2 * mdadm - Intel(R) Matrix Storage Manager Support
4 * Copyright (C) 2002-2008 Intel Corporation
6 * This program is free software; you can redistribute it and/or modify it
7 * under the terms and conditions of the GNU General Public License,
8 * version 2, as published by the Free Software Foundation.
10 * This program is distributed in the hope it will be useful, but WITHOUT
11 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
12 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for
15 * You should have received a copy of the GNU General Public License along with
16 * this program; if not, write to the Free Software Foundation, Inc.,
17 * 51 Franklin St - Fifth Floor, Boston, MA 02110-1301 USA.
20 #define HAVE_STDINT_H 1
24 #include "platform-intel.h"
30 /* MPB == Metadata Parameter Block */
31 #define MPB_SIGNATURE "Intel Raid ISM Cfg Sig. "
32 #define MPB_SIG_LEN (strlen(MPB_SIGNATURE))
33 #define MPB_VERSION_RAID0 "1.0.00"
34 #define MPB_VERSION_RAID1 "1.1.00"
35 #define MPB_VERSION_MANY_VOLUMES_PER_ARRAY "1.2.00"
36 #define MPB_VERSION_3OR4_DISK_ARRAY "1.2.01"
37 #define MPB_VERSION_RAID5 "1.2.02"
38 #define MPB_VERSION_5OR6_DISK_ARRAY "1.2.04"
39 #define MPB_VERSION_CNG "1.2.06"
40 #define MPB_VERSION_ATTRIBS "1.3.00"
41 #define MAX_SIGNATURE_LENGTH 32
42 #define MAX_RAID_SERIAL_LEN 16
45 #define MPB_ATTRIB_RAID0 __cpu_to_le32(0x00000001)
47 #define MPB_ATTRIB_RAID1 __cpu_to_le32(0x00000002)
49 #define MPB_ATTRIB_RAID10 __cpu_to_le32(0x00000004)
51 #define MPB_ATTRIB_RAID1E __cpu_to_le32(0x00000008)
53 #define MPB_ATTRIB_RAID5 __cpu_to_le32(0x00000010)
54 /* supports RAID CNG */
55 #define MPB_ATTRIB_RAIDCNG __cpu_to_le32(0x00000020)
56 /* supports expanded stripe sizes of 256K, 512K and 1MB */
57 #define MPB_ATTRIB_EXP_STRIPE_SIZE __cpu_to_le32(0x00000040)
59 /* The OROM Support RST Caching of Volumes */
60 #define MPB_ATTRIB_NVM __cpu_to_le32(0x02000000)
61 /* The OROM supports creating disks greater than 2TB */
62 #define MPB_ATTRIB_2TB_DISK __cpu_to_le32(0x04000000)
63 /* The OROM supports Bad Block Management */
64 #define MPB_ATTRIB_BBM __cpu_to_le32(0x08000000)
66 /* THe OROM Supports NVM Caching of Volumes */
67 #define MPB_ATTRIB_NEVER_USE2 __cpu_to_le32(0x10000000)
68 /* The OROM supports creating volumes greater than 2TB */
69 #define MPB_ATTRIB_2TB __cpu_to_le32(0x20000000)
70 /* originally for PMP, now it's wasted b/c. Never use this bit! */
71 #define MPB_ATTRIB_NEVER_USE __cpu_to_le32(0x40000000)
72 /* Verify MPB contents against checksum after reading MPB */
73 #define MPB_ATTRIB_CHECKSUM_VERIFY __cpu_to_le32(0x80000000)
75 /* Define all supported attributes that have to be accepted by mdadm
77 #define MPB_ATTRIB_SUPPORTED (MPB_ATTRIB_CHECKSUM_VERIFY | \
79 MPB_ATTRIB_2TB_DISK | \
84 MPB_ATTRIB_EXP_STRIPE_SIZE)
86 /* Define attributes that are unused but not harmful */
87 #define MPB_ATTRIB_IGNORED (MPB_ATTRIB_NEVER_USE)
89 #define MPB_SECTOR_CNT 2210
90 #define IMSM_RESERVED_SECTORS 4096
91 #define NUM_BLOCKS_DIRTY_STRIPE_REGION 2056
92 #define SECT_PER_MB_SHIFT 11
94 /* Disk configuration info. */
95 #define IMSM_MAX_DEVICES 255
97 __u8 serial
[MAX_RAID_SERIAL_LEN
];/* 0xD8 - 0xE7 ascii serial number */
98 __u32 total_blocks
; /* 0xE8 - 0xEB total blocks */
99 __u32 scsi_id
; /* 0xEC - 0xEF scsi ID */
100 #define SPARE_DISK __cpu_to_le32(0x01) /* Spare */
101 #define CONFIGURED_DISK __cpu_to_le32(0x02) /* Member of some RaidDev */
102 #define FAILED_DISK __cpu_to_le32(0x04) /* Permanent failure */
103 __u32 status
; /* 0xF0 - 0xF3 */
104 __u32 owner_cfg_num
; /* which config 0,1,2... owns this disk */
105 #define IMSM_DISK_FILLERS 4
106 __u32 filler
[IMSM_DISK_FILLERS
]; /* 0xF4 - 0x107 MPB_DISK_FILLERS for future expansion */
109 /* RAID map configuration infos. */
111 __u32 pba_of_lba0
; /* start address of partition */
112 __u32 blocks_per_member
;/* blocks per member */
113 __u32 num_data_stripes
; /* number of data stripes */
114 __u16 blocks_per_strip
;
115 __u8 map_state
; /* Normal, Uninitialized, Degraded, Failed */
116 #define IMSM_T_STATE_NORMAL 0
117 #define IMSM_T_STATE_UNINITIALIZED 1
118 #define IMSM_T_STATE_DEGRADED 2
119 #define IMSM_T_STATE_FAILED 3
121 #define IMSM_T_RAID0 0
122 #define IMSM_T_RAID1 1
123 #define IMSM_T_RAID5 5 /* since metadata version 1.2.02 ? */
124 __u8 num_members
; /* number of member disks */
125 __u8 num_domains
; /* number of parity domains */
126 __u8 failed_disk_num
; /* valid only when state is degraded */
128 __u32 filler
[7]; /* expansion area */
129 #define IMSM_ORD_REBUILD (1 << 24)
130 __u32 disk_ord_tbl
[1]; /* disk_ord_tbl[num_members],
131 * top byte contains some flags
133 } __attribute__ ((packed
));
136 __u32 curr_migr_unit
;
137 __u32 checkpoint_id
; /* id to access curr_migr_unit */
138 __u8 migr_state
; /* Normal or Migrating */
140 #define MIGR_REBUILD 1
141 #define MIGR_VERIFY 2 /* analagous to echo check > sync_action */
142 #define MIGR_GEN_MIGR 3
143 #define MIGR_STATE_CHANGE 4
144 #define MIGR_REPAIR 5
145 __u8 migr_type
; /* Initializing, Rebuilding, ... */
147 __u8 fs_state
; /* fast-sync state for CnG (0xff == disabled) */
148 __u16 verify_errors
; /* number of mismatches */
149 __u16 bad_blocks
; /* number of bad blocks during verify */
151 struct imsm_map map
[1];
152 /* here comes another one if migr_state */
153 } __attribute__ ((packed
));
156 __u8 volume
[MAX_RAID_SERIAL_LEN
];
159 #define DEV_BOOTABLE __cpu_to_le32(0x01)
160 #define DEV_BOOT_DEVICE __cpu_to_le32(0x02)
161 #define DEV_READ_COALESCING __cpu_to_le32(0x04)
162 #define DEV_WRITE_COALESCING __cpu_to_le32(0x08)
163 #define DEV_LAST_SHUTDOWN_DIRTY __cpu_to_le32(0x10)
164 #define DEV_HIDDEN_AT_BOOT __cpu_to_le32(0x20)
165 #define DEV_CURRENTLY_HIDDEN __cpu_to_le32(0x40)
166 #define DEV_VERIFY_AND_FIX __cpu_to_le32(0x80)
167 #define DEV_MAP_STATE_UNINIT __cpu_to_le32(0x100)
168 #define DEV_NO_AUTO_RECOVERY __cpu_to_le32(0x200)
169 #define DEV_CLONE_N_GO __cpu_to_le32(0x400)
170 #define DEV_CLONE_MAN_SYNC __cpu_to_le32(0x800)
171 #define DEV_CNG_MASTER_DISK_NUM __cpu_to_le32(0x1000)
172 __u32 status
; /* Persistent RaidDev status */
173 __u32 reserved_blocks
; /* Reserved blocks at beginning of volume */
177 __u8 cng_master_disk
;
181 #define IMSM_DEV_FILLERS 10
182 __u32 filler
[IMSM_DEV_FILLERS
];
184 } __attribute__ ((packed
));
187 __u8 sig
[MAX_SIGNATURE_LENGTH
]; /* 0x00 - 0x1F */
188 __u32 check_sum
; /* 0x20 - 0x23 MPB Checksum */
189 __u32 mpb_size
; /* 0x24 - 0x27 Size of MPB */
190 __u32 family_num
; /* 0x28 - 0x2B Checksum from first time this config was written */
191 __u32 generation_num
; /* 0x2C - 0x2F Incremented each time this array's MPB is written */
192 __u32 error_log_size
; /* 0x30 - 0x33 in bytes */
193 __u32 attributes
; /* 0x34 - 0x37 */
194 __u8 num_disks
; /* 0x38 Number of configured disks */
195 __u8 num_raid_devs
; /* 0x39 Number of configured volumes */
196 __u8 error_log_pos
; /* 0x3A */
197 __u8 fill
[1]; /* 0x3B */
198 __u32 cache_size
; /* 0x3c - 0x40 in mb */
199 __u32 orig_family_num
; /* 0x40 - 0x43 original family num */
200 __u32 pwr_cycle_count
; /* 0x44 - 0x47 simulated power cycle count for array */
201 __u32 bbm_log_size
; /* 0x48 - 0x4B - size of bad Block Mgmt Log in bytes */
202 #define IMSM_FILLERS 35
203 __u32 filler
[IMSM_FILLERS
]; /* 0x4C - 0xD7 RAID_MPB_FILLERS */
204 struct imsm_disk disk
[1]; /* 0xD8 diskTbl[numDisks] */
205 /* here comes imsm_dev[num_raid_devs] */
206 /* here comes BBM logs */
207 } __attribute__ ((packed
));
209 #define BBM_LOG_MAX_ENTRIES 254
211 struct bbm_log_entry
{
212 __u64 defective_block_start
;
213 #define UNREADABLE 0xFFFFFFFF
214 __u32 spare_block_offset
;
215 __u16 remapped_marked_count
;
217 } __attribute__ ((__packed__
));
220 __u32 signature
; /* 0xABADB10C */
222 __u32 reserved_spare_block_count
; /* 0 */
223 __u32 reserved
; /* 0xFFFF */
224 __u64 first_spare_lba
;
225 struct bbm_log_entry mapped_block_entries
[BBM_LOG_MAX_ENTRIES
];
226 } __attribute__ ((__packed__
));
230 static char *map_state_str
[] = { "normal", "uninitialized", "degraded", "failed" };
233 #define RAID_DISK_RESERVED_BLOCKS_IMSM_HI 2209
235 #define GEN_MIGR_AREA_SIZE 2048 /* General Migration Copy Area size in blocks */
237 #define UNIT_SRC_NORMAL 0 /* Source data for curr_migr_unit must
238 * be recovered using srcMap */
239 #define UNIT_SRC_IN_CP_AREA 1 /* Source data for curr_migr_unit has
240 * already been migrated and must
241 * be recovered from checkpoint area */
243 __u32 rec_status
; /* Status used to determine how to restart
244 * migration in case it aborts
246 __u32 curr_migr_unit
; /* 0..numMigrUnits-1 */
247 __u32 family_num
; /* Family number of MPB
248 * containing the RaidDev
249 * that is migrating */
250 __u32 ascending_migr
; /* True if migrating in increasing
252 __u32 blocks_per_unit
; /* Num disk blocks per unit of operation */
253 __u32 dest_depth_per_unit
; /* Num member blocks each destMap
255 * advances per unit-of-operation */
256 __u32 ckpt_area_pba
; /* Pba of first block of ckpt copy area */
257 __u32 dest_1st_member_lba
; /* First member lba on first
258 * stripe of destination */
259 __u32 num_migr_units
; /* Total num migration units-of-op */
260 __u32 post_migr_vol_cap
; /* Size of volume after
261 * migration completes */
262 __u32 post_migr_vol_cap_hi
; /* Expansion space for LBA64 */
263 __u32 ckpt_read_disk_num
; /* Which member disk in destSubMap[0] the
264 * migration ckpt record was read from
265 * (for recovered migrations) */
266 } __attribute__ ((__packed__
));
268 static __u8
migr_type(struct imsm_dev
*dev
)
270 if (dev
->vol
.migr_type
== MIGR_VERIFY
&&
271 dev
->status
& DEV_VERIFY_AND_FIX
)
274 return dev
->vol
.migr_type
;
277 static void set_migr_type(struct imsm_dev
*dev
, __u8 migr_type
)
279 /* for compatibility with older oroms convert MIGR_REPAIR, into
280 * MIGR_VERIFY w/ DEV_VERIFY_AND_FIX status
282 if (migr_type
== MIGR_REPAIR
) {
283 dev
->vol
.migr_type
= MIGR_VERIFY
;
284 dev
->status
|= DEV_VERIFY_AND_FIX
;
286 dev
->vol
.migr_type
= migr_type
;
287 dev
->status
&= ~DEV_VERIFY_AND_FIX
;
291 static unsigned int sector_count(__u32 bytes
)
293 return ((bytes
+ (512-1)) & (~(512-1))) / 512;
296 static unsigned int mpb_sectors(struct imsm_super
*mpb
)
298 return sector_count(__le32_to_cpu(mpb
->mpb_size
));
302 struct imsm_dev
*dev
;
303 struct intel_dev
*next
;
308 enum sys_dev_type type
;
311 struct intel_hba
*next
;
318 /* internal representation of IMSM metadata */
321 void *buf
; /* O_DIRECT buffer for reading/writing metadata */
322 struct imsm_super
*anchor
; /* immovable parameters */
325 void *migr_rec_buf
; /* buffer for I/O operations */
326 struct migr_record
*migr_rec
; /* migration record */
328 size_t len
; /* size of the 'buf' allocation */
329 void *next_buf
; /* for realloc'ing buf from the manager */
331 int updates_pending
; /* count of pending updates for mdmon */
332 int current_vol
; /* index of raid device undergoing creation */
333 __u32 create_offset
; /* common start for 'current_vol' */
334 __u32 random
; /* random data for seeding new family numbers */
335 struct intel_dev
*devlist
;
339 __u8 serial
[MAX_RAID_SERIAL_LEN
];
342 struct imsm_disk disk
;
345 struct extent
*e
; /* for determining freespace @ create */
346 int raiddisk
; /* slot to fill in autolayout */
348 } *disks
, *current_disk
;
349 struct dl
*disk_mgmt_list
; /* list of disks to add/remove while mdmon
351 struct dl
*missing
; /* disks removed while we weren't looking */
352 struct bbm_log
*bbm_log
;
353 struct intel_hba
*hba
; /* device path of the raid controller for this metadata */
354 const struct imsm_orom
*orom
; /* platform firmware support */
355 struct intel_super
*next
; /* (temp) list for disambiguating family_num */
359 struct imsm_disk disk
;
360 #define IMSM_UNKNOWN_OWNER (-1)
362 struct intel_disk
*next
;
366 unsigned long long start
, size
;
369 /* definitions of reshape process types */
370 enum imsm_reshape_type
{
375 /* definition of messages passed to imsm_process_update */
376 enum imsm_update_type
{
377 update_activate_spare
,
381 update_add_remove_disk
,
382 update_reshape_container_disks
,
383 update_reshape_migration
,
385 update_general_migration_checkpoint
,
388 struct imsm_update_activate_spare
{
389 enum imsm_update_type type
;
393 struct imsm_update_activate_spare
*next
;
406 enum takeover_direction
{
410 struct imsm_update_takeover
{
411 enum imsm_update_type type
;
413 enum takeover_direction direction
;
416 struct imsm_update_reshape
{
417 enum imsm_update_type type
;
421 int new_disks
[1]; /* new_raid_disks - old_raid_disks makedev number */
424 struct imsm_update_reshape_migration
{
425 enum imsm_update_type type
;
428 /* fields for array migration changes
435 int new_disks
[1]; /* new_raid_disks - old_raid_disks makedev number */
438 struct imsm_update_general_migration_checkpoint
{
439 enum imsm_update_type type
;
440 __u32 curr_migr_unit
;
444 __u8 serial
[MAX_RAID_SERIAL_LEN
];
447 struct imsm_update_create_array
{
448 enum imsm_update_type type
;
453 struct imsm_update_kill_array
{
454 enum imsm_update_type type
;
458 struct imsm_update_rename_array
{
459 enum imsm_update_type type
;
460 __u8 name
[MAX_RAID_SERIAL_LEN
];
464 struct imsm_update_add_remove_disk
{
465 enum imsm_update_type type
;
469 static const char *_sys_dev_type
[] = {
470 [SYS_DEV_UNKNOWN
] = "Unknown",
471 [SYS_DEV_SAS
] = "SAS",
472 [SYS_DEV_SATA
] = "SATA"
475 const char *get_sys_dev_type(enum sys_dev_type type
)
477 if (type
>= SYS_DEV_MAX
)
478 type
= SYS_DEV_UNKNOWN
;
480 return _sys_dev_type
[type
];
483 static struct intel_hba
* alloc_intel_hba(struct sys_dev
*device
)
485 struct intel_hba
*result
= malloc(sizeof(*result
));
487 result
->type
= device
->type
;
488 result
->path
= strdup(device
->path
);
490 if (result
->path
&& (result
->pci_id
= strrchr(result
->path
, '/')) != NULL
)
496 static struct intel_hba
* find_intel_hba(struct intel_hba
*hba
, struct sys_dev
*device
)
498 struct intel_hba
*result
=NULL
;
499 for (result
= hba
; result
; result
= result
->next
) {
500 if (result
->type
== device
->type
&& strcmp(result
->path
, device
->path
) == 0)
506 static int attach_hba_to_super(struct intel_super
*super
, struct sys_dev
*device
)
508 struct intel_hba
*hba
;
510 /* check if disk attached to Intel HBA */
511 hba
= find_intel_hba(super
->hba
, device
);
514 /* Check if HBA is already attached to super */
515 if (super
->hba
== NULL
) {
516 super
->hba
= alloc_intel_hba(device
);
521 /* Intel metadata allows for all disks attached to the same type HBA.
522 * Do not sypport odf HBA types mixing
524 if (device
->type
!= hba
->type
)
530 hba
->next
= alloc_intel_hba(device
);
534 static struct sys_dev
* find_disk_attached_hba(int fd
, const char *devname
)
536 struct sys_dev
*list
, *elem
, *prev
;
539 if ((list
= find_intel_devices()) == NULL
)
543 disk_path
= (char *) devname
;
545 disk_path
= diskfd_to_devpath(fd
);
552 for (prev
= NULL
, elem
= list
; elem
; prev
= elem
, elem
= elem
->next
) {
553 if (path_attached_to_hba(disk_path
, elem
->path
)) {
557 prev
->next
= elem
->next
;
559 if (disk_path
!= devname
)
565 if (disk_path
!= devname
)
573 static int find_intel_hba_capability(int fd
, struct intel_super
*super
,
576 static struct supertype
*match_metadata_desc_imsm(char *arg
)
578 struct supertype
*st
;
580 if (strcmp(arg
, "imsm") != 0 &&
581 strcmp(arg
, "default") != 0
585 st
= malloc(sizeof(*st
));
588 memset(st
, 0, sizeof(*st
));
589 st
->container_dev
= NoMdDev
;
590 st
->ss
= &super_imsm
;
591 st
->max_devs
= IMSM_MAX_DEVICES
;
592 st
->minor_version
= 0;
598 static __u8
*get_imsm_version(struct imsm_super
*mpb
)
600 return &mpb
->sig
[MPB_SIG_LEN
];
604 /* retrieve a disk directly from the anchor when the anchor is known to be
605 * up-to-date, currently only at load time
607 static struct imsm_disk
*__get_imsm_disk(struct imsm_super
*mpb
, __u8 index
)
609 if (index
>= mpb
->num_disks
)
611 return &mpb
->disk
[index
];
614 /* retrieve the disk description based on a index of the disk
617 static struct dl
*get_imsm_dl_disk(struct intel_super
*super
, __u8 index
)
621 for (d
= super
->disks
; d
; d
= d
->next
)
622 if (d
->index
== index
)
627 /* retrieve a disk from the parsed metadata */
628 static struct imsm_disk
*get_imsm_disk(struct intel_super
*super
, __u8 index
)
632 dl
= get_imsm_dl_disk(super
, index
);
639 /* generate a checksum directly from the anchor when the anchor is known to be
640 * up-to-date, currently only at load or write_super after coalescing
642 static __u32
__gen_imsm_checksum(struct imsm_super
*mpb
)
644 __u32 end
= mpb
->mpb_size
/ sizeof(end
);
645 __u32
*p
= (__u32
*) mpb
;
649 sum
+= __le32_to_cpu(*p
);
653 return sum
- __le32_to_cpu(mpb
->check_sum
);
656 static size_t sizeof_imsm_map(struct imsm_map
*map
)
658 return sizeof(struct imsm_map
) + sizeof(__u32
) * (map
->num_members
- 1);
661 struct imsm_map
*get_imsm_map(struct imsm_dev
*dev
, int second_map
)
663 /* A device can have 2 maps if it is in the middle of a migration.
665 * 0 - we return the first map
666 * 1 - we return the second map if it exists, else NULL
667 * -1 - we return the second map if it exists, else the first
669 struct imsm_map
*map
= &dev
->vol
.map
[0];
671 if (second_map
== 1 && !dev
->vol
.migr_state
)
673 else if (second_map
== 1 ||
674 (second_map
< 0 && dev
->vol
.migr_state
)) {
677 return ptr
+ sizeof_imsm_map(map
);
683 /* return the size of the device.
684 * migr_state increases the returned size if map[0] were to be duplicated
686 static size_t sizeof_imsm_dev(struct imsm_dev
*dev
, int migr_state
)
688 size_t size
= sizeof(*dev
) - sizeof(struct imsm_map
) +
689 sizeof_imsm_map(get_imsm_map(dev
, 0));
691 /* migrating means an additional map */
692 if (dev
->vol
.migr_state
)
693 size
+= sizeof_imsm_map(get_imsm_map(dev
, 1));
695 size
+= sizeof_imsm_map(get_imsm_map(dev
, 0));
701 /* retrieve disk serial number list from a metadata update */
702 static struct disk_info
*get_disk_info(struct imsm_update_create_array
*update
)
705 struct disk_info
*inf
;
707 inf
= u
+ sizeof(*update
) - sizeof(struct imsm_dev
) +
708 sizeof_imsm_dev(&update
->dev
, 0);
714 static struct imsm_dev
*__get_imsm_dev(struct imsm_super
*mpb
, __u8 index
)
720 if (index
>= mpb
->num_raid_devs
)
723 /* devices start after all disks */
724 offset
= ((void *) &mpb
->disk
[mpb
->num_disks
]) - _mpb
;
726 for (i
= 0; i
<= index
; i
++)
728 return _mpb
+ offset
;
730 offset
+= sizeof_imsm_dev(_mpb
+ offset
, 0);
735 static struct imsm_dev
*get_imsm_dev(struct intel_super
*super
, __u8 index
)
737 struct intel_dev
*dv
;
739 if (index
>= super
->anchor
->num_raid_devs
)
741 for (dv
= super
->devlist
; dv
; dv
= dv
->next
)
742 if (dv
->index
== index
)
750 * == 1 get second map
751 * == -1 than get map according to the current migr_state
753 static __u32
get_imsm_ord_tbl_ent(struct imsm_dev
*dev
,
757 struct imsm_map
*map
;
759 map
= get_imsm_map(dev
, second_map
);
761 /* top byte identifies disk under rebuild */
762 return __le32_to_cpu(map
->disk_ord_tbl
[slot
]);
765 #define ord_to_idx(ord) (((ord) << 8) >> 8)
766 static __u32
get_imsm_disk_idx(struct imsm_dev
*dev
, int slot
, int second_map
)
768 __u32 ord
= get_imsm_ord_tbl_ent(dev
, slot
, second_map
);
770 return ord_to_idx(ord
);
773 static void set_imsm_ord_tbl_ent(struct imsm_map
*map
, int slot
, __u32 ord
)
775 map
->disk_ord_tbl
[slot
] = __cpu_to_le32(ord
);
778 static int get_imsm_disk_slot(struct imsm_map
*map
, unsigned idx
)
783 for (slot
= 0; slot
< map
->num_members
; slot
++) {
784 ord
= __le32_to_cpu(map
->disk_ord_tbl
[slot
]);
785 if (ord_to_idx(ord
) == idx
)
792 static int get_imsm_raid_level(struct imsm_map
*map
)
794 if (map
->raid_level
== 1) {
795 if (map
->num_members
== 2)
801 return map
->raid_level
;
804 static int cmp_extent(const void *av
, const void *bv
)
806 const struct extent
*a
= av
;
807 const struct extent
*b
= bv
;
808 if (a
->start
< b
->start
)
810 if (a
->start
> b
->start
)
815 static int count_memberships(struct dl
*dl
, struct intel_super
*super
)
820 for (i
= 0; i
< super
->anchor
->num_raid_devs
; i
++) {
821 struct imsm_dev
*dev
= get_imsm_dev(super
, i
);
822 struct imsm_map
*map
= get_imsm_map(dev
, 0);
824 if (get_imsm_disk_slot(map
, dl
->index
) >= 0)
831 static __u32
imsm_min_reserved_sectors(struct intel_super
*super
);
833 static struct extent
*get_extents(struct intel_super
*super
, struct dl
*dl
)
835 /* find a list of used extents on the given physical device */
836 struct extent
*rv
, *e
;
838 int memberships
= count_memberships(dl
, super
);
841 /* trim the reserved area for spares, so they can join any array
842 * regardless of whether the OROM has assigned sectors from the
843 * IMSM_RESERVED_SECTORS region
846 reservation
= imsm_min_reserved_sectors(super
);
848 reservation
= MPB_SECTOR_CNT
+ IMSM_RESERVED_SECTORS
;
850 rv
= malloc(sizeof(struct extent
) * (memberships
+ 1));
855 for (i
= 0; i
< super
->anchor
->num_raid_devs
; i
++) {
856 struct imsm_dev
*dev
= get_imsm_dev(super
, i
);
857 struct imsm_map
*map
= get_imsm_map(dev
, 0);
859 if (get_imsm_disk_slot(map
, dl
->index
) >= 0) {
860 e
->start
= __le32_to_cpu(map
->pba_of_lba0
);
861 e
->size
= __le32_to_cpu(map
->blocks_per_member
);
865 qsort(rv
, memberships
, sizeof(*rv
), cmp_extent
);
867 /* determine the start of the metadata
868 * when no raid devices are defined use the default
869 * ...otherwise allow the metadata to truncate the value
870 * as is the case with older versions of imsm
873 struct extent
*last
= &rv
[memberships
- 1];
876 remainder
= __le32_to_cpu(dl
->disk
.total_blocks
) -
877 (last
->start
+ last
->size
);
878 /* round down to 1k block to satisfy precision of the kernel
882 /* make sure remainder is still sane */
883 if (remainder
< (unsigned)ROUND_UP(super
->len
, 512) >> 9)
884 remainder
= ROUND_UP(super
->len
, 512) >> 9;
885 if (reservation
> remainder
)
886 reservation
= remainder
;
888 e
->start
= __le32_to_cpu(dl
->disk
.total_blocks
) - reservation
;
893 /* try to determine how much space is reserved for metadata from
894 * the last get_extents() entry, otherwise fallback to the
897 static __u32
imsm_reserved_sectors(struct intel_super
*super
, struct dl
*dl
)
903 /* for spares just return a minimal reservation which will grow
904 * once the spare is picked up by an array
907 return MPB_SECTOR_CNT
;
909 e
= get_extents(super
, dl
);
911 return MPB_SECTOR_CNT
+ IMSM_RESERVED_SECTORS
;
913 /* scroll to last entry */
914 for (i
= 0; e
[i
].size
; i
++)
917 rv
= __le32_to_cpu(dl
->disk
.total_blocks
) - e
[i
].start
;
924 static int is_spare(struct imsm_disk
*disk
)
926 return (disk
->status
& SPARE_DISK
) == SPARE_DISK
;
929 static int is_configured(struct imsm_disk
*disk
)
931 return (disk
->status
& CONFIGURED_DISK
) == CONFIGURED_DISK
;
934 static int is_failed(struct imsm_disk
*disk
)
936 return (disk
->status
& FAILED_DISK
) == FAILED_DISK
;
939 /* try to determine how much space is reserved for metadata from
940 * the last get_extents() entry on the smallest active disk,
941 * otherwise fallback to the default
943 static __u32
imsm_min_reserved_sectors(struct intel_super
*super
)
947 __u32 min_active
, remainder
;
948 __u32 rv
= MPB_SECTOR_CNT
+ IMSM_RESERVED_SECTORS
;
949 struct dl
*dl
, *dl_min
= NULL
;
955 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
958 if (dl
->disk
.total_blocks
< min_active
|| min_active
== 0) {
960 min_active
= dl
->disk
.total_blocks
;
966 /* find last lba used by subarrays on the smallest active disk */
967 e
= get_extents(super
, dl_min
);
970 for (i
= 0; e
[i
].size
; i
++)
973 remainder
= min_active
- e
[i
].start
;
976 /* to give priority to recovery we should not require full
977 IMSM_RESERVED_SECTORS from the spare */
978 rv
= MPB_SECTOR_CNT
+ NUM_BLOCKS_DIRTY_STRIPE_REGION
;
980 /* if real reservation is smaller use that value */
981 return (remainder
< rv
) ? remainder
: rv
;
984 /* Return minimum size of a spare that can be used in this array*/
985 static unsigned long long min_acceptable_spare_size_imsm(struct supertype
*st
)
987 struct intel_super
*super
= st
->sb
;
991 unsigned long long rv
= 0;
995 /* find first active disk in array */
997 while (dl
&& (is_failed(&dl
->disk
) || dl
->index
== -1))
1001 /* find last lba used by subarrays */
1002 e
= get_extents(super
, dl
);
1005 for (i
= 0; e
[i
].size
; i
++)
1008 rv
= e
[i
-1].start
+ e
[i
-1].size
;
1011 /* add the amount of space needed for metadata */
1012 rv
= rv
+ imsm_min_reserved_sectors(super
);
1018 static __u64
blocks_per_migr_unit(struct intel_super
*super
,
1019 struct imsm_dev
*dev
);
1021 static int is_gen_migration(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 ",
1116 __le32_to_cpu(dev
->vol
.curr_migr_unit
));
1117 if ((is_gen_migration(dev
)) && (super
->disks
->index
> 1))
1120 printf("(%llu)", (unsigned long long)
1121 blocks_per_migr_unit(super
, dev
));
1124 printf(" Dirty State : %s\n", dev
->vol
.dirty
? "dirty" : "clean");
1127 static void print_imsm_disk(struct imsm_disk
*disk
, int index
, __u32 reserved
)
1129 char str
[MAX_RAID_SERIAL_LEN
+ 1];
1132 if (index
< -1 || !disk
)
1136 snprintf(str
, MAX_RAID_SERIAL_LEN
+ 1, "%s", disk
->serial
);
1138 printf(" Disk%02d Serial : %s\n", index
, str
);
1140 printf(" Disk Serial : %s\n", str
);
1141 printf(" State :%s%s%s\n", is_spare(disk
) ? " spare" : "",
1142 is_configured(disk
) ? " active" : "",
1143 is_failed(disk
) ? " failed" : "");
1144 printf(" Id : %08x\n", __le32_to_cpu(disk
->scsi_id
));
1145 sz
= __le32_to_cpu(disk
->total_blocks
) - reserved
;
1146 printf(" Usable Size : %llu%s\n", (unsigned long long)sz
,
1147 human_size(sz
* 512));
1150 void examine_migr_rec_imsm(struct intel_super
*super
)
1152 struct migr_record
*migr_rec
= super
->migr_rec
;
1153 struct imsm_super
*mpb
= super
->anchor
;
1156 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
1157 struct imsm_dev
*dev
= __get_imsm_dev(mpb
, i
);
1158 if (is_gen_migration(dev
) == 0)
1161 printf("\nMigration Record Information:");
1162 if (super
->disks
->index
> 1) {
1163 printf(" Empty\n ");
1164 printf("Examine one of first two disks in array\n");
1167 printf("\n Status : ");
1168 if (__le32_to_cpu(migr_rec
->rec_status
) == UNIT_SRC_NORMAL
)
1171 printf("Contains Data\n");
1172 printf(" Current Unit : %u\n",
1173 __le32_to_cpu(migr_rec
->curr_migr_unit
));
1174 printf(" Family : %u\n",
1175 __le32_to_cpu(migr_rec
->family_num
));
1176 printf(" Ascending : %u\n",
1177 __le32_to_cpu(migr_rec
->ascending_migr
));
1178 printf(" Blocks Per Unit : %u\n",
1179 __le32_to_cpu(migr_rec
->blocks_per_unit
));
1180 printf(" Dest. Depth Per Unit : %u\n",
1181 __le32_to_cpu(migr_rec
->dest_depth_per_unit
));
1182 printf(" Checkpoint Area pba : %u\n",
1183 __le32_to_cpu(migr_rec
->ckpt_area_pba
));
1184 printf(" First member lba : %u\n",
1185 __le32_to_cpu(migr_rec
->dest_1st_member_lba
));
1186 printf(" Total Number of Units : %u\n",
1187 __le32_to_cpu(migr_rec
->num_migr_units
));
1188 printf(" Size of volume : %u\n",
1189 __le32_to_cpu(migr_rec
->post_migr_vol_cap
));
1190 printf(" Expansion space for LBA64 : %u\n",
1191 __le32_to_cpu(migr_rec
->post_migr_vol_cap_hi
));
1192 printf(" Record was read from : %u\n",
1193 __le32_to_cpu(migr_rec
->ckpt_read_disk_num
));
1198 #endif /* MDASSEMBLE */
1199 /*******************************************************************************
1200 * function: imsm_check_attributes
1201 * Description: Function checks if features represented by attributes flags
1202 * are supported by mdadm.
