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 ******************************************************************************/
2124 static int load_imsm_migr_rec(struct intel_super
*super
, struct mdinfo
*info
)
2127 struct dl
*dl
= NULL
;
2131 struct imsm_dev
*dev
;
2132 struct imsm_map
*map
= NULL
;
2135 /* find map under migration */
2136 dev
= imsm_get_device_during_migration(super
);
2137 /* nothing to load,no migration in progress?
2141 map
= get_imsm_map(dev
, MAP_0
);
2144 for (sd
= info
->devs
; sd
; sd
= sd
->next
) {
2145 /* skip spare and failed disks
2147 if (sd
->disk
.raid_disk
< 0)
2149 /* read only from one of the first two slots */
2151 slot
= get_imsm_disk_slot(map
,
2152 sd
->disk
.raid_disk
);
2153 if ((map
== NULL
) || (slot
> 1) || (slot
< 0))
2156 sprintf(nm
, "%d:%d", sd
->disk
.major
, sd
->disk
.minor
);
2157 fd
= dev_open(nm
, O_RDONLY
);
2163 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
2164 /* skip spare and failed disks
2168 /* read only from one of the first two slots */
2170 slot
= get_imsm_disk_slot(map
, dl
->index
);
2171 if ((map
== NULL
) || (slot
> 1) || (slot
< 0))
2173 sprintf(nm
, "%d:%d", dl
->major
, dl
->minor
);
2174 fd
= dev_open(nm
, O_RDONLY
);
2181 retval
= read_imsm_migr_rec(fd
, super
);
2190 /*******************************************************************************
2191 * function: imsm_create_metadata_checkpoint_update
2192 * Description: It creates update for checkpoint change.
2194 * super : imsm internal array info
2195 * u : pointer to prepared update
2198 * If length is equal to 0, input pointer u contains no update
2199 ******************************************************************************/
2200 static int imsm_create_metadata_checkpoint_update(
2201 struct intel_super
*super
,
2202 struct imsm_update_general_migration_checkpoint
**u
)
2205 int update_memory_size
= 0;
2207 dprintf("imsm_create_metadata_checkpoint_update(enter)\n");
2213 /* size of all update data without anchor */
2214 update_memory_size
=
2215 sizeof(struct imsm_update_general_migration_checkpoint
);
2217 *u
= calloc(1, update_memory_size
);
2219 dprintf("error: cannot get memory for "
2220 "imsm_create_metadata_checkpoint_update update\n");
2223 (*u
)->type
= update_general_migration_checkpoint
;
2224 (*u
)->curr_migr_unit
= __le32_to_cpu(super
->migr_rec
->curr_migr_unit
);
2225 dprintf("imsm_create_metadata_checkpoint_update: prepared for %u\n",
2226 (*u
)->curr_migr_unit
);
2228 return update_memory_size
;
2232 static void imsm_update_metadata_locally(struct supertype
*st
,
2233 void *buf
, int len
);
2235 /*******************************************************************************
2236 * Function: write_imsm_migr_rec
2237 * Description: Function writes imsm migration record
2238 * (at the last sector of disk)
2240 * super : imsm internal array info
2244 ******************************************************************************/
2245 static int write_imsm_migr_rec(struct supertype
*st
)
2247 struct intel_super
*super
= st
->sb
;
2248 unsigned long long dsize
;
2254 struct imsm_update_general_migration_checkpoint
*u
;
2255 struct imsm_dev
*dev
;
2256 struct imsm_map
*map
= NULL
;
2258 /* find map under migration */
2259 dev
= imsm_get_device_during_migration(super
);
2260 /* if no migration, write buffer anyway to clear migr_record
2261 * on disk based on first available device
2264 dev
= get_imsm_dev(super
, super
->current_vol
< 0 ? 0 :
2265 super
->current_vol
);
2267 map
= get_imsm_map(dev
, MAP_0
);
2269 for (sd
= super
->disks
; sd
; sd
= sd
->next
) {
2272 /* skip failed and spare devices */
2275 /* write to 2 first slots only */
2277 slot
= get_imsm_disk_slot(map
, sd
->index
);
2278 if ((map
== NULL
) || (slot
> 1) || (slot
< 0))
2281 sprintf(nm
, "%d:%d", sd
->major
, sd
->minor
);
2282 fd
= dev_open(nm
, O_RDWR
);
2285 get_dev_size(fd
, NULL
, &dsize
);
2286 if (lseek64(fd
, dsize
- MIGR_REC_POSITION
, SEEK_SET
) < 0) {
2288 Name
": Cannot seek to anchor block: %s\n",
2292 if (write(fd
, super
->migr_rec_buf
, MIGR_REC_BUF_SIZE
) !=
2293 MIGR_REC_BUF_SIZE
) {
2295 Name
": Cannot write migr record block: %s\n",
2302 /* update checkpoint information in metadata */
2303 len
= imsm_create_metadata_checkpoint_update(super
, &u
);
2306 dprintf("imsm: Cannot prepare update\n");
2309 /* update metadata locally */
2310 imsm_update_metadata_locally(st
, u
, len
);
2311 /* and possibly remotely */
2312 if (st
->update_tail
) {
2313 append_metadata_update(st
, u
, len
);
2314 /* during reshape we do all work inside metadata handler
2315 * manage_reshape(), so metadata update has to be triggered
2318 flush_metadata_updates(st
);
2319 st
->update_tail
= &st
->updates
;
2329 #endif /* MDASSEMBLE */
2331 /* spare/missing disks activations are not allowe when
2332 * array/container performs reshape operation, because
2333 * all arrays in container works on the same disks set
2335 int imsm_reshape_blocks_arrays_changes(struct intel_super
*super
)
2338 struct intel_dev
*i_dev
;
2339 struct imsm_dev
*dev
;
2341 /* check whole container
2343 for (i_dev
= super
->devlist
; i_dev
; i_dev
= i_dev
->next
) {
2345 if (is_gen_migration(dev
)) {
2346 /* No repair during any migration in container
2355 static void getinfo_super_imsm_volume(struct supertype
*st
, struct mdinfo
*info
, char *dmap
)
2357 struct intel_super
*super
= st
->sb
;
2358 struct migr_record
*migr_rec
= super
->migr_rec
;
2359 struct imsm_dev
*dev
= get_imsm_dev(super
, super
->current_vol
);
2360 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
2361 struct imsm_map
*prev_map
= get_imsm_map(dev
, MAP_1
);
2362 struct imsm_map
*map_to_analyse
= map
;
2365 unsigned int component_size_alligment
;
2366 int map_disks
= info
->array
.raid_disks
;
2368 memset(info
, 0, sizeof(*info
));
2370 map_to_analyse
= prev_map
;
2372 dl
= super
->current_disk
;
2374 info
->container_member
= super
->current_vol
;
2375 info
->array
.raid_disks
= map
->num_members
;
2376 info
->array
.level
= get_imsm_raid_level(map_to_analyse
);
2377 info
->array
.layout
= imsm_level_to_layout(info
->array
.level
);
2378 info
->array
.md_minor
= -1;
2379 info
->array
.ctime
= 0;
2380 info
->array
.utime
= 0;
2381 info
->array
.chunk_size
=
2382 __le16_to_cpu(map_to_analyse
->blocks_per_strip
) << 9;
2383 info
->array
.state
= !dev
->vol
.dirty
;
2384 info
->custom_array_size
= __le32_to_cpu(dev
->size_high
);
2385 info
->custom_array_size
<<= 32;
2386 info
->custom_array_size
|= __le32_to_cpu(dev
->size_low
);
2387 info
->recovery_blocked
= imsm_reshape_blocks_arrays_changes(st
->sb
);
2389 if (is_gen_migration(dev
)) {
2390 info
->reshape_active
= 1;
2391 info
->new_level
= get_imsm_raid_level(map
);
2392 info
->new_layout
= imsm_level_to_layout(info
->new_level
);
2393 info
->new_chunk
= __le16_to_cpu(map
->blocks_per_strip
) << 9;
2394 info
->delta_disks
= map
->num_members
- prev_map
->num_members
;
2395 if (info
->delta_disks
) {
2396 /* this needs to be applied to every array
2399 info
->reshape_active
= CONTAINER_RESHAPE
;
2401 /* We shape information that we give to md might have to be
2402 * modify to cope with md's requirement for reshaping arrays.
2403 * For example, when reshaping a RAID0, md requires it to be
2404 * presented as a degraded RAID4.
2405 * Also if a RAID0 is migrating to a RAID5 we need to specify
2406 * the array as already being RAID5, but the 'before' layout
2407 * is a RAID4-like layout.
2409 switch (info
->array
.level
) {
2411 switch(info
->new_level
) {
2413 /* conversion is happening as RAID4 */
2414 info
->array
.level
= 4;
2415 info
->array
.raid_disks
+= 1;
2418 /* conversion is happening as RAID5 */
2419 info
->array
.level
= 5;
2420 info
->array
.layout
= ALGORITHM_PARITY_N
;
2421 info
->delta_disks
-= 1;
2424 /* FIXME error message */
2425 info
->array
.level
= UnSet
;
2431 info
->new_level
= UnSet
;
2432 info
->new_layout
= UnSet
;
2433 info
->new_chunk
= info
->array
.chunk_size
;
2434 info
->delta_disks
= 0;
2438 info
->disk
.major
= dl
->major
;
2439 info
->disk
.minor
= dl
->minor
;
2440 info
->disk
.number
= dl
->index
;
2441 info
->disk
.raid_disk
= get_imsm_disk_slot(map_to_analyse
,
2445 info
->data_offset
= __le32_to_cpu(map_to_analyse
->pba_of_lba0
);
2446 info
->component_size
=
2447 __le32_to_cpu(map_to_analyse
->blocks_per_member
);
2449 /* check component size aligment
2451 component_size_alligment
=
2452 info
->component_size
% (info
->array
.chunk_size
/512);
2454 if (component_size_alligment
&&
2455 (info
->array
.level
!= 1) && (info
->array
.level
!= UnSet
)) {
2456 dprintf("imsm: reported component size alligned from %llu ",
2457 info
->component_size
);
2458 info
->component_size
-= component_size_alligment
;
2459 dprintf("to %llu (%i).\n",
2460 info
->component_size
, component_size_alligment
);
2463 memset(info
->uuid
, 0, sizeof(info
->uuid
));
2464 info
->recovery_start
= MaxSector
;
2466 info
->reshape_progress
= 0;
2467 info
->resync_start
= MaxSector
;
2468 if ((map_to_analyse
->map_state
== IMSM_T_STATE_UNINITIALIZED
||
2470 imsm_reshape_blocks_arrays_changes(super
) == 0) {
2471 info
->resync_start
= 0;
2473 if (dev
->vol
.migr_state
) {
2474 switch (migr_type(dev
)) {
2477 __u64 blocks_per_unit
= blocks_per_migr_unit(super
,
2479 __u64 units
= __le32_to_cpu(dev
->vol
.curr_migr_unit
);
2481 info
->resync_start
= blocks_per_unit
* units
;
2484 case MIGR_GEN_MIGR
: {
2485 __u64 blocks_per_unit
= blocks_per_migr_unit(super
,
2487 __u64 units
= __le32_to_cpu(migr_rec
->curr_migr_unit
);
2488 unsigned long long array_blocks
;
2491 if (__le32_to_cpu(migr_rec
->ascending_migr
) &&
2493 (__le32_to_cpu(migr_rec
->num_migr_units
)-1)) &&
2494 (super
->migr_rec
->rec_status
==
2495 __cpu_to_le32(UNIT_SRC_IN_CP_AREA
)))
2498 info
->reshape_progress
= blocks_per_unit
* units
;
2500 dprintf("IMSM: General Migration checkpoint : %llu "
2501 "(%llu) -> read reshape progress : %llu\n",
2502 (unsigned long long)units
,
2503 (unsigned long long)blocks_per_unit
,
2504 info
->reshape_progress
);
2506 used_disks
= imsm_num_data_members(dev
, MAP_1
);
2507 if (used_disks
> 0) {
2508 array_blocks
= map
->blocks_per_member
*
2510 /* round array size down to closest MB
2512 info
->custom_array_size
= (array_blocks
2513 >> SECT_PER_MB_SHIFT
)
2514 << SECT_PER_MB_SHIFT
;
2518 /* we could emulate the checkpointing of
2519 * 'sync_action=check' migrations, but for now
2520 * we just immediately complete them
2523 /* this is handled by container_content_imsm() */
2524 case MIGR_STATE_CHANGE
:
2525 /* FIXME handle other migrations */
2527 /* we are not dirty, so... */
2528 info
->resync_start
= MaxSector
;
2532 strncpy(info
->name
, (char *) dev
->volume
, MAX_RAID_SERIAL_LEN
);
2533 info
->name
[MAX_RAID_SERIAL_LEN
] = 0;
2535 info
->array
.major_version
= -1;
2536 info
->array
.minor_version
= -2;
2537 devname
= devnum2devname(st
->container_dev
);
2538 *info
->text_version
= '\0';
2540 sprintf(info
->text_version
, "/%s/%d", devname
, info
->container_member
);
2542 info
->safe_mode_delay
= 4000; /* 4 secs like the Matrix driver */
2543 uuid_from_super_imsm(st
, info
->uuid
);
2547 for (i
=0; i
<map_disks
; i
++) {
2549 if (i
< info
->array
.raid_disks
) {
2550 struct imsm_disk
*dsk
;
2551 j
= get_imsm_disk_idx(dev
, i
, MAP_X
);
2552 dsk
= get_imsm_disk(super
, j
);
2553 if (dsk
&& (dsk
->status
& CONFIGURED_DISK
))
2560 static __u8
imsm_check_degraded(struct intel_super
*super
, struct imsm_dev
*dev
,
2561 int failed
, int look_in_map
);
2563 static int imsm_count_failed(struct intel_super
*super
, struct imsm_dev
*dev
,
2566 static void manage_second_map(struct intel_super
*super
, struct imsm_dev
*dev
)
2568 if (is_gen_migration(dev
)) {
2571 struct imsm_map
*map2
= get_imsm_map(dev
, MAP_1
);
2573 failed
= imsm_count_failed(super
, dev
, MAP_1
);
2574 map_state
= imsm_check_degraded(super
, dev
, failed
, MAP_1
);
2575 if (map2
->map_state
!= map_state
) {
2576 map2
->map_state
= map_state
;
2577 super
->updates_pending
++;
2582 static struct imsm_disk
*get_imsm_missing(struct intel_super
*super
, __u8 index
)
2586 for (d
= super
->missing
; d
; d
= d
->next
)
2587 if (d
->index
== index
)
2592 static void getinfo_super_imsm(struct supertype
*st
, struct mdinfo
*info
, char *map
)
2594 struct intel_super
*super
= st
->sb
;
2595 struct imsm_disk
*disk
;
2596 int map_disks
= info
->array
.raid_disks
;
2597 int max_enough
= -1;
2599 struct imsm_super
*mpb
;
2601 if (super
->current_vol
>= 0) {
2602 getinfo_super_imsm_volume(st
, info
, map
);
2605 memset(info
, 0, sizeof(*info
));
2607 /* Set raid_disks to zero so that Assemble will always pull in valid
2610 info
->array
.raid_disks
= 0;
2611 info
->array
.level
= LEVEL_CONTAINER
;
2612 info
->array
.layout
= 0;
2613 info
->array
.md_minor
= -1;
2614 info
->array
.ctime
= 0; /* N/A for imsm */
2615 info
->array
.utime
= 0;
2616 info
->array
.chunk_size
= 0;
2618 info
->disk
.major
= 0;
2619 info
->disk
.minor
= 0;
2620 info
->disk
.raid_disk
= -1;
2621 info
->reshape_active
= 0;
2622 info
->array
.major_version
= -1;
2623 info
->array
.minor_version
= -2;
2624 strcpy(info
->text_version
, "imsm");
2625 info
->safe_mode_delay
= 0;
2626 info
->disk
.number
= -1;
2627 info
->disk
.state
= 0;
2629 info
->recovery_start
= MaxSector
;
2630 info
->recovery_blocked
= imsm_reshape_blocks_arrays_changes(st
->sb
);
2632 /* do we have the all the insync disks that we expect? */
2633 mpb
= super
->anchor
;
2635 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
2636 struct imsm_dev
*dev
= get_imsm_dev(super
, i
);
2637 int failed
, enough
, j
, missing
= 0;
2638 struct imsm_map
*map
;
2641 failed
= imsm_count_failed(super
, dev
, MAP_0
);
2642 state
= imsm_check_degraded(super
, dev
, failed
, MAP_0
);
2643 map
= get_imsm_map(dev
, MAP_0
);
2645 /* any newly missing disks?
2646 * (catches single-degraded vs double-degraded)
2648 for (j
= 0; j
< map
->num_members
; j
++) {
2649 __u32 ord
= get_imsm_ord_tbl_ent(dev
, j
, MAP_0
);
2650 __u32 idx
= ord_to_idx(ord
);
2652 if (!(ord
& IMSM_ORD_REBUILD
) &&
2653 get_imsm_missing(super
, idx
)) {
2659 if (state
== IMSM_T_STATE_FAILED
)
2661 else if (state
== IMSM_T_STATE_DEGRADED
&&
2662 (state
!= map
->map_state
|| missing
))
2664 else /* we're normal, or already degraded */
2667 /* in the missing/failed disk case check to see
2668 * if at least one array is runnable
2670 max_enough
= max(max_enough
, enough
);
2672 dprintf("%s: enough: %d\n", __func__
, max_enough
);
2673 info
->container_enough
= max_enough
;
2676 __u32 reserved
= imsm_reserved_sectors(super
, super
->disks
);
2678 disk
= &super
->disks
->disk
;
2679 info
->data_offset
= __le32_to_cpu(disk
->total_blocks
) - reserved
;
2680 info
->component_size
= reserved
;
2681 info
->disk
.state
= is_configured(disk
) ? (1 << MD_DISK_ACTIVE
) : 0;
2682 /* we don't change info->disk.raid_disk here because
2683 * this state will be finalized in mdmon after we have
2684 * found the 'most fresh' version of the metadata
2686 info
->disk
.state
|= is_failed(disk
) ? (1 << MD_DISK_FAULTY
) : 0;
2687 info
->disk
.state
|= is_spare(disk
) ? 0 : (1 << MD_DISK_SYNC
);
2690 /* only call uuid_from_super_imsm when this disk is part of a populated container,
2691 * ->compare_super may have updated the 'num_raid_devs' field for spares
2693 if (info
->disk
.state
& (1 << MD_DISK_SYNC
) || super
->anchor
->num_raid_devs
)
2694 uuid_from_super_imsm(st
, info
->uuid
);
2696 memcpy(info
->uuid
, uuid_zero
, sizeof(uuid_zero
));
2698 /* I don't know how to compute 'map' on imsm, so use safe default */
2701 for (i
= 0; i
< map_disks
; i
++)
2707 /* allocates memory and fills disk in mdinfo structure
2708 * for each disk in array */
2709 struct mdinfo
*getinfo_super_disks_imsm(struct supertype
*st
)
2711 struct mdinfo
*mddev
= NULL
;
2712 struct intel_super
*super
= st
->sb
;
2713 struct imsm_disk
*disk
;
2716 if (!super
|| !super
->disks
)
2719 mddev
= malloc(sizeof(*mddev
));
2721 fprintf(stderr
, Name
": Failed to allocate memory.\n");
2724 memset(mddev
, 0, sizeof(*mddev
));
2728 tmp
= malloc(sizeof(*tmp
));
2730 fprintf(stderr
, Name
": Failed to allocate memory.\n");
2735 memset(tmp
, 0, sizeof(*tmp
));
2737 tmp
->next
= mddev
->devs
;
2739 tmp
->disk
.number
= count
++;
2740 tmp
->disk
.major
= dl
->major
;
2741 tmp
->disk
.minor
= dl
->minor
;
2742 tmp
->disk
.state
= is_configured(disk
) ?
2743 (1 << MD_DISK_ACTIVE
) : 0;
2744 tmp
->disk
.state
|= is_failed(disk
) ? (1 << MD_DISK_FAULTY
) : 0;
2745 tmp
->disk
.state
|= is_spare(disk
) ? 0 : (1 << MD_DISK_SYNC
);
2746 tmp
->disk
.raid_disk
= -1;
2752 static int update_super_imsm(struct supertype
*st
, struct mdinfo
*info
,
2753 char *update
, char *devname
, int verbose
,
2754 int uuid_set
, char *homehost
)
2756 /* For 'assemble' and 'force' we need to return non-zero if any
2757 * change was made. For others, the return value is ignored.
2758 * Update options are:
2759 * force-one : This device looks a bit old but needs to be included,
2760 * update age info appropriately.
2761 * assemble: clear any 'faulty' flag to allow this device to
2763 * force-array: Array is degraded but being forced, mark it clean
2764 * if that will be needed to assemble it.
2766 * newdev: not used ????
2767 * grow: Array has gained a new device - this is currently for
2769 * resync: mark as dirty so a resync will happen.
2770 * name: update the name - preserving the homehost
2771 * uuid: Change the uuid of the array to match watch is given
2773 * Following are not relevant for this imsm:
2774 * sparc2.2 : update from old dodgey metadata
2775 * super-minor: change the preferred_minor number
2776 * summaries: update redundant counters.
2777 * homehost: update the recorded homehost
2778 * _reshape_progress: record new reshape_progress position.
2781 struct intel_super
*super
= st
->sb
;
2782 struct imsm_super
*mpb
;
2784 /* we can only update container info */
2785 if (!super
|| super
->current_vol
>= 0 || !super
->anchor
)
2788 mpb
= super
->anchor
;
2790 if (strcmp(update
, "uuid") == 0 && uuid_set
&& !info
->update_private
)
2792 else if (strcmp(update
, "uuid") == 0 && uuid_set
&& info
->update_private
) {
2793 mpb
->orig_family_num
= *((__u32
*) info
->update_private
);
2795 } else if (strcmp(update
, "uuid") == 0) {
2796 __u32
*new_family
= malloc(sizeof(*new_family
));
2798 /* update orig_family_number with the incoming random
2799 * data, report the new effective uuid, and store the
2800 * new orig_family_num for future updates.
2803 memcpy(&mpb
->orig_family_num
, info
->uuid
, sizeof(__u32
));
2804 uuid_from_super_imsm(st
, info
->uuid
);
2805 *new_family
= mpb
->orig_family_num
;
2806 info
->update_private
= new_family
;
2809 } else if (strcmp(update
, "assemble") == 0)
2814 /* successful update? recompute checksum */
2816 mpb
->check_sum
= __le32_to_cpu(__gen_imsm_checksum(mpb
));
2821 static size_t disks_to_mpb_size(int disks
)
2825 size
= sizeof(struct imsm_super
);
2826 size
+= (disks
- 1) * sizeof(struct imsm_disk
);
2827 size
+= 2 * sizeof(struct imsm_dev
);
2828 /* up to 2 maps per raid device (-2 for imsm_maps in imsm_dev */
2829 size
+= (4 - 2) * sizeof(struct imsm_map
);
2830 /* 4 possible disk_ord_tbl's */
2831 size
+= 4 * (disks
- 1) * sizeof(__u32
);
2836 static __u64
avail_size_imsm(struct supertype
*st
, __u64 devsize
)
2838 if (devsize
< (MPB_SECTOR_CNT
+ IMSM_RESERVED_SECTORS
))
2841 return devsize
- (MPB_SECTOR_CNT
+ IMSM_RESERVED_SECTORS
);
2844 static void free_devlist(struct intel_super
*super
)
2846 struct intel_dev
*dv
;
2848 while (super
->devlist
) {
2849 dv
= super
->devlist
->next
;
2850 free(super
->devlist
->dev
);
2851 free(super
->devlist
);
2852 super
->devlist
= dv
;
2856 static void imsm_copy_dev(struct imsm_dev
*dest
, struct imsm_dev
*src
)
2858 memcpy(dest
, src
, sizeof_imsm_dev(src
, 0));
2861 static int compare_super_imsm(struct supertype
*st
, struct supertype
*tst
)
2865 * 0 same, or first was empty, and second was copied
2866 * 1 second had wrong number
2868 * 3 wrong other info
2870 struct intel_super
*first
= st
->sb
;
2871 struct intel_super
*sec
= tst
->sb
;
2878 /* in platform dependent environment test if the disks
2879 * use the same Intel hba
2881 if (!check_env("IMSM_NO_PLATFORM")) {
2882 if (!first
->hba
|| !sec
->hba
||
2883 (first
->hba
->type
!= sec
->hba
->type
)) {
2885 "HBAs of devices does not match %s != %s\n",
2886 first
->hba
? get_sys_dev_type(first
->hba
->type
) : NULL
,
2887 sec
->hba
? get_sys_dev_type(sec
->hba
->type
) : NULL
);
2892 /* if an anchor does not have num_raid_devs set then it is a free
2895 if (first
->anchor
->num_raid_devs
> 0 &&
2896 sec
->anchor
->num_raid_devs
> 0) {
2897 /* Determine if these disks might ever have been
2898 * related. Further disambiguation can only take place
2899 * in load_super_imsm_all
2901 __u32 first_family
= first
->anchor
->orig_family_num
;
2902 __u32 sec_family
= sec
->anchor
->orig_family_num
;
2904 if (memcmp(first
->anchor
->sig
, sec
->anchor
->sig
,
2905 MAX_SIGNATURE_LENGTH
) != 0)
2908 if (first_family
== 0)
2909 first_family
= first
->anchor
->family_num
;
2910 if (sec_family
== 0)
2911 sec_family
= sec
->anchor
->family_num
;
2913 if (first_family
!= sec_family
)
2919 /* if 'first' is a spare promote it to a populated mpb with sec's
2922 if (first
->anchor
->num_raid_devs
== 0 &&
2923 sec
->anchor
->num_raid_devs
> 0) {
2925 struct intel_dev
*dv
;
2926 struct imsm_dev
*dev
;
2928 /* we need to copy raid device info from sec if an allocation
2929 * fails here we don't associate the spare
2931 for (i
= 0; i
< sec
->anchor
->num_raid_devs
; i
++) {
2932 dv
= malloc(sizeof(*dv
));
2935 dev
= malloc(sizeof_imsm_dev(get_imsm_dev(sec
, i
), 1));
2942 dv
->next
= first
->devlist
;
2943 first
->devlist
= dv
;
2945 if (i
< sec
->anchor
->num_raid_devs
) {
2946 /* allocation failure */
2947 free_devlist(first
);
2948 fprintf(stderr
, "imsm: failed to associate spare\n");
2951 first
->anchor
->num_raid_devs
= sec
->anchor
->num_raid_devs
;
2952 first
->anchor
->orig_family_num
= sec
->anchor
->orig_family_num
;
2953 first
->anchor
->family_num
= sec
->anchor
->family_num
;
2954 memcpy(first
->anchor
->sig
, sec
->anchor
->sig
, MAX_SIGNATURE_LENGTH
);
2955 for (i
= 0; i
< sec
->anchor
->num_raid_devs
; i
++)
2956 imsm_copy_dev(get_imsm_dev(first
, i
), get_imsm_dev(sec
, i
));
2962 static void fd2devname(int fd
, char *name
)
2966 char dname
[PATH_MAX
];
2971 if (fstat(fd
, &st
) != 0)
2973 sprintf(path
, "/sys/dev/block/%d:%d",
2974 major(st
.st_rdev
), minor(st
.st_rdev
));
2976 rv
= readlink(path
, dname
, sizeof(dname
)-1);
2981 nm
= strrchr(dname
, '/');
2984 snprintf(name
, MAX_RAID_SERIAL_LEN
, "/dev/%s", nm
);
2988 extern int scsi_get_serial(int fd
, void *buf
, size_t buf_len
);
2990 static int imsm_read_serial(int fd
, char *devname
,
2991 __u8 serial
[MAX_RAID_SERIAL_LEN
])
2993 unsigned char scsi_serial
[255];
3002 memset(scsi_serial
, 0, sizeof(scsi_serial
));
3004 rv
= scsi_get_serial(fd
, scsi_serial
, sizeof(scsi_serial
));
3006 if (rv
&& check_env("IMSM_DEVNAME_AS_SERIAL")) {
3007 memset(serial
, 0, MAX_RAID_SERIAL_LEN
);
3008 fd2devname(fd
, (char *) serial
);
3015 Name
": Failed to retrieve serial for %s\n",
3020 rsp_len
= scsi_serial
[3];
3024 Name
": Failed to retrieve serial for %s\n",
3028 rsp_buf
= (char *) &scsi_serial
[4];
3030 /* trim all whitespace and non-printable characters and convert
3033 for (i
= 0, dest
= rsp_buf
; i
< rsp_len
; i
++) {
3036 /* ':' is reserved for use in placeholder serial
3037 * numbers for missing disks
3045 len
= dest
- rsp_buf
;
3048 /* truncate leading characters */
3049 if (len
> MAX_RAID_SERIAL_LEN
) {
3050 dest
+= len
- MAX_RAID_SERIAL_LEN
;
3051 len
= MAX_RAID_SERIAL_LEN
;
3054 memset(serial
, 0, MAX_RAID_SERIAL_LEN
);
3055 memcpy(serial
, dest
, len
);
3060 static int serialcmp(__u8
*s1
, __u8
*s2
)
3062 return strncmp((char *) s1
, (char *) s2
, MAX_RAID_SERIAL_LEN
);
3065 static void serialcpy(__u8
*dest
, __u8
*src
)
3067 strncpy((char *) dest
, (char *) src
, MAX_RAID_SERIAL_LEN
);
3070 static struct dl
*serial_to_dl(__u8
*serial
, struct intel_super
*super
)
3074 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
3075 if (serialcmp(dl
->serial
, serial
) == 0)
3081 static struct imsm_disk
*
3082 __serial_to_disk(__u8
*serial
, struct imsm_super
*mpb
, int *idx
)
3086 for (i
= 0; i
< mpb
->num_disks
; i
++) {
3087 struct imsm_disk
*disk
= __get_imsm_disk(mpb
, i
);
3089 if (serialcmp(disk
->serial
, serial
) == 0) {
3100 load_imsm_disk(int fd
, struct intel_super
*super
, char *devname
, int keep_fd
)
3102 struct imsm_disk
*disk
;
3107 __u8 serial
[MAX_RAID_SERIAL_LEN
];
3109 rv
= imsm_read_serial(fd
, devname
, serial
);
3114 dl
= calloc(1, sizeof(*dl
));
3118 Name
": failed to allocate disk buffer for %s\n",
3124 dl
->major
= major(stb
.st_rdev
);
3125 dl
->minor
= minor(stb
.st_rdev
);
3126 dl
->next
= super
->disks
;
3127 dl
->fd
= keep_fd
? fd
: -1;
3128 assert(super
->disks
== NULL
);
3130 serialcpy(dl
->serial
, serial
);
3133 fd2devname(fd
, name
);
3135 dl
->devname
= strdup(devname
);
3137 dl
->devname
= strdup(name
);
3139 /* look up this disk's index in the current anchor */
3140 disk
= __serial_to_disk(dl
->serial
, super
->anchor
, &dl
->index
);
3143 /* only set index on disks that are a member of a
3144 * populated contianer, i.e. one with raid_devs
3146 if (is_failed(&dl
->disk
))
3148 else if (is_spare(&dl
->disk
))
3156 /* When migrating map0 contains the 'destination' state while map1
3157 * contains the current state. When not migrating map0 contains the
3158 * current state. This routine assumes that map[0].map_state is set to
3159 * the current array state before being called.
