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 | \
87 /* Define attributes that are unused but not harmful */
88 #define MPB_ATTRIB_IGNORED (MPB_ATTRIB_NEVER_USE)
90 #define MPB_SECTOR_CNT 2210
91 #define IMSM_RESERVED_SECTORS 4096
92 #define NUM_BLOCKS_DIRTY_STRIPE_REGION 2056
93 #define SECT_PER_MB_SHIFT 11
94 #define MAX_SECTOR_SIZE 4096
96 /* Disk configuration info. */
97 #define IMSM_MAX_DEVICES 255
99 __u8 serial
[MAX_RAID_SERIAL_LEN
];/* 0xD8 - 0xE7 ascii serial number */
100 __u32 total_blocks_lo
; /* 0xE8 - 0xEB total blocks lo */
101 __u32 scsi_id
; /* 0xEC - 0xEF scsi ID */
102 #define SPARE_DISK __cpu_to_le32(0x01) /* Spare */
103 #define CONFIGURED_DISK __cpu_to_le32(0x02) /* Member of some RaidDev */
104 #define FAILED_DISK __cpu_to_le32(0x04) /* Permanent failure */
105 __u32 status
; /* 0xF0 - 0xF3 */
106 __u32 owner_cfg_num
; /* which config 0,1,2... owns this disk */
107 __u32 total_blocks_hi
; /* 0xF4 - 0xF5 total blocks hi */
108 #define IMSM_DISK_FILLERS 3
109 __u32 filler
[IMSM_DISK_FILLERS
]; /* 0xF5 - 0x107 MPB_DISK_FILLERS for future expansion */
112 /* map selector for map managment
118 /* RAID map configuration infos. */
120 __u32 pba_of_lba0_lo
; /* start address of partition */
121 __u32 blocks_per_member_lo
;/* blocks per member */
122 __u32 num_data_stripes_lo
; /* number of data stripes */
123 __u16 blocks_per_strip
;
124 __u8 map_state
; /* Normal, Uninitialized, Degraded, Failed */
125 #define IMSM_T_STATE_NORMAL 0
126 #define IMSM_T_STATE_UNINITIALIZED 1
127 #define IMSM_T_STATE_DEGRADED 2
128 #define IMSM_T_STATE_FAILED 3
130 #define IMSM_T_RAID0 0
131 #define IMSM_T_RAID1 1
132 #define IMSM_T_RAID5 5 /* since metadata version 1.2.02 ? */
133 __u8 num_members
; /* number of member disks */
134 __u8 num_domains
; /* number of parity domains */
135 __u8 failed_disk_num
; /* valid only when state is degraded */
137 __u32 pba_of_lba0_hi
;
138 __u32 blocks_per_member_hi
;
139 __u32 num_data_stripes_hi
;
140 __u32 filler
[4]; /* expansion area */
141 #define IMSM_ORD_REBUILD (1 << 24)
142 __u32 disk_ord_tbl
[1]; /* disk_ord_tbl[num_members],
143 * top byte contains some flags
145 } __attribute__ ((packed
));
148 __u32 curr_migr_unit
;
149 __u32 checkpoint_id
; /* id to access curr_migr_unit */
150 __u8 migr_state
; /* Normal or Migrating */
152 #define MIGR_REBUILD 1
153 #define MIGR_VERIFY 2 /* analagous to echo check > sync_action */
154 #define MIGR_GEN_MIGR 3
155 #define MIGR_STATE_CHANGE 4
156 #define MIGR_REPAIR 5
157 __u8 migr_type
; /* Initializing, Rebuilding, ... */
159 __u8 fs_state
; /* fast-sync state for CnG (0xff == disabled) */
160 __u16 verify_errors
; /* number of mismatches */
161 __u16 bad_blocks
; /* number of bad blocks during verify */
163 struct imsm_map map
[1];
164 /* here comes another one if migr_state */
165 } __attribute__ ((packed
));
168 __u8 volume
[MAX_RAID_SERIAL_LEN
];
171 #define DEV_BOOTABLE __cpu_to_le32(0x01)
172 #define DEV_BOOT_DEVICE __cpu_to_le32(0x02)
173 #define DEV_READ_COALESCING __cpu_to_le32(0x04)
174 #define DEV_WRITE_COALESCING __cpu_to_le32(0x08)
175 #define DEV_LAST_SHUTDOWN_DIRTY __cpu_to_le32(0x10)
176 #define DEV_HIDDEN_AT_BOOT __cpu_to_le32(0x20)
177 #define DEV_CURRENTLY_HIDDEN __cpu_to_le32(0x40)
178 #define DEV_VERIFY_AND_FIX __cpu_to_le32(0x80)
179 #define DEV_MAP_STATE_UNINIT __cpu_to_le32(0x100)
180 #define DEV_NO_AUTO_RECOVERY __cpu_to_le32(0x200)
181 #define DEV_CLONE_N_GO __cpu_to_le32(0x400)
182 #define DEV_CLONE_MAN_SYNC __cpu_to_le32(0x800)
183 #define DEV_CNG_MASTER_DISK_NUM __cpu_to_le32(0x1000)
184 __u32 status
; /* Persistent RaidDev status */
185 __u32 reserved_blocks
; /* Reserved blocks at beginning of volume */
189 __u8 cng_master_disk
;
193 #define IMSM_DEV_FILLERS 10
194 __u32 filler
[IMSM_DEV_FILLERS
];
196 } __attribute__ ((packed
));
199 __u8 sig
[MAX_SIGNATURE_LENGTH
]; /* 0x00 - 0x1F */
200 __u32 check_sum
; /* 0x20 - 0x23 MPB Checksum */
201 __u32 mpb_size
; /* 0x24 - 0x27 Size of MPB */
202 __u32 family_num
; /* 0x28 - 0x2B Checksum from first time this config was written */
203 __u32 generation_num
; /* 0x2C - 0x2F Incremented each time this array's MPB is written */
204 __u32 error_log_size
; /* 0x30 - 0x33 in bytes */
205 __u32 attributes
; /* 0x34 - 0x37 */
206 __u8 num_disks
; /* 0x38 Number of configured disks */
207 __u8 num_raid_devs
; /* 0x39 Number of configured volumes */
208 __u8 error_log_pos
; /* 0x3A */
209 __u8 fill
[1]; /* 0x3B */
210 __u32 cache_size
; /* 0x3c - 0x40 in mb */
211 __u32 orig_family_num
; /* 0x40 - 0x43 original family num */
212 __u32 pwr_cycle_count
; /* 0x44 - 0x47 simulated power cycle count for array */
213 __u32 bbm_log_size
; /* 0x48 - 0x4B - size of bad Block Mgmt Log in bytes */
214 #define IMSM_FILLERS 35
215 __u32 filler
[IMSM_FILLERS
]; /* 0x4C - 0xD7 RAID_MPB_FILLERS */
216 struct imsm_disk disk
[1]; /* 0xD8 diskTbl[numDisks] */
217 /* here comes imsm_dev[num_raid_devs] */
218 /* here comes BBM logs */
219 } __attribute__ ((packed
));
221 #define BBM_LOG_MAX_ENTRIES 254
222 #define BBM_LOG_MAX_LBA_ENTRY_VAL 256 /* Represents 256 LBAs */
223 #define BBM_LOG_SIGNATURE 0xabadb10c
225 struct bbm_log_block_addr
{
228 } __attribute__ ((__packed__
));
230 struct bbm_log_entry
{
231 __u8 marked_count
; /* Number of blocks marked - 1 */
232 __u8 disk_ordinal
; /* Disk entry within the imsm_super */
233 struct bbm_log_block_addr defective_block_start
;
234 } __attribute__ ((__packed__
));
237 __u32 signature
; /* 0xABADB10C */
239 struct bbm_log_entry marked_block_entries
[BBM_LOG_MAX_ENTRIES
];
240 } __attribute__ ((__packed__
));
243 static char *map_state_str
[] = { "normal", "uninitialized", "degraded", "failed" };
246 #define RAID_DISK_RESERVED_BLOCKS_IMSM_HI 2209
248 #define GEN_MIGR_AREA_SIZE 2048 /* General Migration Copy Area size in blocks */
250 #define MIGR_REC_BUF_SECTORS 1 /* size of migr_record i/o buffer in sectors */
251 #define MIGR_REC_SECTOR_POSITION 1 /* migr_record position offset on disk,
252 * MIGR_REC_BUF_SECTORS <= MIGR_REC_SECTOR_POS
255 #define UNIT_SRC_NORMAL 0 /* Source data for curr_migr_unit must
256 * be recovered using srcMap */
257 #define UNIT_SRC_IN_CP_AREA 1 /* Source data for curr_migr_unit has
258 * already been migrated and must
259 * be recovered from checkpoint area */
261 __u32 rec_status
; /* Status used to determine how to restart
262 * migration in case it aborts
264 __u32 curr_migr_unit
; /* 0..numMigrUnits-1 */
265 __u32 family_num
; /* Family number of MPB
266 * containing the RaidDev
267 * that is migrating */
268 __u32 ascending_migr
; /* True if migrating in increasing
270 __u32 blocks_per_unit
; /* Num disk blocks per unit of operation */
271 __u32 dest_depth_per_unit
; /* Num member blocks each destMap
273 * advances per unit-of-operation */
274 __u32 ckpt_area_pba
; /* Pba of first block of ckpt copy area */
275 __u32 dest_1st_member_lba
; /* First member lba on first
276 * stripe of destination */
277 __u32 num_migr_units
; /* Total num migration units-of-op */
278 __u32 post_migr_vol_cap
; /* Size of volume after
279 * migration completes */
280 __u32 post_migr_vol_cap_hi
; /* Expansion space for LBA64 */
281 __u32 ckpt_read_disk_num
; /* Which member disk in destSubMap[0] the
282 * migration ckpt record was read from
283 * (for recovered migrations) */
284 } __attribute__ ((__packed__
));
289 * 2: metadata does not match
297 struct md_list
*next
;
300 #define pr_vrb(fmt, arg...) (void) (verbose && pr_err(fmt, ##arg))
302 static __u8
migr_type(struct imsm_dev
*dev
)
304 if (dev
->vol
.migr_type
== MIGR_VERIFY
&&
305 dev
->status
& DEV_VERIFY_AND_FIX
)
308 return dev
->vol
.migr_type
;
311 static void set_migr_type(struct imsm_dev
*dev
, __u8 migr_type
)
313 /* for compatibility with older oroms convert MIGR_REPAIR, into
314 * MIGR_VERIFY w/ DEV_VERIFY_AND_FIX status
316 if (migr_type
== MIGR_REPAIR
) {
317 dev
->vol
.migr_type
= MIGR_VERIFY
;
318 dev
->status
|= DEV_VERIFY_AND_FIX
;
320 dev
->vol
.migr_type
= migr_type
;
321 dev
->status
&= ~DEV_VERIFY_AND_FIX
;
325 static unsigned int sector_count(__u32 bytes
, unsigned int sector_size
)
327 return ROUND_UP(bytes
, sector_size
) / sector_size
;
330 static unsigned int mpb_sectors(struct imsm_super
*mpb
,
331 unsigned int sector_size
)
333 return sector_count(__le32_to_cpu(mpb
->mpb_size
), sector_size
);
337 struct imsm_dev
*dev
;
338 struct intel_dev
*next
;
343 enum sys_dev_type type
;
346 struct intel_hba
*next
;
353 /* internal representation of IMSM metadata */
356 void *buf
; /* O_DIRECT buffer for reading/writing metadata */
357 struct imsm_super
*anchor
; /* immovable parameters */
360 void *migr_rec_buf
; /* buffer for I/O operations */
361 struct migr_record
*migr_rec
; /* migration record */
363 int clean_migration_record_by_mdmon
; /* when reshape is switched to next
364 array, it indicates that mdmon is allowed to clean migration
366 size_t len
; /* size of the 'buf' allocation */
367 size_t extra_space
; /* extra space in 'buf' that is not used yet */
368 void *next_buf
; /* for realloc'ing buf from the manager */
370 int updates_pending
; /* count of pending updates for mdmon */
371 int current_vol
; /* index of raid device undergoing creation */
372 unsigned long long create_offset
; /* common start for 'current_vol' */
373 __u32 random
; /* random data for seeding new family numbers */
374 struct intel_dev
*devlist
;
375 unsigned int sector_size
; /* sector size of used member drives */
379 __u8 serial
[MAX_RAID_SERIAL_LEN
];
382 struct imsm_disk disk
;
385 struct extent
*e
; /* for determining freespace @ create */
386 int raiddisk
; /* slot to fill in autolayout */
388 } *disks
, *current_disk
;
389 struct dl
*disk_mgmt_list
; /* list of disks to add/remove while mdmon
391 struct dl
*missing
; /* disks removed while we weren't looking */
392 struct bbm_log
*bbm_log
;
393 struct intel_hba
*hba
; /* device path of the raid controller for this metadata */
394 const struct imsm_orom
*orom
; /* platform firmware support */
395 struct intel_super
*next
; /* (temp) list for disambiguating family_num */
396 struct md_bb bb
; /* memory for get_bad_blocks call */
400 struct imsm_disk disk
;
401 #define IMSM_UNKNOWN_OWNER (-1)
403 struct intel_disk
*next
;
407 unsigned long long start
, size
;
410 /* definitions of reshape process types */
411 enum imsm_reshape_type
{
417 /* definition of messages passed to imsm_process_update */
418 enum imsm_update_type
{
419 update_activate_spare
,
423 update_add_remove_disk
,
424 update_reshape_container_disks
,
425 update_reshape_migration
,
427 update_general_migration_checkpoint
,
429 update_prealloc_badblocks_mem
,
432 struct imsm_update_activate_spare
{
433 enum imsm_update_type type
;
437 struct imsm_update_activate_spare
*next
;
443 unsigned long long size
;
450 enum takeover_direction
{
454 struct imsm_update_takeover
{
455 enum imsm_update_type type
;
457 enum takeover_direction direction
;
460 struct imsm_update_reshape
{
461 enum imsm_update_type type
;
465 int new_disks
[1]; /* new_raid_disks - old_raid_disks makedev number */
468 struct imsm_update_reshape_migration
{
469 enum imsm_update_type type
;
472 /* fields for array migration changes
479 int new_disks
[1]; /* new_raid_disks - old_raid_disks makedev number */
482 struct imsm_update_size_change
{
483 enum imsm_update_type type
;
488 struct imsm_update_general_migration_checkpoint
{
489 enum imsm_update_type type
;
490 __u32 curr_migr_unit
;
494 __u8 serial
[MAX_RAID_SERIAL_LEN
];
497 struct imsm_update_create_array
{
498 enum imsm_update_type type
;
503 struct imsm_update_kill_array
{
504 enum imsm_update_type type
;
508 struct imsm_update_rename_array
{
509 enum imsm_update_type type
;
510 __u8 name
[MAX_RAID_SERIAL_LEN
];
514 struct imsm_update_add_remove_disk
{
515 enum imsm_update_type type
;
518 struct imsm_update_prealloc_bb_mem
{
519 enum imsm_update_type type
;
522 static const char *_sys_dev_type
[] = {
523 [SYS_DEV_UNKNOWN
] = "Unknown",
524 [SYS_DEV_SAS
] = "SAS",
525 [SYS_DEV_SATA
] = "SATA",
526 [SYS_DEV_NVME
] = "NVMe",
527 [SYS_DEV_VMD
] = "VMD"
530 const char *get_sys_dev_type(enum sys_dev_type type
)
532 if (type
>= SYS_DEV_MAX
)
533 type
= SYS_DEV_UNKNOWN
;
535 return _sys_dev_type
[type
];
538 static struct intel_hba
* alloc_intel_hba(struct sys_dev
*device
)
540 struct intel_hba
*result
= xmalloc(sizeof(*result
));
542 result
->type
= device
->type
;
543 result
->path
= xstrdup(device
->path
);
545 if (result
->path
&& (result
->pci_id
= strrchr(result
->path
, '/')) != NULL
)
551 static struct intel_hba
* find_intel_hba(struct intel_hba
*hba
, struct sys_dev
*device
)
553 struct intel_hba
*result
;
555 for (result
= hba
; result
; result
= result
->next
) {
556 if (result
->type
== device
->type
&& strcmp(result
->path
, device
->path
) == 0)
562 static int attach_hba_to_super(struct intel_super
*super
, struct sys_dev
*device
)
564 struct intel_hba
*hba
;
566 /* check if disk attached to Intel HBA */
567 hba
= find_intel_hba(super
->hba
, device
);
570 /* Check if HBA is already attached to super */
571 if (super
->hba
== NULL
) {
572 super
->hba
= alloc_intel_hba(device
);
577 /* Intel metadata allows for all disks attached to the same type HBA.
578 * Do not support HBA types mixing
580 if (device
->type
!= hba
->type
)
583 /* Multiple same type HBAs can be used if they share the same OROM */
584 const struct imsm_orom
*device_orom
= get_orom_by_device_id(device
->dev_id
);
586 if (device_orom
!= super
->orom
)
592 hba
->next
= alloc_intel_hba(device
);
596 static struct sys_dev
* find_disk_attached_hba(int fd
, const char *devname
)
598 struct sys_dev
*list
, *elem
;
601 if ((list
= find_intel_devices()) == NULL
)
605 disk_path
= (char *) devname
;
607 disk_path
= diskfd_to_devpath(fd
);
612 for (elem
= list
; elem
; elem
= elem
->next
)
613 if (path_attached_to_hba(disk_path
, elem
->path
))
616 if (disk_path
!= devname
)
622 static int find_intel_hba_capability(int fd
, struct intel_super
*super
,
625 static struct supertype
*match_metadata_desc_imsm(char *arg
)
627 struct supertype
*st
;
629 if (strcmp(arg
, "imsm") != 0 &&
630 strcmp(arg
, "default") != 0
634 st
= xcalloc(1, sizeof(*st
));
635 st
->ss
= &super_imsm
;
636 st
->max_devs
= IMSM_MAX_DEVICES
;
637 st
->minor_version
= 0;
643 static __u8
*get_imsm_version(struct imsm_super
*mpb
)
645 return &mpb
->sig
[MPB_SIG_LEN
];
649 /* retrieve a disk directly from the anchor when the anchor is known to be
650 * up-to-date, currently only at load time
652 static struct imsm_disk
*__get_imsm_disk(struct imsm_super
*mpb
, __u8 index
)
654 if (index
>= mpb
->num_disks
)
656 return &mpb
->disk
[index
];
659 /* retrieve the disk description based on a index of the disk
662 static struct dl
*get_imsm_dl_disk(struct intel_super
*super
, __u8 index
)
666 for (d
= super
->disks
; d
; d
= d
->next
)
667 if (d
->index
== index
)
672 /* retrieve a disk from the parsed metadata */
673 static struct imsm_disk
*get_imsm_disk(struct intel_super
*super
, __u8 index
)
677 dl
= get_imsm_dl_disk(super
, index
);
684 /* generate a checksum directly from the anchor when the anchor is known to be
685 * up-to-date, currently only at load or write_super after coalescing
687 static __u32
__gen_imsm_checksum(struct imsm_super
*mpb
)
689 __u32 end
= mpb
->mpb_size
/ sizeof(end
);
690 __u32
*p
= (__u32
*) mpb
;
694 sum
+= __le32_to_cpu(*p
);
698 return sum
- __le32_to_cpu(mpb
->check_sum
);
701 static size_t sizeof_imsm_map(struct imsm_map
*map
)
703 return sizeof(struct imsm_map
) + sizeof(__u32
) * (map
->num_members
- 1);
706 struct imsm_map
*get_imsm_map(struct imsm_dev
*dev
, int second_map
)
708 /* A device can have 2 maps if it is in the middle of a migration.
710 * MAP_0 - we return the first map
711 * MAP_1 - we return the second map if it exists, else NULL
712 * MAP_X - we return the second map if it exists, else the first
714 struct imsm_map
*map
= &dev
->vol
.map
[0];
715 struct imsm_map
*map2
= NULL
;
717 if (dev
->vol
.migr_state
)
718 map2
= (void *)map
+ sizeof_imsm_map(map
);
720 switch (second_map
) {
737 /* return the size of the device.
