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;
996 /* find first active disk in array */
998 while (dl
&& (is_failed(&dl
->disk
) || dl
->index
== -1))
1002 /* find last lba used by subarrays */
1003 e
= get_extents(super
, dl
);
1006 for (i
= 0; e
[i
].size
; i
++)
1009 rv
= e
[i
-1].start
+ e
[i
-1].size
;
1010 reservation
= __le32_to_cpu(dl
->disk
.total_blocks
) - e
[i
].start
;
1013 /* add the amount of space needed for metadata */
1014 rv
= rv
+ imsm_min_reserved_sectors(super
);
1020 static __u64
blocks_per_migr_unit(struct intel_super
*super
,
1021 struct imsm_dev
*dev
);
1023 static void print_imsm_dev(struct intel_super
*super
,
1024 struct imsm_dev
*dev
,
1030 struct imsm_map
*map
= get_imsm_map(dev
, 0);
1031 struct imsm_map
*map2
= get_imsm_map(dev
, 1);
1035 printf("[%.16s]:\n", dev
->volume
);
1036 printf(" UUID : %s\n", uuid
);
1037 printf(" RAID Level : %d", get_imsm_raid_level(map
));
1039 printf(" <-- %d", get_imsm_raid_level(map2
));
1041 printf(" Members : %d", map
->num_members
);
1043 printf(" <-- %d", map2
->num_members
);
1045 printf(" Slots : [");
1046 for (i
= 0; i
< map
->num_members
; i
++) {
1047 ord
= get_imsm_ord_tbl_ent(dev
, i
, 0);
1048 printf("%s", ord
& IMSM_ORD_REBUILD
? "_" : "U");
1053 for (i
= 0; i
< map2
->num_members
; i
++) {
1054 ord
= get_imsm_ord_tbl_ent(dev
, i
, 1);
1055 printf("%s", ord
& IMSM_ORD_REBUILD
? "_" : "U");
1060 printf(" Failed disk : ");
1061 if (map
->failed_disk_num
== 0xff)
1064 printf("%i", map
->failed_disk_num
);
1066 slot
= get_imsm_disk_slot(map
, disk_idx
);
1068 ord
= get_imsm_ord_tbl_ent(dev
, slot
, -1);
1069 printf(" This Slot : %d%s\n", slot
,
1070 ord
& IMSM_ORD_REBUILD
? " (out-of-sync)" : "");
1072 printf(" This Slot : ?\n");
1073 sz
= __le32_to_cpu(dev
->size_high
);
1075 sz
+= __le32_to_cpu(dev
->size_low
);
1076 printf(" Array Size : %llu%s\n", (unsigned long long)sz
,
1077 human_size(sz
* 512));
1078 sz
= __le32_to_cpu(map
->blocks_per_member
);
1079 printf(" Per Dev Size : %llu%s\n", (unsigned long long)sz
,
1080 human_size(sz
* 512));
1081 printf(" Sector Offset : %u\n",
1082 __le32_to_cpu(map
->pba_of_lba0
));
1083 printf(" Num Stripes : %u\n",
1084 __le32_to_cpu(map
->num_data_stripes
));
1085 printf(" Chunk Size : %u KiB",
1086 __le16_to_cpu(map
->blocks_per_strip
) / 2);
1088 printf(" <-- %u KiB",
1089 __le16_to_cpu(map2
->blocks_per_strip
) / 2);
1091 printf(" Reserved : %d\n", __le32_to_cpu(dev
->reserved_blocks
));
1092 printf(" Migrate State : ");
1093 if (dev
->vol
.migr_state
) {
1094 if (migr_type(dev
) == MIGR_INIT
)
1095 printf("initialize\n");
1096 else if (migr_type(dev
) == MIGR_REBUILD
)
1097 printf("rebuild\n");
1098 else if (migr_type(dev
) == MIGR_VERIFY
)
1100 else if (migr_type(dev
) == MIGR_GEN_MIGR
)
1101 printf("general migration\n");
1102 else if (migr_type(dev
) == MIGR_STATE_CHANGE
)
1103 printf("state change\n");
1104 else if (migr_type(dev
) == MIGR_REPAIR
)
1107 printf("<unknown:%d>\n", migr_type(dev
));
1110 printf(" Map State : %s", map_state_str
[map
->map_state
]);
1111 if (dev
->vol
.migr_state
) {
1112 struct imsm_map
*map
= get_imsm_map(dev
, 1);
1114 printf(" <-- %s", map_state_str
[map
->map_state
]);
1115 printf("\n Checkpoint : %u (%llu)",
1116 __le32_to_cpu(dev
->vol
.curr_migr_unit
),
1117 (unsigned long long)blocks_per_migr_unit(super
, dev
));
1120 printf(" Dirty State : %s\n", dev
->vol
.dirty
? "dirty" : "clean");
1123 static void print_imsm_disk(struct imsm_disk
*disk
, int index
, __u32 reserved
)
1125 char str
[MAX_RAID_SERIAL_LEN
+ 1];
1128 if (index
< -1 || !disk
)
1132 snprintf(str
, MAX_RAID_SERIAL_LEN
+ 1, "%s", disk
->serial
);
1134 printf(" Disk%02d Serial : %s\n", index
, str
);
1136 printf(" Disk Serial : %s\n", str
);
1137 printf(" State :%s%s%s\n", is_spare(disk
) ? " spare" : "",
1138 is_configured(disk
) ? " active" : "",
1139 is_failed(disk
) ? " failed" : "");
1140 printf(" Id : %08x\n", __le32_to_cpu(disk
->scsi_id
));
1141 sz
= __le32_to_cpu(disk
->total_blocks
) - reserved
;
1142 printf(" Usable Size : %llu%s\n", (unsigned long long)sz
,
1143 human_size(sz
* 512));
1146 static int is_gen_migration(struct imsm_dev
*dev
);
1148 void examine_migr_rec_imsm(struct intel_super
*super
)
1150 struct migr_record
*migr_rec
= super
->migr_rec
;
1151 struct imsm_super
*mpb
= super
->anchor
;
1154 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
1155 struct imsm_dev
*dev
= __get_imsm_dev(mpb
, i
);
1156 if (is_gen_migration(dev
) == 0)
1159 printf("\nMigration Record Information:");
1160 if (super
->disks
->index
> 1) {
1161 printf(" Empty\n ");
1162 printf("Examine one of first two disks in array\n");
1165 printf("\n Status : ");
1166 if (__le32_to_cpu(migr_rec
->rec_status
) == UNIT_SRC_NORMAL
)
1169 printf("Contains Data\n");
1170 printf(" Current Unit : %u\n",
1171 __le32_to_cpu(migr_rec
->curr_migr_unit
));
1172 printf(" Family : %u\n",
1173 __le32_to_cpu(migr_rec
->family_num
));
1174 printf(" Ascending : %u\n",
1175 __le32_to_cpu(migr_rec
->ascending_migr
));
1176 printf(" Blocks Per Unit : %u\n",
1177 __le32_to_cpu(migr_rec
->blocks_per_unit
));
1178 printf(" Dest. Depth Per Unit : %u\n",
1179 __le32_to_cpu(migr_rec
->dest_depth_per_unit
));
1180 printf(" Checkpoint Area pba : %u\n",
1181 __le32_to_cpu(migr_rec
->ckpt_area_pba
));
1182 printf(" First member lba : %u\n",
1183 __le32_to_cpu(migr_rec
->dest_1st_member_lba
));
1184 printf(" Total Number of Units : %u\n",
1185 __le32_to_cpu(migr_rec
->num_migr_units
));
1186 printf(" Size of volume : %u\n",
1187 __le32_to_cpu(migr_rec
->post_migr_vol_cap
));
1188 printf(" Expansion space for LBA64 : %u\n",
1189 __le32_to_cpu(migr_rec
->post_migr_vol_cap_hi
));
1190 printf(" Record was read from : %u\n",
1191 __le32_to_cpu(migr_rec
->ckpt_read_disk_num
));
1196 #endif /* MDASSEMBLE */
1197 /*******************************************************************************
1198 * function: imsm_check_attributes
1199 * Description: Function checks if features represented by attributes flags
1200 * are supported by mdadm.
1202 * attributes - Attributes read from metadata
1204 * 0 - passed attributes contains unsupported features flags
1205 * 1 - all features are supported
1206 ******************************************************************************/
1207 static int imsm_check_attributes(__u32 attributes
)
1210 __u32 not_supported
= MPB_ATTRIB_SUPPORTED
^0xffffffff;
1212 not_supported
&= ~MPB_ATTRIB_IGNORED
;
1214 not_supported
&= attributes
;
1215 if (not_supported
) {
1216 fprintf(stderr
, Name
"(IMSM): Unsupported attributes : %x\n",
1217 (unsigned)__le32_to_cpu(not_supported
));
1218 if (not_supported
& MPB_ATTRIB_CHECKSUM_VERIFY
) {
1219 dprintf("\t\tMPB_ATTRIB_CHECKSUM_VERIFY \n");
1220 not_supported
^= MPB_ATTRIB_CHECKSUM_VERIFY
;
1222 if (not_supported
& MPB_ATTRIB_2TB
) {
1223 dprintf("\t\tMPB_ATTRIB_2TB\n");
1224 not_supported
^= MPB_ATTRIB_2TB
;
1226 if (not_supported
& MPB_ATTRIB_RAID0
) {
1227 dprintf("\t\tMPB_ATTRIB_RAID0\n");
1228 not_supported
^= MPB_ATTRIB_RAID0
;
1230 if (not_supported
& MPB_ATTRIB_RAID1
) {
1231 dprintf("\t\tMPB_ATTRIB_RAID1\n");
1232 not_supported
^= MPB_ATTRIB_RAID1
;
1234 if (not_supported
& MPB_ATTRIB_RAID10
) {
1235 dprintf("\t\tMPB_ATTRIB_RAID10\n");
1236 not_supported
^= MPB_ATTRIB_RAID10
;
1238 if (not_supported
& MPB_ATTRIB_RAID1E
) {
1239 dprintf("\t\tMPB_ATTRIB_RAID1E\n");
1240 not_supported
^= MPB_ATTRIB_RAID1E
;
1242 if (not_supported
& MPB_ATTRIB_RAID5
) {
1243 dprintf("\t\tMPB_ATTRIB_RAID5\n");
1244 not_supported
^= MPB_ATTRIB_RAID5
;
1246 if (not_supported
& MPB_ATTRIB_RAIDCNG
) {
1247 dprintf("\t\tMPB_ATTRIB_RAIDCNG\n");
1248 not_supported
^= MPB_ATTRIB_RAIDCNG
;
1250 if (not_supported
& MPB_ATTRIB_BBM
) {
1251 dprintf("\t\tMPB_ATTRIB_BBM\n");
1252 not_supported
^= MPB_ATTRIB_BBM
;
1254 if (not_supported
& MPB_ATTRIB_CHECKSUM_VERIFY
) {
1255 dprintf("\t\tMPB_ATTRIB_CHECKSUM_VERIFY (== MPB_ATTRIB_LEGACY)\n");
1256 not_supported
^= MPB_ATTRIB_CHECKSUM_VERIFY
;
1258 if (not_supported
& MPB_ATTRIB_EXP_STRIPE_SIZE
) {
1259 dprintf("\t\tMPB_ATTRIB_EXP_STRIP_SIZE\n");
1260 not_supported
^= MPB_ATTRIB_EXP_STRIPE_SIZE
;
1262 if (not_supported
& MPB_ATTRIB_2TB_DISK
) {
1263 dprintf("\t\tMPB_ATTRIB_2TB_DISK\n");
1264 not_supported
^= MPB_ATTRIB_2TB_DISK
;
1266 if (not_supported
& MPB_ATTRIB_NEVER_USE2
) {
1267 dprintf("\t\tMPB_ATTRIB_NEVER_USE2\n");
1268 not_supported
^= MPB_ATTRIB_NEVER_USE2
;
1270 if (not_supported
& MPB_ATTRIB_NEVER_USE
) {
1271 dprintf("\t\tMPB_ATTRIB_NEVER_USE\n");
1272 not_supported
^= MPB_ATTRIB_NEVER_USE
;
1276 dprintf(Name
"(IMSM): Unknown attributes : %x\n", not_supported
);
1285 static void getinfo_super_imsm(struct supertype
*st
, struct mdinfo
*info
, char *map
);
1287 static void examine_super_imsm(struct supertype
*st
, char *homehost
)
1289 struct intel_super
*super
= st
->sb
;
1290 struct imsm_super
*mpb
= super
->anchor
;
1291 char str
[MAX_SIGNATURE_LENGTH
];
1296 __u32 reserved
= imsm_reserved_sectors(super
, super
->disks
);
1299 snprintf(str
, MPB_SIG_LEN
, "%s", mpb
->sig
);
1300 printf(" Magic : %s\n", str
);
1301 snprintf(str
, strlen(MPB_VERSION_RAID0
), "%s", get_imsm_version(mpb
));
1302 printf(" Version : %s\n", get_imsm_version(mpb
));
1303 printf(" Orig Family : %08x\n", __le32_to_cpu(mpb
->orig_family_num
));
1304 printf(" Family : %08x\n", __le32_to_cpu(mpb
->family_num
));
1305 printf(" Generation : %08x\n", __le32_to_cpu(mpb
->generation_num
));
1306 printf(" Attributes : ");
1307 if (imsm_check_attributes(mpb
->attributes
))
1308 printf("All supported\n");
1310 printf("not supported\n");
1311 getinfo_super_imsm(st
, &info
, NULL
);
1312 fname_from_uuid(st
, &info
, nbuf
, ':');
1313 printf(" UUID : %s\n", nbuf
+ 5);
1314 sum
= __le32_to_cpu(mpb
->check_sum
);
1315 printf(" Checksum : %08x %s\n", sum
,
1316 __gen_imsm_checksum(mpb
) == sum
? "correct" : "incorrect");
1317 printf(" MPB Sectors : %d\n", mpb_sectors(mpb
));
1318 printf(" Disks : %d\n", mpb
->num_disks
);
1319 printf(" RAID Devices : %d\n", mpb
->num_raid_devs
);
1320 print_imsm_disk(__get_imsm_disk(mpb
, super
->disks
->index
), super
->disks
->index
, reserved
);
1321 if (super
->bbm_log
) {
1322 struct bbm_log
*log
= super
->bbm_log
;
1325 printf("Bad Block Management Log:\n");
1326 printf(" Log Size : %d\n", __le32_to_cpu(mpb
->bbm_log_size
));
1327 printf(" Signature : %x\n", __le32_to_cpu(log
->signature
));
1328 printf(" Entry Count : %d\n", __le32_to_cpu(log
->entry_count
));
1329 printf(" Spare Blocks : %d\n", __le32_to_cpu(log
->reserved_spare_block_count
));
1330 printf(" First Spare : %llx\n",
1331 (unsigned long long) __le64_to_cpu(log
->first_spare_lba
));
1333 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
1335 struct imsm_dev
*dev
= __get_imsm_dev(mpb
, i
);
1337 super
->current_vol
= i
;
1338 getinfo_super_imsm(st
, &info
, NULL
);
1339 fname_from_uuid(st
, &info
, nbuf
, ':');
1340 print_imsm_dev(super
, dev
, nbuf
+ 5, super
->disks
->index
);
1342 for (i
= 0; i
< mpb
->num_disks
; i
++) {
1343 if (i
== super
->disks
->index
)
1345 print_imsm_disk(__get_imsm_disk(mpb
, i
), i
, reserved
);
1348 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
1349 if (dl
->index
== -1)
1350 print_imsm_disk(&dl
->disk
, -1, reserved
);
1352 examine_migr_rec_imsm(super
);
1355 static void brief_examine_super_imsm(struct supertype
*st
, int verbose
)
1357 /* We just write a generic IMSM ARRAY entry */
1360 struct intel_super
*super
= st
->sb
;
1362 if (!super
->anchor
->num_raid_devs
) {
1363 printf("ARRAY metadata=imsm\n");
1367 getinfo_super_imsm(st
, &info
, NULL
);
1368 fname_from_uuid(st
, &info
, nbuf
, ':');
1369 printf("ARRAY metadata=imsm UUID=%s\n", nbuf
+ 5);
1372 static void brief_examine_subarrays_imsm(struct supertype
*st
, int verbose
)
1374 /* We just write a generic IMSM ARRAY entry */
1378 struct intel_super
*super
= st
->sb
;
1381 if (!super
->anchor
->num_raid_devs
)
1384 getinfo_super_imsm(st
, &info
, NULL
);
1385 fname_from_uuid(st
, &info
, nbuf
, ':');
1386 for (i
= 0; i
< super
->anchor
->num_raid_devs
; i
++) {
1387 struct imsm_dev
*dev
= get_imsm_dev(super
, i
);
1389 super
->current_vol
= i
;
1390 getinfo_super_imsm(st
, &info
, NULL
);
1391 fname_from_uuid(st
, &info
, nbuf1
, ':');
1392 printf("ARRAY /dev/md/%.16s container=%s member=%d UUID=%s\n",
1393 dev
->volume
, nbuf
+ 5, i
, nbuf1
+ 5);
1397 static void export_examine_super_imsm(struct supertype
*st
)
1399 struct intel_super
*super
= st
->sb
;
1400 struct imsm_super
*mpb
= super
->anchor
;
1404 getinfo_super_imsm(st
, &info
, NULL
);
1405 fname_from_uuid(st
, &info
, nbuf
, ':');
1406 printf("MD_METADATA=imsm\n");
1407 printf("MD_LEVEL=container\n");
1408 printf("MD_UUID=%s\n", nbuf
+5);
1409 printf("MD_DEVICES=%u\n", mpb
->num_disks
);
1412 static void detail_super_imsm(struct supertype
*st
, char *homehost
)
1417 getinfo_super_imsm(st
, &info
, NULL
);
1418 fname_from_uuid(st
, &info
, nbuf
, ':');
1419 printf("\n UUID : %s\n", nbuf
+ 5);
1422 static void brief_detail_super_imsm(struct supertype
*st
)
1426 getinfo_super_imsm(st
, &info
, NULL
);
1427 fname_from_uuid(st
, &info
, nbuf
, ':');
1428 printf(" UUID=%s", nbuf
+ 5);
1431 static int imsm_read_serial(int fd
, char *devname
, __u8
*serial
);
1432 static void fd2devname(int fd
, char *name
);
1434 static int ahci_enumerate_ports(const char *hba_path
, int port_count
, int host_base
, int verbose
)
1436 /* dump an unsorted list of devices attached to AHCI Intel storage
1437 * controller, as well as non-connected ports
1439 int hba_len
= strlen(hba_path
) + 1;
1444 unsigned long port_mask
= (1 << port_count
) - 1;
1446 if (port_count
> (int)sizeof(port_mask
) * 8) {
1448 fprintf(stderr
, Name
": port_count %d out of range\n", port_count
);
1452 /* scroll through /sys/dev/block looking for devices attached to
1455 dir
= opendir("/sys/dev/block");
1456 for (ent
= dir
? readdir(dir
) : NULL
; ent
; ent
= readdir(dir
)) {
1467 if (sscanf(ent
->d_name
, "%d:%d", &major
, &minor
) != 2)
1469 path
= devt_to_devpath(makedev(major
, minor
));
1472 if (!path_attached_to_hba(path
, hba_path
)) {
1478 /* retrieve the scsi device type */
1479 if (asprintf(&device
, "/sys/dev/block/%d:%d/device/xxxxxxx", major
, minor
) < 0) {
1481 fprintf(stderr
, Name
": failed to allocate 'device'\n");
1485 sprintf(device
, "/sys/dev/block/%d:%d/device/type", major
, minor
);
1486 if (load_sys(device
, buf
) != 0) {
1488 fprintf(stderr
, Name
": failed to read device type for %s\n",
1494 type
= strtoul(buf
, NULL
, 10);
1496 /* if it's not a disk print the vendor and model */
1497 if (!(type
== 0 || type
== 7 || type
== 14)) {
1500 sprintf(device
, "/sys/dev/block/%d:%d/device/vendor", major
, minor
);
1501 if (load_sys(device
, buf
) == 0) {
1502 strncpy(vendor
, buf
, sizeof(vendor
));
1503 vendor
[sizeof(vendor
) - 1] = '\0';
1504 c
= (char *) &vendor
[sizeof(vendor
) - 1];
1505 while (isspace(*c
) || *c
== '\0')
1509 sprintf(device
, "/sys/dev/block/%d:%d/device/model", major
, minor
);
1510 if (load_sys(device
, buf
) == 0) {
1511 strncpy(model
, buf
, sizeof(model
));
1512 model
[sizeof(model
) - 1] = '\0';
1513 c
= (char *) &model
[sizeof(model
) - 1];
1514 while (isspace(*c
) || *c
== '\0')
1518 if (vendor
[0] && model
[0])
1519 sprintf(buf
, "%.64s %.64s", vendor
, model
);
1521 switch (type
) { /* numbers from hald/linux/device.c */
1522 case 1: sprintf(buf
, "tape"); break;
1523 case 2: sprintf(buf
, "printer"); break;
1524 case 3: sprintf(buf
, "processor"); break;
1526 case 5: sprintf(buf
, "cdrom"); break;
1527 case 6: sprintf(buf
, "scanner"); break;
1528 case 8: sprintf(buf
, "media_changer"); break;
1529 case 9: sprintf(buf
, "comm"); break;
1530 case 12: sprintf(buf
, "raid"); break;
1531 default: sprintf(buf
, "unknown");
1537 /* chop device path to 'host%d' and calculate the port number */
1538 c
= strchr(&path
[hba_len
], '/');
1541 fprintf(stderr
, Name
": %s - invalid path name\n", path
+ hba_len
);
1546 if (sscanf(&path
[hba_len
], "host%d", &port
) == 1)
1550 *c
= '/'; /* repair the full string */
1551 fprintf(stderr
, Name
": failed to determine port number for %s\n",
1558 /* mark this port as used */
1559 port_mask
&= ~(1 << port
);
1561 /* print out the device information */
1563 printf(" Port%d : - non-disk device (%s) -\n", port
, buf
);
1567 fd
= dev_open(ent
->d_name
, O_RDONLY
);
1569 printf(" Port%d : - disk info unavailable -\n", port
);
1571 fd2devname(fd
, buf
);
1572 printf(" Port%d : %s", port
, buf
);
1573 if (imsm_read_serial(fd
, NULL
, (__u8
*) buf
) == 0)
1574 printf(" (%.*s)\n", MAX_RAID_SERIAL_LEN
, buf
);
1589 for (i
= 0; i
< port_count
; i
++)
1590 if (port_mask
& (1 << i
))
1591 printf(" Port%d : - no device attached -\n", i
);
1597 static void print_found_intel_controllers(struct sys_dev
*elem
)
1599 for (; elem
; elem
= elem
->next
) {
1600 fprintf(stderr
, Name
": found Intel(R) ");
1601 if (elem
->type
== SYS_DEV_SATA
)
1602 fprintf(stderr
, "SATA ");
1603 else if (elem
->type
== SYS_DEV_SAS
)
1604 fprintf(stderr
, "SAS ");
1605 fprintf(stderr
, "RAID controller");
1607 fprintf(stderr
, " at %s", elem
->pci_id
);
1608 fprintf(stderr
, ".\n");
1613 static int ahci_get_port_count(const char *hba_path
, int *port_count
)
1620 if ((dir
= opendir(hba_path
)) == NULL
)
1623 for (ent
= readdir(dir
); ent
; ent
= readdir(dir
)) {
1626 if (sscanf(ent
->d_name
, "host%d", &host
) != 1)
1628 if (*port_count
== 0)
1630 else if (host
< host_base
)
1633 if (host
+ 1 > *port_count
+ host_base
)
1634 *port_count
= host
+ 1 - host_base
;
1640 static void print_imsm_capability(const struct imsm_orom
*orom
)
1642 printf(" Platform : Intel(R) Matrix Storage Manager\n");
1643 printf(" Version : %d.%d.%d.%d\n", orom
->major_ver
, orom
->minor_ver
,
1644 orom
->hotfix_ver
, orom
->build
);
1645 printf(" RAID Levels :%s%s%s%s%s\n",
1646 imsm_orom_has_raid0(orom
) ? " raid0" : "",
1647 imsm_orom_has_raid1(orom
) ? " raid1" : "",
1648 imsm_orom_has_raid1e(orom
) ? " raid1e" : "",
1649 imsm_orom_has_raid10(orom
) ? " raid10" : "",
1650 imsm_orom_has_raid5(orom
) ? " raid5" : "");
1651 printf(" Chunk Sizes :%s%s%s%s%s%s%s%s%s%s%s%s%s%s%s%s\n",
1652 imsm_orom_has_chunk(orom
, 2) ? " 2k" : "",
1653 imsm_orom_has_chunk(orom
, 4) ? " 4k" : "",
1654 imsm_orom_has_chunk(orom
, 8) ? " 8k" : "",
1655 imsm_orom_has_chunk(orom
, 16) ? " 16k" : "",
1656 imsm_orom_has_chunk(orom
, 32) ? " 32k" : "",
1657 imsm_orom_has_chunk(orom
, 64) ? " 64k" : "",
1658 imsm_orom_has_chunk(orom
, 128) ? " 128k" : "",
1659 imsm_orom_has_chunk(orom
, 256) ? " 256k" : "",
1660 imsm_orom_has_chunk(orom
, 512) ? " 512k" : "",
1661 imsm_orom_has_chunk(orom
, 1024*1) ? " 1M" : "",
1662 imsm_orom_has_chunk(orom
, 1024*2) ? " 2M" : "",
1663 imsm_orom_has_chunk(orom
, 1024*4) ? " 4M" : "",
1664 imsm_orom_has_chunk(orom
, 1024*8) ? " 8M" : "",
1665 imsm_orom_has_chunk(orom
, 1024*16) ? " 16M" : "",
1666 imsm_orom_has_chunk(orom
, 1024*32) ? " 32M" : "",
1667 imsm_orom_has_chunk(orom
, 1024*64) ? " 64M" : "");
1668 printf(" Max Disks : %d\n", orom
->tds
);
1669 printf(" Max Volumes : %d\n", orom
->vpa
);
1673 static int detail_platform_imsm(int verbose
, int enumerate_only
)
1675 /* There are two components to imsm platform support, the ahci SATA
1676 * controller and the option-rom. To find the SATA controller we
1677 * simply look in /sys/bus/pci/drivers/ahci to see if an ahci
1678 * controller with the Intel vendor id is present. This approach
1679 * allows mdadm to leverage the kernel's ahci detection logic, with the
1680 * caveat that if ahci.ko is not loaded mdadm will not be able to
1681 * detect platform raid capabilities. The option-rom resides in a
1682 * platform "Adapter ROM". We scan for its signature to retrieve the
1683 * platform capabilities. If raid support is disabled in the BIOS the
1684 * option-rom capability structure will not be available.
1686 const struct imsm_orom
*orom
;
1687 struct sys_dev
*list
, *hba
;
1692 if (enumerate_only
) {
1693 if (check_env("IMSM_NO_PLATFORM"))
1695 list
= find_intel_devices();
1698 for (hba
= list
; hba
; hba
= hba
->next
) {
1699 orom
= find_imsm_capability(hba
->type
);
1705 free_sys_dev(&list
);
1709 list
= find_intel_devices();
1712 fprintf(stderr
, Name
": no active Intel(R) RAID "
1713 "controller found.\n");
1714 free_sys_dev(&list
);
1717 print_found_intel_controllers(list
);
1719 for (hba
= list
; hba
; hba
= hba
->next
) {
1720 orom
= find_imsm_capability(hba
->type
);
1722 fprintf(stderr
, Name
": imsm capabilities not found for controller: %s (type %s)\n",
1723 hba
->path
, get_sys_dev_type(hba
->type
));
1725 print_imsm_capability(orom
);
1728 for (hba
= list
; hba
; hba
= hba
->next
) {
1729 printf(" I/O Controller : %s (%s)\n",
1730 hba
->path
, get_sys_dev_type(hba
->type
));
1732 if (hba
->type
== SYS_DEV_SATA
) {
1733 host_base
= ahci_get_port_count(hba
->path
, &port_count
);
1734 if (ahci_enumerate_ports(hba
->path
, port_count
, host_base
, verbose
)) {
1736 fprintf(stderr
, Name
": failed to enumerate "
1737 "ports on SATA controller at %s.", hba
->pci_id
);
1743 free_sys_dev(&list
);
1748 static int match_home_imsm(struct supertype
*st
, char *homehost
)
1750 /* the imsm metadata format does not specify any host
1751 * identification information. We return -1 since we can never
1752 * confirm nor deny whether a given array is "meant" for this
1753 * host. We rely on compare_super and the 'family_num' fields to
1754 * exclude member disks that do not belong, and we rely on
1755 * mdadm.conf to specify the arrays that should be assembled.
1756 * Auto-assembly may still pick up "foreign" arrays.
1762 static void uuid_from_super_imsm(struct supertype
*st
, int uuid
[4])
1764 /* The uuid returned here is used for:
1765 * uuid to put into bitmap file (Create, Grow)
1766 * uuid for backup header when saving critical section (Grow)
1767 * comparing uuids when re-adding a device into an array
1768 * In these cases the uuid required is that of the data-array,
1769 * not the device-set.
1770 * uuid to recognise same set when adding a missing device back
1771 * to an array. This is a uuid for the device-set.
1773 * For each of these we can make do with a truncated
1774 * or hashed uuid rather than the original, as long as
1776 * In each case the uuid required is that of the data-array,
1777 * not the device-set.
1779 /* imsm does not track uuid's so we synthesis one using sha1 on
1780 * - The signature (Which is constant for all imsm array, but no matter)
1781 * - the orig_family_num of the container
1782 * - the index number of the volume
1783 * - the 'serial' number of the volume.
1784 * Hopefully these are all constant.
1786 struct intel_super
*super
= st
->sb
;
1789 struct sha1_ctx ctx
;
1790 struct imsm_dev
*dev
= NULL
;
1793 /* some mdadm versions failed to set ->orig_family_num, in which
1794 * case fall back to ->family_num. orig_family_num will be
1795 * fixed up with the first metadata update.
