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
668 * -2 - we return longer map /excluding uninitialized state/
670 struct imsm_map
*map
= &dev
->vol
.map
[0];
671 struct imsm_map
*map2
= NULL
;
673 if (dev
->vol
.migr_state
)
674 map2
= (void *)map
+ sizeof_imsm_map(map
);
676 switch (second_map
) {
688 && map2
->map_state
!= IMSM_T_STATE_UNINITIALIZED
689 && map2
->num_members
> map
->num_members
)
699 /* return the size of the device.
700 * migr_state increases the returned size if map[0] were to be duplicated
702 static size_t sizeof_imsm_dev(struct imsm_dev
*dev
, int migr_state
)
704 size_t size
= sizeof(*dev
) - sizeof(struct imsm_map
) +
705 sizeof_imsm_map(get_imsm_map(dev
, 0));
707 /* migrating means an additional map */
708 if (dev
->vol
.migr_state
)
709 size
+= sizeof_imsm_map(get_imsm_map(dev
, 1));
711 size
+= sizeof_imsm_map(get_imsm_map(dev
, 0));
717 /* retrieve disk serial number list from a metadata update */
718 static struct disk_info
*get_disk_info(struct imsm_update_create_array
*update
)
721 struct disk_info
*inf
;
723 inf
= u
+ sizeof(*update
) - sizeof(struct imsm_dev
) +
724 sizeof_imsm_dev(&update
->dev
, 0);
730 static struct imsm_dev
*__get_imsm_dev(struct imsm_super
*mpb
, __u8 index
)
736 if (index
>= mpb
->num_raid_devs
)
739 /* devices start after all disks */
740 offset
= ((void *) &mpb
->disk
[mpb
->num_disks
]) - _mpb
;
742 for (i
= 0; i
<= index
; i
++)
744 return _mpb
+ offset
;
746 offset
+= sizeof_imsm_dev(_mpb
+ offset
, 0);
751 static struct imsm_dev
*get_imsm_dev(struct intel_super
*super
, __u8 index
)
753 struct intel_dev
*dv
;
755 if (index
>= super
->anchor
->num_raid_devs
)
757 for (dv
= super
->devlist
; dv
; dv
= dv
->next
)
758 if (dv
->index
== index
)
766 * == 1 get second map
767 * == -1 than get map according to the current migr_state
769 static __u32
get_imsm_ord_tbl_ent(struct imsm_dev
*dev
,
773 struct imsm_map
*map
;
775 map
= get_imsm_map(dev
, second_map
);
777 /* top byte identifies disk under rebuild */
778 return __le32_to_cpu(map
->disk_ord_tbl
[slot
]);
781 #define ord_to_idx(ord) (((ord) << 8) >> 8)
782 static __u32
get_imsm_disk_idx(struct imsm_dev
*dev
, int slot
, int second_map
)
784 __u32 ord
= get_imsm_ord_tbl_ent(dev
, slot
, second_map
);
786 return ord_to_idx(ord
);
789 static void set_imsm_ord_tbl_ent(struct imsm_map
*map
, int slot
, __u32 ord
)
791 map
->disk_ord_tbl
[slot
] = __cpu_to_le32(ord
);
794 static int get_imsm_disk_slot(struct imsm_map
*map
, unsigned idx
)
799 for (slot
= 0; slot
< map
->num_members
; slot
++) {
800 ord
= __le32_to_cpu(map
->disk_ord_tbl
[slot
]);
801 if (ord_to_idx(ord
) == idx
)
808 static int get_imsm_raid_level(struct imsm_map
*map
)
810 if (map
->raid_level
== 1) {
811 if (map
->num_members
== 2)
817 return map
->raid_level
;
820 static int cmp_extent(const void *av
, const void *bv
)
822 const struct extent
*a
= av
;
823 const struct extent
*b
= bv
;
824 if (a
->start
< b
->start
)
826 if (a
->start
> b
->start
)
831 static int count_memberships(struct dl
*dl
, struct intel_super
*super
)
836 for (i
= 0; i
< super
->anchor
->num_raid_devs
; i
++) {
837 struct imsm_dev
*dev
= get_imsm_dev(super
, i
);
838 struct imsm_map
*map
= get_imsm_map(dev
, 0);
840 if (get_imsm_disk_slot(map
, dl
->index
) >= 0)
847 static __u32
imsm_min_reserved_sectors(struct intel_super
*super
);
849 static struct extent
*get_extents(struct intel_super
*super
, struct dl
*dl
)
851 /* find a list of used extents on the given physical device */
852 struct extent
*rv
, *e
;
854 int memberships
= count_memberships(dl
, super
);
857 /* trim the reserved area for spares, so they can join any array
858 * regardless of whether the OROM has assigned sectors from the
859 * IMSM_RESERVED_SECTORS region
862 reservation
= imsm_min_reserved_sectors(super
);
864 reservation
= MPB_SECTOR_CNT
+ IMSM_RESERVED_SECTORS
;
866 rv
= malloc(sizeof(struct extent
) * (memberships
+ 1));
871 for (i
= 0; i
< super
->anchor
->num_raid_devs
; i
++) {
872 struct imsm_dev
*dev
= get_imsm_dev(super
, i
);
873 struct imsm_map
*map
= get_imsm_map(dev
, 0);
875 if (get_imsm_disk_slot(map
, dl
->index
) >= 0) {
876 e
->start
= __le32_to_cpu(map
->pba_of_lba0
);
877 e
->size
= __le32_to_cpu(map
->blocks_per_member
);
881 qsort(rv
, memberships
, sizeof(*rv
), cmp_extent
);
883 /* determine the start of the metadata
884 * when no raid devices are defined use the default
885 * ...otherwise allow the metadata to truncate the value
886 * as is the case with older versions of imsm
889 struct extent
*last
= &rv
[memberships
- 1];
892 remainder
= __le32_to_cpu(dl
->disk
.total_blocks
) -
893 (last
->start
+ last
->size
);
894 /* round down to 1k block to satisfy precision of the kernel
898 /* make sure remainder is still sane */
899 if (remainder
< (unsigned)ROUND_UP(super
->len
, 512) >> 9)
900 remainder
= ROUND_UP(super
->len
, 512) >> 9;
901 if (reservation
> remainder
)
902 reservation
= remainder
;
904 e
->start
= __le32_to_cpu(dl
->disk
.total_blocks
) - reservation
;
909 /* try to determine how much space is reserved for metadata from
910 * the last get_extents() entry, otherwise fallback to the
913 static __u32
imsm_reserved_sectors(struct intel_super
*super
, struct dl
*dl
)
919 /* for spares just return a minimal reservation which will grow
920 * once the spare is picked up by an array
923 return MPB_SECTOR_CNT
;
925 e
= get_extents(super
, dl
);
927 return MPB_SECTOR_CNT
+ IMSM_RESERVED_SECTORS
;
929 /* scroll to last entry */
930 for (i
= 0; e
[i
].size
; i
++)
933 rv
= __le32_to_cpu(dl
->disk
.total_blocks
) - e
[i
].start
;
940 static int is_spare(struct imsm_disk
*disk
)
942 return (disk
->status
& SPARE_DISK
) == SPARE_DISK
;
945 static int is_configured(struct imsm_disk
*disk
)
947 return (disk
->status
& CONFIGURED_DISK
) == CONFIGURED_DISK
;
950 static int is_failed(struct imsm_disk
*disk
)
952 return (disk
->status
& FAILED_DISK
) == FAILED_DISK
;
955 /* try to determine how much space is reserved for metadata from
956 * the last get_extents() entry on the smallest active disk,
957 * otherwise fallback to the default
959 static __u32
imsm_min_reserved_sectors(struct intel_super
*super
)
963 __u32 min_active
, remainder
;
964 __u32 rv
= MPB_SECTOR_CNT
+ IMSM_RESERVED_SECTORS
;
965 struct dl
*dl
, *dl_min
= NULL
;
971 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
974 if (dl
->disk
.total_blocks
< min_active
|| min_active
== 0) {
976 min_active
= dl
->disk
.total_blocks
;
982 /* find last lba used by subarrays on the smallest active disk */
983 e
= get_extents(super
, dl_min
);
986 for (i
= 0; e
[i
].size
; i
++)
989 remainder
= min_active
- e
[i
].start
;
992 /* to give priority to recovery we should not require full
993 IMSM_RESERVED_SECTORS from the spare */
994 rv
= MPB_SECTOR_CNT
+ NUM_BLOCKS_DIRTY_STRIPE_REGION
;
996 /* if real reservation is smaller use that value */
997 return (remainder
< rv
) ? remainder
: rv
;
1000 /* Return minimum size of a spare that can be used in this array*/
1001 static unsigned long long min_acceptable_spare_size_imsm(struct supertype
*st
)
1003 struct intel_super
*super
= st
->sb
;
1007 unsigned long long rv
= 0;
1011 /* find first active disk in array */
1013 while (dl
&& (is_failed(&dl
->disk
) || dl
->index
== -1))
1017 /* find last lba used by subarrays */
1018 e
= get_extents(super
, dl
);
1021 for (i
= 0; e
[i
].size
; i
++)
1024 rv
= e
[i
-1].start
+ e
[i
-1].size
;
1027 /* add the amount of space needed for metadata */
1028 rv
= rv
+ imsm_min_reserved_sectors(super
);
1034 static __u64
blocks_per_migr_unit(struct intel_super
*super
,
1035 struct imsm_dev
*dev
);
1037 static int is_gen_migration(struct imsm_dev
*dev
);
1039 static void print_imsm_dev(struct intel_super
*super
,
1040 struct imsm_dev
*dev
,
1046 struct imsm_map
*map
= get_imsm_map(dev
, 0);
1047 struct imsm_map
*map2
= get_imsm_map(dev
, 1);
1051 printf("[%.16s]:\n", dev
->volume
);
1052 printf(" UUID : %s\n", uuid
);
1053 printf(" RAID Level : %d", get_imsm_raid_level(map
));
1055 printf(" <-- %d", get_imsm_raid_level(map2
));
1057 printf(" Members : %d", map
->num_members
);
1059 printf(" <-- %d", map2
->num_members
);
1061 printf(" Slots : [");
1062 for (i
= 0; i
< map
->num_members
; i
++) {
1063 ord
= get_imsm_ord_tbl_ent(dev
, i
, 0);
1064 printf("%s", ord
& IMSM_ORD_REBUILD
? "_" : "U");
1069 for (i
= 0; i
< map2
->num_members
; i
++) {
1070 ord
= get_imsm_ord_tbl_ent(dev
, i
, 1);
1071 printf("%s", ord
& IMSM_ORD_REBUILD
? "_" : "U");
1076 printf(" Failed disk : ");
1077 if (map
->failed_disk_num
== 0xff)
1080 printf("%i", map
->failed_disk_num
);
1082 slot
= get_imsm_disk_slot(map
, disk_idx
);
1084 ord
= get_imsm_ord_tbl_ent(dev
, slot
, -1);
1085 printf(" This Slot : %d%s\n", slot
,
1086 ord
& IMSM_ORD_REBUILD
? " (out-of-sync)" : "");
1088 printf(" This Slot : ?\n");
1089 sz
= __le32_to_cpu(dev
->size_high
);
1091 sz
+= __le32_to_cpu(dev
->size_low
);
1092 printf(" Array Size : %llu%s\n", (unsigned long long)sz
,
1093 human_size(sz
* 512));
1094 sz
= __le32_to_cpu(map
->blocks_per_member
);
1095 printf(" Per Dev Size : %llu%s\n", (unsigned long long)sz
,
1096 human_size(sz
* 512));
1097 printf(" Sector Offset : %u\n",
1098 __le32_to_cpu(map
->pba_of_lba0
));
1099 printf(" Num Stripes : %u\n",
1100 __le32_to_cpu(map
->num_data_stripes
));
1101 printf(" Chunk Size : %u KiB",
1102 __le16_to_cpu(map
->blocks_per_strip
) / 2);
1104 printf(" <-- %u KiB",
1105 __le16_to_cpu(map2
->blocks_per_strip
) / 2);
1107 printf(" Reserved : %d\n", __le32_to_cpu(dev
->reserved_blocks
));
1108 printf(" Migrate State : ");
1109 if (dev
->vol
.migr_state
) {
1110 if (migr_type(dev
) == MIGR_INIT
)
1111 printf("initialize\n");
1112 else if (migr_type(dev
) == MIGR_REBUILD
)
1113 printf("rebuild\n");
1114 else if (migr_type(dev
) == MIGR_VERIFY
)
1116 else if (migr_type(dev
) == MIGR_GEN_MIGR
)
1117 printf("general migration\n");
1118 else if (migr_type(dev
) == MIGR_STATE_CHANGE
)
1119 printf("state change\n");
1120 else if (migr_type(dev
) == MIGR_REPAIR
)
1123 printf("<unknown:%d>\n", migr_type(dev
));
1126 printf(" Map State : %s", map_state_str
[map
->map_state
]);
1127 if (dev
->vol
.migr_state
) {
1128 struct imsm_map
*map
= get_imsm_map(dev
, 1);
1130 printf(" <-- %s", map_state_str
[map
->map_state
]);
1131 printf("\n Checkpoint : %u ",
1132 __le32_to_cpu(dev
->vol
.curr_migr_unit
));
1133 if ((is_gen_migration(dev
)) && (super
->disks
->index
> 1))
1136 printf("(%llu)", (unsigned long long)
1137 blocks_per_migr_unit(super
, dev
));
1140 printf(" Dirty State : %s\n", dev
->vol
.dirty
? "dirty" : "clean");
1143 static void print_imsm_disk(struct imsm_disk
*disk
, int index
, __u32 reserved
)
1145 char str
[MAX_RAID_SERIAL_LEN
+ 1];
1148 if (index
< -1 || !disk
)
1152 snprintf(str
, MAX_RAID_SERIAL_LEN
+ 1, "%s", disk
->serial
);
1154 printf(" Disk%02d Serial : %s\n", index
, str
);
1156 printf(" Disk Serial : %s\n", str
);
1157 printf(" State :%s%s%s\n", is_spare(disk
) ? " spare" : "",
1158 is_configured(disk
) ? " active" : "",
1159 is_failed(disk
) ? " failed" : "");
1160 printf(" Id : %08x\n", __le32_to_cpu(disk
->scsi_id
));
1161 sz
= __le32_to_cpu(disk
->total_blocks
) - reserved
;
1162 printf(" Usable Size : %llu%s\n", (unsigned long long)sz
,
1163 human_size(sz
* 512));
1166 void examine_migr_rec_imsm(struct intel_super
*super
)
1168 struct migr_record
*migr_rec
= super
->migr_rec
;
1169 struct imsm_super
*mpb
= super
->anchor
;
1172 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
1173 struct imsm_dev
*dev
= __get_imsm_dev(mpb
, i
);
1174 if (is_gen_migration(dev
) == 0)
1177 printf("\nMigration Record Information:");
1178 if (super
->disks
->index
> 1) {
1179 printf(" Empty\n ");
1180 printf("Examine one of first two disks in array\n");
1183 printf("\n Status : ");
1184 if (__le32_to_cpu(migr_rec
->rec_status
) == UNIT_SRC_NORMAL
)
1187 printf("Contains Data\n");
1188 printf(" Current Unit : %u\n",
1189 __le32_to_cpu(migr_rec
->curr_migr_unit
));
1190 printf(" Family : %u\n",
1191 __le32_to_cpu(migr_rec
->family_num
));
1192 printf(" Ascending : %u\n",
1193 __le32_to_cpu(migr_rec
->ascending_migr
));
1194 printf(" Blocks Per Unit : %u\n",
1195 __le32_to_cpu(migr_rec
->blocks_per_unit
));
1196 printf(" Dest. Depth Per Unit : %u\n",
1197 __le32_to_cpu(migr_rec
->dest_depth_per_unit
));
1198 printf(" Checkpoint Area pba : %u\n",
1199 __le32_to_cpu(migr_rec
->ckpt_area_pba
));
1200 printf(" First member lba : %u\n",
1201 __le32_to_cpu(migr_rec
->dest_1st_member_lba
));
1202 printf(" Total Number of Units : %u\n",
1203 __le32_to_cpu(migr_rec
->num_migr_units
));
1204 printf(" Size of volume : %u\n",
1205 __le32_to_cpu(migr_rec
->post_migr_vol_cap
));
1206 printf(" Expansion space for LBA64 : %u\n",
1207 __le32_to_cpu(migr_rec
->post_migr_vol_cap_hi
));
1208 printf(" Record was read from : %u\n",
1209 __le32_to_cpu(migr_rec
->ckpt_read_disk_num
));
1214 #endif /* MDASSEMBLE */
1215 /*******************************************************************************
1216 * function: imsm_check_attributes
1217 * Description: Function checks if features represented by attributes flags
1218 * are supported by mdadm.
1220 * attributes - Attributes read from metadata
1222 * 0 - passed attributes contains unsupported features flags
1223 * 1 - all features are supported
1224 ******************************************************************************/
1225 static int imsm_check_attributes(__u32 attributes
)
1228 __u32 not_supported
= MPB_ATTRIB_SUPPORTED
^0xffffffff;
1230 not_supported
&= ~MPB_ATTRIB_IGNORED
;
1232 not_supported
&= attributes
;
1233 if (not_supported
) {
1234 fprintf(stderr
, Name
"(IMSM): Unsupported attributes : %x\n",
1235 (unsigned)__le32_to_cpu(not_supported
));
1236 if (not_supported
& MPB_ATTRIB_CHECKSUM_VERIFY
) {
1237 dprintf("\t\tMPB_ATTRIB_CHECKSUM_VERIFY \n");
1238 not_supported
^= MPB_ATTRIB_CHECKSUM_VERIFY
;
1240 if (not_supported
& MPB_ATTRIB_2TB
) {
1241 dprintf("\t\tMPB_ATTRIB_2TB\n");
1242 not_supported
^= MPB_ATTRIB_2TB
;
1244 if (not_supported
& MPB_ATTRIB_RAID0
) {
1245 dprintf("\t\tMPB_ATTRIB_RAID0\n");
1246 not_supported
^= MPB_ATTRIB_RAID0
;
1248 if (not_supported
& MPB_ATTRIB_RAID1
) {
1249 dprintf("\t\tMPB_ATTRIB_RAID1\n");
1250 not_supported
^= MPB_ATTRIB_RAID1
;
1252 if (not_supported
& MPB_ATTRIB_RAID10
) {
1253 dprintf("\t\tMPB_ATTRIB_RAID10\n");
1254 not_supported
^= MPB_ATTRIB_RAID10
;
1256 if (not_supported
& MPB_ATTRIB_RAID1E
) {
1257 dprintf("\t\tMPB_ATTRIB_RAID1E\n");
1258 not_supported
^= MPB_ATTRIB_RAID1E
;
1260 if (not_supported
& MPB_ATTRIB_RAID5
) {
1261 dprintf("\t\tMPB_ATTRIB_RAID5\n");
1262 not_supported
^= MPB_ATTRIB_RAID5
;
1264 if (not_supported
& MPB_ATTRIB_RAIDCNG
) {
1265 dprintf("\t\tMPB_ATTRIB_RAIDCNG\n");
1266 not_supported
^= MPB_ATTRIB_RAIDCNG
;
1268 if (not_supported
& MPB_ATTRIB_BBM
) {
1269 dprintf("\t\tMPB_ATTRIB_BBM\n");
1270 not_supported
^= MPB_ATTRIB_BBM
;
1272 if (not_supported
& MPB_ATTRIB_CHECKSUM_VERIFY
) {
1273 dprintf("\t\tMPB_ATTRIB_CHECKSUM_VERIFY (== MPB_ATTRIB_LEGACY)\n");
1274 not_supported
^= MPB_ATTRIB_CHECKSUM_VERIFY
;
1276 if (not_supported
& MPB_ATTRIB_EXP_STRIPE_SIZE
) {
1277 dprintf("\t\tMPB_ATTRIB_EXP_STRIP_SIZE\n");
1278 not_supported
^= MPB_ATTRIB_EXP_STRIPE_SIZE
;
1280 if (not_supported
& MPB_ATTRIB_2TB_DISK
) {
1281 dprintf("\t\tMPB_ATTRIB_2TB_DISK\n");
1282 not_supported
^= MPB_ATTRIB_2TB_DISK
;
1284 if (not_supported
& MPB_ATTRIB_NEVER_USE2
) {
1285 dprintf("\t\tMPB_ATTRIB_NEVER_USE2\n");
1286 not_supported
^= MPB_ATTRIB_NEVER_USE2
;
1288 if (not_supported
& MPB_ATTRIB_NEVER_USE
) {
1289 dprintf("\t\tMPB_ATTRIB_NEVER_USE\n");
1290 not_supported
^= MPB_ATTRIB_NEVER_USE
;
1294 dprintf(Name
"(IMSM): Unknown attributes : %x\n", not_supported
);
1303 static void getinfo_super_imsm(struct supertype
*st
, struct mdinfo
*info
, char *map
);
1305 static void examine_super_imsm(struct supertype
*st
, char *homehost
)
1307 struct intel_super
*super
= st
->sb
;
1308 struct imsm_super
*mpb
= super
->anchor
;
1309 char str
[MAX_SIGNATURE_LENGTH
];
1314 __u32 reserved
= imsm_reserved_sectors(super
, super
->disks
);
1317 snprintf(str
, MPB_SIG_LEN
, "%s", mpb
->sig
);
1318 printf(" Magic : %s\n", str
);
1319 snprintf(str
, strlen(MPB_VERSION_RAID0
), "%s", get_imsm_version(mpb
));
1320 printf(" Version : %s\n", get_imsm_version(mpb
));
1321 printf(" Orig Family : %08x\n", __le32_to_cpu(mpb
->orig_family_num
));
1322 printf(" Family : %08x\n", __le32_to_cpu(mpb
->family_num
));
1323 printf(" Generation : %08x\n", __le32_to_cpu(mpb
->generation_num
));
1324 printf(" Attributes : ");
1325 if (imsm_check_attributes(mpb
->attributes
))
1326 printf("All supported\n");
1328 printf("not supported\n");
1329 getinfo_super_imsm(st
, &info
, NULL
);
1330 fname_from_uuid(st
, &info
, nbuf
, ':');
1331 printf(" UUID : %s\n", nbuf
+ 5);
1332 sum
= __le32_to_cpu(mpb
->check_sum
);
1333 printf(" Checksum : %08x %s\n", sum
,
1334 __gen_imsm_checksum(mpb
) == sum
? "correct" : "incorrect");
1335 printf(" MPB Sectors : %d\n", mpb_sectors(mpb
));
1336 printf(" Disks : %d\n", mpb
->num_disks
);
1337 printf(" RAID Devices : %d\n", mpb
->num_raid_devs
);
1338 print_imsm_disk(__get_imsm_disk(mpb
, super
->disks
->index
), super
->disks
->index
, reserved
);
1339 if (super
->bbm_log
) {
1340 struct bbm_log
*log
= super
->bbm_log
;
1343 printf("Bad Block Management Log:\n");
1344 printf(" Log Size : %d\n", __le32_to_cpu(mpb
->bbm_log_size
));
1345 printf(" Signature : %x\n", __le32_to_cpu(log
->signature
));
1346 printf(" Entry Count : %d\n", __le32_to_cpu(log
->entry_count
));
1347 printf(" Spare Blocks : %d\n", __le32_to_cpu(log
->reserved_spare_block_count
));
1348 printf(" First Spare : %llx\n",
1349 (unsigned long long) __le64_to_cpu(log
->first_spare_lba
));
1351 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
1353 struct imsm_dev
*dev
= __get_imsm_dev(mpb
, i
);
1355 super
->current_vol
= i
;
1356 getinfo_super_imsm(st
, &info
, NULL
);
1357 fname_from_uuid(st
, &info
, nbuf
, ':');
1358 print_imsm_dev(super
, dev
, nbuf
+ 5, super
->disks
->index
);
1360 for (i
= 0; i
< mpb
->num_disks
; i
++) {
1361 if (i
== super
->disks
->index
)
1363 print_imsm_disk(__get_imsm_disk(mpb
, i
), i
, reserved
);
1366 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
1367 if (dl
->index
== -1)
1368 print_imsm_disk(&dl
->disk
, -1, reserved
);
1370 examine_migr_rec_imsm(super
);
1373 static void brief_examine_super_imsm(struct supertype
*st
, int verbose
)
1375 /* We just write a generic IMSM ARRAY entry */
1378 struct intel_super
*super
= st
->sb
;
1380 if (!super
->anchor
->num_raid_devs
) {
1381 printf("ARRAY metadata=imsm\n");
1385 getinfo_super_imsm(st
, &info
, NULL
);
1386 fname_from_uuid(st
, &info
, nbuf
, ':');
1387 printf("ARRAY metadata=imsm UUID=%s\n", nbuf
+ 5);
1390 static void brief_examine_subarrays_imsm(struct supertype
*st
, int verbose
)
1392 /* We just write a generic IMSM ARRAY entry */
1396 struct intel_super
*super
= st
->sb
;
1399 if (!super
->anchor
->num_raid_devs
)
1402 getinfo_super_imsm(st
, &info
, NULL
);
1403 fname_from_uuid(st
, &info
, nbuf
, ':');
1404 for (i
= 0; i
< super
->anchor
->num_raid_devs
; i
++) {
1405 struct imsm_dev
*dev
= get_imsm_dev(super
, i
);
1407 super
->current_vol
= i
;
1408 getinfo_super_imsm(st
, &info
, NULL
);
1409 fname_from_uuid(st
, &info
, nbuf1
, ':');
1410 printf("ARRAY /dev/md/%.16s container=%s member=%d UUID=%s\n",
1411 dev
->volume
, nbuf
+ 5, i
, nbuf1
+ 5);
1415 static void export_examine_super_imsm(struct supertype
*st
)
1417 struct intel_super
*super
= st
->sb
;
1418 struct imsm_super
*mpb
= super
->anchor
;
1422 getinfo_super_imsm(st
, &info
, NULL
);
1423 fname_from_uuid(st
, &info
, nbuf
, ':');
1424 printf("MD_METADATA=imsm\n");
1425 printf("MD_LEVEL=container\n");
1426 printf("MD_UUID=%s\n", nbuf
+5);
1427 printf("MD_DEVICES=%u\n", mpb
->num_disks
);
1430 static void detail_super_imsm(struct supertype
*st
, char *homehost
)
1435 getinfo_super_imsm(st
, &info
, NULL
);
1436 fname_from_uuid(st
, &info
, nbuf
, ':');
1437 printf("\n UUID : %s\n", nbuf
+ 5);
1440 static void brief_detail_super_imsm(struct supertype
*st
)
1444 getinfo_super_imsm(st
, &info
, NULL
);
1445 fname_from_uuid(st
, &info
, nbuf
, ':');
1446 printf(" UUID=%s", nbuf
+ 5);
1449 static int imsm_read_serial(int fd
, char *devname
, __u8
*serial
);
1450 static void fd2devname(int fd
, char *name
);
1452 static int ahci_enumerate_ports(const char *hba_path
, int port_count
, int host_base
, int verbose
)
1454 /* dump an unsorted list of devices attached to AHCI Intel storage
1455 * controller, as well as non-connected ports
1457 int hba_len
= strlen(hba_path
) + 1;
1462 unsigned long port_mask
= (1 << port_count
) - 1;
1464 if (port_count
> (int)sizeof(port_mask
) * 8) {
1466 fprintf(stderr
, Name
": port_count %d out of range\n", port_count
);
1470 /* scroll through /sys/dev/block looking for devices attached to
1473 dir
= opendir("/sys/dev/block");
1474 for (ent
= dir
? readdir(dir
) : NULL
; ent
; ent
= readdir(dir
)) {
1485 if (sscanf(ent
->d_name
, "%d:%d", &major
, &minor
) != 2)
1487 path
= devt_to_devpath(makedev(major
, minor
));
1490 if (!path_attached_to_hba(path
, hba_path
)) {
1496 /* retrieve the scsi device type */
1497 if (asprintf(&device
, "/sys/dev/block/%d:%d/device/xxxxxxx", major
, minor
) < 0) {
1499 fprintf(stderr
, Name
": failed to allocate 'device'\n");
1503 sprintf(device
, "/sys/dev/block/%d:%d/device/type", major
, minor
);
1504 if (load_sys(device
, buf
) != 0) {
1506 fprintf(stderr
, Name
": failed to read device type for %s\n",
1512 type
= strtoul(buf
, NULL
, 10);
1514 /* if it's not a disk print the vendor and model */
1515 if (!(type
== 0 || type
== 7 || type
== 14)) {
1518 sprintf(device
, "/sys/dev/block/%d:%d/device/vendor", major
, minor
);
1519 if (load_sys(device
, buf
) == 0) {
1520 strncpy(vendor
, buf
, sizeof(vendor
));
1521 vendor
[sizeof(vendor
) - 1] = '\0';
1522 c
= (char *) &vendor
[sizeof(vendor
) - 1];
1523 while (isspace(*c
) || *c
== '\0')
1527 sprintf(device
, "/sys/dev/block/%d:%d/device/model", major
, minor
);
1528 if (load_sys(device
, buf
) == 0) {
1529 strncpy(model
, buf
, sizeof(model
));
1530 model
[sizeof(model
) - 1] = '\0';
1531 c
= (char *) &model
[sizeof(model
) - 1];
1532 while (isspace(*c
) || *c
== '\0')
1536 if (vendor
[0] && model
[0])
1537 sprintf(buf
, "%.64s %.64s", vendor
, model
);
1539 switch (type
) { /* numbers from hald/linux/device.c */
1540 case 1: sprintf(buf
, "tape"); break;
1541 case 2: sprintf(buf
, "printer"); break;
1542 case 3: sprintf(buf
, "processor"); break;
1544 case 5: sprintf(buf
, "cdrom"); break;
1545 case 6: sprintf(buf
, "scanner"); break;
1546 case 8: sprintf(buf
, "media_changer"); break;
1547 case 9: sprintf(buf
, "comm"); break;
1548 case 12: sprintf(buf
, "raid"); break;
1549 default: sprintf(buf
, "unknown");
1555 /* chop device path to 'host%d' and calculate the port number */
1556 c
= strchr(&path
[hba_len
], '/');
1559 fprintf(stderr
, Name
": %s - invalid path name\n", path
+ hba_len
);
1564 if (sscanf(&path
[hba_len
], "host%d", &port
) == 1)
1568 *c
= '/'; /* repair the full string */
1569 fprintf(stderr
, Name
": failed to determine port number for %s\n",
1576 /* mark this port as used */
1577 port_mask
&= ~(1 << port
);
1579 /* print out the device information */
1581 printf(" Port%d : - non-disk device (%s) -\n", port
, buf
);
1585 fd
= dev_open(ent
->d_name
, O_RDONLY
);
1587 printf(" Port%d : - disk info unavailable -\n", port
);
1589 fd2devname(fd
, buf
);
1590 printf(" Port%d : %s", port
, buf
);
1591 if (imsm_read_serial(fd
, NULL
, (__u8
*) buf
) == 0)
1592 printf(" (%.*s)\n", MAX_RAID_SERIAL_LEN
, buf
);
1607 for (i
= 0; i
< port_count
; i
++)
1608 if (port_mask
& (1 << i
))
1609 printf(" Port%d : - no device attached -\n", i
);
1615 static void print_found_intel_controllers(struct sys_dev
*elem
)
1617 for (; elem
; elem
= elem
->next
) {
1618 fprintf(stderr
, Name
": found Intel(R) ");
1619 if (elem
->type
== SYS_DEV_SATA
)
1620 fprintf(stderr
, "SATA ");
1621 else if (elem
->type
== SYS_DEV_SAS
)
1622 fprintf(stderr
, "SAS ");
1623 fprintf(stderr
, "RAID controller");
1625 fprintf(stderr
, " at %s", elem
->pci_id
);
1626 fprintf(stderr
, ".\n");
1631 static int ahci_get_port_count(const char *hba_path
, int *port_count
)
1638 if ((dir
= opendir(hba_path
)) == NULL
)
1641 for (ent
= readdir(dir
); ent
; ent
= readdir(dir
)) {
1644 if (sscanf(ent
->d_name
, "host%d", &host
) != 1)
1646 if (*port_count
== 0)
1648 else if (host
< host_base
)
1651 if (host
+ 1 > *port_count
+ host_base
)
1652 *port_count
= host
+ 1 - host_base
;
1658 static void print_imsm_capability(const struct imsm_orom
*orom
)
1660 printf(" Platform : Intel(R) Matrix Storage Manager\n");
1661 printf(" Version : %d.%d.%d.%d\n", orom
->major_ver
, orom
->minor_ver
,
1662 orom
->hotfix_ver
, orom
->build
);
1663 printf(" RAID Levels :%s%s%s%s%s\n",
1664 imsm_orom_has_raid0(orom
) ? " raid0" : "",
1665 imsm_orom_has_raid1(orom
) ? " raid1" : "",
1666 imsm_orom_has_raid1e(orom
) ? " raid1e" : "",
1667 imsm_orom_has_raid10(orom
) ? " raid10" : "",
1668 imsm_orom_has_raid5(orom
) ? " raid5" : "");
1669 printf(" Chunk Sizes :%s%s%s%s%s%s%s%s%s%s%s%s%s%s%s%s\n",
1670 imsm_orom_has_chunk(orom
, 2) ? " 2k" : "",
1671 imsm_orom_has_chunk(orom
, 4) ? " 4k" : "",
1672 imsm_orom_has_chunk(orom
, 8) ? " 8k" : "",
1673 imsm_orom_has_chunk(orom
, 16) ? " 16k" : "",
1674 imsm_orom_has_chunk(orom
, 32) ? " 32k" : "",
1675 imsm_orom_has_chunk(orom
, 64) ? " 64k" : "",
1676 imsm_orom_has_chunk(orom
, 128) ? " 128k" : "",
1677 imsm_orom_has_chunk(orom
, 256) ? " 256k" : "",
1678 imsm_orom_has_chunk(orom
, 512) ? " 512k" : "",
1679 imsm_orom_has_chunk(orom
, 1024*1) ? " 1M" : "",
1680 imsm_orom_has_chunk(orom
, 1024*2) ? " 2M" : "",
1681 imsm_orom_has_chunk(orom
, 1024*4) ? " 4M" : "",
1682 imsm_orom_has_chunk(orom
, 1024*8) ? " 8M" : "",
1683 imsm_orom_has_chunk(orom
, 1024*16) ? " 16M" : "",
1684 imsm_orom_has_chunk(orom
, 1024*32) ? " 32M" : "",
1685 imsm_orom_has_chunk(orom
, 1024*64) ? " 64M" : "");
1686 printf(" Max Disks : %d\n", orom
->tds
);
1687 printf(" Max Volumes : %d\n", orom
->vpa
);
1691 static int detail_platform_imsm(int verbose
, int enumerate_only
)
1693 /* There are two components to imsm platform support, the ahci SATA
1694 * controller and the option-rom. To find the SATA controller we
1695 * simply look in /sys/bus/pci/drivers/ahci to see if an ahci
1696 * controller with the Intel vendor id is present. This approach
1697 * allows mdadm to leverage the kernel's ahci detection logic, with the
1698 * caveat that if ahci.ko is not loaded mdadm will not be able to
1699 * detect platform raid capabilities. The option-rom resides in a
1700 * platform "Adapter ROM". We scan for its signature to retrieve the
1701 * platform capabilities. If raid support is disabled in the BIOS the
1702 * option-rom capability structure will not be available.