738 * migr_state increases the returned size if map[0] were to be duplicated
740 static size_t sizeof_imsm_dev(struct imsm_dev
*dev
, int migr_state
)
742 size_t size
= sizeof(*dev
) - sizeof(struct imsm_map
) +
743 sizeof_imsm_map(get_imsm_map(dev
, MAP_0
));
745 /* migrating means an additional map */
746 if (dev
->vol
.migr_state
)
747 size
+= sizeof_imsm_map(get_imsm_map(dev
, MAP_1
));
749 size
+= sizeof_imsm_map(get_imsm_map(dev
, MAP_0
));
755 /* retrieve disk serial number list from a metadata update */
756 static struct disk_info
*get_disk_info(struct imsm_update_create_array
*update
)
759 struct disk_info
*inf
;
761 inf
= u
+ sizeof(*update
) - sizeof(struct imsm_dev
) +
762 sizeof_imsm_dev(&update
->dev
, 0);
768 static struct imsm_dev
*__get_imsm_dev(struct imsm_super
*mpb
, __u8 index
)
774 if (index
>= mpb
->num_raid_devs
)
777 /* devices start after all disks */
778 offset
= ((void *) &mpb
->disk
[mpb
->num_disks
]) - _mpb
;
780 for (i
= 0; i
<= index
; i
++)
782 return _mpb
+ offset
;
784 offset
+= sizeof_imsm_dev(_mpb
+ offset
, 0);
789 static struct imsm_dev
*get_imsm_dev(struct intel_super
*super
, __u8 index
)
791 struct intel_dev
*dv
;
793 if (index
>= super
->anchor
->num_raid_devs
)
795 for (dv
= super
->devlist
; dv
; dv
= dv
->next
)
796 if (dv
->index
== index
)
801 static inline unsigned long long __le48_to_cpu(const struct bbm_log_block_addr
804 return ((((__u64
)__le32_to_cpu(addr
->dw1
)) << 16) |
805 __le16_to_cpu(addr
->w1
));
808 static inline struct bbm_log_block_addr
__cpu_to_le48(unsigned long long sec
)
810 struct bbm_log_block_addr addr
;
812 addr
.w1
= __cpu_to_le16((__u16
)(sec
& 0xffff));
813 addr
.dw1
= __cpu_to_le32((__u32
)(sec
>> 16) & 0xffffffff);
818 /* get size of the bbm log */
819 static __u32
get_imsm_bbm_log_size(struct bbm_log
*log
)
821 if (!log
|| log
->entry_count
== 0)
824 return sizeof(log
->signature
) +
825 sizeof(log
->entry_count
) +
826 log
->entry_count
* sizeof(struct bbm_log_entry
);
829 /* check if bad block is not partially stored in bbm log */
830 static int is_stored_in_bbm(struct bbm_log
*log
, const __u8 idx
, const unsigned
831 long long sector
, const int length
, __u32
*pos
)
835 for (i
= *pos
; i
< log
->entry_count
; i
++) {
836 struct bbm_log_entry
*entry
= &log
->marked_block_entries
[i
];
837 unsigned long long bb_start
;
838 unsigned long long bb_end
;
840 bb_start
= __le48_to_cpu(&entry
->defective_block_start
);
841 bb_end
= bb_start
+ (entry
->marked_count
+ 1);
843 if ((entry
->disk_ordinal
== idx
) && (bb_start
>= sector
) &&
844 (bb_end
<= sector
+ length
)) {
852 /* record new bad block in bbm log */
853 static int record_new_badblock(struct bbm_log
*log
, const __u8 idx
, unsigned
854 long long sector
, int length
)
858 struct bbm_log_entry
*entry
= NULL
;
860 while (is_stored_in_bbm(log
, idx
, sector
, length
, &pos
)) {
861 struct bbm_log_entry
*e
= &log
->marked_block_entries
[pos
];
863 if ((e
->marked_count
+ 1 == BBM_LOG_MAX_LBA_ENTRY_VAL
) &&
864 (__le48_to_cpu(&e
->defective_block_start
) == sector
)) {
865 sector
+= BBM_LOG_MAX_LBA_ENTRY_VAL
;
866 length
-= BBM_LOG_MAX_LBA_ENTRY_VAL
;
875 int cnt
= (length
<= BBM_LOG_MAX_LBA_ENTRY_VAL
) ? length
:
876 BBM_LOG_MAX_LBA_ENTRY_VAL
;
877 entry
->defective_block_start
= __cpu_to_le48(sector
);
878 entry
->marked_count
= cnt
- 1;
885 new_bb
= ROUND_UP(length
, BBM_LOG_MAX_LBA_ENTRY_VAL
) /
886 BBM_LOG_MAX_LBA_ENTRY_VAL
;
887 if (log
->entry_count
+ new_bb
> BBM_LOG_MAX_ENTRIES
)
891 int cnt
= (length
<= BBM_LOG_MAX_LBA_ENTRY_VAL
) ? length
:
892 BBM_LOG_MAX_LBA_ENTRY_VAL
;
893 struct bbm_log_entry
*entry
=
894 &log
->marked_block_entries
[log
->entry_count
];
896 entry
->defective_block_start
= __cpu_to_le48(sector
);
897 entry
->marked_count
= cnt
- 1;
898 entry
->disk_ordinal
= idx
;
909 /* clear all bad blocks for given disk */
910 static void clear_disk_badblocks(struct bbm_log
*log
, const __u8 idx
)
914 while (i
< log
->entry_count
) {
915 struct bbm_log_entry
*entries
= log
->marked_block_entries
;
917 if (entries
[i
].disk_ordinal
== idx
) {
918 if (i
< log
->entry_count
- 1)
919 entries
[i
] = entries
[log
->entry_count
- 1];
927 /* clear given bad block */
928 static int clear_badblock(struct bbm_log
*log
, const __u8 idx
, const unsigned
929 long long sector
, const int length
) {
932 while (i
< log
->entry_count
) {
933 struct bbm_log_entry
*entries
= log
->marked_block_entries
;
935 if ((entries
[i
].disk_ordinal
== idx
) &&
936 (__le48_to_cpu(&entries
[i
].defective_block_start
) ==
937 sector
) && (entries
[i
].marked_count
+ 1 == length
)) {
938 if (i
< log
->entry_count
- 1)
939 entries
[i
] = entries
[log
->entry_count
- 1];
948 #endif /* MDASSEMBLE */
950 /* allocate and load BBM log from metadata */
951 static int load_bbm_log(struct intel_super
*super
)
953 struct imsm_super
*mpb
= super
->anchor
;
954 __u32 bbm_log_size
= __le32_to_cpu(mpb
->bbm_log_size
);
956 super
->bbm_log
= xcalloc(1, sizeof(struct bbm_log
));
961 struct bbm_log
*log
= (void *)mpb
+
962 __le32_to_cpu(mpb
->mpb_size
) - bbm_log_size
;
966 if (bbm_log_size
< sizeof(log
->signature
) +
967 sizeof(log
->entry_count
))
970 entry_count
= __le32_to_cpu(log
->entry_count
);
971 if ((__le32_to_cpu(log
->signature
) != BBM_LOG_SIGNATURE
) ||
972 (entry_count
> BBM_LOG_MAX_ENTRIES
))
976 sizeof(log
->signature
) + sizeof(log
->entry_count
) +
977 entry_count
* sizeof(struct bbm_log_entry
))
980 memcpy(super
->bbm_log
, log
, bbm_log_size
);
982 super
->bbm_log
->signature
= __cpu_to_le32(BBM_LOG_SIGNATURE
);
983 super
->bbm_log
->entry_count
= 0;
989 /* checks if bad block is within volume boundaries */
990 static int is_bad_block_in_volume(const struct bbm_log_entry
*entry
,
991 const unsigned long long start_sector
,
992 const unsigned long long size
)
994 unsigned long long bb_start
;
995 unsigned long long bb_end
;
997 bb_start
= __le48_to_cpu(&entry
->defective_block_start
);
998 bb_end
= bb_start
+ (entry
->marked_count
+ 1);
1000 if (((bb_start
>= start_sector
) && (bb_start
< start_sector
+ size
)) ||
1001 ((bb_end
>= start_sector
) && (bb_end
<= start_sector
+ size
)))
1007 /* get list of bad blocks on a drive for a volume */
1008 static void get_volume_badblocks(const struct bbm_log
*log
, const __u8 idx
,
1009 const unsigned long long start_sector
,
1010 const unsigned long long size
,
1016 for (i
= 0; i
< log
->entry_count
; i
++) {
1017 const struct bbm_log_entry
*ent
=
1018 &log
->marked_block_entries
[i
];
1019 struct md_bb_entry
*bb
;
1021 if ((ent
->disk_ordinal
== idx
) &&
1022 is_bad_block_in_volume(ent
, start_sector
, size
)) {
1024 if (!bbs
->entries
) {
1025 bbs
->entries
= xmalloc(BBM_LOG_MAX_ENTRIES
*
1031 bb
= &bbs
->entries
[count
++];
1032 bb
->sector
= __le48_to_cpu(&ent
->defective_block_start
);
1033 bb
->length
= ent
->marked_count
+ 1;
1041 * == MAP_0 get first map
1042 * == MAP_1 get second map
1043 * == MAP_X than get map according to the current migr_state
1045 static __u32
get_imsm_ord_tbl_ent(struct imsm_dev
*dev
,
1049 struct imsm_map
*map
;
1051 map
= get_imsm_map(dev
, second_map
);
1053 /* top byte identifies disk under rebuild */
1054 return __le32_to_cpu(map
->disk_ord_tbl
[slot
]);
1057 #define ord_to_idx(ord) (((ord) << 8) >> 8)
1058 static __u32
get_imsm_disk_idx(struct imsm_dev
*dev
, int slot
, int second_map
)
1060 __u32 ord
= get_imsm_ord_tbl_ent(dev
, slot
, second_map
);
1062 return ord_to_idx(ord
);
1065 static void set_imsm_ord_tbl_ent(struct imsm_map
*map
, int slot
, __u32 ord
)
1067 map
->disk_ord_tbl
[slot
] = __cpu_to_le32(ord
);
1070 static int get_imsm_disk_slot(struct imsm_map
*map
, unsigned idx
)
1075 for (slot
= 0; slot
< map
->num_members
; slot
++) {
1076 ord
= __le32_to_cpu(map
->disk_ord_tbl
[slot
]);
1077 if (ord_to_idx(ord
) == idx
)
1084 static int get_imsm_raid_level(struct imsm_map
*map
)
1086 if (map
->raid_level
== 1) {
1087 if (map
->num_members
== 2)
1093 return map
->raid_level
;
1096 static int cmp_extent(const void *av
, const void *bv
)
1098 const struct extent
*a
= av
;
1099 const struct extent
*b
= bv
;
1100 if (a
->start
< b
->start
)
1102 if (a
->start
> b
->start
)
1107 static int count_memberships(struct dl
*dl
, struct intel_super
*super
)
1109 int memberships
= 0;
1112 for (i
= 0; i
< super
->anchor
->num_raid_devs
; i
++) {
1113 struct imsm_dev
*dev
= get_imsm_dev(super
, i
);
1114 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
1116 if (get_imsm_disk_slot(map
, dl
->index
) >= 0)
1123 static __u32
imsm_min_reserved_sectors(struct intel_super
*super
);
1125 static int split_ull(unsigned long long n
, __u32
*lo
, __u32
*hi
)
1127 if (lo
== 0 || hi
== 0)
1129 *lo
= __le32_to_cpu((unsigned)n
);
1130 *hi
= __le32_to_cpu((unsigned)(n
>> 32));
1134 static unsigned long long join_u32(__u32 lo
, __u32 hi
)
1136 return (unsigned long long)__le32_to_cpu(lo
) |
1137 (((unsigned long long)__le32_to_cpu(hi
)) << 32);
1140 static unsigned long long total_blocks(struct imsm_disk
*disk
)
1144 return join_u32(disk
->total_blocks_lo
, disk
->total_blocks_hi
);
1147 static unsigned long long pba_of_lba0(struct imsm_map
*map
)
1151 return join_u32(map
->pba_of_lba0_lo
, map
->pba_of_lba0_hi
);
1154 static unsigned long long blocks_per_member(struct imsm_map
*map
)
1158 return join_u32(map
->blocks_per_member_lo
, map
->blocks_per_member_hi
);
1161 static unsigned long long num_data_stripes(struct imsm_map
*map
)
1165 return join_u32(map
->num_data_stripes_lo
, map
->num_data_stripes_hi
);
1168 static void set_total_blocks(struct imsm_disk
*disk
, unsigned long long n
)
1170 split_ull(n
, &disk
->total_blocks_lo
, &disk
->total_blocks_hi
);
1173 static void set_pba_of_lba0(struct imsm_map
*map
, unsigned long long n
)
1175 split_ull(n
, &map
->pba_of_lba0_lo
, &map
->pba_of_lba0_hi
);
1178 static void set_blocks_per_member(struct imsm_map
*map
, unsigned long long n
)
1180 split_ull(n
, &map
->blocks_per_member_lo
, &map
->blocks_per_member_hi
);
1183 static void set_num_data_stripes(struct imsm_map
*map
, unsigned long long n
)
1185 split_ull(n
, &map
->num_data_stripes_lo
, &map
->num_data_stripes_hi
);
1188 static struct extent
*get_extents(struct intel_super
*super
, struct dl
*dl
)
1190 /* find a list of used extents on the given physical device */
1191 struct extent
*rv
, *e
;
1193 int memberships
= count_memberships(dl
, super
);
1196 /* trim the reserved area for spares, so they can join any array
1197 * regardless of whether the OROM has assigned sectors from the
1198 * IMSM_RESERVED_SECTORS region
1200 if (dl
->index
== -1)
1201 reservation
= imsm_min_reserved_sectors(super
);
1203 reservation
= MPB_SECTOR_CNT
+ IMSM_RESERVED_SECTORS
;
1205 rv
= xcalloc(sizeof(struct extent
), (memberships
+ 1));
1208 for (i
= 0; i
< super
->anchor
->num_raid_devs
; i
++) {
1209 struct imsm_dev
*dev
= get_imsm_dev(super
, i
);
1210 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
1212 if (get_imsm_disk_slot(map
, dl
->index
) >= 0) {
1213 e
->start
= pba_of_lba0(map
);
1214 e
->size
= blocks_per_member(map
);
1218 qsort(rv
, memberships
, sizeof(*rv
), cmp_extent
);
1220 /* determine the start of the metadata
1221 * when no raid devices are defined use the default
1222 * ...otherwise allow the metadata to truncate the value
1223 * as is the case with older versions of imsm
1226 struct extent
*last
= &rv
[memberships
- 1];
1227 unsigned long long remainder
;
1229 remainder
= total_blocks(&dl
->disk
) - (last
->start
+ last
->size
);
1230 /* round down to 1k block to satisfy precision of the kernel
1234 /* make sure remainder is still sane */
1235 if (remainder
< (unsigned)ROUND_UP(super
->len
, 512) >> 9)
1236 remainder
= ROUND_UP(super
->len
, 512) >> 9;
1237 if (reservation
> remainder
)
1238 reservation
= remainder
;
1240 e
->start
= total_blocks(&dl
->disk
) - reservation
;
1245 /* try to determine how much space is reserved for metadata from
1246 * the last get_extents() entry, otherwise fallback to the
1249 static __u32
imsm_reserved_sectors(struct intel_super
*super
, struct dl
*dl
)
1255 /* for spares just return a minimal reservation which will grow
1256 * once the spare is picked up by an array
1258 if (dl
->index
== -1)
1259 return MPB_SECTOR_CNT
;
1261 e
= get_extents(super
, dl
);
1263 return MPB_SECTOR_CNT
+ IMSM_RESERVED_SECTORS
;
1265 /* scroll to last entry */
1266 for (i
= 0; e
[i
].size
; i
++)
1269 rv
= total_blocks(&dl
->disk
) - e
[i
].start
;
1276 static int is_spare(struct imsm_disk
*disk
)
1278 return (disk
->status
& SPARE_DISK
) == SPARE_DISK
;
1281 static int is_configured(struct imsm_disk
*disk
)
1283 return (disk
->status
& CONFIGURED_DISK
) == CONFIGURED_DISK
;
1286 static int is_failed(struct imsm_disk
*disk
)
1288 return (disk
->status
& FAILED_DISK
) == FAILED_DISK
;
1291 /* try to determine how much space is reserved for metadata from
1292 * the last get_extents() entry on the smallest active disk,
1293 * otherwise fallback to the default
1295 static __u32
imsm_min_reserved_sectors(struct intel_super
*super
)
1299 unsigned long long min_active
;
1301 __u32 rv
= MPB_SECTOR_CNT
+ IMSM_RESERVED_SECTORS
;
1302 struct dl
*dl
, *dl_min
= NULL
;
1308 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
1311 unsigned long long blocks
= total_blocks(&dl
->disk
);
1312 if (blocks
< min_active
|| min_active
== 0) {
1314 min_active
= blocks
;
1320 /* find last lba used by subarrays on the smallest active disk */
1321 e
= get_extents(super
, dl_min
);
1324 for (i
= 0; e
[i
].size
; i
++)
1327 remainder
= min_active
- e
[i
].start
;
1330 /* to give priority to recovery we should not require full
1331 IMSM_RESERVED_SECTORS from the spare */
1332 rv
= MPB_SECTOR_CNT
+ NUM_BLOCKS_DIRTY_STRIPE_REGION
;
1334 /* if real reservation is smaller use that value */
1335 return (remainder
< rv
) ? remainder
: rv
;
1338 /* Return minimum size of a spare that can be used in this array*/
1339 static unsigned long long min_acceptable_spare_size_imsm(struct supertype
*st
)
1341 struct intel_super
*super
= st
->sb
;
1345 unsigned long long rv
= 0;
1349 /* find first active disk in array */
1351 while (dl
&& (is_failed(&dl
->disk
) || dl
->index
== -1))
1355 /* find last lba used by subarrays */
1356 e
= get_extents(super
, dl
);
1359 for (i
= 0; e
[i
].size
; i
++)
1362 rv
= e
[i
-1].start
+ e
[i
-1].size
;
1365 /* add the amount of space needed for metadata */
1366 rv
= rv
+ imsm_min_reserved_sectors(super
);
1371 static int is_gen_migration(struct imsm_dev
*dev
);
1373 #define IMSM_4K_DIV 8
1376 static __u64
blocks_per_migr_unit(struct intel_super
*super
,
1377 struct imsm_dev
*dev
);
1379 static void print_imsm_dev(struct intel_super
*super
,
1380 struct imsm_dev
*dev
,
1386 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
1387 struct imsm_map
*map2
= get_imsm_map(dev
, MAP_1
);
1391 printf("[%.16s]:\n", dev
->volume
);
1392 printf(" UUID : %s\n", uuid
);
1393 printf(" RAID Level : %d", get_imsm_raid_level(map
));
1395 printf(" <-- %d", get_imsm_raid_level(map2
));
1397 printf(" Members : %d", map
->num_members
);
1399 printf(" <-- %d", map2
->num_members
);
1401 printf(" Slots : [");
1402 for (i
= 0; i
< map
->num_members
; i
++) {
1403 ord
= get_imsm_ord_tbl_ent(dev
, i
, MAP_0
);
1404 printf("%s", ord
& IMSM_ORD_REBUILD
? "_" : "U");
1409 for (i
= 0; i
< map2
->num_members
; i
++) {
1410 ord
= get_imsm_ord_tbl_ent(dev
, i
, MAP_1
);
1411 printf("%s", ord
& IMSM_ORD_REBUILD
? "_" : "U");
1416 printf(" Failed disk : ");
1417 if (map
->failed_disk_num
== 0xff)
1420 printf("%i", map
->failed_disk_num
);
1422 slot
= get_imsm_disk_slot(map
, disk_idx
);
1424 ord
= get_imsm_ord_tbl_ent(dev
, slot
, MAP_X
);
1425 printf(" This Slot : %d%s\n", slot
,
1426 ord
& IMSM_ORD_REBUILD
? " (out-of-sync)" : "");
