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
; /* 0xE8 - 0xEB total blocks */
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 #define IMSM_DISK_FILLERS 4
106 __u32 filler
[IMSM_DISK_FILLERS
]; /* 0xF4 - 0x107 MPB_DISK_FILLERS for future expansion */
109 /* RAID map configuration infos. */
111 __u32 pba_of_lba0
; /* start address of partition */
112 __u32 blocks_per_member
;/* blocks per member */
113 __u32 num_data_stripes
; /* number of data stripes */
114 __u16 blocks_per_strip
;
115 __u8 map_state
; /* Normal, Uninitialized, Degraded, Failed */
116 #define IMSM_T_STATE_NORMAL 0
117 #define IMSM_T_STATE_UNINITIALIZED 1
118 #define IMSM_T_STATE_DEGRADED 2
119 #define IMSM_T_STATE_FAILED 3
121 #define IMSM_T_RAID0 0
122 #define IMSM_T_RAID1 1
123 #define IMSM_T_RAID5 5 /* since metadata version 1.2.02 ? */
124 __u8 num_members
; /* number of member disks */
125 __u8 num_domains
; /* number of parity domains */
126 __u8 failed_disk_num
; /* valid only when state is degraded */
128 __u32 filler
[7]; /* expansion area */
129 #define IMSM_ORD_REBUILD (1 << 24)
130 __u32 disk_ord_tbl
[1]; /* disk_ord_tbl[num_members],
131 * top byte contains some flags
133 } __attribute__ ((packed
));
136 __u32 curr_migr_unit
;
137 __u32 checkpoint_id
; /* id to access curr_migr_unit */
138 __u8 migr_state
; /* Normal or Migrating */
140 #define MIGR_REBUILD 1
141 #define MIGR_VERIFY 2 /* analagous to echo check > sync_action */
142 #define MIGR_GEN_MIGR 3
143 #define MIGR_STATE_CHANGE 4
144 #define MIGR_REPAIR 5
145 __u8 migr_type
; /* Initializing, Rebuilding, ... */
147 __u8 fs_state
; /* fast-sync state for CnG (0xff == disabled) */
148 __u16 verify_errors
; /* number of mismatches */
149 __u16 bad_blocks
; /* number of bad blocks during verify */
151 struct imsm_map map
[1];
152 /* here comes another one if migr_state */
153 } __attribute__ ((packed
));
156 __u8 volume
[MAX_RAID_SERIAL_LEN
];
159 #define DEV_BOOTABLE __cpu_to_le32(0x01)
160 #define DEV_BOOT_DEVICE __cpu_to_le32(0x02)
161 #define DEV_READ_COALESCING __cpu_to_le32(0x04)
162 #define DEV_WRITE_COALESCING __cpu_to_le32(0x08)
163 #define DEV_LAST_SHUTDOWN_DIRTY __cpu_to_le32(0x10)
164 #define DEV_HIDDEN_AT_BOOT __cpu_to_le32(0x20)
165 #define DEV_CURRENTLY_HIDDEN __cpu_to_le32(0x40)
166 #define DEV_VERIFY_AND_FIX __cpu_to_le32(0x80)
167 #define DEV_MAP_STATE_UNINIT __cpu_to_le32(0x100)
168 #define DEV_NO_AUTO_RECOVERY __cpu_to_le32(0x200)
169 #define DEV_CLONE_N_GO __cpu_to_le32(0x400)
170 #define DEV_CLONE_MAN_SYNC __cpu_to_le32(0x800)
171 #define DEV_CNG_MASTER_DISK_NUM __cpu_to_le32(0x1000)
172 __u32 status
; /* Persistent RaidDev status */
173 __u32 reserved_blocks
; /* Reserved blocks at beginning of volume */
177 __u8 cng_master_disk
;
181 #define IMSM_DEV_FILLERS 10
182 __u32 filler
[IMSM_DEV_FILLERS
];
184 } __attribute__ ((packed
));
187 __u8 sig
[MAX_SIGNATURE_LENGTH
]; /* 0x00 - 0x1F */
188 __u32 check_sum
; /* 0x20 - 0x23 MPB Checksum */
189 __u32 mpb_size
; /* 0x24 - 0x27 Size of MPB */
190 __u32 family_num
; /* 0x28 - 0x2B Checksum from first time this config was written */
191 __u32 generation_num
; /* 0x2C - 0x2F Incremented each time this array's MPB is written */
192 __u32 error_log_size
; /* 0x30 - 0x33 in bytes */
193 __u32 attributes
; /* 0x34 - 0x37 */
194 __u8 num_disks
; /* 0x38 Number of configured disks */
195 __u8 num_raid_devs
; /* 0x39 Number of configured volumes */
196 __u8 error_log_pos
; /* 0x3A */
197 __u8 fill
[1]; /* 0x3B */
198 __u32 cache_size
; /* 0x3c - 0x40 in mb */
199 __u32 orig_family_num
; /* 0x40 - 0x43 original family num */
200 __u32 pwr_cycle_count
; /* 0x44 - 0x47 simulated power cycle count for array */
201 __u32 bbm_log_size
; /* 0x48 - 0x4B - size of bad Block Mgmt Log in bytes */
202 #define IMSM_FILLERS 35
203 __u32 filler
[IMSM_FILLERS
]; /* 0x4C - 0xD7 RAID_MPB_FILLERS */
204 struct imsm_disk disk
[1]; /* 0xD8 diskTbl[numDisks] */
205 /* here comes imsm_dev[num_raid_devs] */
206 /* here comes BBM logs */
207 } __attribute__ ((packed
));
209 #define BBM_LOG_MAX_ENTRIES 254
211 struct bbm_log_entry
{
212 __u64 defective_block_start
;
213 #define UNREADABLE 0xFFFFFFFF
214 __u32 spare_block_offset
;
215 __u16 remapped_marked_count
;
217 } __attribute__ ((__packed__
));
220 __u32 signature
; /* 0xABADB10C */
222 __u32 reserved_spare_block_count
; /* 0 */
223 __u32 reserved
; /* 0xFFFF */
224 __u64 first_spare_lba
;
225 struct bbm_log_entry mapped_block_entries
[BBM_LOG_MAX_ENTRIES
];
226 } __attribute__ ((__packed__
));
230 static char *map_state_str
[] = { "normal", "uninitialized", "degraded", "failed" };
233 #define RAID_DISK_RESERVED_BLOCKS_IMSM_HI 2209
235 #define GEN_MIGR_AREA_SIZE 2048 /* General Migration Copy Area size in blocks */
237 #define UNIT_SRC_NORMAL 0 /* Source data for curr_migr_unit must
238 * be recovered using srcMap */
239 #define UNIT_SRC_IN_CP_AREA 1 /* Source data for curr_migr_unit has
240 * already been migrated and must
241 * be recovered from checkpoint area */
243 __u32 rec_status
; /* Status used to determine how to restart
244 * migration in case it aborts
246 __u32 curr_migr_unit
; /* 0..numMigrUnits-1 */
247 __u32 family_num
; /* Family number of MPB
248 * containing the RaidDev
249 * that is migrating */
250 __u32 ascending_migr
; /* True if migrating in increasing
252 __u32 blocks_per_unit
; /* Num disk blocks per unit of operation */
253 __u32 dest_depth_per_unit
; /* Num member blocks each destMap
255 * advances per unit-of-operation */
256 __u32 ckpt_area_pba
; /* Pba of first block of ckpt copy area */
257 __u32 dest_1st_member_lba
; /* First member lba on first
258 * stripe of destination */
259 __u32 num_migr_units
; /* Total num migration units-of-op */
260 __u32 post_migr_vol_cap
; /* Size of volume after
261 * migration completes */
262 __u32 post_migr_vol_cap_hi
; /* Expansion space for LBA64 */
263 __u32 ckpt_read_disk_num
; /* Which member disk in destSubMap[0] the
264 * migration ckpt record was read from
265 * (for recovered migrations) */
266 } __attribute__ ((__packed__
));
268 static __u8
migr_type(struct imsm_dev
*dev
)
270 if (dev
->vol
.migr_type
== MIGR_VERIFY
&&
271 dev
->status
& DEV_VERIFY_AND_FIX
)
274 return dev
->vol
.migr_type
;
277 static void set_migr_type(struct imsm_dev
*dev
, __u8 migr_type
)
279 /* for compatibility with older oroms convert MIGR_REPAIR, into
280 * MIGR_VERIFY w/ DEV_VERIFY_AND_FIX status
282 if (migr_type
== MIGR_REPAIR
) {
283 dev
->vol
.migr_type
= MIGR_VERIFY
;
284 dev
->status
|= DEV_VERIFY_AND_FIX
;
286 dev
->vol
.migr_type
= migr_type
;
287 dev
->status
&= ~DEV_VERIFY_AND_FIX
;
291 static unsigned int sector_count(__u32 bytes
)
293 return ((bytes
+ (512-1)) & (~(512-1))) / 512;
296 static unsigned int mpb_sectors(struct imsm_super
*mpb
)
298 return sector_count(__le32_to_cpu(mpb
->mpb_size
));
302 struct imsm_dev
*dev
;
303 struct intel_dev
*next
;
308 enum sys_dev_type type
;
311 struct intel_hba
*next
;
318 /* internal representation of IMSM metadata */
321 void *buf
; /* O_DIRECT buffer for reading/writing metadata */
322 struct imsm_super
*anchor
; /* immovable parameters */
325 void *migr_rec_buf
; /* buffer for I/O operations */
326 struct migr_record
*migr_rec
; /* migration record */
328 size_t len
; /* size of the 'buf' allocation */
329 void *next_buf
; /* for realloc'ing buf from the manager */
331 int updates_pending
; /* count of pending updates for mdmon */
332 int current_vol
; /* index of raid device undergoing creation */
333 __u32 create_offset
; /* common start for 'current_vol' */
334 __u32 random
; /* random data for seeding new family numbers */
335 struct intel_dev
*devlist
;
339 __u8 serial
[MAX_RAID_SERIAL_LEN
];
342 struct imsm_disk disk
;
345 struct extent
*e
; /* for determining freespace @ create */
346 int raiddisk
; /* slot to fill in autolayout */
348 } *disks
, *current_disk
;
349 struct dl
*disk_mgmt_list
; /* list of disks to add/remove while mdmon
351 struct dl
*missing
; /* disks removed while we weren't looking */
352 struct bbm_log
*bbm_log
;
353 struct intel_hba
*hba
; /* device path of the raid controller for this metadata */
354 const struct imsm_orom
*orom
; /* platform firmware support */
355 struct intel_super
*next
; /* (temp) list for disambiguating family_num */
359 struct imsm_disk disk
;
360 #define IMSM_UNKNOWN_OWNER (-1)
362 struct intel_disk
*next
;
366 unsigned long long start
, size
;
369 /* definitions of reshape process types */
370 enum imsm_reshape_type
{
375 /* definition of messages passed to imsm_process_update */
376 enum imsm_update_type
{
377 update_activate_spare
,
381 update_add_remove_disk
,
382 update_reshape_container_disks
,
383 update_reshape_migration
,
385 update_general_migration_checkpoint
,
388 struct imsm_update_activate_spare
{
389 enum imsm_update_type type
;
393 struct imsm_update_activate_spare
*next
;
406 enum takeover_direction
{
410 struct imsm_update_takeover
{
411 enum imsm_update_type type
;
413 enum takeover_direction direction
;
416 struct imsm_update_reshape
{
417 enum imsm_update_type type
;
421 int new_disks
[1]; /* new_raid_disks - old_raid_disks makedev number */
424 struct imsm_update_reshape_migration
{
425 enum imsm_update_type type
;
428 /* fields for array migration changes
435 int new_disks
[1]; /* new_raid_disks - old_raid_disks makedev number */
438 struct imsm_update_general_migration_checkpoint
{
439 enum imsm_update_type type
;
440 __u32 curr_migr_unit
;
444 __u8 serial
[MAX_RAID_SERIAL_LEN
];
447 struct imsm_update_create_array
{
448 enum imsm_update_type type
;
453 struct imsm_update_kill_array
{
454 enum imsm_update_type type
;
458 struct imsm_update_rename_array
{
459 enum imsm_update_type type
;
460 __u8 name
[MAX_RAID_SERIAL_LEN
];
464 struct imsm_update_add_remove_disk
{
465 enum imsm_update_type type
;
469 static const char *_sys_dev_type
[] = {
470 [SYS_DEV_UNKNOWN
] = "Unknown",
471 [SYS_DEV_SAS
] = "SAS",
472 [SYS_DEV_SATA
] = "SATA"
475 const char *get_sys_dev_type(enum sys_dev_type type
)
477 if (type
>= SYS_DEV_MAX
)
478 type
= SYS_DEV_UNKNOWN
;
480 return _sys_dev_type
[type
];
483 static struct intel_hba
* alloc_intel_hba(struct sys_dev
*device
)
485 struct intel_hba
*result
= malloc(sizeof(*result
));
487 result
->type
= device
->type
;
488 result
->path
= strdup(device
->path
);
490 if (result
->path
&& (result
->pci_id
= strrchr(result
->path
, '/')) != NULL
)
496 static struct intel_hba
* find_intel_hba(struct intel_hba
*hba
, struct sys_dev
*device
)
498 struct intel_hba
*result
=NULL
;
499 for (result
= hba
; result
; result
= result
->next
) {
500 if (result
->type
== device
->type
&& strcmp(result
->path
, device
->path
) == 0)
506 static int attach_hba_to_super(struct intel_super
*super
, struct sys_dev
*device
)
508 struct intel_hba
*hba
;
510 /* check if disk attached to Intel HBA */
511 hba
= find_intel_hba(super
->hba
, device
);
514 /* Check if HBA is already attached to super */
515 if (super
->hba
== NULL
) {
516 super
->hba
= alloc_intel_hba(device
);
521 /* Intel metadata allows for all disks attached to the same type HBA.
522 * Do not sypport odf HBA types mixing
524 if (device
->type
!= hba
->type
)
530 hba
->next
= alloc_intel_hba(device
);
534 static struct sys_dev
* find_disk_attached_hba(int fd
, const char *devname
)
536 struct sys_dev
*list
, *elem
, *prev
;
539 if ((list
= find_intel_devices()) == NULL
)
543 disk_path
= (char *) devname
;
545 disk_path
= diskfd_to_devpath(fd
);
552 for (prev
= NULL
, elem
= list
; elem
; prev
= elem
, elem
= elem
->next
) {
553 if (path_attached_to_hba(disk_path
, elem
->path
)) {
557 prev
->next
= elem
->next
;
559 if (disk_path
!= devname
)
565 if (disk_path
!= devname
)
573 static int find_intel_hba_capability(int fd
, struct intel_super
*super
,
576 static struct supertype
*match_metadata_desc_imsm(char *arg
)
578 struct supertype
*st
;
580 if (strcmp(arg
, "imsm") != 0 &&
581 strcmp(arg
, "default") != 0
585 st
= malloc(sizeof(*st
));
588 memset(st
, 0, sizeof(*st
));
589 st
->container_dev
= NoMdDev
;
590 st
->ss
= &super_imsm
;
591 st
->max_devs
= IMSM_MAX_DEVICES
;
592 st
->minor_version
= 0;
598 static __u8
*get_imsm_version(struct imsm_super
*mpb
)
600 return &mpb
->sig
[MPB_SIG_LEN
];
604 /* retrieve a disk directly from the anchor when the anchor is known to be
605 * up-to-date, currently only at load time
607 static struct imsm_disk
*__get_imsm_disk(struct imsm_super
*mpb
, __u8 index
)
609 if (index
>= mpb
->num_disks
)
611 return &mpb
->disk
[index
];
614 /* retrieve the disk description based on a index of the disk
617 static struct dl
*get_imsm_dl_disk(struct intel_super
*super
, __u8 index
)
621 for (d
= super
->disks
; d
; d
= d
->next
)
622 if (d
->index
== index
)
627 /* retrieve a disk from the parsed metadata */
628 static struct imsm_disk
*get_imsm_disk(struct intel_super
*super
, __u8 index
)
632 dl
= get_imsm_dl_disk(super
, index
);
639 /* generate a checksum directly from the anchor when the anchor is known to be
640 * up-to-date, currently only at load or write_super after coalescing
642 static __u32
__gen_imsm_checksum(struct imsm_super
*mpb
)
644 __u32 end
= mpb
->mpb_size
/ sizeof(end
);
645 __u32
*p
= (__u32
*) mpb
;
649 sum
+= __le32_to_cpu(*p
);
653 return sum
- __le32_to_cpu(mpb
->check_sum
);
656 static size_t sizeof_imsm_map(struct imsm_map
*map
)
658 return sizeof(struct imsm_map
) + sizeof(__u32
) * (map
->num_members
- 1);
661 struct imsm_map
*get_imsm_map(struct imsm_dev
*dev
, int second_map
)
663 /* A device can have 2 maps if it is in the middle of a migration.
665 * 0 - we return the first map
666 * 1 - we return the second map if it exists, else NULL
667 * -1 - we return the second map if it exists, else the first
669 struct imsm_map
*map
= &dev
->vol
.map
[0];
671 if (second_map
== 1 && !dev
->vol
.migr_state
)
673 else if (second_map
== 1 ||
674 (second_map
< 0 && dev
->vol
.migr_state
)) {
677 return ptr
+ sizeof_imsm_map(map
);
683 /* return the size of the device.
684 * migr_state increases the returned size if map[0] were to be duplicated
686 static size_t sizeof_imsm_dev(struct imsm_dev
*dev
, int migr_state
)
688 size_t size
= sizeof(*dev
) - sizeof(struct imsm_map
) +
689 sizeof_imsm_map(get_imsm_map(dev
, 0));
691 /* migrating means an additional map */
692 if (dev
->vol
.migr_state
)
693 size
+= sizeof_imsm_map(get_imsm_map(dev
, 1));
695 size
+= sizeof_imsm_map(get_imsm_map(dev
, 0));
701 /* retrieve disk serial number list from a metadata update */
702 static struct disk_info
*get_disk_info(struct imsm_update_create_array
*update
)
705 struct disk_info
*inf
;
707 inf
= u
+ sizeof(*update
) - sizeof(struct imsm_dev
) +
708 sizeof_imsm_dev(&update
->dev
, 0);
714 static struct imsm_dev
*__get_imsm_dev(struct imsm_super
*mpb
, __u8 index
)
720 if (index
>= mpb
->num_raid_devs
)
723 /* devices start after all disks */
724 offset
= ((void *) &mpb
->disk
[mpb
->num_disks
]) - _mpb
;
726 for (i
= 0; i
<= index
; i
++)
728 return _mpb
+ offset
;
730 offset
+= sizeof_imsm_dev(_mpb
+ offset
, 0);
735 static struct imsm_dev
*get_imsm_dev(struct intel_super
*super
, __u8 index
)
737 struct intel_dev
*dv
;
739 if (index
>= super
->anchor
->num_raid_devs
)
741 for (dv
= super
->devlist
; dv
; dv
= dv
->next
)
742 if (dv
->index
== index
)
750 * == 1 get second map
751 * == -1 than get map according to the current migr_state
753 static __u32
get_imsm_ord_tbl_ent(struct imsm_dev
*dev
,
757 struct imsm_map
*map
;
759 map
= get_imsm_map(dev
, second_map
);
761 /* top byte identifies disk under rebuild */
762 return __le32_to_cpu(map
->disk_ord_tbl
[slot
]);
765 #define ord_to_idx(ord) (((ord) << 8) >> 8)
766 static __u32
get_imsm_disk_idx(struct imsm_dev
*dev
, int slot
, int second_map
)
768 __u32 ord
= get_imsm_ord_tbl_ent(dev
, slot
, second_map
);
770 return ord_to_idx(ord
);
773 static void set_imsm_ord_tbl_ent(struct imsm_map
*map
, int slot
, __u32 ord
)
775 map
->disk_ord_tbl
[slot
] = __cpu_to_le32(ord
);
778 static int get_imsm_disk_slot(struct imsm_map
*map
, unsigned idx
)
783 for (slot
= 0; slot
< map
->num_members
; slot
++) {
784 ord
= __le32_to_cpu(map
->disk_ord_tbl
[slot
]);
785 if (ord_to_idx(ord
) == idx
)
792 static int get_imsm_raid_level(struct imsm_map
*map
)
794 if (map
->raid_level
== 1) {
795 if (map
->num_members
== 2)
801 return map
->raid_level
;
804 static int cmp_extent(const void *av
, const void *bv
)
806 const struct extent
*a
= av
;
807 const struct extent
*b
= bv
;
808 if (a
->start
< b
->start
)
810 if (a
->start
> b
->start
)
815 static int count_memberships(struct dl
*dl
, struct intel_super
*super
)
820 for (i
= 0; i
< super
->anchor
->num_raid_devs
; i
++) {
821 struct imsm_dev
*dev
= get_imsm_dev(super
, i
);
822 struct imsm_map
*map
= get_imsm_map(dev
, 0);
824 if (get_imsm_disk_slot(map
, dl
->index
) >= 0)
831 static __u32
imsm_min_reserved_sectors(struct intel_super
*super
);
833 static struct extent
*get_extents(struct intel_super
*super
, struct dl
*dl
)
835 /* find a list of used extents on the given physical device */
836 struct extent
*rv
, *e
;
838 int memberships
= count_memberships(dl
, super
);
841 /* trim the reserved area for spares, so they can join any array
842 * regardless of whether the OROM has assigned sectors from the
843 * IMSM_RESERVED_SECTORS region
846 reservation
= imsm_min_reserved_sectors(super
);
848 reservation
= MPB_SECTOR_CNT
+ IMSM_RESERVED_SECTORS
;
850 rv
= malloc(sizeof(struct extent
) * (memberships
+ 1));
855 for (i
= 0; i
< super
->anchor
->num_raid_devs
; i
++) {
856 struct imsm_dev
*dev
= get_imsm_dev(super
, i
);
857 struct imsm_map
*map
= get_imsm_map(dev
, 0);
859 if (get_imsm_disk_slot(map
, dl
->index
) >= 0) {
860 e
->start
= __le32_to_cpu(map
->pba_of_lba0
);
861 e
->size
= __le32_to_cpu(map
->blocks_per_member
);
865 qsort(rv
, memberships
, sizeof(*rv
), cmp_extent
);
867 /* determine the start of the metadata
868 * when no raid devices are defined use the default
869 * ...otherwise allow the metadata to truncate the value
870 * as is the case with older versions of imsm
873 struct extent
*last
= &rv
[memberships
- 1];
876 remainder
= __le32_to_cpu(dl
->disk
.total_blocks
) -
877 (last
->start
+ last
->size
);
878 /* round down to 1k block to satisfy precision of the kernel
882 /* make sure remainder is still sane */
883 if (remainder
< (unsigned)ROUND_UP(super
->len
, 512) >> 9)
884 remainder
= ROUND_UP(super
->len
, 512) >> 9;
885 if (reservation
> remainder
)
886 reservation
= remainder
;
888 e
->start
= __le32_to_cpu(dl
->disk
.total_blocks
) - reservation
;
893 /* try to determine how much space is reserved for metadata from
894 * the last get_extents() entry, otherwise fallback to the
897 static __u32
imsm_reserved_sectors(struct intel_super
*super
, struct dl
*dl
)
903 /* for spares just return a minimal reservation which will grow
904 * once the spare is picked up by an array
907 return MPB_SECTOR_CNT
;
909 e
= get_extents(super
, dl
);
911 return MPB_SECTOR_CNT
+ IMSM_RESERVED_SECTORS
;
913 /* scroll to last entry */
914 for (i
= 0; e
[i
].size
; i
++)
917 rv
= __le32_to_cpu(dl
->disk
.total_blocks
) - e
[i
].start
;
924 static int is_spare(struct imsm_disk
*disk
)
926 return (disk
->status
& SPARE_DISK
) == SPARE_DISK
;
929 static int is_configured(struct imsm_disk
*disk
)
931 return (disk
->status
& CONFIGURED_DISK
) == CONFIGURED_DISK
;
934 static int is_failed(struct imsm_disk
*disk
)
936 return (disk
->status
& FAILED_DISK
) == FAILED_DISK
;
939 /* try to determine how much space is reserved for metadata from
940 * the last get_extents() entry on the smallest active disk,
941 * otherwise fallback to the default
943 static __u32
imsm_min_reserved_sectors(struct intel_super
*super
)
947 __u32 min_active
, remainder
;
948 __u32 rv
= MPB_SECTOR_CNT
+ IMSM_RESERVED_SECTORS
;
949 struct dl
*dl
, *dl_min
= NULL
;
955 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
958 if (dl
->disk
.total_blocks
< min_active
|| min_active
== 0) {
960 min_active
= dl
->disk
.total_blocks
;
966 /* find last lba used by subarrays on the smallest active disk */
967 e
= get_extents(super
, dl_min
);
970 for (i
= 0; e
[i
].size
; i
++)
973 remainder
= min_active
- e
[i
].start
;
976 /* to give priority to recovery we should not require full
977 IMSM_RESERVED_SECTORS from the spare */
978 rv
= MPB_SECTOR_CNT
+ NUM_BLOCKS_DIRTY_STRIPE_REGION
;
980 /* if real reservation is smaller use that value */
981 return (remainder
< rv
) ? remainder
: rv
;
984 /* Return minimum size of a spare that can be used in this array*/
985 static unsigned long long min_acceptable_spare_size_imsm(struct supertype
*st
)
987 struct intel_super
*super
= st
->sb
;
991 unsigned long long rv
= 0;
995 /* find first active disk in array */
997 while (dl
&& (is_failed(&dl
->disk
) || dl
->index
== -1))
1001 /* find last lba used by subarrays */
1002 e
= get_extents(super
, dl
);
1005 for (i
= 0; e
[i
].size
; i
++)
1008 rv
= e
[i
-1].start
+ e
[i
-1].size
;
1011 /* add the amount of space needed for metadata */
1012 rv
= rv
+ imsm_min_reserved_sectors(super
);
1018 static __u64
blocks_per_migr_unit(struct intel_super
*super
,
1019 struct imsm_dev
*dev
);
1021 static void print_imsm_dev(struct intel_super
*super
,
1022 struct imsm_dev
*dev
,
1028 struct imsm_map
*map
= get_imsm_map(dev
, 0);
1029 struct imsm_map
*map2
= get_imsm_map(dev
, 1);
1033 printf("[%.16s]:\n", dev
->volume
);
1034 printf(" UUID : %s\n", uuid
);
1035 printf(" RAID Level : %d", get_imsm_raid_level(map
));
1037 printf(" <-- %d", get_imsm_raid_level(map2
));
1039 printf(" Members : %d", map
->num_members
);
1041 printf(" <-- %d", map2
->num_members
);
1043 printf(" Slots : [");
1044 for (i
= 0; i
< map
->num_members
; i
++) {
1045 ord
= get_imsm_ord_tbl_ent(dev
, i
, 0);
1046 printf("%s", ord
& IMSM_ORD_REBUILD
? "_" : "U");
1051 for (i
= 0; i
< map2
->num_members
; i
++) {
1052 ord
= get_imsm_ord_tbl_ent(dev
, i
, 1);
1053 printf("%s", ord
& IMSM_ORD_REBUILD
? "_" : "U");
1058 printf(" Failed disk : ");
1059 if (map
->failed_disk_num
== 0xff)
1062 printf("%i", map
->failed_disk_num
);
1064 slot
= get_imsm_disk_slot(map
, disk_idx
);
1066 ord
= get_imsm_ord_tbl_ent(dev
, slot
, -1);
1067 printf(" This Slot : %d%s\n", slot
,
1068 ord
& IMSM_ORD_REBUILD
? " (out-of-sync)" : "");
1070 printf(" This Slot : ?\n");
1071 sz
= __le32_to_cpu(dev
->size_high
);
1073 sz
+= __le32_to_cpu(dev
->size_low
);
1074 printf(" Array Size : %llu%s\n", (unsigned long long)sz
,
1075 human_size(sz
* 512));
1076 sz
= __le32_to_cpu(map
->blocks_per_member
);
1077 printf(" Per Dev Size : %llu%s\n", (unsigned long long)sz
,
1078 human_size(sz
* 512));
1079 printf(" Sector Offset : %u\n",
1080 __le32_to_cpu(map
->pba_of_lba0
));
1081 printf(" Num Stripes : %u\n",
1082 __le32_to_cpu(map
->num_data_stripes
));
1083 printf(" Chunk Size : %u KiB",
1084 __le16_to_cpu(map
->blocks_per_strip
) / 2);
1086 printf(" <-- %u KiB",
1087 __le16_to_cpu(map2
->blocks_per_strip
) / 2);
1089 printf(" Reserved : %d\n", __le32_to_cpu(dev
->reserved_blocks
));
1090 printf(" Migrate State : ");
1091 if (dev
->vol
.migr_state
) {
1092 if (migr_type(dev
) == MIGR_INIT
)
1093 printf("initialize\n");
1094 else if (migr_type(dev
) == MIGR_REBUILD
)
1095 printf("rebuild\n");
1096 else if (migr_type(dev
) == MIGR_VERIFY
)
1098 else if (migr_type(dev
) == MIGR_GEN_MIGR
)
1099 printf("general migration\n");
1100 else if (migr_type(dev
) == MIGR_STATE_CHANGE
)
1101 printf("state change\n");
1102 else if (migr_type(dev
) == MIGR_REPAIR
)
1105 printf("<unknown:%d>\n", migr_type(dev
));
1108 printf(" Map State : %s", map_state_str
[map
->map_state
]);
1109 if (dev
->vol
.migr_state
) {
1110 struct imsm_map
*map
= get_imsm_map(dev
, 1);
1112 printf(" <-- %s", map_state_str
[map
->map_state
]);
1113 printf("\n Checkpoint : %u (%llu)",
1114 __le32_to_cpu(dev
->vol
.curr_migr_unit
),
1115 (unsigned long long)blocks_per_migr_unit(super
, dev
));
1118 printf(" Dirty State : %s\n", dev
->vol
.dirty
? "dirty" : "clean");
1121 static void print_imsm_disk(struct imsm_disk
*disk
, int index
, __u32 reserved
)
1123 char str
[MAX_RAID_SERIAL_LEN
+ 1];
1126 if (index
< -1 || !disk
)
1130 snprintf(str
, MAX_RAID_SERIAL_LEN
+ 1, "%s", disk
->serial
);
1132 printf(" Disk%02d Serial : %s\n", index
, str
);
1134 printf(" Disk Serial : %s\n", str
);
1135 printf(" State :%s%s%s\n", is_spare(disk
) ? " spare" : "",
1136 is_configured(disk
) ? " active" : "",
1137 is_failed(disk
) ? " failed" : "");
1138 printf(" Id : %08x\n", __le32_to_cpu(disk
->scsi_id
));
1139 sz
= __le32_to_cpu(disk
->total_blocks
) - reserved
;
1140 printf(" Usable Size : %llu%s\n", (unsigned long long)sz
,
1141 human_size(sz
* 512));
1144 static int is_gen_migration(struct imsm_dev
*dev
);
1146 void examine_migr_rec_imsm(struct intel_super
*super
)
1148 struct migr_record
*migr_rec
= super
->migr_rec
;
1149 struct imsm_super
*mpb
= super
->anchor
;
1152 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
1153 struct imsm_dev
*dev
= __get_imsm_dev(mpb
, i
);
1154 if (is_gen_migration(dev
) == 0)
1157 printf("\nMigration Record Information:");
1158 if (super
->disks
->index
> 1) {
1159 printf(" Empty\n ");
1160 printf("Examine one of first two disks in array\n");
1163 printf("\n Status : ");
1164 if (__le32_to_cpu(migr_rec
->rec_status
) == UNIT_SRC_NORMAL
)
1167 printf("Contains Data\n");
1168 printf(" Current Unit : %u\n",
1169 __le32_to_cpu(migr_rec
->curr_migr_unit
));
1170 printf(" Family : %u\n",
1171 __le32_to_cpu(migr_rec
->family_num
));
1172 printf(" Ascending : %u\n",
1173 __le32_to_cpu(migr_rec
->ascending_migr
));
1174 printf(" Blocks Per Unit : %u\n",
1175 __le32_to_cpu(migr_rec
->blocks_per_unit
));
1176 printf(" Dest. Depth Per Unit : %u\n",
1177 __le32_to_cpu(migr_rec
->dest_depth_per_unit
));
1178 printf(" Checkpoint Area pba : %u\n",
1179 __le32_to_cpu(migr_rec
->ckpt_area_pba
));
1180 printf(" First member lba : %u\n",
1181 __le32_to_cpu(migr_rec
->dest_1st_member_lba
));
1182 printf(" Total Number of Units : %u\n",
1183 __le32_to_cpu(migr_rec
->num_migr_units
));
1184 printf(" Size of volume : %u\n",
1185 __le32_to_cpu(migr_rec
->post_migr_vol_cap
));
1186 printf(" Expansion space for LBA64 : %u\n",
1187 __le32_to_cpu(migr_rec
->post_migr_vol_cap_hi
));
1188 printf(" Record was read from : %u\n",
1189 __le32_to_cpu(migr_rec
->ckpt_read_disk_num
));
1194 #endif /* MDASSEMBLE */
1195 /*******************************************************************************
1196 * function: imsm_check_attributes
1197 * Description: Function checks if features represented by attributes flags
1198 * are supported by mdadm.