1704 const struct imsm_orom
*orom
;
1705 struct sys_dev
*list
, *hba
;
1710 if (enumerate_only
) {
1711 if (check_env("IMSM_NO_PLATFORM"))
1713 list
= find_intel_devices();
1716 for (hba
= list
; hba
; hba
= hba
->next
) {
1717 orom
= find_imsm_capability(hba
->type
);
1723 free_sys_dev(&list
);
1727 list
= find_intel_devices();
1730 fprintf(stderr
, Name
": no active Intel(R) RAID "
1731 "controller found.\n");
1732 free_sys_dev(&list
);
1735 print_found_intel_controllers(list
);
1737 for (hba
= list
; hba
; hba
= hba
->next
) {
1738 orom
= find_imsm_capability(hba
->type
);
1740 fprintf(stderr
, Name
": imsm capabilities not found for controller: %s (type %s)\n",
1741 hba
->path
, get_sys_dev_type(hba
->type
));
1743 print_imsm_capability(orom
);
1746 for (hba
= list
; hba
; hba
= hba
->next
) {
1747 printf(" I/O Controller : %s (%s)\n",
1748 hba
->path
, get_sys_dev_type(hba
->type
));
1750 if (hba
->type
== SYS_DEV_SATA
) {
1751 host_base
= ahci_get_port_count(hba
->path
, &port_count
);
1752 if (ahci_enumerate_ports(hba
->path
, port_count
, host_base
, verbose
)) {
1754 fprintf(stderr
, Name
": failed to enumerate "
1755 "ports on SATA controller at %s.", hba
->pci_id
);
1761 free_sys_dev(&list
);
1766 static int match_home_imsm(struct supertype
*st
, char *homehost
)
1768 /* the imsm metadata format does not specify any host
1769 * identification information. We return -1 since we can never
1770 * confirm nor deny whether a given array is "meant" for this
1771 * host. We rely on compare_super and the 'family_num' fields to
1772 * exclude member disks that do not belong, and we rely on
1773 * mdadm.conf to specify the arrays that should be assembled.
1774 * Auto-assembly may still pick up "foreign" arrays.
1780 static void uuid_from_super_imsm(struct supertype
*st
, int uuid
[4])
1782 /* The uuid returned here is used for:
1783 * uuid to put into bitmap file (Create, Grow)
1784 * uuid for backup header when saving critical section (Grow)
1785 * comparing uuids when re-adding a device into an array
1786 * In these cases the uuid required is that of the data-array,
1787 * not the device-set.
1788 * uuid to recognise same set when adding a missing device back
1789 * to an array. This is a uuid for the device-set.
1791 * For each of these we can make do with a truncated
1792 * or hashed uuid rather than the original, as long as
1794 * In each case the uuid required is that of the data-array,
1795 * not the device-set.
1797 /* imsm does not track uuid's so we synthesis one using sha1 on
1798 * - The signature (Which is constant for all imsm array, but no matter)
1799 * - the orig_family_num of the container
1800 * - the index number of the volume
1801 * - the 'serial' number of the volume.
1802 * Hopefully these are all constant.
1804 struct intel_super
*super
= st
->sb
;
1807 struct sha1_ctx ctx
;
1808 struct imsm_dev
*dev
= NULL
;
1811 /* some mdadm versions failed to set ->orig_family_num, in which
1812 * case fall back to ->family_num. orig_family_num will be
1813 * fixed up with the first metadata update.
1815 family_num
= super
->anchor
->orig_family_num
;
1816 if (family_num
== 0)
1817 family_num
= super
->anchor
->family_num
;
1818 sha1_init_ctx(&ctx
);
1819 sha1_process_bytes(super
->anchor
->sig
, MPB_SIG_LEN
, &ctx
);
1820 sha1_process_bytes(&family_num
, sizeof(__u32
), &ctx
);
1821 if (super
->current_vol
>= 0)
1822 dev
= get_imsm_dev(super
, super
->current_vol
);
1824 __u32 vol
= super
->current_vol
;
1825 sha1_process_bytes(&vol
, sizeof(vol
), &ctx
);
1826 sha1_process_bytes(dev
->volume
, MAX_RAID_SERIAL_LEN
, &ctx
);
1828 sha1_finish_ctx(&ctx
, buf
);
1829 memcpy(uuid
, buf
, 4*4);
1834 get_imsm_numerical_version(struct imsm_super
*mpb
, int *m
, int *p
)
1836 __u8
*v
= get_imsm_version(mpb
);
1837 __u8
*end
= mpb
->sig
+ MAX_SIGNATURE_LENGTH
;
1838 char major
[] = { 0, 0, 0 };
1839 char minor
[] = { 0 ,0, 0 };
1840 char patch
[] = { 0, 0, 0 };
1841 char *ver_parse
[] = { major
, minor
, patch
};
1845 while (*v
!= '\0' && v
< end
) {
1846 if (*v
!= '.' && j
< 2)
1847 ver_parse
[i
][j
++] = *v
;
1855 *m
= strtol(minor
, NULL
, 0);
1856 *p
= strtol(patch
, NULL
, 0);
1860 static __u32
migr_strip_blocks_resync(struct imsm_dev
*dev
)
1862 /* migr_strip_size when repairing or initializing parity */
1863 struct imsm_map
*map
= get_imsm_map(dev
, 0);
1864 __u32 chunk
= __le32_to_cpu(map
->blocks_per_strip
);
1866 switch (get_imsm_raid_level(map
)) {
1871 return 128*1024 >> 9;
1875 static __u32
migr_strip_blocks_rebuild(struct imsm_dev
*dev
)
1877 /* migr_strip_size when rebuilding a degraded disk, no idea why
1878 * this is different than migr_strip_size_resync(), but it's good
1881 struct imsm_map
*map
= get_imsm_map(dev
, 1);
1882 __u32 chunk
= __le32_to_cpu(map
->blocks_per_strip
);
1884 switch (get_imsm_raid_level(map
)) {
1887 if (map
->num_members
% map
->num_domains
== 0)
1888 return 128*1024 >> 9;
1892 return max((__u32
) 64*1024 >> 9, chunk
);
1894 return 128*1024 >> 9;
1898 static __u32
num_stripes_per_unit_resync(struct imsm_dev
*dev
)
1900 struct imsm_map
*lo
= get_imsm_map(dev
, 0);
1901 struct imsm_map
*hi
= get_imsm_map(dev
, 1);
1902 __u32 lo_chunk
= __le32_to_cpu(lo
->blocks_per_strip
);
1903 __u32 hi_chunk
= __le32_to_cpu(hi
->blocks_per_strip
);
1905 return max((__u32
) 1, hi_chunk
/ lo_chunk
);
1908 static __u32
num_stripes_per_unit_rebuild(struct imsm_dev
*dev
)
1910 struct imsm_map
*lo
= get_imsm_map(dev
, 0);
1911 int level
= get_imsm_raid_level(lo
);
1913 if (level
== 1 || level
== 10) {
1914 struct imsm_map
*hi
= get_imsm_map(dev
, 1);
1916 return hi
->num_domains
;
1918 return num_stripes_per_unit_resync(dev
);
1921 static __u8
imsm_num_data_members(struct imsm_dev
*dev
, int second_map
)
1923 /* named 'imsm_' because raid0, raid1 and raid10
1924 * counter-intuitively have the same number of data disks
1926 struct imsm_map
*map
= get_imsm_map(dev
, second_map
);
1928 switch (get_imsm_raid_level(map
)) {
1932 return map
->num_members
;
1934 return map
->num_members
- 1;
1936 dprintf("%s: unsupported raid level\n", __func__
);
1941 static __u32
parity_segment_depth(struct imsm_dev
*dev
)
1943 struct imsm_map
*map
= get_imsm_map(dev
, 0);
1944 __u32 chunk
= __le32_to_cpu(map
->blocks_per_strip
);
1946 switch(get_imsm_raid_level(map
)) {
1949 return chunk
* map
->num_domains
;
1951 return chunk
* map
->num_members
;
1957 static __u32
map_migr_block(struct imsm_dev
*dev
, __u32 block
)
1959 struct imsm_map
*map
= get_imsm_map(dev
, 1);
1960 __u32 chunk
= __le32_to_cpu(map
->blocks_per_strip
);
1961 __u32 strip
= block
/ chunk
;
1963 switch (get_imsm_raid_level(map
)) {
1966 __u32 vol_strip
= (strip
* map
->num_domains
) + 1;
1967 __u32 vol_stripe
= vol_strip
/ map
->num_members
;
1969 return vol_stripe
* chunk
+ block
% chunk
;
1971 __u32 stripe
= strip
/ (map
->num_members
- 1);
1973 return stripe
* chunk
+ block
% chunk
;
1980 static __u64
blocks_per_migr_unit(struct intel_super
*super
,
1981 struct imsm_dev
*dev
)
1983 /* calculate the conversion factor between per member 'blocks'
1984 * (md/{resync,rebuild}_start) and imsm migration units, return
1985 * 0 for the 'not migrating' and 'unsupported migration' cases
1987 if (!dev
->vol
.migr_state
)
1990 switch (migr_type(dev
)) {
1991 case MIGR_GEN_MIGR
: {
1992 struct migr_record
*migr_rec
= super
->migr_rec
;
1993 return __le32_to_cpu(migr_rec
->blocks_per_unit
);
1998 struct imsm_map
*map
= get_imsm_map(dev
, 0);
1999 __u32 stripes_per_unit
;
2000 __u32 blocks_per_unit
;
2009 /* yes, this is really the translation of migr_units to
2010 * per-member blocks in the 'resync' case
2012 stripes_per_unit
= num_stripes_per_unit_resync(dev
);
2013 migr_chunk
= migr_strip_blocks_resync(dev
);
2014 disks
= imsm_num_data_members(dev
, 0);
2015 blocks_per_unit
= stripes_per_unit
* migr_chunk
* disks
;
2016 stripe
= __le16_to_cpu(map
->blocks_per_strip
) * disks
;
2017 segment
= blocks_per_unit
/ stripe
;
2018 block_rel
= blocks_per_unit
- segment
* stripe
;
2019 parity_depth
= parity_segment_depth(dev
);
2020 block_map
= map_migr_block(dev
, block_rel
);
2021 return block_map
+ parity_depth
* segment
;
2023 case MIGR_REBUILD
: {
2024 __u32 stripes_per_unit
;
2027 stripes_per_unit
= num_stripes_per_unit_rebuild(dev
);
2028 migr_chunk
= migr_strip_blocks_rebuild(dev
);
2029 return migr_chunk
* stripes_per_unit
;
2031 case MIGR_STATE_CHANGE
:
2037 static int imsm_level_to_layout(int level
)
2045 return ALGORITHM_LEFT_ASYMMETRIC
;
2052 /*******************************************************************************
2053 * Function: read_imsm_migr_rec
2054 * Description: Function reads imsm migration record from last sector of disk
2056 * fd : disk descriptor
2057 * super : metadata info
2061 ******************************************************************************/
2062 static int read_imsm_migr_rec(int fd
, struct intel_super
*super
)
2065 unsigned long long dsize
;
2067 get_dev_size(fd
, NULL
, &dsize
);
2068 if (lseek64(fd
, dsize
- 512, SEEK_SET
) < 0) {
2070 Name
": Cannot seek to anchor block: %s\n",
2074 if (read(fd
, super
->migr_rec_buf
, 512) != 512) {
2076 Name
": Cannot read migr record block: %s\n",
2086 /*******************************************************************************
2087 * Function: load_imsm_migr_rec
2088 * Description: Function reads imsm migration record (it is stored at the last
2091 * super : imsm internal array info
2092 * info : general array info
2096 ******************************************************************************/
2097 static int load_imsm_migr_rec(struct intel_super
*super
, struct mdinfo
*info
)
2100 struct dl
*dl
= NULL
;
2106 for (sd
= info
->devs
; sd
; sd
= sd
->next
) {
2107 /* read only from one of the first two slots */
2108 if ((sd
->disk
.raid_disk
> 1) ||
2109 (sd
->disk
.raid_disk
< 0))
2111 sprintf(nm
, "%d:%d", sd
->disk
.major
, sd
->disk
.minor
);
2112 fd
= dev_open(nm
, O_RDONLY
);
2118 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
2119 /* read only from one of the first two slots */
2122 sprintf(nm
, "%d:%d", dl
->major
, dl
->minor
);
2123 fd
= dev_open(nm
, O_RDONLY
);
2130 retval
= read_imsm_migr_rec(fd
, super
);
2139 /*******************************************************************************
2140 * function: imsm_create_metadata_checkpoint_update
2141 * Description: It creates update for checkpoint change.
2143 * super : imsm internal array info
2144 * u : pointer to prepared update
2147 * If length is equal to 0, input pointer u contains no update
2148 ******************************************************************************/
2149 static int imsm_create_metadata_checkpoint_update(
2150 struct intel_super
*super
,
2151 struct imsm_update_general_migration_checkpoint
**u
)
2154 int update_memory_size
= 0;
2156 dprintf("imsm_create_metadata_checkpoint_update(enter)\n");
2162 /* size of all update data without anchor */
2163 update_memory_size
=
2164 sizeof(struct imsm_update_general_migration_checkpoint
);
2166 *u
= calloc(1, update_memory_size
);
2168 dprintf("error: cannot get memory for "
2169 "imsm_create_metadata_checkpoint_update update\n");
2172 (*u
)->type
= update_general_migration_checkpoint
;
2173 (*u
)->curr_migr_unit
= __le32_to_cpu(super
->migr_rec
->curr_migr_unit
);
2174 dprintf("imsm_create_metadata_checkpoint_update: prepared for %u\n",
2175 (*u
)->curr_migr_unit
);
2177 return update_memory_size
;
2181 static void imsm_update_metadata_locally(struct supertype
*st
,
2182 void *buf
, int len
);
2184 /*******************************************************************************
2185 * Function: write_imsm_migr_rec
2186 * Description: Function writes imsm migration record
2187 * (at the last sector of disk)
2189 * super : imsm internal array info
2193 ******************************************************************************/
2194 static int write_imsm_migr_rec(struct supertype
*st
)
2196 struct intel_super
*super
= st
->sb
;
2197 unsigned long long dsize
;
2203 struct imsm_update_general_migration_checkpoint
*u
;
2205 for (sd
= super
->disks
; sd
; sd
= sd
->next
) {
2206 /* write to 2 first slots only */
2207 if ((sd
->index
< 0) || (sd
->index
> 1))
2209 sprintf(nm
, "%d:%d", sd
->major
, sd
->minor
);
2210 fd
= dev_open(nm
, O_RDWR
);
2213 get_dev_size(fd
, NULL
, &dsize
);
2214 if (lseek64(fd
, dsize
- 512, SEEK_SET
) < 0) {
2216 Name
": Cannot seek to anchor block: %s\n",
2220 if (write(fd
, super
->migr_rec_buf
, 512) != 512) {
2222 Name
": Cannot write migr record block: %s\n",
2229 /* update checkpoint information in metadata */
2230 len
= imsm_create_metadata_checkpoint_update(super
, &u
);
2233 dprintf("imsm: Cannot prepare update\n");
2236 /* update metadata locally */
2237 imsm_update_metadata_locally(st
, u
, len
);
2238 /* and possibly remotely */
2239 if (st
->update_tail
) {
2240 append_metadata_update(st
, u
, len
);
2241 /* during reshape we do all work inside metadata handler
2242 * manage_reshape(), so metadata update has to be triggered
2245 flush_metadata_updates(st
);
2246 st
->update_tail
= &st
->updates
;
2256 #endif /* MDASSEMBLE */
2258 /* spare/missing disks activations are not allowe when
2259 * array/container performs reshape operation, because
2260 * all arrays in container works on the same disks set
2262 int imsm_reshape_blocks_arrays_changes(struct intel_super
*super
)
2265 struct intel_dev
*i_dev
;
2266 struct imsm_dev
*dev
;
2268 /* check whole container
2270 for (i_dev
= super
->devlist
; i_dev
; i_dev
= i_dev
->next
) {
2272 if (is_gen_migration(dev
)) {
2273 /* No repair during any migration in container
2282 static void getinfo_super_imsm_volume(struct supertype
*st
, struct mdinfo
*info
, char *dmap
)
2284 struct intel_super
*super
= st
->sb
;
2285 struct migr_record
*migr_rec
= super
->migr_rec
;
2286 struct imsm_dev
*dev
= get_imsm_dev(super
, super
->current_vol
);
2287 struct imsm_map
*map
= get_imsm_map(dev
, 0);
2288 struct imsm_map
*prev_map
= get_imsm_map(dev
, 1);
2289 struct imsm_map
*map_to_analyse
= map
;
2292 unsigned int component_size_alligment
;
2293 int map_disks
= info
->array
.raid_disks
;
2295 memset(info
, 0, sizeof(*info
));
2297 map_to_analyse
= prev_map
;
2299 dl
= super
->current_disk
;
2301 info
->container_member
= super
->current_vol
;
2302 info
->array
.raid_disks
= map
->num_members
;
2303 info
->array
.level
= get_imsm_raid_level(map_to_analyse
);
2304 info
->array
.layout
= imsm_level_to_layout(info
->array
.level
);
2305 info
->array
.md_minor
= -1;
2306 info
->array
.ctime
= 0;
2307 info
->array
.utime
= 0;
2308 info
->array
.chunk_size
=
2309 __le16_to_cpu(map_to_analyse
->blocks_per_strip
) << 9;
2310 info
->array
.state
= !dev
->vol
.dirty
;
2311 info
->custom_array_size
= __le32_to_cpu(dev
->size_high
);
2312 info
->custom_array_size
<<= 32;
2313 info
->custom_array_size
|= __le32_to_cpu(dev
->size_low
);
2314 info
->recovery_blocked
= imsm_reshape_blocks_arrays_changes(st
->sb
);
2316 if (prev_map
&& map
->map_state
== prev_map
->map_state
&&
2317 (migr_type(dev
) == MIGR_GEN_MIGR
)) {
2318 info
->reshape_active
= 1;
2319 info
->new_level
= get_imsm_raid_level(map
);
2320 info
->new_layout
= imsm_level_to_layout(info
->new_level
);
2321 info
->new_chunk
= __le16_to_cpu(map
->blocks_per_strip
) << 9;
2322 info
->delta_disks
= map
->num_members
- prev_map
->num_members
;
2323 if (info
->delta_disks
) {
2324 /* this needs to be applied to every array
2327 info
->reshape_active
= CONTAINER_RESHAPE
;
2329 /* We shape information that we give to md might have to be
2330 * modify to cope with md's requirement for reshaping arrays.
2331 * For example, when reshaping a RAID0, md requires it to be
2332 * presented as a degraded RAID4.
2333 * Also if a RAID0 is migrating to a RAID5 we need to specify
2334 * the array as already being RAID5, but the 'before' layout
2335 * is a RAID4-like layout.
2337 switch (info
->array
.level
) {
2339 switch(info
->new_level
) {
2341 /* conversion is happening as RAID4 */
2342 info
->array
.level
= 4;
2343 info
->array
.raid_disks
+= 1;
2346 /* conversion is happening as RAID5 */
2347 info
->array
.level
= 5;
2348 info
->array
.layout
= ALGORITHM_PARITY_N
;
2349 info
->delta_disks
-= 1;
2352 /* FIXME error message */
2353 info
->array
.level
= UnSet
;
2359 info
->new_level
= UnSet
;
2360 info
->new_layout
= UnSet
;
2361 info
->new_chunk
= info
->array
.chunk_size
;
2362 info
->delta_disks
= 0;
2366 info
->disk
.major
= dl
->major
;
2367 info
->disk
.minor
= dl
->minor
;
2368 info
->disk
.number
= dl
->index
;
2369 info
->disk
.raid_disk
= get_imsm_disk_slot(map_to_analyse
,
2373 info
->data_offset
= __le32_to_cpu(map_to_analyse
->pba_of_lba0
);
2374 info
->component_size
=
2375 __le32_to_cpu(map_to_analyse
->blocks_per_member
);
2377 /* check component size aligment
2379 component_size_alligment
=
2380 info
->component_size
% (info
->array
.chunk_size
/512);
2382 if (component_size_alligment
&&
2383 (info
->array
.level
!= 1) && (info
->array
.level
!= UnSet
)) {
2384 dprintf("imsm: reported component size alligned from %llu ",
2385 info
->component_size
);
2386 info
->component_size
-= component_size_alligment
;
2387 dprintf("to %llu (%i).\n",
2388 info
->component_size
, component_size_alligment
);
2391 memset(info
->uuid
, 0, sizeof(info
->uuid
));
2392 info
->recovery_start
= MaxSector
;
2394 info
->reshape_progress
= 0;
2395 info
->resync_start
= MaxSector
;
2396 if ((map_to_analyse
->map_state
== IMSM_T_STATE_UNINITIALIZED
||
2398 imsm_reshape_blocks_arrays_changes(super
) == 0) {
2399 info
->resync_start
= 0;
2401 if (dev
->vol
.migr_state
) {
2402 switch (migr_type(dev
)) {
2405 __u64 blocks_per_unit
= blocks_per_migr_unit(super
,
2407 __u64 units
= __le32_to_cpu(dev
->vol
.curr_migr_unit
);
2409 info
->resync_start
= blocks_per_unit
* units
;
2412 case MIGR_GEN_MIGR
: {
2413 __u64 blocks_per_unit
= blocks_per_migr_unit(super
,
2415 __u64 units
= __le32_to_cpu(migr_rec
->curr_migr_unit
);
2416 unsigned long long array_blocks
;
2419 if (__le32_to_cpu(migr_rec
->ascending_migr
) &&
2421 (__le32_to_cpu(migr_rec
->num_migr_units
)-1)) &&
2422 (super
->migr_rec
->rec_status
==
2423 __cpu_to_le32(UNIT_SRC_IN_CP_AREA
)))
2426 info
->reshape_progress
= blocks_per_unit
* units
;
2428 dprintf("IMSM: General Migration checkpoint : %llu "
2429 "(%llu) -> read reshape progress : %llu\n",
2430 (unsigned long long)units
,
2431 (unsigned long long)blocks_per_unit
,
2432 info
->reshape_progress
);
2434 used_disks
= imsm_num_data_members(dev
, 1);
2435 if (used_disks
> 0) {
2436 array_blocks
= map
->blocks_per_member
*
2438 /* round array size down to closest MB
2440 info
->custom_array_size
= (array_blocks
2441 >> SECT_PER_MB_SHIFT
)
2442 << SECT_PER_MB_SHIFT
;
2446 /* we could emulate the checkpointing of
2447 * 'sync_action=check' migrations, but for now
2448 * we just immediately complete them
2451 /* this is handled by container_content_imsm() */
2452 case MIGR_STATE_CHANGE
:
2453 /* FIXME handle other migrations */
2455 /* we are not dirty, so... */
2456 info
->resync_start
= MaxSector
;
2460 strncpy(info
->name
, (char *) dev
->volume
, MAX_RAID_SERIAL_LEN
);
2461 info
->name
[MAX_RAID_SERIAL_LEN
] = 0;
2463 info
->array
.major_version
= -1;
2464 info
->array
.minor_version
= -2;
2465 devname
= devnum2devname(st
->container_dev
);
2466 *info
->text_version
= '\0';
2468 sprintf(info
->text_version
, "/%s/%d", devname
, info
->container_member
);
2470 info
->safe_mode_delay
= 4000; /* 4 secs like the Matrix driver */
2471 uuid_from_super_imsm(st
, info
->uuid
);
2475 for (i
=0; i
<map_disks
; i
++) {
2477 if (i
< info
->array
.raid_disks
) {
2478 struct imsm_disk
*dsk
;
2479 j
= get_imsm_disk_idx(dev
, i
, -1);
2480 dsk
= get_imsm_disk(super
, j
);
2481 if (dsk
&& (dsk
->status
& CONFIGURED_DISK
))
2488 static __u8
imsm_check_degraded(struct intel_super
*super
, struct imsm_dev
*dev
, int failed
);
2489 static int imsm_count_failed(struct intel_super
*super
, struct imsm_dev
*dev
);
2491 static struct imsm_disk
*get_imsm_missing(struct intel_super
*super
, __u8 index
)
2495 for (d
= super
->missing
; d
; d
= d
->next
)
2496 if (d
->index
== index
)
2501 static void getinfo_super_imsm(struct supertype
*st
, struct mdinfo
*info
, char *map
)
2503 struct intel_super
*super
= st
->sb
;
2504 struct imsm_disk
*disk
;
2505 int map_disks
= info
->array
.raid_disks
;
2506 int max_enough
= -1;
2508 struct imsm_super
*mpb
;
2510 if (super
->current_vol
>= 0) {
2511 getinfo_super_imsm_volume(st
, info
, map
);
2514 memset(info
, 0, sizeof(*info
));
2516 /* Set raid_disks to zero so that Assemble will always pull in valid
2519 info
->array
.raid_disks
= 0;
2520 info
->array
.level
= LEVEL_CONTAINER
;
2521 info
->array
.layout
= 0;
2522 info
->array
.md_minor
= -1;
2523 info
->array
.ctime
= 0; /* N/A for imsm */
2524 info
->array
.utime
= 0;
2525 info
->array
.chunk_size
= 0;
2527 info
->disk
.major
= 0;
2528 info
->disk
.minor
= 0;
2529 info
->disk
.raid_disk
= -1;
2530 info
->reshape_active
= 0;
2531 info
->array
.major_version
= -1;
2532 info
->array
.minor_version
= -2;
2533 strcpy(info
->text_version
, "imsm");
2534 info
->safe_mode_delay
= 0;
2535 info
->disk
.number
= -1;
2536 info
->disk
.state
= 0;
2538 info
->recovery_start
= MaxSector
;
2539 info
->recovery_blocked
= imsm_reshape_blocks_arrays_changes(st
->sb
);
2541 /* do we have the all the insync disks that we expect? */
2542 mpb
= super
->anchor
;
2544 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
2545 struct imsm_dev
*dev
= get_imsm_dev(super
, i
);
2546 int failed
, enough
, j
, missing
= 0;
2547 struct imsm_map
*map
;
2550 failed
= imsm_count_failed(super
, dev
);
2551 state
= imsm_check_degraded(super
, dev
, failed
);
2552 map
= get_imsm_map(dev
, 0);
2554 /* any newly missing disks?
2555 * (catches single-degraded vs double-degraded)
2557 for (j
= 0; j
< map
->num_members
; j
++) {
2558 __u32 ord
= get_imsm_ord_tbl_ent(dev
, j
, 0);
2559 __u32 idx
= ord_to_idx(ord
);
2561 if (!(ord
& IMSM_ORD_REBUILD
) &&
2562 get_imsm_missing(super
, idx
)) {
2568 if (state
== IMSM_T_STATE_FAILED
)
2570 else if (state
== IMSM_T_STATE_DEGRADED
&&
2571 (state
!= map
->map_state
|| missing
))
2573 else /* we're normal, or already degraded */
2576 /* in the missing/failed disk case check to see
2577 * if at least one array is runnable
2579 max_enough
= max(max_enough
, enough
);
2581 dprintf("%s: enough: %d\n", __func__
, max_enough
);
2582 info
->container_enough
= max_enough
;
2585 __u32 reserved
= imsm_reserved_sectors(super
, super
->disks
);
2587 disk
= &super
->disks
->disk
;
2588 info
->data_offset
= __le32_to_cpu(disk
->total_blocks
) - reserved
;
2589 info
->component_size
= reserved
;
2590 info
->disk
.state
= is_configured(disk
) ? (1 << MD_DISK_ACTIVE
) : 0;
2591 /* we don't change info->disk.raid_disk here because
2592 * this state will be finalized in mdmon after we have
2593 * found the 'most fresh' version of the metadata
2595 info
->disk
.state
|= is_failed(disk
) ? (1 << MD_DISK_FAULTY
) : 0;
2596 info
->disk
.state
|= is_spare(disk
) ? 0 : (1 << MD_DISK_SYNC
);
2599 /* only call uuid_from_super_imsm when this disk is part of a populated container,
2600 * ->compare_super may have updated the 'num_raid_devs' field for spares
2602 if (info
->disk
.state
& (1 << MD_DISK_SYNC
) || super
->anchor
->num_raid_devs
)
2603 uuid_from_super_imsm(st
, info
->uuid
);
2605 memcpy(info
->uuid
, uuid_zero
, sizeof(uuid_zero
));
2607 /* I don't know how to compute 'map' on imsm, so use safe default */
2610 for (i
= 0; i
< map_disks
; i
++)
2616 /* allocates memory and fills disk in mdinfo structure
2617 * for each disk in array */
2618 struct mdinfo
*getinfo_super_disks_imsm(struct supertype
*st
)
2620 struct mdinfo
*mddev
= NULL
;
2621 struct intel_super
*super
= st
->sb
;
2622 struct imsm_disk
*disk
;
2625 if (!super
|| !super
->disks
)
2628 mddev
= malloc(sizeof(*mddev
));
2630 fprintf(stderr
, Name
": Failed to allocate memory.\n");
2633 memset(mddev
, 0, sizeof(*mddev
));
2637 tmp
= malloc(sizeof(*tmp
));
2639 fprintf(stderr
, Name
": Failed to allocate memory.\n");
2644 memset(tmp
, 0, sizeof(*tmp
));
2646 tmp
->next
= mddev
->devs
;
2648 tmp
->disk
.number
= count
++;
2649 tmp
->disk
.major
= dl
->major
;
2650 tmp
->disk
.minor
= dl
->minor
;
2651 tmp
->disk
.state
= is_configured(disk
) ?
