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
, dev
->vol
.migr_state
);
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
, -1);
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
, '/');
2872 snprintf(name
, MAX_RAID_SERIAL_LEN
, "/dev/%s", nm
);
2875 extern int scsi_get_serial(int fd
, void *buf
, size_t buf_len
);
2877 static int imsm_read_serial(int fd
, char *devname
,
2878 __u8 serial
[MAX_RAID_SERIAL_LEN
])
2880 unsigned char scsi_serial
[255];
2889 memset(scsi_serial
, 0, sizeof(scsi_serial
));
2891 rv
= scsi_get_serial(fd
, scsi_serial
, sizeof(scsi_serial
));
2893 if (rv
&& check_env("IMSM_DEVNAME_AS_SERIAL")) {
2894 memset(serial
, 0, MAX_RAID_SERIAL_LEN
);
2895 fd2devname(fd
, (char *) serial
);
2902 Name
": Failed to retrieve serial for %s\n",
2907 rsp_len
= scsi_serial
[3];
2911 Name
": Failed to retrieve serial for %s\n",
2915 rsp_buf
= (char *) &scsi_serial
[4];
2917 /* trim all whitespace and non-printable characters and convert
2920 for (i
= 0, dest
= rsp_buf
; i
< rsp_len
; i
++) {
2923 /* ':' is reserved for use in placeholder serial
2924 * numbers for missing disks
2932 len
= dest
- rsp_buf
;
2935 /* truncate leading characters */
2936 if (len
> MAX_RAID_SERIAL_LEN
) {
2937 dest
+= len
- MAX_RAID_SERIAL_LEN
;
2938 len
= MAX_RAID_SERIAL_LEN
;
2941 memset(serial
, 0, MAX_RAID_SERIAL_LEN
);
2942 memcpy(serial
, dest
, len
);
2947 static int serialcmp(__u8
*s1
, __u8
*s2
)
2949 return strncmp((char *) s1
, (char *) s2
, MAX_RAID_SERIAL_LEN
);
2952 static void serialcpy(__u8
*dest
, __u8
*src
)
2954 strncpy((char *) dest
, (char *) src
, MAX_RAID_SERIAL_LEN
);
2957 static struct dl
*serial_to_dl(__u8
*serial
, struct intel_super
*super
)
2961 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
2962 if (serialcmp(dl
->serial
, serial
) == 0)
2968 static struct imsm_disk
*
2969 __serial_to_disk(__u8
*serial
, struct imsm_super
*mpb
, int *idx
)
2973 for (i
= 0; i
< mpb
->num_disks
; i
++) {
2974 struct imsm_disk
*disk
= __get_imsm_disk(mpb
, i
);
2976 if (serialcmp(disk
->serial
, serial
) == 0) {
2987 load_imsm_disk(int fd
, struct intel_super
*super
, char *devname
, int keep_fd
)
2989 struct imsm_disk
*disk
;
2994 __u8 serial
[MAX_RAID_SERIAL_LEN
];
2996 rv
= imsm_read_serial(fd
, devname
, serial
);
3001 dl
= calloc(1, sizeof(*dl
));
3005 Name
": failed to allocate disk buffer for %s\n",
3011 dl
->major
= major(stb
.st_rdev
);
3012 dl
->minor
= minor(stb
.st_rdev
);
3013 dl
->next
= super
->disks
;
3014 dl
->fd
= keep_fd
? fd
: -1;
3015 assert(super
->disks
== NULL
);
3017 serialcpy(dl
->serial
, serial
);
3020 fd2devname(fd
, name
);
3022 dl
->devname
= strdup(devname
);
3024 dl
->devname
= strdup(name
);
3026 /* look up this disk's index in the current anchor */
3027 disk
= __serial_to_disk(dl
->serial
, super
->anchor
, &dl
->index
);
3030 /* only set index on disks that are a member of a
3031 * populated contianer, i.e. one with raid_devs
3033 if (is_failed(&dl
->disk
))
3035 else if (is_spare(&dl
->disk
))
3043 /* When migrating map0 contains the 'destination' state while map1
3044 * contains the current state. When not migrating map0 contains the
3045 * current state. This routine assumes that map[0].map_state is set to
3046 * the current array state before being called.
3048 * Migration is indicated by one of the following states
3049 * 1/ Idle (migr_state=0 map0state=normal||unitialized||degraded||failed)
3050 * 2/ Initialize (migr_state=1 migr_type=MIGR_INIT map0state=normal
3051 * map1state=unitialized)
3052 * 3/ Repair (Resync) (migr_state=1 migr_type=MIGR_REPAIR map0state=normal
3054 * 4/ Rebuild (migr_state=1 migr_type=MIGR_REBUILD map0state=normal
3055 * map1state=degraded)
3056 * 5/ Migration (mig_state=1 migr_type=MIGR_GEN_MIGR map0state=normal
3059 static void migrate(struct imsm_dev
*dev
, struct intel_super
*super
,
3060 __u8 to_state
, int migr_type
)
3062 struct imsm_map
*dest
;
3063 struct imsm_map
*src
= get_imsm_map(dev
, 0);
3065 dev
->vol
.migr_state
= 1;
3066 set_migr_type(dev
, migr_type
);
3067 dev
->vol
.curr_migr_unit
= 0;
3068 dest
= get_imsm_map(dev
, 1);
3070 /* duplicate and then set the target end state in map[0] */
3071 memcpy(dest
, src
, sizeof_imsm_map(src
));
3072 if ((migr_type
== MIGR_REBUILD
) ||
3073 (migr_type
== MIGR_GEN_MIGR
)) {
3077 for (i
= 0; i
< src
->num_members
; i
++) {
3078 ord
= __le32_to_cpu(src
->disk_ord_tbl
[i
]);
3079 set_imsm_ord_tbl_ent(src
, i
, ord_to_idx(ord
));
3083 if (migr_type
== MIGR_GEN_MIGR
)
3084 /* Clear migration record */
3085 memset(super
->migr_rec
, 0, sizeof(struct migr_record
));
3087 src
->map_state
= to_state
;
3090 static void end_migration(struct imsm_dev
*dev
, __u8 map_state
)
3092 struct imsm_map
*map
= get_imsm_map(dev
, 0);
3093 struct imsm_map
*prev
= get_imsm_map(dev
, dev
->vol
.migr_state
);
3096 /* merge any IMSM_ORD_REBUILD bits that were not successfully
3097 * completed in the last migration.
3099 * FIXME add support for raid-level-migration
3101 for (i
= 0; i
< prev
->num_members
; i
++)
3102 for (j
= 0; j
< map
->num_members
; j
++)
3103 /* during online capacity expansion
3104 * disks position can be changed if takeover is used
3106 if (ord_to_idx(map
->disk_ord_tbl
[j
]) ==
3107 ord_to_idx(prev
->disk_ord_tbl
[i
])) {
3108 map
->disk_ord_tbl
[j
] |= prev
->disk_ord_tbl
[i
];
3112 dev
->vol
.migr_state
= 0;
3113 set_migr_type(dev
, 0);
3114 dev
->vol
.curr_migr_unit
= 0;
3115 map
->map_state
= map_state
;
3119 static int parse_raid_devices(struct intel_super
*super
)
3122 struct imsm_dev
*dev_new
;
3123 size_t len
, len_migr
;
3125 size_t space_needed
= 0;
3126 struct imsm_super
*mpb
= super
->anchor
;
3128 for (i
= 0; i
< super
->anchor
->num_raid_devs
; i
++) {
3129 struct imsm_dev
*dev_iter
= __get_imsm_dev(super
->anchor
, i
);
3130 struct intel_dev
*dv
;
3132 len
= sizeof_imsm_dev(dev_iter
, 0);
3133 len_migr
= sizeof_imsm_dev(dev_iter
, 1);
3135 space_needed
+= len_migr
- len
;
3137 dv
= malloc(sizeof(*dv
));
3140 if (max_len
< len_migr
)
3142 if (max_len
> len_migr
)
3143 space_needed
+= max_len
- len_migr
;
3144 dev_new
= malloc(max_len
);
3149 imsm_copy_dev(dev_new
, dev_iter
);
3152 dv
->next
= super
->devlist
;
3153 super
->devlist
= dv
;
3156 /* ensure that super->buf is large enough when all raid devices
3159 if (__le32_to_cpu(mpb
->mpb_size
) + space_needed
> super
->len
) {
3162 len
= ROUND_UP(__le32_to_cpu(mpb
->mpb_size
) + space_needed
, 512);
3163 if (posix_memalign(&buf
, 512, len
) != 0)
3166 memcpy(buf
, super
->buf
, super
->len
);
3167 memset(buf
+ super
->len
, 0, len
- super
->len
);
3176 /* retrieve a pointer to the bbm log which starts after all raid devices */
3177 struct bbm_log
*__get_imsm_bbm_log(struct imsm_super
*mpb
)
3181 if (__le32_to_cpu(mpb
->bbm_log_size
)) {
3183 ptr
+= mpb
->mpb_size
- __le32_to_cpu(mpb
->bbm_log_size
);
3189 /*******************************************************************************
3190 * Function: check_mpb_migr_compatibility
3191 * Description: Function checks for unsupported migration features:
3192 * - migration optimization area (pba_of_lba0)
3193 * - descending reshape (ascending_migr)
3195 * super : imsm metadata information
3197 * 0 : migration is compatible
3198 * -1 : migration is not compatible
3199 ******************************************************************************/
3200 int check_mpb_migr_compatibility(struct intel_super
*super
)
3202 struct imsm_map
*map0
, *map1
;
3203 struct migr_record
*migr_rec
= super
->migr_rec
;
3206 for (i
= 0; i
< super
->anchor
->num_raid_devs
; i
++) {
3207 struct imsm_dev
*dev_iter
= __get_imsm_dev(super
->anchor
, i
);
3210 dev_iter
->vol
.migr_state
== 1 &&
3211 dev_iter
->vol
.migr_type
== MIGR_GEN_MIGR
) {
3212 /* This device is migrating */
3213 map0
= get_imsm_map(dev_iter
, 0);
3214 map1
= get_imsm_map(dev_iter
, 1);
3215 if (map0
->pba_of_lba0
!= map1
->pba_of_lba0
)
3216 /* migration optimization area was used */
3218 if (migr_rec
->ascending_migr
== 0
3219 && migr_rec
->dest_depth_per_unit
> 0)
3220 /* descending reshape not supported yet */
3227 static void __free_imsm(struct intel_super
*super
, int free_disks
);
3229 /* load_imsm_mpb - read matrix metadata
3230 * allocates super->mpb to be freed by free_imsm
3232 static int load_imsm_mpb(int fd
, struct intel_super
*super
, char *devname
)
3234 unsigned long long dsize
;
3235 unsigned long long sectors
;
3237 struct imsm_super
*anchor
;
3240 get_dev_size(fd
, NULL
, &dsize
);
3244 Name
": %s: device to small for imsm\n",
3249 if (lseek64(fd
, dsize
- (512 * 2), SEEK_SET
) < 0) {
3251 fprintf(stderr
, Name
3252 ": Cannot seek to anchor block on %s: %s\n",
3253 devname
, strerror(errno
));
3257 if (posix_memalign((void**)&anchor
, 512, 512) != 0) {
3260 Name
": Failed to allocate imsm anchor buffer"
3261 " on %s\n", devname
);
3264 if (read(fd
, anchor
, 512) != 512) {
3267 Name
": Cannot read anchor block on %s: %s\n",
3268 devname
, strerror(errno
));
3273 if (strncmp((char *) anchor
->sig
, MPB_SIGNATURE
, MPB_SIG_LEN
) != 0) {
3276 Name
": no IMSM anchor on %s\n", devname
);
3281 __free_imsm(super
, 0);
3282 /* reload capability and hba */
3284 /* capability and hba must be updated with new super allocation */
3285 find_intel_hba_capability(fd
, super
, devname
);
3286 super
->len
= ROUND_UP(anchor
->mpb_size
, 512);
3287 if (posix_memalign(&super
->buf
, 512, super
->len
) != 0) {
3290 Name
": unable to allocate %zu byte mpb buffer\n",
3295 memcpy(super
->buf
, anchor
, 512);
3297 sectors
= mpb_sectors(anchor
) - 1;
3300 if (posix_memalign(&super
->migr_rec_buf
, 512, 512) != 0) {
3301 fprintf(stderr
, Name
3302 ": %s could not allocate migr_rec buffer\n", __func__
);
3308 check_sum
= __gen_imsm_checksum(super
->anchor
);
3309 if (check_sum
!= __le32_to_cpu(super
->anchor
->check_sum
)) {
3312 Name
": IMSM checksum %x != %x on %s\n",
3314 __le32_to_cpu(super
->anchor
->check_sum
),
3322 /* read the extended mpb */
3323 if (lseek64(fd
, dsize
- (512 * (2 + sectors
)), SEEK_SET
) < 0) {
3326 Name
": Cannot seek to extended mpb on %s: %s\n",
3327 devname
, strerror(errno
));
3331 if ((unsigned)read(fd
, super
->buf
+ 512, super
->len
- 512) != super
->len
- 512) {
3334 Name
": Cannot read extended mpb on %s: %s\n",
3335 devname
, strerror(errno
));
3339 check_sum
= __gen_imsm_checksum(super
->anchor
);
3340 if (check_sum
!= __le32_to_cpu(super
->anchor
->check_sum
)) {
3343 Name
": IMSM checksum %x != %x on %s\n",
3344 check_sum
, __le32_to_cpu(super
->anchor
->check_sum
),
3349 /* FIXME the BBM log is disk specific so we cannot use this global
3350 * buffer for all disks. Ok for now since we only look at the global
3351 * bbm_log_size parameter to gate assembly
3353 super
->bbm_log
= __get_imsm_bbm_log(super
->anchor
);
3358 static int read_imsm_migr_rec(int fd
, struct intel_super
*super
);
3361 load_and_parse_mpb(int fd
, struct intel_super
*super
, char *devname
, int keep_fd
)
3365 err
= load_imsm_mpb(fd
, super
, devname
);
3368 err
= load_imsm_disk(fd
, super
, devname
, keep_fd
);
3371 err
= parse_raid_devices(super
);
3376 static void __free_imsm_disk(struct dl
*d
)
3388 static void free_imsm_disks(struct intel_super
*super
)
3392 while (super
->disks
) {
3394 super
->disks
= d
->next
;
3395 __free_imsm_disk(d
);
3397 while (super
->disk_mgmt_list
) {
3398 d
= super
->disk_mgmt_list
;
3399 super
->disk_mgmt_list
= d
->next
;
3400 __free_imsm_disk(d
);
3402 while (super
->missing
) {
3404 super
->missing
= d
->next
;
3405 __free_imsm_disk(d
);
3410 /* free all the pieces hanging off of a super pointer */
3411 static void __free_imsm(struct intel_super
*super
, int free_disks
)
3413 struct intel_hba
*elem
, *next
;
3419 /* unlink capability description */
3421 if (super
->migr_rec_buf
) {
3422 free(super
->migr_rec_buf
);
3423 super
->migr_rec_buf
= NULL
;
3426 free_imsm_disks(super
);
3427 free_devlist(super
);
3431 free((void *)elem
->path
);
3439 static void free_imsm(struct intel_super
*super
)
3441 __free_imsm(super
, 1);
3445 static void free_super_imsm(struct supertype
*st
)
3447 struct intel_super
*super
= st
->sb
;
3456 static struct intel_super
*alloc_super(void)
3458 struct intel_super
*super
= malloc(sizeof(*super
));
3461 memset(super
, 0, sizeof(*super
));
3462 super
->current_vol
= -1;
3463 super
->create_offset
= ~((__u32
) 0);
3469 * find and allocate hba and OROM/EFI based on valid fd of RAID component device
3471 static int find_intel_hba_capability(int fd
, struct intel_super
*super
, char *devname
)
3473 struct sys_dev
*hba_name
;
3476 if ((fd
< 0) || check_env("IMSM_NO_PLATFORM")) {
3481 hba_name
= find_disk_attached_hba(fd
, NULL
);
3485 Name
": %s is not attached to Intel(R) RAID controller.\n",
3489 rv
= attach_hba_to_super(super
, hba_name
);
3492 struct intel_hba
*hba
= super
->hba
;
3494 fprintf(stderr
, Name
": %s is attached to Intel(R) %s RAID "
3495 "controller (%s),\n"
3496 " but the container is assigned to Intel(R) "
3497 "%s RAID controller (",
3500 hba_name
->pci_id
? : "Err!",
3501 get_sys_dev_type(hba_name
->type
));
3504 fprintf(stderr
, "%s", hba
->pci_id
? : "Err!");
3506 fprintf(stderr
, ", ");
3510 fprintf(stderr
, ").\n"
3511 " Mixing devices attached to different controllers "
3512 "is not allowed.\n");
3514 free_sys_dev(&hba_name
);
3517 super
->orom
= find_imsm_capability(hba_name
->type
);
3518 free_sys_dev(&hba_name
);
3524 /* find_missing - helper routine for load_super_imsm_all that identifies
3525 * disks that have disappeared from the system. This routine relies on
3526 * the mpb being uptodate, which it is at load time.
3528 static int find_missing(struct intel_super
*super
)
3531 struct imsm_super
*mpb
= super
->anchor
;
3533 struct imsm_disk
*disk
;
3535 for (i
= 0; i
< mpb
->num_disks
; i
++) {
3536 disk
= __get_imsm_disk(mpb
, i
);
3537 dl
= serial_to_dl(disk
->serial
, super
);
3541 dl
= malloc(sizeof(*dl
));
3547 dl
->devname
= strdup("missing");
3549 serialcpy(dl
->serial
, disk
->serial
);
3552 dl
->next
= super
->missing
;
3553 super
->missing
= dl
;
3560 static struct intel_disk
*disk_list_get(__u8
*serial
, struct intel_disk
*disk_list
)
3562 struct intel_disk
*idisk
= disk_list
;
3565 if (serialcmp(idisk
->disk
.serial
, serial
) == 0)
3567 idisk
= idisk
->next
;
3573 static int __prep_thunderdome(struct intel_super
**table
, int tbl_size
,
3574 struct intel_super
*super
,
3575 struct intel_disk
**disk_list
)
3577 struct imsm_disk
*d
= &super
->disks
->disk
;
3578 struct imsm_super
*mpb
= super
->anchor
;
3581 for (i
= 0; i
< tbl_size
; i
++) {
3582 struct imsm_super
*tbl_mpb
= table
[i
]->anchor
;
3583 struct imsm_disk
*tbl_d
= &table
[i
]->disks
->disk
;
3585 if (tbl_mpb
->family_num
== mpb
->family_num
) {
3586 if (tbl_mpb
->check_sum
== mpb
->check_sum
) {
3587 dprintf("%s: mpb from %d:%d matches %d:%d\n",
3588 __func__
, super
->disks
->major
,
3589 super
->disks
->minor
,
3590 table
[i
]->disks
->major
,
3591 table
[i
]->disks
->minor
);
3595 if (((is_configured(d
) && !is_configured(tbl_d
)) ||
3596 is_configured(d
) == is_configured(tbl_d
)) &&
3597 tbl_mpb
->generation_num
< mpb
->generation_num
) {
3598 /* current version of the mpb is a
3599 * better candidate than the one in
3600 * super_table, but copy over "cross
3601 * generational" status
3603 struct intel_disk
*idisk
;
3605 dprintf("%s: mpb from %d:%d replaces %d:%d\n",
3606 __func__
, super
->disks
->major
,
3607 super
->disks
->minor
,
3608 table
[i
]->disks
->major
,
3609 table
[i
]->disks
->minor
);
3611 idisk
= disk_list_get(tbl_d
->serial
, *disk_list
);
3612 if (idisk
&& is_failed(&idisk
->disk
))
3613 tbl_d
->status
|= FAILED_DISK
;
3616 struct intel_disk
*idisk
;
3617 struct imsm_disk
*disk
;
3619 /* tbl_mpb is more up to date, but copy
3620 * over cross generational status before
3623 disk
= __serial_to_disk(d
->serial
, mpb
, NULL
);
3624 if (disk
&& is_failed(disk
))
3625 d
->status
|= FAILED_DISK
;
3627 idisk
= disk_list_get(d
->serial
, *disk_list
);
3630 if (disk
&& is_configured(disk
))
3631 idisk
->disk
.status
|= CONFIGURED_DISK
;
3634 dprintf("%s: mpb from %d:%d prefer %d:%d\n",
3635 __func__
, super
->disks
->major
,
3636 super
->disks
->minor
,
3637 table
[i
]->disks
->major
,
3638 table
[i
]->disks
->minor
);
3646 table
[tbl_size
++] = super
;
3650 /* update/extend the merged list of imsm_disk records */
3651 for (j
= 0; j
< mpb
->num_disks
; j
++) {
3652 struct imsm_disk
*disk
= __get_imsm_disk(mpb
, j
);
3653 struct intel_disk
*idisk
;
3655 idisk
= disk_list_get(disk
->serial
, *disk_list
);
3657 idisk
->disk
.status
|= disk
->status
;
3658 if (is_configured(&idisk
->disk
) ||
3659 is_failed(&idisk
->disk
))
3660 idisk
->disk
.status
&= ~(SPARE_DISK
);
3662 idisk
= calloc(1, sizeof(*idisk
));
3665 idisk
->owner
= IMSM_UNKNOWN_OWNER
;
3666 idisk
->disk
= *disk
;
3667 idisk
->next
= *disk_list
;
3671 if (serialcmp(idisk
->disk
.serial
, d
->serial
) == 0)
3678 static struct intel_super
*
3679 validate_members(struct intel_super
*super
, struct intel_disk
*disk_list
,
3682 struct imsm_super
*mpb
= super
->anchor
;
3686 for (i
= 0; i
< mpb
->num_disks
; i
++) {
3687 struct imsm_disk
*disk
= __get_imsm_disk(mpb
, i
);
3688 struct intel_disk
*idisk
;
3690 idisk
= disk_list_get(disk
->serial
, disk_list
);
3692 if (idisk
->owner
== owner
||
3693 idisk
->owner
== IMSM_UNKNOWN_OWNER
)
3696 dprintf("%s: '%.16s' owner %d != %d\n",
3697 __func__
, disk
->serial
, idisk
->owner
,
3700 dprintf("%s: unknown disk %x [%d]: %.16s\n",
3701 __func__
, __le32_to_cpu(mpb
->family_num
), i
,
3707 if (ok_count
== mpb
->num_disks
)
3712 static void show_conflicts(__u32 family_num
, struct intel_super
*super_list
)
3714 struct intel_super
*s
;
3716 for (s
= super_list
; s
; s
= s
->next
) {
3717 if (family_num
!= s
->anchor
->family_num
)
3719 fprintf(stderr
, "Conflict, offlining family %#x on '%s'\n",
3720 __le32_to_cpu(family_num
), s
->disks
->devname
);
3724 static struct intel_super
*
3725 imsm_thunderdome(struct intel_super
**super_list
, int len
)
3727 struct intel_super
*super_table
[len
];
3728 struct intel_disk
*disk_list
= NULL
;
3729 struct intel_super
*champion
, *spare
;
3730 struct intel_super
*s
, **del
;
3735 memset(super_table
, 0, sizeof(super_table
));
3736 for (s
= *super_list
; s
; s
= s
->next
)
3737 tbl_size
= __prep_thunderdome(super_table
, tbl_size
, s
, &disk_list
);
3739 for (i
= 0; i
< tbl_size
; i
++) {
3740 struct imsm_disk
*d
;
3741 struct intel_disk
*idisk
;
3742 struct imsm_super
*mpb
= super_table
[i
]->anchor
;
3745 d
= &s
->disks
->disk
;
3747 /* 'd' must appear in merged disk list for its
3748 * configuration to be valid
3750 idisk
= disk_list_get(d
->serial
, disk_list
);
3751 if (idisk
&& idisk
->owner
== i
)
3752 s
= validate_members(s
, disk_list
, i
);
3757 dprintf("%s: marking family: %#x from %d:%d offline\n",
3758 __func__
, mpb
->family_num
,
3759 super_table
[i
]->disks
->major
,
3760 super_table
[i
]->disks
->minor
);
3764 /* This is where the mdadm implementation differs from the Windows
3765 * driver which has no strict concept of a container. We can only
3766 * assemble one family from a container, so when returning a prodigal
3767 * array member to this system the code will not be able to disambiguate
3768 * the container contents that should be assembled ("foreign" versus
3769 * "local"). It requires user intervention to set the orig_family_num
3770 * to a new value to establish a new container. The Windows driver in
3771 * this situation fixes up the volume name in place and manages the
3772 * foreign array as an independent entity.
