105 while (dl
&& (is_failed(&dl
->disk
) || dl
->index
== -1))
1109 /* find last lba used by subarrays */
1110 e
= get_extents(super
, dl
);
1113 for (i
= 0; e
[i
].size
; i
++)
1116 rv
= e
[i
-1].start
+ e
[i
-1].size
;
1119 /* add the amount of space needed for metadata */
1120 rv
= rv
+ imsm_min_reserved_sectors(super
);
1125 static int is_gen_migration(struct imsm_dev
*dev
);
1128 static __u64
blocks_per_migr_unit(struct intel_super
*super
,
1129 struct imsm_dev
*dev
);
1131 static void print_imsm_dev(struct intel_super
*super
,
1132 struct imsm_dev
*dev
,
1138 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
1139 struct imsm_map
*map2
= get_imsm_map(dev
, MAP_1
);
1143 printf("[%.16s]:\n", dev
->volume
);
1144 printf(" UUID : %s\n", uuid
);
1145 printf(" RAID Level : %d", get_imsm_raid_level(map
));
1147 printf(" <-- %d", get_imsm_raid_level(map2
));
1149 printf(" Members : %d", map
->num_members
);
1151 printf(" <-- %d", map2
->num_members
);
1153 printf(" Slots : [");
1154 for (i
= 0; i
< map
->num_members
; i
++) {
1155 ord
= get_imsm_ord_tbl_ent(dev
, i
, MAP_0
);
1156 printf("%s", ord
& IMSM_ORD_REBUILD
? "_" : "U");
1161 for (i
= 0; i
< map2
->num_members
; i
++) {
1162 ord
= get_imsm_ord_tbl_ent(dev
, i
, MAP_1
);
1163 printf("%s", ord
& IMSM_ORD_REBUILD
? "_" : "U");
1168 printf(" Failed disk : ");
1169 if (map
->failed_disk_num
== 0xff)
1172 printf("%i", map
->failed_disk_num
);
1174 slot
= get_imsm_disk_slot(map
, disk_idx
);
1176 ord
= get_imsm_ord_tbl_ent(dev
, slot
, MAP_X
);
1177 printf(" This Slot : %d%s\n", slot
,
1178 ord
& IMSM_ORD_REBUILD
? " (out-of-sync)" : "");
1180 printf(" This Slot : ?\n");
1181 sz
= __le32_to_cpu(dev
->size_high
);
1183 sz
+= __le32_to_cpu(dev
->size_low
);
1184 printf(" Array Size : %llu%s\n", (unsigned long long)sz
,
1185 human_size(sz
* 512));
1186 sz
= blocks_per_member(map
);
1187 printf(" Per Dev Size : %llu%s\n", (unsigned long long)sz
,
1188 human_size(sz
* 512));
1189 printf(" Sector Offset : %llu\n",
1191 printf(" Num Stripes : %llu\n",
1192 num_data_stripes(map
));
1193 printf(" Chunk Size : %u KiB",
1194 __le16_to_cpu(map
->blocks_per_strip
) / 2);
1196 printf(" <-- %u KiB",
1197 __le16_to_cpu(map2
->blocks_per_strip
) / 2);
1199 printf(" Reserved : %d\n", __le32_to_cpu(dev
->reserved_blocks
));
1200 printf(" Migrate State : ");
1201 if (dev
->vol
.migr_state
) {
1202 if (migr_type(dev
) == MIGR_INIT
)
1203 printf("initialize\n");
1204 else if (migr_type(dev
) == MIGR_REBUILD
)
1205 printf("rebuild\n");
1206 else if (migr_type(dev
) == MIGR_VERIFY
)
1208 else if (migr_type(dev
) == MIGR_GEN_MIGR
)
1209 printf("general migration\n");
1210 else if (migr_type(dev
) == MIGR_STATE_CHANGE
)
1211 printf("state change\n");
1212 else if (migr_type(dev
) == MIGR_REPAIR
)
1215 printf("<unknown:%d>\n", migr_type(dev
));
1218 printf(" Map State : %s", map_state_str
[map
->map_state
]);
1219 if (dev
->vol
.migr_state
) {
1220 struct imsm_map
*map
= get_imsm_map(dev
, MAP_1
);
1222 printf(" <-- %s", map_state_str
[map
->map_state
]);
1223 printf("\n Checkpoint : %u ",
1224 __le32_to_cpu(dev
->vol
.curr_migr_unit
));
1225 if ((is_gen_migration(dev
)) && ((slot
> 1) || (slot
< 0)))
1228 printf("(%llu)", (unsigned long long)
1229 blocks_per_migr_unit(super
, dev
));
1232 printf(" Dirty State : %s\n", dev
->vol
.dirty
? "dirty" : "clean");
1235 static void print_imsm_disk(struct imsm_disk
*disk
, int index
, __u32 reserved
)
1237 char str
[MAX_RAID_SERIAL_LEN
+ 1];
1240 if (index
< -1 || !disk
)
1244 snprintf(str
, MAX_RAID_SERIAL_LEN
+ 1, "%s", disk
->serial
);
1246 printf(" Disk%02d Serial : %s\n", index
, str
);
1248 printf(" Disk Serial : %s\n", str
);
1249 printf(" State :%s%s%s\n", is_spare(disk
) ? " spare" : "",
1250 is_configured(disk
) ? " active" : "",
1251 is_failed(disk
) ? " failed" : "");
1252 printf(" Id : %08x\n", __le32_to_cpu(disk
->scsi_id
));
1253 sz
= total_blocks(disk
) - reserved
;
1254 printf(" Usable Size : %llu%s\n", (unsigned long long)sz
,
1255 human_size(sz
* 512));
1258 void examine_migr_rec_imsm(struct intel_super
*super
)
1260 struct migr_record
*migr_rec
= super
->migr_rec
;
1261 struct imsm_super
*mpb
= super
->anchor
;
1264 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
1265 struct imsm_dev
*dev
= __get_imsm_dev(mpb
, i
);
1266 struct imsm_map
*map
;
1269 if (is_gen_migration(dev
) == 0)
1272 printf("\nMigration Record Information:");
1274 /* first map under migration */
1275 map
= get_imsm_map(dev
, MAP_0
);
1277 slot
= get_imsm_disk_slot(map
, super
->disks
->index
);
1278 if ((map
== NULL
) || (slot
> 1) || (slot
< 0)) {
1279 printf(" Empty\n ");
1280 printf("Examine one of first two disks in array\n");
1283 printf("\n Status : ");
1284 if (__le32_to_cpu(migr_rec
->rec_status
) == UNIT_SRC_NORMAL
)
1287 printf("Contains Data\n");
1288 printf(" Current Unit : %u\n",
1289 __le32_to_cpu(migr_rec
->curr_migr_unit
));
1290 printf(" Family : %u\n",
1291 __le32_to_cpu(migr_rec
->family_num
));
1292 printf(" Ascending : %u\n",
1293 __le32_to_cpu(migr_rec
->ascending_migr
));
1294 printf(" Blocks Per Unit : %u\n",
1295 __le32_to_cpu(migr_rec
->blocks_per_unit
));
1296 printf(" Dest. Depth Per Unit : %u\n",
1297 __le32_to_cpu(migr_rec
->dest_depth_per_unit
));
1298 printf(" Checkpoint Area pba : %u\n",
1299 __le32_to_cpu(migr_rec
->ckpt_area_pba
));
1300 printf(" First member lba : %u\n",
1301 __le32_to_cpu(migr_rec
->dest_1st_member_lba
));
1302 printf(" Total Number of Units : %u\n",
1303 __le32_to_cpu(migr_rec
->num_migr_units
));
1304 printf(" Size of volume : %u\n",
1305 __le32_to_cpu(migr_rec
->post_migr_vol_cap
));
1306 printf(" Expansion space for LBA64 : %u\n",
1307 __le32_to_cpu(migr_rec
->post_migr_vol_cap_hi
));
1308 printf(" Record was read from : %u\n",
1309 __le32_to_cpu(migr_rec
->ckpt_read_disk_num
));
1314 #endif /* MDASSEMBLE */
1315 /*******************************************************************************
1316 * function: imsm_check_attributes
1317 * Description: Function checks if features represented by attributes flags
1318 * are supported by mdadm.
1320 * attributes - Attributes read from metadata
1322 * 0 - passed attributes contains unsupported features flags
1323 * 1 - all features are supported
1324 ******************************************************************************/
1325 static int imsm_check_attributes(__u32 attributes
)
1328 __u32 not_supported
= MPB_ATTRIB_SUPPORTED
^0xffffffff;
1330 not_supported
&= ~MPB_ATTRIB_IGNORED
;
1332 not_supported
&= attributes
;
1333 if (not_supported
) {
1334 fprintf(stderr
, Name
"(IMSM): Unsupported attributes : %x\n",
1335 (unsigned)__le32_to_cpu(not_supported
));
1336 if (not_supported
& MPB_ATTRIB_CHECKSUM_VERIFY
) {
1337 dprintf("\t\tMPB_ATTRIB_CHECKSUM_VERIFY \n");
1338 not_supported
^= MPB_ATTRIB_CHECKSUM_VERIFY
;
1340 if (not_supported
& MPB_ATTRIB_2TB
) {
1341 dprintf("\t\tMPB_ATTRIB_2TB\n");
1342 not_supported
^= MPB_ATTRIB_2TB
;
1344 if (not_supported
& MPB_ATTRIB_RAID0
) {
1345 dprintf("\t\tMPB_ATTRIB_RAID0\n");
1346 not_supported
^= MPB_ATTRIB_RAID0
;
1348 if (not_supported
& MPB_ATTRIB_RAID1
) {
1349 dprintf("\t\tMPB_ATTRIB_RAID1\n");
1350 not_supported
^= MPB_ATTRIB_RAID1
;
1352 if (not_supported
& MPB_ATTRIB_RAID10
) {
1353 dprintf("\t\tMPB_ATTRIB_RAID10\n");
1354 not_supported
^= MPB_ATTRIB_RAID10
;
1356 if (not_supported
& MPB_ATTRIB_RAID1E
) {
1357 dprintf("\t\tMPB_ATTRIB_RAID1E\n");
1358 not_supported
^= MPB_ATTRIB_RAID1E
;
1360 if (not_supported
& MPB_ATTRIB_RAID5
) {
1361 dprintf("\t\tMPB_ATTRIB_RAID5\n");
1362 not_supported
^= MPB_ATTRIB_RAID5
;
1364 if (not_supported
& MPB_ATTRIB_RAIDCNG
) {
1365 dprintf("\t\tMPB_ATTRIB_RAIDCNG\n");
1366 not_supported
^= MPB_ATTRIB_RAIDCNG
;
1368 if (not_supported
& MPB_ATTRIB_BBM
) {
1369 dprintf("\t\tMPB_ATTRIB_BBM\n");
1370 not_supported
^= MPB_ATTRIB_BBM
;
1372 if (not_supported
& MPB_ATTRIB_CHECKSUM_VERIFY
) {
1373 dprintf("\t\tMPB_ATTRIB_CHECKSUM_VERIFY (== MPB_ATTRIB_LEGACY)\n");
1374 not_supported
^= MPB_ATTRIB_CHECKSUM_VERIFY
;
1376 if (not_supported
& MPB_ATTRIB_EXP_STRIPE_SIZE
) {
1377 dprintf("\t\tMPB_ATTRIB_EXP_STRIP_SIZE\n");
1378 not_supported
^= MPB_ATTRIB_EXP_STRIPE_SIZE
;
1380 if (not_supported
& MPB_ATTRIB_2TB_DISK
) {
1381 dprintf("\t\tMPB_ATTRIB_2TB_DISK\n");
1382 not_supported
^= MPB_ATTRIB_2TB_DISK
;
1384 if (not_supported
& MPB_ATTRIB_NEVER_USE2
) {
1385 dprintf("\t\tMPB_ATTRIB_NEVER_USE2\n");
1386 not_supported
^= MPB_ATTRIB_NEVER_USE2
;
1388 if (not_supported
& MPB_ATTRIB_NEVER_USE
) {
1389 dprintf("\t\tMPB_ATTRIB_NEVER_USE\n");
1390 not_supported
^= MPB_ATTRIB_NEVER_USE
;
1394 dprintf(Name
"(IMSM): Unknown attributes : %x\n", not_supported
);
1403 static void getinfo_super_imsm(struct supertype
*st
, struct mdinfo
*info
, char *map
);
1405 static void examine_super_imsm(struct supertype
*st
, char *homehost
)
1407 struct intel_super
*super
= st
->sb
;
1408 struct imsm_super
*mpb
= super
->anchor
;
1409 char str
[MAX_SIGNATURE_LENGTH
];
1414 __u32 reserved
= imsm_reserved_sectors(super
, super
->disks
);
1417 snprintf(str
, MPB_SIG_LEN
, "%s", mpb
->sig
);
1418 printf(" Magic : %s\n", str
);
1419 snprintf(str
, strlen(MPB_VERSION_RAID0
), "%s", get_imsm_version(mpb
));
1420 printf(" Version : %s\n", get_imsm_version(mpb
));
1421 printf(" Orig Family : %08x\n", __le32_to_cpu(mpb
->orig_family_num
));
1422 printf(" Family : %08x\n", __le32_to_cpu(mpb
->family_num
));
1423 printf(" Generation : %08x\n", __le32_to_cpu(mpb
->generation_num
));
1424 printf(" Attributes : ");
1425 if (imsm_check_attributes(mpb
->attributes
))
1426 printf("All supported\n");
1428 printf("not supported\n");
1429 getinfo_super_imsm(st
, &info
, NULL
);
1430 fname_from_uuid(st
, &info
, nbuf
, ':');
1431 printf(" UUID : %s\n", nbuf
+ 5);
1432 sum
= __le32_to_cpu(mpb
->check_sum
);
1433 printf(" Checksum : %08x %s\n", sum
,
1434 __gen_imsm_checksum(mpb
) == sum
? "correct" : "incorrect");
1435 printf(" MPB Sectors : %d\n", mpb_sectors(mpb
));
1436 printf(" Disks : %d\n", mpb
->num_disks
);
1437 printf(" RAID Devices : %d\n", mpb
->num_raid_devs
);
1438 print_imsm_disk(__get_imsm_disk(mpb
, super
->disks
->index
), super
->disks
->index
, reserved
);
1439 if (super
->bbm_log
) {
1440 struct bbm_log
*log
= super
->bbm_log
;
1443 printf("Bad Block Management Log:\n");
1444 printf(" Log Size : %d\n", __le32_to_cpu(mpb
->bbm_log_size
));
1445 printf(" Signature : %x\n", __le32_to_cpu(log
->signature
));
1446 printf(" Entry Count : %d\n", __le32_to_cpu(log
->entry_count
));
1447 printf(" Spare Blocks : %d\n", __le32_to_cpu(log
->reserved_spare_block_count
));
1448 printf(" First Spare : %llx\n",
1449 (unsigned long long) __le64_to_cpu(log
->first_spare_lba
));
1451 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
1453 struct imsm_dev
*dev
= __get_imsm_dev(mpb
, i
);
1455 super
->current_vol
= i
;
1456 getinfo_super_imsm(st
, &info
, NULL
);
1457 fname_from_uuid(st
, &info
, nbuf
, ':');
1458 print_imsm_dev(super
, dev
, nbuf
+ 5, super
->disks
->index
);
1460 for (i
= 0; i
< mpb
->num_disks
; i
++) {
1461 if (i
== super
->disks
->index
)
1463 print_imsm_disk(__get_imsm_disk(mpb
, i
), i
, reserved
);
1466 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
1467 if (dl
->index
== -1)
1468 print_imsm_disk(&dl
->disk
, -1, reserved
);
1470 examine_migr_rec_imsm(super
);
1473 static void brief_examine_super_imsm(struct supertype
*st
, int verbose
)
1475 /* We just write a generic IMSM ARRAY entry */
1478 struct intel_super
*super
= st
->sb
;
1480 if (!super
->anchor
->num_raid_devs
) {
1481 printf("ARRAY metadata=imsm\n");
1485 getinfo_super_imsm(st
, &info
, NULL
);
1486 fname_from_uuid(st
, &info
, nbuf
, ':');
1487 printf("ARRAY metadata=imsm UUID=%s\n", nbuf
+ 5);
1490 static void brief_examine_subarrays_imsm(struct supertype
*st
, int verbose
)
1492 /* We just write a generic IMSM ARRAY entry */
1496 struct intel_super
*super
= st
->sb
;
1499 if (!super
->anchor
->num_raid_devs
)
1502 getinfo_super_imsm(st
, &info
, NULL
);
1503 fname_from_uuid(st
, &info
, nbuf
, ':');
1504 for (i
= 0; i
< super
->anchor
->num_raid_devs
; i
++) {
1505 struct imsm_dev
*dev
= get_imsm_dev(super
, i
);
1507 super
->current_vol
= i
;
1508 getinfo_super_imsm(st
, &info
, NULL
);
1509 fname_from_uuid(st
, &info
, nbuf1
, ':');
1510 printf("ARRAY /dev/md/%.16s container=%s member=%d UUID=%s\n",
1511 dev
->volume
, nbuf
+ 5, i
, nbuf1
+ 5);
1515 static void export_examine_super_imsm(struct supertype
*st
)
1517 struct intel_super
*super
= st
->sb
;
1518 struct imsm_super
*mpb
= super
->anchor
;
1522 getinfo_super_imsm(st
, &info
, NULL
);
1523 fname_from_uuid(st
, &info
, nbuf
, ':');
1524 printf("MD_METADATA=imsm\n");
1525 printf("MD_LEVEL=container\n");
1526 printf("MD_UUID=%s\n", nbuf
+5);
1527 printf("MD_DEVICES=%u\n", mpb
->num_disks
);
1530 static void detail_super_imsm(struct supertype
*st
, char *homehost
)
1535 getinfo_super_imsm(st
, &info
, NULL
);
1536 fname_from_uuid(st
, &info
, nbuf
, ':');
1537 printf("\n UUID : %s\n", nbuf
+ 5);
1540 static void brief_detail_super_imsm(struct supertype
*st
)
1544 getinfo_super_imsm(st
, &info
, NULL
);
1545 fname_from_uuid(st
, &info
, nbuf
, ':');
1546 printf(" UUID=%s", nbuf
+ 5);
1549 static int imsm_read_serial(int fd
, char *devname
, __u8
*serial
);
1550 static void fd2devname(int fd
, char *name
);
1552 static int ahci_enumerate_ports(const char *hba_path
, int port_count
, int host_base
, int verbose
)
1554 /* dump an unsorted list of devices attached to AHCI Intel storage
1555 * controller, as well as non-connected ports
1557 int hba_len
= strlen(hba_path
) + 1;
1562 unsigned long port_mask
= (1 << port_count
) - 1;
1564 if (port_count
> (int)sizeof(port_mask
) * 8) {
1566 fprintf(stderr
, Name
": port_count %d out of range\n", port_count
);
1570 /* scroll through /sys/dev/block looking for devices attached to
1573 dir
= opendir("/sys/dev/block");
1574 for (ent
= dir
? readdir(dir
) : NULL
; ent
; ent
= readdir(dir
)) {
1585 if (sscanf(ent
->d_name
, "%d:%d", &major
, &minor
) != 2)
1587 path
= devt_to_devpath(makedev(major
, minor
));
1590 if (!path_attached_to_hba(path
, hba_path
)) {
1596 /* retrieve the scsi device type */
1597 if (asprintf(&device
, "/sys/dev/block/%d:%d/device/xxxxxxx", major
, minor
) < 0) {
1599 fprintf(stderr
, Name
": failed to allocate 'device'\n");
1603 sprintf(device
, "/sys/dev/block/%d:%d/device/type", major
, minor
);
1604 if (load_sys(device
, buf
) != 0) {
1606 fprintf(stderr
, Name
": failed to read device type for %s\n",
1612 type
= strtoul(buf
, NULL
, 10);
1614 /* if it's not a disk print the vendor and model */
1615 if (!(type
== 0 || type
== 7 || type
== 14)) {
1618 sprintf(device
, "/sys/dev/block/%d:%d/device/vendor", major
, minor
);
1619 if (load_sys(device
, buf
) == 0) {
1620 strncpy(vendor
, buf
, sizeof(vendor
));
1621 vendor
[sizeof(vendor
) - 1] = '\0';
1622 c
= (char *) &vendor
[sizeof(vendor
) - 1];
1623 while (isspace(*c
) || *c
== '\0')
1627 sprintf(device
, "/sys/dev/block/%d:%d/device/model", major
, minor
);
1628 if (load_sys(device
, buf
) == 0) {
1629 strncpy(model
, buf
, sizeof(model
));
1630 model
[sizeof(model
) - 1] = '\0';
1631 c
= (char *) &model
[sizeof(model
) - 1];
1632 while (isspace(*c
) || *c
== '\0')
1636 if (vendor
[0] && model
[0])
1637 sprintf(buf
, "%.64s %.64s", vendor
, model
);
1639 switch (type
) { /* numbers from hald/linux/device.c */
1640 case 1: sprintf(buf
, "tape"); break;
1641 case 2: sprintf(buf
, "printer"); break;
1642 case 3: sprintf(buf
, "processor"); break;
1644 case 5: sprintf(buf
, "cdrom"); break;
1645 case 6: sprintf(buf
, "scanner"); break;
1646 case 8: sprintf(buf
, "media_changer"); break;
1647 case 9: sprintf(buf
, "comm"); break;
1648 case 12: sprintf(buf
, "raid"); break;
1649 default: sprintf(buf
, "unknown");
1655 /* chop device path to 'host%d' and calculate the port number */
1656 c
= strchr(&path
[hba_len
], '/');
1659 fprintf(stderr
, Name
": %s - invalid path name\n", path
+ hba_len
);
1664 if (sscanf(&path
[hba_len
], "host%d", &port
) == 1)
1668 *c
= '/'; /* repair the full string */
1669 fprintf(stderr
, Name
": failed to determine port number for %s\n",
1676 /* mark this port as used */
1677 port_mask
&= ~(1 << port
);
1679 /* print out the device information */
1681 printf(" Port%d : - non-disk device (%s) -\n", port
, buf
);
1685 fd
= dev_open(ent
->d_name
, O_RDONLY
);
1687 printf(" Port%d : - disk info unavailable -\n", port
);
1689 fd2devname(fd
, buf
);
1690 printf(" Port%d : %s", port
, buf
);
1691 if (imsm_read_serial(fd
, NULL
, (__u8
*) buf
) == 0)
1692 printf(" (%.*s)\n", MAX_RAID_SERIAL_LEN
, buf
);
1707 for (i
= 0; i
< port_count
; i
++)
1708 if (port_mask
& (1 << i
))
1709 printf(" Port%d : - no device attached -\n", i
);
1715 static void print_found_intel_controllers(struct sys_dev
*elem
)
1717 for (; elem
; elem
= elem
->next
) {
1718 fprintf(stderr
, Name
": found Intel(R) ");
1719 if (elem
->type
== SYS_DEV_SATA
)
1720 fprintf(stderr
, "SATA ");
1721 else if (elem
->type
== SYS_DEV_SAS
)
1722 fprintf(stderr
, "SAS ");
1723 fprintf(stderr
, "RAID controller");
1725 fprintf(stderr
, " at %s", elem
->pci_id
);
1726 fprintf(stderr
, ".\n");
1731 static int ahci_get_port_count(const char *hba_path
, int *port_count
)
1738 if ((dir
= opendir(hba_path
)) == NULL
)
1741 for (ent
= readdir(dir
); ent
; ent
= readdir(dir
)) {
1744 if (sscanf(ent
->d_name
, "host%d", &host
) != 1)
1746 if (*port_count
== 0)
1748 else if (host
< host_base
)
1751 if (host
+ 1 > *port_count
+ host_base
)
1752 *port_count
= host
+ 1 - host_base
;
1758 static void print_imsm_capability(const struct imsm_orom
*orom
)
1760 printf(" Platform : Intel(R) Matrix Storage Manager\n");
1761 printf(" Version : %d.%d.%d.%d\n", orom
->major_ver
, orom
->minor_ver
,
1762 orom
->hotfix_ver
, orom
->build
);
1763 printf(" RAID Levels :%s%s%s%s%s\n",
1764 imsm_orom_has_raid0(orom
) ? " raid0" : "",
1765 imsm_orom_has_raid1(orom
) ? " raid1" : "",
1766 imsm_orom_has_raid1e(orom
) ? " raid1e" : "",
1767 imsm_orom_has_raid10(orom
) ? " raid10" : "",
1768 imsm_orom_has_raid5(orom
) ? " raid5" : "");
1769 printf(" Chunk Sizes :%s%s%s%s%s%s%s%s%s%s%s%s%s%s%s%s\n",
1770 imsm_orom_has_chunk(orom
, 2) ? " 2k" : "",
1771 imsm_orom_has_chunk(orom
, 4) ? " 4k" : "",
1772 imsm_orom_has_chunk(orom
, 8) ? " 8k" : "",
1773 imsm_orom_has_chunk(orom
, 16) ? " 16k" : "",
1774 imsm_orom_has_chunk(orom
, 32) ? " 32k" : "",
1775 imsm_orom_has_chunk(orom
, 64) ? " 64k" : "",
1776 imsm_orom_has_chunk(orom
, 128) ? " 128k" : "",
1777 imsm_orom_has_chunk(orom
, 256) ? " 256k" : "",
1778 imsm_orom_has_chunk(orom
, 512) ? " 512k" : "",
1779 imsm_orom_has_chunk(orom
, 1024*1) ? " 1M" : "",
1780 imsm_orom_has_chunk(orom
, 1024*2) ? " 2M" : "",
1781 imsm_orom_has_chunk(orom
, 1024*4) ? " 4M" : "",
1782 imsm_orom_has_chunk(orom
, 1024*8) ? " 8M" : "",
1783 imsm_orom_has_chunk(orom
, 1024*16) ? " 16M" : "",
1784 imsm_orom_has_chunk(orom
, 1024*32) ? " 32M" : "",
1785 imsm_orom_has_chunk(orom
, 1024*64) ? " 64M" : "");
1786 printf(" Max Disks : %d\n", orom
->tds
);
1787 printf(" Max Volumes : %d\n", orom
->vpa
);
1791 static int detail_platform_imsm(int verbose
, int enumerate_only
)
1793 /* There are two components to imsm platform support, the ahci SATA
1794 * controller and the option-rom. To find the SATA controller we
1795 * simply look in /sys/bus/pci/drivers/ahci to see if an ahci
1796 * controller with the Intel vendor id is present. This approach
1797 * allows mdadm to leverage the kernel's ahci detection logic, with the
1798 * caveat that if ahci.ko is not loaded mdadm will not be able to
1799 * detect platform raid capabilities. The option-rom resides in a
1800 * platform "Adapter ROM". We scan for its signature to retrieve the
1801 * platform capabilities. If raid support is disabled in the BIOS the
1802 * option-rom capability structure will not be available.
1804 const struct imsm_orom
*orom
;
1805 struct sys_dev
*list
, *hba
;
1810 if (enumerate_only
) {
1811 if (check_env("IMSM_NO_PLATFORM"))
1813 list
= find_intel_devices();
1816 for (hba
= list
; hba
; hba
= hba
->next
) {
1817 orom
= find_imsm_capability(hba
->type
);
1823 free_sys_dev(&list
);
1827 list
= find_intel_devices();
1830 fprintf(stderr
, Name
": no active Intel(R) RAID "
1831 "controller found.\n");
1832 free_sys_dev(&list
);
1835 print_found_intel_controllers(list
);
1837 for (hba
= list
; hba
; hba
= hba
->next
) {
1838 orom
= find_imsm_capability(hba
->type
);
1840 fprintf(stderr
, Name
": imsm capabilities not found for controller: %s (type %s)\n",
1841 hba
->path
, get_sys_dev_type(hba
->type
));
1843 print_imsm_capability(orom
);
1846 for (hba
= list
; hba
; hba
= hba
->next
) {
1847 printf(" I/O Controller : %s (%s)\n",
1848 hba
->path
, get_sys_dev_type(hba
->type
));
1850 if (hba
->type
== SYS_DEV_SATA
) {
1851 host_base
= ahci_get_port_count(hba
->path
, &port_count
);
1852 if (ahci_enumerate_ports(hba
->path
, port_count
, host_base
, verbose
)) {
1854 fprintf(stderr
, Name
": failed to enumerate "
1855 "ports on SATA controller at %s.", hba
->pci_id
);
1861 free_sys_dev(&list
);
1866 static int match_home_imsm(struct supertype
*st
, char *homehost
)
1868 /* the imsm metadata format does not specify any host
1869 * identification information. We return -1 since we can never
1870 * confirm nor deny whether a given array is "meant" for this
1871 * host. We rely on compare_super and the 'family_num' fields to
1872 * exclude member disks that do not belong, and we rely on
1873 * mdadm.conf to specify the arrays that should be assembled.
1874 * Auto-assembly may still pick up "foreign" arrays.
1880 static void uuid_from_super_imsm(struct supertype
*st
, int uuid
[4])
1882 /* The uuid returned here is used for:
1883 * uuid to put into bitmap file (Create, Grow)
1884 * uuid for backup header when saving critical section (Grow)
1885 * comparing uuids when re-adding a device into an array
1886 * In these cases the uuid required is that of the data-array,
1887 * not the device-set.
1888 * uuid to recognise same set when adding a missing device back
1889 * to an array. This is a uuid for the device-set.
1891 * For each of these we can make do with a truncated
1892 * or hashed uuid rather than the original, as long as
1894 * In each case the uuid required is that of the data-array,
1895 * not the device-set.
1897 /* imsm does not track uuid's so we synthesis one using sha1 on
1898 * - The signature (Which is constant for all imsm array, but no matter)
1899 * - the orig_family_num of the container
1900 * - the index number of the volume
1901 * - the 'serial' number of the volume.
1902 * Hopefully these are all constant.
1904 struct intel_super
*super
= st
->sb
;
1907 struct sha1_ctx ctx
;
1908 struct imsm_dev
*dev
= NULL
;
1911 /* some mdadm versions failed to set ->orig_family_num, in which
1912 * case fall back to ->family_num. orig_family_num will be
1913 * fixed up with the first metadata update.