1797 family_num
= super
->anchor
->orig_family_num
;
1798 if (family_num
== 0)
1799 family_num
= super
->anchor
->family_num
;
1800 sha1_init_ctx(&ctx
);
1801 sha1_process_bytes(super
->anchor
->sig
, MPB_SIG_LEN
, &ctx
);
1802 sha1_process_bytes(&family_num
, sizeof(__u32
), &ctx
);
1803 if (super
->current_vol
>= 0)
1804 dev
= get_imsm_dev(super
, super
->current_vol
);
1806 __u32 vol
= super
->current_vol
;
1807 sha1_process_bytes(&vol
, sizeof(vol
), &ctx
);
1808 sha1_process_bytes(dev
->volume
, MAX_RAID_SERIAL_LEN
, &ctx
);
1810 sha1_finish_ctx(&ctx
, buf
);
1811 memcpy(uuid
, buf
, 4*4);
1816 get_imsm_numerical_version(struct imsm_super
*mpb
, int *m
, int *p
)
1818 __u8
*v
= get_imsm_version(mpb
);
1819 __u8
*end
= mpb
->sig
+ MAX_SIGNATURE_LENGTH
;
1820 char major
[] = { 0, 0, 0 };
1821 char minor
[] = { 0 ,0, 0 };
1822 char patch
[] = { 0, 0, 0 };
1823 char *ver_parse
[] = { major
, minor
, patch
};
1827 while (*v
!= '\0' && v
< end
) {
1828 if (*v
!= '.' && j
< 2)
1829 ver_parse
[i
][j
++] = *v
;
1837 *m
= strtol(minor
, NULL
, 0);
1838 *p
= strtol(patch
, NULL
, 0);
1842 static __u32
migr_strip_blocks_resync(struct imsm_dev
*dev
)
1844 /* migr_strip_size when repairing or initializing parity */
1845 struct imsm_map
*map
= get_imsm_map(dev
, 0);
1846 __u32 chunk
= __le32_to_cpu(map
->blocks_per_strip
);
1848 switch (get_imsm_raid_level(map
)) {
1853 return 128*1024 >> 9;
1857 static __u32
migr_strip_blocks_rebuild(struct imsm_dev
*dev
)
1859 /* migr_strip_size when rebuilding a degraded disk, no idea why
1860 * this is different than migr_strip_size_resync(), but it's good
1863 struct imsm_map
*map
= get_imsm_map(dev
, 1);
1864 __u32 chunk
= __le32_to_cpu(map
->blocks_per_strip
);
1866 switch (get_imsm_raid_level(map
)) {
1869 if (map
->num_members
% map
->num_domains
== 0)
1870 return 128*1024 >> 9;
1874 return max((__u32
) 64*1024 >> 9, chunk
);
1876 return 128*1024 >> 9;
1880 static __u32
num_stripes_per_unit_resync(struct imsm_dev
*dev
)
1882 struct imsm_map
*lo
= get_imsm_map(dev
, 0);
1883 struct imsm_map
*hi
= get_imsm_map(dev
, 1);
1884 __u32 lo_chunk
= __le32_to_cpu(lo
->blocks_per_strip
);
1885 __u32 hi_chunk
= __le32_to_cpu(hi
->blocks_per_strip
);
1887 return max((__u32
) 1, hi_chunk
/ lo_chunk
);
1890 static __u32
num_stripes_per_unit_rebuild(struct imsm_dev
*dev
)
1892 struct imsm_map
*lo
= get_imsm_map(dev
, 0);
1893 int level
= get_imsm_raid_level(lo
);
1895 if (level
== 1 || level
== 10) {
1896 struct imsm_map
*hi
= get_imsm_map(dev
, 1);
1898 return hi
->num_domains
;
1900 return num_stripes_per_unit_resync(dev
);
1903 static __u8
imsm_num_data_members(struct imsm_dev
*dev
, int second_map
)
1905 /* named 'imsm_' because raid0, raid1 and raid10
1906 * counter-intuitively have the same number of data disks
1908 struct imsm_map
*map
= get_imsm_map(dev
, second_map
);
1910 switch (get_imsm_raid_level(map
)) {
1914 return map
->num_members
;
1916 return map
->num_members
- 1;
1918 dprintf("%s: unsupported raid level\n", __func__
);
1923 static __u32
parity_segment_depth(struct imsm_dev
*dev
)
1925 struct imsm_map
*map
= get_imsm_map(dev
, 0);
1926 __u32 chunk
= __le32_to_cpu(map
->blocks_per_strip
);
1928 switch(get_imsm_raid_level(map
)) {
1931 return chunk
* map
->num_domains
;
1933 return chunk
* map
->num_members
;
1939 static __u32
map_migr_block(struct imsm_dev
*dev
, __u32 block
)
1941 struct imsm_map
*map
= get_imsm_map(dev
, 1);
1942 __u32 chunk
= __le32_to_cpu(map
->blocks_per_strip
);
1943 __u32 strip
= block
/ chunk
;
1945 switch (get_imsm_raid_level(map
)) {
1948 __u32 vol_strip
= (strip
* map
->num_domains
) + 1;
1949 __u32 vol_stripe
= vol_strip
/ map
->num_members
;
1951 return vol_stripe
* chunk
+ block
% chunk
;
1953 __u32 stripe
= strip
/ (map
->num_members
- 1);
1955 return stripe
* chunk
+ block
% chunk
;
1962 static __u64
blocks_per_migr_unit(struct intel_super
*super
,
1963 struct imsm_dev
*dev
)
1965 /* calculate the conversion factor between per member 'blocks'
1966 * (md/{resync,rebuild}_start) and imsm migration units, return
1967 * 0 for the 'not migrating' and 'unsupported migration' cases
1969 if (!dev
->vol
.migr_state
)
1972 switch (migr_type(dev
)) {
1973 case MIGR_GEN_MIGR
: {
1974 struct migr_record
*migr_rec
= super
->migr_rec
;
1975 return __le32_to_cpu(migr_rec
->blocks_per_unit
);
1980 struct imsm_map
*map
= get_imsm_map(dev
, 0);
1981 __u32 stripes_per_unit
;
1982 __u32 blocks_per_unit
;
1991 /* yes, this is really the translation of migr_units to
1992 * per-member blocks in the 'resync' case
1994 stripes_per_unit
= num_stripes_per_unit_resync(dev
);
1995 migr_chunk
= migr_strip_blocks_resync(dev
);
1996 disks
= imsm_num_data_members(dev
, 0);
1997 blocks_per_unit
= stripes_per_unit
* migr_chunk
* disks
;
1998 stripe
= __le16_to_cpu(map
->blocks_per_strip
) * disks
;
1999 segment
= blocks_per_unit
/ stripe
;
2000 block_rel
= blocks_per_unit
- segment
* stripe
;
2001 parity_depth
= parity_segment_depth(dev
);
2002 block_map
= map_migr_block(dev
, block_rel
);
2003 return block_map
+ parity_depth
* segment
;
2005 case MIGR_REBUILD
: {
2006 __u32 stripes_per_unit
;
2009 stripes_per_unit
= num_stripes_per_unit_rebuild(dev
);
2010 migr_chunk
= migr_strip_blocks_rebuild(dev
);
2011 return migr_chunk
* stripes_per_unit
;
2013 case MIGR_STATE_CHANGE
:
2019 static int imsm_level_to_layout(int level
)
2027 return ALGORITHM_LEFT_ASYMMETRIC
;
2034 /*******************************************************************************
2035 * Function: read_imsm_migr_rec
2036 * Description: Function reads imsm migration record from last sector of disk
2038 * fd : disk descriptor
2039 * super : metadata info
2043 ******************************************************************************/
2044 static int read_imsm_migr_rec(int fd
, struct intel_super
*super
)
2047 unsigned long long dsize
;
2049 get_dev_size(fd
, NULL
, &dsize
);
2050 if (lseek64(fd
, dsize
- 512, SEEK_SET
) < 0) {
2052 Name
": Cannot seek to anchor block: %s\n",
2056 if (read(fd
, super
->migr_rec_buf
, 512) != 512) {
2058 Name
": Cannot read migr record block: %s\n",
2068 /*******************************************************************************
2069 * Function: load_imsm_migr_rec
2070 * Description: Function reads imsm migration record (it is stored at the last
2073 * super : imsm internal array info
2074 * info : general array info
2078 ******************************************************************************/
2079 static int load_imsm_migr_rec(struct intel_super
*super
, struct mdinfo
*info
)
2082 struct dl
*dl
= NULL
;
2088 for (sd
= info
->devs
; sd
; sd
= sd
->next
) {
2089 /* read only from one of the first two slots */
2090 if ((sd
->disk
.raid_disk
> 1) ||
2091 (sd
->disk
.raid_disk
< 0))
2093 sprintf(nm
, "%d:%d", sd
->disk
.major
, sd
->disk
.minor
);
2094 fd
= dev_open(nm
, O_RDONLY
);
2100 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
2101 /* read only from one of the first two slots */
2104 sprintf(nm
, "%d:%d", dl
->major
, dl
->minor
);
2105 fd
= dev_open(nm
, O_RDONLY
);
2112 retval
= read_imsm_migr_rec(fd
, super
);
2121 /*******************************************************************************
2122 * function: imsm_create_metadata_checkpoint_update
2123 * Description: It creates update for checkpoint change.
2125 * super : imsm internal array info
2126 * u : pointer to prepared update
2129 * If length is equal to 0, input pointer u contains no update
2130 ******************************************************************************/
2131 static int imsm_create_metadata_checkpoint_update(
2132 struct intel_super
*super
,
2133 struct imsm_update_general_migration_checkpoint
**u
)
2136 int update_memory_size
= 0;
2138 dprintf("imsm_create_metadata_checkpoint_update(enter)\n");
2144 /* size of all update data without anchor */
2145 update_memory_size
=
2146 sizeof(struct imsm_update_general_migration_checkpoint
);
2148 *u
= calloc(1, update_memory_size
);
2150 dprintf("error: cannot get memory for "
2151 "imsm_create_metadata_checkpoint_update update\n");
2154 (*u
)->type
= update_general_migration_checkpoint
;
2155 (*u
)->curr_migr_unit
= __le32_to_cpu(super
->migr_rec
->curr_migr_unit
);
2156 dprintf("imsm_create_metadata_checkpoint_update: prepared for %u\n",
2157 (*u
)->curr_migr_unit
);
2159 return update_memory_size
;
2163 static void imsm_update_metadata_locally(struct supertype
*st
,
2164 void *buf
, int len
);
2166 /*******************************************************************************
2167 * Function: write_imsm_migr_rec
2168 * Description: Function writes imsm migration record
2169 * (at the last sector of disk)
2171 * super : imsm internal array info
2175 ******************************************************************************/
2176 static int write_imsm_migr_rec(struct supertype
*st
)
2178 struct intel_super
*super
= st
->sb
;
2179 unsigned long long dsize
;
2185 struct imsm_update_general_migration_checkpoint
*u
;
2187 for (sd
= super
->disks
; sd
; sd
= sd
->next
) {
2188 /* write to 2 first slots only */
2189 if ((sd
->index
< 0) || (sd
->index
> 1))
2191 sprintf(nm
, "%d:%d", sd
->major
, sd
->minor
);
2192 fd
= dev_open(nm
, O_RDWR
);
2195 get_dev_size(fd
, NULL
, &dsize
);
2196 if (lseek64(fd
, dsize
- 512, SEEK_SET
) < 0) {
2198 Name
": Cannot seek to anchor block: %s\n",
2202 if (write(fd
, super
->migr_rec_buf
, 512) != 512) {
2204 Name
": Cannot write migr record block: %s\n",
2211 /* update checkpoint information in metadata */
2212 len
= imsm_create_metadata_checkpoint_update(super
, &u
);
2215 dprintf("imsm: Cannot prepare update\n");
2218 /* update metadata locally */
2219 imsm_update_metadata_locally(st
, u
, len
);
2220 /* and possibly remotely */
2221 if (st
->update_tail
) {
2222 append_metadata_update(st
, u
, len
);
2223 /* during reshape we do all work inside metadata handler
2224 * manage_reshape(), so metadata update has to be triggered
2227 flush_metadata_updates(st
);
2228 st
->update_tail
= &st
->updates
;
2238 #endif /* MDASSEMBLE */
2240 static void getinfo_super_imsm_volume(struct supertype
*st
, struct mdinfo
*info
, char *dmap
)
2242 struct intel_super
*super
= st
->sb
;
2243 struct migr_record
*migr_rec
= super
->migr_rec
;
2244 struct imsm_dev
*dev
= get_imsm_dev(super
, super
->current_vol
);
2245 struct imsm_map
*map
= get_imsm_map(dev
, 0);
2246 struct imsm_map
*prev_map
= get_imsm_map(dev
, 1);
2247 struct imsm_map
*map_to_analyse
= map
;
2250 unsigned int component_size_alligment
;
2251 int map_disks
= info
->array
.raid_disks
;
2253 memset(info
, 0, sizeof(*info
));
2255 map_to_analyse
= prev_map
;
2257 dl
= super
->current_disk
;
2259 info
->container_member
= super
->current_vol
;
2260 info
->array
.raid_disks
= map
->num_members
;
2261 info
->array
.level
= get_imsm_raid_level(map_to_analyse
);
2262 info
->array
.layout
= imsm_level_to_layout(info
->array
.level
);
2263 info
->array
.md_minor
= -1;
2264 info
->array
.ctime
= 0;
2265 info
->array
.utime
= 0;
2266 info
->array
.chunk_size
=
2267 __le16_to_cpu(map_to_analyse
->blocks_per_strip
) << 9;
2268 info
->array
.state
= !dev
->vol
.dirty
;
2269 info
->custom_array_size
= __le32_to_cpu(dev
->size_high
);
2270 info
->custom_array_size
<<= 32;
2271 info
->custom_array_size
|= __le32_to_cpu(dev
->size_low
);
2272 if (prev_map
&& map
->map_state
== prev_map
->map_state
) {
2273 info
->reshape_active
= 1;
2274 info
->new_level
= get_imsm_raid_level(map
);
2275 info
->new_layout
= imsm_level_to_layout(info
->new_level
);
2276 info
->new_chunk
= __le16_to_cpu(map
->blocks_per_strip
) << 9;
2277 info
->delta_disks
= map
->num_members
- prev_map
->num_members
;
2278 if (info
->delta_disks
) {
2279 /* this needs to be applied to every array
2282 info
->reshape_active
= 2;
2284 /* We shape information that we give to md might have to be
2285 * modify to cope with md's requirement for reshaping arrays.
2286 * For example, when reshaping a RAID0, md requires it to be
2287 * presented as a degraded RAID4.
2288 * Also if a RAID0 is migrating to a RAID5 we need to specify
2289 * the array as already being RAID5, but the 'before' layout
2290 * is a RAID4-like layout.
2292 switch (info
->array
.level
) {
2294 switch(info
->new_level
) {
2296 /* conversion is happening as RAID4 */
2297 info
->array
.level
= 4;
2298 info
->array
.raid_disks
+= 1;
2301 /* conversion is happening as RAID5 */
2302 info
->array
.level
= 5;
2303 info
->array
.layout
= ALGORITHM_PARITY_N
;
2304 info
->delta_disks
-= 1;
2307 /* FIXME error message */
2308 info
->array
.level
= UnSet
;
2314 info
->new_level
= UnSet
;
2315 info
->new_layout
= UnSet
;
2316 info
->new_chunk
= info
->array
.chunk_size
;
2317 info
->delta_disks
= 0;
2321 info
->disk
.major
= dl
->major
;
2322 info
->disk
.minor
= dl
->minor
;
2323 info
->disk
.number
= dl
->index
;
2324 info
->disk
.raid_disk
= get_imsm_disk_slot(map_to_analyse
,
2328 info
->data_offset
= __le32_to_cpu(map_to_analyse
->pba_of_lba0
);
2329 info
->component_size
=
2330 __le32_to_cpu(map_to_analyse
->blocks_per_member
);
2332 /* check component size aligment
2334 component_size_alligment
=
2335 info
->component_size
% (info
->array
.chunk_size
/512);
2337 if (component_size_alligment
&&
2338 (info
->array
.level
!= 1) && (info
->array
.level
!= UnSet
)) {
2339 dprintf("imsm: reported component size alligned from %llu ",
2340 info
->component_size
);
2341 info
->component_size
-= component_size_alligment
;
2342 dprintf("to %llu (%i).\n",
2343 info
->component_size
, component_size_alligment
);
2346 memset(info
->uuid
, 0, sizeof(info
->uuid
));
2347 info
->recovery_start
= MaxSector
;
2349 info
->reshape_progress
= 0;
2350 info
->resync_start
= MaxSector
;
2351 if (map_to_analyse
->map_state
== IMSM_T_STATE_UNINITIALIZED
||
2353 info
->resync_start
= 0;
2355 if (dev
->vol
.migr_state
) {
2356 switch (migr_type(dev
)) {
2359 __u64 blocks_per_unit
= blocks_per_migr_unit(super
,
2361 __u64 units
= __le32_to_cpu(dev
->vol
.curr_migr_unit
);
2363 info
->resync_start
= blocks_per_unit
* units
;
2366 case MIGR_GEN_MIGR
: {
2367 __u64 blocks_per_unit
= blocks_per_migr_unit(super
,
2369 __u64 units
= __le32_to_cpu(migr_rec
->curr_migr_unit
);
2370 unsigned long long array_blocks
;
2373 if (__le32_to_cpu(migr_rec
->ascending_migr
) &&
2375 (__le32_to_cpu(migr_rec
->num_migr_units
)-1)) &&
2376 (super
->migr_rec
->rec_status
==
2377 __cpu_to_le32(UNIT_SRC_IN_CP_AREA
)))
2380 info
->reshape_progress
= blocks_per_unit
* units
;
2382 dprintf("IMSM: General Migration checkpoint : %llu "
2383 "(%llu) -> read reshape progress : %llu\n",
2384 (unsigned long long)units
,
2385 (unsigned long long)blocks_per_unit
,
2386 info
->reshape_progress
);
2388 used_disks
= imsm_num_data_members(dev
, 1);
2389 if (used_disks
> 0) {
2390 array_blocks
= map
->blocks_per_member
*
2392 /* round array size down to closest MB
2394 info
->custom_array_size
= (array_blocks
2395 >> SECT_PER_MB_SHIFT
)
2396 << SECT_PER_MB_SHIFT
;
2400 /* we could emulate the checkpointing of
2401 * 'sync_action=check' migrations, but for now
2402 * we just immediately complete them
2405 /* this is handled by container_content_imsm() */
2406 case MIGR_STATE_CHANGE
:
2407 /* FIXME handle other migrations */
2409 /* we are not dirty, so... */
2410 info
->resync_start
= MaxSector
;
2414 strncpy(info
->name
, (char *) dev
->volume
, MAX_RAID_SERIAL_LEN
);
2415 info
->name
[MAX_RAID_SERIAL_LEN
] = 0;
2417 info
->array
.major_version
= -1;
2418 info
->array
.minor_version
= -2;
2419 devname
= devnum2devname(st
->container_dev
);
2420 *info
->text_version
= '\0';
2422 sprintf(info
->text_version
, "/%s/%d", devname
, info
->container_member
);
2424 info
->safe_mode_delay
= 4000; /* 4 secs like the Matrix driver */
2425 uuid_from_super_imsm(st
, info
->uuid
);
2429 for (i
=0; i
<map_disks
; i
++) {
2431 if (i
< info
->array
.raid_disks
) {
2432 struct imsm_disk
*dsk
;
2433 j
= get_imsm_disk_idx(dev
, i
, -1);
2434 dsk
= get_imsm_disk(super
, j
);
2435 if (dsk
&& (dsk
->status
& CONFIGURED_DISK
))
2442 static __u8
imsm_check_degraded(struct intel_super
*super
, struct imsm_dev
*dev
, int failed
);
2443 static int imsm_count_failed(struct intel_super
*super
, struct imsm_dev
*dev
);
2445 static struct imsm_disk
*get_imsm_missing(struct intel_super
*super
, __u8 index
)
2449 for (d
= super
->missing
; d
; d
= d
->next
)
2450 if (d
->index
== index
)
2455 static void getinfo_super_imsm(struct supertype
*st
, struct mdinfo
*info
, char *map
)
2457 struct intel_super
*super
= st
->sb
;
2458 struct imsm_disk
*disk
;
2459 int map_disks
= info
->array
.raid_disks
;
2460 int max_enough
= -1;
2462 struct imsm_super
*mpb
;
2464 if (super
->current_vol
>= 0) {
2465 getinfo_super_imsm_volume(st
, info
, map
);
2468 memset(info
, 0, sizeof(*info
));
2470 /* Set raid_disks to zero so that Assemble will always pull in valid
2473 info
->array
.raid_disks
= 0;
2474 info
->array
.level
= LEVEL_CONTAINER
;
2475 info
->array
.layout
= 0;
2476 info
->array
.md_minor
= -1;
2477 info
->array
.ctime
= 0; /* N/A for imsm */
2478 info
->array
.utime
= 0;
2479 info
->array
.chunk_size
= 0;
2481 info
->disk
.major
= 0;
2482 info
->disk
.minor
= 0;
2483 info
->disk
.raid_disk
= -1;
2484 info
->reshape_active
= 0;
2485 info
->array
.major_version
= -1;
2486 info
->array
.minor_version
= -2;
2487 strcpy(info
->text_version
, "imsm");
2488 info
->safe_mode_delay
= 0;
2489 info
->disk
.number
= -1;
2490 info
->disk
.state
= 0;
2492 info
->recovery_start
= MaxSector
;
2494 /* do we have the all the insync disks that we expect? */
2495 mpb
= super
->anchor
;
2497 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
2498 struct imsm_dev
*dev
= get_imsm_dev(super
, i
);
2499 int failed
, enough
, j
, missing
= 0;
2500 struct imsm_map
*map
;
2503 failed
= imsm_count_failed(super
, dev
);
2504 state
= imsm_check_degraded(super
, dev
, failed
);
2505 map
= get_imsm_map(dev
, dev
->vol
.migr_state
);
2507 /* any newly missing disks?
2508 * (catches single-degraded vs double-degraded)
2510 for (j
= 0; j
< map
->num_members
; j
++) {
2511 __u32 ord
= get_imsm_ord_tbl_ent(dev
, i
, -1);
2512 __u32 idx
= ord_to_idx(ord
);
2514 if (!(ord
& IMSM_ORD_REBUILD
) &&
2515 get_imsm_missing(super
, idx
)) {
2521 if (state
== IMSM_T_STATE_FAILED
)
2523 else if (state
== IMSM_T_STATE_DEGRADED
&&
2524 (state
!= map
->map_state
|| missing
))
2526 else /* we're normal, or already degraded */
2529 /* in the missing/failed disk case check to see
2530 * if at least one array is runnable
2532 max_enough
= max(max_enough
, enough
);
2534 dprintf("%s: enough: %d\n", __func__
, max_enough
);
2535 info
->container_enough
= max_enough
;
2538 __u32 reserved
= imsm_reserved_sectors(super
, super
->disks
);
2540 disk
= &super
->disks
->disk
;
2541 info
->data_offset
= __le32_to_cpu(disk
->total_blocks
) - reserved
;
2542 info
->component_size
= reserved
;
2543 info
->disk
.state
= is_configured(disk
) ? (1 << MD_DISK_ACTIVE
) : 0;
2544 /* we don't change info->disk.raid_disk here because
2545 * this state will be finalized in mdmon after we have
2546 * found the 'most fresh' version of the metadata
2548 info
->disk
.state
|= is_failed(disk
) ? (1 << MD_DISK_FAULTY
) : 0;
2549 info
->disk
.state
|= is_spare(disk
) ? 0 : (1 << MD_DISK_SYNC
);
2552 /* only call uuid_from_super_imsm when this disk is part of a populated container,
2553 * ->compare_super may have updated the 'num_raid_devs' field for spares
2555 if (info
->disk
.state
& (1 << MD_DISK_SYNC
) || super
->anchor
->num_raid_devs
)
2556 uuid_from_super_imsm(st
, info
->uuid
);
2558 memcpy(info
->uuid
, uuid_zero
, sizeof(uuid_zero
));
2560 /* I don't know how to compute 'map' on imsm, so use safe default */
2563 for (i
= 0; i
< map_disks
; i
++)
2569 /* allocates memory and fills disk in mdinfo structure
2570 * for each disk in array */
2571 struct mdinfo
*getinfo_super_disks_imsm(struct supertype
*st
)
2573 struct mdinfo
*mddev
= NULL
;
2574 struct intel_super
*super
= st
->sb
;
2575 struct imsm_disk
*disk
;
2578 if (!super
|| !super
->disks
)
2581 mddev
= malloc(sizeof(*mddev
));
2583 fprintf(stderr
, Name
": Failed to allocate memory.\n");
2586 memset(mddev
, 0, sizeof(*mddev
));
2590 tmp
= malloc(sizeof(*tmp
));
2592 fprintf(stderr
, Name
": Failed to allocate memory.\n");
2597 memset(tmp
, 0, sizeof(*tmp
));
2599 tmp
->next
= mddev
->devs
;
2601 tmp
->disk
.number
= count
++;
2602 tmp
->disk
.major
= dl
->major
;
2603 tmp
->disk
.minor
= dl
->minor
;
2604 tmp
->disk
.state
= is_configured(disk
) ?
2605 (1 << MD_DISK_ACTIVE
) : 0;
2606 tmp
->disk
.state
|= is_failed(disk
) ? (1 << MD_DISK_FAULTY
) : 0;
2607 tmp
->disk
.state
|= is_spare(disk
) ? 0 : (1 << MD_DISK_SYNC
);
2608 tmp
->disk
.raid_disk
= -1;
2614 static int update_super_imsm(struct supertype
*st
, struct mdinfo
*info
,
2615 char *update
, char *devname
, int verbose
,
2616 int uuid_set
, char *homehost
)
2618 /* For 'assemble' and 'force' we need to return non-zero if any
2619 * change was made. For others, the return value is ignored.
2620 * Update options are:
2621 * force-one : This device looks a bit old but needs to be included,
2622 * update age info appropriately.
2623 * assemble: clear any 'faulty' flag to allow this device to
2625 * force-array: Array is degraded but being forced, mark it clean
2626 * if that will be needed to assemble it.
2628 * newdev: not used ????
2629 * grow: Array has gained a new device - this is currently for
2631 * resync: mark as dirty so a resync will happen.
2632 * name: update the name - preserving the homehost
2633 * uuid: Change the uuid of the array to match watch is given
2635 * Following are not relevant for this imsm:
2636 * sparc2.2 : update from old dodgey metadata
2637 * super-minor: change the preferred_minor number
2638 * summaries: update redundant counters.
2639 * homehost: update the recorded homehost
2640 * _reshape_progress: record new reshape_progress position.
2643 struct intel_super
*super
= st
->sb
;
2644 struct imsm_super
*mpb
;
2646 /* we can only update container info */
2647 if (!super
|| super
->current_vol
>= 0 || !super
->anchor
)
2650 mpb
= super
->anchor
;
2652 if (strcmp(update
, "uuid") == 0 && uuid_set
&& !info
->update_private
)
2654 else if (strcmp(update
, "uuid") == 0 && uuid_set
&& info
->update_private
) {
2655 mpb
->orig_family_num
= *((__u32
*) info
->update_private
);
2657 } else if (strcmp(update
, "uuid") == 0) {
2658 __u32
*new_family
= malloc(sizeof(*new_family
));
2660 /* update orig_family_number with the incoming random
2661 * data, report the new effective uuid, and store the
2662 * new orig_family_num for future updates.