1204 * attributes - Attributes read from metadata
1206 * 0 - passed attributes contains unsupported features flags
1207 * 1 - all features are supported
1208 ******************************************************************************/
1209 static int imsm_check_attributes(__u32 attributes
)
1212 __u32 not_supported
= MPB_ATTRIB_SUPPORTED
^0xffffffff;
1214 not_supported
&= ~MPB_ATTRIB_IGNORED
;
1216 not_supported
&= attributes
;
1217 if (not_supported
) {
1218 fprintf(stderr
, Name
"(IMSM): Unsupported attributes : %x\n",
1219 (unsigned)__le32_to_cpu(not_supported
));
1220 if (not_supported
& MPB_ATTRIB_CHECKSUM_VERIFY
) {
1221 dprintf("\t\tMPB_ATTRIB_CHECKSUM_VERIFY \n");
1222 not_supported
^= MPB_ATTRIB_CHECKSUM_VERIFY
;
1224 if (not_supported
& MPB_ATTRIB_2TB
) {
1225 dprintf("\t\tMPB_ATTRIB_2TB\n");
1226 not_supported
^= MPB_ATTRIB_2TB
;
1228 if (not_supported
& MPB_ATTRIB_RAID0
) {
1229 dprintf("\t\tMPB_ATTRIB_RAID0\n");
1230 not_supported
^= MPB_ATTRIB_RAID0
;
1232 if (not_supported
& MPB_ATTRIB_RAID1
) {
1233 dprintf("\t\tMPB_ATTRIB_RAID1\n");
1234 not_supported
^= MPB_ATTRIB_RAID1
;
1236 if (not_supported
& MPB_ATTRIB_RAID10
) {
1237 dprintf("\t\tMPB_ATTRIB_RAID10\n");
1238 not_supported
^= MPB_ATTRIB_RAID10
;
1240 if (not_supported
& MPB_ATTRIB_RAID1E
) {
1241 dprintf("\t\tMPB_ATTRIB_RAID1E\n");
1242 not_supported
^= MPB_ATTRIB_RAID1E
;
1244 if (not_supported
& MPB_ATTRIB_RAID5
) {
1245 dprintf("\t\tMPB_ATTRIB_RAID5\n");
1246 not_supported
^= MPB_ATTRIB_RAID5
;
1248 if (not_supported
& MPB_ATTRIB_RAIDCNG
) {
1249 dprintf("\t\tMPB_ATTRIB_RAIDCNG\n");
1250 not_supported
^= MPB_ATTRIB_RAIDCNG
;
1252 if (not_supported
& MPB_ATTRIB_BBM
) {
1253 dprintf("\t\tMPB_ATTRIB_BBM\n");
1254 not_supported
^= MPB_ATTRIB_BBM
;
1256 if (not_supported
& MPB_ATTRIB_CHECKSUM_VERIFY
) {
1257 dprintf("\t\tMPB_ATTRIB_CHECKSUM_VERIFY (== MPB_ATTRIB_LEGACY)\n");
1258 not_supported
^= MPB_ATTRIB_CHECKSUM_VERIFY
;
1260 if (not_supported
& MPB_ATTRIB_EXP_STRIPE_SIZE
) {
1261 dprintf("\t\tMPB_ATTRIB_EXP_STRIP_SIZE\n");
1262 not_supported
^= MPB_ATTRIB_EXP_STRIPE_SIZE
;
1264 if (not_supported
& MPB_ATTRIB_2TB_DISK
) {
1265 dprintf("\t\tMPB_ATTRIB_2TB_DISK\n");
1266 not_supported
^= MPB_ATTRIB_2TB_DISK
;
1268 if (not_supported
& MPB_ATTRIB_NEVER_USE2
) {
1269 dprintf("\t\tMPB_ATTRIB_NEVER_USE2\n");
1270 not_supported
^= MPB_ATTRIB_NEVER_USE2
;
1272 if (not_supported
& MPB_ATTRIB_NEVER_USE
) {
1273 dprintf("\t\tMPB_ATTRIB_NEVER_USE\n");
1274 not_supported
^= MPB_ATTRIB_NEVER_USE
;
1278 dprintf(Name
"(IMSM): Unknown attributes : %x\n", not_supported
);
1287 static void getinfo_super_imsm(struct supertype
*st
, struct mdinfo
*info
, char *map
);
1289 static void examine_super_imsm(struct supertype
*st
, char *homehost
)
1291 struct intel_super
*super
= st
->sb
;
1292 struct imsm_super
*mpb
= super
->anchor
;
1293 char str
[MAX_SIGNATURE_LENGTH
];
1298 __u32 reserved
= imsm_reserved_sectors(super
, super
->disks
);
1301 snprintf(str
, MPB_SIG_LEN
, "%s", mpb
->sig
);
1302 printf(" Magic : %s\n", str
);
1303 snprintf(str
, strlen(MPB_VERSION_RAID0
), "%s", get_imsm_version(mpb
));
1304 printf(" Version : %s\n", get_imsm_version(mpb
));
1305 printf(" Orig Family : %08x\n", __le32_to_cpu(mpb
->orig_family_num
));
1306 printf(" Family : %08x\n", __le32_to_cpu(mpb
->family_num
));
1307 printf(" Generation : %08x\n", __le32_to_cpu(mpb
->generation_num
));
1308 printf(" Attributes : ");
1309 if (imsm_check_attributes(mpb
->attributes
))
1310 printf("All supported\n");
1312 printf("not supported\n");
1313 getinfo_super_imsm(st
, &info
, NULL
);
1314 fname_from_uuid(st
, &info
, nbuf
, ':');
1315 printf(" UUID : %s\n", nbuf
+ 5);
1316 sum
= __le32_to_cpu(mpb
->check_sum
);
1317 printf(" Checksum : %08x %s\n", sum
,
1318 __gen_imsm_checksum(mpb
) == sum
? "correct" : "incorrect");
1319 printf(" MPB Sectors : %d\n", mpb_sectors(mpb
));
1320 printf(" Disks : %d\n", mpb
->num_disks
);
1321 printf(" RAID Devices : %d\n", mpb
->num_raid_devs
);
1322 print_imsm_disk(__get_imsm_disk(mpb
, super
->disks
->index
), super
->disks
->index
, reserved
);
1323 if (super
->bbm_log
) {
1324 struct bbm_log
*log
= super
->bbm_log
;
1327 printf("Bad Block Management Log:\n");
1328 printf(" Log Size : %d\n", __le32_to_cpu(mpb
->bbm_log_size
));
1329 printf(" Signature : %x\n", __le32_to_cpu(log
->signature
));
1330 printf(" Entry Count : %d\n", __le32_to_cpu(log
->entry_count
));
1331 printf(" Spare Blocks : %d\n", __le32_to_cpu(log
->reserved_spare_block_count
));
1332 printf(" First Spare : %llx\n",
1333 (unsigned long long) __le64_to_cpu(log
->first_spare_lba
));
1335 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
1337 struct imsm_dev
*dev
= __get_imsm_dev(mpb
, i
);
1339 super
->current_vol
= i
;
1340 getinfo_super_imsm(st
, &info
, NULL
);
1341 fname_from_uuid(st
, &info
, nbuf
, ':');
1342 print_imsm_dev(super
, dev
, nbuf
+ 5, super
->disks
->index
);
1344 for (i
= 0; i
< mpb
->num_disks
; i
++) {
1345 if (i
== super
->disks
->index
)
1347 print_imsm_disk(__get_imsm_disk(mpb
, i
), i
, reserved
);
1350 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
1351 if (dl
->index
== -1)
1352 print_imsm_disk(&dl
->disk
, -1, reserved
);
1354 examine_migr_rec_imsm(super
);
1357 static void brief_examine_super_imsm(struct supertype
*st
, int verbose
)
1359 /* We just write a generic IMSM ARRAY entry */
1362 struct intel_super
*super
= st
->sb
;
1364 if (!super
->anchor
->num_raid_devs
) {
1365 printf("ARRAY metadata=imsm\n");
1369 getinfo_super_imsm(st
, &info
, NULL
);
1370 fname_from_uuid(st
, &info
, nbuf
, ':');
1371 printf("ARRAY metadata=imsm UUID=%s\n", nbuf
+ 5);
1374 static void brief_examine_subarrays_imsm(struct supertype
*st
, int verbose
)
1376 /* We just write a generic IMSM ARRAY entry */
1380 struct intel_super
*super
= st
->sb
;
1383 if (!super
->anchor
->num_raid_devs
)
1386 getinfo_super_imsm(st
, &info
, NULL
);
1387 fname_from_uuid(st
, &info
, nbuf
, ':');
1388 for (i
= 0; i
< super
->anchor
->num_raid_devs
; i
++) {
1389 struct imsm_dev
*dev
= get_imsm_dev(super
, i
);
1391 super
->current_vol
= i
;
1392 getinfo_super_imsm(st
, &info
, NULL
);
1393 fname_from_uuid(st
, &info
, nbuf1
, ':');
1394 printf("ARRAY /dev/md/%.16s container=%s member=%d UUID=%s\n",
1395 dev
->volume
, nbuf
+ 5, i
, nbuf1
+ 5);
1399 static void export_examine_super_imsm(struct supertype
*st
)
1401 struct intel_super
*super
= st
->sb
;
1402 struct imsm_super
*mpb
= super
->anchor
;
1406 getinfo_super_imsm(st
, &info
, NULL
);
1407 fname_from_uuid(st
, &info
, nbuf
, ':');
1408 printf("MD_METADATA=imsm\n");
1409 printf("MD_LEVEL=container\n");
1410 printf("MD_UUID=%s\n", nbuf
+5);
1411 printf("MD_DEVICES=%u\n", mpb
->num_disks
);
1414 static void detail_super_imsm(struct supertype
*st
, char *homehost
)
1419 getinfo_super_imsm(st
, &info
, NULL
);
1420 fname_from_uuid(st
, &info
, nbuf
, ':');
1421 printf("\n UUID : %s\n", nbuf
+ 5);
1424 static void brief_detail_super_imsm(struct supertype
*st
)
1428 getinfo_super_imsm(st
, &info
, NULL
);
1429 fname_from_uuid(st
, &info
, nbuf
, ':');
1430 printf(" UUID=%s", nbuf
+ 5);
1433 static int imsm_read_serial(int fd
, char *devname
, __u8
*serial
);
1434 static void fd2devname(int fd
, char *name
);
1436 static int ahci_enumerate_ports(const char *hba_path
, int port_count
, int host_base
, int verbose
)
1438 /* dump an unsorted list of devices attached to AHCI Intel storage
1439 * controller, as well as non-connected ports
1441 int hba_len
= strlen(hba_path
) + 1;
1446 unsigned long port_mask
= (1 << port_count
) - 1;
1448 if (port_count
> (int)sizeof(port_mask
) * 8) {
1450 fprintf(stderr
, Name
": port_count %d out of range\n", port_count
);
1454 /* scroll through /sys/dev/block looking for devices attached to
1457 dir
= opendir("/sys/dev/block");
1458 for (ent
= dir
? readdir(dir
) : NULL
; ent
; ent
= readdir(dir
)) {
1469 if (sscanf(ent
->d_name
, "%d:%d", &major
, &minor
) != 2)
1471 path
= devt_to_devpath(makedev(major
, minor
));
1474 if (!path_attached_to_hba(path
, hba_path
)) {
1480 /* retrieve the scsi device type */
1481 if (asprintf(&device
, "/sys/dev/block/%d:%d/device/xxxxxxx", major
, minor
) < 0) {
1483 fprintf(stderr
, Name
": failed to allocate 'device'\n");
1487 sprintf(device
, "/sys/dev/block/%d:%d/device/type", major
, minor
);
1488 if (load_sys(device
, buf
) != 0) {
1490 fprintf(stderr
, Name
": failed to read device type for %s\n",
1496 type
= strtoul(buf
, NULL
, 10);
1498 /* if it's not a disk print the vendor and model */
1499 if (!(type
== 0 || type
== 7 || type
== 14)) {
1502 sprintf(device
, "/sys/dev/block/%d:%d/device/vendor", major
, minor
);
1503 if (load_sys(device
, buf
) == 0) {
1504 strncpy(vendor
, buf
, sizeof(vendor
));
1505 vendor
[sizeof(vendor
) - 1] = '\0';
1506 c
= (char *) &vendor
[sizeof(vendor
) - 1];
1507 while (isspace(*c
) || *c
== '\0')
1511 sprintf(device
, "/sys/dev/block/%d:%d/device/model", major
, minor
);
1512 if (load_sys(device
, buf
) == 0) {
1513 strncpy(model
, buf
, sizeof(model
));
1514 model
[sizeof(model
) - 1] = '\0';
1515 c
= (char *) &model
[sizeof(model
) - 1];
1516 while (isspace(*c
) || *c
== '\0')
1520 if (vendor
[0] && model
[0])
1521 sprintf(buf
, "%.64s %.64s", vendor
, model
);
1523 switch (type
) { /* numbers from hald/linux/device.c */
1524 case 1: sprintf(buf
, "tape"); break;
1525 case 2: sprintf(buf
, "printer"); break;
1526 case 3: sprintf(buf
, "processor"); break;
1528 case 5: sprintf(buf
, "cdrom"); break;
1529 case 6: sprintf(buf
, "scanner"); break;
1530 case 8: sprintf(buf
, "media_changer"); break;
1531 case 9: sprintf(buf
, "comm"); break;
1532 case 12: sprintf(buf
, "raid"); break;
1533 default: sprintf(buf
, "unknown");
1539 /* chop device path to 'host%d' and calculate the port number */
1540 c
= strchr(&path
[hba_len
], '/');
1543 fprintf(stderr
, Name
": %s - invalid path name\n", path
+ hba_len
);
1548 if (sscanf(&path
[hba_len
], "host%d", &port
) == 1)
1552 *c
= '/'; /* repair the full string */
1553 fprintf(stderr
, Name
": failed to determine port number for %s\n",
1560 /* mark this port as used */
1561 port_mask
&= ~(1 << port
);
1563 /* print out the device information */
1565 printf(" Port%d : - non-disk device (%s) -\n", port
, buf
);
1569 fd
= dev_open(ent
->d_name
, O_RDONLY
);
1571 printf(" Port%d : - disk info unavailable -\n", port
);
1573 fd2devname(fd
, buf
);
1574 printf(" Port%d : %s", port
, buf
);
1575 if (imsm_read_serial(fd
, NULL
, (__u8
*) buf
) == 0)
1576 printf(" (%.*s)\n", MAX_RAID_SERIAL_LEN
, buf
);
1591 for (i
= 0; i
< port_count
; i
++)
1592 if (port_mask
& (1 << i
))
1593 printf(" Port%d : - no device attached -\n", i
);
1599 static void print_found_intel_controllers(struct sys_dev
*elem
)
1601 for (; elem
; elem
= elem
->next
) {
1602 fprintf(stderr
, Name
": found Intel(R) ");
1603 if (elem
->type
== SYS_DEV_SATA
)
1604 fprintf(stderr
, "SATA ");
1605 else if (elem
->type
== SYS_DEV_SAS
)
1606 fprintf(stderr
, "SAS ");
1607 fprintf(stderr
, "RAID controller");
1609 fprintf(stderr
, " at %s", elem
->pci_id
);
1610 fprintf(stderr
, ".\n");
1615 static int ahci_get_port_count(const char *hba_path
, int *port_count
)
1622 if ((dir
= opendir(hba_path
)) == NULL
)
1625 for (ent
= readdir(dir
); ent
; ent
= readdir(dir
)) {
1628 if (sscanf(ent
->d_name
, "host%d", &host
) != 1)
1630 if (*port_count
== 0)
1632 else if (host
< host_base
)
1635 if (host
+ 1 > *port_count
+ host_base
)
1636 *port_count
= host
+ 1 - host_base
;
1642 static void print_imsm_capability(const struct imsm_orom
*orom
)
1644 printf(" Platform : Intel(R) Matrix Storage Manager\n");
1645 printf(" Version : %d.%d.%d.%d\n", orom
->major_ver
, orom
->minor_ver
,
1646 orom
->hotfix_ver
, orom
->build
);
1647 printf(" RAID Levels :%s%s%s%s%s\n",
1648 imsm_orom_has_raid0(orom
) ? " raid0" : "",
1649 imsm_orom_has_raid1(orom
) ? " raid1" : "",
1650 imsm_orom_has_raid1e(orom
) ? " raid1e" : "",
1651 imsm_orom_has_raid10(orom
) ? " raid10" : "",
1652 imsm_orom_has_raid5(orom
) ? " raid5" : "");
1653 printf(" Chunk Sizes :%s%s%s%s%s%s%s%s%s%s%s%s%s%s%s%s\n",
1654 imsm_orom_has_chunk(orom
, 2) ? " 2k" : "",
1655 imsm_orom_has_chunk(orom
, 4) ? " 4k" : "",
1656 imsm_orom_has_chunk(orom
, 8) ? " 8k" : "",
1657 imsm_orom_has_chunk(orom
, 16) ? " 16k" : "",
1658 imsm_orom_has_chunk(orom
, 32) ? " 32k" : "",
1659 imsm_orom_has_chunk(orom
, 64) ? " 64k" : "",
1660 imsm_orom_has_chunk(orom
, 128) ? " 128k" : "",
1661 imsm_orom_has_chunk(orom
, 256) ? " 256k" : "",
1662 imsm_orom_has_chunk(orom
, 512) ? " 512k" : "",
1663 imsm_orom_has_chunk(orom
, 1024*1) ? " 1M" : "",
1664 imsm_orom_has_chunk(orom
, 1024*2) ? " 2M" : "",
1665 imsm_orom_has_chunk(orom
, 1024*4) ? " 4M" : "",
1666 imsm_orom_has_chunk(orom
, 1024*8) ? " 8M" : "",
1667 imsm_orom_has_chunk(orom
, 1024*16) ? " 16M" : "",
1668 imsm_orom_has_chunk(orom
, 1024*32) ? " 32M" : "",
1669 imsm_orom_has_chunk(orom
, 1024*64) ? " 64M" : "");
1670 printf(" Max Disks : %d\n", orom
->tds
);
1671 printf(" Max Volumes : %d\n", orom
->vpa
);
1675 static int detail_platform_imsm(int verbose
, int enumerate_only
)
1677 /* There are two components to imsm platform support, the ahci SATA
1678 * controller and the option-rom. To find the SATA controller we
1679 * simply look in /sys/bus/pci/drivers/ahci to see if an ahci
1680 * controller with the Intel vendor id is present. This approach
1681 * allows mdadm to leverage the kernel's ahci detection logic, with the
1682 * caveat that if ahci.ko is not loaded mdadm will not be able to
1683 * detect platform raid capabilities. The option-rom resides in a
1684 * platform "Adapter ROM". We scan for its signature to retrieve the
1685 * platform capabilities. If raid support is disabled in the BIOS the
1686 * option-rom capability structure will not be available.
1688 const struct imsm_orom
*orom
;
1689 struct sys_dev
*list
, *hba
;
1694 if (enumerate_only
) {
1695 if (check_env("IMSM_NO_PLATFORM"))
1697 list
= find_intel_devices();
1700 for (hba
= list
; hba
; hba
= hba
->next
) {
1701 orom
= find_imsm_capability(hba
->type
);
1707 free_sys_dev(&list
);
1711 list
= find_intel_devices();
1714 fprintf(stderr
, Name
": no active Intel(R) RAID "
1715 "controller found.\n");
1716 free_sys_dev(&list
);
1719 print_found_intel_controllers(list
);
1721 for (hba
= list
; hba
; hba
= hba
->next
) {
1722 orom
= find_imsm_capability(hba
->type
);
1724 fprintf(stderr
, Name
": imsm capabilities not found for controller: %s (type %s)\n",
1725 hba
->path
, get_sys_dev_type(hba
->type
));
1727 print_imsm_capability(orom
);
1730 for (hba
= list
; hba
; hba
= hba
->next
) {
1731 printf(" I/O Controller : %s (%s)\n",
1732 hba
->path
, get_sys_dev_type(hba
->type
));
1734 if (hba
->type
== SYS_DEV_SATA
) {
1735 host_base
= ahci_get_port_count(hba
->path
, &port_count
);
1736 if (ahci_enumerate_ports(hba
->path
, port_count
, host_base
, verbose
)) {
1738 fprintf(stderr
, Name
": failed to enumerate "
1739 "ports on SATA controller at %s.", hba
->pci_id
);
1745 free_sys_dev(&list
);
1750 static int match_home_imsm(struct supertype
*st
, char *homehost
)
1752 /* the imsm metadata format does not specify any host
1753 * identification information. We return -1 since we can never
1754 * confirm nor deny whether a given array is "meant" for this
1755 * host. We rely on compare_super and the 'family_num' fields to
1756 * exclude member disks that do not belong, and we rely on
1757 * mdadm.conf to specify the arrays that should be assembled.
1758 * Auto-assembly may still pick up "foreign" arrays.
1764 static void uuid_from_super_imsm(struct supertype
*st
, int uuid
[4])
1766 /* The uuid returned here is used for:
1767 * uuid to put into bitmap file (Create, Grow)
1768 * uuid for backup header when saving critical section (Grow)
1769 * comparing uuids when re-adding a device into an array
1770 * In these cases the uuid required is that of the data-array,
1771 * not the device-set.
1772 * uuid to recognise same set when adding a missing device back
1773 * to an array. This is a uuid for the device-set.
1775 * For each of these we can make do with a truncated
1776 * or hashed uuid rather than the original, as long as
1778 * In each case the uuid required is that of the data-array,
1779 * not the device-set.
1781 /* imsm does not track uuid's so we synthesis one using sha1 on
1782 * - The signature (Which is constant for all imsm array, but no matter)
1783 * - the orig_family_num of the container
1784 * - the index number of the volume
1785 * - the 'serial' number of the volume.
1786 * Hopefully these are all constant.
1788 struct intel_super
*super
= st
->sb
;
1791 struct sha1_ctx ctx
;
1792 struct imsm_dev
*dev
= NULL
;
1795 /* some mdadm versions failed to set ->orig_family_num, in which
1796 * case fall back to ->family_num. orig_family_num will be
1797 * fixed up with the first metadata update.
1799 family_num
= super
->anchor
->orig_family_num
;
1800 if (family_num
== 0)
1801 family_num
= super
->anchor
->family_num
;
1802 sha1_init_ctx(&ctx
);
1803 sha1_process_bytes(super
->anchor
->sig
, MPB_SIG_LEN
, &ctx
);
1804 sha1_process_bytes(&family_num
, sizeof(__u32
), &ctx
);
1805 if (super
->current_vol
>= 0)
1806 dev
= get_imsm_dev(super
, super
->current_vol
);
1808 __u32 vol
= super
->current_vol
;
1809 sha1_process_bytes(&vol
, sizeof(vol
), &ctx
);
1810 sha1_process_bytes(dev
->volume
, MAX_RAID_SERIAL_LEN
, &ctx
);
1812 sha1_finish_ctx(&ctx
, buf
);
1813 memcpy(uuid
, buf
, 4*4);
1818 get_imsm_numerical_version(struct imsm_super
*mpb
, int *m
, int *p
)
1820 __u8
*v
= get_imsm_version(mpb
);
1821 __u8
*end
= mpb
->sig
+ MAX_SIGNATURE_LENGTH
;
1822 char major
[] = { 0, 0, 0 };
1823 char minor
[] = { 0 ,0, 0 };
1824 char patch
[] = { 0, 0, 0 };
1825 char *ver_parse
[] = { major
, minor
, patch
};
1829 while (*v
!= '\0' && v
< end
) {
1830 if (*v
!= '.' && j
< 2)
1831 ver_parse
[i
][j
++] = *v
;
1839 *m
= strtol(minor
, NULL
, 0);
1840 *p
= strtol(patch
, NULL
, 0);
1844 static __u32
migr_strip_blocks_resync(struct imsm_dev
*dev
)
1846 /* migr_strip_size when repairing or initializing parity */
1847 struct imsm_map
*map
= get_imsm_map(dev
, 0);
1848 __u32 chunk
= __le32_to_cpu(map
->blocks_per_strip
);
1850 switch (get_imsm_raid_level(map
)) {
1855 return 128*1024 >> 9;
1859 static __u32
migr_strip_blocks_rebuild(struct imsm_dev
*dev
)
1861 /* migr_strip_size when rebuilding a degraded disk, no idea why
1862 * this is different than migr_strip_size_resync(), but it's good
1865 struct imsm_map
*map
= get_imsm_map(dev
, 1);
1866 __u32 chunk
= __le32_to_cpu(map
->blocks_per_strip
);
1868 switch (get_imsm_raid_level(map
)) {
1871 if (map
->num_members
% map
->num_domains
== 0)
1872 return 128*1024 >> 9;
1876 return max((__u32
) 64*1024 >> 9, chunk
);
1878 return 128*1024 >> 9;
1882 static __u32
num_stripes_per_unit_resync(struct imsm_dev
*dev
)
1884 struct imsm_map
*lo
= get_imsm_map(dev
, 0);
1885 struct imsm_map
*hi
= get_imsm_map(dev
, 1);
1886 __u32 lo_chunk
= __le32_to_cpu(lo
->blocks_per_strip
);
1887 __u32 hi_chunk
= __le32_to_cpu(hi
->blocks_per_strip
);
1889 return max((__u32
) 1, hi_chunk
/ lo_chunk
);
1892 static __u32
num_stripes_per_unit_rebuild(struct imsm_dev
*dev
)
1894 struct imsm_map
*lo
= get_imsm_map(dev
, 0);
1895 int level
= get_imsm_raid_level(lo
);
1897 if (level
== 1 || level
== 10) {
1898 struct imsm_map
*hi
= get_imsm_map(dev
, 1);
1900 return hi
->num_domains
;
1902 return num_stripes_per_unit_resync(dev
);
1905 static __u8
imsm_num_data_members(struct imsm_dev
*dev
, int second_map
)
1907 /* named 'imsm_' because raid0, raid1 and raid10
1908 * counter-intuitively have the same number of data disks
1910 struct imsm_map
*map
= get_imsm_map(dev
, second_map
);
1912 switch (get_imsm_raid_level(map
)) {
1916 return map
->num_members
;
1918 return map
->num_members
- 1;
1920 dprintf("%s: unsupported raid level\n", __func__
);
1925 static __u32
parity_segment_depth(struct imsm_dev
*dev
)
1927 struct imsm_map
*map
= get_imsm_map(dev
, 0);
1928 __u32 chunk
= __le32_to_cpu(map
->blocks_per_strip
);
1930 switch(get_imsm_raid_level(map
)) {
1933 return chunk
* map
->num_domains
;
1935 return chunk
* map
->num_members
;
1941 static __u32
map_migr_block(struct imsm_dev
*dev
, __u32 block
)
1943 struct imsm_map
*map
= get_imsm_map(dev
, 1);
1944 __u32 chunk
= __le32_to_cpu(map
->blocks_per_strip
);
1945 __u32 strip
= block
/ chunk
;
1947 switch (get_imsm_raid_level(map
)) {
1950 __u32 vol_strip
= (strip
* map
->num_domains
) + 1;
1951 __u32 vol_stripe
= vol_strip
/ map
->num_members
;
1953 return vol_stripe
* chunk
+ block
% chunk
;
1955 __u32 stripe
= strip
/ (map
->num_members
- 1);
1957 return stripe
* chunk
+ block
% chunk
;
1964 static __u64
blocks_per_migr_unit(struct intel_super
*super
,
1965 struct imsm_dev
*dev
)
1967 /* calculate the conversion factor between per member 'blocks'
1968 * (md/{resync,rebuild}_start) and imsm migration units, return
1969 * 0 for the 'not migrating' and 'unsupported migration' cases
1971 if (!dev
->vol
.migr_state
)
1974 switch (migr_type(dev
)) {
1975 case MIGR_GEN_MIGR
: {
1976 struct migr_record
*migr_rec
= super
->migr_rec
;
1977 return __le32_to_cpu(migr_rec
->blocks_per_unit
);
1982 struct imsm_map
*map
= get_imsm_map(dev
, 0);
1983 __u32 stripes_per_unit
;
1984 __u32 blocks_per_unit
;
1993 /* yes, this is really the translation of migr_units to
1994 * per-member blocks in the 'resync' case
1996 stripes_per_unit
= num_stripes_per_unit_resync(dev
);
1997 migr_chunk
= migr_strip_blocks_resync(dev
);
1998 disks
= imsm_num_data_members(dev
, 0);
1999 blocks_per_unit
= stripes_per_unit
* migr_chunk
* disks
;
2000 stripe
= __le16_to_cpu(map
->blocks_per_strip
) * disks
;
2001 segment
= blocks_per_unit
/ stripe
;
2002 block_rel
= blocks_per_unit
- segment
* stripe
;
2003 parity_depth
= parity_segment_depth(dev
);
2004 block_map
= map_migr_block(dev
, block_rel
);
2005 return block_map
+ parity_depth
* segment
;
2007 case MIGR_REBUILD
: {
2008 __u32 stripes_per_unit
;
2011 stripes_per_unit
= num_stripes_per_unit_rebuild(dev
);
2012 migr_chunk
= migr_strip_blocks_rebuild(dev
);
2013 return migr_chunk
* stripes_per_unit
;
2015 case MIGR_STATE_CHANGE
:
2021 static int imsm_level_to_layout(int level
)
2029 return ALGORITHM_LEFT_ASYMMETRIC
;
2036 /*******************************************************************************
2037 * Function: read_imsm_migr_rec
2038 * Description: Function reads imsm migration record from last sector of disk
2040 * fd : disk descriptor
2041 * super : metadata info
2045 ******************************************************************************/
2046 static int read_imsm_migr_rec(int fd
, struct intel_super
*super
)
2049 unsigned long long dsize
;
2051 get_dev_size(fd
, NULL
, &dsize
);
2052 if (lseek64(fd
, dsize
- 512, SEEK_SET
) < 0) {
2054 Name
": Cannot seek to anchor block: %s\n",
2058 if (read(fd
, super
->migr_rec_buf
, 512) != 512) {
2060 Name
": Cannot read migr record block: %s\n",
2070 /*******************************************************************************
2071 * Function: load_imsm_migr_rec
2072 * Description: Function reads imsm migration record (it is stored at the last
2075 * super : imsm internal array info
2076 * info : general array info
2080 ******************************************************************************/
2081 static int load_imsm_migr_rec(struct intel_super
*super
, struct mdinfo
*info
)
2084 struct dl
*dl
= NULL
;
2090 for (sd
= info
->devs
; sd
; sd
= sd
->next
) {
2091 /* read only from one of the first two slots */
2092 if ((sd
->disk
.raid_disk
> 1) ||
2093 (sd
->disk
.raid_disk
< 0))
2095 sprintf(nm
, "%d:%d", sd
->disk
.major
, sd
->disk
.minor
);
2096 fd
= dev_open(nm
, O_RDONLY
);
2102 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
2103 /* read only from one of the first two slots */
2106 sprintf(nm
, "%d:%d", dl
->major
, dl
->minor
);
2107 fd
= dev_open(nm
, O_RDONLY
);
2114 retval
= read_imsm_migr_rec(fd
, super
);
2123 /*******************************************************************************
2124 * function: imsm_create_metadata_checkpoint_update
2125 * Description: It creates update for checkpoint change.
2127 * super : imsm internal array info
2128 * u : pointer to prepared update
2131 * If length is equal to 0, input pointer u contains no update
2132 ******************************************************************************/
2133 static int imsm_create_metadata_checkpoint_update(
2134 struct intel_super
*super
,
2135 struct imsm_update_general_migration_checkpoint
**u
)
2138 int update_memory_size
= 0;
2140 dprintf("imsm_create_metadata_checkpoint_update(enter)\n");
2146 /* size of all update data without anchor */
2147 update_memory_size
=
2148 sizeof(struct imsm_update_general_migration_checkpoint
);
2150 *u
= calloc(1, update_memory_size
);
2152 dprintf("error: cannot get memory for "
2153 "imsm_create_metadata_checkpoint_update update\n");
2156 (*u
)->type
= update_general_migration_checkpoint
;
2157 (*u
)->curr_migr_unit
= __le32_to_cpu(super
->migr_rec
->curr_migr_unit
);
2158 dprintf("imsm_create_metadata_checkpoint_update: prepared for %u\n",
2159 (*u
)->curr_migr_unit
);
2161 return update_memory_size
;
2165 static void imsm_update_metadata_locally(struct supertype
*st
,
2166 void *buf
, int len
);
2168 /*******************************************************************************
2169 * Function: write_imsm_migr_rec
2170 * Description: Function writes imsm migration record
2171 * (at the last sector of disk)
2173 * super : imsm internal array info
2177 ******************************************************************************/
2178 static int write_imsm_migr_rec(struct supertype
*st
)
2180 struct intel_super
*super
= st
->sb
;
2181 unsigned long long dsize
;
2187 struct imsm_update_general_migration_checkpoint
*u
;
2189 for (sd
= super
->disks
; sd
; sd
= sd
->next
) {
2190 /* write to 2 first slots only */
2191 if ((sd
->index
< 0) || (sd
->index
> 1))
2193 sprintf(nm
, "%d:%d", sd
->major
, sd
->minor
);
2194 fd
= dev_open(nm
, O_RDWR
);
2197 get_dev_size(fd
, NULL
, &dsize
);
2198 if (lseek64(fd
, dsize
- 512, SEEK_SET
) < 0) {
2200 Name
": Cannot seek to anchor block: %s\n",
2204 if (write(fd
, super
->migr_rec_buf
, 512) != 512) {
2206 Name
": Cannot write migr record block: %s\n",
2213 /* update checkpoint information in metadata */
2214 len
= imsm_create_metadata_checkpoint_update(super
, &u
);
2217 dprintf("imsm: Cannot prepare update\n");
2220 /* update metadata locally */
2221 imsm_update_metadata_locally(st
, u
, len
);
2222 /* and possibly remotely */
2223 if (st
->update_tail
) {
2224 append_metadata_update(st
, u
, len
);
2225 /* during reshape we do all work inside metadata handler
2226 * manage_reshape(), so metadata update has to be triggered
2229 flush_metadata_updates(st
);
2230 st
->update_tail
= &st
->updates
;
2240 #endif /* MDASSEMBLE */
2242 /* spare/missing disks activations are not allowe when
2243 * array/container performs reshape operation, because
2244 * all arrays in container works on the same disks set
2246 int imsm_reshape_blocks_arrays_changes(struct intel_super
*super
)
2249 struct intel_dev
*i_dev
;
2250 struct imsm_dev
*dev
;
2252 /* check whole container
2254 for (i_dev
= super
->devlist
; i_dev
; i_dev
= i_dev
->next
) {
2256 if (is_gen_migration(dev
)) {
2257 /* No repair during any migration in container
2266 static void getinfo_super_imsm_volume(struct supertype
*st
, struct mdinfo
*info
, char *dmap
)
2268 struct intel_super
*super
= st
->sb
;
2269 struct migr_record
*migr_rec
= super
->migr_rec
;
2270 struct imsm_dev
*dev
= get_imsm_dev(super
, super
->current_vol
);
2271 struct imsm_map
*map
= get_imsm_map(dev
, 0);
2272 struct imsm_map
*prev_map
= get_imsm_map(dev
, 1);
2273 struct imsm_map
*map_to_analyse
= map
;
2276 unsigned int component_size_alligment
;
2277 int map_disks
= info
->array
.raid_disks
;
2279 memset(info
, 0, sizeof(*info
));
2281 map_to_analyse
= prev_map
;
2283 dl
= super
->current_disk
;
2285 info
->container_member
= super
->current_vol
;
2286 info
->array
.raid_disks
= map
->num_members
;
2287 info
->array
.level
= get_imsm_raid_level(map_to_analyse
);
2288 info
->array
.layout
= imsm_level_to_layout(info
->array
.level
);
2289 info
->array
.md_minor
= -1;
2290 info
->array
.ctime
= 0;
2291 info
->array
.utime
= 0;
2292 info
->array
.chunk_size
=
2293 __le16_to_cpu(map_to_analyse
->blocks_per_strip
) << 9;
2294 info
->array
.state
= !dev
->vol
.dirty
;
2295 info
->custom_array_size
= __le32_to_cpu(dev
->size_high
);
2296 info
->custom_array_size
<<= 32;
2297 info
->custom_array_size
|= __le32_to_cpu(dev
->size_low
);
2298 info
->recovery_blocked
= imsm_reshape_blocks_arrays_changes(st
->sb
);
2300 if (prev_map
&& map
->map_state
== prev_map
->map_state
&&
2301 (migr_type(dev
) == MIGR_GEN_MIGR
)) {
2302 info
->reshape_active
= 1;
2303 info
->new_level
= get_imsm_raid_level(map
);
2304 info
->new_layout
= imsm_level_to_layout(info
->new_level
);
2305 info
->new_chunk
= __le16_to_cpu(map
->blocks_per_strip
) << 9;
2306 info
->delta_disks
= map
->num_members
- prev_map
->num_members
;
2307 if (info
->delta_disks
) {
2308 /* this needs to be applied to every array
2311 info
->reshape_active
= CONTAINER_RESHAPE
;
2313 /* We shape information that we give to md might have to be
2314 * modify to cope with md's requirement for reshaping arrays.