1915 family_num
= super
->anchor
->orig_family_num
;
1916 if (family_num
== 0)
1917 family_num
= super
->anchor
->family_num
;
1918 sha1_init_ctx(&ctx
);
1919 sha1_process_bytes(super
->anchor
->sig
, MPB_SIG_LEN
, &ctx
);
1920 sha1_process_bytes(&family_num
, sizeof(__u32
), &ctx
);
1921 if (super
->current_vol
>= 0)
1922 dev
= get_imsm_dev(super
, super
->current_vol
);
1924 __u32 vol
= super
->current_vol
;
1925 sha1_process_bytes(&vol
, sizeof(vol
), &ctx
);
1926 sha1_process_bytes(dev
->volume
, MAX_RAID_SERIAL_LEN
, &ctx
);
1928 sha1_finish_ctx(&ctx
, buf
);
1929 memcpy(uuid
, buf
, 4*4);
1934 get_imsm_numerical_version(struct imsm_super
*mpb
, int *m
, int *p
)
1936 __u8
*v
= get_imsm_version(mpb
);
1937 __u8
*end
= mpb
->sig
+ MAX_SIGNATURE_LENGTH
;
1938 char major
[] = { 0, 0, 0 };
1939 char minor
[] = { 0 ,0, 0 };
1940 char patch
[] = { 0, 0, 0 };
1941 char *ver_parse
[] = { major
, minor
, patch
};
1945 while (*v
!= '\0' && v
< end
) {
1946 if (*v
!= '.' && j
< 2)
1947 ver_parse
[i
][j
++] = *v
;
1955 *m
= strtol(minor
, NULL
, 0);
1956 *p
= strtol(patch
, NULL
, 0);
1960 static __u32
migr_strip_blocks_resync(struct imsm_dev
*dev
)
1962 /* migr_strip_size when repairing or initializing parity */
1963 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
1964 __u32 chunk
= __le32_to_cpu(map
->blocks_per_strip
);
1966 switch (get_imsm_raid_level(map
)) {
1971 return 128*1024 >> 9;
1975 static __u32
migr_strip_blocks_rebuild(struct imsm_dev
*dev
)
1977 /* migr_strip_size when rebuilding a degraded disk, no idea why
1978 * this is different than migr_strip_size_resync(), but it's good
1981 struct imsm_map
*map
= get_imsm_map(dev
, MAP_1
);
1982 __u32 chunk
= __le32_to_cpu(map
->blocks_per_strip
);
1984 switch (get_imsm_raid_level(map
)) {
1987 if (map
->num_members
% map
->num_domains
== 0)
1988 return 128*1024 >> 9;
1992 return max((__u32
) 64*1024 >> 9, chunk
);
1994 return 128*1024 >> 9;
1998 static __u32
num_stripes_per_unit_resync(struct imsm_dev
*dev
)
2000 struct imsm_map
*lo
= get_imsm_map(dev
, MAP_0
);
2001 struct imsm_map
*hi
= get_imsm_map(dev
, MAP_1
);
2002 __u32 lo_chunk
= __le32_to_cpu(lo
->blocks_per_strip
);
2003 __u32 hi_chunk
= __le32_to_cpu(hi
->blocks_per_strip
);
2005 return max((__u32
) 1, hi_chunk
/ lo_chunk
);
2008 static __u32
num_stripes_per_unit_rebuild(struct imsm_dev
*dev
)
2010 struct imsm_map
*lo
= get_imsm_map(dev
, MAP_0
);
2011 int level
= get_imsm_raid_level(lo
);
2013 if (level
== 1 || level
== 10) {
2014 struct imsm_map
*hi
= get_imsm_map(dev
, MAP_1
);
2016 return hi
->num_domains
;
2018 return num_stripes_per_unit_resync(dev
);
2021 static __u8
imsm_num_data_members(struct imsm_dev
*dev
, int second_map
)
2023 /* named 'imsm_' because raid0, raid1 and raid10
2024 * counter-intuitively have the same number of data disks
2026 struct imsm_map
*map
= get_imsm_map(dev
, second_map
);
2028 switch (get_imsm_raid_level(map
)) {
2032 return map
->num_members
;
2034 return map
->num_members
- 1;
2036 dprintf("%s: unsupported raid level\n", __func__
);
2041 static __u32
parity_segment_depth(struct imsm_dev
*dev
)
2043 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
2044 __u32 chunk
= __le32_to_cpu(map
->blocks_per_strip
);
2046 switch(get_imsm_raid_level(map
)) {
2049 return chunk
* map
->num_domains
;
2051 return chunk
* map
->num_members
;
2057 static __u32
map_migr_block(struct imsm_dev
*dev
, __u32 block
)
2059 struct imsm_map
*map
= get_imsm_map(dev
, MAP_1
);
2060 __u32 chunk
= __le32_to_cpu(map
->blocks_per_strip
);
2061 __u32 strip
= block
/ chunk
;
2063 switch (get_imsm_raid_level(map
)) {
2066 __u32 vol_strip
= (strip
* map
->num_domains
) + 1;
2067 __u32 vol_stripe
= vol_strip
/ map
->num_members
;
2069 return vol_stripe
* chunk
+ block
% chunk
;
2071 __u32 stripe
= strip
/ (map
->num_members
- 1);
2073 return stripe
* chunk
+ block
% chunk
;
2080 static __u64
blocks_per_migr_unit(struct intel_super
*super
,
2081 struct imsm_dev
*dev
)
2083 /* calculate the conversion factor between per member 'blocks'
2084 * (md/{resync,rebuild}_start) and imsm migration units, return
2085 * 0 for the 'not migrating' and 'unsupported migration' cases
2087 if (!dev
->vol
.migr_state
)
2090 switch (migr_type(dev
)) {
2091 case MIGR_GEN_MIGR
: {
2092 struct migr_record
*migr_rec
= super
->migr_rec
;
2093 return __le32_to_cpu(migr_rec
->blocks_per_unit
);
2098 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
2099 __u32 stripes_per_unit
;
2100 __u32 blocks_per_unit
;
2109 /* yes, this is really the translation of migr_units to
2110 * per-member blocks in the 'resync' case
2112 stripes_per_unit
= num_stripes_per_unit_resync(dev
);
2113 migr_chunk
= migr_strip_blocks_resync(dev
);
2114 disks
= imsm_num_data_members(dev
, MAP_0
);
2115 blocks_per_unit
= stripes_per_unit
* migr_chunk
* disks
;
2116 stripe
= __le16_to_cpu(map
->blocks_per_strip
) * disks
;
2117 segment
= blocks_per_unit
/ stripe
;
2118 block_rel
= blocks_per_unit
- segment
* stripe
;
2119 parity_depth
= parity_segment_depth(dev
);
2120 block_map
= map_migr_block(dev
, block_rel
);
2121 return block_map
+ parity_depth
* segment
;
2123 case MIGR_REBUILD
: {
2124 __u32 stripes_per_unit
;
2127 stripes_per_unit
= num_stripes_per_unit_rebuild(dev
);
2128 migr_chunk
= migr_strip_blocks_rebuild(dev
);
2129 return migr_chunk
* stripes_per_unit
;
2131 case MIGR_STATE_CHANGE
:
2137 static int imsm_level_to_layout(int level
)
2145 return ALGORITHM_LEFT_ASYMMETRIC
;
2152 /*******************************************************************************
2153 * Function: read_imsm_migr_rec
2154 * Description: Function reads imsm migration record from last sector of disk
2156 * fd : disk descriptor
2157 * super : metadata info
2161 ******************************************************************************/
2162 static int read_imsm_migr_rec(int fd
, struct intel_super
*super
)
2165 unsigned long long dsize
;
2167 get_dev_size(fd
, NULL
, &dsize
);
2168 if (lseek64(fd
, dsize
- MIGR_REC_POSITION
, SEEK_SET
) < 0) {
2170 Name
": Cannot seek to anchor block: %s\n",
2174 if (read(fd
, super
->migr_rec_buf
, MIGR_REC_BUF_SIZE
) !=
2175 MIGR_REC_BUF_SIZE
) {
2177 Name
": Cannot read migr record block: %s\n",
2187 static struct imsm_dev
*imsm_get_device_during_migration(
2188 struct intel_super
*super
)
2191 struct intel_dev
*dv
;
2193 for (dv
= super
->devlist
; dv
; dv
= dv
->next
) {
2194 if (is_gen_migration(dv
->dev
))
2200 /*******************************************************************************
2201 * Function: load_imsm_migr_rec
2202 * Description: Function reads imsm migration record (it is stored at the last
2205 * super : imsm internal array info
2206 * info : general array info
2210 * -2 : no migration in progress
2211 ******************************************************************************/
2212 static int load_imsm_migr_rec(struct intel_super
*super
, struct mdinfo
*info
)
2215 struct dl
*dl
= NULL
;
2219 struct imsm_dev
*dev
;
2220 struct imsm_map
*map
= NULL
;
2223 /* find map under migration */
2224 dev
= imsm_get_device_during_migration(super
);
2225 /* nothing to load,no migration in progress?
2229 map
= get_imsm_map(dev
, MAP_0
);
2232 for (sd
= info
->devs
; sd
; sd
= sd
->next
) {
2233 /* skip spare and failed disks
2235 if (sd
->disk
.raid_disk
< 0)
2237 /* read only from one of the first two slots */
2239 slot
= get_imsm_disk_slot(map
,
2240 sd
->disk
.raid_disk
);
2241 if ((map
== NULL
) || (slot
> 1) || (slot
< 0))
2244 sprintf(nm
, "%d:%d", sd
->disk
.major
, sd
->disk
.minor
);
2245 fd
= dev_open(nm
, O_RDONLY
);
2251 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
2252 /* skip spare and failed disks
2256 /* read only from one of the first two slots */
2258 slot
= get_imsm_disk_slot(map
, dl
->index
);
2259 if ((map
== NULL
) || (slot
> 1) || (slot
< 0))
2261 sprintf(nm
, "%d:%d", dl
->major
, dl
->minor
);
2262 fd
= dev_open(nm
, O_RDONLY
);
2269 retval
= read_imsm_migr_rec(fd
, super
);
2278 /*******************************************************************************
2279 * function: imsm_create_metadata_checkpoint_update
2280 * Description: It creates update for checkpoint change.
2282 * super : imsm internal array info
2283 * u : pointer to prepared update
2286 * If length is equal to 0, input pointer u contains no update
2287 ******************************************************************************/
2288 static int imsm_create_metadata_checkpoint_update(
2289 struct intel_super
*super
,
2290 struct imsm_update_general_migration_checkpoint
**u
)
2293 int update_memory_size
= 0;
2295 dprintf("imsm_create_metadata_checkpoint_update(enter)\n");
2301 /* size of all update data without anchor */
2302 update_memory_size
=
2303 sizeof(struct imsm_update_general_migration_checkpoint
);
2305 *u
= calloc(1, update_memory_size
);
2307 dprintf("error: cannot get memory for "
2308 "imsm_create_metadata_checkpoint_update update\n");
2311 (*u
)->type
= update_general_migration_checkpoint
;
2312 (*u
)->curr_migr_unit
= __le32_to_cpu(super
->migr_rec
->curr_migr_unit
);
2313 dprintf("imsm_create_metadata_checkpoint_update: prepared for %u\n",
2314 (*u
)->curr_migr_unit
);
2316 return update_memory_size
;
2320 static void imsm_update_metadata_locally(struct supertype
*st
,
2321 void *buf
, int len
);
2323 /*******************************************************************************
2324 * Function: write_imsm_migr_rec
2325 * Description: Function writes imsm migration record
2326 * (at the last sector of disk)
2328 * super : imsm internal array info
2332 ******************************************************************************/
2333 static int write_imsm_migr_rec(struct supertype
*st
)
2335 struct intel_super
*super
= st
->sb
;
2336 unsigned long long dsize
;
2342 struct imsm_update_general_migration_checkpoint
*u
;
2343 struct imsm_dev
*dev
;
2344 struct imsm_map
*map
= NULL
;
2346 /* find map under migration */
2347 dev
= imsm_get_device_during_migration(super
);
2348 /* if no migration, write buffer anyway to clear migr_record
2349 * on disk based on first available device
2352 dev
= get_imsm_dev(super
, super
->current_vol
< 0 ? 0 :
2353 super
->current_vol
);
2355 map
= get_imsm_map(dev
, MAP_0
);
2357 for (sd
= super
->disks
; sd
; sd
= sd
->next
) {
2360 /* skip failed and spare devices */
2363 /* write to 2 first slots only */
2365 slot
= get_imsm_disk_slot(map
, sd
->index
);
2366 if ((map
== NULL
) || (slot
> 1) || (slot
< 0))
2369 sprintf(nm
, "%d:%d", sd
->major
, sd
->minor
);
2370 fd
= dev_open(nm
, O_RDWR
);
2373 get_dev_size(fd
, NULL
, &dsize
);
2374 if (lseek64(fd
, dsize
- MIGR_REC_POSITION
, SEEK_SET
) < 0) {
2376 Name
": Cannot seek to anchor block: %s\n",
2380 if (write(fd
, super
->migr_rec_buf
, MIGR_REC_BUF_SIZE
) !=
2381 MIGR_REC_BUF_SIZE
) {
2383 Name
": Cannot write migr record block: %s\n",
2390 /* update checkpoint information in metadata */
2391 len
= imsm_create_metadata_checkpoint_update(super
, &u
);
2394 dprintf("imsm: Cannot prepare update\n");
2397 /* update metadata locally */
2398 imsm_update_metadata_locally(st
, u
, len
);
2399 /* and possibly remotely */
2400 if (st
->update_tail
) {
2401 append_metadata_update(st
, u
, len
);
2402 /* during reshape we do all work inside metadata handler
2403 * manage_reshape(), so metadata update has to be triggered
2406 flush_metadata_updates(st
);
2407 st
->update_tail
= &st
->updates
;
2417 #endif /* MDASSEMBLE */
2419 /* spare/missing disks activations are not allowe when
2420 * array/container performs reshape operation, because
2421 * all arrays in container works on the same disks set
2423 int imsm_reshape_blocks_arrays_changes(struct intel_super
*super
)
2426 struct intel_dev
*i_dev
;
2427 struct imsm_dev
*dev
;
2429 /* check whole container
2431 for (i_dev
= super
->devlist
; i_dev
; i_dev
= i_dev
->next
) {
2433 if (is_gen_migration(dev
)) {
2434 /* No repair during any migration in container
2443 static void getinfo_super_imsm_volume(struct supertype
*st
, struct mdinfo
*info
, char *dmap
)
2445 struct intel_super
*super
= st
->sb
;
2446 struct migr_record
*migr_rec
= super
->migr_rec
;
2447 struct imsm_dev
*dev
= get_imsm_dev(super
, super
->current_vol
);
2448 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
2449 struct imsm_map
*prev_map
= get_imsm_map(dev
, MAP_1
);
2450 struct imsm_map
*map_to_analyse
= map
;
2453 unsigned int component_size_alligment
;
2454 int map_disks
= info
->array
.raid_disks
;
2456 memset(info
, 0, sizeof(*info
));
2458 map_to_analyse
= prev_map
;
2460 dl
= super
->current_disk
;
2462 info
->container_member
= super
->current_vol
;
2463 info
->array
.raid_disks
= map
->num_members
;
2464 info
->array
.level
= get_imsm_raid_level(map_to_analyse
);
2465 info
->array
.layout
= imsm_level_to_layout(info
->array
.level
);
2466 info
->array
.md_minor
= -1;
2467 info
->array
.ctime
= 0;
2468 info
->array
.utime
= 0;
2469 info
->array
.chunk_size
=
2470 __le16_to_cpu(map_to_analyse
->blocks_per_strip
) << 9;
2471 info
->array
.state
= !dev
->vol
.dirty
;
2472 info
->custom_array_size
= __le32_to_cpu(dev
->size_high
);
2473 info
->custom_array_size
<<= 32;
2474 info
->custom_array_size
|= __le32_to_cpu(dev
->size_low
);
2475 info
->recovery_blocked
= imsm_reshape_blocks_arrays_changes(st
->sb
);
2477 if (is_gen_migration(dev
)) {
2478 info
->reshape_active
= 1;
2479 info
->new_level
= get_imsm_raid_level(map
);
2480 info
->new_layout
= imsm_level_to_layout(info
->new_level
);
2481 info
->new_chunk
= __le16_to_cpu(map
->blocks_per_strip
) << 9;
2482 info
->delta_disks
= map
->num_members
- prev_map
->num_members
;
2483 if (info
->delta_disks
) {
2484 /* this needs to be applied to every array
2487 info
->reshape_active
= CONTAINER_RESHAPE
;
2489 /* We shape information that we give to md might have to be
2490 * modify to cope with md's requirement for reshaping arrays.
2491 * For example, when reshaping a RAID0, md requires it to be
2492 * presented as a degraded RAID4.
2493 * Also if a RAID0 is migrating to a RAID5 we need to specify
2494 * the array as already being RAID5, but the 'before' layout
2495 * is a RAID4-like layout.
2497 switch (info
->array
.level
) {
2499 switch(info
->new_level
) {
2501 /* conversion is happening as RAID4 */
2502 info
->array
.level
= 4;
2503 info
->array
.raid_disks
+= 1;
2506 /* conversion is happening as RAID5 */
2507 info
->array
.level
= 5;
2508 info
->array
.layout
= ALGORITHM_PARITY_N
;
2509 info
->delta_disks
-= 1;
2512 /* FIXME error message */
2513 info
->array
.level
= UnSet
;
2519 info
->new_level
= UnSet
;
2520 info
->new_layout
= UnSet
;
2521 info
->new_chunk
= info
->array
.chunk_size
;
2522 info
->delta_disks
= 0;
2526 info
->disk
.major
= dl
->major
;
2527 info
->disk
.minor
= dl
->minor
;
2528 info
->disk
.number
= dl
->index
;
2529 info
->disk
.raid_disk
= get_imsm_disk_slot(map_to_analyse
,
2533 info
->data_offset
= pba_of_lba0(map_to_analyse
);
2534 info
->component_size
= blocks_per_member(map_to_analyse
);
2536 /* check component size aligment
2538 component_size_alligment
=
2539 info
->component_size
% (info
->array
.chunk_size
/512);
2541 if (component_size_alligment
&&
2542 (info
->array
.level
!= 1) && (info
->array
.level
!= UnSet
)) {
2543 dprintf("imsm: reported component size alligned from %llu ",
2544 info
->component_size
);
2545 info
->component_size
-= component_size_alligment
;
2546 dprintf("to %llu (%i).\n",
2547 info
->component_size
, component_size_alligment
);
2550 memset(info
->uuid
, 0, sizeof(info
->uuid
));
2551 info
->recovery_start
= MaxSector
;
2553 info
->reshape_progress
= 0;
2554 info
->resync_start
= MaxSector
;
2555 if ((map_to_analyse
->map_state
== IMSM_T_STATE_UNINITIALIZED
||
2557 imsm_reshape_blocks_arrays_changes(super
) == 0) {
2558 info
->resync_start
= 0;
2560 if (dev
->vol
.migr_state
) {
2561 switch (migr_type(dev
)) {
2564 __u64 blocks_per_unit
= blocks_per_migr_unit(super
,
2566 __u64 units
= __le32_to_cpu(dev
->vol
.curr_migr_unit
);
2568 info
->resync_start
= blocks_per_unit
* units
;
2571 case MIGR_GEN_MIGR
: {
2572 __u64 blocks_per_unit
= blocks_per_migr_unit(super
,
2574 __u64 units
= __le32_to_cpu(migr_rec
->curr_migr_unit
);
2575 unsigned long long array_blocks
;
2578 if (__le32_to_cpu(migr_rec
->ascending_migr
) &&
2580 (__le32_to_cpu(migr_rec
->num_migr_units
)-1)) &&
2581 (super
->migr_rec
->rec_status
==
2582 __cpu_to_le32(UNIT_SRC_IN_CP_AREA
)))
2585 info
->reshape_progress
= blocks_per_unit
* units
;
2587 dprintf("IMSM: General Migration checkpoint : %llu "
2588 "(%llu) -> read reshape progress : %llu\n",
2589 (unsigned long long)units
,
2590 (unsigned long long)blocks_per_unit
,
2591 info
->reshape_progress
);
2593 used_disks
= imsm_num_data_members(dev
, MAP_1
);
2594 if (used_disks
> 0) {
2595 array_blocks
= blocks_per_member(map
) *
2597 /* round array size down to closest MB
2599 info
->custom_array_size
= (array_blocks
2600 >> SECT_PER_MB_SHIFT
)
2601 << SECT_PER_MB_SHIFT
;
2605 /* we could emulate the checkpointing of
2606 * 'sync_action=check' migrations, but for now
2607 * we just immediately complete them
2610 /* this is handled by container_content_imsm() */
2611 case MIGR_STATE_CHANGE
:
2612 /* FIXME handle other migrations */
2614 /* we are not dirty, so... */
2615 info
->resync_start
= MaxSector
;
2619 strncpy(info
->name
, (char *) dev
->volume
, MAX_RAID_SERIAL_LEN
);
2620 info
->name
[MAX_RAID_SERIAL_LEN
] = 0;
2622 info
->array
.major_version
= -1;
2623 info
->array
.minor_version
= -2;
2624 devname
= devnum2devname(st
->container_dev
);
2625 *info
->text_version
= '\0';
2627 sprintf(info
->text_version
, "/%s/%d", devname
, info
->container_member
);
2629 info
->safe_mode_delay
= 4000; /* 4 secs like the Matrix driver */
2630 uuid_from_super_imsm(st
, info
->uuid
);
2634 for (i
=0; i
<map_disks
; i
++) {
2636 if (i
< info
->array
.raid_disks
) {
2637 struct imsm_disk
*dsk
;
2638 j
= get_imsm_disk_idx(dev
, i
, MAP_X
);
2639 dsk
= get_imsm_disk(super
, j
);
2640 if (dsk
&& (dsk
->status
& CONFIGURED_DISK
))
2647 static __u8
imsm_check_degraded(struct intel_super
*super
, struct imsm_dev
*dev
,
2648 int failed
, int look_in_map
);
2650 static int imsm_count_failed(struct intel_super
*super
, struct imsm_dev
*dev
,
2655 static void manage_second_map(struct intel_super
*super
, struct imsm_dev
*dev
)
2657 if (is_gen_migration(dev
)) {
2660 struct imsm_map
*map2
= get_imsm_map(dev
, MAP_1
);
2662 failed
= imsm_count_failed(super
, dev
, MAP_1
);
2663 map_state
= imsm_check_degraded(super
, dev
, failed
, MAP_1
);
2664 if (map2
->map_state
!= map_state
) {
2665 map2
->map_state
= map_state
;
2666 super
->updates_pending
++;
2672 static struct imsm_disk
*get_imsm_missing(struct intel_super
*super
, __u8 index
)
2676 for (d
= super
->missing
; d
; d
= d
->next
)
2677 if (d
->index
== index
)
2682 static void getinfo_super_imsm(struct supertype
*st
, struct mdinfo
*info
, char *map
)
2684 struct intel_super
*super
= st
->sb
;
2685 struct imsm_disk
*disk
;
2686 int map_disks
= info
->array
.raid_disks
;
2687 int max_enough
= -1;
2689 struct imsm_super
*mpb
;
2691 if (super
->current_vol
>= 0) {
2692 getinfo_super_imsm_volume(st
, info
, map
);
2695 memset(info
, 0, sizeof(*info
));
2697 /* Set raid_disks to zero so that Assemble will always pull in valid
2700 info
->array
.raid_disks
= 0;
2701 info
->array
.level
= LEVEL_CONTAINER
;
2702 info
->array
.layout
= 0;
2703 info
->array
.md_minor
= -1;
2704 info
->array
.ctime
= 0; /* N/A for imsm */
2705 info
->array
.utime
= 0;
2706 info
->array
.chunk_size
= 0;
2708 info
->disk
.major
= 0;
2709 info
->disk
.minor
= 0;
2710 info
->disk
.raid_disk
= -1;
2711 info
->reshape_active
= 0;
2712 info
->array
.major_version
= -1;
2713 info
->array
.minor_version
= -2;
2714 strcpy(info
->text_version
, "imsm");
2715 info
->safe_mode_delay
= 0;
2716 info
->disk
.number
= -1;
2717 info
->disk
.state
= 0;
2719 info
->recovery_start
= MaxSector
;
2720 info
->recovery_blocked
= imsm_reshape_blocks_arrays_changes(st
->sb
);
2722 /* do we have the all the insync disks that we expect? */
2723 mpb
= super
->anchor
;
2725 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
2726 struct imsm_dev
*dev
= get_imsm_dev(super
, i
);
2727 int failed
, enough
, j
, missing
= 0;
2728 struct imsm_map
*map
;
2731 failed
= imsm_count_failed(super
, dev
, MAP_0
);
2732 state
= imsm_check_degraded(super
, dev
, failed
, MAP_0
);
2733 map
= get_imsm_map(dev
, MAP_0
);
2735 /* any newly missing disks?
2736 * (catches single-degraded vs double-degraded)
2738 for (j
= 0; j
< map
->num_members
; j
++) {
2739 __u32 ord
= get_imsm_ord_tbl_ent(dev
, j
, MAP_0
);
2740 __u32 idx
= ord_to_idx(ord
);
2742 if (!(ord
& IMSM_ORD_REBUILD
) &&
2743 get_imsm_missing(super
, idx
)) {
2749 if (state
== IMSM_T_STATE_FAILED
)
2751 else if (state
== IMSM_T_STATE_DEGRADED
&&
2752 (state
!= map
->map_state
|| missing
))
2754 else /* we're normal, or already degraded */
2756 if (is_gen_migration(dev
) && missing
) {
2757 /* during general migration we need all disks
2758 * that process is running on.
2759 * No new missing disk is allowed.
2763 /* no more checks necessary
2767 /* in the missing/failed disk case check to see
2768 * if at least one array is runnable
2770 max_enough
= max(max_enough
, enough
);
2772 dprintf("%s: enough: %d\n", __func__
, max_enough
);
2773 info
->container_enough
= max_enough
;
2776 __u32 reserved
= imsm_reserved_sectors(super
, super
->disks
);
2778 disk
= &super
->disks
->disk
;
2779 info
->data_offset
= total_blocks(&super
->disks
->disk
) - reserved
;
2780 info
->component_size
= reserved
;
2781 info
->disk
.state
= is_configured(disk
) ? (1 << MD_DISK_ACTIVE
) : 0;
2782 /* we don't change info->disk.raid_disk here because
2783 * this state will be finalized in mdmon after we have
2784 * found the 'most fresh' version of the metadata
2786 info
->disk
.state
|= is_failed(disk
) ? (1 << MD_DISK_FAULTY
) : 0;
2787 info
->disk
.state
|= is_spare(disk
) ? 0 : (1 << MD_DISK_SYNC
);
2790 /* only call uuid_from_super_imsm when this disk is part of a populated container,
2791 * ->compare_super may have updated the 'num_raid_devs' field for spares
2793 if (info
->disk
.state
& (1 << MD_DISK_SYNC
) || super
->anchor
->num_raid_devs
)
2794 uuid_from_super_imsm(st
, info
->uuid
);
2796 memcpy(info
->uuid
, uuid_zero
, sizeof(uuid_zero
));
2798 /* I don't know how to compute 'map' on imsm, so use safe default */
2801 for (i
= 0; i
< map_disks
; i
++)
2807 /* allocates memory and fills disk in mdinfo structure
2808 * for each disk in array */
2809 struct mdinfo
*getinfo_super_disks_imsm(struct supertype
*st
)
2811 struct mdinfo
*mddev
= NULL
;
2812 struct intel_super
*super
= st
->sb
;
2813 struct imsm_disk
*disk
;
2816 if (!super
|| !super
->disks
)
2819 mddev
= malloc(sizeof(*mddev
));
2821 fprintf(stderr
, Name
": Failed to allocate memory.\n");
2824 memset(mddev
, 0, sizeof(*mddev
));
2828 tmp
= malloc(sizeof(*tmp
));
2830 fprintf(stderr
, Name
": Failed to allocate memory.\n");
2835 memset(tmp
, 0, sizeof(*tmp
));
2837 tmp
->next
= mddev
->devs
;
2839 tmp
->disk
.number
= count
++;
2840 tmp
->disk
.major
= dl
->major
;
2841 tmp
->disk
.minor
= dl
->minor
;
2842 tmp
->disk
.state
= is_configured(disk
) ?
2843 (1 << MD_DISK_ACTIVE
) : 0;
2844 tmp
->disk
.state
|= is_failed(disk
) ? (1 << MD_DISK_FAULTY
) : 0;
2845 tmp
->disk
.state
|= is_spare(disk
) ? 0 : (1 << MD_DISK_SYNC
);
2846 tmp
->disk
.raid_disk
= -1;
2852 static int update_super_imsm(struct supertype
*st
, struct mdinfo
*info
,
2853 char *update
, char *devname
, int verbose
,
2854 int uuid_set
, char *homehost
)
2856 /* For 'assemble' and 'force' we need to return non-zero if any
2857 * change was made. For others, the return value is ignored.
2858 * Update options are:
2859 * force-one : This device looks a bit old but needs to be included,
2860 * update age info appropriately.
2861 * assemble: clear any 'faulty' flag to allow this device to
2863 * force-array: Array is degraded but being forced, mark it clean
2864 * if that will be needed to assemble it.
2866 * newdev: not used ????
2867 * grow: Array has gained a new device - this is currently for
2869 * resync: mark as dirty so a resync will happen.
2870 * name: update the name - preserving the homehost
2871 * uuid: Change the uuid of the array to match watch is given
2873 * Following are not relevant for this imsm:
2874 * sparc2.2 : update from old dodgey metadata
2875 * super-minor: change the preferred_minor number
2876 * summaries: update redundant counters.
2877 * homehost: update the recorded homehost
2878 * _reshape_progress: record new reshape_progress position.
2881 struct intel_super
*super
= st
->sb
;
2882 struct imsm_super
*mpb
;
2884 /* we can only update container info */
2885 if (!super
|| super
->current_vol
>= 0 || !super
->anchor
)
2888 mpb
= super
->anchor
;
2890 if (strcmp(update
, "uuid") == 0) {
2891 /* We take this to mean that the family_num should be updated.
2892 * However that is much smaller than the uuid so we cannot really
2893 * allow an explicit uuid to be given. And it is hard to reliably
2895 * So if !uuid_set we know the current uuid is random and just used
2896 * the first 'int' and copy it to the other 3 positions.
2897 * Otherwise we require the 4 'int's to be the same as would be the
2898 * case if we are using a random uuid. So an explicit uuid will be
2899 * accepted as long as all for ints are the same... which shouldn't hurt
2902 info
->uuid
[1] = info
->uuid
[2] = info
->uuid
[3] = info
->uuid
[0];
2905 if (info
->uuid
[0] != info
->uuid
[1] ||
2906 info
->uuid
[1] != info
->uuid
[2] ||
2907 info
->uuid
[2] != info
->uuid
[3])
2913 mpb
->orig_family_num
= info
->uuid
[0];
2914 } else if (strcmp(update
, "assemble") == 0)
2919 /* successful update? recompute checksum */
2921 mpb
->check_sum
= __le32_to_cpu(__gen_imsm_checksum(mpb
));
2926 static size_t disks_to_mpb_size(int disks
)
2930 size
= sizeof(struct imsm_super
);
2931 size
+= (disks
- 1) * sizeof(struct imsm_disk
);
2932 size
+= 2 * sizeof(struct imsm_dev
);
2933 /* up to 2 maps per raid device (-2 for imsm_maps in imsm_dev */
2934 size
+= (4 - 2) * sizeof(struct imsm_map
);
2935 /* 4 possible disk_ord_tbl's */
2936 size
+= 4 * (disks
- 1) * sizeof(__u32
);
2941 static __u64
avail_size_imsm(struct supertype
*st
, __u64 devsize
)
2943 if (devsize
< (MPB_SECTOR_CNT
+ IMSM_RESERVED_SECTORS
))
2946 return devsize
- (MPB_SECTOR_CNT
+ IMSM_RESERVED_SECTORS
);
2949 static void free_devlist(struct intel_super
*super
)
2951 struct intel_dev
*dv
;
2953 while (super
->devlist
) {
2954 dv
= super
->devlist
->next
;
2955 free(super
->devlist
->dev
);
2956 free(super
->devlist
);
2957 super
->devlist
= dv
;
2961 static void imsm_copy_dev(struct imsm_dev
*dest
, struct imsm_dev
*src
)
2963 memcpy(dest
, src
, sizeof_imsm_dev(src
, 0));
2966 static int compare_super_imsm(struct supertype
*st
, struct supertype
*tst
)
2970 * 0 same, or first was empty, and second was copied
2971 * 1 second had wrong number
2973 * 3 wrong other info
2975 struct intel_super
*first
= st
->sb
;
2976 struct intel_super
*sec
= tst
->sb
;
2983 /* in platform dependent environment test if the disks
2984 * use the same Intel hba
2986 if (!check_env("IMSM_NO_PLATFORM")) {
2987 if (!first
->hba
|| !sec
->hba
||
2988 (first
->hba
->type
!= sec
->hba
->type
)) {
2990 "HBAs of devices does not match %s != %s\n",
2991 first
->hba
? get_sys_dev_type(first
->hba
->type
) : NULL
,
2992 sec
->hba
? get_sys_dev_type(sec
->hba
->type
) : NULL
);
2997 /* if an anchor does not have num_raid_devs set then it is a free
3000 if (first
->anchor
->num_raid_devs
> 0 &&
3001 sec
->anchor
->num_raid_devs
> 0) {
3002 /* Determine if these disks might ever have been
3003 * related. Further disambiguation can only take place
3004 * in load_super_imsm_all
3006 __u32 first_family
= first
->anchor
->orig_family_num
;
3007 __u32 sec_family
= sec
->anchor
->orig_family_num
;
3009 if (memcmp(first
->anchor
->sig
, sec
->anchor
->sig
,
3010 MAX_SIGNATURE_LENGTH
) != 0)
3013 if (first_family
== 0)
3014 first_family
= first
->anchor
->family_num
;
3015 if (sec_family
== 0)
3016 sec_family
= sec
->anchor
->family_num
;
3018 if (first_family
!= sec_family
)
3024 /* if 'first' is a spare promote it to a populated mpb with sec's
3027 if (first
->anchor
->num_raid_devs
== 0 &&
3028 sec
->anchor
->num_raid_devs
> 0) {
3030 struct intel_dev
*dv
;
3031 struct imsm_dev
*dev
;
3033 /* we need to copy raid device info from sec if an allocation
3034 * fails here we don't associate the spare
3036 for (i
= 0; i
< sec
->anchor
->num_raid_devs
; i
++) {
3037 dv
= malloc(sizeof(*dv
));
3040 dev
= malloc(sizeof_imsm_dev(get_imsm_dev(sec
, i
), 1));
3047 dv
->next
= first
->devlist
;
3048 first
->devlist
= dv
;
3050 if (i
< sec
->anchor
->num_raid_devs
) {
3051 /* allocation failure */
3052 free_devlist(first
);
3053 fprintf(stderr
, "imsm: failed to associate spare\n");
3056 first
->anchor
->num_raid_devs
= sec
->anchor
->num_raid_devs
;
3057 first
->anchor
->orig_family_num
= sec
->anchor
->orig_family_num
;
3058 first
->anchor
->family_num
= sec
->anchor
->family_num
;
3059 memcpy(first
->anchor
->sig
, sec
->anchor
->sig
, MAX_SIGNATURE_LENGTH
);
3060 for (i
= 0; i
< sec
->anchor
->num_raid_devs
; i
++)
3061 imsm_copy_dev(get_imsm_dev(first
, i
), get_imsm_dev(sec
, i
));
3067 static void fd2devname(int fd
, char *name
)
3071 char dname
[PATH_MAX
];
3076 if (fstat(fd
, &st
) != 0)
3078 sprintf(path
, "/sys/dev/block/%d:%d",
3079 major(st
.st_rdev
), minor(st
.st_rdev
));
3081 rv
= readlink(path
, dname
, sizeof(dname
)-1);
3086 nm
= strrchr(dname
, '/');
3089 snprintf(name
, MAX_RAID_SERIAL_LEN
, "/dev/%s", nm
);
3093 extern int scsi_get_serial(int fd
, void *buf
, size_t buf_len
);
3095 static int imsm_read_serial(int fd
, char *devname
,
3096 __u8 serial
[MAX_RAID_SERIAL_LEN
])
3098 unsigned char scsi_serial
[255];
3107 memset(scsi_serial
, 0, sizeof(scsi_serial
));
3109 rv
= scsi_get_serial(fd
, scsi_serial
, sizeof(scsi_serial
));
3111 if (rv
&& check_env("IMSM_DEVNAME_AS_SERIAL")) {
3112 memset(serial
, 0, MAX_RAID_SERIAL_LEN
);
3113 fd2devname(fd
, (char *) serial
);
3120 Name
": Failed to retrieve serial for %s\n",
3125 rsp_len
= scsi_serial
[3];
3129 Name
": Failed to retrieve serial for %s\n",
3133 rsp_buf
= (char *) &scsi_serial
[4];
3135 /* trim all whitespace and non-printable characters and convert
3138 for (i
= 0, dest
= rsp_buf
; i
< rsp_len
; i
++) {
3141 /* ':' is reserved for use in placeholder serial
3142 * numbers for missing disks
3150 len
= dest
- rsp_buf
;
3153 /* truncate leading characters */
3154 if (len
> MAX_RAID_SERIAL_LEN
) {
3155 dest
+= len
- MAX_RAID_SERIAL_LEN
;
3156 len
= MAX_RAID_SERIAL_LEN
;
3159 memset(serial
, 0, MAX_RAID_SERIAL_LEN
);
3160 memcpy(serial
, dest
, len
);
3165 static int serialcmp(__u8
*s1
, __u8
*s2
)
3167 return strncmp((char *) s1
, (char *) s2
, MAX_RAID_SERIAL_LEN
);
3170 static void serialcpy(__u8
*dest
, __u8
*src
)
3172 strncpy((char *) dest
, (char *) src
, MAX_RAID_SERIAL_LEN
);
3175 static struct dl
*serial_to_dl(__u8
*serial
, struct intel_super
*super
)
3179 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
3180 if (serialcmp(dl
->serial
, serial
) == 0)
3186 static struct imsm_disk
*
3187 __serial_to_disk(__u8
*serial
, struct imsm_super
*mpb
, int *idx
)
3191 for (i
= 0; i
< mpb
->num_disks
; i
++) {
3192 struct imsm_disk
*disk
= __get_imsm_disk(mpb
, i
);
3194 if (serialcmp(disk
->serial
, serial
) == 0) {
3205 load_imsm_disk(int fd
, struct intel_super
*super
, char *devname
, int keep_fd
)
3207 struct imsm_disk
*disk
;
3212 __u8 serial
[MAX_RAID_SERIAL_LEN
];
3214 rv
= imsm_read_serial(fd
, devname
, serial
);
3219 dl
= calloc(1, sizeof(*dl
));
3223 Name
": failed to allocate disk buffer for %s\n",
3229 dl
->major
= major(stb
.st_rdev
);
3230 dl
->minor
= minor(stb
.st_rdev
);
3231 dl
->next
= super
->disks
;
3232 dl
->fd
= keep_fd
? fd
: -1;
3233 assert(super
->disks
== NULL
);
3235 serialcpy(dl
->serial
, serial
);
3238 fd2devname(fd
, name
);
3240 dl
->devname
= strdup(devname
);
3242 dl
->devname
= strdup(name
);
3244 /* look up this disk's index in the current anchor */
3245 disk
= __serial_to_disk(dl
->serial
, super
->anchor
, &dl
->index
);
3248 /* only set index on disks that are a member of a
3249 * populated contianer, i.e. one with raid_devs
3251 if (is_failed(&dl
->disk
))
3253 else if (is_spare(&dl
->disk
))
3261 /* When migrating map0 contains the 'destination' state while map1
3262 * contains the current state. When not migrating map0 contains the
3263 * current state. This routine assumes that map[0].map_state is set to
3264 * the current array state before being called.