2652 (1 << MD_DISK_ACTIVE
) : 0;
2653 tmp
->disk
.state
|= is_failed(disk
) ? (1 << MD_DISK_FAULTY
) : 0;
2654 tmp
->disk
.state
|= is_spare(disk
) ? 0 : (1 << MD_DISK_SYNC
);
2655 tmp
->disk
.raid_disk
= -1;
2661 static int update_super_imsm(struct supertype
*st
, struct mdinfo
*info
,
2662 char *update
, char *devname
, int verbose
,
2663 int uuid_set
, char *homehost
)
2665 /* For 'assemble' and 'force' we need to return non-zero if any
2666 * change was made. For others, the return value is ignored.
2667 * Update options are:
2668 * force-one : This device looks a bit old but needs to be included,
2669 * update age info appropriately.
2670 * assemble: clear any 'faulty' flag to allow this device to
2672 * force-array: Array is degraded but being forced, mark it clean
2673 * if that will be needed to assemble it.
2675 * newdev: not used ????
2676 * grow: Array has gained a new device - this is currently for
2678 * resync: mark as dirty so a resync will happen.
2679 * name: update the name - preserving the homehost
2680 * uuid: Change the uuid of the array to match watch is given
2682 * Following are not relevant for this imsm:
2683 * sparc2.2 : update from old dodgey metadata
2684 * super-minor: change the preferred_minor number
2685 * summaries: update redundant counters.
2686 * homehost: update the recorded homehost
2687 * _reshape_progress: record new reshape_progress position.
2690 struct intel_super
*super
= st
->sb
;
2691 struct imsm_super
*mpb
;
2693 /* we can only update container info */
2694 if (!super
|| super
->current_vol
>= 0 || !super
->anchor
)
2697 mpb
= super
->anchor
;
2699 if (strcmp(update
, "uuid") == 0 && uuid_set
&& !info
->update_private
)
2701 else if (strcmp(update
, "uuid") == 0 && uuid_set
&& info
->update_private
) {
2702 mpb
->orig_family_num
= *((__u32
*) info
->update_private
);
2704 } else if (strcmp(update
, "uuid") == 0) {
2705 __u32
*new_family
= malloc(sizeof(*new_family
));
2707 /* update orig_family_number with the incoming random
2708 * data, report the new effective uuid, and store the
2709 * new orig_family_num for future updates.
2712 memcpy(&mpb
->orig_family_num
, info
->uuid
, sizeof(__u32
));
2713 uuid_from_super_imsm(st
, info
->uuid
);
2714 *new_family
= mpb
->orig_family_num
;
2715 info
->update_private
= new_family
;
2718 } else if (strcmp(update
, "assemble") == 0)
2723 /* successful update? recompute checksum */
2725 mpb
->check_sum
= __le32_to_cpu(__gen_imsm_checksum(mpb
));
2730 static size_t disks_to_mpb_size(int disks
)
2734 size
= sizeof(struct imsm_super
);
2735 size
+= (disks
- 1) * sizeof(struct imsm_disk
);
2736 size
+= 2 * sizeof(struct imsm_dev
);
2737 /* up to 2 maps per raid device (-2 for imsm_maps in imsm_dev */
2738 size
+= (4 - 2) * sizeof(struct imsm_map
);
2739 /* 4 possible disk_ord_tbl's */
2740 size
+= 4 * (disks
- 1) * sizeof(__u32
);
2745 static __u64
avail_size_imsm(struct supertype
*st
, __u64 devsize
)
2747 if (devsize
< (MPB_SECTOR_CNT
+ IMSM_RESERVED_SECTORS
))
2750 return devsize
- (MPB_SECTOR_CNT
+ IMSM_RESERVED_SECTORS
);
2753 static void free_devlist(struct intel_super
*super
)
2755 struct intel_dev
*dv
;
2757 while (super
->devlist
) {
2758 dv
= super
->devlist
->next
;
2759 free(super
->devlist
->dev
);
2760 free(super
->devlist
);
2761 super
->devlist
= dv
;
2765 static void imsm_copy_dev(struct imsm_dev
*dest
, struct imsm_dev
*src
)
2767 memcpy(dest
, src
, sizeof_imsm_dev(src
, 0));
2770 static int compare_super_imsm(struct supertype
*st
, struct supertype
*tst
)
2774 * 0 same, or first was empty, and second was copied
2775 * 1 second had wrong number
2777 * 3 wrong other info
2779 struct intel_super
*first
= st
->sb
;
2780 struct intel_super
*sec
= tst
->sb
;
2787 /* in platform dependent environment test if the disks
2788 * use the same Intel hba
2790 if (!check_env("IMSM_NO_PLATFORM")) {
2791 if (!first
->hba
|| !sec
->hba
||
2792 (first
->hba
->type
!= sec
->hba
->type
)) {
2794 "HBAs of devices does not match %s != %s\n",
2795 first
->hba
? get_sys_dev_type(first
->hba
->type
) : NULL
,
2796 sec
->hba
? get_sys_dev_type(sec
->hba
->type
) : NULL
);
2801 /* if an anchor does not have num_raid_devs set then it is a free
2804 if (first
->anchor
->num_raid_devs
> 0 &&
2805 sec
->anchor
->num_raid_devs
> 0) {
2806 /* Determine if these disks might ever have been
2807 * related. Further disambiguation can only take place
2808 * in load_super_imsm_all
2810 __u32 first_family
= first
->anchor
->orig_family_num
;
2811 __u32 sec_family
= sec
->anchor
->orig_family_num
;
2813 if (memcmp(first
->anchor
->sig
, sec
->anchor
->sig
,
2814 MAX_SIGNATURE_LENGTH
) != 0)
2817 if (first_family
== 0)
2818 first_family
= first
->anchor
->family_num
;
2819 if (sec_family
== 0)
2820 sec_family
= sec
->anchor
->family_num
;
2822 if (first_family
!= sec_family
)
2828 /* if 'first' is a spare promote it to a populated mpb with sec's
2831 if (first
->anchor
->num_raid_devs
== 0 &&
2832 sec
->anchor
->num_raid_devs
> 0) {
2834 struct intel_dev
*dv
;
2835 struct imsm_dev
*dev
;
2837 /* we need to copy raid device info from sec if an allocation
2838 * fails here we don't associate the spare
2840 for (i
= 0; i
< sec
->anchor
->num_raid_devs
; i
++) {
2841 dv
= malloc(sizeof(*dv
));
2844 dev
= malloc(sizeof_imsm_dev(get_imsm_dev(sec
, i
), 1));
2851 dv
->next
= first
->devlist
;
2852 first
->devlist
= dv
;
2854 if (i
< sec
->anchor
->num_raid_devs
) {
2855 /* allocation failure */
2856 free_devlist(first
);
2857 fprintf(stderr
, "imsm: failed to associate spare\n");
2860 first
->anchor
->num_raid_devs
= sec
->anchor
->num_raid_devs
;
2861 first
->anchor
->orig_family_num
= sec
->anchor
->orig_family_num
;
2862 first
->anchor
->family_num
= sec
->anchor
->family_num
;
2863 memcpy(first
->anchor
->sig
, sec
->anchor
->sig
, MAX_SIGNATURE_LENGTH
);
2864 for (i
= 0; i
< sec
->anchor
->num_raid_devs
; i
++)
2865 imsm_copy_dev(get_imsm_dev(first
, i
), get_imsm_dev(sec
, i
));
2871 static void fd2devname(int fd
, char *name
)
2875 char dname
[PATH_MAX
];
2880 if (fstat(fd
, &st
) != 0)
2882 sprintf(path
, "/sys/dev/block/%d:%d",
2883 major(st
.st_rdev
), minor(st
.st_rdev
));
2885 rv
= readlink(path
, dname
, sizeof(dname
)-1);
2890 nm
= strrchr(dname
, '/');
2893 snprintf(name
, MAX_RAID_SERIAL_LEN
, "/dev/%s", nm
);
2897 extern int scsi_get_serial(int fd
, void *buf
, size_t buf_len
);
2899 static int imsm_read_serial(int fd
, char *devname
,
2900 __u8 serial
[MAX_RAID_SERIAL_LEN
])
2902 unsigned char scsi_serial
[255];
2911 memset(scsi_serial
, 0, sizeof(scsi_serial
));
2913 rv
= scsi_get_serial(fd
, scsi_serial
, sizeof(scsi_serial
));
2915 if (rv
&& check_env("IMSM_DEVNAME_AS_SERIAL")) {
2916 memset(serial
, 0, MAX_RAID_SERIAL_LEN
);
2917 fd2devname(fd
, (char *) serial
);
2924 Name
": Failed to retrieve serial for %s\n",
2929 rsp_len
= scsi_serial
[3];
2933 Name
": Failed to retrieve serial for %s\n",
2937 rsp_buf
= (char *) &scsi_serial
[4];
2939 /* trim all whitespace and non-printable characters and convert
2942 for (i
= 0, dest
= rsp_buf
; i
< rsp_len
; i
++) {
2945 /* ':' is reserved for use in placeholder serial
2946 * numbers for missing disks
2954 len
= dest
- rsp_buf
;
2957 /* truncate leading characters */
2958 if (len
> MAX_RAID_SERIAL_LEN
) {
2959 dest
+= len
- MAX_RAID_SERIAL_LEN
;
2960 len
= MAX_RAID_SERIAL_LEN
;
2963 memset(serial
, 0, MAX_RAID_SERIAL_LEN
);
2964 memcpy(serial
, dest
, len
);
2969 static int serialcmp(__u8
*s1
, __u8
*s2
)
2971 return strncmp((char *) s1
, (char *) s2
, MAX_RAID_SERIAL_LEN
);
2974 static void serialcpy(__u8
*dest
, __u8
*src
)
2976 strncpy((char *) dest
, (char *) src
, MAX_RAID_SERIAL_LEN
);
2979 static struct dl
*serial_to_dl(__u8
*serial
, struct intel_super
*super
)
2983 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
2984 if (serialcmp(dl
->serial
, serial
) == 0)
2990 static struct imsm_disk
*
2991 __serial_to_disk(__u8
*serial
, struct imsm_super
*mpb
, int *idx
)
2995 for (i
= 0; i
< mpb
->num_disks
; i
++) {
2996 struct imsm_disk
*disk
= __get_imsm_disk(mpb
, i
);
2998 if (serialcmp(disk
->serial
, serial
) == 0) {
3009 load_imsm_disk(int fd
, struct intel_super
*super
, char *devname
, int keep_fd
)
3011 struct imsm_disk
*disk
;
3016 __u8 serial
[MAX_RAID_SERIAL_LEN
];
3018 rv
= imsm_read_serial(fd
, devname
, serial
);
3023 dl
= calloc(1, sizeof(*dl
));
3027 Name
": failed to allocate disk buffer for %s\n",
3033 dl
->major
= major(stb
.st_rdev
);
3034 dl
->minor
= minor(stb
.st_rdev
);
3035 dl
->next
= super
->disks
;
3036 dl
->fd
= keep_fd
? fd
: -1;
3037 assert(super
->disks
== NULL
);
3039 serialcpy(dl
->serial
, serial
);
3042 fd2devname(fd
, name
);
3044 dl
->devname
= strdup(devname
);
3046 dl
->devname
= strdup(name
);
3048 /* look up this disk's index in the current anchor */
3049 disk
= __serial_to_disk(dl
->serial
, super
->anchor
, &dl
->index
);
3052 /* only set index on disks that are a member of a
3053 * populated contianer, i.e. one with raid_devs
3055 if (is_failed(&dl
->disk
))
3057 else if (is_spare(&dl
->disk
))
3065 /* When migrating map0 contains the 'destination' state while map1
3066 * contains the current state. When not migrating map0 contains the
3067 * current state. This routine assumes that map[0].map_state is set to
3068 * the current array state before being called.
3070 * Migration is indicated by one of the following states
3071 * 1/ Idle (migr_state=0 map0state=normal||unitialized||degraded||failed)
3072 * 2/ Initialize (migr_state=1 migr_type=MIGR_INIT map0state=normal
3073 * map1state=unitialized)
3074 * 3/ Repair (Resync) (migr_state=1 migr_type=MIGR_REPAIR map0state=normal
3076 * 4/ Rebuild (migr_state=1 migr_type=MIGR_REBUILD map0state=normal
3077 * map1state=degraded)
3078 * 5/ Migration (mig_state=1 migr_type=MIGR_GEN_MIGR map0state=normal
3081 static void migrate(struct imsm_dev
*dev
, struct intel_super
*super
,
3082 __u8 to_state
, int migr_type
)
3084 struct imsm_map
*dest
;
3085 struct imsm_map
*src
= get_imsm_map(dev
, 0);
3087 dev
->vol
.migr_state
= 1;
3088 set_migr_type(dev
, migr_type
);
3089 dev
->vol
.curr_migr_unit
= 0;
3090 dest
= get_imsm_map(dev
, 1);
3092 /* duplicate and then set the target end state in map[0] */
3093 memcpy(dest
, src
, sizeof_imsm_map(src
));
3094 if ((migr_type
== MIGR_REBUILD
) ||
3095 (migr_type
== MIGR_GEN_MIGR
)) {
3099 for (i
= 0; i
< src
->num_members
; i
++) {
3100 ord
= __le32_to_cpu(src
->disk_ord_tbl
[i
]);
3101 set_imsm_ord_tbl_ent(src
, i
, ord_to_idx(ord
));
3105 if (migr_type
== MIGR_GEN_MIGR
)
3106 /* Clear migration record */
3107 memset(super
->migr_rec
, 0, sizeof(struct migr_record
));
3109 src
->map_state
= to_state
;
3112 static void end_migration(struct imsm_dev
*dev
, __u8 map_state
)
3114 struct imsm_map
*map
= get_imsm_map(dev
, 0);
3115 struct imsm_map
*prev
= get_imsm_map(dev
, dev
->vol
.migr_state
);
3118 /* merge any IMSM_ORD_REBUILD bits that were not successfully
3119 * completed in the last migration.
3121 * FIXME add support for raid-level-migration
3123 for (i
= 0; i
< prev
->num_members
; i
++)
3124 for (j
= 0; j
< map
->num_members
; j
++)
3125 /* during online capacity expansion
3126 * disks position can be changed if takeover is used
3128 if (ord_to_idx(map
->disk_ord_tbl
[j
]) ==
3129 ord_to_idx(prev
->disk_ord_tbl
[i
])) {
3130 map
->disk_ord_tbl
[j
] |= prev
->disk_ord_tbl
[i
];
3134 dev
->vol
.migr_state
= 0;
3135 set_migr_type(dev
, 0);
3136 dev
->vol
.curr_migr_unit
= 0;
3137 map
->map_state
= map_state
;
3141 static int parse_raid_devices(struct intel_super
*super
)
3144 struct imsm_dev
*dev_new
;
3145 size_t len
, len_migr
;
3147 size_t space_needed
= 0;
3148 struct imsm_super
*mpb
= super
->anchor
;
3150 for (i
= 0; i
< super
->anchor
->num_raid_devs
; i
++) {
3151 struct imsm_dev
*dev_iter
= __get_imsm_dev(super
->anchor
, i
);
3152 struct intel_dev
*dv
;
3154 len
= sizeof_imsm_dev(dev_iter
, 0);
3155 len_migr
= sizeof_imsm_dev(dev_iter
, 1);
3157 space_needed
+= len_migr
- len
;
3159 dv
= malloc(sizeof(*dv
));
3162 if (max_len
< len_migr
)
3164 if (max_len
> len_migr
)
3165 space_needed
+= max_len
- len_migr
;
3166 dev_new
= malloc(max_len
);
3171 imsm_copy_dev(dev_new
, dev_iter
);
3174 dv
->next
= super
->devlist
;
3175 super
->devlist
= dv
;
3178 /* ensure that super->buf is large enough when all raid devices
3181 if (__le32_to_cpu(mpb
->mpb_size
) + space_needed
> super
->len
) {
3184 len
= ROUND_UP(__le32_to_cpu(mpb
->mpb_size
) + space_needed
, 512);
3185 if (posix_memalign(&buf
, 512, len
) != 0)
3188 memcpy(buf
, super
->buf
, super
->len
);
3189 memset(buf
+ super
->len
, 0, len
- super
->len
);
3198 /* retrieve a pointer to the bbm log which starts after all raid devices */
3199 struct bbm_log
*__get_imsm_bbm_log(struct imsm_super
*mpb
)
3203 if (__le32_to_cpu(mpb
->bbm_log_size
)) {
3205 ptr
+= mpb
->mpb_size
- __le32_to_cpu(mpb
->bbm_log_size
);
3211 /*******************************************************************************
3212 * Function: check_mpb_migr_compatibility
3213 * Description: Function checks for unsupported migration features:
3214 * - migration optimization area (pba_of_lba0)
3215 * - descending reshape (ascending_migr)
3217 * super : imsm metadata information
3219 * 0 : migration is compatible
3220 * -1 : migration is not compatible
3221 ******************************************************************************/
3222 int check_mpb_migr_compatibility(struct intel_super
*super
)
3224 struct imsm_map
*map0
, *map1
;
3225 struct migr_record
*migr_rec
= super
->migr_rec
;
3228 for (i
= 0; i
< super
->anchor
->num_raid_devs
; i
++) {
3229 struct imsm_dev
*dev_iter
= __get_imsm_dev(super
->anchor
, i
);
3232 dev_iter
->vol
.migr_state
== 1 &&
3233 dev_iter
->vol
.migr_type
== MIGR_GEN_MIGR
) {
3234 /* This device is migrating */
3235 map0
= get_imsm_map(dev_iter
, 0);
3236 map1
= get_imsm_map(dev_iter
, 1);
3237 if (map0
->pba_of_lba0
!= map1
->pba_of_lba0
)
3238 /* migration optimization area was used */
3240 if (migr_rec
->ascending_migr
== 0
3241 && migr_rec
->dest_depth_per_unit
> 0)
3242 /* descending reshape not supported yet */
3249 static void __free_imsm(struct intel_super
*super
, int free_disks
);
3251 /* load_imsm_mpb - read matrix metadata
3252 * allocates super->mpb to be freed by free_imsm
3254 static int load_imsm_mpb(int fd
, struct intel_super
*super
, char *devname
)
3256 unsigned long long dsize
;
3257 unsigned long long sectors
;
3259 struct imsm_super
*anchor
;
3262 get_dev_size(fd
, NULL
, &dsize
);
3266 Name
": %s: device to small for imsm\n",
3271 if (lseek64(fd
, dsize
- (512 * 2), SEEK_SET
) < 0) {
3273 fprintf(stderr
, Name
3274 ": Cannot seek to anchor block on %s: %s\n",
3275 devname
, strerror(errno
));
3279 if (posix_memalign((void**)&anchor
, 512, 512) != 0) {
3282 Name
": Failed to allocate imsm anchor buffer"
3283 " on %s\n", devname
);
3286 if (read(fd
, anchor
, 512) != 512) {
3289 Name
": Cannot read anchor block on %s: %s\n",
3290 devname
, strerror(errno
));
3295 if (strncmp((char *) anchor
->sig
, MPB_SIGNATURE
, MPB_SIG_LEN
) != 0) {
3298 Name
": no IMSM anchor on %s\n", devname
);
3303 __free_imsm(super
, 0);
3304 /* reload capability and hba */
3306 /* capability and hba must be updated with new super allocation */
3307 find_intel_hba_capability(fd
, super
, devname
);
3308 super
->len
= ROUND_UP(anchor
->mpb_size
, 512);
3309 if (posix_memalign(&super
->buf
, 512, super
->len
) != 0) {
3312 Name
": unable to allocate %zu byte mpb buffer\n",
3317 memcpy(super
->buf
, anchor
, 512);
3319 sectors
= mpb_sectors(anchor
) - 1;
3322 if (posix_memalign(&super
->migr_rec_buf
, 512, 512) != 0) {
3323 fprintf(stderr
, Name
3324 ": %s could not allocate migr_rec buffer\n", __func__
);
3330 check_sum
= __gen_imsm_checksum(super
->anchor
);
3331 if (check_sum
!= __le32_to_cpu(super
->anchor
->check_sum
)) {
3334 Name
": IMSM checksum %x != %x on %s\n",
3336 __le32_to_cpu(super
->anchor
->check_sum
),
3344 /* read the extended mpb */
3345 if (lseek64(fd
, dsize
- (512 * (2 + sectors
)), SEEK_SET
) < 0) {
3348 Name
": Cannot seek to extended mpb on %s: %s\n",
3349 devname
, strerror(errno
));
3353 if ((unsigned)read(fd
, super
->buf
+ 512, super
->len
- 512) != super
->len
- 512) {
3356 Name
": Cannot read extended mpb on %s: %s\n",
3357 devname
, strerror(errno
));
3361 check_sum
= __gen_imsm_checksum(super
->anchor
);
3362 if (check_sum
!= __le32_to_cpu(super
->anchor
->check_sum
)) {
3365 Name
": IMSM checksum %x != %x on %s\n",
3366 check_sum
, __le32_to_cpu(super
->anchor
->check_sum
),
3371 /* FIXME the BBM log is disk specific so we cannot use this global
3372 * buffer for all disks. Ok for now since we only look at the global
3373 * bbm_log_size parameter to gate assembly
3375 super
->bbm_log
= __get_imsm_bbm_log(super
->anchor
);
3380 static int read_imsm_migr_rec(int fd
, struct intel_super
*super
);
3383 load_and_parse_mpb(int fd
, struct intel_super
*super
, char *devname
, int keep_fd
)
3387 err
= load_imsm_mpb(fd
, super
, devname
);
3390 err
= load_imsm_disk(fd
, super
, devname
, keep_fd
);
3393 err
= parse_raid_devices(super
);
3398 static void __free_imsm_disk(struct dl
*d
)
3410 static void free_imsm_disks(struct intel_super
*super
)
3414 while (super
->disks
) {
3416 super
->disks
= d
->next
;
3417 __free_imsm_disk(d
);
3419 while (super
->disk_mgmt_list
) {
3420 d
= super
->disk_mgmt_list
;
3421 super
->disk_mgmt_list
= d
->next
;
3422 __free_imsm_disk(d
);
3424 while (super
->missing
) {
3426 super
->missing
= d
->next
;
3427 __free_imsm_disk(d
);
3432 /* free all the pieces hanging off of a super pointer */
3433 static void __free_imsm(struct intel_super
*super
, int free_disks
)
3435 struct intel_hba
*elem
, *next
;
3441 /* unlink capability description */
3443 if (super
->migr_rec_buf
) {
3444 free(super
->migr_rec_buf
);
3445 super
->migr_rec_buf
= NULL
;
3448 free_imsm_disks(super
);
3449 free_devlist(super
);
3453 free((void *)elem
->path
);
3461 static void free_imsm(struct intel_super
*super
)
3463 __free_imsm(super
, 1);
3467 static void free_super_imsm(struct supertype
*st
)
3469 struct intel_super
*super
= st
->sb
;
3478 static struct intel_super
*alloc_super(void)
3480 struct intel_super
*super
= malloc(sizeof(*super
));
3483 memset(super
, 0, sizeof(*super
));
3484 super
->current_vol
= -1;
3485 super
->create_offset
= ~((__u32
) 0);
3491 * find and allocate hba and OROM/EFI based on valid fd of RAID component device
3493 static int find_intel_hba_capability(int fd
, struct intel_super
*super
, char *devname
)
3495 struct sys_dev
*hba_name
;
3498 if ((fd
< 0) || check_env("IMSM_NO_PLATFORM")) {
3503 hba_name
= find_disk_attached_hba(fd
, NULL
);
3507 Name
": %s is not attached to Intel(R) RAID controller.\n",
3511 rv
= attach_hba_to_super(super
, hba_name
);
3514 struct intel_hba
*hba
= super
->hba
;
3516 fprintf(stderr
, Name
": %s is attached to Intel(R) %s RAID "
3517 "controller (%s),\n"
3518 " but the container is assigned to Intel(R) "
3519 "%s RAID controller (",
3522 hba_name
->pci_id
? : "Err!",
3523 get_sys_dev_type(hba_name
->type
));
3526 fprintf(stderr
, "%s", hba
->pci_id
? : "Err!");
3528 fprintf(stderr
, ", ");
3532 fprintf(stderr
, ").\n"
3533 " Mixing devices attached to different controllers "
3534 "is not allowed.\n");
3536 free_sys_dev(&hba_name
);
3539 super
->orom
= find_imsm_capability(hba_name
->type
);
3540 free_sys_dev(&hba_name
);
3546 /* find_missing - helper routine for load_super_imsm_all that identifies
3547 * disks that have disappeared from the system. This routine relies on
3548 * the mpb being uptodate, which it is at load time.
3550 static int find_missing(struct intel_super
*super
)
3553 struct imsm_super
*mpb
= super
->anchor
;
3555 struct imsm_disk
*disk
;
3557 for (i
= 0; i
< mpb
->num_disks
; i
++) {
3558 disk
= __get_imsm_disk(mpb
, i
);
3559 dl
= serial_to_dl(disk
->serial
, super
);
3563 dl
= malloc(sizeof(*dl
));
3569 dl
->devname
= strdup("missing");
3571 serialcpy(dl
->serial
, disk
->serial
);
3574 dl
->next
= super
->missing
;
3575 super
->missing
= dl
;
3582 static struct intel_disk
*disk_list_get(__u8
*serial
, struct intel_disk
*disk_list
)
3584 struct intel_disk
*idisk
= disk_list
;
3587 if (serialcmp(idisk
->disk
.serial
, serial
) == 0)
3589 idisk
= idisk
->next
;
3595 static int __prep_thunderdome(struct intel_super
**table
, int tbl_size
,
3596 struct intel_super
*super
,
3597 struct intel_disk
**disk_list
)
3599 struct imsm_disk
*d
= &super
->disks
->disk
;
3600 struct imsm_super
*mpb
= super
->anchor
;
3603 for (i
= 0; i
< tbl_size
; i
++) {
3604 struct imsm_super
*tbl_mpb
= table
[i
]->anchor
;
3605 struct imsm_disk
*tbl_d
= &table
[i
]->disks
->disk
;
3607 if (tbl_mpb
->family_num
== mpb
->family_num
) {
3608 if (tbl_mpb
->check_sum
== mpb
->check_sum
) {
3609 dprintf("%s: mpb from %d:%d matches %d:%d\n",
3610 __func__
, super
->disks
->major
,
3611 super
->disks
->minor
,
3612 table
[i
]->disks
->major
,
3613 table
[i
]->disks
->minor
);
3617 if (((is_configured(d
) && !is_configured(tbl_d
)) ||
3618 is_configured(d
) == is_configured(tbl_d
)) &&
3619 tbl_mpb
->generation_num
< mpb
->generation_num
) {
3620 /* current version of the mpb is a
3621 * better candidate than the one in
3622 * super_table, but copy over "cross
3623 * generational" status
3625 struct intel_disk
*idisk
;
3627 dprintf("%s: mpb from %d:%d replaces %d:%d\n",
3628 __func__
, super
->disks
->major
,
3629 super
->disks
->minor
,
3630 table
[i
]->disks
->major
,
3631 table
[i
]->disks
->minor
);
3633 idisk
= disk_list_get(tbl_d
->serial
, *disk_list
);
3634 if (idisk
&& is_failed(&idisk
->disk
))
3635 tbl_d
->status
|= FAILED_DISK
;
3638 struct intel_disk
*idisk
;
3639 struct imsm_disk
*disk
;
3641 /* tbl_mpb is more up to date, but copy
3642 * over cross generational status before
3645 disk
= __serial_to_disk(d
->serial
, mpb
, NULL
);
3646 if (disk
&& is_failed(disk
))
3647 d
->status
|= FAILED_DISK
;
3649 idisk
= disk_list_get(d
->serial
, *disk_list
);
3652 if (disk
&& is_configured(disk
))
3653 idisk
->disk
.status
|= CONFIGURED_DISK
;
3656 dprintf("%s: mpb from %d:%d prefer %d:%d\n",
3657 __func__
, super
->disks
->major
,
3658 super
->disks
->minor
,
3659 table
[i
]->disks
->major
,
3660 table
[i
]->disks
->minor
);
3668 table
[tbl_size
++] = super
;
3672 /* update/extend the merged list of imsm_disk records */
3673 for (j
= 0; j
< mpb
->num_disks
; j
++) {
3674 struct imsm_disk
*disk
= __get_imsm_disk(mpb
, j
);
3675 struct intel_disk
*idisk
;
3677 idisk
= disk_list_get(disk
->serial
, *disk_list
);
3679 idisk
->disk
.status
|= disk
->status
;
3680 if (is_configured(&idisk
->disk
) ||
3681 is_failed(&idisk
->disk
))
3682 idisk
->disk
.status
&= ~(SPARE_DISK
);
3684 idisk
= calloc(1, sizeof(*idisk
));
3687 idisk
->owner
= IMSM_UNKNOWN_OWNER
;
3688 idisk
->disk
= *disk
;
3689 idisk
->next
= *disk_list
;
3693 if (serialcmp(idisk
->disk
.serial
, d
->serial
) == 0)
3700 static struct intel_super
*
3701 validate_members(struct intel_super
*super
, struct intel_disk
*disk_list
,
3704 struct imsm_super
*mpb
= super
->anchor
;
3708 for (i
= 0; i
< mpb
->num_disks
; i
++) {
3709 struct imsm_disk
*disk
= __get_imsm_disk(mpb
, i
);
3710 struct intel_disk
*idisk
;
3712 idisk
= disk_list_get(disk
->serial
, disk_list
);
3714 if (idisk
->owner
== owner
||
3715 idisk
->owner
== IMSM_UNKNOWN_OWNER
)
3718 dprintf("%s: '%.16s' owner %d != %d\n",
3719 __func__
, disk
->serial
, idisk
->owner
,
3722 dprintf("%s: unknown disk %x [%d]: %.16s\n",
3723 __func__
, __le32_to_cpu(mpb
->family_num
), i
,
3729 if (ok_count
== mpb
->num_disks
)
3734 static void show_conflicts(__u32 family_num
, struct intel_super
*super_list
)
3736 struct intel_super
*s
;
3738 for (s
= super_list
; s
; s
= s
->next
) {
3739 if (family_num
!= s
->anchor
->family_num
)
3741 fprintf(stderr
, "Conflict, offlining family %#x on '%s'\n",
3742 __le32_to_cpu(family_num
), s
->disks
->devname
);
3746 static struct intel_super
*
3747 imsm_thunderdome(struct intel_super
**super_list
, int len
)
3749 struct intel_super
*super_table
[len
];
3750 struct intel_disk
*disk_list
= NULL
;
3751 struct intel_super
*champion
, *spare
;
3752 struct intel_super
*s
, **del
;
3757 memset(super_table
, 0, sizeof(super_table
));
3758 for (s
= *super_list
; s
; s
= s
->next
)
3759 tbl_size
= __prep_thunderdome(super_table
, tbl_size
, s
, &disk_list
);
3761 for (i
= 0; i
< tbl_size
; i
++) {
3762 struct imsm_disk
*d
;
3763 struct intel_disk
*idisk
;
3764 struct imsm_super
*mpb
= super_table
[i
]->anchor
;
3767 d
= &s
->disks
->disk
;
3769 /* 'd' must appear in merged disk list for its
3770 * configuration to be valid
3772 idisk
= disk_list_get(d
->serial
, disk_list
);
3773 if (idisk
&& idisk
->owner
== i
)
3774 s
= validate_members(s
, disk_list
, i
);
3779 dprintf("%s: marking family: %#x from %d:%d offline\n",
3780 __func__
, mpb
->family_num
,
3781 super_table
[i
]->disks
->major
,
3782 super_table
[i
]->disks
->minor
);
3786 /* This is where the mdadm implementation differs from the Windows
3787 * driver which has no strict concept of a container. We can only
3788 * assemble one family from a container, so when returning a prodigal
3789 * array member to this system the code will not be able to disambiguate
3790 * the container contents that should be assembled ("foreign" versus
3791 * "local"). It requires user intervention to set the orig_family_num
3792 * to a new value to establish a new container. The Windows driver in
3793 * this situation fixes up the volume name in place and manages the
3794 * foreign array as an independent entity.