2315 * For example, when reshaping a RAID0, md requires it to be
2316 * presented as a degraded RAID4.
2317 * Also if a RAID0 is migrating to a RAID5 we need to specify
2318 * the array as already being RAID5, but the 'before' layout
2319 * is a RAID4-like layout.
2321 switch (info
->array
.level
) {
2323 switch(info
->new_level
) {
2325 /* conversion is happening as RAID4 */
2326 info
->array
.level
= 4;
2327 info
->array
.raid_disks
+= 1;
2330 /* conversion is happening as RAID5 */
2331 info
->array
.level
= 5;
2332 info
->array
.layout
= ALGORITHM_PARITY_N
;
2333 info
->delta_disks
-= 1;
2336 /* FIXME error message */
2337 info
->array
.level
= UnSet
;
2343 info
->new_level
= UnSet
;
2344 info
->new_layout
= UnSet
;
2345 info
->new_chunk
= info
->array
.chunk_size
;
2346 info
->delta_disks
= 0;
2350 info
->disk
.major
= dl
->major
;
2351 info
->disk
.minor
= dl
->minor
;
2352 info
->disk
.number
= dl
->index
;
2353 info
->disk
.raid_disk
= get_imsm_disk_slot(map_to_analyse
,
2357 info
->data_offset
= __le32_to_cpu(map_to_analyse
->pba_of_lba0
);
2358 info
->component_size
=
2359 __le32_to_cpu(map_to_analyse
->blocks_per_member
);
2361 /* check component size aligment
2363 component_size_alligment
=
2364 info
->component_size
% (info
->array
.chunk_size
/512);
2366 if (component_size_alligment
&&
2367 (info
->array
.level
!= 1) && (info
->array
.level
!= UnSet
)) {
2368 dprintf("imsm: reported component size alligned from %llu ",
2369 info
->component_size
);
2370 info
->component_size
-= component_size_alligment
;
2371 dprintf("to %llu (%i).\n",
2372 info
->component_size
, component_size_alligment
);
2375 memset(info
->uuid
, 0, sizeof(info
->uuid
));
2376 info
->recovery_start
= MaxSector
;
2378 info
->reshape_progress
= 0;
2379 info
->resync_start
= MaxSector
;
2380 if ((map_to_analyse
->map_state
== IMSM_T_STATE_UNINITIALIZED
||
2382 imsm_reshape_blocks_arrays_changes(super
) == 0) {
2383 info
->resync_start
= 0;
2385 if (dev
->vol
.migr_state
) {
2386 switch (migr_type(dev
)) {
2389 __u64 blocks_per_unit
= blocks_per_migr_unit(super
,
2391 __u64 units
= __le32_to_cpu(dev
->vol
.curr_migr_unit
);
2393 info
->resync_start
= blocks_per_unit
* units
;
2396 case MIGR_GEN_MIGR
: {
2397 __u64 blocks_per_unit
= blocks_per_migr_unit(super
,
2399 __u64 units
= __le32_to_cpu(migr_rec
->curr_migr_unit
);
2400 unsigned long long array_blocks
;
2403 if (__le32_to_cpu(migr_rec
->ascending_migr
) &&
2405 (__le32_to_cpu(migr_rec
->num_migr_units
)-1)) &&
2406 (super
->migr_rec
->rec_status
==
2407 __cpu_to_le32(UNIT_SRC_IN_CP_AREA
)))
2410 info
->reshape_progress
= blocks_per_unit
* units
;
2412 dprintf("IMSM: General Migration checkpoint : %llu "
2413 "(%llu) -> read reshape progress : %llu\n",
2414 (unsigned long long)units
,
2415 (unsigned long long)blocks_per_unit
,
2416 info
->reshape_progress
);
2418 used_disks
= imsm_num_data_members(dev
, 1);
2419 if (used_disks
> 0) {
2420 array_blocks
= map
->blocks_per_member
*
2422 /* round array size down to closest MB
2424 info
->custom_array_size
= (array_blocks
2425 >> SECT_PER_MB_SHIFT
)
2426 << SECT_PER_MB_SHIFT
;
2430 /* we could emulate the checkpointing of
2431 * 'sync_action=check' migrations, but for now
2432 * we just immediately complete them
2435 /* this is handled by container_content_imsm() */
2436 case MIGR_STATE_CHANGE
:
2437 /* FIXME handle other migrations */
2439 /* we are not dirty, so... */
2440 info
->resync_start
= MaxSector
;
2444 strncpy(info
->name
, (char *) dev
->volume
, MAX_RAID_SERIAL_LEN
);
2445 info
->name
[MAX_RAID_SERIAL_LEN
] = 0;
2447 info
->array
.major_version
= -1;
2448 info
->array
.minor_version
= -2;
2449 devname
= devnum2devname(st
->container_dev
);
2450 *info
->text_version
= '\0';
2452 sprintf(info
->text_version
, "/%s/%d", devname
, info
->container_member
);
2454 info
->safe_mode_delay
= 4000; /* 4 secs like the Matrix driver */
2455 uuid_from_super_imsm(st
, info
->uuid
);
2459 for (i
=0; i
<map_disks
; i
++) {
2461 if (i
< info
->array
.raid_disks
) {
2462 struct imsm_disk
*dsk
;
2463 j
= get_imsm_disk_idx(dev
, i
, -1);
2464 dsk
= get_imsm_disk(super
, j
);
2465 if (dsk
&& (dsk
->status
& CONFIGURED_DISK
))
2472 static __u8
imsm_check_degraded(struct intel_super
*super
, struct imsm_dev
*dev
, int failed
);
2473 static int imsm_count_failed(struct intel_super
*super
, struct imsm_dev
*dev
);
2475 static struct imsm_disk
*get_imsm_missing(struct intel_super
*super
, __u8 index
)
2479 for (d
= super
->missing
; d
; d
= d
->next
)
2480 if (d
->index
== index
)
2485 static void getinfo_super_imsm(struct supertype
*st
, struct mdinfo
*info
, char *map
)
2487 struct intel_super
*super
= st
->sb
;
2488 struct imsm_disk
*disk
;
2489 int map_disks
= info
->array
.raid_disks
;
2490 int max_enough
= -1;
2492 struct imsm_super
*mpb
;
2494 if (super
->current_vol
>= 0) {
2495 getinfo_super_imsm_volume(st
, info
, map
);
2498 memset(info
, 0, sizeof(*info
));
2500 /* Set raid_disks to zero so that Assemble will always pull in valid
2503 info
->array
.raid_disks
= 0;
2504 info
->array
.level
= LEVEL_CONTAINER
;
2505 info
->array
.layout
= 0;
2506 info
->array
.md_minor
= -1;
2507 info
->array
.ctime
= 0; /* N/A for imsm */
2508 info
->array
.utime
= 0;
2509 info
->array
.chunk_size
= 0;
2511 info
->disk
.major
= 0;
2512 info
->disk
.minor
= 0;
2513 info
->disk
.raid_disk
= -1;
2514 info
->reshape_active
= 0;
2515 info
->array
.major_version
= -1;
2516 info
->array
.minor_version
= -2;
2517 strcpy(info
->text_version
, "imsm");
2518 info
->safe_mode_delay
= 0;
2519 info
->disk
.number
= -1;
2520 info
->disk
.state
= 0;
2522 info
->recovery_start
= MaxSector
;
2523 info
->recovery_blocked
= imsm_reshape_blocks_arrays_changes(st
->sb
);
2525 /* do we have the all the insync disks that we expect? */
2526 mpb
= super
->anchor
;
2528 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
2529 struct imsm_dev
*dev
= get_imsm_dev(super
, i
);
2530 int failed
, enough
, j
, missing
= 0;
2531 struct imsm_map
*map
;
2534 failed
= imsm_count_failed(super
, dev
);
2535 state
= imsm_check_degraded(super
, dev
, failed
);
2536 map
= get_imsm_map(dev
, 0);
2538 /* any newly missing disks?
2539 * (catches single-degraded vs double-degraded)
2541 for (j
= 0; j
< map
->num_members
; j
++) {
2542 __u32 ord
= get_imsm_ord_tbl_ent(dev
, j
, 0);
2543 __u32 idx
= ord_to_idx(ord
);
2545 if (!(ord
& IMSM_ORD_REBUILD
) &&
2546 get_imsm_missing(super
, idx
)) {
2552 if (state
== IMSM_T_STATE_FAILED
)
2554 else if (state
== IMSM_T_STATE_DEGRADED
&&
2555 (state
!= map
->map_state
|| missing
))
2557 else /* we're normal, or already degraded */
2560 /* in the missing/failed disk case check to see
2561 * if at least one array is runnable
2563 max_enough
= max(max_enough
, enough
);
2565 dprintf("%s: enough: %d\n", __func__
, max_enough
);
2566 info
->container_enough
= max_enough
;
2569 __u32 reserved
= imsm_reserved_sectors(super
, super
->disks
);
2571 disk
= &super
->disks
->disk
;
2572 info
->data_offset
= __le32_to_cpu(disk
->total_blocks
) - reserved
;
2573 info
->component_size
= reserved
;
2574 info
->disk
.state
= is_configured(disk
) ? (1 << MD_DISK_ACTIVE
) : 0;
2575 /* we don't change info->disk.raid_disk here because
2576 * this state will be finalized in mdmon after we have
2577 * found the 'most fresh' version of the metadata
2579 info
->disk
.state
|= is_failed(disk
) ? (1 << MD_DISK_FAULTY
) : 0;
2580 info
->disk
.state
|= is_spare(disk
) ? 0 : (1 << MD_DISK_SYNC
);
2583 /* only call uuid_from_super_imsm when this disk is part of a populated container,
2584 * ->compare_super may have updated the 'num_raid_devs' field for spares
2586 if (info
->disk
.state
& (1 << MD_DISK_SYNC
) || super
->anchor
->num_raid_devs
)
2587 uuid_from_super_imsm(st
, info
->uuid
);
2589 memcpy(info
->uuid
, uuid_zero
, sizeof(uuid_zero
));
2591 /* I don't know how to compute 'map' on imsm, so use safe default */
2594 for (i
= 0; i
< map_disks
; i
++)
2600 /* allocates memory and fills disk in mdinfo structure
2601 * for each disk in array */
2602 struct mdinfo
*getinfo_super_disks_imsm(struct supertype
*st
)
2604 struct mdinfo
*mddev
= NULL
;
2605 struct intel_super
*super
= st
->sb
;
2606 struct imsm_disk
*disk
;
2609 if (!super
|| !super
->disks
)
2612 mddev
= malloc(sizeof(*mddev
));
2614 fprintf(stderr
, Name
": Failed to allocate memory.\n");
2617 memset(mddev
, 0, sizeof(*mddev
));
2621 tmp
= malloc(sizeof(*tmp
));
2623 fprintf(stderr
, Name
": Failed to allocate memory.\n");
2628 memset(tmp
, 0, sizeof(*tmp
));
2630 tmp
->next
= mddev
->devs
;
2632 tmp
->disk
.number
= count
++;
2633 tmp
->disk
.major
= dl
->major
;
2634 tmp
->disk
.minor
= dl
->minor
;
2635 tmp
->disk
.state
= is_configured(disk
) ?
2636 (1 << MD_DISK_ACTIVE
) : 0;
2637 tmp
->disk
.state
|= is_failed(disk
) ? (1 << MD_DISK_FAULTY
) : 0;
2638 tmp
->disk
.state
|= is_spare(disk
) ? 0 : (1 << MD_DISK_SYNC
);
2639 tmp
->disk
.raid_disk
= -1;
2645 static int update_super_imsm(struct supertype
*st
, struct mdinfo
*info
,
2646 char *update
, char *devname
, int verbose
,
2647 int uuid_set
, char *homehost
)
2649 /* For 'assemble' and 'force' we need to return non-zero if any
2650 * change was made. For others, the return value is ignored.
2651 * Update options are:
2652 * force-one : This device looks a bit old but needs to be included,
2653 * update age info appropriately.
2654 * assemble: clear any 'faulty' flag to allow this device to
2656 * force-array: Array is degraded but being forced, mark it clean
2657 * if that will be needed to assemble it.
2659 * newdev: not used ????
2660 * grow: Array has gained a new device - this is currently for
2662 * resync: mark as dirty so a resync will happen.
2663 * name: update the name - preserving the homehost
2664 * uuid: Change the uuid of the array to match watch is given
2666 * Following are not relevant for this imsm:
2667 * sparc2.2 : update from old dodgey metadata
2668 * super-minor: change the preferred_minor number
2669 * summaries: update redundant counters.
2670 * homehost: update the recorded homehost
2671 * _reshape_progress: record new reshape_progress position.
2674 struct intel_super
*super
= st
->sb
;
2675 struct imsm_super
*mpb
;
2677 /* we can only update container info */
2678 if (!super
|| super
->current_vol
>= 0 || !super
->anchor
)
2681 mpb
= super
->anchor
;
2683 if (strcmp(update
, "uuid") == 0 && uuid_set
&& !info
->update_private
)
2685 else if (strcmp(update
, "uuid") == 0 && uuid_set
&& info
->update_private
) {
2686 mpb
->orig_family_num
= *((__u32
*) info
->update_private
);
2688 } else if (strcmp(update
, "uuid") == 0) {
2689 __u32
*new_family
= malloc(sizeof(*new_family
));
2691 /* update orig_family_number with the incoming random
2692 * data, report the new effective uuid, and store the
2693 * new orig_family_num for future updates.
2696 memcpy(&mpb
->orig_family_num
, info
->uuid
, sizeof(__u32
));
2697 uuid_from_super_imsm(st
, info
->uuid
);
2698 *new_family
= mpb
->orig_family_num
;
2699 info
->update_private
= new_family
;
2702 } else if (strcmp(update
, "assemble") == 0)
2707 /* successful update? recompute checksum */
2709 mpb
->check_sum
= __le32_to_cpu(__gen_imsm_checksum(mpb
));
2714 static size_t disks_to_mpb_size(int disks
)
2718 size
= sizeof(struct imsm_super
);
2719 size
+= (disks
- 1) * sizeof(struct imsm_disk
);
2720 size
+= 2 * sizeof(struct imsm_dev
);
2721 /* up to 2 maps per raid device (-2 for imsm_maps in imsm_dev */
2722 size
+= (4 - 2) * sizeof(struct imsm_map
);
2723 /* 4 possible disk_ord_tbl's */
2724 size
+= 4 * (disks
- 1) * sizeof(__u32
);
2729 static __u64
avail_size_imsm(struct supertype
*st
, __u64 devsize
)
2731 if (devsize
< (MPB_SECTOR_CNT
+ IMSM_RESERVED_SECTORS
))
2734 return devsize
- (MPB_SECTOR_CNT
+ IMSM_RESERVED_SECTORS
);
2737 static void free_devlist(struct intel_super
*super
)
2739 struct intel_dev
*dv
;
2741 while (super
->devlist
) {
2742 dv
= super
->devlist
->next
;
2743 free(super
->devlist
->dev
);
2744 free(super
->devlist
);
2745 super
->devlist
= dv
;
2749 static void imsm_copy_dev(struct imsm_dev
*dest
, struct imsm_dev
*src
)
2751 memcpy(dest
, src
, sizeof_imsm_dev(src
, 0));
2754 static int compare_super_imsm(struct supertype
*st
, struct supertype
*tst
)
2758 * 0 same, or first was empty, and second was copied
2759 * 1 second had wrong number
2761 * 3 wrong other info
2763 struct intel_super
*first
= st
->sb
;
2764 struct intel_super
*sec
= tst
->sb
;
2771 /* in platform dependent environment test if the disks
2772 * use the same Intel hba
2774 if (!check_env("IMSM_NO_PLATFORM")) {
2775 if (!first
->hba
|| !sec
->hba
||
2776 (first
->hba
->type
!= sec
->hba
->type
)) {
2778 "HBAs of devices does not match %s != %s\n",
2779 first
->hba
? get_sys_dev_type(first
->hba
->type
) : NULL
,
2780 sec
->hba
? get_sys_dev_type(sec
->hba
->type
) : NULL
);
2785 /* if an anchor does not have num_raid_devs set then it is a free
2788 if (first
->anchor
->num_raid_devs
> 0 &&
2789 sec
->anchor
->num_raid_devs
> 0) {
2790 /* Determine if these disks might ever have been
2791 * related. Further disambiguation can only take place
2792 * in load_super_imsm_all
2794 __u32 first_family
= first
->anchor
->orig_family_num
;
2795 __u32 sec_family
= sec
->anchor
->orig_family_num
;
2797 if (memcmp(first
->anchor
->sig
, sec
->anchor
->sig
,
2798 MAX_SIGNATURE_LENGTH
) != 0)
2801 if (first_family
== 0)
2802 first_family
= first
->anchor
->family_num
;
2803 if (sec_family
== 0)
2804 sec_family
= sec
->anchor
->family_num
;
2806 if (first_family
!= sec_family
)
2812 /* if 'first' is a spare promote it to a populated mpb with sec's
2815 if (first
->anchor
->num_raid_devs
== 0 &&
2816 sec
->anchor
->num_raid_devs
> 0) {
2818 struct intel_dev
*dv
;
2819 struct imsm_dev
*dev
;
2821 /* we need to copy raid device info from sec if an allocation
2822 * fails here we don't associate the spare
2824 for (i
= 0; i
< sec
->anchor
->num_raid_devs
; i
++) {
2825 dv
= malloc(sizeof(*dv
));
2828 dev
= malloc(sizeof_imsm_dev(get_imsm_dev(sec
, i
), 1));
2835 dv
->next
= first
->devlist
;
2836 first
->devlist
= dv
;
2838 if (i
< sec
->anchor
->num_raid_devs
) {
2839 /* allocation failure */
2840 free_devlist(first
);
2841 fprintf(stderr
, "imsm: failed to associate spare\n");
2844 first
->anchor
->num_raid_devs
= sec
->anchor
->num_raid_devs
;
2845 first
->anchor
->orig_family_num
= sec
->anchor
->orig_family_num
;
2846 first
->anchor
->family_num
= sec
->anchor
->family_num
;
2847 memcpy(first
->anchor
->sig
, sec
->anchor
->sig
, MAX_SIGNATURE_LENGTH
);
2848 for (i
= 0; i
< sec
->anchor
->num_raid_devs
; i
++)
2849 imsm_copy_dev(get_imsm_dev(first
, i
), get_imsm_dev(sec
, i
));
2855 static void fd2devname(int fd
, char *name
)
2859 char dname
[PATH_MAX
];
2864 if (fstat(fd
, &st
) != 0)
2866 sprintf(path
, "/sys/dev/block/%d:%d",
2867 major(st
.st_rdev
), minor(st
.st_rdev
));
2869 rv
= readlink(path
, dname
, sizeof(dname
)-1);
2874 nm
= strrchr(dname
, '/');
2877 snprintf(name
, MAX_RAID_SERIAL_LEN
, "/dev/%s", nm
);
2881 extern int scsi_get_serial(int fd
, void *buf
, size_t buf_len
);
2883 static int imsm_read_serial(int fd
, char *devname
,
2884 __u8 serial
[MAX_RAID_SERIAL_LEN
])
2886 unsigned char scsi_serial
[255];
2895 memset(scsi_serial
, 0, sizeof(scsi_serial
));
2897 rv
= scsi_get_serial(fd
, scsi_serial
, sizeof(scsi_serial
));
2899 if (rv
&& check_env("IMSM_DEVNAME_AS_SERIAL")) {
2900 memset(serial
, 0, MAX_RAID_SERIAL_LEN
);
2901 fd2devname(fd
, (char *) serial
);
2908 Name
": Failed to retrieve serial for %s\n",
2913 rsp_len
= scsi_serial
[3];
2917 Name
": Failed to retrieve serial for %s\n",
2921 rsp_buf
= (char *) &scsi_serial
[4];
2923 /* trim all whitespace and non-printable characters and convert
2926 for (i
= 0, dest
= rsp_buf
; i
< rsp_len
; i
++) {
2929 /* ':' is reserved for use in placeholder serial
2930 * numbers for missing disks
2938 len
= dest
- rsp_buf
;
2941 /* truncate leading characters */
2942 if (len
> MAX_RAID_SERIAL_LEN
) {
2943 dest
+= len
- MAX_RAID_SERIAL_LEN
;
2944 len
= MAX_RAID_SERIAL_LEN
;
2947 memset(serial
, 0, MAX_RAID_SERIAL_LEN
);
2948 memcpy(serial
, dest
, len
);
2953 static int serialcmp(__u8
*s1
, __u8
*s2
)
2955 return strncmp((char *) s1
, (char *) s2
, MAX_RAID_SERIAL_LEN
);
2958 static void serialcpy(__u8
*dest
, __u8
*src
)
2960 strncpy((char *) dest
, (char *) src
, MAX_RAID_SERIAL_LEN
);
2963 static struct dl
*serial_to_dl(__u8
*serial
, struct intel_super
*super
)
2967 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
2968 if (serialcmp(dl
->serial
, serial
) == 0)
2974 static struct imsm_disk
*
2975 __serial_to_disk(__u8
*serial
, struct imsm_super
*mpb
, int *idx
)
2979 for (i
= 0; i
< mpb
->num_disks
; i
++) {
2980 struct imsm_disk
*disk
= __get_imsm_disk(mpb
, i
);
2982 if (serialcmp(disk
->serial
, serial
) == 0) {
2993 load_imsm_disk(int fd
, struct intel_super
*super
, char *devname
, int keep_fd
)
2995 struct imsm_disk
*disk
;
3000 __u8 serial
[MAX_RAID_SERIAL_LEN
];
3002 rv
= imsm_read_serial(fd
, devname
, serial
);
3007 dl
= calloc(1, sizeof(*dl
));
3011 Name
": failed to allocate disk buffer for %s\n",
3017 dl
->major
= major(stb
.st_rdev
);
3018 dl
->minor
= minor(stb
.st_rdev
);
3019 dl
->next
= super
->disks
;
3020 dl
->fd
= keep_fd
? fd
: -1;
3021 assert(super
->disks
== NULL
);
3023 serialcpy(dl
->serial
, serial
);
3026 fd2devname(fd
, name
);
3028 dl
->devname
= strdup(devname
);
3030 dl
->devname
= strdup(name
);
3032 /* look up this disk's index in the current anchor */
3033 disk
= __serial_to_disk(dl
->serial
, super
->anchor
, &dl
->index
);
3036 /* only set index on disks that are a member of a
3037 * populated contianer, i.e. one with raid_devs
3039 if (is_failed(&dl
->disk
))
3041 else if (is_spare(&dl
->disk
))
3049 /* When migrating map0 contains the 'destination' state while map1
3050 * contains the current state. When not migrating map0 contains the
3051 * current state. This routine assumes that map[0].map_state is set to
3052 * the current array state before being called.
3054 * Migration is indicated by one of the following states
3055 * 1/ Idle (migr_state=0 map0state=normal||unitialized||degraded||failed)
3056 * 2/ Initialize (migr_state=1 migr_type=MIGR_INIT map0state=normal
3057 * map1state=unitialized)
3058 * 3/ Repair (Resync) (migr_state=1 migr_type=MIGR_REPAIR map0state=normal
3060 * 4/ Rebuild (migr_state=1 migr_type=MIGR_REBUILD map0state=normal
3061 * map1state=degraded)
3062 * 5/ Migration (mig_state=1 migr_type=MIGR_GEN_MIGR map0state=normal
3065 static void migrate(struct imsm_dev
*dev
, struct intel_super
*super
,
3066 __u8 to_state
, int migr_type
)
3068 struct imsm_map
*dest
;
3069 struct imsm_map
*src
= get_imsm_map(dev
, 0);
3071 dev
->vol
.migr_state
= 1;
3072 set_migr_type(dev
, migr_type
);
3073 dev
->vol
.curr_migr_unit
= 0;
3074 dest
= get_imsm_map(dev
, 1);
3076 /* duplicate and then set the target end state in map[0] */
3077 memcpy(dest
, src
, sizeof_imsm_map(src
));
3078 if ((migr_type
== MIGR_REBUILD
) ||
3079 (migr_type
== MIGR_GEN_MIGR
)) {
3083 for (i
= 0; i
< src
->num_members
; i
++) {
3084 ord
= __le32_to_cpu(src
->disk_ord_tbl
[i
]);
3085 set_imsm_ord_tbl_ent(src
, i
, ord_to_idx(ord
));
3089 if (migr_type
== MIGR_GEN_MIGR
)
3090 /* Clear migration record */
3091 memset(super
->migr_rec
, 0, sizeof(struct migr_record
));
3093 src
->map_state
= to_state
;
3096 static void end_migration(struct imsm_dev
*dev
, __u8 map_state
)
3098 struct imsm_map
*map
= get_imsm_map(dev
, 0);
3099 struct imsm_map
*prev
= get_imsm_map(dev
, dev
->vol
.migr_state
);
3102 /* merge any IMSM_ORD_REBUILD bits that were not successfully
3103 * completed in the last migration.
3105 * FIXME add support for raid-level-migration
3107 for (i
= 0; i
< prev
->num_members
; i
++)
3108 for (j
= 0; j
< map
->num_members
; j
++)
3109 /* during online capacity expansion
3110 * disks position can be changed if takeover is used
3112 if (ord_to_idx(map
->disk_ord_tbl
[j
]) ==
3113 ord_to_idx(prev
->disk_ord_tbl
[i
])) {
3114 map
->disk_ord_tbl
[j
] |= prev
->disk_ord_tbl
[i
];
3118 dev
->vol
.migr_state
= 0;
3119 set_migr_type(dev
, 0);
3120 dev
->vol
.curr_migr_unit
= 0;
3121 map
->map_state
= map_state
;
3125 static int parse_raid_devices(struct intel_super
*super
)
3128 struct imsm_dev
*dev_new
;
3129 size_t len
, len_migr
;
3131 size_t space_needed
= 0;
3132 struct imsm_super
*mpb
= super
->anchor
;
3134 for (i
= 0; i
< super
->anchor
->num_raid_devs
; i
++) {
3135 struct imsm_dev
*dev_iter
= __get_imsm_dev(super
->anchor
, i
);
3136 struct intel_dev
*dv
;
3138 len
= sizeof_imsm_dev(dev_iter
, 0);
3139 len_migr
= sizeof_imsm_dev(dev_iter
, 1);
3141 space_needed
+= len_migr
- len
;
3143 dv
= malloc(sizeof(*dv
));
3146 if (max_len
< len_migr
)
3148 if (max_len
> len_migr
)
3149 space_needed
+= max_len
- len_migr
;
3150 dev_new
= malloc(max_len
);
3155 imsm_copy_dev(dev_new
, dev_iter
);
3158 dv
->next
= super
->devlist
;
3159 super
->devlist
= dv
;
3162 /* ensure that super->buf is large enough when all raid devices
3165 if (__le32_to_cpu(mpb
->mpb_size
) + space_needed
> super
->len
) {
3168 len
= ROUND_UP(__le32_to_cpu(mpb
->mpb_size
) + space_needed
, 512);
3169 if (posix_memalign(&buf
, 512, len
) != 0)
3172 memcpy(buf
, super
->buf
, super
->len
);
3173 memset(buf
+ super
->len
, 0, len
- super
->len
);
3182 /* retrieve a pointer to the bbm log which starts after all raid devices */
3183 struct bbm_log
*__get_imsm_bbm_log(struct imsm_super
*mpb
)
3187 if (__le32_to_cpu(mpb
->bbm_log_size
)) {
3189 ptr
+= mpb
->mpb_size
- __le32_to_cpu(mpb
->bbm_log_size
);
3195 /*******************************************************************************
3196 * Function: check_mpb_migr_compatibility
3197 * Description: Function checks for unsupported migration features:
3198 * - migration optimization area (pba_of_lba0)
3199 * - descending reshape (ascending_migr)
3201 * super : imsm metadata information
3203 * 0 : migration is compatible
3204 * -1 : migration is not compatible
3205 ******************************************************************************/
3206 int check_mpb_migr_compatibility(struct intel_super
*super
)
3208 struct imsm_map
*map0
, *map1
;
3209 struct migr_record
*migr_rec
= super
->migr_rec
;
3212 for (i
= 0; i
< super
->anchor
->num_raid_devs
; i
++) {
3213 struct imsm_dev
*dev_iter
= __get_imsm_dev(super
->anchor
, i
);
3216 dev_iter
->vol
.migr_state
== 1 &&
3217 dev_iter
->vol
.migr_type
== MIGR_GEN_MIGR
) {
3218 /* This device is migrating */
3219 map0
= get_imsm_map(dev_iter
, 0);
3220 map1
= get_imsm_map(dev_iter
, 1);
3221 if (map0
->pba_of_lba0
!= map1
->pba_of_lba0
)
3222 /* migration optimization area was used */
3224 if (migr_rec
->ascending_migr
== 0
3225 && migr_rec
->dest_depth_per_unit
> 0)
3226 /* descending reshape not supported yet */
3233 static void __free_imsm(struct intel_super
*super
, int free_disks
);
3235 /* load_imsm_mpb - read matrix metadata
3236 * allocates super->mpb to be freed by free_imsm
3238 static int load_imsm_mpb(int fd
, struct intel_super
*super
, char *devname
)
3240 unsigned long long dsize
;
3241 unsigned long long sectors
;
3243 struct imsm_super
*anchor
;
3246 get_dev_size(fd
, NULL
, &dsize
);
3250 Name
": %s: device to small for imsm\n",
3255 if (lseek64(fd
, dsize
- (512 * 2), SEEK_SET
) < 0) {
3257 fprintf(stderr
, Name
3258 ": Cannot seek to anchor block on %s: %s\n",
3259 devname
, strerror(errno
));
3263 if (posix_memalign((void**)&anchor
, 512, 512) != 0) {
3266 Name
": Failed to allocate imsm anchor buffer"
3267 " on %s\n", devname
);
3270 if (read(fd
, anchor
, 512) != 512) {
3273 Name
": Cannot read anchor block on %s: %s\n",
3274 devname
, strerror(errno
));
3279 if (strncmp((char *) anchor
->sig
, MPB_SIGNATURE
, MPB_SIG_LEN
) != 0) {
3282 Name
": no IMSM anchor on %s\n", devname
);
3287 __free_imsm(super
, 0);
3288 /* reload capability and hba */
3290 /* capability and hba must be updated with new super allocation */
3291 find_intel_hba_capability(fd
, super
, devname
);
3292 super
->len
= ROUND_UP(anchor
->mpb_size
, 512);
3293 if (posix_memalign(&super
->buf
, 512, super
->len
) != 0) {
3296 Name
": unable to allocate %zu byte mpb buffer\n",
3301 memcpy(super
->buf
, anchor
, 512);
3303 sectors
= mpb_sectors(anchor
) - 1;
3306 if (posix_memalign(&super
->migr_rec_buf
, 512, 512) != 0) {
3307 fprintf(stderr
, Name
3308 ": %s could not allocate migr_rec buffer\n", __func__
);
3314 check_sum
= __gen_imsm_checksum(super
->anchor
);
3315 if (check_sum
!= __le32_to_cpu(super
->anchor
->check_sum
)) {
3318 Name
": IMSM checksum %x != %x on %s\n",
3320 __le32_to_cpu(super
->anchor
->check_sum
),
3328 /* read the extended mpb */
3329 if (lseek64(fd
, dsize
- (512 * (2 + sectors
)), SEEK_SET
) < 0) {
3332 Name
": Cannot seek to extended mpb on %s: %s\n",
3333 devname
, strerror(errno
));
3337 if ((unsigned)read(fd
, super
->buf
+ 512, super
->len
- 512) != super
->len
- 512) {
3340 Name
": Cannot read extended mpb on %s: %s\n",
3341 devname
, strerror(errno
));
3345 check_sum
= __gen_imsm_checksum(super
->anchor
);
3346 if (check_sum
!= __le32_to_cpu(super
->anchor
->check_sum
)) {
3349 Name
": IMSM checksum %x != %x on %s\n",
3350 check_sum
, __le32_to_cpu(super
->anchor
->check_sum
),
3355 /* FIXME the BBM log is disk specific so we cannot use this global
3356 * buffer for all disks. Ok for now since we only look at the global
3357 * bbm_log_size parameter to gate assembly
3359 super
->bbm_log
= __get_imsm_bbm_log(super
->anchor
);
3364 static int read_imsm_migr_rec(int fd
, struct intel_super
*super
);
3367 load_and_parse_mpb(int fd
, struct intel_super
*super
, char *devname
, int keep_fd
)
3371 err
= load_imsm_mpb(fd
, super
, devname
);
3374 err
= load_imsm_disk(fd
, super
, devname
, keep_fd
);
3377 err
= parse_raid_devices(super
);
3382 static void __free_imsm_disk(struct dl
*d
)
3394 static void free_imsm_disks(struct intel_super
*super
)
3398 while (super
->disks
) {
3400 super
->disks
= d
->next
;
3401 __free_imsm_disk(d
);
3403 while (super
->disk_mgmt_list
) {
3404 d
= super
->disk_mgmt_list
;
3405 super
->disk_mgmt_list
= d
->next
;
3406 __free_imsm_disk(d
);
3408 while (super
->missing
) {
3410 super
->missing
= d
->next
;
3411 __free_imsm_disk(d
);
3416 /* free all the pieces hanging off of a super pointer */
3417 static void __free_imsm(struct intel_super
*super
, int free_disks
)
3419 struct intel_hba
*elem
, *next
;
3425 /* unlink capability description */
3427 if (super
->migr_rec_buf
) {
3428 free(super
->migr_rec_buf
);
3429 super
->migr_rec_buf
= NULL
;
3432 free_imsm_disks(super
);
3433 free_devlist(super
);
3437 free((void *)elem
->path
);
3445 static void free_imsm(struct intel_super
*super
)
3447 __free_imsm(super
, 1);
3451 static void free_super_imsm(struct supertype
*st
)
3453 struct intel_super
*super
= st
->sb
;
3462 static struct intel_super
*alloc_super(void)
3464 struct intel_super
*super
= malloc(sizeof(*super
));
3467 memset(super
, 0, sizeof(*super
));
3468 super
->current_vol
= -1;
3469 super
->create_offset
= ~((__u32
) 0);
3475 * find and allocate hba and OROM/EFI based on valid fd of RAID component device
3477 static int find_intel_hba_capability(int fd
, struct intel_super
*super
, char *devname
)
3479 struct sys_dev
*hba_name
;
3482 if ((fd
< 0) || check_env("IMSM_NO_PLATFORM")) {
3487 hba_name
= find_disk_attached_hba(fd
, NULL
);
3491 Name
": %s is not attached to Intel(R) RAID controller.\n",
3495 rv
= attach_hba_to_super(super
, hba_name
);
3498 struct intel_hba
*hba
= super
->hba
;
3500 fprintf(stderr
, Name
": %s is attached to Intel(R) %s RAID "
3501 "controller (%s),\n"
3502 " but the container is assigned to Intel(R) "
3503 "%s RAID controller (",
3506 hba_name
->pci_id
? : "Err!",
3507 get_sys_dev_type(hba_name
->type
));
3510 fprintf(stderr
, "%s", hba
->pci_id
? : "Err!");
3512 fprintf(stderr
, ", ");
3516 fprintf(stderr
, ").\n"
3517 " Mixing devices attached to different controllers "
3518 "is not allowed.\n");
3520 free_sys_dev(&hba_name
);
3523 super
->orom
= find_imsm_capability(hba_name
->type
);
3524 free_sys_dev(&hba_name
);
3530 /* find_missing - helper routine for load_super_imsm_all that identifies
3531 * disks that have disappeared from the system. This routine relies on
3532 * the mpb being uptodate, which it is at load time.
