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
);
3594 load_and_parse_mpb(int fd
, struct intel_super
*super
, char *devname
, int keep_fd
)
3598 err
= load_imsm_mpb(fd
, super
, devname
);
3601 err
= load_imsm_disk(fd
, super
, devname
, keep_fd
);
3604 err
= parse_raid_devices(super
);
3609 static void __free_imsm_disk(struct dl
*d
)
3621 static void free_imsm_disks(struct intel_super
*super
)
3625 while (super
->disks
) {
3627 super
->disks
= d
->next
;
3628 __free_imsm_disk(d
);
3630 while (super
->disk_mgmt_list
) {
3631 d
= super
->disk_mgmt_list
;
3632 super
->disk_mgmt_list
= d
->next
;
3633 __free_imsm_disk(d
);
3635 while (super
->missing
) {
3637 super
->missing
= d
->next
;
3638 __free_imsm_disk(d
);
3643 /* free all the pieces hanging off of a super pointer */
3644 static void __free_imsm(struct intel_super
*super
, int free_disks
)
3646 struct intel_hba
*elem
, *next
;
3652 /* unlink capability description */
3654 if (super
->migr_rec_buf
) {
3655 free(super
->migr_rec_buf
);
3656 super
->migr_rec_buf
= NULL
;
3659 free_imsm_disks(super
);
3660 free_devlist(super
);
3664 free((void *)elem
->path
);
3672 static void free_imsm(struct intel_super
*super
)
3674 __free_imsm(super
, 1);
3678 static void free_super_imsm(struct supertype
*st
)
3680 struct intel_super
*super
= st
->sb
;
3689 static struct intel_super
*alloc_super(void)
3691 struct intel_super
*super
= malloc(sizeof(*super
));
3694 memset(super
, 0, sizeof(*super
));
3695 super
->current_vol
= -1;
3696 super
->create_offset
= ~((unsigned long long) 0);
3702 * find and allocate hba and OROM/EFI based on valid fd of RAID component device
3704 static int find_intel_hba_capability(int fd
, struct intel_super
*super
, char *devname
)
3706 struct sys_dev
*hba_name
;
3709 if ((fd
< 0) || check_env("IMSM_NO_PLATFORM")) {
3714 hba_name
= find_disk_attached_hba(fd
, NULL
);
3718 Name
": %s is not attached to Intel(R) RAID controller.\n",
3722 rv
= attach_hba_to_super(super
, hba_name
);
3725 struct intel_hba
*hba
= super
->hba
;
3727 fprintf(stderr
, Name
": %s is attached to Intel(R) %s RAID "
3728 "controller (%s),\n"
3729 " but the container is assigned to Intel(R) "
3730 "%s RAID controller (",
3733 hba_name
->pci_id
? : "Err!",
3734 get_sys_dev_type(hba_name
->type
));
3737 fprintf(stderr
, "%s", hba
->pci_id
? : "Err!");
3739 fprintf(stderr
, ", ");
3743 fprintf(stderr
, ").\n"
3744 " Mixing devices attached to different controllers "
3745 "is not allowed.\n");
3747 free_sys_dev(&hba_name
);
3750 super
->orom
= find_imsm_capability(hba_name
->type
);
3751 free_sys_dev(&hba_name
);
3757 /* find_missing - helper routine for load_super_imsm_all that identifies
3758 * disks that have disappeared from the system. This routine relies on
3759 * the mpb being uptodate, which it is at load time.
3761 static int find_missing(struct intel_super
*super
)
3764 struct imsm_super
*mpb
= super
->anchor
;
3766 struct imsm_disk
*disk
;
3768 for (i
= 0; i
< mpb
->num_disks
; i
++) {
3769 disk
= __get_imsm_disk(mpb
, i
);
3770 dl
= serial_to_dl(disk
->serial
, super
);
3774 dl
= malloc(sizeof(*dl
));
3780 dl
->devname
= strdup("missing");
3782 serialcpy(dl
->serial
, disk
->serial
);
3785 dl
->next
= super
->missing
;
3786 super
->missing
= dl
;
3793 static struct intel_disk
*disk_list_get(__u8
*serial
, struct intel_disk
*disk_list
)
3795 struct intel_disk
*idisk
= disk_list
;
3798 if (serialcmp(idisk
->disk
.serial
, serial
) == 0)
3800 idisk
= idisk
->next
;
3806 static int __prep_thunderdome(struct intel_super
**table
, int tbl_size
,
3807 struct intel_super
*super
,
3808 struct intel_disk
**disk_list
)
3810 struct imsm_disk
*d
= &super
->disks
->disk
;
3811 struct imsm_super
*mpb
= super
->anchor
;
3814 for (i
= 0; i
< tbl_size
; i
++) {
3815 struct imsm_super
*tbl_mpb
= table
[i
]->anchor
;
3816 struct imsm_disk
*tbl_d
= &table
[i
]->disks
->disk
;
3818 if (tbl_mpb
->family_num
== mpb
->family_num
) {
3819 if (tbl_mpb
->check_sum
== mpb
->check_sum
) {
3820 dprintf("%s: mpb from %d:%d matches %d:%d\n",
3821 __func__
, super
->disks
->major
,
3822 super
->disks
->minor
,
3823 table
[i
]->disks
->major
,
3824 table
[i
]->disks
->minor
);
3828 if (((is_configured(d
) && !is_configured(tbl_d
)) ||
3829 is_configured(d
) == is_configured(tbl_d
)) &&
3830 tbl_mpb
->generation_num
< mpb
->generation_num
) {
3831 /* current version of the mpb is a
3832 * better candidate than the one in
3833 * super_table, but copy over "cross
3834 * generational" status
3836 struct intel_disk
*idisk
;
3838 dprintf("%s: mpb from %d:%d replaces %d:%d\n",
3839 __func__
, super
->disks
->major
,
3840 super
->disks
->minor
,
3841 table
[i
]->disks
->major
,
3842 table
[i
]->disks
->minor
);
3844 idisk
= disk_list_get(tbl_d
->serial
, *disk_list
);
3845 if (idisk
&& is_failed(&idisk
->disk
))
3846 tbl_d
->status
|= FAILED_DISK
;
3849 struct intel_disk
*idisk
;
3850 struct imsm_disk
*disk
;
3852 /* tbl_mpb is more up to date, but copy
3853 * over cross generational status before
3856 disk
= __serial_to_disk(d
->serial
, mpb
, NULL
);
3857 if (disk
&& is_failed(disk
))
3858 d
->status
|= FAILED_DISK
;
3860 idisk
= disk_list_get(d
->serial
, *disk_list
);
3863 if (disk
&& is_configured(disk
))
3864 idisk
->disk
.status
|= CONFIGURED_DISK
;
3867 dprintf("%s: mpb from %d:%d prefer %d:%d\n",
3868 __func__
, super
->disks
->major
,
3869 super
->disks
->minor
,
3870 table
[i
]->disks
->major
,
3871 table
[i
]->disks
->minor
);
3879 table
[tbl_size
++] = super
;
3883 /* update/extend the merged list of imsm_disk records */
3884 for (j
= 0; j
< mpb
->num_disks
; j
++) {
3885 struct imsm_disk
*disk
= __get_imsm_disk(mpb
, j
);
3886 struct intel_disk
*idisk
;
3888 idisk
= disk_list_get(disk
->serial
, *disk_list
);
3890 idisk
->disk
.status
|= disk
->status
;
3891 if (is_configured(&idisk
->disk
) ||
3892 is_failed(&idisk
->disk
))
3893 idisk
->disk
.status
&= ~(SPARE_DISK
);
3895 idisk
= calloc(1, sizeof(*idisk
));
3898 idisk
->owner
= IMSM_UNKNOWN_OWNER
;
3899 idisk
->disk
= *disk
;
3900 idisk
->next
= *disk_list
;
3904 if (serialcmp(idisk
->disk
.serial
, d
->serial
) == 0)
3911 static struct intel_super
*
3912 validate_members(struct intel_super
*super
, struct intel_disk
*disk_list
,
3915 struct imsm_super
*mpb
= super
->anchor
;
3919 for (i
= 0; i
< mpb
->num_disks
; i
++) {
3920 struct imsm_disk
*disk
= __get_imsm_disk(mpb
, i
);
3921 struct intel_disk
*idisk
;
3923 idisk
= disk_list_get(disk
->serial
, disk_list
);
3925 if (idisk
->owner
== owner
||
3926 idisk
->owner
== IMSM_UNKNOWN_OWNER
)
3929 dprintf("%s: '%.16s' owner %d != %d\n",
3930 __func__
, disk
->serial
, idisk
->owner
,
3933 dprintf("%s: unknown disk %x [%d]: %.16s\n",
3934 __func__
, __le32_to_cpu(mpb
->family_num
), i
,
3940 if (ok_count
== mpb
->num_disks
)
3945 static void show_conflicts(__u32 family_num
, struct intel_super
*super_list
)
3947 struct intel_super
*s
;
3949 for (s
= super_list
; s
; s
= s
->next
) {
3950 if (family_num
!= s
->anchor
->family_num
)
3952 fprintf(stderr
, "Conflict, offlining family %#x on '%s'\n",
3953 __le32_to_cpu(family_num
), s
->disks
->devname
);
3957 static struct intel_super
*
3958 imsm_thunderdome(struct intel_super
**super_list
, int len
)
3960 struct intel_super
*super_table
[len
];
3961 struct intel_disk
*disk_list
= NULL
;
3962 struct intel_super
*champion
, *spare
;
3963 struct intel_super
*s
, **del
;
3968 memset(super_table
, 0, sizeof(super_table
));
3969 for (s
= *super_list
; s
; s
= s
->next
)
3970 tbl_size
= __prep_thunderdome(super_table
, tbl_size
, s
, &disk_list
);
3972 for (i
= 0; i
< tbl_size
; i
++) {
3973 struct imsm_disk
*d
;
3974 struct intel_disk
*idisk
;
3975 struct imsm_super
*mpb
= super_table
[i
]->anchor
;
3978 d
= &s
->disks
->disk
;
3980 /* 'd' must appear in merged disk list for its
3981 * configuration to be valid
3983 idisk
= disk_list_get(d
->serial
, disk_list
);
3984 if (idisk
&& idisk
->owner
== i
)
3985 s
= validate_members(s
, disk_list
, i
);
3990 dprintf("%s: marking family: %#x from %d:%d offline\n",
3991 __func__
, mpb
->family_num
,
3992 super_table
[i
]->disks
->major
,
3993 super_table
[i
]->disks
->minor
);
3997 /* This is where the mdadm implementation differs from the Windows
3998 * driver which has no strict concept of a container. We can only
3999 * assemble one family from a container, so when returning a prodigal
4000 * array member to this system the code will not be able to disambiguate
4001 * the container contents that should be assembled ("foreign" versus
4002 * "local"). It requires user intervention to set the orig_family_num
4003 * to a new value to establish a new container. The Windows driver in
4004 * this situation fixes up the volume name in place and manages the
4005 * foreign array as an independent entity.
4010 for (i
= 0; i
< tbl_size
; i
++) {
4011 struct intel_super
*tbl_ent
= super_table
[i
];
4017 if (tbl_ent
->anchor
->num_raid_devs
== 0) {
4022 if (s
&& !is_spare
) {
4023 show_conflicts(tbl_ent
->anchor
->family_num
, *super_list
);
4025 } else if (!s
&& !is_spare
)
4038 fprintf(stderr
, "Chose family %#x on '%s', "
4039 "assemble conflicts to new container with '--update=uuid'\n",
4040 __le32_to_cpu(s
->anchor
->family_num
), s
->disks
->devname
);
4042 /* collect all dl's onto 'champion', and update them to
4043 * champion's version of the status
4045 for (s
= *super_list
; s
; s
= s
->next
) {
4046 struct imsm_super
*mpb
= champion
->anchor
;
4047 struct dl
*dl
= s
->disks
;
4052 for (i
= 0; i
< mpb
->num_disks
; i
++) {
4053 struct imsm_disk
*disk
;
4055 disk
= __serial_to_disk(dl
->serial
, mpb
, &dl
->index
);
4058 /* only set index on disks that are a member of
4059 * a populated contianer, i.e. one with
4062 if (is_failed(&dl
->disk
))
4064 else if (is_spare(&dl
->disk
))
4070 if (i
>= mpb
->num_disks
) {
4071 struct intel_disk
*idisk
;
4073 idisk
= disk_list_get(dl
->serial
, disk_list
);
4074 if (idisk
&& is_spare(&idisk
->disk
) &&
4075 !is_failed(&idisk
->disk
) && !is_configured(&idisk
->disk
))
4083 dl
->next
= champion
->disks
;
4084 champion
->disks
= dl
;
4088 /* delete 'champion' from super_list */
4089 for (del
= super_list
; *del
; ) {
4090 if (*del
== champion
) {
4091 *del
= (*del
)->next
;
4094 del
= &(*del
)->next
;
4096 champion
->next
= NULL
;
4100 struct intel_disk
*idisk
= disk_list
;
4102 disk_list
= disk_list
->next
;
4111 get_sra_super_block(int fd
, struct intel_super
**super_list
, char *devname
, int *max
, int keep_fd
);
4112 static int get_super_block(struct intel_super
**super_list
, int devnum
, char *devname
,
4113 int major
, int minor
, int keep_fd
);
4115 get_devlist_super_block(struct md_list
*devlist
, struct intel_super
**super_list
,
4116 int *max
, int keep_fd
);
4119 static int load_super_imsm_all(struct supertype
*st
, int fd
, void **sbp
,
4120 char *devname
, struct md_list
*devlist
,
4123 struct intel_super
*super_list
= NULL
;
4124 struct intel_super
*super
= NULL
;
4129 /* 'fd' is an opened container */
4130 err
= get_sra_super_block(fd
, &super_list
, devname
, &i
, keep_fd
);
4132 /* get super block from devlist devices */
4133 err
= get_devlist_super_block(devlist
, &super_list
, &i
, keep_fd
);
4136 /* all mpbs enter, maybe one leaves */
4137 super
= imsm_thunderdome(&super_list
, i
);
4143 if (find_missing(super
) != 0) {
4149 /* load migration record */
4150 err
= load_imsm_migr_rec(super
, NULL
);
4152 /* migration is in progress,
4153 * but migr_rec cannot be loaded,
4159 /* Check migration compatibility */
4160 if ((err
== 0) && (check_mpb_migr_compatibility(super
) != 0)) {
4161 fprintf(stderr
, Name
": Unsupported migration detected");
4163 fprintf(stderr
, " on %s\n", devname
);
4165 fprintf(stderr
, " (IMSM).\n");
4174 while (super_list
) {
4175 struct intel_super
*s
= super_list
;
4177 super_list
= super_list
->next
;
4187 st
->container_dev
= fd2devnum(fd
);
4189 st
->container_dev
= NoMdDev
;
4190 if (err
== 0 && st
->ss
== NULL
) {
4191 st
->ss
= &super_imsm
;
4192 st
->minor_version
= 0;
4193 st
->max_devs
= IMSM_MAX_DEVICES
;
4200 get_devlist_super_block(struct md_list
*devlist
, struct intel_super
**super_list
,
4201 int *max
, int keep_fd
)
4203 struct md_list
*tmpdev
;
4207 for (i
= 0, tmpdev
= devlist
; tmpdev
; tmpdev
= tmpdev
->next
) {
4208 if (tmpdev
->used
!= 1)
4210 if (tmpdev
->container
== 1) {
4212 int fd
= dev_open(tmpdev
->devname
, O_RDONLY
|O_EXCL
);
4214 fprintf(stderr
, Name
": cannot open device %s: %s\n",
4215 tmpdev
->devname
, strerror(errno
));
4219 err
= get_sra_super_block(fd
, super_list
,
4220 tmpdev
->devname
, &lmax
,
4229 int major
= major(tmpdev
->st_rdev
);
4230 int minor
= minor(tmpdev
->st_rdev
);
4231 err
= get_super_block(super_list
,
4248 static int get_super_block(struct intel_super
**super_list
, int devnum
, char *devname
,
4249 int major
, int minor
, int keep_fd
)
4251 struct intel_super
*s
= NULL
;
4264 sprintf(nm
, "%d:%d", major
, minor
);
4265 dfd
= dev_open(nm
, O_RDWR
);
4271 rv
= find_intel_hba_capability(dfd
, s
, devname
);
4272 /* no orom/efi or non-intel hba of the disk */
4278 err
= load_and_parse_mpb(dfd
, s
, NULL
, keep_fd
);
4280 /* retry the load if we might have raced against mdmon */
4281 if (err
== 3 && (devnum
!= -1) && mdmon_running(devnum
))
4282 for (retry
= 0; retry
< 3; retry
++) {
4284 err
= load_and_parse_mpb(dfd
, s
, NULL
, keep_fd
);
4290 s
->next
= *super_list
;
4298 if ((dfd
>= 0) && (!keep_fd
))
4305 get_sra_super_block(int fd
, struct intel_super
**super_list
, char *devname
, int *max
, int keep_fd
)
4312 sra
= sysfs_read(fd
, 0, GET_LEVEL
|GET_VERSION
|GET_DEVS
|GET_STATE
);
4316 if (sra
->array
.major_version
!= -1 ||
4317 sra
->array
.minor_version
!= -2 ||
4318 strcmp(sra
->text_version
, "imsm") != 0) {
4323 devnum
= fd2devnum(fd
);
4324 for (sd
= sra
->devs
, i
= 0; sd
; sd
= sd
->next
, i
++) {
4325 if (get_super_block(super_list
, devnum
, devname
,
4326 sd
->disk
.major
, sd
->disk
.minor
, keep_fd
) != 0) {
4337 static int load_container_imsm(struct supertype
*st
, int fd
, char *devname
)
4339 return load_super_imsm_all(st
, fd
, &st
->sb
, devname
, NULL
, 1);
4343 static int load_super_imsm(struct supertype
*st
, int fd
, char *devname
)
4345 struct intel_super
*super
;
4348 if (test_partition(fd
))
4349 /* IMSM not allowed on partitions */
4352 free_super_imsm(st
);
4354 super
= alloc_super();
4357 Name
": malloc of %zu failed.\n",
4361 /* Load hba and capabilities if they exist.
4362 * But do not preclude loading metadata in case capabilities or hba are
4363 * non-compliant and ignore_hw_compat is set.
4365 rv
= find_intel_hba_capability(fd
, super
, devname
);
4366 /* no orom/efi or non-intel hba of the disk */
4367 if ((rv
!= 0) && (st
->ignore_hw_compat
== 0)) {
4370 Name
": No OROM/EFI properties for %s\n", devname
);
4374 rv
= load_and_parse_mpb(fd
, super
, devname
, 0);
4379 Name
": Failed to load all information "
4380 "sections on %s\n", devname
);
4386 if (st
->ss
== NULL
) {
4387 st
->ss
= &super_imsm
;
4388 st
->minor_version
= 0;
4389 st
->max_devs
= IMSM_MAX_DEVICES
;
4392 /* load migration record */
4393 if (load_imsm_migr_rec(super
, NULL
) == 0) {
4394 /* Check for unsupported migration features */
4395 if (check_mpb_migr_compatibility(super
) != 0) {
4397 Name
": Unsupported migration detected");
4399 fprintf(stderr
, " on %s\n", devname
);
4401 fprintf(stderr
, " (IMSM).\n");
4409 static __u16
info_to_blocks_per_strip(mdu_array_info_t
*info
)
4411 if (info
->level
== 1)
4413 return info
->chunk_size
>> 9;
4416 static unsigned long long info_to_blocks_per_member(mdu_array_info_t
*info
,
4417 unsigned long long size
)
4419 if (info
->level
== 1)
4422 return (size
* 2) & ~(info_to_blocks_per_strip(info
) - 1);
4425 static void imsm_update_version_info(struct intel_super
*super
)
4427 /* update the version and attributes */
4428 struct imsm_super
*mpb
= super
->anchor
;
4430 struct imsm_dev
*dev
;
4431 struct imsm_map
*map
;
4434 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
4435 dev
= get_imsm_dev(super
, i
);
4436 map
= get_imsm_map(dev
, MAP_0
);
4437 if (__le32_to_cpu(dev
->size_high
) > 0)
4438 mpb
->attributes
|= MPB_ATTRIB_2TB
;
4440 /* FIXME detect when an array spans a port multiplier */
4442 mpb
->attributes
|= MPB_ATTRIB_PM
;
4445 if (mpb
->num_raid_devs
> 1 ||
4446 mpb
->attributes
!= MPB_ATTRIB_CHECKSUM_VERIFY
) {
4447 version
= MPB_VERSION_ATTRIBS
;
4448 switch (get_imsm_raid_level(map
)) {
4449 case 0: mpb
->attributes
|= MPB_ATTRIB_RAID0
; break;
4450 case 1: mpb
->attributes
|= MPB_ATTRIB_RAID1
; break;
4451 case 10: mpb
->attributes
|= MPB_ATTRIB_RAID10
; break;
4452 case 5: mpb
->attributes
|= MPB_ATTRIB_RAID5
; break;
4455 if (map
->num_members
>= 5)
4456 version
= MPB_VERSION_5OR6_DISK_ARRAY
;
4457 else if (dev
->status
== DEV_CLONE_N_GO
)
4458 version
= MPB_VERSION_CNG
;
4459 else if (get_imsm_raid_level(map
) == 5)
4460 version
= MPB_VERSION_RAID5
;
4461 else if (map
->num_members
>= 3)
4462 version
= MPB_VERSION_3OR4_DISK_ARRAY
;
4463 else if (get_imsm_raid_level(map
) == 1)
4464 version
= MPB_VERSION_RAID1
;
4466 version
= MPB_VERSION_RAID0
;
4468 strcpy(((char *) mpb
->sig
) + strlen(MPB_SIGNATURE
), version
);
4472 static int check_name(struct intel_super
*super
, char *name
, int quiet
)
4474 struct imsm_super
*mpb
= super
->anchor
;
4475 char *reason
= NULL
;
4478 if (strlen(name
) > MAX_RAID_SERIAL_LEN
)
4479 reason
= "must be 16 characters or less";
4481 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
4482 struct imsm_dev
*dev
= get_imsm_dev(super
, i
);
4484 if (strncmp((char *) dev
->volume
, name
, MAX_RAID_SERIAL_LEN
) == 0) {
4485 reason
= "already exists";
4490 if (reason
&& !quiet
)
4491 fprintf(stderr
, Name
": imsm volume name %s\n", reason
);
4496 static int init_super_imsm_volume(struct supertype
*st
, mdu_array_info_t
*info
,
4497 unsigned long long size
, char *name
,
4498 char *homehost
, int *uuid
)
4500 /* We are creating a volume inside a pre-existing container.
4501 * so st->sb is already set.