3799 for (i
= 0; i
< tbl_size
; i
++) {
3800 struct intel_super
*tbl_ent
= super_table
[i
];
3806 if (tbl_ent
->anchor
->num_raid_devs
== 0) {
3811 if (s
&& !is_spare
) {
3812 show_conflicts(tbl_ent
->anchor
->family_num
, *super_list
);
3814 } else if (!s
&& !is_spare
)
3827 fprintf(stderr
, "Chose family %#x on '%s', "
3828 "assemble conflicts to new container with '--update=uuid'\n",
3829 __le32_to_cpu(s
->anchor
->family_num
), s
->disks
->devname
);
3831 /* collect all dl's onto 'champion', and update them to
3832 * champion's version of the status
3834 for (s
= *super_list
; s
; s
= s
->next
) {
3835 struct imsm_super
*mpb
= champion
->anchor
;
3836 struct dl
*dl
= s
->disks
;
3841 for (i
= 0; i
< mpb
->num_disks
; i
++) {
3842 struct imsm_disk
*disk
;
3844 disk
= __serial_to_disk(dl
->serial
, mpb
, &dl
->index
);
3847 /* only set index on disks that are a member of
3848 * a populated contianer, i.e. one with
3851 if (is_failed(&dl
->disk
))
3853 else if (is_spare(&dl
->disk
))
3859 if (i
>= mpb
->num_disks
) {
3860 struct intel_disk
*idisk
;
3862 idisk
= disk_list_get(dl
->serial
, disk_list
);
3863 if (idisk
&& is_spare(&idisk
->disk
) &&
3864 !is_failed(&idisk
->disk
) && !is_configured(&idisk
->disk
))
3872 dl
->next
= champion
->disks
;
3873 champion
->disks
= dl
;
3877 /* delete 'champion' from super_list */
3878 for (del
= super_list
; *del
; ) {
3879 if (*del
== champion
) {
3880 *del
= (*del
)->next
;
3883 del
= &(*del
)->next
;
3885 champion
->next
= NULL
;
3889 struct intel_disk
*idisk
= disk_list
;
3891 disk_list
= disk_list
->next
;
3898 static int load_super_imsm_all(struct supertype
*st
, int fd
, void **sbp
,
3902 struct intel_super
*super_list
= NULL
;
3903 struct intel_super
*super
= NULL
;
3904 int devnum
= fd2devnum(fd
);
3910 /* check if 'fd' an opened container */
3911 sra
= sysfs_read(fd
, 0, GET_LEVEL
|GET_VERSION
|GET_DEVS
|GET_STATE
);
3915 if (sra
->array
.major_version
!= -1 ||
3916 sra
->array
.minor_version
!= -2 ||
3917 strcmp(sra
->text_version
, "imsm") != 0) {
3922 for (sd
= sra
->devs
, i
= 0; sd
; sd
= sd
->next
, i
++) {
3923 struct intel_super
*s
= alloc_super();
3931 s
->next
= super_list
;
3935 sprintf(nm
, "%d:%d", sd
->disk
.major
, sd
->disk
.minor
);
3936 dfd
= dev_open(nm
, O_RDWR
);
3940 rv
= find_intel_hba_capability(dfd
, s
, devname
);
3941 /* no orom/efi or non-intel hba of the disk */
3945 err
= load_and_parse_mpb(dfd
, s
, NULL
, 1);
3947 /* retry the load if we might have raced against mdmon */
3948 if (err
== 3 && mdmon_running(devnum
))
3949 for (retry
= 0; retry
< 3; retry
++) {
3951 err
= load_and_parse_mpb(dfd
, s
, NULL
, 1);
3959 /* all mpbs enter, maybe one leaves */
3960 super
= imsm_thunderdome(&super_list
, i
);
3966 if (find_missing(super
) != 0) {
3972 /* load migration record */
3973 err
= load_imsm_migr_rec(super
, NULL
);
3979 /* Check migration compatibility */
3980 if (check_mpb_migr_compatibility(super
) != 0) {
3981 fprintf(stderr
, Name
": Unsupported migration detected");
3983 fprintf(stderr
, " on %s\n", devname
);
3985 fprintf(stderr
, " (IMSM).\n");
3994 while (super_list
) {
3995 struct intel_super
*s
= super_list
;
3997 super_list
= super_list
->next
;
4006 st
->container_dev
= devnum
;
4007 if (err
== 0 && st
->ss
== NULL
) {
4008 st
->ss
= &super_imsm
;
4009 st
->minor_version
= 0;
4010 st
->max_devs
= IMSM_MAX_DEVICES
;
4015 static int load_container_imsm(struct supertype
*st
, int fd
, char *devname
)
4017 return load_super_imsm_all(st
, fd
, &st
->sb
, devname
);
4021 static int load_super_imsm(struct supertype
*st
, int fd
, char *devname
)
4023 struct intel_super
*super
;
4026 if (test_partition(fd
))
4027 /* IMSM not allowed on partitions */
4030 free_super_imsm(st
);
4032 super
= alloc_super();
4035 Name
": malloc of %zu failed.\n",
4039 /* Load hba and capabilities if they exist.
4040 * But do not preclude loading metadata in case capabilities or hba are
4041 * non-compliant and ignore_hw_compat is set.
4043 rv
= find_intel_hba_capability(fd
, super
, devname
);
4044 /* no orom/efi or non-intel hba of the disk */
4045 if ((rv
!= 0) && (st
->ignore_hw_compat
== 0)) {
4048 Name
": No OROM/EFI properties for %s\n", devname
);
4052 rv
= load_and_parse_mpb(fd
, super
, devname
, 0);
4057 Name
": Failed to load all information "
4058 "sections on %s\n", devname
);
4064 if (st
->ss
== NULL
) {
4065 st
->ss
= &super_imsm
;
4066 st
->minor_version
= 0;
4067 st
->max_devs
= IMSM_MAX_DEVICES
;
4070 /* load migration record */
4071 if (load_imsm_migr_rec(super
, NULL
) == 0) {
4072 /* Check for unsupported migration features */
4073 if (check_mpb_migr_compatibility(super
) != 0) {
4075 Name
": Unsupported migration detected");
4077 fprintf(stderr
, " on %s\n", devname
);
4079 fprintf(stderr
, " (IMSM).\n");
4087 static __u16
info_to_blocks_per_strip(mdu_array_info_t
*info
)
4089 if (info
->level
== 1)
4091 return info
->chunk_size
>> 9;
4094 static __u32
info_to_num_data_stripes(mdu_array_info_t
*info
, int num_domains
)
4098 num_stripes
= (info
->size
* 2) / info_to_blocks_per_strip(info
);
4099 num_stripes
/= num_domains
;
4104 static __u32
info_to_blocks_per_member(mdu_array_info_t
*info
)
4106 if (info
->level
== 1)
4107 return info
->size
* 2;
4109 return (info
->size
* 2) & ~(info_to_blocks_per_strip(info
) - 1);
4112 static void imsm_update_version_info(struct intel_super
*super
)
4114 /* update the version and attributes */
4115 struct imsm_super
*mpb
= super
->anchor
;
4117 struct imsm_dev
*dev
;
4118 struct imsm_map
*map
;
4121 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
4122 dev
= get_imsm_dev(super
, i
);
4123 map
= get_imsm_map(dev
, 0);
4124 if (__le32_to_cpu(dev
->size_high
) > 0)
4125 mpb
->attributes
|= MPB_ATTRIB_2TB
;
4127 /* FIXME detect when an array spans a port multiplier */
4129 mpb
->attributes
|= MPB_ATTRIB_PM
;
4132 if (mpb
->num_raid_devs
> 1 ||
4133 mpb
->attributes
!= MPB_ATTRIB_CHECKSUM_VERIFY
) {
4134 version
= MPB_VERSION_ATTRIBS
;
4135 switch (get_imsm_raid_level(map
)) {
4136 case 0: mpb
->attributes
|= MPB_ATTRIB_RAID0
; break;
4137 case 1: mpb
->attributes
|= MPB_ATTRIB_RAID1
; break;
4138 case 10: mpb
->attributes
|= MPB_ATTRIB_RAID10
; break;
4139 case 5: mpb
->attributes
|= MPB_ATTRIB_RAID5
; break;
4142 if (map
->num_members
>= 5)
4143 version
= MPB_VERSION_5OR6_DISK_ARRAY
;
4144 else if (dev
->status
== DEV_CLONE_N_GO
)
4145 version
= MPB_VERSION_CNG
;
4146 else if (get_imsm_raid_level(map
) == 5)
4147 version
= MPB_VERSION_RAID5
;
4148 else if (map
->num_members
>= 3)
4149 version
= MPB_VERSION_3OR4_DISK_ARRAY
;
4150 else if (get_imsm_raid_level(map
) == 1)
4151 version
= MPB_VERSION_RAID1
;
4153 version
= MPB_VERSION_RAID0
;
4155 strcpy(((char *) mpb
->sig
) + strlen(MPB_SIGNATURE
), version
);
4159 static int check_name(struct intel_super
*super
, char *name
, int quiet
)
4161 struct imsm_super
*mpb
= super
->anchor
;
4162 char *reason
= NULL
;
4165 if (strlen(name
) > MAX_RAID_SERIAL_LEN
)
4166 reason
= "must be 16 characters or less";
4168 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
4169 struct imsm_dev
*dev
= get_imsm_dev(super
, i
);
4171 if (strncmp((char *) dev
->volume
, name
, MAX_RAID_SERIAL_LEN
) == 0) {
4172 reason
= "already exists";
4177 if (reason
&& !quiet
)
4178 fprintf(stderr
, Name
": imsm volume name %s\n", reason
);
4183 static int init_super_imsm_volume(struct supertype
*st
, mdu_array_info_t
*info
,
4184 unsigned long long size
, char *name
,
4185 char *homehost
, int *uuid
)
4187 /* We are creating a volume inside a pre-existing container.
4188 * so st->sb is already set.
4190 struct intel_super
*super
= st
->sb
;
4191 struct imsm_super
*mpb
= super
->anchor
;
4192 struct intel_dev
*dv
;
4193 struct imsm_dev
*dev
;
4194 struct imsm_vol
*vol
;
4195 struct imsm_map
*map
;
4196 int idx
= mpb
->num_raid_devs
;
4198 unsigned long long array_blocks
;
4199 size_t size_old
, size_new
;
4200 __u32 num_data_stripes
;
4202 if (super
->orom
&& mpb
->num_raid_devs
>= super
->orom
->vpa
) {
4203 fprintf(stderr
, Name
": This imsm-container already has the "
4204 "maximum of %d volumes\n", super
->orom
->vpa
);
4208 /* ensure the mpb is large enough for the new data */
4209 size_old
= __le32_to_cpu(mpb
->mpb_size
);
4210 size_new
= disks_to_mpb_size(info
->nr_disks
);
4211 if (size_new
> size_old
) {
4213 size_t size_round
= ROUND_UP(size_new
, 512);
4215 if (posix_memalign(&mpb_new
, 512, size_round
) != 0) {
4216 fprintf(stderr
, Name
": could not allocate new mpb\n");
4219 if (posix_memalign(&super
->migr_rec_buf
, 512, 512) != 0) {
4220 fprintf(stderr
, Name
4221 ": %s could not allocate migr_rec buffer\n",
4228 memcpy(mpb_new
, mpb
, size_old
);
4231 super
->anchor
= mpb_new
;
4232 mpb
->mpb_size
= __cpu_to_le32(size_new
);
4233 memset(mpb_new
+ size_old
, 0, size_round
- size_old
);
4235 super
->current_vol
= idx
;
4237 /* handle 'failed_disks' by either:
4238 * a) create dummy disk entries in the table if this the first
4239 * volume in the array. We add them here as this is the only
4240 * opportunity to add them. add_to_super_imsm_volume()
4241 * handles the non-failed disks and continues incrementing
4243 * b) validate that 'failed_disks' matches the current number
4244 * of missing disks if the container is populated
4246 if (super
->current_vol
== 0) {
4248 for (i
= 0; i
< info
->failed_disks
; i
++) {
4249 struct imsm_disk
*disk
;
4252 disk
= __get_imsm_disk(mpb
, i
);
4253 disk
->status
= CONFIGURED_DISK
| FAILED_DISK
;
4254 disk
->scsi_id
= __cpu_to_le32(~(__u32
)0);
4255 snprintf((char *) disk
->serial
, MAX_RAID_SERIAL_LEN
,
4258 find_missing(super
);
4263 for (d
= super
->missing
; d
; d
= d
->next
)
4265 if (info
->failed_disks
> missing
) {
4266 fprintf(stderr
, Name
": unable to add 'missing' disk to container\n");
4271 if (!check_name(super
, name
, 0))
4273 dv
= malloc(sizeof(*dv
));
4275 fprintf(stderr
, Name
": failed to allocate device list entry\n");
4278 dev
= calloc(1, sizeof(*dev
) + sizeof(__u32
) * (info
->raid_disks
- 1));
4281 fprintf(stderr
, Name
": could not allocate raid device\n");
4285 strncpy((char *) dev
->volume
, name
, MAX_RAID_SERIAL_LEN
);
4286 if (info
->level
== 1)
4287 array_blocks
= info_to_blocks_per_member(info
);
4289 array_blocks
= calc_array_size(info
->level
, info
->raid_disks
,
4290 info
->layout
, info
->chunk_size
,
4292 /* round array size down to closest MB */
4293 array_blocks
= (array_blocks
>> SECT_PER_MB_SHIFT
) << SECT_PER_MB_SHIFT
;
4295 dev
->size_low
= __cpu_to_le32((__u32
) array_blocks
);
4296 dev
->size_high
= __cpu_to_le32((__u32
) (array_blocks
>> 32));
4297 dev
->status
= (DEV_READ_COALESCING
| DEV_WRITE_COALESCING
);
4299 vol
->migr_state
= 0;
4300 set_migr_type(dev
, MIGR_INIT
);
4301 vol
->dirty
= !info
->state
;
4302 vol
->curr_migr_unit
= 0;
4303 map
= get_imsm_map(dev
, 0);
4304 map
->pba_of_lba0
= __cpu_to_le32(super
->create_offset
);
4305 map
->blocks_per_member
= __cpu_to_le32(info_to_blocks_per_member(info
));
4306 map
->blocks_per_strip
= __cpu_to_le16(info_to_blocks_per_strip(info
));
4307 map
->failed_disk_num
= ~0;
4308 map
->map_state
= info
->failed_disks
? IMSM_T_STATE_DEGRADED
: IMSM_T_STATE_NORMAL
;
4311 if (info
->level
== 1 && info
->raid_disks
> 2) {
4314 fprintf(stderr
, Name
": imsm does not support more than 2 disks"
4315 "in a raid1 volume\n");
4319 map
->raid_level
= info
->level
;
4320 if (info
->level
== 10) {
4321 map
->raid_level
= 1;
4322 map
->num_domains
= info
->raid_disks
/ 2;
4323 } else if (info
->level
== 1)
4324 map
->num_domains
= info
->raid_disks
;
4326 map
->num_domains
= 1;
4328 num_data_stripes
= info_to_num_data_stripes(info
, map
->num_domains
);
4329 map
->num_data_stripes
= __cpu_to_le32(num_data_stripes
);
4331 map
->num_members
= info
->raid_disks
;
4332 for (i
= 0; i
< map
->num_members
; i
++) {
4333 /* initialized in add_to_super */
4334 set_imsm_ord_tbl_ent(map
, i
, IMSM_ORD_REBUILD
);
4336 mpb
->num_raid_devs
++;
4339 dv
->index
= super
->current_vol
;
4340 dv
->next
= super
->devlist
;
4341 super
->devlist
= dv
;
4343 imsm_update_version_info(super
);
4348 static int init_super_imsm(struct supertype
*st
, mdu_array_info_t
*info
,
4349 unsigned long long size
, char *name
,
4350 char *homehost
, int *uuid
)
4352 /* This is primarily called by Create when creating a new array.
4353 * We will then get add_to_super called for each component, and then
4354 * write_init_super called to write it out to each device.
4355 * For IMSM, Create can create on fresh devices or on a pre-existing
4357 * To create on a pre-existing array a different method will be called.
4358 * This one is just for fresh drives.
4360 struct intel_super
*super
;
4361 struct imsm_super
*mpb
;
4366 return init_super_imsm_volume(st
, info
, size
, name
, homehost
, uuid
);
4369 mpb_size
= disks_to_mpb_size(info
->nr_disks
);
4373 super
= alloc_super();
4374 if (super
&& posix_memalign(&super
->buf
, 512, mpb_size
) != 0) {
4379 fprintf(stderr
, Name
4380 ": %s could not allocate superblock\n", __func__
);
4383 if (posix_memalign(&super
->migr_rec_buf
, 512, 512) != 0) {
4384 fprintf(stderr
, Name
4385 ": %s could not allocate migr_rec buffer\n", __func__
);
4390 memset(super
->buf
, 0, mpb_size
);
4392 mpb
->mpb_size
= __cpu_to_le32(mpb_size
);
4396 /* zeroing superblock */
4400 mpb
->attributes
= MPB_ATTRIB_CHECKSUM_VERIFY
;
4402 version
= (char *) mpb
->sig
;
4403 strcpy(version
, MPB_SIGNATURE
);
4404 version
+= strlen(MPB_SIGNATURE
);
4405 strcpy(version
, MPB_VERSION_RAID0
);
4411 static int add_to_super_imsm_volume(struct supertype
*st
, mdu_disk_info_t
*dk
,
4412 int fd
, char *devname
)
4414 struct intel_super
*super
= st
->sb
;
4415 struct imsm_super
*mpb
= super
->anchor
;
4416 struct imsm_disk
*_disk
;
4417 struct imsm_dev
*dev
;
4418 struct imsm_map
*map
;
4422 dev
= get_imsm_dev(super
, super
->current_vol
);
4423 map
= get_imsm_map(dev
, 0);
4425 if (! (dk
->state
& (1<<MD_DISK_SYNC
))) {
4426 fprintf(stderr
, Name
": %s: Cannot add spare devices to IMSM volume\n",
4432 /* we're doing autolayout so grab the pre-marked (in
4433 * validate_geometry) raid_disk
4435 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
4436 if (dl
->raiddisk
== dk
->raid_disk
)
4439 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
4440 if (dl
->major
== dk
->major
&&
4441 dl
->minor
== dk
->minor
)
4446 fprintf(stderr
, Name
": %s is not a member of the same container\n", devname
);
4450 /* add a pristine spare to the metadata */
4451 if (dl
->index
< 0) {
4452 dl
->index
= super
->anchor
->num_disks
;
4453 super
->anchor
->num_disks
++;
4455 /* Check the device has not already been added */
4456 slot
= get_imsm_disk_slot(map
, dl
->index
);
4458 (get_imsm_ord_tbl_ent(dev
, slot
, -1) & IMSM_ORD_REBUILD
) == 0) {
4459 fprintf(stderr
, Name
": %s has been included in this array twice\n",
4463 set_imsm_ord_tbl_ent(map
, dk
->raid_disk
, dl
->index
);
4464 dl
->disk
.status
= CONFIGURED_DISK
;
4466 /* update size of 'missing' disks to be at least as large as the
4467 * largest acitve member (we only have dummy missing disks when
4468 * creating the first volume)
4470 if (super
->current_vol
== 0) {
4471 for (df
= super
->missing
; df
; df
= df
->next
) {
4472 if (dl
->disk
.total_blocks
> df
->disk
.total_blocks
)
4473 df
->disk
.total_blocks
= dl
->disk
.total_blocks
;
4474 _disk
= __get_imsm_disk(mpb
, df
->index
);
4479 /* refresh unset/failed slots to point to valid 'missing' entries */
4480 for (df
= super
->missing
; df
; df
= df
->next
)
4481 for (slot
= 0; slot
< mpb
->num_disks
; slot
++) {
4482 __u32 ord
= get_imsm_ord_tbl_ent(dev
, slot
, -1);
4484 if ((ord
& IMSM_ORD_REBUILD
) == 0)
4486 set_imsm_ord_tbl_ent(map
, slot
, df
->index
| IMSM_ORD_REBUILD
);
4487 dprintf("set slot:%d to missing disk:%d\n", slot
, df
->index
);
4491 /* if we are creating the first raid device update the family number */
4492 if (super
->current_vol
== 0) {
4494 struct imsm_dev
*_dev
= __get_imsm_dev(mpb
, 0);
4496 _disk
= __get_imsm_disk(mpb
, dl
->index
);
4497 if (!_dev
|| !_disk
) {
4498 fprintf(stderr
, Name
": BUG mpb setup error\n");
4504 sum
+= __gen_imsm_checksum(mpb
);
4505 mpb
->family_num
= __cpu_to_le32(sum
);
4506 mpb
->orig_family_num
= mpb
->family_num
;
4508 super
->current_disk
= dl
;
4513 * Function marks disk as spare and restores disk serial
4514 * in case it was previously marked as failed by takeover operation
4516 * -1 : critical error
4517 * 0 : disk is marked as spare but serial is not set
4520 int mark_spare(struct dl
*disk
)
4522 __u8 serial
[MAX_RAID_SERIAL_LEN
];
4529 if (!imsm_read_serial(disk
->fd
, NULL
, serial
)) {
4530 /* Restore disk serial number, because takeover marks disk
4531 * as failed and adds to serial ':0' before it becomes
4534 serialcpy(disk
->serial
, serial
);
4535 serialcpy(disk
->disk
.serial
, serial
);
4538 disk
->disk
.status
= SPARE_DISK
;
4544 static int add_to_super_imsm(struct supertype
*st
, mdu_disk_info_t
*dk
,
4545 int fd
, char *devname
)
4547 struct intel_super
*super
= st
->sb
;
4549 unsigned long long size
;
4554 /* If we are on an RAID enabled platform check that the disk is
4555 * attached to the raid controller.
4556 * We do not need to test disks attachment for container based additions,
4557 * they shall be already tested when container was created/assembled.
4559 rv
= find_intel_hba_capability(fd
, super
, devname
);
4560 /* no orom/efi or non-intel hba of the disk */
4562 dprintf("capability: %p fd: %d ret: %d\n",
4563 super
->orom
, fd
, rv
);
4567 if (super
->current_vol
>= 0)
4568 return add_to_super_imsm_volume(st
, dk
, fd
, devname
);
4571 dd
= malloc(sizeof(*dd
));
4574 Name
": malloc failed %s:%d.\n", __func__
, __LINE__
);
4577 memset(dd
, 0, sizeof(*dd
));
4578 dd
->major
= major(stb
.st_rdev
);
4579 dd
->minor
= minor(stb
.st_rdev
);
4580 dd
->devname
= devname
? strdup(devname
) : NULL
;
4583 dd
->action
= DISK_ADD
;
4584 rv
= imsm_read_serial(fd
, devname
, dd
->serial
);
4587 Name
": failed to retrieve scsi serial, aborting\n");
4592 get_dev_size(fd
, NULL
, &size
);
4594 serialcpy(dd
->disk
.serial
, dd
->serial
);
4595 dd
->disk
.total_blocks
= __cpu_to_le32(size
);
4597 if (sysfs_disk_to_scsi_id(fd
, &id
) == 0)
4598 dd
->disk
.scsi_id
= __cpu_to_le32(id
);
4600 dd
->disk
.scsi_id
= __cpu_to_le32(0);
4602 if (st
->update_tail
) {
4603 dd
->next
= super
->disk_mgmt_list
;
4604 super
->disk_mgmt_list
= dd
;
4606 dd
->next
= super
->disks
;
4608 super
->updates_pending
++;
4615 static int remove_from_super_imsm(struct supertype
*st
, mdu_disk_info_t
*dk
)
4617 struct intel_super
*super
= st
->sb
;
4620 /* remove from super works only in mdmon - for communication
4621 * manager - monitor. Check if communication memory buffer
4624 if (!st
->update_tail
) {
4626 Name
": %s shall be used in mdmon context only"
4627 "(line %d).\n", __func__
, __LINE__
);
4630 dd
= malloc(sizeof(*dd
));
4633 Name
": malloc failed %s:%d.\n", __func__
, __LINE__
);
4636 memset(dd
, 0, sizeof(*dd
));
4637 dd
->major
= dk
->major
;
4638 dd
->minor
= dk
->minor
;
4641 dd
->action
= DISK_REMOVE
;
4643 dd
->next
= super
->disk_mgmt_list
;
4644 super
->disk_mgmt_list
= dd
;
4650 static int store_imsm_mpb(int fd
, struct imsm_super
*mpb
);
4654 struct imsm_super anchor
;
4655 } spare_record
__attribute__ ((aligned(512)));
4657 /* spare records have their own family number and do not have any defined raid
4660 static int write_super_imsm_spares(struct intel_super
*super
, int doclose
)
4662 struct imsm_super
*mpb
= super
->anchor
;
4663 struct imsm_super
*spare
= &spare_record
.anchor
;
4667 spare
->mpb_size
= __cpu_to_le32(sizeof(struct imsm_super
)),
4668 spare
->generation_num
= __cpu_to_le32(1UL),
4669 spare
->attributes
= MPB_ATTRIB_CHECKSUM_VERIFY
;
4670 spare
->num_disks
= 1,
4671 spare
->num_raid_devs
= 0,
4672 spare
->cache_size
= mpb
->cache_size
,
4673 spare
->pwr_cycle_count
= __cpu_to_le32(1),
4675 snprintf((char *) spare
->sig
, MAX_SIGNATURE_LENGTH
,
4676 MPB_SIGNATURE MPB_VERSION_RAID0
);
4678 for (d
= super
->disks
; d
; d
= d
->next
) {
4682 spare
->disk
[0] = d
->disk
;
4683 sum
= __gen_imsm_checksum(spare
);
4684 spare
->family_num
= __cpu_to_le32(sum
);
4685 spare
->orig_family_num
= 0;
4686 sum
= __gen_imsm_checksum(spare
);
4687 spare
->check_sum
= __cpu_to_le32(sum
);
4689 if (store_imsm_mpb(d
->fd
, spare
)) {
4690 fprintf(stderr
, "%s: failed for device %d:%d %s\n",
4691 __func__
, d
->major
, d
->minor
, strerror(errno
));
4703 static int write_super_imsm(struct supertype
*st
, int doclose
)
4705 struct intel_super
*super
= st
->sb
;
4706 struct imsm_super
*mpb
= super
->anchor
;
4712 __u32 mpb_size
= sizeof(struct imsm_super
) - sizeof(struct imsm_disk
);
4714 int clear_migration_record
= 1;
4716 /* 'generation' is incremented everytime the metadata is written */
4717 generation
= __le32_to_cpu(mpb
->generation_num
);
4719 mpb
->generation_num
= __cpu_to_le32(generation
);
4721 /* fix up cases where previous mdadm releases failed to set
4724 if (mpb
->orig_family_num
== 0)
4725 mpb
->orig_family_num
= mpb
->family_num
;
4727 for (d
= super
->disks
; d
; d
= d
->next
) {
4731 mpb
->disk
[d
->index
] = d
->disk
;
4735 for (d
= super
->missing
; d
; d
= d
->next
) {
4736 mpb
->disk
[d
->index
] = d
->disk
;
4739 mpb
->num_disks
= num_disks
;
4740 mpb_size
+= sizeof(struct imsm_disk
) * mpb
->num_disks
;
4742 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
4743 struct imsm_dev
*dev
= __get_imsm_dev(mpb
, i
);
4744 struct imsm_dev
*dev2
= get_imsm_dev(super
, i
);
4746 imsm_copy_dev(dev
, dev2
);
4747 mpb_size
+= sizeof_imsm_dev(dev
, 0);
4749 if (is_gen_migration(dev2
))
4750 clear_migration_record
= 0;
4752 mpb_size
+= __le32_to_cpu(mpb
->bbm_log_size
);
4753 mpb
->mpb_size
= __cpu_to_le32(mpb_size
);
4755 /* recalculate checksum */
4756 sum
= __gen_imsm_checksum(mpb
);
4757 mpb
->check_sum
= __cpu_to_le32(sum
);
4759 if (clear_migration_record
)
4760 memset(super
->migr_rec_buf
, 0, 512);
4762 /* write the mpb for disks that compose raid devices */
4763 for (d
= super
->disks
; d
; d
= d
->next
) {
4764 if (d
->index
< 0 || is_failed(&d
->disk
))
4766 if (store_imsm_mpb(d
->fd
, mpb
))
4767 fprintf(stderr
, "%s: failed for device %d:%d %s\n",
4768 __func__
, d
->major
, d
->minor
, strerror(errno
));
4769 if (clear_migration_record
) {
4770 unsigned long long dsize
;
4772 get_dev_size(d
->fd
, NULL
, &dsize
);
4773 if (lseek64(d
->fd
, dsize
- 512, SEEK_SET
) >= 0) {
4774 if (write(d
->fd
, super
->migr_rec_buf
, 512) != 512)
4775 perror("Write migr_rec failed");
4785 return write_super_imsm_spares(super
, doclose
);
4791 static int create_array(struct supertype
*st
, int dev_idx
)
4794 struct imsm_update_create_array
*u
;
4795 struct intel_super
*super
= st
->sb
;
4796 struct imsm_dev
*dev
= get_imsm_dev(super
, dev_idx
);
4797 struct imsm_map
*map
= get_imsm_map(dev
, 0);
4798 struct disk_info
*inf
;
4799 struct imsm_disk
*disk
;
4802 len
= sizeof(*u
) - sizeof(*dev
) + sizeof_imsm_dev(dev
, 0) +
4803 sizeof(*inf
) * map
->num_members
;
4806 fprintf(stderr
, "%s: failed to allocate update buffer\n",
4811 u
->type
= update_create_array
;
4812 u
->dev_idx
= dev_idx
;
4813 imsm_copy_dev(&u
->dev
, dev
);
4814 inf
= get_disk_info(u
);
4815 for (i
= 0; i
< map
->num_members
; i
++) {
4816 int idx
= get_imsm_disk_idx(dev
, i
, -1);
4818 disk
= get_imsm_disk(super
, idx
);
4819 serialcpy(inf
[i
].serial
, disk
->serial
);
4821 append_metadata_update(st
, u
, len
);
4826 static int mgmt_disk(struct supertype
*st
)
4828 struct intel_super
*super
= st
->sb
;
4830 struct imsm_update_add_remove_disk
*u
;
4832 if (!super
->disk_mgmt_list
)
4838 fprintf(stderr
, "%s: failed to allocate update buffer\n",
4843 u
->type
= update_add_remove_disk
;
4844 append_metadata_update(st
, u
, len
);
4849 static int write_init_super_imsm(struct supertype
*st
)
4851 struct intel_super
*super
= st
->sb
;
4852 int current_vol
= super
->current_vol
;
4854 /* we are done with current_vol reset it to point st at the container */
4855 super
->current_vol
= -1;
4857 if (st
->update_tail
) {
4858 /* queue the recently created array / added disk
4859 * as a metadata update */
4862 /* determine if we are creating a volume or adding a disk */
4863 if (current_vol
< 0) {
4864 /* in the mgmt (add/remove) disk case we are running
4865 * in mdmon context, so don't close fd's
4867 return mgmt_disk(st
);
4869 rv
= create_array(st
, current_vol
);
4874 for (d
= super
->disks
; d
; d
= d
->next
)
4875 Kill(d
->devname
, NULL
, 0, 1, 1);
4876 return write_super_imsm(st
, 1);
4881 static int store_super_imsm(struct supertype
*st
, int fd
)
4883 struct intel_super
*super
= st
->sb
;
4884 struct imsm_super
*mpb
= super
? super
->anchor
: NULL
;
4890 return store_imsm_mpb(fd
, mpb
);
4896 static int imsm_bbm_log_size(struct imsm_super
*mpb
)
4898 return __le32_to_cpu(mpb
->bbm_log_size
);
4902 static int validate_geometry_imsm_container(struct supertype
*st
, int level
,
4903 int layout
, int raiddisks
, int chunk
,
4904 unsigned long long size
, char *dev
,
4905 unsigned long long *freesize
,
4909 unsigned long long ldsize
;
4910 struct intel_super
*super
=NULL
;
4913 if (level
!= LEVEL_CONTAINER
)
4918 fd
= open(dev
, O_RDONLY
|O_EXCL
, 0);
4921 fprintf(stderr
, Name
": imsm: Cannot open %s: %s\n",
4922 dev
, strerror(errno
));
4925 if (!get_dev_size(fd
, dev
, &ldsize
)) {
4930 /* capabilities retrieve could be possible
4931 * note that there is no fd for the disks in array.