3534 static int find_missing(struct intel_super
*super
)
3537 struct imsm_super
*mpb
= super
->anchor
;
3539 struct imsm_disk
*disk
;
3541 for (i
= 0; i
< mpb
->num_disks
; i
++) {
3542 disk
= __get_imsm_disk(mpb
, i
);
3543 dl
= serial_to_dl(disk
->serial
, super
);
3547 dl
= malloc(sizeof(*dl
));
3553 dl
->devname
= strdup("missing");
3555 serialcpy(dl
->serial
, disk
->serial
);
3558 dl
->next
= super
->missing
;
3559 super
->missing
= dl
;
3566 static struct intel_disk
*disk_list_get(__u8
*serial
, struct intel_disk
*disk_list
)
3568 struct intel_disk
*idisk
= disk_list
;
3571 if (serialcmp(idisk
->disk
.serial
, serial
) == 0)
3573 idisk
= idisk
->next
;
3579 static int __prep_thunderdome(struct intel_super
**table
, int tbl_size
,
3580 struct intel_super
*super
,
3581 struct intel_disk
**disk_list
)
3583 struct imsm_disk
*d
= &super
->disks
->disk
;
3584 struct imsm_super
*mpb
= super
->anchor
;
3587 for (i
= 0; i
< tbl_size
; i
++) {
3588 struct imsm_super
*tbl_mpb
= table
[i
]->anchor
;
3589 struct imsm_disk
*tbl_d
= &table
[i
]->disks
->disk
;
3591 if (tbl_mpb
->family_num
== mpb
->family_num
) {
3592 if (tbl_mpb
->check_sum
== mpb
->check_sum
) {
3593 dprintf("%s: mpb from %d:%d matches %d:%d\n",
3594 __func__
, super
->disks
->major
,
3595 super
->disks
->minor
,
3596 table
[i
]->disks
->major
,
3597 table
[i
]->disks
->minor
);
3601 if (((is_configured(d
) && !is_configured(tbl_d
)) ||
3602 is_configured(d
) == is_configured(tbl_d
)) &&
3603 tbl_mpb
->generation_num
< mpb
->generation_num
) {
3604 /* current version of the mpb is a
3605 * better candidate than the one in
3606 * super_table, but copy over "cross
3607 * generational" status
3609 struct intel_disk
*idisk
;
3611 dprintf("%s: mpb from %d:%d replaces %d:%d\n",
3612 __func__
, super
->disks
->major
,
3613 super
->disks
->minor
,
3614 table
[i
]->disks
->major
,
3615 table
[i
]->disks
->minor
);
3617 idisk
= disk_list_get(tbl_d
->serial
, *disk_list
);
3618 if (idisk
&& is_failed(&idisk
->disk
))
3619 tbl_d
->status
|= FAILED_DISK
;
3622 struct intel_disk
*idisk
;
3623 struct imsm_disk
*disk
;
3625 /* tbl_mpb is more up to date, but copy
3626 * over cross generational status before
3629 disk
= __serial_to_disk(d
->serial
, mpb
, NULL
);
3630 if (disk
&& is_failed(disk
))
3631 d
->status
|= FAILED_DISK
;
3633 idisk
= disk_list_get(d
->serial
, *disk_list
);
3636 if (disk
&& is_configured(disk
))
3637 idisk
->disk
.status
|= CONFIGURED_DISK
;
3640 dprintf("%s: mpb from %d:%d prefer %d:%d\n",
3641 __func__
, super
->disks
->major
,
3642 super
->disks
->minor
,
3643 table
[i
]->disks
->major
,
3644 table
[i
]->disks
->minor
);
3652 table
[tbl_size
++] = super
;
3656 /* update/extend the merged list of imsm_disk records */
3657 for (j
= 0; j
< mpb
->num_disks
; j
++) {
3658 struct imsm_disk
*disk
= __get_imsm_disk(mpb
, j
);
3659 struct intel_disk
*idisk
;
3661 idisk
= disk_list_get(disk
->serial
, *disk_list
);
3663 idisk
->disk
.status
|= disk
->status
;
3664 if (is_configured(&idisk
->disk
) ||
3665 is_failed(&idisk
->disk
))
3666 idisk
->disk
.status
&= ~(SPARE_DISK
);
3668 idisk
= calloc(1, sizeof(*idisk
));
3671 idisk
->owner
= IMSM_UNKNOWN_OWNER
;
3672 idisk
->disk
= *disk
;
3673 idisk
->next
= *disk_list
;
3677 if (serialcmp(idisk
->disk
.serial
, d
->serial
) == 0)
3684 static struct intel_super
*
3685 validate_members(struct intel_super
*super
, struct intel_disk
*disk_list
,
3688 struct imsm_super
*mpb
= super
->anchor
;
3692 for (i
= 0; i
< mpb
->num_disks
; i
++) {
3693 struct imsm_disk
*disk
= __get_imsm_disk(mpb
, i
);
3694 struct intel_disk
*idisk
;
3696 idisk
= disk_list_get(disk
->serial
, disk_list
);
3698 if (idisk
->owner
== owner
||
3699 idisk
->owner
== IMSM_UNKNOWN_OWNER
)
3702 dprintf("%s: '%.16s' owner %d != %d\n",
3703 __func__
, disk
->serial
, idisk
->owner
,
3706 dprintf("%s: unknown disk %x [%d]: %.16s\n",
3707 __func__
, __le32_to_cpu(mpb
->family_num
), i
,
3713 if (ok_count
== mpb
->num_disks
)
3718 static void show_conflicts(__u32 family_num
, struct intel_super
*super_list
)
3720 struct intel_super
*s
;
3722 for (s
= super_list
; s
; s
= s
->next
) {
3723 if (family_num
!= s
->anchor
->family_num
)
3725 fprintf(stderr
, "Conflict, offlining family %#x on '%s'\n",
3726 __le32_to_cpu(family_num
), s
->disks
->devname
);
3730 static struct intel_super
*
3731 imsm_thunderdome(struct intel_super
**super_list
, int len
)
3733 struct intel_super
*super_table
[len
];
3734 struct intel_disk
*disk_list
= NULL
;
3735 struct intel_super
*champion
, *spare
;
3736 struct intel_super
*s
, **del
;
3741 memset(super_table
, 0, sizeof(super_table
));
3742 for (s
= *super_list
; s
; s
= s
->next
)
3743 tbl_size
= __prep_thunderdome(super_table
, tbl_size
, s
, &disk_list
);
3745 for (i
= 0; i
< tbl_size
; i
++) {
3746 struct imsm_disk
*d
;
3747 struct intel_disk
*idisk
;
3748 struct imsm_super
*mpb
= super_table
[i
]->anchor
;
3751 d
= &s
->disks
->disk
;
3753 /* 'd' must appear in merged disk list for its
3754 * configuration to be valid
3756 idisk
= disk_list_get(d
->serial
, disk_list
);
3757 if (idisk
&& idisk
->owner
== i
)
3758 s
= validate_members(s
, disk_list
, i
);
3763 dprintf("%s: marking family: %#x from %d:%d offline\n",
3764 __func__
, mpb
->family_num
,
3765 super_table
[i
]->disks
->major
,
3766 super_table
[i
]->disks
->minor
);
3770 /* This is where the mdadm implementation differs from the Windows
3771 * driver which has no strict concept of a container. We can only
3772 * assemble one family from a container, so when returning a prodigal
3773 * array member to this system the code will not be able to disambiguate
3774 * the container contents that should be assembled ("foreign" versus
3775 * "local"). It requires user intervention to set the orig_family_num
3776 * to a new value to establish a new container. The Windows driver in
3777 * this situation fixes up the volume name in place and manages the
3778 * foreign array as an independent entity.
3783 for (i
= 0; i
< tbl_size
; i
++) {
3784 struct intel_super
*tbl_ent
= super_table
[i
];
3790 if (tbl_ent
->anchor
->num_raid_devs
== 0) {
3795 if (s
&& !is_spare
) {
3796 show_conflicts(tbl_ent
->anchor
->family_num
, *super_list
);
3798 } else if (!s
&& !is_spare
)
3811 fprintf(stderr
, "Chose family %#x on '%s', "
3812 "assemble conflicts to new container with '--update=uuid'\n",
3813 __le32_to_cpu(s
->anchor
->family_num
), s
->disks
->devname
);
3815 /* collect all dl's onto 'champion', and update them to
3816 * champion's version of the status
3818 for (s
= *super_list
; s
; s
= s
->next
) {
3819 struct imsm_super
*mpb
= champion
->anchor
;
3820 struct dl
*dl
= s
->disks
;
3825 for (i
= 0; i
< mpb
->num_disks
; i
++) {
3826 struct imsm_disk
*disk
;
3828 disk
= __serial_to_disk(dl
->serial
, mpb
, &dl
->index
);
3831 /* only set index on disks that are a member of
3832 * a populated contianer, i.e. one with
3835 if (is_failed(&dl
->disk
))
3837 else if (is_spare(&dl
->disk
))
3843 if (i
>= mpb
->num_disks
) {
3844 struct intel_disk
*idisk
;
3846 idisk
= disk_list_get(dl
->serial
, disk_list
);
3847 if (idisk
&& is_spare(&idisk
->disk
) &&
3848 !is_failed(&idisk
->disk
) && !is_configured(&idisk
->disk
))
3856 dl
->next
= champion
->disks
;
3857 champion
->disks
= dl
;
3861 /* delete 'champion' from super_list */
3862 for (del
= super_list
; *del
; ) {
3863 if (*del
== champion
) {
3864 *del
= (*del
)->next
;
3867 del
= &(*del
)->next
;
3869 champion
->next
= NULL
;
3873 struct intel_disk
*idisk
= disk_list
;
3875 disk_list
= disk_list
->next
;
3882 static int load_super_imsm_all(struct supertype
*st
, int fd
, void **sbp
,
3886 struct intel_super
*super_list
= NULL
;
3887 struct intel_super
*super
= NULL
;
3888 int devnum
= fd2devnum(fd
);
3894 /* check if 'fd' an opened container */
3895 sra
= sysfs_read(fd
, 0, GET_LEVEL
|GET_VERSION
|GET_DEVS
|GET_STATE
);
3899 if (sra
->array
.major_version
!= -1 ||
3900 sra
->array
.minor_version
!= -2 ||
3901 strcmp(sra
->text_version
, "imsm") != 0) {
3906 for (sd
= sra
->devs
, i
= 0; sd
; sd
= sd
->next
, i
++) {
3907 struct intel_super
*s
= alloc_super();
3915 s
->next
= super_list
;
3919 sprintf(nm
, "%d:%d", sd
->disk
.major
, sd
->disk
.minor
);
3920 dfd
= dev_open(nm
, O_RDWR
);
3924 rv
= find_intel_hba_capability(dfd
, s
, devname
);
3925 /* no orom/efi or non-intel hba of the disk */
3929 err
= load_and_parse_mpb(dfd
, s
, NULL
, 1);
3931 /* retry the load if we might have raced against mdmon */
3932 if (err
== 3 && mdmon_running(devnum
))
3933 for (retry
= 0; retry
< 3; retry
++) {
3935 err
= load_and_parse_mpb(dfd
, s
, NULL
, 1);
3943 /* all mpbs enter, maybe one leaves */
3944 super
= imsm_thunderdome(&super_list
, i
);
3950 if (find_missing(super
) != 0) {
3956 /* load migration record */
3957 err
= load_imsm_migr_rec(super
, NULL
);
3963 /* Check migration compatibility */
3964 if (check_mpb_migr_compatibility(super
) != 0) {
3965 fprintf(stderr
, Name
": Unsupported migration detected");
3967 fprintf(stderr
, " on %s\n", devname
);
3969 fprintf(stderr
, " (IMSM).\n");
3978 while (super_list
) {
3979 struct intel_super
*s
= super_list
;
3981 super_list
= super_list
->next
;
3990 st
->container_dev
= devnum
;
3991 if (err
== 0 && st
->ss
== NULL
) {
3992 st
->ss
= &super_imsm
;
3993 st
->minor_version
= 0;
3994 st
->max_devs
= IMSM_MAX_DEVICES
;
3999 static int load_container_imsm(struct supertype
*st
, int fd
, char *devname
)
4001 return load_super_imsm_all(st
, fd
, &st
->sb
, devname
);
4005 static int load_super_imsm(struct supertype
*st
, int fd
, char *devname
)
4007 struct intel_super
*super
;
4010 if (test_partition(fd
))
4011 /* IMSM not allowed on partitions */
4014 free_super_imsm(st
);
4016 super
= alloc_super();
4019 Name
": malloc of %zu failed.\n",
4023 /* Load hba and capabilities if they exist.
4024 * But do not preclude loading metadata in case capabilities or hba are
4025 * non-compliant and ignore_hw_compat is set.
4027 rv
= find_intel_hba_capability(fd
, super
, devname
);
4028 /* no orom/efi or non-intel hba of the disk */
4029 if ((rv
!= 0) && (st
->ignore_hw_compat
== 0)) {
4032 Name
": No OROM/EFI properties for %s\n", devname
);
4036 rv
= load_and_parse_mpb(fd
, super
, devname
, 0);
4041 Name
": Failed to load all information "
4042 "sections on %s\n", devname
);
4048 if (st
->ss
== NULL
) {
4049 st
->ss
= &super_imsm
;
4050 st
->minor_version
= 0;
4051 st
->max_devs
= IMSM_MAX_DEVICES
;
4054 /* load migration record */
4055 if (load_imsm_migr_rec(super
, NULL
) == 0) {
4056 /* Check for unsupported migration features */
4057 if (check_mpb_migr_compatibility(super
) != 0) {
4059 Name
": Unsupported migration detected");
4061 fprintf(stderr
, " on %s\n", devname
);
4063 fprintf(stderr
, " (IMSM).\n");
4071 static __u16
info_to_blocks_per_strip(mdu_array_info_t
*info
)
4073 if (info
->level
== 1)
4075 return info
->chunk_size
>> 9;
4078 static __u32
info_to_num_data_stripes(mdu_array_info_t
*info
, int num_domains
)
4082 num_stripes
= (info
->size
* 2) / info_to_blocks_per_strip(info
);
4083 num_stripes
/= num_domains
;
4088 static __u32
info_to_blocks_per_member(mdu_array_info_t
*info
)
4090 if (info
->level
== 1)
4091 return info
->size
* 2;
4093 return (info
->size
* 2) & ~(info_to_blocks_per_strip(info
) - 1);
4096 static void imsm_update_version_info(struct intel_super
*super
)
4098 /* update the version and attributes */
4099 struct imsm_super
*mpb
= super
->anchor
;
4101 struct imsm_dev
*dev
;
4102 struct imsm_map
*map
;
4105 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
4106 dev
= get_imsm_dev(super
, i
);
4107 map
= get_imsm_map(dev
, 0);
4108 if (__le32_to_cpu(dev
->size_high
) > 0)
4109 mpb
->attributes
|= MPB_ATTRIB_2TB
;
4111 /* FIXME detect when an array spans a port multiplier */
4113 mpb
->attributes
|= MPB_ATTRIB_PM
;
4116 if (mpb
->num_raid_devs
> 1 ||
4117 mpb
->attributes
!= MPB_ATTRIB_CHECKSUM_VERIFY
) {
4118 version
= MPB_VERSION_ATTRIBS
;
4119 switch (get_imsm_raid_level(map
)) {
4120 case 0: mpb
->attributes
|= MPB_ATTRIB_RAID0
; break;
4121 case 1: mpb
->attributes
|= MPB_ATTRIB_RAID1
; break;
4122 case 10: mpb
->attributes
|= MPB_ATTRIB_RAID10
; break;
4123 case 5: mpb
->attributes
|= MPB_ATTRIB_RAID5
; break;
4126 if (map
->num_members
>= 5)
4127 version
= MPB_VERSION_5OR6_DISK_ARRAY
;
4128 else if (dev
->status
== DEV_CLONE_N_GO
)
4129 version
= MPB_VERSION_CNG
;
4130 else if (get_imsm_raid_level(map
) == 5)
4131 version
= MPB_VERSION_RAID5
;
4132 else if (map
->num_members
>= 3)
4133 version
= MPB_VERSION_3OR4_DISK_ARRAY
;
4134 else if (get_imsm_raid_level(map
) == 1)
4135 version
= MPB_VERSION_RAID1
;
4137 version
= MPB_VERSION_RAID0
;
4139 strcpy(((char *) mpb
->sig
) + strlen(MPB_SIGNATURE
), version
);
4143 static int check_name(struct intel_super
*super
, char *name
, int quiet
)
4145 struct imsm_super
*mpb
= super
->anchor
;
4146 char *reason
= NULL
;
4149 if (strlen(name
) > MAX_RAID_SERIAL_LEN
)
4150 reason
= "must be 16 characters or less";
4152 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
4153 struct imsm_dev
*dev
= get_imsm_dev(super
, i
);
4155 if (strncmp((char *) dev
->volume
, name
, MAX_RAID_SERIAL_LEN
) == 0) {
4156 reason
= "already exists";
4161 if (reason
&& !quiet
)
4162 fprintf(stderr
, Name
": imsm volume name %s\n", reason
);
4167 static int init_super_imsm_volume(struct supertype
*st
, mdu_array_info_t
*info
,
4168 unsigned long long size
, char *name
,
4169 char *homehost
, int *uuid
)
4171 /* We are creating a volume inside a pre-existing container.
4172 * so st->sb is already set.
4174 struct intel_super
*super
= st
->sb
;
4175 struct imsm_super
*mpb
= super
->anchor
;
4176 struct intel_dev
*dv
;
4177 struct imsm_dev
*dev
;
4178 struct imsm_vol
*vol
;
4179 struct imsm_map
*map
;
4180 int idx
= mpb
->num_raid_devs
;
4182 unsigned long long array_blocks
;
4183 size_t size_old
, size_new
;
4184 __u32 num_data_stripes
;
4186 if (super
->orom
&& mpb
->num_raid_devs
>= super
->orom
->vpa
) {
4187 fprintf(stderr
, Name
": This imsm-container already has the "
4188 "maximum of %d volumes\n", super
->orom
->vpa
);
4192 /* ensure the mpb is large enough for the new data */
4193 size_old
= __le32_to_cpu(mpb
->mpb_size
);
4194 size_new
= disks_to_mpb_size(info
->nr_disks
);
4195 if (size_new
> size_old
) {
4197 size_t size_round
= ROUND_UP(size_new
, 512);
4199 if (posix_memalign(&mpb_new
, 512, size_round
) != 0) {
4200 fprintf(stderr
, Name
": could not allocate new mpb\n");
4203 if (posix_memalign(&super
->migr_rec_buf
, 512, 512) != 0) {
4204 fprintf(stderr
, Name
4205 ": %s could not allocate migr_rec buffer\n",
4212 memcpy(mpb_new
, mpb
, size_old
);
4215 super
->anchor
= mpb_new
;
4216 mpb
->mpb_size
= __cpu_to_le32(size_new
);
4217 memset(mpb_new
+ size_old
, 0, size_round
- size_old
);
4219 super
->current_vol
= idx
;
4221 /* handle 'failed_disks' by either:
4222 * a) create dummy disk entries in the table if this the first
4223 * volume in the array. We add them here as this is the only
4224 * opportunity to add them. add_to_super_imsm_volume()
4225 * handles the non-failed disks and continues incrementing
4227 * b) validate that 'failed_disks' matches the current number
4228 * of missing disks if the container is populated
4230 if (super
->current_vol
== 0) {
4232 for (i
= 0; i
< info
->failed_disks
; i
++) {
4233 struct imsm_disk
*disk
;
4236 disk
= __get_imsm_disk(mpb
, i
);
4237 disk
->status
= CONFIGURED_DISK
| FAILED_DISK
;
4238 disk
->scsi_id
= __cpu_to_le32(~(__u32
)0);
4239 snprintf((char *) disk
->serial
, MAX_RAID_SERIAL_LEN
,
4242 find_missing(super
);
4247 for (d
= super
->missing
; d
; d
= d
->next
)
4249 if (info
->failed_disks
> missing
) {
4250 fprintf(stderr
, Name
": unable to add 'missing' disk to container\n");
4255 if (!check_name(super
, name
, 0))
4257 dv
= malloc(sizeof(*dv
));
4259 fprintf(stderr
, Name
": failed to allocate device list entry\n");
4262 dev
= calloc(1, sizeof(*dev
) + sizeof(__u32
) * (info
->raid_disks
- 1));
4265 fprintf(stderr
, Name
": could not allocate raid device\n");
4269 strncpy((char *) dev
->volume
, name
, MAX_RAID_SERIAL_LEN
);
4270 if (info
->level
== 1)
4271 array_blocks
= info_to_blocks_per_member(info
);
4273 array_blocks
= calc_array_size(info
->level
, info
->raid_disks
,
4274 info
->layout
, info
->chunk_size
,
4276 /* round array size down to closest MB */
4277 array_blocks
= (array_blocks
>> SECT_PER_MB_SHIFT
) << SECT_PER_MB_SHIFT
;
4279 dev
->size_low
= __cpu_to_le32((__u32
) array_blocks
);
4280 dev
->size_high
= __cpu_to_le32((__u32
) (array_blocks
>> 32));
4281 dev
->status
= (DEV_READ_COALESCING
| DEV_WRITE_COALESCING
);
4283 vol
->migr_state
= 0;
4284 set_migr_type(dev
, MIGR_INIT
);
4285 vol
->dirty
= !info
->state
;
4286 vol
->curr_migr_unit
= 0;
4287 map
= get_imsm_map(dev
, 0);
4288 map
->pba_of_lba0
= __cpu_to_le32(super
->create_offset
);
4289 map
->blocks_per_member
= __cpu_to_le32(info_to_blocks_per_member(info
));
4290 map
->blocks_per_strip
= __cpu_to_le16(info_to_blocks_per_strip(info
));
4291 map
->failed_disk_num
= ~0;
4292 map
->map_state
= info
->failed_disks
? IMSM_T_STATE_DEGRADED
: IMSM_T_STATE_NORMAL
;
4295 if (info
->level
== 1 && info
->raid_disks
> 2) {
4298 fprintf(stderr
, Name
": imsm does not support more than 2 disks"
4299 "in a raid1 volume\n");
4303 map
->raid_level
= info
->level
;
4304 if (info
->level
== 10) {
4305 map
->raid_level
= 1;
4306 map
->num_domains
= info
->raid_disks
/ 2;
4307 } else if (info
->level
== 1)
4308 map
->num_domains
= info
->raid_disks
;
4310 map
->num_domains
= 1;
4312 num_data_stripes
= info_to_num_data_stripes(info
, map
->num_domains
);
4313 map
->num_data_stripes
= __cpu_to_le32(num_data_stripes
);
4315 map
->num_members
= info
->raid_disks
;
4316 for (i
= 0; i
< map
->num_members
; i
++) {
4317 /* initialized in add_to_super */
4318 set_imsm_ord_tbl_ent(map
, i
, IMSM_ORD_REBUILD
);
4320 mpb
->num_raid_devs
++;
4323 dv
->index
= super
->current_vol
;
4324 dv
->next
= super
->devlist
;
4325 super
->devlist
= dv
;
4327 imsm_update_version_info(super
);
4332 static int init_super_imsm(struct supertype
*st
, mdu_array_info_t
*info
,
4333 unsigned long long size
, char *name
,
4334 char *homehost
, int *uuid
)
4336 /* This is primarily called by Create when creating a new array.
4337 * We will then get add_to_super called for each component, and then
4338 * write_init_super called to write it out to each device.
4339 * For IMSM, Create can create on fresh devices or on a pre-existing
4341 * To create on a pre-existing array a different method will be called.
4342 * This one is just for fresh drives.
4344 struct intel_super
*super
;
4345 struct imsm_super
*mpb
;
4350 return init_super_imsm_volume(st
, info
, size
, name
, homehost
, uuid
);
4353 mpb_size
= disks_to_mpb_size(info
->nr_disks
);
4357 super
= alloc_super();
4358 if (super
&& posix_memalign(&super
->buf
, 512, mpb_size
) != 0) {
4363 fprintf(stderr
, Name
4364 ": %s could not allocate superblock\n", __func__
);
4367 if (posix_memalign(&super
->migr_rec_buf
, 512, 512) != 0) {
4368 fprintf(stderr
, Name
4369 ": %s could not allocate migr_rec buffer\n", __func__
);
4374 memset(super
->buf
, 0, mpb_size
);
4376 mpb
->mpb_size
= __cpu_to_le32(mpb_size
);
4380 /* zeroing superblock */
4384 mpb
->attributes
= MPB_ATTRIB_CHECKSUM_VERIFY
;
4386 version
= (char *) mpb
->sig
;
4387 strcpy(version
, MPB_SIGNATURE
);
4388 version
+= strlen(MPB_SIGNATURE
);
4389 strcpy(version
, MPB_VERSION_RAID0
);
4395 static int add_to_super_imsm_volume(struct supertype
*st
, mdu_disk_info_t
*dk
,
4396 int fd
, char *devname
)
4398 struct intel_super
*super
= st
->sb
;
4399 struct imsm_super
*mpb
= super
->anchor
;
4400 struct imsm_disk
*_disk
;
4401 struct imsm_dev
*dev
;
4402 struct imsm_map
*map
;
4406 dev
= get_imsm_dev(super
, super
->current_vol
);
4407 map
= get_imsm_map(dev
, 0);
4409 if (! (dk
->state
& (1<<MD_DISK_SYNC
))) {
4410 fprintf(stderr
, Name
": %s: Cannot add spare devices to IMSM volume\n",
4416 /* we're doing autolayout so grab the pre-marked (in
4417 * validate_geometry) raid_disk
4419 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
4420 if (dl
->raiddisk
== dk
->raid_disk
)
4423 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
4424 if (dl
->major
== dk
->major
&&
4425 dl
->minor
== dk
->minor
)
4430 fprintf(stderr
, Name
": %s is not a member of the same container\n", devname
);
4434 /* add a pristine spare to the metadata */
4435 if (dl
->index
< 0) {
4436 dl
->index
= super
->anchor
->num_disks
;
4437 super
->anchor
->num_disks
++;
4439 /* Check the device has not already been added */
4440 slot
= get_imsm_disk_slot(map
, dl
->index
);
4442 (get_imsm_ord_tbl_ent(dev
, slot
, -1) & IMSM_ORD_REBUILD
) == 0) {
4443 fprintf(stderr
, Name
": %s has been included in this array twice\n",
4447 set_imsm_ord_tbl_ent(map
, dk
->raid_disk
, dl
->index
);
4448 dl
->disk
.status
= CONFIGURED_DISK
;
4450 /* update size of 'missing' disks to be at least as large as the
4451 * largest acitve member (we only have dummy missing disks when
4452 * creating the first volume)
4454 if (super
->current_vol
== 0) {
4455 for (df
= super
->missing
; df
; df
= df
->next
) {
4456 if (dl
->disk
.total_blocks
> df
->disk
.total_blocks
)
4457 df
->disk
.total_blocks
= dl
->disk
.total_blocks
;
4458 _disk
= __get_imsm_disk(mpb
, df
->index
);
4463 /* refresh unset/failed slots to point to valid 'missing' entries */
4464 for (df
= super
->missing
; df
; df
= df
->next
)
4465 for (slot
= 0; slot
< mpb
->num_disks
; slot
++) {
4466 __u32 ord
= get_imsm_ord_tbl_ent(dev
, slot
, -1);
4468 if ((ord
& IMSM_ORD_REBUILD
) == 0)
4470 set_imsm_ord_tbl_ent(map
, slot
, df
->index
| IMSM_ORD_REBUILD
);
4471 dprintf("set slot:%d to missing disk:%d\n", slot
, df
->index
);
4475 /* if we are creating the first raid device update the family number */
4476 if (super
->current_vol
== 0) {
4478 struct imsm_dev
*_dev
= __get_imsm_dev(mpb
, 0);
4480 _disk
= __get_imsm_disk(mpb
, dl
->index
);
4481 if (!_dev
|| !_disk
) {
4482 fprintf(stderr
, Name
": BUG mpb setup error\n");
4488 sum
+= __gen_imsm_checksum(mpb
);
4489 mpb
->family_num
= __cpu_to_le32(sum
);
4490 mpb
->orig_family_num
= mpb
->family_num
;
4492 super
->current_disk
= dl
;
4497 * Function marks disk as spare and restores disk serial
4498 * in case it was previously marked as failed by takeover operation
4500 * -1 : critical error
4501 * 0 : disk is marked as spare but serial is not set
4504 int mark_spare(struct dl
*disk
)
4506 __u8 serial
[MAX_RAID_SERIAL_LEN
];
4513 if (!imsm_read_serial(disk
->fd
, NULL
, serial
)) {
4514 /* Restore disk serial number, because takeover marks disk
4515 * as failed and adds to serial ':0' before it becomes
4518 serialcpy(disk
->serial
, serial
);
4519 serialcpy(disk
->disk
.serial
, serial
);
4522 disk
->disk
.status
= SPARE_DISK
;
4528 static int add_to_super_imsm(struct supertype
*st
, mdu_disk_info_t
*dk
,
4529 int fd
, char *devname
)
4531 struct intel_super
*super
= st
->sb
;
4533 unsigned long long size
;
4538 /* If we are on an RAID enabled platform check that the disk is
4539 * attached to the raid controller.
4540 * We do not need to test disks attachment for container based additions,
4541 * they shall be already tested when container was created/assembled.