3161 * Migration is indicated by one of the following states
3162 * 1/ Idle (migr_state=0 map0state=normal||unitialized||degraded||failed)
3163 * 2/ Initialize (migr_state=1 migr_type=MIGR_INIT map0state=normal
3164 * map1state=unitialized)
3165 * 3/ Repair (Resync) (migr_state=1 migr_type=MIGR_REPAIR map0state=normal
3167 * 4/ Rebuild (migr_state=1 migr_type=MIGR_REBUILD map0state=normal
3168 * map1state=degraded)
3169 * 5/ Migration (mig_state=1 migr_type=MIGR_GEN_MIGR map0state=normal
3172 static void migrate(struct imsm_dev
*dev
, struct intel_super
*super
,
3173 __u8 to_state
, int migr_type
)
3175 struct imsm_map
*dest
;
3176 struct imsm_map
*src
= get_imsm_map(dev
, MAP_0
);
3178 dev
->vol
.migr_state
= 1;
3179 set_migr_type(dev
, migr_type
);
3180 dev
->vol
.curr_migr_unit
= 0;
3181 dest
= get_imsm_map(dev
, MAP_1
);
3183 /* duplicate and then set the target end state in map[0] */
3184 memcpy(dest
, src
, sizeof_imsm_map(src
));
3185 if ((migr_type
== MIGR_REBUILD
) ||
3186 (migr_type
== MIGR_GEN_MIGR
)) {
3190 for (i
= 0; i
< src
->num_members
; i
++) {
3191 ord
= __le32_to_cpu(src
->disk_ord_tbl
[i
]);
3192 set_imsm_ord_tbl_ent(src
, i
, ord_to_idx(ord
));
3196 if (migr_type
== MIGR_GEN_MIGR
)
3197 /* Clear migration record */
3198 memset(super
->migr_rec
, 0, sizeof(struct migr_record
));
3200 src
->map_state
= to_state
;
3203 static void end_migration(struct imsm_dev
*dev
, struct intel_super
*super
,
3206 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
3207 struct imsm_map
*prev
= get_imsm_map(dev
, dev
->vol
.migr_state
== 0 ?
3211 /* merge any IMSM_ORD_REBUILD bits that were not successfully
3212 * completed in the last migration.
3214 * FIXME add support for raid-level-migration
3216 if ((map_state
!= map
->map_state
) && (is_gen_migration(dev
) == 0) &&
3217 (prev
->map_state
!= IMSM_T_STATE_UNINITIALIZED
)) {
3218 /* when final map state is other than expected
3219 * merge maps (not for migration)
3223 for (i
= 0; i
< prev
->num_members
; i
++)
3224 for (j
= 0; j
< map
->num_members
; j
++)
3225 /* during online capacity expansion
3226 * disks position can be changed
3227 * if takeover is used
3229 if (ord_to_idx(map
->disk_ord_tbl
[j
]) ==
3230 ord_to_idx(prev
->disk_ord_tbl
[i
])) {
3231 map
->disk_ord_tbl
[j
] |=
3232 prev
->disk_ord_tbl
[i
];
3235 failed
= imsm_count_failed(super
, dev
, MAP_0
);
3236 map_state
= imsm_check_degraded(super
, dev
, failed
, MAP_0
);
3239 dev
->vol
.migr_state
= 0;
3240 set_migr_type(dev
, 0);
3241 dev
->vol
.curr_migr_unit
= 0;
3242 map
->map_state
= map_state
;
3246 static int parse_raid_devices(struct intel_super
*super
)
3249 struct imsm_dev
*dev_new
;
3250 size_t len
, len_migr
;
3252 size_t space_needed
= 0;
3253 struct imsm_super
*mpb
= super
->anchor
;
3255 for (i
= 0; i
< super
->anchor
->num_raid_devs
; i
++) {
3256 struct imsm_dev
*dev_iter
= __get_imsm_dev(super
->anchor
, i
);
3257 struct intel_dev
*dv
;
3259 len
= sizeof_imsm_dev(dev_iter
, 0);
3260 len_migr
= sizeof_imsm_dev(dev_iter
, 1);
3262 space_needed
+= len_migr
- len
;
3264 dv
= malloc(sizeof(*dv
));
3267 if (max_len
< len_migr
)
3269 if (max_len
> len_migr
)
3270 space_needed
+= max_len
- len_migr
;
3271 dev_new
= malloc(max_len
);
3276 imsm_copy_dev(dev_new
, dev_iter
);
3279 dv
->next
= super
->devlist
;
3280 super
->devlist
= dv
;
3283 /* ensure that super->buf is large enough when all raid devices
3286 if (__le32_to_cpu(mpb
->mpb_size
) + space_needed
> super
->len
) {
3289 len
= ROUND_UP(__le32_to_cpu(mpb
->mpb_size
) + space_needed
, 512);
3290 if (posix_memalign(&buf
, 512, len
) != 0)
3293 memcpy(buf
, super
->buf
, super
->len
);
3294 memset(buf
+ super
->len
, 0, len
- super
->len
);
3303 /* retrieve a pointer to the bbm log which starts after all raid devices */
3304 struct bbm_log
*__get_imsm_bbm_log(struct imsm_super
*mpb
)
3308 if (__le32_to_cpu(mpb
->bbm_log_size
)) {
3310 ptr
+= mpb
->mpb_size
- __le32_to_cpu(mpb
->bbm_log_size
);
3316 /*******************************************************************************
3317 * Function: check_mpb_migr_compatibility
3318 * Description: Function checks for unsupported migration features:
3319 * - migration optimization area (pba_of_lba0)
3320 * - descending reshape (ascending_migr)
3322 * super : imsm metadata information
3324 * 0 : migration is compatible
3325 * -1 : migration is not compatible
3326 ******************************************************************************/
3327 int check_mpb_migr_compatibility(struct intel_super
*super
)
3329 struct imsm_map
*map0
, *map1
;
3330 struct migr_record
*migr_rec
= super
->migr_rec
;
3333 for (i
= 0; i
< super
->anchor
->num_raid_devs
; i
++) {
3334 struct imsm_dev
*dev_iter
= __get_imsm_dev(super
->anchor
, i
);
3337 dev_iter
->vol
.migr_state
== 1 &&
3338 dev_iter
->vol
.migr_type
== MIGR_GEN_MIGR
) {
3339 /* This device is migrating */
3340 map0
= get_imsm_map(dev_iter
, MAP_0
);
3341 map1
= get_imsm_map(dev_iter
, MAP_1
);
3342 if (map0
->pba_of_lba0
!= map1
->pba_of_lba0
)
3343 /* migration optimization area was used */
3345 if (migr_rec
->ascending_migr
== 0
3346 && migr_rec
->dest_depth_per_unit
> 0)
3347 /* descending reshape not supported yet */
3354 static void __free_imsm(struct intel_super
*super
, int free_disks
);
3356 /* load_imsm_mpb - read matrix metadata
3357 * allocates super->mpb to be freed by free_imsm
3359 static int load_imsm_mpb(int fd
, struct intel_super
*super
, char *devname
)
3361 unsigned long long dsize
;
3362 unsigned long long sectors
;
3364 struct imsm_super
*anchor
;
3367 get_dev_size(fd
, NULL
, &dsize
);
3371 Name
": %s: device to small for imsm\n",
3376 if (lseek64(fd
, dsize
- (512 * 2), SEEK_SET
) < 0) {
3378 fprintf(stderr
, Name
3379 ": Cannot seek to anchor block on %s: %s\n",
3380 devname
, strerror(errno
));
3384 if (posix_memalign((void**)&anchor
, 512, 512) != 0) {
3387 Name
": Failed to allocate imsm anchor buffer"
3388 " on %s\n", devname
);
3391 if (read(fd
, anchor
, 512) != 512) {
3394 Name
": Cannot read anchor block on %s: %s\n",
3395 devname
, strerror(errno
));
3400 if (strncmp((char *) anchor
->sig
, MPB_SIGNATURE
, MPB_SIG_LEN
) != 0) {
3403 Name
": no IMSM anchor on %s\n", devname
);
3408 __free_imsm(super
, 0);
3409 /* reload capability and hba */
3411 /* capability and hba must be updated with new super allocation */
3412 find_intel_hba_capability(fd
, super
, devname
);
3413 super
->len
= ROUND_UP(anchor
->mpb_size
, 512);
3414 if (posix_memalign(&super
->buf
, 512, super
->len
) != 0) {
3417 Name
": unable to allocate %zu byte mpb buffer\n",
3422 memcpy(super
->buf
, anchor
, 512);
3424 sectors
= mpb_sectors(anchor
) - 1;
3427 if (posix_memalign(&super
->migr_rec_buf
, 512, MIGR_REC_BUF_SIZE
) != 0) {
3428 fprintf(stderr
, Name
3429 ": %s could not allocate migr_rec buffer\n", __func__
);
3435 check_sum
= __gen_imsm_checksum(super
->anchor
);
3436 if (check_sum
!= __le32_to_cpu(super
->anchor
->check_sum
)) {
3439 Name
": IMSM checksum %x != %x on %s\n",
3441 __le32_to_cpu(super
->anchor
->check_sum
),
3449 /* read the extended mpb */
3450 if (lseek64(fd
, dsize
- (512 * (2 + sectors
)), SEEK_SET
) < 0) {
3453 Name
": Cannot seek to extended mpb on %s: %s\n",
3454 devname
, strerror(errno
));
3458 if ((unsigned)read(fd
, super
->buf
+ 512, super
->len
- 512) != super
->len
- 512) {
3461 Name
": Cannot read extended mpb on %s: %s\n",
3462 devname
, strerror(errno
));
3466 check_sum
= __gen_imsm_checksum(super
->anchor
);
3467 if (check_sum
!= __le32_to_cpu(super
->anchor
->check_sum
)) {
3470 Name
": IMSM checksum %x != %x on %s\n",
3471 check_sum
, __le32_to_cpu(super
->anchor
->check_sum
),
3476 /* FIXME the BBM log is disk specific so we cannot use this global
3477 * buffer for all disks. Ok for now since we only look at the global
3478 * bbm_log_size parameter to gate assembly
3480 super
->bbm_log
= __get_imsm_bbm_log(super
->anchor
);
3485 static int read_imsm_migr_rec(int fd
, struct intel_super
*super
);
3488 load_and_parse_mpb(int fd
, struct intel_super
*super
, char *devname
, int keep_fd
)
3492 err
= load_imsm_mpb(fd
, super
, devname
);
3495 err
= load_imsm_disk(fd
, super
, devname
, keep_fd
);
3498 err
= parse_raid_devices(super
);
3503 static void __free_imsm_disk(struct dl
*d
)
3515 static void free_imsm_disks(struct intel_super
*super
)
3519 while (super
->disks
) {
3521 super
->disks
= d
->next
;
3522 __free_imsm_disk(d
);
3524 while (super
->disk_mgmt_list
) {
3525 d
= super
->disk_mgmt_list
;
3526 super
->disk_mgmt_list
= d
->next
;
3527 __free_imsm_disk(d
);
3529 while (super
->missing
) {
3531 super
->missing
= d
->next
;
3532 __free_imsm_disk(d
);
3537 /* free all the pieces hanging off of a super pointer */
3538 static void __free_imsm(struct intel_super
*super
, int free_disks
)
3540 struct intel_hba
*elem
, *next
;
3546 /* unlink capability description */
3548 if (super
->migr_rec_buf
) {
3549 free(super
->migr_rec_buf
);
3550 super
->migr_rec_buf
= NULL
;
3553 free_imsm_disks(super
);
3554 free_devlist(super
);
3558 free((void *)elem
->path
);
3566 static void free_imsm(struct intel_super
*super
)
3568 __free_imsm(super
, 1);
3572 static void free_super_imsm(struct supertype
*st
)
3574 struct intel_super
*super
= st
->sb
;
3583 static struct intel_super
*alloc_super(void)
3585 struct intel_super
*super
= malloc(sizeof(*super
));
3588 memset(super
, 0, sizeof(*super
));
3589 super
->current_vol
= -1;
3590 super
->create_offset
= ~((__u32
) 0);
3596 * find and allocate hba and OROM/EFI based on valid fd of RAID component device
3598 static int find_intel_hba_capability(int fd
, struct intel_super
*super
, char *devname
)
3600 struct sys_dev
*hba_name
;
3603 if ((fd
< 0) || check_env("IMSM_NO_PLATFORM")) {
3608 hba_name
= find_disk_attached_hba(fd
, NULL
);
3612 Name
": %s is not attached to Intel(R) RAID controller.\n",
3616 rv
= attach_hba_to_super(super
, hba_name
);
3619 struct intel_hba
*hba
= super
->hba
;
3621 fprintf(stderr
, Name
": %s is attached to Intel(R) %s RAID "
3622 "controller (%s),\n"
3623 " but the container is assigned to Intel(R) "
3624 "%s RAID controller (",
3627 hba_name
->pci_id
? : "Err!",
3628 get_sys_dev_type(hba_name
->type
));
3631 fprintf(stderr
, "%s", hba
->pci_id
? : "Err!");
3633 fprintf(stderr
, ", ");
3637 fprintf(stderr
, ").\n"
3638 " Mixing devices attached to different controllers "
3639 "is not allowed.\n");
3641 free_sys_dev(&hba_name
);
3644 super
->orom
= find_imsm_capability(hba_name
->type
);
3645 free_sys_dev(&hba_name
);
3651 /* find_missing - helper routine for load_super_imsm_all that identifies
3652 * disks that have disappeared from the system. This routine relies on
3653 * the mpb being uptodate, which it is at load time.
3655 static int find_missing(struct intel_super
*super
)
3658 struct imsm_super
*mpb
= super
->anchor
;
3660 struct imsm_disk
*disk
;
3662 for (i
= 0; i
< mpb
->num_disks
; i
++) {
3663 disk
= __get_imsm_disk(mpb
, i
);
3664 dl
= serial_to_dl(disk
->serial
, super
);
3668 dl
= malloc(sizeof(*dl
));
3674 dl
->devname
= strdup("missing");
3676 serialcpy(dl
->serial
, disk
->serial
);
3679 dl
->next
= super
->missing
;
3680 super
->missing
= dl
;
3687 static struct intel_disk
*disk_list_get(__u8
*serial
, struct intel_disk
*disk_list
)
3689 struct intel_disk
*idisk
= disk_list
;
3692 if (serialcmp(idisk
->disk
.serial
, serial
) == 0)
3694 idisk
= idisk
->next
;
3700 static int __prep_thunderdome(struct intel_super
**table
, int tbl_size
,
3701 struct intel_super
*super
,
3702 struct intel_disk
**disk_list
)
3704 struct imsm_disk
*d
= &super
->disks
->disk
;
3705 struct imsm_super
*mpb
= super
->anchor
;
3708 for (i
= 0; i
< tbl_size
; i
++) {
3709 struct imsm_super
*tbl_mpb
= table
[i
]->anchor
;
3710 struct imsm_disk
*tbl_d
= &table
[i
]->disks
->disk
;
3712 if (tbl_mpb
->family_num
== mpb
->family_num
) {
3713 if (tbl_mpb
->check_sum
== mpb
->check_sum
) {
3714 dprintf("%s: mpb from %d:%d matches %d:%d\n",
3715 __func__
, super
->disks
->major
,
3716 super
->disks
->minor
,
3717 table
[i
]->disks
->major
,
3718 table
[i
]->disks
->minor
);
3722 if (((is_configured(d
) && !is_configured(tbl_d
)) ||
3723 is_configured(d
) == is_configured(tbl_d
)) &&
3724 tbl_mpb
->generation_num
< mpb
->generation_num
) {
3725 /* current version of the mpb is a
3726 * better candidate than the one in
3727 * super_table, but copy over "cross
3728 * generational" status
3730 struct intel_disk
*idisk
;
3732 dprintf("%s: mpb from %d:%d replaces %d:%d\n",
3733 __func__
, super
->disks
->major
,
3734 super
->disks
->minor
,
3735 table
[i
]->disks
->major
,
3736 table
[i
]->disks
->minor
);
3738 idisk
= disk_list_get(tbl_d
->serial
, *disk_list
);
3739 if (idisk
&& is_failed(&idisk
->disk
))
3740 tbl_d
->status
|= FAILED_DISK
;
3743 struct intel_disk
*idisk
;
3744 struct imsm_disk
*disk
;
3746 /* tbl_mpb is more up to date, but copy
3747 * over cross generational status before
3750 disk
= __serial_to_disk(d
->serial
, mpb
, NULL
);
3751 if (disk
&& is_failed(disk
))
3752 d
->status
|= FAILED_DISK
;
3754 idisk
= disk_list_get(d
->serial
, *disk_list
);
3757 if (disk
&& is_configured(disk
))
3758 idisk
->disk
.status
|= CONFIGURED_DISK
;
3761 dprintf("%s: mpb from %d:%d prefer %d:%d\n",
3762 __func__
, super
->disks
->major
,
3763 super
->disks
->minor
,
3764 table
[i
]->disks
->major
,
3765 table
[i
]->disks
->minor
);
3773 table
[tbl_size
++] = super
;
3777 /* update/extend the merged list of imsm_disk records */
3778 for (j
= 0; j
< mpb
->num_disks
; j
++) {
3779 struct imsm_disk
*disk
= __get_imsm_disk(mpb
, j
);
3780 struct intel_disk
*idisk
;
3782 idisk
= disk_list_get(disk
->serial
, *disk_list
);
3784 idisk
->disk
.status
|= disk
->status
;
3785 if (is_configured(&idisk
->disk
) ||
3786 is_failed(&idisk
->disk
))
3787 idisk
->disk
.status
&= ~(SPARE_DISK
);
3789 idisk
= calloc(1, sizeof(*idisk
));
3792 idisk
->owner
= IMSM_UNKNOWN_OWNER
;
3793 idisk
->disk
= *disk
;
3794 idisk
->next
= *disk_list
;
3798 if (serialcmp(idisk
->disk
.serial
, d
->serial
) == 0)
3805 static struct intel_super
*
3806 validate_members(struct intel_super
*super
, struct intel_disk
*disk_list
,
3809 struct imsm_super
*mpb
= super
->anchor
;
3813 for (i
= 0; i
< mpb
->num_disks
; i
++) {
3814 struct imsm_disk
*disk
= __get_imsm_disk(mpb
, i
);
3815 struct intel_disk
*idisk
;
3817 idisk
= disk_list_get(disk
->serial
, disk_list
);
3819 if (idisk
->owner
== owner
||
3820 idisk
->owner
== IMSM_UNKNOWN_OWNER
)
3823 dprintf("%s: '%.16s' owner %d != %d\n",
3824 __func__
, disk
->serial
, idisk
->owner
,
3827 dprintf("%s: unknown disk %x [%d]: %.16s\n",
3828 __func__
, __le32_to_cpu(mpb
->family_num
), i
,
3834 if (ok_count
== mpb
->num_disks
)
3839 static void show_conflicts(__u32 family_num
, struct intel_super
*super_list
)
3841 struct intel_super
*s
;
3843 for (s
= super_list
; s
; s
= s
->next
) {
3844 if (family_num
!= s
->anchor
->family_num
)
3846 fprintf(stderr
, "Conflict, offlining family %#x on '%s'\n",
3847 __le32_to_cpu(family_num
), s
->disks
->devname
);
3851 static struct intel_super
*
3852 imsm_thunderdome(struct intel_super
**super_list
, int len
)
3854 struct intel_super
*super_table
[len
];
3855 struct intel_disk
*disk_list
= NULL
;
3856 struct intel_super
*champion
, *spare
;
3857 struct intel_super
*s
, **del
;
3862 memset(super_table
, 0, sizeof(super_table
));
3863 for (s
= *super_list
; s
; s
= s
->next
)
3864 tbl_size
= __prep_thunderdome(super_table
, tbl_size
, s
, &disk_list
);
3866 for (i
= 0; i
< tbl_size
; i
++) {
3867 struct imsm_disk
*d
;
3868 struct intel_disk
*idisk
;
3869 struct imsm_super
*mpb
= super_table
[i
]->anchor
;
3872 d
= &s
->disks
->disk
;
3874 /* 'd' must appear in merged disk list for its
3875 * configuration to be valid
3877 idisk
= disk_list_get(d
->serial
, disk_list
);
3878 if (idisk
&& idisk
->owner
== i
)
3879 s
= validate_members(s
, disk_list
, i
);
3884 dprintf("%s: marking family: %#x from %d:%d offline\n",
3885 __func__
, mpb
->family_num
,
3886 super_table
[i
]->disks
->major
,
3887 super_table
[i
]->disks
->minor
);
3891 /* This is where the mdadm implementation differs from the Windows
3892 * driver which has no strict concept of a container. We can only
3893 * assemble one family from a container, so when returning a prodigal
3894 * array member to this system the code will not be able to disambiguate
3895 * the container contents that should be assembled ("foreign" versus
3896 * "local"). It requires user intervention to set the orig_family_num
3897 * to a new value to establish a new container. The Windows driver in
3898 * this situation fixes up the volume name in place and manages the
3899 * foreign array as an independent entity.
3904 for (i
= 0; i
< tbl_size
; i
++) {
3905 struct intel_super
*tbl_ent
= super_table
[i
];
3911 if (tbl_ent
->anchor
->num_raid_devs
== 0) {
3916 if (s
&& !is_spare
) {
3917 show_conflicts(tbl_ent
->anchor
->family_num
, *super_list
);
3919 } else if (!s
&& !is_spare
)
3932 fprintf(stderr
, "Chose family %#x on '%s', "
3933 "assemble conflicts to new container with '--update=uuid'\n",
3934 __le32_to_cpu(s
->anchor
->family_num
), s
->disks
->devname
);
3936 /* collect all dl's onto 'champion', and update them to
3937 * champion's version of the status
3939 for (s
= *super_list
; s
; s
= s
->next
) {
3940 struct imsm_super
*mpb
= champion
->anchor
;
3941 struct dl
*dl
= s
->disks
;
3946 for (i
= 0; i
< mpb
->num_disks
; i
++) {
3947 struct imsm_disk
*disk
;
3949 disk
= __serial_to_disk(dl
->serial
, mpb
, &dl
->index
);
3952 /* only set index on disks that are a member of
3953 * a populated contianer, i.e. one with
3956 if (is_failed(&dl
->disk
))
3958 else if (is_spare(&dl
->disk
))
3964 if (i
>= mpb
->num_disks
) {
3965 struct intel_disk
*idisk
;
3967 idisk
= disk_list_get(dl
->serial
, disk_list
);
3968 if (idisk
&& is_spare(&idisk
->disk
) &&
3969 !is_failed(&idisk
->disk
) && !is_configured(&idisk
->disk
))
3977 dl
->next
= champion
->disks
;
3978 champion
->disks
= dl
;
3982 /* delete 'champion' from super_list */
3983 for (del
= super_list
; *del
; ) {
3984 if (*del
== champion
) {
3985 *del
= (*del
)->next
;
3988 del
= &(*del
)->next
;
3990 champion
->next
= NULL
;
3994 struct intel_disk
*idisk
= disk_list
;
3996 disk_list
= disk_list
->next
;
4003 static int load_super_imsm_all(struct supertype
*st
, int fd
, void **sbp
,
4007 struct intel_super
*super_list
= NULL
;
4008 struct intel_super
*super
= NULL
;
4009 int devnum
= fd2devnum(fd
);
4015 /* check if 'fd' an opened container */
4016 sra
= sysfs_read(fd
, 0, GET_LEVEL
|GET_VERSION
|GET_DEVS
|GET_STATE
);
4020 if (sra
->array
.major_version
!= -1 ||
4021 sra
->array
.minor_version
!= -2 ||
4022 strcmp(sra
->text_version
, "imsm") != 0) {
4027 for (sd
= sra
->devs
, i
= 0; sd
; sd
= sd
->next
, i
++) {
4028 struct intel_super
*s
= alloc_super();
4036 s
->next
= super_list
;
4040 sprintf(nm
, "%d:%d", sd
->disk
.major
, sd
->disk
.minor
);
4041 dfd
= dev_open(nm
, O_RDWR
);
4045 rv
= find_intel_hba_capability(dfd
, s
, devname
);
4046 /* no orom/efi or non-intel hba of the disk */
4050 err
= load_and_parse_mpb(dfd
, s
, NULL
, 1);
4052 /* retry the load if we might have raced against mdmon */
4053 if (err
== 3 && mdmon_running(devnum
))
4054 for (retry
= 0; retry
< 3; retry
++) {
4056 err
= load_and_parse_mpb(dfd
, s
, NULL
, 1);
4064 /* all mpbs enter, maybe one leaves */
4065 super
= imsm_thunderdome(&super_list
, i
);
4071 if (find_missing(super
) != 0) {
4077 /* load migration record */
4078 err
= load_imsm_migr_rec(super
, NULL
);
4084 /* Check migration compatibility */
4085 if (check_mpb_migr_compatibility(super
) != 0) {
4086 fprintf(stderr
, Name
": Unsupported migration detected");
4088 fprintf(stderr
, " on %s\n", devname
);
4090 fprintf(stderr
, " (IMSM).\n");
4099 while (super_list
) {
4100 struct intel_super
*s
= super_list
;
4102 super_list
= super_list
->next
;
4111 st
->container_dev
= devnum
;
4112 if (err
== 0 && st
->ss
== NULL
) {
4113 st
->ss
= &super_imsm
;
4114 st
->minor_version
= 0;
4115 st
->max_devs
= IMSM_MAX_DEVICES
;
4120 static int load_container_imsm(struct supertype
*st
, int fd
, char *devname
)
4122 return load_super_imsm_all(st
, fd
, &st
->sb
, devname
);
4126 static int load_super_imsm(struct supertype
*st
, int fd
, char *devname
)
4128 struct intel_super
*super
;
4131 if (test_partition(fd
))
4132 /* IMSM not allowed on partitions */
4135 free_super_imsm(st
);
4137 super
= alloc_super();
4140 Name
": malloc of %zu failed.\n",
4144 /* Load hba and capabilities if they exist.