2665 memcpy(&mpb
->orig_family_num
, info
->uuid
, sizeof(__u32
));
2666 uuid_from_super_imsm(st
, info
->uuid
);
2667 *new_family
= mpb
->orig_family_num
;
2668 info
->update_private
= new_family
;
2671 } else if (strcmp(update
, "assemble") == 0)
2676 /* successful update? recompute checksum */
2678 mpb
->check_sum
= __le32_to_cpu(__gen_imsm_checksum(mpb
));
2683 static size_t disks_to_mpb_size(int disks
)
2687 size
= sizeof(struct imsm_super
);
2688 size
+= (disks
- 1) * sizeof(struct imsm_disk
);
2689 size
+= 2 * sizeof(struct imsm_dev
);
2690 /* up to 2 maps per raid device (-2 for imsm_maps in imsm_dev */
2691 size
+= (4 - 2) * sizeof(struct imsm_map
);
2692 /* 4 possible disk_ord_tbl's */
2693 size
+= 4 * (disks
- 1) * sizeof(__u32
);
2698 static __u64
avail_size_imsm(struct supertype
*st
, __u64 devsize
)
2700 if (devsize
< (MPB_SECTOR_CNT
+ IMSM_RESERVED_SECTORS
))
2703 return devsize
- (MPB_SECTOR_CNT
+ IMSM_RESERVED_SECTORS
);
2706 static void free_devlist(struct intel_super
*super
)
2708 struct intel_dev
*dv
;
2710 while (super
->devlist
) {
2711 dv
= super
->devlist
->next
;
2712 free(super
->devlist
->dev
);
2713 free(super
->devlist
);
2714 super
->devlist
= dv
;
2718 static void imsm_copy_dev(struct imsm_dev
*dest
, struct imsm_dev
*src
)
2720 memcpy(dest
, src
, sizeof_imsm_dev(src
, 0));
2723 static int compare_super_imsm(struct supertype
*st
, struct supertype
*tst
)
2727 * 0 same, or first was empty, and second was copied
2728 * 1 second had wrong number
2730 * 3 wrong other info
2732 struct intel_super
*first
= st
->sb
;
2733 struct intel_super
*sec
= tst
->sb
;
2740 /* in platform dependent environment test if the disks
2741 * use the same Intel hba
2743 if (!check_env("IMSM_NO_PLATFORM")) {
2744 if (!first
->hba
|| !sec
->hba
||
2745 (first
->hba
->type
!= sec
->hba
->type
)) {
2747 "HBAs of devices does not match %s != %s\n",
2748 first
->hba
? get_sys_dev_type(first
->hba
->type
) : NULL
,
2749 sec
->hba
? get_sys_dev_type(sec
->hba
->type
) : NULL
);
2754 /* if an anchor does not have num_raid_devs set then it is a free
2757 if (first
->anchor
->num_raid_devs
> 0 &&
2758 sec
->anchor
->num_raid_devs
> 0) {
2759 /* Determine if these disks might ever have been
2760 * related. Further disambiguation can only take place
2761 * in load_super_imsm_all
2763 __u32 first_family
= first
->anchor
->orig_family_num
;
2764 __u32 sec_family
= sec
->anchor
->orig_family_num
;
2766 if (memcmp(first
->anchor
->sig
, sec
->anchor
->sig
,
2767 MAX_SIGNATURE_LENGTH
) != 0)
2770 if (first_family
== 0)
2771 first_family
= first
->anchor
->family_num
;
2772 if (sec_family
== 0)
2773 sec_family
= sec
->anchor
->family_num
;
2775 if (first_family
!= sec_family
)
2781 /* if 'first' is a spare promote it to a populated mpb with sec's
2784 if (first
->anchor
->num_raid_devs
== 0 &&
2785 sec
->anchor
->num_raid_devs
> 0) {
2787 struct intel_dev
*dv
;
2788 struct imsm_dev
*dev
;
2790 /* we need to copy raid device info from sec if an allocation
2791 * fails here we don't associate the spare
2793 for (i
= 0; i
< sec
->anchor
->num_raid_devs
; i
++) {
2794 dv
= malloc(sizeof(*dv
));
2797 dev
= malloc(sizeof_imsm_dev(get_imsm_dev(sec
, i
), 1));
2804 dv
->next
= first
->devlist
;
2805 first
->devlist
= dv
;
2807 if (i
< sec
->anchor
->num_raid_devs
) {
2808 /* allocation failure */
2809 free_devlist(first
);
2810 fprintf(stderr
, "imsm: failed to associate spare\n");
2813 first
->anchor
->num_raid_devs
= sec
->anchor
->num_raid_devs
;
2814 first
->anchor
->orig_family_num
= sec
->anchor
->orig_family_num
;
2815 first
->anchor
->family_num
= sec
->anchor
->family_num
;
2816 memcpy(first
->anchor
->sig
, sec
->anchor
->sig
, MAX_SIGNATURE_LENGTH
);
2817 for (i
= 0; i
< sec
->anchor
->num_raid_devs
; i
++)
2818 imsm_copy_dev(get_imsm_dev(first
, i
), get_imsm_dev(sec
, i
));
2824 static void fd2devname(int fd
, char *name
)
2828 char dname
[PATH_MAX
];
2833 if (fstat(fd
, &st
) != 0)
2835 sprintf(path
, "/sys/dev/block/%d:%d",
2836 major(st
.st_rdev
), minor(st
.st_rdev
));
2838 rv
= readlink(path
, dname
, sizeof(dname
));
2843 nm
= strrchr(dname
, '/');
2845 snprintf(name
, MAX_RAID_SERIAL_LEN
, "/dev/%s", nm
);
2848 extern int scsi_get_serial(int fd
, void *buf
, size_t buf_len
);
2850 static int imsm_read_serial(int fd
, char *devname
,
2851 __u8 serial
[MAX_RAID_SERIAL_LEN
])
2853 unsigned char scsi_serial
[255];
2862 memset(scsi_serial
, 0, sizeof(scsi_serial
));
2864 rv
= scsi_get_serial(fd
, scsi_serial
, sizeof(scsi_serial
));
2866 if (rv
&& check_env("IMSM_DEVNAME_AS_SERIAL")) {
2867 memset(serial
, 0, MAX_RAID_SERIAL_LEN
);
2868 fd2devname(fd
, (char *) serial
);
2875 Name
": Failed to retrieve serial for %s\n",
2880 rsp_len
= scsi_serial
[3];
2884 Name
": Failed to retrieve serial for %s\n",
2888 rsp_buf
= (char *) &scsi_serial
[4];
2890 /* trim all whitespace and non-printable characters and convert
2893 for (i
= 0, dest
= rsp_buf
; i
< rsp_len
; i
++) {
2896 /* ':' is reserved for use in placeholder serial
2897 * numbers for missing disks
2905 len
= dest
- rsp_buf
;
2908 /* truncate leading characters */
2909 if (len
> MAX_RAID_SERIAL_LEN
) {
2910 dest
+= len
- MAX_RAID_SERIAL_LEN
;
2911 len
= MAX_RAID_SERIAL_LEN
;
2914 memset(serial
, 0, MAX_RAID_SERIAL_LEN
);
2915 memcpy(serial
, dest
, len
);
2920 static int serialcmp(__u8
*s1
, __u8
*s2
)
2922 return strncmp((char *) s1
, (char *) s2
, MAX_RAID_SERIAL_LEN
);
2925 static void serialcpy(__u8
*dest
, __u8
*src
)
2927 strncpy((char *) dest
, (char *) src
, MAX_RAID_SERIAL_LEN
);
2930 static struct dl
*serial_to_dl(__u8
*serial
, struct intel_super
*super
)
2934 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
2935 if (serialcmp(dl
->serial
, serial
) == 0)
2941 static struct imsm_disk
*
2942 __serial_to_disk(__u8
*serial
, struct imsm_super
*mpb
, int *idx
)
2946 for (i
= 0; i
< mpb
->num_disks
; i
++) {
2947 struct imsm_disk
*disk
= __get_imsm_disk(mpb
, i
);
2949 if (serialcmp(disk
->serial
, serial
) == 0) {
2960 load_imsm_disk(int fd
, struct intel_super
*super
, char *devname
, int keep_fd
)
2962 struct imsm_disk
*disk
;
2967 __u8 serial
[MAX_RAID_SERIAL_LEN
];
2969 rv
= imsm_read_serial(fd
, devname
, serial
);
2974 dl
= calloc(1, sizeof(*dl
));
2978 Name
": failed to allocate disk buffer for %s\n",
2984 dl
->major
= major(stb
.st_rdev
);
2985 dl
->minor
= minor(stb
.st_rdev
);
2986 dl
->next
= super
->disks
;
2987 dl
->fd
= keep_fd
? fd
: -1;
2988 assert(super
->disks
== NULL
);
2990 serialcpy(dl
->serial
, serial
);
2993 fd2devname(fd
, name
);
2995 dl
->devname
= strdup(devname
);
2997 dl
->devname
= strdup(name
);
2999 /* look up this disk's index in the current anchor */
3000 disk
= __serial_to_disk(dl
->serial
, super
->anchor
, &dl
->index
);
3003 /* only set index on disks that are a member of a
3004 * populated contianer, i.e. one with raid_devs
3006 if (is_failed(&dl
->disk
))
3008 else if (is_spare(&dl
->disk
))
3016 /* When migrating map0 contains the 'destination' state while map1
3017 * contains the current state. When not migrating map0 contains the
3018 * current state. This routine assumes that map[0].map_state is set to
3019 * the current array state before being called.
3021 * Migration is indicated by one of the following states
3022 * 1/ Idle (migr_state=0 map0state=normal||unitialized||degraded||failed)
3023 * 2/ Initialize (migr_state=1 migr_type=MIGR_INIT map0state=normal
3024 * map1state=unitialized)
3025 * 3/ Repair (Resync) (migr_state=1 migr_type=MIGR_REPAIR map0state=normal
3027 * 4/ Rebuild (migr_state=1 migr_type=MIGR_REBUILD map0state=normal
3028 * map1state=degraded)
3029 * 5/ Migration (mig_state=1 migr_type=MIGR_GEN_MIGR map0state=normal
3032 static void migrate(struct imsm_dev
*dev
, struct intel_super
*super
,
3033 __u8 to_state
, int migr_type
)
3035 struct imsm_map
*dest
;
3036 struct imsm_map
*src
= get_imsm_map(dev
, 0);
3038 dev
->vol
.migr_state
= 1;
3039 set_migr_type(dev
, migr_type
);
3040 dev
->vol
.curr_migr_unit
= 0;
3041 dest
= get_imsm_map(dev
, 1);
3043 /* duplicate and then set the target end state in map[0] */
3044 memcpy(dest
, src
, sizeof_imsm_map(src
));
3045 if ((migr_type
== MIGR_REBUILD
) ||
3046 (migr_type
== MIGR_GEN_MIGR
)) {
3050 for (i
= 0; i
< src
->num_members
; i
++) {
3051 ord
= __le32_to_cpu(src
->disk_ord_tbl
[i
]);
3052 set_imsm_ord_tbl_ent(src
, i
, ord_to_idx(ord
));
3056 if (migr_type
== MIGR_GEN_MIGR
)
3057 /* Clear migration record */
3058 memset(super
->migr_rec
, 0, sizeof(struct migr_record
));
3060 src
->map_state
= to_state
;
3063 static void end_migration(struct imsm_dev
*dev
, __u8 map_state
)
3065 struct imsm_map
*map
= get_imsm_map(dev
, 0);
3066 struct imsm_map
*prev
= get_imsm_map(dev
, dev
->vol
.migr_state
);
3069 /* merge any IMSM_ORD_REBUILD bits that were not successfully
3070 * completed in the last migration.
3072 * FIXME add support for raid-level-migration
3074 for (i
= 0; i
< prev
->num_members
; i
++)
3075 for (j
= 0; j
< map
->num_members
; j
++)
3076 /* during online capacity expansion
3077 * disks position can be changed if takeover is used
3079 if (ord_to_idx(map
->disk_ord_tbl
[j
]) ==
3080 ord_to_idx(prev
->disk_ord_tbl
[i
])) {
3081 map
->disk_ord_tbl
[j
] |= prev
->disk_ord_tbl
[i
];
3085 dev
->vol
.migr_state
= 0;
3086 dev
->vol
.migr_type
= 0;
3087 dev
->vol
.curr_migr_unit
= 0;
3088 map
->map_state
= map_state
;
3092 static int parse_raid_devices(struct intel_super
*super
)
3095 struct imsm_dev
*dev_new
;
3096 size_t len
, len_migr
;
3098 size_t space_needed
= 0;
3099 struct imsm_super
*mpb
= super
->anchor
;
3101 for (i
= 0; i
< super
->anchor
->num_raid_devs
; i
++) {
3102 struct imsm_dev
*dev_iter
= __get_imsm_dev(super
->anchor
, i
);
3103 struct intel_dev
*dv
;
3105 len
= sizeof_imsm_dev(dev_iter
, 0);
3106 len_migr
= sizeof_imsm_dev(dev_iter
, 1);
3108 space_needed
+= len_migr
- len
;
3110 dv
= malloc(sizeof(*dv
));
3113 if (max_len
< len_migr
)
3115 if (max_len
> len_migr
)
3116 space_needed
+= max_len
- len_migr
;
3117 dev_new
= malloc(max_len
);
3122 imsm_copy_dev(dev_new
, dev_iter
);
3125 dv
->next
= super
->devlist
;
3126 super
->devlist
= dv
;
3129 /* ensure that super->buf is large enough when all raid devices
3132 if (__le32_to_cpu(mpb
->mpb_size
) + space_needed
> super
->len
) {
3135 len
= ROUND_UP(__le32_to_cpu(mpb
->mpb_size
) + space_needed
, 512);
3136 if (posix_memalign(&buf
, 512, len
) != 0)
3139 memcpy(buf
, super
->buf
, super
->len
);
3140 memset(buf
+ super
->len
, 0, len
- super
->len
);
3149 /* retrieve a pointer to the bbm log which starts after all raid devices */
3150 struct bbm_log
*__get_imsm_bbm_log(struct imsm_super
*mpb
)
3154 if (__le32_to_cpu(mpb
->bbm_log_size
)) {
3156 ptr
+= mpb
->mpb_size
- __le32_to_cpu(mpb
->bbm_log_size
);
3162 /*******************************************************************************
3163 * Function: check_mpb_migr_compatibility
3164 * Description: Function checks for unsupported migration features:
3165 * - migration optimization area (pba_of_lba0)
3166 * - descending reshape (ascending_migr)
3168 * super : imsm metadata information
3170 * 0 : migration is compatible
3171 * -1 : migration is not compatible
3172 ******************************************************************************/
3173 int check_mpb_migr_compatibility(struct intel_super
*super
)
3175 struct imsm_map
*map0
, *map1
;
3176 struct migr_record
*migr_rec
= super
->migr_rec
;
3179 for (i
= 0; i
< super
->anchor
->num_raid_devs
; i
++) {
3180 struct imsm_dev
*dev_iter
= __get_imsm_dev(super
->anchor
, i
);
3183 dev_iter
->vol
.migr_state
== 1 &&
3184 dev_iter
->vol
.migr_type
== MIGR_GEN_MIGR
) {
3185 /* This device is migrating */
3186 map0
= get_imsm_map(dev_iter
, 0);
3187 map1
= get_imsm_map(dev_iter
, 1);
3188 if (map0
->pba_of_lba0
!= map1
->pba_of_lba0
)
3189 /* migration optimization area was used */
3191 if (migr_rec
->ascending_migr
== 0
3192 && migr_rec
->dest_depth_per_unit
> 0)
3193 /* descending reshape not supported yet */
3200 static void __free_imsm(struct intel_super
*super
, int free_disks
);
3202 /* load_imsm_mpb - read matrix metadata
3203 * allocates super->mpb to be freed by free_imsm
3205 static int load_imsm_mpb(int fd
, struct intel_super
*super
, char *devname
)
3207 unsigned long long dsize
;
3208 unsigned long long sectors
;
3210 struct imsm_super
*anchor
;
3213 get_dev_size(fd
, NULL
, &dsize
);
3217 Name
": %s: device to small for imsm\n",
3222 if (lseek64(fd
, dsize
- (512 * 2), SEEK_SET
) < 0) {
3224 fprintf(stderr
, Name
3225 ": Cannot seek to anchor block on %s: %s\n",
3226 devname
, strerror(errno
));
3230 if (posix_memalign((void**)&anchor
, 512, 512) != 0) {
3233 Name
": Failed to allocate imsm anchor buffer"
3234 " on %s\n", devname
);
3237 if (read(fd
, anchor
, 512) != 512) {
3240 Name
": Cannot read anchor block on %s: %s\n",
3241 devname
, strerror(errno
));
3246 if (strncmp((char *) anchor
->sig
, MPB_SIGNATURE
, MPB_SIG_LEN
) != 0) {
3249 Name
": no IMSM anchor on %s\n", devname
);
3254 __free_imsm(super
, 0);
3255 /* reload capability and hba */
3257 /* capability and hba must be updated with new super allocation */
3258 find_intel_hba_capability(fd
, super
, devname
);
3259 super
->len
= ROUND_UP(anchor
->mpb_size
, 512);
3260 if (posix_memalign(&super
->buf
, 512, super
->len
) != 0) {
3263 Name
": unable to allocate %zu byte mpb buffer\n",
3268 memcpy(super
->buf
, anchor
, 512);
3270 sectors
= mpb_sectors(anchor
) - 1;
3273 if (posix_memalign(&super
->migr_rec_buf
, 512, 512) != 0) {
3274 fprintf(stderr
, Name
3275 ": %s could not allocate migr_rec buffer\n", __func__
);
3281 check_sum
= __gen_imsm_checksum(super
->anchor
);
3282 if (check_sum
!= __le32_to_cpu(super
->anchor
->check_sum
)) {
3285 Name
": IMSM checksum %x != %x on %s\n",
3287 __le32_to_cpu(super
->anchor
->check_sum
),
3295 /* read the extended mpb */
3296 if (lseek64(fd
, dsize
- (512 * (2 + sectors
)), SEEK_SET
) < 0) {
3299 Name
": Cannot seek to extended mpb on %s: %s\n",
3300 devname
, strerror(errno
));
3304 if ((unsigned)read(fd
, super
->buf
+ 512, super
->len
- 512) != super
->len
- 512) {
3307 Name
": Cannot read extended mpb on %s: %s\n",
3308 devname
, strerror(errno
));
3312 check_sum
= __gen_imsm_checksum(super
->anchor
);
3313 if (check_sum
!= __le32_to_cpu(super
->anchor
->check_sum
)) {
3316 Name
": IMSM checksum %x != %x on %s\n",
3317 check_sum
, __le32_to_cpu(super
->anchor
->check_sum
),
3322 /* FIXME the BBM log is disk specific so we cannot use this global
3323 * buffer for all disks. Ok for now since we only look at the global
3324 * bbm_log_size parameter to gate assembly
3326 super
->bbm_log
= __get_imsm_bbm_log(super
->anchor
);
3331 static int read_imsm_migr_rec(int fd
, struct intel_super
*super
);
3334 load_and_parse_mpb(int fd
, struct intel_super
*super
, char *devname
, int keep_fd
)
3338 err
= load_imsm_mpb(fd
, super
, devname
);
3341 err
= load_imsm_disk(fd
, super
, devname
, keep_fd
);
3344 err
= parse_raid_devices(super
);
3349 static void __free_imsm_disk(struct dl
*d
)
3361 static void free_imsm_disks(struct intel_super
*super
)
3365 while (super
->disks
) {
3367 super
->disks
= d
->next
;
3368 __free_imsm_disk(d
);
3370 while (super
->disk_mgmt_list
) {
3371 d
= super
->disk_mgmt_list
;
3372 super
->disk_mgmt_list
= d
->next
;
3373 __free_imsm_disk(d
);
3375 while (super
->missing
) {
3377 super
->missing
= d
->next
;
3378 __free_imsm_disk(d
);
3383 /* free all the pieces hanging off of a super pointer */
3384 static void __free_imsm(struct intel_super
*super
, int free_disks
)
3386 struct intel_hba
*elem
, *next
;
3392 /* unlink capability description */
3394 if (super
->migr_rec_buf
) {
3395 free(super
->migr_rec_buf
);
3396 super
->migr_rec_buf
= NULL
;
3399 free_imsm_disks(super
);
3400 free_devlist(super
);
3404 free((void *)elem
->path
);
3412 static void free_imsm(struct intel_super
*super
)
3414 __free_imsm(super
, 1);
3418 static void free_super_imsm(struct supertype
*st
)
3420 struct intel_super
*super
= st
->sb
;
3429 static struct intel_super
*alloc_super(void)
3431 struct intel_super
*super
= malloc(sizeof(*super
));
3434 memset(super
, 0, sizeof(*super
));
3435 super
->current_vol
= -1;
3436 super
->create_offset
= ~((__u32
) 0);
3442 * find and allocate hba and OROM/EFI based on valid fd of RAID component device
3444 static int find_intel_hba_capability(int fd
, struct intel_super
*super
, char *devname
)
3446 struct sys_dev
*hba_name
;
3449 if ((fd
< 0) || check_env("IMSM_NO_PLATFORM")) {
3454 hba_name
= find_disk_attached_hba(fd
, NULL
);
3458 Name
": %s is not attached to Intel(R) RAID controller.\n",
3462 rv
= attach_hba_to_super(super
, hba_name
);
3465 struct intel_hba
*hba
= super
->hba
;
3467 fprintf(stderr
, Name
": %s is attached to Intel(R) %s RAID "
3468 "controller (%s),\n"
3469 " but the container is assigned to Intel(R) "
3470 "%s RAID controller (",
3473 hba_name
->pci_id
? : "Err!",
3474 get_sys_dev_type(hba_name
->type
));
3477 fprintf(stderr
, "%s", hba
->pci_id
? : "Err!");
3479 fprintf(stderr
, ", ");
3483 fprintf(stderr
, ").\n"
3484 " Mixing devices attached to different controllers "
3485 "is not allowed.\n");
3487 free_sys_dev(&hba_name
);
3490 super
->orom
= find_imsm_capability(hba_name
->type
);
3491 free_sys_dev(&hba_name
);
3497 /* find_missing - helper routine for load_super_imsm_all that identifies
3498 * disks that have disappeared from the system. This routine relies on
3499 * the mpb being uptodate, which it is at load time.
3501 static int find_missing(struct intel_super
*super
)
3504 struct imsm_super
*mpb
= super
->anchor
;
3506 struct imsm_disk
*disk
;
3508 for (i
= 0; i
< mpb
->num_disks
; i
++) {
3509 disk
= __get_imsm_disk(mpb
, i
);
3510 dl
= serial_to_dl(disk
->serial
, super
);
3514 dl
= malloc(sizeof(*dl
));
3520 dl
->devname
= strdup("missing");
3522 serialcpy(dl
->serial
, disk
->serial
);
3525 dl
->next
= super
->missing
;
3526 super
->missing
= dl
;
3533 static struct intel_disk
*disk_list_get(__u8
*serial
, struct intel_disk
*disk_list
)
3535 struct intel_disk
*idisk
= disk_list
;
3538 if (serialcmp(idisk
->disk
.serial
, serial
) == 0)
3540 idisk
= idisk
->next
;
3546 static int __prep_thunderdome(struct intel_super
**table
, int tbl_size
,
3547 struct intel_super
*super
,
3548 struct intel_disk
**disk_list
)
3550 struct imsm_disk
*d
= &super
->disks
->disk
;
3551 struct imsm_super
*mpb
= super
->anchor
;
3554 for (i
= 0; i
< tbl_size
; i
++) {
3555 struct imsm_super
*tbl_mpb
= table
[i
]->anchor
;
3556 struct imsm_disk
*tbl_d
= &table
[i
]->disks
->disk
;
3558 if (tbl_mpb
->family_num
== mpb
->family_num
) {
3559 if (tbl_mpb
->check_sum
== mpb
->check_sum
) {
3560 dprintf("%s: mpb from %d:%d matches %d:%d\n",
3561 __func__
, super
->disks
->major
,
3562 super
->disks
->minor
,
3563 table
[i
]->disks
->major
,
3564 table
[i
]->disks
->minor
);
3568 if (((is_configured(d
) && !is_configured(tbl_d
)) ||
3569 is_configured(d
) == is_configured(tbl_d
)) &&
3570 tbl_mpb
->generation_num
< mpb
->generation_num
) {
3571 /* current version of the mpb is a
3572 * better candidate than the one in
3573 * super_table, but copy over "cross
3574 * generational" status
3576 struct intel_disk
*idisk
;
3578 dprintf("%s: mpb from %d:%d replaces %d:%d\n",
3579 __func__
, super
->disks
->major
,
3580 super
->disks
->minor
,
3581 table
[i
]->disks
->major
,
3582 table
[i
]->disks
->minor
);
3584 idisk
= disk_list_get(tbl_d
->serial
, *disk_list
);
3585 if (idisk
&& is_failed(&idisk
->disk
))
3586 tbl_d
->status
|= FAILED_DISK
;
3589 struct intel_disk
*idisk
;
3590 struct imsm_disk
*disk
;
3592 /* tbl_mpb is more up to date, but copy
3593 * over cross generational status before
3596 disk
= __serial_to_disk(d
->serial
, mpb
, NULL
);
3597 if (disk
&& is_failed(disk
))
3598 d
->status
|= FAILED_DISK
;
3600 idisk
= disk_list_get(d
->serial
, *disk_list
);
3603 if (disk
&& is_configured(disk
))
3604 idisk
->disk
.status
|= CONFIGURED_DISK
;
3607 dprintf("%s: mpb from %d:%d prefer %d:%d\n",
3608 __func__
, super
->disks
->major
,
3609 super
->disks
->minor
,
3610 table
[i
]->disks
->major
,
3611 table
[i
]->disks
->minor
);
3619 table
[tbl_size
++] = super
;
3623 /* update/extend the merged list of imsm_disk records */
3624 for (j
= 0; j
< mpb
->num_disks
; j
++) {
3625 struct imsm_disk
*disk
= __get_imsm_disk(mpb
, j
);
3626 struct intel_disk
*idisk
;
3628 idisk
= disk_list_get(disk
->serial
, *disk_list
);
3630 idisk
->disk
.status
|= disk
->status
;
3631 if (is_configured(&idisk
->disk
) ||
3632 is_failed(&idisk
->disk
))
3633 idisk
->disk
.status
&= ~(SPARE_DISK
);
3635 idisk
= calloc(1, sizeof(*idisk
));
3638 idisk
->owner
= IMSM_UNKNOWN_OWNER
;
3639 idisk
->disk
= *disk
;
3640 idisk
->next
= *disk_list
;
3644 if (serialcmp(idisk
->disk
.serial
, d
->serial
) == 0)
3651 static struct intel_super
*
3652 validate_members(struct intel_super
*super
, struct intel_disk
*disk_list
,
3655 struct imsm_super
*mpb
= super
->anchor
;
3659 for (i
= 0; i
< mpb
->num_disks
; i
++) {
3660 struct imsm_disk
*disk
= __get_imsm_disk(mpb
, i
);
3661 struct intel_disk
*idisk
;
3663 idisk
= disk_list_get(disk
->serial
, disk_list
);
3665 if (idisk
->owner
== owner
||
3666 idisk
->owner
== IMSM_UNKNOWN_OWNER
)
3669 dprintf("%s: '%.16s' owner %d != %d\n",
3670 __func__
, disk
->serial
, idisk
->owner
,
3673 dprintf("%s: unknown disk %x [%d]: %.16s\n",
3674 __func__
, __le32_to_cpu(mpb
->family_num
), i
,
3680 if (ok_count
== mpb
->num_disks
)
3685 static void show_conflicts(__u32 family_num
, struct intel_super
*super_list
)
3687 struct intel_super
*s
;
3689 for (s
= super_list
; s
; s
= s
->next
) {
3690 if (family_num
!= s
->anchor
->family_num
)
3692 fprintf(stderr
, "Conflict, offlining family %#x on '%s'\n",
3693 __le32_to_cpu(family_num
), s
->disks
->devname
);
3697 static struct intel_super
*
3698 imsm_thunderdome(struct intel_super
**super_list
, int len
)
3700 struct intel_super
*super_table
[len
];
3701 struct intel_disk
*disk_list
= NULL
;
3702 struct intel_super
*champion
, *spare
;
3703 struct intel_super
*s
, **del
;
3708 memset(super_table
, 0, sizeof(super_table
));
3709 for (s
= *super_list
; s
; s
= s
->next
)
3710 tbl_size
= __prep_thunderdome(super_table
, tbl_size
, s
, &disk_list
);
3712 for (i
= 0; i
< tbl_size
; i
++) {
3713 struct imsm_disk
*d
;
3714 struct intel_disk
*idisk
;
3715 struct imsm_super
*mpb
= super_table
[i
]->anchor
;
3718 d
= &s
->disks
->disk
;
3720 /* 'd' must appear in merged disk list for its
3721 * configuration to be valid
3723 idisk
= disk_list_get(d
->serial
, disk_list
);
3724 if (idisk
&& idisk
->owner
== i
)
3725 s
= validate_members(s
, disk_list
, i
);
3730 dprintf("%s: marking family: %#x from %d:%d offline\n",
3731 __func__
, mpb
->family_num
,
3732 super_table
[i
]->disks
->major
,
3733 super_table
[i
]->disks
->minor
);
3737 /* This is where the mdadm implementation differs from the Windows
3738 * driver which has no strict concept of a container. We can only
3739 * assemble one family from a container, so when returning a prodigal
3740 * array member to this system the code will not be able to disambiguate
3741 * the container contents that should be assembled ("foreign" versus
3742 * "local"). It requires user intervention to set the orig_family_num
3743 * to a new value to establish a new container. The Windows driver in
3744 * this situation fixes up the volume name in place and manages the
3745 * foreign array as an independent entity.
3750 for (i
= 0; i
< tbl_size
; i
++) {
3751 struct intel_super
*tbl_ent
= super_table
[i
];
3757 if (tbl_ent
->anchor
->num_raid_devs
== 0) {
3762 if (s
&& !is_spare
) {
3763 show_conflicts(tbl_ent
->anchor
->family_num
, *super_list
);
3765 } else if (!s
&& !is_spare
)
3778 fprintf(stderr
, "Chose family %#x on '%s', "
3779 "assemble conflicts to new container with '--update=uuid'\n",
3780 __le32_to_cpu(s
->anchor
->family_num
), s
->disks
->devname
);
3782 /* collect all dl's onto 'champion', and update them to
3783 * champion's version of the status
3785 for (s
= *super_list
; s
; s
= s
->next
) {
3786 struct imsm_super
*mpb
= champion
->anchor
;
3787 struct dl
*dl
= s
->disks
;
3792 for (i
= 0; i
< mpb
->num_disks
; i
++) {
3793 struct imsm_disk
*disk
;
3795 disk
= __serial_to_disk(dl
->serial
, mpb
, &dl
->index
);
3798 /* only set index on disks that are a member of
3799 * a populated contianer, i.e. one with
3802 if (is_failed(&dl
->disk
))
3804 else if (is_spare(&dl
->disk
))
3810 if (i
>= mpb
->num_disks
) {
3811 struct intel_disk
*idisk
;
3813 idisk
= disk_list_get(dl
->serial
, disk_list
);
3814 if (idisk
&& is_spare(&idisk
->disk
) &&
3815 !is_failed(&idisk
->disk
) && !is_configured(&idisk
->disk
))
3823 dl
->next
= champion
->disks
;
3824 champion
->disks
= dl
;
3828 /* delete 'champion' from super_list */
3829 for (del
= super_list
; *del
; ) {
3830 if (*del
== champion
) {
3831 *del
= (*del
)->next
;
3834 del
= &(*del
)->next
;
3836 champion
->next
= NULL
;
3840 struct intel_disk
*idisk
= disk_list
;
3842 disk_list
= disk_list
->next
;
3849 static int load_super_imsm_all(struct supertype
*st
, int fd
, void **sbp
,
3853 struct intel_super
*super_list
= NULL
;
3854 struct intel_super
*super
= NULL
;
3855 int devnum
= fd2devnum(fd
);
3861 /* check if 'fd' an opened container */
3862 sra
= sysfs_read(fd
, 0, GET_LEVEL
|GET_VERSION
|GET_DEVS
|GET_STATE
);
3866 if (sra
->array
.major_version
!= -1 ||
3867 sra
->array
.minor_version
!= -2 ||
3868 strcmp(sra
->text_version
, "imsm") != 0) {
3873 for (sd
= sra
->devs
, i
= 0; sd
; sd
= sd
->next
, i
++) {
3874 struct intel_super
*s
= alloc_super();
3882 s
->next
= super_list
;
3886 sprintf(nm
, "%d:%d", sd
->disk
.major
, sd
->disk
.minor
);
3887 dfd
= dev_open(nm
, O_RDWR
);
3891 rv
= find_intel_hba_capability(dfd
, s
, devname
);
3892 /* no orom/efi or non-intel hba of the disk */
3896 err
= load_and_parse_mpb(dfd
, s
, NULL
, 1);
3898 /* retry the load if we might have raced against mdmon */
3899 if (err
== 3 && mdmon_running(devnum
))
3900 for (retry
= 0; retry
< 3; retry
++) {
3902 err
= load_and_parse_mpb(dfd
, s
, NULL
, 1);
3910 /* all mpbs enter, maybe one leaves */
3911 super
= imsm_thunderdome(&super_list
, i
);
3917 if (find_missing(super
) != 0) {
3923 /* load migration record */
3924 err
= load_imsm_migr_rec(super
, NULL
);
3930 /* Check migration compatibility */
3931 if (check_mpb_migr_compatibility(super
) != 0) {
3932 fprintf(stderr
, Name
": Unsupported migration detected");
3934 fprintf(stderr
, " on %s\n", devname
);
3936 fprintf(stderr
, " (IMSM).\n");
3945 while (super_list
) {
3946 struct intel_super
*s
= super_list
;
3948 super_list
= super_list
->next
;
3957 st
->container_dev
= devnum
;
3958 if (err
== 0 && st
->ss
== NULL
) {
3959 st
->ss
= &super_imsm
;
3960 st
->minor_version
= 0;
3961 st
->max_devs
= IMSM_MAX_DEVICES
;
3966 static int load_container_imsm(struct supertype
*st
, int fd
, char *devname
)
3968 return load_super_imsm_all(st
, fd
, &st
->sb
, devname
);
3972 static int load_super_imsm(struct supertype
*st
, int fd
, char *devname
)
3974 struct intel_super
*super
;
3977 if (test_partition(fd
))
3978 /* IMSM not allowed on partitions */
3981 free_super_imsm(st
);
3983 super
= alloc_super();
3986 Name
": malloc of %zu failed.\n",
3990 /* Load hba and capabilities if they exist.
3991 * But do not preclude loading metadata in case capabilities or hba are
3992 * non-compliant and ignore_hw_compat is set.