4543 rv
= find_intel_hba_capability(fd
, super
, devname
);
4544 /* no orom/efi or non-intel hba of the disk */
4546 dprintf("capability: %p fd: %d ret: %d\n",
4547 super
->orom
, fd
, rv
);
4551 if (super
->current_vol
>= 0)
4552 return add_to_super_imsm_volume(st
, dk
, fd
, devname
);
4555 dd
= malloc(sizeof(*dd
));
4558 Name
": malloc failed %s:%d.\n", __func__
, __LINE__
);
4561 memset(dd
, 0, sizeof(*dd
));
4562 dd
->major
= major(stb
.st_rdev
);
4563 dd
->minor
= minor(stb
.st_rdev
);
4564 dd
->devname
= devname
? strdup(devname
) : NULL
;
4567 dd
->action
= DISK_ADD
;
4568 rv
= imsm_read_serial(fd
, devname
, dd
->serial
);
4571 Name
": failed to retrieve scsi serial, aborting\n");
4576 get_dev_size(fd
, NULL
, &size
);
4578 serialcpy(dd
->disk
.serial
, dd
->serial
);
4579 dd
->disk
.total_blocks
= __cpu_to_le32(size
);
4581 if (sysfs_disk_to_scsi_id(fd
, &id
) == 0)
4582 dd
->disk
.scsi_id
= __cpu_to_le32(id
);
4584 dd
->disk
.scsi_id
= __cpu_to_le32(0);
4586 if (st
->update_tail
) {
4587 dd
->next
= super
->disk_mgmt_list
;
4588 super
->disk_mgmt_list
= dd
;
4590 dd
->next
= super
->disks
;
4592 super
->updates_pending
++;
4599 static int remove_from_super_imsm(struct supertype
*st
, mdu_disk_info_t
*dk
)
4601 struct intel_super
*super
= st
->sb
;
4604 /* remove from super works only in mdmon - for communication
4605 * manager - monitor. Check if communication memory buffer
4608 if (!st
->update_tail
) {
4610 Name
": %s shall be used in mdmon context only"
4611 "(line %d).\n", __func__
, __LINE__
);
4614 dd
= malloc(sizeof(*dd
));
4617 Name
": malloc failed %s:%d.\n", __func__
, __LINE__
);
4620 memset(dd
, 0, sizeof(*dd
));
4621 dd
->major
= dk
->major
;
4622 dd
->minor
= dk
->minor
;
4625 dd
->action
= DISK_REMOVE
;
4627 dd
->next
= super
->disk_mgmt_list
;
4628 super
->disk_mgmt_list
= dd
;
4634 static int store_imsm_mpb(int fd
, struct imsm_super
*mpb
);
4638 struct imsm_super anchor
;
4639 } spare_record
__attribute__ ((aligned(512)));
4641 /* spare records have their own family number and do not have any defined raid
4644 static int write_super_imsm_spares(struct intel_super
*super
, int doclose
)
4646 struct imsm_super
*mpb
= super
->anchor
;
4647 struct imsm_super
*spare
= &spare_record
.anchor
;
4651 spare
->mpb_size
= __cpu_to_le32(sizeof(struct imsm_super
)),
4652 spare
->generation_num
= __cpu_to_le32(1UL),
4653 spare
->attributes
= MPB_ATTRIB_CHECKSUM_VERIFY
;
4654 spare
->num_disks
= 1,
4655 spare
->num_raid_devs
= 0,
4656 spare
->cache_size
= mpb
->cache_size
,
4657 spare
->pwr_cycle_count
= __cpu_to_le32(1),
4659 snprintf((char *) spare
->sig
, MAX_SIGNATURE_LENGTH
,
4660 MPB_SIGNATURE MPB_VERSION_RAID0
);
4662 for (d
= super
->disks
; d
; d
= d
->next
) {
4666 spare
->disk
[0] = d
->disk
;
4667 sum
= __gen_imsm_checksum(spare
);
4668 spare
->family_num
= __cpu_to_le32(sum
);
4669 spare
->orig_family_num
= 0;
4670 sum
= __gen_imsm_checksum(spare
);
4671 spare
->check_sum
= __cpu_to_le32(sum
);
4673 if (store_imsm_mpb(d
->fd
, spare
)) {
4674 fprintf(stderr
, "%s: failed for device %d:%d %s\n",
4675 __func__
, d
->major
, d
->minor
, strerror(errno
));
4687 static int write_super_imsm(struct supertype
*st
, int doclose
)
4689 struct intel_super
*super
= st
->sb
;
4690 struct imsm_super
*mpb
= super
->anchor
;
4696 __u32 mpb_size
= sizeof(struct imsm_super
) - sizeof(struct imsm_disk
);
4698 int clear_migration_record
= 1;
4700 /* 'generation' is incremented everytime the metadata is written */
4701 generation
= __le32_to_cpu(mpb
->generation_num
);
4703 mpb
->generation_num
= __cpu_to_le32(generation
);
4705 /* fix up cases where previous mdadm releases failed to set
4708 if (mpb
->orig_family_num
== 0)
4709 mpb
->orig_family_num
= mpb
->family_num
;
4711 for (d
= super
->disks
; d
; d
= d
->next
) {
4715 mpb
->disk
[d
->index
] = d
->disk
;
4719 for (d
= super
->missing
; d
; d
= d
->next
) {
4720 mpb
->disk
[d
->index
] = d
->disk
;
4723 mpb
->num_disks
= num_disks
;
4724 mpb_size
+= sizeof(struct imsm_disk
) * mpb
->num_disks
;
4726 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
4727 struct imsm_dev
*dev
= __get_imsm_dev(mpb
, i
);
4728 struct imsm_dev
*dev2
= get_imsm_dev(super
, i
);
4730 imsm_copy_dev(dev
, dev2
);
4731 mpb_size
+= sizeof_imsm_dev(dev
, 0);
4733 if (is_gen_migration(dev2
))
4734 clear_migration_record
= 0;
4736 mpb_size
+= __le32_to_cpu(mpb
->bbm_log_size
);
4737 mpb
->mpb_size
= __cpu_to_le32(mpb_size
);
4739 /* recalculate checksum */
4740 sum
= __gen_imsm_checksum(mpb
);
4741 mpb
->check_sum
= __cpu_to_le32(sum
);
4743 if (clear_migration_record
)
4744 memset(super
->migr_rec_buf
, 0, 512);
4746 /* write the mpb for disks that compose raid devices */
4747 for (d
= super
->disks
; d
; d
= d
->next
) {
4748 if (d
->index
< 0 || is_failed(&d
->disk
))
4750 if (store_imsm_mpb(d
->fd
, mpb
))
4751 fprintf(stderr
, "%s: failed for device %d:%d %s\n",
4752 __func__
, d
->major
, d
->minor
, strerror(errno
));
4753 if (clear_migration_record
) {
4754 unsigned long long dsize
;
4756 get_dev_size(d
->fd
, NULL
, &dsize
);
4757 if (lseek64(d
->fd
, dsize
- 512, SEEK_SET
) >= 0) {
4758 if (write(d
->fd
, super
->migr_rec_buf
, 512) != 512)
4759 perror("Write migr_rec failed");
4769 return write_super_imsm_spares(super
, doclose
);
4775 static int create_array(struct supertype
*st
, int dev_idx
)
4778 struct imsm_update_create_array
*u
;
4779 struct intel_super
*super
= st
->sb
;
4780 struct imsm_dev
*dev
= get_imsm_dev(super
, dev_idx
);
4781 struct imsm_map
*map
= get_imsm_map(dev
, 0);
4782 struct disk_info
*inf
;
4783 struct imsm_disk
*disk
;
4786 len
= sizeof(*u
) - sizeof(*dev
) + sizeof_imsm_dev(dev
, 0) +
4787 sizeof(*inf
) * map
->num_members
;
4790 fprintf(stderr
, "%s: failed to allocate update buffer\n",
4795 u
->type
= update_create_array
;
4796 u
->dev_idx
= dev_idx
;
4797 imsm_copy_dev(&u
->dev
, dev
);
4798 inf
= get_disk_info(u
);
4799 for (i
= 0; i
< map
->num_members
; i
++) {
4800 int idx
= get_imsm_disk_idx(dev
, i
, -1);
4802 disk
= get_imsm_disk(super
, idx
);
4803 serialcpy(inf
[i
].serial
, disk
->serial
);
4805 append_metadata_update(st
, u
, len
);
4810 static int mgmt_disk(struct supertype
*st
)
4812 struct intel_super
*super
= st
->sb
;
4814 struct imsm_update_add_remove_disk
*u
;
4816 if (!super
->disk_mgmt_list
)
4822 fprintf(stderr
, "%s: failed to allocate update buffer\n",
4827 u
->type
= update_add_remove_disk
;
4828 append_metadata_update(st
, u
, len
);
4833 static int write_init_super_imsm(struct supertype
*st
)
4835 struct intel_super
*super
= st
->sb
;
4836 int current_vol
= super
->current_vol
;
4838 /* we are done with current_vol reset it to point st at the container */
4839 super
->current_vol
= -1;
4841 if (st
->update_tail
) {
4842 /* queue the recently created array / added disk
4843 * as a metadata update */
4846 /* determine if we are creating a volume or adding a disk */
4847 if (current_vol
< 0) {
4848 /* in the mgmt (add/remove) disk case we are running
4849 * in mdmon context, so don't close fd's
4851 return mgmt_disk(st
);
4853 rv
= create_array(st
, current_vol
);
4858 for (d
= super
->disks
; d
; d
= d
->next
)
4859 Kill(d
->devname
, NULL
, 0, 1, 1);
4860 return write_super_imsm(st
, 1);
4865 static int store_super_imsm(struct supertype
*st
, int fd
)
4867 struct intel_super
*super
= st
->sb
;
4868 struct imsm_super
*mpb
= super
? super
->anchor
: NULL
;
4874 return store_imsm_mpb(fd
, mpb
);
4880 static int imsm_bbm_log_size(struct imsm_super
*mpb
)
4882 return __le32_to_cpu(mpb
->bbm_log_size
);
4886 static int validate_geometry_imsm_container(struct supertype
*st
, int level
,
4887 int layout
, int raiddisks
, int chunk
,
4888 unsigned long long size
, char *dev
,
4889 unsigned long long *freesize
,
4893 unsigned long long ldsize
;
4894 struct intel_super
*super
=NULL
;
4897 if (level
!= LEVEL_CONTAINER
)
4902 fd
= open(dev
, O_RDONLY
|O_EXCL
, 0);
4905 fprintf(stderr
, Name
": imsm: Cannot open %s: %s\n",
4906 dev
, strerror(errno
));
4909 if (!get_dev_size(fd
, dev
, &ldsize
)) {
4914 /* capabilities retrieve could be possible
4915 * note that there is no fd for the disks in array.
4917 super
= alloc_super();
4920 Name
": malloc of %zu failed.\n",
4926 rv
= find_intel_hba_capability(fd
, super
, verbose
? dev
: NULL
);
4930 fd2devname(fd
, str
);
4931 dprintf("validate_geometry_imsm_container: fd: %d %s orom: %p rv: %d raiddisk: %d\n",
4932 fd
, str
, super
->orom
, rv
, raiddisks
);
4934 /* no orom/efi or non-intel hba of the disk */
4940 if (super
->orom
&& raiddisks
> super
->orom
->tds
) {
4942 fprintf(stderr
, Name
": %d exceeds maximum number of"
4943 " platform supported disks: %d\n",
4944 raiddisks
, super
->orom
->tds
);
4950 *freesize
= avail_size_imsm(st
, ldsize
>> 9);
4956 static unsigned long long find_size(struct extent
*e
, int *idx
, int num_extents
)
4958 const unsigned long long base_start
= e
[*idx
].start
;
4959 unsigned long long end
= base_start
+ e
[*idx
].size
;
4962 if (base_start
== end
)
4966 for (i
= *idx
; i
< num_extents
; i
++) {
4967 /* extend overlapping extents */
4968 if (e
[i
].start
>= base_start
&&
4969 e
[i
].start
<= end
) {
4972 if (e
[i
].start
+ e
[i
].size
> end
)
4973 end
= e
[i
].start
+ e
[i
].size
;
4974 } else if (e
[i
].start
> end
) {
4980 return end
- base_start
;
4983 static unsigned long long merge_extents(struct intel_super
*super
, int sum_extents
)
4985 /* build a composite disk with all known extents and generate a new
4986 * 'maxsize' given the "all disks in an array must share a common start
4987 * offset" constraint
4989 struct extent
*e
= calloc(sum_extents
, sizeof(*e
));
4993 unsigned long long pos
;
4994 unsigned long long start
= 0;
4995 unsigned long long maxsize
;
4996 unsigned long reserve
;
5001 /* coalesce and sort all extents. also, check to see if we need to
5002 * reserve space between member arrays
5005 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
5008 for (i
= 0; i
< dl
->extent_cnt
; i
++)
5011 qsort(e
, sum_extents
, sizeof(*e
), cmp_extent
);
5016 while (i
< sum_extents
) {
5017 e
[j
].start
= e
[i
].start
;
5018 e
[j
].size
= find_size(e
, &i
, sum_extents
);
5020 if (e
[j
-1].size
== 0)
5029 unsigned long long esize
;
5031 esize
= e
[i
].start
- pos
;
5032 if (esize
>= maxsize
) {
5037 pos
= e
[i
].start
+ e
[i
].size
;
5039 } while (e
[i
-1].size
);
5045 /* FIXME assumes volume at offset 0 is the first volume in a
5048 if (start_extent
> 0)
5049 reserve
= IMSM_RESERVED_SECTORS
; /* gap between raid regions */
5053 if (maxsize
< reserve
)
5056 super
->create_offset
= ~((__u32
) 0);
5057 if (start
+ reserve
> super
->create_offset
)
5058 return 0; /* start overflows create_offset */
5059 super
->create_offset
= start
+ reserve
;
5061 return maxsize
- reserve
;
5064 static int is_raid_level_supported(const struct imsm_orom
*orom
, int level
, int raiddisks
)
5066 if (level
< 0 || level
== 6 || level
== 4)
5069 /* if we have an orom prevent invalid raid levels */
5072 case 0: return imsm_orom_has_raid0(orom
);
5075 return imsm_orom_has_raid1e(orom
);
5076 return imsm_orom_has_raid1(orom
) && raiddisks
== 2;
5077 case 10: return imsm_orom_has_raid10(orom
) && raiddisks
== 4;
5078 case 5: return imsm_orom_has_raid5(orom
) && raiddisks
> 2;
5081 return 1; /* not on an Intel RAID platform so anything goes */
5086 static int imsm_default_chunk(const struct imsm_orom
*orom
)
5088 /* up to 512 if the plaform supports it, otherwise the platform max.
5089 * 128 if no platform detected
5091 int fs
= max(7, orom
? fls(orom
->sss
) : 0);
5093 return min(512, (1 << fs
));
5096 #define pr_vrb(fmt, arg...) (void) (verbose && fprintf(stderr, Name fmt, ##arg))
5098 validate_geometry_imsm_orom(struct intel_super
*super
, int level
, int layout
,
5099 int raiddisks
, int *chunk
, int verbose
)
5101 /* check/set platform and metadata limits/defaults */
5102 if (super
->orom
&& raiddisks
> super
->orom
->dpa
) {
5103 pr_vrb(": platform supports a maximum of %d disks per array\n",
5108 /* capabilities of OROM tested - copied from validate_geometry_imsm_volume */
5109 if (!is_raid_level_supported(super
->orom
, level
, raiddisks
)) {
5110 pr_vrb(": platform does not support raid%d with %d disk%s\n",
5111 level
, raiddisks
, raiddisks
> 1 ? "s" : "");
5115 if (chunk
&& (*chunk
== 0 || *chunk
== UnSet
))
5116 *chunk
= imsm_default_chunk(super
->orom
);
5118 if (super
->orom
&& chunk
&& !imsm_orom_has_chunk(super
->orom
, *chunk
)) {
5119 pr_vrb(": platform does not support a chunk size of: "
5124 if (layout
!= imsm_level_to_layout(level
)) {
5126 pr_vrb(": imsm raid 5 only supports the left-asymmetric layout\n");
5127 else if (level
== 10)
5128 pr_vrb(": imsm raid 10 only supports the n2 layout\n");
5130 pr_vrb(": imsm unknown layout %#x for this raid level %d\n",
5137 /* validate_geometry_imsm_volume - lifted from validate_geometry_ddf_bvd
5138 * FIX ME add ahci details
5140 static int validate_geometry_imsm_volume(struct supertype
*st
, int level
,
5141 int layout
, int raiddisks
, int *chunk
,
5142 unsigned long long size
, char *dev
,
5143 unsigned long long *freesize
,
5147 struct intel_super
*super
= st
->sb
;
5148 struct imsm_super
*mpb
;
5150 unsigned long long pos
= 0;
5151 unsigned long long maxsize
;
5155 /* We must have the container info already read in. */
5159 mpb
= super
->anchor
;
5161 if (mpb
->num_raid_devs
> 0 && mpb
->num_disks
!= raiddisks
) {
5162 fprintf(stderr
, Name
": the option-rom requires all "
5163 "member disks to be a member of all volumes.\n");
5167 if (!validate_geometry_imsm_orom(super
, level
, layout
, raiddisks
, chunk
, verbose
)) {
5168 fprintf(stderr
, Name
": RAID gemetry validation failed. "
5169 "Cannot proceed with the action(s).\n");
5173 /* General test: make sure there is space for
5174 * 'raiddisks' device extents of size 'size' at a given
5177 unsigned long long minsize
= size
;
5178 unsigned long long start_offset
= MaxSector
;
5181 minsize
= MPB_SECTOR_CNT
+ IMSM_RESERVED_SECTORS
;
5182 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
5187 e
= get_extents(super
, dl
);
5190 unsigned long long esize
;
5191 esize
= e
[i
].start
- pos
;
5192 if (esize
>= minsize
)
5194 if (found
&& start_offset
== MaxSector
) {
5197 } else if (found
&& pos
!= start_offset
) {
5201 pos
= e
[i
].start
+ e
[i
].size
;
5203 } while (e
[i
-1].size
);
5208 if (dcnt
< raiddisks
) {
5210 fprintf(stderr
, Name
": imsm: Not enough "
5211 "devices with space for this array "
5219 /* This device must be a member of the set */
5220 if (stat(dev
, &stb
) < 0)
5222 if ((S_IFMT
& stb
.st_mode
) != S_IFBLK
)
5224 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
5225 if (dl
->major
== (int)major(stb
.st_rdev
) &&
5226 dl
->minor
== (int)minor(stb
.st_rdev
))
5231 fprintf(stderr
, Name
": %s is not in the "
5232 "same imsm set\n", dev
);
5234 } else if (super
->orom
&& dl
->index
< 0 && mpb
->num_raid_devs
) {
5235 /* If a volume is present then the current creation attempt
5236 * cannot incorporate new spares because the orom may not
5237 * understand this configuration (all member disks must be
5238 * members of each array in the container).
5240 fprintf(stderr
, Name
": %s is a spare and a volume"
5241 " is already defined for this container\n", dev
);
5242 fprintf(stderr
, Name
": The option-rom requires all member"
5243 " disks to be a member of all volumes\n");
5247 /* retrieve the largest free space block */
5248 e
= get_extents(super
, dl
);
5253 unsigned long long esize
;
5255 esize
= e
[i
].start
- pos
;
5256 if (esize
>= maxsize
)
5258 pos
= e
[i
].start
+ e
[i
].size
;
5260 } while (e
[i
-1].size
);
5265 fprintf(stderr
, Name
": unable to determine free space for: %s\n",
5269 if (maxsize
< size
) {
5271 fprintf(stderr
, Name
": %s not enough space (%llu < %llu)\n",
5272 dev
, maxsize
, size
);
5276 /* count total number of extents for merge */
5278 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
5280 i
+= dl
->extent_cnt
;
5282 maxsize
= merge_extents(super
, i
);
5284 if (!check_env("IMSM_NO_PLATFORM") &&
5285 mpb
->num_raid_devs
> 0 && size
&& size
!= maxsize
) {
5286 fprintf(stderr
, Name
": attempting to create a second "
5287 "volume with size less then remaining space. "
5292 if (maxsize
< size
|| maxsize
== 0) {
5294 fprintf(stderr
, Name
": not enough space after merge (%llu < %llu)\n",
5299 *freesize
= maxsize
;
5304 static int reserve_space(struct supertype
*st
, int raiddisks
,
5305 unsigned long long size
, int chunk
,
5306 unsigned long long *freesize
)
5308 struct intel_super
*super
= st
->sb
;
5309 struct imsm_super
*mpb
= super
->anchor
;
5314 unsigned long long maxsize
;
5315 unsigned long long minsize
;
5319 /* find the largest common start free region of the possible disks */
5323 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
5329 /* don't activate new spares if we are orom constrained
5330 * and there is already a volume active in the container
5332 if (super
->orom
&& dl
->index
< 0 && mpb
->num_raid_devs
)
5335 e
= get_extents(super
, dl
);
5338 for (i
= 1; e
[i
-1].size
; i
++)
5346 maxsize
= merge_extents(super
, extent_cnt
);
5350 minsize
= chunk
* 2;
5352 if (cnt
< raiddisks
||
5353 (super
->orom
&& used
&& used
!= raiddisks
) ||
5354 maxsize
< minsize
||
5356 fprintf(stderr
, Name
": not enough devices with space to create array.\n");
5357 return 0; /* No enough free spaces large enough */
5369 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
5371 dl
->raiddisk
= cnt
++;
5378 static int validate_geometry_imsm(struct supertype
*st
, int level
, int layout
,
5379 int raiddisks
, int *chunk
, unsigned long long size
,
5380 char *dev
, unsigned long long *freesize
,
5388 * if given unused devices create a container
5389 * if given given devices in a container create a member volume
5391 if (level
== LEVEL_CONTAINER
) {
5392 /* Must be a fresh device to add to a container */
5393 return validate_geometry_imsm_container(st
, level
, layout
,
5395 chunk
?*chunk
:0, size
,
5402 if (!validate_geometry_imsm_orom(st
->sb
, level
, layout
,
5406 /* we are being asked to automatically layout a
5407 * new volume based on the current contents of
5408 * the container. If the the parameters can be
5409 * satisfied reserve_space will record the disks,
5410 * start offset, and size of the volume to be
5411 * created. add_to_super and getinfo_super
5412 * detect when autolayout is in progress.
5415 return reserve_space(st
, raiddisks
, size
,
5416 chunk
?*chunk
:0, freesize
);
5421 /* creating in a given container */
5422 return validate_geometry_imsm_volume(st
, level
, layout
,
5423 raiddisks
, chunk
, size
,
5424 dev
, freesize
, verbose
);
5427 /* This device needs to be a device in an 'imsm' container */
5428 fd
= open(dev
, O_RDONLY
|O_EXCL
, 0);
5432 Name
": Cannot create this array on device %s\n",
5437 if (errno
!= EBUSY
|| (fd
= open(dev
, O_RDONLY
, 0)) < 0) {
5439 fprintf(stderr
, Name
": Cannot open %s: %s\n",
5440 dev
, strerror(errno
));
5443 /* Well, it is in use by someone, maybe an 'imsm' container. */
5444 cfd
= open_container(fd
);
5448 fprintf(stderr
, Name
": Cannot use %s: It is busy\n",
5452 sra
= sysfs_read(cfd
, 0, GET_VERSION
);
5453 if (sra
&& sra
->array
.major_version
== -1 &&
5454 strcmp(sra
->text_version
, "imsm") == 0)
5458 /* This is a member of a imsm container. Load the container
5459 * and try to create a volume
5461 struct intel_super
*super
;
5463 if (load_super_imsm_all(st
, cfd
, (void **) &super
, NULL
) == 0) {
5465 st
->container_dev
= fd2devnum(cfd
);
5467 return validate_geometry_imsm_volume(st
, level
, layout
,
5476 fprintf(stderr
, Name
": failed container membership check\n");
5482 static void default_geometry_imsm(struct supertype
*st
, int *level
, int *layout
, int *chunk
)
5484 struct intel_super
*super
= st
->sb
;
5486 if (level
&& *level
== UnSet
)
5487 *level
= LEVEL_CONTAINER
;
5489 if (level
&& layout
&& *layout
== UnSet
)
5490 *layout
= imsm_level_to_layout(*level
);
5492 if (chunk
&& (*chunk
== UnSet
|| *chunk
== 0))
5493 *chunk
= imsm_default_chunk(super
->orom
);
5496 static void handle_missing(struct intel_super
*super
, struct imsm_dev
*dev
);
5498 static int kill_subarray_imsm(struct supertype
*st
)
5500 /* remove the subarray currently referenced by ->current_vol */
5502 struct intel_dev
**dp
;
5503 struct intel_super
*super
= st
->sb
;
5504 __u8 current_vol
= super
->current_vol
;
5505 struct imsm_super
*mpb
= super
->anchor
;
5507 if (super
->current_vol
< 0)
5509 super
->current_vol
= -1; /* invalidate subarray cursor */
5511 /* block deletions that would change the uuid of active subarrays
5513 * FIXME when immutable ids are available, but note that we'll
5514 * also need to fixup the invalidated/active subarray indexes in
5517 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
5520 if (i
< current_vol
)
5522 sprintf(subarray
, "%u", i
);
5523 if (is_subarray_active(subarray
, st
->devname
)) {
5525 Name
": deleting subarray-%d would change the UUID of active subarray-%d, aborting\n",
5532 if (st
->update_tail
) {
5533 struct imsm_update_kill_array
*u
= malloc(sizeof(*u
));
5537 u
->type
= update_kill_array
;
5538 u
->dev_idx
= current_vol
;
5539 append_metadata_update(st
, u
, sizeof(*u
));
5544 for (dp
= &super
->devlist
; *dp
;)
5545 if ((*dp
)->index
== current_vol
) {
5548 handle_missing(super
, (*dp
)->dev
);
5549 if ((*dp
)->index
> current_vol
)
5554 /* no more raid devices, all active components are now spares,
5555 * but of course failed are still failed
5557 if (--mpb
->num_raid_devs
== 0) {
5560 for (d
= super
->disks
; d
; d
= d
->next
)
5565 super
->updates_pending
++;
5570 static int update_subarray_imsm(struct supertype
*st
, char *subarray
,
5571 char *update
, struct mddev_ident
*ident
)
5573 /* update the subarray currently referenced by ->current_vol */
5574 struct intel_super
*super
= st
->sb
;
5575 struct imsm_super
*mpb
= super
->anchor
;
5577 if (strcmp(update
, "name") == 0) {
5578 char *name
= ident
->name
;
5582 if (is_subarray_active(subarray
, st
->devname
)) {
5584 Name
": Unable to update name of active subarray\n");
5588 if (!check_name(super
, name
, 0))
5591 vol
= strtoul(subarray
, &ep
, 10);
5592 if (*ep
!= '\0' || vol
>= super
->anchor
->num_raid_devs
)
5595 if (st
->update_tail
) {
5596 struct imsm_update_rename_array
*u
= malloc(sizeof(*u
));
5600 u
->type
= update_rename_array
;
5602 snprintf((char *) u
->name
, MAX_RAID_SERIAL_LEN
, "%s", name
);
5603 append_metadata_update(st
, u
, sizeof(*u
));
5605 struct imsm_dev
*dev
;
5608 dev
= get_imsm_dev(super
, vol
);
5609 snprintf((char *) dev
->volume
, MAX_RAID_SERIAL_LEN
, "%s", name
);
5610 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
5611 dev
= get_imsm_dev(super
, i
);
5612 handle_missing(super
, dev
);
5614 super
->updates_pending
++;
5622 static int is_gen_migration(struct imsm_dev
*dev
)
5627 if (!dev
->vol
.migr_state
)
5630 if (migr_type(dev
) == MIGR_GEN_MIGR
)
5635 #endif /* MDASSEMBLE */
5637 static int is_rebuilding(struct imsm_dev
*dev
)
5639 struct imsm_map
*migr_map
;
5641 if (!dev
->vol
.migr_state
)
5644 if (migr_type(dev
) != MIGR_REBUILD
)
5647 migr_map
= get_imsm_map(dev
, 1);
5649 if (migr_map
->map_state
== IMSM_T_STATE_DEGRADED
)
5655 static void update_recovery_start(struct intel_super
*super
,
5656 struct imsm_dev
*dev
,
5657 struct mdinfo
*array
)
5659 struct mdinfo
*rebuild
= NULL
;
5663 if (!is_rebuilding(dev
))
5666 /* Find the rebuild target, but punt on the dual rebuild case */
5667 for (d
= array
->devs
; d
; d
= d
->next
)
5668 if (d
->recovery_start
== 0) {
5675 /* (?) none of the disks are marked with
5676 * IMSM_ORD_REBUILD, so assume they are missing and the
5677 * disk_ord_tbl was not correctly updated
5679 dprintf("%s: failed to locate out-of-sync disk\n", __func__
);
5683 units
= __le32_to_cpu(dev
->vol
.curr_migr_unit
);
5684 rebuild
->recovery_start
= units
* blocks_per_migr_unit(super
, dev
);
5688 static int recover_backup_imsm(struct supertype
*st
, struct mdinfo
*info
);
5691 static struct mdinfo
*container_content_imsm(struct supertype
*st
, char *subarray
)
5693 /* Given a container loaded by load_super_imsm_all,
5694 * extract information about all the arrays into
5696 * If 'subarray' is given, just extract info about that array.
5698 * For each imsm_dev create an mdinfo, fill it in,
5699 * then look for matching devices in super->disks
5700 * and create appropriate device mdinfo.
5702 struct intel_super
*super
= st
->sb
;
5703 struct imsm_super
*mpb
= super
->anchor
;
5704 struct mdinfo
*rest
= NULL
;
5708 int spare_disks
= 0;
5710 /* do not assemble arrays when not all attributes are supported */
5711 if (imsm_check_attributes(mpb
->attributes
) == 0) {
5713 fprintf(stderr
, Name
": Unsupported attributes in IMSM metadata."
5714 "Arrays activation is blocked.\n");
5717 /* check for bad blocks */
5718 if (imsm_bbm_log_size(super
->anchor
)) {
5719 fprintf(stderr
, Name
": BBM log found in IMSM metadata."
5720 "Arrays activation is blocked.\n");
5725 /* count spare devices, not used in maps
5727 for (d
= super
->disks
; d
; d
= d
->next
)
5731 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
5732 struct imsm_dev
*dev
;
5733 struct imsm_map
*map
;
5734 struct imsm_map
*map2
;
5735 struct mdinfo
*this;
5740 (i
!= strtoul(subarray
, &ep
, 10) || *ep
!= '\0'))
5743 dev
= get_imsm_dev(super
, i
);
5744 map
= get_imsm_map(dev
, 0);
5745 map2
= get_imsm_map(dev
, 1);
5747 /* do not publish arrays that are in the middle of an
5748 * unsupported migration
5750 if (dev
->vol
.migr_state
&&
5751 (migr_type(dev
) == MIGR_STATE_CHANGE
)) {
5752 fprintf(stderr
, Name
": cannot assemble volume '%.16s':"
5753 " unsupported migration in progress\n",
5757 /* do not publish arrays that are not support by controller's
5761 chunk
= __le16_to_cpu(map
->blocks_per_strip
) >> 1;
5762 this = malloc(sizeof(*this));
5764 fprintf(stderr
, Name
": failed to allocate %zu bytes\n",
5769 super
->current_vol
= i
;
5770 getinfo_super_imsm_volume(st
, this, NULL
);
5773 /* mdadm does not support all metadata features- set the bit in all arrays state */
5774 if (!validate_geometry_imsm_orom(super
,
5775 get_imsm_raid_level(map
), /* RAID level */
5776 imsm_level_to_layout(get_imsm_raid_level(map
)),
5777 map
->num_members
, /* raid disks */
5780 fprintf(stderr
, Name
": IMSM RAID geometry validation"
5781 " failed. Array %s activation is blocked.\n",
5783 this->array
.state
|=
5784 (1<<MD_SB_BLOCK_CONTAINER_RESHAPE
) |
5785 (1<<MD_SB_BLOCK_VOLUME
);
5789 /* if array has bad blocks, set suitable bit in all arrays state */
5791 this->array
.state
|=
5792 (1<<MD_SB_BLOCK_CONTAINER_RESHAPE
) |
5793 (1<<MD_SB_BLOCK_VOLUME
);
5795 for (slot
= 0 ; slot
< map
->num_members
; slot
++) {
5796 unsigned long long recovery_start
;
5797 struct mdinfo
*info_d
;
5804 idx
= get_imsm_disk_idx(dev
, slot
, 0);
5805 ord
= get_imsm_ord_tbl_ent(dev
, slot
, -1);
5806 for (d
= super
->disks
; d
; d
= d
->next
)
5807 if (d
->index
== idx
)
5810 recovery_start
= MaxSector
;
5813 if (d
&& is_failed(&d
->disk
))
5815 if (ord
& IMSM_ORD_REBUILD
)
5819 * if we skip some disks the array will be assmebled degraded;
5820 * reset resync start to avoid a dirty-degraded
5821 * situation when performing the intial sync
5823 * FIXME handle dirty degraded
5825 if ((skip
|| recovery_start
== 0) && !dev
->vol
.dirty
)
5826 this->resync_start
= MaxSector
;
5830 info_d
= calloc(1, sizeof(*info_d
));
5832 fprintf(stderr
, Name
": failed to allocate disk"
5833 " for volume %.16s\n", dev
->volume
);
5834 info_d
= this->devs
;
5836 struct mdinfo
*d
= info_d
->next
;
5845 info_d
->next
= this->devs
;
5846 this->devs
= info_d
;
5848 info_d
->disk
.number
= d
->index
;
5849 info_d
->disk
.major
= d
->major
;
5850 info_d
->disk
.minor
= d
->minor
;
5851 info_d
->disk
.raid_disk
= slot
;
5852 info_d
->recovery_start
= recovery_start
;
5854 if (slot
< map2
->num_members
)
5855 info_d
->disk
.state
= (1 << MD_DISK_ACTIVE
);
5857 this->array
.spare_disks
++;
5859 if (slot
< map
->num_members
)
5860 info_d
->disk
.state
= (1 << MD_DISK_ACTIVE
);
5862 this->array
.spare_disks
++;
5864 if (info_d
->recovery_start
== MaxSector
)
5865 this->array
.working_disks
++;
5867 info_d
->events
= __le32_to_cpu(mpb
->generation_num
);
5868 info_d
->data_offset
= __le32_to_cpu(map
->pba_of_lba0
);
5869 info_d
->component_size
= __le32_to_cpu(map
->blocks_per_member
);
5871 /* now that the disk list is up-to-date fixup recovery_start */
5872 update_recovery_start(super
, dev
, this);
5873 this->array
.spare_disks
+= spare_disks
;
5876 /* check for reshape */
5877 if (this->reshape_active
== 1)
5878 recover_backup_imsm(st
, this);
5887 static __u8
imsm_check_degraded(struct intel_super
*super
, struct imsm_dev
*dev
, int failed
)
5889 struct imsm_map
*map
= get_imsm_map(dev
, 0);
5892 return map
->map_state
== IMSM_T_STATE_UNINITIALIZED
?
