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) {
2795 /* We take this to mean that the family_num should be updated.
2796 * However that is much smaller than the uuid so we cannot really
2797 * allow an explicit uuid to be given. And it is hard to reliably
2799 * So if !uuid_set we know the current uuid is random and just used
2800 * the first 'int' and copy it to the other 3 positions.
2801 * Otherwise we require the 4 'int's to be the same as would be the
2802 * case if we are using a random uuid. So an explicit uuid will be
2803 * accepted as long as all for ints are the same... which shouldn't hurt
2806 info
->uuid
[1] = info
->uuid
[2] = info
->uuid
[3] = info
->uuid
[0];
2809 if (info
->uuid
[0] != info
->uuid
[1] ||
2810 info
->uuid
[1] != info
->uuid
[2] ||
2811 info
->uuid
[2] != info
->uuid
[3])
2817 mpb
->orig_family_num
= info
->uuid
[0];
2818 } else if (strcmp(update
, "assemble") == 0)
2823 /* successful update? recompute checksum */
2825 mpb
->check_sum
= __le32_to_cpu(__gen_imsm_checksum(mpb
));
2830 static size_t disks_to_mpb_size(int disks
)
2834 size
= sizeof(struct imsm_super
);
2835 size
+= (disks
- 1) * sizeof(struct imsm_disk
);
2836 size
+= 2 * sizeof(struct imsm_dev
);
2837 /* up to 2 maps per raid device (-2 for imsm_maps in imsm_dev */
2838 size
+= (4 - 2) * sizeof(struct imsm_map
);
2839 /* 4 possible disk_ord_tbl's */
2840 size
+= 4 * (disks
- 1) * sizeof(__u32
);
2845 static __u64
avail_size_imsm(struct supertype
*st
, __u64 devsize
)
2847 if (devsize
< (MPB_SECTOR_CNT
+ IMSM_RESERVED_SECTORS
))
2850 return devsize
- (MPB_SECTOR_CNT
+ IMSM_RESERVED_SECTORS
);
2853 static void free_devlist(struct intel_super
*super
)
2855 struct intel_dev
*dv
;
2857 while (super
->devlist
) {
2858 dv
= super
->devlist
->next
;
2859 free(super
->devlist
->dev
);
2860 free(super
->devlist
);
2861 super
->devlist
= dv
;
2865 static void imsm_copy_dev(struct imsm_dev
*dest
, struct imsm_dev
*src
)
2867 memcpy(dest
, src
, sizeof_imsm_dev(src
, 0));
2870 static int compare_super_imsm(struct supertype
*st
, struct supertype
*tst
)
2874 * 0 same, or first was empty, and second was copied
2875 * 1 second had wrong number
2877 * 3 wrong other info
2879 struct intel_super
*first
= st
->sb
;
2880 struct intel_super
*sec
= tst
->sb
;
2887 /* in platform dependent environment test if the disks
2888 * use the same Intel hba
2890 if (!check_env("IMSM_NO_PLATFORM")) {
2891 if (!first
->hba
|| !sec
->hba
||
2892 (first
->hba
->type
!= sec
->hba
->type
)) {
2894 "HBAs of devices does not match %s != %s\n",
2895 first
->hba
? get_sys_dev_type(first
->hba
->type
) : NULL
,
2896 sec
->hba
? get_sys_dev_type(sec
->hba
->type
) : NULL
);
2901 /* if an anchor does not have num_raid_devs set then it is a free
2904 if (first
->anchor
->num_raid_devs
> 0 &&
2905 sec
->anchor
->num_raid_devs
> 0) {
2906 /* Determine if these disks might ever have been
2907 * related. Further disambiguation can only take place
2908 * in load_super_imsm_all
2910 __u32 first_family
= first
->anchor
->orig_family_num
;
2911 __u32 sec_family
= sec
->anchor
->orig_family_num
;
2913 if (memcmp(first
->anchor
->sig
, sec
->anchor
->sig
,
2914 MAX_SIGNATURE_LENGTH
) != 0)
2917 if (first_family
== 0)
2918 first_family
= first
->anchor
->family_num
;
2919 if (sec_family
== 0)
2920 sec_family
= sec
->anchor
->family_num
;
2922 if (first_family
!= sec_family
)
2928 /* if 'first' is a spare promote it to a populated mpb with sec's
2931 if (first
->anchor
->num_raid_devs
== 0 &&
2932 sec
->anchor
->num_raid_devs
> 0) {
2934 struct intel_dev
*dv
;
2935 struct imsm_dev
*dev
;
2937 /* we need to copy raid device info from sec if an allocation
2938 * fails here we don't associate the spare
2940 for (i
= 0; i
< sec
->anchor
->num_raid_devs
; i
++) {
2941 dv
= malloc(sizeof(*dv
));
2944 dev
= malloc(sizeof_imsm_dev(get_imsm_dev(sec
, i
), 1));
2951 dv
->next
= first
->devlist
;
2952 first
->devlist
= dv
;
2954 if (i
< sec
->anchor
->num_raid_devs
) {
2955 /* allocation failure */
2956 free_devlist(first
);
2957 fprintf(stderr
, "imsm: failed to associate spare\n");
2960 first
->anchor
->num_raid_devs
= sec
->anchor
->num_raid_devs
;
2961 first
->anchor
->orig_family_num
= sec
->anchor
->orig_family_num
;
2962 first
->anchor
->family_num
= sec
->anchor
->family_num
;
2963 memcpy(first
->anchor
->sig
, sec
->anchor
->sig
, MAX_SIGNATURE_LENGTH
);
2964 for (i
= 0; i
< sec
->anchor
->num_raid_devs
; i
++)
2965 imsm_copy_dev(get_imsm_dev(first
, i
), get_imsm_dev(sec
, i
));
2971 static void fd2devname(int fd
, char *name
)
2975 char dname
[PATH_MAX
];
2980 if (fstat(fd
, &st
) != 0)
2982 sprintf(path
, "/sys/dev/block/%d:%d",
2983 major(st
.st_rdev
), minor(st
.st_rdev
));
2985 rv
= readlink(path
, dname
, sizeof(dname
)-1);
2990 nm
= strrchr(dname
, '/');
2993 snprintf(name
, MAX_RAID_SERIAL_LEN
, "/dev/%s", nm
);
2997 extern int scsi_get_serial(int fd
, void *buf
, size_t buf_len
);
2999 static int imsm_read_serial(int fd
, char *devname
,
3000 __u8 serial
[MAX_RAID_SERIAL_LEN
])
3002 unsigned char scsi_serial
[255];
3011 memset(scsi_serial
, 0, sizeof(scsi_serial
));
3013 rv
= scsi_get_serial(fd
, scsi_serial
, sizeof(scsi_serial
));
3015 if (rv
&& check_env("IMSM_DEVNAME_AS_SERIAL")) {
3016 memset(serial
, 0, MAX_RAID_SERIAL_LEN
);
3017 fd2devname(fd
, (char *) serial
);
3024 Name
": Failed to retrieve serial for %s\n",
3029 rsp_len
= scsi_serial
[3];
3033 Name
": Failed to retrieve serial for %s\n",
3037 rsp_buf
= (char *) &scsi_serial
[4];
3039 /* trim all whitespace and non-printable characters and convert
3042 for (i
= 0, dest
= rsp_buf
; i
< rsp_len
; i
++) {
3045 /* ':' is reserved for use in placeholder serial
3046 * numbers for missing disks
3054 len
= dest
- rsp_buf
;
3057 /* truncate leading characters */
3058 if (len
> MAX_RAID_SERIAL_LEN
) {
3059 dest
+= len
- MAX_RAID_SERIAL_LEN
;
3060 len
= MAX_RAID_SERIAL_LEN
;
3063 memset(serial
, 0, MAX_RAID_SERIAL_LEN
);
3064 memcpy(serial
, dest
, len
);
3069 static int serialcmp(__u8
*s1
, __u8
*s2
)
3071 return strncmp((char *) s1
, (char *) s2
, MAX_RAID_SERIAL_LEN
);
3074 static void serialcpy(__u8
*dest
, __u8
*src
)
3076 strncpy((char *) dest
, (char *) src
, MAX_RAID_SERIAL_LEN
);
3079 static struct dl
*serial_to_dl(__u8
*serial
, struct intel_super
*super
)
3083 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
3084 if (serialcmp(dl
->serial
, serial
) == 0)
3090 static struct imsm_disk
*
3091 __serial_to_disk(__u8
*serial
, struct imsm_super
*mpb
, int *idx
)
3095 for (i
= 0; i
< mpb
->num_disks
; i
++) {
3096 struct imsm_disk
*disk
= __get_imsm_disk(mpb
, i
);
3098 if (serialcmp(disk
->serial
, serial
) == 0) {
3109 load_imsm_disk(int fd
, struct intel_super
*super
, char *devname
, int keep_fd
)
3111 struct imsm_disk
*disk
;
3116 __u8 serial
[MAX_RAID_SERIAL_LEN
];
3118 rv
= imsm_read_serial(fd
, devname
, serial
);
3123 dl
= calloc(1, sizeof(*dl
));
3127 Name
": failed to allocate disk buffer for %s\n",
3133 dl
->major
= major(stb
.st_rdev
);
3134 dl
->minor
= minor(stb
.st_rdev
);
3135 dl
->next
= super
->disks
;
3136 dl
->fd
= keep_fd
? fd
: -1;
3137 assert(super
->disks
== NULL
);
3139 serialcpy(dl
->serial
, serial
);
3142 fd2devname(fd
, name
);
3144 dl
->devname
= strdup(devname
);
3146 dl
->devname
= strdup(name
);
3148 /* look up this disk's index in the current anchor */
3149 disk
= __serial_to_disk(dl
->serial
, super
->anchor
, &dl
->index
);
3152 /* only set index on disks that are a member of a
3153 * populated contianer, i.e. one with raid_devs
3155 if (is_failed(&dl
->disk
))
3157 else if (is_spare(&dl
->disk
))
3165 /* When migrating map0 contains the 'destination' state while map1
3166 * contains the current state. When not migrating map0 contains the
3167 * current state. This routine assumes that map[0].map_state is set to
3168 * the current array state before being called.
3170 * Migration is indicated by one of the following states
3171 * 1/ Idle (migr_state=0 map0state=normal||unitialized||degraded||failed)
3172 * 2/ Initialize (migr_state=1 migr_type=MIGR_INIT map0state=normal
3173 * map1state=unitialized)
3174 * 3/ Repair (Resync) (migr_state=1 migr_type=MIGR_REPAIR map0state=normal
3176 * 4/ Rebuild (migr_state=1 migr_type=MIGR_REBUILD map0state=normal
3177 * map1state=degraded)
3178 * 5/ Migration (mig_state=1 migr_type=MIGR_GEN_MIGR map0state=normal
3181 static void migrate(struct imsm_dev
*dev
, struct intel_super
*super
,
3182 __u8 to_state
, int migr_type
)
3184 struct imsm_map
*dest
;
3185 struct imsm_map
*src
= get_imsm_map(dev
, MAP_0
);
3187 dev
->vol
.migr_state
= 1;
3188 set_migr_type(dev
, migr_type
);
3189 dev
->vol
.curr_migr_unit
= 0;
3190 dest
= get_imsm_map(dev
, MAP_1
);
3192 /* duplicate and then set the target end state in map[0] */
3193 memcpy(dest
, src
, sizeof_imsm_map(src
));
3194 if ((migr_type
== MIGR_REBUILD
) ||
3195 (migr_type
== MIGR_GEN_MIGR
)) {
3199 for (i
= 0; i
< src
->num_members
; i
++) {
3200 ord
= __le32_to_cpu(src
->disk_ord_tbl
[i
]);
3201 set_imsm_ord_tbl_ent(src
, i
, ord_to_idx(ord
));
3205 if (migr_type
== MIGR_GEN_MIGR
)
3206 /* Clear migration record */
3207 memset(super
->migr_rec
, 0, sizeof(struct migr_record
));
3209 src
->map_state
= to_state
;
3212 static void end_migration(struct imsm_dev
*dev
, struct intel_super
*super
,
3215 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
3216 struct imsm_map
*prev
= get_imsm_map(dev
, dev
->vol
.migr_state
== 0 ?
3220 /* merge any IMSM_ORD_REBUILD bits that were not successfully
3221 * completed in the last migration.
3223 * FIXME add support for raid-level-migration
3225 if ((map_state
!= map
->map_state
) && (is_gen_migration(dev
) == 0) &&
3226 (prev
->map_state
!= IMSM_T_STATE_UNINITIALIZED
)) {
3227 /* when final map state is other than expected
3228 * merge maps (not for migration)
3232 for (i
= 0; i
< prev
->num_members
; i
++)
3233 for (j
= 0; j
< map
->num_members
; j
++)
3234 /* during online capacity expansion
3235 * disks position can be changed
3236 * if takeover is used
3238 if (ord_to_idx(map
->disk_ord_tbl
[j
]) ==
3239 ord_to_idx(prev
->disk_ord_tbl
[i
])) {
3240 map
->disk_ord_tbl
[j
] |=
3241 prev
->disk_ord_tbl
[i
];
3244 failed
= imsm_count_failed(super
, dev
, MAP_0
);
3245 map_state
= imsm_check_degraded(super
, dev
, failed
, MAP_0
);
3248 dev
->vol
.migr_state
= 0;
3249 set_migr_type(dev
, 0);
3250 dev
->vol
.curr_migr_unit
= 0;
3251 map
->map_state
= map_state
;
3255 static int parse_raid_devices(struct intel_super
*super
)
3258 struct imsm_dev
*dev_new
;
3259 size_t len
, len_migr
;
3261 size_t space_needed
= 0;
3262 struct imsm_super
*mpb
= super
->anchor
;
3264 for (i
= 0; i
< super
->anchor
->num_raid_devs
; i
++) {
3265 struct imsm_dev
*dev_iter
= __get_imsm_dev(super
->anchor
, i
);
3266 struct intel_dev
*dv
;
3268 len
= sizeof_imsm_dev(dev_iter
, 0);
3269 len_migr
= sizeof_imsm_dev(dev_iter
, 1);
3271 space_needed
+= len_migr
- len
;
3273 dv
= malloc(sizeof(*dv
));
3276 if (max_len
< len_migr
)
3278 if (max_len
> len_migr
)
3279 space_needed
+= max_len
- len_migr
;
3280 dev_new
= malloc(max_len
);
3285 imsm_copy_dev(dev_new
, dev_iter
);
3288 dv
->next
= super
->devlist
;
3289 super
->devlist
= dv
;
3292 /* ensure that super->buf is large enough when all raid devices
3295 if (__le32_to_cpu(mpb
->mpb_size
) + space_needed
> super
->len
) {
3298 len
= ROUND_UP(__le32_to_cpu(mpb
->mpb_size
) + space_needed
, 512);
3299 if (posix_memalign(&buf
, 512, len
) != 0)
3302 memcpy(buf
, super
->buf
, super
->len
);
3303 memset(buf
+ super
->len
, 0, len
- super
->len
);
3312 /* retrieve a pointer to the bbm log which starts after all raid devices */
3313 struct bbm_log
*__get_imsm_bbm_log(struct imsm_super
*mpb
)
3317 if (__le32_to_cpu(mpb
->bbm_log_size
)) {
3319 ptr
+= mpb
->mpb_size
- __le32_to_cpu(mpb
->bbm_log_size
);
3325 /*******************************************************************************
3326 * Function: check_mpb_migr_compatibility
3327 * Description: Function checks for unsupported migration features:
3328 * - migration optimization area (pba_of_lba0)
3329 * - descending reshape (ascending_migr)
3331 * super : imsm metadata information
3333 * 0 : migration is compatible
3334 * -1 : migration is not compatible
3335 ******************************************************************************/
3336 int check_mpb_migr_compatibility(struct intel_super
*super
)
3338 struct imsm_map
*map0
, *map1
;
3339 struct migr_record
*migr_rec
= super
->migr_rec
;
3342 for (i
= 0; i
< super
->anchor
->num_raid_devs
; i
++) {
3343 struct imsm_dev
*dev_iter
= __get_imsm_dev(super
->anchor
, i
);
3346 dev_iter
->vol
.migr_state
== 1 &&
3347 dev_iter
->vol
.migr_type
== MIGR_GEN_MIGR
) {
3348 /* This device is migrating */
3349 map0
= get_imsm_map(dev_iter
, MAP_0
);
3350 map1
= get_imsm_map(dev_iter
, MAP_1
);
3351 if (map0
->pba_of_lba0
!= map1
->pba_of_lba0
)
3352 /* migration optimization area was used */
3354 if (migr_rec
->ascending_migr
== 0
3355 && migr_rec
->dest_depth_per_unit
> 0)
3356 /* descending reshape not supported yet */
3363 static void __free_imsm(struct intel_super
*super
, int free_disks
);
3365 /* load_imsm_mpb - read matrix metadata
3366 * allocates super->mpb to be freed by free_imsm
3368 static int load_imsm_mpb(int fd
, struct intel_super
*super
, char *devname
)
3370 unsigned long long dsize
;
3371 unsigned long long sectors
;
3373 struct imsm_super
*anchor
;
3376 get_dev_size(fd
, NULL
, &dsize
);
3380 Name
": %s: device to small for imsm\n",
3385 if (lseek64(fd
, dsize
- (512 * 2), SEEK_SET
) < 0) {
3387 fprintf(stderr
, Name
3388 ": Cannot seek to anchor block on %s: %s\n",
3389 devname
, strerror(errno
));
3393 if (posix_memalign((void**)&anchor
, 512, 512) != 0) {
3396 Name
": Failed to allocate imsm anchor buffer"
3397 " on %s\n", devname
);
3400 if (read(fd
, anchor
, 512) != 512) {
3403 Name
": Cannot read anchor block on %s: %s\n",
3404 devname
, strerror(errno
));
3409 if (strncmp((char *) anchor
->sig
, MPB_SIGNATURE
, MPB_SIG_LEN
) != 0) {
3412 Name
": no IMSM anchor on %s\n", devname
);
3417 __free_imsm(super
, 0);
3418 /* reload capability and hba */
3420 /* capability and hba must be updated with new super allocation */
3421 find_intel_hba_capability(fd
, super
, devname
);
3422 super
->len
= ROUND_UP(anchor
->mpb_size
, 512);
3423 if (posix_memalign(&super
->buf
, 512, super
->len
) != 0) {
3426 Name
": unable to allocate %zu byte mpb buffer\n",
3431 memcpy(super
->buf
, anchor
, 512);
3433 sectors
= mpb_sectors(anchor
) - 1;
3436 if (posix_memalign(&super
->migr_rec_buf
, 512, MIGR_REC_BUF_SIZE
) != 0) {
3437 fprintf(stderr
, Name
3438 ": %s could not allocate migr_rec buffer\n", __func__
);
3444 check_sum
= __gen_imsm_checksum(super
->anchor
);
3445 if (check_sum
!= __le32_to_cpu(super
->anchor
->check_sum
)) {
3448 Name
": IMSM checksum %x != %x on %s\n",
3450 __le32_to_cpu(super
->anchor
->check_sum
),
3458 /* read the extended mpb */
3459 if (lseek64(fd
, dsize
- (512 * (2 + sectors
)), SEEK_SET
) < 0) {
3462 Name
": Cannot seek to extended mpb on %s: %s\n",
3463 devname
, strerror(errno
));
3467 if ((unsigned)read(fd
, super
->buf
+ 512, super
->len
- 512) != super
->len
- 512) {
3470 Name
": Cannot read extended mpb on %s: %s\n",
3471 devname
, strerror(errno
));
3475 check_sum
= __gen_imsm_checksum(super
->anchor
);
3476 if (check_sum
!= __le32_to_cpu(super
->anchor
->check_sum
)) {
3479 Name
": IMSM checksum %x != %x on %s\n",
3480 check_sum
, __le32_to_cpu(super
->anchor
->check_sum
),
3485 /* FIXME the BBM log is disk specific so we cannot use this global
3486 * buffer for all disks. Ok for now since we only look at the global
3487 * bbm_log_size parameter to gate assembly
3489 super
->bbm_log
= __get_imsm_bbm_log(super
->anchor
);
3494 static int read_imsm_migr_rec(int fd
, struct intel_super
*super
);
3497 load_and_parse_mpb(int fd
, struct intel_super
*super
, char *devname
, int keep_fd
)
3501 err
= load_imsm_mpb(fd
, super
, devname
);
3504 err
= load_imsm_disk(fd
, super
, devname
, keep_fd
);
3507 err
= parse_raid_devices(super
);
3512 static void __free_imsm_disk(struct dl
*d
)
3524 static void free_imsm_disks(struct intel_super
*super
)
3528 while (super
->disks
) {
3530 super
->disks
= d
->next
;
3531 __free_imsm_disk(d
);
3533 while (super
->disk_mgmt_list
) {
3534 d
= super
->disk_mgmt_list
;
3535 super
->disk_mgmt_list
= d
->next
;
3536 __free_imsm_disk(d
);
3538 while (super
->missing
) {
3540 super
->missing
= d
->next
;
3541 __free_imsm_disk(d
);
3546 /* free all the pieces hanging off of a super pointer */
3547 static void __free_imsm(struct intel_super
*super
, int free_disks
)
3549 struct intel_hba
*elem
, *next
;
3555 /* unlink capability description */
3557 if (super
->migr_rec_buf
) {
3558 free(super
->migr_rec_buf
);
3559 super
->migr_rec_buf
= NULL
;
3562 free_imsm_disks(super
);
3563 free_devlist(super
);
3567 free((void *)elem
->path
);
3575 static void free_imsm(struct intel_super
*super
)
3577 __free_imsm(super
, 1);
3581 static void free_super_imsm(struct supertype
*st
)
3583 struct intel_super
*super
= st
->sb
;
3592 static struct intel_super
*alloc_super(void)
3594 struct intel_super
*super
= malloc(sizeof(*super
));
3597 memset(super
, 0, sizeof(*super
));
3598 super
->current_vol
= -1;
3599 super
->create_offset
= ~((__u32
) 0);
3605 * find and allocate hba and OROM/EFI based on valid fd of RAID component device
3607 static int find_intel_hba_capability(int fd
, struct intel_super
*super
, char *devname
)
3609 struct sys_dev
*hba_name
;
3612 if ((fd
< 0) || check_env("IMSM_NO_PLATFORM")) {
3617 hba_name
= find_disk_attached_hba(fd
, NULL
);
3621 Name
": %s is not attached to Intel(R) RAID controller.\n",
3625 rv
= attach_hba_to_super(super
, hba_name
);
3628 struct intel_hba
*hba
= super
->hba
;
3630 fprintf(stderr
, Name
": %s is attached to Intel(R) %s RAID "
3631 "controller (%s),\n"
3632 " but the container is assigned to Intel(R) "
3633 "%s RAID controller (",
3636 hba_name
->pci_id
? : "Err!",
3637 get_sys_dev_type(hba_name
->type
));
3640 fprintf(stderr
, "%s", hba
->pci_id
? : "Err!");
3642 fprintf(stderr
, ", ");
3646 fprintf(stderr
, ").\n"
3647 " Mixing devices attached to different controllers "
3648 "is not allowed.\n");
3650 free_sys_dev(&hba_name
);
3653 super
->orom
= find_imsm_capability(hba_name
->type
);
3654 free_sys_dev(&hba_name
);
3660 /* find_missing - helper routine for load_super_imsm_all that identifies
3661 * disks that have disappeared from the system. This routine relies on
3662 * the mpb being uptodate, which it is at load time.