1428 printf(" This Slot : ?\n");
1429 sz
= __le32_to_cpu(dev
->size_high
);
1431 sz
+= __le32_to_cpu(dev
->size_low
);
1432 printf(" Array Size : %llu%s\n", (unsigned long long)sz
,
1433 human_size(sz
* 512));
1434 sz
= blocks_per_member(map
);
1435 printf(" Per Dev Size : %llu%s\n", (unsigned long long)sz
,
1436 human_size(sz
* 512));
1437 printf(" Sector Offset : %llu\n",
1439 printf(" Num Stripes : %llu\n",
1440 num_data_stripes(map
));
1441 printf(" Chunk Size : %u KiB",
1442 __le16_to_cpu(map
->blocks_per_strip
) / 2);
1444 printf(" <-- %u KiB",
1445 __le16_to_cpu(map2
->blocks_per_strip
) / 2);
1447 printf(" Reserved : %d\n", __le32_to_cpu(dev
->reserved_blocks
));
1448 printf(" Migrate State : ");
1449 if (dev
->vol
.migr_state
) {
1450 if (migr_type(dev
) == MIGR_INIT
)
1451 printf("initialize\n");
1452 else if (migr_type(dev
) == MIGR_REBUILD
)
1453 printf("rebuild\n");
1454 else if (migr_type(dev
) == MIGR_VERIFY
)
1456 else if (migr_type(dev
) == MIGR_GEN_MIGR
)
1457 printf("general migration\n");
1458 else if (migr_type(dev
) == MIGR_STATE_CHANGE
)
1459 printf("state change\n");
1460 else if (migr_type(dev
) == MIGR_REPAIR
)
1463 printf("<unknown:%d>\n", migr_type(dev
));
1466 printf(" Map State : %s", map_state_str
[map
->map_state
]);
1467 if (dev
->vol
.migr_state
) {
1468 struct imsm_map
*map
= get_imsm_map(dev
, MAP_1
);
1470 printf(" <-- %s", map_state_str
[map
->map_state
]);
1471 printf("\n Checkpoint : %u ",
1472 __le32_to_cpu(dev
->vol
.curr_migr_unit
));
1473 if (is_gen_migration(dev
) && (slot
> 1 || slot
< 0))
1476 printf("(%llu)", (unsigned long long)
1477 blocks_per_migr_unit(super
, dev
));
1480 printf(" Dirty State : %s\n", dev
->vol
.dirty
? "dirty" : "clean");
1483 static void print_imsm_disk(struct imsm_disk
*disk
,
1486 unsigned int sector_size
) {
1487 char str
[MAX_RAID_SERIAL_LEN
+ 1];
1490 if (index
< -1 || !disk
)
1494 snprintf(str
, MAX_RAID_SERIAL_LEN
+ 1, "%s", disk
->serial
);
1496 printf(" Disk%02d Serial : %s\n", index
, str
);
1498 printf(" Disk Serial : %s\n", str
);
1499 printf(" State :%s%s%s\n", is_spare(disk
) ? " spare" : "",
1500 is_configured(disk
) ? " active" : "",
1501 is_failed(disk
) ? " failed" : "");
1502 printf(" Id : %08x\n", __le32_to_cpu(disk
->scsi_id
));
1503 sz
= total_blocks(disk
) - reserved
;
1504 printf(" Usable Size : %llu%s\n",
1505 (unsigned long long)sz
* 512 / sector_size
,
1506 human_size(sz
* 512));
1509 void convert_to_4k_imsm_migr_rec(struct intel_super
*super
)
1511 struct migr_record
*migr_rec
= super
->migr_rec
;
1513 migr_rec
->blocks_per_unit
/= IMSM_4K_DIV
;
1514 migr_rec
->ckpt_area_pba
/= IMSM_4K_DIV
;
1515 migr_rec
->dest_1st_member_lba
/= IMSM_4K_DIV
;
1516 migr_rec
->dest_depth_per_unit
/= IMSM_4K_DIV
;
1517 split_ull((join_u32(migr_rec
->post_migr_vol_cap
,
1518 migr_rec
->post_migr_vol_cap_hi
) / IMSM_4K_DIV
),
1519 &migr_rec
->post_migr_vol_cap
, &migr_rec
->post_migr_vol_cap_hi
);
1522 void convert_to_4k_imsm_disk(struct imsm_disk
*disk
)
1524 set_total_blocks(disk
, (total_blocks(disk
)/IMSM_4K_DIV
));
1527 void convert_to_4k(struct intel_super
*super
)
1529 struct imsm_super
*mpb
= super
->anchor
;
1530 struct imsm_disk
*disk
;
1532 __u32 bbm_log_size
= __le32_to_cpu(mpb
->bbm_log_size
);
1534 for (i
= 0; i
< mpb
->num_disks
; i
++) {
1535 disk
= __get_imsm_disk(mpb
, i
);
1537 convert_to_4k_imsm_disk(disk
);
1539 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
1540 struct imsm_dev
*dev
= __get_imsm_dev(mpb
, i
);
1541 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
1543 split_ull((join_u32(dev
->size_low
, dev
->size_high
)/IMSM_4K_DIV
),
1544 &dev
->size_low
, &dev
->size_high
);
1545 dev
->vol
.curr_migr_unit
/= IMSM_4K_DIV
;
1548 set_blocks_per_member(map
, blocks_per_member(map
)/IMSM_4K_DIV
);
1549 map
->blocks_per_strip
/= IMSM_4K_DIV
;
1550 set_pba_of_lba0(map
, pba_of_lba0(map
)/IMSM_4K_DIV
);
1552 if (dev
->vol
.migr_state
) {
1554 map
= get_imsm_map(dev
, MAP_1
);
1555 set_blocks_per_member(map
,
1556 blocks_per_member(map
)/IMSM_4K_DIV
);
1557 map
->blocks_per_strip
/= IMSM_4K_DIV
;
1558 set_pba_of_lba0(map
, pba_of_lba0(map
)/IMSM_4K_DIV
);
1562 struct bbm_log
*log
= (void *)mpb
+
1563 __le32_to_cpu(mpb
->mpb_size
) - bbm_log_size
;
1566 for (i
= 0; i
< log
->entry_count
; i
++) {
1567 struct bbm_log_entry
*entry
=
1568 &log
->marked_block_entries
[i
];
1570 __u8 count
= entry
->marked_count
+ 1;
1571 unsigned long long sector
=
1572 __le48_to_cpu(&entry
->defective_block_start
);
1574 entry
->defective_block_start
=
1575 __cpu_to_le48(sector
/IMSM_4K_DIV
);
1576 entry
->marked_count
= max(count
/IMSM_4K_DIV
, 1) - 1;
1580 mpb
->check_sum
= __gen_imsm_checksum(mpb
);
1583 void examine_migr_rec_imsm(struct intel_super
*super
)
1585 struct migr_record
*migr_rec
= super
->migr_rec
;
1586 struct imsm_super
*mpb
= super
->anchor
;
1589 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
1590 struct imsm_dev
*dev
= __get_imsm_dev(mpb
, i
);
1591 struct imsm_map
*map
;
1594 if (is_gen_migration(dev
) == 0)
1597 printf("\nMigration Record Information:");
1599 /* first map under migration */
1600 map
= get_imsm_map(dev
, MAP_0
);
1602 slot
= get_imsm_disk_slot(map
, super
->disks
->index
);
1603 if (map
== NULL
|| slot
> 1 || slot
< 0) {
1604 printf(" Empty\n ");
1605 printf("Examine one of first two disks in array\n");
1608 printf("\n Status : ");
1609 if (__le32_to_cpu(migr_rec
->rec_status
) == UNIT_SRC_NORMAL
)
1612 printf("Contains Data\n");
1613 printf(" Current Unit : %u\n",
1614 __le32_to_cpu(migr_rec
->curr_migr_unit
));
1615 printf(" Family : %u\n",
1616 __le32_to_cpu(migr_rec
->family_num
));
1617 printf(" Ascending : %u\n",
1618 __le32_to_cpu(migr_rec
->ascending_migr
));
1619 printf(" Blocks Per Unit : %u\n",
1620 __le32_to_cpu(migr_rec
->blocks_per_unit
));
1621 printf(" Dest. Depth Per Unit : %u\n",
1622 __le32_to_cpu(migr_rec
->dest_depth_per_unit
));
1623 printf(" Checkpoint Area pba : %u\n",
1624 __le32_to_cpu(migr_rec
->ckpt_area_pba
));
1625 printf(" First member lba : %u\n",
1626 __le32_to_cpu(migr_rec
->dest_1st_member_lba
));
1627 printf(" Total Number of Units : %u\n",
1628 __le32_to_cpu(migr_rec
->num_migr_units
));
1629 printf(" Size of volume : %u\n",
1630 __le32_to_cpu(migr_rec
->post_migr_vol_cap
));
1631 printf(" Expansion space for LBA64 : %u\n",
1632 __le32_to_cpu(migr_rec
->post_migr_vol_cap_hi
));
1633 printf(" Record was read from : %u\n",
1634 __le32_to_cpu(migr_rec
->ckpt_read_disk_num
));
1639 #endif /* MDASSEMBLE */
1641 void convert_from_4k_imsm_migr_rec(struct intel_super
*super
)
1643 struct migr_record
*migr_rec
= super
->migr_rec
;
1645 migr_rec
->blocks_per_unit
*= IMSM_4K_DIV
;
1646 migr_rec
->ckpt_area_pba
*= IMSM_4K_DIV
;
1647 migr_rec
->dest_1st_member_lba
*= IMSM_4K_DIV
;
1648 migr_rec
->dest_depth_per_unit
*= IMSM_4K_DIV
;
1649 split_ull((join_u32(migr_rec
->post_migr_vol_cap
,
1650 migr_rec
->post_migr_vol_cap_hi
) * IMSM_4K_DIV
),
1651 &migr_rec
->post_migr_vol_cap
,
1652 &migr_rec
->post_migr_vol_cap_hi
);
1655 void convert_from_4k(struct intel_super
*super
)
1657 struct imsm_super
*mpb
= super
->anchor
;
1658 struct imsm_disk
*disk
;
1660 __u32 bbm_log_size
= __le32_to_cpu(mpb
->bbm_log_size
);
1662 for (i
= 0; i
< mpb
->num_disks
; i
++) {
1663 disk
= __get_imsm_disk(mpb
, i
);
1665 set_total_blocks(disk
, (total_blocks(disk
)*IMSM_4K_DIV
));
1668 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
1669 struct imsm_dev
*dev
= __get_imsm_dev(mpb
, i
);
1670 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
1672 split_ull((join_u32(dev
->size_low
, dev
->size_high
)*IMSM_4K_DIV
),
1673 &dev
->size_low
, &dev
->size_high
);
1674 dev
->vol
.curr_migr_unit
*= IMSM_4K_DIV
;
1677 set_blocks_per_member(map
, blocks_per_member(map
)*IMSM_4K_DIV
);
1678 map
->blocks_per_strip
*= IMSM_4K_DIV
;
1679 set_pba_of_lba0(map
, pba_of_lba0(map
)*IMSM_4K_DIV
);
1681 if (dev
->vol
.migr_state
) {
1683 map
= get_imsm_map(dev
, MAP_1
);
1684 set_blocks_per_member(map
,
1685 blocks_per_member(map
)*IMSM_4K_DIV
);
1686 map
->blocks_per_strip
*= IMSM_4K_DIV
;
1687 set_pba_of_lba0(map
, pba_of_lba0(map
)*IMSM_4K_DIV
);
1691 struct bbm_log
*log
= (void *)mpb
+
1692 __le32_to_cpu(mpb
->mpb_size
) - bbm_log_size
;
1695 for (i
= 0; i
< log
->entry_count
; i
++) {
1696 struct bbm_log_entry
*entry
=
1697 &log
->marked_block_entries
[i
];
1699 __u8 count
= entry
->marked_count
+ 1;
1700 unsigned long long sector
=
1701 __le48_to_cpu(&entry
->defective_block_start
);
1703 entry
->defective_block_start
=
1704 __cpu_to_le48(sector
*IMSM_4K_DIV
);
1705 entry
->marked_count
= count
*IMSM_4K_DIV
- 1;
1709 mpb
->check_sum
= __gen_imsm_checksum(mpb
);
1712 /*******************************************************************************
1713 * function: imsm_check_attributes
1714 * Description: Function checks if features represented by attributes flags
1715 * are supported by mdadm.
1717 * attributes - Attributes read from metadata
1719 * 0 - passed attributes contains unsupported features flags
1720 * 1 - all features are supported
1721 ******************************************************************************/
1722 static int imsm_check_attributes(__u32 attributes
)
1725 __u32 not_supported
= MPB_ATTRIB_SUPPORTED
^0xffffffff;
1727 not_supported
&= ~MPB_ATTRIB_IGNORED
;
1729 not_supported
&= attributes
;
1730 if (not_supported
) {
1731 pr_err("(IMSM): Unsupported attributes : %x\n",
1732 (unsigned)__le32_to_cpu(not_supported
));
1733 if (not_supported
& MPB_ATTRIB_CHECKSUM_VERIFY
) {
1734 dprintf("\t\tMPB_ATTRIB_CHECKSUM_VERIFY \n");
1735 not_supported
^= MPB_ATTRIB_CHECKSUM_VERIFY
;
1737 if (not_supported
& MPB_ATTRIB_2TB
) {
1738 dprintf("\t\tMPB_ATTRIB_2TB\n");
1739 not_supported
^= MPB_ATTRIB_2TB
;
1741 if (not_supported
& MPB_ATTRIB_RAID0
) {
1742 dprintf("\t\tMPB_ATTRIB_RAID0\n");
1743 not_supported
^= MPB_ATTRIB_RAID0
;
1745 if (not_supported
& MPB_ATTRIB_RAID1
) {
1746 dprintf("\t\tMPB_ATTRIB_RAID1\n");
1747 not_supported
^= MPB_ATTRIB_RAID1
;
1749 if (not_supported
& MPB_ATTRIB_RAID10
) {
1750 dprintf("\t\tMPB_ATTRIB_RAID10\n");
1751 not_supported
^= MPB_ATTRIB_RAID10
;
1753 if (not_supported
& MPB_ATTRIB_RAID1E
) {
1754 dprintf("\t\tMPB_ATTRIB_RAID1E\n");
1755 not_supported
^= MPB_ATTRIB_RAID1E
;
1757 if (not_supported
& MPB_ATTRIB_RAID5
) {
1758 dprintf("\t\tMPB_ATTRIB_RAID5\n");
1759 not_supported
^= MPB_ATTRIB_RAID5
;
1761 if (not_supported
& MPB_ATTRIB_RAIDCNG
) {
1762 dprintf("\t\tMPB_ATTRIB_RAIDCNG\n");
1763 not_supported
^= MPB_ATTRIB_RAIDCNG
;
1765 if (not_supported
& MPB_ATTRIB_BBM
) {
1766 dprintf("\t\tMPB_ATTRIB_BBM\n");
1767 not_supported
^= MPB_ATTRIB_BBM
;
1769 if (not_supported
& MPB_ATTRIB_CHECKSUM_VERIFY
) {
1770 dprintf("\t\tMPB_ATTRIB_CHECKSUM_VERIFY (== MPB_ATTRIB_LEGACY)\n");
1771 not_supported
^= MPB_ATTRIB_CHECKSUM_VERIFY
;
1773 if (not_supported
& MPB_ATTRIB_EXP_STRIPE_SIZE
) {
1774 dprintf("\t\tMPB_ATTRIB_EXP_STRIP_SIZE\n");
1775 not_supported
^= MPB_ATTRIB_EXP_STRIPE_SIZE
;
1777 if (not_supported
& MPB_ATTRIB_2TB_DISK
) {
1778 dprintf("\t\tMPB_ATTRIB_2TB_DISK\n");
1779 not_supported
^= MPB_ATTRIB_2TB_DISK
;
1781 if (not_supported
& MPB_ATTRIB_NEVER_USE2
) {
1782 dprintf("\t\tMPB_ATTRIB_NEVER_USE2\n");
1783 not_supported
^= MPB_ATTRIB_NEVER_USE2
;
1785 if (not_supported
& MPB_ATTRIB_NEVER_USE
) {
1786 dprintf("\t\tMPB_ATTRIB_NEVER_USE\n");
1787 not_supported
^= MPB_ATTRIB_NEVER_USE
;
1791 dprintf("(IMSM): Unknown attributes : %x\n", not_supported
);
1800 static void getinfo_super_imsm(struct supertype
*st
, struct mdinfo
*info
, char *map
);
1802 static void examine_super_imsm(struct supertype
*st
, char *homehost
)
1804 struct intel_super
*super
= st
->sb
;
1805 struct imsm_super
*mpb
= super
->anchor
;
1806 char str
[MAX_SIGNATURE_LENGTH
];
1811 __u32 reserved
= imsm_reserved_sectors(super
, super
->disks
);
1814 strncpy(str
, (char *)mpb
->sig
, MPB_SIG_LEN
);
1815 str
[MPB_SIG_LEN
-1] = '\0';
1816 printf(" Magic : %s\n", str
);
1817 snprintf(str
, strlen(MPB_VERSION_RAID0
), "%s", get_imsm_version(mpb
));
1818 printf(" Version : %s\n", get_imsm_version(mpb
));
1819 printf(" Orig Family : %08x\n", __le32_to_cpu(mpb
->orig_family_num
));
1820 printf(" Family : %08x\n", __le32_to_cpu(mpb
->family_num
));
1821 printf(" Generation : %08x\n", __le32_to_cpu(mpb
->generation_num
));
1822 printf(" Attributes : ");
1823 if (imsm_check_attributes(mpb
->attributes
))
1824 printf("All supported\n");
1826 printf("not supported\n");
1827 getinfo_super_imsm(st
, &info
, NULL
);
1828 fname_from_uuid(st
, &info
, nbuf
, ':');
1829 printf(" UUID : %s\n", nbuf
+ 5);
1830 sum
= __le32_to_cpu(mpb
->check_sum
);
1831 printf(" Checksum : %08x %s\n", sum
,
1832 __gen_imsm_checksum(mpb
) == sum
? "correct" : "incorrect");
1833 printf(" MPB Sectors : %d\n", mpb_sectors(mpb
, super
->sector_size
));
1834 printf(" Disks : %d\n", mpb
->num_disks
);
1835 printf(" RAID Devices : %d\n", mpb
->num_raid_devs
);
1836 print_imsm_disk(__get_imsm_disk(mpb
, super
->disks
->index
),
1837 super
->disks
->index
, reserved
, super
->sector_size
);
1838 if (get_imsm_bbm_log_size(super
->bbm_log
)) {
1839 struct bbm_log
*log
= super
->bbm_log
;
1842 printf("Bad Block Management Log:\n");
1843 printf(" Log Size : %d\n", __le32_to_cpu(mpb
->bbm_log_size
));
1844 printf(" Signature : %x\n", __le32_to_cpu(log
->signature
));
1845 printf(" Entry Count : %d\n", __le32_to_cpu(log
->entry_count
));
1847 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
1849 struct imsm_dev
*dev
= __get_imsm_dev(mpb
, i
);
1851 super
->current_vol
= i
;
1852 getinfo_super_imsm(st
, &info
, NULL
);
1853 fname_from_uuid(st
, &info
, nbuf
, ':');
1854 print_imsm_dev(super
, dev
, nbuf
+ 5, super
->disks
->index
);
1856 for (i
= 0; i
< mpb
->num_disks
; i
++) {
1857 if (i
== super
->disks
->index
)
1859 print_imsm_disk(__get_imsm_disk(mpb
, i
), i
, reserved
,
1860 super
->sector_size
);
1863 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
1864 if (dl
->index
== -1)
1865 print_imsm_disk(&dl
->disk
, -1, reserved
,
1866 super
->sector_size
);
1868 examine_migr_rec_imsm(super
);
1871 static void brief_examine_super_imsm(struct supertype
*st
, int verbose
)
1873 /* We just write a generic IMSM ARRAY entry */
1876 struct intel_super
*super
= st
->sb
;
1878 if (!super
->anchor
->num_raid_devs
) {
1879 printf("ARRAY metadata=imsm\n");
1883 getinfo_super_imsm(st
, &info
, NULL
);
1884 fname_from_uuid(st
, &info
, nbuf
, ':');
1885 printf("ARRAY metadata=imsm UUID=%s\n", nbuf
+ 5);
1888 static void brief_examine_subarrays_imsm(struct supertype
*st
, int verbose
)
1890 /* We just write a generic IMSM ARRAY entry */
1894 struct intel_super
*super
= st
->sb
;
1897 if (!super
->anchor
->num_raid_devs
)
1900 getinfo_super_imsm(st
, &info
, NULL
);
1901 fname_from_uuid(st
, &info
, nbuf
, ':');
1902 for (i
= 0; i
< super
->anchor
->num_raid_devs
; i
++) {
1903 struct imsm_dev
*dev
= get_imsm_dev(super
, i
);
1905 super
->current_vol
= i
;
1906 getinfo_super_imsm(st
, &info
, NULL
);
1907 fname_from_uuid(st
, &info
, nbuf1
, ':');
1908 printf("ARRAY /dev/md/%.16s container=%s member=%d UUID=%s\n",
1909 dev
->volume
, nbuf
+ 5, i
, nbuf1
+ 5);
1913 static void export_examine_super_imsm(struct supertype
*st
)
1915 struct intel_super
*super
= st
->sb
;
1916 struct imsm_super
*mpb
= super
->anchor
;
1920 getinfo_super_imsm(st
, &info
, NULL
);
1921 fname_from_uuid(st
, &info
, nbuf
, ':');
1922 printf("MD_METADATA=imsm\n");
1923 printf("MD_LEVEL=container\n");
1924 printf("MD_UUID=%s\n", nbuf
+5);
1925 printf("MD_DEVICES=%u\n", mpb
->num_disks
);
1928 static int copy_metadata_imsm(struct supertype
*st
, int from
, int to
)
1930 /* The second last sector of the device contains
1931 * the "struct imsm_super" metadata.