4145 * But do not preclude loading metadata in case capabilities or hba are
4146 * non-compliant and ignore_hw_compat is set.
4148 rv
= find_intel_hba_capability(fd
, super
, devname
);
4149 /* no orom/efi or non-intel hba of the disk */
4150 if ((rv
!= 0) && (st
->ignore_hw_compat
== 0)) {
4153 Name
": No OROM/EFI properties for %s\n", devname
);
4157 rv
= load_and_parse_mpb(fd
, super
, devname
, 0);
4162 Name
": Failed to load all information "
4163 "sections on %s\n", devname
);
4169 if (st
->ss
== NULL
) {
4170 st
->ss
= &super_imsm
;
4171 st
->minor_version
= 0;
4172 st
->max_devs
= IMSM_MAX_DEVICES
;
4175 /* load migration record */
4176 if (load_imsm_migr_rec(super
, NULL
) == 0) {
4177 /* Check for unsupported migration features */
4178 if (check_mpb_migr_compatibility(super
) != 0) {
4180 Name
": Unsupported migration detected");
4182 fprintf(stderr
, " on %s\n", devname
);
4184 fprintf(stderr
, " (IMSM).\n");
4192 static __u16
info_to_blocks_per_strip(mdu_array_info_t
*info
)
4194 if (info
->level
== 1)
4196 return info
->chunk_size
>> 9;
4199 static __u32
info_to_num_data_stripes(mdu_array_info_t
*info
, int num_domains
)
4203 num_stripes
= (info
->size
* 2) / info_to_blocks_per_strip(info
);
4204 num_stripes
/= num_domains
;
4209 static __u32
info_to_blocks_per_member(mdu_array_info_t
*info
)
4211 if (info
->level
== 1)
4212 return info
->size
* 2;
4214 return (info
->size
* 2) & ~(info_to_blocks_per_strip(info
) - 1);
4217 static void imsm_update_version_info(struct intel_super
*super
)
4219 /* update the version and attributes */
4220 struct imsm_super
*mpb
= super
->anchor
;
4222 struct imsm_dev
*dev
;
4223 struct imsm_map
*map
;
4226 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
4227 dev
= get_imsm_dev(super
, i
);
4228 map
= get_imsm_map(dev
, MAP_0
);
4229 if (__le32_to_cpu(dev
->size_high
) > 0)
4230 mpb
->attributes
|= MPB_ATTRIB_2TB
;
4232 /* FIXME detect when an array spans a port multiplier */
4234 mpb
->attributes
|= MPB_ATTRIB_PM
;
4237 if (mpb
->num_raid_devs
> 1 ||
4238 mpb
->attributes
!= MPB_ATTRIB_CHECKSUM_VERIFY
) {
4239 version
= MPB_VERSION_ATTRIBS
;
4240 switch (get_imsm_raid_level(map
)) {
4241 case 0: mpb
->attributes
|= MPB_ATTRIB_RAID0
; break;
4242 case 1: mpb
->attributes
|= MPB_ATTRIB_RAID1
; break;
4243 case 10: mpb
->attributes
|= MPB_ATTRIB_RAID10
; break;
4244 case 5: mpb
->attributes
|= MPB_ATTRIB_RAID5
; break;
4247 if (map
->num_members
>= 5)
4248 version
= MPB_VERSION_5OR6_DISK_ARRAY
;
4249 else if (dev
->status
== DEV_CLONE_N_GO
)
4250 version
= MPB_VERSION_CNG
;
4251 else if (get_imsm_raid_level(map
) == 5)
4252 version
= MPB_VERSION_RAID5
;
4253 else if (map
->num_members
>= 3)
4254 version
= MPB_VERSION_3OR4_DISK_ARRAY
;
4255 else if (get_imsm_raid_level(map
) == 1)
4256 version
= MPB_VERSION_RAID1
;
4258 version
= MPB_VERSION_RAID0
;
4260 strcpy(((char *) mpb
->sig
) + strlen(MPB_SIGNATURE
), version
);
4264 static int check_name(struct intel_super
*super
, char *name
, int quiet
)
4266 struct imsm_super
*mpb
= super
->anchor
;
4267 char *reason
= NULL
;
4270 if (strlen(name
) > MAX_RAID_SERIAL_LEN
)
4271 reason
= "must be 16 characters or less";
4273 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
4274 struct imsm_dev
*dev
= get_imsm_dev(super
, i
);
4276 if (strncmp((char *) dev
->volume
, name
, MAX_RAID_SERIAL_LEN
) == 0) {
4277 reason
= "already exists";
4282 if (reason
&& !quiet
)
4283 fprintf(stderr
, Name
": imsm volume name %s\n", reason
);
4288 static int init_super_imsm_volume(struct supertype
*st
, mdu_array_info_t
*info
,
4289 unsigned long long size
, char *name
,
4290 char *homehost
, int *uuid
)
4292 /* We are creating a volume inside a pre-existing container.
4293 * so st->sb is already set.
4295 struct intel_super
*super
= st
->sb
;
4296 struct imsm_super
*mpb
= super
->anchor
;
4297 struct intel_dev
*dv
;
4298 struct imsm_dev
*dev
;
4299 struct imsm_vol
*vol
;
4300 struct imsm_map
*map
;
4301 int idx
= mpb
->num_raid_devs
;
4303 unsigned long long array_blocks
;
4304 size_t size_old
, size_new
;
4305 __u32 num_data_stripes
;
4307 if (super
->orom
&& mpb
->num_raid_devs
>= super
->orom
->vpa
) {
4308 fprintf(stderr
, Name
": This imsm-container already has the "
4309 "maximum of %d volumes\n", super
->orom
->vpa
);
4313 /* ensure the mpb is large enough for the new data */
4314 size_old
= __le32_to_cpu(mpb
->mpb_size
);
4315 size_new
= disks_to_mpb_size(info
->nr_disks
);
4316 if (size_new
> size_old
) {
4318 size_t size_round
= ROUND_UP(size_new
, 512);
4320 if (posix_memalign(&mpb_new
, 512, size_round
) != 0) {
4321 fprintf(stderr
, Name
": could not allocate new mpb\n");
4324 if (posix_memalign(&super
->migr_rec_buf
, 512,
4325 MIGR_REC_BUF_SIZE
) != 0) {
4326 fprintf(stderr
, Name
4327 ": %s could not allocate migr_rec buffer\n",
4334 memcpy(mpb_new
, mpb
, size_old
);
4337 super
->anchor
= mpb_new
;
4338 mpb
->mpb_size
= __cpu_to_le32(size_new
);
4339 memset(mpb_new
+ size_old
, 0, size_round
- size_old
);
4341 super
->current_vol
= idx
;
4343 /* handle 'failed_disks' by either:
4344 * a) create dummy disk entries in the table if this the first
4345 * volume in the array. We add them here as this is the only
4346 * opportunity to add them. add_to_super_imsm_volume()
4347 * handles the non-failed disks and continues incrementing
4349 * b) validate that 'failed_disks' matches the current number
4350 * of missing disks if the container is populated
4352 if (super
->current_vol
== 0) {
4354 for (i
= 0; i
< info
->failed_disks
; i
++) {
4355 struct imsm_disk
*disk
;
4358 disk
= __get_imsm_disk(mpb
, i
);
4359 disk
->status
= CONFIGURED_DISK
| FAILED_DISK
;
4360 disk
->scsi_id
= __cpu_to_le32(~(__u32
)0);
4361 snprintf((char *) disk
->serial
, MAX_RAID_SERIAL_LEN
,
4364 find_missing(super
);
4369 for (d
= super
->missing
; d
; d
= d
->next
)
4371 if (info
->failed_disks
> missing
) {
4372 fprintf(stderr
, Name
": unable to add 'missing' disk to container\n");
4377 if (!check_name(super
, name
, 0))
4379 dv
= malloc(sizeof(*dv
));
4381 fprintf(stderr
, Name
": failed to allocate device list entry\n");
4384 dev
= calloc(1, sizeof(*dev
) + sizeof(__u32
) * (info
->raid_disks
- 1));
4387 fprintf(stderr
, Name
": could not allocate raid device\n");
4391 strncpy((char *) dev
->volume
, name
, MAX_RAID_SERIAL_LEN
);
4392 if (info
->level
== 1)
4393 array_blocks
= info_to_blocks_per_member(info
);
4395 array_blocks
= calc_array_size(info
->level
, info
->raid_disks
,
4396 info
->layout
, info
->chunk_size
,
4398 /* round array size down to closest MB */
4399 array_blocks
= (array_blocks
>> SECT_PER_MB_SHIFT
) << SECT_PER_MB_SHIFT
;
4401 dev
->size_low
= __cpu_to_le32((__u32
) array_blocks
);
4402 dev
->size_high
= __cpu_to_le32((__u32
) (array_blocks
>> 32));
4403 dev
->status
= (DEV_READ_COALESCING
| DEV_WRITE_COALESCING
);
4405 vol
->migr_state
= 0;
4406 set_migr_type(dev
, MIGR_INIT
);
4407 vol
->dirty
= !info
->state
;
4408 vol
->curr_migr_unit
= 0;
4409 map
= get_imsm_map(dev
, MAP_0
);
4410 map
->pba_of_lba0
= __cpu_to_le32(super
->create_offset
);
4411 map
->blocks_per_member
= __cpu_to_le32(info_to_blocks_per_member(info
));
4412 map
->blocks_per_strip
= __cpu_to_le16(info_to_blocks_per_strip(info
));
4413 map
->failed_disk_num
= ~0;
4414 if (info
->level
> 0)
4415 map
->map_state
= IMSM_T_STATE_UNINITIALIZED
;
4417 map
->map_state
= info
->failed_disks
? IMSM_T_STATE_FAILED
:
4418 IMSM_T_STATE_NORMAL
;
4421 if (info
->level
== 1 && info
->raid_disks
> 2) {
4424 fprintf(stderr
, Name
": imsm does not support more than 2 disks"
4425 "in a raid1 volume\n");
4429 map
->raid_level
= info
->level
;
4430 if (info
->level
== 10) {
4431 map
->raid_level
= 1;
4432 map
->num_domains
= info
->raid_disks
/ 2;
4433 } else if (info
->level
== 1)
4434 map
->num_domains
= info
->raid_disks
;
4436 map
->num_domains
= 1;
4438 num_data_stripes
= info_to_num_data_stripes(info
, map
->num_domains
);
4439 map
->num_data_stripes
= __cpu_to_le32(num_data_stripes
);
4441 map
->num_members
= info
->raid_disks
;
4442 for (i
= 0; i
< map
->num_members
; i
++) {
4443 /* initialized in add_to_super */
4444 set_imsm_ord_tbl_ent(map
, i
, IMSM_ORD_REBUILD
);
4446 mpb
->num_raid_devs
++;
4449 dv
->index
= super
->current_vol
;
4450 dv
->next
= super
->devlist
;
4451 super
->devlist
= dv
;
4453 imsm_update_version_info(super
);
4458 static int init_super_imsm(struct supertype
*st
, mdu_array_info_t
*info
,
4459 unsigned long long size
, char *name
,
4460 char *homehost
, int *uuid
)
4462 /* This is primarily called by Create when creating a new array.
4463 * We will then get add_to_super called for each component, and then
4464 * write_init_super called to write it out to each device.
4465 * For IMSM, Create can create on fresh devices or on a pre-existing
4467 * To create on a pre-existing array a different method will be called.
4468 * This one is just for fresh drives.
4470 struct intel_super
*super
;
4471 struct imsm_super
*mpb
;
4476 return init_super_imsm_volume(st
, info
, size
, name
, homehost
, uuid
);
4479 mpb_size
= disks_to_mpb_size(info
->nr_disks
);
4483 super
= alloc_super();
4484 if (super
&& posix_memalign(&super
->buf
, 512, mpb_size
) != 0) {
4489 fprintf(stderr
, Name
4490 ": %s could not allocate superblock\n", __func__
);
4493 if (posix_memalign(&super
->migr_rec_buf
, 512, MIGR_REC_BUF_SIZE
) != 0) {
4494 fprintf(stderr
, Name
4495 ": %s could not allocate migr_rec buffer\n", __func__
);
4500 memset(super
->buf
, 0, mpb_size
);
4502 mpb
->mpb_size
= __cpu_to_le32(mpb_size
);
4506 /* zeroing superblock */
4510 mpb
->attributes
= MPB_ATTRIB_CHECKSUM_VERIFY
;
4512 version
= (char *) mpb
->sig
;
4513 strcpy(version
, MPB_SIGNATURE
);
4514 version
+= strlen(MPB_SIGNATURE
);
4515 strcpy(version
, MPB_VERSION_RAID0
);
4521 static int add_to_super_imsm_volume(struct supertype
*st
, mdu_disk_info_t
*dk
,
4522 int fd
, char *devname
)
4524 struct intel_super
*super
= st
->sb
;
4525 struct imsm_super
*mpb
= super
->anchor
;
4526 struct imsm_disk
*_disk
;
4527 struct imsm_dev
*dev
;
4528 struct imsm_map
*map
;
4532 dev
= get_imsm_dev(super
, super
->current_vol
);
4533 map
= get_imsm_map(dev
, MAP_0
);
4535 if (! (dk
->state
& (1<<MD_DISK_SYNC
))) {
4536 fprintf(stderr
, Name
": %s: Cannot add spare devices to IMSM volume\n",
4542 /* we're doing autolayout so grab the pre-marked (in
4543 * validate_geometry) raid_disk
4545 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
4546 if (dl
->raiddisk
== dk
->raid_disk
)
4549 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
4550 if (dl
->major
== dk
->major
&&
4551 dl
->minor
== dk
->minor
)
4556 fprintf(stderr
, Name
": %s is not a member of the same container\n", devname
);
4560 /* add a pristine spare to the metadata */
4561 if (dl
->index
< 0) {
4562 dl
->index
= super
->anchor
->num_disks
;
4563 super
->anchor
->num_disks
++;
4565 /* Check the device has not already been added */
4566 slot
= get_imsm_disk_slot(map
, dl
->index
);
4568 (get_imsm_ord_tbl_ent(dev
, slot
, MAP_X
) & IMSM_ORD_REBUILD
) == 0) {
4569 fprintf(stderr
, Name
": %s has been included in this array twice\n",
4573 set_imsm_ord_tbl_ent(map
, dk
->raid_disk
, dl
->index
);
4574 dl
->disk
.status
= CONFIGURED_DISK
;
4576 /* update size of 'missing' disks to be at least as large as the
4577 * largest acitve member (we only have dummy missing disks when
4578 * creating the first volume)
4580 if (super
->current_vol
== 0) {
4581 for (df
= super
->missing
; df
; df
= df
->next
) {
4582 if (dl
->disk
.total_blocks
> df
->disk
.total_blocks
)
4583 df
->disk
.total_blocks
= dl
->disk
.total_blocks
;
4584 _disk
= __get_imsm_disk(mpb
, df
->index
);
4589 /* refresh unset/failed slots to point to valid 'missing' entries */
4590 for (df
= super
->missing
; df
; df
= df
->next
)
4591 for (slot
= 0; slot
< mpb
->num_disks
; slot
++) {
4592 __u32 ord
= get_imsm_ord_tbl_ent(dev
, slot
, MAP_X
);
4594 if ((ord
& IMSM_ORD_REBUILD
) == 0)
4596 set_imsm_ord_tbl_ent(map
, slot
, df
->index
| IMSM_ORD_REBUILD
);
4597 if (is_gen_migration(dev
)) {
4598 struct imsm_map
*map2
= get_imsm_map(dev
,
4600 int slot2
= get_imsm_disk_slot(map2
, df
->index
);
4601 if ((slot2
< map2
->num_members
) &&
4603 __u32 ord2
= get_imsm_ord_tbl_ent(dev
,
4606 if ((unsigned)df
->index
==
4608 set_imsm_ord_tbl_ent(map2
,
4614 dprintf("set slot:%d to missing disk:%d\n", slot
, df
->index
);
4618 /* if we are creating the first raid device update the family number */
4619 if (super
->current_vol
== 0) {
4621 struct imsm_dev
*_dev
= __get_imsm_dev(mpb
, 0);
4623 _disk
= __get_imsm_disk(mpb
, dl
->index
);
4624 if (!_dev
|| !_disk
) {
4625 fprintf(stderr
, Name
": BUG mpb setup error\n");
4631 sum
+= __gen_imsm_checksum(mpb
);
4632 mpb
->family_num
= __cpu_to_le32(sum
);
4633 mpb
->orig_family_num
= mpb
->family_num
;
4635 super
->current_disk
= dl
;
4640 * Function marks disk as spare and restores disk serial
4641 * in case it was previously marked as failed by takeover operation
4643 * -1 : critical error
4644 * 0 : disk is marked as spare but serial is not set
4647 int mark_spare(struct dl
*disk
)
4649 __u8 serial
[MAX_RAID_SERIAL_LEN
];
4656 if (!imsm_read_serial(disk
->fd
, NULL
, serial
)) {
4657 /* Restore disk serial number, because takeover marks disk
4658 * as failed and adds to serial ':0' before it becomes
4661 serialcpy(disk
->serial
, serial
);
4662 serialcpy(disk
->disk
.serial
, serial
);
4665 disk
->disk
.status
= SPARE_DISK
;
4671 static int add_to_super_imsm(struct supertype
*st
, mdu_disk_info_t
*dk
,
4672 int fd
, char *devname
)
4674 struct intel_super
*super
= st
->sb
;
4676 unsigned long long size
;
4681 /* If we are on an RAID enabled platform check that the disk is
4682 * attached to the raid controller.
4683 * We do not need to test disks attachment for container based additions,
4684 * they shall be already tested when container was created/assembled.
4686 rv
= find_intel_hba_capability(fd
, super
, devname
);
4687 /* no orom/efi or non-intel hba of the disk */
4689 dprintf("capability: %p fd: %d ret: %d\n",
4690 super
->orom
, fd
, rv
);
4694 if (super
->current_vol
>= 0)
4695 return add_to_super_imsm_volume(st
, dk
, fd
, devname
);
4698 dd
= malloc(sizeof(*dd
));
4701 Name
": malloc failed %s:%d.\n", __func__
, __LINE__
);
4704 memset(dd
, 0, sizeof(*dd
));
4705 dd
->major
= major(stb
.st_rdev
);
4706 dd
->minor
= minor(stb
.st_rdev
);
4707 dd
->devname
= devname
? strdup(devname
) : NULL
;
4710 dd
->action
= DISK_ADD
;
4711 rv
= imsm_read_serial(fd
, devname
, dd
->serial
);
4714 Name
": failed to retrieve scsi serial, aborting\n");
4719 get_dev_size(fd
, NULL
, &size
);
4721 serialcpy(dd
->disk
.serial
, dd
->serial
);
4722 dd
->disk
.total_blocks
= __cpu_to_le32(size
);
4724 if (sysfs_disk_to_scsi_id(fd
, &id
) == 0)
4725 dd
->disk
.scsi_id
= __cpu_to_le32(id
);
4727 dd
->disk
.scsi_id
= __cpu_to_le32(0);
4729 if (st
->update_tail
) {
4730 dd
->next
= super
->disk_mgmt_list
;
4731 super
->disk_mgmt_list
= dd
;
4733 dd
->next
= super
->disks
;
4735 super
->updates_pending
++;
4742 static int remove_from_super_imsm(struct supertype
*st
, mdu_disk_info_t
*dk
)
4744 struct intel_super
*super
= st
->sb
;
4747 /* remove from super works only in mdmon - for communication
4748 * manager - monitor. Check if communication memory buffer
4751 if (!st
->update_tail
) {
4753 Name
": %s shall be used in mdmon context only"
4754 "(line %d).\n", __func__
, __LINE__
);
4757 dd
= malloc(sizeof(*dd
));
4760 Name
": malloc failed %s:%d.\n", __func__
, __LINE__
);
4763 memset(dd
, 0, sizeof(*dd
));
4764 dd
->major
= dk
->major
;
4765 dd
->minor
= dk
->minor
;
4768 dd
->action
= DISK_REMOVE
;
4770 dd
->next
= super
->disk_mgmt_list
;
4771 super
->disk_mgmt_list
= dd
;
4777 static int store_imsm_mpb(int fd
, struct imsm_super
*mpb
);
4781 struct imsm_super anchor
;
4782 } spare_record
__attribute__ ((aligned(512)));
4784 /* spare records have their own family number and do not have any defined raid
4787 static int write_super_imsm_spares(struct intel_super
*super
, int doclose
)
4789 struct imsm_super
*mpb
= super
->anchor
;
4790 struct imsm_super
*spare
= &spare_record
.anchor
;
4794 spare
->mpb_size
= __cpu_to_le32(sizeof(struct imsm_super
)),
4795 spare
->generation_num
= __cpu_to_le32(1UL),
4796 spare
->attributes
= MPB_ATTRIB_CHECKSUM_VERIFY
;
4797 spare
->num_disks
= 1,
4798 spare
->num_raid_devs
= 0,
4799 spare
->cache_size
= mpb
->cache_size
,
4800 spare
->pwr_cycle_count
= __cpu_to_le32(1),
4802 snprintf((char *) spare
->sig
, MAX_SIGNATURE_LENGTH
,
4803 MPB_SIGNATURE MPB_VERSION_RAID0
);
4805 for (d
= super
->disks
; d
; d
= d
->next
) {
4809 spare
->disk
[0] = d
->disk
;
4810 sum
= __gen_imsm_checksum(spare
);
4811 spare
->family_num
= __cpu_to_le32(sum
);
4812 spare
->orig_family_num
= 0;
4813 sum
= __gen_imsm_checksum(spare
);
4814 spare
->check_sum
= __cpu_to_le32(sum
);
4816 if (store_imsm_mpb(d
->fd
, spare
)) {
4817 fprintf(stderr
, "%s: failed for device %d:%d %s\n",
4818 __func__
, d
->major
, d
->minor
, strerror(errno
));
4830 static int write_super_imsm(struct supertype
*st
, int doclose
)
4832 struct intel_super
*super
= st
->sb
;
4833 struct imsm_super
*mpb
= super
->anchor
;
4839 __u32 mpb_size
= sizeof(struct imsm_super
) - sizeof(struct imsm_disk
);
4841 int clear_migration_record
= 1;
4843 /* 'generation' is incremented everytime the metadata is written */
4844 generation
= __le32_to_cpu(mpb
->generation_num
);
4846 mpb
->generation_num
= __cpu_to_le32(generation
);
4848 /* fix up cases where previous mdadm releases failed to set
4851 if (mpb
->orig_family_num
== 0)
4852 mpb
->orig_family_num
= mpb
->family_num
;
4854 for (d
= super
->disks
; d
; d
= d
->next
) {
4858 mpb
->disk
[d
->index
] = d
->disk
;
4862 for (d
= super
->missing
; d
; d
= d
->next
) {
4863 mpb
->disk
[d
->index
] = d
->disk
;
4866 mpb
->num_disks
= num_disks
;
4867 mpb_size
+= sizeof(struct imsm_disk
) * mpb
->num_disks
;
4869 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
4870 struct imsm_dev
*dev
= __get_imsm_dev(mpb
, i
);
4871 struct imsm_dev
*dev2
= get_imsm_dev(super
, i
);
4873 imsm_copy_dev(dev
, dev2
);
4874 mpb_size
+= sizeof_imsm_dev(dev
, 0);
4876 if (is_gen_migration(dev2
))
4877 clear_migration_record
= 0;
4879 mpb_size
+= __le32_to_cpu(mpb
->bbm_log_size
);
4880 mpb
->mpb_size
= __cpu_to_le32(mpb_size
);
4882 /* recalculate checksum */
4883 sum
= __gen_imsm_checksum(mpb
);
4884 mpb
->check_sum
= __cpu_to_le32(sum
);
4886 if (clear_migration_record
)
4887 memset(super
->migr_rec_buf
, 0, MIGR_REC_BUF_SIZE
);
4889 /* write the mpb for disks that compose raid devices */
4890 for (d
= super
->disks
; d
; d
= d
->next
) {
4891 if (d
->index
< 0 || is_failed(&d
->disk
))
4893 if (store_imsm_mpb(d
->fd
, mpb
))
4894 fprintf(stderr
, "%s: failed for device %d:%d %s\n",
4895 __func__
, d
->major
, d
->minor
, strerror(errno
));
4896 if (clear_migration_record
) {
4897 unsigned long long dsize
;
4899 get_dev_size(d
->fd
, NULL
, &dsize
);
4900 if (lseek64(d
->fd
, dsize
- 512, SEEK_SET
) >= 0) {
4901 if (write(d
->fd
, super
->migr_rec_buf
,
4902 MIGR_REC_BUF_SIZE
) != MIGR_REC_BUF_SIZE
)
4903 perror("Write migr_rec failed");
4913 return write_super_imsm_spares(super
, doclose
);
4919 static int create_array(struct supertype
*st
, int dev_idx
)
4922 struct imsm_update_create_array
*u
;
4923 struct intel_super
*super
= st
->sb
;
4924 struct imsm_dev
*dev
= get_imsm_dev(super
, dev_idx
);
4925 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
4926 struct disk_info
*inf
;
4927 struct imsm_disk
*disk
;
4930 len
= sizeof(*u
) - sizeof(*dev
) + sizeof_imsm_dev(dev
, 0) +
4931 sizeof(*inf
) * map
->num_members
;
4934 fprintf(stderr
, "%s: failed to allocate update buffer\n",
4939 u
->type
= update_create_array
;
4940 u
->dev_idx
= dev_idx
;
4941 imsm_copy_dev(&u
->dev
, dev
);
4942 inf
= get_disk_info(u
);
4943 for (i
= 0; i
< map
->num_members
; i
++) {
4944 int idx
= get_imsm_disk_idx(dev
, i
, MAP_X
);
4946 disk
= get_imsm_disk(super
, idx
);
4947 serialcpy(inf
[i
].serial
, disk
->serial
);
4949 append_metadata_update(st
, u
, len
);
4954 static int mgmt_disk(struct supertype
*st
)
4956 struct intel_super
*super
= st
->sb
;
4958 struct imsm_update_add_remove_disk
*u
;
4960 if (!super
->disk_mgmt_list
)
4966 fprintf(stderr
, "%s: failed to allocate update buffer\n",
4971 u
->type
= update_add_remove_disk
;
4972 append_metadata_update(st
, u
, len
);
4977 static int write_init_super_imsm(struct supertype
*st
)
4979 struct intel_super
*super
= st
->sb
;
4980 int current_vol
= super
->current_vol
;
4982 /* we are done with current_vol reset it to point st at the container */
4983 super
->current_vol
= -1;
4985 if (st
->update_tail
) {
4986 /* queue the recently created array / added disk
4987 * as a metadata update */
4990 /* determine if we are creating a volume or adding a disk */
4991 if (current_vol
< 0) {
4992 /* in the mgmt (add/remove) disk case we are running
4993 * in mdmon context, so don't close fd's
4995 return mgmt_disk(st
);
4997 rv
= create_array(st
, current_vol
);
5002 for (d
= super
->disks
; d
; d
= d
->next
)
5003 Kill(d
->devname
, NULL
, 0, 1, 1);
5004 return write_super_imsm(st
, 1);
5009 static int store_super_imsm(struct supertype
*st
, int fd
)
5011 struct intel_super
*super
= st
->sb
;
5012 struct imsm_super
*mpb
= super
? super
->anchor
: NULL
;
5018 return store_imsm_mpb(fd
, mpb
);
5024 static int imsm_bbm_log_size(struct imsm_super
*mpb
)
5026 return __le32_to_cpu(mpb
->bbm_log_size
);
5030 static int validate_geometry_imsm_container(struct supertype
*st
, int level
,
5031 int layout
, int raiddisks
, int chunk
,
5032 unsigned long long size
, char *dev
,
5033 unsigned long long *freesize
,
5037 unsigned long long ldsize
;
5038 struct intel_super
*super
=NULL
;
5041 if (level
!= LEVEL_CONTAINER
)
5046 fd
= open(dev
, O_RDONLY
|O_EXCL
, 0);
5049 fprintf(stderr
, Name
": imsm: Cannot open %s: %s\n",
5050 dev
, strerror(errno
));
5053 if (!get_dev_size(fd
, dev
, &ldsize
)) {
5058 /* capabilities retrieve could be possible
5059 * note that there is no fd for the disks in array.