3266 * Migration is indicated by one of the following states
3267 * 1/ Idle (migr_state=0 map0state=normal||unitialized||degraded||failed)
3268 * 2/ Initialize (migr_state=1 migr_type=MIGR_INIT map0state=normal
3269 * map1state=unitialized)
3270 * 3/ Repair (Resync) (migr_state=1 migr_type=MIGR_REPAIR map0state=normal
3272 * 4/ Rebuild (migr_state=1 migr_type=MIGR_REBUILD map0state=normal
3273 * map1state=degraded)
3274 * 5/ Migration (mig_state=1 migr_type=MIGR_GEN_MIGR map0state=normal
3277 static void migrate(struct imsm_dev
*dev
, struct intel_super
*super
,
3278 __u8 to_state
, int migr_type
)
3280 struct imsm_map
*dest
;
3281 struct imsm_map
*src
= get_imsm_map(dev
, MAP_0
);
3283 dev
->vol
.migr_state
= 1;
3284 set_migr_type(dev
, migr_type
);
3285 dev
->vol
.curr_migr_unit
= 0;
3286 dest
= get_imsm_map(dev
, MAP_1
);
3288 /* duplicate and then set the target end state in map[0] */
3289 memcpy(dest
, src
, sizeof_imsm_map(src
));
3290 if ((migr_type
== MIGR_REBUILD
) ||
3291 (migr_type
== MIGR_GEN_MIGR
)) {
3295 for (i
= 0; i
< src
->num_members
; i
++) {
3296 ord
= __le32_to_cpu(src
->disk_ord_tbl
[i
]);
3297 set_imsm_ord_tbl_ent(src
, i
, ord_to_idx(ord
));
3301 if (migr_type
== MIGR_GEN_MIGR
)
3302 /* Clear migration record */
3303 memset(super
->migr_rec
, 0, sizeof(struct migr_record
));
3305 src
->map_state
= to_state
;
3308 static void end_migration(struct imsm_dev
*dev
, struct intel_super
*super
,
3311 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
3312 struct imsm_map
*prev
= get_imsm_map(dev
, dev
->vol
.migr_state
== 0 ?
3316 /* merge any IMSM_ORD_REBUILD bits that were not successfully
3317 * completed in the last migration.
3319 * FIXME add support for raid-level-migration
3321 if ((map_state
!= map
->map_state
) && (is_gen_migration(dev
) == 0) &&
3322 (prev
->map_state
!= IMSM_T_STATE_UNINITIALIZED
)) {
3323 /* when final map state is other than expected
3324 * merge maps (not for migration)
3328 for (i
= 0; i
< prev
->num_members
; i
++)
3329 for (j
= 0; j
< map
->num_members
; j
++)
3330 /* during online capacity expansion
3331 * disks position can be changed
3332 * if takeover is used
3334 if (ord_to_idx(map
->disk_ord_tbl
[j
]) ==
3335 ord_to_idx(prev
->disk_ord_tbl
[i
])) {
3336 map
->disk_ord_tbl
[j
] |=
3337 prev
->disk_ord_tbl
[i
];
3340 failed
= imsm_count_failed(super
, dev
, MAP_0
);
3341 map_state
= imsm_check_degraded(super
, dev
, failed
, MAP_0
);
3344 dev
->vol
.migr_state
= 0;
3345 set_migr_type(dev
, 0);
3346 dev
->vol
.curr_migr_unit
= 0;
3347 map
->map_state
= map_state
;
3351 static int parse_raid_devices(struct intel_super
*super
)
3354 struct imsm_dev
*dev_new
;
3355 size_t len
, len_migr
;
3357 size_t space_needed
= 0;
3358 struct imsm_super
*mpb
= super
->anchor
;
3360 for (i
= 0; i
< super
->anchor
->num_raid_devs
; i
++) {
3361 struct imsm_dev
*dev_iter
= __get_imsm_dev(super
->anchor
, i
);
3362 struct intel_dev
*dv
;
3364 len
= sizeof_imsm_dev(dev_iter
, 0);
3365 len_migr
= sizeof_imsm_dev(dev_iter
, 1);
3367 space_needed
+= len_migr
- len
;
3369 dv
= malloc(sizeof(*dv
));
3372 if (max_len
< len_migr
)
3374 if (max_len
> len_migr
)
3375 space_needed
+= max_len
- len_migr
;
3376 dev_new
= malloc(max_len
);
3381 imsm_copy_dev(dev_new
, dev_iter
);
3384 dv
->next
= super
->devlist
;
3385 super
->devlist
= dv
;
3388 /* ensure that super->buf is large enough when all raid devices
3391 if (__le32_to_cpu(mpb
->mpb_size
) + space_needed
> super
->len
) {
3394 len
= ROUND_UP(__le32_to_cpu(mpb
->mpb_size
) + space_needed
, 512);
3395 if (posix_memalign(&buf
, 512, len
) != 0)
3398 memcpy(buf
, super
->buf
, super
->len
);
3399 memset(buf
+ super
->len
, 0, len
- super
->len
);
3408 /* retrieve a pointer to the bbm log which starts after all raid devices */
3409 struct bbm_log
*__get_imsm_bbm_log(struct imsm_super
*mpb
)
3413 if (__le32_to_cpu(mpb
->bbm_log_size
)) {
3415 ptr
+= mpb
->mpb_size
- __le32_to_cpu(mpb
->bbm_log_size
);
3421 /*******************************************************************************
3422 * Function: check_mpb_migr_compatibility
3423 * Description: Function checks for unsupported migration features:
3424 * - migration optimization area (pba_of_lba0)
3425 * - descending reshape (ascending_migr)
3427 * super : imsm metadata information
3429 * 0 : migration is compatible
3430 * -1 : migration is not compatible
3431 ******************************************************************************/
3432 int check_mpb_migr_compatibility(struct intel_super
*super
)
3434 struct imsm_map
*map0
, *map1
;
3435 struct migr_record
*migr_rec
= super
->migr_rec
;
3438 for (i
= 0; i
< super
->anchor
->num_raid_devs
; i
++) {
3439 struct imsm_dev
*dev_iter
= __get_imsm_dev(super
->anchor
, i
);
3442 dev_iter
->vol
.migr_state
== 1 &&
3443 dev_iter
->vol
.migr_type
== MIGR_GEN_MIGR
) {
3444 /* This device is migrating */
3445 map0
= get_imsm_map(dev_iter
, MAP_0
);
3446 map1
= get_imsm_map(dev_iter
, MAP_1
);
3447 if (pba_of_lba0(map0
) != pba_of_lba0(map1
))
3448 /* migration optimization area was used */
3450 if (migr_rec
->ascending_migr
== 0
3451 && migr_rec
->dest_depth_per_unit
> 0)
3452 /* descending reshape not supported yet */
3459 static void __free_imsm(struct intel_super
*super
, int free_disks
);
3461 /* load_imsm_mpb - read matrix metadata
3462 * allocates super->mpb to be freed by free_imsm
3464 static int load_imsm_mpb(int fd
, struct intel_super
*super
, char *devname
)
3466 unsigned long long dsize
;
3467 unsigned long long sectors
;
3469 struct imsm_super
*anchor
;
3472 get_dev_size(fd
, NULL
, &dsize
);
3476 Name
": %s: device to small for imsm\n",
3481 if (lseek64(fd
, dsize
- (512 * 2), SEEK_SET
) < 0) {
3483 fprintf(stderr
, Name
3484 ": Cannot seek to anchor block on %s: %s\n",
3485 devname
, strerror(errno
));
3489 if (posix_memalign((void**)&anchor
, 512, 512) != 0) {
3492 Name
": Failed to allocate imsm anchor buffer"
3493 " on %s\n", devname
);
3496 if (read(fd
, anchor
, 512) != 512) {
3499 Name
": Cannot read anchor block on %s: %s\n",
3500 devname
, strerror(errno
));
3505 if (strncmp((char *) anchor
->sig
, MPB_SIGNATURE
, MPB_SIG_LEN
) != 0) {
3508 Name
": no IMSM anchor on %s\n", devname
);
3513 __free_imsm(super
, 0);
3514 /* reload capability and hba */
3516 /* capability and hba must be updated with new super allocation */
3517 find_intel_hba_capability(fd
, super
, devname
);
3518 super
->len
= ROUND_UP(anchor
->mpb_size
, 512);
3519 if (posix_memalign(&super
->buf
, 512, super
->len
) != 0) {
3522 Name
": unable to allocate %zu byte mpb buffer\n",
3527 memcpy(super
->buf
, anchor
, 512);
3529 sectors
= mpb_sectors(anchor
) - 1;
3532 if (posix_memalign(&super
->migr_rec_buf
, 512, MIGR_REC_BUF_SIZE
) != 0) {
3533 fprintf(stderr
, Name
3534 ": %s could not allocate migr_rec buffer\n", __func__
);
3538 super
->clean_migration_record_by_mdmon
= 0;
3541 check_sum
= __gen_imsm_checksum(super
->anchor
);
3542 if (check_sum
!= __le32_to_cpu(super
->anchor
->check_sum
)) {
3545 Name
": IMSM checksum %x != %x on %s\n",
3547 __le32_to_cpu(super
->anchor
->check_sum
),
3555 /* read the extended mpb */
3556 if (lseek64(fd
, dsize
- (512 * (2 + sectors
)), SEEK_SET
) < 0) {
3559 Name
": Cannot seek to extended mpb on %s: %s\n",
3560 devname
, strerror(errno
));
3564 if ((unsigned)read(fd
, super
->buf
+ 512, super
->len
- 512) != super
->len
- 512) {
3567 Name
": Cannot read extended mpb on %s: %s\n",
3568 devname
, strerror(errno
));
3572 check_sum
= __gen_imsm_checksum(super
->anchor
);
3573 if (check_sum
!= __le32_to_cpu(super
->anchor
->check_sum
)) {
3576 Name
": IMSM checksum %x != %x on %s\n",
3577 check_sum
, __le32_to_cpu(super
->anchor
->check_sum
),
3582 /* FIXME the BBM log is disk specific so we cannot use this global
3583 * buffer for all disks. Ok for now since we only look at the global
3584 * bbm_log_size parameter to gate assembly
3586 super
->bbm_log
= __get_imsm_bbm_log(super
->anchor
);
3591 static int read_imsm_migr_rec(int fd
, struct intel_super
*super
);
3593 /* clears hi bits in metadata if MPB_ATTRIB_2TB_DISK not set */
3594 static void clear_hi(struct intel_super
*super
)
3596 struct imsm_super
*mpb
= super
->anchor
;
3598 if (mpb
->attributes
& MPB_ATTRIB_2TB_DISK
)
3600 for (i
= 0; i
< mpb
->num_disks
; ++i
) {
3601 struct imsm_disk
*disk
= &mpb
->disk
[i
];
3602 disk
->total_blocks_hi
= 0;
3604 for (i
= 0; i
< mpb
->num_raid_devs
; ++i
) {
3605 struct imsm_dev
*dev
= get_imsm_dev(super
, i
);
3608 for (n
= 0; n
< 2; ++n
) {
3609 struct imsm_map
*map
= get_imsm_map(dev
, n
);
3612 map
->pba_of_lba0_hi
= 0;
3613 map
->blocks_per_member_hi
= 0;
3614 map
->num_data_stripes_hi
= 0;
3620 load_and_parse_mpb(int fd
, struct intel_super
*super
, char *devname
, int keep_fd
)
3624 err
= load_imsm_mpb(fd
, super
, devname
);
3627 err
= load_imsm_disk(fd
, super
, devname
, keep_fd
);
3630 err
= parse_raid_devices(super
);
3635 static void __free_imsm_disk(struct dl
*d
)
3647 static void free_imsm_disks(struct intel_super
*super
)
3651 while (super
->disks
) {
3653 super
->disks
= d
->next
;
3654 __free_imsm_disk(d
);
3656 while (super
->disk_mgmt_list
) {
3657 d
= super
->disk_mgmt_list
;
3658 super
->disk_mgmt_list
= d
->next
;
3659 __free_imsm_disk(d
);
3661 while (super
->missing
) {
3663 super
->missing
= d
->next
;
3664 __free_imsm_disk(d
);
3669 /* free all the pieces hanging off of a super pointer */
3670 static void __free_imsm(struct intel_super
*super
, int free_disks
)
3672 struct intel_hba
*elem
, *next
;
3678 /* unlink capability description */
3680 if (super
->migr_rec_buf
) {
3681 free(super
->migr_rec_buf
);
3682 super
->migr_rec_buf
= NULL
;
3685 free_imsm_disks(super
);
3686 free_devlist(super
);
3690 free((void *)elem
->path
);
3698 static void free_imsm(struct intel_super
*super
)
3700 __free_imsm(super
, 1);
3704 static void free_super_imsm(struct supertype
*st
)
3706 struct intel_super
*super
= st
->sb
;
3715 static struct intel_super
*alloc_super(void)
3717 struct intel_super
*super
= malloc(sizeof(*super
));
3720 memset(super
, 0, sizeof(*super
));
3721 super
->current_vol
= -1;
3722 super
->create_offset
= ~((unsigned long long) 0);
3728 * find and allocate hba and OROM/EFI based on valid fd of RAID component device
3730 static int find_intel_hba_capability(int fd
, struct intel_super
*super
, char *devname
)
3732 struct sys_dev
*hba_name
;
3735 if ((fd
< 0) || check_env("IMSM_NO_PLATFORM")) {
3740 hba_name
= find_disk_attached_hba(fd
, NULL
);
3744 Name
": %s is not attached to Intel(R) RAID controller.\n",
3748 rv
= attach_hba_to_super(super
, hba_name
);
3751 struct intel_hba
*hba
= super
->hba
;
3753 fprintf(stderr
, Name
": %s is attached to Intel(R) %s RAID "
3754 "controller (%s),\n"
3755 " but the container is assigned to Intel(R) "
3756 "%s RAID controller (",
3759 hba_name
->pci_id
? : "Err!",
3760 get_sys_dev_type(hba_name
->type
));
3763 fprintf(stderr
, "%s", hba
->pci_id
? : "Err!");
3765 fprintf(stderr
, ", ");
3769 fprintf(stderr
, ").\n"
3770 " Mixing devices attached to different controllers "
3771 "is not allowed.\n");
3773 free_sys_dev(&hba_name
);
3776 super
->orom
= find_imsm_capability(hba_name
->type
);
3777 free_sys_dev(&hba_name
);
3783 /* find_missing - helper routine for load_super_imsm_all that identifies
3784 * disks that have disappeared from the system. This routine relies on
3785 * the mpb being uptodate, which it is at load time.
3787 static int find_missing(struct intel_super
*super
)
3790 struct imsm_super
*mpb
= super
->anchor
;
3792 struct imsm_disk
*disk
;
3794 for (i
= 0; i
< mpb
->num_disks
; i
++) {
3795 disk
= __get_imsm_disk(mpb
, i
);
3796 dl
= serial_to_dl(disk
->serial
, super
);
3800 dl
= malloc(sizeof(*dl
));
3806 dl
->devname
= strdup("missing");
3808 serialcpy(dl
->serial
, disk
->serial
);
3811 dl
->next
= super
->missing
;
3812 super
->missing
= dl
;
3819 static struct intel_disk
*disk_list_get(__u8
*serial
, struct intel_disk
*disk_list
)
3821 struct intel_disk
*idisk
= disk_list
;
3824 if (serialcmp(idisk
->disk
.serial
, serial
) == 0)
3826 idisk
= idisk
->next
;
3832 static int __prep_thunderdome(struct intel_super
**table
, int tbl_size
,
3833 struct intel_super
*super
,
3834 struct intel_disk
**disk_list
)
3836 struct imsm_disk
*d
= &super
->disks
->disk
;
3837 struct imsm_super
*mpb
= super
->anchor
;
3840 for (i
= 0; i
< tbl_size
; i
++) {
3841 struct imsm_super
*tbl_mpb
= table
[i
]->anchor
;
3842 struct imsm_disk
*tbl_d
= &table
[i
]->disks
->disk
;
3844 if (tbl_mpb
->family_num
== mpb
->family_num
) {
3845 if (tbl_mpb
->check_sum
== mpb
->check_sum
) {
3846 dprintf("%s: mpb from %d:%d matches %d:%d\n",
3847 __func__
, super
->disks
->major
,
3848 super
->disks
->minor
,
3849 table
[i
]->disks
->major
,
3850 table
[i
]->disks
->minor
);
3854 if (((is_configured(d
) && !is_configured(tbl_d
)) ||
3855 is_configured(d
) == is_configured(tbl_d
)) &&
3856 tbl_mpb
->generation_num
< mpb
->generation_num
) {
3857 /* current version of the mpb is a
3858 * better candidate than the one in
3859 * super_table, but copy over "cross
3860 * generational" status
3862 struct intel_disk
*idisk
;
3864 dprintf("%s: mpb from %d:%d replaces %d:%d\n",
3865 __func__
, super
->disks
->major
,
3866 super
->disks
->minor
,
3867 table
[i
]->disks
->major
,
3868 table
[i
]->disks
->minor
);
3870 idisk
= disk_list_get(tbl_d
->serial
, *disk_list
);
3871 if (idisk
&& is_failed(&idisk
->disk
))
3872 tbl_d
->status
|= FAILED_DISK
;
3875 struct intel_disk
*idisk
;
3876 struct imsm_disk
*disk
;
3878 /* tbl_mpb is more up to date, but copy
3879 * over cross generational status before
3882 disk
= __serial_to_disk(d
->serial
, mpb
, NULL
);
3883 if (disk
&& is_failed(disk
))
3884 d
->status
|= FAILED_DISK
;
3886 idisk
= disk_list_get(d
->serial
, *disk_list
);
3889 if (disk
&& is_configured(disk
))
3890 idisk
->disk
.status
|= CONFIGURED_DISK
;
3893 dprintf("%s: mpb from %d:%d prefer %d:%d\n",
3894 __func__
, super
->disks
->major
,
3895 super
->disks
->minor
,
3896 table
[i
]->disks
->major
,
3897 table
[i
]->disks
->minor
);
3905 table
[tbl_size
++] = super
;
3909 /* update/extend the merged list of imsm_disk records */
3910 for (j
= 0; j
< mpb
->num_disks
; j
++) {
3911 struct imsm_disk
*disk
= __get_imsm_disk(mpb
, j
);
3912 struct intel_disk
*idisk
;
3914 idisk
= disk_list_get(disk
->serial
, *disk_list
);
3916 idisk
->disk
.status
|= disk
->status
;
3917 if (is_configured(&idisk
->disk
) ||
3918 is_failed(&idisk
->disk
))
3919 idisk
->disk
.status
&= ~(SPARE_DISK
);
3921 idisk
= calloc(1, sizeof(*idisk
));
3924 idisk
->owner
= IMSM_UNKNOWN_OWNER
;
3925 idisk
->disk
= *disk
;
3926 idisk
->next
= *disk_list
;
3930 if (serialcmp(idisk
->disk
.serial
, d
->serial
) == 0)
3937 static struct intel_super
*
3938 validate_members(struct intel_super
*super
, struct intel_disk
*disk_list
,
3941 struct imsm_super
*mpb
= super
->anchor
;
3945 for (i
= 0; i
< mpb
->num_disks
; i
++) {
3946 struct imsm_disk
*disk
= __get_imsm_disk(mpb
, i
);
3947 struct intel_disk
*idisk
;
3949 idisk
= disk_list_get(disk
->serial
, disk_list
);
3951 if (idisk
->owner
== owner
||
3952 idisk
->owner
== IMSM_UNKNOWN_OWNER
)
3955 dprintf("%s: '%.16s' owner %d != %d\n",
3956 __func__
, disk
->serial
, idisk
->owner
,
3959 dprintf("%s: unknown disk %x [%d]: %.16s\n",
3960 __func__
, __le32_to_cpu(mpb
->family_num
), i
,
3966 if (ok_count
== mpb
->num_disks
)
3971 static void show_conflicts(__u32 family_num
, struct intel_super
*super_list
)
3973 struct intel_super
*s
;
3975 for (s
= super_list
; s
; s
= s
->next
) {
3976 if (family_num
!= s
->anchor
->family_num
)
3978 fprintf(stderr
, "Conflict, offlining family %#x on '%s'\n",
3979 __le32_to_cpu(family_num
), s
->disks
->devname
);
3983 static struct intel_super
*
3984 imsm_thunderdome(struct intel_super
**super_list
, int len
)
3986 struct intel_super
*super_table
[len
];
3987 struct intel_disk
*disk_list
= NULL
;
3988 struct intel_super
*champion
, *spare
;
3989 struct intel_super
*s
, **del
;
3994 memset(super_table
, 0, sizeof(super_table
));
3995 for (s
= *super_list
; s
; s
= s
->next
)
3996 tbl_size
= __prep_thunderdome(super_table
, tbl_size
, s
, &disk_list
);
3998 for (i
= 0; i
< tbl_size
; i
++) {
3999 struct imsm_disk
*d
;
4000 struct intel_disk
*idisk
;
4001 struct imsm_super
*mpb
= super_table
[i
]->anchor
;
4004 d
= &s
->disks
->disk
;
4006 /* 'd' must appear in merged disk list for its
4007 * configuration to be valid
4009 idisk
= disk_list_get(d
->serial
, disk_list
);
4010 if (idisk
&& idisk
->owner
== i
)
4011 s
= validate_members(s
, disk_list
, i
);
4016 dprintf("%s: marking family: %#x from %d:%d offline\n",
4017 __func__
, mpb
->family_num
,
4018 super_table
[i
]->disks
->major
,
4019 super_table
[i
]->disks
->minor
);
4023 /* This is where the mdadm implementation differs from the Windows
4024 * driver which has no strict concept of a container. We can only
4025 * assemble one family from a container, so when returning a prodigal
4026 * array member to this system the code will not be able to disambiguate
4027 * the container contents that should be assembled ("foreign" versus
4028 * "local"). It requires user intervention to set the orig_family_num
4029 * to a new value to establish a new container. The Windows driver in
4030 * this situation fixes up the volume name in place and manages the
4031 * foreign array as an independent entity.
4036 for (i
= 0; i
< tbl_size
; i
++) {
4037 struct intel_super
*tbl_ent
= super_table
[i
];
4043 if (tbl_ent
->anchor
->num_raid_devs
== 0) {
4048 if (s
&& !is_spare
) {
4049 show_conflicts(tbl_ent
->anchor
->family_num
, *super_list
);
4051 } else if (!s
&& !is_spare
)
4064 fprintf(stderr
, "Chose family %#x on '%s', "
4065 "assemble conflicts to new container with '--update=uuid'\n",
4066 __le32_to_cpu(s
->anchor
->family_num
), s
->disks
->devname
);
4068 /* collect all dl's onto 'champion', and update them to
4069 * champion's version of the status
4071 for (s
= *super_list
; s
; s
= s
->next
) {
4072 struct imsm_super
*mpb
= champion
->anchor
;
4073 struct dl
*dl
= s
->disks
;
4078 for (i
= 0; i
< mpb
->num_disks
; i
++) {
4079 struct imsm_disk
*disk
;
4081 disk
= __serial_to_disk(dl
->serial
, mpb
, &dl
->index
);
4084 /* only set index on disks that are a member of
4085 * a populated contianer, i.e. one with
4088 if (is_failed(&dl
->disk
))
4090 else if (is_spare(&dl
->disk
))
4096 if (i
>= mpb
->num_disks
) {
4097 struct intel_disk
*idisk
;
4099 idisk
= disk_list_get(dl
->serial
, disk_list
);
4100 if (idisk
&& is_spare(&idisk
->disk
) &&
4101 !is_failed(&idisk
->disk
) && !is_configured(&idisk
->disk
))
4109 dl
->next
= champion
->disks
;
4110 champion
->disks
= dl
;
4114 /* delete 'champion' from super_list */
4115 for (del
= super_list
; *del
; ) {
4116 if (*del
== champion
) {
4117 *del
= (*del
)->next
;
4120 del
= &(*del
)->next
;
4122 champion
->next
= NULL
;
4126 struct intel_disk
*idisk
= disk_list
;
4128 disk_list
= disk_list
->next
;
4137 get_sra_super_block(int fd
, struct intel_super
**super_list
, char *devname
, int *max
, int keep_fd
);
4138 static int get_super_block(struct intel_super
**super_list
, int devnum
, char *devname
,
4139 int major
, int minor
, int keep_fd
);
4141 get_devlist_super_block(struct md_list
*devlist
, struct intel_super
**super_list
,
4142 int *max
, int keep_fd
);
4145 static int load_super_imsm_all(struct supertype
*st
, int fd
, void **sbp
,
4146 char *devname
, struct md_list
*devlist
,
4149 struct intel_super
*super_list
= NULL
;
4150 struct intel_super
*super
= NULL
;
4155 /* 'fd' is an opened container */
4156 err
= get_sra_super_block(fd
, &super_list
, devname
, &i
, keep_fd
);
4158 /* get super block from devlist devices */
4159 err
= get_devlist_super_block(devlist
, &super_list
, &i
, keep_fd
);
4162 /* all mpbs enter, maybe one leaves */
4163 super
= imsm_thunderdome(&super_list
, i
);
4169 if (find_missing(super
) != 0) {
4175 /* load migration record */
4176 err
= load_imsm_migr_rec(super
, NULL
);
4178 /* migration is in progress,
4179 * but migr_rec cannot be loaded,
4185 /* Check migration compatibility */
4186 if ((err
== 0) && (check_mpb_migr_compatibility(super
) != 0)) {
4187 fprintf(stderr
, Name
": Unsupported migration detected");
4189 fprintf(stderr
, " on %s\n", devname
);
4191 fprintf(stderr
, " (IMSM).\n");
4200 while (super_list
) {
4201 struct intel_super
*s
= super_list
;
4203 super_list
= super_list
->next
;
4213 st
->container_dev
= fd2devnum(fd
);
4215 st
->container_dev
= NoMdDev
;
4216 if (err
== 0 && st
->ss
== NULL
) {
4217 st
->ss
= &super_imsm
;
4218 st
->minor_version
= 0;
4219 st
->max_devs
= IMSM_MAX_DEVICES
;
4226 get_devlist_super_block(struct md_list
*devlist
, struct intel_super
**super_list
,
4227 int *max
, int keep_fd
)
4229 struct md_list
*tmpdev
;
4233 for (i
= 0, tmpdev
= devlist
; tmpdev
; tmpdev
= tmpdev
->next
) {
4234 if (tmpdev
->used
!= 1)
4236 if (tmpdev
->container
== 1) {
4238 int fd
= dev_open(tmpdev
->devname
, O_RDONLY
|O_EXCL
);
4240 fprintf(stderr
, Name
": cannot open device %s: %s\n",
4241 tmpdev
->devname
, strerror(errno
));
4245 err
= get_sra_super_block(fd
, super_list
,
4246 tmpdev
->devname
, &lmax
,
4255 int major
= major(tmpdev
->st_rdev
);
4256 int minor
= minor(tmpdev
->st_rdev
);
4257 err
= get_super_block(super_list
,
4274 static int get_super_block(struct intel_super
**super_list
, int devnum
, char *devname
,
4275 int major
, int minor
, int keep_fd
)
4277 struct intel_super
*s
= NULL
;
4290 sprintf(nm
, "%d:%d", major
, minor
);
4291 dfd
= dev_open(nm
, O_RDWR
);
4297 rv
= find_intel_hba_capability(dfd
, s
, devname
);
4298 /* no orom/efi or non-intel hba of the disk */
4304 err
= load_and_parse_mpb(dfd
, s
, NULL
, keep_fd
);
4306 /* retry the load if we might have raced against mdmon */
4307 if (err
== 3 && (devnum
!= -1) && mdmon_running(devnum
))
4308 for (retry
= 0; retry
< 3; retry
++) {
4310 err
= load_and_parse_mpb(dfd
, s
, NULL
, keep_fd
);
4316 s
->next
= *super_list
;
4324 if ((dfd
>= 0) && (!keep_fd
))
4331 get_sra_super_block(int fd
, struct intel_super
**super_list
, char *devname
, int *max
, int keep_fd
)
4338 sra
= sysfs_read(fd
, 0, GET_LEVEL
|GET_VERSION
|GET_DEVS
|GET_STATE
);
4342 if (sra
->array
.major_version
!= -1 ||
4343 sra
->array
.minor_version
!= -2 ||
4344 strcmp(sra
->text_version
, "imsm") != 0) {
4349 devnum
= fd2devnum(fd
);
4350 for (sd
= sra
->devs
, i
= 0; sd
; sd
= sd
->next
, i
++) {
4351 if (get_super_block(super_list
, devnum
, devname
,
4352 sd
->disk
.major
, sd
->disk
.minor
, keep_fd
) != 0) {
4363 static int load_container_imsm(struct supertype
*st
, int fd
, char *devname
)
4365 return load_super_imsm_all(st
, fd
, &st
->sb
, devname
, NULL
, 1);
4369 static int load_super_imsm(struct supertype
*st
, int fd
, char *devname
)
4371 struct intel_super
*super
;
4374 if (test_partition(fd
))
4375 /* IMSM not allowed on partitions */
4378 free_super_imsm(st
);
4380 super
= alloc_super();
4383 Name
": malloc of %zu failed.\n",
4387 /* Load hba and capabilities if they exist.
4388 * But do not preclude loading metadata in case capabilities or hba are
4389 * non-compliant and ignore_hw_compat is set.
4391 rv
= find_intel_hba_capability(fd
, super
, devname
);
4392 /* no orom/efi or non-intel hba of the disk */
4393 if ((rv
!= 0) && (st
->ignore_hw_compat
== 0)) {
4396 Name
": No OROM/EFI properties for %s\n", devname
);
4400 rv
= load_and_parse_mpb(fd
, super
, devname
, 0);
4405 Name
": Failed to load all information "
4406 "sections on %s\n", devname
);
4412 if (st
->ss
== NULL
) {
4413 st
->ss
= &super_imsm
;
4414 st
->minor_version
= 0;
4415 st
->max_devs
= IMSM_MAX_DEVICES
;
4418 /* load migration record */
4419 if (load_imsm_migr_rec(super
, NULL
) == 0) {
4420 /* Check for unsupported migration features */
4421 if (check_mpb_migr_compatibility(super
) != 0) {
4423 Name
": Unsupported migration detected");
4425 fprintf(stderr
, " on %s\n", devname
);
4427 fprintf(stderr
, " (IMSM).\n");
4435 static __u16
info_to_blocks_per_strip(mdu_array_info_t
*info
)
4437 if (info
->level
== 1)
4439 return info
->chunk_size
>> 9;
4442 static unsigned long long info_to_blocks_per_member(mdu_array_info_t
*info
,
4443 unsigned long long size
)
4445 if (info
->level
== 1)
4448 return (size
* 2) & ~(info_to_blocks_per_strip(info
) - 1);
4451 static void imsm_update_version_info(struct intel_super
*super
)
4453 /* update the version and attributes */
4454 struct imsm_super
*mpb
= super
->anchor
;
4456 struct imsm_dev
*dev
;
4457 struct imsm_map
*map
;
4460 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
4461 dev
= get_imsm_dev(super
, i
);
4462 map
= get_imsm_map(dev
, MAP_0
);
4463 if (__le32_to_cpu(dev
->size_high
) > 0)
4464 mpb
->attributes
|= MPB_ATTRIB_2TB
;
4466 /* FIXME detect when an array spans a port multiplier */
4468 mpb
->attributes
|= MPB_ATTRIB_PM
;
4471 if (mpb
->num_raid_devs
> 1 ||
4472 mpb
->attributes
!= MPB_ATTRIB_CHECKSUM_VERIFY
) {
4473 version
= MPB_VERSION_ATTRIBS
;
4474 switch (get_imsm_raid_level(map
)) {
4475 case 0: mpb
->attributes
|= MPB_ATTRIB_RAID0
; break;
4476 case 1: mpb
->attributes
|= MPB_ATTRIB_RAID1
; break;
4477 case 10: mpb
->attributes
|= MPB_ATTRIB_RAID10
; break;
4478 case 5: mpb
->attributes
|= MPB_ATTRIB_RAID5
; break;
4481 if (map
->num_members
>= 5)
4482 version
= MPB_VERSION_5OR6_DISK_ARRAY
;
4483 else if (dev
->status
== DEV_CLONE_N_GO
)
4484 version
= MPB_VERSION_CNG
;
4485 else if (get_imsm_raid_level(map
) == 5)
4486 version
= MPB_VERSION_RAID5
;
4487 else if (map
->num_members
>= 3)
4488 version
= MPB_VERSION_3OR4_DISK_ARRAY
;
4489 else if (get_imsm_raid_level(map
) == 1)
4490 version
= MPB_VERSION_RAID1
;
4492 version
= MPB_VERSION_RAID0
;
4494 strcpy(((char *) mpb
->sig
) + strlen(MPB_SIGNATURE
), version
);
4498 static int check_name(struct intel_super
*super
, char *name
, int quiet
)
4500 struct imsm_super
*mpb
= super
->anchor
;
4501 char *reason
= NULL
;
4504 if (strlen(name
) > MAX_RAID_SERIAL_LEN
)
4505 reason
= "must be 16 characters or less";
4507 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
4508 struct imsm_dev
*dev
= get_imsm_dev(super
, i
);
4510 if (strncmp((char *) dev
->volume
, name
, MAX_RAID_SERIAL_LEN
) == 0) {
4511 reason
= "already exists";
4516 if (reason
&& !quiet
)
4517 fprintf(stderr
, Name
": imsm volume name %s\n", reason
);
4522 static int init_super_imsm_volume(struct supertype
*st
, mdu_array_info_t
*info
,
4523 unsigned long long size
, char *name
,
4524 char *homehost
, int *uuid
)
4526 /* We are creating a volume inside a pre-existing container.