3994 rv
= find_intel_hba_capability(fd
, super
, devname
);
3995 /* no orom/efi or non-intel hba of the disk */
3996 if ((rv
!= 0) && (st
->ignore_hw_compat
== 0)) {
3999 Name
": No OROM/EFI properties for %s\n", devname
);
4003 rv
= load_and_parse_mpb(fd
, super
, devname
, 0);
4008 Name
": Failed to load all information "
4009 "sections on %s\n", devname
);
4015 if (st
->ss
== NULL
) {
4016 st
->ss
= &super_imsm
;
4017 st
->minor_version
= 0;
4018 st
->max_devs
= IMSM_MAX_DEVICES
;
4021 /* load migration record */
4022 if (load_imsm_migr_rec(super
, NULL
) == 0) {
4023 /* Check for unsupported migration features */
4024 if (check_mpb_migr_compatibility(super
) != 0) {
4026 Name
": Unsupported migration detected");
4028 fprintf(stderr
, " on %s\n", devname
);
4030 fprintf(stderr
, " (IMSM).\n");
4038 static __u16
info_to_blocks_per_strip(mdu_array_info_t
*info
)
4040 if (info
->level
== 1)
4042 return info
->chunk_size
>> 9;
4045 static __u32
info_to_num_data_stripes(mdu_array_info_t
*info
, int num_domains
)
4049 num_stripes
= (info
->size
* 2) / info_to_blocks_per_strip(info
);
4050 num_stripes
/= num_domains
;
4055 static __u32
info_to_blocks_per_member(mdu_array_info_t
*info
)
4057 if (info
->level
== 1)
4058 return info
->size
* 2;
4060 return (info
->size
* 2) & ~(info_to_blocks_per_strip(info
) - 1);
4063 static void imsm_update_version_info(struct intel_super
*super
)
4065 /* update the version and attributes */
4066 struct imsm_super
*mpb
= super
->anchor
;
4068 struct imsm_dev
*dev
;
4069 struct imsm_map
*map
;
4072 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
4073 dev
= get_imsm_dev(super
, i
);
4074 map
= get_imsm_map(dev
, 0);
4075 if (__le32_to_cpu(dev
->size_high
) > 0)
4076 mpb
->attributes
|= MPB_ATTRIB_2TB
;
4078 /* FIXME detect when an array spans a port multiplier */
4080 mpb
->attributes
|= MPB_ATTRIB_PM
;
4083 if (mpb
->num_raid_devs
> 1 ||
4084 mpb
->attributes
!= MPB_ATTRIB_CHECKSUM_VERIFY
) {
4085 version
= MPB_VERSION_ATTRIBS
;
4086 switch (get_imsm_raid_level(map
)) {
4087 case 0: mpb
->attributes
|= MPB_ATTRIB_RAID0
; break;
4088 case 1: mpb
->attributes
|= MPB_ATTRIB_RAID1
; break;
4089 case 10: mpb
->attributes
|= MPB_ATTRIB_RAID10
; break;
4090 case 5: mpb
->attributes
|= MPB_ATTRIB_RAID5
; break;
4093 if (map
->num_members
>= 5)
4094 version
= MPB_VERSION_5OR6_DISK_ARRAY
;
4095 else if (dev
->status
== DEV_CLONE_N_GO
)
4096 version
= MPB_VERSION_CNG
;
4097 else if (get_imsm_raid_level(map
) == 5)
4098 version
= MPB_VERSION_RAID5
;
4099 else if (map
->num_members
>= 3)
4100 version
= MPB_VERSION_3OR4_DISK_ARRAY
;
4101 else if (get_imsm_raid_level(map
) == 1)
4102 version
= MPB_VERSION_RAID1
;
4104 version
= MPB_VERSION_RAID0
;
4106 strcpy(((char *) mpb
->sig
) + strlen(MPB_SIGNATURE
), version
);
4110 static int check_name(struct intel_super
*super
, char *name
, int quiet
)
4112 struct imsm_super
*mpb
= super
->anchor
;
4113 char *reason
= NULL
;
4116 if (strlen(name
) > MAX_RAID_SERIAL_LEN
)
4117 reason
= "must be 16 characters or less";
4119 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
4120 struct imsm_dev
*dev
= get_imsm_dev(super
, i
);
4122 if (strncmp((char *) dev
->volume
, name
, MAX_RAID_SERIAL_LEN
) == 0) {
4123 reason
= "already exists";
4128 if (reason
&& !quiet
)
4129 fprintf(stderr
, Name
": imsm volume name %s\n", reason
);
4134 static int init_super_imsm_volume(struct supertype
*st
, mdu_array_info_t
*info
,
4135 unsigned long long size
, char *name
,
4136 char *homehost
, int *uuid
)
4138 /* We are creating a volume inside a pre-existing container.
4139 * so st->sb is already set.
4141 struct intel_super
*super
= st
->sb
;
4142 struct imsm_super
*mpb
= super
->anchor
;
4143 struct intel_dev
*dv
;
4144 struct imsm_dev
*dev
;
4145 struct imsm_vol
*vol
;
4146 struct imsm_map
*map
;
4147 int idx
= mpb
->num_raid_devs
;
4149 unsigned long long array_blocks
;
4150 size_t size_old
, size_new
;
4151 __u32 num_data_stripes
;
4153 if (super
->orom
&& mpb
->num_raid_devs
>= super
->orom
->vpa
) {
4154 fprintf(stderr
, Name
": This imsm-container already has the "
4155 "maximum of %d volumes\n", super
->orom
->vpa
);
4159 /* ensure the mpb is large enough for the new data */
4160 size_old
= __le32_to_cpu(mpb
->mpb_size
);
4161 size_new
= disks_to_mpb_size(info
->nr_disks
);
4162 if (size_new
> size_old
) {
4164 size_t size_round
= ROUND_UP(size_new
, 512);
4166 if (posix_memalign(&mpb_new
, 512, size_round
) != 0) {
4167 fprintf(stderr
, Name
": could not allocate new mpb\n");
4170 if (posix_memalign(&super
->migr_rec_buf
, 512, 512) != 0) {
4171 fprintf(stderr
, Name
4172 ": %s could not allocate migr_rec buffer\n",
4178 memcpy(mpb_new
, mpb
, size_old
);
4181 super
->anchor
= mpb_new
;
4182 mpb
->mpb_size
= __cpu_to_le32(size_new
);
4183 memset(mpb_new
+ size_old
, 0, size_round
- size_old
);
4185 super
->current_vol
= idx
;
4187 /* handle 'failed_disks' by either:
4188 * a) create dummy disk entries in the table if this the first
4189 * volume in the array. We add them here as this is the only
4190 * opportunity to add them. add_to_super_imsm_volume()
4191 * handles the non-failed disks and continues incrementing
4193 * b) validate that 'failed_disks' matches the current number
4194 * of missing disks if the container is populated
4196 if (super
->current_vol
== 0) {
4198 for (i
= 0; i
< info
->failed_disks
; i
++) {
4199 struct imsm_disk
*disk
;
4202 disk
= __get_imsm_disk(mpb
, i
);
4203 disk
->status
= CONFIGURED_DISK
| FAILED_DISK
;
4204 disk
->scsi_id
= __cpu_to_le32(~(__u32
)0);
4205 snprintf((char *) disk
->serial
, MAX_RAID_SERIAL_LEN
,
4208 find_missing(super
);
4213 for (d
= super
->missing
; d
; d
= d
->next
)
4215 if (info
->failed_disks
> missing
) {
4216 fprintf(stderr
, Name
": unable to add 'missing' disk to container\n");
4221 if (!check_name(super
, name
, 0))
4223 dv
= malloc(sizeof(*dv
));
4225 fprintf(stderr
, Name
": failed to allocate device list entry\n");
4228 dev
= calloc(1, sizeof(*dev
) + sizeof(__u32
) * (info
->raid_disks
- 1));
4231 fprintf(stderr
, Name
": could not allocate raid device\n");
4235 strncpy((char *) dev
->volume
, name
, MAX_RAID_SERIAL_LEN
);
4236 if (info
->level
== 1)
4237 array_blocks
= info_to_blocks_per_member(info
);
4239 array_blocks
= calc_array_size(info
->level
, info
->raid_disks
,
4240 info
->layout
, info
->chunk_size
,
4242 /* round array size down to closest MB */
4243 array_blocks
= (array_blocks
>> SECT_PER_MB_SHIFT
) << SECT_PER_MB_SHIFT
;
4245 dev
->size_low
= __cpu_to_le32((__u32
) array_blocks
);
4246 dev
->size_high
= __cpu_to_le32((__u32
) (array_blocks
>> 32));
4247 dev
->status
= (DEV_READ_COALESCING
| DEV_WRITE_COALESCING
);
4249 vol
->migr_state
= 0;
4250 set_migr_type(dev
, MIGR_INIT
);
4251 vol
->dirty
= !info
->state
;
4252 vol
->curr_migr_unit
= 0;
4253 map
= get_imsm_map(dev
, 0);
4254 map
->pba_of_lba0
= __cpu_to_le32(super
->create_offset
);
4255 map
->blocks_per_member
= __cpu_to_le32(info_to_blocks_per_member(info
));
4256 map
->blocks_per_strip
= __cpu_to_le16(info_to_blocks_per_strip(info
));
4257 map
->failed_disk_num
= ~0;
4258 map
->map_state
= info
->failed_disks
? IMSM_T_STATE_DEGRADED
: IMSM_T_STATE_NORMAL
;
4261 if (info
->level
== 1 && info
->raid_disks
> 2) {
4264 fprintf(stderr
, Name
": imsm does not support more than 2 disks"
4265 "in a raid1 volume\n");
4269 map
->raid_level
= info
->level
;
4270 if (info
->level
== 10) {
4271 map
->raid_level
= 1;
4272 map
->num_domains
= info
->raid_disks
/ 2;
4273 } else if (info
->level
== 1)
4274 map
->num_domains
= info
->raid_disks
;
4276 map
->num_domains
= 1;
4278 num_data_stripes
= info_to_num_data_stripes(info
, map
->num_domains
);
4279 map
->num_data_stripes
= __cpu_to_le32(num_data_stripes
);
4281 map
->num_members
= info
->raid_disks
;
4282 for (i
= 0; i
< map
->num_members
; i
++) {
4283 /* initialized in add_to_super */
4284 set_imsm_ord_tbl_ent(map
, i
, IMSM_ORD_REBUILD
);
4286 mpb
->num_raid_devs
++;
4289 dv
->index
= super
->current_vol
;
4290 dv
->next
= super
->devlist
;
4291 super
->devlist
= dv
;
4293 imsm_update_version_info(super
);
4298 static int init_super_imsm(struct supertype
*st
, mdu_array_info_t
*info
,
4299 unsigned long long size
, char *name
,
4300 char *homehost
, int *uuid
)
4302 /* This is primarily called by Create when creating a new array.
4303 * We will then get add_to_super called for each component, and then
4304 * write_init_super called to write it out to each device.
4305 * For IMSM, Create can create on fresh devices or on a pre-existing
4307 * To create on a pre-existing array a different method will be called.
4308 * This one is just for fresh drives.
4310 struct intel_super
*super
;
4311 struct imsm_super
*mpb
;
4316 return init_super_imsm_volume(st
, info
, size
, name
, homehost
, uuid
);
4319 mpb_size
= disks_to_mpb_size(info
->nr_disks
);
4323 super
= alloc_super();
4324 if (super
&& posix_memalign(&super
->buf
, 512, mpb_size
) != 0) {
4329 fprintf(stderr
, Name
4330 ": %s could not allocate superblock\n", __func__
);
4333 if (posix_memalign(&super
->migr_rec_buf
, 512, 512) != 0) {
4334 fprintf(stderr
, Name
4335 ": %s could not allocate migr_rec buffer\n", __func__
);
4340 memset(super
->buf
, 0, mpb_size
);
4342 mpb
->mpb_size
= __cpu_to_le32(mpb_size
);
4346 /* zeroing superblock */
4350 mpb
->attributes
= MPB_ATTRIB_CHECKSUM_VERIFY
;
4352 version
= (char *) mpb
->sig
;
4353 strcpy(version
, MPB_SIGNATURE
);
4354 version
+= strlen(MPB_SIGNATURE
);
4355 strcpy(version
, MPB_VERSION_RAID0
);
4361 static int add_to_super_imsm_volume(struct supertype
*st
, mdu_disk_info_t
*dk
,
4362 int fd
, char *devname
)
4364 struct intel_super
*super
= st
->sb
;
4365 struct imsm_super
*mpb
= super
->anchor
;
4366 struct imsm_disk
*_disk
;
4367 struct imsm_dev
*dev
;
4368 struct imsm_map
*map
;
4372 dev
= get_imsm_dev(super
, super
->current_vol
);
4373 map
= get_imsm_map(dev
, 0);
4375 if (! (dk
->state
& (1<<MD_DISK_SYNC
))) {
4376 fprintf(stderr
, Name
": %s: Cannot add spare devices to IMSM volume\n",
4382 /* we're doing autolayout so grab the pre-marked (in
4383 * validate_geometry) raid_disk
4385 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
4386 if (dl
->raiddisk
== dk
->raid_disk
)
4389 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
4390 if (dl
->major
== dk
->major
&&
4391 dl
->minor
== dk
->minor
)
4396 fprintf(stderr
, Name
": %s is not a member of the same container\n", devname
);
4400 /* add a pristine spare to the metadata */
4401 if (dl
->index
< 0) {
4402 dl
->index
= super
->anchor
->num_disks
;
4403 super
->anchor
->num_disks
++;
4405 /* Check the device has not already been added */
4406 slot
= get_imsm_disk_slot(map
, dl
->index
);
4408 (get_imsm_ord_tbl_ent(dev
, slot
, -1) & IMSM_ORD_REBUILD
) == 0) {
4409 fprintf(stderr
, Name
": %s has been included in this array twice\n",
4413 set_imsm_ord_tbl_ent(map
, dk
->raid_disk
, dl
->index
);
4414 dl
->disk
.status
= CONFIGURED_DISK
;
4416 /* update size of 'missing' disks to be at least as large as the
4417 * largest acitve member (we only have dummy missing disks when
4418 * creating the first volume)
4420 if (super
->current_vol
== 0) {
4421 for (df
= super
->missing
; df
; df
= df
->next
) {
4422 if (dl
->disk
.total_blocks
> df
->disk
.total_blocks
)
4423 df
->disk
.total_blocks
= dl
->disk
.total_blocks
;
4424 _disk
= __get_imsm_disk(mpb
, df
->index
);
4429 /* refresh unset/failed slots to point to valid 'missing' entries */
4430 for (df
= super
->missing
; df
; df
= df
->next
)
4431 for (slot
= 0; slot
< mpb
->num_disks
; slot
++) {
4432 __u32 ord
= get_imsm_ord_tbl_ent(dev
, slot
, -1);
4434 if ((ord
& IMSM_ORD_REBUILD
) == 0)
4436 set_imsm_ord_tbl_ent(map
, slot
, df
->index
| IMSM_ORD_REBUILD
);
4437 dprintf("set slot:%d to missing disk:%d\n", slot
, df
->index
);
4441 /* if we are creating the first raid device update the family number */
4442 if (super
->current_vol
== 0) {
4444 struct imsm_dev
*_dev
= __get_imsm_dev(mpb
, 0);
4446 _disk
= __get_imsm_disk(mpb
, dl
->index
);
4447 if (!_dev
|| !_disk
) {
4448 fprintf(stderr
, Name
": BUG mpb setup error\n");
4454 sum
+= __gen_imsm_checksum(mpb
);
4455 mpb
->family_num
= __cpu_to_le32(sum
);
4456 mpb
->orig_family_num
= mpb
->family_num
;
4458 super
->current_disk
= dl
;
4463 * Function marks disk as spare and restores disk serial
4464 * in case it was previously marked as failed by takeover operation
4466 * -1 : critical error
4467 * 0 : disk is marked as spare but serial is not set
4470 int mark_spare(struct dl
*disk
)
4472 __u8 serial
[MAX_RAID_SERIAL_LEN
];
4479 if (!imsm_read_serial(disk
->fd
, NULL
, serial
)) {
4480 /* Restore disk serial number, because takeover marks disk
4481 * as failed and adds to serial ':0' before it becomes
4484 serialcpy(disk
->serial
, serial
);
4485 serialcpy(disk
->disk
.serial
, serial
);
4488 disk
->disk
.status
= SPARE_DISK
;
4494 static int add_to_super_imsm(struct supertype
*st
, mdu_disk_info_t
*dk
,
4495 int fd
, char *devname
)
4497 struct intel_super
*super
= st
->sb
;
4499 unsigned long long size
;
4504 /* If we are on an RAID enabled platform check that the disk is
4505 * attached to the raid controller.
4506 * We do not need to test disks attachment for container based additions,
4507 * they shall be already tested when container was created/assembled.
4509 rv
= find_intel_hba_capability(fd
, super
, devname
);
4510 /* no orom/efi or non-intel hba of the disk */
4512 dprintf("capability: %p fd: %d ret: %d\n",
4513 super
->orom
, fd
, rv
);
4517 if (super
->current_vol
>= 0)
4518 return add_to_super_imsm_volume(st
, dk
, fd
, devname
);
4521 dd
= malloc(sizeof(*dd
));
4524 Name
": malloc failed %s:%d.\n", __func__
, __LINE__
);
4527 memset(dd
, 0, sizeof(*dd
));
4528 dd
->major
= major(stb
.st_rdev
);
4529 dd
->minor
= minor(stb
.st_rdev
);
4530 dd
->devname
= devname
? strdup(devname
) : NULL
;
4533 dd
->action
= DISK_ADD
;
4534 rv
= imsm_read_serial(fd
, devname
, dd
->serial
);
4537 Name
": failed to retrieve scsi serial, aborting\n");
4542 get_dev_size(fd
, NULL
, &size
);
4544 serialcpy(dd
->disk
.serial
, dd
->serial
);
4545 dd
->disk
.total_blocks
= __cpu_to_le32(size
);
4547 if (sysfs_disk_to_scsi_id(fd
, &id
) == 0)
4548 dd
->disk
.scsi_id
= __cpu_to_le32(id
);
4550 dd
->disk
.scsi_id
= __cpu_to_le32(0);
4552 if (st
->update_tail
) {
4553 dd
->next
= super
->disk_mgmt_list
;
4554 super
->disk_mgmt_list
= dd
;
4556 dd
->next
= super
->disks
;
4558 super
->updates_pending
++;
4565 static int remove_from_super_imsm(struct supertype
*st
, mdu_disk_info_t
*dk
)
4567 struct intel_super
*super
= st
->sb
;
4570 /* remove from super works only in mdmon - for communication
4571 * manager - monitor. Check if communication memory buffer
4574 if (!st
->update_tail
) {
4576 Name
": %s shall be used in mdmon context only"
4577 "(line %d).\n", __func__
, __LINE__
);
4580 dd
= malloc(sizeof(*dd
));
4583 Name
": malloc failed %s:%d.\n", __func__
, __LINE__
);
4586 memset(dd
, 0, sizeof(*dd
));
4587 dd
->major
= dk
->major
;
4588 dd
->minor
= dk
->minor
;
4591 dd
->action
= DISK_REMOVE
;
4593 dd
->next
= super
->disk_mgmt_list
;
4594 super
->disk_mgmt_list
= dd
;
4600 static int store_imsm_mpb(int fd
, struct imsm_super
*mpb
);
4604 struct imsm_super anchor
;
4605 } spare_record
__attribute__ ((aligned(512)));
4607 /* spare records have their own family number and do not have any defined raid
4610 static int write_super_imsm_spares(struct intel_super
*super
, int doclose
)
4612 struct imsm_super
*mpb
= super
->anchor
;
4613 struct imsm_super
*spare
= &spare_record
.anchor
;
4617 spare
->mpb_size
= __cpu_to_le32(sizeof(struct imsm_super
)),
4618 spare
->generation_num
= __cpu_to_le32(1UL),
4619 spare
->attributes
= MPB_ATTRIB_CHECKSUM_VERIFY
;
4620 spare
->num_disks
= 1,
4621 spare
->num_raid_devs
= 0,
4622 spare
->cache_size
= mpb
->cache_size
,
4623 spare
->pwr_cycle_count
= __cpu_to_le32(1),
4625 snprintf((char *) spare
->sig
, MAX_SIGNATURE_LENGTH
,
4626 MPB_SIGNATURE MPB_VERSION_RAID0
);
4628 for (d
= super
->disks
; d
; d
= d
->next
) {
4632 spare
->disk
[0] = d
->disk
;
4633 sum
= __gen_imsm_checksum(spare
);
4634 spare
->family_num
= __cpu_to_le32(sum
);
4635 spare
->orig_family_num
= 0;
4636 sum
= __gen_imsm_checksum(spare
);
4637 spare
->check_sum
= __cpu_to_le32(sum
);
4639 if (store_imsm_mpb(d
->fd
, spare
)) {
4640 fprintf(stderr
, "%s: failed for device %d:%d %s\n",
4641 __func__
, d
->major
, d
->minor
, strerror(errno
));
4653 static int write_super_imsm(struct supertype
*st
, int doclose
)
4655 struct intel_super
*super
= st
->sb
;
4656 struct imsm_super
*mpb
= super
->anchor
;
4662 __u32 mpb_size
= sizeof(struct imsm_super
) - sizeof(struct imsm_disk
);
4664 int clear_migration_record
= 1;
4666 /* 'generation' is incremented everytime the metadata is written */
4667 generation
= __le32_to_cpu(mpb
->generation_num
);
4669 mpb
->generation_num
= __cpu_to_le32(generation
);
4671 /* fix up cases where previous mdadm releases failed to set
4674 if (mpb
->orig_family_num
== 0)
4675 mpb
->orig_family_num
= mpb
->family_num
;
4677 for (d
= super
->disks
; d
; d
= d
->next
) {
4681 mpb
->disk
[d
->index
] = d
->disk
;
4685 for (d
= super
->missing
; d
; d
= d
->next
) {
4686 mpb
->disk
[d
->index
] = d
->disk
;
4689 mpb
->num_disks
= num_disks
;
4690 mpb_size
+= sizeof(struct imsm_disk
) * mpb
->num_disks
;
4692 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
4693 struct imsm_dev
*dev
= __get_imsm_dev(mpb
, i
);
4694 struct imsm_dev
*dev2
= get_imsm_dev(super
, i
);
4696 imsm_copy_dev(dev
, dev2
);
4697 mpb_size
+= sizeof_imsm_dev(dev
, 0);
4699 if (is_gen_migration(dev2
))
4700 clear_migration_record
= 0;
4702 mpb_size
+= __le32_to_cpu(mpb
->bbm_log_size
);
4703 mpb
->mpb_size
= __cpu_to_le32(mpb_size
);
4705 /* recalculate checksum */
4706 sum
= __gen_imsm_checksum(mpb
);
4707 mpb
->check_sum
= __cpu_to_le32(sum
);
4709 if (clear_migration_record
)
4710 memset(super
->migr_rec_buf
, 0, 512);
4712 /* write the mpb for disks that compose raid devices */
4713 for (d
= super
->disks
; d
; d
= d
->next
) {
4714 if (d
->index
< 0 || is_failed(&d
->disk
))
4716 if (store_imsm_mpb(d
->fd
, mpb
))
4717 fprintf(stderr
, "%s: failed for device %d:%d %s\n",
4718 __func__
, d
->major
, d
->minor
, strerror(errno
));
4719 if (clear_migration_record
) {
4720 unsigned long long dsize
;
4722 get_dev_size(d
->fd
, NULL
, &dsize
);
4723 if (lseek64(d
->fd
, dsize
- 512, SEEK_SET
) >= 0) {
4724 if (write(d
->fd
, super
->migr_rec_buf
, 512) != 512)
4725 perror("Write migr_rec failed");
4735 return write_super_imsm_spares(super
, doclose
);
4741 static int create_array(struct supertype
*st
, int dev_idx
)
4744 struct imsm_update_create_array
*u
;
4745 struct intel_super
*super
= st
->sb
;
4746 struct imsm_dev
*dev
= get_imsm_dev(super
, dev_idx
);
4747 struct imsm_map
*map
= get_imsm_map(dev
, 0);
4748 struct disk_info
*inf
;
4749 struct imsm_disk
*disk
;
4752 len
= sizeof(*u
) - sizeof(*dev
) + sizeof_imsm_dev(dev
, 0) +
4753 sizeof(*inf
) * map
->num_members
;
4756 fprintf(stderr
, "%s: failed to allocate update buffer\n",
4761 u
->type
= update_create_array
;
4762 u
->dev_idx
= dev_idx
;
4763 imsm_copy_dev(&u
->dev
, dev
);
4764 inf
= get_disk_info(u
);
4765 for (i
= 0; i
< map
->num_members
; i
++) {
4766 int idx
= get_imsm_disk_idx(dev
, i
, -1);
4768 disk
= get_imsm_disk(super
, idx
);
4769 serialcpy(inf
[i
].serial
, disk
->serial
);
4771 append_metadata_update(st
, u
, len
);
4776 static int mgmt_disk(struct supertype
*st
)
4778 struct intel_super
*super
= st
->sb
;
4780 struct imsm_update_add_remove_disk
*u
;
4782 if (!super
->disk_mgmt_list
)
4788 fprintf(stderr
, "%s: failed to allocate update buffer\n",
4793 u
->type
= update_add_remove_disk
;
4794 append_metadata_update(st
, u
, len
);
4799 static int write_init_super_imsm(struct supertype
*st
)
4801 struct intel_super
*super
= st
->sb
;
4802 int current_vol
= super
->current_vol
;
4804 /* we are done with current_vol reset it to point st at the container */
4805 super
->current_vol
= -1;
4807 if (st
->update_tail
) {
4808 /* queue the recently created array / added disk
4809 * as a metadata update */
4812 /* determine if we are creating a volume or adding a disk */
4813 if (current_vol
< 0) {
4814 /* in the mgmt (add/remove) disk case we are running
4815 * in mdmon context, so don't close fd's
4817 return mgmt_disk(st
);
4819 rv
= create_array(st
, current_vol
);
4824 for (d
= super
->disks
; d
; d
= d
->next
)
4825 Kill(d
->devname
, NULL
, 0, 1, 1);
4826 return write_super_imsm(st
, 1);
4831 static int store_super_imsm(struct supertype
*st
, int fd
)
4833 struct intel_super
*super
= st
->sb
;
4834 struct imsm_super
*mpb
= super
? super
->anchor
: NULL
;
4840 return store_imsm_mpb(fd
, mpb
);
4846 static int imsm_bbm_log_size(struct imsm_super
*mpb
)
4848 return __le32_to_cpu(mpb
->bbm_log_size
);
4852 static int validate_geometry_imsm_container(struct supertype
*st
, int level
,
4853 int layout
, int raiddisks
, int chunk
,
4854 unsigned long long size
, char *dev
,
4855 unsigned long long *freesize
,
4859 unsigned long long ldsize
;
4860 struct intel_super
*super
=NULL
;
4863 if (level
!= LEVEL_CONTAINER
)
4868 fd
= open(dev
, O_RDONLY
|O_EXCL
, 0);
4871 fprintf(stderr
, Name
": imsm: Cannot open %s: %s\n",
4872 dev
, strerror(errno
));
4875 if (!get_dev_size(fd
, dev
, &ldsize
)) {
4880 /* capabilities retrieve could be possible
4881 * note that there is no fd for the disks in array.
4883 super
= alloc_super();
4886 Name
": malloc of %zu failed.\n",
4892 rv
= find_intel_hba_capability(fd
, super
, verbose
? dev
: NULL
);
4896 fd2devname(fd
, str
);
4897 dprintf("validate_geometry_imsm_container: fd: %d %s orom: %p rv: %d raiddisk: %d\n",
4898 fd
, str
, super
->orom
, rv
, raiddisks
);
4900 /* no orom/efi or non-intel hba of the disk */
4906 if (super
->orom
&& raiddisks
> super
->orom
->tds
) {
4908 fprintf(stderr
, Name
": %d exceeds maximum number of"
4909 " platform supported disks: %d\n",
4910 raiddisks
, super
->orom
->tds
);
4916 *freesize
= avail_size_imsm(st
, ldsize
>> 9);
4922 static unsigned long long find_size(struct extent
*e
, int *idx
, int num_extents
)
4924 const unsigned long long base_start
= e
[*idx
].start
;
4925 unsigned long long end
= base_start
+ e
[*idx
].size
;
4928 if (base_start
== end
)
4932 for (i
= *idx
; i
< num_extents
; i
++) {
4933 /* extend overlapping extents */
4934 if (e
[i
].start
>= base_start
&&
4935 e
[i
].start
<= end
) {
4938 if (e
[i
].start
+ e
[i
].size
> end
)
4939 end
= e
[i
].start
+ e
[i
].size
;
4940 } else if (e
[i
].start
> end
) {
4946 return end
- base_start
;
4949 static unsigned long long merge_extents(struct intel_super
*super
, int sum_extents
)
4951 /* build a composite disk with all known extents and generate a new
4952 * 'maxsize' given the "all disks in an array must share a common start
4953 * offset" constraint
4955 struct extent
*e
= calloc(sum_extents
, sizeof(*e
));
4959 unsigned long long pos
;
4960 unsigned long long start
= 0;
4961 unsigned long long maxsize
;
4962 unsigned long reserve
;
4967 /* coalesce and sort all extents. also, check to see if we need to
4968 * reserve space between member arrays
4971 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
4974 for (i
= 0; i
< dl
->extent_cnt
; i
++)
4977 qsort(e
, sum_extents
, sizeof(*e
), cmp_extent
);
4982 while (i
< sum_extents
) {
4983 e
[j
].start
= e
[i
].start
;
4984 e
[j
].size
= find_size(e
, &i
, sum_extents
);
4986 if (e
[j
-1].size
== 0)
4995 unsigned long long esize
;
4997 esize
= e
[i
].start
- pos
;
4998 if (esize
>= maxsize
) {
5003 pos
= e
[i
].start
+ e
[i
].size
;
5005 } while (e
[i
-1].size
);
5011 /* FIXME assumes volume at offset 0 is the first volume in a
5014 if (start_extent
> 0)
5015 reserve
= IMSM_RESERVED_SECTORS
; /* gap between raid regions */
5019 if (maxsize
< reserve
)
5022 super
->create_offset
= ~((__u32
) 0);
5023 if (start
+ reserve
> super
->create_offset
)
5024 return 0; /* start overflows create_offset */
5025 super
->create_offset
= start
+ reserve
;
5027 return maxsize
- reserve
;
5030 static int is_raid_level_supported(const struct imsm_orom
*orom
, int level
, int raiddisks
)
5032 if (level
< 0 || level
== 6 || level
== 4)
5035 /* if we have an orom prevent invalid raid levels */
5038 case 0: return imsm_orom_has_raid0(orom
);
5041 return imsm_orom_has_raid1e(orom
);
5042 return imsm_orom_has_raid1(orom
) && raiddisks
== 2;
5043 case 10: return imsm_orom_has_raid10(orom
) && raiddisks
== 4;
5044 case 5: return imsm_orom_has_raid5(orom
) && raiddisks
> 2;
5047 return 1; /* not on an Intel RAID platform so anything goes */
5052 static int imsm_default_chunk(const struct imsm_orom
*orom
)
5054 /* up to 512 if the plaform supports it, otherwise the platform max.