5061 super
= alloc_super();
5064 Name
": malloc of %zu failed.\n",
5070 rv
= find_intel_hba_capability(fd
, super
, verbose
? dev
: NULL
);
5074 fd2devname(fd
, str
);
5075 dprintf("validate_geometry_imsm_container: fd: %d %s orom: %p rv: %d raiddisk: %d\n",
5076 fd
, str
, super
->orom
, rv
, raiddisks
);
5078 /* no orom/efi or non-intel hba of the disk */
5084 if (super
->orom
&& raiddisks
> super
->orom
->tds
) {
5086 fprintf(stderr
, Name
": %d exceeds maximum number of"
5087 " platform supported disks: %d\n",
5088 raiddisks
, super
->orom
->tds
);
5094 *freesize
= avail_size_imsm(st
, ldsize
>> 9);
5100 static unsigned long long find_size(struct extent
*e
, int *idx
, int num_extents
)
5102 const unsigned long long base_start
= e
[*idx
].start
;
5103 unsigned long long end
= base_start
+ e
[*idx
].size
;
5106 if (base_start
== end
)
5110 for (i
= *idx
; i
< num_extents
; i
++) {
5111 /* extend overlapping extents */
5112 if (e
[i
].start
>= base_start
&&
5113 e
[i
].start
<= end
) {
5116 if (e
[i
].start
+ e
[i
].size
> end
)
5117 end
= e
[i
].start
+ e
[i
].size
;
5118 } else if (e
[i
].start
> end
) {
5124 return end
- base_start
;
5127 static unsigned long long merge_extents(struct intel_super
*super
, int sum_extents
)
5129 /* build a composite disk with all known extents and generate a new
5130 * 'maxsize' given the "all disks in an array must share a common start
5131 * offset" constraint
5133 struct extent
*e
= calloc(sum_extents
, sizeof(*e
));
5137 unsigned long long pos
;
5138 unsigned long long start
= 0;
5139 unsigned long long maxsize
;
5140 unsigned long reserve
;
5145 /* coalesce and sort all extents. also, check to see if we need to
5146 * reserve space between member arrays
5149 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
5152 for (i
= 0; i
< dl
->extent_cnt
; i
++)
5155 qsort(e
, sum_extents
, sizeof(*e
), cmp_extent
);
5160 while (i
< sum_extents
) {
5161 e
[j
].start
= e
[i
].start
;
5162 e
[j
].size
= find_size(e
, &i
, sum_extents
);
5164 if (e
[j
-1].size
== 0)
5173 unsigned long long esize
;
5175 esize
= e
[i
].start
- pos
;
5176 if (esize
>= maxsize
) {
5181 pos
= e
[i
].start
+ e
[i
].size
;
5183 } while (e
[i
-1].size
);
5189 /* FIXME assumes volume at offset 0 is the first volume in a
5192 if (start_extent
> 0)
5193 reserve
= IMSM_RESERVED_SECTORS
; /* gap between raid regions */
5197 if (maxsize
< reserve
)
5200 super
->create_offset
= ~((__u32
) 0);
5201 if (start
+ reserve
> super
->create_offset
)
5202 return 0; /* start overflows create_offset */
5203 super
->create_offset
= start
+ reserve
;
5205 return maxsize
- reserve
;
5208 static int is_raid_level_supported(const struct imsm_orom
*orom
, int level
, int raiddisks
)
5210 if (level
< 0 || level
== 6 || level
== 4)
5213 /* if we have an orom prevent invalid raid levels */
5216 case 0: return imsm_orom_has_raid0(orom
);
5219 return imsm_orom_has_raid1e(orom
);
5220 return imsm_orom_has_raid1(orom
) && raiddisks
== 2;
5221 case 10: return imsm_orom_has_raid10(orom
) && raiddisks
== 4;
5222 case 5: return imsm_orom_has_raid5(orom
) && raiddisks
> 2;
5225 return 1; /* not on an Intel RAID platform so anything goes */
5230 static int imsm_default_chunk(const struct imsm_orom
*orom
)
5232 /* up to 512 if the plaform supports it, otherwise the platform max.
5233 * 128 if no platform detected
5235 int fs
= max(7, orom
? fls(orom
->sss
) : 0);
5237 return min(512, (1 << fs
));
5240 #define pr_vrb(fmt, arg...) (void) (verbose && fprintf(stderr, Name fmt, ##arg))
5242 validate_geometry_imsm_orom(struct intel_super
*super
, int level
, int layout
,
5243 int raiddisks
, int *chunk
, int verbose
)
5245 /* check/set platform and metadata limits/defaults */
5246 if (super
->orom
&& raiddisks
> super
->orom
->dpa
) {
5247 pr_vrb(": platform supports a maximum of %d disks per array\n",
5252 /* capabilities of OROM tested - copied from validate_geometry_imsm_volume */
5253 if (!is_raid_level_supported(super
->orom
, level
, raiddisks
)) {
5254 pr_vrb(": platform does not support raid%d with %d disk%s\n",
5255 level
, raiddisks
, raiddisks
> 1 ? "s" : "");
5259 if (chunk
&& (*chunk
== 0 || *chunk
== UnSet
))
5260 *chunk
= imsm_default_chunk(super
->orom
);
5262 if (super
->orom
&& chunk
&& !imsm_orom_has_chunk(super
->orom
, *chunk
)) {
5263 pr_vrb(": platform does not support a chunk size of: "
5268 if (layout
!= imsm_level_to_layout(level
)) {
5270 pr_vrb(": imsm raid 5 only supports the left-asymmetric layout\n");
5271 else if (level
== 10)
5272 pr_vrb(": imsm raid 10 only supports the n2 layout\n");
5274 pr_vrb(": imsm unknown layout %#x for this raid level %d\n",
5281 /* validate_geometry_imsm_volume - lifted from validate_geometry_ddf_bvd
5282 * FIX ME add ahci details
5284 static int validate_geometry_imsm_volume(struct supertype
*st
, int level
,
5285 int layout
, int raiddisks
, int *chunk
,
5286 unsigned long long size
, char *dev
,
5287 unsigned long long *freesize
,
5291 struct intel_super
*super
= st
->sb
;
5292 struct imsm_super
*mpb
;
5294 unsigned long long pos
= 0;
5295 unsigned long long maxsize
;
5299 /* We must have the container info already read in. */
5303 mpb
= super
->anchor
;
5305 if (mpb
->num_raid_devs
> 0 && mpb
->num_disks
!= raiddisks
) {
5306 fprintf(stderr
, Name
": the option-rom requires all "
5307 "member disks to be a member of all volumes.\n");
5311 if (!validate_geometry_imsm_orom(super
, level
, layout
, raiddisks
, chunk
, verbose
)) {
5312 fprintf(stderr
, Name
": RAID gemetry validation failed. "
5313 "Cannot proceed with the action(s).\n");
5317 /* General test: make sure there is space for
5318 * 'raiddisks' device extents of size 'size' at a given
5321 unsigned long long minsize
= size
;
5322 unsigned long long start_offset
= MaxSector
;
5325 minsize
= MPB_SECTOR_CNT
+ IMSM_RESERVED_SECTORS
;
5326 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
5331 e
= get_extents(super
, dl
);
5334 unsigned long long esize
;
5335 esize
= e
[i
].start
- pos
;
5336 if (esize
>= minsize
)
5338 if (found
&& start_offset
== MaxSector
) {
5341 } else if (found
&& pos
!= start_offset
) {
5345 pos
= e
[i
].start
+ e
[i
].size
;
5347 } while (e
[i
-1].size
);
5352 if (dcnt
< raiddisks
) {
5354 fprintf(stderr
, Name
": imsm: Not enough "
5355 "devices with space for this array "
5363 /* This device must be a member of the set */
5364 if (stat(dev
, &stb
) < 0)
5366 if ((S_IFMT
& stb
.st_mode
) != S_IFBLK
)
5368 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
5369 if (dl
->major
== (int)major(stb
.st_rdev
) &&
5370 dl
->minor
== (int)minor(stb
.st_rdev
))
5375 fprintf(stderr
, Name
": %s is not in the "
5376 "same imsm set\n", dev
);
5378 } else if (super
->orom
&& dl
->index
< 0 && mpb
->num_raid_devs
) {
5379 /* If a volume is present then the current creation attempt
5380 * cannot incorporate new spares because the orom may not
5381 * understand this configuration (all member disks must be
5382 * members of each array in the container).
5384 fprintf(stderr
, Name
": %s is a spare and a volume"
5385 " is already defined for this container\n", dev
);
5386 fprintf(stderr
, Name
": The option-rom requires all member"
5387 " disks to be a member of all volumes\n");
5391 /* retrieve the largest free space block */
5392 e
= get_extents(super
, dl
);
5397 unsigned long long esize
;
5399 esize
= e
[i
].start
- pos
;
5400 if (esize
>= maxsize
)
5402 pos
= e
[i
].start
+ e
[i
].size
;
5404 } while (e
[i
-1].size
);
5409 fprintf(stderr
, Name
": unable to determine free space for: %s\n",
5413 if (maxsize
< size
) {
5415 fprintf(stderr
, Name
": %s not enough space (%llu < %llu)\n",
5416 dev
, maxsize
, size
);
5420 /* count total number of extents for merge */
5422 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
5424 i
+= dl
->extent_cnt
;
5426 maxsize
= merge_extents(super
, i
);
5428 if (!check_env("IMSM_NO_PLATFORM") &&
5429 mpb
->num_raid_devs
> 0 && size
&& size
!= maxsize
) {
5430 fprintf(stderr
, Name
": attempting to create a second "
5431 "volume with size less then remaining space. "
5436 if (maxsize
< size
|| maxsize
== 0) {
5438 fprintf(stderr
, Name
": not enough space after merge (%llu < %llu)\n",
5443 *freesize
= maxsize
;
5448 static int reserve_space(struct supertype
*st
, int raiddisks
,
5449 unsigned long long size
, int chunk
,
5450 unsigned long long *freesize
)
5452 struct intel_super
*super
= st
->sb
;
5453 struct imsm_super
*mpb
= super
->anchor
;
5458 unsigned long long maxsize
;
5459 unsigned long long minsize
;
5463 /* find the largest common start free region of the possible disks */
5467 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
5473 /* don't activate new spares if we are orom constrained
5474 * and there is already a volume active in the container
5476 if (super
->orom
&& dl
->index
< 0 && mpb
->num_raid_devs
)
5479 e
= get_extents(super
, dl
);
5482 for (i
= 1; e
[i
-1].size
; i
++)
5490 maxsize
= merge_extents(super
, extent_cnt
);
5494 minsize
= chunk
* 2;
5496 if (cnt
< raiddisks
||
5497 (super
->orom
&& used
&& used
!= raiddisks
) ||
5498 maxsize
< minsize
||
5500 fprintf(stderr
, Name
": not enough devices with space to create array.\n");
5501 return 0; /* No enough free spaces large enough */
5513 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
5515 dl
->raiddisk
= cnt
++;
5522 static int validate_geometry_imsm(struct supertype
*st
, int level
, int layout
,
5523 int raiddisks
, int *chunk
, unsigned long long size
,
5524 char *dev
, unsigned long long *freesize
,
5532 * if given unused devices create a container
5533 * if given given devices in a container create a member volume
5535 if (level
== LEVEL_CONTAINER
) {
5536 /* Must be a fresh device to add to a container */
5537 return validate_geometry_imsm_container(st
, level
, layout
,
5539 chunk
?*chunk
:0, size
,
5546 if (!validate_geometry_imsm_orom(st
->sb
, level
, layout
,
5550 /* we are being asked to automatically layout a
5551 * new volume based on the current contents of
5552 * the container. If the the parameters can be
5553 * satisfied reserve_space will record the disks,
5554 * start offset, and size of the volume to be
5555 * created. add_to_super and getinfo_super
5556 * detect when autolayout is in progress.
5559 return reserve_space(st
, raiddisks
, size
,
5560 chunk
?*chunk
:0, freesize
);
5565 /* creating in a given container */
5566 return validate_geometry_imsm_volume(st
, level
, layout
,
5567 raiddisks
, chunk
, size
,
5568 dev
, freesize
, verbose
);
5571 /* This device needs to be a device in an 'imsm' container */
5572 fd
= open(dev
, O_RDONLY
|O_EXCL
, 0);
5576 Name
": Cannot create this array on device %s\n",
5581 if (errno
!= EBUSY
|| (fd
= open(dev
, O_RDONLY
, 0)) < 0) {
5583 fprintf(stderr
, Name
": Cannot open %s: %s\n",
5584 dev
, strerror(errno
));
5587 /* Well, it is in use by someone, maybe an 'imsm' container. */
5588 cfd
= open_container(fd
);
5592 fprintf(stderr
, Name
": Cannot use %s: It is busy\n",
5596 sra
= sysfs_read(cfd
, 0, GET_VERSION
);
5597 if (sra
&& sra
->array
.major_version
== -1 &&
5598 strcmp(sra
->text_version
, "imsm") == 0)
5602 /* This is a member of a imsm container. Load the container
5603 * and try to create a volume
5605 struct intel_super
*super
;
5607 if (load_super_imsm_all(st
, cfd
, (void **) &super
, NULL
) == 0) {
5609 st
->container_dev
= fd2devnum(cfd
);
5611 return validate_geometry_imsm_volume(st
, level
, layout
,
5620 fprintf(stderr
, Name
": failed container membership check\n");
5626 static void default_geometry_imsm(struct supertype
*st
, int *level
, int *layout
, int *chunk
)
5628 struct intel_super
*super
= st
->sb
;
5630 if (level
&& *level
== UnSet
)
5631 *level
= LEVEL_CONTAINER
;
5633 if (level
&& layout
&& *layout
== UnSet
)
5634 *layout
= imsm_level_to_layout(*level
);
5636 if (chunk
&& (*chunk
== UnSet
|| *chunk
== 0))
5637 *chunk
= imsm_default_chunk(super
->orom
);
5640 static void handle_missing(struct intel_super
*super
, struct imsm_dev
*dev
);
5642 static int kill_subarray_imsm(struct supertype
*st
)
5644 /* remove the subarray currently referenced by ->current_vol */
5646 struct intel_dev
**dp
;
5647 struct intel_super
*super
= st
->sb
;
5648 __u8 current_vol
= super
->current_vol
;
5649 struct imsm_super
*mpb
= super
->anchor
;
5651 if (super
->current_vol
< 0)
5653 super
->current_vol
= -1; /* invalidate subarray cursor */
5655 /* block deletions that would change the uuid of active subarrays
5657 * FIXME when immutable ids are available, but note that we'll
5658 * also need to fixup the invalidated/active subarray indexes in
5661 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
5664 if (i
< current_vol
)
5666 sprintf(subarray
, "%u", i
);
5667 if (is_subarray_active(subarray
, st
->devname
)) {
5669 Name
": deleting subarray-%d would change the UUID of active subarray-%d, aborting\n",
5676 if (st
->update_tail
) {
5677 struct imsm_update_kill_array
*u
= malloc(sizeof(*u
));
5681 u
->type
= update_kill_array
;
5682 u
->dev_idx
= current_vol
;
5683 append_metadata_update(st
, u
, sizeof(*u
));
5688 for (dp
= &super
->devlist
; *dp
;)
5689 if ((*dp
)->index
== current_vol
) {
5692 handle_missing(super
, (*dp
)->dev
);
5693 if ((*dp
)->index
> current_vol
)
5698 /* no more raid devices, all active components are now spares,
5699 * but of course failed are still failed
5701 if (--mpb
->num_raid_devs
== 0) {
5704 for (d
= super
->disks
; d
; d
= d
->next
)
5709 super
->updates_pending
++;
5714 static int update_subarray_imsm(struct supertype
*st
, char *subarray
,
5715 char *update
, struct mddev_ident
*ident
)
5717 /* update the subarray currently referenced by ->current_vol */
5718 struct intel_super
*super
= st
->sb
;
5719 struct imsm_super
*mpb
= super
->anchor
;
5721 if (strcmp(update
, "name") == 0) {
5722 char *name
= ident
->name
;
5726 if (is_subarray_active(subarray
, st
->devname
)) {
5728 Name
": Unable to update name of active subarray\n");
5732 if (!check_name(super
, name
, 0))
5735 vol
= strtoul(subarray
, &ep
, 10);
5736 if (*ep
!= '\0' || vol
>= super
->anchor
->num_raid_devs
)
5739 if (st
->update_tail
) {
5740 struct imsm_update_rename_array
*u
= malloc(sizeof(*u
));
5744 u
->type
= update_rename_array
;
5746 snprintf((char *) u
->name
, MAX_RAID_SERIAL_LEN
, "%s", name
);
5747 append_metadata_update(st
, u
, sizeof(*u
));
5749 struct imsm_dev
*dev
;
5752 dev
= get_imsm_dev(super
, vol
);
5753 snprintf((char *) dev
->volume
, MAX_RAID_SERIAL_LEN
, "%s", name
);
5754 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
5755 dev
= get_imsm_dev(super
, i
);
5756 handle_missing(super
, dev
);
5758 super
->updates_pending
++;
5765 #endif /* MDASSEMBLE */
5767 static int is_gen_migration(struct imsm_dev
*dev
)
5772 if (!dev
->vol
.migr_state
)
5775 if (migr_type(dev
) == MIGR_GEN_MIGR
)
5781 static int is_rebuilding(struct imsm_dev
*dev
)
5783 struct imsm_map
*migr_map
;
5785 if (!dev
->vol
.migr_state
)
5788 if (migr_type(dev
) != MIGR_REBUILD
)
5791 migr_map
= get_imsm_map(dev
, MAP_1
);
5793 if (migr_map
->map_state
== IMSM_T_STATE_DEGRADED
)
5799 static int is_initializing(struct imsm_dev
*dev
)
5801 struct imsm_map
*migr_map
;
5803 if (!dev
->vol
.migr_state
)
5806 if (migr_type(dev
) != MIGR_INIT
)
5809 migr_map
= get_imsm_map(dev
, MAP_1
);
5811 if (migr_map
->map_state
== IMSM_T_STATE_UNINITIALIZED
)
5818 static void update_recovery_start(struct intel_super
*super
,
5819 struct imsm_dev
*dev
,
5820 struct mdinfo
*array
)
5822 struct mdinfo
*rebuild
= NULL
;
5826 if (!is_rebuilding(dev
))
5829 /* Find the rebuild target, but punt on the dual rebuild case */
5830 for (d
= array
->devs
; d
; d
= d
->next
)
5831 if (d
->recovery_start
== 0) {
5838 /* (?) none of the disks are marked with
5839 * IMSM_ORD_REBUILD, so assume they are missing and the
5840 * disk_ord_tbl was not correctly updated
5842 dprintf("%s: failed to locate out-of-sync disk\n", __func__
);
5846 units
= __le32_to_cpu(dev
->vol
.curr_migr_unit
);
5847 rebuild
->recovery_start
= units
* blocks_per_migr_unit(super
, dev
);
5851 static int recover_backup_imsm(struct supertype
*st
, struct mdinfo
*info
);
5854 static struct mdinfo
*container_content_imsm(struct supertype
*st
, char *subarray
)
5856 /* Given a container loaded by load_super_imsm_all,
5857 * extract information about all the arrays into
5859 * If 'subarray' is given, just extract info about that array.
5861 * For each imsm_dev create an mdinfo, fill it in,
5862 * then look for matching devices in super->disks
5863 * and create appropriate device mdinfo.
5865 struct intel_super
*super
= st
->sb
;
5866 struct imsm_super
*mpb
= super
->anchor
;
5867 struct mdinfo
*rest
= NULL
;
5871 int spare_disks
= 0;
5873 /* do not assemble arrays when not all attributes are supported */
5874 if (imsm_check_attributes(mpb
->attributes
) == 0) {
5876 fprintf(stderr
, Name
": Unsupported attributes in IMSM metadata."
5877 "Arrays activation is blocked.\n");
5880 /* check for bad blocks */
5881 if (imsm_bbm_log_size(super
->anchor
)) {
5882 fprintf(stderr
, Name
": BBM log found in IMSM metadata."
5883 "Arrays activation is blocked.\n");
5888 /* count spare devices, not used in maps
5890 for (d
= super
->disks
; d
; d
= d
->next
)
5894 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
5895 struct imsm_dev
*dev
;
5896 struct imsm_map
*map
;
5897 struct imsm_map
*map2
;
5898 struct mdinfo
*this;
5903 (i
!= strtoul(subarray
, &ep
, 10) || *ep
!= '\0'))
5906 dev
= get_imsm_dev(super
, i
);
5907 map
= get_imsm_map(dev
, MAP_0
);
5908 map2
= get_imsm_map(dev
, MAP_1
);
5910 /* do not publish arrays that are in the middle of an
5911 * unsupported migration
5913 if (dev
->vol
.migr_state
&&
5914 (migr_type(dev
) == MIGR_STATE_CHANGE
)) {
5915 fprintf(stderr
, Name
": cannot assemble volume '%.16s':"
5916 " unsupported migration in progress\n",
5920 /* do not publish arrays that are not support by controller's
5924 chunk
= __le16_to_cpu(map
->blocks_per_strip
) >> 1;
5925 this = malloc(sizeof(*this));
5927 fprintf(stderr
, Name
": failed to allocate %zu bytes\n",
5932 super
->current_vol
= i
;
5933 getinfo_super_imsm_volume(st
, this, NULL
);
5936 /* mdadm does not support all metadata features- set the bit in all arrays state */
5937 if (!validate_geometry_imsm_orom(super
,
5938 get_imsm_raid_level(map
), /* RAID level */
5939 imsm_level_to_layout(get_imsm_raid_level(map
)),
5940 map
->num_members
, /* raid disks */
5943 fprintf(stderr
, Name
": IMSM RAID geometry validation"
5944 " failed. Array %s activation is blocked.\n",
5946 this->array
.state
|=
5947 (1<<MD_SB_BLOCK_CONTAINER_RESHAPE
) |
5948 (1<<MD_SB_BLOCK_VOLUME
);
5952 /* if array has bad blocks, set suitable bit in all arrays state */
5954 this->array
.state
|=
5955 (1<<MD_SB_BLOCK_CONTAINER_RESHAPE
) |
5956 (1<<MD_SB_BLOCK_VOLUME
);
5958 for (slot
= 0 ; slot
< map
->num_members
; slot
++) {
5959 unsigned long long recovery_start
;
5960 struct mdinfo
*info_d
;
5967 idx
= get_imsm_disk_idx(dev
, slot
, MAP_0
);
5968 ord
= get_imsm_ord_tbl_ent(dev
, slot
, MAP_X
);
5969 for (d
= super
->disks
; d
; d
= d
->next
)
5970 if (d
->index
== idx
)
5973 recovery_start
= MaxSector
;
5976 if (d
&& is_failed(&d
->disk
))
5978 if (ord
& IMSM_ORD_REBUILD
)
5982 * if we skip some disks the array will be assmebled degraded;
5983 * reset resync start to avoid a dirty-degraded
5984 * situation when performing the intial sync
5986 * FIXME handle dirty degraded
5988 if ((skip
|| recovery_start
== 0) && !dev
->vol
.dirty
)
5989 this->resync_start
= MaxSector
;
5993 info_d
= calloc(1, sizeof(*info_d
));
5995 fprintf(stderr
, Name
": failed to allocate disk"
5996 " for volume %.16s\n", dev
->volume
);
5997 info_d
= this->devs
;
5999 struct mdinfo
*d
= info_d
->next
;
6008 info_d
->next
= this->devs
;
6009 this->devs
= info_d
;
6011 info_d
->disk
.number
= d
->index
;
6012 info_d
->disk
.major
= d
->major
;
6013 info_d
->disk
.minor
= d
->minor
;
6014 info_d
->disk
.raid_disk
= slot
;
6015 info_d
->recovery_start
= recovery_start
;
6017 if (slot
< map2
->num_members
)
6018 info_d
->disk
.state
= (1 << MD_DISK_ACTIVE
);
6020 this->array
.spare_disks
++;
6022 if (slot
< map
->num_members
)
6023 info_d
->disk
.state
= (1 << MD_DISK_ACTIVE
);
6025 this->array
.spare_disks
++;
6027 if (info_d
->recovery_start
== MaxSector
)
6028 this->array
.working_disks
++;
6030 info_d
->events
= __le32_to_cpu(mpb
->generation_num
);
6031 info_d
->data_offset
= __le32_to_cpu(map
->pba_of_lba0
);
6032 info_d
->component_size
= __le32_to_cpu(map
->blocks_per_member
);
6034 /* now that the disk list is up-to-date fixup recovery_start */
6035 update_recovery_start(super
, dev
, this);
6036 this->array
.spare_disks
+= spare_disks
;
6039 /* check for reshape */
6040 if (this->reshape_active
== 1)
6041 recover_backup_imsm(st
, this);
6050 static __u8
imsm_check_degraded(struct intel_super
*super
, struct imsm_dev
*dev
,
6051 int failed
, int look_in_map
)
6053 struct imsm_map
*map
;
6055 map
= get_imsm_map(dev
, look_in_map
);
6058 return map
->map_state
== IMSM_T_STATE_UNINITIALIZED
?