3777 for (i
= 0; i
< tbl_size
; i
++) {
3778 struct intel_super
*tbl_ent
= super_table
[i
];
3784 if (tbl_ent
->anchor
->num_raid_devs
== 0) {
3789 if (s
&& !is_spare
) {
3790 show_conflicts(tbl_ent
->anchor
->family_num
, *super_list
);
3792 } else if (!s
&& !is_spare
)
3805 fprintf(stderr
, "Chose family %#x on '%s', "
3806 "assemble conflicts to new container with '--update=uuid'\n",
3807 __le32_to_cpu(s
->anchor
->family_num
), s
->disks
->devname
);
3809 /* collect all dl's onto 'champion', and update them to
3810 * champion's version of the status
3812 for (s
= *super_list
; s
; s
= s
->next
) {
3813 struct imsm_super
*mpb
= champion
->anchor
;
3814 struct dl
*dl
= s
->disks
;
3819 for (i
= 0; i
< mpb
->num_disks
; i
++) {
3820 struct imsm_disk
*disk
;
3822 disk
= __serial_to_disk(dl
->serial
, mpb
, &dl
->index
);
3825 /* only set index on disks that are a member of
3826 * a populated contianer, i.e. one with
3829 if (is_failed(&dl
->disk
))
3831 else if (is_spare(&dl
->disk
))
3837 if (i
>= mpb
->num_disks
) {
3838 struct intel_disk
*idisk
;
3840 idisk
= disk_list_get(dl
->serial
, disk_list
);
3841 if (idisk
&& is_spare(&idisk
->disk
) &&
3842 !is_failed(&idisk
->disk
) && !is_configured(&idisk
->disk
))
3850 dl
->next
= champion
->disks
;
3851 champion
->disks
= dl
;
3855 /* delete 'champion' from super_list */
3856 for (del
= super_list
; *del
; ) {
3857 if (*del
== champion
) {
3858 *del
= (*del
)->next
;
3861 del
= &(*del
)->next
;
3863 champion
->next
= NULL
;
3867 struct intel_disk
*idisk
= disk_list
;
3869 disk_list
= disk_list
->next
;
3876 static int load_super_imsm_all(struct supertype
*st
, int fd
, void **sbp
,
3880 struct intel_super
*super_list
= NULL
;
3881 struct intel_super
*super
= NULL
;
3882 int devnum
= fd2devnum(fd
);
3888 /* check if 'fd' an opened container */
3889 sra
= sysfs_read(fd
, 0, GET_LEVEL
|GET_VERSION
|GET_DEVS
|GET_STATE
);
3893 if (sra
->array
.major_version
!= -1 ||
3894 sra
->array
.minor_version
!= -2 ||
3895 strcmp(sra
->text_version
, "imsm") != 0) {
3900 for (sd
= sra
->devs
, i
= 0; sd
; sd
= sd
->next
, i
++) {
3901 struct intel_super
*s
= alloc_super();
3909 s
->next
= super_list
;
3913 sprintf(nm
, "%d:%d", sd
->disk
.major
, sd
->disk
.minor
);
3914 dfd
= dev_open(nm
, O_RDWR
);
3918 rv
= find_intel_hba_capability(dfd
, s
, devname
);
3919 /* no orom/efi or non-intel hba of the disk */
3923 err
= load_and_parse_mpb(dfd
, s
, NULL
, 1);
3925 /* retry the load if we might have raced against mdmon */
3926 if (err
== 3 && mdmon_running(devnum
))
3927 for (retry
= 0; retry
< 3; retry
++) {
3929 err
= load_and_parse_mpb(dfd
, s
, NULL
, 1);
3937 /* all mpbs enter, maybe one leaves */
3938 super
= imsm_thunderdome(&super_list
, i
);
3944 if (find_missing(super
) != 0) {
3950 /* load migration record */
3951 err
= load_imsm_migr_rec(super
, NULL
);
3957 /* Check migration compatibility */
3958 if (check_mpb_migr_compatibility(super
) != 0) {
3959 fprintf(stderr
, Name
": Unsupported migration detected");
3961 fprintf(stderr
, " on %s\n", devname
);
3963 fprintf(stderr
, " (IMSM).\n");
3972 while (super_list
) {
3973 struct intel_super
*s
= super_list
;
3975 super_list
= super_list
->next
;
3984 st
->container_dev
= devnum
;
3985 if (err
== 0 && st
->ss
== NULL
) {
3986 st
->ss
= &super_imsm
;
3987 st
->minor_version
= 0;
3988 st
->max_devs
= IMSM_MAX_DEVICES
;
3993 static int load_container_imsm(struct supertype
*st
, int fd
, char *devname
)
3995 return load_super_imsm_all(st
, fd
, &st
->sb
, devname
);
3999 static int load_super_imsm(struct supertype
*st
, int fd
, char *devname
)
4001 struct intel_super
*super
;
4004 if (test_partition(fd
))
4005 /* IMSM not allowed on partitions */
4008 free_super_imsm(st
);
4010 super
= alloc_super();
4013 Name
": malloc of %zu failed.\n",
4017 /* Load hba and capabilities if they exist.
4018 * But do not preclude loading metadata in case capabilities or hba are
4019 * non-compliant and ignore_hw_compat is set.
4021 rv
= find_intel_hba_capability(fd
, super
, devname
);
4022 /* no orom/efi or non-intel hba of the disk */
4023 if ((rv
!= 0) && (st
->ignore_hw_compat
== 0)) {
4026 Name
": No OROM/EFI properties for %s\n", devname
);
4030 rv
= load_and_parse_mpb(fd
, super
, devname
, 0);
4035 Name
": Failed to load all information "
4036 "sections on %s\n", devname
);
4042 if (st
->ss
== NULL
) {
4043 st
->ss
= &super_imsm
;
4044 st
->minor_version
= 0;
4045 st
->max_devs
= IMSM_MAX_DEVICES
;
4048 /* load migration record */
4049 if (load_imsm_migr_rec(super
, NULL
) == 0) {
4050 /* Check for unsupported migration features */
4051 if (check_mpb_migr_compatibility(super
) != 0) {
4053 Name
": Unsupported migration detected");
4055 fprintf(stderr
, " on %s\n", devname
);
4057 fprintf(stderr
, " (IMSM).\n");
4065 static __u16
info_to_blocks_per_strip(mdu_array_info_t
*info
)
4067 if (info
->level
== 1)
4069 return info
->chunk_size
>> 9;
4072 static __u32
info_to_num_data_stripes(mdu_array_info_t
*info
, int num_domains
)
4076 num_stripes
= (info
->size
* 2) / info_to_blocks_per_strip(info
);
4077 num_stripes
/= num_domains
;
4082 static __u32
info_to_blocks_per_member(mdu_array_info_t
*info
)
4084 if (info
->level
== 1)
4085 return info
->size
* 2;
4087 return (info
->size
* 2) & ~(info_to_blocks_per_strip(info
) - 1);
4090 static void imsm_update_version_info(struct intel_super
*super
)
4092 /* update the version and attributes */
4093 struct imsm_super
*mpb
= super
->anchor
;
4095 struct imsm_dev
*dev
;
4096 struct imsm_map
*map
;
4099 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
4100 dev
= get_imsm_dev(super
, i
);
4101 map
= get_imsm_map(dev
, 0);
4102 if (__le32_to_cpu(dev
->size_high
) > 0)
4103 mpb
->attributes
|= MPB_ATTRIB_2TB
;
4105 /* FIXME detect when an array spans a port multiplier */
4107 mpb
->attributes
|= MPB_ATTRIB_PM
;
4110 if (mpb
->num_raid_devs
> 1 ||
4111 mpb
->attributes
!= MPB_ATTRIB_CHECKSUM_VERIFY
) {
4112 version
= MPB_VERSION_ATTRIBS
;
4113 switch (get_imsm_raid_level(map
)) {
4114 case 0: mpb
->attributes
|= MPB_ATTRIB_RAID0
; break;
4115 case 1: mpb
->attributes
|= MPB_ATTRIB_RAID1
; break;
4116 case 10: mpb
->attributes
|= MPB_ATTRIB_RAID10
; break;
4117 case 5: mpb
->attributes
|= MPB_ATTRIB_RAID5
; break;
4120 if (map
->num_members
>= 5)
4121 version
= MPB_VERSION_5OR6_DISK_ARRAY
;
4122 else if (dev
->status
== DEV_CLONE_N_GO
)
4123 version
= MPB_VERSION_CNG
;
4124 else if (get_imsm_raid_level(map
) == 5)
4125 version
= MPB_VERSION_RAID5
;
4126 else if (map
->num_members
>= 3)
4127 version
= MPB_VERSION_3OR4_DISK_ARRAY
;
4128 else if (get_imsm_raid_level(map
) == 1)
4129 version
= MPB_VERSION_RAID1
;
4131 version
= MPB_VERSION_RAID0
;
4133 strcpy(((char *) mpb
->sig
) + strlen(MPB_SIGNATURE
), version
);
4137 static int check_name(struct intel_super
*super
, char *name
, int quiet
)
4139 struct imsm_super
*mpb
= super
->anchor
;
4140 char *reason
= NULL
;
4143 if (strlen(name
) > MAX_RAID_SERIAL_LEN
)
4144 reason
= "must be 16 characters or less";
4146 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
4147 struct imsm_dev
*dev
= get_imsm_dev(super
, i
);
4149 if (strncmp((char *) dev
->volume
, name
, MAX_RAID_SERIAL_LEN
) == 0) {
4150 reason
= "already exists";
4155 if (reason
&& !quiet
)
4156 fprintf(stderr
, Name
": imsm volume name %s\n", reason
);
4161 static int init_super_imsm_volume(struct supertype
*st
, mdu_array_info_t
*info
,
4162 unsigned long long size
, char *name
,
4163 char *homehost
, int *uuid
)
4165 /* We are creating a volume inside a pre-existing container.
4166 * so st->sb is already set.
4168 struct intel_super
*super
= st
->sb
;
4169 struct imsm_super
*mpb
= super
->anchor
;
4170 struct intel_dev
*dv
;
4171 struct imsm_dev
*dev
;
4172 struct imsm_vol
*vol
;
4173 struct imsm_map
*map
;
4174 int idx
= mpb
->num_raid_devs
;
4176 unsigned long long array_blocks
;
4177 size_t size_old
, size_new
;
4178 __u32 num_data_stripes
;
4180 if (super
->orom
&& mpb
->num_raid_devs
>= super
->orom
->vpa
) {
4181 fprintf(stderr
, Name
": This imsm-container already has the "
4182 "maximum of %d volumes\n", super
->orom
->vpa
);
4186 /* ensure the mpb is large enough for the new data */
4187 size_old
= __le32_to_cpu(mpb
->mpb_size
);
4188 size_new
= disks_to_mpb_size(info
->nr_disks
);
4189 if (size_new
> size_old
) {
4191 size_t size_round
= ROUND_UP(size_new
, 512);
4193 if (posix_memalign(&mpb_new
, 512, size_round
) != 0) {
4194 fprintf(stderr
, Name
": could not allocate new mpb\n");
4197 if (posix_memalign(&super
->migr_rec_buf
, 512, 512) != 0) {
4198 fprintf(stderr
, Name
4199 ": %s could not allocate migr_rec buffer\n",
4206 memcpy(mpb_new
, mpb
, size_old
);
4209 super
->anchor
= mpb_new
;
4210 mpb
->mpb_size
= __cpu_to_le32(size_new
);
4211 memset(mpb_new
+ size_old
, 0, size_round
- size_old
);
4213 super
->current_vol
= idx
;
4215 /* handle 'failed_disks' by either:
4216 * a) create dummy disk entries in the table if this the first
4217 * volume in the array. We add them here as this is the only
4218 * opportunity to add them. add_to_super_imsm_volume()
4219 * handles the non-failed disks and continues incrementing
4221 * b) validate that 'failed_disks' matches the current number
4222 * of missing disks if the container is populated
4224 if (super
->current_vol
== 0) {
4226 for (i
= 0; i
< info
->failed_disks
; i
++) {
4227 struct imsm_disk
*disk
;
4230 disk
= __get_imsm_disk(mpb
, i
);
4231 disk
->status
= CONFIGURED_DISK
| FAILED_DISK
;
4232 disk
->scsi_id
= __cpu_to_le32(~(__u32
)0);
4233 snprintf((char *) disk
->serial
, MAX_RAID_SERIAL_LEN
,
4236 find_missing(super
);
4241 for (d
= super
->missing
; d
; d
= d
->next
)
4243 if (info
->failed_disks
> missing
) {
4244 fprintf(stderr
, Name
": unable to add 'missing' disk to container\n");
4249 if (!check_name(super
, name
, 0))
4251 dv
= malloc(sizeof(*dv
));
4253 fprintf(stderr
, Name
": failed to allocate device list entry\n");
4256 dev
= calloc(1, sizeof(*dev
) + sizeof(__u32
) * (info
->raid_disks
- 1));
4259 fprintf(stderr
, Name
": could not allocate raid device\n");
4263 strncpy((char *) dev
->volume
, name
, MAX_RAID_SERIAL_LEN
);
4264 if (info
->level
== 1)
4265 array_blocks
= info_to_blocks_per_member(info
);
4267 array_blocks
= calc_array_size(info
->level
, info
->raid_disks
,
4268 info
->layout
, info
->chunk_size
,
4270 /* round array size down to closest MB */
4271 array_blocks
= (array_blocks
>> SECT_PER_MB_SHIFT
) << SECT_PER_MB_SHIFT
;
4273 dev
->size_low
= __cpu_to_le32((__u32
) array_blocks
);
4274 dev
->size_high
= __cpu_to_le32((__u32
) (array_blocks
>> 32));
4275 dev
->status
= (DEV_READ_COALESCING
| DEV_WRITE_COALESCING
);
4277 vol
->migr_state
= 0;
4278 set_migr_type(dev
, MIGR_INIT
);
4279 vol
->dirty
= !info
->state
;
4280 vol
->curr_migr_unit
= 0;
4281 map
= get_imsm_map(dev
, 0);
4282 map
->pba_of_lba0
= __cpu_to_le32(super
->create_offset
);
4283 map
->blocks_per_member
= __cpu_to_le32(info_to_blocks_per_member(info
));
4284 map
->blocks_per_strip
= __cpu_to_le16(info_to_blocks_per_strip(info
));
4285 map
->failed_disk_num
= ~0;
4286 map
->map_state
= info
->failed_disks
? IMSM_T_STATE_DEGRADED
: IMSM_T_STATE_NORMAL
;
4289 if (info
->level
== 1 && info
->raid_disks
> 2) {
4292 fprintf(stderr
, Name
": imsm does not support more than 2 disks"
4293 "in a raid1 volume\n");
4297 map
->raid_level
= info
->level
;
4298 if (info
->level
== 10) {
4299 map
->raid_level
= 1;
4300 map
->num_domains
= info
->raid_disks
/ 2;
4301 } else if (info
->level
== 1)
4302 map
->num_domains
= info
->raid_disks
;
4304 map
->num_domains
= 1;
4306 num_data_stripes
= info_to_num_data_stripes(info
, map
->num_domains
);
4307 map
->num_data_stripes
= __cpu_to_le32(num_data_stripes
);
4309 map
->num_members
= info
->raid_disks
;
4310 for (i
= 0; i
< map
->num_members
; i
++) {
4311 /* initialized in add_to_super */
4312 set_imsm_ord_tbl_ent(map
, i
, IMSM_ORD_REBUILD
);
4314 mpb
->num_raid_devs
++;
4317 dv
->index
= super
->current_vol
;
4318 dv
->next
= super
->devlist
;
4319 super
->devlist
= dv
;
4321 imsm_update_version_info(super
);
4326 static int init_super_imsm(struct supertype
*st
, mdu_array_info_t
*info
,
4327 unsigned long long size
, char *name
,
4328 char *homehost
, int *uuid
)
4330 /* This is primarily called by Create when creating a new array.
4331 * We will then get add_to_super called for each component, and then
4332 * write_init_super called to write it out to each device.
4333 * For IMSM, Create can create on fresh devices or on a pre-existing
4335 * To create on a pre-existing array a different method will be called.
4336 * This one is just for fresh drives.
4338 struct intel_super
*super
;
4339 struct imsm_super
*mpb
;
4344 return init_super_imsm_volume(st
, info
, size
, name
, homehost
, uuid
);
4347 mpb_size
= disks_to_mpb_size(info
->nr_disks
);
4351 super
= alloc_super();
4352 if (super
&& posix_memalign(&super
->buf
, 512, mpb_size
) != 0) {
4357 fprintf(stderr
, Name
4358 ": %s could not allocate superblock\n", __func__
);
4361 if (posix_memalign(&super
->migr_rec_buf
, 512, 512) != 0) {
4362 fprintf(stderr
, Name
4363 ": %s could not allocate migr_rec buffer\n", __func__
);
4368 memset(super
->buf
, 0, mpb_size
);
4370 mpb
->mpb_size
= __cpu_to_le32(mpb_size
);
4374 /* zeroing superblock */
4378 mpb
->attributes
= MPB_ATTRIB_CHECKSUM_VERIFY
;
4380 version
= (char *) mpb
->sig
;
4381 strcpy(version
, MPB_SIGNATURE
);
4382 version
+= strlen(MPB_SIGNATURE
);
4383 strcpy(version
, MPB_VERSION_RAID0
);
4389 static int add_to_super_imsm_volume(struct supertype
*st
, mdu_disk_info_t
*dk
,
4390 int fd
, char *devname
)
4392 struct intel_super
*super
= st
->sb
;
4393 struct imsm_super
*mpb
= super
->anchor
;
4394 struct imsm_disk
*_disk
;
4395 struct imsm_dev
*dev
;
4396 struct imsm_map
*map
;
4400 dev
= get_imsm_dev(super
, super
->current_vol
);
4401 map
= get_imsm_map(dev
, 0);
4403 if (! (dk
->state
& (1<<MD_DISK_SYNC
))) {
4404 fprintf(stderr
, Name
": %s: Cannot add spare devices to IMSM volume\n",
4410 /* we're doing autolayout so grab the pre-marked (in
4411 * validate_geometry) raid_disk
4413 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
4414 if (dl
->raiddisk
== dk
->raid_disk
)
4417 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
4418 if (dl
->major
== dk
->major
&&
4419 dl
->minor
== dk
->minor
)
4424 fprintf(stderr
, Name
": %s is not a member of the same container\n", devname
);
4428 /* add a pristine spare to the metadata */
4429 if (dl
->index
< 0) {
4430 dl
->index
= super
->anchor
->num_disks
;
4431 super
->anchor
->num_disks
++;
4433 /* Check the device has not already been added */
4434 slot
= get_imsm_disk_slot(map
, dl
->index
);
4436 (get_imsm_ord_tbl_ent(dev
, slot
, -1) & IMSM_ORD_REBUILD
) == 0) {
4437 fprintf(stderr
, Name
": %s has been included in this array twice\n",
4441 set_imsm_ord_tbl_ent(map
, dk
->raid_disk
, dl
->index
);
4442 dl
->disk
.status
= CONFIGURED_DISK
;
4444 /* update size of 'missing' disks to be at least as large as the
4445 * largest acitve member (we only have dummy missing disks when
4446 * creating the first volume)
4448 if (super
->current_vol
== 0) {
4449 for (df
= super
->missing
; df
; df
= df
->next
) {
4450 if (dl
->disk
.total_blocks
> df
->disk
.total_blocks
)
4451 df
->disk
.total_blocks
= dl
->disk
.total_blocks
;
4452 _disk
= __get_imsm_disk(mpb
, df
->index
);
4457 /* refresh unset/failed slots to point to valid 'missing' entries */
4458 for (df
= super
->missing
; df
; df
= df
->next
)
4459 for (slot
= 0; slot
< mpb
->num_disks
; slot
++) {
4460 __u32 ord
= get_imsm_ord_tbl_ent(dev
, slot
, -1);
4462 if ((ord
& IMSM_ORD_REBUILD
) == 0)
4464 set_imsm_ord_tbl_ent(map
, slot
, df
->index
| IMSM_ORD_REBUILD
);
4465 dprintf("set slot:%d to missing disk:%d\n", slot
, df
->index
);
4469 /* if we are creating the first raid device update the family number */
4470 if (super
->current_vol
== 0) {
4472 struct imsm_dev
*_dev
= __get_imsm_dev(mpb
, 0);
4474 _disk
= __get_imsm_disk(mpb
, dl
->index
);
4475 if (!_dev
|| !_disk
) {
4476 fprintf(stderr
, Name
": BUG mpb setup error\n");
4482 sum
+= __gen_imsm_checksum(mpb
);
4483 mpb
->family_num
= __cpu_to_le32(sum
);
4484 mpb
->orig_family_num
= mpb
->family_num
;
4486 super
->current_disk
= dl
;
4491 * Function marks disk as spare and restores disk serial
4492 * in case it was previously marked as failed by takeover operation
4494 * -1 : critical error
4495 * 0 : disk is marked as spare but serial is not set
4498 int mark_spare(struct dl
*disk
)
4500 __u8 serial
[MAX_RAID_SERIAL_LEN
];
4507 if (!imsm_read_serial(disk
->fd
, NULL
, serial
)) {
4508 /* Restore disk serial number, because takeover marks disk
4509 * as failed and adds to serial ':0' before it becomes
4512 serialcpy(disk
->serial
, serial
);
4513 serialcpy(disk
->disk
.serial
, serial
);
4516 disk
->disk
.status
= SPARE_DISK
;
4522 static int add_to_super_imsm(struct supertype
*st
, mdu_disk_info_t
*dk
,
4523 int fd
, char *devname
)
4525 struct intel_super
*super
= st
->sb
;
4527 unsigned long long size
;
4532 /* If we are on an RAID enabled platform check that the disk is
4533 * attached to the raid controller.
4534 * We do not need to test disks attachment for container based additions,
4535 * they shall be already tested when container was created/assembled.