3664 static int find_missing(struct intel_super
*super
)
3667 struct imsm_super
*mpb
= super
->anchor
;
3669 struct imsm_disk
*disk
;
3671 for (i
= 0; i
< mpb
->num_disks
; i
++) {
3672 disk
= __get_imsm_disk(mpb
, i
);
3673 dl
= serial_to_dl(disk
->serial
, super
);
3677 dl
= malloc(sizeof(*dl
));
3683 dl
->devname
= strdup("missing");
3685 serialcpy(dl
->serial
, disk
->serial
);
3688 dl
->next
= super
->missing
;
3689 super
->missing
= dl
;
3696 static struct intel_disk
*disk_list_get(__u8
*serial
, struct intel_disk
*disk_list
)
3698 struct intel_disk
*idisk
= disk_list
;
3701 if (serialcmp(idisk
->disk
.serial
, serial
) == 0)
3703 idisk
= idisk
->next
;
3709 static int __prep_thunderdome(struct intel_super
**table
, int tbl_size
,
3710 struct intel_super
*super
,
3711 struct intel_disk
**disk_list
)
3713 struct imsm_disk
*d
= &super
->disks
->disk
;
3714 struct imsm_super
*mpb
= super
->anchor
;
3717 for (i
= 0; i
< tbl_size
; i
++) {
3718 struct imsm_super
*tbl_mpb
= table
[i
]->anchor
;
3719 struct imsm_disk
*tbl_d
= &table
[i
]->disks
->disk
;
3721 if (tbl_mpb
->family_num
== mpb
->family_num
) {
3722 if (tbl_mpb
->check_sum
== mpb
->check_sum
) {
3723 dprintf("%s: mpb from %d:%d matches %d:%d\n",
3724 __func__
, super
->disks
->major
,
3725 super
->disks
->minor
,
3726 table
[i
]->disks
->major
,
3727 table
[i
]->disks
->minor
);
3731 if (((is_configured(d
) && !is_configured(tbl_d
)) ||
3732 is_configured(d
) == is_configured(tbl_d
)) &&
3733 tbl_mpb
->generation_num
< mpb
->generation_num
) {
3734 /* current version of the mpb is a
3735 * better candidate than the one in
3736 * super_table, but copy over "cross
3737 * generational" status
3739 struct intel_disk
*idisk
;
3741 dprintf("%s: mpb from %d:%d replaces %d:%d\n",
3742 __func__
, super
->disks
->major
,
3743 super
->disks
->minor
,
3744 table
[i
]->disks
->major
,
3745 table
[i
]->disks
->minor
);
3747 idisk
= disk_list_get(tbl_d
->serial
, *disk_list
);
3748 if (idisk
&& is_failed(&idisk
->disk
))
3749 tbl_d
->status
|= FAILED_DISK
;
3752 struct intel_disk
*idisk
;
3753 struct imsm_disk
*disk
;
3755 /* tbl_mpb is more up to date, but copy
3756 * over cross generational status before
3759 disk
= __serial_to_disk(d
->serial
, mpb
, NULL
);
3760 if (disk
&& is_failed(disk
))
3761 d
->status
|= FAILED_DISK
;
3763 idisk
= disk_list_get(d
->serial
, *disk_list
);
3766 if (disk
&& is_configured(disk
))
3767 idisk
->disk
.status
|= CONFIGURED_DISK
;
3770 dprintf("%s: mpb from %d:%d prefer %d:%d\n",
3771 __func__
, super
->disks
->major
,
3772 super
->disks
->minor
,
3773 table
[i
]->disks
->major
,
3774 table
[i
]->disks
->minor
);
3782 table
[tbl_size
++] = super
;
3786 /* update/extend the merged list of imsm_disk records */
3787 for (j
= 0; j
< mpb
->num_disks
; j
++) {
3788 struct imsm_disk
*disk
= __get_imsm_disk(mpb
, j
);
3789 struct intel_disk
*idisk
;
3791 idisk
= disk_list_get(disk
->serial
, *disk_list
);
3793 idisk
->disk
.status
|= disk
->status
;
3794 if (is_configured(&idisk
->disk
) ||
3795 is_failed(&idisk
->disk
))
3796 idisk
->disk
.status
&= ~(SPARE_DISK
);
3798 idisk
= calloc(1, sizeof(*idisk
));
3801 idisk
->owner
= IMSM_UNKNOWN_OWNER
;
3802 idisk
->disk
= *disk
;
3803 idisk
->next
= *disk_list
;
3807 if (serialcmp(idisk
->disk
.serial
, d
->serial
) == 0)
3814 static struct intel_super
*
3815 validate_members(struct intel_super
*super
, struct intel_disk
*disk_list
,
3818 struct imsm_super
*mpb
= super
->anchor
;
3822 for (i
= 0; i
< mpb
->num_disks
; i
++) {
3823 struct imsm_disk
*disk
= __get_imsm_disk(mpb
, i
);
3824 struct intel_disk
*idisk
;
3826 idisk
= disk_list_get(disk
->serial
, disk_list
);
3828 if (idisk
->owner
== owner
||
3829 idisk
->owner
== IMSM_UNKNOWN_OWNER
)
3832 dprintf("%s: '%.16s' owner %d != %d\n",
3833 __func__
, disk
->serial
, idisk
->owner
,
3836 dprintf("%s: unknown disk %x [%d]: %.16s\n",
3837 __func__
, __le32_to_cpu(mpb
->family_num
), i
,
3843 if (ok_count
== mpb
->num_disks
)
3848 static void show_conflicts(__u32 family_num
, struct intel_super
*super_list
)
3850 struct intel_super
*s
;
3852 for (s
= super_list
; s
; s
= s
->next
) {
3853 if (family_num
!= s
->anchor
->family_num
)
3855 fprintf(stderr
, "Conflict, offlining family %#x on '%s'\n",
3856 __le32_to_cpu(family_num
), s
->disks
->devname
);
3860 static struct intel_super
*
3861 imsm_thunderdome(struct intel_super
**super_list
, int len
)
3863 struct intel_super
*super_table
[len
];
3864 struct intel_disk
*disk_list
= NULL
;
3865 struct intel_super
*champion
, *spare
;
3866 struct intel_super
*s
, **del
;
3871 memset(super_table
, 0, sizeof(super_table
));
3872 for (s
= *super_list
; s
; s
= s
->next
)
3873 tbl_size
= __prep_thunderdome(super_table
, tbl_size
, s
, &disk_list
);
3875 for (i
= 0; i
< tbl_size
; i
++) {
3876 struct imsm_disk
*d
;
3877 struct intel_disk
*idisk
;
3878 struct imsm_super
*mpb
= super_table
[i
]->anchor
;
3881 d
= &s
->disks
->disk
;
3883 /* 'd' must appear in merged disk list for its
3884 * configuration to be valid
3886 idisk
= disk_list_get(d
->serial
, disk_list
);
3887 if (idisk
&& idisk
->owner
== i
)
3888 s
= validate_members(s
, disk_list
, i
);
3893 dprintf("%s: marking family: %#x from %d:%d offline\n",
3894 __func__
, mpb
->family_num
,
3895 super_table
[i
]->disks
->major
,
3896 super_table
[i
]->disks
->minor
);
3900 /* This is where the mdadm implementation differs from the Windows
3901 * driver which has no strict concept of a container. We can only
3902 * assemble one family from a container, so when returning a prodigal
3903 * array member to this system the code will not be able to disambiguate
3904 * the container contents that should be assembled ("foreign" versus
3905 * "local"). It requires user intervention to set the orig_family_num
3906 * to a new value to establish a new container. The Windows driver in
3907 * this situation fixes up the volume name in place and manages the
3908 * foreign array as an independent entity.
3913 for (i
= 0; i
< tbl_size
; i
++) {
3914 struct intel_super
*tbl_ent
= super_table
[i
];
3920 if (tbl_ent
->anchor
->num_raid_devs
== 0) {
3925 if (s
&& !is_spare
) {
3926 show_conflicts(tbl_ent
->anchor
->family_num
, *super_list
);
3928 } else if (!s
&& !is_spare
)
3941 fprintf(stderr
, "Chose family %#x on '%s', "
3942 "assemble conflicts to new container with '--update=uuid'\n",
3943 __le32_to_cpu(s
->anchor
->family_num
), s
->disks
->devname
);
3945 /* collect all dl's onto 'champion', and update them to
3946 * champion's version of the status
3948 for (s
= *super_list
; s
; s
= s
->next
) {
3949 struct imsm_super
*mpb
= champion
->anchor
;
3950 struct dl
*dl
= s
->disks
;
3955 for (i
= 0; i
< mpb
->num_disks
; i
++) {
3956 struct imsm_disk
*disk
;
3958 disk
= __serial_to_disk(dl
->serial
, mpb
, &dl
->index
);
3961 /* only set index on disks that are a member of
3962 * a populated contianer, i.e. one with
3965 if (is_failed(&dl
->disk
))
3967 else if (is_spare(&dl
->disk
))
3973 if (i
>= mpb
->num_disks
) {
3974 struct intel_disk
*idisk
;
3976 idisk
= disk_list_get(dl
->serial
, disk_list
);
3977 if (idisk
&& is_spare(&idisk
->disk
) &&
3978 !is_failed(&idisk
->disk
) && !is_configured(&idisk
->disk
))
3986 dl
->next
= champion
->disks
;
3987 champion
->disks
= dl
;
3991 /* delete 'champion' from super_list */
3992 for (del
= super_list
; *del
; ) {
3993 if (*del
== champion
) {
3994 *del
= (*del
)->next
;
3997 del
= &(*del
)->next
;
3999 champion
->next
= NULL
;
4003 struct intel_disk
*idisk
= disk_list
;
4005 disk_list
= disk_list
->next
;
4012 static int load_super_imsm_all(struct supertype
*st
, int fd
, void **sbp
,
4016 struct intel_super
*super_list
= NULL
;
4017 struct intel_super
*super
= NULL
;
4018 int devnum
= fd2devnum(fd
);
4024 /* check if 'fd' an opened container */
4025 sra
= sysfs_read(fd
, 0, GET_LEVEL
|GET_VERSION
|GET_DEVS
|GET_STATE
);
4029 if (sra
->array
.major_version
!= -1 ||
4030 sra
->array
.minor_version
!= -2 ||
4031 strcmp(sra
->text_version
, "imsm") != 0) {
4036 for (sd
= sra
->devs
, i
= 0; sd
; sd
= sd
->next
, i
++) {
4037 struct intel_super
*s
= alloc_super();
4045 s
->next
= super_list
;
4049 sprintf(nm
, "%d:%d", sd
->disk
.major
, sd
->disk
.minor
);
4050 dfd
= dev_open(nm
, O_RDWR
);
4054 rv
= find_intel_hba_capability(dfd
, s
, devname
);
4055 /* no orom/efi or non-intel hba of the disk */
4059 err
= load_and_parse_mpb(dfd
, s
, NULL
, 1);
4061 /* retry the load if we might have raced against mdmon */
4062 if (err
== 3 && mdmon_running(devnum
))
4063 for (retry
= 0; retry
< 3; retry
++) {
4065 err
= load_and_parse_mpb(dfd
, s
, NULL
, 1);
4073 /* all mpbs enter, maybe one leaves */
4074 super
= imsm_thunderdome(&super_list
, i
);
4080 if (find_missing(super
) != 0) {
4086 /* load migration record */
4087 err
= load_imsm_migr_rec(super
, NULL
);
4089 /* migration is in progress,
4090 * but migr_rec cannot be loaded,
4096 /* Check migration compatibility */
4097 if ((err
== 0) && (check_mpb_migr_compatibility(super
) != 0)) {
4098 fprintf(stderr
, Name
": Unsupported migration detected");
4100 fprintf(stderr
, " on %s\n", devname
);
4102 fprintf(stderr
, " (IMSM).\n");
4111 while (super_list
) {
4112 struct intel_super
*s
= super_list
;
4114 super_list
= super_list
->next
;
4123 st
->container_dev
= devnum
;
4124 if (err
== 0 && st
->ss
== NULL
) {
4125 st
->ss
= &super_imsm
;
4126 st
->minor_version
= 0;
4127 st
->max_devs
= IMSM_MAX_DEVICES
;
4132 static int load_container_imsm(struct supertype
*st
, int fd
, char *devname
)
4134 return load_super_imsm_all(st
, fd
, &st
->sb
, devname
);
4138 static int load_super_imsm(struct supertype
*st
, int fd
, char *devname
)
4140 struct intel_super
*super
;
4143 if (test_partition(fd
))
4144 /* IMSM not allowed on partitions */
4147 free_super_imsm(st
);
4149 super
= alloc_super();
4152 Name
": malloc of %zu failed.\n",
4156 /* Load hba and capabilities if they exist.
4157 * But do not preclude loading metadata in case capabilities or hba are
4158 * non-compliant and ignore_hw_compat is set.
4160 rv
= find_intel_hba_capability(fd
, super
, devname
);
4161 /* no orom/efi or non-intel hba of the disk */
4162 if ((rv
!= 0) && (st
->ignore_hw_compat
== 0)) {
4165 Name
": No OROM/EFI properties for %s\n", devname
);
4169 rv
= load_and_parse_mpb(fd
, super
, devname
, 0);
4174 Name
": Failed to load all information "
4175 "sections on %s\n", devname
);
4181 if (st
->ss
== NULL
) {
4182 st
->ss
= &super_imsm
;
4183 st
->minor_version
= 0;
4184 st
->max_devs
= IMSM_MAX_DEVICES
;
4187 /* load migration record */
4188 if (load_imsm_migr_rec(super
, NULL
) == 0) {
4189 /* Check for unsupported migration features */
4190 if (check_mpb_migr_compatibility(super
) != 0) {
4192 Name
": Unsupported migration detected");
4194 fprintf(stderr
, " on %s\n", devname
);
4196 fprintf(stderr
, " (IMSM).\n");
4204 static __u16
info_to_blocks_per_strip(mdu_array_info_t
*info
)
4206 if (info
->level
== 1)
4208 return info
->chunk_size
>> 9;
4211 static __u32
info_to_num_data_stripes(mdu_array_info_t
*info
, int num_domains
)
4215 num_stripes
= (info
->size
* 2) / info_to_blocks_per_strip(info
);
4216 num_stripes
/= num_domains
;
4221 static __u32
info_to_blocks_per_member(mdu_array_info_t
*info
)
4223 if (info
->level
== 1)
4224 return info
->size
* 2;
4226 return (info
->size
* 2) & ~(info_to_blocks_per_strip(info
) - 1);
4229 static void imsm_update_version_info(struct intel_super
*super
)
4231 /* update the version and attributes */
4232 struct imsm_super
*mpb
= super
->anchor
;
4234 struct imsm_dev
*dev
;
4235 struct imsm_map
*map
;
4238 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
4239 dev
= get_imsm_dev(super
, i
);
4240 map
= get_imsm_map(dev
, MAP_0
);
4241 if (__le32_to_cpu(dev
->size_high
) > 0)
4242 mpb
->attributes
|= MPB_ATTRIB_2TB
;
4244 /* FIXME detect when an array spans a port multiplier */
4246 mpb
->attributes
|= MPB_ATTRIB_PM
;
4249 if (mpb
->num_raid_devs
> 1 ||
4250 mpb
->attributes
!= MPB_ATTRIB_CHECKSUM_VERIFY
) {
4251 version
= MPB_VERSION_ATTRIBS
;
4252 switch (get_imsm_raid_level(map
)) {
4253 case 0: mpb
->attributes
|= MPB_ATTRIB_RAID0
; break;
4254 case 1: mpb
->attributes
|= MPB_ATTRIB_RAID1
; break;
4255 case 10: mpb
->attributes
|= MPB_ATTRIB_RAID10
; break;
4256 case 5: mpb
->attributes
|= MPB_ATTRIB_RAID5
; break;
4259 if (map
->num_members
>= 5)
4260 version
= MPB_VERSION_5OR6_DISK_ARRAY
;
4261 else if (dev
->status
== DEV_CLONE_N_GO
)
4262 version
= MPB_VERSION_CNG
;
4263 else if (get_imsm_raid_level(map
) == 5)
4264 version
= MPB_VERSION_RAID5
;
4265 else if (map
->num_members
>= 3)
4266 version
= MPB_VERSION_3OR4_DISK_ARRAY
;
4267 else if (get_imsm_raid_level(map
) == 1)
4268 version
= MPB_VERSION_RAID1
;
4270 version
= MPB_VERSION_RAID0
;
4272 strcpy(((char *) mpb
->sig
) + strlen(MPB_SIGNATURE
), version
);
4276 static int check_name(struct intel_super
*super
, char *name
, int quiet
)
4278 struct imsm_super
*mpb
= super
->anchor
;
4279 char *reason
= NULL
;
4282 if (strlen(name
) > MAX_RAID_SERIAL_LEN
)
4283 reason
= "must be 16 characters or less";
4285 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
4286 struct imsm_dev
*dev
= get_imsm_dev(super
, i
);
4288 if (strncmp((char *) dev
->volume
, name
, MAX_RAID_SERIAL_LEN
) == 0) {
4289 reason
= "already exists";
4294 if (reason
&& !quiet
)
4295 fprintf(stderr
, Name
": imsm volume name %s\n", reason
);
4300 static int init_super_imsm_volume(struct supertype
*st
, mdu_array_info_t
*info
,
4301 unsigned long long size
, char *name
,
4302 char *homehost
, int *uuid
)
4304 /* We are creating a volume inside a pre-existing container.
4305 * so st->sb is already set.
4307 struct intel_super
*super
= st
->sb
;
4308 struct imsm_super
*mpb
= super
->anchor
;
4309 struct intel_dev
*dv
;
4310 struct imsm_dev
*dev
;
4311 struct imsm_vol
*vol
;
4312 struct imsm_map
*map
;
4313 int idx
= mpb
->num_raid_devs
;
4315 unsigned long long array_blocks
;
4316 size_t size_old
, size_new
;
4317 __u32 num_data_stripes
;
4319 if (super
->orom
&& mpb
->num_raid_devs
>= super
->orom
->vpa
) {
4320 fprintf(stderr
, Name
": This imsm-container already has the "
4321 "maximum of %d volumes\n", super
->orom
->vpa
);
4325 /* ensure the mpb is large enough for the new data */
4326 size_old
= __le32_to_cpu(mpb
->mpb_size
);
4327 size_new
= disks_to_mpb_size(info
->nr_disks
);
4328 if (size_new
> size_old
) {
4330 size_t size_round
= ROUND_UP(size_new
, 512);
4332 if (posix_memalign(&mpb_new
, 512, size_round
) != 0) {
4333 fprintf(stderr
, Name
": could not allocate new mpb\n");
4336 if (posix_memalign(&super
->migr_rec_buf
, 512,
4337 MIGR_REC_BUF_SIZE
) != 0) {
4338 fprintf(stderr
, Name
4339 ": %s could not allocate migr_rec buffer\n",
4346 memcpy(mpb_new
, mpb
, size_old
);
4349 super
->anchor
= mpb_new
;
4350 mpb
->mpb_size
= __cpu_to_le32(size_new
);
4351 memset(mpb_new
+ size_old
, 0, size_round
- size_old
);
4353 super
->current_vol
= idx
;
4355 /* handle 'failed_disks' by either:
4356 * a) create dummy disk entries in the table if this the first
4357 * volume in the array. We add them here as this is the only
4358 * opportunity to add them. add_to_super_imsm_volume()
4359 * handles the non-failed disks and continues incrementing
4361 * b) validate that 'failed_disks' matches the current number
4362 * of missing disks if the container is populated
4364 if (super
->current_vol
== 0) {
4366 for (i
= 0; i
< info
->failed_disks
; i
++) {
4367 struct imsm_disk
*disk
;
4370 disk
= __get_imsm_disk(mpb
, i
);
4371 disk
->status
= CONFIGURED_DISK
| FAILED_DISK
;
4372 disk
->scsi_id
= __cpu_to_le32(~(__u32
)0);
4373 snprintf((char *) disk
->serial
, MAX_RAID_SERIAL_LEN
,
4376 find_missing(super
);
4381 for (d
= super
->missing
; d
; d
= d
->next
)
4383 if (info
->failed_disks
> missing
) {
4384 fprintf(stderr
, Name
": unable to add 'missing' disk to container\n");
4389 if (!check_name(super
, name
, 0))
4391 dv
= malloc(sizeof(*dv
));
4393 fprintf(stderr
, Name
": failed to allocate device list entry\n");
4396 dev
= calloc(1, sizeof(*dev
) + sizeof(__u32
) * (info
->raid_disks
- 1));
4399 fprintf(stderr
, Name
": could not allocate raid device\n");
4403 strncpy((char *) dev
->volume
, name
, MAX_RAID_SERIAL_LEN
);
4404 if (info
->level
== 1)
4405 array_blocks
= info_to_blocks_per_member(info
);
4407 array_blocks
= calc_array_size(info
->level
, info
->raid_disks
,
4408 info
->layout
, info
->chunk_size
,
4410 /* round array size down to closest MB */
4411 array_blocks
= (array_blocks
>> SECT_PER_MB_SHIFT
) << SECT_PER_MB_SHIFT
;
4413 dev
->size_low
= __cpu_to_le32((__u32
) array_blocks
);
4414 dev
->size_high
= __cpu_to_le32((__u32
) (array_blocks
>> 32));
4415 dev
->status
= (DEV_READ_COALESCING
| DEV_WRITE_COALESCING
);
4417 vol
->migr_state
= 0;
4418 set_migr_type(dev
, MIGR_INIT
);
4419 vol
->dirty
= !info
->state
;
4420 vol
->curr_migr_unit
= 0;
4421 map
= get_imsm_map(dev
, MAP_0
);
4422 map
->pba_of_lba0
= __cpu_to_le32(super
->create_offset
);
4423 map
->blocks_per_member
= __cpu_to_le32(info_to_blocks_per_member(info
));
4424 map
->blocks_per_strip
= __cpu_to_le16(info_to_blocks_per_strip(info
));
4425 map
->failed_disk_num
= ~0;
4426 if (info
->level
> 0)
4427 map
->map_state
= IMSM_T_STATE_UNINITIALIZED
;
4429 map
->map_state
= info
->failed_disks
? IMSM_T_STATE_FAILED
:
4430 IMSM_T_STATE_NORMAL
;
4433 if (info
->level
== 1 && info
->raid_disks
> 2) {
4436 fprintf(stderr
, Name
": imsm does not support more than 2 disks"
4437 "in a raid1 volume\n");
4441 map
->raid_level
= info
->level
;
4442 if (info
->level
== 10) {
4443 map
->raid_level
= 1;
4444 map
->num_domains
= info
->raid_disks
/ 2;
4445 } else if (info
->level
== 1)
4446 map
->num_domains
= info
->raid_disks
;
4448 map
->num_domains
= 1;
4450 num_data_stripes
= info_to_num_data_stripes(info
, map
->num_domains
);
4451 map
->num_data_stripes
= __cpu_to_le32(num_data_stripes
);
4453 map
->num_members
= info
->raid_disks
;
4454 for (i
= 0; i
< map
->num_members
; i
++) {
4455 /* initialized in add_to_super */
4456 set_imsm_ord_tbl_ent(map
, i
, IMSM_ORD_REBUILD
);
4458 mpb
->num_raid_devs
++;
4461 dv
->index
= super
->current_vol
;
4462 dv
->next
= super
->devlist
;
4463 super
->devlist
= dv
;
4465 imsm_update_version_info(super
);
4470 static int init_super_imsm(struct supertype
*st
, mdu_array_info_t
*info
,
4471 unsigned long long size
, char *name
,
4472 char *homehost
, int *uuid
)
4474 /* This is primarily called by Create when creating a new array.
4475 * We will then get add_to_super called for each component, and then
4476 * write_init_super called to write it out to each device.
4477 * For IMSM, Create can create on fresh devices or on a pre-existing
4479 * To create on a pre-existing array a different method will be called.
4480 * This one is just for fresh drives.