6059 IMSM_T_STATE_UNINITIALIZED
: IMSM_T_STATE_NORMAL
;
6061 switch (get_imsm_raid_level(map
)) {
6063 return IMSM_T_STATE_FAILED
;
6066 if (failed
< map
->num_members
)
6067 return IMSM_T_STATE_DEGRADED
;
6069 return IMSM_T_STATE_FAILED
;
6074 * check to see if any mirrors have failed, otherwise we
6075 * are degraded. Even numbered slots are mirrored on
6079 /* gcc -Os complains that this is unused */
6080 int insync
= insync
;
6082 for (i
= 0; i
< map
->num_members
; i
++) {
6083 __u32 ord
= get_imsm_ord_tbl_ent(dev
, i
, MAP_X
);
6084 int idx
= ord_to_idx(ord
);
6085 struct imsm_disk
*disk
;
6087 /* reset the potential in-sync count on even-numbered
6088 * slots. num_copies is always 2 for imsm raid10
6093 disk
= get_imsm_disk(super
, idx
);
6094 if (!disk
|| is_failed(disk
) || ord
& IMSM_ORD_REBUILD
)
6097 /* no in-sync disks left in this mirror the
6101 return IMSM_T_STATE_FAILED
;
6104 return IMSM_T_STATE_DEGRADED
;
6108 return IMSM_T_STATE_DEGRADED
;
6110 return IMSM_T_STATE_FAILED
;
6116 return map
->map_state
;
6119 static int imsm_count_failed(struct intel_super
*super
, struct imsm_dev
*dev
,
6124 struct imsm_disk
*disk
;
6125 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
6126 struct imsm_map
*prev
= get_imsm_map(dev
, MAP_1
);
6127 struct imsm_map
*map_for_loop
;
6132 /* at the beginning of migration we set IMSM_ORD_REBUILD on
6133 * disks that are being rebuilt. New failures are recorded to
6134 * map[0]. So we look through all the disks we started with and
6135 * see if any failures are still present, or if any new ones
6139 if (prev
&& (map
->num_members
< prev
->num_members
))
6140 map_for_loop
= prev
;
6142 for (i
= 0; i
< map_for_loop
->num_members
; i
++) {
6144 /* when MAP_X is passed both maps failures are counted
6147 ((look_in_map
== MAP_1
) || (look_in_map
== MAP_X
)) &&
6148 (i
< prev
->num_members
)) {
6149 ord
= __le32_to_cpu(prev
->disk_ord_tbl
[i
]);
6150 idx_1
= ord_to_idx(ord
);
6152 disk
= get_imsm_disk(super
, idx_1
);
6153 if (!disk
|| is_failed(disk
) || ord
& IMSM_ORD_REBUILD
)
6156 if (((look_in_map
== MAP_0
) || (look_in_map
== MAP_X
)) &&
6157 (i
< map
->num_members
)) {
6158 ord
= __le32_to_cpu(map
->disk_ord_tbl
[i
]);
6159 idx
= ord_to_idx(ord
);
6162 disk
= get_imsm_disk(super
, idx
);
6163 if (!disk
|| is_failed(disk
) ||
6164 ord
& IMSM_ORD_REBUILD
)
6174 static int imsm_open_new(struct supertype
*c
, struct active_array
*a
,
6177 struct intel_super
*super
= c
->sb
;
6178 struct imsm_super
*mpb
= super
->anchor
;
6180 if (atoi(inst
) >= mpb
->num_raid_devs
) {
6181 fprintf(stderr
, "%s: subarry index %d, out of range\n",
6182 __func__
, atoi(inst
));
6186 dprintf("imsm: open_new %s\n", inst
);
6187 a
->info
.container_member
= atoi(inst
);
6191 static int is_resyncing(struct imsm_dev
*dev
)
6193 struct imsm_map
*migr_map
;
6195 if (!dev
->vol
.migr_state
)
6198 if (migr_type(dev
) == MIGR_INIT
||
6199 migr_type(dev
) == MIGR_REPAIR
)
6202 if (migr_type(dev
) == MIGR_GEN_MIGR
)
6205 migr_map
= get_imsm_map(dev
, MAP_1
);
6207 if ((migr_map
->map_state
== IMSM_T_STATE_NORMAL
) &&
6208 (dev
->vol
.migr_type
!= MIGR_GEN_MIGR
))
6214 /* return true if we recorded new information */
6215 static int mark_failure(struct imsm_dev
*dev
, struct imsm_disk
*disk
, int idx
)
6219 struct imsm_map
*map
;
6220 char buf
[MAX_RAID_SERIAL_LEN
+3];
6221 unsigned int len
, shift
= 0;
6223 /* new failures are always set in map[0] */
6224 map
= get_imsm_map(dev
, MAP_0
);
6226 slot
= get_imsm_disk_slot(map
, idx
);
6230 ord
= __le32_to_cpu(map
->disk_ord_tbl
[slot
]);
6231 if (is_failed(disk
) && (ord
& IMSM_ORD_REBUILD
))
6234 memcpy(buf
, disk
->serial
, MAX_RAID_SERIAL_LEN
);
6235 buf
[MAX_RAID_SERIAL_LEN
] = '\000';
6237 if ((len
= strlen(buf
)) >= MAX_RAID_SERIAL_LEN
)
6238 shift
= len
- MAX_RAID_SERIAL_LEN
+ 1;
6239 strncpy((char *)disk
->serial
, &buf
[shift
], MAX_RAID_SERIAL_LEN
);
6241 disk
->status
|= FAILED_DISK
;
6242 set_imsm_ord_tbl_ent(map
, slot
, idx
| IMSM_ORD_REBUILD
);
6243 /* mark failures in second map if second map exists and this disk
6245 * This is valid for migration, initialization and rebuild
6247 if (dev
->vol
.migr_state
) {
6248 struct imsm_map
*map2
= get_imsm_map(dev
, MAP_1
);
6249 int slot2
= get_imsm_disk_slot(map2
, idx
);
6251 if ((slot2
< map2
->num_members
) &&
6253 set_imsm_ord_tbl_ent(map2
, slot2
,
6254 idx
| IMSM_ORD_REBUILD
);
6256 if (map
->failed_disk_num
== 0xff)
6257 map
->failed_disk_num
= slot
;
6261 static void mark_missing(struct imsm_dev
*dev
, struct imsm_disk
*disk
, int idx
)
6263 mark_failure(dev
, disk
, idx
);
6265 if (disk
->scsi_id
== __cpu_to_le32(~(__u32
)0))
6268 disk
->scsi_id
= __cpu_to_le32(~(__u32
)0);
6269 memmove(&disk
->serial
[0], &disk
->serial
[1], MAX_RAID_SERIAL_LEN
- 1);
6272 static void handle_missing(struct intel_super
*super
, struct imsm_dev
*dev
)
6276 if (!super
->missing
)
6279 dprintf("imsm: mark missing\n");
6280 /* end process for initialization and rebuild only
6282 if (is_gen_migration(dev
) == 0) {
6286 failed
= imsm_count_failed(super
, dev
, MAP_0
);
6287 map_state
= imsm_check_degraded(super
, dev
, failed
, MAP_0
);
6289 end_migration(dev
, super
, map_state
);
6291 for (dl
= super
->missing
; dl
; dl
= dl
->next
)
6292 mark_missing(dev
, &dl
->disk
, dl
->index
);
6293 super
->updates_pending
++;
6296 static unsigned long long imsm_set_array_size(struct imsm_dev
*dev
)
6298 int used_disks
= imsm_num_data_members(dev
, MAP_0
);
6299 unsigned long long array_blocks
;
6300 struct imsm_map
*map
;
6302 if (used_disks
== 0) {
6303 /* when problems occures
6304 * return current array_blocks value
6306 array_blocks
= __le32_to_cpu(dev
->size_high
);
6307 array_blocks
= array_blocks
<< 32;
6308 array_blocks
+= __le32_to_cpu(dev
->size_low
);
6310 return array_blocks
;
6313 /* set array size in metadata
6315 map
= get_imsm_map(dev
, MAP_0
);
6316 array_blocks
= map
->blocks_per_member
* used_disks
;
6318 /* round array size down to closest MB
6320 array_blocks
= (array_blocks
>> SECT_PER_MB_SHIFT
) << SECT_PER_MB_SHIFT
;
6321 dev
->size_low
= __cpu_to_le32((__u32
)array_blocks
);
6322 dev
->size_high
= __cpu_to_le32((__u32
)(array_blocks
>> 32));
6324 return array_blocks
;
6327 static void imsm_set_disk(struct active_array
*a
, int n
, int state
);
6329 static void imsm_progress_container_reshape(struct intel_super
*super
)
6331 /* if no device has a migr_state, but some device has a
6332 * different number of members than the previous device, start
6333 * changing the number of devices in this device to match
6336 struct imsm_super
*mpb
= super
->anchor
;
6337 int prev_disks
= -1;
6341 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
6342 struct imsm_dev
*dev
= get_imsm_dev(super
, i
);
6343 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
6344 struct imsm_map
*map2
;
6345 int prev_num_members
;
6347 if (dev
->vol
.migr_state
)
6350 if (prev_disks
== -1)
6351 prev_disks
= map
->num_members
;
6352 if (prev_disks
== map
->num_members
)
6355 /* OK, this array needs to enter reshape mode.
6356 * i.e it needs a migr_state
6359 copy_map_size
= sizeof_imsm_map(map
);
6360 prev_num_members
= map
->num_members
;
6361 map
->num_members
= prev_disks
;
6362 dev
->vol
.migr_state
= 1;
6363 dev
->vol
.curr_migr_unit
= 0;
6364 set_migr_type(dev
, MIGR_GEN_MIGR
);
6365 for (i
= prev_num_members
;
6366 i
< map
->num_members
; i
++)
6367 set_imsm_ord_tbl_ent(map
, i
, i
);
6368 map2
= get_imsm_map(dev
, MAP_1
);
6369 /* Copy the current map */
6370 memcpy(map2
, map
, copy_map_size
);
6371 map2
->num_members
= prev_num_members
;
6373 imsm_set_array_size(dev
);
6374 super
->updates_pending
++;
6378 /* Handle dirty -> clean transititions, resync and reshape. Degraded and rebuild
6379 * states are handled in imsm_set_disk() with one exception, when a
6380 * resync is stopped due to a new failure this routine will set the
6381 * 'degraded' state for the array.
6383 static int imsm_set_array_state(struct active_array
*a
, int consistent
)
6385 int inst
= a
->info
.container_member
;
6386 struct intel_super
*super
= a
->container
->sb
;
6387 struct imsm_dev
*dev
= get_imsm_dev(super
, inst
);
6388 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
6389 int failed
= imsm_count_failed(super
, dev
, MAP_0
);
6390 __u8 map_state
= imsm_check_degraded(super
, dev
, failed
, MAP_0
);
6391 __u32 blocks_per_unit
;
6393 if (dev
->vol
.migr_state
&&
6394 dev
->vol
.migr_type
== MIGR_GEN_MIGR
) {
6395 /* array state change is blocked due to reshape action
6397 * - abort the reshape (if last_checkpoint is 0 and action!= reshape)
6398 * - finish the reshape (if last_checkpoint is big and action != reshape)
6399 * - update curr_migr_unit
6401 if (a
->curr_action
== reshape
) {
6402 /* still reshaping, maybe update curr_migr_unit */
6403 goto mark_checkpoint
;
6405 if (a
->last_checkpoint
== 0 && a
->prev_action
== reshape
) {
6406 /* for some reason we aborted the reshape.
6408 * disable automatic metadata rollback
6409 * user action is required to recover process
6412 struct imsm_map
*map2
=
6413 get_imsm_map(dev
, MAP_1
);
6414 dev
->vol
.migr_state
= 0;
6415 set_migr_type(dev
, 0);
6416 dev
->vol
.curr_migr_unit
= 0;
6418 sizeof_imsm_map(map2
));
6419 super
->updates_pending
++;
6422 if (a
->last_checkpoint
>= a
->info
.component_size
) {
6423 unsigned long long array_blocks
;
6427 used_disks
= imsm_num_data_members(dev
, MAP_0
);
6428 if (used_disks
> 0) {
6430 map
->blocks_per_member
*
6432 /* round array size down to closest MB
6434 array_blocks
= (array_blocks
6435 >> SECT_PER_MB_SHIFT
)
6436 << SECT_PER_MB_SHIFT
;
6437 a
->info
.custom_array_size
= array_blocks
;
6438 /* encourage manager to update array
6442 a
->check_reshape
= 1;
6444 /* finalize online capacity expansion/reshape */
6445 for (mdi
= a
->info
.devs
; mdi
; mdi
= mdi
->next
)
6447 mdi
->disk
.raid_disk
,
6450 imsm_progress_container_reshape(super
);
6455 /* before we activate this array handle any missing disks */
6456 if (consistent
== 2)
6457 handle_missing(super
, dev
);
6459 if (consistent
== 2 &&
6460 (!is_resync_complete(&a
->info
) ||
6461 map_state
!= IMSM_T_STATE_NORMAL
||
6462 dev
->vol
.migr_state
))
6465 if (is_resync_complete(&a
->info
)) {
6466 /* complete intialization / resync,
6467 * recovery and interrupted recovery is completed in
6470 if (is_resyncing(dev
)) {
6471 dprintf("imsm: mark resync done\n");
6472 end_migration(dev
, super
, map_state
);
6473 super
->updates_pending
++;
6474 a
->last_checkpoint
= 0;
6476 } else if ((!is_resyncing(dev
) && !failed
) &&
6477 (imsm_reshape_blocks_arrays_changes(super
) == 0)) {
6478 /* mark the start of the init process if nothing is failed */
6479 dprintf("imsm: mark resync start\n");
6480 if (map
->map_state
== IMSM_T_STATE_UNINITIALIZED
)
6481 migrate(dev
, super
, IMSM_T_STATE_NORMAL
, MIGR_INIT
);
6483 migrate(dev
, super
, IMSM_T_STATE_NORMAL
, MIGR_REPAIR
);
6484 super
->updates_pending
++;
6488 /* skip checkpointing for general migration,
6489 * it is controlled in mdadm
6491 if (is_gen_migration(dev
))
6492 goto skip_mark_checkpoint
;
6494 /* check if we can update curr_migr_unit from resync_start, recovery_start */
6495 blocks_per_unit
= blocks_per_migr_unit(super
, dev
);
6496 if (blocks_per_unit
) {
6500 units
= a
->last_checkpoint
/ blocks_per_unit
;
6503 /* check that we did not overflow 32-bits, and that
6504 * curr_migr_unit needs updating
6506 if (units32
== units
&&
6508 __le32_to_cpu(dev
->vol
.curr_migr_unit
) != units32
) {
6509 dprintf("imsm: mark checkpoint (%u)\n", units32
);
6510 dev
->vol
.curr_migr_unit
= __cpu_to_le32(units32
);
6511 super
->updates_pending
++;
6515 skip_mark_checkpoint
:
6516 /* mark dirty / clean */
6517 if (dev
->vol
.dirty
!= !consistent
) {
6518 dprintf("imsm: mark '%s'\n", consistent
? "clean" : "dirty");
6523 super
->updates_pending
++;
6529 static void imsm_set_disk(struct active_array
*a
, int n
, int state
)
6531 int inst
= a
->info
.container_member
;
6532 struct intel_super
*super
= a
->container
->sb
;
6533 struct imsm_dev
*dev
= get_imsm_dev(super
, inst
);
6534 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
6535 struct imsm_disk
*disk
;
6540 if (n
> map
->num_members
)
6541 fprintf(stderr
, "imsm: set_disk %d out of range 0..%d\n",
6542 n
, map
->num_members
- 1);
6547 dprintf("imsm: set_disk %d:%x\n", n
, state
);
6549 ord
= get_imsm_ord_tbl_ent(dev
, n
, MAP_0
);
6550 disk
= get_imsm_disk(super
, ord_to_idx(ord
));
6552 /* check for new failures */
6553 if (state
& DS_FAULTY
) {
6554 if (mark_failure(dev
, disk
, ord_to_idx(ord
)))
6555 super
->updates_pending
++;
6558 /* check if in_sync */
6559 if (state
& DS_INSYNC
&& ord
& IMSM_ORD_REBUILD
&& is_rebuilding(dev
)) {
6560 struct imsm_map
*migr_map
= get_imsm_map(dev
, MAP_1
);
6562 set_imsm_ord_tbl_ent(migr_map
, n
, ord_to_idx(ord
));
6563 super
->updates_pending
++;
6566 failed
= imsm_count_failed(super
, dev
, MAP_0
);
6567 map_state
= imsm_check_degraded(super
, dev
, failed
, MAP_0
);
6569 /* check if recovery complete, newly degraded, or failed */
6570 dprintf("imsm: Detected transition to state ");
6571 switch (map_state
) {
6572 case IMSM_T_STATE_NORMAL
: /* transition to normal state */
6573 dprintf("normal: ");
6574 if (is_rebuilding(dev
)) {
6575 dprintf("while rebuilding");
6576 end_migration(dev
, super
, map_state
);
6577 map
= get_imsm_map(dev
, MAP_0
);
6578 map
->failed_disk_num
= ~0;
6579 super
->updates_pending
++;
6580 a
->last_checkpoint
= 0;
6583 if (is_gen_migration(dev
)) {
6584 dprintf("while general migration");
6585 if (a
->last_checkpoint
>= a
->info
.component_size
)
6586 end_migration(dev
, super
, map_state
);
6588 map
->map_state
= map_state
;
6589 map
= get_imsm_map(dev
, MAP_0
);
6590 map
->failed_disk_num
= ~0;
6591 super
->updates_pending
++;
6595 case IMSM_T_STATE_DEGRADED
: /* transition to degraded state */
6596 dprintf("degraded: ");
6597 if ((map
->map_state
!= map_state
) &&
6598 !dev
->vol
.migr_state
) {
6599 dprintf("mark degraded");
6600 map
->map_state
= map_state
;
6601 super
->updates_pending
++;
6602 a
->last_checkpoint
= 0;
6605 if (is_rebuilding(dev
)) {
6606 dprintf("while rebuilding.");
6607 if (map
->map_state
!= map_state
) {
6608 dprintf(" Map state change");
6609 end_migration(dev
, super
, map_state
);
6610 super
->updates_pending
++;
6614 if (is_gen_migration(dev
)) {
6615 dprintf("while general migration");
6616 if (a
->last_checkpoint
>= a
->info
.component_size
)
6617 end_migration(dev
, super
, map_state
);
6619 map
->map_state
= map_state
;
6620 manage_second_map(super
, dev
);
6622 super
->updates_pending
++;
6625 if (is_initializing(dev
)) {
6626 dprintf("while initialization.");
6627 map
->map_state
= map_state
;
6628 super
->updates_pending
++;
6632 case IMSM_T_STATE_FAILED
: /* transition to failed state */
6633 dprintf("failed: ");
6634 if (is_gen_migration(dev
)) {
6635 dprintf("while general migration");
6636 map
->map_state
= map_state
;
6637 super
->updates_pending
++;
6640 if (map
->map_state
!= map_state
) {
6641 dprintf("mark failed");
6642 end_migration(dev
, super
, map_state
);
6643 super
->updates_pending
++;
6644 a
->last_checkpoint
= 0;
6649 dprintf("state %i\n", map_state
);
6655 static int store_imsm_mpb(int fd
, struct imsm_super
*mpb
)
6658 __u32 mpb_size
= __le32_to_cpu(mpb
->mpb_size
);
6659 unsigned long long dsize
;
6660 unsigned long long sectors
;
6662 get_dev_size(fd
, NULL
, &dsize
);
6664 if (mpb_size
> 512) {
6665 /* -1 to account for anchor */
6666 sectors
= mpb_sectors(mpb
) - 1;
6668 /* write the extended mpb to the sectors preceeding the anchor */
6669 if (lseek64(fd
, dsize
- (512 * (2 + sectors
)), SEEK_SET
) < 0)
6672 if ((unsigned long long)write(fd
, buf
+ 512, 512 * sectors
)
6677 /* first block is stored on second to last sector of the disk */
6678 if (lseek64(fd
, dsize
- (512 * 2), SEEK_SET
) < 0)
6681 if (write(fd
, buf
, 512) != 512)
6687 static void imsm_sync_metadata(struct supertype
*container
)
6689 struct intel_super
*super
= container
->sb
;
6691 dprintf("sync metadata: %d\n", super
->updates_pending
);
6692 if (!super
->updates_pending
)
6695 write_super_imsm(container
, 0);
6697 super
->updates_pending
= 0;
6700 static struct dl
*imsm_readd(struct intel_super
*super
, int idx
, struct active_array
*a
)
6702 struct imsm_dev
*dev
= get_imsm_dev(super
, a
->info
.container_member
);
6703 int i
= get_imsm_disk_idx(dev
, idx
, MAP_X
);
6706 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
6710 if (dl
&& is_failed(&dl
->disk
))
6714 dprintf("%s: found %x:%x\n", __func__
, dl
->major
, dl
->minor
);
6719 static struct dl
*imsm_add_spare(struct intel_super
*super
, int slot
,
6720 struct active_array
*a
, int activate_new
,
6721 struct mdinfo
*additional_test_list
)
6723 struct imsm_dev
*dev
= get_imsm_dev(super
, a
->info
.container_member
);
6724 int idx
= get_imsm_disk_idx(dev
, slot
, MAP_X
);
6725 struct imsm_super
*mpb
= super
->anchor
;
6726 struct imsm_map
*map
;
6727 unsigned long long pos
;
6732 __u32 array_start
= 0;
6733 __u32 array_end
= 0;
6735 struct mdinfo
*test_list
;
6737 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
6738 /* If in this array, skip */
6739 for (d
= a
->info
.devs
; d
; d
= d
->next
)
6740 if (d
->state_fd
>= 0 &&
6741 d
->disk
.major
== dl
->major
&&
6742 d
->disk
.minor
== dl
->minor
) {
6743 dprintf("%x:%x already in array\n",
6744 dl
->major
, dl
->minor
);
6749 test_list
= additional_test_list
;
6751 if (test_list
->disk
.major
== dl
->major
&&
6752 test_list
->disk
.minor
== dl
->minor
) {
6753 dprintf("%x:%x already in additional test list\n",
6754 dl
->major
, dl
->minor
);
6757 test_list
= test_list
->next
;
6762 /* skip in use or failed drives */
6763 if (is_failed(&dl
->disk
) || idx
== dl
->index
||
6765 dprintf("%x:%x status (failed: %d index: %d)\n",
6766 dl
->major
, dl
->minor
, is_failed(&dl
->disk
), idx
);
6770 /* skip pure spares when we are looking for partially
6771 * assimilated drives
6773 if (dl
->index
== -1 && !activate_new
)
6776 /* Does this unused device have the requisite free space?
6777 * It needs to be able to cover all member volumes
6779 ex
= get_extents(super
, dl
);
6781 dprintf("cannot get extents\n");
6784 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
6785 dev
= get_imsm_dev(super
, i
);
6786 map
= get_imsm_map(dev
, MAP_0
);
6788 /* check if this disk is already a member of
6791 if (get_imsm_disk_slot(map
, dl
->index
) >= 0)
6797 array_start
= __le32_to_cpu(map
->pba_of_lba0
);
6798 array_end
= array_start
+
6799 __le32_to_cpu(map
->blocks_per_member
) - 1;
6802 /* check that we can start at pba_of_lba0 with
6803 * blocks_per_member of space
6805 if (array_start
>= pos
&& array_end
< ex
[j
].start
) {
6809 pos
= ex
[j
].start
+ ex
[j
].size
;
6811 } while (ex
[j
-1].size
);
6818 if (i
< mpb
->num_raid_devs
) {
6819 dprintf("%x:%x does not have %u to %u available\n",
6820 dl
->major
, dl
->minor
, array_start
, array_end
);
6831 static int imsm_rebuild_allowed(struct supertype
*cont
, int dev_idx
, int failed
)
6833 struct imsm_dev
*dev2
;
6834 struct imsm_map
*map
;
6840 dev2
= get_imsm_dev(cont
->sb
, dev_idx
);
6842 state
= imsm_check_degraded(cont
->sb
, dev2
, failed
, MAP_0
);
6843 if (state
== IMSM_T_STATE_FAILED
) {
6844 map
= get_imsm_map(dev2
, MAP_0
);
6847 for (slot
= 0; slot
< map
->num_members
; slot
++) {
6849 * Check if failed disks are deleted from intel
6850 * disk list or are marked to be deleted
6852 idx
= get_imsm_disk_idx(dev2
, slot
, MAP_X
);
6853 idisk
= get_imsm_dl_disk(cont
->sb
, idx
);
6855 * Do not rebuild the array if failed disks
6856 * from failed sub-array are not removed from
6860 is_failed(&idisk
->disk
) &&
6861 (idisk
->action
!= DISK_REMOVE
))
6869 static struct mdinfo
*imsm_activate_spare(struct active_array
*a
,
6870 struct metadata_update
**updates
)
6873 * Find a device with unused free space and use it to replace a
6874 * failed/vacant region in an array. We replace failed regions one a
6875 * array at a time. The result is that a new spare disk will be added
6876 * to the first failed array and after the monitor has finished
6877 * propagating failures the remainder will be consumed.
6879 * FIXME add a capability for mdmon to request spares from another
6883 struct intel_super
*super
= a
->container
->sb
;
6884 int inst
= a
->info
.container_member
;
6885 struct imsm_dev
*dev
= get_imsm_dev(super
, inst
);
6886 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
6887 int failed
= a
->info
.array
.raid_disks
;
6888 struct mdinfo
*rv
= NULL
;
6891 struct metadata_update
*mu
;
6893 struct imsm_update_activate_spare
*u
;
6898 for (d
= a
->info
.devs
; d
; d
= d
->next
) {
6899 if ((d
->curr_state
& DS_FAULTY
) &&
6901 /* wait for Removal to happen */
6903 if (d
->state_fd
>= 0)
6907 dprintf("imsm: activate spare: inst=%d failed=%d (%d) level=%d\n",
6908 inst
, failed
, a
->info
.array
.raid_disks
, a
->info
.array
.level
);
6910 if (imsm_reshape_blocks_arrays_changes(super
))
6913 /* Cannot activate another spare if rebuild is in progress already
6915 if (is_rebuilding(dev
)) {
6916 dprintf("imsm: No spare activation allowed. "
6917 "Rebuild in progress already.\n");
6921 if (a
->info
.array
.level
== 4)
6922 /* No repair for takeovered array
6923 * imsm doesn't support raid4
6927 if (imsm_check_degraded(super
, dev
, failed
, MAP_0
) !=
6928 IMSM_T_STATE_DEGRADED
)
6932 * If there are any failed disks check state of the other volume.
6933 * Block rebuild if the another one is failed until failed disks
6934 * are removed from container.