4933 super
= alloc_super();
4936 Name
": malloc of %zu failed.\n",
4942 rv
= find_intel_hba_capability(fd
, super
, verbose
? dev
: NULL
);
4946 fd2devname(fd
, str
);
4947 dprintf("validate_geometry_imsm_container: fd: %d %s orom: %p rv: %d raiddisk: %d\n",
4948 fd
, str
, super
->orom
, rv
, raiddisks
);
4950 /* no orom/efi or non-intel hba of the disk */
4956 if (super
->orom
&& raiddisks
> super
->orom
->tds
) {
4958 fprintf(stderr
, Name
": %d exceeds maximum number of"
4959 " platform supported disks: %d\n",
4960 raiddisks
, super
->orom
->tds
);
4966 *freesize
= avail_size_imsm(st
, ldsize
>> 9);
4972 static unsigned long long find_size(struct extent
*e
, int *idx
, int num_extents
)
4974 const unsigned long long base_start
= e
[*idx
].start
;
4975 unsigned long long end
= base_start
+ e
[*idx
].size
;
4978 if (base_start
== end
)
4982 for (i
= *idx
; i
< num_extents
; i
++) {
4983 /* extend overlapping extents */
4984 if (e
[i
].start
>= base_start
&&
4985 e
[i
].start
<= end
) {
4988 if (e
[i
].start
+ e
[i
].size
> end
)
4989 end
= e
[i
].start
+ e
[i
].size
;
4990 } else if (e
[i
].start
> end
) {
4996 return end
- base_start
;
4999 static unsigned long long merge_extents(struct intel_super
*super
, int sum_extents
)
5001 /* build a composite disk with all known extents and generate a new
5002 * 'maxsize' given the "all disks in an array must share a common start
5003 * offset" constraint
5005 struct extent
*e
= calloc(sum_extents
, sizeof(*e
));
5009 unsigned long long pos
;
5010 unsigned long long start
= 0;
5011 unsigned long long maxsize
;
5012 unsigned long reserve
;
5017 /* coalesce and sort all extents. also, check to see if we need to
5018 * reserve space between member arrays
5021 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
5024 for (i
= 0; i
< dl
->extent_cnt
; i
++)
5027 qsort(e
, sum_extents
, sizeof(*e
), cmp_extent
);
5032 while (i
< sum_extents
) {
5033 e
[j
].start
= e
[i
].start
;
5034 e
[j
].size
= find_size(e
, &i
, sum_extents
);
5036 if (e
[j
-1].size
== 0)
5045 unsigned long long esize
;
5047 esize
= e
[i
].start
- pos
;
5048 if (esize
>= maxsize
) {
5053 pos
= e
[i
].start
+ e
[i
].size
;
5055 } while (e
[i
-1].size
);
5061 /* FIXME assumes volume at offset 0 is the first volume in a
5064 if (start_extent
> 0)
5065 reserve
= IMSM_RESERVED_SECTORS
; /* gap between raid regions */
5069 if (maxsize
< reserve
)
5072 super
->create_offset
= ~((__u32
) 0);
5073 if (start
+ reserve
> super
->create_offset
)
5074 return 0; /* start overflows create_offset */
5075 super
->create_offset
= start
+ reserve
;
5077 return maxsize
- reserve
;
5080 static int is_raid_level_supported(const struct imsm_orom
*orom
, int level
, int raiddisks
)
5082 if (level
< 0 || level
== 6 || level
== 4)
5085 /* if we have an orom prevent invalid raid levels */
5088 case 0: return imsm_orom_has_raid0(orom
);
5091 return imsm_orom_has_raid1e(orom
);
5092 return imsm_orom_has_raid1(orom
) && raiddisks
== 2;
5093 case 10: return imsm_orom_has_raid10(orom
) && raiddisks
== 4;
5094 case 5: return imsm_orom_has_raid5(orom
) && raiddisks
> 2;
5097 return 1; /* not on an Intel RAID platform so anything goes */
5102 static int imsm_default_chunk(const struct imsm_orom
*orom
)
5104 /* up to 512 if the plaform supports it, otherwise the platform max.
5105 * 128 if no platform detected
5107 int fs
= max(7, orom
? fls(orom
->sss
) : 0);
5109 return min(512, (1 << fs
));
5112 #define pr_vrb(fmt, arg...) (void) (verbose && fprintf(stderr, Name fmt, ##arg))
5114 validate_geometry_imsm_orom(struct intel_super
*super
, int level
, int layout
,
5115 int raiddisks
, int *chunk
, int verbose
)
5117 /* check/set platform and metadata limits/defaults */
5118 if (super
->orom
&& raiddisks
> super
->orom
->dpa
) {
5119 pr_vrb(": platform supports a maximum of %d disks per array\n",
5124 /* capabilities of OROM tested - copied from validate_geometry_imsm_volume */
5125 if (!is_raid_level_supported(super
->orom
, level
, raiddisks
)) {
5126 pr_vrb(": platform does not support raid%d with %d disk%s\n",
5127 level
, raiddisks
, raiddisks
> 1 ? "s" : "");
5131 if (chunk
&& (*chunk
== 0 || *chunk
== UnSet
))
5132 *chunk
= imsm_default_chunk(super
->orom
);
5134 if (super
->orom
&& chunk
&& !imsm_orom_has_chunk(super
->orom
, *chunk
)) {
5135 pr_vrb(": platform does not support a chunk size of: "
5140 if (layout
!= imsm_level_to_layout(level
)) {
5142 pr_vrb(": imsm raid 5 only supports the left-asymmetric layout\n");
5143 else if (level
== 10)
5144 pr_vrb(": imsm raid 10 only supports the n2 layout\n");
5146 pr_vrb(": imsm unknown layout %#x for this raid level %d\n",
5153 /* validate_geometry_imsm_volume - lifted from validate_geometry_ddf_bvd
5154 * FIX ME add ahci details
5156 static int validate_geometry_imsm_volume(struct supertype
*st
, int level
,
5157 int layout
, int raiddisks
, int *chunk
,
5158 unsigned long long size
, char *dev
,
5159 unsigned long long *freesize
,
5163 struct intel_super
*super
= st
->sb
;
5164 struct imsm_super
*mpb
;
5166 unsigned long long pos
= 0;
5167 unsigned long long maxsize
;
5171 /* We must have the container info already read in. */
5175 mpb
= super
->anchor
;
5177 if (mpb
->num_raid_devs
> 0 && mpb
->num_disks
!= raiddisks
) {
5178 fprintf(stderr
, Name
": the option-rom requires all "
5179 "member disks to be a member of all volumes.\n");
5183 if (!validate_geometry_imsm_orom(super
, level
, layout
, raiddisks
, chunk
, verbose
)) {
5184 fprintf(stderr
, Name
": RAID gemetry validation failed. "
5185 "Cannot proceed with the action(s).\n");
5189 /* General test: make sure there is space for
5190 * 'raiddisks' device extents of size 'size' at a given
5193 unsigned long long minsize
= size
;
5194 unsigned long long start_offset
= MaxSector
;
5197 minsize
= MPB_SECTOR_CNT
+ IMSM_RESERVED_SECTORS
;
5198 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
5203 e
= get_extents(super
, dl
);
5206 unsigned long long esize
;
5207 esize
= e
[i
].start
- pos
;
5208 if (esize
>= minsize
)
5210 if (found
&& start_offset
== MaxSector
) {
5213 } else if (found
&& pos
!= start_offset
) {
5217 pos
= e
[i
].start
+ e
[i
].size
;
5219 } while (e
[i
-1].size
);
5224 if (dcnt
< raiddisks
) {
5226 fprintf(stderr
, Name
": imsm: Not enough "
5227 "devices with space for this array "
5235 /* This device must be a member of the set */
5236 if (stat(dev
, &stb
) < 0)
5238 if ((S_IFMT
& stb
.st_mode
) != S_IFBLK
)
5240 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
5241 if (dl
->major
== (int)major(stb
.st_rdev
) &&
5242 dl
->minor
== (int)minor(stb
.st_rdev
))
5247 fprintf(stderr
, Name
": %s is not in the "
5248 "same imsm set\n", dev
);
5250 } else if (super
->orom
&& dl
->index
< 0 && mpb
->num_raid_devs
) {
5251 /* If a volume is present then the current creation attempt
5252 * cannot incorporate new spares because the orom may not
5253 * understand this configuration (all member disks must be
5254 * members of each array in the container).
5256 fprintf(stderr
, Name
": %s is a spare and a volume"
5257 " is already defined for this container\n", dev
);
5258 fprintf(stderr
, Name
": The option-rom requires all member"
5259 " disks to be a member of all volumes\n");
5263 /* retrieve the largest free space block */
5264 e
= get_extents(super
, dl
);
5269 unsigned long long esize
;
5271 esize
= e
[i
].start
- pos
;
5272 if (esize
>= maxsize
)
5274 pos
= e
[i
].start
+ e
[i
].size
;
5276 } while (e
[i
-1].size
);
5281 fprintf(stderr
, Name
": unable to determine free space for: %s\n",
5285 if (maxsize
< size
) {
5287 fprintf(stderr
, Name
": %s not enough space (%llu < %llu)\n",
5288 dev
, maxsize
, size
);
5292 /* count total number of extents for merge */
5294 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
5296 i
+= dl
->extent_cnt
;
5298 maxsize
= merge_extents(super
, i
);
5300 if (!check_env("IMSM_NO_PLATFORM") &&
5301 mpb
->num_raid_devs
> 0 && size
&& size
!= maxsize
) {
5302 fprintf(stderr
, Name
": attempting to create a second "
5303 "volume with size less then remaining space. "
5308 if (maxsize
< size
|| maxsize
== 0) {
5310 fprintf(stderr
, Name
": not enough space after merge (%llu < %llu)\n",
5315 *freesize
= maxsize
;
5320 static int reserve_space(struct supertype
*st
, int raiddisks
,
5321 unsigned long long size
, int chunk
,
5322 unsigned long long *freesize
)
5324 struct intel_super
*super
= st
->sb
;
5325 struct imsm_super
*mpb
= super
->anchor
;
5330 unsigned long long maxsize
;
5331 unsigned long long minsize
;
5335 /* find the largest common start free region of the possible disks */
5339 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
5345 /* don't activate new spares if we are orom constrained
5346 * and there is already a volume active in the container
5348 if (super
->orom
&& dl
->index
< 0 && mpb
->num_raid_devs
)
5351 e
= get_extents(super
, dl
);
5354 for (i
= 1; e
[i
-1].size
; i
++)
5362 maxsize
= merge_extents(super
, extent_cnt
);
5366 minsize
= chunk
* 2;
5368 if (cnt
< raiddisks
||
5369 (super
->orom
&& used
&& used
!= raiddisks
) ||
5370 maxsize
< minsize
||
5372 fprintf(stderr
, Name
": not enough devices with space to create array.\n");
5373 return 0; /* No enough free spaces large enough */
5385 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
5387 dl
->raiddisk
= cnt
++;
5394 static int validate_geometry_imsm(struct supertype
*st
, int level
, int layout
,
5395 int raiddisks
, int *chunk
, unsigned long long size
,
5396 char *dev
, unsigned long long *freesize
,
5404 * if given unused devices create a container
5405 * if given given devices in a container create a member volume
5407 if (level
== LEVEL_CONTAINER
) {
5408 /* Must be a fresh device to add to a container */
5409 return validate_geometry_imsm_container(st
, level
, layout
,
5411 chunk
?*chunk
:0, size
,
5418 if (!validate_geometry_imsm_orom(st
->sb
, level
, layout
,
5422 /* we are being asked to automatically layout a
5423 * new volume based on the current contents of
5424 * the container. If the the parameters can be
5425 * satisfied reserve_space will record the disks,
5426 * start offset, and size of the volume to be
5427 * created. add_to_super and getinfo_super
5428 * detect when autolayout is in progress.
5431 return reserve_space(st
, raiddisks
, size
,
5432 chunk
?*chunk
:0, freesize
);
5437 /* creating in a given container */
5438 return validate_geometry_imsm_volume(st
, level
, layout
,
5439 raiddisks
, chunk
, size
,
5440 dev
, freesize
, verbose
);
5443 /* This device needs to be a device in an 'imsm' container */
5444 fd
= open(dev
, O_RDONLY
|O_EXCL
, 0);
5448 Name
": Cannot create this array on device %s\n",
5453 if (errno
!= EBUSY
|| (fd
= open(dev
, O_RDONLY
, 0)) < 0) {
5455 fprintf(stderr
, Name
": Cannot open %s: %s\n",
5456 dev
, strerror(errno
));
5459 /* Well, it is in use by someone, maybe an 'imsm' container. */
5460 cfd
= open_container(fd
);
5464 fprintf(stderr
, Name
": Cannot use %s: It is busy\n",
5468 sra
= sysfs_read(cfd
, 0, GET_VERSION
);
5469 if (sra
&& sra
->array
.major_version
== -1 &&
5470 strcmp(sra
->text_version
, "imsm") == 0)
5474 /* This is a member of a imsm container. Load the container
5475 * and try to create a volume
5477 struct intel_super
*super
;
5479 if (load_super_imsm_all(st
, cfd
, (void **) &super
, NULL
) == 0) {
5481 st
->container_dev
= fd2devnum(cfd
);
5483 return validate_geometry_imsm_volume(st
, level
, layout
,
5492 fprintf(stderr
, Name
": failed container membership check\n");
5498 static void default_geometry_imsm(struct supertype
*st
, int *level
, int *layout
, int *chunk
)
5500 struct intel_super
*super
= st
->sb
;
5502 if (level
&& *level
== UnSet
)
5503 *level
= LEVEL_CONTAINER
;
5505 if (level
&& layout
&& *layout
== UnSet
)
5506 *layout
= imsm_level_to_layout(*level
);
5508 if (chunk
&& (*chunk
== UnSet
|| *chunk
== 0))
5509 *chunk
= imsm_default_chunk(super
->orom
);
5512 static void handle_missing(struct intel_super
*super
, struct imsm_dev
*dev
);
5514 static int kill_subarray_imsm(struct supertype
*st
)
5516 /* remove the subarray currently referenced by ->current_vol */
5518 struct intel_dev
**dp
;
5519 struct intel_super
*super
= st
->sb
;
5520 __u8 current_vol
= super
->current_vol
;
5521 struct imsm_super
*mpb
= super
->anchor
;
5523 if (super
->current_vol
< 0)
5525 super
->current_vol
= -1; /* invalidate subarray cursor */
5527 /* block deletions that would change the uuid of active subarrays
5529 * FIXME when immutable ids are available, but note that we'll
5530 * also need to fixup the invalidated/active subarray indexes in
5533 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
5536 if (i
< current_vol
)
5538 sprintf(subarray
, "%u", i
);
5539 if (is_subarray_active(subarray
, st
->devname
)) {
5541 Name
": deleting subarray-%d would change the UUID of active subarray-%d, aborting\n",
5548 if (st
->update_tail
) {
5549 struct imsm_update_kill_array
*u
= malloc(sizeof(*u
));
5553 u
->type
= update_kill_array
;
5554 u
->dev_idx
= current_vol
;
5555 append_metadata_update(st
, u
, sizeof(*u
));
5560 for (dp
= &super
->devlist
; *dp
;)
5561 if ((*dp
)->index
== current_vol
) {
5564 handle_missing(super
, (*dp
)->dev
);
5565 if ((*dp
)->index
> current_vol
)
5570 /* no more raid devices, all active components are now spares,
5571 * but of course failed are still failed
5573 if (--mpb
->num_raid_devs
== 0) {
5576 for (d
= super
->disks
; d
; d
= d
->next
)
5581 super
->updates_pending
++;
5586 static int update_subarray_imsm(struct supertype
*st
, char *subarray
,
5587 char *update
, struct mddev_ident
*ident
)
5589 /* update the subarray currently referenced by ->current_vol */
5590 struct intel_super
*super
= st
->sb
;
5591 struct imsm_super
*mpb
= super
->anchor
;
5593 if (strcmp(update
, "name") == 0) {
5594 char *name
= ident
->name
;
5598 if (is_subarray_active(subarray
, st
->devname
)) {
5600 Name
": Unable to update name of active subarray\n");
5604 if (!check_name(super
, name
, 0))
5607 vol
= strtoul(subarray
, &ep
, 10);
5608 if (*ep
!= '\0' || vol
>= super
->anchor
->num_raid_devs
)
5611 if (st
->update_tail
) {
5612 struct imsm_update_rename_array
*u
= malloc(sizeof(*u
));
5616 u
->type
= update_rename_array
;
5618 snprintf((char *) u
->name
, MAX_RAID_SERIAL_LEN
, "%s", name
);
5619 append_metadata_update(st
, u
, sizeof(*u
));
5621 struct imsm_dev
*dev
;
5624 dev
= get_imsm_dev(super
, vol
);
5625 snprintf((char *) dev
->volume
, MAX_RAID_SERIAL_LEN
, "%s", name
);
5626 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
5627 dev
= get_imsm_dev(super
, i
);
5628 handle_missing(super
, dev
);
5630 super
->updates_pending
++;
5638 static int is_gen_migration(struct imsm_dev
*dev
)
5643 if (!dev
->vol
.migr_state
)
5646 if (migr_type(dev
) == MIGR_GEN_MIGR
)
5651 #endif /* MDASSEMBLE */
5653 static int is_rebuilding(struct imsm_dev
*dev
)
5655 struct imsm_map
*migr_map
;
5657 if (!dev
->vol
.migr_state
)
5660 if (migr_type(dev
) != MIGR_REBUILD
)
5663 migr_map
= get_imsm_map(dev
, 1);
5665 if (migr_map
->map_state
== IMSM_T_STATE_DEGRADED
)
5671 static void update_recovery_start(struct intel_super
*super
,
5672 struct imsm_dev
*dev
,
5673 struct mdinfo
*array
)
5675 struct mdinfo
*rebuild
= NULL
;
5679 if (!is_rebuilding(dev
))
5682 /* Find the rebuild target, but punt on the dual rebuild case */
5683 for (d
= array
->devs
; d
; d
= d
->next
)
5684 if (d
->recovery_start
== 0) {
5691 /* (?) none of the disks are marked with
5692 * IMSM_ORD_REBUILD, so assume they are missing and the
5693 * disk_ord_tbl was not correctly updated
5695 dprintf("%s: failed to locate out-of-sync disk\n", __func__
);
5699 units
= __le32_to_cpu(dev
->vol
.curr_migr_unit
);
5700 rebuild
->recovery_start
= units
* blocks_per_migr_unit(super
, dev
);
5704 static int recover_backup_imsm(struct supertype
*st
, struct mdinfo
*info
);
5707 static struct mdinfo
*container_content_imsm(struct supertype
*st
, char *subarray
)
5709 /* Given a container loaded by load_super_imsm_all,
5710 * extract information about all the arrays into
5712 * If 'subarray' is given, just extract info about that array.
5714 * For each imsm_dev create an mdinfo, fill it in,
5715 * then look for matching devices in super->disks
5716 * and create appropriate device mdinfo.
5718 struct intel_super
*super
= st
->sb
;
5719 struct imsm_super
*mpb
= super
->anchor
;
5720 struct mdinfo
*rest
= NULL
;
5724 int spare_disks
= 0;
5726 /* do not assemble arrays when not all attributes are supported */
5727 if (imsm_check_attributes(mpb
->attributes
) == 0) {
5729 fprintf(stderr
, Name
": Unsupported attributes in IMSM metadata."
5730 "Arrays activation is blocked.\n");
5733 /* check for bad blocks */
5734 if (imsm_bbm_log_size(super
->anchor
)) {
5735 fprintf(stderr
, Name
": BBM log found in IMSM metadata."
5736 "Arrays activation is blocked.\n");
5741 /* count spare devices, not used in maps
5743 for (d
= super
->disks
; d
; d
= d
->next
)
5747 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
5748 struct imsm_dev
*dev
;
5749 struct imsm_map
*map
;
5750 struct imsm_map
*map2
;
5751 struct mdinfo
*this;
5756 (i
!= strtoul(subarray
, &ep
, 10) || *ep
!= '\0'))
5759 dev
= get_imsm_dev(super
, i
);
5760 map
= get_imsm_map(dev
, 0);
5761 map2
= get_imsm_map(dev
, 1);
5763 /* do not publish arrays that are in the middle of an
5764 * unsupported migration
5766 if (dev
->vol
.migr_state
&&
5767 (migr_type(dev
) == MIGR_STATE_CHANGE
)) {
5768 fprintf(stderr
, Name
": cannot assemble volume '%.16s':"
5769 " unsupported migration in progress\n",
5773 /* do not publish arrays that are not support by controller's
5777 chunk
= __le16_to_cpu(map
->blocks_per_strip
) >> 1;
5778 this = malloc(sizeof(*this));
5780 fprintf(stderr
, Name
": failed to allocate %zu bytes\n",
5785 super
->current_vol
= i
;
5786 getinfo_super_imsm_volume(st
, this, NULL
);
5789 /* mdadm does not support all metadata features- set the bit in all arrays state */
5790 if (!validate_geometry_imsm_orom(super
,
5791 get_imsm_raid_level(map
), /* RAID level */
5792 imsm_level_to_layout(get_imsm_raid_level(map
)),
5793 map
->num_members
, /* raid disks */
5796 fprintf(stderr
, Name
": IMSM RAID geometry validation"
5797 " failed. Array %s activation is blocked.\n",
5799 this->array
.state
|=
5800 (1<<MD_SB_BLOCK_CONTAINER_RESHAPE
) |
5801 (1<<MD_SB_BLOCK_VOLUME
);
5805 /* if array has bad blocks, set suitable bit in all arrays state */
5807 this->array
.state
|=
5808 (1<<MD_SB_BLOCK_CONTAINER_RESHAPE
) |
5809 (1<<MD_SB_BLOCK_VOLUME
);
5811 for (slot
= 0 ; slot
< map
->num_members
; slot
++) {
5812 unsigned long long recovery_start
;
5813 struct mdinfo
*info_d
;
5820 idx
= get_imsm_disk_idx(dev
, slot
, 0);
5821 ord
= get_imsm_ord_tbl_ent(dev
, slot
, -1);
5822 for (d
= super
->disks
; d
; d
= d
->next
)
5823 if (d
->index
== idx
)
5826 recovery_start
= MaxSector
;
5829 if (d
&& is_failed(&d
->disk
))
5831 if (ord
& IMSM_ORD_REBUILD
)
5835 * if we skip some disks the array will be assmebled degraded;
5836 * reset resync start to avoid a dirty-degraded
5837 * situation when performing the intial sync
5839 * FIXME handle dirty degraded
5841 if ((skip
|| recovery_start
== 0) && !dev
->vol
.dirty
)
5842 this->resync_start
= MaxSector
;
5846 info_d
= calloc(1, sizeof(*info_d
));
5848 fprintf(stderr
, Name
": failed to allocate disk"
5849 " for volume %.16s\n", dev
->volume
);
5850 info_d
= this->devs
;
5852 struct mdinfo
*d
= info_d
->next
;
5861 info_d
->next
= this->devs
;
5862 this->devs
= info_d
;
5864 info_d
->disk
.number
= d
->index
;
5865 info_d
->disk
.major
= d
->major
;
5866 info_d
->disk
.minor
= d
->minor
;
5867 info_d
->disk
.raid_disk
= slot
;
5868 info_d
->recovery_start
= recovery_start
;
5870 if (slot
< map2
->num_members
)
5871 info_d
->disk
.state
= (1 << MD_DISK_ACTIVE
);
5873 this->array
.spare_disks
++;
5875 if (slot
< map
->num_members
)
5876 info_d
->disk
.state
= (1 << MD_DISK_ACTIVE
);
5878 this->array
.spare_disks
++;
5880 if (info_d
->recovery_start
== MaxSector
)
5881 this->array
.working_disks
++;
5883 info_d
->events
= __le32_to_cpu(mpb
->generation_num
);
5884 info_d
->data_offset
= __le32_to_cpu(map
->pba_of_lba0
);
5885 info_d
->component_size
= __le32_to_cpu(map
->blocks_per_member
);
5887 /* now that the disk list is up-to-date fixup recovery_start */
5888 update_recovery_start(super
, dev
, this);
5889 this->array
.spare_disks
+= spare_disks
;
5892 /* check for reshape */
5893 if (this->reshape_active
== 1)
5894 recover_backup_imsm(st
, this);
5903 static __u8
imsm_check_degraded(struct intel_super
*super
, struct imsm_dev
*dev
, int failed
)
5905 struct imsm_map
*map
= get_imsm_map(dev
, 0);
5908 return map
->map_state
== IMSM_T_STATE_UNINITIALIZED
?
5909 IMSM_T_STATE_UNINITIALIZED
: IMSM_T_STATE_NORMAL
;
5911 switch (get_imsm_raid_level(map
)) {
5913 return IMSM_T_STATE_FAILED
;
5916 if (failed
< map
->num_members
)
5917 return IMSM_T_STATE_DEGRADED
;
5919 return IMSM_T_STATE_FAILED
;
5924 * check to see if any mirrors have failed, otherwise we
5925 * are degraded. Even numbered slots are mirrored on
5929 /* gcc -Os complains that this is unused */
5930 int insync
= insync
;
5932 for (i
= 0; i
< map
->num_members
; i
++) {
5933 __u32 ord
= get_imsm_ord_tbl_ent(dev
, i
, -1);
5934 int idx
= ord_to_idx(ord
);
5935 struct imsm_disk
*disk
;
5937 /* reset the potential in-sync count on even-numbered
5938 * slots. num_copies is always 2 for imsm raid10
5943 disk
= get_imsm_disk(super
, idx
);
5944 if (!disk
|| is_failed(disk
) || ord
& IMSM_ORD_REBUILD
)
5947 /* no in-sync disks left in this mirror the
5951 return IMSM_T_STATE_FAILED
;
5954 return IMSM_T_STATE_DEGRADED
;
5958 return IMSM_T_STATE_DEGRADED
;
5960 return IMSM_T_STATE_FAILED
;
5966 return map
->map_state
;
5969 static int imsm_count_failed(struct intel_super
*super
, struct imsm_dev
*dev
)
5973 struct imsm_disk
*disk
;
5974 struct imsm_map
*map
= get_imsm_map(dev
, 0);
5975 struct imsm_map
*prev
= get_imsm_map(dev
, dev
->vol
.migr_state
);
5976 struct imsm_map
*map_for_loop
;
5980 /* at the beginning of migration we set IMSM_ORD_REBUILD on
5981 * disks that are being rebuilt. New failures are recorded to
5982 * map[0]. So we look through all the disks we started with and
5983 * see if any failures are still present, or if any new ones
5986 map_for_loop
= prev
;
5987 if (is_gen_migration(dev
))
5988 if (prev
&& (map
->num_members
> prev
->num_members
))
5991 for (i
= 0; i
< map_for_loop
->num_members
; i
++) {
5993 if (i
< prev
->num_members
)
5994 ord
|= __le32_to_cpu(prev
->disk_ord_tbl
[i
]);
5995 if (i
< map
->num_members
)
5996 ord
|= __le32_to_cpu(map
->disk_ord_tbl
[i
]);
5997 idx
= ord_to_idx(ord
);
5999 disk
= get_imsm_disk(super
, idx
);
6000 if (!disk
|| is_failed(disk
) || ord
& IMSM_ORD_REBUILD
)
6008 static int imsm_open_new(struct supertype
*c
, struct active_array
*a
,
6011 struct intel_super
*super
= c
->sb
;
6012 struct imsm_super
*mpb
= super
->anchor
;
6014 if (atoi(inst
) >= mpb
->num_raid_devs
) {
6015 fprintf(stderr
, "%s: subarry index %d, out of range\n",
6016 __func__
, atoi(inst
));
6020 dprintf("imsm: open_new %s\n", inst
);
6021 a
->info
.container_member
= atoi(inst
);
6025 static int is_resyncing(struct imsm_dev
*dev
)
6027 struct imsm_map
*migr_map
;
6029 if (!dev
->vol
.migr_state
)
6032 if (migr_type(dev
) == MIGR_INIT
||
6033 migr_type(dev
) == MIGR_REPAIR
)
6036 if (migr_type(dev
) == MIGR_GEN_MIGR
)
6039 migr_map
= get_imsm_map(dev
, 1);
6041 if ((migr_map
->map_state
== IMSM_T_STATE_NORMAL
) &&
6042 (dev
->vol
.migr_type
!= MIGR_GEN_MIGR
))
6048 /* return true if we recorded new information */
6049 static int mark_failure(struct imsm_dev
*dev
, struct imsm_disk
*disk
, int idx
)
6053 struct imsm_map
*map
;
6054 char buf
[MAX_RAID_SERIAL_LEN
+3];
6055 unsigned int len
, shift
= 0;
6057 /* new failures are always set in map[0] */
6058 map
= get_imsm_map(dev
, 0);
6060 slot
= get_imsm_disk_slot(map
, idx
);
6064 ord
= __le32_to_cpu(map
->disk_ord_tbl
[slot
]);
6065 if (is_failed(disk
) && (ord
& IMSM_ORD_REBUILD
))
6068 memcpy(buf
, disk
->serial
, MAX_RAID_SERIAL_LEN
);
6069 buf
[MAX_RAID_SERIAL_LEN
] = '\000';
6071 if ((len
= strlen(buf
)) >= MAX_RAID_SERIAL_LEN
)
6072 shift
= len
- MAX_RAID_SERIAL_LEN
+ 1;
6073 strncpy((char *)disk
->serial
, &buf
[shift
], MAX_RAID_SERIAL_LEN
);
6075 disk
->status
|= FAILED_DISK
;
6076 set_imsm_ord_tbl_ent(map
, slot
, idx
| IMSM_ORD_REBUILD
);
6077 if (map
->failed_disk_num
== 0xff)
6078 map
->failed_disk_num
= slot
;
6082 static void mark_missing(struct imsm_dev
*dev
, struct imsm_disk
*disk
, int idx
)
6084 mark_failure(dev
, disk
, idx
);
6086 if (disk
->scsi_id
== __cpu_to_le32(~(__u32
)0))
6089 disk
->scsi_id
= __cpu_to_le32(~(__u32
)0);
6090 memmove(&disk
->serial
[0], &disk
->serial
[1], MAX_RAID_SERIAL_LEN
- 1);
6093 static void handle_missing(struct intel_super
*super
, struct imsm_dev
*dev
)
6099 if (!super
->missing
)
6101 failed
= imsm_count_failed(super
, dev
);
6102 map_state
= imsm_check_degraded(super
, dev
, failed
);
6104 dprintf("imsm: mark missing\n");
6105 end_migration(dev
, map_state
);
6106 for (dl
= super
->missing
; dl
; dl
= dl
->next
)
6107 mark_missing(dev
, &dl
->disk
, dl
->index
);
6108 super
->updates_pending
++;
6111 static unsigned long long imsm_set_array_size(struct imsm_dev
*dev
)
6113 int used_disks
= imsm_num_data_members(dev
, 0);
6114 unsigned long long array_blocks
;
6115 struct imsm_map
*map
;
6117 if (used_disks
== 0) {
6118 /* when problems occures
6119 * return current array_blocks value
6121 array_blocks
= __le32_to_cpu(dev
->size_high
);
6122 array_blocks
= array_blocks
<< 32;
6123 array_blocks
+= __le32_to_cpu(dev
->size_low
);
6125 return array_blocks
;
6128 /* set array size in metadata
6130 map
= get_imsm_map(dev
, 0);
6131 array_blocks
= map
->blocks_per_member
* used_disks
;
6133 /* round array size down to closest MB
6135 array_blocks
= (array_blocks
>> SECT_PER_MB_SHIFT
) << SECT_PER_MB_SHIFT
;
6136 dev
->size_low
= __cpu_to_le32((__u32
)array_blocks
);
6137 dev
->size_high
= __cpu_to_le32((__u32
)(array_blocks
>> 32));
6139 return array_blocks
;
6142 static void imsm_set_disk(struct active_array
*a
, int n
, int state
);
6144 static void imsm_progress_container_reshape(struct intel_super
*super
)
6146 /* if no device has a migr_state, but some device has a
6147 * different number of members than the previous device, start
6148 * changing the number of devices in this device to match
6151 struct imsm_super
*mpb
= super
->anchor
;
6152 int prev_disks
= -1;
6156 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
6157 struct imsm_dev
*dev
= get_imsm_dev(super
, i
);
6158 struct imsm_map
*map
= get_imsm_map(dev
, 0);
6159 struct imsm_map
*map2
;
6160 int prev_num_members
;
6162 if (dev
->vol
.migr_state
)
6165 if (prev_disks
== -1)
6166 prev_disks
= map
->num_members
;
6167 if (prev_disks
== map
->num_members
)
6170 /* OK, this array needs to enter reshape mode.