4527 * so st->sb is already set.
4529 struct intel_super
*super
= st
->sb
;
4530 struct imsm_super
*mpb
= super
->anchor
;
4531 struct intel_dev
*dv
;
4532 struct imsm_dev
*dev
;
4533 struct imsm_vol
*vol
;
4534 struct imsm_map
*map
;
4535 int idx
= mpb
->num_raid_devs
;
4537 unsigned long long array_blocks
;
4538 size_t size_old
, size_new
;
4539 unsigned long long num_data_stripes
;
4541 if (super
->orom
&& mpb
->num_raid_devs
>= super
->orom
->vpa
) {
4542 fprintf(stderr
, Name
": This imsm-container already has the "
4543 "maximum of %d volumes\n", super
->orom
->vpa
);
4547 /* ensure the mpb is large enough for the new data */
4548 size_old
= __le32_to_cpu(mpb
->mpb_size
);
4549 size_new
= disks_to_mpb_size(info
->nr_disks
);
4550 if (size_new
> size_old
) {
4552 size_t size_round
= ROUND_UP(size_new
, 512);
4554 if (posix_memalign(&mpb_new
, 512, size_round
) != 0) {
4555 fprintf(stderr
, Name
": could not allocate new mpb\n");
4558 if (posix_memalign(&super
->migr_rec_buf
, 512,
4559 MIGR_REC_BUF_SIZE
) != 0) {
4560 fprintf(stderr
, Name
4561 ": %s could not allocate migr_rec buffer\n",
4568 memcpy(mpb_new
, mpb
, size_old
);
4571 super
->anchor
= mpb_new
;
4572 mpb
->mpb_size
= __cpu_to_le32(size_new
);
4573 memset(mpb_new
+ size_old
, 0, size_round
- size_old
);
4575 super
->current_vol
= idx
;
4577 /* handle 'failed_disks' by either:
4578 * a) create dummy disk entries in the table if this the first
4579 * volume in the array. We add them here as this is the only
4580 * opportunity to add them. add_to_super_imsm_volume()
4581 * handles the non-failed disks and continues incrementing
4583 * b) validate that 'failed_disks' matches the current number
4584 * of missing disks if the container is populated
4586 if (super
->current_vol
== 0) {
4588 for (i
= 0; i
< info
->failed_disks
; i
++) {
4589 struct imsm_disk
*disk
;
4592 disk
= __get_imsm_disk(mpb
, i
);
4593 disk
->status
= CONFIGURED_DISK
| FAILED_DISK
;
4594 disk
->scsi_id
= __cpu_to_le32(~(__u32
)0);
4595 snprintf((char *) disk
->serial
, MAX_RAID_SERIAL_LEN
,
4598 find_missing(super
);
4603 for (d
= super
->missing
; d
; d
= d
->next
)
4605 if (info
->failed_disks
> missing
) {
4606 fprintf(stderr
, Name
": unable to add 'missing' disk to container\n");
4611 if (!check_name(super
, name
, 0))
4613 dv
= malloc(sizeof(*dv
));
4615 fprintf(stderr
, Name
": failed to allocate device list entry\n");
4618 dev
= calloc(1, sizeof(*dev
) + sizeof(__u32
) * (info
->raid_disks
- 1));
4621 fprintf(stderr
, Name
": could not allocate raid device\n");
4625 strncpy((char *) dev
->volume
, name
, MAX_RAID_SERIAL_LEN
);
4626 array_blocks
= calc_array_size(info
->level
, info
->raid_disks
,
4627 info
->layout
, info
->chunk_size
,
4629 /* round array size down to closest MB */
4630 array_blocks
= (array_blocks
>> SECT_PER_MB_SHIFT
) << SECT_PER_MB_SHIFT
;
4632 dev
->size_low
= __cpu_to_le32((__u32
) array_blocks
);
4633 dev
->size_high
= __cpu_to_le32((__u32
) (array_blocks
>> 32));
4634 dev
->status
= (DEV_READ_COALESCING
| DEV_WRITE_COALESCING
);
4636 vol
->migr_state
= 0;
4637 set_migr_type(dev
, MIGR_INIT
);
4638 vol
->dirty
= !info
->state
;
4639 vol
->curr_migr_unit
= 0;
4640 map
= get_imsm_map(dev
, MAP_0
);
4641 set_pba_of_lba0(map
, super
->create_offset
);
4642 set_blocks_per_member(map
, info_to_blocks_per_member(info
, size
));
4643 map
->blocks_per_strip
= __cpu_to_le16(info_to_blocks_per_strip(info
));
4644 map
->failed_disk_num
= ~0;
4645 if (info
->level
> 0)
4646 map
->map_state
= IMSM_T_STATE_UNINITIALIZED
;
4648 map
->map_state
= info
->failed_disks
? IMSM_T_STATE_FAILED
:
4649 IMSM_T_STATE_NORMAL
;
4652 if (info
->level
== 1 && info
->raid_disks
> 2) {
4655 fprintf(stderr
, Name
": imsm does not support more than 2 disks"
4656 "in a raid1 volume\n");
4660 map
->raid_level
= info
->level
;
4661 if (info
->level
== 10) {
4662 map
->raid_level
= 1;
4663 map
->num_domains
= info
->raid_disks
/ 2;
4664 } else if (info
->level
== 1)
4665 map
->num_domains
= info
->raid_disks
;
4667 map
->num_domains
= 1;
4669 /* info->size is only int so use the 'size' parameter instead */
4670 num_data_stripes
= (size
* 2) / info_to_blocks_per_strip(info
);
4671 num_data_stripes
/= map
->num_domains
;
4672 set_num_data_stripes(map
, num_data_stripes
);
4674 map
->num_members
= info
->raid_disks
;
4675 for (i
= 0; i
< map
->num_members
; i
++) {
4676 /* initialized in add_to_super */
4677 set_imsm_ord_tbl_ent(map
, i
, IMSM_ORD_REBUILD
);
4679 mpb
->num_raid_devs
++;
4682 dv
->index
= super
->current_vol
;
4683 dv
->next
= super
->devlist
;
4684 super
->devlist
= dv
;
4686 imsm_update_version_info(super
);
4691 static int init_super_imsm(struct supertype
*st
, mdu_array_info_t
*info
,
4692 unsigned long long size
, char *name
,
4693 char *homehost
, int *uuid
)
4695 /* This is primarily called by Create when creating a new array.
4696 * We will then get add_to_super called for each component, and then
4697 * write_init_super called to write it out to each device.
4698 * For IMSM, Create can create on fresh devices or on a pre-existing
4700 * To create on a pre-existing array a different method will be called.
4701 * This one is just for fresh drives.
4703 struct intel_super
*super
;
4704 struct imsm_super
*mpb
;
4709 return init_super_imsm_volume(st
, info
, size
, name
, homehost
, uuid
);
4712 mpb_size
= disks_to_mpb_size(info
->nr_disks
);
4716 super
= alloc_super();
4717 if (super
&& posix_memalign(&super
->buf
, 512, mpb_size
) != 0) {
4722 fprintf(stderr
, Name
4723 ": %s could not allocate superblock\n", __func__
);
4726 if (posix_memalign(&super
->migr_rec_buf
, 512, MIGR_REC_BUF_SIZE
) != 0) {
4727 fprintf(stderr
, Name
4728 ": %s could not allocate migr_rec buffer\n", __func__
);
4733 memset(super
->buf
, 0, mpb_size
);
4735 mpb
->mpb_size
= __cpu_to_le32(mpb_size
);
4739 /* zeroing superblock */
4743 mpb
->attributes
= MPB_ATTRIB_CHECKSUM_VERIFY
;
4745 version
= (char *) mpb
->sig
;
4746 strcpy(version
, MPB_SIGNATURE
);
4747 version
+= strlen(MPB_SIGNATURE
);
4748 strcpy(version
, MPB_VERSION_RAID0
);
4754 static int add_to_super_imsm_volume(struct supertype
*st
, mdu_disk_info_t
*dk
,
4755 int fd
, char *devname
)
4757 struct intel_super
*super
= st
->sb
;
4758 struct imsm_super
*mpb
= super
->anchor
;
4759 struct imsm_disk
*_disk
;
4760 struct imsm_dev
*dev
;
4761 struct imsm_map
*map
;
4765 dev
= get_imsm_dev(super
, super
->current_vol
);
4766 map
= get_imsm_map(dev
, MAP_0
);
4768 if (! (dk
->state
& (1<<MD_DISK_SYNC
))) {
4769 fprintf(stderr
, Name
": %s: Cannot add spare devices to IMSM volume\n",
4775 /* we're doing autolayout so grab the pre-marked (in
4776 * validate_geometry) raid_disk
4778 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
4779 if (dl
->raiddisk
== dk
->raid_disk
)
4782 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
4783 if (dl
->major
== dk
->major
&&
4784 dl
->minor
== dk
->minor
)
4789 fprintf(stderr
, Name
": %s is not a member of the same container\n", devname
);
4793 /* add a pristine spare to the metadata */
4794 if (dl
->index
< 0) {
4795 dl
->index
= super
->anchor
->num_disks
;
4796 super
->anchor
->num_disks
++;
4798 /* Check the device has not already been added */
4799 slot
= get_imsm_disk_slot(map
, dl
->index
);
4801 (get_imsm_ord_tbl_ent(dev
, slot
, MAP_X
) & IMSM_ORD_REBUILD
) == 0) {
4802 fprintf(stderr
, Name
": %s has been included in this array twice\n",
4806 set_imsm_ord_tbl_ent(map
, dk
->raid_disk
, dl
->index
);
4807 dl
->disk
.status
= CONFIGURED_DISK
;
4809 /* update size of 'missing' disks to be at least as large as the
4810 * largest acitve member (we only have dummy missing disks when
4811 * creating the first volume)
4813 if (super
->current_vol
== 0) {
4814 for (df
= super
->missing
; df
; df
= df
->next
) {
4815 if (total_blocks(&dl
->disk
) > total_blocks(&df
->disk
))
4816 set_total_blocks(&df
->disk
, total_blocks(&dl
->disk
));
4817 _disk
= __get_imsm_disk(mpb
, df
->index
);
4822 /* refresh unset/failed slots to point to valid 'missing' entries */
4823 for (df
= super
->missing
; df
; df
= df
->next
)
4824 for (slot
= 0; slot
< mpb
->num_disks
; slot
++) {
4825 __u32 ord
= get_imsm_ord_tbl_ent(dev
, slot
, MAP_X
);
4827 if ((ord
& IMSM_ORD_REBUILD
) == 0)
4829 set_imsm_ord_tbl_ent(map
, slot
, df
->index
| IMSM_ORD_REBUILD
);
4830 if (is_gen_migration(dev
)) {
4831 struct imsm_map
*map2
= get_imsm_map(dev
,
4833 int slot2
= get_imsm_disk_slot(map2
, df
->index
);
4834 if ((slot2
< map2
->num_members
) &&
4836 __u32 ord2
= get_imsm_ord_tbl_ent(dev
,
4839 if ((unsigned)df
->index
==
4841 set_imsm_ord_tbl_ent(map2
,
4847 dprintf("set slot:%d to missing disk:%d\n", slot
, df
->index
);
4851 /* if we are creating the first raid device update the family number */
4852 if (super
->current_vol
== 0) {
4854 struct imsm_dev
*_dev
= __get_imsm_dev(mpb
, 0);
4856 _disk
= __get_imsm_disk(mpb
, dl
->index
);
4857 if (!_dev
|| !_disk
) {
4858 fprintf(stderr
, Name
": BUG mpb setup error\n");
4864 sum
+= __gen_imsm_checksum(mpb
);
4865 mpb
->family_num
= __cpu_to_le32(sum
);
4866 mpb
->orig_family_num
= mpb
->family_num
;
4868 super
->current_disk
= dl
;
4873 * Function marks disk as spare and restores disk serial
4874 * in case it was previously marked as failed by takeover operation
4876 * -1 : critical error
4877 * 0 : disk is marked as spare but serial is not set
4880 int mark_spare(struct dl
*disk
)
4882 __u8 serial
[MAX_RAID_SERIAL_LEN
];
4889 if (!imsm_read_serial(disk
->fd
, NULL
, serial
)) {
4890 /* Restore disk serial number, because takeover marks disk
4891 * as failed and adds to serial ':0' before it becomes
4894 serialcpy(disk
->serial
, serial
);
4895 serialcpy(disk
->disk
.serial
, serial
);
4898 disk
->disk
.status
= SPARE_DISK
;
4904 static int add_to_super_imsm(struct supertype
*st
, mdu_disk_info_t
*dk
,
4905 int fd
, char *devname
)
4907 struct intel_super
*super
= st
->sb
;
4909 unsigned long long size
;
4914 /* If we are on an RAID enabled platform check that the disk is
4915 * attached to the raid controller.
4916 * We do not need to test disks attachment for container based additions,
4917 * they shall be already tested when container was created/assembled.
4919 rv
= find_intel_hba_capability(fd
, super
, devname
);
4920 /* no orom/efi or non-intel hba of the disk */
4922 dprintf("capability: %p fd: %d ret: %d\n",
4923 super
->orom
, fd
, rv
);
4927 if (super
->current_vol
>= 0)
4928 return add_to_super_imsm_volume(st
, dk
, fd
, devname
);
4931 dd
= malloc(sizeof(*dd
));
4934 Name
": malloc failed %s:%d.\n", __func__
, __LINE__
);
4937 memset(dd
, 0, sizeof(*dd
));
4938 dd
->major
= major(stb
.st_rdev
);
4939 dd
->minor
= minor(stb
.st_rdev
);
4940 dd
->devname
= devname
? strdup(devname
) : NULL
;
4943 dd
->action
= DISK_ADD
;
4944 rv
= imsm_read_serial(fd
, devname
, dd
->serial
);
4947 Name
": failed to retrieve scsi serial, aborting\n");
4952 get_dev_size(fd
, NULL
, &size
);
4954 serialcpy(dd
->disk
.serial
, dd
->serial
);
4955 set_total_blocks(&dd
->disk
, size
);
4956 if (__le32_to_cpu(dd
->disk
.total_blocks_hi
) > 0) {
4957 struct imsm_super
*mpb
= super
->anchor
;
4958 mpb
->attributes
|= MPB_ATTRIB_2TB_DISK
;
4961 if (sysfs_disk_to_scsi_id(fd
, &id
) == 0)
4962 dd
->disk
.scsi_id
= __cpu_to_le32(id
);
4964 dd
->disk
.scsi_id
= __cpu_to_le32(0);
4966 if (st
->update_tail
) {
4967 dd
->next
= super
->disk_mgmt_list
;
4968 super
->disk_mgmt_list
= dd
;
4970 dd
->next
= super
->disks
;
4972 super
->updates_pending
++;
4979 static int remove_from_super_imsm(struct supertype
*st
, mdu_disk_info_t
*dk
)
4981 struct intel_super
*super
= st
->sb
;
4984 /* remove from super works only in mdmon - for communication
4985 * manager - monitor. Check if communication memory buffer
4988 if (!st
->update_tail
) {
4990 Name
": %s shall be used in mdmon context only"
4991 "(line %d).\n", __func__
, __LINE__
);
4994 dd
= malloc(sizeof(*dd
));
4997 Name
": malloc failed %s:%d.\n", __func__
, __LINE__
);
5000 memset(dd
, 0, sizeof(*dd
));
5001 dd
->major
= dk
->major
;
5002 dd
->minor
= dk
->minor
;
5005 dd
->action
= DISK_REMOVE
;
5007 dd
->next
= super
->disk_mgmt_list
;
5008 super
->disk_mgmt_list
= dd
;
5014 static int store_imsm_mpb(int fd
, struct imsm_super
*mpb
);
5018 struct imsm_super anchor
;
5019 } spare_record
__attribute__ ((aligned(512)));
5021 /* spare records have their own family number and do not have any defined raid
5024 static int write_super_imsm_spares(struct intel_super
*super
, int doclose
)
5026 struct imsm_super
*mpb
= super
->anchor
;
5027 struct imsm_super
*spare
= &spare_record
.anchor
;
5031 spare
->mpb_size
= __cpu_to_le32(sizeof(struct imsm_super
)),
5032 spare
->generation_num
= __cpu_to_le32(1UL),
5033 spare
->attributes
= MPB_ATTRIB_CHECKSUM_VERIFY
;
5034 spare
->num_disks
= 1,
5035 spare
->num_raid_devs
= 0,
5036 spare
->cache_size
= mpb
->cache_size
,
5037 spare
->pwr_cycle_count
= __cpu_to_le32(1),
5039 snprintf((char *) spare
->sig
, MAX_SIGNATURE_LENGTH
,
5040 MPB_SIGNATURE MPB_VERSION_RAID0
);
5042 for (d
= super
->disks
; d
; d
= d
->next
) {
5046 spare
->disk
[0] = d
->disk
;
5047 sum
= __gen_imsm_checksum(spare
);
5048 spare
->family_num
= __cpu_to_le32(sum
);
5049 spare
->orig_family_num
= 0;
5050 sum
= __gen_imsm_checksum(spare
);
5051 spare
->check_sum
= __cpu_to_le32(sum
);
5053 if (store_imsm_mpb(d
->fd
, spare
)) {
5054 fprintf(stderr
, "%s: failed for device %d:%d %s\n",
5055 __func__
, d
->major
, d
->minor
, strerror(errno
));
5067 static int write_super_imsm(struct supertype
*st
, int doclose
)
5069 struct intel_super
*super
= st
->sb
;
5070 struct imsm_super
*mpb
= super
->anchor
;
5076 __u32 mpb_size
= sizeof(struct imsm_super
) - sizeof(struct imsm_disk
);
5078 int clear_migration_record
= 1;
5080 /* 'generation' is incremented everytime the metadata is written */
5081 generation
= __le32_to_cpu(mpb
->generation_num
);
5083 mpb
->generation_num
= __cpu_to_le32(generation
);
5085 /* fix up cases where previous mdadm releases failed to set
5088 if (mpb
->orig_family_num
== 0)
5089 mpb
->orig_family_num
= mpb
->family_num
;
5091 for (d
= super
->disks
; d
; d
= d
->next
) {
5095 mpb
->disk
[d
->index
] = d
->disk
;
5099 for (d
= super
->missing
; d
; d
= d
->next
) {
5100 mpb
->disk
[d
->index
] = d
->disk
;
5103 mpb
->num_disks
= num_disks
;
5104 mpb_size
+= sizeof(struct imsm_disk
) * mpb
->num_disks
;
5106 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
5107 struct imsm_dev
*dev
= __get_imsm_dev(mpb
, i
);
5108 struct imsm_dev
*dev2
= get_imsm_dev(super
, i
);
5110 imsm_copy_dev(dev
, dev2
);
5111 mpb_size
+= sizeof_imsm_dev(dev
, 0);
5113 if (is_gen_migration(dev2
))
5114 clear_migration_record
= 0;
5116 mpb_size
+= __le32_to_cpu(mpb
->bbm_log_size
);
5117 mpb
->mpb_size
= __cpu_to_le32(mpb_size
);
5119 /* recalculate checksum */
5120 sum
= __gen_imsm_checksum(mpb
);
5121 mpb
->check_sum
= __cpu_to_le32(sum
);
5123 if (super
->clean_migration_record_by_mdmon
) {
5124 clear_migration_record
= 1;
5125 super
->clean_migration_record_by_mdmon
= 0;
5127 if (clear_migration_record
)
5128 memset(super
->migr_rec_buf
, 0, MIGR_REC_BUF_SIZE
);
5130 /* write the mpb for disks that compose raid devices */
5131 for (d
= super
->disks
; d
; d
= d
->next
) {
5132 if (d
->index
< 0 || is_failed(&d
->disk
))
5135 if (clear_migration_record
) {
5136 unsigned long long dsize
;
5138 get_dev_size(d
->fd
, NULL
, &dsize
);
5139 if (lseek64(d
->fd
, dsize
- 512, SEEK_SET
) >= 0) {
5140 if (write(d
->fd
, super
->migr_rec_buf
,
5141 MIGR_REC_BUF_SIZE
) != MIGR_REC_BUF_SIZE
)
5142 perror("Write migr_rec failed");
5146 if (store_imsm_mpb(d
->fd
, mpb
))
5148 "%s: failed for device %d:%d (fd: %d)%s\n",
5149 __func__
, d
->major
, d
->minor
,
5150 d
->fd
, strerror(errno
));
5159 return write_super_imsm_spares(super
, doclose
);
5165 static int create_array(struct supertype
*st
, int dev_idx
)
5168 struct imsm_update_create_array
*u
;
5169 struct intel_super
*super
= st
->sb
;
5170 struct imsm_dev
*dev
= get_imsm_dev(super
, dev_idx
);
5171 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
5172 struct disk_info
*inf
;
5173 struct imsm_disk
*disk
;
5176 len
= sizeof(*u
) - sizeof(*dev
) + sizeof_imsm_dev(dev
, 0) +
5177 sizeof(*inf
) * map
->num_members
;
5180 fprintf(stderr
, "%s: failed to allocate update buffer\n",
5185 u
->type
= update_create_array
;
5186 u
->dev_idx
= dev_idx
;
5187 imsm_copy_dev(&u
->dev
, dev
);
5188 inf
= get_disk_info(u
);
5189 for (i
= 0; i
< map
->num_members
; i
++) {
5190 int idx
= get_imsm_disk_idx(dev
, i
, MAP_X
);
5192 disk
= get_imsm_disk(super
, idx
);
5193 serialcpy(inf
[i
].serial
, disk
->serial
);
5195 append_metadata_update(st
, u
, len
);
5200 static int mgmt_disk(struct supertype
*st
)
5202 struct intel_super
*super
= st
->sb
;
5204 struct imsm_update_add_remove_disk
*u
;
5206 if (!super
->disk_mgmt_list
)
5212 fprintf(stderr
, "%s: failed to allocate update buffer\n",
5217 u
->type
= update_add_remove_disk
;
5218 append_metadata_update(st
, u
, len
);
5223 static int write_init_super_imsm(struct supertype
*st
)
5225 struct intel_super
*super
= st
->sb
;
5226 int current_vol
= super
->current_vol
;
5228 /* we are done with current_vol reset it to point st at the container */
5229 super
->current_vol
= -1;
5231 if (st
->update_tail
) {
5232 /* queue the recently created array / added disk
5233 * as a metadata update */
5236 /* determine if we are creating a volume or adding a disk */
5237 if (current_vol
< 0) {
5238 /* in the mgmt (add/remove) disk case we are running
5239 * in mdmon context, so don't close fd's
5241 return mgmt_disk(st
);
5243 rv
= create_array(st
, current_vol
);
5248 for (d
= super
->disks
; d
; d
= d
->next
)
5249 Kill(d
->devname
, NULL
, 0, 1, 1);
5250 return write_super_imsm(st
, 1);
5255 static int store_super_imsm(struct supertype
*st
, int fd
)
5257 struct intel_super
*super
= st
->sb
;
5258 struct imsm_super
*mpb
= super
? super
->anchor
: NULL
;
5264 return store_imsm_mpb(fd
, mpb
);
5270 static int imsm_bbm_log_size(struct imsm_super
*mpb
)
5272 return __le32_to_cpu(mpb
->bbm_log_size
);
5276 static int validate_geometry_imsm_container(struct supertype
*st
, int level
,
5277 int layout
, int raiddisks
, int chunk
,
5278 unsigned long long size
, char *dev
,
5279 unsigned long long *freesize
,
5283 unsigned long long ldsize
;
5284 struct intel_super
*super
=NULL
;
5287 if (level
!= LEVEL_CONTAINER
)
5292 fd
= open(dev
, O_RDONLY
|O_EXCL
, 0);
5295 fprintf(stderr
, Name
": imsm: Cannot open %s: %s\n",
5296 dev
, strerror(errno
));
5299 if (!get_dev_size(fd
, dev
, &ldsize
)) {
5304 /* capabilities retrieve could be possible
5305 * note that there is no fd for the disks in array.
5307 super
= alloc_super();
5310 Name
": malloc of %zu failed.\n",
5316 rv
= find_intel_hba_capability(fd
, super
, verbose
? dev
: NULL
);
5320 fd2devname(fd
, str
);
5321 dprintf("validate_geometry_imsm_container: fd: %d %s orom: %p rv: %d raiddisk: %d\n",
5322 fd
, str
, super
->orom
, rv
, raiddisks
);
5324 /* no orom/efi or non-intel hba of the disk */
5330 if (super
->orom
&& raiddisks
> super
->orom
->tds
) {
5332 fprintf(stderr
, Name
": %d exceeds maximum number of"
5333 " platform supported disks: %d\n",
5334 raiddisks
, super
->orom
->tds
);
5340 *freesize
= avail_size_imsm(st
, ldsize
>> 9);
5346 static unsigned long long find_size(struct extent
*e
, int *idx
, int num_extents
)
5348 const unsigned long long base_start
= e
[*idx
].start
;
5349 unsigned long long end
= base_start
+ e
[*idx
].size
;
5352 if (base_start
== end
)
5356 for (i
= *idx
; i
< num_extents
; i
++) {
5357 /* extend overlapping extents */
5358 if (e
[i
].start
>= base_start
&&
5359 e
[i
].start
<= end
) {
5362 if (e
[i
].start
+ e
[i
].size
> end
)
5363 end
= e
[i
].start
+ e
[i
].size
;
5364 } else if (e
[i
].start
> end
) {
5370 return end
- base_start
;
5373 static unsigned long long merge_extents(struct intel_super
*super
, int sum_extents
)
5375 /* build a composite disk with all known extents and generate a new
5376 * 'maxsize' given the "all disks in an array must share a common start
5377 * offset" constraint
5379 struct extent
*e
= calloc(sum_extents
, sizeof(*e
));
5383 unsigned long long pos
;
5384 unsigned long long start
= 0;
5385 unsigned long long maxsize
;
5386 unsigned long reserve
;
5391 /* coalesce and sort all extents. also, check to see if we need to
5392 * reserve space between member arrays
5395 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
5398 for (i
= 0; i
< dl
->extent_cnt
; i
++)
5401 qsort(e
, sum_extents
, sizeof(*e
), cmp_extent
);
5406 while (i
< sum_extents
) {
5407 e
[j
].start
= e
[i
].start
;
5408 e
[j
].size
= find_size(e
, &i
, sum_extents
);
5410 if (e
[j
-1].size
== 0)
5419 unsigned long long esize
;
5421 esize
= e
[i
].start
- pos
;
5422 if (esize
>= maxsize
) {
5427 pos
= e
[i
].start
+ e
[i
].size
;
5429 } while (e
[i
-1].size
);
5435 /* FIXME assumes volume at offset 0 is the first volume in a
5438 if (start_extent
> 0)
5439 reserve
= IMSM_RESERVED_SECTORS
; /* gap between raid regions */
5443 if (maxsize
< reserve
)
5446 super
->create_offset
= ~((unsigned long long) 0);
5447 if (start
+ reserve
> super
->create_offset
)
5448 return 0; /* start overflows create_offset */
5449 super
->create_offset
= start
+ reserve
;
5451 return maxsize
- reserve
;
5454 static int is_raid_level_supported(const struct imsm_orom
*orom
, int level
, int raiddisks
)
5456 if (level
< 0 || level
== 6 || level
== 4)
5459 /* if we have an orom prevent invalid raid levels */
5462 case 0: return imsm_orom_has_raid0(orom
);
5465 return imsm_orom_has_raid1e(orom
);
5466 return imsm_orom_has_raid1(orom
) && raiddisks
== 2;
5467 case 10: return imsm_orom_has_raid10(orom
) && raiddisks
== 4;
5468 case 5: return imsm_orom_has_raid5(orom
) && raiddisks
> 2;
5471 return 1; /* not on an Intel RAID platform so anything goes */
5478 active_arrays_by_format(char *name
, char* hba
, struct md_list
**devlist
,
5479 int dpa
, int verbose
)
5481 struct mdstat_ent
*mdstat
= mdstat_read(0, 0);
5482 struct mdstat_ent
*memb
= NULL
;
5485 struct md_list
*dv
= NULL
;
5488 for (memb
= mdstat
; memb
; memb
= memb
->next
) {
5489 if (memb
->metadata_version
&&
5490 (strncmp(memb
->metadata_version
, "external:", 9) == 0) &&
5491 (strcmp(&memb
->metadata_version
[9], name
) == 0) &&
5492 !is_subarray(memb
->metadata_version
+9) &&
5494 struct dev_member
*dev
= memb
->members
;
5496 while(dev
&& (fd
< 0)) {
5497 char *path
= malloc(strlen(dev
->name
) + strlen("/dev/") + 1);
5499 num
= sprintf(path
, "%s%s", "/dev/", dev
->name
);
5501 fd
= open(path
, O_RDONLY
, 0);
5502 if ((num
<= 0) || (fd
< 0)) {
5503 pr_vrb(": Cannot open %s: %s\n",
5504 dev
->name
, strerror(errno
));
5511 if ((fd
>= 0) && disk_attached_to_hba(fd
, hba
)) {
5512 struct mdstat_ent
*vol
;
5513 for (vol
= mdstat
; vol
; vol
= vol
->next
) {
5514 if ((vol
->active
> 0) &&
5515 vol
->metadata_version
&&
5516 is_container_member(vol
, memb
->dev
)) {
5521 if (*devlist
&& (found
< dpa
)) {
5522 dv
= calloc(1, sizeof(*dv
));
5524 fprintf(stderr
, Name
": calloc failed\n");
5526 dv
->devname
= malloc(strlen(memb
->dev
) + strlen("/dev/") + 1);
5527 if (dv
->devname
!= NULL
) {
5528 sprintf(dv
->devname
, "%s%s", "/dev/", memb
->dev
);
5531 dv
->next
= *devlist
;
5542 free_mdstat(mdstat
);
5547 static struct md_list
*
5548 get_loop_devices(void)
5551 struct md_list
*devlist
= NULL
;
5552 struct md_list
*dv
= NULL
;
5554 for(i
= 0; i
< 12; i
++) {
5555 dv
= calloc(1, sizeof(*dv
));
5557 fprintf(stderr
, Name
": calloc failed\n");
5560 dv
->devname
= malloc(40);
5561 if (dv
->devname
== NULL
) {
5562 fprintf(stderr
, Name
": malloc failed\n");
5566 sprintf(dv
->devname
, "/dev/loop%d", i
);
5574 static struct md_list
*
5575 get_devices(const char *hba_path
)
5577 struct md_list
*devlist
= NULL
;
5578 struct md_list
*dv
= NULL
;
5584 devlist
= get_loop_devices();
5587 /* scroll through /sys/dev/block looking for devices attached to
5590 dir
= opendir("/sys/dev/block");
5591 for (ent
= dir
? readdir(dir
) : NULL
; ent
; ent
= readdir(dir
)) {
5596 if (sscanf(ent
->d_name
, "%d:%d", &major
, &minor
) != 2)
5598 path
= devt_to_devpath(makedev(major
, minor
));
5601 if (!path_attached_to_hba(path
, hba_path
)) {
5608 fd
= dev_open(ent
->d_name
, O_RDONLY
);
5610 fd2devname(fd
, buf
);
5613 fprintf(stderr
, Name
": cannot open device: %s\n",
5619 dv
= calloc(1, sizeof(*dv
));
5621 fprintf(stderr
, Name
": malloc failed\n");
5625 dv
->devname
= strdup(buf
);
5626 if (dv
->devname
== NULL
) {
5627 fprintf(stderr
, Name
": malloc failed\n");
5638 devlist
= devlist
->next
;
5647 count_volumes_list(struct md_list
*devlist
, char *homehost
,
5648 int verbose
, int *found
)
5650 struct md_list
*tmpdev
;
5652 struct supertype
*st
= NULL
;
5654 /* first walk the list of devices to find a consistent set
5655 * that match the criterea, if that is possible.
5656 * We flag the ones we like with 'used'.
