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 /* map selector for map managment
114 /* RAID map configuration infos. */
116 __u32 pba_of_lba0
; /* start address of partition */
117 __u32 blocks_per_member
;/* blocks per member */
118 __u32 num_data_stripes
; /* number of data stripes */
119 __u16 blocks_per_strip
;
120 __u8 map_state
; /* Normal, Uninitialized, Degraded, Failed */
121 #define IMSM_T_STATE_NORMAL 0
122 #define IMSM_T_STATE_UNINITIALIZED 1
123 #define IMSM_T_STATE_DEGRADED 2
124 #define IMSM_T_STATE_FAILED 3
126 #define IMSM_T_RAID0 0
127 #define IMSM_T_RAID1 1
128 #define IMSM_T_RAID5 5 /* since metadata version 1.2.02 ? */
129 __u8 num_members
; /* number of member disks */
130 __u8 num_domains
; /* number of parity domains */
131 __u8 failed_disk_num
; /* valid only when state is degraded */
133 __u32 filler
[7]; /* expansion area */
134 #define IMSM_ORD_REBUILD (1 << 24)
135 __u32 disk_ord_tbl
[1]; /* disk_ord_tbl[num_members],
136 * top byte contains some flags
138 } __attribute__ ((packed
));
141 __u32 curr_migr_unit
;
142 __u32 checkpoint_id
; /* id to access curr_migr_unit */
143 __u8 migr_state
; /* Normal or Migrating */
145 #define MIGR_REBUILD 1
146 #define MIGR_VERIFY 2 /* analagous to echo check > sync_action */
147 #define MIGR_GEN_MIGR 3
148 #define MIGR_STATE_CHANGE 4
149 #define MIGR_REPAIR 5
150 __u8 migr_type
; /* Initializing, Rebuilding, ... */
152 __u8 fs_state
; /* fast-sync state for CnG (0xff == disabled) */
153 __u16 verify_errors
; /* number of mismatches */
154 __u16 bad_blocks
; /* number of bad blocks during verify */
156 struct imsm_map map
[1];
157 /* here comes another one if migr_state */
158 } __attribute__ ((packed
));
161 __u8 volume
[MAX_RAID_SERIAL_LEN
];
164 #define DEV_BOOTABLE __cpu_to_le32(0x01)
165 #define DEV_BOOT_DEVICE __cpu_to_le32(0x02)
166 #define DEV_READ_COALESCING __cpu_to_le32(0x04)
167 #define DEV_WRITE_COALESCING __cpu_to_le32(0x08)
168 #define DEV_LAST_SHUTDOWN_DIRTY __cpu_to_le32(0x10)
169 #define DEV_HIDDEN_AT_BOOT __cpu_to_le32(0x20)
170 #define DEV_CURRENTLY_HIDDEN __cpu_to_le32(0x40)
171 #define DEV_VERIFY_AND_FIX __cpu_to_le32(0x80)
172 #define DEV_MAP_STATE_UNINIT __cpu_to_le32(0x100)
173 #define DEV_NO_AUTO_RECOVERY __cpu_to_le32(0x200)
174 #define DEV_CLONE_N_GO __cpu_to_le32(0x400)
175 #define DEV_CLONE_MAN_SYNC __cpu_to_le32(0x800)
176 #define DEV_CNG_MASTER_DISK_NUM __cpu_to_le32(0x1000)
177 __u32 status
; /* Persistent RaidDev status */
178 __u32 reserved_blocks
; /* Reserved blocks at beginning of volume */
182 __u8 cng_master_disk
;
186 #define IMSM_DEV_FILLERS 10
187 __u32 filler
[IMSM_DEV_FILLERS
];
189 } __attribute__ ((packed
));
192 __u8 sig
[MAX_SIGNATURE_LENGTH
]; /* 0x00 - 0x1F */
193 __u32 check_sum
; /* 0x20 - 0x23 MPB Checksum */
194 __u32 mpb_size
; /* 0x24 - 0x27 Size of MPB */
195 __u32 family_num
; /* 0x28 - 0x2B Checksum from first time this config was written */
196 __u32 generation_num
; /* 0x2C - 0x2F Incremented each time this array's MPB is written */
197 __u32 error_log_size
; /* 0x30 - 0x33 in bytes */
198 __u32 attributes
; /* 0x34 - 0x37 */
199 __u8 num_disks
; /* 0x38 Number of configured disks */
200 __u8 num_raid_devs
; /* 0x39 Number of configured volumes */
201 __u8 error_log_pos
; /* 0x3A */
202 __u8 fill
[1]; /* 0x3B */
203 __u32 cache_size
; /* 0x3c - 0x40 in mb */
204 __u32 orig_family_num
; /* 0x40 - 0x43 original family num */
205 __u32 pwr_cycle_count
; /* 0x44 - 0x47 simulated power cycle count for array */
206 __u32 bbm_log_size
; /* 0x48 - 0x4B - size of bad Block Mgmt Log in bytes */
207 #define IMSM_FILLERS 35
208 __u32 filler
[IMSM_FILLERS
]; /* 0x4C - 0xD7 RAID_MPB_FILLERS */
209 struct imsm_disk disk
[1]; /* 0xD8 diskTbl[numDisks] */
210 /* here comes imsm_dev[num_raid_devs] */
211 /* here comes BBM logs */
212 } __attribute__ ((packed
));
214 #define BBM_LOG_MAX_ENTRIES 254
216 struct bbm_log_entry
{
217 __u64 defective_block_start
;
218 #define UNREADABLE 0xFFFFFFFF
219 __u32 spare_block_offset
;
220 __u16 remapped_marked_count
;
222 } __attribute__ ((__packed__
));
225 __u32 signature
; /* 0xABADB10C */
227 __u32 reserved_spare_block_count
; /* 0 */
228 __u32 reserved
; /* 0xFFFF */
229 __u64 first_spare_lba
;
230 struct bbm_log_entry mapped_block_entries
[BBM_LOG_MAX_ENTRIES
];
231 } __attribute__ ((__packed__
));
235 static char *map_state_str
[] = { "normal", "uninitialized", "degraded", "failed" };
238 #define RAID_DISK_RESERVED_BLOCKS_IMSM_HI 2209
240 #define GEN_MIGR_AREA_SIZE 2048 /* General Migration Copy Area size in blocks */
242 #define UNIT_SRC_NORMAL 0 /* Source data for curr_migr_unit must
243 * be recovered using srcMap */
244 #define UNIT_SRC_IN_CP_AREA 1 /* Source data for curr_migr_unit has
245 * already been migrated and must
246 * be recovered from checkpoint area */
248 __u32 rec_status
; /* Status used to determine how to restart
249 * migration in case it aborts
251 __u32 curr_migr_unit
; /* 0..numMigrUnits-1 */
252 __u32 family_num
; /* Family number of MPB
253 * containing the RaidDev
254 * that is migrating */
255 __u32 ascending_migr
; /* True if migrating in increasing
257 __u32 blocks_per_unit
; /* Num disk blocks per unit of operation */
258 __u32 dest_depth_per_unit
; /* Num member blocks each destMap
260 * advances per unit-of-operation */
261 __u32 ckpt_area_pba
; /* Pba of first block of ckpt copy area */
262 __u32 dest_1st_member_lba
; /* First member lba on first
263 * stripe of destination */
264 __u32 num_migr_units
; /* Total num migration units-of-op */
265 __u32 post_migr_vol_cap
; /* Size of volume after
266 * migration completes */
267 __u32 post_migr_vol_cap_hi
; /* Expansion space for LBA64 */
268 __u32 ckpt_read_disk_num
; /* Which member disk in destSubMap[0] the
269 * migration ckpt record was read from
270 * (for recovered migrations) */
271 } __attribute__ ((__packed__
));
273 static __u8
migr_type(struct imsm_dev
*dev
)
275 if (dev
->vol
.migr_type
== MIGR_VERIFY
&&
276 dev
->status
& DEV_VERIFY_AND_FIX
)
279 return dev
->vol
.migr_type
;
282 static void set_migr_type(struct imsm_dev
*dev
, __u8 migr_type
)
284 /* for compatibility with older oroms convert MIGR_REPAIR, into
285 * MIGR_VERIFY w/ DEV_VERIFY_AND_FIX status
287 if (migr_type
== MIGR_REPAIR
) {
288 dev
->vol
.migr_type
= MIGR_VERIFY
;
289 dev
->status
|= DEV_VERIFY_AND_FIX
;
291 dev
->vol
.migr_type
= migr_type
;
292 dev
->status
&= ~DEV_VERIFY_AND_FIX
;
296 static unsigned int sector_count(__u32 bytes
)
298 return ((bytes
+ (512-1)) & (~(512-1))) / 512;
301 static unsigned int mpb_sectors(struct imsm_super
*mpb
)
303 return sector_count(__le32_to_cpu(mpb
->mpb_size
));
307 struct imsm_dev
*dev
;
308 struct intel_dev
*next
;
313 enum sys_dev_type type
;
316 struct intel_hba
*next
;
323 /* internal representation of IMSM metadata */
326 void *buf
; /* O_DIRECT buffer for reading/writing metadata */
327 struct imsm_super
*anchor
; /* immovable parameters */
330 void *migr_rec_buf
; /* buffer for I/O operations */
331 struct migr_record
*migr_rec
; /* migration record */
333 size_t len
; /* size of the 'buf' allocation */
334 void *next_buf
; /* for realloc'ing buf from the manager */
336 int updates_pending
; /* count of pending updates for mdmon */
337 int current_vol
; /* index of raid device undergoing creation */
338 __u32 create_offset
; /* common start for 'current_vol' */
339 __u32 random
; /* random data for seeding new family numbers */
340 struct intel_dev
*devlist
;
344 __u8 serial
[MAX_RAID_SERIAL_LEN
];
347 struct imsm_disk disk
;
350 struct extent
*e
; /* for determining freespace @ create */
351 int raiddisk
; /* slot to fill in autolayout */
353 } *disks
, *current_disk
;
354 struct dl
*disk_mgmt_list
; /* list of disks to add/remove while mdmon
356 struct dl
*missing
; /* disks removed while we weren't looking */
357 struct bbm_log
*bbm_log
;
358 struct intel_hba
*hba
; /* device path of the raid controller for this metadata */
359 const struct imsm_orom
*orom
; /* platform firmware support */
360 struct intel_super
*next
; /* (temp) list for disambiguating family_num */
364 struct imsm_disk disk
;
365 #define IMSM_UNKNOWN_OWNER (-1)
367 struct intel_disk
*next
;
371 unsigned long long start
, size
;
374 /* definitions of reshape process types */
375 enum imsm_reshape_type
{
380 /* definition of messages passed to imsm_process_update */
381 enum imsm_update_type
{
382 update_activate_spare
,
386 update_add_remove_disk
,
387 update_reshape_container_disks
,
388 update_reshape_migration
,
390 update_general_migration_checkpoint
,
393 struct imsm_update_activate_spare
{
394 enum imsm_update_type type
;
398 struct imsm_update_activate_spare
*next
;
411 enum takeover_direction
{
415 struct imsm_update_takeover
{
416 enum imsm_update_type type
;
418 enum takeover_direction direction
;
421 struct imsm_update_reshape
{
422 enum imsm_update_type type
;
426 int new_disks
[1]; /* new_raid_disks - old_raid_disks makedev number */
429 struct imsm_update_reshape_migration
{
430 enum imsm_update_type type
;
433 /* fields for array migration changes
440 int new_disks
[1]; /* new_raid_disks - old_raid_disks makedev number */
443 struct imsm_update_general_migration_checkpoint
{
444 enum imsm_update_type type
;
445 __u32 curr_migr_unit
;
449 __u8 serial
[MAX_RAID_SERIAL_LEN
];
452 struct imsm_update_create_array
{
453 enum imsm_update_type type
;
458 struct imsm_update_kill_array
{
459 enum imsm_update_type type
;
463 struct imsm_update_rename_array
{
464 enum imsm_update_type type
;
465 __u8 name
[MAX_RAID_SERIAL_LEN
];
469 struct imsm_update_add_remove_disk
{
470 enum imsm_update_type type
;
474 static const char *_sys_dev_type
[] = {
475 [SYS_DEV_UNKNOWN
] = "Unknown",
476 [SYS_DEV_SAS
] = "SAS",
477 [SYS_DEV_SATA
] = "SATA"
480 const char *get_sys_dev_type(enum sys_dev_type type
)
482 if (type
>= SYS_DEV_MAX
)
483 type
= SYS_DEV_UNKNOWN
;
485 return _sys_dev_type
[type
];
488 static struct intel_hba
* alloc_intel_hba(struct sys_dev
*device
)
490 struct intel_hba
*result
= malloc(sizeof(*result
));
492 result
->type
= device
->type
;
493 result
->path
= strdup(device
->path
);
495 if (result
->path
&& (result
->pci_id
= strrchr(result
->path
, '/')) != NULL
)
501 static struct intel_hba
* find_intel_hba(struct intel_hba
*hba
, struct sys_dev
*device
)
503 struct intel_hba
*result
=NULL
;
504 for (result
= hba
; result
; result
= result
->next
) {
505 if (result
->type
== device
->type
&& strcmp(result
->path
, device
->path
) == 0)
511 static int attach_hba_to_super(struct intel_super
*super
, struct sys_dev
*device
)
513 struct intel_hba
*hba
;
515 /* check if disk attached to Intel HBA */
516 hba
= find_intel_hba(super
->hba
, device
);
519 /* Check if HBA is already attached to super */
520 if (super
->hba
== NULL
) {
521 super
->hba
= alloc_intel_hba(device
);
526 /* Intel metadata allows for all disks attached to the same type HBA.
527 * Do not sypport odf HBA types mixing
529 if (device
->type
!= hba
->type
)
535 hba
->next
= alloc_intel_hba(device
);
539 static struct sys_dev
* find_disk_attached_hba(int fd
, const char *devname
)
541 struct sys_dev
*list
, *elem
, *prev
;
544 if ((list
= find_intel_devices()) == NULL
)
548 disk_path
= (char *) devname
;
550 disk_path
= diskfd_to_devpath(fd
);
557 for (prev
= NULL
, elem
= list
; elem
; prev
= elem
, elem
= elem
->next
) {
558 if (path_attached_to_hba(disk_path
, elem
->path
)) {
562 prev
->next
= elem
->next
;
564 if (disk_path
!= devname
)
570 if (disk_path
!= devname
)
578 static int find_intel_hba_capability(int fd
, struct intel_super
*super
,
581 static struct supertype
*match_metadata_desc_imsm(char *arg
)
583 struct supertype
*st
;
585 if (strcmp(arg
, "imsm") != 0 &&
586 strcmp(arg
, "default") != 0
590 st
= malloc(sizeof(*st
));
593 memset(st
, 0, sizeof(*st
));
594 st
->container_dev
= NoMdDev
;
595 st
->ss
= &super_imsm
;
596 st
->max_devs
= IMSM_MAX_DEVICES
;
597 st
->minor_version
= 0;
603 static __u8
*get_imsm_version(struct imsm_super
*mpb
)
605 return &mpb
->sig
[MPB_SIG_LEN
];
609 /* retrieve a disk directly from the anchor when the anchor is known to be
610 * up-to-date, currently only at load time
612 static struct imsm_disk
*__get_imsm_disk(struct imsm_super
*mpb
, __u8 index
)
614 if (index
>= mpb
->num_disks
)
616 return &mpb
->disk
[index
];
619 /* retrieve the disk description based on a index of the disk
622 static struct dl
*get_imsm_dl_disk(struct intel_super
*super
, __u8 index
)
626 for (d
= super
->disks
; d
; d
= d
->next
)
627 if (d
->index
== index
)
632 /* retrieve a disk from the parsed metadata */
633 static struct imsm_disk
*get_imsm_disk(struct intel_super
*super
, __u8 index
)
637 dl
= get_imsm_dl_disk(super
, index
);
644 /* generate a checksum directly from the anchor when the anchor is known to be
645 * up-to-date, currently only at load or write_super after coalescing
647 static __u32
__gen_imsm_checksum(struct imsm_super
*mpb
)
649 __u32 end
= mpb
->mpb_size
/ sizeof(end
);
650 __u32
*p
= (__u32
*) mpb
;
654 sum
+= __le32_to_cpu(*p
);
658 return sum
- __le32_to_cpu(mpb
->check_sum
);
661 static size_t sizeof_imsm_map(struct imsm_map
*map
)
663 return sizeof(struct imsm_map
) + sizeof(__u32
) * (map
->num_members
- 1);
666 struct imsm_map
*get_imsm_map(struct imsm_dev
*dev
, int second_map
)
668 /* A device can have 2 maps if it is in the middle of a migration.
670 * MAP_0 or 0 - we return the first map
671 * MAP_1 or 1 - we return the second map if it exists, else NULL
672 * -1 - we return the second map if it exists, else the first
674 struct imsm_map
*map
= &dev
->vol
.map
[0];
675 struct imsm_map
*map2
= NULL
;
677 if (dev
->vol
.migr_state
)
678 map2
= (void *)map
+ sizeof_imsm_map(map
);
680 switch (second_map
) {
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
);
1033 static int is_gen_migration(struct imsm_dev
*dev
);
1036 static __u64
blocks_per_migr_unit(struct intel_super
*super
,
1037 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 (is_gen_migration(dev
)) {
2317 info
->reshape_active
= 1;
2318 info
->new_level
= get_imsm_raid_level(map
);
2319 info
->new_layout
= imsm_level_to_layout(info
->new_level
);
2320 info
->new_chunk
= __le16_to_cpu(map
->blocks_per_strip
) << 9;
2321 info
->delta_disks
= map
->num_members
- prev_map
->num_members
;
2322 if (info
->delta_disks
) {
2323 /* this needs to be applied to every array
2326 info
->reshape_active
= CONTAINER_RESHAPE
;
2328 /* We shape information that we give to md might have to be
2329 * modify to cope with md's requirement for reshaping arrays.
2330 * For example, when reshaping a RAID0, md requires it to be
2331 * presented as a degraded RAID4.
2332 * Also if a RAID0 is migrating to a RAID5 we need to specify
2333 * the array as already being RAID5, but the 'before' layout
2334 * is a RAID4-like layout.
2336 switch (info
->array
.level
) {
2338 switch(info
->new_level
) {
2340 /* conversion is happening as RAID4 */
2341 info
->array
.level
= 4;
2342 info
->array
.raid_disks
+= 1;
2345 /* conversion is happening as RAID5 */
2346 info
->array
.level
= 5;
2347 info
->array
.layout
= ALGORITHM_PARITY_N
;
2348 info
->delta_disks
-= 1;
2351 /* FIXME error message */
2352 info
->array
.level
= UnSet
;
2358 info
->new_level
= UnSet
;
2359 info
->new_layout
= UnSet
;
2360 info
->new_chunk
= info
->array
.chunk_size
;
2361 info
->delta_disks
= 0;
2365 info
->disk
.major
= dl
->major
;
2366 info
->disk
.minor
= dl
->minor
;
2367 info
->disk
.number
= dl
->index
;
2368 info
->disk
.raid_disk
= get_imsm_disk_slot(map_to_analyse
,
2372 info
->data_offset
= __le32_to_cpu(map_to_analyse
->pba_of_lba0
);
2373 info
->component_size
=
2374 __le32_to_cpu(map_to_analyse
->blocks_per_member
);
2376 /* check component size aligment
2378 component_size_alligment
=
2379 info
->component_size
% (info
->array
.chunk_size
/512);
2381 if (component_size_alligment
&&
2382 (info
->array
.level
!= 1) && (info
->array
.level
!= UnSet
)) {
2383 dprintf("imsm: reported component size alligned from %llu ",
2384 info
->component_size
);
2385 info
->component_size
-= component_size_alligment
;
2386 dprintf("to %llu (%i).\n",
2387 info
->component_size
, component_size_alligment
);
2390 memset(info
->uuid
, 0, sizeof(info
->uuid
));
2391 info
->recovery_start
= MaxSector
;
2393 info
->reshape_progress
= 0;
2394 info
->resync_start
= MaxSector
;
2395 if ((map_to_analyse
->map_state
== IMSM_T_STATE_UNINITIALIZED
||
2397 imsm_reshape_blocks_arrays_changes(super
) == 0) {
2398 info
->resync_start
= 0;
2400 if (dev
->vol
.migr_state
) {
2401 switch (migr_type(dev
)) {
2404 __u64 blocks_per_unit
= blocks_per_migr_unit(super
,
2406 __u64 units
= __le32_to_cpu(dev
->vol
.curr_migr_unit
);
2408 info
->resync_start
= blocks_per_unit
* units
;
2411 case MIGR_GEN_MIGR
: {
2412 __u64 blocks_per_unit
= blocks_per_migr_unit(super
,
2414 __u64 units
= __le32_to_cpu(migr_rec
->curr_migr_unit
);
2415 unsigned long long array_blocks
;
2418 if (__le32_to_cpu(migr_rec
->ascending_migr
) &&
2420 (__le32_to_cpu(migr_rec
->num_migr_units
)-1)) &&
2421 (super
->migr_rec
->rec_status
==
2422 __cpu_to_le32(UNIT_SRC_IN_CP_AREA
)))
2425 info
->reshape_progress
= blocks_per_unit
* units
;
2427 dprintf("IMSM: General Migration checkpoint : %llu "
2428 "(%llu) -> read reshape progress : %llu\n",
2429 (unsigned long long)units
,
2430 (unsigned long long)blocks_per_unit
,
2431 info
->reshape_progress
);
2433 used_disks
= imsm_num_data_members(dev
, 1);
2434 if (used_disks
> 0) {
2435 array_blocks
= map
->blocks_per_member
*
2437 /* round array size down to closest MB
2439 info
->custom_array_size
= (array_blocks
2440 >> SECT_PER_MB_SHIFT
)
2441 << SECT_PER_MB_SHIFT
;
2445 /* we could emulate the checkpointing of
2446 * 'sync_action=check' migrations, but for now
2447 * we just immediately complete them
2450 /* this is handled by container_content_imsm() */
2451 case MIGR_STATE_CHANGE
:
2452 /* FIXME handle other migrations */
2454 /* we are not dirty, so... */
2455 info
->resync_start
= MaxSector
;
2459 strncpy(info
->name
, (char *) dev
->volume
, MAX_RAID_SERIAL_LEN
);
2460 info
->name
[MAX_RAID_SERIAL_LEN
] = 0;
2462 info
->array
.major_version
= -1;
2463 info
->array
.minor_version
= -2;
2464 devname
= devnum2devname(st
->container_dev
);
2465 *info
->text_version
= '\0';
2467 sprintf(info
->text_version
, "/%s/%d", devname
, info
->container_member
);
2469 info
->safe_mode_delay
= 4000; /* 4 secs like the Matrix driver */
2470 uuid_from_super_imsm(st
, info
->uuid
);
2474 for (i
=0; i
<map_disks
; i
++) {
2476 if (i
< info
->array
.raid_disks
) {
2477 struct imsm_disk
*dsk
;
2478 j
= get_imsm_disk_idx(dev
, i
, -1);
2479 dsk
= get_imsm_disk(super
, j
);
2480 if (dsk
&& (dsk
->status
& CONFIGURED_DISK
))
2487 static __u8
imsm_check_degraded(struct intel_super
*super
, struct imsm_dev
*dev
,
2488 int failed
, int look_in_map
);
2490 static int imsm_count_failed(struct intel_super
*super
, struct imsm_dev
*dev
,
2493 static void manage_second_map(struct intel_super
*super
, struct imsm_dev
*dev
)
2495 if (is_gen_migration(dev
)) {
2498 struct imsm_map
*map2
= get_imsm_map(dev
, MAP_1
);
2500 failed
= imsm_count_failed(super
, dev
, MAP_1
);
2501 map_state
= imsm_check_degraded(super
, dev
, failed
,
2503 if (map2
->map_state
!= map_state
) {
2504 map2
->map_state
= map_state
;
2505 super
->updates_pending
++;
2510 static struct imsm_disk
*get_imsm_missing(struct intel_super
*super
, __u8 index
)
2514 for (d
= super
->missing
; d
; d
= d
->next
)
2515 if (d
->index
== index
)
2520 static void getinfo_super_imsm(struct supertype
*st
, struct mdinfo
*info
, char *map
)
2522 struct intel_super
*super
= st
->sb
;
2523 struct imsm_disk
*disk
;
2524 int map_disks
= info
->array
.raid_disks
;
2525 int max_enough
= -1;
2527 struct imsm_super
*mpb
;
2529 if (super
->current_vol
>= 0) {
2530 getinfo_super_imsm_volume(st
, info
, map
);
2533 memset(info
, 0, sizeof(*info
));
2535 /* Set raid_disks to zero so that Assemble will always pull in valid
2538 info
->array
.raid_disks
= 0;
2539 info
->array
.level
= LEVEL_CONTAINER
;
2540 info
->array
.layout
= 0;
2541 info
->array
.md_minor
= -1;
2542 info
->array
.ctime
= 0; /* N/A for imsm */
2543 info
->array
.utime
= 0;
2544 info
->array
.chunk_size
= 0;
2546 info
->disk
.major
= 0;
2547 info
->disk
.minor
= 0;
2548 info
->disk
.raid_disk
= -1;
2549 info
->reshape_active
= 0;
2550 info
->array
.major_version
= -1;
2551 info
->array
.minor_version
= -2;
2552 strcpy(info
->text_version
, "imsm");
2553 info
->safe_mode_delay
= 0;
2554 info
->disk
.number
= -1;
2555 info
->disk
.state
= 0;
2557 info
->recovery_start
= MaxSector
;
2558 info
->recovery_blocked
= imsm_reshape_blocks_arrays_changes(st
->sb
);
2560 /* do we have the all the insync disks that we expect? */
2561 mpb
= super
->anchor
;
2563 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
2564 struct imsm_dev
*dev
= get_imsm_dev(super
, i
);
2565 int failed
, enough
, j
, missing
= 0;
2566 struct imsm_map
*map
;
2569 failed
= imsm_count_failed(super
, dev
, MAP_0
);
2570 state
= imsm_check_degraded(super
, dev
, failed
, MAP_0
);
2571 map
= get_imsm_map(dev
, 0);
2573 /* any newly missing disks?
2574 * (catches single-degraded vs double-degraded)
2576 for (j
= 0; j
< map
->num_members
; j
++) {
2577 __u32 ord
= get_imsm_ord_tbl_ent(dev
, j
, 0);
2578 __u32 idx
= ord_to_idx(ord
);
2580 if (!(ord
& IMSM_ORD_REBUILD
) &&
2581 get_imsm_missing(super
, idx
)) {
2587 if (state
== IMSM_T_STATE_FAILED
)
2589 else if (state
== IMSM_T_STATE_DEGRADED
&&
2590 (state
!= map
->map_state
|| missing
))
2592 else /* we're normal, or already degraded */
2595 /* in the missing/failed disk case check to see
2596 * if at least one array is runnable
2598 max_enough
= max(max_enough
, enough
);
2600 dprintf("%s: enough: %d\n", __func__
, max_enough
);
2601 info
->container_enough
= max_enough
;
2604 __u32 reserved
= imsm_reserved_sectors(super
, super
->disks
);
2606 disk
= &super
->disks
->disk
;
2607 info
->data_offset
= __le32_to_cpu(disk
->total_blocks
) - reserved
;
2608 info
->component_size
= reserved
;
2609 info
->disk
.state
= is_configured(disk
) ? (1 << MD_DISK_ACTIVE
) : 0;
2610 /* we don't change info->disk.raid_disk here because
2611 * this state will be finalized in mdmon after we have
2612 * found the 'most fresh' version of the metadata
2614 info
->disk
.state
|= is_failed(disk
) ? (1 << MD_DISK_FAULTY
) : 0;
2615 info
->disk
.state
|= is_spare(disk
) ? 0 : (1 << MD_DISK_SYNC
);
2618 /* only call uuid_from_super_imsm when this disk is part of a populated container,
2619 * ->compare_super may have updated the 'num_raid_devs' field for spares
2621 if (info
->disk
.state
& (1 << MD_DISK_SYNC
) || super
->anchor
->num_raid_devs
)
2622 uuid_from_super_imsm(st
, info
->uuid
);
2624 memcpy(info
->uuid
, uuid_zero
, sizeof(uuid_zero
));
2626 /* I don't know how to compute 'map' on imsm, so use safe default */
2629 for (i
= 0; i
< map_disks
; i
++)
2635 /* allocates memory and fills disk in mdinfo structure
2636 * for each disk in array */
2637 struct mdinfo
*getinfo_super_disks_imsm(struct supertype
*st
)
2639 struct mdinfo
*mddev
= NULL
;
2640 struct intel_super
*super
= st
->sb
;
2641 struct imsm_disk
*disk
;
2644 if (!super
|| !super
->disks
)
2647 mddev
= malloc(sizeof(*mddev
));
2649 fprintf(stderr
, Name
": Failed to allocate memory.\n");
2652 memset(mddev
, 0, sizeof(*mddev
));
2656 tmp
= malloc(sizeof(*tmp
));
2658 fprintf(stderr
, Name
": Failed to allocate memory.\n");
2663 memset(tmp
, 0, sizeof(*tmp
));
2665 tmp
->next
= mddev
->devs
;
2667 tmp
->disk
.number
= count
++;
2668 tmp
->disk
.major
= dl
->major
;
2669 tmp
->disk
.minor
= dl
->minor
;
2670 tmp
->disk
.state
= is_configured(disk
) ?