6171 * i.e it needs a migr_state
6174 copy_map_size
= sizeof_imsm_map(map
);
6175 prev_num_members
= map
->num_members
;
6176 map
->num_members
= prev_disks
;
6177 dev
->vol
.migr_state
= 1;
6178 dev
->vol
.curr_migr_unit
= 0;
6179 set_migr_type(dev
, MIGR_GEN_MIGR
);
6180 for (i
= prev_num_members
;
6181 i
< map
->num_members
; i
++)
6182 set_imsm_ord_tbl_ent(map
, i
, i
);
6183 map2
= get_imsm_map(dev
, 1);
6184 /* Copy the current map */
6185 memcpy(map2
, map
, copy_map_size
);
6186 map2
->num_members
= prev_num_members
;
6188 imsm_set_array_size(dev
);
6189 super
->updates_pending
++;
6193 /* Handle dirty -> clean transititions, resync and reshape. Degraded and rebuild
6194 * states are handled in imsm_set_disk() with one exception, when a
6195 * resync is stopped due to a new failure this routine will set the
6196 * 'degraded' state for the array.
6198 static int imsm_set_array_state(struct active_array
*a
, int consistent
)
6200 int inst
= a
->info
.container_member
;
6201 struct intel_super
*super
= a
->container
->sb
;
6202 struct imsm_dev
*dev
= get_imsm_dev(super
, inst
);
6203 struct imsm_map
*map
= get_imsm_map(dev
, 0);
6204 int failed
= imsm_count_failed(super
, dev
);
6205 __u8 map_state
= imsm_check_degraded(super
, dev
, failed
);
6206 __u32 blocks_per_unit
;
6208 if (dev
->vol
.migr_state
&&
6209 dev
->vol
.migr_type
== MIGR_GEN_MIGR
) {
6210 /* array state change is blocked due to reshape action
6212 * - abort the reshape (if last_checkpoint is 0 and action!= reshape)
6213 * - finish the reshape (if last_checkpoint is big and action != reshape)
6214 * - update curr_migr_unit
6216 if (a
->curr_action
== reshape
) {
6217 /* still reshaping, maybe update curr_migr_unit */
6218 goto mark_checkpoint
;
6220 if (a
->last_checkpoint
== 0 && a
->prev_action
== reshape
) {
6221 /* for some reason we aborted the reshape.
6223 * disable automatic metadata rollback
6224 * user action is required to recover process
6227 struct imsm_map
*map2
= get_imsm_map(dev
, 1);
6228 dev
->vol
.migr_state
= 0;
6229 set_migr_type(dev
, 0);
6230 dev
->vol
.curr_migr_unit
= 0;
6231 memcpy(map
, map2
, sizeof_imsm_map(map2
));
6232 super
->updates_pending
++;
6235 if (a
->last_checkpoint
>= a
->info
.component_size
) {
6236 unsigned long long array_blocks
;
6240 used_disks
= imsm_num_data_members(dev
, 0);
6241 if (used_disks
> 0) {
6243 map
->blocks_per_member
*
6245 /* round array size down to closest MB
6247 array_blocks
= (array_blocks
6248 >> SECT_PER_MB_SHIFT
)
6249 << SECT_PER_MB_SHIFT
;
6250 a
->info
.custom_array_size
= array_blocks
;
6251 /* encourage manager to update array
6255 a
->check_reshape
= 1;
6257 /* finalize online capacity expansion/reshape */
6258 for (mdi
= a
->info
.devs
; mdi
; mdi
= mdi
->next
)
6260 mdi
->disk
.raid_disk
,
6263 imsm_progress_container_reshape(super
);
6268 /* before we activate this array handle any missing disks */
6269 if (consistent
== 2)
6270 handle_missing(super
, dev
);
6272 if (consistent
== 2 &&
6273 (!is_resync_complete(&a
->info
) ||
6274 map_state
!= IMSM_T_STATE_NORMAL
||
6275 dev
->vol
.migr_state
))
6278 if (is_resync_complete(&a
->info
)) {
6279 /* complete intialization / resync,
6280 * recovery and interrupted recovery is completed in
6283 if (is_resyncing(dev
)) {
6284 dprintf("imsm: mark resync done\n");
6285 end_migration(dev
, map_state
);
6286 super
->updates_pending
++;
6287 a
->last_checkpoint
= 0;
6289 } else if ((!is_resyncing(dev
) && !failed
) &&
6290 (imsm_reshape_blocks_arrays_changes(super
) == 0)) {
6291 /* mark the start of the init process if nothing is failed */
6292 dprintf("imsm: mark resync start\n");
6293 if (map
->map_state
== IMSM_T_STATE_UNINITIALIZED
)
6294 migrate(dev
, super
, IMSM_T_STATE_NORMAL
, MIGR_INIT
);
6296 migrate(dev
, super
, IMSM_T_STATE_NORMAL
, MIGR_REPAIR
);
6297 super
->updates_pending
++;
6301 /* skip checkpointing for general migration,
6302 * it is controlled in mdadm
6304 if (is_gen_migration(dev
))
6305 goto skip_mark_checkpoint
;
6307 /* check if we can update curr_migr_unit from resync_start, recovery_start */
6308 blocks_per_unit
= blocks_per_migr_unit(super
, dev
);
6309 if (blocks_per_unit
) {
6313 units
= a
->last_checkpoint
/ blocks_per_unit
;
6316 /* check that we did not overflow 32-bits, and that
6317 * curr_migr_unit needs updating
6319 if (units32
== units
&&
6321 __le32_to_cpu(dev
->vol
.curr_migr_unit
) != units32
) {
6322 dprintf("imsm: mark checkpoint (%u)\n", units32
);
6323 dev
->vol
.curr_migr_unit
= __cpu_to_le32(units32
);
6324 super
->updates_pending
++;
6328 skip_mark_checkpoint
:
6329 /* mark dirty / clean */
6330 if (dev
->vol
.dirty
!= !consistent
) {
6331 dprintf("imsm: mark '%s'\n", consistent
? "clean" : "dirty");
6336 super
->updates_pending
++;
6342 static void imsm_set_disk(struct active_array
*a
, int n
, int state
)
6344 int inst
= a
->info
.container_member
;
6345 struct intel_super
*super
= a
->container
->sb
;
6346 struct imsm_dev
*dev
= get_imsm_dev(super
, inst
);
6347 struct imsm_map
*map
= get_imsm_map(dev
, 0);
6348 struct imsm_disk
*disk
;
6353 if (n
> map
->num_members
)
6354 fprintf(stderr
, "imsm: set_disk %d out of range 0..%d\n",
6355 n
, map
->num_members
- 1);
6360 dprintf("imsm: set_disk %d:%x\n", n
, state
);
6362 ord
= get_imsm_ord_tbl_ent(dev
, n
, -2);
6363 disk
= get_imsm_disk(super
, ord_to_idx(ord
));
6365 /* check for new failures */
6366 if (state
& DS_FAULTY
) {
6367 if (mark_failure(dev
, disk
, ord_to_idx(ord
)))
6368 super
->updates_pending
++;
6371 /* check if in_sync */
6372 if (state
& DS_INSYNC
&& ord
& IMSM_ORD_REBUILD
&& is_rebuilding(dev
)) {
6373 struct imsm_map
*migr_map
= get_imsm_map(dev
, 1);
6375 set_imsm_ord_tbl_ent(migr_map
, n
, ord_to_idx(ord
));
6376 super
->updates_pending
++;
6379 failed
= imsm_count_failed(super
, dev
);
6380 map_state
= imsm_check_degraded(super
, dev
, failed
);
6382 /* check if recovery complete, newly degraded, or failed */
6383 if (map_state
== IMSM_T_STATE_NORMAL
&& is_rebuilding(dev
)) {
6384 end_migration(dev
, map_state
);
6385 map
= get_imsm_map(dev
, 0);
6386 map
->failed_disk_num
= ~0;
6387 super
->updates_pending
++;
6388 a
->last_checkpoint
= 0;
6389 } else if (map_state
== IMSM_T_STATE_DEGRADED
&&
6390 map
->map_state
!= map_state
&&
6391 !dev
->vol
.migr_state
) {
6392 dprintf("imsm: mark degraded\n");
6393 map
->map_state
= map_state
;
6394 super
->updates_pending
++;
6395 a
->last_checkpoint
= 0;
6396 } else if (map_state
== IMSM_T_STATE_FAILED
&&
6397 map
->map_state
!= map_state
) {
6398 dprintf("imsm: mark failed\n");
6399 end_migration(dev
, map_state
);
6400 super
->updates_pending
++;
6401 a
->last_checkpoint
= 0;
6402 } else if (is_gen_migration(dev
)) {
6403 dprintf("imsm: Detected General Migration in state: ");
6404 if (map_state
== IMSM_T_STATE_NORMAL
) {
6405 end_migration(dev
, map_state
);
6406 map
= get_imsm_map(dev
, 0);
6407 map
->failed_disk_num
= ~0;
6408 dprintf("normal\n");
6410 if (map_state
== IMSM_T_STATE_DEGRADED
) {
6411 printf("degraded\n");
6412 end_migration(dev
, map_state
);
6414 dprintf("failed\n");
6416 map
->map_state
= map_state
;
6418 super
->updates_pending
++;
6422 static int store_imsm_mpb(int fd
, struct imsm_super
*mpb
)
6425 __u32 mpb_size
= __le32_to_cpu(mpb
->mpb_size
);
6426 unsigned long long dsize
;
6427 unsigned long long sectors
;
6429 get_dev_size(fd
, NULL
, &dsize
);
6431 if (mpb_size
> 512) {
6432 /* -1 to account for anchor */
6433 sectors
= mpb_sectors(mpb
) - 1;
6435 /* write the extended mpb to the sectors preceeding the anchor */
6436 if (lseek64(fd
, dsize
- (512 * (2 + sectors
)), SEEK_SET
) < 0)
6439 if ((unsigned long long)write(fd
, buf
+ 512, 512 * sectors
)
6444 /* first block is stored on second to last sector of the disk */
6445 if (lseek64(fd
, dsize
- (512 * 2), SEEK_SET
) < 0)
6448 if (write(fd
, buf
, 512) != 512)
6454 static void imsm_sync_metadata(struct supertype
*container
)
6456 struct intel_super
*super
= container
->sb
;
6458 dprintf("sync metadata: %d\n", super
->updates_pending
);
6459 if (!super
->updates_pending
)
6462 write_super_imsm(container
, 0);
6464 super
->updates_pending
= 0;
6467 static struct dl
*imsm_readd(struct intel_super
*super
, int idx
, struct active_array
*a
)
6469 struct imsm_dev
*dev
= get_imsm_dev(super
, a
->info
.container_member
);
6470 int i
= get_imsm_disk_idx(dev
, idx
, -1);
6473 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
6477 if (dl
&& is_failed(&dl
->disk
))
6481 dprintf("%s: found %x:%x\n", __func__
, dl
->major
, dl
->minor
);
6486 static struct dl
*imsm_add_spare(struct intel_super
*super
, int slot
,
6487 struct active_array
*a
, int activate_new
,
6488 struct mdinfo
*additional_test_list
)
6490 struct imsm_dev
*dev
= get_imsm_dev(super
, a
->info
.container_member
);
6491 int idx
= get_imsm_disk_idx(dev
, slot
, -1);
6492 struct imsm_super
*mpb
= super
->anchor
;
6493 struct imsm_map
*map
;
6494 unsigned long long pos
;
6499 __u32 array_start
= 0;
6500 __u32 array_end
= 0;
6502 struct mdinfo
*test_list
;
6504 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
6505 /* If in this array, skip */
6506 for (d
= a
->info
.devs
; d
; d
= d
->next
)
6507 if (d
->state_fd
>= 0 &&
6508 d
->disk
.major
== dl
->major
&&
6509 d
->disk
.minor
== dl
->minor
) {
6510 dprintf("%x:%x already in array\n",
6511 dl
->major
, dl
->minor
);
6516 test_list
= additional_test_list
;
6518 if (test_list
->disk
.major
== dl
->major
&&
6519 test_list
->disk
.minor
== dl
->minor
) {
6520 dprintf("%x:%x already in additional test list\n",
6521 dl
->major
, dl
->minor
);
6524 test_list
= test_list
->next
;
6529 /* skip in use or failed drives */
6530 if (is_failed(&dl
->disk
) || idx
== dl
->index
||
6532 dprintf("%x:%x status (failed: %d index: %d)\n",
6533 dl
->major
, dl
->minor
, is_failed(&dl
->disk
), idx
);
6537 /* skip pure spares when we are looking for partially
6538 * assimilated drives
6540 if (dl
->index
== -1 && !activate_new
)
6543 /* Does this unused device have the requisite free space?
6544 * It needs to be able to cover all member volumes
6546 ex
= get_extents(super
, dl
);
6548 dprintf("cannot get extents\n");
6551 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
6552 dev
= get_imsm_dev(super
, i
);
6553 map
= get_imsm_map(dev
, 0);
6555 /* check if this disk is already a member of
6558 if (get_imsm_disk_slot(map
, dl
->index
) >= 0)
6564 array_start
= __le32_to_cpu(map
->pba_of_lba0
);
6565 array_end
= array_start
+
6566 __le32_to_cpu(map
->blocks_per_member
) - 1;
6569 /* check that we can start at pba_of_lba0 with
6570 * blocks_per_member of space
6572 if (array_start
>= pos
&& array_end
< ex
[j
].start
) {
6576 pos
= ex
[j
].start
+ ex
[j
].size
;
6578 } while (ex
[j
-1].size
);
6585 if (i
< mpb
->num_raid_devs
) {
6586 dprintf("%x:%x does not have %u to %u available\n",
6587 dl
->major
, dl
->minor
, array_start
, array_end
);
6598 static int imsm_rebuild_allowed(struct supertype
*cont
, int dev_idx
, int failed
)
6600 struct imsm_dev
*dev2
;
6601 struct imsm_map
*map
;
6607 dev2
= get_imsm_dev(cont
->sb
, dev_idx
);
6609 state
= imsm_check_degraded(cont
->sb
, dev2
, failed
);
6610 if (state
== IMSM_T_STATE_FAILED
) {
6611 map
= get_imsm_map(dev2
, 0);
6614 for (slot
= 0; slot
< map
->num_members
; slot
++) {
6616 * Check if failed disks are deleted from intel
6617 * disk list or are marked to be deleted
6619 idx
= get_imsm_disk_idx(dev2
, slot
, -1);
6620 idisk
= get_imsm_dl_disk(cont
->sb
, idx
);
6622 * Do not rebuild the array if failed disks
6623 * from failed sub-array are not removed from
6627 is_failed(&idisk
->disk
) &&
6628 (idisk
->action
!= DISK_REMOVE
))
6636 static struct mdinfo
*imsm_activate_spare(struct active_array
*a
,
6637 struct metadata_update
**updates
)
6640 * Find a device with unused free space and use it to replace a
6641 * failed/vacant region in an array. We replace failed regions one a
6642 * array at a time. The result is that a new spare disk will be added
6643 * to the first failed array and after the monitor has finished
6644 * propagating failures the remainder will be consumed.
6646 * FIXME add a capability for mdmon to request spares from another
6650 struct intel_super
*super
= a
->container
->sb
;
6651 int inst
= a
->info
.container_member
;
6652 struct imsm_dev
*dev
= get_imsm_dev(super
, inst
);
6653 struct imsm_map
*map
= get_imsm_map(dev
, 0);
6654 int failed
= a
->info
.array
.raid_disks
;
6655 struct mdinfo
*rv
= NULL
;
6658 struct metadata_update
*mu
;
6660 struct imsm_update_activate_spare
*u
;
6665 for (d
= a
->info
.devs
; d
; d
= d
->next
) {
6666 if ((d
->curr_state
& DS_FAULTY
) &&
6668 /* wait for Removal to happen */
6670 if (d
->state_fd
>= 0)
6674 dprintf("imsm: activate spare: inst=%d failed=%d (%d) level=%d\n",
6675 inst
, failed
, a
->info
.array
.raid_disks
, a
->info
.array
.level
);
6677 if (imsm_reshape_blocks_arrays_changes(super
))
6680 if (a
->info
.array
.level
== 4)
6681 /* No repair for takeovered array
6682 * imsm doesn't support raid4
6686 if (imsm_check_degraded(super
, dev
, failed
) != IMSM_T_STATE_DEGRADED
)
6690 * If there are any failed disks check state of the other volume.
6691 * Block rebuild if the another one is failed until failed disks
6692 * are removed from container.
6695 dprintf("found failed disks in %.*s, check if there another"
6696 "failed sub-array.\n",
6697 MAX_RAID_SERIAL_LEN
, dev
->volume
);
6698 /* check if states of the other volumes allow for rebuild */
6699 for (i
= 0; i
< super
->anchor
->num_raid_devs
; i
++) {
6701 allowed
= imsm_rebuild_allowed(a
->container
,
6709 /* For each slot, if it is not working, find a spare */
6710 for (i
= 0; i
< a
->info
.array
.raid_disks
; i
++) {
6711 for (d
= a
->info
.devs
; d
; d
= d
->next
)
6712 if (d
->disk
.raid_disk
== i
)
6714 dprintf("found %d: %p %x\n", i
, d
, d
?d
->curr_state
:0);
6715 if (d
&& (d
->state_fd
>= 0))
6719 * OK, this device needs recovery. Try to re-add the
6720 * previous occupant of this slot, if this fails see if
6721 * we can continue the assimilation of a spare that was
6722 * partially assimilated, finally try to activate a new
6725 dl
= imsm_readd(super
, i
, a
);
6727 dl
= imsm_add_spare(super
, i
, a
, 0, rv
);
6729 dl
= imsm_add_spare(super
, i
, a
, 1, rv
);
6733 /* found a usable disk with enough space */
6734 di
= malloc(sizeof(*di
));
6737 memset(di
, 0, sizeof(*di
));
6739 /* dl->index will be -1 in the case we are activating a
6740 * pristine spare. imsm_process_update() will create a
6741 * new index in this case. Once a disk is found to be
6742 * failed in all member arrays it is kicked from the
6745 di
->disk
.number
= dl
->index
;
6747 /* (ab)use di->devs to store a pointer to the device
6750 di
->devs
= (struct mdinfo
*) dl
;
6752 di
->disk
.raid_disk
= i
;
6753 di
->disk
.major
= dl
->major
;
6754 di
->disk
.minor
= dl
->minor
;
6756 di
->recovery_start
= 0;
6757 di
->data_offset
= __le32_to_cpu(map
->pba_of_lba0
);
6758 di
->component_size
= a
->info
.component_size
;
6759 di
->container_member
= inst
;
6760 super
->random
= random32();
6764 dprintf("%x:%x to be %d at %llu\n", dl
->major
, dl
->minor
,
6765 i
, di
->data_offset
);
6769 /* No spares found */
6771 /* Now 'rv' has a list of devices to return.
6772 * Create a metadata_update record to update the
6773 * disk_ord_tbl for the array
6775 mu
= malloc(sizeof(*mu
));
6777 mu
->buf
= malloc(sizeof(struct imsm_update_activate_spare
) * num_spares
);
6778 if (mu
->buf
== NULL
) {
6785 struct mdinfo
*n
= rv
->next
;
6794 mu
->space_list
= NULL
;
6795 mu
->len
= sizeof(struct imsm_update_activate_spare
) * num_spares
;
6796 mu
->next
= *updates
;
6797 u
= (struct imsm_update_activate_spare
*) mu
->buf
;
6799 for (di
= rv
; di
; di
= di
->next
) {
6800 u
->type
= update_activate_spare
;
6801 u
->dl
= (struct dl
*) di
->devs
;
6803 u
->slot
= di
->disk
.raid_disk
;
6814 static int disks_overlap(struct intel_super
*super
, int idx
, struct imsm_update_create_array
*u
)
6816 struct imsm_dev
*dev
= get_imsm_dev(super
, idx
);
6817 struct imsm_map
*map
= get_imsm_map(dev
, 0);
6818 struct imsm_map
*new_map
= get_imsm_map(&u
->dev
, 0);
6819 struct disk_info
*inf
= get_disk_info(u
);
6820 struct imsm_disk
*disk
;
6824 for (i
= 0; i
< map
->num_members
; i
++) {
6825 disk
= get_imsm_disk(super
, get_imsm_disk_idx(dev
, i
, -1));
6826 for (j
= 0; j
< new_map
->num_members
; j
++)
6827 if (serialcmp(disk
->serial
, inf
[j
].serial
) == 0)
6835 static struct dl
*get_disk_super(struct intel_super
*super
, int major
, int minor
)
6837 struct dl
*dl
= NULL
;
6838 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
6839 if ((dl
->major
== major
) && (dl
->minor
== minor
))
6844 static int remove_disk_super(struct intel_super
*super
, int major
, int minor
)
6846 struct dl
*prev
= NULL
;
6850 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
6851 if ((dl
->major
== major
) && (dl
->minor
== minor
)) {
6854 prev
->next
= dl
->next
;
6856 super
->disks
= dl
->next
;
6858 __free_imsm_disk(dl
);
6859 dprintf("%s: removed %x:%x\n",
6860 __func__
, major
, minor
);
6868 static void imsm_delete(struct intel_super
*super
, struct dl
**dlp
, unsigned index
);
6870 static int add_remove_disk_update(struct intel_super
*super
)
6872 int check_degraded
= 0;
6873 struct dl
*disk
= NULL
;
6874 /* add/remove some spares to/from the metadata/contrainer */
6875 while (super
->disk_mgmt_list
) {
6876 struct dl
*disk_cfg
;
6878 disk_cfg
= super
->disk_mgmt_list
;
6879 super
->disk_mgmt_list
= disk_cfg
->next
;
6880 disk_cfg
->next
= NULL
;
6882 if (disk_cfg
->action
== DISK_ADD
) {
6883 disk_cfg
->next
= super
->disks
;
6884 super
->disks
= disk_cfg
;
6886 dprintf("%s: added %x:%x\n",
6887 __func__
, disk_cfg
->major
,
6889 } else if (disk_cfg
->action
== DISK_REMOVE
) {
6890 dprintf("Disk remove action processed: %x.%x\n",
6891 disk_cfg
->major
, disk_cfg
->minor
);
6892 disk
= get_disk_super(super
,
6896 /* store action status */
6897 disk
->action
= DISK_REMOVE
;
6898 /* remove spare disks only */
6899 if (disk
->index
== -1) {
6900 remove_disk_super(super
,
6905 /* release allocate disk structure */
6906 __free_imsm_disk(disk_cfg
);
6909 return check_degraded
;
6913 static int apply_reshape_migration_update(struct imsm_update_reshape_migration
*u
,
6914 struct intel_super
*super
,
6917 struct intel_dev
*id
;
6918 void **tofree
= NULL
;
6921 dprintf("apply_reshape_migration_update()\n");
6922 if ((u
->subdev
< 0) ||
6924 dprintf("imsm: Error: Wrong subdev: %i\n", u
->subdev
);
6927 if ((space_list
== NULL
) || (*space_list
== NULL
)) {
6928 dprintf("imsm: Error: Memory is not allocated\n");
6932 for (id
= super
->devlist
; id
; id
= id
->next
) {
6933 if (id
->index
== (unsigned)u
->subdev
) {
6934 struct imsm_dev
*dev
= get_imsm_dev(super
, u
->subdev
);
6935 struct imsm_map
*map
;
6936 struct imsm_dev
*new_dev
=
6937 (struct imsm_dev
*)*space_list
;
6938 struct imsm_map
*migr_map
= get_imsm_map(dev
, 1);
6940 struct dl
*new_disk
;
6942 if (new_dev
== NULL
)
6944 *space_list
= **space_list
;
6945 memcpy(new_dev
, dev
, sizeof_imsm_dev(dev
, 0));
6946 map
= get_imsm_map(new_dev
, 0);
6948 dprintf("imsm: Error: migration in progress");
6952 to_state
= map
->map_state
;
6953 if ((u
->new_level
== 5) && (map
->raid_level
== 0)) {
6955 /* this should not happen */
6956 if (u
->new_disks
[0] < 0) {
6957 map
->failed_disk_num
=
6958 map
->num_members
- 1;
6959 to_state
= IMSM_T_STATE_DEGRADED
;
6961 to_state
= IMSM_T_STATE_NORMAL
;
6963 migrate(new_dev
, super
, to_state
, MIGR_GEN_MIGR
);
6964 if (u
->new_level
> -1)
6965 map
->raid_level
= u
->new_level
;
6966 migr_map
= get_imsm_map(new_dev
, 1);
6967 if ((u
->new_level
== 5) &&
6968 (migr_map
->raid_level
== 0)) {
6969 int ord
= map
->num_members
- 1;
6970 migr_map
->num_members
--;
6971 if (u
->new_disks
[0] < 0)
6972 ord
|= IMSM_ORD_REBUILD
;
6973 set_imsm_ord_tbl_ent(map
,
6974 map
->num_members
- 1,
6978 tofree
= (void **)dev
;
6980 /* update chunk size
6982 if (u
->new_chunksize
> 0)
6983 map
->blocks_per_strip
=
6984 __cpu_to_le16(u
->new_chunksize
* 2);
6988 if ((u
->new_level
!= 5) ||
6989 (migr_map
->raid_level
!= 0) ||
6990 (migr_map
->raid_level
== map
->raid_level
))
6993 if (u
->new_disks
[0] >= 0) {
6996 new_disk
= get_disk_super(super
,
6997 major(u
->new_disks
[0]),
6998 minor(u
->new_disks
[0]));
6999 dprintf("imsm: new disk for reshape is: %i:%i "
7000 "(%p, index = %i)\n",
7001 major(u
->new_disks
[0]),
7002 minor(u
->new_disks
[0]),
7003 new_disk
, new_disk
->index
);
7004 if (new_disk
== NULL
)
7005 goto error_disk_add
;
7007 new_disk
->index
= map
->num_members
- 1;
7008 /* slot to fill in autolayout
7010 new_disk
->raiddisk
= new_disk
->index
;
7011 new_disk
->disk
.status
|= CONFIGURED_DISK
;
7012 new_disk
->disk
.status
&= ~SPARE_DISK
;
7014 goto error_disk_add
;
7017 *tofree
= *space_list
;
7018 /* calculate new size
7020 imsm_set_array_size(new_dev
);
7027 *space_list
= tofree
;
7031 dprintf("Error: imsm: Cannot find disk.\n");
7035 static int apply_update_activate_spare(struct imsm_update_activate_spare
*u
,
7036 struct intel_super
*super
,
7037 struct active_array
*active_array
)
7039 struct imsm_super
*mpb
= super
->anchor
;
7040 struct imsm_dev
*dev
= get_imsm_dev(super
, u
->array
);
7041 struct imsm_map
*map
= get_imsm_map(dev
, 0);
7042 struct imsm_map
*migr_map
;
7043 struct active_array
*a
;
7044 struct imsm_disk
*disk
;
7051 int second_map_created
= 0;
7053 for (; u
; u
= u
->next
) {
7054 victim
= get_imsm_disk_idx(dev
, u
->slot
, -1);
7059 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
7064 fprintf(stderr
, "error: imsm_activate_spare passed "
7065 "an unknown disk (index: %d)\n",
7070 /* count failures (excluding rebuilds and the victim)
7071 * to determine map[0] state
7074 for (i
= 0; i
< map
->num_members
; i
++) {
7077 disk
= get_imsm_disk(super
,
7078 get_imsm_disk_idx(dev
, i
, -1));
7079 if (!disk
|| is_failed(disk
))
7083 /* adding a pristine spare, assign a new index */
7084 if (dl
->index
< 0) {
7085 dl
->index
= super
->anchor
->num_disks
;
7086 super
->anchor
->num_disks
++;
7089 disk
->status
|= CONFIGURED_DISK
;
7090 disk
->status
&= ~SPARE_DISK
;
7093 to_state
= imsm_check_degraded(super
, dev
, failed
);
7094 if (!second_map_created
) {
7095 second_map_created
= 1;
7096 map
->map_state
= IMSM_T_STATE_DEGRADED
;
7097 migrate(dev
, super
, to_state
, MIGR_REBUILD
);
7099 map
->map_state
= to_state
;
7100 migr_map
= get_imsm_map(dev
, 1);
7101 set_imsm_ord_tbl_ent(map
, u
->slot
, dl
->index
);
7102 set_imsm_ord_tbl_ent(migr_map
, u
->slot
,
7103 dl
->index
| IMSM_ORD_REBUILD
);
7105 /* update the family_num to mark a new container
7106 * generation, being careful to record the existing
7107 * family_num in orig_family_num to clean up after
7108 * earlier mdadm versions that neglected to set it.