1932 * This contains mpb_size which is the size in bytes of the
1933 * extended metadata. This is located immediately before
1935 * We want to read all that, plus the last sector which
1936 * may contain a migration record, and write it all
1940 unsigned long long dsize
, offset
;
1942 struct imsm_super
*sb
;
1943 struct intel_super
*super
= st
->sb
;
1944 unsigned int sector_size
= super
->sector_size
;
1945 unsigned int written
= 0;
1947 if (posix_memalign(&buf
, MAX_SECTOR_SIZE
, MAX_SECTOR_SIZE
) != 0)
1950 if (!get_dev_size(from
, NULL
, &dsize
))
1953 if (lseek64(from
, dsize
-(2*sector_size
), 0) < 0)
1955 if ((unsigned int)read(from
, buf
, sector_size
) != sector_size
)
1958 if (strncmp((char*)sb
->sig
, MPB_SIGNATURE
, MPB_SIG_LEN
) != 0)
1961 sectors
= mpb_sectors(sb
, sector_size
) + 2;
1962 offset
= dsize
- sectors
* sector_size
;
1963 if (lseek64(from
, offset
, 0) < 0 ||
1964 lseek64(to
, offset
, 0) < 0)
1966 while (written
< sectors
* sector_size
) {
1967 int n
= sectors
*sector_size
- written
;
1970 if (read(from
, buf
, n
) != n
)
1972 if (write(to
, buf
, n
) != n
)
1983 static void detail_super_imsm(struct supertype
*st
, char *homehost
)
1988 getinfo_super_imsm(st
, &info
, NULL
);
1989 fname_from_uuid(st
, &info
, nbuf
, ':');
1990 printf("\n UUID : %s\n", nbuf
+ 5);
1993 static void brief_detail_super_imsm(struct supertype
*st
)
1997 getinfo_super_imsm(st
, &info
, NULL
);
1998 fname_from_uuid(st
, &info
, nbuf
, ':');
1999 printf(" UUID=%s", nbuf
+ 5);
2002 static int imsm_read_serial(int fd
, char *devname
, __u8
*serial
);
2003 static void fd2devname(int fd
, char *name
);
2005 static int ahci_enumerate_ports(const char *hba_path
, int port_count
, int host_base
, int verbose
)
2007 /* dump an unsorted list of devices attached to AHCI Intel storage
2008 * controller, as well as non-connected ports
2010 int hba_len
= strlen(hba_path
) + 1;
2015 unsigned long port_mask
= (1 << port_count
) - 1;
2017 if (port_count
> (int)sizeof(port_mask
) * 8) {
2019 pr_err("port_count %d out of range\n", port_count
);
2023 /* scroll through /sys/dev/block looking for devices attached to
2026 dir
= opendir("/sys/dev/block");
2030 for (ent
= readdir(dir
); ent
; ent
= readdir(dir
)) {
2041 if (sscanf(ent
->d_name
, "%d:%d", &major
, &minor
) != 2)
2043 path
= devt_to_devpath(makedev(major
, minor
));
2046 if (!path_attached_to_hba(path
, hba_path
)) {
2052 /* retrieve the scsi device type */
2053 if (asprintf(&device
, "/sys/dev/block/%d:%d/device/xxxxxxx", major
, minor
) < 0) {
2055 pr_err("failed to allocate 'device'\n");
2059 sprintf(device
, "/sys/dev/block/%d:%d/device/type", major
, minor
);
2060 if (load_sys(device
, buf
, sizeof(buf
)) != 0) {
2062 pr_err("failed to read device type for %s\n",
2068 type
= strtoul(buf
, NULL
, 10);
2070 /* if it's not a disk print the vendor and model */
2071 if (!(type
== 0 || type
== 7 || type
== 14)) {
2074 sprintf(device
, "/sys/dev/block/%d:%d/device/vendor", major
, minor
);
2075 if (load_sys(device
, buf
, sizeof(buf
)) == 0) {
2076 strncpy(vendor
, buf
, sizeof(vendor
));
2077 vendor
[sizeof(vendor
) - 1] = '\0';
2078 c
= (char *) &vendor
[sizeof(vendor
) - 1];
2079 while (isspace(*c
) || *c
== '\0')
2083 sprintf(device
, "/sys/dev/block/%d:%d/device/model", major
, minor
);
2084 if (load_sys(device
, buf
, sizeof(buf
)) == 0) {
2085 strncpy(model
, buf
, sizeof(model
));
2086 model
[sizeof(model
) - 1] = '\0';
2087 c
= (char *) &model
[sizeof(model
) - 1];
2088 while (isspace(*c
) || *c
== '\0')
2092 if (vendor
[0] && model
[0])
2093 sprintf(buf
, "%.64s %.64s", vendor
, model
);
2095 switch (type
) { /* numbers from hald/linux/device.c */
2096 case 1: sprintf(buf
, "tape"); break;
2097 case 2: sprintf(buf
, "printer"); break;
2098 case 3: sprintf(buf
, "processor"); break;
2100 case 5: sprintf(buf
, "cdrom"); break;
2101 case 6: sprintf(buf
, "scanner"); break;
2102 case 8: sprintf(buf
, "media_changer"); break;
2103 case 9: sprintf(buf
, "comm"); break;
2104 case 12: sprintf(buf
, "raid"); break;
2105 default: sprintf(buf
, "unknown");
2111 /* chop device path to 'host%d' and calculate the port number */
2112 c
= strchr(&path
[hba_len
], '/');
2115 pr_err("%s - invalid path name\n", path
+ hba_len
);
2120 if ((sscanf(&path
[hba_len
], "ata%d", &port
) == 1) ||
2121 ((sscanf(&path
[hba_len
], "host%d", &port
) == 1)))
2125 *c
= '/'; /* repair the full string */
2126 pr_err("failed to determine port number for %s\n",
2133 /* mark this port as used */
2134 port_mask
&= ~(1 << port
);
2136 /* print out the device information */
2138 printf(" Port%d : - non-disk device (%s) -\n", port
, buf
);
2142 fd
= dev_open(ent
->d_name
, O_RDONLY
);
2144 printf(" Port%d : - disk info unavailable -\n", port
);
2146 fd2devname(fd
, buf
);
2147 printf(" Port%d : %s", port
, buf
);
2148 if (imsm_read_serial(fd
, NULL
, (__u8
*) buf
) == 0)
2149 printf(" (%.*s)\n", MAX_RAID_SERIAL_LEN
, buf
);
2164 for (i
= 0; i
< port_count
; i
++)
2165 if (port_mask
& (1 << i
))
2166 printf(" Port%d : - no device attached -\n", i
);
2172 static int print_vmd_attached_devs(struct sys_dev
*hba
)
2180 if (hba
->type
!= SYS_DEV_VMD
)
2183 /* scroll through /sys/dev/block looking for devices attached to
2186 dir
= opendir("/sys/bus/pci/drivers/nvme");
2190 for (ent
= readdir(dir
); ent
; ent
= readdir(dir
)) {
2193 /* is 'ent' a device? check that the 'subsystem' link exists and
2194 * that its target matches 'bus'
2196 sprintf(path
, "/sys/bus/pci/drivers/nvme/%s/subsystem",
2198 n
= readlink(path
, link
, sizeof(link
));
2199 if (n
< 0 || n
>= (int)sizeof(link
))
2202 c
= strrchr(link
, '/');
2205 if (strncmp("pci", c
+1, strlen("pci")) != 0)
2208 sprintf(path
, "/sys/bus/pci/drivers/nvme/%s", ent
->d_name
);
2210 rp
= realpath(path
, NULL
);
2214 if (path_attached_to_hba(rp
, hba
->path
)) {
2215 printf(" NVMe under VMD : %s\n", rp
);
2224 static void print_found_intel_controllers(struct sys_dev
*elem
)
2226 for (; elem
; elem
= elem
->next
) {
2227 pr_err("found Intel(R) ");
2228 if (elem
->type
== SYS_DEV_SATA
)
2229 fprintf(stderr
, "SATA ");
2230 else if (elem
->type
== SYS_DEV_SAS
)
2231 fprintf(stderr
, "SAS ");
2232 else if (elem
->type
== SYS_DEV_NVME
)
2233 fprintf(stderr
, "NVMe ");
2235 if (elem
->type
== SYS_DEV_VMD
)
2236 fprintf(stderr
, "VMD domain");
2238 fprintf(stderr
, "RAID controller");
2241 fprintf(stderr
, " at %s", elem
->pci_id
);
2242 fprintf(stderr
, ".\n");
2247 static int ahci_get_port_count(const char *hba_path
, int *port_count
)
2254 if ((dir
= opendir(hba_path
)) == NULL
)
2257 for (ent
= readdir(dir
); ent
; ent
= readdir(dir
)) {
2260 if ((sscanf(ent
->d_name
, "ata%d", &host
) != 1) &&
2261 ((sscanf(ent
->d_name
, "host%d", &host
) != 1)))
2263 if (*port_count
== 0)
2265 else if (host
< host_base
)
2268 if (host
+ 1 > *port_count
+ host_base
)
2269 *port_count
= host
+ 1 - host_base
;
2275 static void print_imsm_capability(const struct imsm_orom
*orom
)
2277 printf(" Platform : Intel(R) ");
2278 if (orom
->capabilities
== 0 && orom
->driver_features
== 0)
2279 printf("Matrix Storage Manager\n");
2281 printf("Rapid Storage Technology%s\n",
2282 imsm_orom_is_enterprise(orom
) ? " enterprise" : "");
2283 if (orom
->major_ver
|| orom
->minor_ver
|| orom
->hotfix_ver
|| orom
->build
)
2284 printf(" Version : %d.%d.%d.%d\n", orom
->major_ver
,
2285 orom
->minor_ver
, orom
->hotfix_ver
, orom
->build
);
2286 printf(" RAID Levels :%s%s%s%s%s\n",
2287 imsm_orom_has_raid0(orom
) ? " raid0" : "",
2288 imsm_orom_has_raid1(orom
) ? " raid1" : "",
2289 imsm_orom_has_raid1e(orom
) ? " raid1e" : "",
2290 imsm_orom_has_raid10(orom
) ? " raid10" : "",
2291 imsm_orom_has_raid5(orom
) ? " raid5" : "");
2292 printf(" Chunk Sizes :%s%s%s%s%s%s%s%s%s%s%s%s%s%s%s%s\n",
2293 imsm_orom_has_chunk(orom
, 2) ? " 2k" : "",
2294 imsm_orom_has_chunk(orom
, 4) ? " 4k" : "",
2295 imsm_orom_has_chunk(orom
, 8) ? " 8k" : "",
2296 imsm_orom_has_chunk(orom
, 16) ? " 16k" : "",
2297 imsm_orom_has_chunk(orom
, 32) ? " 32k" : "",
2298 imsm_orom_has_chunk(orom
, 64) ? " 64k" : "",
2299 imsm_orom_has_chunk(orom
, 128) ? " 128k" : "",
2300 imsm_orom_has_chunk(orom
, 256) ? " 256k" : "",
2301 imsm_orom_has_chunk(orom
, 512) ? " 512k" : "",
2302 imsm_orom_has_chunk(orom
, 1024*1) ? " 1M" : "",
2303 imsm_orom_has_chunk(orom
, 1024*2) ? " 2M" : "",
2304 imsm_orom_has_chunk(orom
, 1024*4) ? " 4M" : "",
2305 imsm_orom_has_chunk(orom
, 1024*8) ? " 8M" : "",
2306 imsm_orom_has_chunk(orom
, 1024*16) ? " 16M" : "",
2307 imsm_orom_has_chunk(orom
, 1024*32) ? " 32M" : "",
2308 imsm_orom_has_chunk(orom
, 1024*64) ? " 64M" : "");
2309 printf(" 2TB volumes :%s supported\n",
2310 (orom
->attr
& IMSM_OROM_ATTR_2TB
)?"":" not");
2311 printf(" 2TB disks :%s supported\n",
2312 (orom
->attr
& IMSM_OROM_ATTR_2TB_DISK
)?"":" not");
2313 printf(" Max Disks : %d\n", orom
->tds
);
2314 printf(" Max Volumes : %d per array, %d per %s\n",
2315 orom
->vpa
, orom
->vphba
,
2316 imsm_orom_is_nvme(orom
) ? "platform" : "controller");
2320 static void print_imsm_capability_export(const struct imsm_orom
*orom
)
2322 printf("MD_FIRMWARE_TYPE=imsm\n");
2323 if (orom
->major_ver
|| orom
->minor_ver
|| orom
->hotfix_ver
|| orom
->build
)
2324 printf("IMSM_VERSION=%d.%d.%d.%d\n", orom
->major_ver
, orom
->minor_ver
,
2325 orom
->hotfix_ver
, orom
->build
);
2326 printf("IMSM_SUPPORTED_RAID_LEVELS=%s%s%s%s%s\n",
2327 imsm_orom_has_raid0(orom
) ? "raid0 " : "",
2328 imsm_orom_has_raid1(orom
) ? "raid1 " : "",
2329 imsm_orom_has_raid1e(orom
) ? "raid1e " : "",
2330 imsm_orom_has_raid5(orom
) ? "raid10 " : "",
2331 imsm_orom_has_raid10(orom
) ? "raid5 " : "");
2332 printf("IMSM_SUPPORTED_CHUNK_SIZES=%s%s%s%s%s%s%s%s%s%s%s%s%s%s%s%s\n",
2333 imsm_orom_has_chunk(orom
, 2) ? "2k " : "",
2334 imsm_orom_has_chunk(orom
, 4) ? "4k " : "",
2335 imsm_orom_has_chunk(orom
, 8) ? "8k " : "",
2336 imsm_orom_has_chunk(orom
, 16) ? "16k " : "",
2337 imsm_orom_has_chunk(orom
, 32) ? "32k " : "",
2338 imsm_orom_has_chunk(orom
, 64) ? "64k " : "",
2339 imsm_orom_has_chunk(orom
, 128) ? "128k " : "",
2340 imsm_orom_has_chunk(orom
, 256) ? "256k " : "",
2341 imsm_orom_has_chunk(orom
, 512) ? "512k " : "",
2342 imsm_orom_has_chunk(orom
, 1024*1) ? "1M " : "",
2343 imsm_orom_has_chunk(orom
, 1024*2) ? "2M " : "",
2344 imsm_orom_has_chunk(orom
, 1024*4) ? "4M " : "",
2345 imsm_orom_has_chunk(orom
, 1024*8) ? "8M " : "",
2346 imsm_orom_has_chunk(orom
, 1024*16) ? "16M " : "",
2347 imsm_orom_has_chunk(orom
, 1024*32) ? "32M " : "",
2348 imsm_orom_has_chunk(orom
, 1024*64) ? "64M " : "");
2349 printf("IMSM_2TB_VOLUMES=%s\n",(orom
->attr
& IMSM_OROM_ATTR_2TB
) ? "yes" : "no");
2350 printf("IMSM_2TB_DISKS=%s\n",(orom
->attr
& IMSM_OROM_ATTR_2TB_DISK
) ? "yes" : "no");
2351 printf("IMSM_MAX_DISKS=%d\n",orom
->tds
);
2352 printf("IMSM_MAX_VOLUMES_PER_ARRAY=%d\n",orom
->vpa
);
2353 printf("IMSM_MAX_VOLUMES_PER_CONTROLLER=%d\n",orom
->vphba
);
2356 static int detail_platform_imsm(int verbose
, int enumerate_only
, char *controller_path
)
2358 /* There are two components to imsm platform support, the ahci SATA
2359 * controller and the option-rom. To find the SATA controller we
2360 * simply look in /sys/bus/pci/drivers/ahci to see if an ahci
2361 * controller with the Intel vendor id is present. This approach
2362 * allows mdadm to leverage the kernel's ahci detection logic, with the
2363 * caveat that if ahci.ko is not loaded mdadm will not be able to
2364 * detect platform raid capabilities. The option-rom resides in a
2365 * platform "Adapter ROM". We scan for its signature to retrieve the
2366 * platform capabilities. If raid support is disabled in the BIOS the
2367 * option-rom capability structure will not be available.
2369 struct sys_dev
*list
, *hba
;
2374 if (enumerate_only
) {
2375 if (check_env("IMSM_NO_PLATFORM"))
2377 list
= find_intel_devices();
2380 for (hba
= list
; hba
; hba
= hba
->next
) {
2381 if (find_imsm_capability(hba
)) {
2391 list
= find_intel_devices();
2394 pr_err("no active Intel(R) RAID controller found.\n");
2396 } else if (verbose
> 0)
2397 print_found_intel_controllers(list
);
2399 for (hba
= list
; hba
; hba
= hba
->next
) {
2400 if (controller_path
&& (compare_paths(hba
->path
, controller_path
) != 0))
2402 if (!find_imsm_capability(hba
)) {
2404 pr_err("imsm capabilities not found for controller: %s (type %s)\n",
2405 hba
->type
== SYS_DEV_VMD
? vmd_domain_to_controller(hba
, buf
) : hba
->path
,
2406 get_sys_dev_type(hba
->type
));
2412 if (controller_path
&& result
== 1) {
2413 pr_err("no active Intel(R) RAID controller found under %s\n",
2418 const struct orom_entry
*entry
;
2420 for (entry
= orom_entries
; entry
; entry
= entry
->next
) {
2421 if (entry
->type
== SYS_DEV_VMD
) {
2422 print_imsm_capability(&entry
->orom
);
2423 printf(" 3rd party NVMe :%s supported\n",
2424 imsm_orom_has_tpv_support(&entry
->orom
)?"":" not");
2425 for (hba
= list
; hba
; hba
= hba
->next
) {
2426 if (hba
->type
== SYS_DEV_VMD
) {
2428 printf(" I/O Controller : %s (%s)\n",
2429 vmd_domain_to_controller(hba
, buf
), get_sys_dev_type(hba
->type
));
2430 if (print_vmd_attached_devs(hba
)) {
2432 pr_err("failed to get devices attached to VMD domain.\n");
2441 print_imsm_capability(&entry
->orom
);
2442 if (entry
->type
== SYS_DEV_NVME
) {
2443 for (hba
= list
; hba
; hba
= hba
->next
) {
2444 if (hba
->type
== SYS_DEV_NVME
)
2445 printf(" NVMe Device : %s\n", hba
->path
);
2451 struct devid_list
*devid
;
2452 for (devid
= entry
->devid_list
; devid
; devid
= devid
->next
) {
2453 hba
= device_by_id(devid
->devid
);
2457 printf(" I/O Controller : %s (%s)\n",
2458 hba
->path
, get_sys_dev_type(hba
->type
));
2459 if (hba
->type
== SYS_DEV_SATA
) {
2460 host_base
= ahci_get_port_count(hba
->path
, &port_count
);
2461 if (ahci_enumerate_ports(hba
->path
, port_count
, host_base
, verbose
)) {
2463 pr_err("failed to enumerate ports on SATA controller at %s.\n", hba
->pci_id
);
2474 static int export_detail_platform_imsm(int verbose
, char *controller_path
)
2476 struct sys_dev
*list
, *hba
;
2479 list
= find_intel_devices();
2482 pr_err("IMSM_DETAIL_PLATFORM_ERROR=NO_INTEL_DEVICES\n");
2487 for (hba
= list
; hba
; hba
= hba
->next
) {
2488 if (controller_path
&& (compare_paths(hba
->path
,controller_path
) != 0))
2490 if (!find_imsm_capability(hba
) && verbose
> 0) {
2492 pr_err("IMSM_DETAIL_PLATFORM_ERROR=NO_IMSM_CAPABLE_DEVICE_UNDER_%s\n",
2493 hba
->type
== SYS_DEV_VMD
? vmd_domain_to_controller(hba
, buf
) : hba
->path
);
2499 const struct orom_entry
*entry
;
2501 for (entry
= orom_entries
; entry
; entry
= entry
->next
) {
2502 if (entry
->type
== SYS_DEV_VMD
) {
2503 for (hba
= list
; hba
; hba
= hba
->next
)
2504 print_imsm_capability_export(&entry
->orom
);
2507 print_imsm_capability_export(&entry
->orom
);
2515 static int match_home_imsm(struct supertype
*st
, char *homehost
)
2517 /* the imsm metadata format does not specify any host
2518 * identification information. We return -1 since we can never
2519 * confirm nor deny whether a given array is "meant" for this
2520 * host. We rely on compare_super and the 'family_num' fields to
2521 * exclude member disks that do not belong, and we rely on
2522 * mdadm.conf to specify the arrays that should be assembled.
2523 * Auto-assembly may still pick up "foreign" arrays.
2529 static void uuid_from_super_imsm(struct supertype
*st
, int uuid
[4])
2531 /* The uuid returned here is used for:
2532 * uuid to put into bitmap file (Create, Grow)
2533 * uuid for backup header when saving critical section (Grow)
2534 * comparing uuids when re-adding a device into an array
2535 * In these cases the uuid required is that of the data-array,
2536 * not the device-set.
2537 * uuid to recognise same set when adding a missing device back
2538 * to an array. This is a uuid for the device-set.
2540 * For each of these we can make do with a truncated
2541 * or hashed uuid rather than the original, as long as
2543 * In each case the uuid required is that of the data-array,
2544 * not the device-set.
2546 /* imsm does not track uuid's so we synthesis one using sha1 on
2547 * - The signature (Which is constant for all imsm array, but no matter)
2548 * - the orig_family_num of the container
2549 * - the index number of the volume
2550 * - the 'serial' number of the volume.
2551 * Hopefully these are all constant.
2553 struct intel_super
*super
= st
->sb
;
2556 struct sha1_ctx ctx
;
2557 struct imsm_dev
*dev
= NULL
;
2560 /* some mdadm versions failed to set ->orig_family_num, in which
2561 * case fall back to ->family_num. orig_family_num will be
2562 * fixed up with the first metadata update.
