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
115 /* RAID map configuration infos. */
117 __u32 pba_of_lba0
; /* start address of partition */
118 __u32 blocks_per_member
;/* blocks per member */
119 __u32 num_data_stripes
; /* number of data stripes */
120 __u16 blocks_per_strip
;
121 __u8 map_state
; /* Normal, Uninitialized, Degraded, Failed */
122 #define IMSM_T_STATE_NORMAL 0
123 #define IMSM_T_STATE_UNINITIALIZED 1
124 #define IMSM_T_STATE_DEGRADED 2
125 #define IMSM_T_STATE_FAILED 3
127 #define IMSM_T_RAID0 0
128 #define IMSM_T_RAID1 1
129 #define IMSM_T_RAID5 5 /* since metadata version 1.2.02 ? */
130 __u8 num_members
; /* number of member disks */
131 __u8 num_domains
; /* number of parity domains */
132 __u8 failed_disk_num
; /* valid only when state is degraded */
134 __u32 filler
[7]; /* expansion area */
135 #define IMSM_ORD_REBUILD (1 << 24)
136 __u32 disk_ord_tbl
[1]; /* disk_ord_tbl[num_members],
137 * top byte contains some flags
139 } __attribute__ ((packed
));
142 __u32 curr_migr_unit
;
143 __u32 checkpoint_id
; /* id to access curr_migr_unit */
144 __u8 migr_state
; /* Normal or Migrating */
146 #define MIGR_REBUILD 1
147 #define MIGR_VERIFY 2 /* analagous to echo check > sync_action */
148 #define MIGR_GEN_MIGR 3
149 #define MIGR_STATE_CHANGE 4
150 #define MIGR_REPAIR 5
151 __u8 migr_type
; /* Initializing, Rebuilding, ... */
153 __u8 fs_state
; /* fast-sync state for CnG (0xff == disabled) */
154 __u16 verify_errors
; /* number of mismatches */
155 __u16 bad_blocks
; /* number of bad blocks during verify */
157 struct imsm_map map
[1];
158 /* here comes another one if migr_state */
159 } __attribute__ ((packed
));
162 __u8 volume
[MAX_RAID_SERIAL_LEN
];
165 #define DEV_BOOTABLE __cpu_to_le32(0x01)
166 #define DEV_BOOT_DEVICE __cpu_to_le32(0x02)
167 #define DEV_READ_COALESCING __cpu_to_le32(0x04)
168 #define DEV_WRITE_COALESCING __cpu_to_le32(0x08)
169 #define DEV_LAST_SHUTDOWN_DIRTY __cpu_to_le32(0x10)
170 #define DEV_HIDDEN_AT_BOOT __cpu_to_le32(0x20)
171 #define DEV_CURRENTLY_HIDDEN __cpu_to_le32(0x40)
172 #define DEV_VERIFY_AND_FIX __cpu_to_le32(0x80)
173 #define DEV_MAP_STATE_UNINIT __cpu_to_le32(0x100)
174 #define DEV_NO_AUTO_RECOVERY __cpu_to_le32(0x200)
175 #define DEV_CLONE_N_GO __cpu_to_le32(0x400)
176 #define DEV_CLONE_MAN_SYNC __cpu_to_le32(0x800)
177 #define DEV_CNG_MASTER_DISK_NUM __cpu_to_le32(0x1000)
178 __u32 status
; /* Persistent RaidDev status */
179 __u32 reserved_blocks
; /* Reserved blocks at beginning of volume */
183 __u8 cng_master_disk
;
187 #define IMSM_DEV_FILLERS 10
188 __u32 filler
[IMSM_DEV_FILLERS
];
190 } __attribute__ ((packed
));
193 __u8 sig
[MAX_SIGNATURE_LENGTH
]; /* 0x00 - 0x1F */
194 __u32 check_sum
; /* 0x20 - 0x23 MPB Checksum */
195 __u32 mpb_size
; /* 0x24 - 0x27 Size of MPB */
196 __u32 family_num
; /* 0x28 - 0x2B Checksum from first time this config was written */
197 __u32 generation_num
; /* 0x2C - 0x2F Incremented each time this array's MPB is written */
198 __u32 error_log_size
; /* 0x30 - 0x33 in bytes */
199 __u32 attributes
; /* 0x34 - 0x37 */
200 __u8 num_disks
; /* 0x38 Number of configured disks */
201 __u8 num_raid_devs
; /* 0x39 Number of configured volumes */
202 __u8 error_log_pos
; /* 0x3A */
203 __u8 fill
[1]; /* 0x3B */
204 __u32 cache_size
; /* 0x3c - 0x40 in mb */
205 __u32 orig_family_num
; /* 0x40 - 0x43 original family num */
206 __u32 pwr_cycle_count
; /* 0x44 - 0x47 simulated power cycle count for array */
207 __u32 bbm_log_size
; /* 0x48 - 0x4B - size of bad Block Mgmt Log in bytes */
208 #define IMSM_FILLERS 35
209 __u32 filler
[IMSM_FILLERS
]; /* 0x4C - 0xD7 RAID_MPB_FILLERS */
210 struct imsm_disk disk
[1]; /* 0xD8 diskTbl[numDisks] */
211 /* here comes imsm_dev[num_raid_devs] */
212 /* here comes BBM logs */
213 } __attribute__ ((packed
));
215 #define BBM_LOG_MAX_ENTRIES 254
217 struct bbm_log_entry
{
218 __u64 defective_block_start
;
219 #define UNREADABLE 0xFFFFFFFF
220 __u32 spare_block_offset
;
221 __u16 remapped_marked_count
;
223 } __attribute__ ((__packed__
));
226 __u32 signature
; /* 0xABADB10C */
228 __u32 reserved_spare_block_count
; /* 0 */
229 __u32 reserved
; /* 0xFFFF */
230 __u64 first_spare_lba
;
231 struct bbm_log_entry mapped_block_entries
[BBM_LOG_MAX_ENTRIES
];
232 } __attribute__ ((__packed__
));
236 static char *map_state_str
[] = { "normal", "uninitialized", "degraded", "failed" };
239 #define RAID_DISK_RESERVED_BLOCKS_IMSM_HI 2209
241 #define GEN_MIGR_AREA_SIZE 2048 /* General Migration Copy Area size in blocks */
243 #define MIGR_REC_BUF_SIZE 512 /* size of migr_record i/o buffer */
244 #define MIGR_REC_POSITION 512 /* migr_record position offset on disk,
245 * MIGR_REC_BUF_SIZE <= MIGR_REC_POSITION
249 #define UNIT_SRC_NORMAL 0 /* Source data for curr_migr_unit must
250 * be recovered using srcMap */
251 #define UNIT_SRC_IN_CP_AREA 1 /* Source data for curr_migr_unit has
252 * already been migrated and must
253 * be recovered from checkpoint area */
255 __u32 rec_status
; /* Status used to determine how to restart
256 * migration in case it aborts
258 __u32 curr_migr_unit
; /* 0..numMigrUnits-1 */
259 __u32 family_num
; /* Family number of MPB
260 * containing the RaidDev
261 * that is migrating */
262 __u32 ascending_migr
; /* True if migrating in increasing
264 __u32 blocks_per_unit
; /* Num disk blocks per unit of operation */
265 __u32 dest_depth_per_unit
; /* Num member blocks each destMap
267 * advances per unit-of-operation */
268 __u32 ckpt_area_pba
; /* Pba of first block of ckpt copy area */
269 __u32 dest_1st_member_lba
; /* First member lba on first
270 * stripe of destination */
271 __u32 num_migr_units
; /* Total num migration units-of-op */
272 __u32 post_migr_vol_cap
; /* Size of volume after
273 * migration completes */
274 __u32 post_migr_vol_cap_hi
; /* Expansion space for LBA64 */
275 __u32 ckpt_read_disk_num
; /* Which member disk in destSubMap[0] the
276 * migration ckpt record was read from
277 * (for recovered migrations) */
278 } __attribute__ ((__packed__
));
280 static __u8
migr_type(struct imsm_dev
*dev
)
282 if (dev
->vol
.migr_type
== MIGR_VERIFY
&&
283 dev
->status
& DEV_VERIFY_AND_FIX
)
286 return dev
->vol
.migr_type
;
289 static void set_migr_type(struct imsm_dev
*dev
, __u8 migr_type
)
291 /* for compatibility with older oroms convert MIGR_REPAIR, into
292 * MIGR_VERIFY w/ DEV_VERIFY_AND_FIX status
294 if (migr_type
== MIGR_REPAIR
) {
295 dev
->vol
.migr_type
= MIGR_VERIFY
;
296 dev
->status
|= DEV_VERIFY_AND_FIX
;
298 dev
->vol
.migr_type
= migr_type
;
299 dev
->status
&= ~DEV_VERIFY_AND_FIX
;
303 static unsigned int sector_count(__u32 bytes
)
305 return ((bytes
+ (512-1)) & (~(512-1))) / 512;
308 static unsigned int mpb_sectors(struct imsm_super
*mpb
)
310 return sector_count(__le32_to_cpu(mpb
->mpb_size
));
314 struct imsm_dev
*dev
;
315 struct intel_dev
*next
;
320 enum sys_dev_type type
;
323 struct intel_hba
*next
;
330 /* internal representation of IMSM metadata */
333 void *buf
; /* O_DIRECT buffer for reading/writing metadata */
334 struct imsm_super
*anchor
; /* immovable parameters */
337 void *migr_rec_buf
; /* buffer for I/O operations */
338 struct migr_record
*migr_rec
; /* migration record */
340 size_t len
; /* size of the 'buf' allocation */
341 void *next_buf
; /* for realloc'ing buf from the manager */
343 int updates_pending
; /* count of pending updates for mdmon */
344 int current_vol
; /* index of raid device undergoing creation */
345 __u32 create_offset
; /* common start for 'current_vol' */
346 __u32 random
; /* random data for seeding new family numbers */
347 struct intel_dev
*devlist
;
351 __u8 serial
[MAX_RAID_SERIAL_LEN
];
354 struct imsm_disk disk
;
357 struct extent
*e
; /* for determining freespace @ create */
358 int raiddisk
; /* slot to fill in autolayout */
360 } *disks
, *current_disk
;
361 struct dl
*disk_mgmt_list
; /* list of disks to add/remove while mdmon
363 struct dl
*missing
; /* disks removed while we weren't looking */
364 struct bbm_log
*bbm_log
;
365 struct intel_hba
*hba
; /* device path of the raid controller for this metadata */
366 const struct imsm_orom
*orom
; /* platform firmware support */
367 struct intel_super
*next
; /* (temp) list for disambiguating family_num */
371 struct imsm_disk disk
;
372 #define IMSM_UNKNOWN_OWNER (-1)
374 struct intel_disk
*next
;
378 unsigned long long start
, size
;
381 /* definitions of reshape process types */
382 enum imsm_reshape_type
{
387 /* definition of messages passed to imsm_process_update */
388 enum imsm_update_type
{
389 update_activate_spare
,
393 update_add_remove_disk
,
394 update_reshape_container_disks
,
395 update_reshape_migration
,
397 update_general_migration_checkpoint
,
400 struct imsm_update_activate_spare
{
401 enum imsm_update_type type
;
405 struct imsm_update_activate_spare
*next
;
418 enum takeover_direction
{
422 struct imsm_update_takeover
{
423 enum imsm_update_type type
;
425 enum takeover_direction direction
;
428 struct imsm_update_reshape
{
429 enum imsm_update_type type
;
433 int new_disks
[1]; /* new_raid_disks - old_raid_disks makedev number */
436 struct imsm_update_reshape_migration
{
437 enum imsm_update_type type
;
440 /* fields for array migration changes
447 int new_disks
[1]; /* new_raid_disks - old_raid_disks makedev number */
450 struct imsm_update_general_migration_checkpoint
{
451 enum imsm_update_type type
;
452 __u32 curr_migr_unit
;
456 __u8 serial
[MAX_RAID_SERIAL_LEN
];
459 struct imsm_update_create_array
{
460 enum imsm_update_type type
;
465 struct imsm_update_kill_array
{
466 enum imsm_update_type type
;
470 struct imsm_update_rename_array
{
471 enum imsm_update_type type
;
472 __u8 name
[MAX_RAID_SERIAL_LEN
];
476 struct imsm_update_add_remove_disk
{
477 enum imsm_update_type type
;
481 static const char *_sys_dev_type
[] = {
482 [SYS_DEV_UNKNOWN
] = "Unknown",
483 [SYS_DEV_SAS
] = "SAS",
484 [SYS_DEV_SATA
] = "SATA"
487 const char *get_sys_dev_type(enum sys_dev_type type
)
489 if (type
>= SYS_DEV_MAX
)
490 type
= SYS_DEV_UNKNOWN
;
492 return _sys_dev_type
[type
];
495 static struct intel_hba
* alloc_intel_hba(struct sys_dev
*device
)
497 struct intel_hba
*result
= malloc(sizeof(*result
));
499 result
->type
= device
->type
;
500 result
->path
= strdup(device
->path
);
502 if (result
->path
&& (result
->pci_id
= strrchr(result
->path
, '/')) != NULL
)
508 static struct intel_hba
* find_intel_hba(struct intel_hba
*hba
, struct sys_dev
*device
)
510 struct intel_hba
*result
=NULL
;
511 for (result
= hba
; result
; result
= result
->next
) {
512 if (result
->type
== device
->type
&& strcmp(result
->path
, device
->path
) == 0)
518 static int attach_hba_to_super(struct intel_super
*super
, struct sys_dev
*device
)
520 struct intel_hba
*hba
;
522 /* check if disk attached to Intel HBA */
523 hba
= find_intel_hba(super
->hba
, device
);
526 /* Check if HBA is already attached to super */
527 if (super
->hba
== NULL
) {
528 super
->hba
= alloc_intel_hba(device
);
533 /* Intel metadata allows for all disks attached to the same type HBA.
534 * Do not sypport odf HBA types mixing
536 if (device
->type
!= hba
->type
)
542 hba
->next
= alloc_intel_hba(device
);
546 static struct sys_dev
* find_disk_attached_hba(int fd
, const char *devname
)
548 struct sys_dev
*list
, *elem
, *prev
;
551 if ((list
= find_intel_devices()) == NULL
)
555 disk_path
= (char *) devname
;
557 disk_path
= diskfd_to_devpath(fd
);
564 for (prev
= NULL
, elem
= list
; elem
; prev
= elem
, elem
= elem
->next
) {
565 if (path_attached_to_hba(disk_path
, elem
->path
)) {
569 prev
->next
= elem
->next
;
571 if (disk_path
!= devname
)
577 if (disk_path
!= devname
)
585 static int find_intel_hba_capability(int fd
, struct intel_super
*super
,
588 static struct supertype
*match_metadata_desc_imsm(char *arg
)
590 struct supertype
*st
;
592 if (strcmp(arg
, "imsm") != 0 &&
593 strcmp(arg
, "default") != 0
597 st
= malloc(sizeof(*st
));
600 memset(st
, 0, sizeof(*st
));
601 st
->container_dev
= NoMdDev
;
602 st
->ss
= &super_imsm
;
603 st
->max_devs
= IMSM_MAX_DEVICES
;
604 st
->minor_version
= 0;
610 static __u8
*get_imsm_version(struct imsm_super
*mpb
)
612 return &mpb
->sig
[MPB_SIG_LEN
];
616 /* retrieve a disk directly from the anchor when the anchor is known to be
617 * up-to-date, currently only at load time
619 static struct imsm_disk
*__get_imsm_disk(struct imsm_super
*mpb
, __u8 index
)
621 if (index
>= mpb
->num_disks
)
623 return &mpb
->disk
[index
];
626 /* retrieve the disk description based on a index of the disk
629 static struct dl
*get_imsm_dl_disk(struct intel_super
*super
, __u8 index
)
633 for (d
= super
->disks
; d
; d
= d
->next
)
634 if (d
->index
== index
)
639 /* retrieve a disk from the parsed metadata */
640 static struct imsm_disk
*get_imsm_disk(struct intel_super
*super
, __u8 index
)
644 dl
= get_imsm_dl_disk(super
, index
);
651 /* generate a checksum directly from the anchor when the anchor is known to be
652 * up-to-date, currently only at load or write_super after coalescing
654 static __u32
__gen_imsm_checksum(struct imsm_super
*mpb
)
656 __u32 end
= mpb
->mpb_size
/ sizeof(end
);
657 __u32
*p
= (__u32
*) mpb
;
661 sum
+= __le32_to_cpu(*p
);
665 return sum
- __le32_to_cpu(mpb
->check_sum
);
668 static size_t sizeof_imsm_map(struct imsm_map
*map
)
670 return sizeof(struct imsm_map
) + sizeof(__u32
) * (map
->num_members
- 1);
673 struct imsm_map
*get_imsm_map(struct imsm_dev
*dev
, int second_map
)
675 /* A device can have 2 maps if it is in the middle of a migration.
677 * MAP_0 - we return the first map
678 * MAP_1 - we return the second map if it exists, else NULL
679 * MAP_X - we return the second map if it exists, else the first
681 struct imsm_map
*map
= &dev
->vol
.map
[0];
682 struct imsm_map
*map2
= NULL
;
684 if (dev
->vol
.migr_state
)
685 map2
= (void *)map
+ sizeof_imsm_map(map
);
687 switch (second_map
) {
704 /* return the size of the device.
705 * migr_state increases the returned size if map[0] were to be duplicated
707 static size_t sizeof_imsm_dev(struct imsm_dev
*dev
, int migr_state
)
709 size_t size
= sizeof(*dev
) - sizeof(struct imsm_map
) +
710 sizeof_imsm_map(get_imsm_map(dev
, MAP_0
));
712 /* migrating means an additional map */
713 if (dev
->vol
.migr_state
)
714 size
+= sizeof_imsm_map(get_imsm_map(dev
, MAP_1
));
716 size
+= sizeof_imsm_map(get_imsm_map(dev
, MAP_0
));
722 /* retrieve disk serial number list from a metadata update */
723 static struct disk_info
*get_disk_info(struct imsm_update_create_array
*update
)
726 struct disk_info
*inf
;
728 inf
= u
+ sizeof(*update
) - sizeof(struct imsm_dev
) +
729 sizeof_imsm_dev(&update
->dev
, 0);
735 static struct imsm_dev
*__get_imsm_dev(struct imsm_super
*mpb
, __u8 index
)
741 if (index
>= mpb
->num_raid_devs
)
744 /* devices start after all disks */
745 offset
= ((void *) &mpb
->disk
[mpb
->num_disks
]) - _mpb
;
747 for (i
= 0; i
<= index
; i
++)
749 return _mpb
+ offset
;
751 offset
+= sizeof_imsm_dev(_mpb
+ offset
, 0);
756 static struct imsm_dev
*get_imsm_dev(struct intel_super
*super
, __u8 index
)
758 struct intel_dev
*dv
;
760 if (index
>= super
->anchor
->num_raid_devs
)
762 for (dv
= super
->devlist
; dv
; dv
= dv
->next
)
763 if (dv
->index
== index
)
770 * == MAP_0 get first map
771 * == MAP_1 get second map
772 * == MAP_X than get map according to the current migr_state
774 static __u32
get_imsm_ord_tbl_ent(struct imsm_dev
*dev
,
778 struct imsm_map
*map
;
780 map
= get_imsm_map(dev
, second_map
);
782 /* top byte identifies disk under rebuild */
783 return __le32_to_cpu(map
->disk_ord_tbl
[slot
]);
786 #define ord_to_idx(ord) (((ord) << 8) >> 8)
787 static __u32
get_imsm_disk_idx(struct imsm_dev
*dev
, int slot
, int second_map
)
789 __u32 ord
= get_imsm_ord_tbl_ent(dev
, slot
, second_map
);
791 return ord_to_idx(ord
);
794 static void set_imsm_ord_tbl_ent(struct imsm_map
*map
, int slot
, __u32 ord
)
796 map
->disk_ord_tbl
[slot
] = __cpu_to_le32(ord
);
799 static int get_imsm_disk_slot(struct imsm_map
*map
, unsigned idx
)
804 for (slot
= 0; slot
< map
->num_members
; slot
++) {
805 ord
= __le32_to_cpu(map
->disk_ord_tbl
[slot
]);
806 if (ord_to_idx(ord
) == idx
)
813 static int get_imsm_raid_level(struct imsm_map
*map
)
815 if (map
->raid_level
== 1) {
816 if (map
->num_members
== 2)
822 return map
->raid_level
;
825 static int cmp_extent(const void *av
, const void *bv
)
827 const struct extent
*a
= av
;
828 const struct extent
*b
= bv
;
829 if (a
->start
< b
->start
)
831 if (a
->start
> b
->start
)
836 static int count_memberships(struct dl
*dl
, struct intel_super
*super
)
841 for (i
= 0; i
< super
->anchor
->num_raid_devs
; i
++) {
842 struct imsm_dev
*dev
= get_imsm_dev(super
, i
);
843 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
845 if (get_imsm_disk_slot(map
, dl
->index
) >= 0)
852 static __u32
imsm_min_reserved_sectors(struct intel_super
*super
);
854 static struct extent
*get_extents(struct intel_super
*super
, struct dl
*dl
)
856 /* find a list of used extents on the given physical device */
857 struct extent
*rv
, *e
;
859 int memberships
= count_memberships(dl
, super
);
862 /* trim the reserved area for spares, so they can join any array
863 * regardless of whether the OROM has assigned sectors from the
864 * IMSM_RESERVED_SECTORS region
867 reservation
= imsm_min_reserved_sectors(super
);
869 reservation
= MPB_SECTOR_CNT
+ IMSM_RESERVED_SECTORS
;
871 rv
= malloc(sizeof(struct extent
) * (memberships
+ 1));
876 for (i
= 0; i
< super
->anchor
->num_raid_devs
; i
++) {
877 struct imsm_dev
*dev
= get_imsm_dev(super
, i
);
878 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
880 if (get_imsm_disk_slot(map
, dl
->index
) >= 0) {
881 e
->start
= __le32_to_cpu(map
->pba_of_lba0
);
882 e
->size
= __le32_to_cpu(map
->blocks_per_member
);
886 qsort(rv
, memberships
, sizeof(*rv
), cmp_extent
);
888 /* determine the start of the metadata
889 * when no raid devices are defined use the default
890 * ...otherwise allow the metadata to truncate the value
891 * as is the case with older versions of imsm
894 struct extent
*last
= &rv
[memberships
- 1];
897 remainder
= __le32_to_cpu(dl
->disk
.total_blocks
) -
898 (last
->start
+ last
->size
);
899 /* round down to 1k block to satisfy precision of the kernel
903 /* make sure remainder is still sane */
904 if (remainder
< (unsigned)ROUND_UP(super
->len
, 512) >> 9)
905 remainder
= ROUND_UP(super
->len
, 512) >> 9;
906 if (reservation
> remainder
)
907 reservation
= remainder
;
909 e
->start
= __le32_to_cpu(dl
->disk
.total_blocks
) - reservation
;
914 /* try to determine how much space is reserved for metadata from
915 * the last get_extents() entry, otherwise fallback to the
918 static __u32
imsm_reserved_sectors(struct intel_super
*super
, struct dl
*dl
)
924 /* for spares just return a minimal reservation which will grow
925 * once the spare is picked up by an array
928 return MPB_SECTOR_CNT
;
930 e
= get_extents(super
, dl
);
932 return MPB_SECTOR_CNT
+ IMSM_RESERVED_SECTORS
;
934 /* scroll to last entry */
935 for (i
= 0; e
[i
].size
; i
++)
938 rv
= __le32_to_cpu(dl
->disk
.total_blocks
) - e
[i
].start
;
945 static int is_spare(struct imsm_disk
*disk
)
947 return (disk
->status
& SPARE_DISK
) == SPARE_DISK
;
950 static int is_configured(struct imsm_disk
*disk
)
952 return (disk
->status
& CONFIGURED_DISK
) == CONFIGURED_DISK
;
955 static int is_failed(struct imsm_disk
*disk
)
957 return (disk
->status
& FAILED_DISK
) == FAILED_DISK
;
960 /* try to determine how much space is reserved for metadata from
961 * the last get_extents() entry on the smallest active disk,
962 * otherwise fallback to the default
964 static __u32
imsm_min_reserved_sectors(struct intel_super
*super
)
968 __u32 min_active
, remainder
;
969 __u32 rv
= MPB_SECTOR_CNT
+ IMSM_RESERVED_SECTORS
;
970 struct dl
*dl
, *dl_min
= NULL
;
976 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
979 if (dl
->disk
.total_blocks
< min_active
|| min_active
== 0) {
981 min_active
= dl
->disk
.total_blocks
;
987 /* find last lba used by subarrays on the smallest active disk */
988 e
= get_extents(super
, dl_min
);
991 for (i
= 0; e
[i
].size
; i
++)
994 remainder
= min_active
- e
[i
].start
;
997 /* to give priority to recovery we should not require full
998 IMSM_RESERVED_SECTORS from the spare */
999 rv
= MPB_SECTOR_CNT
+ NUM_BLOCKS_DIRTY_STRIPE_REGION
;
1001 /* if real reservation is smaller use that value */
1002 return (remainder
< rv
) ? remainder
: rv
;
1005 /* Return minimum size of a spare that can be used in this array*/
1006 static unsigned long long min_acceptable_spare_size_imsm(struct supertype
*st
)
1008 struct intel_super
*super
= st
->sb
;
1012 unsigned long long rv
= 0;
1016 /* find first active disk in array */
1018 while (dl
&& (is_failed(&dl
->disk
) || dl
->index
== -1))
1022 /* find last lba used by subarrays */
1023 e
= get_extents(super
, dl
);
1026 for (i
= 0; e
[i
].size
; i
++)
1029 rv
= e
[i
-1].start
+ e
[i
-1].size
;
1032 /* add the amount of space needed for metadata */
1033 rv
= rv
+ imsm_min_reserved_sectors(super
);
1038 static int is_gen_migration(struct imsm_dev
*dev
);
1041 static __u64
blocks_per_migr_unit(struct intel_super
*super
,
1042 struct imsm_dev
*dev
);
1044 static void print_imsm_dev(struct intel_super
*super
,
1045 struct imsm_dev
*dev
,
1051 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
1052 struct imsm_map
*map2
= get_imsm_map(dev
, MAP_1
);
1056 printf("[%.16s]:\n", dev
->volume
);
1057 printf(" UUID : %s\n", uuid
);
1058 printf(" RAID Level : %d", get_imsm_raid_level(map
));
1060 printf(" <-- %d", get_imsm_raid_level(map2
));
1062 printf(" Members : %d", map
->num_members
);
1064 printf(" <-- %d", map2
->num_members
);
1066 printf(" Slots : [");
1067 for (i
= 0; i
< map
->num_members
; i
++) {
1068 ord
= get_imsm_ord_tbl_ent(dev
, i
, MAP_0
);
1069 printf("%s", ord
& IMSM_ORD_REBUILD
? "_" : "U");
1074 for (i
= 0; i
< map2
->num_members
; i
++) {
1075 ord
= get_imsm_ord_tbl_ent(dev
, i
, MAP_1
);
1076 printf("%s", ord
& IMSM_ORD_REBUILD
? "_" : "U");
1081 printf(" Failed disk : ");
1082 if (map
->failed_disk_num
== 0xff)
1085 printf("%i", map
->failed_disk_num
);
1087 slot
= get_imsm_disk_slot(map
, disk_idx
);
1089 ord
= get_imsm_ord_tbl_ent(dev
, slot
, MAP_X
);
1090 printf(" This Slot : %d%s\n", slot
,
1091 ord
& IMSM_ORD_REBUILD
? " (out-of-sync)" : "");
1093 printf(" This Slot : ?\n");
1094 sz
= __le32_to_cpu(dev
->size_high
);
1096 sz
+= __le32_to_cpu(dev
->size_low
);
1097 printf(" Array Size : %llu%s\n", (unsigned long long)sz
,
1098 human_size(sz
* 512));
1099 sz
= __le32_to_cpu(map
->blocks_per_member
);
1100 printf(" Per Dev Size : %llu%s\n", (unsigned long long)sz
,
1101 human_size(sz
* 512));
1102 printf(" Sector Offset : %u\n",
1103 __le32_to_cpu(map
->pba_of_lba0
));
1104 printf(" Num Stripes : %u\n",
1105 __le32_to_cpu(map
->num_data_stripes
));
1106 printf(" Chunk Size : %u KiB",
1107 __le16_to_cpu(map
->blocks_per_strip
) / 2);
1109 printf(" <-- %u KiB",
1110 __le16_to_cpu(map2
->blocks_per_strip
) / 2);
1112 printf(" Reserved : %d\n", __le32_to_cpu(dev
->reserved_blocks
));
1113 printf(" Migrate State : ");
1114 if (dev
->vol
.migr_state
) {
1115 if (migr_type(dev
) == MIGR_INIT
)
1116 printf("initialize\n");
1117 else if (migr_type(dev
) == MIGR_REBUILD
)
1118 printf("rebuild\n");
1119 else if (migr_type(dev
) == MIGR_VERIFY
)
1121 else if (migr_type(dev
) == MIGR_GEN_MIGR
)
1122 printf("general migration\n");
1123 else if (migr_type(dev
) == MIGR_STATE_CHANGE
)
1124 printf("state change\n");
1125 else if (migr_type(dev
) == MIGR_REPAIR
)
1128 printf("<unknown:%d>\n", migr_type(dev
));
1131 printf(" Map State : %s", map_state_str
[map
->map_state
]);
1132 if (dev
->vol
.migr_state
) {
1133 struct imsm_map
*map
= get_imsm_map(dev
, MAP_1
);
1135 printf(" <-- %s", map_state_str
[map
->map_state
]);
1136 printf("\n Checkpoint : %u ",
1137 __le32_to_cpu(dev
->vol
.curr_migr_unit
));
1138 if ((is_gen_migration(dev
)) && ((slot
> 1) || (slot
< 0)))
1141 printf("(%llu)", (unsigned long long)
1142 blocks_per_migr_unit(super
, dev
));
1145 printf(" Dirty State : %s\n", dev
->vol
.dirty
? "dirty" : "clean");
1148 static void print_imsm_disk(struct imsm_disk
*disk
, int index
, __u32 reserved
)
1150 char str
[MAX_RAID_SERIAL_LEN
+ 1];
1153 if (index
< -1 || !disk
)
1157 snprintf(str
, MAX_RAID_SERIAL_LEN
+ 1, "%s", disk
->serial
);
1159 printf(" Disk%02d Serial : %s\n", index
, str
);
1161 printf(" Disk Serial : %s\n", str
);
1162 printf(" State :%s%s%s\n", is_spare(disk
) ? " spare" : "",
1163 is_configured(disk
) ? " active" : "",
1164 is_failed(disk
) ? " failed" : "");
1165 printf(" Id : %08x\n", __le32_to_cpu(disk
->scsi_id
));
1166 sz
= __le32_to_cpu(disk
->total_blocks
) - reserved
;
1167 printf(" Usable Size : %llu%s\n", (unsigned long long)sz
,
1168 human_size(sz
* 512));
1171 void examine_migr_rec_imsm(struct intel_super
*super
)
1173 struct migr_record
*migr_rec
= super
->migr_rec
;
1174 struct imsm_super
*mpb
= super
->anchor
;
1177 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
1178 struct imsm_dev
*dev
= __get_imsm_dev(mpb
, i
);
1179 struct imsm_map
*map
;
1182 if (is_gen_migration(dev
) == 0)
1185 printf("\nMigration Record Information:");
1187 /* first map under migration */
1188 map
= get_imsm_map(dev
, MAP_0
);
1190 slot
= get_imsm_disk_slot(map
, super
->disks
->index
);
1191 if ((map
== NULL
) || (slot
> 1) || (slot
< 0)) {
1192 printf(" Empty\n ");
1193 printf("Examine one of first two disks in array\n");
1196 printf("\n Status : ");
1197 if (__le32_to_cpu(migr_rec
->rec_status
) == UNIT_SRC_NORMAL
)
1200 printf("Contains Data\n");
1201 printf(" Current Unit : %u\n",
1202 __le32_to_cpu(migr_rec
->curr_migr_unit
));
1203 printf(" Family : %u\n",
1204 __le32_to_cpu(migr_rec
->family_num
));
1205 printf(" Ascending : %u\n",
1206 __le32_to_cpu(migr_rec
->ascending_migr
));
1207 printf(" Blocks Per Unit : %u\n",
1208 __le32_to_cpu(migr_rec
->blocks_per_unit
));
1209 printf(" Dest. Depth Per Unit : %u\n",
1210 __le32_to_cpu(migr_rec
->dest_depth_per_unit
));
1211 printf(" Checkpoint Area pba : %u\n",
1212 __le32_to_cpu(migr_rec
->ckpt_area_pba
));
1213 printf(" First member lba : %u\n",
1214 __le32_to_cpu(migr_rec
->dest_1st_member_lba
));
1215 printf(" Total Number of Units : %u\n",
1216 __le32_to_cpu(migr_rec
->num_migr_units
));
1217 printf(" Size of volume : %u\n",
1218 __le32_to_cpu(migr_rec
->post_migr_vol_cap
));
1219 printf(" Expansion space for LBA64 : %u\n",
1220 __le32_to_cpu(migr_rec
->post_migr_vol_cap_hi
));
1221 printf(" Record was read from : %u\n",
1222 __le32_to_cpu(migr_rec
->ckpt_read_disk_num
));
1227 #endif /* MDASSEMBLE */
1228 /*******************************************************************************
1229 * function: imsm_check_attributes
1230 * Description: Function checks if features represented by attributes flags
1231 * are supported by mdadm.