4482 struct intel_super
*super
;
4483 struct imsm_super
*mpb
;
4488 return init_super_imsm_volume(st
, info
, size
, name
, homehost
, uuid
);
4491 mpb_size
= disks_to_mpb_size(info
->nr_disks
);
4495 super
= alloc_super();
4496 if (super
&& posix_memalign(&super
->buf
, 512, mpb_size
) != 0) {
4501 fprintf(stderr
, Name
4502 ": %s could not allocate superblock\n", __func__
);
4505 if (posix_memalign(&super
->migr_rec_buf
, 512, MIGR_REC_BUF_SIZE
) != 0) {
4506 fprintf(stderr
, Name
4507 ": %s could not allocate migr_rec buffer\n", __func__
);
4512 memset(super
->buf
, 0, mpb_size
);
4514 mpb
->mpb_size
= __cpu_to_le32(mpb_size
);
4518 /* zeroing superblock */
4522 mpb
->attributes
= MPB_ATTRIB_CHECKSUM_VERIFY
;
4524 version
= (char *) mpb
->sig
;
4525 strcpy(version
, MPB_SIGNATURE
);
4526 version
+= strlen(MPB_SIGNATURE
);
4527 strcpy(version
, MPB_VERSION_RAID0
);
4533 static int add_to_super_imsm_volume(struct supertype
*st
, mdu_disk_info_t
*dk
,
4534 int fd
, char *devname
)
4536 struct intel_super
*super
= st
->sb
;
4537 struct imsm_super
*mpb
= super
->anchor
;
4538 struct imsm_disk
*_disk
;
4539 struct imsm_dev
*dev
;
4540 struct imsm_map
*map
;
4544 dev
= get_imsm_dev(super
, super
->current_vol
);
4545 map
= get_imsm_map(dev
, MAP_0
);
4547 if (! (dk
->state
& (1<<MD_DISK_SYNC
))) {
4548 fprintf(stderr
, Name
": %s: Cannot add spare devices to IMSM volume\n",
4554 /* we're doing autolayout so grab the pre-marked (in
4555 * validate_geometry) raid_disk
4557 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
4558 if (dl
->raiddisk
== dk
->raid_disk
)
4561 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
4562 if (dl
->major
== dk
->major
&&
4563 dl
->minor
== dk
->minor
)
4568 fprintf(stderr
, Name
": %s is not a member of the same container\n", devname
);
4572 /* add a pristine spare to the metadata */
4573 if (dl
->index
< 0) {
4574 dl
->index
= super
->anchor
->num_disks
;
4575 super
->anchor
->num_disks
++;
4577 /* Check the device has not already been added */
4578 slot
= get_imsm_disk_slot(map
, dl
->index
);
4580 (get_imsm_ord_tbl_ent(dev
, slot
, MAP_X
) & IMSM_ORD_REBUILD
) == 0) {
4581 fprintf(stderr
, Name
": %s has been included in this array twice\n",
4585 set_imsm_ord_tbl_ent(map
, dk
->raid_disk
, dl
->index
);
4586 dl
->disk
.status
= CONFIGURED_DISK
;
4588 /* update size of 'missing' disks to be at least as large as the
4589 * largest acitve member (we only have dummy missing disks when
4590 * creating the first volume)
4592 if (super
->current_vol
== 0) {
4593 for (df
= super
->missing
; df
; df
= df
->next
) {
4594 if (dl
->disk
.total_blocks
> df
->disk
.total_blocks
)
4595 df
->disk
.total_blocks
= dl
->disk
.total_blocks
;
4596 _disk
= __get_imsm_disk(mpb
, df
->index
);
4601 /* refresh unset/failed slots to point to valid 'missing' entries */
4602 for (df
= super
->missing
; df
; df
= df
->next
)
4603 for (slot
= 0; slot
< mpb
->num_disks
; slot
++) {
4604 __u32 ord
= get_imsm_ord_tbl_ent(dev
, slot
, MAP_X
);
4606 if ((ord
& IMSM_ORD_REBUILD
) == 0)
4608 set_imsm_ord_tbl_ent(map
, slot
, df
->index
| IMSM_ORD_REBUILD
);
4609 if (is_gen_migration(dev
)) {
4610 struct imsm_map
*map2
= get_imsm_map(dev
,
4612 int slot2
= get_imsm_disk_slot(map2
, df
->index
);
4613 if ((slot2
< map2
->num_members
) &&
4615 __u32 ord2
= get_imsm_ord_tbl_ent(dev
,
4618 if ((unsigned)df
->index
==
4620 set_imsm_ord_tbl_ent(map2
,
4626 dprintf("set slot:%d to missing disk:%d\n", slot
, df
->index
);
4630 /* if we are creating the first raid device update the family number */
4631 if (super
->current_vol
== 0) {
4633 struct imsm_dev
*_dev
= __get_imsm_dev(mpb
, 0);
4635 _disk
= __get_imsm_disk(mpb
, dl
->index
);
4636 if (!_dev
|| !_disk
) {
4637 fprintf(stderr
, Name
": BUG mpb setup error\n");
4643 sum
+= __gen_imsm_checksum(mpb
);
4644 mpb
->family_num
= __cpu_to_le32(sum
);
4645 mpb
->orig_family_num
= mpb
->family_num
;
4647 super
->current_disk
= dl
;
4652 * Function marks disk as spare and restores disk serial
4653 * in case it was previously marked as failed by takeover operation
4655 * -1 : critical error
4656 * 0 : disk is marked as spare but serial is not set
4659 int mark_spare(struct dl
*disk
)
4661 __u8 serial
[MAX_RAID_SERIAL_LEN
];
4668 if (!imsm_read_serial(disk
->fd
, NULL
, serial
)) {
4669 /* Restore disk serial number, because takeover marks disk
4670 * as failed and adds to serial ':0' before it becomes
4673 serialcpy(disk
->serial
, serial
);
4674 serialcpy(disk
->disk
.serial
, serial
);
4677 disk
->disk
.status
= SPARE_DISK
;
4683 static int add_to_super_imsm(struct supertype
*st
, mdu_disk_info_t
*dk
,
4684 int fd
, char *devname
)
4686 struct intel_super
*super
= st
->sb
;
4688 unsigned long long size
;
4693 /* If we are on an RAID enabled platform check that the disk is
4694 * attached to the raid controller.
4695 * We do not need to test disks attachment for container based additions,
4696 * they shall be already tested when container was created/assembled.
4698 rv
= find_intel_hba_capability(fd
, super
, devname
);
4699 /* no orom/efi or non-intel hba of the disk */
4701 dprintf("capability: %p fd: %d ret: %d\n",
4702 super
->orom
, fd
, rv
);
4706 if (super
->current_vol
>= 0)
4707 return add_to_super_imsm_volume(st
, dk
, fd
, devname
);
4710 dd
= malloc(sizeof(*dd
));
4713 Name
": malloc failed %s:%d.\n", __func__
, __LINE__
);
4716 memset(dd
, 0, sizeof(*dd
));
4717 dd
->major
= major(stb
.st_rdev
);
4718 dd
->minor
= minor(stb
.st_rdev
);
4719 dd
->devname
= devname
? strdup(devname
) : NULL
;
4722 dd
->action
= DISK_ADD
;
4723 rv
= imsm_read_serial(fd
, devname
, dd
->serial
);
4726 Name
": failed to retrieve scsi serial, aborting\n");
4731 get_dev_size(fd
, NULL
, &size
);
4733 serialcpy(dd
->disk
.serial
, dd
->serial
);
4734 dd
->disk
.total_blocks
= __cpu_to_le32(size
);
4736 if (sysfs_disk_to_scsi_id(fd
, &id
) == 0)
4737 dd
->disk
.scsi_id
= __cpu_to_le32(id
);
4739 dd
->disk
.scsi_id
= __cpu_to_le32(0);
4741 if (st
->update_tail
) {
4742 dd
->next
= super
->disk_mgmt_list
;
4743 super
->disk_mgmt_list
= dd
;
4745 dd
->next
= super
->disks
;
4747 super
->updates_pending
++;
4754 static int remove_from_super_imsm(struct supertype
*st
, mdu_disk_info_t
*dk
)
4756 struct intel_super
*super
= st
->sb
;
4759 /* remove from super works only in mdmon - for communication
4760 * manager - monitor. Check if communication memory buffer
4763 if (!st
->update_tail
) {
4765 Name
": %s shall be used in mdmon context only"
4766 "(line %d).\n", __func__
, __LINE__
);
4769 dd
= malloc(sizeof(*dd
));
4772 Name
": malloc failed %s:%d.\n", __func__
, __LINE__
);
4775 memset(dd
, 0, sizeof(*dd
));
4776 dd
->major
= dk
->major
;
4777 dd
->minor
= dk
->minor
;
4780 dd
->action
= DISK_REMOVE
;
4782 dd
->next
= super
->disk_mgmt_list
;
4783 super
->disk_mgmt_list
= dd
;
4789 static int store_imsm_mpb(int fd
, struct imsm_super
*mpb
);
4793 struct imsm_super anchor
;
4794 } spare_record
__attribute__ ((aligned(512)));
4796 /* spare records have their own family number and do not have any defined raid
4799 static int write_super_imsm_spares(struct intel_super
*super
, int doclose
)
4801 struct imsm_super
*mpb
= super
->anchor
;
4802 struct imsm_super
*spare
= &spare_record
.anchor
;
4806 spare
->mpb_size
= __cpu_to_le32(sizeof(struct imsm_super
)),
4807 spare
->generation_num
= __cpu_to_le32(1UL),
4808 spare
->attributes
= MPB_ATTRIB_CHECKSUM_VERIFY
;
4809 spare
->num_disks
= 1,
4810 spare
->num_raid_devs
= 0,
4811 spare
->cache_size
= mpb
->cache_size
,
4812 spare
->pwr_cycle_count
= __cpu_to_le32(1),
4814 snprintf((char *) spare
->sig
, MAX_SIGNATURE_LENGTH
,
4815 MPB_SIGNATURE MPB_VERSION_RAID0
);
4817 for (d
= super
->disks
; d
; d
= d
->next
) {
4821 spare
->disk
[0] = d
->disk
;
4822 sum
= __gen_imsm_checksum(spare
);
4823 spare
->family_num
= __cpu_to_le32(sum
);
4824 spare
->orig_family_num
= 0;
4825 sum
= __gen_imsm_checksum(spare
);
4826 spare
->check_sum
= __cpu_to_le32(sum
);
4828 if (store_imsm_mpb(d
->fd
, spare
)) {
4829 fprintf(stderr
, "%s: failed for device %d:%d %s\n",
4830 __func__
, d
->major
, d
->minor
, strerror(errno
));
4842 static int write_super_imsm(struct supertype
*st
, int doclose
)
4844 struct intel_super
*super
= st
->sb
;
4845 struct imsm_super
*mpb
= super
->anchor
;
4851 __u32 mpb_size
= sizeof(struct imsm_super
) - sizeof(struct imsm_disk
);
4853 int clear_migration_record
= 1;
4855 /* 'generation' is incremented everytime the metadata is written */
4856 generation
= __le32_to_cpu(mpb
->generation_num
);
4858 mpb
->generation_num
= __cpu_to_le32(generation
);
4860 /* fix up cases where previous mdadm releases failed to set
4863 if (mpb
->orig_family_num
== 0)
4864 mpb
->orig_family_num
= mpb
->family_num
;
4866 for (d
= super
->disks
; d
; d
= d
->next
) {
4870 mpb
->disk
[d
->index
] = d
->disk
;
4874 for (d
= super
->missing
; d
; d
= d
->next
) {
4875 mpb
->disk
[d
->index
] = d
->disk
;
4878 mpb
->num_disks
= num_disks
;
4879 mpb_size
+= sizeof(struct imsm_disk
) * mpb
->num_disks
;
4881 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
4882 struct imsm_dev
*dev
= __get_imsm_dev(mpb
, i
);
4883 struct imsm_dev
*dev2
= get_imsm_dev(super
, i
);
4885 imsm_copy_dev(dev
, dev2
);
4886 mpb_size
+= sizeof_imsm_dev(dev
, 0);
4888 if (is_gen_migration(dev2
))
4889 clear_migration_record
= 0;
4891 mpb_size
+= __le32_to_cpu(mpb
->bbm_log_size
);
4892 mpb
->mpb_size
= __cpu_to_le32(mpb_size
);
4894 /* recalculate checksum */
4895 sum
= __gen_imsm_checksum(mpb
);
4896 mpb
->check_sum
= __cpu_to_le32(sum
);
4898 if (clear_migration_record
)
4899 memset(super
->migr_rec_buf
, 0, MIGR_REC_BUF_SIZE
);
4901 /* write the mpb for disks that compose raid devices */
4902 for (d
= super
->disks
; d
; d
= d
->next
) {
4903 if (d
->index
< 0 || is_failed(&d
->disk
))
4905 if (store_imsm_mpb(d
->fd
, mpb
))
4906 fprintf(stderr
, "%s: failed for device %d:%d %s\n",
4907 __func__
, d
->major
, d
->minor
, strerror(errno
));
4908 if (clear_migration_record
) {
4909 unsigned long long dsize
;
4911 get_dev_size(d
->fd
, NULL
, &dsize
);
4912 if (lseek64(d
->fd
, dsize
- 512, SEEK_SET
) >= 0) {
4913 if (write(d
->fd
, super
->migr_rec_buf
,
4914 MIGR_REC_BUF_SIZE
) != MIGR_REC_BUF_SIZE
)
4915 perror("Write migr_rec failed");
4925 return write_super_imsm_spares(super
, doclose
);
4931 static int create_array(struct supertype
*st
, int dev_idx
)
4934 struct imsm_update_create_array
*u
;
4935 struct intel_super
*super
= st
->sb
;
4936 struct imsm_dev
*dev
= get_imsm_dev(super
, dev_idx
);
4937 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
4938 struct disk_info
*inf
;
4939 struct imsm_disk
*disk
;
4942 len
= sizeof(*u
) - sizeof(*dev
) + sizeof_imsm_dev(dev
, 0) +
4943 sizeof(*inf
) * map
->num_members
;
4946 fprintf(stderr
, "%s: failed to allocate update buffer\n",
4951 u
->type
= update_create_array
;
4952 u
->dev_idx
= dev_idx
;
4953 imsm_copy_dev(&u
->dev
, dev
);
4954 inf
= get_disk_info(u
);
4955 for (i
= 0; i
< map
->num_members
; i
++) {
4956 int idx
= get_imsm_disk_idx(dev
, i
, MAP_X
);
4958 disk
= get_imsm_disk(super
, idx
);
4959 serialcpy(inf
[i
].serial
, disk
->serial
);
4961 append_metadata_update(st
, u
, len
);
4966 static int mgmt_disk(struct supertype
*st
)
4968 struct intel_super
*super
= st
->sb
;
4970 struct imsm_update_add_remove_disk
*u
;
4972 if (!super
->disk_mgmt_list
)
4978 fprintf(stderr
, "%s: failed to allocate update buffer\n",
4983 u
->type
= update_add_remove_disk
;
4984 append_metadata_update(st
, u
, len
);
4989 static int write_init_super_imsm(struct supertype
*st
)
4991 struct intel_super
*super
= st
->sb
;
4992 int current_vol
= super
->current_vol
;
4994 /* we are done with current_vol reset it to point st at the container */
4995 super
->current_vol
= -1;
4997 if (st
->update_tail
) {
4998 /* queue the recently created array / added disk
4999 * as a metadata update */
5002 /* determine if we are creating a volume or adding a disk */
5003 if (current_vol
< 0) {
5004 /* in the mgmt (add/remove) disk case we are running
5005 * in mdmon context, so don't close fd's
5007 return mgmt_disk(st
);
5009 rv
= create_array(st
, current_vol
);
5014 for (d
= super
->disks
; d
; d
= d
->next
)
5015 Kill(d
->devname
, NULL
, 0, 1, 1);
5016 return write_super_imsm(st
, 1);
5021 static int store_super_imsm(struct supertype
*st
, int fd
)
5023 struct intel_super
*super
= st
->sb
;
5024 struct imsm_super
*mpb
= super
? super
->anchor
: NULL
;
5030 return store_imsm_mpb(fd
, mpb
);
5036 static int imsm_bbm_log_size(struct imsm_super
*mpb
)
5038 return __le32_to_cpu(mpb
->bbm_log_size
);
5042 static int validate_geometry_imsm_container(struct supertype
*st
, int level
,
5043 int layout
, int raiddisks
, int chunk
,
5044 unsigned long long size
, char *dev
,
5045 unsigned long long *freesize
,
5049 unsigned long long ldsize
;
5050 struct intel_super
*super
=NULL
;
5053 if (level
!= LEVEL_CONTAINER
)
5058 fd
= open(dev
, O_RDONLY
|O_EXCL
, 0);
5061 fprintf(stderr
, Name
": imsm: Cannot open %s: %s\n",
5062 dev
, strerror(errno
));
5065 if (!get_dev_size(fd
, dev
, &ldsize
)) {
5070 /* capabilities retrieve could be possible
5071 * note that there is no fd for the disks in array.
5073 super
= alloc_super();
5076 Name
": malloc of %zu failed.\n",
5082 rv
= find_intel_hba_capability(fd
, super
, verbose
? dev
: NULL
);
5086 fd2devname(fd
, str
);
5087 dprintf("validate_geometry_imsm_container: fd: %d %s orom: %p rv: %d raiddisk: %d\n",
5088 fd
, str
, super
->orom
, rv
, raiddisks
);
5090 /* no orom/efi or non-intel hba of the disk */
5096 if (super
->orom
&& raiddisks
> super
->orom
->tds
) {
5098 fprintf(stderr
, Name
": %d exceeds maximum number of"
5099 " platform supported disks: %d\n",
5100 raiddisks
, super
->orom
->tds
);
5106 *freesize
= avail_size_imsm(st
, ldsize
>> 9);
5112 static unsigned long long find_size(struct extent
*e
, int *idx
, int num_extents
)
5114 const unsigned long long base_start
= e
[*idx
].start
;
5115 unsigned long long end
= base_start
+ e
[*idx
].size
;
5118 if (base_start
== end
)
5122 for (i
= *idx
; i
< num_extents
; i
++) {
5123 /* extend overlapping extents */
5124 if (e
[i
].start
>= base_start
&&
5125 e
[i
].start
<= end
) {
5128 if (e
[i
].start
+ e
[i
].size
> end
)
5129 end
= e
[i
].start
+ e
[i
].size
;
5130 } else if (e
[i
].start
> end
) {
5136 return end
- base_start
;
5139 static unsigned long long merge_extents(struct intel_super
*super
, int sum_extents
)
5141 /* build a composite disk with all known extents and generate a new
5142 * 'maxsize' given the "all disks in an array must share a common start
5143 * offset" constraint
5145 struct extent
*e
= calloc(sum_extents
, sizeof(*e
));
5149 unsigned long long pos
;
5150 unsigned long long start
= 0;
5151 unsigned long long maxsize
;
5152 unsigned long reserve
;
5157 /* coalesce and sort all extents. also, check to see if we need to
5158 * reserve space between member arrays
5161 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
5164 for (i
= 0; i
< dl
->extent_cnt
; i
++)
5167 qsort(e
, sum_extents
, sizeof(*e
), cmp_extent
);
5172 while (i
< sum_extents
) {
5173 e
[j
].start
= e
[i
].start
;
5174 e
[j
].size
= find_size(e
, &i
, sum_extents
);
5176 if (e
[j
-1].size
== 0)
5185 unsigned long long esize
;
5187 esize
= e
[i
].start
- pos
;
5188 if (esize
>= maxsize
) {
5193 pos
= e
[i
].start
+ e
[i
].size
;
5195 } while (e
[i
-1].size
);
5201 /* FIXME assumes volume at offset 0 is the first volume in a
5204 if (start_extent
> 0)
5205 reserve
= IMSM_RESERVED_SECTORS
; /* gap between raid regions */
5209 if (maxsize
< reserve
)
5212 super
->create_offset
= ~((__u32
) 0);
5213 if (start
+ reserve
> super
->create_offset
)
5214 return 0; /* start overflows create_offset */
5215 super
->create_offset
= start
+ reserve
;
5217 return maxsize
- reserve
;
5220 static int is_raid_level_supported(const struct imsm_orom
*orom
, int level
, int raiddisks
)
5222 if (level
< 0 || level
== 6 || level
== 4)
5225 /* if we have an orom prevent invalid raid levels */
5228 case 0: return imsm_orom_has_raid0(orom
);
5231 return imsm_orom_has_raid1e(orom
);
5232 return imsm_orom_has_raid1(orom
) && raiddisks
== 2;
5233 case 10: return imsm_orom_has_raid10(orom
) && raiddisks
== 4;
5234 case 5: return imsm_orom_has_raid5(orom
) && raiddisks
> 2;
5237 return 1; /* not on an Intel RAID platform so anything goes */
5242 static int imsm_default_chunk(const struct imsm_orom
*orom
)
5244 /* up to 512 if the plaform supports it, otherwise the platform max.
5245 * 128 if no platform detected
5247 int fs
= max(7, orom
? fls(orom
->sss
) : 0);
5249 return min(512, (1 << fs
));
5252 #define pr_vrb(fmt, arg...) (void) (verbose && fprintf(stderr, Name fmt, ##arg))
5254 validate_geometry_imsm_orom(struct intel_super
*super
, int level
, int layout
,
5255 int raiddisks
, int *chunk
, int verbose
)
5257 /* check/set platform and metadata limits/defaults */
5258 if (super
->orom
&& raiddisks
> super
->orom
->dpa
) {
5259 pr_vrb(": platform supports a maximum of %d disks per array\n",
5264 /* capabilities of OROM tested - copied from validate_geometry_imsm_volume */
5265 if (!is_raid_level_supported(super
->orom
, level
, raiddisks
)) {
5266 pr_vrb(": platform does not support raid%d with %d disk%s\n",
5267 level
, raiddisks
, raiddisks
> 1 ? "s" : "");
5271 if (chunk
&& (*chunk
== 0 || *chunk
== UnSet
))
5272 *chunk
= imsm_default_chunk(super
->orom
);
5274 if (super
->orom
&& chunk
&& !imsm_orom_has_chunk(super
->orom
, *chunk
)) {
5275 pr_vrb(": platform does not support a chunk size of: "
5280 if (layout
!= imsm_level_to_layout(level
)) {
5282 pr_vrb(": imsm raid 5 only supports the left-asymmetric layout\n");
5283 else if (level
== 10)
5284 pr_vrb(": imsm raid 10 only supports the n2 layout\n");
5286 pr_vrb(": imsm unknown layout %#x for this raid level %d\n",
5293 /* validate_geometry_imsm_volume - lifted from validate_geometry_ddf_bvd
5294 * FIX ME add ahci details
5296 static int validate_geometry_imsm_volume(struct supertype
*st
, int level
,
5297 int layout
, int raiddisks
, int *chunk
,
5298 unsigned long long size
, char *dev
,
5299 unsigned long long *freesize
,
5303 struct intel_super
*super
= st
->sb
;
5304 struct imsm_super
*mpb
;
5306 unsigned long long pos
= 0;
5307 unsigned long long maxsize
;
5311 /* We must have the container info already read in. */
5315 mpb
= super
->anchor
;
5317 if (!validate_geometry_imsm_orom(super
, level
, layout
, raiddisks
, chunk
, verbose
)) {
5318 fprintf(stderr
, Name
": RAID gemetry validation failed. "
5319 "Cannot proceed with the action(s).\n");
5323 /* General test: make sure there is space for
5324 * 'raiddisks' device extents of size 'size' at a given
5327 unsigned long long minsize
= size
;
5328 unsigned long long start_offset
= MaxSector
;
5331 minsize
= MPB_SECTOR_CNT
+ IMSM_RESERVED_SECTORS
;
5332 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
5337 e
= get_extents(super
, dl
);
5340 unsigned long long esize
;
5341 esize
= e
[i
].start
- pos
;
5342 if (esize
>= minsize
)
5344 if (found
&& start_offset
== MaxSector
) {
5347 } else if (found
&& pos
!= start_offset
) {
5351 pos
= e
[i
].start
+ e
[i
].size
;
5353 } while (e
[i
-1].size
);
5358 if (dcnt
< raiddisks
) {
5360 fprintf(stderr
, Name
": imsm: Not enough "
5361 "devices with space for this array "
5369 /* This device must be a member of the set */
5370 if (stat(dev
, &stb
) < 0)
5372 if ((S_IFMT
& stb
.st_mode
) != S_IFBLK
)
5374 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
5375 if (dl
->major
== (int)major(stb
.st_rdev
) &&
5376 dl
->minor
== (int)minor(stb
.st_rdev
))
5381 fprintf(stderr
, Name
": %s is not in the "
5382 "same imsm set\n", dev
);
5384 } else if (super
->orom
&& dl
->index
< 0 && mpb
->num_raid_devs
) {
5385 /* If a volume is present then the current creation attempt
5386 * cannot incorporate new spares because the orom may not
5387 * understand this configuration (all member disks must be
5388 * members of each array in the container).
5390 fprintf(stderr
, Name
": %s is a spare and a volume"
5391 " is already defined for this container\n", dev
);
5392 fprintf(stderr
, Name
": The option-rom requires all member"
5393 " disks to be a member of all volumes\n");
5395 } else if (super
->orom
&& mpb
->num_raid_devs
> 0 &&
5396 mpb
->num_disks
!= raiddisks
) {
5397 fprintf(stderr
, Name
": The option-rom requires all member"
5398 " disks to be a member of all volumes\n");
5402 /* retrieve the largest free space block */
5403 e
= get_extents(super
, dl
);
5408 unsigned long long esize
;
5410 esize
= e
[i
].start
- pos
;
5411 if (esize
>= maxsize
)
5413 pos
= e
[i
].start
+ e
[i
].size
;
5415 } while (e
[i
-1].size
);
5420 fprintf(stderr
, Name
": unable to determine free space for: %s\n",
5424 if (maxsize
< size
) {
5426 fprintf(stderr
, Name
": %s not enough space (%llu < %llu)\n",
5427 dev
, maxsize
, size
);
5431 /* count total number of extents for merge */
5433 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
5435 i
+= dl
->extent_cnt
;
5437 maxsize
= merge_extents(super
, i
);
5439 if (!check_env("IMSM_NO_PLATFORM") &&
5440 mpb
->num_raid_devs
> 0 && size
&& size
!= maxsize
) {
5441 fprintf(stderr
, Name
": attempting to create a second "
5442 "volume with size less then remaining space. "
5447 if (maxsize
< size
|| maxsize
== 0) {
5450 fprintf(stderr
, Name
": no free space"
5451 " left on device. Aborting...\n");
5453 fprintf(stderr
, Name
": not enough space"
5454 " to create volume of given size"
5455 " (%llu < %llu). Aborting...\n",
5461 *freesize
= maxsize
;
5466 static int reserve_space(struct supertype
*st
, int raiddisks
,
5467 unsigned long long size
, int chunk
,
5468 unsigned long long *freesize
)
5470 struct intel_super
*super
= st
->sb
;
5471 struct imsm_super
*mpb
= super
->anchor
;
5476 unsigned long long maxsize
;
5477 unsigned long long minsize
;
5481 /* find the largest common start free region of the possible disks */
5485 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
5491 /* don't activate new spares if we are orom constrained
5492 * and there is already a volume active in the container
5494 if (super
->orom
&& dl
->index
< 0 && mpb
->num_raid_devs
)
5497 e
= get_extents(super
, dl
);
5500 for (i
= 1; e
[i
-1].size
; i
++)
5508 maxsize
= merge_extents(super
, extent_cnt
);
5512 minsize
= chunk
* 2;
5514 if (cnt
< raiddisks
||
5515 (super
->orom
&& used
&& used
!= raiddisks
) ||
5516 maxsize
< minsize
||
5518 fprintf(stderr
, Name
": not enough devices with space to create array.\n");
5519 return 0; /* No enough free spaces large enough */
5531 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
5533 dl
->raiddisk
= cnt
++;
5540 static int validate_geometry_imsm(struct supertype
*st
, int level
, int layout
,
5541 int raiddisks
, int *chunk
, unsigned long long size
,
5542 char *dev
, unsigned long long *freesize
,
5550 * if given unused devices create a container
5551 * if given given devices in a container create a member volume
5553 if (level
== LEVEL_CONTAINER
) {
5554 /* Must be a fresh device to add to a container */
5555 return validate_geometry_imsm_container(st
, level
, layout
,
5557 chunk
?*chunk
:0, size
,
5564 if (!validate_geometry_imsm_orom(st
->sb
, level
, layout
,
5568 /* we are being asked to automatically layout a
5569 * new volume based on the current contents of
5570 * the container. If the the parameters can be
5571 * satisfied reserve_space will record the disks,
5572 * start offset, and size of the volume to be
5573 * created. add_to_super and getinfo_super
5574 * detect when autolayout is in progress.