4537 rv
= find_intel_hba_capability(fd
, super
, devname
);
4538 /* no orom/efi or non-intel hba of the disk */
4540 dprintf("capability: %p fd: %d ret: %d\n",
4541 super
->orom
, fd
, rv
);
4545 if (super
->current_vol
>= 0)
4546 return add_to_super_imsm_volume(st
, dk
, fd
, devname
);
4549 dd
= malloc(sizeof(*dd
));
4552 Name
": malloc failed %s:%d.\n", __func__
, __LINE__
);
4555 memset(dd
, 0, sizeof(*dd
));
4556 dd
->major
= major(stb
.st_rdev
);
4557 dd
->minor
= minor(stb
.st_rdev
);
4558 dd
->devname
= devname
? strdup(devname
) : NULL
;
4561 dd
->action
= DISK_ADD
;
4562 rv
= imsm_read_serial(fd
, devname
, dd
->serial
);
4565 Name
": failed to retrieve scsi serial, aborting\n");
4570 get_dev_size(fd
, NULL
, &size
);
4572 serialcpy(dd
->disk
.serial
, dd
->serial
);
4573 dd
->disk
.total_blocks
= __cpu_to_le32(size
);
4575 if (sysfs_disk_to_scsi_id(fd
, &id
) == 0)
4576 dd
->disk
.scsi_id
= __cpu_to_le32(id
);
4578 dd
->disk
.scsi_id
= __cpu_to_le32(0);
4580 if (st
->update_tail
) {
4581 dd
->next
= super
->disk_mgmt_list
;
4582 super
->disk_mgmt_list
= dd
;
4584 dd
->next
= super
->disks
;
4586 super
->updates_pending
++;
4593 static int remove_from_super_imsm(struct supertype
*st
, mdu_disk_info_t
*dk
)
4595 struct intel_super
*super
= st
->sb
;
4598 /* remove from super works only in mdmon - for communication
4599 * manager - monitor. Check if communication memory buffer
4602 if (!st
->update_tail
) {
4604 Name
": %s shall be used in mdmon context only"
4605 "(line %d).\n", __func__
, __LINE__
);
4608 dd
= malloc(sizeof(*dd
));
4611 Name
": malloc failed %s:%d.\n", __func__
, __LINE__
);
4614 memset(dd
, 0, sizeof(*dd
));
4615 dd
->major
= dk
->major
;
4616 dd
->minor
= dk
->minor
;
4619 dd
->action
= DISK_REMOVE
;
4621 dd
->next
= super
->disk_mgmt_list
;
4622 super
->disk_mgmt_list
= dd
;
4628 static int store_imsm_mpb(int fd
, struct imsm_super
*mpb
);
4632 struct imsm_super anchor
;
4633 } spare_record
__attribute__ ((aligned(512)));
4635 /* spare records have their own family number and do not have any defined raid
4638 static int write_super_imsm_spares(struct intel_super
*super
, int doclose
)
4640 struct imsm_super
*mpb
= super
->anchor
;
4641 struct imsm_super
*spare
= &spare_record
.anchor
;
4645 spare
->mpb_size
= __cpu_to_le32(sizeof(struct imsm_super
)),
4646 spare
->generation_num
= __cpu_to_le32(1UL),
4647 spare
->attributes
= MPB_ATTRIB_CHECKSUM_VERIFY
;
4648 spare
->num_disks
= 1,
4649 spare
->num_raid_devs
= 0,
4650 spare
->cache_size
= mpb
->cache_size
,
4651 spare
->pwr_cycle_count
= __cpu_to_le32(1),
4653 snprintf((char *) spare
->sig
, MAX_SIGNATURE_LENGTH
,
4654 MPB_SIGNATURE MPB_VERSION_RAID0
);
4656 for (d
= super
->disks
; d
; d
= d
->next
) {
4660 spare
->disk
[0] = d
->disk
;
4661 sum
= __gen_imsm_checksum(spare
);
4662 spare
->family_num
= __cpu_to_le32(sum
);
4663 spare
->orig_family_num
= 0;
4664 sum
= __gen_imsm_checksum(spare
);
4665 spare
->check_sum
= __cpu_to_le32(sum
);
4667 if (store_imsm_mpb(d
->fd
, spare
)) {
4668 fprintf(stderr
, "%s: failed for device %d:%d %s\n",
4669 __func__
, d
->major
, d
->minor
, strerror(errno
));
4681 static int write_super_imsm(struct supertype
*st
, int doclose
)
4683 struct intel_super
*super
= st
->sb
;
4684 struct imsm_super
*mpb
= super
->anchor
;
4690 __u32 mpb_size
= sizeof(struct imsm_super
) - sizeof(struct imsm_disk
);
4692 int clear_migration_record
= 1;
4694 /* 'generation' is incremented everytime the metadata is written */
4695 generation
= __le32_to_cpu(mpb
->generation_num
);
4697 mpb
->generation_num
= __cpu_to_le32(generation
);
4699 /* fix up cases where previous mdadm releases failed to set
4702 if (mpb
->orig_family_num
== 0)
4703 mpb
->orig_family_num
= mpb
->family_num
;
4705 for (d
= super
->disks
; d
; d
= d
->next
) {
4709 mpb
->disk
[d
->index
] = d
->disk
;
4713 for (d
= super
->missing
; d
; d
= d
->next
) {
4714 mpb
->disk
[d
->index
] = d
->disk
;
4717 mpb
->num_disks
= num_disks
;
4718 mpb_size
+= sizeof(struct imsm_disk
) * mpb
->num_disks
;
4720 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
4721 struct imsm_dev
*dev
= __get_imsm_dev(mpb
, i
);
4722 struct imsm_dev
*dev2
= get_imsm_dev(super
, i
);
4724 imsm_copy_dev(dev
, dev2
);
4725 mpb_size
+= sizeof_imsm_dev(dev
, 0);
4727 if (is_gen_migration(dev2
))
4728 clear_migration_record
= 0;
4730 mpb_size
+= __le32_to_cpu(mpb
->bbm_log_size
);
4731 mpb
->mpb_size
= __cpu_to_le32(mpb_size
);
4733 /* recalculate checksum */
4734 sum
= __gen_imsm_checksum(mpb
);
4735 mpb
->check_sum
= __cpu_to_le32(sum
);
4737 if (clear_migration_record
)
4738 memset(super
->migr_rec_buf
, 0, 512);
4740 /* write the mpb for disks that compose raid devices */
4741 for (d
= super
->disks
; d
; d
= d
->next
) {
4742 if (d
->index
< 0 || is_failed(&d
->disk
))
4744 if (store_imsm_mpb(d
->fd
, mpb
))
4745 fprintf(stderr
, "%s: failed for device %d:%d %s\n",
4746 __func__
, d
->major
, d
->minor
, strerror(errno
));
4747 if (clear_migration_record
) {
4748 unsigned long long dsize
;
4750 get_dev_size(d
->fd
, NULL
, &dsize
);
4751 if (lseek64(d
->fd
, dsize
- 512, SEEK_SET
) >= 0) {
4752 if (write(d
->fd
, super
->migr_rec_buf
, 512) != 512)
4753 perror("Write migr_rec failed");
4763 return write_super_imsm_spares(super
, doclose
);
4769 static int create_array(struct supertype
*st
, int dev_idx
)
4772 struct imsm_update_create_array
*u
;
4773 struct intel_super
*super
= st
->sb
;
4774 struct imsm_dev
*dev
= get_imsm_dev(super
, dev_idx
);
4775 struct imsm_map
*map
= get_imsm_map(dev
, 0);
4776 struct disk_info
*inf
;
4777 struct imsm_disk
*disk
;
4780 len
= sizeof(*u
) - sizeof(*dev
) + sizeof_imsm_dev(dev
, 0) +
4781 sizeof(*inf
) * map
->num_members
;
4784 fprintf(stderr
, "%s: failed to allocate update buffer\n",
4789 u
->type
= update_create_array
;
4790 u
->dev_idx
= dev_idx
;
4791 imsm_copy_dev(&u
->dev
, dev
);
4792 inf
= get_disk_info(u
);
4793 for (i
= 0; i
< map
->num_members
; i
++) {
4794 int idx
= get_imsm_disk_idx(dev
, i
, -1);
4796 disk
= get_imsm_disk(super
, idx
);
4797 serialcpy(inf
[i
].serial
, disk
->serial
);
4799 append_metadata_update(st
, u
, len
);
4804 static int mgmt_disk(struct supertype
*st
)
4806 struct intel_super
*super
= st
->sb
;
4808 struct imsm_update_add_remove_disk
*u
;
4810 if (!super
->disk_mgmt_list
)
4816 fprintf(stderr
, "%s: failed to allocate update buffer\n",
4821 u
->type
= update_add_remove_disk
;
4822 append_metadata_update(st
, u
, len
);
4827 static int write_init_super_imsm(struct supertype
*st
)
4829 struct intel_super
*super
= st
->sb
;
4830 int current_vol
= super
->current_vol
;
4832 /* we are done with current_vol reset it to point st at the container */
4833 super
->current_vol
= -1;
4835 if (st
->update_tail
) {
4836 /* queue the recently created array / added disk
4837 * as a metadata update */
4840 /* determine if we are creating a volume or adding a disk */
4841 if (current_vol
< 0) {
4842 /* in the mgmt (add/remove) disk case we are running
4843 * in mdmon context, so don't close fd's
4845 return mgmt_disk(st
);
4847 rv
= create_array(st
, current_vol
);
4852 for (d
= super
->disks
; d
; d
= d
->next
)
4853 Kill(d
->devname
, NULL
, 0, 1, 1);
4854 return write_super_imsm(st
, 1);
4859 static int store_super_imsm(struct supertype
*st
, int fd
)
4861 struct intel_super
*super
= st
->sb
;
4862 struct imsm_super
*mpb
= super
? super
->anchor
: NULL
;
4868 return store_imsm_mpb(fd
, mpb
);
4874 static int imsm_bbm_log_size(struct imsm_super
*mpb
)
4876 return __le32_to_cpu(mpb
->bbm_log_size
);
4880 static int validate_geometry_imsm_container(struct supertype
*st
, int level
,
4881 int layout
, int raiddisks
, int chunk
,
4882 unsigned long long size
, char *dev
,
4883 unsigned long long *freesize
,
4887 unsigned long long ldsize
;
4888 struct intel_super
*super
=NULL
;
4891 if (level
!= LEVEL_CONTAINER
)
4896 fd
= open(dev
, O_RDONLY
|O_EXCL
, 0);
4899 fprintf(stderr
, Name
": imsm: Cannot open %s: %s\n",
4900 dev
, strerror(errno
));
4903 if (!get_dev_size(fd
, dev
, &ldsize
)) {
4908 /* capabilities retrieve could be possible
4909 * note that there is no fd for the disks in array.
4911 super
= alloc_super();
4914 Name
": malloc of %zu failed.\n",
4920 rv
= find_intel_hba_capability(fd
, super
, verbose
? dev
: NULL
);
4924 fd2devname(fd
, str
);
4925 dprintf("validate_geometry_imsm_container: fd: %d %s orom: %p rv: %d raiddisk: %d\n",
4926 fd
, str
, super
->orom
, rv
, raiddisks
);
4928 /* no orom/efi or non-intel hba of the disk */
4934 if (super
->orom
&& raiddisks
> super
->orom
->tds
) {
4936 fprintf(stderr
, Name
": %d exceeds maximum number of"
4937 " platform supported disks: %d\n",
4938 raiddisks
, super
->orom
->tds
);
4944 *freesize
= avail_size_imsm(st
, ldsize
>> 9);
4950 static unsigned long long find_size(struct extent
*e
, int *idx
, int num_extents
)
4952 const unsigned long long base_start
= e
[*idx
].start
;
4953 unsigned long long end
= base_start
+ e
[*idx
].size
;
4956 if (base_start
== end
)
4960 for (i
= *idx
; i
< num_extents
; i
++) {
4961 /* extend overlapping extents */
4962 if (e
[i
].start
>= base_start
&&
4963 e
[i
].start
<= end
) {
4966 if (e
[i
].start
+ e
[i
].size
> end
)
4967 end
= e
[i
].start
+ e
[i
].size
;
4968 } else if (e
[i
].start
> end
) {
4974 return end
- base_start
;
4977 static unsigned long long merge_extents(struct intel_super
*super
, int sum_extents
)
4979 /* build a composite disk with all known extents and generate a new
4980 * 'maxsize' given the "all disks in an array must share a common start
4981 * offset" constraint
4983 struct extent
*e
= calloc(sum_extents
, sizeof(*e
));
4987 unsigned long long pos
;
4988 unsigned long long start
= 0;
4989 unsigned long long maxsize
;
4990 unsigned long reserve
;
4995 /* coalesce and sort all extents. also, check to see if we need to
4996 * reserve space between member arrays
4999 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
5002 for (i
= 0; i
< dl
->extent_cnt
; i
++)
5005 qsort(e
, sum_extents
, sizeof(*e
), cmp_extent
);
5010 while (i
< sum_extents
) {
5011 e
[j
].start
= e
[i
].start
;
5012 e
[j
].size
= find_size(e
, &i
, sum_extents
);
5014 if (e
[j
-1].size
== 0)
5023 unsigned long long esize
;
5025 esize
= e
[i
].start
- pos
;
5026 if (esize
>= maxsize
) {
5031 pos
= e
[i
].start
+ e
[i
].size
;
5033 } while (e
[i
-1].size
);
5039 /* FIXME assumes volume at offset 0 is the first volume in a
5042 if (start_extent
> 0)
5043 reserve
= IMSM_RESERVED_SECTORS
; /* gap between raid regions */
5047 if (maxsize
< reserve
)
5050 super
->create_offset
= ~((__u32
) 0);
5051 if (start
+ reserve
> super
->create_offset
)
5052 return 0; /* start overflows create_offset */
5053 super
->create_offset
= start
+ reserve
;
5055 return maxsize
- reserve
;
5058 static int is_raid_level_supported(const struct imsm_orom
*orom
, int level
, int raiddisks
)
5060 if (level
< 0 || level
== 6 || level
== 4)
5063 /* if we have an orom prevent invalid raid levels */
5066 case 0: return imsm_orom_has_raid0(orom
);
5069 return imsm_orom_has_raid1e(orom
);
5070 return imsm_orom_has_raid1(orom
) && raiddisks
== 2;
5071 case 10: return imsm_orom_has_raid10(orom
) && raiddisks
== 4;
5072 case 5: return imsm_orom_has_raid5(orom
) && raiddisks
> 2;
5075 return 1; /* not on an Intel RAID platform so anything goes */
5080 static int imsm_default_chunk(const struct imsm_orom
*orom
)
5082 /* up to 512 if the plaform supports it, otherwise the platform max.
5083 * 128 if no platform detected
5085 int fs
= max(7, orom
? fls(orom
->sss
) : 0);
5087 return min(512, (1 << fs
));
5090 #define pr_vrb(fmt, arg...) (void) (verbose && fprintf(stderr, Name fmt, ##arg))
5092 validate_geometry_imsm_orom(struct intel_super
*super
, int level
, int layout
,
5093 int raiddisks
, int *chunk
, int verbose
)
5095 /* check/set platform and metadata limits/defaults */
5096 if (super
->orom
&& raiddisks
> super
->orom
->dpa
) {
5097 pr_vrb(": platform supports a maximum of %d disks per array\n",
5102 /* capabilities of OROM tested - copied from validate_geometry_imsm_volume */
5103 if (!is_raid_level_supported(super
->orom
, level
, raiddisks
)) {
5104 pr_vrb(": platform does not support raid%d with %d disk%s\n",
5105 level
, raiddisks
, raiddisks
> 1 ? "s" : "");
5109 if (chunk
&& (*chunk
== 0 || *chunk
== UnSet
))
5110 *chunk
= imsm_default_chunk(super
->orom
);
5112 if (super
->orom
&& chunk
&& !imsm_orom_has_chunk(super
->orom
, *chunk
)) {
5113 pr_vrb(": platform does not support a chunk size of: "
5118 if (layout
!= imsm_level_to_layout(level
)) {
5120 pr_vrb(": imsm raid 5 only supports the left-asymmetric layout\n");
5121 else if (level
== 10)
5122 pr_vrb(": imsm raid 10 only supports the n2 layout\n");
5124 pr_vrb(": imsm unknown layout %#x for this raid level %d\n",
5131 /* validate_geometry_imsm_volume - lifted from validate_geometry_ddf_bvd
5132 * FIX ME add ahci details
5134 static int validate_geometry_imsm_volume(struct supertype
*st
, int level
,
5135 int layout
, int raiddisks
, int *chunk
,
5136 unsigned long long size
, char *dev
,
5137 unsigned long long *freesize
,
5141 struct intel_super
*super
= st
->sb
;
5142 struct imsm_super
*mpb
;
5144 unsigned long long pos
= 0;
5145 unsigned long long maxsize
;
5149 /* We must have the container info already read in. */
5153 mpb
= super
->anchor
;
5155 if (mpb
->num_raid_devs
> 0 && mpb
->num_disks
!= raiddisks
) {
5156 fprintf(stderr
, Name
": the option-rom requires all "
5157 "member disks to be a member of all volumes.\n");
5161 if (!validate_geometry_imsm_orom(super
, level
, layout
, raiddisks
, chunk
, verbose
)) {
5162 fprintf(stderr
, Name
": RAID gemetry validation failed. "
5163 "Cannot proceed with the action(s).\n");
5167 /* General test: make sure there is space for
5168 * 'raiddisks' device extents of size 'size' at a given
5171 unsigned long long minsize
= size
;
5172 unsigned long long start_offset
= MaxSector
;
5175 minsize
= MPB_SECTOR_CNT
+ IMSM_RESERVED_SECTORS
;
5176 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
5181 e
= get_extents(super
, dl
);
5184 unsigned long long esize
;
5185 esize
= e
[i
].start
- pos
;
5186 if (esize
>= minsize
)
5188 if (found
&& start_offset
== MaxSector
) {
5191 } else if (found
&& pos
!= start_offset
) {
5195 pos
= e
[i
].start
+ e
[i
].size
;
5197 } while (e
[i
-1].size
);
5202 if (dcnt
< raiddisks
) {
5204 fprintf(stderr
, Name
": imsm: Not enough "
5205 "devices with space for this array "
5213 /* This device must be a member of the set */
5214 if (stat(dev
, &stb
) < 0)
5216 if ((S_IFMT
& stb
.st_mode
) != S_IFBLK
)
5218 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
5219 if (dl
->major
== (int)major(stb
.st_rdev
) &&
5220 dl
->minor
== (int)minor(stb
.st_rdev
))
5225 fprintf(stderr
, Name
": %s is not in the "
5226 "same imsm set\n", dev
);
5228 } else if (super
->orom
&& dl
->index
< 0 && mpb
->num_raid_devs
) {
5229 /* If a volume is present then the current creation attempt
5230 * cannot incorporate new spares because the orom may not
5231 * understand this configuration (all member disks must be
5232 * members of each array in the container).
5234 fprintf(stderr
, Name
": %s is a spare and a volume"
5235 " is already defined for this container\n", dev
);
5236 fprintf(stderr
, Name
": The option-rom requires all member"
5237 " disks to be a member of all volumes\n");
5241 /* retrieve the largest free space block */
5242 e
= get_extents(super
, dl
);
5247 unsigned long long esize
;
5249 esize
= e
[i
].start
- pos
;
5250 if (esize
>= maxsize
)
5252 pos
= e
[i
].start
+ e
[i
].size
;
5254 } while (e
[i
-1].size
);
5259 fprintf(stderr
, Name
": unable to determine free space for: %s\n",
5263 if (maxsize
< size
) {
5265 fprintf(stderr
, Name
": %s not enough space (%llu < %llu)\n",
5266 dev
, maxsize
, size
);
5270 /* count total number of extents for merge */
5272 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
5274 i
+= dl
->extent_cnt
;
5276 maxsize
= merge_extents(super
, i
);
5277 if (maxsize
< size
|| maxsize
== 0) {
5279 fprintf(stderr
, Name
": not enough space after merge (%llu < %llu)\n",
5284 *freesize
= maxsize
;
5289 static int reserve_space(struct supertype
*st
, int raiddisks
,
5290 unsigned long long size
, int chunk
,
5291 unsigned long long *freesize
)
5293 struct intel_super
*super
= st
->sb
;
5294 struct imsm_super
*mpb
= super
->anchor
;
5299 unsigned long long maxsize
;
5300 unsigned long long minsize
;
5304 /* find the largest common start free region of the possible disks */
5308 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
5314 /* don't activate new spares if we are orom constrained
5315 * and there is already a volume active in the container
5317 if (super
->orom
&& dl
->index
< 0 && mpb
->num_raid_devs
)
5320 e
= get_extents(super
, dl
);
5323 for (i
= 1; e
[i
-1].size
; i
++)
5331 maxsize
= merge_extents(super
, extent_cnt
);
5335 minsize
= chunk
* 2;
5337 if (cnt
< raiddisks
||
5338 (super
->orom
&& used
&& used
!= raiddisks
) ||
5339 maxsize
< minsize
||
5341 fprintf(stderr
, Name
": not enough devices with space to create array.\n");
5342 return 0; /* No enough free spaces large enough */
5354 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
5356 dl
->raiddisk
= cnt
++;
5363 static int validate_geometry_imsm(struct supertype
*st
, int level
, int layout
,
5364 int raiddisks
, int *chunk
, unsigned long long size
,
5365 char *dev
, unsigned long long *freesize
,
5373 * if given unused devices create a container
5374 * if given given devices in a container create a member volume
5376 if (level
== LEVEL_CONTAINER
) {
5377 /* Must be a fresh device to add to a container */
5378 return validate_geometry_imsm_container(st
, level
, layout
,
5380 chunk
?*chunk
:0, size
,
5386 if (st
->sb
&& freesize
) {
5387 /* we are being asked to automatically layout a
5388 * new volume based on the current contents of
5389 * the container. If the the parameters can be
5390 * satisfied reserve_space will record the disks,
5391 * start offset, and size of the volume to be
5392 * created. add_to_super and getinfo_super
5393 * detect when autolayout is in progress.