1233 * attributes - Attributes read from metadata
1235 * 0 - passed attributes contains unsupported features flags
1236 * 1 - all features are supported
1237 ******************************************************************************/
1238 static int imsm_check_attributes(__u32 attributes
)
1241 __u32 not_supported
= MPB_ATTRIB_SUPPORTED
^0xffffffff;
1243 not_supported
&= ~MPB_ATTRIB_IGNORED
;
1245 not_supported
&= attributes
;
1246 if (not_supported
) {
1247 fprintf(stderr
, Name
"(IMSM): Unsupported attributes : %x\n",
1248 (unsigned)__le32_to_cpu(not_supported
));
1249 if (not_supported
& MPB_ATTRIB_CHECKSUM_VERIFY
) {
1250 dprintf("\t\tMPB_ATTRIB_CHECKSUM_VERIFY \n");
1251 not_supported
^= MPB_ATTRIB_CHECKSUM_VERIFY
;
1253 if (not_supported
& MPB_ATTRIB_2TB
) {
1254 dprintf("\t\tMPB_ATTRIB_2TB\n");
1255 not_supported
^= MPB_ATTRIB_2TB
;
1257 if (not_supported
& MPB_ATTRIB_RAID0
) {
1258 dprintf("\t\tMPB_ATTRIB_RAID0\n");
1259 not_supported
^= MPB_ATTRIB_RAID0
;
1261 if (not_supported
& MPB_ATTRIB_RAID1
) {
1262 dprintf("\t\tMPB_ATTRIB_RAID1\n");
1263 not_supported
^= MPB_ATTRIB_RAID1
;
1265 if (not_supported
& MPB_ATTRIB_RAID10
) {
1266 dprintf("\t\tMPB_ATTRIB_RAID10\n");
1267 not_supported
^= MPB_ATTRIB_RAID10
;
1269 if (not_supported
& MPB_ATTRIB_RAID1E
) {
1270 dprintf("\t\tMPB_ATTRIB_RAID1E\n");
1271 not_supported
^= MPB_ATTRIB_RAID1E
;
1273 if (not_supported
& MPB_ATTRIB_RAID5
) {
1274 dprintf("\t\tMPB_ATTRIB_RAID5\n");
1275 not_supported
^= MPB_ATTRIB_RAID5
;
1277 if (not_supported
& MPB_ATTRIB_RAIDCNG
) {
1278 dprintf("\t\tMPB_ATTRIB_RAIDCNG\n");
1279 not_supported
^= MPB_ATTRIB_RAIDCNG
;
1281 if (not_supported
& MPB_ATTRIB_BBM
) {
1282 dprintf("\t\tMPB_ATTRIB_BBM\n");
1283 not_supported
^= MPB_ATTRIB_BBM
;
1285 if (not_supported
& MPB_ATTRIB_CHECKSUM_VERIFY
) {
1286 dprintf("\t\tMPB_ATTRIB_CHECKSUM_VERIFY (== MPB_ATTRIB_LEGACY)\n");
1287 not_supported
^= MPB_ATTRIB_CHECKSUM_VERIFY
;
1289 if (not_supported
& MPB_ATTRIB_EXP_STRIPE_SIZE
) {
1290 dprintf("\t\tMPB_ATTRIB_EXP_STRIP_SIZE\n");
1291 not_supported
^= MPB_ATTRIB_EXP_STRIPE_SIZE
;
1293 if (not_supported
& MPB_ATTRIB_2TB_DISK
) {
1294 dprintf("\t\tMPB_ATTRIB_2TB_DISK\n");
1295 not_supported
^= MPB_ATTRIB_2TB_DISK
;
1297 if (not_supported
& MPB_ATTRIB_NEVER_USE2
) {
1298 dprintf("\t\tMPB_ATTRIB_NEVER_USE2\n");
1299 not_supported
^= MPB_ATTRIB_NEVER_USE2
;
1301 if (not_supported
& MPB_ATTRIB_NEVER_USE
) {
1302 dprintf("\t\tMPB_ATTRIB_NEVER_USE\n");
1303 not_supported
^= MPB_ATTRIB_NEVER_USE
;
1307 dprintf(Name
"(IMSM): Unknown attributes : %x\n", not_supported
);
1316 static void getinfo_super_imsm(struct supertype
*st
, struct mdinfo
*info
, char *map
);
1318 static void examine_super_imsm(struct supertype
*st
, char *homehost
)
1320 struct intel_super
*super
= st
->sb
;
1321 struct imsm_super
*mpb
= super
->anchor
;
1322 char str
[MAX_SIGNATURE_LENGTH
];
1327 __u32 reserved
= imsm_reserved_sectors(super
, super
->disks
);
1330 snprintf(str
, MPB_SIG_LEN
, "%s", mpb
->sig
);
1331 printf(" Magic : %s\n", str
);
1332 snprintf(str
, strlen(MPB_VERSION_RAID0
), "%s", get_imsm_version(mpb
));
1333 printf(" Version : %s\n", get_imsm_version(mpb
));
1334 printf(" Orig Family : %08x\n", __le32_to_cpu(mpb
->orig_family_num
));
1335 printf(" Family : %08x\n", __le32_to_cpu(mpb
->family_num
));
1336 printf(" Generation : %08x\n", __le32_to_cpu(mpb
->generation_num
));
1337 printf(" Attributes : ");
1338 if (imsm_check_attributes(mpb
->attributes
))
1339 printf("All supported\n");
1341 printf("not supported\n");
1342 getinfo_super_imsm(st
, &info
, NULL
);
1343 fname_from_uuid(st
, &info
, nbuf
, ':');
1344 printf(" UUID : %s\n", nbuf
+ 5);
1345 sum
= __le32_to_cpu(mpb
->check_sum
);
1346 printf(" Checksum : %08x %s\n", sum
,
1347 __gen_imsm_checksum(mpb
) == sum
? "correct" : "incorrect");
1348 printf(" MPB Sectors : %d\n", mpb_sectors(mpb
));
1349 printf(" Disks : %d\n", mpb
->num_disks
);
1350 printf(" RAID Devices : %d\n", mpb
->num_raid_devs
);
1351 print_imsm_disk(__get_imsm_disk(mpb
, super
->disks
->index
), super
->disks
->index
, reserved
);
1352 if (super
->bbm_log
) {
1353 struct bbm_log
*log
= super
->bbm_log
;
1356 printf("Bad Block Management Log:\n");
1357 printf(" Log Size : %d\n", __le32_to_cpu(mpb
->bbm_log_size
));
1358 printf(" Signature : %x\n", __le32_to_cpu(log
->signature
));
1359 printf(" Entry Count : %d\n", __le32_to_cpu(log
->entry_count
));
1360 printf(" Spare Blocks : %d\n", __le32_to_cpu(log
->reserved_spare_block_count
));
1361 printf(" First Spare : %llx\n",
1362 (unsigned long long) __le64_to_cpu(log
->first_spare_lba
));
1364 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
1366 struct imsm_dev
*dev
= __get_imsm_dev(mpb
, i
);
1368 super
->current_vol
= i
;
1369 getinfo_super_imsm(st
, &info
, NULL
);
1370 fname_from_uuid(st
, &info
, nbuf
, ':');
1371 print_imsm_dev(super
, dev
, nbuf
+ 5, super
->disks
->index
);
1373 for (i
= 0; i
< mpb
->num_disks
; i
++) {
1374 if (i
== super
->disks
->index
)
1376 print_imsm_disk(__get_imsm_disk(mpb
, i
), i
, reserved
);
1379 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
1380 if (dl
->index
== -1)
1381 print_imsm_disk(&dl
->disk
, -1, reserved
);
1383 examine_migr_rec_imsm(super
);
1386 static void brief_examine_super_imsm(struct supertype
*st
, int verbose
)
1388 /* We just write a generic IMSM ARRAY entry */
1391 struct intel_super
*super
= st
->sb
;
1393 if (!super
->anchor
->num_raid_devs
) {
1394 printf("ARRAY metadata=imsm\n");
1398 getinfo_super_imsm(st
, &info
, NULL
);
1399 fname_from_uuid(st
, &info
, nbuf
, ':');
1400 printf("ARRAY metadata=imsm UUID=%s\n", nbuf
+ 5);
1403 static void brief_examine_subarrays_imsm(struct supertype
*st
, int verbose
)
1405 /* We just write a generic IMSM ARRAY entry */
1409 struct intel_super
*super
= st
->sb
;
1412 if (!super
->anchor
->num_raid_devs
)
1415 getinfo_super_imsm(st
, &info
, NULL
);
1416 fname_from_uuid(st
, &info
, nbuf
, ':');
1417 for (i
= 0; i
< super
->anchor
->num_raid_devs
; i
++) {
1418 struct imsm_dev
*dev
= get_imsm_dev(super
, i
);
1420 super
->current_vol
= i
;
1421 getinfo_super_imsm(st
, &info
, NULL
);
1422 fname_from_uuid(st
, &info
, nbuf1
, ':');
1423 printf("ARRAY /dev/md/%.16s container=%s member=%d UUID=%s\n",
1424 dev
->volume
, nbuf
+ 5, i
, nbuf1
+ 5);
1428 static void export_examine_super_imsm(struct supertype
*st
)
1430 struct intel_super
*super
= st
->sb
;
1431 struct imsm_super
*mpb
= super
->anchor
;
1435 getinfo_super_imsm(st
, &info
, NULL
);
1436 fname_from_uuid(st
, &info
, nbuf
, ':');
1437 printf("MD_METADATA=imsm\n");
1438 printf("MD_LEVEL=container\n");
1439 printf("MD_UUID=%s\n", nbuf
+5);
1440 printf("MD_DEVICES=%u\n", mpb
->num_disks
);
1443 static void detail_super_imsm(struct supertype
*st
, char *homehost
)
1448 getinfo_super_imsm(st
, &info
, NULL
);
1449 fname_from_uuid(st
, &info
, nbuf
, ':');
1450 printf("\n UUID : %s\n", nbuf
+ 5);
1453 static void brief_detail_super_imsm(struct supertype
*st
)
1457 getinfo_super_imsm(st
, &info
, NULL
);
1458 fname_from_uuid(st
, &info
, nbuf
, ':');
1459 printf(" UUID=%s", nbuf
+ 5);
1462 static int imsm_read_serial(int fd
, char *devname
, __u8
*serial
);
1463 static void fd2devname(int fd
, char *name
);
1465 static int ahci_enumerate_ports(const char *hba_path
, int port_count
, int host_base
, int verbose
)
1467 /* dump an unsorted list of devices attached to AHCI Intel storage
1468 * controller, as well as non-connected ports
1470 int hba_len
= strlen(hba_path
) + 1;
1475 unsigned long port_mask
= (1 << port_count
) - 1;
1477 if (port_count
> (int)sizeof(port_mask
) * 8) {
1479 fprintf(stderr
, Name
": port_count %d out of range\n", port_count
);
1483 /* scroll through /sys/dev/block looking for devices attached to
1486 dir
= opendir("/sys/dev/block");
1487 for (ent
= dir
? readdir(dir
) : NULL
; ent
; ent
= readdir(dir
)) {
1498 if (sscanf(ent
->d_name
, "%d:%d", &major
, &minor
) != 2)
1500 path
= devt_to_devpath(makedev(major
, minor
));
1503 if (!path_attached_to_hba(path
, hba_path
)) {
1509 /* retrieve the scsi device type */
1510 if (asprintf(&device
, "/sys/dev/block/%d:%d/device/xxxxxxx", major
, minor
) < 0) {
1512 fprintf(stderr
, Name
": failed to allocate 'device'\n");
1516 sprintf(device
, "/sys/dev/block/%d:%d/device/type", major
, minor
);
1517 if (load_sys(device
, buf
) != 0) {
1519 fprintf(stderr
, Name
": failed to read device type for %s\n",
1525 type
= strtoul(buf
, NULL
, 10);
1527 /* if it's not a disk print the vendor and model */
1528 if (!(type
== 0 || type
== 7 || type
== 14)) {
1531 sprintf(device
, "/sys/dev/block/%d:%d/device/vendor", major
, minor
);
1532 if (load_sys(device
, buf
) == 0) {
1533 strncpy(vendor
, buf
, sizeof(vendor
));
1534 vendor
[sizeof(vendor
) - 1] = '\0';
1535 c
= (char *) &vendor
[sizeof(vendor
) - 1];
1536 while (isspace(*c
) || *c
== '\0')
1540 sprintf(device
, "/sys/dev/block/%d:%d/device/model", major
, minor
);
1541 if (load_sys(device
, buf
) == 0) {
1542 strncpy(model
, buf
, sizeof(model
));
1543 model
[sizeof(model
) - 1] = '\0';
1544 c
= (char *) &model
[sizeof(model
) - 1];
1545 while (isspace(*c
) || *c
== '\0')
1549 if (vendor
[0] && model
[0])
1550 sprintf(buf
, "%.64s %.64s", vendor
, model
);
1552 switch (type
) { /* numbers from hald/linux/device.c */
1553 case 1: sprintf(buf
, "tape"); break;
1554 case 2: sprintf(buf
, "printer"); break;
1555 case 3: sprintf(buf
, "processor"); break;
1557 case 5: sprintf(buf
, "cdrom"); break;
1558 case 6: sprintf(buf
, "scanner"); break;
1559 case 8: sprintf(buf
, "media_changer"); break;
1560 case 9: sprintf(buf
, "comm"); break;
1561 case 12: sprintf(buf
, "raid"); break;
1562 default: sprintf(buf
, "unknown");
1568 /* chop device path to 'host%d' and calculate the port number */
1569 c
= strchr(&path
[hba_len
], '/');
1572 fprintf(stderr
, Name
": %s - invalid path name\n", path
+ hba_len
);
1577 if (sscanf(&path
[hba_len
], "host%d", &port
) == 1)
1581 *c
= '/'; /* repair the full string */
1582 fprintf(stderr
, Name
": failed to determine port number for %s\n",
1589 /* mark this port as used */
1590 port_mask
&= ~(1 << port
);
1592 /* print out the device information */
1594 printf(" Port%d : - non-disk device (%s) -\n", port
, buf
);
1598 fd
= dev_open(ent
->d_name
, O_RDONLY
);
1600 printf(" Port%d : - disk info unavailable -\n", port
);
1602 fd2devname(fd
, buf
);
1603 printf(" Port%d : %s", port
, buf
);
1604 if (imsm_read_serial(fd
, NULL
, (__u8
*) buf
) == 0)
1605 printf(" (%.*s)\n", MAX_RAID_SERIAL_LEN
, buf
);
1620 for (i
= 0; i
< port_count
; i
++)
1621 if (port_mask
& (1 << i
))
1622 printf(" Port%d : - no device attached -\n", i
);
1628 static void print_found_intel_controllers(struct sys_dev
*elem
)
1630 for (; elem
; elem
= elem
->next
) {
1631 fprintf(stderr
, Name
": found Intel(R) ");
1632 if (elem
->type
== SYS_DEV_SATA
)
1633 fprintf(stderr
, "SATA ");
1634 else if (elem
->type
== SYS_DEV_SAS
)
1635 fprintf(stderr
, "SAS ");
1636 fprintf(stderr
, "RAID controller");
1638 fprintf(stderr
, " at %s", elem
->pci_id
);
1639 fprintf(stderr
, ".\n");
1644 static int ahci_get_port_count(const char *hba_path
, int *port_count
)
1651 if ((dir
= opendir(hba_path
)) == NULL
)
1654 for (ent
= readdir(dir
); ent
; ent
= readdir(dir
)) {
1657 if (sscanf(ent
->d_name
, "host%d", &host
) != 1)
1659 if (*port_count
== 0)
1661 else if (host
< host_base
)
1664 if (host
+ 1 > *port_count
+ host_base
)
1665 *port_count
= host
+ 1 - host_base
;
1671 static void print_imsm_capability(const struct imsm_orom
*orom
)
1673 printf(" Platform : Intel(R) Matrix Storage Manager\n");
1674 printf(" Version : %d.%d.%d.%d\n", orom
->major_ver
, orom
->minor_ver
,
1675 orom
->hotfix_ver
, orom
->build
);
1676 printf(" RAID Levels :%s%s%s%s%s\n",
1677 imsm_orom_has_raid0(orom
) ? " raid0" : "",
1678 imsm_orom_has_raid1(orom
) ? " raid1" : "",
1679 imsm_orom_has_raid1e(orom
) ? " raid1e" : "",
1680 imsm_orom_has_raid10(orom
) ? " raid10" : "",
1681 imsm_orom_has_raid5(orom
) ? " raid5" : "");
1682 printf(" Chunk Sizes :%s%s%s%s%s%s%s%s%s%s%s%s%s%s%s%s\n",
1683 imsm_orom_has_chunk(orom
, 2) ? " 2k" : "",
1684 imsm_orom_has_chunk(orom
, 4) ? " 4k" : "",
1685 imsm_orom_has_chunk(orom
, 8) ? " 8k" : "",
1686 imsm_orom_has_chunk(orom
, 16) ? " 16k" : "",
1687 imsm_orom_has_chunk(orom
, 32) ? " 32k" : "",
1688 imsm_orom_has_chunk(orom
, 64) ? " 64k" : "",
1689 imsm_orom_has_chunk(orom
, 128) ? " 128k" : "",
1690 imsm_orom_has_chunk(orom
, 256) ? " 256k" : "",
1691 imsm_orom_has_chunk(orom
, 512) ? " 512k" : "",
1692 imsm_orom_has_chunk(orom
, 1024*1) ? " 1M" : "",
1693 imsm_orom_has_chunk(orom
, 1024*2) ? " 2M" : "",
1694 imsm_orom_has_chunk(orom
, 1024*4) ? " 4M" : "",
1695 imsm_orom_has_chunk(orom
, 1024*8) ? " 8M" : "",
1696 imsm_orom_has_chunk(orom
, 1024*16) ? " 16M" : "",
1697 imsm_orom_has_chunk(orom
, 1024*32) ? " 32M" : "",
1698 imsm_orom_has_chunk(orom
, 1024*64) ? " 64M" : "");
1699 printf(" Max Disks : %d\n", orom
->tds
);
1700 printf(" Max Volumes : %d\n", orom
->vpa
);
1704 static int detail_platform_imsm(int verbose
, int enumerate_only
)
1706 /* There are two components to imsm platform support, the ahci SATA
1707 * controller and the option-rom. To find the SATA controller we
1708 * simply look in /sys/bus/pci/drivers/ahci to see if an ahci
1709 * controller with the Intel vendor id is present. This approach
1710 * allows mdadm to leverage the kernel's ahci detection logic, with the
1711 * caveat that if ahci.ko is not loaded mdadm will not be able to
1712 * detect platform raid capabilities. The option-rom resides in a
1713 * platform "Adapter ROM". We scan for its signature to retrieve the
1714 * platform capabilities. If raid support is disabled in the BIOS the
1715 * option-rom capability structure will not be available.
1717 const struct imsm_orom
*orom
;
1718 struct sys_dev
*list
, *hba
;
1723 if (enumerate_only
) {
1724 if (check_env("IMSM_NO_PLATFORM"))
1726 list
= find_intel_devices();
1729 for (hba
= list
; hba
; hba
= hba
->next
) {
1730 orom
= find_imsm_capability(hba
->type
);
1736 free_sys_dev(&list
);
1740 list
= find_intel_devices();
1743 fprintf(stderr
, Name
": no active Intel(R) RAID "
1744 "controller found.\n");
1745 free_sys_dev(&list
);
1748 print_found_intel_controllers(list
);
1750 for (hba
= list
; hba
; hba
= hba
->next
) {
1751 orom
= find_imsm_capability(hba
->type
);
1753 fprintf(stderr
, Name
": imsm capabilities not found for controller: %s (type %s)\n",
1754 hba
->path
, get_sys_dev_type(hba
->type
));
1756 print_imsm_capability(orom
);
1759 for (hba
= list
; hba
; hba
= hba
->next
) {
1760 printf(" I/O Controller : %s (%s)\n",
1761 hba
->path
, get_sys_dev_type(hba
->type
));
1763 if (hba
->type
== SYS_DEV_SATA
) {
1764 host_base
= ahci_get_port_count(hba
->path
, &port_count
);
1765 if (ahci_enumerate_ports(hba
->path
, port_count
, host_base
, verbose
)) {
1767 fprintf(stderr
, Name
": failed to enumerate "
1768 "ports on SATA controller at %s.", hba
->pci_id
);
1774 free_sys_dev(&list
);
1779 static int match_home_imsm(struct supertype
*st
, char *homehost
)
1781 /* the imsm metadata format does not specify any host
1782 * identification information. We return -1 since we can never
1783 * confirm nor deny whether a given array is "meant" for this
1784 * host. We rely on compare_super and the 'family_num' fields to
1785 * exclude member disks that do not belong, and we rely on
1786 * mdadm.conf to specify the arrays that should be assembled.
1787 * Auto-assembly may still pick up "foreign" arrays.
1793 static void uuid_from_super_imsm(struct supertype
*st
, int uuid
[4])
1795 /* The uuid returned here is used for:
1796 * uuid to put into bitmap file (Create, Grow)
1797 * uuid for backup header when saving critical section (Grow)
1798 * comparing uuids when re-adding a device into an array
1799 * In these cases the uuid required is that of the data-array,
1800 * not the device-set.
1801 * uuid to recognise same set when adding a missing device back
1802 * to an array. This is a uuid for the device-set.
1804 * For each of these we can make do with a truncated
1805 * or hashed uuid rather than the original, as long as
1807 * In each case the uuid required is that of the data-array,
1808 * not the device-set.
1810 /* imsm does not track uuid's so we synthesis one using sha1 on
1811 * - The signature (Which is constant for all imsm array, but no matter)
1812 * - the orig_family_num of the container
1813 * - the index number of the volume
1814 * - the 'serial' number of the volume.
1815 * Hopefully these are all constant.
1817 struct intel_super
*super
= st
->sb
;
1820 struct sha1_ctx ctx
;
1821 struct imsm_dev
*dev
= NULL
;
1824 /* some mdadm versions failed to set ->orig_family_num, in which
1825 * case fall back to ->family_num. orig_family_num will be
1826 * fixed up with the first metadata update.
1828 family_num
= super
->anchor
->orig_family_num
;
1829 if (family_num
== 0)
1830 family_num
= super
->anchor
->family_num
;
1831 sha1_init_ctx(&ctx
);
1832 sha1_process_bytes(super
->anchor
->sig
, MPB_SIG_LEN
, &ctx
);
1833 sha1_process_bytes(&family_num
, sizeof(__u32
), &ctx
);
1834 if (super
->current_vol
>= 0)
1835 dev
= get_imsm_dev(super
, super
->current_vol
);
1837 __u32 vol
= super
->current_vol
;
1838 sha1_process_bytes(&vol
, sizeof(vol
), &ctx
);
1839 sha1_process_bytes(dev
->volume
, MAX_RAID_SERIAL_LEN
, &ctx
);
1841 sha1_finish_ctx(&ctx
, buf
);
1842 memcpy(uuid
, buf
, 4*4);
1847 get_imsm_numerical_version(struct imsm_super
*mpb
, int *m
, int *p
)
1849 __u8
*v
= get_imsm_version(mpb
);
1850 __u8
*end
= mpb
->sig
+ MAX_SIGNATURE_LENGTH
;
1851 char major
[] = { 0, 0, 0 };
1852 char minor
[] = { 0 ,0, 0 };
1853 char patch
[] = { 0, 0, 0 };
1854 char *ver_parse
[] = { major
, minor
, patch
};
1858 while (*v
!= '\0' && v
< end
) {
1859 if (*v
!= '.' && j
< 2)
1860 ver_parse
[i
][j
++] = *v
;
1868 *m
= strtol(minor
, NULL
, 0);
1869 *p
= strtol(patch
, NULL
, 0);
1873 static __u32
migr_strip_blocks_resync(struct imsm_dev
*dev
)
1875 /* migr_strip_size when repairing or initializing parity */
1876 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
1877 __u32 chunk
= __le32_to_cpu(map
->blocks_per_strip
);
1879 switch (get_imsm_raid_level(map
)) {
1884 return 128*1024 >> 9;
1888 static __u32
migr_strip_blocks_rebuild(struct imsm_dev
*dev
)
1890 /* migr_strip_size when rebuilding a degraded disk, no idea why
1891 * this is different than migr_strip_size_resync(), but it's good
1894 struct imsm_map
*map
= get_imsm_map(dev
, MAP_1
);
1895 __u32 chunk
= __le32_to_cpu(map
->blocks_per_strip
);
1897 switch (get_imsm_raid_level(map
)) {
1900 if (map
->num_members
% map
->num_domains
== 0)
1901 return 128*1024 >> 9;
1905 return max((__u32
) 64*1024 >> 9, chunk
);
1907 return 128*1024 >> 9;
1911 static __u32
num_stripes_per_unit_resync(struct imsm_dev
*dev
)
1913 struct imsm_map
*lo
= get_imsm_map(dev
, MAP_0
);
1914 struct imsm_map
*hi
= get_imsm_map(dev
, MAP_1
);
1915 __u32 lo_chunk
= __le32_to_cpu(lo
->blocks_per_strip
);
1916 __u32 hi_chunk
= __le32_to_cpu(hi
->blocks_per_strip
);
1918 return max((__u32
) 1, hi_chunk
/ lo_chunk
);
1921 static __u32
num_stripes_per_unit_rebuild(struct imsm_dev
*dev
)
1923 struct imsm_map
*lo
= get_imsm_map(dev
, MAP_0
);
1924 int level
= get_imsm_raid_level(lo
);
1926 if (level
== 1 || level
== 10) {
1927 struct imsm_map
*hi
= get_imsm_map(dev
, MAP_1
);
1929 return hi
->num_domains
;
1931 return num_stripes_per_unit_resync(dev
);
1934 static __u8
imsm_num_data_members(struct imsm_dev
*dev
, int second_map
)
1936 /* named 'imsm_' because raid0, raid1 and raid10
1937 * counter-intuitively have the same number of data disks
1939 struct imsm_map
*map
= get_imsm_map(dev
, second_map
);
1941 switch (get_imsm_raid_level(map
)) {
1945 return map
->num_members
;
1947 return map
->num_members
- 1;
1949 dprintf("%s: unsupported raid level\n", __func__
);
1954 static __u32
parity_segment_depth(struct imsm_dev
*dev
)
1956 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
1957 __u32 chunk
= __le32_to_cpu(map
->blocks_per_strip
);
1959 switch(get_imsm_raid_level(map
)) {
1962 return chunk
* map
->num_domains
;
1964 return chunk
* map
->num_members
;
1970 static __u32
map_migr_block(struct imsm_dev
*dev
, __u32 block
)
1972 struct imsm_map
*map
= get_imsm_map(dev
, MAP_1
);
1973 __u32 chunk
= __le32_to_cpu(map
->blocks_per_strip
);
1974 __u32 strip
= block
/ chunk
;
1976 switch (get_imsm_raid_level(map
)) {
1979 __u32 vol_strip
= (strip
* map
->num_domains
) + 1;
1980 __u32 vol_stripe
= vol_strip
/ map
->num_members
;
1982 return vol_stripe
* chunk
+ block
% chunk
;
1984 __u32 stripe
= strip
/ (map
->num_members
- 1);
1986 return stripe
* chunk
+ block
% chunk
;
1993 static __u64
blocks_per_migr_unit(struct intel_super
*super
,
1994 struct imsm_dev
*dev
)
1996 /* calculate the conversion factor between per member 'blocks'
1997 * (md/{resync,rebuild}_start) and imsm migration units, return
1998 * 0 for the 'not migrating' and 'unsupported migration' cases
2000 if (!dev
->vol
.migr_state
)
2003 switch (migr_type(dev
)) {
2004 case MIGR_GEN_MIGR
: {
2005 struct migr_record
*migr_rec
= super
->migr_rec
;
2006 return __le32_to_cpu(migr_rec
->blocks_per_unit
);
2011 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
2012 __u32 stripes_per_unit
;
2013 __u32 blocks_per_unit
;
2022 /* yes, this is really the translation of migr_units to
2023 * per-member blocks in the 'resync' case
2025 stripes_per_unit
= num_stripes_per_unit_resync(dev
);
2026 migr_chunk
= migr_strip_blocks_resync(dev
);
2027 disks
= imsm_num_data_members(dev
, MAP_0
);
2028 blocks_per_unit
= stripes_per_unit
* migr_chunk
* disks
;
2029 stripe
= __le16_to_cpu(map
->blocks_per_strip
) * disks
;
2030 segment
= blocks_per_unit
/ stripe
;
2031 block_rel
= blocks_per_unit
- segment
* stripe
;
2032 parity_depth
= parity_segment_depth(dev
);
2033 block_map
= map_migr_block(dev
, block_rel
);
2034 return block_map
+ parity_depth
* segment
;
2036 case MIGR_REBUILD
: {
2037 __u32 stripes_per_unit
;
2040 stripes_per_unit
= num_stripes_per_unit_rebuild(dev
);
2041 migr_chunk
= migr_strip_blocks_rebuild(dev
);
2042 return migr_chunk
* stripes_per_unit
;
2044 case MIGR_STATE_CHANGE
:
2050 static int imsm_level_to_layout(int level
)
2058 return ALGORITHM_LEFT_ASYMMETRIC
;
2065 /*******************************************************************************
2066 * Function: read_imsm_migr_rec
2067 * Description: Function reads imsm migration record from last sector of disk
2069 * fd : disk descriptor
2070 * super : metadata info
2074 ******************************************************************************/
2075 static int read_imsm_migr_rec(int fd
, struct intel_super
*super
)
2078 unsigned long long dsize
;
2080 get_dev_size(fd
, NULL
, &dsize
);
2081 if (lseek64(fd
, dsize
- MIGR_REC_POSITION
, SEEK_SET
) < 0) {
2083 Name
": Cannot seek to anchor block: %s\n",
2087 if (read(fd
, super
->migr_rec_buf
, MIGR_REC_BUF_SIZE
) !=
2088 MIGR_REC_BUF_SIZE
) {
2090 Name
": Cannot read migr record block: %s\n",
2100 static struct imsm_dev
*imsm_get_device_during_migration(
2101 struct intel_super
*super
)
2104 struct intel_dev
*dv
;
2106 for (dv
= super
->devlist
; dv
; dv
= dv
->next
) {
2107 if (is_gen_migration(dv
->dev
))
2113 /*******************************************************************************
2114 * Function: load_imsm_migr_rec
2115 * Description: Function reads imsm migration record (it is stored at the last
2118 * super : imsm internal array info
2119 * info : general array info
2123 * -2 : no migration in progress
2124 ******************************************************************************/
2125 static int load_imsm_migr_rec(struct intel_super
*super
, struct mdinfo
*info
)
2128 struct dl
*dl
= NULL
;
2132 struct imsm_dev
*dev
;
2133 struct imsm_map
*map
= NULL
;
2136 /* find map under migration */
2137 dev
= imsm_get_device_during_migration(super
);
2138 /* nothing to load,no migration in progress?
2142 map
= get_imsm_map(dev
, MAP_0
);
2145 for (sd
= info
->devs
; sd
; sd
= sd
->next
) {
2146 /* skip spare and failed disks
2148 if (sd
->disk
.raid_disk
< 0)
2150 /* read only from one of the first two slots */
2152 slot
= get_imsm_disk_slot(map
,
2153 sd
->disk
.raid_disk
);
2154 if ((map
== NULL
) || (slot
> 1) || (slot
< 0))
2157 sprintf(nm
, "%d:%d", sd
->disk
.major
, sd
->disk
.minor
);
2158 fd
= dev_open(nm
, O_RDONLY
);
2164 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
2165 /* skip spare and failed disks
2169 /* read only from one of the first two slots */
2171 slot
= get_imsm_disk_slot(map
, dl
->index
);
2172 if ((map
== NULL
) || (slot
> 1) || (slot
< 0))
2174 sprintf(nm
, "%d:%d", dl
->major
, dl
->minor
);
2175 fd
= dev_open(nm
, O_RDONLY
);
2182 retval
= read_imsm_migr_rec(fd
, super
);
2191 /*******************************************************************************
2192 * function: imsm_create_metadata_checkpoint_update
2193 * Description: It creates update for checkpoint change.
2195 * super : imsm internal array info
2196 * u : pointer to prepared update
2199 * If length is equal to 0, input pointer u contains no update
2200 ******************************************************************************/
2201 static int imsm_create_metadata_checkpoint_update(
2202 struct intel_super
*super
,
2203 struct imsm_update_general_migration_checkpoint
**u
)
2206 int update_memory_size
= 0;
2208 dprintf("imsm_create_metadata_checkpoint_update(enter)\n");
2214 /* size of all update data without anchor */
2215 update_memory_size
=
2216 sizeof(struct imsm_update_general_migration_checkpoint
);
2218 *u
= calloc(1, update_memory_size
);
2220 dprintf("error: cannot get memory for "
2221 "imsm_create_metadata_checkpoint_update update\n");
2224 (*u
)->type
= update_general_migration_checkpoint
;
2225 (*u
)->curr_migr_unit
= __le32_to_cpu(super
->migr_rec
->curr_migr_unit
);
2226 dprintf("imsm_create_metadata_checkpoint_update: prepared for %u\n",
2227 (*u
)->curr_migr_unit
);
2229 return update_memory_size
;
2233 static void imsm_update_metadata_locally(struct supertype
*st
,
2234 void *buf
, int len
);
2236 /*******************************************************************************
2237 * Function: write_imsm_migr_rec
2238 * Description: Function writes imsm migration record
2239 * (at the last sector of disk)
2241 * super : imsm internal array info
2245 ******************************************************************************/
2246 static int write_imsm_migr_rec(struct supertype
*st
)
2248 struct intel_super
*super
= st
->sb
;
2249 unsigned long long dsize
;
2255 struct imsm_update_general_migration_checkpoint
*u
;
2256 struct imsm_dev
*dev
;
2257 struct imsm_map
*map
= NULL
;
2259 /* find map under migration */
2260 dev
= imsm_get_device_during_migration(super
);
2261 /* if no migration, write buffer anyway to clear migr_record
2262 * on disk based on first available device
2265 dev
= get_imsm_dev(super
, super
->current_vol
< 0 ? 0 :
2266 super
->current_vol
);
2268 map
= get_imsm_map(dev
, MAP_0
);
2270 for (sd
= super
->disks
; sd
; sd
= sd
->next
) {
2273 /* skip failed and spare devices */
2276 /* write to 2 first slots only */
2278 slot
= get_imsm_disk_slot(map
, sd
->index
);
2279 if ((map
== NULL
) || (slot
> 1) || (slot
< 0))
2282 sprintf(nm
, "%d:%d", sd
->major
, sd
->minor
);
2283 fd
= dev_open(nm
, O_RDWR
);
2286 get_dev_size(fd
, NULL
, &dsize
);
2287 if (lseek64(fd
, dsize
- MIGR_REC_POSITION
, SEEK_SET
) < 0) {
2289 Name
": Cannot seek to anchor block: %s\n",
2293 if (write(fd
, super
->migr_rec_buf
, MIGR_REC_BUF_SIZE
) !=
2294 MIGR_REC_BUF_SIZE
) {
2296 Name
": Cannot write migr record block: %s\n",
2303 /* update checkpoint information in metadata */
2304 len
= imsm_create_metadata_checkpoint_update(super
, &u
);
2307 dprintf("imsm: Cannot prepare update\n");
2310 /* update metadata locally */
2311 imsm_update_metadata_locally(st
, u
, len
);
2312 /* and possibly remotely */
2313 if (st
->update_tail
) {
2314 append_metadata_update(st
, u
, len
);
2315 /* during reshape we do all work inside metadata handler
2316 * manage_reshape(), so metadata update has to be triggered
2319 flush_metadata_updates(st
);
2320 st
->update_tail
= &st
->updates
;
2330 #endif /* MDASSEMBLE */
2332 /* spare/missing disks activations are not allowe when
2333 * array/container performs reshape operation, because
2334 * all arrays in container works on the same disks set
2336 int imsm_reshape_blocks_arrays_changes(struct intel_super
*super
)
2339 struct intel_dev
*i_dev
;
2340 struct imsm_dev
*dev
;
2342 /* check whole container
2344 for (i_dev
= super
->devlist
; i_dev
; i_dev
= i_dev
->next
) {
2346 if (is_gen_migration(dev
)) {
2347 /* No repair during any migration in container
2356 static void getinfo_super_imsm_volume(struct supertype
*st
, struct mdinfo
*info
, char *dmap
)
2358 struct intel_super
*super
= st
->sb
;
2359 struct migr_record
*migr_rec
= super
->migr_rec
;
2360 struct imsm_dev
*dev
= get_imsm_dev(super
, super
->current_vol
);
2361 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
2362 struct imsm_map
*prev_map
= get_imsm_map(dev
, MAP_1
);
2363 struct imsm_map
*map_to_analyse
= map
;
2366 unsigned int component_size_alligment
;
2367 int map_disks
= info
->array
.raid_disks
;
2369 memset(info
, 0, sizeof(*info
));
2371 map_to_analyse
= prev_map
;
2373 dl
= super
->current_disk
;
2375 info
->container_member
= super
->current_vol
;
2376 info
->array
.raid_disks
= map
->num_members
;
2377 info
->array
.level
= get_imsm_raid_level(map_to_analyse
);
2378 info
->array
.layout
= imsm_level_to_layout(info
->array
.level
);
2379 info
->array
.md_minor
= -1;
2380 info
->array
.ctime
= 0;
2381 info
->array
.utime
= 0;
2382 info
->array
.chunk_size
=
2383 __le16_to_cpu(map_to_analyse
->blocks_per_strip
) << 9;
2384 info
->array
.state
= !dev
->vol
.dirty
;
2385 info
->custom_array_size
= __le32_to_cpu(dev
->size_high
);
2386 info
->custom_array_size
<<= 32;
2387 info
->custom_array_size
|= __le32_to_cpu(dev
->size_low
);
2388 info
->recovery_blocked
= imsm_reshape_blocks_arrays_changes(st
->sb
);
2390 if (is_gen_migration(dev
)) {
2391 info
->reshape_active
= 1;
2392 info
->new_level
= get_imsm_raid_level(map
);
2393 info
->new_layout
= imsm_level_to_layout(info
->new_level
);
2394 info
->new_chunk
= __le16_to_cpu(map
->blocks_per_strip
) << 9;
2395 info
->delta_disks
= map
->num_members
- prev_map
->num_members
;
2396 if (info
->delta_disks
) {
2397 /* this needs to be applied to every array
2400 info
->reshape_active
= CONTAINER_RESHAPE
;
2402 /* We shape information that we give to md might have to be
2403 * modify to cope with md's requirement for reshaping arrays.