5577 return reserve_space(st
, raiddisks
, size
,
5578 chunk
?*chunk
:0, freesize
);
5583 /* creating in a given container */
5584 return validate_geometry_imsm_volume(st
, level
, layout
,
5585 raiddisks
, chunk
, size
,
5586 dev
, freesize
, verbose
);
5589 /* This device needs to be a device in an 'imsm' container */
5590 fd
= open(dev
, O_RDONLY
|O_EXCL
, 0);
5594 Name
": Cannot create this array on device %s\n",
5599 if (errno
!= EBUSY
|| (fd
= open(dev
, O_RDONLY
, 0)) < 0) {
5601 fprintf(stderr
, Name
": Cannot open %s: %s\n",
5602 dev
, strerror(errno
));
5605 /* Well, it is in use by someone, maybe an 'imsm' container. */
5606 cfd
= open_container(fd
);
5610 fprintf(stderr
, Name
": Cannot use %s: It is busy\n",
5614 sra
= sysfs_read(cfd
, 0, GET_VERSION
);
5615 if (sra
&& sra
->array
.major_version
== -1 &&
5616 strcmp(sra
->text_version
, "imsm") == 0)
5620 /* This is a member of a imsm container. Load the container
5621 * and try to create a volume
5623 struct intel_super
*super
;
5625 if (load_super_imsm_all(st
, cfd
, (void **) &super
, NULL
) == 0) {
5627 st
->container_dev
= fd2devnum(cfd
);
5629 return validate_geometry_imsm_volume(st
, level
, layout
,
5638 fprintf(stderr
, Name
": failed container membership check\n");
5644 static void default_geometry_imsm(struct supertype
*st
, int *level
, int *layout
, int *chunk
)
5646 struct intel_super
*super
= st
->sb
;
5648 if (level
&& *level
== UnSet
)
5649 *level
= LEVEL_CONTAINER
;
5651 if (level
&& layout
&& *layout
== UnSet
)
5652 *layout
= imsm_level_to_layout(*level
);
5654 if (chunk
&& (*chunk
== UnSet
|| *chunk
== 0))
5655 *chunk
= imsm_default_chunk(super
->orom
);
5658 static void handle_missing(struct intel_super
*super
, struct imsm_dev
*dev
);
5660 static int kill_subarray_imsm(struct supertype
*st
)
5662 /* remove the subarray currently referenced by ->current_vol */
5664 struct intel_dev
**dp
;
5665 struct intel_super
*super
= st
->sb
;
5666 __u8 current_vol
= super
->current_vol
;
5667 struct imsm_super
*mpb
= super
->anchor
;
5669 if (super
->current_vol
< 0)
5671 super
->current_vol
= -1; /* invalidate subarray cursor */
5673 /* block deletions that would change the uuid of active subarrays
5675 * FIXME when immutable ids are available, but note that we'll
5676 * also need to fixup the invalidated/active subarray indexes in
5679 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
5682 if (i
< current_vol
)
5684 sprintf(subarray
, "%u", i
);
5685 if (is_subarray_active(subarray
, st
->devname
)) {
5687 Name
": deleting subarray-%d would change the UUID of active subarray-%d, aborting\n",
5694 if (st
->update_tail
) {
5695 struct imsm_update_kill_array
*u
= malloc(sizeof(*u
));
5699 u
->type
= update_kill_array
;
5700 u
->dev_idx
= current_vol
;
5701 append_metadata_update(st
, u
, sizeof(*u
));
5706 for (dp
= &super
->devlist
; *dp
;)
5707 if ((*dp
)->index
== current_vol
) {
5710 handle_missing(super
, (*dp
)->dev
);
5711 if ((*dp
)->index
> current_vol
)
5716 /* no more raid devices, all active components are now spares,
5717 * but of course failed are still failed
5719 if (--mpb
->num_raid_devs
== 0) {
5722 for (d
= super
->disks
; d
; d
= d
->next
)
5727 super
->updates_pending
++;
5732 static int update_subarray_imsm(struct supertype
*st
, char *subarray
,
5733 char *update
, struct mddev_ident
*ident
)
5735 /* update the subarray currently referenced by ->current_vol */
5736 struct intel_super
*super
= st
->sb
;
5737 struct imsm_super
*mpb
= super
->anchor
;
5739 if (strcmp(update
, "name") == 0) {
5740 char *name
= ident
->name
;
5744 if (is_subarray_active(subarray
, st
->devname
)) {
5746 Name
": Unable to update name of active subarray\n");
5750 if (!check_name(super
, name
, 0))
5753 vol
= strtoul(subarray
, &ep
, 10);
5754 if (*ep
!= '\0' || vol
>= super
->anchor
->num_raid_devs
)
5757 if (st
->update_tail
) {
5758 struct imsm_update_rename_array
*u
= malloc(sizeof(*u
));
5762 u
->type
= update_rename_array
;
5764 snprintf((char *) u
->name
, MAX_RAID_SERIAL_LEN
, "%s", name
);
5765 append_metadata_update(st
, u
, sizeof(*u
));
5767 struct imsm_dev
*dev
;
5770 dev
= get_imsm_dev(super
, vol
);
5771 snprintf((char *) dev
->volume
, MAX_RAID_SERIAL_LEN
, "%s", name
);
5772 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
5773 dev
= get_imsm_dev(super
, i
);
5774 handle_missing(super
, dev
);
5776 super
->updates_pending
++;
5783 #endif /* MDASSEMBLE */
5785 static int is_gen_migration(struct imsm_dev
*dev
)
5790 if (!dev
->vol
.migr_state
)
5793 if (migr_type(dev
) == MIGR_GEN_MIGR
)
5799 static int is_rebuilding(struct imsm_dev
*dev
)
5801 struct imsm_map
*migr_map
;
5803 if (!dev
->vol
.migr_state
)
5806 if (migr_type(dev
) != MIGR_REBUILD
)
5809 migr_map
= get_imsm_map(dev
, MAP_1
);
5811 if (migr_map
->map_state
== IMSM_T_STATE_DEGRADED
)
5818 static int is_initializing(struct imsm_dev
*dev
)
5820 struct imsm_map
*migr_map
;
5822 if (!dev
->vol
.migr_state
)
5825 if (migr_type(dev
) != MIGR_INIT
)
5828 migr_map
= get_imsm_map(dev
, MAP_1
);
5830 if (migr_map
->map_state
== IMSM_T_STATE_UNINITIALIZED
)
5837 static void update_recovery_start(struct intel_super
*super
,
5838 struct imsm_dev
*dev
,
5839 struct mdinfo
*array
)
5841 struct mdinfo
*rebuild
= NULL
;
5845 if (!is_rebuilding(dev
))
5848 /* Find the rebuild target, but punt on the dual rebuild case */
5849 for (d
= array
->devs
; d
; d
= d
->next
)
5850 if (d
->recovery_start
== 0) {
5857 /* (?) none of the disks are marked with
5858 * IMSM_ORD_REBUILD, so assume they are missing and the
5859 * disk_ord_tbl was not correctly updated
5861 dprintf("%s: failed to locate out-of-sync disk\n", __func__
);
5865 units
= __le32_to_cpu(dev
->vol
.curr_migr_unit
);
5866 rebuild
->recovery_start
= units
* blocks_per_migr_unit(super
, dev
);
5870 static int recover_backup_imsm(struct supertype
*st
, struct mdinfo
*info
);
5873 static struct mdinfo
*container_content_imsm(struct supertype
*st
, char *subarray
)
5875 /* Given a container loaded by load_super_imsm_all,
5876 * extract information about all the arrays into
5878 * If 'subarray' is given, just extract info about that array.
5880 * For each imsm_dev create an mdinfo, fill it in,
5881 * then look for matching devices in super->disks
5882 * and create appropriate device mdinfo.
5884 struct intel_super
*super
= st
->sb
;
5885 struct imsm_super
*mpb
= super
->anchor
;
5886 struct mdinfo
*rest
= NULL
;
5890 int spare_disks
= 0;
5892 /* do not assemble arrays when not all attributes are supported */
5893 if (imsm_check_attributes(mpb
->attributes
) == 0) {
5895 fprintf(stderr
, Name
": Unsupported attributes in IMSM metadata."
5896 "Arrays activation is blocked.\n");
5899 /* check for bad blocks */
5900 if (imsm_bbm_log_size(super
->anchor
)) {
5901 fprintf(stderr
, Name
": BBM log found in IMSM metadata."
5902 "Arrays activation is blocked.\n");
5907 /* count spare devices, not used in maps
5909 for (d
= super
->disks
; d
; d
= d
->next
)
5913 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
5914 struct imsm_dev
*dev
;
5915 struct imsm_map
*map
;
5916 struct imsm_map
*map2
;
5917 struct mdinfo
*this;
5925 (i
!= strtoul(subarray
, &ep
, 10) || *ep
!= '\0'))
5928 dev
= get_imsm_dev(super
, i
);
5929 map
= get_imsm_map(dev
, MAP_0
);
5930 map2
= get_imsm_map(dev
, MAP_1
);
5932 /* do not publish arrays that are in the middle of an
5933 * unsupported migration
5935 if (dev
->vol
.migr_state
&&
5936 (migr_type(dev
) == MIGR_STATE_CHANGE
)) {
5937 fprintf(stderr
, Name
": cannot assemble volume '%.16s':"
5938 " unsupported migration in progress\n",
5942 /* do not publish arrays that are not support by controller's
5946 this = malloc(sizeof(*this));
5948 fprintf(stderr
, Name
": failed to allocate %zu bytes\n",
5953 super
->current_vol
= i
;
5954 getinfo_super_imsm_volume(st
, this, NULL
);
5957 chunk
= __le16_to_cpu(map
->blocks_per_strip
) >> 1;
5958 /* mdadm does not support all metadata features- set the bit in all arrays state */
5959 if (!validate_geometry_imsm_orom(super
,
5960 get_imsm_raid_level(map
), /* RAID level */
5961 imsm_level_to_layout(get_imsm_raid_level(map
)),
5962 map
->num_members
, /* raid disks */
5965 fprintf(stderr
, Name
": IMSM RAID geometry validation"
5966 " failed. Array %s activation is blocked.\n",
5968 this->array
.state
|=
5969 (1<<MD_SB_BLOCK_CONTAINER_RESHAPE
) |
5970 (1<<MD_SB_BLOCK_VOLUME
);
5974 /* if array has bad blocks, set suitable bit in all arrays state */
5976 this->array
.state
|=
5977 (1<<MD_SB_BLOCK_CONTAINER_RESHAPE
) |
5978 (1<<MD_SB_BLOCK_VOLUME
);
5980 for (slot
= 0 ; slot
< map
->num_members
; slot
++) {
5981 unsigned long long recovery_start
;
5982 struct mdinfo
*info_d
;
5989 idx
= get_imsm_disk_idx(dev
, slot
, MAP_0
);
5990 ord
= get_imsm_ord_tbl_ent(dev
, slot
, MAP_X
);
5991 for (d
= super
->disks
; d
; d
= d
->next
)
5992 if (d
->index
== idx
)
5995 recovery_start
= MaxSector
;
5998 if (d
&& is_failed(&d
->disk
))
6000 if (ord
& IMSM_ORD_REBUILD
)
6004 * if we skip some disks the array will be assmebled degraded;
6005 * reset resync start to avoid a dirty-degraded
6006 * situation when performing the intial sync
6008 * FIXME handle dirty degraded
6010 if ((skip
|| recovery_start
== 0) && !dev
->vol
.dirty
)
6011 this->resync_start
= MaxSector
;
6015 info_d
= calloc(1, sizeof(*info_d
));
6017 fprintf(stderr
, Name
": failed to allocate disk"
6018 " for volume %.16s\n", dev
->volume
);
6019 info_d
= this->devs
;
6021 struct mdinfo
*d
= info_d
->next
;
6030 info_d
->next
= this->devs
;
6031 this->devs
= info_d
;
6033 info_d
->disk
.number
= d
->index
;
6034 info_d
->disk
.major
= d
->major
;
6035 info_d
->disk
.minor
= d
->minor
;
6036 info_d
->disk
.raid_disk
= slot
;
6037 info_d
->recovery_start
= recovery_start
;
6039 if (slot
< map2
->num_members
)
6040 info_d
->disk
.state
= (1 << MD_DISK_ACTIVE
);
6042 this->array
.spare_disks
++;
6044 if (slot
< map
->num_members
)
6045 info_d
->disk
.state
= (1 << MD_DISK_ACTIVE
);
6047 this->array
.spare_disks
++;
6049 if (info_d
->recovery_start
== MaxSector
)
6050 this->array
.working_disks
++;
6052 info_d
->events
= __le32_to_cpu(mpb
->generation_num
);
6053 info_d
->data_offset
= __le32_to_cpu(map
->pba_of_lba0
);
6054 info_d
->component_size
= __le32_to_cpu(map
->blocks_per_member
);
6056 /* now that the disk list is up-to-date fixup recovery_start */
6057 update_recovery_start(super
, dev
, this);
6058 this->array
.spare_disks
+= spare_disks
;
6061 /* check for reshape */
6062 if (this->reshape_active
== 1)
6063 recover_backup_imsm(st
, this);
6072 static __u8
imsm_check_degraded(struct intel_super
*super
, struct imsm_dev
*dev
,
6073 int failed
, int look_in_map
)
6075 struct imsm_map
*map
;
6077 map
= get_imsm_map(dev
, look_in_map
);
6080 return map
->map_state
== IMSM_T_STATE_UNINITIALIZED
?
6081 IMSM_T_STATE_UNINITIALIZED
: IMSM_T_STATE_NORMAL
;
6083 switch (get_imsm_raid_level(map
)) {
6085 return IMSM_T_STATE_FAILED
;
6088 if (failed
< map
->num_members
)
6089 return IMSM_T_STATE_DEGRADED
;
6091 return IMSM_T_STATE_FAILED
;
6096 * check to see if any mirrors have failed, otherwise we
6097 * are degraded. Even numbered slots are mirrored on
6101 /* gcc -Os complains that this is unused */
6102 int insync
= insync
;
6104 for (i
= 0; i
< map
->num_members
; i
++) {
6105 __u32 ord
= get_imsm_ord_tbl_ent(dev
, i
, MAP_X
);
6106 int idx
= ord_to_idx(ord
);
6107 struct imsm_disk
*disk
;
6109 /* reset the potential in-sync count on even-numbered
6110 * slots. num_copies is always 2 for imsm raid10
6115 disk
= get_imsm_disk(super
, idx
);
6116 if (!disk
|| is_failed(disk
) || ord
& IMSM_ORD_REBUILD
)
6119 /* no in-sync disks left in this mirror the
6123 return IMSM_T_STATE_FAILED
;
6126 return IMSM_T_STATE_DEGRADED
;
6130 return IMSM_T_STATE_DEGRADED
;
6132 return IMSM_T_STATE_FAILED
;
6138 return map
->map_state
;
6141 static int imsm_count_failed(struct intel_super
*super
, struct imsm_dev
*dev
,
6146 struct imsm_disk
*disk
;
6147 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
6148 struct imsm_map
*prev
= get_imsm_map(dev
, MAP_1
);
6149 struct imsm_map
*map_for_loop
;
6154 /* at the beginning of migration we set IMSM_ORD_REBUILD on
6155 * disks that are being rebuilt. New failures are recorded to
6156 * map[0]. So we look through all the disks we started with and
6157 * see if any failures are still present, or if any new ones
6161 if (prev
&& (map
->num_members
< prev
->num_members
))
6162 map_for_loop
= prev
;
6164 for (i
= 0; i
< map_for_loop
->num_members
; i
++) {
6166 /* when MAP_X is passed both maps failures are counted
6169 ((look_in_map
== MAP_1
) || (look_in_map
== MAP_X
)) &&
6170 (i
< prev
->num_members
)) {
6171 ord
= __le32_to_cpu(prev
->disk_ord_tbl
[i
]);
6172 idx_1
= ord_to_idx(ord
);
6174 disk
= get_imsm_disk(super
, idx_1
);
6175 if (!disk
|| is_failed(disk
) || ord
& IMSM_ORD_REBUILD
)
6178 if (((look_in_map
== MAP_0
) || (look_in_map
== MAP_X
)) &&
6179 (i
< map
->num_members
)) {
6180 ord
= __le32_to_cpu(map
->disk_ord_tbl
[i
]);
6181 idx
= ord_to_idx(ord
);
6184 disk
= get_imsm_disk(super
, idx
);
6185 if (!disk
|| is_failed(disk
) ||
6186 ord
& IMSM_ORD_REBUILD
)
6196 static int imsm_open_new(struct supertype
*c
, struct active_array
*a
,
6199 struct intel_super
*super
= c
->sb
;
6200 struct imsm_super
*mpb
= super
->anchor
;
6202 if (atoi(inst
) >= mpb
->num_raid_devs
) {
6203 fprintf(stderr
, "%s: subarry index %d, out of range\n",
6204 __func__
, atoi(inst
));
6208 dprintf("imsm: open_new %s\n", inst
);
6209 a
->info
.container_member
= atoi(inst
);
6213 static int is_resyncing(struct imsm_dev
*dev
)
6215 struct imsm_map
*migr_map
;
6217 if (!dev
->vol
.migr_state
)
6220 if (migr_type(dev
) == MIGR_INIT
||
6221 migr_type(dev
) == MIGR_REPAIR
)
6224 if (migr_type(dev
) == MIGR_GEN_MIGR
)
6227 migr_map
= get_imsm_map(dev
, MAP_1
);
6229 if ((migr_map
->map_state
== IMSM_T_STATE_NORMAL
) &&
6230 (dev
->vol
.migr_type
!= MIGR_GEN_MIGR
))
6236 /* return true if we recorded new information */
6237 static int mark_failure(struct imsm_dev
*dev
, struct imsm_disk
*disk
, int idx
)
6241 struct imsm_map
*map
;
6242 char buf
[MAX_RAID_SERIAL_LEN
+3];
6243 unsigned int len
, shift
= 0;
6245 /* new failures are always set in map[0] */
6246 map
= get_imsm_map(dev
, MAP_0
);
6248 slot
= get_imsm_disk_slot(map
, idx
);
6252 ord
= __le32_to_cpu(map
->disk_ord_tbl
[slot
]);
6253 if (is_failed(disk
) && (ord
& IMSM_ORD_REBUILD
))
6256 memcpy(buf
, disk
->serial
, MAX_RAID_SERIAL_LEN
);
6257 buf
[MAX_RAID_SERIAL_LEN
] = '\000';
6259 if ((len
= strlen(buf
)) >= MAX_RAID_SERIAL_LEN
)
6260 shift
= len
- MAX_RAID_SERIAL_LEN
+ 1;
6261 strncpy((char *)disk
->serial
, &buf
[shift
], MAX_RAID_SERIAL_LEN
);
6263 disk
->status
|= FAILED_DISK
;
6264 set_imsm_ord_tbl_ent(map
, slot
, idx
| IMSM_ORD_REBUILD
);
6265 /* mark failures in second map if second map exists and this disk
6267 * This is valid for migration, initialization and rebuild
6269 if (dev
->vol
.migr_state
) {
6270 struct imsm_map
*map2
= get_imsm_map(dev
, MAP_1
);
6271 int slot2
= get_imsm_disk_slot(map2
, idx
);
6273 if ((slot2
< map2
->num_members
) &&
6275 set_imsm_ord_tbl_ent(map2
, slot2
,
6276 idx
| IMSM_ORD_REBUILD
);
6278 if (map
->failed_disk_num
== 0xff)
6279 map
->failed_disk_num
= slot
;
6283 static void mark_missing(struct imsm_dev
*dev
, struct imsm_disk
*disk
, int idx
)
6285 mark_failure(dev
, disk
, idx
);
6287 if (disk
->scsi_id
== __cpu_to_le32(~(__u32
)0))
6290 disk
->scsi_id
= __cpu_to_le32(~(__u32
)0);
6291 memmove(&disk
->serial
[0], &disk
->serial
[1], MAX_RAID_SERIAL_LEN
- 1);
6294 static void handle_missing(struct intel_super
*super
, struct imsm_dev
*dev
)
6298 if (!super
->missing
)
6301 dprintf("imsm: mark missing\n");
6302 /* end process for initialization and rebuild only
6304 if (is_gen_migration(dev
) == 0) {
6308 failed
= imsm_count_failed(super
, dev
, MAP_0
);
6309 map_state
= imsm_check_degraded(super
, dev
, failed
, MAP_0
);
6311 end_migration(dev
, super
, map_state
);
6313 for (dl
= super
->missing
; dl
; dl
= dl
->next
)
6314 mark_missing(dev
, &dl
->disk
, dl
->index
);
6315 super
->updates_pending
++;
6318 static unsigned long long imsm_set_array_size(struct imsm_dev
*dev
)
6320 int used_disks
= imsm_num_data_members(dev
, MAP_0
);
6321 unsigned long long array_blocks
;
6322 struct imsm_map
*map
;
6324 if (used_disks
== 0) {
6325 /* when problems occures
6326 * return current array_blocks value
6328 array_blocks
= __le32_to_cpu(dev
->size_high
);
6329 array_blocks
= array_blocks
<< 32;
6330 array_blocks
+= __le32_to_cpu(dev
->size_low
);
6332 return array_blocks
;
6335 /* set array size in metadata
6337 map
= get_imsm_map(dev
, MAP_0
);
6338 array_blocks
= map
->blocks_per_member
* used_disks
;
6340 /* round array size down to closest MB
6342 array_blocks
= (array_blocks
>> SECT_PER_MB_SHIFT
) << SECT_PER_MB_SHIFT
;
6343 dev
->size_low
= __cpu_to_le32((__u32
)array_blocks
);
6344 dev
->size_high
= __cpu_to_le32((__u32
)(array_blocks
>> 32));
6346 return array_blocks
;
6349 static void imsm_set_disk(struct active_array
*a
, int n
, int state
);
6351 static void imsm_progress_container_reshape(struct intel_super
*super
)
6353 /* if no device has a migr_state, but some device has a
6354 * different number of members than the previous device, start
6355 * changing the number of devices in this device to match
6358 struct imsm_super
*mpb
= super
->anchor
;
6359 int prev_disks
= -1;
6363 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
6364 struct imsm_dev
*dev
= get_imsm_dev(super
, i
);
6365 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
6366 struct imsm_map
*map2
;
6367 int prev_num_members
;
6369 if (dev
->vol
.migr_state
)
6372 if (prev_disks
== -1)
6373 prev_disks
= map
->num_members
;
6374 if (prev_disks
== map
->num_members
)
6377 /* OK, this array needs to enter reshape mode.
6378 * i.e it needs a migr_state
6381 copy_map_size
= sizeof_imsm_map(map
);
6382 prev_num_members
= map
->num_members
;
6383 map
->num_members
= prev_disks
;
6384 dev
->vol
.migr_state
= 1;
6385 dev
->vol
.curr_migr_unit
= 0;
6386 set_migr_type(dev
, MIGR_GEN_MIGR
);
6387 for (i
= prev_num_members
;
6388 i
< map
->num_members
; i
++)
6389 set_imsm_ord_tbl_ent(map
, i
, i
);
6390 map2
= get_imsm_map(dev
, MAP_1
);
6391 /* Copy the current map */
6392 memcpy(map2
, map
, copy_map_size
);
6393 map2
->num_members
= prev_num_members
;
6395 imsm_set_array_size(dev
);
6396 super
->updates_pending
++;
6400 /* Handle dirty -> clean transititions, resync and reshape. Degraded and rebuild
6401 * states are handled in imsm_set_disk() with one exception, when a
6402 * resync is stopped due to a new failure this routine will set the
6403 * 'degraded' state for the array.