1200 * attributes - Attributes read from metadata
1202 * 0 - passed attributes contains unsupported features flags
1203 * 1 - all features are supported
1204 ******************************************************************************/
1205 static int imsm_check_attributes(__u32 attributes
)
1208 __u32 not_supported
= MPB_ATTRIB_SUPPORTED
^0xffffffff;
1210 not_supported
&= ~MPB_ATTRIB_IGNORED
;
1212 not_supported
&= attributes
;
1213 if (not_supported
) {
1214 fprintf(stderr
, Name
"(IMSM): Unsupported attributes : %x\n",
1215 (unsigned)__le32_to_cpu(not_supported
));
1216 if (not_supported
& MPB_ATTRIB_CHECKSUM_VERIFY
) {
1217 dprintf("\t\tMPB_ATTRIB_CHECKSUM_VERIFY \n");
1218 not_supported
^= MPB_ATTRIB_CHECKSUM_VERIFY
;
1220 if (not_supported
& MPB_ATTRIB_2TB
) {
1221 dprintf("\t\tMPB_ATTRIB_2TB\n");
1222 not_supported
^= MPB_ATTRIB_2TB
;
1224 if (not_supported
& MPB_ATTRIB_RAID0
) {
1225 dprintf("\t\tMPB_ATTRIB_RAID0\n");
1226 not_supported
^= MPB_ATTRIB_RAID0
;
1228 if (not_supported
& MPB_ATTRIB_RAID1
) {
1229 dprintf("\t\tMPB_ATTRIB_RAID1\n");
1230 not_supported
^= MPB_ATTRIB_RAID1
;
1232 if (not_supported
& MPB_ATTRIB_RAID10
) {
1233 dprintf("\t\tMPB_ATTRIB_RAID10\n");
1234 not_supported
^= MPB_ATTRIB_RAID10
;
1236 if (not_supported
& MPB_ATTRIB_RAID1E
) {
1237 dprintf("\t\tMPB_ATTRIB_RAID1E\n");
1238 not_supported
^= MPB_ATTRIB_RAID1E
;
1240 if (not_supported
& MPB_ATTRIB_RAID5
) {
1241 dprintf("\t\tMPB_ATTRIB_RAID5\n");
1242 not_supported
^= MPB_ATTRIB_RAID5
;
1244 if (not_supported
& MPB_ATTRIB_RAIDCNG
) {
1245 dprintf("\t\tMPB_ATTRIB_RAIDCNG\n");
1246 not_supported
^= MPB_ATTRIB_RAIDCNG
;
1248 if (not_supported
& MPB_ATTRIB_BBM
) {
1249 dprintf("\t\tMPB_ATTRIB_BBM\n");
1250 not_supported
^= MPB_ATTRIB_BBM
;
1252 if (not_supported
& MPB_ATTRIB_CHECKSUM_VERIFY
) {
1253 dprintf("\t\tMPB_ATTRIB_CHECKSUM_VERIFY (== MPB_ATTRIB_LEGACY)\n");
1254 not_supported
^= MPB_ATTRIB_CHECKSUM_VERIFY
;
1256 if (not_supported
& MPB_ATTRIB_EXP_STRIPE_SIZE
) {
1257 dprintf("\t\tMPB_ATTRIB_EXP_STRIP_SIZE\n");
1258 not_supported
^= MPB_ATTRIB_EXP_STRIPE_SIZE
;
1260 if (not_supported
& MPB_ATTRIB_2TB_DISK
) {
1261 dprintf("\t\tMPB_ATTRIB_2TB_DISK\n");
1262 not_supported
^= MPB_ATTRIB_2TB_DISK
;
1264 if (not_supported
& MPB_ATTRIB_NEVER_USE2
) {
1265 dprintf("\t\tMPB_ATTRIB_NEVER_USE2\n");
1266 not_supported
^= MPB_ATTRIB_NEVER_USE2
;
1268 if (not_supported
& MPB_ATTRIB_NEVER_USE
) {
1269 dprintf("\t\tMPB_ATTRIB_NEVER_USE\n");
1270 not_supported
^= MPB_ATTRIB_NEVER_USE
;
1274 dprintf(Name
"(IMSM): Unknown attributes : %x\n", not_supported
);
1283 static void getinfo_super_imsm(struct supertype
*st
, struct mdinfo
*info
, char *map
);
1285 static void examine_super_imsm(struct supertype
*st
, char *homehost
)
1287 struct intel_super
*super
= st
->sb
;
1288 struct imsm_super
*mpb
= super
->anchor
;
1289 char str
[MAX_SIGNATURE_LENGTH
];
1294 __u32 reserved
= imsm_reserved_sectors(super
, super
->disks
);
1297 snprintf(str
, MPB_SIG_LEN
, "%s", mpb
->sig
);
1298 printf(" Magic : %s\n", str
);
1299 snprintf(str
, strlen(MPB_VERSION_RAID0
), "%s", get_imsm_version(mpb
));
1300 printf(" Version : %s\n", get_imsm_version(mpb
));
1301 printf(" Orig Family : %08x\n", __le32_to_cpu(mpb
->orig_family_num
));
1302 printf(" Family : %08x\n", __le32_to_cpu(mpb
->family_num
));
1303 printf(" Generation : %08x\n", __le32_to_cpu(mpb
->generation_num
));
1304 printf(" Attributes : ");
1305 if (imsm_check_attributes(mpb
->attributes
))
1306 printf("All supported\n");
1308 printf("not supported\n");
1309 getinfo_super_imsm(st
, &info
, NULL
);
1310 fname_from_uuid(st
, &info
, nbuf
, ':');
1311 printf(" UUID : %s\n", nbuf
+ 5);
1312 sum
= __le32_to_cpu(mpb
->check_sum
);
1313 printf(" Checksum : %08x %s\n", sum
,
1314 __gen_imsm_checksum(mpb
) == sum
? "correct" : "incorrect");
1315 printf(" MPB Sectors : %d\n", mpb_sectors(mpb
));
1316 printf(" Disks : %d\n", mpb
->num_disks
);
1317 printf(" RAID Devices : %d\n", mpb
->num_raid_devs
);
1318 print_imsm_disk(__get_imsm_disk(mpb
, super
->disks
->index
), super
->disks
->index
, reserved
);
1319 if (super
->bbm_log
) {
1320 struct bbm_log
*log
= super
->bbm_log
;
1323 printf("Bad Block Management Log:\n");
1324 printf(" Log Size : %d\n", __le32_to_cpu(mpb
->bbm_log_size
));
1325 printf(" Signature : %x\n", __le32_to_cpu(log
->signature
));
1326 printf(" Entry Count : %d\n", __le32_to_cpu(log
->entry_count
));
1327 printf(" Spare Blocks : %d\n", __le32_to_cpu(log
->reserved_spare_block_count
));
1328 printf(" First Spare : %llx\n",
1329 (unsigned long long) __le64_to_cpu(log
->first_spare_lba
));
1331 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
1333 struct imsm_dev
*dev
= __get_imsm_dev(mpb
, i
);
1335 super
->current_vol
= i
;
1336 getinfo_super_imsm(st
, &info
, NULL
);
1337 fname_from_uuid(st
, &info
, nbuf
, ':');
1338 print_imsm_dev(super
, dev
, nbuf
+ 5, super
->disks
->index
);
1340 for (i
= 0; i
< mpb
->num_disks
; i
++) {
1341 if (i
== super
->disks
->index
)
1343 print_imsm_disk(__get_imsm_disk(mpb
, i
), i
, reserved
);
1346 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
1347 if (dl
->index
== -1)
1348 print_imsm_disk(&dl
->disk
, -1, reserved
);
1350 examine_migr_rec_imsm(super
);
1353 static void brief_examine_super_imsm(struct supertype
*st
, int verbose
)
1355 /* We just write a generic IMSM ARRAY entry */
1358 struct intel_super
*super
= st
->sb
;
1360 if (!super
->anchor
->num_raid_devs
) {
1361 printf("ARRAY metadata=imsm\n");
1365 getinfo_super_imsm(st
, &info
, NULL
);
1366 fname_from_uuid(st
, &info
, nbuf
, ':');
1367 printf("ARRAY metadata=imsm UUID=%s\n", nbuf
+ 5);
1370 static void brief_examine_subarrays_imsm(struct supertype
*st
, int verbose
)
1372 /* We just write a generic IMSM ARRAY entry */
1376 struct intel_super
*super
= st
->sb
;
1379 if (!super
->anchor
->num_raid_devs
)
1382 getinfo_super_imsm(st
, &info
, NULL
);
1383 fname_from_uuid(st
, &info
, nbuf
, ':');
1384 for (i
= 0; i
< super
->anchor
->num_raid_devs
; i
++) {
1385 struct imsm_dev
*dev
= get_imsm_dev(super
, i
);
1387 super
->current_vol
= i
;
1388 getinfo_super_imsm(st
, &info
, NULL
);
1389 fname_from_uuid(st
, &info
, nbuf1
, ':');
1390 printf("ARRAY /dev/md/%.16s container=%s member=%d UUID=%s\n",
1391 dev
->volume
, nbuf
+ 5, i
, nbuf1
+ 5);
1395 static void export_examine_super_imsm(struct supertype
*st
)
1397 struct intel_super
*super
= st
->sb
;
1398 struct imsm_super
*mpb
= super
->anchor
;
1402 getinfo_super_imsm(st
, &info
, NULL
);
1403 fname_from_uuid(st
, &info
, nbuf
, ':');
1404 printf("MD_METADATA=imsm\n");
1405 printf("MD_LEVEL=container\n");
1406 printf("MD_UUID=%s\n", nbuf
+5);
1407 printf("MD_DEVICES=%u\n", mpb
->num_disks
);
1410 static void detail_super_imsm(struct supertype
*st
, char *homehost
)
1415 getinfo_super_imsm(st
, &info
, NULL
);
1416 fname_from_uuid(st
, &info
, nbuf
, ':');
1417 printf("\n UUID : %s\n", nbuf
+ 5);
1420 static void brief_detail_super_imsm(struct supertype
*st
)
1424 getinfo_super_imsm(st
, &info
, NULL
);
1425 fname_from_uuid(st
, &info
, nbuf
, ':');
1426 printf(" UUID=%s", nbuf
+ 5);
1429 static int imsm_read_serial(int fd
, char *devname
, __u8
*serial
);
1430 static void fd2devname(int fd
, char *name
);
1432 static int ahci_enumerate_ports(const char *hba_path
, int port_count
, int host_base
, int verbose
)
1434 /* dump an unsorted list of devices attached to AHCI Intel storage
1435 * controller, as well as non-connected ports
1437 int hba_len
= strlen(hba_path
) + 1;
1442 unsigned long port_mask
= (1 << port_count
) - 1;
1444 if (port_count
> (int)sizeof(port_mask
) * 8) {
1446 fprintf(stderr
, Name
": port_count %d out of range\n", port_count
);
1450 /* scroll through /sys/dev/block looking for devices attached to
1453 dir
= opendir("/sys/dev/block");
1454 for (ent
= dir
? readdir(dir
) : NULL
; ent
; ent
= readdir(dir
)) {
1465 if (sscanf(ent
->d_name
, "%d:%d", &major
, &minor
) != 2)
1467 path
= devt_to_devpath(makedev(major
, minor
));
1470 if (!path_attached_to_hba(path
, hba_path
)) {
1476 /* retrieve the scsi device type */
1477 if (asprintf(&device
, "/sys/dev/block/%d:%d/device/xxxxxxx", major
, minor
) < 0) {
1479 fprintf(stderr
, Name
": failed to allocate 'device'\n");
1483 sprintf(device
, "/sys/dev/block/%d:%d/device/type", major
, minor
);
1484 if (load_sys(device
, buf
) != 0) {
1486 fprintf(stderr
, Name
": failed to read device type for %s\n",
1492 type
= strtoul(buf
, NULL
, 10);
1494 /* if it's not a disk print the vendor and model */
1495 if (!(type
== 0 || type
== 7 || type
== 14)) {
1498 sprintf(device
, "/sys/dev/block/%d:%d/device/vendor", major
, minor
);
1499 if (load_sys(device
, buf
) == 0) {
1500 strncpy(vendor
, buf
, sizeof(vendor
));
1501 vendor
[sizeof(vendor
) - 1] = '\0';
1502 c
= (char *) &vendor
[sizeof(vendor
) - 1];
1503 while (isspace(*c
) || *c
== '\0')
1507 sprintf(device
, "/sys/dev/block/%d:%d/device/model", major
, minor
);
1508 if (load_sys(device
, buf
) == 0) {
1509 strncpy(model
, buf
, sizeof(model
));
1510 model
[sizeof(model
) - 1] = '\0';
1511 c
= (char *) &model
[sizeof(model
) - 1];
1512 while (isspace(*c
) || *c
== '\0')
1516 if (vendor
[0] && model
[0])
1517 sprintf(buf
, "%.64s %.64s", vendor
, model
);
1519 switch (type
) { /* numbers from hald/linux/device.c */
1520 case 1: sprintf(buf
, "tape"); break;
1521 case 2: sprintf(buf
, "printer"); break;
1522 case 3: sprintf(buf
, "processor"); break;
1524 case 5: sprintf(buf
, "cdrom"); break;
1525 case 6: sprintf(buf
, "scanner"); break;
1526 case 8: sprintf(buf
, "media_changer"); break;
1527 case 9: sprintf(buf
, "comm"); break;
1528 case 12: sprintf(buf
, "raid"); break;
1529 default: sprintf(buf
, "unknown");
1535 /* chop device path to 'host%d' and calculate the port number */
1536 c
= strchr(&path
[hba_len
], '/');
1539 fprintf(stderr
, Name
": %s - invalid path name\n", path
+ hba_len
);
1544 if (sscanf(&path
[hba_len
], "host%d", &port
) == 1)
1548 *c
= '/'; /* repair the full string */
1549 fprintf(stderr
, Name
": failed to determine port number for %s\n",
1556 /* mark this port as used */
1557 port_mask
&= ~(1 << port
);
1559 /* print out the device information */
1561 printf(" Port%d : - non-disk device (%s) -\n", port
, buf
);
1565 fd
= dev_open(ent
->d_name
, O_RDONLY
);
1567 printf(" Port%d : - disk info unavailable -\n", port
);
1569 fd2devname(fd
, buf
);
1570 printf(" Port%d : %s", port
, buf
);
1571 if (imsm_read_serial(fd
, NULL
, (__u8
*) buf
) == 0)
1572 printf(" (%.*s)\n", MAX_RAID_SERIAL_LEN
, buf
);
1587 for (i
= 0; i
< port_count
; i
++)
1588 if (port_mask
& (1 << i
))
1589 printf(" Port%d : - no device attached -\n", i
);
1595 static void print_found_intel_controllers(struct sys_dev
*elem
)
1597 for (; elem
; elem
= elem
->next
) {
1598 fprintf(stderr
, Name
": found Intel(R) ");
1599 if (elem
->type
== SYS_DEV_SATA
)
1600 fprintf(stderr
, "SATA ");
1601 else if (elem
->type
== SYS_DEV_SAS
)
1602 fprintf(stderr
, "SAS ");
1603 fprintf(stderr
, "RAID controller");
1605 fprintf(stderr
, " at %s", elem
->pci_id
);
1606 fprintf(stderr
, ".\n");
1611 static int ahci_get_port_count(const char *hba_path
, int *port_count
)
1618 if ((dir
= opendir(hba_path
)) == NULL
)
1621 for (ent
= readdir(dir
); ent
; ent
= readdir(dir
)) {
1624 if (sscanf(ent
->d_name
, "host%d", &host
) != 1)
1626 if (*port_count
== 0)
1628 else if (host
< host_base
)
1631 if (host
+ 1 > *port_count
+ host_base
)
1632 *port_count
= host
+ 1 - host_base
;
1638 static void print_imsm_capability(const struct imsm_orom
*orom
)
1640 printf(" Platform : Intel(R) Matrix Storage Manager\n");
1641 printf(" Version : %d.%d.%d.%d\n", orom
->major_ver
, orom
->minor_ver
,
1642 orom
->hotfix_ver
, orom
->build
);
1643 printf(" RAID Levels :%s%s%s%s%s\n",
1644 imsm_orom_has_raid0(orom
) ? " raid0" : "",
1645 imsm_orom_has_raid1(orom
) ? " raid1" : "",
1646 imsm_orom_has_raid1e(orom
) ? " raid1e" : "",
1647 imsm_orom_has_raid10(orom
) ? " raid10" : "",
1648 imsm_orom_has_raid5(orom
) ? " raid5" : "");
1649 printf(" Chunk Sizes :%s%s%s%s%s%s%s%s%s%s%s%s%s%s%s%s\n",
1650 imsm_orom_has_chunk(orom
, 2) ? " 2k" : "",
1651 imsm_orom_has_chunk(orom
, 4) ? " 4k" : "",
1652 imsm_orom_has_chunk(orom
, 8) ? " 8k" : "",
1653 imsm_orom_has_chunk(orom
, 16) ? " 16k" : "",
1654 imsm_orom_has_chunk(orom
, 32) ? " 32k" : "",
1655 imsm_orom_has_chunk(orom
, 64) ? " 64k" : "",
1656 imsm_orom_has_chunk(orom
, 128) ? " 128k" : "",
1657 imsm_orom_has_chunk(orom
, 256) ? " 256k" : "",
1658 imsm_orom_has_chunk(orom
, 512) ? " 512k" : "",
1659 imsm_orom_has_chunk(orom
, 1024*1) ? " 1M" : "",
1660 imsm_orom_has_chunk(orom
, 1024*2) ? " 2M" : "",
1661 imsm_orom_has_chunk(orom
, 1024*4) ? " 4M" : "",
1662 imsm_orom_has_chunk(orom
, 1024*8) ? " 8M" : "",
1663 imsm_orom_has_chunk(orom
, 1024*16) ? " 16M" : "",
1664 imsm_orom_has_chunk(orom
, 1024*32) ? " 32M" : "",
1665 imsm_orom_has_chunk(orom
, 1024*64) ? " 64M" : "");
1666 printf(" Max Disks : %d\n", orom
->tds
);
1667 printf(" Max Volumes : %d\n", orom
->vpa
);
1671 static int detail_platform_imsm(int verbose
, int enumerate_only
)
1673 /* There are two components to imsm platform support, the ahci SATA
1674 * controller and the option-rom. To find the SATA controller we
1675 * simply look in /sys/bus/pci/drivers/ahci to see if an ahci
1676 * controller with the Intel vendor id is present. This approach
1677 * allows mdadm to leverage the kernel's ahci detection logic, with the
1678 * caveat that if ahci.ko is not loaded mdadm will not be able to
1679 * detect platform raid capabilities. The option-rom resides in a
1680 * platform "Adapter ROM". We scan for its signature to retrieve the
1681 * platform capabilities. If raid support is disabled in the BIOS the
1682 * option-rom capability structure will not be available.
1684 const struct imsm_orom
*orom
;
1685 struct sys_dev
*list
, *hba
;
1690 if (enumerate_only
) {
1691 if (check_env("IMSM_NO_PLATFORM"))
1693 list
= find_intel_devices();
1696 for (hba
= list
; hba
; hba
= hba
->next
) {
1697 orom
= find_imsm_capability(hba
->type
);
1703 free_sys_dev(&list
);
1707 list
= find_intel_devices();
1710 fprintf(stderr
, Name
": no active Intel(R) RAID "
1711 "controller found.\n");
1712 free_sys_dev(&list
);
1715 print_found_intel_controllers(list
);
1717 for (hba
= list
; hba
; hba
= hba
->next
) {
1718 orom
= find_imsm_capability(hba
->type
);
1720 fprintf(stderr
, Name
": imsm capabilities not found for controller: %s (type %s)\n",
1721 hba
->path
, get_sys_dev_type(hba
->type
));
1723 print_imsm_capability(orom
);
1726 for (hba
= list
; hba
; hba
= hba
->next
) {
1727 printf(" I/O Controller : %s (%s)\n",
1728 hba
->path
, get_sys_dev_type(hba
->type
));
1730 if (hba
->type
== SYS_DEV_SATA
) {
1731 host_base
= ahci_get_port_count(hba
->path
, &port_count
);
1732 if (ahci_enumerate_ports(hba
->path
, port_count
, host_base
, verbose
)) {
1734 fprintf(stderr
, Name
": failed to enumerate "
1735 "ports on SATA controller at %s.", hba
->pci_id
);
1741 free_sys_dev(&list
);
1746 static int match_home_imsm(struct supertype
*st
, char *homehost
)
1748 /* the imsm metadata format does not specify any host
1749 * identification information. We return -1 since we can never
1750 * confirm nor deny whether a given array is "meant" for this
1751 * host. We rely on compare_super and the 'family_num' fields to
1752 * exclude member disks that do not belong, and we rely on
1753 * mdadm.conf to specify the arrays that should be assembled.
1754 * Auto-assembly may still pick up "foreign" arrays.
1760 static void uuid_from_super_imsm(struct supertype
*st
, int uuid
[4])
1762 /* The uuid returned here is used for:
1763 * uuid to put into bitmap file (Create, Grow)
1764 * uuid for backup header when saving critical section (Grow)
1765 * comparing uuids when re-adding a device into an array
1766 * In these cases the uuid required is that of the data-array,
1767 * not the device-set.
1768 * uuid to recognise same set when adding a missing device back
1769 * to an array. This is a uuid for the device-set.
1771 * For each of these we can make do with a truncated
1772 * or hashed uuid rather than the original, as long as
1774 * In each case the uuid required is that of the data-array,
1775 * not the device-set.
1777 /* imsm does not track uuid's so we synthesis one using sha1 on
1778 * - The signature (Which is constant for all imsm array, but no matter)
1779 * - the orig_family_num of the container
1780 * - the index number of the volume
1781 * - the 'serial' number of the volume.
1782 * Hopefully these are all constant.
1784 struct intel_super
*super
= st
->sb
;
1787 struct sha1_ctx ctx
;
1788 struct imsm_dev
*dev
= NULL
;
1791 /* some mdadm versions failed to set ->orig_family_num, in which
1792 * case fall back to ->family_num. orig_family_num will be
1793 * fixed up with the first metadata update.
1795 family_num
= super
->anchor
->orig_family_num
;
1796 if (family_num
== 0)
1797 family_num
= super
->anchor
->family_num
;
1798 sha1_init_ctx(&ctx
);
1799 sha1_process_bytes(super
->anchor
->sig
, MPB_SIG_LEN
, &ctx
);
1800 sha1_process_bytes(&family_num
, sizeof(__u32
), &ctx
);
1801 if (super
->current_vol
>= 0)
1802 dev
= get_imsm_dev(super
, super
->current_vol
);
1804 __u32 vol
= super
->current_vol
;
1805 sha1_process_bytes(&vol
, sizeof(vol
), &ctx
);
1806 sha1_process_bytes(dev
->volume
, MAX_RAID_SERIAL_LEN
, &ctx
);
1808 sha1_finish_ctx(&ctx
, buf
);
1809 memcpy(uuid
, buf
, 4*4);
1814 get_imsm_numerical_version(struct imsm_super
*mpb
, int *m
, int *p
)
1816 __u8
*v
= get_imsm_version(mpb
);
1817 __u8
*end
= mpb
->sig
+ MAX_SIGNATURE_LENGTH
;
1818 char major
[] = { 0, 0, 0 };
1819 char minor
[] = { 0 ,0, 0 };
1820 char patch
[] = { 0, 0, 0 };
1821 char *ver_parse
[] = { major
, minor
, patch
};
1825 while (*v
!= '\0' && v
< end
) {
1826 if (*v
!= '.' && j
< 2)
1827 ver_parse
[i
][j
++] = *v
;
1835 *m
= strtol(minor
, NULL
, 0);
1836 *p
= strtol(patch
, NULL
, 0);
1840 static __u32
migr_strip_blocks_resync(struct imsm_dev
*dev
)
1842 /* migr_strip_size when repairing or initializing parity */
1843 struct imsm_map
*map
= get_imsm_map(dev
, 0);
1844 __u32 chunk
= __le32_to_cpu(map
->blocks_per_strip
);
1846 switch (get_imsm_raid_level(map
)) {
1851 return 128*1024 >> 9;
1855 static __u32
migr_strip_blocks_rebuild(struct imsm_dev
*dev
)
1857 /* migr_strip_size when rebuilding a degraded disk, no idea why
1858 * this is different than migr_strip_size_resync(), but it's good
1861 struct imsm_map
*map
= get_imsm_map(dev
, 1);
1862 __u32 chunk
= __le32_to_cpu(map
->blocks_per_strip
);
1864 switch (get_imsm_raid_level(map
)) {
1867 if (map
->num_members
% map
->num_domains
== 0)
1868 return 128*1024 >> 9;
1872 return max((__u32
) 64*1024 >> 9, chunk
);
1874 return 128*1024 >> 9;
1878 static __u32
num_stripes_per_unit_resync(struct imsm_dev
*dev
)
1880 struct imsm_map
*lo
= get_imsm_map(dev
, 0);
1881 struct imsm_map
*hi
= get_imsm_map(dev
, 1);
1882 __u32 lo_chunk
= __le32_to_cpu(lo
->blocks_per_strip
);
1883 __u32 hi_chunk
= __le32_to_cpu(hi
->blocks_per_strip
);
1885 return max((__u32
) 1, hi_chunk
/ lo_chunk
);
1888 static __u32
num_stripes_per_unit_rebuild(struct imsm_dev
*dev
)
1890 struct imsm_map
*lo
= get_imsm_map(dev
, 0);
1891 int level
= get_imsm_raid_level(lo
);
1893 if (level
== 1 || level
== 10) {
1894 struct imsm_map
*hi
= get_imsm_map(dev
, 1);
1896 return hi
->num_domains
;
1898 return num_stripes_per_unit_resync(dev
);
1901 static __u8
imsm_num_data_members(struct imsm_dev
*dev
, int second_map
)
1903 /* named 'imsm_' because raid0, raid1 and raid10
1904 * counter-intuitively have the same number of data disks
1906 struct imsm_map
*map
= get_imsm_map(dev
, second_map
);
1908 switch (get_imsm_raid_level(map
)) {
1912 return map
->num_members
;
1914 return map
->num_members
- 1;
1916 dprintf("%s: unsupported raid level\n", __func__
);
1921 static __u32
parity_segment_depth(struct imsm_dev
*dev
)
1923 struct imsm_map
*map
= get_imsm_map(dev
, 0);
1924 __u32 chunk
= __le32_to_cpu(map
->blocks_per_strip
);
1926 switch(get_imsm_raid_level(map
)) {
1929 return chunk
* map
->num_domains
;
1931 return chunk
* map
->num_members
;
1937 static __u32
map_migr_block(struct imsm_dev
*dev
, __u32 block
)
1939 struct imsm_map
*map
= get_imsm_map(dev
, 1);
1940 __u32 chunk
= __le32_to_cpu(map
->blocks_per_strip
);
1941 __u32 strip
= block
/ chunk
;
1943 switch (get_imsm_raid_level(map
)) {
1946 __u32 vol_strip
= (strip
* map
->num_domains
) + 1;
1947 __u32 vol_stripe
= vol_strip
/ map
->num_members
;
1949 return vol_stripe
* chunk
+ block
% chunk
;
1951 __u32 stripe
= strip
/ (map
->num_members
- 1);
1953 return stripe
* chunk
+ block
% chunk
;
1960 static __u64
blocks_per_migr_unit(struct intel_super
*super
,
1961 struct imsm_dev
*dev
)
1963 /* calculate the conversion factor between per member 'blocks'
1964 * (md/{resync,rebuild}_start) and imsm migration units, return
1965 * 0 for the 'not migrating' and 'unsupported migration' cases
1967 if (!dev
->vol
.migr_state
)
1970 switch (migr_type(dev
)) {
1971 case MIGR_GEN_MIGR
: {
1972 struct migr_record
*migr_rec
= super
->migr_rec
;
1973 return __le32_to_cpu(migr_rec
->blocks_per_unit
);
1978 struct imsm_map
*map
= get_imsm_map(dev
, 0);
1979 __u32 stripes_per_unit
;
1980 __u32 blocks_per_unit
;
1989 /* yes, this is really the translation of migr_units to
1990 * per-member blocks in the 'resync' case
1992 stripes_per_unit
= num_stripes_per_unit_resync(dev
);
1993 migr_chunk
= migr_strip_blocks_resync(dev
);
1994 disks
= imsm_num_data_members(dev
, 0);
1995 blocks_per_unit
= stripes_per_unit
* migr_chunk
* disks
;
1996 stripe
= __le16_to_cpu(map
->blocks_per_strip
) * disks
;
1997 segment
= blocks_per_unit
/ stripe
;
1998 block_rel
= blocks_per_unit
- segment
* stripe
;
1999 parity_depth
= parity_segment_depth(dev
);
2000 block_map
= map_migr_block(dev
, block_rel
);
2001 return block_map
+ parity_depth
* segment
;
2003 case MIGR_REBUILD
: {
2004 __u32 stripes_per_unit
;
2007 stripes_per_unit
= num_stripes_per_unit_rebuild(dev
);
2008 migr_chunk
= migr_strip_blocks_rebuild(dev
);
2009 return migr_chunk
* stripes_per_unit
;
2011 case MIGR_STATE_CHANGE
:
2017 static int imsm_level_to_layout(int level
)
2025 return ALGORITHM_LEFT_ASYMMETRIC
;
2032 /*******************************************************************************
2033 * Function: read_imsm_migr_rec
2034 * Description: Function reads imsm migration record from last sector of disk
2036 * fd : disk descriptor
2037 * super : metadata info
2041 ******************************************************************************/
2042 static int read_imsm_migr_rec(int fd
, struct intel_super
*super
)
2045 unsigned long long dsize
;
2047 get_dev_size(fd
, NULL
, &dsize
);
2048 if (lseek64(fd
, dsize
- 512, SEEK_SET
) < 0) {
2050 Name
": Cannot seek to anchor block: %s\n",
2054 if (read(fd
, super
->migr_rec_buf
, 512) != 512) {
2056 Name
": Cannot read migr record block: %s\n",
2066 /*******************************************************************************
2067 * Function: load_imsm_migr_rec
2068 * Description: Function reads imsm migration record (it is stored at the last
2071 * super : imsm internal array info
2072 * info : general array info
2076 ******************************************************************************/
2077 static int load_imsm_migr_rec(struct intel_super
*super
, struct mdinfo
*info
)
2080 struct dl
*dl
= NULL
;
2086 for (sd
= info
->devs
; sd
; sd
= sd
->next
) {
2087 /* read only from one of the first two slots */
2088 if ((sd
->disk
.raid_disk
> 1) ||
2089 (sd
->disk
.raid_disk
< 0))
2091 sprintf(nm
, "%d:%d", sd
->disk
.major
, sd
->disk
.minor
);
2092 fd
= dev_open(nm
, O_RDONLY
);
2098 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
2099 /* read only from one of the first two slots */
2102 sprintf(nm
, "%d:%d", dl
->major
, dl
->minor
);
2103 fd
= dev_open(nm
, O_RDONLY
);
2110 retval
= read_imsm_migr_rec(fd
, super
);
2119 /*******************************************************************************
2120 * function: imsm_create_metadata_checkpoint_update
2121 * Description: It creates update for checkpoint change.
2123 * super : imsm internal array info
2124 * u : pointer to prepared update
2127 * If length is equal to 0, input pointer u contains no update
2128 ******************************************************************************/
2129 static int imsm_create_metadata_checkpoint_update(
2130 struct intel_super
*super
,
2131 struct imsm_update_general_migration_checkpoint
**u
)
2134 int update_memory_size
= 0;
2136 dprintf("imsm_create_metadata_checkpoint_update(enter)\n");
2142 /* size of all update data without anchor */
2143 update_memory_size
=
2144 sizeof(struct imsm_update_general_migration_checkpoint
);
2146 *u
= calloc(1, update_memory_size
);
2148 dprintf("error: cannot get memory for "
2149 "imsm_create_metadata_checkpoint_update update\n");
2152 (*u
)->type
= update_general_migration_checkpoint
;
2153 (*u
)->curr_migr_unit
= __le32_to_cpu(super
->migr_rec
->curr_migr_unit
);
2154 dprintf("imsm_create_metadata_checkpoint_update: prepared for %u\n",
2155 (*u
)->curr_migr_unit
);
2157 return update_memory_size
;
2161 static void imsm_update_metadata_locally(struct supertype
*st
,
2162 void *buf
, int len
);
2164 /*******************************************************************************
2165 * Function: write_imsm_migr_rec
2166 * Description: Function writes imsm migration record
2167 * (at the last sector of disk)
2169 * super : imsm internal array info
2173 ******************************************************************************/
2174 static int write_imsm_migr_rec(struct supertype
*st
)
2176 struct intel_super
*super
= st
->sb
;
2177 unsigned long long dsize
;
2183 struct imsm_update_general_migration_checkpoint
*u
;
2185 for (sd
= super
->disks
; sd
; sd
= sd
->next
) {
2186 /* write to 2 first slots only */
2187 if ((sd
->index
< 0) || (sd
->index
> 1))
2189 sprintf(nm
, "%d:%d", sd
->major
, sd
->minor
);
2190 fd
= dev_open(nm
, O_RDWR
);
2193 get_dev_size(fd
, NULL
, &dsize
);
2194 if (lseek64(fd
, dsize
- 512, SEEK_SET
) < 0) {
2196 Name
": Cannot seek to anchor block: %s\n",
2200 if (write(fd
, super
->migr_rec_buf
, 512) != 512) {
2202 Name
": Cannot write migr record block: %s\n",
2209 /* update checkpoint information in metadata */
2210 len
= imsm_create_metadata_checkpoint_update(super
, &u
);
2213 dprintf("imsm: Cannot prepare update\n");
2216 /* update metadata locally */
2217 imsm_update_metadata_locally(st
, u
, len
);
2218 /* and possibly remotely */
2219 if (st
->update_tail
) {
2220 append_metadata_update(st
, u
, len
);
2221 /* during reshape we do all work inside metadata handler
2222 * manage_reshape(), so metadata update has to be triggered
2225 flush_metadata_updates(st
);
2226 st
->update_tail
= &st
->updates
;
2236 #endif /* MDASSEMBLE */
2238 /* spare/missing disks activations are not allowe when
2239 * array/container performs reshape operation, because
2240 * all arrays in container works on the same disks set
2242 int imsm_reshape_blocks_arrays_changes(struct intel_super
*super
)
2245 struct intel_dev
*i_dev
;
2246 struct imsm_dev
*dev
;
2248 /* check whole container
2250 for (i_dev
= super
->devlist
; i_dev
; i_dev
= i_dev
->next
) {
2252 if (is_gen_migration(dev
)) {
2253 /* No repair during any migration in container
2262 static void getinfo_super_imsm_volume(struct supertype
*st
, struct mdinfo
*info
, char *dmap
)
2264 struct intel_super
*super
= st
->sb
;
2265 struct migr_record
*migr_rec
= super
->migr_rec
;
2266 struct imsm_dev
*dev
= get_imsm_dev(super
, super
->current_vol
);
2267 struct imsm_map
*map
= get_imsm_map(dev
, 0);
2268 struct imsm_map
*prev_map
= get_imsm_map(dev
, 1);
2269 struct imsm_map
*map_to_analyse
= map
;
2272 unsigned int component_size_alligment
;
2273 int map_disks
= info
->array
.raid_disks
;
2275 memset(info
, 0, sizeof(*info
));
2277 map_to_analyse
= prev_map
;
2279 dl
= super
->current_disk
;
2281 info
->container_member
= super
->current_vol
;
2282 info
->array
.raid_disks
= map
->num_members
;
2283 info
->array
.level
= get_imsm_raid_level(map_to_analyse
);
2284 info
->array
.layout
= imsm_level_to_layout(info
->array
.level
);
2285 info
->array
.md_minor
= -1;
2286 info
->array
.ctime
= 0;
2287 info
->array
.utime
= 0;
2288 info
->array
.chunk_size
=
2289 __le16_to_cpu(map_to_analyse
->blocks_per_strip
) << 9;
2290 info
->array
.state
= !dev
->vol
.dirty
;
2291 info
->custom_array_size
= __le32_to_cpu(dev
->size_high
);
2292 info
->custom_array_size
<<= 32;
2293 info
->custom_array_size
|= __le32_to_cpu(dev
->size_low
);
2294 info
->recovery_blocked
= imsm_reshape_blocks_arrays_changes(st
->sb
);
2296 if (prev_map
&& map
->map_state
== prev_map
->map_state
&&
2297 (migr_type(dev
) == MIGR_GEN_MIGR
)) {
2298 info
->reshape_active
= 1;
2299 info
->new_level
= get_imsm_raid_level(map
);
2300 info
->new_layout
= imsm_level_to_layout(info
->new_level
);
2301 info
->new_chunk
= __le16_to_cpu(map
->blocks_per_strip
) << 9;
2302 info
->delta_disks
= map
->num_members
- prev_map
->num_members
;
2303 if (info
->delta_disks
) {
2304 /* this needs to be applied to every array
2307 info
->reshape_active
= CONTAINER_RESHAPE
;
2309 /* We shape information that we give to md might have to be
2310 * modify to cope with md's requirement for reshaping arrays.