2404 * For example, when reshaping a RAID0, md requires it to be
2405 * presented as a degraded RAID4.
2406 * Also if a RAID0 is migrating to a RAID5 we need to specify
2407 * the array as already being RAID5, but the 'before' layout
2408 * is a RAID4-like layout.
2410 switch (info
->array
.level
) {
2412 switch(info
->new_level
) {
2414 /* conversion is happening as RAID4 */
2415 info
->array
.level
= 4;
2416 info
->array
.raid_disks
+= 1;
2419 /* conversion is happening as RAID5 */
2420 info
->array
.level
= 5;
2421 info
->array
.layout
= ALGORITHM_PARITY_N
;
2422 info
->delta_disks
-= 1;
2425 /* FIXME error message */
2426 info
->array
.level
= UnSet
;
2432 info
->new_level
= UnSet
;
2433 info
->new_layout
= UnSet
;
2434 info
->new_chunk
= info
->array
.chunk_size
;
2435 info
->delta_disks
= 0;
2439 info
->disk
.major
= dl
->major
;
2440 info
->disk
.minor
= dl
->minor
;
2441 info
->disk
.number
= dl
->index
;
2442 info
->disk
.raid_disk
= get_imsm_disk_slot(map_to_analyse
,
2446 info
->data_offset
= __le32_to_cpu(map_to_analyse
->pba_of_lba0
);
2447 info
->component_size
=
2448 __le32_to_cpu(map_to_analyse
->blocks_per_member
);
2450 /* check component size aligment
2452 component_size_alligment
=
2453 info
->component_size
% (info
->array
.chunk_size
/512);
2455 if (component_size_alligment
&&
2456 (info
->array
.level
!= 1) && (info
->array
.level
!= UnSet
)) {
2457 dprintf("imsm: reported component size alligned from %llu ",
2458 info
->component_size
);
2459 info
->component_size
-= component_size_alligment
;
2460 dprintf("to %llu (%i).\n",
2461 info
->component_size
, component_size_alligment
);
2464 memset(info
->uuid
, 0, sizeof(info
->uuid
));
2465 info
->recovery_start
= MaxSector
;
2467 info
->reshape_progress
= 0;
2468 info
->resync_start
= MaxSector
;
2469 if ((map_to_analyse
->map_state
== IMSM_T_STATE_UNINITIALIZED
||
2471 imsm_reshape_blocks_arrays_changes(super
) == 0) {
2472 info
->resync_start
= 0;
2474 if (dev
->vol
.migr_state
) {
2475 switch (migr_type(dev
)) {
2478 __u64 blocks_per_unit
= blocks_per_migr_unit(super
,
2480 __u64 units
= __le32_to_cpu(dev
->vol
.curr_migr_unit
);
2482 info
->resync_start
= blocks_per_unit
* units
;
2485 case MIGR_GEN_MIGR
: {
2486 __u64 blocks_per_unit
= blocks_per_migr_unit(super
,
2488 __u64 units
= __le32_to_cpu(migr_rec
->curr_migr_unit
);
2489 unsigned long long array_blocks
;
2492 if (__le32_to_cpu(migr_rec
->ascending_migr
) &&
2494 (__le32_to_cpu(migr_rec
->num_migr_units
)-1)) &&
2495 (super
->migr_rec
->rec_status
==
2496 __cpu_to_le32(UNIT_SRC_IN_CP_AREA
)))
2499 info
->reshape_progress
= blocks_per_unit
* units
;
2501 dprintf("IMSM: General Migration checkpoint : %llu "
2502 "(%llu) -> read reshape progress : %llu\n",
2503 (unsigned long long)units
,
2504 (unsigned long long)blocks_per_unit
,
2505 info
->reshape_progress
);
2507 used_disks
= imsm_num_data_members(dev
, MAP_1
);
2508 if (used_disks
> 0) {
2509 array_blocks
= map
->blocks_per_member
*
2511 /* round array size down to closest MB
2513 info
->custom_array_size
= (array_blocks
2514 >> SECT_PER_MB_SHIFT
)
2515 << SECT_PER_MB_SHIFT
;
2519 /* we could emulate the checkpointing of
2520 * 'sync_action=check' migrations, but for now
2521 * we just immediately complete them
2524 /* this is handled by container_content_imsm() */
2525 case MIGR_STATE_CHANGE
:
2526 /* FIXME handle other migrations */
2528 /* we are not dirty, so... */
2529 info
->resync_start
= MaxSector
;
2533 strncpy(info
->name
, (char *) dev
->volume
, MAX_RAID_SERIAL_LEN
);
2534 info
->name
[MAX_RAID_SERIAL_LEN
] = 0;
2536 info
->array
.major_version
= -1;
2537 info
->array
.minor_version
= -2;
2538 devname
= devnum2devname(st
->container_dev
);
2539 *info
->text_version
= '\0';
2541 sprintf(info
->text_version
, "/%s/%d", devname
, info
->container_member
);
2543 info
->safe_mode_delay
= 4000; /* 4 secs like the Matrix driver */
2544 uuid_from_super_imsm(st
, info
->uuid
);
2548 for (i
=0; i
<map_disks
; i
++) {
2550 if (i
< info
->array
.raid_disks
) {
2551 struct imsm_disk
*dsk
;
2552 j
= get_imsm_disk_idx(dev
, i
, MAP_X
);
2553 dsk
= get_imsm_disk(super
, j
);
2554 if (dsk
&& (dsk
->status
& CONFIGURED_DISK
))
2561 static __u8
imsm_check_degraded(struct intel_super
*super
, struct imsm_dev
*dev
,
2562 int failed
, int look_in_map
);
2564 static int imsm_count_failed(struct intel_super
*super
, struct imsm_dev
*dev
,
2569 static void manage_second_map(struct intel_super
*super
, struct imsm_dev
*dev
)
2571 if (is_gen_migration(dev
)) {
2574 struct imsm_map
*map2
= get_imsm_map(dev
, MAP_1
);
2576 failed
= imsm_count_failed(super
, dev
, MAP_1
);
2577 map_state
= imsm_check_degraded(super
, dev
, failed
, MAP_1
);
2578 if (map2
->map_state
!= map_state
) {
2579 map2
->map_state
= map_state
;
2580 super
->updates_pending
++;
2586 static struct imsm_disk
*get_imsm_missing(struct intel_super
*super
, __u8 index
)
2590 for (d
= super
->missing
; d
; d
= d
->next
)
2591 if (d
->index
== index
)
2596 static void getinfo_super_imsm(struct supertype
*st
, struct mdinfo
*info
, char *map
)
2598 struct intel_super
*super
= st
->sb
;
2599 struct imsm_disk
*disk
;
2600 int map_disks
= info
->array
.raid_disks
;
2601 int max_enough
= -1;
2603 struct imsm_super
*mpb
;
2605 if (super
->current_vol
>= 0) {
2606 getinfo_super_imsm_volume(st
, info
, map
);
2609 memset(info
, 0, sizeof(*info
));
2611 /* Set raid_disks to zero so that Assemble will always pull in valid
2614 info
->array
.raid_disks
= 0;
2615 info
->array
.level
= LEVEL_CONTAINER
;
2616 info
->array
.layout
= 0;
2617 info
->array
.md_minor
= -1;
2618 info
->array
.ctime
= 0; /* N/A for imsm */
2619 info
->array
.utime
= 0;
2620 info
->array
.chunk_size
= 0;
2622 info
->disk
.major
= 0;
2623 info
->disk
.minor
= 0;
2624 info
->disk
.raid_disk
= -1;
2625 info
->reshape_active
= 0;
2626 info
->array
.major_version
= -1;
2627 info
->array
.minor_version
= -2;
2628 strcpy(info
->text_version
, "imsm");
2629 info
->safe_mode_delay
= 0;
2630 info
->disk
.number
= -1;
2631 info
->disk
.state
= 0;
2633 info
->recovery_start
= MaxSector
;
2634 info
->recovery_blocked
= imsm_reshape_blocks_arrays_changes(st
->sb
);
2636 /* do we have the all the insync disks that we expect? */
2637 mpb
= super
->anchor
;
2639 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
2640 struct imsm_dev
*dev
= get_imsm_dev(super
, i
);
2641 int failed
, enough
, j
, missing
= 0;
2642 struct imsm_map
*map
;
2645 failed
= imsm_count_failed(super
, dev
, MAP_0
);
2646 state
= imsm_check_degraded(super
, dev
, failed
, MAP_0
);
2647 map
= get_imsm_map(dev
, MAP_0
);
2649 /* any newly missing disks?
2650 * (catches single-degraded vs double-degraded)
2652 for (j
= 0; j
< map
->num_members
; j
++) {
2653 __u32 ord
= get_imsm_ord_tbl_ent(dev
, j
, MAP_0
);
2654 __u32 idx
= ord_to_idx(ord
);
2656 if (!(ord
& IMSM_ORD_REBUILD
) &&
2657 get_imsm_missing(super
, idx
)) {
2663 if (state
== IMSM_T_STATE_FAILED
)
2665 else if (state
== IMSM_T_STATE_DEGRADED
&&
2666 (state
!= map
->map_state
|| missing
))
2668 else /* we're normal, or already degraded */
2671 /* in the missing/failed disk case check to see
2672 * if at least one array is runnable
2674 max_enough
= max(max_enough
, enough
);
2676 dprintf("%s: enough: %d\n", __func__
, max_enough
);
2677 info
->container_enough
= max_enough
;
2680 __u32 reserved
= imsm_reserved_sectors(super
, super
->disks
);
2682 disk
= &super
->disks
->disk
;
2683 info
->data_offset
= __le32_to_cpu(disk
->total_blocks
) - reserved
;
2684 info
->component_size
= reserved
;
2685 info
->disk
.state
= is_configured(disk
) ? (1 << MD_DISK_ACTIVE
) : 0;
2686 /* we don't change info->disk.raid_disk here because
2687 * this state will be finalized in mdmon after we have
2688 * found the 'most fresh' version of the metadata
2690 info
->disk
.state
|= is_failed(disk
) ? (1 << MD_DISK_FAULTY
) : 0;
2691 info
->disk
.state
|= is_spare(disk
) ? 0 : (1 << MD_DISK_SYNC
);
2694 /* only call uuid_from_super_imsm when this disk is part of a populated container,
2695 * ->compare_super may have updated the 'num_raid_devs' field for spares
2697 if (info
->disk
.state
& (1 << MD_DISK_SYNC
) || super
->anchor
->num_raid_devs
)
2698 uuid_from_super_imsm(st
, info
->uuid
);
2700 memcpy(info
->uuid
, uuid_zero
, sizeof(uuid_zero
));
2702 /* I don't know how to compute 'map' on imsm, so use safe default */
2705 for (i
= 0; i
< map_disks
; i
++)
2711 /* allocates memory and fills disk in mdinfo structure
2712 * for each disk in array */
2713 struct mdinfo
*getinfo_super_disks_imsm(struct supertype
*st
)
2715 struct mdinfo
*mddev
= NULL
;
2716 struct intel_super
*super
= st
->sb
;
2717 struct imsm_disk
*disk
;
2720 if (!super
|| !super
->disks
)
2723 mddev
= malloc(sizeof(*mddev
));
2725 fprintf(stderr
, Name
": Failed to allocate memory.\n");
2728 memset(mddev
, 0, sizeof(*mddev
));
2732 tmp
= malloc(sizeof(*tmp
));
2734 fprintf(stderr
, Name
": Failed to allocate memory.\n");
2739 memset(tmp
, 0, sizeof(*tmp
));
2741 tmp
->next
= mddev
->devs
;
2743 tmp
->disk
.number
= count
++;
2744 tmp
->disk
.major
= dl
->major
;
2745 tmp
->disk
.minor
= dl
->minor
;
2746 tmp
->disk
.state
= is_configured(disk
) ?
2747 (1 << MD_DISK_ACTIVE
) : 0;
2748 tmp
->disk
.state
|= is_failed(disk
) ? (1 << MD_DISK_FAULTY
) : 0;
2749 tmp
->disk
.state
|= is_spare(disk
) ? 0 : (1 << MD_DISK_SYNC
);
2750 tmp
->disk
.raid_disk
= -1;
2756 static int update_super_imsm(struct supertype
*st
, struct mdinfo
*info
,
2757 char *update
, char *devname
, int verbose
,
2758 int uuid_set
, char *homehost
)
2760 /* For 'assemble' and 'force' we need to return non-zero if any
2761 * change was made. For others, the return value is ignored.
2762 * Update options are:
2763 * force-one : This device looks a bit old but needs to be included,
2764 * update age info appropriately.
2765 * assemble: clear any 'faulty' flag to allow this device to
2767 * force-array: Array is degraded but being forced, mark it clean
2768 * if that will be needed to assemble it.
2770 * newdev: not used ????
2771 * grow: Array has gained a new device - this is currently for
2773 * resync: mark as dirty so a resync will happen.
2774 * name: update the name - preserving the homehost
2775 * uuid: Change the uuid of the array to match watch is given
2777 * Following are not relevant for this imsm:
2778 * sparc2.2 : update from old dodgey metadata
2779 * super-minor: change the preferred_minor number
2780 * summaries: update redundant counters.
2781 * homehost: update the recorded homehost
2782 * _reshape_progress: record new reshape_progress position.
2785 struct intel_super
*super
= st
->sb
;
2786 struct imsm_super
*mpb
;
2788 /* we can only update container info */
2789 if (!super
|| super
->current_vol
>= 0 || !super
->anchor
)
2792 mpb
= super
->anchor
;
2794 if (strcmp(update
, "uuid") == 0 && uuid_set
&& !info
->update_private
)
2796 else if (strcmp(update
, "uuid") == 0 && uuid_set
&& info
->update_private
) {
2797 mpb
->orig_family_num
= *((__u32
*) info
->update_private
);
2799 } else if (strcmp(update
, "uuid") == 0) {
2800 __u32
*new_family
= malloc(sizeof(*new_family
));
2802 /* update orig_family_number with the incoming random
2803 * data, report the new effective uuid, and store the
2804 * new orig_family_num for future updates.
2807 memcpy(&mpb
->orig_family_num
, info
->uuid
, sizeof(__u32
));
2808 uuid_from_super_imsm(st
, info
->uuid
);
2809 *new_family
= mpb
->orig_family_num
;
2810 info
->update_private
= new_family
;
2813 } else if (strcmp(update
, "assemble") == 0)
2818 /* successful update? recompute checksum */
2820 mpb
->check_sum
= __le32_to_cpu(__gen_imsm_checksum(mpb
));
2825 static size_t disks_to_mpb_size(int disks
)
2829 size
= sizeof(struct imsm_super
);
2830 size
+= (disks
- 1) * sizeof(struct imsm_disk
);
2831 size
+= 2 * sizeof(struct imsm_dev
);
2832 /* up to 2 maps per raid device (-2 for imsm_maps in imsm_dev */
2833 size
+= (4 - 2) * sizeof(struct imsm_map
);
2834 /* 4 possible disk_ord_tbl's */
2835 size
+= 4 * (disks
- 1) * sizeof(__u32
);
2840 static __u64
avail_size_imsm(struct supertype
*st
, __u64 devsize
)
2842 if (devsize
< (MPB_SECTOR_CNT
+ IMSM_RESERVED_SECTORS
))
2845 return devsize
- (MPB_SECTOR_CNT
+ IMSM_RESERVED_SECTORS
);
2848 static void free_devlist(struct intel_super
*super
)
2850 struct intel_dev
*dv
;
2852 while (super
->devlist
) {
2853 dv
= super
->devlist
->next
;
2854 free(super
->devlist
->dev
);
2855 free(super
->devlist
);
2856 super
->devlist
= dv
;
2860 static void imsm_copy_dev(struct imsm_dev
*dest
, struct imsm_dev
*src
)
2862 memcpy(dest
, src
, sizeof_imsm_dev(src
, 0));
2865 static int compare_super_imsm(struct supertype
*st
, struct supertype
*tst
)
2869 * 0 same, or first was empty, and second was copied
2870 * 1 second had wrong number
2872 * 3 wrong other info
2874 struct intel_super
*first
= st
->sb
;
2875 struct intel_super
*sec
= tst
->sb
;
2882 /* in platform dependent environment test if the disks
2883 * use the same Intel hba
2885 if (!check_env("IMSM_NO_PLATFORM")) {
2886 if (!first
->hba
|| !sec
->hba
||
2887 (first
->hba
->type
!= sec
->hba
->type
)) {
2889 "HBAs of devices does not match %s != %s\n",
2890 first
->hba
? get_sys_dev_type(first
->hba
->type
) : NULL
,
2891 sec
->hba
? get_sys_dev_type(sec
->hba
->type
) : NULL
);
2896 /* if an anchor does not have num_raid_devs set then it is a free
2899 if (first
->anchor
->num_raid_devs
> 0 &&
2900 sec
->anchor
->num_raid_devs
> 0) {
2901 /* Determine if these disks might ever have been
2902 * related. Further disambiguation can only take place
2903 * in load_super_imsm_all
2905 __u32 first_family
= first
->anchor
->orig_family_num
;
2906 __u32 sec_family
= sec
->anchor
->orig_family_num
;
2908 if (memcmp(first
->anchor
->sig
, sec
->anchor
->sig
,
2909 MAX_SIGNATURE_LENGTH
) != 0)
2912 if (first_family
== 0)
2913 first_family
= first
->anchor
->family_num
;
2914 if (sec_family
== 0)
2915 sec_family
= sec
->anchor
->family_num
;
2917 if (first_family
!= sec_family
)
2923 /* if 'first' is a spare promote it to a populated mpb with sec's
2926 if (first
->anchor
->num_raid_devs
== 0 &&
2927 sec
->anchor
->num_raid_devs
> 0) {
2929 struct intel_dev
*dv
;
2930 struct imsm_dev
*dev
;
2932 /* we need to copy raid device info from sec if an allocation
2933 * fails here we don't associate the spare
2935 for (i
= 0; i
< sec
->anchor
->num_raid_devs
; i
++) {
2936 dv
= malloc(sizeof(*dv
));
2939 dev
= malloc(sizeof_imsm_dev(get_imsm_dev(sec
, i
), 1));
2946 dv
->next
= first
->devlist
;
2947 first
->devlist
= dv
;
2949 if (i
< sec
->anchor
->num_raid_devs
) {
2950 /* allocation failure */
2951 free_devlist(first
);
2952 fprintf(stderr
, "imsm: failed to associate spare\n");
2955 first
->anchor
->num_raid_devs
= sec
->anchor
->num_raid_devs
;
2956 first
->anchor
->orig_family_num
= sec
->anchor
->orig_family_num
;
2957 first
->anchor
->family_num
= sec
->anchor
->family_num
;
2958 memcpy(first
->anchor
->sig
, sec
->anchor
->sig
, MAX_SIGNATURE_LENGTH
);
2959 for (i
= 0; i
< sec
->anchor
->num_raid_devs
; i
++)
2960 imsm_copy_dev(get_imsm_dev(first
, i
), get_imsm_dev(sec
, i
));
2966 static void fd2devname(int fd
, char *name
)
2970 char dname
[PATH_MAX
];
2975 if (fstat(fd
, &st
) != 0)
2977 sprintf(path
, "/sys/dev/block/%d:%d",
2978 major(st
.st_rdev
), minor(st
.st_rdev
));
2980 rv
= readlink(path
, dname
, sizeof(dname
)-1);
2985 nm
= strrchr(dname
, '/');
2988 snprintf(name
, MAX_RAID_SERIAL_LEN
, "/dev/%s", nm
);
2992 extern int scsi_get_serial(int fd
, void *buf
, size_t buf_len
);
2994 static int imsm_read_serial(int fd
, char *devname
,
2995 __u8 serial
[MAX_RAID_SERIAL_LEN
])
2997 unsigned char scsi_serial
[255];
3006 memset(scsi_serial
, 0, sizeof(scsi_serial
));
3008 rv
= scsi_get_serial(fd
, scsi_serial
, sizeof(scsi_serial
));
3010 if (rv
&& check_env("IMSM_DEVNAME_AS_SERIAL")) {
3011 memset(serial
, 0, MAX_RAID_SERIAL_LEN
);
3012 fd2devname(fd
, (char *) serial
);
3019 Name
": Failed to retrieve serial for %s\n",
3024 rsp_len
= scsi_serial
[3];
3028 Name
": Failed to retrieve serial for %s\n",
3032 rsp_buf
= (char *) &scsi_serial
[4];
3034 /* trim all whitespace and non-printable characters and convert
3037 for (i
= 0, dest
= rsp_buf
; i
< rsp_len
; i
++) {
3040 /* ':' is reserved for use in placeholder serial
3041 * numbers for missing disks
3049 len
= dest
- rsp_buf
;
3052 /* truncate leading characters */
3053 if (len
> MAX_RAID_SERIAL_LEN
) {
3054 dest
+= len
- MAX_RAID_SERIAL_LEN
;
3055 len
= MAX_RAID_SERIAL_LEN
;
3058 memset(serial
, 0, MAX_RAID_SERIAL_LEN
);
3059 memcpy(serial
, dest
, len
);
3064 static int serialcmp(__u8
*s1
, __u8
*s2
)
3066 return strncmp((char *) s1
, (char *) s2
, MAX_RAID_SERIAL_LEN
);
3069 static void serialcpy(__u8
*dest
, __u8
*src
)
3071 strncpy((char *) dest
, (char *) src
, MAX_RAID_SERIAL_LEN
);
3074 static struct dl
*serial_to_dl(__u8
*serial
, struct intel_super
*super
)
3078 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
3079 if (serialcmp(dl
->serial
, serial
) == 0)
3085 static struct imsm_disk
*
3086 __serial_to_disk(__u8
*serial
, struct imsm_super
*mpb
, int *idx
)
3090 for (i
= 0; i
< mpb
->num_disks
; i
++) {
3091 struct imsm_disk
*disk
= __get_imsm_disk(mpb
, i
);
3093 if (serialcmp(disk
->serial
, serial
) == 0) {
3104 load_imsm_disk(int fd
, struct intel_super
*super
, char *devname
, int keep_fd
)
3106 struct imsm_disk
*disk
;
3111 __u8 serial
[MAX_RAID_SERIAL_LEN
];
3113 rv
= imsm_read_serial(fd
, devname
, serial
);
3118 dl
= calloc(1, sizeof(*dl
));
3122 Name
": failed to allocate disk buffer for %s\n",
3128 dl
->major
= major(stb
.st_rdev
);
3129 dl
->minor
= minor(stb
.st_rdev
);
3130 dl
->next
= super
->disks
;
3131 dl
->fd
= keep_fd
? fd
: -1;
3132 assert(super
->disks
== NULL
);
3134 serialcpy(dl
->serial
, serial
);
3137 fd2devname(fd
, name
);
3139 dl
->devname
= strdup(devname
);
3141 dl
->devname
= strdup(name
);
3143 /* look up this disk's index in the current anchor */
3144 disk
= __serial_to_disk(dl
->serial
, super
->anchor
, &dl
->index
);
3147 /* only set index on disks that are a member of a
3148 * populated contianer, i.e. one with raid_devs
3150 if (is_failed(&dl
->disk
))
3152 else if (is_spare(&dl
->disk
))
3160 /* When migrating map0 contains the 'destination' state while map1
3161 * contains the current state. When not migrating map0 contains the
3162 * current state. This routine assumes that map[0].map_state is set to
3163 * the current array state before being called.
3165 * Migration is indicated by one of the following states
3166 * 1/ Idle (migr_state=0 map0state=normal||unitialized||degraded||failed)
3167 * 2/ Initialize (migr_state=1 migr_type=MIGR_INIT map0state=normal
3168 * map1state=unitialized)
3169 * 3/ Repair (Resync) (migr_state=1 migr_type=MIGR_REPAIR map0state=normal
3171 * 4/ Rebuild (migr_state=1 migr_type=MIGR_REBUILD map0state=normal
3172 * map1state=degraded)
3173 * 5/ Migration (mig_state=1 migr_type=MIGR_GEN_MIGR map0state=normal
3176 static void migrate(struct imsm_dev
*dev
, struct intel_super
*super
,
3177 __u8 to_state
, int migr_type
)
3179 struct imsm_map
*dest
;
3180 struct imsm_map
*src
= get_imsm_map(dev
, MAP_0
);
3182 dev
->vol
.migr_state
= 1;
3183 set_migr_type(dev
, migr_type
);
3184 dev
->vol
.curr_migr_unit
= 0;
3185 dest
= get_imsm_map(dev
, MAP_1
);
3187 /* duplicate and then set the target end state in map[0] */
3188 memcpy(dest
, src
, sizeof_imsm_map(src
));
3189 if ((migr_type
== MIGR_REBUILD
) ||
3190 (migr_type
== MIGR_GEN_MIGR
)) {
3194 for (i
= 0; i
< src
->num_members
; i
++) {
3195 ord
= __le32_to_cpu(src
->disk_ord_tbl
[i
]);
3196 set_imsm_ord_tbl_ent(src
, i
, ord_to_idx(ord
));
3200 if (migr_type
== MIGR_GEN_MIGR
)
3201 /* Clear migration record */
3202 memset(super
->migr_rec
, 0, sizeof(struct migr_record
));
3204 src
->map_state
= to_state
;
3207 static void end_migration(struct imsm_dev
*dev
, struct intel_super
*super
,
3210 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
3211 struct imsm_map
*prev
= get_imsm_map(dev
, dev
->vol
.migr_state
== 0 ?
3215 /* merge any IMSM_ORD_REBUILD bits that were not successfully
3216 * completed in the last migration.
3218 * FIXME add support for raid-level-migration
3220 if ((map_state
!= map
->map_state
) && (is_gen_migration(dev
) == 0) &&
3221 (prev
->map_state
!= IMSM_T_STATE_UNINITIALIZED
)) {
3222 /* when final map state is other than expected
3223 * merge maps (not for migration)
3227 for (i
= 0; i
< prev
->num_members
; i
++)
3228 for (j
= 0; j
< map
->num_members
; j
++)
3229 /* during online capacity expansion
3230 * disks position can be changed
3231 * if takeover is used
3233 if (ord_to_idx(map
->disk_ord_tbl
[j
]) ==
3234 ord_to_idx(prev
->disk_ord_tbl
[i
])) {
3235 map
->disk_ord_tbl
[j
] |=
3236 prev
->disk_ord_tbl
[i
];
3239 failed
= imsm_count_failed(super
, dev
, MAP_0
);
3240 map_state
= imsm_check_degraded(super
, dev
, failed
, MAP_0
);
3243 dev
->vol
.migr_state
= 0;
3244 set_migr_type(dev
, 0);
3245 dev
->vol
.curr_migr_unit
= 0;
3246 map
->map_state
= map_state
;
3250 static int parse_raid_devices(struct intel_super
*super
)
3253 struct imsm_dev
*dev_new
;
3254 size_t len
, len_migr
;
3256 size_t space_needed
= 0;
3257 struct imsm_super
*mpb
= super
->anchor
;
3259 for (i
= 0; i
< super
->anchor
->num_raid_devs
; i
++) {
3260 struct imsm_dev
*dev_iter
= __get_imsm_dev(super
->anchor
, i
);
3261 struct intel_dev
*dv
;
3263 len
= sizeof_imsm_dev(dev_iter
, 0);
3264 len_migr
= sizeof_imsm_dev(dev_iter
, 1);
3266 space_needed
+= len_migr
- len
;
3268 dv
= malloc(sizeof(*dv
));
3271 if (max_len
< len_migr
)
3273 if (max_len
> len_migr
)
3274 space_needed
+= max_len
- len_migr
;
3275 dev_new
= malloc(max_len
);
3280 imsm_copy_dev(dev_new
, dev_iter
);
3283 dv
->next
= super
->devlist
;
3284 super
->devlist
= dv
;
3287 /* ensure that super->buf is large enough when all raid devices
3290 if (__le32_to_cpu(mpb
->mpb_size
) + space_needed
> super
->len
) {
3293 len
= ROUND_UP(__le32_to_cpu(mpb
->mpb_size
) + space_needed
, 512);
3294 if (posix_memalign(&buf
, 512, len
) != 0)
3297 memcpy(buf
, super
->buf
, super
->len
);
3298 memset(buf
+ super
->len
, 0, len
- super
->len
);
3307 /* retrieve a pointer to the bbm log which starts after all raid devices */
3308 struct bbm_log
*__get_imsm_bbm_log(struct imsm_super
*mpb
)
3312 if (__le32_to_cpu(mpb
->bbm_log_size
)) {
3314 ptr
+= mpb
->mpb_size
- __le32_to_cpu(mpb
->bbm_log_size
);
3320 /*******************************************************************************
3321 * Function: check_mpb_migr_compatibility
3322 * Description: Function checks for unsupported migration features:
3323 * - migration optimization area (pba_of_lba0)
3324 * - descending reshape (ascending_migr)
3326 * super : imsm metadata information
3328 * 0 : migration is compatible
3329 * -1 : migration is not compatible
3330 ******************************************************************************/
3331 int check_mpb_migr_compatibility(struct intel_super
*super
)
3333 struct imsm_map
*map0
, *map1
;
3334 struct migr_record
*migr_rec
= super
->migr_rec
;
3337 for (i
= 0; i
< super
->anchor
->num_raid_devs
; i
++) {
3338 struct imsm_dev
*dev_iter
= __get_imsm_dev(super
->anchor
, i
);
3341 dev_iter
->vol
.migr_state
== 1 &&
3342 dev_iter
->vol
.migr_type
== MIGR_GEN_MIGR
) {
3343 /* This device is migrating */
3344 map0
= get_imsm_map(dev_iter
, MAP_0
);
3345 map1
= get_imsm_map(dev_iter
, MAP_1
);
3346 if (map0
->pba_of_lba0
!= map1
->pba_of_lba0
)
3347 /* migration optimization area was used */
3349 if (migr_rec
->ascending_migr
== 0
3350 && migr_rec
->dest_depth_per_unit
> 0)
3351 /* descending reshape not supported yet */
3358 static void __free_imsm(struct intel_super
*super
, int free_disks
);
3360 /* load_imsm_mpb - read matrix metadata
3361 * allocates super->mpb to be freed by free_imsm
3363 static int load_imsm_mpb(int fd
, struct intel_super
*super
, char *devname
)
3365 unsigned long long dsize
;
3366 unsigned long long sectors
;
3368 struct imsm_super
*anchor
;
3371 get_dev_size(fd
, NULL
, &dsize
);
3375 Name
": %s: device to small for imsm\n",
3380 if (lseek64(fd
, dsize
- (512 * 2), SEEK_SET
) < 0) {
3382 fprintf(stderr
, Name
3383 ": Cannot seek to anchor block on %s: %s\n",
3384 devname
, strerror(errno
));
3388 if (posix_memalign((void**)&anchor
, 512, 512) != 0) {
3391 Name
": Failed to allocate imsm anchor buffer"
3392 " on %s\n", devname
);
3395 if (read(fd
, anchor
, 512) != 512) {
3398 Name
": Cannot read anchor block on %s: %s\n",
3399 devname
, strerror(errno
));
3404 if (strncmp((char *) anchor
->sig
, MPB_SIGNATURE
, MPB_SIG_LEN
) != 0) {
3407 Name
": no IMSM anchor on %s\n", devname
);
3412 __free_imsm(super
, 0);
3413 /* reload capability and hba */
3415 /* capability and hba must be updated with new super allocation */
3416 find_intel_hba_capability(fd
, super
, devname
);
3417 super
->len
= ROUND_UP(anchor
->mpb_size
, 512);
3418 if (posix_memalign(&super
->buf
, 512, super
->len
) != 0) {
3421 Name
": unable to allocate %zu byte mpb buffer\n",
3426 memcpy(super
->buf
, anchor
, 512);
3428 sectors
= mpb_sectors(anchor
) - 1;
3431 if (posix_memalign(&super
->migr_rec_buf
, 512, MIGR_REC_BUF_SIZE
) != 0) {
3432 fprintf(stderr
, Name
3433 ": %s could not allocate migr_rec buffer\n", __func__
);
3439 check_sum
= __gen_imsm_checksum(super
->anchor
);
3440 if (check_sum
!= __le32_to_cpu(super
->anchor
->check_sum
)) {
3443 Name
": IMSM checksum %x != %x on %s\n",
3445 __le32_to_cpu(super
->anchor
->check_sum
),
3453 /* read the extended mpb */
3454 if (lseek64(fd
, dsize
- (512 * (2 + sectors
)), SEEK_SET
) < 0) {
3457 Name
": Cannot seek to extended mpb on %s: %s\n",
3458 devname
, strerror(errno
));
3462 if ((unsigned)read(fd
, super
->buf
+ 512, super
->len
- 512) != super
->len
- 512) {
3465 Name
": Cannot read extended mpb on %s: %s\n",
3466 devname
, strerror(errno
));
3470 check_sum
= __gen_imsm_checksum(super
->anchor
);
3471 if (check_sum
!= __le32_to_cpu(super
->anchor
->check_sum
)) {
3474 Name
": IMSM checksum %x != %x on %s\n",
3475 check_sum
, __le32_to_cpu(super
->anchor
->check_sum
),
3480 /* FIXME the BBM log is disk specific so we cannot use this global
3481 * buffer for all disks. Ok for now since we only look at the global
3482 * bbm_log_size parameter to gate assembly
3484 super
->bbm_log
= __get_imsm_bbm_log(super
->anchor
);
3489 static int read_imsm_migr_rec(int fd
, struct intel_super
*super
);
3492 load_and_parse_mpb(int fd
, struct intel_super
*super
, char *devname
, int keep_fd
)
3496 err
= load_imsm_mpb(fd
, super
, devname
);
3499 err
= load_imsm_disk(fd
, super
, devname
, keep_fd
);
3502 err
= parse_raid_devices(super
);
3507 static void __free_imsm_disk(struct dl
*d
)
3519 static void free_imsm_disks(struct intel_super
*super
)
3523 while (super
->disks
) {
3525 super
->disks
= d
->next
;
3526 __free_imsm_disk(d
);
3528 while (super
->disk_mgmt_list
) {
3529 d
= super
->disk_mgmt_list
;
3530 super
->disk_mgmt_list
= d
->next
;
3531 __free_imsm_disk(d
);
3533 while (super
->missing
) {
3535 super
->missing
= d
->next
;
3536 __free_imsm_disk(d
);
3541 /* free all the pieces hanging off of a super pointer */
3542 static void __free_imsm(struct intel_super
*super
, int free_disks
)
3544 struct intel_hba
*elem
, *next
;
3550 /* unlink capability description */
3552 if (super
->migr_rec_buf
) {
3553 free(super
->migr_rec_buf
);
3554 super
->migr_rec_buf
= NULL
;
3557 free_imsm_disks(super
);
3558 free_devlist(super
);
3562 free((void *)elem
->path
);
3570 static void free_imsm(struct intel_super
*super
)
3572 __free_imsm(super
, 1);
3576 static void free_super_imsm(struct supertype
*st
)
3578 struct intel_super
*super
= st
->sb
;
3587 static struct intel_super
*alloc_super(void)
3589 struct intel_super
*super
= malloc(sizeof(*super
));
3592 memset(super
, 0, sizeof(*super
));
3593 super
->current_vol
= -1;
3594 super
->create_offset
= ~((__u32
) 0);
3600 * find and allocate hba and OROM/EFI based on valid fd of RAID component device
3602 static int find_intel_hba_capability(int fd
, struct intel_super
*super
, char *devname
)
3604 struct sys_dev
*hba_name
;
3607 if ((fd
< 0) || check_env("IMSM_NO_PLATFORM")) {
3612 hba_name
= find_disk_attached_hba(fd
, NULL
);
3616 Name
": %s is not attached to Intel(R) RAID controller.\n",
3620 rv
= attach_hba_to_super(super
, hba_name
);
3623 struct intel_hba
*hba
= super
->hba
;
3625 fprintf(stderr
, Name
": %s is attached to Intel(R) %s RAID "
3626 "controller (%s),\n"
3627 " but the container is assigned to Intel(R) "
3628 "%s RAID controller (",
3631 hba_name
->pci_id
? : "Err!",
3632 get_sys_dev_type(hba_name
->type
));
3635 fprintf(stderr
, "%s", hba
->pci_id
? : "Err!");
3637 fprintf(stderr
, ", ");
3641 fprintf(stderr
, ").\n"
3642 " Mixing devices attached to different controllers "
3643 "is not allowed.\n");
3645 free_sys_dev(&hba_name
);
3648 super
->orom
= find_imsm_capability(hba_name
->type
);
3649 free_sys_dev(&hba_name
);
3655 /* find_missing - helper routine for load_super_imsm_all that identifies
3656 * disks that have disappeared from the system. This routine relies on
3657 * the mpb being uptodate, which it is at load time.