6405 static int imsm_set_array_state(struct active_array
*a
, int consistent
)
6407 int inst
= a
->info
.container_member
;
6408 struct intel_super
*super
= a
->container
->sb
;
6409 struct imsm_dev
*dev
= get_imsm_dev(super
, inst
);
6410 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
6411 int failed
= imsm_count_failed(super
, dev
, MAP_0
);
6412 __u8 map_state
= imsm_check_degraded(super
, dev
, failed
, MAP_0
);
6413 __u32 blocks_per_unit
;
6415 if (dev
->vol
.migr_state
&&
6416 dev
->vol
.migr_type
== MIGR_GEN_MIGR
) {
6417 /* array state change is blocked due to reshape action
6419 * - abort the reshape (if last_checkpoint is 0 and action!= reshape)
6420 * - finish the reshape (if last_checkpoint is big and action != reshape)
6421 * - update curr_migr_unit
6423 if (a
->curr_action
== reshape
) {
6424 /* still reshaping, maybe update curr_migr_unit */
6425 goto mark_checkpoint
;
6427 if (a
->last_checkpoint
== 0 && a
->prev_action
== reshape
) {
6428 /* for some reason we aborted the reshape.
6430 * disable automatic metadata rollback
6431 * user action is required to recover process
6434 struct imsm_map
*map2
=
6435 get_imsm_map(dev
, MAP_1
);
6436 dev
->vol
.migr_state
= 0;
6437 set_migr_type(dev
, 0);
6438 dev
->vol
.curr_migr_unit
= 0;
6440 sizeof_imsm_map(map2
));
6441 super
->updates_pending
++;
6444 if (a
->last_checkpoint
>= a
->info
.component_size
) {
6445 unsigned long long array_blocks
;
6449 used_disks
= imsm_num_data_members(dev
, MAP_0
);
6450 if (used_disks
> 0) {
6452 map
->blocks_per_member
*
6454 /* round array size down to closest MB
6456 array_blocks
= (array_blocks
6457 >> SECT_PER_MB_SHIFT
)
6458 << SECT_PER_MB_SHIFT
;
6459 a
->info
.custom_array_size
= array_blocks
;
6460 /* encourage manager to update array
6464 a
->check_reshape
= 1;
6466 /* finalize online capacity expansion/reshape */
6467 for (mdi
= a
->info
.devs
; mdi
; mdi
= mdi
->next
)
6469 mdi
->disk
.raid_disk
,
6472 imsm_progress_container_reshape(super
);
6477 /* before we activate this array handle any missing disks */
6478 if (consistent
== 2)
6479 handle_missing(super
, dev
);
6481 if (consistent
== 2 &&
6482 (!is_resync_complete(&a
->info
) ||
6483 map_state
!= IMSM_T_STATE_NORMAL
||
6484 dev
->vol
.migr_state
))
6487 if (is_resync_complete(&a
->info
)) {
6488 /* complete intialization / resync,
6489 * recovery and interrupted recovery is completed in
6492 if (is_resyncing(dev
)) {
6493 dprintf("imsm: mark resync done\n");
6494 end_migration(dev
, super
, map_state
);
6495 super
->updates_pending
++;
6496 a
->last_checkpoint
= 0;
6498 } else if ((!is_resyncing(dev
) && !failed
) &&
6499 (imsm_reshape_blocks_arrays_changes(super
) == 0)) {
6500 /* mark the start of the init process if nothing is failed */
6501 dprintf("imsm: mark resync start\n");
6502 if (map
->map_state
== IMSM_T_STATE_UNINITIALIZED
)
6503 migrate(dev
, super
, IMSM_T_STATE_NORMAL
, MIGR_INIT
);
6505 migrate(dev
, super
, IMSM_T_STATE_NORMAL
, MIGR_REPAIR
);
6506 super
->updates_pending
++;
6510 /* skip checkpointing for general migration,
6511 * it is controlled in mdadm
6513 if (is_gen_migration(dev
))
6514 goto skip_mark_checkpoint
;
6516 /* check if we can update curr_migr_unit from resync_start, recovery_start */
6517 blocks_per_unit
= blocks_per_migr_unit(super
, dev
);
6518 if (blocks_per_unit
) {
6522 units
= a
->last_checkpoint
/ blocks_per_unit
;
6525 /* check that we did not overflow 32-bits, and that
6526 * curr_migr_unit needs updating
6528 if (units32
== units
&&
6530 __le32_to_cpu(dev
->vol
.curr_migr_unit
) != units32
) {
6531 dprintf("imsm: mark checkpoint (%u)\n", units32
);
6532 dev
->vol
.curr_migr_unit
= __cpu_to_le32(units32
);
6533 super
->updates_pending
++;
6537 skip_mark_checkpoint
:
6538 /* mark dirty / clean */
6539 if (dev
->vol
.dirty
!= !consistent
) {
6540 dprintf("imsm: mark '%s'\n", consistent
? "clean" : "dirty");
6545 super
->updates_pending
++;
6551 static void imsm_set_disk(struct active_array
*a
, int n
, int state
)
6553 int inst
= a
->info
.container_member
;
6554 struct intel_super
*super
= a
->container
->sb
;
6555 struct imsm_dev
*dev
= get_imsm_dev(super
, inst
);
6556 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
6557 struct imsm_disk
*disk
;
6562 if (n
> map
->num_members
)
6563 fprintf(stderr
, "imsm: set_disk %d out of range 0..%d\n",
6564 n
, map
->num_members
- 1);
6569 dprintf("imsm: set_disk %d:%x\n", n
, state
);
6571 ord
= get_imsm_ord_tbl_ent(dev
, n
, MAP_0
);
6572 disk
= get_imsm_disk(super
, ord_to_idx(ord
));
6574 /* check for new failures */
6575 if (state
& DS_FAULTY
) {
6576 if (mark_failure(dev
, disk
, ord_to_idx(ord
)))
6577 super
->updates_pending
++;
6580 /* check if in_sync */
6581 if (state
& DS_INSYNC
&& ord
& IMSM_ORD_REBUILD
&& is_rebuilding(dev
)) {
6582 struct imsm_map
*migr_map
= get_imsm_map(dev
, MAP_1
);
6584 set_imsm_ord_tbl_ent(migr_map
, n
, ord_to_idx(ord
));
6585 super
->updates_pending
++;
6588 failed
= imsm_count_failed(super
, dev
, MAP_0
);
6589 map_state
= imsm_check_degraded(super
, dev
, failed
, MAP_0
);
6591 /* check if recovery complete, newly degraded, or failed */
6592 dprintf("imsm: Detected transition to state ");
6593 switch (map_state
) {
6594 case IMSM_T_STATE_NORMAL
: /* transition to normal state */
6595 dprintf("normal: ");
6596 if (is_rebuilding(dev
)) {
6597 dprintf("while rebuilding");
6598 end_migration(dev
, super
, map_state
);
6599 map
= get_imsm_map(dev
, MAP_0
);
6600 map
->failed_disk_num
= ~0;
6601 super
->updates_pending
++;
6602 a
->last_checkpoint
= 0;
6605 if (is_gen_migration(dev
)) {
6606 dprintf("while general migration");
6607 if (a
->last_checkpoint
>= a
->info
.component_size
)
6608 end_migration(dev
, super
, map_state
);
6610 map
->map_state
= map_state
;
6611 map
= get_imsm_map(dev
, MAP_0
);
6612 map
->failed_disk_num
= ~0;
6613 super
->updates_pending
++;
6617 case IMSM_T_STATE_DEGRADED
: /* transition to degraded state */
6618 dprintf("degraded: ");
6619 if ((map
->map_state
!= map_state
) &&
6620 !dev
->vol
.migr_state
) {
6621 dprintf("mark degraded");
6622 map
->map_state
= map_state
;
6623 super
->updates_pending
++;
6624 a
->last_checkpoint
= 0;
6627 if (is_rebuilding(dev
)) {
6628 dprintf("while rebuilding.");
6629 if (map
->map_state
!= map_state
) {
6630 dprintf(" Map state change");
6631 end_migration(dev
, super
, map_state
);
6632 super
->updates_pending
++;
6636 if (is_gen_migration(dev
)) {
6637 dprintf("while general migration");
6638 if (a
->last_checkpoint
>= a
->info
.component_size
)
6639 end_migration(dev
, super
, map_state
);
6641 map
->map_state
= map_state
;
6642 manage_second_map(super
, dev
);
6644 super
->updates_pending
++;
6647 if (is_initializing(dev
)) {
6648 dprintf("while initialization.");
6649 map
->map_state
= map_state
;
6650 super
->updates_pending
++;
6654 case IMSM_T_STATE_FAILED
: /* transition to failed state */
6655 dprintf("failed: ");
6656 if (is_gen_migration(dev
)) {
6657 dprintf("while general migration");
6658 map
->map_state
= map_state
;
6659 super
->updates_pending
++;
6662 if (map
->map_state
!= map_state
) {
6663 dprintf("mark failed");
6664 end_migration(dev
, super
, map_state
);
6665 super
->updates_pending
++;
6666 a
->last_checkpoint
= 0;
6671 dprintf("state %i\n", map_state
);
6677 static int store_imsm_mpb(int fd
, struct imsm_super
*mpb
)
6680 __u32 mpb_size
= __le32_to_cpu(mpb
->mpb_size
);
6681 unsigned long long dsize
;
6682 unsigned long long sectors
;
6684 get_dev_size(fd
, NULL
, &dsize
);
6686 if (mpb_size
> 512) {
6687 /* -1 to account for anchor */
6688 sectors
= mpb_sectors(mpb
) - 1;
6690 /* write the extended mpb to the sectors preceeding the anchor */
6691 if (lseek64(fd
, dsize
- (512 * (2 + sectors
)), SEEK_SET
) < 0)
6694 if ((unsigned long long)write(fd
, buf
+ 512, 512 * sectors
)
6699 /* first block is stored on second to last sector of the disk */
6700 if (lseek64(fd
, dsize
- (512 * 2), SEEK_SET
) < 0)
6703 if (write(fd
, buf
, 512) != 512)
6709 static void imsm_sync_metadata(struct supertype
*container
)
6711 struct intel_super
*super
= container
->sb
;
6713 dprintf("sync metadata: %d\n", super
->updates_pending
);
6714 if (!super
->updates_pending
)
6717 write_super_imsm(container
, 0);
6719 super
->updates_pending
= 0;
6722 static struct dl
*imsm_readd(struct intel_super
*super
, int idx
, struct active_array
*a
)
6724 struct imsm_dev
*dev
= get_imsm_dev(super
, a
->info
.container_member
);
6725 int i
= get_imsm_disk_idx(dev
, idx
, MAP_X
);
6728 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
6732 if (dl
&& is_failed(&dl
->disk
))
6736 dprintf("%s: found %x:%x\n", __func__
, dl
->major
, dl
->minor
);
6741 static struct dl
*imsm_add_spare(struct intel_super
*super
, int slot
,
6742 struct active_array
*a
, int activate_new
,
6743 struct mdinfo
*additional_test_list
)
6745 struct imsm_dev
*dev
= get_imsm_dev(super
, a
->info
.container_member
);
6746 int idx
= get_imsm_disk_idx(dev
, slot
, MAP_X
);
6747 struct imsm_super
*mpb
= super
->anchor
;
6748 struct imsm_map
*map
;
6749 unsigned long long pos
;
6754 __u32 array_start
= 0;
6755 __u32 array_end
= 0;
6757 struct mdinfo
*test_list
;
6759 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
6760 /* If in this array, skip */
6761 for (d
= a
->info
.devs
; d
; d
= d
->next
)
6762 if (d
->state_fd
>= 0 &&
6763 d
->disk
.major
== dl
->major
&&
6764 d
->disk
.minor
== dl
->minor
) {
6765 dprintf("%x:%x already in array\n",
6766 dl
->major
, dl
->minor
);
6771 test_list
= additional_test_list
;
6773 if (test_list
->disk
.major
== dl
->major
&&
6774 test_list
->disk
.minor
== dl
->minor
) {
6775 dprintf("%x:%x already in additional test list\n",
6776 dl
->major
, dl
->minor
);
6779 test_list
= test_list
->next
;
6784 /* skip in use or failed drives */
6785 if (is_failed(&dl
->disk
) || idx
== dl
->index
||
6787 dprintf("%x:%x status (failed: %d index: %d)\n",
6788 dl
->major
, dl
->minor
, is_failed(&dl
->disk
), idx
);
6792 /* skip pure spares when we are looking for partially
6793 * assimilated drives
6795 if (dl
->index
== -1 && !activate_new
)
6798 /* Does this unused device have the requisite free space?
6799 * It needs to be able to cover all member volumes
6801 ex
= get_extents(super
, dl
);
6803 dprintf("cannot get extents\n");
6806 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
6807 dev
= get_imsm_dev(super
, i
);
6808 map
= get_imsm_map(dev
, MAP_0
);
6810 /* check if this disk is already a member of
6813 if (get_imsm_disk_slot(map
, dl
->index
) >= 0)
6819 array_start
= __le32_to_cpu(map
->pba_of_lba0
);
6820 array_end
= array_start
+
6821 __le32_to_cpu(map
->blocks_per_member
) - 1;
6824 /* check that we can start at pba_of_lba0 with
6825 * blocks_per_member of space
6827 if (array_start
>= pos
&& array_end
< ex
[j
].start
) {
6831 pos
= ex
[j
].start
+ ex
[j
].size
;
6833 } while (ex
[j
-1].size
);
6840 if (i
< mpb
->num_raid_devs
) {
6841 dprintf("%x:%x does not have %u to %u available\n",
6842 dl
->major
, dl
->minor
, array_start
, array_end
);
6853 static int imsm_rebuild_allowed(struct supertype
*cont
, int dev_idx
, int failed
)
6855 struct imsm_dev
*dev2
;
6856 struct imsm_map
*map
;
6862 dev2
= get_imsm_dev(cont
->sb
, dev_idx
);
6864 state
= imsm_check_degraded(cont
->sb
, dev2
, failed
, MAP_0
);
6865 if (state
== IMSM_T_STATE_FAILED
) {
6866 map
= get_imsm_map(dev2
, MAP_0
);
6869 for (slot
= 0; slot
< map
->num_members
; slot
++) {
6871 * Check if failed disks are deleted from intel
6872 * disk list or are marked to be deleted
6874 idx
= get_imsm_disk_idx(dev2
, slot
, MAP_X
);
6875 idisk
= get_imsm_dl_disk(cont
->sb
, idx
);
6877 * Do not rebuild the array if failed disks
6878 * from failed sub-array are not removed from
6882 is_failed(&idisk
->disk
) &&
6883 (idisk
->action
!= DISK_REMOVE
))
6891 static struct mdinfo
*imsm_activate_spare(struct active_array
*a
,
6892 struct metadata_update
**updates
)
6895 * Find a device with unused free space and use it to replace a
6896 * failed/vacant region in an array. We replace failed regions one a
6897 * array at a time. The result is that a new spare disk will be added
6898 * to the first failed array and after the monitor has finished
6899 * propagating failures the remainder will be consumed.
6901 * FIXME add a capability for mdmon to request spares from another
6905 struct intel_super
*super
= a
->container
->sb
;
6906 int inst
= a
->info
.container_member
;
6907 struct imsm_dev
*dev
= get_imsm_dev(super
, inst
);
6908 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
6909 int failed
= a
->info
.array
.raid_disks
;
6910 struct mdinfo
*rv
= NULL
;
6913 struct metadata_update
*mu
;
6915 struct imsm_update_activate_spare
*u
;
6920 for (d
= a
->info
.devs
; d
; d
= d
->next
) {
6921 if ((d
->curr_state
& DS_FAULTY
) &&
6923 /* wait for Removal to happen */
6925 if (d
->state_fd
>= 0)
6929 dprintf("imsm: activate spare: inst=%d failed=%d (%d) level=%d\n",
6930 inst
, failed
, a
->info
.array
.raid_disks
, a
->info
.array
.level
);
6932 if (imsm_reshape_blocks_arrays_changes(super
))
6935 /* Cannot activate another spare if rebuild is in progress already
6937 if (is_rebuilding(dev
)) {
6938 dprintf("imsm: No spare activation allowed. "
6939 "Rebuild in progress already.\n");
6943 if (a
->info
.array
.level
== 4)
6944 /* No repair for takeovered array
6945 * imsm doesn't support raid4
6949 if (imsm_check_degraded(super
, dev
, failed
, MAP_0
) !=
6950 IMSM_T_STATE_DEGRADED
)
6954 * If there are any failed disks check state of the other volume.
6955 * Block rebuild if the another one is failed until failed disks
6956 * are removed from container.
6959 dprintf("found failed disks in %.*s, check if there another"
6960 "failed sub-array.\n",
6961 MAX_RAID_SERIAL_LEN
, dev
->volume
);
6962 /* check if states of the other volumes allow for rebuild */
6963 for (i
= 0; i
< super
->anchor
->num_raid_devs
; i
++) {
6965 allowed
= imsm_rebuild_allowed(a
->container
,
6973 /* For each slot, if it is not working, find a spare */
6974 for (i
= 0; i
< a
->info
.array
.raid_disks
; i
++) {
6975 for (d
= a
->info
.devs
; d
; d
= d
->next
)
6976 if (d
->disk
.raid_disk
== i
)
6978 dprintf("found %d: %p %x\n", i
, d
, d
?d
->curr_state
:0);
6979 if (d
&& (d
->state_fd
>= 0))
6983 * OK, this device needs recovery. Try to re-add the
6984 * previous occupant of this slot, if this fails see if
6985 * we can continue the assimilation of a spare that was
6986 * partially assimilated, finally try to activate a new
6989 dl
= imsm_readd(super
, i
, a
);
6991 dl
= imsm_add_spare(super
, i
, a
, 0, rv
);
6993 dl
= imsm_add_spare(super
, i
, a
, 1, rv
);
6997 /* found a usable disk with enough space */
6998 di
= malloc(sizeof(*di
));
7001 memset(di
, 0, sizeof(*di
));
7003 /* dl->index will be -1 in the case we are activating a
7004 * pristine spare. imsm_process_update() will create a
7005 * new index in this case. Once a disk is found to be
7006 * failed in all member arrays it is kicked from the
7009 di
->disk
.number
= dl
->index
;
7011 /* (ab)use di->devs to store a pointer to the device
7014 di
->devs
= (struct mdinfo
*) dl
;
7016 di
->disk
.raid_disk
= i
;
7017 di
->disk
.major
= dl
->major
;
7018 di
->disk
.minor
= dl
->minor
;
7020 di
->recovery_start
= 0;
7021 di
->data_offset
= __le32_to_cpu(map
->pba_of_lba0
);
7022 di
->component_size
= a
->info
.component_size
;
7023 di
->container_member
= inst
;
7024 super
->random
= random32();
7028 dprintf("%x:%x to be %d at %llu\n", dl
->major
, dl
->minor
,
7029 i
, di
->data_offset
);
7033 /* No spares found */
7035 /* Now 'rv' has a list of devices to return.
7036 * Create a metadata_update record to update the
7037 * disk_ord_tbl for the array
7039 mu
= malloc(sizeof(*mu
));
7041 mu
->buf
= malloc(sizeof(struct imsm_update_activate_spare
) * num_spares
);
7042 if (mu
->buf
== NULL
) {
7049 struct mdinfo
*n
= rv
->next
;
7058 mu
->space_list
= NULL
;
7059 mu
->len
= sizeof(struct imsm_update_activate_spare
) * num_spares
;
7060 mu
->next
= *updates
;
7061 u
= (struct imsm_update_activate_spare
*) mu
->buf
;
7063 for (di
= rv
; di
; di
= di
->next
) {
7064 u
->type
= update_activate_spare
;
7065 u
->dl
= (struct dl
*) di
->devs
;
7067 u
->slot
= di
->disk
.raid_disk
;
7078 static int disks_overlap(struct intel_super
*super
, int idx
, struct imsm_update_create_array
*u
)
7080 struct imsm_dev
*dev
= get_imsm_dev(super
, idx
);
7081 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
7082 struct imsm_map
*new_map
= get_imsm_map(&u
->dev
, MAP_0
);
7083 struct disk_info
*inf
= get_disk_info(u
);
7084 struct imsm_disk
*disk
;
7088 for (i
= 0; i
< map
->num_members
; i
++) {
7089 disk
= get_imsm_disk(super
, get_imsm_disk_idx(dev
, i
, MAP_X
));
7090 for (j
= 0; j
< new_map
->num_members
; j
++)
7091 if (serialcmp(disk
->serial
, inf
[j
].serial
) == 0)
7099 static struct dl
*get_disk_super(struct intel_super
*super
, int major
, int minor
)
7101 struct dl
*dl
= NULL
;
7102 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
7103 if ((dl
->major
== major
) && (dl
->minor
== minor
))
7108 static int remove_disk_super(struct intel_super
*super
, int major
, int minor
)
7110 struct dl
*prev
= NULL
;
7114 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
7115 if ((dl
->major
== major
) && (dl
->minor
== minor
)) {
7118 prev
->next
= dl
->next
;
7120 super
->disks
= dl
->next
;
7122 __free_imsm_disk(dl
);
7123 dprintf("%s: removed %x:%x\n",
7124 __func__
, major
, minor
);
7132 static void imsm_delete(struct intel_super
*super
, struct dl
**dlp
, unsigned index
);
7134 static int add_remove_disk_update(struct intel_super
*super
)
7136 int check_degraded
= 0;
7137 struct dl
*disk
= NULL
;
7138 /* add/remove some spares to/from the metadata/contrainer */
7139 while (super
->disk_mgmt_list
) {
7140 struct dl
*disk_cfg
;
7142 disk_cfg
= super
->disk_mgmt_list
;
7143 super
->disk_mgmt_list
= disk_cfg
->next
;
7144 disk_cfg
->next
= NULL
;
7146 if (disk_cfg
->action
== DISK_ADD
) {
7147 disk_cfg
->next
= super
->disks
;
7148 super
->disks
= disk_cfg
;
7150 dprintf("%s: added %x:%x\n",
7151 __func__
, disk_cfg
->major
,
7153 } else if (disk_cfg
->action
== DISK_REMOVE
) {
7154 dprintf("Disk remove action processed: %x.%x\n",
7155 disk_cfg
->major
, disk_cfg
->minor
);
7156 disk
= get_disk_super(super
,
7160 /* store action status */
7161 disk
->action
= DISK_REMOVE
;
7162 /* remove spare disks only */
7163 if (disk
->index
== -1) {
7164 remove_disk_super(super
,
7169 /* release allocate disk structure */
7170 __free_imsm_disk(disk_cfg
);
7173 return check_degraded
;
7177 static int apply_reshape_migration_update(struct imsm_update_reshape_migration
*u
,
7178 struct intel_super
*super
,
7181 struct intel_dev
*id
;
7182 void **tofree
= NULL
;
7185 dprintf("apply_reshape_migration_update()\n");
7186 if ((u
->subdev
< 0) ||
7188 dprintf("imsm: Error: Wrong subdev: %i\n", u
->subdev
);
7191 if ((space_list
== NULL
) || (*space_list
== NULL
)) {
7192 dprintf("imsm: Error: Memory is not allocated\n");
7196 for (id
= super
->devlist
; id
; id
= id
->next
) {
7197 if (id
->index
== (unsigned)u
->subdev
) {
7198 struct imsm_dev
*dev
= get_imsm_dev(super
, u
->subdev
);
7199 struct imsm_map
*map
;
7200 struct imsm_dev
*new_dev
=
7201 (struct imsm_dev
*)*space_list
;
7202 struct imsm_map
*migr_map
= get_imsm_map(dev
, MAP_1
);
7204 struct dl
*new_disk
;
7206 if (new_dev
== NULL
)
7208 *space_list
= **space_list
;
7209 memcpy(new_dev
, dev
, sizeof_imsm_dev(dev
, 0));
7210 map
= get_imsm_map(new_dev
, MAP_0
);
7212 dprintf("imsm: Error: migration in progress");
7216 to_state
= map
->map_state
;
7217 if ((u
->new_level
== 5) && (map
->raid_level
== 0)) {
7219 /* this should not happen */
7220 if (u
->new_disks
[0] < 0) {
7221 map
->failed_disk_num
=
7222 map
->num_members
- 1;
7223 to_state
= IMSM_T_STATE_DEGRADED
;
7225 to_state
= IMSM_T_STATE_NORMAL
;
7227 migrate(new_dev
, super
, to_state
, MIGR_GEN_MIGR
);
7228 if (u
->new_level
> -1)
7229 map
->raid_level
= u
->new_level
;
7230 migr_map
= get_imsm_map(new_dev
, MAP_1
);
7231 if ((u
->new_level
== 5) &&
7232 (migr_map
->raid_level
== 0)) {
7233 int ord
= map
->num_members
- 1;
7234 migr_map
->num_members
--;
7235 if (u
->new_disks
[0] < 0)
7236 ord
|= IMSM_ORD_REBUILD
;
7237 set_imsm_ord_tbl_ent(map
,
7238 map
->num_members
- 1,
7242 tofree
= (void **)dev
;
7244 /* update chunk size
7246 if (u
->new_chunksize
> 0)
7247 map
->blocks_per_strip
=
7248 __cpu_to_le16(u
->new_chunksize
* 2);
7252 if ((u
->new_level
!= 5) ||
7253 (migr_map
->raid_level
!= 0) ||
7254 (migr_map
->raid_level
== map
->raid_level
))
7257 if (u
->new_disks
[0] >= 0) {
7260 new_disk
= get_disk_super(super
,
7261 major(u
->new_disks
[0]),
7262 minor(u
->new_disks
[0]));
7263 dprintf("imsm: new disk for reshape is: %i:%i "
7264 "(%p, index = %i)\n",
7265 major(u
->new_disks
[0]),
7266 minor(u
->new_disks
[0]),
7267 new_disk
, new_disk
->index
);
7268 if (new_disk
== NULL
)
7269 goto error_disk_add
;
7271 new_disk
->index
= map
->num_members
- 1;
7272 /* slot to fill in autolayout
7274 new_disk
->raiddisk
= new_disk
->index
;
7275 new_disk
->disk
.status
|= CONFIGURED_DISK
;
7276 new_disk
->disk
.status
&= ~SPARE_DISK
;
7278 goto error_disk_add
;
7281 *tofree
= *space_list
;
7282 /* calculate new size
7284 imsm_set_array_size(new_dev
);
7291 *space_list
= tofree
;
7295 dprintf("Error: imsm: Cannot find disk.\n");
7299 static int apply_update_activate_spare(struct imsm_update_activate_spare
*u
,
7300 struct intel_super
*super
,
7301 struct active_array
*active_array
)
7303 struct imsm_super
*mpb
= super
->anchor
;
7304 struct imsm_dev
*dev
= get_imsm_dev(super
, u
->array
);
7305 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
7306 struct imsm_map
*migr_map
;
7307 struct active_array
*a
;
7308 struct imsm_disk
*disk
;
7315 int second_map_created
= 0;
7317 for (; u
; u
= u
->next
) {
7318 victim
= get_imsm_disk_idx(dev
, u
->slot
, MAP_X
);
7323 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
7328 fprintf(stderr
, "error: imsm_activate_spare passed "
7329 "an unknown disk (index: %d)\n",
7334 /* count failures (excluding rebuilds and the victim)
7335 * to determine map[0] state
7338 for (i
= 0; i
< map
->num_members
; i
++) {
7341 disk
= get_imsm_disk(super
,
7342 get_imsm_disk_idx(dev
, i
, MAP_X
));
7343 if (!disk
|| is_failed(disk
))
7347 /* adding a pristine spare, assign a new index */
7348 if (dl
->index
< 0) {
7349 dl
->index
= super
->anchor
->num_disks
;
7350 super
->anchor
->num_disks
++;
7353 disk
->status
|= CONFIGURED_DISK
;
7354 disk
->status
&= ~SPARE_DISK
;
7357 to_state
= imsm_check_degraded(super
, dev
, failed
, MAP_0
);
7358 if (!second_map_created
) {
7359 second_map_created
= 1;
7360 map
->map_state
= IMSM_T_STATE_DEGRADED
;
7361 migrate(dev
, super
, to_state
, MIGR_REBUILD
);
7363 map
->map_state
= to_state
;
7364 migr_map
= get_imsm_map(dev
, MAP_1
);
7365 set_imsm_ord_tbl_ent(map
, u
->slot
, dl
->index
);
7366 set_imsm_ord_tbl_ent(migr_map
, u
->slot
,
7367 dl
->index
| IMSM_ORD_REBUILD
);
7369 /* update the family_num to mark a new container
7370 * generation, being careful to record the existing
7371 * family_num in orig_family_num to clean up after
7372 * earlier mdadm versions that neglected to set it.