2671 (1 << MD_DISK_ACTIVE
) : 0;
2672 tmp
->disk
.state
|= is_failed(disk
) ? (1 << MD_DISK_FAULTY
) : 0;
2673 tmp
->disk
.state
|= is_spare(disk
) ? 0 : (1 << MD_DISK_SYNC
);
2674 tmp
->disk
.raid_disk
= -1;
2680 static int update_super_imsm(struct supertype
*st
, struct mdinfo
*info
,
2681 char *update
, char *devname
, int verbose
,
2682 int uuid_set
, char *homehost
)
2684 /* For 'assemble' and 'force' we need to return non-zero if any
2685 * change was made. For others, the return value is ignored.
2686 * Update options are:
2687 * force-one : This device looks a bit old but needs to be included,
2688 * update age info appropriately.
2689 * assemble: clear any 'faulty' flag to allow this device to
2691 * force-array: Array is degraded but being forced, mark it clean
2692 * if that will be needed to assemble it.
2694 * newdev: not used ????
2695 * grow: Array has gained a new device - this is currently for
2697 * resync: mark as dirty so a resync will happen.
2698 * name: update the name - preserving the homehost
2699 * uuid: Change the uuid of the array to match watch is given
2701 * Following are not relevant for this imsm:
2702 * sparc2.2 : update from old dodgey metadata
2703 * super-minor: change the preferred_minor number
2704 * summaries: update redundant counters.
2705 * homehost: update the recorded homehost
2706 * _reshape_progress: record new reshape_progress position.
2709 struct intel_super
*super
= st
->sb
;
2710 struct imsm_super
*mpb
;
2712 /* we can only update container info */
2713 if (!super
|| super
->current_vol
>= 0 || !super
->anchor
)
2716 mpb
= super
->anchor
;
2718 if (strcmp(update
, "uuid") == 0 && uuid_set
&& !info
->update_private
)
2720 else if (strcmp(update
, "uuid") == 0 && uuid_set
&& info
->update_private
) {
2721 mpb
->orig_family_num
= *((__u32
*) info
->update_private
);
2723 } else if (strcmp(update
, "uuid") == 0) {
2724 __u32
*new_family
= malloc(sizeof(*new_family
));
2726 /* update orig_family_number with the incoming random
2727 * data, report the new effective uuid, and store the
2728 * new orig_family_num for future updates.
2731 memcpy(&mpb
->orig_family_num
, info
->uuid
, sizeof(__u32
));
2732 uuid_from_super_imsm(st
, info
->uuid
);
2733 *new_family
= mpb
->orig_family_num
;
2734 info
->update_private
= new_family
;
2737 } else if (strcmp(update
, "assemble") == 0)
2742 /* successful update? recompute checksum */
2744 mpb
->check_sum
= __le32_to_cpu(__gen_imsm_checksum(mpb
));
2749 static size_t disks_to_mpb_size(int disks
)
2753 size
= sizeof(struct imsm_super
);
2754 size
+= (disks
- 1) * sizeof(struct imsm_disk
);
2755 size
+= 2 * sizeof(struct imsm_dev
);
2756 /* up to 2 maps per raid device (-2 for imsm_maps in imsm_dev */
2757 size
+= (4 - 2) * sizeof(struct imsm_map
);
2758 /* 4 possible disk_ord_tbl's */
2759 size
+= 4 * (disks
- 1) * sizeof(__u32
);
2764 static __u64
avail_size_imsm(struct supertype
*st
, __u64 devsize
)
2766 if (devsize
< (MPB_SECTOR_CNT
+ IMSM_RESERVED_SECTORS
))
2769 return devsize
- (MPB_SECTOR_CNT
+ IMSM_RESERVED_SECTORS
);
2772 static void free_devlist(struct intel_super
*super
)
2774 struct intel_dev
*dv
;
2776 while (super
->devlist
) {
2777 dv
= super
->devlist
->next
;
2778 free(super
->devlist
->dev
);
2779 free(super
->devlist
);
2780 super
->devlist
= dv
;
2784 static void imsm_copy_dev(struct imsm_dev
*dest
, struct imsm_dev
*src
)
2786 memcpy(dest
, src
, sizeof_imsm_dev(src
, 0));
2789 static int compare_super_imsm(struct supertype
*st
, struct supertype
*tst
)
2793 * 0 same, or first was empty, and second was copied
2794 * 1 second had wrong number
2796 * 3 wrong other info
2798 struct intel_super
*first
= st
->sb
;
2799 struct intel_super
*sec
= tst
->sb
;
2806 /* in platform dependent environment test if the disks
2807 * use the same Intel hba
2809 if (!check_env("IMSM_NO_PLATFORM")) {
2810 if (!first
->hba
|| !sec
->hba
||
2811 (first
->hba
->type
!= sec
->hba
->type
)) {
2813 "HBAs of devices does not match %s != %s\n",
2814 first
->hba
? get_sys_dev_type(first
->hba
->type
) : NULL
,
2815 sec
->hba
? get_sys_dev_type(sec
->hba
->type
) : NULL
);
2820 /* if an anchor does not have num_raid_devs set then it is a free
2823 if (first
->anchor
->num_raid_devs
> 0 &&
2824 sec
->anchor
->num_raid_devs
> 0) {
2825 /* Determine if these disks might ever have been
2826 * related. Further disambiguation can only take place
2827 * in load_super_imsm_all
2829 __u32 first_family
= first
->anchor
->orig_family_num
;
2830 __u32 sec_family
= sec
->anchor
->orig_family_num
;
2832 if (memcmp(first
->anchor
->sig
, sec
->anchor
->sig
,
2833 MAX_SIGNATURE_LENGTH
) != 0)
2836 if (first_family
== 0)
2837 first_family
= first
->anchor
->family_num
;
2838 if (sec_family
== 0)
2839 sec_family
= sec
->anchor
->family_num
;
2841 if (first_family
!= sec_family
)
2847 /* if 'first' is a spare promote it to a populated mpb with sec's
2850 if (first
->anchor
->num_raid_devs
== 0 &&
2851 sec
->anchor
->num_raid_devs
> 0) {
2853 struct intel_dev
*dv
;
2854 struct imsm_dev
*dev
;
2856 /* we need to copy raid device info from sec if an allocation
2857 * fails here we don't associate the spare
2859 for (i
= 0; i
< sec
->anchor
->num_raid_devs
; i
++) {
2860 dv
= malloc(sizeof(*dv
));
2863 dev
= malloc(sizeof_imsm_dev(get_imsm_dev(sec
, i
), 1));
2870 dv
->next
= first
->devlist
;
2871 first
->devlist
= dv
;
2873 if (i
< sec
->anchor
->num_raid_devs
) {
2874 /* allocation failure */
2875 free_devlist(first
);
2876 fprintf(stderr
, "imsm: failed to associate spare\n");
2879 first
->anchor
->num_raid_devs
= sec
->anchor
->num_raid_devs
;
2880 first
->anchor
->orig_family_num
= sec
->anchor
->orig_family_num
;
2881 first
->anchor
->family_num
= sec
->anchor
->family_num
;
2882 memcpy(first
->anchor
->sig
, sec
->anchor
->sig
, MAX_SIGNATURE_LENGTH
);
2883 for (i
= 0; i
< sec
->anchor
->num_raid_devs
; i
++)
2884 imsm_copy_dev(get_imsm_dev(first
, i
), get_imsm_dev(sec
, i
));
2890 static void fd2devname(int fd
, char *name
)
2894 char dname
[PATH_MAX
];
2899 if (fstat(fd
, &st
) != 0)
2901 sprintf(path
, "/sys/dev/block/%d:%d",
2902 major(st
.st_rdev
), minor(st
.st_rdev
));
2904 rv
= readlink(path
, dname
, sizeof(dname
)-1);
2909 nm
= strrchr(dname
, '/');
2912 snprintf(name
, MAX_RAID_SERIAL_LEN
, "/dev/%s", nm
);
2916 extern int scsi_get_serial(int fd
, void *buf
, size_t buf_len
);
2918 static int imsm_read_serial(int fd
, char *devname
,
2919 __u8 serial
[MAX_RAID_SERIAL_LEN
])
2921 unsigned char scsi_serial
[255];
2930 memset(scsi_serial
, 0, sizeof(scsi_serial
));
2932 rv
= scsi_get_serial(fd
, scsi_serial
, sizeof(scsi_serial
));
2934 if (rv
&& check_env("IMSM_DEVNAME_AS_SERIAL")) {
2935 memset(serial
, 0, MAX_RAID_SERIAL_LEN
);
2936 fd2devname(fd
, (char *) serial
);
2943 Name
": Failed to retrieve serial for %s\n",
2948 rsp_len
= scsi_serial
[3];
2952 Name
": Failed to retrieve serial for %s\n",
2956 rsp_buf
= (char *) &scsi_serial
[4];
2958 /* trim all whitespace and non-printable characters and convert
2961 for (i
= 0, dest
= rsp_buf
; i
< rsp_len
; i
++) {
2964 /* ':' is reserved for use in placeholder serial
2965 * numbers for missing disks
2973 len
= dest
- rsp_buf
;
2976 /* truncate leading characters */
2977 if (len
> MAX_RAID_SERIAL_LEN
) {
2978 dest
+= len
- MAX_RAID_SERIAL_LEN
;
2979 len
= MAX_RAID_SERIAL_LEN
;
2982 memset(serial
, 0, MAX_RAID_SERIAL_LEN
);
2983 memcpy(serial
, dest
, len
);
2988 static int serialcmp(__u8
*s1
, __u8
*s2
)
2990 return strncmp((char *) s1
, (char *) s2
, MAX_RAID_SERIAL_LEN
);
2993 static void serialcpy(__u8
*dest
, __u8
*src
)
2995 strncpy((char *) dest
, (char *) src
, MAX_RAID_SERIAL_LEN
);
2998 static struct dl
*serial_to_dl(__u8
*serial
, struct intel_super
*super
)
3002 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
3003 if (serialcmp(dl
->serial
, serial
) == 0)
3009 static struct imsm_disk
*
3010 __serial_to_disk(__u8
*serial
, struct imsm_super
*mpb
, int *idx
)
3014 for (i
= 0; i
< mpb
->num_disks
; i
++) {
3015 struct imsm_disk
*disk
= __get_imsm_disk(mpb
, i
);
3017 if (serialcmp(disk
->serial
, serial
) == 0) {
3028 load_imsm_disk(int fd
, struct intel_super
*super
, char *devname
, int keep_fd
)
3030 struct imsm_disk
*disk
;
3035 __u8 serial
[MAX_RAID_SERIAL_LEN
];
3037 rv
= imsm_read_serial(fd
, devname
, serial
);
3042 dl
= calloc(1, sizeof(*dl
));
3046 Name
": failed to allocate disk buffer for %s\n",
3052 dl
->major
= major(stb
.st_rdev
);
3053 dl
->minor
= minor(stb
.st_rdev
);
3054 dl
->next
= super
->disks
;
3055 dl
->fd
= keep_fd
? fd
: -1;
3056 assert(super
->disks
== NULL
);
3058 serialcpy(dl
->serial
, serial
);
3061 fd2devname(fd
, name
);
3063 dl
->devname
= strdup(devname
);
3065 dl
->devname
= strdup(name
);
3067 /* look up this disk's index in the current anchor */
3068 disk
= __serial_to_disk(dl
->serial
, super
->anchor
, &dl
->index
);
3071 /* only set index on disks that are a member of a
3072 * populated contianer, i.e. one with raid_devs
3074 if (is_failed(&dl
->disk
))
3076 else if (is_spare(&dl
->disk
))
3084 /* When migrating map0 contains the 'destination' state while map1
3085 * contains the current state. When not migrating map0 contains the
3086 * current state. This routine assumes that map[0].map_state is set to
3087 * the current array state before being called.
3089 * Migration is indicated by one of the following states
3090 * 1/ Idle (migr_state=0 map0state=normal||unitialized||degraded||failed)
3091 * 2/ Initialize (migr_state=1 migr_type=MIGR_INIT map0state=normal
3092 * map1state=unitialized)
3093 * 3/ Repair (Resync) (migr_state=1 migr_type=MIGR_REPAIR map0state=normal
3095 * 4/ Rebuild (migr_state=1 migr_type=MIGR_REBUILD map0state=normal
3096 * map1state=degraded)
3097 * 5/ Migration (mig_state=1 migr_type=MIGR_GEN_MIGR map0state=normal
3100 static void migrate(struct imsm_dev
*dev
, struct intel_super
*super
,
3101 __u8 to_state
, int migr_type
)
3103 struct imsm_map
*dest
;
3104 struct imsm_map
*src
= get_imsm_map(dev
, 0);
3106 dev
->vol
.migr_state
= 1;
3107 set_migr_type(dev
, migr_type
);
3108 dev
->vol
.curr_migr_unit
= 0;
3109 dest
= get_imsm_map(dev
, 1);
3111 /* duplicate and then set the target end state in map[0] */
3112 memcpy(dest
, src
, sizeof_imsm_map(src
));
3113 if ((migr_type
== MIGR_REBUILD
) ||
3114 (migr_type
== MIGR_GEN_MIGR
)) {
3118 for (i
= 0; i
< src
->num_members
; i
++) {
3119 ord
= __le32_to_cpu(src
->disk_ord_tbl
[i
]);
3120 set_imsm_ord_tbl_ent(src
, i
, ord_to_idx(ord
));
3124 if (migr_type
== MIGR_GEN_MIGR
)
3125 /* Clear migration record */
3126 memset(super
->migr_rec
, 0, sizeof(struct migr_record
));
3128 src
->map_state
= to_state
;
3131 static void end_migration(struct imsm_dev
*dev
, struct intel_super
*super
,
3134 struct imsm_map
*map
= get_imsm_map(dev
, 0);
3135 struct imsm_map
*prev
= get_imsm_map(dev
, dev
->vol
.migr_state
);
3138 /* merge any IMSM_ORD_REBUILD bits that were not successfully
3139 * completed in the last migration.
3141 * FIXME add support for raid-level-migration
3143 if ((map_state
!= map
->map_state
) && (is_gen_migration(dev
) == 0) &&
3144 (prev
->map_state
!= IMSM_T_STATE_UNINITIALIZED
)) {
3145 /* when final map state is other than expected
3146 * merge maps (not for migration)
3150 for (i
= 0; i
< prev
->num_members
; i
++)
3151 for (j
= 0; j
< map
->num_members
; j
++)
3152 /* during online capacity expansion
3153 * disks position can be changed
3154 * if takeover is used
3156 if (ord_to_idx(map
->disk_ord_tbl
[j
]) ==
3157 ord_to_idx(prev
->disk_ord_tbl
[i
])) {
3158 map
->disk_ord_tbl
[j
] |=
3159 prev
->disk_ord_tbl
[i
];
3162 failed
= imsm_count_failed(super
, dev
, MAP_0
);
3163 map_state
= imsm_check_degraded(super
, dev
, failed
, MAP_0
);
3166 dev
->vol
.migr_state
= 0;
3167 set_migr_type(dev
, 0);
3168 dev
->vol
.curr_migr_unit
= 0;
3169 map
->map_state
= map_state
;
3173 static int parse_raid_devices(struct intel_super
*super
)
3176 struct imsm_dev
*dev_new
;
3177 size_t len
, len_migr
;
3179 size_t space_needed
= 0;
3180 struct imsm_super
*mpb
= super
->anchor
;
3182 for (i
= 0; i
< super
->anchor
->num_raid_devs
; i
++) {
3183 struct imsm_dev
*dev_iter
= __get_imsm_dev(super
->anchor
, i
);
3184 struct intel_dev
*dv
;
3186 len
= sizeof_imsm_dev(dev_iter
, 0);
3187 len_migr
= sizeof_imsm_dev(dev_iter
, 1);
3189 space_needed
+= len_migr
- len
;
3191 dv
= malloc(sizeof(*dv
));
3194 if (max_len
< len_migr
)
3196 if (max_len
> len_migr
)
3197 space_needed
+= max_len
- len_migr
;
3198 dev_new
= malloc(max_len
);
3203 imsm_copy_dev(dev_new
, dev_iter
);
3206 dv
->next
= super
->devlist
;
3207 super
->devlist
= dv
;
3210 /* ensure that super->buf is large enough when all raid devices
3213 if (__le32_to_cpu(mpb
->mpb_size
) + space_needed
> super
->len
) {
3216 len
= ROUND_UP(__le32_to_cpu(mpb
->mpb_size
) + space_needed
, 512);
3217 if (posix_memalign(&buf
, 512, len
) != 0)
3220 memcpy(buf
, super
->buf
, super
->len
);
3221 memset(buf
+ super
->len
, 0, len
- super
->len
);
3230 /* retrieve a pointer to the bbm log which starts after all raid devices */
3231 struct bbm_log
*__get_imsm_bbm_log(struct imsm_super
*mpb
)
3235 if (__le32_to_cpu(mpb
->bbm_log_size
)) {
3237 ptr
+= mpb
->mpb_size
- __le32_to_cpu(mpb
->bbm_log_size
);
3243 /*******************************************************************************
3244 * Function: check_mpb_migr_compatibility
3245 * Description: Function checks for unsupported migration features:
3246 * - migration optimization area (pba_of_lba0)
3247 * - descending reshape (ascending_migr)
3249 * super : imsm metadata information
3251 * 0 : migration is compatible
3252 * -1 : migration is not compatible
3253 ******************************************************************************/
3254 int check_mpb_migr_compatibility(struct intel_super
*super
)
3256 struct imsm_map
*map0
, *map1
;
3257 struct migr_record
*migr_rec
= super
->migr_rec
;
3260 for (i
= 0; i
< super
->anchor
->num_raid_devs
; i
++) {
3261 struct imsm_dev
*dev_iter
= __get_imsm_dev(super
->anchor
, i
);
3264 dev_iter
->vol
.migr_state
== 1 &&
3265 dev_iter
->vol
.migr_type
== MIGR_GEN_MIGR
) {
3266 /* This device is migrating */
3267 map0
= get_imsm_map(dev_iter
, 0);
3268 map1
= get_imsm_map(dev_iter
, 1);
3269 if (map0
->pba_of_lba0
!= map1
->pba_of_lba0
)
3270 /* migration optimization area was used */
3272 if (migr_rec
->ascending_migr
== 0
3273 && migr_rec
->dest_depth_per_unit
> 0)
3274 /* descending reshape not supported yet */
3281 static void __free_imsm(struct intel_super
*super
, int free_disks
);
3283 /* load_imsm_mpb - read matrix metadata
3284 * allocates super->mpb to be freed by free_imsm
3286 static int load_imsm_mpb(int fd
, struct intel_super
*super
, char *devname
)
3288 unsigned long long dsize
;
3289 unsigned long long sectors
;
3291 struct imsm_super
*anchor
;
3294 get_dev_size(fd
, NULL
, &dsize
);
3298 Name
": %s: device to small for imsm\n",
3303 if (lseek64(fd
, dsize
- (512 * 2), SEEK_SET
) < 0) {
3305 fprintf(stderr
, Name
3306 ": Cannot seek to anchor block on %s: %s\n",
3307 devname
, strerror(errno
));
3311 if (posix_memalign((void**)&anchor
, 512, 512) != 0) {
3314 Name
": Failed to allocate imsm anchor buffer"
3315 " on %s\n", devname
);
3318 if (read(fd
, anchor
, 512) != 512) {
3321 Name
": Cannot read anchor block on %s: %s\n",
3322 devname
, strerror(errno
));
3327 if (strncmp((char *) anchor
->sig
, MPB_SIGNATURE
, MPB_SIG_LEN
) != 0) {
3330 Name
": no IMSM anchor on %s\n", devname
);
3335 __free_imsm(super
, 0);
3336 /* reload capability and hba */
3338 /* capability and hba must be updated with new super allocation */
3339 find_intel_hba_capability(fd
, super
, devname
);
3340 super
->len
= ROUND_UP(anchor
->mpb_size
, 512);
3341 if (posix_memalign(&super
->buf
, 512, super
->len
) != 0) {
3344 Name
": unable to allocate %zu byte mpb buffer\n",
3349 memcpy(super
->buf
, anchor
, 512);
3351 sectors
= mpb_sectors(anchor
) - 1;
3354 if (posix_memalign(&super
->migr_rec_buf
, 512, 512) != 0) {
3355 fprintf(stderr
, Name
3356 ": %s could not allocate migr_rec buffer\n", __func__
);
3362 check_sum
= __gen_imsm_checksum(super
->anchor
);
3363 if (check_sum
!= __le32_to_cpu(super
->anchor
->check_sum
)) {
3366 Name
": IMSM checksum %x != %x on %s\n",
3368 __le32_to_cpu(super
->anchor
->check_sum
),
3376 /* read the extended mpb */
3377 if (lseek64(fd
, dsize
- (512 * (2 + sectors
)), SEEK_SET
) < 0) {
3380 Name
": Cannot seek to extended mpb on %s: %s\n",
3381 devname
, strerror(errno
));
3385 if ((unsigned)read(fd
, super
->buf
+ 512, super
->len
- 512) != super
->len
- 512) {
3388 Name
": Cannot read extended mpb on %s: %s\n",
3389 devname
, strerror(errno
));
3393 check_sum
= __gen_imsm_checksum(super
->anchor
);
3394 if (check_sum
!= __le32_to_cpu(super
->anchor
->check_sum
)) {
3397 Name
": IMSM checksum %x != %x on %s\n",
3398 check_sum
, __le32_to_cpu(super
->anchor
->check_sum
),
3403 /* FIXME the BBM log is disk specific so we cannot use this global
3404 * buffer for all disks. Ok for now since we only look at the global
3405 * bbm_log_size parameter to gate assembly
3407 super
->bbm_log
= __get_imsm_bbm_log(super
->anchor
);
3412 static int read_imsm_migr_rec(int fd
, struct intel_super
*super
);
3415 load_and_parse_mpb(int fd
, struct intel_super
*super
, char *devname
, int keep_fd
)
3419 err
= load_imsm_mpb(fd
, super
, devname
);
3422 err
= load_imsm_disk(fd
, super
, devname
, keep_fd
);
3425 err
= parse_raid_devices(super
);
3430 static void __free_imsm_disk(struct dl
*d
)
3442 static void free_imsm_disks(struct intel_super
*super
)
3446 while (super
->disks
) {
3448 super
->disks
= d
->next
;
3449 __free_imsm_disk(d
);
3451 while (super
->disk_mgmt_list
) {
3452 d
= super
->disk_mgmt_list
;
3453 super
->disk_mgmt_list
= d
->next
;
3454 __free_imsm_disk(d
);
3456 while (super
->missing
) {
3458 super
->missing
= d
->next
;
3459 __free_imsm_disk(d
);
3464 /* free all the pieces hanging off of a super pointer */
3465 static void __free_imsm(struct intel_super
*super
, int free_disks
)
3467 struct intel_hba
*elem
, *next
;
3473 /* unlink capability description */
3475 if (super
->migr_rec_buf
) {
3476 free(super
->migr_rec_buf
);
3477 super
->migr_rec_buf
= NULL
;
3480 free_imsm_disks(super
);
3481 free_devlist(super
);
3485 free((void *)elem
->path
);
3493 static void free_imsm(struct intel_super
*super
)
3495 __free_imsm(super
, 1);
3499 static void free_super_imsm(struct supertype
*st
)
3501 struct intel_super
*super
= st
->sb
;
3510 static struct intel_super
*alloc_super(void)
3512 struct intel_super
*super
= malloc(sizeof(*super
));
3515 memset(super
, 0, sizeof(*super
));
3516 super
->current_vol
= -1;
3517 super
->create_offset
= ~((__u32
) 0);
3523 * find and allocate hba and OROM/EFI based on valid fd of RAID component device
3525 static int find_intel_hba_capability(int fd
, struct intel_super
*super
, char *devname
)
3527 struct sys_dev
*hba_name
;
3530 if ((fd
< 0) || check_env("IMSM_NO_PLATFORM")) {
3535 hba_name
= find_disk_attached_hba(fd
, NULL
);
3539 Name
": %s is not attached to Intel(R) RAID controller.\n",
3543 rv
= attach_hba_to_super(super
, hba_name
);
3546 struct intel_hba
*hba
= super
->hba
;
3548 fprintf(stderr
, Name
": %s is attached to Intel(R) %s RAID "
3549 "controller (%s),\n"
3550 " but the container is assigned to Intel(R) "
3551 "%s RAID controller (",
3554 hba_name
->pci_id
? : "Err!",
3555 get_sys_dev_type(hba_name
->type
));
3558 fprintf(stderr
, "%s", hba
->pci_id
? : "Err!");
3560 fprintf(stderr
, ", ");
3564 fprintf(stderr
, ").\n"
3565 " Mixing devices attached to different controllers "
3566 "is not allowed.\n");
3568 free_sys_dev(&hba_name
);
3571 super
->orom
= find_imsm_capability(hba_name
->type
);
3572 free_sys_dev(&hba_name
);
3578 /* find_missing - helper routine for load_super_imsm_all that identifies
3579 * disks that have disappeared from the system. This routine relies on
3580 * the mpb being uptodate, which it is at load time.