4503 struct intel_super
*super
= st
->sb
;
4504 struct imsm_super
*mpb
= super
->anchor
;
4505 struct intel_dev
*dv
;
4506 struct imsm_dev
*dev
;
4507 struct imsm_vol
*vol
;
4508 struct imsm_map
*map
;
4509 int idx
= mpb
->num_raid_devs
;
4511 unsigned long long array_blocks
;
4512 size_t size_old
, size_new
;
4513 unsigned long long num_data_stripes
;
4515 if (super
->orom
&& mpb
->num_raid_devs
>= super
->orom
->vpa
) {
4516 fprintf(stderr
, Name
": This imsm-container already has the "
4517 "maximum of %d volumes\n", super
->orom
->vpa
);
4521 /* ensure the mpb is large enough for the new data */
4522 size_old
= __le32_to_cpu(mpb
->mpb_size
);
4523 size_new
= disks_to_mpb_size(info
->nr_disks
);
4524 if (size_new
> size_old
) {
4526 size_t size_round
= ROUND_UP(size_new
, 512);
4528 if (posix_memalign(&mpb_new
, 512, size_round
) != 0) {
4529 fprintf(stderr
, Name
": could not allocate new mpb\n");
4532 if (posix_memalign(&super
->migr_rec_buf
, 512,
4533 MIGR_REC_BUF_SIZE
) != 0) {
4534 fprintf(stderr
, Name
4535 ": %s could not allocate migr_rec buffer\n",
4542 memcpy(mpb_new
, mpb
, size_old
);
4545 super
->anchor
= mpb_new
;
4546 mpb
->mpb_size
= __cpu_to_le32(size_new
);
4547 memset(mpb_new
+ size_old
, 0, size_round
- size_old
);
4549 super
->current_vol
= idx
;
4551 /* handle 'failed_disks' by either:
4552 * a) create dummy disk entries in the table if this the first
4553 * volume in the array. We add them here as this is the only
4554 * opportunity to add them. add_to_super_imsm_volume()
4555 * handles the non-failed disks and continues incrementing
4557 * b) validate that 'failed_disks' matches the current number
4558 * of missing disks if the container is populated
4560 if (super
->current_vol
== 0) {
4562 for (i
= 0; i
< info
->failed_disks
; i
++) {
4563 struct imsm_disk
*disk
;
4566 disk
= __get_imsm_disk(mpb
, i
);
4567 disk
->status
= CONFIGURED_DISK
| FAILED_DISK
;
4568 disk
->scsi_id
= __cpu_to_le32(~(__u32
)0);
4569 snprintf((char *) disk
->serial
, MAX_RAID_SERIAL_LEN
,
4572 find_missing(super
);
4577 for (d
= super
->missing
; d
; d
= d
->next
)
4579 if (info
->failed_disks
> missing
) {
4580 fprintf(stderr
, Name
": unable to add 'missing' disk to container\n");
4585 if (!check_name(super
, name
, 0))
4587 dv
= malloc(sizeof(*dv
));
4589 fprintf(stderr
, Name
": failed to allocate device list entry\n");
4592 dev
= calloc(1, sizeof(*dev
) + sizeof(__u32
) * (info
->raid_disks
- 1));
4595 fprintf(stderr
, Name
": could not allocate raid device\n");
4599 strncpy((char *) dev
->volume
, name
, MAX_RAID_SERIAL_LEN
);
4600 if (info
->level
== 1)
4601 array_blocks
= info_to_blocks_per_member(info
, size
);
4603 array_blocks
= calc_array_size(info
->level
, info
->raid_disks
,
4604 info
->layout
, info
->chunk_size
,
4606 /* round array size down to closest MB */
4607 array_blocks
= (array_blocks
>> SECT_PER_MB_SHIFT
) << SECT_PER_MB_SHIFT
;
4609 dev
->size_low
= __cpu_to_le32((__u32
) array_blocks
);
4610 dev
->size_high
= __cpu_to_le32((__u32
) (array_blocks
>> 32));
4611 dev
->status
= (DEV_READ_COALESCING
| DEV_WRITE_COALESCING
);
4613 vol
->migr_state
= 0;
4614 set_migr_type(dev
, MIGR_INIT
);
4615 vol
->dirty
= !info
->state
;
4616 vol
->curr_migr_unit
= 0;
4617 map
= get_imsm_map(dev
, MAP_0
);
4618 set_pba_of_lba0(map
, super
->create_offset
);
4619 set_blocks_per_member(map
, info_to_blocks_per_member(info
, size
));
4620 map
->blocks_per_strip
= __cpu_to_le16(info_to_blocks_per_strip(info
));
4621 map
->failed_disk_num
= ~0;
4622 if (info
->level
> 0)
4623 map
->map_state
= IMSM_T_STATE_UNINITIALIZED
;
4625 map
->map_state
= info
->failed_disks
? IMSM_T_STATE_FAILED
:
4626 IMSM_T_STATE_NORMAL
;
4629 if (info
->level
== 1 && info
->raid_disks
> 2) {
4632 fprintf(stderr
, Name
": imsm does not support more than 2 disks"
4633 "in a raid1 volume\n");
4637 map
->raid_level
= info
->level
;
4638 if (info
->level
== 10) {
4639 map
->raid_level
= 1;
4640 map
->num_domains
= info
->raid_disks
/ 2;
4641 } else if (info
->level
== 1)
4642 map
->num_domains
= info
->raid_disks
;
4644 map
->num_domains
= 1;
4646 /* info->size is only int so use the 'size' parameter instead */
4647 num_data_stripes
= (size
* 2) / info_to_blocks_per_strip(info
);
4648 num_data_stripes
/= map
->num_domains
;
4649 set_num_data_stripes(map
, num_data_stripes
);
4651 map
->num_members
= info
->raid_disks
;
4652 for (i
= 0; i
< map
->num_members
; i
++) {
4653 /* initialized in add_to_super */
4654 set_imsm_ord_tbl_ent(map
, i
, IMSM_ORD_REBUILD
);
4656 mpb
->num_raid_devs
++;
4659 dv
->index
= super
->current_vol
;
4660 dv
->next
= super
->devlist
;
4661 super
->devlist
= dv
;
4663 imsm_update_version_info(super
);
4668 static int init_super_imsm(struct supertype
*st
, mdu_array_info_t
*info
,
4669 unsigned long long size
, char *name
,
4670 char *homehost
, int *uuid
)
4672 /* This is primarily called by Create when creating a new array.
4673 * We will then get add_to_super called for each component, and then
4674 * write_init_super called to write it out to each device.
4675 * For IMSM, Create can create on fresh devices or on a pre-existing
4677 * To create on a pre-existing array a different method will be called.
4678 * This one is just for fresh drives.
4680 struct intel_super
*super
;
4681 struct imsm_super
*mpb
;
4686 return init_super_imsm_volume(st
, info
, size
, name
, homehost
, uuid
);
4689 mpb_size
= disks_to_mpb_size(info
->nr_disks
);
4693 super
= alloc_super();
4694 if (super
&& posix_memalign(&super
->buf
, 512, mpb_size
) != 0) {
4699 fprintf(stderr
, Name
4700 ": %s could not allocate superblock\n", __func__
);
4703 if (posix_memalign(&super
->migr_rec_buf
, 512, MIGR_REC_BUF_SIZE
) != 0) {
4704 fprintf(stderr
, Name
4705 ": %s could not allocate migr_rec buffer\n", __func__
);
4710 memset(super
->buf
, 0, mpb_size
);
4712 mpb
->mpb_size
= __cpu_to_le32(mpb_size
);
4716 /* zeroing superblock */
4720 mpb
->attributes
= MPB_ATTRIB_CHECKSUM_VERIFY
;
4722 version
= (char *) mpb
->sig
;
4723 strcpy(version
, MPB_SIGNATURE
);
4724 version
+= strlen(MPB_SIGNATURE
);
4725 strcpy(version
, MPB_VERSION_RAID0
);
4731 static int add_to_super_imsm_volume(struct supertype
*st
, mdu_disk_info_t
*dk
,
4732 int fd
, char *devname
)
4734 struct intel_super
*super
= st
->sb
;
4735 struct imsm_super
*mpb
= super
->anchor
;
4736 struct imsm_disk
*_disk
;
4737 struct imsm_dev
*dev
;
4738 struct imsm_map
*map
;
4742 dev
= get_imsm_dev(super
, super
->current_vol
);
4743 map
= get_imsm_map(dev
, MAP_0
);
4745 if (! (dk
->state
& (1<<MD_DISK_SYNC
))) {
4746 fprintf(stderr
, Name
": %s: Cannot add spare devices to IMSM volume\n",
4752 /* we're doing autolayout so grab the pre-marked (in
4753 * validate_geometry) raid_disk
4755 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
4756 if (dl
->raiddisk
== dk
->raid_disk
)
4759 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
4760 if (dl
->major
== dk
->major
&&
4761 dl
->minor
== dk
->minor
)
4766 fprintf(stderr
, Name
": %s is not a member of the same container\n", devname
);
4770 /* add a pristine spare to the metadata */
4771 if (dl
->index
< 0) {
4772 dl
->index
= super
->anchor
->num_disks
;
4773 super
->anchor
->num_disks
++;
4775 /* Check the device has not already been added */
4776 slot
= get_imsm_disk_slot(map
, dl
->index
);
4778 (get_imsm_ord_tbl_ent(dev
, slot
, MAP_X
) & IMSM_ORD_REBUILD
) == 0) {
4779 fprintf(stderr
, Name
": %s has been included in this array twice\n",
4783 set_imsm_ord_tbl_ent(map
, dk
->raid_disk
, dl
->index
);
4784 dl
->disk
.status
= CONFIGURED_DISK
;
4786 /* update size of 'missing' disks to be at least as large as the
4787 * largest acitve member (we only have dummy missing disks when
4788 * creating the first volume)
4790 if (super
->current_vol
== 0) {
4791 for (df
= super
->missing
; df
; df
= df
->next
) {
4792 if (total_blocks(&dl
->disk
) > total_blocks(&df
->disk
))
4793 set_total_blocks(&df
->disk
, total_blocks(&dl
->disk
));
4794 _disk
= __get_imsm_disk(mpb
, df
->index
);
4799 /* refresh unset/failed slots to point to valid 'missing' entries */
4800 for (df
= super
->missing
; df
; df
= df
->next
)
4801 for (slot
= 0; slot
< mpb
->num_disks
; slot
++) {
4802 __u32 ord
= get_imsm_ord_tbl_ent(dev
, slot
, MAP_X
);
4804 if ((ord
& IMSM_ORD_REBUILD
) == 0)
4806 set_imsm_ord_tbl_ent(map
, slot
, df
->index
| IMSM_ORD_REBUILD
);
4807 if (is_gen_migration(dev
)) {
4808 struct imsm_map
*map2
= get_imsm_map(dev
,
4810 int slot2
= get_imsm_disk_slot(map2
, df
->index
);
4811 if ((slot2
< map2
->num_members
) &&
4813 __u32 ord2
= get_imsm_ord_tbl_ent(dev
,
4816 if ((unsigned)df
->index
==
4818 set_imsm_ord_tbl_ent(map2
,
4824 dprintf("set slot:%d to missing disk:%d\n", slot
, df
->index
);
4828 /* if we are creating the first raid device update the family number */
4829 if (super
->current_vol
== 0) {
4831 struct imsm_dev
*_dev
= __get_imsm_dev(mpb
, 0);
4833 _disk
= __get_imsm_disk(mpb
, dl
->index
);
4834 if (!_dev
|| !_disk
) {
4835 fprintf(stderr
, Name
": BUG mpb setup error\n");
4841 sum
+= __gen_imsm_checksum(mpb
);
4842 mpb
->family_num
= __cpu_to_le32(sum
);
4843 mpb
->orig_family_num
= mpb
->family_num
;
4845 super
->current_disk
= dl
;
4850 * Function marks disk as spare and restores disk serial
4851 * in case it was previously marked as failed by takeover operation
4853 * -1 : critical error
4854 * 0 : disk is marked as spare but serial is not set
4857 int mark_spare(struct dl
*disk
)
4859 __u8 serial
[MAX_RAID_SERIAL_LEN
];
4866 if (!imsm_read_serial(disk
->fd
, NULL
, serial
)) {
4867 /* Restore disk serial number, because takeover marks disk
4868 * as failed and adds to serial ':0' before it becomes
4871 serialcpy(disk
->serial
, serial
);
4872 serialcpy(disk
->disk
.serial
, serial
);
4875 disk
->disk
.status
= SPARE_DISK
;
4881 static int add_to_super_imsm(struct supertype
*st
, mdu_disk_info_t
*dk
,
4882 int fd
, char *devname
)
4884 struct intel_super
*super
= st
->sb
;
4886 unsigned long long size
;
4891 /* If we are on an RAID enabled platform check that the disk is
4892 * attached to the raid controller.
4893 * We do not need to test disks attachment for container based additions,
4894 * they shall be already tested when container was created/assembled.
4896 rv
= find_intel_hba_capability(fd
, super
, devname
);
4897 /* no orom/efi or non-intel hba of the disk */
4899 dprintf("capability: %p fd: %d ret: %d\n",
4900 super
->orom
, fd
, rv
);
4904 if (super
->current_vol
>= 0)
4905 return add_to_super_imsm_volume(st
, dk
, fd
, devname
);
4908 dd
= malloc(sizeof(*dd
));
4911 Name
": malloc failed %s:%d.\n", __func__
, __LINE__
);
4914 memset(dd
, 0, sizeof(*dd
));
4915 dd
->major
= major(stb
.st_rdev
);
4916 dd
->minor
= minor(stb
.st_rdev
);
4917 dd
->devname
= devname
? strdup(devname
) : NULL
;
4920 dd
->action
= DISK_ADD
;
4921 rv
= imsm_read_serial(fd
, devname
, dd
->serial
);
4924 Name
": failed to retrieve scsi serial, aborting\n");
4929 get_dev_size(fd
, NULL
, &size
);
4931 serialcpy(dd
->disk
.serial
, dd
->serial
);
4932 set_total_blocks(&dd
->disk
, size
);
4933 if (__le32_to_cpu(dd
->disk
.total_blocks_hi
) > 0) {
4934 struct imsm_super
*mpb
= super
->anchor
;
4935 mpb
->attributes
|= MPB_ATTRIB_2TB_DISK
;
4938 if (sysfs_disk_to_scsi_id(fd
, &id
) == 0)
4939 dd
->disk
.scsi_id
= __cpu_to_le32(id
);
4941 dd
->disk
.scsi_id
= __cpu_to_le32(0);
4943 if (st
->update_tail
) {
4944 dd
->next
= super
->disk_mgmt_list
;
4945 super
->disk_mgmt_list
= dd
;
4947 dd
->next
= super
->disks
;
4949 super
->updates_pending
++;
4956 static int remove_from_super_imsm(struct supertype
*st
, mdu_disk_info_t
*dk
)
4958 struct intel_super
*super
= st
->sb
;
4961 /* remove from super works only in mdmon - for communication
4962 * manager - monitor. Check if communication memory buffer
4965 if (!st
->update_tail
) {
4967 Name
": %s shall be used in mdmon context only"
4968 "(line %d).\n", __func__
, __LINE__
);
4971 dd
= malloc(sizeof(*dd
));
4974 Name
": malloc failed %s:%d.\n", __func__
, __LINE__
);
4977 memset(dd
, 0, sizeof(*dd
));
4978 dd
->major
= dk
->major
;
4979 dd
->minor
= dk
->minor
;
4982 dd
->action
= DISK_REMOVE
;
4984 dd
->next
= super
->disk_mgmt_list
;
4985 super
->disk_mgmt_list
= dd
;
4991 static int store_imsm_mpb(int fd
, struct imsm_super
*mpb
);
4995 struct imsm_super anchor
;
4996 } spare_record
__attribute__ ((aligned(512)));
4998 /* spare records have their own family number and do not have any defined raid
5001 static int write_super_imsm_spares(struct intel_super
*super
, int doclose
)
5003 struct imsm_super
*mpb
= super
->anchor
;
5004 struct imsm_super
*spare
= &spare_record
.anchor
;
5008 spare
->mpb_size
= __cpu_to_le32(sizeof(struct imsm_super
)),
5009 spare
->generation_num
= __cpu_to_le32(1UL),
5010 spare
->attributes
= MPB_ATTRIB_CHECKSUM_VERIFY
;
5011 spare
->num_disks
= 1,
5012 spare
->num_raid_devs
= 0,
5013 spare
->cache_size
= mpb
->cache_size
,
5014 spare
->pwr_cycle_count
= __cpu_to_le32(1),
5016 snprintf((char *) spare
->sig
, MAX_SIGNATURE_LENGTH
,
5017 MPB_SIGNATURE MPB_VERSION_RAID0
);
5019 for (d
= super
->disks
; d
; d
= d
->next
) {
5023 spare
->disk
[0] = d
->disk
;
5024 sum
= __gen_imsm_checksum(spare
);
5025 spare
->family_num
= __cpu_to_le32(sum
);
5026 spare
->orig_family_num
= 0;
5027 sum
= __gen_imsm_checksum(spare
);
5028 spare
->check_sum
= __cpu_to_le32(sum
);
5030 if (store_imsm_mpb(d
->fd
, spare
)) {
5031 fprintf(stderr
, "%s: failed for device %d:%d %s\n",
5032 __func__
, d
->major
, d
->minor
, strerror(errno
));
5044 static int write_super_imsm(struct supertype
*st
, int doclose
)
5046 struct intel_super
*super
= st
->sb
;
5047 struct imsm_super
*mpb
= super
->anchor
;
5053 __u32 mpb_size
= sizeof(struct imsm_super
) - sizeof(struct imsm_disk
);
5055 int clear_migration_record
= 1;
5057 /* 'generation' is incremented everytime the metadata is written */
5058 generation
= __le32_to_cpu(mpb
->generation_num
);
5060 mpb
->generation_num
= __cpu_to_le32(generation
);
5062 /* fix up cases where previous mdadm releases failed to set
5065 if (mpb
->orig_family_num
== 0)
5066 mpb
->orig_family_num
= mpb
->family_num
;
5068 for (d
= super
->disks
; d
; d
= d
->next
) {
5072 mpb
->disk
[d
->index
] = d
->disk
;
5076 for (d
= super
->missing
; d
; d
= d
->next
) {
5077 mpb
->disk
[d
->index
] = d
->disk
;
5080 mpb
->num_disks
= num_disks
;
5081 mpb_size
+= sizeof(struct imsm_disk
) * mpb
->num_disks
;
5083 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
5084 struct imsm_dev
*dev
= __get_imsm_dev(mpb
, i
);
5085 struct imsm_dev
*dev2
= get_imsm_dev(super
, i
);
5087 imsm_copy_dev(dev
, dev2
);
5088 mpb_size
+= sizeof_imsm_dev(dev
, 0);
5090 if (is_gen_migration(dev2
))
5091 clear_migration_record
= 0;
5093 mpb_size
+= __le32_to_cpu(mpb
->bbm_log_size
);
5094 mpb
->mpb_size
= __cpu_to_le32(mpb_size
);
5096 /* recalculate checksum */
5097 sum
= __gen_imsm_checksum(mpb
);
5098 mpb
->check_sum
= __cpu_to_le32(sum
);
5100 if (super
->clean_migration_record_by_mdmon
) {
5101 clear_migration_record
= 1;
5102 super
->clean_migration_record_by_mdmon
= 0;
5104 if (clear_migration_record
)
5105 memset(super
->migr_rec_buf
, 0, MIGR_REC_BUF_SIZE
);
5107 /* write the mpb for disks that compose raid devices */
5108 for (d
= super
->disks
; d
; d
= d
->next
) {
5109 if (d
->index
< 0 || is_failed(&d
->disk
))
5112 if (clear_migration_record
) {
5113 unsigned long long dsize
;
5115 get_dev_size(d
->fd
, NULL
, &dsize
);
5116 if (lseek64(d
->fd
, dsize
- 512, SEEK_SET
) >= 0) {
5117 if (write(d
->fd
, super
->migr_rec_buf
,
5118 MIGR_REC_BUF_SIZE
) != MIGR_REC_BUF_SIZE
)
5119 perror("Write migr_rec failed");
5123 if (store_imsm_mpb(d
->fd
, mpb
))
5125 "%s: failed for device %d:%d (fd: %d)%s\n",
5126 __func__
, d
->major
, d
->minor
,
5127 d
->fd
, strerror(errno
));
5136 return write_super_imsm_spares(super
, doclose
);
5142 static int create_array(struct supertype
*st
, int dev_idx
)
5145 struct imsm_update_create_array
*u
;
5146 struct intel_super
*super
= st
->sb
;
5147 struct imsm_dev
*dev
= get_imsm_dev(super
, dev_idx
);
5148 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
5149 struct disk_info
*inf
;
5150 struct imsm_disk
*disk
;
5153 len
= sizeof(*u
) - sizeof(*dev
) + sizeof_imsm_dev(dev
, 0) +
5154 sizeof(*inf
) * map
->num_members
;
5157 fprintf(stderr
, "%s: failed to allocate update buffer\n",
5162 u
->type
= update_create_array
;
5163 u
->dev_idx
= dev_idx
;
5164 imsm_copy_dev(&u
->dev
, dev
);
5165 inf
= get_disk_info(u
);
5166 for (i
= 0; i
< map
->num_members
; i
++) {
5167 int idx
= get_imsm_disk_idx(dev
, i
, MAP_X
);
5169 disk
= get_imsm_disk(super
, idx
);
5170 serialcpy(inf
[i
].serial
, disk
->serial
);
5172 append_metadata_update(st
, u
, len
);
5177 static int mgmt_disk(struct supertype
*st
)
5179 struct intel_super
*super
= st
->sb
;
5181 struct imsm_update_add_remove_disk
*u
;
5183 if (!super
->disk_mgmt_list
)
5189 fprintf(stderr
, "%s: failed to allocate update buffer\n",
5194 u
->type
= update_add_remove_disk
;
5195 append_metadata_update(st
, u
, len
);
5200 static int write_init_super_imsm(struct supertype
*st
)
5202 struct intel_super
*super
= st
->sb
;
5203 int current_vol
= super
->current_vol
;
5205 /* we are done with current_vol reset it to point st at the container */
5206 super
->current_vol
= -1;
5208 if (st
->update_tail
) {
5209 /* queue the recently created array / added disk
5210 * as a metadata update */
5213 /* determine if we are creating a volume or adding a disk */
5214 if (current_vol
< 0) {
5215 /* in the mgmt (add/remove) disk case we are running
5216 * in mdmon context, so don't close fd's
5218 return mgmt_disk(st
);
5220 rv
= create_array(st
, current_vol
);
5225 for (d
= super
->disks
; d
; d
= d
->next
)
5226 Kill(d
->devname
, NULL
, 0, 1, 1);
5227 return write_super_imsm(st
, 1);
5232 static int store_super_imsm(struct supertype
*st
, int fd
)
5234 struct intel_super
*super
= st
->sb
;
5235 struct imsm_super
*mpb
= super
? super
->anchor
: NULL
;
5241 return store_imsm_mpb(fd
, mpb
);
5247 static int imsm_bbm_log_size(struct imsm_super
*mpb
)
5249 return __le32_to_cpu(mpb
->bbm_log_size
);
5253 static int validate_geometry_imsm_container(struct supertype
*st
, int level
,
5254 int layout
, int raiddisks
, int chunk
,
5255 unsigned long long size
, char *dev
,
5256 unsigned long long *freesize
,
5260 unsigned long long ldsize
;
5261 struct intel_super
*super
=NULL
;
5264 if (level
!= LEVEL_CONTAINER
)
5269 fd
= open(dev
, O_RDONLY
|O_EXCL
, 0);
5272 fprintf(stderr
, Name
": imsm: Cannot open %s: %s\n",
5273 dev
, strerror(errno
));
5276 if (!get_dev_size(fd
, dev
, &ldsize
)) {
5281 /* capabilities retrieve could be possible
5282 * note that there is no fd for the disks in array.