5893 IMSM_T_STATE_UNINITIALIZED
: IMSM_T_STATE_NORMAL
;
5895 switch (get_imsm_raid_level(map
)) {
5897 return IMSM_T_STATE_FAILED
;
5900 if (failed
< map
->num_members
)
5901 return IMSM_T_STATE_DEGRADED
;
5903 return IMSM_T_STATE_FAILED
;
5908 * check to see if any mirrors have failed, otherwise we
5909 * are degraded. Even numbered slots are mirrored on
5913 /* gcc -Os complains that this is unused */
5914 int insync
= insync
;
5916 for (i
= 0; i
< map
->num_members
; i
++) {
5917 __u32 ord
= get_imsm_ord_tbl_ent(dev
, i
, -1);
5918 int idx
= ord_to_idx(ord
);
5919 struct imsm_disk
*disk
;
5921 /* reset the potential in-sync count on even-numbered
5922 * slots. num_copies is always 2 for imsm raid10
5927 disk
= get_imsm_disk(super
, idx
);
5928 if (!disk
|| is_failed(disk
) || ord
& IMSM_ORD_REBUILD
)
5931 /* no in-sync disks left in this mirror the
5935 return IMSM_T_STATE_FAILED
;
5938 return IMSM_T_STATE_DEGRADED
;
5942 return IMSM_T_STATE_DEGRADED
;
5944 return IMSM_T_STATE_FAILED
;
5950 return map
->map_state
;
5953 static int imsm_count_failed(struct intel_super
*super
, struct imsm_dev
*dev
)
5957 struct imsm_disk
*disk
;
5958 struct imsm_map
*map
= get_imsm_map(dev
, 0);
5959 struct imsm_map
*prev
= get_imsm_map(dev
, dev
->vol
.migr_state
);
5960 struct imsm_map
*map_for_loop
;
5964 /* at the beginning of migration we set IMSM_ORD_REBUILD on
5965 * disks that are being rebuilt. New failures are recorded to
5966 * map[0]. So we look through all the disks we started with and
5967 * see if any failures are still present, or if any new ones
5970 map_for_loop
= prev
;
5971 if (is_gen_migration(dev
))
5972 if (prev
&& (map
->num_members
> prev
->num_members
))
5975 for (i
= 0; i
< map_for_loop
->num_members
; i
++) {
5977 if (i
< prev
->num_members
)
5978 ord
|= __le32_to_cpu(prev
->disk_ord_tbl
[i
]);
5979 if (i
< map
->num_members
)
5980 ord
|= __le32_to_cpu(map
->disk_ord_tbl
[i
]);
5981 idx
= ord_to_idx(ord
);
5983 disk
= get_imsm_disk(super
, idx
);
5984 if (!disk
|| is_failed(disk
) || ord
& IMSM_ORD_REBUILD
)
5992 static int imsm_open_new(struct supertype
*c
, struct active_array
*a
,
5995 struct intel_super
*super
= c
->sb
;
5996 struct imsm_super
*mpb
= super
->anchor
;
5998 if (atoi(inst
) >= mpb
->num_raid_devs
) {
5999 fprintf(stderr
, "%s: subarry index %d, out of range\n",
6000 __func__
, atoi(inst
));
6004 dprintf("imsm: open_new %s\n", inst
);
6005 a
->info
.container_member
= atoi(inst
);
6009 static int is_resyncing(struct imsm_dev
*dev
)
6011 struct imsm_map
*migr_map
;
6013 if (!dev
->vol
.migr_state
)
6016 if (migr_type(dev
) == MIGR_INIT
||
6017 migr_type(dev
) == MIGR_REPAIR
)
6020 if (migr_type(dev
) == MIGR_GEN_MIGR
)
6023 migr_map
= get_imsm_map(dev
, 1);
6025 if ((migr_map
->map_state
== IMSM_T_STATE_NORMAL
) &&
6026 (dev
->vol
.migr_type
!= MIGR_GEN_MIGR
))
6032 /* return true if we recorded new information */
6033 static int mark_failure(struct imsm_dev
*dev
, struct imsm_disk
*disk
, int idx
)
6037 struct imsm_map
*map
;
6038 char buf
[MAX_RAID_SERIAL_LEN
+3];
6039 unsigned int len
, shift
= 0;
6041 /* new failures are always set in map[0] */
6042 map
= get_imsm_map(dev
, 0);
6044 slot
= get_imsm_disk_slot(map
, idx
);
6048 ord
= __le32_to_cpu(map
->disk_ord_tbl
[slot
]);
6049 if (is_failed(disk
) && (ord
& IMSM_ORD_REBUILD
))
6052 memcpy(buf
, disk
->serial
, MAX_RAID_SERIAL_LEN
);
6053 buf
[MAX_RAID_SERIAL_LEN
] = '\000';
6055 if ((len
= strlen(buf
)) >= MAX_RAID_SERIAL_LEN
)
6056 shift
= len
- MAX_RAID_SERIAL_LEN
+ 1;
6057 strncpy((char *)disk
->serial
, &buf
[shift
], MAX_RAID_SERIAL_LEN
);
6059 disk
->status
|= FAILED_DISK
;
6060 set_imsm_ord_tbl_ent(map
, slot
, idx
| IMSM_ORD_REBUILD
);
6061 if (map
->failed_disk_num
== 0xff)
6062 map
->failed_disk_num
= slot
;
6066 static void mark_missing(struct imsm_dev
*dev
, struct imsm_disk
*disk
, int idx
)
6068 mark_failure(dev
, disk
, idx
);
6070 if (disk
->scsi_id
== __cpu_to_le32(~(__u32
)0))
6073 disk
->scsi_id
= __cpu_to_le32(~(__u32
)0);
6074 memmove(&disk
->serial
[0], &disk
->serial
[1], MAX_RAID_SERIAL_LEN
- 1);
6077 static void handle_missing(struct intel_super
*super
, struct imsm_dev
*dev
)
6083 if (!super
->missing
)
6085 failed
= imsm_count_failed(super
, dev
);
6086 map_state
= imsm_check_degraded(super
, dev
, failed
);
6088 dprintf("imsm: mark missing\n");
6089 end_migration(dev
, map_state
);
6090 for (dl
= super
->missing
; dl
; dl
= dl
->next
)
6091 mark_missing(dev
, &dl
->disk
, dl
->index
);
6092 super
->updates_pending
++;
6095 static unsigned long long imsm_set_array_size(struct imsm_dev
*dev
)
6097 int used_disks
= imsm_num_data_members(dev
, 0);
6098 unsigned long long array_blocks
;
6099 struct imsm_map
*map
;
6101 if (used_disks
== 0) {
6102 /* when problems occures
6103 * return current array_blocks value
6105 array_blocks
= __le32_to_cpu(dev
->size_high
);
6106 array_blocks
= array_blocks
<< 32;
6107 array_blocks
+= __le32_to_cpu(dev
->size_low
);
6109 return array_blocks
;
6112 /* set array size in metadata
6114 map
= get_imsm_map(dev
, 0);
6115 array_blocks
= map
->blocks_per_member
* used_disks
;
6117 /* round array size down to closest MB
6119 array_blocks
= (array_blocks
>> SECT_PER_MB_SHIFT
) << SECT_PER_MB_SHIFT
;
6120 dev
->size_low
= __cpu_to_le32((__u32
)array_blocks
);
6121 dev
->size_high
= __cpu_to_le32((__u32
)(array_blocks
>> 32));
6123 return array_blocks
;
6126 static void imsm_set_disk(struct active_array
*a
, int n
, int state
);
6128 static void imsm_progress_container_reshape(struct intel_super
*super
)
6130 /* if no device has a migr_state, but some device has a
6131 * different number of members than the previous device, start
6132 * changing the number of devices in this device to match
6135 struct imsm_super
*mpb
= super
->anchor
;
6136 int prev_disks
= -1;
6140 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
6141 struct imsm_dev
*dev
= get_imsm_dev(super
, i
);
6142 struct imsm_map
*map
= get_imsm_map(dev
, 0);
6143 struct imsm_map
*map2
;
6144 int prev_num_members
;
6146 if (dev
->vol
.migr_state
)
6149 if (prev_disks
== -1)
6150 prev_disks
= map
->num_members
;
6151 if (prev_disks
== map
->num_members
)
6154 /* OK, this array needs to enter reshape mode.
6155 * i.e it needs a migr_state
6158 copy_map_size
= sizeof_imsm_map(map
);
6159 prev_num_members
= map
->num_members
;
6160 map
->num_members
= prev_disks
;
6161 dev
->vol
.migr_state
= 1;
6162 dev
->vol
.curr_migr_unit
= 0;
6163 set_migr_type(dev
, MIGR_GEN_MIGR
);
6164 for (i
= prev_num_members
;
6165 i
< map
->num_members
; i
++)
6166 set_imsm_ord_tbl_ent(map
, i
, i
);
6167 map2
= get_imsm_map(dev
, 1);
6168 /* Copy the current map */
6169 memcpy(map2
, map
, copy_map_size
);
6170 map2
->num_members
= prev_num_members
;
6172 imsm_set_array_size(dev
);
6173 super
->updates_pending
++;
6177 /* Handle dirty -> clean transititions, resync and reshape. Degraded and rebuild
6178 * states are handled in imsm_set_disk() with one exception, when a
6179 * resync is stopped due to a new failure this routine will set the
6180 * 'degraded' state for the array.
6182 static int imsm_set_array_state(struct active_array
*a
, int consistent
)
6184 int inst
= a
->info
.container_member
;
6185 struct intel_super
*super
= a
->container
->sb
;
6186 struct imsm_dev
*dev
= get_imsm_dev(super
, inst
);
6187 struct imsm_map
*map
= get_imsm_map(dev
, 0);
6188 int failed
= imsm_count_failed(super
, dev
);
6189 __u8 map_state
= imsm_check_degraded(super
, dev
, failed
);
6190 __u32 blocks_per_unit
;
6192 if (dev
->vol
.migr_state
&&
6193 dev
->vol
.migr_type
== MIGR_GEN_MIGR
) {
6194 /* array state change is blocked due to reshape action
6196 * - abort the reshape (if last_checkpoint is 0 and action!= reshape)
6197 * - finish the reshape (if last_checkpoint is big and action != reshape)
6198 * - update curr_migr_unit
6200 if (a
->curr_action
== reshape
) {
6201 /* still reshaping, maybe update curr_migr_unit */
6202 goto mark_checkpoint
;
6204 if (a
->last_checkpoint
== 0 && a
->prev_action
== reshape
) {
6205 /* for some reason we aborted the reshape.
6207 * disable automatic metadata rollback
6208 * user action is required to recover process
6211 struct imsm_map
*map2
= get_imsm_map(dev
, 1);
6212 dev
->vol
.migr_state
= 0;
6213 set_migr_type(dev
, 0);
6214 dev
->vol
.curr_migr_unit
= 0;
6215 memcpy(map
, map2
, sizeof_imsm_map(map2
));
6216 super
->updates_pending
++;
6219 if (a
->last_checkpoint
>= a
->info
.component_size
) {
6220 unsigned long long array_blocks
;
6224 used_disks
= imsm_num_data_members(dev
, 0);
6225 if (used_disks
> 0) {
6227 map
->blocks_per_member
*
6229 /* round array size down to closest MB
6231 array_blocks
= (array_blocks
6232 >> SECT_PER_MB_SHIFT
)
6233 << SECT_PER_MB_SHIFT
;
6234 a
->info
.custom_array_size
= array_blocks
;
6235 /* encourage manager to update array
6239 a
->check_reshape
= 1;
6241 /* finalize online capacity expansion/reshape */
6242 for (mdi
= a
->info
.devs
; mdi
; mdi
= mdi
->next
)
6244 mdi
->disk
.raid_disk
,
6247 imsm_progress_container_reshape(super
);
6252 /* before we activate this array handle any missing disks */
6253 if (consistent
== 2)
6254 handle_missing(super
, dev
);
6256 if (consistent
== 2 &&
6257 (!is_resync_complete(&a
->info
) ||
6258 map_state
!= IMSM_T_STATE_NORMAL
||
6259 dev
->vol
.migr_state
))
6262 if (is_resync_complete(&a
->info
)) {
6263 /* complete intialization / resync,
6264 * recovery and interrupted recovery is completed in
6267 if (is_resyncing(dev
)) {
6268 dprintf("imsm: mark resync done\n");
6269 end_migration(dev
, map_state
);
6270 super
->updates_pending
++;
6271 a
->last_checkpoint
= 0;
6273 } else if ((!is_resyncing(dev
) && !failed
) &&
6274 (imsm_reshape_blocks_arrays_changes(super
) == 0)) {
6275 /* mark the start of the init process if nothing is failed */
6276 dprintf("imsm: mark resync start\n");
6277 if (map
->map_state
== IMSM_T_STATE_UNINITIALIZED
)
6278 migrate(dev
, super
, IMSM_T_STATE_NORMAL
, MIGR_INIT
);
6280 migrate(dev
, super
, IMSM_T_STATE_NORMAL
, MIGR_REPAIR
);
6281 super
->updates_pending
++;
6285 /* skip checkpointing for general migration,
6286 * it is controlled in mdadm
6288 if (is_gen_migration(dev
))
6289 goto skip_mark_checkpoint
;
6291 /* check if we can update curr_migr_unit from resync_start, recovery_start */
6292 blocks_per_unit
= blocks_per_migr_unit(super
, dev
);
6293 if (blocks_per_unit
) {
6297 units
= a
->last_checkpoint
/ blocks_per_unit
;
6300 /* check that we did not overflow 32-bits, and that
6301 * curr_migr_unit needs updating
6303 if (units32
== units
&&
6305 __le32_to_cpu(dev
->vol
.curr_migr_unit
) != units32
) {
6306 dprintf("imsm: mark checkpoint (%u)\n", units32
);
6307 dev
->vol
.curr_migr_unit
= __cpu_to_le32(units32
);
6308 super
->updates_pending
++;
6312 skip_mark_checkpoint
:
6313 /* mark dirty / clean */
6314 if (dev
->vol
.dirty
!= !consistent
) {
6315 dprintf("imsm: mark '%s'\n", consistent
? "clean" : "dirty");
6320 super
->updates_pending
++;
6326 static void imsm_set_disk(struct active_array
*a
, int n
, int state
)
6328 int inst
= a
->info
.container_member
;
6329 struct intel_super
*super
= a
->container
->sb
;
6330 struct imsm_dev
*dev
= get_imsm_dev(super
, inst
);
6331 struct imsm_map
*map
= get_imsm_map(dev
, 0);
6332 struct imsm_disk
*disk
;
6337 if (n
> map
->num_members
)
6338 fprintf(stderr
, "imsm: set_disk %d out of range 0..%d\n",
6339 n
, map
->num_members
- 1);
6344 dprintf("imsm: set_disk %d:%x\n", n
, state
);
6346 ord
= get_imsm_ord_tbl_ent(dev
, n
, -1);
6347 disk
= get_imsm_disk(super
, ord_to_idx(ord
));
6349 /* check for new failures */
6350 if (state
& DS_FAULTY
) {
6351 if (mark_failure(dev
, disk
, ord_to_idx(ord
)))
6352 super
->updates_pending
++;
6355 /* check if in_sync */
6356 if (state
& DS_INSYNC
&& ord
& IMSM_ORD_REBUILD
&& is_rebuilding(dev
)) {
6357 struct imsm_map
*migr_map
= get_imsm_map(dev
, 1);
6359 set_imsm_ord_tbl_ent(migr_map
, n
, ord_to_idx(ord
));
6360 super
->updates_pending
++;
6363 failed
= imsm_count_failed(super
, dev
);
6364 map_state
= imsm_check_degraded(super
, dev
, failed
);
6366 /* check if recovery complete, newly degraded, or failed */
6367 if (map_state
== IMSM_T_STATE_NORMAL
&& is_rebuilding(dev
)) {
6368 end_migration(dev
, map_state
);
6369 map
= get_imsm_map(dev
, 0);
6370 map
->failed_disk_num
= ~0;
6371 super
->updates_pending
++;
6372 a
->last_checkpoint
= 0;
6373 } else if (map_state
== IMSM_T_STATE_DEGRADED
&&
6374 map
->map_state
!= map_state
&&
6375 !dev
->vol
.migr_state
) {
6376 dprintf("imsm: mark degraded\n");
6377 map
->map_state
= map_state
;
6378 super
->updates_pending
++;
6379 a
->last_checkpoint
= 0;
6380 } else if (map_state
== IMSM_T_STATE_FAILED
&&
6381 map
->map_state
!= map_state
) {
6382 dprintf("imsm: mark failed\n");
6383 end_migration(dev
, map_state
);
6384 super
->updates_pending
++;
6385 a
->last_checkpoint
= 0;
6386 } else if (is_gen_migration(dev
)) {
6387 dprintf("imsm: Detected General Migration in state: ");
6388 if (map_state
== IMSM_T_STATE_NORMAL
) {
6389 end_migration(dev
, map_state
);
6390 map
= get_imsm_map(dev
, 0);
6391 map
->failed_disk_num
= ~0;
6392 dprintf("normal\n");
6394 if (map_state
== IMSM_T_STATE_DEGRADED
) {
6395 printf("degraded\n");
6396 end_migration(dev
, map_state
);
6398 dprintf("failed\n");
6400 map
->map_state
= map_state
;
6402 super
->updates_pending
++;
6406 static int store_imsm_mpb(int fd
, struct imsm_super
*mpb
)
6409 __u32 mpb_size
= __le32_to_cpu(mpb
->mpb_size
);
6410 unsigned long long dsize
;
6411 unsigned long long sectors
;
6413 get_dev_size(fd
, NULL
, &dsize
);
6415 if (mpb_size
> 512) {
6416 /* -1 to account for anchor */
6417 sectors
= mpb_sectors(mpb
) - 1;
6419 /* write the extended mpb to the sectors preceeding the anchor */
6420 if (lseek64(fd
, dsize
- (512 * (2 + sectors
)), SEEK_SET
) < 0)
6423 if ((unsigned long long)write(fd
, buf
+ 512, 512 * sectors
)
6428 /* first block is stored on second to last sector of the disk */
6429 if (lseek64(fd
, dsize
- (512 * 2), SEEK_SET
) < 0)
6432 if (write(fd
, buf
, 512) != 512)
6438 static void imsm_sync_metadata(struct supertype
*container
)
6440 struct intel_super
*super
= container
->sb
;
6442 dprintf("sync metadata: %d\n", super
->updates_pending
);
6443 if (!super
->updates_pending
)
6446 write_super_imsm(container
, 0);
6448 super
->updates_pending
= 0;
6451 static struct dl
*imsm_readd(struct intel_super
*super
, int idx
, struct active_array
*a
)
6453 struct imsm_dev
*dev
= get_imsm_dev(super
, a
->info
.container_member
);
6454 int i
= get_imsm_disk_idx(dev
, idx
, -1);
6457 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
6461 if (dl
&& is_failed(&dl
->disk
))
6465 dprintf("%s: found %x:%x\n", __func__
, dl
->major
, dl
->minor
);
6470 static struct dl
*imsm_add_spare(struct intel_super
*super
, int slot
,
6471 struct active_array
*a
, int activate_new
,
6472 struct mdinfo
*additional_test_list
)
6474 struct imsm_dev
*dev
= get_imsm_dev(super
, a
->info
.container_member
);
6475 int idx
= get_imsm_disk_idx(dev
, slot
, -1);
6476 struct imsm_super
*mpb
= super
->anchor
;
6477 struct imsm_map
*map
;
6478 unsigned long long pos
;
6483 __u32 array_start
= 0;
6484 __u32 array_end
= 0;
6486 struct mdinfo
*test_list
;
6488 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
6489 /* If in this array, skip */
6490 for (d
= a
->info
.devs
; d
; d
= d
->next
)
6491 if (d
->state_fd
>= 0 &&
6492 d
->disk
.major
== dl
->major
&&
6493 d
->disk
.minor
== dl
->minor
) {
6494 dprintf("%x:%x already in array\n",
6495 dl
->major
, dl
->minor
);
6500 test_list
= additional_test_list
;
6502 if (test_list
->disk
.major
== dl
->major
&&
6503 test_list
->disk
.minor
== dl
->minor
) {
6504 dprintf("%x:%x already in additional test list\n",
6505 dl
->major
, dl
->minor
);
6508 test_list
= test_list
->next
;
6513 /* skip in use or failed drives */
6514 if (is_failed(&dl
->disk
) || idx
== dl
->index
||
6516 dprintf("%x:%x status (failed: %d index: %d)\n",
6517 dl
->major
, dl
->minor
, is_failed(&dl
->disk
), idx
);
6521 /* skip pure spares when we are looking for partially
6522 * assimilated drives
6524 if (dl
->index
== -1 && !activate_new
)
6527 /* Does this unused device have the requisite free space?
6528 * It needs to be able to cover all member volumes
6530 ex
= get_extents(super
, dl
);
6532 dprintf("cannot get extents\n");
6535 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
6536 dev
= get_imsm_dev(super
, i
);
6537 map
= get_imsm_map(dev
, 0);
6539 /* check if this disk is already a member of
6542 if (get_imsm_disk_slot(map
, dl
->index
) >= 0)
6548 array_start
= __le32_to_cpu(map
->pba_of_lba0
);
6549 array_end
= array_start
+
6550 __le32_to_cpu(map
->blocks_per_member
) - 1;
6553 /* check that we can start at pba_of_lba0 with
6554 * blocks_per_member of space
6556 if (array_start
>= pos
&& array_end
< ex
[j
].start
) {
6560 pos
= ex
[j
].start
+ ex
[j
].size
;
6562 } while (ex
[j
-1].size
);
6569 if (i
< mpb
->num_raid_devs
) {
6570 dprintf("%x:%x does not have %u to %u available\n",
6571 dl
->major
, dl
->minor
, array_start
, array_end
);
6582 static int imsm_rebuild_allowed(struct supertype
*cont
, int dev_idx
, int failed
)
6584 struct imsm_dev
*dev2
;
6585 struct imsm_map
*map
;
6591 dev2
= get_imsm_dev(cont
->sb
, dev_idx
);
6593 state
= imsm_check_degraded(cont
->sb
, dev2
, failed
);
6594 if (state
== IMSM_T_STATE_FAILED
) {
6595 map
= get_imsm_map(dev2
, 0);
6598 for (slot
= 0; slot
< map
->num_members
; slot
++) {
6600 * Check if failed disks are deleted from intel
6601 * disk list or are marked to be deleted
6603 idx
= get_imsm_disk_idx(dev2
, slot
, -1);
6604 idisk
= get_imsm_dl_disk(cont
->sb
, idx
);
6606 * Do not rebuild the array if failed disks
6607 * from failed sub-array are not removed from
6611 is_failed(&idisk
->disk
) &&
6612 (idisk
->action
!= DISK_REMOVE
))
6620 static struct mdinfo
*imsm_activate_spare(struct active_array
*a
,
6621 struct metadata_update
**updates
)
6624 * Find a device with unused free space and use it to replace a
6625 * failed/vacant region in an array. We replace failed regions one a
6626 * array at a time. The result is that a new spare disk will be added
6627 * to the first failed array and after the monitor has finished
6628 * propagating failures the remainder will be consumed.
6630 * FIXME add a capability for mdmon to request spares from another
6634 struct intel_super
*super
= a
->container
->sb
;
6635 int inst
= a
->info
.container_member
;
6636 struct imsm_dev
*dev
= get_imsm_dev(super
, inst
);
6637 struct imsm_map
*map
= get_imsm_map(dev
, 0);
6638 int failed
= a
->info
.array
.raid_disks
;
6639 struct mdinfo
*rv
= NULL
;
6642 struct metadata_update
*mu
;
6644 struct imsm_update_activate_spare
*u
;
6649 for (d
= a
->info
.devs
; d
; d
= d
->next
) {
6650 if ((d
->curr_state
& DS_FAULTY
) &&
6652 /* wait for Removal to happen */
6654 if (d
->state_fd
>= 0)
6658 dprintf("imsm: activate spare: inst=%d failed=%d (%d) level=%d\n",
6659 inst
, failed
, a
->info
.array
.raid_disks
, a
->info
.array
.level
);
6661 if (imsm_reshape_blocks_arrays_changes(super
))
6664 if (a
->info
.array
.level
== 4)
6665 /* No repair for takeovered array
6666 * imsm doesn't support raid4
6670 if (imsm_check_degraded(super
, dev
, failed
) != IMSM_T_STATE_DEGRADED
)
6674 * If there are any failed disks check state of the other volume.
6675 * Block rebuild if the another one is failed until failed disks
6676 * are removed from container.
6679 dprintf("found failed disks in %.*s, check if there another"
6680 "failed sub-array.\n",
6681 MAX_RAID_SERIAL_LEN
, dev
->volume
);
6682 /* check if states of the other volumes allow for rebuild */
6683 for (i
= 0; i
< super
->anchor
->num_raid_devs
; i
++) {
6685 allowed
= imsm_rebuild_allowed(a
->container
,
6693 /* For each slot, if it is not working, find a spare */
6694 for (i
= 0; i
< a
->info
.array
.raid_disks
; i
++) {
6695 for (d
= a
->info
.devs
; d
; d
= d
->next
)
6696 if (d
->disk
.raid_disk
== i
)
6698 dprintf("found %d: %p %x\n", i
, d
, d
?d
->curr_state
:0);
6699 if (d
&& (d
->state_fd
>= 0))
6703 * OK, this device needs recovery. Try to re-add the
6704 * previous occupant of this slot, if this fails see if
6705 * we can continue the assimilation of a spare that was
6706 * partially assimilated, finally try to activate a new
6709 dl
= imsm_readd(super
, i
, a
);
6711 dl
= imsm_add_spare(super
, i
, a
, 0, rv
);
6713 dl
= imsm_add_spare(super
, i
, a
, 1, rv
);
6717 /* found a usable disk with enough space */
6718 di
= malloc(sizeof(*di
));
6721 memset(di
, 0, sizeof(*di
));
6723 /* dl->index will be -1 in the case we are activating a
6724 * pristine spare. imsm_process_update() will create a
6725 * new index in this case. Once a disk is found to be
6726 * failed in all member arrays it is kicked from the
6729 di
->disk
.number
= dl
->index
;
6731 /* (ab)use di->devs to store a pointer to the device
6734 di
->devs
= (struct mdinfo
*) dl
;
6736 di
->disk
.raid_disk
= i
;
6737 di
->disk
.major
= dl
->major
;
6738 di
->disk
.minor
= dl
->minor
;
6740 di
->recovery_start
= 0;
6741 di
->data_offset
= __le32_to_cpu(map
->pba_of_lba0
);
6742 di
->component_size
= a
->info
.component_size
;
6743 di
->container_member
= inst
;
6744 super
->random
= random32();
6748 dprintf("%x:%x to be %d at %llu\n", dl
->major
, dl
->minor
,
6749 i
, di
->data_offset
);
6753 /* No spares found */
6755 /* Now 'rv' has a list of devices to return.
6756 * Create a metadata_update record to update the
6757 * disk_ord_tbl for the array
6759 mu
= malloc(sizeof(*mu
));
6761 mu
->buf
= malloc(sizeof(struct imsm_update_activate_spare
) * num_spares
);
6762 if (mu
->buf
== NULL
) {
6769 struct mdinfo
*n
= rv
->next
;
6778 mu
->space_list
= NULL
;
6779 mu
->len
= sizeof(struct imsm_update_activate_spare
) * num_spares
;
6780 mu
->next
= *updates
;
6781 u
= (struct imsm_update_activate_spare
*) mu
->buf
;
6783 for (di
= rv
; di
; di
= di
->next
) {
6784 u
->type
= update_activate_spare
;
6785 u
->dl
= (struct dl
*) di
->devs
;
6787 u
->slot
= di
->disk
.raid_disk
;
6798 static int disks_overlap(struct intel_super
*super
, int idx
, struct imsm_update_create_array
*u
)
6800 struct imsm_dev
*dev
= get_imsm_dev(super
, idx
);
6801 struct imsm_map
*map
= get_imsm_map(dev
, 0);
6802 struct imsm_map
*new_map
= get_imsm_map(&u
->dev
, 0);
6803 struct disk_info
*inf
= get_disk_info(u
);
6804 struct imsm_disk
*disk
;
6808 for (i
= 0; i
< map
->num_members
; i
++) {
6809 disk
= get_imsm_disk(super
, get_imsm_disk_idx(dev
, i
, -1));
6810 for (j
= 0; j
< new_map
->num_members
; j
++)
6811 if (serialcmp(disk
->serial
, inf
[j
].serial
) == 0)
6819 static struct dl
*get_disk_super(struct intel_super
*super
, int major
, int minor
)
6821 struct dl
*dl
= NULL
;
6822 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
6823 if ((dl
->major
== major
) && (dl
->minor
== minor
))
6828 static int remove_disk_super(struct intel_super
*super
, int major
, int minor
)
6830 struct dl
*prev
= NULL
;
6834 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
6835 if ((dl
->major
== major
) && (dl
->minor
== minor
)) {
6838 prev
->next
= dl
->next
;
6840 super
->disks
= dl
->next
;
6842 __free_imsm_disk(dl
);
6843 dprintf("%s: removed %x:%x\n",
6844 __func__
, major
, minor
);
6852 static void imsm_delete(struct intel_super
*super
, struct dl
**dlp
, unsigned index
);
6854 static int add_remove_disk_update(struct intel_super
*super
)
6856 int check_degraded
= 0;
6857 struct dl
*disk
= NULL
;
6858 /* add/remove some spares to/from the metadata/contrainer */
6859 while (super
->disk_mgmt_list
) {
6860 struct dl
*disk_cfg
;
6862 disk_cfg
= super
->disk_mgmt_list
;
6863 super
->disk_mgmt_list
= disk_cfg
->next
;
6864 disk_cfg
->next
= NULL
;
6866 if (disk_cfg
->action
== DISK_ADD
) {
6867 disk_cfg
->next
= super
->disks
;
6868 super
->disks
= disk_cfg
;
6870 dprintf("%s: added %x:%x\n",
6871 __func__
, disk_cfg
->major
,
6873 } else if (disk_cfg
->action
== DISK_REMOVE
) {
6874 dprintf("Disk remove action processed: %x.%x\n",
6875 disk_cfg
->major
, disk_cfg
->minor
);
6876 disk
= get_disk_super(super
,
6880 /* store action status */
6881 disk
->action
= DISK_REMOVE
;
6882 /* remove spare disks only */
6883 if (disk
->index
== -1) {
6884 remove_disk_super(super
,
6889 /* release allocate disk structure */
6890 __free_imsm_disk(disk_cfg
);
6893 return check_degraded
;
6897 static int apply_reshape_migration_update(struct imsm_update_reshape_migration
*u
,
6898 struct intel_super
*super
,
6901 struct intel_dev
*id
;
6902 void **tofree
= NULL
;
6905 dprintf("apply_reshape_migration_update()\n");
6906 if ((u
->subdev
< 0) ||
6908 dprintf("imsm: Error: Wrong subdev: %i\n", u
->subdev
);
6911 if ((space_list
== NULL
) || (*space_list
== NULL
)) {
6912 dprintf("imsm: Error: Memory is not allocated\n");
6916 for (id
= super
->devlist
; id
; id
= id
->next
) {
6917 if (id
->index
== (unsigned)u
->subdev
) {
6918 struct imsm_dev
*dev
= get_imsm_dev(super
, u
->subdev
);
6919 struct imsm_map
*map
;
6920 struct imsm_dev
*new_dev
=
6921 (struct imsm_dev
*)*space_list
;
6922 struct imsm_map
*migr_map
= get_imsm_map(dev
, 1);
6924 struct dl
*new_disk
;
6926 if (new_dev
== NULL
)
6928 *space_list
= **space_list
;
6929 memcpy(new_dev
, dev
, sizeof_imsm_dev(dev
, 0));
6930 map
= get_imsm_map(new_dev
, 0);
6932 dprintf("imsm: Error: migration in progress");
6936 to_state
= map
->map_state
;
6937 if ((u
->new_level
== 5) && (map
->raid_level
== 0)) {
6939 /* this should not happen */
6940 if (u
->new_disks
[0] < 0) {
6941 map
->failed_disk_num
=
6942 map
->num_members
- 1;
6943 to_state
= IMSM_T_STATE_DEGRADED
;
6945 to_state
= IMSM_T_STATE_NORMAL
;
6947 migrate(new_dev
, super
, to_state
, MIGR_GEN_MIGR
);
6948 if (u
->new_level
> -1)
6949 map
->raid_level
= u
->new_level
;
6950 migr_map
= get_imsm_map(new_dev
, 1);
6951 if ((u
->new_level
== 5) &&
6952 (migr_map
->raid_level
== 0)) {
6953 int ord
= map
->num_members
- 1;
6954 migr_map
->num_members
--;
6955 if (u
->new_disks
[0] < 0)
6956 ord
|= IMSM_ORD_REBUILD
;
6957 set_imsm_ord_tbl_ent(map
,
6958 map
->num_members
- 1,
6962 tofree
= (void **)dev
;
6964 /* update chunk size
6966 if (u
->new_chunksize
> 0)
6967 map
->blocks_per_strip
=
6968 __cpu_to_le16(u
->new_chunksize
* 2);
6972 if ((u
->new_level
!= 5) ||
6973 (migr_map
->raid_level
!= 0) ||
6974 (migr_map
->raid_level
== map
->raid_level
))
6977 if (u
->new_disks
[0] >= 0) {
6980 new_disk
= get_disk_super(super
,
6981 major(u
->new_disks
[0]),
6982 minor(u
->new_disks
[0]));
6983 dprintf("imsm: new disk for reshape is: %i:%i "
6984 "(%p, index = %i)\n",
6985 major(u
->new_disks
[0]),
6986 minor(u
->new_disks
[0]),
6987 new_disk
, new_disk
->index
);
6988 if (new_disk
== NULL
)
6989 goto error_disk_add
;
6991 new_disk
->index
= map
->num_members
- 1;
6992 /* slot to fill in autolayout
6994 new_disk
->raiddisk
= new_disk
->index
;
6995 new_disk
->disk
.status
|= CONFIGURED_DISK
;
6996 new_disk
->disk
.status
&= ~SPARE_DISK
;
6998 goto error_disk_add
;
7001 *tofree
= *space_list
;
7002 /* calculate new size
7004 imsm_set_array_size(new_dev
);
7011 *space_list
= tofree
;
7015 dprintf("Error: imsm: Cannot find disk.\n");
7019 static int apply_update_activate_spare(struct imsm_update_activate_spare
*u
,
7020 struct intel_super
*super
,
7021 struct active_array
*active_array
)
7023 struct imsm_super
*mpb
= super
->anchor
;
7024 struct imsm_dev
*dev
= get_imsm_dev(super
, u
->array
);
7025 struct imsm_map
*map
= get_imsm_map(dev
, 0);
7026 struct imsm_map
*migr_map
;
7027 struct active_array
*a
;
7028 struct imsm_disk
*disk
;
7035 int second_map_created
= 0;
7037 for (; u
; u
= u
->next
) {
7038 victim
= get_imsm_disk_idx(dev
, u
->slot
, -1);
7043 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
7048 fprintf(stderr
, "error: imsm_activate_spare passed "
7049 "an unknown disk (index: %d)\n",
7054 /* count failures (excluding rebuilds and the victim)
7055 * to determine map[0] state
7058 for (i
= 0; i
< map
->num_members
; i
++) {
7061 disk
= get_imsm_disk(super
,
7062 get_imsm_disk_idx(dev
, i
, -1));
7063 if (!disk
|| is_failed(disk
))
7067 /* adding a pristine spare, assign a new index */
7068 if (dl
->index
< 0) {
7069 dl
->index
= super
->anchor
->num_disks
;
7070 super
->anchor
->num_disks
++;
7073 disk
->status
|= CONFIGURED_DISK
;
7074 disk
->status
&= ~SPARE_DISK
;
7077 to_state
= imsm_check_degraded(super
, dev
, failed
);
7078 if (!second_map_created
) {
7079 second_map_created
= 1;
7080 map
->map_state
= IMSM_T_STATE_DEGRADED
;
7081 migrate(dev
, super
, to_state
, MIGR_REBUILD
);
7083 map
->map_state
= to_state
;
7084 migr_map
= get_imsm_map(dev
, 1);
7085 set_imsm_ord_tbl_ent(map
, u
->slot
, dl
->index
);
7086 set_imsm_ord_tbl_ent(migr_map
, u
->slot
,
7087 dl
->index
| IMSM_ORD_REBUILD
);
7089 /* update the family_num to mark a new container
7090 * generation, being careful to record the existing
7091 * family_num in orig_family_num to clean up after
7092 * earlier mdadm versions that neglected to set it.