3582 static int find_missing(struct intel_super
*super
)
3585 struct imsm_super
*mpb
= super
->anchor
;
3587 struct imsm_disk
*disk
;
3589 for (i
= 0; i
< mpb
->num_disks
; i
++) {
3590 disk
= __get_imsm_disk(mpb
, i
);
3591 dl
= serial_to_dl(disk
->serial
, super
);
3595 dl
= malloc(sizeof(*dl
));
3601 dl
->devname
= strdup("missing");
3603 serialcpy(dl
->serial
, disk
->serial
);
3606 dl
->next
= super
->missing
;
3607 super
->missing
= dl
;
3614 static struct intel_disk
*disk_list_get(__u8
*serial
, struct intel_disk
*disk_list
)
3616 struct intel_disk
*idisk
= disk_list
;
3619 if (serialcmp(idisk
->disk
.serial
, serial
) == 0)
3621 idisk
= idisk
->next
;
3627 static int __prep_thunderdome(struct intel_super
**table
, int tbl_size
,
3628 struct intel_super
*super
,
3629 struct intel_disk
**disk_list
)
3631 struct imsm_disk
*d
= &super
->disks
->disk
;
3632 struct imsm_super
*mpb
= super
->anchor
;
3635 for (i
= 0; i
< tbl_size
; i
++) {
3636 struct imsm_super
*tbl_mpb
= table
[i
]->anchor
;
3637 struct imsm_disk
*tbl_d
= &table
[i
]->disks
->disk
;
3639 if (tbl_mpb
->family_num
== mpb
->family_num
) {
3640 if (tbl_mpb
->check_sum
== mpb
->check_sum
) {
3641 dprintf("%s: mpb from %d:%d matches %d:%d\n",
3642 __func__
, super
->disks
->major
,
3643 super
->disks
->minor
,
3644 table
[i
]->disks
->major
,
3645 table
[i
]->disks
->minor
);
3649 if (((is_configured(d
) && !is_configured(tbl_d
)) ||
3650 is_configured(d
) == is_configured(tbl_d
)) &&
3651 tbl_mpb
->generation_num
< mpb
->generation_num
) {
3652 /* current version of the mpb is a
3653 * better candidate than the one in
3654 * super_table, but copy over "cross
3655 * generational" status
3657 struct intel_disk
*idisk
;
3659 dprintf("%s: mpb from %d:%d replaces %d:%d\n",
3660 __func__
, super
->disks
->major
,
3661 super
->disks
->minor
,
3662 table
[i
]->disks
->major
,
3663 table
[i
]->disks
->minor
);
3665 idisk
= disk_list_get(tbl_d
->serial
, *disk_list
);
3666 if (idisk
&& is_failed(&idisk
->disk
))
3667 tbl_d
->status
|= FAILED_DISK
;
3670 struct intel_disk
*idisk
;
3671 struct imsm_disk
*disk
;
3673 /* tbl_mpb is more up to date, but copy
3674 * over cross generational status before
3677 disk
= __serial_to_disk(d
->serial
, mpb
, NULL
);
3678 if (disk
&& is_failed(disk
))
3679 d
->status
|= FAILED_DISK
;
3681 idisk
= disk_list_get(d
->serial
, *disk_list
);
3684 if (disk
&& is_configured(disk
))
3685 idisk
->disk
.status
|= CONFIGURED_DISK
;
3688 dprintf("%s: mpb from %d:%d prefer %d:%d\n",
3689 __func__
, super
->disks
->major
,
3690 super
->disks
->minor
,
3691 table
[i
]->disks
->major
,
3692 table
[i
]->disks
->minor
);
3700 table
[tbl_size
++] = super
;
3704 /* update/extend the merged list of imsm_disk records */
3705 for (j
= 0; j
< mpb
->num_disks
; j
++) {
3706 struct imsm_disk
*disk
= __get_imsm_disk(mpb
, j
);
3707 struct intel_disk
*idisk
;
3709 idisk
= disk_list_get(disk
->serial
, *disk_list
);
3711 idisk
->disk
.status
|= disk
->status
;
3712 if (is_configured(&idisk
->disk
) ||
3713 is_failed(&idisk
->disk
))
3714 idisk
->disk
.status
&= ~(SPARE_DISK
);
3716 idisk
= calloc(1, sizeof(*idisk
));
3719 idisk
->owner
= IMSM_UNKNOWN_OWNER
;
3720 idisk
->disk
= *disk
;
3721 idisk
->next
= *disk_list
;
3725 if (serialcmp(idisk
->disk
.serial
, d
->serial
) == 0)
3732 static struct intel_super
*
3733 validate_members(struct intel_super
*super
, struct intel_disk
*disk_list
,
3736 struct imsm_super
*mpb
= super
->anchor
;
3740 for (i
= 0; i
< mpb
->num_disks
; i
++) {
3741 struct imsm_disk
*disk
= __get_imsm_disk(mpb
, i
);
3742 struct intel_disk
*idisk
;
3744 idisk
= disk_list_get(disk
->serial
, disk_list
);
3746 if (idisk
->owner
== owner
||
3747 idisk
->owner
== IMSM_UNKNOWN_OWNER
)
3750 dprintf("%s: '%.16s' owner %d != %d\n",
3751 __func__
, disk
->serial
, idisk
->owner
,
3754 dprintf("%s: unknown disk %x [%d]: %.16s\n",
3755 __func__
, __le32_to_cpu(mpb
->family_num
), i
,
3761 if (ok_count
== mpb
->num_disks
)
3766 static void show_conflicts(__u32 family_num
, struct intel_super
*super_list
)
3768 struct intel_super
*s
;
3770 for (s
= super_list
; s
; s
= s
->next
) {
3771 if (family_num
!= s
->anchor
->family_num
)
3773 fprintf(stderr
, "Conflict, offlining family %#x on '%s'\n",
3774 __le32_to_cpu(family_num
), s
->disks
->devname
);
3778 static struct intel_super
*
3779 imsm_thunderdome(struct intel_super
**super_list
, int len
)
3781 struct intel_super
*super_table
[len
];
3782 struct intel_disk
*disk_list
= NULL
;
3783 struct intel_super
*champion
, *spare
;
3784 struct intel_super
*s
, **del
;
3789 memset(super_table
, 0, sizeof(super_table
));
3790 for (s
= *super_list
; s
; s
= s
->next
)
3791 tbl_size
= __prep_thunderdome(super_table
, tbl_size
, s
, &disk_list
);
3793 for (i
= 0; i
< tbl_size
; i
++) {
3794 struct imsm_disk
*d
;
3795 struct intel_disk
*idisk
;
3796 struct imsm_super
*mpb
= super_table
[i
]->anchor
;
3799 d
= &s
->disks
->disk
;
3801 /* 'd' must appear in merged disk list for its
3802 * configuration to be valid
3804 idisk
= disk_list_get(d
->serial
, disk_list
);
3805 if (idisk
&& idisk
->owner
== i
)
3806 s
= validate_members(s
, disk_list
, i
);
3811 dprintf("%s: marking family: %#x from %d:%d offline\n",
3812 __func__
, mpb
->family_num
,
3813 super_table
[i
]->disks
->major
,
3814 super_table
[i
]->disks
->minor
);
3818 /* This is where the mdadm implementation differs from the Windows
3819 * driver which has no strict concept of a container. We can only
3820 * assemble one family from a container, so when returning a prodigal
3821 * array member to this system the code will not be able to disambiguate
3822 * the container contents that should be assembled ("foreign" versus
3823 * "local"). It requires user intervention to set the orig_family_num
3824 * to a new value to establish a new container. The Windows driver in
3825 * this situation fixes up the volume name in place and manages the
3826 * foreign array as an independent entity.
3831 for (i
= 0; i
< tbl_size
; i
++) {
3832 struct intel_super
*tbl_ent
= super_table
[i
];
3838 if (tbl_ent
->anchor
->num_raid_devs
== 0) {
3843 if (s
&& !is_spare
) {
3844 show_conflicts(tbl_ent
->anchor
->family_num
, *super_list
);
3846 } else if (!s
&& !is_spare
)
3859 fprintf(stderr
, "Chose family %#x on '%s', "
3860 "assemble conflicts to new container with '--update=uuid'\n",
3861 __le32_to_cpu(s
->anchor
->family_num
), s
->disks
->devname
);
3863 /* collect all dl's onto 'champion', and update them to
3864 * champion's version of the status
3866 for (s
= *super_list
; s
; s
= s
->next
) {
3867 struct imsm_super
*mpb
= champion
->anchor
;
3868 struct dl
*dl
= s
->disks
;
3873 for (i
= 0; i
< mpb
->num_disks
; i
++) {
3874 struct imsm_disk
*disk
;
3876 disk
= __serial_to_disk(dl
->serial
, mpb
, &dl
->index
);
3879 /* only set index on disks that are a member of
3880 * a populated contianer, i.e. one with
3883 if (is_failed(&dl
->disk
))
3885 else if (is_spare(&dl
->disk
))
3891 if (i
>= mpb
->num_disks
) {
3892 struct intel_disk
*idisk
;
3894 idisk
= disk_list_get(dl
->serial
, disk_list
);
3895 if (idisk
&& is_spare(&idisk
->disk
) &&
3896 !is_failed(&idisk
->disk
) && !is_configured(&idisk
->disk
))
3904 dl
->next
= champion
->disks
;
3905 champion
->disks
= dl
;
3909 /* delete 'champion' from super_list */
3910 for (del
= super_list
; *del
; ) {
3911 if (*del
== champion
) {
3912 *del
= (*del
)->next
;
3915 del
= &(*del
)->next
;
3917 champion
->next
= NULL
;
3921 struct intel_disk
*idisk
= disk_list
;
3923 disk_list
= disk_list
->next
;
3930 static int load_super_imsm_all(struct supertype
*st
, int fd
, void **sbp
,
3934 struct intel_super
*super_list
= NULL
;
3935 struct intel_super
*super
= NULL
;
3936 int devnum
= fd2devnum(fd
);
3942 /* check if 'fd' an opened container */
3943 sra
= sysfs_read(fd
, 0, GET_LEVEL
|GET_VERSION
|GET_DEVS
|GET_STATE
);
3947 if (sra
->array
.major_version
!= -1 ||
3948 sra
->array
.minor_version
!= -2 ||
3949 strcmp(sra
->text_version
, "imsm") != 0) {
3954 for (sd
= sra
->devs
, i
= 0; sd
; sd
= sd
->next
, i
++) {
3955 struct intel_super
*s
= alloc_super();
3963 s
->next
= super_list
;
3967 sprintf(nm
, "%d:%d", sd
->disk
.major
, sd
->disk
.minor
);
3968 dfd
= dev_open(nm
, O_RDWR
);
3972 rv
= find_intel_hba_capability(dfd
, s
, devname
);
3973 /* no orom/efi or non-intel hba of the disk */
3977 err
= load_and_parse_mpb(dfd
, s
, NULL
, 1);
3979 /* retry the load if we might have raced against mdmon */
3980 if (err
== 3 && mdmon_running(devnum
))
3981 for (retry
= 0; retry
< 3; retry
++) {
3983 err
= load_and_parse_mpb(dfd
, s
, NULL
, 1);
3991 /* all mpbs enter, maybe one leaves */
3992 super
= imsm_thunderdome(&super_list
, i
);
3998 if (find_missing(super
) != 0) {
4004 /* load migration record */
4005 err
= load_imsm_migr_rec(super
, NULL
);
4011 /* Check migration compatibility */
4012 if (check_mpb_migr_compatibility(super
) != 0) {
4013 fprintf(stderr
, Name
": Unsupported migration detected");
4015 fprintf(stderr
, " on %s\n", devname
);
4017 fprintf(stderr
, " (IMSM).\n");
4026 while (super_list
) {
4027 struct intel_super
*s
= super_list
;
4029 super_list
= super_list
->next
;
4038 st
->container_dev
= devnum
;
4039 if (err
== 0 && st
->ss
== NULL
) {
4040 st
->ss
= &super_imsm
;
4041 st
->minor_version
= 0;
4042 st
->max_devs
= IMSM_MAX_DEVICES
;
4047 static int load_container_imsm(struct supertype
*st
, int fd
, char *devname
)
4049 return load_super_imsm_all(st
, fd
, &st
->sb
, devname
);
4053 static int load_super_imsm(struct supertype
*st
, int fd
, char *devname
)
4055 struct intel_super
*super
;
4058 if (test_partition(fd
))
4059 /* IMSM not allowed on partitions */
4062 free_super_imsm(st
);
4064 super
= alloc_super();
4067 Name
": malloc of %zu failed.\n",
4071 /* Load hba and capabilities if they exist.
4072 * But do not preclude loading metadata in case capabilities or hba are
4073 * non-compliant and ignore_hw_compat is set.
4075 rv
= find_intel_hba_capability(fd
, super
, devname
);
4076 /* no orom/efi or non-intel hba of the disk */
4077 if ((rv
!= 0) && (st
->ignore_hw_compat
== 0)) {
4080 Name
": No OROM/EFI properties for %s\n", devname
);
4084 rv
= load_and_parse_mpb(fd
, super
, devname
, 0);
4089 Name
": Failed to load all information "
4090 "sections on %s\n", devname
);
4096 if (st
->ss
== NULL
) {
4097 st
->ss
= &super_imsm
;
4098 st
->minor_version
= 0;
4099 st
->max_devs
= IMSM_MAX_DEVICES
;
4102 /* load migration record */
4103 if (load_imsm_migr_rec(super
, NULL
) == 0) {
4104 /* Check for unsupported migration features */
4105 if (check_mpb_migr_compatibility(super
) != 0) {
4107 Name
": Unsupported migration detected");
4109 fprintf(stderr
, " on %s\n", devname
);
4111 fprintf(stderr
, " (IMSM).\n");
4119 static __u16
info_to_blocks_per_strip(mdu_array_info_t
*info
)
4121 if (info
->level
== 1)
4123 return info
->chunk_size
>> 9;
4126 static __u32
info_to_num_data_stripes(mdu_array_info_t
*info
, int num_domains
)
4130 num_stripes
= (info
->size
* 2) / info_to_blocks_per_strip(info
);
4131 num_stripes
/= num_domains
;
4136 static __u32
info_to_blocks_per_member(mdu_array_info_t
*info
)
4138 if (info
->level
== 1)
4139 return info
->size
* 2;
4141 return (info
->size
* 2) & ~(info_to_blocks_per_strip(info
) - 1);
4144 static void imsm_update_version_info(struct intel_super
*super
)
4146 /* update the version and attributes */
4147 struct imsm_super
*mpb
= super
->anchor
;
4149 struct imsm_dev
*dev
;
4150 struct imsm_map
*map
;
4153 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
4154 dev
= get_imsm_dev(super
, i
);
4155 map
= get_imsm_map(dev
, 0);
4156 if (__le32_to_cpu(dev
->size_high
) > 0)
4157 mpb
->attributes
|= MPB_ATTRIB_2TB
;
4159 /* FIXME detect when an array spans a port multiplier */
4161 mpb
->attributes
|= MPB_ATTRIB_PM
;
4164 if (mpb
->num_raid_devs
> 1 ||
4165 mpb
->attributes
!= MPB_ATTRIB_CHECKSUM_VERIFY
) {
4166 version
= MPB_VERSION_ATTRIBS
;
4167 switch (get_imsm_raid_level(map
)) {
4168 case 0: mpb
->attributes
|= MPB_ATTRIB_RAID0
; break;
4169 case 1: mpb
->attributes
|= MPB_ATTRIB_RAID1
; break;
4170 case 10: mpb
->attributes
|= MPB_ATTRIB_RAID10
; break;
4171 case 5: mpb
->attributes
|= MPB_ATTRIB_RAID5
; break;
4174 if (map
->num_members
>= 5)
4175 version
= MPB_VERSION_5OR6_DISK_ARRAY
;
4176 else if (dev
->status
== DEV_CLONE_N_GO
)
4177 version
= MPB_VERSION_CNG
;
4178 else if (get_imsm_raid_level(map
) == 5)
4179 version
= MPB_VERSION_RAID5
;
4180 else if (map
->num_members
>= 3)
4181 version
= MPB_VERSION_3OR4_DISK_ARRAY
;
4182 else if (get_imsm_raid_level(map
) == 1)
4183 version
= MPB_VERSION_RAID1
;
4185 version
= MPB_VERSION_RAID0
;
4187 strcpy(((char *) mpb
->sig
) + strlen(MPB_SIGNATURE
), version
);
4191 static int check_name(struct intel_super
*super
, char *name
, int quiet
)
4193 struct imsm_super
*mpb
= super
->anchor
;
4194 char *reason
= NULL
;
4197 if (strlen(name
) > MAX_RAID_SERIAL_LEN
)
4198 reason
= "must be 16 characters or less";
4200 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
4201 struct imsm_dev
*dev
= get_imsm_dev(super
, i
);
4203 if (strncmp((char *) dev
->volume
, name
, MAX_RAID_SERIAL_LEN
) == 0) {
4204 reason
= "already exists";
4209 if (reason
&& !quiet
)
4210 fprintf(stderr
, Name
": imsm volume name %s\n", reason
);
4215 static int init_super_imsm_volume(struct supertype
*st
, mdu_array_info_t
*info
,
4216 unsigned long long size
, char *name
,
4217 char *homehost
, int *uuid
)
4219 /* We are creating a volume inside a pre-existing container.
4220 * so st->sb is already set.
4222 struct intel_super
*super
= st
->sb
;
4223 struct imsm_super
*mpb
= super
->anchor
;
4224 struct intel_dev
*dv
;
4225 struct imsm_dev
*dev
;
4226 struct imsm_vol
*vol
;
4227 struct imsm_map
*map
;
4228 int idx
= mpb
->num_raid_devs
;
4230 unsigned long long array_blocks
;
4231 size_t size_old
, size_new
;
4232 __u32 num_data_stripes
;
4234 if (super
->orom
&& mpb
->num_raid_devs
>= super
->orom
->vpa
) {
4235 fprintf(stderr
, Name
": This imsm-container already has the "
4236 "maximum of %d volumes\n", super
->orom
->vpa
);
4240 /* ensure the mpb is large enough for the new data */
4241 size_old
= __le32_to_cpu(mpb
->mpb_size
);
4242 size_new
= disks_to_mpb_size(info
->nr_disks
);
4243 if (size_new
> size_old
) {
4245 size_t size_round
= ROUND_UP(size_new
, 512);
4247 if (posix_memalign(&mpb_new
, 512, size_round
) != 0) {
4248 fprintf(stderr
, Name
": could not allocate new mpb\n");
4251 if (posix_memalign(&super
->migr_rec_buf
, 512, 512) != 0) {
4252 fprintf(stderr
, Name
4253 ": %s could not allocate migr_rec buffer\n",
4260 memcpy(mpb_new
, mpb
, size_old
);
4263 super
->anchor
= mpb_new
;
4264 mpb
->mpb_size
= __cpu_to_le32(size_new
);
4265 memset(mpb_new
+ size_old
, 0, size_round
- size_old
);
4267 super
->current_vol
= idx
;
4269 /* handle 'failed_disks' by either:
4270 * a) create dummy disk entries in the table if this the first
4271 * volume in the array. We add them here as this is the only
4272 * opportunity to add them. add_to_super_imsm_volume()
4273 * handles the non-failed disks and continues incrementing
4275 * b) validate that 'failed_disks' matches the current number
4276 * of missing disks if the container is populated
4278 if (super
->current_vol
== 0) {
4280 for (i
= 0; i
< info
->failed_disks
; i
++) {
4281 struct imsm_disk
*disk
;
4284 disk
= __get_imsm_disk(mpb
, i
);
4285 disk
->status
= CONFIGURED_DISK
| FAILED_DISK
;
4286 disk
->scsi_id
= __cpu_to_le32(~(__u32
)0);
4287 snprintf((char *) disk
->serial
, MAX_RAID_SERIAL_LEN
,
4290 find_missing(super
);
4295 for (d
= super
->missing
; d
; d
= d
->next
)
4297 if (info
->failed_disks
> missing
) {
4298 fprintf(stderr
, Name
": unable to add 'missing' disk to container\n");
4303 if (!check_name(super
, name
, 0))
4305 dv
= malloc(sizeof(*dv
));
4307 fprintf(stderr
, Name
": failed to allocate device list entry\n");
4310 dev
= calloc(1, sizeof(*dev
) + sizeof(__u32
) * (info
->raid_disks
- 1));
4313 fprintf(stderr
, Name
": could not allocate raid device\n");
4317 strncpy((char *) dev
->volume
, name
, MAX_RAID_SERIAL_LEN
);
4318 if (info
->level
== 1)
4319 array_blocks
= info_to_blocks_per_member(info
);
4321 array_blocks
= calc_array_size(info
->level
, info
->raid_disks
,
4322 info
->layout
, info
->chunk_size
,
4324 /* round array size down to closest MB */
4325 array_blocks
= (array_blocks
>> SECT_PER_MB_SHIFT
) << SECT_PER_MB_SHIFT
;
4327 dev
->size_low
= __cpu_to_le32((__u32
) array_blocks
);
4328 dev
->size_high
= __cpu_to_le32((__u32
) (array_blocks
>> 32));
4329 dev
->status
= (DEV_READ_COALESCING
| DEV_WRITE_COALESCING
);
4331 vol
->migr_state
= 0;
4332 set_migr_type(dev
, MIGR_INIT
);
4333 vol
->dirty
= !info
->state
;
4334 vol
->curr_migr_unit
= 0;
4335 map
= get_imsm_map(dev
, 0);
4336 map
->pba_of_lba0
= __cpu_to_le32(super
->create_offset
);
4337 map
->blocks_per_member
= __cpu_to_le32(info_to_blocks_per_member(info
));
4338 map
->blocks_per_strip
= __cpu_to_le16(info_to_blocks_per_strip(info
));
4339 map
->failed_disk_num
= ~0;
4340 if (info
->level
> 0)
4341 map
->map_state
= IMSM_T_STATE_UNINITIALIZED
;
4343 map
->map_state
= info
->failed_disks
? IMSM_T_STATE_FAILED
:
4344 IMSM_T_STATE_NORMAL
;
4347 if (info
->level
== 1 && info
->raid_disks
> 2) {
4350 fprintf(stderr
, Name
": imsm does not support more than 2 disks"
4351 "in a raid1 volume\n");
4355 map
->raid_level
= info
->level
;
4356 if (info
->level
== 10) {
4357 map
->raid_level
= 1;
4358 map
->num_domains
= info
->raid_disks
/ 2;
4359 } else if (info
->level
== 1)
4360 map
->num_domains
= info
->raid_disks
;
4362 map
->num_domains
= 1;
4364 num_data_stripes
= info_to_num_data_stripes(info
, map
->num_domains
);
4365 map
->num_data_stripes
= __cpu_to_le32(num_data_stripes
);
4367 map
->num_members
= info
->raid_disks
;
4368 for (i
= 0; i
< map
->num_members
; i
++) {
4369 /* initialized in add_to_super */
4370 set_imsm_ord_tbl_ent(map
, i
, IMSM_ORD_REBUILD
);
4372 mpb
->num_raid_devs
++;
4375 dv
->index
= super
->current_vol
;
4376 dv
->next
= super
->devlist
;
4377 super
->devlist
= dv
;
4379 imsm_update_version_info(super
);
4384 static int init_super_imsm(struct supertype
*st
, mdu_array_info_t
*info
,
4385 unsigned long long size
, char *name
,
4386 char *homehost
, int *uuid
)
4388 /* This is primarily called by Create when creating a new array.
4389 * We will then get add_to_super called for each component, and then
4390 * write_init_super called to write it out to each device.
4391 * For IMSM, Create can create on fresh devices or on a pre-existing
4393 * To create on a pre-existing array a different method will be called.
4394 * This one is just for fresh drives.
4396 struct intel_super
*super
;
4397 struct imsm_super
*mpb
;
4402 return init_super_imsm_volume(st
, info
, size
, name
, homehost
, uuid
);
4405 mpb_size
= disks_to_mpb_size(info
->nr_disks
);
4409 super
= alloc_super();
4410 if (super
&& posix_memalign(&super
->buf
, 512, mpb_size
) != 0) {
4415 fprintf(stderr
, Name
4416 ": %s could not allocate superblock\n", __func__
);
4419 if (posix_memalign(&super
->migr_rec_buf
, 512, 512) != 0) {
4420 fprintf(stderr
, Name
4421 ": %s could not allocate migr_rec buffer\n", __func__
);
4426 memset(super
->buf
, 0, mpb_size
);
4428 mpb
->mpb_size
= __cpu_to_le32(mpb_size
);
4432 /* zeroing superblock */
4436 mpb
->attributes
= MPB_ATTRIB_CHECKSUM_VERIFY
;
4438 version
= (char *) mpb
->sig
;
4439 strcpy(version
, MPB_SIGNATURE
);
4440 version
+= strlen(MPB_SIGNATURE
);
4441 strcpy(version
, MPB_VERSION_RAID0
);
4447 static int add_to_super_imsm_volume(struct supertype
*st
, mdu_disk_info_t
*dk
,
4448 int fd
, char *devname
)
4450 struct intel_super
*super
= st
->sb
;
4451 struct imsm_super
*mpb
= super
->anchor
;
4452 struct imsm_disk
*_disk
;
4453 struct imsm_dev
*dev
;
4454 struct imsm_map
*map
;
4458 dev
= get_imsm_dev(super
, super
->current_vol
);
4459 map
= get_imsm_map(dev
, 0);
4461 if (! (dk
->state
& (1<<MD_DISK_SYNC
))) {
4462 fprintf(stderr
, Name
": %s: Cannot add spare devices to IMSM volume\n",
4468 /* we're doing autolayout so grab the pre-marked (in
4469 * validate_geometry) raid_disk
4471 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
4472 if (dl
->raiddisk
== dk
->raid_disk
)
4475 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
4476 if (dl
->major
== dk
->major
&&
4477 dl
->minor
== dk
->minor
)
4482 fprintf(stderr
, Name
": %s is not a member of the same container\n", devname
);
4486 /* add a pristine spare to the metadata */
4487 if (dl
->index
< 0) {
4488 dl
->index
= super
->anchor
->num_disks
;
4489 super
->anchor
->num_disks
++;
4491 /* Check the device has not already been added */
4492 slot
= get_imsm_disk_slot(map
, dl
->index
);
4494 (get_imsm_ord_tbl_ent(dev
, slot
, -1) & IMSM_ORD_REBUILD
) == 0) {
4495 fprintf(stderr
, Name
": %s has been included in this array twice\n",
4499 set_imsm_ord_tbl_ent(map
, dk
->raid_disk
, dl
->index
);
4500 dl
->disk
.status
= CONFIGURED_DISK
;
4502 /* update size of 'missing' disks to be at least as large as the
4503 * largest acitve member (we only have dummy missing disks when
4504 * creating the first volume)
4506 if (super
->current_vol
== 0) {
4507 for (df
= super
->missing
; df
; df
= df
->next
) {
4508 if (dl
->disk
.total_blocks
> df
->disk
.total_blocks
)
4509 df
->disk
.total_blocks
= dl
->disk
.total_blocks
;
4510 _disk
= __get_imsm_disk(mpb
, df
->index
);
4515 /* refresh unset/failed slots to point to valid 'missing' entries */
4516 for (df
= super
->missing
; df
; df
= df
->next
)
4517 for (slot
= 0; slot
< mpb
->num_disks
; slot
++) {
4518 __u32 ord
= get_imsm_ord_tbl_ent(dev
, slot
, -1);
4520 if ((ord
& IMSM_ORD_REBUILD
) == 0)
4522 set_imsm_ord_tbl_ent(map
, slot
, df
->index
| IMSM_ORD_REBUILD
);
4523 if (is_gen_migration(dev
)) {
4524 struct imsm_map
*map2
= get_imsm_map(dev
, 1);
4525 int slot2
= get_imsm_disk_slot(map2
, df
->index
);
4526 if ((slot2
< map2
->num_members
) &&
4528 __u32 ord2
= get_imsm_ord_tbl_ent(dev
,
4531 if ((unsigned)df
->index
==
4533 set_imsm_ord_tbl_ent(map2
,
4539 dprintf("set slot:%d to missing disk:%d\n", slot
, df
->index
);
4543 /* if we are creating the first raid device update the family number */
4544 if (super
->current_vol
== 0) {
4546 struct imsm_dev
*_dev
= __get_imsm_dev(mpb
, 0);
4548 _disk
= __get_imsm_disk(mpb
, dl
->index
);
4549 if (!_dev
|| !_disk
) {
4550 fprintf(stderr
, Name
": BUG mpb setup error\n");
4556 sum
+= __gen_imsm_checksum(mpb
);
4557 mpb
->family_num
= __cpu_to_le32(sum
);
4558 mpb
->orig_family_num
= mpb
->family_num
;
4560 super
->current_disk
= dl
;
4565 * Function marks disk as spare and restores disk serial
4566 * in case it was previously marked as failed by takeover operation
4568 * -1 : critical error
4569 * 0 : disk is marked as spare but serial is not set
4572 int mark_spare(struct dl
*disk
)
4574 __u8 serial
[MAX_RAID_SERIAL_LEN
];
4581 if (!imsm_read_serial(disk
->fd
, NULL
, serial
)) {
4582 /* Restore disk serial number, because takeover marks disk
4583 * as failed and adds to serial ':0' before it becomes
4586 serialcpy(disk
->serial
, serial
);
4587 serialcpy(disk
->disk
.serial
, serial
);
4590 disk
->disk
.status
= SPARE_DISK
;
4596 static int add_to_super_imsm(struct supertype
*st
, mdu_disk_info_t
*dk
,
4597 int fd
, char *devname
)
4599 struct intel_super
*super
= st
->sb
;
4601 unsigned long long size
;
4606 /* If we are on an RAID enabled platform check that the disk is
4607 * attached to the raid controller.
4608 * We do not need to test disks attachment for container based additions,
4609 * they shall be already tested when container was created/assembled.