2311 * For example, when reshaping a RAID0, md requires it to be
2312 * presented as a degraded RAID4.
2313 * Also if a RAID0 is migrating to a RAID5 we need to specify
2314 * the array as already being RAID5, but the 'before' layout
2315 * is a RAID4-like layout.
2317 switch (info
->array
.level
) {
2319 switch(info
->new_level
) {
2321 /* conversion is happening as RAID4 */
2322 info
->array
.level
= 4;
2323 info
->array
.raid_disks
+= 1;
2326 /* conversion is happening as RAID5 */
2327 info
->array
.level
= 5;
2328 info
->array
.layout
= ALGORITHM_PARITY_N
;
2329 info
->delta_disks
-= 1;
2332 /* FIXME error message */
2333 info
->array
.level
= UnSet
;
2339 info
->new_level
= UnSet
;
2340 info
->new_layout
= UnSet
;
2341 info
->new_chunk
= info
->array
.chunk_size
;
2342 info
->delta_disks
= 0;
2346 info
->disk
.major
= dl
->major
;
2347 info
->disk
.minor
= dl
->minor
;
2348 info
->disk
.number
= dl
->index
;
2349 info
->disk
.raid_disk
= get_imsm_disk_slot(map_to_analyse
,
2353 info
->data_offset
= __le32_to_cpu(map_to_analyse
->pba_of_lba0
);
2354 info
->component_size
=
2355 __le32_to_cpu(map_to_analyse
->blocks_per_member
);
2357 /* check component size aligment
2359 component_size_alligment
=
2360 info
->component_size
% (info
->array
.chunk_size
/512);
2362 if (component_size_alligment
&&
2363 (info
->array
.level
!= 1) && (info
->array
.level
!= UnSet
)) {
2364 dprintf("imsm: reported component size alligned from %llu ",
2365 info
->component_size
);
2366 info
->component_size
-= component_size_alligment
;
2367 dprintf("to %llu (%i).\n",
2368 info
->component_size
, component_size_alligment
);
2371 memset(info
->uuid
, 0, sizeof(info
->uuid
));
2372 info
->recovery_start
= MaxSector
;
2374 info
->reshape_progress
= 0;
2375 info
->resync_start
= MaxSector
;
2376 if ((map_to_analyse
->map_state
== IMSM_T_STATE_UNINITIALIZED
||
2378 imsm_reshape_blocks_arrays_changes(super
) == 0) {
2379 info
->resync_start
= 0;
2381 if (dev
->vol
.migr_state
) {
2382 switch (migr_type(dev
)) {
2385 __u64 blocks_per_unit
= blocks_per_migr_unit(super
,
2387 __u64 units
= __le32_to_cpu(dev
->vol
.curr_migr_unit
);
2389 info
->resync_start
= blocks_per_unit
* units
;
2392 case MIGR_GEN_MIGR
: {
2393 __u64 blocks_per_unit
= blocks_per_migr_unit(super
,
2395 __u64 units
= __le32_to_cpu(migr_rec
->curr_migr_unit
);
2396 unsigned long long array_blocks
;
2399 if (__le32_to_cpu(migr_rec
->ascending_migr
) &&
2401 (__le32_to_cpu(migr_rec
->num_migr_units
)-1)) &&
2402 (super
->migr_rec
->rec_status
==
2403 __cpu_to_le32(UNIT_SRC_IN_CP_AREA
)))
2406 info
->reshape_progress
= blocks_per_unit
* units
;
2408 dprintf("IMSM: General Migration checkpoint : %llu "
2409 "(%llu) -> read reshape progress : %llu\n",
2410 (unsigned long long)units
,
2411 (unsigned long long)blocks_per_unit
,
2412 info
->reshape_progress
);
2414 used_disks
= imsm_num_data_members(dev
, 1);
2415 if (used_disks
> 0) {
2416 array_blocks
= map
->blocks_per_member
*
2418 /* round array size down to closest MB
2420 info
->custom_array_size
= (array_blocks
2421 >> SECT_PER_MB_SHIFT
)
2422 << SECT_PER_MB_SHIFT
;
2426 /* we could emulate the checkpointing of
2427 * 'sync_action=check' migrations, but for now
2428 * we just immediately complete them
2431 /* this is handled by container_content_imsm() */
2432 case MIGR_STATE_CHANGE
:
2433 /* FIXME handle other migrations */
2435 /* we are not dirty, so... */
2436 info
->resync_start
= MaxSector
;
2440 strncpy(info
->name
, (char *) dev
->volume
, MAX_RAID_SERIAL_LEN
);
2441 info
->name
[MAX_RAID_SERIAL_LEN
] = 0;
2443 info
->array
.major_version
= -1;
2444 info
->array
.minor_version
= -2;
2445 devname
= devnum2devname(st
->container_dev
);
2446 *info
->text_version
= '\0';
2448 sprintf(info
->text_version
, "/%s/%d", devname
, info
->container_member
);
2450 info
->safe_mode_delay
= 4000; /* 4 secs like the Matrix driver */
2451 uuid_from_super_imsm(st
, info
->uuid
);
2455 for (i
=0; i
<map_disks
; i
++) {
2457 if (i
< info
->array
.raid_disks
) {
2458 struct imsm_disk
*dsk
;
2459 j
= get_imsm_disk_idx(dev
, i
, -1);
2460 dsk
= get_imsm_disk(super
, j
);
2461 if (dsk
&& (dsk
->status
& CONFIGURED_DISK
))
2468 static __u8
imsm_check_degraded(struct intel_super
*super
, struct imsm_dev
*dev
, int failed
);
2469 static int imsm_count_failed(struct intel_super
*super
, struct imsm_dev
*dev
);
2471 static struct imsm_disk
*get_imsm_missing(struct intel_super
*super
, __u8 index
)
2475 for (d
= super
->missing
; d
; d
= d
->next
)
2476 if (d
->index
== index
)
2481 static void getinfo_super_imsm(struct supertype
*st
, struct mdinfo
*info
, char *map
)
2483 struct intel_super
*super
= st
->sb
;
2484 struct imsm_disk
*disk
;
2485 int map_disks
= info
->array
.raid_disks
;
2486 int max_enough
= -1;
2488 struct imsm_super
*mpb
;
2490 if (super
->current_vol
>= 0) {
2491 getinfo_super_imsm_volume(st
, info
, map
);
2494 memset(info
, 0, sizeof(*info
));
2496 /* Set raid_disks to zero so that Assemble will always pull in valid
2499 info
->array
.raid_disks
= 0;
2500 info
->array
.level
= LEVEL_CONTAINER
;
2501 info
->array
.layout
= 0;
2502 info
->array
.md_minor
= -1;
2503 info
->array
.ctime
= 0; /* N/A for imsm */
2504 info
->array
.utime
= 0;
2505 info
->array
.chunk_size
= 0;
2507 info
->disk
.major
= 0;
2508 info
->disk
.minor
= 0;
2509 info
->disk
.raid_disk
= -1;
2510 info
->reshape_active
= 0;
2511 info
->array
.major_version
= -1;
2512 info
->array
.minor_version
= -2;
2513 strcpy(info
->text_version
, "imsm");
2514 info
->safe_mode_delay
= 0;
2515 info
->disk
.number
= -1;
2516 info
->disk
.state
= 0;
2518 info
->recovery_start
= MaxSector
;
2519 info
->recovery_blocked
= imsm_reshape_blocks_arrays_changes(st
->sb
);
2521 /* do we have the all the insync disks that we expect? */
2522 mpb
= super
->anchor
;
2524 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
2525 struct imsm_dev
*dev
= get_imsm_dev(super
, i
);
2526 int failed
, enough
, j
, missing
= 0;
2527 struct imsm_map
*map
;
2530 failed
= imsm_count_failed(super
, dev
);
2531 state
= imsm_check_degraded(super
, dev
, failed
);
2532 map
= get_imsm_map(dev
, 0);
2534 /* any newly missing disks?
2535 * (catches single-degraded vs double-degraded)
2537 for (j
= 0; j
< map
->num_members
; j
++) {
2538 __u32 ord
= get_imsm_ord_tbl_ent(dev
, i
, 0);
2539 __u32 idx
= ord_to_idx(ord
);
2541 if (!(ord
& IMSM_ORD_REBUILD
) &&
2542 get_imsm_missing(super
, idx
)) {
2548 if (state
== IMSM_T_STATE_FAILED
)
2550 else if (state
== IMSM_T_STATE_DEGRADED
&&
2551 (state
!= map
->map_state
|| missing
))
2553 else /* we're normal, or already degraded */
2556 /* in the missing/failed disk case check to see
2557 * if at least one array is runnable
2559 max_enough
= max(max_enough
, enough
);
2561 dprintf("%s: enough: %d\n", __func__
, max_enough
);
2562 info
->container_enough
= max_enough
;
2565 __u32 reserved
= imsm_reserved_sectors(super
, super
->disks
);
2567 disk
= &super
->disks
->disk
;
2568 info
->data_offset
= __le32_to_cpu(disk
->total_blocks
) - reserved
;
2569 info
->component_size
= reserved
;
2570 info
->disk
.state
= is_configured(disk
) ? (1 << MD_DISK_ACTIVE
) : 0;
2571 /* we don't change info->disk.raid_disk here because
2572 * this state will be finalized in mdmon after we have
2573 * found the 'most fresh' version of the metadata
2575 info
->disk
.state
|= is_failed(disk
) ? (1 << MD_DISK_FAULTY
) : 0;
2576 info
->disk
.state
|= is_spare(disk
) ? 0 : (1 << MD_DISK_SYNC
);
2579 /* only call uuid_from_super_imsm when this disk is part of a populated container,
2580 * ->compare_super may have updated the 'num_raid_devs' field for spares
2582 if (info
->disk
.state
& (1 << MD_DISK_SYNC
) || super
->anchor
->num_raid_devs
)
2583 uuid_from_super_imsm(st
, info
->uuid
);
2585 memcpy(info
->uuid
, uuid_zero
, sizeof(uuid_zero
));
2587 /* I don't know how to compute 'map' on imsm, so use safe default */
2590 for (i
= 0; i
< map_disks
; i
++)
2596 /* allocates memory and fills disk in mdinfo structure
2597 * for each disk in array */
2598 struct mdinfo
*getinfo_super_disks_imsm(struct supertype
*st
)
2600 struct mdinfo
*mddev
= NULL
;
2601 struct intel_super
*super
= st
->sb
;
2602 struct imsm_disk
*disk
;
2605 if (!super
|| !super
->disks
)
2608 mddev
= malloc(sizeof(*mddev
));
2610 fprintf(stderr
, Name
": Failed to allocate memory.\n");
2613 memset(mddev
, 0, sizeof(*mddev
));
2617 tmp
= malloc(sizeof(*tmp
));
2619 fprintf(stderr
, Name
": Failed to allocate memory.\n");
2624 memset(tmp
, 0, sizeof(*tmp
));
2626 tmp
->next
= mddev
->devs
;
2628 tmp
->disk
.number
= count
++;
2629 tmp
->disk
.major
= dl
->major
;
2630 tmp
->disk
.minor
= dl
->minor
;
2631 tmp
->disk
.state
= is_configured(disk
) ?
2632 (1 << MD_DISK_ACTIVE
) : 0;
2633 tmp
->disk
.state
|= is_failed(disk
) ? (1 << MD_DISK_FAULTY
) : 0;
2634 tmp
->disk
.state
|= is_spare(disk
) ? 0 : (1 << MD_DISK_SYNC
);
2635 tmp
->disk
.raid_disk
= -1;
2641 static int update_super_imsm(struct supertype
*st
, struct mdinfo
*info
,
2642 char *update
, char *devname
, int verbose
,
2643 int uuid_set
, char *homehost
)
2645 /* For 'assemble' and 'force' we need to return non-zero if any
2646 * change was made. For others, the return value is ignored.
2647 * Update options are:
2648 * force-one : This device looks a bit old but needs to be included,
2649 * update age info appropriately.
2650 * assemble: clear any 'faulty' flag to allow this device to
2652 * force-array: Array is degraded but being forced, mark it clean
2653 * if that will be needed to assemble it.
2655 * newdev: not used ????
2656 * grow: Array has gained a new device - this is currently for
2658 * resync: mark as dirty so a resync will happen.
2659 * name: update the name - preserving the homehost
2660 * uuid: Change the uuid of the array to match watch is given
2662 * Following are not relevant for this imsm:
2663 * sparc2.2 : update from old dodgey metadata
2664 * super-minor: change the preferred_minor number
2665 * summaries: update redundant counters.
2666 * homehost: update the recorded homehost
2667 * _reshape_progress: record new reshape_progress position.
2670 struct intel_super
*super
= st
->sb
;
2671 struct imsm_super
*mpb
;
2673 /* we can only update container info */
2674 if (!super
|| super
->current_vol
>= 0 || !super
->anchor
)
2677 mpb
= super
->anchor
;
2679 if (strcmp(update
, "uuid") == 0 && uuid_set
&& !info
->update_private
)
2681 else if (strcmp(update
, "uuid") == 0 && uuid_set
&& info
->update_private
) {
2682 mpb
->orig_family_num
= *((__u32
*) info
->update_private
);
2684 } else if (strcmp(update
, "uuid") == 0) {
2685 __u32
*new_family
= malloc(sizeof(*new_family
));
2687 /* update orig_family_number with the incoming random
2688 * data, report the new effective uuid, and store the
2689 * new orig_family_num for future updates.
2692 memcpy(&mpb
->orig_family_num
, info
->uuid
, sizeof(__u32
));
2693 uuid_from_super_imsm(st
, info
->uuid
);
2694 *new_family
= mpb
->orig_family_num
;
2695 info
->update_private
= new_family
;
2698 } else if (strcmp(update
, "assemble") == 0)
2703 /* successful update? recompute checksum */
2705 mpb
->check_sum
= __le32_to_cpu(__gen_imsm_checksum(mpb
));
2710 static size_t disks_to_mpb_size(int disks
)
2714 size
= sizeof(struct imsm_super
);
2715 size
+= (disks
- 1) * sizeof(struct imsm_disk
);
2716 size
+= 2 * sizeof(struct imsm_dev
);
2717 /* up to 2 maps per raid device (-2 for imsm_maps in imsm_dev */
2718 size
+= (4 - 2) * sizeof(struct imsm_map
);
2719 /* 4 possible disk_ord_tbl's */
2720 size
+= 4 * (disks
- 1) * sizeof(__u32
);
2725 static __u64
avail_size_imsm(struct supertype
*st
, __u64 devsize
)
2727 if (devsize
< (MPB_SECTOR_CNT
+ IMSM_RESERVED_SECTORS
))
2730 return devsize
- (MPB_SECTOR_CNT
+ IMSM_RESERVED_SECTORS
);
2733 static void free_devlist(struct intel_super
*super
)
2735 struct intel_dev
*dv
;
2737 while (super
->devlist
) {
2738 dv
= super
->devlist
->next
;
2739 free(super
->devlist
->dev
);
2740 free(super
->devlist
);
2741 super
->devlist
= dv
;
2745 static void imsm_copy_dev(struct imsm_dev
*dest
, struct imsm_dev
*src
)
2747 memcpy(dest
, src
, sizeof_imsm_dev(src
, 0));
2750 static int compare_super_imsm(struct supertype
*st
, struct supertype
*tst
)
2754 * 0 same, or first was empty, and second was copied
2755 * 1 second had wrong number
2757 * 3 wrong other info
2759 struct intel_super
*first
= st
->sb
;
2760 struct intel_super
*sec
= tst
->sb
;
2767 /* in platform dependent environment test if the disks
2768 * use the same Intel hba
2770 if (!check_env("IMSM_NO_PLATFORM")) {
2771 if (!first
->hba
|| !sec
->hba
||
2772 (first
->hba
->type
!= sec
->hba
->type
)) {
2774 "HBAs of devices does not match %s != %s\n",
2775 first
->hba
? get_sys_dev_type(first
->hba
->type
) : NULL
,
2776 sec
->hba
? get_sys_dev_type(sec
->hba
->type
) : NULL
);
2781 /* if an anchor does not have num_raid_devs set then it is a free
2784 if (first
->anchor
->num_raid_devs
> 0 &&
2785 sec
->anchor
->num_raid_devs
> 0) {
2786 /* Determine if these disks might ever have been
2787 * related. Further disambiguation can only take place
2788 * in load_super_imsm_all
2790 __u32 first_family
= first
->anchor
->orig_family_num
;
2791 __u32 sec_family
= sec
->anchor
->orig_family_num
;
2793 if (memcmp(first
->anchor
->sig
, sec
->anchor
->sig
,
2794 MAX_SIGNATURE_LENGTH
) != 0)
2797 if (first_family
== 0)
2798 first_family
= first
->anchor
->family_num
;
2799 if (sec_family
== 0)
2800 sec_family
= sec
->anchor
->family_num
;
2802 if (first_family
!= sec_family
)
2808 /* if 'first' is a spare promote it to a populated mpb with sec's
2811 if (first
->anchor
->num_raid_devs
== 0 &&
2812 sec
->anchor
->num_raid_devs
> 0) {
2814 struct intel_dev
*dv
;
2815 struct imsm_dev
*dev
;
2817 /* we need to copy raid device info from sec if an allocation
2818 * fails here we don't associate the spare
2820 for (i
= 0; i
< sec
->anchor
->num_raid_devs
; i
++) {
2821 dv
= malloc(sizeof(*dv
));
2824 dev
= malloc(sizeof_imsm_dev(get_imsm_dev(sec
, i
), 1));
2831 dv
->next
= first
->devlist
;
2832 first
->devlist
= dv
;
2834 if (i
< sec
->anchor
->num_raid_devs
) {
2835 /* allocation failure */
2836 free_devlist(first
);
2837 fprintf(stderr
, "imsm: failed to associate spare\n");
2840 first
->anchor
->num_raid_devs
= sec
->anchor
->num_raid_devs
;
2841 first
->anchor
->orig_family_num
= sec
->anchor
->orig_family_num
;
2842 first
->anchor
->family_num
= sec
->anchor
->family_num
;
2843 memcpy(first
->anchor
->sig
, sec
->anchor
->sig
, MAX_SIGNATURE_LENGTH
);
2844 for (i
= 0; i
< sec
->anchor
->num_raid_devs
; i
++)
2845 imsm_copy_dev(get_imsm_dev(first
, i
), get_imsm_dev(sec
, i
));
2851 static void fd2devname(int fd
, char *name
)
2855 char dname
[PATH_MAX
];
2860 if (fstat(fd
, &st
) != 0)
2862 sprintf(path
, "/sys/dev/block/%d:%d",
2863 major(st
.st_rdev
), minor(st
.st_rdev
));
2865 rv
= readlink(path
, dname
, sizeof(dname
)-1);
2870 nm
= strrchr(dname
, '/');
2873 snprintf(name
, MAX_RAID_SERIAL_LEN
, "/dev/%s", nm
);
2877 extern int scsi_get_serial(int fd
, void *buf
, size_t buf_len
);
2879 static int imsm_read_serial(int fd
, char *devname
,
2880 __u8 serial
[MAX_RAID_SERIAL_LEN
])
2882 unsigned char scsi_serial
[255];
2891 memset(scsi_serial
, 0, sizeof(scsi_serial
));
2893 rv
= scsi_get_serial(fd
, scsi_serial
, sizeof(scsi_serial
));
2895 if (rv
&& check_env("IMSM_DEVNAME_AS_SERIAL")) {
2896 memset(serial
, 0, MAX_RAID_SERIAL_LEN
);
2897 fd2devname(fd
, (char *) serial
);
2904 Name
": Failed to retrieve serial for %s\n",
2909 rsp_len
= scsi_serial
[3];
2913 Name
": Failed to retrieve serial for %s\n",
2917 rsp_buf
= (char *) &scsi_serial
[4];
2919 /* trim all whitespace and non-printable characters and convert
2922 for (i
= 0, dest
= rsp_buf
; i
< rsp_len
; i
++) {
2925 /* ':' is reserved for use in placeholder serial
2926 * numbers for missing disks
2934 len
= dest
- rsp_buf
;
2937 /* truncate leading characters */
2938 if (len
> MAX_RAID_SERIAL_LEN
) {
2939 dest
+= len
- MAX_RAID_SERIAL_LEN
;
2940 len
= MAX_RAID_SERIAL_LEN
;
2943 memset(serial
, 0, MAX_RAID_SERIAL_LEN
);
2944 memcpy(serial
, dest
, len
);
2949 static int serialcmp(__u8
*s1
, __u8
*s2
)
2951 return strncmp((char *) s1
, (char *) s2
, MAX_RAID_SERIAL_LEN
);
2954 static void serialcpy(__u8
*dest
, __u8
*src
)
2956 strncpy((char *) dest
, (char *) src
, MAX_RAID_SERIAL_LEN
);
2959 static struct dl
*serial_to_dl(__u8
*serial
, struct intel_super
*super
)
2963 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
2964 if (serialcmp(dl
->serial
, serial
) == 0)
2970 static struct imsm_disk
*
2971 __serial_to_disk(__u8
*serial
, struct imsm_super
*mpb
, int *idx
)
2975 for (i
= 0; i
< mpb
->num_disks
; i
++) {
2976 struct imsm_disk
*disk
= __get_imsm_disk(mpb
, i
);
2978 if (serialcmp(disk
->serial
, serial
) == 0) {
2989 load_imsm_disk(int fd
, struct intel_super
*super
, char *devname
, int keep_fd
)
2991 struct imsm_disk
*disk
;
2996 __u8 serial
[MAX_RAID_SERIAL_LEN
];
2998 rv
= imsm_read_serial(fd
, devname
, serial
);
3003 dl
= calloc(1, sizeof(*dl
));
3007 Name
": failed to allocate disk buffer for %s\n",
3013 dl
->major
= major(stb
.st_rdev
);
3014 dl
->minor
= minor(stb
.st_rdev
);
3015 dl
->next
= super
->disks
;
3016 dl
->fd
= keep_fd
? fd
: -1;
3017 assert(super
->disks
== NULL
);
3019 serialcpy(dl
->serial
, serial
);
3022 fd2devname(fd
, name
);
3024 dl
->devname
= strdup(devname
);
3026 dl
->devname
= strdup(name
);
3028 /* look up this disk's index in the current anchor */
3029 disk
= __serial_to_disk(dl
->serial
, super
->anchor
, &dl
->index
);
3032 /* only set index on disks that are a member of a
3033 * populated contianer, i.e. one with raid_devs
3035 if (is_failed(&dl
->disk
))
3037 else if (is_spare(&dl
->disk
))
3045 /* When migrating map0 contains the 'destination' state while map1
3046 * contains the current state. When not migrating map0 contains the
3047 * current state. This routine assumes that map[0].map_state is set to
3048 * the current array state before being called.
3050 * Migration is indicated by one of the following states
3051 * 1/ Idle (migr_state=0 map0state=normal||unitialized||degraded||failed)
3052 * 2/ Initialize (migr_state=1 migr_type=MIGR_INIT map0state=normal
3053 * map1state=unitialized)
3054 * 3/ Repair (Resync) (migr_state=1 migr_type=MIGR_REPAIR map0state=normal
3056 * 4/ Rebuild (migr_state=1 migr_type=MIGR_REBUILD map0state=normal
3057 * map1state=degraded)
3058 * 5/ Migration (mig_state=1 migr_type=MIGR_GEN_MIGR map0state=normal
3061 static void migrate(struct imsm_dev
*dev
, struct intel_super
*super
,
3062 __u8 to_state
, int migr_type
)
3064 struct imsm_map
*dest
;
3065 struct imsm_map
*src
= get_imsm_map(dev
, 0);
3067 dev
->vol
.migr_state
= 1;
3068 set_migr_type(dev
, migr_type
);
3069 dev
->vol
.curr_migr_unit
= 0;
3070 dest
= get_imsm_map(dev
, 1);
3072 /* duplicate and then set the target end state in map[0] */
3073 memcpy(dest
, src
, sizeof_imsm_map(src
));
3074 if ((migr_type
== MIGR_REBUILD
) ||
3075 (migr_type
== MIGR_GEN_MIGR
)) {
3079 for (i
= 0; i
< src
->num_members
; i
++) {
3080 ord
= __le32_to_cpu(src
->disk_ord_tbl
[i
]);
3081 set_imsm_ord_tbl_ent(src
, i
, ord_to_idx(ord
));
3085 if (migr_type
== MIGR_GEN_MIGR
)
3086 /* Clear migration record */
3087 memset(super
->migr_rec
, 0, sizeof(struct migr_record
));
3089 src
->map_state
= to_state
;
3092 static void end_migration(struct imsm_dev
*dev
, __u8 map_state
)
3094 struct imsm_map
*map
= get_imsm_map(dev
, 0);
3095 struct imsm_map
*prev
= get_imsm_map(dev
, dev
->vol
.migr_state
);
3098 /* merge any IMSM_ORD_REBUILD bits that were not successfully
3099 * completed in the last migration.
3101 * FIXME add support for raid-level-migration
3103 for (i
= 0; i
< prev
->num_members
; i
++)
3104 for (j
= 0; j
< map
->num_members
; j
++)
3105 /* during online capacity expansion
3106 * disks position can be changed if takeover is used
3108 if (ord_to_idx(map
->disk_ord_tbl
[j
]) ==
3109 ord_to_idx(prev
->disk_ord_tbl
[i
])) {
3110 map
->disk_ord_tbl
[j
] |= prev
->disk_ord_tbl
[i
];
3114 dev
->vol
.migr_state
= 0;
3115 set_migr_type(dev
, 0);
3116 dev
->vol
.curr_migr_unit
= 0;
3117 map
->map_state
= map_state
;
3121 static int parse_raid_devices(struct intel_super
*super
)
3124 struct imsm_dev
*dev_new
;
3125 size_t len
, len_migr
;
3127 size_t space_needed
= 0;
3128 struct imsm_super
*mpb
= super
->anchor
;
3130 for (i
= 0; i
< super
->anchor
->num_raid_devs
; i
++) {
3131 struct imsm_dev
*dev_iter
= __get_imsm_dev(super
->anchor
, i
);
3132 struct intel_dev
*dv
;
3134 len
= sizeof_imsm_dev(dev_iter
, 0);
3135 len_migr
= sizeof_imsm_dev(dev_iter
, 1);
3137 space_needed
+= len_migr
- len
;
3139 dv
= malloc(sizeof(*dv
));
3142 if (max_len
< len_migr
)
3144 if (max_len
> len_migr
)
3145 space_needed
+= max_len
- len_migr
;
3146 dev_new
= malloc(max_len
);
3151 imsm_copy_dev(dev_new
, dev_iter
);
3154 dv
->next
= super
->devlist
;
3155 super
->devlist
= dv
;
3158 /* ensure that super->buf is large enough when all raid devices
3161 if (__le32_to_cpu(mpb
->mpb_size
) + space_needed
> super
->len
) {
3164 len
= ROUND_UP(__le32_to_cpu(mpb
->mpb_size
) + space_needed
, 512);
3165 if (posix_memalign(&buf
, 512, len
) != 0)
3168 memcpy(buf
, super
->buf
, super
->len
);
3169 memset(buf
+ super
->len
, 0, len
- super
->len
);
3178 /* retrieve a pointer to the bbm log which starts after all raid devices */
3179 struct bbm_log
*__get_imsm_bbm_log(struct imsm_super
*mpb
)
3183 if (__le32_to_cpu(mpb
->bbm_log_size
)) {
3185 ptr
+= mpb
->mpb_size
- __le32_to_cpu(mpb
->bbm_log_size
);
3191 /*******************************************************************************
3192 * Function: check_mpb_migr_compatibility
3193 * Description: Function checks for unsupported migration features:
3194 * - migration optimization area (pba_of_lba0)
3195 * - descending reshape (ascending_migr)
3197 * super : imsm metadata information
3199 * 0 : migration is compatible
3200 * -1 : migration is not compatible
3201 ******************************************************************************/
3202 int check_mpb_migr_compatibility(struct intel_super
*super
)
3204 struct imsm_map
*map0
, *map1
;
3205 struct migr_record
*migr_rec
= super
->migr_rec
;
3208 for (i
= 0; i
< super
->anchor
->num_raid_devs
; i
++) {
3209 struct imsm_dev
*dev_iter
= __get_imsm_dev(super
->anchor
, i
);
3212 dev_iter
->vol
.migr_state
== 1 &&
3213 dev_iter
->vol
.migr_type
== MIGR_GEN_MIGR
) {
3214 /* This device is migrating */
3215 map0
= get_imsm_map(dev_iter
, 0);
3216 map1
= get_imsm_map(dev_iter
, 1);
3217 if (map0
->pba_of_lba0
!= map1
->pba_of_lba0
)
3218 /* migration optimization area was used */
3220 if (migr_rec
->ascending_migr
== 0
3221 && migr_rec
->dest_depth_per_unit
> 0)
3222 /* descending reshape not supported yet */
3229 static void __free_imsm(struct intel_super
*super
, int free_disks
);
3231 /* load_imsm_mpb - read matrix metadata
3232 * allocates super->mpb to be freed by free_imsm
3234 static int load_imsm_mpb(int fd
, struct intel_super
*super
, char *devname
)
3236 unsigned long long dsize
;
3237 unsigned long long sectors
;
3239 struct imsm_super
*anchor
;
3242 get_dev_size(fd
, NULL
, &dsize
);
3246 Name
": %s: device to small for imsm\n",
3251 if (lseek64(fd
, dsize
- (512 * 2), SEEK_SET
) < 0) {
3253 fprintf(stderr
, Name
3254 ": Cannot seek to anchor block on %s: %s\n",
3255 devname
, strerror(errno
));
3259 if (posix_memalign((void**)&anchor
, 512, 512) != 0) {
3262 Name
": Failed to allocate imsm anchor buffer"
3263 " on %s\n", devname
);
3266 if (read(fd
, anchor
, 512) != 512) {
3269 Name
": Cannot read anchor block on %s: %s\n",
3270 devname
, strerror(errno
));
3275 if (strncmp((char *) anchor
->sig
, MPB_SIGNATURE
, MPB_SIG_LEN
) != 0) {
3278 Name
": no IMSM anchor on %s\n", devname
);
3283 __free_imsm(super
, 0);
3284 /* reload capability and hba */
3286 /* capability and hba must be updated with new super allocation */
3287 find_intel_hba_capability(fd
, super
, devname
);
3288 super
->len
= ROUND_UP(anchor
->mpb_size
, 512);
3289 if (posix_memalign(&super
->buf
, 512, super
->len
) != 0) {
3292 Name
": unable to allocate %zu byte mpb buffer\n",
3297 memcpy(super
->buf
, anchor
, 512);
3299 sectors
= mpb_sectors(anchor
) - 1;
3302 if (posix_memalign(&super
->migr_rec_buf
, 512, 512) != 0) {
3303 fprintf(stderr
, Name
3304 ": %s could not allocate migr_rec buffer\n", __func__
);
3310 check_sum
= __gen_imsm_checksum(super
->anchor
);
3311 if (check_sum
!= __le32_to_cpu(super
->anchor
->check_sum
)) {
3314 Name
": IMSM checksum %x != %x on %s\n",
3316 __le32_to_cpu(super
->anchor
->check_sum
),
3324 /* read the extended mpb */
3325 if (lseek64(fd
, dsize
- (512 * (2 + sectors
)), SEEK_SET
) < 0) {
3328 Name
": Cannot seek to extended mpb on %s: %s\n",
3329 devname
, strerror(errno
));
3333 if ((unsigned)read(fd
, super
->buf
+ 512, super
->len
- 512) != super
->len
- 512) {
3336 Name
": Cannot read extended mpb on %s: %s\n",
3337 devname
, strerror(errno
));
3341 check_sum
= __gen_imsm_checksum(super
->anchor
);
3342 if (check_sum
!= __le32_to_cpu(super
->anchor
->check_sum
)) {
3345 Name
": IMSM checksum %x != %x on %s\n",
3346 check_sum
, __le32_to_cpu(super
->anchor
->check_sum
),
3351 /* FIXME the BBM log is disk specific so we cannot use this global
3352 * buffer for all disks. Ok for now since we only look at the global
3353 * bbm_log_size parameter to gate assembly
3355 super
->bbm_log
= __get_imsm_bbm_log(super
->anchor
);
3360 static int read_imsm_migr_rec(int fd
, struct intel_super
*super
);
3363 load_and_parse_mpb(int fd
, struct intel_super
*super
, char *devname
, int keep_fd
)
3367 err
= load_imsm_mpb(fd
, super
, devname
);
3370 err
= load_imsm_disk(fd
, super
, devname
, keep_fd
);
3373 err
= parse_raid_devices(super
);
3378 static void __free_imsm_disk(struct dl
*d
)
3390 static void free_imsm_disks(struct intel_super
*super
)
3394 while (super
->disks
) {
3396 super
->disks
= d
->next
;
3397 __free_imsm_disk(d
);
3399 while (super
->disk_mgmt_list
) {
3400 d
= super
->disk_mgmt_list
;
3401 super
->disk_mgmt_list
= d
->next
;
3402 __free_imsm_disk(d
);
3404 while (super
->missing
) {
3406 super
->missing
= d
->next
;
3407 __free_imsm_disk(d
);
3412 /* free all the pieces hanging off of a super pointer */
3413 static void __free_imsm(struct intel_super
*super
, int free_disks
)
3415 struct intel_hba
*elem
, *next
;
3421 /* unlink capability description */
3423 if (super
->migr_rec_buf
) {
3424 free(super
->migr_rec_buf
);
3425 super
->migr_rec_buf
= NULL
;
3428 free_imsm_disks(super
);
3429 free_devlist(super
);
3433 free((void *)elem
->path
);
3441 static void free_imsm(struct intel_super
*super
)
3443 __free_imsm(super
, 1);
3447 static void free_super_imsm(struct supertype
*st
)
3449 struct intel_super
*super
= st
->sb
;
3458 static struct intel_super
*alloc_super(void)
3460 struct intel_super
*super
= malloc(sizeof(*super
));
3463 memset(super
, 0, sizeof(*super
));
3464 super
->current_vol
= -1;
3465 super
->create_offset
= ~((__u32
) 0);
3471 * find and allocate hba and OROM/EFI based on valid fd of RAID component device
3473 static int find_intel_hba_capability(int fd
, struct intel_super
*super
, char *devname
)
3475 struct sys_dev
*hba_name
;
3478 if ((fd
< 0) || check_env("IMSM_NO_PLATFORM")) {
3483 hba_name
= find_disk_attached_hba(fd
, NULL
);
3487 Name
": %s is not attached to Intel(R) RAID controller.\n",
3491 rv
= attach_hba_to_super(super
, hba_name
);
3494 struct intel_hba
*hba
= super
->hba
;
3496 fprintf(stderr
, Name
": %s is attached to Intel(R) %s RAID "
3497 "controller (%s),\n"
3498 " but the container is assigned to Intel(R) "
3499 "%s RAID controller (",
3502 hba_name
->pci_id
? : "Err!",
3503 get_sys_dev_type(hba_name
->type
));
3506 fprintf(stderr
, "%s", hba
->pci_id
? : "Err!");
3508 fprintf(stderr
, ", ");
3512 fprintf(stderr
, ").\n"
3513 " Mixing devices attached to different controllers "
3514 "is not allowed.\n");
3516 free_sys_dev(&hba_name
);
3519 super
->orom
= find_imsm_capability(hba_name
->type
);
3520 free_sys_dev(&hba_name
);
3526 /* find_missing - helper routine for load_super_imsm_all that identifies
3527 * disks that have disappeared from the system. This routine relies on
3528 * the mpb being uptodate, which it is at load time.