7094 if (mpb
->orig_family_num
== 0)
7095 mpb
->orig_family_num
= mpb
->family_num
;
7096 mpb
->family_num
+= super
->random
;
7098 /* count arrays using the victim in the metadata */
7100 for (a
= active_array
; a
; a
= a
->next
) {
7101 dev
= get_imsm_dev(super
, a
->info
.container_member
);
7102 map
= get_imsm_map(dev
, 0);
7104 if (get_imsm_disk_slot(map
, victim
) >= 0)
7108 /* delete the victim if it is no longer being
7114 /* We know that 'manager' isn't touching anything,
7115 * so it is safe to delete
7117 for (dlp
= &super
->disks
; *dlp
; dlp
= &(*dlp
)->next
)
7118 if ((*dlp
)->index
== victim
)
7121 /* victim may be on the missing list */
7123 for (dlp
= &super
->missing
; *dlp
;
7124 dlp
= &(*dlp
)->next
)
7125 if ((*dlp
)->index
== victim
)
7127 imsm_delete(super
, dlp
, victim
);
7134 static int apply_reshape_container_disks_update(struct imsm_update_reshape
*u
,
7135 struct intel_super
*super
,
7138 struct dl
*new_disk
;
7139 struct intel_dev
*id
;
7141 int delta_disks
= u
->new_raid_disks
- u
->old_raid_disks
;
7142 int disk_count
= u
->old_raid_disks
;
7143 void **tofree
= NULL
;
7144 int devices_to_reshape
= 1;
7145 struct imsm_super
*mpb
= super
->anchor
;
7147 unsigned int dev_id
;
7149 dprintf("imsm: apply_reshape_container_disks_update()\n");
7151 /* enable spares to use in array */
7152 for (i
= 0; i
< delta_disks
; i
++) {
7153 new_disk
= get_disk_super(super
,
7154 major(u
->new_disks
[i
]),
7155 minor(u
->new_disks
[i
]));
7156 dprintf("imsm: new disk for reshape is: %i:%i "
7157 "(%p, index = %i)\n",
7158 major(u
->new_disks
[i
]), minor(u
->new_disks
[i
]),
7159 new_disk
, new_disk
->index
);
7160 if ((new_disk
== NULL
) ||
7161 ((new_disk
->index
>= 0) &&
7162 (new_disk
->index
< u
->old_raid_disks
)))
7163 goto update_reshape_exit
;
7164 new_disk
->index
= disk_count
++;
7165 /* slot to fill in autolayout
7167 new_disk
->raiddisk
= new_disk
->index
;
7168 new_disk
->disk
.status
|=
7170 new_disk
->disk
.status
&= ~SPARE_DISK
;
7173 dprintf("imsm: volume set mpb->num_raid_devs = %i\n",
7174 mpb
->num_raid_devs
);
7175 /* manage changes in volume
7177 for (dev_id
= 0; dev_id
< mpb
->num_raid_devs
; dev_id
++) {
7178 void **sp
= *space_list
;
7179 struct imsm_dev
*newdev
;
7180 struct imsm_map
*newmap
, *oldmap
;
7182 for (id
= super
->devlist
; id
; id
= id
->next
) {
7183 if (id
->index
== dev_id
)
7192 /* Copy the dev, but not (all of) the map */
7193 memcpy(newdev
, id
->dev
, sizeof(*newdev
));
7194 oldmap
= get_imsm_map(id
->dev
, 0);
7195 newmap
= get_imsm_map(newdev
, 0);
7196 /* Copy the current map */
7197 memcpy(newmap
, oldmap
, sizeof_imsm_map(oldmap
));
7198 /* update one device only
7200 if (devices_to_reshape
) {
7201 dprintf("imsm: modifying subdev: %i\n",
7203 devices_to_reshape
--;
7204 newdev
->vol
.migr_state
= 1;
7205 newdev
->vol
.curr_migr_unit
= 0;
7206 set_migr_type(newdev
, MIGR_GEN_MIGR
);
7207 newmap
->num_members
= u
->new_raid_disks
;
7208 for (i
= 0; i
< delta_disks
; i
++) {
7209 set_imsm_ord_tbl_ent(newmap
,
7210 u
->old_raid_disks
+ i
,
7211 u
->old_raid_disks
+ i
);
7213 /* New map is correct, now need to save old map
7215 newmap
= get_imsm_map(newdev
, 1);
7216 memcpy(newmap
, oldmap
, sizeof_imsm_map(oldmap
));
7218 imsm_set_array_size(newdev
);
7221 sp
= (void **)id
->dev
;
7226 /* Clear migration record */
7227 memset(super
->migr_rec
, 0, sizeof(struct migr_record
));
7230 *space_list
= tofree
;
7233 update_reshape_exit
:
7238 static int apply_takeover_update(struct imsm_update_takeover
*u
,
7239 struct intel_super
*super
,
7242 struct imsm_dev
*dev
= NULL
;
7243 struct intel_dev
*dv
;
7244 struct imsm_dev
*dev_new
;
7245 struct imsm_map
*map
;
7249 for (dv
= super
->devlist
; dv
; dv
= dv
->next
)
7250 if (dv
->index
== (unsigned int)u
->subarray
) {
7258 map
= get_imsm_map(dev
, 0);
7260 if (u
->direction
== R10_TO_R0
) {
7261 /* Number of failed disks must be half of initial disk number */
7262 if (imsm_count_failed(super
, dev
) != (map
->num_members
/ 2))
7265 /* iterate through devices to mark removed disks as spare */
7266 for (dm
= super
->disks
; dm
; dm
= dm
->next
) {
7267 if (dm
->disk
.status
& FAILED_DISK
) {
7268 int idx
= dm
->index
;
7269 /* update indexes on the disk list */
7270 /* FIXME this loop-with-the-loop looks wrong, I'm not convinced
7271 the index values will end up being correct.... NB */
7272 for (du
= super
->disks
; du
; du
= du
->next
)
7273 if (du
->index
> idx
)
7275 /* mark as spare disk */
7280 map
->num_members
= map
->num_members
/ 2;
7281 map
->map_state
= IMSM_T_STATE_NORMAL
;
7282 map
->num_domains
= 1;
7283 map
->raid_level
= 0;
7284 map
->failed_disk_num
= -1;
7287 if (u
->direction
== R0_TO_R10
) {
7289 /* update slots in current disk list */
7290 for (dm
= super
->disks
; dm
; dm
= dm
->next
) {
7294 /* create new *missing* disks */
7295 for (i
= 0; i
< map
->num_members
; i
++) {
7296 space
= *space_list
;
7299 *space_list
= *space
;
7301 memcpy(du
, super
->disks
, sizeof(*du
));
7305 du
->index
= (i
* 2) + 1;
7306 sprintf((char *)du
->disk
.serial
,
7307 " MISSING_%d", du
->index
);
7308 sprintf((char *)du
->serial
,
7309 "MISSING_%d", du
->index
);
7310 du
->next
= super
->missing
;
7311 super
->missing
= du
;
7313 /* create new dev and map */
7314 space
= *space_list
;
7317 *space_list
= *space
;
7318 dev_new
= (void *)space
;
7319 memcpy(dev_new
, dev
, sizeof(*dev
));
7320 /* update new map */
7321 map
= get_imsm_map(dev_new
, 0);
7322 map
->num_members
= map
->num_members
* 2;
7323 map
->map_state
= IMSM_T_STATE_DEGRADED
;
7324 map
->num_domains
= 2;
7325 map
->raid_level
= 1;
7326 /* replace dev<->dev_new */
7329 /* update disk order table */
7330 for (du
= super
->disks
; du
; du
= du
->next
)
7332 set_imsm_ord_tbl_ent(map
, du
->index
, du
->index
);
7333 for (du
= super
->missing
; du
; du
= du
->next
)
7334 if (du
->index
>= 0) {
7335 set_imsm_ord_tbl_ent(map
, du
->index
, du
->index
);
7336 mark_missing(dv
->dev
, &du
->disk
, du
->index
);
7342 static void imsm_process_update(struct supertype
*st
,
7343 struct metadata_update
*update
)
7346 * crack open the metadata_update envelope to find the update record
7347 * update can be one of:
7348 * update_reshape_container_disks - all the arrays in the container
7349 * are being reshaped to have more devices. We need to mark
7350 * the arrays for general migration and convert selected spares
7351 * into active devices.
7352 * update_activate_spare - a spare device has replaced a failed
7353 * device in an array, update the disk_ord_tbl. If this disk is
7354 * present in all member arrays then also clear the SPARE_DISK
7356 * update_create_array
7358 * update_rename_array
7359 * update_add_remove_disk
7361 struct intel_super
*super
= st
->sb
;
7362 struct imsm_super
*mpb
;
7363 enum imsm_update_type type
= *(enum imsm_update_type
*) update
->buf
;
7365 /* update requires a larger buf but the allocation failed */
7366 if (super
->next_len
&& !super
->next_buf
) {
7367 super
->next_len
= 0;
7371 if (super
->next_buf
) {
7372 memcpy(super
->next_buf
, super
->buf
, super
->len
);
7374 super
->len
= super
->next_len
;
7375 super
->buf
= super
->next_buf
;
7377 super
->next_len
= 0;
7378 super
->next_buf
= NULL
;
7381 mpb
= super
->anchor
;
7384 case update_general_migration_checkpoint
: {
7385 struct intel_dev
*id
;
7386 struct imsm_update_general_migration_checkpoint
*u
=
7387 (void *)update
->buf
;
7389 dprintf("imsm: process_update() "
7390 "for update_general_migration_checkpoint called\n");
7392 /* find device under general migration */
7393 for (id
= super
->devlist
; id
; id
= id
->next
) {
7394 if (is_gen_migration(id
->dev
)) {
7395 id
->dev
->vol
.curr_migr_unit
=
7396 __cpu_to_le32(u
->curr_migr_unit
);
7397 super
->updates_pending
++;
7402 case update_takeover
: {
7403 struct imsm_update_takeover
*u
= (void *)update
->buf
;
7404 if (apply_takeover_update(u
, super
, &update
->space_list
)) {
7405 imsm_update_version_info(super
);
7406 super
->updates_pending
++;
7411 case update_reshape_container_disks
: {
7412 struct imsm_update_reshape
*u
= (void *)update
->buf
;
7413 if (apply_reshape_container_disks_update(
7414 u
, super
, &update
->space_list
))
7415 super
->updates_pending
++;
7418 case update_reshape_migration
: {
7419 struct imsm_update_reshape_migration
*u
= (void *)update
->buf
;
7420 if (apply_reshape_migration_update(
7421 u
, super
, &update
->space_list
))
7422 super
->updates_pending
++;
7425 case update_activate_spare
: {
7426 struct imsm_update_activate_spare
*u
= (void *) update
->buf
;
7427 if (apply_update_activate_spare(u
, super
, st
->arrays
))
7428 super
->updates_pending
++;
7431 case update_create_array
: {
7432 /* someone wants to create a new array, we need to be aware of
7433 * a few races/collisions:
7434 * 1/ 'Create' called by two separate instances of mdadm
7435 * 2/ 'Create' versus 'activate_spare': mdadm has chosen
7436 * devices that have since been assimilated via
7438 * In the event this update can not be carried out mdadm will
7439 * (FIX ME) notice that its update did not take hold.
7441 struct imsm_update_create_array
*u
= (void *) update
->buf
;
7442 struct intel_dev
*dv
;
7443 struct imsm_dev
*dev
;
7444 struct imsm_map
*map
, *new_map
;
7445 unsigned long long start
, end
;
7446 unsigned long long new_start
, new_end
;
7448 struct disk_info
*inf
;
7451 /* handle racing creates: first come first serve */
7452 if (u
->dev_idx
< mpb
->num_raid_devs
) {
7453 dprintf("%s: subarray %d already defined\n",
7454 __func__
, u
->dev_idx
);
7458 /* check update is next in sequence */
7459 if (u
->dev_idx
!= mpb
->num_raid_devs
) {
7460 dprintf("%s: can not create array %d expected index %d\n",
7461 __func__
, u
->dev_idx
, mpb
->num_raid_devs
);
7465 new_map
= get_imsm_map(&u
->dev
, 0);
7466 new_start
= __le32_to_cpu(new_map
->pba_of_lba0
);
7467 new_end
= new_start
+ __le32_to_cpu(new_map
->blocks_per_member
);
7468 inf
= get_disk_info(u
);
7470 /* handle activate_spare versus create race:
7471 * check to make sure that overlapping arrays do not include
7474 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
7475 dev
= get_imsm_dev(super
, i
);
7476 map
= get_imsm_map(dev
, 0);
7477 start
= __le32_to_cpu(map
->pba_of_lba0
);
7478 end
= start
+ __le32_to_cpu(map
->blocks_per_member
);
7479 if ((new_start
>= start
&& new_start
<= end
) ||
7480 (start
>= new_start
&& start
<= new_end
))
7485 if (disks_overlap(super
, i
, u
)) {
7486 dprintf("%s: arrays overlap\n", __func__
);
7491 /* check that prepare update was successful */
7492 if (!update
->space
) {
7493 dprintf("%s: prepare update failed\n", __func__
);
7497 /* check that all disks are still active before committing
7498 * changes. FIXME: could we instead handle this by creating a
7499 * degraded array? That's probably not what the user expects,
7500 * so better to drop this update on the floor.
7502 for (i
= 0; i
< new_map
->num_members
; i
++) {
7503 dl
= serial_to_dl(inf
[i
].serial
, super
);
7505 dprintf("%s: disk disappeared\n", __func__
);
7510 super
->updates_pending
++;
7512 /* convert spares to members and fixup ord_tbl */
7513 for (i
= 0; i
< new_map
->num_members
; i
++) {
7514 dl
= serial_to_dl(inf
[i
].serial
, super
);
7515 if (dl
->index
== -1) {
7516 dl
->index
= mpb
->num_disks
;
7518 dl
->disk
.status
|= CONFIGURED_DISK
;
7519 dl
->disk
.status
&= ~SPARE_DISK
;
7521 set_imsm_ord_tbl_ent(new_map
, i
, dl
->index
);
7526 update
->space
= NULL
;
7527 imsm_copy_dev(dev
, &u
->dev
);
7528 dv
->index
= u
->dev_idx
;
7529 dv
->next
= super
->devlist
;
7530 super
->devlist
= dv
;
7531 mpb
->num_raid_devs
++;
7533 imsm_update_version_info(super
);
7536 /* mdmon knows how to release update->space, but not
7537 * ((struct intel_dev *) update->space)->dev
7539 if (update
->space
) {
7545 case update_kill_array
: {
7546 struct imsm_update_kill_array
*u
= (void *) update
->buf
;
7547 int victim
= u
->dev_idx
;
7548 struct active_array
*a
;
7549 struct intel_dev
**dp
;
7550 struct imsm_dev
*dev
;
7552 /* sanity check that we are not affecting the uuid of
7553 * active arrays, or deleting an active array
7555 * FIXME when immutable ids are available, but note that
7556 * we'll also need to fixup the invalidated/active
7557 * subarray indexes in mdstat
7559 for (a
= st
->arrays
; a
; a
= a
->next
)
7560 if (a
->info
.container_member
>= victim
)
7562 /* by definition if mdmon is running at least one array
7563 * is active in the container, so checking
7564 * mpb->num_raid_devs is just extra paranoia
7566 dev
= get_imsm_dev(super
, victim
);
7567 if (a
|| !dev
|| mpb
->num_raid_devs
== 1) {
7568 dprintf("failed to delete subarray-%d\n", victim
);
7572 for (dp
= &super
->devlist
; *dp
;)
7573 if ((*dp
)->index
== (unsigned)super
->current_vol
) {
7576 if ((*dp
)->index
> (unsigned)victim
)
7580 mpb
->num_raid_devs
--;
7581 super
->updates_pending
++;
7584 case update_rename_array
: {
7585 struct imsm_update_rename_array
*u
= (void *) update
->buf
;
7586 char name
[MAX_RAID_SERIAL_LEN
+1];
7587 int target
= u
->dev_idx
;
7588 struct active_array
*a
;
7589 struct imsm_dev
*dev
;
7591 /* sanity check that we are not affecting the uuid of
7594 snprintf(name
, MAX_RAID_SERIAL_LEN
, "%s", (char *) u
->name
);
7595 name
[MAX_RAID_SERIAL_LEN
] = '\0';
7596 for (a
= st
->arrays
; a
; a
= a
->next
)
7597 if (a
->info
.container_member
== target
)
7599 dev
= get_imsm_dev(super
, u
->dev_idx
);
7600 if (a
|| !dev
|| !check_name(super
, name
, 1)) {
7601 dprintf("failed to rename subarray-%d\n", target
);
7605 snprintf((char *) dev
->volume
, MAX_RAID_SERIAL_LEN
, "%s", name
);
7606 super
->updates_pending
++;
7609 case update_add_remove_disk
: {
7610 /* we may be able to repair some arrays if disks are
7611 * being added, check teh status of add_remove_disk
7612 * if discs has been added.
7614 if (add_remove_disk_update(super
)) {
7615 struct active_array
*a
;
7617 super
->updates_pending
++;
7618 for (a
= st
->arrays
; a
; a
= a
->next
)
7619 a
->check_degraded
= 1;
7624 fprintf(stderr
, "error: unsuported process update type:"
7625 "(type: %d)\n", type
);
7629 static struct mdinfo
*get_spares_for_grow(struct supertype
*st
);
7631 static void imsm_prepare_update(struct supertype
*st
,
7632 struct metadata_update
*update
)
7635 * Allocate space to hold new disk entries, raid-device entries or a new
7636 * mpb if necessary. The manager synchronously waits for updates to
7637 * complete in the monitor, so new mpb buffers allocated here can be
7638 * integrated by the monitor thread without worrying about live pointers
7639 * in the manager thread.
7641 enum imsm_update_type type
= *(enum imsm_update_type
*) update
->buf
;
7642 struct intel_super
*super
= st
->sb
;
7643 struct imsm_super
*mpb
= super
->anchor
;
7648 case update_general_migration_checkpoint
:
7649 dprintf("imsm: prepare_update() "
7650 "for update_general_migration_checkpoint called\n");
7652 case update_takeover
: {
7653 struct imsm_update_takeover
*u
= (void *)update
->buf
;
7654 if (u
->direction
== R0_TO_R10
) {
7655 void **tail
= (void **)&update
->space_list
;
7656 struct imsm_dev
*dev
= get_imsm_dev(super
, u
->subarray
);
7657 struct imsm_map
*map
= get_imsm_map(dev
, 0);
7658 int num_members
= map
->num_members
;
7662 /* allocate memory for added disks */
7663 for (i
= 0; i
< num_members
; i
++) {
7664 size
= sizeof(struct dl
);
7665 space
= malloc(size
);
7674 /* allocate memory for new device */
7675 size
= sizeof_imsm_dev(super
->devlist
->dev
, 0) +
7676 (num_members
* sizeof(__u32
));
7677 space
= malloc(size
);
7686 len
= disks_to_mpb_size(num_members
* 2);
7688 /* if allocation didn't success, free buffer */
7689 while (update
->space_list
) {
7690 void **sp
= update
->space_list
;
7691 update
->space_list
= *sp
;
7699 case update_reshape_container_disks
: {
7700 /* Every raid device in the container is about to
7701 * gain some more devices, and we will enter a
7703 * So each 'imsm_map' will be bigger, and the imsm_vol
7704 * will now hold 2 of them.
7705 * Thus we need new 'struct imsm_dev' allocations sized
7706 * as sizeof_imsm_dev but with more devices in both maps.
7708 struct imsm_update_reshape
*u
= (void *)update
->buf
;
7709 struct intel_dev
*dl
;
7710 void **space_tail
= (void**)&update
->space_list
;
7712 dprintf("imsm: imsm_prepare_update() for update_reshape\n");
7714 for (dl
= super
->devlist
; dl
; dl
= dl
->next
) {
7715 int size
= sizeof_imsm_dev(dl
->dev
, 1);
7717 if (u
->new_raid_disks
> u
->old_raid_disks
)
7718 size
+= sizeof(__u32
)*2*
7719 (u
->new_raid_disks
- u
->old_raid_disks
);
7728 len
= disks_to_mpb_size(u
->new_raid_disks
);
7729 dprintf("New anchor length is %llu\n", (unsigned long long)len
);
7732 case update_reshape_migration
: {
7733 /* for migration level 0->5 we need to add disks
7734 * so the same as for container operation we will copy
7735 * device to the bigger location.
7736 * in memory prepared device and new disk area are prepared
7737 * for usage in process update
7739 struct imsm_update_reshape_migration
*u
= (void *)update
->buf
;
7740 struct intel_dev
*id
;
7741 void **space_tail
= (void **)&update
->space_list
;
7744 int current_level
= -1;
7746 dprintf("imsm: imsm_prepare_update() for update_reshape\n");
7748 /* add space for bigger array in update
7750 for (id
= super
->devlist
; id
; id
= id
->next
) {
7751 if (id
->index
== (unsigned)u
->subdev
) {
7752 size
= sizeof_imsm_dev(id
->dev
, 1);
7753 if (u
->new_raid_disks
> u
->old_raid_disks
)
7754 size
+= sizeof(__u32
)*2*
7755 (u
->new_raid_disks
- u
->old_raid_disks
);
7765 if (update
->space_list
== NULL
)
7768 /* add space for disk in update
7770 size
= sizeof(struct dl
);
7773 free(update
->space_list
);
7774 update
->space_list
= NULL
;
7781 /* add spare device to update
7783 for (id
= super
->devlist
; id
; id
= id
->next
)
7784 if (id
->index
== (unsigned)u
->subdev
) {
7785 struct imsm_dev
*dev
;
7786 struct imsm_map
*map
;
7788 dev
= get_imsm_dev(super
, u
->subdev
);
7789 map
= get_imsm_map(dev
, 0);
7790 current_level
= map
->raid_level
;
7793 if ((u
->new_level
== 5) && (u
->new_level
!= current_level
)) {
7794 struct mdinfo
*spares
;
7796 spares
= get_spares_for_grow(st
);
7804 makedev(dev
->disk
.major
,
7806 dl
= get_disk_super(super
,
7809 dl
->index
= u
->old_raid_disks
;
7815 len
= disks_to_mpb_size(u
->new_raid_disks
);
7816 dprintf("New anchor length is %llu\n", (unsigned long long)len
);
7819 case update_create_array
: {
7820 struct imsm_update_create_array
*u
= (void *) update
->buf
;
7821 struct intel_dev
*dv
;
7822 struct imsm_dev
*dev
= &u
->dev
;
7823 struct imsm_map
*map
= get_imsm_map(dev
, 0);
7825 struct disk_info
*inf
;
7829 inf
= get_disk_info(u
);
7830 len
= sizeof_imsm_dev(dev
, 1);
7831 /* allocate a new super->devlist entry */
7832 dv
= malloc(sizeof(*dv
));
7834 dv
->dev
= malloc(len
);
7839 update
->space
= NULL
;
7843 /* count how many spares will be converted to members */
7844 for (i
= 0; i
< map
->num_members
; i
++) {
7845 dl
= serial_to_dl(inf
[i
].serial
, super
);
7847 /* hmm maybe it failed?, nothing we can do about
7852 if (count_memberships(dl
, super
) == 0)
7855 len
+= activate
* sizeof(struct imsm_disk
);
7862 /* check if we need a larger metadata buffer */
7863 if (super
->next_buf
)
7864 buf_len
= super
->next_len
;
7866 buf_len
= super
->len
;
7868 if (__le32_to_cpu(mpb
->mpb_size
) + len
> buf_len
) {
7869 /* ok we need a larger buf than what is currently allocated
7870 * if this allocation fails process_update will notice that
7871 * ->next_len is set and ->next_buf is NULL
7873 buf_len
= ROUND_UP(__le32_to_cpu(mpb
->mpb_size
) + len
, 512);
7874 if (super
->next_buf
)
7875 free(super
->next_buf
);
7877 super
->next_len
= buf_len
;
7878 if (posix_memalign(&super
->next_buf
, 512, buf_len
) == 0)
7879 memset(super
->next_buf
, 0, buf_len
);
7881 super
->next_buf
= NULL
;
7885 /* must be called while manager is quiesced */
7886 static void imsm_delete(struct intel_super
*super
, struct dl
**dlp
, unsigned index
)
7888 struct imsm_super
*mpb
= super
->anchor
;
7890 struct imsm_dev
*dev
;
7891 struct imsm_map
*map
;
7892 int i
, j
, num_members
;
7895 dprintf("%s: deleting device[%d] from imsm_super\n",
7898 /* shift all indexes down one */
7899 for (iter
= super
->disks
; iter
; iter
= iter
->next
)
7900 if (iter
->index
> (int)index
)
7902 for (iter
= super
->missing
; iter
; iter
= iter
->next
)
7903 if (iter
->index
> (int)index
)
7906 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
7907 dev
= get_imsm_dev(super
, i
);
7908 map
= get_imsm_map(dev
, 0);
7909 num_members
= map
->num_members
;
7910 for (j
= 0; j
< num_members
; j
++) {
7911 /* update ord entries being careful not to propagate
7912 * ord-flags to the first map
7914 ord
= get_imsm_ord_tbl_ent(dev
, j
, -1);
7916 if (ord_to_idx(ord
) <= index
)
7919 map
= get_imsm_map(dev
, 0);
7920 set_imsm_ord_tbl_ent(map
, j
, ord_to_idx(ord
- 1));
7921 map
= get_imsm_map(dev
, 1);
7923 set_imsm_ord_tbl_ent(map
, j
, ord
- 1);
7928 super
->updates_pending
++;
7930 struct dl
*dl
= *dlp
;
7932 *dlp
= (*dlp
)->next
;
7933 __free_imsm_disk(dl
);
7936 #endif /* MDASSEMBLE */
7937 /*******************************************************************************
7938 * Function: open_backup_targets
7939 * Description: Function opens file descriptors for all devices given in
7942 * info : general array info
7943 * raid_disks : number of disks
7944 * raid_fds : table of device's file descriptors
7948 ******************************************************************************/
7949 int open_backup_targets(struct mdinfo
*info
, int raid_disks
, int *raid_fds
)
7953 for (sd
= info
->devs
; sd
; sd
= sd
->next
) {
7956 if (sd
->disk
.state
& (1<<MD_DISK_FAULTY
)) {
7957 dprintf("disk is faulty!!\n");
7961 if ((sd
->disk
.raid_disk
>= raid_disks
) ||
7962 (sd
->disk
.raid_disk
< 0))
7965 dn
= map_dev(sd
->disk
.major
,
7967 raid_fds
[sd
->disk
.raid_disk
] = dev_open(dn
, O_RDWR
);
7968 if (raid_fds
[sd
->disk
.raid_disk
] < 0) {
7969 fprintf(stderr
, "cannot open component\n");
7977 /*******************************************************************************
7978 * Function: init_migr_record_imsm
7979 * Description: Function inits imsm migration record
7981 * super : imsm internal array info
7982 * dev : device under migration
7983 * info : general array info to find the smallest device
7986 ******************************************************************************/
7987 void init_migr_record_imsm(struct supertype
*st
, struct imsm_dev
*dev
,
7988 struct mdinfo
*info
)
7990 struct intel_super
*super
= st
->sb
;
7991 struct migr_record
*migr_rec
= super
->migr_rec
;
7993 unsigned long long dsize
, dev_sectors
;
7994 long long unsigned min_dev_sectors
= -1LLU;
7998 struct imsm_map
*map_dest
= get_imsm_map(dev
, 0);
7999 struct imsm_map
*map_src
= get_imsm_map(dev
, 1);
8000 unsigned long long num_migr_units
;
8001 unsigned long long array_blocks
;
8003 memset(migr_rec
, 0, sizeof(struct migr_record
));
8004 migr_rec
->family_num
= __cpu_to_le32(super
->anchor
->family_num
);
8006 /* only ascending reshape supported now */
8007 migr_rec
->ascending_migr
= __cpu_to_le32(1);
8009 migr_rec
->dest_depth_per_unit
= GEN_MIGR_AREA_SIZE
/
8010 max(map_dest
->blocks_per_strip
, map_src
->blocks_per_strip
);
8011 migr_rec
->dest_depth_per_unit
*= map_dest
->blocks_per_strip
;
8012 new_data_disks
= imsm_num_data_members(dev
, 0);
8013 migr_rec
->blocks_per_unit
=
8014 __cpu_to_le32(migr_rec
->dest_depth_per_unit
* new_data_disks
);
8015 migr_rec
->dest_depth_per_unit
=
8016 __cpu_to_le32(migr_rec
->dest_depth_per_unit
);
8017 array_blocks
= info
->component_size
* new_data_disks
;
8019 array_blocks
/ __le32_to_cpu(migr_rec
->blocks_per_unit
);
8021 if (array_blocks
% __le32_to_cpu(migr_rec
->blocks_per_unit
))
8023 migr_rec
->num_migr_units
= __cpu_to_le32(num_migr_units
);
8025 migr_rec
->post_migr_vol_cap
= dev
->size_low
;
8026 migr_rec
->post_migr_vol_cap_hi
= dev
->size_high
;
8029 /* Find the smallest dev */
8030 for (sd
= info
->devs
; sd
; sd
= sd
->next
) {
8031 sprintf(nm
, "%d:%d", sd
->disk
.major
, sd
->disk
.minor
);
8032 fd
= dev_open(nm
, O_RDONLY
);
8035 get_dev_size(fd
, NULL
, &dsize
);
8036 dev_sectors
= dsize
/ 512;
8037 if (dev_sectors
< min_dev_sectors
)
8038 min_dev_sectors
= dev_sectors
;
8041 migr_rec
->ckpt_area_pba
= __cpu_to_le32(min_dev_sectors
-
8042 RAID_DISK_RESERVED_BLOCKS_IMSM_HI
);
8044 write_imsm_migr_rec(st
);
8049 /*******************************************************************************
8050 * Function: save_backup_imsm
8051 * Description: Function saves critical data stripes to Migration Copy Area
8052 * and updates the current migration unit status.
8053 * Use restore_stripes() to form a destination stripe,
8054 * and to write it to the Copy Area.
8056 * st : supertype information
8057 * dev : imsm device that backup is saved for
8058 * info : general array info
8059 * buf : input buffer
8060 * length : length of data to backup (blocks_per_unit)
8064 ******************************************************************************/
8065 int save_backup_imsm(struct supertype
*st
,
8066 struct imsm_dev
*dev
,
8067 struct mdinfo
*info
,
8072 struct intel_super
*super
= st
->sb
;
8073 unsigned long long *target_offsets
= NULL
;
8074 int *targets
= NULL
;
8076 struct imsm_map
*map_dest
= get_imsm_map(dev
, 0);
8077 int new_disks
= map_dest
->num_members
;
8078 int dest_layout
= 0;
8080 unsigned long long start
;
8081 int data_disks
= imsm_num_data_members(dev
, 0);
8083 targets
= malloc(new_disks
* sizeof(int));
8087 for (i
= 0; i
< new_disks
; i
++)
8090 target_offsets
= malloc(new_disks
* sizeof(unsigned long long));
8091 if (!target_offsets
)
8094 start
= info
->reshape_progress
* 512;
8095 for (i
= 0; i
< new_disks
; i
++) {
8096 target_offsets
[i
] = (unsigned long long)
8097 __le32_to_cpu(super
->migr_rec
->ckpt_area_pba
) * 512;
8098 /* move back copy area adderss, it will be moved forward
8099 * in restore_stripes() using start input variable
8101 target_offsets
[i
] -= start
/data_disks
;
8104 if (open_backup_targets(info
, new_disks
, targets
))
8107 dest_layout
= imsm_level_to_layout(map_dest
->raid_level
);
8108 dest_chunk
= __le16_to_cpu(map_dest
->blocks_per_strip
) * 512;
8110 if (restore_stripes(targets
, /* list of dest devices */
8111 target_offsets
, /* migration record offsets */
8114 map_dest
->raid_level
,
8116 -1, /* source backup file descriptor */
8117 0, /* input buf offset
8118 * always 0 buf is already offseted */
8122 fprintf(stderr
, Name
": Error restoring stripes\n");
8130 for (i
= 0; i
< new_disks
; i
++)
8131 if (targets
[i
] >= 0)
8135 free(target_offsets
);
8140 /*******************************************************************************
8141 * Function: save_checkpoint_imsm
8142 * Description: Function called for current unit status update
8143 * in the migration record. It writes it to disk.
8145 * super : imsm internal array info
8146 * info : general array info
8150 * 2: failure, means no valid migration record
8151 * / no general migration in progress /
8152 ******************************************************************************/
8153 int save_checkpoint_imsm(struct supertype
*st
, struct mdinfo
*info
, int state
)
8155 struct intel_super
*super
= st
->sb
;
8156 unsigned long long blocks_per_unit
;
8157 unsigned long long curr_migr_unit
;
8159 if (load_imsm_migr_rec(super
, info
) != 0) {
8160 dprintf("imsm: ERROR: Cannot read migration record "
8161 "for checkpoint save.\n");
8165 blocks_per_unit
= __le32_to_cpu(super
->migr_rec
->blocks_per_unit
);
8166 if (blocks_per_unit
== 0) {
8167 dprintf("imsm: no migration in progress.\n");
8170 curr_migr_unit
= info
->reshape_progress
/ blocks_per_unit
;
8171 /* check if array is alligned to copy area
8172 * if it is not alligned, add one to current migration unit value
8173 * this can happend on array reshape finish only
8175 if (info
->reshape_progress
% blocks_per_unit
)
8178 super
->migr_rec
->curr_migr_unit
=
8179 __cpu_to_le32(curr_migr_unit
);
8180 super
->migr_rec
->rec_status
= __cpu_to_le32(state
);
8181 super
->migr_rec
->dest_1st_member_lba
=
8182 __cpu_to_le32(curr_migr_unit
*
8183 __le32_to_cpu(super
->migr_rec
->dest_depth_per_unit
));
8184 if (write_imsm_migr_rec(st
) < 0) {
8185 dprintf("imsm: Cannot write migration record "
8186 "outside backup area\n");
8193 /*******************************************************************************
8194 * Function: recover_backup_imsm
8195 * Description: Function recovers critical data from the Migration Copy Area
8196 * while assembling an array.