3659 static int find_missing(struct intel_super
*super
)
3662 struct imsm_super
*mpb
= super
->anchor
;
3664 struct imsm_disk
*disk
;
3666 for (i
= 0; i
< mpb
->num_disks
; i
++) {
3667 disk
= __get_imsm_disk(mpb
, i
);
3668 dl
= serial_to_dl(disk
->serial
, super
);
3672 dl
= malloc(sizeof(*dl
));
3678 dl
->devname
= strdup("missing");
3680 serialcpy(dl
->serial
, disk
->serial
);
3683 dl
->next
= super
->missing
;
3684 super
->missing
= dl
;
3691 static struct intel_disk
*disk_list_get(__u8
*serial
, struct intel_disk
*disk_list
)
3693 struct intel_disk
*idisk
= disk_list
;
3696 if (serialcmp(idisk
->disk
.serial
, serial
) == 0)
3698 idisk
= idisk
->next
;
3704 static int __prep_thunderdome(struct intel_super
**table
, int tbl_size
,
3705 struct intel_super
*super
,
3706 struct intel_disk
**disk_list
)
3708 struct imsm_disk
*d
= &super
->disks
->disk
;
3709 struct imsm_super
*mpb
= super
->anchor
;
3712 for (i
= 0; i
< tbl_size
; i
++) {
3713 struct imsm_super
*tbl_mpb
= table
[i
]->anchor
;
3714 struct imsm_disk
*tbl_d
= &table
[i
]->disks
->disk
;
3716 if (tbl_mpb
->family_num
== mpb
->family_num
) {
3717 if (tbl_mpb
->check_sum
== mpb
->check_sum
) {
3718 dprintf("%s: mpb from %d:%d matches %d:%d\n",
3719 __func__
, super
->disks
->major
,
3720 super
->disks
->minor
,
3721 table
[i
]->disks
->major
,
3722 table
[i
]->disks
->minor
);
3726 if (((is_configured(d
) && !is_configured(tbl_d
)) ||
3727 is_configured(d
) == is_configured(tbl_d
)) &&
3728 tbl_mpb
->generation_num
< mpb
->generation_num
) {
3729 /* current version of the mpb is a
3730 * better candidate than the one in
3731 * super_table, but copy over "cross
3732 * generational" status
3734 struct intel_disk
*idisk
;
3736 dprintf("%s: mpb from %d:%d replaces %d:%d\n",
3737 __func__
, super
->disks
->major
,
3738 super
->disks
->minor
,
3739 table
[i
]->disks
->major
,
3740 table
[i
]->disks
->minor
);
3742 idisk
= disk_list_get(tbl_d
->serial
, *disk_list
);
3743 if (idisk
&& is_failed(&idisk
->disk
))
3744 tbl_d
->status
|= FAILED_DISK
;
3747 struct intel_disk
*idisk
;
3748 struct imsm_disk
*disk
;
3750 /* tbl_mpb is more up to date, but copy
3751 * over cross generational status before
3754 disk
= __serial_to_disk(d
->serial
, mpb
, NULL
);
3755 if (disk
&& is_failed(disk
))
3756 d
->status
|= FAILED_DISK
;
3758 idisk
= disk_list_get(d
->serial
, *disk_list
);
3761 if (disk
&& is_configured(disk
))
3762 idisk
->disk
.status
|= CONFIGURED_DISK
;
3765 dprintf("%s: mpb from %d:%d prefer %d:%d\n",
3766 __func__
, super
->disks
->major
,
3767 super
->disks
->minor
,
3768 table
[i
]->disks
->major
,
3769 table
[i
]->disks
->minor
);
3777 table
[tbl_size
++] = super
;
3781 /* update/extend the merged list of imsm_disk records */
3782 for (j
= 0; j
< mpb
->num_disks
; j
++) {
3783 struct imsm_disk
*disk
= __get_imsm_disk(mpb
, j
);
3784 struct intel_disk
*idisk
;
3786 idisk
= disk_list_get(disk
->serial
, *disk_list
);
3788 idisk
->disk
.status
|= disk
->status
;
3789 if (is_configured(&idisk
->disk
) ||
3790 is_failed(&idisk
->disk
))
3791 idisk
->disk
.status
&= ~(SPARE_DISK
);
3793 idisk
= calloc(1, sizeof(*idisk
));
3796 idisk
->owner
= IMSM_UNKNOWN_OWNER
;
3797 idisk
->disk
= *disk
;
3798 idisk
->next
= *disk_list
;
3802 if (serialcmp(idisk
->disk
.serial
, d
->serial
) == 0)
3809 static struct intel_super
*
3810 validate_members(struct intel_super
*super
, struct intel_disk
*disk_list
,
3813 struct imsm_super
*mpb
= super
->anchor
;
3817 for (i
= 0; i
< mpb
->num_disks
; i
++) {
3818 struct imsm_disk
*disk
= __get_imsm_disk(mpb
, i
);
3819 struct intel_disk
*idisk
;
3821 idisk
= disk_list_get(disk
->serial
, disk_list
);
3823 if (idisk
->owner
== owner
||
3824 idisk
->owner
== IMSM_UNKNOWN_OWNER
)
3827 dprintf("%s: '%.16s' owner %d != %d\n",
3828 __func__
, disk
->serial
, idisk
->owner
,
3831 dprintf("%s: unknown disk %x [%d]: %.16s\n",
3832 __func__
, __le32_to_cpu(mpb
->family_num
), i
,
3838 if (ok_count
== mpb
->num_disks
)
3843 static void show_conflicts(__u32 family_num
, struct intel_super
*super_list
)
3845 struct intel_super
*s
;
3847 for (s
= super_list
; s
; s
= s
->next
) {
3848 if (family_num
!= s
->anchor
->family_num
)
3850 fprintf(stderr
, "Conflict, offlining family %#x on '%s'\n",
3851 __le32_to_cpu(family_num
), s
->disks
->devname
);
3855 static struct intel_super
*
3856 imsm_thunderdome(struct intel_super
**super_list
, int len
)
3858 struct intel_super
*super_table
[len
];
3859 struct intel_disk
*disk_list
= NULL
;
3860 struct intel_super
*champion
, *spare
;
3861 struct intel_super
*s
, **del
;
3866 memset(super_table
, 0, sizeof(super_table
));
3867 for (s
= *super_list
; s
; s
= s
->next
)
3868 tbl_size
= __prep_thunderdome(super_table
, tbl_size
, s
, &disk_list
);
3870 for (i
= 0; i
< tbl_size
; i
++) {
3871 struct imsm_disk
*d
;
3872 struct intel_disk
*idisk
;
3873 struct imsm_super
*mpb
= super_table
[i
]->anchor
;
3876 d
= &s
->disks
->disk
;
3878 /* 'd' must appear in merged disk list for its
3879 * configuration to be valid
3881 idisk
= disk_list_get(d
->serial
, disk_list
);
3882 if (idisk
&& idisk
->owner
== i
)
3883 s
= validate_members(s
, disk_list
, i
);
3888 dprintf("%s: marking family: %#x from %d:%d offline\n",
3889 __func__
, mpb
->family_num
,
3890 super_table
[i
]->disks
->major
,
3891 super_table
[i
]->disks
->minor
);
3895 /* This is where the mdadm implementation differs from the Windows
3896 * driver which has no strict concept of a container. We can only
3897 * assemble one family from a container, so when returning a prodigal
3898 * array member to this system the code will not be able to disambiguate
3899 * the container contents that should be assembled ("foreign" versus
3900 * "local"). It requires user intervention to set the orig_family_num
3901 * to a new value to establish a new container. The Windows driver in
3902 * this situation fixes up the volume name in place and manages the
3903 * foreign array as an independent entity.
3908 for (i
= 0; i
< tbl_size
; i
++) {
3909 struct intel_super
*tbl_ent
= super_table
[i
];
3915 if (tbl_ent
->anchor
->num_raid_devs
== 0) {
3920 if (s
&& !is_spare
) {
3921 show_conflicts(tbl_ent
->anchor
->family_num
, *super_list
);
3923 } else if (!s
&& !is_spare
)
3936 fprintf(stderr
, "Chose family %#x on '%s', "
3937 "assemble conflicts to new container with '--update=uuid'\n",
3938 __le32_to_cpu(s
->anchor
->family_num
), s
->disks
->devname
);
3940 /* collect all dl's onto 'champion', and update them to
3941 * champion's version of the status
3943 for (s
= *super_list
; s
; s
= s
->next
) {
3944 struct imsm_super
*mpb
= champion
->anchor
;
3945 struct dl
*dl
= s
->disks
;
3950 for (i
= 0; i
< mpb
->num_disks
; i
++) {
3951 struct imsm_disk
*disk
;
3953 disk
= __serial_to_disk(dl
->serial
, mpb
, &dl
->index
);
3956 /* only set index on disks that are a member of
3957 * a populated contianer, i.e. one with
3960 if (is_failed(&dl
->disk
))
3962 else if (is_spare(&dl
->disk
))
3968 if (i
>= mpb
->num_disks
) {
3969 struct intel_disk
*idisk
;
3971 idisk
= disk_list_get(dl
->serial
, disk_list
);
3972 if (idisk
&& is_spare(&idisk
->disk
) &&
3973 !is_failed(&idisk
->disk
) && !is_configured(&idisk
->disk
))
3981 dl
->next
= champion
->disks
;
3982 champion
->disks
= dl
;
3986 /* delete 'champion' from super_list */
3987 for (del
= super_list
; *del
; ) {
3988 if (*del
== champion
) {
3989 *del
= (*del
)->next
;
3992 del
= &(*del
)->next
;
3994 champion
->next
= NULL
;
3998 struct intel_disk
*idisk
= disk_list
;
4000 disk_list
= disk_list
->next
;
4007 static int load_super_imsm_all(struct supertype
*st
, int fd
, void **sbp
,
4011 struct intel_super
*super_list
= NULL
;
4012 struct intel_super
*super
= NULL
;
4013 int devnum
= fd2devnum(fd
);
4019 /* check if 'fd' an opened container */
4020 sra
= sysfs_read(fd
, 0, GET_LEVEL
|GET_VERSION
|GET_DEVS
|GET_STATE
);
4024 if (sra
->array
.major_version
!= -1 ||
4025 sra
->array
.minor_version
!= -2 ||
4026 strcmp(sra
->text_version
, "imsm") != 0) {
4031 for (sd
= sra
->devs
, i
= 0; sd
; sd
= sd
->next
, i
++) {
4032 struct intel_super
*s
= alloc_super();
4040 s
->next
= super_list
;
4044 sprintf(nm
, "%d:%d", sd
->disk
.major
, sd
->disk
.minor
);
4045 dfd
= dev_open(nm
, O_RDWR
);
4049 rv
= find_intel_hba_capability(dfd
, s
, devname
);
4050 /* no orom/efi or non-intel hba of the disk */
4054 err
= load_and_parse_mpb(dfd
, s
, NULL
, 1);
4056 /* retry the load if we might have raced against mdmon */
4057 if (err
== 3 && mdmon_running(devnum
))
4058 for (retry
= 0; retry
< 3; retry
++) {
4060 err
= load_and_parse_mpb(dfd
, s
, NULL
, 1);
4068 /* all mpbs enter, maybe one leaves */
4069 super
= imsm_thunderdome(&super_list
, i
);
4075 if (find_missing(super
) != 0) {
4081 /* load migration record */
4082 err
= load_imsm_migr_rec(super
, NULL
);
4084 /* migration is in progress,
4085 * but migr_rec cannot be loaded,
4091 /* Check migration compatibility */
4092 if ((err
== 0) && (check_mpb_migr_compatibility(super
) != 0)) {
4093 fprintf(stderr
, Name
": Unsupported migration detected");
4095 fprintf(stderr
, " on %s\n", devname
);
4097 fprintf(stderr
, " (IMSM).\n");
4106 while (super_list
) {
4107 struct intel_super
*s
= super_list
;
4109 super_list
= super_list
->next
;
4118 st
->container_dev
= devnum
;
4119 if (err
== 0 && st
->ss
== NULL
) {
4120 st
->ss
= &super_imsm
;
4121 st
->minor_version
= 0;
4122 st
->max_devs
= IMSM_MAX_DEVICES
;
4127 static int load_container_imsm(struct supertype
*st
, int fd
, char *devname
)
4129 return load_super_imsm_all(st
, fd
, &st
->sb
, devname
);
4133 static int load_super_imsm(struct supertype
*st
, int fd
, char *devname
)
4135 struct intel_super
*super
;
4138 if (test_partition(fd
))
4139 /* IMSM not allowed on partitions */
4142 free_super_imsm(st
);
4144 super
= alloc_super();
4147 Name
": malloc of %zu failed.\n",
4151 /* Load hba and capabilities if they exist.
4152 * But do not preclude loading metadata in case capabilities or hba are
4153 * non-compliant and ignore_hw_compat is set.
4155 rv
= find_intel_hba_capability(fd
, super
, devname
);
4156 /* no orom/efi or non-intel hba of the disk */
4157 if ((rv
!= 0) && (st
->ignore_hw_compat
== 0)) {
4160 Name
": No OROM/EFI properties for %s\n", devname
);
4164 rv
= load_and_parse_mpb(fd
, super
, devname
, 0);
4169 Name
": Failed to load all information "
4170 "sections on %s\n", devname
);
4176 if (st
->ss
== NULL
) {
4177 st
->ss
= &super_imsm
;
4178 st
->minor_version
= 0;
4179 st
->max_devs
= IMSM_MAX_DEVICES
;
4182 /* load migration record */
4183 if (load_imsm_migr_rec(super
, NULL
) == 0) {
4184 /* Check for unsupported migration features */
4185 if (check_mpb_migr_compatibility(super
) != 0) {
4187 Name
": Unsupported migration detected");
4189 fprintf(stderr
, " on %s\n", devname
);
4191 fprintf(stderr
, " (IMSM).\n");
4199 static __u16
info_to_blocks_per_strip(mdu_array_info_t
*info
)
4201 if (info
->level
== 1)
4203 return info
->chunk_size
>> 9;
4206 static __u32
info_to_num_data_stripes(mdu_array_info_t
*info
, int num_domains
)
4210 num_stripes
= (info
->size
* 2) / info_to_blocks_per_strip(info
);
4211 num_stripes
/= num_domains
;
4216 static __u32
info_to_blocks_per_member(mdu_array_info_t
*info
)
4218 if (info
->level
== 1)
4219 return info
->size
* 2;
4221 return (info
->size
* 2) & ~(info_to_blocks_per_strip(info
) - 1);
4224 static void imsm_update_version_info(struct intel_super
*super
)
4226 /* update the version and attributes */
4227 struct imsm_super
*mpb
= super
->anchor
;
4229 struct imsm_dev
*dev
;
4230 struct imsm_map
*map
;
4233 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
4234 dev
= get_imsm_dev(super
, i
);
4235 map
= get_imsm_map(dev
, MAP_0
);
4236 if (__le32_to_cpu(dev
->size_high
) > 0)
4237 mpb
->attributes
|= MPB_ATTRIB_2TB
;
4239 /* FIXME detect when an array spans a port multiplier */
4241 mpb
->attributes
|= MPB_ATTRIB_PM
;
4244 if (mpb
->num_raid_devs
> 1 ||
4245 mpb
->attributes
!= MPB_ATTRIB_CHECKSUM_VERIFY
) {
4246 version
= MPB_VERSION_ATTRIBS
;
4247 switch (get_imsm_raid_level(map
)) {
4248 case 0: mpb
->attributes
|= MPB_ATTRIB_RAID0
; break;
4249 case 1: mpb
->attributes
|= MPB_ATTRIB_RAID1
; break;
4250 case 10: mpb
->attributes
|= MPB_ATTRIB_RAID10
; break;
4251 case 5: mpb
->attributes
|= MPB_ATTRIB_RAID5
; break;
4254 if (map
->num_members
>= 5)
4255 version
= MPB_VERSION_5OR6_DISK_ARRAY
;
4256 else if (dev
->status
== DEV_CLONE_N_GO
)
4257 version
= MPB_VERSION_CNG
;
4258 else if (get_imsm_raid_level(map
) == 5)
4259 version
= MPB_VERSION_RAID5
;
4260 else if (map
->num_members
>= 3)
4261 version
= MPB_VERSION_3OR4_DISK_ARRAY
;
4262 else if (get_imsm_raid_level(map
) == 1)
4263 version
= MPB_VERSION_RAID1
;
4265 version
= MPB_VERSION_RAID0
;
4267 strcpy(((char *) mpb
->sig
) + strlen(MPB_SIGNATURE
), version
);
4271 static int check_name(struct intel_super
*super
, char *name
, int quiet
)
4273 struct imsm_super
*mpb
= super
->anchor
;
4274 char *reason
= NULL
;
4277 if (strlen(name
) > MAX_RAID_SERIAL_LEN
)
4278 reason
= "must be 16 characters or less";
4280 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
4281 struct imsm_dev
*dev
= get_imsm_dev(super
, i
);
4283 if (strncmp((char *) dev
->volume
, name
, MAX_RAID_SERIAL_LEN
) == 0) {
4284 reason
= "already exists";
4289 if (reason
&& !quiet
)
4290 fprintf(stderr
, Name
": imsm volume name %s\n", reason
);
4295 static int init_super_imsm_volume(struct supertype
*st
, mdu_array_info_t
*info
,
4296 unsigned long long size
, char *name
,
4297 char *homehost
, int *uuid
)
4299 /* We are creating a volume inside a pre-existing container.
4300 * so st->sb is already set.
4302 struct intel_super
*super
= st
->sb
;
4303 struct imsm_super
*mpb
= super
->anchor
;
4304 struct intel_dev
*dv
;
4305 struct imsm_dev
*dev
;
4306 struct imsm_vol
*vol
;
4307 struct imsm_map
*map
;
4308 int idx
= mpb
->num_raid_devs
;
4310 unsigned long long array_blocks
;
4311 size_t size_old
, size_new
;
4312 __u32 num_data_stripes
;
4314 if (super
->orom
&& mpb
->num_raid_devs
>= super
->orom
->vpa
) {
4315 fprintf(stderr
, Name
": This imsm-container already has the "
4316 "maximum of %d volumes\n", super
->orom
->vpa
);
4320 /* ensure the mpb is large enough for the new data */
4321 size_old
= __le32_to_cpu(mpb
->mpb_size
);
4322 size_new
= disks_to_mpb_size(info
->nr_disks
);
4323 if (size_new
> size_old
) {
4325 size_t size_round
= ROUND_UP(size_new
, 512);
4327 if (posix_memalign(&mpb_new
, 512, size_round
) != 0) {
4328 fprintf(stderr
, Name
": could not allocate new mpb\n");
4331 if (posix_memalign(&super
->migr_rec_buf
, 512,
4332 MIGR_REC_BUF_SIZE
) != 0) {
4333 fprintf(stderr
, Name
4334 ": %s could not allocate migr_rec buffer\n",
4341 memcpy(mpb_new
, mpb
, size_old
);
4344 super
->anchor
= mpb_new
;
4345 mpb
->mpb_size
= __cpu_to_le32(size_new
);
4346 memset(mpb_new
+ size_old
, 0, size_round
- size_old
);
4348 super
->current_vol
= idx
;
4350 /* handle 'failed_disks' by either:
4351 * a) create dummy disk entries in the table if this the first
4352 * volume in the array. We add them here as this is the only
4353 * opportunity to add them. add_to_super_imsm_volume()
4354 * handles the non-failed disks and continues incrementing
4356 * b) validate that 'failed_disks' matches the current number
4357 * of missing disks if the container is populated
4359 if (super
->current_vol
== 0) {
4361 for (i
= 0; i
< info
->failed_disks
; i
++) {
4362 struct imsm_disk
*disk
;
4365 disk
= __get_imsm_disk(mpb
, i
);
4366 disk
->status
= CONFIGURED_DISK
| FAILED_DISK
;
4367 disk
->scsi_id
= __cpu_to_le32(~(__u32
)0);
4368 snprintf((char *) disk
->serial
, MAX_RAID_SERIAL_LEN
,
4371 find_missing(super
);
4376 for (d
= super
->missing
; d
; d
= d
->next
)
4378 if (info
->failed_disks
> missing
) {
4379 fprintf(stderr
, Name
": unable to add 'missing' disk to container\n");
4384 if (!check_name(super
, name
, 0))
4386 dv
= malloc(sizeof(*dv
));
4388 fprintf(stderr
, Name
": failed to allocate device list entry\n");
4391 dev
= calloc(1, sizeof(*dev
) + sizeof(__u32
) * (info
->raid_disks
- 1));
4394 fprintf(stderr
, Name
": could not allocate raid device\n");
4398 strncpy((char *) dev
->volume
, name
, MAX_RAID_SERIAL_LEN
);
4399 if (info
->level
== 1)
4400 array_blocks
= info_to_blocks_per_member(info
);
4402 array_blocks
= calc_array_size(info
->level
, info
->raid_disks
,
4403 info
->layout
, info
->chunk_size
,
4405 /* round array size down to closest MB */
4406 array_blocks
= (array_blocks
>> SECT_PER_MB_SHIFT
) << SECT_PER_MB_SHIFT
;
4408 dev
->size_low
= __cpu_to_le32((__u32
) array_blocks
);
4409 dev
->size_high
= __cpu_to_le32((__u32
) (array_blocks
>> 32));
4410 dev
->status
= (DEV_READ_COALESCING
| DEV_WRITE_COALESCING
);
4412 vol
->migr_state
= 0;
4413 set_migr_type(dev
, MIGR_INIT
);
4414 vol
->dirty
= !info
->state
;
4415 vol
->curr_migr_unit
= 0;
4416 map
= get_imsm_map(dev
, MAP_0
);
4417 map
->pba_of_lba0
= __cpu_to_le32(super
->create_offset
);
4418 map
->blocks_per_member
= __cpu_to_le32(info_to_blocks_per_member(info
));
4419 map
->blocks_per_strip
= __cpu_to_le16(info_to_blocks_per_strip(info
));
4420 map
->failed_disk_num
= ~0;
4421 if (info
->level
> 0)
4422 map
->map_state
= IMSM_T_STATE_UNINITIALIZED
;
4424 map
->map_state
= info
->failed_disks
? IMSM_T_STATE_FAILED
:
4425 IMSM_T_STATE_NORMAL
;
4428 if (info
->level
== 1 && info
->raid_disks
> 2) {
4431 fprintf(stderr
, Name
": imsm does not support more than 2 disks"
4432 "in a raid1 volume\n");
4436 map
->raid_level
= info
->level
;
4437 if (info
->level
== 10) {
4438 map
->raid_level
= 1;
4439 map
->num_domains
= info
->raid_disks
/ 2;
4440 } else if (info
->level
== 1)
4441 map
->num_domains
= info
->raid_disks
;
4443 map
->num_domains
= 1;
4445 num_data_stripes
= info_to_num_data_stripes(info
, map
->num_domains
);
4446 map
->num_data_stripes
= __cpu_to_le32(num_data_stripes
);
4448 map
->num_members
= info
->raid_disks
;
4449 for (i
= 0; i
< map
->num_members
; i
++) {
4450 /* initialized in add_to_super */
4451 set_imsm_ord_tbl_ent(map
, i
, IMSM_ORD_REBUILD
);
4453 mpb
->num_raid_devs
++;
4456 dv
->index
= super
->current_vol
;
4457 dv
->next
= super
->devlist
;
4458 super
->devlist
= dv
;
4460 imsm_update_version_info(super
);
4465 static int init_super_imsm(struct supertype
*st
, mdu_array_info_t
*info
,
4466 unsigned long long size
, char *name
,
4467 char *homehost
, int *uuid
)
4469 /* This is primarily called by Create when creating a new array.
4470 * We will then get add_to_super called for each component, and then
4471 * write_init_super called to write it out to each device.
4472 * For IMSM, Create can create on fresh devices or on a pre-existing
4474 * To create on a pre-existing array a different method will be called.
4475 * This one is just for fresh drives.
4477 struct intel_super
*super
;
4478 struct imsm_super
*mpb
;
4483 return init_super_imsm_volume(st
, info
, size
, name
, homehost
, uuid
);
4486 mpb_size
= disks_to_mpb_size(info
->nr_disks
);
4490 super
= alloc_super();
4491 if (super
&& posix_memalign(&super
->buf
, 512, mpb_size
) != 0) {
4496 fprintf(stderr
, Name
4497 ": %s could not allocate superblock\n", __func__
);
4500 if (posix_memalign(&super
->migr_rec_buf
, 512, MIGR_REC_BUF_SIZE
) != 0) {
4501 fprintf(stderr
, Name
4502 ": %s could not allocate migr_rec buffer\n", __func__
);
4507 memset(super
->buf
, 0, mpb_size
);
4509 mpb
->mpb_size
= __cpu_to_le32(mpb_size
);
4513 /* zeroing superblock */
4517 mpb
->attributes
= MPB_ATTRIB_CHECKSUM_VERIFY
;
4519 version
= (char *) mpb
->sig
;
4520 strcpy(version
, MPB_SIGNATURE
);
4521 version
+= strlen(MPB_SIGNATURE
);
4522 strcpy(version
, MPB_VERSION_RAID0
);
4528 static int add_to_super_imsm_volume(struct supertype
*st
, mdu_disk_info_t
*dk
,
4529 int fd
, char *devname
)
4531 struct intel_super
*super
= st
->sb
;
4532 struct imsm_super
*mpb
= super
->anchor
;
4533 struct imsm_disk
*_disk
;
4534 struct imsm_dev
*dev
;
4535 struct imsm_map
*map
;
4539 dev
= get_imsm_dev(super
, super
->current_vol
);
4540 map
= get_imsm_map(dev
, MAP_0
);
4542 if (! (dk
->state
& (1<<MD_DISK_SYNC
))) {
4543 fprintf(stderr
, Name
": %s: Cannot add spare devices to IMSM volume\n",
4549 /* we're doing autolayout so grab the pre-marked (in
4550 * validate_geometry) raid_disk
4552 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
4553 if (dl
->raiddisk
== dk
->raid_disk
)
4556 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
4557 if (dl
->major
== dk
->major
&&
4558 dl
->minor
== dk
->minor
)
4563 fprintf(stderr
, Name
": %s is not a member of the same container\n", devname
);
4567 /* add a pristine spare to the metadata */
4568 if (dl
->index
< 0) {
4569 dl
->index
= super
->anchor
->num_disks
;
4570 super
->anchor
->num_disks
++;
4572 /* Check the device has not already been added */
4573 slot
= get_imsm_disk_slot(map
, dl
->index
);
4575 (get_imsm_ord_tbl_ent(dev
, slot
, MAP_X
) & IMSM_ORD_REBUILD
) == 0) {
4576 fprintf(stderr
, Name
": %s has been included in this array twice\n",
4580 set_imsm_ord_tbl_ent(map
, dk
->raid_disk
, dl
->index
);
4581 dl
->disk
.status
= CONFIGURED_DISK
;
4583 /* update size of 'missing' disks to be at least as large as the
4584 * largest acitve member (we only have dummy missing disks when
4585 * creating the first volume)
4587 if (super
->current_vol
== 0) {
4588 for (df
= super
->missing
; df
; df
= df
->next
) {
4589 if (dl
->disk
.total_blocks
> df
->disk
.total_blocks
)
4590 df
->disk
.total_blocks
= dl
->disk
.total_blocks
;
4591 _disk
= __get_imsm_disk(mpb
, df
->index
);
4596 /* refresh unset/failed slots to point to valid 'missing' entries */
4597 for (df
= super
->missing
; df
; df
= df
->next
)
4598 for (slot
= 0; slot
< mpb
->num_disks
; slot
++) {
4599 __u32 ord
= get_imsm_ord_tbl_ent(dev
, slot
, MAP_X
);
4601 if ((ord
& IMSM_ORD_REBUILD
) == 0)
4603 set_imsm_ord_tbl_ent(map
, slot
, df
->index
| IMSM_ORD_REBUILD
);
4604 if (is_gen_migration(dev
)) {
4605 struct imsm_map
*map2
= get_imsm_map(dev
,
4607 int slot2
= get_imsm_disk_slot(map2
, df
->index
);
4608 if ((slot2
< map2
->num_members
) &&
4610 __u32 ord2
= get_imsm_ord_tbl_ent(dev
,
4613 if ((unsigned)df
->index
==
4615 set_imsm_ord_tbl_ent(map2
,
4621 dprintf("set slot:%d to missing disk:%d\n", slot
, df
->index
);
4625 /* if we are creating the first raid device update the family number */
4626 if (super
->current_vol
== 0) {
4628 struct imsm_dev
*_dev
= __get_imsm_dev(mpb
, 0);
4630 _disk
= __get_imsm_disk(mpb
, dl
->index
);
4631 if (!_dev
|| !_disk
) {
4632 fprintf(stderr
, Name
": BUG mpb setup error\n");
4638 sum
+= __gen_imsm_checksum(mpb
);
4639 mpb
->family_num
= __cpu_to_le32(sum
);
4640 mpb
->orig_family_num
= mpb
->family_num
;
4642 super
->current_disk
= dl
;
4647 * Function marks disk as spare and restores disk serial
4648 * in case it was previously marked as failed by takeover operation
4650 * -1 : critical error
4651 * 0 : disk is marked as spare but serial is not set
4654 int mark_spare(struct dl
*disk
)
4656 __u8 serial
[MAX_RAID_SERIAL_LEN
];
4663 if (!imsm_read_serial(disk
->fd
, NULL
, serial
)) {
4664 /* Restore disk serial number, because takeover marks disk
4665 * as failed and adds to serial ':0' before it becomes
4668 serialcpy(disk
->serial
, serial
);
4669 serialcpy(disk
->disk
.serial
, serial
);
4672 disk
->disk
.status
= SPARE_DISK
;
4678 static int add_to_super_imsm(struct supertype
*st
, mdu_disk_info_t
*dk
,
4679 int fd
, char *devname
)
4681 struct intel_super
*super
= st
->sb
;
4683 unsigned long long size
;
4688 /* If we are on an RAID enabled platform check that the disk is
4689 * attached to the raid controller.
4690 * We do not need to test disks attachment for container based additions,
4691 * they shall be already tested when container was created/assembled.