2564 family_num
= super
->anchor
->orig_family_num
;
2565 if (family_num
== 0)
2566 family_num
= super
->anchor
->family_num
;
2567 sha1_init_ctx(&ctx
);
2568 sha1_process_bytes(super
->anchor
->sig
, MPB_SIG_LEN
, &ctx
);
2569 sha1_process_bytes(&family_num
, sizeof(__u32
), &ctx
);
2570 if (super
->current_vol
>= 0)
2571 dev
= get_imsm_dev(super
, super
->current_vol
);
2573 __u32 vol
= super
->current_vol
;
2574 sha1_process_bytes(&vol
, sizeof(vol
), &ctx
);
2575 sha1_process_bytes(dev
->volume
, MAX_RAID_SERIAL_LEN
, &ctx
);
2577 sha1_finish_ctx(&ctx
, buf
);
2578 memcpy(uuid
, buf
, 4*4);
2583 get_imsm_numerical_version(struct imsm_super
*mpb
, int *m
, int *p
)
2585 __u8
*v
= get_imsm_version(mpb
);
2586 __u8
*end
= mpb
->sig
+ MAX_SIGNATURE_LENGTH
;
2587 char major
[] = { 0, 0, 0 };
2588 char minor
[] = { 0 ,0, 0 };
2589 char patch
[] = { 0, 0, 0 };
2590 char *ver_parse
[] = { major
, minor
, patch
};
2594 while (*v
!= '\0' && v
< end
) {
2595 if (*v
!= '.' && j
< 2)
2596 ver_parse
[i
][j
++] = *v
;
2604 *m
= strtol(minor
, NULL
, 0);
2605 *p
= strtol(patch
, NULL
, 0);
2609 static __u32
migr_strip_blocks_resync(struct imsm_dev
*dev
)
2611 /* migr_strip_size when repairing or initializing parity */
2612 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
2613 __u32 chunk
= __le32_to_cpu(map
->blocks_per_strip
);
2615 switch (get_imsm_raid_level(map
)) {
2620 return 128*1024 >> 9;
2624 static __u32
migr_strip_blocks_rebuild(struct imsm_dev
*dev
)
2626 /* migr_strip_size when rebuilding a degraded disk, no idea why
2627 * this is different than migr_strip_size_resync(), but it's good
2630 struct imsm_map
*map
= get_imsm_map(dev
, MAP_1
);
2631 __u32 chunk
= __le32_to_cpu(map
->blocks_per_strip
);
2633 switch (get_imsm_raid_level(map
)) {
2636 if (map
->num_members
% map
->num_domains
== 0)
2637 return 128*1024 >> 9;
2641 return max((__u32
) 64*1024 >> 9, chunk
);
2643 return 128*1024 >> 9;
2647 static __u32
num_stripes_per_unit_resync(struct imsm_dev
*dev
)
2649 struct imsm_map
*lo
= get_imsm_map(dev
, MAP_0
);
2650 struct imsm_map
*hi
= get_imsm_map(dev
, MAP_1
);
2651 __u32 lo_chunk
= __le32_to_cpu(lo
->blocks_per_strip
);
2652 __u32 hi_chunk
= __le32_to_cpu(hi
->blocks_per_strip
);
2654 return max((__u32
) 1, hi_chunk
/ lo_chunk
);
2657 static __u32
num_stripes_per_unit_rebuild(struct imsm_dev
*dev
)
2659 struct imsm_map
*lo
= get_imsm_map(dev
, MAP_0
);
2660 int level
= get_imsm_raid_level(lo
);
2662 if (level
== 1 || level
== 10) {
2663 struct imsm_map
*hi
= get_imsm_map(dev
, MAP_1
);
2665 return hi
->num_domains
;
2667 return num_stripes_per_unit_resync(dev
);
2670 static __u8
imsm_num_data_members(struct imsm_dev
*dev
, int second_map
)
2672 /* named 'imsm_' because raid0, raid1 and raid10
2673 * counter-intuitively have the same number of data disks
2675 struct imsm_map
*map
= get_imsm_map(dev
, second_map
);
2677 switch (get_imsm_raid_level(map
)) {
2679 return map
->num_members
;
2683 return map
->num_members
/2;
2685 return map
->num_members
- 1;
2687 dprintf("unsupported raid level\n");
2692 static __u32
parity_segment_depth(struct imsm_dev
*dev
)
2694 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
2695 __u32 chunk
= __le32_to_cpu(map
->blocks_per_strip
);
2697 switch(get_imsm_raid_level(map
)) {
2700 return chunk
* map
->num_domains
;
2702 return chunk
* map
->num_members
;
2708 static __u32
map_migr_block(struct imsm_dev
*dev
, __u32 block
)
2710 struct imsm_map
*map
= get_imsm_map(dev
, MAP_1
);
2711 __u32 chunk
= __le32_to_cpu(map
->blocks_per_strip
);
2712 __u32 strip
= block
/ chunk
;
2714 switch (get_imsm_raid_level(map
)) {
2717 __u32 vol_strip
= (strip
* map
->num_domains
) + 1;
2718 __u32 vol_stripe
= vol_strip
/ map
->num_members
;
2720 return vol_stripe
* chunk
+ block
% chunk
;
2722 __u32 stripe
= strip
/ (map
->num_members
- 1);
2724 return stripe
* chunk
+ block
% chunk
;
2731 static __u64
blocks_per_migr_unit(struct intel_super
*super
,
2732 struct imsm_dev
*dev
)
2734 /* calculate the conversion factor between per member 'blocks'
2735 * (md/{resync,rebuild}_start) and imsm migration units, return
2736 * 0 for the 'not migrating' and 'unsupported migration' cases
2738 if (!dev
->vol
.migr_state
)
2741 switch (migr_type(dev
)) {
2742 case MIGR_GEN_MIGR
: {
2743 struct migr_record
*migr_rec
= super
->migr_rec
;
2744 return __le32_to_cpu(migr_rec
->blocks_per_unit
);
2749 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
2750 __u32 stripes_per_unit
;
2751 __u32 blocks_per_unit
;
2760 /* yes, this is really the translation of migr_units to
2761 * per-member blocks in the 'resync' case
2763 stripes_per_unit
= num_stripes_per_unit_resync(dev
);
2764 migr_chunk
= migr_strip_blocks_resync(dev
);
2765 disks
= imsm_num_data_members(dev
, MAP_0
);
2766 blocks_per_unit
= stripes_per_unit
* migr_chunk
* disks
;
2767 stripe
= __le16_to_cpu(map
->blocks_per_strip
) * disks
;
2768 segment
= blocks_per_unit
/ stripe
;
2769 block_rel
= blocks_per_unit
- segment
* stripe
;
2770 parity_depth
= parity_segment_depth(dev
);
2771 block_map
= map_migr_block(dev
, block_rel
);
2772 return block_map
+ parity_depth
* segment
;
2774 case MIGR_REBUILD
: {
2775 __u32 stripes_per_unit
;
2778 stripes_per_unit
= num_stripes_per_unit_rebuild(dev
);
2779 migr_chunk
= migr_strip_blocks_rebuild(dev
);
2780 return migr_chunk
* stripes_per_unit
;
2782 case MIGR_STATE_CHANGE
:
2788 static int imsm_level_to_layout(int level
)
2796 return ALGORITHM_LEFT_ASYMMETRIC
;
2803 /*******************************************************************************
2804 * Function: read_imsm_migr_rec
2805 * Description: Function reads imsm migration record from last sector of disk
2807 * fd : disk descriptor
2808 * super : metadata info
2812 ******************************************************************************/
2813 static int read_imsm_migr_rec(int fd
, struct intel_super
*super
)
2816 unsigned int sector_size
= super
->sector_size
;
2817 unsigned long long dsize
;
2819 get_dev_size(fd
, NULL
, &dsize
);
2820 if (lseek64(fd
, dsize
- (sector_size
*MIGR_REC_SECTOR_POSITION
),
2822 pr_err("Cannot seek to anchor block: %s\n",
2826 if ((unsigned int)read(fd
, super
->migr_rec_buf
,
2827 MIGR_REC_BUF_SECTORS
*sector_size
) !=
2828 MIGR_REC_BUF_SECTORS
*sector_size
) {
2829 pr_err("Cannot read migr record block: %s\n",
2834 if (sector_size
== 4096)
2835 convert_from_4k_imsm_migr_rec(super
);
2841 static struct imsm_dev
*imsm_get_device_during_migration(
2842 struct intel_super
*super
)
2845 struct intel_dev
*dv
;
2847 for (dv
= super
->devlist
; dv
; dv
= dv
->next
) {
2848 if (is_gen_migration(dv
->dev
))
2854 /*******************************************************************************
2855 * Function: load_imsm_migr_rec
2856 * Description: Function reads imsm migration record (it is stored at the last
2859 * super : imsm internal array info
2860 * info : general array info
2864 * -2 : no migration in progress
2865 ******************************************************************************/
2866 static int load_imsm_migr_rec(struct intel_super
*super
, struct mdinfo
*info
)
2873 struct imsm_dev
*dev
;
2874 struct imsm_map
*map
;
2877 /* find map under migration */
2878 dev
= imsm_get_device_during_migration(super
);
2879 /* nothing to load,no migration in progress?
2885 for (sd
= info
->devs
; sd
; sd
= sd
->next
) {
2886 /* read only from one of the first two slots */
2887 if ((sd
->disk
.raid_disk
< 0) ||
2888 (sd
->disk
.raid_disk
> 1))
2891 sprintf(nm
, "%d:%d", sd
->disk
.major
, sd
->disk
.minor
);
2892 fd
= dev_open(nm
, O_RDONLY
);
2898 map
= get_imsm_map(dev
, MAP_0
);
2899 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
2900 /* skip spare and failed disks
2904 /* read only from one of the first two slots */
2906 slot
= get_imsm_disk_slot(map
, dl
->index
);
2907 if (map
== NULL
|| slot
> 1 || slot
< 0)
2909 sprintf(nm
, "%d:%d", dl
->major
, dl
->minor
);
2910 fd
= dev_open(nm
, O_RDONLY
);
2917 retval
= read_imsm_migr_rec(fd
, super
);
2926 /*******************************************************************************
2927 * function: imsm_create_metadata_checkpoint_update
2928 * Description: It creates update for checkpoint change.
2930 * super : imsm internal array info
2931 * u : pointer to prepared update
2934 * If length is equal to 0, input pointer u contains no update
2935 ******************************************************************************/
2936 static int imsm_create_metadata_checkpoint_update(
2937 struct intel_super
*super
,
2938 struct imsm_update_general_migration_checkpoint
**u
)
2941 int update_memory_size
= 0;
2943 dprintf("(enter)\n");
2949 /* size of all update data without anchor */
2950 update_memory_size
=
2951 sizeof(struct imsm_update_general_migration_checkpoint
);
2953 *u
= xcalloc(1, update_memory_size
);
2955 dprintf("error: cannot get memory\n");
2958 (*u
)->type
= update_general_migration_checkpoint
;
2959 (*u
)->curr_migr_unit
= __le32_to_cpu(super
->migr_rec
->curr_migr_unit
);
2960 dprintf("prepared for %u\n", (*u
)->curr_migr_unit
);
2962 return update_memory_size
;
2965 static void imsm_update_metadata_locally(struct supertype
*st
,
2966 void *buf
, int len
);
2968 /*******************************************************************************
2969 * Function: write_imsm_migr_rec
2970 * Description: Function writes imsm migration record
2971 * (at the last sector of disk)
2973 * super : imsm internal array info
2977 ******************************************************************************/
2978 static int write_imsm_migr_rec(struct supertype
*st
)
2980 struct intel_super
*super
= st
->sb
;
2981 unsigned int sector_size
= super
->sector_size
;
2982 unsigned long long dsize
;
2988 struct imsm_update_general_migration_checkpoint
*u
;
2989 struct imsm_dev
*dev
;
2990 struct imsm_map
*map
;
2992 /* find map under migration */
2993 dev
= imsm_get_device_during_migration(super
);
2994 /* if no migration, write buffer anyway to clear migr_record
2995 * on disk based on first available device
2998 dev
= get_imsm_dev(super
, super
->current_vol
< 0 ? 0 :
2999 super
->current_vol
);
3001 map
= get_imsm_map(dev
, MAP_0
);
3003 if (sector_size
== 4096)
3004 convert_to_4k_imsm_migr_rec(super
);
3005 for (sd
= super
->disks
; sd
; sd
= sd
->next
) {
3008 /* skip failed and spare devices */
3011 /* write to 2 first slots only */
3013 slot
= get_imsm_disk_slot(map
, sd
->index
);
3014 if (map
== NULL
|| slot
> 1 || slot
< 0)
3017 sprintf(nm
, "%d:%d", sd
->major
, sd
->minor
);
3018 fd
= dev_open(nm
, O_RDWR
);
3021 get_dev_size(fd
, NULL
, &dsize
);
3022 if (lseek64(fd
, dsize
- (MIGR_REC_SECTOR_POSITION
*sector_size
),
3024 pr_err("Cannot seek to anchor block: %s\n",
3028 if ((unsigned int)write(fd
, super
->migr_rec_buf
,
3029 MIGR_REC_BUF_SECTORS
*sector_size
) !=
3030 MIGR_REC_BUF_SECTORS
*sector_size
) {
3031 pr_err("Cannot write migr record block: %s\n",
3038 if (sector_size
== 4096)
3039 convert_from_4k_imsm_migr_rec(super
);
3040 /* update checkpoint information in metadata */
3041 len
= imsm_create_metadata_checkpoint_update(super
, &u
);
3043 dprintf("imsm: Cannot prepare update\n");
3046 /* update metadata locally */
3047 imsm_update_metadata_locally(st
, u
, len
);
3048 /* and possibly remotely */
3049 if (st
->update_tail
) {
3050 append_metadata_update(st
, u
, len
);
3051 /* during reshape we do all work inside metadata handler
3052 * manage_reshape(), so metadata update has to be triggered
3055 flush_metadata_updates(st
);
3056 st
->update_tail
= &st
->updates
;
3066 #endif /* MDASSEMBLE */
3068 /* spare/missing disks activations are not allowe when
3069 * array/container performs reshape operation, because
3070 * all arrays in container works on the same disks set
3072 int imsm_reshape_blocks_arrays_changes(struct intel_super
*super
)
3075 struct intel_dev
*i_dev
;
3076 struct imsm_dev
*dev
;
3078 /* check whole container
3080 for (i_dev
= super
->devlist
; i_dev
; i_dev
= i_dev
->next
) {
3082 if (is_gen_migration(dev
)) {
3083 /* No repair during any migration in container
3091 static unsigned long long imsm_component_size_aligment_check(int level
,
3093 unsigned int sector_size
,
3094 unsigned long long component_size
)
3096 unsigned int component_size_alligment
;
3098 /* check component size aligment
3100 component_size_alligment
= component_size
% (chunk_size
/sector_size
);
3102 dprintf("(Level: %i, chunk_size = %i, component_size = %llu), component_size_alligment = %u\n",
3103 level
, chunk_size
, component_size
,
3104 component_size_alligment
);
3106 if (component_size_alligment
&& (level
!= 1) && (level
!= UnSet
)) {
3107 dprintf("imsm: reported component size alligned from %llu ",
3109 component_size
-= component_size_alligment
;
3110 dprintf_cont("to %llu (%i).\n",
3111 component_size
, component_size_alligment
);
3114 return component_size
;
3117 static void getinfo_super_imsm_volume(struct supertype
*st
, struct mdinfo
*info
, char *dmap
)
3119 struct intel_super
*super
= st
->sb
;
3120 struct migr_record
*migr_rec
= super
->migr_rec
;
3121 struct imsm_dev
*dev
= get_imsm_dev(super
, super
->current_vol
);
3122 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
3123 struct imsm_map
*prev_map
= get_imsm_map(dev
, MAP_1
);
3124 struct imsm_map
*map_to_analyse
= map
;
3126 int map_disks
= info
->array
.raid_disks
;
3128 memset(info
, 0, sizeof(*info
));
3130 map_to_analyse
= prev_map
;
3132 dl
= super
->current_disk
;
3134 info
->container_member
= super
->current_vol
;
3135 info
->array
.raid_disks
= map
->num_members
;
3136 info
->array
.level
= get_imsm_raid_level(map_to_analyse
);
3137 info
->array
.layout
= imsm_level_to_layout(info
->array
.level
);
3138 info
->array
.md_minor
= -1;
3139 info
->array
.ctime
= 0;
3140 info
->array
.utime
= 0;
3141 info
->array
.chunk_size
=
3142 __le16_to_cpu(map_to_analyse
->blocks_per_strip
) << 9;
3143 info
->array
.state
= !dev
->vol
.dirty
;
3144 info
->custom_array_size
= __le32_to_cpu(dev
->size_high
);
3145 info
->custom_array_size
<<= 32;
3146 info
->custom_array_size
|= __le32_to_cpu(dev
->size_low
);
3147 info
->recovery_blocked
= imsm_reshape_blocks_arrays_changes(st
->sb
);
3149 if (is_gen_migration(dev
)) {
3150 info
->reshape_active
= 1;
3151 info
->new_level
= get_imsm_raid_level(map
);
3152 info
->new_layout
= imsm_level_to_layout(info
->new_level
);
3153 info
->new_chunk
= __le16_to_cpu(map
->blocks_per_strip
) << 9;
3154 info
->delta_disks
= map
->num_members
- prev_map
->num_members
;
3155 if (info
->delta_disks
) {
3156 /* this needs to be applied to every array
3159 info
->reshape_active
= CONTAINER_RESHAPE
;
3161 /* We shape information that we give to md might have to be
3162 * modify to cope with md's requirement for reshaping arrays.
3163 * For example, when reshaping a RAID0, md requires it to be
3164 * presented as a degraded RAID4.
3165 * Also if a RAID0 is migrating to a RAID5 we need to specify
3166 * the array as already being RAID5, but the 'before' layout
3167 * is a RAID4-like layout.
3169 switch (info
->array
.level
) {
3171 switch(info
->new_level
) {
3173 /* conversion is happening as RAID4 */
3174 info
->array
.level
= 4;
3175 info
->array
.raid_disks
+= 1;
3178 /* conversion is happening as RAID5 */
3179 info
->array
.level
= 5;
3180 info
->array
.layout
= ALGORITHM_PARITY_N
;
3181 info
->delta_disks
-= 1;
3184 /* FIXME error message */
3185 info
->array
.level
= UnSet
;
3191 info
->new_level
= UnSet
;
3192 info
->new_layout
= UnSet
;
3193 info
->new_chunk
= info
->array
.chunk_size
;
3194 info
->delta_disks
= 0;
3198 info
->disk
.major
= dl
->major
;
3199 info
->disk
.minor
= dl
->minor
;
3200 info
->disk
.number
= dl
->index
;
3201 info
->disk
.raid_disk
= get_imsm_disk_slot(map_to_analyse
,
3205 info
->data_offset
= pba_of_lba0(map_to_analyse
);
3207 if (info
->array
.level
== 5) {
3208 info
->component_size
= num_data_stripes(map_to_analyse
) *
3209 map_to_analyse
->blocks_per_strip
;
3211 info
->component_size
= blocks_per_member(map_to_analyse
);
3214 info
->component_size
= imsm_component_size_aligment_check(
3216 info
->array
.chunk_size
,
3218 info
->component_size
);
3219 info
->bb
.supported
= 1;
3221 memset(info
->uuid
, 0, sizeof(info
->uuid
));
3222 info
->recovery_start
= MaxSector
;
3224 info
->reshape_progress
= 0;
3225 info
->resync_start
= MaxSector
;
3226 if ((map_to_analyse
->map_state
== IMSM_T_STATE_UNINITIALIZED
||
3228 imsm_reshape_blocks_arrays_changes(super
) == 0) {
3229 info
->resync_start
= 0;
3231 if (dev
->vol
.migr_state
) {
3232 switch (migr_type(dev
)) {
3235 __u64 blocks_per_unit
= blocks_per_migr_unit(super
,
3237 __u64 units
= __le32_to_cpu(dev
->vol
.curr_migr_unit
);
3239 info
->resync_start
= blocks_per_unit
* units
;
3242 case MIGR_GEN_MIGR
: {
3243 __u64 blocks_per_unit
= blocks_per_migr_unit(super
,
3245 __u64 units
= __le32_to_cpu(migr_rec
->curr_migr_unit
);
3246 unsigned long long array_blocks
;
3249 if (__le32_to_cpu(migr_rec
->ascending_migr
) &&
3251 (__le32_to_cpu(migr_rec
->num_migr_units
)-1)) &&
3252 (super
->migr_rec
->rec_status
==
3253 __cpu_to_le32(UNIT_SRC_IN_CP_AREA
)))
3256 info
->reshape_progress
= blocks_per_unit
* units
;
3258 dprintf("IMSM: General Migration checkpoint : %llu (%llu) -> read reshape progress : %llu\n",
3259 (unsigned long long)units
,
3260 (unsigned long long)blocks_per_unit
,
3261 info
->reshape_progress
);
3263 used_disks
= imsm_num_data_members(dev
, MAP_1
);
3264 if (used_disks
> 0) {
3265 array_blocks
= blocks_per_member(map
) *
3267 /* round array size down to closest MB
3269 info
->custom_array_size
= (array_blocks
3270 >> SECT_PER_MB_SHIFT
)
3271 << SECT_PER_MB_SHIFT
;
3275 /* we could emulate the checkpointing of
3276 * 'sync_action=check' migrations, but for now
3277 * we just immediately complete them
3280 /* this is handled by container_content_imsm() */
3281 case MIGR_STATE_CHANGE
:
3282 /* FIXME handle other migrations */
3284 /* we are not dirty, so... */
3285 info
->resync_start
= MaxSector
;
3289 strncpy(info
->name
, (char *) dev
->volume
, MAX_RAID_SERIAL_LEN
);
3290 info
->name
[MAX_RAID_SERIAL_LEN
] = 0;
3292 info
->array
.major_version
= -1;
3293 info
->array
.minor_version
= -2;
3294 sprintf(info
->text_version
, "/%s/%d", st
->container_devnm
, info
->container_member
);
3295 info
->safe_mode_delay
= 4000; /* 4 secs like the Matrix driver */
3296 uuid_from_super_imsm(st
, info
->uuid
);
3300 for (i
=0; i
<map_disks
; i
++) {
3302 if (i
< info
->array
.raid_disks
) {
3303 struct imsm_disk
*dsk
;
3304 j
= get_imsm_disk_idx(dev
, i
, MAP_X
);
3305 dsk
= get_imsm_disk(super
, j
);
3306 if (dsk
&& (dsk
->status
& CONFIGURED_DISK
))
3313 static __u8
imsm_check_degraded(struct intel_super
*super
, struct imsm_dev
*dev
,
3314 int failed
, int look_in_map
);
3316 static int imsm_count_failed(struct intel_super
*super
, struct imsm_dev
*dev
,
3320 static void manage_second_map(struct intel_super
*super
, struct imsm_dev
*dev
)
3322 if (is_gen_migration(dev
)) {
3325 struct imsm_map
*map2
= get_imsm_map(dev
, MAP_1
);
3327 failed
= imsm_count_failed(super
, dev
, MAP_1
);
3328 map_state
= imsm_check_degraded(super
, dev
, failed
, MAP_1
);
3329 if (map2
->map_state
!= map_state
) {
3330 map2
->map_state
= map_state
;
3331 super
->updates_pending
++;
3337 static struct imsm_disk
*get_imsm_missing(struct intel_super
*super
, __u8 index
)
3341 for (d
= super
->missing
; d
; d
= d
->next
)
3342 if (d
->index
== index
)
3347 static void getinfo_super_imsm(struct supertype
*st
, struct mdinfo
*info
, char *map
)
3349 struct intel_super
*super
= st
->sb
;
3350 struct imsm_disk
*disk
;
3351 int map_disks
= info
->array
.raid_disks
;
3352 int max_enough
= -1;
3354 struct imsm_super
*mpb
;
3356 if (super
->current_vol
>= 0) {
3357 getinfo_super_imsm_volume(st
, info
, map
);
3360 memset(info
, 0, sizeof(*info
));
3362 /* Set raid_disks to zero so that Assemble will always pull in valid
3365 info
->array
.raid_disks
= 0;
3366 info
->array
.level
= LEVEL_CONTAINER
;
3367 info
->array
.layout
= 0;
3368 info
->array
.md_minor
= -1;
3369 info
->array
.ctime
= 0; /* N/A for imsm */
3370 info
->array
.utime
= 0;
3371 info
->array
.chunk_size
= 0;
3373 info
->disk
.major
= 0;
3374 info
->disk
.minor
= 0;
3375 info
->disk
.raid_disk
= -1;
3376 info
->reshape_active
= 0;
3377 info
->array
.major_version
= -1;
3378 info
->array
.minor_version
= -2;
3379 strcpy(info
->text_version
, "imsm");
3380 info
->safe_mode_delay
= 0;
3381 info
->disk
.number
= -1;
3382 info
->disk
.state
= 0;
3384 info
->recovery_start
= MaxSector
;
3385 info
->recovery_blocked
= imsm_reshape_blocks_arrays_changes(st
->sb
);
3386 info
->bb
.supported
= 1;
3388 /* do we have the all the insync disks that we expect? */
3389 mpb
= super
->anchor
;
3390 info
->events
= __le32_to_cpu(mpb
->generation_num
);
3392 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
3393 struct imsm_dev
*dev
= get_imsm_dev(super
, i
);
3394 int failed
, enough
, j
, missing
= 0;
3395 struct imsm_map
*map
;
3398 failed
= imsm_count_failed(super
, dev
, MAP_0
);
3399 state
= imsm_check_degraded(super
, dev
, failed
, MAP_0
);
3400 map
= get_imsm_map(dev
, MAP_0
);
3402 /* any newly missing disks?