8198 * super : imsm internal array info
8199 * info : general array info
8201 * 0 : success (or there is no data to recover)
8203 ******************************************************************************/
8204 int recover_backup_imsm(struct supertype
*st
, struct mdinfo
*info
)
8206 struct intel_super
*super
= st
->sb
;
8207 struct migr_record
*migr_rec
= super
->migr_rec
;
8208 struct imsm_map
*map_dest
= NULL
;
8209 struct intel_dev
*id
= NULL
;
8210 unsigned long long read_offset
;
8211 unsigned long long write_offset
;
8213 int *targets
= NULL
;
8214 int new_disks
, i
, err
;
8217 unsigned long curr_migr_unit
= __le32_to_cpu(migr_rec
->curr_migr_unit
);
8218 unsigned long num_migr_units
= __le32_to_cpu(migr_rec
->num_migr_units
);
8220 int skipped_disks
= 0;
8221 int max_degradation
;
8223 err
= sysfs_get_str(info
, NULL
, "array_state", (char *)buffer
, 20);
8227 /* recover data only during assemblation */
8228 if (strncmp(buffer
, "inactive", 8) != 0)
8230 /* no data to recover */
8231 if (__le32_to_cpu(migr_rec
->rec_status
) == UNIT_SRC_NORMAL
)
8233 if (curr_migr_unit
>= num_migr_units
)
8236 /* find device during reshape */
8237 for (id
= super
->devlist
; id
; id
= id
->next
)
8238 if (is_gen_migration(id
->dev
))
8243 map_dest
= get_imsm_map(id
->dev
, 0);
8244 new_disks
= map_dest
->num_members
;
8245 max_degradation
= new_disks
- imsm_num_data_members(id
->dev
, 0);
8247 read_offset
= (unsigned long long)
8248 __le32_to_cpu(migr_rec
->ckpt_area_pba
) * 512;
8250 write_offset
= ((unsigned long long)
8251 __le32_to_cpu(migr_rec
->dest_1st_member_lba
) +
8252 __le32_to_cpu(map_dest
->pba_of_lba0
)) * 512;
8254 unit_len
= __le32_to_cpu(migr_rec
->dest_depth_per_unit
) * 512;
8255 if (posix_memalign((void **)&buf
, 512, unit_len
) != 0)
8257 targets
= malloc(new_disks
* sizeof(int));
8261 open_backup_targets(info
, new_disks
, targets
);
8263 for (i
= 0; i
< new_disks
; i
++) {
8264 if (targets
[i
] < 0) {
8268 if (lseek64(targets
[i
], read_offset
, SEEK_SET
) < 0) {
8270 Name
": Cannot seek to block: %s\n",
8274 if ((unsigned)read(targets
[i
], buf
, unit_len
) != unit_len
) {
8276 Name
": Cannot read copy area block: %s\n",
8280 if (lseek64(targets
[i
], write_offset
, SEEK_SET
) < 0) {
8282 Name
": Cannot seek to block: %s\n",
8286 if ((unsigned)write(targets
[i
], buf
, unit_len
) != unit_len
) {
8288 Name
": Cannot restore block: %s\n",
8294 if (skipped_disks
> max_degradation
) {
8296 Name
": Cannot restore data from backup."
8297 " Too many failed disks\n");
8301 if (save_checkpoint_imsm(st
, info
, UNIT_SRC_NORMAL
)) {
8302 /* ignore error == 2, this can mean end of reshape here
8304 dprintf("imsm: Cannot write checkpoint to "
8305 "migration record (UNIT_SRC_NORMAL) during restart\n");
8311 for (i
= 0; i
< new_disks
; i
++)
8320 static char disk_by_path
[] = "/dev/disk/by-path/";
8322 static const char *imsm_get_disk_controller_domain(const char *path
)
8324 char disk_path
[PATH_MAX
];
8328 strncpy(disk_path
, disk_by_path
, PATH_MAX
- 1);
8329 strncat(disk_path
, path
, PATH_MAX
- strlen(disk_path
) - 1);
8330 if (stat(disk_path
, &st
) == 0) {
8331 struct sys_dev
* hba
;
8334 path
= devt_to_devpath(st
.st_rdev
);
8337 hba
= find_disk_attached_hba(-1, path
);
8338 if (hba
&& hba
->type
== SYS_DEV_SAS
)
8340 else if (hba
&& hba
->type
== SYS_DEV_SATA
)
8344 dprintf("path: %s hba: %s attached: %s\n",
8345 path
, (hba
) ? hba
->path
: "NULL", drv
);
8353 static int imsm_find_array_minor_by_subdev(int subdev
, int container
, int *minor
)
8355 char subdev_name
[20];
8356 struct mdstat_ent
*mdstat
;
8358 sprintf(subdev_name
, "%d", subdev
);
8359 mdstat
= mdstat_by_subdev(subdev_name
, container
);
8363 *minor
= mdstat
->devnum
;
8364 free_mdstat(mdstat
);
8368 static int imsm_reshape_is_allowed_on_container(struct supertype
*st
,
8369 struct geo_params
*geo
,
8370 int *old_raid_disks
)
8372 /* currently we only support increasing the number of devices
8373 * for a container. This increases the number of device for each
8374 * member array. They must all be RAID0 or RAID5.
8377 struct mdinfo
*info
, *member
;
8378 int devices_that_can_grow
= 0;
8380 dprintf("imsm: imsm_reshape_is_allowed_on_container(ENTER): "
8381 "st->devnum = (%i)\n",
8384 if (geo
->size
!= -1 ||
8385 geo
->level
!= UnSet
||
8386 geo
->layout
!= UnSet
||
8387 geo
->chunksize
!= 0 ||
8388 geo
->raid_disks
== UnSet
) {
8389 dprintf("imsm: Container operation is allowed for "
8390 "raid disks number change only.\n");
8394 info
= container_content_imsm(st
, NULL
);
8395 for (member
= info
; member
; member
= member
->next
) {
8399 dprintf("imsm: checking device_num: %i\n",
8400 member
->container_member
);
8402 if (geo
->raid_disks
<= member
->array
.raid_disks
) {
8403 /* we work on container for Online Capacity Expansion
8404 * only so raid_disks has to grow
8406 dprintf("imsm: for container operation raid disks "
8407 "increase is required\n");
8411 if ((info
->array
.level
!= 0) &&
8412 (info
->array
.level
!= 5)) {
8413 /* we cannot use this container with other raid level
8415 dprintf("imsm: for container operation wrong"
8416 " raid level (%i) detected\n",
8420 /* check for platform support
8421 * for this raid level configuration
8423 struct intel_super
*super
= st
->sb
;
8424 if (!is_raid_level_supported(super
->orom
,
8425 member
->array
.level
,
8427 dprintf("platform does not support raid%d with"
8431 geo
->raid_disks
> 1 ? "s" : "");
8434 /* check if component size is aligned to chunk size
8436 if (info
->component_size
%
8437 (info
->array
.chunk_size
/512)) {
8438 dprintf("Component size is not aligned to "
8444 if (*old_raid_disks
&&
8445 info
->array
.raid_disks
!= *old_raid_disks
)
8447 *old_raid_disks
= info
->array
.raid_disks
;
8449 /* All raid5 and raid0 volumes in container
8450 * have to be ready for Online Capacity Expansion
8451 * so they need to be assembled. We have already
8452 * checked that no recovery etc is happening.
8454 result
= imsm_find_array_minor_by_subdev(member
->container_member
,
8458 dprintf("imsm: cannot find array\n");
8461 devices_that_can_grow
++;
8464 if (!member
&& devices_that_can_grow
)
8468 dprintf("\tContainer operation allowed\n");
8470 dprintf("\tError: %i\n", ret_val
);
8475 /* Function: get_spares_for_grow
8476 * Description: Allocates memory and creates list of spare devices
8477 * avaliable in container. Checks if spare drive size is acceptable.
8478 * Parameters: Pointer to the supertype structure
8479 * Returns: Pointer to the list of spare devices (mdinfo structure) on success,
8482 static struct mdinfo
*get_spares_for_grow(struct supertype
*st
)
8484 unsigned long long min_size
= min_acceptable_spare_size_imsm(st
);
8485 return container_choose_spares(st
, min_size
, NULL
, NULL
, NULL
, 0);
8488 /******************************************************************************
8489 * function: imsm_create_metadata_update_for_reshape
8490 * Function creates update for whole IMSM container.
8492 ******************************************************************************/
8493 static int imsm_create_metadata_update_for_reshape(
8494 struct supertype
*st
,
8495 struct geo_params
*geo
,
8497 struct imsm_update_reshape
**updatep
)
8499 struct intel_super
*super
= st
->sb
;
8500 struct imsm_super
*mpb
= super
->anchor
;
8501 int update_memory_size
= 0;
8502 struct imsm_update_reshape
*u
= NULL
;
8503 struct mdinfo
*spares
= NULL
;
8505 int delta_disks
= 0;
8508 dprintf("imsm_update_metadata_for_reshape(enter) raid_disks = %i\n",
8511 delta_disks
= geo
->raid_disks
- old_raid_disks
;
8513 /* size of all update data without anchor */
8514 update_memory_size
= sizeof(struct imsm_update_reshape
);
8516 /* now add space for spare disks that we need to add. */
8517 update_memory_size
+= sizeof(u
->new_disks
[0]) * (delta_disks
- 1);
8519 u
= calloc(1, update_memory_size
);
8522 "cannot get memory for imsm_update_reshape update\n");
8525 u
->type
= update_reshape_container_disks
;
8526 u
->old_raid_disks
= old_raid_disks
;
8527 u
->new_raid_disks
= geo
->raid_disks
;
8529 /* now get spare disks list
8531 spares
= get_spares_for_grow(st
);
8534 || delta_disks
> spares
->array
.spare_disks
) {
8535 fprintf(stderr
, Name
": imsm: ERROR: Cannot get spare devices "
8536 "for %s.\n", geo
->dev_name
);
8541 /* we have got spares
8542 * update disk list in imsm_disk list table in anchor
8544 dprintf("imsm: %i spares are available.\n\n",
8545 spares
->array
.spare_disks
);
8548 for (i
= 0; i
< delta_disks
; i
++) {
8553 u
->new_disks
[i
] = makedev(dev
->disk
.major
,
8555 dl
= get_disk_super(super
, dev
->disk
.major
, dev
->disk
.minor
);
8556 dl
->index
= mpb
->num_disks
;
8566 dprintf("imsm: reshape update preparation :");
8567 if (i
== delta_disks
) {
8570 return update_memory_size
;
8573 dprintf(" Error\n");
8578 /******************************************************************************
8579 * function: imsm_create_metadata_update_for_migration()
8580 * Creates update for IMSM array.
8582 ******************************************************************************/
8583 static int imsm_create_metadata_update_for_migration(
8584 struct supertype
*st
,
8585 struct geo_params
*geo
,
8586 struct imsm_update_reshape_migration
**updatep
)
8588 struct intel_super
*super
= st
->sb
;
8589 int update_memory_size
= 0;
8590 struct imsm_update_reshape_migration
*u
= NULL
;
8591 struct imsm_dev
*dev
;
8592 int previous_level
= -1;
8594 dprintf("imsm_create_metadata_update_for_migration(enter)"
8595 " New Level = %i\n", geo
->level
);
8597 /* size of all update data without anchor */
8598 update_memory_size
= sizeof(struct imsm_update_reshape_migration
);
8600 u
= calloc(1, update_memory_size
);
8602 dprintf("error: cannot get memory for "
8603 "imsm_create_metadata_update_for_migration\n");
8606 u
->type
= update_reshape_migration
;
8607 u
->subdev
= super
->current_vol
;
8608 u
->new_level
= geo
->level
;
8609 u
->new_layout
= geo
->layout
;
8610 u
->new_raid_disks
= u
->old_raid_disks
= geo
->raid_disks
;
8611 u
->new_disks
[0] = -1;
8612 u
->new_chunksize
= -1;
8614 dev
= get_imsm_dev(super
, u
->subdev
);
8616 struct imsm_map
*map
;
8618 map
= get_imsm_map(dev
, 0);
8620 int current_chunk_size
=
8621 __le16_to_cpu(map
->blocks_per_strip
) / 2;
8623 if (geo
->chunksize
!= current_chunk_size
) {
8624 u
->new_chunksize
= geo
->chunksize
/ 1024;
8626 "chunk size change from %i to %i\n",
8627 current_chunk_size
, u
->new_chunksize
);
8629 previous_level
= map
->raid_level
;
8632 if ((geo
->level
== 5) && (previous_level
== 0)) {
8633 struct mdinfo
*spares
= NULL
;
8635 u
->new_raid_disks
++;
8636 spares
= get_spares_for_grow(st
);
8637 if ((spares
== NULL
) || (spares
->array
.spare_disks
< 1)) {
8640 update_memory_size
= 0;
8641 dprintf("error: cannot get spare device "
8642 "for requested migration");
8647 dprintf("imsm: reshape update preparation : OK\n");
8650 return update_memory_size
;
8653 static void imsm_update_metadata_locally(struct supertype
*st
,
8656 struct metadata_update mu
;
8661 mu
.space_list
= NULL
;
8663 imsm_prepare_update(st
, &mu
);
8664 imsm_process_update(st
, &mu
);
8666 while (mu
.space_list
) {
8667 void **space
= mu
.space_list
;
8668 mu
.space_list
= *space
;
8673 /***************************************************************************
8674 * Function: imsm_analyze_change
8675 * Description: Function analyze change for single volume
8676 * and validate if transition is supported
8677 * Parameters: Geometry parameters, supertype structure
8678 * Returns: Operation type code on success, -1 if fail
8679 ****************************************************************************/
8680 enum imsm_reshape_type
imsm_analyze_change(struct supertype
*st
,
8681 struct geo_params
*geo
)
8690 getinfo_super_imsm_volume(st
, &info
, NULL
);
8691 if ((geo
->level
!= info
.array
.level
) &&
8692 (geo
->level
>= 0) &&
8693 (geo
->level
!= UnSet
)) {
8694 switch (info
.array
.level
) {
8696 if (geo
->level
== 5) {
8697 change
= CH_MIGRATION
;
8698 if (geo
->layout
!= ALGORITHM_LEFT_ASYMMETRIC
) {
8700 Name
" Error. Requested Layout "
8701 "not supported (left-asymmetric layout "
8702 "is supported only)!\n");
8704 goto analyse_change_exit
;
8706 layout
= geo
->layout
;
8708 devNumChange
= 1; /* parity disk added */
8709 } else if (geo
->level
== 10) {
8710 change
= CH_TAKEOVER
;
8712 devNumChange
= 2; /* two mirrors added */
8713 layout
= 0x102; /* imsm supported layout */
8718 if (geo
->level
== 0) {
8719 change
= CH_TAKEOVER
;
8721 devNumChange
= -(geo
->raid_disks
/2);
8722 layout
= 0; /* imsm raid0 layout */
8728 Name
" Error. Level Migration from %d to %d "
8730 info
.array
.level
, geo
->level
);
8731 goto analyse_change_exit
;
8734 geo
->level
= info
.array
.level
;
8736 if ((geo
->layout
!= info
.array
.layout
)
8737 && ((geo
->layout
!= UnSet
) && (geo
->layout
!= -1))) {
8738 change
= CH_MIGRATION
;
8739 if ((info
.array
.layout
== 0)
8740 && (info
.array
.level
== 5)
8741 && (geo
->layout
== 5)) {
8742 /* reshape 5 -> 4 */
8743 } else if ((info
.array
.layout
== 5)
8744 && (info
.array
.level
== 5)
8745 && (geo
->layout
== 0)) {
8746 /* reshape 4 -> 5 */
8751 Name
" Error. Layout Migration from %d to %d "
8753 info
.array
.layout
, geo
->layout
);
8755 goto analyse_change_exit
;
8758 geo
->layout
= info
.array
.layout
;
8760 if ((geo
->chunksize
> 0) && (geo
->chunksize
!= UnSet
)
8761 && (geo
->chunksize
!= info
.array
.chunk_size
))
8762 change
= CH_MIGRATION
;
8764 geo
->chunksize
= info
.array
.chunk_size
;
8766 chunk
= geo
->chunksize
/ 1024;
8767 if (!validate_geometry_imsm(st
,
8770 geo
->raid_disks
+ devNumChange
,
8777 struct intel_super
*super
= st
->sb
;
8778 struct imsm_super
*mpb
= super
->anchor
;
8780 if (mpb
->num_raid_devs
> 1) {
8782 Name
" Error. Cannot perform operation on %s"
8783 "- for this operation it MUST be single "
8784 "array in container\n",
8790 analyse_change_exit
:
8795 int imsm_takeover(struct supertype
*st
, struct geo_params
*geo
)
8797 struct intel_super
*super
= st
->sb
;
8798 struct imsm_update_takeover
*u
;
8800 u
= malloc(sizeof(struct imsm_update_takeover
));
8804 u
->type
= update_takeover
;
8805 u
->subarray
= super
->current_vol
;
8807 /* 10->0 transition */
8808 if (geo
->level
== 0)
8809 u
->direction
= R10_TO_R0
;
8811 /* 0->10 transition */
8812 if (geo
->level
== 10)
8813 u
->direction
= R0_TO_R10
;
8815 /* update metadata locally */
8816 imsm_update_metadata_locally(st
, u
,
8817 sizeof(struct imsm_update_takeover
));
8818 /* and possibly remotely */
8819 if (st
->update_tail
)
8820 append_metadata_update(st
, u
,
8821 sizeof(struct imsm_update_takeover
));
8828 static int imsm_reshape_super(struct supertype
*st
, long long size
, int level
,
8829 int layout
, int chunksize
, int raid_disks
,
8830 int delta_disks
, char *backup
, char *dev
,
8834 struct geo_params geo
;
8836 dprintf("imsm: reshape_super called.\n");
8838 memset(&geo
, 0, sizeof(struct geo_params
));
8841 geo
.dev_id
= st
->devnum
;
8844 geo
.layout
= layout
;
8845 geo
.chunksize
= chunksize
;
8846 geo
.raid_disks
= raid_disks
;
8847 if (delta_disks
!= UnSet
)
8848 geo
.raid_disks
+= delta_disks
;
8850 dprintf("\tfor level : %i\n", geo
.level
);
8851 dprintf("\tfor raid_disks : %i\n", geo
.raid_disks
);
8853 if (experimental() == 0)
8856 if (st
->container_dev
== st
->devnum
) {
8857 /* On container level we can only increase number of devices. */
8858 dprintf("imsm: info: Container operation\n");
8859 int old_raid_disks
= 0;
8861 if (imsm_reshape_is_allowed_on_container(
8862 st
, &geo
, &old_raid_disks
)) {
8863 struct imsm_update_reshape
*u
= NULL
;
8866 len
= imsm_create_metadata_update_for_reshape(
8867 st
, &geo
, old_raid_disks
, &u
);
8870 dprintf("imsm: Cannot prepare update\n");
8871 goto exit_imsm_reshape_super
;
8875 /* update metadata locally */
8876 imsm_update_metadata_locally(st
, u
, len
);
8877 /* and possibly remotely */
8878 if (st
->update_tail
)
8879 append_metadata_update(st
, u
, len
);
8884 fprintf(stderr
, Name
": (imsm) Operation "
8885 "is not allowed on this container\n");
8888 /* On volume level we support following operations
8889 * - takeover: raid10 -> raid0; raid0 -> raid10
8890 * - chunk size migration
8891 * - migration: raid5 -> raid0; raid0 -> raid5
8893 struct intel_super
*super
= st
->sb
;
8894 struct intel_dev
*dev
= super
->devlist
;
8896 dprintf("imsm: info: Volume operation\n");
8897 /* find requested device */
8899 if (imsm_find_array_minor_by_subdev(
8900 dev
->index
, st
->container_dev
, &devnum
) == 0
8901 && devnum
== geo
.dev_id
)
8906 fprintf(stderr
, Name
" Cannot find %s (%i) subarray\n",
8907 geo
.dev_name
, geo
.dev_id
);
8908 goto exit_imsm_reshape_super
;
8910 super
->current_vol
= dev
->index
;
8911 change
= imsm_analyze_change(st
, &geo
);
8914 ret_val
= imsm_takeover(st
, &geo
);
8916 case CH_MIGRATION
: {
8917 struct imsm_update_reshape_migration
*u
= NULL
;
8919 imsm_create_metadata_update_for_migration(
8923 "Cannot prepare update\n");
8927 /* update metadata locally */
8928 imsm_update_metadata_locally(st
, u
, len
);
8929 /* and possibly remotely */
8930 if (st
->update_tail
)
8931 append_metadata_update(st
, u
, len
);
8941 exit_imsm_reshape_super
:
8942 dprintf("imsm: reshape_super Exit code = %i\n", ret_val
);
8946 /*******************************************************************************
8947 * Function: wait_for_reshape_imsm
8948 * Description: Function writes new sync_max value and waits until
8949 * reshape process reach new position
8951 * sra : general array info
8952 * ndata : number of disks in new array's layout
8955 * 1 : there is no reshape in progress,
8957 ******************************************************************************/
8958 int wait_for_reshape_imsm(struct mdinfo
*sra
, int ndata
)
8960 int fd
= sysfs_get_fd(sra
, NULL
, "reshape_position");
8961 unsigned long long completed
;
8962 /* to_complete : new sync_max position */
8963 unsigned long long to_complete
= sra
->reshape_progress
;
8964 unsigned long long position_to_set
= to_complete
/ ndata
;
8967 dprintf("imsm: wait_for_reshape_imsm() "
8968 "cannot open reshape_position\n");
8972 if (sysfs_fd_get_ll(fd
, &completed
) < 0) {
8973 dprintf("imsm: wait_for_reshape_imsm() "
8974 "cannot read reshape_position (no reshape in progres)\n");
8979 if (completed
> to_complete
) {
8980 dprintf("imsm: wait_for_reshape_imsm() "
8981 "wrong next position to set %llu (%llu)\n",
8982 to_complete
, completed
);
8986 dprintf("Position set: %llu\n", position_to_set
);
8987 if (sysfs_set_num(sra
, NULL
, "sync_max",
8988 position_to_set
) != 0) {
8989 dprintf("imsm: wait_for_reshape_imsm() "
8990 "cannot set reshape position to %llu\n",
9001 select(fd
+1, &rfds
, NULL
, NULL
, NULL
);
9002 if (sysfs_get_str(sra
, NULL
, "sync_action",
9004 strncmp(action
, "reshape", 7) != 0)
9006 if (sysfs_fd_get_ll(fd
, &completed
) < 0) {
9007 dprintf("imsm: wait_for_reshape_imsm() "
9008 "cannot read reshape_position (in loop)\n");
9012 } while (completed
< to_complete
);
9018 /*******************************************************************************
9019 * Function: check_degradation_change
9020 * Description: Check that array hasn't become failed.
9022 * info : for sysfs access
9023 * sources : source disks descriptors
9024 * degraded: previous degradation level
9027 ******************************************************************************/
9028 int check_degradation_change(struct mdinfo
*info
,
9032 unsigned long long new_degraded
;
9033 sysfs_get_ll(info
, NULL
, "degraded", &new_degraded
);
9034 if (new_degraded
!= (unsigned long long)degraded
) {
9035 /* check each device to ensure it is still working */
9038 for (sd
= info
->devs
; sd
; sd
= sd
->next
) {
9039 if (sd
->disk
.state
& (1<<MD_DISK_FAULTY
))
9041 if (sd
->disk
.state
& (1<<MD_DISK_SYNC
)) {
9043 if (sysfs_get_str(info
,
9044 sd
, "state", sbuf
, 20) < 0 ||
9045 strstr(sbuf
, "faulty") ||
9046 strstr(sbuf
, "in_sync") == NULL
) {
9047 /* this device is dead */
9048 sd
->disk
.state
= (1<<MD_DISK_FAULTY
);
9049 if (sd
->disk
.raid_disk
>= 0 &&
9050 sources
[sd
->disk
.raid_disk
] >= 0) {
9052 sd
->disk
.raid_disk
]);
9053 sources
[sd
->disk
.raid_disk
] =
9062 return new_degraded
;
9065 /*******************************************************************************
9066 * Function: imsm_manage_reshape
9067 * Description: Function finds array under reshape and it manages reshape
9068 * process. It creates stripes backups (if required) and sets
9071 * afd : Backup handle (nattive) - not used
9072 * sra : general array info
9073 * reshape : reshape parameters - not used
9074 * st : supertype structure
9075 * blocks : size of critical section [blocks]
9076 * fds : table of source device descriptor
9077 * offsets : start of array (offest per devices)
9079 * destfd : table of destination device descriptor
9080 * destoffsets : table of destination offsets (per device)
9082 * 1 : success, reshape is done
9084 ******************************************************************************/
9085 static int imsm_manage_reshape(
9086 int afd
, struct mdinfo
*sra
, struct reshape
*reshape
,
9087 struct supertype
*st
, unsigned long backup_blocks
,
9088 int *fds
, unsigned long long *offsets
,
9089 int dests
, int *destfd
, unsigned long long *destoffsets
)
9092 struct intel_super
*super
= st
->sb
;
9093 struct intel_dev
*dv
= NULL
;
9094 struct imsm_dev
*dev
= NULL
;
9095 struct imsm_map
*map_src
;
9096 int migr_vol_qan
= 0;
9097 int ndata
, odata
; /* [bytes] */
9098 int chunk
; /* [bytes] */
9099 struct migr_record
*migr_rec
;
9101 unsigned int buf_size
; /* [bytes] */
9102 unsigned long long max_position
; /* array size [bytes] */
9103 unsigned long long next_step
; /* [blocks]/[bytes] */
9104 unsigned long long old_data_stripe_length
;
9105 unsigned long long start_src
; /* [bytes] */
9106 unsigned long long start
; /* [bytes] */
9107 unsigned long long start_buf_shift
; /* [bytes] */
9109 int source_layout
= 0;
9111 if (!fds
|| !offsets
|| !sra
)
9114 /* Find volume during the reshape */
9115 for (dv
= super
->devlist
; dv
; dv
= dv
->next
) {
9116 if (dv
->dev
->vol
.migr_type
== MIGR_GEN_MIGR
9117 && dv
->dev
->vol
.migr_state
== 1) {
9122 /* Only one volume can migrate at the same time */
9123 if (migr_vol_qan
!= 1) {
9124 fprintf(stderr
, Name
" : %s", migr_vol_qan
?
9125 "Number of migrating volumes greater than 1\n" :
9126 "There is no volume during migrationg\n");
9130 map_src
= get_imsm_map(dev
, 1);
9131 if (map_src
== NULL
)
9134 ndata
= imsm_num_data_members(dev
, 0);
9135 odata
= imsm_num_data_members(dev
, 1);
9137 chunk
= __le16_to_cpu(map_src
->blocks_per_strip
) * 512;
9138 old_data_stripe_length
= odata
* chunk
;
9140 migr_rec
= super
->migr_rec
;
9142 /* initialize migration record for start condition */
9143 if (sra
->reshape_progress
== 0)
9144 init_migr_record_imsm(st
, dev
, sra
);
9146 if (__le32_to_cpu(migr_rec
->rec_status
) != UNIT_SRC_NORMAL
) {
9147 dprintf("imsm: cannot restart migration when data "
9148 "are present in copy area.\n");
9154 buf_size
= __le32_to_cpu(migr_rec
->blocks_per_unit
) * 512;
9155 /* extend buffer size for parity disk */
9156 buf_size
+= __le32_to_cpu(migr_rec
->dest_depth_per_unit
) * 512;
9157 /* add space for stripe aligment */
9158 buf_size
+= old_data_stripe_length
;
9159 if (posix_memalign((void **)&buf
, 4096, buf_size
)) {
9160 dprintf("imsm: Cannot allocate checpoint buffer\n");
9164 max_position
= sra
->component_size
* ndata
;
9165 source_layout
= imsm_level_to_layout(map_src
->raid_level
);
9167 while (__le32_to_cpu(migr_rec
->curr_migr_unit
) <
9168 __le32_to_cpu(migr_rec
->num_migr_units
)) {
9169 /* current reshape position [blocks] */
9170 unsigned long long current_position
=
9171 __le32_to_cpu(migr_rec
->blocks_per_unit
)
9172 * __le32_to_cpu(migr_rec
->curr_migr_unit
);
9173 unsigned long long border
;
9175 /* Check that array hasn't become failed.
9177 degraded
= check_degradation_change(sra
, fds
, degraded
);
9179 dprintf("imsm: Abort reshape due to degradation"
9180 " level (%i)\n", degraded
);
9184 next_step
= __le32_to_cpu(migr_rec
->blocks_per_unit
);
9186 if ((current_position
+ next_step
) > max_position
)
9187 next_step
= max_position
- current_position
;
9189 start
= current_position
* 512;
9191 /* allign reading start to old geometry */
9192 start_buf_shift
= start
% old_data_stripe_length
;
9193 start_src
= start
- start_buf_shift
;
9195 border
= (start_src
/ odata
) - (start
/ ndata
);
9197 if (border
<= __le32_to_cpu(migr_rec
->dest_depth_per_unit
)) {
9198 /* save critical stripes to buf
9199 * start - start address of current unit
9201 * start_src - start address of current unit
9202 * to backup alligned to source array
9205 unsigned long long next_step_filler
= 0;
9206 unsigned long long copy_length
= next_step
* 512;
9208 /* allign copy area length to stripe in old geometry */
9209 next_step_filler
= ((copy_length
+ start_buf_shift
)
9210 % old_data_stripe_length
);
9211 if (next_step_filler
)
9212 next_step_filler
= (old_data_stripe_length
9213 - next_step_filler
);
9214 dprintf("save_stripes() parameters: start = %llu,"
9215 "\tstart_src = %llu,\tnext_step*512 = %llu,"
9216 "\tstart_in_buf_shift = %llu,"
9217 "\tnext_step_filler = %llu\n",
9218 start
, start_src
, copy_length
,
9219 start_buf_shift
, next_step_filler
);
9221 if (save_stripes(fds
, offsets
, map_src
->num_members
,
9222 chunk
, map_src
->raid_level
,
9223 source_layout
, 0, NULL
, start_src
,
9225 next_step_filler
+ start_buf_shift
,
9227 dprintf("imsm: Cannot save stripes"
9231 /* Convert data to destination format and store it
9232 * in backup general migration area
9234 if (save_backup_imsm(st
, dev
, sra
,
9235 buf
+ start_buf_shift
, copy_length
)) {
9236 dprintf("imsm: Cannot save stripes to "
9237 "target devices\n");
9240 if (save_checkpoint_imsm(st
, sra
,
9241 UNIT_SRC_IN_CP_AREA
)) {
9242 dprintf("imsm: Cannot write checkpoint to "
9243 "migration record (UNIT_SRC_IN_CP_AREA)\n");
9247 /* set next step to use whole border area */
9248 border
/= next_step
;
9250 next_step
*= border
;
9252 /* When data backed up, checkpoint stored,
9253 * kick the kernel to reshape unit of data
9255 next_step
= next_step
+ sra
->reshape_progress
;
9256 /* limit next step to array max position */
9257 if (next_step
> max_position
)
9258 next_step
= max_position
;
9259 sysfs_set_num(sra
, NULL
, "suspend_lo", sra
->reshape_progress
);
9260 sysfs_set_num(sra
, NULL
, "suspend_hi", next_step
);
9261 sra
->reshape_progress
= next_step
;
9263 /* wait until reshape finish */
9264 if (wait_for_reshape_imsm(sra
, ndata
) < 0) {
9265 dprintf("wait_for_reshape_imsm returned error!\n");
9269 if (save_checkpoint_imsm(st
, sra
, UNIT_SRC_NORMAL
) == 1) {
9270 /* ignore error == 2, this can mean end of reshape here
9272 dprintf("imsm: Cannot write checkpoint to "
9273 "migration record (UNIT_SRC_NORMAL)\n");
9279 /* return '1' if done */
9287 #endif /* MDASSEMBLE */
9289 struct superswitch super_imsm
= {
9291 .examine_super
= examine_super_imsm
,
9292 .brief_examine_super
= brief_examine_super_imsm
,
9293 .brief_examine_subarrays
= brief_examine_subarrays_imsm
,
9294 .export_examine_super
= export_examine_super_imsm
,
9295 .detail_super
= detail_super_imsm
,
9296 .brief_detail_super
= brief_detail_super_imsm
,
9297 .write_init_super
= write_init_super_imsm
,
9298 .validate_geometry
= validate_geometry_imsm
,
9299 .add_to_super
= add_to_super_imsm
,
9300 .remove_from_super
= remove_from_super_imsm
,
9301 .detail_platform
= detail_platform_imsm
,
9302 .kill_subarray
= kill_subarray_imsm
,
9303 .update_subarray
= update_subarray_imsm
,
9304 .load_container
= load_container_imsm
,
9305 .default_geometry
= default_geometry_imsm
,
9306 .get_disk_controller_domain
= imsm_get_disk_controller_domain
,
9307 .reshape_super
= imsm_reshape_super
,
9308 .manage_reshape
= imsm_manage_reshape
,
9309 .recover_backup
= recover_backup_imsm
,
9311 .match_home
= match_home_imsm
,
9312 .uuid_from_super
= uuid_from_super_imsm
,
9313 .getinfo_super
= getinfo_super_imsm
,
9314 .getinfo_super_disks
= getinfo_super_disks_imsm
,
9315 .update_super
= update_super_imsm
,
9317 .avail_size
= avail_size_imsm
,
9318 .min_acceptable_spare_size
= min_acceptable_spare_size_imsm
,
9320 .compare_super
= compare_super_imsm
,
9322 .load_super
= load_super_imsm
,
9323 .init_super
= init_super_imsm
,
9324 .store_super
= store_super_imsm
,
9325 .free_super
= free_super_imsm
,
9326 .match_metadata_desc
= match_metadata_desc_imsm
,
9327 .container_content
= container_content_imsm
,
9335 .open_new
= imsm_open_new
,
9336 .set_array_state
= imsm_set_array_state
,
9337 .set_disk
= imsm_set_disk
,
9338 .sync_metadata
= imsm_sync_metadata
,
9339 .activate_spare
= imsm_activate_spare
,
9340 .process_update
= imsm_process_update
,
9341 .prepare_update
= imsm_prepare_update
,
9342 #endif /* MDASSEMBLE */