5284 super
= alloc_super();
5287 Name
": malloc of %zu failed.\n",
5293 rv
= find_intel_hba_capability(fd
, super
, verbose
? dev
: NULL
);
5297 fd2devname(fd
, str
);
5298 dprintf("validate_geometry_imsm_container: fd: %d %s orom: %p rv: %d raiddisk: %d\n",
5299 fd
, str
, super
->orom
, rv
, raiddisks
);
5301 /* no orom/efi or non-intel hba of the disk */
5307 if (super
->orom
&& raiddisks
> super
->orom
->tds
) {
5309 fprintf(stderr
, Name
": %d exceeds maximum number of"
5310 " platform supported disks: %d\n",
5311 raiddisks
, super
->orom
->tds
);
5317 *freesize
= avail_size_imsm(st
, ldsize
>> 9);
5323 static unsigned long long find_size(struct extent
*e
, int *idx
, int num_extents
)
5325 const unsigned long long base_start
= e
[*idx
].start
;
5326 unsigned long long end
= base_start
+ e
[*idx
].size
;
5329 if (base_start
== end
)
5333 for (i
= *idx
; i
< num_extents
; i
++) {
5334 /* extend overlapping extents */
5335 if (e
[i
].start
>= base_start
&&
5336 e
[i
].start
<= end
) {
5339 if (e
[i
].start
+ e
[i
].size
> end
)
5340 end
= e
[i
].start
+ e
[i
].size
;
5341 } else if (e
[i
].start
> end
) {
5347 return end
- base_start
;
5350 static unsigned long long merge_extents(struct intel_super
*super
, int sum_extents
)
5352 /* build a composite disk with all known extents and generate a new
5353 * 'maxsize' given the "all disks in an array must share a common start
5354 * offset" constraint
5356 struct extent
*e
= calloc(sum_extents
, sizeof(*e
));
5360 unsigned long long pos
;
5361 unsigned long long start
= 0;
5362 unsigned long long maxsize
;
5363 unsigned long reserve
;
5368 /* coalesce and sort all extents. also, check to see if we need to
5369 * reserve space between member arrays
5372 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
5375 for (i
= 0; i
< dl
->extent_cnt
; i
++)
5378 qsort(e
, sum_extents
, sizeof(*e
), cmp_extent
);
5383 while (i
< sum_extents
) {
5384 e
[j
].start
= e
[i
].start
;
5385 e
[j
].size
= find_size(e
, &i
, sum_extents
);
5387 if (e
[j
-1].size
== 0)
5396 unsigned long long esize
;
5398 esize
= e
[i
].start
- pos
;
5399 if (esize
>= maxsize
) {
5404 pos
= e
[i
].start
+ e
[i
].size
;
5406 } while (e
[i
-1].size
);
5412 /* FIXME assumes volume at offset 0 is the first volume in a
5415 if (start_extent
> 0)
5416 reserve
= IMSM_RESERVED_SECTORS
; /* gap between raid regions */
5420 if (maxsize
< reserve
)
5423 super
->create_offset
= ~((unsigned long long) 0);
5424 if (start
+ reserve
> super
->create_offset
)
5425 return 0; /* start overflows create_offset */
5426 super
->create_offset
= start
+ reserve
;
5428 return maxsize
- reserve
;
5431 static int is_raid_level_supported(const struct imsm_orom
*orom
, int level
, int raiddisks
)
5433 if (level
< 0 || level
== 6 || level
== 4)
5436 /* if we have an orom prevent invalid raid levels */
5439 case 0: return imsm_orom_has_raid0(orom
);
5442 return imsm_orom_has_raid1e(orom
);
5443 return imsm_orom_has_raid1(orom
) && raiddisks
== 2;
5444 case 10: return imsm_orom_has_raid10(orom
) && raiddisks
== 4;
5445 case 5: return imsm_orom_has_raid5(orom
) && raiddisks
> 2;
5448 return 1; /* not on an Intel RAID platform so anything goes */
5455 active_arrays_by_format(char *name
, char* hba
, struct md_list
**devlist
,
5456 int dpa
, int verbose
)
5458 struct mdstat_ent
*mdstat
= mdstat_read(0, 0);
5459 struct mdstat_ent
*memb
= NULL
;
5462 struct md_list
*dv
= NULL
;
5465 for (memb
= mdstat
; memb
; memb
= memb
->next
) {
5466 if (memb
->metadata_version
&&
5467 (strncmp(memb
->metadata_version
, "external:", 9) == 0) &&
5468 (strcmp(&memb
->metadata_version
[9], name
) == 0) &&
5469 !is_subarray(memb
->metadata_version
+9) &&
5471 struct dev_member
*dev
= memb
->members
;
5473 while(dev
&& (fd
< 0)) {
5474 char *path
= malloc(strlen(dev
->name
) + strlen("/dev/") + 1);
5476 num
= sprintf(path
, "%s%s", "/dev/", dev
->name
);
5478 fd
= open(path
, O_RDONLY
, 0);
5479 if ((num
<= 0) || (fd
< 0)) {
5480 pr_vrb(": Cannot open %s: %s\n",
5481 dev
->name
, strerror(errno
));
5488 if ((fd
>= 0) && disk_attached_to_hba(fd
, hba
)) {
5489 struct mdstat_ent
*vol
;
5490 for (vol
= mdstat
; vol
; vol
= vol
->next
) {
5491 if ((vol
->active
> 0) &&
5492 vol
->metadata_version
&&
5493 is_container_member(vol
, memb
->dev
)) {
5498 if (*devlist
&& (found
< dpa
)) {
5499 dv
= calloc(1, sizeof(*dv
));
5501 fprintf(stderr
, Name
": calloc failed\n");
5503 dv
->devname
= malloc(strlen(memb
->dev
) + strlen("/dev/") + 1);
5504 if (dv
->devname
!= NULL
) {
5505 sprintf(dv
->devname
, "%s%s", "/dev/", memb
->dev
);
5508 dv
->next
= *devlist
;
5519 free_mdstat(mdstat
);
5524 static struct md_list
*
5525 get_loop_devices(void)
5528 struct md_list
*devlist
= NULL
;
5529 struct md_list
*dv
= NULL
;
5531 for(i
= 0; i
< 12; i
++) {
5532 dv
= calloc(1, sizeof(*dv
));
5534 fprintf(stderr
, Name
": calloc failed\n");
5537 dv
->devname
= malloc(40);
5538 if (dv
->devname
== NULL
) {
5539 fprintf(stderr
, Name
": malloc failed\n");
5543 sprintf(dv
->devname
, "/dev/loop%d", i
);
5551 static struct md_list
*
5552 get_devices(const char *hba_path
)
5554 struct md_list
*devlist
= NULL
;
5555 struct md_list
*dv
= NULL
;
5561 devlist
= get_loop_devices();
5564 /* scroll through /sys/dev/block looking for devices attached to
5567 dir
= opendir("/sys/dev/block");
5568 for (ent
= dir
? readdir(dir
) : NULL
; ent
; ent
= readdir(dir
)) {
5573 if (sscanf(ent
->d_name
, "%d:%d", &major
, &minor
) != 2)
5575 path
= devt_to_devpath(makedev(major
, minor
));
5578 if (!path_attached_to_hba(path
, hba_path
)) {
5585 fd
= dev_open(ent
->d_name
, O_RDONLY
);
5587 fd2devname(fd
, buf
);
5590 fprintf(stderr
, Name
": cannot open device: %s\n",
5596 dv
= calloc(1, sizeof(*dv
));
5598 fprintf(stderr
, Name
": malloc failed\n");
5602 dv
->devname
= strdup(buf
);
5603 if (dv
->devname
== NULL
) {
5604 fprintf(stderr
, Name
": malloc failed\n");
5615 devlist
= devlist
->next
;
5624 count_volumes_list(struct md_list
*devlist
, char *homehost
,
5625 int verbose
, int *found
)
5627 struct md_list
*tmpdev
;
5629 struct supertype
*st
= NULL
;
5631 /* first walk the list of devices to find a consistent set
5632 * that match the criterea, if that is possible.
5633 * We flag the ones we like with 'used'.
5636 st
= match_metadata_desc_imsm("imsm");
5638 pr_vrb(": cannot allocate memory for imsm supertype\n");
5642 for (tmpdev
= devlist
; tmpdev
; tmpdev
= tmpdev
->next
) {
5643 char *devname
= tmpdev
->devname
;
5645 struct supertype
*tst
;
5647 if (tmpdev
->used
> 1)
5649 tst
= dup_super(st
);
5651 pr_vrb(": cannot allocate memory for imsm supertype\n");
5654 tmpdev
->container
= 0;
5655 dfd
= dev_open(devname
, O_RDONLY
|O_EXCL
);
5657 dprintf(": cannot open device %s: %s\n",
5658 devname
, strerror(errno
));
5660 } else if (fstat(dfd
, &stb
)< 0) {
5662 dprintf(": fstat failed for %s: %s\n",
5663 devname
, strerror(errno
));
5665 } else if ((stb
.st_mode
& S_IFMT
) != S_IFBLK
) {
5666 dprintf(": %s is not a block device.\n",
5669 } else if (must_be_container(dfd
)) {
5670 struct supertype
*cst
;
5671 cst
= super_by_fd(dfd
, NULL
);
5673 dprintf(": cannot recognize container type %s\n",
5676 } else if (tst
->ss
!= st
->ss
) {
5677 dprintf(": non-imsm container - ignore it: %s\n",
5680 } else if (!tst
->ss
->load_container
||
5681 tst
->ss
->load_container(tst
, dfd
, NULL
))
5684 tmpdev
->container
= 1;
5687 cst
->ss
->free_super(cst
);
5689 tmpdev
->st_rdev
= stb
.st_rdev
;
5690 if (tst
->ss
->load_super(tst
,dfd
, NULL
)) {
5691 dprintf(": no RAID superblock on %s\n",
5694 } else if (tst
->ss
->compare_super
== NULL
) {
5695 dprintf(": Cannot assemble %s metadata on %s\n",
5696 tst
->ss
->name
, devname
);
5702 if (tmpdev
->used
== 2 || tmpdev
->used
== 4) {
5703 /* Ignore unrecognised devices during auto-assembly */
5708 tst
->ss
->getinfo_super(tst
, &info
, NULL
);
5710 if (st
->minor_version
== -1)
5711 st
->minor_version
= tst
->minor_version
;
5713 if (memcmp(info
.uuid
, uuid_zero
,
5714 sizeof(int[4])) == 0) {
5715 /* this is a floating spare. It cannot define
5716 * an array unless there are no more arrays of
5717 * this type to be found. It can be included
5718 * in an array of this type though.
5724 if (st
->ss
!= tst
->ss
||
5725 st
->minor_version
!= tst
->minor_version
||
5726 st
->ss
->compare_super(st
, tst
) != 0) {
5727 /* Some mismatch. If exactly one array matches this host,
5728 * we can resolve on that one.
5729 * Or, if we are auto assembling, we just ignore the second
5732 dprintf(": superblock on %s doesn't match others - assembly aborted\n",
5738 dprintf("found: devname: %s\n", devname
);
5742 tst
->ss
->free_super(tst
);
5746 if ((err
= load_super_imsm_all(st
, -1, &st
->sb
, NULL
, devlist
, 0)) == 0) {
5747 struct mdinfo
*iter
, *head
= st
->ss
->container_content(st
, NULL
);
5748 for (iter
= head
; iter
; iter
= iter
->next
) {
5749 dprintf("content->text_version: %s vol\n",
5750 iter
->text_version
);
5751 if (iter
->array
.state
& (1<<MD_SB_BLOCK_VOLUME
)) {
5752 /* do not assemble arrays with unsupported
5754 dprintf(": Cannot activate member %s.\n",
5755 iter
->text_version
);
5762 dprintf(" no valid super block on device list: err: %d %p\n",
5766 dprintf(" no more devices to examin\n");
5769 for (tmpdev
= devlist
; tmpdev
; tmpdev
= tmpdev
->next
) {
5770 if ((tmpdev
->used
== 1) && (tmpdev
->found
)) {
5772 if (count
< tmpdev
->found
)
5775 count
-= tmpdev
->found
;
5778 if (tmpdev
->used
== 1)
5783 st
->ss
->free_super(st
);
5789 count_volumes(char *hba
, int dpa
, int verbose
)
5791 struct md_list
*devlist
= NULL
;
5795 devlist
= get_devices(hba
);
5796 /* if no intel devices return zero volumes */
5797 if (devlist
== NULL
)
5800 count
= active_arrays_by_format("imsm", hba
, &devlist
, dpa
, verbose
);
5801 dprintf(" path: %s active arrays: %d\n", hba
, count
);
5802 if (devlist
== NULL
)
5806 count
+= count_volumes_list(devlist
,
5810 dprintf("found %d count: %d\n", found
, count
);
5813 dprintf("path: %s total number of volumes: %d\n", hba
, count
);
5816 struct md_list
*dv
= devlist
;
5817 devlist
= devlist
->next
;
5824 static int imsm_default_chunk(const struct imsm_orom
*orom
)
5826 /* up to 512 if the plaform supports it, otherwise the platform max.
5827 * 128 if no platform detected
5829 int fs
= max(7, orom
? fls(orom
->sss
) : 0);
5831 return min(512, (1 << fs
));
5835 validate_geometry_imsm_orom(struct intel_super
*super
, int level
, int layout
,
5836 int raiddisks
, int *chunk
, int verbose
)
5838 /* check/set platform and metadata limits/defaults */
5839 if (super
->orom
&& raiddisks
> super
->orom
->dpa
) {
5840 pr_vrb(": platform supports a maximum of %d disks per array\n",
5845 /* capabilities of OROM tested - copied from validate_geometry_imsm_volume */
5846 if (!is_raid_level_supported(super
->orom
, level
, raiddisks
)) {
5847 pr_vrb(": platform does not support raid%d with %d disk%s\n",
5848 level
, raiddisks
, raiddisks
> 1 ? "s" : "");
5852 if (chunk
&& (*chunk
== 0 || *chunk
== UnSet
))
5853 *chunk
= imsm_default_chunk(super
->orom
);
5855 if (super
->orom
&& chunk
&& !imsm_orom_has_chunk(super
->orom
, *chunk
)) {
5856 pr_vrb(": platform does not support a chunk size of: "
5861 if (layout
!= imsm_level_to_layout(level
)) {
5863 pr_vrb(": imsm raid 5 only supports the left-asymmetric layout\n");
5864 else if (level
== 10)
5865 pr_vrb(": imsm raid 10 only supports the n2 layout\n");
5867 pr_vrb(": imsm unknown layout %#x for this raid level %d\n",
5874 /* validate_geometry_imsm_volume - lifted from validate_geometry_ddf_bvd
5875 * FIX ME add ahci details
5877 static int validate_geometry_imsm_volume(struct supertype
*st
, int level
,
5878 int layout
, int raiddisks
, int *chunk
,
5879 unsigned long long size
, char *dev
,
5880 unsigned long long *freesize
,
5884 struct intel_super
*super
= st
->sb
;
5885 struct imsm_super
*mpb
;
5887 unsigned long long pos
= 0;
5888 unsigned long long maxsize
;
5892 /* We must have the container info already read in. */
5896 mpb
= super
->anchor
;
5898 if (!validate_geometry_imsm_orom(super
, level
, layout
, raiddisks
, chunk
, verbose
)) {
5899 fprintf(stderr
, Name
": RAID gemetry validation failed. "
5900 "Cannot proceed with the action(s).\n");
5904 /* General test: make sure there is space for
5905 * 'raiddisks' device extents of size 'size' at a given
5908 unsigned long long minsize
= size
;
5909 unsigned long long start_offset
= MaxSector
;
5912 minsize
= MPB_SECTOR_CNT
+ IMSM_RESERVED_SECTORS
;
5913 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
5918 e
= get_extents(super
, dl
);
5921 unsigned long long esize
;
5922 esize
= e
[i
].start
- pos
;
5923 if (esize
>= minsize
)
5925 if (found
&& start_offset
== MaxSector
) {
5928 } else if (found
&& pos
!= start_offset
) {
5932 pos
= e
[i
].start
+ e
[i
].size
;
5934 } while (e
[i
-1].size
);
5939 if (dcnt
< raiddisks
) {
5941 fprintf(stderr
, Name
": imsm: Not enough "
5942 "devices with space for this array "
5950 /* This device must be a member of the set */
5951 if (stat(dev
, &stb
) < 0)
5953 if ((S_IFMT
& stb
.st_mode
) != S_IFBLK
)
5955 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
5956 if (dl
->major
== (int)major(stb
.st_rdev
) &&
5957 dl
->minor
== (int)minor(stb
.st_rdev
))
5962 fprintf(stderr
, Name
": %s is not in the "
5963 "same imsm set\n", dev
);
5965 } else if (super
->orom
&& dl
->index
< 0 && mpb
->num_raid_devs
) {
5966 /* If a volume is present then the current creation attempt
5967 * cannot incorporate new spares because the orom may not
5968 * understand this configuration (all member disks must be
5969 * members of each array in the container).
5971 fprintf(stderr
, Name
": %s is a spare and a volume"
5972 " is already defined for this container\n", dev
);
5973 fprintf(stderr
, Name
": The option-rom requires all member"
5974 " disks to be a member of all volumes\n");
5976 } else if (super
->orom
&& mpb
->num_raid_devs
> 0 &&
5977 mpb
->num_disks
!= raiddisks
) {
5978 fprintf(stderr
, Name
": The option-rom requires all member"
5979 " disks to be a member of all volumes\n");
5983 /* retrieve the largest free space block */
5984 e
= get_extents(super
, dl
);
5989 unsigned long long esize
;
5991 esize
= e
[i
].start
- pos
;
5992 if (esize
>= maxsize
)
5994 pos
= e
[i
].start
+ e
[i
].size
;
5996 } while (e
[i
-1].size
);
6001 fprintf(stderr
, Name
": unable to determine free space for: %s\n",
6005 if (maxsize
< size
) {
6007 fprintf(stderr
, Name
": %s not enough space (%llu < %llu)\n",
6008 dev
, maxsize
, size
);
6012 /* count total number of extents for merge */
6014 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
6016 i
+= dl
->extent_cnt
;
6018 maxsize
= merge_extents(super
, i
);
6020 if (!check_env("IMSM_NO_PLATFORM") &&
6021 mpb
->num_raid_devs
> 0 && size
&& size
!= maxsize
) {
6022 fprintf(stderr
, Name
": attempting to create a second "
6023 "volume with size less then remaining space. "
6028 if (maxsize
< size
|| maxsize
== 0) {
6031 fprintf(stderr
, Name
": no free space"
6032 " left on device. Aborting...\n");
6034 fprintf(stderr
, Name
": not enough space"
6035 " to create volume of given size"
6036 " (%llu < %llu). Aborting...\n",
6042 *freesize
= maxsize
;
6045 int count
= count_volumes(super
->hba
->path
,
6046 super
->orom
->dpa
, verbose
);
6047 if (super
->orom
->vphba
<= count
) {
6048 pr_vrb(": platform does not support more then %d raid volumes.\n",
6049 super
->orom
->vphba
);
6056 static int reserve_space(struct supertype
*st
, int raiddisks
,
6057 unsigned long long size
, int chunk
,
6058 unsigned long long *freesize
)
6060 struct intel_super
*super
= st
->sb
;
6061 struct imsm_super
*mpb
= super
->anchor
;
6066 unsigned long long maxsize
;
6067 unsigned long long minsize
;
6071 /* find the largest common start free region of the possible disks */
6075 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
6081 /* don't activate new spares if we are orom constrained
6082 * and there is already a volume active in the container
6084 if (super
->orom
&& dl
->index
< 0 && mpb
->num_raid_devs
)
6087 e
= get_extents(super
, dl
);
6090 for (i
= 1; e
[i
-1].size
; i
++)
6098 maxsize
= merge_extents(super
, extent_cnt
);
6102 minsize
= chunk
* 2;
6104 if (cnt
< raiddisks
||
6105 (super
->orom
&& used
&& used
!= raiddisks
) ||
6106 maxsize
< minsize
||
6108 fprintf(stderr
, Name
": not enough devices with space to create array.\n");
6109 return 0; /* No enough free spaces large enough */
6120 if (!check_env("IMSM_NO_PLATFORM") &&
6121 mpb
->num_raid_devs
> 0 && size
&& size
!= maxsize
) {
6122 fprintf(stderr
, Name
": attempting to create a second "
6123 "volume with size less then remaining space. "
6128 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
6130 dl
->raiddisk
= cnt
++;
6137 static int validate_geometry_imsm(struct supertype
*st
, int level
, int layout
,
6138 int raiddisks
, int *chunk
, unsigned long long size
,
6139 char *dev
, unsigned long long *freesize
,
6147 * if given unused devices create a container
6148 * if given given devices in a container create a member volume
6150 if (level
== LEVEL_CONTAINER
) {
6151 /* Must be a fresh device to add to a container */
6152 return validate_geometry_imsm_container(st
, level
, layout
,
6154 chunk
?*chunk
:0, size
,
6161 struct intel_super
*super
= st
->sb
;
6162 if (!validate_geometry_imsm_orom(st
->sb
, level
, layout
,
6166 /* we are being asked to automatically layout a
6167 * new volume based on the current contents of
6168 * the container. If the the parameters can be
6169 * satisfied reserve_space will record the disks,
6170 * start offset, and size of the volume to be
6171 * created. add_to_super and getinfo_super
6172 * detect when autolayout is in progress.