3530 static int find_missing(struct intel_super
*super
)
3533 struct imsm_super
*mpb
= super
->anchor
;
3535 struct imsm_disk
*disk
;
3537 for (i
= 0; i
< mpb
->num_disks
; i
++) {
3538 disk
= __get_imsm_disk(mpb
, i
);
3539 dl
= serial_to_dl(disk
->serial
, super
);
3543 dl
= malloc(sizeof(*dl
));
3549 dl
->devname
= strdup("missing");
3551 serialcpy(dl
->serial
, disk
->serial
);
3554 dl
->next
= super
->missing
;
3555 super
->missing
= dl
;
3562 static struct intel_disk
*disk_list_get(__u8
*serial
, struct intel_disk
*disk_list
)
3564 struct intel_disk
*idisk
= disk_list
;
3567 if (serialcmp(idisk
->disk
.serial
, serial
) == 0)
3569 idisk
= idisk
->next
;
3575 static int __prep_thunderdome(struct intel_super
**table
, int tbl_size
,
3576 struct intel_super
*super
,
3577 struct intel_disk
**disk_list
)
3579 struct imsm_disk
*d
= &super
->disks
->disk
;
3580 struct imsm_super
*mpb
= super
->anchor
;
3583 for (i
= 0; i
< tbl_size
; i
++) {
3584 struct imsm_super
*tbl_mpb
= table
[i
]->anchor
;
3585 struct imsm_disk
*tbl_d
= &table
[i
]->disks
->disk
;
3587 if (tbl_mpb
->family_num
== mpb
->family_num
) {
3588 if (tbl_mpb
->check_sum
== mpb
->check_sum
) {
3589 dprintf("%s: mpb from %d:%d matches %d:%d\n",
3590 __func__
, super
->disks
->major
,
3591 super
->disks
->minor
,
3592 table
[i
]->disks
->major
,
3593 table
[i
]->disks
->minor
);
3597 if (((is_configured(d
) && !is_configured(tbl_d
)) ||
3598 is_configured(d
) == is_configured(tbl_d
)) &&
3599 tbl_mpb
->generation_num
< mpb
->generation_num
) {
3600 /* current version of the mpb is a
3601 * better candidate than the one in
3602 * super_table, but copy over "cross
3603 * generational" status
3605 struct intel_disk
*idisk
;
3607 dprintf("%s: mpb from %d:%d replaces %d:%d\n",
3608 __func__
, super
->disks
->major
,
3609 super
->disks
->minor
,
3610 table
[i
]->disks
->major
,
3611 table
[i
]->disks
->minor
);
3613 idisk
= disk_list_get(tbl_d
->serial
, *disk_list
);
3614 if (idisk
&& is_failed(&idisk
->disk
))
3615 tbl_d
->status
|= FAILED_DISK
;
3618 struct intel_disk
*idisk
;
3619 struct imsm_disk
*disk
;
3621 /* tbl_mpb is more up to date, but copy
3622 * over cross generational status before
3625 disk
= __serial_to_disk(d
->serial
, mpb
, NULL
);
3626 if (disk
&& is_failed(disk
))
3627 d
->status
|= FAILED_DISK
;
3629 idisk
= disk_list_get(d
->serial
, *disk_list
);
3632 if (disk
&& is_configured(disk
))
3633 idisk
->disk
.status
|= CONFIGURED_DISK
;
3636 dprintf("%s: mpb from %d:%d prefer %d:%d\n",
3637 __func__
, super
->disks
->major
,
3638 super
->disks
->minor
,
3639 table
[i
]->disks
->major
,
3640 table
[i
]->disks
->minor
);
3648 table
[tbl_size
++] = super
;
3652 /* update/extend the merged list of imsm_disk records */
3653 for (j
= 0; j
< mpb
->num_disks
; j
++) {
3654 struct imsm_disk
*disk
= __get_imsm_disk(mpb
, j
);
3655 struct intel_disk
*idisk
;
3657 idisk
= disk_list_get(disk
->serial
, *disk_list
);
3659 idisk
->disk
.status
|= disk
->status
;
3660 if (is_configured(&idisk
->disk
) ||
3661 is_failed(&idisk
->disk
))
3662 idisk
->disk
.status
&= ~(SPARE_DISK
);
3664 idisk
= calloc(1, sizeof(*idisk
));
3667 idisk
->owner
= IMSM_UNKNOWN_OWNER
;
3668 idisk
->disk
= *disk
;
3669 idisk
->next
= *disk_list
;
3673 if (serialcmp(idisk
->disk
.serial
, d
->serial
) == 0)
3680 static struct intel_super
*
3681 validate_members(struct intel_super
*super
, struct intel_disk
*disk_list
,
3684 struct imsm_super
*mpb
= super
->anchor
;
3688 for (i
= 0; i
< mpb
->num_disks
; i
++) {
3689 struct imsm_disk
*disk
= __get_imsm_disk(mpb
, i
);
3690 struct intel_disk
*idisk
;
3692 idisk
= disk_list_get(disk
->serial
, disk_list
);
3694 if (idisk
->owner
== owner
||
3695 idisk
->owner
== IMSM_UNKNOWN_OWNER
)
3698 dprintf("%s: '%.16s' owner %d != %d\n",
3699 __func__
, disk
->serial
, idisk
->owner
,
3702 dprintf("%s: unknown disk %x [%d]: %.16s\n",
3703 __func__
, __le32_to_cpu(mpb
->family_num
), i
,
3709 if (ok_count
== mpb
->num_disks
)
3714 static void show_conflicts(__u32 family_num
, struct intel_super
*super_list
)
3716 struct intel_super
*s
;
3718 for (s
= super_list
; s
; s
= s
->next
) {
3719 if (family_num
!= s
->anchor
->family_num
)
3721 fprintf(stderr
, "Conflict, offlining family %#x on '%s'\n",
3722 __le32_to_cpu(family_num
), s
->disks
->devname
);
3726 static struct intel_super
*
3727 imsm_thunderdome(struct intel_super
**super_list
, int len
)
3729 struct intel_super
*super_table
[len
];
3730 struct intel_disk
*disk_list
= NULL
;
3731 struct intel_super
*champion
, *spare
;
3732 struct intel_super
*s
, **del
;
3737 memset(super_table
, 0, sizeof(super_table
));
3738 for (s
= *super_list
; s
; s
= s
->next
)
3739 tbl_size
= __prep_thunderdome(super_table
, tbl_size
, s
, &disk_list
);
3741 for (i
= 0; i
< tbl_size
; i
++) {
3742 struct imsm_disk
*d
;
3743 struct intel_disk
*idisk
;
3744 struct imsm_super
*mpb
= super_table
[i
]->anchor
;
3747 d
= &s
->disks
->disk
;
3749 /* 'd' must appear in merged disk list for its
3750 * configuration to be valid
3752 idisk
= disk_list_get(d
->serial
, disk_list
);
3753 if (idisk
&& idisk
->owner
== i
)
3754 s
= validate_members(s
, disk_list
, i
);
3759 dprintf("%s: marking family: %#x from %d:%d offline\n",
3760 __func__
, mpb
->family_num
,
3761 super_table
[i
]->disks
->major
,
3762 super_table
[i
]->disks
->minor
);
3766 /* This is where the mdadm implementation differs from the Windows
3767 * driver which has no strict concept of a container. We can only
3768 * assemble one family from a container, so when returning a prodigal
3769 * array member to this system the code will not be able to disambiguate
3770 * the container contents that should be assembled ("foreign" versus
3771 * "local"). It requires user intervention to set the orig_family_num
3772 * to a new value to establish a new container. The Windows driver in
3773 * this situation fixes up the volume name in place and manages the
3774 * foreign array as an independent entity.
3779 for (i
= 0; i
< tbl_size
; i
++) {
3780 struct intel_super
*tbl_ent
= super_table
[i
];
3786 if (tbl_ent
->anchor
->num_raid_devs
== 0) {
3791 if (s
&& !is_spare
) {
3792 show_conflicts(tbl_ent
->anchor
->family_num
, *super_list
);
3794 } else if (!s
&& !is_spare
)
3807 fprintf(stderr
, "Chose family %#x on '%s', "
3808 "assemble conflicts to new container with '--update=uuid'\n",
3809 __le32_to_cpu(s
->anchor
->family_num
), s
->disks
->devname
);
3811 /* collect all dl's onto 'champion', and update them to
3812 * champion's version of the status
3814 for (s
= *super_list
; s
; s
= s
->next
) {
3815 struct imsm_super
*mpb
= champion
->anchor
;
3816 struct dl
*dl
= s
->disks
;
3821 for (i
= 0; i
< mpb
->num_disks
; i
++) {
3822 struct imsm_disk
*disk
;
3824 disk
= __serial_to_disk(dl
->serial
, mpb
, &dl
->index
);
3827 /* only set index on disks that are a member of
3828 * a populated contianer, i.e. one with
3831 if (is_failed(&dl
->disk
))
3833 else if (is_spare(&dl
->disk
))
3839 if (i
>= mpb
->num_disks
) {
3840 struct intel_disk
*idisk
;
3842 idisk
= disk_list_get(dl
->serial
, disk_list
);
3843 if (idisk
&& is_spare(&idisk
->disk
) &&
3844 !is_failed(&idisk
->disk
) && !is_configured(&idisk
->disk
))
3852 dl
->next
= champion
->disks
;
3853 champion
->disks
= dl
;
3857 /* delete 'champion' from super_list */
3858 for (del
= super_list
; *del
; ) {
3859 if (*del
== champion
) {
3860 *del
= (*del
)->next
;
3863 del
= &(*del
)->next
;
3865 champion
->next
= NULL
;
3869 struct intel_disk
*idisk
= disk_list
;
3871 disk_list
= disk_list
->next
;
3878 static int load_super_imsm_all(struct supertype
*st
, int fd
, void **sbp
,
3882 struct intel_super
*super_list
= NULL
;
3883 struct intel_super
*super
= NULL
;
3884 int devnum
= fd2devnum(fd
);
3890 /* check if 'fd' an opened container */
3891 sra
= sysfs_read(fd
, 0, GET_LEVEL
|GET_VERSION
|GET_DEVS
|GET_STATE
);
3895 if (sra
->array
.major_version
!= -1 ||
3896 sra
->array
.minor_version
!= -2 ||
3897 strcmp(sra
->text_version
, "imsm") != 0) {
3902 for (sd
= sra
->devs
, i
= 0; sd
; sd
= sd
->next
, i
++) {
3903 struct intel_super
*s
= alloc_super();
3911 s
->next
= super_list
;
3915 sprintf(nm
, "%d:%d", sd
->disk
.major
, sd
->disk
.minor
);
3916 dfd
= dev_open(nm
, O_RDWR
);
3920 rv
= find_intel_hba_capability(dfd
, s
, devname
);
3921 /* no orom/efi or non-intel hba of the disk */
3925 err
= load_and_parse_mpb(dfd
, s
, NULL
, 1);
3927 /* retry the load if we might have raced against mdmon */
3928 if (err
== 3 && mdmon_running(devnum
))
3929 for (retry
= 0; retry
< 3; retry
++) {
3931 err
= load_and_parse_mpb(dfd
, s
, NULL
, 1);
3939 /* all mpbs enter, maybe one leaves */
3940 super
= imsm_thunderdome(&super_list
, i
);
3946 if (find_missing(super
) != 0) {
3952 /* load migration record */
3953 err
= load_imsm_migr_rec(super
, NULL
);
3959 /* Check migration compatibility */
3960 if (check_mpb_migr_compatibility(super
) != 0) {
3961 fprintf(stderr
, Name
": Unsupported migration detected");
3963 fprintf(stderr
, " on %s\n", devname
);
3965 fprintf(stderr
, " (IMSM).\n");
3974 while (super_list
) {
3975 struct intel_super
*s
= super_list
;
3977 super_list
= super_list
->next
;
3986 st
->container_dev
= devnum
;
3987 if (err
== 0 && st
->ss
== NULL
) {
3988 st
->ss
= &super_imsm
;
3989 st
->minor_version
= 0;
3990 st
->max_devs
= IMSM_MAX_DEVICES
;
3995 static int load_container_imsm(struct supertype
*st
, int fd
, char *devname
)
3997 return load_super_imsm_all(st
, fd
, &st
->sb
, devname
);
4001 static int load_super_imsm(struct supertype
*st
, int fd
, char *devname
)
4003 struct intel_super
*super
;
4006 if (test_partition(fd
))
4007 /* IMSM not allowed on partitions */
4010 free_super_imsm(st
);
4012 super
= alloc_super();
4015 Name
": malloc of %zu failed.\n",
4019 /* Load hba and capabilities if they exist.
4020 * But do not preclude loading metadata in case capabilities or hba are
4021 * non-compliant and ignore_hw_compat is set.
4023 rv
= find_intel_hba_capability(fd
, super
, devname
);
4024 /* no orom/efi or non-intel hba of the disk */
4025 if ((rv
!= 0) && (st
->ignore_hw_compat
== 0)) {
4028 Name
": No OROM/EFI properties for %s\n", devname
);
4032 rv
= load_and_parse_mpb(fd
, super
, devname
, 0);
4037 Name
": Failed to load all information "
4038 "sections on %s\n", devname
);
4044 if (st
->ss
== NULL
) {
4045 st
->ss
= &super_imsm
;
4046 st
->minor_version
= 0;
4047 st
->max_devs
= IMSM_MAX_DEVICES
;
4050 /* load migration record */
4051 if (load_imsm_migr_rec(super
, NULL
) == 0) {
4052 /* Check for unsupported migration features */
4053 if (check_mpb_migr_compatibility(super
) != 0) {
4055 Name
": Unsupported migration detected");
4057 fprintf(stderr
, " on %s\n", devname
);
4059 fprintf(stderr
, " (IMSM).\n");
4067 static __u16
info_to_blocks_per_strip(mdu_array_info_t
*info
)
4069 if (info
->level
== 1)
4071 return info
->chunk_size
>> 9;
4074 static __u32
info_to_num_data_stripes(mdu_array_info_t
*info
, int num_domains
)
4078 num_stripes
= (info
->size
* 2) / info_to_blocks_per_strip(info
);
4079 num_stripes
/= num_domains
;
4084 static __u32
info_to_blocks_per_member(mdu_array_info_t
*info
)
4086 if (info
->level
== 1)
4087 return info
->size
* 2;
4089 return (info
->size
* 2) & ~(info_to_blocks_per_strip(info
) - 1);
4092 static void imsm_update_version_info(struct intel_super
*super
)
4094 /* update the version and attributes */
4095 struct imsm_super
*mpb
= super
->anchor
;
4097 struct imsm_dev
*dev
;
4098 struct imsm_map
*map
;
4101 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
4102 dev
= get_imsm_dev(super
, i
);
4103 map
= get_imsm_map(dev
, 0);
4104 if (__le32_to_cpu(dev
->size_high
) > 0)
4105 mpb
->attributes
|= MPB_ATTRIB_2TB
;
4107 /* FIXME detect when an array spans a port multiplier */
4109 mpb
->attributes
|= MPB_ATTRIB_PM
;
4112 if (mpb
->num_raid_devs
> 1 ||
4113 mpb
->attributes
!= MPB_ATTRIB_CHECKSUM_VERIFY
) {
4114 version
= MPB_VERSION_ATTRIBS
;
4115 switch (get_imsm_raid_level(map
)) {
4116 case 0: mpb
->attributes
|= MPB_ATTRIB_RAID0
; break;
4117 case 1: mpb
->attributes
|= MPB_ATTRIB_RAID1
; break;
4118 case 10: mpb
->attributes
|= MPB_ATTRIB_RAID10
; break;
4119 case 5: mpb
->attributes
|= MPB_ATTRIB_RAID5
; break;
4122 if (map
->num_members
>= 5)
4123 version
= MPB_VERSION_5OR6_DISK_ARRAY
;
4124 else if (dev
->status
== DEV_CLONE_N_GO
)
4125 version
= MPB_VERSION_CNG
;
4126 else if (get_imsm_raid_level(map
) == 5)
4127 version
= MPB_VERSION_RAID5
;
4128 else if (map
->num_members
>= 3)
4129 version
= MPB_VERSION_3OR4_DISK_ARRAY
;
4130 else if (get_imsm_raid_level(map
) == 1)
4131 version
= MPB_VERSION_RAID1
;
4133 version
= MPB_VERSION_RAID0
;
4135 strcpy(((char *) mpb
->sig
) + strlen(MPB_SIGNATURE
), version
);
4139 static int check_name(struct intel_super
*super
, char *name
, int quiet
)
4141 struct imsm_super
*mpb
= super
->anchor
;
4142 char *reason
= NULL
;
4145 if (strlen(name
) > MAX_RAID_SERIAL_LEN
)
4146 reason
= "must be 16 characters or less";
4148 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
4149 struct imsm_dev
*dev
= get_imsm_dev(super
, i
);
4151 if (strncmp((char *) dev
->volume
, name
, MAX_RAID_SERIAL_LEN
) == 0) {
4152 reason
= "already exists";
4157 if (reason
&& !quiet
)
4158 fprintf(stderr
, Name
": imsm volume name %s\n", reason
);
4163 static int init_super_imsm_volume(struct supertype
*st
, mdu_array_info_t
*info
,
4164 unsigned long long size
, char *name
,
4165 char *homehost
, int *uuid
)
4167 /* We are creating a volume inside a pre-existing container.
4168 * so st->sb is already set.
4170 struct intel_super
*super
= st
->sb
;
4171 struct imsm_super
*mpb
= super
->anchor
;
4172 struct intel_dev
*dv
;
4173 struct imsm_dev
*dev
;
4174 struct imsm_vol
*vol
;
4175 struct imsm_map
*map
;
4176 int idx
= mpb
->num_raid_devs
;
4178 unsigned long long array_blocks
;
4179 size_t size_old
, size_new
;
4180 __u32 num_data_stripes
;
4182 if (super
->orom
&& mpb
->num_raid_devs
>= super
->orom
->vpa
) {
4183 fprintf(stderr
, Name
": This imsm-container already has the "
4184 "maximum of %d volumes\n", super
->orom
->vpa
);
4188 /* ensure the mpb is large enough for the new data */
4189 size_old
= __le32_to_cpu(mpb
->mpb_size
);
4190 size_new
= disks_to_mpb_size(info
->nr_disks
);
4191 if (size_new
> size_old
) {
4193 size_t size_round
= ROUND_UP(size_new
, 512);
4195 if (posix_memalign(&mpb_new
, 512, size_round
) != 0) {
4196 fprintf(stderr
, Name
": could not allocate new mpb\n");
4199 if (posix_memalign(&super
->migr_rec_buf
, 512, 512) != 0) {
4200 fprintf(stderr
, Name
4201 ": %s could not allocate migr_rec buffer\n",
4208 memcpy(mpb_new
, mpb
, size_old
);
4211 super
->anchor
= mpb_new
;
4212 mpb
->mpb_size
= __cpu_to_le32(size_new
);
4213 memset(mpb_new
+ size_old
, 0, size_round
- size_old
);
4215 super
->current_vol
= idx
;
4217 /* handle 'failed_disks' by either:
4218 * a) create dummy disk entries in the table if this the first
4219 * volume in the array. We add them here as this is the only
4220 * opportunity to add them. add_to_super_imsm_volume()
4221 * handles the non-failed disks and continues incrementing
4223 * b) validate that 'failed_disks' matches the current number
4224 * of missing disks if the container is populated
4226 if (super
->current_vol
== 0) {
4228 for (i
= 0; i
< info
->failed_disks
; i
++) {
4229 struct imsm_disk
*disk
;
4232 disk
= __get_imsm_disk(mpb
, i
);
4233 disk
->status
= CONFIGURED_DISK
| FAILED_DISK
;
4234 disk
->scsi_id
= __cpu_to_le32(~(__u32
)0);
4235 snprintf((char *) disk
->serial
, MAX_RAID_SERIAL_LEN
,
4238 find_missing(super
);
4243 for (d
= super
->missing
; d
; d
= d
->next
)
4245 if (info
->failed_disks
> missing
) {
4246 fprintf(stderr
, Name
": unable to add 'missing' disk to container\n");
4251 if (!check_name(super
, name
, 0))
4253 dv
= malloc(sizeof(*dv
));
4255 fprintf(stderr
, Name
": failed to allocate device list entry\n");
4258 dev
= calloc(1, sizeof(*dev
) + sizeof(__u32
) * (info
->raid_disks
- 1));
4261 fprintf(stderr
, Name
": could not allocate raid device\n");
4265 strncpy((char *) dev
->volume
, name
, MAX_RAID_SERIAL_LEN
);
4266 if (info
->level
== 1)
4267 array_blocks
= info_to_blocks_per_member(info
);
4269 array_blocks
= calc_array_size(info
->level
, info
->raid_disks
,
4270 info
->layout
, info
->chunk_size
,
4272 /* round array size down to closest MB */
4273 array_blocks
= (array_blocks
>> SECT_PER_MB_SHIFT
) << SECT_PER_MB_SHIFT
;
4275 dev
->size_low
= __cpu_to_le32((__u32
) array_blocks
);
4276 dev
->size_high
= __cpu_to_le32((__u32
) (array_blocks
>> 32));
4277 dev
->status
= (DEV_READ_COALESCING
| DEV_WRITE_COALESCING
);
4279 vol
->migr_state
= 0;
4280 set_migr_type(dev
, MIGR_INIT
);
4281 vol
->dirty
= !info
->state
;
4282 vol
->curr_migr_unit
= 0;
4283 map
= get_imsm_map(dev
, 0);
4284 map
->pba_of_lba0
= __cpu_to_le32(super
->create_offset
);
4285 map
->blocks_per_member
= __cpu_to_le32(info_to_blocks_per_member(info
));
4286 map
->blocks_per_strip
= __cpu_to_le16(info_to_blocks_per_strip(info
));
4287 map
->failed_disk_num
= ~0;
4288 map
->map_state
= info
->failed_disks
? IMSM_T_STATE_DEGRADED
: IMSM_T_STATE_NORMAL
;
4291 if (info
->level
== 1 && info
->raid_disks
> 2) {
4294 fprintf(stderr
, Name
": imsm does not support more than 2 disks"
4295 "in a raid1 volume\n");
4299 map
->raid_level
= info
->level
;
4300 if (info
->level
== 10) {
4301 map
->raid_level
= 1;
4302 map
->num_domains
= info
->raid_disks
/ 2;
4303 } else if (info
->level
== 1)
4304 map
->num_domains
= info
->raid_disks
;
4306 map
->num_domains
= 1;
4308 num_data_stripes
= info_to_num_data_stripes(info
, map
->num_domains
);
4309 map
->num_data_stripes
= __cpu_to_le32(num_data_stripes
);
4311 map
->num_members
= info
->raid_disks
;
4312 for (i
= 0; i
< map
->num_members
; i
++) {
4313 /* initialized in add_to_super */
4314 set_imsm_ord_tbl_ent(map
, i
, IMSM_ORD_REBUILD
);
4316 mpb
->num_raid_devs
++;
4319 dv
->index
= super
->current_vol
;
4320 dv
->next
= super
->devlist
;
4321 super
->devlist
= dv
;
4323 imsm_update_version_info(super
);
4328 static int init_super_imsm(struct supertype
*st
, mdu_array_info_t
*info
,
4329 unsigned long long size
, char *name
,
4330 char *homehost
, int *uuid
)
4332 /* This is primarily called by Create when creating a new array.
4333 * We will then get add_to_super called for each component, and then
4334 * write_init_super called to write it out to each device.
4335 * For IMSM, Create can create on fresh devices or on a pre-existing
4337 * To create on a pre-existing array a different method will be called.
4338 * This one is just for fresh drives.
4340 struct intel_super
*super
;
4341 struct imsm_super
*mpb
;
4346 return init_super_imsm_volume(st
, info
, size
, name
, homehost
, uuid
);
4349 mpb_size
= disks_to_mpb_size(info
->nr_disks
);
4353 super
= alloc_super();
4354 if (super
&& posix_memalign(&super
->buf
, 512, mpb_size
) != 0) {
4359 fprintf(stderr
, Name
4360 ": %s could not allocate superblock\n", __func__
);
4363 if (posix_memalign(&super
->migr_rec_buf
, 512, 512) != 0) {
4364 fprintf(stderr
, Name
4365 ": %s could not allocate migr_rec buffer\n", __func__
);
4370 memset(super
->buf
, 0, mpb_size
);
4372 mpb
->mpb_size
= __cpu_to_le32(mpb_size
);
4376 /* zeroing superblock */
4380 mpb
->attributes
= MPB_ATTRIB_CHECKSUM_VERIFY
;
4382 version
= (char *) mpb
->sig
;
4383 strcpy(version
, MPB_SIGNATURE
);
4384 version
+= strlen(MPB_SIGNATURE
);
4385 strcpy(version
, MPB_VERSION_RAID0
);
4391 static int add_to_super_imsm_volume(struct supertype
*st
, mdu_disk_info_t
*dk
,
4392 int fd
, char *devname
)
4394 struct intel_super
*super
= st
->sb
;
4395 struct imsm_super
*mpb
= super
->anchor
;
4396 struct imsm_disk
*_disk
;
4397 struct imsm_dev
*dev
;
4398 struct imsm_map
*map
;
4402 dev
= get_imsm_dev(super
, super
->current_vol
);
4403 map
= get_imsm_map(dev
, 0);
4405 if (! (dk
->state
& (1<<MD_DISK_SYNC
))) {
4406 fprintf(stderr
, Name
": %s: Cannot add spare devices to IMSM volume\n",
4412 /* we're doing autolayout so grab the pre-marked (in
4413 * validate_geometry) raid_disk
4415 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
4416 if (dl
->raiddisk
== dk
->raid_disk
)
4419 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
4420 if (dl
->major
== dk
->major
&&
4421 dl
->minor
== dk
->minor
)
4426 fprintf(stderr
, Name
": %s is not a member of the same container\n", devname
);
4430 /* add a pristine spare to the metadata */
4431 if (dl
->index
< 0) {
4432 dl
->index
= super
->anchor
->num_disks
;
4433 super
->anchor
->num_disks
++;
4435 /* Check the device has not already been added */
4436 slot
= get_imsm_disk_slot(map
, dl
->index
);
4438 (get_imsm_ord_tbl_ent(dev
, slot
, -1) & IMSM_ORD_REBUILD
) == 0) {
4439 fprintf(stderr
, Name
": %s has been included in this array twice\n",
4443 set_imsm_ord_tbl_ent(map
, dk
->raid_disk
, dl
->index
);
4444 dl
->disk
.status
= CONFIGURED_DISK
;
4446 /* update size of 'missing' disks to be at least as large as the
4447 * largest acitve member (we only have dummy missing disks when
4448 * creating the first volume)
4450 if (super
->current_vol
== 0) {
4451 for (df
= super
->missing
; df
; df
= df
->next
) {
4452 if (dl
->disk
.total_blocks
> df
->disk
.total_blocks
)
4453 df
->disk
.total_blocks
= dl
->disk
.total_blocks
;
4454 _disk
= __get_imsm_disk(mpb
, df
->index
);
4459 /* refresh unset/failed slots to point to valid 'missing' entries */
4460 for (df
= super
->missing
; df
; df
= df
->next
)
4461 for (slot
= 0; slot
< mpb
->num_disks
; slot
++) {
4462 __u32 ord
= get_imsm_ord_tbl_ent(dev
, slot
, -1);
4464 if ((ord
& IMSM_ORD_REBUILD
) == 0)
4466 set_imsm_ord_tbl_ent(map
, slot
, df
->index
| IMSM_ORD_REBUILD
);
4467 dprintf("set slot:%d to missing disk:%d\n", slot
, df
->index
);
4471 /* if we are creating the first raid device update the family number */
4472 if (super
->current_vol
== 0) {
4474 struct imsm_dev
*_dev
= __get_imsm_dev(mpb
, 0);
4476 _disk
= __get_imsm_disk(mpb
, dl
->index
);
4477 if (!_dev
|| !_disk
) {
4478 fprintf(stderr
, Name
": BUG mpb setup error\n");
4484 sum
+= __gen_imsm_checksum(mpb
);
4485 mpb
->family_num
= __cpu_to_le32(sum
);
4486 mpb
->orig_family_num
= mpb
->family_num
;
4488 super
->current_disk
= dl
;
4493 * Function marks disk as spare and restores disk serial
4494 * in case it was previously marked as failed by takeover operation
4496 * -1 : critical error
4497 * 0 : disk is marked as spare but serial is not set
4500 int mark_spare(struct dl
*disk
)
4502 __u8 serial
[MAX_RAID_SERIAL_LEN
];
4509 if (!imsm_read_serial(disk
->fd
, NULL
, serial
)) {
4510 /* Restore disk serial number, because takeover marks disk
4511 * as failed and adds to serial ':0' before it becomes
4514 serialcpy(disk
->serial
, serial
);
4515 serialcpy(disk
->disk
.serial
, serial
);
4518 disk
->disk
.status
= SPARE_DISK
;
4524 static int add_to_super_imsm(struct supertype
*st
, mdu_disk_info_t
*dk
,
4525 int fd
, char *devname
)
4527 struct intel_super
*super
= st
->sb
;
4529 unsigned long long size
;
4534 /* If we are on an RAID enabled platform check that the disk is
4535 * attached to the raid controller.
4536 * We do not need to test disks attachment for container based additions,
4537 * they shall be already tested when container was created/assembled.