5659 st
= match_metadata_desc_imsm("imsm");
5661 pr_vrb(": cannot allocate memory for imsm supertype\n");
5665 for (tmpdev
= devlist
; tmpdev
; tmpdev
= tmpdev
->next
) {
5666 char *devname
= tmpdev
->devname
;
5668 struct supertype
*tst
;
5670 if (tmpdev
->used
> 1)
5672 tst
= dup_super(st
);
5674 pr_vrb(": cannot allocate memory for imsm supertype\n");
5677 tmpdev
->container
= 0;
5678 dfd
= dev_open(devname
, O_RDONLY
|O_EXCL
);
5680 dprintf(": cannot open device %s: %s\n",
5681 devname
, strerror(errno
));
5683 } else if (fstat(dfd
, &stb
)< 0) {
5685 dprintf(": fstat failed for %s: %s\n",
5686 devname
, strerror(errno
));
5688 } else if ((stb
.st_mode
& S_IFMT
) != S_IFBLK
) {
5689 dprintf(": %s is not a block device.\n",
5692 } else if (must_be_container(dfd
)) {
5693 struct supertype
*cst
;
5694 cst
= super_by_fd(dfd
, NULL
);
5696 dprintf(": cannot recognize container type %s\n",
5699 } else if (tst
->ss
!= st
->ss
) {
5700 dprintf(": non-imsm container - ignore it: %s\n",
5703 } else if (!tst
->ss
->load_container
||
5704 tst
->ss
->load_container(tst
, dfd
, NULL
))
5707 tmpdev
->container
= 1;
5710 cst
->ss
->free_super(cst
);
5712 tmpdev
->st_rdev
= stb
.st_rdev
;
5713 if (tst
->ss
->load_super(tst
,dfd
, NULL
)) {
5714 dprintf(": no RAID superblock on %s\n",
5717 } else if (tst
->ss
->compare_super
== NULL
) {
5718 dprintf(": Cannot assemble %s metadata on %s\n",
5719 tst
->ss
->name
, devname
);
5725 if (tmpdev
->used
== 2 || tmpdev
->used
== 4) {
5726 /* Ignore unrecognised devices during auto-assembly */
5731 tst
->ss
->getinfo_super(tst
, &info
, NULL
);
5733 if (st
->minor_version
== -1)
5734 st
->minor_version
= tst
->minor_version
;
5736 if (memcmp(info
.uuid
, uuid_zero
,
5737 sizeof(int[4])) == 0) {
5738 /* this is a floating spare. It cannot define
5739 * an array unless there are no more arrays of
5740 * this type to be found. It can be included
5741 * in an array of this type though.
5747 if (st
->ss
!= tst
->ss
||
5748 st
->minor_version
!= tst
->minor_version
||
5749 st
->ss
->compare_super(st
, tst
) != 0) {
5750 /* Some mismatch. If exactly one array matches this host,
5751 * we can resolve on that one.
5752 * Or, if we are auto assembling, we just ignore the second
5755 dprintf(": superblock on %s doesn't match others - assembly aborted\n",
5761 dprintf("found: devname: %s\n", devname
);
5765 tst
->ss
->free_super(tst
);
5769 if ((err
= load_super_imsm_all(st
, -1, &st
->sb
, NULL
, devlist
, 0)) == 0) {
5770 struct mdinfo
*iter
, *head
= st
->ss
->container_content(st
, NULL
);
5771 for (iter
= head
; iter
; iter
= iter
->next
) {
5772 dprintf("content->text_version: %s vol\n",
5773 iter
->text_version
);
5774 if (iter
->array
.state
& (1<<MD_SB_BLOCK_VOLUME
)) {
5775 /* do not assemble arrays with unsupported
5777 dprintf(": Cannot activate member %s.\n",
5778 iter
->text_version
);
5785 dprintf(" no valid super block on device list: err: %d %p\n",
5789 dprintf(" no more devices to examin\n");
5792 for (tmpdev
= devlist
; tmpdev
; tmpdev
= tmpdev
->next
) {
5793 if ((tmpdev
->used
== 1) && (tmpdev
->found
)) {
5795 if (count
< tmpdev
->found
)
5798 count
-= tmpdev
->found
;
5801 if (tmpdev
->used
== 1)
5806 st
->ss
->free_super(st
);
5812 count_volumes(char *hba
, int dpa
, int verbose
)
5814 struct md_list
*devlist
= NULL
;
5818 devlist
= get_devices(hba
);
5819 /* if no intel devices return zero volumes */
5820 if (devlist
== NULL
)
5823 count
= active_arrays_by_format("imsm", hba
, &devlist
, dpa
, verbose
);
5824 dprintf(" path: %s active arrays: %d\n", hba
, count
);
5825 if (devlist
== NULL
)
5829 count
+= count_volumes_list(devlist
,
5833 dprintf("found %d count: %d\n", found
, count
);
5836 dprintf("path: %s total number of volumes: %d\n", hba
, count
);
5839 struct md_list
*dv
= devlist
;
5840 devlist
= devlist
->next
;
5847 static int imsm_default_chunk(const struct imsm_orom
*orom
)
5849 /* up to 512 if the plaform supports it, otherwise the platform max.
5850 * 128 if no platform detected
5852 int fs
= max(7, orom
? fls(orom
->sss
) : 0);
5854 return min(512, (1 << fs
));
5858 validate_geometry_imsm_orom(struct intel_super
*super
, int level
, int layout
,
5859 int raiddisks
, int *chunk
, int verbose
)
5861 /* check/set platform and metadata limits/defaults */
5862 if (super
->orom
&& raiddisks
> super
->orom
->dpa
) {
5863 pr_vrb(": platform supports a maximum of %d disks per array\n",
5868 /* capabilities of OROM tested - copied from validate_geometry_imsm_volume */
5869 if (!is_raid_level_supported(super
->orom
, level
, raiddisks
)) {
5870 pr_vrb(": platform does not support raid%d with %d disk%s\n",
5871 level
, raiddisks
, raiddisks
> 1 ? "s" : "");
5875 if (chunk
&& (*chunk
== 0 || *chunk
== UnSet
))
5876 *chunk
= imsm_default_chunk(super
->orom
);
5878 if (super
->orom
&& chunk
&& !imsm_orom_has_chunk(super
->orom
, *chunk
)) {
5879 pr_vrb(": platform does not support a chunk size of: "
5884 if (layout
!= imsm_level_to_layout(level
)) {
5886 pr_vrb(": imsm raid 5 only supports the left-asymmetric layout\n");
5887 else if (level
== 10)
5888 pr_vrb(": imsm raid 10 only supports the n2 layout\n");
5890 pr_vrb(": imsm unknown layout %#x for this raid level %d\n",
5897 /* validate_geometry_imsm_volume - lifted from validate_geometry_ddf_bvd
5898 * FIX ME add ahci details
5900 static int validate_geometry_imsm_volume(struct supertype
*st
, int level
,
5901 int layout
, int raiddisks
, int *chunk
,
5902 unsigned long long size
, char *dev
,
5903 unsigned long long *freesize
,
5907 struct intel_super
*super
= st
->sb
;
5908 struct imsm_super
*mpb
;
5910 unsigned long long pos
= 0;
5911 unsigned long long maxsize
;
5915 /* We must have the container info already read in. */
5919 mpb
= super
->anchor
;
5921 if (!validate_geometry_imsm_orom(super
, level
, layout
, raiddisks
, chunk
, verbose
)) {
5922 fprintf(stderr
, Name
": RAID gemetry validation failed. "
5923 "Cannot proceed with the action(s).\n");
5927 /* General test: make sure there is space for
5928 * 'raiddisks' device extents of size 'size' at a given
5931 unsigned long long minsize
= size
;
5932 unsigned long long start_offset
= MaxSector
;
5935 minsize
= MPB_SECTOR_CNT
+ IMSM_RESERVED_SECTORS
;
5936 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
5941 e
= get_extents(super
, dl
);
5944 unsigned long long esize
;
5945 esize
= e
[i
].start
- pos
;
5946 if (esize
>= minsize
)
5948 if (found
&& start_offset
== MaxSector
) {
5951 } else if (found
&& pos
!= start_offset
) {
5955 pos
= e
[i
].start
+ e
[i
].size
;
5957 } while (e
[i
-1].size
);
5962 if (dcnt
< raiddisks
) {
5964 fprintf(stderr
, Name
": imsm: Not enough "
5965 "devices with space for this array "
5973 /* This device must be a member of the set */
5974 if (stat(dev
, &stb
) < 0)
5976 if ((S_IFMT
& stb
.st_mode
) != S_IFBLK
)
5978 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
5979 if (dl
->major
== (int)major(stb
.st_rdev
) &&
5980 dl
->minor
== (int)minor(stb
.st_rdev
))
5985 fprintf(stderr
, Name
": %s is not in the "
5986 "same imsm set\n", dev
);
5988 } else if (super
->orom
&& dl
->index
< 0 && mpb
->num_raid_devs
) {
5989 /* If a volume is present then the current creation attempt
5990 * cannot incorporate new spares because the orom may not
5991 * understand this configuration (all member disks must be
5992 * members of each array in the container).
5994 fprintf(stderr
, Name
": %s is a spare and a volume"
5995 " is already defined for this container\n", dev
);
5996 fprintf(stderr
, Name
": The option-rom requires all member"
5997 " disks to be a member of all volumes\n");
5999 } else if (super
->orom
&& mpb
->num_raid_devs
> 0 &&
6000 mpb
->num_disks
!= raiddisks
) {
6001 fprintf(stderr
, Name
": The option-rom requires all member"
6002 " disks to be a member of all volumes\n");
6006 /* retrieve the largest free space block */
6007 e
= get_extents(super
, dl
);
6012 unsigned long long esize
;
6014 esize
= e
[i
].start
- pos
;
6015 if (esize
>= maxsize
)
6017 pos
= e
[i
].start
+ e
[i
].size
;
6019 } while (e
[i
-1].size
);
6024 fprintf(stderr
, Name
": unable to determine free space for: %s\n",
6028 if (maxsize
< size
) {
6030 fprintf(stderr
, Name
": %s not enough space (%llu < %llu)\n",
6031 dev
, maxsize
, size
);
6035 /* count total number of extents for merge */
6037 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
6039 i
+= dl
->extent_cnt
;
6041 maxsize
= merge_extents(super
, i
);
6043 if (!check_env("IMSM_NO_PLATFORM") &&
6044 mpb
->num_raid_devs
> 0 && size
&& size
!= maxsize
) {
6045 fprintf(stderr
, Name
": attempting to create a second "
6046 "volume with size less then remaining space. "
6051 if (maxsize
< size
|| maxsize
== 0) {
6054 fprintf(stderr
, Name
": no free space"
6055 " left on device. Aborting...\n");
6057 fprintf(stderr
, Name
": not enough space"
6058 " to create volume of given size"
6059 " (%llu < %llu). Aborting...\n",
6065 *freesize
= maxsize
;
6068 int count
= count_volumes(super
->hba
->path
,
6069 super
->orom
->dpa
, verbose
);
6070 if (super
->orom
->vphba
<= count
) {
6071 pr_vrb(": platform does not support more then %d raid volumes.\n",
6072 super
->orom
->vphba
);
6079 static int reserve_space(struct supertype
*st
, int raiddisks
,
6080 unsigned long long size
, int chunk
,
6081 unsigned long long *freesize
)
6083 struct intel_super
*super
= st
->sb
;
6084 struct imsm_super
*mpb
= super
->anchor
;
6089 unsigned long long maxsize
;
6090 unsigned long long minsize
;
6094 /* find the largest common start free region of the possible disks */
6098 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
6104 /* don't activate new spares if we are orom constrained
6105 * and there is already a volume active in the container
6107 if (super
->orom
&& dl
->index
< 0 && mpb
->num_raid_devs
)
6110 e
= get_extents(super
, dl
);
6113 for (i
= 1; e
[i
-1].size
; i
++)
6121 maxsize
= merge_extents(super
, extent_cnt
);
6125 minsize
= chunk
* 2;
6127 if (cnt
< raiddisks
||
6128 (super
->orom
&& used
&& used
!= raiddisks
) ||
6129 maxsize
< minsize
||
6131 fprintf(stderr
, Name
": not enough devices with space to create array.\n");
6132 return 0; /* No enough free spaces large enough */
6143 if (!check_env("IMSM_NO_PLATFORM") &&
6144 mpb
->num_raid_devs
> 0 && size
&& size
!= maxsize
) {
6145 fprintf(stderr
, Name
": attempting to create a second "
6146 "volume with size less then remaining space. "
6151 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
6153 dl
->raiddisk
= cnt
++;
6160 static int validate_geometry_imsm(struct supertype
*st
, int level
, int layout
,
6161 int raiddisks
, int *chunk
, unsigned long long size
,
6162 char *dev
, unsigned long long *freesize
,
6170 * if given unused devices create a container
6171 * if given given devices in a container create a member volume
6173 if (level
== LEVEL_CONTAINER
) {
6174 /* Must be a fresh device to add to a container */
6175 return validate_geometry_imsm_container(st
, level
, layout
,
6177 chunk
?*chunk
:0, size
,
6184 struct intel_super
*super
= st
->sb
;
6185 if (!validate_geometry_imsm_orom(st
->sb
, level
, layout
,
6189 /* we are being asked to automatically layout a
6190 * new volume based on the current contents of
6191 * the container. If the the parameters can be
6192 * satisfied reserve_space will record the disks,
6193 * start offset, and size of the volume to be
6194 * created. add_to_super and getinfo_super
6195 * detect when autolayout is in progress.
6197 /* assuming that freesize is always given when array is
6199 if (super
->orom
&& freesize
) {
6201 count
= count_volumes(super
->hba
->path
,
6202 super
->orom
->dpa
, verbose
);
6203 if (super
->orom
->vphba
<= count
) {
6204 pr_vrb(": platform does not support more"
6205 "then %d raid volumes.\n",
6206 super
->orom
->vphba
);
6211 return reserve_space(st
, raiddisks
, size
,
6212 chunk
?*chunk
:0, freesize
);
6217 /* creating in a given container */
6218 return validate_geometry_imsm_volume(st
, level
, layout
,
6219 raiddisks
, chunk
, size
,
6220 dev
, freesize
, verbose
);
6223 /* This device needs to be a device in an 'imsm' container */
6224 fd
= open(dev
, O_RDONLY
|O_EXCL
, 0);
6228 Name
": Cannot create this array on device %s\n",
6233 if (errno
!= EBUSY
|| (fd
= open(dev
, O_RDONLY
, 0)) < 0) {
6235 fprintf(stderr
, Name
": Cannot open %s: %s\n",
6236 dev
, strerror(errno
));
6239 /* Well, it is in use by someone, maybe an 'imsm' container. */
6240 cfd
= open_container(fd
);
6244 fprintf(stderr
, Name
": Cannot use %s: It is busy\n",
6248 sra
= sysfs_read(cfd
, 0, GET_VERSION
);
6249 if (sra
&& sra
->array
.major_version
== -1 &&
6250 strcmp(sra
->text_version
, "imsm") == 0)
6254 /* This is a member of a imsm container. Load the container
6255 * and try to create a volume
6257 struct intel_super
*super
;
6259 if (load_super_imsm_all(st
, cfd
, (void **) &super
, NULL
, NULL
, 1) == 0) {
6261 st
->container_dev
= fd2devnum(cfd
);
6263 return validate_geometry_imsm_volume(st
, level
, layout
,
6272 fprintf(stderr
, Name
": failed container membership check\n");
6278 static void default_geometry_imsm(struct supertype
*st
, int *level
, int *layout
, int *chunk
)
6280 struct intel_super
*super
= st
->sb
;
6282 if (level
&& *level
== UnSet
)
6283 *level
= LEVEL_CONTAINER
;
6285 if (level
&& layout
&& *layout
== UnSet
)
6286 *layout
= imsm_level_to_layout(*level
);
6288 if (chunk
&& (*chunk
== UnSet
|| *chunk
== 0))
6289 *chunk
= imsm_default_chunk(super
->orom
);
6292 static void handle_missing(struct intel_super
*super
, struct imsm_dev
*dev
);
6294 static int kill_subarray_imsm(struct supertype
*st
)
6296 /* remove the subarray currently referenced by ->current_vol */
6298 struct intel_dev
**dp
;
6299 struct intel_super
*super
= st
->sb
;
6300 __u8 current_vol
= super
->current_vol
;
6301 struct imsm_super
*mpb
= super
->anchor
;
6303 if (super
->current_vol
< 0)
6305 super
->current_vol
= -1; /* invalidate subarray cursor */
6307 /* block deletions that would change the uuid of active subarrays
6309 * FIXME when immutable ids are available, but note that we'll
6310 * also need to fixup the invalidated/active subarray indexes in
6313 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
6316 if (i
< current_vol
)
6318 sprintf(subarray
, "%u", i
);
6319 if (is_subarray_active(subarray
, st
->devname
)) {
6321 Name
": deleting subarray-%d would change the UUID of active subarray-%d, aborting\n",
6328 if (st
->update_tail
) {
6329 struct imsm_update_kill_array
*u
= malloc(sizeof(*u
));
6333 u
->type
= update_kill_array
;
6334 u
->dev_idx
= current_vol
;
6335 append_metadata_update(st
, u
, sizeof(*u
));
6340 for (dp
= &super
->devlist
; *dp
;)
6341 if ((*dp
)->index
== current_vol
) {
6344 handle_missing(super
, (*dp
)->dev
);
6345 if ((*dp
)->index
> current_vol
)
6350 /* no more raid devices, all active components are now spares,
6351 * but of course failed are still failed
6353 if (--mpb
->num_raid_devs
== 0) {
6356 for (d
= super
->disks
; d
; d
= d
->next
)
6361 super
->updates_pending
++;
6366 static int update_subarray_imsm(struct supertype
*st
, char *subarray
,
6367 char *update
, struct mddev_ident
*ident
)
6369 /* update the subarray currently referenced by ->current_vol */
6370 struct intel_super
*super
= st
->sb
;
6371 struct imsm_super
*mpb
= super
->anchor
;
6373 if (strcmp(update
, "name") == 0) {
6374 char *name
= ident
->name
;
6378 if (is_subarray_active(subarray
, st
->devname
)) {
6380 Name
": Unable to update name of active subarray\n");
6384 if (!check_name(super
, name
, 0))
6387 vol
= strtoul(subarray
, &ep
, 10);
6388 if (*ep
!= '\0' || vol
>= super
->anchor
->num_raid_devs
)
6391 if (st
->update_tail
) {
6392 struct imsm_update_rename_array
*u
= malloc(sizeof(*u
));
6396 u
->type
= update_rename_array
;
6398 snprintf((char *) u
->name
, MAX_RAID_SERIAL_LEN
, "%s", name
);
6399 append_metadata_update(st
, u
, sizeof(*u
));
6401 struct imsm_dev
*dev
;
6404 dev
= get_imsm_dev(super
, vol
);
6405 snprintf((char *) dev
->volume
, MAX_RAID_SERIAL_LEN
, "%s", name
);
6406 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
6407 dev
= get_imsm_dev(super
, i
);
6408 handle_missing(super
, dev
);
6410 super
->updates_pending
++;
6417 #endif /* MDASSEMBLE */
6419 static int is_gen_migration(struct imsm_dev
*dev
)
6424 if (!dev
->vol
.migr_state
)
6427 if (migr_type(dev
) == MIGR_GEN_MIGR
)
6433 static int is_rebuilding(struct imsm_dev
*dev
)
6435 struct imsm_map
*migr_map
;
6437 if (!dev
->vol
.migr_state
)
6440 if (migr_type(dev
) != MIGR_REBUILD
)
6443 migr_map
= get_imsm_map(dev
, MAP_1
);
6445 if (migr_map
->map_state
== IMSM_T_STATE_DEGRADED
)
6452 static int is_initializing(struct imsm_dev
*dev
)
6454 struct imsm_map
*migr_map
;
6456 if (!dev
->vol
.migr_state
)
6459 if (migr_type(dev
) != MIGR_INIT
)
6462 migr_map
= get_imsm_map(dev
, MAP_1
);
6464 if (migr_map
->map_state
== IMSM_T_STATE_UNINITIALIZED
)
6471 static void update_recovery_start(struct intel_super
*super
,
6472 struct imsm_dev
*dev
,
6473 struct mdinfo
*array
)
6475 struct mdinfo
*rebuild
= NULL
;
6479 if (!is_rebuilding(dev
))
6482 /* Find the rebuild target, but punt on the dual rebuild case */
6483 for (d
= array
->devs
; d
; d
= d
->next
)
6484 if (d
->recovery_start
== 0) {
6491 /* (?) none of the disks are marked with
6492 * IMSM_ORD_REBUILD, so assume they are missing and the
6493 * disk_ord_tbl was not correctly updated
6495 dprintf("%s: failed to locate out-of-sync disk\n", __func__
);
6499 units
= __le32_to_cpu(dev
->vol
.curr_migr_unit
);
6500 rebuild
->recovery_start
= units
* blocks_per_migr_unit(super
, dev
);
6504 static int recover_backup_imsm(struct supertype
*st
, struct mdinfo
*info
);
6507 static struct mdinfo
*container_content_imsm(struct supertype
*st
, char *subarray
)
6509 /* Given a container loaded by load_super_imsm_all,
6510 * extract information about all the arrays into
6512 * If 'subarray' is given, just extract info about that array.
6514 * For each imsm_dev create an mdinfo, fill it in,
6515 * then look for matching devices in super->disks
6516 * and create appropriate device mdinfo.
6518 struct intel_super
*super
= st
->sb
;
6519 struct imsm_super
*mpb
= super
->anchor
;
6520 struct mdinfo
*rest
= NULL
;
6524 int spare_disks
= 0;
6526 /* do not assemble arrays when not all attributes are supported */
6527 if (imsm_check_attributes(mpb
->attributes
) == 0) {
6529 fprintf(stderr
, Name
": Unsupported attributes in IMSM metadata."
6530 "Arrays activation is blocked.\n");
6533 /* check for bad blocks */
6534 if (imsm_bbm_log_size(super
->anchor
)) {
6535 fprintf(stderr
, Name
": BBM log found in IMSM metadata."
6536 "Arrays activation is blocked.\n");
6541 /* count spare devices, not used in maps
6543 for (d
= super
->disks
; d
; d
= d
->next
)
6547 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
6548 struct imsm_dev
*dev
;
6549 struct imsm_map
*map
;
6550 struct imsm_map
*map2
;
6551 struct mdinfo
*this;
6559 (i
!= strtoul(subarray
, &ep
, 10) || *ep
!= '\0'))
6562 dev
= get_imsm_dev(super
, i
);
6563 map
= get_imsm_map(dev
, MAP_0
);
6564 map2
= get_imsm_map(dev
, MAP_1
);
6566 /* do not publish arrays that are in the middle of an
6567 * unsupported migration
6569 if (dev
->vol
.migr_state
&&
6570 (migr_type(dev
) == MIGR_STATE_CHANGE
)) {
6571 fprintf(stderr
, Name
": cannot assemble volume '%.16s':"
6572 " unsupported migration in progress\n",
6576 /* do not publish arrays that are not support by controller's
6580 this = malloc(sizeof(*this));
6582 fprintf(stderr
, Name
": failed to allocate %zu bytes\n",
6587 super
->current_vol
= i
;
6588 getinfo_super_imsm_volume(st
, this, NULL
);
6591 chunk
= __le16_to_cpu(map
->blocks_per_strip
) >> 1;
6592 /* mdadm does not support all metadata features- set the bit in all arrays state */
6593 if (!validate_geometry_imsm_orom(super
,
6594 get_imsm_raid_level(map
), /* RAID level */
6595 imsm_level_to_layout(get_imsm_raid_level(map
)),
6596 map
->num_members
, /* raid disks */
6599 fprintf(stderr
, Name
": IMSM RAID geometry validation"
6600 " failed. Array %s activation is blocked.\n",
6602 this->array
.state
|=
6603 (1<<MD_SB_BLOCK_CONTAINER_RESHAPE
) |
6604 (1<<MD_SB_BLOCK_VOLUME
);
6608 /* if array has bad blocks, set suitable bit in all arrays state */
6610 this->array
.state
|=
6611 (1<<MD_SB_BLOCK_CONTAINER_RESHAPE
) |
6612 (1<<MD_SB_BLOCK_VOLUME
);
6614 for (slot
= 0 ; slot
< map
->num_members
; slot
++) {
6615 unsigned long long recovery_start
;
6616 struct mdinfo
*info_d
;
6623 idx
= get_imsm_disk_idx(dev
, slot
, MAP_0
);
6624 ord
= get_imsm_ord_tbl_ent(dev
, slot
, MAP_X
);
6625 for (d
= super
->disks
; d
; d
= d
->next
)
6626 if (d
->index
== idx
)
6629 recovery_start
= MaxSector
;
6632 if (d
&& is_failed(&d
->disk
))
6634 if (ord
& IMSM_ORD_REBUILD
)
6638 * if we skip some disks the array will be assmebled degraded;
6639 * reset resync start to avoid a dirty-degraded
6640 * situation when performing the intial sync
6642 * FIXME handle dirty degraded
6644 if ((skip
|| recovery_start
== 0) && !dev
->vol
.dirty
)
6645 this->resync_start
= MaxSector
;
6649 info_d
= calloc(1, sizeof(*info_d
));
6651 fprintf(stderr
, Name
": failed to allocate disk"
6652 " for volume %.16s\n", dev
->volume
);
6653 info_d
= this->devs
;
6655 struct mdinfo
*d
= info_d
->next
;
6664 info_d
->next
= this->devs
;
6665 this->devs
= info_d
;
6667 info_d
->disk
.number
= d
->index
;
6668 info_d
->disk
.major
= d
->major
;
6669 info_d
->disk
.minor
= d
->minor
;
6670 info_d
->disk
.raid_disk
= slot
;
6671 info_d
->recovery_start
= recovery_start
;
6673 if (slot
< map2
->num_members
)
6674 info_d
->disk
.state
= (1 << MD_DISK_ACTIVE
);
6676 this->array
.spare_disks
++;
6678 if (slot
< map
->num_members
)
6679 info_d
->disk
.state
= (1 << MD_DISK_ACTIVE
);
6681 this->array
.spare_disks
++;
6683 if (info_d
->recovery_start
== MaxSector
)
6684 this->array
.working_disks
++;
6686 info_d
->events
= __le32_to_cpu(mpb
->generation_num
);
6687 info_d
->data_offset
= pba_of_lba0(map
);
6688 info_d
->component_size
= blocks_per_member(map
);
6690 /* now that the disk list is up-to-date fixup recovery_start */
6691 update_recovery_start(super
, dev
, this);
6692 this->array
.spare_disks
+= spare_disks
;
6695 /* check for reshape */
6696 if (this->reshape_active
== 1)
6697 recover_backup_imsm(st
, this);
6706 static __u8
imsm_check_degraded(struct intel_super
*super
, struct imsm_dev
*dev
,
6707 int failed
, int look_in_map
)
6709 struct imsm_map
*map
;
6711 map
= get_imsm_map(dev
, look_in_map
);
6714 return map
->map_state
== IMSM_T_STATE_UNINITIALIZED
?
6715 IMSM_T_STATE_UNINITIALIZED
: IMSM_T_STATE_NORMAL
;
6717 switch (get_imsm_raid_level(map
)) {
6719 return IMSM_T_STATE_FAILED
;
6722 if (failed
< map
->num_members
)
6723 return IMSM_T_STATE_DEGRADED
;
6725 return IMSM_T_STATE_FAILED
;
6730 * check to see if any mirrors have failed, otherwise we
6731 * are degraded. Even numbered slots are mirrored on
6735 /* gcc -Os complains that this is unused */
6736 int insync
= insync
;
6738 for (i
= 0; i
< map
->num_members
; i
++) {
6739 __u32 ord
= get_imsm_ord_tbl_ent(dev
, i
, MAP_X
);
6740 int idx
= ord_to_idx(ord
);
6741 struct imsm_disk
*disk
;
6743 /* reset the potential in-sync count on even-numbered
6744 * slots. num_copies is always 2 for imsm raid10
6749 disk
= get_imsm_disk(super
, idx
);
6750 if (!disk
|| is_failed(disk
) || ord
& IMSM_ORD_REBUILD
)
6753 /* no in-sync disks left in this mirror the
6757 return IMSM_T_STATE_FAILED
;
6760 return IMSM_T_STATE_DEGRADED
;
6764 return IMSM_T_STATE_DEGRADED
;
6766 return IMSM_T_STATE_FAILED
;
6772 return map
->map_state
;
6775 static int imsm_count_failed(struct intel_super
*super
, struct imsm_dev
*dev
,
6780 struct imsm_disk
*disk
;
6781 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
6782 struct imsm_map
*prev
= get_imsm_map(dev
, MAP_1
);
6783 struct imsm_map
*map_for_loop
;
6788 /* at the beginning of migration we set IMSM_ORD_REBUILD on
6789 * disks that are being rebuilt. New failures are recorded to
6790 * map[0]. So we look through all the disks we started with and
6791 * see if any failures are still present, or if any new ones
6795 if (prev
&& (map
->num_members
< prev
->num_members
))
6796 map_for_loop
= prev
;
6798 for (i
= 0; i
< map_for_loop
->num_members
; i
++) {
6800 /* when MAP_X is passed both maps failures are counted
6803 ((look_in_map
== MAP_1
) || (look_in_map
== MAP_X
)) &&
6804 (i
< prev
->num_members
)) {
6805 ord
= __le32_to_cpu(prev
->disk_ord_tbl
[i
]);
6806 idx_1
= ord_to_idx(ord
);
6808 disk
= get_imsm_disk(super
, idx_1
);
6809 if (!disk
|| is_failed(disk
) || ord
& IMSM_ORD_REBUILD
)
6812 if (((look_in_map
== MAP_0
) || (look_in_map
== MAP_X
)) &&
6813 (i
< map
->num_members
)) {
6814 ord
= __le32_to_cpu(map
->disk_ord_tbl
[i
]);
6815 idx
= ord_to_idx(ord
);
6818 disk
= get_imsm_disk(super
, idx
);
6819 if (!disk
|| is_failed(disk
) ||
6820 ord
& IMSM_ORD_REBUILD
)
6830 static int imsm_open_new(struct supertype
*c
, struct active_array
*a
,
6833 struct intel_super
*super
= c
->sb
;
6834 struct imsm_super
*mpb
= super
->anchor
;
6836 if (atoi(inst
) >= mpb
->num_raid_devs
) {
6837 fprintf(stderr
, "%s: subarry index %d, out of range\n",
6838 __func__
, atoi(inst
));
6842 dprintf("imsm: open_new %s\n", inst
);
6843 a
->info
.container_member
= atoi(inst
);
6847 static int is_resyncing(struct imsm_dev
*dev
)
6849 struct imsm_map
*migr_map
;
6851 if (!dev
->vol
.migr_state
)
6854 if (migr_type(dev
) == MIGR_INIT
||
6855 migr_type(dev
) == MIGR_REPAIR
)
6858 if (migr_type(dev
) == MIGR_GEN_MIGR
)
6861 migr_map
= get_imsm_map(dev
, MAP_1
);
6863 if ((migr_map
->map_state
== IMSM_T_STATE_NORMAL
) &&
6864 (dev
->vol
.migr_type
!= MIGR_GEN_MIGR
))
6870 /* return true if we recorded new information */
6871 static int mark_failure(struct imsm_dev
*dev
, struct imsm_disk
*disk
, int idx
)
6875 struct imsm_map
*map
;
6876 char buf
[MAX_RAID_SERIAL_LEN
+3];
6877 unsigned int len
, shift
= 0;
6879 /* new failures are always set in map[0] */
6880 map
= get_imsm_map(dev
, MAP_0
);
6882 slot
= get_imsm_disk_slot(map
, idx
);
6886 ord
= __le32_to_cpu(map
->disk_ord_tbl
[slot
]);
6887 if (is_failed(disk
) && (ord
& IMSM_ORD_REBUILD
))
6890 memcpy(buf
, disk
->serial
, MAX_RAID_SERIAL_LEN
);
6891 buf
[MAX_RAID_SERIAL_LEN
] = '\000';
6893 if ((len
= strlen(buf
)) >= MAX_RAID_SERIAL_LEN
)
6894 shift
= len
- MAX_RAID_SERIAL_LEN
+ 1;
6895 strncpy((char *)disk
->serial
, &buf
[shift
], MAX_RAID_SERIAL_LEN
);
6897 disk
->status
|= FAILED_DISK
;
6898 set_imsm_ord_tbl_ent(map
, slot
, idx
| IMSM_ORD_REBUILD
);
6899 /* mark failures in second map if second map exists and this disk
6901 * This is valid for migration, initialization and rebuild
6903 if (dev
->vol
.migr_state
) {
6904 struct imsm_map
*map2
= get_imsm_map(dev
, MAP_1
);
6905 int slot2
= get_imsm_disk_slot(map2
, idx
);
6907 if ((slot2
< map2
->num_members
) &&
6909 set_imsm_ord_tbl_ent(map2
, slot2
,
6910 idx
| IMSM_ORD_REBUILD
);
6912 if (map
->failed_disk_num
== 0xff)
6913 map
->failed_disk_num
= slot
;
6917 static void mark_missing(struct imsm_dev
*dev
, struct imsm_disk
*disk
, int idx
)
6919 mark_failure(dev
, disk
, idx
);
6921 if (disk
->scsi_id
== __cpu_to_le32(~(__u32
)0))
6924 disk
->scsi_id
= __cpu_to_le32(~(__u32
)0);
6925 memmove(&disk
->serial
[0], &disk
->serial
[1], MAX_RAID_SERIAL_LEN
- 1);
6928 static void handle_missing(struct intel_super
*super
, struct imsm_dev
*dev
)
6932 if (!super
->missing
)
6935 dprintf("imsm: mark missing\n");
6936 /* end process for initialization and rebuild only
6938 if (is_gen_migration(dev
) == 0) {
6942 failed
= imsm_count_failed(super
, dev
, MAP_0
);
6943 map_state
= imsm_check_degraded(super
, dev
, failed
, MAP_0
);
6945 end_migration(dev
, super
, map_state
);
6947 for (dl
= super
->missing
; dl
; dl
= dl
->next
)
6948 mark_missing(dev
, &dl
->disk
, dl
->index
);
6949 super
->updates_pending
++;
6952 static unsigned long long imsm_set_array_size(struct imsm_dev
*dev
)
6954 int used_disks
= imsm_num_data_members(dev
, MAP_0
);
6955 unsigned long long array_blocks
;
6956 struct imsm_map
*map
;
6958 if (used_disks
== 0) {
6959 /* when problems occures
6960 * return current array_blocks value
6962 array_blocks
= __le32_to_cpu(dev
->size_high
);
6963 array_blocks
= array_blocks
<< 32;
6964 array_blocks
+= __le32_to_cpu(dev
->size_low
);
6966 return array_blocks
;
6969 /* set array size in metadata
6971 map
= get_imsm_map(dev
, MAP_0
);
6972 array_blocks
= blocks_per_member(map
) * used_disks
;
6974 /* round array size down to closest MB
6976 array_blocks
= (array_blocks
>> SECT_PER_MB_SHIFT
) << SECT_PER_MB_SHIFT
;
6977 dev
->size_low
= __cpu_to_le32((__u32
)array_blocks
);
6978 dev
->size_high
= __cpu_to_le32((__u32
)(array_blocks
>> 32));
6980 return array_blocks
;
6983 static void imsm_set_disk(struct active_array
*a
, int n
, int state
);
6985 static void imsm_progress_container_reshape(struct intel_super
*super
)
6987 /* if no device has a migr_state, but some device has a
6988 * different number of members than the previous device, start
6989 * changing the number of devices in this device to match
6992 struct imsm_super
*mpb
= super
->anchor
;
6993 int prev_disks
= -1;
6997 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
6998 struct imsm_dev
*dev
= get_imsm_dev(super
, i
);
6999 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
7000 struct imsm_map
*map2
;
7001 int prev_num_members
;
7003 if (dev
->vol
.migr_state
)
7006 if (prev_disks
== -1)
7007 prev_disks
= map
->num_members
;
7008 if (prev_disks
== map
->num_members
)
7011 /* OK, this array needs to enter reshape mode.