5395 if (!validate_geometry_imsm_orom(st
->sb
, level
, layout
,
5399 return reserve_space(st
, raiddisks
, size
,
5400 chunk
?*chunk
:0, freesize
);
5405 /* creating in a given container */
5406 return validate_geometry_imsm_volume(st
, level
, layout
,
5407 raiddisks
, chunk
, size
,
5408 dev
, freesize
, verbose
);
5411 /* This device needs to be a device in an 'imsm' container */
5412 fd
= open(dev
, O_RDONLY
|O_EXCL
, 0);
5416 Name
": Cannot create this array on device %s\n",
5421 if (errno
!= EBUSY
|| (fd
= open(dev
, O_RDONLY
, 0)) < 0) {
5423 fprintf(stderr
, Name
": Cannot open %s: %s\n",
5424 dev
, strerror(errno
));
5427 /* Well, it is in use by someone, maybe an 'imsm' container. */
5428 cfd
= open_container(fd
);
5432 fprintf(stderr
, Name
": Cannot use %s: It is busy\n",
5436 sra
= sysfs_read(cfd
, 0, GET_VERSION
);
5437 if (sra
&& sra
->array
.major_version
== -1 &&
5438 strcmp(sra
->text_version
, "imsm") == 0)
5442 /* This is a member of a imsm container. Load the container
5443 * and try to create a volume
5445 struct intel_super
*super
;
5447 if (load_super_imsm_all(st
, cfd
, (void **) &super
, NULL
) == 0) {
5449 st
->container_dev
= fd2devnum(cfd
);
5451 return validate_geometry_imsm_volume(st
, level
, layout
,
5460 fprintf(stderr
, Name
": failed container membership check\n");
5466 static void default_geometry_imsm(struct supertype
*st
, int *level
, int *layout
, int *chunk
)
5468 struct intel_super
*super
= st
->sb
;
5470 if (level
&& *level
== UnSet
)
5471 *level
= LEVEL_CONTAINER
;
5473 if (level
&& layout
&& *layout
== UnSet
)
5474 *layout
= imsm_level_to_layout(*level
);
5476 if (chunk
&& (*chunk
== UnSet
|| *chunk
== 0))
5477 *chunk
= imsm_default_chunk(super
->orom
);
5480 static void handle_missing(struct intel_super
*super
, struct imsm_dev
*dev
);
5482 static int kill_subarray_imsm(struct supertype
*st
)
5484 /* remove the subarray currently referenced by ->current_vol */
5486 struct intel_dev
**dp
;
5487 struct intel_super
*super
= st
->sb
;
5488 __u8 current_vol
= super
->current_vol
;
5489 struct imsm_super
*mpb
= super
->anchor
;
5491 if (super
->current_vol
< 0)
5493 super
->current_vol
= -1; /* invalidate subarray cursor */
5495 /* block deletions that would change the uuid of active subarrays
5497 * FIXME when immutable ids are available, but note that we'll
5498 * also need to fixup the invalidated/active subarray indexes in
5501 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
5504 if (i
< current_vol
)
5506 sprintf(subarray
, "%u", i
);
5507 if (is_subarray_active(subarray
, st
->devname
)) {
5509 Name
": deleting subarray-%d would change the UUID of active subarray-%d, aborting\n",
5516 if (st
->update_tail
) {
5517 struct imsm_update_kill_array
*u
= malloc(sizeof(*u
));
5521 u
->type
= update_kill_array
;
5522 u
->dev_idx
= current_vol
;
5523 append_metadata_update(st
, u
, sizeof(*u
));
5528 for (dp
= &super
->devlist
; *dp
;)
5529 if ((*dp
)->index
== current_vol
) {
5532 handle_missing(super
, (*dp
)->dev
);
5533 if ((*dp
)->index
> current_vol
)
5538 /* no more raid devices, all active components are now spares,
5539 * but of course failed are still failed
5541 if (--mpb
->num_raid_devs
== 0) {
5544 for (d
= super
->disks
; d
; d
= d
->next
)
5549 super
->updates_pending
++;
5554 static int update_subarray_imsm(struct supertype
*st
, char *subarray
,
5555 char *update
, struct mddev_ident
*ident
)
5557 /* update the subarray currently referenced by ->current_vol */
5558 struct intel_super
*super
= st
->sb
;
5559 struct imsm_super
*mpb
= super
->anchor
;
5561 if (strcmp(update
, "name") == 0) {
5562 char *name
= ident
->name
;
5566 if (is_subarray_active(subarray
, st
->devname
)) {
5568 Name
": Unable to update name of active subarray\n");
5572 if (!check_name(super
, name
, 0))
5575 vol
= strtoul(subarray
, &ep
, 10);
5576 if (*ep
!= '\0' || vol
>= super
->anchor
->num_raid_devs
)
5579 if (st
->update_tail
) {
5580 struct imsm_update_rename_array
*u
= malloc(sizeof(*u
));
5584 u
->type
= update_rename_array
;
5586 snprintf((char *) u
->name
, MAX_RAID_SERIAL_LEN
, "%s", name
);
5587 append_metadata_update(st
, u
, sizeof(*u
));
5589 struct imsm_dev
*dev
;
5592 dev
= get_imsm_dev(super
, vol
);
5593 snprintf((char *) dev
->volume
, MAX_RAID_SERIAL_LEN
, "%s", name
);
5594 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
5595 dev
= get_imsm_dev(super
, i
);
5596 handle_missing(super
, dev
);
5598 super
->updates_pending
++;
5606 static int is_gen_migration(struct imsm_dev
*dev
)
5611 if (!dev
->vol
.migr_state
)
5614 if (migr_type(dev
) == MIGR_GEN_MIGR
)
5619 #endif /* MDASSEMBLE */
5621 static int is_rebuilding(struct imsm_dev
*dev
)
5623 struct imsm_map
*migr_map
;
5625 if (!dev
->vol
.migr_state
)
5628 if (migr_type(dev
) != MIGR_REBUILD
)
5631 migr_map
= get_imsm_map(dev
, 1);
5633 if (migr_map
->map_state
== IMSM_T_STATE_DEGRADED
)
5639 static void update_recovery_start(struct intel_super
*super
,
5640 struct imsm_dev
*dev
,
5641 struct mdinfo
*array
)
5643 struct mdinfo
*rebuild
= NULL
;
5647 if (!is_rebuilding(dev
))
5650 /* Find the rebuild target, but punt on the dual rebuild case */
5651 for (d
= array
->devs
; d
; d
= d
->next
)
5652 if (d
->recovery_start
== 0) {
5659 /* (?) none of the disks are marked with
5660 * IMSM_ORD_REBUILD, so assume they are missing and the
5661 * disk_ord_tbl was not correctly updated
5663 dprintf("%s: failed to locate out-of-sync disk\n", __func__
);
5667 units
= __le32_to_cpu(dev
->vol
.curr_migr_unit
);
5668 rebuild
->recovery_start
= units
* blocks_per_migr_unit(super
, dev
);
5672 static int recover_backup_imsm(struct supertype
*st
, struct mdinfo
*info
);
5675 static struct mdinfo
*container_content_imsm(struct supertype
*st
, char *subarray
)
5677 /* Given a container loaded by load_super_imsm_all,
5678 * extract information about all the arrays into
5680 * If 'subarray' is given, just extract info about that array.
5682 * For each imsm_dev create an mdinfo, fill it in,
5683 * then look for matching devices in super->disks
5684 * and create appropriate device mdinfo.
5686 struct intel_super
*super
= st
->sb
;
5687 struct imsm_super
*mpb
= super
->anchor
;
5688 struct mdinfo
*rest
= NULL
;
5692 int spare_disks
= 0;
5694 /* do not assemble arrays when not all attributes are supported */
5695 if (imsm_check_attributes(mpb
->attributes
) == 0) {
5697 fprintf(stderr
, Name
": Unsupported attributes in IMSM metadata."
5698 "Arrays activation is blocked.\n");
5701 /* check for bad blocks */
5702 if (imsm_bbm_log_size(super
->anchor
)) {
5703 fprintf(stderr
, Name
": BBM log found in IMSM metadata."
5704 "Arrays activation is blocked.\n");
5709 /* count spare devices, not used in maps
5711 for (d
= super
->disks
; d
; d
= d
->next
)
5715 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
5716 struct imsm_dev
*dev
;
5717 struct imsm_map
*map
;
5718 struct imsm_map
*map2
;
5719 struct mdinfo
*this;
5724 (i
!= strtoul(subarray
, &ep
, 10) || *ep
!= '\0'))
5727 dev
= get_imsm_dev(super
, i
);
5728 map
= get_imsm_map(dev
, 0);
5729 map2
= get_imsm_map(dev
, 1);
5731 /* do not publish arrays that are in the middle of an
5732 * unsupported migration
5734 if (dev
->vol
.migr_state
&&
5735 (migr_type(dev
) == MIGR_STATE_CHANGE
)) {
5736 fprintf(stderr
, Name
": cannot assemble volume '%.16s':"
5737 " unsupported migration in progress\n",
5741 /* do not publish arrays that are not support by controller's
5745 chunk
= __le16_to_cpu(map
->blocks_per_strip
) >> 1;
5746 this = malloc(sizeof(*this));
5748 fprintf(stderr
, Name
": failed to allocate %zu bytes\n",
5753 super
->current_vol
= i
;
5754 getinfo_super_imsm_volume(st
, this, NULL
);
5757 /* mdadm does not support all metadata features- set the bit in all arrays state */
5758 if (!validate_geometry_imsm_orom(super
,
5759 get_imsm_raid_level(map
), /* RAID level */
5760 imsm_level_to_layout(get_imsm_raid_level(map
)),
5761 map
->num_members
, /* raid disks */
5764 fprintf(stderr
, Name
": IMSM RAID geometry validation"
5765 " failed. Array %s activation is blocked.\n",
5767 this->array
.state
|=
5768 (1<<MD_SB_BLOCK_CONTAINER_RESHAPE
) |
5769 (1<<MD_SB_BLOCK_VOLUME
);
5773 /* if array has bad blocks, set suitable bit in all arrays state */
5775 this->array
.state
|=
5776 (1<<MD_SB_BLOCK_CONTAINER_RESHAPE
) |
5777 (1<<MD_SB_BLOCK_VOLUME
);
5779 for (slot
= 0 ; slot
< map
->num_members
; slot
++) {
5780 unsigned long long recovery_start
;
5781 struct mdinfo
*info_d
;
5788 idx
= get_imsm_disk_idx(dev
, slot
, 0);
5789 ord
= get_imsm_ord_tbl_ent(dev
, slot
, -1);
5790 for (d
= super
->disks
; d
; d
= d
->next
)
5791 if (d
->index
== idx
)
5794 recovery_start
= MaxSector
;
5797 if (d
&& is_failed(&d
->disk
))
5799 if (ord
& IMSM_ORD_REBUILD
)
5803 * if we skip some disks the array will be assmebled degraded;
5804 * reset resync start to avoid a dirty-degraded
5805 * situation when performing the intial sync
5807 * FIXME handle dirty degraded
5809 if ((skip
|| recovery_start
== 0) && !dev
->vol
.dirty
)
5810 this->resync_start
= MaxSector
;
5814 info_d
= calloc(1, sizeof(*info_d
));
5816 fprintf(stderr
, Name
": failed to allocate disk"
5817 " for volume %.16s\n", dev
->volume
);
5818 info_d
= this->devs
;
5820 struct mdinfo
*d
= info_d
->next
;
5829 info_d
->next
= this->devs
;
5830 this->devs
= info_d
;
5832 info_d
->disk
.number
= d
->index
;
5833 info_d
->disk
.major
= d
->major
;
5834 info_d
->disk
.minor
= d
->minor
;
5835 info_d
->disk
.raid_disk
= slot
;
5836 info_d
->recovery_start
= recovery_start
;
5838 if (slot
< map2
->num_members
)
5839 info_d
->disk
.state
= (1 << MD_DISK_ACTIVE
);
5841 this->array
.spare_disks
++;
5843 if (slot
< map
->num_members
)
5844 info_d
->disk
.state
= (1 << MD_DISK_ACTIVE
);
5846 this->array
.spare_disks
++;
5848 if (info_d
->recovery_start
== MaxSector
)
5849 this->array
.working_disks
++;
5851 info_d
->events
= __le32_to_cpu(mpb
->generation_num
);
5852 info_d
->data_offset
= __le32_to_cpu(map
->pba_of_lba0
);
5853 info_d
->component_size
= __le32_to_cpu(map
->blocks_per_member
);
5855 /* now that the disk list is up-to-date fixup recovery_start */
5856 update_recovery_start(super
, dev
, this);
5857 this->array
.spare_disks
+= spare_disks
;
5860 /* check for reshape */
5861 if (this->reshape_active
== 1)
5862 recover_backup_imsm(st
, this);
5871 static __u8
imsm_check_degraded(struct intel_super
*super
, struct imsm_dev
*dev
, int failed
)
5873 struct imsm_map
*map
= get_imsm_map(dev
, 0);
5876 return map
->map_state
== IMSM_T_STATE_UNINITIALIZED
?
5877 IMSM_T_STATE_UNINITIALIZED
: IMSM_T_STATE_NORMAL
;
5879 switch (get_imsm_raid_level(map
)) {
5881 return IMSM_T_STATE_FAILED
;
5884 if (failed
< map
->num_members
)
5885 return IMSM_T_STATE_DEGRADED
;
5887 return IMSM_T_STATE_FAILED
;
5892 * check to see if any mirrors have failed, otherwise we
5893 * are degraded. Even numbered slots are mirrored on
5897 /* gcc -Os complains that this is unused */
5898 int insync
= insync
;
5900 for (i
= 0; i
< map
->num_members
; i
++) {
5901 __u32 ord
= get_imsm_ord_tbl_ent(dev
, i
, -1);
5902 int idx
= ord_to_idx(ord
);
5903 struct imsm_disk
*disk
;
5905 /* reset the potential in-sync count on even-numbered
5906 * slots. num_copies is always 2 for imsm raid10
5911 disk
= get_imsm_disk(super
, idx
);
5912 if (!disk
|| is_failed(disk
) || ord
& IMSM_ORD_REBUILD
)
5915 /* no in-sync disks left in this mirror the
5919 return IMSM_T_STATE_FAILED
;
5922 return IMSM_T_STATE_DEGRADED
;
5926 return IMSM_T_STATE_DEGRADED
;
5928 return IMSM_T_STATE_FAILED
;
5934 return map
->map_state
;
5937 static int imsm_count_failed(struct intel_super
*super
, struct imsm_dev
*dev
)
5941 struct imsm_disk
*disk
;
5942 struct imsm_map
*map
= get_imsm_map(dev
, 0);
5943 struct imsm_map
*prev
= get_imsm_map(dev
, dev
->vol
.migr_state
);
5947 /* at the beginning of migration we set IMSM_ORD_REBUILD on
5948 * disks that are being rebuilt. New failures are recorded to
5949 * map[0]. So we look through all the disks we started with and
5950 * see if any failures are still present, or if any new ones
5953 * FIXME add support for online capacity expansion and
5954 * raid-level-migration
5956 for (i
= 0; i
< prev
->num_members
; i
++) {
5957 ord
= __le32_to_cpu(prev
->disk_ord_tbl
[i
]);
5958 ord
|= __le32_to_cpu(map
->disk_ord_tbl
[i
]);
5959 idx
= ord_to_idx(ord
);
5961 disk
= get_imsm_disk(super
, idx
);
5962 if (!disk
|| is_failed(disk
) || ord
& IMSM_ORD_REBUILD
)
5970 static int imsm_open_new(struct supertype
*c
, struct active_array
*a
,
5973 struct intel_super
*super
= c
->sb
;
5974 struct imsm_super
*mpb
= super
->anchor
;
5976 if (atoi(inst
) >= mpb
->num_raid_devs
) {
5977 fprintf(stderr
, "%s: subarry index %d, out of range\n",
5978 __func__
, atoi(inst
));
5982 dprintf("imsm: open_new %s\n", inst
);
5983 a
->info
.container_member
= atoi(inst
);
5987 static int is_resyncing(struct imsm_dev
*dev
)
5989 struct imsm_map
*migr_map
;
5991 if (!dev
->vol
.migr_state
)
5994 if (migr_type(dev
) == MIGR_INIT
||
5995 migr_type(dev
) == MIGR_REPAIR
)
5998 if (migr_type(dev
) == MIGR_GEN_MIGR
)
6001 migr_map
= get_imsm_map(dev
, 1);
6003 if ((migr_map
->map_state
== IMSM_T_STATE_NORMAL
) &&
6004 (dev
->vol
.migr_type
!= MIGR_GEN_MIGR
))
6010 /* return true if we recorded new information */
6011 static int mark_failure(struct imsm_dev
*dev
, struct imsm_disk
*disk
, int idx
)
6015 struct imsm_map
*map
;
6016 char buf
[MAX_RAID_SERIAL_LEN
+3];
6017 unsigned int len
, shift
= 0;
6019 /* new failures are always set in map[0] */
6020 map
= get_imsm_map(dev
, 0);
6022 slot
= get_imsm_disk_slot(map
, idx
);
6026 ord
= __le32_to_cpu(map
->disk_ord_tbl
[slot
]);
6027 if (is_failed(disk
) && (ord
& IMSM_ORD_REBUILD
))
6030 memcpy(buf
, disk
->serial
, MAX_RAID_SERIAL_LEN
);
6031 buf
[MAX_RAID_SERIAL_LEN
] = '\000';
6033 if ((len
= strlen(buf
)) >= MAX_RAID_SERIAL_LEN
)
6034 shift
= len
- MAX_RAID_SERIAL_LEN
+ 1;
6035 strncpy((char *)disk
->serial
, &buf
[shift
], MAX_RAID_SERIAL_LEN
);
6037 disk
->status
|= FAILED_DISK
;
6038 set_imsm_ord_tbl_ent(map
, slot
, idx
| IMSM_ORD_REBUILD
);
6039 if (map
->failed_disk_num
== 0xff)
6040 map
->failed_disk_num
= slot
;
6044 static void mark_missing(struct imsm_dev
*dev
, struct imsm_disk
*disk
, int idx
)
6046 mark_failure(dev
, disk
, idx
);
6048 if (disk
->scsi_id
== __cpu_to_le32(~(__u32
)0))
6051 disk
->scsi_id
= __cpu_to_le32(~(__u32
)0);
6052 memmove(&disk
->serial
[0], &disk
->serial
[1], MAX_RAID_SERIAL_LEN
- 1);
6055 static void handle_missing(struct intel_super
*super
, struct imsm_dev
*dev
)
6061 if (!super
->missing
)
6063 failed
= imsm_count_failed(super
, dev
);
6064 map_state
= imsm_check_degraded(super
, dev
, failed
);
6066 dprintf("imsm: mark missing\n");
6067 end_migration(dev
, map_state
);
6068 for (dl
= super
->missing
; dl
; dl
= dl
->next
)
6069 mark_missing(dev
, &dl
->disk
, dl
->index
);
6070 super
->updates_pending
++;
6073 static unsigned long long imsm_set_array_size(struct imsm_dev
*dev
)
6075 int used_disks
= imsm_num_data_members(dev
, 0);
6076 unsigned long long array_blocks
;
6077 struct imsm_map
*map
;
6079 if (used_disks
== 0) {
6080 /* when problems occures
6081 * return current array_blocks value
6083 array_blocks
= __le32_to_cpu(dev
->size_high
);
6084 array_blocks
= array_blocks
<< 32;
6085 array_blocks
+= __le32_to_cpu(dev
->size_low
);
6087 return array_blocks
;
6090 /* set array size in metadata
6092 map
= get_imsm_map(dev
, 0);
6093 array_blocks
= map
->blocks_per_member
* used_disks
;
6095 /* round array size down to closest MB
6097 array_blocks
= (array_blocks
>> SECT_PER_MB_SHIFT
) << SECT_PER_MB_SHIFT
;
6098 dev
->size_low
= __cpu_to_le32((__u32
)array_blocks
);
6099 dev
->size_high
= __cpu_to_le32((__u32
)(array_blocks
>> 32));
6101 return array_blocks
;
6104 static void imsm_set_disk(struct active_array
*a
, int n
, int state
);
6106 static void imsm_progress_container_reshape(struct intel_super
*super
)
6108 /* if no device has a migr_state, but some device has a
6109 * different number of members than the previous device, start
6110 * changing the number of devices in this device to match
6113 struct imsm_super
*mpb
= super
->anchor
;
6114 int prev_disks
= -1;
6118 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
6119 struct imsm_dev
*dev
= get_imsm_dev(super
, i
);
6120 struct imsm_map
*map
= get_imsm_map(dev
, 0);
6121 struct imsm_map
*map2
;
6122 int prev_num_members
;
6124 if (dev
->vol
.migr_state
)
6127 if (prev_disks
== -1)
6128 prev_disks
= map
->num_members
;
6129 if (prev_disks
== map
->num_members
)
6132 /* OK, this array needs to enter reshape mode.
6133 * i.e it needs a migr_state
6136 copy_map_size
= sizeof_imsm_map(map
);
6137 prev_num_members
= map
->num_members
;
6138 map
->num_members
= prev_disks
;
6139 dev
->vol
.migr_state
= 1;
6140 dev
->vol
.curr_migr_unit
= 0;
6141 set_migr_type(dev
, MIGR_GEN_MIGR
);
6142 for (i
= prev_num_members
;
6143 i
< map
->num_members
; i
++)
6144 set_imsm_ord_tbl_ent(map
, i
, i
);
6145 map2
= get_imsm_map(dev
, 1);
6146 /* Copy the current map */
6147 memcpy(map2
, map
, copy_map_size
);
6148 map2
->num_members
= prev_num_members
;
6150 imsm_set_array_size(dev
);
6151 super
->updates_pending
++;
6155 /* Handle dirty -> clean transititions, resync and reshape. Degraded and rebuild
6156 * states are handled in imsm_set_disk() with one exception, when a
6157 * resync is stopped due to a new failure this routine will set the
6158 * 'degraded' state for the array.
6160 static int imsm_set_array_state(struct active_array
*a
, int consistent
)
6162 int inst
= a
->info
.container_member
;
6163 struct intel_super
*super
= a
->container
->sb
;
6164 struct imsm_dev
*dev
= get_imsm_dev(super
, inst
);
6165 struct imsm_map
*map
= get_imsm_map(dev
, 0);
6166 int failed
= imsm_count_failed(super
, dev
);
6167 __u8 map_state
= imsm_check_degraded(super
, dev
, failed
);
6168 __u32 blocks_per_unit
;
6170 if (dev
->vol
.migr_state
&&
6171 dev
->vol
.migr_type
== MIGR_GEN_MIGR
) {
6172 /* array state change is blocked due to reshape action
6174 * - abort the reshape (if last_checkpoint is 0 and action!= reshape)
6175 * - finish the reshape (if last_checkpoint is big and action != reshape)
6176 * - update curr_migr_unit
6178 if (a
->curr_action
== reshape
) {
6179 /* still reshaping, maybe update curr_migr_unit */
6180 goto mark_checkpoint
;
6182 if (a
->last_checkpoint
== 0 && a
->prev_action
== reshape
) {
6183 /* for some reason we aborted the reshape.