5055 * 128 if no platform detected
5057 int fs
= max(7, orom
? fls(orom
->sss
) : 0);
5059 return min(512, (1 << fs
));
5062 #define pr_vrb(fmt, arg...) (void) (verbose && fprintf(stderr, Name fmt, ##arg))
5064 validate_geometry_imsm_orom(struct intel_super
*super
, int level
, int layout
,
5065 int raiddisks
, int *chunk
, int verbose
)
5067 /* check/set platform and metadata limits/defaults */
5068 if (super
->orom
&& raiddisks
> super
->orom
->dpa
) {
5069 pr_vrb(": platform supports a maximum of %d disks per array\n",
5074 /* capabilities of OROM tested - copied from validate_geometry_imsm_volume */
5075 if (!is_raid_level_supported(super
->orom
, level
, raiddisks
)) {
5076 pr_vrb(": platform does not support raid%d with %d disk%s\n",
5077 level
, raiddisks
, raiddisks
> 1 ? "s" : "");
5081 if (chunk
&& (*chunk
== 0 || *chunk
== UnSet
))
5082 *chunk
= imsm_default_chunk(super
->orom
);
5084 if (super
->orom
&& chunk
&& !imsm_orom_has_chunk(super
->orom
, *chunk
)) {
5085 pr_vrb(": platform does not support a chunk size of: "
5090 if (layout
!= imsm_level_to_layout(level
)) {
5092 pr_vrb(": imsm raid 5 only supports the left-asymmetric layout\n");
5093 else if (level
== 10)
5094 pr_vrb(": imsm raid 10 only supports the n2 layout\n");
5096 pr_vrb(": imsm unknown layout %#x for this raid level %d\n",
5103 /* validate_geometry_imsm_volume - lifted from validate_geometry_ddf_bvd
5104 * FIX ME add ahci details
5106 static int validate_geometry_imsm_volume(struct supertype
*st
, int level
,
5107 int layout
, int raiddisks
, int *chunk
,
5108 unsigned long long size
, char *dev
,
5109 unsigned long long *freesize
,
5113 struct intel_super
*super
= st
->sb
;
5114 struct imsm_super
*mpb
= super
->anchor
;
5116 unsigned long long pos
= 0;
5117 unsigned long long maxsize
;
5121 /* We must have the container info already read in. */
5125 if (mpb
->num_raid_devs
> 0 && mpb
->num_disks
!= raiddisks
) {
5126 fprintf(stderr
, Name
": the option-rom requires all "
5127 "member disks to be a member of all volumes.\n");
5131 if (!validate_geometry_imsm_orom(super
, level
, layout
, raiddisks
, chunk
, verbose
)) {
5132 fprintf(stderr
, Name
": RAID gemetry validation failed. "
5133 "Cannot proceed with the action(s).\n");
5137 /* General test: make sure there is space for
5138 * 'raiddisks' device extents of size 'size' at a given
5141 unsigned long long minsize
= size
;
5142 unsigned long long start_offset
= MaxSector
;
5145 minsize
= MPB_SECTOR_CNT
+ IMSM_RESERVED_SECTORS
;
5146 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
5151 e
= get_extents(super
, dl
);
5154 unsigned long long esize
;
5155 esize
= e
[i
].start
- pos
;
5156 if (esize
>= minsize
)
5158 if (found
&& start_offset
== MaxSector
) {
5161 } else if (found
&& pos
!= start_offset
) {
5165 pos
= e
[i
].start
+ e
[i
].size
;
5167 } while (e
[i
-1].size
);
5172 if (dcnt
< raiddisks
) {
5174 fprintf(stderr
, Name
": imsm: Not enough "
5175 "devices with space for this array "
5183 /* This device must be a member of the set */
5184 if (stat(dev
, &stb
) < 0)
5186 if ((S_IFMT
& stb
.st_mode
) != S_IFBLK
)
5188 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
5189 if (dl
->major
== (int)major(stb
.st_rdev
) &&
5190 dl
->minor
== (int)minor(stb
.st_rdev
))
5195 fprintf(stderr
, Name
": %s is not in the "
5196 "same imsm set\n", dev
);
5198 } else if (super
->orom
&& dl
->index
< 0 && mpb
->num_raid_devs
) {
5199 /* If a volume is present then the current creation attempt
5200 * cannot incorporate new spares because the orom may not
5201 * understand this configuration (all member disks must be
5202 * members of each array in the container).
5204 fprintf(stderr
, Name
": %s is a spare and a volume"
5205 " is already defined for this container\n", dev
);
5206 fprintf(stderr
, Name
": The option-rom requires all member"
5207 " disks to be a member of all volumes\n");
5211 /* retrieve the largest free space block */
5212 e
= get_extents(super
, dl
);
5217 unsigned long long esize
;
5219 esize
= e
[i
].start
- pos
;
5220 if (esize
>= maxsize
)
5222 pos
= e
[i
].start
+ e
[i
].size
;
5224 } while (e
[i
-1].size
);
5229 fprintf(stderr
, Name
": unable to determine free space for: %s\n",
5233 if (maxsize
< size
) {
5235 fprintf(stderr
, Name
": %s not enough space (%llu < %llu)\n",
5236 dev
, maxsize
, size
);
5240 /* count total number of extents for merge */
5242 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
5244 i
+= dl
->extent_cnt
;
5246 maxsize
= merge_extents(super
, i
);
5247 if (maxsize
< size
|| maxsize
== 0) {
5249 fprintf(stderr
, Name
": not enough space after merge (%llu < %llu)\n",
5254 *freesize
= maxsize
;
5259 static int reserve_space(struct supertype
*st
, int raiddisks
,
5260 unsigned long long size
, int chunk
,
5261 unsigned long long *freesize
)
5263 struct intel_super
*super
= st
->sb
;
5264 struct imsm_super
*mpb
= super
->anchor
;
5269 unsigned long long maxsize
;
5270 unsigned long long minsize
;
5274 /* find the largest common start free region of the possible disks */
5278 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
5284 /* don't activate new spares if we are orom constrained
5285 * and there is already a volume active in the container
5287 if (super
->orom
&& dl
->index
< 0 && mpb
->num_raid_devs
)
5290 e
= get_extents(super
, dl
);
5293 for (i
= 1; e
[i
-1].size
; i
++)
5301 maxsize
= merge_extents(super
, extent_cnt
);
5305 minsize
= chunk
* 2;
5307 if (cnt
< raiddisks
||
5308 (super
->orom
&& used
&& used
!= raiddisks
) ||
5309 maxsize
< minsize
||
5311 fprintf(stderr
, Name
": not enough devices with space to create array.\n");
5312 return 0; /* No enough free spaces large enough */
5324 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
5326 dl
->raiddisk
= cnt
++;
5333 static int validate_geometry_imsm(struct supertype
*st
, int level
, int layout
,
5334 int raiddisks
, int *chunk
, unsigned long long size
,
5335 char *dev
, unsigned long long *freesize
,
5343 * if given unused devices create a container
5344 * if given given devices in a container create a member volume
5346 if (level
== LEVEL_CONTAINER
) {
5347 /* Must be a fresh device to add to a container */
5348 return validate_geometry_imsm_container(st
, level
, layout
,
5350 chunk
?*chunk
:0, size
,
5356 if (st
->sb
&& freesize
) {
5357 /* we are being asked to automatically layout a
5358 * new volume based on the current contents of
5359 * the container. If the the parameters can be
5360 * satisfied reserve_space will record the disks,
5361 * start offset, and size of the volume to be
5362 * created. add_to_super and getinfo_super
5363 * detect when autolayout is in progress.
5365 if (!validate_geometry_imsm_orom(st
->sb
, level
, layout
,
5369 return reserve_space(st
, raiddisks
, size
,
5370 chunk
?*chunk
:0, freesize
);
5375 /* creating in a given container */
5376 return validate_geometry_imsm_volume(st
, level
, layout
,
5377 raiddisks
, chunk
, size
,
5378 dev
, freesize
, verbose
);
5381 /* This device needs to be a device in an 'imsm' container */
5382 fd
= open(dev
, O_RDONLY
|O_EXCL
, 0);
5386 Name
": Cannot create this array on device %s\n",
5391 if (errno
!= EBUSY
|| (fd
= open(dev
, O_RDONLY
, 0)) < 0) {
5393 fprintf(stderr
, Name
": Cannot open %s: %s\n",
5394 dev
, strerror(errno
));
5397 /* Well, it is in use by someone, maybe an 'imsm' container. */
5398 cfd
= open_container(fd
);
5402 fprintf(stderr
, Name
": Cannot use %s: It is busy\n",
5406 sra
= sysfs_read(cfd
, 0, GET_VERSION
);
5407 if (sra
&& sra
->array
.major_version
== -1 &&
5408 strcmp(sra
->text_version
, "imsm") == 0)
5412 /* This is a member of a imsm container. Load the container
5413 * and try to create a volume
5415 struct intel_super
*super
;
5417 if (load_super_imsm_all(st
, cfd
, (void **) &super
, NULL
) == 0) {
5419 st
->container_dev
= fd2devnum(cfd
);
5421 return validate_geometry_imsm_volume(st
, level
, layout
,
5430 fprintf(stderr
, Name
": failed container membership check\n");
5436 static void default_geometry_imsm(struct supertype
*st
, int *level
, int *layout
, int *chunk
)
5438 struct intel_super
*super
= st
->sb
;
5440 if (level
&& *level
== UnSet
)
5441 *level
= LEVEL_CONTAINER
;
5443 if (level
&& layout
&& *layout
== UnSet
)
5444 *layout
= imsm_level_to_layout(*level
);
5446 if (chunk
&& (*chunk
== UnSet
|| *chunk
== 0))
5447 *chunk
= imsm_default_chunk(super
->orom
);
5450 static void handle_missing(struct intel_super
*super
, struct imsm_dev
*dev
);
5452 static int kill_subarray_imsm(struct supertype
*st
)
5454 /* remove the subarray currently referenced by ->current_vol */
5456 struct intel_dev
**dp
;
5457 struct intel_super
*super
= st
->sb
;
5458 __u8 current_vol
= super
->current_vol
;
5459 struct imsm_super
*mpb
= super
->anchor
;
5461 if (super
->current_vol
< 0)
5463 super
->current_vol
= -1; /* invalidate subarray cursor */
5465 /* block deletions that would change the uuid of active subarrays
5467 * FIXME when immutable ids are available, but note that we'll
5468 * also need to fixup the invalidated/active subarray indexes in
5471 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
5474 if (i
< current_vol
)
5476 sprintf(subarray
, "%u", i
);
5477 if (is_subarray_active(subarray
, st
->devname
)) {
5479 Name
": deleting subarray-%d would change the UUID of active subarray-%d, aborting\n",
5486 if (st
->update_tail
) {
5487 struct imsm_update_kill_array
*u
= malloc(sizeof(*u
));
5491 u
->type
= update_kill_array
;
5492 u
->dev_idx
= current_vol
;
5493 append_metadata_update(st
, u
, sizeof(*u
));
5498 for (dp
= &super
->devlist
; *dp
;)
5499 if ((*dp
)->index
== current_vol
) {
5502 handle_missing(super
, (*dp
)->dev
);
5503 if ((*dp
)->index
> current_vol
)
5508 /* no more raid devices, all active components are now spares,
5509 * but of course failed are still failed
5511 if (--mpb
->num_raid_devs
== 0) {
5514 for (d
= super
->disks
; d
; d
= d
->next
)
5519 super
->updates_pending
++;
5524 static int update_subarray_imsm(struct supertype
*st
, char *subarray
,
5525 char *update
, struct mddev_ident
*ident
)
5527 /* update the subarray currently referenced by ->current_vol */
5528 struct intel_super
*super
= st
->sb
;
5529 struct imsm_super
*mpb
= super
->anchor
;
5531 if (strcmp(update
, "name") == 0) {
5532 char *name
= ident
->name
;
5536 if (is_subarray_active(subarray
, st
->devname
)) {
5538 Name
": Unable to update name of active subarray\n");
5542 if (!check_name(super
, name
, 0))
5545 vol
= strtoul(subarray
, &ep
, 10);
5546 if (*ep
!= '\0' || vol
>= super
->anchor
->num_raid_devs
)
5549 if (st
->update_tail
) {
5550 struct imsm_update_rename_array
*u
= malloc(sizeof(*u
));
5554 u
->type
= update_rename_array
;
5556 snprintf((char *) u
->name
, MAX_RAID_SERIAL_LEN
, "%s", name
);
5557 append_metadata_update(st
, u
, sizeof(*u
));
5559 struct imsm_dev
*dev
;
5562 dev
= get_imsm_dev(super
, vol
);
5563 snprintf((char *) dev
->volume
, MAX_RAID_SERIAL_LEN
, "%s", name
);
5564 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
5565 dev
= get_imsm_dev(super
, i
);
5566 handle_missing(super
, dev
);
5568 super
->updates_pending
++;
5576 static int is_gen_migration(struct imsm_dev
*dev
)
5581 if (!dev
->vol
.migr_state
)
5584 if (migr_type(dev
) == MIGR_GEN_MIGR
)
5589 #endif /* MDASSEMBLE */
5591 static int is_rebuilding(struct imsm_dev
*dev
)
5593 struct imsm_map
*migr_map
;
5595 if (!dev
->vol
.migr_state
)
5598 if (migr_type(dev
) != MIGR_REBUILD
)
5601 migr_map
= get_imsm_map(dev
, 1);
5603 if (migr_map
->map_state
== IMSM_T_STATE_DEGRADED
)
5609 static void update_recovery_start(struct intel_super
*super
,
5610 struct imsm_dev
*dev
,
5611 struct mdinfo
*array
)
5613 struct mdinfo
*rebuild
= NULL
;
5617 if (!is_rebuilding(dev
))
5620 /* Find the rebuild target, but punt on the dual rebuild case */
5621 for (d
= array
->devs
; d
; d
= d
->next
)
5622 if (d
->recovery_start
== 0) {
5629 /* (?) none of the disks are marked with
5630 * IMSM_ORD_REBUILD, so assume they are missing and the
5631 * disk_ord_tbl was not correctly updated
5633 dprintf("%s: failed to locate out-of-sync disk\n", __func__
);
5637 units
= __le32_to_cpu(dev
->vol
.curr_migr_unit
);
5638 rebuild
->recovery_start
= units
* blocks_per_migr_unit(super
, dev
);
5642 static int recover_backup_imsm(struct supertype
*st
, struct mdinfo
*info
);
5645 static struct mdinfo
*container_content_imsm(struct supertype
*st
, char *subarray
)
5647 /* Given a container loaded by load_super_imsm_all,
5648 * extract information about all the arrays into
5650 * If 'subarray' is given, just extract info about that array.
5652 * For each imsm_dev create an mdinfo, fill it in,
5653 * then look for matching devices in super->disks
5654 * and create appropriate device mdinfo.
5656 struct intel_super
*super
= st
->sb
;
5657 struct imsm_super
*mpb
= super
->anchor
;
5658 struct mdinfo
*rest
= NULL
;
5662 int spare_disks
= 0;
5664 /* do not assemble arrays when not all attributes are supported */
5665 if (imsm_check_attributes(mpb
->attributes
) == 0) {
5666 fprintf(stderr
, Name
": IMSM metadata loading not allowed "
5667 "due to attributes incompatibility.\n");
5671 /* check for bad blocks */
5672 if (imsm_bbm_log_size(super
->anchor
))
5675 /* count spare devices, not used in maps
5677 for (d
= super
->disks
; d
; d
= d
->next
)
5681 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
5682 struct imsm_dev
*dev
;
5683 struct imsm_map
*map
;
5684 struct imsm_map
*map2
;
5685 struct mdinfo
*this;
5690 (i
!= strtoul(subarray
, &ep
, 10) || *ep
!= '\0'))
5693 dev
= get_imsm_dev(super
, i
);
5694 map
= get_imsm_map(dev
, 0);
5695 map2
= get_imsm_map(dev
, 1);
5697 /* do not publish arrays that are in the middle of an
5698 * unsupported migration
5700 if (dev
->vol
.migr_state
&&
5701 (migr_type(dev
) == MIGR_STATE_CHANGE
)) {
5702 fprintf(stderr
, Name
": cannot assemble volume '%.16s':"
5703 " unsupported migration in progress\n",
5707 /* do not publish arrays that are not support by controller's
5711 chunk
= __le16_to_cpu(map
->blocks_per_strip
) >> 1;
5713 if (!validate_geometry_imsm_orom(super
,
5714 get_imsm_raid_level(map
), /* RAID level */
5715 imsm_level_to_layout(get_imsm_raid_level(map
)),
5716 map
->num_members
, /* raid disks */
5719 fprintf(stderr
, Name
": RAID gemetry validation failed. "
5720 "Cannot proceed with the action(s).\n");
5723 #endif /* MDASSEMBLE */
5724 this = malloc(sizeof(*this));
5726 fprintf(stderr
, Name
": failed to allocate %zu bytes\n",
5731 super
->current_vol
= i
;
5732 getinfo_super_imsm_volume(st
, this, NULL
);
5734 for (slot
= 0 ; slot
< map
->num_members
; slot
++) {
5735 unsigned long long recovery_start
;
5736 struct mdinfo
*info_d
;
5743 idx
= get_imsm_disk_idx(dev
, slot
, 0);
5744 ord
= get_imsm_ord_tbl_ent(dev
, slot
, -1);
5745 for (d
= super
->disks
; d
; d
= d
->next
)
5746 if (d
->index
== idx
)
5749 recovery_start
= MaxSector
;
5752 if (d
&& is_failed(&d
->disk
))
5754 if (ord
& IMSM_ORD_REBUILD
)
5758 * if we skip some disks the array will be assmebled degraded;
5759 * reset resync start to avoid a dirty-degraded
5760 * situation when performing the intial sync
5762 * FIXME handle dirty degraded
5764 if ((skip
|| recovery_start
== 0) && !dev
->vol
.dirty
)
5765 this->resync_start
= MaxSector
;
5769 info_d
= calloc(1, sizeof(*info_d
));
5771 fprintf(stderr
, Name
": failed to allocate disk"
5772 " for volume %.16s\n", dev
->volume
);
5773 info_d
= this->devs
;
5775 struct mdinfo
*d
= info_d
->next
;
5784 info_d
->next
= this->devs
;
5785 this->devs
= info_d
;
5787 info_d
->disk
.number
= d
->index
;
5788 info_d
->disk
.major
= d
->major
;
5789 info_d
->disk
.minor
= d
->minor
;
5790 info_d
->disk
.raid_disk
= slot
;
5791 info_d
->recovery_start
= recovery_start
;
5793 if (slot
< map2
->num_members
)
5794 info_d
->disk
.state
= (1 << MD_DISK_ACTIVE
);
5796 this->array
.spare_disks
++;
5798 if (slot
< map
->num_members
)
5799 info_d
->disk
.state
= (1 << MD_DISK_ACTIVE
);
5801 this->array
.spare_disks
++;
5803 if (info_d
->recovery_start
== MaxSector
)
5804 this->array
.working_disks
++;
5806 info_d
->events
= __le32_to_cpu(mpb
->generation_num
);
5807 info_d
->data_offset
= __le32_to_cpu(map
->pba_of_lba0
);
5808 info_d
->component_size
= __le32_to_cpu(map
->blocks_per_member
);
5810 /* now that the disk list is up-to-date fixup recovery_start */
5811 update_recovery_start(super
, dev
, this);
5812 this->array
.spare_disks
+= spare_disks
;
5815 /* check for reshape */
5816 if (this->reshape_active
== 1)
5817 recover_backup_imsm(st
, this);
5822 /* if array has bad blocks, set suitable bit in array status */
5824 rest
->array
.state
|= (1<<MD_SB_BBM_ERRORS
);
5830 static __u8
imsm_check_degraded(struct intel_super
*super
, struct imsm_dev
*dev
, int failed
)
5832 struct imsm_map
*map
= get_imsm_map(dev
, 0);
5835 return map
->map_state
== IMSM_T_STATE_UNINITIALIZED
?
5836 IMSM_T_STATE_UNINITIALIZED
: IMSM_T_STATE_NORMAL
;
5838 switch (get_imsm_raid_level(map
)) {
5840 return IMSM_T_STATE_FAILED
;
5843 if (failed
< map
->num_members
)
5844 return IMSM_T_STATE_DEGRADED
;
5846 return IMSM_T_STATE_FAILED
;
5851 * check to see if any mirrors have failed, otherwise we
5852 * are degraded. Even numbered slots are mirrored on
5856 /* gcc -Os complains that this is unused */
5857 int insync
= insync
;
5859 for (i
= 0; i
< map
->num_members
; i
++) {
5860 __u32 ord
= get_imsm_ord_tbl_ent(dev
, i
, -1);
5861 int idx
= ord_to_idx(ord
);
5862 struct imsm_disk
*disk
;
5864 /* reset the potential in-sync count on even-numbered
5865 * slots. num_copies is always 2 for imsm raid10
5870 disk
= get_imsm_disk(super
, idx
);
5871 if (!disk
|| is_failed(disk
) || ord
& IMSM_ORD_REBUILD
)
5874 /* no in-sync disks left in this mirror the
5878 return IMSM_T_STATE_FAILED
;
5881 return IMSM_T_STATE_DEGRADED
;
5885 return IMSM_T_STATE_DEGRADED
;
5887 return IMSM_T_STATE_FAILED
;
5893 return map
->map_state
;
5896 static int imsm_count_failed(struct intel_super
*super
, struct imsm_dev
*dev
)
5900 struct imsm_disk
*disk
;
5901 struct imsm_map
*map
= get_imsm_map(dev
, 0);
5902 struct imsm_map
*prev
= get_imsm_map(dev
, dev
->vol
.migr_state
);
5906 /* at the beginning of migration we set IMSM_ORD_REBUILD on
5907 * disks that are being rebuilt. New failures are recorded to
5908 * map[0]. So we look through all the disks we started with and
5909 * see if any failures are still present, or if any new ones
5912 * FIXME add support for online capacity expansion and
5913 * raid-level-migration
5915 for (i
= 0; i
< prev
->num_members
; i
++) {
5916 ord
= __le32_to_cpu(prev
->disk_ord_tbl
[i
]);
5917 ord
|= __le32_to_cpu(map
->disk_ord_tbl
[i
]);
5918 idx
= ord_to_idx(ord
);
5920 disk
= get_imsm_disk(super
, idx
);
5921 if (!disk
|| is_failed(disk
) || ord
& IMSM_ORD_REBUILD
)
5929 static int imsm_open_new(struct supertype
*c
, struct active_array
*a
,
5932 struct intel_super
*super
= c
->sb
;
5933 struct imsm_super
*mpb
= super
->anchor
;
5935 if (atoi(inst
) >= mpb
->num_raid_devs
) {
5936 fprintf(stderr
, "%s: subarry index %d, out of range\n",
5937 __func__
, atoi(inst
));
5941 dprintf("imsm: open_new %s\n", inst
);
5942 a
->info
.container_member
= atoi(inst
);
5946 static int is_resyncing(struct imsm_dev
*dev
)
5948 struct imsm_map
*migr_map
;
5950 if (!dev
->vol
.migr_state
)
5953 if (migr_type(dev
) == MIGR_INIT
||
5954 migr_type(dev
) == MIGR_REPAIR
)
5957 if (migr_type(dev
) == MIGR_GEN_MIGR
)
5960 migr_map
= get_imsm_map(dev
, 1);
5962 if ((migr_map
->map_state
== IMSM_T_STATE_NORMAL
) &&
5963 (dev
->vol
.migr_type
!= MIGR_GEN_MIGR
))
5969 /* return true if we recorded new information */
5970 static int mark_failure(struct imsm_dev
*dev
, struct imsm_disk
*disk
, int idx
)
5974 struct imsm_map
*map
;
5975 char buf
[MAX_RAID_SERIAL_LEN
+3];
5976 unsigned int len
, shift
= 0;
5978 /* new failures are always set in map[0] */
5979 map
= get_imsm_map(dev
, 0);
5981 slot
= get_imsm_disk_slot(map
, idx
);
5985 ord
= __le32_to_cpu(map
->disk_ord_tbl
[slot
]);
5986 if (is_failed(disk
) && (ord
& IMSM_ORD_REBUILD
))
5989 sprintf(buf
, "%s:0", disk
->serial
);
5990 if ((len
= strlen(buf
)) >= MAX_RAID_SERIAL_LEN
)
5991 shift
= len
- MAX_RAID_SERIAL_LEN
+ 1;
5992 strncpy((char *)disk
->serial
, &buf
[shift
], MAX_RAID_SERIAL_LEN
);
5994 disk
->status
|= FAILED_DISK
;
5995 set_imsm_ord_tbl_ent(map
, slot
, idx
| IMSM_ORD_REBUILD
);
5996 if (map
->failed_disk_num
== 0xff)
5997 map
->failed_disk_num
= slot
;
6001 static void mark_missing(struct imsm_dev
*dev
, struct imsm_disk
*disk
, int idx
)
6003 mark_failure(dev
, disk
, idx
);
6005 if (disk
->scsi_id
== __cpu_to_le32(~(__u32
)0))
6008 disk
->scsi_id
= __cpu_to_le32(~(__u32
)0);
6009 memmove(&disk
->serial
[0], &disk
->serial
[1], MAX_RAID_SERIAL_LEN
- 1);
6012 static void handle_missing(struct intel_super
*super
, struct imsm_dev
*dev
)
6018 if (!super
->missing
)
6020 failed
= imsm_count_failed(super
, dev
);
6021 map_state
= imsm_check_degraded(super
, dev
, failed
);
6023 dprintf("imsm: mark missing\n");
6024 end_migration(dev
, map_state
);
6025 for (dl
= super
->missing
; dl
; dl
= dl
->next
)
6026 mark_missing(dev
, &dl
->disk
, dl
->index
);
6027 super
->updates_pending
++;
6030 static unsigned long long imsm_set_array_size(struct imsm_dev
*dev
)
6032 int used_disks
= imsm_num_data_members(dev
, 0);
6033 unsigned long long array_blocks
;
6034 struct imsm_map
*map
;
6036 if (used_disks
== 0) {
6037 /* when problems occures
6038 * return current array_blocks value
6040 array_blocks
= __le32_to_cpu(dev
->size_high
);
6041 array_blocks
= array_blocks
<< 32;
6042 array_blocks
+= __le32_to_cpu(dev
->size_low
);
6044 return array_blocks
;
6047 /* set array size in metadata
6049 map
= get_imsm_map(dev
, 0);
6050 array_blocks
= map
->blocks_per_member
* used_disks
;
6052 /* round array size down to closest MB
6054 array_blocks
= (array_blocks
>> SECT_PER_MB_SHIFT
) << SECT_PER_MB_SHIFT
;
6055 dev
->size_low
= __cpu_to_le32((__u32
)array_blocks
);
6056 dev
->size_high
= __cpu_to_le32((__u32
)(array_blocks
>> 32));
6058 return array_blocks
;
6061 static void imsm_set_disk(struct active_array
*a
, int n
, int state
);
6063 static void imsm_progress_container_reshape(struct intel_super
*super
)
6065 /* if no device has a migr_state, but some device has a
6066 * different number of members than the previous device, start
6067 * changing the number of devices in this device to match
6070 struct imsm_super
*mpb
= super
->anchor
;
6071 int prev_disks
= -1;
6075 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
6076 struct imsm_dev
*dev
= get_imsm_dev(super
, i
);
6077 struct imsm_map
*map
= get_imsm_map(dev
, 0);
6078 struct imsm_map
*map2
;
6079 int prev_num_members
;
6081 if (dev
->vol
.migr_state
)
6084 if (prev_disks
== -1)
6085 prev_disks
= map
->num_members
;
6086 if (prev_disks
== map
->num_members
)
6089 /* OK, this array needs to enter reshape mode.
6090 * i.e it needs a migr_state
6093 copy_map_size
= sizeof_imsm_map(map
);
6094 prev_num_members
= map
->num_members
;
6095 map
->num_members
= prev_disks
;
6096 dev
->vol
.migr_state
= 1;
6097 dev
->vol
.curr_migr_unit
= 0;
6098 dev
->vol
.migr_type
= MIGR_GEN_MIGR
;
6099 for (i
= prev_num_members
;
6100 i
< map
->num_members
; i
++)
6101 set_imsm_ord_tbl_ent(map
, i
, i
);
6102 map2
= get_imsm_map(dev
, 1);
6103 /* Copy the current map */
6104 memcpy(map2
, map
, copy_map_size
);
6105 map2
->num_members
= prev_num_members
;
6107 imsm_set_array_size(dev
);
6108 super
->updates_pending
++;
6112 /* Handle dirty -> clean transititions, resync and reshape. Degraded and rebuild
6113 * states are handled in imsm_set_disk() with one exception, when a
6114 * resync is stopped due to a new failure this routine will set the
6115 * 'degraded' state for the array.
6117 static int imsm_set_array_state(struct active_array
*a
, int consistent
)
6119 int inst
= a
->info
.container_member
;
6120 struct intel_super
*super
= a
->container
->sb
;
6121 struct imsm_dev
*dev
= get_imsm_dev(super
, inst
);
6122 struct imsm_map
*map
= get_imsm_map(dev
, 0);
6123 int failed
= imsm_count_failed(super
, dev
);
6124 __u8 map_state
= imsm_check_degraded(super
, dev
, failed
);
6125 __u32 blocks_per_unit
;
6127 if (dev
->vol
.migr_state
&&
6128 dev
->vol
.migr_type
== MIGR_GEN_MIGR
) {
6129 /* array state change is blocked due to reshape action
6131 * - abort the reshape (if last_checkpoint is 0 and action!= reshape)
6132 * - finish the reshape (if last_checkpoint is big and action != reshape)
6133 * - update curr_migr_unit
6135 if (a
->curr_action
== reshape
) {
6136 /* still reshaping, maybe update curr_migr_unit */
6137 goto mark_checkpoint
;
6139 if (a
->last_checkpoint
== 0 && a
->prev_action
== reshape
) {
6140 /* for some reason we aborted the reshape.