7012 * i.e it needs a migr_state
7015 copy_map_size
= sizeof_imsm_map(map
);
7016 prev_num_members
= map
->num_members
;
7017 map
->num_members
= prev_disks
;
7018 dev
->vol
.migr_state
= 1;
7019 dev
->vol
.curr_migr_unit
= 0;
7020 set_migr_type(dev
, MIGR_GEN_MIGR
);
7021 for (i
= prev_num_members
;
7022 i
< map
->num_members
; i
++)
7023 set_imsm_ord_tbl_ent(map
, i
, i
);
7024 map2
= get_imsm_map(dev
, MAP_1
);
7025 /* Copy the current map */
7026 memcpy(map2
, map
, copy_map_size
);
7027 map2
->num_members
= prev_num_members
;
7029 imsm_set_array_size(dev
);
7030 super
->clean_migration_record_by_mdmon
= 1;
7031 super
->updates_pending
++;
7035 /* Handle dirty -> clean transititions, resync and reshape. Degraded and rebuild
7036 * states are handled in imsm_set_disk() with one exception, when a
7037 * resync is stopped due to a new failure this routine will set the
7038 * 'degraded' state for the array.
7040 static int imsm_set_array_state(struct active_array
*a
, int consistent
)
7042 int inst
= a
->info
.container_member
;
7043 struct intel_super
*super
= a
->container
->sb
;
7044 struct imsm_dev
*dev
= get_imsm_dev(super
, inst
);
7045 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
7046 int failed
= imsm_count_failed(super
, dev
, MAP_0
);
7047 __u8 map_state
= imsm_check_degraded(super
, dev
, failed
, MAP_0
);
7048 __u32 blocks_per_unit
;
7050 if (dev
->vol
.migr_state
&&
7051 dev
->vol
.migr_type
== MIGR_GEN_MIGR
) {
7052 /* array state change is blocked due to reshape action
7054 * - abort the reshape (if last_checkpoint is 0 and action!= reshape)
7055 * - finish the reshape (if last_checkpoint is big and action != reshape)
7056 * - update curr_migr_unit
7058 if (a
->curr_action
== reshape
) {
7059 /* still reshaping, maybe update curr_migr_unit */
7060 goto mark_checkpoint
;
7062 if (a
->last_checkpoint
== 0 && a
->prev_action
== reshape
) {
7063 /* for some reason we aborted the reshape.
7065 * disable automatic metadata rollback
7066 * user action is required to recover process
7069 struct imsm_map
*map2
=
7070 get_imsm_map(dev
, MAP_1
);
7071 dev
->vol
.migr_state
= 0;
7072 set_migr_type(dev
, 0);
7073 dev
->vol
.curr_migr_unit
= 0;
7075 sizeof_imsm_map(map2
));
7076 super
->updates_pending
++;
7079 if (a
->last_checkpoint
>= a
->info
.component_size
) {
7080 unsigned long long array_blocks
;
7084 used_disks
= imsm_num_data_members(dev
, MAP_0
);
7085 if (used_disks
> 0) {
7087 blocks_per_member(map
) *
7089 /* round array size down to closest MB
7091 array_blocks
= (array_blocks
7092 >> SECT_PER_MB_SHIFT
)
7093 << SECT_PER_MB_SHIFT
;
7094 a
->info
.custom_array_size
= array_blocks
;
7095 /* encourage manager to update array
7099 a
->check_reshape
= 1;
7101 /* finalize online capacity expansion/reshape */
7102 for (mdi
= a
->info
.devs
; mdi
; mdi
= mdi
->next
)
7104 mdi
->disk
.raid_disk
,
7107 imsm_progress_container_reshape(super
);
7112 /* before we activate this array handle any missing disks */
7113 if (consistent
== 2)
7114 handle_missing(super
, dev
);
7116 if (consistent
== 2 &&
7117 (!is_resync_complete(&a
->info
) ||
7118 map_state
!= IMSM_T_STATE_NORMAL
||
7119 dev
->vol
.migr_state
))
7122 if (is_resync_complete(&a
->info
)) {
7123 /* complete intialization / resync,
7124 * recovery and interrupted recovery is completed in
7127 if (is_resyncing(dev
)) {
7128 dprintf("imsm: mark resync done\n");
7129 end_migration(dev
, super
, map_state
);
7130 super
->updates_pending
++;
7131 a
->last_checkpoint
= 0;
7133 } else if ((!is_resyncing(dev
) && !failed
) &&
7134 (imsm_reshape_blocks_arrays_changes(super
) == 0)) {
7135 /* mark the start of the init process if nothing is failed */
7136 dprintf("imsm: mark resync start\n");
7137 if (map
->map_state
== IMSM_T_STATE_UNINITIALIZED
)
7138 migrate(dev
, super
, IMSM_T_STATE_NORMAL
, MIGR_INIT
);
7140 migrate(dev
, super
, IMSM_T_STATE_NORMAL
, MIGR_REPAIR
);
7141 super
->updates_pending
++;
7145 /* skip checkpointing for general migration,
7146 * it is controlled in mdadm
7148 if (is_gen_migration(dev
))
7149 goto skip_mark_checkpoint
;
7151 /* check if we can update curr_migr_unit from resync_start, recovery_start */
7152 blocks_per_unit
= blocks_per_migr_unit(super
, dev
);
7153 if (blocks_per_unit
) {
7157 units
= a
->last_checkpoint
/ blocks_per_unit
;
7160 /* check that we did not overflow 32-bits, and that
7161 * curr_migr_unit needs updating
7163 if (units32
== units
&&
7165 __le32_to_cpu(dev
->vol
.curr_migr_unit
) != units32
) {
7166 dprintf("imsm: mark checkpoint (%u)\n", units32
);
7167 dev
->vol
.curr_migr_unit
= __cpu_to_le32(units32
);
7168 super
->updates_pending
++;
7172 skip_mark_checkpoint
:
7173 /* mark dirty / clean */
7174 if (dev
->vol
.dirty
!= !consistent
) {
7175 dprintf("imsm: mark '%s'\n", consistent
? "clean" : "dirty");
7180 super
->updates_pending
++;
7186 static void imsm_set_disk(struct active_array
*a
, int n
, int state
)
7188 int inst
= a
->info
.container_member
;
7189 struct intel_super
*super
= a
->container
->sb
;
7190 struct imsm_dev
*dev
= get_imsm_dev(super
, inst
);
7191 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
7192 struct imsm_disk
*disk
;
7197 if (n
> map
->num_members
)
7198 fprintf(stderr
, "imsm: set_disk %d out of range 0..%d\n",
7199 n
, map
->num_members
- 1);
7204 dprintf("imsm: set_disk %d:%x\n", n
, state
);
7206 ord
= get_imsm_ord_tbl_ent(dev
, n
, MAP_0
);
7207 disk
= get_imsm_disk(super
, ord_to_idx(ord
));
7209 /* check for new failures */
7210 if (state
& DS_FAULTY
) {
7211 if (mark_failure(dev
, disk
, ord_to_idx(ord
)))
7212 super
->updates_pending
++;
7215 /* check if in_sync */
7216 if (state
& DS_INSYNC
&& ord
& IMSM_ORD_REBUILD
&& is_rebuilding(dev
)) {
7217 struct imsm_map
*migr_map
= get_imsm_map(dev
, MAP_1
);
7219 set_imsm_ord_tbl_ent(migr_map
, n
, ord_to_idx(ord
));
7220 super
->updates_pending
++;
7223 failed
= imsm_count_failed(super
, dev
, MAP_0
);
7224 map_state
= imsm_check_degraded(super
, dev
, failed
, MAP_0
);
7226 /* check if recovery complete, newly degraded, or failed */
7227 dprintf("imsm: Detected transition to state ");
7228 switch (map_state
) {
7229 case IMSM_T_STATE_NORMAL
: /* transition to normal state */
7230 dprintf("normal: ");
7231 if (is_rebuilding(dev
)) {
7232 dprintf("while rebuilding");
7233 end_migration(dev
, super
, map_state
);
7234 map
= get_imsm_map(dev
, MAP_0
);
7235 map
->failed_disk_num
= ~0;
7236 super
->updates_pending
++;
7237 a
->last_checkpoint
= 0;
7240 if (is_gen_migration(dev
)) {
7241 dprintf("while general migration");
7242 if (a
->last_checkpoint
>= a
->info
.component_size
)
7243 end_migration(dev
, super
, map_state
);
7245 map
->map_state
= map_state
;
7246 map
= get_imsm_map(dev
, MAP_0
);
7247 map
->failed_disk_num
= ~0;
7248 super
->updates_pending
++;
7252 case IMSM_T_STATE_DEGRADED
: /* transition to degraded state */
7253 dprintf("degraded: ");
7254 if ((map
->map_state
!= map_state
) &&
7255 !dev
->vol
.migr_state
) {
7256 dprintf("mark degraded");
7257 map
->map_state
= map_state
;
7258 super
->updates_pending
++;
7259 a
->last_checkpoint
= 0;
7262 if (is_rebuilding(dev
)) {
7263 dprintf("while rebuilding.");
7264 if (map
->map_state
!= map_state
) {
7265 dprintf(" Map state change");
7266 end_migration(dev
, super
, map_state
);
7267 super
->updates_pending
++;
7271 if (is_gen_migration(dev
)) {
7272 dprintf("while general migration");
7273 if (a
->last_checkpoint
>= a
->info
.component_size
)
7274 end_migration(dev
, super
, map_state
);
7276 map
->map_state
= map_state
;
7277 manage_second_map(super
, dev
);
7279 super
->updates_pending
++;
7282 if (is_initializing(dev
)) {
7283 dprintf("while initialization.");
7284 map
->map_state
= map_state
;
7285 super
->updates_pending
++;
7289 case IMSM_T_STATE_FAILED
: /* transition to failed state */
7290 dprintf("failed: ");
7291 if (is_gen_migration(dev
)) {
7292 dprintf("while general migration");
7293 map
->map_state
= map_state
;
7294 super
->updates_pending
++;
7297 if (map
->map_state
!= map_state
) {
7298 dprintf("mark failed");
7299 end_migration(dev
, super
, map_state
);
7300 super
->updates_pending
++;
7301 a
->last_checkpoint
= 0;
7306 dprintf("state %i\n", map_state
);
7312 static int store_imsm_mpb(int fd
, struct imsm_super
*mpb
)
7315 __u32 mpb_size
= __le32_to_cpu(mpb
->mpb_size
);
7316 unsigned long long dsize
;
7317 unsigned long long sectors
;
7319 get_dev_size(fd
, NULL
, &dsize
);
7321 if (mpb_size
> 512) {
7322 /* -1 to account for anchor */
7323 sectors
= mpb_sectors(mpb
) - 1;
7325 /* write the extended mpb to the sectors preceeding the anchor */
7326 if (lseek64(fd
, dsize
- (512 * (2 + sectors
)), SEEK_SET
) < 0)
7329 if ((unsigned long long)write(fd
, buf
+ 512, 512 * sectors
)
7334 /* first block is stored on second to last sector of the disk */
7335 if (lseek64(fd
, dsize
- (512 * 2), SEEK_SET
) < 0)
7338 if (write(fd
, buf
, 512) != 512)
7344 static void imsm_sync_metadata(struct supertype
*container
)
7346 struct intel_super
*super
= container
->sb
;
7348 dprintf("sync metadata: %d\n", super
->updates_pending
);
7349 if (!super
->updates_pending
)
7352 write_super_imsm(container
, 0);
7354 super
->updates_pending
= 0;
7357 static struct dl
*imsm_readd(struct intel_super
*super
, int idx
, struct active_array
*a
)
7359 struct imsm_dev
*dev
= get_imsm_dev(super
, a
->info
.container_member
);
7360 int i
= get_imsm_disk_idx(dev
, idx
, MAP_X
);
7363 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
7367 if (dl
&& is_failed(&dl
->disk
))
7371 dprintf("%s: found %x:%x\n", __func__
, dl
->major
, dl
->minor
);
7376 static struct dl
*imsm_add_spare(struct intel_super
*super
, int slot
,
7377 struct active_array
*a
, int activate_new
,
7378 struct mdinfo
*additional_test_list
)
7380 struct imsm_dev
*dev
= get_imsm_dev(super
, a
->info
.container_member
);
7381 int idx
= get_imsm_disk_idx(dev
, slot
, MAP_X
);
7382 struct imsm_super
*mpb
= super
->anchor
;
7383 struct imsm_map
*map
;
7384 unsigned long long pos
;
7389 __u32 array_start
= 0;
7390 __u32 array_end
= 0;
7392 struct mdinfo
*test_list
;
7394 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
7395 /* If in this array, skip */
7396 for (d
= a
->info
.devs
; d
; d
= d
->next
)
7397 if (d
->state_fd
>= 0 &&
7398 d
->disk
.major
== dl
->major
&&
7399 d
->disk
.minor
== dl
->minor
) {
7400 dprintf("%x:%x already in array\n",
7401 dl
->major
, dl
->minor
);
7406 test_list
= additional_test_list
;
7408 if (test_list
->disk
.major
== dl
->major
&&
7409 test_list
->disk
.minor
== dl
->minor
) {
7410 dprintf("%x:%x already in additional test list\n",
7411 dl
->major
, dl
->minor
);
7414 test_list
= test_list
->next
;
7419 /* skip in use or failed drives */
7420 if (is_failed(&dl
->disk
) || idx
== dl
->index
||
7422 dprintf("%x:%x status (failed: %d index: %d)\n",
7423 dl
->major
, dl
->minor
, is_failed(&dl
->disk
), idx
);
7427 /* skip pure spares when we are looking for partially
7428 * assimilated drives
7430 if (dl
->index
== -1 && !activate_new
)
7433 /* Does this unused device have the requisite free space?
7434 * It needs to be able to cover all member volumes
7436 ex
= get_extents(super
, dl
);
7438 dprintf("cannot get extents\n");
7441 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
7442 dev
= get_imsm_dev(super
, i
);
7443 map
= get_imsm_map(dev
, MAP_0
);
7445 /* check if this disk is already a member of
7448 if (get_imsm_disk_slot(map
, dl
->index
) >= 0)
7454 array_start
= pba_of_lba0(map
);
7455 array_end
= array_start
+
7456 blocks_per_member(map
) - 1;
7459 /* check that we can start at pba_of_lba0 with
7460 * blocks_per_member of space
7462 if (array_start
>= pos
&& array_end
< ex
[j
].start
) {
7466 pos
= ex
[j
].start
+ ex
[j
].size
;
7468 } while (ex
[j
-1].size
);
7475 if (i
< mpb
->num_raid_devs
) {
7476 dprintf("%x:%x does not have %u to %u available\n",
7477 dl
->major
, dl
->minor
, array_start
, array_end
);
7488 static int imsm_rebuild_allowed(struct supertype
*cont
, int dev_idx
, int failed
)
7490 struct imsm_dev
*dev2
;
7491 struct imsm_map
*map
;
7497 dev2
= get_imsm_dev(cont
->sb
, dev_idx
);
7499 state
= imsm_check_degraded(cont
->sb
, dev2
, failed
, MAP_0
);
7500 if (state
== IMSM_T_STATE_FAILED
) {
7501 map
= get_imsm_map(dev2
, MAP_0
);
7504 for (slot
= 0; slot
< map
->num_members
; slot
++) {
7506 * Check if failed disks are deleted from intel
7507 * disk list or are marked to be deleted
7509 idx
= get_imsm_disk_idx(dev2
, slot
, MAP_X
);
7510 idisk
= get_imsm_dl_disk(cont
->sb
, idx
);
7512 * Do not rebuild the array if failed disks
7513 * from failed sub-array are not removed from
7517 is_failed(&idisk
->disk
) &&
7518 (idisk
->action
!= DISK_REMOVE
))
7526 static struct mdinfo
*imsm_activate_spare(struct active_array
*a
,
7527 struct metadata_update
**updates
)
7530 * Find a device with unused free space and use it to replace a
7531 * failed/vacant region in an array. We replace failed regions one a
7532 * array at a time. The result is that a new spare disk will be added
7533 * to the first failed array and after the monitor has finished
7534 * propagating failures the remainder will be consumed.
7536 * FIXME add a capability for mdmon to request spares from another
7540 struct intel_super
*super
= a
->container
->sb
;
7541 int inst
= a
->info
.container_member
;
7542 struct imsm_dev
*dev
= get_imsm_dev(super
, inst
);
7543 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
7544 int failed
= a
->info
.array
.raid_disks
;
7545 struct mdinfo
*rv
= NULL
;
7548 struct metadata_update
*mu
;
7550 struct imsm_update_activate_spare
*u
;
7555 for (d
= a
->info
.devs
; d
; d
= d
->next
) {
7556 if ((d
->curr_state
& DS_FAULTY
) &&
7558 /* wait for Removal to happen */
7560 if (d
->state_fd
>= 0)
7564 dprintf("imsm: activate spare: inst=%d failed=%d (%d) level=%d\n",
7565 inst
, failed
, a
->info
.array
.raid_disks
, a
->info
.array
.level
);
7567 if (imsm_reshape_blocks_arrays_changes(super
))
7570 /* Cannot activate another spare if rebuild is in progress already
7572 if (is_rebuilding(dev
)) {
7573 dprintf("imsm: No spare activation allowed. "
7574 "Rebuild in progress already.\n");
7578 if (a
->info
.array
.level
== 4)
7579 /* No repair for takeovered array
7580 * imsm doesn't support raid4
7584 if (imsm_check_degraded(super
, dev
, failed
, MAP_0
) !=
7585 IMSM_T_STATE_DEGRADED
)
7589 * If there are any failed disks check state of the other volume.
7590 * Block rebuild if the another one is failed until failed disks
7591 * are removed from container.
7594 dprintf("found failed disks in %.*s, check if there another"
7595 "failed sub-array.\n",
7596 MAX_RAID_SERIAL_LEN
, dev
->volume
);
7597 /* check if states of the other volumes allow for rebuild */
7598 for (i
= 0; i
< super
->anchor
->num_raid_devs
; i
++) {
7600 allowed
= imsm_rebuild_allowed(a
->container
,
7608 /* For each slot, if it is not working, find a spare */
7609 for (i
= 0; i
< a
->info
.array
.raid_disks
; i
++) {
7610 for (d
= a
->info
.devs
; d
; d
= d
->next
)
7611 if (d
->disk
.raid_disk
== i
)
7613 dprintf("found %d: %p %x\n", i
, d
, d
?d
->curr_state
:0);
7614 if (d
&& (d
->state_fd
>= 0))
7618 * OK, this device needs recovery. Try to re-add the
7619 * previous occupant of this slot, if this fails see if
7620 * we can continue the assimilation of a spare that was
7621 * partially assimilated, finally try to activate a new
7624 dl
= imsm_readd(super
, i
, a
);
7626 dl
= imsm_add_spare(super
, i
, a
, 0, rv
);
7628 dl
= imsm_add_spare(super
, i
, a
, 1, rv
);
7632 /* found a usable disk with enough space */
7633 di
= malloc(sizeof(*di
));
7636 memset(di
, 0, sizeof(*di
));
7638 /* dl->index will be -1 in the case we are activating a
7639 * pristine spare. imsm_process_update() will create a
7640 * new index in this case. Once a disk is found to be
7641 * failed in all member arrays it is kicked from the
7644 di
->disk
.number
= dl
->index
;
7646 /* (ab)use di->devs to store a pointer to the device
7649 di
->devs
= (struct mdinfo
*) dl
;
7651 di
->disk
.raid_disk
= i
;
7652 di
->disk
.major
= dl
->major
;
7653 di
->disk
.minor
= dl
->minor
;
7655 di
->recovery_start
= 0;
7656 di
->data_offset
= pba_of_lba0(map
);
7657 di
->component_size
= a
->info
.component_size
;
7658 di
->container_member
= inst
;
7659 super
->random
= random32();
7663 dprintf("%x:%x to be %d at %llu\n", dl
->major
, dl
->minor
,
7664 i
, di
->data_offset
);
7668 /* No spares found */
7670 /* Now 'rv' has a list of devices to return.
7671 * Create a metadata_update record to update the
7672 * disk_ord_tbl for the array
7674 mu
= malloc(sizeof(*mu
));
7676 mu
->buf
= malloc(sizeof(struct imsm_update_activate_spare
) * num_spares
);
7677 if (mu
->buf
== NULL
) {
7684 struct mdinfo
*n
= rv
->next
;
7693 mu
->space_list
= NULL
;
7694 mu
->len
= sizeof(struct imsm_update_activate_spare
) * num_spares
;
7695 mu
->next
= *updates
;
7696 u
= (struct imsm_update_activate_spare
*) mu
->buf
;
7698 for (di
= rv
; di
; di
= di
->next
) {
7699 u
->type
= update_activate_spare
;
7700 u
->dl
= (struct dl
*) di
->devs
;
7702 u
->slot
= di
->disk
.raid_disk
;
7713 static int disks_overlap(struct intel_super
*super
, int idx
, struct imsm_update_create_array
*u
)
7715 struct imsm_dev
*dev
= get_imsm_dev(super
, idx
);
7716 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
7717 struct imsm_map
*new_map
= get_imsm_map(&u
->dev
, MAP_0
);
7718 struct disk_info
*inf
= get_disk_info(u
);
7719 struct imsm_disk
*disk
;
7723 for (i
= 0; i
< map
->num_members
; i
++) {
7724 disk
= get_imsm_disk(super
, get_imsm_disk_idx(dev
, i
, MAP_X
));
7725 for (j
= 0; j
< new_map
->num_members
; j
++)
7726 if (serialcmp(disk
->serial
, inf
[j
].serial
) == 0)
7734 static struct dl
*get_disk_super(struct intel_super
*super
, int major
, int minor
)
7736 struct dl
*dl
= NULL
;
7737 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
7738 if ((dl
->major
== major
) && (dl
->minor
== minor
))
7743 static int remove_disk_super(struct intel_super
*super
, int major
, int minor
)
7745 struct dl
*prev
= NULL
;
7749 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
7750 if ((dl
->major
== major
) && (dl
->minor
== minor
)) {
7753 prev
->next
= dl
->next
;
7755 super
->disks
= dl
->next
;
7757 __free_imsm_disk(dl
);
7758 dprintf("%s: removed %x:%x\n",
7759 __func__
, major
, minor
);
7767 static void imsm_delete(struct intel_super
*super
, struct dl
**dlp
, unsigned index
);
7769 static int add_remove_disk_update(struct intel_super
*super
)
7771 int check_degraded
= 0;
7772 struct dl
*disk
= NULL
;
7773 /* add/remove some spares to/from the metadata/contrainer */
7774 while (super
->disk_mgmt_list
) {
7775 struct dl
*disk_cfg
;
7777 disk_cfg
= super
->disk_mgmt_list
;
7778 super
->disk_mgmt_list
= disk_cfg
->next
;
7779 disk_cfg
->next
= NULL
;
7781 if (disk_cfg
->action
== DISK_ADD
) {
7782 disk_cfg
->next
= super
->disks
;
7783 super
->disks
= disk_cfg
;
7785 dprintf("%s: added %x:%x\n",
7786 __func__
, disk_cfg
->major
,
7788 } else if (disk_cfg
->action
== DISK_REMOVE
) {
7789 dprintf("Disk remove action processed: %x.%x\n",
7790 disk_cfg
->major
, disk_cfg
->minor
);
7791 disk
= get_disk_super(super
,
7795 /* store action status */
7796 disk
->action
= DISK_REMOVE
;
7797 /* remove spare disks only */
7798 if (disk
->index
== -1) {
7799 remove_disk_super(super
,
7804 /* release allocate disk structure */
7805 __free_imsm_disk(disk_cfg
);
7808 return check_degraded
;
7812 static int apply_reshape_migration_update(struct imsm_update_reshape_migration
*u
,
7813 struct intel_super
*super
,
7816 struct intel_dev
*id
;
7817 void **tofree
= NULL
;
7820 dprintf("apply_reshape_migration_update()\n");
7821 if ((u
->subdev
< 0) ||
7823 dprintf("imsm: Error: Wrong subdev: %i\n", u
->subdev
);
7826 if ((space_list
== NULL
) || (*space_list
== NULL
)) {
7827 dprintf("imsm: Error: Memory is not allocated\n");
7831 for (id
= super
->devlist
; id
; id
= id
->next
) {
7832 if (id
->index
== (unsigned)u
->subdev
) {
7833 struct imsm_dev
*dev
= get_imsm_dev(super
, u
->subdev
);
7834 struct imsm_map
*map
;
7835 struct imsm_dev
*new_dev
=
7836 (struct imsm_dev
*)*space_list
;
7837 struct imsm_map
*migr_map
= get_imsm_map(dev
, MAP_1
);
7839 struct dl
*new_disk
;
7841 if (new_dev
== NULL
)
7843 *space_list
= **space_list
;
7844 memcpy(new_dev
, dev
, sizeof_imsm_dev(dev
, 0));
7845 map
= get_imsm_map(new_dev
, MAP_0
);
7847 dprintf("imsm: Error: migration in progress");
7851 to_state
= map
->map_state
;
7852 if ((u
->new_level
== 5) && (map
->raid_level
== 0)) {
7854 /* this should not happen */
7855 if (u
->new_disks
[0] < 0) {
7856 map
->failed_disk_num
=
7857 map
->num_members
- 1;
7858 to_state
= IMSM_T_STATE_DEGRADED
;
7860 to_state
= IMSM_T_STATE_NORMAL
;
7862 migrate(new_dev
, super
, to_state
, MIGR_GEN_MIGR
);
7863 if (u
->new_level
> -1)
7864 map
->raid_level
= u
->new_level
;
7865 migr_map
= get_imsm_map(new_dev
, MAP_1
);
7866 if ((u
->new_level
== 5) &&
7867 (migr_map
->raid_level
== 0)) {
7868 int ord
= map
->num_members
- 1;
7869 migr_map
->num_members
--;
7870 if (u
->new_disks
[0] < 0)
7871 ord
|= IMSM_ORD_REBUILD
;
7872 set_imsm_ord_tbl_ent(map
,
7873 map
->num_members
- 1,
7877 tofree
= (void **)dev
;
7879 /* update chunk size
7881 if (u
->new_chunksize
> 0)
7882 map
->blocks_per_strip
=
7883 __cpu_to_le16(u
->new_chunksize
* 2);
7887 if ((u
->new_level
!= 5) ||
7888 (migr_map
->raid_level
!= 0) ||
7889 (migr_map
->raid_level
== map
->raid_level
))
7892 if (u
->new_disks
[0] >= 0) {
7895 new_disk
= get_disk_super(super
,
7896 major(u
->new_disks
[0]),
7897 minor(u
->new_disks
[0]));
7898 dprintf("imsm: new disk for reshape is: %i:%i "
7899 "(%p, index = %i)\n",
7900 major(u
->new_disks
[0]),
7901 minor(u
->new_disks
[0]),
7902 new_disk
, new_disk
->index
);
7903 if (new_disk
== NULL
)
7904 goto error_disk_add
;
7906 new_disk
->index
= map
->num_members
- 1;
7907 /* slot to fill in autolayout
7909 new_disk
->raiddisk
= new_disk
->index
;
7910 new_disk
->disk
.status
|= CONFIGURED_DISK
;
7911 new_disk
->disk
.status
&= ~SPARE_DISK
;
7913 goto error_disk_add
;
7916 *tofree
= *space_list
;
7917 /* calculate new size
7919 imsm_set_array_size(new_dev
);
7926 *space_list
= tofree
;
7930 dprintf("Error: imsm: Cannot find disk.\n");
7934 static int apply_update_activate_spare(struct imsm_update_activate_spare
*u
,
7935 struct intel_super
*super
,
7936 struct active_array
*active_array
)
7938 struct imsm_super
*mpb
= super
->anchor
;
7939 struct imsm_dev
*dev
= get_imsm_dev(super
, u
->array
);
7940 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
7941 struct imsm_map
*migr_map
;
7942 struct active_array
*a
;
7943 struct imsm_disk
*disk
;
7950 int second_map_created
= 0;
7952 for (; u
; u
= u
->next
) {
7953 victim
= get_imsm_disk_idx(dev
, u
->slot
, MAP_X
);
7958 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
7963 fprintf(stderr
, "error: imsm_activate_spare passed "
7964 "an unknown disk (index: %d)\n",
7969 /* count failures (excluding rebuilds and the victim)
7970 * to determine map[0] state
7973 for (i
= 0; i
< map
->num_members
; i
++) {
7976 disk
= get_imsm_disk(super
,
7977 get_imsm_disk_idx(dev
, i
, MAP_X
));
7978 if (!disk
|| is_failed(disk
))
7982 /* adding a pristine spare, assign a new index */
7983 if (dl
->index
< 0) {
7984 dl
->index
= super
->anchor
->num_disks
;
7985 super
->anchor
->num_disks
++;
7988 disk
->status
|= CONFIGURED_DISK
;
7989 disk
->status
&= ~SPARE_DISK
;
7992 to_state
= imsm_check_degraded(super
, dev
, failed
, MAP_0
);
7993 if (!second_map_created
) {
7994 second_map_created
= 1;
7995 map
->map_state
= IMSM_T_STATE_DEGRADED
;
7996 migrate(dev
, super
, to_state
, MIGR_REBUILD
);
7998 map
->map_state
= to_state
;
7999 migr_map
= get_imsm_map(dev
, MAP_1
);
8000 set_imsm_ord_tbl_ent(map
, u
->slot
, dl
->index
);
8001 set_imsm_ord_tbl_ent(migr_map
, u
->slot
,
8002 dl
->index
| IMSM_ORD_REBUILD
);
8004 /* update the family_num to mark a new container
8005 * generation, being careful to record the existing
8006 * family_num in orig_family_num to clean up after
8007 * earlier mdadm versions that neglected to set it.