3403 * (catches single-degraded vs double-degraded)
3405 for (j
= 0; j
< map
->num_members
; j
++) {
3406 __u32 ord
= get_imsm_ord_tbl_ent(dev
, j
, MAP_0
);
3407 __u32 idx
= ord_to_idx(ord
);
3409 if (!(ord
& IMSM_ORD_REBUILD
) &&
3410 get_imsm_missing(super
, idx
)) {
3416 if (state
== IMSM_T_STATE_FAILED
)
3418 else if (state
== IMSM_T_STATE_DEGRADED
&&
3419 (state
!= map
->map_state
|| missing
))
3421 else /* we're normal, or already degraded */
3423 if (is_gen_migration(dev
) && missing
) {
3424 /* during general migration we need all disks
3425 * that process is running on.
3426 * No new missing disk is allowed.
3430 /* no more checks necessary
3434 /* in the missing/failed disk case check to see
3435 * if at least one array is runnable
3437 max_enough
= max(max_enough
, enough
);
3439 dprintf("enough: %d\n", max_enough
);
3440 info
->container_enough
= max_enough
;
3443 __u32 reserved
= imsm_reserved_sectors(super
, super
->disks
);
3445 disk
= &super
->disks
->disk
;
3446 info
->data_offset
= total_blocks(&super
->disks
->disk
) - reserved
;
3447 info
->component_size
= reserved
;
3448 info
->disk
.state
= is_configured(disk
) ? (1 << MD_DISK_ACTIVE
) : 0;
3449 /* we don't change info->disk.raid_disk here because
3450 * this state will be finalized in mdmon after we have
3451 * found the 'most fresh' version of the metadata
3453 info
->disk
.state
|= is_failed(disk
) ? (1 << MD_DISK_FAULTY
) : 0;
3454 info
->disk
.state
|= is_spare(disk
) ? 0 : (1 << MD_DISK_SYNC
);
3457 /* only call uuid_from_super_imsm when this disk is part of a populated container,
3458 * ->compare_super may have updated the 'num_raid_devs' field for spares
3460 if (info
->disk
.state
& (1 << MD_DISK_SYNC
) || super
->anchor
->num_raid_devs
)
3461 uuid_from_super_imsm(st
, info
->uuid
);
3463 memcpy(info
->uuid
, uuid_zero
, sizeof(uuid_zero
));
3465 /* I don't know how to compute 'map' on imsm, so use safe default */
3468 for (i
= 0; i
< map_disks
; i
++)
3474 /* allocates memory and fills disk in mdinfo structure
3475 * for each disk in array */
3476 struct mdinfo
*getinfo_super_disks_imsm(struct supertype
*st
)
3478 struct mdinfo
*mddev
;
3479 struct intel_super
*super
= st
->sb
;
3480 struct imsm_disk
*disk
;
3483 if (!super
|| !super
->disks
)
3486 mddev
= xcalloc(1, sizeof(*mddev
));
3490 tmp
= xcalloc(1, sizeof(*tmp
));
3492 tmp
->next
= mddev
->devs
;
3494 tmp
->disk
.number
= count
++;
3495 tmp
->disk
.major
= dl
->major
;
3496 tmp
->disk
.minor
= dl
->minor
;
3497 tmp
->disk
.state
= is_configured(disk
) ?
3498 (1 << MD_DISK_ACTIVE
) : 0;
3499 tmp
->disk
.state
|= is_failed(disk
) ? (1 << MD_DISK_FAULTY
) : 0;
3500 tmp
->disk
.state
|= is_spare(disk
) ? 0 : (1 << MD_DISK_SYNC
);
3501 tmp
->disk
.raid_disk
= -1;
3507 static int update_super_imsm(struct supertype
*st
, struct mdinfo
*info
,
3508 char *update
, char *devname
, int verbose
,
3509 int uuid_set
, char *homehost
)
3511 /* For 'assemble' and 'force' we need to return non-zero if any
3512 * change was made. For others, the return value is ignored.
3513 * Update options are:
3514 * force-one : This device looks a bit old but needs to be included,
3515 * update age info appropriately.
3516 * assemble: clear any 'faulty' flag to allow this device to
3518 * force-array: Array is degraded but being forced, mark it clean
3519 * if that will be needed to assemble it.
3521 * newdev: not used ????
3522 * grow: Array has gained a new device - this is currently for
3524 * resync: mark as dirty so a resync will happen.
3525 * name: update the name - preserving the homehost
3526 * uuid: Change the uuid of the array to match watch is given
3528 * Following are not relevant for this imsm:
3529 * sparc2.2 : update from old dodgey metadata
3530 * super-minor: change the preferred_minor number
3531 * summaries: update redundant counters.
3532 * homehost: update the recorded homehost
3533 * _reshape_progress: record new reshape_progress position.
3536 struct intel_super
*super
= st
->sb
;
3537 struct imsm_super
*mpb
;
3539 /* we can only update container info */
3540 if (!super
|| super
->current_vol
>= 0 || !super
->anchor
)
3543 mpb
= super
->anchor
;
3545 if (strcmp(update
, "uuid") == 0) {
3546 /* We take this to mean that the family_num should be updated.
3547 * However that is much smaller than the uuid so we cannot really
3548 * allow an explicit uuid to be given. And it is hard to reliably
3550 * So if !uuid_set we know the current uuid is random and just used
3551 * the first 'int' and copy it to the other 3 positions.
3552 * Otherwise we require the 4 'int's to be the same as would be the
3553 * case if we are using a random uuid. So an explicit uuid will be
3554 * accepted as long as all for ints are the same... which shouldn't hurt
3557 info
->uuid
[1] = info
->uuid
[2] = info
->uuid
[3] = info
->uuid
[0];
3560 if (info
->uuid
[0] != info
->uuid
[1] ||
3561 info
->uuid
[1] != info
->uuid
[2] ||
3562 info
->uuid
[2] != info
->uuid
[3])
3568 mpb
->orig_family_num
= info
->uuid
[0];
3569 } else if (strcmp(update
, "assemble") == 0)
3574 /* successful update? recompute checksum */
3576 mpb
->check_sum
= __le32_to_cpu(__gen_imsm_checksum(mpb
));
3581 static size_t disks_to_mpb_size(int disks
)
3585 size
= sizeof(struct imsm_super
);
3586 size
+= (disks
- 1) * sizeof(struct imsm_disk
);
3587 size
+= 2 * sizeof(struct imsm_dev
);
3588 /* up to 2 maps per raid device (-2 for imsm_maps in imsm_dev */
3589 size
+= (4 - 2) * sizeof(struct imsm_map
);
3590 /* 4 possible disk_ord_tbl's */
3591 size
+= 4 * (disks
- 1) * sizeof(__u32
);
3592 /* maximum bbm log */
3593 size
+= sizeof(struct bbm_log
);
3598 static __u64
avail_size_imsm(struct supertype
*st
, __u64 devsize
,
3599 unsigned long long data_offset
)
3601 if (devsize
< (MPB_SECTOR_CNT
+ IMSM_RESERVED_SECTORS
))
3604 return devsize
- (MPB_SECTOR_CNT
+ IMSM_RESERVED_SECTORS
);
3607 static void free_devlist(struct intel_super
*super
)
3609 struct intel_dev
*dv
;
3611 while (super
->devlist
) {
3612 dv
= super
->devlist
->next
;
3613 free(super
->devlist
->dev
);
3614 free(super
->devlist
);
3615 super
->devlist
= dv
;
3619 static void imsm_copy_dev(struct imsm_dev
*dest
, struct imsm_dev
*src
)
3621 memcpy(dest
, src
, sizeof_imsm_dev(src
, 0));
3624 static int compare_super_imsm(struct supertype
*st
, struct supertype
*tst
)
3628 * 0 same, or first was empty, and second was copied
3629 * 1 second had wrong number
3631 * 3 wrong other info
3633 struct intel_super
*first
= st
->sb
;
3634 struct intel_super
*sec
= tst
->sb
;
3641 /* in platform dependent environment test if the disks
3642 * use the same Intel hba
3643 * If not on Intel hba at all, allow anything.
3645 if (!check_env("IMSM_NO_PLATFORM") && first
->hba
&& sec
->hba
) {
3646 if (first
->hba
->type
!= sec
->hba
->type
) {
3648 "HBAs of devices do not match %s != %s\n",
3649 get_sys_dev_type(first
->hba
->type
),
3650 get_sys_dev_type(sec
->hba
->type
));
3653 if (first
->orom
!= sec
->orom
) {
3655 "HBAs of devices do not match %s != %s\n",
3656 first
->hba
->pci_id
, sec
->hba
->pci_id
);
3661 /* if an anchor does not have num_raid_devs set then it is a free
3664 if (first
->anchor
->num_raid_devs
> 0 &&
3665 sec
->anchor
->num_raid_devs
> 0) {
3666 /* Determine if these disks might ever have been
3667 * related. Further disambiguation can only take place
3668 * in load_super_imsm_all
3670 __u32 first_family
= first
->anchor
->orig_family_num
;
3671 __u32 sec_family
= sec
->anchor
->orig_family_num
;
3673 if (memcmp(first
->anchor
->sig
, sec
->anchor
->sig
,
3674 MAX_SIGNATURE_LENGTH
) != 0)
3677 if (first_family
== 0)
3678 first_family
= first
->anchor
->family_num
;
3679 if (sec_family
== 0)
3680 sec_family
= sec
->anchor
->family_num
;
3682 if (first_family
!= sec_family
)
3687 /* if 'first' is a spare promote it to a populated mpb with sec's
3690 if (first
->anchor
->num_raid_devs
== 0 &&
3691 sec
->anchor
->num_raid_devs
> 0) {
3693 struct intel_dev
*dv
;
3694 struct imsm_dev
*dev
;
3696 /* we need to copy raid device info from sec if an allocation
3697 * fails here we don't associate the spare
3699 for (i
= 0; i
< sec
->anchor
->num_raid_devs
; i
++) {
3700 dv
= xmalloc(sizeof(*dv
));
3701 dev
= xmalloc(sizeof_imsm_dev(get_imsm_dev(sec
, i
), 1));
3704 dv
->next
= first
->devlist
;
3705 first
->devlist
= dv
;
3707 if (i
< sec
->anchor
->num_raid_devs
) {
3708 /* allocation failure */
3709 free_devlist(first
);
3710 pr_err("imsm: failed to associate spare\n");
3713 first
->anchor
->num_raid_devs
= sec
->anchor
->num_raid_devs
;
3714 first
->anchor
->orig_family_num
= sec
->anchor
->orig_family_num
;
3715 first
->anchor
->family_num
= sec
->anchor
->family_num
;
3716 memcpy(first
->anchor
->sig
, sec
->anchor
->sig
, MAX_SIGNATURE_LENGTH
);
3717 for (i
= 0; i
< sec
->anchor
->num_raid_devs
; i
++)
3718 imsm_copy_dev(get_imsm_dev(first
, i
), get_imsm_dev(sec
, i
));
3724 static void fd2devname(int fd
, char *name
)
3728 char dname
[PATH_MAX
];
3733 if (fstat(fd
, &st
) != 0)
3735 sprintf(path
, "/sys/dev/block/%d:%d",
3736 major(st
.st_rdev
), minor(st
.st_rdev
));
3738 rv
= readlink(path
, dname
, sizeof(dname
)-1);
3743 nm
= strrchr(dname
, '/');
3746 snprintf(name
, MAX_RAID_SERIAL_LEN
, "/dev/%s", nm
);
3750 static int nvme_get_serial(int fd
, void *buf
, size_t buf_len
)
3753 char *name
= fd2kname(fd
);
3758 if (strncmp(name
, "nvme", 4) != 0)
3761 snprintf(path
, sizeof(path
) - 1, "/sys/block/%s/device/serial", name
);
3763 return load_sys(path
, buf
, buf_len
);
3766 extern int scsi_get_serial(int fd
, void *buf
, size_t buf_len
);
3768 static int imsm_read_serial(int fd
, char *devname
,
3769 __u8 serial
[MAX_RAID_SERIAL_LEN
])
3778 memset(buf
, 0, sizeof(buf
));
3780 rv
= nvme_get_serial(fd
, buf
, sizeof(buf
));
3783 rv
= scsi_get_serial(fd
, buf
, sizeof(buf
));
3785 if (rv
&& check_env("IMSM_DEVNAME_AS_SERIAL")) {
3786 memset(serial
, 0, MAX_RAID_SERIAL_LEN
);
3787 fd2devname(fd
, (char *) serial
);
3793 pr_err("Failed to retrieve serial for %s\n",
3798 /* trim all whitespace and non-printable characters and convert
3801 for (i
= 0, dest
= buf
; i
< sizeof(buf
) && buf
[i
]; i
++) {
3804 /* ':' is reserved for use in placeholder serial
3805 * numbers for missing disks
3816 /* truncate leading characters */
3817 if (len
> MAX_RAID_SERIAL_LEN
) {
3818 dest
+= len
- MAX_RAID_SERIAL_LEN
;
3819 len
= MAX_RAID_SERIAL_LEN
;
3822 memset(serial
, 0, MAX_RAID_SERIAL_LEN
);
3823 memcpy(serial
, dest
, len
);
3828 static int serialcmp(__u8
*s1
, __u8
*s2
)
3830 return strncmp((char *) s1
, (char *) s2
, MAX_RAID_SERIAL_LEN
);
3833 static void serialcpy(__u8
*dest
, __u8
*src
)
3835 strncpy((char *) dest
, (char *) src
, MAX_RAID_SERIAL_LEN
);
3838 static struct dl
*serial_to_dl(__u8
*serial
, struct intel_super
*super
)
3842 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
3843 if (serialcmp(dl
->serial
, serial
) == 0)
3849 static struct imsm_disk
*
3850 __serial_to_disk(__u8
*serial
, struct imsm_super
*mpb
, int *idx
)
3854 for (i
= 0; i
< mpb
->num_disks
; i
++) {
3855 struct imsm_disk
*disk
= __get_imsm_disk(mpb
, i
);
3857 if (serialcmp(disk
->serial
, serial
) == 0) {
3868 load_imsm_disk(int fd
, struct intel_super
*super
, char *devname
, int keep_fd
)
3870 struct imsm_disk
*disk
;
3875 __u8 serial
[MAX_RAID_SERIAL_LEN
];
3877 rv
= imsm_read_serial(fd
, devname
, serial
);
3882 dl
= xcalloc(1, sizeof(*dl
));
3885 dl
->major
= major(stb
.st_rdev
);
3886 dl
->minor
= minor(stb
.st_rdev
);
3887 dl
->next
= super
->disks
;
3888 dl
->fd
= keep_fd
? fd
: -1;
3889 assert(super
->disks
== NULL
);
3891 serialcpy(dl
->serial
, serial
);
3894 fd2devname(fd
, name
);
3896 dl
->devname
= xstrdup(devname
);
3898 dl
->devname
= xstrdup(name
);
3900 /* look up this disk's index in the current anchor */
3901 disk
= __serial_to_disk(dl
->serial
, super
->anchor
, &dl
->index
);
3904 /* only set index on disks that are a member of a
3905 * populated contianer, i.e. one with raid_devs
3907 if (is_failed(&dl
->disk
))
3909 else if (is_spare(&dl
->disk
))
3917 /* When migrating map0 contains the 'destination' state while map1
3918 * contains the current state. When not migrating map0 contains the
3919 * current state. This routine assumes that map[0].map_state is set to
3920 * the current array state before being called.
3922 * Migration is indicated by one of the following states
3923 * 1/ Idle (migr_state=0 map0state=normal||unitialized||degraded||failed)
3924 * 2/ Initialize (migr_state=1 migr_type=MIGR_INIT map0state=normal
3925 * map1state=unitialized)
3926 * 3/ Repair (Resync) (migr_state=1 migr_type=MIGR_REPAIR map0state=normal
3928 * 4/ Rebuild (migr_state=1 migr_type=MIGR_REBUILD map0state=normal
3929 * map1state=degraded)
3930 * 5/ Migration (mig_state=1 migr_type=MIGR_GEN_MIGR map0state=normal
3933 static void migrate(struct imsm_dev
*dev
, struct intel_super
*super
,
3934 __u8 to_state
, int migr_type
)
3936 struct imsm_map
*dest
;
3937 struct imsm_map
*src
= get_imsm_map(dev
, MAP_0
);
3939 dev
->vol
.migr_state
= 1;
3940 set_migr_type(dev
, migr_type
);
3941 dev
->vol
.curr_migr_unit
= 0;
3942 dest
= get_imsm_map(dev
, MAP_1
);
3944 /* duplicate and then set the target end state in map[0] */
3945 memcpy(dest
, src
, sizeof_imsm_map(src
));
3946 if (migr_type
== MIGR_REBUILD
|| migr_type
== MIGR_GEN_MIGR
) {
3950 for (i
= 0; i
< src
->num_members
; i
++) {
3951 ord
= __le32_to_cpu(src
->disk_ord_tbl
[i
]);
3952 set_imsm_ord_tbl_ent(src
, i
, ord_to_idx(ord
));
3956 if (migr_type
== MIGR_GEN_MIGR
)
3957 /* Clear migration record */
3958 memset(super
->migr_rec
, 0, sizeof(struct migr_record
));
3960 src
->map_state
= to_state
;
3963 static void end_migration(struct imsm_dev
*dev
, struct intel_super
*super
,
3966 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
3967 struct imsm_map
*prev
= get_imsm_map(dev
, dev
->vol
.migr_state
== 0 ?
3971 /* merge any IMSM_ORD_REBUILD bits that were not successfully
3972 * completed in the last migration.