4611 rv
= find_intel_hba_capability(fd
, super
, devname
);
4612 /* no orom/efi or non-intel hba of the disk */
4614 dprintf("capability: %p fd: %d ret: %d\n",
4615 super
->orom
, fd
, rv
);
4619 if (super
->current_vol
>= 0)
4620 return add_to_super_imsm_volume(st
, dk
, fd
, devname
);
4623 dd
= malloc(sizeof(*dd
));
4626 Name
": malloc failed %s:%d.\n", __func__
, __LINE__
);
4629 memset(dd
, 0, sizeof(*dd
));
4630 dd
->major
= major(stb
.st_rdev
);
4631 dd
->minor
= minor(stb
.st_rdev
);
4632 dd
->devname
= devname
? strdup(devname
) : NULL
;
4635 dd
->action
= DISK_ADD
;
4636 rv
= imsm_read_serial(fd
, devname
, dd
->serial
);
4639 Name
": failed to retrieve scsi serial, aborting\n");
4644 get_dev_size(fd
, NULL
, &size
);
4646 serialcpy(dd
->disk
.serial
, dd
->serial
);
4647 dd
->disk
.total_blocks
= __cpu_to_le32(size
);
4649 if (sysfs_disk_to_scsi_id(fd
, &id
) == 0)
4650 dd
->disk
.scsi_id
= __cpu_to_le32(id
);
4652 dd
->disk
.scsi_id
= __cpu_to_le32(0);
4654 if (st
->update_tail
) {
4655 dd
->next
= super
->disk_mgmt_list
;
4656 super
->disk_mgmt_list
= dd
;
4658 dd
->next
= super
->disks
;
4660 super
->updates_pending
++;
4667 static int remove_from_super_imsm(struct supertype
*st
, mdu_disk_info_t
*dk
)
4669 struct intel_super
*super
= st
->sb
;
4672 /* remove from super works only in mdmon - for communication
4673 * manager - monitor. Check if communication memory buffer
4676 if (!st
->update_tail
) {
4678 Name
": %s shall be used in mdmon context only"
4679 "(line %d).\n", __func__
, __LINE__
);
4682 dd
= malloc(sizeof(*dd
));
4685 Name
": malloc failed %s:%d.\n", __func__
, __LINE__
);
4688 memset(dd
, 0, sizeof(*dd
));
4689 dd
->major
= dk
->major
;
4690 dd
->minor
= dk
->minor
;
4693 dd
->action
= DISK_REMOVE
;
4695 dd
->next
= super
->disk_mgmt_list
;
4696 super
->disk_mgmt_list
= dd
;
4702 static int store_imsm_mpb(int fd
, struct imsm_super
*mpb
);
4706 struct imsm_super anchor
;
4707 } spare_record
__attribute__ ((aligned(512)));
4709 /* spare records have their own family number and do not have any defined raid
4712 static int write_super_imsm_spares(struct intel_super
*super
, int doclose
)
4714 struct imsm_super
*mpb
= super
->anchor
;
4715 struct imsm_super
*spare
= &spare_record
.anchor
;
4719 spare
->mpb_size
= __cpu_to_le32(sizeof(struct imsm_super
)),
4720 spare
->generation_num
= __cpu_to_le32(1UL),
4721 spare
->attributes
= MPB_ATTRIB_CHECKSUM_VERIFY
;
4722 spare
->num_disks
= 1,
4723 spare
->num_raid_devs
= 0,
4724 spare
->cache_size
= mpb
->cache_size
,
4725 spare
->pwr_cycle_count
= __cpu_to_le32(1),
4727 snprintf((char *) spare
->sig
, MAX_SIGNATURE_LENGTH
,
4728 MPB_SIGNATURE MPB_VERSION_RAID0
);
4730 for (d
= super
->disks
; d
; d
= d
->next
) {
4734 spare
->disk
[0] = d
->disk
;
4735 sum
= __gen_imsm_checksum(spare
);
4736 spare
->family_num
= __cpu_to_le32(sum
);
4737 spare
->orig_family_num
= 0;
4738 sum
= __gen_imsm_checksum(spare
);
4739 spare
->check_sum
= __cpu_to_le32(sum
);
4741 if (store_imsm_mpb(d
->fd
, spare
)) {
4742 fprintf(stderr
, "%s: failed for device %d:%d %s\n",
4743 __func__
, d
->major
, d
->minor
, strerror(errno
));
4755 static int write_super_imsm(struct supertype
*st
, int doclose
)
4757 struct intel_super
*super
= st
->sb
;
4758 struct imsm_super
*mpb
= super
->anchor
;
4764 __u32 mpb_size
= sizeof(struct imsm_super
) - sizeof(struct imsm_disk
);
4766 int clear_migration_record
= 1;
4768 /* 'generation' is incremented everytime the metadata is written */
4769 generation
= __le32_to_cpu(mpb
->generation_num
);
4771 mpb
->generation_num
= __cpu_to_le32(generation
);
4773 /* fix up cases where previous mdadm releases failed to set
4776 if (mpb
->orig_family_num
== 0)
4777 mpb
->orig_family_num
= mpb
->family_num
;
4779 for (d
= super
->disks
; d
; d
= d
->next
) {
4783 mpb
->disk
[d
->index
] = d
->disk
;
4787 for (d
= super
->missing
; d
; d
= d
->next
) {
4788 mpb
->disk
[d
->index
] = d
->disk
;
4791 mpb
->num_disks
= num_disks
;
4792 mpb_size
+= sizeof(struct imsm_disk
) * mpb
->num_disks
;
4794 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
4795 struct imsm_dev
*dev
= __get_imsm_dev(mpb
, i
);
4796 struct imsm_dev
*dev2
= get_imsm_dev(super
, i
);
4798 imsm_copy_dev(dev
, dev2
);
4799 mpb_size
+= sizeof_imsm_dev(dev
, 0);
4801 if (is_gen_migration(dev2
))
4802 clear_migration_record
= 0;
4804 mpb_size
+= __le32_to_cpu(mpb
->bbm_log_size
);
4805 mpb
->mpb_size
= __cpu_to_le32(mpb_size
);
4807 /* recalculate checksum */
4808 sum
= __gen_imsm_checksum(mpb
);
4809 mpb
->check_sum
= __cpu_to_le32(sum
);
4811 if (clear_migration_record
)
4812 memset(super
->migr_rec_buf
, 0, 512);
4814 /* write the mpb for disks that compose raid devices */
4815 for (d
= super
->disks
; d
; d
= d
->next
) {
4816 if (d
->index
< 0 || is_failed(&d
->disk
))
4818 if (store_imsm_mpb(d
->fd
, mpb
))
4819 fprintf(stderr
, "%s: failed for device %d:%d %s\n",
4820 __func__
, d
->major
, d
->minor
, strerror(errno
));
4821 if (clear_migration_record
) {
4822 unsigned long long dsize
;
4824 get_dev_size(d
->fd
, NULL
, &dsize
);
4825 if (lseek64(d
->fd
, dsize
- 512, SEEK_SET
) >= 0) {
4826 if (write(d
->fd
, super
->migr_rec_buf
, 512) != 512)
4827 perror("Write migr_rec failed");
4837 return write_super_imsm_spares(super
, doclose
);
4843 static int create_array(struct supertype
*st
, int dev_idx
)
4846 struct imsm_update_create_array
*u
;
4847 struct intel_super
*super
= st
->sb
;
4848 struct imsm_dev
*dev
= get_imsm_dev(super
, dev_idx
);
4849 struct imsm_map
*map
= get_imsm_map(dev
, 0);
4850 struct disk_info
*inf
;
4851 struct imsm_disk
*disk
;
4854 len
= sizeof(*u
) - sizeof(*dev
) + sizeof_imsm_dev(dev
, 0) +
4855 sizeof(*inf
) * map
->num_members
;
4858 fprintf(stderr
, "%s: failed to allocate update buffer\n",
4863 u
->type
= update_create_array
;
4864 u
->dev_idx
= dev_idx
;
4865 imsm_copy_dev(&u
->dev
, dev
);
4866 inf
= get_disk_info(u
);
4867 for (i
= 0; i
< map
->num_members
; i
++) {
4868 int idx
= get_imsm_disk_idx(dev
, i
, -1);
4870 disk
= get_imsm_disk(super
, idx
);
4871 serialcpy(inf
[i
].serial
, disk
->serial
);
4873 append_metadata_update(st
, u
, len
);
4878 static int mgmt_disk(struct supertype
*st
)
4880 struct intel_super
*super
= st
->sb
;
4882 struct imsm_update_add_remove_disk
*u
;
4884 if (!super
->disk_mgmt_list
)
4890 fprintf(stderr
, "%s: failed to allocate update buffer\n",
4895 u
->type
= update_add_remove_disk
;
4896 append_metadata_update(st
, u
, len
);
4901 static int write_init_super_imsm(struct supertype
*st
)
4903 struct intel_super
*super
= st
->sb
;
4904 int current_vol
= super
->current_vol
;
4906 /* we are done with current_vol reset it to point st at the container */
4907 super
->current_vol
= -1;
4909 if (st
->update_tail
) {
4910 /* queue the recently created array / added disk
4911 * as a metadata update */
4914 /* determine if we are creating a volume or adding a disk */
4915 if (current_vol
< 0) {
4916 /* in the mgmt (add/remove) disk case we are running
4917 * in mdmon context, so don't close fd's
4919 return mgmt_disk(st
);
4921 rv
= create_array(st
, current_vol
);
4926 for (d
= super
->disks
; d
; d
= d
->next
)
4927 Kill(d
->devname
, NULL
, 0, 1, 1);
4928 return write_super_imsm(st
, 1);
4933 static int store_super_imsm(struct supertype
*st
, int fd
)
4935 struct intel_super
*super
= st
->sb
;
4936 struct imsm_super
*mpb
= super
? super
->anchor
: NULL
;
4942 return store_imsm_mpb(fd
, mpb
);
4948 static int imsm_bbm_log_size(struct imsm_super
*mpb
)
4950 return __le32_to_cpu(mpb
->bbm_log_size
);
4954 static int validate_geometry_imsm_container(struct supertype
*st
, int level
,
4955 int layout
, int raiddisks
, int chunk
,
4956 unsigned long long size
, char *dev
,
4957 unsigned long long *freesize
,
4961 unsigned long long ldsize
;
4962 struct intel_super
*super
=NULL
;
4965 if (level
!= LEVEL_CONTAINER
)
4970 fd
= open(dev
, O_RDONLY
|O_EXCL
, 0);
4973 fprintf(stderr
, Name
": imsm: Cannot open %s: %s\n",
4974 dev
, strerror(errno
));
4977 if (!get_dev_size(fd
, dev
, &ldsize
)) {
4982 /* capabilities retrieve could be possible
4983 * note that there is no fd for the disks in array.
4985 super
= alloc_super();
4988 Name
": malloc of %zu failed.\n",
4994 rv
= find_intel_hba_capability(fd
, super
, verbose
? dev
: NULL
);
4998 fd2devname(fd
, str
);
4999 dprintf("validate_geometry_imsm_container: fd: %d %s orom: %p rv: %d raiddisk: %d\n",
5000 fd
, str
, super
->orom
, rv
, raiddisks
);
5002 /* no orom/efi or non-intel hba of the disk */
5008 if (super
->orom
&& raiddisks
> super
->orom
->tds
) {
5010 fprintf(stderr
, Name
": %d exceeds maximum number of"
5011 " platform supported disks: %d\n",
5012 raiddisks
, super
->orom
->tds
);
5018 *freesize
= avail_size_imsm(st
, ldsize
>> 9);
5024 static unsigned long long find_size(struct extent
*e
, int *idx
, int num_extents
)
5026 const unsigned long long base_start
= e
[*idx
].start
;
5027 unsigned long long end
= base_start
+ e
[*idx
].size
;
5030 if (base_start
== end
)
5034 for (i
= *idx
; i
< num_extents
; i
++) {
5035 /* extend overlapping extents */
5036 if (e
[i
].start
>= base_start
&&
5037 e
[i
].start
<= end
) {
5040 if (e
[i
].start
+ e
[i
].size
> end
)
5041 end
= e
[i
].start
+ e
[i
].size
;
5042 } else if (e
[i
].start
> end
) {
5048 return end
- base_start
;
5051 static unsigned long long merge_extents(struct intel_super
*super
, int sum_extents
)
5053 /* build a composite disk with all known extents and generate a new
5054 * 'maxsize' given the "all disks in an array must share a common start
5055 * offset" constraint
5057 struct extent
*e
= calloc(sum_extents
, sizeof(*e
));
5061 unsigned long long pos
;
5062 unsigned long long start
= 0;
5063 unsigned long long maxsize
;
5064 unsigned long reserve
;
5069 /* coalesce and sort all extents. also, check to see if we need to
5070 * reserve space between member arrays
5073 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
5076 for (i
= 0; i
< dl
->extent_cnt
; i
++)
5079 qsort(e
, sum_extents
, sizeof(*e
), cmp_extent
);
5084 while (i
< sum_extents
) {
5085 e
[j
].start
= e
[i
].start
;
5086 e
[j
].size
= find_size(e
, &i
, sum_extents
);
5088 if (e
[j
-1].size
== 0)
5097 unsigned long long esize
;
5099 esize
= e
[i
].start
- pos
;
5100 if (esize
>= maxsize
) {
5105 pos
= e
[i
].start
+ e
[i
].size
;
5107 } while (e
[i
-1].size
);
5113 /* FIXME assumes volume at offset 0 is the first volume in a
5116 if (start_extent
> 0)
5117 reserve
= IMSM_RESERVED_SECTORS
; /* gap between raid regions */
5121 if (maxsize
< reserve
)
5124 super
->create_offset
= ~((__u32
) 0);
5125 if (start
+ reserve
> super
->create_offset
)
5126 return 0; /* start overflows create_offset */
5127 super
->create_offset
= start
+ reserve
;
5129 return maxsize
- reserve
;
5132 static int is_raid_level_supported(const struct imsm_orom
*orom
, int level
, int raiddisks
)
5134 if (level
< 0 || level
== 6 || level
== 4)
5137 /* if we have an orom prevent invalid raid levels */
5140 case 0: return imsm_orom_has_raid0(orom
);
5143 return imsm_orom_has_raid1e(orom
);
5144 return imsm_orom_has_raid1(orom
) && raiddisks
== 2;
5145 case 10: return imsm_orom_has_raid10(orom
) && raiddisks
== 4;
5146 case 5: return imsm_orom_has_raid5(orom
) && raiddisks
> 2;
5149 return 1; /* not on an Intel RAID platform so anything goes */
5154 static int imsm_default_chunk(const struct imsm_orom
*orom
)
5156 /* up to 512 if the plaform supports it, otherwise the platform max.
5157 * 128 if no platform detected
5159 int fs
= max(7, orom
? fls(orom
->sss
) : 0);
5161 return min(512, (1 << fs
));
5164 #define pr_vrb(fmt, arg...) (void) (verbose && fprintf(stderr, Name fmt, ##arg))
5166 validate_geometry_imsm_orom(struct intel_super
*super
, int level
, int layout
,
5167 int raiddisks
, int *chunk
, int verbose
)
5169 /* check/set platform and metadata limits/defaults */
5170 if (super
->orom
&& raiddisks
> super
->orom
->dpa
) {
5171 pr_vrb(": platform supports a maximum of %d disks per array\n",
5176 /* capabilities of OROM tested - copied from validate_geometry_imsm_volume */
5177 if (!is_raid_level_supported(super
->orom
, level
, raiddisks
)) {
5178 pr_vrb(": platform does not support raid%d with %d disk%s\n",
5179 level
, raiddisks
, raiddisks
> 1 ? "s" : "");
5183 if (chunk
&& (*chunk
== 0 || *chunk
== UnSet
))
5184 *chunk
= imsm_default_chunk(super
->orom
);
5186 if (super
->orom
&& chunk
&& !imsm_orom_has_chunk(super
->orom
, *chunk
)) {
5187 pr_vrb(": platform does not support a chunk size of: "
5192 if (layout
!= imsm_level_to_layout(level
)) {
5194 pr_vrb(": imsm raid 5 only supports the left-asymmetric layout\n");
5195 else if (level
== 10)
5196 pr_vrb(": imsm raid 10 only supports the n2 layout\n");
5198 pr_vrb(": imsm unknown layout %#x for this raid level %d\n",
5205 /* validate_geometry_imsm_volume - lifted from validate_geometry_ddf_bvd
5206 * FIX ME add ahci details
5208 static int validate_geometry_imsm_volume(struct supertype
*st
, int level
,
5209 int layout
, int raiddisks
, int *chunk
,
5210 unsigned long long size
, char *dev
,
5211 unsigned long long *freesize
,
5215 struct intel_super
*super
= st
->sb
;
5216 struct imsm_super
*mpb
;
5218 unsigned long long pos
= 0;
5219 unsigned long long maxsize
;
5223 /* We must have the container info already read in. */
5227 mpb
= super
->anchor
;
5229 if (mpb
->num_raid_devs
> 0 && mpb
->num_disks
!= raiddisks
) {
5230 fprintf(stderr
, Name
": the option-rom requires all "
5231 "member disks to be a member of all volumes.\n");
5235 if (!validate_geometry_imsm_orom(super
, level
, layout
, raiddisks
, chunk
, verbose
)) {
5236 fprintf(stderr
, Name
": RAID gemetry validation failed. "
5237 "Cannot proceed with the action(s).\n");
5241 /* General test: make sure there is space for
5242 * 'raiddisks' device extents of size 'size' at a given
5245 unsigned long long minsize
= size
;
5246 unsigned long long start_offset
= MaxSector
;
5249 minsize
= MPB_SECTOR_CNT
+ IMSM_RESERVED_SECTORS
;
5250 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
5255 e
= get_extents(super
, dl
);
5258 unsigned long long esize
;
5259 esize
= e
[i
].start
- pos
;
5260 if (esize
>= minsize
)
5262 if (found
&& start_offset
== MaxSector
) {
5265 } else if (found
&& pos
!= start_offset
) {
5269 pos
= e
[i
].start
+ e
[i
].size
;
5271 } while (e
[i
-1].size
);
5276 if (dcnt
< raiddisks
) {
5278 fprintf(stderr
, Name
": imsm: Not enough "
5279 "devices with space for this array "
5287 /* This device must be a member of the set */
5288 if (stat(dev
, &stb
) < 0)
5290 if ((S_IFMT
& stb
.st_mode
) != S_IFBLK
)
5292 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
5293 if (dl
->major
== (int)major(stb
.st_rdev
) &&
5294 dl
->minor
== (int)minor(stb
.st_rdev
))
5299 fprintf(stderr
, Name
": %s is not in the "
5300 "same imsm set\n", dev
);
5302 } else if (super
->orom
&& dl
->index
< 0 && mpb
->num_raid_devs
) {
5303 /* If a volume is present then the current creation attempt
5304 * cannot incorporate new spares because the orom may not
5305 * understand this configuration (all member disks must be
5306 * members of each array in the container).
5308 fprintf(stderr
, Name
": %s is a spare and a volume"
5309 " is already defined for this container\n", dev
);
5310 fprintf(stderr
, Name
": The option-rom requires all member"
5311 " disks to be a member of all volumes\n");
5315 /* retrieve the largest free space block */
5316 e
= get_extents(super
, dl
);
5321 unsigned long long esize
;
5323 esize
= e
[i
].start
- pos
;
5324 if (esize
>= maxsize
)
5326 pos
= e
[i
].start
+ e
[i
].size
;
5328 } while (e
[i
-1].size
);
5333 fprintf(stderr
, Name
": unable to determine free space for: %s\n",
5337 if (maxsize
< size
) {
5339 fprintf(stderr
, Name
": %s not enough space (%llu < %llu)\n",
5340 dev
, maxsize
, size
);
5344 /* count total number of extents for merge */
5346 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
5348 i
+= dl
->extent_cnt
;
5350 maxsize
= merge_extents(super
, i
);
5352 if (!check_env("IMSM_NO_PLATFORM") &&
5353 mpb
->num_raid_devs
> 0 && size
&& size
!= maxsize
) {
5354 fprintf(stderr
, Name
": attempting to create a second "
5355 "volume with size less then remaining space. "
5360 if (maxsize
< size
|| maxsize
== 0) {
5362 fprintf(stderr
, Name
": not enough space after merge (%llu < %llu)\n",
5367 *freesize
= maxsize
;
5372 static int reserve_space(struct supertype
*st
, int raiddisks
,
5373 unsigned long long size
, int chunk
,
5374 unsigned long long *freesize
)
5376 struct intel_super
*super
= st
->sb
;
5377 struct imsm_super
*mpb
= super
->anchor
;
5382 unsigned long long maxsize
;
5383 unsigned long long minsize
;
5387 /* find the largest common start free region of the possible disks */
5391 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
5397 /* don't activate new spares if we are orom constrained
5398 * and there is already a volume active in the container
5400 if (super
->orom
&& dl
->index
< 0 && mpb
->num_raid_devs
)
5403 e
= get_extents(super
, dl
);
5406 for (i
= 1; e
[i
-1].size
; i
++)
5414 maxsize
= merge_extents(super
, extent_cnt
);
5418 minsize
= chunk
* 2;
5420 if (cnt
< raiddisks
||
5421 (super
->orom
&& used
&& used
!= raiddisks
) ||
5422 maxsize
< minsize
||
5424 fprintf(stderr
, Name
": not enough devices with space to create array.\n");
5425 return 0; /* No enough free spaces large enough */
5437 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
5439 dl
->raiddisk
= cnt
++;
5446 static int validate_geometry_imsm(struct supertype
*st
, int level
, int layout
,
5447 int raiddisks
, int *chunk
, unsigned long long size
,
5448 char *dev
, unsigned long long *freesize
,
5456 * if given unused devices create a container
5457 * if given given devices in a container create a member volume
5459 if (level
== LEVEL_CONTAINER
) {
5460 /* Must be a fresh device to add to a container */
5461 return validate_geometry_imsm_container(st
, level
, layout
,
5463 chunk
?*chunk
:0, size
,
5470 if (!validate_geometry_imsm_orom(st
->sb
, level
, layout
,
5474 /* we are being asked to automatically layout a
5475 * new volume based on the current contents of
5476 * the container. If the the parameters can be
5477 * satisfied reserve_space will record the disks,
5478 * start offset, and size of the volume to be
5479 * created. add_to_super and getinfo_super
5480 * detect when autolayout is in progress.
5483 return reserve_space(st
, raiddisks
, size
,
5484 chunk
?*chunk
:0, freesize
);
5489 /* creating in a given container */
5490 return validate_geometry_imsm_volume(st
, level
, layout
,
5491 raiddisks
, chunk
, size
,
5492 dev
, freesize
, verbose
);
5495 /* This device needs to be a device in an 'imsm' container */
5496 fd
= open(dev
, O_RDONLY
|O_EXCL
, 0);
5500 Name
": Cannot create this array on device %s\n",
5505 if (errno
!= EBUSY
|| (fd
= open(dev
, O_RDONLY
, 0)) < 0) {
5507 fprintf(stderr
, Name
": Cannot open %s: %s\n",
5508 dev
, strerror(errno
));
5511 /* Well, it is in use by someone, maybe an 'imsm' container. */
5512 cfd
= open_container(fd
);
5516 fprintf(stderr
, Name
": Cannot use %s: It is busy\n",
5520 sra
= sysfs_read(cfd
, 0, GET_VERSION
);
5521 if (sra
&& sra
->array
.major_version
== -1 &&
5522 strcmp(sra
->text_version
, "imsm") == 0)
5526 /* This is a member of a imsm container. Load the container
5527 * and try to create a volume
5529 struct intel_super
*super
;
5531 if (load_super_imsm_all(st
, cfd
, (void **) &super
, NULL
) == 0) {
5533 st
->container_dev
= fd2devnum(cfd
);
5535 return validate_geometry_imsm_volume(st
, level
, layout
,
5544 fprintf(stderr
, Name
": failed container membership check\n");
5550 static void default_geometry_imsm(struct supertype
*st
, int *level
, int *layout
, int *chunk
)
5552 struct intel_super
*super
= st
->sb
;
5554 if (level
&& *level
== UnSet
)
5555 *level
= LEVEL_CONTAINER
;
5557 if (level
&& layout
&& *layout
== UnSet
)
5558 *layout
= imsm_level_to_layout(*level
);
5560 if (chunk
&& (*chunk
== UnSet
|| *chunk
== 0))
5561 *chunk
= imsm_default_chunk(super
->orom
);
5564 static void handle_missing(struct intel_super
*super
, struct imsm_dev
*dev
);
5566 static int kill_subarray_imsm(struct supertype
*st
)
5568 /* remove the subarray currently referenced by ->current_vol */
5570 struct intel_dev
**dp
;
5571 struct intel_super
*super
= st
->sb
;
5572 __u8 current_vol
= super
->current_vol
;
5573 struct imsm_super
*mpb
= super
->anchor
;
5575 if (super
->current_vol
< 0)
5577 super
->current_vol
= -1; /* invalidate subarray cursor */
5579 /* block deletions that would change the uuid of active subarrays
5581 * FIXME when immutable ids are available, but note that we'll
5582 * also need to fixup the invalidated/active subarray indexes in
5585 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
5588 if (i
< current_vol
)
5590 sprintf(subarray
, "%u", i
);
5591 if (is_subarray_active(subarray
, st
->devname
)) {
5593 Name
": deleting subarray-%d would change the UUID of active subarray-%d, aborting\n",
5600 if (st
->update_tail
) {
5601 struct imsm_update_kill_array
*u
= malloc(sizeof(*u
));
5605 u
->type
= update_kill_array
;
5606 u
->dev_idx
= current_vol
;
5607 append_metadata_update(st
, u
, sizeof(*u
));
5612 for (dp
= &super
->devlist
; *dp
;)
5613 if ((*dp
)->index
== current_vol
) {
5616 handle_missing(super
, (*dp
)->dev
);
5617 if ((*dp
)->index
> current_vol
)
5622 /* no more raid devices, all active components are now spares,
5623 * but of course failed are still failed
5625 if (--mpb
->num_raid_devs
== 0) {
5628 for (d
= super
->disks
; d
; d
= d
->next
)
5633 super
->updates_pending
++;
5638 static int update_subarray_imsm(struct supertype
*st
, char *subarray
,
5639 char *update
, struct mddev_ident
*ident
)
5641 /* update the subarray currently referenced by ->current_vol */
5642 struct intel_super
*super
= st
->sb
;
5643 struct imsm_super
*mpb
= super
->anchor
;
5645 if (strcmp(update
, "name") == 0) {
5646 char *name
= ident
->name
;
5650 if (is_subarray_active(subarray
, st
->devname
)) {
5652 Name
": Unable to update name of active subarray\n");
5656 if (!check_name(super
, name
, 0))
5659 vol
= strtoul(subarray
, &ep
, 10);
5660 if (*ep
!= '\0' || vol
>= super
->anchor
->num_raid_devs
)
5663 if (st
->update_tail
) {
5664 struct imsm_update_rename_array
*u
= malloc(sizeof(*u
));
5668 u
->type
= update_rename_array
;
5670 snprintf((char *) u
->name
, MAX_RAID_SERIAL_LEN
, "%s", name
);
5671 append_metadata_update(st
, u
, sizeof(*u
));
5673 struct imsm_dev
*dev
;
5676 dev
= get_imsm_dev(super
, vol
);
5677 snprintf((char *) dev
->volume
, MAX_RAID_SERIAL_LEN
, "%s", name
);
5678 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
5679 dev
= get_imsm_dev(super
, i
);
5680 handle_missing(super
, dev
);
5682 super
->updates_pending
++;
5689 #endif /* MDASSEMBLE */
5691 static int is_gen_migration(struct imsm_dev
*dev
)
5696 if (!dev
->vol
.migr_state
)
5699 if (migr_type(dev
) == MIGR_GEN_MIGR
)
5705 static int is_rebuilding(struct imsm_dev
*dev
)
5707 struct imsm_map
*migr_map
;
5709 if (!dev
->vol
.migr_state
)
5712 if (migr_type(dev
) != MIGR_REBUILD
)
5715 migr_map
= get_imsm_map(dev
, 1);
5717 if (migr_map
->map_state
== IMSM_T_STATE_DEGRADED
)
5723 static int is_initializing(struct imsm_dev
*dev
)
5725 struct imsm_map
*migr_map
;
5727 if (!dev
->vol
.migr_state
)
5730 if (migr_type(dev
) != MIGR_INIT
)
5733 migr_map
= get_imsm_map(dev
, 1);
5735 if (migr_map
->map_state
== IMSM_T_STATE_UNINITIALIZED
)
5742 static void update_recovery_start(struct intel_super
*super
,
5743 struct imsm_dev
*dev
,
5744 struct mdinfo
*array
)
5746 struct mdinfo
*rebuild
= NULL
;
5750 if (!is_rebuilding(dev
))
5753 /* Find the rebuild target, but punt on the dual rebuild case */
5754 for (d
= array
->devs
; d
; d
= d
->next
)
5755 if (d
->recovery_start
== 0) {
5762 /* (?) none of the disks are marked with
5763 * IMSM_ORD_REBUILD, so assume they are missing and the
5764 * disk_ord_tbl was not correctly updated
5766 dprintf("%s: failed to locate out-of-sync disk\n", __func__
);
5770 units
= __le32_to_cpu(dev
->vol
.curr_migr_unit
);
5771 rebuild
->recovery_start
= units
* blocks_per_migr_unit(super
, dev
);
5775 static int recover_backup_imsm(struct supertype
*st
, struct mdinfo
*info
);
5778 static struct mdinfo
*container_content_imsm(struct supertype
*st
, char *subarray
)
5780 /* Given a container loaded by load_super_imsm_all,
5781 * extract information about all the arrays into
5783 * If 'subarray' is given, just extract info about that array.
5785 * For each imsm_dev create an mdinfo, fill it in,
5786 * then look for matching devices in super->disks
5787 * and create appropriate device mdinfo.