7374 if (mpb
->orig_family_num
== 0)
7375 mpb
->orig_family_num
= mpb
->family_num
;
7376 mpb
->family_num
+= super
->random
;
7378 /* count arrays using the victim in the metadata */
7380 for (a
= active_array
; a
; a
= a
->next
) {
7381 dev
= get_imsm_dev(super
, a
->info
.container_member
);
7382 map
= get_imsm_map(dev
, MAP_0
);
7384 if (get_imsm_disk_slot(map
, victim
) >= 0)
7388 /* delete the victim if it is no longer being
7394 /* We know that 'manager' isn't touching anything,
7395 * so it is safe to delete
7397 for (dlp
= &super
->disks
; *dlp
; dlp
= &(*dlp
)->next
)
7398 if ((*dlp
)->index
== victim
)
7401 /* victim may be on the missing list */
7403 for (dlp
= &super
->missing
; *dlp
;
7404 dlp
= &(*dlp
)->next
)
7405 if ((*dlp
)->index
== victim
)
7407 imsm_delete(super
, dlp
, victim
);
7414 static int apply_reshape_container_disks_update(struct imsm_update_reshape
*u
,
7415 struct intel_super
*super
,
7418 struct dl
*new_disk
;
7419 struct intel_dev
*id
;
7421 int delta_disks
= u
->new_raid_disks
- u
->old_raid_disks
;
7422 int disk_count
= u
->old_raid_disks
;
7423 void **tofree
= NULL
;
7424 int devices_to_reshape
= 1;
7425 struct imsm_super
*mpb
= super
->anchor
;
7427 unsigned int dev_id
;
7429 dprintf("imsm: apply_reshape_container_disks_update()\n");
7431 /* enable spares to use in array */
7432 for (i
= 0; i
< delta_disks
; i
++) {
7433 new_disk
= get_disk_super(super
,
7434 major(u
->new_disks
[i
]),
7435 minor(u
->new_disks
[i
]));
7436 dprintf("imsm: new disk for reshape is: %i:%i "
7437 "(%p, index = %i)\n",
7438 major(u
->new_disks
[i
]), minor(u
->new_disks
[i
]),
7439 new_disk
, new_disk
->index
);
7440 if ((new_disk
== NULL
) ||
7441 ((new_disk
->index
>= 0) &&
7442 (new_disk
->index
< u
->old_raid_disks
)))
7443 goto update_reshape_exit
;
7444 new_disk
->index
= disk_count
++;
7445 /* slot to fill in autolayout
7447 new_disk
->raiddisk
= new_disk
->index
;
7448 new_disk
->disk
.status
|=
7450 new_disk
->disk
.status
&= ~SPARE_DISK
;
7453 dprintf("imsm: volume set mpb->num_raid_devs = %i\n",
7454 mpb
->num_raid_devs
);
7455 /* manage changes in volume
7457 for (dev_id
= 0; dev_id
< mpb
->num_raid_devs
; dev_id
++) {
7458 void **sp
= *space_list
;
7459 struct imsm_dev
*newdev
;
7460 struct imsm_map
*newmap
, *oldmap
;
7462 for (id
= super
->devlist
; id
; id
= id
->next
) {
7463 if (id
->index
== dev_id
)
7472 /* Copy the dev, but not (all of) the map */
7473 memcpy(newdev
, id
->dev
, sizeof(*newdev
));
7474 oldmap
= get_imsm_map(id
->dev
, MAP_0
);
7475 newmap
= get_imsm_map(newdev
, MAP_0
);
7476 /* Copy the current map */
7477 memcpy(newmap
, oldmap
, sizeof_imsm_map(oldmap
));
7478 /* update one device only
7480 if (devices_to_reshape
) {
7481 dprintf("imsm: modifying subdev: %i\n",
7483 devices_to_reshape
--;
7484 newdev
->vol
.migr_state
= 1;
7485 newdev
->vol
.curr_migr_unit
= 0;
7486 set_migr_type(newdev
, MIGR_GEN_MIGR
);
7487 newmap
->num_members
= u
->new_raid_disks
;
7488 for (i
= 0; i
< delta_disks
; i
++) {
7489 set_imsm_ord_tbl_ent(newmap
,
7490 u
->old_raid_disks
+ i
,
7491 u
->old_raid_disks
+ i
);
7493 /* New map is correct, now need to save old map
7495 newmap
= get_imsm_map(newdev
, MAP_1
);
7496 memcpy(newmap
, oldmap
, sizeof_imsm_map(oldmap
));
7498 imsm_set_array_size(newdev
);
7501 sp
= (void **)id
->dev
;
7506 /* Clear migration record */
7507 memset(super
->migr_rec
, 0, sizeof(struct migr_record
));
7510 *space_list
= tofree
;
7513 update_reshape_exit
:
7518 static int apply_takeover_update(struct imsm_update_takeover
*u
,
7519 struct intel_super
*super
,
7522 struct imsm_dev
*dev
= NULL
;
7523 struct intel_dev
*dv
;
7524 struct imsm_dev
*dev_new
;
7525 struct imsm_map
*map
;
7529 for (dv
= super
->devlist
; dv
; dv
= dv
->next
)
7530 if (dv
->index
== (unsigned int)u
->subarray
) {
7538 map
= get_imsm_map(dev
, MAP_0
);
7540 if (u
->direction
== R10_TO_R0
) {
7541 /* Number of failed disks must be half of initial disk number */
7542 if (imsm_count_failed(super
, dev
, MAP_0
) !=
7543 (map
->num_members
/ 2))
7546 /* iterate through devices to mark removed disks as spare */
7547 for (dm
= super
->disks
; dm
; dm
= dm
->next
) {
7548 if (dm
->disk
.status
& FAILED_DISK
) {
7549 int idx
= dm
->index
;
7550 /* update indexes on the disk list */
7551 /* FIXME this loop-with-the-loop looks wrong, I'm not convinced
7552 the index values will end up being correct.... NB */
7553 for (du
= super
->disks
; du
; du
= du
->next
)
7554 if (du
->index
> idx
)
7556 /* mark as spare disk */
7561 map
->num_members
= map
->num_members
/ 2;
7562 map
->map_state
= IMSM_T_STATE_NORMAL
;
7563 map
->num_domains
= 1;
7564 map
->raid_level
= 0;
7565 map
->failed_disk_num
= -1;
7568 if (u
->direction
== R0_TO_R10
) {
7570 /* update slots in current disk list */
7571 for (dm
= super
->disks
; dm
; dm
= dm
->next
) {
7575 /* create new *missing* disks */
7576 for (i
= 0; i
< map
->num_members
; i
++) {
7577 space
= *space_list
;
7580 *space_list
= *space
;
7582 memcpy(du
, super
->disks
, sizeof(*du
));
7586 du
->index
= (i
* 2) + 1;
7587 sprintf((char *)du
->disk
.serial
,
7588 " MISSING_%d", du
->index
);
7589 sprintf((char *)du
->serial
,
7590 "MISSING_%d", du
->index
);
7591 du
->next
= super
->missing
;
7592 super
->missing
= du
;
7594 /* create new dev and map */
7595 space
= *space_list
;
7598 *space_list
= *space
;
7599 dev_new
= (void *)space
;
7600 memcpy(dev_new
, dev
, sizeof(*dev
));
7601 /* update new map */
7602 map
= get_imsm_map(dev_new
, MAP_0
);
7603 map
->num_members
= map
->num_members
* 2;
7604 map
->map_state
= IMSM_T_STATE_DEGRADED
;
7605 map
->num_domains
= 2;
7606 map
->raid_level
= 1;
7607 /* replace dev<->dev_new */
7610 /* update disk order table */
7611 for (du
= super
->disks
; du
; du
= du
->next
)
7613 set_imsm_ord_tbl_ent(map
, du
->index
, du
->index
);
7614 for (du
= super
->missing
; du
; du
= du
->next
)
7615 if (du
->index
>= 0) {
7616 set_imsm_ord_tbl_ent(map
, du
->index
, du
->index
);
7617 mark_missing(dv
->dev
, &du
->disk
, du
->index
);
7623 static void imsm_process_update(struct supertype
*st
,
7624 struct metadata_update
*update
)
7627 * crack open the metadata_update envelope to find the update record
7628 * update can be one of:
7629 * update_reshape_container_disks - all the arrays in the container
7630 * are being reshaped to have more devices. We need to mark
7631 * the arrays for general migration and convert selected spares
7632 * into active devices.
7633 * update_activate_spare - a spare device has replaced a failed
7634 * device in an array, update the disk_ord_tbl. If this disk is
7635 * present in all member arrays then also clear the SPARE_DISK
7637 * update_create_array
7639 * update_rename_array
7640 * update_add_remove_disk
7642 struct intel_super
*super
= st
->sb
;
7643 struct imsm_super
*mpb
;
7644 enum imsm_update_type type
= *(enum imsm_update_type
*) update
->buf
;
7646 /* update requires a larger buf but the allocation failed */
7647 if (super
->next_len
&& !super
->next_buf
) {
7648 super
->next_len
= 0;
7652 if (super
->next_buf
) {
7653 memcpy(super
->next_buf
, super
->buf
, super
->len
);
7655 super
->len
= super
->next_len
;
7656 super
->buf
= super
->next_buf
;
7658 super
->next_len
= 0;
7659 super
->next_buf
= NULL
;
7662 mpb
= super
->anchor
;
7665 case update_general_migration_checkpoint
: {
7666 struct intel_dev
*id
;
7667 struct imsm_update_general_migration_checkpoint
*u
=
7668 (void *)update
->buf
;
7670 dprintf("imsm: process_update() "
7671 "for update_general_migration_checkpoint called\n");
7673 /* find device under general migration */
7674 for (id
= super
->devlist
; id
; id
= id
->next
) {
7675 if (is_gen_migration(id
->dev
)) {
7676 id
->dev
->vol
.curr_migr_unit
=
7677 __cpu_to_le32(u
->curr_migr_unit
);
7678 super
->updates_pending
++;
7683 case update_takeover
: {
7684 struct imsm_update_takeover
*u
= (void *)update
->buf
;
7685 if (apply_takeover_update(u
, super
, &update
->space_list
)) {
7686 imsm_update_version_info(super
);
7687 super
->updates_pending
++;
7692 case update_reshape_container_disks
: {
7693 struct imsm_update_reshape
*u
= (void *)update
->buf
;
7694 if (apply_reshape_container_disks_update(
7695 u
, super
, &update
->space_list
))
7696 super
->updates_pending
++;
7699 case update_reshape_migration
: {
7700 struct imsm_update_reshape_migration
*u
= (void *)update
->buf
;
7701 if (apply_reshape_migration_update(
7702 u
, super
, &update
->space_list
))
7703 super
->updates_pending
++;
7706 case update_activate_spare
: {
7707 struct imsm_update_activate_spare
*u
= (void *) update
->buf
;
7708 if (apply_update_activate_spare(u
, super
, st
->arrays
))
7709 super
->updates_pending
++;
7712 case update_create_array
: {
7713 /* someone wants to create a new array, we need to be aware of
7714 * a few races/collisions:
7715 * 1/ 'Create' called by two separate instances of mdadm
7716 * 2/ 'Create' versus 'activate_spare': mdadm has chosen
7717 * devices that have since been assimilated via
7719 * In the event this update can not be carried out mdadm will
7720 * (FIX ME) notice that its update did not take hold.
7722 struct imsm_update_create_array
*u
= (void *) update
->buf
;
7723 struct intel_dev
*dv
;
7724 struct imsm_dev
*dev
;
7725 struct imsm_map
*map
, *new_map
;
7726 unsigned long long start
, end
;
7727 unsigned long long new_start
, new_end
;
7729 struct disk_info
*inf
;
7732 /* handle racing creates: first come first serve */
7733 if (u
->dev_idx
< mpb
->num_raid_devs
) {
7734 dprintf("%s: subarray %d already defined\n",
7735 __func__
, u
->dev_idx
);
7739 /* check update is next in sequence */
7740 if (u
->dev_idx
!= mpb
->num_raid_devs
) {
7741 dprintf("%s: can not create array %d expected index %d\n",
7742 __func__
, u
->dev_idx
, mpb
->num_raid_devs
);
7746 new_map
= get_imsm_map(&u
->dev
, MAP_0
);
7747 new_start
= __le32_to_cpu(new_map
->pba_of_lba0
);
7748 new_end
= new_start
+ __le32_to_cpu(new_map
->blocks_per_member
);
7749 inf
= get_disk_info(u
);
7751 /* handle activate_spare versus create race:
7752 * check to make sure that overlapping arrays do not include
7755 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
7756 dev
= get_imsm_dev(super
, i
);
7757 map
= get_imsm_map(dev
, MAP_0
);
7758 start
= __le32_to_cpu(map
->pba_of_lba0
);
7759 end
= start
+ __le32_to_cpu(map
->blocks_per_member
);
7760 if ((new_start
>= start
&& new_start
<= end
) ||
7761 (start
>= new_start
&& start
<= new_end
))
7766 if (disks_overlap(super
, i
, u
)) {
7767 dprintf("%s: arrays overlap\n", __func__
);
7772 /* check that prepare update was successful */
7773 if (!update
->space
) {
7774 dprintf("%s: prepare update failed\n", __func__
);
7778 /* check that all disks are still active before committing
7779 * changes. FIXME: could we instead handle this by creating a
7780 * degraded array? That's probably not what the user expects,
7781 * so better to drop this update on the floor.
7783 for (i
= 0; i
< new_map
->num_members
; i
++) {
7784 dl
= serial_to_dl(inf
[i
].serial
, super
);
7786 dprintf("%s: disk disappeared\n", __func__
);
7791 super
->updates_pending
++;
7793 /* convert spares to members and fixup ord_tbl */
7794 for (i
= 0; i
< new_map
->num_members
; i
++) {
7795 dl
= serial_to_dl(inf
[i
].serial
, super
);
7796 if (dl
->index
== -1) {
7797 dl
->index
= mpb
->num_disks
;
7799 dl
->disk
.status
|= CONFIGURED_DISK
;
7800 dl
->disk
.status
&= ~SPARE_DISK
;
7802 set_imsm_ord_tbl_ent(new_map
, i
, dl
->index
);
7807 update
->space
= NULL
;
7808 imsm_copy_dev(dev
, &u
->dev
);
7809 dv
->index
= u
->dev_idx
;
7810 dv
->next
= super
->devlist
;
7811 super
->devlist
= dv
;
7812 mpb
->num_raid_devs
++;
7814 imsm_update_version_info(super
);
7817 /* mdmon knows how to release update->space, but not
7818 * ((struct intel_dev *) update->space)->dev
7820 if (update
->space
) {
7826 case update_kill_array
: {
7827 struct imsm_update_kill_array
*u
= (void *) update
->buf
;
7828 int victim
= u
->dev_idx
;
7829 struct active_array
*a
;
7830 struct intel_dev
**dp
;
7831 struct imsm_dev
*dev
;
7833 /* sanity check that we are not affecting the uuid of
7834 * active arrays, or deleting an active array
7836 * FIXME when immutable ids are available, but note that
7837 * we'll also need to fixup the invalidated/active
7838 * subarray indexes in mdstat
7840 for (a
= st
->arrays
; a
; a
= a
->next
)
7841 if (a
->info
.container_member
>= victim
)
7843 /* by definition if mdmon is running at least one array
7844 * is active in the container, so checking
7845 * mpb->num_raid_devs is just extra paranoia
7847 dev
= get_imsm_dev(super
, victim
);
7848 if (a
|| !dev
|| mpb
->num_raid_devs
== 1) {
7849 dprintf("failed to delete subarray-%d\n", victim
);
7853 for (dp
= &super
->devlist
; *dp
;)
7854 if ((*dp
)->index
== (unsigned)super
->current_vol
) {
7857 if ((*dp
)->index
> (unsigned)victim
)
7861 mpb
->num_raid_devs
--;
7862 super
->updates_pending
++;
7865 case update_rename_array
: {
7866 struct imsm_update_rename_array
*u
= (void *) update
->buf
;
7867 char name
[MAX_RAID_SERIAL_LEN
+1];
7868 int target
= u
->dev_idx
;
7869 struct active_array
*a
;
7870 struct imsm_dev
*dev
;
7872 /* sanity check that we are not affecting the uuid of
7875 snprintf(name
, MAX_RAID_SERIAL_LEN
, "%s", (char *) u
->name
);
7876 name
[MAX_RAID_SERIAL_LEN
] = '\0';
7877 for (a
= st
->arrays
; a
; a
= a
->next
)
7878 if (a
->info
.container_member
== target
)
7880 dev
= get_imsm_dev(super
, u
->dev_idx
);
7881 if (a
|| !dev
|| !check_name(super
, name
, 1)) {
7882 dprintf("failed to rename subarray-%d\n", target
);
7886 snprintf((char *) dev
->volume
, MAX_RAID_SERIAL_LEN
, "%s", name
);
7887 super
->updates_pending
++;
7890 case update_add_remove_disk
: {
7891 /* we may be able to repair some arrays if disks are
7892 * being added, check teh status of add_remove_disk
7893 * if discs has been added.
7895 if (add_remove_disk_update(super
)) {
7896 struct active_array
*a
;
7898 super
->updates_pending
++;
7899 for (a
= st
->arrays
; a
; a
= a
->next
)
7900 a
->check_degraded
= 1;
7905 fprintf(stderr
, "error: unsuported process update type:"
7906 "(type: %d)\n", type
);
7910 static struct mdinfo
*get_spares_for_grow(struct supertype
*st
);
7912 static void imsm_prepare_update(struct supertype
*st
,
7913 struct metadata_update
*update
)
7916 * Allocate space to hold new disk entries, raid-device entries or a new
7917 * mpb if necessary. The manager synchronously waits for updates to
7918 * complete in the monitor, so new mpb buffers allocated here can be
7919 * integrated by the monitor thread without worrying about live pointers
7920 * in the manager thread.
7922 enum imsm_update_type type
= *(enum imsm_update_type
*) update
->buf
;
7923 struct intel_super
*super
= st
->sb
;
7924 struct imsm_super
*mpb
= super
->anchor
;
7929 case update_general_migration_checkpoint
:
7930 dprintf("imsm: prepare_update() "
7931 "for update_general_migration_checkpoint called\n");
7933 case update_takeover
: {
7934 struct imsm_update_takeover
*u
= (void *)update
->buf
;
7935 if (u
->direction
== R0_TO_R10
) {
7936 void **tail
= (void **)&update
->space_list
;
7937 struct imsm_dev
*dev
= get_imsm_dev(super
, u
->subarray
);
7938 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
7939 int num_members
= map
->num_members
;
7943 /* allocate memory for added disks */
7944 for (i
= 0; i
< num_members
; i
++) {
7945 size
= sizeof(struct dl
);
7946 space
= malloc(size
);
7955 /* allocate memory for new device */
7956 size
= sizeof_imsm_dev(super
->devlist
->dev
, 0) +
7957 (num_members
* sizeof(__u32
));
7958 space
= malloc(size
);
7967 len
= disks_to_mpb_size(num_members
* 2);
7969 /* if allocation didn't success, free buffer */
7970 while (update
->space_list
) {
7971 void **sp
= update
->space_list
;
7972 update
->space_list
= *sp
;
7980 case update_reshape_container_disks
: {
7981 /* Every raid device in the container is about to
7982 * gain some more devices, and we will enter a
7984 * So each 'imsm_map' will be bigger, and the imsm_vol
7985 * will now hold 2 of them.
7986 * Thus we need new 'struct imsm_dev' allocations sized
7987 * as sizeof_imsm_dev but with more devices in both maps.
7989 struct imsm_update_reshape
*u
= (void *)update
->buf
;
7990 struct intel_dev
*dl
;
7991 void **space_tail
= (void**)&update
->space_list
;
7993 dprintf("imsm: imsm_prepare_update() for update_reshape\n");
7995 for (dl
= super
->devlist
; dl
; dl
= dl
->next
) {
7996 int size
= sizeof_imsm_dev(dl
->dev
, 1);
7998 if (u
->new_raid_disks
> u
->old_raid_disks
)
7999 size
+= sizeof(__u32
)*2*
8000 (u
->new_raid_disks
- u
->old_raid_disks
);
8009 len
= disks_to_mpb_size(u
->new_raid_disks
);
8010 dprintf("New anchor length is %llu\n", (unsigned long long)len
);
8013 case update_reshape_migration
: {
8014 /* for migration level 0->5 we need to add disks
8015 * so the same as for container operation we will copy
8016 * device to the bigger location.