6174 /* assuming that freesize is always given when array is
6176 if (super
->orom
&& freesize
) {
6178 count
= count_volumes(super
->hba
->path
,
6179 super
->orom
->dpa
, verbose
);
6180 if (super
->orom
->vphba
<= count
) {
6181 pr_vrb(": platform does not support more"
6182 "then %d raid volumes.\n",
6183 super
->orom
->vphba
);
6188 return reserve_space(st
, raiddisks
, size
,
6189 chunk
?*chunk
:0, freesize
);
6194 /* creating in a given container */
6195 return validate_geometry_imsm_volume(st
, level
, layout
,
6196 raiddisks
, chunk
, size
,
6197 dev
, freesize
, verbose
);
6200 /* This device needs to be a device in an 'imsm' container */
6201 fd
= open(dev
, O_RDONLY
|O_EXCL
, 0);
6205 Name
": Cannot create this array on device %s\n",
6210 if (errno
!= EBUSY
|| (fd
= open(dev
, O_RDONLY
, 0)) < 0) {
6212 fprintf(stderr
, Name
": Cannot open %s: %s\n",
6213 dev
, strerror(errno
));
6216 /* Well, it is in use by someone, maybe an 'imsm' container. */
6217 cfd
= open_container(fd
);
6221 fprintf(stderr
, Name
": Cannot use %s: It is busy\n",
6225 sra
= sysfs_read(cfd
, 0, GET_VERSION
);
6226 if (sra
&& sra
->array
.major_version
== -1 &&
6227 strcmp(sra
->text_version
, "imsm") == 0)
6231 /* This is a member of a imsm container. Load the container
6232 * and try to create a volume
6234 struct intel_super
*super
;
6236 if (load_super_imsm_all(st
, cfd
, (void **) &super
, NULL
, NULL
, 1) == 0) {
6238 st
->container_dev
= fd2devnum(cfd
);
6240 return validate_geometry_imsm_volume(st
, level
, layout
,
6249 fprintf(stderr
, Name
": failed container membership check\n");
6255 static void default_geometry_imsm(struct supertype
*st
, int *level
, int *layout
, int *chunk
)
6257 struct intel_super
*super
= st
->sb
;
6259 if (level
&& *level
== UnSet
)
6260 *level
= LEVEL_CONTAINER
;
6262 if (level
&& layout
&& *layout
== UnSet
)
6263 *layout
= imsm_level_to_layout(*level
);
6265 if (chunk
&& (*chunk
== UnSet
|| *chunk
== 0))
6266 *chunk
= imsm_default_chunk(super
->orom
);
6269 static void handle_missing(struct intel_super
*super
, struct imsm_dev
*dev
);
6271 static int kill_subarray_imsm(struct supertype
*st
)
6273 /* remove the subarray currently referenced by ->current_vol */
6275 struct intel_dev
**dp
;
6276 struct intel_super
*super
= st
->sb
;
6277 __u8 current_vol
= super
->current_vol
;
6278 struct imsm_super
*mpb
= super
->anchor
;
6280 if (super
->current_vol
< 0)
6282 super
->current_vol
= -1; /* invalidate subarray cursor */
6284 /* block deletions that would change the uuid of active subarrays
6286 * FIXME when immutable ids are available, but note that we'll
6287 * also need to fixup the invalidated/active subarray indexes in
6290 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
6293 if (i
< current_vol
)
6295 sprintf(subarray
, "%u", i
);
6296 if (is_subarray_active(subarray
, st
->devname
)) {
6298 Name
": deleting subarray-%d would change the UUID of active subarray-%d, aborting\n",
6305 if (st
->update_tail
) {
6306 struct imsm_update_kill_array
*u
= malloc(sizeof(*u
));
6310 u
->type
= update_kill_array
;
6311 u
->dev_idx
= current_vol
;
6312 append_metadata_update(st
, u
, sizeof(*u
));
6317 for (dp
= &super
->devlist
; *dp
;)
6318 if ((*dp
)->index
== current_vol
) {
6321 handle_missing(super
, (*dp
)->dev
);
6322 if ((*dp
)->index
> current_vol
)
6327 /* no more raid devices, all active components are now spares,
6328 * but of course failed are still failed
6330 if (--mpb
->num_raid_devs
== 0) {
6333 for (d
= super
->disks
; d
; d
= d
->next
)
6338 super
->updates_pending
++;
6343 static int update_subarray_imsm(struct supertype
*st
, char *subarray
,
6344 char *update
, struct mddev_ident
*ident
)
6346 /* update the subarray currently referenced by ->current_vol */
6347 struct intel_super
*super
= st
->sb
;
6348 struct imsm_super
*mpb
= super
->anchor
;
6350 if (strcmp(update
, "name") == 0) {
6351 char *name
= ident
->name
;
6355 if (is_subarray_active(subarray
, st
->devname
)) {
6357 Name
": Unable to update name of active subarray\n");
6361 if (!check_name(super
, name
, 0))
6364 vol
= strtoul(subarray
, &ep
, 10);
6365 if (*ep
!= '\0' || vol
>= super
->anchor
->num_raid_devs
)
6368 if (st
->update_tail
) {
6369 struct imsm_update_rename_array
*u
= malloc(sizeof(*u
));
6373 u
->type
= update_rename_array
;
6375 snprintf((char *) u
->name
, MAX_RAID_SERIAL_LEN
, "%s", name
);
6376 append_metadata_update(st
, u
, sizeof(*u
));
6378 struct imsm_dev
*dev
;
6381 dev
= get_imsm_dev(super
, vol
);
6382 snprintf((char *) dev
->volume
, MAX_RAID_SERIAL_LEN
, "%s", name
);
6383 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
6384 dev
= get_imsm_dev(super
, i
);
6385 handle_missing(super
, dev
);
6387 super
->updates_pending
++;
6394 #endif /* MDASSEMBLE */
6396 static int is_gen_migration(struct imsm_dev
*dev
)
6401 if (!dev
->vol
.migr_state
)
6404 if (migr_type(dev
) == MIGR_GEN_MIGR
)
6410 static int is_rebuilding(struct imsm_dev
*dev
)
6412 struct imsm_map
*migr_map
;
6414 if (!dev
->vol
.migr_state
)
6417 if (migr_type(dev
) != MIGR_REBUILD
)
6420 migr_map
= get_imsm_map(dev
, MAP_1
);
6422 if (migr_map
->map_state
== IMSM_T_STATE_DEGRADED
)
6429 static int is_initializing(struct imsm_dev
*dev
)
6431 struct imsm_map
*migr_map
;
6433 if (!dev
->vol
.migr_state
)
6436 if (migr_type(dev
) != MIGR_INIT
)
6439 migr_map
= get_imsm_map(dev
, MAP_1
);
6441 if (migr_map
->map_state
== IMSM_T_STATE_UNINITIALIZED
)
6448 static void update_recovery_start(struct intel_super
*super
,
6449 struct imsm_dev
*dev
,
6450 struct mdinfo
*array
)
6452 struct mdinfo
*rebuild
= NULL
;
6456 if (!is_rebuilding(dev
))
6459 /* Find the rebuild target, but punt on the dual rebuild case */
6460 for (d
= array
->devs
; d
; d
= d
->next
)
6461 if (d
->recovery_start
== 0) {
6468 /* (?) none of the disks are marked with
6469 * IMSM_ORD_REBUILD, so assume they are missing and the
6470 * disk_ord_tbl was not correctly updated
6472 dprintf("%s: failed to locate out-of-sync disk\n", __func__
);
6476 units
= __le32_to_cpu(dev
->vol
.curr_migr_unit
);
6477 rebuild
->recovery_start
= units
* blocks_per_migr_unit(super
, dev
);
6481 static int recover_backup_imsm(struct supertype
*st
, struct mdinfo
*info
);
6484 static struct mdinfo
*container_content_imsm(struct supertype
*st
, char *subarray
)
6486 /* Given a container loaded by load_super_imsm_all,
6487 * extract information about all the arrays into
6489 * If 'subarray' is given, just extract info about that array.
6491 * For each imsm_dev create an mdinfo, fill it in,
6492 * then look for matching devices in super->disks
6493 * and create appropriate device mdinfo.
6495 struct intel_super
*super
= st
->sb
;
6496 struct imsm_super
*mpb
= super
->anchor
;
6497 struct mdinfo
*rest
= NULL
;
6501 int spare_disks
= 0;
6503 /* do not assemble arrays when not all attributes are supported */
6504 if (imsm_check_attributes(mpb
->attributes
) == 0) {
6506 fprintf(stderr
, Name
": Unsupported attributes in IMSM metadata."
6507 "Arrays activation is blocked.\n");
6510 /* check for bad blocks */
6511 if (imsm_bbm_log_size(super
->anchor
)) {
6512 fprintf(stderr
, Name
": BBM log found in IMSM metadata."
6513 "Arrays activation is blocked.\n");
6518 /* count spare devices, not used in maps
6520 for (d
= super
->disks
; d
; d
= d
->next
)
6524 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
6525 struct imsm_dev
*dev
;
6526 struct imsm_map
*map
;
6527 struct imsm_map
*map2
;
6528 struct mdinfo
*this;
6536 (i
!= strtoul(subarray
, &ep
, 10) || *ep
!= '\0'))
6539 dev
= get_imsm_dev(super
, i
);
6540 map
= get_imsm_map(dev
, MAP_0
);
6541 map2
= get_imsm_map(dev
, MAP_1
);
6543 /* do not publish arrays that are in the middle of an
6544 * unsupported migration
6546 if (dev
->vol
.migr_state
&&
6547 (migr_type(dev
) == MIGR_STATE_CHANGE
)) {
6548 fprintf(stderr
, Name
": cannot assemble volume '%.16s':"
6549 " unsupported migration in progress\n",
6553 /* do not publish arrays that are not support by controller's
6557 this = malloc(sizeof(*this));
6559 fprintf(stderr
, Name
": failed to allocate %zu bytes\n",
6564 super
->current_vol
= i
;
6565 getinfo_super_imsm_volume(st
, this, NULL
);
6568 chunk
= __le16_to_cpu(map
->blocks_per_strip
) >> 1;
6569 /* mdadm does not support all metadata features- set the bit in all arrays state */
6570 if (!validate_geometry_imsm_orom(super
,
6571 get_imsm_raid_level(map
), /* RAID level */
6572 imsm_level_to_layout(get_imsm_raid_level(map
)),
6573 map
->num_members
, /* raid disks */
6576 fprintf(stderr
, Name
": IMSM RAID geometry validation"
6577 " failed. Array %s activation is blocked.\n",
6579 this->array
.state
|=
6580 (1<<MD_SB_BLOCK_CONTAINER_RESHAPE
) |
6581 (1<<MD_SB_BLOCK_VOLUME
);
6585 /* if array has bad blocks, set suitable bit in all arrays state */
6587 this->array
.state
|=
6588 (1<<MD_SB_BLOCK_CONTAINER_RESHAPE
) |
6589 (1<<MD_SB_BLOCK_VOLUME
);
6591 for (slot
= 0 ; slot
< map
->num_members
; slot
++) {
6592 unsigned long long recovery_start
;
6593 struct mdinfo
*info_d
;
6600 idx
= get_imsm_disk_idx(dev
, slot
, MAP_0
);
6601 ord
= get_imsm_ord_tbl_ent(dev
, slot
, MAP_X
);
6602 for (d
= super
->disks
; d
; d
= d
->next
)
6603 if (d
->index
== idx
)
6606 recovery_start
= MaxSector
;
6609 if (d
&& is_failed(&d
->disk
))
6611 if (ord
& IMSM_ORD_REBUILD
)
6615 * if we skip some disks the array will be assmebled degraded;
6616 * reset resync start to avoid a dirty-degraded
6617 * situation when performing the intial sync
6619 * FIXME handle dirty degraded
6621 if ((skip
|| recovery_start
== 0) && !dev
->vol
.dirty
)
6622 this->resync_start
= MaxSector
;
6626 info_d
= calloc(1, sizeof(*info_d
));
6628 fprintf(stderr
, Name
": failed to allocate disk"
6629 " for volume %.16s\n", dev
->volume
);
6630 info_d
= this->devs
;
6632 struct mdinfo
*d
= info_d
->next
;
6641 info_d
->next
= this->devs
;
6642 this->devs
= info_d
;
6644 info_d
->disk
.number
= d
->index
;
6645 info_d
->disk
.major
= d
->major
;
6646 info_d
->disk
.minor
= d
->minor
;
6647 info_d
->disk
.raid_disk
= slot
;
6648 info_d
->recovery_start
= recovery_start
;
6650 if (slot
< map2
->num_members
)
6651 info_d
->disk
.state
= (1 << MD_DISK_ACTIVE
);
6653 this->array
.spare_disks
++;
6655 if (slot
< map
->num_members
)
6656 info_d
->disk
.state
= (1 << MD_DISK_ACTIVE
);
6658 this->array
.spare_disks
++;
6660 if (info_d
->recovery_start
== MaxSector
)
6661 this->array
.working_disks
++;
6663 info_d
->events
= __le32_to_cpu(mpb
->generation_num
);
6664 info_d
->data_offset
= pba_of_lba0(map
);
6665 info_d
->component_size
= blocks_per_member(map
);
6667 /* now that the disk list is up-to-date fixup recovery_start */
6668 update_recovery_start(super
, dev
, this);
6669 this->array
.spare_disks
+= spare_disks
;
6672 /* check for reshape */
6673 if (this->reshape_active
== 1)
6674 recover_backup_imsm(st
, this);
6683 static __u8
imsm_check_degraded(struct intel_super
*super
, struct imsm_dev
*dev
,
6684 int failed
, int look_in_map
)
6686 struct imsm_map
*map
;
6688 map
= get_imsm_map(dev
, look_in_map
);
6691 return map
->map_state
== IMSM_T_STATE_UNINITIALIZED
?
6692 IMSM_T_STATE_UNINITIALIZED
: IMSM_T_STATE_NORMAL
;
6694 switch (get_imsm_raid_level(map
)) {
6696 return IMSM_T_STATE_FAILED
;
6699 if (failed
< map
->num_members
)
6700 return IMSM_T_STATE_DEGRADED
;
6702 return IMSM_T_STATE_FAILED
;
6707 * check to see if any mirrors have failed, otherwise we
6708 * are degraded. Even numbered slots are mirrored on
6712 /* gcc -Os complains that this is unused */
6713 int insync
= insync
;
6715 for (i
= 0; i
< map
->num_members
; i
++) {
6716 __u32 ord
= get_imsm_ord_tbl_ent(dev
, i
, MAP_X
);
6717 int idx
= ord_to_idx(ord
);
6718 struct imsm_disk
*disk
;
6720 /* reset the potential in-sync count on even-numbered
6721 * slots. num_copies is always 2 for imsm raid10
6726 disk
= get_imsm_disk(super
, idx
);
6727 if (!disk
|| is_failed(disk
) || ord
& IMSM_ORD_REBUILD
)
6730 /* no in-sync disks left in this mirror the
6734 return IMSM_T_STATE_FAILED
;
6737 return IMSM_T_STATE_DEGRADED
;
6741 return IMSM_T_STATE_DEGRADED
;
6743 return IMSM_T_STATE_FAILED
;
6749 return map
->map_state
;
6752 static int imsm_count_failed(struct intel_super
*super
, struct imsm_dev
*dev
,
6757 struct imsm_disk
*disk
;
6758 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
6759 struct imsm_map
*prev
= get_imsm_map(dev
, MAP_1
);
6760 struct imsm_map
*map_for_loop
;
6765 /* at the beginning of migration we set IMSM_ORD_REBUILD on
6766 * disks that are being rebuilt. New failures are recorded to
6767 * map[0]. So we look through all the disks we started with and
6768 * see if any failures are still present, or if any new ones
6772 if (prev
&& (map
->num_members
< prev
->num_members
))
6773 map_for_loop
= prev
;
6775 for (i
= 0; i
< map_for_loop
->num_members
; i
++) {
6777 /* when MAP_X is passed both maps failures are counted
6780 ((look_in_map
== MAP_1
) || (look_in_map
== MAP_X
)) &&
6781 (i
< prev
->num_members
)) {
6782 ord
= __le32_to_cpu(prev
->disk_ord_tbl
[i
]);
6783 idx_1
= ord_to_idx(ord
);
6785 disk
= get_imsm_disk(super
, idx_1
);
6786 if (!disk
|| is_failed(disk
) || ord
& IMSM_ORD_REBUILD
)
6789 if (((look_in_map
== MAP_0
) || (look_in_map
== MAP_X
)) &&
6790 (i
< map
->num_members
)) {
6791 ord
= __le32_to_cpu(map
->disk_ord_tbl
[i
]);
6792 idx
= ord_to_idx(ord
);
6795 disk
= get_imsm_disk(super
, idx
);
6796 if (!disk
|| is_failed(disk
) ||
6797 ord
& IMSM_ORD_REBUILD
)
6807 static int imsm_open_new(struct supertype
*c
, struct active_array
*a
,
6810 struct intel_super
*super
= c
->sb
;
6811 struct imsm_super
*mpb
= super
->anchor
;
6813 if (atoi(inst
) >= mpb
->num_raid_devs
) {
6814 fprintf(stderr
, "%s: subarry index %d, out of range\n",
6815 __func__
, atoi(inst
));
6819 dprintf("imsm: open_new %s\n", inst
);
6820 a
->info
.container_member
= atoi(inst
);
6824 static int is_resyncing(struct imsm_dev
*dev
)
6826 struct imsm_map
*migr_map
;
6828 if (!dev
->vol
.migr_state
)
6831 if (migr_type(dev
) == MIGR_INIT
||
6832 migr_type(dev
) == MIGR_REPAIR
)
6835 if (migr_type(dev
) == MIGR_GEN_MIGR
)
6838 migr_map
= get_imsm_map(dev
, MAP_1
);
6840 if ((migr_map
->map_state
== IMSM_T_STATE_NORMAL
) &&
6841 (dev
->vol
.migr_type
!= MIGR_GEN_MIGR
))
6847 /* return true if we recorded new information */
6848 static int mark_failure(struct imsm_dev
*dev
, struct imsm_disk
*disk
, int idx
)
6852 struct imsm_map
*map
;
6853 char buf
[MAX_RAID_SERIAL_LEN
+3];
6854 unsigned int len
, shift
= 0;
6856 /* new failures are always set in map[0] */
6857 map
= get_imsm_map(dev
, MAP_0
);
6859 slot
= get_imsm_disk_slot(map
, idx
);
6863 ord
= __le32_to_cpu(map
->disk_ord_tbl
[slot
]);
6864 if (is_failed(disk
) && (ord
& IMSM_ORD_REBUILD
))
6867 memcpy(buf
, disk
->serial
, MAX_RAID_SERIAL_LEN
);
6868 buf
[MAX_RAID_SERIAL_LEN
] = '\000';
6870 if ((len
= strlen(buf
)) >= MAX_RAID_SERIAL_LEN
)
6871 shift
= len
- MAX_RAID_SERIAL_LEN
+ 1;
6872 strncpy((char *)disk
->serial
, &buf
[shift
], MAX_RAID_SERIAL_LEN
);
6874 disk
->status
|= FAILED_DISK
;
6875 set_imsm_ord_tbl_ent(map
, slot
, idx
| IMSM_ORD_REBUILD
);
6876 /* mark failures in second map if second map exists and this disk
6878 * This is valid for migration, initialization and rebuild
6880 if (dev
->vol
.migr_state
) {
6881 struct imsm_map
*map2
= get_imsm_map(dev
, MAP_1
);
6882 int slot2
= get_imsm_disk_slot(map2
, idx
);
6884 if ((slot2
< map2
->num_members
) &&
6886 set_imsm_ord_tbl_ent(map2
, slot2
,
6887 idx
| IMSM_ORD_REBUILD
);
6889 if (map
->failed_disk_num
== 0xff)
6890 map
->failed_disk_num
= slot
;
6894 static void mark_missing(struct imsm_dev
*dev
, struct imsm_disk
*disk
, int idx
)
6896 mark_failure(dev
, disk
, idx
);
6898 if (disk
->scsi_id
== __cpu_to_le32(~(__u32
)0))
6901 disk
->scsi_id
= __cpu_to_le32(~(__u32
)0);
6902 memmove(&disk
->serial
[0], &disk
->serial
[1], MAX_RAID_SERIAL_LEN
- 1);
6905 static void handle_missing(struct intel_super
*super
, struct imsm_dev
*dev
)
6909 if (!super
->missing
)
6912 dprintf("imsm: mark missing\n");
6913 /* end process for initialization and rebuild only
6915 if (is_gen_migration(dev
) == 0) {
6919 failed
= imsm_count_failed(super
, dev
, MAP_0
);
6920 map_state
= imsm_check_degraded(super
, dev
, failed
, MAP_0
);
6922 end_migration(dev
, super
, map_state
);
6924 for (dl
= super
->missing
; dl
; dl
= dl
->next
)
6925 mark_missing(dev
, &dl
->disk
, dl
->index
);
6926 super
->updates_pending
++;
6929 static unsigned long long imsm_set_array_size(struct imsm_dev
*dev
)
6931 int used_disks
= imsm_num_data_members(dev
, MAP_0
);
6932 unsigned long long array_blocks
;
6933 struct imsm_map
*map
;
6935 if (used_disks
== 0) {
6936 /* when problems occures
6937 * return current array_blocks value
6939 array_blocks
= __le32_to_cpu(dev
->size_high
);
6940 array_blocks
= array_blocks
<< 32;
6941 array_blocks
+= __le32_to_cpu(dev
->size_low
);
6943 return array_blocks
;
6946 /* set array size in metadata
6948 map
= get_imsm_map(dev
, MAP_0
);
6949 array_blocks
= blocks_per_member(map
) * used_disks
;
6951 /* round array size down to closest MB
6953 array_blocks
= (array_blocks
>> SECT_PER_MB_SHIFT
) << SECT_PER_MB_SHIFT
;
6954 dev
->size_low
= __cpu_to_le32((__u32
)array_blocks
);
6955 dev
->size_high
= __cpu_to_le32((__u32
)(array_blocks
>> 32));
6957 return array_blocks
;
6960 static void imsm_set_disk(struct active_array
*a
, int n
, int state
);
6962 static void imsm_progress_container_reshape(struct intel_super
*super
)
6964 /* if no device has a migr_state, but some device has a
6965 * different number of members than the previous device, start
6966 * changing the number of devices in this device to match
6969 struct imsm_super
*mpb
= super
->anchor
;
6970 int prev_disks
= -1;
6974 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
6975 struct imsm_dev
*dev
= get_imsm_dev(super
, i
);
6976 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
6977 struct imsm_map
*map2
;
6978 int prev_num_members
;
6980 if (dev
->vol
.migr_state
)
6983 if (prev_disks
== -1)
6984 prev_disks
= map
->num_members
;
6985 if (prev_disks
== map
->num_members
)
6988 /* OK, this array needs to enter reshape mode.
6989 * i.e it needs a migr_state
6992 copy_map_size
= sizeof_imsm_map(map
);
6993 prev_num_members
= map
->num_members
;
6994 map
->num_members
= prev_disks
;
6995 dev
->vol
.migr_state
= 1;
6996 dev
->vol
.curr_migr_unit
= 0;
6997 set_migr_type(dev
, MIGR_GEN_MIGR
);
6998 for (i
= prev_num_members
;
6999 i
< map
->num_members
; i
++)
7000 set_imsm_ord_tbl_ent(map
, i
, i
);
7001 map2
= get_imsm_map(dev
, MAP_1
);
7002 /* Copy the current map */
7003 memcpy(map2
, map
, copy_map_size
);
7004 map2
->num_members
= prev_num_members
;
7006 imsm_set_array_size(dev
);
7007 super
->clean_migration_record_by_mdmon
= 1;
7008 super
->updates_pending
++;
7012 /* Handle dirty -> clean transititions, resync and reshape. Degraded and rebuild
7013 * states are handled in imsm_set_disk() with one exception, when a
7014 * resync is stopped due to a new failure this routine will set the
7015 * 'degraded' state for the array.
7017 static int imsm_set_array_state(struct active_array
*a
, int consistent
)
7019 int inst
= a
->info
.container_member
;
7020 struct intel_super
*super
= a
->container
->sb
;
7021 struct imsm_dev
*dev
= get_imsm_dev(super
, inst
);
7022 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
7023 int failed
= imsm_count_failed(super
, dev
, MAP_0
);
7024 __u8 map_state
= imsm_check_degraded(super
, dev
, failed
, MAP_0
);
7025 __u32 blocks_per_unit
;
7027 if (dev
->vol
.migr_state
&&
7028 dev
->vol
.migr_type
== MIGR_GEN_MIGR
) {
7029 /* array state change is blocked due to reshape action
7031 * - abort the reshape (if last_checkpoint is 0 and action!= reshape)
7032 * - finish the reshape (if last_checkpoint is big and action != reshape)
7033 * - update curr_migr_unit
7035 if (a
->curr_action
== reshape
) {
7036 /* still reshaping, maybe update curr_migr_unit */
7037 goto mark_checkpoint
;
7039 if (a
->last_checkpoint
== 0 && a
->prev_action
== reshape
) {
7040 /* for some reason we aborted the reshape.