7110 if (mpb
->orig_family_num
== 0)
7111 mpb
->orig_family_num
= mpb
->family_num
;
7112 mpb
->family_num
+= super
->random
;
7114 /* count arrays using the victim in the metadata */
7116 for (a
= active_array
; a
; a
= a
->next
) {
7117 dev
= get_imsm_dev(super
, a
->info
.container_member
);
7118 map
= get_imsm_map(dev
, 0);
7120 if (get_imsm_disk_slot(map
, victim
) >= 0)
7124 /* delete the victim if it is no longer being
7130 /* We know that 'manager' isn't touching anything,
7131 * so it is safe to delete
7133 for (dlp
= &super
->disks
; *dlp
; dlp
= &(*dlp
)->next
)
7134 if ((*dlp
)->index
== victim
)
7137 /* victim may be on the missing list */
7139 for (dlp
= &super
->missing
; *dlp
;
7140 dlp
= &(*dlp
)->next
)
7141 if ((*dlp
)->index
== victim
)
7143 imsm_delete(super
, dlp
, victim
);
7150 static int apply_reshape_container_disks_update(struct imsm_update_reshape
*u
,
7151 struct intel_super
*super
,
7154 struct dl
*new_disk
;
7155 struct intel_dev
*id
;
7157 int delta_disks
= u
->new_raid_disks
- u
->old_raid_disks
;
7158 int disk_count
= u
->old_raid_disks
;
7159 void **tofree
= NULL
;
7160 int devices_to_reshape
= 1;
7161 struct imsm_super
*mpb
= super
->anchor
;
7163 unsigned int dev_id
;
7165 dprintf("imsm: apply_reshape_container_disks_update()\n");
7167 /* enable spares to use in array */
7168 for (i
= 0; i
< delta_disks
; i
++) {
7169 new_disk
= get_disk_super(super
,
7170 major(u
->new_disks
[i
]),
7171 minor(u
->new_disks
[i
]));
7172 dprintf("imsm: new disk for reshape is: %i:%i "
7173 "(%p, index = %i)\n",
7174 major(u
->new_disks
[i
]), minor(u
->new_disks
[i
]),
7175 new_disk
, new_disk
->index
);
7176 if ((new_disk
== NULL
) ||
7177 ((new_disk
->index
>= 0) &&
7178 (new_disk
->index
< u
->old_raid_disks
)))
7179 goto update_reshape_exit
;
7180 new_disk
->index
= disk_count
++;
7181 /* slot to fill in autolayout
7183 new_disk
->raiddisk
= new_disk
->index
;
7184 new_disk
->disk
.status
|=
7186 new_disk
->disk
.status
&= ~SPARE_DISK
;
7189 dprintf("imsm: volume set mpb->num_raid_devs = %i\n",
7190 mpb
->num_raid_devs
);
7191 /* manage changes in volume
7193 for (dev_id
= 0; dev_id
< mpb
->num_raid_devs
; dev_id
++) {
7194 void **sp
= *space_list
;
7195 struct imsm_dev
*newdev
;
7196 struct imsm_map
*newmap
, *oldmap
;
7198 for (id
= super
->devlist
; id
; id
= id
->next
) {
7199 if (id
->index
== dev_id
)
7208 /* Copy the dev, but not (all of) the map */
7209 memcpy(newdev
, id
->dev
, sizeof(*newdev
));
7210 oldmap
= get_imsm_map(id
->dev
, 0);
7211 newmap
= get_imsm_map(newdev
, 0);
7212 /* Copy the current map */
7213 memcpy(newmap
, oldmap
, sizeof_imsm_map(oldmap
));
7214 /* update one device only
7216 if (devices_to_reshape
) {
7217 dprintf("imsm: modifying subdev: %i\n",
7219 devices_to_reshape
--;
7220 newdev
->vol
.migr_state
= 1;
7221 newdev
->vol
.curr_migr_unit
= 0;
7222 set_migr_type(newdev
, MIGR_GEN_MIGR
);
7223 newmap
->num_members
= u
->new_raid_disks
;
7224 for (i
= 0; i
< delta_disks
; i
++) {
7225 set_imsm_ord_tbl_ent(newmap
,
7226 u
->old_raid_disks
+ i
,
7227 u
->old_raid_disks
+ i
);
7229 /* New map is correct, now need to save old map
7231 newmap
= get_imsm_map(newdev
, 1);
7232 memcpy(newmap
, oldmap
, sizeof_imsm_map(oldmap
));
7234 imsm_set_array_size(newdev
);
7237 sp
= (void **)id
->dev
;
7242 /* Clear migration record */
7243 memset(super
->migr_rec
, 0, sizeof(struct migr_record
));
7246 *space_list
= tofree
;
7249 update_reshape_exit
:
7254 static int apply_takeover_update(struct imsm_update_takeover
*u
,
7255 struct intel_super
*super
,
7258 struct imsm_dev
*dev
= NULL
;
7259 struct intel_dev
*dv
;
7260 struct imsm_dev
*dev_new
;
7261 struct imsm_map
*map
;
7265 for (dv
= super
->devlist
; dv
; dv
= dv
->next
)
7266 if (dv
->index
== (unsigned int)u
->subarray
) {
7274 map
= get_imsm_map(dev
, 0);
7276 if (u
->direction
== R10_TO_R0
) {
7277 /* Number of failed disks must be half of initial disk number */
7278 if (imsm_count_failed(super
, dev
) != (map
->num_members
/ 2))
7281 /* iterate through devices to mark removed disks as spare */
7282 for (dm
= super
->disks
; dm
; dm
= dm
->next
) {
7283 if (dm
->disk
.status
& FAILED_DISK
) {
7284 int idx
= dm
->index
;
7285 /* update indexes on the disk list */
7286 /* FIXME this loop-with-the-loop looks wrong, I'm not convinced
7287 the index values will end up being correct.... NB */
7288 for (du
= super
->disks
; du
; du
= du
->next
)
7289 if (du
->index
> idx
)
7291 /* mark as spare disk */
7296 map
->num_members
= map
->num_members
/ 2;
7297 map
->map_state
= IMSM_T_STATE_NORMAL
;
7298 map
->num_domains
= 1;
7299 map
->raid_level
= 0;
7300 map
->failed_disk_num
= -1;
7303 if (u
->direction
== R0_TO_R10
) {
7305 /* update slots in current disk list */
7306 for (dm
= super
->disks
; dm
; dm
= dm
->next
) {
7310 /* create new *missing* disks */
7311 for (i
= 0; i
< map
->num_members
; i
++) {
7312 space
= *space_list
;
7315 *space_list
= *space
;
7317 memcpy(du
, super
->disks
, sizeof(*du
));
7321 du
->index
= (i
* 2) + 1;
7322 sprintf((char *)du
->disk
.serial
,
7323 " MISSING_%d", du
->index
);
7324 sprintf((char *)du
->serial
,
7325 "MISSING_%d", du
->index
);
7326 du
->next
= super
->missing
;
7327 super
->missing
= du
;
7329 /* create new dev and map */
7330 space
= *space_list
;
7333 *space_list
= *space
;
7334 dev_new
= (void *)space
;
7335 memcpy(dev_new
, dev
, sizeof(*dev
));
7336 /* update new map */
7337 map
= get_imsm_map(dev_new
, 0);
7338 map
->num_members
= map
->num_members
* 2;
7339 map
->map_state
= IMSM_T_STATE_DEGRADED
;
7340 map
->num_domains
= 2;
7341 map
->raid_level
= 1;
7342 /* replace dev<->dev_new */
7345 /* update disk order table */
7346 for (du
= super
->disks
; du
; du
= du
->next
)
7348 set_imsm_ord_tbl_ent(map
, du
->index
, du
->index
);
7349 for (du
= super
->missing
; du
; du
= du
->next
)
7350 if (du
->index
>= 0) {
7351 set_imsm_ord_tbl_ent(map
, du
->index
, du
->index
);
7352 mark_missing(dv
->dev
, &du
->disk
, du
->index
);
7358 static void imsm_process_update(struct supertype
*st
,
7359 struct metadata_update
*update
)
7362 * crack open the metadata_update envelope to find the update record
7363 * update can be one of:
7364 * update_reshape_container_disks - all the arrays in the container
7365 * are being reshaped to have more devices. We need to mark
7366 * the arrays for general migration and convert selected spares
7367 * into active devices.
7368 * update_activate_spare - a spare device has replaced a failed
7369 * device in an array, update the disk_ord_tbl. If this disk is
7370 * present in all member arrays then also clear the SPARE_DISK
7372 * update_create_array
7374 * update_rename_array
7375 * update_add_remove_disk
7377 struct intel_super
*super
= st
->sb
;
7378 struct imsm_super
*mpb
;
7379 enum imsm_update_type type
= *(enum imsm_update_type
*) update
->buf
;
7381 /* update requires a larger buf but the allocation failed */
7382 if (super
->next_len
&& !super
->next_buf
) {
7383 super
->next_len
= 0;
7387 if (super
->next_buf
) {
7388 memcpy(super
->next_buf
, super
->buf
, super
->len
);
7390 super
->len
= super
->next_len
;
7391 super
->buf
= super
->next_buf
;
7393 super
->next_len
= 0;
7394 super
->next_buf
= NULL
;
7397 mpb
= super
->anchor
;
7400 case update_general_migration_checkpoint
: {
7401 struct intel_dev
*id
;
7402 struct imsm_update_general_migration_checkpoint
*u
=
7403 (void *)update
->buf
;
7405 dprintf("imsm: process_update() "
7406 "for update_general_migration_checkpoint called\n");
7408 /* find device under general migration */
7409 for (id
= super
->devlist
; id
; id
= id
->next
) {
7410 if (is_gen_migration(id
->dev
)) {
7411 id
->dev
->vol
.curr_migr_unit
=
7412 __cpu_to_le32(u
->curr_migr_unit
);
7413 super
->updates_pending
++;
7418 case update_takeover
: {
7419 struct imsm_update_takeover
*u
= (void *)update
->buf
;
7420 if (apply_takeover_update(u
, super
, &update
->space_list
)) {
7421 imsm_update_version_info(super
);
7422 super
->updates_pending
++;
7427 case update_reshape_container_disks
: {
7428 struct imsm_update_reshape
*u
= (void *)update
->buf
;
7429 if (apply_reshape_container_disks_update(
7430 u
, super
, &update
->space_list
))
7431 super
->updates_pending
++;
7434 case update_reshape_migration
: {
7435 struct imsm_update_reshape_migration
*u
= (void *)update
->buf
;
7436 if (apply_reshape_migration_update(
7437 u
, super
, &update
->space_list
))
7438 super
->updates_pending
++;
7441 case update_activate_spare
: {
7442 struct imsm_update_activate_spare
*u
= (void *) update
->buf
;
7443 if (apply_update_activate_spare(u
, super
, st
->arrays
))
7444 super
->updates_pending
++;
7447 case update_create_array
: {
7448 /* someone wants to create a new array, we need to be aware of
7449 * a few races/collisions:
7450 * 1/ 'Create' called by two separate instances of mdadm
7451 * 2/ 'Create' versus 'activate_spare': mdadm has chosen
7452 * devices that have since been assimilated via
7454 * In the event this update can not be carried out mdadm will
7455 * (FIX ME) notice that its update did not take hold.
7457 struct imsm_update_create_array
*u
= (void *) update
->buf
;
7458 struct intel_dev
*dv
;
7459 struct imsm_dev
*dev
;
7460 struct imsm_map
*map
, *new_map
;
7461 unsigned long long start
, end
;
7462 unsigned long long new_start
, new_end
;
7464 struct disk_info
*inf
;
7467 /* handle racing creates: first come first serve */
7468 if (u
->dev_idx
< mpb
->num_raid_devs
) {
7469 dprintf("%s: subarray %d already defined\n",
7470 __func__
, u
->dev_idx
);
7474 /* check update is next in sequence */
7475 if (u
->dev_idx
!= mpb
->num_raid_devs
) {
7476 dprintf("%s: can not create array %d expected index %d\n",
7477 __func__
, u
->dev_idx
, mpb
->num_raid_devs
);
7481 new_map
= get_imsm_map(&u
->dev
, 0);
7482 new_start
= __le32_to_cpu(new_map
->pba_of_lba0
);
7483 new_end
= new_start
+ __le32_to_cpu(new_map
->blocks_per_member
);
7484 inf
= get_disk_info(u
);
7486 /* handle activate_spare versus create race:
7487 * check to make sure that overlapping arrays do not include
7490 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
7491 dev
= get_imsm_dev(super
, i
);
7492 map
= get_imsm_map(dev
, 0);
7493 start
= __le32_to_cpu(map
->pba_of_lba0
);
7494 end
= start
+ __le32_to_cpu(map
->blocks_per_member
);
7495 if ((new_start
>= start
&& new_start
<= end
) ||
7496 (start
>= new_start
&& start
<= new_end
))
7501 if (disks_overlap(super
, i
, u
)) {
7502 dprintf("%s: arrays overlap\n", __func__
);
7507 /* check that prepare update was successful */
7508 if (!update
->space
) {
7509 dprintf("%s: prepare update failed\n", __func__
);
7513 /* check that all disks are still active before committing
7514 * changes. FIXME: could we instead handle this by creating a
7515 * degraded array? That's probably not what the user expects,
7516 * so better to drop this update on the floor.
7518 for (i
= 0; i
< new_map
->num_members
; i
++) {
7519 dl
= serial_to_dl(inf
[i
].serial
, super
);
7521 dprintf("%s: disk disappeared\n", __func__
);
7526 super
->updates_pending
++;
7528 /* convert spares to members and fixup ord_tbl */
7529 for (i
= 0; i
< new_map
->num_members
; i
++) {
7530 dl
= serial_to_dl(inf
[i
].serial
, super
);
7531 if (dl
->index
== -1) {
7532 dl
->index
= mpb
->num_disks
;
7534 dl
->disk
.status
|= CONFIGURED_DISK
;
7535 dl
->disk
.status
&= ~SPARE_DISK
;
7537 set_imsm_ord_tbl_ent(new_map
, i
, dl
->index
);
7542 update
->space
= NULL
;
7543 imsm_copy_dev(dev
, &u
->dev
);
7544 dv
->index
= u
->dev_idx
;
7545 dv
->next
= super
->devlist
;
7546 super
->devlist
= dv
;
7547 mpb
->num_raid_devs
++;
7549 imsm_update_version_info(super
);
7552 /* mdmon knows how to release update->space, but not
7553 * ((struct intel_dev *) update->space)->dev
7555 if (update
->space
) {
7561 case update_kill_array
: {
7562 struct imsm_update_kill_array
*u
= (void *) update
->buf
;
7563 int victim
= u
->dev_idx
;
7564 struct active_array
*a
;
7565 struct intel_dev
**dp
;
7566 struct imsm_dev
*dev
;
7568 /* sanity check that we are not affecting the uuid of
7569 * active arrays, or deleting an active array
7571 * FIXME when immutable ids are available, but note that
7572 * we'll also need to fixup the invalidated/active
7573 * subarray indexes in mdstat
7575 for (a
= st
->arrays
; a
; a
= a
->next
)
7576 if (a
->info
.container_member
>= victim
)
7578 /* by definition if mdmon is running at least one array
7579 * is active in the container, so checking
7580 * mpb->num_raid_devs is just extra paranoia
7582 dev
= get_imsm_dev(super
, victim
);
7583 if (a
|| !dev
|| mpb
->num_raid_devs
== 1) {
7584 dprintf("failed to delete subarray-%d\n", victim
);
7588 for (dp
= &super
->devlist
; *dp
;)
7589 if ((*dp
)->index
== (unsigned)super
->current_vol
) {
7592 if ((*dp
)->index
> (unsigned)victim
)
7596 mpb
->num_raid_devs
--;
7597 super
->updates_pending
++;
7600 case update_rename_array
: {
7601 struct imsm_update_rename_array
*u
= (void *) update
->buf
;
7602 char name
[MAX_RAID_SERIAL_LEN
+1];
7603 int target
= u
->dev_idx
;
7604 struct active_array
*a
;
7605 struct imsm_dev
*dev
;
7607 /* sanity check that we are not affecting the uuid of
7610 snprintf(name
, MAX_RAID_SERIAL_LEN
, "%s", (char *) u
->name
);
7611 name
[MAX_RAID_SERIAL_LEN
] = '\0';
7612 for (a
= st
->arrays
; a
; a
= a
->next
)
7613 if (a
->info
.container_member
== target
)
7615 dev
= get_imsm_dev(super
, u
->dev_idx
);
7616 if (a
|| !dev
|| !check_name(super
, name
, 1)) {
7617 dprintf("failed to rename subarray-%d\n", target
);
7621 snprintf((char *) dev
->volume
, MAX_RAID_SERIAL_LEN
, "%s", name
);
7622 super
->updates_pending
++;
7625 case update_add_remove_disk
: {
7626 /* we may be able to repair some arrays if disks are
7627 * being added, check teh status of add_remove_disk
7628 * if discs has been added.
7630 if (add_remove_disk_update(super
)) {
7631 struct active_array
*a
;
7633 super
->updates_pending
++;
7634 for (a
= st
->arrays
; a
; a
= a
->next
)
7635 a
->check_degraded
= 1;
7640 fprintf(stderr
, "error: unsuported process update type:"
7641 "(type: %d)\n", type
);
7645 static struct mdinfo
*get_spares_for_grow(struct supertype
*st
);
7647 static void imsm_prepare_update(struct supertype
*st
,
7648 struct metadata_update
*update
)
7651 * Allocate space to hold new disk entries, raid-device entries or a new
7652 * mpb if necessary. The manager synchronously waits for updates to
7653 * complete in the monitor, so new mpb buffers allocated here can be
7654 * integrated by the monitor thread without worrying about live pointers
7655 * in the manager thread.
7657 enum imsm_update_type type
= *(enum imsm_update_type
*) update
->buf
;
7658 struct intel_super
*super
= st
->sb
;
7659 struct imsm_super
*mpb
= super
->anchor
;
7664 case update_general_migration_checkpoint
:
7665 dprintf("imsm: prepare_update() "
7666 "for update_general_migration_checkpoint called\n");
7668 case update_takeover
: {
7669 struct imsm_update_takeover
*u
= (void *)update
->buf
;
7670 if (u
->direction
== R0_TO_R10
) {
7671 void **tail
= (void **)&update
->space_list
;
7672 struct imsm_dev
*dev
= get_imsm_dev(super
, u
->subarray
);
7673 struct imsm_map
*map
= get_imsm_map(dev
, 0);
7674 int num_members
= map
->num_members
;
7678 /* allocate memory for added disks */
7679 for (i
= 0; i
< num_members
; i
++) {
7680 size
= sizeof(struct dl
);
7681 space
= malloc(size
);
7690 /* allocate memory for new device */
7691 size
= sizeof_imsm_dev(super
->devlist
->dev
, 0) +
7692 (num_members
* sizeof(__u32
));
7693 space
= malloc(size
);
7702 len
= disks_to_mpb_size(num_members
* 2);
7704 /* if allocation didn't success, free buffer */
7705 while (update
->space_list
) {
7706 void **sp
= update
->space_list
;
7707 update
->space_list
= *sp
;
7715 case update_reshape_container_disks
: {
7716 /* Every raid device in the container is about to
7717 * gain some more devices, and we will enter a
7719 * So each 'imsm_map' will be bigger, and the imsm_vol
7720 * will now hold 2 of them.
7721 * Thus we need new 'struct imsm_dev' allocations sized
7722 * as sizeof_imsm_dev but with more devices in both maps.
7724 struct imsm_update_reshape
*u
= (void *)update
->buf
;
7725 struct intel_dev
*dl
;
7726 void **space_tail
= (void**)&update
->space_list
;
7728 dprintf("imsm: imsm_prepare_update() for update_reshape\n");
7730 for (dl
= super
->devlist
; dl
; dl
= dl
->next
) {
7731 int size
= sizeof_imsm_dev(dl
->dev
, 1);
7733 if (u
->new_raid_disks
> u
->old_raid_disks
)
7734 size
+= sizeof(__u32
)*2*
7735 (u
->new_raid_disks
- u
->old_raid_disks
);
7744 len
= disks_to_mpb_size(u
->new_raid_disks
);
7745 dprintf("New anchor length is %llu\n", (unsigned long long)len
);
7748 case update_reshape_migration
: {
7749 /* for migration level 0->5 we need to add disks
7750 * so the same as for container operation we will copy
7751 * device to the bigger location.
7752 * in memory prepared device and new disk area are prepared
7753 * for usage in process update
7755 struct imsm_update_reshape_migration
*u
= (void *)update
->buf
;
7756 struct intel_dev
*id
;
7757 void **space_tail
= (void **)&update
->space_list
;
7760 int current_level
= -1;
7762 dprintf("imsm: imsm_prepare_update() for update_reshape\n");
7764 /* add space for bigger array in update
7766 for (id
= super
->devlist
; id
; id
= id
->next
) {
7767 if (id
->index
== (unsigned)u
->subdev
) {
7768 size
= sizeof_imsm_dev(id
->dev
, 1);
7769 if (u
->new_raid_disks
> u
->old_raid_disks
)
7770 size
+= sizeof(__u32
)*2*
7771 (u
->new_raid_disks
- u
->old_raid_disks
);
7781 if (update
->space_list
== NULL
)
7784 /* add space for disk in update
7786 size
= sizeof(struct dl
);
7789 free(update
->space_list
);
7790 update
->space_list
= NULL
;
7797 /* add spare device to update
7799 for (id
= super
->devlist
; id
; id
= id
->next
)
7800 if (id
->index
== (unsigned)u
->subdev
) {
7801 struct imsm_dev
*dev
;
7802 struct imsm_map
*map
;
7804 dev
= get_imsm_dev(super
, u
->subdev
);
7805 map
= get_imsm_map(dev
, 0);
7806 current_level
= map
->raid_level
;
7809 if ((u
->new_level
== 5) && (u
->new_level
!= current_level
)) {
7810 struct mdinfo
*spares
;
7812 spares
= get_spares_for_grow(st
);
7820 makedev(dev
->disk
.major
,
7822 dl
= get_disk_super(super
,
7825 dl
->index
= u
->old_raid_disks
;
7831 len
= disks_to_mpb_size(u
->new_raid_disks
);
7832 dprintf("New anchor length is %llu\n", (unsigned long long)len
);
7835 case update_create_array
: {
7836 struct imsm_update_create_array
*u
= (void *) update
->buf
;
7837 struct intel_dev
*dv
;
7838 struct imsm_dev
*dev
= &u
->dev
;
7839 struct imsm_map
*map
= get_imsm_map(dev
, 0);
7841 struct disk_info
*inf
;
7845 inf
= get_disk_info(u
);
7846 len
= sizeof_imsm_dev(dev
, 1);
7847 /* allocate a new super->devlist entry */
7848 dv
= malloc(sizeof(*dv
));
7850 dv
->dev
= malloc(len
);
7855 update
->space
= NULL
;
7859 /* count how many spares will be converted to members */
7860 for (i
= 0; i
< map
->num_members
; i
++) {
7861 dl
= serial_to_dl(inf
[i
].serial
, super
);
7863 /* hmm maybe it failed?, nothing we can do about
7868 if (count_memberships(dl
, super
) == 0)
7871 len
+= activate
* sizeof(struct imsm_disk
);
7878 /* check if we need a larger metadata buffer */
7879 if (super
->next_buf
)
7880 buf_len
= super
->next_len
;
7882 buf_len
= super
->len
;
7884 if (__le32_to_cpu(mpb
->mpb_size
) + len
> buf_len
) {
7885 /* ok we need a larger buf than what is currently allocated
7886 * if this allocation fails process_update will notice that
7887 * ->next_len is set and ->next_buf is NULL
7889 buf_len
= ROUND_UP(__le32_to_cpu(mpb
->mpb_size
) + len
, 512);
7890 if (super
->next_buf
)
7891 free(super
->next_buf
);
7893 super
->next_len
= buf_len
;
7894 if (posix_memalign(&super
->next_buf
, 512, buf_len
) == 0)
7895 memset(super
->next_buf
, 0, buf_len
);
7897 super
->next_buf
= NULL
;
7901 /* must be called while manager is quiesced */
7902 static void imsm_delete(struct intel_super
*super
, struct dl
**dlp
, unsigned index
)
7904 struct imsm_super
*mpb
= super
->anchor
;
7906 struct imsm_dev
*dev
;
7907 struct imsm_map
*map
;
7908 int i
, j
, num_members
;
7911 dprintf("%s: deleting device[%d] from imsm_super\n",
7914 /* shift all indexes down one */
7915 for (iter
= super
->disks
; iter
; iter
= iter
->next
)
7916 if (iter
->index
> (int)index
)
7918 for (iter
= super
->missing
; iter
; iter
= iter
->next
)
7919 if (iter
->index
> (int)index
)
7922 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
7923 dev
= get_imsm_dev(super
, i
);
7924 map
= get_imsm_map(dev
, 0);
7925 num_members
= map
->num_members
;
7926 for (j
= 0; j
< num_members
; j
++) {
7927 /* update ord entries being careful not to propagate
7928 * ord-flags to the first map
7930 ord
= get_imsm_ord_tbl_ent(dev
, j
, -1);
7932 if (ord_to_idx(ord
) <= index
)
7935 map
= get_imsm_map(dev
, 0);
7936 set_imsm_ord_tbl_ent(map
, j
, ord_to_idx(ord
- 1));
7937 map
= get_imsm_map(dev
, 1);
7939 set_imsm_ord_tbl_ent(map
, j
, ord
- 1);
7944 super
->updates_pending
++;
7946 struct dl
*dl
= *dlp
;
7948 *dlp
= (*dlp
)->next
;
7949 __free_imsm_disk(dl
);
7952 #endif /* MDASSEMBLE */
7953 /*******************************************************************************
7954 * Function: open_backup_targets
7955 * Description: Function opens file descriptors for all devices given in
7958 * info : general array info
7959 * raid_disks : number of disks
7960 * raid_fds : table of device's file descriptors
7964 ******************************************************************************/
7965 int open_backup_targets(struct mdinfo
*info
, int raid_disks
, int *raid_fds
)
7969 for (sd
= info
->devs
; sd
; sd
= sd
->next
) {
7972 if (sd
->disk
.state
& (1<<MD_DISK_FAULTY
)) {
7973 dprintf("disk is faulty!!\n");
7977 if ((sd
->disk
.raid_disk
>= raid_disks
) ||
7978 (sd
->disk
.raid_disk
< 0))
7981 dn
= map_dev(sd
->disk
.major
,
7983 raid_fds
[sd
->disk
.raid_disk
] = dev_open(dn
, O_RDWR
);
7984 if (raid_fds
[sd
->disk
.raid_disk
] < 0) {
7985 fprintf(stderr
, "cannot open component\n");
7993 /*******************************************************************************
7994 * Function: init_migr_record_imsm
7995 * Description: Function inits imsm migration record
7997 * super : imsm internal array info
7998 * dev : device under migration
7999 * info : general array info to find the smallest device
8002 ******************************************************************************/
8003 void init_migr_record_imsm(struct supertype
*st
, struct imsm_dev
*dev
,
8004 struct mdinfo
*info
)
8006 struct intel_super
*super
= st
->sb
;
8007 struct migr_record
*migr_rec
= super
->migr_rec
;
8009 unsigned long long dsize
, dev_sectors
;
8010 long long unsigned min_dev_sectors
= -1LLU;
8014 struct imsm_map
*map_dest
= get_imsm_map(dev
, 0);
8015 struct imsm_map
*map_src
= get_imsm_map(dev
, 1);
8016 unsigned long long num_migr_units
;
8017 unsigned long long array_blocks
;
8019 memset(migr_rec
, 0, sizeof(struct migr_record
));
8020 migr_rec
->family_num
= __cpu_to_le32(super
->anchor
->family_num
);
8022 /* only ascending reshape supported now */
8023 migr_rec
->ascending_migr
= __cpu_to_le32(1);
8025 migr_rec
->dest_depth_per_unit
= GEN_MIGR_AREA_SIZE
/
8026 max(map_dest
->blocks_per_strip
, map_src
->blocks_per_strip
);
8027 migr_rec
->dest_depth_per_unit
*= map_dest
->blocks_per_strip
;
8028 new_data_disks
= imsm_num_data_members(dev
, 0);
8029 migr_rec
->blocks_per_unit
=
8030 __cpu_to_le32(migr_rec
->dest_depth_per_unit
* new_data_disks
);
8031 migr_rec
->dest_depth_per_unit
=
8032 __cpu_to_le32(migr_rec
->dest_depth_per_unit
);
8033 array_blocks
= info
->component_size
* new_data_disks
;
8035 array_blocks
/ __le32_to_cpu(migr_rec
->blocks_per_unit
);
8037 if (array_blocks
% __le32_to_cpu(migr_rec
->blocks_per_unit
))
8039 migr_rec
->num_migr_units
= __cpu_to_le32(num_migr_units
);
8041 migr_rec
->post_migr_vol_cap
= dev
->size_low
;
8042 migr_rec
->post_migr_vol_cap_hi
= dev
->size_high
;
8045 /* Find the smallest dev */
8046 for (sd
= info
->devs
; sd
; sd
= sd
->next
) {
8047 sprintf(nm
, "%d:%d", sd
->disk
.major
, sd
->disk
.minor
);
8048 fd
= dev_open(nm
, O_RDONLY
);
8051 get_dev_size(fd
, NULL
, &dsize
);
8052 dev_sectors
= dsize
/ 512;
8053 if (dev_sectors
< min_dev_sectors
)
8054 min_dev_sectors
= dev_sectors
;
8057 migr_rec
->ckpt_area_pba
= __cpu_to_le32(min_dev_sectors
-
8058 RAID_DISK_RESERVED_BLOCKS_IMSM_HI
);
8060 write_imsm_migr_rec(st
);
8065 /*******************************************************************************
8066 * Function: save_backup_imsm
8067 * Description: Function saves critical data stripes to Migration Copy Area
8068 * and updates the current migration unit status.
8069 * Use restore_stripes() to form a destination stripe,
8070 * and to write it to the Copy Area.
8072 * st : supertype information
8073 * dev : imsm device that backup is saved for
8074 * info : general array info
8075 * buf : input buffer
8076 * length : length of data to backup (blocks_per_unit)
8080 ******************************************************************************/
8081 int save_backup_imsm(struct supertype
*st
,
8082 struct imsm_dev
*dev
,
8083 struct mdinfo
*info
,
8088 struct intel_super
*super
= st
->sb
;
8089 unsigned long long *target_offsets
= NULL
;
8090 int *targets
= NULL
;
8092 struct imsm_map
*map_dest
= get_imsm_map(dev
, 0);
8093 int new_disks
= map_dest
->num_members
;
8094 int dest_layout
= 0;
8096 unsigned long long start
;
8097 int data_disks
= imsm_num_data_members(dev
, 0);
8099 targets
= malloc(new_disks
* sizeof(int));
8103 for (i
= 0; i
< new_disks
; i
++)
8106 target_offsets
= malloc(new_disks
* sizeof(unsigned long long));
8107 if (!target_offsets
)
8110 start
= info
->reshape_progress
* 512;
8111 for (i
= 0; i
< new_disks
; i
++) {
8112 target_offsets
[i
] = (unsigned long long)
8113 __le32_to_cpu(super
->migr_rec
->ckpt_area_pba
) * 512;
8114 /* move back copy area adderss, it will be moved forward
8115 * in restore_stripes() using start input variable
8117 target_offsets
[i
] -= start
/data_disks
;
8120 if (open_backup_targets(info
, new_disks
, targets
))
8123 dest_layout
= imsm_level_to_layout(map_dest
->raid_level
);
8124 dest_chunk
= __le16_to_cpu(map_dest
->blocks_per_strip
) * 512;
8126 if (restore_stripes(targets
, /* list of dest devices */
8127 target_offsets
, /* migration record offsets */
8130 map_dest
->raid_level
,
8132 -1, /* source backup file descriptor */
8133 0, /* input buf offset
8134 * always 0 buf is already offseted */
8138 fprintf(stderr
, Name
": Error restoring stripes\n");
8146 for (i
= 0; i
< new_disks
; i
++)
8147 if (targets
[i
] >= 0)
8151 free(target_offsets
);
8156 /*******************************************************************************
8157 * Function: save_checkpoint_imsm
8158 * Description: Function called for current unit status update
8159 * in the migration record. It writes it to disk.
8161 * super : imsm internal array info
8162 * info : general array info
8166 * 2: failure, means no valid migration record
8167 * / no general migration in progress /
8168 ******************************************************************************/
8169 int save_checkpoint_imsm(struct supertype
*st
, struct mdinfo
*info
, int state
)
8171 struct intel_super
*super
= st
->sb
;
8172 unsigned long long blocks_per_unit
;
8173 unsigned long long curr_migr_unit
;
8175 if (load_imsm_migr_rec(super
, info
) != 0) {
8176 dprintf("imsm: ERROR: Cannot read migration record "
8177 "for checkpoint save.\n");
8181 blocks_per_unit
= __le32_to_cpu(super
->migr_rec
->blocks_per_unit
);
8182 if (blocks_per_unit
== 0) {
8183 dprintf("imsm: no migration in progress.\n");
8186 curr_migr_unit
= info
->reshape_progress
/ blocks_per_unit
;
8187 /* check if array is alligned to copy area
8188 * if it is not alligned, add one to current migration unit value
8189 * this can happend on array reshape finish only
8191 if (info
->reshape_progress
% blocks_per_unit
)
8194 super
->migr_rec
->curr_migr_unit
=
8195 __cpu_to_le32(curr_migr_unit
);
8196 super
->migr_rec
->rec_status
= __cpu_to_le32(state
);
8197 super
->migr_rec
->dest_1st_member_lba
=
8198 __cpu_to_le32(curr_migr_unit
*
8199 __le32_to_cpu(super
->migr_rec
->dest_depth_per_unit
));
8200 if (write_imsm_migr_rec(st
) < 0) {
8201 dprintf("imsm: Cannot write migration record "
8202 "outside backup area\n");
8209 /*******************************************************************************
8210 * Function: recover_backup_imsm
8211 * Description: Function recovers critical data from the Migration Copy Area
8212 * while assembling an array.
8214 * super : imsm internal array info
8215 * info : general array info
8217 * 0 : success (or there is no data to recover)
8219 ******************************************************************************/
8220 int recover_backup_imsm(struct supertype
*st
, struct mdinfo
*info
)
8222 struct intel_super
*super
= st
->sb
;
8223 struct migr_record
*migr_rec
= super
->migr_rec
;
8224 struct imsm_map
*map_dest
= NULL
;
8225 struct intel_dev
*id
= NULL
;
8226 unsigned long long read_offset
;
8227 unsigned long long write_offset
;
8229 int *targets
= NULL
;
8230 int new_disks
, i
, err
;
8233 unsigned long curr_migr_unit
= __le32_to_cpu(migr_rec
->curr_migr_unit
);
8234 unsigned long num_migr_units
= __le32_to_cpu(migr_rec
->num_migr_units
);
8236 int skipped_disks
= 0;
8237 int max_degradation
;
8239 err
= sysfs_get_str(info
, NULL
, "array_state", (char *)buffer
, 20);
8243 /* recover data only during assemblation */
8244 if (strncmp(buffer
, "inactive", 8) != 0)
8246 /* no data to recover */
8247 if (__le32_to_cpu(migr_rec
->rec_status
) == UNIT_SRC_NORMAL
)
8249 if (curr_migr_unit
>= num_migr_units
)
8252 /* find device during reshape */
8253 for (id
= super
->devlist
; id
; id
= id
->next
)
8254 if (is_gen_migration(id
->dev
))
8259 map_dest
= get_imsm_map(id
->dev
, 0);
8260 new_disks
= map_dest
->num_members
;
8261 max_degradation
= new_disks
- imsm_num_data_members(id
->dev
, 0);
8263 read_offset
= (unsigned long long)
8264 __le32_to_cpu(migr_rec
->ckpt_area_pba
) * 512;
8266 write_offset
= ((unsigned long long)
8267 __le32_to_cpu(migr_rec
->dest_1st_member_lba
) +
8268 __le32_to_cpu(map_dest
->pba_of_lba0
)) * 512;
8270 unit_len
= __le32_to_cpu(migr_rec
->dest_depth_per_unit
) * 512;
8271 if (posix_memalign((void **)&buf
, 512, unit_len
) != 0)
8273 targets
= malloc(new_disks
* sizeof(int));
8277 open_backup_targets(info
, new_disks
, targets
);
8279 for (i
= 0; i
< new_disks
; i
++) {
8280 if (targets
[i
] < 0) {
8284 if (lseek64(targets
[i
], read_offset
, SEEK_SET
) < 0) {
8286 Name
": Cannot seek to block: %s\n",
8290 if ((unsigned)read(targets
[i
], buf
, unit_len
) != unit_len
) {
8292 Name
": Cannot read copy area block: %s\n",
8296 if (lseek64(targets
[i
], write_offset
, SEEK_SET
) < 0) {
8298 Name
": Cannot seek to block: %s\n",
8302 if ((unsigned)write(targets
[i
], buf
, unit_len
) != unit_len
) {
8304 Name
": Cannot restore block: %s\n",
8310 if (skipped_disks
> max_degradation
) {
8312 Name
": Cannot restore data from backup."