8017 * in memory prepared device and new disk area are prepared
8018 * for usage in process update
8020 struct imsm_update_reshape_migration
*u
= (void *)update
->buf
;
8021 struct intel_dev
*id
;
8022 void **space_tail
= (void **)&update
->space_list
;
8025 int current_level
= -1;
8027 dprintf("imsm: imsm_prepare_update() for update_reshape\n");
8029 /* add space for bigger array in update
8031 for (id
= super
->devlist
; id
; id
= id
->next
) {
8032 if (id
->index
== (unsigned)u
->subdev
) {
8033 size
= sizeof_imsm_dev(id
->dev
, 1);
8034 if (u
->new_raid_disks
> u
->old_raid_disks
)
8035 size
+= sizeof(__u32
)*2*
8036 (u
->new_raid_disks
- u
->old_raid_disks
);
8046 if (update
->space_list
== NULL
)
8049 /* add space for disk in update
8051 size
= sizeof(struct dl
);
8054 free(update
->space_list
);
8055 update
->space_list
= NULL
;
8062 /* add spare device to update
8064 for (id
= super
->devlist
; id
; id
= id
->next
)
8065 if (id
->index
== (unsigned)u
->subdev
) {
8066 struct imsm_dev
*dev
;
8067 struct imsm_map
*map
;
8069 dev
= get_imsm_dev(super
, u
->subdev
);
8070 map
= get_imsm_map(dev
, MAP_0
);
8071 current_level
= map
->raid_level
;
8074 if ((u
->new_level
== 5) && (u
->new_level
!= current_level
)) {
8075 struct mdinfo
*spares
;
8077 spares
= get_spares_for_grow(st
);
8085 makedev(dev
->disk
.major
,
8087 dl
= get_disk_super(super
,
8090 dl
->index
= u
->old_raid_disks
;
8096 len
= disks_to_mpb_size(u
->new_raid_disks
);
8097 dprintf("New anchor length is %llu\n", (unsigned long long)len
);
8100 case update_create_array
: {
8101 struct imsm_update_create_array
*u
= (void *) update
->buf
;
8102 struct intel_dev
*dv
;
8103 struct imsm_dev
*dev
= &u
->dev
;
8104 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
8106 struct disk_info
*inf
;
8110 inf
= get_disk_info(u
);
8111 len
= sizeof_imsm_dev(dev
, 1);
8112 /* allocate a new super->devlist entry */
8113 dv
= malloc(sizeof(*dv
));
8115 dv
->dev
= malloc(len
);
8120 update
->space
= NULL
;
8124 /* count how many spares will be converted to members */
8125 for (i
= 0; i
< map
->num_members
; i
++) {
8126 dl
= serial_to_dl(inf
[i
].serial
, super
);
8128 /* hmm maybe it failed?, nothing we can do about
8133 if (count_memberships(dl
, super
) == 0)
8136 len
+= activate
* sizeof(struct imsm_disk
);
8143 /* check if we need a larger metadata buffer */
8144 if (super
->next_buf
)
8145 buf_len
= super
->next_len
;
8147 buf_len
= super
->len
;
8149 if (__le32_to_cpu(mpb
->mpb_size
) + len
> buf_len
) {
8150 /* ok we need a larger buf than what is currently allocated
8151 * if this allocation fails process_update will notice that
8152 * ->next_len is set and ->next_buf is NULL
8154 buf_len
= ROUND_UP(__le32_to_cpu(mpb
->mpb_size
) + len
, 512);
8155 if (super
->next_buf
)
8156 free(super
->next_buf
);
8158 super
->next_len
= buf_len
;
8159 if (posix_memalign(&super
->next_buf
, 512, buf_len
) == 0)
8160 memset(super
->next_buf
, 0, buf_len
);
8162 super
->next_buf
= NULL
;
8166 /* must be called while manager is quiesced */
8167 static void imsm_delete(struct intel_super
*super
, struct dl
**dlp
, unsigned index
)
8169 struct imsm_super
*mpb
= super
->anchor
;
8171 struct imsm_dev
*dev
;
8172 struct imsm_map
*map
;
8173 int i
, j
, num_members
;
8176 dprintf("%s: deleting device[%d] from imsm_super\n",
8179 /* shift all indexes down one */
8180 for (iter
= super
->disks
; iter
; iter
= iter
->next
)
8181 if (iter
->index
> (int)index
)
8183 for (iter
= super
->missing
; iter
; iter
= iter
->next
)
8184 if (iter
->index
> (int)index
)
8187 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
8188 dev
= get_imsm_dev(super
, i
);
8189 map
= get_imsm_map(dev
, MAP_0
);
8190 num_members
= map
->num_members
;
8191 for (j
= 0; j
< num_members
; j
++) {
8192 /* update ord entries being careful not to propagate
8193 * ord-flags to the first map
8195 ord
= get_imsm_ord_tbl_ent(dev
, j
, MAP_X
);
8197 if (ord_to_idx(ord
) <= index
)
8200 map
= get_imsm_map(dev
, MAP_0
);
8201 set_imsm_ord_tbl_ent(map
, j
, ord_to_idx(ord
- 1));
8202 map
= get_imsm_map(dev
, MAP_1
);
8204 set_imsm_ord_tbl_ent(map
, j
, ord
- 1);
8209 super
->updates_pending
++;
8211 struct dl
*dl
= *dlp
;
8213 *dlp
= (*dlp
)->next
;
8214 __free_imsm_disk(dl
);
8217 #endif /* MDASSEMBLE */
8219 static void close_targets(int *targets
, int new_disks
)
8226 for (i
= 0; i
< new_disks
; i
++) {
8227 if (targets
[i
] >= 0) {
8234 static int imsm_get_allowed_degradation(int level
, int raid_disks
,
8235 struct intel_super
*super
,
8236 struct imsm_dev
*dev
)
8241 struct imsm_map
*map
;
8244 ret_val
= raid_disks
/2;
8245 /* check map if all disks pairs not failed
8248 map
= get_imsm_map(dev
, MAP_0
);
8249 for (i
= 0; i
< ret_val
; i
++) {
8250 int degradation
= 0;
8251 if (get_imsm_disk(super
, i
) == NULL
)
8253 if (get_imsm_disk(super
, i
+ 1) == NULL
)
8255 if (degradation
== 2)
8258 map
= get_imsm_map(dev
, MAP_1
);
8259 /* if there is no second map
8260 * result can be returned
8264 /* check degradation in second map
8266 for (i
= 0; i
< ret_val
; i
++) {
8267 int degradation
= 0;
8268 if (get_imsm_disk(super
, i
) == NULL
)
8270 if (get_imsm_disk(super
, i
+ 1) == NULL
)
8272 if (degradation
== 2)
8287 /*******************************************************************************
8288 * Function: open_backup_targets
8289 * Description: Function opens file descriptors for all devices given in
8292 * info : general array info
8293 * raid_disks : number of disks
8294 * raid_fds : table of device's file descriptors
8295 * super : intel super for raid10 degradation check
8296 * dev : intel device for raid10 degradation check
8300 ******************************************************************************/
8301 int open_backup_targets(struct mdinfo
*info
, int raid_disks
, int *raid_fds
,
8302 struct intel_super
*super
, struct imsm_dev
*dev
)
8308 for (i
= 0; i
< raid_disks
; i
++)
8311 for (sd
= info
->devs
; sd
; sd
= sd
->next
) {
8314 if (sd
->disk
.state
& (1<<MD_DISK_FAULTY
)) {
8315 dprintf("disk is faulty!!\n");
8319 if ((sd
->disk
.raid_disk
>= raid_disks
) ||
8320 (sd
->disk
.raid_disk
< 0))
8323 dn
= map_dev(sd
->disk
.major
,
8325 raid_fds
[sd
->disk
.raid_disk
] = dev_open(dn
, O_RDWR
);
8326 if (raid_fds
[sd
->disk
.raid_disk
] < 0) {
8327 fprintf(stderr
, "cannot open component\n");
8332 /* check if maximum array degradation level is not exceeded
8334 if ((raid_disks
- opened
) >
8335 imsm_get_allowed_degradation(info
->new_level
,
8338 fprintf(stderr
, "Not enough disks can be opened.\n");
8339 close_targets(raid_fds
, raid_disks
);
8346 /*******************************************************************************
8347 * Function: init_migr_record_imsm
8348 * Description: Function inits imsm migration record
8350 * super : imsm internal array info
8351 * dev : device under migration
8352 * info : general array info to find the smallest device
8355 ******************************************************************************/
8356 void init_migr_record_imsm(struct supertype
*st
, struct imsm_dev
*dev
,
8357 struct mdinfo
*info
)
8359 struct intel_super
*super
= st
->sb
;
8360 struct migr_record
*migr_rec
= super
->migr_rec
;
8362 unsigned long long dsize
, dev_sectors
;
8363 long long unsigned min_dev_sectors
= -1LLU;
8367 struct imsm_map
*map_dest
= get_imsm_map(dev
, MAP_0
);
8368 struct imsm_map
*map_src
= get_imsm_map(dev
, MAP_1
);
8369 unsigned long long num_migr_units
;
8370 unsigned long long array_blocks
;
8372 memset(migr_rec
, 0, sizeof(struct migr_record
));
8373 migr_rec
->family_num
= __cpu_to_le32(super
->anchor
->family_num
);
8375 /* only ascending reshape supported now */
8376 migr_rec
->ascending_migr
= __cpu_to_le32(1);
8378 migr_rec
->dest_depth_per_unit
= GEN_MIGR_AREA_SIZE
/
8379 max(map_dest
->blocks_per_strip
, map_src
->blocks_per_strip
);
8380 migr_rec
->dest_depth_per_unit
*= map_dest
->blocks_per_strip
;
8381 new_data_disks
= imsm_num_data_members(dev
, MAP_0
);
8382 migr_rec
->blocks_per_unit
=
8383 __cpu_to_le32(migr_rec
->dest_depth_per_unit
* new_data_disks
);
8384 migr_rec
->dest_depth_per_unit
=
8385 __cpu_to_le32(migr_rec
->dest_depth_per_unit
);
8386 array_blocks
= info
->component_size
* new_data_disks
;
8388 array_blocks
/ __le32_to_cpu(migr_rec
->blocks_per_unit
);
8390 if (array_blocks
% __le32_to_cpu(migr_rec
->blocks_per_unit
))
8392 migr_rec
->num_migr_units
= __cpu_to_le32(num_migr_units
);
8394 migr_rec
->post_migr_vol_cap
= dev
->size_low
;
8395 migr_rec
->post_migr_vol_cap_hi
= dev
->size_high
;
8398 /* Find the smallest dev */
8399 for (sd
= info
->devs
; sd
; sd
= sd
->next
) {
8400 sprintf(nm
, "%d:%d", sd
->disk
.major
, sd
->disk
.minor
);
8401 fd
= dev_open(nm
, O_RDONLY
);
8404 get_dev_size(fd
, NULL
, &dsize
);
8405 dev_sectors
= dsize
/ 512;
8406 if (dev_sectors
< min_dev_sectors
)
8407 min_dev_sectors
= dev_sectors
;
8410 migr_rec
->ckpt_area_pba
= __cpu_to_le32(min_dev_sectors
-
8411 RAID_DISK_RESERVED_BLOCKS_IMSM_HI
);
8413 write_imsm_migr_rec(st
);
8418 /*******************************************************************************
8419 * Function: save_backup_imsm
8420 * Description: Function saves critical data stripes to Migration Copy Area
8421 * and updates the current migration unit status.
8422 * Use restore_stripes() to form a destination stripe,
8423 * and to write it to the Copy Area.
8425 * st : supertype information
8426 * dev : imsm device that backup is saved for
8427 * info : general array info
8428 * buf : input buffer
8429 * length : length of data to backup (blocks_per_unit)
8433 ******************************************************************************/
8434 int save_backup_imsm(struct supertype
*st
,
8435 struct imsm_dev
*dev
,
8436 struct mdinfo
*info
,
8441 struct intel_super
*super
= st
->sb
;
8442 unsigned long long *target_offsets
= NULL
;
8443 int *targets
= NULL
;
8445 struct imsm_map
*map_dest
= get_imsm_map(dev
, MAP_0
);
8446 int new_disks
= map_dest
->num_members
;
8447 int dest_layout
= 0;
8449 unsigned long long start
;
8450 int data_disks
= imsm_num_data_members(dev
, MAP_0
);
8452 targets
= malloc(new_disks
* sizeof(int));
8456 for (i
= 0; i
< new_disks
; i
++)
8459 target_offsets
= malloc(new_disks
* sizeof(unsigned long long));
8460 if (!target_offsets
)
8463 start
= info
->reshape_progress
* 512;
8464 for (i
= 0; i
< new_disks
; i
++) {
8465 target_offsets
[i
] = (unsigned long long)
8466 __le32_to_cpu(super
->migr_rec
->ckpt_area_pba
) * 512;
8467 /* move back copy area adderss, it will be moved forward
8468 * in restore_stripes() using start input variable
8470 target_offsets
[i
] -= start
/data_disks
;
8473 if (open_backup_targets(info
, new_disks
, targets
,
8477 dest_layout
= imsm_level_to_layout(map_dest
->raid_level
);
8478 dest_chunk
= __le16_to_cpu(map_dest
->blocks_per_strip
) * 512;
8480 if (restore_stripes(targets
, /* list of dest devices */
8481 target_offsets
, /* migration record offsets */
8484 map_dest
->raid_level
,
8486 -1, /* source backup file descriptor */
8487 0, /* input buf offset
8488 * always 0 buf is already offseted */
8492 fprintf(stderr
, Name
": Error restoring stripes\n");
8500 close_targets(targets
, new_disks
);
8503 free(target_offsets
);
8508 /*******************************************************************************
8509 * Function: save_checkpoint_imsm
8510 * Description: Function called for current unit status update
8511 * in the migration record. It writes it to disk.
8513 * super : imsm internal array info
8514 * info : general array info
8518 * 2: failure, means no valid migration record
8519 * / no general migration in progress /
8520 ******************************************************************************/
8521 int save_checkpoint_imsm(struct supertype
*st
, struct mdinfo
*info
, int state
)
8523 struct intel_super
*super
= st
->sb
;
8524 unsigned long long blocks_per_unit
;
8525 unsigned long long curr_migr_unit
;
8527 if (load_imsm_migr_rec(super
, info
) != 0) {
8528 dprintf("imsm: ERROR: Cannot read migration record "
8529 "for checkpoint save.\n");
8533 blocks_per_unit
= __le32_to_cpu(super
->migr_rec
->blocks_per_unit
);
8534 if (blocks_per_unit
== 0) {
8535 dprintf("imsm: no migration in progress.\n");
8538 curr_migr_unit
= info
->reshape_progress
/ blocks_per_unit
;
8539 /* check if array is alligned to copy area
8540 * if it is not alligned, add one to current migration unit value
8541 * this can happend on array reshape finish only
8543 if (info
->reshape_progress
% blocks_per_unit
)
8546 super
->migr_rec
->curr_migr_unit
=
8547 __cpu_to_le32(curr_migr_unit
);
8548 super
->migr_rec
->rec_status
= __cpu_to_le32(state
);
8549 super
->migr_rec
->dest_1st_member_lba
=
8550 __cpu_to_le32(curr_migr_unit
*
8551 __le32_to_cpu(super
->migr_rec
->dest_depth_per_unit
));
8552 if (write_imsm_migr_rec(st
) < 0) {
8553 dprintf("imsm: Cannot write migration record "
8554 "outside backup area\n");
8561 /*******************************************************************************
8562 * Function: recover_backup_imsm
8563 * Description: Function recovers critical data from the Migration Copy Area
8564 * while assembling an array.
8566 * super : imsm internal array info
8567 * info : general array info
8569 * 0 : success (or there is no data to recover)
8571 ******************************************************************************/
8572 int recover_backup_imsm(struct supertype
*st
, struct mdinfo
*info
)
8574 struct intel_super
*super
= st
->sb
;
8575 struct migr_record
*migr_rec
= super
->migr_rec
;
8576 struct imsm_map
*map_dest
= NULL
;
8577 struct intel_dev
*id
= NULL
;
8578 unsigned long long read_offset
;
8579 unsigned long long write_offset
;
8581 int *targets
= NULL
;
8582 int new_disks
, i
, err
;
8585 unsigned long curr_migr_unit
= __le32_to_cpu(migr_rec
->curr_migr_unit
);
8586 unsigned long num_migr_units
= __le32_to_cpu(migr_rec
->num_migr_units
);
8588 int skipped_disks
= 0;
8590 err
= sysfs_get_str(info
, NULL
, "array_state", (char *)buffer
, 20);
8594 /* recover data only during assemblation */
8595 if (strncmp(buffer
, "inactive", 8) != 0)
8597 /* no data to recover */
8598 if (__le32_to_cpu(migr_rec
->rec_status
) == UNIT_SRC_NORMAL
)
8600 if (curr_migr_unit
>= num_migr_units
)
8603 /* find device during reshape */
8604 for (id
= super
->devlist
; id
; id
= id
->next
)
8605 if (is_gen_migration(id
->dev
))
8610 map_dest
= get_imsm_map(id
->dev
, MAP_0
);
8611 new_disks
= map_dest
->num_members
;
8613 read_offset
= (unsigned long long)
8614 __le32_to_cpu(migr_rec
->ckpt_area_pba
) * 512;
8616 write_offset
= ((unsigned long long)
8617 __le32_to_cpu(migr_rec
->dest_1st_member_lba
) +
8618 __le32_to_cpu(map_dest
->pba_of_lba0
)) * 512;
8620 unit_len
= __le32_to_cpu(migr_rec
->dest_depth_per_unit
) * 512;
8621 if (posix_memalign((void **)&buf
, 512, unit_len
) != 0)
8623 targets
= malloc(new_disks
* sizeof(int));
8627 if (open_backup_targets(info
, new_disks
, targets
, super
, id
->dev
)) {
8629 Name
": Cannot open some devices belonging to array.\n");
8633 for (i
= 0; i
< new_disks
; i
++) {
8634 if (targets
[i
] < 0) {
8638 if (lseek64(targets
[i
], read_offset
, SEEK_SET
) < 0) {
8640 Name
": Cannot seek to block: %s\n",
8645 if ((unsigned)read(targets
[i
], buf
, unit_len
) != unit_len
) {
8647 Name
": Cannot read copy area block: %s\n",
8652 if (lseek64(targets
[i
], write_offset
, SEEK_SET
) < 0) {
8654 Name
": Cannot seek to block: %s\n",
8659 if ((unsigned)write(targets
[i
], buf
, unit_len
) != unit_len
) {
8661 Name
": Cannot restore block: %s\n",
8668 if (skipped_disks
> imsm_get_allowed_degradation(info
->new_level
,
8673 Name
": Cannot restore data from backup."
8674 " Too many failed disks\n");
8678 if (save_checkpoint_imsm(st
, info
, UNIT_SRC_NORMAL
)) {
8679 /* ignore error == 2, this can mean end of reshape here
8681 dprintf("imsm: Cannot write checkpoint to "
8682 "migration record (UNIT_SRC_NORMAL) during restart\n");
8688 for (i
= 0; i
< new_disks
; i
++)
8697 static char disk_by_path
[] = "/dev/disk/by-path/";
8699 static const char *imsm_get_disk_controller_domain(const char *path
)
8701 char disk_path
[PATH_MAX
];
8705 strncpy(disk_path
, disk_by_path
, PATH_MAX
- 1);
8706 strncat(disk_path
, path
, PATH_MAX
- strlen(disk_path
) - 1);
8707 if (stat(disk_path
, &st
) == 0) {
8708 struct sys_dev
* hba
;
8711 path
= devt_to_devpath(st
.st_rdev
);
8714 hba
= find_disk_attached_hba(-1, path
);
8715 if (hba
&& hba
->type
== SYS_DEV_SAS
)
8717 else if (hba
&& hba
->type
== SYS_DEV_SATA
)
8721 dprintf("path: %s hba: %s attached: %s\n",
8722 path
, (hba
) ? hba
->path
: "NULL", drv
);
8730 static int imsm_find_array_minor_by_subdev(int subdev
, int container
, int *minor
)
8732 char subdev_name
[20];
8733 struct mdstat_ent
*mdstat
;
8735 sprintf(subdev_name
, "%d", subdev
);
8736 mdstat
= mdstat_by_subdev(subdev_name
, container
);
8740 *minor
= mdstat
->devnum
;
8741 free_mdstat(mdstat
);
8745 static int imsm_reshape_is_allowed_on_container(struct supertype
*st
,
8746 struct geo_params
*geo
,
8747 int *old_raid_disks
)
8749 /* currently we only support increasing the number of devices
8750 * for a container. This increases the number of device for each
8751 * member array. They must all be RAID0 or RAID5.
8754 struct mdinfo
*info
, *member
;
8755 int devices_that_can_grow
= 0;
8757 dprintf("imsm: imsm_reshape_is_allowed_on_container(ENTER): "
8758 "st->devnum = (%i)\n",
8761 if (geo
->size
!= -1 ||
8762 geo
->level
!= UnSet
||
8763 geo
->layout
!= UnSet
||
8764 geo
->chunksize
!= 0 ||
8765 geo
->raid_disks
== UnSet
) {
8766 dprintf("imsm: Container operation is allowed for "
8767 "raid disks number change only.\n");
8771 info
= container_content_imsm(st
, NULL
);
8772 for (member
= info
; member
; member
= member
->next
) {
8776 dprintf("imsm: checking device_num: %i\n",
8777 member
->container_member
);
8779 if (geo
->raid_disks
<= member
->array
.raid_disks
) {
8780 /* we work on container for Online Capacity Expansion
8781 * only so raid_disks has to grow
8783 dprintf("imsm: for container operation raid disks "
8784 "increase is required\n");
8788 if ((info
->array
.level
!= 0) &&
8789 (info
->array
.level
!= 5)) {
8790 /* we cannot use this container with other raid level
8792 dprintf("imsm: for container operation wrong"
8793 " raid level (%i) detected\n",
8797 /* check for platform support
8798 * for this raid level configuration
8800 struct intel_super
*super
= st
->sb
;
8801 if (!is_raid_level_supported(super
->orom
,
8802 member
->array
.level
,
8804 dprintf("platform does not support raid%d with"
8808 geo
->raid_disks
> 1 ? "s" : "");
8811 /* check if component size is aligned to chunk size
8813 if (info
->component_size
%
8814 (info
->array
.chunk_size
/512)) {
8815 dprintf("Component size is not aligned to "
8821 if (*old_raid_disks
&&
8822 info
->array
.raid_disks
!= *old_raid_disks
)
8824 *old_raid_disks
= info
->array
.raid_disks
;
8826 /* All raid5 and raid0 volumes in container
8827 * have to be ready for Online Capacity Expansion
8828 * so they need to be assembled. We have already
8829 * checked that no recovery etc is happening.
8831 result
= imsm_find_array_minor_by_subdev(member
->container_member
,
8835 dprintf("imsm: cannot find array\n");
8838 devices_that_can_grow
++;
8841 if (!member
&& devices_that_can_grow
)
8845 dprintf("\tContainer operation allowed\n");
8847 dprintf("\tError: %i\n", ret_val
);
8852 /* Function: get_spares_for_grow
8853 * Description: Allocates memory and creates list of spare devices
8854 * avaliable in container. Checks if spare drive size is acceptable.
8855 * Parameters: Pointer to the supertype structure
8856 * Returns: Pointer to the list of spare devices (mdinfo structure) on success,
8859 static struct mdinfo
*get_spares_for_grow(struct supertype
*st
)
8861 unsigned long long min_size
= min_acceptable_spare_size_imsm(st
);
8862 return container_choose_spares(st
, min_size
, NULL
, NULL
, NULL
, 0);
8865 /******************************************************************************
8866 * function: imsm_create_metadata_update_for_reshape
8867 * Function creates update for whole IMSM container.
8869 ******************************************************************************/
8870 static int imsm_create_metadata_update_for_reshape(
8871 struct supertype
*st
,
8872 struct geo_params
*geo
,
8874 struct imsm_update_reshape
**updatep
)
8876 struct intel_super
*super
= st
->sb
;
8877 struct imsm_super
*mpb
= super
->anchor
;
8878 int update_memory_size
= 0;
8879 struct imsm_update_reshape
*u
= NULL
;
8880 struct mdinfo
*spares
= NULL
;
8882 int delta_disks
= 0;
8885 dprintf("imsm_update_metadata_for_reshape(enter) raid_disks = %i\n",
8888 delta_disks
= geo
->raid_disks
- old_raid_disks
;
8890 /* size of all update data without anchor */
8891 update_memory_size
= sizeof(struct imsm_update_reshape
);
8893 /* now add space for spare disks that we need to add. */
8894 update_memory_size
+= sizeof(u
->new_disks
[0]) * (delta_disks
- 1);
8896 u
= calloc(1, update_memory_size
);
8899 "cannot get memory for imsm_update_reshape update\n");
8902 u
->type
= update_reshape_container_disks
;
8903 u
->old_raid_disks
= old_raid_disks
;
8904 u
->new_raid_disks
= geo
->raid_disks
;
8906 /* now get spare disks list
8908 spares
= get_spares_for_grow(st
);
8911 || delta_disks
> spares
->array
.spare_disks
) {
8912 fprintf(stderr
, Name
": imsm: ERROR: Cannot get spare devices "
8913 "for %s.\n", geo
->dev_name
);
8918 /* we have got spares
8919 * update disk list in imsm_disk list table in anchor
8921 dprintf("imsm: %i spares are available.\n\n",
8922 spares
->array
.spare_disks
);
8925 for (i
= 0; i
< delta_disks
; i
++) {
8930 u
->new_disks
[i
] = makedev(dev
->disk
.major
,
8932 dl
= get_disk_super(super
, dev
->disk
.major
, dev
->disk
.minor
);
8933 dl
->index
= mpb
->num_disks
;
8943 dprintf("imsm: reshape update preparation :");
8944 if (i
== delta_disks
) {
8947 return update_memory_size
;
8950 dprintf(" Error\n");
8955 /******************************************************************************
8956 * function: imsm_create_metadata_update_for_migration()
8957 * Creates update for IMSM array.