4539 rv
= find_intel_hba_capability(fd
, super
, devname
);
4540 /* no orom/efi or non-intel hba of the disk */
4542 dprintf("capability: %p fd: %d ret: %d\n",
4543 super
->orom
, fd
, rv
);
4547 if (super
->current_vol
>= 0)
4548 return add_to_super_imsm_volume(st
, dk
, fd
, devname
);
4551 dd
= malloc(sizeof(*dd
));
4554 Name
": malloc failed %s:%d.\n", __func__
, __LINE__
);
4557 memset(dd
, 0, sizeof(*dd
));
4558 dd
->major
= major(stb
.st_rdev
);
4559 dd
->minor
= minor(stb
.st_rdev
);
4560 dd
->devname
= devname
? strdup(devname
) : NULL
;
4563 dd
->action
= DISK_ADD
;
4564 rv
= imsm_read_serial(fd
, devname
, dd
->serial
);
4567 Name
": failed to retrieve scsi serial, aborting\n");
4572 get_dev_size(fd
, NULL
, &size
);
4574 serialcpy(dd
->disk
.serial
, dd
->serial
);
4575 dd
->disk
.total_blocks
= __cpu_to_le32(size
);
4577 if (sysfs_disk_to_scsi_id(fd
, &id
) == 0)
4578 dd
->disk
.scsi_id
= __cpu_to_le32(id
);
4580 dd
->disk
.scsi_id
= __cpu_to_le32(0);
4582 if (st
->update_tail
) {
4583 dd
->next
= super
->disk_mgmt_list
;
4584 super
->disk_mgmt_list
= dd
;
4586 dd
->next
= super
->disks
;
4588 super
->updates_pending
++;
4595 static int remove_from_super_imsm(struct supertype
*st
, mdu_disk_info_t
*dk
)
4597 struct intel_super
*super
= st
->sb
;
4600 /* remove from super works only in mdmon - for communication
4601 * manager - monitor. Check if communication memory buffer
4604 if (!st
->update_tail
) {
4606 Name
": %s shall be used in mdmon context only"
4607 "(line %d).\n", __func__
, __LINE__
);
4610 dd
= malloc(sizeof(*dd
));
4613 Name
": malloc failed %s:%d.\n", __func__
, __LINE__
);
4616 memset(dd
, 0, sizeof(*dd
));
4617 dd
->major
= dk
->major
;
4618 dd
->minor
= dk
->minor
;
4621 dd
->action
= DISK_REMOVE
;
4623 dd
->next
= super
->disk_mgmt_list
;
4624 super
->disk_mgmt_list
= dd
;
4630 static int store_imsm_mpb(int fd
, struct imsm_super
*mpb
);
4634 struct imsm_super anchor
;
4635 } spare_record
__attribute__ ((aligned(512)));
4637 /* spare records have their own family number and do not have any defined raid
4640 static int write_super_imsm_spares(struct intel_super
*super
, int doclose
)
4642 struct imsm_super
*mpb
= super
->anchor
;
4643 struct imsm_super
*spare
= &spare_record
.anchor
;
4647 spare
->mpb_size
= __cpu_to_le32(sizeof(struct imsm_super
)),
4648 spare
->generation_num
= __cpu_to_le32(1UL),
4649 spare
->attributes
= MPB_ATTRIB_CHECKSUM_VERIFY
;
4650 spare
->num_disks
= 1,
4651 spare
->num_raid_devs
= 0,
4652 spare
->cache_size
= mpb
->cache_size
,
4653 spare
->pwr_cycle_count
= __cpu_to_le32(1),
4655 snprintf((char *) spare
->sig
, MAX_SIGNATURE_LENGTH
,
4656 MPB_SIGNATURE MPB_VERSION_RAID0
);
4658 for (d
= super
->disks
; d
; d
= d
->next
) {
4662 spare
->disk
[0] = d
->disk
;
4663 sum
= __gen_imsm_checksum(spare
);
4664 spare
->family_num
= __cpu_to_le32(sum
);
4665 spare
->orig_family_num
= 0;
4666 sum
= __gen_imsm_checksum(spare
);
4667 spare
->check_sum
= __cpu_to_le32(sum
);
4669 if (store_imsm_mpb(d
->fd
, spare
)) {
4670 fprintf(stderr
, "%s: failed for device %d:%d %s\n",
4671 __func__
, d
->major
, d
->minor
, strerror(errno
));
4683 static int write_super_imsm(struct supertype
*st
, int doclose
)
4685 struct intel_super
*super
= st
->sb
;
4686 struct imsm_super
*mpb
= super
->anchor
;
4692 __u32 mpb_size
= sizeof(struct imsm_super
) - sizeof(struct imsm_disk
);
4694 int clear_migration_record
= 1;
4696 /* 'generation' is incremented everytime the metadata is written */
4697 generation
= __le32_to_cpu(mpb
->generation_num
);
4699 mpb
->generation_num
= __cpu_to_le32(generation
);
4701 /* fix up cases where previous mdadm releases failed to set
4704 if (mpb
->orig_family_num
== 0)
4705 mpb
->orig_family_num
= mpb
->family_num
;
4707 for (d
= super
->disks
; d
; d
= d
->next
) {
4711 mpb
->disk
[d
->index
] = d
->disk
;
4715 for (d
= super
->missing
; d
; d
= d
->next
) {
4716 mpb
->disk
[d
->index
] = d
->disk
;
4719 mpb
->num_disks
= num_disks
;
4720 mpb_size
+= sizeof(struct imsm_disk
) * mpb
->num_disks
;
4722 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
4723 struct imsm_dev
*dev
= __get_imsm_dev(mpb
, i
);
4724 struct imsm_dev
*dev2
= get_imsm_dev(super
, i
);
4726 imsm_copy_dev(dev
, dev2
);
4727 mpb_size
+= sizeof_imsm_dev(dev
, 0);
4729 if (is_gen_migration(dev2
))
4730 clear_migration_record
= 0;
4732 mpb_size
+= __le32_to_cpu(mpb
->bbm_log_size
);
4733 mpb
->mpb_size
= __cpu_to_le32(mpb_size
);
4735 /* recalculate checksum */
4736 sum
= __gen_imsm_checksum(mpb
);
4737 mpb
->check_sum
= __cpu_to_le32(sum
);
4739 if (clear_migration_record
)
4740 memset(super
->migr_rec_buf
, 0, 512);
4742 /* write the mpb for disks that compose raid devices */
4743 for (d
= super
->disks
; d
; d
= d
->next
) {
4744 if (d
->index
< 0 || is_failed(&d
->disk
))
4746 if (store_imsm_mpb(d
->fd
, mpb
))
4747 fprintf(stderr
, "%s: failed for device %d:%d %s\n",
4748 __func__
, d
->major
, d
->minor
, strerror(errno
));
4749 if (clear_migration_record
) {
4750 unsigned long long dsize
;
4752 get_dev_size(d
->fd
, NULL
, &dsize
);
4753 if (lseek64(d
->fd
, dsize
- 512, SEEK_SET
) >= 0) {
4754 if (write(d
->fd
, super
->migr_rec_buf
, 512) != 512)
4755 perror("Write migr_rec failed");
4765 return write_super_imsm_spares(super
, doclose
);
4771 static int create_array(struct supertype
*st
, int dev_idx
)
4774 struct imsm_update_create_array
*u
;
4775 struct intel_super
*super
= st
->sb
;
4776 struct imsm_dev
*dev
= get_imsm_dev(super
, dev_idx
);
4777 struct imsm_map
*map
= get_imsm_map(dev
, 0);
4778 struct disk_info
*inf
;
4779 struct imsm_disk
*disk
;
4782 len
= sizeof(*u
) - sizeof(*dev
) + sizeof_imsm_dev(dev
, 0) +
4783 sizeof(*inf
) * map
->num_members
;
4786 fprintf(stderr
, "%s: failed to allocate update buffer\n",
4791 u
->type
= update_create_array
;
4792 u
->dev_idx
= dev_idx
;
4793 imsm_copy_dev(&u
->dev
, dev
);
4794 inf
= get_disk_info(u
);
4795 for (i
= 0; i
< map
->num_members
; i
++) {
4796 int idx
= get_imsm_disk_idx(dev
, i
, -1);
4798 disk
= get_imsm_disk(super
, idx
);
4799 serialcpy(inf
[i
].serial
, disk
->serial
);
4801 append_metadata_update(st
, u
, len
);
4806 static int mgmt_disk(struct supertype
*st
)
4808 struct intel_super
*super
= st
->sb
;
4810 struct imsm_update_add_remove_disk
*u
;
4812 if (!super
->disk_mgmt_list
)
4818 fprintf(stderr
, "%s: failed to allocate update buffer\n",
4823 u
->type
= update_add_remove_disk
;
4824 append_metadata_update(st
, u
, len
);
4829 static int write_init_super_imsm(struct supertype
*st
)
4831 struct intel_super
*super
= st
->sb
;
4832 int current_vol
= super
->current_vol
;
4834 /* we are done with current_vol reset it to point st at the container */
4835 super
->current_vol
= -1;
4837 if (st
->update_tail
) {
4838 /* queue the recently created array / added disk
4839 * as a metadata update */
4842 /* determine if we are creating a volume or adding a disk */
4843 if (current_vol
< 0) {
4844 /* in the mgmt (add/remove) disk case we are running
4845 * in mdmon context, so don't close fd's
4847 return mgmt_disk(st
);
4849 rv
= create_array(st
, current_vol
);
4854 for (d
= super
->disks
; d
; d
= d
->next
)
4855 Kill(d
->devname
, NULL
, 0, 1, 1);
4856 return write_super_imsm(st
, 1);
4861 static int store_super_imsm(struct supertype
*st
, int fd
)
4863 struct intel_super
*super
= st
->sb
;
4864 struct imsm_super
*mpb
= super
? super
->anchor
: NULL
;
4870 return store_imsm_mpb(fd
, mpb
);
4876 static int imsm_bbm_log_size(struct imsm_super
*mpb
)
4878 return __le32_to_cpu(mpb
->bbm_log_size
);
4882 static int validate_geometry_imsm_container(struct supertype
*st
, int level
,
4883 int layout
, int raiddisks
, int chunk
,
4884 unsigned long long size
, char *dev
,
4885 unsigned long long *freesize
,
4889 unsigned long long ldsize
;
4890 struct intel_super
*super
=NULL
;
4893 if (level
!= LEVEL_CONTAINER
)
4898 fd
= open(dev
, O_RDONLY
|O_EXCL
, 0);
4901 fprintf(stderr
, Name
": imsm: Cannot open %s: %s\n",
4902 dev
, strerror(errno
));
4905 if (!get_dev_size(fd
, dev
, &ldsize
)) {
4910 /* capabilities retrieve could be possible
4911 * note that there is no fd for the disks in array.
4913 super
= alloc_super();
4916 Name
": malloc of %zu failed.\n",
4922 rv
= find_intel_hba_capability(fd
, super
, verbose
? dev
: NULL
);
4926 fd2devname(fd
, str
);
4927 dprintf("validate_geometry_imsm_container: fd: %d %s orom: %p rv: %d raiddisk: %d\n",
4928 fd
, str
, super
->orom
, rv
, raiddisks
);
4930 /* no orom/efi or non-intel hba of the disk */
4936 if (super
->orom
&& raiddisks
> super
->orom
->tds
) {
4938 fprintf(stderr
, Name
": %d exceeds maximum number of"
4939 " platform supported disks: %d\n",
4940 raiddisks
, super
->orom
->tds
);
4946 *freesize
= avail_size_imsm(st
, ldsize
>> 9);
4952 static unsigned long long find_size(struct extent
*e
, int *idx
, int num_extents
)
4954 const unsigned long long base_start
= e
[*idx
].start
;
4955 unsigned long long end
= base_start
+ e
[*idx
].size
;
4958 if (base_start
== end
)
4962 for (i
= *idx
; i
< num_extents
; i
++) {
4963 /* extend overlapping extents */
4964 if (e
[i
].start
>= base_start
&&
4965 e
[i
].start
<= end
) {
4968 if (e
[i
].start
+ e
[i
].size
> end
)
4969 end
= e
[i
].start
+ e
[i
].size
;
4970 } else if (e
[i
].start
> end
) {
4976 return end
- base_start
;
4979 static unsigned long long merge_extents(struct intel_super
*super
, int sum_extents
)
4981 /* build a composite disk with all known extents and generate a new
4982 * 'maxsize' given the "all disks in an array must share a common start
4983 * offset" constraint
4985 struct extent
*e
= calloc(sum_extents
, sizeof(*e
));
4989 unsigned long long pos
;
4990 unsigned long long start
= 0;
4991 unsigned long long maxsize
;
4992 unsigned long reserve
;
4997 /* coalesce and sort all extents. also, check to see if we need to
4998 * reserve space between member arrays
5001 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
5004 for (i
= 0; i
< dl
->extent_cnt
; i
++)
5007 qsort(e
, sum_extents
, sizeof(*e
), cmp_extent
);
5012 while (i
< sum_extents
) {
5013 e
[j
].start
= e
[i
].start
;
5014 e
[j
].size
= find_size(e
, &i
, sum_extents
);
5016 if (e
[j
-1].size
== 0)
5025 unsigned long long esize
;
5027 esize
= e
[i
].start
- pos
;
5028 if (esize
>= maxsize
) {
5033 pos
= e
[i
].start
+ e
[i
].size
;
5035 } while (e
[i
-1].size
);
5041 /* FIXME assumes volume at offset 0 is the first volume in a
5044 if (start_extent
> 0)
5045 reserve
= IMSM_RESERVED_SECTORS
; /* gap between raid regions */
5049 if (maxsize
< reserve
)
5052 super
->create_offset
= ~((__u32
) 0);
5053 if (start
+ reserve
> super
->create_offset
)
5054 return 0; /* start overflows create_offset */
5055 super
->create_offset
= start
+ reserve
;
5057 return maxsize
- reserve
;
5060 static int is_raid_level_supported(const struct imsm_orom
*orom
, int level
, int raiddisks
)
5062 if (level
< 0 || level
== 6 || level
== 4)
5065 /* if we have an orom prevent invalid raid levels */
5068 case 0: return imsm_orom_has_raid0(orom
);
5071 return imsm_orom_has_raid1e(orom
);
5072 return imsm_orom_has_raid1(orom
) && raiddisks
== 2;
5073 case 10: return imsm_orom_has_raid10(orom
) && raiddisks
== 4;
5074 case 5: return imsm_orom_has_raid5(orom
) && raiddisks
> 2;
5077 return 1; /* not on an Intel RAID platform so anything goes */
5082 static int imsm_default_chunk(const struct imsm_orom
*orom
)
5084 /* up to 512 if the plaform supports it, otherwise the platform max.
5085 * 128 if no platform detected
5087 int fs
= max(7, orom
? fls(orom
->sss
) : 0);
5089 return min(512, (1 << fs
));
5092 #define pr_vrb(fmt, arg...) (void) (verbose && fprintf(stderr, Name fmt, ##arg))
5094 validate_geometry_imsm_orom(struct intel_super
*super
, int level
, int layout
,
5095 int raiddisks
, int *chunk
, int verbose
)
5097 /* check/set platform and metadata limits/defaults */
5098 if (super
->orom
&& raiddisks
> super
->orom
->dpa
) {
5099 pr_vrb(": platform supports a maximum of %d disks per array\n",
5104 /* capabilities of OROM tested - copied from validate_geometry_imsm_volume */
5105 if (!is_raid_level_supported(super
->orom
, level
, raiddisks
)) {
5106 pr_vrb(": platform does not support raid%d with %d disk%s\n",
5107 level
, raiddisks
, raiddisks
> 1 ? "s" : "");
5111 if (chunk
&& (*chunk
== 0 || *chunk
== UnSet
))
5112 *chunk
= imsm_default_chunk(super
->orom
);
5114 if (super
->orom
&& chunk
&& !imsm_orom_has_chunk(super
->orom
, *chunk
)) {
5115 pr_vrb(": platform does not support a chunk size of: "
5120 if (layout
!= imsm_level_to_layout(level
)) {
5122 pr_vrb(": imsm raid 5 only supports the left-asymmetric layout\n");
5123 else if (level
== 10)
5124 pr_vrb(": imsm raid 10 only supports the n2 layout\n");
5126 pr_vrb(": imsm unknown layout %#x for this raid level %d\n",
5133 /* validate_geometry_imsm_volume - lifted from validate_geometry_ddf_bvd
5134 * FIX ME add ahci details
5136 static int validate_geometry_imsm_volume(struct supertype
*st
, int level
,
5137 int layout
, int raiddisks
, int *chunk
,
5138 unsigned long long size
, char *dev
,
5139 unsigned long long *freesize
,
5143 struct intel_super
*super
= st
->sb
;
5144 struct imsm_super
*mpb
;
5146 unsigned long long pos
= 0;
5147 unsigned long long maxsize
;
5151 /* We must have the container info already read in. */
5155 mpb
= super
->anchor
;
5157 if (mpb
->num_raid_devs
> 0 && mpb
->num_disks
!= raiddisks
) {
5158 fprintf(stderr
, Name
": the option-rom requires all "
5159 "member disks to be a member of all volumes.\n");
5163 if (!validate_geometry_imsm_orom(super
, level
, layout
, raiddisks
, chunk
, verbose
)) {
5164 fprintf(stderr
, Name
": RAID gemetry validation failed. "
5165 "Cannot proceed with the action(s).\n");
5169 /* General test: make sure there is space for
5170 * 'raiddisks' device extents of size 'size' at a given
5173 unsigned long long minsize
= size
;
5174 unsigned long long start_offset
= MaxSector
;
5177 minsize
= MPB_SECTOR_CNT
+ IMSM_RESERVED_SECTORS
;
5178 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
5183 e
= get_extents(super
, dl
);
5186 unsigned long long esize
;
5187 esize
= e
[i
].start
- pos
;
5188 if (esize
>= minsize
)
5190 if (found
&& start_offset
== MaxSector
) {
5193 } else if (found
&& pos
!= start_offset
) {
5197 pos
= e
[i
].start
+ e
[i
].size
;
5199 } while (e
[i
-1].size
);
5204 if (dcnt
< raiddisks
) {
5206 fprintf(stderr
, Name
": imsm: Not enough "
5207 "devices with space for this array "
5215 /* This device must be a member of the set */
5216 if (stat(dev
, &stb
) < 0)
5218 if ((S_IFMT
& stb
.st_mode
) != S_IFBLK
)
5220 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
5221 if (dl
->major
== (int)major(stb
.st_rdev
) &&
5222 dl
->minor
== (int)minor(stb
.st_rdev
))
5227 fprintf(stderr
, Name
": %s is not in the "
5228 "same imsm set\n", dev
);
5230 } else if (super
->orom
&& dl
->index
< 0 && mpb
->num_raid_devs
) {
5231 /* If a volume is present then the current creation attempt
5232 * cannot incorporate new spares because the orom may not
5233 * understand this configuration (all member disks must be
5234 * members of each array in the container).
5236 fprintf(stderr
, Name
": %s is a spare and a volume"
5237 " is already defined for this container\n", dev
);
5238 fprintf(stderr
, Name
": The option-rom requires all member"
5239 " disks to be a member of all volumes\n");
5243 /* retrieve the largest free space block */
5244 e
= get_extents(super
, dl
);
5249 unsigned long long esize
;
5251 esize
= e
[i
].start
- pos
;
5252 if (esize
>= maxsize
)
5254 pos
= e
[i
].start
+ e
[i
].size
;
5256 } while (e
[i
-1].size
);
5261 fprintf(stderr
, Name
": unable to determine free space for: %s\n",
5265 if (maxsize
< size
) {
5267 fprintf(stderr
, Name
": %s not enough space (%llu < %llu)\n",
5268 dev
, maxsize
, size
);
5272 /* count total number of extents for merge */
5274 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
5276 i
+= dl
->extent_cnt
;
5278 maxsize
= merge_extents(super
, i
);
5280 if (!check_env("IMSM_NO_PLATFORM") &&
5281 mpb
->num_raid_devs
> 0 && size
&& size
!= maxsize
) {
5282 fprintf(stderr
, Name
": attempting to create a second "
5283 "volume with size less then remaining space. "
5288 if (maxsize
< size
|| maxsize
== 0) {
5290 fprintf(stderr
, Name
": not enough space after merge (%llu < %llu)\n",
5295 *freesize
= maxsize
;
5300 static int reserve_space(struct supertype
*st
, int raiddisks
,
5301 unsigned long long size
, int chunk
,
5302 unsigned long long *freesize
)
5304 struct intel_super
*super
= st
->sb
;
5305 struct imsm_super
*mpb
= super
->anchor
;
5310 unsigned long long maxsize
;
5311 unsigned long long minsize
;
5315 /* find the largest common start free region of the possible disks */
5319 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
5325 /* don't activate new spares if we are orom constrained
5326 * and there is already a volume active in the container
5328 if (super
->orom
&& dl
->index
< 0 && mpb
->num_raid_devs
)
5331 e
= get_extents(super
, dl
);
5334 for (i
= 1; e
[i
-1].size
; i
++)
5342 maxsize
= merge_extents(super
, extent_cnt
);
5346 minsize
= chunk
* 2;
5348 if (cnt
< raiddisks
||
5349 (super
->orom
&& used
&& used
!= raiddisks
) ||
5350 maxsize
< minsize
||
5352 fprintf(stderr
, Name
": not enough devices with space to create array.\n");
5353 return 0; /* No enough free spaces large enough */
5365 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
5367 dl
->raiddisk
= cnt
++;
5374 static int validate_geometry_imsm(struct supertype
*st
, int level
, int layout
,
5375 int raiddisks
, int *chunk
, unsigned long long size
,
5376 char *dev
, unsigned long long *freesize
,
5384 * if given unused devices create a container
5385 * if given given devices in a container create a member volume
5387 if (level
== LEVEL_CONTAINER
) {
5388 /* Must be a fresh device to add to a container */
5389 return validate_geometry_imsm_container(st
, level
, layout
,
5391 chunk
?*chunk
:0, size
,
5398 if (!validate_geometry_imsm_orom(st
->sb
, level
, layout
,
5402 /* we are being asked to automatically layout a
5403 * new volume based on the current contents of
5404 * the container. If the the parameters can be
5405 * satisfied reserve_space will record the disks,
5406 * start offset, and size of the volume to be
5407 * created. add_to_super and getinfo_super
5408 * detect when autolayout is in progress.
5411 return reserve_space(st
, raiddisks
, size
,
5412 chunk
?*chunk
:0, freesize
);
5417 /* creating in a given container */
5418 return validate_geometry_imsm_volume(st
, level
, layout
,
5419 raiddisks
, chunk
, size
,
5420 dev
, freesize
, verbose
);
5423 /* This device needs to be a device in an 'imsm' container */
5424 fd
= open(dev
, O_RDONLY
|O_EXCL
, 0);
5428 Name
": Cannot create this array on device %s\n",
5433 if (errno
!= EBUSY
|| (fd
= open(dev
, O_RDONLY
, 0)) < 0) {
5435 fprintf(stderr
, Name
": Cannot open %s: %s\n",
5436 dev
, strerror(errno
));
5439 /* Well, it is in use by someone, maybe an 'imsm' container. */
5440 cfd
= open_container(fd
);
5444 fprintf(stderr
, Name
": Cannot use %s: It is busy\n",
5448 sra
= sysfs_read(cfd
, 0, GET_VERSION
);
5449 if (sra
&& sra
->array
.major_version
== -1 &&
5450 strcmp(sra
->text_version
, "imsm") == 0)
5454 /* This is a member of a imsm container. Load the container
5455 * and try to create a volume
5457 struct intel_super
*super
;
5459 if (load_super_imsm_all(st
, cfd
, (void **) &super
, NULL
) == 0) {
5461 st
->container_dev
= fd2devnum(cfd
);
5463 return validate_geometry_imsm_volume(st
, level
, layout
,
5472 fprintf(stderr
, Name
": failed container membership check\n");
5478 static void default_geometry_imsm(struct supertype
*st
, int *level
, int *layout
, int *chunk
)
5480 struct intel_super
*super
= st
->sb
;
5482 if (level
&& *level
== UnSet
)
5483 *level
= LEVEL_CONTAINER
;
5485 if (level
&& layout
&& *layout
== UnSet
)
5486 *layout
= imsm_level_to_layout(*level
);
5488 if (chunk
&& (*chunk
== UnSet
|| *chunk
== 0))
5489 *chunk
= imsm_default_chunk(super
->orom
);
5492 static void handle_missing(struct intel_super
*super
, struct imsm_dev
*dev
);
5494 static int kill_subarray_imsm(struct supertype
*st
)
5496 /* remove the subarray currently referenced by ->current_vol */
5498 struct intel_dev
**dp
;
5499 struct intel_super
*super
= st
->sb
;
5500 __u8 current_vol
= super
->current_vol
;
5501 struct imsm_super
*mpb
= super
->anchor
;
5503 if (super
->current_vol
< 0)
5505 super
->current_vol
= -1; /* invalidate subarray cursor */
5507 /* block deletions that would change the uuid of active subarrays
5509 * FIXME when immutable ids are available, but note that we'll
5510 * also need to fixup the invalidated/active subarray indexes in
5513 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
5516 if (i
< current_vol
)
5518 sprintf(subarray
, "%u", i
);
5519 if (is_subarray_active(subarray
, st
->devname
)) {
5521 Name
": deleting subarray-%d would change the UUID of active subarray-%d, aborting\n",
5528 if (st
->update_tail
) {
5529 struct imsm_update_kill_array
*u
= malloc(sizeof(*u
));
5533 u
->type
= update_kill_array
;
5534 u
->dev_idx
= current_vol
;
5535 append_metadata_update(st
, u
, sizeof(*u
));
5540 for (dp
= &super
->devlist
; *dp
;)
5541 if ((*dp
)->index
== current_vol
) {
5544 handle_missing(super
, (*dp
)->dev
);
5545 if ((*dp
)->index
> current_vol
)
5550 /* no more raid devices, all active components are now spares,
5551 * but of course failed are still failed
5553 if (--mpb
->num_raid_devs
== 0) {
5556 for (d
= super
->disks
; d
; d
= d
->next
)
5561 super
->updates_pending
++;
5566 static int update_subarray_imsm(struct supertype
*st
, char *subarray
,
5567 char *update
, struct mddev_ident
*ident
)
5569 /* update the subarray currently referenced by ->current_vol */
5570 struct intel_super
*super
= st
->sb
;
5571 struct imsm_super
*mpb
= super
->anchor
;
5573 if (strcmp(update
, "name") == 0) {
5574 char *name
= ident
->name
;
5578 if (is_subarray_active(subarray
, st
->devname
)) {
5580 Name
": Unable to update name of active subarray\n");
5584 if (!check_name(super
, name
, 0))
5587 vol
= strtoul(subarray
, &ep
, 10);
5588 if (*ep
!= '\0' || vol
>= super
->anchor
->num_raid_devs
)
5591 if (st
->update_tail
) {
5592 struct imsm_update_rename_array
*u
= malloc(sizeof(*u
));
5596 u
->type
= update_rename_array
;
5598 snprintf((char *) u
->name
, MAX_RAID_SERIAL_LEN
, "%s", name
);
5599 append_metadata_update(st
, u
, sizeof(*u
));
5601 struct imsm_dev
*dev
;
5604 dev
= get_imsm_dev(super
, vol
);
5605 snprintf((char *) dev
->volume
, MAX_RAID_SERIAL_LEN
, "%s", name
);
5606 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
5607 dev
= get_imsm_dev(super
, i
);
5608 handle_missing(super
, dev
);
5610 super
->updates_pending
++;
5618 static int is_gen_migration(struct imsm_dev
*dev
)
5623 if (!dev
->vol
.migr_state
)
5626 if (migr_type(dev
) == MIGR_GEN_MIGR
)
5631 #endif /* MDASSEMBLE */
5633 static int is_rebuilding(struct imsm_dev
*dev
)
5635 struct imsm_map
*migr_map
;
5637 if (!dev
->vol
.migr_state
)
5640 if (migr_type(dev
) != MIGR_REBUILD
)
5643 migr_map
= get_imsm_map(dev
, 1);
5645 if (migr_map
->map_state
== IMSM_T_STATE_DEGRADED
)
5651 static void update_recovery_start(struct intel_super
*super
,
5652 struct imsm_dev
*dev
,
5653 struct mdinfo
*array
)
5655 struct mdinfo
*rebuild
= NULL
;
5659 if (!is_rebuilding(dev
))
5662 /* Find the rebuild target, but punt on the dual rebuild case */
5663 for (d
= array
->devs
; d
; d
= d
->next
)
5664 if (d
->recovery_start
== 0) {
5671 /* (?) none of the disks are marked with
5672 * IMSM_ORD_REBUILD, so assume they are missing and the
5673 * disk_ord_tbl was not correctly updated
5675 dprintf("%s: failed to locate out-of-sync disk\n", __func__
);
5679 units
= __le32_to_cpu(dev
->vol
.curr_migr_unit
);
5680 rebuild
->recovery_start
= units
* blocks_per_migr_unit(super
, dev
);
5684 static int recover_backup_imsm(struct supertype
*st
, struct mdinfo
*info
);
5687 static struct mdinfo
*container_content_imsm(struct supertype
*st
, char *subarray
)
5689 /* Given a container loaded by load_super_imsm_all,
5690 * extract information about all the arrays into
5692 * If 'subarray' is given, just extract info about that array.
5694 * For each imsm_dev create an mdinfo, fill it in,
5695 * then look for matching devices in super->disks
5696 * and create appropriate device mdinfo.
5698 struct intel_super
*super
= st
->sb
;
5699 struct imsm_super
*mpb
= super
->anchor
;
5700 struct mdinfo
*rest
= NULL
;
5704 int spare_disks
= 0;
5706 /* do not assemble arrays when not all attributes are supported */
5707 if (imsm_check_attributes(mpb
->attributes
) == 0) {
5709 fprintf(stderr
, Name
": Unsupported attributes in IMSM metadata."
5710 "Arrays activation is blocked.\n");
5713 /* check for bad blocks */
5714 if (imsm_bbm_log_size(super
->anchor
)) {
5715 fprintf(stderr
, Name
": BBM log found in IMSM metadata."
5716 "Arrays activation is blocked.\n");
5721 /* count spare devices, not used in maps
5723 for (d
= super
->disks
; d
; d
= d
->next
)
5727 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
5728 struct imsm_dev
*dev
;
5729 struct imsm_map
*map
;
5730 struct imsm_map
*map2
;
5731 struct mdinfo
*this;
5736 (i
!= strtoul(subarray
, &ep
, 10) || *ep
!= '\0'))
5739 dev
= get_imsm_dev(super
, i
);
5740 map
= get_imsm_map(dev
, 0);
5741 map2
= get_imsm_map(dev
, 1);
5743 /* do not publish arrays that are in the middle of an
5744 * unsupported migration
5746 if (dev
->vol
.migr_state
&&
5747 (migr_type(dev
) == MIGR_STATE_CHANGE
)) {
5748 fprintf(stderr
, Name
": cannot assemble volume '%.16s':"
5749 " unsupported migration in progress\n",
5753 /* do not publish arrays that are not support by controller's
5757 chunk
= __le16_to_cpu(map
->blocks_per_strip
) >> 1;
5758 this = malloc(sizeof(*this));
5760 fprintf(stderr
, Name
": failed to allocate %zu bytes\n",
5765 super
->current_vol
= i
;
5766 getinfo_super_imsm_volume(st
, this, NULL
);
5769 /* mdadm does not support all metadata features- set the bit in all arrays state */
5770 if (!validate_geometry_imsm_orom(super
,
5771 get_imsm_raid_level(map
), /* RAID level */
5772 imsm_level_to_layout(get_imsm_raid_level(map
)),
5773 map
->num_members
, /* raid disks */
5776 fprintf(stderr
, Name
": IMSM RAID geometry validation"
5777 " failed. Array %s activation is blocked.\n",
5779 this->array
.state
|=
5780 (1<<MD_SB_BLOCK_CONTAINER_RESHAPE
) |
5781 (1<<MD_SB_BLOCK_VOLUME
);
5785 /* if array has bad blocks, set suitable bit in all arrays state */
5787 this->array
.state
|=
5788 (1<<MD_SB_BLOCK_CONTAINER_RESHAPE
) |
5789 (1<<MD_SB_BLOCK_VOLUME
);
5791 for (slot
= 0 ; slot
< map
->num_members
; slot
++) {
5792 unsigned long long recovery_start
;
5793 struct mdinfo
*info_d
;
5800 idx
= get_imsm_disk_idx(dev
, slot
, 0);
5801 ord
= get_imsm_ord_tbl_ent(dev
, slot
, -1);
5802 for (d
= super
->disks
; d
; d
= d
->next
)
5803 if (d
->index
== idx
)
5806 recovery_start
= MaxSector
;
5809 if (d
&& is_failed(&d
->disk
))
5811 if (ord
& IMSM_ORD_REBUILD
)
5815 * if we skip some disks the array will be assmebled degraded;
5816 * reset resync start to avoid a dirty-degraded
5817 * situation when performing the intial sync
5819 * FIXME handle dirty degraded
5821 if ((skip
|| recovery_start
== 0) && !dev
->vol
.dirty
)
5822 this->resync_start
= MaxSector
;
5826 info_d
= calloc(1, sizeof(*info_d
));
5828 fprintf(stderr
, Name
": failed to allocate disk"
5829 " for volume %.16s\n", dev
->volume
);
5830 info_d
= this->devs
;
5832 struct mdinfo
*d
= info_d
->next
;
5841 info_d
->next
= this->devs
;
5842 this->devs
= info_d
;
5844 info_d
->disk
.number
= d
->index
;
5845 info_d
->disk
.major
= d
->major
;
5846 info_d
->disk
.minor
= d
->minor
;
5847 info_d
->disk
.raid_disk
= slot
;
5848 info_d
->recovery_start
= recovery_start
;
5850 if (slot
< map2
->num_members
)
5851 info_d
->disk
.state
= (1 << MD_DISK_ACTIVE
);
5853 this->array
.spare_disks
++;
5855 if (slot
< map
->num_members
)
5856 info_d
->disk
.state
= (1 << MD_DISK_ACTIVE
);
5858 this->array
.spare_disks
++;
5860 if (info_d
->recovery_start
== MaxSector
)
5861 this->array
.working_disks
++;
5863 info_d
->events
= __le32_to_cpu(mpb
->generation_num
);
5864 info_d
->data_offset
= __le32_to_cpu(map
->pba_of_lba0
);
5865 info_d
->component_size
= __le32_to_cpu(map
->blocks_per_member
);
5867 /* now that the disk list is up-to-date fixup recovery_start */
5868 update_recovery_start(super
, dev
, this);
5869 this->array
.spare_disks
+= spare_disks
;
5872 /* check for reshape */
5873 if (this->reshape_active
== 1)
5874 recover_backup_imsm(st
, this);
5883 static __u8
imsm_check_degraded(struct intel_super
*super
, struct imsm_dev
*dev
, int failed
)
5885 struct imsm_map
*map
= get_imsm_map(dev
, 0);
5888 return map
->map_state
== IMSM_T_STATE_UNINITIALIZED
?