6142 * disable automatic metadata rollback
6143 * user action is required to recover process
6146 struct imsm_map
*map2
= get_imsm_map(dev
, 1);
6147 dev
->vol
.migr_state
= 0;
6148 dev
->vol
.migr_type
= 0;
6149 dev
->vol
.curr_migr_unit
= 0;
6150 memcpy(map
, map2
, sizeof_imsm_map(map2
));
6151 super
->updates_pending
++;
6154 if (a
->last_checkpoint
>= a
->info
.component_size
) {
6155 unsigned long long array_blocks
;
6159 used_disks
= imsm_num_data_members(dev
, 0);
6160 if (used_disks
> 0) {
6162 map
->blocks_per_member
*
6164 /* round array size down to closest MB
6166 array_blocks
= (array_blocks
6167 >> SECT_PER_MB_SHIFT
)
6168 << SECT_PER_MB_SHIFT
;
6169 a
->info
.custom_array_size
= array_blocks
;
6170 /* encourage manager to update array
6174 a
->check_reshape
= 1;
6176 /* finalize online capacity expansion/reshape */
6177 for (mdi
= a
->info
.devs
; mdi
; mdi
= mdi
->next
)
6179 mdi
->disk
.raid_disk
,
6182 imsm_progress_container_reshape(super
);
6187 /* before we activate this array handle any missing disks */
6188 if (consistent
== 2)
6189 handle_missing(super
, dev
);
6191 if (consistent
== 2 &&
6192 (!is_resync_complete(&a
->info
) ||
6193 map_state
!= IMSM_T_STATE_NORMAL
||
6194 dev
->vol
.migr_state
))
6197 if (is_resync_complete(&a
->info
)) {
6198 /* complete intialization / resync,
6199 * recovery and interrupted recovery is completed in
6202 if (is_resyncing(dev
)) {
6203 dprintf("imsm: mark resync done\n");
6204 end_migration(dev
, map_state
);
6205 super
->updates_pending
++;
6206 a
->last_checkpoint
= 0;
6208 } else if (!is_resyncing(dev
) && !failed
) {
6209 /* mark the start of the init process if nothing is failed */
6210 dprintf("imsm: mark resync start\n");
6211 if (map
->map_state
== IMSM_T_STATE_UNINITIALIZED
)
6212 migrate(dev
, super
, IMSM_T_STATE_NORMAL
, MIGR_INIT
);
6214 migrate(dev
, super
, IMSM_T_STATE_NORMAL
, MIGR_REPAIR
);
6215 super
->updates_pending
++;
6219 /* skip checkpointing for general migration,
6220 * it is controlled in mdadm
6222 if (is_gen_migration(dev
))
6223 goto skip_mark_checkpoint
;
6225 /* check if we can update curr_migr_unit from resync_start, recovery_start */
6226 blocks_per_unit
= blocks_per_migr_unit(super
, dev
);
6227 if (blocks_per_unit
) {
6231 units
= a
->last_checkpoint
/ blocks_per_unit
;
6234 /* check that we did not overflow 32-bits, and that
6235 * curr_migr_unit needs updating
6237 if (units32
== units
&&
6239 __le32_to_cpu(dev
->vol
.curr_migr_unit
) != units32
) {
6240 dprintf("imsm: mark checkpoint (%u)\n", units32
);
6241 dev
->vol
.curr_migr_unit
= __cpu_to_le32(units32
);
6242 super
->updates_pending
++;
6246 skip_mark_checkpoint
:
6247 /* mark dirty / clean */
6248 if (dev
->vol
.dirty
!= !consistent
) {
6249 dprintf("imsm: mark '%s'\n", consistent
? "clean" : "dirty");
6254 super
->updates_pending
++;
6260 static void imsm_set_disk(struct active_array
*a
, int n
, int state
)
6262 int inst
= a
->info
.container_member
;
6263 struct intel_super
*super
= a
->container
->sb
;
6264 struct imsm_dev
*dev
= get_imsm_dev(super
, inst
);
6265 struct imsm_map
*map
= get_imsm_map(dev
, 0);
6266 struct imsm_disk
*disk
;
6271 if (n
> map
->num_members
)
6272 fprintf(stderr
, "imsm: set_disk %d out of range 0..%d\n",
6273 n
, map
->num_members
- 1);
6278 dprintf("imsm: set_disk %d:%x\n", n
, state
);
6280 ord
= get_imsm_ord_tbl_ent(dev
, n
, -1);
6281 disk
= get_imsm_disk(super
, ord_to_idx(ord
));
6283 /* check for new failures */
6284 if (state
& DS_FAULTY
) {
6285 if (mark_failure(dev
, disk
, ord_to_idx(ord
)))
6286 super
->updates_pending
++;
6289 /* check if in_sync */
6290 if (state
& DS_INSYNC
&& ord
& IMSM_ORD_REBUILD
&& is_rebuilding(dev
)) {
6291 struct imsm_map
*migr_map
= get_imsm_map(dev
, 1);
6293 set_imsm_ord_tbl_ent(migr_map
, n
, ord_to_idx(ord
));
6294 super
->updates_pending
++;
6297 failed
= imsm_count_failed(super
, dev
);
6298 map_state
= imsm_check_degraded(super
, dev
, failed
);
6300 /* check if recovery complete, newly degraded, or failed */
6301 if (map_state
== IMSM_T_STATE_NORMAL
&& is_rebuilding(dev
)) {
6302 end_migration(dev
, map_state
);
6303 map
= get_imsm_map(dev
, 0);
6304 map
->failed_disk_num
= ~0;
6305 super
->updates_pending
++;
6306 a
->last_checkpoint
= 0;
6307 } else if (map_state
== IMSM_T_STATE_DEGRADED
&&
6308 map
->map_state
!= map_state
&&
6309 !dev
->vol
.migr_state
) {
6310 dprintf("imsm: mark degraded\n");
6311 map
->map_state
= map_state
;
6312 super
->updates_pending
++;
6313 a
->last_checkpoint
= 0;
6314 } else if (map_state
== IMSM_T_STATE_FAILED
&&
6315 map
->map_state
!= map_state
) {
6316 dprintf("imsm: mark failed\n");
6317 end_migration(dev
, map_state
);
6318 super
->updates_pending
++;
6319 a
->last_checkpoint
= 0;
6320 } else if (is_gen_migration(dev
)) {
6321 dprintf("imsm: Detected General Migration in state: ");
6322 if (map_state
== IMSM_T_STATE_NORMAL
) {
6323 end_migration(dev
, map_state
);
6324 map
= get_imsm_map(dev
, 0);
6325 map
->failed_disk_num
= ~0;
6326 dprintf("normal\n");
6328 if (map_state
== IMSM_T_STATE_DEGRADED
) {
6329 printf("degraded\n");
6330 end_migration(dev
, map_state
);
6332 dprintf("failed\n");
6334 map
->map_state
= map_state
;
6336 super
->updates_pending
++;
6340 static int store_imsm_mpb(int fd
, struct imsm_super
*mpb
)
6343 __u32 mpb_size
= __le32_to_cpu(mpb
->mpb_size
);
6344 unsigned long long dsize
;
6345 unsigned long long sectors
;
6347 get_dev_size(fd
, NULL
, &dsize
);
6349 if (mpb_size
> 512) {
6350 /* -1 to account for anchor */
6351 sectors
= mpb_sectors(mpb
) - 1;
6353 /* write the extended mpb to the sectors preceeding the anchor */
6354 if (lseek64(fd
, dsize
- (512 * (2 + sectors
)), SEEK_SET
) < 0)
6357 if ((unsigned long long)write(fd
, buf
+ 512, 512 * sectors
)
6362 /* first block is stored on second to last sector of the disk */
6363 if (lseek64(fd
, dsize
- (512 * 2), SEEK_SET
) < 0)
6366 if (write(fd
, buf
, 512) != 512)
6372 static void imsm_sync_metadata(struct supertype
*container
)
6374 struct intel_super
*super
= container
->sb
;
6376 dprintf("sync metadata: %d\n", super
->updates_pending
);
6377 if (!super
->updates_pending
)
6380 write_super_imsm(container
, 0);
6382 super
->updates_pending
= 0;
6385 static struct dl
*imsm_readd(struct intel_super
*super
, int idx
, struct active_array
*a
)
6387 struct imsm_dev
*dev
= get_imsm_dev(super
, a
->info
.container_member
);
6388 int i
= get_imsm_disk_idx(dev
, idx
, -1);
6391 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
6395 if (dl
&& is_failed(&dl
->disk
))
6399 dprintf("%s: found %x:%x\n", __func__
, dl
->major
, dl
->minor
);
6404 static struct dl
*imsm_add_spare(struct intel_super
*super
, int slot
,
6405 struct active_array
*a
, int activate_new
,
6406 struct mdinfo
*additional_test_list
)
6408 struct imsm_dev
*dev
= get_imsm_dev(super
, a
->info
.container_member
);
6409 int idx
= get_imsm_disk_idx(dev
, slot
, -1);
6410 struct imsm_super
*mpb
= super
->anchor
;
6411 struct imsm_map
*map
;
6412 unsigned long long pos
;
6417 __u32 array_start
= 0;
6418 __u32 array_end
= 0;
6420 struct mdinfo
*test_list
;
6422 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
6423 /* If in this array, skip */
6424 for (d
= a
->info
.devs
; d
; d
= d
->next
)
6425 if (d
->state_fd
>= 0 &&
6426 d
->disk
.major
== dl
->major
&&
6427 d
->disk
.minor
== dl
->minor
) {
6428 dprintf("%x:%x already in array\n",
6429 dl
->major
, dl
->minor
);
6434 test_list
= additional_test_list
;
6436 if (test_list
->disk
.major
== dl
->major
&&
6437 test_list
->disk
.minor
== dl
->minor
) {
6438 dprintf("%x:%x already in additional test list\n",
6439 dl
->major
, dl
->minor
);
6442 test_list
= test_list
->next
;
6447 /* skip in use or failed drives */
6448 if (is_failed(&dl
->disk
) || idx
== dl
->index
||
6450 dprintf("%x:%x status (failed: %d index: %d)\n",
6451 dl
->major
, dl
->minor
, is_failed(&dl
->disk
), idx
);
6455 /* skip pure spares when we are looking for partially
6456 * assimilated drives
6458 if (dl
->index
== -1 && !activate_new
)
6461 /* Does this unused device have the requisite free space?
6462 * It needs to be able to cover all member volumes
6464 ex
= get_extents(super
, dl
);
6466 dprintf("cannot get extents\n");
6469 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
6470 dev
= get_imsm_dev(super
, i
);
6471 map
= get_imsm_map(dev
, 0);
6473 /* check if this disk is already a member of
6476 if (get_imsm_disk_slot(map
, dl
->index
) >= 0)
6482 array_start
= __le32_to_cpu(map
->pba_of_lba0
);
6483 array_end
= array_start
+
6484 __le32_to_cpu(map
->blocks_per_member
) - 1;
6487 /* check that we can start at pba_of_lba0 with
6488 * blocks_per_member of space
6490 if (array_start
>= pos
&& array_end
< ex
[j
].start
) {
6494 pos
= ex
[j
].start
+ ex
[j
].size
;
6496 } while (ex
[j
-1].size
);
6503 if (i
< mpb
->num_raid_devs
) {
6504 dprintf("%x:%x does not have %u to %u available\n",
6505 dl
->major
, dl
->minor
, array_start
, array_end
);
6516 static int imsm_rebuild_allowed(struct supertype
*cont
, int dev_idx
, int failed
)
6518 struct imsm_dev
*dev2
;
6519 struct imsm_map
*map
;
6525 dev2
= get_imsm_dev(cont
->sb
, dev_idx
);
6527 state
= imsm_check_degraded(cont
->sb
, dev2
, failed
);
6528 if (state
== IMSM_T_STATE_FAILED
) {
6529 map
= get_imsm_map(dev2
, 0);
6532 for (slot
= 0; slot
< map
->num_members
; slot
++) {
6534 * Check if failed disks are deleted from intel
6535 * disk list or are marked to be deleted
6537 idx
= get_imsm_disk_idx(dev2
, slot
, -1);
6538 idisk
= get_imsm_dl_disk(cont
->sb
, idx
);
6540 * Do not rebuild the array if failed disks
6541 * from failed sub-array are not removed from
6545 is_failed(&idisk
->disk
) &&
6546 (idisk
->action
!= DISK_REMOVE
))
6554 static struct mdinfo
*imsm_activate_spare(struct active_array
*a
,
6555 struct metadata_update
**updates
)
6558 * Find a device with unused free space and use it to replace a
6559 * failed/vacant region in an array. We replace failed regions one a
6560 * array at a time. The result is that a new spare disk will be added
6561 * to the first failed array and after the monitor has finished
6562 * propagating failures the remainder will be consumed.
6564 * FIXME add a capability for mdmon to request spares from another
6568 struct intel_super
*super
= a
->container
->sb
;
6569 int inst
= a
->info
.container_member
;
6570 struct imsm_dev
*dev
= get_imsm_dev(super
, inst
);
6571 struct imsm_map
*map
= get_imsm_map(dev
, 0);
6572 int failed
= a
->info
.array
.raid_disks
;
6573 struct mdinfo
*rv
= NULL
;
6576 struct metadata_update
*mu
;
6578 struct imsm_update_activate_spare
*u
;
6583 for (d
= a
->info
.devs
; d
; d
= d
->next
) {
6584 if ((d
->curr_state
& DS_FAULTY
) &&
6586 /* wait for Removal to happen */
6588 if (d
->state_fd
>= 0)
6592 dprintf("imsm: activate spare: inst=%d failed=%d (%d) level=%d\n",
6593 inst
, failed
, a
->info
.array
.raid_disks
, a
->info
.array
.level
);
6595 if (dev
->vol
.migr_state
&&
6596 dev
->vol
.migr_type
== MIGR_GEN_MIGR
)
6597 /* No repair during migration */
6600 if (a
->info
.array
.level
== 4)
6601 /* No repair for takeovered array
6602 * imsm doesn't support raid4
6606 if (imsm_check_degraded(super
, dev
, failed
) != IMSM_T_STATE_DEGRADED
)
6610 * If there are any failed disks check state of the other volume.
6611 * Block rebuild if the another one is failed until failed disks
6612 * are removed from container.
6615 dprintf("found failed disks in %s, check if there another"
6616 "failed sub-array.\n",
6618 /* check if states of the other volumes allow for rebuild */
6619 for (i
= 0; i
< super
->anchor
->num_raid_devs
; i
++) {
6621 allowed
= imsm_rebuild_allowed(a
->container
,
6629 /* For each slot, if it is not working, find a spare */
6630 for (i
= 0; i
< a
->info
.array
.raid_disks
; i
++) {
6631 for (d
= a
->info
.devs
; d
; d
= d
->next
)
6632 if (d
->disk
.raid_disk
== i
)
6634 dprintf("found %d: %p %x\n", i
, d
, d
?d
->curr_state
:0);
6635 if (d
&& (d
->state_fd
>= 0))
6639 * OK, this device needs recovery. Try to re-add the
6640 * previous occupant of this slot, if this fails see if
6641 * we can continue the assimilation of a spare that was
6642 * partially assimilated, finally try to activate a new
6645 dl
= imsm_readd(super
, i
, a
);
6647 dl
= imsm_add_spare(super
, i
, a
, 0, NULL
);
6649 dl
= imsm_add_spare(super
, i
, a
, 1, NULL
);
6653 /* found a usable disk with enough space */
6654 di
= malloc(sizeof(*di
));
6657 memset(di
, 0, sizeof(*di
));
6659 /* dl->index will be -1 in the case we are activating a
6660 * pristine spare. imsm_process_update() will create a
6661 * new index in this case. Once a disk is found to be
6662 * failed in all member arrays it is kicked from the
6665 di
->disk
.number
= dl
->index
;
6667 /* (ab)use di->devs to store a pointer to the device
6670 di
->devs
= (struct mdinfo
*) dl
;
6672 di
->disk
.raid_disk
= i
;
6673 di
->disk
.major
= dl
->major
;
6674 di
->disk
.minor
= dl
->minor
;
6676 di
->recovery_start
= 0;
6677 di
->data_offset
= __le32_to_cpu(map
->pba_of_lba0
);
6678 di
->component_size
= a
->info
.component_size
;
6679 di
->container_member
= inst
;
6680 super
->random
= random32();
6684 dprintf("%x:%x to be %d at %llu\n", dl
->major
, dl
->minor
,
6685 i
, di
->data_offset
);
6691 /* No spares found */
6693 /* Now 'rv' has a list of devices to return.
6694 * Create a metadata_update record to update the
6695 * disk_ord_tbl for the array
6697 mu
= malloc(sizeof(*mu
));
6699 mu
->buf
= malloc(sizeof(struct imsm_update_activate_spare
) * num_spares
);
6700 if (mu
->buf
== NULL
) {
6707 struct mdinfo
*n
= rv
->next
;
6716 mu
->space_list
= NULL
;
6717 mu
->len
= sizeof(struct imsm_update_activate_spare
) * num_spares
;
6718 mu
->next
= *updates
;
6719 u
= (struct imsm_update_activate_spare
*) mu
->buf
;
6721 for (di
= rv
; di
; di
= di
->next
) {
6722 u
->type
= update_activate_spare
;
6723 u
->dl
= (struct dl
*) di
->devs
;
6725 u
->slot
= di
->disk
.raid_disk
;
6736 static int disks_overlap(struct intel_super
*super
, int idx
, struct imsm_update_create_array
*u
)
6738 struct imsm_dev
*dev
= get_imsm_dev(super
, idx
);
6739 struct imsm_map
*map
= get_imsm_map(dev
, 0);
6740 struct imsm_map
*new_map
= get_imsm_map(&u
->dev
, 0);
6741 struct disk_info
*inf
= get_disk_info(u
);
6742 struct imsm_disk
*disk
;
6746 for (i
= 0; i
< map
->num_members
; i
++) {
6747 disk
= get_imsm_disk(super
, get_imsm_disk_idx(dev
, i
, -1));
6748 for (j
= 0; j
< new_map
->num_members
; j
++)
6749 if (serialcmp(disk
->serial
, inf
[j
].serial
) == 0)
6757 static struct dl
*get_disk_super(struct intel_super
*super
, int major
, int minor
)
6759 struct dl
*dl
= NULL
;
6760 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
6761 if ((dl
->major
== major
) && (dl
->minor
== minor
))
6766 static int remove_disk_super(struct intel_super
*super
, int major
, int minor
)
6768 struct dl
*prev
= NULL
;
6772 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
6773 if ((dl
->major
== major
) && (dl
->minor
== minor
)) {
6776 prev
->next
= dl
->next
;
6778 super
->disks
= dl
->next
;
6780 __free_imsm_disk(dl
);
6781 dprintf("%s: removed %x:%x\n",
6782 __func__
, major
, minor
);
6790 static void imsm_delete(struct intel_super
*super
, struct dl
**dlp
, unsigned index
);
6792 static int add_remove_disk_update(struct intel_super
*super
)
6794 int check_degraded
= 0;
6795 struct dl
*disk
= NULL
;
6796 /* add/remove some spares to/from the metadata/contrainer */
6797 while (super
->disk_mgmt_list
) {
6798 struct dl
*disk_cfg
;
6800 disk_cfg
= super
->disk_mgmt_list
;
6801 super
->disk_mgmt_list
= disk_cfg
->next
;
6802 disk_cfg
->next
= NULL
;
6804 if (disk_cfg
->action
== DISK_ADD
) {
6805 disk_cfg
->next
= super
->disks
;
6806 super
->disks
= disk_cfg
;
6808 dprintf("%s: added %x:%x\n",
6809 __func__
, disk_cfg
->major
,
6811 } else if (disk_cfg
->action
== DISK_REMOVE
) {
6812 dprintf("Disk remove action processed: %x.%x\n",
6813 disk_cfg
->major
, disk_cfg
->minor
);
6814 disk
= get_disk_super(super
,
6818 /* store action status */
6819 disk
->action
= DISK_REMOVE
;
6820 /* remove spare disks only */
6821 if (disk
->index
== -1) {
6822 remove_disk_super(super
,
6827 /* release allocate disk structure */
6828 __free_imsm_disk(disk_cfg
);
6831 return check_degraded
;
6835 static int apply_reshape_migration_update(struct imsm_update_reshape_migration
*u
,
6836 struct intel_super
*super
,
6839 struct intel_dev
*id
;
6840 void **tofree
= NULL
;
6843 dprintf("apply_reshape_migration_update()\n");
6844 if ((u
->subdev
< 0) ||
6846 dprintf("imsm: Error: Wrong subdev: %i\n", u
->subdev
);
6849 if ((space_list
== NULL
) || (*space_list
== NULL
)) {
6850 dprintf("imsm: Error: Memory is not allocated\n");
6854 for (id
= super
->devlist
; id
; id
= id
->next
) {
6855 if (id
->index
== (unsigned)u
->subdev
) {
6856 struct imsm_dev
*dev
= get_imsm_dev(super
, u
->subdev
);
6857 struct imsm_map
*map
;
6858 struct imsm_dev
*new_dev
=
6859 (struct imsm_dev
*)*space_list
;
6860 struct imsm_map
*migr_map
= get_imsm_map(dev
, 1);
6862 struct dl
*new_disk
;
6864 if (new_dev
== NULL
)
6866 *space_list
= **space_list
;
6867 memcpy(new_dev
, dev
, sizeof_imsm_dev(dev
, 0));
6868 map
= get_imsm_map(new_dev
, 0);
6870 dprintf("imsm: Error: migration in progress");
6874 to_state
= map
->map_state
;
6875 if ((u
->new_level
== 5) && (map
->raid_level
== 0)) {
6877 /* this should not happen */
6878 if (u
->new_disks
[0] < 0) {
6879 map
->failed_disk_num
=
6880 map
->num_members
- 1;
6881 to_state
= IMSM_T_STATE_DEGRADED
;
6883 to_state
= IMSM_T_STATE_NORMAL
;
6885 migrate(new_dev
, super
, to_state
, MIGR_GEN_MIGR
);
6886 if (u
->new_level
> -1)
6887 map
->raid_level
= u
->new_level
;
6888 migr_map
= get_imsm_map(new_dev
, 1);
6889 if ((u
->new_level
== 5) &&
6890 (migr_map
->raid_level
== 0)) {
6891 int ord
= map
->num_members
- 1;
6892 migr_map
->num_members
--;
6893 if (u
->new_disks
[0] < 0)
6894 ord
|= IMSM_ORD_REBUILD
;
6895 set_imsm_ord_tbl_ent(map
,
6896 map
->num_members
- 1,
6900 tofree
= (void **)dev
;
6902 /* update chunk size
6904 if (u
->new_chunksize
> 0)
6905 map
->blocks_per_strip
=
6906 __cpu_to_le16(u
->new_chunksize
* 2);
6910 if ((u
->new_level
!= 5) ||
6911 (migr_map
->raid_level
!= 0) ||
6912 (migr_map
->raid_level
== map
->raid_level
))
6915 if (u
->new_disks
[0] >= 0) {
6918 new_disk
= get_disk_super(super
,
6919 major(u
->new_disks
[0]),
6920 minor(u
->new_disks
[0]));
6921 dprintf("imsm: new disk for reshape is: %i:%i "
6922 "(%p, index = %i)\n",
6923 major(u
->new_disks
[0]),
6924 minor(u
->new_disks
[0]),
6925 new_disk
, new_disk
->index
);
6926 if (new_disk
== NULL
)
6927 goto error_disk_add
;
6929 new_disk
->index
= map
->num_members
- 1;
6930 /* slot to fill in autolayout
6932 new_disk
->raiddisk
= new_disk
->index
;
6933 new_disk
->disk
.status
|= CONFIGURED_DISK
;
6934 new_disk
->disk
.status
&= ~SPARE_DISK
;
6936 goto error_disk_add
;
6939 *tofree
= *space_list
;
6940 /* calculate new size
6942 imsm_set_array_size(new_dev
);
6949 *space_list
= tofree
;
6953 dprintf("Error: imsm: Cannot find disk.\n");
6958 static int apply_reshape_container_disks_update(struct imsm_update_reshape
*u
,
6959 struct intel_super
*super
,
6962 struct dl
*new_disk
;
6963 struct intel_dev
*id
;
6965 int delta_disks
= u
->new_raid_disks
- u
->old_raid_disks
;
6966 int disk_count
= u
->old_raid_disks
;
6967 void **tofree
= NULL
;
6968 int devices_to_reshape
= 1;
6969 struct imsm_super
*mpb
= super
->anchor
;
6971 unsigned int dev_id
;
6973 dprintf("imsm: apply_reshape_container_disks_update()\n");
6975 /* enable spares to use in array */
6976 for (i
= 0; i
< delta_disks
; i
++) {
6977 new_disk
= get_disk_super(super
,
6978 major(u
->new_disks
[i
]),
6979 minor(u
->new_disks
[i
]));
6980 dprintf("imsm: new disk for reshape is: %i:%i "
6981 "(%p, index = %i)\n",
6982 major(u
->new_disks
[i
]), minor(u
->new_disks
[i
]),
6983 new_disk
, new_disk
->index
);
6984 if ((new_disk
== NULL
) ||
6985 ((new_disk
->index
>= 0) &&
6986 (new_disk
->index
< u
->old_raid_disks
)))
6987 goto update_reshape_exit
;
6988 new_disk
->index
= disk_count
++;
6989 /* slot to fill in autolayout
6991 new_disk
->raiddisk
= new_disk
->index
;
6992 new_disk
->disk
.status
|=
6994 new_disk
->disk
.status
&= ~SPARE_DISK
;
6997 dprintf("imsm: volume set mpb->num_raid_devs = %i\n",
6998 mpb
->num_raid_devs
);
6999 /* manage changes in volume
7001 for (dev_id
= 0; dev_id
< mpb
->num_raid_devs
; dev_id
++) {
7002 void **sp
= *space_list
;
7003 struct imsm_dev
*newdev
;
7004 struct imsm_map
*newmap
, *oldmap
;
7006 for (id
= super
->devlist
; id
; id
= id
->next
) {
7007 if (id
->index
== dev_id
)
7016 /* Copy the dev, but not (all of) the map */
7017 memcpy(newdev
, id
->dev
, sizeof(*newdev
));
7018 oldmap
= get_imsm_map(id
->dev
, 0);
7019 newmap
= get_imsm_map(newdev
, 0);
7020 /* Copy the current map */
7021 memcpy(newmap
, oldmap
, sizeof_imsm_map(oldmap
));
7022 /* update one device only
7024 if (devices_to_reshape
) {
7025 dprintf("imsm: modifying subdev: %i\n",
7027 devices_to_reshape
--;
7028 newdev
->vol
.migr_state
= 1;
7029 newdev
->vol
.curr_migr_unit
= 0;
7030 newdev
->vol
.migr_type
= MIGR_GEN_MIGR
;
7031 newmap
->num_members
= u
->new_raid_disks
;
7032 for (i
= 0; i
< delta_disks
; i
++) {
7033 set_imsm_ord_tbl_ent(newmap
,
7034 u
->old_raid_disks
+ i
,
7035 u
->old_raid_disks
+ i
);
7037 /* New map is correct, now need to save old map
7039 newmap
= get_imsm_map(newdev
, 1);
7040 memcpy(newmap
, oldmap
, sizeof_imsm_map(oldmap
));
7042 imsm_set_array_size(newdev
);
7045 sp
= (void **)id
->dev
;
7050 /* Clear migration record */
7051 memset(super
->migr_rec
, 0, sizeof(struct migr_record
));
7054 *space_list
= tofree
;
7057 update_reshape_exit
:
7062 static int apply_takeover_update(struct imsm_update_takeover
*u
,
7063 struct intel_super
*super
,
7066 struct imsm_dev
*dev
= NULL
;
7067 struct intel_dev
*dv
;
7068 struct imsm_dev
*dev_new
;
7069 struct imsm_map
*map
;
7073 for (dv
= super
->devlist
; dv
; dv
= dv
->next
)
7074 if (dv
->index
== (unsigned int)u
->subarray
) {
7082 map
= get_imsm_map(dev
, 0);
7084 if (u
->direction
== R10_TO_R0
) {
7085 /* Number of failed disks must be half of initial disk number */
7086 if (imsm_count_failed(super
, dev
) != (map
->num_members
/ 2))
7089 /* iterate through devices to mark removed disks as spare */
7090 for (dm
= super
->disks
; dm
; dm
= dm
->next
) {
7091 if (dm
->disk
.status
& FAILED_DISK
) {
7092 int idx
= dm
->index
;
7093 /* update indexes on the disk list */
7094 /* FIXME this loop-with-the-loop looks wrong, I'm not convinced
7095 the index values will end up being correct.... NB */
7096 for (du
= super
->disks
; du
; du
= du
->next
)
7097 if (du
->index
> idx
)
7099 /* mark as spare disk */
7104 map
->num_members
= map
->num_members
/ 2;
7105 map
->map_state
= IMSM_T_STATE_NORMAL
;
7106 map
->num_domains
= 1;
7107 map
->raid_level
= 0;
7108 map
->failed_disk_num
= -1;
7111 if (u
->direction
== R0_TO_R10
) {
7113 /* update slots in current disk list */
7114 for (dm
= super
->disks
; dm
; dm
= dm
->next
) {
7118 /* create new *missing* disks */
7119 for (i
= 0; i
< map
->num_members
; i
++) {
7120 space
= *space_list
;
7123 *space_list
= *space
;
7125 memcpy(du
, super
->disks
, sizeof(*du
));
7129 du
->index
= (i
* 2) + 1;
7130 sprintf((char *)du
->disk
.serial
,
7131 " MISSING_%d", du
->index
);
7132 sprintf((char *)du
->serial
,
7133 "MISSING_%d", du
->index
);
7134 du
->next
= super
->missing
;
7135 super
->missing
= du
;
7137 /* create new dev and map */
7138 space
= *space_list
;
7141 *space_list
= *space
;
7142 dev_new
= (void *)space
;
7143 memcpy(dev_new
, dev
, sizeof(*dev
));
7144 /* update new map */
7145 map
= get_imsm_map(dev_new
, 0);
7146 map
->num_members
= map
->num_members
* 2;
7147 map
->map_state
= IMSM_T_STATE_DEGRADED
;
7148 map
->num_domains
= 2;
7149 map
->raid_level
= 1;
7150 /* replace dev<->dev_new */
7153 /* update disk order table */
7154 for (du
= super
->disks
; du
; du
= du
->next
)
7156 set_imsm_ord_tbl_ent(map
, du
->index
, du
->index
);
7157 for (du
= super
->missing
; du
; du
= du
->next
)
7158 if (du
->index
>= 0) {
7159 set_imsm_ord_tbl_ent(map
, du
->index
, du
->index
);
7160 mark_missing(dv
->dev
, &du
->disk
, du
->index
);
7166 static void imsm_process_update(struct supertype
*st
,
7167 struct metadata_update
*update
)
7170 * crack open the metadata_update envelope to find the update record
7171 * update can be one of:
7172 * update_reshape_container_disks - all the arrays in the container
7173 * are being reshaped to have more devices. We need to mark
7174 * the arrays for general migration and convert selected spares
7175 * into active devices.