3974 * FIXME add support for raid-level-migration
3976 if (map_state
!= map
->map_state
&& (is_gen_migration(dev
) == 0) &&
3977 prev
->map_state
!= IMSM_T_STATE_UNINITIALIZED
) {
3978 /* when final map state is other than expected
3979 * merge maps (not for migration)
3983 for (i
= 0; i
< prev
->num_members
; i
++)
3984 for (j
= 0; j
< map
->num_members
; j
++)
3985 /* during online capacity expansion
3986 * disks position can be changed
3987 * if takeover is used
3989 if (ord_to_idx(map
->disk_ord_tbl
[j
]) ==
3990 ord_to_idx(prev
->disk_ord_tbl
[i
])) {
3991 map
->disk_ord_tbl
[j
] |=
3992 prev
->disk_ord_tbl
[i
];
3995 failed
= imsm_count_failed(super
, dev
, MAP_0
);
3996 map_state
= imsm_check_degraded(super
, dev
, failed
, MAP_0
);
3999 dev
->vol
.migr_state
= 0;
4000 set_migr_type(dev
, 0);
4001 dev
->vol
.curr_migr_unit
= 0;
4002 map
->map_state
= map_state
;
4006 static int parse_raid_devices(struct intel_super
*super
)
4009 struct imsm_dev
*dev_new
;
4010 size_t len
, len_migr
;
4012 size_t space_needed
= 0;
4013 struct imsm_super
*mpb
= super
->anchor
;
4015 for (i
= 0; i
< super
->anchor
->num_raid_devs
; i
++) {
4016 struct imsm_dev
*dev_iter
= __get_imsm_dev(super
->anchor
, i
);
4017 struct intel_dev
*dv
;
4019 len
= sizeof_imsm_dev(dev_iter
, 0);
4020 len_migr
= sizeof_imsm_dev(dev_iter
, 1);
4022 space_needed
+= len_migr
- len
;
4024 dv
= xmalloc(sizeof(*dv
));
4025 if (max_len
< len_migr
)
4027 if (max_len
> len_migr
)
4028 space_needed
+= max_len
- len_migr
;
4029 dev_new
= xmalloc(max_len
);
4030 imsm_copy_dev(dev_new
, dev_iter
);
4033 dv
->next
= super
->devlist
;
4034 super
->devlist
= dv
;
4037 /* ensure that super->buf is large enough when all raid devices
4040 if (__le32_to_cpu(mpb
->mpb_size
) + space_needed
> super
->len
) {
4043 len
= ROUND_UP(__le32_to_cpu(mpb
->mpb_size
) + space_needed
,
4044 super
->sector_size
);
4045 if (posix_memalign(&buf
, MAX_SECTOR_SIZE
, len
) != 0)
4048 memcpy(buf
, super
->buf
, super
->len
);
4049 memset(buf
+ super
->len
, 0, len
- super
->len
);
4055 super
->extra_space
+= space_needed
;
4060 /*******************************************************************************
4061 * Function: check_mpb_migr_compatibility
4062 * Description: Function checks for unsupported migration features:
4063 * - migration optimization area (pba_of_lba0)
4064 * - descending reshape (ascending_migr)
4066 * super : imsm metadata information
4068 * 0 : migration is compatible
4069 * -1 : migration is not compatible
4070 ******************************************************************************/
4071 int check_mpb_migr_compatibility(struct intel_super
*super
)
4073 struct imsm_map
*map0
, *map1
;
4074 struct migr_record
*migr_rec
= super
->migr_rec
;
4077 for (i
= 0; i
< super
->anchor
->num_raid_devs
; i
++) {
4078 struct imsm_dev
*dev_iter
= __get_imsm_dev(super
->anchor
, i
);
4081 dev_iter
->vol
.migr_state
== 1 &&
4082 dev_iter
->vol
.migr_type
== MIGR_GEN_MIGR
) {
4083 /* This device is migrating */
4084 map0
= get_imsm_map(dev_iter
, MAP_0
);
4085 map1
= get_imsm_map(dev_iter
, MAP_1
);
4086 if (pba_of_lba0(map0
) != pba_of_lba0(map1
))
4087 /* migration optimization area was used */
4089 if (migr_rec
->ascending_migr
== 0
4090 && migr_rec
->dest_depth_per_unit
> 0)
4091 /* descending reshape not supported yet */
4098 static void __free_imsm(struct intel_super
*super
, int free_disks
);
4100 /* load_imsm_mpb - read matrix metadata
4101 * allocates super->mpb to be freed by free_imsm
4103 static int load_imsm_mpb(int fd
, struct intel_super
*super
, char *devname
)
4105 unsigned long long dsize
;
4106 unsigned long long sectors
;
4107 unsigned int sector_size
= super
->sector_size
;
4109 struct imsm_super
*anchor
;
4112 get_dev_size(fd
, NULL
, &dsize
);
4113 if (dsize
< 2*sector_size
) {
4115 pr_err("%s: device to small for imsm\n",
4120 if (lseek64(fd
, dsize
- (sector_size
* 2), SEEK_SET
) < 0) {
4122 pr_err("Cannot seek to anchor block on %s: %s\n",
4123 devname
, strerror(errno
));
4127 if (posix_memalign((void **)&anchor
, sector_size
, sector_size
) != 0) {
4129 pr_err("Failed to allocate imsm anchor buffer on %s\n", devname
);
4132 if ((unsigned int)read(fd
, anchor
, sector_size
) != sector_size
) {
4134 pr_err("Cannot read anchor block on %s: %s\n",
4135 devname
, strerror(errno
));
4140 if (strncmp((char *) anchor
->sig
, MPB_SIGNATURE
, MPB_SIG_LEN
) != 0) {
4142 pr_err("no IMSM anchor on %s\n", devname
);
4147 __free_imsm(super
, 0);
4148 /* reload capability and hba */
4150 /* capability and hba must be updated with new super allocation */
4151 find_intel_hba_capability(fd
, super
, devname
);
4152 super
->len
= ROUND_UP(anchor
->mpb_size
, sector_size
);
4153 if (posix_memalign(&super
->buf
, MAX_SECTOR_SIZE
, super
->len
) != 0) {
4155 pr_err("unable to allocate %zu byte mpb buffer\n",
4160 memcpy(super
->buf
, anchor
, sector_size
);
4162 sectors
= mpb_sectors(anchor
, sector_size
) - 1;
4165 if (posix_memalign(&super
->migr_rec_buf
, sector_size
,
4166 MIGR_REC_BUF_SECTORS
*sector_size
) != 0) {
4167 pr_err("could not allocate migr_rec buffer\n");
4171 super
->clean_migration_record_by_mdmon
= 0;
4174 check_sum
= __gen_imsm_checksum(super
->anchor
);
4175 if (check_sum
!= __le32_to_cpu(super
->anchor
->check_sum
)) {
4177 pr_err("IMSM checksum %x != %x on %s\n",
4179 __le32_to_cpu(super
->anchor
->check_sum
),
4187 /* read the extended mpb */
4188 if (lseek64(fd
, dsize
- (sector_size
* (2 + sectors
)), SEEK_SET
) < 0) {
4190 pr_err("Cannot seek to extended mpb on %s: %s\n",
4191 devname
, strerror(errno
));
4195 if ((unsigned int)read(fd
, super
->buf
+ sector_size
,
4196 super
->len
- sector_size
) != super
->len
- sector_size
) {
4198 pr_err("Cannot read extended mpb on %s: %s\n",
4199 devname
, strerror(errno
));
4203 check_sum
= __gen_imsm_checksum(super
->anchor
);
4204 if (check_sum
!= __le32_to_cpu(super
->anchor
->check_sum
)) {
4206 pr_err("IMSM checksum %x != %x on %s\n",
4207 check_sum
, __le32_to_cpu(super
->anchor
->check_sum
),
4215 static int read_imsm_migr_rec(int fd
, struct intel_super
*super
);
4217 /* clears hi bits in metadata if MPB_ATTRIB_2TB_DISK not set */
4218 static void clear_hi(struct intel_super
*super
)
4220 struct imsm_super
*mpb
= super
->anchor
;
4222 if (mpb
->attributes
& MPB_ATTRIB_2TB_DISK
)
4224 for (i
= 0; i
< mpb
->num_disks
; ++i
) {
4225 struct imsm_disk
*disk
= &mpb
->disk
[i
];
4226 disk
->total_blocks_hi
= 0;
4228 for (i
= 0; i
< mpb
->num_raid_devs
; ++i
) {
4229 struct imsm_dev
*dev
= get_imsm_dev(super
, i
);
4232 for (n
= 0; n
< 2; ++n
) {
4233 struct imsm_map
*map
= get_imsm_map(dev
, n
);
4236 map
->pba_of_lba0_hi
= 0;
4237 map
->blocks_per_member_hi
= 0;
4238 map
->num_data_stripes_hi
= 0;
4244 load_and_parse_mpb(int fd
, struct intel_super
*super
, char *devname
, int keep_fd
)
4248 err
= load_imsm_mpb(fd
, super
, devname
);
4251 if (super
->sector_size
== 4096)
4252 convert_from_4k(super
);
4253 err
= load_imsm_disk(fd
, super
, devname
, keep_fd
);
4256 err
= parse_raid_devices(super
);
4259 err
= load_bbm_log(super
);
4264 static void __free_imsm_disk(struct dl
*d
)
4276 static void free_imsm_disks(struct intel_super
*super
)
4280 while (super
->disks
) {
4282 super
->disks
= d
->next
;
4283 __free_imsm_disk(d
);
4285 while (super
->disk_mgmt_list
) {
4286 d
= super
->disk_mgmt_list
;
4287 super
->disk_mgmt_list
= d
->next
;
4288 __free_imsm_disk(d
);
4290 while (super
->missing
) {
4292 super
->missing
= d
->next
;
4293 __free_imsm_disk(d
);
4298 /* free all the pieces hanging off of a super pointer */
4299 static void __free_imsm(struct intel_super
*super
, int free_disks
)
4301 struct intel_hba
*elem
, *next
;
4307 /* unlink capability description */
4309 if (super
->migr_rec_buf
) {
4310 free(super
->migr_rec_buf
);
4311 super
->migr_rec_buf
= NULL
;
4314 free_imsm_disks(super
);
4315 free_devlist(super
);
4319 free((void *)elem
->path
);
4325 free(super
->bbm_log
);
4329 static void free_imsm(struct intel_super
*super
)
4331 __free_imsm(super
, 1);
4332 free(super
->bb
.entries
);
4336 static void free_super_imsm(struct supertype
*st
)
4338 struct intel_super
*super
= st
->sb
;
4347 static struct intel_super
*alloc_super(void)
4349 struct intel_super
*super
= xcalloc(1, sizeof(*super
));
4351 super
->current_vol
= -1;
4352 super
->create_offset
= ~((unsigned long long) 0);
4354 super
->bb
.entries
= xmalloc(BBM_LOG_MAX_ENTRIES
*
4355 sizeof(struct md_bb_entry
));
4356 if (!super
->bb
.entries
) {
4365 * find and allocate hba and OROM/EFI based on valid fd of RAID component device
4367 static int find_intel_hba_capability(int fd
, struct intel_super
*super
, char *devname
)
4369 struct sys_dev
*hba_name
;
4372 if (fd
< 0 || check_env("IMSM_NO_PLATFORM")) {
4377 hba_name
= find_disk_attached_hba(fd
, NULL
);
4380 pr_err("%s is not attached to Intel(R) RAID controller.\n",
4384 rv
= attach_hba_to_super(super
, hba_name
);
4387 struct intel_hba
*hba
= super
->hba
;
4389 pr_err("%s is attached to Intel(R) %s %s (%s),\n"
4390 " but the container is assigned to Intel(R) %s %s (",
4392 get_sys_dev_type(hba_name
->type
),
4393 hba_name
->type
== SYS_DEV_VMD
? "domain" : "RAID controller",
4394 hba_name
->pci_id
? : "Err!",
4395 get_sys_dev_type(super
->hba
->type
),
4396 hba
->type
== SYS_DEV_VMD
? "domain" : "RAID controller");
4399 fprintf(stderr
, "%s", hba
->pci_id
? : "Err!");
4401 fprintf(stderr
, ", ");
4404 fprintf(stderr
, ").\n"
4405 " Mixing devices attached to different %s is not allowed.\n",
4406 hba_name
->type
== SYS_DEV_VMD
? "VMD domains" : "controllers");
4410 super
->orom
= find_imsm_capability(hba_name
);
4417 /* find_missing - helper routine for load_super_imsm_all that identifies
4418 * disks that have disappeared from the system. This routine relies on
4419 * the mpb being uptodate, which it is at load time.
4421 static int find_missing(struct intel_super
*super
)
4424 struct imsm_super
*mpb
= super
->anchor
;
4426 struct imsm_disk
*disk
;
4428 for (i
= 0; i
< mpb
->num_disks
; i
++) {
4429 disk
= __get_imsm_disk(mpb
, i
);
4430 dl
= serial_to_dl(disk
->serial
, super
);
4434 dl
= xmalloc(sizeof(*dl
));
4438 dl
->devname
= xstrdup("missing");
4440 serialcpy(dl
->serial
, disk
->serial
);
4443 dl
->next
= super
->missing
;
4444 super
->missing
= dl
;
4451 static struct intel_disk
*disk_list_get(__u8
*serial
, struct intel_disk
*disk_list
)
4453 struct intel_disk
*idisk
= disk_list
;
4456 if (serialcmp(idisk
->disk
.serial
, serial
) == 0)
4458 idisk
= idisk
->next
;
4464 static int __prep_thunderdome(struct intel_super
**table
, int tbl_size
,
4465 struct intel_super
*super
,
4466 struct intel_disk
**disk_list
)
4468 struct imsm_disk
*d
= &super
->disks
->disk
;
4469 struct imsm_super
*mpb
= super
->anchor
;
4472 for (i
= 0; i
< tbl_size
; i
++) {
4473 struct imsm_super
*tbl_mpb
= table
[i
]->anchor
;
4474 struct imsm_disk
*tbl_d
= &table
[i
]->disks
->disk
;
4476 if (tbl_mpb
->family_num
== mpb
->family_num
) {
4477 if (tbl_mpb
->check_sum
== mpb
->check_sum
) {
4478 dprintf("mpb from %d:%d matches %d:%d\n",
4479 super
->disks
->major
,
4480 super
->disks
->minor
,
4481 table
[i
]->disks
->major
,
4482 table
[i
]->disks
->minor
);
4486 if (((is_configured(d
) && !is_configured(tbl_d
)) ||
4487 is_configured(d
) == is_configured(tbl_d
)) &&
4488 tbl_mpb
->generation_num
< mpb
->generation_num
) {
4489 /* current version of the mpb is a
4490 * better candidate than the one in
4491 * super_table, but copy over "cross
4492 * generational" status
4494 struct intel_disk
*idisk
;
4496 dprintf("mpb from %d:%d replaces %d:%d\n",
4497 super
->disks
->major
,
4498 super
->disks
->minor
,
4499 table
[i
]->disks
->major
,
4500 table
[i
]->disks
->minor
);
4502 idisk
= disk_list_get(tbl_d
->serial
, *disk_list
);
4503 if (idisk
&& is_failed(&idisk
->disk
))
4504 tbl_d
->status
|= FAILED_DISK
;
4507 struct intel_disk
*idisk
;
4508 struct imsm_disk
*disk
;
4510 /* tbl_mpb is more up to date, but copy
4511 * over cross generational status before
4514 disk
= __serial_to_disk(d
->serial
, mpb
, NULL
);
4515 if (disk
&& is_failed(disk
))
4516 d
->status
|= FAILED_DISK
;
4518 idisk
= disk_list_get(d
->serial
, *disk_list
);
4521 if (disk
&& is_configured(disk
))
4522 idisk
->disk
.status
|= CONFIGURED_DISK
;
4525 dprintf("mpb from %d:%d prefer %d:%d\n",
4526 super
->disks
->major
,
4527 super
->disks
->minor
,
4528 table
[i
]->disks
->major
,
4529 table
[i
]->disks
->minor
);
4537 table
[tbl_size
++] = super
;
4541 /* update/extend the merged list of imsm_disk records */
4542 for (j
= 0; j
< mpb
->num_disks
; j
++) {
4543 struct imsm_disk
*disk
= __get_imsm_disk(mpb
, j
);
4544 struct intel_disk
*idisk
;
4546 idisk
= disk_list_get(disk
->serial
, *disk_list
);
4548 idisk
->disk
.status
|= disk
->status
;
4549 if (is_configured(&idisk
->disk
) ||
4550 is_failed(&idisk
->disk
))
4551 idisk
->disk
.status
&= ~(SPARE_DISK
);
4553 idisk
= xcalloc(1, sizeof(*idisk
));
4554 idisk
->owner
= IMSM_UNKNOWN_OWNER
;
4555 idisk
->disk
= *disk
;
4556 idisk
->next
= *disk_list
;
4560 if (serialcmp(idisk
->disk
.serial
, d
->serial
) == 0)
4567 static struct intel_super
*
4568 validate_members(struct intel_super
*super
, struct intel_disk
*disk_list
,
4571 struct imsm_super
*mpb
= super
->anchor
;
4575 for (i
= 0; i
< mpb
->num_disks
; i
++) {
4576 struct imsm_disk
*disk
= __get_imsm_disk(mpb
, i
);
4577 struct intel_disk
*idisk
;
4579 idisk
= disk_list_get(disk
->serial
, disk_list
);
4581 if (idisk
->owner
== owner
||
4582 idisk
->owner
== IMSM_UNKNOWN_OWNER
)
4585 dprintf("'%.16s' owner %d != %d\n",
4586 disk
->serial
, idisk
->owner
,
4589 dprintf("unknown disk %x [%d]: %.16s\n",
4590 __le32_to_cpu(mpb
->family_num
), i
,
4596 if (ok_count
== mpb
->num_disks
)
4601 static void show_conflicts(__u32 family_num
, struct intel_super
*super_list
)
4603 struct intel_super
*s
;
4605 for (s
= super_list
; s
; s
= s
->next
) {
4606 if (family_num
!= s
->anchor
->family_num
)
4608 pr_err("Conflict, offlining family %#x on '%s'\n",
4609 __le32_to_cpu(family_num
), s
->disks
->devname
);
4613 static struct intel_super
*
4614 imsm_thunderdome(struct intel_super
**super_list
, int len
)
4616 struct intel_super
*super_table
[len
];
4617 struct intel_disk
*disk_list
= NULL
;
4618 struct intel_super
*champion
, *spare
;
4619 struct intel_super
*s
, **del
;
4624 memset(super_table
, 0, sizeof(super_table
));
4625 for (s
= *super_list
; s
; s
= s
->next
)
4626 tbl_size
= __prep_thunderdome(super_table
, tbl_size
, s
, &disk_list
);
4628 for (i
= 0; i
< tbl_size
; i
++) {
4629 struct imsm_disk
*d
;
4630 struct intel_disk
*idisk
;
4631 struct imsm_super
*mpb
= super_table
[i
]->anchor
;
4634 d
= &s
->disks
->disk
;
4636 /* 'd' must appear in merged disk list for its
4637 * configuration to be valid
4639 idisk
= disk_list_get(d
->serial
, disk_list
);
4640 if (idisk
&& idisk
->owner
== i
)
4641 s
= validate_members(s
, disk_list
, i
);
4646 dprintf("marking family: %#x from %d:%d offline\n",
4648 super_table
[i
]->disks
->major
,
4649 super_table
[i
]->disks
->minor
);
4653 /* This is where the mdadm implementation differs from the Windows
4654 * driver which has no strict concept of a container. We can only
4655 * assemble one family from a container, so when returning a prodigal
4656 * array member to this system the code will not be able to disambiguate
4657 * the container contents that should be assembled ("foreign" versus
4658 * "local"). It requires user intervention to set the orig_family_num
4659 * to a new value to establish a new container. The Windows driver in
4660 * this situation fixes up the volume name in place and manages the
4661 * foreign array as an independent entity.
4666 for (i
= 0; i
< tbl_size
; i
++) {
4667 struct intel_super
*tbl_ent
= super_table
[i
];
4673 if (tbl_ent
->anchor
->num_raid_devs
== 0) {
4678 if (s
&& !is_spare
) {
4679 show_conflicts(tbl_ent
->anchor
->family_num
, *super_list
);
4681 } else if (!s
&& !is_spare
)
4694 pr_err("Chose family %#x on '%s', assemble conflicts to new container with '--update=uuid'\n",
4695 __le32_to_cpu(s
->anchor
->family_num
), s
->disks
->devname
);
4697 /* collect all dl's onto 'champion', and update them to
4698 * champion's version of the status
4700 for (s
= *super_list
; s
; s
= s
->next
) {
4701 struct imsm_super
*mpb
= champion
->anchor
;
4702 struct dl
*dl
= s
->disks
;
4707 mpb
->attributes
|= s
->anchor
->attributes
& MPB_ATTRIB_2TB_DISK
;
4709 for (i
= 0; i
< mpb
->num_disks
; i
++) {
4710 struct imsm_disk
*disk
;
4712 disk
= __serial_to_disk(dl
->serial
, mpb
, &dl
->index
);
4715 /* only set index on disks that are a member of
4716 * a populated contianer, i.e. one with
4719 if (is_failed(&dl
->disk
))
4721 else if (is_spare(&dl
->disk
))
4727 if (i
>= mpb
->num_disks
) {
4728 struct intel_disk
*idisk
;
4730 idisk
= disk_list_get(dl
->serial
, disk_list
);
4731 if (idisk
&& is_spare(&idisk
->disk
) &&
4732 !is_failed(&idisk
->disk
) && !is_configured(&idisk
->disk
))
4740 dl
->next
= champion
->disks
;
4741 champion
->disks
= dl
;
4745 /* delete 'champion' from super_list */
4746 for (del
= super_list
; *del
; ) {
4747 if (*del
== champion
) {
4748 *del
= (*del
)->next
;
4751 del
= &(*del
)->next
;
4753 champion
->next
= NULL
;
4757 struct intel_disk
*idisk
= disk_list
;
4759 disk_list
= disk_list
->next
;
4767 get_sra_super_block(int fd
, struct intel_super
**super_list
, char *devname
, int *max
, int keep_fd
);
4768 static int get_super_block(struct intel_super
**super_list
, char *devnm
, char *devname
,
4769 int major
, int minor
, int keep_fd
);
4771 get_devlist_super_block(struct md_list
*devlist
, struct intel_super
**super_list
,
4772 int *max
, int keep_fd
);
4774 static int load_super_imsm_all(struct supertype
*st
, int fd
, void **sbp
,
4775 char *devname
, struct md_list
*devlist
,
4778 struct intel_super
*super_list
= NULL
;
4779 struct intel_super
*super
= NULL
;
4784 /* 'fd' is an opened container */
4785 err
= get_sra_super_block(fd
, &super_list
, devname
, &i
, keep_fd
);
4787 /* get super block from devlist devices */
4788 err
= get_devlist_super_block(devlist
, &super_list
, &i
, keep_fd
);
4791 /* all mpbs enter, maybe one leaves */
4792 super
= imsm_thunderdome(&super_list
, i
);
4798 if (find_missing(super
) != 0) {
4804 /* load migration record */
4805 err
= load_imsm_migr_rec(super
, NULL
);
4807 /* migration is in progress,
4808 * but migr_rec cannot be loaded,
4814 /* Check migration compatibility */
4815 if (err
== 0 && check_mpb_migr_compatibility(super
) != 0) {
4816 pr_err("Unsupported migration detected");
4818 fprintf(stderr
, " on %s\n", devname
);
4820 fprintf(stderr
, " (IMSM).\n");
4829 while (super_list
) {
4830 struct intel_super
*s
= super_list
;
4832 super_list
= super_list
->next
;
4841 strcpy(st
->container_devnm
, fd2devnm(fd
));
4843 st
->container_devnm
[0] = 0;
4844 if (err
== 0 && st
->ss
== NULL
) {
4845 st
->ss
= &super_imsm
;
4846 st
->minor_version
= 0;
4847 st
->max_devs
= IMSM_MAX_DEVICES
;
4853 get_devlist_super_block(struct md_list
*devlist
, struct intel_super
**super_list
,
4854 int *max
, int keep_fd
)
4856 struct md_list
*tmpdev
;
4860 for (i
= 0, tmpdev
= devlist
; tmpdev
; tmpdev
= tmpdev
->next
) {
4861 if (tmpdev
->used
!= 1)
4863 if (tmpdev
->container
== 1) {
4865 int fd
= dev_open(tmpdev
->devname
, O_RDONLY
|O_EXCL
);
4867 pr_err("cannot open device %s: %s\n",
4868 tmpdev
->devname
, strerror(errno
));
4872 err
= get_sra_super_block(fd
, super_list
,
4873 tmpdev
->devname
, &lmax
,
4882 int major
= major(tmpdev
->st_rdev
);
4883 int minor
= minor(tmpdev
->st_rdev
);
4884 err
= get_super_block(super_list
,
4901 static int get_super_block(struct intel_super
**super_list
, char *devnm
, char *devname
,
4902 int major
, int minor
, int keep_fd
)
4904 struct intel_super
*s
;
4916 sprintf(nm
, "%d:%d", major
, minor
);
4917 dfd
= dev_open(nm
, O_RDWR
);
4923 get_dev_sector_size(dfd
, NULL
, &s
->sector_size
);
4924 find_intel_hba_capability(dfd
, s
, devname
);
4925 err
= load_and_parse_mpb(dfd
, s
, NULL
, keep_fd
);
4927 /* retry the load if we might have raced against mdmon */
4928 if (err
== 3 && devnm
&& mdmon_running(devnm
))
4929 for (retry
= 0; retry
< 3; retry
++) {
4931 err
= load_and_parse_mpb(dfd
, s
, NULL
, keep_fd
);
4937 s
->next
= *super_list
;
4945 if (dfd
>= 0 && !keep_fd
)
4952 get_sra_super_block(int fd
, struct intel_super
**super_list
, char *devname
, int *max
, int keep_fd
)
4959 sra
= sysfs_read(fd
, NULL
, GET_LEVEL
|GET_VERSION
|GET_DEVS
|GET_STATE
);
4963 if (sra
->array
.major_version
!= -1 ||
4964 sra
->array
.minor_version
!= -2 ||
4965 strcmp(sra
->text_version
, "imsm") != 0) {
4970 devnm
= fd2devnm(fd
);
4971 for (sd
= sra
->devs
, i
= 0; sd
; sd
= sd
->next
, i
++) {
4972 if (get_super_block(super_list
, devnm
, devname
,
4973 sd
->disk
.major
, sd
->disk
.minor
, keep_fd
) != 0) {
4984 static int load_container_imsm(struct supertype
*st
, int fd
, char *devname
)
4986 return load_super_imsm_all(st
, fd
, &st
->sb
, devname
, NULL
, 1);
4990 static int load_super_imsm(struct supertype
*st
, int fd
, char *devname
)
4992 struct intel_super
*super
;
4996 if (test_partition(fd
))
4997 /* IMSM not allowed on partitions */
5000 free_super_imsm(st
);
5002 super
= alloc_super();
5003 get_dev_sector_size(fd
, NULL
, &super
->sector_size
);
5006 /* Load hba and capabilities if they exist.