6185 * disable automatic metadata rollback
6186 * user action is required to recover process
6189 struct imsm_map
*map2
= get_imsm_map(dev
, 1);
6190 dev
->vol
.migr_state
= 0;
6191 set_migr_type(dev
, 0);
6192 dev
->vol
.curr_migr_unit
= 0;
6193 memcpy(map
, map2
, sizeof_imsm_map(map2
));
6194 super
->updates_pending
++;
6197 if (a
->last_checkpoint
>= a
->info
.component_size
) {
6198 unsigned long long array_blocks
;
6202 used_disks
= imsm_num_data_members(dev
, 0);
6203 if (used_disks
> 0) {
6205 map
->blocks_per_member
*
6207 /* round array size down to closest MB
6209 array_blocks
= (array_blocks
6210 >> SECT_PER_MB_SHIFT
)
6211 << SECT_PER_MB_SHIFT
;
6212 a
->info
.custom_array_size
= array_blocks
;
6213 /* encourage manager to update array
6217 a
->check_reshape
= 1;
6219 /* finalize online capacity expansion/reshape */
6220 for (mdi
= a
->info
.devs
; mdi
; mdi
= mdi
->next
)
6222 mdi
->disk
.raid_disk
,
6225 imsm_progress_container_reshape(super
);
6230 /* before we activate this array handle any missing disks */
6231 if (consistent
== 2)
6232 handle_missing(super
, dev
);
6234 if (consistent
== 2 &&
6235 (!is_resync_complete(&a
->info
) ||
6236 map_state
!= IMSM_T_STATE_NORMAL
||
6237 dev
->vol
.migr_state
))
6240 if (is_resync_complete(&a
->info
)) {
6241 /* complete intialization / resync,
6242 * recovery and interrupted recovery is completed in
6245 if (is_resyncing(dev
)) {
6246 dprintf("imsm: mark resync done\n");
6247 end_migration(dev
, map_state
);
6248 super
->updates_pending
++;
6249 a
->last_checkpoint
= 0;
6251 } else if ((!is_resyncing(dev
) && !failed
) &&
6252 (imsm_reshape_blocks_arrays_changes(super
) == 0)) {
6253 /* mark the start of the init process if nothing is failed */
6254 dprintf("imsm: mark resync start\n");
6255 if (map
->map_state
== IMSM_T_STATE_UNINITIALIZED
)
6256 migrate(dev
, super
, IMSM_T_STATE_NORMAL
, MIGR_INIT
);
6258 migrate(dev
, super
, IMSM_T_STATE_NORMAL
, MIGR_REPAIR
);
6259 super
->updates_pending
++;
6263 /* skip checkpointing for general migration,
6264 * it is controlled in mdadm
6266 if (is_gen_migration(dev
))
6267 goto skip_mark_checkpoint
;
6269 /* check if we can update curr_migr_unit from resync_start, recovery_start */
6270 blocks_per_unit
= blocks_per_migr_unit(super
, dev
);
6271 if (blocks_per_unit
) {
6275 units
= a
->last_checkpoint
/ blocks_per_unit
;
6278 /* check that we did not overflow 32-bits, and that
6279 * curr_migr_unit needs updating
6281 if (units32
== units
&&
6283 __le32_to_cpu(dev
->vol
.curr_migr_unit
) != units32
) {
6284 dprintf("imsm: mark checkpoint (%u)\n", units32
);
6285 dev
->vol
.curr_migr_unit
= __cpu_to_le32(units32
);
6286 super
->updates_pending
++;
6290 skip_mark_checkpoint
:
6291 /* mark dirty / clean */
6292 if (dev
->vol
.dirty
!= !consistent
) {
6293 dprintf("imsm: mark '%s'\n", consistent
? "clean" : "dirty");
6298 super
->updates_pending
++;
6304 static void imsm_set_disk(struct active_array
*a
, int n
, int state
)
6306 int inst
= a
->info
.container_member
;
6307 struct intel_super
*super
= a
->container
->sb
;
6308 struct imsm_dev
*dev
= get_imsm_dev(super
, inst
);
6309 struct imsm_map
*map
= get_imsm_map(dev
, 0);
6310 struct imsm_disk
*disk
;
6315 if (n
> map
->num_members
)
6316 fprintf(stderr
, "imsm: set_disk %d out of range 0..%d\n",
6317 n
, map
->num_members
- 1);
6322 dprintf("imsm: set_disk %d:%x\n", n
, state
);
6324 ord
= get_imsm_ord_tbl_ent(dev
, n
, -1);
6325 disk
= get_imsm_disk(super
, ord_to_idx(ord
));
6327 /* check for new failures */
6328 if (state
& DS_FAULTY
) {
6329 if (mark_failure(dev
, disk
, ord_to_idx(ord
)))
6330 super
->updates_pending
++;
6333 /* check if in_sync */
6334 if (state
& DS_INSYNC
&& ord
& IMSM_ORD_REBUILD
&& is_rebuilding(dev
)) {
6335 struct imsm_map
*migr_map
= get_imsm_map(dev
, 1);
6337 set_imsm_ord_tbl_ent(migr_map
, n
, ord_to_idx(ord
));
6338 super
->updates_pending
++;
6341 failed
= imsm_count_failed(super
, dev
);
6342 map_state
= imsm_check_degraded(super
, dev
, failed
);
6344 /* check if recovery complete, newly degraded, or failed */
6345 if (map_state
== IMSM_T_STATE_NORMAL
&& is_rebuilding(dev
)) {
6346 end_migration(dev
, map_state
);
6347 map
= get_imsm_map(dev
, 0);
6348 map
->failed_disk_num
= ~0;
6349 super
->updates_pending
++;
6350 a
->last_checkpoint
= 0;
6351 } else if (map_state
== IMSM_T_STATE_DEGRADED
&&
6352 map
->map_state
!= map_state
&&
6353 !dev
->vol
.migr_state
) {
6354 dprintf("imsm: mark degraded\n");
6355 map
->map_state
= map_state
;
6356 super
->updates_pending
++;
6357 a
->last_checkpoint
= 0;
6358 } else if (map_state
== IMSM_T_STATE_FAILED
&&
6359 map
->map_state
!= map_state
) {
6360 dprintf("imsm: mark failed\n");
6361 end_migration(dev
, map_state
);
6362 super
->updates_pending
++;
6363 a
->last_checkpoint
= 0;
6364 } else if (is_gen_migration(dev
)) {
6365 dprintf("imsm: Detected General Migration in state: ");
6366 if (map_state
== IMSM_T_STATE_NORMAL
) {
6367 end_migration(dev
, map_state
);
6368 map
= get_imsm_map(dev
, 0);
6369 map
->failed_disk_num
= ~0;
6370 dprintf("normal\n");
6372 if (map_state
== IMSM_T_STATE_DEGRADED
) {
6373 printf("degraded\n");
6374 end_migration(dev
, map_state
);
6376 dprintf("failed\n");
6378 map
->map_state
= map_state
;
6380 super
->updates_pending
++;
6384 static int store_imsm_mpb(int fd
, struct imsm_super
*mpb
)
6387 __u32 mpb_size
= __le32_to_cpu(mpb
->mpb_size
);
6388 unsigned long long dsize
;
6389 unsigned long long sectors
;
6391 get_dev_size(fd
, NULL
, &dsize
);
6393 if (mpb_size
> 512) {
6394 /* -1 to account for anchor */
6395 sectors
= mpb_sectors(mpb
) - 1;
6397 /* write the extended mpb to the sectors preceeding the anchor */
6398 if (lseek64(fd
, dsize
- (512 * (2 + sectors
)), SEEK_SET
) < 0)
6401 if ((unsigned long long)write(fd
, buf
+ 512, 512 * sectors
)
6406 /* first block is stored on second to last sector of the disk */
6407 if (lseek64(fd
, dsize
- (512 * 2), SEEK_SET
) < 0)
6410 if (write(fd
, buf
, 512) != 512)
6416 static void imsm_sync_metadata(struct supertype
*container
)
6418 struct intel_super
*super
= container
->sb
;
6420 dprintf("sync metadata: %d\n", super
->updates_pending
);
6421 if (!super
->updates_pending
)
6424 write_super_imsm(container
, 0);
6426 super
->updates_pending
= 0;
6429 static struct dl
*imsm_readd(struct intel_super
*super
, int idx
, struct active_array
*a
)
6431 struct imsm_dev
*dev
= get_imsm_dev(super
, a
->info
.container_member
);
6432 int i
= get_imsm_disk_idx(dev
, idx
, -1);
6435 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
6439 if (dl
&& is_failed(&dl
->disk
))
6443 dprintf("%s: found %x:%x\n", __func__
, dl
->major
, dl
->minor
);
6448 static struct dl
*imsm_add_spare(struct intel_super
*super
, int slot
,
6449 struct active_array
*a
, int activate_new
,
6450 struct mdinfo
*additional_test_list
)
6452 struct imsm_dev
*dev
= get_imsm_dev(super
, a
->info
.container_member
);
6453 int idx
= get_imsm_disk_idx(dev
, slot
, -1);
6454 struct imsm_super
*mpb
= super
->anchor
;
6455 struct imsm_map
*map
;
6456 unsigned long long pos
;
6461 __u32 array_start
= 0;
6462 __u32 array_end
= 0;
6464 struct mdinfo
*test_list
;
6466 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
6467 /* If in this array, skip */
6468 for (d
= a
->info
.devs
; d
; d
= d
->next
)
6469 if (d
->state_fd
>= 0 &&
6470 d
->disk
.major
== dl
->major
&&
6471 d
->disk
.minor
== dl
->minor
) {
6472 dprintf("%x:%x already in array\n",
6473 dl
->major
, dl
->minor
);
6478 test_list
= additional_test_list
;
6480 if (test_list
->disk
.major
== dl
->major
&&
6481 test_list
->disk
.minor
== dl
->minor
) {
6482 dprintf("%x:%x already in additional test list\n",
6483 dl
->major
, dl
->minor
);
6486 test_list
= test_list
->next
;
6491 /* skip in use or failed drives */
6492 if (is_failed(&dl
->disk
) || idx
== dl
->index
||
6494 dprintf("%x:%x status (failed: %d index: %d)\n",
6495 dl
->major
, dl
->minor
, is_failed(&dl
->disk
), idx
);
6499 /* skip pure spares when we are looking for partially
6500 * assimilated drives
6502 if (dl
->index
== -1 && !activate_new
)
6505 /* Does this unused device have the requisite free space?
6506 * It needs to be able to cover all member volumes
6508 ex
= get_extents(super
, dl
);
6510 dprintf("cannot get extents\n");
6513 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
6514 dev
= get_imsm_dev(super
, i
);
6515 map
= get_imsm_map(dev
, 0);
6517 /* check if this disk is already a member of
6520 if (get_imsm_disk_slot(map
, dl
->index
) >= 0)
6526 array_start
= __le32_to_cpu(map
->pba_of_lba0
);
6527 array_end
= array_start
+
6528 __le32_to_cpu(map
->blocks_per_member
) - 1;
6531 /* check that we can start at pba_of_lba0 with
6532 * blocks_per_member of space
6534 if (array_start
>= pos
&& array_end
< ex
[j
].start
) {
6538 pos
= ex
[j
].start
+ ex
[j
].size
;
6540 } while (ex
[j
-1].size
);
6547 if (i
< mpb
->num_raid_devs
) {
6548 dprintf("%x:%x does not have %u to %u available\n",
6549 dl
->major
, dl
->minor
, array_start
, array_end
);
6560 static int imsm_rebuild_allowed(struct supertype
*cont
, int dev_idx
, int failed
)
6562 struct imsm_dev
*dev2
;
6563 struct imsm_map
*map
;
6569 dev2
= get_imsm_dev(cont
->sb
, dev_idx
);
6571 state
= imsm_check_degraded(cont
->sb
, dev2
, failed
);
6572 if (state
== IMSM_T_STATE_FAILED
) {
6573 map
= get_imsm_map(dev2
, 0);
6576 for (slot
= 0; slot
< map
->num_members
; slot
++) {
6578 * Check if failed disks are deleted from intel
6579 * disk list or are marked to be deleted
6581 idx
= get_imsm_disk_idx(dev2
, slot
, -1);
6582 idisk
= get_imsm_dl_disk(cont
->sb
, idx
);
6584 * Do not rebuild the array if failed disks
6585 * from failed sub-array are not removed from
6589 is_failed(&idisk
->disk
) &&
6590 (idisk
->action
!= DISK_REMOVE
))
6598 static struct mdinfo
*imsm_activate_spare(struct active_array
*a
,
6599 struct metadata_update
**updates
)
6602 * Find a device with unused free space and use it to replace a
6603 * failed/vacant region in an array. We replace failed regions one a
6604 * array at a time. The result is that a new spare disk will be added
6605 * to the first failed array and after the monitor has finished
6606 * propagating failures the remainder will be consumed.
6608 * FIXME add a capability for mdmon to request spares from another
6612 struct intel_super
*super
= a
->container
->sb
;
6613 int inst
= a
->info
.container_member
;
6614 struct imsm_dev
*dev
= get_imsm_dev(super
, inst
);
6615 struct imsm_map
*map
= get_imsm_map(dev
, 0);
6616 int failed
= a
->info
.array
.raid_disks
;
6617 struct mdinfo
*rv
= NULL
;
6620 struct metadata_update
*mu
;
6622 struct imsm_update_activate_spare
*u
;
6627 for (d
= a
->info
.devs
; d
; d
= d
->next
) {
6628 if ((d
->curr_state
& DS_FAULTY
) &&
6630 /* wait for Removal to happen */
6632 if (d
->state_fd
>= 0)
6636 dprintf("imsm: activate spare: inst=%d failed=%d (%d) level=%d\n",
6637 inst
, failed
, a
->info
.array
.raid_disks
, a
->info
.array
.level
);
6639 if (imsm_reshape_blocks_arrays_changes(super
))
6642 if (a
->info
.array
.level
== 4)
6643 /* No repair for takeovered array
6644 * imsm doesn't support raid4
6648 if (imsm_check_degraded(super
, dev
, failed
) != IMSM_T_STATE_DEGRADED
)
6652 * If there are any failed disks check state of the other volume.
6653 * Block rebuild if the another one is failed until failed disks
6654 * are removed from container.
6657 dprintf("found failed disks in %.*s, check if there another"
6658 "failed sub-array.\n",
6659 MAX_RAID_SERIAL_LEN
, dev
->volume
);
6660 /* check if states of the other volumes allow for rebuild */
6661 for (i
= 0; i
< super
->anchor
->num_raid_devs
; i
++) {
6663 allowed
= imsm_rebuild_allowed(a
->container
,
6671 /* For each slot, if it is not working, find a spare */
6672 for (i
= 0; i
< a
->info
.array
.raid_disks
; i
++) {
6673 for (d
= a
->info
.devs
; d
; d
= d
->next
)
6674 if (d
->disk
.raid_disk
== i
)
6676 dprintf("found %d: %p %x\n", i
, d
, d
?d
->curr_state
:0);
6677 if (d
&& (d
->state_fd
>= 0))
6681 * OK, this device needs recovery. Try to re-add the
6682 * previous occupant of this slot, if this fails see if
6683 * we can continue the assimilation of a spare that was
6684 * partially assimilated, finally try to activate a new
6687 dl
= imsm_readd(super
, i
, a
);
6689 dl
= imsm_add_spare(super
, i
, a
, 0, rv
);
6691 dl
= imsm_add_spare(super
, i
, a
, 1, rv
);
6695 /* found a usable disk with enough space */
6696 di
= malloc(sizeof(*di
));
6699 memset(di
, 0, sizeof(*di
));
6701 /* dl->index will be -1 in the case we are activating a
6702 * pristine spare. imsm_process_update() will create a
6703 * new index in this case. Once a disk is found to be
6704 * failed in all member arrays it is kicked from the
6707 di
->disk
.number
= dl
->index
;
6709 /* (ab)use di->devs to store a pointer to the device
6712 di
->devs
= (struct mdinfo
*) dl
;
6714 di
->disk
.raid_disk
= i
;
6715 di
->disk
.major
= dl
->major
;
6716 di
->disk
.minor
= dl
->minor
;
6718 di
->recovery_start
= 0;
6719 di
->data_offset
= __le32_to_cpu(map
->pba_of_lba0
);
6720 di
->component_size
= a
->info
.component_size
;
6721 di
->container_member
= inst
;
6722 super
->random
= random32();
6726 dprintf("%x:%x to be %d at %llu\n", dl
->major
, dl
->minor
,
6727 i
, di
->data_offset
);
6731 /* No spares found */
6733 /* Now 'rv' has a list of devices to return.
6734 * Create a metadata_update record to update the
6735 * disk_ord_tbl for the array
6737 mu
= malloc(sizeof(*mu
));
6739 mu
->buf
= malloc(sizeof(struct imsm_update_activate_spare
) * num_spares
);
6740 if (mu
->buf
== NULL
) {
6747 struct mdinfo
*n
= rv
->next
;
6756 mu
->space_list
= NULL
;
6757 mu
->len
= sizeof(struct imsm_update_activate_spare
) * num_spares
;
6758 mu
->next
= *updates
;
6759 u
= (struct imsm_update_activate_spare
*) mu
->buf
;
6761 for (di
= rv
; di
; di
= di
->next
) {
6762 u
->type
= update_activate_spare
;
6763 u
->dl
= (struct dl
*) di
->devs
;
6765 u
->slot
= di
->disk
.raid_disk
;
6776 static int disks_overlap(struct intel_super
*super
, int idx
, struct imsm_update_create_array
*u
)
6778 struct imsm_dev
*dev
= get_imsm_dev(super
, idx
);
6779 struct imsm_map
*map
= get_imsm_map(dev
, 0);
6780 struct imsm_map
*new_map
= get_imsm_map(&u
->dev
, 0);
6781 struct disk_info
*inf
= get_disk_info(u
);
6782 struct imsm_disk
*disk
;
6786 for (i
= 0; i
< map
->num_members
; i
++) {
6787 disk
= get_imsm_disk(super
, get_imsm_disk_idx(dev
, i
, -1));
6788 for (j
= 0; j
< new_map
->num_members
; j
++)
6789 if (serialcmp(disk
->serial
, inf
[j
].serial
) == 0)
6797 static struct dl
*get_disk_super(struct intel_super
*super
, int major
, int minor
)
6799 struct dl
*dl
= NULL
;
6800 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
6801 if ((dl
->major
== major
) && (dl
->minor
== minor
))
6806 static int remove_disk_super(struct intel_super
*super
, int major
, int minor
)
6808 struct dl
*prev
= NULL
;
6812 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
6813 if ((dl
->major
== major
) && (dl
->minor
== minor
)) {
6816 prev
->next
= dl
->next
;
6818 super
->disks
= dl
->next
;
6820 __free_imsm_disk(dl
);
6821 dprintf("%s: removed %x:%x\n",
6822 __func__
, major
, minor
);
6830 static void imsm_delete(struct intel_super
*super
, struct dl
**dlp
, unsigned index
);
6832 static int add_remove_disk_update(struct intel_super
*super
)
6834 int check_degraded
= 0;
6835 struct dl
*disk
= NULL
;
6836 /* add/remove some spares to/from the metadata/contrainer */
6837 while (super
->disk_mgmt_list
) {
6838 struct dl
*disk_cfg
;
6840 disk_cfg
= super
->disk_mgmt_list
;
6841 super
->disk_mgmt_list
= disk_cfg
->next
;
6842 disk_cfg
->next
= NULL
;
6844 if (disk_cfg
->action
== DISK_ADD
) {
6845 disk_cfg
->next
= super
->disks
;
6846 super
->disks
= disk_cfg
;
6848 dprintf("%s: added %x:%x\n",
6849 __func__
, disk_cfg
->major
,
6851 } else if (disk_cfg
->action
== DISK_REMOVE
) {
6852 dprintf("Disk remove action processed: %x.%x\n",
6853 disk_cfg
->major
, disk_cfg
->minor
);
6854 disk
= get_disk_super(super
,
6858 /* store action status */
6859 disk
->action
= DISK_REMOVE
;
6860 /* remove spare disks only */
6861 if (disk
->index
== -1) {
6862 remove_disk_super(super
,
6867 /* release allocate disk structure */
6868 __free_imsm_disk(disk_cfg
);
6871 return check_degraded
;
6875 static int apply_reshape_migration_update(struct imsm_update_reshape_migration
*u
,
6876 struct intel_super
*super
,
6879 struct intel_dev
*id
;
6880 void **tofree
= NULL
;
6883 dprintf("apply_reshape_migration_update()\n");
6884 if ((u
->subdev
< 0) ||
6886 dprintf("imsm: Error: Wrong subdev: %i\n", u
->subdev
);
6889 if ((space_list
== NULL
) || (*space_list
== NULL
)) {
6890 dprintf("imsm: Error: Memory is not allocated\n");
6894 for (id
= super
->devlist
; id
; id
= id
->next
) {
6895 if (id
->index
== (unsigned)u
->subdev
) {
6896 struct imsm_dev
*dev
= get_imsm_dev(super
, u
->subdev
);
6897 struct imsm_map
*map
;
6898 struct imsm_dev
*new_dev
=
6899 (struct imsm_dev
*)*space_list
;
6900 struct imsm_map
*migr_map
= get_imsm_map(dev
, 1);
6902 struct dl
*new_disk
;
6904 if (new_dev
== NULL
)
6906 *space_list
= **space_list
;
6907 memcpy(new_dev
, dev
, sizeof_imsm_dev(dev
, 0));
6908 map
= get_imsm_map(new_dev
, 0);
6910 dprintf("imsm: Error: migration in progress");
6914 to_state
= map
->map_state
;
6915 if ((u
->new_level
== 5) && (map
->raid_level
== 0)) {
6917 /* this should not happen */
6918 if (u
->new_disks
[0] < 0) {
6919 map
->failed_disk_num
=
6920 map
->num_members
- 1;
6921 to_state
= IMSM_T_STATE_DEGRADED
;
6923 to_state
= IMSM_T_STATE_NORMAL
;
6925 migrate(new_dev
, super
, to_state
, MIGR_GEN_MIGR
);
6926 if (u
->new_level
> -1)
6927 map
->raid_level
= u
->new_level
;
6928 migr_map
= get_imsm_map(new_dev
, 1);
6929 if ((u
->new_level
== 5) &&
6930 (migr_map
->raid_level
== 0)) {
6931 int ord
= map
->num_members
- 1;
6932 migr_map
->num_members
--;
6933 if (u
->new_disks
[0] < 0)
6934 ord
|= IMSM_ORD_REBUILD
;
6935 set_imsm_ord_tbl_ent(map
,
6936 map
->num_members
- 1,
6940 tofree
= (void **)dev
;
6942 /* update chunk size
6944 if (u
->new_chunksize
> 0)
6945 map
->blocks_per_strip
=
6946 __cpu_to_le16(u
->new_chunksize
* 2);
6950 if ((u
->new_level
!= 5) ||
6951 (migr_map
->raid_level
!= 0) ||
6952 (migr_map
->raid_level
== map
->raid_level
))
6955 if (u
->new_disks
[0] >= 0) {
6958 new_disk
= get_disk_super(super
,
6959 major(u
->new_disks
[0]),
6960 minor(u
->new_disks
[0]));
6961 dprintf("imsm: new disk for reshape is: %i:%i "
6962 "(%p, index = %i)\n",
6963 major(u
->new_disks
[0]),
6964 minor(u
->new_disks
[0]),
6965 new_disk
, new_disk
->index
);
6966 if (new_disk
== NULL
)
6967 goto error_disk_add
;
6969 new_disk
->index
= map
->num_members
- 1;
6970 /* slot to fill in autolayout
6972 new_disk
->raiddisk
= new_disk
->index
;
6973 new_disk
->disk
.status
|= CONFIGURED_DISK
;
6974 new_disk
->disk
.status
&= ~SPARE_DISK
;
6976 goto error_disk_add
;
6979 *tofree
= *space_list
;
6980 /* calculate new size
6982 imsm_set_array_size(new_dev
);
6989 *space_list
= tofree
;
6993 dprintf("Error: imsm: Cannot find disk.\n");
6997 static int apply_update_activate_spare(struct imsm_update_activate_spare
*u
,
6998 struct intel_super
*super
,
6999 struct active_array
*active_array
)
7001 struct imsm_super
*mpb
= super
->anchor
;
7002 struct imsm_dev
*dev
= get_imsm_dev(super
, u
->array
);
7003 struct imsm_map
*map
= get_imsm_map(dev
, 0);
7004 struct imsm_map
*migr_map
;
7005 struct active_array
*a
;
7006 struct imsm_disk
*disk
;
7013 int second_map_created
= 0;
7015 for (; u
; u
= u
->next
) {
7016 victim
= get_imsm_disk_idx(dev
, u
->slot
, -1);
7021 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
7026 fprintf(stderr
, "error: imsm_activate_spare passed "
7027 "an unknown disk (index: %d)\n",
7032 /* count failures (excluding rebuilds and the victim)
7033 * to determine map[0] state
7036 for (i
= 0; i
< map
->num_members
; i
++) {
7039 disk
= get_imsm_disk(super
,
7040 get_imsm_disk_idx(dev
, i
, -1));
7041 if (!disk
|| is_failed(disk
))
7045 /* adding a pristine spare, assign a new index */
7046 if (dl
->index
< 0) {
7047 dl
->index
= super
->anchor
->num_disks
;
7048 super
->anchor
->num_disks
++;
7051 disk
->status
|= CONFIGURED_DISK
;
7052 disk
->status
&= ~SPARE_DISK
;
7055 to_state
= imsm_check_degraded(super
, dev
, failed
);
7056 if (!second_map_created
) {
7057 second_map_created
= 1;
7058 map
->map_state
= IMSM_T_STATE_DEGRADED
;
7059 migrate(dev
, super
, to_state
, MIGR_REBUILD
);
7061 map
->map_state
= to_state
;
7062 migr_map
= get_imsm_map(dev
, 1);
7063 set_imsm_ord_tbl_ent(map
, u
->slot
, dl
->index
);
7064 set_imsm_ord_tbl_ent(migr_map
, u
->slot
,
7065 dl
->index
| IMSM_ORD_REBUILD
);
7067 /* update the family_num to mark a new container
7068 * generation, being careful to record the existing
7069 * family_num in orig_family_num to clean up after
7070 * earlier mdadm versions that neglected to set it.