4693 rv
= find_intel_hba_capability(fd
, super
, devname
);
4694 /* no orom/efi or non-intel hba of the disk */
4696 dprintf("capability: %p fd: %d ret: %d\n",
4697 super
->orom
, fd
, rv
);
4701 if (super
->current_vol
>= 0)
4702 return add_to_super_imsm_volume(st
, dk
, fd
, devname
);
4705 dd
= malloc(sizeof(*dd
));
4708 Name
": malloc failed %s:%d.\n", __func__
, __LINE__
);
4711 memset(dd
, 0, sizeof(*dd
));
4712 dd
->major
= major(stb
.st_rdev
);
4713 dd
->minor
= minor(stb
.st_rdev
);
4714 dd
->devname
= devname
? strdup(devname
) : NULL
;
4717 dd
->action
= DISK_ADD
;
4718 rv
= imsm_read_serial(fd
, devname
, dd
->serial
);
4721 Name
": failed to retrieve scsi serial, aborting\n");
4726 get_dev_size(fd
, NULL
, &size
);
4728 serialcpy(dd
->disk
.serial
, dd
->serial
);
4729 dd
->disk
.total_blocks
= __cpu_to_le32(size
);
4731 if (sysfs_disk_to_scsi_id(fd
, &id
) == 0)
4732 dd
->disk
.scsi_id
= __cpu_to_le32(id
);
4734 dd
->disk
.scsi_id
= __cpu_to_le32(0);
4736 if (st
->update_tail
) {
4737 dd
->next
= super
->disk_mgmt_list
;
4738 super
->disk_mgmt_list
= dd
;
4740 dd
->next
= super
->disks
;
4742 super
->updates_pending
++;
4749 static int remove_from_super_imsm(struct supertype
*st
, mdu_disk_info_t
*dk
)
4751 struct intel_super
*super
= st
->sb
;
4754 /* remove from super works only in mdmon - for communication
4755 * manager - monitor. Check if communication memory buffer
4758 if (!st
->update_tail
) {
4760 Name
": %s shall be used in mdmon context only"
4761 "(line %d).\n", __func__
, __LINE__
);
4764 dd
= malloc(sizeof(*dd
));
4767 Name
": malloc failed %s:%d.\n", __func__
, __LINE__
);
4770 memset(dd
, 0, sizeof(*dd
));
4771 dd
->major
= dk
->major
;
4772 dd
->minor
= dk
->minor
;
4775 dd
->action
= DISK_REMOVE
;
4777 dd
->next
= super
->disk_mgmt_list
;
4778 super
->disk_mgmt_list
= dd
;
4784 static int store_imsm_mpb(int fd
, struct imsm_super
*mpb
);
4788 struct imsm_super anchor
;
4789 } spare_record
__attribute__ ((aligned(512)));
4791 /* spare records have their own family number and do not have any defined raid
4794 static int write_super_imsm_spares(struct intel_super
*super
, int doclose
)
4796 struct imsm_super
*mpb
= super
->anchor
;
4797 struct imsm_super
*spare
= &spare_record
.anchor
;
4801 spare
->mpb_size
= __cpu_to_le32(sizeof(struct imsm_super
)),
4802 spare
->generation_num
= __cpu_to_le32(1UL),
4803 spare
->attributes
= MPB_ATTRIB_CHECKSUM_VERIFY
;
4804 spare
->num_disks
= 1,
4805 spare
->num_raid_devs
= 0,
4806 spare
->cache_size
= mpb
->cache_size
,
4807 spare
->pwr_cycle_count
= __cpu_to_le32(1),
4809 snprintf((char *) spare
->sig
, MAX_SIGNATURE_LENGTH
,
4810 MPB_SIGNATURE MPB_VERSION_RAID0
);
4812 for (d
= super
->disks
; d
; d
= d
->next
) {
4816 spare
->disk
[0] = d
->disk
;
4817 sum
= __gen_imsm_checksum(spare
);
4818 spare
->family_num
= __cpu_to_le32(sum
);
4819 spare
->orig_family_num
= 0;
4820 sum
= __gen_imsm_checksum(spare
);
4821 spare
->check_sum
= __cpu_to_le32(sum
);
4823 if (store_imsm_mpb(d
->fd
, spare
)) {
4824 fprintf(stderr
, "%s: failed for device %d:%d %s\n",
4825 __func__
, d
->major
, d
->minor
, strerror(errno
));
4837 static int write_super_imsm(struct supertype
*st
, int doclose
)
4839 struct intel_super
*super
= st
->sb
;
4840 struct imsm_super
*mpb
= super
->anchor
;
4846 __u32 mpb_size
= sizeof(struct imsm_super
) - sizeof(struct imsm_disk
);
4848 int clear_migration_record
= 1;
4850 /* 'generation' is incremented everytime the metadata is written */
4851 generation
= __le32_to_cpu(mpb
->generation_num
);
4853 mpb
->generation_num
= __cpu_to_le32(generation
);
4855 /* fix up cases where previous mdadm releases failed to set
4858 if (mpb
->orig_family_num
== 0)
4859 mpb
->orig_family_num
= mpb
->family_num
;
4861 for (d
= super
->disks
; d
; d
= d
->next
) {
4865 mpb
->disk
[d
->index
] = d
->disk
;
4869 for (d
= super
->missing
; d
; d
= d
->next
) {
4870 mpb
->disk
[d
->index
] = d
->disk
;
4873 mpb
->num_disks
= num_disks
;
4874 mpb_size
+= sizeof(struct imsm_disk
) * mpb
->num_disks
;
4876 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
4877 struct imsm_dev
*dev
= __get_imsm_dev(mpb
, i
);
4878 struct imsm_dev
*dev2
= get_imsm_dev(super
, i
);
4880 imsm_copy_dev(dev
, dev2
);
4881 mpb_size
+= sizeof_imsm_dev(dev
, 0);
4883 if (is_gen_migration(dev2
))
4884 clear_migration_record
= 0;
4886 mpb_size
+= __le32_to_cpu(mpb
->bbm_log_size
);
4887 mpb
->mpb_size
= __cpu_to_le32(mpb_size
);
4889 /* recalculate checksum */
4890 sum
= __gen_imsm_checksum(mpb
);
4891 mpb
->check_sum
= __cpu_to_le32(sum
);
4893 if (clear_migration_record
)
4894 memset(super
->migr_rec_buf
, 0, MIGR_REC_BUF_SIZE
);
4896 /* write the mpb for disks that compose raid devices */
4897 for (d
= super
->disks
; d
; d
= d
->next
) {
4898 if (d
->index
< 0 || is_failed(&d
->disk
))
4900 if (store_imsm_mpb(d
->fd
, mpb
))
4901 fprintf(stderr
, "%s: failed for device %d:%d %s\n",
4902 __func__
, d
->major
, d
->minor
, strerror(errno
));
4903 if (clear_migration_record
) {
4904 unsigned long long dsize
;
4906 get_dev_size(d
->fd
, NULL
, &dsize
);
4907 if (lseek64(d
->fd
, dsize
- 512, SEEK_SET
) >= 0) {
4908 if (write(d
->fd
, super
->migr_rec_buf
,
4909 MIGR_REC_BUF_SIZE
) != MIGR_REC_BUF_SIZE
)
4910 perror("Write migr_rec failed");
4920 return write_super_imsm_spares(super
, doclose
);
4926 static int create_array(struct supertype
*st
, int dev_idx
)
4929 struct imsm_update_create_array
*u
;
4930 struct intel_super
*super
= st
->sb
;
4931 struct imsm_dev
*dev
= get_imsm_dev(super
, dev_idx
);
4932 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
4933 struct disk_info
*inf
;
4934 struct imsm_disk
*disk
;
4937 len
= sizeof(*u
) - sizeof(*dev
) + sizeof_imsm_dev(dev
, 0) +
4938 sizeof(*inf
) * map
->num_members
;
4941 fprintf(stderr
, "%s: failed to allocate update buffer\n",
4946 u
->type
= update_create_array
;
4947 u
->dev_idx
= dev_idx
;
4948 imsm_copy_dev(&u
->dev
, dev
);
4949 inf
= get_disk_info(u
);
4950 for (i
= 0; i
< map
->num_members
; i
++) {
4951 int idx
= get_imsm_disk_idx(dev
, i
, MAP_X
);
4953 disk
= get_imsm_disk(super
, idx
);
4954 serialcpy(inf
[i
].serial
, disk
->serial
);
4956 append_metadata_update(st
, u
, len
);
4961 static int mgmt_disk(struct supertype
*st
)
4963 struct intel_super
*super
= st
->sb
;
4965 struct imsm_update_add_remove_disk
*u
;
4967 if (!super
->disk_mgmt_list
)
4973 fprintf(stderr
, "%s: failed to allocate update buffer\n",
4978 u
->type
= update_add_remove_disk
;
4979 append_metadata_update(st
, u
, len
);
4984 static int write_init_super_imsm(struct supertype
*st
)
4986 struct intel_super
*super
= st
->sb
;
4987 int current_vol
= super
->current_vol
;
4989 /* we are done with current_vol reset it to point st at the container */
4990 super
->current_vol
= -1;
4992 if (st
->update_tail
) {
4993 /* queue the recently created array / added disk
4994 * as a metadata update */
4997 /* determine if we are creating a volume or adding a disk */
4998 if (current_vol
< 0) {
4999 /* in the mgmt (add/remove) disk case we are running
5000 * in mdmon context, so don't close fd's
5002 return mgmt_disk(st
);
5004 rv
= create_array(st
, current_vol
);
5009 for (d
= super
->disks
; d
; d
= d
->next
)
5010 Kill(d
->devname
, NULL
, 0, 1, 1);
5011 return write_super_imsm(st
, 1);
5016 static int store_super_imsm(struct supertype
*st
, int fd
)
5018 struct intel_super
*super
= st
->sb
;
5019 struct imsm_super
*mpb
= super
? super
->anchor
: NULL
;
5025 return store_imsm_mpb(fd
, mpb
);
5031 static int imsm_bbm_log_size(struct imsm_super
*mpb
)
5033 return __le32_to_cpu(mpb
->bbm_log_size
);
5037 static int validate_geometry_imsm_container(struct supertype
*st
, int level
,
5038 int layout
, int raiddisks
, int chunk
,
5039 unsigned long long size
, char *dev
,
5040 unsigned long long *freesize
,
5044 unsigned long long ldsize
;
5045 struct intel_super
*super
=NULL
;
5048 if (level
!= LEVEL_CONTAINER
)
5053 fd
= open(dev
, O_RDONLY
|O_EXCL
, 0);
5056 fprintf(stderr
, Name
": imsm: Cannot open %s: %s\n",
5057 dev
, strerror(errno
));
5060 if (!get_dev_size(fd
, dev
, &ldsize
)) {
5065 /* capabilities retrieve could be possible
5066 * note that there is no fd for the disks in array.
5068 super
= alloc_super();
5071 Name
": malloc of %zu failed.\n",
5077 rv
= find_intel_hba_capability(fd
, super
, verbose
? dev
: NULL
);
5081 fd2devname(fd
, str
);
5082 dprintf("validate_geometry_imsm_container: fd: %d %s orom: %p rv: %d raiddisk: %d\n",
5083 fd
, str
, super
->orom
, rv
, raiddisks
);
5085 /* no orom/efi or non-intel hba of the disk */
5091 if (super
->orom
&& raiddisks
> super
->orom
->tds
) {
5093 fprintf(stderr
, Name
": %d exceeds maximum number of"
5094 " platform supported disks: %d\n",
5095 raiddisks
, super
->orom
->tds
);
5101 *freesize
= avail_size_imsm(st
, ldsize
>> 9);
5107 static unsigned long long find_size(struct extent
*e
, int *idx
, int num_extents
)
5109 const unsigned long long base_start
= e
[*idx
].start
;
5110 unsigned long long end
= base_start
+ e
[*idx
].size
;
5113 if (base_start
== end
)
5117 for (i
= *idx
; i
< num_extents
; i
++) {
5118 /* extend overlapping extents */
5119 if (e
[i
].start
>= base_start
&&
5120 e
[i
].start
<= end
) {
5123 if (e
[i
].start
+ e
[i
].size
> end
)
5124 end
= e
[i
].start
+ e
[i
].size
;
5125 } else if (e
[i
].start
> end
) {
5131 return end
- base_start
;
5134 static unsigned long long merge_extents(struct intel_super
*super
, int sum_extents
)
5136 /* build a composite disk with all known extents and generate a new
5137 * 'maxsize' given the "all disks in an array must share a common start
5138 * offset" constraint
5140 struct extent
*e
= calloc(sum_extents
, sizeof(*e
));
5144 unsigned long long pos
;
5145 unsigned long long start
= 0;
5146 unsigned long long maxsize
;
5147 unsigned long reserve
;
5152 /* coalesce and sort all extents. also, check to see if we need to
5153 * reserve space between member arrays
5156 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
5159 for (i
= 0; i
< dl
->extent_cnt
; i
++)
5162 qsort(e
, sum_extents
, sizeof(*e
), cmp_extent
);
5167 while (i
< sum_extents
) {
5168 e
[j
].start
= e
[i
].start
;
5169 e
[j
].size
= find_size(e
, &i
, sum_extents
);
5171 if (e
[j
-1].size
== 0)
5180 unsigned long long esize
;
5182 esize
= e
[i
].start
- pos
;
5183 if (esize
>= maxsize
) {
5188 pos
= e
[i
].start
+ e
[i
].size
;
5190 } while (e
[i
-1].size
);
5196 /* FIXME assumes volume at offset 0 is the first volume in a
5199 if (start_extent
> 0)
5200 reserve
= IMSM_RESERVED_SECTORS
; /* gap between raid regions */
5204 if (maxsize
< reserve
)
5207 super
->create_offset
= ~((__u32
) 0);
5208 if (start
+ reserve
> super
->create_offset
)
5209 return 0; /* start overflows create_offset */
5210 super
->create_offset
= start
+ reserve
;
5212 return maxsize
- reserve
;
5215 static int is_raid_level_supported(const struct imsm_orom
*orom
, int level
, int raiddisks
)
5217 if (level
< 0 || level
== 6 || level
== 4)
5220 /* if we have an orom prevent invalid raid levels */
5223 case 0: return imsm_orom_has_raid0(orom
);
5226 return imsm_orom_has_raid1e(orom
);
5227 return imsm_orom_has_raid1(orom
) && raiddisks
== 2;
5228 case 10: return imsm_orom_has_raid10(orom
) && raiddisks
== 4;
5229 case 5: return imsm_orom_has_raid5(orom
) && raiddisks
> 2;
5232 return 1; /* not on an Intel RAID platform so anything goes */
5237 static int imsm_default_chunk(const struct imsm_orom
*orom
)
5239 /* up to 512 if the plaform supports it, otherwise the platform max.
5240 * 128 if no platform detected
5242 int fs
= max(7, orom
? fls(orom
->sss
) : 0);
5244 return min(512, (1 << fs
));
5247 #define pr_vrb(fmt, arg...) (void) (verbose && fprintf(stderr, Name fmt, ##arg))
5249 validate_geometry_imsm_orom(struct intel_super
*super
, int level
, int layout
,
5250 int raiddisks
, int *chunk
, int verbose
)
5252 /* check/set platform and metadata limits/defaults */
5253 if (super
->orom
&& raiddisks
> super
->orom
->dpa
) {
5254 pr_vrb(": platform supports a maximum of %d disks per array\n",
5259 /* capabilities of OROM tested - copied from validate_geometry_imsm_volume */
5260 if (!is_raid_level_supported(super
->orom
, level
, raiddisks
)) {
5261 pr_vrb(": platform does not support raid%d with %d disk%s\n",
5262 level
, raiddisks
, raiddisks
> 1 ? "s" : "");
5266 if (chunk
&& (*chunk
== 0 || *chunk
== UnSet
))
5267 *chunk
= imsm_default_chunk(super
->orom
);
5269 if (super
->orom
&& chunk
&& !imsm_orom_has_chunk(super
->orom
, *chunk
)) {
5270 pr_vrb(": platform does not support a chunk size of: "
5275 if (layout
!= imsm_level_to_layout(level
)) {
5277 pr_vrb(": imsm raid 5 only supports the left-asymmetric layout\n");
5278 else if (level
== 10)
5279 pr_vrb(": imsm raid 10 only supports the n2 layout\n");
5281 pr_vrb(": imsm unknown layout %#x for this raid level %d\n",
5288 /* validate_geometry_imsm_volume - lifted from validate_geometry_ddf_bvd
5289 * FIX ME add ahci details
5291 static int validate_geometry_imsm_volume(struct supertype
*st
, int level
,
5292 int layout
, int raiddisks
, int *chunk
,
5293 unsigned long long size
, char *dev
,
5294 unsigned long long *freesize
,
5298 struct intel_super
*super
= st
->sb
;
5299 struct imsm_super
*mpb
;
5301 unsigned long long pos
= 0;
5302 unsigned long long maxsize
;
5306 /* We must have the container info already read in. */
5310 mpb
= super
->anchor
;
5312 if (mpb
->num_raid_devs
> 0 && mpb
->num_disks
!= raiddisks
) {
5313 fprintf(stderr
, Name
": the option-rom requires all "
5314 "member disks to be a member of all volumes.\n");
5318 if (!validate_geometry_imsm_orom(super
, level
, layout
, raiddisks
, chunk
, verbose
)) {
5319 fprintf(stderr
, Name
": RAID gemetry validation failed. "
5320 "Cannot proceed with the action(s).\n");
5324 /* General test: make sure there is space for
5325 * 'raiddisks' device extents of size 'size' at a given
5328 unsigned long long minsize
= size
;
5329 unsigned long long start_offset
= MaxSector
;
5332 minsize
= MPB_SECTOR_CNT
+ IMSM_RESERVED_SECTORS
;
5333 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
5338 e
= get_extents(super
, dl
);
5341 unsigned long long esize
;
5342 esize
= e
[i
].start
- pos
;
5343 if (esize
>= minsize
)
5345 if (found
&& start_offset
== MaxSector
) {
5348 } else if (found
&& pos
!= start_offset
) {
5352 pos
= e
[i
].start
+ e
[i
].size
;
5354 } while (e
[i
-1].size
);
5359 if (dcnt
< raiddisks
) {
5361 fprintf(stderr
, Name
": imsm: Not enough "
5362 "devices with space for this array "
5370 /* This device must be a member of the set */
5371 if (stat(dev
, &stb
) < 0)
5373 if ((S_IFMT
& stb
.st_mode
) != S_IFBLK
)
5375 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
5376 if (dl
->major
== (int)major(stb
.st_rdev
) &&
5377 dl
->minor
== (int)minor(stb
.st_rdev
))
5382 fprintf(stderr
, Name
": %s is not in the "
5383 "same imsm set\n", dev
);
5385 } else if (super
->orom
&& dl
->index
< 0 && mpb
->num_raid_devs
) {
5386 /* If a volume is present then the current creation attempt
5387 * cannot incorporate new spares because the orom may not
5388 * understand this configuration (all member disks must be
5389 * members of each array in the container).
5391 fprintf(stderr
, Name
": %s is a spare and a volume"
5392 " is already defined for this container\n", dev
);
5393 fprintf(stderr
, Name
": The option-rom requires all member"
5394 " disks to be a member of all volumes\n");
5398 /* retrieve the largest free space block */
5399 e
= get_extents(super
, dl
);
5404 unsigned long long esize
;
5406 esize
= e
[i
].start
- pos
;
5407 if (esize
>= maxsize
)
5409 pos
= e
[i
].start
+ e
[i
].size
;
5411 } while (e
[i
-1].size
);
5416 fprintf(stderr
, Name
": unable to determine free space for: %s\n",
5420 if (maxsize
< size
) {
5422 fprintf(stderr
, Name
": %s not enough space (%llu < %llu)\n",
5423 dev
, maxsize
, size
);
5427 /* count total number of extents for merge */
5429 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
5431 i
+= dl
->extent_cnt
;
5433 maxsize
= merge_extents(super
, i
);
5435 if (!check_env("IMSM_NO_PLATFORM") &&
5436 mpb
->num_raid_devs
> 0 && size
&& size
!= maxsize
) {
5437 fprintf(stderr
, Name
": attempting to create a second "
5438 "volume with size less then remaining space. "
5443 if (maxsize
< size
|| maxsize
== 0) {
5445 fprintf(stderr
, Name
": not enough space after merge (%llu < %llu)\n",
5450 *freesize
= maxsize
;
5455 static int reserve_space(struct supertype
*st
, int raiddisks
,
5456 unsigned long long size
, int chunk
,
5457 unsigned long long *freesize
)
5459 struct intel_super
*super
= st
->sb
;
5460 struct imsm_super
*mpb
= super
->anchor
;
5465 unsigned long long maxsize
;
5466 unsigned long long minsize
;
5470 /* find the largest common start free region of the possible disks */
5474 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
5480 /* don't activate new spares if we are orom constrained
5481 * and there is already a volume active in the container
5483 if (super
->orom
&& dl
->index
< 0 && mpb
->num_raid_devs
)
5486 e
= get_extents(super
, dl
);
5489 for (i
= 1; e
[i
-1].size
; i
++)
5497 maxsize
= merge_extents(super
, extent_cnt
);
5501 minsize
= chunk
* 2;
5503 if (cnt
< raiddisks
||
5504 (super
->orom
&& used
&& used
!= raiddisks
) ||
5505 maxsize
< minsize
||
5507 fprintf(stderr
, Name
": not enough devices with space to create array.\n");
5508 return 0; /* No enough free spaces large enough */
5520 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
5522 dl
->raiddisk
= cnt
++;
5529 static int validate_geometry_imsm(struct supertype
*st
, int level
, int layout
,
5530 int raiddisks
, int *chunk
, unsigned long long size
,
5531 char *dev
, unsigned long long *freesize
,
5539 * if given unused devices create a container
5540 * if given given devices in a container create a member volume
5542 if (level
== LEVEL_CONTAINER
) {
5543 /* Must be a fresh device to add to a container */
5544 return validate_geometry_imsm_container(st
, level
, layout
,
5546 chunk
?*chunk
:0, size
,
5553 if (!validate_geometry_imsm_orom(st
->sb
, level
, layout
,
5557 /* we are being asked to automatically layout a
5558 * new volume based on the current contents of
5559 * the container. If the the parameters can be
5560 * satisfied reserve_space will record the disks,
5561 * start offset, and size of the volume to be
5562 * created. add_to_super and getinfo_super
5563 * detect when autolayout is in progress.
5566 return reserve_space(st
, raiddisks
, size
,
5567 chunk
?*chunk
:0, freesize
);
5572 /* creating in a given container */
5573 return validate_geometry_imsm_volume(st
, level
, layout
,
5574 raiddisks
, chunk
, size
,
5575 dev
, freesize
, verbose
);
5578 /* This device needs to be a device in an 'imsm' container */
5579 fd
= open(dev
, O_RDONLY
|O_EXCL
, 0);
5583 Name
": Cannot create this array on device %s\n",
5588 if (errno
!= EBUSY
|| (fd
= open(dev
, O_RDONLY
, 0)) < 0) {
5590 fprintf(stderr
, Name
": Cannot open %s: %s\n",
5591 dev
, strerror(errno
));
5594 /* Well, it is in use by someone, maybe an 'imsm' container. */
5595 cfd
= open_container(fd
);
5599 fprintf(stderr
, Name
": Cannot use %s: It is busy\n",
5603 sra
= sysfs_read(cfd
, 0, GET_VERSION
);
5604 if (sra
&& sra
->array
.major_version
== -1 &&
5605 strcmp(sra
->text_version
, "imsm") == 0)
5609 /* This is a member of a imsm container. Load the container
5610 * and try to create a volume
5612 struct intel_super
*super
;
5614 if (load_super_imsm_all(st
, cfd
, (void **) &super
, NULL
) == 0) {
5616 st
->container_dev
= fd2devnum(cfd
);
5618 return validate_geometry_imsm_volume(st
, level
, layout
,
5627 fprintf(stderr
, Name
": failed container membership check\n");
5633 static void default_geometry_imsm(struct supertype
*st
, int *level
, int *layout
, int *chunk
)
5635 struct intel_super
*super
= st
->sb
;
5637 if (level
&& *level
== UnSet
)
5638 *level
= LEVEL_CONTAINER
;
5640 if (level
&& layout
&& *layout
== UnSet
)
5641 *layout
= imsm_level_to_layout(*level
);
5643 if (chunk
&& (*chunk
== UnSet
|| *chunk
== 0))
5644 *chunk
= imsm_default_chunk(super
->orom
);
5647 static void handle_missing(struct intel_super
*super
, struct imsm_dev
*dev
);
5649 static int kill_subarray_imsm(struct supertype
*st
)
5651 /* remove the subarray currently referenced by ->current_vol */
5653 struct intel_dev
**dp
;
5654 struct intel_super
*super
= st
->sb
;
5655 __u8 current_vol
= super
->current_vol
;
5656 struct imsm_super
*mpb
= super
->anchor
;
5658 if (super
->current_vol
< 0)
5660 super
->current_vol
= -1; /* invalidate subarray cursor */
5662 /* block deletions that would change the uuid of active subarrays
5664 * FIXME when immutable ids are available, but note that we'll
5665 * also need to fixup the invalidated/active subarray indexes in
5668 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
5671 if (i
< current_vol
)
5673 sprintf(subarray
, "%u", i
);
5674 if (is_subarray_active(subarray
, st
->devname
)) {
5676 Name
": deleting subarray-%d would change the UUID of active subarray-%d, aborting\n",
5683 if (st
->update_tail
) {
5684 struct imsm_update_kill_array
*u
= malloc(sizeof(*u
));
5688 u
->type
= update_kill_array
;
5689 u
->dev_idx
= current_vol
;
5690 append_metadata_update(st
, u
, sizeof(*u
));
5695 for (dp
= &super
->devlist
; *dp
;)
5696 if ((*dp
)->index
== current_vol
) {
5699 handle_missing(super
, (*dp
)->dev
);
5700 if ((*dp
)->index
> current_vol
)
5705 /* no more raid devices, all active components are now spares,
5706 * but of course failed are still failed
5708 if (--mpb
->num_raid_devs
== 0) {
5711 for (d
= super
->disks
; d
; d
= d
->next
)
5716 super
->updates_pending
++;
5721 static int update_subarray_imsm(struct supertype
*st
, char *subarray
,
5722 char *update
, struct mddev_ident
*ident
)
5724 /* update the subarray currently referenced by ->current_vol */
5725 struct intel_super
*super
= st
->sb
;
5726 struct imsm_super
*mpb
= super
->anchor
;
5728 if (strcmp(update
, "name") == 0) {
5729 char *name
= ident
->name
;
5733 if (is_subarray_active(subarray
, st
->devname
)) {
5735 Name
": Unable to update name of active subarray\n");
5739 if (!check_name(super
, name
, 0))
5742 vol
= strtoul(subarray
, &ep
, 10);
5743 if (*ep
!= '\0' || vol
>= super
->anchor
->num_raid_devs
)
5746 if (st
->update_tail
) {
5747 struct imsm_update_rename_array
*u
= malloc(sizeof(*u
));
5751 u
->type
= update_rename_array
;
5753 snprintf((char *) u
->name
, MAX_RAID_SERIAL_LEN
, "%s", name
);
5754 append_metadata_update(st
, u
, sizeof(*u
));
5756 struct imsm_dev
*dev
;
5759 dev
= get_imsm_dev(super
, vol
);
5760 snprintf((char *) dev
->volume
, MAX_RAID_SERIAL_LEN
, "%s", name
);
5761 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
5762 dev
= get_imsm_dev(super
, i
);
5763 handle_missing(super
, dev
);
5765 super
->updates_pending
++;
5772 #endif /* MDASSEMBLE */
5774 static int is_gen_migration(struct imsm_dev
*dev
)
5779 if (!dev
->vol
.migr_state
)
5782 if (migr_type(dev
) == MIGR_GEN_MIGR
)
5788 static int is_rebuilding(struct imsm_dev
*dev
)
5790 struct imsm_map
*migr_map
;
5792 if (!dev
->vol
.migr_state
)
5795 if (migr_type(dev
) != MIGR_REBUILD
)
5798 migr_map
= get_imsm_map(dev
, MAP_1
);
5800 if (migr_map
->map_state
== IMSM_T_STATE_DEGRADED
)
5807 static int is_initializing(struct imsm_dev
*dev
)
5809 struct imsm_map
*migr_map
;
5811 if (!dev
->vol
.migr_state
)
5814 if (migr_type(dev
) != MIGR_INIT
)
5817 migr_map
= get_imsm_map(dev
, MAP_1
);
5819 if (migr_map
->map_state
== IMSM_T_STATE_UNINITIALIZED
)
5826 static void update_recovery_start(struct intel_super
*super
,
5827 struct imsm_dev
*dev
,
5828 struct mdinfo
*array
)
5830 struct mdinfo
*rebuild
= NULL
;
5834 if (!is_rebuilding(dev
))
5837 /* Find the rebuild target, but punt on the dual rebuild case */
5838 for (d
= array
->devs
; d
; d
= d
->next
)
5839 if (d
->recovery_start
== 0) {
5846 /* (?) none of the disks are marked with
5847 * IMSM_ORD_REBUILD, so assume they are missing and the
5848 * disk_ord_tbl was not correctly updated
5850 dprintf("%s: failed to locate out-of-sync disk\n", __func__
);
5854 units
= __le32_to_cpu(dev
->vol
.curr_migr_unit
);
5855 rebuild
->recovery_start
= units
* blocks_per_migr_unit(super
, dev
);
5859 static int recover_backup_imsm(struct supertype
*st
, struct mdinfo
*info
);
5862 static struct mdinfo
*container_content_imsm(struct supertype
*st
, char *subarray
)
5864 /* Given a container loaded by load_super_imsm_all,
5865 * extract information about all the arrays into
5867 * If 'subarray' is given, just extract info about that array.
5869 * For each imsm_dev create an mdinfo, fill it in,
5870 * then look for matching devices in super->disks
5871 * and create appropriate device mdinfo.
5873 struct intel_super
*super
= st
->sb
;
5874 struct imsm_super
*mpb
= super
->anchor
;
5875 struct mdinfo
*rest
= NULL
;
5879 int spare_disks
= 0;
5881 /* do not assemble arrays when not all attributes are supported */
5882 if (imsm_check_attributes(mpb
->attributes
) == 0) {
5884 fprintf(stderr
, Name
": Unsupported attributes in IMSM metadata."
5885 "Arrays activation is blocked.\n");
5888 /* check for bad blocks */
5889 if (imsm_bbm_log_size(super
->anchor
)) {
5890 fprintf(stderr
, Name
": BBM log found in IMSM metadata."
5891 "Arrays activation is blocked.\n");
5896 /* count spare devices, not used in maps
5898 for (d
= super
->disks
; d
; d
= d
->next
)
5902 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
5903 struct imsm_dev
*dev
;
5904 struct imsm_map
*map
;
5905 struct imsm_map
*map2
;
5906 struct mdinfo
*this;
5914 (i
!= strtoul(subarray
, &ep
, 10) || *ep
!= '\0'))
5917 dev
= get_imsm_dev(super
, i
);
5918 map
= get_imsm_map(dev
, MAP_0
);
5919 map2
= get_imsm_map(dev
, MAP_1
);
5921 /* do not publish arrays that are in the middle of an
5922 * unsupported migration
5924 if (dev
->vol
.migr_state
&&
5925 (migr_type(dev
) == MIGR_STATE_CHANGE
)) {
5926 fprintf(stderr
, Name
": cannot assemble volume '%.16s':"
5927 " unsupported migration in progress\n",
5931 /* do not publish arrays that are not support by controller's
5935 this = malloc(sizeof(*this));
5937 fprintf(stderr
, Name
": failed to allocate %zu bytes\n",
5942 super
->current_vol
= i
;
5943 getinfo_super_imsm_volume(st
, this, NULL
);
5946 chunk
= __le16_to_cpu(map
->blocks_per_strip
) >> 1;
5947 /* mdadm does not support all metadata features- set the bit in all arrays state */
5948 if (!validate_geometry_imsm_orom(super
,
5949 get_imsm_raid_level(map
), /* RAID level */
5950 imsm_level_to_layout(get_imsm_raid_level(map
)),
5951 map
->num_members
, /* raid disks */
5954 fprintf(stderr
, Name
": IMSM RAID geometry validation"
5955 " failed. Array %s activation is blocked.\n",
5957 this->array
.state
|=
5958 (1<<MD_SB_BLOCK_CONTAINER_RESHAPE
) |
5959 (1<<MD_SB_BLOCK_VOLUME
);
5963 /* if array has bad blocks, set suitable bit in all arrays state */
5965 this->array
.state
|=
5966 (1<<MD_SB_BLOCK_CONTAINER_RESHAPE
) |
5967 (1<<MD_SB_BLOCK_VOLUME
);
5969 for (slot
= 0 ; slot
< map
->num_members
; slot
++) {
5970 unsigned long long recovery_start
;
5971 struct mdinfo
*info_d
;
5978 idx
= get_imsm_disk_idx(dev
, slot
, MAP_0
);
5979 ord
= get_imsm_ord_tbl_ent(dev
, slot
, MAP_X
);
5980 for (d
= super
->disks
; d
; d
= d
->next
)
5981 if (d
->index
== idx
)
5984 recovery_start
= MaxSector
;
5987 if (d
&& is_failed(&d
->disk
))
5989 if (ord
& IMSM_ORD_REBUILD
)
5993 * if we skip some disks the array will be assmebled degraded;
5994 * reset resync start to avoid a dirty-degraded
5995 * situation when performing the intial sync
5997 * FIXME handle dirty degraded
5999 if ((skip
|| recovery_start
== 0) && !dev
->vol
.dirty
)
6000 this->resync_start
= MaxSector
;
6004 info_d
= calloc(1, sizeof(*info_d
));
6006 fprintf(stderr
, Name
": failed to allocate disk"
6007 " for volume %.16s\n", dev
->volume
);
6008 info_d
= this->devs
;
6010 struct mdinfo
*d
= info_d
->next
;
6019 info_d
->next
= this->devs
;
6020 this->devs
= info_d
;
6022 info_d
->disk
.number
= d
->index
;
6023 info_d
->disk
.major
= d
->major
;
6024 info_d
->disk
.minor
= d
->minor
;
6025 info_d
->disk
.raid_disk
= slot
;
6026 info_d
->recovery_start
= recovery_start
;
6028 if (slot
< map2
->num_members
)
6029 info_d
->disk
.state
= (1 << MD_DISK_ACTIVE
);
6031 this->array
.spare_disks
++;
6033 if (slot
< map
->num_members
)
6034 info_d
->disk
.state
= (1 << MD_DISK_ACTIVE
);
6036 this->array
.spare_disks
++;
6038 if (info_d
->recovery_start
== MaxSector
)
6039 this->array
.working_disks
++;
6041 info_d
->events
= __le32_to_cpu(mpb
->generation_num
);
6042 info_d
->data_offset
= __le32_to_cpu(map
->pba_of_lba0
);
6043 info_d
->component_size
= __le32_to_cpu(map
->blocks_per_member
);
6045 /* now that the disk list is up-to-date fixup recovery_start */
6046 update_recovery_start(super
, dev
, this);
6047 this->array
.spare_disks
+= spare_disks
;
6050 /* check for reshape */
6051 if (this->reshape_active
== 1)
6052 recover_backup_imsm(st
, this);
6061 static __u8
imsm_check_degraded(struct intel_super
*super
, struct imsm_dev
*dev
,
6062 int failed
, int look_in_map
)
6064 struct imsm_map
*map
;
6066 map
= get_imsm_map(dev
, look_in_map
);
6069 return map
->map_state
== IMSM_T_STATE_UNINITIALIZED
?