6937 dprintf("found failed disks in %.*s, check if there another"
6938 "failed sub-array.\n",
6939 MAX_RAID_SERIAL_LEN
, dev
->volume
);
6940 /* check if states of the other volumes allow for rebuild */
6941 for (i
= 0; i
< super
->anchor
->num_raid_devs
; i
++) {
6943 allowed
= imsm_rebuild_allowed(a
->container
,
6951 /* For each slot, if it is not working, find a spare */
6952 for (i
= 0; i
< a
->info
.array
.raid_disks
; i
++) {
6953 for (d
= a
->info
.devs
; d
; d
= d
->next
)
6954 if (d
->disk
.raid_disk
== i
)
6956 dprintf("found %d: %p %x\n", i
, d
, d
?d
->curr_state
:0);
6957 if (d
&& (d
->state_fd
>= 0))
6961 * OK, this device needs recovery. Try to re-add the
6962 * previous occupant of this slot, if this fails see if
6963 * we can continue the assimilation of a spare that was
6964 * partially assimilated, finally try to activate a new
6967 dl
= imsm_readd(super
, i
, a
);
6969 dl
= imsm_add_spare(super
, i
, a
, 0, rv
);
6971 dl
= imsm_add_spare(super
, i
, a
, 1, rv
);
6975 /* found a usable disk with enough space */
6976 di
= malloc(sizeof(*di
));
6979 memset(di
, 0, sizeof(*di
));
6981 /* dl->index will be -1 in the case we are activating a
6982 * pristine spare. imsm_process_update() will create a
6983 * new index in this case. Once a disk is found to be
6984 * failed in all member arrays it is kicked from the
6987 di
->disk
.number
= dl
->index
;
6989 /* (ab)use di->devs to store a pointer to the device
6992 di
->devs
= (struct mdinfo
*) dl
;
6994 di
->disk
.raid_disk
= i
;
6995 di
->disk
.major
= dl
->major
;
6996 di
->disk
.minor
= dl
->minor
;
6998 di
->recovery_start
= 0;
6999 di
->data_offset
= __le32_to_cpu(map
->pba_of_lba0
);
7000 di
->component_size
= a
->info
.component_size
;
7001 di
->container_member
= inst
;
7002 super
->random
= random32();
7006 dprintf("%x:%x to be %d at %llu\n", dl
->major
, dl
->minor
,
7007 i
, di
->data_offset
);
7011 /* No spares found */
7013 /* Now 'rv' has a list of devices to return.
7014 * Create a metadata_update record to update the
7015 * disk_ord_tbl for the array
7017 mu
= malloc(sizeof(*mu
));
7019 mu
->buf
= malloc(sizeof(struct imsm_update_activate_spare
) * num_spares
);
7020 if (mu
->buf
== NULL
) {
7027 struct mdinfo
*n
= rv
->next
;
7036 mu
->space_list
= NULL
;
7037 mu
->len
= sizeof(struct imsm_update_activate_spare
) * num_spares
;
7038 mu
->next
= *updates
;
7039 u
= (struct imsm_update_activate_spare
*) mu
->buf
;
7041 for (di
= rv
; di
; di
= di
->next
) {
7042 u
->type
= update_activate_spare
;
7043 u
->dl
= (struct dl
*) di
->devs
;
7045 u
->slot
= di
->disk
.raid_disk
;
7056 static int disks_overlap(struct intel_super
*super
, int idx
, struct imsm_update_create_array
*u
)
7058 struct imsm_dev
*dev
= get_imsm_dev(super
, idx
);
7059 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
7060 struct imsm_map
*new_map
= get_imsm_map(&u
->dev
, MAP_0
);
7061 struct disk_info
*inf
= get_disk_info(u
);
7062 struct imsm_disk
*disk
;
7066 for (i
= 0; i
< map
->num_members
; i
++) {
7067 disk
= get_imsm_disk(super
, get_imsm_disk_idx(dev
, i
, MAP_X
));
7068 for (j
= 0; j
< new_map
->num_members
; j
++)
7069 if (serialcmp(disk
->serial
, inf
[j
].serial
) == 0)
7077 static struct dl
*get_disk_super(struct intel_super
*super
, int major
, int minor
)
7079 struct dl
*dl
= NULL
;
7080 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
7081 if ((dl
->major
== major
) && (dl
->minor
== minor
))
7086 static int remove_disk_super(struct intel_super
*super
, int major
, int minor
)
7088 struct dl
*prev
= NULL
;
7092 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
7093 if ((dl
->major
== major
) && (dl
->minor
== minor
)) {
7096 prev
->next
= dl
->next
;
7098 super
->disks
= dl
->next
;
7100 __free_imsm_disk(dl
);
7101 dprintf("%s: removed %x:%x\n",
7102 __func__
, major
, minor
);
7110 static void imsm_delete(struct intel_super
*super
, struct dl
**dlp
, unsigned index
);
7112 static int add_remove_disk_update(struct intel_super
*super
)
7114 int check_degraded
= 0;
7115 struct dl
*disk
= NULL
;
7116 /* add/remove some spares to/from the metadata/contrainer */
7117 while (super
->disk_mgmt_list
) {
7118 struct dl
*disk_cfg
;
7120 disk_cfg
= super
->disk_mgmt_list
;
7121 super
->disk_mgmt_list
= disk_cfg
->next
;
7122 disk_cfg
->next
= NULL
;
7124 if (disk_cfg
->action
== DISK_ADD
) {
7125 disk_cfg
->next
= super
->disks
;
7126 super
->disks
= disk_cfg
;
7128 dprintf("%s: added %x:%x\n",
7129 __func__
, disk_cfg
->major
,
7131 } else if (disk_cfg
->action
== DISK_REMOVE
) {
7132 dprintf("Disk remove action processed: %x.%x\n",
7133 disk_cfg
->major
, disk_cfg
->minor
);
7134 disk
= get_disk_super(super
,
7138 /* store action status */
7139 disk
->action
= DISK_REMOVE
;
7140 /* remove spare disks only */
7141 if (disk
->index
== -1) {
7142 remove_disk_super(super
,
7147 /* release allocate disk structure */
7148 __free_imsm_disk(disk_cfg
);
7151 return check_degraded
;
7155 static int apply_reshape_migration_update(struct imsm_update_reshape_migration
*u
,
7156 struct intel_super
*super
,
7159 struct intel_dev
*id
;
7160 void **tofree
= NULL
;
7163 dprintf("apply_reshape_migration_update()\n");
7164 if ((u
->subdev
< 0) ||
7166 dprintf("imsm: Error: Wrong subdev: %i\n", u
->subdev
);
7169 if ((space_list
== NULL
) || (*space_list
== NULL
)) {
7170 dprintf("imsm: Error: Memory is not allocated\n");
7174 for (id
= super
->devlist
; id
; id
= id
->next
) {
7175 if (id
->index
== (unsigned)u
->subdev
) {
7176 struct imsm_dev
*dev
= get_imsm_dev(super
, u
->subdev
);
7177 struct imsm_map
*map
;
7178 struct imsm_dev
*new_dev
=
7179 (struct imsm_dev
*)*space_list
;
7180 struct imsm_map
*migr_map
= get_imsm_map(dev
, MAP_1
);
7182 struct dl
*new_disk
;
7184 if (new_dev
== NULL
)
7186 *space_list
= **space_list
;
7187 memcpy(new_dev
, dev
, sizeof_imsm_dev(dev
, 0));
7188 map
= get_imsm_map(new_dev
, MAP_0
);
7190 dprintf("imsm: Error: migration in progress");
7194 to_state
= map
->map_state
;
7195 if ((u
->new_level
== 5) && (map
->raid_level
== 0)) {
7197 /* this should not happen */
7198 if (u
->new_disks
[0] < 0) {
7199 map
->failed_disk_num
=
7200 map
->num_members
- 1;
7201 to_state
= IMSM_T_STATE_DEGRADED
;
7203 to_state
= IMSM_T_STATE_NORMAL
;
7205 migrate(new_dev
, super
, to_state
, MIGR_GEN_MIGR
);
7206 if (u
->new_level
> -1)
7207 map
->raid_level
= u
->new_level
;
7208 migr_map
= get_imsm_map(new_dev
, MAP_1
);
7209 if ((u
->new_level
== 5) &&
7210 (migr_map
->raid_level
== 0)) {
7211 int ord
= map
->num_members
- 1;
7212 migr_map
->num_members
--;
7213 if (u
->new_disks
[0] < 0)
7214 ord
|= IMSM_ORD_REBUILD
;
7215 set_imsm_ord_tbl_ent(map
,
7216 map
->num_members
- 1,
7220 tofree
= (void **)dev
;
7222 /* update chunk size
7224 if (u
->new_chunksize
> 0)
7225 map
->blocks_per_strip
=
7226 __cpu_to_le16(u
->new_chunksize
* 2);
7230 if ((u
->new_level
!= 5) ||
7231 (migr_map
->raid_level
!= 0) ||
7232 (migr_map
->raid_level
== map
->raid_level
))
7235 if (u
->new_disks
[0] >= 0) {
7238 new_disk
= get_disk_super(super
,
7239 major(u
->new_disks
[0]),
7240 minor(u
->new_disks
[0]));
7241 dprintf("imsm: new disk for reshape is: %i:%i "
7242 "(%p, index = %i)\n",
7243 major(u
->new_disks
[0]),
7244 minor(u
->new_disks
[0]),
7245 new_disk
, new_disk
->index
);
7246 if (new_disk
== NULL
)
7247 goto error_disk_add
;
7249 new_disk
->index
= map
->num_members
- 1;
7250 /* slot to fill in autolayout
7252 new_disk
->raiddisk
= new_disk
->index
;
7253 new_disk
->disk
.status
|= CONFIGURED_DISK
;
7254 new_disk
->disk
.status
&= ~SPARE_DISK
;
7256 goto error_disk_add
;
7259 *tofree
= *space_list
;
7260 /* calculate new size
7262 imsm_set_array_size(new_dev
);
7269 *space_list
= tofree
;
7273 dprintf("Error: imsm: Cannot find disk.\n");
7277 static int apply_update_activate_spare(struct imsm_update_activate_spare
*u
,
7278 struct intel_super
*super
,
7279 struct active_array
*active_array
)
7281 struct imsm_super
*mpb
= super
->anchor
;
7282 struct imsm_dev
*dev
= get_imsm_dev(super
, u
->array
);
7283 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
7284 struct imsm_map
*migr_map
;
7285 struct active_array
*a
;
7286 struct imsm_disk
*disk
;
7293 int second_map_created
= 0;
7295 for (; u
; u
= u
->next
) {
7296 victim
= get_imsm_disk_idx(dev
, u
->slot
, MAP_X
);
7301 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
7306 fprintf(stderr
, "error: imsm_activate_spare passed "
7307 "an unknown disk (index: %d)\n",
7312 /* count failures (excluding rebuilds and the victim)
7313 * to determine map[0] state
7316 for (i
= 0; i
< map
->num_members
; i
++) {
7319 disk
= get_imsm_disk(super
,
7320 get_imsm_disk_idx(dev
, i
, MAP_X
));
7321 if (!disk
|| is_failed(disk
))
7325 /* adding a pristine spare, assign a new index */
7326 if (dl
->index
< 0) {
7327 dl
->index
= super
->anchor
->num_disks
;
7328 super
->anchor
->num_disks
++;
7331 disk
->status
|= CONFIGURED_DISK
;
7332 disk
->status
&= ~SPARE_DISK
;
7335 to_state
= imsm_check_degraded(super
, dev
, failed
, MAP_0
);
7336 if (!second_map_created
) {
7337 second_map_created
= 1;
7338 map
->map_state
= IMSM_T_STATE_DEGRADED
;
7339 migrate(dev
, super
, to_state
, MIGR_REBUILD
);
7341 map
->map_state
= to_state
;
7342 migr_map
= get_imsm_map(dev
, MAP_1
);
7343 set_imsm_ord_tbl_ent(map
, u
->slot
, dl
->index
);
7344 set_imsm_ord_tbl_ent(migr_map
, u
->slot
,
7345 dl
->index
| IMSM_ORD_REBUILD
);
7347 /* update the family_num to mark a new container
7348 * generation, being careful to record the existing
7349 * family_num in orig_family_num to clean up after
7350 * earlier mdadm versions that neglected to set it.
7352 if (mpb
->orig_family_num
== 0)
7353 mpb
->orig_family_num
= mpb
->family_num
;
7354 mpb
->family_num
+= super
->random
;
7356 /* count arrays using the victim in the metadata */
7358 for (a
= active_array
; a
; a
= a
->next
) {
7359 dev
= get_imsm_dev(super
, a
->info
.container_member
);
7360 map
= get_imsm_map(dev
, MAP_0
);
7362 if (get_imsm_disk_slot(map
, victim
) >= 0)
7366 /* delete the victim if it is no longer being
7372 /* We know that 'manager' isn't touching anything,
7373 * so it is safe to delete
7375 for (dlp
= &super
->disks
; *dlp
; dlp
= &(*dlp
)->next
)
7376 if ((*dlp
)->index
== victim
)
7379 /* victim may be on the missing list */
7381 for (dlp
= &super
->missing
; *dlp
;
7382 dlp
= &(*dlp
)->next
)
7383 if ((*dlp
)->index
== victim
)
7385 imsm_delete(super
, dlp
, victim
);
7392 static int apply_reshape_container_disks_update(struct imsm_update_reshape
*u
,
7393 struct intel_super
*super
,
7396 struct dl
*new_disk
;
7397 struct intel_dev
*id
;
7399 int delta_disks
= u
->new_raid_disks
- u
->old_raid_disks
;
7400 int disk_count
= u
->old_raid_disks
;
7401 void **tofree
= NULL
;
7402 int devices_to_reshape
= 1;
7403 struct imsm_super
*mpb
= super
->anchor
;
7405 unsigned int dev_id
;
7407 dprintf("imsm: apply_reshape_container_disks_update()\n");
7409 /* enable spares to use in array */
7410 for (i
= 0; i
< delta_disks
; i
++) {
7411 new_disk
= get_disk_super(super
,
7412 major(u
->new_disks
[i
]),
7413 minor(u
->new_disks
[i
]));
7414 dprintf("imsm: new disk for reshape is: %i:%i "
7415 "(%p, index = %i)\n",
7416 major(u
->new_disks
[i
]), minor(u
->new_disks
[i
]),
7417 new_disk
, new_disk
->index
);
7418 if ((new_disk
== NULL
) ||
7419 ((new_disk
->index
>= 0) &&
7420 (new_disk
->index
< u
->old_raid_disks
)))
7421 goto update_reshape_exit
;
7422 new_disk
->index
= disk_count
++;
7423 /* slot to fill in autolayout
7425 new_disk
->raiddisk
= new_disk
->index
;
7426 new_disk
->disk
.status
|=
7428 new_disk
->disk
.status
&= ~SPARE_DISK
;
7431 dprintf("imsm: volume set mpb->num_raid_devs = %i\n",
7432 mpb
->num_raid_devs
);
7433 /* manage changes in volume
7435 for (dev_id
= 0; dev_id
< mpb
->num_raid_devs
; dev_id
++) {
7436 void **sp
= *space_list
;
7437 struct imsm_dev
*newdev
;
7438 struct imsm_map
*newmap
, *oldmap
;
7440 for (id
= super
->devlist
; id
; id
= id
->next
) {
7441 if (id
->index
== dev_id
)
7450 /* Copy the dev, but not (all of) the map */
7451 memcpy(newdev
, id
->dev
, sizeof(*newdev
));
7452 oldmap
= get_imsm_map(id
->dev
, MAP_0
);
7453 newmap
= get_imsm_map(newdev
, MAP_0
);
7454 /* Copy the current map */
7455 memcpy(newmap
, oldmap
, sizeof_imsm_map(oldmap
));
7456 /* update one device only
7458 if (devices_to_reshape
) {
7459 dprintf("imsm: modifying subdev: %i\n",
7461 devices_to_reshape
--;
7462 newdev
->vol
.migr_state
= 1;
7463 newdev
->vol
.curr_migr_unit
= 0;
7464 set_migr_type(newdev
, MIGR_GEN_MIGR
);
7465 newmap
->num_members
= u
->new_raid_disks
;
7466 for (i
= 0; i
< delta_disks
; i
++) {
7467 set_imsm_ord_tbl_ent(newmap
,
7468 u
->old_raid_disks
+ i
,
7469 u
->old_raid_disks
+ i
);
7471 /* New map is correct, now need to save old map
7473 newmap
= get_imsm_map(newdev
, MAP_1
);
7474 memcpy(newmap
, oldmap
, sizeof_imsm_map(oldmap
));
7476 imsm_set_array_size(newdev
);
7479 sp
= (void **)id
->dev
;
7484 /* Clear migration record */
7485 memset(super
->migr_rec
, 0, sizeof(struct migr_record
));
7488 *space_list
= tofree
;
7491 update_reshape_exit
:
7496 static int apply_takeover_update(struct imsm_update_takeover
*u
,
7497 struct intel_super
*super
,
7500 struct imsm_dev
*dev
= NULL
;
7501 struct intel_dev
*dv
;
7502 struct imsm_dev
*dev_new
;
7503 struct imsm_map
*map
;
7507 for (dv
= super
->devlist
; dv
; dv
= dv
->next
)
7508 if (dv
->index
== (unsigned int)u
->subarray
) {
7516 map
= get_imsm_map(dev
, MAP_0
);
7518 if (u
->direction
== R10_TO_R0
) {
7519 /* Number of failed disks must be half of initial disk number */
7520 if (imsm_count_failed(super
, dev
, MAP_0
) !=
7521 (map
->num_members
/ 2))
7524 /* iterate through devices to mark removed disks as spare */
7525 for (dm
= super
->disks
; dm
; dm
= dm
->next
) {
7526 if (dm
->disk
.status
& FAILED_DISK
) {
7527 int idx
= dm
->index
;
7528 /* update indexes on the disk list */
7529 /* FIXME this loop-with-the-loop looks wrong, I'm not convinced
7530 the index values will end up being correct.... NB */
7531 for (du
= super
->disks
; du
; du
= du
->next
)
7532 if (du
->index
> idx
)
7534 /* mark as spare disk */
7539 map
->num_members
= map
->num_members
/ 2;
7540 map
->map_state
= IMSM_T_STATE_NORMAL
;
7541 map
->num_domains
= 1;
7542 map
->raid_level
= 0;
7543 map
->failed_disk_num
= -1;
7546 if (u
->direction
== R0_TO_R10
) {
7548 /* update slots in current disk list */
7549 for (dm
= super
->disks
; dm
; dm
= dm
->next
) {
7553 /* create new *missing* disks */
7554 for (i
= 0; i
< map
->num_members
; i
++) {
7555 space
= *space_list
;
7558 *space_list
= *space
;
7560 memcpy(du
, super
->disks
, sizeof(*du
));
7564 du
->index
= (i
* 2) + 1;
7565 sprintf((char *)du
->disk
.serial
,
7566 " MISSING_%d", du
->index
);
7567 sprintf((char *)du
->serial
,
7568 "MISSING_%d", du
->index
);
7569 du
->next
= super
->missing
;
7570 super
->missing
= du
;
7572 /* create new dev and map */
7573 space
= *space_list
;
7576 *space_list
= *space
;
7577 dev_new
= (void *)space
;
7578 memcpy(dev_new
, dev
, sizeof(*dev
));
7579 /* update new map */
7580 map
= get_imsm_map(dev_new
, MAP_0
);
7581 map
->num_members
= map
->num_members
* 2;
7582 map
->map_state
= IMSM_T_STATE_DEGRADED
;
7583 map
->num_domains
= 2;
7584 map
->raid_level
= 1;
7585 /* replace dev<->dev_new */
7588 /* update disk order table */
7589 for (du
= super
->disks
; du
; du
= du
->next
)
7591 set_imsm_ord_tbl_ent(map
, du
->index
, du
->index
);
7592 for (du
= super
->missing
; du
; du
= du
->next
)
7593 if (du
->index
>= 0) {
7594 set_imsm_ord_tbl_ent(map
, du
->index
, du
->index
);
7595 mark_missing(dv
->dev
, &du
->disk
, du
->index
);
7601 static void imsm_process_update(struct supertype
*st
,
7602 struct metadata_update
*update
)
7605 * crack open the metadata_update envelope to find the update record
7606 * update can be one of:
7607 * update_reshape_container_disks - all the arrays in the container
7608 * are being reshaped to have more devices. We need to mark
7609 * the arrays for general migration and convert selected spares
7610 * into active devices.
7611 * update_activate_spare - a spare device has replaced a failed
7612 * device in an array, update the disk_ord_tbl. If this disk is
7613 * present in all member arrays then also clear the SPARE_DISK
7615 * update_create_array
7617 * update_rename_array
7618 * update_add_remove_disk
7620 struct intel_super
*super
= st
->sb
;
7621 struct imsm_super
*mpb
;
7622 enum imsm_update_type type
= *(enum imsm_update_type
*) update
->buf
;
7624 /* update requires a larger buf but the allocation failed */
7625 if (super
->next_len
&& !super
->next_buf
) {
7626 super
->next_len
= 0;
7630 if (super
->next_buf
) {
7631 memcpy(super
->next_buf
, super
->buf
, super
->len
);
7633 super
->len
= super
->next_len
;
7634 super
->buf
= super
->next_buf
;
7636 super
->next_len
= 0;
7637 super
->next_buf
= NULL
;
7640 mpb
= super
->anchor
;
7643 case update_general_migration_checkpoint
: {
7644 struct intel_dev
*id
;
7645 struct imsm_update_general_migration_checkpoint
*u
=
7646 (void *)update
->buf
;
7648 dprintf("imsm: process_update() "
7649 "for update_general_migration_checkpoint called\n");
7651 /* find device under general migration */
7652 for (id
= super
->devlist
; id
; id
= id
->next
) {
7653 if (is_gen_migration(id
->dev
)) {
7654 id
->dev
->vol
.curr_migr_unit
=
7655 __cpu_to_le32(u
->curr_migr_unit
);
7656 super
->updates_pending
++;
7661 case update_takeover
: {
7662 struct imsm_update_takeover
*u
= (void *)update
->buf
;
7663 if (apply_takeover_update(u
, super
, &update
->space_list
)) {
7664 imsm_update_version_info(super
);
7665 super
->updates_pending
++;
7670 case update_reshape_container_disks
: {
7671 struct imsm_update_reshape
*u
= (void *)update
->buf
;
7672 if (apply_reshape_container_disks_update(
7673 u
, super
, &update
->space_list
))
7674 super
->updates_pending
++;
7677 case update_reshape_migration
: {
7678 struct imsm_update_reshape_migration
*u
= (void *)update
->buf
;
7679 if (apply_reshape_migration_update(
7680 u
, super
, &update
->space_list
))
7681 super
->updates_pending
++;
7684 case update_activate_spare
: {
7685 struct imsm_update_activate_spare
*u
= (void *) update
->buf
;
7686 if (apply_update_activate_spare(u
, super
, st
->arrays
))
7687 super
->updates_pending
++;
7690 case update_create_array
: {
7691 /* someone wants to create a new array, we need to be aware of
7692 * a few races/collisions:
7693 * 1/ 'Create' called by two separate instances of mdadm
7694 * 2/ 'Create' versus 'activate_spare': mdadm has chosen
7695 * devices that have since been assimilated via
7697 * In the event this update can not be carried out mdadm will
7698 * (FIX ME) notice that its update did not take hold.
7700 struct imsm_update_create_array
*u
= (void *) update
->buf
;
7701 struct intel_dev
*dv
;
7702 struct imsm_dev
*dev
;
7703 struct imsm_map
*map
, *new_map
;
7704 unsigned long long start
, end
;
7705 unsigned long long new_start
, new_end
;
7707 struct disk_info
*inf
;
7710 /* handle racing creates: first come first serve */
7711 if (u
->dev_idx
< mpb
->num_raid_devs
) {
7712 dprintf("%s: subarray %d already defined\n",
7713 __func__
, u
->dev_idx
);
7717 /* check update is next in sequence */
7718 if (u
->dev_idx
!= mpb
->num_raid_devs
) {
7719 dprintf("%s: can not create array %d expected index %d\n",
7720 __func__
, u
->dev_idx
, mpb
->num_raid_devs
);
7724 new_map
= get_imsm_map(&u
->dev
, MAP_0
);
7725 new_start
= __le32_to_cpu(new_map
->pba_of_lba0
);
7726 new_end
= new_start
+ __le32_to_cpu(new_map
->blocks_per_member
);
7727 inf
= get_disk_info(u
);
7729 /* handle activate_spare versus create race:
7730 * check to make sure that overlapping arrays do not include
7733 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
7734 dev
= get_imsm_dev(super
, i
);
7735 map
= get_imsm_map(dev
, MAP_0
);
7736 start
= __le32_to_cpu(map
->pba_of_lba0
);
7737 end
= start
+ __le32_to_cpu(map
->blocks_per_member
);
7738 if ((new_start
>= start
&& new_start
<= end
) ||
7739 (start
>= new_start
&& start
<= new_end
))
7744 if (disks_overlap(super
, i
, u
)) {
7745 dprintf("%s: arrays overlap\n", __func__
);
7750 /* check that prepare update was successful */
7751 if (!update
->space
) {
7752 dprintf("%s: prepare update failed\n", __func__
);
7756 /* check that all disks are still active before committing
7757 * changes. FIXME: could we instead handle this by creating a
7758 * degraded array? That's probably not what the user expects,
7759 * so better to drop this update on the floor.
7761 for (i
= 0; i
< new_map
->num_members
; i
++) {
7762 dl
= serial_to_dl(inf
[i
].serial
, super
);
7764 dprintf("%s: disk disappeared\n", __func__
);
7769 super
->updates_pending
++;
7771 /* convert spares to members and fixup ord_tbl */
7772 for (i
= 0; i
< new_map
->num_members
; i
++) {
7773 dl
= serial_to_dl(inf
[i
].serial
, super
);
7774 if (dl
->index
== -1) {
7775 dl
->index
= mpb
->num_disks
;
7777 dl
->disk
.status
|= CONFIGURED_DISK
;
7778 dl
->disk
.status
&= ~SPARE_DISK
;
7780 set_imsm_ord_tbl_ent(new_map
, i
, dl
->index
);
7785 update
->space
= NULL
;
7786 imsm_copy_dev(dev
, &u
->dev
);
7787 dv
->index
= u
->dev_idx
;
7788 dv
->next
= super
->devlist
;
7789 super
->devlist
= dv
;
7790 mpb
->num_raid_devs
++;
7792 imsm_update_version_info(super
);
7795 /* mdmon knows how to release update->space, but not
7796 * ((struct intel_dev *) update->space)->dev
7798 if (update
->space
) {
7804 case update_kill_array
: {
7805 struct imsm_update_kill_array
*u
= (void *) update
->buf
;
7806 int victim
= u
->dev_idx
;
7807 struct active_array
*a
;
7808 struct intel_dev
**dp
;
7809 struct imsm_dev
*dev
;
7811 /* sanity check that we are not affecting the uuid of
7812 * active arrays, or deleting an active array
7814 * FIXME when immutable ids are available, but note that
7815 * we'll also need to fixup the invalidated/active
7816 * subarray indexes in mdstat
7818 for (a
= st
->arrays
; a
; a
= a
->next
)
7819 if (a
->info
.container_member
>= victim
)
7821 /* by definition if mdmon is running at least one array
7822 * is active in the container, so checking
7823 * mpb->num_raid_devs is just extra paranoia
7825 dev
= get_imsm_dev(super
, victim
);
7826 if (a
|| !dev
|| mpb
->num_raid_devs
== 1) {
7827 dprintf("failed to delete subarray-%d\n", victim
);
7831 for (dp
= &super
->devlist
; *dp
;)
7832 if ((*dp
)->index
== (unsigned)super
->current_vol
) {
7835 if ((*dp
)->index
> (unsigned)victim
)
7839 mpb
->num_raid_devs
--;
7840 super
->updates_pending
++;
7843 case update_rename_array
: {
7844 struct imsm_update_rename_array
*u
= (void *) update
->buf
;
7845 char name
[MAX_RAID_SERIAL_LEN
+1];
7846 int target
= u
->dev_idx
;
7847 struct active_array
*a
;
7848 struct imsm_dev
*dev
;
7850 /* sanity check that we are not affecting the uuid of
7853 snprintf(name
, MAX_RAID_SERIAL_LEN
, "%s", (char *) u
->name
);
7854 name
[MAX_RAID_SERIAL_LEN
] = '\0';
7855 for (a
= st
->arrays
; a
; a
= a
->next
)
7856 if (a
->info
.container_member
== target
)
7858 dev
= get_imsm_dev(super
, u
->dev_idx
);
7859 if (a
|| !dev
|| !check_name(super
, name
, 1)) {
7860 dprintf("failed to rename subarray-%d\n", target
);
7864 snprintf((char *) dev
->volume
, MAX_RAID_SERIAL_LEN
, "%s", name
);
7865 super
->updates_pending
++;
7868 case update_add_remove_disk
: {
7869 /* we may be able to repair some arrays if disks are
7870 * being added, check teh status of add_remove_disk
7871 * if discs has been added.