8009 if (mpb
->orig_family_num
== 0)
8010 mpb
->orig_family_num
= mpb
->family_num
;
8011 mpb
->family_num
+= super
->random
;
8013 /* count arrays using the victim in the metadata */
8015 for (a
= active_array
; a
; a
= a
->next
) {
8016 dev
= get_imsm_dev(super
, a
->info
.container_member
);
8017 map
= get_imsm_map(dev
, MAP_0
);
8019 if (get_imsm_disk_slot(map
, victim
) >= 0)
8023 /* delete the victim if it is no longer being
8029 /* We know that 'manager' isn't touching anything,
8030 * so it is safe to delete
8032 for (dlp
= &super
->disks
; *dlp
; dlp
= &(*dlp
)->next
)
8033 if ((*dlp
)->index
== victim
)
8036 /* victim may be on the missing list */
8038 for (dlp
= &super
->missing
; *dlp
;
8039 dlp
= &(*dlp
)->next
)
8040 if ((*dlp
)->index
== victim
)
8042 imsm_delete(super
, dlp
, victim
);
8049 static int apply_reshape_container_disks_update(struct imsm_update_reshape
*u
,
8050 struct intel_super
*super
,
8053 struct dl
*new_disk
;
8054 struct intel_dev
*id
;
8056 int delta_disks
= u
->new_raid_disks
- u
->old_raid_disks
;
8057 int disk_count
= u
->old_raid_disks
;
8058 void **tofree
= NULL
;
8059 int devices_to_reshape
= 1;
8060 struct imsm_super
*mpb
= super
->anchor
;
8062 unsigned int dev_id
;
8064 dprintf("imsm: apply_reshape_container_disks_update()\n");
8066 /* enable spares to use in array */
8067 for (i
= 0; i
< delta_disks
; i
++) {
8068 new_disk
= get_disk_super(super
,
8069 major(u
->new_disks
[i
]),
8070 minor(u
->new_disks
[i
]));
8071 dprintf("imsm: new disk for reshape is: %i:%i "
8072 "(%p, index = %i)\n",
8073 major(u
->new_disks
[i
]), minor(u
->new_disks
[i
]),
8074 new_disk
, new_disk
->index
);
8075 if ((new_disk
== NULL
) ||
8076 ((new_disk
->index
>= 0) &&
8077 (new_disk
->index
< u
->old_raid_disks
)))
8078 goto update_reshape_exit
;
8079 new_disk
->index
= disk_count
++;
8080 /* slot to fill in autolayout
8082 new_disk
->raiddisk
= new_disk
->index
;
8083 new_disk
->disk
.status
|=
8085 new_disk
->disk
.status
&= ~SPARE_DISK
;
8088 dprintf("imsm: volume set mpb->num_raid_devs = %i\n",
8089 mpb
->num_raid_devs
);
8090 /* manage changes in volume
8092 for (dev_id
= 0; dev_id
< mpb
->num_raid_devs
; dev_id
++) {
8093 void **sp
= *space_list
;
8094 struct imsm_dev
*newdev
;
8095 struct imsm_map
*newmap
, *oldmap
;
8097 for (id
= super
->devlist
; id
; id
= id
->next
) {
8098 if (id
->index
== dev_id
)
8107 /* Copy the dev, but not (all of) the map */
8108 memcpy(newdev
, id
->dev
, sizeof(*newdev
));
8109 oldmap
= get_imsm_map(id
->dev
, MAP_0
);
8110 newmap
= get_imsm_map(newdev
, MAP_0
);
8111 /* Copy the current map */
8112 memcpy(newmap
, oldmap
, sizeof_imsm_map(oldmap
));
8113 /* update one device only
8115 if (devices_to_reshape
) {
8116 dprintf("imsm: modifying subdev: %i\n",
8118 devices_to_reshape
--;
8119 newdev
->vol
.migr_state
= 1;
8120 newdev
->vol
.curr_migr_unit
= 0;
8121 set_migr_type(newdev
, MIGR_GEN_MIGR
);
8122 newmap
->num_members
= u
->new_raid_disks
;
8123 for (i
= 0; i
< delta_disks
; i
++) {
8124 set_imsm_ord_tbl_ent(newmap
,
8125 u
->old_raid_disks
+ i
,
8126 u
->old_raid_disks
+ i
);
8128 /* New map is correct, now need to save old map
8130 newmap
= get_imsm_map(newdev
, MAP_1
);
8131 memcpy(newmap
, oldmap
, sizeof_imsm_map(oldmap
));
8133 imsm_set_array_size(newdev
);
8136 sp
= (void **)id
->dev
;
8141 /* Clear migration record */
8142 memset(super
->migr_rec
, 0, sizeof(struct migr_record
));
8145 *space_list
= tofree
;
8148 update_reshape_exit
:
8153 static int apply_takeover_update(struct imsm_update_takeover
*u
,
8154 struct intel_super
*super
,
8157 struct imsm_dev
*dev
= NULL
;
8158 struct intel_dev
*dv
;
8159 struct imsm_dev
*dev_new
;
8160 struct imsm_map
*map
;
8164 for (dv
= super
->devlist
; dv
; dv
= dv
->next
)
8165 if (dv
->index
== (unsigned int)u
->subarray
) {
8173 map
= get_imsm_map(dev
, MAP_0
);
8175 if (u
->direction
== R10_TO_R0
) {
8176 /* Number of failed disks must be half of initial disk number */
8177 if (imsm_count_failed(super
, dev
, MAP_0
) !=
8178 (map
->num_members
/ 2))
8181 /* iterate through devices to mark removed disks as spare */
8182 for (dm
= super
->disks
; dm
; dm
= dm
->next
) {
8183 if (dm
->disk
.status
& FAILED_DISK
) {
8184 int idx
= dm
->index
;
8185 /* update indexes on the disk list */
8186 /* FIXME this loop-with-the-loop looks wrong, I'm not convinced
8187 the index values will end up being correct.... NB */
8188 for (du
= super
->disks
; du
; du
= du
->next
)
8189 if (du
->index
> idx
)
8191 /* mark as spare disk */
8196 map
->num_members
= map
->num_members
/ 2;
8197 map
->map_state
= IMSM_T_STATE_NORMAL
;
8198 map
->num_domains
= 1;
8199 map
->raid_level
= 0;
8200 map
->failed_disk_num
= -1;
8203 if (u
->direction
== R0_TO_R10
) {
8205 /* update slots in current disk list */
8206 for (dm
= super
->disks
; dm
; dm
= dm
->next
) {
8210 /* create new *missing* disks */
8211 for (i
= 0; i
< map
->num_members
; i
++) {
8212 space
= *space_list
;
8215 *space_list
= *space
;
8217 memcpy(du
, super
->disks
, sizeof(*du
));
8221 du
->index
= (i
* 2) + 1;
8222 sprintf((char *)du
->disk
.serial
,
8223 " MISSING_%d", du
->index
);
8224 sprintf((char *)du
->serial
,
8225 "MISSING_%d", du
->index
);
8226 du
->next
= super
->missing
;
8227 super
->missing
= du
;
8229 /* create new dev and map */
8230 space
= *space_list
;
8233 *space_list
= *space
;
8234 dev_new
= (void *)space
;
8235 memcpy(dev_new
, dev
, sizeof(*dev
));
8236 /* update new map */
8237 map
= get_imsm_map(dev_new
, MAP_0
);
8238 map
->num_members
= map
->num_members
* 2;
8239 map
->map_state
= IMSM_T_STATE_DEGRADED
;
8240 map
->num_domains
= 2;
8241 map
->raid_level
= 1;
8242 /* replace dev<->dev_new */
8245 /* update disk order table */
8246 for (du
= super
->disks
; du
; du
= du
->next
)
8248 set_imsm_ord_tbl_ent(map
, du
->index
, du
->index
);
8249 for (du
= super
->missing
; du
; du
= du
->next
)
8250 if (du
->index
>= 0) {
8251 set_imsm_ord_tbl_ent(map
, du
->index
, du
->index
);
8252 mark_missing(dv
->dev
, &du
->disk
, du
->index
);
8258 static void imsm_process_update(struct supertype
*st
,
8259 struct metadata_update
*update
)
8262 * crack open the metadata_update envelope to find the update record
8263 * update can be one of:
8264 * update_reshape_container_disks - all the arrays in the container
8265 * are being reshaped to have more devices. We need to mark
8266 * the arrays for general migration and convert selected spares
8267 * into active devices.
8268 * update_activate_spare - a spare device has replaced a failed
8269 * device in an array, update the disk_ord_tbl. If this disk is
8270 * present in all member arrays then also clear the SPARE_DISK
8272 * update_create_array
8274 * update_rename_array
8275 * update_add_remove_disk
8277 struct intel_super
*super
= st
->sb
;
8278 struct imsm_super
*mpb
;
8279 enum imsm_update_type type
= *(enum imsm_update_type
*) update
->buf
;
8281 /* update requires a larger buf but the allocation failed */
8282 if (super
->next_len
&& !super
->next_buf
) {
8283 super
->next_len
= 0;
8287 if (super
->next_buf
) {
8288 memcpy(super
->next_buf
, super
->buf
, super
->len
);
8290 super
->len
= super
->next_len
;
8291 super
->buf
= super
->next_buf
;
8293 super
->next_len
= 0;
8294 super
->next_buf
= NULL
;
8297 mpb
= super
->anchor
;
8300 case update_general_migration_checkpoint
: {
8301 struct intel_dev
*id
;
8302 struct imsm_update_general_migration_checkpoint
*u
=
8303 (void *)update
->buf
;
8305 dprintf("imsm: process_update() "
8306 "for update_general_migration_checkpoint called\n");
8308 /* find device under general migration */
8309 for (id
= super
->devlist
; id
; id
= id
->next
) {
8310 if (is_gen_migration(id
->dev
)) {
8311 id
->dev
->vol
.curr_migr_unit
=
8312 __cpu_to_le32(u
->curr_migr_unit
);
8313 super
->updates_pending
++;
8318 case update_takeover
: {
8319 struct imsm_update_takeover
*u
= (void *)update
->buf
;
8320 if (apply_takeover_update(u
, super
, &update
->space_list
)) {
8321 imsm_update_version_info(super
);
8322 super
->updates_pending
++;
8327 case update_reshape_container_disks
: {
8328 struct imsm_update_reshape
*u
= (void *)update
->buf
;
8329 if (apply_reshape_container_disks_update(
8330 u
, super
, &update
->space_list
))
8331 super
->updates_pending
++;
8334 case update_reshape_migration
: {
8335 struct imsm_update_reshape_migration
*u
= (void *)update
->buf
;
8336 if (apply_reshape_migration_update(
8337 u
, super
, &update
->space_list
))
8338 super
->updates_pending
++;
8341 case update_activate_spare
: {
8342 struct imsm_update_activate_spare
*u
= (void *) update
->buf
;
8343 if (apply_update_activate_spare(u
, super
, st
->arrays
))
8344 super
->updates_pending
++;
8347 case update_create_array
: {
8348 /* someone wants to create a new array, we need to be aware of
8349 * a few races/collisions:
8350 * 1/ 'Create' called by two separate instances of mdadm
8351 * 2/ 'Create' versus 'activate_spare': mdadm has chosen
8352 * devices that have since been assimilated via
8354 * In the event this update can not be carried out mdadm will
8355 * (FIX ME) notice that its update did not take hold.
8357 struct imsm_update_create_array
*u
= (void *) update
->buf
;
8358 struct intel_dev
*dv
;
8359 struct imsm_dev
*dev
;
8360 struct imsm_map
*map
, *new_map
;
8361 unsigned long long start
, end
;
8362 unsigned long long new_start
, new_end
;
8364 struct disk_info
*inf
;
8367 /* handle racing creates: first come first serve */
8368 if (u
->dev_idx
< mpb
->num_raid_devs
) {
8369 dprintf("%s: subarray %d already defined\n",
8370 __func__
, u
->dev_idx
);
8374 /* check update is next in sequence */
8375 if (u
->dev_idx
!= mpb
->num_raid_devs
) {
8376 dprintf("%s: can not create array %d expected index %d\n",
8377 __func__
, u
->dev_idx
, mpb
->num_raid_devs
);
8381 new_map
= get_imsm_map(&u
->dev
, MAP_0
);
8382 new_start
= pba_of_lba0(new_map
);
8383 new_end
= new_start
+ blocks_per_member(new_map
);
8384 inf
= get_disk_info(u
);
8386 /* handle activate_spare versus create race:
8387 * check to make sure that overlapping arrays do not include
8390 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
8391 dev
= get_imsm_dev(super
, i
);
8392 map
= get_imsm_map(dev
, MAP_0
);
8393 start
= pba_of_lba0(map
);
8394 end
= start
+ blocks_per_member(map
);
8395 if ((new_start
>= start
&& new_start
<= end
) ||
8396 (start
>= new_start
&& start
<= new_end
))
8401 if (disks_overlap(super
, i
, u
)) {
8402 dprintf("%s: arrays overlap\n", __func__
);
8407 /* check that prepare update was successful */
8408 if (!update
->space
) {
8409 dprintf("%s: prepare update failed\n", __func__
);
8413 /* check that all disks are still active before committing
8414 * changes. FIXME: could we instead handle this by creating a
8415 * degraded array? That's probably not what the user expects,
8416 * so better to drop this update on the floor.
8418 for (i
= 0; i
< new_map
->num_members
; i
++) {
8419 dl
= serial_to_dl(inf
[i
].serial
, super
);
8421 dprintf("%s: disk disappeared\n", __func__
);
8426 super
->updates_pending
++;
8428 /* convert spares to members and fixup ord_tbl */
8429 for (i
= 0; i
< new_map
->num_members
; i
++) {
8430 dl
= serial_to_dl(inf
[i
].serial
, super
);
8431 if (dl
->index
== -1) {
8432 dl
->index
= mpb
->num_disks
;
8434 dl
->disk
.status
|= CONFIGURED_DISK
;
8435 dl
->disk
.status
&= ~SPARE_DISK
;
8437 set_imsm_ord_tbl_ent(new_map
, i
, dl
->index
);
8442 update
->space
= NULL
;
8443 imsm_copy_dev(dev
, &u
->dev
);
8444 dv
->index
= u
->dev_idx
;
8445 dv
->next
= super
->devlist
;
8446 super
->devlist
= dv
;
8447 mpb
->num_raid_devs
++;
8449 imsm_update_version_info(super
);
8452 /* mdmon knows how to release update->space, but not
8453 * ((struct intel_dev *) update->space)->dev
8455 if (update
->space
) {
8461 case update_kill_array
: {
8462 struct imsm_update_kill_array
*u
= (void *) update
->buf
;
8463 int victim
= u
->dev_idx
;
8464 struct active_array
*a
;
8465 struct intel_dev
**dp
;
8466 struct imsm_dev
*dev
;
8468 /* sanity check that we are not affecting the uuid of
8469 * active arrays, or deleting an active array
8471 * FIXME when immutable ids are available, but note that
8472 * we'll also need to fixup the invalidated/active
8473 * subarray indexes in mdstat
8475 for (a
= st
->arrays
; a
; a
= a
->next
)
8476 if (a
->info
.container_member
>= victim
)
8478 /* by definition if mdmon is running at least one array
8479 * is active in the container, so checking
8480 * mpb->num_raid_devs is just extra paranoia
8482 dev
= get_imsm_dev(super
, victim
);
8483 if (a
|| !dev
|| mpb
->num_raid_devs
== 1) {
8484 dprintf("failed to delete subarray-%d\n", victim
);
8488 for (dp
= &super
->devlist
; *dp
;)
8489 if ((*dp
)->index
== (unsigned)super
->current_vol
) {
8492 if ((*dp
)->index
> (unsigned)victim
)
8496 mpb
->num_raid_devs
--;
8497 super
->updates_pending
++;
8500 case update_rename_array
: {
8501 struct imsm_update_rename_array
*u
= (void *) update
->buf
;
8502 char name
[MAX_RAID_SERIAL_LEN
+1];
8503 int target
= u
->dev_idx
;
8504 struct active_array
*a
;
8505 struct imsm_dev
*dev
;
8507 /* sanity check that we are not affecting the uuid of
8510 snprintf(name
, MAX_RAID_SERIAL_LEN
, "%s", (char *) u
->name
);
8511 name
[MAX_RAID_SERIAL_LEN
] = '\0';
8512 for (a
= st
->arrays
; a
; a
= a
->next
)
8513 if (a
->info
.container_member
== target
)
8515 dev
= get_imsm_dev(super
, u
->dev_idx
);
8516 if (a
|| !dev
|| !check_name(super
, name
, 1)) {
8517 dprintf("failed to rename subarray-%d\n", target
);
8521 snprintf((char *) dev
->volume
, MAX_RAID_SERIAL_LEN
, "%s", name
);
8522 super
->updates_pending
++;
8525 case update_add_remove_disk
: {
8526 /* we may be able to repair some arrays if disks are
8527 * being added, check teh status of add_remove_disk
8528 * if discs has been added.
8530 if (add_remove_disk_update(super
)) {
8531 struct active_array
*a
;
8533 super
->updates_pending
++;
8534 for (a
= st
->arrays
; a
; a
= a
->next
)
8535 a
->check_degraded
= 1;
8540 fprintf(stderr
, "error: unsuported process update type:"
8541 "(type: %d)\n", type
);
8545 static struct mdinfo
*get_spares_for_grow(struct supertype
*st
);
8547 static void imsm_prepare_update(struct supertype
*st
,
8548 struct metadata_update
*update
)
8551 * Allocate space to hold new disk entries, raid-device entries or a new
8552 * mpb if necessary. The manager synchronously waits for updates to
8553 * complete in the monitor, so new mpb buffers allocated here can be
8554 * integrated by the monitor thread without worrying about live pointers
8555 * in the manager thread.
8557 enum imsm_update_type type
= *(enum imsm_update_type
*) update
->buf
;
8558 struct intel_super
*super
= st
->sb
;
8559 struct imsm_super
*mpb
= super
->anchor
;
8564 case update_general_migration_checkpoint
:
8565 dprintf("imsm: prepare_update() "
8566 "for update_general_migration_checkpoint called\n");
8568 case update_takeover
: {
8569 struct imsm_update_takeover
*u
= (void *)update
->buf
;
8570 if (u
->direction
== R0_TO_R10
) {
8571 void **tail
= (void **)&update
->space_list
;
8572 struct imsm_dev
*dev
= get_imsm_dev(super
, u
->subarray
);
8573 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
8574 int num_members
= map
->num_members
;
8578 /* allocate memory for added disks */
8579 for (i
= 0; i
< num_members
; i
++) {
8580 size
= sizeof(struct dl
);
8581 space
= malloc(size
);
8590 /* allocate memory for new device */
8591 size
= sizeof_imsm_dev(super
->devlist
->dev
, 0) +
8592 (num_members
* sizeof(__u32
));
8593 space
= malloc(size
);
8602 len
= disks_to_mpb_size(num_members
* 2);
8604 /* if allocation didn't success, free buffer */
8605 while (update
->space_list
) {
8606 void **sp
= update
->space_list
;
8607 update
->space_list
= *sp
;
8615 case update_reshape_container_disks
: {
8616 /* Every raid device in the container is about to
8617 * gain some more devices, and we will enter a
8619 * So each 'imsm_map' will be bigger, and the imsm_vol
8620 * will now hold 2 of them.
8621 * Thus we need new 'struct imsm_dev' allocations sized
8622 * as sizeof_imsm_dev but with more devices in both maps.
8624 struct imsm_update_reshape
*u
= (void *)update
->buf
;
8625 struct intel_dev
*dl
;
8626 void **space_tail
= (void**)&update
->space_list
;
8628 dprintf("imsm: imsm_prepare_update() for update_reshape\n");
8630 for (dl
= super
->devlist
; dl
; dl
= dl
->next
) {
8631 int size
= sizeof_imsm_dev(dl
->dev
, 1);
8633 if (u
->new_raid_disks
> u
->old_raid_disks
)
8634 size
+= sizeof(__u32
)*2*
8635 (u
->new_raid_disks
- u
->old_raid_disks
);
8644 len
= disks_to_mpb_size(u
->new_raid_disks
);
8645 dprintf("New anchor length is %llu\n", (unsigned long long)len
);
8648 case update_reshape_migration
: {
8649 /* for migration level 0->5 we need to add disks
8650 * so the same as for container operation we will copy
8651 * device to the bigger location.
8652 * in memory prepared device and new disk area are prepared
8653 * for usage in process update
8655 struct imsm_update_reshape_migration
*u
= (void *)update
->buf
;
8656 struct intel_dev
*id
;
8657 void **space_tail
= (void **)&update
->space_list
;
8660 int current_level
= -1;
8662 dprintf("imsm: imsm_prepare_update() for update_reshape\n");
8664 /* add space for bigger array in update
8666 for (id
= super
->devlist
; id
; id
= id
->next
) {
8667 if (id
->index
== (unsigned)u
->subdev
) {
8668 size
= sizeof_imsm_dev(id
->dev
, 1);
8669 if (u
->new_raid_disks
> u
->old_raid_disks
)
8670 size
+= sizeof(__u32
)*2*
8671 (u
->new_raid_disks
- u
->old_raid_disks
);
8681 if (update
->space_list
== NULL
)
8684 /* add space for disk in update
8686 size
= sizeof(struct dl
);
8689 free(update
->space_list
);
8690 update
->space_list
= NULL
;
8697 /* add spare device to update
8699 for (id
= super
->devlist
; id
; id
= id
->next
)
8700 if (id
->index
== (unsigned)u
->subdev
) {
8701 struct imsm_dev
*dev
;
8702 struct imsm_map
*map
;
8704 dev
= get_imsm_dev(super
, u
->subdev
);
8705 map
= get_imsm_map(dev
, MAP_0
);
8706 current_level
= map
->raid_level
;
8709 if ((u
->new_level
== 5) && (u
->new_level
!= current_level
)) {
8710 struct mdinfo
*spares
;
8712 spares
= get_spares_for_grow(st
);
8720 makedev(dev
->disk
.major
,
8722 dl
= get_disk_super(super
,
8725 dl
->index
= u
->old_raid_disks
;
8731 len
= disks_to_mpb_size(u
->new_raid_disks
);
8732 dprintf("New anchor length is %llu\n", (unsigned long long)len
);
8735 case update_create_array
: {
8736 struct imsm_update_create_array
*u
= (void *) update
->buf
;
8737 struct intel_dev
*dv
;
8738 struct imsm_dev
*dev
= &u
->dev
;
8739 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
8741 struct disk_info
*inf
;
8745 inf
= get_disk_info(u
);
8746 len
= sizeof_imsm_dev(dev
, 1);
8747 /* allocate a new super->devlist entry */
8748 dv
= malloc(sizeof(*dv
));
8750 dv
->dev
= malloc(len
);
8755 update
->space
= NULL
;
8759 /* count how many spares will be converted to members */
8760 for (i
= 0; i
< map
->num_members
; i
++) {
8761 dl
= serial_to_dl(inf
[i
].serial
, super
);
8763 /* hmm maybe it failed?, nothing we can do about
8768 if (count_memberships(dl
, super
) == 0)
8771 len
+= activate
* sizeof(struct imsm_disk
);
8778 /* check if we need a larger metadata buffer */
8779 if (super
->next_buf
)
8780 buf_len
= super
->next_len
;
8782 buf_len
= super
->len
;
8784 if (__le32_to_cpu(mpb
->mpb_size
) + len
> buf_len
) {
8785 /* ok we need a larger buf than what is currently allocated
8786 * if this allocation fails process_update will notice that
8787 * ->next_len is set and ->next_buf is NULL
8789 buf_len
= ROUND_UP(__le32_to_cpu(mpb
->mpb_size
) + len
, 512);
8790 if (super
->next_buf
)
8791 free(super
->next_buf
);
8793 super
->next_len
= buf_len
;
8794 if (posix_memalign(&super
->next_buf
, 512, buf_len
) == 0)
8795 memset(super
->next_buf
, 0, buf_len
);
8797 super
->next_buf
= NULL
;
8801 /* must be called while manager is quiesced */
8802 static void imsm_delete(struct intel_super
*super
, struct dl
**dlp
, unsigned index
)
8804 struct imsm_super
*mpb
= super
->anchor
;
8806 struct imsm_dev
*dev
;
8807 struct imsm_map
*map
;
8808 int i
, j
, num_members
;
8811 dprintf("%s: deleting device[%d] from imsm_super\n",
8814 /* shift all indexes down one */
8815 for (iter
= super
->disks
; iter
; iter
= iter
->next
)
8816 if (iter
->index
> (int)index
)
8818 for (iter
= super
->missing
; iter
; iter
= iter
->next
)
8819 if (iter
->index
> (int)index
)
8822 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
8823 dev
= get_imsm_dev(super
, i
);
8824 map
= get_imsm_map(dev
, MAP_0
);
8825 num_members
= map
->num_members
;
8826 for (j
= 0; j
< num_members
; j
++) {
8827 /* update ord entries being careful not to propagate
8828 * ord-flags to the first map
8830 ord
= get_imsm_ord_tbl_ent(dev
, j
, MAP_X
);
8832 if (ord_to_idx(ord
) <= index
)
8835 map
= get_imsm_map(dev
, MAP_0
);
8836 set_imsm_ord_tbl_ent(map
, j
, ord_to_idx(ord
- 1));
8837 map
= get_imsm_map(dev
, MAP_1
);
8839 set_imsm_ord_tbl_ent(map
, j
, ord
- 1);
8844 super
->updates_pending
++;
8846 struct dl
*dl
= *dlp
;
8848 *dlp
= (*dlp
)->next
;
8849 __free_imsm_disk(dl
);
8852 #endif /* MDASSEMBLE */
8854 static void close_targets(int *targets
, int new_disks
)
8861 for (i
= 0; i
< new_disks
; i
++) {
8862 if (targets
[i
] >= 0) {
8869 static int imsm_get_allowed_degradation(int level
, int raid_disks
,
8870 struct intel_super
*super
,
8871 struct imsm_dev
*dev
)
8877 struct imsm_map
*map
;
8880 ret_val
= raid_disks
/2;
8881 /* check map if all disks pairs not failed
8884 map
= get_imsm_map(dev
, MAP_0
);
8885 for (i
= 0; i
< ret_val
; i
++) {
8886 int degradation
= 0;
8887 if (get_imsm_disk(super
, i
) == NULL
)
8889 if (get_imsm_disk(super
, i
+ 1) == NULL
)
8891 if (degradation
== 2)
8894 map
= get_imsm_map(dev
, MAP_1
);
8895 /* if there is no second map
8896 * result can be returned
8900 /* check degradation in second map
8902 for (i
= 0; i
< ret_val
; i
++) {
8903 int degradation
= 0;
8904 if (get_imsm_disk(super
, i
) == NULL
)
8906 if (get_imsm_disk(super
, i
+ 1) == NULL
)
8908 if (degradation
== 2)
8923 /*******************************************************************************
8924 * Function: open_backup_targets
8925 * Description: Function opens file descriptors for all devices given in
8928 * info : general array info
8929 * raid_disks : number of disks
8930 * raid_fds : table of device's file descriptors
8931 * super : intel super for raid10 degradation check
8932 * dev : intel device for raid10 degradation check
8936 ******************************************************************************/
8937 int open_backup_targets(struct mdinfo
*info
, int raid_disks
, int *raid_fds
,
8938 struct intel_super
*super
, struct imsm_dev
*dev
)
8944 for (i
= 0; i
< raid_disks
; i
++)
8947 for (sd
= info
->devs
; sd
; sd
= sd
->next
) {
8950 if (sd
->disk
.state
& (1<<MD_DISK_FAULTY
)) {
8951 dprintf("disk is faulty!!\n");
8955 if ((sd
->disk
.raid_disk
>= raid_disks
) ||
8956 (sd
->disk
.raid_disk
< 0))
8959 dn
= map_dev(sd
->disk
.major
,
8961 raid_fds
[sd
->disk
.raid_disk
] = dev_open(dn
, O_RDWR
);
8962 if (raid_fds
[sd
->disk
.raid_disk
] < 0) {
8963 fprintf(stderr
, "cannot open component\n");
8968 /* check if maximum array degradation level is not exceeded
8970 if ((raid_disks
- opened
) >
8971 imsm_get_allowed_degradation(info
->new_level
,
8974 fprintf(stderr
, "Not enough disks can be opened.\n");
8975 close_targets(raid_fds
, raid_disks
);
8982 /*******************************************************************************
8983 * Function: init_migr_record_imsm
8984 * Description: Function inits imsm migration record
8986 * super : imsm internal array info
8987 * dev : device under migration
8988 * info : general array info to find the smallest device
8991 ******************************************************************************/
8992 void init_migr_record_imsm(struct supertype
*st
, struct imsm_dev
*dev
,
8993 struct mdinfo
*info
)
8995 struct intel_super
*super
= st
->sb
;
8996 struct migr_record
*migr_rec
= super
->migr_rec
;
8998 unsigned long long dsize
, dev_sectors
;
8999 long long unsigned min_dev_sectors
= -1LLU;
9003 struct imsm_map
*map_dest
= get_imsm_map(dev
, MAP_0
);
9004 struct imsm_map
*map_src
= get_imsm_map(dev
, MAP_1
);
9005 unsigned long long num_migr_units
;
9006 unsigned long long array_blocks
;
9008 memset(migr_rec
, 0, sizeof(struct migr_record
));
9009 migr_rec
->family_num
= __cpu_to_le32(super
->anchor
->family_num
);
9011 /* only ascending reshape supported now */
9012 migr_rec
->ascending_migr
= __cpu_to_le32(1);
9014 migr_rec
->dest_depth_per_unit
= GEN_MIGR_AREA_SIZE
/
9015 max(map_dest
->blocks_per_strip
, map_src
->blocks_per_strip
);
9016 migr_rec
->dest_depth_per_unit
*=
9017 max(map_dest
->blocks_per_strip
, map_src
->blocks_per_strip
);
9018 new_data_disks
= imsm_num_data_members(dev
, MAP_0
);
9019 migr_rec
->blocks_per_unit
=
9020 __cpu_to_le32(migr_rec
->dest_depth_per_unit
* new_data_disks
);
9021 migr_rec
->dest_depth_per_unit
=
9022 __cpu_to_le32(migr_rec
->dest_depth_per_unit
);
9023 array_blocks
= info
->component_size
* new_data_disks
;
9025 array_blocks
/ __le32_to_cpu(migr_rec
->blocks_per_unit
);
9027 if (array_blocks
% __le32_to_cpu(migr_rec
->blocks_per_unit
))
9029 migr_rec
->num_migr_units
= __cpu_to_le32(num_migr_units
);
9031 migr_rec
->post_migr_vol_cap
= dev
->size_low
;
9032 migr_rec
->post_migr_vol_cap_hi
= dev
->size_high
;
9035 /* Find the smallest dev */
9036 for (sd
= info
->devs
; sd
; sd
= sd
->next
) {
9037 sprintf(nm
, "%d:%d", sd
->disk
.major
, sd
->disk
.minor
);
9038 fd
= dev_open(nm
, O_RDONLY
);
9041 get_dev_size(fd
, NULL
, &dsize
);
9042 dev_sectors
= dsize
/ 512;
9043 if (dev_sectors
< min_dev_sectors
)
9044 min_dev_sectors
= dev_sectors
;
9047 migr_rec
->ckpt_area_pba
= __cpu_to_le32(min_dev_sectors
-
9048 RAID_DISK_RESERVED_BLOCKS_IMSM_HI
);
9050 write_imsm_migr_rec(st
);
9055 /*******************************************************************************
9056 * Function: save_backup_imsm
9057 * Description: Function saves critical data stripes to Migration Copy Area
9058 * and updates the current migration unit status.
9059 * Use restore_stripes() to form a destination stripe,
9060 * and to write it to the Copy Area.
9062 * st : supertype information
9063 * dev : imsm device that backup is saved for
9064 * info : general array info
9065 * buf : input buffer
9066 * length : length of data to backup (blocks_per_unit)
9070 ******************************************************************************/
9071 int save_backup_imsm(struct supertype
*st
,
9072 struct imsm_dev
*dev
,
9073 struct mdinfo
*info
,
9078 struct intel_super
*super
= st
->sb
;
9079 unsigned long long *target_offsets
= NULL
;
9080 int *targets
= NULL
;
9082 struct imsm_map
*map_dest
= get_imsm_map(dev
, MAP_0
);
9083 int new_disks
= map_dest
->num_members
;
9084 int dest_layout
= 0;
9086 unsigned long long start
;
9087 int data_disks
= imsm_num_data_members(dev
, MAP_0
);
9089 targets
= malloc(new_disks
* sizeof(int));
9093 for (i
= 0; i
< new_disks
; i
++)
9096 target_offsets
= malloc(new_disks
* sizeof(unsigned long long));
9097 if (!target_offsets
)
9100 start
= info
->reshape_progress
* 512;
9101 for (i
= 0; i
< new_disks
; i
++) {
9102 target_offsets
[i
] = (unsigned long long)
9103 __le32_to_cpu(super
->migr_rec
->ckpt_area_pba
) * 512;
9104 /* move back copy area adderss, it will be moved forward
9105 * in restore_stripes() using start input variable
9107 target_offsets
[i
] -= start
/data_disks
;
9110 if (open_backup_targets(info
, new_disks
, targets
,
9114 dest_layout
= imsm_level_to_layout(map_dest
->raid_level
);
9115 dest_chunk
= __le16_to_cpu(map_dest
->blocks_per_strip
) * 512;
9117 if (restore_stripes(targets
, /* list of dest devices */
9118 target_offsets
, /* migration record offsets */
9121 map_dest
->raid_level
,
9123 -1, /* source backup file descriptor */
9124 0, /* input buf offset
9125 * always 0 buf is already offseted */
9129 fprintf(stderr
, Name
": Error restoring stripes\n");
9137 close_targets(targets
, new_disks
);
9140 free(target_offsets
);
9145 /*******************************************************************************
9146 * Function: save_checkpoint_imsm
9147 * Description: Function called for current unit status update
9148 * in the migration record. It writes it to disk.
9150 * super : imsm internal array info
9151 * info : general array info
9155 * 2: failure, means no valid migration record
9156 * / no general migration in progress /
9157 ******************************************************************************/
9158 int save_checkpoint_imsm(struct supertype
*st
, struct mdinfo
*info
, int state
)
9160 struct intel_super
*super
= st
->sb
;
9161 unsigned long long blocks_per_unit
;
9162 unsigned long long curr_migr_unit
;
9164 if (load_imsm_migr_rec(super
, info
) != 0) {
9165 dprintf("imsm: ERROR: Cannot read migration record "
9166 "for checkpoint save.\n");
9170 blocks_per_unit
= __le32_to_cpu(super
->migr_rec
->blocks_per_unit
);
9171 if (blocks_per_unit
== 0) {
9172 dprintf("imsm: no migration in progress.\n");
9175 curr_migr_unit
= info
->reshape_progress
/ blocks_per_unit
;
9176 /* check if array is alligned to copy area
9177 * if it is not alligned, add one to current migration unit value
9178 * this can happend on array reshape finish only
9180 if (info
->reshape_progress
% blocks_per_unit
)
9183 super
->migr_rec
->curr_migr_unit
=
9184 __cpu_to_le32(curr_migr_unit
);
9185 super
->migr_rec
->rec_status
= __cpu_to_le32(state
);
9186 super
->migr_rec
->dest_1st_member_lba
=
9187 __cpu_to_le32(curr_migr_unit
*
9188 __le32_to_cpu(super
->migr_rec
->dest_depth_per_unit
));
9189 if (write_imsm_migr_rec(st
) < 0) {
9190 dprintf("imsm: Cannot write migration record "
9191 "outside backup area\n");
9198 /*******************************************************************************
9199 * Function: recover_backup_imsm
9200 * Description: Function recovers critical data from the Migration Copy Area
9201 * while assembling an array.