7042 * disable automatic metadata rollback
7043 * user action is required to recover process
7046 struct imsm_map
*map2
=
7047 get_imsm_map(dev
, MAP_1
);
7048 dev
->vol
.migr_state
= 0;
7049 set_migr_type(dev
, 0);
7050 dev
->vol
.curr_migr_unit
= 0;
7052 sizeof_imsm_map(map2
));
7053 super
->updates_pending
++;
7056 if (a
->last_checkpoint
>= a
->info
.component_size
) {
7057 unsigned long long array_blocks
;
7061 used_disks
= imsm_num_data_members(dev
, MAP_0
);
7062 if (used_disks
> 0) {
7064 blocks_per_member(map
) *
7066 /* round array size down to closest MB
7068 array_blocks
= (array_blocks
7069 >> SECT_PER_MB_SHIFT
)
7070 << SECT_PER_MB_SHIFT
;
7071 a
->info
.custom_array_size
= array_blocks
;
7072 /* encourage manager to update array
7076 a
->check_reshape
= 1;
7078 /* finalize online capacity expansion/reshape */
7079 for (mdi
= a
->info
.devs
; mdi
; mdi
= mdi
->next
)
7081 mdi
->disk
.raid_disk
,
7084 imsm_progress_container_reshape(super
);
7089 /* before we activate this array handle any missing disks */
7090 if (consistent
== 2)
7091 handle_missing(super
, dev
);
7093 if (consistent
== 2 &&
7094 (!is_resync_complete(&a
->info
) ||
7095 map_state
!= IMSM_T_STATE_NORMAL
||
7096 dev
->vol
.migr_state
))
7099 if (is_resync_complete(&a
->info
)) {
7100 /* complete intialization / resync,
7101 * recovery and interrupted recovery is completed in
7104 if (is_resyncing(dev
)) {
7105 dprintf("imsm: mark resync done\n");
7106 end_migration(dev
, super
, map_state
);
7107 super
->updates_pending
++;
7108 a
->last_checkpoint
= 0;
7110 } else if ((!is_resyncing(dev
) && !failed
) &&
7111 (imsm_reshape_blocks_arrays_changes(super
) == 0)) {
7112 /* mark the start of the init process if nothing is failed */
7113 dprintf("imsm: mark resync start\n");
7114 if (map
->map_state
== IMSM_T_STATE_UNINITIALIZED
)
7115 migrate(dev
, super
, IMSM_T_STATE_NORMAL
, MIGR_INIT
);
7117 migrate(dev
, super
, IMSM_T_STATE_NORMAL
, MIGR_REPAIR
);
7118 super
->updates_pending
++;
7122 /* skip checkpointing for general migration,
7123 * it is controlled in mdadm
7125 if (is_gen_migration(dev
))
7126 goto skip_mark_checkpoint
;
7128 /* check if we can update curr_migr_unit from resync_start, recovery_start */
7129 blocks_per_unit
= blocks_per_migr_unit(super
, dev
);
7130 if (blocks_per_unit
) {
7134 units
= a
->last_checkpoint
/ blocks_per_unit
;
7137 /* check that we did not overflow 32-bits, and that
7138 * curr_migr_unit needs updating
7140 if (units32
== units
&&
7142 __le32_to_cpu(dev
->vol
.curr_migr_unit
) != units32
) {
7143 dprintf("imsm: mark checkpoint (%u)\n", units32
);
7144 dev
->vol
.curr_migr_unit
= __cpu_to_le32(units32
);
7145 super
->updates_pending
++;
7149 skip_mark_checkpoint
:
7150 /* mark dirty / clean */
7151 if (dev
->vol
.dirty
!= !consistent
) {
7152 dprintf("imsm: mark '%s'\n", consistent
? "clean" : "dirty");
7157 super
->updates_pending
++;
7163 static void imsm_set_disk(struct active_array
*a
, int n
, int state
)
7165 int inst
= a
->info
.container_member
;
7166 struct intel_super
*super
= a
->container
->sb
;
7167 struct imsm_dev
*dev
= get_imsm_dev(super
, inst
);
7168 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
7169 struct imsm_disk
*disk
;
7174 if (n
> map
->num_members
)
7175 fprintf(stderr
, "imsm: set_disk %d out of range 0..%d\n",
7176 n
, map
->num_members
- 1);
7181 dprintf("imsm: set_disk %d:%x\n", n
, state
);
7183 ord
= get_imsm_ord_tbl_ent(dev
, n
, MAP_0
);
7184 disk
= get_imsm_disk(super
, ord_to_idx(ord
));
7186 /* check for new failures */
7187 if (state
& DS_FAULTY
) {
7188 if (mark_failure(dev
, disk
, ord_to_idx(ord
)))
7189 super
->updates_pending
++;
7192 /* check if in_sync */
7193 if (state
& DS_INSYNC
&& ord
& IMSM_ORD_REBUILD
&& is_rebuilding(dev
)) {
7194 struct imsm_map
*migr_map
= get_imsm_map(dev
, MAP_1
);
7196 set_imsm_ord_tbl_ent(migr_map
, n
, ord_to_idx(ord
));
7197 super
->updates_pending
++;
7200 failed
= imsm_count_failed(super
, dev
, MAP_0
);
7201 map_state
= imsm_check_degraded(super
, dev
, failed
, MAP_0
);
7203 /* check if recovery complete, newly degraded, or failed */
7204 dprintf("imsm: Detected transition to state ");
7205 switch (map_state
) {
7206 case IMSM_T_STATE_NORMAL
: /* transition to normal state */
7207 dprintf("normal: ");
7208 if (is_rebuilding(dev
)) {
7209 dprintf("while rebuilding");
7210 end_migration(dev
, super
, map_state
);
7211 map
= get_imsm_map(dev
, MAP_0
);
7212 map
->failed_disk_num
= ~0;
7213 super
->updates_pending
++;
7214 a
->last_checkpoint
= 0;
7217 if (is_gen_migration(dev
)) {
7218 dprintf("while general migration");
7219 if (a
->last_checkpoint
>= a
->info
.component_size
)
7220 end_migration(dev
, super
, map_state
);
7222 map
->map_state
= map_state
;
7223 map
= get_imsm_map(dev
, MAP_0
);
7224 map
->failed_disk_num
= ~0;
7225 super
->updates_pending
++;
7229 case IMSM_T_STATE_DEGRADED
: /* transition to degraded state */
7230 dprintf("degraded: ");
7231 if ((map
->map_state
!= map_state
) &&
7232 !dev
->vol
.migr_state
) {
7233 dprintf("mark degraded");
7234 map
->map_state
= map_state
;
7235 super
->updates_pending
++;
7236 a
->last_checkpoint
= 0;
7239 if (is_rebuilding(dev
)) {
7240 dprintf("while rebuilding.");
7241 if (map
->map_state
!= map_state
) {
7242 dprintf(" Map state change");
7243 end_migration(dev
, super
, map_state
);
7244 super
->updates_pending
++;
7248 if (is_gen_migration(dev
)) {
7249 dprintf("while general migration");
7250 if (a
->last_checkpoint
>= a
->info
.component_size
)
7251 end_migration(dev
, super
, map_state
);
7253 map
->map_state
= map_state
;
7254 manage_second_map(super
, dev
);
7256 super
->updates_pending
++;
7259 if (is_initializing(dev
)) {
7260 dprintf("while initialization.");
7261 map
->map_state
= map_state
;
7262 super
->updates_pending
++;
7266 case IMSM_T_STATE_FAILED
: /* transition to failed state */
7267 dprintf("failed: ");
7268 if (is_gen_migration(dev
)) {
7269 dprintf("while general migration");
7270 map
->map_state
= map_state
;
7271 super
->updates_pending
++;
7274 if (map
->map_state
!= map_state
) {
7275 dprintf("mark failed");
7276 end_migration(dev
, super
, map_state
);
7277 super
->updates_pending
++;
7278 a
->last_checkpoint
= 0;
7283 dprintf("state %i\n", map_state
);
7289 static int store_imsm_mpb(int fd
, struct imsm_super
*mpb
)
7292 __u32 mpb_size
= __le32_to_cpu(mpb
->mpb_size
);
7293 unsigned long long dsize
;
7294 unsigned long long sectors
;
7296 get_dev_size(fd
, NULL
, &dsize
);
7298 if (mpb_size
> 512) {
7299 /* -1 to account for anchor */
7300 sectors
= mpb_sectors(mpb
) - 1;
7302 /* write the extended mpb to the sectors preceeding the anchor */
7303 if (lseek64(fd
, dsize
- (512 * (2 + sectors
)), SEEK_SET
) < 0)
7306 if ((unsigned long long)write(fd
, buf
+ 512, 512 * sectors
)
7311 /* first block is stored on second to last sector of the disk */
7312 if (lseek64(fd
, dsize
- (512 * 2), SEEK_SET
) < 0)
7315 if (write(fd
, buf
, 512) != 512)
7321 static void imsm_sync_metadata(struct supertype
*container
)
7323 struct intel_super
*super
= container
->sb
;
7325 dprintf("sync metadata: %d\n", super
->updates_pending
);
7326 if (!super
->updates_pending
)
7329 write_super_imsm(container
, 0);
7331 super
->updates_pending
= 0;
7334 static struct dl
*imsm_readd(struct intel_super
*super
, int idx
, struct active_array
*a
)
7336 struct imsm_dev
*dev
= get_imsm_dev(super
, a
->info
.container_member
);
7337 int i
= get_imsm_disk_idx(dev
, idx
, MAP_X
);
7340 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
7344 if (dl
&& is_failed(&dl
->disk
))
7348 dprintf("%s: found %x:%x\n", __func__
, dl
->major
, dl
->minor
);
7353 static struct dl
*imsm_add_spare(struct intel_super
*super
, int slot
,
7354 struct active_array
*a
, int activate_new
,
7355 struct mdinfo
*additional_test_list
)
7357 struct imsm_dev
*dev
= get_imsm_dev(super
, a
->info
.container_member
);
7358 int idx
= get_imsm_disk_idx(dev
, slot
, MAP_X
);
7359 struct imsm_super
*mpb
= super
->anchor
;
7360 struct imsm_map
*map
;
7361 unsigned long long pos
;
7366 __u32 array_start
= 0;
7367 __u32 array_end
= 0;
7369 struct mdinfo
*test_list
;
7371 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
7372 /* If in this array, skip */
7373 for (d
= a
->info
.devs
; d
; d
= d
->next
)
7374 if (d
->state_fd
>= 0 &&
7375 d
->disk
.major
== dl
->major
&&
7376 d
->disk
.minor
== dl
->minor
) {
7377 dprintf("%x:%x already in array\n",
7378 dl
->major
, dl
->minor
);
7383 test_list
= additional_test_list
;
7385 if (test_list
->disk
.major
== dl
->major
&&
7386 test_list
->disk
.minor
== dl
->minor
) {
7387 dprintf("%x:%x already in additional test list\n",
7388 dl
->major
, dl
->minor
);
7391 test_list
= test_list
->next
;
7396 /* skip in use or failed drives */
7397 if (is_failed(&dl
->disk
) || idx
== dl
->index
||
7399 dprintf("%x:%x status (failed: %d index: %d)\n",
7400 dl
->major
, dl
->minor
, is_failed(&dl
->disk
), idx
);
7404 /* skip pure spares when we are looking for partially
7405 * assimilated drives
7407 if (dl
->index
== -1 && !activate_new
)
7410 /* Does this unused device have the requisite free space?
7411 * It needs to be able to cover all member volumes
7413 ex
= get_extents(super
, dl
);
7415 dprintf("cannot get extents\n");
7418 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
7419 dev
= get_imsm_dev(super
, i
);
7420 map
= get_imsm_map(dev
, MAP_0
);
7422 /* check if this disk is already a member of
7425 if (get_imsm_disk_slot(map
, dl
->index
) >= 0)
7431 array_start
= pba_of_lba0(map
);
7432 array_end
= array_start
+
7433 blocks_per_member(map
) - 1;
7436 /* check that we can start at pba_of_lba0 with
7437 * blocks_per_member of space
7439 if (array_start
>= pos
&& array_end
< ex
[j
].start
) {
7443 pos
= ex
[j
].start
+ ex
[j
].size
;
7445 } while (ex
[j
-1].size
);
7452 if (i
< mpb
->num_raid_devs
) {
7453 dprintf("%x:%x does not have %u to %u available\n",
7454 dl
->major
, dl
->minor
, array_start
, array_end
);
7465 static int imsm_rebuild_allowed(struct supertype
*cont
, int dev_idx
, int failed
)
7467 struct imsm_dev
*dev2
;
7468 struct imsm_map
*map
;
7474 dev2
= get_imsm_dev(cont
->sb
, dev_idx
);
7476 state
= imsm_check_degraded(cont
->sb
, dev2
, failed
, MAP_0
);
7477 if (state
== IMSM_T_STATE_FAILED
) {
7478 map
= get_imsm_map(dev2
, MAP_0
);
7481 for (slot
= 0; slot
< map
->num_members
; slot
++) {
7483 * Check if failed disks are deleted from intel
7484 * disk list or are marked to be deleted
7486 idx
= get_imsm_disk_idx(dev2
, slot
, MAP_X
);
7487 idisk
= get_imsm_dl_disk(cont
->sb
, idx
);
7489 * Do not rebuild the array if failed disks
7490 * from failed sub-array are not removed from
7494 is_failed(&idisk
->disk
) &&
7495 (idisk
->action
!= DISK_REMOVE
))
7503 static struct mdinfo
*imsm_activate_spare(struct active_array
*a
,
7504 struct metadata_update
**updates
)
7507 * Find a device with unused free space and use it to replace a
7508 * failed/vacant region in an array. We replace failed regions one a
7509 * array at a time. The result is that a new spare disk will be added
7510 * to the first failed array and after the monitor has finished
7511 * propagating failures the remainder will be consumed.
7513 * FIXME add a capability for mdmon to request spares from another
7517 struct intel_super
*super
= a
->container
->sb
;
7518 int inst
= a
->info
.container_member
;
7519 struct imsm_dev
*dev
= get_imsm_dev(super
, inst
);
7520 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
7521 int failed
= a
->info
.array
.raid_disks
;
7522 struct mdinfo
*rv
= NULL
;
7525 struct metadata_update
*mu
;
7527 struct imsm_update_activate_spare
*u
;
7532 for (d
= a
->info
.devs
; d
; d
= d
->next
) {
7533 if ((d
->curr_state
& DS_FAULTY
) &&
7535 /* wait for Removal to happen */
7537 if (d
->state_fd
>= 0)
7541 dprintf("imsm: activate spare: inst=%d failed=%d (%d) level=%d\n",
7542 inst
, failed
, a
->info
.array
.raid_disks
, a
->info
.array
.level
);
7544 if (imsm_reshape_blocks_arrays_changes(super
))
7547 /* Cannot activate another spare if rebuild is in progress already
7549 if (is_rebuilding(dev
)) {
7550 dprintf("imsm: No spare activation allowed. "
7551 "Rebuild in progress already.\n");
7555 if (a
->info
.array
.level
== 4)
7556 /* No repair for takeovered array
7557 * imsm doesn't support raid4
7561 if (imsm_check_degraded(super
, dev
, failed
, MAP_0
) !=
7562 IMSM_T_STATE_DEGRADED
)
7566 * If there are any failed disks check state of the other volume.
7567 * Block rebuild if the another one is failed until failed disks
7568 * are removed from container.
7571 dprintf("found failed disks in %.*s, check if there another"
7572 "failed sub-array.\n",
7573 MAX_RAID_SERIAL_LEN
, dev
->volume
);
7574 /* check if states of the other volumes allow for rebuild */
7575 for (i
= 0; i
< super
->anchor
->num_raid_devs
; i
++) {
7577 allowed
= imsm_rebuild_allowed(a
->container
,
7585 /* For each slot, if it is not working, find a spare */
7586 for (i
= 0; i
< a
->info
.array
.raid_disks
; i
++) {
7587 for (d
= a
->info
.devs
; d
; d
= d
->next
)
7588 if (d
->disk
.raid_disk
== i
)
7590 dprintf("found %d: %p %x\n", i
, d
, d
?d
->curr_state
:0);
7591 if (d
&& (d
->state_fd
>= 0))
7595 * OK, this device needs recovery. Try to re-add the
7596 * previous occupant of this slot, if this fails see if
7597 * we can continue the assimilation of a spare that was
7598 * partially assimilated, finally try to activate a new
7601 dl
= imsm_readd(super
, i
, a
);
7603 dl
= imsm_add_spare(super
, i
, a
, 0, rv
);
7605 dl
= imsm_add_spare(super
, i
, a
, 1, rv
);
7609 /* found a usable disk with enough space */
7610 di
= malloc(sizeof(*di
));
7613 memset(di
, 0, sizeof(*di
));
7615 /* dl->index will be -1 in the case we are activating a
7616 * pristine spare. imsm_process_update() will create a
7617 * new index in this case. Once a disk is found to be
7618 * failed in all member arrays it is kicked from the
7621 di
->disk
.number
= dl
->index
;
7623 /* (ab)use di->devs to store a pointer to the device
7626 di
->devs
= (struct mdinfo
*) dl
;
7628 di
->disk
.raid_disk
= i
;
7629 di
->disk
.major
= dl
->major
;
7630 di
->disk
.minor
= dl
->minor
;
7632 di
->recovery_start
= 0;
7633 di
->data_offset
= pba_of_lba0(map
);
7634 di
->component_size
= a
->info
.component_size
;
7635 di
->container_member
= inst
;
7636 super
->random
= random32();
7640 dprintf("%x:%x to be %d at %llu\n", dl
->major
, dl
->minor
,
7641 i
, di
->data_offset
);
7645 /* No spares found */
7647 /* Now 'rv' has a list of devices to return.
7648 * Create a metadata_update record to update the
7649 * disk_ord_tbl for the array
7651 mu
= malloc(sizeof(*mu
));
7653 mu
->buf
= malloc(sizeof(struct imsm_update_activate_spare
) * num_spares
);
7654 if (mu
->buf
== NULL
) {
7661 struct mdinfo
*n
= rv
->next
;
7670 mu
->space_list
= NULL
;
7671 mu
->len
= sizeof(struct imsm_update_activate_spare
) * num_spares
;
7672 mu
->next
= *updates
;
7673 u
= (struct imsm_update_activate_spare
*) mu
->buf
;
7675 for (di
= rv
; di
; di
= di
->next
) {
7676 u
->type
= update_activate_spare
;
7677 u
->dl
= (struct dl
*) di
->devs
;
7679 u
->slot
= di
->disk
.raid_disk
;
7690 static int disks_overlap(struct intel_super
*super
, int idx
, struct imsm_update_create_array
*u
)
7692 struct imsm_dev
*dev
= get_imsm_dev(super
, idx
);
7693 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
7694 struct imsm_map
*new_map
= get_imsm_map(&u
->dev
, MAP_0
);
7695 struct disk_info
*inf
= get_disk_info(u
);
7696 struct imsm_disk
*disk
;
7700 for (i
= 0; i
< map
->num_members
; i
++) {
7701 disk
= get_imsm_disk(super
, get_imsm_disk_idx(dev
, i
, MAP_X
));
7702 for (j
= 0; j
< new_map
->num_members
; j
++)
7703 if (serialcmp(disk
->serial
, inf
[j
].serial
) == 0)
7711 static struct dl
*get_disk_super(struct intel_super
*super
, int major
, int minor
)
7713 struct dl
*dl
= NULL
;
7714 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
7715 if ((dl
->major
== major
) && (dl
->minor
== minor
))
7720 static int remove_disk_super(struct intel_super
*super
, int major
, int minor
)
7722 struct dl
*prev
= NULL
;
7726 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
7727 if ((dl
->major
== major
) && (dl
->minor
== minor
)) {
7730 prev
->next
= dl
->next
;
7732 super
->disks
= dl
->next
;
7734 __free_imsm_disk(dl
);
7735 dprintf("%s: removed %x:%x\n",
7736 __func__
, major
, minor
);
7744 static void imsm_delete(struct intel_super
*super
, struct dl
**dlp
, unsigned index
);
7746 static int add_remove_disk_update(struct intel_super
*super
)
7748 int check_degraded
= 0;
7749 struct dl
*disk
= NULL
;
7750 /* add/remove some spares to/from the metadata/contrainer */
7751 while (super
->disk_mgmt_list
) {
7752 struct dl
*disk_cfg
;
7754 disk_cfg
= super
->disk_mgmt_list
;
7755 super
->disk_mgmt_list
= disk_cfg
->next
;
7756 disk_cfg
->next
= NULL
;
7758 if (disk_cfg
->action
== DISK_ADD
) {
7759 disk_cfg
->next
= super
->disks
;
7760 super
->disks
= disk_cfg
;
7762 dprintf("%s: added %x:%x\n",
7763 __func__
, disk_cfg
->major
,
7765 } else if (disk_cfg
->action
== DISK_REMOVE
) {
7766 dprintf("Disk remove action processed: %x.%x\n",
7767 disk_cfg
->major
, disk_cfg
->minor
);
7768 disk
= get_disk_super(super
,
7772 /* store action status */
7773 disk
->action
= DISK_REMOVE
;
7774 /* remove spare disks only */
7775 if (disk
->index
== -1) {
7776 remove_disk_super(super
,
7781 /* release allocate disk structure */
7782 __free_imsm_disk(disk_cfg
);
7785 return check_degraded
;
7789 static int apply_reshape_migration_update(struct imsm_update_reshape_migration
*u
,
7790 struct intel_super
*super
,
7793 struct intel_dev
*id
;
7794 void **tofree
= NULL
;
7797 dprintf("apply_reshape_migration_update()\n");
7798 if ((u
->subdev
< 0) ||
7800 dprintf("imsm: Error: Wrong subdev: %i\n", u
->subdev
);
7803 if ((space_list
== NULL
) || (*space_list
== NULL
)) {
7804 dprintf("imsm: Error: Memory is not allocated\n");
7808 for (id
= super
->devlist
; id
; id
= id
->next
) {
7809 if (id
->index
== (unsigned)u
->subdev
) {
7810 struct imsm_dev
*dev
= get_imsm_dev(super
, u
->subdev
);
7811 struct imsm_map
*map
;
7812 struct imsm_dev
*new_dev
=
7813 (struct imsm_dev
*)*space_list
;
7814 struct imsm_map
*migr_map
= get_imsm_map(dev
, MAP_1
);
7816 struct dl
*new_disk
;
7818 if (new_dev
== NULL
)
7820 *space_list
= **space_list
;
7821 memcpy(new_dev
, dev
, sizeof_imsm_dev(dev
, 0));
7822 map
= get_imsm_map(new_dev
, MAP_0
);
7824 dprintf("imsm: Error: migration in progress");
7828 to_state
= map
->map_state
;
7829 if ((u
->new_level
== 5) && (map
->raid_level
== 0)) {
7831 /* this should not happen */
7832 if (u
->new_disks
[0] < 0) {
7833 map
->failed_disk_num
=
7834 map
->num_members
- 1;
7835 to_state
= IMSM_T_STATE_DEGRADED
;
7837 to_state
= IMSM_T_STATE_NORMAL
;
7839 migrate(new_dev
, super
, to_state
, MIGR_GEN_MIGR
);
7840 if (u
->new_level
> -1)
7841 map
->raid_level
= u
->new_level
;
7842 migr_map
= get_imsm_map(new_dev
, MAP_1
);
7843 if ((u
->new_level
== 5) &&
7844 (migr_map
->raid_level
== 0)) {
7845 int ord
= map
->num_members
- 1;
7846 migr_map
->num_members
--;
7847 if (u
->new_disks
[0] < 0)
7848 ord
|= IMSM_ORD_REBUILD
;
7849 set_imsm_ord_tbl_ent(map
,
7850 map
->num_members
- 1,
7854 tofree
= (void **)dev
;
7856 /* update chunk size
7858 if (u
->new_chunksize
> 0)
7859 map
->blocks_per_strip
=
7860 __cpu_to_le16(u
->new_chunksize
* 2);
7864 if ((u
->new_level
!= 5) ||
7865 (migr_map
->raid_level
!= 0) ||
7866 (migr_map
->raid_level
== map
->raid_level
))
7869 if (u
->new_disks
[0] >= 0) {
7872 new_disk
= get_disk_super(super
,
7873 major(u
->new_disks
[0]),
7874 minor(u
->new_disks
[0]));
7875 dprintf("imsm: new disk for reshape is: %i:%i "
7876 "(%p, index = %i)\n",
7877 major(u
->new_disks
[0]),
7878 minor(u
->new_disks
[0]),
7879 new_disk
, new_disk
->index
);
7880 if (new_disk
== NULL
)
7881 goto error_disk_add
;
7883 new_disk
->index
= map
->num_members
- 1;
7884 /* slot to fill in autolayout
7886 new_disk
->raiddisk
= new_disk
->index
;
7887 new_disk
->disk
.status
|= CONFIGURED_DISK
;
7888 new_disk
->disk
.status
&= ~SPARE_DISK
;
7890 goto error_disk_add
;
7893 *tofree
= *space_list
;
7894 /* calculate new size
7896 imsm_set_array_size(new_dev
);
7903 *space_list
= tofree
;
7907 dprintf("Error: imsm: Cannot find disk.\n");
7911 static int apply_update_activate_spare(struct imsm_update_activate_spare
*u
,
7912 struct intel_super
*super
,
7913 struct active_array
*active_array
)
7915 struct imsm_super
*mpb
= super
->anchor
;
7916 struct imsm_dev
*dev
= get_imsm_dev(super
, u
->array
);
7917 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
7918 struct imsm_map
*migr_map
;
7919 struct active_array
*a
;
7920 struct imsm_disk
*disk
;
7927 int second_map_created
= 0;
7929 for (; u
; u
= u
->next
) {
7930 victim
= get_imsm_disk_idx(dev
, u
->slot
, MAP_X
);
7935 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
7940 fprintf(stderr
, "error: imsm_activate_spare passed "
7941 "an unknown disk (index: %d)\n",
7946 /* count failures (excluding rebuilds and the victim)
7947 * to determine map[0] state
7950 for (i
= 0; i
< map
->num_members
; i
++) {
7953 disk
= get_imsm_disk(super
,
7954 get_imsm_disk_idx(dev
, i
, MAP_X
));
7955 if (!disk
|| is_failed(disk
))
7959 /* adding a pristine spare, assign a new index */
7960 if (dl
->index
< 0) {
7961 dl
->index
= super
->anchor
->num_disks
;
7962 super
->anchor
->num_disks
++;
7965 disk
->status
|= CONFIGURED_DISK
;
7966 disk
->status
&= ~SPARE_DISK
;
7969 to_state
= imsm_check_degraded(super
, dev
, failed
, MAP_0
);
7970 if (!second_map_created
) {
7971 second_map_created
= 1;
7972 map
->map_state
= IMSM_T_STATE_DEGRADED
;
7973 migrate(dev
, super
, to_state
, MIGR_REBUILD
);
7975 map
->map_state
= to_state
;
7976 migr_map
= get_imsm_map(dev
, MAP_1
);
7977 set_imsm_ord_tbl_ent(map
, u
->slot
, dl
->index
);
7978 set_imsm_ord_tbl_ent(migr_map
, u
->slot
,
7979 dl
->index
| IMSM_ORD_REBUILD
);
7981 /* update the family_num to mark a new container
7982 * generation, being careful to record the existing
7983 * family_num in orig_family_num to clean up after
7984 * earlier mdadm versions that neglected to set it.