6070 IMSM_T_STATE_UNINITIALIZED
: IMSM_T_STATE_NORMAL
;
6072 switch (get_imsm_raid_level(map
)) {
6074 return IMSM_T_STATE_FAILED
;
6077 if (failed
< map
->num_members
)
6078 return IMSM_T_STATE_DEGRADED
;
6080 return IMSM_T_STATE_FAILED
;
6085 * check to see if any mirrors have failed, otherwise we
6086 * are degraded. Even numbered slots are mirrored on
6090 /* gcc -Os complains that this is unused */
6091 int insync
= insync
;
6093 for (i
= 0; i
< map
->num_members
; i
++) {
6094 __u32 ord
= get_imsm_ord_tbl_ent(dev
, i
, MAP_X
);
6095 int idx
= ord_to_idx(ord
);
6096 struct imsm_disk
*disk
;
6098 /* reset the potential in-sync count on even-numbered
6099 * slots. num_copies is always 2 for imsm raid10
6104 disk
= get_imsm_disk(super
, idx
);
6105 if (!disk
|| is_failed(disk
) || ord
& IMSM_ORD_REBUILD
)
6108 /* no in-sync disks left in this mirror the
6112 return IMSM_T_STATE_FAILED
;
6115 return IMSM_T_STATE_DEGRADED
;
6119 return IMSM_T_STATE_DEGRADED
;
6121 return IMSM_T_STATE_FAILED
;
6127 return map
->map_state
;
6130 static int imsm_count_failed(struct intel_super
*super
, struct imsm_dev
*dev
,
6135 struct imsm_disk
*disk
;
6136 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
6137 struct imsm_map
*prev
= get_imsm_map(dev
, MAP_1
);
6138 struct imsm_map
*map_for_loop
;
6143 /* at the beginning of migration we set IMSM_ORD_REBUILD on
6144 * disks that are being rebuilt. New failures are recorded to
6145 * map[0]. So we look through all the disks we started with and
6146 * see if any failures are still present, or if any new ones
6150 if (prev
&& (map
->num_members
< prev
->num_members
))
6151 map_for_loop
= prev
;
6153 for (i
= 0; i
< map_for_loop
->num_members
; i
++) {
6155 /* when MAP_X is passed both maps failures are counted
6158 ((look_in_map
== MAP_1
) || (look_in_map
== MAP_X
)) &&
6159 (i
< prev
->num_members
)) {
6160 ord
= __le32_to_cpu(prev
->disk_ord_tbl
[i
]);
6161 idx_1
= ord_to_idx(ord
);
6163 disk
= get_imsm_disk(super
, idx_1
);
6164 if (!disk
|| is_failed(disk
) || ord
& IMSM_ORD_REBUILD
)
6167 if (((look_in_map
== MAP_0
) || (look_in_map
== MAP_X
)) &&
6168 (i
< map
->num_members
)) {
6169 ord
= __le32_to_cpu(map
->disk_ord_tbl
[i
]);
6170 idx
= ord_to_idx(ord
);
6173 disk
= get_imsm_disk(super
, idx
);
6174 if (!disk
|| is_failed(disk
) ||
6175 ord
& IMSM_ORD_REBUILD
)
6185 static int imsm_open_new(struct supertype
*c
, struct active_array
*a
,
6188 struct intel_super
*super
= c
->sb
;
6189 struct imsm_super
*mpb
= super
->anchor
;
6191 if (atoi(inst
) >= mpb
->num_raid_devs
) {
6192 fprintf(stderr
, "%s: subarry index %d, out of range\n",
6193 __func__
, atoi(inst
));
6197 dprintf("imsm: open_new %s\n", inst
);
6198 a
->info
.container_member
= atoi(inst
);
6202 static int is_resyncing(struct imsm_dev
*dev
)
6204 struct imsm_map
*migr_map
;
6206 if (!dev
->vol
.migr_state
)
6209 if (migr_type(dev
) == MIGR_INIT
||
6210 migr_type(dev
) == MIGR_REPAIR
)
6213 if (migr_type(dev
) == MIGR_GEN_MIGR
)
6216 migr_map
= get_imsm_map(dev
, MAP_1
);
6218 if ((migr_map
->map_state
== IMSM_T_STATE_NORMAL
) &&
6219 (dev
->vol
.migr_type
!= MIGR_GEN_MIGR
))
6225 /* return true if we recorded new information */
6226 static int mark_failure(struct imsm_dev
*dev
, struct imsm_disk
*disk
, int idx
)
6230 struct imsm_map
*map
;
6231 char buf
[MAX_RAID_SERIAL_LEN
+3];
6232 unsigned int len
, shift
= 0;
6234 /* new failures are always set in map[0] */
6235 map
= get_imsm_map(dev
, MAP_0
);
6237 slot
= get_imsm_disk_slot(map
, idx
);
6241 ord
= __le32_to_cpu(map
->disk_ord_tbl
[slot
]);
6242 if (is_failed(disk
) && (ord
& IMSM_ORD_REBUILD
))
6245 memcpy(buf
, disk
->serial
, MAX_RAID_SERIAL_LEN
);
6246 buf
[MAX_RAID_SERIAL_LEN
] = '\000';
6248 if ((len
= strlen(buf
)) >= MAX_RAID_SERIAL_LEN
)
6249 shift
= len
- MAX_RAID_SERIAL_LEN
+ 1;
6250 strncpy((char *)disk
->serial
, &buf
[shift
], MAX_RAID_SERIAL_LEN
);
6252 disk
->status
|= FAILED_DISK
;
6253 set_imsm_ord_tbl_ent(map
, slot
, idx
| IMSM_ORD_REBUILD
);
6254 /* mark failures in second map if second map exists and this disk
6256 * This is valid for migration, initialization and rebuild
6258 if (dev
->vol
.migr_state
) {
6259 struct imsm_map
*map2
= get_imsm_map(dev
, MAP_1
);
6260 int slot2
= get_imsm_disk_slot(map2
, idx
);
6262 if ((slot2
< map2
->num_members
) &&
6264 set_imsm_ord_tbl_ent(map2
, slot2
,
6265 idx
| IMSM_ORD_REBUILD
);
6267 if (map
->failed_disk_num
== 0xff)
6268 map
->failed_disk_num
= slot
;
6272 static void mark_missing(struct imsm_dev
*dev
, struct imsm_disk
*disk
, int idx
)
6274 mark_failure(dev
, disk
, idx
);
6276 if (disk
->scsi_id
== __cpu_to_le32(~(__u32
)0))
6279 disk
->scsi_id
= __cpu_to_le32(~(__u32
)0);
6280 memmove(&disk
->serial
[0], &disk
->serial
[1], MAX_RAID_SERIAL_LEN
- 1);
6283 static void handle_missing(struct intel_super
*super
, struct imsm_dev
*dev
)
6287 if (!super
->missing
)
6290 dprintf("imsm: mark missing\n");
6291 /* end process for initialization and rebuild only
6293 if (is_gen_migration(dev
) == 0) {
6297 failed
= imsm_count_failed(super
, dev
, MAP_0
);
6298 map_state
= imsm_check_degraded(super
, dev
, failed
, MAP_0
);
6300 end_migration(dev
, super
, map_state
);
6302 for (dl
= super
->missing
; dl
; dl
= dl
->next
)
6303 mark_missing(dev
, &dl
->disk
, dl
->index
);
6304 super
->updates_pending
++;
6307 static unsigned long long imsm_set_array_size(struct imsm_dev
*dev
)
6309 int used_disks
= imsm_num_data_members(dev
, MAP_0
);
6310 unsigned long long array_blocks
;
6311 struct imsm_map
*map
;
6313 if (used_disks
== 0) {
6314 /* when problems occures
6315 * return current array_blocks value
6317 array_blocks
= __le32_to_cpu(dev
->size_high
);
6318 array_blocks
= array_blocks
<< 32;
6319 array_blocks
+= __le32_to_cpu(dev
->size_low
);
6321 return array_blocks
;
6324 /* set array size in metadata
6326 map
= get_imsm_map(dev
, MAP_0
);
6327 array_blocks
= map
->blocks_per_member
* used_disks
;
6329 /* round array size down to closest MB
6331 array_blocks
= (array_blocks
>> SECT_PER_MB_SHIFT
) << SECT_PER_MB_SHIFT
;
6332 dev
->size_low
= __cpu_to_le32((__u32
)array_blocks
);
6333 dev
->size_high
= __cpu_to_le32((__u32
)(array_blocks
>> 32));
6335 return array_blocks
;
6338 static void imsm_set_disk(struct active_array
*a
, int n
, int state
);
6340 static void imsm_progress_container_reshape(struct intel_super
*super
)
6342 /* if no device has a migr_state, but some device has a
6343 * different number of members than the previous device, start
6344 * changing the number of devices in this device to match
6347 struct imsm_super
*mpb
= super
->anchor
;
6348 int prev_disks
= -1;
6352 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
6353 struct imsm_dev
*dev
= get_imsm_dev(super
, i
);
6354 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
6355 struct imsm_map
*map2
;
6356 int prev_num_members
;
6358 if (dev
->vol
.migr_state
)
6361 if (prev_disks
== -1)
6362 prev_disks
= map
->num_members
;
6363 if (prev_disks
== map
->num_members
)
6366 /* OK, this array needs to enter reshape mode.
6367 * i.e it needs a migr_state
6370 copy_map_size
= sizeof_imsm_map(map
);
6371 prev_num_members
= map
->num_members
;
6372 map
->num_members
= prev_disks
;
6373 dev
->vol
.migr_state
= 1;
6374 dev
->vol
.curr_migr_unit
= 0;
6375 set_migr_type(dev
, MIGR_GEN_MIGR
);
6376 for (i
= prev_num_members
;
6377 i
< map
->num_members
; i
++)
6378 set_imsm_ord_tbl_ent(map
, i
, i
);
6379 map2
= get_imsm_map(dev
, MAP_1
);
6380 /* Copy the current map */
6381 memcpy(map2
, map
, copy_map_size
);
6382 map2
->num_members
= prev_num_members
;
6384 imsm_set_array_size(dev
);
6385 super
->updates_pending
++;
6389 /* Handle dirty -> clean transititions, resync and reshape. Degraded and rebuild
6390 * states are handled in imsm_set_disk() with one exception, when a
6391 * resync is stopped due to a new failure this routine will set the
6392 * 'degraded' state for the array.
6394 static int imsm_set_array_state(struct active_array
*a
, int consistent
)
6396 int inst
= a
->info
.container_member
;
6397 struct intel_super
*super
= a
->container
->sb
;
6398 struct imsm_dev
*dev
= get_imsm_dev(super
, inst
);
6399 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
6400 int failed
= imsm_count_failed(super
, dev
, MAP_0
);
6401 __u8 map_state
= imsm_check_degraded(super
, dev
, failed
, MAP_0
);
6402 __u32 blocks_per_unit
;
6404 if (dev
->vol
.migr_state
&&
6405 dev
->vol
.migr_type
== MIGR_GEN_MIGR
) {
6406 /* array state change is blocked due to reshape action
6408 * - abort the reshape (if last_checkpoint is 0 and action!= reshape)
6409 * - finish the reshape (if last_checkpoint is big and action != reshape)
6410 * - update curr_migr_unit
6412 if (a
->curr_action
== reshape
) {
6413 /* still reshaping, maybe update curr_migr_unit */
6414 goto mark_checkpoint
;
6416 if (a
->last_checkpoint
== 0 && a
->prev_action
== reshape
) {
6417 /* for some reason we aborted the reshape.
6419 * disable automatic metadata rollback
6420 * user action is required to recover process
6423 struct imsm_map
*map2
=
6424 get_imsm_map(dev
, MAP_1
);
6425 dev
->vol
.migr_state
= 0;
6426 set_migr_type(dev
, 0);
6427 dev
->vol
.curr_migr_unit
= 0;
6429 sizeof_imsm_map(map2
));
6430 super
->updates_pending
++;
6433 if (a
->last_checkpoint
>= a
->info
.component_size
) {
6434 unsigned long long array_blocks
;
6438 used_disks
= imsm_num_data_members(dev
, MAP_0
);
6439 if (used_disks
> 0) {
6441 map
->blocks_per_member
*
6443 /* round array size down to closest MB
6445 array_blocks
= (array_blocks
6446 >> SECT_PER_MB_SHIFT
)
6447 << SECT_PER_MB_SHIFT
;
6448 a
->info
.custom_array_size
= array_blocks
;
6449 /* encourage manager to update array
6453 a
->check_reshape
= 1;
6455 /* finalize online capacity expansion/reshape */
6456 for (mdi
= a
->info
.devs
; mdi
; mdi
= mdi
->next
)
6458 mdi
->disk
.raid_disk
,
6461 imsm_progress_container_reshape(super
);
6466 /* before we activate this array handle any missing disks */
6467 if (consistent
== 2)
6468 handle_missing(super
, dev
);
6470 if (consistent
== 2 &&
6471 (!is_resync_complete(&a
->info
) ||
6472 map_state
!= IMSM_T_STATE_NORMAL
||
6473 dev
->vol
.migr_state
))
6476 if (is_resync_complete(&a
->info
)) {
6477 /* complete intialization / resync,
6478 * recovery and interrupted recovery is completed in
6481 if (is_resyncing(dev
)) {
6482 dprintf("imsm: mark resync done\n");
6483 end_migration(dev
, super
, map_state
);
6484 super
->updates_pending
++;
6485 a
->last_checkpoint
= 0;
6487 } else if ((!is_resyncing(dev
) && !failed
) &&
6488 (imsm_reshape_blocks_arrays_changes(super
) == 0)) {
6489 /* mark the start of the init process if nothing is failed */
6490 dprintf("imsm: mark resync start\n");
6491 if (map
->map_state
== IMSM_T_STATE_UNINITIALIZED
)
6492 migrate(dev
, super
, IMSM_T_STATE_NORMAL
, MIGR_INIT
);
6494 migrate(dev
, super
, IMSM_T_STATE_NORMAL
, MIGR_REPAIR
);
6495 super
->updates_pending
++;
6499 /* skip checkpointing for general migration,
6500 * it is controlled in mdadm
6502 if (is_gen_migration(dev
))
6503 goto skip_mark_checkpoint
;
6505 /* check if we can update curr_migr_unit from resync_start, recovery_start */
6506 blocks_per_unit
= blocks_per_migr_unit(super
, dev
);
6507 if (blocks_per_unit
) {
6511 units
= a
->last_checkpoint
/ blocks_per_unit
;
6514 /* check that we did not overflow 32-bits, and that
6515 * curr_migr_unit needs updating
6517 if (units32
== units
&&
6519 __le32_to_cpu(dev
->vol
.curr_migr_unit
) != units32
) {
6520 dprintf("imsm: mark checkpoint (%u)\n", units32
);
6521 dev
->vol
.curr_migr_unit
= __cpu_to_le32(units32
);
6522 super
->updates_pending
++;
6526 skip_mark_checkpoint
:
6527 /* mark dirty / clean */
6528 if (dev
->vol
.dirty
!= !consistent
) {
6529 dprintf("imsm: mark '%s'\n", consistent
? "clean" : "dirty");
6534 super
->updates_pending
++;
6540 static void imsm_set_disk(struct active_array
*a
, int n
, int state
)
6542 int inst
= a
->info
.container_member
;
6543 struct intel_super
*super
= a
->container
->sb
;
6544 struct imsm_dev
*dev
= get_imsm_dev(super
, inst
);
6545 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
6546 struct imsm_disk
*disk
;
6551 if (n
> map
->num_members
)
6552 fprintf(stderr
, "imsm: set_disk %d out of range 0..%d\n",
6553 n
, map
->num_members
- 1);
6558 dprintf("imsm: set_disk %d:%x\n", n
, state
);
6560 ord
= get_imsm_ord_tbl_ent(dev
, n
, MAP_0
);
6561 disk
= get_imsm_disk(super
, ord_to_idx(ord
));
6563 /* check for new failures */
6564 if (state
& DS_FAULTY
) {
6565 if (mark_failure(dev
, disk
, ord_to_idx(ord
)))
6566 super
->updates_pending
++;
6569 /* check if in_sync */
6570 if (state
& DS_INSYNC
&& ord
& IMSM_ORD_REBUILD
&& is_rebuilding(dev
)) {
6571 struct imsm_map
*migr_map
= get_imsm_map(dev
, MAP_1
);
6573 set_imsm_ord_tbl_ent(migr_map
, n
, ord_to_idx(ord
));
6574 super
->updates_pending
++;
6577 failed
= imsm_count_failed(super
, dev
, MAP_0
);
6578 map_state
= imsm_check_degraded(super
, dev
, failed
, MAP_0
);
6580 /* check if recovery complete, newly degraded, or failed */
6581 dprintf("imsm: Detected transition to state ");
6582 switch (map_state
) {
6583 case IMSM_T_STATE_NORMAL
: /* transition to normal state */
6584 dprintf("normal: ");
6585 if (is_rebuilding(dev
)) {
6586 dprintf("while rebuilding");
6587 end_migration(dev
, super
, map_state
);
6588 map
= get_imsm_map(dev
, MAP_0
);
6589 map
->failed_disk_num
= ~0;
6590 super
->updates_pending
++;
6591 a
->last_checkpoint
= 0;
6594 if (is_gen_migration(dev
)) {
6595 dprintf("while general migration");
6596 if (a
->last_checkpoint
>= a
->info
.component_size
)
6597 end_migration(dev
, super
, map_state
);
6599 map
->map_state
= map_state
;
6600 map
= get_imsm_map(dev
, MAP_0
);
6601 map
->failed_disk_num
= ~0;
6602 super
->updates_pending
++;
6606 case IMSM_T_STATE_DEGRADED
: /* transition to degraded state */
6607 dprintf("degraded: ");
6608 if ((map
->map_state
!= map_state
) &&
6609 !dev
->vol
.migr_state
) {
6610 dprintf("mark degraded");
6611 map
->map_state
= map_state
;
6612 super
->updates_pending
++;
6613 a
->last_checkpoint
= 0;
6616 if (is_rebuilding(dev
)) {
6617 dprintf("while rebuilding.");
6618 if (map
->map_state
!= map_state
) {
6619 dprintf(" Map state change");
6620 end_migration(dev
, super
, map_state
);
6621 super
->updates_pending
++;
6625 if (is_gen_migration(dev
)) {
6626 dprintf("while general migration");
6627 if (a
->last_checkpoint
>= a
->info
.component_size
)
6628 end_migration(dev
, super
, map_state
);
6630 map
->map_state
= map_state
;
6631 manage_second_map(super
, dev
);
6633 super
->updates_pending
++;
6636 if (is_initializing(dev
)) {
6637 dprintf("while initialization.");
6638 map
->map_state
= map_state
;
6639 super
->updates_pending
++;
6643 case IMSM_T_STATE_FAILED
: /* transition to failed state */
6644 dprintf("failed: ");
6645 if (is_gen_migration(dev
)) {
6646 dprintf("while general migration");
6647 map
->map_state
= map_state
;
6648 super
->updates_pending
++;
6651 if (map
->map_state
!= map_state
) {
6652 dprintf("mark failed");
6653 end_migration(dev
, super
, map_state
);
6654 super
->updates_pending
++;
6655 a
->last_checkpoint
= 0;
6660 dprintf("state %i\n", map_state
);
6666 static int store_imsm_mpb(int fd
, struct imsm_super
*mpb
)
6669 __u32 mpb_size
= __le32_to_cpu(mpb
->mpb_size
);
6670 unsigned long long dsize
;
6671 unsigned long long sectors
;
6673 get_dev_size(fd
, NULL
, &dsize
);
6675 if (mpb_size
> 512) {
6676 /* -1 to account for anchor */
6677 sectors
= mpb_sectors(mpb
) - 1;
6679 /* write the extended mpb to the sectors preceeding the anchor */
6680 if (lseek64(fd
, dsize
- (512 * (2 + sectors
)), SEEK_SET
) < 0)
6683 if ((unsigned long long)write(fd
, buf
+ 512, 512 * sectors
)
6688 /* first block is stored on second to last sector of the disk */
6689 if (lseek64(fd
, dsize
- (512 * 2), SEEK_SET
) < 0)
6692 if (write(fd
, buf
, 512) != 512)
6698 static void imsm_sync_metadata(struct supertype
*container
)
6700 struct intel_super
*super
= container
->sb
;
6702 dprintf("sync metadata: %d\n", super
->updates_pending
);
6703 if (!super
->updates_pending
)
6706 write_super_imsm(container
, 0);
6708 super
->updates_pending
= 0;
6711 static struct dl
*imsm_readd(struct intel_super
*super
, int idx
, struct active_array
*a
)
6713 struct imsm_dev
*dev
= get_imsm_dev(super
, a
->info
.container_member
);
6714 int i
= get_imsm_disk_idx(dev
, idx
, MAP_X
);
6717 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
6721 if (dl
&& is_failed(&dl
->disk
))
6725 dprintf("%s: found %x:%x\n", __func__
, dl
->major
, dl
->minor
);
6730 static struct dl
*imsm_add_spare(struct intel_super
*super
, int slot
,
6731 struct active_array
*a
, int activate_new
,
6732 struct mdinfo
*additional_test_list
)
6734 struct imsm_dev
*dev
= get_imsm_dev(super
, a
->info
.container_member
);
6735 int idx
= get_imsm_disk_idx(dev
, slot
, MAP_X
);
6736 struct imsm_super
*mpb
= super
->anchor
;
6737 struct imsm_map
*map
;
6738 unsigned long long pos
;
6743 __u32 array_start
= 0;
6744 __u32 array_end
= 0;
6746 struct mdinfo
*test_list
;
6748 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
6749 /* If in this array, skip */
6750 for (d
= a
->info
.devs
; d
; d
= d
->next
)
6751 if (d
->state_fd
>= 0 &&
6752 d
->disk
.major
== dl
->major
&&
6753 d
->disk
.minor
== dl
->minor
) {
6754 dprintf("%x:%x already in array\n",
6755 dl
->major
, dl
->minor
);
6760 test_list
= additional_test_list
;
6762 if (test_list
->disk
.major
== dl
->major
&&
6763 test_list
->disk
.minor
== dl
->minor
) {
6764 dprintf("%x:%x already in additional test list\n",
6765 dl
->major
, dl
->minor
);
6768 test_list
= test_list
->next
;
6773 /* skip in use or failed drives */
6774 if (is_failed(&dl
->disk
) || idx
== dl
->index
||
6776 dprintf("%x:%x status (failed: %d index: %d)\n",
6777 dl
->major
, dl
->minor
, is_failed(&dl
->disk
), idx
);
6781 /* skip pure spares when we are looking for partially
6782 * assimilated drives
6784 if (dl
->index
== -1 && !activate_new
)
6787 /* Does this unused device have the requisite free space?
6788 * It needs to be able to cover all member volumes
6790 ex
= get_extents(super
, dl
);
6792 dprintf("cannot get extents\n");
6795 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
6796 dev
= get_imsm_dev(super
, i
);
6797 map
= get_imsm_map(dev
, MAP_0
);
6799 /* check if this disk is already a member of
6802 if (get_imsm_disk_slot(map
, dl
->index
) >= 0)
6808 array_start
= __le32_to_cpu(map
->pba_of_lba0
);
6809 array_end
= array_start
+
6810 __le32_to_cpu(map
->blocks_per_member
) - 1;
6813 /* check that we can start at pba_of_lba0 with
6814 * blocks_per_member of space
6816 if (array_start
>= pos
&& array_end
< ex
[j
].start
) {
6820 pos
= ex
[j
].start
+ ex
[j
].size
;
6822 } while (ex
[j
-1].size
);
6829 if (i
< mpb
->num_raid_devs
) {
6830 dprintf("%x:%x does not have %u to %u available\n",
6831 dl
->major
, dl
->minor
, array_start
, array_end
);
6842 static int imsm_rebuild_allowed(struct supertype
*cont
, int dev_idx
, int failed
)
6844 struct imsm_dev
*dev2
;
6845 struct imsm_map
*map
;
6851 dev2
= get_imsm_dev(cont
->sb
, dev_idx
);
6853 state
= imsm_check_degraded(cont
->sb
, dev2
, failed
, MAP_0
);
6854 if (state
== IMSM_T_STATE_FAILED
) {
6855 map
= get_imsm_map(dev2
, MAP_0
);
6858 for (slot
= 0; slot
< map
->num_members
; slot
++) {
6860 * Check if failed disks are deleted from intel
6861 * disk list or are marked to be deleted
6863 idx
= get_imsm_disk_idx(dev2
, slot
, MAP_X
);
6864 idisk
= get_imsm_dl_disk(cont
->sb
, idx
);
6866 * Do not rebuild the array if failed disks
6867 * from failed sub-array are not removed from
6871 is_failed(&idisk
->disk
) &&
6872 (idisk
->action
!= DISK_REMOVE
))
6880 static struct mdinfo
*imsm_activate_spare(struct active_array
*a
,
6881 struct metadata_update
**updates
)
6884 * Find a device with unused free space and use it to replace a
6885 * failed/vacant region in an array. We replace failed regions one a
6886 * array at a time. The result is that a new spare disk will be added
6887 * to the first failed array and after the monitor has finished
6888 * propagating failures the remainder will be consumed.
6890 * FIXME add a capability for mdmon to request spares from another
6894 struct intel_super
*super
= a
->container
->sb
;
6895 int inst
= a
->info
.container_member
;
6896 struct imsm_dev
*dev
= get_imsm_dev(super
, inst
);
6897 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
6898 int failed
= a
->info
.array
.raid_disks
;
6899 struct mdinfo
*rv
= NULL
;
6902 struct metadata_update
*mu
;
6904 struct imsm_update_activate_spare
*u
;
6909 for (d
= a
->info
.devs
; d
; d
= d
->next
) {
6910 if ((d
->curr_state
& DS_FAULTY
) &&
6912 /* wait for Removal to happen */
6914 if (d
->state_fd
>= 0)
6918 dprintf("imsm: activate spare: inst=%d failed=%d (%d) level=%d\n",
6919 inst
, failed
, a
->info
.array
.raid_disks
, a
->info
.array
.level
);
6921 if (imsm_reshape_blocks_arrays_changes(super
))
6924 /* Cannot activate another spare if rebuild is in progress already
6926 if (is_rebuilding(dev
)) {
6927 dprintf("imsm: No spare activation allowed. "
6928 "Rebuild in progress already.\n");
6932 if (a
->info
.array
.level
== 4)
6933 /* No repair for takeovered array
6934 * imsm doesn't support raid4
6938 if (imsm_check_degraded(super
, dev
, failed
, MAP_0
) !=
6939 IMSM_T_STATE_DEGRADED
)
6943 * If there are any failed disks check state of the other volume.
6944 * Block rebuild if the another one is failed until failed disks
6945 * are removed from container.
6948 dprintf("found failed disks in %.*s, check if there another"
6949 "failed sub-array.\n",
6950 MAX_RAID_SERIAL_LEN
, dev
->volume
);
6951 /* check if states of the other volumes allow for rebuild */
6952 for (i
= 0; i
< super
->anchor
->num_raid_devs
; i
++) {
6954 allowed
= imsm_rebuild_allowed(a
->container
,
6962 /* For each slot, if it is not working, find a spare */
6963 for (i
= 0; i
< a
->info
.array
.raid_disks
; i
++) {
6964 for (d
= a
->info
.devs
; d
; d
= d
->next
)
6965 if (d
->disk
.raid_disk
== i
)
6967 dprintf("found %d: %p %x\n", i
, d
, d
?d
->curr_state
:0);
6968 if (d
&& (d
->state_fd
>= 0))
6972 * OK, this device needs recovery. Try to re-add the
6973 * previous occupant of this slot, if this fails see if
6974 * we can continue the assimilation of a spare that was
6975 * partially assimilated, finally try to activate a new
6978 dl
= imsm_readd(super
, i
, a
);
6980 dl
= imsm_add_spare(super
, i
, a
, 0, rv
);
6982 dl
= imsm_add_spare(super
, i
, a
, 1, rv
);
6986 /* found a usable disk with enough space */
6987 di
= malloc(sizeof(*di
));
6990 memset(di
, 0, sizeof(*di
));
6992 /* dl->index will be -1 in the case we are activating a
6993 * pristine spare. imsm_process_update() will create a
6994 * new index in this case. Once a disk is found to be
6995 * failed in all member arrays it is kicked from the
6998 di
->disk
.number
= dl
->index
;
7000 /* (ab)use di->devs to store a pointer to the device
7003 di
->devs
= (struct mdinfo
*) dl
;
7005 di
->disk
.raid_disk
= i
;
7006 di
->disk
.major
= dl
->major
;
7007 di
->disk
.minor
= dl
->minor
;
7009 di
->recovery_start
= 0;
7010 di
->data_offset
= __le32_to_cpu(map
->pba_of_lba0
);
7011 di
->component_size
= a
->info
.component_size
;
7012 di
->container_member
= inst
;
7013 super
->random
= random32();
7017 dprintf("%x:%x to be %d at %llu\n", dl
->major
, dl
->minor
,
7018 i
, di
->data_offset
);
7022 /* No spares found */
7024 /* Now 'rv' has a list of devices to return.
7025 * Create a metadata_update record to update the
7026 * disk_ord_tbl for the array
7028 mu
= malloc(sizeof(*mu
));
7030 mu
->buf
= malloc(sizeof(struct imsm_update_activate_spare
) * num_spares
);
7031 if (mu
->buf
== NULL
) {
7038 struct mdinfo
*n
= rv
->next
;
7047 mu
->space_list
= NULL
;
7048 mu
->len
= sizeof(struct imsm_update_activate_spare
) * num_spares
;
7049 mu
->next
= *updates
;
7050 u
= (struct imsm_update_activate_spare
*) mu
->buf
;
7052 for (di
= rv
; di
; di
= di
->next
) {
7053 u
->type
= update_activate_spare
;
7054 u
->dl
= (struct dl
*) di
->devs
;
7056 u
->slot
= di
->disk
.raid_disk
;
7067 static int disks_overlap(struct intel_super
*super
, int idx
, struct imsm_update_create_array
*u
)
7069 struct imsm_dev
*dev
= get_imsm_dev(super
, idx
);
7070 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
7071 struct imsm_map
*new_map
= get_imsm_map(&u
->dev
, MAP_0
);
7072 struct disk_info
*inf
= get_disk_info(u
);
7073 struct imsm_disk
*disk
;
7077 for (i
= 0; i
< map
->num_members
; i
++) {
7078 disk
= get_imsm_disk(super
, get_imsm_disk_idx(dev
, i
, MAP_X
));
7079 for (j
= 0; j
< new_map
->num_members
; j
++)
7080 if (serialcmp(disk
->serial
, inf
[j
].serial
) == 0)
7088 static struct dl
*get_disk_super(struct intel_super
*super
, int major
, int minor
)
7090 struct dl
*dl
= NULL
;
7091 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
7092 if ((dl
->major
== major
) && (dl
->minor
== minor
))
7097 static int remove_disk_super(struct intel_super
*super
, int major
, int minor
)
7099 struct dl
*prev
= NULL
;
7103 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
7104 if ((dl
->major
== major
) && (dl
->minor
== minor
)) {
7107 prev
->next
= dl
->next
;
7109 super
->disks
= dl
->next
;
7111 __free_imsm_disk(dl
);
7112 dprintf("%s: removed %x:%x\n",
7113 __func__
, major
, minor
);
7121 static void imsm_delete(struct intel_super
*super
, struct dl
**dlp
, unsigned index
);
7123 static int add_remove_disk_update(struct intel_super
*super
)
7125 int check_degraded
= 0;
7126 struct dl
*disk
= NULL
;
7127 /* add/remove some spares to/from the metadata/contrainer */
7128 while (super
->disk_mgmt_list
) {
7129 struct dl
*disk_cfg
;
7131 disk_cfg
= super
->disk_mgmt_list
;
7132 super
->disk_mgmt_list
= disk_cfg
->next
;
7133 disk_cfg
->next
= NULL
;
7135 if (disk_cfg
->action
== DISK_ADD
) {
7136 disk_cfg
->next
= super
->disks
;
7137 super
->disks
= disk_cfg
;
7139 dprintf("%s: added %x:%x\n",
7140 __func__
, disk_cfg
->major
,
7142 } else if (disk_cfg
->action
== DISK_REMOVE
) {
7143 dprintf("Disk remove action processed: %x.%x\n",
7144 disk_cfg
->major
, disk_cfg
->minor
);
7145 disk
= get_disk_super(super
,
7149 /* store action status */
7150 disk
->action
= DISK_REMOVE
;
7151 /* remove spare disks only */
7152 if (disk
->index
== -1) {
7153 remove_disk_super(super
,
7158 /* release allocate disk structure */
7159 __free_imsm_disk(disk_cfg
);
7162 return check_degraded
;
7166 static int apply_reshape_migration_update(struct imsm_update_reshape_migration
*u
,
7167 struct intel_super
*super
,
7170 struct intel_dev
*id
;
7171 void **tofree
= NULL
;
7174 dprintf("apply_reshape_migration_update()\n");
7175 if ((u
->subdev
< 0) ||
7177 dprintf("imsm: Error: Wrong subdev: %i\n", u
->subdev
);
7180 if ((space_list
== NULL
) || (*space_list
== NULL
)) {
7181 dprintf("imsm: Error: Memory is not allocated\n");
7185 for (id
= super
->devlist
; id
; id
= id
->next
) {
7186 if (id
->index
== (unsigned)u
->subdev
) {
7187 struct imsm_dev
*dev
= get_imsm_dev(super
, u
->subdev
);
7188 struct imsm_map
*map
;
7189 struct imsm_dev
*new_dev
=
7190 (struct imsm_dev
*)*space_list
;
7191 struct imsm_map
*migr_map
= get_imsm_map(dev
, MAP_1
);
7193 struct dl
*new_disk
;
7195 if (new_dev
== NULL
)
7197 *space_list
= **space_list
;
7198 memcpy(new_dev
, dev
, sizeof_imsm_dev(dev
, 0));
7199 map
= get_imsm_map(new_dev
, MAP_0
);
7201 dprintf("imsm: Error: migration in progress");
7205 to_state
= map
->map_state
;
7206 if ((u
->new_level
== 5) && (map
->raid_level
== 0)) {
7208 /* this should not happen */
7209 if (u
->new_disks
[0] < 0) {
7210 map
->failed_disk_num
=
7211 map
->num_members
- 1;
7212 to_state
= IMSM_T_STATE_DEGRADED
;
7214 to_state
= IMSM_T_STATE_NORMAL
;
7216 migrate(new_dev
, super
, to_state
, MIGR_GEN_MIGR
);
7217 if (u
->new_level
> -1)
7218 map
->raid_level
= u
->new_level
;
7219 migr_map
= get_imsm_map(new_dev
, MAP_1
);
7220 if ((u
->new_level
== 5) &&
7221 (migr_map
->raid_level
== 0)) {
7222 int ord
= map
->num_members
- 1;
7223 migr_map
->num_members
--;
7224 if (u
->new_disks
[0] < 0)
7225 ord
|= IMSM_ORD_REBUILD
;
7226 set_imsm_ord_tbl_ent(map
,
7227 map
->num_members
- 1,
7231 tofree
= (void **)dev
;
7233 /* update chunk size
7235 if (u
->new_chunksize
> 0)
7236 map
->blocks_per_strip
=
7237 __cpu_to_le16(u
->new_chunksize
* 2);
7241 if ((u
->new_level
!= 5) ||
7242 (migr_map
->raid_level
!= 0) ||
7243 (migr_map
->raid_level
== map
->raid_level
))
7246 if (u
->new_disks
[0] >= 0) {
7249 new_disk
= get_disk_super(super
,
7250 major(u
->new_disks
[0]),
7251 minor(u
->new_disks
[0]));
7252 dprintf("imsm: new disk for reshape is: %i:%i "
7253 "(%p, index = %i)\n",
7254 major(u
->new_disks
[0]),
7255 minor(u
->new_disks
[0]),
7256 new_disk
, new_disk
->index
);
7257 if (new_disk
== NULL
)
7258 goto error_disk_add
;
7260 new_disk
->index
= map
->num_members
- 1;
7261 /* slot to fill in autolayout
7263 new_disk
->raiddisk
= new_disk
->index
;
7264 new_disk
->disk
.status
|= CONFIGURED_DISK
;
7265 new_disk
->disk
.status
&= ~SPARE_DISK
;
7267 goto error_disk_add
;
7270 *tofree
= *space_list
;
7271 /* calculate new size
7273 imsm_set_array_size(new_dev
);
7280 *space_list
= tofree
;
7284 dprintf("Error: imsm: Cannot find disk.\n");
7288 static int apply_update_activate_spare(struct imsm_update_activate_spare
*u
,
7289 struct intel_super
*super
,
7290 struct active_array
*active_array
)
7292 struct imsm_super
*mpb
= super
->anchor
;
7293 struct imsm_dev
*dev
= get_imsm_dev(super
, u
->array
);
7294 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
7295 struct imsm_map
*migr_map
;
7296 struct active_array
*a
;
7297 struct imsm_disk
*disk
;
7304 int second_map_created
= 0;
7306 for (; u
; u
= u
->next
) {
7307 victim
= get_imsm_disk_idx(dev
, u
->slot
, MAP_X
);
7312 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
7317 fprintf(stderr
, "error: imsm_activate_spare passed "
7318 "an unknown disk (index: %d)\n",
7323 /* count failures (excluding rebuilds and the victim)
7324 * to determine map[0] state
7327 for (i
= 0; i
< map
->num_members
; i
++) {
7330 disk
= get_imsm_disk(super
,
7331 get_imsm_disk_idx(dev
, i
, MAP_X
));
7332 if (!disk
|| is_failed(disk
))
7336 /* adding a pristine spare, assign a new index */
7337 if (dl
->index
< 0) {
7338 dl
->index
= super
->anchor
->num_disks
;
7339 super
->anchor
->num_disks
++;
7342 disk
->status
|= CONFIGURED_DISK
;
7343 disk
->status
&= ~SPARE_DISK
;
7346 to_state
= imsm_check_degraded(super
, dev
, failed
, MAP_0
);
7347 if (!second_map_created
) {
7348 second_map_created
= 1;
7349 map
->map_state
= IMSM_T_STATE_DEGRADED
;
7350 migrate(dev
, super
, to_state
, MIGR_REBUILD
);
7352 map
->map_state
= to_state
;
7353 migr_map
= get_imsm_map(dev
, MAP_1
);
7354 set_imsm_ord_tbl_ent(map
, u
->slot
, dl
->index
);
7355 set_imsm_ord_tbl_ent(migr_map
, u
->slot
,
7356 dl
->index
| IMSM_ORD_REBUILD
);
7358 /* update the family_num to mark a new container
7359 * generation, being careful to record the existing
7360 * family_num in orig_family_num to clean up after
7361 * earlier mdadm versions that neglected to set it.