7072 if (mpb
->orig_family_num
== 0)
7073 mpb
->orig_family_num
= mpb
->family_num
;
7074 mpb
->family_num
+= super
->random
;
7076 /* count arrays using the victim in the metadata */
7078 for (a
= active_array
; a
; a
= a
->next
) {
7079 dev
= get_imsm_dev(super
, a
->info
.container_member
);
7080 map
= get_imsm_map(dev
, 0);
7082 if (get_imsm_disk_slot(map
, victim
) >= 0)
7086 /* delete the victim if it is no longer being
7092 /* We know that 'manager' isn't touching anything,
7093 * so it is safe to delete
7095 for (dlp
= &super
->disks
; *dlp
; dlp
= &(*dlp
)->next
)
7096 if ((*dlp
)->index
== victim
)
7099 /* victim may be on the missing list */
7101 for (dlp
= &super
->missing
; *dlp
;
7102 dlp
= &(*dlp
)->next
)
7103 if ((*dlp
)->index
== victim
)
7105 imsm_delete(super
, dlp
, victim
);
7112 static int apply_reshape_container_disks_update(struct imsm_update_reshape
*u
,
7113 struct intel_super
*super
,
7116 struct dl
*new_disk
;
7117 struct intel_dev
*id
;
7119 int delta_disks
= u
->new_raid_disks
- u
->old_raid_disks
;
7120 int disk_count
= u
->old_raid_disks
;
7121 void **tofree
= NULL
;
7122 int devices_to_reshape
= 1;
7123 struct imsm_super
*mpb
= super
->anchor
;
7125 unsigned int dev_id
;
7127 dprintf("imsm: apply_reshape_container_disks_update()\n");
7129 /* enable spares to use in array */
7130 for (i
= 0; i
< delta_disks
; i
++) {
7131 new_disk
= get_disk_super(super
,
7132 major(u
->new_disks
[i
]),
7133 minor(u
->new_disks
[i
]));
7134 dprintf("imsm: new disk for reshape is: %i:%i "
7135 "(%p, index = %i)\n",
7136 major(u
->new_disks
[i
]), minor(u
->new_disks
[i
]),
7137 new_disk
, new_disk
->index
);
7138 if ((new_disk
== NULL
) ||
7139 ((new_disk
->index
>= 0) &&
7140 (new_disk
->index
< u
->old_raid_disks
)))
7141 goto update_reshape_exit
;
7142 new_disk
->index
= disk_count
++;
7143 /* slot to fill in autolayout
7145 new_disk
->raiddisk
= new_disk
->index
;
7146 new_disk
->disk
.status
|=
7148 new_disk
->disk
.status
&= ~SPARE_DISK
;
7151 dprintf("imsm: volume set mpb->num_raid_devs = %i\n",
7152 mpb
->num_raid_devs
);
7153 /* manage changes in volume
7155 for (dev_id
= 0; dev_id
< mpb
->num_raid_devs
; dev_id
++) {
7156 void **sp
= *space_list
;
7157 struct imsm_dev
*newdev
;
7158 struct imsm_map
*newmap
, *oldmap
;
7160 for (id
= super
->devlist
; id
; id
= id
->next
) {
7161 if (id
->index
== dev_id
)
7170 /* Copy the dev, but not (all of) the map */
7171 memcpy(newdev
, id
->dev
, sizeof(*newdev
));
7172 oldmap
= get_imsm_map(id
->dev
, 0);
7173 newmap
= get_imsm_map(newdev
, 0);
7174 /* Copy the current map */
7175 memcpy(newmap
, oldmap
, sizeof_imsm_map(oldmap
));
7176 /* update one device only
7178 if (devices_to_reshape
) {
7179 dprintf("imsm: modifying subdev: %i\n",
7181 devices_to_reshape
--;
7182 newdev
->vol
.migr_state
= 1;
7183 newdev
->vol
.curr_migr_unit
= 0;
7184 set_migr_type(newdev
, MIGR_GEN_MIGR
);
7185 newmap
->num_members
= u
->new_raid_disks
;
7186 for (i
= 0; i
< delta_disks
; i
++) {
7187 set_imsm_ord_tbl_ent(newmap
,
7188 u
->old_raid_disks
+ i
,
7189 u
->old_raid_disks
+ i
);
7191 /* New map is correct, now need to save old map
7193 newmap
= get_imsm_map(newdev
, 1);
7194 memcpy(newmap
, oldmap
, sizeof_imsm_map(oldmap
));
7196 imsm_set_array_size(newdev
);
7199 sp
= (void **)id
->dev
;
7204 /* Clear migration record */
7205 memset(super
->migr_rec
, 0, sizeof(struct migr_record
));
7208 *space_list
= tofree
;
7211 update_reshape_exit
:
7216 static int apply_takeover_update(struct imsm_update_takeover
*u
,
7217 struct intel_super
*super
,
7220 struct imsm_dev
*dev
= NULL
;
7221 struct intel_dev
*dv
;
7222 struct imsm_dev
*dev_new
;
7223 struct imsm_map
*map
;
7227 for (dv
= super
->devlist
; dv
; dv
= dv
->next
)
7228 if (dv
->index
== (unsigned int)u
->subarray
) {
7236 map
= get_imsm_map(dev
, 0);
7238 if (u
->direction
== R10_TO_R0
) {
7239 /* Number of failed disks must be half of initial disk number */
7240 if (imsm_count_failed(super
, dev
) != (map
->num_members
/ 2))
7243 /* iterate through devices to mark removed disks as spare */
7244 for (dm
= super
->disks
; dm
; dm
= dm
->next
) {
7245 if (dm
->disk
.status
& FAILED_DISK
) {
7246 int idx
= dm
->index
;
7247 /* update indexes on the disk list */
7248 /* FIXME this loop-with-the-loop looks wrong, I'm not convinced
7249 the index values will end up being correct.... NB */
7250 for (du
= super
->disks
; du
; du
= du
->next
)
7251 if (du
->index
> idx
)
7253 /* mark as spare disk */
7258 map
->num_members
= map
->num_members
/ 2;
7259 map
->map_state
= IMSM_T_STATE_NORMAL
;
7260 map
->num_domains
= 1;
7261 map
->raid_level
= 0;
7262 map
->failed_disk_num
= -1;
7265 if (u
->direction
== R0_TO_R10
) {
7267 /* update slots in current disk list */
7268 for (dm
= super
->disks
; dm
; dm
= dm
->next
) {
7272 /* create new *missing* disks */
7273 for (i
= 0; i
< map
->num_members
; i
++) {
7274 space
= *space_list
;
7277 *space_list
= *space
;
7279 memcpy(du
, super
->disks
, sizeof(*du
));
7283 du
->index
= (i
* 2) + 1;
7284 sprintf((char *)du
->disk
.serial
,
7285 " MISSING_%d", du
->index
);
7286 sprintf((char *)du
->serial
,
7287 "MISSING_%d", du
->index
);
7288 du
->next
= super
->missing
;
7289 super
->missing
= du
;
7291 /* create new dev and map */
7292 space
= *space_list
;
7295 *space_list
= *space
;
7296 dev_new
= (void *)space
;
7297 memcpy(dev_new
, dev
, sizeof(*dev
));
7298 /* update new map */
7299 map
= get_imsm_map(dev_new
, 0);
7300 map
->num_members
= map
->num_members
* 2;
7301 map
->map_state
= IMSM_T_STATE_DEGRADED
;
7302 map
->num_domains
= 2;
7303 map
->raid_level
= 1;
7304 /* replace dev<->dev_new */
7307 /* update disk order table */
7308 for (du
= super
->disks
; du
; du
= du
->next
)
7310 set_imsm_ord_tbl_ent(map
, du
->index
, du
->index
);
7311 for (du
= super
->missing
; du
; du
= du
->next
)
7312 if (du
->index
>= 0) {
7313 set_imsm_ord_tbl_ent(map
, du
->index
, du
->index
);
7314 mark_missing(dv
->dev
, &du
->disk
, du
->index
);
7320 static void imsm_process_update(struct supertype
*st
,
7321 struct metadata_update
*update
)
7324 * crack open the metadata_update envelope to find the update record
7325 * update can be one of:
7326 * update_reshape_container_disks - all the arrays in the container
7327 * are being reshaped to have more devices. We need to mark
7328 * the arrays for general migration and convert selected spares
7329 * into active devices.
7330 * update_activate_spare - a spare device has replaced a failed
7331 * device in an array, update the disk_ord_tbl. If this disk is
7332 * present in all member arrays then also clear the SPARE_DISK
7334 * update_create_array
7336 * update_rename_array
7337 * update_add_remove_disk
7339 struct intel_super
*super
= st
->sb
;
7340 struct imsm_super
*mpb
;
7341 enum imsm_update_type type
= *(enum imsm_update_type
*) update
->buf
;
7343 /* update requires a larger buf but the allocation failed */
7344 if (super
->next_len
&& !super
->next_buf
) {
7345 super
->next_len
= 0;
7349 if (super
->next_buf
) {
7350 memcpy(super
->next_buf
, super
->buf
, super
->len
);
7352 super
->len
= super
->next_len
;
7353 super
->buf
= super
->next_buf
;
7355 super
->next_len
= 0;
7356 super
->next_buf
= NULL
;
7359 mpb
= super
->anchor
;
7362 case update_general_migration_checkpoint
: {
7363 struct intel_dev
*id
;
7364 struct imsm_update_general_migration_checkpoint
*u
=
7365 (void *)update
->buf
;
7367 dprintf("imsm: process_update() "
7368 "for update_general_migration_checkpoint called\n");
7370 /* find device under general migration */
7371 for (id
= super
->devlist
; id
; id
= id
->next
) {
7372 if (is_gen_migration(id
->dev
)) {
7373 id
->dev
->vol
.curr_migr_unit
=
7374 __cpu_to_le32(u
->curr_migr_unit
);
7375 super
->updates_pending
++;
7380 case update_takeover
: {
7381 struct imsm_update_takeover
*u
= (void *)update
->buf
;
7382 if (apply_takeover_update(u
, super
, &update
->space_list
)) {
7383 imsm_update_version_info(super
);
7384 super
->updates_pending
++;
7389 case update_reshape_container_disks
: {
7390 struct imsm_update_reshape
*u
= (void *)update
->buf
;
7391 if (apply_reshape_container_disks_update(
7392 u
, super
, &update
->space_list
))
7393 super
->updates_pending
++;
7396 case update_reshape_migration
: {
7397 struct imsm_update_reshape_migration
*u
= (void *)update
->buf
;
7398 if (apply_reshape_migration_update(
7399 u
, super
, &update
->space_list
))
7400 super
->updates_pending
++;
7403 case update_activate_spare
: {
7404 struct imsm_update_activate_spare
*u
= (void *) update
->buf
;
7405 if (apply_update_activate_spare(u
, super
, st
->arrays
))
7406 super
->updates_pending
++;
7409 case update_create_array
: {
7410 /* someone wants to create a new array, we need to be aware of
7411 * a few races/collisions:
7412 * 1/ 'Create' called by two separate instances of mdadm
7413 * 2/ 'Create' versus 'activate_spare': mdadm has chosen
7414 * devices that have since been assimilated via
7416 * In the event this update can not be carried out mdadm will
7417 * (FIX ME) notice that its update did not take hold.
7419 struct imsm_update_create_array
*u
= (void *) update
->buf
;
7420 struct intel_dev
*dv
;
7421 struct imsm_dev
*dev
;
7422 struct imsm_map
*map
, *new_map
;
7423 unsigned long long start
, end
;
7424 unsigned long long new_start
, new_end
;
7426 struct disk_info
*inf
;
7429 /* handle racing creates: first come first serve */
7430 if (u
->dev_idx
< mpb
->num_raid_devs
) {
7431 dprintf("%s: subarray %d already defined\n",
7432 __func__
, u
->dev_idx
);
7436 /* check update is next in sequence */
7437 if (u
->dev_idx
!= mpb
->num_raid_devs
) {
7438 dprintf("%s: can not create array %d expected index %d\n",
7439 __func__
, u
->dev_idx
, mpb
->num_raid_devs
);
7443 new_map
= get_imsm_map(&u
->dev
, 0);
7444 new_start
= __le32_to_cpu(new_map
->pba_of_lba0
);
7445 new_end
= new_start
+ __le32_to_cpu(new_map
->blocks_per_member
);
7446 inf
= get_disk_info(u
);
7448 /* handle activate_spare versus create race:
7449 * check to make sure that overlapping arrays do not include
7452 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
7453 dev
= get_imsm_dev(super
, i
);
7454 map
= get_imsm_map(dev
, 0);
7455 start
= __le32_to_cpu(map
->pba_of_lba0
);
7456 end
= start
+ __le32_to_cpu(map
->blocks_per_member
);
7457 if ((new_start
>= start
&& new_start
<= end
) ||
7458 (start
>= new_start
&& start
<= new_end
))
7463 if (disks_overlap(super
, i
, u
)) {
7464 dprintf("%s: arrays overlap\n", __func__
);
7469 /* check that prepare update was successful */
7470 if (!update
->space
) {
7471 dprintf("%s: prepare update failed\n", __func__
);
7475 /* check that all disks are still active before committing
7476 * changes. FIXME: could we instead handle this by creating a
7477 * degraded array? That's probably not what the user expects,
7478 * so better to drop this update on the floor.
7480 for (i
= 0; i
< new_map
->num_members
; i
++) {
7481 dl
= serial_to_dl(inf
[i
].serial
, super
);
7483 dprintf("%s: disk disappeared\n", __func__
);
7488 super
->updates_pending
++;
7490 /* convert spares to members and fixup ord_tbl */
7491 for (i
= 0; i
< new_map
->num_members
; i
++) {
7492 dl
= serial_to_dl(inf
[i
].serial
, super
);
7493 if (dl
->index
== -1) {
7494 dl
->index
= mpb
->num_disks
;
7496 dl
->disk
.status
|= CONFIGURED_DISK
;
7497 dl
->disk
.status
&= ~SPARE_DISK
;
7499 set_imsm_ord_tbl_ent(new_map
, i
, dl
->index
);
7504 update
->space
= NULL
;
7505 imsm_copy_dev(dev
, &u
->dev
);
7506 dv
->index
= u
->dev_idx
;
7507 dv
->next
= super
->devlist
;
7508 super
->devlist
= dv
;
7509 mpb
->num_raid_devs
++;
7511 imsm_update_version_info(super
);
7514 /* mdmon knows how to release update->space, but not
7515 * ((struct intel_dev *) update->space)->dev
7517 if (update
->space
) {
7523 case update_kill_array
: {
7524 struct imsm_update_kill_array
*u
= (void *) update
->buf
;
7525 int victim
= u
->dev_idx
;
7526 struct active_array
*a
;
7527 struct intel_dev
**dp
;
7528 struct imsm_dev
*dev
;
7530 /* sanity check that we are not affecting the uuid of
7531 * active arrays, or deleting an active array
7533 * FIXME when immutable ids are available, but note that
7534 * we'll also need to fixup the invalidated/active
7535 * subarray indexes in mdstat
7537 for (a
= st
->arrays
; a
; a
= a
->next
)
7538 if (a
->info
.container_member
>= victim
)
7540 /* by definition if mdmon is running at least one array
7541 * is active in the container, so checking
7542 * mpb->num_raid_devs is just extra paranoia
7544 dev
= get_imsm_dev(super
, victim
);
7545 if (a
|| !dev
|| mpb
->num_raid_devs
== 1) {
7546 dprintf("failed to delete subarray-%d\n", victim
);
7550 for (dp
= &super
->devlist
; *dp
;)
7551 if ((*dp
)->index
== (unsigned)super
->current_vol
) {
7554 if ((*dp
)->index
> (unsigned)victim
)
7558 mpb
->num_raid_devs
--;
7559 super
->updates_pending
++;
7562 case update_rename_array
: {
7563 struct imsm_update_rename_array
*u
= (void *) update
->buf
;
7564 char name
[MAX_RAID_SERIAL_LEN
+1];
7565 int target
= u
->dev_idx
;
7566 struct active_array
*a
;
7567 struct imsm_dev
*dev
;
7569 /* sanity check that we are not affecting the uuid of
7572 snprintf(name
, MAX_RAID_SERIAL_LEN
, "%s", (char *) u
->name
);
7573 name
[MAX_RAID_SERIAL_LEN
] = '\0';
7574 for (a
= st
->arrays
; a
; a
= a
->next
)
7575 if (a
->info
.container_member
== target
)
7577 dev
= get_imsm_dev(super
, u
->dev_idx
);
7578 if (a
|| !dev
|| !check_name(super
, name
, 1)) {
7579 dprintf("failed to rename subarray-%d\n", target
);
7583 snprintf((char *) dev
->volume
, MAX_RAID_SERIAL_LEN
, "%s", name
);
7584 super
->updates_pending
++;
7587 case update_add_remove_disk
: {
7588 /* we may be able to repair some arrays if disks are
7589 * being added, check teh status of add_remove_disk
7590 * if discs has been added.
7592 if (add_remove_disk_update(super
)) {
7593 struct active_array
*a
;
7595 super
->updates_pending
++;
7596 for (a
= st
->arrays
; a
; a
= a
->next
)
7597 a
->check_degraded
= 1;
7602 fprintf(stderr
, "error: unsuported process update type:"
7603 "(type: %d)\n", type
);
7607 static struct mdinfo
*get_spares_for_grow(struct supertype
*st
);
7609 static void imsm_prepare_update(struct supertype
*st
,
7610 struct metadata_update
*update
)
7613 * Allocate space to hold new disk entries, raid-device entries or a new
7614 * mpb if necessary. The manager synchronously waits for updates to
7615 * complete in the monitor, so new mpb buffers allocated here can be
7616 * integrated by the monitor thread without worrying about live pointers
7617 * in the manager thread.
7619 enum imsm_update_type type
= *(enum imsm_update_type
*) update
->buf
;
7620 struct intel_super
*super
= st
->sb
;
7621 struct imsm_super
*mpb
= super
->anchor
;
7626 case update_general_migration_checkpoint
:
7627 dprintf("imsm: prepare_update() "
7628 "for update_general_migration_checkpoint called\n");
7630 case update_takeover
: {
7631 struct imsm_update_takeover
*u
= (void *)update
->buf
;
7632 if (u
->direction
== R0_TO_R10
) {
7633 void **tail
= (void **)&update
->space_list
;
7634 struct imsm_dev
*dev
= get_imsm_dev(super
, u
->subarray
);
7635 struct imsm_map
*map
= get_imsm_map(dev
, 0);
7636 int num_members
= map
->num_members
;
7640 /* allocate memory for added disks */
7641 for (i
= 0; i
< num_members
; i
++) {
7642 size
= sizeof(struct dl
);
7643 space
= malloc(size
);
7652 /* allocate memory for new device */
7653 size
= sizeof_imsm_dev(super
->devlist
->dev
, 0) +
7654 (num_members
* sizeof(__u32
));
7655 space
= malloc(size
);
7664 len
= disks_to_mpb_size(num_members
* 2);
7666 /* if allocation didn't success, free buffer */
7667 while (update
->space_list
) {
7668 void **sp
= update
->space_list
;
7669 update
->space_list
= *sp
;
7677 case update_reshape_container_disks
: {
7678 /* Every raid device in the container is about to
7679 * gain some more devices, and we will enter a
7681 * So each 'imsm_map' will be bigger, and the imsm_vol
7682 * will now hold 2 of them.
7683 * Thus we need new 'struct imsm_dev' allocations sized
7684 * as sizeof_imsm_dev but with more devices in both maps.
7686 struct imsm_update_reshape
*u
= (void *)update
->buf
;
7687 struct intel_dev
*dl
;
7688 void **space_tail
= (void**)&update
->space_list
;
7690 dprintf("imsm: imsm_prepare_update() for update_reshape\n");
7692 for (dl
= super
->devlist
; dl
; dl
= dl
->next
) {
7693 int size
= sizeof_imsm_dev(dl
->dev
, 1);
7695 if (u
->new_raid_disks
> u
->old_raid_disks
)
7696 size
+= sizeof(__u32
)*2*
7697 (u
->new_raid_disks
- u
->old_raid_disks
);
7706 len
= disks_to_mpb_size(u
->new_raid_disks
);
7707 dprintf("New anchor length is %llu\n", (unsigned long long)len
);
7710 case update_reshape_migration
: {
7711 /* for migration level 0->5 we need to add disks
7712 * so the same as for container operation we will copy
7713 * device to the bigger location.