5789 struct intel_super
*super
= st
->sb
;
5790 struct imsm_super
*mpb
= super
->anchor
;
5791 struct mdinfo
*rest
= NULL
;
5795 int spare_disks
= 0;
5797 /* do not assemble arrays when not all attributes are supported */
5798 if (imsm_check_attributes(mpb
->attributes
) == 0) {
5800 fprintf(stderr
, Name
": Unsupported attributes in IMSM metadata."
5801 "Arrays activation is blocked.\n");
5804 /* check for bad blocks */
5805 if (imsm_bbm_log_size(super
->anchor
)) {
5806 fprintf(stderr
, Name
": BBM log found in IMSM metadata."
5807 "Arrays activation is blocked.\n");
5812 /* count spare devices, not used in maps
5814 for (d
= super
->disks
; d
; d
= d
->next
)
5818 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
5819 struct imsm_dev
*dev
;
5820 struct imsm_map
*map
;
5821 struct imsm_map
*map2
;
5822 struct mdinfo
*this;
5827 (i
!= strtoul(subarray
, &ep
, 10) || *ep
!= '\0'))
5830 dev
= get_imsm_dev(super
, i
);
5831 map
= get_imsm_map(dev
, 0);
5832 map2
= get_imsm_map(dev
, 1);
5834 /* do not publish arrays that are in the middle of an
5835 * unsupported migration
5837 if (dev
->vol
.migr_state
&&
5838 (migr_type(dev
) == MIGR_STATE_CHANGE
)) {
5839 fprintf(stderr
, Name
": cannot assemble volume '%.16s':"
5840 " unsupported migration in progress\n",
5844 /* do not publish arrays that are not support by controller's
5848 chunk
= __le16_to_cpu(map
->blocks_per_strip
) >> 1;
5849 this = malloc(sizeof(*this));
5851 fprintf(stderr
, Name
": failed to allocate %zu bytes\n",
5856 super
->current_vol
= i
;
5857 getinfo_super_imsm_volume(st
, this, NULL
);
5860 /* mdadm does not support all metadata features- set the bit in all arrays state */
5861 if (!validate_geometry_imsm_orom(super
,
5862 get_imsm_raid_level(map
), /* RAID level */
5863 imsm_level_to_layout(get_imsm_raid_level(map
)),
5864 map
->num_members
, /* raid disks */
5867 fprintf(stderr
, Name
": IMSM RAID geometry validation"
5868 " failed. Array %s activation is blocked.\n",
5870 this->array
.state
|=
5871 (1<<MD_SB_BLOCK_CONTAINER_RESHAPE
) |
5872 (1<<MD_SB_BLOCK_VOLUME
);
5876 /* if array has bad blocks, set suitable bit in all arrays state */
5878 this->array
.state
|=
5879 (1<<MD_SB_BLOCK_CONTAINER_RESHAPE
) |
5880 (1<<MD_SB_BLOCK_VOLUME
);
5882 for (slot
= 0 ; slot
< map
->num_members
; slot
++) {
5883 unsigned long long recovery_start
;
5884 struct mdinfo
*info_d
;
5891 idx
= get_imsm_disk_idx(dev
, slot
, 0);
5892 ord
= get_imsm_ord_tbl_ent(dev
, slot
, -1);
5893 for (d
= super
->disks
; d
; d
= d
->next
)
5894 if (d
->index
== idx
)
5897 recovery_start
= MaxSector
;
5900 if (d
&& is_failed(&d
->disk
))
5902 if (ord
& IMSM_ORD_REBUILD
)
5906 * if we skip some disks the array will be assmebled degraded;
5907 * reset resync start to avoid a dirty-degraded
5908 * situation when performing the intial sync
5910 * FIXME handle dirty degraded
5912 if ((skip
|| recovery_start
== 0) && !dev
->vol
.dirty
)
5913 this->resync_start
= MaxSector
;
5917 info_d
= calloc(1, sizeof(*info_d
));
5919 fprintf(stderr
, Name
": failed to allocate disk"
5920 " for volume %.16s\n", dev
->volume
);
5921 info_d
= this->devs
;
5923 struct mdinfo
*d
= info_d
->next
;
5932 info_d
->next
= this->devs
;
5933 this->devs
= info_d
;
5935 info_d
->disk
.number
= d
->index
;
5936 info_d
->disk
.major
= d
->major
;
5937 info_d
->disk
.minor
= d
->minor
;
5938 info_d
->disk
.raid_disk
= slot
;
5939 info_d
->recovery_start
= recovery_start
;
5941 if (slot
< map2
->num_members
)
5942 info_d
->disk
.state
= (1 << MD_DISK_ACTIVE
);
5944 this->array
.spare_disks
++;
5946 if (slot
< map
->num_members
)
5947 info_d
->disk
.state
= (1 << MD_DISK_ACTIVE
);
5949 this->array
.spare_disks
++;
5951 if (info_d
->recovery_start
== MaxSector
)
5952 this->array
.working_disks
++;
5954 info_d
->events
= __le32_to_cpu(mpb
->generation_num
);
5955 info_d
->data_offset
= __le32_to_cpu(map
->pba_of_lba0
);
5956 info_d
->component_size
= __le32_to_cpu(map
->blocks_per_member
);
5958 /* now that the disk list is up-to-date fixup recovery_start */
5959 update_recovery_start(super
, dev
, this);
5960 this->array
.spare_disks
+= spare_disks
;
5963 /* check for reshape */
5964 if (this->reshape_active
== 1)
5965 recover_backup_imsm(st
, this);
5974 static __u8
imsm_check_degraded(struct intel_super
*super
, struct imsm_dev
*dev
,
5975 int failed
, int look_in_map
)
5977 struct imsm_map
*map
;
5979 map
= get_imsm_map(dev
, look_in_map
);
5982 return map
->map_state
== IMSM_T_STATE_UNINITIALIZED
?
5983 IMSM_T_STATE_UNINITIALIZED
: IMSM_T_STATE_NORMAL
;
5985 switch (get_imsm_raid_level(map
)) {
5987 return IMSM_T_STATE_FAILED
;
5990 if (failed
< map
->num_members
)
5991 return IMSM_T_STATE_DEGRADED
;
5993 return IMSM_T_STATE_FAILED
;
5998 * check to see if any mirrors have failed, otherwise we
5999 * are degraded. Even numbered slots are mirrored on
6003 /* gcc -Os complains that this is unused */
6004 int insync
= insync
;
6006 for (i
= 0; i
< map
->num_members
; i
++) {
6007 __u32 ord
= get_imsm_ord_tbl_ent(dev
, i
, -1);
6008 int idx
= ord_to_idx(ord
);
6009 struct imsm_disk
*disk
;
6011 /* reset the potential in-sync count on even-numbered
6012 * slots. num_copies is always 2 for imsm raid10
6017 disk
= get_imsm_disk(super
, idx
);
6018 if (!disk
|| is_failed(disk
) || ord
& IMSM_ORD_REBUILD
)
6021 /* no in-sync disks left in this mirror the
6025 return IMSM_T_STATE_FAILED
;
6028 return IMSM_T_STATE_DEGRADED
;
6032 return IMSM_T_STATE_DEGRADED
;
6034 return IMSM_T_STATE_FAILED
;
6040 return map
->map_state
;
6043 static int imsm_count_failed(struct intel_super
*super
, struct imsm_dev
*dev
,
6048 struct imsm_disk
*disk
;
6049 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
6050 struct imsm_map
*prev
= get_imsm_map(dev
, MAP_1
);
6051 struct imsm_map
*map_for_loop
;
6056 /* at the beginning of migration we set IMSM_ORD_REBUILD on
6057 * disks that are being rebuilt. New failures are recorded to
6058 * map[0]. So we look through all the disks we started with and
6059 * see if any failures are still present, or if any new ones
6063 if (prev
&& (map
->num_members
< prev
->num_members
))
6064 map_for_loop
= prev
;
6066 for (i
= 0; i
< map_for_loop
->num_members
; i
++) {
6069 (look_in_map
& MAP_1
) && (i
< prev
->num_members
)) {
6070 ord
= __le32_to_cpu(prev
->disk_ord_tbl
[i
]);
6071 idx_1
= ord_to_idx(ord
);
6073 disk
= get_imsm_disk(super
, idx_1
);
6074 if (!disk
|| is_failed(disk
) || ord
& IMSM_ORD_REBUILD
)
6077 if ((look_in_map
& MAP_0
) && (i
< map
->num_members
)) {
6078 ord
= __le32_to_cpu(map
->disk_ord_tbl
[i
]);
6079 idx
= ord_to_idx(ord
);
6082 disk
= get_imsm_disk(super
, idx
);
6083 if (!disk
|| is_failed(disk
) ||
6084 ord
& IMSM_ORD_REBUILD
)
6094 static int imsm_open_new(struct supertype
*c
, struct active_array
*a
,
6097 struct intel_super
*super
= c
->sb
;
6098 struct imsm_super
*mpb
= super
->anchor
;
6100 if (atoi(inst
) >= mpb
->num_raid_devs
) {
6101 fprintf(stderr
, "%s: subarry index %d, out of range\n",
6102 __func__
, atoi(inst
));
6106 dprintf("imsm: open_new %s\n", inst
);
6107 a
->info
.container_member
= atoi(inst
);
6111 static int is_resyncing(struct imsm_dev
*dev
)
6113 struct imsm_map
*migr_map
;
6115 if (!dev
->vol
.migr_state
)
6118 if (migr_type(dev
) == MIGR_INIT
||
6119 migr_type(dev
) == MIGR_REPAIR
)
6122 if (migr_type(dev
) == MIGR_GEN_MIGR
)
6125 migr_map
= get_imsm_map(dev
, 1);
6127 if ((migr_map
->map_state
== IMSM_T_STATE_NORMAL
) &&
6128 (dev
->vol
.migr_type
!= MIGR_GEN_MIGR
))
6134 /* return true if we recorded new information */
6135 static int mark_failure(struct imsm_dev
*dev
, struct imsm_disk
*disk
, int idx
)
6139 struct imsm_map
*map
;
6140 char buf
[MAX_RAID_SERIAL_LEN
+3];
6141 unsigned int len
, shift
= 0;
6143 /* new failures are always set in map[0] */
6144 map
= get_imsm_map(dev
, 0);
6146 slot
= get_imsm_disk_slot(map
, idx
);
6150 ord
= __le32_to_cpu(map
->disk_ord_tbl
[slot
]);
6151 if (is_failed(disk
) && (ord
& IMSM_ORD_REBUILD
))
6154 memcpy(buf
, disk
->serial
, MAX_RAID_SERIAL_LEN
);
6155 buf
[MAX_RAID_SERIAL_LEN
] = '\000';
6157 if ((len
= strlen(buf
)) >= MAX_RAID_SERIAL_LEN
)
6158 shift
= len
- MAX_RAID_SERIAL_LEN
+ 1;
6159 strncpy((char *)disk
->serial
, &buf
[shift
], MAX_RAID_SERIAL_LEN
);
6161 disk
->status
|= FAILED_DISK
;
6162 set_imsm_ord_tbl_ent(map
, slot
, idx
| IMSM_ORD_REBUILD
);
6163 /* mark failures in second map if second map exists and this disk
6165 * This is valid for migration, initialization and rebuild
6167 if (dev
->vol
.migr_state
) {
6168 struct imsm_map
*map2
= get_imsm_map(dev
, 1);
6169 int slot2
= get_imsm_disk_slot(map2
, idx
);
6171 if ((slot2
< map2
->num_members
) &&
6173 set_imsm_ord_tbl_ent(map2
, slot2
,
6174 idx
| IMSM_ORD_REBUILD
);
6176 if (map
->failed_disk_num
== 0xff)
6177 map
->failed_disk_num
= slot
;
6181 static void mark_missing(struct imsm_dev
*dev
, struct imsm_disk
*disk
, int idx
)
6183 mark_failure(dev
, disk
, idx
);
6185 if (disk
->scsi_id
== __cpu_to_le32(~(__u32
)0))
6188 disk
->scsi_id
= __cpu_to_le32(~(__u32
)0);
6189 memmove(&disk
->serial
[0], &disk
->serial
[1], MAX_RAID_SERIAL_LEN
- 1);
6192 static void handle_missing(struct intel_super
*super
, struct imsm_dev
*dev
)
6196 if (!super
->missing
)
6199 dprintf("imsm: mark missing\n");
6200 /* end process for initialization and rebuild only
6202 if (is_gen_migration(dev
) == 0) {
6206 failed
= imsm_count_failed(super
, dev
, MAP_0
);
6207 map_state
= imsm_check_degraded(super
, dev
, failed
, MAP_0
);
6209 end_migration(dev
, super
, map_state
);
6211 for (dl
= super
->missing
; dl
; dl
= dl
->next
)
6212 mark_missing(dev
, &dl
->disk
, dl
->index
);
6213 super
->updates_pending
++;
6216 static unsigned long long imsm_set_array_size(struct imsm_dev
*dev
)
6218 int used_disks
= imsm_num_data_members(dev
, 0);
6219 unsigned long long array_blocks
;
6220 struct imsm_map
*map
;
6222 if (used_disks
== 0) {
6223 /* when problems occures
6224 * return current array_blocks value
6226 array_blocks
= __le32_to_cpu(dev
->size_high
);
6227 array_blocks
= array_blocks
<< 32;
6228 array_blocks
+= __le32_to_cpu(dev
->size_low
);
6230 return array_blocks
;
6233 /* set array size in metadata
6235 map
= get_imsm_map(dev
, 0);
6236 array_blocks
= map
->blocks_per_member
* used_disks
;
6238 /* round array size down to closest MB
6240 array_blocks
= (array_blocks
>> SECT_PER_MB_SHIFT
) << SECT_PER_MB_SHIFT
;
6241 dev
->size_low
= __cpu_to_le32((__u32
)array_blocks
);
6242 dev
->size_high
= __cpu_to_le32((__u32
)(array_blocks
>> 32));
6244 return array_blocks
;
6247 static void imsm_set_disk(struct active_array
*a
, int n
, int state
);
6249 static void imsm_progress_container_reshape(struct intel_super
*super
)
6251 /* if no device has a migr_state, but some device has a
6252 * different number of members than the previous device, start
6253 * changing the number of devices in this device to match
6256 struct imsm_super
*mpb
= super
->anchor
;
6257 int prev_disks
= -1;
6261 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
6262 struct imsm_dev
*dev
= get_imsm_dev(super
, i
);
6263 struct imsm_map
*map
= get_imsm_map(dev
, 0);
6264 struct imsm_map
*map2
;
6265 int prev_num_members
;
6267 if (dev
->vol
.migr_state
)
6270 if (prev_disks
== -1)
6271 prev_disks
= map
->num_members
;
6272 if (prev_disks
== map
->num_members
)
6275 /* OK, this array needs to enter reshape mode.
6276 * i.e it needs a migr_state
6279 copy_map_size
= sizeof_imsm_map(map
);
6280 prev_num_members
= map
->num_members
;
6281 map
->num_members
= prev_disks
;
6282 dev
->vol
.migr_state
= 1;
6283 dev
->vol
.curr_migr_unit
= 0;
6284 set_migr_type(dev
, MIGR_GEN_MIGR
);
6285 for (i
= prev_num_members
;
6286 i
< map
->num_members
; i
++)
6287 set_imsm_ord_tbl_ent(map
, i
, i
);
6288 map2
= get_imsm_map(dev
, 1);
6289 /* Copy the current map */
6290 memcpy(map2
, map
, copy_map_size
);
6291 map2
->num_members
= prev_num_members
;
6293 imsm_set_array_size(dev
);
6294 super
->updates_pending
++;
6298 /* Handle dirty -> clean transititions, resync and reshape. Degraded and rebuild
6299 * states are handled in imsm_set_disk() with one exception, when a
6300 * resync is stopped due to a new failure this routine will set the
6301 * 'degraded' state for the array.
6303 static int imsm_set_array_state(struct active_array
*a
, int consistent
)
6305 int inst
= a
->info
.container_member
;
6306 struct intel_super
*super
= a
->container
->sb
;
6307 struct imsm_dev
*dev
= get_imsm_dev(super
, inst
);
6308 struct imsm_map
*map
= get_imsm_map(dev
, 0);
6309 int failed
= imsm_count_failed(super
, dev
, MAP_0
);
6310 __u8 map_state
= imsm_check_degraded(super
, dev
, failed
, MAP_0
);
6311 __u32 blocks_per_unit
;
6313 if (dev
->vol
.migr_state
&&
6314 dev
->vol
.migr_type
== MIGR_GEN_MIGR
) {
6315 /* array state change is blocked due to reshape action
6317 * - abort the reshape (if last_checkpoint is 0 and action!= reshape)
6318 * - finish the reshape (if last_checkpoint is big and action != reshape)
6319 * - update curr_migr_unit
6321 if (a
->curr_action
== reshape
) {
6322 /* still reshaping, maybe update curr_migr_unit */
6323 goto mark_checkpoint
;
6325 if (a
->last_checkpoint
== 0 && a
->prev_action
== reshape
) {
6326 /* for some reason we aborted the reshape.
6328 * disable automatic metadata rollback
6329 * user action is required to recover process
6332 struct imsm_map
*map2
= get_imsm_map(dev
, 1);
6333 dev
->vol
.migr_state
= 0;
6334 set_migr_type(dev
, 0);
6335 dev
->vol
.curr_migr_unit
= 0;
6336 memcpy(map
, map2
, sizeof_imsm_map(map2
));
6337 super
->updates_pending
++;
6340 if (a
->last_checkpoint
>= a
->info
.component_size
) {
6341 unsigned long long array_blocks
;
6345 used_disks
= imsm_num_data_members(dev
, 0);
6346 if (used_disks
> 0) {
6348 map
->blocks_per_member
*
6350 /* round array size down to closest MB
6352 array_blocks
= (array_blocks
6353 >> SECT_PER_MB_SHIFT
)
6354 << SECT_PER_MB_SHIFT
;
6355 a
->info
.custom_array_size
= array_blocks
;
6356 /* encourage manager to update array
6360 a
->check_reshape
= 1;
6362 /* finalize online capacity expansion/reshape */
6363 for (mdi
= a
->info
.devs
; mdi
; mdi
= mdi
->next
)
6365 mdi
->disk
.raid_disk
,
6368 imsm_progress_container_reshape(super
);
6373 /* before we activate this array handle any missing disks */
6374 if (consistent
== 2)
6375 handle_missing(super
, dev
);
6377 if (consistent
== 2 &&
6378 (!is_resync_complete(&a
->info
) ||
6379 map_state
!= IMSM_T_STATE_NORMAL
||
6380 dev
->vol
.migr_state
))
6383 if (is_resync_complete(&a
->info
)) {
6384 /* complete intialization / resync,
6385 * recovery and interrupted recovery is completed in
6388 if (is_resyncing(dev
)) {
6389 dprintf("imsm: mark resync done\n");
6390 end_migration(dev
, super
, map_state
);
6391 super
->updates_pending
++;
6392 a
->last_checkpoint
= 0;
6394 } else if ((!is_resyncing(dev
) && !failed
) &&
6395 (imsm_reshape_blocks_arrays_changes(super
) == 0)) {
6396 /* mark the start of the init process if nothing is failed */
6397 dprintf("imsm: mark resync start\n");
6398 if (map
->map_state
== IMSM_T_STATE_UNINITIALIZED
)
6399 migrate(dev
, super
, IMSM_T_STATE_NORMAL
, MIGR_INIT
);
6401 migrate(dev
, super
, IMSM_T_STATE_NORMAL
, MIGR_REPAIR
);
6402 super
->updates_pending
++;
6406 /* skip checkpointing for general migration,
6407 * it is controlled in mdadm
6409 if (is_gen_migration(dev
))
6410 goto skip_mark_checkpoint
;
6412 /* check if we can update curr_migr_unit from resync_start, recovery_start */
6413 blocks_per_unit
= blocks_per_migr_unit(super
, dev
);
6414 if (blocks_per_unit
) {
6418 units
= a
->last_checkpoint
/ blocks_per_unit
;
6421 /* check that we did not overflow 32-bits, and that
6422 * curr_migr_unit needs updating
6424 if (units32
== units
&&
6426 __le32_to_cpu(dev
->vol
.curr_migr_unit
) != units32
) {
6427 dprintf("imsm: mark checkpoint (%u)\n", units32
);
6428 dev
->vol
.curr_migr_unit
= __cpu_to_le32(units32
);
6429 super
->updates_pending
++;
6433 skip_mark_checkpoint
:
6434 /* mark dirty / clean */
6435 if (dev
->vol
.dirty
!= !consistent
) {
6436 dprintf("imsm: mark '%s'\n", consistent
? "clean" : "dirty");
6441 super
->updates_pending
++;
6447 static void imsm_set_disk(struct active_array
*a
, int n
, int state
)
6449 int inst
= a
->info
.container_member
;
6450 struct intel_super
*super
= a
->container
->sb
;
6451 struct imsm_dev
*dev
= get_imsm_dev(super
, inst
);
6452 struct imsm_map
*map
= get_imsm_map(dev
, 0);
6453 struct imsm_disk
*disk
;
6458 if (n
> map
->num_members
)
6459 fprintf(stderr
, "imsm: set_disk %d out of range 0..%d\n",
6460 n
, map
->num_members
- 1);
6465 dprintf("imsm: set_disk %d:%x\n", n
, state
);
6467 ord
= get_imsm_ord_tbl_ent(dev
, n
, 0);
6468 disk
= get_imsm_disk(super
, ord_to_idx(ord
));
6470 /* check for new failures */
6471 if (state
& DS_FAULTY
) {
6472 if (mark_failure(dev
, disk
, ord_to_idx(ord
)))
6473 super
->updates_pending
++;
6476 /* check if in_sync */
6477 if (state
& DS_INSYNC
&& ord
& IMSM_ORD_REBUILD
&& is_rebuilding(dev
)) {
6478 struct imsm_map
*migr_map
= get_imsm_map(dev
, 1);
6480 set_imsm_ord_tbl_ent(migr_map
, n
, ord_to_idx(ord
));
6481 super
->updates_pending
++;
6484 failed
= imsm_count_failed(super
, dev
, MAP_0
);
6485 map_state
= imsm_check_degraded(super
, dev
, failed
, MAP_0
);
6487 /* check if recovery complete, newly degraded, or failed */
6488 dprintf("imsm: Detected transition to state ");
6489 switch (map_state
) {
6490 case IMSM_T_STATE_NORMAL
: /* transition to normal state */
6491 dprintf("normal: ");
6492 if (is_rebuilding(dev
)) {
6493 dprintf("while rebuilding");
6494 end_migration(dev
, super
, map_state
);
6495 map
= get_imsm_map(dev
, 0);
6496 map
->failed_disk_num
= ~0;
6497 super
->updates_pending
++;
6498 a
->last_checkpoint
= 0;
6501 if (is_gen_migration(dev
)) {
6502 dprintf("while general migration");
6503 if (a
->last_checkpoint
>= a
->info
.component_size
)
6504 end_migration(dev
, super
, map_state
);
6506 map
->map_state
= map_state
;
6507 map
= get_imsm_map(dev
, 0);
6508 map
->failed_disk_num
= ~0;
6509 super
->updates_pending
++;
6513 case IMSM_T_STATE_DEGRADED
: /* transition to degraded state */
6514 dprintf("degraded: ");
6515 if ((map
->map_state
!= map_state
) &&
6516 !dev
->vol
.migr_state
) {
6517 dprintf("mark degraded");
6518 map
->map_state
= map_state
;
6519 super
->updates_pending
++;
6520 a
->last_checkpoint
= 0;
6523 if (is_rebuilding(dev
)) {
6524 dprintf("while rebuilding.");
6525 if (map
->map_state
!= map_state
) {
6526 dprintf(" Map state change");
6527 end_migration(dev
, super
, map_state
);
6528 super
->updates_pending
++;
6532 if (is_gen_migration(dev
)) {
6533 dprintf("while general migration");
6534 if (a
->last_checkpoint
>= a
->info
.component_size
)
6535 end_migration(dev
, super
, map_state
);
6537 map
->map_state
= map_state
;
6538 manage_second_map(super
, dev
);
6540 super
->updates_pending
++;
6543 if (is_initializing(dev
)) {
6544 dprintf("while initialization.");
6545 map
->map_state
= map_state
;
6546 super
->updates_pending
++;
6550 case IMSM_T_STATE_FAILED
: /* transition to failed state */
6551 dprintf("failed: ");
6552 if (is_gen_migration(dev
)) {
6553 dprintf("while general migration");
6554 map
->map_state
= map_state
;
6555 super
->updates_pending
++;
6558 if (map
->map_state
!= map_state
) {
6559 dprintf("mark failed");
6560 end_migration(dev
, super
, map_state
);
6561 super
->updates_pending
++;
6562 a
->last_checkpoint
= 0;
6567 dprintf("state %i\n", map_state
);
6573 static int store_imsm_mpb(int fd
, struct imsm_super
*mpb
)
6576 __u32 mpb_size
= __le32_to_cpu(mpb
->mpb_size
);
6577 unsigned long long dsize
;
6578 unsigned long long sectors
;
6580 get_dev_size(fd
, NULL
, &dsize
);
6582 if (mpb_size
> 512) {
6583 /* -1 to account for anchor */
6584 sectors
= mpb_sectors(mpb
) - 1;
6586 /* write the extended mpb to the sectors preceeding the anchor */
6587 if (lseek64(fd
, dsize
- (512 * (2 + sectors
)), SEEK_SET
) < 0)
6590 if ((unsigned long long)write(fd
, buf
+ 512, 512 * sectors
)
6595 /* first block is stored on second to last sector of the disk */
6596 if (lseek64(fd
, dsize
- (512 * 2), SEEK_SET
) < 0)
6599 if (write(fd
, buf
, 512) != 512)
6605 static void imsm_sync_metadata(struct supertype
*container
)
6607 struct intel_super
*super
= container
->sb
;
6609 dprintf("sync metadata: %d\n", super
->updates_pending
);
6610 if (!super
->updates_pending
)
6613 write_super_imsm(container
, 0);
6615 super
->updates_pending
= 0;
6618 static struct dl
*imsm_readd(struct intel_super
*super
, int idx
, struct active_array
*a
)
6620 struct imsm_dev
*dev
= get_imsm_dev(super
, a
->info
.container_member
);
6621 int i
= get_imsm_disk_idx(dev
, idx
, -1);
6624 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
6628 if (dl
&& is_failed(&dl
->disk
))
6632 dprintf("%s: found %x:%x\n", __func__
, dl
->major
, dl
->minor
);
6637 static struct dl
*imsm_add_spare(struct intel_super
*super
, int slot
,
6638 struct active_array
*a
, int activate_new
,
6639 struct mdinfo
*additional_test_list
)
6641 struct imsm_dev
*dev
= get_imsm_dev(super
, a
->info
.container_member
);
6642 int idx
= get_imsm_disk_idx(dev
, slot
, -1);
6643 struct imsm_super
*mpb
= super
->anchor
;
6644 struct imsm_map
*map
;
6645 unsigned long long pos
;
6650 __u32 array_start
= 0;
6651 __u32 array_end
= 0;
6653 struct mdinfo
*test_list
;
6655 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
6656 /* If in this array, skip */
6657 for (d
= a
->info
.devs
; d
; d
= d
->next
)
6658 if (d
->state_fd
>= 0 &&
6659 d
->disk
.major
== dl
->major
&&
6660 d
->disk
.minor
== dl
->minor
) {
6661 dprintf("%x:%x already in array\n",
6662 dl
->major
, dl
->minor
);
6667 test_list
= additional_test_list
;
6669 if (test_list
->disk
.major
== dl
->major
&&
6670 test_list
->disk
.minor
== dl
->minor
) {
6671 dprintf("%x:%x already in additional test list\n",
6672 dl
->major
, dl
->minor
);
6675 test_list
= test_list
->next
;
6680 /* skip in use or failed drives */
6681 if (is_failed(&dl
->disk
) || idx
== dl
->index
||
6683 dprintf("%x:%x status (failed: %d index: %d)\n",
6684 dl
->major
, dl
->minor
, is_failed(&dl
->disk
), idx
);
6688 /* skip pure spares when we are looking for partially
6689 * assimilated drives
6691 if (dl
->index
== -1 && !activate_new
)
6694 /* Does this unused device have the requisite free space?