7176 * update_activate_spare - a spare device has replaced a failed
7177 * device in an array, update the disk_ord_tbl. If this disk is
7178 * present in all member arrays then also clear the SPARE_DISK
7180 * update_create_array
7182 * update_rename_array
7183 * update_add_remove_disk
7185 struct intel_super
*super
= st
->sb
;
7186 struct imsm_super
*mpb
;
7187 enum imsm_update_type type
= *(enum imsm_update_type
*) update
->buf
;
7189 /* update requires a larger buf but the allocation failed */
7190 if (super
->next_len
&& !super
->next_buf
) {
7191 super
->next_len
= 0;
7195 if (super
->next_buf
) {
7196 memcpy(super
->next_buf
, super
->buf
, super
->len
);
7198 super
->len
= super
->next_len
;
7199 super
->buf
= super
->next_buf
;
7201 super
->next_len
= 0;
7202 super
->next_buf
= NULL
;
7205 mpb
= super
->anchor
;
7208 case update_general_migration_checkpoint
: {
7209 struct intel_dev
*id
;
7210 struct imsm_update_general_migration_checkpoint
*u
=
7211 (void *)update
->buf
;
7213 dprintf("imsm: process_update() "
7214 "for update_general_migration_checkpoint called\n");
7216 /* find device under general migration */
7217 for (id
= super
->devlist
; id
; id
= id
->next
) {
7218 if (is_gen_migration(id
->dev
)) {
7219 id
->dev
->vol
.curr_migr_unit
=
7220 __cpu_to_le32(u
->curr_migr_unit
);
7221 super
->updates_pending
++;
7226 case update_takeover
: {
7227 struct imsm_update_takeover
*u
= (void *)update
->buf
;
7228 if (apply_takeover_update(u
, super
, &update
->space_list
)) {
7229 imsm_update_version_info(super
);
7230 super
->updates_pending
++;
7235 case update_reshape_container_disks
: {
7236 struct imsm_update_reshape
*u
= (void *)update
->buf
;
7237 if (apply_reshape_container_disks_update(
7238 u
, super
, &update
->space_list
))
7239 super
->updates_pending
++;
7242 case update_reshape_migration
: {
7243 struct imsm_update_reshape_migration
*u
= (void *)update
->buf
;
7244 if (apply_reshape_migration_update(
7245 u
, super
, &update
->space_list
))
7246 super
->updates_pending
++;
7249 case update_activate_spare
: {
7250 struct imsm_update_activate_spare
*u
= (void *) update
->buf
;
7251 struct imsm_dev
*dev
= get_imsm_dev(super
, u
->array
);
7252 struct imsm_map
*map
= get_imsm_map(dev
, 0);
7253 struct imsm_map
*migr_map
;
7254 struct active_array
*a
;
7255 struct imsm_disk
*disk
;
7260 int victim
= get_imsm_disk_idx(dev
, u
->slot
, -1);
7263 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
7268 fprintf(stderr
, "error: imsm_activate_spare passed "
7269 "an unknown disk (index: %d)\n",
7274 super
->updates_pending
++;
7275 /* count failures (excluding rebuilds and the victim)
7276 * to determine map[0] state
7279 for (i
= 0; i
< map
->num_members
; i
++) {
7282 disk
= get_imsm_disk(super
,
7283 get_imsm_disk_idx(dev
, i
, -1));
7284 if (!disk
|| is_failed(disk
))
7288 /* adding a pristine spare, assign a new index */
7289 if (dl
->index
< 0) {
7290 dl
->index
= super
->anchor
->num_disks
;
7291 super
->anchor
->num_disks
++;
7294 disk
->status
|= CONFIGURED_DISK
;
7295 disk
->status
&= ~SPARE_DISK
;
7298 to_state
= imsm_check_degraded(super
, dev
, failed
);
7299 map
->map_state
= IMSM_T_STATE_DEGRADED
;
7300 migrate(dev
, super
, to_state
, MIGR_REBUILD
);
7301 migr_map
= get_imsm_map(dev
, 1);
7302 set_imsm_ord_tbl_ent(map
, u
->slot
, dl
->index
);
7303 set_imsm_ord_tbl_ent(migr_map
, u
->slot
, dl
->index
| IMSM_ORD_REBUILD
);
7305 /* update the family_num to mark a new container
7306 * generation, being careful to record the existing
7307 * family_num in orig_family_num to clean up after
7308 * earlier mdadm versions that neglected to set it.
7310 if (mpb
->orig_family_num
== 0)
7311 mpb
->orig_family_num
= mpb
->family_num
;
7312 mpb
->family_num
+= super
->random
;
7314 /* count arrays using the victim in the metadata */
7316 for (a
= st
->arrays
; a
; a
= a
->next
) {
7317 dev
= get_imsm_dev(super
, a
->info
.container_member
);
7318 map
= get_imsm_map(dev
, 0);
7320 if (get_imsm_disk_slot(map
, victim
) >= 0)
7324 /* delete the victim if it is no longer being
7330 /* We know that 'manager' isn't touching anything,
7331 * so it is safe to delete
7333 for (dlp
= &super
->disks
; *dlp
; dlp
= &(*dlp
)->next
)
7334 if ((*dlp
)->index
== victim
)
7337 /* victim may be on the missing list */
7339 for (dlp
= &super
->missing
; *dlp
; dlp
= &(*dlp
)->next
)
7340 if ((*dlp
)->index
== victim
)
7342 imsm_delete(super
, dlp
, victim
);
7346 case update_create_array
: {
7347 /* someone wants to create a new array, we need to be aware of
7348 * a few races/collisions:
7349 * 1/ 'Create' called by two separate instances of mdadm
7350 * 2/ 'Create' versus 'activate_spare': mdadm has chosen
7351 * devices that have since been assimilated via
7353 * In the event this update can not be carried out mdadm will
7354 * (FIX ME) notice that its update did not take hold.
7356 struct imsm_update_create_array
*u
= (void *) update
->buf
;
7357 struct intel_dev
*dv
;
7358 struct imsm_dev
*dev
;
7359 struct imsm_map
*map
, *new_map
;
7360 unsigned long long start
, end
;
7361 unsigned long long new_start
, new_end
;
7363 struct disk_info
*inf
;
7366 /* handle racing creates: first come first serve */
7367 if (u
->dev_idx
< mpb
->num_raid_devs
) {
7368 dprintf("%s: subarray %d already defined\n",
7369 __func__
, u
->dev_idx
);
7373 /* check update is next in sequence */
7374 if (u
->dev_idx
!= mpb
->num_raid_devs
) {
7375 dprintf("%s: can not create array %d expected index %d\n",
7376 __func__
, u
->dev_idx
, mpb
->num_raid_devs
);
7380 new_map
= get_imsm_map(&u
->dev
, 0);
7381 new_start
= __le32_to_cpu(new_map
->pba_of_lba0
);
7382 new_end
= new_start
+ __le32_to_cpu(new_map
->blocks_per_member
);
7383 inf
= get_disk_info(u
);
7385 /* handle activate_spare versus create race:
7386 * check to make sure that overlapping arrays do not include
7389 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
7390 dev
= get_imsm_dev(super
, i
);
7391 map
= get_imsm_map(dev
, 0);
7392 start
= __le32_to_cpu(map
->pba_of_lba0
);
7393 end
= start
+ __le32_to_cpu(map
->blocks_per_member
);
7394 if ((new_start
>= start
&& new_start
<= end
) ||
7395 (start
>= new_start
&& start
<= new_end
))
7400 if (disks_overlap(super
, i
, u
)) {
7401 dprintf("%s: arrays overlap\n", __func__
);
7406 /* check that prepare update was successful */
7407 if (!update
->space
) {
7408 dprintf("%s: prepare update failed\n", __func__
);
7412 /* check that all disks are still active before committing
7413 * changes. FIXME: could we instead handle this by creating a
7414 * degraded array? That's probably not what the user expects,
7415 * so better to drop this update on the floor.
7417 for (i
= 0; i
< new_map
->num_members
; i
++) {
7418 dl
= serial_to_dl(inf
[i
].serial
, super
);
7420 dprintf("%s: disk disappeared\n", __func__
);
7425 super
->updates_pending
++;
7427 /* convert spares to members and fixup ord_tbl */
7428 for (i
= 0; i
< new_map
->num_members
; i
++) {
7429 dl
= serial_to_dl(inf
[i
].serial
, super
);
7430 if (dl
->index
== -1) {
7431 dl
->index
= mpb
->num_disks
;
7433 dl
->disk
.status
|= CONFIGURED_DISK
;
7434 dl
->disk
.status
&= ~SPARE_DISK
;
7436 set_imsm_ord_tbl_ent(new_map
, i
, dl
->index
);
7441 update
->space
= NULL
;
7442 imsm_copy_dev(dev
, &u
->dev
);
7443 dv
->index
= u
->dev_idx
;
7444 dv
->next
= super
->devlist
;
7445 super
->devlist
= dv
;
7446 mpb
->num_raid_devs
++;
7448 imsm_update_version_info(super
);
7451 /* mdmon knows how to release update->space, but not
7452 * ((struct intel_dev *) update->space)->dev
7454 if (update
->space
) {
7460 case update_kill_array
: {
7461 struct imsm_update_kill_array
*u
= (void *) update
->buf
;
7462 int victim
= u
->dev_idx
;
7463 struct active_array
*a
;
7464 struct intel_dev
**dp
;
7465 struct imsm_dev
*dev
;
7467 /* sanity check that we are not affecting the uuid of
7468 * active arrays, or deleting an active array
7470 * FIXME when immutable ids are available, but note that
7471 * we'll also need to fixup the invalidated/active
7472 * subarray indexes in mdstat
7474 for (a
= st
->arrays
; a
; a
= a
->next
)
7475 if (a
->info
.container_member
>= victim
)
7477 /* by definition if mdmon is running at least one array
7478 * is active in the container, so checking
7479 * mpb->num_raid_devs is just extra paranoia
7481 dev
= get_imsm_dev(super
, victim
);
7482 if (a
|| !dev
|| mpb
->num_raid_devs
== 1) {
7483 dprintf("failed to delete subarray-%d\n", victim
);
7487 for (dp
= &super
->devlist
; *dp
;)
7488 if ((*dp
)->index
== (unsigned)super
->current_vol
) {
7491 if ((*dp
)->index
> (unsigned)victim
)
7495 mpb
->num_raid_devs
--;
7496 super
->updates_pending
++;
7499 case update_rename_array
: {
7500 struct imsm_update_rename_array
*u
= (void *) update
->buf
;
7501 char name
[MAX_RAID_SERIAL_LEN
+1];
7502 int target
= u
->dev_idx
;
7503 struct active_array
*a
;
7504 struct imsm_dev
*dev
;
7506 /* sanity check that we are not affecting the uuid of
7509 snprintf(name
, MAX_RAID_SERIAL_LEN
, "%s", (char *) u
->name
);
7510 name
[MAX_RAID_SERIAL_LEN
] = '\0';
7511 for (a
= st
->arrays
; a
; a
= a
->next
)
7512 if (a
->info
.container_member
== target
)
7514 dev
= get_imsm_dev(super
, u
->dev_idx
);
7515 if (a
|| !dev
|| !check_name(super
, name
, 1)) {
7516 dprintf("failed to rename subarray-%d\n", target
);
7520 snprintf((char *) dev
->volume
, MAX_RAID_SERIAL_LEN
, "%s", name
);
7521 super
->updates_pending
++;
7524 case update_add_remove_disk
: {
7525 /* we may be able to repair some arrays if disks are
7526 * being added, check teh status of add_remove_disk
7527 * if discs has been added.
7529 if (add_remove_disk_update(super
)) {
7530 struct active_array
*a
;
7532 super
->updates_pending
++;
7533 for (a
= st
->arrays
; a
; a
= a
->next
)
7534 a
->check_degraded
= 1;
7539 fprintf(stderr
, "error: unsuported process update type:"
7540 "(type: %d)\n", type
);
7544 static struct mdinfo
*get_spares_for_grow(struct supertype
*st
);
7546 static void imsm_prepare_update(struct supertype
*st
,
7547 struct metadata_update
*update
)
7550 * Allocate space to hold new disk entries, raid-device entries or a new
7551 * mpb if necessary. The manager synchronously waits for updates to
7552 * complete in the monitor, so new mpb buffers allocated here can be
7553 * integrated by the monitor thread without worrying about live pointers
7554 * in the manager thread.
7556 enum imsm_update_type type
= *(enum imsm_update_type
*) update
->buf
;
7557 struct intel_super
*super
= st
->sb
;
7558 struct imsm_super
*mpb
= super
->anchor
;
7563 case update_general_migration_checkpoint
:
7564 dprintf("imsm: prepare_update() "
7565 "for update_general_migration_checkpoint called\n");
7567 case update_takeover
: {
7568 struct imsm_update_takeover
*u
= (void *)update
->buf
;
7569 if (u
->direction
== R0_TO_R10
) {
7570 void **tail
= (void **)&update
->space_list
;
7571 struct imsm_dev
*dev
= get_imsm_dev(super
, u
->subarray
);
7572 struct imsm_map
*map
= get_imsm_map(dev
, 0);
7573 int num_members
= map
->num_members
;
7577 /* allocate memory for added disks */
7578 for (i
= 0; i
< num_members
; i
++) {
7579 size
= sizeof(struct dl
);
7580 space
= malloc(size
);
7589 /* allocate memory for new device */
7590 size
= sizeof_imsm_dev(super
->devlist
->dev
, 0) +
7591 (num_members
* sizeof(__u32
));
7592 space
= malloc(size
);
7601 len
= disks_to_mpb_size(num_members
* 2);
7603 /* if allocation didn't success, free buffer */
7604 while (update
->space_list
) {
7605 void **sp
= update
->space_list
;
7606 update
->space_list
= *sp
;
7614 case update_reshape_container_disks
: {
7615 /* Every raid device in the container is about to
7616 * gain some more devices, and we will enter a
7618 * So each 'imsm_map' will be bigger, and the imsm_vol
7619 * will now hold 2 of them.
7620 * Thus we need new 'struct imsm_dev' allocations sized
7621 * as sizeof_imsm_dev but with more devices in both maps.
7623 struct imsm_update_reshape
*u
= (void *)update
->buf
;
7624 struct intel_dev
*dl
;
7625 void **space_tail
= (void**)&update
->space_list
;
7627 dprintf("imsm: imsm_prepare_update() for update_reshape\n");
7629 for (dl
= super
->devlist
; dl
; dl
= dl
->next
) {
7630 int size
= sizeof_imsm_dev(dl
->dev
, 1);
7632 if (u
->new_raid_disks
> u
->old_raid_disks
)
7633 size
+= sizeof(__u32
)*2*
7634 (u
->new_raid_disks
- u
->old_raid_disks
);
7643 len
= disks_to_mpb_size(u
->new_raid_disks
);
7644 dprintf("New anchor length is %llu\n", (unsigned long long)len
);
7647 case update_reshape_migration
: {
7648 /* for migration level 0->5 we need to add disks
7649 * so the same as for container operation we will copy
7650 * device to the bigger location.
7651 * in memory prepared device and new disk area are prepared
7652 * for usage in process update
7654 struct imsm_update_reshape_migration
*u
= (void *)update
->buf
;
7655 struct intel_dev
*id
;
7656 void **space_tail
= (void **)&update
->space_list
;
7659 int current_level
= -1;
7661 dprintf("imsm: imsm_prepare_update() for update_reshape\n");
7663 /* add space for bigger array in update
7665 for (id
= super
->devlist
; id
; id
= id
->next
) {
7666 if (id
->index
== (unsigned)u
->subdev
) {
7667 size
= sizeof_imsm_dev(id
->dev
, 1);
7668 if (u
->new_raid_disks
> u
->old_raid_disks
)
7669 size
+= sizeof(__u32
)*2*
7670 (u
->new_raid_disks
- u
->old_raid_disks
);
7680 if (update
->space_list
== NULL
)
7683 /* add space for disk in update
7685 size
= sizeof(struct dl
);
7688 free(update
->space_list
);
7689 update
->space_list
= NULL
;
7696 /* add spare device to update
7698 for (id
= super
->devlist
; id
; id
= id
->next
)
7699 if (id
->index
== (unsigned)u
->subdev
) {
7700 struct imsm_dev
*dev
;
7701 struct imsm_map
*map
;
7703 dev
= get_imsm_dev(super
, u
->subdev
);
7704 map
= get_imsm_map(dev
, 0);
7705 current_level
= map
->raid_level
;
7708 if ((u
->new_level
== 5) && (u
->new_level
!= current_level
)) {
7709 struct mdinfo
*spares
;
7711 spares
= get_spares_for_grow(st
);
7719 makedev(dev
->disk
.major
,
7721 dl
= get_disk_super(super
,
7724 dl
->index
= u
->old_raid_disks
;
7730 len
= disks_to_mpb_size(u
->new_raid_disks
);
7731 dprintf("New anchor length is %llu\n", (unsigned long long)len
);
7734 case update_create_array
: {
7735 struct imsm_update_create_array
*u
= (void *) update
->buf
;
7736 struct intel_dev
*dv
;
7737 struct imsm_dev
*dev
= &u
->dev
;
7738 struct imsm_map
*map
= get_imsm_map(dev
, 0);
7740 struct disk_info
*inf
;
7744 inf
= get_disk_info(u
);
7745 len
= sizeof_imsm_dev(dev
, 1);
7746 /* allocate a new super->devlist entry */
7747 dv
= malloc(sizeof(*dv
));
7749 dv
->dev
= malloc(len
);
7754 update
->space
= NULL
;
7758 /* count how many spares will be converted to members */
7759 for (i
= 0; i
< map
->num_members
; i
++) {
7760 dl
= serial_to_dl(inf
[i
].serial
, super
);
7762 /* hmm maybe it failed?, nothing we can do about
7767 if (count_memberships(dl
, super
) == 0)
7770 len
+= activate
* sizeof(struct imsm_disk
);
7777 /* check if we need a larger metadata buffer */
7778 if (super
->next_buf
)
7779 buf_len
= super
->next_len
;
7781 buf_len
= super
->len
;
7783 if (__le32_to_cpu(mpb
->mpb_size
) + len
> buf_len
) {
7784 /* ok we need a larger buf than what is currently allocated
7785 * if this allocation fails process_update will notice that
7786 * ->next_len is set and ->next_buf is NULL
7788 buf_len
= ROUND_UP(__le32_to_cpu(mpb
->mpb_size
) + len
, 512);
7789 if (super
->next_buf
)
7790 free(super
->next_buf
);
7792 super
->next_len
= buf_len
;
7793 if (posix_memalign(&super
->next_buf
, 512, buf_len
) == 0)
7794 memset(super
->next_buf
, 0, buf_len
);
7796 super
->next_buf
= NULL
;
7800 /* must be called while manager is quiesced */
7801 static void imsm_delete(struct intel_super
*super
, struct dl
**dlp
, unsigned index
)
7803 struct imsm_super
*mpb
= super
->anchor
;
7805 struct imsm_dev
*dev
;
7806 struct imsm_map
*map
;
7807 int i
, j
, num_members
;
7810 dprintf("%s: deleting device[%d] from imsm_super\n",
7813 /* shift all indexes down one */
7814 for (iter
= super
->disks
; iter
; iter
= iter
->next
)
7815 if (iter
->index
> (int)index
)
7817 for (iter
= super
->missing
; iter
; iter
= iter
->next
)
7818 if (iter
->index
> (int)index
)
7821 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
7822 dev
= get_imsm_dev(super
, i
);
7823 map
= get_imsm_map(dev
, 0);
7824 num_members
= map
->num_members
;
7825 for (j
= 0; j
< num_members
; j
++) {
7826 /* update ord entries being careful not to propagate
7827 * ord-flags to the first map
7829 ord
= get_imsm_ord_tbl_ent(dev
, j
, -1);
7831 if (ord_to_idx(ord
) <= index
)
7834 map
= get_imsm_map(dev
, 0);
7835 set_imsm_ord_tbl_ent(map
, j
, ord_to_idx(ord
- 1));
7836 map
= get_imsm_map(dev
, 1);
7838 set_imsm_ord_tbl_ent(map
, j
, ord
- 1);
7843 super
->updates_pending
++;
7845 struct dl
*dl
= *dlp
;
7847 *dlp
= (*dlp
)->next
;
7848 __free_imsm_disk(dl
);
7851 #endif /* MDASSEMBLE */
7852 /*******************************************************************************
7853 * Function: open_backup_targets
7854 * Description: Function opens file descriptors for all devices given in
7857 * info : general array info
7858 * raid_disks : number of disks
7859 * raid_fds : table of device's file descriptors
7863 ******************************************************************************/
7864 int open_backup_targets(struct mdinfo
*info
, int raid_disks
, int *raid_fds
)
7868 for (sd
= info
->devs
; sd
; sd
= sd
->next
) {
7871 if (sd
->disk
.state
& (1<<MD_DISK_FAULTY
)) {
7872 dprintf("disk is faulty!!\n");
7876 if ((sd
->disk
.raid_disk
>= raid_disks
) ||
7877 (sd
->disk
.raid_disk
< 0))
7880 dn
= map_dev(sd
->disk
.major
,
7882 raid_fds
[sd
->disk
.raid_disk
] = dev_open(dn
, O_RDWR
);
7883 if (raid_fds
[sd
->disk
.raid_disk
] < 0) {
7884 fprintf(stderr
, "cannot open component\n");
7892 /*******************************************************************************
7893 * Function: init_migr_record_imsm
7894 * Description: Function inits imsm migration record
7896 * super : imsm internal array info
7897 * dev : device under migration
7898 * info : general array info to find the smallest device
7901 ******************************************************************************/
7902 void init_migr_record_imsm(struct supertype
*st
, struct imsm_dev
*dev
,
7903 struct mdinfo
*info
)
7905 struct intel_super
*super
= st
->sb
;
7906 struct migr_record
*migr_rec
= super
->migr_rec
;
7908 unsigned long long dsize
, dev_sectors
;
7909 long long unsigned min_dev_sectors
= -1LLU;
7913 struct imsm_map
*map_dest
= get_imsm_map(dev
, 0);
7914 struct imsm_map
*map_src
= get_imsm_map(dev
, 1);
7915 unsigned long long num_migr_units
;
7916 unsigned long long array_blocks
;
7918 memset(migr_rec
, 0, sizeof(struct migr_record
));
7919 migr_rec
->family_num
= __cpu_to_le32(super
->anchor
->family_num
);
7921 /* only ascending reshape supported now */
7922 migr_rec
->ascending_migr
= __cpu_to_le32(1);
7924 migr_rec
->dest_depth_per_unit
= GEN_MIGR_AREA_SIZE
/
7925 max(map_dest
->blocks_per_strip
, map_src
->blocks_per_strip
);
7926 migr_rec
->dest_depth_per_unit
*= map_dest
->blocks_per_strip
;
7927 new_data_disks
= imsm_num_data_members(dev
, 0);
7928 migr_rec
->blocks_per_unit
=
7929 __cpu_to_le32(migr_rec
->dest_depth_per_unit
* new_data_disks
);
7930 migr_rec
->dest_depth_per_unit
=
7931 __cpu_to_le32(migr_rec
->dest_depth_per_unit
);
7932 array_blocks
= info
->component_size
* new_data_disks
;
7934 array_blocks
/ __le32_to_cpu(migr_rec
->blocks_per_unit
);
7936 if (array_blocks
% __le32_to_cpu(migr_rec
->blocks_per_unit
))
7938 migr_rec
->num_migr_units
= __cpu_to_le32(num_migr_units
);
7940 migr_rec
->post_migr_vol_cap
= dev
->size_low
;
7941 migr_rec
->post_migr_vol_cap_hi
= dev
->size_high
;
7944 /* Find the smallest dev */
7945 for (sd
= info
->devs
; sd
; sd
= sd
->next
) {
7946 sprintf(nm
, "%d:%d", sd
->disk
.major
, sd
->disk
.minor
);
7947 fd
= dev_open(nm
, O_RDONLY
);
7950 get_dev_size(fd
, NULL
, &dsize
);
7951 dev_sectors
= dsize
/ 512;
7952 if (dev_sectors
< min_dev_sectors
)
7953 min_dev_sectors
= dev_sectors
;
7956 migr_rec
->ckpt_area_pba
= __cpu_to_le32(min_dev_sectors
-
7957 RAID_DISK_RESERVED_BLOCKS_IMSM_HI
);
7959 write_imsm_migr_rec(st
);
7964 /*******************************************************************************
7965 * Function: save_backup_imsm
7966 * Description: Function saves critical data stripes to Migration Copy Area
7967 * and updates the current migration unit status.
7968 * Use restore_stripes() to form a destination stripe,
7969 * and to write it to the Copy Area.
7971 * st : supertype information
7972 * dev : imsm device that backup is saved for
7973 * info : general array info
7974 * buf : input buffer
7975 * length : length of data to backup (blocks_per_unit)
7979 ******************************************************************************/
7980 int save_backup_imsm(struct supertype
*st
,
7981 struct imsm_dev
*dev
,
7982 struct mdinfo
*info
,
7987 struct intel_super
*super
= st
->sb
;
7988 unsigned long long *target_offsets
= NULL
;
7989 int *targets
= NULL
;
7991 struct imsm_map
*map_dest
= get_imsm_map(dev
, 0);
7992 int new_disks
= map_dest
->num_members
;
7993 int dest_layout
= 0;
7995 unsigned long long start
;
7996 int data_disks
= imsm_num_data_members(dev
, 0);
7998 targets
= malloc(new_disks
* sizeof(int));
8002 for (i
= 0; i
< new_disks
; i
++)
8005 target_offsets
= malloc(new_disks
* sizeof(unsigned long long));
8006 if (!target_offsets
)
8009 start
= info
->reshape_progress
* 512;
8010 for (i
= 0; i
< new_disks
; i
++) {
8011 target_offsets
[i
] = (unsigned long long)
8012 __le32_to_cpu(super
->migr_rec
->ckpt_area_pba
) * 512;
8013 /* move back copy area adderss, it will be moved forward
8014 * in restore_stripes() using start input variable
8016 target_offsets
[i
] -= start
/data_disks
;
8019 if (open_backup_targets(info
, new_disks
, targets
))
8022 dest_layout
= imsm_level_to_layout(map_dest
->raid_level
);
8023 dest_chunk
= __le16_to_cpu(map_dest
->blocks_per_strip
) * 512;
8025 if (restore_stripes(targets
, /* list of dest devices */
8026 target_offsets
, /* migration record offsets */
8029 map_dest
->raid_level
,
8031 -1, /* source backup file descriptor */
8032 0, /* input buf offset
8033 * always 0 buf is already offseted */
8037 fprintf(stderr
, Name
": Error restoring stripes\n");
8045 for (i
= 0; i
< new_disks
; i
++)
8046 if (targets
[i
] >= 0)
8050 free(target_offsets
);
8055 /*******************************************************************************
8056 * Function: save_checkpoint_imsm
8057 * Description: Function called for current unit status update
8058 * in the migration record. It writes it to disk.
8060 * super : imsm internal array info
8061 * info : general array info
8065 * 2: failure, means no valid migration record
8066 * / no general migration in progress /
8067 ******************************************************************************/
8068 int save_checkpoint_imsm(struct supertype
*st
, struct mdinfo
*info
, int state
)
8070 struct intel_super
*super
= st
->sb
;
8071 unsigned long long blocks_per_unit
;
8072 unsigned long long curr_migr_unit
;
8074 if (load_imsm_migr_rec(super
, info
) != 0) {
8075 dprintf("imsm: ERROR: Cannot read migration record "
8076 "for checkpoint save.\n");
8080 blocks_per_unit
= __le32_to_cpu(super
->migr_rec
->blocks_per_unit
);
8081 if (blocks_per_unit
== 0) {
8082 dprintf("imsm: no migration in progress.\n");
8085 curr_migr_unit
= info
->reshape_progress
/ blocks_per_unit
;
8086 /* check if array is alligned to copy area
8087 * if it is not alligned, add one to current migration unit value
8088 * this can happend on array reshape finish only
8090 if (info
->reshape_progress
% blocks_per_unit
)
8093 super
->migr_rec
->curr_migr_unit
=
8094 __cpu_to_le32(curr_migr_unit
);
8095 super
->migr_rec
->rec_status
= __cpu_to_le32(state
);
8096 super
->migr_rec
->dest_1st_member_lba
=
8097 __cpu_to_le32(curr_migr_unit
*
8098 __le32_to_cpu(super
->migr_rec
->dest_depth_per_unit
));
8099 if (write_imsm_migr_rec(st
) < 0) {
8100 dprintf("imsm: Cannot write migration record "
8101 "outside backup area\n");
8108 /*******************************************************************************
8109 * Function: recover_backup_imsm
8110 * Description: Function recovers critical data from the Migration Copy Area
8111 * while assembling an array.
8113 * super : imsm internal array info
8114 * info : general array info
8116 * 0 : success (or there is no data to recover)
8118 ******************************************************************************/
8119 int recover_backup_imsm(struct supertype
*st
, struct mdinfo
*info
)
8121 struct intel_super
*super
= st
->sb
;
8122 struct migr_record
*migr_rec
= super
->migr_rec
;
8123 struct imsm_map
*map_dest
= NULL
;
8124 struct intel_dev
*id
= NULL
;
8125 unsigned long long read_offset
;
8126 unsigned long long write_offset
;
8128 int *targets
= NULL
;
8129 int new_disks
, i
, err
;
8132 unsigned long curr_migr_unit
= __le32_to_cpu(migr_rec
->curr_migr_unit
);
8133 unsigned long num_migr_units
= __le32_to_cpu(migr_rec
->num_migr_units
);
8135 int skipped_disks
= 0;
8136 int max_degradation
;
8138 err
= sysfs_get_str(info
, NULL
, "array_state", (char *)buffer
, 20);
8142 /* recover data only during assemblation */
8143 if (strncmp(buffer
, "inactive", 8) != 0)
8145 /* no data to recover */
8146 if (__le32_to_cpu(migr_rec
->rec_status
) == UNIT_SRC_NORMAL
)
8148 if (curr_migr_unit
>= num_migr_units
)
8151 /* find device during reshape */
8152 for (id
= super
->devlist
; id
; id
= id
->next
)
8153 if (is_gen_migration(id
->dev
))
8158 map_dest
= get_imsm_map(id
->dev
, 0);
8159 new_disks
= map_dest
->num_members
;
8160 max_degradation
= new_disks
- imsm_num_data_members(id
->dev
, 0);
8162 read_offset
= (unsigned long long)
8163 __le32_to_cpu(migr_rec
->ckpt_area_pba
) * 512;
8165 write_offset
= ((unsigned long long)
8166 __le32_to_cpu(migr_rec
->dest_1st_member_lba
) +
8167 __le32_to_cpu(map_dest
->pba_of_lba0
)) * 512;
8169 unit_len
= __le32_to_cpu(migr_rec
->dest_depth_per_unit
) * 512;
8170 if (posix_memalign((void **)&buf
, 512, unit_len
) != 0)
8172 targets
= malloc(new_disks
* sizeof(int));
8176 open_backup_targets(info
, new_disks
, targets
);
8178 for (i
= 0; i
< new_disks
; i
++) {
8179 if (targets
[i
] < 0) {
8183 if (lseek64(targets
[i
], read_offset
, SEEK_SET
) < 0) {
8185 Name
": Cannot seek to block: %s\n",
8189 if ((unsigned)read(targets
[i
], buf
, unit_len
) != unit_len
) {
8191 Name
": Cannot read copy area block: %s\n",
8195 if (lseek64(targets
[i
], write_offset
, SEEK_SET
) < 0) {
8197 Name
": Cannot seek to block: %s\n",
8201 if ((unsigned)write(targets
[i
], buf
, unit_len
) != unit_len
) {
8203 Name
": Cannot restore block: %s\n",
8209 if (skipped_disks
> max_degradation
) {
8211 Name
": Cannot restore data from backup."