7873 if (add_remove_disk_update(super
)) {
7874 struct active_array
*a
;
7876 super
->updates_pending
++;
7877 for (a
= st
->arrays
; a
; a
= a
->next
)
7878 a
->check_degraded
= 1;
7883 fprintf(stderr
, "error: unsuported process update type:"
7884 "(type: %d)\n", type
);
7888 static struct mdinfo
*get_spares_for_grow(struct supertype
*st
);
7890 static void imsm_prepare_update(struct supertype
*st
,
7891 struct metadata_update
*update
)
7894 * Allocate space to hold new disk entries, raid-device entries or a new
7895 * mpb if necessary. The manager synchronously waits for updates to
7896 * complete in the monitor, so new mpb buffers allocated here can be
7897 * integrated by the monitor thread without worrying about live pointers
7898 * in the manager thread.
7900 enum imsm_update_type type
= *(enum imsm_update_type
*) update
->buf
;
7901 struct intel_super
*super
= st
->sb
;
7902 struct imsm_super
*mpb
= super
->anchor
;
7907 case update_general_migration_checkpoint
:
7908 dprintf("imsm: prepare_update() "
7909 "for update_general_migration_checkpoint called\n");
7911 case update_takeover
: {
7912 struct imsm_update_takeover
*u
= (void *)update
->buf
;
7913 if (u
->direction
== R0_TO_R10
) {
7914 void **tail
= (void **)&update
->space_list
;
7915 struct imsm_dev
*dev
= get_imsm_dev(super
, u
->subarray
);
7916 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
7917 int num_members
= map
->num_members
;
7921 /* allocate memory for added disks */
7922 for (i
= 0; i
< num_members
; i
++) {
7923 size
= sizeof(struct dl
);
7924 space
= malloc(size
);
7933 /* allocate memory for new device */
7934 size
= sizeof_imsm_dev(super
->devlist
->dev
, 0) +
7935 (num_members
* sizeof(__u32
));
7936 space
= malloc(size
);
7945 len
= disks_to_mpb_size(num_members
* 2);
7947 /* if allocation didn't success, free buffer */
7948 while (update
->space_list
) {
7949 void **sp
= update
->space_list
;
7950 update
->space_list
= *sp
;
7958 case update_reshape_container_disks
: {
7959 /* Every raid device in the container is about to
7960 * gain some more devices, and we will enter a
7962 * So each 'imsm_map' will be bigger, and the imsm_vol
7963 * will now hold 2 of them.
7964 * Thus we need new 'struct imsm_dev' allocations sized
7965 * as sizeof_imsm_dev but with more devices in both maps.
7967 struct imsm_update_reshape
*u
= (void *)update
->buf
;
7968 struct intel_dev
*dl
;
7969 void **space_tail
= (void**)&update
->space_list
;
7971 dprintf("imsm: imsm_prepare_update() for update_reshape\n");
7973 for (dl
= super
->devlist
; dl
; dl
= dl
->next
) {
7974 int size
= sizeof_imsm_dev(dl
->dev
, 1);
7976 if (u
->new_raid_disks
> u
->old_raid_disks
)
7977 size
+= sizeof(__u32
)*2*
7978 (u
->new_raid_disks
- u
->old_raid_disks
);
7987 len
= disks_to_mpb_size(u
->new_raid_disks
);
7988 dprintf("New anchor length is %llu\n", (unsigned long long)len
);
7991 case update_reshape_migration
: {
7992 /* for migration level 0->5 we need to add disks
7993 * so the same as for container operation we will copy
7994 * device to the bigger location.
7995 * in memory prepared device and new disk area are prepared
7996 * for usage in process update
7998 struct imsm_update_reshape_migration
*u
= (void *)update
->buf
;
7999 struct intel_dev
*id
;
8000 void **space_tail
= (void **)&update
->space_list
;
8003 int current_level
= -1;
8005 dprintf("imsm: imsm_prepare_update() for update_reshape\n");
8007 /* add space for bigger array in update
8009 for (id
= super
->devlist
; id
; id
= id
->next
) {
8010 if (id
->index
== (unsigned)u
->subdev
) {
8011 size
= sizeof_imsm_dev(id
->dev
, 1);
8012 if (u
->new_raid_disks
> u
->old_raid_disks
)
8013 size
+= sizeof(__u32
)*2*
8014 (u
->new_raid_disks
- u
->old_raid_disks
);
8024 if (update
->space_list
== NULL
)
8027 /* add space for disk in update
8029 size
= sizeof(struct dl
);
8032 free(update
->space_list
);
8033 update
->space_list
= NULL
;
8040 /* add spare device to update
8042 for (id
= super
->devlist
; id
; id
= id
->next
)
8043 if (id
->index
== (unsigned)u
->subdev
) {
8044 struct imsm_dev
*dev
;
8045 struct imsm_map
*map
;
8047 dev
= get_imsm_dev(super
, u
->subdev
);
8048 map
= get_imsm_map(dev
, MAP_0
);
8049 current_level
= map
->raid_level
;
8052 if ((u
->new_level
== 5) && (u
->new_level
!= current_level
)) {
8053 struct mdinfo
*spares
;
8055 spares
= get_spares_for_grow(st
);
8063 makedev(dev
->disk
.major
,
8065 dl
= get_disk_super(super
,
8068 dl
->index
= u
->old_raid_disks
;
8074 len
= disks_to_mpb_size(u
->new_raid_disks
);
8075 dprintf("New anchor length is %llu\n", (unsigned long long)len
);
8078 case update_create_array
: {
8079 struct imsm_update_create_array
*u
= (void *) update
->buf
;
8080 struct intel_dev
*dv
;
8081 struct imsm_dev
*dev
= &u
->dev
;
8082 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
8084 struct disk_info
*inf
;
8088 inf
= get_disk_info(u
);
8089 len
= sizeof_imsm_dev(dev
, 1);
8090 /* allocate a new super->devlist entry */
8091 dv
= malloc(sizeof(*dv
));
8093 dv
->dev
= malloc(len
);
8098 update
->space
= NULL
;
8102 /* count how many spares will be converted to members */
8103 for (i
= 0; i
< map
->num_members
; i
++) {
8104 dl
= serial_to_dl(inf
[i
].serial
, super
);
8106 /* hmm maybe it failed?, nothing we can do about
8111 if (count_memberships(dl
, super
) == 0)
8114 len
+= activate
* sizeof(struct imsm_disk
);
8121 /* check if we need a larger metadata buffer */
8122 if (super
->next_buf
)
8123 buf_len
= super
->next_len
;
8125 buf_len
= super
->len
;
8127 if (__le32_to_cpu(mpb
->mpb_size
) + len
> buf_len
) {
8128 /* ok we need a larger buf than what is currently allocated
8129 * if this allocation fails process_update will notice that
8130 * ->next_len is set and ->next_buf is NULL
8132 buf_len
= ROUND_UP(__le32_to_cpu(mpb
->mpb_size
) + len
, 512);
8133 if (super
->next_buf
)
8134 free(super
->next_buf
);
8136 super
->next_len
= buf_len
;
8137 if (posix_memalign(&super
->next_buf
, 512, buf_len
) == 0)
8138 memset(super
->next_buf
, 0, buf_len
);
8140 super
->next_buf
= NULL
;
8144 /* must be called while manager is quiesced */
8145 static void imsm_delete(struct intel_super
*super
, struct dl
**dlp
, unsigned index
)
8147 struct imsm_super
*mpb
= super
->anchor
;
8149 struct imsm_dev
*dev
;
8150 struct imsm_map
*map
;
8151 int i
, j
, num_members
;
8154 dprintf("%s: deleting device[%d] from imsm_super\n",
8157 /* shift all indexes down one */
8158 for (iter
= super
->disks
; iter
; iter
= iter
->next
)
8159 if (iter
->index
> (int)index
)
8161 for (iter
= super
->missing
; iter
; iter
= iter
->next
)
8162 if (iter
->index
> (int)index
)
8165 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
8166 dev
= get_imsm_dev(super
, i
);
8167 map
= get_imsm_map(dev
, MAP_0
);
8168 num_members
= map
->num_members
;
8169 for (j
= 0; j
< num_members
; j
++) {
8170 /* update ord entries being careful not to propagate
8171 * ord-flags to the first map
8173 ord
= get_imsm_ord_tbl_ent(dev
, j
, MAP_X
);
8175 if (ord_to_idx(ord
) <= index
)
8178 map
= get_imsm_map(dev
, MAP_0
);
8179 set_imsm_ord_tbl_ent(map
, j
, ord_to_idx(ord
- 1));
8180 map
= get_imsm_map(dev
, MAP_1
);
8182 set_imsm_ord_tbl_ent(map
, j
, ord
- 1);
8187 super
->updates_pending
++;
8189 struct dl
*dl
= *dlp
;
8191 *dlp
= (*dlp
)->next
;
8192 __free_imsm_disk(dl
);
8195 #endif /* MDASSEMBLE */
8197 static void close_targets(int *targets
, int new_disks
)
8204 for (i
= 0; i
< new_disks
; i
++) {
8205 if (targets
[i
] >= 0) {
8212 static int imsm_get_allowed_degradation(int level
, int raid_disks
,
8213 struct intel_super
*super
,
8214 struct imsm_dev
*dev
)
8219 struct imsm_map
*map
;
8222 ret_val
= raid_disks
/2;
8223 /* check map if all disks pairs not failed
8226 map
= get_imsm_map(dev
, MAP_0
);
8227 for (i
= 0; i
< ret_val
; i
++) {
8228 int degradation
= 0;
8229 if (get_imsm_disk(super
, i
) == NULL
)
8231 if (get_imsm_disk(super
, i
+ 1) == NULL
)
8233 if (degradation
== 2)
8236 map
= get_imsm_map(dev
, MAP_1
);
8237 /* if there is no second map
8238 * result can be returned
8242 /* check degradation in second map
8244 for (i
= 0; i
< ret_val
; i
++) {
8245 int degradation
= 0;
8246 if (get_imsm_disk(super
, i
) == NULL
)
8248 if (get_imsm_disk(super
, i
+ 1) == NULL
)
8250 if (degradation
== 2)
8265 /*******************************************************************************
8266 * Function: open_backup_targets
8267 * Description: Function opens file descriptors for all devices given in
8270 * info : general array info
8271 * raid_disks : number of disks
8272 * raid_fds : table of device's file descriptors
8273 * super : intel super for raid10 degradation check
8274 * dev : intel device for raid10 degradation check
8278 ******************************************************************************/
8279 int open_backup_targets(struct mdinfo
*info
, int raid_disks
, int *raid_fds
,
8280 struct intel_super
*super
, struct imsm_dev
*dev
)
8286 for (i
= 0; i
< raid_disks
; i
++)
8289 for (sd
= info
->devs
; sd
; sd
= sd
->next
) {
8292 if (sd
->disk
.state
& (1<<MD_DISK_FAULTY
)) {
8293 dprintf("disk is faulty!!\n");
8297 if ((sd
->disk
.raid_disk
>= raid_disks
) ||
8298 (sd
->disk
.raid_disk
< 0))
8301 dn
= map_dev(sd
->disk
.major
,
8303 raid_fds
[sd
->disk
.raid_disk
] = dev_open(dn
, O_RDWR
);
8304 if (raid_fds
[sd
->disk
.raid_disk
] < 0) {
8305 fprintf(stderr
, "cannot open component\n");
8310 /* check if maximum array degradation level is not exceeded
8312 if ((raid_disks
- opened
) >
8313 imsm_get_allowed_degradation(info
->new_level
,
8316 fprintf(stderr
, "Not enough disks can be opened.\n");
8317 close_targets(raid_fds
, raid_disks
);
8324 /*******************************************************************************
8325 * Function: init_migr_record_imsm
8326 * Description: Function inits imsm migration record
8328 * super : imsm internal array info
8329 * dev : device under migration
8330 * info : general array info to find the smallest device
8333 ******************************************************************************/
8334 void init_migr_record_imsm(struct supertype
*st
, struct imsm_dev
*dev
,
8335 struct mdinfo
*info
)
8337 struct intel_super
*super
= st
->sb
;
8338 struct migr_record
*migr_rec
= super
->migr_rec
;
8340 unsigned long long dsize
, dev_sectors
;
8341 long long unsigned min_dev_sectors
= -1LLU;
8345 struct imsm_map
*map_dest
= get_imsm_map(dev
, MAP_0
);
8346 struct imsm_map
*map_src
= get_imsm_map(dev
, MAP_1
);
8347 unsigned long long num_migr_units
;
8348 unsigned long long array_blocks
;
8350 memset(migr_rec
, 0, sizeof(struct migr_record
));
8351 migr_rec
->family_num
= __cpu_to_le32(super
->anchor
->family_num
);
8353 /* only ascending reshape supported now */
8354 migr_rec
->ascending_migr
= __cpu_to_le32(1);
8356 migr_rec
->dest_depth_per_unit
= GEN_MIGR_AREA_SIZE
/
8357 max(map_dest
->blocks_per_strip
, map_src
->blocks_per_strip
);
8358 migr_rec
->dest_depth_per_unit
*= map_dest
->blocks_per_strip
;
8359 new_data_disks
= imsm_num_data_members(dev
, MAP_0
);
8360 migr_rec
->blocks_per_unit
=
8361 __cpu_to_le32(migr_rec
->dest_depth_per_unit
* new_data_disks
);
8362 migr_rec
->dest_depth_per_unit
=
8363 __cpu_to_le32(migr_rec
->dest_depth_per_unit
);
8364 array_blocks
= info
->component_size
* new_data_disks
;
8366 array_blocks
/ __le32_to_cpu(migr_rec
->blocks_per_unit
);
8368 if (array_blocks
% __le32_to_cpu(migr_rec
->blocks_per_unit
))
8370 migr_rec
->num_migr_units
= __cpu_to_le32(num_migr_units
);
8372 migr_rec
->post_migr_vol_cap
= dev
->size_low
;
8373 migr_rec
->post_migr_vol_cap_hi
= dev
->size_high
;
8376 /* Find the smallest dev */
8377 for (sd
= info
->devs
; sd
; sd
= sd
->next
) {
8378 sprintf(nm
, "%d:%d", sd
->disk
.major
, sd
->disk
.minor
);
8379 fd
= dev_open(nm
, O_RDONLY
);
8382 get_dev_size(fd
, NULL
, &dsize
);
8383 dev_sectors
= dsize
/ 512;
8384 if (dev_sectors
< min_dev_sectors
)
8385 min_dev_sectors
= dev_sectors
;
8388 migr_rec
->ckpt_area_pba
= __cpu_to_le32(min_dev_sectors
-
8389 RAID_DISK_RESERVED_BLOCKS_IMSM_HI
);
8391 write_imsm_migr_rec(st
);
8396 /*******************************************************************************
8397 * Function: save_backup_imsm
8398 * Description: Function saves critical data stripes to Migration Copy Area
8399 * and updates the current migration unit status.
8400 * Use restore_stripes() to form a destination stripe,
8401 * and to write it to the Copy Area.
8403 * st : supertype information
8404 * dev : imsm device that backup is saved for
8405 * info : general array info
8406 * buf : input buffer
8407 * length : length of data to backup (blocks_per_unit)
8411 ******************************************************************************/
8412 int save_backup_imsm(struct supertype
*st
,
8413 struct imsm_dev
*dev
,
8414 struct mdinfo
*info
,
8419 struct intel_super
*super
= st
->sb
;
8420 unsigned long long *target_offsets
= NULL
;
8421 int *targets
= NULL
;
8423 struct imsm_map
*map_dest
= get_imsm_map(dev
, MAP_0
);
8424 int new_disks
= map_dest
->num_members
;
8425 int dest_layout
= 0;
8427 unsigned long long start
;
8428 int data_disks
= imsm_num_data_members(dev
, MAP_0
);
8430 targets
= malloc(new_disks
* sizeof(int));
8434 for (i
= 0; i
< new_disks
; i
++)
8437 target_offsets
= malloc(new_disks
* sizeof(unsigned long long));
8438 if (!target_offsets
)
8441 start
= info
->reshape_progress
* 512;
8442 for (i
= 0; i
< new_disks
; i
++) {
8443 target_offsets
[i
] = (unsigned long long)
8444 __le32_to_cpu(super
->migr_rec
->ckpt_area_pba
) * 512;
8445 /* move back copy area adderss, it will be moved forward
8446 * in restore_stripes() using start input variable
8448 target_offsets
[i
] -= start
/data_disks
;
8451 if (open_backup_targets(info
, new_disks
, targets
,
8455 dest_layout
= imsm_level_to_layout(map_dest
->raid_level
);
8456 dest_chunk
= __le16_to_cpu(map_dest
->blocks_per_strip
) * 512;
8458 if (restore_stripes(targets
, /* list of dest devices */
8459 target_offsets
, /* migration record offsets */
8462 map_dest
->raid_level
,
8464 -1, /* source backup file descriptor */
8465 0, /* input buf offset
8466 * always 0 buf is already offseted */
8470 fprintf(stderr
, Name
": Error restoring stripes\n");
8478 close_targets(targets
, new_disks
);
8481 free(target_offsets
);
8486 /*******************************************************************************
8487 * Function: save_checkpoint_imsm
8488 * Description: Function called for current unit status update
8489 * in the migration record. It writes it to disk.
8491 * super : imsm internal array info
8492 * info : general array info
8496 * 2: failure, means no valid migration record
8497 * / no general migration in progress /
8498 ******************************************************************************/
8499 int save_checkpoint_imsm(struct supertype
*st
, struct mdinfo
*info
, int state
)
8501 struct intel_super
*super
= st
->sb
;
8502 unsigned long long blocks_per_unit
;
8503 unsigned long long curr_migr_unit
;
8505 if (load_imsm_migr_rec(super
, info
) != 0) {
8506 dprintf("imsm: ERROR: Cannot read migration record "
8507 "for checkpoint save.\n");
8511 blocks_per_unit
= __le32_to_cpu(super
->migr_rec
->blocks_per_unit
);
8512 if (blocks_per_unit
== 0) {
8513 dprintf("imsm: no migration in progress.\n");
8516 curr_migr_unit
= info
->reshape_progress
/ blocks_per_unit
;
8517 /* check if array is alligned to copy area
8518 * if it is not alligned, add one to current migration unit value
8519 * this can happend on array reshape finish only
8521 if (info
->reshape_progress
% blocks_per_unit
)
8524 super
->migr_rec
->curr_migr_unit
=
8525 __cpu_to_le32(curr_migr_unit
);
8526 super
->migr_rec
->rec_status
= __cpu_to_le32(state
);
8527 super
->migr_rec
->dest_1st_member_lba
=
8528 __cpu_to_le32(curr_migr_unit
*
8529 __le32_to_cpu(super
->migr_rec
->dest_depth_per_unit
));
8530 if (write_imsm_migr_rec(st
) < 0) {
8531 dprintf("imsm: Cannot write migration record "
8532 "outside backup area\n");
8539 /*******************************************************************************
8540 * Function: recover_backup_imsm
8541 * Description: Function recovers critical data from the Migration Copy Area
8542 * while assembling an array.
8544 * super : imsm internal array info
8545 * info : general array info
8547 * 0 : success (or there is no data to recover)
8549 ******************************************************************************/
8550 int recover_backup_imsm(struct supertype
*st
, struct mdinfo
*info
)
8552 struct intel_super
*super
= st
->sb
;
8553 struct migr_record
*migr_rec
= super
->migr_rec
;
8554 struct imsm_map
*map_dest
= NULL
;
8555 struct intel_dev
*id
= NULL
;
8556 unsigned long long read_offset
;
8557 unsigned long long write_offset
;
8559 int *targets
= NULL
;
8560 int new_disks
, i
, err
;
8563 unsigned long curr_migr_unit
= __le32_to_cpu(migr_rec
->curr_migr_unit
);
8564 unsigned long num_migr_units
= __le32_to_cpu(migr_rec
->num_migr_units
);
8566 int skipped_disks
= 0;
8568 err
= sysfs_get_str(info
, NULL
, "array_state", (char *)buffer
, 20);
8572 /* recover data only during assemblation */
8573 if (strncmp(buffer
, "inactive", 8) != 0)
8575 /* no data to recover */
8576 if (__le32_to_cpu(migr_rec
->rec_status
) == UNIT_SRC_NORMAL
)
8578 if (curr_migr_unit
>= num_migr_units
)
8581 /* find device during reshape */
8582 for (id
= super
->devlist
; id
; id
= id
->next
)
8583 if (is_gen_migration(id
->dev
))
8588 map_dest
= get_imsm_map(id
->dev
, MAP_0
);
8589 new_disks
= map_dest
->num_members
;
8591 read_offset
= (unsigned long long)
8592 __le32_to_cpu(migr_rec
->ckpt_area_pba
) * 512;
8594 write_offset
= ((unsigned long long)
8595 __le32_to_cpu(migr_rec
->dest_1st_member_lba
) +
8596 __le32_to_cpu(map_dest
->pba_of_lba0
)) * 512;
8598 unit_len
= __le32_to_cpu(migr_rec
->dest_depth_per_unit
) * 512;
8599 if (posix_memalign((void **)&buf
, 512, unit_len
) != 0)
8601 targets
= malloc(new_disks
* sizeof(int));
8605 if (open_backup_targets(info
, new_disks
, targets
, super
, id
->dev
)) {
8607 Name
": Cannot open some devices belonging to array.\n");
8611 for (i
= 0; i
< new_disks
; i
++) {
8612 if (targets
[i
] < 0) {
8616 if (lseek64(targets
[i
], read_offset
, SEEK_SET
) < 0) {
8618 Name
": Cannot seek to block: %s\n",
8623 if ((unsigned)read(targets
[i
], buf
, unit_len
) != unit_len
) {
8625 Name
": Cannot read copy area block: %s\n",
8630 if (lseek64(targets
[i
], write_offset
, SEEK_SET
) < 0) {
8632 Name
": Cannot seek to block: %s\n",
8637 if ((unsigned)write(targets
[i
], buf
, unit_len
) != unit_len
) {
8639 Name
": Cannot restore block: %s\n",
8646 if (skipped_disks
> imsm_get_allowed_degradation(info
->new_level
,
8651 Name
": Cannot restore data from backup."
8652 " Too many failed disks\n");
8656 if (save_checkpoint_imsm(st
, info
, UNIT_SRC_NORMAL
)) {
8657 /* ignore error == 2, this can mean end of reshape here
8659 dprintf("imsm: Cannot write checkpoint to "
8660 "migration record (UNIT_SRC_NORMAL) during restart\n");
8666 for (i
= 0; i
< new_disks
; i
++)
8675 static char disk_by_path
[] = "/dev/disk/by-path/";
8677 static const char *imsm_get_disk_controller_domain(const char *path
)
8679 char disk_path
[PATH_MAX
];
8683 strncpy(disk_path
, disk_by_path
, PATH_MAX
- 1);
8684 strncat(disk_path
, path
, PATH_MAX
- strlen(disk_path
) - 1);
8685 if (stat(disk_path
, &st
) == 0) {
8686 struct sys_dev
* hba
;
8689 path
= devt_to_devpath(st
.st_rdev
);
8692 hba
= find_disk_attached_hba(-1, path
);
8693 if (hba
&& hba
->type
== SYS_DEV_SAS
)
8695 else if (hba
&& hba
->type
== SYS_DEV_SATA
)
8699 dprintf("path: %s hba: %s attached: %s\n",
8700 path
, (hba
) ? hba
->path
: "NULL", drv
);
8708 static int imsm_find_array_minor_by_subdev(int subdev
, int container
, int *minor
)
8710 char subdev_name
[20];
8711 struct mdstat_ent
*mdstat
;
8713 sprintf(subdev_name
, "%d", subdev
);
8714 mdstat
= mdstat_by_subdev(subdev_name
, container
);
8718 *minor
= mdstat
->devnum
;
8719 free_mdstat(mdstat
);
8723 static int imsm_reshape_is_allowed_on_container(struct supertype
*st
,
8724 struct geo_params
*geo
,
8725 int *old_raid_disks
)
8727 /* currently we only support increasing the number of devices
8728 * for a container. This increases the number of device for each
8729 * member array. They must all be RAID0 or RAID5.
8732 struct mdinfo
*info
, *member
;
8733 int devices_that_can_grow
= 0;
8735 dprintf("imsm: imsm_reshape_is_allowed_on_container(ENTER): "
8736 "st->devnum = (%i)\n",
8739 if (geo
->size
!= -1 ||
8740 geo
->level
!= UnSet
||
8741 geo
->layout
!= UnSet
||
8742 geo
->chunksize
!= 0 ||
8743 geo
->raid_disks
== UnSet
) {
8744 dprintf("imsm: Container operation is allowed for "
8745 "raid disks number change only.\n");
8749 info
= container_content_imsm(st
, NULL
);
8750 for (member
= info
; member
; member
= member
->next
) {
8754 dprintf("imsm: checking device_num: %i\n",
8755 member
->container_member
);
8757 if (geo
->raid_disks
<= member
->array
.raid_disks
) {
8758 /* we work on container for Online Capacity Expansion
8759 * only so raid_disks has to grow
8761 dprintf("imsm: for container operation raid disks "
8762 "increase is required\n");
8766 if ((info
->array
.level
!= 0) &&
8767 (info
->array
.level
!= 5)) {
8768 /* we cannot use this container with other raid level
8770 dprintf("imsm: for container operation wrong"
8771 " raid level (%i) detected\n",
8775 /* check for platform support
8776 * for this raid level configuration
8778 struct intel_super
*super
= st
->sb
;
8779 if (!is_raid_level_supported(super
->orom
,
8780 member
->array
.level
,
8782 dprintf("platform does not support raid%d with"
8786 geo
->raid_disks
> 1 ? "s" : "");
8789 /* check if component size is aligned to chunk size
8791 if (info
->component_size
%
8792 (info
->array
.chunk_size
/512)) {
8793 dprintf("Component size is not aligned to "
8799 if (*old_raid_disks
&&
8800 info
->array
.raid_disks
!= *old_raid_disks
)
8802 *old_raid_disks
= info
->array
.raid_disks
;
8804 /* All raid5 and raid0 volumes in container
8805 * have to be ready for Online Capacity Expansion
8806 * so they need to be assembled. We have already
8807 * checked that no recovery etc is happening.