6695 * It needs to be able to cover all member volumes
6697 ex
= get_extents(super
, dl
);
6699 dprintf("cannot get extents\n");
6702 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
6703 dev
= get_imsm_dev(super
, i
);
6704 map
= get_imsm_map(dev
, 0);
6706 /* check if this disk is already a member of
6709 if (get_imsm_disk_slot(map
, dl
->index
) >= 0)
6715 array_start
= __le32_to_cpu(map
->pba_of_lba0
);
6716 array_end
= array_start
+
6717 __le32_to_cpu(map
->blocks_per_member
) - 1;
6720 /* check that we can start at pba_of_lba0 with
6721 * blocks_per_member of space
6723 if (array_start
>= pos
&& array_end
< ex
[j
].start
) {
6727 pos
= ex
[j
].start
+ ex
[j
].size
;
6729 } while (ex
[j
-1].size
);
6736 if (i
< mpb
->num_raid_devs
) {
6737 dprintf("%x:%x does not have %u to %u available\n",
6738 dl
->major
, dl
->minor
, array_start
, array_end
);
6749 static int imsm_rebuild_allowed(struct supertype
*cont
, int dev_idx
, int failed
)
6751 struct imsm_dev
*dev2
;
6752 struct imsm_map
*map
;
6758 dev2
= get_imsm_dev(cont
->sb
, dev_idx
);
6760 state
= imsm_check_degraded(cont
->sb
, dev2
, failed
,
6762 if (state
== IMSM_T_STATE_FAILED
) {
6763 map
= get_imsm_map(dev2
, 0);
6766 for (slot
= 0; slot
< map
->num_members
; slot
++) {
6768 * Check if failed disks are deleted from intel
6769 * disk list or are marked to be deleted
6771 idx
= get_imsm_disk_idx(dev2
, slot
, -1);
6772 idisk
= get_imsm_dl_disk(cont
->sb
, idx
);
6774 * Do not rebuild the array if failed disks
6775 * from failed sub-array are not removed from
6779 is_failed(&idisk
->disk
) &&
6780 (idisk
->action
!= DISK_REMOVE
))
6788 static struct mdinfo
*imsm_activate_spare(struct active_array
*a
,
6789 struct metadata_update
**updates
)
6792 * Find a device with unused free space and use it to replace a
6793 * failed/vacant region in an array. We replace failed regions one a
6794 * array at a time. The result is that a new spare disk will be added
6795 * to the first failed array and after the monitor has finished
6796 * propagating failures the remainder will be consumed.
6798 * FIXME add a capability for mdmon to request spares from another
6802 struct intel_super
*super
= a
->container
->sb
;
6803 int inst
= a
->info
.container_member
;
6804 struct imsm_dev
*dev
= get_imsm_dev(super
, inst
);
6805 struct imsm_map
*map
= get_imsm_map(dev
, 0);
6806 int failed
= a
->info
.array
.raid_disks
;
6807 struct mdinfo
*rv
= NULL
;
6810 struct metadata_update
*mu
;
6812 struct imsm_update_activate_spare
*u
;
6817 for (d
= a
->info
.devs
; d
; d
= d
->next
) {
6818 if ((d
->curr_state
& DS_FAULTY
) &&
6820 /* wait for Removal to happen */
6822 if (d
->state_fd
>= 0)
6826 dprintf("imsm: activate spare: inst=%d failed=%d (%d) level=%d\n",
6827 inst
, failed
, a
->info
.array
.raid_disks
, a
->info
.array
.level
);
6829 if (imsm_reshape_blocks_arrays_changes(super
))
6832 /* Cannot activate another spare if rebuild is in progress already
6834 if (is_rebuilding(dev
)) {
6835 dprintf("imsm: No spare activation allowed. "
6836 "Rebuild in progress already.\n");
6840 if (a
->info
.array
.level
== 4)
6841 /* No repair for takeovered array
6842 * imsm doesn't support raid4
6846 if (imsm_check_degraded(super
, dev
, failed
, MAP_0
) !=
6847 IMSM_T_STATE_DEGRADED
)
6851 * If there are any failed disks check state of the other volume.
6852 * Block rebuild if the another one is failed until failed disks
6853 * are removed from container.
6856 dprintf("found failed disks in %.*s, check if there another"
6857 "failed sub-array.\n",
6858 MAX_RAID_SERIAL_LEN
, dev
->volume
);
6859 /* check if states of the other volumes allow for rebuild */
6860 for (i
= 0; i
< super
->anchor
->num_raid_devs
; i
++) {
6862 allowed
= imsm_rebuild_allowed(a
->container
,
6870 /* For each slot, if it is not working, find a spare */
6871 for (i
= 0; i
< a
->info
.array
.raid_disks
; i
++) {
6872 for (d
= a
->info
.devs
; d
; d
= d
->next
)
6873 if (d
->disk
.raid_disk
== i
)
6875 dprintf("found %d: %p %x\n", i
, d
, d
?d
->curr_state
:0);
6876 if (d
&& (d
->state_fd
>= 0))
6880 * OK, this device needs recovery. Try to re-add the
6881 * previous occupant of this slot, if this fails see if
6882 * we can continue the assimilation of a spare that was
6883 * partially assimilated, finally try to activate a new
6886 dl
= imsm_readd(super
, i
, a
);
6888 dl
= imsm_add_spare(super
, i
, a
, 0, rv
);
6890 dl
= imsm_add_spare(super
, i
, a
, 1, rv
);
6894 /* found a usable disk with enough space */
6895 di
= malloc(sizeof(*di
));
6898 memset(di
, 0, sizeof(*di
));
6900 /* dl->index will be -1 in the case we are activating a
6901 * pristine spare. imsm_process_update() will create a
6902 * new index in this case. Once a disk is found to be
6903 * failed in all member arrays it is kicked from the
6906 di
->disk
.number
= dl
->index
;
6908 /* (ab)use di->devs to store a pointer to the device
6911 di
->devs
= (struct mdinfo
*) dl
;
6913 di
->disk
.raid_disk
= i
;
6914 di
->disk
.major
= dl
->major
;
6915 di
->disk
.minor
= dl
->minor
;
6917 di
->recovery_start
= 0;
6918 di
->data_offset
= __le32_to_cpu(map
->pba_of_lba0
);
6919 di
->component_size
= a
->info
.component_size
;
6920 di
->container_member
= inst
;
6921 super
->random
= random32();
6925 dprintf("%x:%x to be %d at %llu\n", dl
->major
, dl
->minor
,
6926 i
, di
->data_offset
);
6930 /* No spares found */
6932 /* Now 'rv' has a list of devices to return.
6933 * Create a metadata_update record to update the
6934 * disk_ord_tbl for the array
6936 mu
= malloc(sizeof(*mu
));
6938 mu
->buf
= malloc(sizeof(struct imsm_update_activate_spare
) * num_spares
);
6939 if (mu
->buf
== NULL
) {
6946 struct mdinfo
*n
= rv
->next
;
6955 mu
->space_list
= NULL
;
6956 mu
->len
= sizeof(struct imsm_update_activate_spare
) * num_spares
;
6957 mu
->next
= *updates
;
6958 u
= (struct imsm_update_activate_spare
*) mu
->buf
;
6960 for (di
= rv
; di
; di
= di
->next
) {
6961 u
->type
= update_activate_spare
;
6962 u
->dl
= (struct dl
*) di
->devs
;
6964 u
->slot
= di
->disk
.raid_disk
;
6975 static int disks_overlap(struct intel_super
*super
, int idx
, struct imsm_update_create_array
*u
)
6977 struct imsm_dev
*dev
= get_imsm_dev(super
, idx
);
6978 struct imsm_map
*map
= get_imsm_map(dev
, 0);
6979 struct imsm_map
*new_map
= get_imsm_map(&u
->dev
, 0);
6980 struct disk_info
*inf
= get_disk_info(u
);
6981 struct imsm_disk
*disk
;
6985 for (i
= 0; i
< map
->num_members
; i
++) {
6986 disk
= get_imsm_disk(super
, get_imsm_disk_idx(dev
, i
, -1));
6987 for (j
= 0; j
< new_map
->num_members
; j
++)
6988 if (serialcmp(disk
->serial
, inf
[j
].serial
) == 0)
6996 static struct dl
*get_disk_super(struct intel_super
*super
, int major
, int minor
)
6998 struct dl
*dl
= NULL
;
6999 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
7000 if ((dl
->major
== major
) && (dl
->minor
== minor
))
7005 static int remove_disk_super(struct intel_super
*super
, int major
, int minor
)
7007 struct dl
*prev
= NULL
;
7011 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
7012 if ((dl
->major
== major
) && (dl
->minor
== minor
)) {
7015 prev
->next
= dl
->next
;
7017 super
->disks
= dl
->next
;
7019 __free_imsm_disk(dl
);
7020 dprintf("%s: removed %x:%x\n",
7021 __func__
, major
, minor
);
7029 static void imsm_delete(struct intel_super
*super
, struct dl
**dlp
, unsigned index
);
7031 static int add_remove_disk_update(struct intel_super
*super
)
7033 int check_degraded
= 0;
7034 struct dl
*disk
= NULL
;
7035 /* add/remove some spares to/from the metadata/contrainer */
7036 while (super
->disk_mgmt_list
) {
7037 struct dl
*disk_cfg
;
7039 disk_cfg
= super
->disk_mgmt_list
;
7040 super
->disk_mgmt_list
= disk_cfg
->next
;
7041 disk_cfg
->next
= NULL
;
7043 if (disk_cfg
->action
== DISK_ADD
) {
7044 disk_cfg
->next
= super
->disks
;
7045 super
->disks
= disk_cfg
;
7047 dprintf("%s: added %x:%x\n",
7048 __func__
, disk_cfg
->major
,
7050 } else if (disk_cfg
->action
== DISK_REMOVE
) {
7051 dprintf("Disk remove action processed: %x.%x\n",
7052 disk_cfg
->major
, disk_cfg
->minor
);
7053 disk
= get_disk_super(super
,
7057 /* store action status */
7058 disk
->action
= DISK_REMOVE
;
7059 /* remove spare disks only */
7060 if (disk
->index
== -1) {
7061 remove_disk_super(super
,
7066 /* release allocate disk structure */
7067 __free_imsm_disk(disk_cfg
);
7070 return check_degraded
;
7074 static int apply_reshape_migration_update(struct imsm_update_reshape_migration
*u
,
7075 struct intel_super
*super
,
7078 struct intel_dev
*id
;
7079 void **tofree
= NULL
;
7082 dprintf("apply_reshape_migration_update()\n");
7083 if ((u
->subdev
< 0) ||
7085 dprintf("imsm: Error: Wrong subdev: %i\n", u
->subdev
);
7088 if ((space_list
== NULL
) || (*space_list
== NULL
)) {
7089 dprintf("imsm: Error: Memory is not allocated\n");
7093 for (id
= super
->devlist
; id
; id
= id
->next
) {
7094 if (id
->index
== (unsigned)u
->subdev
) {
7095 struct imsm_dev
*dev
= get_imsm_dev(super
, u
->subdev
);
7096 struct imsm_map
*map
;
7097 struct imsm_dev
*new_dev
=
7098 (struct imsm_dev
*)*space_list
;
7099 struct imsm_map
*migr_map
= get_imsm_map(dev
, 1);
7101 struct dl
*new_disk
;
7103 if (new_dev
== NULL
)
7105 *space_list
= **space_list
;
7106 memcpy(new_dev
, dev
, sizeof_imsm_dev(dev
, 0));
7107 map
= get_imsm_map(new_dev
, 0);
7109 dprintf("imsm: Error: migration in progress");
7113 to_state
= map
->map_state
;
7114 if ((u
->new_level
== 5) && (map
->raid_level
== 0)) {
7116 /* this should not happen */
7117 if (u
->new_disks
[0] < 0) {
7118 map
->failed_disk_num
=
7119 map
->num_members
- 1;
7120 to_state
= IMSM_T_STATE_DEGRADED
;
7122 to_state
= IMSM_T_STATE_NORMAL
;
7124 migrate(new_dev
, super
, to_state
, MIGR_GEN_MIGR
);
7125 if (u
->new_level
> -1)
7126 map
->raid_level
= u
->new_level
;
7127 migr_map
= get_imsm_map(new_dev
, 1);
7128 if ((u
->new_level
== 5) &&
7129 (migr_map
->raid_level
== 0)) {
7130 int ord
= map
->num_members
- 1;
7131 migr_map
->num_members
--;
7132 if (u
->new_disks
[0] < 0)
7133 ord
|= IMSM_ORD_REBUILD
;
7134 set_imsm_ord_tbl_ent(map
,
7135 map
->num_members
- 1,
7139 tofree
= (void **)dev
;
7141 /* update chunk size
7143 if (u
->new_chunksize
> 0)
7144 map
->blocks_per_strip
=
7145 __cpu_to_le16(u
->new_chunksize
* 2);
7149 if ((u
->new_level
!= 5) ||
7150 (migr_map
->raid_level
!= 0) ||
7151 (migr_map
->raid_level
== map
->raid_level
))
7154 if (u
->new_disks
[0] >= 0) {
7157 new_disk
= get_disk_super(super
,
7158 major(u
->new_disks
[0]),
7159 minor(u
->new_disks
[0]));
7160 dprintf("imsm: new disk for reshape is: %i:%i "
7161 "(%p, index = %i)\n",
7162 major(u
->new_disks
[0]),
7163 minor(u
->new_disks
[0]),
7164 new_disk
, new_disk
->index
);
7165 if (new_disk
== NULL
)
7166 goto error_disk_add
;
7168 new_disk
->index
= map
->num_members
- 1;
7169 /* slot to fill in autolayout
7171 new_disk
->raiddisk
= new_disk
->index
;
7172 new_disk
->disk
.status
|= CONFIGURED_DISK
;
7173 new_disk
->disk
.status
&= ~SPARE_DISK
;
7175 goto error_disk_add
;
7178 *tofree
= *space_list
;
7179 /* calculate new size
7181 imsm_set_array_size(new_dev
);
7188 *space_list
= tofree
;
7192 dprintf("Error: imsm: Cannot find disk.\n");
7196 static int apply_update_activate_spare(struct imsm_update_activate_spare
*u
,
7197 struct intel_super
*super
,
7198 struct active_array
*active_array
)
7200 struct imsm_super
*mpb
= super
->anchor
;
7201 struct imsm_dev
*dev
= get_imsm_dev(super
, u
->array
);
7202 struct imsm_map
*map
= get_imsm_map(dev
, 0);
7203 struct imsm_map
*migr_map
;
7204 struct active_array
*a
;
7205 struct imsm_disk
*disk
;
7212 int second_map_created
= 0;
7214 for (; u
; u
= u
->next
) {
7215 victim
= get_imsm_disk_idx(dev
, u
->slot
, -1);
7220 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
7225 fprintf(stderr
, "error: imsm_activate_spare passed "
7226 "an unknown disk (index: %d)\n",
7231 /* count failures (excluding rebuilds and the victim)
7232 * to determine map[0] state
7235 for (i
= 0; i
< map
->num_members
; i
++) {
7238 disk
= get_imsm_disk(super
,
7239 get_imsm_disk_idx(dev
, i
, -1));
7240 if (!disk
|| is_failed(disk
))
7244 /* adding a pristine spare, assign a new index */
7245 if (dl
->index
< 0) {
7246 dl
->index
= super
->anchor
->num_disks
;
7247 super
->anchor
->num_disks
++;
7250 disk
->status
|= CONFIGURED_DISK
;
7251 disk
->status
&= ~SPARE_DISK
;
7254 to_state
= imsm_check_degraded(super
, dev
, failed
,
7256 if (!second_map_created
) {
7257 second_map_created
= 1;
7258 map
->map_state
= IMSM_T_STATE_DEGRADED
;
7259 migrate(dev
, super
, to_state
, MIGR_REBUILD
);
7261 map
->map_state
= to_state
;
7262 migr_map
= get_imsm_map(dev
, 1);
7263 set_imsm_ord_tbl_ent(map
, u
->slot
, dl
->index
);
7264 set_imsm_ord_tbl_ent(migr_map
, u
->slot
,
7265 dl
->index
| IMSM_ORD_REBUILD
);
7267 /* update the family_num to mark a new container
7268 * generation, being careful to record the existing
7269 * family_num in orig_family_num to clean up after
7270 * earlier mdadm versions that neglected to set it.
7272 if (mpb
->orig_family_num
== 0)
7273 mpb
->orig_family_num
= mpb
->family_num
;
7274 mpb
->family_num
+= super
->random
;
7276 /* count arrays using the victim in the metadata */
7278 for (a
= active_array
; a
; a
= a
->next
) {
7279 dev
= get_imsm_dev(super
, a
->info
.container_member
);
7280 map
= get_imsm_map(dev
, 0);
7282 if (get_imsm_disk_slot(map
, victim
) >= 0)
7286 /* delete the victim if it is no longer being
7292 /* We know that 'manager' isn't touching anything,
7293 * so it is safe to delete
7295 for (dlp
= &super
->disks
; *dlp
; dlp
= &(*dlp
)->next
)
7296 if ((*dlp
)->index
== victim
)
7299 /* victim may be on the missing list */
7301 for (dlp
= &super
->missing
; *dlp
;
7302 dlp
= &(*dlp
)->next
)
7303 if ((*dlp
)->index
== victim
)
7305 imsm_delete(super
, dlp
, victim
);
7312 static int apply_reshape_container_disks_update(struct imsm_update_reshape
*u
,
7313 struct intel_super
*super
,
7316 struct dl
*new_disk
;
7317 struct intel_dev
*id
;
7319 int delta_disks
= u
->new_raid_disks
- u
->old_raid_disks
;
7320 int disk_count
= u
->old_raid_disks
;
7321 void **tofree
= NULL
;
7322 int devices_to_reshape
= 1;
7323 struct imsm_super
*mpb
= super
->anchor
;
7325 unsigned int dev_id
;
7327 dprintf("imsm: apply_reshape_container_disks_update()\n");
7329 /* enable spares to use in array */
7330 for (i
= 0; i
< delta_disks
; i
++) {
7331 new_disk
= get_disk_super(super
,
7332 major(u
->new_disks
[i
]),
7333 minor(u
->new_disks
[i
]));
7334 dprintf("imsm: new disk for reshape is: %i:%i "
7335 "(%p, index = %i)\n",
7336 major(u
->new_disks
[i
]), minor(u
->new_disks
[i
]),
7337 new_disk
, new_disk
->index
);
7338 if ((new_disk
== NULL
) ||
7339 ((new_disk
->index
>= 0) &&
7340 (new_disk
->index
< u
->old_raid_disks
)))
7341 goto update_reshape_exit
;
7342 new_disk
->index
= disk_count
++;
7343 /* slot to fill in autolayout
7345 new_disk
->raiddisk
= new_disk
->index
;
7346 new_disk
->disk
.status
|=
7348 new_disk
->disk
.status
&= ~SPARE_DISK
;
7351 dprintf("imsm: volume set mpb->num_raid_devs = %i\n",
7352 mpb
->num_raid_devs
);
7353 /* manage changes in volume
7355 for (dev_id
= 0; dev_id
< mpb
->num_raid_devs
; dev_id
++) {
7356 void **sp
= *space_list
;
7357 struct imsm_dev
*newdev
;
7358 struct imsm_map
*newmap
, *oldmap
;
7360 for (id
= super
->devlist
; id
; id
= id
->next
) {
7361 if (id
->index
== dev_id
)
7370 /* Copy the dev, but not (all of) the map */
7371 memcpy(newdev
, id
->dev
, sizeof(*newdev
));
7372 oldmap
= get_imsm_map(id
->dev
, 0);
7373 newmap
= get_imsm_map(newdev
, 0);
7374 /* Copy the current map */
7375 memcpy(newmap
, oldmap
, sizeof_imsm_map(oldmap
));
7376 /* update one device only
7378 if (devices_to_reshape
) {
7379 dprintf("imsm: modifying subdev: %i\n",
7381 devices_to_reshape
--;
7382 newdev
->vol
.migr_state
= 1;
7383 newdev
->vol
.curr_migr_unit
= 0;
7384 set_migr_type(newdev
, MIGR_GEN_MIGR
);
7385 newmap
->num_members
= u
->new_raid_disks
;
7386 for (i
= 0; i
< delta_disks
; i
++) {
7387 set_imsm_ord_tbl_ent(newmap
,
7388 u
->old_raid_disks
+ i
,
7389 u
->old_raid_disks
+ i
);
7391 /* New map is correct, now need to save old map
7393 newmap
= get_imsm_map(newdev
, 1);
7394 memcpy(newmap
, oldmap
, sizeof_imsm_map(oldmap
));
7396 imsm_set_array_size(newdev
);
7399 sp
= (void **)id
->dev
;
7404 /* Clear migration record */
7405 memset(super
->migr_rec
, 0, sizeof(struct migr_record
));
7408 *space_list
= tofree
;
7411 update_reshape_exit
:
7416 static int apply_takeover_update(struct imsm_update_takeover
*u
,
7417 struct intel_super
*super
,
7420 struct imsm_dev
*dev
= NULL
;
7421 struct intel_dev
*dv
;
7422 struct imsm_dev
*dev_new
;
7423 struct imsm_map
*map
;
7427 for (dv
= super
->devlist
; dv
; dv
= dv
->next
)
7428 if (dv
->index
== (unsigned int)u
->subarray
) {
7436 map
= get_imsm_map(dev
, 0);
7438 if (u
->direction
== R10_TO_R0
) {
7439 /* Number of failed disks must be half of initial disk number */
7440 if (imsm_count_failed(super
, dev
, MAP_0
) !=
7441 (map
->num_members
/ 2))
7444 /* iterate through devices to mark removed disks as spare */
7445 for (dm
= super
->disks
; dm
; dm
= dm
->next
) {
7446 if (dm
->disk
.status
& FAILED_DISK
) {
7447 int idx
= dm
->index
;
7448 /* update indexes on the disk list */
7449 /* FIXME this loop-with-the-loop looks wrong, I'm not convinced
7450 the index values will end up being correct.... NB */
7451 for (du
= super
->disks
; du
; du
= du
->next
)
7452 if (du
->index
> idx
)
7454 /* mark as spare disk */
7459 map
->num_members
= map
->num_members
/ 2;
7460 map
->map_state
= IMSM_T_STATE_NORMAL
;
7461 map
->num_domains
= 1;
7462 map
->raid_level
= 0;
7463 map
->failed_disk_num
= -1;
7466 if (u
->direction
== R0_TO_R10
) {
7468 /* update slots in current disk list */
7469 for (dm
= super
->disks
; dm
; dm
= dm
->next
) {
7473 /* create new *missing* disks */
7474 for (i
= 0; i
< map
->num_members
; i
++) {
7475 space
= *space_list
;
7478 *space_list
= *space
;
7480 memcpy(du
, super
->disks
, sizeof(*du
));
7484 du
->index
= (i
* 2) + 1;
7485 sprintf((char *)du
->disk
.serial
,
7486 " MISSING_%d", du
->index
);
7487 sprintf((char *)du
->serial
,
7488 "MISSING_%d", du
->index
);
7489 du
->next
= super
->missing
;
7490 super
->missing
= du
;
7492 /* create new dev and map */
7493 space
= *space_list
;
7496 *space_list
= *space
;
7497 dev_new
= (void *)space
;
7498 memcpy(dev_new
, dev
, sizeof(*dev
));
7499 /* update new map */
7500 map
= get_imsm_map(dev_new
, 0);
7501 map
->num_members
= map
->num_members
* 2;
7502 map
->map_state
= IMSM_T_STATE_DEGRADED
;
7503 map
->num_domains
= 2;
7504 map
->raid_level
= 1;
7505 /* replace dev<->dev_new */
7508 /* update disk order table */
7509 for (du
= super
->disks
; du
; du
= du
->next
)
7511 set_imsm_ord_tbl_ent(map
, du
->index
, du
->index
);
7512 for (du
= super
->missing
; du
; du
= du
->next
)
7513 if (du
->index
>= 0) {
7514 set_imsm_ord_tbl_ent(map
, du
->index
, du
->index
);
7515 mark_missing(dv
->dev
, &du
->disk
, du
->index
);
7521 static void imsm_process_update(struct supertype
*st
,
7522 struct metadata_update
*update
)
7525 * crack open the metadata_update envelope to find the update record
7526 * update can be one of:
7527 * update_reshape_container_disks - all the arrays in the container
7528 * are being reshaped to have more devices. We need to mark
7529 * the arrays for general migration and convert selected spares
7530 * into active devices.
7531 * update_activate_spare - a spare device has replaced a failed
7532 * device in an array, update the disk_ord_tbl. If this disk is
7533 * present in all member arrays then also clear the SPARE_DISK
7535 * update_create_array
7537 * update_rename_array
7538 * update_add_remove_disk
7540 struct intel_super
*super
= st
->sb
;
7541 struct imsm_super
*mpb
;
7542 enum imsm_update_type type
= *(enum imsm_update_type
*) update
->buf
;
7544 /* update requires a larger buf but the allocation failed */
7545 if (super
->next_len
&& !super
->next_buf
) {
7546 super
->next_len
= 0;
7550 if (super
->next_buf
) {
7551 memcpy(super
->next_buf
, super
->buf
, super
->len
);
7553 super
->len
= super
->next_len
;
7554 super
->buf
= super
->next_buf
;
7556 super
->next_len
= 0;
7557 super
->next_buf
= NULL
;
7560 mpb
= super
->anchor
;
7563 case update_general_migration_checkpoint
: {
7564 struct intel_dev
*id
;
7565 struct imsm_update_general_migration_checkpoint
*u
=
7566 (void *)update
->buf
;
7568 dprintf("imsm: process_update() "
7569 "for update_general_migration_checkpoint called\n");
7571 /* find device under general migration */
7572 for (id
= super
->devlist
; id
; id
= id
->next
) {
7573 if (is_gen_migration(id
->dev
)) {
7574 id
->dev
->vol
.curr_migr_unit
=
7575 __cpu_to_le32(u
->curr_migr_unit
);
7576 super
->updates_pending
++;
7581 case update_takeover
: {
7582 struct imsm_update_takeover
*u
= (void *)update
->buf
;
7583 if (apply_takeover_update(u
, super
, &update
->space_list
)) {
7584 imsm_update_version_info(super
);
7585 super
->updates_pending
++;
7590 case update_reshape_container_disks
: {
7591 struct imsm_update_reshape
*u
= (void *)update
->buf
;
7592 if (apply_reshape_container_disks_update(
7593 u
, super
, &update
->space_list
))
7594 super
->updates_pending
++;
7597 case update_reshape_migration
: {
7598 struct imsm_update_reshape_migration
*u
= (void *)update
->buf
;
7599 if (apply_reshape_migration_update(
7600 u
, super
, &update
->space_list
))
7601 super
->updates_pending
++;
7604 case update_activate_spare
: {
7605 struct imsm_update_activate_spare
*u
= (void *) update
->buf
;
7606 if (apply_update_activate_spare(u
, super
, st
->arrays
))
7607 super
->updates_pending
++;
7610 case update_create_array
: {
7611 /* someone wants to create a new array, we need to be aware of
7612 * a few races/collisions:
7613 * 1/ 'Create' called by two separate instances of mdadm
7614 * 2/ 'Create' versus 'activate_spare': mdadm has chosen
7615 * devices that have since been assimilated via
7617 * In the event this update can not be carried out mdadm will
7618 * (FIX ME) notice that its update did not take hold.
7620 struct imsm_update_create_array
*u
= (void *) update
->buf
;
7621 struct intel_dev
*dv
;
7622 struct imsm_dev
*dev
;
7623 struct imsm_map
*map
, *new_map
;
7624 unsigned long long start
, end
;
7625 unsigned long long new_start
, new_end
;
7627 struct disk_info
*inf
;
7630 /* handle racing creates: first come first serve */
7631 if (u
->dev_idx
< mpb
->num_raid_devs
) {
7632 dprintf("%s: subarray %d already defined\n",
7633 __func__
, u
->dev_idx
);
7637 /* check update is next in sequence */
7638 if (u
->dev_idx
!= mpb
->num_raid_devs
) {
7639 dprintf("%s: can not create array %d expected index %d\n",
7640 __func__
, u
->dev_idx
, mpb
->num_raid_devs
);
7644 new_map
= get_imsm_map(&u
->dev
, 0);
7645 new_start
= __le32_to_cpu(new_map
->pba_of_lba0
);
7646 new_end
= new_start
+ __le32_to_cpu(new_map
->blocks_per_member
);
7647 inf
= get_disk_info(u
);
7649 /* handle activate_spare versus create race:
7650 * check to make sure that overlapping arrays do not include
7653 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
7654 dev
= get_imsm_dev(super
, i
);
7655 map
= get_imsm_map(dev
, 0);
7656 start
= __le32_to_cpu(map
->pba_of_lba0
);
7657 end
= start
+ __le32_to_cpu(map
->blocks_per_member
);
7658 if ((new_start
>= start
&& new_start
<= end
) ||
7659 (start
>= new_start
&& start
<= new_end
))
7664 if (disks_overlap(super
, i
, u
)) {
7665 dprintf("%s: arrays overlap\n", __func__
);
7670 /* check that prepare update was successful */
7671 if (!update
->space
) {
7672 dprintf("%s: prepare update failed\n", __func__
);
7676 /* check that all disks are still active before committing
7677 * changes. FIXME: could we instead handle this by creating a
7678 * degraded array? That's probably not what the user expects,
7679 * so better to drop this update on the floor.