5889 IMSM_T_STATE_UNINITIALIZED
: IMSM_T_STATE_NORMAL
;
5891 switch (get_imsm_raid_level(map
)) {
5893 return IMSM_T_STATE_FAILED
;
5896 if (failed
< map
->num_members
)
5897 return IMSM_T_STATE_DEGRADED
;
5899 return IMSM_T_STATE_FAILED
;
5904 * check to see if any mirrors have failed, otherwise we
5905 * are degraded. Even numbered slots are mirrored on
5909 /* gcc -Os complains that this is unused */
5910 int insync
= insync
;
5912 for (i
= 0; i
< map
->num_members
; i
++) {
5913 __u32 ord
= get_imsm_ord_tbl_ent(dev
, i
, -1);
5914 int idx
= ord_to_idx(ord
);
5915 struct imsm_disk
*disk
;
5917 /* reset the potential in-sync count on even-numbered
5918 * slots. num_copies is always 2 for imsm raid10
5923 disk
= get_imsm_disk(super
, idx
);
5924 if (!disk
|| is_failed(disk
) || ord
& IMSM_ORD_REBUILD
)
5927 /* no in-sync disks left in this mirror the
5931 return IMSM_T_STATE_FAILED
;
5934 return IMSM_T_STATE_DEGRADED
;
5938 return IMSM_T_STATE_DEGRADED
;
5940 return IMSM_T_STATE_FAILED
;
5946 return map
->map_state
;
5949 static int imsm_count_failed(struct intel_super
*super
, struct imsm_dev
*dev
)
5953 struct imsm_disk
*disk
;
5954 struct imsm_map
*map
= get_imsm_map(dev
, 0);
5955 struct imsm_map
*prev
= get_imsm_map(dev
, dev
->vol
.migr_state
);
5956 struct imsm_map
*map_for_loop
;
5960 /* at the beginning of migration we set IMSM_ORD_REBUILD on
5961 * disks that are being rebuilt. New failures are recorded to
5962 * map[0]. So we look through all the disks we started with and
5963 * see if any failures are still present, or if any new ones
5966 map_for_loop
= prev
;
5967 if (is_gen_migration(dev
))
5968 if (prev
&& (map
->num_members
> prev
->num_members
))
5971 for (i
= 0; i
< map_for_loop
->num_members
; i
++) {
5973 if (i
< prev
->num_members
)
5974 ord
|= __le32_to_cpu(prev
->disk_ord_tbl
[i
]);
5975 if (i
< map
->num_members
)
5976 ord
|= __le32_to_cpu(map
->disk_ord_tbl
[i
]);
5977 idx
= ord_to_idx(ord
);
5979 disk
= get_imsm_disk(super
, idx
);
5980 if (!disk
|| is_failed(disk
) || ord
& IMSM_ORD_REBUILD
)
5988 static int imsm_open_new(struct supertype
*c
, struct active_array
*a
,
5991 struct intel_super
*super
= c
->sb
;
5992 struct imsm_super
*mpb
= super
->anchor
;
5994 if (atoi(inst
) >= mpb
->num_raid_devs
) {
5995 fprintf(stderr
, "%s: subarry index %d, out of range\n",
5996 __func__
, atoi(inst
));
6000 dprintf("imsm: open_new %s\n", inst
);
6001 a
->info
.container_member
= atoi(inst
);
6005 static int is_resyncing(struct imsm_dev
*dev
)
6007 struct imsm_map
*migr_map
;
6009 if (!dev
->vol
.migr_state
)
6012 if (migr_type(dev
) == MIGR_INIT
||
6013 migr_type(dev
) == MIGR_REPAIR
)
6016 if (migr_type(dev
) == MIGR_GEN_MIGR
)
6019 migr_map
= get_imsm_map(dev
, 1);
6021 if ((migr_map
->map_state
== IMSM_T_STATE_NORMAL
) &&
6022 (dev
->vol
.migr_type
!= MIGR_GEN_MIGR
))
6028 /* return true if we recorded new information */
6029 static int mark_failure(struct imsm_dev
*dev
, struct imsm_disk
*disk
, int idx
)
6033 struct imsm_map
*map
;
6034 char buf
[MAX_RAID_SERIAL_LEN
+3];
6035 unsigned int len
, shift
= 0;
6037 /* new failures are always set in map[0] */
6038 map
= get_imsm_map(dev
, 0);
6040 slot
= get_imsm_disk_slot(map
, idx
);
6044 ord
= __le32_to_cpu(map
->disk_ord_tbl
[slot
]);
6045 if (is_failed(disk
) && (ord
& IMSM_ORD_REBUILD
))
6048 memcpy(buf
, disk
->serial
, MAX_RAID_SERIAL_LEN
);
6049 buf
[MAX_RAID_SERIAL_LEN
] = '\000';
6051 if ((len
= strlen(buf
)) >= MAX_RAID_SERIAL_LEN
)
6052 shift
= len
- MAX_RAID_SERIAL_LEN
+ 1;
6053 strncpy((char *)disk
->serial
, &buf
[shift
], MAX_RAID_SERIAL_LEN
);
6055 disk
->status
|= FAILED_DISK
;
6056 set_imsm_ord_tbl_ent(map
, slot
, idx
| IMSM_ORD_REBUILD
);
6057 if (map
->failed_disk_num
== 0xff)
6058 map
->failed_disk_num
= slot
;
6062 static void mark_missing(struct imsm_dev
*dev
, struct imsm_disk
*disk
, int idx
)
6064 mark_failure(dev
, disk
, idx
);
6066 if (disk
->scsi_id
== __cpu_to_le32(~(__u32
)0))
6069 disk
->scsi_id
= __cpu_to_le32(~(__u32
)0);
6070 memmove(&disk
->serial
[0], &disk
->serial
[1], MAX_RAID_SERIAL_LEN
- 1);
6073 static void handle_missing(struct intel_super
*super
, struct imsm_dev
*dev
)
6079 if (!super
->missing
)
6081 failed
= imsm_count_failed(super
, dev
);
6082 map_state
= imsm_check_degraded(super
, dev
, failed
);
6084 dprintf("imsm: mark missing\n");
6085 end_migration(dev
, map_state
);
6086 for (dl
= super
->missing
; dl
; dl
= dl
->next
)
6087 mark_missing(dev
, &dl
->disk
, dl
->index
);
6088 super
->updates_pending
++;
6091 static unsigned long long imsm_set_array_size(struct imsm_dev
*dev
)
6093 int used_disks
= imsm_num_data_members(dev
, 0);
6094 unsigned long long array_blocks
;
6095 struct imsm_map
*map
;
6097 if (used_disks
== 0) {
6098 /* when problems occures
6099 * return current array_blocks value
6101 array_blocks
= __le32_to_cpu(dev
->size_high
);
6102 array_blocks
= array_blocks
<< 32;
6103 array_blocks
+= __le32_to_cpu(dev
->size_low
);
6105 return array_blocks
;
6108 /* set array size in metadata
6110 map
= get_imsm_map(dev
, 0);
6111 array_blocks
= map
->blocks_per_member
* used_disks
;
6113 /* round array size down to closest MB
6115 array_blocks
= (array_blocks
>> SECT_PER_MB_SHIFT
) << SECT_PER_MB_SHIFT
;
6116 dev
->size_low
= __cpu_to_le32((__u32
)array_blocks
);
6117 dev
->size_high
= __cpu_to_le32((__u32
)(array_blocks
>> 32));
6119 return array_blocks
;
6122 static void imsm_set_disk(struct active_array
*a
, int n
, int state
);
6124 static void imsm_progress_container_reshape(struct intel_super
*super
)
6126 /* if no device has a migr_state, but some device has a
6127 * different number of members than the previous device, start
6128 * changing the number of devices in this device to match
6131 struct imsm_super
*mpb
= super
->anchor
;
6132 int prev_disks
= -1;
6136 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
6137 struct imsm_dev
*dev
= get_imsm_dev(super
, i
);
6138 struct imsm_map
*map
= get_imsm_map(dev
, 0);
6139 struct imsm_map
*map2
;
6140 int prev_num_members
;
6142 if (dev
->vol
.migr_state
)
6145 if (prev_disks
== -1)
6146 prev_disks
= map
->num_members
;
6147 if (prev_disks
== map
->num_members
)
6150 /* OK, this array needs to enter reshape mode.
6151 * i.e it needs a migr_state
6154 copy_map_size
= sizeof_imsm_map(map
);
6155 prev_num_members
= map
->num_members
;
6156 map
->num_members
= prev_disks
;
6157 dev
->vol
.migr_state
= 1;
6158 dev
->vol
.curr_migr_unit
= 0;
6159 set_migr_type(dev
, MIGR_GEN_MIGR
);
6160 for (i
= prev_num_members
;
6161 i
< map
->num_members
; i
++)
6162 set_imsm_ord_tbl_ent(map
, i
, i
);
6163 map2
= get_imsm_map(dev
, 1);
6164 /* Copy the current map */
6165 memcpy(map2
, map
, copy_map_size
);
6166 map2
->num_members
= prev_num_members
;
6168 imsm_set_array_size(dev
);
6169 super
->updates_pending
++;
6173 /* Handle dirty -> clean transititions, resync and reshape. Degraded and rebuild
6174 * states are handled in imsm_set_disk() with one exception, when a
6175 * resync is stopped due to a new failure this routine will set the
6176 * 'degraded' state for the array.
6178 static int imsm_set_array_state(struct active_array
*a
, int consistent
)
6180 int inst
= a
->info
.container_member
;
6181 struct intel_super
*super
= a
->container
->sb
;
6182 struct imsm_dev
*dev
= get_imsm_dev(super
, inst
);
6183 struct imsm_map
*map
= get_imsm_map(dev
, 0);
6184 int failed
= imsm_count_failed(super
, dev
);
6185 __u8 map_state
= imsm_check_degraded(super
, dev
, failed
);
6186 __u32 blocks_per_unit
;
6188 if (dev
->vol
.migr_state
&&
6189 dev
->vol
.migr_type
== MIGR_GEN_MIGR
) {
6190 /* array state change is blocked due to reshape action
6192 * - abort the reshape (if last_checkpoint is 0 and action!= reshape)
6193 * - finish the reshape (if last_checkpoint is big and action != reshape)
6194 * - update curr_migr_unit
6196 if (a
->curr_action
== reshape
) {
6197 /* still reshaping, maybe update curr_migr_unit */
6198 goto mark_checkpoint
;
6200 if (a
->last_checkpoint
== 0 && a
->prev_action
== reshape
) {
6201 /* for some reason we aborted the reshape.
6203 * disable automatic metadata rollback
6204 * user action is required to recover process
6207 struct imsm_map
*map2
= get_imsm_map(dev
, 1);
6208 dev
->vol
.migr_state
= 0;
6209 set_migr_type(dev
, 0);
6210 dev
->vol
.curr_migr_unit
= 0;
6211 memcpy(map
, map2
, sizeof_imsm_map(map2
));
6212 super
->updates_pending
++;
6215 if (a
->last_checkpoint
>= a
->info
.component_size
) {
6216 unsigned long long array_blocks
;
6220 used_disks
= imsm_num_data_members(dev
, 0);
6221 if (used_disks
> 0) {
6223 map
->blocks_per_member
*
6225 /* round array size down to closest MB
6227 array_blocks
= (array_blocks
6228 >> SECT_PER_MB_SHIFT
)
6229 << SECT_PER_MB_SHIFT
;
6230 a
->info
.custom_array_size
= array_blocks
;
6231 /* encourage manager to update array
6235 a
->check_reshape
= 1;
6237 /* finalize online capacity expansion/reshape */
6238 for (mdi
= a
->info
.devs
; mdi
; mdi
= mdi
->next
)
6240 mdi
->disk
.raid_disk
,
6243 imsm_progress_container_reshape(super
);
6248 /* before we activate this array handle any missing disks */
6249 if (consistent
== 2)
6250 handle_missing(super
, dev
);
6252 if (consistent
== 2 &&
6253 (!is_resync_complete(&a
->info
) ||
6254 map_state
!= IMSM_T_STATE_NORMAL
||
6255 dev
->vol
.migr_state
))
6258 if (is_resync_complete(&a
->info
)) {
6259 /* complete intialization / resync,
6260 * recovery and interrupted recovery is completed in
6263 if (is_resyncing(dev
)) {
6264 dprintf("imsm: mark resync done\n");
6265 end_migration(dev
, map_state
);
6266 super
->updates_pending
++;
6267 a
->last_checkpoint
= 0;
6269 } else if ((!is_resyncing(dev
) && !failed
) &&
6270 (imsm_reshape_blocks_arrays_changes(super
) == 0)) {
6271 /* mark the start of the init process if nothing is failed */
6272 dprintf("imsm: mark resync start\n");
6273 if (map
->map_state
== IMSM_T_STATE_UNINITIALIZED
)
6274 migrate(dev
, super
, IMSM_T_STATE_NORMAL
, MIGR_INIT
);
6276 migrate(dev
, super
, IMSM_T_STATE_NORMAL
, MIGR_REPAIR
);
6277 super
->updates_pending
++;
6281 /* skip checkpointing for general migration,
6282 * it is controlled in mdadm
6284 if (is_gen_migration(dev
))
6285 goto skip_mark_checkpoint
;
6287 /* check if we can update curr_migr_unit from resync_start, recovery_start */
6288 blocks_per_unit
= blocks_per_migr_unit(super
, dev
);
6289 if (blocks_per_unit
) {
6293 units
= a
->last_checkpoint
/ blocks_per_unit
;
6296 /* check that we did not overflow 32-bits, and that
6297 * curr_migr_unit needs updating
6299 if (units32
== units
&&
6301 __le32_to_cpu(dev
->vol
.curr_migr_unit
) != units32
) {
6302 dprintf("imsm: mark checkpoint (%u)\n", units32
);
6303 dev
->vol
.curr_migr_unit
= __cpu_to_le32(units32
);
6304 super
->updates_pending
++;
6308 skip_mark_checkpoint
:
6309 /* mark dirty / clean */
6310 if (dev
->vol
.dirty
!= !consistent
) {
6311 dprintf("imsm: mark '%s'\n", consistent
? "clean" : "dirty");
6316 super
->updates_pending
++;
6322 static void imsm_set_disk(struct active_array
*a
, int n
, int state
)
6324 int inst
= a
->info
.container_member
;
6325 struct intel_super
*super
= a
->container
->sb
;
6326 struct imsm_dev
*dev
= get_imsm_dev(super
, inst
);
6327 struct imsm_map
*map
= get_imsm_map(dev
, 0);
6328 struct imsm_disk
*disk
;
6333 if (n
> map
->num_members
)
6334 fprintf(stderr
, "imsm: set_disk %d out of range 0..%d\n",
6335 n
, map
->num_members
- 1);
6340 dprintf("imsm: set_disk %d:%x\n", n
, state
);
6342 ord
= get_imsm_ord_tbl_ent(dev
, n
, -1);
6343 disk
= get_imsm_disk(super
, ord_to_idx(ord
));
6345 /* check for new failures */
6346 if (state
& DS_FAULTY
) {
6347 if (mark_failure(dev
, disk
, ord_to_idx(ord
)))
6348 super
->updates_pending
++;
6351 /* check if in_sync */
6352 if (state
& DS_INSYNC
&& ord
& IMSM_ORD_REBUILD
&& is_rebuilding(dev
)) {
6353 struct imsm_map
*migr_map
= get_imsm_map(dev
, 1);
6355 set_imsm_ord_tbl_ent(migr_map
, n
, ord_to_idx(ord
));
6356 super
->updates_pending
++;
6359 failed
= imsm_count_failed(super
, dev
);
6360 map_state
= imsm_check_degraded(super
, dev
, failed
);
6362 /* check if recovery complete, newly degraded, or failed */
6363 if (map_state
== IMSM_T_STATE_NORMAL
&& is_rebuilding(dev
)) {
6364 end_migration(dev
, map_state
);
6365 map
= get_imsm_map(dev
, 0);
6366 map
->failed_disk_num
= ~0;
6367 super
->updates_pending
++;
6368 a
->last_checkpoint
= 0;
6369 } else if (map_state
== IMSM_T_STATE_DEGRADED
&&
6370 map
->map_state
!= map_state
&&
6371 !dev
->vol
.migr_state
) {
6372 dprintf("imsm: mark degraded\n");
6373 map
->map_state
= map_state
;
6374 super
->updates_pending
++;
6375 a
->last_checkpoint
= 0;
6376 } else if (map_state
== IMSM_T_STATE_FAILED
&&
6377 map
->map_state
!= map_state
) {
6378 dprintf("imsm: mark failed\n");
6379 end_migration(dev
, map_state
);
6380 super
->updates_pending
++;
6381 a
->last_checkpoint
= 0;
6382 } else if (is_gen_migration(dev
)) {
6383 dprintf("imsm: Detected General Migration in state: ");
6384 if (map_state
== IMSM_T_STATE_NORMAL
) {
6385 end_migration(dev
, map_state
);
6386 map
= get_imsm_map(dev
, 0);
6387 map
->failed_disk_num
= ~0;
6388 dprintf("normal\n");
6390 if (map_state
== IMSM_T_STATE_DEGRADED
) {
6391 printf("degraded\n");
6392 end_migration(dev
, map_state
);
6394 dprintf("failed\n");
6396 map
->map_state
= map_state
;
6398 super
->updates_pending
++;
6402 static int store_imsm_mpb(int fd
, struct imsm_super
*mpb
)
6405 __u32 mpb_size
= __le32_to_cpu(mpb
->mpb_size
);
6406 unsigned long long dsize
;
6407 unsigned long long sectors
;
6409 get_dev_size(fd
, NULL
, &dsize
);
6411 if (mpb_size
> 512) {
6412 /* -1 to account for anchor */
6413 sectors
= mpb_sectors(mpb
) - 1;
6415 /* write the extended mpb to the sectors preceeding the anchor */
6416 if (lseek64(fd
, dsize
- (512 * (2 + sectors
)), SEEK_SET
) < 0)
6419 if ((unsigned long long)write(fd
, buf
+ 512, 512 * sectors
)
6424 /* first block is stored on second to last sector of the disk */
6425 if (lseek64(fd
, dsize
- (512 * 2), SEEK_SET
) < 0)
6428 if (write(fd
, buf
, 512) != 512)
6434 static void imsm_sync_metadata(struct supertype
*container
)
6436 struct intel_super
*super
= container
->sb
;
6438 dprintf("sync metadata: %d\n", super
->updates_pending
);
6439 if (!super
->updates_pending
)
6442 write_super_imsm(container
, 0);
6444 super
->updates_pending
= 0;
6447 static struct dl
*imsm_readd(struct intel_super
*super
, int idx
, struct active_array
*a
)
6449 struct imsm_dev
*dev
= get_imsm_dev(super
, a
->info
.container_member
);
6450 int i
= get_imsm_disk_idx(dev
, idx
, -1);
6453 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
6457 if (dl
&& is_failed(&dl
->disk
))
6461 dprintf("%s: found %x:%x\n", __func__
, dl
->major
, dl
->minor
);
6466 static struct dl
*imsm_add_spare(struct intel_super
*super
, int slot
,
6467 struct active_array
*a
, int activate_new
,
6468 struct mdinfo
*additional_test_list
)
6470 struct imsm_dev
*dev
= get_imsm_dev(super
, a
->info
.container_member
);
6471 int idx
= get_imsm_disk_idx(dev
, slot
, -1);
6472 struct imsm_super
*mpb
= super
->anchor
;
6473 struct imsm_map
*map
;
6474 unsigned long long pos
;
6479 __u32 array_start
= 0;
6480 __u32 array_end
= 0;
6482 struct mdinfo
*test_list
;
6484 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
6485 /* If in this array, skip */
6486 for (d
= a
->info
.devs
; d
; d
= d
->next
)
6487 if (d
->state_fd
>= 0 &&
6488 d
->disk
.major
== dl
->major
&&
6489 d
->disk
.minor
== dl
->minor
) {
6490 dprintf("%x:%x already in array\n",
6491 dl
->major
, dl
->minor
);
6496 test_list
= additional_test_list
;
6498 if (test_list
->disk
.major
== dl
->major
&&
6499 test_list
->disk
.minor
== dl
->minor
) {
6500 dprintf("%x:%x already in additional test list\n",
6501 dl
->major
, dl
->minor
);
6504 test_list
= test_list
->next
;
6509 /* skip in use or failed drives */
6510 if (is_failed(&dl
->disk
) || idx
== dl
->index
||
6512 dprintf("%x:%x status (failed: %d index: %d)\n",
6513 dl
->major
, dl
->minor
, is_failed(&dl
->disk
), idx
);
6517 /* skip pure spares when we are looking for partially
6518 * assimilated drives
6520 if (dl
->index
== -1 && !activate_new
)
6523 /* Does this unused device have the requisite free space?
6524 * It needs to be able to cover all member volumes
6526 ex
= get_extents(super
, dl
);
6528 dprintf("cannot get extents\n");
6531 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
6532 dev
= get_imsm_dev(super
, i
);
6533 map
= get_imsm_map(dev
, 0);
6535 /* check if this disk is already a member of
6538 if (get_imsm_disk_slot(map
, dl
->index
) >= 0)
6544 array_start
= __le32_to_cpu(map
->pba_of_lba0
);
6545 array_end
= array_start
+
6546 __le32_to_cpu(map
->blocks_per_member
) - 1;
6549 /* check that we can start at pba_of_lba0 with
6550 * blocks_per_member of space
6552 if (array_start
>= pos
&& array_end
< ex
[j
].start
) {
6556 pos
= ex
[j
].start
+ ex
[j
].size
;
6558 } while (ex
[j
-1].size
);
6565 if (i
< mpb
->num_raid_devs
) {
6566 dprintf("%x:%x does not have %u to %u available\n",
6567 dl
->major
, dl
->minor
, array_start
, array_end
);
6578 static int imsm_rebuild_allowed(struct supertype
*cont
, int dev_idx
, int failed
)
6580 struct imsm_dev
*dev2
;
6581 struct imsm_map
*map
;
6587 dev2
= get_imsm_dev(cont
->sb
, dev_idx
);
6589 state
= imsm_check_degraded(cont
->sb
, dev2
, failed
);
6590 if (state
== IMSM_T_STATE_FAILED
) {
6591 map
= get_imsm_map(dev2
, 0);
6594 for (slot
= 0; slot
< map
->num_members
; slot
++) {
6596 * Check if failed disks are deleted from intel
6597 * disk list or are marked to be deleted
6599 idx
= get_imsm_disk_idx(dev2
, slot
, -1);
6600 idisk
= get_imsm_dl_disk(cont
->sb
, idx
);
6602 * Do not rebuild the array if failed disks
6603 * from failed sub-array are not removed from
6607 is_failed(&idisk
->disk
) &&
6608 (idisk
->action
!= DISK_REMOVE
))
6616 static struct mdinfo
*imsm_activate_spare(struct active_array
*a
,
6617 struct metadata_update
**updates
)
6620 * Find a device with unused free space and use it to replace a
6621 * failed/vacant region in an array. We replace failed regions one a
6622 * array at a time. The result is that a new spare disk will be added
6623 * to the first failed array and after the monitor has finished
6624 * propagating failures the remainder will be consumed.
6626 * FIXME add a capability for mdmon to request spares from another
6630 struct intel_super
*super
= a
->container
->sb
;
6631 int inst
= a
->info
.container_member
;
6632 struct imsm_dev
*dev
= get_imsm_dev(super
, inst
);
6633 struct imsm_map
*map
= get_imsm_map(dev
, 0);
6634 int failed
= a
->info
.array
.raid_disks
;
6635 struct mdinfo
*rv
= NULL
;
6638 struct metadata_update
*mu
;
6640 struct imsm_update_activate_spare
*u
;
6645 for (d
= a
->info
.devs
; d
; d
= d
->next
) {
6646 if ((d
->curr_state
& DS_FAULTY
) &&
6648 /* wait for Removal to happen */
6650 if (d
->state_fd
>= 0)
6654 dprintf("imsm: activate spare: inst=%d failed=%d (%d) level=%d\n",
6655 inst
, failed
, a
->info
.array
.raid_disks
, a
->info
.array
.level
);
6657 if (imsm_reshape_blocks_arrays_changes(super
))
6660 if (a
->info
.array
.level
== 4)
6661 /* No repair for takeovered array
6662 * imsm doesn't support raid4
6666 if (imsm_check_degraded(super
, dev
, failed
) != IMSM_T_STATE_DEGRADED
)
6670 * If there are any failed disks check state of the other volume.
6671 * Block rebuild if the another one is failed until failed disks
6672 * are removed from container.
6675 dprintf("found failed disks in %.*s, check if there another"
6676 "failed sub-array.\n",
6677 MAX_RAID_SERIAL_LEN
, dev
->volume
);
6678 /* check if states of the other volumes allow for rebuild */
6679 for (i
= 0; i
< super
->anchor
->num_raid_devs
; i
++) {
6681 allowed
= imsm_rebuild_allowed(a
->container
,
6689 /* For each slot, if it is not working, find a spare */
6690 for (i
= 0; i
< a
->info
.array
.raid_disks
; i
++) {
6691 for (d
= a
->info
.devs
; d
; d
= d
->next
)
6692 if (d
->disk
.raid_disk
== i
)
6694 dprintf("found %d: %p %x\n", i
, d
, d
?d
->curr_state
:0);
6695 if (d
&& (d
->state_fd
>= 0))
6699 * OK, this device needs recovery. Try to re-add the
6700 * previous occupant of this slot, if this fails see if
6701 * we can continue the assimilation of a spare that was
6702 * partially assimilated, finally try to activate a new
6705 dl
= imsm_readd(super
, i
, a
);
6707 dl
= imsm_add_spare(super
, i
, a
, 0, rv
);
6709 dl
= imsm_add_spare(super
, i
, a
, 1, rv
);
6713 /* found a usable disk with enough space */
6714 di
= malloc(sizeof(*di
));
6717 memset(di
, 0, sizeof(*di
));
6719 /* dl->index will be -1 in the case we are activating a
6720 * pristine spare. imsm_process_update() will create a
6721 * new index in this case. Once a disk is found to be
6722 * failed in all member arrays it is kicked from the
6725 di
->disk
.number
= dl
->index
;
6727 /* (ab)use di->devs to store a pointer to the device
6730 di
->devs
= (struct mdinfo
*) dl
;
6732 di
->disk
.raid_disk
= i
;
6733 di
->disk
.major
= dl
->major
;
6734 di
->disk
.minor
= dl
->minor
;
6736 di
->recovery_start
= 0;
6737 di
->data_offset
= __le32_to_cpu(map
->pba_of_lba0
);
6738 di
->component_size
= a
->info
.component_size
;
6739 di
->container_member
= inst
;
6740 super
->random
= random32();
6744 dprintf("%x:%x to be %d at %llu\n", dl
->major
, dl
->minor
,
6745 i
, di
->data_offset
);
6749 /* No spares found */
6751 /* Now 'rv' has a list of devices to return.
6752 * Create a metadata_update record to update the
6753 * disk_ord_tbl for the array
6755 mu
= malloc(sizeof(*mu
));
6757 mu
->buf
= malloc(sizeof(struct imsm_update_activate_spare
) * num_spares
);
6758 if (mu
->buf
== NULL
) {
6765 struct mdinfo
*n
= rv
->next
;
6774 mu
->space_list
= NULL
;
6775 mu
->len
= sizeof(struct imsm_update_activate_spare
) * num_spares
;
6776 mu
->next
= *updates
;
6777 u
= (struct imsm_update_activate_spare
*) mu
->buf
;
6779 for (di
= rv
; di
; di
= di
->next
) {
6780 u
->type
= update_activate_spare
;
6781 u
->dl
= (struct dl
*) di
->devs
;
6783 u
->slot
= di
->disk
.raid_disk
;
6794 static int disks_overlap(struct intel_super
*super
, int idx
, struct imsm_update_create_array
*u
)
6796 struct imsm_dev
*dev
= get_imsm_dev(super
, idx
);
6797 struct imsm_map
*map
= get_imsm_map(dev
, 0);
6798 struct imsm_map
*new_map
= get_imsm_map(&u
->dev
, 0);
6799 struct disk_info
*inf
= get_disk_info(u
);
6800 struct imsm_disk
*disk
;
6804 for (i
= 0; i
< map
->num_members
; i
++) {
6805 disk
= get_imsm_disk(super
, get_imsm_disk_idx(dev
, i
, -1));
6806 for (j
= 0; j
< new_map
->num_members
; j
++)
6807 if (serialcmp(disk
->serial
, inf
[j
].serial
) == 0)
6815 static struct dl
*get_disk_super(struct intel_super
*super
, int major
, int minor
)
6817 struct dl
*dl
= NULL
;
6818 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
6819 if ((dl
->major
== major
) && (dl
->minor
== minor
))
6824 static int remove_disk_super(struct intel_super
*super
, int major
, int minor
)
6826 struct dl
*prev
= NULL
;
6830 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
6831 if ((dl
->major
== major
) && (dl
->minor
== minor
)) {
6834 prev
->next
= dl
->next
;
6836 super
->disks
= dl
->next
;
6838 __free_imsm_disk(dl
);
6839 dprintf("%s: removed %x:%x\n",
6840 __func__
, major
, minor
);
6848 static void imsm_delete(struct intel_super
*super
, struct dl
**dlp
, unsigned index
);
6850 static int add_remove_disk_update(struct intel_super
*super
)
6852 int check_degraded
= 0;
6853 struct dl
*disk
= NULL
;
6854 /* add/remove some spares to/from the metadata/contrainer */
6855 while (super
->disk_mgmt_list
) {
6856 struct dl
*disk_cfg
;
6858 disk_cfg
= super
->disk_mgmt_list
;
6859 super
->disk_mgmt_list
= disk_cfg
->next
;
6860 disk_cfg
->next
= NULL
;
6862 if (disk_cfg
->action
== DISK_ADD
) {
6863 disk_cfg
->next
= super
->disks
;
6864 super
->disks
= disk_cfg
;
6866 dprintf("%s: added %x:%x\n",
6867 __func__
, disk_cfg
->major
,
6869 } else if (disk_cfg
->action
== DISK_REMOVE
) {
6870 dprintf("Disk remove action processed: %x.%x\n",
6871 disk_cfg
->major
, disk_cfg
->minor
);
6872 disk
= get_disk_super(super
,
6876 /* store action status */
6877 disk
->action
= DISK_REMOVE
;
6878 /* remove spare disks only */
6879 if (disk
->index
== -1) {
6880 remove_disk_super(super
,
6885 /* release allocate disk structure */
6886 __free_imsm_disk(disk_cfg
);
6889 return check_degraded
;
6893 static int apply_reshape_migration_update(struct imsm_update_reshape_migration
*u
,
6894 struct intel_super
*super
,
6897 struct intel_dev
*id
;
6898 void **tofree
= NULL
;
6901 dprintf("apply_reshape_migration_update()\n");
6902 if ((u
->subdev
< 0) ||
6904 dprintf("imsm: Error: Wrong subdev: %i\n", u
->subdev
);
6907 if ((space_list
== NULL
) || (*space_list
== NULL
)) {
6908 dprintf("imsm: Error: Memory is not allocated\n");
6912 for (id
= super
->devlist
; id
; id
= id
->next
) {
6913 if (id
->index
== (unsigned)u
->subdev
) {
6914 struct imsm_dev
*dev
= get_imsm_dev(super
, u
->subdev
);
6915 struct imsm_map
*map
;
6916 struct imsm_dev
*new_dev
=
6917 (struct imsm_dev
*)*space_list
;
6918 struct imsm_map
*migr_map
= get_imsm_map(dev
, 1);
6920 struct dl
*new_disk
;
6922 if (new_dev
== NULL
)
6924 *space_list
= **space_list
;
6925 memcpy(new_dev
, dev
, sizeof_imsm_dev(dev
, 0));
6926 map
= get_imsm_map(new_dev
, 0);
6928 dprintf("imsm: Error: migration in progress");
6932 to_state
= map
->map_state
;
6933 if ((u
->new_level
== 5) && (map
->raid_level
== 0)) {
6935 /* this should not happen */
6936 if (u
->new_disks
[0] < 0) {
6937 map
->failed_disk_num
=
6938 map
->num_members
- 1;
6939 to_state
= IMSM_T_STATE_DEGRADED
;
6941 to_state
= IMSM_T_STATE_NORMAL
;
6943 migrate(new_dev
, super
, to_state
, MIGR_GEN_MIGR
);
6944 if (u
->new_level
> -1)
6945 map
->raid_level
= u
->new_level
;
6946 migr_map
= get_imsm_map(new_dev
, 1);
6947 if ((u
->new_level
== 5) &&
6948 (migr_map
->raid_level
== 0)) {
6949 int ord
= map
->num_members
- 1;
6950 migr_map
->num_members
--;
6951 if (u
->new_disks
[0] < 0)
6952 ord
|= IMSM_ORD_REBUILD
;
6953 set_imsm_ord_tbl_ent(map
,
6954 map
->num_members
- 1,
6958 tofree
= (void **)dev
;
6960 /* update chunk size
6962 if (u
->new_chunksize
> 0)
6963 map
->blocks_per_strip
=
6964 __cpu_to_le16(u
->new_chunksize
* 2);
6968 if ((u
->new_level
!= 5) ||
6969 (migr_map
->raid_level
!= 0) ||
6970 (migr_map
->raid_level
== map
->raid_level
))
6973 if (u
->new_disks
[0] >= 0) {
6976 new_disk
= get_disk_super(super
,
6977 major(u
->new_disks
[0]),
6978 minor(u
->new_disks
[0]));
6979 dprintf("imsm: new disk for reshape is: %i:%i "
6980 "(%p, index = %i)\n",
6981 major(u
->new_disks
[0]),
6982 minor(u
->new_disks
[0]),
6983 new_disk
, new_disk
->index
);
6984 if (new_disk
== NULL
)
6985 goto error_disk_add
;
6987 new_disk
->index
= map
->num_members
- 1;
6988 /* slot to fill in autolayout
6990 new_disk
->raiddisk
= new_disk
->index
;
6991 new_disk
->disk
.status
|= CONFIGURED_DISK
;
6992 new_disk
->disk
.status
&= ~SPARE_DISK
;
6994 goto error_disk_add
;
6997 *tofree
= *space_list
;
6998 /* calculate new size
7000 imsm_set_array_size(new_dev
);
7007 *space_list
= tofree
;
7011 dprintf("Error: imsm: Cannot find disk.\n");
7015 static int apply_update_activate_spare(struct imsm_update_activate_spare
*u
,
7016 struct intel_super
*super
,
7017 struct active_array
*active_array
)
7019 struct imsm_super
*mpb
= super
->anchor
;
7020 struct imsm_dev
*dev
= get_imsm_dev(super
, u
->array
);
7021 struct imsm_map
*map
= get_imsm_map(dev
, 0);
7022 struct imsm_map
*migr_map
;
7023 struct active_array
*a
;
7024 struct imsm_disk
*disk
;
7031 int second_map_created
= 0;
7033 for (; u
; u
= u
->next
) {
7034 victim
= get_imsm_disk_idx(dev
, u
->slot
, -1);
7039 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
7044 fprintf(stderr
, "error: imsm_activate_spare passed "
7045 "an unknown disk (index: %d)\n",
7050 /* count failures (excluding rebuilds and the victim)
7051 * to determine map[0] state
7054 for (i
= 0; i
< map
->num_members
; i
++) {
7057 disk
= get_imsm_disk(super
,
7058 get_imsm_disk_idx(dev
, i
, -1));
7059 if (!disk
|| is_failed(disk
))
7063 /* adding a pristine spare, assign a new index */
7064 if (dl
->index
< 0) {
7065 dl
->index
= super
->anchor
->num_disks
;
7066 super
->anchor
->num_disks
++;
7069 disk
->status
|= CONFIGURED_DISK
;
7070 disk
->status
&= ~SPARE_DISK
;
7073 to_state
= imsm_check_degraded(super
, dev
, failed
);
7074 if (!second_map_created
) {
7075 second_map_created
= 1;
7076 map
->map_state
= IMSM_T_STATE_DEGRADED
;
7077 migrate(dev
, super
, to_state
, MIGR_REBUILD
);
7079 map
->map_state
= to_state
;
7080 migr_map
= get_imsm_map(dev
, 1);
7081 set_imsm_ord_tbl_ent(map
, u
->slot
, dl
->index
);
7082 set_imsm_ord_tbl_ent(migr_map
, u
->slot
,
7083 dl
->index
| IMSM_ORD_REBUILD
);
7085 /* update the family_num to mark a new container
7086 * generation, being careful to record the existing
7087 * family_num in orig_family_num to clean up after
7088 * earlier mdadm versions that neglected to set it.