7986 if (mpb
->orig_family_num
== 0)
7987 mpb
->orig_family_num
= mpb
->family_num
;
7988 mpb
->family_num
+= super
->random
;
7990 /* count arrays using the victim in the metadata */
7992 for (a
= active_array
; a
; a
= a
->next
) {
7993 dev
= get_imsm_dev(super
, a
->info
.container_member
);
7994 map
= get_imsm_map(dev
, MAP_0
);
7996 if (get_imsm_disk_slot(map
, victim
) >= 0)
8000 /* delete the victim if it is no longer being
8006 /* We know that 'manager' isn't touching anything,
8007 * so it is safe to delete
8009 for (dlp
= &super
->disks
; *dlp
; dlp
= &(*dlp
)->next
)
8010 if ((*dlp
)->index
== victim
)
8013 /* victim may be on the missing list */
8015 for (dlp
= &super
->missing
; *dlp
;
8016 dlp
= &(*dlp
)->next
)
8017 if ((*dlp
)->index
== victim
)
8019 imsm_delete(super
, dlp
, victim
);
8026 static int apply_reshape_container_disks_update(struct imsm_update_reshape
*u
,
8027 struct intel_super
*super
,
8030 struct dl
*new_disk
;
8031 struct intel_dev
*id
;
8033 int delta_disks
= u
->new_raid_disks
- u
->old_raid_disks
;
8034 int disk_count
= u
->old_raid_disks
;
8035 void **tofree
= NULL
;
8036 int devices_to_reshape
= 1;
8037 struct imsm_super
*mpb
= super
->anchor
;
8039 unsigned int dev_id
;
8041 dprintf("imsm: apply_reshape_container_disks_update()\n");
8043 /* enable spares to use in array */
8044 for (i
= 0; i
< delta_disks
; i
++) {
8045 new_disk
= get_disk_super(super
,
8046 major(u
->new_disks
[i
]),
8047 minor(u
->new_disks
[i
]));
8048 dprintf("imsm: new disk for reshape is: %i:%i "
8049 "(%p, index = %i)\n",
8050 major(u
->new_disks
[i
]), minor(u
->new_disks
[i
]),
8051 new_disk
, new_disk
->index
);
8052 if ((new_disk
== NULL
) ||
8053 ((new_disk
->index
>= 0) &&
8054 (new_disk
->index
< u
->old_raid_disks
)))
8055 goto update_reshape_exit
;
8056 new_disk
->index
= disk_count
++;
8057 /* slot to fill in autolayout
8059 new_disk
->raiddisk
= new_disk
->index
;
8060 new_disk
->disk
.status
|=
8062 new_disk
->disk
.status
&= ~SPARE_DISK
;
8065 dprintf("imsm: volume set mpb->num_raid_devs = %i\n",
8066 mpb
->num_raid_devs
);
8067 /* manage changes in volume
8069 for (dev_id
= 0; dev_id
< mpb
->num_raid_devs
; dev_id
++) {
8070 void **sp
= *space_list
;
8071 struct imsm_dev
*newdev
;
8072 struct imsm_map
*newmap
, *oldmap
;
8074 for (id
= super
->devlist
; id
; id
= id
->next
) {
8075 if (id
->index
== dev_id
)
8084 /* Copy the dev, but not (all of) the map */
8085 memcpy(newdev
, id
->dev
, sizeof(*newdev
));
8086 oldmap
= get_imsm_map(id
->dev
, MAP_0
);
8087 newmap
= get_imsm_map(newdev
, MAP_0
);
8088 /* Copy the current map */
8089 memcpy(newmap
, oldmap
, sizeof_imsm_map(oldmap
));
8090 /* update one device only
8092 if (devices_to_reshape
) {
8093 dprintf("imsm: modifying subdev: %i\n",
8095 devices_to_reshape
--;
8096 newdev
->vol
.migr_state
= 1;
8097 newdev
->vol
.curr_migr_unit
= 0;
8098 set_migr_type(newdev
, MIGR_GEN_MIGR
);
8099 newmap
->num_members
= u
->new_raid_disks
;
8100 for (i
= 0; i
< delta_disks
; i
++) {
8101 set_imsm_ord_tbl_ent(newmap
,
8102 u
->old_raid_disks
+ i
,
8103 u
->old_raid_disks
+ i
);
8105 /* New map is correct, now need to save old map
8107 newmap
= get_imsm_map(newdev
, MAP_1
);
8108 memcpy(newmap
, oldmap
, sizeof_imsm_map(oldmap
));
8110 imsm_set_array_size(newdev
);
8113 sp
= (void **)id
->dev
;
8118 /* Clear migration record */
8119 memset(super
->migr_rec
, 0, sizeof(struct migr_record
));
8122 *space_list
= tofree
;
8125 update_reshape_exit
:
8130 static int apply_takeover_update(struct imsm_update_takeover
*u
,
8131 struct intel_super
*super
,
8134 struct imsm_dev
*dev
= NULL
;
8135 struct intel_dev
*dv
;
8136 struct imsm_dev
*dev_new
;
8137 struct imsm_map
*map
;
8141 for (dv
= super
->devlist
; dv
; dv
= dv
->next
)
8142 if (dv
->index
== (unsigned int)u
->subarray
) {
8150 map
= get_imsm_map(dev
, MAP_0
);
8152 if (u
->direction
== R10_TO_R0
) {
8153 /* Number of failed disks must be half of initial disk number */
8154 if (imsm_count_failed(super
, dev
, MAP_0
) !=
8155 (map
->num_members
/ 2))
8158 /* iterate through devices to mark removed disks as spare */
8159 for (dm
= super
->disks
; dm
; dm
= dm
->next
) {
8160 if (dm
->disk
.status
& FAILED_DISK
) {
8161 int idx
= dm
->index
;
8162 /* update indexes on the disk list */
8163 /* FIXME this loop-with-the-loop looks wrong, I'm not convinced
8164 the index values will end up being correct.... NB */
8165 for (du
= super
->disks
; du
; du
= du
->next
)
8166 if (du
->index
> idx
)
8168 /* mark as spare disk */
8173 map
->num_members
= map
->num_members
/ 2;
8174 map
->map_state
= IMSM_T_STATE_NORMAL
;
8175 map
->num_domains
= 1;
8176 map
->raid_level
= 0;
8177 map
->failed_disk_num
= -1;
8180 if (u
->direction
== R0_TO_R10
) {
8182 /* update slots in current disk list */
8183 for (dm
= super
->disks
; dm
; dm
= dm
->next
) {
8187 /* create new *missing* disks */
8188 for (i
= 0; i
< map
->num_members
; i
++) {
8189 space
= *space_list
;
8192 *space_list
= *space
;
8194 memcpy(du
, super
->disks
, sizeof(*du
));
8198 du
->index
= (i
* 2) + 1;
8199 sprintf((char *)du
->disk
.serial
,
8200 " MISSING_%d", du
->index
);
8201 sprintf((char *)du
->serial
,
8202 "MISSING_%d", du
->index
);
8203 du
->next
= super
->missing
;
8204 super
->missing
= du
;
8206 /* create new dev and map */
8207 space
= *space_list
;
8210 *space_list
= *space
;
8211 dev_new
= (void *)space
;
8212 memcpy(dev_new
, dev
, sizeof(*dev
));
8213 /* update new map */
8214 map
= get_imsm_map(dev_new
, MAP_0
);
8215 map
->num_members
= map
->num_members
* 2;
8216 map
->map_state
= IMSM_T_STATE_DEGRADED
;
8217 map
->num_domains
= 2;
8218 map
->raid_level
= 1;
8219 /* replace dev<->dev_new */
8222 /* update disk order table */
8223 for (du
= super
->disks
; du
; du
= du
->next
)
8225 set_imsm_ord_tbl_ent(map
, du
->index
, du
->index
);
8226 for (du
= super
->missing
; du
; du
= du
->next
)
8227 if (du
->index
>= 0) {
8228 set_imsm_ord_tbl_ent(map
, du
->index
, du
->index
);
8229 mark_missing(dv
->dev
, &du
->disk
, du
->index
);
8235 static void imsm_process_update(struct supertype
*st
,
8236 struct metadata_update
*update
)
8239 * crack open the metadata_update envelope to find the update record
8240 * update can be one of:
8241 * update_reshape_container_disks - all the arrays in the container
8242 * are being reshaped to have more devices. We need to mark
8243 * the arrays for general migration and convert selected spares
8244 * into active devices.
8245 * update_activate_spare - a spare device has replaced a failed
8246 * device in an array, update the disk_ord_tbl. If this disk is
8247 * present in all member arrays then also clear the SPARE_DISK
8249 * update_create_array
8251 * update_rename_array
8252 * update_add_remove_disk
8254 struct intel_super
*super
= st
->sb
;
8255 struct imsm_super
*mpb
;
8256 enum imsm_update_type type
= *(enum imsm_update_type
*) update
->buf
;
8258 /* update requires a larger buf but the allocation failed */
8259 if (super
->next_len
&& !super
->next_buf
) {
8260 super
->next_len
= 0;
8264 if (super
->next_buf
) {
8265 memcpy(super
->next_buf
, super
->buf
, super
->len
);
8267 super
->len
= super
->next_len
;
8268 super
->buf
= super
->next_buf
;
8270 super
->next_len
= 0;
8271 super
->next_buf
= NULL
;
8274 mpb
= super
->anchor
;
8277 case update_general_migration_checkpoint
: {
8278 struct intel_dev
*id
;
8279 struct imsm_update_general_migration_checkpoint
*u
=
8280 (void *)update
->buf
;
8282 dprintf("imsm: process_update() "
8283 "for update_general_migration_checkpoint called\n");
8285 /* find device under general migration */
8286 for (id
= super
->devlist
; id
; id
= id
->next
) {
8287 if (is_gen_migration(id
->dev
)) {
8288 id
->dev
->vol
.curr_migr_unit
=
8289 __cpu_to_le32(u
->curr_migr_unit
);
8290 super
->updates_pending
++;
8295 case update_takeover
: {
8296 struct imsm_update_takeover
*u
= (void *)update
->buf
;
8297 if (apply_takeover_update(u
, super
, &update
->space_list
)) {
8298 imsm_update_version_info(super
);
8299 super
->updates_pending
++;
8304 case update_reshape_container_disks
: {
8305 struct imsm_update_reshape
*u
= (void *)update
->buf
;
8306 if (apply_reshape_container_disks_update(
8307 u
, super
, &update
->space_list
))
8308 super
->updates_pending
++;
8311 case update_reshape_migration
: {
8312 struct imsm_update_reshape_migration
*u
= (void *)update
->buf
;
8313 if (apply_reshape_migration_update(
8314 u
, super
, &update
->space_list
))
8315 super
->updates_pending
++;
8318 case update_activate_spare
: {
8319 struct imsm_update_activate_spare
*u
= (void *) update
->buf
;
8320 if (apply_update_activate_spare(u
, super
, st
->arrays
))
8321 super
->updates_pending
++;
8324 case update_create_array
: {
8325 /* someone wants to create a new array, we need to be aware of
8326 * a few races/collisions:
8327 * 1/ 'Create' called by two separate instances of mdadm
8328 * 2/ 'Create' versus 'activate_spare': mdadm has chosen
8329 * devices that have since been assimilated via
8331 * In the event this update can not be carried out mdadm will
8332 * (FIX ME) notice that its update did not take hold.
8334 struct imsm_update_create_array
*u
= (void *) update
->buf
;
8335 struct intel_dev
*dv
;
8336 struct imsm_dev
*dev
;
8337 struct imsm_map
*map
, *new_map
;
8338 unsigned long long start
, end
;
8339 unsigned long long new_start
, new_end
;
8341 struct disk_info
*inf
;
8344 /* handle racing creates: first come first serve */
8345 if (u
->dev_idx
< mpb
->num_raid_devs
) {
8346 dprintf("%s: subarray %d already defined\n",
8347 __func__
, u
->dev_idx
);
8351 /* check update is next in sequence */
8352 if (u
->dev_idx
!= mpb
->num_raid_devs
) {
8353 dprintf("%s: can not create array %d expected index %d\n",
8354 __func__
, u
->dev_idx
, mpb
->num_raid_devs
);
8358 new_map
= get_imsm_map(&u
->dev
, MAP_0
);
8359 new_start
= pba_of_lba0(new_map
);
8360 new_end
= new_start
+ blocks_per_member(new_map
);
8361 inf
= get_disk_info(u
);
8363 /* handle activate_spare versus create race:
8364 * check to make sure that overlapping arrays do not include
8367 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
8368 dev
= get_imsm_dev(super
, i
);
8369 map
= get_imsm_map(dev
, MAP_0
);
8370 start
= pba_of_lba0(map
);
8371 end
= start
+ blocks_per_member(map
);
8372 if ((new_start
>= start
&& new_start
<= end
) ||
8373 (start
>= new_start
&& start
<= new_end
))
8378 if (disks_overlap(super
, i
, u
)) {
8379 dprintf("%s: arrays overlap\n", __func__
);
8384 /* check that prepare update was successful */
8385 if (!update
->space
) {
8386 dprintf("%s: prepare update failed\n", __func__
);
8390 /* check that all disks are still active before committing
8391 * changes. FIXME: could we instead handle this by creating a
8392 * degraded array? That's probably not what the user expects,
8393 * so better to drop this update on the floor.
8395 for (i
= 0; i
< new_map
->num_members
; i
++) {
8396 dl
= serial_to_dl(inf
[i
].serial
, super
);
8398 dprintf("%s: disk disappeared\n", __func__
);
8403 super
->updates_pending
++;
8405 /* convert spares to members and fixup ord_tbl */
8406 for (i
= 0; i
< new_map
->num_members
; i
++) {
8407 dl
= serial_to_dl(inf
[i
].serial
, super
);
8408 if (dl
->index
== -1) {
8409 dl
->index
= mpb
->num_disks
;
8411 dl
->disk
.status
|= CONFIGURED_DISK
;
8412 dl
->disk
.status
&= ~SPARE_DISK
;
8414 set_imsm_ord_tbl_ent(new_map
, i
, dl
->index
);
8419 update
->space
= NULL
;
8420 imsm_copy_dev(dev
, &u
->dev
);
8421 dv
->index
= u
->dev_idx
;
8422 dv
->next
= super
->devlist
;
8423 super
->devlist
= dv
;
8424 mpb
->num_raid_devs
++;
8426 imsm_update_version_info(super
);
8429 /* mdmon knows how to release update->space, but not
8430 * ((struct intel_dev *) update->space)->dev
8432 if (update
->space
) {
8438 case update_kill_array
: {
8439 struct imsm_update_kill_array
*u
= (void *) update
->buf
;
8440 int victim
= u
->dev_idx
;
8441 struct active_array
*a
;
8442 struct intel_dev
**dp
;
8443 struct imsm_dev
*dev
;
8445 /* sanity check that we are not affecting the uuid of
8446 * active arrays, or deleting an active array
8448 * FIXME when immutable ids are available, but note that
8449 * we'll also need to fixup the invalidated/active
8450 * subarray indexes in mdstat
8452 for (a
= st
->arrays
; a
; a
= a
->next
)
8453 if (a
->info
.container_member
>= victim
)
8455 /* by definition if mdmon is running at least one array
8456 * is active in the container, so checking
8457 * mpb->num_raid_devs is just extra paranoia
8459 dev
= get_imsm_dev(super
, victim
);
8460 if (a
|| !dev
|| mpb
->num_raid_devs
== 1) {
8461 dprintf("failed to delete subarray-%d\n", victim
);
8465 for (dp
= &super
->devlist
; *dp
;)
8466 if ((*dp
)->index
== (unsigned)super
->current_vol
) {
8469 if ((*dp
)->index
> (unsigned)victim
)
8473 mpb
->num_raid_devs
--;
8474 super
->updates_pending
++;
8477 case update_rename_array
: {
8478 struct imsm_update_rename_array
*u
= (void *) update
->buf
;
8479 char name
[MAX_RAID_SERIAL_LEN
+1];
8480 int target
= u
->dev_idx
;
8481 struct active_array
*a
;
8482 struct imsm_dev
*dev
;
8484 /* sanity check that we are not affecting the uuid of
8487 snprintf(name
, MAX_RAID_SERIAL_LEN
, "%s", (char *) u
->name
);
8488 name
[MAX_RAID_SERIAL_LEN
] = '\0';
8489 for (a
= st
->arrays
; a
; a
= a
->next
)
8490 if (a
->info
.container_member
== target
)
8492 dev
= get_imsm_dev(super
, u
->dev_idx
);
8493 if (a
|| !dev
|| !check_name(super
, name
, 1)) {
8494 dprintf("failed to rename subarray-%d\n", target
);
8498 snprintf((char *) dev
->volume
, MAX_RAID_SERIAL_LEN
, "%s", name
);
8499 super
->updates_pending
++;
8502 case update_add_remove_disk
: {
8503 /* we may be able to repair some arrays if disks are
8504 * being added, check teh status of add_remove_disk
8505 * if discs has been added.
8507 if (add_remove_disk_update(super
)) {
8508 struct active_array
*a
;
8510 super
->updates_pending
++;
8511 for (a
= st
->arrays
; a
; a
= a
->next
)
8512 a
->check_degraded
= 1;
8517 fprintf(stderr
, "error: unsuported process update type:"
8518 "(type: %d)\n", type
);
8522 static struct mdinfo
*get_spares_for_grow(struct supertype
*st
);
8524 static void imsm_prepare_update(struct supertype
*st
,
8525 struct metadata_update
*update
)
8528 * Allocate space to hold new disk entries, raid-device entries or a new
8529 * mpb if necessary. The manager synchronously waits for updates to
8530 * complete in the monitor, so new mpb buffers allocated here can be
8531 * integrated by the monitor thread without worrying about live pointers
8532 * in the manager thread.
8534 enum imsm_update_type type
= *(enum imsm_update_type
*) update
->buf
;
8535 struct intel_super
*super
= st
->sb
;
8536 struct imsm_super
*mpb
= super
->anchor
;
8541 case update_general_migration_checkpoint
:
8542 dprintf("imsm: prepare_update() "
8543 "for update_general_migration_checkpoint called\n");
8545 case update_takeover
: {
8546 struct imsm_update_takeover
*u
= (void *)update
->buf
;
8547 if (u
->direction
== R0_TO_R10
) {
8548 void **tail
= (void **)&update
->space_list
;
8549 struct imsm_dev
*dev
= get_imsm_dev(super
, u
->subarray
);
8550 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
8551 int num_members
= map
->num_members
;
8555 /* allocate memory for added disks */
8556 for (i
= 0; i
< num_members
; i
++) {
8557 size
= sizeof(struct dl
);
8558 space
= malloc(size
);
8567 /* allocate memory for new device */
8568 size
= sizeof_imsm_dev(super
->devlist
->dev
, 0) +
8569 (num_members
* sizeof(__u32
));
8570 space
= malloc(size
);
8579 len
= disks_to_mpb_size(num_members
* 2);
8581 /* if allocation didn't success, free buffer */
8582 while (update
->space_list
) {
8583 void **sp
= update
->space_list
;
8584 update
->space_list
= *sp
;
8592 case update_reshape_container_disks
: {
8593 /* Every raid device in the container is about to
8594 * gain some more devices, and we will enter a
8596 * So each 'imsm_map' will be bigger, and the imsm_vol
8597 * will now hold 2 of them.
8598 * Thus we need new 'struct imsm_dev' allocations sized
8599 * as sizeof_imsm_dev but with more devices in both maps.
8601 struct imsm_update_reshape
*u
= (void *)update
->buf
;
8602 struct intel_dev
*dl
;
8603 void **space_tail
= (void**)&update
->space_list
;
8605 dprintf("imsm: imsm_prepare_update() for update_reshape\n");
8607 for (dl
= super
->devlist
; dl
; dl
= dl
->next
) {
8608 int size
= sizeof_imsm_dev(dl
->dev
, 1);
8610 if (u
->new_raid_disks
> u
->old_raid_disks
)
8611 size
+= sizeof(__u32
)*2*
8612 (u
->new_raid_disks
- u
->old_raid_disks
);
8621 len
= disks_to_mpb_size(u
->new_raid_disks
);
8622 dprintf("New anchor length is %llu\n", (unsigned long long)len
);
8625 case update_reshape_migration
: {
8626 /* for migration level 0->5 we need to add disks
8627 * so the same as for container operation we will copy
8628 * device to the bigger location.