7681 for (i
= 0; i
< new_map
->num_members
; i
++) {
7682 dl
= serial_to_dl(inf
[i
].serial
, super
);
7684 dprintf("%s: disk disappeared\n", __func__
);
7689 super
->updates_pending
++;
7691 /* convert spares to members and fixup ord_tbl */
7692 for (i
= 0; i
< new_map
->num_members
; i
++) {
7693 dl
= serial_to_dl(inf
[i
].serial
, super
);
7694 if (dl
->index
== -1) {
7695 dl
->index
= mpb
->num_disks
;
7697 dl
->disk
.status
|= CONFIGURED_DISK
;
7698 dl
->disk
.status
&= ~SPARE_DISK
;
7700 set_imsm_ord_tbl_ent(new_map
, i
, dl
->index
);
7705 update
->space
= NULL
;
7706 imsm_copy_dev(dev
, &u
->dev
);
7707 dv
->index
= u
->dev_idx
;
7708 dv
->next
= super
->devlist
;
7709 super
->devlist
= dv
;
7710 mpb
->num_raid_devs
++;
7712 imsm_update_version_info(super
);
7715 /* mdmon knows how to release update->space, but not
7716 * ((struct intel_dev *) update->space)->dev
7718 if (update
->space
) {
7724 case update_kill_array
: {
7725 struct imsm_update_kill_array
*u
= (void *) update
->buf
;
7726 int victim
= u
->dev_idx
;
7727 struct active_array
*a
;
7728 struct intel_dev
**dp
;
7729 struct imsm_dev
*dev
;
7731 /* sanity check that we are not affecting the uuid of
7732 * active arrays, or deleting an active array
7734 * FIXME when immutable ids are available, but note that
7735 * we'll also need to fixup the invalidated/active
7736 * subarray indexes in mdstat
7738 for (a
= st
->arrays
; a
; a
= a
->next
)
7739 if (a
->info
.container_member
>= victim
)
7741 /* by definition if mdmon is running at least one array
7742 * is active in the container, so checking
7743 * mpb->num_raid_devs is just extra paranoia
7745 dev
= get_imsm_dev(super
, victim
);
7746 if (a
|| !dev
|| mpb
->num_raid_devs
== 1) {
7747 dprintf("failed to delete subarray-%d\n", victim
);
7751 for (dp
= &super
->devlist
; *dp
;)
7752 if ((*dp
)->index
== (unsigned)super
->current_vol
) {
7755 if ((*dp
)->index
> (unsigned)victim
)
7759 mpb
->num_raid_devs
--;
7760 super
->updates_pending
++;
7763 case update_rename_array
: {
7764 struct imsm_update_rename_array
*u
= (void *) update
->buf
;
7765 char name
[MAX_RAID_SERIAL_LEN
+1];
7766 int target
= u
->dev_idx
;
7767 struct active_array
*a
;
7768 struct imsm_dev
*dev
;
7770 /* sanity check that we are not affecting the uuid of
7773 snprintf(name
, MAX_RAID_SERIAL_LEN
, "%s", (char *) u
->name
);
7774 name
[MAX_RAID_SERIAL_LEN
] = '\0';
7775 for (a
= st
->arrays
; a
; a
= a
->next
)
7776 if (a
->info
.container_member
== target
)
7778 dev
= get_imsm_dev(super
, u
->dev_idx
);
7779 if (a
|| !dev
|| !check_name(super
, name
, 1)) {
7780 dprintf("failed to rename subarray-%d\n", target
);
7784 snprintf((char *) dev
->volume
, MAX_RAID_SERIAL_LEN
, "%s", name
);
7785 super
->updates_pending
++;
7788 case update_add_remove_disk
: {
7789 /* we may be able to repair some arrays if disks are
7790 * being added, check teh status of add_remove_disk
7791 * if discs has been added.
7793 if (add_remove_disk_update(super
)) {
7794 struct active_array
*a
;
7796 super
->updates_pending
++;
7797 for (a
= st
->arrays
; a
; a
= a
->next
)
7798 a
->check_degraded
= 1;
7803 fprintf(stderr
, "error: unsuported process update type:"
7804 "(type: %d)\n", type
);
7808 static struct mdinfo
*get_spares_for_grow(struct supertype
*st
);
7810 static void imsm_prepare_update(struct supertype
*st
,
7811 struct metadata_update
*update
)
7814 * Allocate space to hold new disk entries, raid-device entries or a new
7815 * mpb if necessary. The manager synchronously waits for updates to
7816 * complete in the monitor, so new mpb buffers allocated here can be
7817 * integrated by the monitor thread without worrying about live pointers
7818 * in the manager thread.
7820 enum imsm_update_type type
= *(enum imsm_update_type
*) update
->buf
;
7821 struct intel_super
*super
= st
->sb
;
7822 struct imsm_super
*mpb
= super
->anchor
;
7827 case update_general_migration_checkpoint
:
7828 dprintf("imsm: prepare_update() "
7829 "for update_general_migration_checkpoint called\n");
7831 case update_takeover
: {
7832 struct imsm_update_takeover
*u
= (void *)update
->buf
;
7833 if (u
->direction
== R0_TO_R10
) {
7834 void **tail
= (void **)&update
->space_list
;
7835 struct imsm_dev
*dev
= get_imsm_dev(super
, u
->subarray
);
7836 struct imsm_map
*map
= get_imsm_map(dev
, 0);
7837 int num_members
= map
->num_members
;
7841 /* allocate memory for added disks */
7842 for (i
= 0; i
< num_members
; i
++) {
7843 size
= sizeof(struct dl
);
7844 space
= malloc(size
);
7853 /* allocate memory for new device */
7854 size
= sizeof_imsm_dev(super
->devlist
->dev
, 0) +
7855 (num_members
* sizeof(__u32
));
7856 space
= malloc(size
);
7865 len
= disks_to_mpb_size(num_members
* 2);
7867 /* if allocation didn't success, free buffer */
7868 while (update
->space_list
) {
7869 void **sp
= update
->space_list
;
7870 update
->space_list
= *sp
;
7878 case update_reshape_container_disks
: {
7879 /* Every raid device in the container is about to
7880 * gain some more devices, and we will enter a
7882 * So each 'imsm_map' will be bigger, and the imsm_vol
7883 * will now hold 2 of them.
7884 * Thus we need new 'struct imsm_dev' allocations sized
7885 * as sizeof_imsm_dev but with more devices in both maps.
7887 struct imsm_update_reshape
*u
= (void *)update
->buf
;
7888 struct intel_dev
*dl
;
7889 void **space_tail
= (void**)&update
->space_list
;
7891 dprintf("imsm: imsm_prepare_update() for update_reshape\n");
7893 for (dl
= super
->devlist
; dl
; dl
= dl
->next
) {
7894 int size
= sizeof_imsm_dev(dl
->dev
, 1);
7896 if (u
->new_raid_disks
> u
->old_raid_disks
)
7897 size
+= sizeof(__u32
)*2*
7898 (u
->new_raid_disks
- u
->old_raid_disks
);
7907 len
= disks_to_mpb_size(u
->new_raid_disks
);
7908 dprintf("New anchor length is %llu\n", (unsigned long long)len
);
7911 case update_reshape_migration
: {
7912 /* for migration level 0->5 we need to add disks
7913 * so the same as for container operation we will copy
7914 * device to the bigger location.
7915 * in memory prepared device and new disk area are prepared
7916 * for usage in process update
7918 struct imsm_update_reshape_migration
*u
= (void *)update
->buf
;
7919 struct intel_dev
*id
;
7920 void **space_tail
= (void **)&update
->space_list
;
7923 int current_level
= -1;
7925 dprintf("imsm: imsm_prepare_update() for update_reshape\n");
7927 /* add space for bigger array in update
7929 for (id
= super
->devlist
; id
; id
= id
->next
) {
7930 if (id
->index
== (unsigned)u
->subdev
) {
7931 size
= sizeof_imsm_dev(id
->dev
, 1);
7932 if (u
->new_raid_disks
> u
->old_raid_disks
)
7933 size
+= sizeof(__u32
)*2*
7934 (u
->new_raid_disks
- u
->old_raid_disks
);
7944 if (update
->space_list
== NULL
)
7947 /* add space for disk in update
7949 size
= sizeof(struct dl
);
7952 free(update
->space_list
);
7953 update
->space_list
= NULL
;
7960 /* add spare device to update
7962 for (id
= super
->devlist
; id
; id
= id
->next
)
7963 if (id
->index
== (unsigned)u
->subdev
) {
7964 struct imsm_dev
*dev
;
7965 struct imsm_map
*map
;
7967 dev
= get_imsm_dev(super
, u
->subdev
);
7968 map
= get_imsm_map(dev
, 0);
7969 current_level
= map
->raid_level
;
7972 if ((u
->new_level
== 5) && (u
->new_level
!= current_level
)) {
7973 struct mdinfo
*spares
;
7975 spares
= get_spares_for_grow(st
);
7983 makedev(dev
->disk
.major
,
7985 dl
= get_disk_super(super
,
7988 dl
->index
= u
->old_raid_disks
;
7994 len
= disks_to_mpb_size(u
->new_raid_disks
);
7995 dprintf("New anchor length is %llu\n", (unsigned long long)len
);
7998 case update_create_array
: {
7999 struct imsm_update_create_array
*u
= (void *) update
->buf
;
8000 struct intel_dev
*dv
;
8001 struct imsm_dev
*dev
= &u
->dev
;
8002 struct imsm_map
*map
= get_imsm_map(dev
, 0);
8004 struct disk_info
*inf
;
8008 inf
= get_disk_info(u
);
8009 len
= sizeof_imsm_dev(dev
, 1);
8010 /* allocate a new super->devlist entry */
8011 dv
= malloc(sizeof(*dv
));
8013 dv
->dev
= malloc(len
);
8018 update
->space
= NULL
;
8022 /* count how many spares will be converted to members */
8023 for (i
= 0; i
< map
->num_members
; i
++) {
8024 dl
= serial_to_dl(inf
[i
].serial
, super
);
8026 /* hmm maybe it failed?, nothing we can do about
8031 if (count_memberships(dl
, super
) == 0)
8034 len
+= activate
* sizeof(struct imsm_disk
);
8041 /* check if we need a larger metadata buffer */
8042 if (super
->next_buf
)
8043 buf_len
= super
->next_len
;
8045 buf_len
= super
->len
;
8047 if (__le32_to_cpu(mpb
->mpb_size
) + len
> buf_len
) {
8048 /* ok we need a larger buf than what is currently allocated
8049 * if this allocation fails process_update will notice that
8050 * ->next_len is set and ->next_buf is NULL
8052 buf_len
= ROUND_UP(__le32_to_cpu(mpb
->mpb_size
) + len
, 512);
8053 if (super
->next_buf
)
8054 free(super
->next_buf
);
8056 super
->next_len
= buf_len
;
8057 if (posix_memalign(&super
->next_buf
, 512, buf_len
) == 0)
8058 memset(super
->next_buf
, 0, buf_len
);
8060 super
->next_buf
= NULL
;
8064 /* must be called while manager is quiesced */
8065 static void imsm_delete(struct intel_super
*super
, struct dl
**dlp
, unsigned index
)
8067 struct imsm_super
*mpb
= super
->anchor
;
8069 struct imsm_dev
*dev
;
8070 struct imsm_map
*map
;
8071 int i
, j
, num_members
;
8074 dprintf("%s: deleting device[%d] from imsm_super\n",
8077 /* shift all indexes down one */
8078 for (iter
= super
->disks
; iter
; iter
= iter
->next
)
8079 if (iter
->index
> (int)index
)
8081 for (iter
= super
->missing
; iter
; iter
= iter
->next
)
8082 if (iter
->index
> (int)index
)
8085 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
8086 dev
= get_imsm_dev(super
, i
);
8087 map
= get_imsm_map(dev
, 0);
8088 num_members
= map
->num_members
;
8089 for (j
= 0; j
< num_members
; j
++) {
8090 /* update ord entries being careful not to propagate
8091 * ord-flags to the first map
8093 ord
= get_imsm_ord_tbl_ent(dev
, j
, -1);
8095 if (ord_to_idx(ord
) <= index
)
8098 map
= get_imsm_map(dev
, 0);
8099 set_imsm_ord_tbl_ent(map
, j
, ord_to_idx(ord
- 1));
8100 map
= get_imsm_map(dev
, 1);
8102 set_imsm_ord_tbl_ent(map
, j
, ord
- 1);
8107 super
->updates_pending
++;
8109 struct dl
*dl
= *dlp
;
8111 *dlp
= (*dlp
)->next
;
8112 __free_imsm_disk(dl
);
8115 #endif /* MDASSEMBLE */
8117 static void close_targets(int *targets
, int new_disks
)
8124 for (i
= 0; i
< new_disks
; i
++) {
8125 if (targets
[i
] >= 0) {
8132 static int imsm_get_allowed_degradation(int level
, int raid_disks
,
8133 struct intel_super
*super
,
8134 struct imsm_dev
*dev
)
8139 struct imsm_map
*map
;
8142 ret_val
= raid_disks
/2;
8143 /* check map if all disks pairs not failed
8146 map
= get_imsm_map(dev
, 0);
8147 for (i
= 0; i
< ret_val
; i
++) {
8148 int degradation
= 0;
8149 if (get_imsm_disk(super
, i
) == NULL
)
8151 if (get_imsm_disk(super
, i
+ 1) == NULL
)
8153 if (degradation
== 2)
8156 map
= get_imsm_map(dev
, 1);
8157 /* if there is no second map
8158 * result can be returned
8162 /* check degradation in second map
8164 for (i
= 0; i
< ret_val
; i
++) {
8165 int degradation
= 0;
8166 if (get_imsm_disk(super
, i
) == NULL
)
8168 if (get_imsm_disk(super
, i
+ 1) == NULL
)
8170 if (degradation
== 2)
8185 /*******************************************************************************
8186 * Function: open_backup_targets
8187 * Description: Function opens file descriptors for all devices given in
8190 * info : general array info
8191 * raid_disks : number of disks
8192 * raid_fds : table of device's file descriptors
8193 * super : intel super for raid10 degradation check
8194 * dev : intel device for raid10 degradation check
8198 ******************************************************************************/
8199 int open_backup_targets(struct mdinfo
*info
, int raid_disks
, int *raid_fds
,
8200 struct intel_super
*super
, struct imsm_dev
*dev
)
8206 for (i
= 0; i
< raid_disks
; i
++)
8209 for (sd
= info
->devs
; sd
; sd
= sd
->next
) {
8212 if (sd
->disk
.state
& (1<<MD_DISK_FAULTY
)) {
8213 dprintf("disk is faulty!!\n");
8217 if ((sd
->disk
.raid_disk
>= raid_disks
) ||
8218 (sd
->disk
.raid_disk
< 0))
8221 dn
= map_dev(sd
->disk
.major
,
8223 raid_fds
[sd
->disk
.raid_disk
] = dev_open(dn
, O_RDWR
);
8224 if (raid_fds
[sd
->disk
.raid_disk
] < 0) {
8225 fprintf(stderr
, "cannot open component\n");
8230 /* check if maximum array degradation level is not exceeded
8232 if ((raid_disks
- opened
) >
8233 imsm_get_allowed_degradation(info
->new_level
,
8236 fprintf(stderr
, "Not enough disks can be opened.\n");
8237 close_targets(raid_fds
, raid_disks
);
8244 /*******************************************************************************
8245 * Function: init_migr_record_imsm
8246 * Description: Function inits imsm migration record
8248 * super : imsm internal array info
8249 * dev : device under migration
8250 * info : general array info to find the smallest device
8253 ******************************************************************************/
8254 void init_migr_record_imsm(struct supertype
*st
, struct imsm_dev
*dev
,
8255 struct mdinfo
*info
)
8257 struct intel_super
*super
= st
->sb
;
8258 struct migr_record
*migr_rec
= super
->migr_rec
;
8260 unsigned long long dsize
, dev_sectors
;
8261 long long unsigned min_dev_sectors
= -1LLU;
8265 struct imsm_map
*map_dest
= get_imsm_map(dev
, 0);
8266 struct imsm_map
*map_src
= get_imsm_map(dev
, 1);
8267 unsigned long long num_migr_units
;
8268 unsigned long long array_blocks
;
8270 memset(migr_rec
, 0, sizeof(struct migr_record
));
8271 migr_rec
->family_num
= __cpu_to_le32(super
->anchor
->family_num
);
8273 /* only ascending reshape supported now */
8274 migr_rec
->ascending_migr
= __cpu_to_le32(1);
8276 migr_rec
->dest_depth_per_unit
= GEN_MIGR_AREA_SIZE
/
8277 max(map_dest
->blocks_per_strip
, map_src
->blocks_per_strip
);
8278 migr_rec
->dest_depth_per_unit
*= map_dest
->blocks_per_strip
;
8279 new_data_disks
= imsm_num_data_members(dev
, 0);
8280 migr_rec
->blocks_per_unit
=
8281 __cpu_to_le32(migr_rec
->dest_depth_per_unit
* new_data_disks
);
8282 migr_rec
->dest_depth_per_unit
=
8283 __cpu_to_le32(migr_rec
->dest_depth_per_unit
);
8284 array_blocks
= info
->component_size
* new_data_disks
;
8286 array_blocks
/ __le32_to_cpu(migr_rec
->blocks_per_unit
);
8288 if (array_blocks
% __le32_to_cpu(migr_rec
->blocks_per_unit
))
8290 migr_rec
->num_migr_units
= __cpu_to_le32(num_migr_units
);
8292 migr_rec
->post_migr_vol_cap
= dev
->size_low
;
8293 migr_rec
->post_migr_vol_cap_hi
= dev
->size_high
;
8296 /* Find the smallest dev */
8297 for (sd
= info
->devs
; sd
; sd
= sd
->next
) {
8298 sprintf(nm
, "%d:%d", sd
->disk
.major
, sd
->disk
.minor
);
8299 fd
= dev_open(nm
, O_RDONLY
);
8302 get_dev_size(fd
, NULL
, &dsize
);
8303 dev_sectors
= dsize
/ 512;
8304 if (dev_sectors
< min_dev_sectors
)
8305 min_dev_sectors
= dev_sectors
;
8308 migr_rec
->ckpt_area_pba
= __cpu_to_le32(min_dev_sectors
-
8309 RAID_DISK_RESERVED_BLOCKS_IMSM_HI
);
8311 write_imsm_migr_rec(st
);
8316 /*******************************************************************************
8317 * Function: save_backup_imsm
8318 * Description: Function saves critical data stripes to Migration Copy Area
8319 * and updates the current migration unit status.
8320 * Use restore_stripes() to form a destination stripe,
8321 * and to write it to the Copy Area.
8323 * st : supertype information
8324 * dev : imsm device that backup is saved for
8325 * info : general array info
8326 * buf : input buffer
8327 * length : length of data to backup (blocks_per_unit)
8331 ******************************************************************************/
8332 int save_backup_imsm(struct supertype
*st
,
8333 struct imsm_dev
*dev
,
8334 struct mdinfo
*info
,
8339 struct intel_super
*super
= st
->sb
;
8340 unsigned long long *target_offsets
= NULL
;
8341 int *targets
= NULL
;
8343 struct imsm_map
*map_dest
= get_imsm_map(dev
, 0);
8344 int new_disks
= map_dest
->num_members
;
8345 int dest_layout
= 0;
8347 unsigned long long start
;
8348 int data_disks
= imsm_num_data_members(dev
, 0);
8350 targets
= malloc(new_disks
* sizeof(int));
8354 for (i
= 0; i
< new_disks
; i
++)
8357 target_offsets
= malloc(new_disks
* sizeof(unsigned long long));
8358 if (!target_offsets
)
8361 start
= info
->reshape_progress
* 512;
8362 for (i
= 0; i
< new_disks
; i
++) {
8363 target_offsets
[i
] = (unsigned long long)
8364 __le32_to_cpu(super
->migr_rec
->ckpt_area_pba
) * 512;
8365 /* move back copy area adderss, it will be moved forward
8366 * in restore_stripes() using start input variable
8368 target_offsets
[i
] -= start
/data_disks
;
8371 if (open_backup_targets(info
, new_disks
, targets
,
8375 dest_layout
= imsm_level_to_layout(map_dest
->raid_level
);
8376 dest_chunk
= __le16_to_cpu(map_dest
->blocks_per_strip
) * 512;
8378 if (restore_stripes(targets
, /* list of dest devices */
8379 target_offsets
, /* migration record offsets */
8382 map_dest
->raid_level
,
8384 -1, /* source backup file descriptor */
8385 0, /* input buf offset
8386 * always 0 buf is already offseted */
8390 fprintf(stderr
, Name
": Error restoring stripes\n");
8398 close_targets(targets
, new_disks
);
8401 free(target_offsets
);
8406 /*******************************************************************************
8407 * Function: save_checkpoint_imsm
8408 * Description: Function called for current unit status update
8409 * in the migration record. It writes it to disk.
8411 * super : imsm internal array info
8412 * info : general array info
8416 * 2: failure, means no valid migration record
8417 * / no general migration in progress /
8418 ******************************************************************************/
8419 int save_checkpoint_imsm(struct supertype
*st
, struct mdinfo
*info
, int state
)
8421 struct intel_super
*super
= st
->sb
;
8422 unsigned long long blocks_per_unit
;
8423 unsigned long long curr_migr_unit
;
8425 if (load_imsm_migr_rec(super
, info
) != 0) {
8426 dprintf("imsm: ERROR: Cannot read migration record "
8427 "for checkpoint save.\n");
8431 blocks_per_unit
= __le32_to_cpu(super
->migr_rec
->blocks_per_unit
);
8432 if (blocks_per_unit
== 0) {
8433 dprintf("imsm: no migration in progress.\n");
8436 curr_migr_unit
= info
->reshape_progress
/ blocks_per_unit
;
8437 /* check if array is alligned to copy area
8438 * if it is not alligned, add one to current migration unit value
8439 * this can happend on array reshape finish only
8441 if (info
->reshape_progress
% blocks_per_unit
)
8444 super
->migr_rec
->curr_migr_unit
=
8445 __cpu_to_le32(curr_migr_unit
);
8446 super
->migr_rec
->rec_status
= __cpu_to_le32(state
);
8447 super
->migr_rec
->dest_1st_member_lba
=
8448 __cpu_to_le32(curr_migr_unit
*
8449 __le32_to_cpu(super
->migr_rec
->dest_depth_per_unit
));
8450 if (write_imsm_migr_rec(st
) < 0) {
8451 dprintf("imsm: Cannot write migration record "
8452 "outside backup area\n");
8459 /*******************************************************************************
8460 * Function: recover_backup_imsm
8461 * Description: Function recovers critical data from the Migration Copy Area
8462 * while assembling an array.
8464 * super : imsm internal array info
8465 * info : general array info
8467 * 0 : success (or there is no data to recover)
8469 ******************************************************************************/
8470 int recover_backup_imsm(struct supertype
*st
, struct mdinfo
*info
)
8472 struct intel_super
*super
= st
->sb
;
8473 struct migr_record
*migr_rec
= super
->migr_rec
;
8474 struct imsm_map
*map_dest
= NULL
;
8475 struct intel_dev
*id
= NULL
;
8476 unsigned long long read_offset
;
8477 unsigned long long write_offset
;
8479 int *targets
= NULL
;
8480 int new_disks
, i
, err
;
8483 unsigned long curr_migr_unit
= __le32_to_cpu(migr_rec
->curr_migr_unit
);
8484 unsigned long num_migr_units
= __le32_to_cpu(migr_rec
->num_migr_units
);
8486 int skipped_disks
= 0;
8488 err
= sysfs_get_str(info
, NULL
, "array_state", (char *)buffer
, 20);
8492 /* recover data only during assemblation */
8493 if (strncmp(buffer
, "inactive", 8) != 0)
8495 /* no data to recover */
8496 if (__le32_to_cpu(migr_rec
->rec_status
) == UNIT_SRC_NORMAL
)
8498 if (curr_migr_unit
>= num_migr_units
)
8501 /* find device during reshape */
8502 for (id
= super
->devlist
; id
; id
= id
->next
)
8503 if (is_gen_migration(id
->dev
))
8508 map_dest
= get_imsm_map(id
->dev
, 0);
8509 new_disks
= map_dest
->num_members
;
8511 read_offset
= (unsigned long long)
8512 __le32_to_cpu(migr_rec
->ckpt_area_pba
) * 512;
8514 write_offset
= ((unsigned long long)
8515 __le32_to_cpu(migr_rec
->dest_1st_member_lba
) +
8516 __le32_to_cpu(map_dest
->pba_of_lba0
)) * 512;
8518 unit_len
= __le32_to_cpu(migr_rec
->dest_depth_per_unit
) * 512;
8519 if (posix_memalign((void **)&buf
, 512, unit_len
) != 0)
8521 targets
= malloc(new_disks
* sizeof(int));
8525 if (open_backup_targets(info
, new_disks
, targets
, super
, id
->dev
)) {
8527 Name
": Cannot open some devices belonging to array.\n");
8531 for (i
= 0; i
< new_disks
; i
++) {
8532 if (targets
[i
] < 0) {
8536 if (lseek64(targets
[i
], read_offset
, SEEK_SET
) < 0) {
8538 Name
": Cannot seek to block: %s\n",
8543 if ((unsigned)read(targets
[i
], buf
, unit_len
) != unit_len
) {
8545 Name
": Cannot read copy area block: %s\n",
8550 if (lseek64(targets
[i
], write_offset
, SEEK_SET
) < 0) {
8552 Name
": Cannot seek to block: %s\n",
8557 if ((unsigned)write(targets
[i
], buf
, unit_len
) != unit_len
) {
8559 Name
": Cannot restore block: %s\n",
8566 if (skipped_disks
> imsm_get_allowed_degradation(info
->new_level
,
8571 Name
": Cannot restore data from backup."
8572 " Too many failed disks\n");
8576 if (save_checkpoint_imsm(st
, info
, UNIT_SRC_NORMAL
)) {
8577 /* ignore error == 2, this can mean end of reshape here
8579 dprintf("imsm: Cannot write checkpoint to "
8580 "migration record (UNIT_SRC_NORMAL) during restart\n");
8586 for (i
= 0; i
< new_disks
; i
++)
8595 static char disk_by_path
[] = "/dev/disk/by-path/";
8597 static const char *imsm_get_disk_controller_domain(const char *path
)
8599 char disk_path
[PATH_MAX
];
8603 strncpy(disk_path
, disk_by_path
, PATH_MAX
- 1);
8604 strncat(disk_path
, path
, PATH_MAX
- strlen(disk_path
) - 1);
8605 if (stat(disk_path
, &st
) == 0) {
8606 struct sys_dev
* hba
;
8609 path
= devt_to_devpath(st
.st_rdev
);
8612 hba
= find_disk_attached_hba(-1, path
);
8613 if (hba
&& hba
->type
== SYS_DEV_SAS
)
8615 else if (hba
&& hba
->type
== SYS_DEV_SATA
)
8619 dprintf("path: %s hba: %s attached: %s\n",
8620 path
, (hba
) ? hba
->path
: "NULL", drv
);
8628 static int imsm_find_array_minor_by_subdev(int subdev
, int container
, int *minor
)
8630 char subdev_name
[20];
8631 struct mdstat_ent
*mdstat
;
8633 sprintf(subdev_name
, "%d", subdev
);
8634 mdstat
= mdstat_by_subdev(subdev_name
, container
);
8638 *minor
= mdstat
->devnum
;
8639 free_mdstat(mdstat
);
8643 static int imsm_reshape_is_allowed_on_container(struct supertype
*st
,
8644 struct geo_params
*geo
,
8645 int *old_raid_disks
)
8647 /* currently we only support increasing the number of devices
8648 * for a container. This increases the number of device for each
8649 * member array. They must all be RAID0 or RAID5.