7363 if (mpb
->orig_family_num
== 0)
7364 mpb
->orig_family_num
= mpb
->family_num
;
7365 mpb
->family_num
+= super
->random
;
7367 /* count arrays using the victim in the metadata */
7369 for (a
= active_array
; a
; a
= a
->next
) {
7370 dev
= get_imsm_dev(super
, a
->info
.container_member
);
7371 map
= get_imsm_map(dev
, MAP_0
);
7373 if (get_imsm_disk_slot(map
, victim
) >= 0)
7377 /* delete the victim if it is no longer being
7383 /* We know that 'manager' isn't touching anything,
7384 * so it is safe to delete
7386 for (dlp
= &super
->disks
; *dlp
; dlp
= &(*dlp
)->next
)
7387 if ((*dlp
)->index
== victim
)
7390 /* victim may be on the missing list */
7392 for (dlp
= &super
->missing
; *dlp
;
7393 dlp
= &(*dlp
)->next
)
7394 if ((*dlp
)->index
== victim
)
7396 imsm_delete(super
, dlp
, victim
);
7403 static int apply_reshape_container_disks_update(struct imsm_update_reshape
*u
,
7404 struct intel_super
*super
,
7407 struct dl
*new_disk
;
7408 struct intel_dev
*id
;
7410 int delta_disks
= u
->new_raid_disks
- u
->old_raid_disks
;
7411 int disk_count
= u
->old_raid_disks
;
7412 void **tofree
= NULL
;
7413 int devices_to_reshape
= 1;
7414 struct imsm_super
*mpb
= super
->anchor
;
7416 unsigned int dev_id
;
7418 dprintf("imsm: apply_reshape_container_disks_update()\n");
7420 /* enable spares to use in array */
7421 for (i
= 0; i
< delta_disks
; i
++) {
7422 new_disk
= get_disk_super(super
,
7423 major(u
->new_disks
[i
]),
7424 minor(u
->new_disks
[i
]));
7425 dprintf("imsm: new disk for reshape is: %i:%i "
7426 "(%p, index = %i)\n",
7427 major(u
->new_disks
[i
]), minor(u
->new_disks
[i
]),
7428 new_disk
, new_disk
->index
);
7429 if ((new_disk
== NULL
) ||
7430 ((new_disk
->index
>= 0) &&
7431 (new_disk
->index
< u
->old_raid_disks
)))
7432 goto update_reshape_exit
;
7433 new_disk
->index
= disk_count
++;
7434 /* slot to fill in autolayout
7436 new_disk
->raiddisk
= new_disk
->index
;
7437 new_disk
->disk
.status
|=
7439 new_disk
->disk
.status
&= ~SPARE_DISK
;
7442 dprintf("imsm: volume set mpb->num_raid_devs = %i\n",
7443 mpb
->num_raid_devs
);
7444 /* manage changes in volume
7446 for (dev_id
= 0; dev_id
< mpb
->num_raid_devs
; dev_id
++) {
7447 void **sp
= *space_list
;
7448 struct imsm_dev
*newdev
;
7449 struct imsm_map
*newmap
, *oldmap
;
7451 for (id
= super
->devlist
; id
; id
= id
->next
) {
7452 if (id
->index
== dev_id
)
7461 /* Copy the dev, but not (all of) the map */
7462 memcpy(newdev
, id
->dev
, sizeof(*newdev
));
7463 oldmap
= get_imsm_map(id
->dev
, MAP_0
);
7464 newmap
= get_imsm_map(newdev
, MAP_0
);
7465 /* Copy the current map */
7466 memcpy(newmap
, oldmap
, sizeof_imsm_map(oldmap
));
7467 /* update one device only
7469 if (devices_to_reshape
) {
7470 dprintf("imsm: modifying subdev: %i\n",
7472 devices_to_reshape
--;
7473 newdev
->vol
.migr_state
= 1;
7474 newdev
->vol
.curr_migr_unit
= 0;
7475 set_migr_type(newdev
, MIGR_GEN_MIGR
);
7476 newmap
->num_members
= u
->new_raid_disks
;
7477 for (i
= 0; i
< delta_disks
; i
++) {
7478 set_imsm_ord_tbl_ent(newmap
,
7479 u
->old_raid_disks
+ i
,
7480 u
->old_raid_disks
+ i
);
7482 /* New map is correct, now need to save old map
7484 newmap
= get_imsm_map(newdev
, MAP_1
);
7485 memcpy(newmap
, oldmap
, sizeof_imsm_map(oldmap
));
7487 imsm_set_array_size(newdev
);
7490 sp
= (void **)id
->dev
;
7495 /* Clear migration record */
7496 memset(super
->migr_rec
, 0, sizeof(struct migr_record
));
7499 *space_list
= tofree
;
7502 update_reshape_exit
:
7507 static int apply_takeover_update(struct imsm_update_takeover
*u
,
7508 struct intel_super
*super
,
7511 struct imsm_dev
*dev
= NULL
;
7512 struct intel_dev
*dv
;
7513 struct imsm_dev
*dev_new
;
7514 struct imsm_map
*map
;
7518 for (dv
= super
->devlist
; dv
; dv
= dv
->next
)
7519 if (dv
->index
== (unsigned int)u
->subarray
) {
7527 map
= get_imsm_map(dev
, MAP_0
);
7529 if (u
->direction
== R10_TO_R0
) {
7530 /* Number of failed disks must be half of initial disk number */
7531 if (imsm_count_failed(super
, dev
, MAP_0
) !=
7532 (map
->num_members
/ 2))
7535 /* iterate through devices to mark removed disks as spare */
7536 for (dm
= super
->disks
; dm
; dm
= dm
->next
) {
7537 if (dm
->disk
.status
& FAILED_DISK
) {
7538 int idx
= dm
->index
;
7539 /* update indexes on the disk list */
7540 /* FIXME this loop-with-the-loop looks wrong, I'm not convinced
7541 the index values will end up being correct.... NB */
7542 for (du
= super
->disks
; du
; du
= du
->next
)
7543 if (du
->index
> idx
)
7545 /* mark as spare disk */
7550 map
->num_members
= map
->num_members
/ 2;
7551 map
->map_state
= IMSM_T_STATE_NORMAL
;
7552 map
->num_domains
= 1;
7553 map
->raid_level
= 0;
7554 map
->failed_disk_num
= -1;
7557 if (u
->direction
== R0_TO_R10
) {
7559 /* update slots in current disk list */
7560 for (dm
= super
->disks
; dm
; dm
= dm
->next
) {
7564 /* create new *missing* disks */
7565 for (i
= 0; i
< map
->num_members
; i
++) {
7566 space
= *space_list
;
7569 *space_list
= *space
;
7571 memcpy(du
, super
->disks
, sizeof(*du
));
7575 du
->index
= (i
* 2) + 1;
7576 sprintf((char *)du
->disk
.serial
,
7577 " MISSING_%d", du
->index
);
7578 sprintf((char *)du
->serial
,
7579 "MISSING_%d", du
->index
);
7580 du
->next
= super
->missing
;
7581 super
->missing
= du
;
7583 /* create new dev and map */
7584 space
= *space_list
;
7587 *space_list
= *space
;
7588 dev_new
= (void *)space
;
7589 memcpy(dev_new
, dev
, sizeof(*dev
));
7590 /* update new map */
7591 map
= get_imsm_map(dev_new
, MAP_0
);
7592 map
->num_members
= map
->num_members
* 2;
7593 map
->map_state
= IMSM_T_STATE_DEGRADED
;
7594 map
->num_domains
= 2;
7595 map
->raid_level
= 1;
7596 /* replace dev<->dev_new */
7599 /* update disk order table */
7600 for (du
= super
->disks
; du
; du
= du
->next
)
7602 set_imsm_ord_tbl_ent(map
, du
->index
, du
->index
);
7603 for (du
= super
->missing
; du
; du
= du
->next
)
7604 if (du
->index
>= 0) {
7605 set_imsm_ord_tbl_ent(map
, du
->index
, du
->index
);
7606 mark_missing(dv
->dev
, &du
->disk
, du
->index
);
7612 static void imsm_process_update(struct supertype
*st
,
7613 struct metadata_update
*update
)
7616 * crack open the metadata_update envelope to find the update record
7617 * update can be one of:
7618 * update_reshape_container_disks - all the arrays in the container
7619 * are being reshaped to have more devices. We need to mark
7620 * the arrays for general migration and convert selected spares
7621 * into active devices.
7622 * update_activate_spare - a spare device has replaced a failed
7623 * device in an array, update the disk_ord_tbl. If this disk is
7624 * present in all member arrays then also clear the SPARE_DISK
7626 * update_create_array
7628 * update_rename_array
7629 * update_add_remove_disk
7631 struct intel_super
*super
= st
->sb
;
7632 struct imsm_super
*mpb
;
7633 enum imsm_update_type type
= *(enum imsm_update_type
*) update
->buf
;
7635 /* update requires a larger buf but the allocation failed */
7636 if (super
->next_len
&& !super
->next_buf
) {
7637 super
->next_len
= 0;
7641 if (super
->next_buf
) {
7642 memcpy(super
->next_buf
, super
->buf
, super
->len
);
7644 super
->len
= super
->next_len
;
7645 super
->buf
= super
->next_buf
;
7647 super
->next_len
= 0;
7648 super
->next_buf
= NULL
;
7651 mpb
= super
->anchor
;
7654 case update_general_migration_checkpoint
: {
7655 struct intel_dev
*id
;
7656 struct imsm_update_general_migration_checkpoint
*u
=
7657 (void *)update
->buf
;
7659 dprintf("imsm: process_update() "
7660 "for update_general_migration_checkpoint called\n");
7662 /* find device under general migration */
7663 for (id
= super
->devlist
; id
; id
= id
->next
) {
7664 if (is_gen_migration(id
->dev
)) {
7665 id
->dev
->vol
.curr_migr_unit
=
7666 __cpu_to_le32(u
->curr_migr_unit
);
7667 super
->updates_pending
++;
7672 case update_takeover
: {
7673 struct imsm_update_takeover
*u
= (void *)update
->buf
;
7674 if (apply_takeover_update(u
, super
, &update
->space_list
)) {
7675 imsm_update_version_info(super
);
7676 super
->updates_pending
++;
7681 case update_reshape_container_disks
: {
7682 struct imsm_update_reshape
*u
= (void *)update
->buf
;
7683 if (apply_reshape_container_disks_update(
7684 u
, super
, &update
->space_list
))
7685 super
->updates_pending
++;
7688 case update_reshape_migration
: {
7689 struct imsm_update_reshape_migration
*u
= (void *)update
->buf
;
7690 if (apply_reshape_migration_update(
7691 u
, super
, &update
->space_list
))
7692 super
->updates_pending
++;
7695 case update_activate_spare
: {
7696 struct imsm_update_activate_spare
*u
= (void *) update
->buf
;
7697 if (apply_update_activate_spare(u
, super
, st
->arrays
))
7698 super
->updates_pending
++;
7701 case update_create_array
: {
7702 /* someone wants to create a new array, we need to be aware of
7703 * a few races/collisions:
7704 * 1/ 'Create' called by two separate instances of mdadm
7705 * 2/ 'Create' versus 'activate_spare': mdadm has chosen
7706 * devices that have since been assimilated via
7708 * In the event this update can not be carried out mdadm will
7709 * (FIX ME) notice that its update did not take hold.
7711 struct imsm_update_create_array
*u
= (void *) update
->buf
;
7712 struct intel_dev
*dv
;
7713 struct imsm_dev
*dev
;
7714 struct imsm_map
*map
, *new_map
;
7715 unsigned long long start
, end
;
7716 unsigned long long new_start
, new_end
;
7718 struct disk_info
*inf
;
7721 /* handle racing creates: first come first serve */
7722 if (u
->dev_idx
< mpb
->num_raid_devs
) {
7723 dprintf("%s: subarray %d already defined\n",
7724 __func__
, u
->dev_idx
);
7728 /* check update is next in sequence */
7729 if (u
->dev_idx
!= mpb
->num_raid_devs
) {
7730 dprintf("%s: can not create array %d expected index %d\n",
7731 __func__
, u
->dev_idx
, mpb
->num_raid_devs
);
7735 new_map
= get_imsm_map(&u
->dev
, MAP_0
);
7736 new_start
= __le32_to_cpu(new_map
->pba_of_lba0
);
7737 new_end
= new_start
+ __le32_to_cpu(new_map
->blocks_per_member
);
7738 inf
= get_disk_info(u
);
7740 /* handle activate_spare versus create race:
7741 * check to make sure that overlapping arrays do not include
7744 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
7745 dev
= get_imsm_dev(super
, i
);
7746 map
= get_imsm_map(dev
, MAP_0
);
7747 start
= __le32_to_cpu(map
->pba_of_lba0
);
7748 end
= start
+ __le32_to_cpu(map
->blocks_per_member
);
7749 if ((new_start
>= start
&& new_start
<= end
) ||
7750 (start
>= new_start
&& start
<= new_end
))
7755 if (disks_overlap(super
, i
, u
)) {
7756 dprintf("%s: arrays overlap\n", __func__
);
7761 /* check that prepare update was successful */
7762 if (!update
->space
) {
7763 dprintf("%s: prepare update failed\n", __func__
);
7767 /* check that all disks are still active before committing
7768 * changes. FIXME: could we instead handle this by creating a
7769 * degraded array? That's probably not what the user expects,
7770 * so better to drop this update on the floor.
7772 for (i
= 0; i
< new_map
->num_members
; i
++) {
7773 dl
= serial_to_dl(inf
[i
].serial
, super
);
7775 dprintf("%s: disk disappeared\n", __func__
);
7780 super
->updates_pending
++;
7782 /* convert spares to members and fixup ord_tbl */
7783 for (i
= 0; i
< new_map
->num_members
; i
++) {
7784 dl
= serial_to_dl(inf
[i
].serial
, super
);
7785 if (dl
->index
== -1) {
7786 dl
->index
= mpb
->num_disks
;
7788 dl
->disk
.status
|= CONFIGURED_DISK
;
7789 dl
->disk
.status
&= ~SPARE_DISK
;
7791 set_imsm_ord_tbl_ent(new_map
, i
, dl
->index
);
7796 update
->space
= NULL
;
7797 imsm_copy_dev(dev
, &u
->dev
);
7798 dv
->index
= u
->dev_idx
;
7799 dv
->next
= super
->devlist
;
7800 super
->devlist
= dv
;
7801 mpb
->num_raid_devs
++;
7803 imsm_update_version_info(super
);
7806 /* mdmon knows how to release update->space, but not
7807 * ((struct intel_dev *) update->space)->dev
7809 if (update
->space
) {
7815 case update_kill_array
: {
7816 struct imsm_update_kill_array
*u
= (void *) update
->buf
;
7817 int victim
= u
->dev_idx
;
7818 struct active_array
*a
;
7819 struct intel_dev
**dp
;
7820 struct imsm_dev
*dev
;
7822 /* sanity check that we are not affecting the uuid of
7823 * active arrays, or deleting an active array
7825 * FIXME when immutable ids are available, but note that
7826 * we'll also need to fixup the invalidated/active
7827 * subarray indexes in mdstat
7829 for (a
= st
->arrays
; a
; a
= a
->next
)
7830 if (a
->info
.container_member
>= victim
)
7832 /* by definition if mdmon is running at least one array
7833 * is active in the container, so checking
7834 * mpb->num_raid_devs is just extra paranoia
7836 dev
= get_imsm_dev(super
, victim
);
7837 if (a
|| !dev
|| mpb
->num_raid_devs
== 1) {
7838 dprintf("failed to delete subarray-%d\n", victim
);
7842 for (dp
= &super
->devlist
; *dp
;)
7843 if ((*dp
)->index
== (unsigned)super
->current_vol
) {
7846 if ((*dp
)->index
> (unsigned)victim
)
7850 mpb
->num_raid_devs
--;
7851 super
->updates_pending
++;
7854 case update_rename_array
: {
7855 struct imsm_update_rename_array
*u
= (void *) update
->buf
;
7856 char name
[MAX_RAID_SERIAL_LEN
+1];
7857 int target
= u
->dev_idx
;
7858 struct active_array
*a
;
7859 struct imsm_dev
*dev
;
7861 /* sanity check that we are not affecting the uuid of
7864 snprintf(name
, MAX_RAID_SERIAL_LEN
, "%s", (char *) u
->name
);
7865 name
[MAX_RAID_SERIAL_LEN
] = '\0';
7866 for (a
= st
->arrays
; a
; a
= a
->next
)
7867 if (a
->info
.container_member
== target
)
7869 dev
= get_imsm_dev(super
, u
->dev_idx
);
7870 if (a
|| !dev
|| !check_name(super
, name
, 1)) {
7871 dprintf("failed to rename subarray-%d\n", target
);
7875 snprintf((char *) dev
->volume
, MAX_RAID_SERIAL_LEN
, "%s", name
);
7876 super
->updates_pending
++;
7879 case update_add_remove_disk
: {
7880 /* we may be able to repair some arrays if disks are
7881 * being added, check teh status of add_remove_disk
7882 * if discs has been added.
7884 if (add_remove_disk_update(super
)) {
7885 struct active_array
*a
;
7887 super
->updates_pending
++;
7888 for (a
= st
->arrays
; a
; a
= a
->next
)
7889 a
->check_degraded
= 1;
7894 fprintf(stderr
, "error: unsuported process update type:"
7895 "(type: %d)\n", type
);
7899 static struct mdinfo
*get_spares_for_grow(struct supertype
*st
);
7901 static void imsm_prepare_update(struct supertype
*st
,
7902 struct metadata_update
*update
)
7905 * Allocate space to hold new disk entries, raid-device entries or a new
7906 * mpb if necessary. The manager synchronously waits for updates to
7907 * complete in the monitor, so new mpb buffers allocated here can be
7908 * integrated by the monitor thread without worrying about live pointers
7909 * in the manager thread.
7911 enum imsm_update_type type
= *(enum imsm_update_type
*) update
->buf
;
7912 struct intel_super
*super
= st
->sb
;
7913 struct imsm_super
*mpb
= super
->anchor
;
7918 case update_general_migration_checkpoint
:
7919 dprintf("imsm: prepare_update() "
7920 "for update_general_migration_checkpoint called\n");
7922 case update_takeover
: {
7923 struct imsm_update_takeover
*u
= (void *)update
->buf
;
7924 if (u
->direction
== R0_TO_R10
) {
7925 void **tail
= (void **)&update
->space_list
;
7926 struct imsm_dev
*dev
= get_imsm_dev(super
, u
->subarray
);
7927 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
7928 int num_members
= map
->num_members
;
7932 /* allocate memory for added disks */
7933 for (i
= 0; i
< num_members
; i
++) {
7934 size
= sizeof(struct dl
);
7935 space
= malloc(size
);
7944 /* allocate memory for new device */
7945 size
= sizeof_imsm_dev(super
->devlist
->dev
, 0) +
7946 (num_members
* sizeof(__u32
));
7947 space
= malloc(size
);
7956 len
= disks_to_mpb_size(num_members
* 2);
7958 /* if allocation didn't success, free buffer */
7959 while (update
->space_list
) {
7960 void **sp
= update
->space_list
;
7961 update
->space_list
= *sp
;
7969 case update_reshape_container_disks
: {
7970 /* Every raid device in the container is about to
7971 * gain some more devices, and we will enter a
7973 * So each 'imsm_map' will be bigger, and the imsm_vol
7974 * will now hold 2 of them.
7975 * Thus we need new 'struct imsm_dev' allocations sized
7976 * as sizeof_imsm_dev but with more devices in both maps.
7978 struct imsm_update_reshape
*u
= (void *)update
->buf
;
7979 struct intel_dev
*dl
;
7980 void **space_tail
= (void**)&update
->space_list
;
7982 dprintf("imsm: imsm_prepare_update() for update_reshape\n");
7984 for (dl
= super
->devlist
; dl
; dl
= dl
->next
) {
7985 int size
= sizeof_imsm_dev(dl
->dev
, 1);
7987 if (u
->new_raid_disks
> u
->old_raid_disks
)
7988 size
+= sizeof(__u32
)*2*
7989 (u
->new_raid_disks
- u
->old_raid_disks
);
7998 len
= disks_to_mpb_size(u
->new_raid_disks
);
7999 dprintf("New anchor length is %llu\n", (unsigned long long)len
);
8002 case update_reshape_migration
: {
8003 /* for migration level 0->5 we need to add disks
8004 * so the same as for container operation we will copy
8005 * device to the bigger location.
8006 * in memory prepared device and new disk area are prepared
8007 * for usage in process update
8009 struct imsm_update_reshape_migration
*u
= (void *)update
->buf
;
8010 struct intel_dev
*id
;
8011 void **space_tail
= (void **)&update
->space_list
;
8014 int current_level
= -1;
8016 dprintf("imsm: imsm_prepare_update() for update_reshape\n");
8018 /* add space for bigger array in update
8020 for (id
= super
->devlist
; id
; id
= id
->next
) {
8021 if (id
->index
== (unsigned)u
->subdev
) {
8022 size
= sizeof_imsm_dev(id
->dev
, 1);
8023 if (u
->new_raid_disks
> u
->old_raid_disks
)
8024 size
+= sizeof(__u32
)*2*
8025 (u
->new_raid_disks
- u
->old_raid_disks
);
8035 if (update
->space_list
== NULL
)
8038 /* add space for disk in update
8040 size
= sizeof(struct dl
);
8043 free(update
->space_list
);
8044 update
->space_list
= NULL
;
8051 /* add spare device to update
8053 for (id
= super
->devlist
; id
; id
= id
->next
)
8054 if (id
->index
== (unsigned)u
->subdev
) {
8055 struct imsm_dev
*dev
;
8056 struct imsm_map
*map
;
8058 dev
= get_imsm_dev(super
, u
->subdev
);
8059 map
= get_imsm_map(dev
, MAP_0
);
8060 current_level
= map
->raid_level
;
8063 if ((u
->new_level
== 5) && (u
->new_level
!= current_level
)) {
8064 struct mdinfo
*spares
;
8066 spares
= get_spares_for_grow(st
);
8074 makedev(dev
->disk
.major
,
8076 dl
= get_disk_super(super
,
8079 dl
->index
= u
->old_raid_disks
;
8085 len
= disks_to_mpb_size(u
->new_raid_disks
);
8086 dprintf("New anchor length is %llu\n", (unsigned long long)len
);
8089 case update_create_array
: {
8090 struct imsm_update_create_array
*u
= (void *) update
->buf
;
8091 struct intel_dev
*dv
;
8092 struct imsm_dev
*dev
= &u
->dev
;
8093 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
8095 struct disk_info
*inf
;
8099 inf
= get_disk_info(u
);
8100 len
= sizeof_imsm_dev(dev
, 1);
8101 /* allocate a new super->devlist entry */
8102 dv
= malloc(sizeof(*dv
));
8104 dv
->dev
= malloc(len
);
8109 update
->space
= NULL
;
8113 /* count how many spares will be converted to members */
8114 for (i
= 0; i
< map
->num_members
; i
++) {
8115 dl
= serial_to_dl(inf
[i
].serial
, super
);
8117 /* hmm maybe it failed?, nothing we can do about
8122 if (count_memberships(dl
, super
) == 0)
8125 len
+= activate
* sizeof(struct imsm_disk
);
8132 /* check if we need a larger metadata buffer */
8133 if (super
->next_buf
)
8134 buf_len
= super
->next_len
;
8136 buf_len
= super
->len
;
8138 if (__le32_to_cpu(mpb
->mpb_size
) + len
> buf_len
) {
8139 /* ok we need a larger buf than what is currently allocated
8140 * if this allocation fails process_update will notice that
8141 * ->next_len is set and ->next_buf is NULL
8143 buf_len
= ROUND_UP(__le32_to_cpu(mpb
->mpb_size
) + len
, 512);
8144 if (super
->next_buf
)
8145 free(super
->next_buf
);
8147 super
->next_len
= buf_len
;
8148 if (posix_memalign(&super
->next_buf
, 512, buf_len
) == 0)
8149 memset(super
->next_buf
, 0, buf_len
);
8151 super
->next_buf
= NULL
;
8155 /* must be called while manager is quiesced */
8156 static void imsm_delete(struct intel_super
*super
, struct dl
**dlp
, unsigned index
)
8158 struct imsm_super
*mpb
= super
->anchor
;
8160 struct imsm_dev
*dev
;
8161 struct imsm_map
*map
;
8162 int i
, j
, num_members
;
8165 dprintf("%s: deleting device[%d] from imsm_super\n",
8168 /* shift all indexes down one */
8169 for (iter
= super
->disks
; iter
; iter
= iter
->next
)
8170 if (iter
->index
> (int)index
)
8172 for (iter
= super
->missing
; iter
; iter
= iter
->next
)
8173 if (iter
->index
> (int)index
)
8176 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
8177 dev
= get_imsm_dev(super
, i
);
8178 map
= get_imsm_map(dev
, MAP_0
);
8179 num_members
= map
->num_members
;
8180 for (j
= 0; j
< num_members
; j
++) {
8181 /* update ord entries being careful not to propagate
8182 * ord-flags to the first map
8184 ord
= get_imsm_ord_tbl_ent(dev
, j
, MAP_X
);
8186 if (ord_to_idx(ord
) <= index
)
8189 map
= get_imsm_map(dev
, MAP_0
);
8190 set_imsm_ord_tbl_ent(map
, j
, ord_to_idx(ord
- 1));
8191 map
= get_imsm_map(dev
, MAP_1
);
8193 set_imsm_ord_tbl_ent(map
, j
, ord
- 1);
8198 super
->updates_pending
++;
8200 struct dl
*dl
= *dlp
;
8202 *dlp
= (*dlp
)->next
;
8203 __free_imsm_disk(dl
);
8206 #endif /* MDASSEMBLE */
8208 static void close_targets(int *targets
, int new_disks
)
8215 for (i
= 0; i
< new_disks
; i
++) {
8216 if (targets
[i
] >= 0) {
8223 static int imsm_get_allowed_degradation(int level
, int raid_disks
,
8224 struct intel_super
*super
,
8225 struct imsm_dev
*dev
)
8230 struct imsm_map
*map
;
8233 ret_val
= raid_disks
/2;
8234 /* check map if all disks pairs not failed
8237 map
= get_imsm_map(dev
, MAP_0
);
8238 for (i
= 0; i
< ret_val
; i
++) {
8239 int degradation
= 0;
8240 if (get_imsm_disk(super
, i
) == NULL
)
8242 if (get_imsm_disk(super
, i
+ 1) == NULL
)
8244 if (degradation
== 2)
8247 map
= get_imsm_map(dev
, MAP_1
);
8248 /* if there is no second map
8249 * result can be returned
8253 /* check degradation in second map
8255 for (i
= 0; i
< ret_val
; i
++) {
8256 int degradation
= 0;
8257 if (get_imsm_disk(super
, i
) == NULL
)
8259 if (get_imsm_disk(super
, i
+ 1) == NULL
)
8261 if (degradation
== 2)
8276 /*******************************************************************************
8277 * Function: open_backup_targets
8278 * Description: Function opens file descriptors for all devices given in
8281 * info : general array info
8282 * raid_disks : number of disks
8283 * raid_fds : table of device's file descriptors
8284 * super : intel super for raid10 degradation check
8285 * dev : intel device for raid10 degradation check
8289 ******************************************************************************/
8290 int open_backup_targets(struct mdinfo
*info
, int raid_disks
, int *raid_fds
,
8291 struct intel_super
*super
, struct imsm_dev
*dev
)
8297 for (i
= 0; i
< raid_disks
; i
++)
8300 for (sd
= info
->devs
; sd
; sd
= sd
->next
) {
8303 if (sd
->disk
.state
& (1<<MD_DISK_FAULTY
)) {
8304 dprintf("disk is faulty!!\n");
8308 if ((sd
->disk
.raid_disk
>= raid_disks
) ||
8309 (sd
->disk
.raid_disk
< 0))
8312 dn
= map_dev(sd
->disk
.major
,
8314 raid_fds
[sd
->disk
.raid_disk
] = dev_open(dn
, O_RDWR
);
8315 if (raid_fds
[sd
->disk
.raid_disk
] < 0) {
8316 fprintf(stderr
, "cannot open component\n");
8321 /* check if maximum array degradation level is not exceeded
8323 if ((raid_disks
- opened
) >
8324 imsm_get_allowed_degradation(info
->new_level
,
8327 fprintf(stderr
, "Not enough disks can be opened.\n");
8328 close_targets(raid_fds
, raid_disks
);
8335 /*******************************************************************************
8336 * Function: init_migr_record_imsm
8337 * Description: Function inits imsm migration record
8339 * super : imsm internal array info
8340 * dev : device under migration
8341 * info : general array info to find the smallest device
8344 ******************************************************************************/
8345 void init_migr_record_imsm(struct supertype
*st
, struct imsm_dev
*dev
,
8346 struct mdinfo
*info
)
8348 struct intel_super
*super
= st
->sb
;
8349 struct migr_record
*migr_rec
= super
->migr_rec
;
8351 unsigned long long dsize
, dev_sectors
;
8352 long long unsigned min_dev_sectors
= -1LLU;
8356 struct imsm_map
*map_dest
= get_imsm_map(dev
, MAP_0
);
8357 struct imsm_map
*map_src
= get_imsm_map(dev
, MAP_1
);
8358 unsigned long long num_migr_units
;
8359 unsigned long long array_blocks
;
8361 memset(migr_rec
, 0, sizeof(struct migr_record
));
8362 migr_rec
->family_num
= __cpu_to_le32(super
->anchor
->family_num
);
8364 /* only ascending reshape supported now */
8365 migr_rec
->ascending_migr
= __cpu_to_le32(1);
8367 migr_rec
->dest_depth_per_unit
= GEN_MIGR_AREA_SIZE
/
8368 max(map_dest
->blocks_per_strip
, map_src
->blocks_per_strip
);
8369 migr_rec
->dest_depth_per_unit
*= map_dest
->blocks_per_strip
;
8370 new_data_disks
= imsm_num_data_members(dev
, MAP_0
);
8371 migr_rec
->blocks_per_unit
=
8372 __cpu_to_le32(migr_rec
->dest_depth_per_unit
* new_data_disks
);
8373 migr_rec
->dest_depth_per_unit
=
8374 __cpu_to_le32(migr_rec
->dest_depth_per_unit
);
8375 array_blocks
= info
->component_size
* new_data_disks
;
8377 array_blocks
/ __le32_to_cpu(migr_rec
->blocks_per_unit
);
8379 if (array_blocks
% __le32_to_cpu(migr_rec
->blocks_per_unit
))
8381 migr_rec
->num_migr_units
= __cpu_to_le32(num_migr_units
);
8383 migr_rec
->post_migr_vol_cap
= dev
->size_low
;
8384 migr_rec
->post_migr_vol_cap_hi
= dev
->size_high
;
8387 /* Find the smallest dev */
8388 for (sd
= info
->devs
; sd
; sd
= sd
->next
) {
8389 sprintf(nm
, "%d:%d", sd
->disk
.major
, sd
->disk
.minor
);
8390 fd
= dev_open(nm
, O_RDONLY
);
8393 get_dev_size(fd
, NULL
, &dsize
);
8394 dev_sectors
= dsize
/ 512;
8395 if (dev_sectors
< min_dev_sectors
)
8396 min_dev_sectors
= dev_sectors
;
8399 migr_rec
->ckpt_area_pba
= __cpu_to_le32(min_dev_sectors
-
8400 RAID_DISK_RESERVED_BLOCKS_IMSM_HI
);
8402 write_imsm_migr_rec(st
);
8407 /*******************************************************************************
8408 * Function: save_backup_imsm
8409 * Description: Function saves critical data stripes to Migration Copy Area
8410 * and updates the current migration unit status.
8411 * Use restore_stripes() to form a destination stripe,
8412 * and to write it to the Copy Area.
8414 * st : supertype information
8415 * dev : imsm device that backup is saved for
8416 * info : general array info
8417 * buf : input buffer
8418 * length : length of data to backup (blocks_per_unit)
8422 ******************************************************************************/
8423 int save_backup_imsm(struct supertype
*st
,
8424 struct imsm_dev
*dev
,
8425 struct mdinfo
*info
,
8430 struct intel_super
*super
= st
->sb
;
8431 unsigned long long *target_offsets
= NULL
;
8432 int *targets
= NULL
;
8434 struct imsm_map
*map_dest
= get_imsm_map(dev
, MAP_0
);
8435 int new_disks
= map_dest
->num_members
;
8436 int dest_layout
= 0;
8438 unsigned long long start
;
8439 int data_disks
= imsm_num_data_members(dev
, MAP_0
);
8441 targets
= malloc(new_disks
* sizeof(int));
8445 for (i
= 0; i
< new_disks
; i
++)
8448 target_offsets
= malloc(new_disks
* sizeof(unsigned long long));
8449 if (!target_offsets
)
8452 start
= info
->reshape_progress
* 512;
8453 for (i
= 0; i
< new_disks
; i
++) {
8454 target_offsets
[i
] = (unsigned long long)
8455 __le32_to_cpu(super
->migr_rec
->ckpt_area_pba
) * 512;
8456 /* move back copy area adderss, it will be moved forward
8457 * in restore_stripes() using start input variable
8459 target_offsets
[i
] -= start
/data_disks
;
8462 if (open_backup_targets(info
, new_disks
, targets
,
8466 dest_layout
= imsm_level_to_layout(map_dest
->raid_level
);
8467 dest_chunk
= __le16_to_cpu(map_dest
->blocks_per_strip
) * 512;
8469 if (restore_stripes(targets
, /* list of dest devices */
8470 target_offsets
, /* migration record offsets */
8473 map_dest
->raid_level
,
8475 -1, /* source backup file descriptor */
8476 0, /* input buf offset
8477 * always 0 buf is already offseted */
8481 fprintf(stderr
, Name
": Error restoring stripes\n");
8489 close_targets(targets
, new_disks
);
8492 free(target_offsets
);
8497 /*******************************************************************************
8498 * Function: save_checkpoint_imsm
8499 * Description: Function called for current unit status update
8500 * in the migration record. It writes it to disk.