8629 * in memory prepared device and new disk area are prepared
8630 * for usage in process update
8632 struct imsm_update_reshape_migration
*u
= (void *)update
->buf
;
8633 struct intel_dev
*id
;
8634 void **space_tail
= (void **)&update
->space_list
;
8637 int current_level
= -1;
8639 dprintf("imsm: imsm_prepare_update() for update_reshape\n");
8641 /* add space for bigger array in update
8643 for (id
= super
->devlist
; id
; id
= id
->next
) {
8644 if (id
->index
== (unsigned)u
->subdev
) {
8645 size
= sizeof_imsm_dev(id
->dev
, 1);
8646 if (u
->new_raid_disks
> u
->old_raid_disks
)
8647 size
+= sizeof(__u32
)*2*
8648 (u
->new_raid_disks
- u
->old_raid_disks
);
8658 if (update
->space_list
== NULL
)
8661 /* add space for disk in update
8663 size
= sizeof(struct dl
);
8666 free(update
->space_list
);
8667 update
->space_list
= NULL
;
8674 /* add spare device to update
8676 for (id
= super
->devlist
; id
; id
= id
->next
)
8677 if (id
->index
== (unsigned)u
->subdev
) {
8678 struct imsm_dev
*dev
;
8679 struct imsm_map
*map
;
8681 dev
= get_imsm_dev(super
, u
->subdev
);
8682 map
= get_imsm_map(dev
, MAP_0
);
8683 current_level
= map
->raid_level
;
8686 if ((u
->new_level
== 5) && (u
->new_level
!= current_level
)) {
8687 struct mdinfo
*spares
;
8689 spares
= get_spares_for_grow(st
);
8697 makedev(dev
->disk
.major
,
8699 dl
= get_disk_super(super
,
8702 dl
->index
= u
->old_raid_disks
;
8708 len
= disks_to_mpb_size(u
->new_raid_disks
);
8709 dprintf("New anchor length is %llu\n", (unsigned long long)len
);
8712 case update_create_array
: {
8713 struct imsm_update_create_array
*u
= (void *) update
->buf
;
8714 struct intel_dev
*dv
;
8715 struct imsm_dev
*dev
= &u
->dev
;
8716 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
8718 struct disk_info
*inf
;
8722 inf
= get_disk_info(u
);
8723 len
= sizeof_imsm_dev(dev
, 1);
8724 /* allocate a new super->devlist entry */
8725 dv
= malloc(sizeof(*dv
));
8727 dv
->dev
= malloc(len
);
8732 update
->space
= NULL
;
8736 /* count how many spares will be converted to members */
8737 for (i
= 0; i
< map
->num_members
; i
++) {
8738 dl
= serial_to_dl(inf
[i
].serial
, super
);
8740 /* hmm maybe it failed?, nothing we can do about
8745 if (count_memberships(dl
, super
) == 0)
8748 len
+= activate
* sizeof(struct imsm_disk
);
8755 /* check if we need a larger metadata buffer */
8756 if (super
->next_buf
)
8757 buf_len
= super
->next_len
;
8759 buf_len
= super
->len
;
8761 if (__le32_to_cpu(mpb
->mpb_size
) + len
> buf_len
) {
8762 /* ok we need a larger buf than what is currently allocated
8763 * if this allocation fails process_update will notice that
8764 * ->next_len is set and ->next_buf is NULL
8766 buf_len
= ROUND_UP(__le32_to_cpu(mpb
->mpb_size
) + len
, 512);
8767 if (super
->next_buf
)
8768 free(super
->next_buf
);
8770 super
->next_len
= buf_len
;
8771 if (posix_memalign(&super
->next_buf
, 512, buf_len
) == 0)
8772 memset(super
->next_buf
, 0, buf_len
);
8774 super
->next_buf
= NULL
;
8778 /* must be called while manager is quiesced */
8779 static void imsm_delete(struct intel_super
*super
, struct dl
**dlp
, unsigned index
)
8781 struct imsm_super
*mpb
= super
->anchor
;
8783 struct imsm_dev
*dev
;
8784 struct imsm_map
*map
;
8785 int i
, j
, num_members
;
8788 dprintf("%s: deleting device[%d] from imsm_super\n",
8791 /* shift all indexes down one */
8792 for (iter
= super
->disks
; iter
; iter
= iter
->next
)
8793 if (iter
->index
> (int)index
)
8795 for (iter
= super
->missing
; iter
; iter
= iter
->next
)
8796 if (iter
->index
> (int)index
)
8799 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
8800 dev
= get_imsm_dev(super
, i
);
8801 map
= get_imsm_map(dev
, MAP_0
);
8802 num_members
= map
->num_members
;
8803 for (j
= 0; j
< num_members
; j
++) {
8804 /* update ord entries being careful not to propagate
8805 * ord-flags to the first map
8807 ord
= get_imsm_ord_tbl_ent(dev
, j
, MAP_X
);
8809 if (ord_to_idx(ord
) <= index
)
8812 map
= get_imsm_map(dev
, MAP_0
);
8813 set_imsm_ord_tbl_ent(map
, j
, ord_to_idx(ord
- 1));
8814 map
= get_imsm_map(dev
, MAP_1
);
8816 set_imsm_ord_tbl_ent(map
, j
, ord
- 1);
8821 super
->updates_pending
++;
8823 struct dl
*dl
= *dlp
;
8825 *dlp
= (*dlp
)->next
;
8826 __free_imsm_disk(dl
);
8829 #endif /* MDASSEMBLE */
8831 static void close_targets(int *targets
, int new_disks
)
8838 for (i
= 0; i
< new_disks
; i
++) {
8839 if (targets
[i
] >= 0) {
8846 static int imsm_get_allowed_degradation(int level
, int raid_disks
,
8847 struct intel_super
*super
,
8848 struct imsm_dev
*dev
)
8854 struct imsm_map
*map
;
8857 ret_val
= raid_disks
/2;
8858 /* check map if all disks pairs not failed
8861 map
= get_imsm_map(dev
, MAP_0
);
8862 for (i
= 0; i
< ret_val
; i
++) {
8863 int degradation
= 0;
8864 if (get_imsm_disk(super
, i
) == NULL
)
8866 if (get_imsm_disk(super
, i
+ 1) == NULL
)
8868 if (degradation
== 2)
8871 map
= get_imsm_map(dev
, MAP_1
);
8872 /* if there is no second map
8873 * result can be returned
8877 /* check degradation in second map
8879 for (i
= 0; i
< ret_val
; i
++) {
8880 int degradation
= 0;
8881 if (get_imsm_disk(super
, i
) == NULL
)
8883 if (get_imsm_disk(super
, i
+ 1) == NULL
)
8885 if (degradation
== 2)
8900 /*******************************************************************************
8901 * Function: open_backup_targets
8902 * Description: Function opens file descriptors for all devices given in
8905 * info : general array info
8906 * raid_disks : number of disks
8907 * raid_fds : table of device's file descriptors
8908 * super : intel super for raid10 degradation check
8909 * dev : intel device for raid10 degradation check
8913 ******************************************************************************/
8914 int open_backup_targets(struct mdinfo
*info
, int raid_disks
, int *raid_fds
,
8915 struct intel_super
*super
, struct imsm_dev
*dev
)
8921 for (i
= 0; i
< raid_disks
; i
++)
8924 for (sd
= info
->devs
; sd
; sd
= sd
->next
) {
8927 if (sd
->disk
.state
& (1<<MD_DISK_FAULTY
)) {
8928 dprintf("disk is faulty!!\n");
8932 if ((sd
->disk
.raid_disk
>= raid_disks
) ||
8933 (sd
->disk
.raid_disk
< 0))
8936 dn
= map_dev(sd
->disk
.major
,
8938 raid_fds
[sd
->disk
.raid_disk
] = dev_open(dn
, O_RDWR
);
8939 if (raid_fds
[sd
->disk
.raid_disk
] < 0) {
8940 fprintf(stderr
, "cannot open component\n");
8945 /* check if maximum array degradation level is not exceeded
8947 if ((raid_disks
- opened
) >
8948 imsm_get_allowed_degradation(info
->new_level
,
8951 fprintf(stderr
, "Not enough disks can be opened.\n");
8952 close_targets(raid_fds
, raid_disks
);
8959 /*******************************************************************************
8960 * Function: init_migr_record_imsm
8961 * Description: Function inits imsm migration record
8963 * super : imsm internal array info
8964 * dev : device under migration
8965 * info : general array info to find the smallest device
8968 ******************************************************************************/
8969 void init_migr_record_imsm(struct supertype
*st
, struct imsm_dev
*dev
,
8970 struct mdinfo
*info
)
8972 struct intel_super
*super
= st
->sb
;
8973 struct migr_record
*migr_rec
= super
->migr_rec
;
8975 unsigned long long dsize
, dev_sectors
;
8976 long long unsigned min_dev_sectors
= -1LLU;
8980 struct imsm_map
*map_dest
= get_imsm_map(dev
, MAP_0
);
8981 struct imsm_map
*map_src
= get_imsm_map(dev
, MAP_1
);
8982 unsigned long long num_migr_units
;
8983 unsigned long long array_blocks
;
8985 memset(migr_rec
, 0, sizeof(struct migr_record
));
8986 migr_rec
->family_num
= __cpu_to_le32(super
->anchor
->family_num
);
8988 /* only ascending reshape supported now */
8989 migr_rec
->ascending_migr
= __cpu_to_le32(1);
8991 migr_rec
->dest_depth_per_unit
= GEN_MIGR_AREA_SIZE
/
8992 max(map_dest
->blocks_per_strip
, map_src
->blocks_per_strip
);
8993 migr_rec
->dest_depth_per_unit
*=
8994 max(map_dest
->blocks_per_strip
, map_src
->blocks_per_strip
);
8995 new_data_disks
= imsm_num_data_members(dev
, MAP_0
);
8996 migr_rec
->blocks_per_unit
=
8997 __cpu_to_le32(migr_rec
->dest_depth_per_unit
* new_data_disks
);
8998 migr_rec
->dest_depth_per_unit
=
8999 __cpu_to_le32(migr_rec
->dest_depth_per_unit
);
9000 array_blocks
= info
->component_size
* new_data_disks
;
9002 array_blocks
/ __le32_to_cpu(migr_rec
->blocks_per_unit
);
9004 if (array_blocks
% __le32_to_cpu(migr_rec
->blocks_per_unit
))
9006 migr_rec
->num_migr_units
= __cpu_to_le32(num_migr_units
);
9008 migr_rec
->post_migr_vol_cap
= dev
->size_low
;
9009 migr_rec
->post_migr_vol_cap_hi
= dev
->size_high
;
9012 /* Find the smallest dev */
9013 for (sd
= info
->devs
; sd
; sd
= sd
->next
) {
9014 sprintf(nm
, "%d:%d", sd
->disk
.major
, sd
->disk
.minor
);
9015 fd
= dev_open(nm
, O_RDONLY
);
9018 get_dev_size(fd
, NULL
, &dsize
);
9019 dev_sectors
= dsize
/ 512;
9020 if (dev_sectors
< min_dev_sectors
)
9021 min_dev_sectors
= dev_sectors
;
9024 migr_rec
->ckpt_area_pba
= __cpu_to_le32(min_dev_sectors
-
9025 RAID_DISK_RESERVED_BLOCKS_IMSM_HI
);
9027 write_imsm_migr_rec(st
);
9032 /*******************************************************************************
9033 * Function: save_backup_imsm
9034 * Description: Function saves critical data stripes to Migration Copy Area
9035 * and updates the current migration unit status.
9036 * Use restore_stripes() to form a destination stripe,
9037 * and to write it to the Copy Area.
9039 * st : supertype information
9040 * dev : imsm device that backup is saved for
9041 * info : general array info
9042 * buf : input buffer
9043 * length : length of data to backup (blocks_per_unit)
9047 ******************************************************************************/
9048 int save_backup_imsm(struct supertype
*st
,
9049 struct imsm_dev
*dev
,
9050 struct mdinfo
*info
,
9055 struct intel_super
*super
= st
->sb
;
9056 unsigned long long *target_offsets
= NULL
;
9057 int *targets
= NULL
;
9059 struct imsm_map
*map_dest
= get_imsm_map(dev
, MAP_0
);
9060 int new_disks
= map_dest
->num_members
;
9061 int dest_layout
= 0;
9063 unsigned long long start
;
9064 int data_disks
= imsm_num_data_members(dev
, MAP_0
);
9066 targets
= malloc(new_disks
* sizeof(int));
9070 for (i
= 0; i
< new_disks
; i
++)
9073 target_offsets
= malloc(new_disks
* sizeof(unsigned long long));
9074 if (!target_offsets
)
9077 start
= info
->reshape_progress
* 512;
9078 for (i
= 0; i
< new_disks
; i
++) {
9079 target_offsets
[i
] = (unsigned long long)
9080 __le32_to_cpu(super
->migr_rec
->ckpt_area_pba
) * 512;
9081 /* move back copy area adderss, it will be moved forward
9082 * in restore_stripes() using start input variable
9084 target_offsets
[i
] -= start
/data_disks
;
9087 if (open_backup_targets(info
, new_disks
, targets
,
9091 dest_layout
= imsm_level_to_layout(map_dest
->raid_level
);
9092 dest_chunk
= __le16_to_cpu(map_dest
->blocks_per_strip
) * 512;
9094 if (restore_stripes(targets
, /* list of dest devices */
9095 target_offsets
, /* migration record offsets */
9098 map_dest
->raid_level
,
9100 -1, /* source backup file descriptor */
9101 0, /* input buf offset
9102 * always 0 buf is already offseted */
9106 fprintf(stderr
, Name
": Error restoring stripes\n");
9114 close_targets(targets
, new_disks
);
9117 free(target_offsets
);
9122 /*******************************************************************************
9123 * Function: save_checkpoint_imsm
9124 * Description: Function called for current unit status update
9125 * in the migration record. It writes it to disk.
9127 * super : imsm internal array info
9128 * info : general array info
9132 * 2: failure, means no valid migration record
9133 * / no general migration in progress /
9134 ******************************************************************************/
9135 int save_checkpoint_imsm(struct supertype
*st
, struct mdinfo
*info
, int state
)
9137 struct intel_super
*super
= st
->sb
;
9138 unsigned long long blocks_per_unit
;
9139 unsigned long long curr_migr_unit
;
9141 if (load_imsm_migr_rec(super
, info
) != 0) {
9142 dprintf("imsm: ERROR: Cannot read migration record "
9143 "for checkpoint save.\n");
9147 blocks_per_unit
= __le32_to_cpu(super
->migr_rec
->blocks_per_unit
);
9148 if (blocks_per_unit
== 0) {
9149 dprintf("imsm: no migration in progress.\n");
9152 curr_migr_unit
= info
->reshape_progress
/ blocks_per_unit
;
9153 /* check if array is alligned to copy area
9154 * if it is not alligned, add one to current migration unit value
9155 * this can happend on array reshape finish only
9157 if (info
->reshape_progress
% blocks_per_unit
)
9160 super
->migr_rec
->curr_migr_unit
=
9161 __cpu_to_le32(curr_migr_unit
);
9162 super
->migr_rec
->rec_status
= __cpu_to_le32(state
);
9163 super
->migr_rec
->dest_1st_member_lba
=
9164 __cpu_to_le32(curr_migr_unit
*
9165 __le32_to_cpu(super
->migr_rec
->dest_depth_per_unit
));
9166 if (write_imsm_migr_rec(st
) < 0) {
9167 dprintf("imsm: Cannot write migration record "
9168 "outside backup area\n");
9175 /*******************************************************************************
9176 * Function: recover_backup_imsm
9177 * Description: Function recovers critical data from the Migration Copy Area
9178 * while assembling an array.
9180 * super : imsm internal array info
9181 * info : general array info
9183 * 0 : success (or there is no data to recover)
9185 ******************************************************************************/
9186 int recover_backup_imsm(struct supertype
*st
, struct mdinfo
*info
)
9188 struct intel_super
*super
= st
->sb
;
9189 struct migr_record
*migr_rec
= super
->migr_rec
;
9190 struct imsm_map
*map_dest
= NULL
;
9191 struct intel_dev
*id
= NULL
;
9192 unsigned long long read_offset
;
9193 unsigned long long write_offset
;
9195 int *targets
= NULL
;
9196 int new_disks
, i
, err
;
9199 unsigned long curr_migr_unit
= __le32_to_cpu(migr_rec
->curr_migr_unit
);
9200 unsigned long num_migr_units
= __le32_to_cpu(migr_rec
->num_migr_units
);
9202 int skipped_disks
= 0;
9204 err
= sysfs_get_str(info
, NULL
, "array_state", (char *)buffer
, 20);
9208 /* recover data only during assemblation */
9209 if (strncmp(buffer
, "inactive", 8) != 0)
9211 /* no data to recover */
9212 if (__le32_to_cpu(migr_rec
->rec_status
) == UNIT_SRC_NORMAL
)
9214 if (curr_migr_unit
>= num_migr_units
)
9217 /* find device during reshape */
9218 for (id
= super
->devlist
; id
; id
= id
->next
)
9219 if (is_gen_migration(id
->dev
))
9224 map_dest
= get_imsm_map(id
->dev
, MAP_0
);
9225 new_disks
= map_dest
->num_members
;
9227 read_offset
= (unsigned long long)
9228 __le32_to_cpu(migr_rec
->ckpt_area_pba
) * 512;
9230 write_offset
= ((unsigned long long)
9231 __le32_to_cpu(migr_rec
->dest_1st_member_lba
) +
9232 pba_of_lba0(map_dest
)) * 512;
9234 unit_len
= __le32_to_cpu(migr_rec
->dest_depth_per_unit
) * 512;
9235 if (posix_memalign((void **)&buf
, 512, unit_len
) != 0)
9237 targets
= malloc(new_disks
* sizeof(int));
9241 if (open_backup_targets(info
, new_disks
, targets
, super
, id
->dev
)) {
9243 Name
": Cannot open some devices belonging to array.\n");
9247 for (i
= 0; i
< new_disks
; i
++) {
9248 if (targets
[i
] < 0) {
9252 if (lseek64(targets
[i
], read_offset
, SEEK_SET
) < 0) {
9254 Name
": Cannot seek to block: %s\n",
9259 if ((unsigned)read(targets
[i
], buf
, unit_len
) != unit_len
) {
9261 Name
": Cannot read copy area block: %s\n",
9266 if (lseek64(targets
[i
], write_offset
, SEEK_SET
) < 0) {
9268 Name
": Cannot seek to block: %s\n",
9273 if ((unsigned)write(targets
[i
], buf
, unit_len
) != unit_len
) {
9275 Name
": Cannot restore block: %s\n",
9282 if (skipped_disks
> imsm_get_allowed_degradation(info
->new_level
,
9287 Name
": Cannot restore data from backup."
9288 " Too many failed disks\n");
9292 if (save_checkpoint_imsm(st
, info
, UNIT_SRC_NORMAL
)) {
9293 /* ignore error == 2, this can mean end of reshape here
9295 dprintf("imsm: Cannot write checkpoint to "
9296 "migration record (UNIT_SRC_NORMAL) during restart\n");
9302 for (i
= 0; i
< new_disks
; i
++)
9311 static char disk_by_path
[] = "/dev/disk/by-path/";
9313 static const char *imsm_get_disk_controller_domain(const char *path
)
9315 char disk_path
[PATH_MAX
];
9319 strncpy(disk_path
, disk_by_path
, PATH_MAX
- 1);
9320 strncat(disk_path
, path
, PATH_MAX
- strlen(disk_path
) - 1);
9321 if (stat(disk_path
, &st
) == 0) {
9322 struct sys_dev
* hba
;
9325 path
= devt_to_devpath(st
.st_rdev
);
9328 hba
= find_disk_attached_hba(-1, path
);
9329 if (hba
&& hba
->type
== SYS_DEV_SAS
)
9331 else if (hba
&& hba
->type
== SYS_DEV_SATA
)
9335 dprintf("path: %s hba: %s attached: %s\n",
9336 path
, (hba
) ? hba
->path
: "NULL", drv
);
9344 static int imsm_find_array_minor_by_subdev(int subdev
, int container
, int *minor
)
9346 char subdev_name
[20];
9347 struct mdstat_ent
*mdstat
;
9349 sprintf(subdev_name
, "%d", subdev
);
9350 mdstat
= mdstat_by_subdev(subdev_name
, container
);
9354 *minor
= mdstat
->devnum
;
9355 free_mdstat(mdstat
);
9359 static int imsm_reshape_is_allowed_on_container(struct supertype
*st
,
9360 struct geo_params
*geo
,
9361 int *old_raid_disks
)
9363 /* currently we only support increasing the number of devices
9364 * for a container. This increases the number of device for each
9365 * member array. They must all be RAID0 or RAID5.
9368 struct mdinfo
*info
, *member
;
9369 int devices_that_can_grow
= 0;
9371 dprintf("imsm: imsm_reshape_is_allowed_on_container(ENTER): "
9372 "st->devnum = (%i)\n",
9375 if (geo
->size
!= -1 ||
9376 geo
->level
!= UnSet
||
9377 geo
->layout
!= UnSet
||
9378 geo
->chunksize
!= 0 ||
9379 geo
->raid_disks
== UnSet
) {
9380 dprintf("imsm: Container operation is allowed for "
9381 "raid disks number change only.\n");
9385 info
= container_content_imsm(st
, NULL
);
9386 for (member
= info
; member
; member
= member
->next
) {
9390 dprintf("imsm: checking device_num: %i\n",
9391 member
->container_member
);
9393 if (geo
->raid_disks
<= member
->array
.raid_disks
) {
9394 /* we work on container for Online Capacity Expansion
9395 * only so raid_disks has to grow
9397 dprintf("imsm: for container operation raid disks "
9398 "increase is required\n");
9402 if ((info
->array
.level
!= 0) &&
9403 (info
->array
.level
!= 5)) {
9404 /* we cannot use this container with other raid level
9406 dprintf("imsm: for container operation wrong"
9407 " raid level (%i) detected\n",
9411 /* check for platform support
9412 * for this raid level configuration
9414 struct intel_super
*super
= st
->sb
;
9415 if (!is_raid_level_supported(super
->orom
,
9416 member
->array
.level
,
9418 dprintf("platform does not support raid%d with"
9422 geo
->raid_disks
> 1 ? "s" : "");
9425 /* check if component size is aligned to chunk size
9427 if (info
->component_size
%
9428 (info
->array
.chunk_size
/512)) {
9429 dprintf("Component size is not aligned to "
9435 if (*old_raid_disks
&&
9436 info
->array
.raid_disks
!= *old_raid_disks
)
9438 *old_raid_disks
= info
->array
.raid_disks
;
9440 /* All raid5 and raid0 volumes in container
9441 * have to be ready for Online Capacity Expansion
9442 * so they need to be assembled. We have already
9443 * checked that no recovery etc is happening.
9445 result
= imsm_find_array_minor_by_subdev(member
->container_member
,
9449 dprintf("imsm: cannot find array\n");
9452 devices_that_can_grow
++;
9455 if (!member
&& devices_that_can_grow
)
9459 dprintf("\tContainer operation allowed\n");
9461 dprintf("\tError: %i\n", ret_val
);
9466 /* Function: get_spares_for_grow
9467 * Description: Allocates memory and creates list of spare devices
9468 * avaliable in container. Checks if spare drive size is acceptable.
9469 * Parameters: Pointer to the supertype structure
9470 * Returns: Pointer to the list of spare devices (mdinfo structure) on success,
9473 static struct mdinfo
*get_spares_for_grow(struct supertype
*st
)
9475 unsigned long long min_size
= min_acceptable_spare_size_imsm(st
);
9476 return container_choose_spares(st
, min_size
, NULL
, NULL
, NULL
, 0);
9479 /******************************************************************************
9480 * function: imsm_create_metadata_update_for_reshape
9481 * Function creates update for whole IMSM container.
9483 ******************************************************************************/
9484 static int imsm_create_metadata_update_for_reshape(
9485 struct supertype
*st
,
9486 struct geo_params
*geo
,
9488 struct imsm_update_reshape
**updatep
)
9490 struct intel_super
*super
= st
->sb
;
9491 struct imsm_super
*mpb
= super
->anchor
;
9492 int update_memory_size
= 0;
9493 struct imsm_update_reshape
*u
= NULL
;
9494 struct mdinfo
*spares
= NULL
;
9496 int delta_disks
= 0;
9499 dprintf("imsm_update_metadata_for_reshape(enter) raid_disks = %i\n",
9502 delta_disks
= geo
->raid_disks
- old_raid_disks
;
9504 /* size of all update data without anchor */
9505 update_memory_size
= sizeof(struct imsm_update_reshape
);
9507 /* now add space for spare disks that we need to add. */
9508 update_memory_size
+= sizeof(u
->new_disks
[0]) * (delta_disks
- 1);
9510 u
= calloc(1, update_memory_size
);
9513 "cannot get memory for imsm_update_reshape update\n");
9516 u
->type
= update_reshape_container_disks
;
9517 u
->old_raid_disks
= old_raid_disks
;
9518 u
->new_raid_disks
= geo
->raid_disks
;
9520 /* now get spare disks list
9522 spares
= get_spares_for_grow(st
);
9525 || delta_disks
> spares
->array
.spare_disks
) {
9526 fprintf(stderr
, Name
": imsm: ERROR: Cannot get spare devices "
9527 "for %s.\n", geo
->dev_name
);
9532 /* we have got spares
9533 * update disk list in imsm_disk list table in anchor
9535 dprintf("imsm: %i spares are available.\n\n",
9536 spares
->array
.spare_disks
);
9539 for (i
= 0; i
< delta_disks
; i
++) {
9544 u
->new_disks
[i
] = makedev(dev
->disk
.major
,
9546 dl
= get_disk_super(super
, dev
->disk
.major
, dev
->disk
.minor
);
9547 dl
->index
= mpb
->num_disks
;
9557 dprintf("imsm: reshape update preparation :");
9558 if (i
== delta_disks
) {
9561 return update_memory_size
;
9564 dprintf(" Error\n");
9569 /******************************************************************************
9570 * function: imsm_create_metadata_update_for_migration()
9571 * Creates update for IMSM array.