7714 * in memory prepared device and new disk area are prepared
7715 * for usage in process update
7717 struct imsm_update_reshape_migration
*u
= (void *)update
->buf
;
7718 struct intel_dev
*id
;
7719 void **space_tail
= (void **)&update
->space_list
;
7722 int current_level
= -1;
7724 dprintf("imsm: imsm_prepare_update() for update_reshape\n");
7726 /* add space for bigger array in update
7728 for (id
= super
->devlist
; id
; id
= id
->next
) {
7729 if (id
->index
== (unsigned)u
->subdev
) {
7730 size
= sizeof_imsm_dev(id
->dev
, 1);
7731 if (u
->new_raid_disks
> u
->old_raid_disks
)
7732 size
+= sizeof(__u32
)*2*
7733 (u
->new_raid_disks
- u
->old_raid_disks
);
7743 if (update
->space_list
== NULL
)
7746 /* add space for disk in update
7748 size
= sizeof(struct dl
);
7751 free(update
->space_list
);
7752 update
->space_list
= NULL
;
7759 /* add spare device to update
7761 for (id
= super
->devlist
; id
; id
= id
->next
)
7762 if (id
->index
== (unsigned)u
->subdev
) {
7763 struct imsm_dev
*dev
;
7764 struct imsm_map
*map
;
7766 dev
= get_imsm_dev(super
, u
->subdev
);
7767 map
= get_imsm_map(dev
, 0);
7768 current_level
= map
->raid_level
;
7771 if ((u
->new_level
== 5) && (u
->new_level
!= current_level
)) {
7772 struct mdinfo
*spares
;
7774 spares
= get_spares_for_grow(st
);
7782 makedev(dev
->disk
.major
,
7784 dl
= get_disk_super(super
,
7787 dl
->index
= u
->old_raid_disks
;
7793 len
= disks_to_mpb_size(u
->new_raid_disks
);
7794 dprintf("New anchor length is %llu\n", (unsigned long long)len
);
7797 case update_create_array
: {
7798 struct imsm_update_create_array
*u
= (void *) update
->buf
;
7799 struct intel_dev
*dv
;
7800 struct imsm_dev
*dev
= &u
->dev
;
7801 struct imsm_map
*map
= get_imsm_map(dev
, 0);
7803 struct disk_info
*inf
;
7807 inf
= get_disk_info(u
);
7808 len
= sizeof_imsm_dev(dev
, 1);
7809 /* allocate a new super->devlist entry */
7810 dv
= malloc(sizeof(*dv
));
7812 dv
->dev
= malloc(len
);
7817 update
->space
= NULL
;
7821 /* count how many spares will be converted to members */
7822 for (i
= 0; i
< map
->num_members
; i
++) {
7823 dl
= serial_to_dl(inf
[i
].serial
, super
);
7825 /* hmm maybe it failed?, nothing we can do about
7830 if (count_memberships(dl
, super
) == 0)
7833 len
+= activate
* sizeof(struct imsm_disk
);
7840 /* check if we need a larger metadata buffer */
7841 if (super
->next_buf
)
7842 buf_len
= super
->next_len
;
7844 buf_len
= super
->len
;
7846 if (__le32_to_cpu(mpb
->mpb_size
) + len
> buf_len
) {
7847 /* ok we need a larger buf than what is currently allocated
7848 * if this allocation fails process_update will notice that
7849 * ->next_len is set and ->next_buf is NULL
7851 buf_len
= ROUND_UP(__le32_to_cpu(mpb
->mpb_size
) + len
, 512);
7852 if (super
->next_buf
)
7853 free(super
->next_buf
);
7855 super
->next_len
= buf_len
;
7856 if (posix_memalign(&super
->next_buf
, 512, buf_len
) == 0)
7857 memset(super
->next_buf
, 0, buf_len
);
7859 super
->next_buf
= NULL
;
7863 /* must be called while manager is quiesced */
7864 static void imsm_delete(struct intel_super
*super
, struct dl
**dlp
, unsigned index
)
7866 struct imsm_super
*mpb
= super
->anchor
;
7868 struct imsm_dev
*dev
;
7869 struct imsm_map
*map
;
7870 int i
, j
, num_members
;
7873 dprintf("%s: deleting device[%d] from imsm_super\n",
7876 /* shift all indexes down one */
7877 for (iter
= super
->disks
; iter
; iter
= iter
->next
)
7878 if (iter
->index
> (int)index
)
7880 for (iter
= super
->missing
; iter
; iter
= iter
->next
)
7881 if (iter
->index
> (int)index
)
7884 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
7885 dev
= get_imsm_dev(super
, i
);
7886 map
= get_imsm_map(dev
, 0);
7887 num_members
= map
->num_members
;
7888 for (j
= 0; j
< num_members
; j
++) {
7889 /* update ord entries being careful not to propagate
7890 * ord-flags to the first map
7892 ord
= get_imsm_ord_tbl_ent(dev
, j
, -1);
7894 if (ord_to_idx(ord
) <= index
)
7897 map
= get_imsm_map(dev
, 0);
7898 set_imsm_ord_tbl_ent(map
, j
, ord_to_idx(ord
- 1));
7899 map
= get_imsm_map(dev
, 1);
7901 set_imsm_ord_tbl_ent(map
, j
, ord
- 1);
7906 super
->updates_pending
++;
7908 struct dl
*dl
= *dlp
;
7910 *dlp
= (*dlp
)->next
;
7911 __free_imsm_disk(dl
);
7914 #endif /* MDASSEMBLE */
7915 /*******************************************************************************
7916 * Function: open_backup_targets
7917 * Description: Function opens file descriptors for all devices given in
7920 * info : general array info
7921 * raid_disks : number of disks
7922 * raid_fds : table of device's file descriptors
7926 ******************************************************************************/
7927 int open_backup_targets(struct mdinfo
*info
, int raid_disks
, int *raid_fds
)
7931 for (sd
= info
->devs
; sd
; sd
= sd
->next
) {
7934 if (sd
->disk
.state
& (1<<MD_DISK_FAULTY
)) {
7935 dprintf("disk is faulty!!\n");
7939 if ((sd
->disk
.raid_disk
>= raid_disks
) ||
7940 (sd
->disk
.raid_disk
< 0))
7943 dn
= map_dev(sd
->disk
.major
,
7945 raid_fds
[sd
->disk
.raid_disk
] = dev_open(dn
, O_RDWR
);
7946 if (raid_fds
[sd
->disk
.raid_disk
] < 0) {
7947 fprintf(stderr
, "cannot open component\n");
7955 /*******************************************************************************
7956 * Function: init_migr_record_imsm
7957 * Description: Function inits imsm migration record
7959 * super : imsm internal array info
7960 * dev : device under migration
7961 * info : general array info to find the smallest device
7964 ******************************************************************************/
7965 void init_migr_record_imsm(struct supertype
*st
, struct imsm_dev
*dev
,
7966 struct mdinfo
*info
)
7968 struct intel_super
*super
= st
->sb
;
7969 struct migr_record
*migr_rec
= super
->migr_rec
;
7971 unsigned long long dsize
, dev_sectors
;
7972 long long unsigned min_dev_sectors
= -1LLU;
7976 struct imsm_map
*map_dest
= get_imsm_map(dev
, 0);
7977 struct imsm_map
*map_src
= get_imsm_map(dev
, 1);
7978 unsigned long long num_migr_units
;
7979 unsigned long long array_blocks
;
7981 memset(migr_rec
, 0, sizeof(struct migr_record
));
7982 migr_rec
->family_num
= __cpu_to_le32(super
->anchor
->family_num
);
7984 /* only ascending reshape supported now */
7985 migr_rec
->ascending_migr
= __cpu_to_le32(1);
7987 migr_rec
->dest_depth_per_unit
= GEN_MIGR_AREA_SIZE
/
7988 max(map_dest
->blocks_per_strip
, map_src
->blocks_per_strip
);
7989 migr_rec
->dest_depth_per_unit
*= map_dest
->blocks_per_strip
;
7990 new_data_disks
= imsm_num_data_members(dev
, 0);
7991 migr_rec
->blocks_per_unit
=
7992 __cpu_to_le32(migr_rec
->dest_depth_per_unit
* new_data_disks
);
7993 migr_rec
->dest_depth_per_unit
=
7994 __cpu_to_le32(migr_rec
->dest_depth_per_unit
);
7995 array_blocks
= info
->component_size
* new_data_disks
;
7997 array_blocks
/ __le32_to_cpu(migr_rec
->blocks_per_unit
);
7999 if (array_blocks
% __le32_to_cpu(migr_rec
->blocks_per_unit
))
8001 migr_rec
->num_migr_units
= __cpu_to_le32(num_migr_units
);
8003 migr_rec
->post_migr_vol_cap
= dev
->size_low
;
8004 migr_rec
->post_migr_vol_cap_hi
= dev
->size_high
;
8007 /* Find the smallest dev */
8008 for (sd
= info
->devs
; sd
; sd
= sd
->next
) {
8009 sprintf(nm
, "%d:%d", sd
->disk
.major
, sd
->disk
.minor
);
8010 fd
= dev_open(nm
, O_RDONLY
);
8013 get_dev_size(fd
, NULL
, &dsize
);
8014 dev_sectors
= dsize
/ 512;
8015 if (dev_sectors
< min_dev_sectors
)
8016 min_dev_sectors
= dev_sectors
;
8019 migr_rec
->ckpt_area_pba
= __cpu_to_le32(min_dev_sectors
-
8020 RAID_DISK_RESERVED_BLOCKS_IMSM_HI
);
8022 write_imsm_migr_rec(st
);
8027 /*******************************************************************************
8028 * Function: save_backup_imsm
8029 * Description: Function saves critical data stripes to Migration Copy Area
8030 * and updates the current migration unit status.
8031 * Use restore_stripes() to form a destination stripe,
8032 * and to write it to the Copy Area.
8034 * st : supertype information
8035 * dev : imsm device that backup is saved for
8036 * info : general array info
8037 * buf : input buffer
8038 * length : length of data to backup (blocks_per_unit)
8042 ******************************************************************************/
8043 int save_backup_imsm(struct supertype
*st
,
8044 struct imsm_dev
*dev
,
8045 struct mdinfo
*info
,
8050 struct intel_super
*super
= st
->sb
;
8051 unsigned long long *target_offsets
= NULL
;
8052 int *targets
= NULL
;
8054 struct imsm_map
*map_dest
= get_imsm_map(dev
, 0);
8055 int new_disks
= map_dest
->num_members
;
8056 int dest_layout
= 0;
8058 unsigned long long start
;
8059 int data_disks
= imsm_num_data_members(dev
, 0);
8061 targets
= malloc(new_disks
* sizeof(int));
8065 for (i
= 0; i
< new_disks
; i
++)
8068 target_offsets
= malloc(new_disks
* sizeof(unsigned long long));
8069 if (!target_offsets
)
8072 start
= info
->reshape_progress
* 512;
8073 for (i
= 0; i
< new_disks
; i
++) {
8074 target_offsets
[i
] = (unsigned long long)
8075 __le32_to_cpu(super
->migr_rec
->ckpt_area_pba
) * 512;
8076 /* move back copy area adderss, it will be moved forward
8077 * in restore_stripes() using start input variable
8079 target_offsets
[i
] -= start
/data_disks
;
8082 if (open_backup_targets(info
, new_disks
, targets
))
8085 dest_layout
= imsm_level_to_layout(map_dest
->raid_level
);
8086 dest_chunk
= __le16_to_cpu(map_dest
->blocks_per_strip
) * 512;
8088 if (restore_stripes(targets
, /* list of dest devices */
8089 target_offsets
, /* migration record offsets */
8092 map_dest
->raid_level
,
8094 -1, /* source backup file descriptor */
8095 0, /* input buf offset
8096 * always 0 buf is already offseted */
8100 fprintf(stderr
, Name
": Error restoring stripes\n");
8108 for (i
= 0; i
< new_disks
; i
++)
8109 if (targets
[i
] >= 0)
8113 free(target_offsets
);
8118 /*******************************************************************************
8119 * Function: save_checkpoint_imsm
8120 * Description: Function called for current unit status update
8121 * in the migration record. It writes it to disk.
8123 * super : imsm internal array info
8124 * info : general array info
8128 * 2: failure, means no valid migration record
8129 * / no general migration in progress /
8130 ******************************************************************************/
8131 int save_checkpoint_imsm(struct supertype
*st
, struct mdinfo
*info
, int state
)
8133 struct intel_super
*super
= st
->sb
;
8134 unsigned long long blocks_per_unit
;
8135 unsigned long long curr_migr_unit
;
8137 if (load_imsm_migr_rec(super
, info
) != 0) {
8138 dprintf("imsm: ERROR: Cannot read migration record "
8139 "for checkpoint save.\n");
8143 blocks_per_unit
= __le32_to_cpu(super
->migr_rec
->blocks_per_unit
);
8144 if (blocks_per_unit
== 0) {
8145 dprintf("imsm: no migration in progress.\n");
8148 curr_migr_unit
= info
->reshape_progress
/ blocks_per_unit
;
8149 /* check if array is alligned to copy area
8150 * if it is not alligned, add one to current migration unit value
8151 * this can happend on array reshape finish only
8153 if (info
->reshape_progress
% blocks_per_unit
)
8156 super
->migr_rec
->curr_migr_unit
=
8157 __cpu_to_le32(curr_migr_unit
);
8158 super
->migr_rec
->rec_status
= __cpu_to_le32(state
);
8159 super
->migr_rec
->dest_1st_member_lba
=
8160 __cpu_to_le32(curr_migr_unit
*
8161 __le32_to_cpu(super
->migr_rec
->dest_depth_per_unit
));
8162 if (write_imsm_migr_rec(st
) < 0) {
8163 dprintf("imsm: Cannot write migration record "
8164 "outside backup area\n");
8171 /*******************************************************************************
8172 * Function: recover_backup_imsm
8173 * Description: Function recovers critical data from the Migration Copy Area
8174 * while assembling an array.
8176 * super : imsm internal array info
8177 * info : general array info
8179 * 0 : success (or there is no data to recover)
8181 ******************************************************************************/
8182 int recover_backup_imsm(struct supertype
*st
, struct mdinfo
*info
)
8184 struct intel_super
*super
= st
->sb
;
8185 struct migr_record
*migr_rec
= super
->migr_rec
;
8186 struct imsm_map
*map_dest
= NULL
;
8187 struct intel_dev
*id
= NULL
;
8188 unsigned long long read_offset
;
8189 unsigned long long write_offset
;
8191 int *targets
= NULL
;
8192 int new_disks
, i
, err
;
8195 unsigned long curr_migr_unit
= __le32_to_cpu(migr_rec
->curr_migr_unit
);
8196 unsigned long num_migr_units
= __le32_to_cpu(migr_rec
->num_migr_units
);
8198 int skipped_disks
= 0;
8199 int max_degradation
;
8201 err
= sysfs_get_str(info
, NULL
, "array_state", (char *)buffer
, 20);
8205 /* recover data only during assemblation */
8206 if (strncmp(buffer
, "inactive", 8) != 0)
8208 /* no data to recover */
8209 if (__le32_to_cpu(migr_rec
->rec_status
) == UNIT_SRC_NORMAL
)
8211 if (curr_migr_unit
>= num_migr_units
)
8214 /* find device during reshape */
8215 for (id
= super
->devlist
; id
; id
= id
->next
)
8216 if (is_gen_migration(id
->dev
))
8221 map_dest
= get_imsm_map(id
->dev
, 0);
8222 new_disks
= map_dest
->num_members
;
8223 max_degradation
= new_disks
- imsm_num_data_members(id
->dev
, 0);
8225 read_offset
= (unsigned long long)
8226 __le32_to_cpu(migr_rec
->ckpt_area_pba
) * 512;
8228 write_offset
= ((unsigned long long)
8229 __le32_to_cpu(migr_rec
->dest_1st_member_lba
) +
8230 __le32_to_cpu(map_dest
->pba_of_lba0
)) * 512;
8232 unit_len
= __le32_to_cpu(migr_rec
->dest_depth_per_unit
) * 512;
8233 if (posix_memalign((void **)&buf
, 512, unit_len
) != 0)
8235 targets
= malloc(new_disks
* sizeof(int));
8239 open_backup_targets(info
, new_disks
, targets
);
8241 for (i
= 0; i
< new_disks
; i
++) {
8242 if (targets
[i
] < 0) {
8246 if (lseek64(targets
[i
], read_offset
, SEEK_SET
) < 0) {
8248 Name
": Cannot seek to block: %s\n",
8252 if ((unsigned)read(targets
[i
], buf
, unit_len
) != unit_len
) {
8254 Name
": Cannot read copy area block: %s\n",
8258 if (lseek64(targets
[i
], write_offset
, SEEK_SET
) < 0) {
8260 Name
": Cannot seek to block: %s\n",
8264 if ((unsigned)write(targets
[i
], buf
, unit_len
) != unit_len
) {
8266 Name
": Cannot restore block: %s\n",
8272 if (skipped_disks
> max_degradation
) {
8274 Name
": Cannot restore data from backup."
8275 " Too many failed disks\n");
8279 if (save_checkpoint_imsm(st
, info
, UNIT_SRC_NORMAL
)) {
8280 /* ignore error == 2, this can mean end of reshape here
8282 dprintf("imsm: Cannot write checkpoint to "
8283 "migration record (UNIT_SRC_NORMAL) during restart\n");
8289 for (i
= 0; i
< new_disks
; i
++)
8298 static char disk_by_path
[] = "/dev/disk/by-path/";
8300 static const char *imsm_get_disk_controller_domain(const char *path
)
8302 char disk_path
[PATH_MAX
];
8306 strncpy(disk_path
, disk_by_path
, PATH_MAX
- 1);
8307 strncat(disk_path
, path
, PATH_MAX
- strlen(disk_path
) - 1);
8308 if (stat(disk_path
, &st
) == 0) {
8309 struct sys_dev
* hba
;
8312 path
= devt_to_devpath(st
.st_rdev
);
8315 hba
= find_disk_attached_hba(-1, path
);
8316 if (hba
&& hba
->type
== SYS_DEV_SAS
)
8318 else if (hba
&& hba
->type
== SYS_DEV_SATA
)
8322 dprintf("path: %s hba: %s attached: %s\n",
8323 path
, (hba
) ? hba
->path
: "NULL", drv
);
8331 static int imsm_find_array_minor_by_subdev(int subdev
, int container
, int *minor
)
8333 char subdev_name
[20];
8334 struct mdstat_ent
*mdstat
;
8336 sprintf(subdev_name
, "%d", subdev
);
8337 mdstat
= mdstat_by_subdev(subdev_name
, container
);
8341 *minor
= mdstat
->devnum
;
8342 free_mdstat(mdstat
);
8346 static int imsm_reshape_is_allowed_on_container(struct supertype
*st
,
8347 struct geo_params
*geo
,
8348 int *old_raid_disks
)
8350 /* currently we only support increasing the number of devices
8351 * for a container. This increases the number of device for each
8352 * member array. They must all be RAID0 or RAID5.
8355 struct mdinfo
*info
, *member
;
8356 int devices_that_can_grow
= 0;
8358 dprintf("imsm: imsm_reshape_is_allowed_on_container(ENTER): "
8359 "st->devnum = (%i)\n",
8362 if (geo
->size
!= -1 ||
8363 geo
->level
!= UnSet
||
8364 geo
->layout
!= UnSet
||
8365 geo
->chunksize
!= 0 ||
8366 geo
->raid_disks
== UnSet
) {
8367 dprintf("imsm: Container operation is allowed for "
8368 "raid disks number change only.\n");
8372 info
= container_content_imsm(st
, NULL
);
8373 for (member
= info
; member
; member
= member
->next
) {
8377 dprintf("imsm: checking device_num: %i\n",
8378 member
->container_member
);
8380 if (geo
->raid_disks
<= member
->array
.raid_disks
) {
8381 /* we work on container for Online Capacity Expansion
8382 * only so raid_disks has to grow
8384 dprintf("imsm: for container operation raid disks "
8385 "increase is required\n");
8389 if ((info
->array
.level
!= 0) &&
8390 (info
->array
.level
!= 5)) {
8391 /* we cannot use this container with other raid level
8393 dprintf("imsm: for container operation wrong"
8394 " raid level (%i) detected\n",
8398 /* check for platform support
8399 * for this raid level configuration
8401 struct intel_super
*super
= st
->sb
;
8402 if (!is_raid_level_supported(super
->orom
,
8403 member
->array
.level
,
8405 dprintf("platform does not support raid%d with"
8409 geo
->raid_disks
> 1 ? "s" : "");
8412 /* check if component size is aligned to chunk size
8414 if (info
->component_size
%
8415 (info
->array
.chunk_size
/512)) {
8416 dprintf("Component size is not aligned to "
8422 if (*old_raid_disks
&&
8423 info
->array
.raid_disks
!= *old_raid_disks
)
8425 *old_raid_disks
= info
->array
.raid_disks
;
8427 /* All raid5 and raid0 volumes in container
8428 * have to be ready for Online Capacity Expansion
8429 * so they need to be assembled. We have already
8430 * checked that no recovery etc is happening.
8432 result
= imsm_find_array_minor_by_subdev(member
->container_member
,
8436 dprintf("imsm: cannot find array\n");
8439 devices_that_can_grow
++;
8442 if (!member
&& devices_that_can_grow
)
8446 dprintf("\tContainer operation allowed\n");
8448 dprintf("\tError: %i\n", ret_val
);
8453 /* Function: get_spares_for_grow
8454 * Description: Allocates memory and creates list of spare devices
8455 * avaliable in container. Checks if spare drive size is acceptable.
8456 * Parameters: Pointer to the supertype structure
8457 * Returns: Pointer to the list of spare devices (mdinfo structure) on success,
8460 static struct mdinfo
*get_spares_for_grow(struct supertype
*st
)
8462 unsigned long long min_size
= min_acceptable_spare_size_imsm(st
);
8463 return container_choose_spares(st
, min_size
, NULL
, NULL
, NULL
, 0);
8466 /******************************************************************************
8467 * function: imsm_create_metadata_update_for_reshape
8468 * Function creates update for whole IMSM container.
8470 ******************************************************************************/
8471 static int imsm_create_metadata_update_for_reshape(
8472 struct supertype
*st
,
8473 struct geo_params
*geo
,
8475 struct imsm_update_reshape
**updatep
)
8477 struct intel_super
*super
= st
->sb
;
8478 struct imsm_super
*mpb
= super
->anchor
;
8479 int update_memory_size
= 0;
8480 struct imsm_update_reshape
*u
= NULL
;
8481 struct mdinfo
*spares
= NULL
;
8483 int delta_disks
= 0;
8486 dprintf("imsm_update_metadata_for_reshape(enter) raid_disks = %i\n",
8489 delta_disks
= geo
->raid_disks
- old_raid_disks
;
8491 /* size of all update data without anchor */
8492 update_memory_size
= sizeof(struct imsm_update_reshape
);
8494 /* now add space for spare disks that we need to add. */
8495 update_memory_size
+= sizeof(u
->new_disks
[0]) * (delta_disks
- 1);
8497 u
= calloc(1, update_memory_size
);
8500 "cannot get memory for imsm_update_reshape update\n");
8503 u
->type
= update_reshape_container_disks
;
8504 u
->old_raid_disks
= old_raid_disks
;
8505 u
->new_raid_disks
= geo
->raid_disks
;
8507 /* now get spare disks list
8509 spares
= get_spares_for_grow(st
);
8512 || delta_disks
> spares
->array
.spare_disks
) {
8513 fprintf(stderr
, Name
": imsm: ERROR: Cannot get spare devices "
8514 "for %s.\n", geo
->dev_name
);
8519 /* we have got spares
8520 * update disk list in imsm_disk list table in anchor
8522 dprintf("imsm: %i spares are available.\n\n",
8523 spares
->array
.spare_disks
);
8526 for (i
= 0; i
< delta_disks
; i
++) {
8531 u
->new_disks
[i
] = makedev(dev
->disk
.major
,
8533 dl
= get_disk_super(super
, dev
->disk
.major
, dev
->disk
.minor
);
8534 dl
->index
= mpb
->num_disks
;
8544 dprintf("imsm: reshape update preparation :");
8545 if (i
== delta_disks
) {
8548 return update_memory_size
;
8551 dprintf(" Error\n");
8556 /******************************************************************************
8557 * function: imsm_create_metadata_update_for_migration()
8558 * Creates update for IMSM array.