8652 struct mdinfo
*info
, *member
;
8653 int devices_that_can_grow
= 0;
8655 dprintf("imsm: imsm_reshape_is_allowed_on_container(ENTER): "
8656 "st->devnum = (%i)\n",
8659 if (geo
->size
!= -1 ||
8660 geo
->level
!= UnSet
||
8661 geo
->layout
!= UnSet
||
8662 geo
->chunksize
!= 0 ||
8663 geo
->raid_disks
== UnSet
) {
8664 dprintf("imsm: Container operation is allowed for "
8665 "raid disks number change only.\n");
8669 info
= container_content_imsm(st
, NULL
);
8670 for (member
= info
; member
; member
= member
->next
) {
8674 dprintf("imsm: checking device_num: %i\n",
8675 member
->container_member
);
8677 if (geo
->raid_disks
<= member
->array
.raid_disks
) {
8678 /* we work on container for Online Capacity Expansion
8679 * only so raid_disks has to grow
8681 dprintf("imsm: for container operation raid disks "
8682 "increase is required\n");
8686 if ((info
->array
.level
!= 0) &&
8687 (info
->array
.level
!= 5)) {
8688 /* we cannot use this container with other raid level
8690 dprintf("imsm: for container operation wrong"
8691 " raid level (%i) detected\n",
8695 /* check for platform support
8696 * for this raid level configuration
8698 struct intel_super
*super
= st
->sb
;
8699 if (!is_raid_level_supported(super
->orom
,
8700 member
->array
.level
,
8702 dprintf("platform does not support raid%d with"
8706 geo
->raid_disks
> 1 ? "s" : "");
8709 /* check if component size is aligned to chunk size
8711 if (info
->component_size
%
8712 (info
->array
.chunk_size
/512)) {
8713 dprintf("Component size is not aligned to "
8719 if (*old_raid_disks
&&
8720 info
->array
.raid_disks
!= *old_raid_disks
)
8722 *old_raid_disks
= info
->array
.raid_disks
;
8724 /* All raid5 and raid0 volumes in container
8725 * have to be ready for Online Capacity Expansion
8726 * so they need to be assembled. We have already
8727 * checked that no recovery etc is happening.
8729 result
= imsm_find_array_minor_by_subdev(member
->container_member
,
8733 dprintf("imsm: cannot find array\n");
8736 devices_that_can_grow
++;
8739 if (!member
&& devices_that_can_grow
)
8743 dprintf("\tContainer operation allowed\n");
8745 dprintf("\tError: %i\n", ret_val
);
8750 /* Function: get_spares_for_grow
8751 * Description: Allocates memory and creates list of spare devices
8752 * avaliable in container. Checks if spare drive size is acceptable.
8753 * Parameters: Pointer to the supertype structure
8754 * Returns: Pointer to the list of spare devices (mdinfo structure) on success,
8757 static struct mdinfo
*get_spares_for_grow(struct supertype
*st
)
8759 unsigned long long min_size
= min_acceptable_spare_size_imsm(st
);
8760 return container_choose_spares(st
, min_size
, NULL
, NULL
, NULL
, 0);
8763 /******************************************************************************
8764 * function: imsm_create_metadata_update_for_reshape
8765 * Function creates update for whole IMSM container.
8767 ******************************************************************************/
8768 static int imsm_create_metadata_update_for_reshape(
8769 struct supertype
*st
,
8770 struct geo_params
*geo
,
8772 struct imsm_update_reshape
**updatep
)
8774 struct intel_super
*super
= st
->sb
;
8775 struct imsm_super
*mpb
= super
->anchor
;
8776 int update_memory_size
= 0;
8777 struct imsm_update_reshape
*u
= NULL
;
8778 struct mdinfo
*spares
= NULL
;
8780 int delta_disks
= 0;
8783 dprintf("imsm_update_metadata_for_reshape(enter) raid_disks = %i\n",
8786 delta_disks
= geo
->raid_disks
- old_raid_disks
;
8788 /* size of all update data without anchor */
8789 update_memory_size
= sizeof(struct imsm_update_reshape
);
8791 /* now add space for spare disks that we need to add. */
8792 update_memory_size
+= sizeof(u
->new_disks
[0]) * (delta_disks
- 1);
8794 u
= calloc(1, update_memory_size
);
8797 "cannot get memory for imsm_update_reshape update\n");
8800 u
->type
= update_reshape_container_disks
;
8801 u
->old_raid_disks
= old_raid_disks
;
8802 u
->new_raid_disks
= geo
->raid_disks
;
8804 /* now get spare disks list
8806 spares
= get_spares_for_grow(st
);
8809 || delta_disks
> spares
->array
.spare_disks
) {
8810 fprintf(stderr
, Name
": imsm: ERROR: Cannot get spare devices "
8811 "for %s.\n", geo
->dev_name
);
8816 /* we have got spares
8817 * update disk list in imsm_disk list table in anchor
8819 dprintf("imsm: %i spares are available.\n\n",
8820 spares
->array
.spare_disks
);
8823 for (i
= 0; i
< delta_disks
; i
++) {
8828 u
->new_disks
[i
] = makedev(dev
->disk
.major
,
8830 dl
= get_disk_super(super
, dev
->disk
.major
, dev
->disk
.minor
);
8831 dl
->index
= mpb
->num_disks
;
8841 dprintf("imsm: reshape update preparation :");
8842 if (i
== delta_disks
) {
8845 return update_memory_size
;
8848 dprintf(" Error\n");
8853 /******************************************************************************
8854 * function: imsm_create_metadata_update_for_migration()
8855 * Creates update for IMSM array.
8857 ******************************************************************************/
8858 static int imsm_create_metadata_update_for_migration(
8859 struct supertype
*st
,
8860 struct geo_params
*geo
,
8861 struct imsm_update_reshape_migration
**updatep
)
8863 struct intel_super
*super
= st
->sb
;
8864 int update_memory_size
= 0;
8865 struct imsm_update_reshape_migration
*u
= NULL
;
8866 struct imsm_dev
*dev
;
8867 int previous_level
= -1;
8869 dprintf("imsm_create_metadata_update_for_migration(enter)"
8870 " New Level = %i\n", geo
->level
);
8872 /* size of all update data without anchor */
8873 update_memory_size
= sizeof(struct imsm_update_reshape_migration
);
8875 u
= calloc(1, update_memory_size
);
8877 dprintf("error: cannot get memory for "
8878 "imsm_create_metadata_update_for_migration\n");
8881 u
->type
= update_reshape_migration
;
8882 u
->subdev
= super
->current_vol
;
8883 u
->new_level
= geo
->level
;
8884 u
->new_layout
= geo
->layout
;
8885 u
->new_raid_disks
= u
->old_raid_disks
= geo
->raid_disks
;
8886 u
->new_disks
[0] = -1;
8887 u
->new_chunksize
= -1;
8889 dev
= get_imsm_dev(super
, u
->subdev
);
8891 struct imsm_map
*map
;
8893 map
= get_imsm_map(dev
, 0);
8895 int current_chunk_size
=
8896 __le16_to_cpu(map
->blocks_per_strip
) / 2;
8898 if (geo
->chunksize
!= current_chunk_size
) {
8899 u
->new_chunksize
= geo
->chunksize
/ 1024;
8901 "chunk size change from %i to %i\n",
8902 current_chunk_size
, u
->new_chunksize
);
8904 previous_level
= map
->raid_level
;
8907 if ((geo
->level
== 5) && (previous_level
== 0)) {
8908 struct mdinfo
*spares
= NULL
;
8910 u
->new_raid_disks
++;
8911 spares
= get_spares_for_grow(st
);
8912 if ((spares
== NULL
) || (spares
->array
.spare_disks
< 1)) {
8915 update_memory_size
= 0;
8916 dprintf("error: cannot get spare device "
8917 "for requested migration");
8922 dprintf("imsm: reshape update preparation : OK\n");
8925 return update_memory_size
;
8928 static void imsm_update_metadata_locally(struct supertype
*st
,
8931 struct metadata_update mu
;
8936 mu
.space_list
= NULL
;
8938 imsm_prepare_update(st
, &mu
);
8939 imsm_process_update(st
, &mu
);
8941 while (mu
.space_list
) {
8942 void **space
= mu
.space_list
;
8943 mu
.space_list
= *space
;
8948 /***************************************************************************
8949 * Function: imsm_analyze_change
8950 * Description: Function analyze change for single volume
8951 * and validate if transition is supported
8952 * Parameters: Geometry parameters, supertype structure
8953 * Returns: Operation type code on success, -1 if fail
8954 ****************************************************************************/
8955 enum imsm_reshape_type
imsm_analyze_change(struct supertype
*st
,
8956 struct geo_params
*geo
)
8965 getinfo_super_imsm_volume(st
, &info
, NULL
);
8966 if ((geo
->level
!= info
.array
.level
) &&
8967 (geo
->level
>= 0) &&
8968 (geo
->level
!= UnSet
)) {
8969 switch (info
.array
.level
) {
8971 if (geo
->level
== 5) {
8972 change
= CH_MIGRATION
;
8973 if (geo
->layout
!= ALGORITHM_LEFT_ASYMMETRIC
) {
8975 Name
" Error. Requested Layout "
8976 "not supported (left-asymmetric layout "
8977 "is supported only)!\n");
8979 goto analyse_change_exit
;
8981 layout
= geo
->layout
;
8983 devNumChange
= 1; /* parity disk added */
8984 } else if (geo
->level
== 10) {
8985 change
= CH_TAKEOVER
;
8987 devNumChange
= 2; /* two mirrors added */
8988 layout
= 0x102; /* imsm supported layout */
8993 if (geo
->level
== 0) {
8994 change
= CH_TAKEOVER
;
8996 devNumChange
= -(geo
->raid_disks
/2);
8997 layout
= 0; /* imsm raid0 layout */
9003 Name
" Error. Level Migration from %d to %d "
9005 info
.array
.level
, geo
->level
);
9006 goto analyse_change_exit
;
9009 geo
->level
= info
.array
.level
;
9011 if ((geo
->layout
!= info
.array
.layout
)
9012 && ((geo
->layout
!= UnSet
) && (geo
->layout
!= -1))) {
9013 change
= CH_MIGRATION
;
9014 if ((info
.array
.layout
== 0)
9015 && (info
.array
.level
== 5)
9016 && (geo
->layout
== 5)) {
9017 /* reshape 5 -> 4 */
9018 } else if ((info
.array
.layout
== 5)
9019 && (info
.array
.level
== 5)
9020 && (geo
->layout
== 0)) {
9021 /* reshape 4 -> 5 */
9026 Name
" Error. Layout Migration from %d to %d "
9028 info
.array
.layout
, geo
->layout
);
9030 goto analyse_change_exit
;
9033 geo
->layout
= info
.array
.layout
;
9035 if ((geo
->chunksize
> 0) && (geo
->chunksize
!= UnSet
)
9036 && (geo
->chunksize
!= info
.array
.chunk_size
))
9037 change
= CH_MIGRATION
;
9039 geo
->chunksize
= info
.array
.chunk_size
;
9041 chunk
= geo
->chunksize
/ 1024;
9042 if (!validate_geometry_imsm(st
,
9045 geo
->raid_disks
+ devNumChange
,
9052 struct intel_super
*super
= st
->sb
;
9053 struct imsm_super
*mpb
= super
->anchor
;
9055 if (mpb
->num_raid_devs
> 1) {
9057 Name
" Error. Cannot perform operation on %s"
9058 "- for this operation it MUST be single "
9059 "array in container\n",
9065 analyse_change_exit
:
9070 int imsm_takeover(struct supertype
*st
, struct geo_params
*geo
)
9072 struct intel_super
*super
= st
->sb
;
9073 struct imsm_update_takeover
*u
;
9075 u
= malloc(sizeof(struct imsm_update_takeover
));
9079 u
->type
= update_takeover
;
9080 u
->subarray
= super
->current_vol
;
9082 /* 10->0 transition */
9083 if (geo
->level
== 0)
9084 u
->direction
= R10_TO_R0
;
9086 /* 0->10 transition */
9087 if (geo
->level
== 10)
9088 u
->direction
= R0_TO_R10
;
9090 /* update metadata locally */
9091 imsm_update_metadata_locally(st
, u
,
9092 sizeof(struct imsm_update_takeover
));
9093 /* and possibly remotely */
9094 if (st
->update_tail
)
9095 append_metadata_update(st
, u
,
9096 sizeof(struct imsm_update_takeover
));
9103 static int imsm_reshape_super(struct supertype
*st
, long long size
, int level
,
9104 int layout
, int chunksize
, int raid_disks
,
9105 int delta_disks
, char *backup
, char *dev
,
9109 struct geo_params geo
;
9111 dprintf("imsm: reshape_super called.\n");
9113 memset(&geo
, 0, sizeof(struct geo_params
));
9116 geo
.dev_id
= st
->devnum
;
9119 geo
.layout
= layout
;
9120 geo
.chunksize
= chunksize
;
9121 geo
.raid_disks
= raid_disks
;
9122 if (delta_disks
!= UnSet
)
9123 geo
.raid_disks
+= delta_disks
;
9125 dprintf("\tfor level : %i\n", geo
.level
);
9126 dprintf("\tfor raid_disks : %i\n", geo
.raid_disks
);
9128 if (experimental() == 0)
9131 if (st
->container_dev
== st
->devnum
) {
9132 /* On container level we can only increase number of devices. */
9133 dprintf("imsm: info: Container operation\n");
9134 int old_raid_disks
= 0;
9136 if (imsm_reshape_is_allowed_on_container(
9137 st
, &geo
, &old_raid_disks
)) {
9138 struct imsm_update_reshape
*u
= NULL
;
9141 len
= imsm_create_metadata_update_for_reshape(
9142 st
, &geo
, old_raid_disks
, &u
);
9145 dprintf("imsm: Cannot prepare update\n");
9146 goto exit_imsm_reshape_super
;
9150 /* update metadata locally */
9151 imsm_update_metadata_locally(st
, u
, len
);
9152 /* and possibly remotely */
9153 if (st
->update_tail
)
9154 append_metadata_update(st
, u
, len
);
9159 fprintf(stderr
, Name
": (imsm) Operation "
9160 "is not allowed on this container\n");
9163 /* On volume level we support following operations
9164 * - takeover: raid10 -> raid0; raid0 -> raid10
9165 * - chunk size migration
9166 * - migration: raid5 -> raid0; raid0 -> raid5
9168 struct intel_super
*super
= st
->sb
;
9169 struct intel_dev
*dev
= super
->devlist
;
9171 dprintf("imsm: info: Volume operation\n");
9172 /* find requested device */
9174 if (imsm_find_array_minor_by_subdev(
9175 dev
->index
, st
->container_dev
, &devnum
) == 0
9176 && devnum
== geo
.dev_id
)
9181 fprintf(stderr
, Name
" Cannot find %s (%i) subarray\n",
9182 geo
.dev_name
, geo
.dev_id
);
9183 goto exit_imsm_reshape_super
;
9185 super
->current_vol
= dev
->index
;
9186 change
= imsm_analyze_change(st
, &geo
);
9189 ret_val
= imsm_takeover(st
, &geo
);
9191 case CH_MIGRATION
: {
9192 struct imsm_update_reshape_migration
*u
= NULL
;
9194 imsm_create_metadata_update_for_migration(
9198 "Cannot prepare update\n");
9202 /* update metadata locally */
9203 imsm_update_metadata_locally(st
, u
, len
);
9204 /* and possibly remotely */
9205 if (st
->update_tail
)
9206 append_metadata_update(st
, u
, len
);
9216 exit_imsm_reshape_super
:
9217 dprintf("imsm: reshape_super Exit code = %i\n", ret_val
);
9221 /*******************************************************************************
9222 * Function: wait_for_reshape_imsm
9223 * Description: Function writes new sync_max value and waits until
9224 * reshape process reach new position
9226 * sra : general array info
9227 * ndata : number of disks in new array's layout
9230 * 1 : there is no reshape in progress,
9232 ******************************************************************************/
9233 int wait_for_reshape_imsm(struct mdinfo
*sra
, int ndata
)
9235 int fd
= sysfs_get_fd(sra
, NULL
, "reshape_position");
9236 unsigned long long completed
;
9237 /* to_complete : new sync_max position */
9238 unsigned long long to_complete
= sra
->reshape_progress
;
9239 unsigned long long position_to_set
= to_complete
/ ndata
;
9242 dprintf("imsm: wait_for_reshape_imsm() "
9243 "cannot open reshape_position\n");
9247 if (sysfs_fd_get_ll(fd
, &completed
) < 0) {
9248 dprintf("imsm: wait_for_reshape_imsm() "
9249 "cannot read reshape_position (no reshape in progres)\n");
9254 if (completed
> to_complete
) {
9255 dprintf("imsm: wait_for_reshape_imsm() "
9256 "wrong next position to set %llu (%llu)\n",
9257 to_complete
, completed
);
9261 dprintf("Position set: %llu\n", position_to_set
);
9262 if (sysfs_set_num(sra
, NULL
, "sync_max",
9263 position_to_set
) != 0) {
9264 dprintf("imsm: wait_for_reshape_imsm() "
9265 "cannot set reshape position to %llu\n",
9276 select(fd
+1, &rfds
, NULL
, NULL
, NULL
);
9277 if (sysfs_get_str(sra
, NULL
, "sync_action",
9279 strncmp(action
, "reshape", 7) != 0)
9281 if (sysfs_fd_get_ll(fd
, &completed
) < 0) {
9282 dprintf("imsm: wait_for_reshape_imsm() "
9283 "cannot read reshape_position (in loop)\n");
9287 } while (completed
< to_complete
);
9293 /*******************************************************************************
9294 * Function: check_degradation_change
9295 * Description: Check that array hasn't become failed.
9297 * info : for sysfs access
9298 * sources : source disks descriptors
9299 * degraded: previous degradation level
9302 ******************************************************************************/
9303 int check_degradation_change(struct mdinfo
*info
,
9307 unsigned long long new_degraded
;
9308 sysfs_get_ll(info
, NULL
, "degraded", &new_degraded
);
9309 if (new_degraded
!= (unsigned long long)degraded
) {
9310 /* check each device to ensure it is still working */
9313 for (sd
= info
->devs
; sd
; sd
= sd
->next
) {
9314 if (sd
->disk
.state
& (1<<MD_DISK_FAULTY
))
9316 if (sd
->disk
.state
& (1<<MD_DISK_SYNC
)) {
9318 if (sysfs_get_str(info
,
9319 sd
, "state", sbuf
, 20) < 0 ||
9320 strstr(sbuf
, "faulty") ||
9321 strstr(sbuf
, "in_sync") == NULL
) {
9322 /* this device is dead */
9323 sd
->disk
.state
= (1<<MD_DISK_FAULTY
);
9324 if (sd
->disk
.raid_disk
>= 0 &&
9325 sources
[sd
->disk
.raid_disk
] >= 0) {
9327 sd
->disk
.raid_disk
]);
9328 sources
[sd
->disk
.raid_disk
] =
9337 return new_degraded
;
9340 /*******************************************************************************
9341 * Function: imsm_manage_reshape
9342 * Description: Function finds array under reshape and it manages reshape
9343 * process. It creates stripes backups (if required) and sets
9346 * afd : Backup handle (nattive) - not used
9347 * sra : general array info
9348 * reshape : reshape parameters - not used
9349 * st : supertype structure
9350 * blocks : size of critical section [blocks]
9351 * fds : table of source device descriptor
9352 * offsets : start of array (offest per devices)
9354 * destfd : table of destination device descriptor
9355 * destoffsets : table of destination offsets (per device)
9357 * 1 : success, reshape is done
9359 ******************************************************************************/
9360 static int imsm_manage_reshape(
9361 int afd
, struct mdinfo
*sra
, struct reshape
*reshape
,
9362 struct supertype
*st
, unsigned long backup_blocks
,
9363 int *fds
, unsigned long long *offsets
,
9364 int dests
, int *destfd
, unsigned long long *destoffsets
)
9367 struct intel_super
*super
= st
->sb
;
9368 struct intel_dev
*dv
= NULL
;
9369 struct imsm_dev
*dev
= NULL
;
9370 struct imsm_map
*map_src
;
9371 int migr_vol_qan
= 0;
9372 int ndata
, odata
; /* [bytes] */
9373 int chunk
; /* [bytes] */
9374 struct migr_record
*migr_rec
;
9376 unsigned int buf_size
; /* [bytes] */
9377 unsigned long long max_position
; /* array size [bytes] */
9378 unsigned long long next_step
; /* [blocks]/[bytes] */
9379 unsigned long long old_data_stripe_length
;
9380 unsigned long long start_src
; /* [bytes] */
9381 unsigned long long start
; /* [bytes] */
9382 unsigned long long start_buf_shift
; /* [bytes] */
9384 int source_layout
= 0;
9386 if (!fds
|| !offsets
|| !sra
)
9389 /* Find volume during the reshape */
9390 for (dv
= super
->devlist
; dv
; dv
= dv
->next
) {
9391 if (dv
->dev
->vol
.migr_type
== MIGR_GEN_MIGR
9392 && dv
->dev
->vol
.migr_state
== 1) {
9397 /* Only one volume can migrate at the same time */
9398 if (migr_vol_qan
!= 1) {
9399 fprintf(stderr
, Name
" : %s", migr_vol_qan
?
9400 "Number of migrating volumes greater than 1\n" :
9401 "There is no volume during migrationg\n");
9405 map_src
= get_imsm_map(dev
, 1);
9406 if (map_src
== NULL
)
9409 ndata
= imsm_num_data_members(dev
, 0);
9410 odata
= imsm_num_data_members(dev
, 1);
9412 chunk
= __le16_to_cpu(map_src
->blocks_per_strip
) * 512;
9413 old_data_stripe_length
= odata
* chunk
;
9415 migr_rec
= super
->migr_rec
;
9417 /* initialize migration record for start condition */
9418 if (sra
->reshape_progress
== 0)
9419 init_migr_record_imsm(st
, dev
, sra
);
9421 if (__le32_to_cpu(migr_rec
->rec_status
) != UNIT_SRC_NORMAL
) {
9422 dprintf("imsm: cannot restart migration when data "
9423 "are present in copy area.\n");
9429 buf_size
= __le32_to_cpu(migr_rec
->blocks_per_unit
) * 512;
9430 /* extend buffer size for parity disk */
9431 buf_size
+= __le32_to_cpu(migr_rec
->dest_depth_per_unit
) * 512;
9432 /* add space for stripe aligment */
9433 buf_size
+= old_data_stripe_length
;
9434 if (posix_memalign((void **)&buf
, 4096, buf_size
)) {
9435 dprintf("imsm: Cannot allocate checpoint buffer\n");
9439 max_position
= sra
->component_size
* ndata
;
9440 source_layout
= imsm_level_to_layout(map_src
->raid_level
);
9442 while (__le32_to_cpu(migr_rec
->curr_migr_unit
) <
9443 __le32_to_cpu(migr_rec
->num_migr_units
)) {
9444 /* current reshape position [blocks] */
9445 unsigned long long current_position
=
9446 __le32_to_cpu(migr_rec
->blocks_per_unit
)
9447 * __le32_to_cpu(migr_rec
->curr_migr_unit
);
9448 unsigned long long border
;
9450 /* Check that array hasn't become failed.
9452 degraded
= check_degradation_change(sra
, fds
, degraded
);
9454 dprintf("imsm: Abort reshape due to degradation"
9455 " level (%i)\n", degraded
);
9459 next_step
= __le32_to_cpu(migr_rec
->blocks_per_unit
);
9461 if ((current_position
+ next_step
) > max_position
)
9462 next_step
= max_position
- current_position
;
9464 start
= current_position
* 512;
9466 /* allign reading start to old geometry */
9467 start_buf_shift
= start
% old_data_stripe_length
;
9468 start_src
= start
- start_buf_shift
;
9470 border
= (start_src
/ odata
) - (start
/ ndata
);
9472 if (border
<= __le32_to_cpu(migr_rec
->dest_depth_per_unit
)) {
9473 /* save critical stripes to buf
9474 * start - start address of current unit
9476 * start_src - start address of current unit
9477 * to backup alligned to source array
9480 unsigned long long next_step_filler
= 0;
9481 unsigned long long copy_length
= next_step
* 512;
9483 /* allign copy area length to stripe in old geometry */
9484 next_step_filler
= ((copy_length
+ start_buf_shift
)
9485 % old_data_stripe_length
);
9486 if (next_step_filler
)
9487 next_step_filler
= (old_data_stripe_length
9488 - next_step_filler
);
9489 dprintf("save_stripes() parameters: start = %llu,"
9490 "\tstart_src = %llu,\tnext_step*512 = %llu,"
9491 "\tstart_in_buf_shift = %llu,"
9492 "\tnext_step_filler = %llu\n",
9493 start
, start_src
, copy_length
,
9494 start_buf_shift
, next_step_filler
);
9496 if (save_stripes(fds
, offsets
, map_src
->num_members
,
9497 chunk
, map_src
->raid_level
,
9498 source_layout
, 0, NULL
, start_src
,
9500 next_step_filler
+ start_buf_shift
,
9502 dprintf("imsm: Cannot save stripes"
9506 /* Convert data to destination format and store it
9507 * in backup general migration area
9509 if (save_backup_imsm(st
, dev
, sra
,
9510 buf
+ start_buf_shift
, copy_length
)) {
9511 dprintf("imsm: Cannot save stripes to "
9512 "target devices\n");
9515 if (save_checkpoint_imsm(st
, sra
,
9516 UNIT_SRC_IN_CP_AREA
)) {
9517 dprintf("imsm: Cannot write checkpoint to "
9518 "migration record (UNIT_SRC_IN_CP_AREA)\n");
9522 /* set next step to use whole border area */
9523 border
/= next_step
;
9525 next_step
*= border
;
9527 /* When data backed up, checkpoint stored,
9528 * kick the kernel to reshape unit of data
9530 next_step
= next_step
+ sra
->reshape_progress
;
9531 /* limit next step to array max position */
9532 if (next_step
> max_position
)
9533 next_step
= max_position
;
9534 sysfs_set_num(sra
, NULL
, "suspend_lo", sra
->reshape_progress
);
9535 sysfs_set_num(sra
, NULL
, "suspend_hi", next_step
);
9536 sra
->reshape_progress
= next_step
;
9538 /* wait until reshape finish */
9539 if (wait_for_reshape_imsm(sra
, ndata
) < 0) {
9540 dprintf("wait_for_reshape_imsm returned error!\n");
9544 if (save_checkpoint_imsm(st
, sra
, UNIT_SRC_NORMAL
) == 1) {
9545 /* ignore error == 2, this can mean end of reshape here
9547 dprintf("imsm: Cannot write checkpoint to "
9548 "migration record (UNIT_SRC_NORMAL)\n");
9554 /* return '1' if done */
9562 #endif /* MDASSEMBLE */
9564 struct superswitch super_imsm
= {
9566 .examine_super
= examine_super_imsm
,
9567 .brief_examine_super
= brief_examine_super_imsm
,
9568 .brief_examine_subarrays
= brief_examine_subarrays_imsm
,
9569 .export_examine_super
= export_examine_super_imsm
,
9570 .detail_super
= detail_super_imsm
,
9571 .brief_detail_super
= brief_detail_super_imsm
,
9572 .write_init_super
= write_init_super_imsm
,
9573 .validate_geometry
= validate_geometry_imsm
,
9574 .add_to_super
= add_to_super_imsm
,
9575 .remove_from_super
= remove_from_super_imsm
,
9576 .detail_platform
= detail_platform_imsm
,
9577 .kill_subarray
= kill_subarray_imsm
,
9578 .update_subarray
= update_subarray_imsm
,
9579 .load_container
= load_container_imsm
,
9580 .default_geometry
= default_geometry_imsm
,
9581 .get_disk_controller_domain
= imsm_get_disk_controller_domain
,
9582 .reshape_super
= imsm_reshape_super
,
9583 .manage_reshape
= imsm_manage_reshape
,
9584 .recover_backup
= recover_backup_imsm
,
9586 .match_home
= match_home_imsm
,
9587 .uuid_from_super
= uuid_from_super_imsm
,
9588 .getinfo_super
= getinfo_super_imsm
,
9589 .getinfo_super_disks
= getinfo_super_disks_imsm
,
9590 .update_super
= update_super_imsm
,
9592 .avail_size
= avail_size_imsm
,
9593 .min_acceptable_spare_size
= min_acceptable_spare_size_imsm
,
9595 .compare_super
= compare_super_imsm
,
9597 .load_super
= load_super_imsm
,
9598 .init_super
= init_super_imsm
,
9599 .store_super
= store_super_imsm
,
9600 .free_super
= free_super_imsm
,
9601 .match_metadata_desc
= match_metadata_desc_imsm
,
9602 .container_content
= container_content_imsm
,
9610 .open_new
= imsm_open_new
,
9611 .set_array_state
= imsm_set_array_state
,
9612 .set_disk
= imsm_set_disk
,
9613 .sync_metadata
= imsm_sync_metadata
,
9614 .activate_spare
= imsm_activate_spare
,
9615 .process_update
= imsm_process_update
,
9616 .prepare_update
= imsm_prepare_update
,
9617 #endif /* MDASSEMBLE */