8959 ******************************************************************************/
8960 static int imsm_create_metadata_update_for_migration(
8961 struct supertype
*st
,
8962 struct geo_params
*geo
,
8963 struct imsm_update_reshape_migration
**updatep
)
8965 struct intel_super
*super
= st
->sb
;
8966 int update_memory_size
= 0;
8967 struct imsm_update_reshape_migration
*u
= NULL
;
8968 struct imsm_dev
*dev
;
8969 int previous_level
= -1;
8971 dprintf("imsm_create_metadata_update_for_migration(enter)"
8972 " New Level = %i\n", geo
->level
);
8974 /* size of all update data without anchor */
8975 update_memory_size
= sizeof(struct imsm_update_reshape_migration
);
8977 u
= calloc(1, update_memory_size
);
8979 dprintf("error: cannot get memory for "
8980 "imsm_create_metadata_update_for_migration\n");
8983 u
->type
= update_reshape_migration
;
8984 u
->subdev
= super
->current_vol
;
8985 u
->new_level
= geo
->level
;
8986 u
->new_layout
= geo
->layout
;
8987 u
->new_raid_disks
= u
->old_raid_disks
= geo
->raid_disks
;
8988 u
->new_disks
[0] = -1;
8989 u
->new_chunksize
= -1;
8991 dev
= get_imsm_dev(super
, u
->subdev
);
8993 struct imsm_map
*map
;
8995 map
= get_imsm_map(dev
, MAP_0
);
8997 int current_chunk_size
=
8998 __le16_to_cpu(map
->blocks_per_strip
) / 2;
9000 if (geo
->chunksize
!= current_chunk_size
) {
9001 u
->new_chunksize
= geo
->chunksize
/ 1024;
9003 "chunk size change from %i to %i\n",
9004 current_chunk_size
, u
->new_chunksize
);
9006 previous_level
= map
->raid_level
;
9009 if ((geo
->level
== 5) && (previous_level
== 0)) {
9010 struct mdinfo
*spares
= NULL
;
9012 u
->new_raid_disks
++;
9013 spares
= get_spares_for_grow(st
);
9014 if ((spares
== NULL
) || (spares
->array
.spare_disks
< 1)) {
9017 update_memory_size
= 0;
9018 dprintf("error: cannot get spare device "
9019 "for requested migration");
9024 dprintf("imsm: reshape update preparation : OK\n");
9027 return update_memory_size
;
9030 static void imsm_update_metadata_locally(struct supertype
*st
,
9033 struct metadata_update mu
;
9038 mu
.space_list
= NULL
;
9040 imsm_prepare_update(st
, &mu
);
9041 imsm_process_update(st
, &mu
);
9043 while (mu
.space_list
) {
9044 void **space
= mu
.space_list
;
9045 mu
.space_list
= *space
;
9050 /***************************************************************************
9051 * Function: imsm_analyze_change
9052 * Description: Function analyze change for single volume
9053 * and validate if transition is supported
9054 * Parameters: Geometry parameters, supertype structure
9055 * Returns: Operation type code on success, -1 if fail
9056 ****************************************************************************/
9057 enum imsm_reshape_type
imsm_analyze_change(struct supertype
*st
,
9058 struct geo_params
*geo
)
9064 /* number of added/removed disks in operation result */
9065 int devNumChange
= 0;
9066 /* imsm compatible layout value for array geometry verification */
9067 int imsm_layout
= -1;
9069 getinfo_super_imsm_volume(st
, &info
, NULL
);
9070 if ((geo
->level
!= info
.array
.level
) &&
9071 (geo
->level
>= 0) &&
9072 (geo
->level
!= UnSet
)) {
9073 switch (info
.array
.level
) {
9075 if (geo
->level
== 5) {
9076 change
= CH_MIGRATION
;
9077 if (geo
->layout
!= ALGORITHM_LEFT_ASYMMETRIC
) {
9079 Name
" Error. Requested Layout "
9080 "not supported (left-asymmetric layout "
9081 "is supported only)!\n");
9083 goto analyse_change_exit
;
9085 imsm_layout
= geo
->layout
;
9087 devNumChange
= 1; /* parity disk added */
9088 } else if (geo
->level
== 10) {
9089 change
= CH_TAKEOVER
;
9091 devNumChange
= 2; /* two mirrors added */
9092 imsm_layout
= 0x102; /* imsm supported layout */
9097 if (geo
->level
== 0) {
9098 change
= CH_TAKEOVER
;
9100 devNumChange
= -(geo
->raid_disks
/2);
9101 imsm_layout
= 0; /* imsm raid0 layout */
9107 Name
" Error. Level Migration from %d to %d "
9109 info
.array
.level
, geo
->level
);
9110 goto analyse_change_exit
;
9113 geo
->level
= info
.array
.level
;
9115 if ((geo
->layout
!= info
.array
.layout
)
9116 && ((geo
->layout
!= UnSet
) && (geo
->layout
!= -1))) {
9117 change
= CH_MIGRATION
;
9118 if ((info
.array
.layout
== 0)
9119 && (info
.array
.level
== 5)
9120 && (geo
->layout
== 5)) {
9121 /* reshape 5 -> 4 */
9122 } else if ((info
.array
.layout
== 5)
9123 && (info
.array
.level
== 5)
9124 && (geo
->layout
== 0)) {
9125 /* reshape 4 -> 5 */
9130 Name
" Error. Layout Migration from %d to %d "
9132 info
.array
.layout
, geo
->layout
);
9134 goto analyse_change_exit
;
9137 geo
->layout
= info
.array
.layout
;
9138 if (imsm_layout
== -1)
9139 imsm_layout
= info
.array
.layout
;
9142 if ((geo
->chunksize
> 0) && (geo
->chunksize
!= UnSet
)
9143 && (geo
->chunksize
!= info
.array
.chunk_size
))
9144 change
= CH_MIGRATION
;
9146 geo
->chunksize
= info
.array
.chunk_size
;
9148 chunk
= geo
->chunksize
/ 1024;
9149 if (!validate_geometry_imsm(st
,
9152 geo
->raid_disks
+ devNumChange
,
9159 struct intel_super
*super
= st
->sb
;
9160 struct imsm_super
*mpb
= super
->anchor
;
9162 if (mpb
->num_raid_devs
> 1) {
9164 Name
" Error. Cannot perform operation on %s"
9165 "- for this operation it MUST be single "
9166 "array in container\n",
9172 analyse_change_exit
:
9177 int imsm_takeover(struct supertype
*st
, struct geo_params
*geo
)
9179 struct intel_super
*super
= st
->sb
;
9180 struct imsm_update_takeover
*u
;
9182 u
= malloc(sizeof(struct imsm_update_takeover
));
9186 u
->type
= update_takeover
;
9187 u
->subarray
= super
->current_vol
;
9189 /* 10->0 transition */
9190 if (geo
->level
== 0)
9191 u
->direction
= R10_TO_R0
;
9193 /* 0->10 transition */
9194 if (geo
->level
== 10)
9195 u
->direction
= R0_TO_R10
;
9197 /* update metadata locally */
9198 imsm_update_metadata_locally(st
, u
,
9199 sizeof(struct imsm_update_takeover
));
9200 /* and possibly remotely */
9201 if (st
->update_tail
)
9202 append_metadata_update(st
, u
,
9203 sizeof(struct imsm_update_takeover
));
9210 static int imsm_reshape_super(struct supertype
*st
, long long size
, int level
,
9211 int layout
, int chunksize
, int raid_disks
,
9212 int delta_disks
, char *backup
, char *dev
,
9216 struct geo_params geo
;
9218 dprintf("imsm: reshape_super called.\n");
9220 memset(&geo
, 0, sizeof(struct geo_params
));
9223 geo
.dev_id
= st
->devnum
;
9226 geo
.layout
= layout
;
9227 geo
.chunksize
= chunksize
;
9228 geo
.raid_disks
= raid_disks
;
9229 if (delta_disks
!= UnSet
)
9230 geo
.raid_disks
+= delta_disks
;
9232 dprintf("\tfor level : %i\n", geo
.level
);
9233 dprintf("\tfor raid_disks : %i\n", geo
.raid_disks
);
9235 if (experimental() == 0)
9238 if (st
->container_dev
== st
->devnum
) {
9239 /* On container level we can only increase number of devices. */
9240 dprintf("imsm: info: Container operation\n");
9241 int old_raid_disks
= 0;
9243 if (imsm_reshape_is_allowed_on_container(
9244 st
, &geo
, &old_raid_disks
)) {
9245 struct imsm_update_reshape
*u
= NULL
;
9248 len
= imsm_create_metadata_update_for_reshape(
9249 st
, &geo
, old_raid_disks
, &u
);
9252 dprintf("imsm: Cannot prepare update\n");
9253 goto exit_imsm_reshape_super
;
9257 /* update metadata locally */
9258 imsm_update_metadata_locally(st
, u
, len
);
9259 /* and possibly remotely */
9260 if (st
->update_tail
)
9261 append_metadata_update(st
, u
, len
);
9266 fprintf(stderr
, Name
": (imsm) Operation "
9267 "is not allowed on this container\n");
9270 /* On volume level we support following operations
9271 * - takeover: raid10 -> raid0; raid0 -> raid10
9272 * - chunk size migration
9273 * - migration: raid5 -> raid0; raid0 -> raid5
9275 struct intel_super
*super
= st
->sb
;
9276 struct intel_dev
*dev
= super
->devlist
;
9278 dprintf("imsm: info: Volume operation\n");
9279 /* find requested device */
9281 if (imsm_find_array_minor_by_subdev(
9282 dev
->index
, st
->container_dev
, &devnum
) == 0
9283 && devnum
== geo
.dev_id
)
9288 fprintf(stderr
, Name
" Cannot find %s (%i) subarray\n",
9289 geo
.dev_name
, geo
.dev_id
);
9290 goto exit_imsm_reshape_super
;
9292 super
->current_vol
= dev
->index
;
9293 change
= imsm_analyze_change(st
, &geo
);
9296 ret_val
= imsm_takeover(st
, &geo
);
9298 case CH_MIGRATION
: {
9299 struct imsm_update_reshape_migration
*u
= NULL
;
9301 imsm_create_metadata_update_for_migration(
9305 "Cannot prepare update\n");
9309 /* update metadata locally */
9310 imsm_update_metadata_locally(st
, u
, len
);
9311 /* and possibly remotely */
9312 if (st
->update_tail
)
9313 append_metadata_update(st
, u
, len
);
9323 exit_imsm_reshape_super
:
9324 dprintf("imsm: reshape_super Exit code = %i\n", ret_val
);
9328 /*******************************************************************************
9329 * Function: wait_for_reshape_imsm
9330 * Description: Function writes new sync_max value and waits until
9331 * reshape process reach new position
9333 * sra : general array info
9334 * ndata : number of disks in new array's layout
9337 * 1 : there is no reshape in progress,
9339 ******************************************************************************/
9340 int wait_for_reshape_imsm(struct mdinfo
*sra
, int ndata
)
9342 int fd
= sysfs_get_fd(sra
, NULL
, "reshape_position");
9343 unsigned long long completed
;
9344 /* to_complete : new sync_max position */
9345 unsigned long long to_complete
= sra
->reshape_progress
;
9346 unsigned long long position_to_set
= to_complete
/ ndata
;
9349 dprintf("imsm: wait_for_reshape_imsm() "
9350 "cannot open reshape_position\n");
9354 if (sysfs_fd_get_ll(fd
, &completed
) < 0) {
9355 dprintf("imsm: wait_for_reshape_imsm() "
9356 "cannot read reshape_position (no reshape in progres)\n");
9361 if (completed
> to_complete
) {
9362 dprintf("imsm: wait_for_reshape_imsm() "
9363 "wrong next position to set %llu (%llu)\n",
9364 to_complete
, completed
);
9368 dprintf("Position set: %llu\n", position_to_set
);
9369 if (sysfs_set_num(sra
, NULL
, "sync_max",
9370 position_to_set
) != 0) {
9371 dprintf("imsm: wait_for_reshape_imsm() "
9372 "cannot set reshape position to %llu\n",
9383 select(fd
+1, &rfds
, NULL
, NULL
, NULL
);
9384 if (sysfs_get_str(sra
, NULL
, "sync_action",
9386 strncmp(action
, "reshape", 7) != 0)
9388 if (sysfs_fd_get_ll(fd
, &completed
) < 0) {
9389 dprintf("imsm: wait_for_reshape_imsm() "
9390 "cannot read reshape_position (in loop)\n");
9394 } while (completed
< to_complete
);
9400 /*******************************************************************************
9401 * Function: check_degradation_change
9402 * Description: Check that array hasn't become failed.
9404 * info : for sysfs access
9405 * sources : source disks descriptors
9406 * degraded: previous degradation level
9409 ******************************************************************************/
9410 int check_degradation_change(struct mdinfo
*info
,
9414 unsigned long long new_degraded
;
9415 sysfs_get_ll(info
, NULL
, "degraded", &new_degraded
);
9416 if (new_degraded
!= (unsigned long long)degraded
) {
9417 /* check each device to ensure it is still working */
9420 for (sd
= info
->devs
; sd
; sd
= sd
->next
) {
9421 if (sd
->disk
.state
& (1<<MD_DISK_FAULTY
))
9423 if (sd
->disk
.state
& (1<<MD_DISK_SYNC
)) {
9425 if (sysfs_get_str(info
,
9426 sd
, "state", sbuf
, 20) < 0 ||
9427 strstr(sbuf
, "faulty") ||
9428 strstr(sbuf
, "in_sync") == NULL
) {
9429 /* this device is dead */
9430 sd
->disk
.state
= (1<<MD_DISK_FAULTY
);
9431 if (sd
->disk
.raid_disk
>= 0 &&
9432 sources
[sd
->disk
.raid_disk
] >= 0) {
9434 sd
->disk
.raid_disk
]);
9435 sources
[sd
->disk
.raid_disk
] =
9444 return new_degraded
;
9447 /*******************************************************************************
9448 * Function: imsm_manage_reshape
9449 * Description: Function finds array under reshape and it manages reshape
9450 * process. It creates stripes backups (if required) and sets
9453 * afd : Backup handle (nattive) - not used
9454 * sra : general array info
9455 * reshape : reshape parameters - not used
9456 * st : supertype structure
9457 * blocks : size of critical section [blocks]
9458 * fds : table of source device descriptor
9459 * offsets : start of array (offest per devices)
9461 * destfd : table of destination device descriptor
9462 * destoffsets : table of destination offsets (per device)
9464 * 1 : success, reshape is done
9466 ******************************************************************************/
9467 static int imsm_manage_reshape(
9468 int afd
, struct mdinfo
*sra
, struct reshape
*reshape
,
9469 struct supertype
*st
, unsigned long backup_blocks
,
9470 int *fds
, unsigned long long *offsets
,
9471 int dests
, int *destfd
, unsigned long long *destoffsets
)
9474 struct intel_super
*super
= st
->sb
;
9475 struct intel_dev
*dv
= NULL
;
9476 struct imsm_dev
*dev
= NULL
;
9477 struct imsm_map
*map_src
;
9478 int migr_vol_qan
= 0;
9479 int ndata
, odata
; /* [bytes] */
9480 int chunk
; /* [bytes] */
9481 struct migr_record
*migr_rec
;
9483 unsigned int buf_size
; /* [bytes] */
9484 unsigned long long max_position
; /* array size [bytes] */
9485 unsigned long long next_step
; /* [blocks]/[bytes] */
9486 unsigned long long old_data_stripe_length
;
9487 unsigned long long start_src
; /* [bytes] */
9488 unsigned long long start
; /* [bytes] */
9489 unsigned long long start_buf_shift
; /* [bytes] */
9491 int source_layout
= 0;
9493 if (!fds
|| !offsets
|| !sra
)
9496 /* Find volume during the reshape */
9497 for (dv
= super
->devlist
; dv
; dv
= dv
->next
) {
9498 if (dv
->dev
->vol
.migr_type
== MIGR_GEN_MIGR
9499 && dv
->dev
->vol
.migr_state
== 1) {
9504 /* Only one volume can migrate at the same time */
9505 if (migr_vol_qan
!= 1) {
9506 fprintf(stderr
, Name
" : %s", migr_vol_qan
?
9507 "Number of migrating volumes greater than 1\n" :
9508 "There is no volume during migrationg\n");
9512 map_src
= get_imsm_map(dev
, MAP_1
);
9513 if (map_src
== NULL
)
9516 ndata
= imsm_num_data_members(dev
, MAP_0
);
9517 odata
= imsm_num_data_members(dev
, MAP_1
);
9519 chunk
= __le16_to_cpu(map_src
->blocks_per_strip
) * 512;
9520 old_data_stripe_length
= odata
* chunk
;
9522 migr_rec
= super
->migr_rec
;
9524 /* initialize migration record for start condition */
9525 if (sra
->reshape_progress
== 0)
9526 init_migr_record_imsm(st
, dev
, sra
);
9528 if (__le32_to_cpu(migr_rec
->rec_status
) != UNIT_SRC_NORMAL
) {
9529 dprintf("imsm: cannot restart migration when data "
9530 "are present in copy area.\n");
9536 buf_size
= __le32_to_cpu(migr_rec
->blocks_per_unit
) * 512;
9537 /* extend buffer size for parity disk */
9538 buf_size
+= __le32_to_cpu(migr_rec
->dest_depth_per_unit
) * 512;
9539 /* add space for stripe aligment */
9540 buf_size
+= old_data_stripe_length
;
9541 if (posix_memalign((void **)&buf
, 4096, buf_size
)) {
9542 dprintf("imsm: Cannot allocate checpoint buffer\n");
9546 max_position
= sra
->component_size
* ndata
;
9547 source_layout
= imsm_level_to_layout(map_src
->raid_level
);
9549 while (__le32_to_cpu(migr_rec
->curr_migr_unit
) <
9550 __le32_to_cpu(migr_rec
->num_migr_units
)) {
9551 /* current reshape position [blocks] */
9552 unsigned long long current_position
=
9553 __le32_to_cpu(migr_rec
->blocks_per_unit
)
9554 * __le32_to_cpu(migr_rec
->curr_migr_unit
);
9555 unsigned long long border
;
9557 /* Check that array hasn't become failed.
9559 degraded
= check_degradation_change(sra
, fds
, degraded
);
9561 dprintf("imsm: Abort reshape due to degradation"
9562 " level (%i)\n", degraded
);
9566 next_step
= __le32_to_cpu(migr_rec
->blocks_per_unit
);
9568 if ((current_position
+ next_step
) > max_position
)
9569 next_step
= max_position
- current_position
;
9571 start
= current_position
* 512;
9573 /* allign reading start to old geometry */
9574 start_buf_shift
= start
% old_data_stripe_length
;
9575 start_src
= start
- start_buf_shift
;
9577 border
= (start_src
/ odata
) - (start
/ ndata
);
9579 if (border
<= __le32_to_cpu(migr_rec
->dest_depth_per_unit
)) {
9580 /* save critical stripes to buf
9581 * start - start address of current unit
9583 * start_src - start address of current unit
9584 * to backup alligned to source array
9587 unsigned long long next_step_filler
= 0;
9588 unsigned long long copy_length
= next_step
* 512;
9590 /* allign copy area length to stripe in old geometry */
9591 next_step_filler
= ((copy_length
+ start_buf_shift
)
9592 % old_data_stripe_length
);
9593 if (next_step_filler
)
9594 next_step_filler
= (old_data_stripe_length
9595 - next_step_filler
);
9596 dprintf("save_stripes() parameters: start = %llu,"
9597 "\tstart_src = %llu,\tnext_step*512 = %llu,"
9598 "\tstart_in_buf_shift = %llu,"
9599 "\tnext_step_filler = %llu\n",
9600 start
, start_src
, copy_length
,
9601 start_buf_shift
, next_step_filler
);
9603 if (save_stripes(fds
, offsets
, map_src
->num_members
,
9604 chunk
, map_src
->raid_level
,
9605 source_layout
, 0, NULL
, start_src
,
9607 next_step_filler
+ start_buf_shift
,
9609 dprintf("imsm: Cannot save stripes"
9613 /* Convert data to destination format and store it
9614 * in backup general migration area
9616 if (save_backup_imsm(st
, dev
, sra
,
9617 buf
+ start_buf_shift
, copy_length
)) {
9618 dprintf("imsm: Cannot save stripes to "
9619 "target devices\n");
9622 if (save_checkpoint_imsm(st
, sra
,
9623 UNIT_SRC_IN_CP_AREA
)) {
9624 dprintf("imsm: Cannot write checkpoint to "
9625 "migration record (UNIT_SRC_IN_CP_AREA)\n");
9629 /* set next step to use whole border area */
9630 border
/= next_step
;
9632 next_step
*= border
;
9634 /* When data backed up, checkpoint stored,
9635 * kick the kernel to reshape unit of data
9637 next_step
= next_step
+ sra
->reshape_progress
;
9638 /* limit next step to array max position */
9639 if (next_step
> max_position
)
9640 next_step
= max_position
;
9641 sysfs_set_num(sra
, NULL
, "suspend_lo", sra
->reshape_progress
);
9642 sysfs_set_num(sra
, NULL
, "suspend_hi", next_step
);
9643 sra
->reshape_progress
= next_step
;
9645 /* wait until reshape finish */
9646 if (wait_for_reshape_imsm(sra
, ndata
) < 0) {
9647 dprintf("wait_for_reshape_imsm returned error!\n");
9651 if (save_checkpoint_imsm(st
, sra
, UNIT_SRC_NORMAL
) == 1) {
9652 /* ignore error == 2, this can mean end of reshape here
9654 dprintf("imsm: Cannot write checkpoint to "
9655 "migration record (UNIT_SRC_NORMAL)\n");
9661 /* return '1' if done */
9669 #endif /* MDASSEMBLE */
9671 struct superswitch super_imsm
= {
9673 .examine_super
= examine_super_imsm
,
9674 .brief_examine_super
= brief_examine_super_imsm
,
9675 .brief_examine_subarrays
= brief_examine_subarrays_imsm
,
9676 .export_examine_super
= export_examine_super_imsm
,
9677 .detail_super
= detail_super_imsm
,
9678 .brief_detail_super
= brief_detail_super_imsm
,
9679 .write_init_super
= write_init_super_imsm
,
9680 .validate_geometry
= validate_geometry_imsm
,
9681 .add_to_super
= add_to_super_imsm
,
9682 .remove_from_super
= remove_from_super_imsm
,
9683 .detail_platform
= detail_platform_imsm
,
9684 .kill_subarray
= kill_subarray_imsm
,
9685 .update_subarray
= update_subarray_imsm
,
9686 .load_container
= load_container_imsm
,
9687 .default_geometry
= default_geometry_imsm
,
9688 .get_disk_controller_domain
= imsm_get_disk_controller_domain
,
9689 .reshape_super
= imsm_reshape_super
,
9690 .manage_reshape
= imsm_manage_reshape
,
9691 .recover_backup
= recover_backup_imsm
,
9693 .match_home
= match_home_imsm
,
9694 .uuid_from_super
= uuid_from_super_imsm
,
9695 .getinfo_super
= getinfo_super_imsm
,
9696 .getinfo_super_disks
= getinfo_super_disks_imsm
,
9697 .update_super
= update_super_imsm
,
9699 .avail_size
= avail_size_imsm
,
9700 .min_acceptable_spare_size
= min_acceptable_spare_size_imsm
,
9702 .compare_super
= compare_super_imsm
,
9704 .load_super
= load_super_imsm
,
9705 .init_super
= init_super_imsm
,
9706 .store_super
= store_super_imsm
,
9707 .free_super
= free_super_imsm
,
9708 .match_metadata_desc
= match_metadata_desc_imsm
,
9709 .container_content
= container_content_imsm
,
9717 .open_new
= imsm_open_new
,
9718 .set_array_state
= imsm_set_array_state
,
9719 .set_disk
= imsm_set_disk
,
9720 .sync_metadata
= imsm_sync_metadata
,
9721 .activate_spare
= imsm_activate_spare
,
9722 .process_update
= imsm_process_update
,
9723 .prepare_update
= imsm_prepare_update
,
9724 #endif /* MDASSEMBLE */