9203 * super : imsm internal array info
9204 * info : general array info
9206 * 0 : success (or there is no data to recover)
9208 ******************************************************************************/
9209 int recover_backup_imsm(struct supertype
*st
, struct mdinfo
*info
)
9211 struct intel_super
*super
= st
->sb
;
9212 struct migr_record
*migr_rec
= super
->migr_rec
;
9213 struct imsm_map
*map_dest
= NULL
;
9214 struct intel_dev
*id
= NULL
;
9215 unsigned long long read_offset
;
9216 unsigned long long write_offset
;
9218 int *targets
= NULL
;
9219 int new_disks
, i
, err
;
9222 unsigned long curr_migr_unit
= __le32_to_cpu(migr_rec
->curr_migr_unit
);
9223 unsigned long num_migr_units
= __le32_to_cpu(migr_rec
->num_migr_units
);
9225 int skipped_disks
= 0;
9227 err
= sysfs_get_str(info
, NULL
, "array_state", (char *)buffer
, 20);
9231 /* recover data only during assemblation */
9232 if (strncmp(buffer
, "inactive", 8) != 0)
9234 /* no data to recover */
9235 if (__le32_to_cpu(migr_rec
->rec_status
) == UNIT_SRC_NORMAL
)
9237 if (curr_migr_unit
>= num_migr_units
)
9240 /* find device during reshape */
9241 for (id
= super
->devlist
; id
; id
= id
->next
)
9242 if (is_gen_migration(id
->dev
))
9247 map_dest
= get_imsm_map(id
->dev
, MAP_0
);
9248 new_disks
= map_dest
->num_members
;
9250 read_offset
= (unsigned long long)
9251 __le32_to_cpu(migr_rec
->ckpt_area_pba
) * 512;
9253 write_offset
= ((unsigned long long)
9254 __le32_to_cpu(migr_rec
->dest_1st_member_lba
) +
9255 pba_of_lba0(map_dest
)) * 512;
9257 unit_len
= __le32_to_cpu(migr_rec
->dest_depth_per_unit
) * 512;
9258 if (posix_memalign((void **)&buf
, 512, unit_len
) != 0)
9260 targets
= malloc(new_disks
* sizeof(int));
9264 if (open_backup_targets(info
, new_disks
, targets
, super
, id
->dev
)) {
9266 Name
": Cannot open some devices belonging to array.\n");
9270 for (i
= 0; i
< new_disks
; i
++) {
9271 if (targets
[i
] < 0) {
9275 if (lseek64(targets
[i
], read_offset
, SEEK_SET
) < 0) {
9277 Name
": Cannot seek to block: %s\n",
9282 if ((unsigned)read(targets
[i
], buf
, unit_len
) != unit_len
) {
9284 Name
": Cannot read copy area block: %s\n",
9289 if (lseek64(targets
[i
], write_offset
, SEEK_SET
) < 0) {
9291 Name
": Cannot seek to block: %s\n",
9296 if ((unsigned)write(targets
[i
], buf
, unit_len
) != unit_len
) {
9298 Name
": Cannot restore block: %s\n",
9305 if (skipped_disks
> imsm_get_allowed_degradation(info
->new_level
,
9310 Name
": Cannot restore data from backup."
9311 " Too many failed disks\n");
9315 if (save_checkpoint_imsm(st
, info
, UNIT_SRC_NORMAL
)) {
9316 /* ignore error == 2, this can mean end of reshape here
9318 dprintf("imsm: Cannot write checkpoint to "
9319 "migration record (UNIT_SRC_NORMAL) during restart\n");
9325 for (i
= 0; i
< new_disks
; i
++)
9334 static char disk_by_path
[] = "/dev/disk/by-path/";
9336 static const char *imsm_get_disk_controller_domain(const char *path
)
9338 char disk_path
[PATH_MAX
];
9342 strncpy(disk_path
, disk_by_path
, PATH_MAX
- 1);
9343 strncat(disk_path
, path
, PATH_MAX
- strlen(disk_path
) - 1);
9344 if (stat(disk_path
, &st
) == 0) {
9345 struct sys_dev
* hba
;
9348 path
= devt_to_devpath(st
.st_rdev
);
9351 hba
= find_disk_attached_hba(-1, path
);
9352 if (hba
&& hba
->type
== SYS_DEV_SAS
)
9354 else if (hba
&& hba
->type
== SYS_DEV_SATA
)
9358 dprintf("path: %s hba: %s attached: %s\n",
9359 path
, (hba
) ? hba
->path
: "NULL", drv
);
9367 static int imsm_find_array_minor_by_subdev(int subdev
, int container
, int *minor
)
9369 char subdev_name
[20];
9370 struct mdstat_ent
*mdstat
;
9372 sprintf(subdev_name
, "%d", subdev
);
9373 mdstat
= mdstat_by_subdev(subdev_name
, container
);
9377 *minor
= mdstat
->devnum
;
9378 free_mdstat(mdstat
);
9382 static int imsm_reshape_is_allowed_on_container(struct supertype
*st
,
9383 struct geo_params
*geo
,
9384 int *old_raid_disks
)
9386 /* currently we only support increasing the number of devices
9387 * for a container. This increases the number of device for each
9388 * member array. They must all be RAID0 or RAID5.
9391 struct mdinfo
*info
, *member
;
9392 int devices_that_can_grow
= 0;
9394 dprintf("imsm: imsm_reshape_is_allowed_on_container(ENTER): "
9395 "st->devnum = (%i)\n",
9398 if (geo
->size
!= -1 ||
9399 geo
->level
!= UnSet
||
9400 geo
->layout
!= UnSet
||
9401 geo
->chunksize
!= 0 ||
9402 geo
->raid_disks
== UnSet
) {
9403 dprintf("imsm: Container operation is allowed for "
9404 "raid disks number change only.\n");
9408 info
= container_content_imsm(st
, NULL
);
9409 for (member
= info
; member
; member
= member
->next
) {
9413 dprintf("imsm: checking device_num: %i\n",
9414 member
->container_member
);
9416 if (geo
->raid_disks
<= member
->array
.raid_disks
) {
9417 /* we work on container for Online Capacity Expansion
9418 * only so raid_disks has to grow
9420 dprintf("imsm: for container operation raid disks "
9421 "increase is required\n");
9425 if ((info
->array
.level
!= 0) &&
9426 (info
->array
.level
!= 5)) {
9427 /* we cannot use this container with other raid level
9429 dprintf("imsm: for container operation wrong"
9430 " raid level (%i) detected\n",
9434 /* check for platform support
9435 * for this raid level configuration
9437 struct intel_super
*super
= st
->sb
;
9438 if (!is_raid_level_supported(super
->orom
,
9439 member
->array
.level
,
9441 dprintf("platform does not support raid%d with"
9445 geo
->raid_disks
> 1 ? "s" : "");
9448 /* check if component size is aligned to chunk size
9450 if (info
->component_size
%
9451 (info
->array
.chunk_size
/512)) {
9452 dprintf("Component size is not aligned to "
9458 if (*old_raid_disks
&&
9459 info
->array
.raid_disks
!= *old_raid_disks
)
9461 *old_raid_disks
= info
->array
.raid_disks
;
9463 /* All raid5 and raid0 volumes in container
9464 * have to be ready for Online Capacity Expansion
9465 * so they need to be assembled. We have already
9466 * checked that no recovery etc is happening.
9468 result
= imsm_find_array_minor_by_subdev(member
->container_member
,
9472 dprintf("imsm: cannot find array\n");
9475 devices_that_can_grow
++;
9478 if (!member
&& devices_that_can_grow
)
9482 dprintf("\tContainer operation allowed\n");
9484 dprintf("\tError: %i\n", ret_val
);
9489 /* Function: get_spares_for_grow
9490 * Description: Allocates memory and creates list of spare devices
9491 * avaliable in container. Checks if spare drive size is acceptable.
9492 * Parameters: Pointer to the supertype structure
9493 * Returns: Pointer to the list of spare devices (mdinfo structure) on success,
9496 static struct mdinfo
*get_spares_for_grow(struct supertype
*st
)
9498 unsigned long long min_size
= min_acceptable_spare_size_imsm(st
);
9499 return container_choose_spares(st
, min_size
, NULL
, NULL
, NULL
, 0);
9502 /******************************************************************************
9503 * function: imsm_create_metadata_update_for_reshape
9504 * Function creates update for whole IMSM container.
9506 ******************************************************************************/
9507 static int imsm_create_metadata_update_for_reshape(
9508 struct supertype
*st
,
9509 struct geo_params
*geo
,
9511 struct imsm_update_reshape
**updatep
)
9513 struct intel_super
*super
= st
->sb
;
9514 struct imsm_super
*mpb
= super
->anchor
;
9515 int update_memory_size
= 0;
9516 struct imsm_update_reshape
*u
= NULL
;
9517 struct mdinfo
*spares
= NULL
;
9519 int delta_disks
= 0;
9522 dprintf("imsm_update_metadata_for_reshape(enter) raid_disks = %i\n",
9525 delta_disks
= geo
->raid_disks
- old_raid_disks
;
9527 /* size of all update data without anchor */
9528 update_memory_size
= sizeof(struct imsm_update_reshape
);
9530 /* now add space for spare disks that we need to add. */
9531 update_memory_size
+= sizeof(u
->new_disks
[0]) * (delta_disks
- 1);
9533 u
= calloc(1, update_memory_size
);
9536 "cannot get memory for imsm_update_reshape update\n");
9539 u
->type
= update_reshape_container_disks
;
9540 u
->old_raid_disks
= old_raid_disks
;
9541 u
->new_raid_disks
= geo
->raid_disks
;
9543 /* now get spare disks list
9545 spares
= get_spares_for_grow(st
);
9548 || delta_disks
> spares
->array
.spare_disks
) {
9549 fprintf(stderr
, Name
": imsm: ERROR: Cannot get spare devices "
9550 "for %s.\n", geo
->dev_name
);
9555 /* we have got spares
9556 * update disk list in imsm_disk list table in anchor
9558 dprintf("imsm: %i spares are available.\n\n",
9559 spares
->array
.spare_disks
);
9562 for (i
= 0; i
< delta_disks
; i
++) {
9567 u
->new_disks
[i
] = makedev(dev
->disk
.major
,
9569 dl
= get_disk_super(super
, dev
->disk
.major
, dev
->disk
.minor
);
9570 dl
->index
= mpb
->num_disks
;
9580 dprintf("imsm: reshape update preparation :");
9581 if (i
== delta_disks
) {
9584 return update_memory_size
;
9587 dprintf(" Error\n");
9592 /******************************************************************************
9593 * function: imsm_create_metadata_update_for_migration()
9594 * Creates update for IMSM array.
9596 ******************************************************************************/
9597 static int imsm_create_metadata_update_for_migration(
9598 struct supertype
*st
,
9599 struct geo_params
*geo
,
9600 struct imsm_update_reshape_migration
**updatep
)
9602 struct intel_super
*super
= st
->sb
;
9603 int update_memory_size
= 0;
9604 struct imsm_update_reshape_migration
*u
= NULL
;
9605 struct imsm_dev
*dev
;
9606 int previous_level
= -1;
9608 dprintf("imsm_create_metadata_update_for_migration(enter)"
9609 " New Level = %i\n", geo
->level
);
9611 /* size of all update data without anchor */
9612 update_memory_size
= sizeof(struct imsm_update_reshape_migration
);
9614 u
= calloc(1, update_memory_size
);
9616 dprintf("error: cannot get memory for "
9617 "imsm_create_metadata_update_for_migration\n");
9620 u
->type
= update_reshape_migration
;
9621 u
->subdev
= super
->current_vol
;
9622 u
->new_level
= geo
->level
;
9623 u
->new_layout
= geo
->layout
;
9624 u
->new_raid_disks
= u
->old_raid_disks
= geo
->raid_disks
;
9625 u
->new_disks
[0] = -1;
9626 u
->new_chunksize
= -1;
9628 dev
= get_imsm_dev(super
, u
->subdev
);
9630 struct imsm_map
*map
;
9632 map
= get_imsm_map(dev
, MAP_0
);
9634 int current_chunk_size
=
9635 __le16_to_cpu(map
->blocks_per_strip
) / 2;
9637 if (geo
->chunksize
!= current_chunk_size
) {
9638 u
->new_chunksize
= geo
->chunksize
/ 1024;
9640 "chunk size change from %i to %i\n",
9641 current_chunk_size
, u
->new_chunksize
);
9643 previous_level
= map
->raid_level
;
9646 if ((geo
->level
== 5) && (previous_level
== 0)) {
9647 struct mdinfo
*spares
= NULL
;
9649 u
->new_raid_disks
++;
9650 spares
= get_spares_for_grow(st
);
9651 if ((spares
== NULL
) || (spares
->array
.spare_disks
< 1)) {
9654 update_memory_size
= 0;
9655 dprintf("error: cannot get spare device "
9656 "for requested migration");
9661 dprintf("imsm: reshape update preparation : OK\n");
9664 return update_memory_size
;
9667 static void imsm_update_metadata_locally(struct supertype
*st
,
9670 struct metadata_update mu
;
9675 mu
.space_list
= NULL
;
9677 imsm_prepare_update(st
, &mu
);
9678 imsm_process_update(st
, &mu
);
9680 while (mu
.space_list
) {
9681 void **space
= mu
.space_list
;
9682 mu
.space_list
= *space
;
9687 /***************************************************************************
9688 * Function: imsm_analyze_change
9689 * Description: Function analyze change for single volume
9690 * and validate if transition is supported
9691 * Parameters: Geometry parameters, supertype structure
9692 * Returns: Operation type code on success, -1 if fail
9693 ****************************************************************************/
9694 enum imsm_reshape_type
imsm_analyze_change(struct supertype
*st
,
9695 struct geo_params
*geo
)
9701 /* number of added/removed disks in operation result */
9702 int devNumChange
= 0;
9703 /* imsm compatible layout value for array geometry verification */
9704 int imsm_layout
= -1;
9706 getinfo_super_imsm_volume(st
, &info
, NULL
);
9707 if ((geo
->level
!= info
.array
.level
) &&
9708 (geo
->level
>= 0) &&
9709 (geo
->level
!= UnSet
)) {
9710 switch (info
.array
.level
) {
9712 if (geo
->level
== 5) {
9713 change
= CH_MIGRATION
;
9714 if (geo
->layout
!= ALGORITHM_LEFT_ASYMMETRIC
) {
9716 Name
" Error. Requested Layout "
9717 "not supported (left-asymmetric layout "
9718 "is supported only)!\n");
9720 goto analyse_change_exit
;
9722 imsm_layout
= geo
->layout
;
9724 devNumChange
= 1; /* parity disk added */
9725 } else if (geo
->level
== 10) {
9726 change
= CH_TAKEOVER
;
9728 devNumChange
= 2; /* two mirrors added */
9729 imsm_layout
= 0x102; /* imsm supported layout */
9734 if (geo
->level
== 0) {
9735 change
= CH_TAKEOVER
;
9737 devNumChange
= -(geo
->raid_disks
/2);
9738 imsm_layout
= 0; /* imsm raid0 layout */
9744 Name
" Error. Level Migration from %d to %d "
9746 info
.array
.level
, geo
->level
);
9747 goto analyse_change_exit
;
9750 geo
->level
= info
.array
.level
;
9752 if ((geo
->layout
!= info
.array
.layout
)
9753 && ((geo
->layout
!= UnSet
) && (geo
->layout
!= -1))) {
9754 change
= CH_MIGRATION
;
9755 if ((info
.array
.layout
== 0)
9756 && (info
.array
.level
== 5)
9757 && (geo
->layout
== 5)) {
9758 /* reshape 5 -> 4 */
9759 } else if ((info
.array
.layout
== 5)
9760 && (info
.array
.level
== 5)
9761 && (geo
->layout
== 0)) {
9762 /* reshape 4 -> 5 */
9767 Name
" Error. Layout Migration from %d to %d "
9769 info
.array
.layout
, geo
->layout
);
9771 goto analyse_change_exit
;
9774 geo
->layout
= info
.array
.layout
;
9775 if (imsm_layout
== -1)
9776 imsm_layout
= info
.array
.layout
;
9779 if ((geo
->chunksize
> 0) && (geo
->chunksize
!= UnSet
)
9780 && (geo
->chunksize
!= info
.array
.chunk_size
))
9781 change
= CH_MIGRATION
;
9783 geo
->chunksize
= info
.array
.chunk_size
;
9785 chunk
= geo
->chunksize
/ 1024;
9786 if (!validate_geometry_imsm(st
,
9789 geo
->raid_disks
+ devNumChange
,
9796 struct intel_super
*super
= st
->sb
;
9797 struct imsm_super
*mpb
= super
->anchor
;
9799 if (mpb
->num_raid_devs
> 1) {
9801 Name
" Error. Cannot perform operation on %s"
9802 "- for this operation it MUST be single "
9803 "array in container\n",
9809 analyse_change_exit
:
9814 int imsm_takeover(struct supertype
*st
, struct geo_params
*geo
)
9816 struct intel_super
*super
= st
->sb
;
9817 struct imsm_update_takeover
*u
;
9819 u
= malloc(sizeof(struct imsm_update_takeover
));
9823 u
->type
= update_takeover
;
9824 u
->subarray
= super
->current_vol
;
9826 /* 10->0 transition */
9827 if (geo
->level
== 0)
9828 u
->direction
= R10_TO_R0
;
9830 /* 0->10 transition */
9831 if (geo
->level
== 10)
9832 u
->direction
= R0_TO_R10
;
9834 /* update metadata locally */
9835 imsm_update_metadata_locally(st
, u
,
9836 sizeof(struct imsm_update_takeover
));
9837 /* and possibly remotely */
9838 if (st
->update_tail
)
9839 append_metadata_update(st
, u
,
9840 sizeof(struct imsm_update_takeover
));
9847 static int imsm_reshape_super(struct supertype
*st
, long long size
, int level
,
9848 int layout
, int chunksize
, int raid_disks
,
9849 int delta_disks
, char *backup
, char *dev
,
9853 struct geo_params geo
;
9855 dprintf("imsm: reshape_super called.\n");
9857 memset(&geo
, 0, sizeof(struct geo_params
));
9860 geo
.dev_id
= st
->devnum
;
9863 geo
.layout
= layout
;
9864 geo
.chunksize
= chunksize
;
9865 geo
.raid_disks
= raid_disks
;
9866 if (delta_disks
!= UnSet
)
9867 geo
.raid_disks
+= delta_disks
;
9869 dprintf("\tfor level : %i\n", geo
.level
);
9870 dprintf("\tfor raid_disks : %i\n", geo
.raid_disks
);
9872 if (experimental() == 0)
9875 if (st
->container_dev
== st
->devnum
) {
9876 /* On container level we can only increase number of devices. */
9877 dprintf("imsm: info: Container operation\n");
9878 int old_raid_disks
= 0;
9880 if (imsm_reshape_is_allowed_on_container(
9881 st
, &geo
, &old_raid_disks
)) {
9882 struct imsm_update_reshape
*u
= NULL
;
9885 len
= imsm_create_metadata_update_for_reshape(
9886 st
, &geo
, old_raid_disks
, &u
);
9889 dprintf("imsm: Cannot prepare update\n");
9890 goto exit_imsm_reshape_super
;
9894 /* update metadata locally */
9895 imsm_update_metadata_locally(st
, u
, len
);
9896 /* and possibly remotely */
9897 if (st
->update_tail
)
9898 append_metadata_update(st
, u
, len
);
9903 fprintf(stderr
, Name
": (imsm) Operation "
9904 "is not allowed on this container\n");
9907 /* On volume level we support following operations
9908 * - takeover: raid10 -> raid0; raid0 -> raid10
9909 * - chunk size migration
9910 * - migration: raid5 -> raid0; raid0 -> raid5
9912 struct intel_super
*super
= st
->sb
;
9913 struct intel_dev
*dev
= super
->devlist
;
9915 dprintf("imsm: info: Volume operation\n");
9916 /* find requested device */
9918 if (imsm_find_array_minor_by_subdev(
9919 dev
->index
, st
->container_dev
, &devnum
) == 0
9920 && devnum
== geo
.dev_id
)
9925 fprintf(stderr
, Name
" Cannot find %s (%i) subarray\n",
9926 geo
.dev_name
, geo
.dev_id
);
9927 goto exit_imsm_reshape_super
;
9929 super
->current_vol
= dev
->index
;
9930 change
= imsm_analyze_change(st
, &geo
);
9933 ret_val
= imsm_takeover(st
, &geo
);
9935 case CH_MIGRATION
: {
9936 struct imsm_update_reshape_migration
*u
= NULL
;
9938 imsm_create_metadata_update_for_migration(
9942 "Cannot prepare update\n");
9946 /* update metadata locally */
9947 imsm_update_metadata_locally(st
, u
, len
);
9948 /* and possibly remotely */
9949 if (st
->update_tail
)
9950 append_metadata_update(st
, u
, len
);
9960 exit_imsm_reshape_super
:
9961 dprintf("imsm: reshape_super Exit code = %i\n", ret_val
);
9965 /*******************************************************************************
9966 * Function: wait_for_reshape_imsm
9967 * Description: Function writes new sync_max value and waits until
9968 * reshape process reach new position
9970 * sra : general array info
9971 * ndata : number of disks in new array's layout
9974 * 1 : there is no reshape in progress,
9976 ******************************************************************************/
9977 int wait_for_reshape_imsm(struct mdinfo
*sra
, int ndata
)
9979 int fd
= sysfs_get_fd(sra
, NULL
, "reshape_position");
9980 unsigned long long completed
;
9981 /* to_complete : new sync_max position */
9982 unsigned long long to_complete
= sra
->reshape_progress
;
9983 unsigned long long position_to_set
= to_complete
/ ndata
;
9986 dprintf("imsm: wait_for_reshape_imsm() "
9987 "cannot open reshape_position\n");
9991 if (sysfs_fd_get_ll(fd
, &completed
) < 0) {
9992 dprintf("imsm: wait_for_reshape_imsm() "
9993 "cannot read reshape_position (no reshape in progres)\n");
9998 if (completed
> to_complete
) {
9999 dprintf("imsm: wait_for_reshape_imsm() "
10000 "wrong next position to set %llu (%llu)\n",
10001 to_complete
, completed
);
10005 dprintf("Position set: %llu\n", position_to_set
);
10006 if (sysfs_set_num(sra
, NULL
, "sync_max",
10007 position_to_set
) != 0) {
10008 dprintf("imsm: wait_for_reshape_imsm() "
10009 "cannot set reshape position to %llu\n",
10020 select(fd
+1, &rfds
, NULL
, NULL
, NULL
);
10021 if (sysfs_get_str(sra
, NULL
, "sync_action",
10023 strncmp(action
, "reshape", 7) != 0)
10025 if (sysfs_fd_get_ll(fd
, &completed
) < 0) {
10026 dprintf("imsm: wait_for_reshape_imsm() "
10027 "cannot read reshape_position (in loop)\n");
10031 } while (completed
< to_complete
);
10037 /*******************************************************************************
10038 * Function: check_degradation_change
10039 * Description: Check that array hasn't become failed.
10041 * info : for sysfs access
10042 * sources : source disks descriptors
10043 * degraded: previous degradation level
10045 * degradation level
10046 ******************************************************************************/
10047 int check_degradation_change(struct mdinfo
*info
,
10051 unsigned long long new_degraded
;
10052 sysfs_get_ll(info
, NULL
, "degraded", &new_degraded
);
10053 if (new_degraded
!= (unsigned long long)degraded
) {
10054 /* check each device to ensure it is still working */
10057 for (sd
= info
->devs
; sd
; sd
= sd
->next
) {
10058 if (sd
->disk
.state
& (1<<MD_DISK_FAULTY
))
10060 if (sd
->disk
.state
& (1<<MD_DISK_SYNC
)) {
10062 if (sysfs_get_str(info
,
10063 sd
, "state", sbuf
, 20) < 0 ||
10064 strstr(sbuf
, "faulty") ||
10065 strstr(sbuf
, "in_sync") == NULL
) {
10066 /* this device is dead */
10067 sd
->disk
.state
= (1<<MD_DISK_FAULTY
);
10068 if (sd
->disk
.raid_disk
>= 0 &&
10069 sources
[sd
->disk
.raid_disk
] >= 0) {
10071 sd
->disk
.raid_disk
]);
10072 sources
[sd
->disk
.raid_disk
] =
10081 return new_degraded
;
10084 /*******************************************************************************
10085 * Function: imsm_manage_reshape
10086 * Description: Function finds array under reshape and it manages reshape
10087 * process. It creates stripes backups (if required) and sets
10090 * afd : Backup handle (nattive) - not used
10091 * sra : general array info
10092 * reshape : reshape parameters - not used
10093 * st : supertype structure
10094 * blocks : size of critical section [blocks]
10095 * fds : table of source device descriptor
10096 * offsets : start of array (offest per devices)
10098 * destfd : table of destination device descriptor
10099 * destoffsets : table of destination offsets (per device)
10101 * 1 : success, reshape is done
10103 ******************************************************************************/
10104 static int imsm_manage_reshape(
10105 int afd
, struct mdinfo
*sra
, struct reshape
*reshape
,
10106 struct supertype
*st
, unsigned long backup_blocks
,
10107 int *fds
, unsigned long long *offsets
,
10108 int dests
, int *destfd
, unsigned long long *destoffsets
)
10111 struct intel_super
*super
= st
->sb
;
10112 struct intel_dev
*dv
= NULL
;
10113 struct imsm_dev
*dev
= NULL
;
10114 struct imsm_map
*map_src
;
10115 int migr_vol_qan
= 0;
10116 int ndata
, odata
; /* [bytes] */
10117 int chunk
; /* [bytes] */
10118 struct migr_record
*migr_rec
;
10120 unsigned int buf_size
; /* [bytes] */
10121 unsigned long long max_position
; /* array size [bytes] */
10122 unsigned long long next_step
; /* [blocks]/[bytes] */
10123 unsigned long long old_data_stripe_length
;
10124 unsigned long long start_src
; /* [bytes] */
10125 unsigned long long start
; /* [bytes] */
10126 unsigned long long start_buf_shift
; /* [bytes] */
10128 int source_layout
= 0;
10130 if (!fds
|| !offsets
|| !sra
)
10133 /* Find volume during the reshape */
10134 for (dv
= super
->devlist
; dv
; dv
= dv
->next
) {
10135 if (dv
->dev
->vol
.migr_type
== MIGR_GEN_MIGR
10136 && dv
->dev
->vol
.migr_state
== 1) {
10141 /* Only one volume can migrate at the same time */
10142 if (migr_vol_qan
!= 1) {
10143 fprintf(stderr
, Name
" : %s", migr_vol_qan
?
10144 "Number of migrating volumes greater than 1\n" :
10145 "There is no volume during migrationg\n");
10149 map_src
= get_imsm_map(dev
, MAP_1
);
10150 if (map_src
== NULL
)
10153 ndata
= imsm_num_data_members(dev
, MAP_0
);
10154 odata
= imsm_num_data_members(dev
, MAP_1
);
10156 chunk
= __le16_to_cpu(map_src
->blocks_per_strip
) * 512;
10157 old_data_stripe_length
= odata
* chunk
;
10159 migr_rec
= super
->migr_rec
;
10161 /* initialize migration record for start condition */
10162 if (sra
->reshape_progress
== 0)
10163 init_migr_record_imsm(st
, dev
, sra
);
10165 if (__le32_to_cpu(migr_rec
->rec_status
) != UNIT_SRC_NORMAL
) {
10166 dprintf("imsm: cannot restart migration when data "
10167 "are present in copy area.\n");
10170 /* Save checkpoint to update migration record for current
10171 * reshape position (in md). It can be farther than current
10172 * reshape position in metadata.
10174 if (save_checkpoint_imsm(st
, sra
, UNIT_SRC_NORMAL
) == 1) {
10175 /* ignore error == 2, this can mean end of reshape here
10177 dprintf("imsm: Cannot write checkpoint to "
10178 "migration record (UNIT_SRC_NORMAL, "
10179 "initial save)\n");
10184 /* size for data */
10185 buf_size
= __le32_to_cpu(migr_rec
->blocks_per_unit
) * 512;
10186 /* extend buffer size for parity disk */
10187 buf_size
+= __le32_to_cpu(migr_rec
->dest_depth_per_unit
) * 512;
10188 /* add space for stripe aligment */
10189 buf_size
+= old_data_stripe_length
;
10190 if (posix_memalign((void **)&buf
, 4096, buf_size
)) {
10191 dprintf("imsm: Cannot allocate checpoint buffer\n");
10195 max_position
= sra
->component_size
* ndata
;
10196 source_layout
= imsm_level_to_layout(map_src
->raid_level
);
10198 while (__le32_to_cpu(migr_rec
->curr_migr_unit
) <
10199 __le32_to_cpu(migr_rec
->num_migr_units
)) {
10200 /* current reshape position [blocks] */
10201 unsigned long long current_position
=
10202 __le32_to_cpu(migr_rec
->blocks_per_unit
)
10203 * __le32_to_cpu(migr_rec
->curr_migr_unit
);
10204 unsigned long long border
;
10206 /* Check that array hasn't become failed.
10208 degraded
= check_degradation_change(sra
, fds
, degraded
);
10209 if (degraded
> 1) {
10210 dprintf("imsm: Abort reshape due to degradation"
10211 " level (%i)\n", degraded
);
10215 next_step
= __le32_to_cpu(migr_rec
->blocks_per_unit
);
10217 if ((current_position
+ next_step
) > max_position
)
10218 next_step
= max_position
- current_position
;
10220 start
= current_position
* 512;
10222 /* allign reading start to old geometry */
10223 start_buf_shift
= start
% old_data_stripe_length
;
10224 start_src
= start
- start_buf_shift
;
10226 border
= (start_src
/ odata
) - (start
/ ndata
);
10228 if (border
<= __le32_to_cpu(migr_rec
->dest_depth_per_unit
)) {
10229 /* save critical stripes to buf
10230 * start - start address of current unit
10231 * to backup [bytes]
10232 * start_src - start address of current unit
10233 * to backup alligned to source array
10236 unsigned long long next_step_filler
= 0;
10237 unsigned long long copy_length
= next_step
* 512;
10239 /* allign copy area length to stripe in old geometry */
10240 next_step_filler
= ((copy_length
+ start_buf_shift
)
10241 % old_data_stripe_length
);
10242 if (next_step_filler
)
10243 next_step_filler
= (old_data_stripe_length
10244 - next_step_filler
);
10245 dprintf("save_stripes() parameters: start = %llu,"
10246 "\tstart_src = %llu,\tnext_step*512 = %llu,"
10247 "\tstart_in_buf_shift = %llu,"
10248 "\tnext_step_filler = %llu\n",
10249 start
, start_src
, copy_length
,
10250 start_buf_shift
, next_step_filler
);
10252 if (save_stripes(fds
, offsets
, map_src
->num_members
,
10253 chunk
, map_src
->raid_level
,
10254 source_layout
, 0, NULL
, start_src
,
10256 next_step_filler
+ start_buf_shift
,
10258 dprintf("imsm: Cannot save stripes"
10262 /* Convert data to destination format and store it
10263 * in backup general migration area
10265 if (save_backup_imsm(st
, dev
, sra
,
10266 buf
+ start_buf_shift
, copy_length
)) {
10267 dprintf("imsm: Cannot save stripes to "
10268 "target devices\n");
10271 if (save_checkpoint_imsm(st
, sra
,
10272 UNIT_SRC_IN_CP_AREA
)) {
10273 dprintf("imsm: Cannot write checkpoint to "
10274 "migration record (UNIT_SRC_IN_CP_AREA)\n");
10278 /* set next step to use whole border area */
10279 border
/= next_step
;
10281 next_step
*= border
;
10283 /* When data backed up, checkpoint stored,
10284 * kick the kernel to reshape unit of data
10286 next_step
= next_step
+ sra
->reshape_progress
;
10287 /* limit next step to array max position */
10288 if (next_step
> max_position
)
10289 next_step
= max_position
;
10290 sysfs_set_num(sra
, NULL
, "suspend_lo", sra
->reshape_progress
);
10291 sysfs_set_num(sra
, NULL
, "suspend_hi", next_step
);
10292 sra
->reshape_progress
= next_step
;
10294 /* wait until reshape finish */
10295 if (wait_for_reshape_imsm(sra
, ndata
) < 0) {
10296 dprintf("wait_for_reshape_imsm returned error!\n");
10300 if (save_checkpoint_imsm(st
, sra
, UNIT_SRC_NORMAL
) == 1) {
10301 /* ignore error == 2, this can mean end of reshape here
10303 dprintf("imsm: Cannot write checkpoint to "
10304 "migration record (UNIT_SRC_NORMAL)\n");
10310 /* return '1' if done */
10314 abort_reshape(sra
);
10318 #endif /* MDASSEMBLE */
10320 struct superswitch super_imsm
= {
10322 .examine_super
= examine_super_imsm
,
10323 .brief_examine_super
= brief_examine_super_imsm
,
10324 .brief_examine_subarrays
= brief_examine_subarrays_imsm
,
10325 .export_examine_super
= export_examine_super_imsm
,
10326 .detail_super
= detail_super_imsm
,
10327 .brief_detail_super
= brief_detail_super_imsm
,
10328 .write_init_super
= write_init_super_imsm
,
10329 .validate_geometry
= validate_geometry_imsm
,
10330 .add_to_super
= add_to_super_imsm
,
10331 .remove_from_super
= remove_from_super_imsm
,
10332 .detail_platform
= detail_platform_imsm
,
10333 .kill_subarray
= kill_subarray_imsm
,
10334 .update_subarray
= update_subarray_imsm
,
10335 .load_container
= load_container_imsm
,
10336 .default_geometry
= default_geometry_imsm
,
10337 .get_disk_controller_domain
= imsm_get_disk_controller_domain
,
10338 .reshape_super
= imsm_reshape_super
,
10339 .manage_reshape
= imsm_manage_reshape
,
10340 .recover_backup
= recover_backup_imsm
,
10342 .match_home
= match_home_imsm
,
10343 .uuid_from_super
= uuid_from_super_imsm
,
10344 .getinfo_super
= getinfo_super_imsm
,
10345 .getinfo_super_disks
= getinfo_super_disks_imsm
,
10346 .update_super
= update_super_imsm
,
10348 .avail_size
= avail_size_imsm
,
10349 .min_acceptable_spare_size
= min_acceptable_spare_size_imsm
,
10351 .compare_super
= compare_super_imsm
,
10353 .load_super
= load_super_imsm
,
10354 .init_super
= init_super_imsm
,
10355 .store_super
= store_super_imsm
,
10356 .free_super
= free_super_imsm
,
10357 .match_metadata_desc
= match_metadata_desc_imsm
,
10358 .container_content
= container_content_imsm
,
10366 .open_new
= imsm_open_new
,
10367 .set_array_state
= imsm_set_array_state
,
10368 .set_disk
= imsm_set_disk
,
10369 .sync_metadata
= imsm_sync_metadata
,
10370 .activate_spare
= imsm_activate_spare
,
10371 .process_update
= imsm_process_update
,
10372 .prepare_update
= imsm_prepare_update
,
10373 #endif /* MDASSEMBLE */