8809 result
= imsm_find_array_minor_by_subdev(member
->container_member
,
8813 dprintf("imsm: cannot find array\n");
8816 devices_that_can_grow
++;
8819 if (!member
&& devices_that_can_grow
)
8823 dprintf("\tContainer operation allowed\n");
8825 dprintf("\tError: %i\n", ret_val
);
8830 /* Function: get_spares_for_grow
8831 * Description: Allocates memory and creates list of spare devices
8832 * avaliable in container. Checks if spare drive size is acceptable.
8833 * Parameters: Pointer to the supertype structure
8834 * Returns: Pointer to the list of spare devices (mdinfo structure) on success,
8837 static struct mdinfo
*get_spares_for_grow(struct supertype
*st
)
8839 unsigned long long min_size
= min_acceptable_spare_size_imsm(st
);
8840 return container_choose_spares(st
, min_size
, NULL
, NULL
, NULL
, 0);
8843 /******************************************************************************
8844 * function: imsm_create_metadata_update_for_reshape
8845 * Function creates update for whole IMSM container.
8847 ******************************************************************************/
8848 static int imsm_create_metadata_update_for_reshape(
8849 struct supertype
*st
,
8850 struct geo_params
*geo
,
8852 struct imsm_update_reshape
**updatep
)
8854 struct intel_super
*super
= st
->sb
;
8855 struct imsm_super
*mpb
= super
->anchor
;
8856 int update_memory_size
= 0;
8857 struct imsm_update_reshape
*u
= NULL
;
8858 struct mdinfo
*spares
= NULL
;
8860 int delta_disks
= 0;
8863 dprintf("imsm_update_metadata_for_reshape(enter) raid_disks = %i\n",
8866 delta_disks
= geo
->raid_disks
- old_raid_disks
;
8868 /* size of all update data without anchor */
8869 update_memory_size
= sizeof(struct imsm_update_reshape
);
8871 /* now add space for spare disks that we need to add. */
8872 update_memory_size
+= sizeof(u
->new_disks
[0]) * (delta_disks
- 1);
8874 u
= calloc(1, update_memory_size
);
8877 "cannot get memory for imsm_update_reshape update\n");
8880 u
->type
= update_reshape_container_disks
;
8881 u
->old_raid_disks
= old_raid_disks
;
8882 u
->new_raid_disks
= geo
->raid_disks
;
8884 /* now get spare disks list
8886 spares
= get_spares_for_grow(st
);
8889 || delta_disks
> spares
->array
.spare_disks
) {
8890 fprintf(stderr
, Name
": imsm: ERROR: Cannot get spare devices "
8891 "for %s.\n", geo
->dev_name
);
8896 /* we have got spares
8897 * update disk list in imsm_disk list table in anchor
8899 dprintf("imsm: %i spares are available.\n\n",
8900 spares
->array
.spare_disks
);
8903 for (i
= 0; i
< delta_disks
; i
++) {
8908 u
->new_disks
[i
] = makedev(dev
->disk
.major
,
8910 dl
= get_disk_super(super
, dev
->disk
.major
, dev
->disk
.minor
);
8911 dl
->index
= mpb
->num_disks
;
8921 dprintf("imsm: reshape update preparation :");
8922 if (i
== delta_disks
) {
8925 return update_memory_size
;
8928 dprintf(" Error\n");
8933 /******************************************************************************
8934 * function: imsm_create_metadata_update_for_migration()
8935 * Creates update for IMSM array.
8937 ******************************************************************************/
8938 static int imsm_create_metadata_update_for_migration(
8939 struct supertype
*st
,
8940 struct geo_params
*geo
,
8941 struct imsm_update_reshape_migration
**updatep
)
8943 struct intel_super
*super
= st
->sb
;
8944 int update_memory_size
= 0;
8945 struct imsm_update_reshape_migration
*u
= NULL
;
8946 struct imsm_dev
*dev
;
8947 int previous_level
= -1;
8949 dprintf("imsm_create_metadata_update_for_migration(enter)"
8950 " New Level = %i\n", geo
->level
);
8952 /* size of all update data without anchor */
8953 update_memory_size
= sizeof(struct imsm_update_reshape_migration
);
8955 u
= calloc(1, update_memory_size
);
8957 dprintf("error: cannot get memory for "
8958 "imsm_create_metadata_update_for_migration\n");
8961 u
->type
= update_reshape_migration
;
8962 u
->subdev
= super
->current_vol
;
8963 u
->new_level
= geo
->level
;
8964 u
->new_layout
= geo
->layout
;
8965 u
->new_raid_disks
= u
->old_raid_disks
= geo
->raid_disks
;
8966 u
->new_disks
[0] = -1;
8967 u
->new_chunksize
= -1;
8969 dev
= get_imsm_dev(super
, u
->subdev
);
8971 struct imsm_map
*map
;
8973 map
= get_imsm_map(dev
, MAP_0
);
8975 int current_chunk_size
=
8976 __le16_to_cpu(map
->blocks_per_strip
) / 2;
8978 if (geo
->chunksize
!= current_chunk_size
) {
8979 u
->new_chunksize
= geo
->chunksize
/ 1024;
8981 "chunk size change from %i to %i\n",
8982 current_chunk_size
, u
->new_chunksize
);
8984 previous_level
= map
->raid_level
;
8987 if ((geo
->level
== 5) && (previous_level
== 0)) {
8988 struct mdinfo
*spares
= NULL
;
8990 u
->new_raid_disks
++;
8991 spares
= get_spares_for_grow(st
);
8992 if ((spares
== NULL
) || (spares
->array
.spare_disks
< 1)) {
8995 update_memory_size
= 0;
8996 dprintf("error: cannot get spare device "
8997 "for requested migration");
9002 dprintf("imsm: reshape update preparation : OK\n");
9005 return update_memory_size
;
9008 static void imsm_update_metadata_locally(struct supertype
*st
,
9011 struct metadata_update mu
;
9016 mu
.space_list
= NULL
;
9018 imsm_prepare_update(st
, &mu
);
9019 imsm_process_update(st
, &mu
);
9021 while (mu
.space_list
) {
9022 void **space
= mu
.space_list
;
9023 mu
.space_list
= *space
;
9028 /***************************************************************************
9029 * Function: imsm_analyze_change
9030 * Description: Function analyze change for single volume
9031 * and validate if transition is supported
9032 * Parameters: Geometry parameters, supertype structure
9033 * Returns: Operation type code on success, -1 if fail
9034 ****************************************************************************/
9035 enum imsm_reshape_type
imsm_analyze_change(struct supertype
*st
,
9036 struct geo_params
*geo
)
9045 getinfo_super_imsm_volume(st
, &info
, NULL
);
9046 if ((geo
->level
!= info
.array
.level
) &&
9047 (geo
->level
>= 0) &&
9048 (geo
->level
!= UnSet
)) {
9049 switch (info
.array
.level
) {
9051 if (geo
->level
== 5) {
9052 change
= CH_MIGRATION
;
9053 if (geo
->layout
!= ALGORITHM_LEFT_ASYMMETRIC
) {
9055 Name
" Error. Requested Layout "
9056 "not supported (left-asymmetric layout "
9057 "is supported only)!\n");
9059 goto analyse_change_exit
;
9061 layout
= geo
->layout
;
9063 devNumChange
= 1; /* parity disk added */
9064 } else if (geo
->level
== 10) {
9065 change
= CH_TAKEOVER
;
9067 devNumChange
= 2; /* two mirrors added */
9068 layout
= 0x102; /* imsm supported layout */
9073 if (geo
->level
== 0) {
9074 change
= CH_TAKEOVER
;
9076 devNumChange
= -(geo
->raid_disks
/2);
9077 layout
= 0; /* imsm raid0 layout */
9083 Name
" Error. Level Migration from %d to %d "
9085 info
.array
.level
, geo
->level
);
9086 goto analyse_change_exit
;
9089 geo
->level
= info
.array
.level
;
9091 if ((geo
->layout
!= info
.array
.layout
)
9092 && ((geo
->layout
!= UnSet
) && (geo
->layout
!= -1))) {
9093 change
= CH_MIGRATION
;
9094 if ((info
.array
.layout
== 0)
9095 && (info
.array
.level
== 5)
9096 && (geo
->layout
== 5)) {
9097 /* reshape 5 -> 4 */
9098 } else if ((info
.array
.layout
== 5)
9099 && (info
.array
.level
== 5)
9100 && (geo
->layout
== 0)) {
9101 /* reshape 4 -> 5 */
9106 Name
" Error. Layout Migration from %d to %d "
9108 info
.array
.layout
, geo
->layout
);
9110 goto analyse_change_exit
;
9113 geo
->layout
= info
.array
.layout
;
9115 if ((geo
->chunksize
> 0) && (geo
->chunksize
!= UnSet
)
9116 && (geo
->chunksize
!= info
.array
.chunk_size
))
9117 change
= CH_MIGRATION
;
9119 geo
->chunksize
= info
.array
.chunk_size
;
9121 chunk
= geo
->chunksize
/ 1024;
9122 if (!validate_geometry_imsm(st
,
9125 geo
->raid_disks
+ devNumChange
,
9132 struct intel_super
*super
= st
->sb
;
9133 struct imsm_super
*mpb
= super
->anchor
;
9135 if (mpb
->num_raid_devs
> 1) {
9137 Name
" Error. Cannot perform operation on %s"
9138 "- for this operation it MUST be single "
9139 "array in container\n",
9145 analyse_change_exit
:
9150 int imsm_takeover(struct supertype
*st
, struct geo_params
*geo
)
9152 struct intel_super
*super
= st
->sb
;
9153 struct imsm_update_takeover
*u
;
9155 u
= malloc(sizeof(struct imsm_update_takeover
));
9159 u
->type
= update_takeover
;
9160 u
->subarray
= super
->current_vol
;
9162 /* 10->0 transition */
9163 if (geo
->level
== 0)
9164 u
->direction
= R10_TO_R0
;
9166 /* 0->10 transition */
9167 if (geo
->level
== 10)
9168 u
->direction
= R0_TO_R10
;
9170 /* update metadata locally */
9171 imsm_update_metadata_locally(st
, u
,
9172 sizeof(struct imsm_update_takeover
));
9173 /* and possibly remotely */
9174 if (st
->update_tail
)
9175 append_metadata_update(st
, u
,
9176 sizeof(struct imsm_update_takeover
));
9183 static int imsm_reshape_super(struct supertype
*st
, long long size
, int level
,
9184 int layout
, int chunksize
, int raid_disks
,
9185 int delta_disks
, char *backup
, char *dev
,
9189 struct geo_params geo
;
9191 dprintf("imsm: reshape_super called.\n");
9193 memset(&geo
, 0, sizeof(struct geo_params
));
9196 geo
.dev_id
= st
->devnum
;
9199 geo
.layout
= layout
;
9200 geo
.chunksize
= chunksize
;
9201 geo
.raid_disks
= raid_disks
;
9202 if (delta_disks
!= UnSet
)
9203 geo
.raid_disks
+= delta_disks
;
9205 dprintf("\tfor level : %i\n", geo
.level
);
9206 dprintf("\tfor raid_disks : %i\n", geo
.raid_disks
);
9208 if (experimental() == 0)
9211 if (st
->container_dev
== st
->devnum
) {
9212 /* On container level we can only increase number of devices. */
9213 dprintf("imsm: info: Container operation\n");
9214 int old_raid_disks
= 0;
9216 if (imsm_reshape_is_allowed_on_container(
9217 st
, &geo
, &old_raid_disks
)) {
9218 struct imsm_update_reshape
*u
= NULL
;
9221 len
= imsm_create_metadata_update_for_reshape(
9222 st
, &geo
, old_raid_disks
, &u
);
9225 dprintf("imsm: Cannot prepare update\n");
9226 goto exit_imsm_reshape_super
;
9230 /* update metadata locally */
9231 imsm_update_metadata_locally(st
, u
, len
);
9232 /* and possibly remotely */
9233 if (st
->update_tail
)
9234 append_metadata_update(st
, u
, len
);
9239 fprintf(stderr
, Name
": (imsm) Operation "
9240 "is not allowed on this container\n");
9243 /* On volume level we support following operations
9244 * - takeover: raid10 -> raid0; raid0 -> raid10
9245 * - chunk size migration
9246 * - migration: raid5 -> raid0; raid0 -> raid5
9248 struct intel_super
*super
= st
->sb
;
9249 struct intel_dev
*dev
= super
->devlist
;
9251 dprintf("imsm: info: Volume operation\n");
9252 /* find requested device */
9254 if (imsm_find_array_minor_by_subdev(
9255 dev
->index
, st
->container_dev
, &devnum
) == 0
9256 && devnum
== geo
.dev_id
)
9261 fprintf(stderr
, Name
" Cannot find %s (%i) subarray\n",
9262 geo
.dev_name
, geo
.dev_id
);
9263 goto exit_imsm_reshape_super
;
9265 super
->current_vol
= dev
->index
;
9266 change
= imsm_analyze_change(st
, &geo
);
9269 ret_val
= imsm_takeover(st
, &geo
);
9271 case CH_MIGRATION
: {
9272 struct imsm_update_reshape_migration
*u
= NULL
;
9274 imsm_create_metadata_update_for_migration(
9278 "Cannot prepare update\n");
9282 /* update metadata locally */
9283 imsm_update_metadata_locally(st
, u
, len
);
9284 /* and possibly remotely */
9285 if (st
->update_tail
)
9286 append_metadata_update(st
, u
, len
);
9296 exit_imsm_reshape_super
:
9297 dprintf("imsm: reshape_super Exit code = %i\n", ret_val
);
9301 /*******************************************************************************
9302 * Function: wait_for_reshape_imsm
9303 * Description: Function writes new sync_max value and waits until
9304 * reshape process reach new position
9306 * sra : general array info
9307 * ndata : number of disks in new array's layout
9310 * 1 : there is no reshape in progress,
9312 ******************************************************************************/
9313 int wait_for_reshape_imsm(struct mdinfo
*sra
, int ndata
)
9315 int fd
= sysfs_get_fd(sra
, NULL
, "reshape_position");
9316 unsigned long long completed
;
9317 /* to_complete : new sync_max position */
9318 unsigned long long to_complete
= sra
->reshape_progress
;
9319 unsigned long long position_to_set
= to_complete
/ ndata
;
9322 dprintf("imsm: wait_for_reshape_imsm() "
9323 "cannot open reshape_position\n");
9327 if (sysfs_fd_get_ll(fd
, &completed
) < 0) {
9328 dprintf("imsm: wait_for_reshape_imsm() "
9329 "cannot read reshape_position (no reshape in progres)\n");
9334 if (completed
> to_complete
) {
9335 dprintf("imsm: wait_for_reshape_imsm() "
9336 "wrong next position to set %llu (%llu)\n",
9337 to_complete
, completed
);
9341 dprintf("Position set: %llu\n", position_to_set
);
9342 if (sysfs_set_num(sra
, NULL
, "sync_max",
9343 position_to_set
) != 0) {
9344 dprintf("imsm: wait_for_reshape_imsm() "
9345 "cannot set reshape position to %llu\n",
9356 select(fd
+1, &rfds
, NULL
, NULL
, NULL
);
9357 if (sysfs_get_str(sra
, NULL
, "sync_action",
9359 strncmp(action
, "reshape", 7) != 0)
9361 if (sysfs_fd_get_ll(fd
, &completed
) < 0) {
9362 dprintf("imsm: wait_for_reshape_imsm() "
9363 "cannot read reshape_position (in loop)\n");
9367 } while (completed
< to_complete
);
9373 /*******************************************************************************
9374 * Function: check_degradation_change
9375 * Description: Check that array hasn't become failed.
9377 * info : for sysfs access
9378 * sources : source disks descriptors
9379 * degraded: previous degradation level
9382 ******************************************************************************/
9383 int check_degradation_change(struct mdinfo
*info
,
9387 unsigned long long new_degraded
;
9388 sysfs_get_ll(info
, NULL
, "degraded", &new_degraded
);
9389 if (new_degraded
!= (unsigned long long)degraded
) {
9390 /* check each device to ensure it is still working */
9393 for (sd
= info
->devs
; sd
; sd
= sd
->next
) {
9394 if (sd
->disk
.state
& (1<<MD_DISK_FAULTY
))
9396 if (sd
->disk
.state
& (1<<MD_DISK_SYNC
)) {
9398 if (sysfs_get_str(info
,
9399 sd
, "state", sbuf
, 20) < 0 ||
9400 strstr(sbuf
, "faulty") ||
9401 strstr(sbuf
, "in_sync") == NULL
) {
9402 /* this device is dead */
9403 sd
->disk
.state
= (1<<MD_DISK_FAULTY
);
9404 if (sd
->disk
.raid_disk
>= 0 &&
9405 sources
[sd
->disk
.raid_disk
] >= 0) {
9407 sd
->disk
.raid_disk
]);
9408 sources
[sd
->disk
.raid_disk
] =
9417 return new_degraded
;
9420 /*******************************************************************************
9421 * Function: imsm_manage_reshape
9422 * Description: Function finds array under reshape and it manages reshape
9423 * process. It creates stripes backups (if required) and sets
9426 * afd : Backup handle (nattive) - not used
9427 * sra : general array info
9428 * reshape : reshape parameters - not used
9429 * st : supertype structure
9430 * blocks : size of critical section [blocks]
9431 * fds : table of source device descriptor
9432 * offsets : start of array (offest per devices)
9434 * destfd : table of destination device descriptor
9435 * destoffsets : table of destination offsets (per device)
9437 * 1 : success, reshape is done
9439 ******************************************************************************/
9440 static int imsm_manage_reshape(
9441 int afd
, struct mdinfo
*sra
, struct reshape
*reshape
,
9442 struct supertype
*st
, unsigned long backup_blocks
,
9443 int *fds
, unsigned long long *offsets
,
9444 int dests
, int *destfd
, unsigned long long *destoffsets
)
9447 struct intel_super
*super
= st
->sb
;
9448 struct intel_dev
*dv
= NULL
;
9449 struct imsm_dev
*dev
= NULL
;
9450 struct imsm_map
*map_src
;
9451 int migr_vol_qan
= 0;
9452 int ndata
, odata
; /* [bytes] */
9453 int chunk
; /* [bytes] */
9454 struct migr_record
*migr_rec
;
9456 unsigned int buf_size
; /* [bytes] */
9457 unsigned long long max_position
; /* array size [bytes] */
9458 unsigned long long next_step
; /* [blocks]/[bytes] */
9459 unsigned long long old_data_stripe_length
;
9460 unsigned long long start_src
; /* [bytes] */
9461 unsigned long long start
; /* [bytes] */
9462 unsigned long long start_buf_shift
; /* [bytes] */
9464 int source_layout
= 0;
9466 if (!fds
|| !offsets
|| !sra
)
9469 /* Find volume during the reshape */
9470 for (dv
= super
->devlist
; dv
; dv
= dv
->next
) {
9471 if (dv
->dev
->vol
.migr_type
== MIGR_GEN_MIGR
9472 && dv
->dev
->vol
.migr_state
== 1) {
9477 /* Only one volume can migrate at the same time */
9478 if (migr_vol_qan
!= 1) {
9479 fprintf(stderr
, Name
" : %s", migr_vol_qan
?
9480 "Number of migrating volumes greater than 1\n" :
9481 "There is no volume during migrationg\n");
9485 map_src
= get_imsm_map(dev
, MAP_1
);
9486 if (map_src
== NULL
)
9489 ndata
= imsm_num_data_members(dev
, MAP_0
);
9490 odata
= imsm_num_data_members(dev
, MAP_1
);
9492 chunk
= __le16_to_cpu(map_src
->blocks_per_strip
) * 512;
9493 old_data_stripe_length
= odata
* chunk
;
9495 migr_rec
= super
->migr_rec
;
9497 /* initialize migration record for start condition */
9498 if (sra
->reshape_progress
== 0)
9499 init_migr_record_imsm(st
, dev
, sra
);
9501 if (__le32_to_cpu(migr_rec
->rec_status
) != UNIT_SRC_NORMAL
) {
9502 dprintf("imsm: cannot restart migration when data "
9503 "are present in copy area.\n");
9509 buf_size
= __le32_to_cpu(migr_rec
->blocks_per_unit
) * 512;
9510 /* extend buffer size for parity disk */
9511 buf_size
+= __le32_to_cpu(migr_rec
->dest_depth_per_unit
) * 512;
9512 /* add space for stripe aligment */
9513 buf_size
+= old_data_stripe_length
;
9514 if (posix_memalign((void **)&buf
, 4096, buf_size
)) {
9515 dprintf("imsm: Cannot allocate checpoint buffer\n");
9519 max_position
= sra
->component_size
* ndata
;
9520 source_layout
= imsm_level_to_layout(map_src
->raid_level
);
9522 while (__le32_to_cpu(migr_rec
->curr_migr_unit
) <
9523 __le32_to_cpu(migr_rec
->num_migr_units
)) {
9524 /* current reshape position [blocks] */
9525 unsigned long long current_position
=
9526 __le32_to_cpu(migr_rec
->blocks_per_unit
)
9527 * __le32_to_cpu(migr_rec
->curr_migr_unit
);
9528 unsigned long long border
;
9530 /* Check that array hasn't become failed.
9532 degraded
= check_degradation_change(sra
, fds
, degraded
);
9534 dprintf("imsm: Abort reshape due to degradation"
9535 " level (%i)\n", degraded
);
9539 next_step
= __le32_to_cpu(migr_rec
->blocks_per_unit
);
9541 if ((current_position
+ next_step
) > max_position
)
9542 next_step
= max_position
- current_position
;
9544 start
= current_position
* 512;
9546 /* allign reading start to old geometry */
9547 start_buf_shift
= start
% old_data_stripe_length
;
9548 start_src
= start
- start_buf_shift
;
9550 border
= (start_src
/ odata
) - (start
/ ndata
);
9552 if (border
<= __le32_to_cpu(migr_rec
->dest_depth_per_unit
)) {
9553 /* save critical stripes to buf
9554 * start - start address of current unit
9556 * start_src - start address of current unit
9557 * to backup alligned to source array
9560 unsigned long long next_step_filler
= 0;
9561 unsigned long long copy_length
= next_step
* 512;
9563 /* allign copy area length to stripe in old geometry */
9564 next_step_filler
= ((copy_length
+ start_buf_shift
)
9565 % old_data_stripe_length
);
9566 if (next_step_filler
)
9567 next_step_filler
= (old_data_stripe_length
9568 - next_step_filler
);
9569 dprintf("save_stripes() parameters: start = %llu,"
9570 "\tstart_src = %llu,\tnext_step*512 = %llu,"
9571 "\tstart_in_buf_shift = %llu,"
9572 "\tnext_step_filler = %llu\n",
9573 start
, start_src
, copy_length
,
9574 start_buf_shift
, next_step_filler
);
9576 if (save_stripes(fds
, offsets
, map_src
->num_members
,
9577 chunk
, map_src
->raid_level
,
9578 source_layout
, 0, NULL
, start_src
,
9580 next_step_filler
+ start_buf_shift
,
9582 dprintf("imsm: Cannot save stripes"
9586 /* Convert data to destination format and store it
9587 * in backup general migration area
9589 if (save_backup_imsm(st
, dev
, sra
,
9590 buf
+ start_buf_shift
, copy_length
)) {
9591 dprintf("imsm: Cannot save stripes to "
9592 "target devices\n");
9595 if (save_checkpoint_imsm(st
, sra
,
9596 UNIT_SRC_IN_CP_AREA
)) {
9597 dprintf("imsm: Cannot write checkpoint to "
9598 "migration record (UNIT_SRC_IN_CP_AREA)\n");
9602 /* set next step to use whole border area */
9603 border
/= next_step
;
9605 next_step
*= border
;
9607 /* When data backed up, checkpoint stored,
9608 * kick the kernel to reshape unit of data
9610 next_step
= next_step
+ sra
->reshape_progress
;
9611 /* limit next step to array max position */
9612 if (next_step
> max_position
)
9613 next_step
= max_position
;
9614 sysfs_set_num(sra
, NULL
, "suspend_lo", sra
->reshape_progress
);
9615 sysfs_set_num(sra
, NULL
, "suspend_hi", next_step
);
9616 sra
->reshape_progress
= next_step
;
9618 /* wait until reshape finish */
9619 if (wait_for_reshape_imsm(sra
, ndata
) < 0) {
9620 dprintf("wait_for_reshape_imsm returned error!\n");
9624 if (save_checkpoint_imsm(st
, sra
, UNIT_SRC_NORMAL
) == 1) {
9625 /* ignore error == 2, this can mean end of reshape here
9627 dprintf("imsm: Cannot write checkpoint to "
9628 "migration record (UNIT_SRC_NORMAL)\n");
9634 /* return '1' if done */
9642 #endif /* MDASSEMBLE */
9644 struct superswitch super_imsm
= {
9646 .examine_super
= examine_super_imsm
,
9647 .brief_examine_super
= brief_examine_super_imsm
,
9648 .brief_examine_subarrays
= brief_examine_subarrays_imsm
,
9649 .export_examine_super
= export_examine_super_imsm
,
9650 .detail_super
= detail_super_imsm
,
9651 .brief_detail_super
= brief_detail_super_imsm
,
9652 .write_init_super
= write_init_super_imsm
,
9653 .validate_geometry
= validate_geometry_imsm
,
9654 .add_to_super
= add_to_super_imsm
,
9655 .remove_from_super
= remove_from_super_imsm
,
9656 .detail_platform
= detail_platform_imsm
,
9657 .kill_subarray
= kill_subarray_imsm
,
9658 .update_subarray
= update_subarray_imsm
,
9659 .load_container
= load_container_imsm
,
9660 .default_geometry
= default_geometry_imsm
,
9661 .get_disk_controller_domain
= imsm_get_disk_controller_domain
,
9662 .reshape_super
= imsm_reshape_super
,
9663 .manage_reshape
= imsm_manage_reshape
,
9664 .recover_backup
= recover_backup_imsm
,
9666 .match_home
= match_home_imsm
,
9667 .uuid_from_super
= uuid_from_super_imsm
,
9668 .getinfo_super
= getinfo_super_imsm
,
9669 .getinfo_super_disks
= getinfo_super_disks_imsm
,
9670 .update_super
= update_super_imsm
,
9672 .avail_size
= avail_size_imsm
,
9673 .min_acceptable_spare_size
= min_acceptable_spare_size_imsm
,
9675 .compare_super
= compare_super_imsm
,
9677 .load_super
= load_super_imsm
,
9678 .init_super
= init_super_imsm
,
9679 .store_super
= store_super_imsm
,
9680 .free_super
= free_super_imsm
,
9681 .match_metadata_desc
= match_metadata_desc_imsm
,
9682 .container_content
= container_content_imsm
,
9690 .open_new
= imsm_open_new
,
9691 .set_array_state
= imsm_set_array_state
,
9692 .set_disk
= imsm_set_disk
,
9693 .sync_metadata
= imsm_sync_metadata
,
9694 .activate_spare
= imsm_activate_spare
,
9695 .process_update
= imsm_process_update
,
9696 .prepare_update
= imsm_prepare_update
,
9697 #endif /* MDASSEMBLE */