8502 * super : imsm internal array info
8503 * info : general array info
8507 * 2: failure, means no valid migration record
8508 * / no general migration in progress /
8509 ******************************************************************************/
8510 int save_checkpoint_imsm(struct supertype
*st
, struct mdinfo
*info
, int state
)
8512 struct intel_super
*super
= st
->sb
;
8513 unsigned long long blocks_per_unit
;
8514 unsigned long long curr_migr_unit
;
8516 if (load_imsm_migr_rec(super
, info
) != 0) {
8517 dprintf("imsm: ERROR: Cannot read migration record "
8518 "for checkpoint save.\n");
8522 blocks_per_unit
= __le32_to_cpu(super
->migr_rec
->blocks_per_unit
);
8523 if (blocks_per_unit
== 0) {
8524 dprintf("imsm: no migration in progress.\n");
8527 curr_migr_unit
= info
->reshape_progress
/ blocks_per_unit
;
8528 /* check if array is alligned to copy area
8529 * if it is not alligned, add one to current migration unit value
8530 * this can happend on array reshape finish only
8532 if (info
->reshape_progress
% blocks_per_unit
)
8535 super
->migr_rec
->curr_migr_unit
=
8536 __cpu_to_le32(curr_migr_unit
);
8537 super
->migr_rec
->rec_status
= __cpu_to_le32(state
);
8538 super
->migr_rec
->dest_1st_member_lba
=
8539 __cpu_to_le32(curr_migr_unit
*
8540 __le32_to_cpu(super
->migr_rec
->dest_depth_per_unit
));
8541 if (write_imsm_migr_rec(st
) < 0) {
8542 dprintf("imsm: Cannot write migration record "
8543 "outside backup area\n");
8550 /*******************************************************************************
8551 * Function: recover_backup_imsm
8552 * Description: Function recovers critical data from the Migration Copy Area
8553 * while assembling an array.
8555 * super : imsm internal array info
8556 * info : general array info
8558 * 0 : success (or there is no data to recover)
8560 ******************************************************************************/
8561 int recover_backup_imsm(struct supertype
*st
, struct mdinfo
*info
)
8563 struct intel_super
*super
= st
->sb
;
8564 struct migr_record
*migr_rec
= super
->migr_rec
;
8565 struct imsm_map
*map_dest
= NULL
;
8566 struct intel_dev
*id
= NULL
;
8567 unsigned long long read_offset
;
8568 unsigned long long write_offset
;
8570 int *targets
= NULL
;
8571 int new_disks
, i
, err
;
8574 unsigned long curr_migr_unit
= __le32_to_cpu(migr_rec
->curr_migr_unit
);
8575 unsigned long num_migr_units
= __le32_to_cpu(migr_rec
->num_migr_units
);
8577 int skipped_disks
= 0;
8579 err
= sysfs_get_str(info
, NULL
, "array_state", (char *)buffer
, 20);
8583 /* recover data only during assemblation */
8584 if (strncmp(buffer
, "inactive", 8) != 0)
8586 /* no data to recover */
8587 if (__le32_to_cpu(migr_rec
->rec_status
) == UNIT_SRC_NORMAL
)
8589 if (curr_migr_unit
>= num_migr_units
)
8592 /* find device during reshape */
8593 for (id
= super
->devlist
; id
; id
= id
->next
)
8594 if (is_gen_migration(id
->dev
))
8599 map_dest
= get_imsm_map(id
->dev
, MAP_0
);
8600 new_disks
= map_dest
->num_members
;
8602 read_offset
= (unsigned long long)
8603 __le32_to_cpu(migr_rec
->ckpt_area_pba
) * 512;
8605 write_offset
= ((unsigned long long)
8606 __le32_to_cpu(migr_rec
->dest_1st_member_lba
) +
8607 __le32_to_cpu(map_dest
->pba_of_lba0
)) * 512;
8609 unit_len
= __le32_to_cpu(migr_rec
->dest_depth_per_unit
) * 512;
8610 if (posix_memalign((void **)&buf
, 512, unit_len
) != 0)
8612 targets
= malloc(new_disks
* sizeof(int));
8616 if (open_backup_targets(info
, new_disks
, targets
, super
, id
->dev
)) {
8618 Name
": Cannot open some devices belonging to array.\n");
8622 for (i
= 0; i
< new_disks
; i
++) {
8623 if (targets
[i
] < 0) {
8627 if (lseek64(targets
[i
], read_offset
, SEEK_SET
) < 0) {
8629 Name
": Cannot seek to block: %s\n",
8634 if ((unsigned)read(targets
[i
], buf
, unit_len
) != unit_len
) {
8636 Name
": Cannot read copy area block: %s\n",
8641 if (lseek64(targets
[i
], write_offset
, SEEK_SET
) < 0) {
8643 Name
": Cannot seek to block: %s\n",
8648 if ((unsigned)write(targets
[i
], buf
, unit_len
) != unit_len
) {
8650 Name
": Cannot restore block: %s\n",
8657 if (skipped_disks
> imsm_get_allowed_degradation(info
->new_level
,
8662 Name
": Cannot restore data from backup."
8663 " Too many failed disks\n");
8667 if (save_checkpoint_imsm(st
, info
, UNIT_SRC_NORMAL
)) {
8668 /* ignore error == 2, this can mean end of reshape here
8670 dprintf("imsm: Cannot write checkpoint to "
8671 "migration record (UNIT_SRC_NORMAL) during restart\n");
8677 for (i
= 0; i
< new_disks
; i
++)
8686 static char disk_by_path
[] = "/dev/disk/by-path/";
8688 static const char *imsm_get_disk_controller_domain(const char *path
)
8690 char disk_path
[PATH_MAX
];
8694 strncpy(disk_path
, disk_by_path
, PATH_MAX
- 1);
8695 strncat(disk_path
, path
, PATH_MAX
- strlen(disk_path
) - 1);
8696 if (stat(disk_path
, &st
) == 0) {
8697 struct sys_dev
* hba
;
8700 path
= devt_to_devpath(st
.st_rdev
);
8703 hba
= find_disk_attached_hba(-1, path
);
8704 if (hba
&& hba
->type
== SYS_DEV_SAS
)
8706 else if (hba
&& hba
->type
== SYS_DEV_SATA
)
8710 dprintf("path: %s hba: %s attached: %s\n",
8711 path
, (hba
) ? hba
->path
: "NULL", drv
);
8719 static int imsm_find_array_minor_by_subdev(int subdev
, int container
, int *minor
)
8721 char subdev_name
[20];
8722 struct mdstat_ent
*mdstat
;
8724 sprintf(subdev_name
, "%d", subdev
);
8725 mdstat
= mdstat_by_subdev(subdev_name
, container
);
8729 *minor
= mdstat
->devnum
;
8730 free_mdstat(mdstat
);
8734 static int imsm_reshape_is_allowed_on_container(struct supertype
*st
,
8735 struct geo_params
*geo
,
8736 int *old_raid_disks
)
8738 /* currently we only support increasing the number of devices
8739 * for a container. This increases the number of device for each
8740 * member array. They must all be RAID0 or RAID5.
8743 struct mdinfo
*info
, *member
;
8744 int devices_that_can_grow
= 0;
8746 dprintf("imsm: imsm_reshape_is_allowed_on_container(ENTER): "
8747 "st->devnum = (%i)\n",
8750 if (geo
->size
!= -1 ||
8751 geo
->level
!= UnSet
||
8752 geo
->layout
!= UnSet
||
8753 geo
->chunksize
!= 0 ||
8754 geo
->raid_disks
== UnSet
) {
8755 dprintf("imsm: Container operation is allowed for "
8756 "raid disks number change only.\n");
8760 info
= container_content_imsm(st
, NULL
);
8761 for (member
= info
; member
; member
= member
->next
) {
8765 dprintf("imsm: checking device_num: %i\n",
8766 member
->container_member
);
8768 if (geo
->raid_disks
<= member
->array
.raid_disks
) {
8769 /* we work on container for Online Capacity Expansion
8770 * only so raid_disks has to grow
8772 dprintf("imsm: for container operation raid disks "
8773 "increase is required\n");
8777 if ((info
->array
.level
!= 0) &&
8778 (info
->array
.level
!= 5)) {
8779 /* we cannot use this container with other raid level
8781 dprintf("imsm: for container operation wrong"
8782 " raid level (%i) detected\n",
8786 /* check for platform support
8787 * for this raid level configuration
8789 struct intel_super
*super
= st
->sb
;
8790 if (!is_raid_level_supported(super
->orom
,
8791 member
->array
.level
,
8793 dprintf("platform does not support raid%d with"
8797 geo
->raid_disks
> 1 ? "s" : "");
8800 /* check if component size is aligned to chunk size
8802 if (info
->component_size
%
8803 (info
->array
.chunk_size
/512)) {
8804 dprintf("Component size is not aligned to "
8810 if (*old_raid_disks
&&
8811 info
->array
.raid_disks
!= *old_raid_disks
)
8813 *old_raid_disks
= info
->array
.raid_disks
;
8815 /* All raid5 and raid0 volumes in container
8816 * have to be ready for Online Capacity Expansion
8817 * so they need to be assembled. We have already
8818 * checked that no recovery etc is happening.
8820 result
= imsm_find_array_minor_by_subdev(member
->container_member
,
8824 dprintf("imsm: cannot find array\n");
8827 devices_that_can_grow
++;
8830 if (!member
&& devices_that_can_grow
)
8834 dprintf("\tContainer operation allowed\n");
8836 dprintf("\tError: %i\n", ret_val
);
8841 /* Function: get_spares_for_grow
8842 * Description: Allocates memory and creates list of spare devices
8843 * avaliable in container. Checks if spare drive size is acceptable.
8844 * Parameters: Pointer to the supertype structure
8845 * Returns: Pointer to the list of spare devices (mdinfo structure) on success,
8848 static struct mdinfo
*get_spares_for_grow(struct supertype
*st
)
8850 unsigned long long min_size
= min_acceptable_spare_size_imsm(st
);
8851 return container_choose_spares(st
, min_size
, NULL
, NULL
, NULL
, 0);
8854 /******************************************************************************
8855 * function: imsm_create_metadata_update_for_reshape
8856 * Function creates update for whole IMSM container.
8858 ******************************************************************************/
8859 static int imsm_create_metadata_update_for_reshape(
8860 struct supertype
*st
,
8861 struct geo_params
*geo
,
8863 struct imsm_update_reshape
**updatep
)
8865 struct intel_super
*super
= st
->sb
;
8866 struct imsm_super
*mpb
= super
->anchor
;
8867 int update_memory_size
= 0;
8868 struct imsm_update_reshape
*u
= NULL
;
8869 struct mdinfo
*spares
= NULL
;
8871 int delta_disks
= 0;
8874 dprintf("imsm_update_metadata_for_reshape(enter) raid_disks = %i\n",
8877 delta_disks
= geo
->raid_disks
- old_raid_disks
;
8879 /* size of all update data without anchor */
8880 update_memory_size
= sizeof(struct imsm_update_reshape
);
8882 /* now add space for spare disks that we need to add. */
8883 update_memory_size
+= sizeof(u
->new_disks
[0]) * (delta_disks
- 1);
8885 u
= calloc(1, update_memory_size
);
8888 "cannot get memory for imsm_update_reshape update\n");
8891 u
->type
= update_reshape_container_disks
;
8892 u
->old_raid_disks
= old_raid_disks
;
8893 u
->new_raid_disks
= geo
->raid_disks
;
8895 /* now get spare disks list
8897 spares
= get_spares_for_grow(st
);
8900 || delta_disks
> spares
->array
.spare_disks
) {
8901 fprintf(stderr
, Name
": imsm: ERROR: Cannot get spare devices "
8902 "for %s.\n", geo
->dev_name
);
8907 /* we have got spares
8908 * update disk list in imsm_disk list table in anchor
8910 dprintf("imsm: %i spares are available.\n\n",
8911 spares
->array
.spare_disks
);
8914 for (i
= 0; i
< delta_disks
; i
++) {
8919 u
->new_disks
[i
] = makedev(dev
->disk
.major
,
8921 dl
= get_disk_super(super
, dev
->disk
.major
, dev
->disk
.minor
);
8922 dl
->index
= mpb
->num_disks
;
8932 dprintf("imsm: reshape update preparation :");
8933 if (i
== delta_disks
) {
8936 return update_memory_size
;
8939 dprintf(" Error\n");
8944 /******************************************************************************
8945 * function: imsm_create_metadata_update_for_migration()
8946 * Creates update for IMSM array.
8948 ******************************************************************************/
8949 static int imsm_create_metadata_update_for_migration(
8950 struct supertype
*st
,
8951 struct geo_params
*geo
,
8952 struct imsm_update_reshape_migration
**updatep
)
8954 struct intel_super
*super
= st
->sb
;
8955 int update_memory_size
= 0;
8956 struct imsm_update_reshape_migration
*u
= NULL
;
8957 struct imsm_dev
*dev
;
8958 int previous_level
= -1;
8960 dprintf("imsm_create_metadata_update_for_migration(enter)"
8961 " New Level = %i\n", geo
->level
);
8963 /* size of all update data without anchor */
8964 update_memory_size
= sizeof(struct imsm_update_reshape_migration
);
8966 u
= calloc(1, update_memory_size
);
8968 dprintf("error: cannot get memory for "
8969 "imsm_create_metadata_update_for_migration\n");
8972 u
->type
= update_reshape_migration
;
8973 u
->subdev
= super
->current_vol
;
8974 u
->new_level
= geo
->level
;
8975 u
->new_layout
= geo
->layout
;
8976 u
->new_raid_disks
= u
->old_raid_disks
= geo
->raid_disks
;
8977 u
->new_disks
[0] = -1;
8978 u
->new_chunksize
= -1;
8980 dev
= get_imsm_dev(super
, u
->subdev
);
8982 struct imsm_map
*map
;
8984 map
= get_imsm_map(dev
, MAP_0
);
8986 int current_chunk_size
=
8987 __le16_to_cpu(map
->blocks_per_strip
) / 2;
8989 if (geo
->chunksize
!= current_chunk_size
) {
8990 u
->new_chunksize
= geo
->chunksize
/ 1024;
8992 "chunk size change from %i to %i\n",
8993 current_chunk_size
, u
->new_chunksize
);
8995 previous_level
= map
->raid_level
;
8998 if ((geo
->level
== 5) && (previous_level
== 0)) {
8999 struct mdinfo
*spares
= NULL
;
9001 u
->new_raid_disks
++;
9002 spares
= get_spares_for_grow(st
);
9003 if ((spares
== NULL
) || (spares
->array
.spare_disks
< 1)) {
9006 update_memory_size
= 0;
9007 dprintf("error: cannot get spare device "
9008 "for requested migration");
9013 dprintf("imsm: reshape update preparation : OK\n");
9016 return update_memory_size
;
9019 static void imsm_update_metadata_locally(struct supertype
*st
,
9022 struct metadata_update mu
;
9027 mu
.space_list
= NULL
;
9029 imsm_prepare_update(st
, &mu
);
9030 imsm_process_update(st
, &mu
);
9032 while (mu
.space_list
) {
9033 void **space
= mu
.space_list
;
9034 mu
.space_list
= *space
;
9039 /***************************************************************************
9040 * Function: imsm_analyze_change
9041 * Description: Function analyze change for single volume
9042 * and validate if transition is supported
9043 * Parameters: Geometry parameters, supertype structure
9044 * Returns: Operation type code on success, -1 if fail
9045 ****************************************************************************/
9046 enum imsm_reshape_type
imsm_analyze_change(struct supertype
*st
,
9047 struct geo_params
*geo
)
9053 /* number of added/removed disks in operation result */
9054 int devNumChange
= 0;
9055 /* imsm compatible layout value for array geometry verification */
9056 int imsm_layout
= -1;
9058 getinfo_super_imsm_volume(st
, &info
, NULL
);
9059 if ((geo
->level
!= info
.array
.level
) &&
9060 (geo
->level
>= 0) &&
9061 (geo
->level
!= UnSet
)) {
9062 switch (info
.array
.level
) {
9064 if (geo
->level
== 5) {
9065 change
= CH_MIGRATION
;
9066 if (geo
->layout
!= ALGORITHM_LEFT_ASYMMETRIC
) {
9068 Name
" Error. Requested Layout "
9069 "not supported (left-asymmetric layout "
9070 "is supported only)!\n");
9072 goto analyse_change_exit
;
9074 imsm_layout
= geo
->layout
;
9076 devNumChange
= 1; /* parity disk added */
9077 } else if (geo
->level
== 10) {
9078 change
= CH_TAKEOVER
;
9080 devNumChange
= 2; /* two mirrors added */
9081 imsm_layout
= 0x102; /* imsm supported layout */
9086 if (geo
->level
== 0) {
9087 change
= CH_TAKEOVER
;
9089 devNumChange
= -(geo
->raid_disks
/2);
9090 imsm_layout
= 0; /* imsm raid0 layout */
9096 Name
" Error. Level Migration from %d to %d "
9098 info
.array
.level
, geo
->level
);
9099 goto analyse_change_exit
;
9102 geo
->level
= info
.array
.level
;
9104 if ((geo
->layout
!= info
.array
.layout
)
9105 && ((geo
->layout
!= UnSet
) && (geo
->layout
!= -1))) {
9106 change
= CH_MIGRATION
;
9107 if ((info
.array
.layout
== 0)
9108 && (info
.array
.level
== 5)
9109 && (geo
->layout
== 5)) {
9110 /* reshape 5 -> 4 */
9111 } else if ((info
.array
.layout
== 5)
9112 && (info
.array
.level
== 5)
9113 && (geo
->layout
== 0)) {
9114 /* reshape 4 -> 5 */
9119 Name
" Error. Layout Migration from %d to %d "
9121 info
.array
.layout
, geo
->layout
);
9123 goto analyse_change_exit
;
9126 geo
->layout
= info
.array
.layout
;
9127 if (imsm_layout
== -1)
9128 imsm_layout
= info
.array
.layout
;
9131 if ((geo
->chunksize
> 0) && (geo
->chunksize
!= UnSet
)
9132 && (geo
->chunksize
!= info
.array
.chunk_size
))
9133 change
= CH_MIGRATION
;
9135 geo
->chunksize
= info
.array
.chunk_size
;
9137 chunk
= geo
->chunksize
/ 1024;
9138 if (!validate_geometry_imsm(st
,
9141 geo
->raid_disks
+ devNumChange
,
9148 struct intel_super
*super
= st
->sb
;
9149 struct imsm_super
*mpb
= super
->anchor
;
9151 if (mpb
->num_raid_devs
> 1) {
9153 Name
" Error. Cannot perform operation on %s"
9154 "- for this operation it MUST be single "
9155 "array in container\n",
9161 analyse_change_exit
:
9166 int imsm_takeover(struct supertype
*st
, struct geo_params
*geo
)
9168 struct intel_super
*super
= st
->sb
;
9169 struct imsm_update_takeover
*u
;
9171 u
= malloc(sizeof(struct imsm_update_takeover
));
9175 u
->type
= update_takeover
;
9176 u
->subarray
= super
->current_vol
;
9178 /* 10->0 transition */
9179 if (geo
->level
== 0)
9180 u
->direction
= R10_TO_R0
;
9182 /* 0->10 transition */
9183 if (geo
->level
== 10)
9184 u
->direction
= R0_TO_R10
;
9186 /* update metadata locally */
9187 imsm_update_metadata_locally(st
, u
,
9188 sizeof(struct imsm_update_takeover
));
9189 /* and possibly remotely */
9190 if (st
->update_tail
)
9191 append_metadata_update(st
, u
,
9192 sizeof(struct imsm_update_takeover
));
9199 static int imsm_reshape_super(struct supertype
*st
, long long size
, int level
,
9200 int layout
, int chunksize
, int raid_disks
,
9201 int delta_disks
, char *backup
, char *dev
,
9205 struct geo_params geo
;
9207 dprintf("imsm: reshape_super called.\n");
9209 memset(&geo
, 0, sizeof(struct geo_params
));
9212 geo
.dev_id
= st
->devnum
;
9215 geo
.layout
= layout
;
9216 geo
.chunksize
= chunksize
;
9217 geo
.raid_disks
= raid_disks
;
9218 if (delta_disks
!= UnSet
)
9219 geo
.raid_disks
+= delta_disks
;
9221 dprintf("\tfor level : %i\n", geo
.level
);
9222 dprintf("\tfor raid_disks : %i\n", geo
.raid_disks
);
9224 if (experimental() == 0)
9227 if (st
->container_dev
== st
->devnum
) {
9228 /* On container level we can only increase number of devices. */
9229 dprintf("imsm: info: Container operation\n");
9230 int old_raid_disks
= 0;
9232 if (imsm_reshape_is_allowed_on_container(
9233 st
, &geo
, &old_raid_disks
)) {
9234 struct imsm_update_reshape
*u
= NULL
;
9237 len
= imsm_create_metadata_update_for_reshape(
9238 st
, &geo
, old_raid_disks
, &u
);
9241 dprintf("imsm: Cannot prepare update\n");
9242 goto exit_imsm_reshape_super
;
9246 /* update metadata locally */
9247 imsm_update_metadata_locally(st
, u
, len
);
9248 /* and possibly remotely */
9249 if (st
->update_tail
)
9250 append_metadata_update(st
, u
, len
);
9255 fprintf(stderr
, Name
": (imsm) Operation "
9256 "is not allowed on this container\n");
9259 /* On volume level we support following operations
9260 * - takeover: raid10 -> raid0; raid0 -> raid10
9261 * - chunk size migration
9262 * - migration: raid5 -> raid0; raid0 -> raid5
9264 struct intel_super
*super
= st
->sb
;
9265 struct intel_dev
*dev
= super
->devlist
;
9267 dprintf("imsm: info: Volume operation\n");
9268 /* find requested device */
9270 if (imsm_find_array_minor_by_subdev(
9271 dev
->index
, st
->container_dev
, &devnum
) == 0
9272 && devnum
== geo
.dev_id
)
9277 fprintf(stderr
, Name
" Cannot find %s (%i) subarray\n",
9278 geo
.dev_name
, geo
.dev_id
);
9279 goto exit_imsm_reshape_super
;
9281 super
->current_vol
= dev
->index
;
9282 change
= imsm_analyze_change(st
, &geo
);
9285 ret_val
= imsm_takeover(st
, &geo
);
9287 case CH_MIGRATION
: {
9288 struct imsm_update_reshape_migration
*u
= NULL
;
9290 imsm_create_metadata_update_for_migration(
9294 "Cannot prepare update\n");
9298 /* update metadata locally */
9299 imsm_update_metadata_locally(st
, u
, len
);
9300 /* and possibly remotely */
9301 if (st
->update_tail
)
9302 append_metadata_update(st
, u
, len
);
9312 exit_imsm_reshape_super
:
9313 dprintf("imsm: reshape_super Exit code = %i\n", ret_val
);
9317 /*******************************************************************************
9318 * Function: wait_for_reshape_imsm
9319 * Description: Function writes new sync_max value and waits until
9320 * reshape process reach new position
9322 * sra : general array info
9323 * ndata : number of disks in new array's layout
9326 * 1 : there is no reshape in progress,
9328 ******************************************************************************/
9329 int wait_for_reshape_imsm(struct mdinfo
*sra
, int ndata
)
9331 int fd
= sysfs_get_fd(sra
, NULL
, "reshape_position");
9332 unsigned long long completed
;
9333 /* to_complete : new sync_max position */
9334 unsigned long long to_complete
= sra
->reshape_progress
;
9335 unsigned long long position_to_set
= to_complete
/ ndata
;
9338 dprintf("imsm: wait_for_reshape_imsm() "
9339 "cannot open reshape_position\n");
9343 if (sysfs_fd_get_ll(fd
, &completed
) < 0) {
9344 dprintf("imsm: wait_for_reshape_imsm() "
9345 "cannot read reshape_position (no reshape in progres)\n");
9350 if (completed
> to_complete
) {
9351 dprintf("imsm: wait_for_reshape_imsm() "
9352 "wrong next position to set %llu (%llu)\n",
9353 to_complete
, completed
);
9357 dprintf("Position set: %llu\n", position_to_set
);
9358 if (sysfs_set_num(sra
, NULL
, "sync_max",
9359 position_to_set
) != 0) {
9360 dprintf("imsm: wait_for_reshape_imsm() "
9361 "cannot set reshape position to %llu\n",
9372 select(fd
+1, &rfds
, NULL
, NULL
, NULL
);
9373 if (sysfs_get_str(sra
, NULL
, "sync_action",
9375 strncmp(action
, "reshape", 7) != 0)
9377 if (sysfs_fd_get_ll(fd
, &completed
) < 0) {
9378 dprintf("imsm: wait_for_reshape_imsm() "
9379 "cannot read reshape_position (in loop)\n");
9383 } while (completed
< to_complete
);
9389 /*******************************************************************************
9390 * Function: check_degradation_change
9391 * Description: Check that array hasn't become failed.
9393 * info : for sysfs access
9394 * sources : source disks descriptors
9395 * degraded: previous degradation level
9398 ******************************************************************************/
9399 int check_degradation_change(struct mdinfo
*info
,
9403 unsigned long long new_degraded
;
9404 sysfs_get_ll(info
, NULL
, "degraded", &new_degraded
);
9405 if (new_degraded
!= (unsigned long long)degraded
) {
9406 /* check each device to ensure it is still working */
9409 for (sd
= info
->devs
; sd
; sd
= sd
->next
) {
9410 if (sd
->disk
.state
& (1<<MD_DISK_FAULTY
))
9412 if (sd
->disk
.state
& (1<<MD_DISK_SYNC
)) {
9414 if (sysfs_get_str(info
,
9415 sd
, "state", sbuf
, 20) < 0 ||
9416 strstr(sbuf
, "faulty") ||
9417 strstr(sbuf
, "in_sync") == NULL
) {
9418 /* this device is dead */
9419 sd
->disk
.state
= (1<<MD_DISK_FAULTY
);
9420 if (sd
->disk
.raid_disk
>= 0 &&
9421 sources
[sd
->disk
.raid_disk
] >= 0) {
9423 sd
->disk
.raid_disk
]);
9424 sources
[sd
->disk
.raid_disk
] =
9433 return new_degraded
;
9436 /*******************************************************************************
9437 * Function: imsm_manage_reshape
9438 * Description: Function finds array under reshape and it manages reshape
9439 * process. It creates stripes backups (if required) and sets
9442 * afd : Backup handle (nattive) - not used
9443 * sra : general array info
9444 * reshape : reshape parameters - not used
9445 * st : supertype structure
9446 * blocks : size of critical section [blocks]
9447 * fds : table of source device descriptor
9448 * offsets : start of array (offest per devices)
9450 * destfd : table of destination device descriptor
9451 * destoffsets : table of destination offsets (per device)
9453 * 1 : success, reshape is done
9455 ******************************************************************************/
9456 static int imsm_manage_reshape(
9457 int afd
, struct mdinfo
*sra
, struct reshape
*reshape
,
9458 struct supertype
*st
, unsigned long backup_blocks
,
9459 int *fds
, unsigned long long *offsets
,
9460 int dests
, int *destfd
, unsigned long long *destoffsets
)
9463 struct intel_super
*super
= st
->sb
;
9464 struct intel_dev
*dv
= NULL
;
9465 struct imsm_dev
*dev
= NULL
;
9466 struct imsm_map
*map_src
;
9467 int migr_vol_qan
= 0;
9468 int ndata
, odata
; /* [bytes] */
9469 int chunk
; /* [bytes] */
9470 struct migr_record
*migr_rec
;
9472 unsigned int buf_size
; /* [bytes] */
9473 unsigned long long max_position
; /* array size [bytes] */
9474 unsigned long long next_step
; /* [blocks]/[bytes] */
9475 unsigned long long old_data_stripe_length
;
9476 unsigned long long start_src
; /* [bytes] */
9477 unsigned long long start
; /* [bytes] */
9478 unsigned long long start_buf_shift
; /* [bytes] */
9480 int source_layout
= 0;
9482 if (!fds
|| !offsets
|| !sra
)
9485 /* Find volume during the reshape */
9486 for (dv
= super
->devlist
; dv
; dv
= dv
->next
) {
9487 if (dv
->dev
->vol
.migr_type
== MIGR_GEN_MIGR
9488 && dv
->dev
->vol
.migr_state
== 1) {
9493 /* Only one volume can migrate at the same time */
9494 if (migr_vol_qan
!= 1) {
9495 fprintf(stderr
, Name
" : %s", migr_vol_qan
?
9496 "Number of migrating volumes greater than 1\n" :
9497 "There is no volume during migrationg\n");
9501 map_src
= get_imsm_map(dev
, MAP_1
);
9502 if (map_src
== NULL
)
9505 ndata
= imsm_num_data_members(dev
, MAP_0
);
9506 odata
= imsm_num_data_members(dev
, MAP_1
);
9508 chunk
= __le16_to_cpu(map_src
->blocks_per_strip
) * 512;
9509 old_data_stripe_length
= odata
* chunk
;
9511 migr_rec
= super
->migr_rec
;
9513 /* initialize migration record for start condition */
9514 if (sra
->reshape_progress
== 0)
9515 init_migr_record_imsm(st
, dev
, sra
);
9517 if (__le32_to_cpu(migr_rec
->rec_status
) != UNIT_SRC_NORMAL
) {
9518 dprintf("imsm: cannot restart migration when data "
9519 "are present in copy area.\n");
9525 buf_size
= __le32_to_cpu(migr_rec
->blocks_per_unit
) * 512;
9526 /* extend buffer size for parity disk */
9527 buf_size
+= __le32_to_cpu(migr_rec
->dest_depth_per_unit
) * 512;
9528 /* add space for stripe aligment */
9529 buf_size
+= old_data_stripe_length
;
9530 if (posix_memalign((void **)&buf
, 4096, buf_size
)) {
9531 dprintf("imsm: Cannot allocate checpoint buffer\n");
9535 max_position
= sra
->component_size
* ndata
;
9536 source_layout
= imsm_level_to_layout(map_src
->raid_level
);
9538 while (__le32_to_cpu(migr_rec
->curr_migr_unit
) <
9539 __le32_to_cpu(migr_rec
->num_migr_units
)) {
9540 /* current reshape position [blocks] */
9541 unsigned long long current_position
=
9542 __le32_to_cpu(migr_rec
->blocks_per_unit
)
9543 * __le32_to_cpu(migr_rec
->curr_migr_unit
);
9544 unsigned long long border
;
9546 /* Check that array hasn't become failed.
9548 degraded
= check_degradation_change(sra
, fds
, degraded
);
9550 dprintf("imsm: Abort reshape due to degradation"
9551 " level (%i)\n", degraded
);
9555 next_step
= __le32_to_cpu(migr_rec
->blocks_per_unit
);
9557 if ((current_position
+ next_step
) > max_position
)
9558 next_step
= max_position
- current_position
;
9560 start
= current_position
* 512;
9562 /* allign reading start to old geometry */
9563 start_buf_shift
= start
% old_data_stripe_length
;
9564 start_src
= start
- start_buf_shift
;
9566 border
= (start_src
/ odata
) - (start
/ ndata
);
9568 if (border
<= __le32_to_cpu(migr_rec
->dest_depth_per_unit
)) {
9569 /* save critical stripes to buf
9570 * start - start address of current unit
9572 * start_src - start address of current unit
9573 * to backup alligned to source array
9576 unsigned long long next_step_filler
= 0;
9577 unsigned long long copy_length
= next_step
* 512;
9579 /* allign copy area length to stripe in old geometry */
9580 next_step_filler
= ((copy_length
+ start_buf_shift
)
9581 % old_data_stripe_length
);
9582 if (next_step_filler
)
9583 next_step_filler
= (old_data_stripe_length
9584 - next_step_filler
);
9585 dprintf("save_stripes() parameters: start = %llu,"
9586 "\tstart_src = %llu,\tnext_step*512 = %llu,"
9587 "\tstart_in_buf_shift = %llu,"
9588 "\tnext_step_filler = %llu\n",
9589 start
, start_src
, copy_length
,
9590 start_buf_shift
, next_step_filler
);
9592 if (save_stripes(fds
, offsets
, map_src
->num_members
,
9593 chunk
, map_src
->raid_level
,
9594 source_layout
, 0, NULL
, start_src
,
9596 next_step_filler
+ start_buf_shift
,
9598 dprintf("imsm: Cannot save stripes"
9602 /* Convert data to destination format and store it
9603 * in backup general migration area
9605 if (save_backup_imsm(st
, dev
, sra
,
9606 buf
+ start_buf_shift
, copy_length
)) {
9607 dprintf("imsm: Cannot save stripes to "
9608 "target devices\n");
9611 if (save_checkpoint_imsm(st
, sra
,
9612 UNIT_SRC_IN_CP_AREA
)) {
9613 dprintf("imsm: Cannot write checkpoint to "
9614 "migration record (UNIT_SRC_IN_CP_AREA)\n");
9618 /* set next step to use whole border area */
9619 border
/= next_step
;
9621 next_step
*= border
;
9623 /* When data backed up, checkpoint stored,
9624 * kick the kernel to reshape unit of data
9626 next_step
= next_step
+ sra
->reshape_progress
;
9627 /* limit next step to array max position */
9628 if (next_step
> max_position
)
9629 next_step
= max_position
;
9630 sysfs_set_num(sra
, NULL
, "suspend_lo", sra
->reshape_progress
);
9631 sysfs_set_num(sra
, NULL
, "suspend_hi", next_step
);
9632 sra
->reshape_progress
= next_step
;
9634 /* wait until reshape finish */
9635 if (wait_for_reshape_imsm(sra
, ndata
) < 0) {
9636 dprintf("wait_for_reshape_imsm returned error!\n");
9640 if (save_checkpoint_imsm(st
, sra
, UNIT_SRC_NORMAL
) == 1) {
9641 /* ignore error == 2, this can mean end of reshape here
9643 dprintf("imsm: Cannot write checkpoint to "
9644 "migration record (UNIT_SRC_NORMAL)\n");
9650 /* return '1' if done */
9658 #endif /* MDASSEMBLE */
9660 struct superswitch super_imsm
= {
9662 .examine_super
= examine_super_imsm
,
9663 .brief_examine_super
= brief_examine_super_imsm
,
9664 .brief_examine_subarrays
= brief_examine_subarrays_imsm
,
9665 .export_examine_super
= export_examine_super_imsm
,
9666 .detail_super
= detail_super_imsm
,
9667 .brief_detail_super
= brief_detail_super_imsm
,
9668 .write_init_super
= write_init_super_imsm
,
9669 .validate_geometry
= validate_geometry_imsm
,
9670 .add_to_super
= add_to_super_imsm
,
9671 .remove_from_super
= remove_from_super_imsm
,
9672 .detail_platform
= detail_platform_imsm
,
9673 .kill_subarray
= kill_subarray_imsm
,
9674 .update_subarray
= update_subarray_imsm
,
9675 .load_container
= load_container_imsm
,
9676 .default_geometry
= default_geometry_imsm
,
9677 .get_disk_controller_domain
= imsm_get_disk_controller_domain
,
9678 .reshape_super
= imsm_reshape_super
,
9679 .manage_reshape
= imsm_manage_reshape
,
9680 .recover_backup
= recover_backup_imsm
,
9682 .match_home
= match_home_imsm
,
9683 .uuid_from_super
= uuid_from_super_imsm
,
9684 .getinfo_super
= getinfo_super_imsm
,
9685 .getinfo_super_disks
= getinfo_super_disks_imsm
,
9686 .update_super
= update_super_imsm
,
9688 .avail_size
= avail_size_imsm
,
9689 .min_acceptable_spare_size
= min_acceptable_spare_size_imsm
,
9691 .compare_super
= compare_super_imsm
,
9693 .load_super
= load_super_imsm
,
9694 .init_super
= init_super_imsm
,
9695 .store_super
= store_super_imsm
,
9696 .free_super
= free_super_imsm
,
9697 .match_metadata_desc
= match_metadata_desc_imsm
,
9698 .container_content
= container_content_imsm
,
9706 .open_new
= imsm_open_new
,
9707 .set_array_state
= imsm_set_array_state
,
9708 .set_disk
= imsm_set_disk
,
9709 .sync_metadata
= imsm_sync_metadata
,
9710 .activate_spare
= imsm_activate_spare
,
9711 .process_update
= imsm_process_update
,
9712 .prepare_update
= imsm_prepare_update
,
9713 #endif /* MDASSEMBLE */