8212 " Too many failed disks\n");
8216 if (save_checkpoint_imsm(st
, info
, UNIT_SRC_NORMAL
)) {
8217 /* ignore error == 2, this can mean end of reshape here
8219 dprintf("imsm: Cannot write checkpoint to "
8220 "migration record (UNIT_SRC_NORMAL) during restart\n");
8226 for (i
= 0; i
< new_disks
; i
++)
8235 static char disk_by_path
[] = "/dev/disk/by-path/";
8237 static const char *imsm_get_disk_controller_domain(const char *path
)
8239 char disk_path
[PATH_MAX
];
8243 strncpy(disk_path
, disk_by_path
, PATH_MAX
- 1);
8244 strncat(disk_path
, path
, PATH_MAX
- strlen(disk_path
) - 1);
8245 if (stat(disk_path
, &st
) == 0) {
8246 struct sys_dev
* hba
;
8249 path
= devt_to_devpath(st
.st_rdev
);
8252 hba
= find_disk_attached_hba(-1, path
);
8253 if (hba
&& hba
->type
== SYS_DEV_SAS
)
8255 else if (hba
&& hba
->type
== SYS_DEV_SATA
)
8259 dprintf("path: %s hba: %s attached: %s\n",
8260 path
, (hba
) ? hba
->path
: "NULL", drv
);
8268 static int imsm_find_array_minor_by_subdev(int subdev
, int container
, int *minor
)
8270 char subdev_name
[20];
8271 struct mdstat_ent
*mdstat
;
8273 sprintf(subdev_name
, "%d", subdev
);
8274 mdstat
= mdstat_by_subdev(subdev_name
, container
);
8278 *minor
= mdstat
->devnum
;
8279 free_mdstat(mdstat
);
8283 static int imsm_reshape_is_allowed_on_container(struct supertype
*st
,
8284 struct geo_params
*geo
,
8285 int *old_raid_disks
)
8287 /* currently we only support increasing the number of devices
8288 * for a container. This increases the number of device for each
8289 * member array. They must all be RAID0 or RAID5.
8292 struct mdinfo
*info
, *member
;
8293 int devices_that_can_grow
= 0;
8295 dprintf("imsm: imsm_reshape_is_allowed_on_container(ENTER): "
8296 "st->devnum = (%i)\n",
8299 if (geo
->size
!= -1 ||
8300 geo
->level
!= UnSet
||
8301 geo
->layout
!= UnSet
||
8302 geo
->chunksize
!= 0 ||
8303 geo
->raid_disks
== UnSet
) {
8304 dprintf("imsm: Container operation is allowed for "
8305 "raid disks number change only.\n");
8309 info
= container_content_imsm(st
, NULL
);
8310 for (member
= info
; member
; member
= member
->next
) {
8314 dprintf("imsm: checking device_num: %i\n",
8315 member
->container_member
);
8317 if (geo
->raid_disks
<= member
->array
.raid_disks
) {
8318 /* we work on container for Online Capacity Expansion
8319 * only so raid_disks has to grow
8321 dprintf("imsm: for container operation raid disks "
8322 "increase is required\n");
8326 if ((info
->array
.level
!= 0) &&
8327 (info
->array
.level
!= 5)) {
8328 /* we cannot use this container with other raid level
8330 dprintf("imsm: for container operation wrong"
8331 " raid level (%i) detected\n",
8335 /* check for platform support
8336 * for this raid level configuration
8338 struct intel_super
*super
= st
->sb
;
8339 if (!is_raid_level_supported(super
->orom
,
8340 member
->array
.level
,
8342 dprintf("platform does not support raid%d with"
8346 geo
->raid_disks
> 1 ? "s" : "");
8349 /* check if component size is aligned to chunk size
8351 if (info
->component_size
%
8352 (info
->array
.chunk_size
/512)) {
8353 dprintf("Component size is not aligned to "
8359 if (*old_raid_disks
&&
8360 info
->array
.raid_disks
!= *old_raid_disks
)
8362 *old_raid_disks
= info
->array
.raid_disks
;
8364 /* All raid5 and raid0 volumes in container
8365 * have to be ready for Online Capacity Expansion
8366 * so they need to be assembled. We have already
8367 * checked that no recovery etc is happening.
8369 result
= imsm_find_array_minor_by_subdev(member
->container_member
,
8373 dprintf("imsm: cannot find array\n");
8376 devices_that_can_grow
++;
8379 if (!member
&& devices_that_can_grow
)
8383 dprintf("\tContainer operation allowed\n");
8385 dprintf("\tError: %i\n", ret_val
);
8390 /* Function: get_spares_for_grow
8391 * Description: Allocates memory and creates list of spare devices
8392 * avaliable in container. Checks if spare drive size is acceptable.
8393 * Parameters: Pointer to the supertype structure
8394 * Returns: Pointer to the list of spare devices (mdinfo structure) on success,
8397 static struct mdinfo
*get_spares_for_grow(struct supertype
*st
)
8399 unsigned long long min_size
= min_acceptable_spare_size_imsm(st
);
8400 return container_choose_spares(st
, min_size
, NULL
, NULL
, NULL
, 0);
8403 /******************************************************************************
8404 * function: imsm_create_metadata_update_for_reshape
8405 * Function creates update for whole IMSM container.
8407 ******************************************************************************/
8408 static int imsm_create_metadata_update_for_reshape(
8409 struct supertype
*st
,
8410 struct geo_params
*geo
,
8412 struct imsm_update_reshape
**updatep
)
8414 struct intel_super
*super
= st
->sb
;
8415 struct imsm_super
*mpb
= super
->anchor
;
8416 int update_memory_size
= 0;
8417 struct imsm_update_reshape
*u
= NULL
;
8418 struct mdinfo
*spares
= NULL
;
8420 int delta_disks
= 0;
8423 dprintf("imsm_update_metadata_for_reshape(enter) raid_disks = %i\n",
8426 delta_disks
= geo
->raid_disks
- old_raid_disks
;
8428 /* size of all update data without anchor */
8429 update_memory_size
= sizeof(struct imsm_update_reshape
);
8431 /* now add space for spare disks that we need to add. */
8432 update_memory_size
+= sizeof(u
->new_disks
[0]) * (delta_disks
- 1);
8434 u
= calloc(1, update_memory_size
);
8437 "cannot get memory for imsm_update_reshape update\n");
8440 u
->type
= update_reshape_container_disks
;
8441 u
->old_raid_disks
= old_raid_disks
;
8442 u
->new_raid_disks
= geo
->raid_disks
;
8444 /* now get spare disks list
8446 spares
= get_spares_for_grow(st
);
8449 || delta_disks
> spares
->array
.spare_disks
) {
8450 fprintf(stderr
, Name
": imsm: ERROR: Cannot get spare devices "
8451 "for %s.\n", geo
->dev_name
);
8456 /* we have got spares
8457 * update disk list in imsm_disk list table in anchor
8459 dprintf("imsm: %i spares are available.\n\n",
8460 spares
->array
.spare_disks
);
8463 for (i
= 0; i
< delta_disks
; i
++) {
8468 u
->new_disks
[i
] = makedev(dev
->disk
.major
,
8470 dl
= get_disk_super(super
, dev
->disk
.major
, dev
->disk
.minor
);
8471 dl
->index
= mpb
->num_disks
;
8481 dprintf("imsm: reshape update preparation :");
8482 if (i
== delta_disks
) {
8485 return update_memory_size
;
8488 dprintf(" Error\n");
8493 /******************************************************************************
8494 * function: imsm_create_metadata_update_for_migration()
8495 * Creates update for IMSM array.
8497 ******************************************************************************/
8498 static int imsm_create_metadata_update_for_migration(
8499 struct supertype
*st
,
8500 struct geo_params
*geo
,
8501 struct imsm_update_reshape_migration
**updatep
)
8503 struct intel_super
*super
= st
->sb
;
8504 int update_memory_size
= 0;
8505 struct imsm_update_reshape_migration
*u
= NULL
;
8506 struct imsm_dev
*dev
;
8507 int previous_level
= -1;
8509 dprintf("imsm_create_metadata_update_for_migration(enter)"
8510 " New Level = %i\n", geo
->level
);
8512 /* size of all update data without anchor */
8513 update_memory_size
= sizeof(struct imsm_update_reshape_migration
);
8515 u
= calloc(1, update_memory_size
);
8517 dprintf("error: cannot get memory for "
8518 "imsm_create_metadata_update_for_migration\n");
8521 u
->type
= update_reshape_migration
;
8522 u
->subdev
= super
->current_vol
;
8523 u
->new_level
= geo
->level
;
8524 u
->new_layout
= geo
->layout
;
8525 u
->new_raid_disks
= u
->old_raid_disks
= geo
->raid_disks
;
8526 u
->new_disks
[0] = -1;
8527 u
->new_chunksize
= -1;
8529 dev
= get_imsm_dev(super
, u
->subdev
);
8531 struct imsm_map
*map
;
8533 map
= get_imsm_map(dev
, 0);
8535 int current_chunk_size
=
8536 __le16_to_cpu(map
->blocks_per_strip
) / 2;
8538 if (geo
->chunksize
!= current_chunk_size
) {
8539 u
->new_chunksize
= geo
->chunksize
/ 1024;
8541 "chunk size change from %i to %i\n",
8542 current_chunk_size
, u
->new_chunksize
);
8544 previous_level
= map
->raid_level
;
8547 if ((geo
->level
== 5) && (previous_level
== 0)) {
8548 struct mdinfo
*spares
= NULL
;
8550 u
->new_raid_disks
++;
8551 spares
= get_spares_for_grow(st
);
8552 if ((spares
== NULL
) || (spares
->array
.spare_disks
< 1)) {
8555 update_memory_size
= 0;
8556 dprintf("error: cannot get spare device "
8557 "for requested migration");
8562 dprintf("imsm: reshape update preparation : OK\n");
8565 return update_memory_size
;
8568 static void imsm_update_metadata_locally(struct supertype
*st
,
8571 struct metadata_update mu
;
8576 mu
.space_list
= NULL
;
8578 imsm_prepare_update(st
, &mu
);
8579 imsm_process_update(st
, &mu
);
8581 while (mu
.space_list
) {
8582 void **space
= mu
.space_list
;
8583 mu
.space_list
= *space
;
8588 /***************************************************************************
8589 * Function: imsm_analyze_change
8590 * Description: Function analyze change for single volume
8591 * and validate if transition is supported
8592 * Parameters: Geometry parameters, supertype structure
8593 * Returns: Operation type code on success, -1 if fail
8594 ****************************************************************************/
8595 enum imsm_reshape_type
imsm_analyze_change(struct supertype
*st
,
8596 struct geo_params
*geo
)
8603 getinfo_super_imsm_volume(st
, &info
, NULL
);
8604 if ((geo
->level
!= info
.array
.level
) &&
8605 (geo
->level
>= 0) &&
8606 (geo
->level
!= UnSet
)) {
8607 switch (info
.array
.level
) {
8609 if (geo
->level
== 5) {
8610 change
= CH_MIGRATION
;
8611 if (geo
->layout
!= ALGORITHM_LEFT_ASYMMETRIC
) {
8613 Name
" Error. Requested Layout "
8614 "not supported (left-asymmetric layout "
8615 "is supported only)!\n");
8617 goto analyse_change_exit
;
8621 if (geo
->level
== 10) {
8622 change
= CH_TAKEOVER
;
8627 if (geo
->level
== 0) {
8628 change
= CH_TAKEOVER
;
8633 if (geo
->level
== 0) {
8634 change
= CH_TAKEOVER
;
8641 Name
" Error. Level Migration from %d to %d "
8643 info
.array
.level
, geo
->level
);
8644 goto analyse_change_exit
;
8647 geo
->level
= info
.array
.level
;
8649 if ((geo
->layout
!= info
.array
.layout
)
8650 && ((geo
->layout
!= UnSet
) && (geo
->layout
!= -1))) {
8651 change
= CH_MIGRATION
;
8652 if ((info
.array
.layout
== 0)
8653 && (info
.array
.level
== 5)
8654 && (geo
->layout
== 5)) {
8655 /* reshape 5 -> 4 */
8656 } else if ((info
.array
.layout
== 5)
8657 && (info
.array
.level
== 5)
8658 && (geo
->layout
== 0)) {
8659 /* reshape 4 -> 5 */
8664 Name
" Error. Layout Migration from %d to %d "
8666 info
.array
.layout
, geo
->layout
);
8668 goto analyse_change_exit
;
8671 geo
->layout
= info
.array
.layout
;
8673 if ((geo
->chunksize
> 0) && (geo
->chunksize
!= UnSet
)
8674 && (geo
->chunksize
!= info
.array
.chunk_size
))
8675 change
= CH_MIGRATION
;
8677 geo
->chunksize
= info
.array
.chunk_size
;
8679 chunk
= geo
->chunksize
/ 1024;
8680 if (!validate_geometry_imsm(st
,
8690 struct intel_super
*super
= st
->sb
;
8691 struct imsm_super
*mpb
= super
->anchor
;
8693 if (mpb
->num_raid_devs
> 1) {
8695 Name
" Error. Cannot perform operation on %s"
8696 "- for this operation it MUST be single "
8697 "array in container\n",
8703 analyse_change_exit
:
8708 int imsm_takeover(struct supertype
*st
, struct geo_params
*geo
)
8710 struct intel_super
*super
= st
->sb
;
8711 struct imsm_update_takeover
*u
;
8713 u
= malloc(sizeof(struct imsm_update_takeover
));
8717 u
->type
= update_takeover
;
8718 u
->subarray
= super
->current_vol
;
8720 /* 10->0 transition */
8721 if (geo
->level
== 0)
8722 u
->direction
= R10_TO_R0
;
8724 /* 0->10 transition */
8725 if (geo
->level
== 10)
8726 u
->direction
= R0_TO_R10
;
8728 /* update metadata locally */
8729 imsm_update_metadata_locally(st
, u
,
8730 sizeof(struct imsm_update_takeover
));
8731 /* and possibly remotely */
8732 if (st
->update_tail
)
8733 append_metadata_update(st
, u
,
8734 sizeof(struct imsm_update_takeover
));
8741 static int imsm_reshape_super(struct supertype
*st
, long long size
, int level
,
8742 int layout
, int chunksize
, int raid_disks
,
8743 int delta_disks
, char *backup
, char *dev
,
8747 struct geo_params geo
;
8749 dprintf("imsm: reshape_super called.\n");
8751 memset(&geo
, 0, sizeof(struct geo_params
));
8754 geo
.dev_id
= st
->devnum
;
8757 geo
.layout
= layout
;
8758 geo
.chunksize
= chunksize
;
8759 geo
.raid_disks
= raid_disks
;
8760 if (delta_disks
!= UnSet
)
8761 geo
.raid_disks
+= delta_disks
;
8763 dprintf("\tfor level : %i\n", geo
.level
);
8764 dprintf("\tfor raid_disks : %i\n", geo
.raid_disks
);
8766 if (experimental() == 0)
8769 if (st
->container_dev
== st
->devnum
) {
8770 /* On container level we can only increase number of devices. */
8771 dprintf("imsm: info: Container operation\n");
8772 int old_raid_disks
= 0;
8774 if (imsm_reshape_is_allowed_on_container(
8775 st
, &geo
, &old_raid_disks
)) {
8776 struct imsm_update_reshape
*u
= NULL
;
8779 len
= imsm_create_metadata_update_for_reshape(
8780 st
, &geo
, old_raid_disks
, &u
);
8783 dprintf("imsm: Cannot prepare update\n");
8784 goto exit_imsm_reshape_super
;
8788 /* update metadata locally */
8789 imsm_update_metadata_locally(st
, u
, len
);
8790 /* and possibly remotely */
8791 if (st
->update_tail
)
8792 append_metadata_update(st
, u
, len
);
8797 fprintf(stderr
, Name
": (imsm) Operation "
8798 "is not allowed on this container\n");
8801 /* On volume level we support following operations
8802 * - takeover: raid10 -> raid0; raid0 -> raid10
8803 * - chunk size migration
8804 * - migration: raid5 -> raid0; raid0 -> raid5
8806 struct intel_super
*super
= st
->sb
;
8807 struct intel_dev
*dev
= super
->devlist
;
8809 dprintf("imsm: info: Volume operation\n");
8810 /* find requested device */
8812 if (imsm_find_array_minor_by_subdev(
8813 dev
->index
, st
->container_dev
, &devnum
) == 0
8814 && devnum
== geo
.dev_id
)
8819 fprintf(stderr
, Name
" Cannot find %s (%i) subarray\n",
8820 geo
.dev_name
, geo
.dev_id
);
8821 goto exit_imsm_reshape_super
;
8823 super
->current_vol
= dev
->index
;
8824 change
= imsm_analyze_change(st
, &geo
);
8827 ret_val
= imsm_takeover(st
, &geo
);
8829 case CH_MIGRATION
: {
8830 struct imsm_update_reshape_migration
*u
= NULL
;
8832 imsm_create_metadata_update_for_migration(
8836 "Cannot prepare update\n");
8840 /* update metadata locally */
8841 imsm_update_metadata_locally(st
, u
, len
);
8842 /* and possibly remotely */
8843 if (st
->update_tail
)
8844 append_metadata_update(st
, u
, len
);
8854 exit_imsm_reshape_super
:
8855 dprintf("imsm: reshape_super Exit code = %i\n", ret_val
);
8859 /*******************************************************************************
8860 * Function: wait_for_reshape_imsm
8861 * Description: Function writes new sync_max value and waits until
8862 * reshape process reach new position
8864 * sra : general array info
8865 * ndata : number of disks in new array's layout
8868 * 1 : there is no reshape in progress,
8870 ******************************************************************************/
8871 int wait_for_reshape_imsm(struct mdinfo
*sra
, int ndata
)
8873 int fd
= sysfs_get_fd(sra
, NULL
, "reshape_position");
8874 unsigned long long completed
;
8875 /* to_complete : new sync_max position */
8876 unsigned long long to_complete
= sra
->reshape_progress
;
8877 unsigned long long position_to_set
= to_complete
/ ndata
;
8880 dprintf("imsm: wait_for_reshape_imsm() "
8881 "cannot open reshape_position\n");
8885 if (sysfs_fd_get_ll(fd
, &completed
) < 0) {
8886 dprintf("imsm: wait_for_reshape_imsm() "
8887 "cannot read reshape_position (no reshape in progres)\n");
8892 if (completed
> to_complete
) {
8893 dprintf("imsm: wait_for_reshape_imsm() "
8894 "wrong next position to set %llu (%llu)\n",
8895 to_complete
, completed
);
8899 dprintf("Position set: %llu\n", position_to_set
);
8900 if (sysfs_set_num(sra
, NULL
, "sync_max",
8901 position_to_set
) != 0) {
8902 dprintf("imsm: wait_for_reshape_imsm() "
8903 "cannot set reshape position to %llu\n",
8914 select(fd
+1, &rfds
, NULL
, NULL
, NULL
);
8915 if (sysfs_get_str(sra
, NULL
, "sync_action",
8917 strncmp(action
, "reshape", 7) != 0)
8919 if (sysfs_fd_get_ll(fd
, &completed
) < 0) {
8920 dprintf("imsm: wait_for_reshape_imsm() "
8921 "cannot read reshape_position (in loop)\n");
8925 } while (completed
< to_complete
);
8931 /*******************************************************************************
8932 * Function: check_degradation_change
8933 * Description: Check that array hasn't become failed.
8935 * info : for sysfs access
8936 * sources : source disks descriptors
8937 * degraded: previous degradation level
8940 ******************************************************************************/
8941 int check_degradation_change(struct mdinfo
*info
,
8945 unsigned long long new_degraded
;
8946 sysfs_get_ll(info
, NULL
, "degraded", &new_degraded
);
8947 if (new_degraded
!= (unsigned long long)degraded
) {
8948 /* check each device to ensure it is still working */
8951 for (sd
= info
->devs
; sd
; sd
= sd
->next
) {
8952 if (sd
->disk
.state
& (1<<MD_DISK_FAULTY
))
8954 if (sd
->disk
.state
& (1<<MD_DISK_SYNC
)) {
8956 if (sysfs_get_str(info
,
8957 sd
, "state", sbuf
, 20) < 0 ||
8958 strstr(sbuf
, "faulty") ||
8959 strstr(sbuf
, "in_sync") == NULL
) {
8960 /* this device is dead */
8961 sd
->disk
.state
= (1<<MD_DISK_FAULTY
);
8962 if (sd
->disk
.raid_disk
>= 0 &&
8963 sources
[sd
->disk
.raid_disk
] >= 0) {
8965 sd
->disk
.raid_disk
]);
8966 sources
[sd
->disk
.raid_disk
] =
8975 return new_degraded
;
8978 /*******************************************************************************
8979 * Function: imsm_manage_reshape
8980 * Description: Function finds array under reshape and it manages reshape
8981 * process. It creates stripes backups (if required) and sets
8984 * afd : Backup handle (nattive) - not used
8985 * sra : general array info
8986 * reshape : reshape parameters - not used
8987 * st : supertype structure
8988 * blocks : size of critical section [blocks]
8989 * fds : table of source device descriptor
8990 * offsets : start of array (offest per devices)
8992 * destfd : table of destination device descriptor
8993 * destoffsets : table of destination offsets (per device)
8995 * 1 : success, reshape is done
8997 ******************************************************************************/
8998 static int imsm_manage_reshape(
8999 int afd
, struct mdinfo
*sra
, struct reshape
*reshape
,
9000 struct supertype
*st
, unsigned long backup_blocks
,
9001 int *fds
, unsigned long long *offsets
,
9002 int dests
, int *destfd
, unsigned long long *destoffsets
)
9005 struct intel_super
*super
= st
->sb
;
9006 struct intel_dev
*dv
= NULL
;
9007 struct imsm_dev
*dev
= NULL
;
9008 struct imsm_map
*map_src
;
9009 int migr_vol_qan
= 0;
9010 int ndata
, odata
; /* [bytes] */
9011 int chunk
; /* [bytes] */
9012 struct migr_record
*migr_rec
;
9014 unsigned int buf_size
; /* [bytes] */
9015 unsigned long long max_position
; /* array size [bytes] */
9016 unsigned long long next_step
; /* [blocks]/[bytes] */
9017 unsigned long long old_data_stripe_length
;
9018 unsigned long long start_src
; /* [bytes] */
9019 unsigned long long start
; /* [bytes] */
9020 unsigned long long start_buf_shift
; /* [bytes] */
9022 int source_layout
= 0;
9024 if (!fds
|| !offsets
|| !sra
)
9027 /* Find volume during the reshape */
9028 for (dv
= super
->devlist
; dv
; dv
= dv
->next
) {
9029 if (dv
->dev
->vol
.migr_type
== MIGR_GEN_MIGR
9030 && dv
->dev
->vol
.migr_state
== 1) {
9035 /* Only one volume can migrate at the same time */
9036 if (migr_vol_qan
!= 1) {
9037 fprintf(stderr
, Name
" : %s", migr_vol_qan
?
9038 "Number of migrating volumes greater than 1\n" :
9039 "There is no volume during migrationg\n");
9043 map_src
= get_imsm_map(dev
, 1);
9044 if (map_src
== NULL
)
9047 ndata
= imsm_num_data_members(dev
, 0);
9048 odata
= imsm_num_data_members(dev
, 1);
9050 chunk
= __le16_to_cpu(map_src
->blocks_per_strip
) * 512;
9051 old_data_stripe_length
= odata
* chunk
;
9053 migr_rec
= super
->migr_rec
;
9055 /* initialize migration record for start condition */
9056 if (sra
->reshape_progress
== 0)
9057 init_migr_record_imsm(st
, dev
, sra
);
9059 if (__le32_to_cpu(migr_rec
->rec_status
) != UNIT_SRC_NORMAL
) {
9060 dprintf("imsm: cannot restart migration when data "
9061 "are present in copy area.\n");
9067 buf_size
= __le32_to_cpu(migr_rec
->blocks_per_unit
) * 512;
9068 /* extend buffer size for parity disk */
9069 buf_size
+= __le32_to_cpu(migr_rec
->dest_depth_per_unit
) * 512;
9070 /* add space for stripe aligment */
9071 buf_size
+= old_data_stripe_length
;
9072 if (posix_memalign((void **)&buf
, 4096, buf_size
)) {
9073 dprintf("imsm: Cannot allocate checpoint buffer\n");
9077 max_position
= sra
->component_size
* ndata
;
9078 source_layout
= imsm_level_to_layout(map_src
->raid_level
);
9080 while (__le32_to_cpu(migr_rec
->curr_migr_unit
) <
9081 __le32_to_cpu(migr_rec
->num_migr_units
)) {
9082 /* current reshape position [blocks] */
9083 unsigned long long current_position
=
9084 __le32_to_cpu(migr_rec
->blocks_per_unit
)
9085 * __le32_to_cpu(migr_rec
->curr_migr_unit
);
9086 unsigned long long border
;
9088 /* Check that array hasn't become failed.
9090 degraded
= check_degradation_change(sra
, fds
, degraded
);
9092 dprintf("imsm: Abort reshape due to degradation"
9093 " level (%i)\n", degraded
);
9097 next_step
= __le32_to_cpu(migr_rec
->blocks_per_unit
);
9099 if ((current_position
+ next_step
) > max_position
)
9100 next_step
= max_position
- current_position
;
9102 start
= current_position
* 512;
9104 /* allign reading start to old geometry */
9105 start_buf_shift
= start
% old_data_stripe_length
;
9106 start_src
= start
- start_buf_shift
;
9108 border
= (start_src
/ odata
) - (start
/ ndata
);
9110 if (border
<= __le32_to_cpu(migr_rec
->dest_depth_per_unit
)) {
9111 /* save critical stripes to buf
9112 * start - start address of current unit
9114 * start_src - start address of current unit
9115 * to backup alligned to source array
9118 unsigned long long next_step_filler
= 0;
9119 unsigned long long copy_length
= next_step
* 512;
9121 /* allign copy area length to stripe in old geometry */
9122 next_step_filler
= ((copy_length
+ start_buf_shift
)
9123 % old_data_stripe_length
);
9124 if (next_step_filler
)
9125 next_step_filler
= (old_data_stripe_length
9126 - next_step_filler
);
9127 dprintf("save_stripes() parameters: start = %llu,"
9128 "\tstart_src = %llu,\tnext_step*512 = %llu,"
9129 "\tstart_in_buf_shift = %llu,"
9130 "\tnext_step_filler = %llu\n",
9131 start
, start_src
, copy_length
,
9132 start_buf_shift
, next_step_filler
);
9134 if (save_stripes(fds
, offsets
, map_src
->num_members
,
9135 chunk
, map_src
->raid_level
,
9136 source_layout
, 0, NULL
, start_src
,
9138 next_step_filler
+ start_buf_shift
,
9140 dprintf("imsm: Cannot save stripes"
9144 /* Convert data to destination format and store it
9145 * in backup general migration area
9147 if (save_backup_imsm(st
, dev
, sra
,
9148 buf
+ start_buf_shift
, copy_length
)) {
9149 dprintf("imsm: Cannot save stripes to "
9150 "target devices\n");
9153 if (save_checkpoint_imsm(st
, sra
,
9154 UNIT_SRC_IN_CP_AREA
)) {
9155 dprintf("imsm: Cannot write checkpoint to "
9156 "migration record (UNIT_SRC_IN_CP_AREA)\n");
9160 /* set next step to use whole border area */
9161 border
/= next_step
;
9163 next_step
*= border
;
9165 /* When data backed up, checkpoint stored,
9166 * kick the kernel to reshape unit of data
9168 next_step
= next_step
+ sra
->reshape_progress
;
9169 /* limit next step to array max position */
9170 if (next_step
> max_position
)
9171 next_step
= max_position
;
9172 sysfs_set_num(sra
, NULL
, "suspend_lo", sra
->reshape_progress
);
9173 sysfs_set_num(sra
, NULL
, "suspend_hi", next_step
);
9174 sra
->reshape_progress
= next_step
;
9176 /* wait until reshape finish */
9177 if (wait_for_reshape_imsm(sra
, ndata
) < 0) {
9178 dprintf("wait_for_reshape_imsm returned error!\n");
9182 if (save_checkpoint_imsm(st
, sra
, UNIT_SRC_NORMAL
) == 1) {
9183 /* ignore error == 2, this can mean end of reshape here
9185 dprintf("imsm: Cannot write checkpoint to "
9186 "migration record (UNIT_SRC_NORMAL)\n");
9192 /* return '1' if done */
9200 #endif /* MDASSEMBLE */
9202 struct superswitch super_imsm
= {
9204 .examine_super
= examine_super_imsm
,
9205 .brief_examine_super
= brief_examine_super_imsm
,
9206 .brief_examine_subarrays
= brief_examine_subarrays_imsm
,
9207 .export_examine_super
= export_examine_super_imsm
,
9208 .detail_super
= detail_super_imsm
,
9209 .brief_detail_super
= brief_detail_super_imsm
,
9210 .write_init_super
= write_init_super_imsm
,
9211 .validate_geometry
= validate_geometry_imsm
,
9212 .add_to_super
= add_to_super_imsm
,
9213 .remove_from_super
= remove_from_super_imsm
,
9214 .detail_platform
= detail_platform_imsm
,
9215 .kill_subarray
= kill_subarray_imsm
,
9216 .update_subarray
= update_subarray_imsm
,
9217 .load_container
= load_container_imsm
,
9218 .default_geometry
= default_geometry_imsm
,
9219 .get_disk_controller_domain
= imsm_get_disk_controller_domain
,
9220 .reshape_super
= imsm_reshape_super
,
9221 .manage_reshape
= imsm_manage_reshape
,
9222 .recover_backup
= recover_backup_imsm
,
9224 .match_home
= match_home_imsm
,
9225 .uuid_from_super
= uuid_from_super_imsm
,
9226 .getinfo_super
= getinfo_super_imsm
,
9227 .getinfo_super_disks
= getinfo_super_disks_imsm
,
9228 .update_super
= update_super_imsm
,
9230 .avail_size
= avail_size_imsm
,
9231 .min_acceptable_spare_size
= min_acceptable_spare_size_imsm
,
9233 .compare_super
= compare_super_imsm
,
9235 .load_super
= load_super_imsm
,
9236 .init_super
= init_super_imsm
,
9237 .store_super
= store_super_imsm
,
9238 .free_super
= free_super_imsm
,
9239 .match_metadata_desc
= match_metadata_desc_imsm
,
9240 .container_content
= container_content_imsm
,
9248 .open_new
= imsm_open_new
,
9249 .set_array_state
= imsm_set_array_state
,
9250 .set_disk
= imsm_set_disk
,
9251 .sync_metadata
= imsm_sync_metadata
,
9252 .activate_spare
= imsm_activate_spare
,
9253 .process_update
= imsm_process_update
,
9254 .prepare_update
= imsm_prepare_update
,
9255 #endif /* MDASSEMBLE */