7090 if (mpb
->orig_family_num
== 0)
7091 mpb
->orig_family_num
= mpb
->family_num
;
7092 mpb
->family_num
+= super
->random
;
7094 /* count arrays using the victim in the metadata */
7096 for (a
= active_array
; a
; a
= a
->next
) {
7097 dev
= get_imsm_dev(super
, a
->info
.container_member
);
7098 map
= get_imsm_map(dev
, 0);
7100 if (get_imsm_disk_slot(map
, victim
) >= 0)
7104 /* delete the victim if it is no longer being
7110 /* We know that 'manager' isn't touching anything,
7111 * so it is safe to delete
7113 for (dlp
= &super
->disks
; *dlp
; dlp
= &(*dlp
)->next
)
7114 if ((*dlp
)->index
== victim
)
7117 /* victim may be on the missing list */
7119 for (dlp
= &super
->missing
; *dlp
;
7120 dlp
= &(*dlp
)->next
)
7121 if ((*dlp
)->index
== victim
)
7123 imsm_delete(super
, dlp
, victim
);
7130 static int apply_reshape_container_disks_update(struct imsm_update_reshape
*u
,
7131 struct intel_super
*super
,
7134 struct dl
*new_disk
;
7135 struct intel_dev
*id
;
7137 int delta_disks
= u
->new_raid_disks
- u
->old_raid_disks
;
7138 int disk_count
= u
->old_raid_disks
;
7139 void **tofree
= NULL
;
7140 int devices_to_reshape
= 1;
7141 struct imsm_super
*mpb
= super
->anchor
;
7143 unsigned int dev_id
;
7145 dprintf("imsm: apply_reshape_container_disks_update()\n");
7147 /* enable spares to use in array */
7148 for (i
= 0; i
< delta_disks
; i
++) {
7149 new_disk
= get_disk_super(super
,
7150 major(u
->new_disks
[i
]),
7151 minor(u
->new_disks
[i
]));
7152 dprintf("imsm: new disk for reshape is: %i:%i "
7153 "(%p, index = %i)\n",
7154 major(u
->new_disks
[i
]), minor(u
->new_disks
[i
]),
7155 new_disk
, new_disk
->index
);
7156 if ((new_disk
== NULL
) ||
7157 ((new_disk
->index
>= 0) &&
7158 (new_disk
->index
< u
->old_raid_disks
)))
7159 goto update_reshape_exit
;
7160 new_disk
->index
= disk_count
++;
7161 /* slot to fill in autolayout
7163 new_disk
->raiddisk
= new_disk
->index
;
7164 new_disk
->disk
.status
|=
7166 new_disk
->disk
.status
&= ~SPARE_DISK
;
7169 dprintf("imsm: volume set mpb->num_raid_devs = %i\n",
7170 mpb
->num_raid_devs
);
7171 /* manage changes in volume
7173 for (dev_id
= 0; dev_id
< mpb
->num_raid_devs
; dev_id
++) {
7174 void **sp
= *space_list
;
7175 struct imsm_dev
*newdev
;
7176 struct imsm_map
*newmap
, *oldmap
;
7178 for (id
= super
->devlist
; id
; id
= id
->next
) {
7179 if (id
->index
== dev_id
)
7188 /* Copy the dev, but not (all of) the map */
7189 memcpy(newdev
, id
->dev
, sizeof(*newdev
));
7190 oldmap
= get_imsm_map(id
->dev
, 0);
7191 newmap
= get_imsm_map(newdev
, 0);
7192 /* Copy the current map */
7193 memcpy(newmap
, oldmap
, sizeof_imsm_map(oldmap
));
7194 /* update one device only
7196 if (devices_to_reshape
) {
7197 dprintf("imsm: modifying subdev: %i\n",
7199 devices_to_reshape
--;
7200 newdev
->vol
.migr_state
= 1;
7201 newdev
->vol
.curr_migr_unit
= 0;
7202 set_migr_type(newdev
, MIGR_GEN_MIGR
);
7203 newmap
->num_members
= u
->new_raid_disks
;
7204 for (i
= 0; i
< delta_disks
; i
++) {
7205 set_imsm_ord_tbl_ent(newmap
,
7206 u
->old_raid_disks
+ i
,
7207 u
->old_raid_disks
+ i
);
7209 /* New map is correct, now need to save old map
7211 newmap
= get_imsm_map(newdev
, 1);
7212 memcpy(newmap
, oldmap
, sizeof_imsm_map(oldmap
));
7214 imsm_set_array_size(newdev
);
7217 sp
= (void **)id
->dev
;
7222 /* Clear migration record */
7223 memset(super
->migr_rec
, 0, sizeof(struct migr_record
));
7226 *space_list
= tofree
;
7229 update_reshape_exit
:
7234 static int apply_takeover_update(struct imsm_update_takeover
*u
,
7235 struct intel_super
*super
,
7238 struct imsm_dev
*dev
= NULL
;
7239 struct intel_dev
*dv
;
7240 struct imsm_dev
*dev_new
;
7241 struct imsm_map
*map
;
7245 for (dv
= super
->devlist
; dv
; dv
= dv
->next
)
7246 if (dv
->index
== (unsigned int)u
->subarray
) {
7254 map
= get_imsm_map(dev
, 0);
7256 if (u
->direction
== R10_TO_R0
) {
7257 /* Number of failed disks must be half of initial disk number */
7258 if (imsm_count_failed(super
, dev
) != (map
->num_members
/ 2))
7261 /* iterate through devices to mark removed disks as spare */
7262 for (dm
= super
->disks
; dm
; dm
= dm
->next
) {
7263 if (dm
->disk
.status
& FAILED_DISK
) {
7264 int idx
= dm
->index
;
7265 /* update indexes on the disk list */
7266 /* FIXME this loop-with-the-loop looks wrong, I'm not convinced
7267 the index values will end up being correct.... NB */
7268 for (du
= super
->disks
; du
; du
= du
->next
)
7269 if (du
->index
> idx
)
7271 /* mark as spare disk */
7276 map
->num_members
= map
->num_members
/ 2;
7277 map
->map_state
= IMSM_T_STATE_NORMAL
;
7278 map
->num_domains
= 1;
7279 map
->raid_level
= 0;
7280 map
->failed_disk_num
= -1;
7283 if (u
->direction
== R0_TO_R10
) {
7285 /* update slots in current disk list */
7286 for (dm
= super
->disks
; dm
; dm
= dm
->next
) {
7290 /* create new *missing* disks */
7291 for (i
= 0; i
< map
->num_members
; i
++) {
7292 space
= *space_list
;
7295 *space_list
= *space
;
7297 memcpy(du
, super
->disks
, sizeof(*du
));
7301 du
->index
= (i
* 2) + 1;
7302 sprintf((char *)du
->disk
.serial
,
7303 " MISSING_%d", du
->index
);
7304 sprintf((char *)du
->serial
,
7305 "MISSING_%d", du
->index
);
7306 du
->next
= super
->missing
;
7307 super
->missing
= du
;
7309 /* create new dev and map */
7310 space
= *space_list
;
7313 *space_list
= *space
;
7314 dev_new
= (void *)space
;
7315 memcpy(dev_new
, dev
, sizeof(*dev
));
7316 /* update new map */
7317 map
= get_imsm_map(dev_new
, 0);
7318 map
->num_members
= map
->num_members
* 2;
7319 map
->map_state
= IMSM_T_STATE_DEGRADED
;
7320 map
->num_domains
= 2;
7321 map
->raid_level
= 1;
7322 /* replace dev<->dev_new */
7325 /* update disk order table */
7326 for (du
= super
->disks
; du
; du
= du
->next
)
7328 set_imsm_ord_tbl_ent(map
, du
->index
, du
->index
);
7329 for (du
= super
->missing
; du
; du
= du
->next
)
7330 if (du
->index
>= 0) {
7331 set_imsm_ord_tbl_ent(map
, du
->index
, du
->index
);
7332 mark_missing(dv
->dev
, &du
->disk
, du
->index
);
7338 static void imsm_process_update(struct supertype
*st
,
7339 struct metadata_update
*update
)
7342 * crack open the metadata_update envelope to find the update record
7343 * update can be one of:
7344 * update_reshape_container_disks - all the arrays in the container
7345 * are being reshaped to have more devices. We need to mark
7346 * the arrays for general migration and convert selected spares
7347 * into active devices.
7348 * update_activate_spare - a spare device has replaced a failed
7349 * device in an array, update the disk_ord_tbl. If this disk is
7350 * present in all member arrays then also clear the SPARE_DISK
7352 * update_create_array
7354 * update_rename_array
7355 * update_add_remove_disk
7357 struct intel_super
*super
= st
->sb
;
7358 struct imsm_super
*mpb
;
7359 enum imsm_update_type type
= *(enum imsm_update_type
*) update
->buf
;
7361 /* update requires a larger buf but the allocation failed */
7362 if (super
->next_len
&& !super
->next_buf
) {
7363 super
->next_len
= 0;
7367 if (super
->next_buf
) {
7368 memcpy(super
->next_buf
, super
->buf
, super
->len
);
7370 super
->len
= super
->next_len
;
7371 super
->buf
= super
->next_buf
;
7373 super
->next_len
= 0;
7374 super
->next_buf
= NULL
;
7377 mpb
= super
->anchor
;
7380 case update_general_migration_checkpoint
: {
7381 struct intel_dev
*id
;
7382 struct imsm_update_general_migration_checkpoint
*u
=
7383 (void *)update
->buf
;
7385 dprintf("imsm: process_update() "
7386 "for update_general_migration_checkpoint called\n");
7388 /* find device under general migration */
7389 for (id
= super
->devlist
; id
; id
= id
->next
) {
7390 if (is_gen_migration(id
->dev
)) {
7391 id
->dev
->vol
.curr_migr_unit
=
7392 __cpu_to_le32(u
->curr_migr_unit
);
7393 super
->updates_pending
++;
7398 case update_takeover
: {
7399 struct imsm_update_takeover
*u
= (void *)update
->buf
;
7400 if (apply_takeover_update(u
, super
, &update
->space_list
)) {
7401 imsm_update_version_info(super
);
7402 super
->updates_pending
++;
7407 case update_reshape_container_disks
: {
7408 struct imsm_update_reshape
*u
= (void *)update
->buf
;
7409 if (apply_reshape_container_disks_update(
7410 u
, super
, &update
->space_list
))
7411 super
->updates_pending
++;
7414 case update_reshape_migration
: {
7415 struct imsm_update_reshape_migration
*u
= (void *)update
->buf
;
7416 if (apply_reshape_migration_update(
7417 u
, super
, &update
->space_list
))
7418 super
->updates_pending
++;
7421 case update_activate_spare
: {
7422 struct imsm_update_activate_spare
*u
= (void *) update
->buf
;
7423 if (apply_update_activate_spare(u
, super
, st
->arrays
))
7424 super
->updates_pending
++;
7427 case update_create_array
: {
7428 /* someone wants to create a new array, we need to be aware of
7429 * a few races/collisions:
7430 * 1/ 'Create' called by two separate instances of mdadm
7431 * 2/ 'Create' versus 'activate_spare': mdadm has chosen
7432 * devices that have since been assimilated via
7434 * In the event this update can not be carried out mdadm will
7435 * (FIX ME) notice that its update did not take hold.
7437 struct imsm_update_create_array
*u
= (void *) update
->buf
;
7438 struct intel_dev
*dv
;
7439 struct imsm_dev
*dev
;
7440 struct imsm_map
*map
, *new_map
;
7441 unsigned long long start
, end
;
7442 unsigned long long new_start
, new_end
;
7444 struct disk_info
*inf
;
7447 /* handle racing creates: first come first serve */
7448 if (u
->dev_idx
< mpb
->num_raid_devs
) {
7449 dprintf("%s: subarray %d already defined\n",
7450 __func__
, u
->dev_idx
);
7454 /* check update is next in sequence */
7455 if (u
->dev_idx
!= mpb
->num_raid_devs
) {
7456 dprintf("%s: can not create array %d expected index %d\n",
7457 __func__
, u
->dev_idx
, mpb
->num_raid_devs
);
7461 new_map
= get_imsm_map(&u
->dev
, 0);
7462 new_start
= __le32_to_cpu(new_map
->pba_of_lba0
);
7463 new_end
= new_start
+ __le32_to_cpu(new_map
->blocks_per_member
);
7464 inf
= get_disk_info(u
);
7466 /* handle activate_spare versus create race:
7467 * check to make sure that overlapping arrays do not include
7470 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
7471 dev
= get_imsm_dev(super
, i
);
7472 map
= get_imsm_map(dev
, 0);
7473 start
= __le32_to_cpu(map
->pba_of_lba0
);
7474 end
= start
+ __le32_to_cpu(map
->blocks_per_member
);
7475 if ((new_start
>= start
&& new_start
<= end
) ||
7476 (start
>= new_start
&& start
<= new_end
))
7481 if (disks_overlap(super
, i
, u
)) {
7482 dprintf("%s: arrays overlap\n", __func__
);
7487 /* check that prepare update was successful */
7488 if (!update
->space
) {
7489 dprintf("%s: prepare update failed\n", __func__
);
7493 /* check that all disks are still active before committing
7494 * changes. FIXME: could we instead handle this by creating a
7495 * degraded array? That's probably not what the user expects,
7496 * so better to drop this update on the floor.
7498 for (i
= 0; i
< new_map
->num_members
; i
++) {
7499 dl
= serial_to_dl(inf
[i
].serial
, super
);
7501 dprintf("%s: disk disappeared\n", __func__
);
7506 super
->updates_pending
++;
7508 /* convert spares to members and fixup ord_tbl */
7509 for (i
= 0; i
< new_map
->num_members
; i
++) {
7510 dl
= serial_to_dl(inf
[i
].serial
, super
);
7511 if (dl
->index
== -1) {
7512 dl
->index
= mpb
->num_disks
;
7514 dl
->disk
.status
|= CONFIGURED_DISK
;
7515 dl
->disk
.status
&= ~SPARE_DISK
;
7517 set_imsm_ord_tbl_ent(new_map
, i
, dl
->index
);
7522 update
->space
= NULL
;
7523 imsm_copy_dev(dev
, &u
->dev
);
7524 dv
->index
= u
->dev_idx
;
7525 dv
->next
= super
->devlist
;
7526 super
->devlist
= dv
;
7527 mpb
->num_raid_devs
++;
7529 imsm_update_version_info(super
);
7532 /* mdmon knows how to release update->space, but not
7533 * ((struct intel_dev *) update->space)->dev
7535 if (update
->space
) {
7541 case update_kill_array
: {
7542 struct imsm_update_kill_array
*u
= (void *) update
->buf
;
7543 int victim
= u
->dev_idx
;
7544 struct active_array
*a
;
7545 struct intel_dev
**dp
;
7546 struct imsm_dev
*dev
;
7548 /* sanity check that we are not affecting the uuid of
7549 * active arrays, or deleting an active array
7551 * FIXME when immutable ids are available, but note that
7552 * we'll also need to fixup the invalidated/active
7553 * subarray indexes in mdstat
7555 for (a
= st
->arrays
; a
; a
= a
->next
)
7556 if (a
->info
.container_member
>= victim
)
7558 /* by definition if mdmon is running at least one array
7559 * is active in the container, so checking
7560 * mpb->num_raid_devs is just extra paranoia
7562 dev
= get_imsm_dev(super
, victim
);
7563 if (a
|| !dev
|| mpb
->num_raid_devs
== 1) {
7564 dprintf("failed to delete subarray-%d\n", victim
);
7568 for (dp
= &super
->devlist
; *dp
;)
7569 if ((*dp
)->index
== (unsigned)super
->current_vol
) {
7572 if ((*dp
)->index
> (unsigned)victim
)
7576 mpb
->num_raid_devs
--;
7577 super
->updates_pending
++;
7580 case update_rename_array
: {
7581 struct imsm_update_rename_array
*u
= (void *) update
->buf
;
7582 char name
[MAX_RAID_SERIAL_LEN
+1];
7583 int target
= u
->dev_idx
;
7584 struct active_array
*a
;
7585 struct imsm_dev
*dev
;
7587 /* sanity check that we are not affecting the uuid of
7590 snprintf(name
, MAX_RAID_SERIAL_LEN
, "%s", (char *) u
->name
);
7591 name
[MAX_RAID_SERIAL_LEN
] = '\0';
7592 for (a
= st
->arrays
; a
; a
= a
->next
)
7593 if (a
->info
.container_member
== target
)
7595 dev
= get_imsm_dev(super
, u
->dev_idx
);
7596 if (a
|| !dev
|| !check_name(super
, name
, 1)) {
7597 dprintf("failed to rename subarray-%d\n", target
);
7601 snprintf((char *) dev
->volume
, MAX_RAID_SERIAL_LEN
, "%s", name
);
7602 super
->updates_pending
++;
7605 case update_add_remove_disk
: {
7606 /* we may be able to repair some arrays if disks are
7607 * being added, check teh status of add_remove_disk
7608 * if discs has been added.
7610 if (add_remove_disk_update(super
)) {
7611 struct active_array
*a
;
7613 super
->updates_pending
++;
7614 for (a
= st
->arrays
; a
; a
= a
->next
)
7615 a
->check_degraded
= 1;
7620 fprintf(stderr
, "error: unsuported process update type:"
7621 "(type: %d)\n", type
);
7625 static struct mdinfo
*get_spares_for_grow(struct supertype
*st
);
7627 static void imsm_prepare_update(struct supertype
*st
,
7628 struct metadata_update
*update
)
7631 * Allocate space to hold new disk entries, raid-device entries or a new
7632 * mpb if necessary. The manager synchronously waits for updates to
7633 * complete in the monitor, so new mpb buffers allocated here can be
7634 * integrated by the monitor thread without worrying about live pointers
7635 * in the manager thread.
7637 enum imsm_update_type type
= *(enum imsm_update_type
*) update
->buf
;
7638 struct intel_super
*super
= st
->sb
;
7639 struct imsm_super
*mpb
= super
->anchor
;
7644 case update_general_migration_checkpoint
:
7645 dprintf("imsm: prepare_update() "
7646 "for update_general_migration_checkpoint called\n");
7648 case update_takeover
: {
7649 struct imsm_update_takeover
*u
= (void *)update
->buf
;
7650 if (u
->direction
== R0_TO_R10
) {
7651 void **tail
= (void **)&update
->space_list
;
7652 struct imsm_dev
*dev
= get_imsm_dev(super
, u
->subarray
);
7653 struct imsm_map
*map
= get_imsm_map(dev
, 0);
7654 int num_members
= map
->num_members
;
7658 /* allocate memory for added disks */
7659 for (i
= 0; i
< num_members
; i
++) {
7660 size
= sizeof(struct dl
);
7661 space
= malloc(size
);
7670 /* allocate memory for new device */
7671 size
= sizeof_imsm_dev(super
->devlist
->dev
, 0) +
7672 (num_members
* sizeof(__u32
));
7673 space
= malloc(size
);
7682 len
= disks_to_mpb_size(num_members
* 2);
7684 /* if allocation didn't success, free buffer */
7685 while (update
->space_list
) {
7686 void **sp
= update
->space_list
;
7687 update
->space_list
= *sp
;
7695 case update_reshape_container_disks
: {
7696 /* Every raid device in the container is about to
7697 * gain some more devices, and we will enter a
7699 * So each 'imsm_map' will be bigger, and the imsm_vol
7700 * will now hold 2 of them.
7701 * Thus we need new 'struct imsm_dev' allocations sized
7702 * as sizeof_imsm_dev but with more devices in both maps.
7704 struct imsm_update_reshape
*u
= (void *)update
->buf
;
7705 struct intel_dev
*dl
;
7706 void **space_tail
= (void**)&update
->space_list
;
7708 dprintf("imsm: imsm_prepare_update() for update_reshape\n");
7710 for (dl
= super
->devlist
; dl
; dl
= dl
->next
) {
7711 int size
= sizeof_imsm_dev(dl
->dev
, 1);
7713 if (u
->new_raid_disks
> u
->old_raid_disks
)
7714 size
+= sizeof(__u32
)*2*
7715 (u
->new_raid_disks
- u
->old_raid_disks
);
7724 len
= disks_to_mpb_size(u
->new_raid_disks
);
7725 dprintf("New anchor length is %llu\n", (unsigned long long)len
);
7728 case update_reshape_migration
: {
7729 /* for migration level 0->5 we need to add disks
7730 * so the same as for container operation we will copy
7731 * device to the bigger location.
7732 * in memory prepared device and new disk area are prepared
7733 * for usage in process update
7735 struct imsm_update_reshape_migration
*u
= (void *)update
->buf
;
7736 struct intel_dev
*id
;
7737 void **space_tail
= (void **)&update
->space_list
;
7740 int current_level
= -1;
7742 dprintf("imsm: imsm_prepare_update() for update_reshape\n");
7744 /* add space for bigger array in update
7746 for (id
= super
->devlist
; id
; id
= id
->next
) {
7747 if (id
->index
== (unsigned)u
->subdev
) {
7748 size
= sizeof_imsm_dev(id
->dev
, 1);
7749 if (u
->new_raid_disks
> u
->old_raid_disks
)
7750 size
+= sizeof(__u32
)*2*
7751 (u
->new_raid_disks
- u
->old_raid_disks
);
7761 if (update
->space_list
== NULL
)
7764 /* add space for disk in update
7766 size
= sizeof(struct dl
);
7769 free(update
->space_list
);
7770 update
->space_list
= NULL
;
7777 /* add spare device to update
7779 for (id
= super
->devlist
; id
; id
= id
->next
)
7780 if (id
->index
== (unsigned)u
->subdev
) {
7781 struct imsm_dev
*dev
;
7782 struct imsm_map
*map
;
7784 dev
= get_imsm_dev(super
, u
->subdev
);
7785 map
= get_imsm_map(dev
, 0);
7786 current_level
= map
->raid_level
;
7789 if ((u
->new_level
== 5) && (u
->new_level
!= current_level
)) {
7790 struct mdinfo
*spares
;
7792 spares
= get_spares_for_grow(st
);
7800 makedev(dev
->disk
.major
,
7802 dl
= get_disk_super(super
,
7805 dl
->index
= u
->old_raid_disks
;
7811 len
= disks_to_mpb_size(u
->new_raid_disks
);
7812 dprintf("New anchor length is %llu\n", (unsigned long long)len
);
7815 case update_create_array
: {
7816 struct imsm_update_create_array
*u
= (void *) update
->buf
;
7817 struct intel_dev
*dv
;
7818 struct imsm_dev
*dev
= &u
->dev
;
7819 struct imsm_map
*map
= get_imsm_map(dev
, 0);
7821 struct disk_info
*inf
;
7825 inf
= get_disk_info(u
);
7826 len
= sizeof_imsm_dev(dev
, 1);
7827 /* allocate a new super->devlist entry */
7828 dv
= malloc(sizeof(*dv
));
7830 dv
->dev
= malloc(len
);
7835 update
->space
= NULL
;
7839 /* count how many spares will be converted to members */
7840 for (i
= 0; i
< map
->num_members
; i
++) {
7841 dl
= serial_to_dl(inf
[i
].serial
, super
);
7843 /* hmm maybe it failed?, nothing we can do about
7848 if (count_memberships(dl
, super
) == 0)
7851 len
+= activate
* sizeof(struct imsm_disk
);
7858 /* check if we need a larger metadata buffer */
7859 if (super
->next_buf
)
7860 buf_len
= super
->next_len
;
7862 buf_len
= super
->len
;
7864 if (__le32_to_cpu(mpb
->mpb_size
) + len
> buf_len
) {
7865 /* ok we need a larger buf than what is currently allocated
7866 * if this allocation fails process_update will notice that
7867 * ->next_len is set and ->next_buf is NULL
7869 buf_len
= ROUND_UP(__le32_to_cpu(mpb
->mpb_size
) + len
, 512);
7870 if (super
->next_buf
)
7871 free(super
->next_buf
);
7873 super
->next_len
= buf_len
;
7874 if (posix_memalign(&super
->next_buf
, 512, buf_len
) == 0)
7875 memset(super
->next_buf
, 0, buf_len
);
7877 super
->next_buf
= NULL
;
7881 /* must be called while manager is quiesced */
7882 static void imsm_delete(struct intel_super
*super
, struct dl
**dlp
, unsigned index
)
7884 struct imsm_super
*mpb
= super
->anchor
;
7886 struct imsm_dev
*dev
;
7887 struct imsm_map
*map
;
7888 int i
, j
, num_members
;
7891 dprintf("%s: deleting device[%d] from imsm_super\n",
7894 /* shift all indexes down one */
7895 for (iter
= super
->disks
; iter
; iter
= iter
->next
)
7896 if (iter
->index
> (int)index
)
7898 for (iter
= super
->missing
; iter
; iter
= iter
->next
)
7899 if (iter
->index
> (int)index
)
7902 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
7903 dev
= get_imsm_dev(super
, i
);
7904 map
= get_imsm_map(dev
, 0);
7905 num_members
= map
->num_members
;
7906 for (j
= 0; j
< num_members
; j
++) {
7907 /* update ord entries being careful not to propagate
7908 * ord-flags to the first map
7910 ord
= get_imsm_ord_tbl_ent(dev
, j
, -1);
7912 if (ord_to_idx(ord
) <= index
)
7915 map
= get_imsm_map(dev
, 0);
7916 set_imsm_ord_tbl_ent(map
, j
, ord_to_idx(ord
- 1));
7917 map
= get_imsm_map(dev
, 1);
7919 set_imsm_ord_tbl_ent(map
, j
, ord
- 1);
7924 super
->updates_pending
++;
7926 struct dl
*dl
= *dlp
;
7928 *dlp
= (*dlp
)->next
;
7929 __free_imsm_disk(dl
);
7932 #endif /* MDASSEMBLE */
7933 /*******************************************************************************
7934 * Function: open_backup_targets
7935 * Description: Function opens file descriptors for all devices given in
7938 * info : general array info
7939 * raid_disks : number of disks
7940 * raid_fds : table of device's file descriptors
7944 ******************************************************************************/
7945 int open_backup_targets(struct mdinfo
*info
, int raid_disks
, int *raid_fds
)
7949 for (sd
= info
->devs
; sd
; sd
= sd
->next
) {
7952 if (sd
->disk
.state
& (1<<MD_DISK_FAULTY
)) {
7953 dprintf("disk is faulty!!\n");
7957 if ((sd
->disk
.raid_disk
>= raid_disks
) ||
7958 (sd
->disk
.raid_disk
< 0))
7961 dn
= map_dev(sd
->disk
.major
,
7963 raid_fds
[sd
->disk
.raid_disk
] = dev_open(dn
, O_RDWR
);
7964 if (raid_fds
[sd
->disk
.raid_disk
] < 0) {
7965 fprintf(stderr
, "cannot open component\n");
7973 /*******************************************************************************
7974 * Function: init_migr_record_imsm
7975 * Description: Function inits imsm migration record
7977 * super : imsm internal array info
7978 * dev : device under migration
7979 * info : general array info to find the smallest device
7982 ******************************************************************************/
7983 void init_migr_record_imsm(struct supertype
*st
, struct imsm_dev
*dev
,
7984 struct mdinfo
*info
)
7986 struct intel_super
*super
= st
->sb
;
7987 struct migr_record
*migr_rec
= super
->migr_rec
;
7989 unsigned long long dsize
, dev_sectors
;
7990 long long unsigned min_dev_sectors
= -1LLU;
7994 struct imsm_map
*map_dest
= get_imsm_map(dev
, 0);
7995 struct imsm_map
*map_src
= get_imsm_map(dev
, 1);
7996 unsigned long long num_migr_units
;
7997 unsigned long long array_blocks
;
7999 memset(migr_rec
, 0, sizeof(struct migr_record
));
8000 migr_rec
->family_num
= __cpu_to_le32(super
->anchor
->family_num
);
8002 /* only ascending reshape supported now */
8003 migr_rec
->ascending_migr
= __cpu_to_le32(1);
8005 migr_rec
->dest_depth_per_unit
= GEN_MIGR_AREA_SIZE
/
8006 max(map_dest
->blocks_per_strip
, map_src
->blocks_per_strip
);
8007 migr_rec
->dest_depth_per_unit
*= map_dest
->blocks_per_strip
;
8008 new_data_disks
= imsm_num_data_members(dev
, 0);
8009 migr_rec
->blocks_per_unit
=
8010 __cpu_to_le32(migr_rec
->dest_depth_per_unit
* new_data_disks
);
8011 migr_rec
->dest_depth_per_unit
=
8012 __cpu_to_le32(migr_rec
->dest_depth_per_unit
);
8013 array_blocks
= info
->component_size
* new_data_disks
;
8015 array_blocks
/ __le32_to_cpu(migr_rec
->blocks_per_unit
);
8017 if (array_blocks
% __le32_to_cpu(migr_rec
->blocks_per_unit
))
8019 migr_rec
->num_migr_units
= __cpu_to_le32(num_migr_units
);
8021 migr_rec
->post_migr_vol_cap
= dev
->size_low
;
8022 migr_rec
->post_migr_vol_cap_hi
= dev
->size_high
;
8025 /* Find the smallest dev */
8026 for (sd
= info
->devs
; sd
; sd
= sd
->next
) {
8027 sprintf(nm
, "%d:%d", sd
->disk
.major
, sd
->disk
.minor
);
8028 fd
= dev_open(nm
, O_RDONLY
);
8031 get_dev_size(fd
, NULL
, &dsize
);
8032 dev_sectors
= dsize
/ 512;
8033 if (dev_sectors
< min_dev_sectors
)
8034 min_dev_sectors
= dev_sectors
;
8037 migr_rec
->ckpt_area_pba
= __cpu_to_le32(min_dev_sectors
-
8038 RAID_DISK_RESERVED_BLOCKS_IMSM_HI
);
8040 write_imsm_migr_rec(st
);
8045 /*******************************************************************************
8046 * Function: save_backup_imsm
8047 * Description: Function saves critical data stripes to Migration Copy Area
8048 * and updates the current migration unit status.
8049 * Use restore_stripes() to form a destination stripe,
8050 * and to write it to the Copy Area.
8052 * st : supertype information
8053 * dev : imsm device that backup is saved for
8054 * info : general array info
8055 * buf : input buffer
8056 * length : length of data to backup (blocks_per_unit)
8060 ******************************************************************************/
8061 int save_backup_imsm(struct supertype
*st
,
8062 struct imsm_dev
*dev
,
8063 struct mdinfo
*info
,
8068 struct intel_super
*super
= st
->sb
;
8069 unsigned long long *target_offsets
= NULL
;
8070 int *targets
= NULL
;
8072 struct imsm_map
*map_dest
= get_imsm_map(dev
, 0);
8073 int new_disks
= map_dest
->num_members
;
8074 int dest_layout
= 0;
8076 unsigned long long start
;
8077 int data_disks
= imsm_num_data_members(dev
, 0);
8079 targets
= malloc(new_disks
* sizeof(int));
8083 for (i
= 0; i
< new_disks
; i
++)
8086 target_offsets
= malloc(new_disks
* sizeof(unsigned long long));
8087 if (!target_offsets
)
8090 start
= info
->reshape_progress
* 512;
8091 for (i
= 0; i
< new_disks
; i
++) {
8092 target_offsets
[i
] = (unsigned long long)
8093 __le32_to_cpu(super
->migr_rec
->ckpt_area_pba
) * 512;
8094 /* move back copy area adderss, it will be moved forward
8095 * in restore_stripes() using start input variable
8097 target_offsets
[i
] -= start
/data_disks
;
8100 if (open_backup_targets(info
, new_disks
, targets
))
8103 dest_layout
= imsm_level_to_layout(map_dest
->raid_level
);
8104 dest_chunk
= __le16_to_cpu(map_dest
->blocks_per_strip
) * 512;
8106 if (restore_stripes(targets
, /* list of dest devices */
8107 target_offsets
, /* migration record offsets */
8110 map_dest
->raid_level
,
8112 -1, /* source backup file descriptor */
8113 0, /* input buf offset
8114 * always 0 buf is already offseted */
8118 fprintf(stderr
, Name
": Error restoring stripes\n");
8126 for (i
= 0; i
< new_disks
; i
++)
8127 if (targets
[i
] >= 0)
8131 free(target_offsets
);
8136 /*******************************************************************************
8137 * Function: save_checkpoint_imsm
8138 * Description: Function called for current unit status update
8139 * in the migration record. It writes it to disk.
8141 * super : imsm internal array info
8142 * info : general array info
8146 * 2: failure, means no valid migration record
8147 * / no general migration in progress /
8148 ******************************************************************************/
8149 int save_checkpoint_imsm(struct supertype
*st
, struct mdinfo
*info
, int state
)
8151 struct intel_super
*super
= st
->sb
;
8152 unsigned long long blocks_per_unit
;
8153 unsigned long long curr_migr_unit
;
8155 if (load_imsm_migr_rec(super
, info
) != 0) {
8156 dprintf("imsm: ERROR: Cannot read migration record "
8157 "for checkpoint save.\n");
8161 blocks_per_unit
= __le32_to_cpu(super
->migr_rec
->blocks_per_unit
);
8162 if (blocks_per_unit
== 0) {
8163 dprintf("imsm: no migration in progress.\n");
8166 curr_migr_unit
= info
->reshape_progress
/ blocks_per_unit
;
8167 /* check if array is alligned to copy area
8168 * if it is not alligned, add one to current migration unit value
8169 * this can happend on array reshape finish only
8171 if (info
->reshape_progress
% blocks_per_unit
)
8174 super
->migr_rec
->curr_migr_unit
=
8175 __cpu_to_le32(curr_migr_unit
);
8176 super
->migr_rec
->rec_status
= __cpu_to_le32(state
);
8177 super
->migr_rec
->dest_1st_member_lba
=
8178 __cpu_to_le32(curr_migr_unit
*
8179 __le32_to_cpu(super
->migr_rec
->dest_depth_per_unit
));
8180 if (write_imsm_migr_rec(st
) < 0) {
8181 dprintf("imsm: Cannot write migration record "
8182 "outside backup area\n");
8189 /*******************************************************************************
8190 * Function: recover_backup_imsm
8191 * Description: Function recovers critical data from the Migration Copy Area
8192 * while assembling an array.