8560 ******************************************************************************/
8561 static int imsm_create_metadata_update_for_migration(
8562 struct supertype
*st
,
8563 struct geo_params
*geo
,
8564 struct imsm_update_reshape_migration
**updatep
)
8566 struct intel_super
*super
= st
->sb
;
8567 int update_memory_size
= 0;
8568 struct imsm_update_reshape_migration
*u
= NULL
;
8569 struct imsm_dev
*dev
;
8570 int previous_level
= -1;
8572 dprintf("imsm_create_metadata_update_for_migration(enter)"
8573 " New Level = %i\n", geo
->level
);
8575 /* size of all update data without anchor */
8576 update_memory_size
= sizeof(struct imsm_update_reshape_migration
);
8578 u
= calloc(1, update_memory_size
);
8580 dprintf("error: cannot get memory for "
8581 "imsm_create_metadata_update_for_migration\n");
8584 u
->type
= update_reshape_migration
;
8585 u
->subdev
= super
->current_vol
;
8586 u
->new_level
= geo
->level
;
8587 u
->new_layout
= geo
->layout
;
8588 u
->new_raid_disks
= u
->old_raid_disks
= geo
->raid_disks
;
8589 u
->new_disks
[0] = -1;
8590 u
->new_chunksize
= -1;
8592 dev
= get_imsm_dev(super
, u
->subdev
);
8594 struct imsm_map
*map
;
8596 map
= get_imsm_map(dev
, 0);
8598 int current_chunk_size
=
8599 __le16_to_cpu(map
->blocks_per_strip
) / 2;
8601 if (geo
->chunksize
!= current_chunk_size
) {
8602 u
->new_chunksize
= geo
->chunksize
/ 1024;
8604 "chunk size change from %i to %i\n",
8605 current_chunk_size
, u
->new_chunksize
);
8607 previous_level
= map
->raid_level
;
8610 if ((geo
->level
== 5) && (previous_level
== 0)) {
8611 struct mdinfo
*spares
= NULL
;
8613 u
->new_raid_disks
++;
8614 spares
= get_spares_for_grow(st
);
8615 if ((spares
== NULL
) || (spares
->array
.spare_disks
< 1)) {
8618 update_memory_size
= 0;
8619 dprintf("error: cannot get spare device "
8620 "for requested migration");
8625 dprintf("imsm: reshape update preparation : OK\n");
8628 return update_memory_size
;
8631 static void imsm_update_metadata_locally(struct supertype
*st
,
8634 struct metadata_update mu
;
8639 mu
.space_list
= NULL
;
8641 imsm_prepare_update(st
, &mu
);
8642 imsm_process_update(st
, &mu
);
8644 while (mu
.space_list
) {
8645 void **space
= mu
.space_list
;
8646 mu
.space_list
= *space
;
8651 /***************************************************************************
8652 * Function: imsm_analyze_change
8653 * Description: Function analyze change for single volume
8654 * and validate if transition is supported
8655 * Parameters: Geometry parameters, supertype structure
8656 * Returns: Operation type code on success, -1 if fail
8657 ****************************************************************************/
8658 enum imsm_reshape_type
imsm_analyze_change(struct supertype
*st
,
8659 struct geo_params
*geo
)
8666 getinfo_super_imsm_volume(st
, &info
, NULL
);
8667 if ((geo
->level
!= info
.array
.level
) &&
8668 (geo
->level
>= 0) &&
8669 (geo
->level
!= UnSet
)) {
8670 switch (info
.array
.level
) {
8672 if (geo
->level
== 5) {
8673 change
= CH_MIGRATION
;
8674 if (geo
->layout
!= ALGORITHM_LEFT_ASYMMETRIC
) {
8676 Name
" Error. Requested Layout "
8677 "not supported (left-asymmetric layout "
8678 "is supported only)!\n");
8680 goto analyse_change_exit
;
8684 if (geo
->level
== 10) {
8685 change
= CH_TAKEOVER
;
8690 if (geo
->level
== 0) {
8691 change
= CH_TAKEOVER
;
8696 if (geo
->level
== 0) {
8697 change
= CH_TAKEOVER
;
8704 Name
" Error. Level Migration from %d to %d "
8706 info
.array
.level
, geo
->level
);
8707 goto analyse_change_exit
;
8710 geo
->level
= info
.array
.level
;
8712 if ((geo
->layout
!= info
.array
.layout
)
8713 && ((geo
->layout
!= UnSet
) && (geo
->layout
!= -1))) {
8714 change
= CH_MIGRATION
;
8715 if ((info
.array
.layout
== 0)
8716 && (info
.array
.level
== 5)
8717 && (geo
->layout
== 5)) {
8718 /* reshape 5 -> 4 */
8719 } else if ((info
.array
.layout
== 5)
8720 && (info
.array
.level
== 5)
8721 && (geo
->layout
== 0)) {
8722 /* reshape 4 -> 5 */
8727 Name
" Error. Layout Migration from %d to %d "
8729 info
.array
.layout
, geo
->layout
);
8731 goto analyse_change_exit
;
8734 geo
->layout
= info
.array
.layout
;
8736 if ((geo
->chunksize
> 0) && (geo
->chunksize
!= UnSet
)
8737 && (geo
->chunksize
!= info
.array
.chunk_size
))
8738 change
= CH_MIGRATION
;
8740 geo
->chunksize
= info
.array
.chunk_size
;
8742 chunk
= geo
->chunksize
/ 1024;
8743 if (!validate_geometry_imsm(st
,
8753 struct intel_super
*super
= st
->sb
;
8754 struct imsm_super
*mpb
= super
->anchor
;
8756 if (mpb
->num_raid_devs
> 1) {
8758 Name
" Error. Cannot perform operation on %s"
8759 "- for this operation it MUST be single "
8760 "array in container\n",
8766 analyse_change_exit
:
8771 int imsm_takeover(struct supertype
*st
, struct geo_params
*geo
)
8773 struct intel_super
*super
= st
->sb
;
8774 struct imsm_update_takeover
*u
;
8776 u
= malloc(sizeof(struct imsm_update_takeover
));
8780 u
->type
= update_takeover
;
8781 u
->subarray
= super
->current_vol
;
8783 /* 10->0 transition */
8784 if (geo
->level
== 0)
8785 u
->direction
= R10_TO_R0
;
8787 /* 0->10 transition */
8788 if (geo
->level
== 10)
8789 u
->direction
= R0_TO_R10
;
8791 /* update metadata locally */
8792 imsm_update_metadata_locally(st
, u
,
8793 sizeof(struct imsm_update_takeover
));
8794 /* and possibly remotely */
8795 if (st
->update_tail
)
8796 append_metadata_update(st
, u
,
8797 sizeof(struct imsm_update_takeover
));
8804 static int imsm_reshape_super(struct supertype
*st
, long long size
, int level
,
8805 int layout
, int chunksize
, int raid_disks
,
8806 int delta_disks
, char *backup
, char *dev
,
8810 struct geo_params geo
;
8812 dprintf("imsm: reshape_super called.\n");
8814 memset(&geo
, 0, sizeof(struct geo_params
));
8817 geo
.dev_id
= st
->devnum
;
8820 geo
.layout
= layout
;
8821 geo
.chunksize
= chunksize
;
8822 geo
.raid_disks
= raid_disks
;
8823 if (delta_disks
!= UnSet
)
8824 geo
.raid_disks
+= delta_disks
;
8826 dprintf("\tfor level : %i\n", geo
.level
);
8827 dprintf("\tfor raid_disks : %i\n", geo
.raid_disks
);
8829 if (experimental() == 0)
8832 if (st
->container_dev
== st
->devnum
) {
8833 /* On container level we can only increase number of devices. */
8834 dprintf("imsm: info: Container operation\n");
8835 int old_raid_disks
= 0;
8837 if (imsm_reshape_is_allowed_on_container(
8838 st
, &geo
, &old_raid_disks
)) {
8839 struct imsm_update_reshape
*u
= NULL
;
8842 len
= imsm_create_metadata_update_for_reshape(
8843 st
, &geo
, old_raid_disks
, &u
);
8846 dprintf("imsm: Cannot prepare update\n");
8847 goto exit_imsm_reshape_super
;
8851 /* update metadata locally */
8852 imsm_update_metadata_locally(st
, u
, len
);
8853 /* and possibly remotely */
8854 if (st
->update_tail
)
8855 append_metadata_update(st
, u
, len
);
8860 fprintf(stderr
, Name
": (imsm) Operation "
8861 "is not allowed on this container\n");
8864 /* On volume level we support following operations
8865 * - takeover: raid10 -> raid0; raid0 -> raid10
8866 * - chunk size migration
8867 * - migration: raid5 -> raid0; raid0 -> raid5
8869 struct intel_super
*super
= st
->sb
;
8870 struct intel_dev
*dev
= super
->devlist
;
8872 dprintf("imsm: info: Volume operation\n");
8873 /* find requested device */
8875 if (imsm_find_array_minor_by_subdev(
8876 dev
->index
, st
->container_dev
, &devnum
) == 0
8877 && devnum
== geo
.dev_id
)
8882 fprintf(stderr
, Name
" Cannot find %s (%i) subarray\n",
8883 geo
.dev_name
, geo
.dev_id
);
8884 goto exit_imsm_reshape_super
;
8886 super
->current_vol
= dev
->index
;
8887 change
= imsm_analyze_change(st
, &geo
);
8890 ret_val
= imsm_takeover(st
, &geo
);
8892 case CH_MIGRATION
: {
8893 struct imsm_update_reshape_migration
*u
= NULL
;
8895 imsm_create_metadata_update_for_migration(
8899 "Cannot prepare update\n");
8903 /* update metadata locally */
8904 imsm_update_metadata_locally(st
, u
, len
);
8905 /* and possibly remotely */
8906 if (st
->update_tail
)
8907 append_metadata_update(st
, u
, len
);
8917 exit_imsm_reshape_super
:
8918 dprintf("imsm: reshape_super Exit code = %i\n", ret_val
);
8922 /*******************************************************************************
8923 * Function: wait_for_reshape_imsm
8924 * Description: Function writes new sync_max value and waits until
8925 * reshape process reach new position
8927 * sra : general array info
8928 * ndata : number of disks in new array's layout
8931 * 1 : there is no reshape in progress,
8933 ******************************************************************************/
8934 int wait_for_reshape_imsm(struct mdinfo
*sra
, int ndata
)
8936 int fd
= sysfs_get_fd(sra
, NULL
, "reshape_position");
8937 unsigned long long completed
;
8938 /* to_complete : new sync_max position */
8939 unsigned long long to_complete
= sra
->reshape_progress
;
8940 unsigned long long position_to_set
= to_complete
/ ndata
;
8943 dprintf("imsm: wait_for_reshape_imsm() "
8944 "cannot open reshape_position\n");
8948 if (sysfs_fd_get_ll(fd
, &completed
) < 0) {
8949 dprintf("imsm: wait_for_reshape_imsm() "
8950 "cannot read reshape_position (no reshape in progres)\n");
8955 if (completed
> to_complete
) {
8956 dprintf("imsm: wait_for_reshape_imsm() "
8957 "wrong next position to set %llu (%llu)\n",
8958 to_complete
, completed
);
8962 dprintf("Position set: %llu\n", position_to_set
);
8963 if (sysfs_set_num(sra
, NULL
, "sync_max",
8964 position_to_set
) != 0) {
8965 dprintf("imsm: wait_for_reshape_imsm() "
8966 "cannot set reshape position to %llu\n",
8977 select(fd
+1, &rfds
, NULL
, NULL
, NULL
);
8978 if (sysfs_get_str(sra
, NULL
, "sync_action",
8980 strncmp(action
, "reshape", 7) != 0)
8982 if (sysfs_fd_get_ll(fd
, &completed
) < 0) {
8983 dprintf("imsm: wait_for_reshape_imsm() "
8984 "cannot read reshape_position (in loop)\n");
8988 } while (completed
< to_complete
);
8994 /*******************************************************************************
8995 * Function: check_degradation_change
8996 * Description: Check that array hasn't become failed.
8998 * info : for sysfs access
8999 * sources : source disks descriptors
9000 * degraded: previous degradation level
9003 ******************************************************************************/
9004 int check_degradation_change(struct mdinfo
*info
,
9008 unsigned long long new_degraded
;
9009 sysfs_get_ll(info
, NULL
, "degraded", &new_degraded
);
9010 if (new_degraded
!= (unsigned long long)degraded
) {
9011 /* check each device to ensure it is still working */
9014 for (sd
= info
->devs
; sd
; sd
= sd
->next
) {
9015 if (sd
->disk
.state
& (1<<MD_DISK_FAULTY
))
9017 if (sd
->disk
.state
& (1<<MD_DISK_SYNC
)) {
9019 if (sysfs_get_str(info
,
9020 sd
, "state", sbuf
, 20) < 0 ||
9021 strstr(sbuf
, "faulty") ||
9022 strstr(sbuf
, "in_sync") == NULL
) {
9023 /* this device is dead */
9024 sd
->disk
.state
= (1<<MD_DISK_FAULTY
);
9025 if (sd
->disk
.raid_disk
>= 0 &&
9026 sources
[sd
->disk
.raid_disk
] >= 0) {
9028 sd
->disk
.raid_disk
]);
9029 sources
[sd
->disk
.raid_disk
] =
9038 return new_degraded
;
9041 /*******************************************************************************
9042 * Function: imsm_manage_reshape
9043 * Description: Function finds array under reshape and it manages reshape
9044 * process. It creates stripes backups (if required) and sets
9047 * afd : Backup handle (nattive) - not used
9048 * sra : general array info
9049 * reshape : reshape parameters - not used
9050 * st : supertype structure
9051 * blocks : size of critical section [blocks]
9052 * fds : table of source device descriptor
9053 * offsets : start of array (offest per devices)
9055 * destfd : table of destination device descriptor
9056 * destoffsets : table of destination offsets (per device)
9058 * 1 : success, reshape is done
9060 ******************************************************************************/
9061 static int imsm_manage_reshape(
9062 int afd
, struct mdinfo
*sra
, struct reshape
*reshape
,
9063 struct supertype
*st
, unsigned long backup_blocks
,
9064 int *fds
, unsigned long long *offsets
,
9065 int dests
, int *destfd
, unsigned long long *destoffsets
)
9068 struct intel_super
*super
= st
->sb
;
9069 struct intel_dev
*dv
= NULL
;
9070 struct imsm_dev
*dev
= NULL
;
9071 struct imsm_map
*map_src
;
9072 int migr_vol_qan
= 0;
9073 int ndata
, odata
; /* [bytes] */
9074 int chunk
; /* [bytes] */
9075 struct migr_record
*migr_rec
;
9077 unsigned int buf_size
; /* [bytes] */
9078 unsigned long long max_position
; /* array size [bytes] */
9079 unsigned long long next_step
; /* [blocks]/[bytes] */
9080 unsigned long long old_data_stripe_length
;
9081 unsigned long long start_src
; /* [bytes] */
9082 unsigned long long start
; /* [bytes] */
9083 unsigned long long start_buf_shift
; /* [bytes] */
9085 int source_layout
= 0;
9087 if (!fds
|| !offsets
|| !sra
)
9090 /* Find volume during the reshape */
9091 for (dv
= super
->devlist
; dv
; dv
= dv
->next
) {
9092 if (dv
->dev
->vol
.migr_type
== MIGR_GEN_MIGR
9093 && dv
->dev
->vol
.migr_state
== 1) {
9098 /* Only one volume can migrate at the same time */
9099 if (migr_vol_qan
!= 1) {
9100 fprintf(stderr
, Name
" : %s", migr_vol_qan
?
9101 "Number of migrating volumes greater than 1\n" :
9102 "There is no volume during migrationg\n");
9106 map_src
= get_imsm_map(dev
, 1);
9107 if (map_src
== NULL
)
9110 ndata
= imsm_num_data_members(dev
, 0);
9111 odata
= imsm_num_data_members(dev
, 1);
9113 chunk
= __le16_to_cpu(map_src
->blocks_per_strip
) * 512;
9114 old_data_stripe_length
= odata
* chunk
;
9116 migr_rec
= super
->migr_rec
;
9118 /* initialize migration record for start condition */
9119 if (sra
->reshape_progress
== 0)
9120 init_migr_record_imsm(st
, dev
, sra
);
9122 if (__le32_to_cpu(migr_rec
->rec_status
) != UNIT_SRC_NORMAL
) {
9123 dprintf("imsm: cannot restart migration when data "
9124 "are present in copy area.\n");
9130 buf_size
= __le32_to_cpu(migr_rec
->blocks_per_unit
) * 512;
9131 /* extend buffer size for parity disk */
9132 buf_size
+= __le32_to_cpu(migr_rec
->dest_depth_per_unit
) * 512;
9133 /* add space for stripe aligment */
9134 buf_size
+= old_data_stripe_length
;
9135 if (posix_memalign((void **)&buf
, 4096, buf_size
)) {
9136 dprintf("imsm: Cannot allocate checpoint buffer\n");
9140 max_position
= sra
->component_size
* ndata
;
9141 source_layout
= imsm_level_to_layout(map_src
->raid_level
);
9143 while (__le32_to_cpu(migr_rec
->curr_migr_unit
) <
9144 __le32_to_cpu(migr_rec
->num_migr_units
)) {
9145 /* current reshape position [blocks] */
9146 unsigned long long current_position
=
9147 __le32_to_cpu(migr_rec
->blocks_per_unit
)
9148 * __le32_to_cpu(migr_rec
->curr_migr_unit
);
9149 unsigned long long border
;
9151 /* Check that array hasn't become failed.
9153 degraded
= check_degradation_change(sra
, fds
, degraded
);
9155 dprintf("imsm: Abort reshape due to degradation"
9156 " level (%i)\n", degraded
);
9160 next_step
= __le32_to_cpu(migr_rec
->blocks_per_unit
);
9162 if ((current_position
+ next_step
) > max_position
)
9163 next_step
= max_position
- current_position
;
9165 start
= current_position
* 512;
9167 /* allign reading start to old geometry */
9168 start_buf_shift
= start
% old_data_stripe_length
;
9169 start_src
= start
- start_buf_shift
;
9171 border
= (start_src
/ odata
) - (start
/ ndata
);
9173 if (border
<= __le32_to_cpu(migr_rec
->dest_depth_per_unit
)) {
9174 /* save critical stripes to buf
9175 * start - start address of current unit
9177 * start_src - start address of current unit
9178 * to backup alligned to source array
9181 unsigned long long next_step_filler
= 0;
9182 unsigned long long copy_length
= next_step
* 512;
9184 /* allign copy area length to stripe in old geometry */
9185 next_step_filler
= ((copy_length
+ start_buf_shift
)
9186 % old_data_stripe_length
);
9187 if (next_step_filler
)
9188 next_step_filler
= (old_data_stripe_length
9189 - next_step_filler
);
9190 dprintf("save_stripes() parameters: start = %llu,"
9191 "\tstart_src = %llu,\tnext_step*512 = %llu,"
9192 "\tstart_in_buf_shift = %llu,"
9193 "\tnext_step_filler = %llu\n",
9194 start
, start_src
, copy_length
,
9195 start_buf_shift
, next_step_filler
);
9197 if (save_stripes(fds
, offsets
, map_src
->num_members
,
9198 chunk
, map_src
->raid_level
,
9199 source_layout
, 0, NULL
, start_src
,
9201 next_step_filler
+ start_buf_shift
,
9203 dprintf("imsm: Cannot save stripes"
9207 /* Convert data to destination format and store it
9208 * in backup general migration area
9210 if (save_backup_imsm(st
, dev
, sra
,
9211 buf
+ start_buf_shift
, copy_length
)) {
9212 dprintf("imsm: Cannot save stripes to "
9213 "target devices\n");
9216 if (save_checkpoint_imsm(st
, sra
,
9217 UNIT_SRC_IN_CP_AREA
)) {
9218 dprintf("imsm: Cannot write checkpoint to "
9219 "migration record (UNIT_SRC_IN_CP_AREA)\n");
9223 /* set next step to use whole border area */
9224 border
/= next_step
;
9226 next_step
*= border
;
9228 /* When data backed up, checkpoint stored,
9229 * kick the kernel to reshape unit of data
9231 next_step
= next_step
+ sra
->reshape_progress
;
9232 /* limit next step to array max position */
9233 if (next_step
> max_position
)
9234 next_step
= max_position
;
9235 sysfs_set_num(sra
, NULL
, "suspend_lo", sra
->reshape_progress
);
9236 sysfs_set_num(sra
, NULL
, "suspend_hi", next_step
);
9237 sra
->reshape_progress
= next_step
;
9239 /* wait until reshape finish */
9240 if (wait_for_reshape_imsm(sra
, ndata
) < 0) {
9241 dprintf("wait_for_reshape_imsm returned error!\n");
9245 if (save_checkpoint_imsm(st
, sra
, UNIT_SRC_NORMAL
) == 1) {
9246 /* ignore error == 2, this can mean end of reshape here
9248 dprintf("imsm: Cannot write checkpoint to "
9249 "migration record (UNIT_SRC_NORMAL)\n");
9255 /* return '1' if done */
9263 #endif /* MDASSEMBLE */
9265 struct superswitch super_imsm
= {
9267 .examine_super
= examine_super_imsm
,
9268 .brief_examine_super
= brief_examine_super_imsm
,
9269 .brief_examine_subarrays
= brief_examine_subarrays_imsm
,
9270 .export_examine_super
= export_examine_super_imsm
,
9271 .detail_super
= detail_super_imsm
,
9272 .brief_detail_super
= brief_detail_super_imsm
,
9273 .write_init_super
= write_init_super_imsm
,
9274 .validate_geometry
= validate_geometry_imsm
,
9275 .add_to_super
= add_to_super_imsm
,
9276 .remove_from_super
= remove_from_super_imsm
,
9277 .detail_platform
= detail_platform_imsm
,
9278 .kill_subarray
= kill_subarray_imsm
,
9279 .update_subarray
= update_subarray_imsm
,
9280 .load_container
= load_container_imsm
,
9281 .default_geometry
= default_geometry_imsm
,
9282 .get_disk_controller_domain
= imsm_get_disk_controller_domain
,
9283 .reshape_super
= imsm_reshape_super
,
9284 .manage_reshape
= imsm_manage_reshape
,
9285 .recover_backup
= recover_backup_imsm
,
9287 .match_home
= match_home_imsm
,
9288 .uuid_from_super
= uuid_from_super_imsm
,
9289 .getinfo_super
= getinfo_super_imsm
,
9290 .getinfo_super_disks
= getinfo_super_disks_imsm
,
9291 .update_super
= update_super_imsm
,
9293 .avail_size
= avail_size_imsm
,
9294 .min_acceptable_spare_size
= min_acceptable_spare_size_imsm
,
9296 .compare_super
= compare_super_imsm
,
9298 .load_super
= load_super_imsm
,
9299 .init_super
= init_super_imsm
,
9300 .store_super
= store_super_imsm
,
9301 .free_super
= free_super_imsm
,
9302 .match_metadata_desc
= match_metadata_desc_imsm
,
9303 .container_content
= container_content_imsm
,
9311 .open_new
= imsm_open_new
,
9312 .set_array_state
= imsm_set_array_state
,
9313 .set_disk
= imsm_set_disk
,
9314 .sync_metadata
= imsm_sync_metadata
,
9315 .activate_spare
= imsm_activate_spare
,
9316 .process_update
= imsm_process_update
,
9317 .prepare_update
= imsm_prepare_update
,
9318 #endif /* MDASSEMBLE */