8313 " Too many failed disks\n");
8317 if (save_checkpoint_imsm(st
, info
, UNIT_SRC_NORMAL
)) {
8318 /* ignore error == 2, this can mean end of reshape here
8320 dprintf("imsm: Cannot write checkpoint to "
8321 "migration record (UNIT_SRC_NORMAL) during restart\n");
8327 for (i
= 0; i
< new_disks
; i
++)
8336 static char disk_by_path
[] = "/dev/disk/by-path/";
8338 static const char *imsm_get_disk_controller_domain(const char *path
)
8340 char disk_path
[PATH_MAX
];
8344 strncpy(disk_path
, disk_by_path
, PATH_MAX
- 1);
8345 strncat(disk_path
, path
, PATH_MAX
- strlen(disk_path
) - 1);
8346 if (stat(disk_path
, &st
) == 0) {
8347 struct sys_dev
* hba
;
8350 path
= devt_to_devpath(st
.st_rdev
);
8353 hba
= find_disk_attached_hba(-1, path
);
8354 if (hba
&& hba
->type
== SYS_DEV_SAS
)
8356 else if (hba
&& hba
->type
== SYS_DEV_SATA
)
8360 dprintf("path: %s hba: %s attached: %s\n",
8361 path
, (hba
) ? hba
->path
: "NULL", drv
);
8369 static int imsm_find_array_minor_by_subdev(int subdev
, int container
, int *minor
)
8371 char subdev_name
[20];
8372 struct mdstat_ent
*mdstat
;
8374 sprintf(subdev_name
, "%d", subdev
);
8375 mdstat
= mdstat_by_subdev(subdev_name
, container
);
8379 *minor
= mdstat
->devnum
;
8380 free_mdstat(mdstat
);
8384 static int imsm_reshape_is_allowed_on_container(struct supertype
*st
,
8385 struct geo_params
*geo
,
8386 int *old_raid_disks
)
8388 /* currently we only support increasing the number of devices
8389 * for a container. This increases the number of device for each
8390 * member array. They must all be RAID0 or RAID5.
8393 struct mdinfo
*info
, *member
;
8394 int devices_that_can_grow
= 0;
8396 dprintf("imsm: imsm_reshape_is_allowed_on_container(ENTER): "
8397 "st->devnum = (%i)\n",
8400 if (geo
->size
!= -1 ||
8401 geo
->level
!= UnSet
||
8402 geo
->layout
!= UnSet
||
8403 geo
->chunksize
!= 0 ||
8404 geo
->raid_disks
== UnSet
) {
8405 dprintf("imsm: Container operation is allowed for "
8406 "raid disks number change only.\n");
8410 info
= container_content_imsm(st
, NULL
);
8411 for (member
= info
; member
; member
= member
->next
) {
8415 dprintf("imsm: checking device_num: %i\n",
8416 member
->container_member
);
8418 if (geo
->raid_disks
<= member
->array
.raid_disks
) {
8419 /* we work on container for Online Capacity Expansion
8420 * only so raid_disks has to grow
8422 dprintf("imsm: for container operation raid disks "
8423 "increase is required\n");
8427 if ((info
->array
.level
!= 0) &&
8428 (info
->array
.level
!= 5)) {
8429 /* we cannot use this container with other raid level
8431 dprintf("imsm: for container operation wrong"
8432 " raid level (%i) detected\n",
8436 /* check for platform support
8437 * for this raid level configuration
8439 struct intel_super
*super
= st
->sb
;
8440 if (!is_raid_level_supported(super
->orom
,
8441 member
->array
.level
,
8443 dprintf("platform does not support raid%d with"
8447 geo
->raid_disks
> 1 ? "s" : "");
8450 /* check if component size is aligned to chunk size
8452 if (info
->component_size
%
8453 (info
->array
.chunk_size
/512)) {
8454 dprintf("Component size is not aligned to "
8460 if (*old_raid_disks
&&
8461 info
->array
.raid_disks
!= *old_raid_disks
)
8463 *old_raid_disks
= info
->array
.raid_disks
;
8465 /* All raid5 and raid0 volumes in container
8466 * have to be ready for Online Capacity Expansion
8467 * so they need to be assembled. We have already
8468 * checked that no recovery etc is happening.
8470 result
= imsm_find_array_minor_by_subdev(member
->container_member
,
8474 dprintf("imsm: cannot find array\n");
8477 devices_that_can_grow
++;
8480 if (!member
&& devices_that_can_grow
)
8484 dprintf("\tContainer operation allowed\n");
8486 dprintf("\tError: %i\n", ret_val
);
8491 /* Function: get_spares_for_grow
8492 * Description: Allocates memory and creates list of spare devices
8493 * avaliable in container. Checks if spare drive size is acceptable.
8494 * Parameters: Pointer to the supertype structure
8495 * Returns: Pointer to the list of spare devices (mdinfo structure) on success,
8498 static struct mdinfo
*get_spares_for_grow(struct supertype
*st
)
8500 unsigned long long min_size
= min_acceptable_spare_size_imsm(st
);
8501 return container_choose_spares(st
, min_size
, NULL
, NULL
, NULL
, 0);
8504 /******************************************************************************
8505 * function: imsm_create_metadata_update_for_reshape
8506 * Function creates update for whole IMSM container.
8508 ******************************************************************************/
8509 static int imsm_create_metadata_update_for_reshape(
8510 struct supertype
*st
,
8511 struct geo_params
*geo
,
8513 struct imsm_update_reshape
**updatep
)
8515 struct intel_super
*super
= st
->sb
;
8516 struct imsm_super
*mpb
= super
->anchor
;
8517 int update_memory_size
= 0;
8518 struct imsm_update_reshape
*u
= NULL
;
8519 struct mdinfo
*spares
= NULL
;
8521 int delta_disks
= 0;
8524 dprintf("imsm_update_metadata_for_reshape(enter) raid_disks = %i\n",
8527 delta_disks
= geo
->raid_disks
- old_raid_disks
;
8529 /* size of all update data without anchor */
8530 update_memory_size
= sizeof(struct imsm_update_reshape
);
8532 /* now add space for spare disks that we need to add. */
8533 update_memory_size
+= sizeof(u
->new_disks
[0]) * (delta_disks
- 1);
8535 u
= calloc(1, update_memory_size
);
8538 "cannot get memory for imsm_update_reshape update\n");
8541 u
->type
= update_reshape_container_disks
;
8542 u
->old_raid_disks
= old_raid_disks
;
8543 u
->new_raid_disks
= geo
->raid_disks
;
8545 /* now get spare disks list
8547 spares
= get_spares_for_grow(st
);
8550 || delta_disks
> spares
->array
.spare_disks
) {
8551 fprintf(stderr
, Name
": imsm: ERROR: Cannot get spare devices "
8552 "for %s.\n", geo
->dev_name
);
8557 /* we have got spares
8558 * update disk list in imsm_disk list table in anchor
8560 dprintf("imsm: %i spares are available.\n\n",
8561 spares
->array
.spare_disks
);
8564 for (i
= 0; i
< delta_disks
; i
++) {
8569 u
->new_disks
[i
] = makedev(dev
->disk
.major
,
8571 dl
= get_disk_super(super
, dev
->disk
.major
, dev
->disk
.minor
);
8572 dl
->index
= mpb
->num_disks
;
8582 dprintf("imsm: reshape update preparation :");
8583 if (i
== delta_disks
) {
8586 return update_memory_size
;
8589 dprintf(" Error\n");
8594 /******************************************************************************
8595 * function: imsm_create_metadata_update_for_migration()
8596 * Creates update for IMSM array.
8598 ******************************************************************************/
8599 static int imsm_create_metadata_update_for_migration(
8600 struct supertype
*st
,
8601 struct geo_params
*geo
,
8602 struct imsm_update_reshape_migration
**updatep
)
8604 struct intel_super
*super
= st
->sb
;
8605 int update_memory_size
= 0;
8606 struct imsm_update_reshape_migration
*u
= NULL
;
8607 struct imsm_dev
*dev
;
8608 int previous_level
= -1;
8610 dprintf("imsm_create_metadata_update_for_migration(enter)"
8611 " New Level = %i\n", geo
->level
);
8613 /* size of all update data without anchor */
8614 update_memory_size
= sizeof(struct imsm_update_reshape_migration
);
8616 u
= calloc(1, update_memory_size
);
8618 dprintf("error: cannot get memory for "
8619 "imsm_create_metadata_update_for_migration\n");
8622 u
->type
= update_reshape_migration
;
8623 u
->subdev
= super
->current_vol
;
8624 u
->new_level
= geo
->level
;
8625 u
->new_layout
= geo
->layout
;
8626 u
->new_raid_disks
= u
->old_raid_disks
= geo
->raid_disks
;
8627 u
->new_disks
[0] = -1;
8628 u
->new_chunksize
= -1;
8630 dev
= get_imsm_dev(super
, u
->subdev
);
8632 struct imsm_map
*map
;
8634 map
= get_imsm_map(dev
, 0);
8636 int current_chunk_size
=
8637 __le16_to_cpu(map
->blocks_per_strip
) / 2;
8639 if (geo
->chunksize
!= current_chunk_size
) {
8640 u
->new_chunksize
= geo
->chunksize
/ 1024;
8642 "chunk size change from %i to %i\n",
8643 current_chunk_size
, u
->new_chunksize
);
8645 previous_level
= map
->raid_level
;
8648 if ((geo
->level
== 5) && (previous_level
== 0)) {
8649 struct mdinfo
*spares
= NULL
;
8651 u
->new_raid_disks
++;
8652 spares
= get_spares_for_grow(st
);
8653 if ((spares
== NULL
) || (spares
->array
.spare_disks
< 1)) {
8656 update_memory_size
= 0;
8657 dprintf("error: cannot get spare device "
8658 "for requested migration");
8663 dprintf("imsm: reshape update preparation : OK\n");
8666 return update_memory_size
;
8669 static void imsm_update_metadata_locally(struct supertype
*st
,
8672 struct metadata_update mu
;
8677 mu
.space_list
= NULL
;
8679 imsm_prepare_update(st
, &mu
);
8680 imsm_process_update(st
, &mu
);
8682 while (mu
.space_list
) {
8683 void **space
= mu
.space_list
;
8684 mu
.space_list
= *space
;
8689 /***************************************************************************
8690 * Function: imsm_analyze_change
8691 * Description: Function analyze change for single volume
8692 * and validate if transition is supported
8693 * Parameters: Geometry parameters, supertype structure
8694 * Returns: Operation type code on success, -1 if fail
8695 ****************************************************************************/
8696 enum imsm_reshape_type
imsm_analyze_change(struct supertype
*st
,
8697 struct geo_params
*geo
)
8706 getinfo_super_imsm_volume(st
, &info
, NULL
);
8707 if ((geo
->level
!= info
.array
.level
) &&
8708 (geo
->level
>= 0) &&
8709 (geo
->level
!= UnSet
)) {
8710 switch (info
.array
.level
) {
8712 if (geo
->level
== 5) {
8713 change
= CH_MIGRATION
;
8714 if (geo
->layout
!= ALGORITHM_LEFT_ASYMMETRIC
) {
8716 Name
" Error. Requested Layout "
8717 "not supported (left-asymmetric layout "
8718 "is supported only)!\n");
8720 goto analyse_change_exit
;
8722 layout
= geo
->layout
;
8724 devNumChange
= 1; /* parity disk added */
8725 } else if (geo
->level
== 10) {
8726 change
= CH_TAKEOVER
;
8728 devNumChange
= 2; /* two mirrors added */
8729 layout
= 0x102; /* imsm supported layout */
8734 if (geo
->level
== 0) {
8735 change
= CH_TAKEOVER
;
8737 devNumChange
= -(geo
->raid_disks
/2);
8738 layout
= 0; /* imsm raid0 layout */
8744 Name
" Error. Level Migration from %d to %d "
8746 info
.array
.level
, geo
->level
);
8747 goto analyse_change_exit
;
8750 geo
->level
= info
.array
.level
;
8752 if ((geo
->layout
!= info
.array
.layout
)
8753 && ((geo
->layout
!= UnSet
) && (geo
->layout
!= -1))) {
8754 change
= CH_MIGRATION
;
8755 if ((info
.array
.layout
== 0)
8756 && (info
.array
.level
== 5)
8757 && (geo
->layout
== 5)) {
8758 /* reshape 5 -> 4 */
8759 } else if ((info
.array
.layout
== 5)
8760 && (info
.array
.level
== 5)
8761 && (geo
->layout
== 0)) {
8762 /* reshape 4 -> 5 */
8767 Name
" Error. Layout Migration from %d to %d "
8769 info
.array
.layout
, geo
->layout
);
8771 goto analyse_change_exit
;
8774 geo
->layout
= info
.array
.layout
;
8776 if ((geo
->chunksize
> 0) && (geo
->chunksize
!= UnSet
)
8777 && (geo
->chunksize
!= info
.array
.chunk_size
))
8778 change
= CH_MIGRATION
;
8780 geo
->chunksize
= info
.array
.chunk_size
;
8782 chunk
= geo
->chunksize
/ 1024;
8783 if (!validate_geometry_imsm(st
,
8786 geo
->raid_disks
+ devNumChange
,
8793 struct intel_super
*super
= st
->sb
;
8794 struct imsm_super
*mpb
= super
->anchor
;
8796 if (mpb
->num_raid_devs
> 1) {
8798 Name
" Error. Cannot perform operation on %s"
8799 "- for this operation it MUST be single "
8800 "array in container\n",
8806 analyse_change_exit
:
8811 int imsm_takeover(struct supertype
*st
, struct geo_params
*geo
)
8813 struct intel_super
*super
= st
->sb
;
8814 struct imsm_update_takeover
*u
;
8816 u
= malloc(sizeof(struct imsm_update_takeover
));
8820 u
->type
= update_takeover
;
8821 u
->subarray
= super
->current_vol
;
8823 /* 10->0 transition */
8824 if (geo
->level
== 0)
8825 u
->direction
= R10_TO_R0
;
8827 /* 0->10 transition */
8828 if (geo
->level
== 10)
8829 u
->direction
= R0_TO_R10
;
8831 /* update metadata locally */
8832 imsm_update_metadata_locally(st
, u
,
8833 sizeof(struct imsm_update_takeover
));
8834 /* and possibly remotely */
8835 if (st
->update_tail
)
8836 append_metadata_update(st
, u
,
8837 sizeof(struct imsm_update_takeover
));
8844 static int imsm_reshape_super(struct supertype
*st
, long long size
, int level
,
8845 int layout
, int chunksize
, int raid_disks
,
8846 int delta_disks
, char *backup
, char *dev
,
8850 struct geo_params geo
;
8852 dprintf("imsm: reshape_super called.\n");
8854 memset(&geo
, 0, sizeof(struct geo_params
));
8857 geo
.dev_id
= st
->devnum
;
8860 geo
.layout
= layout
;
8861 geo
.chunksize
= chunksize
;
8862 geo
.raid_disks
= raid_disks
;
8863 if (delta_disks
!= UnSet
)
8864 geo
.raid_disks
+= delta_disks
;
8866 dprintf("\tfor level : %i\n", geo
.level
);
8867 dprintf("\tfor raid_disks : %i\n", geo
.raid_disks
);
8869 if (experimental() == 0)
8872 if (st
->container_dev
== st
->devnum
) {
8873 /* On container level we can only increase number of devices. */
8874 dprintf("imsm: info: Container operation\n");
8875 int old_raid_disks
= 0;
8877 if (imsm_reshape_is_allowed_on_container(
8878 st
, &geo
, &old_raid_disks
)) {
8879 struct imsm_update_reshape
*u
= NULL
;
8882 len
= imsm_create_metadata_update_for_reshape(
8883 st
, &geo
, old_raid_disks
, &u
);
8886 dprintf("imsm: Cannot prepare update\n");
8887 goto exit_imsm_reshape_super
;
8891 /* update metadata locally */
8892 imsm_update_metadata_locally(st
, u
, len
);
8893 /* and possibly remotely */
8894 if (st
->update_tail
)
8895 append_metadata_update(st
, u
, len
);
8900 fprintf(stderr
, Name
": (imsm) Operation "
8901 "is not allowed on this container\n");
8904 /* On volume level we support following operations
8905 * - takeover: raid10 -> raid0; raid0 -> raid10
8906 * - chunk size migration
8907 * - migration: raid5 -> raid0; raid0 -> raid5
8909 struct intel_super
*super
= st
->sb
;
8910 struct intel_dev
*dev
= super
->devlist
;
8912 dprintf("imsm: info: Volume operation\n");
8913 /* find requested device */
8915 if (imsm_find_array_minor_by_subdev(
8916 dev
->index
, st
->container_dev
, &devnum
) == 0
8917 && devnum
== geo
.dev_id
)
8922 fprintf(stderr
, Name
" Cannot find %s (%i) subarray\n",
8923 geo
.dev_name
, geo
.dev_id
);
8924 goto exit_imsm_reshape_super
;
8926 super
->current_vol
= dev
->index
;
8927 change
= imsm_analyze_change(st
, &geo
);
8930 ret_val
= imsm_takeover(st
, &geo
);
8932 case CH_MIGRATION
: {
8933 struct imsm_update_reshape_migration
*u
= NULL
;
8935 imsm_create_metadata_update_for_migration(
8939 "Cannot prepare update\n");
8943 /* update metadata locally */
8944 imsm_update_metadata_locally(st
, u
, len
);
8945 /* and possibly remotely */
8946 if (st
->update_tail
)
8947 append_metadata_update(st
, u
, len
);
8957 exit_imsm_reshape_super
:
8958 dprintf("imsm: reshape_super Exit code = %i\n", ret_val
);
8962 /*******************************************************************************
8963 * Function: wait_for_reshape_imsm
8964 * Description: Function writes new sync_max value and waits until
8965 * reshape process reach new position
8967 * sra : general array info
8968 * ndata : number of disks in new array's layout
8971 * 1 : there is no reshape in progress,
8973 ******************************************************************************/
8974 int wait_for_reshape_imsm(struct mdinfo
*sra
, int ndata
)
8976 int fd
= sysfs_get_fd(sra
, NULL
, "reshape_position");
8977 unsigned long long completed
;
8978 /* to_complete : new sync_max position */
8979 unsigned long long to_complete
= sra
->reshape_progress
;
8980 unsigned long long position_to_set
= to_complete
/ ndata
;
8983 dprintf("imsm: wait_for_reshape_imsm() "
8984 "cannot open reshape_position\n");
8988 if (sysfs_fd_get_ll(fd
, &completed
) < 0) {
8989 dprintf("imsm: wait_for_reshape_imsm() "
8990 "cannot read reshape_position (no reshape in progres)\n");
8995 if (completed
> to_complete
) {
8996 dprintf("imsm: wait_for_reshape_imsm() "
8997 "wrong next position to set %llu (%llu)\n",
8998 to_complete
, completed
);
9002 dprintf("Position set: %llu\n", position_to_set
);
9003 if (sysfs_set_num(sra
, NULL
, "sync_max",
9004 position_to_set
) != 0) {
9005 dprintf("imsm: wait_for_reshape_imsm() "
9006 "cannot set reshape position to %llu\n",
9017 select(fd
+1, &rfds
, NULL
, NULL
, NULL
);
9018 if (sysfs_get_str(sra
, NULL
, "sync_action",
9020 strncmp(action
, "reshape", 7) != 0)
9022 if (sysfs_fd_get_ll(fd
, &completed
) < 0) {
9023 dprintf("imsm: wait_for_reshape_imsm() "
9024 "cannot read reshape_position (in loop)\n");
9028 } while (completed
< to_complete
);
9034 /*******************************************************************************
9035 * Function: check_degradation_change
9036 * Description: Check that array hasn't become failed.
9038 * info : for sysfs access
9039 * sources : source disks descriptors
9040 * degraded: previous degradation level
9043 ******************************************************************************/
9044 int check_degradation_change(struct mdinfo
*info
,
9048 unsigned long long new_degraded
;
9049 sysfs_get_ll(info
, NULL
, "degraded", &new_degraded
);
9050 if (new_degraded
!= (unsigned long long)degraded
) {
9051 /* check each device to ensure it is still working */
9054 for (sd
= info
->devs
; sd
; sd
= sd
->next
) {
9055 if (sd
->disk
.state
& (1<<MD_DISK_FAULTY
))
9057 if (sd
->disk
.state
& (1<<MD_DISK_SYNC
)) {
9059 if (sysfs_get_str(info
,
9060 sd
, "state", sbuf
, 20) < 0 ||
9061 strstr(sbuf
, "faulty") ||
9062 strstr(sbuf
, "in_sync") == NULL
) {
9063 /* this device is dead */
9064 sd
->disk
.state
= (1<<MD_DISK_FAULTY
);
9065 if (sd
->disk
.raid_disk
>= 0 &&
9066 sources
[sd
->disk
.raid_disk
] >= 0) {
9068 sd
->disk
.raid_disk
]);
9069 sources
[sd
->disk
.raid_disk
] =
9078 return new_degraded
;
9081 /*******************************************************************************
9082 * Function: imsm_manage_reshape
9083 * Description: Function finds array under reshape and it manages reshape
9084 * process. It creates stripes backups (if required) and sets
9087 * afd : Backup handle (nattive) - not used
9088 * sra : general array info
9089 * reshape : reshape parameters - not used
9090 * st : supertype structure
9091 * blocks : size of critical section [blocks]
9092 * fds : table of source device descriptor
9093 * offsets : start of array (offest per devices)
9095 * destfd : table of destination device descriptor
9096 * destoffsets : table of destination offsets (per device)
9098 * 1 : success, reshape is done
9100 ******************************************************************************/
9101 static int imsm_manage_reshape(
9102 int afd
, struct mdinfo
*sra
, struct reshape
*reshape
,
9103 struct supertype
*st
, unsigned long backup_blocks
,
9104 int *fds
, unsigned long long *offsets
,
9105 int dests
, int *destfd
, unsigned long long *destoffsets
)
9108 struct intel_super
*super
= st
->sb
;
9109 struct intel_dev
*dv
= NULL
;
9110 struct imsm_dev
*dev
= NULL
;
9111 struct imsm_map
*map_src
;
9112 int migr_vol_qan
= 0;
9113 int ndata
, odata
; /* [bytes] */
9114 int chunk
; /* [bytes] */
9115 struct migr_record
*migr_rec
;
9117 unsigned int buf_size
; /* [bytes] */
9118 unsigned long long max_position
; /* array size [bytes] */
9119 unsigned long long next_step
; /* [blocks]/[bytes] */
9120 unsigned long long old_data_stripe_length
;
9121 unsigned long long start_src
; /* [bytes] */
9122 unsigned long long start
; /* [bytes] */
9123 unsigned long long start_buf_shift
; /* [bytes] */
9125 int source_layout
= 0;
9127 if (!fds
|| !offsets
|| !sra
)
9130 /* Find volume during the reshape */
9131 for (dv
= super
->devlist
; dv
; dv
= dv
->next
) {
9132 if (dv
->dev
->vol
.migr_type
== MIGR_GEN_MIGR
9133 && dv
->dev
->vol
.migr_state
== 1) {
9138 /* Only one volume can migrate at the same time */
9139 if (migr_vol_qan
!= 1) {
9140 fprintf(stderr
, Name
" : %s", migr_vol_qan
?
9141 "Number of migrating volumes greater than 1\n" :
9142 "There is no volume during migrationg\n");
9146 map_src
= get_imsm_map(dev
, 1);
9147 if (map_src
== NULL
)
9150 ndata
= imsm_num_data_members(dev
, 0);
9151 odata
= imsm_num_data_members(dev
, 1);
9153 chunk
= __le16_to_cpu(map_src
->blocks_per_strip
) * 512;
9154 old_data_stripe_length
= odata
* chunk
;
9156 migr_rec
= super
->migr_rec
;
9158 /* initialize migration record for start condition */
9159 if (sra
->reshape_progress
== 0)
9160 init_migr_record_imsm(st
, dev
, sra
);
9162 if (__le32_to_cpu(migr_rec
->rec_status
) != UNIT_SRC_NORMAL
) {
9163 dprintf("imsm: cannot restart migration when data "
9164 "are present in copy area.\n");
9170 buf_size
= __le32_to_cpu(migr_rec
->blocks_per_unit
) * 512;
9171 /* extend buffer size for parity disk */
9172 buf_size
+= __le32_to_cpu(migr_rec
->dest_depth_per_unit
) * 512;
9173 /* add space for stripe aligment */
9174 buf_size
+= old_data_stripe_length
;
9175 if (posix_memalign((void **)&buf
, 4096, buf_size
)) {
9176 dprintf("imsm: Cannot allocate checpoint buffer\n");
9180 max_position
= sra
->component_size
* ndata
;
9181 source_layout
= imsm_level_to_layout(map_src
->raid_level
);
9183 while (__le32_to_cpu(migr_rec
->curr_migr_unit
) <
9184 __le32_to_cpu(migr_rec
->num_migr_units
)) {
9185 /* current reshape position [blocks] */
9186 unsigned long long current_position
=
9187 __le32_to_cpu(migr_rec
->blocks_per_unit
)
9188 * __le32_to_cpu(migr_rec
->curr_migr_unit
);
9189 unsigned long long border
;
9191 /* Check that array hasn't become failed.
9193 degraded
= check_degradation_change(sra
, fds
, degraded
);
9195 dprintf("imsm: Abort reshape due to degradation"
9196 " level (%i)\n", degraded
);
9200 next_step
= __le32_to_cpu(migr_rec
->blocks_per_unit
);
9202 if ((current_position
+ next_step
) > max_position
)
9203 next_step
= max_position
- current_position
;
9205 start
= current_position
* 512;
9207 /* allign reading start to old geometry */
9208 start_buf_shift
= start
% old_data_stripe_length
;
9209 start_src
= start
- start_buf_shift
;
9211 border
= (start_src
/ odata
) - (start
/ ndata
);
9213 if (border
<= __le32_to_cpu(migr_rec
->dest_depth_per_unit
)) {
9214 /* save critical stripes to buf
9215 * start - start address of current unit
9217 * start_src - start address of current unit
9218 * to backup alligned to source array
9221 unsigned long long next_step_filler
= 0;
9222 unsigned long long copy_length
= next_step
* 512;
9224 /* allign copy area length to stripe in old geometry */
9225 next_step_filler
= ((copy_length
+ start_buf_shift
)
9226 % old_data_stripe_length
);
9227 if (next_step_filler
)
9228 next_step_filler
= (old_data_stripe_length
9229 - next_step_filler
);
9230 dprintf("save_stripes() parameters: start = %llu,"
9231 "\tstart_src = %llu,\tnext_step*512 = %llu,"
9232 "\tstart_in_buf_shift = %llu,"
9233 "\tnext_step_filler = %llu\n",
9234 start
, start_src
, copy_length
,
9235 start_buf_shift
, next_step_filler
);
9237 if (save_stripes(fds
, offsets
, map_src
->num_members
,
9238 chunk
, map_src
->raid_level
,
9239 source_layout
, 0, NULL
, start_src
,
9241 next_step_filler
+ start_buf_shift
,
9243 dprintf("imsm: Cannot save stripes"
9247 /* Convert data to destination format and store it
9248 * in backup general migration area
9250 if (save_backup_imsm(st
, dev
, sra
,
9251 buf
+ start_buf_shift
, copy_length
)) {
9252 dprintf("imsm: Cannot save stripes to "
9253 "target devices\n");
9256 if (save_checkpoint_imsm(st
, sra
,
9257 UNIT_SRC_IN_CP_AREA
)) {
9258 dprintf("imsm: Cannot write checkpoint to "
9259 "migration record (UNIT_SRC_IN_CP_AREA)\n");
9263 /* set next step to use whole border area */
9264 border
/= next_step
;
9266 next_step
*= border
;
9268 /* When data backed up, checkpoint stored,
9269 * kick the kernel to reshape unit of data
9271 next_step
= next_step
+ sra
->reshape_progress
;
9272 /* limit next step to array max position */
9273 if (next_step
> max_position
)
9274 next_step
= max_position
;
9275 sysfs_set_num(sra
, NULL
, "suspend_lo", sra
->reshape_progress
);
9276 sysfs_set_num(sra
, NULL
, "suspend_hi", next_step
);
9277 sra
->reshape_progress
= next_step
;
9279 /* wait until reshape finish */
9280 if (wait_for_reshape_imsm(sra
, ndata
) < 0) {
9281 dprintf("wait_for_reshape_imsm returned error!\n");
9285 if (save_checkpoint_imsm(st
, sra
, UNIT_SRC_NORMAL
) == 1) {
9286 /* ignore error == 2, this can mean end of reshape here
9288 dprintf("imsm: Cannot write checkpoint to "
9289 "migration record (UNIT_SRC_NORMAL)\n");
9295 /* return '1' if done */
9303 #endif /* MDASSEMBLE */
9305 struct superswitch super_imsm
= {
9307 .examine_super
= examine_super_imsm
,
9308 .brief_examine_super
= brief_examine_super_imsm
,
9309 .brief_examine_subarrays
= brief_examine_subarrays_imsm
,
9310 .export_examine_super
= export_examine_super_imsm
,
9311 .detail_super
= detail_super_imsm
,
9312 .brief_detail_super
= brief_detail_super_imsm
,
9313 .write_init_super
= write_init_super_imsm
,
9314 .validate_geometry
= validate_geometry_imsm
,
9315 .add_to_super
= add_to_super_imsm
,
9316 .remove_from_super
= remove_from_super_imsm
,
9317 .detail_platform
= detail_platform_imsm
,
9318 .kill_subarray
= kill_subarray_imsm
,
9319 .update_subarray
= update_subarray_imsm
,
9320 .load_container
= load_container_imsm
,
9321 .default_geometry
= default_geometry_imsm
,
9322 .get_disk_controller_domain
= imsm_get_disk_controller_domain
,
9323 .reshape_super
= imsm_reshape_super
,
9324 .manage_reshape
= imsm_manage_reshape
,
9325 .recover_backup
= recover_backup_imsm
,
9327 .match_home
= match_home_imsm
,
9328 .uuid_from_super
= uuid_from_super_imsm
,
9329 .getinfo_super
= getinfo_super_imsm
,
9330 .getinfo_super_disks
= getinfo_super_disks_imsm
,
9331 .update_super
= update_super_imsm
,
9333 .avail_size
= avail_size_imsm
,
9334 .min_acceptable_spare_size
= min_acceptable_spare_size_imsm
,
9336 .compare_super
= compare_super_imsm
,
9338 .load_super
= load_super_imsm
,
9339 .init_super
= init_super_imsm
,
9340 .store_super
= store_super_imsm
,
9341 .free_super
= free_super_imsm
,
9342 .match_metadata_desc
= match_metadata_desc_imsm
,
9343 .container_content
= container_content_imsm
,
9351 .open_new
= imsm_open_new
,
9352 .set_array_state
= imsm_set_array_state
,
9353 .set_disk
= imsm_set_disk
,
9354 .sync_metadata
= imsm_sync_metadata
,
9355 .activate_spare
= imsm_activate_spare
,
9356 .process_update
= imsm_process_update
,
9357 .prepare_update
= imsm_prepare_update
,
9358 #endif /* MDASSEMBLE */