8194 * super : imsm internal array info
8195 * info : general array info
8197 * 0 : success (or there is no data to recover)
8199 ******************************************************************************/
8200 int recover_backup_imsm(struct supertype
*st
, struct mdinfo
*info
)
8202 struct intel_super
*super
= st
->sb
;
8203 struct migr_record
*migr_rec
= super
->migr_rec
;
8204 struct imsm_map
*map_dest
= NULL
;
8205 struct intel_dev
*id
= NULL
;
8206 unsigned long long read_offset
;
8207 unsigned long long write_offset
;
8209 int *targets
= NULL
;
8210 int new_disks
, i
, err
;
8213 unsigned long curr_migr_unit
= __le32_to_cpu(migr_rec
->curr_migr_unit
);
8214 unsigned long num_migr_units
= __le32_to_cpu(migr_rec
->num_migr_units
);
8216 int skipped_disks
= 0;
8217 int max_degradation
;
8219 err
= sysfs_get_str(info
, NULL
, "array_state", (char *)buffer
, 20);
8223 /* recover data only during assemblation */
8224 if (strncmp(buffer
, "inactive", 8) != 0)
8226 /* no data to recover */
8227 if (__le32_to_cpu(migr_rec
->rec_status
) == UNIT_SRC_NORMAL
)
8229 if (curr_migr_unit
>= num_migr_units
)
8232 /* find device during reshape */
8233 for (id
= super
->devlist
; id
; id
= id
->next
)
8234 if (is_gen_migration(id
->dev
))
8239 map_dest
= get_imsm_map(id
->dev
, 0);
8240 new_disks
= map_dest
->num_members
;
8241 max_degradation
= new_disks
- imsm_num_data_members(id
->dev
, 0);
8243 read_offset
= (unsigned long long)
8244 __le32_to_cpu(migr_rec
->ckpt_area_pba
) * 512;
8246 write_offset
= ((unsigned long long)
8247 __le32_to_cpu(migr_rec
->dest_1st_member_lba
) +
8248 __le32_to_cpu(map_dest
->pba_of_lba0
)) * 512;
8250 unit_len
= __le32_to_cpu(migr_rec
->dest_depth_per_unit
) * 512;
8251 if (posix_memalign((void **)&buf
, 512, unit_len
) != 0)
8253 targets
= malloc(new_disks
* sizeof(int));
8257 open_backup_targets(info
, new_disks
, targets
);
8259 for (i
= 0; i
< new_disks
; i
++) {
8260 if (targets
[i
] < 0) {
8264 if (lseek64(targets
[i
], read_offset
, SEEK_SET
) < 0) {
8266 Name
": Cannot seek to block: %s\n",
8270 if ((unsigned)read(targets
[i
], buf
, unit_len
) != unit_len
) {
8272 Name
": Cannot read copy area block: %s\n",
8276 if (lseek64(targets
[i
], write_offset
, SEEK_SET
) < 0) {
8278 Name
": Cannot seek to block: %s\n",
8282 if ((unsigned)write(targets
[i
], buf
, unit_len
) != unit_len
) {
8284 Name
": Cannot restore block: %s\n",
8290 if (skipped_disks
> max_degradation
) {
8292 Name
": Cannot restore data from backup."
8293 " Too many failed disks\n");
8297 if (save_checkpoint_imsm(st
, info
, UNIT_SRC_NORMAL
)) {
8298 /* ignore error == 2, this can mean end of reshape here
8300 dprintf("imsm: Cannot write checkpoint to "
8301 "migration record (UNIT_SRC_NORMAL) during restart\n");
8307 for (i
= 0; i
< new_disks
; i
++)
8316 static char disk_by_path
[] = "/dev/disk/by-path/";
8318 static const char *imsm_get_disk_controller_domain(const char *path
)
8320 char disk_path
[PATH_MAX
];
8324 strncpy(disk_path
, disk_by_path
, PATH_MAX
- 1);
8325 strncat(disk_path
, path
, PATH_MAX
- strlen(disk_path
) - 1);
8326 if (stat(disk_path
, &st
) == 0) {
8327 struct sys_dev
* hba
;
8330 path
= devt_to_devpath(st
.st_rdev
);
8333 hba
= find_disk_attached_hba(-1, path
);
8334 if (hba
&& hba
->type
== SYS_DEV_SAS
)
8336 else if (hba
&& hba
->type
== SYS_DEV_SATA
)
8340 dprintf("path: %s hba: %s attached: %s\n",
8341 path
, (hba
) ? hba
->path
: "NULL", drv
);
8349 static int imsm_find_array_minor_by_subdev(int subdev
, int container
, int *minor
)
8351 char subdev_name
[20];
8352 struct mdstat_ent
*mdstat
;
8354 sprintf(subdev_name
, "%d", subdev
);
8355 mdstat
= mdstat_by_subdev(subdev_name
, container
);
8359 *minor
= mdstat
->devnum
;
8360 free_mdstat(mdstat
);
8364 static int imsm_reshape_is_allowed_on_container(struct supertype
*st
,
8365 struct geo_params
*geo
,
8366 int *old_raid_disks
)
8368 /* currently we only support increasing the number of devices
8369 * for a container. This increases the number of device for each
8370 * member array. They must all be RAID0 or RAID5.
8373 struct mdinfo
*info
, *member
;
8374 int devices_that_can_grow
= 0;
8376 dprintf("imsm: imsm_reshape_is_allowed_on_container(ENTER): "
8377 "st->devnum = (%i)\n",
8380 if (geo
->size
!= -1 ||
8381 geo
->level
!= UnSet
||
8382 geo
->layout
!= UnSet
||
8383 geo
->chunksize
!= 0 ||
8384 geo
->raid_disks
== UnSet
) {
8385 dprintf("imsm: Container operation is allowed for "
8386 "raid disks number change only.\n");
8390 info
= container_content_imsm(st
, NULL
);
8391 for (member
= info
; member
; member
= member
->next
) {
8395 dprintf("imsm: checking device_num: %i\n",
8396 member
->container_member
);
8398 if (geo
->raid_disks
<= member
->array
.raid_disks
) {
8399 /* we work on container for Online Capacity Expansion
8400 * only so raid_disks has to grow
8402 dprintf("imsm: for container operation raid disks "
8403 "increase is required\n");
8407 if ((info
->array
.level
!= 0) &&
8408 (info
->array
.level
!= 5)) {
8409 /* we cannot use this container with other raid level
8411 dprintf("imsm: for container operation wrong"
8412 " raid level (%i) detected\n",
8416 /* check for platform support
8417 * for this raid level configuration
8419 struct intel_super
*super
= st
->sb
;
8420 if (!is_raid_level_supported(super
->orom
,
8421 member
->array
.level
,
8423 dprintf("platform does not support raid%d with"
8427 geo
->raid_disks
> 1 ? "s" : "");
8430 /* check if component size is aligned to chunk size
8432 if (info
->component_size
%
8433 (info
->array
.chunk_size
/512)) {
8434 dprintf("Component size is not aligned to "
8440 if (*old_raid_disks
&&
8441 info
->array
.raid_disks
!= *old_raid_disks
)
8443 *old_raid_disks
= info
->array
.raid_disks
;
8445 /* All raid5 and raid0 volumes in container
8446 * have to be ready for Online Capacity Expansion
8447 * so they need to be assembled. We have already
8448 * checked that no recovery etc is happening.
8450 result
= imsm_find_array_minor_by_subdev(member
->container_member
,
8454 dprintf("imsm: cannot find array\n");
8457 devices_that_can_grow
++;
8460 if (!member
&& devices_that_can_grow
)
8464 dprintf("\tContainer operation allowed\n");
8466 dprintf("\tError: %i\n", ret_val
);
8471 /* Function: get_spares_for_grow
8472 * Description: Allocates memory and creates list of spare devices
8473 * avaliable in container. Checks if spare drive size is acceptable.
8474 * Parameters: Pointer to the supertype structure
8475 * Returns: Pointer to the list of spare devices (mdinfo structure) on success,
8478 static struct mdinfo
*get_spares_for_grow(struct supertype
*st
)
8480 unsigned long long min_size
= min_acceptable_spare_size_imsm(st
);
8481 return container_choose_spares(st
, min_size
, NULL
, NULL
, NULL
, 0);
8484 /******************************************************************************
8485 * function: imsm_create_metadata_update_for_reshape
8486 * Function creates update for whole IMSM container.
8488 ******************************************************************************/
8489 static int imsm_create_metadata_update_for_reshape(
8490 struct supertype
*st
,
8491 struct geo_params
*geo
,
8493 struct imsm_update_reshape
**updatep
)
8495 struct intel_super
*super
= st
->sb
;
8496 struct imsm_super
*mpb
= super
->anchor
;
8497 int update_memory_size
= 0;
8498 struct imsm_update_reshape
*u
= NULL
;
8499 struct mdinfo
*spares
= NULL
;
8501 int delta_disks
= 0;
8504 dprintf("imsm_update_metadata_for_reshape(enter) raid_disks = %i\n",
8507 delta_disks
= geo
->raid_disks
- old_raid_disks
;
8509 /* size of all update data without anchor */
8510 update_memory_size
= sizeof(struct imsm_update_reshape
);
8512 /* now add space for spare disks that we need to add. */
8513 update_memory_size
+= sizeof(u
->new_disks
[0]) * (delta_disks
- 1);
8515 u
= calloc(1, update_memory_size
);
8518 "cannot get memory for imsm_update_reshape update\n");
8521 u
->type
= update_reshape_container_disks
;
8522 u
->old_raid_disks
= old_raid_disks
;
8523 u
->new_raid_disks
= geo
->raid_disks
;
8525 /* now get spare disks list
8527 spares
= get_spares_for_grow(st
);
8530 || delta_disks
> spares
->array
.spare_disks
) {
8531 fprintf(stderr
, Name
": imsm: ERROR: Cannot get spare devices "
8532 "for %s.\n", geo
->dev_name
);
8537 /* we have got spares
8538 * update disk list in imsm_disk list table in anchor
8540 dprintf("imsm: %i spares are available.\n\n",
8541 spares
->array
.spare_disks
);
8544 for (i
= 0; i
< delta_disks
; i
++) {
8549 u
->new_disks
[i
] = makedev(dev
->disk
.major
,
8551 dl
= get_disk_super(super
, dev
->disk
.major
, dev
->disk
.minor
);
8552 dl
->index
= mpb
->num_disks
;
8562 dprintf("imsm: reshape update preparation :");
8563 if (i
== delta_disks
) {
8566 return update_memory_size
;
8569 dprintf(" Error\n");
8574 /******************************************************************************
8575 * function: imsm_create_metadata_update_for_migration()
8576 * Creates update for IMSM array.
8578 ******************************************************************************/
8579 static int imsm_create_metadata_update_for_migration(
8580 struct supertype
*st
,
8581 struct geo_params
*geo
,
8582 struct imsm_update_reshape_migration
**updatep
)
8584 struct intel_super
*super
= st
->sb
;
8585 int update_memory_size
= 0;
8586 struct imsm_update_reshape_migration
*u
= NULL
;
8587 struct imsm_dev
*dev
;
8588 int previous_level
= -1;
8590 dprintf("imsm_create_metadata_update_for_migration(enter)"
8591 " New Level = %i\n", geo
->level
);
8593 /* size of all update data without anchor */
8594 update_memory_size
= sizeof(struct imsm_update_reshape_migration
);
8596 u
= calloc(1, update_memory_size
);
8598 dprintf("error: cannot get memory for "
8599 "imsm_create_metadata_update_for_migration\n");
8602 u
->type
= update_reshape_migration
;
8603 u
->subdev
= super
->current_vol
;
8604 u
->new_level
= geo
->level
;
8605 u
->new_layout
= geo
->layout
;
8606 u
->new_raid_disks
= u
->old_raid_disks
= geo
->raid_disks
;
8607 u
->new_disks
[0] = -1;
8608 u
->new_chunksize
= -1;
8610 dev
= get_imsm_dev(super
, u
->subdev
);
8612 struct imsm_map
*map
;
8614 map
= get_imsm_map(dev
, 0);
8616 int current_chunk_size
=
8617 __le16_to_cpu(map
->blocks_per_strip
) / 2;
8619 if (geo
->chunksize
!= current_chunk_size
) {
8620 u
->new_chunksize
= geo
->chunksize
/ 1024;
8622 "chunk size change from %i to %i\n",
8623 current_chunk_size
, u
->new_chunksize
);
8625 previous_level
= map
->raid_level
;
8628 if ((geo
->level
== 5) && (previous_level
== 0)) {
8629 struct mdinfo
*spares
= NULL
;
8631 u
->new_raid_disks
++;
8632 spares
= get_spares_for_grow(st
);
8633 if ((spares
== NULL
) || (spares
->array
.spare_disks
< 1)) {
8636 update_memory_size
= 0;
8637 dprintf("error: cannot get spare device "
8638 "for requested migration");
8643 dprintf("imsm: reshape update preparation : OK\n");
8646 return update_memory_size
;
8649 static void imsm_update_metadata_locally(struct supertype
*st
,
8652 struct metadata_update mu
;
8657 mu
.space_list
= NULL
;
8659 imsm_prepare_update(st
, &mu
);
8660 imsm_process_update(st
, &mu
);
8662 while (mu
.space_list
) {
8663 void **space
= mu
.space_list
;
8664 mu
.space_list
= *space
;
8669 /***************************************************************************
8670 * Function: imsm_analyze_change
8671 * Description: Function analyze change for single volume
8672 * and validate if transition is supported
8673 * Parameters: Geometry parameters, supertype structure
8674 * Returns: Operation type code on success, -1 if fail
8675 ****************************************************************************/
8676 enum imsm_reshape_type
imsm_analyze_change(struct supertype
*st
,
8677 struct geo_params
*geo
)
8686 getinfo_super_imsm_volume(st
, &info
, NULL
);
8687 if ((geo
->level
!= info
.array
.level
) &&
8688 (geo
->level
>= 0) &&
8689 (geo
->level
!= UnSet
)) {
8690 switch (info
.array
.level
) {
8692 if (geo
->level
== 5) {
8693 change
= CH_MIGRATION
;
8694 if (geo
->layout
!= ALGORITHM_LEFT_ASYMMETRIC
) {
8696 Name
" Error. Requested Layout "
8697 "not supported (left-asymmetric layout "
8698 "is supported only)!\n");
8700 goto analyse_change_exit
;
8702 layout
= geo
->layout
;
8704 devNumChange
= 1; /* parity disk added */
8705 } else if (geo
->level
== 10) {
8706 change
= CH_TAKEOVER
;
8708 devNumChange
= 2; /* two mirrors added */
8709 layout
= 0x102; /* imsm supported layout */
8714 if (geo
->level
== 0) {
8715 change
= CH_TAKEOVER
;
8717 devNumChange
= -(geo
->raid_disks
/2);
8718 layout
= 0; /* imsm raid0 layout */
8724 Name
" Error. Level Migration from %d to %d "
8726 info
.array
.level
, geo
->level
);
8727 goto analyse_change_exit
;
8730 geo
->level
= info
.array
.level
;
8732 if ((geo
->layout
!= info
.array
.layout
)
8733 && ((geo
->layout
!= UnSet
) && (geo
->layout
!= -1))) {
8734 change
= CH_MIGRATION
;
8735 if ((info
.array
.layout
== 0)
8736 && (info
.array
.level
== 5)
8737 && (geo
->layout
== 5)) {
8738 /* reshape 5 -> 4 */
8739 } else if ((info
.array
.layout
== 5)
8740 && (info
.array
.level
== 5)
8741 && (geo
->layout
== 0)) {
8742 /* reshape 4 -> 5 */
8747 Name
" Error. Layout Migration from %d to %d "
8749 info
.array
.layout
, geo
->layout
);
8751 goto analyse_change_exit
;
8754 geo
->layout
= info
.array
.layout
;
8756 if ((geo
->chunksize
> 0) && (geo
->chunksize
!= UnSet
)
8757 && (geo
->chunksize
!= info
.array
.chunk_size
))
8758 change
= CH_MIGRATION
;
8760 geo
->chunksize
= info
.array
.chunk_size
;
8762 chunk
= geo
->chunksize
/ 1024;
8763 if (!validate_geometry_imsm(st
,
8766 geo
->raid_disks
+ devNumChange
,
8773 struct intel_super
*super
= st
->sb
;
8774 struct imsm_super
*mpb
= super
->anchor
;
8776 if (mpb
->num_raid_devs
> 1) {
8778 Name
" Error. Cannot perform operation on %s"
8779 "- for this operation it MUST be single "
8780 "array in container\n",
8786 analyse_change_exit
:
8791 int imsm_takeover(struct supertype
*st
, struct geo_params
*geo
)
8793 struct intel_super
*super
= st
->sb
;
8794 struct imsm_update_takeover
*u
;
8796 u
= malloc(sizeof(struct imsm_update_takeover
));
8800 u
->type
= update_takeover
;
8801 u
->subarray
= super
->current_vol
;
8803 /* 10->0 transition */
8804 if (geo
->level
== 0)
8805 u
->direction
= R10_TO_R0
;
8807 /* 0->10 transition */
8808 if (geo
->level
== 10)
8809 u
->direction
= R0_TO_R10
;
8811 /* update metadata locally */
8812 imsm_update_metadata_locally(st
, u
,
8813 sizeof(struct imsm_update_takeover
));
8814 /* and possibly remotely */
8815 if (st
->update_tail
)
8816 append_metadata_update(st
, u
,
8817 sizeof(struct imsm_update_takeover
));
8824 static int imsm_reshape_super(struct supertype
*st
, long long size
, int level
,
8825 int layout
, int chunksize
, int raid_disks
,
8826 int delta_disks
, char *backup
, char *dev
,
8830 struct geo_params geo
;
8832 dprintf("imsm: reshape_super called.\n");
8834 memset(&geo
, 0, sizeof(struct geo_params
));
8837 geo
.dev_id
= st
->devnum
;
8840 geo
.layout
= layout
;
8841 geo
.chunksize
= chunksize
;
8842 geo
.raid_disks
= raid_disks
;
8843 if (delta_disks
!= UnSet
)
8844 geo
.raid_disks
+= delta_disks
;
8846 dprintf("\tfor level : %i\n", geo
.level
);
8847 dprintf("\tfor raid_disks : %i\n", geo
.raid_disks
);
8849 if (experimental() == 0)
8852 if (st
->container_dev
== st
->devnum
) {
8853 /* On container level we can only increase number of devices. */
8854 dprintf("imsm: info: Container operation\n");
8855 int old_raid_disks
= 0;
8857 if (imsm_reshape_is_allowed_on_container(
8858 st
, &geo
, &old_raid_disks
)) {
8859 struct imsm_update_reshape
*u
= NULL
;
8862 len
= imsm_create_metadata_update_for_reshape(
8863 st
, &geo
, old_raid_disks
, &u
);
8866 dprintf("imsm: Cannot prepare update\n");
8867 goto exit_imsm_reshape_super
;
8871 /* update metadata locally */
8872 imsm_update_metadata_locally(st
, u
, len
);
8873 /* and possibly remotely */
8874 if (st
->update_tail
)
8875 append_metadata_update(st
, u
, len
);
8880 fprintf(stderr
, Name
": (imsm) Operation "
8881 "is not allowed on this container\n");
8884 /* On volume level we support following operations
8885 * - takeover: raid10 -> raid0; raid0 -> raid10
8886 * - chunk size migration
8887 * - migration: raid5 -> raid0; raid0 -> raid5
8889 struct intel_super
*super
= st
->sb
;
8890 struct intel_dev
*dev
= super
->devlist
;
8892 dprintf("imsm: info: Volume operation\n");
8893 /* find requested device */
8895 if (imsm_find_array_minor_by_subdev(
8896 dev
->index
, st
->container_dev
, &devnum
) == 0
8897 && devnum
== geo
.dev_id
)
8902 fprintf(stderr
, Name
" Cannot find %s (%i) subarray\n",
8903 geo
.dev_name
, geo
.dev_id
);
8904 goto exit_imsm_reshape_super
;
8906 super
->current_vol
= dev
->index
;
8907 change
= imsm_analyze_change(st
, &geo
);
8910 ret_val
= imsm_takeover(st
, &geo
);
8912 case CH_MIGRATION
: {
8913 struct imsm_update_reshape_migration
*u
= NULL
;
8915 imsm_create_metadata_update_for_migration(
8919 "Cannot prepare update\n");
8923 /* update metadata locally */
8924 imsm_update_metadata_locally(st
, u
, len
);
8925 /* and possibly remotely */
8926 if (st
->update_tail
)
8927 append_metadata_update(st
, u
, len
);
8937 exit_imsm_reshape_super
:
8938 dprintf("imsm: reshape_super Exit code = %i\n", ret_val
);
8942 /*******************************************************************************
8943 * Function: wait_for_reshape_imsm
8944 * Description: Function writes new sync_max value and waits until
8945 * reshape process reach new position
8947 * sra : general array info
8948 * ndata : number of disks in new array's layout
8951 * 1 : there is no reshape in progress,
8953 ******************************************************************************/
8954 int wait_for_reshape_imsm(struct mdinfo
*sra
, int ndata
)
8956 int fd
= sysfs_get_fd(sra
, NULL
, "reshape_position");
8957 unsigned long long completed
;
8958 /* to_complete : new sync_max position */
8959 unsigned long long to_complete
= sra
->reshape_progress
;
8960 unsigned long long position_to_set
= to_complete
/ ndata
;
8963 dprintf("imsm: wait_for_reshape_imsm() "
8964 "cannot open reshape_position\n");
8968 if (sysfs_fd_get_ll(fd
, &completed
) < 0) {
8969 dprintf("imsm: wait_for_reshape_imsm() "
8970 "cannot read reshape_position (no reshape in progres)\n");
8975 if (completed
> to_complete
) {
8976 dprintf("imsm: wait_for_reshape_imsm() "
8977 "wrong next position to set %llu (%llu)\n",
8978 to_complete
, completed
);
8982 dprintf("Position set: %llu\n", position_to_set
);
8983 if (sysfs_set_num(sra
, NULL
, "sync_max",
8984 position_to_set
) != 0) {
8985 dprintf("imsm: wait_for_reshape_imsm() "
8986 "cannot set reshape position to %llu\n",
8997 select(fd
+1, &rfds
, NULL
, NULL
, NULL
);
8998 if (sysfs_get_str(sra
, NULL
, "sync_action",
9000 strncmp(action
, "reshape", 7) != 0)
9002 if (sysfs_fd_get_ll(fd
, &completed
) < 0) {
9003 dprintf("imsm: wait_for_reshape_imsm() "
9004 "cannot read reshape_position (in loop)\n");
9008 } while (completed
< to_complete
);
9014 /*******************************************************************************
9015 * Function: check_degradation_change
9016 * Description: Check that array hasn't become failed.
9018 * info : for sysfs access
9019 * sources : source disks descriptors
9020 * degraded: previous degradation level
9023 ******************************************************************************/
9024 int check_degradation_change(struct mdinfo
*info
,
9028 unsigned long long new_degraded
;
9029 sysfs_get_ll(info
, NULL
, "degraded", &new_degraded
);
9030 if (new_degraded
!= (unsigned long long)degraded
) {
9031 /* check each device to ensure it is still working */
9034 for (sd
= info
->devs
; sd
; sd
= sd
->next
) {
9035 if (sd
->disk
.state
& (1<<MD_DISK_FAULTY
))
9037 if (sd
->disk
.state
& (1<<MD_DISK_SYNC
)) {
9039 if (sysfs_get_str(info
,
9040 sd
, "state", sbuf
, 20) < 0 ||
9041 strstr(sbuf
, "faulty") ||
9042 strstr(sbuf
, "in_sync") == NULL
) {
9043 /* this device is dead */
9044 sd
->disk
.state
= (1<<MD_DISK_FAULTY
);
9045 if (sd
->disk
.raid_disk
>= 0 &&
9046 sources
[sd
->disk
.raid_disk
] >= 0) {
9048 sd
->disk
.raid_disk
]);
9049 sources
[sd
->disk
.raid_disk
] =
9058 return new_degraded
;
9061 /*******************************************************************************
9062 * Function: imsm_manage_reshape
9063 * Description: Function finds array under reshape and it manages reshape
9064 * process. It creates stripes backups (if required) and sets
9067 * afd : Backup handle (nattive) - not used
9068 * sra : general array info
9069 * reshape : reshape parameters - not used
9070 * st : supertype structure
9071 * blocks : size of critical section [blocks]
9072 * fds : table of source device descriptor
9073 * offsets : start of array (offest per devices)
9075 * destfd : table of destination device descriptor
9076 * destoffsets : table of destination offsets (per device)
9078 * 1 : success, reshape is done
9080 ******************************************************************************/
9081 static int imsm_manage_reshape(
9082 int afd
, struct mdinfo
*sra
, struct reshape
*reshape
,
9083 struct supertype
*st
, unsigned long backup_blocks
,
9084 int *fds
, unsigned long long *offsets
,
9085 int dests
, int *destfd
, unsigned long long *destoffsets
)
9088 struct intel_super
*super
= st
->sb
;
9089 struct intel_dev
*dv
= NULL
;
9090 struct imsm_dev
*dev
= NULL
;
9091 struct imsm_map
*map_src
;
9092 int migr_vol_qan
= 0;
9093 int ndata
, odata
; /* [bytes] */
9094 int chunk
; /* [bytes] */
9095 struct migr_record
*migr_rec
;
9097 unsigned int buf_size
; /* [bytes] */
9098 unsigned long long max_position
; /* array size [bytes] */
9099 unsigned long long next_step
; /* [blocks]/[bytes] */
9100 unsigned long long old_data_stripe_length
;
9101 unsigned long long start_src
; /* [bytes] */
9102 unsigned long long start
; /* [bytes] */
9103 unsigned long long start_buf_shift
; /* [bytes] */
9105 int source_layout
= 0;
9107 if (!fds
|| !offsets
|| !sra
)
9110 /* Find volume during the reshape */
9111 for (dv
= super
->devlist
; dv
; dv
= dv
->next
) {
9112 if (dv
->dev
->vol
.migr_type
== MIGR_GEN_MIGR
9113 && dv
->dev
->vol
.migr_state
== 1) {
9118 /* Only one volume can migrate at the same time */
9119 if (migr_vol_qan
!= 1) {
9120 fprintf(stderr
, Name
" : %s", migr_vol_qan
?
9121 "Number of migrating volumes greater than 1\n" :
9122 "There is no volume during migrationg\n");
9126 map_src
= get_imsm_map(dev
, 1);
9127 if (map_src
== NULL
)
9130 ndata
= imsm_num_data_members(dev
, 0);
9131 odata
= imsm_num_data_members(dev
, 1);
9133 chunk
= __le16_to_cpu(map_src
->blocks_per_strip
) * 512;
9134 old_data_stripe_length
= odata
* chunk
;
9136 migr_rec
= super
->migr_rec
;
9138 /* initialize migration record for start condition */
9139 if (sra
->reshape_progress
== 0)
9140 init_migr_record_imsm(st
, dev
, sra
);
9142 if (__le32_to_cpu(migr_rec
->rec_status
) != UNIT_SRC_NORMAL
) {
9143 dprintf("imsm: cannot restart migration when data "
9144 "are present in copy area.\n");
9150 buf_size
= __le32_to_cpu(migr_rec
->blocks_per_unit
) * 512;
9151 /* extend buffer size for parity disk */
9152 buf_size
+= __le32_to_cpu(migr_rec
->dest_depth_per_unit
) * 512;
9153 /* add space for stripe aligment */
9154 buf_size
+= old_data_stripe_length
;
9155 if (posix_memalign((void **)&buf
, 4096, buf_size
)) {
9156 dprintf("imsm: Cannot allocate checpoint buffer\n");
9160 max_position
= sra
->component_size
* ndata
;
9161 source_layout
= imsm_level_to_layout(map_src
->raid_level
);
9163 while (__le32_to_cpu(migr_rec
->curr_migr_unit
) <
9164 __le32_to_cpu(migr_rec
->num_migr_units
)) {
9165 /* current reshape position [blocks] */
9166 unsigned long long current_position
=
9167 __le32_to_cpu(migr_rec
->blocks_per_unit
)
9168 * __le32_to_cpu(migr_rec
->curr_migr_unit
);
9169 unsigned long long border
;
9171 /* Check that array hasn't become failed.
9173 degraded
= check_degradation_change(sra
, fds
, degraded
);
9175 dprintf("imsm: Abort reshape due to degradation"
9176 " level (%i)\n", degraded
);
9180 next_step
= __le32_to_cpu(migr_rec
->blocks_per_unit
);
9182 if ((current_position
+ next_step
) > max_position
)
9183 next_step
= max_position
- current_position
;
9185 start
= current_position
* 512;
9187 /* allign reading start to old geometry */
9188 start_buf_shift
= start
% old_data_stripe_length
;
9189 start_src
= start
- start_buf_shift
;
9191 border
= (start_src
/ odata
) - (start
/ ndata
);
9193 if (border
<= __le32_to_cpu(migr_rec
->dest_depth_per_unit
)) {
9194 /* save critical stripes to buf
9195 * start - start address of current unit
9197 * start_src - start address of current unit
9198 * to backup alligned to source array
9201 unsigned long long next_step_filler
= 0;
9202 unsigned long long copy_length
= next_step
* 512;
9204 /* allign copy area length to stripe in old geometry */
9205 next_step_filler
= ((copy_length
+ start_buf_shift
)
9206 % old_data_stripe_length
);
9207 if (next_step_filler
)
9208 next_step_filler
= (old_data_stripe_length
9209 - next_step_filler
);
9210 dprintf("save_stripes() parameters: start = %llu,"
9211 "\tstart_src = %llu,\tnext_step*512 = %llu,"
9212 "\tstart_in_buf_shift = %llu,"
9213 "\tnext_step_filler = %llu\n",
9214 start
, start_src
, copy_length
,
9215 start_buf_shift
, next_step_filler
);
9217 if (save_stripes(fds
, offsets
, map_src
->num_members
,
9218 chunk
, map_src
->raid_level
,
9219 source_layout
, 0, NULL
, start_src
,
9221 next_step_filler
+ start_buf_shift
,
9223 dprintf("imsm: Cannot save stripes"
9227 /* Convert data to destination format and store it
9228 * in backup general migration area
9230 if (save_backup_imsm(st
, dev
, sra
,
9231 buf
+ start_buf_shift
, copy_length
)) {
9232 dprintf("imsm: Cannot save stripes to "
9233 "target devices\n");
9236 if (save_checkpoint_imsm(st
, sra
,
9237 UNIT_SRC_IN_CP_AREA
)) {
9238 dprintf("imsm: Cannot write checkpoint to "
9239 "migration record (UNIT_SRC_IN_CP_AREA)\n");
9243 /* set next step to use whole border area */
9244 border
/= next_step
;
9246 next_step
*= border
;
9248 /* When data backed up, checkpoint stored,
9249 * kick the kernel to reshape unit of data
9251 next_step
= next_step
+ sra
->reshape_progress
;
9252 /* limit next step to array max position */
9253 if (next_step
> max_position
)
9254 next_step
= max_position
;
9255 sysfs_set_num(sra
, NULL
, "suspend_lo", sra
->reshape_progress
);
9256 sysfs_set_num(sra
, NULL
, "suspend_hi", next_step
);
9257 sra
->reshape_progress
= next_step
;
9259 /* wait until reshape finish */
9260 if (wait_for_reshape_imsm(sra
, ndata
) < 0) {
9261 dprintf("wait_for_reshape_imsm returned error!\n");
9265 if (save_checkpoint_imsm(st
, sra
, UNIT_SRC_NORMAL
) == 1) {
9266 /* ignore error == 2, this can mean end of reshape here
9268 dprintf("imsm: Cannot write checkpoint to "
9269 "migration record (UNIT_SRC_NORMAL)\n");
9275 /* return '1' if done */
9283 #endif /* MDASSEMBLE */
9285 struct superswitch super_imsm
= {
9287 .examine_super
= examine_super_imsm
,
9288 .brief_examine_super
= brief_examine_super_imsm
,
9289 .brief_examine_subarrays
= brief_examine_subarrays_imsm
,
9290 .export_examine_super
= export_examine_super_imsm
,
9291 .detail_super
= detail_super_imsm
,
9292 .brief_detail_super
= brief_detail_super_imsm
,
9293 .write_init_super
= write_init_super_imsm
,
9294 .validate_geometry
= validate_geometry_imsm
,
9295 .add_to_super
= add_to_super_imsm
,
9296 .remove_from_super
= remove_from_super_imsm
,
9297 .detail_platform
= detail_platform_imsm
,
9298 .kill_subarray
= kill_subarray_imsm
,
9299 .update_subarray
= update_subarray_imsm
,
9300 .load_container
= load_container_imsm
,
9301 .default_geometry
= default_geometry_imsm
,
9302 .get_disk_controller_domain
= imsm_get_disk_controller_domain
,
9303 .reshape_super
= imsm_reshape_super
,
9304 .manage_reshape
= imsm_manage_reshape
,
9305 .recover_backup
= recover_backup_imsm
,
9307 .match_home
= match_home_imsm
,
9308 .uuid_from_super
= uuid_from_super_imsm
,
9309 .getinfo_super
= getinfo_super_imsm
,
9310 .getinfo_super_disks
= getinfo_super_disks_imsm
,
9311 .update_super
= update_super_imsm
,
9313 .avail_size
= avail_size_imsm
,
9314 .min_acceptable_spare_size
= min_acceptable_spare_size_imsm
,
9316 .compare_super
= compare_super_imsm
,
9318 .load_super
= load_super_imsm
,
9319 .init_super
= init_super_imsm
,
9320 .store_super
= store_super_imsm
,
9321 .free_super
= free_super_imsm
,
9322 .match_metadata_desc
= match_metadata_desc_imsm
,
9323 .container_content
= container_content_imsm
,
9331 .open_new
= imsm_open_new
,
9332 .set_array_state
= imsm_set_array_state
,
9333 .set_disk
= imsm_set_disk
,
9334 .sync_metadata
= imsm_sync_metadata
,
9335 .activate_spare
= imsm_activate_spare
,
9336 .process_update
= imsm_process_update
,
9337 .prepare_update
= imsm_prepare_update
,
9338 #endif /* MDASSEMBLE */