9573 ******************************************************************************/
9574 static int imsm_create_metadata_update_for_migration(
9575 struct supertype
*st
,
9576 struct geo_params
*geo
,
9577 struct imsm_update_reshape_migration
**updatep
)
9579 struct intel_super
*super
= st
->sb
;
9580 int update_memory_size
= 0;
9581 struct imsm_update_reshape_migration
*u
= NULL
;
9582 struct imsm_dev
*dev
;
9583 int previous_level
= -1;
9585 dprintf("imsm_create_metadata_update_for_migration(enter)"
9586 " New Level = %i\n", geo
->level
);
9588 /* size of all update data without anchor */
9589 update_memory_size
= sizeof(struct imsm_update_reshape_migration
);
9591 u
= calloc(1, update_memory_size
);
9593 dprintf("error: cannot get memory for "
9594 "imsm_create_metadata_update_for_migration\n");
9597 u
->type
= update_reshape_migration
;
9598 u
->subdev
= super
->current_vol
;
9599 u
->new_level
= geo
->level
;
9600 u
->new_layout
= geo
->layout
;
9601 u
->new_raid_disks
= u
->old_raid_disks
= geo
->raid_disks
;
9602 u
->new_disks
[0] = -1;
9603 u
->new_chunksize
= -1;
9605 dev
= get_imsm_dev(super
, u
->subdev
);
9607 struct imsm_map
*map
;
9609 map
= get_imsm_map(dev
, MAP_0
);
9611 int current_chunk_size
=
9612 __le16_to_cpu(map
->blocks_per_strip
) / 2;
9614 if (geo
->chunksize
!= current_chunk_size
) {
9615 u
->new_chunksize
= geo
->chunksize
/ 1024;
9617 "chunk size change from %i to %i\n",
9618 current_chunk_size
, u
->new_chunksize
);
9620 previous_level
= map
->raid_level
;
9623 if ((geo
->level
== 5) && (previous_level
== 0)) {
9624 struct mdinfo
*spares
= NULL
;
9626 u
->new_raid_disks
++;
9627 spares
= get_spares_for_grow(st
);
9628 if ((spares
== NULL
) || (spares
->array
.spare_disks
< 1)) {
9631 update_memory_size
= 0;
9632 dprintf("error: cannot get spare device "
9633 "for requested migration");
9638 dprintf("imsm: reshape update preparation : OK\n");
9641 return update_memory_size
;
9644 static void imsm_update_metadata_locally(struct supertype
*st
,
9647 struct metadata_update mu
;
9652 mu
.space_list
= NULL
;
9654 imsm_prepare_update(st
, &mu
);
9655 imsm_process_update(st
, &mu
);
9657 while (mu
.space_list
) {
9658 void **space
= mu
.space_list
;
9659 mu
.space_list
= *space
;
9664 /***************************************************************************
9665 * Function: imsm_analyze_change
9666 * Description: Function analyze change for single volume
9667 * and validate if transition is supported
9668 * Parameters: Geometry parameters, supertype structure
9669 * Returns: Operation type code on success, -1 if fail
9670 ****************************************************************************/
9671 enum imsm_reshape_type
imsm_analyze_change(struct supertype
*st
,
9672 struct geo_params
*geo
)
9678 /* number of added/removed disks in operation result */
9679 int devNumChange
= 0;
9680 /* imsm compatible layout value for array geometry verification */
9681 int imsm_layout
= -1;
9683 getinfo_super_imsm_volume(st
, &info
, NULL
);
9684 if ((geo
->level
!= info
.array
.level
) &&
9685 (geo
->level
>= 0) &&
9686 (geo
->level
!= UnSet
)) {
9687 switch (info
.array
.level
) {
9689 if (geo
->level
== 5) {
9690 change
= CH_MIGRATION
;
9691 if (geo
->layout
!= ALGORITHM_LEFT_ASYMMETRIC
) {
9693 Name
" Error. Requested Layout "
9694 "not supported (left-asymmetric layout "
9695 "is supported only)!\n");
9697 goto analyse_change_exit
;
9699 imsm_layout
= geo
->layout
;
9701 devNumChange
= 1; /* parity disk added */
9702 } else if (geo
->level
== 10) {
9703 change
= CH_TAKEOVER
;
9705 devNumChange
= 2; /* two mirrors added */
9706 imsm_layout
= 0x102; /* imsm supported layout */
9711 if (geo
->level
== 0) {
9712 change
= CH_TAKEOVER
;
9714 devNumChange
= -(geo
->raid_disks
/2);
9715 imsm_layout
= 0; /* imsm raid0 layout */
9721 Name
" Error. Level Migration from %d to %d "
9723 info
.array
.level
, geo
->level
);
9724 goto analyse_change_exit
;
9727 geo
->level
= info
.array
.level
;
9729 if ((geo
->layout
!= info
.array
.layout
)
9730 && ((geo
->layout
!= UnSet
) && (geo
->layout
!= -1))) {
9731 change
= CH_MIGRATION
;
9732 if ((info
.array
.layout
== 0)
9733 && (info
.array
.level
== 5)
9734 && (geo
->layout
== 5)) {
9735 /* reshape 5 -> 4 */
9736 } else if ((info
.array
.layout
== 5)
9737 && (info
.array
.level
== 5)
9738 && (geo
->layout
== 0)) {
9739 /* reshape 4 -> 5 */
9744 Name
" Error. Layout Migration from %d to %d "
9746 info
.array
.layout
, geo
->layout
);
9748 goto analyse_change_exit
;
9751 geo
->layout
= info
.array
.layout
;
9752 if (imsm_layout
== -1)
9753 imsm_layout
= info
.array
.layout
;
9756 if ((geo
->chunksize
> 0) && (geo
->chunksize
!= UnSet
)
9757 && (geo
->chunksize
!= info
.array
.chunk_size
))
9758 change
= CH_MIGRATION
;
9760 geo
->chunksize
= info
.array
.chunk_size
;
9762 chunk
= geo
->chunksize
/ 1024;
9763 if (!validate_geometry_imsm(st
,
9766 geo
->raid_disks
+ devNumChange
,
9773 struct intel_super
*super
= st
->sb
;
9774 struct imsm_super
*mpb
= super
->anchor
;
9776 if (mpb
->num_raid_devs
> 1) {
9778 Name
" Error. Cannot perform operation on %s"
9779 "- for this operation it MUST be single "
9780 "array in container\n",
9786 analyse_change_exit
:
9791 int imsm_takeover(struct supertype
*st
, struct geo_params
*geo
)
9793 struct intel_super
*super
= st
->sb
;
9794 struct imsm_update_takeover
*u
;
9796 u
= malloc(sizeof(struct imsm_update_takeover
));
9800 u
->type
= update_takeover
;
9801 u
->subarray
= super
->current_vol
;
9803 /* 10->0 transition */
9804 if (geo
->level
== 0)
9805 u
->direction
= R10_TO_R0
;
9807 /* 0->10 transition */
9808 if (geo
->level
== 10)
9809 u
->direction
= R0_TO_R10
;
9811 /* update metadata locally */
9812 imsm_update_metadata_locally(st
, u
,
9813 sizeof(struct imsm_update_takeover
));
9814 /* and possibly remotely */
9815 if (st
->update_tail
)
9816 append_metadata_update(st
, u
,
9817 sizeof(struct imsm_update_takeover
));
9824 static int imsm_reshape_super(struct supertype
*st
, long long size
, int level
,
9825 int layout
, int chunksize
, int raid_disks
,
9826 int delta_disks
, char *backup
, char *dev
,
9830 struct geo_params geo
;
9832 dprintf("imsm: reshape_super called.\n");
9834 memset(&geo
, 0, sizeof(struct geo_params
));
9837 geo
.dev_id
= st
->devnum
;
9840 geo
.layout
= layout
;
9841 geo
.chunksize
= chunksize
;
9842 geo
.raid_disks
= raid_disks
;
9843 if (delta_disks
!= UnSet
)
9844 geo
.raid_disks
+= delta_disks
;
9846 dprintf("\tfor level : %i\n", geo
.level
);
9847 dprintf("\tfor raid_disks : %i\n", geo
.raid_disks
);
9849 if (experimental() == 0)
9852 if (st
->container_dev
== st
->devnum
) {
9853 /* On container level we can only increase number of devices. */
9854 dprintf("imsm: info: Container operation\n");
9855 int old_raid_disks
= 0;
9857 if (imsm_reshape_is_allowed_on_container(
9858 st
, &geo
, &old_raid_disks
)) {
9859 struct imsm_update_reshape
*u
= NULL
;
9862 len
= imsm_create_metadata_update_for_reshape(
9863 st
, &geo
, old_raid_disks
, &u
);
9866 dprintf("imsm: Cannot prepare update\n");
9867 goto exit_imsm_reshape_super
;
9871 /* update metadata locally */
9872 imsm_update_metadata_locally(st
, u
, len
);
9873 /* and possibly remotely */
9874 if (st
->update_tail
)
9875 append_metadata_update(st
, u
, len
);
9880 fprintf(stderr
, Name
": (imsm) Operation "
9881 "is not allowed on this container\n");
9884 /* On volume level we support following operations
9885 * - takeover: raid10 -> raid0; raid0 -> raid10
9886 * - chunk size migration
9887 * - migration: raid5 -> raid0; raid0 -> raid5
9889 struct intel_super
*super
= st
->sb
;
9890 struct intel_dev
*dev
= super
->devlist
;
9892 dprintf("imsm: info: Volume operation\n");
9893 /* find requested device */
9895 if (imsm_find_array_minor_by_subdev(
9896 dev
->index
, st
->container_dev
, &devnum
) == 0
9897 && devnum
== geo
.dev_id
)
9902 fprintf(stderr
, Name
" Cannot find %s (%i) subarray\n",
9903 geo
.dev_name
, geo
.dev_id
);
9904 goto exit_imsm_reshape_super
;
9906 super
->current_vol
= dev
->index
;
9907 change
= imsm_analyze_change(st
, &geo
);
9910 ret_val
= imsm_takeover(st
, &geo
);
9912 case CH_MIGRATION
: {
9913 struct imsm_update_reshape_migration
*u
= NULL
;
9915 imsm_create_metadata_update_for_migration(
9919 "Cannot prepare update\n");
9923 /* update metadata locally */
9924 imsm_update_metadata_locally(st
, u
, len
);
9925 /* and possibly remotely */
9926 if (st
->update_tail
)
9927 append_metadata_update(st
, u
, len
);
9937 exit_imsm_reshape_super
:
9938 dprintf("imsm: reshape_super Exit code = %i\n", ret_val
);
9942 /*******************************************************************************
9943 * Function: wait_for_reshape_imsm
9944 * Description: Function writes new sync_max value and waits until
9945 * reshape process reach new position
9947 * sra : general array info
9948 * ndata : number of disks in new array's layout
9951 * 1 : there is no reshape in progress,
9953 ******************************************************************************/
9954 int wait_for_reshape_imsm(struct mdinfo
*sra
, int ndata
)
9956 int fd
= sysfs_get_fd(sra
, NULL
, "reshape_position");
9957 unsigned long long completed
;
9958 /* to_complete : new sync_max position */
9959 unsigned long long to_complete
= sra
->reshape_progress
;
9960 unsigned long long position_to_set
= to_complete
/ ndata
;
9963 dprintf("imsm: wait_for_reshape_imsm() "
9964 "cannot open reshape_position\n");
9968 if (sysfs_fd_get_ll(fd
, &completed
) < 0) {
9969 dprintf("imsm: wait_for_reshape_imsm() "
9970 "cannot read reshape_position (no reshape in progres)\n");
9975 if (completed
> to_complete
) {
9976 dprintf("imsm: wait_for_reshape_imsm() "
9977 "wrong next position to set %llu (%llu)\n",
9978 to_complete
, completed
);
9982 dprintf("Position set: %llu\n", position_to_set
);
9983 if (sysfs_set_num(sra
, NULL
, "sync_max",
9984 position_to_set
) != 0) {
9985 dprintf("imsm: wait_for_reshape_imsm() "
9986 "cannot set reshape position to %llu\n",
9997 select(fd
+1, &rfds
, NULL
, NULL
, NULL
);
9998 if (sysfs_get_str(sra
, NULL
, "sync_action",
10000 strncmp(action
, "reshape", 7) != 0)
10002 if (sysfs_fd_get_ll(fd
, &completed
) < 0) {
10003 dprintf("imsm: wait_for_reshape_imsm() "
10004 "cannot read reshape_position (in loop)\n");
10008 } while (completed
< to_complete
);
10014 /*******************************************************************************
10015 * Function: check_degradation_change
10016 * Description: Check that array hasn't become failed.
10018 * info : for sysfs access
10019 * sources : source disks descriptors
10020 * degraded: previous degradation level
10022 * degradation level
10023 ******************************************************************************/
10024 int check_degradation_change(struct mdinfo
*info
,
10028 unsigned long long new_degraded
;
10029 sysfs_get_ll(info
, NULL
, "degraded", &new_degraded
);
10030 if (new_degraded
!= (unsigned long long)degraded
) {
10031 /* check each device to ensure it is still working */
10034 for (sd
= info
->devs
; sd
; sd
= sd
->next
) {
10035 if (sd
->disk
.state
& (1<<MD_DISK_FAULTY
))
10037 if (sd
->disk
.state
& (1<<MD_DISK_SYNC
)) {
10039 if (sysfs_get_str(info
,
10040 sd
, "state", sbuf
, 20) < 0 ||
10041 strstr(sbuf
, "faulty") ||
10042 strstr(sbuf
, "in_sync") == NULL
) {
10043 /* this device is dead */
10044 sd
->disk
.state
= (1<<MD_DISK_FAULTY
);
10045 if (sd
->disk
.raid_disk
>= 0 &&
10046 sources
[sd
->disk
.raid_disk
] >= 0) {
10048 sd
->disk
.raid_disk
]);
10049 sources
[sd
->disk
.raid_disk
] =
10058 return new_degraded
;
10061 /*******************************************************************************
10062 * Function: imsm_manage_reshape
10063 * Description: Function finds array under reshape and it manages reshape
10064 * process. It creates stripes backups (if required) and sets
10067 * afd : Backup handle (nattive) - not used
10068 * sra : general array info
10069 * reshape : reshape parameters - not used
10070 * st : supertype structure
10071 * blocks : size of critical section [blocks]
10072 * fds : table of source device descriptor
10073 * offsets : start of array (offest per devices)
10075 * destfd : table of destination device descriptor
10076 * destoffsets : table of destination offsets (per device)
10078 * 1 : success, reshape is done
10080 ******************************************************************************/
10081 static int imsm_manage_reshape(
10082 int afd
, struct mdinfo
*sra
, struct reshape
*reshape
,
10083 struct supertype
*st
, unsigned long backup_blocks
,
10084 int *fds
, unsigned long long *offsets
,
10085 int dests
, int *destfd
, unsigned long long *destoffsets
)
10088 struct intel_super
*super
= st
->sb
;
10089 struct intel_dev
*dv
= NULL
;
10090 struct imsm_dev
*dev
= NULL
;
10091 struct imsm_map
*map_src
;
10092 int migr_vol_qan
= 0;
10093 int ndata
, odata
; /* [bytes] */
10094 int chunk
; /* [bytes] */
10095 struct migr_record
*migr_rec
;
10097 unsigned int buf_size
; /* [bytes] */
10098 unsigned long long max_position
; /* array size [bytes] */
10099 unsigned long long next_step
; /* [blocks]/[bytes] */
10100 unsigned long long old_data_stripe_length
;
10101 unsigned long long start_src
; /* [bytes] */
10102 unsigned long long start
; /* [bytes] */
10103 unsigned long long start_buf_shift
; /* [bytes] */
10105 int source_layout
= 0;
10107 if (!fds
|| !offsets
|| !sra
)
10110 /* Find volume during the reshape */
10111 for (dv
= super
->devlist
; dv
; dv
= dv
->next
) {
10112 if (dv
->dev
->vol
.migr_type
== MIGR_GEN_MIGR
10113 && dv
->dev
->vol
.migr_state
== 1) {
10118 /* Only one volume can migrate at the same time */
10119 if (migr_vol_qan
!= 1) {
10120 fprintf(stderr
, Name
" : %s", migr_vol_qan
?
10121 "Number of migrating volumes greater than 1\n" :
10122 "There is no volume during migrationg\n");
10126 map_src
= get_imsm_map(dev
, MAP_1
);
10127 if (map_src
== NULL
)
10130 ndata
= imsm_num_data_members(dev
, MAP_0
);
10131 odata
= imsm_num_data_members(dev
, MAP_1
);
10133 chunk
= __le16_to_cpu(map_src
->blocks_per_strip
) * 512;
10134 old_data_stripe_length
= odata
* chunk
;
10136 migr_rec
= super
->migr_rec
;
10138 /* initialize migration record for start condition */
10139 if (sra
->reshape_progress
== 0)
10140 init_migr_record_imsm(st
, dev
, sra
);
10142 if (__le32_to_cpu(migr_rec
->rec_status
) != UNIT_SRC_NORMAL
) {
10143 dprintf("imsm: cannot restart migration when data "
10144 "are present in copy area.\n");
10147 /* Save checkpoint to update migration record for current
10148 * reshape position (in md). It can be farther than current
10149 * reshape position in metadata.
10151 if (save_checkpoint_imsm(st
, sra
, UNIT_SRC_NORMAL
) == 1) {
10152 /* ignore error == 2, this can mean end of reshape here
10154 dprintf("imsm: Cannot write checkpoint to "
10155 "migration record (UNIT_SRC_NORMAL, "
10156 "initial save)\n");
10161 /* size for data */
10162 buf_size
= __le32_to_cpu(migr_rec
->blocks_per_unit
) * 512;
10163 /* extend buffer size for parity disk */
10164 buf_size
+= __le32_to_cpu(migr_rec
->dest_depth_per_unit
) * 512;
10165 /* add space for stripe aligment */
10166 buf_size
+= old_data_stripe_length
;
10167 if (posix_memalign((void **)&buf
, 4096, buf_size
)) {
10168 dprintf("imsm: Cannot allocate checpoint buffer\n");
10172 max_position
= sra
->component_size
* ndata
;
10173 source_layout
= imsm_level_to_layout(map_src
->raid_level
);
10175 while (__le32_to_cpu(migr_rec
->curr_migr_unit
) <
10176 __le32_to_cpu(migr_rec
->num_migr_units
)) {
10177 /* current reshape position [blocks] */
10178 unsigned long long current_position
=
10179 __le32_to_cpu(migr_rec
->blocks_per_unit
)
10180 * __le32_to_cpu(migr_rec
->curr_migr_unit
);
10181 unsigned long long border
;
10183 /* Check that array hasn't become failed.
10185 degraded
= check_degradation_change(sra
, fds
, degraded
);
10186 if (degraded
> 1) {
10187 dprintf("imsm: Abort reshape due to degradation"
10188 " level (%i)\n", degraded
);
10192 next_step
= __le32_to_cpu(migr_rec
->blocks_per_unit
);
10194 if ((current_position
+ next_step
) > max_position
)
10195 next_step
= max_position
- current_position
;
10197 start
= current_position
* 512;
10199 /* allign reading start to old geometry */
10200 start_buf_shift
= start
% old_data_stripe_length
;
10201 start_src
= start
- start_buf_shift
;
10203 border
= (start_src
/ odata
) - (start
/ ndata
);
10205 if (border
<= __le32_to_cpu(migr_rec
->dest_depth_per_unit
)) {
10206 /* save critical stripes to buf
10207 * start - start address of current unit
10208 * to backup [bytes]
10209 * start_src - start address of current unit
10210 * to backup alligned to source array
10213 unsigned long long next_step_filler
= 0;
10214 unsigned long long copy_length
= next_step
* 512;
10216 /* allign copy area length to stripe in old geometry */
10217 next_step_filler
= ((copy_length
+ start_buf_shift
)
10218 % old_data_stripe_length
);
10219 if (next_step_filler
)
10220 next_step_filler
= (old_data_stripe_length
10221 - next_step_filler
);
10222 dprintf("save_stripes() parameters: start = %llu,"
10223 "\tstart_src = %llu,\tnext_step*512 = %llu,"
10224 "\tstart_in_buf_shift = %llu,"
10225 "\tnext_step_filler = %llu\n",
10226 start
, start_src
, copy_length
,
10227 start_buf_shift
, next_step_filler
);
10229 if (save_stripes(fds
, offsets
, map_src
->num_members
,
10230 chunk
, map_src
->raid_level
,
10231 source_layout
, 0, NULL
, start_src
,
10233 next_step_filler
+ start_buf_shift
,
10235 dprintf("imsm: Cannot save stripes"
10239 /* Convert data to destination format and store it
10240 * in backup general migration area
10242 if (save_backup_imsm(st
, dev
, sra
,
10243 buf
+ start_buf_shift
, copy_length
)) {
10244 dprintf("imsm: Cannot save stripes to "
10245 "target devices\n");
10248 if (save_checkpoint_imsm(st
, sra
,
10249 UNIT_SRC_IN_CP_AREA
)) {
10250 dprintf("imsm: Cannot write checkpoint to "
10251 "migration record (UNIT_SRC_IN_CP_AREA)\n");
10255 /* set next step to use whole border area */
10256 border
/= next_step
;
10258 next_step
*= border
;
10260 /* When data backed up, checkpoint stored,
10261 * kick the kernel to reshape unit of data
10263 next_step
= next_step
+ sra
->reshape_progress
;
10264 /* limit next step to array max position */
10265 if (next_step
> max_position
)
10266 next_step
= max_position
;
10267 sysfs_set_num(sra
, NULL
, "suspend_lo", sra
->reshape_progress
);
10268 sysfs_set_num(sra
, NULL
, "suspend_hi", next_step
);
10269 sra
->reshape_progress
= next_step
;
10271 /* wait until reshape finish */
10272 if (wait_for_reshape_imsm(sra
, ndata
) < 0) {
10273 dprintf("wait_for_reshape_imsm returned error!\n");
10277 if (save_checkpoint_imsm(st
, sra
, UNIT_SRC_NORMAL
) == 1) {
10278 /* ignore error == 2, this can mean end of reshape here
10280 dprintf("imsm: Cannot write checkpoint to "
10281 "migration record (UNIT_SRC_NORMAL)\n");
10287 /* return '1' if done */
10291 abort_reshape(sra
);
10295 #endif /* MDASSEMBLE */
10297 struct superswitch super_imsm
= {
10299 .examine_super
= examine_super_imsm
,
10300 .brief_examine_super
= brief_examine_super_imsm
,
10301 .brief_examine_subarrays
= brief_examine_subarrays_imsm
,
10302 .export_examine_super
= export_examine_super_imsm
,
10303 .detail_super
= detail_super_imsm
,
10304 .brief_detail_super
= brief_detail_super_imsm
,
10305 .write_init_super
= write_init_super_imsm
,
10306 .validate_geometry
= validate_geometry_imsm
,
10307 .add_to_super
= add_to_super_imsm
,
10308 .remove_from_super
= remove_from_super_imsm
,
10309 .detail_platform
= detail_platform_imsm
,
10310 .kill_subarray
= kill_subarray_imsm
,
10311 .update_subarray
= update_subarray_imsm
,
10312 .load_container
= load_container_imsm
,
10313 .default_geometry
= default_geometry_imsm
,
10314 .get_disk_controller_domain
= imsm_get_disk_controller_domain
,
10315 .reshape_super
= imsm_reshape_super
,
10316 .manage_reshape
= imsm_manage_reshape
,
10317 .recover_backup
= recover_backup_imsm
,
10319 .match_home
= match_home_imsm
,
10320 .uuid_from_super
= uuid_from_super_imsm
,
10321 .getinfo_super
= getinfo_super_imsm
,
10322 .getinfo_super_disks
= getinfo_super_disks_imsm
,
10323 .update_super
= update_super_imsm
,
10325 .avail_size
= avail_size_imsm
,
10326 .min_acceptable_spare_size
= min_acceptable_spare_size_imsm
,
10328 .compare_super
= compare_super_imsm
,
10330 .load_super
= load_super_imsm
,
10331 .init_super
= init_super_imsm
,
10332 .store_super
= store_super_imsm
,
10333 .free_super
= free_super_imsm
,
10334 .match_metadata_desc
= match_metadata_desc_imsm
,
10335 .container_content
= container_content_imsm
,
10343 .open_new
= imsm_open_new
,
10344 .set_array_state
= imsm_set_array_state
,
10345 .set_disk
= imsm_set_disk
,
10346 .sync_metadata
= imsm_sync_metadata
,
10347 .activate_spare
= imsm_activate_spare
,
10348 .process_update
= imsm_process_update
,
10349 .prepare_update
= imsm_prepare_update
,
10350 #endif /* MDASSEMBLE */