5007 * But do not preclude loading metadata in case capabilities or hba are
5008 * non-compliant and ignore_hw_compat is set.
5010 rv
= find_intel_hba_capability(fd
, super
, devname
);
5011 /* no orom/efi or non-intel hba of the disk */
5012 if (rv
!= 0 && st
->ignore_hw_compat
== 0) {
5014 pr_err("No OROM/EFI properties for %s\n", devname
);
5018 rv
= load_and_parse_mpb(fd
, super
, devname
, 0);
5020 /* retry the load if we might have raced against mdmon */
5022 struct mdstat_ent
*mdstat
= NULL
;
5023 char *name
= fd2kname(fd
);
5026 mdstat
= mdstat_by_component(name
);
5028 if (mdstat
&& mdmon_running(mdstat
->devnm
) && getpid() != mdmon_pid(mdstat
->devnm
)) {
5029 for (retry
= 0; retry
< 3; retry
++) {
5031 rv
= load_and_parse_mpb(fd
, super
, devname
, 0);
5037 free_mdstat(mdstat
);
5042 pr_err("Failed to load all information sections on %s\n", devname
);
5048 if (st
->ss
== NULL
) {
5049 st
->ss
= &super_imsm
;
5050 st
->minor_version
= 0;
5051 st
->max_devs
= IMSM_MAX_DEVICES
;
5054 /* load migration record */
5055 if (load_imsm_migr_rec(super
, NULL
) == 0) {
5056 /* Check for unsupported migration features */
5057 if (check_mpb_migr_compatibility(super
) != 0) {
5058 pr_err("Unsupported migration detected");
5060 fprintf(stderr
, " on %s\n", devname
);
5062 fprintf(stderr
, " (IMSM).\n");
5070 static __u16
info_to_blocks_per_strip(mdu_array_info_t
*info
)
5072 if (info
->level
== 1)
5074 return info
->chunk_size
>> 9;
5077 static unsigned long long info_to_blocks_per_member(mdu_array_info_t
*info
,
5078 unsigned long long size
)
5080 if (info
->level
== 1)
5083 return (size
* 2) & ~(info_to_blocks_per_strip(info
) - 1);
5086 static void imsm_update_version_info(struct intel_super
*super
)
5088 /* update the version and attributes */
5089 struct imsm_super
*mpb
= super
->anchor
;
5091 struct imsm_dev
*dev
;
5092 struct imsm_map
*map
;
5095 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
5096 dev
= get_imsm_dev(super
, i
);
5097 map
= get_imsm_map(dev
, MAP_0
);
5098 if (__le32_to_cpu(dev
->size_high
) > 0)
5099 mpb
->attributes
|= MPB_ATTRIB_2TB
;
5101 /* FIXME detect when an array spans a port multiplier */
5103 mpb
->attributes
|= MPB_ATTRIB_PM
;
5106 if (mpb
->num_raid_devs
> 1 ||
5107 mpb
->attributes
!= MPB_ATTRIB_CHECKSUM_VERIFY
) {
5108 version
= MPB_VERSION_ATTRIBS
;
5109 switch (get_imsm_raid_level(map
)) {
5110 case 0: mpb
->attributes
|= MPB_ATTRIB_RAID0
; break;
5111 case 1: mpb
->attributes
|= MPB_ATTRIB_RAID1
; break;
5112 case 10: mpb
->attributes
|= MPB_ATTRIB_RAID10
; break;
5113 case 5: mpb
->attributes
|= MPB_ATTRIB_RAID5
; break;
5116 if (map
->num_members
>= 5)
5117 version
= MPB_VERSION_5OR6_DISK_ARRAY
;
5118 else if (dev
->status
== DEV_CLONE_N_GO
)
5119 version
= MPB_VERSION_CNG
;
5120 else if (get_imsm_raid_level(map
) == 5)
5121 version
= MPB_VERSION_RAID5
;
5122 else if (map
->num_members
>= 3)
5123 version
= MPB_VERSION_3OR4_DISK_ARRAY
;
5124 else if (get_imsm_raid_level(map
) == 1)
5125 version
= MPB_VERSION_RAID1
;
5127 version
= MPB_VERSION_RAID0
;
5129 strcpy(((char *) mpb
->sig
) + strlen(MPB_SIGNATURE
), version
);
5133 static int check_name(struct intel_super
*super
, char *name
, int quiet
)
5135 struct imsm_super
*mpb
= super
->anchor
;
5136 char *reason
= NULL
;
5139 if (strlen(name
) > MAX_RAID_SERIAL_LEN
)
5140 reason
= "must be 16 characters or less";
5142 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
5143 struct imsm_dev
*dev
= get_imsm_dev(super
, i
);
5145 if (strncmp((char *) dev
->volume
, name
, MAX_RAID_SERIAL_LEN
) == 0) {
5146 reason
= "already exists";
5151 if (reason
&& !quiet
)
5152 pr_err("imsm volume name %s\n", reason
);
5157 static int init_super_imsm_volume(struct supertype
*st
, mdu_array_info_t
*info
,
5158 unsigned long long size
, char *name
,
5159 char *homehost
, int *uuid
,
5160 long long data_offset
)
5162 /* We are creating a volume inside a pre-existing container.
5163 * so st->sb is already set.
5165 struct intel_super
*super
= st
->sb
;
5166 unsigned int sector_size
= super
->sector_size
;
5167 struct imsm_super
*mpb
= super
->anchor
;
5168 struct intel_dev
*dv
;
5169 struct imsm_dev
*dev
;
5170 struct imsm_vol
*vol
;
5171 struct imsm_map
*map
;
5172 int idx
= mpb
->num_raid_devs
;
5174 unsigned long long array_blocks
;
5175 size_t size_old
, size_new
;
5176 unsigned long long num_data_stripes
;
5178 if (super
->orom
&& mpb
->num_raid_devs
>= super
->orom
->vpa
) {
5179 pr_err("This imsm-container already has the maximum of %d volumes\n", super
->orom
->vpa
);
5183 /* ensure the mpb is large enough for the new data */
5184 size_old
= __le32_to_cpu(mpb
->mpb_size
);
5185 size_new
= disks_to_mpb_size(info
->nr_disks
);
5186 if (size_new
> size_old
) {
5188 size_t size_round
= ROUND_UP(size_new
, sector_size
);
5190 if (posix_memalign(&mpb_new
, sector_size
, size_round
) != 0) {
5191 pr_err("could not allocate new mpb\n");
5194 if (posix_memalign(&super
->migr_rec_buf
, sector_size
,
5195 MIGR_REC_BUF_SECTORS
*sector_size
) != 0) {
5196 pr_err("could not allocate migr_rec buffer\n");
5202 memcpy(mpb_new
, mpb
, size_old
);
5205 super
->anchor
= mpb_new
;
5206 mpb
->mpb_size
= __cpu_to_le32(size_new
);
5207 memset(mpb_new
+ size_old
, 0, size_round
- size_old
);
5208 super
->len
= size_round
;
5210 super
->current_vol
= idx
;
5212 /* handle 'failed_disks' by either:
5213 * a) create dummy disk entries in the table if this the first
5214 * volume in the array. We add them here as this is the only
5215 * opportunity to add them. add_to_super_imsm_volume()
5216 * handles the non-failed disks and continues incrementing
5218 * b) validate that 'failed_disks' matches the current number
5219 * of missing disks if the container is populated
5221 if (super
->current_vol
== 0) {
5223 for (i
= 0; i
< info
->failed_disks
; i
++) {
5224 struct imsm_disk
*disk
;
5227 disk
= __get_imsm_disk(mpb
, i
);
5228 disk
->status
= CONFIGURED_DISK
| FAILED_DISK
;
5229 disk
->scsi_id
= __cpu_to_le32(~(__u32
)0);
5230 snprintf((char *) disk
->serial
, MAX_RAID_SERIAL_LEN
,
5233 find_missing(super
);
5238 for (d
= super
->missing
; d
; d
= d
->next
)
5240 if (info
->failed_disks
> missing
) {
5241 pr_err("unable to add 'missing' disk to container\n");
5246 if (!check_name(super
, name
, 0))
5248 dv
= xmalloc(sizeof(*dv
));
5249 dev
= xcalloc(1, sizeof(*dev
) + sizeof(__u32
) * (info
->raid_disks
- 1));
5250 strncpy((char *) dev
->volume
, name
, MAX_RAID_SERIAL_LEN
);
5251 array_blocks
= calc_array_size(info
->level
, info
->raid_disks
,
5252 info
->layout
, info
->chunk_size
,
5254 /* round array size down to closest MB */
5255 array_blocks
= (array_blocks
>> SECT_PER_MB_SHIFT
) << SECT_PER_MB_SHIFT
;
5257 dev
->size_low
= __cpu_to_le32((__u32
) array_blocks
);
5258 dev
->size_high
= __cpu_to_le32((__u32
) (array_blocks
>> 32));
5259 dev
->status
= (DEV_READ_COALESCING
| DEV_WRITE_COALESCING
);
5261 vol
->migr_state
= 0;
5262 set_migr_type(dev
, MIGR_INIT
);
5263 vol
->dirty
= !info
->state
;
5264 vol
->curr_migr_unit
= 0;
5265 map
= get_imsm_map(dev
, MAP_0
);
5266 set_pba_of_lba0(map
, super
->create_offset
);
5267 set_blocks_per_member(map
, info_to_blocks_per_member(info
, size
));
5268 map
->blocks_per_strip
= __cpu_to_le16(info_to_blocks_per_strip(info
));
5269 map
->failed_disk_num
= ~0;
5270 if (info
->level
> 0)
5271 map
->map_state
= (info
->state
? IMSM_T_STATE_NORMAL
5272 : IMSM_T_STATE_UNINITIALIZED
);
5274 map
->map_state
= info
->failed_disks
? IMSM_T_STATE_FAILED
:
5275 IMSM_T_STATE_NORMAL
;
5278 if (info
->level
== 1 && info
->raid_disks
> 2) {
5281 pr_err("imsm does not support more than 2 disksin a raid1 volume\n");
5285 map
->raid_level
= info
->level
;
5286 if (info
->level
== 10) {
5287 map
->raid_level
= 1;
5288 map
->num_domains
= info
->raid_disks
/ 2;
5289 } else if (info
->level
== 1)
5290 map
->num_domains
= info
->raid_disks
;
5292 map
->num_domains
= 1;
5294 /* info->size is only int so use the 'size' parameter instead */
5295 num_data_stripes
= (size
* 2) / info_to_blocks_per_strip(info
);
5296 num_data_stripes
/= map
->num_domains
;
5297 set_num_data_stripes(map
, num_data_stripes
);
5299 map
->num_members
= info
->raid_disks
;
5300 for (i
= 0; i
< map
->num_members
; i
++) {
5301 /* initialized in add_to_super */
5302 set_imsm_ord_tbl_ent(map
, i
, IMSM_ORD_REBUILD
);
5304 mpb
->num_raid_devs
++;
5307 dv
->index
= super
->current_vol
;
5308 dv
->next
= super
->devlist
;
5309 super
->devlist
= dv
;
5311 imsm_update_version_info(super
);
5316 static int init_super_imsm(struct supertype
*st
, mdu_array_info_t
*info
,
5317 unsigned long long size
, char *name
,
5318 char *homehost
, int *uuid
,
5319 unsigned long long data_offset
)
5321 /* This is primarily called by Create when creating a new array.
5322 * We will then get add_to_super called for each component, and then
5323 * write_init_super called to write it out to each device.
5324 * For IMSM, Create can create on fresh devices or on a pre-existing
5326 * To create on a pre-existing array a different method will be called.
5327 * This one is just for fresh drives.
5329 struct intel_super
*super
;
5330 struct imsm_super
*mpb
;
5334 if (data_offset
!= INVALID_SECTORS
) {
5335 pr_err("data-offset not supported by imsm\n");
5340 return init_super_imsm_volume(st
, info
, size
, name
, homehost
, uuid
,
5344 mpb_size
= disks_to_mpb_size(info
->nr_disks
);
5346 mpb_size
= MAX_SECTOR_SIZE
;
5348 super
= alloc_super();
5350 posix_memalign(&super
->buf
, MAX_SECTOR_SIZE
, mpb_size
) != 0) {
5355 pr_err("could not allocate superblock\n");
5358 if (posix_memalign(&super
->migr_rec_buf
, MAX_SECTOR_SIZE
,
5359 MIGR_REC_BUF_SECTORS
*MAX_SECTOR_SIZE
) != 0) {
5360 pr_err("could not allocate migr_rec buffer\n");
5365 memset(super
->buf
, 0, mpb_size
);
5367 mpb
->mpb_size
= __cpu_to_le32(mpb_size
);
5371 /* zeroing superblock */
5375 mpb
->attributes
= MPB_ATTRIB_CHECKSUM_VERIFY
;
5377 version
= (char *) mpb
->sig
;
5378 strcpy(version
, MPB_SIGNATURE
);
5379 version
+= strlen(MPB_SIGNATURE
);
5380 strcpy(version
, MPB_VERSION_RAID0
);
5386 static int add_to_super_imsm_volume(struct supertype
*st
, mdu_disk_info_t
*dk
,
5387 int fd
, char *devname
)
5389 struct intel_super
*super
= st
->sb
;
5390 struct imsm_super
*mpb
= super
->anchor
;
5391 struct imsm_disk
*_disk
;
5392 struct imsm_dev
*dev
;
5393 struct imsm_map
*map
;
5397 dev
= get_imsm_dev(super
, super
->current_vol
);
5398 map
= get_imsm_map(dev
, MAP_0
);
5400 if (! (dk
->state
& (1<<MD_DISK_SYNC
))) {
5401 pr_err("%s: Cannot add spare devices to IMSM volume\n",
5407 /* we're doing autolayout so grab the pre-marked (in
5408 * validate_geometry) raid_disk
5410 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
5411 if (dl
->raiddisk
== dk
->raid_disk
)
5414 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
5415 if (dl
->major
== dk
->major
&&
5416 dl
->minor
== dk
->minor
)
5421 pr_err("%s is not a member of the same container\n", devname
);
5425 /* add a pristine spare to the metadata */
5426 if (dl
->index
< 0) {
5427 dl
->index
= super
->anchor
->num_disks
;
5428 super
->anchor
->num_disks
++;
5430 /* Check the device has not already been added */
5431 slot
= get_imsm_disk_slot(map
, dl
->index
);
5433 (get_imsm_ord_tbl_ent(dev
, slot
, MAP_X
) & IMSM_ORD_REBUILD
) == 0) {
5434 pr_err("%s has been included in this array twice\n",
5438 set_imsm_ord_tbl_ent(map
, dk
->raid_disk
, dl
->index
);
5439 dl
->disk
.status
= CONFIGURED_DISK
;
5441 /* update size of 'missing' disks to be at least as large as the
5442 * largest acitve member (we only have dummy missing disks when
5443 * creating the first volume)
5445 if (super
->current_vol
== 0) {
5446 for (df
= super
->missing
; df
; df
= df
->next
) {
5447 if (total_blocks(&dl
->disk
) > total_blocks(&df
->disk
))
5448 set_total_blocks(&df
->disk
, total_blocks(&dl
->disk
));
5449 _disk
= __get_imsm_disk(mpb
, df
->index
);
5454 /* refresh unset/failed slots to point to valid 'missing' entries */
5455 for (df
= super
->missing
; df
; df
= df
->next
)
5456 for (slot
= 0; slot
< mpb
->num_disks
; slot
++) {
5457 __u32 ord
= get_imsm_ord_tbl_ent(dev
, slot
, MAP_X
);
5459 if ((ord
& IMSM_ORD_REBUILD
) == 0)
5461 set_imsm_ord_tbl_ent(map
, slot
, df
->index
| IMSM_ORD_REBUILD
);
5462 if (is_gen_migration(dev
)) {
5463 struct imsm_map
*map2
= get_imsm_map(dev
,
5465 int slot2
= get_imsm_disk_slot(map2
, df
->index
);
5466 if (slot2
< map2
->num_members
&& slot2
>= 0) {
5467 __u32 ord2
= get_imsm_ord_tbl_ent(dev
,
5470 if ((unsigned)df
->index
==
5472 set_imsm_ord_tbl_ent(map2
,
5478 dprintf("set slot:%d to missing disk:%d\n", slot
, df
->index
);
5482 /* if we are creating the first raid device update the family number */
5483 if (super
->current_vol
== 0) {
5485 struct imsm_dev
*_dev
= __get_imsm_dev(mpb
, 0);
5487 _disk
= __get_imsm_disk(mpb
, dl
->index
);
5488 if (!_dev
|| !_disk
) {
5489 pr_err("BUG mpb setup error\n");
5495 sum
+= __gen_imsm_checksum(mpb
);
5496 mpb
->family_num
= __cpu_to_le32(sum
);
5497 mpb
->orig_family_num
= mpb
->family_num
;
5499 super
->current_disk
= dl
;
5504 * Function marks disk as spare and restores disk serial
5505 * in case it was previously marked as failed by takeover operation
5507 * -1 : critical error
5508 * 0 : disk is marked as spare but serial is not set
5511 int mark_spare(struct dl
*disk
)
5513 __u8 serial
[MAX_RAID_SERIAL_LEN
];
5520 if (!imsm_read_serial(disk
->fd
, NULL
, serial
)) {
5521 /* Restore disk serial number, because takeover marks disk
5522 * as failed and adds to serial ':0' before it becomes
5525 serialcpy(disk
->serial
, serial
);
5526 serialcpy(disk
->disk
.serial
, serial
);
5529 disk
->disk
.status
= SPARE_DISK
;
5535 static int add_to_super_imsm(struct supertype
*st
, mdu_disk_info_t
*dk
,
5536 int fd
, char *devname
,
5537 unsigned long long data_offset
)
5539 struct intel_super
*super
= st
->sb
;
5541 unsigned long long size
;
5542 unsigned int member_sector_size
;
5547 /* If we are on an RAID enabled platform check that the disk is
5548 * attached to the raid controller.
5549 * We do not need to test disks attachment for container based additions,
5550 * they shall be already tested when container was created/assembled.
5552 rv
= find_intel_hba_capability(fd
, super
, devname
);
5553 /* no orom/efi or non-intel hba of the disk */
5555 dprintf("capability: %p fd: %d ret: %d\n",
5556 super
->orom
, fd
, rv
);
5560 if (super
->current_vol
>= 0)
5561 return add_to_super_imsm_volume(st
, dk
, fd
, devname
);
5564 dd
= xcalloc(sizeof(*dd
), 1);
5565 dd
->major
= major(stb
.st_rdev
);
5566 dd
->minor
= minor(stb
.st_rdev
);
5567 dd
->devname
= devname
? xstrdup(devname
) : NULL
;
5570 dd
->action
= DISK_ADD
;
5571 rv
= imsm_read_serial(fd
, devname
, dd
->serial
);
5573 pr_err("failed to retrieve scsi serial, aborting\n");
5579 if (super
->hba
&& ((super
->hba
->type
== SYS_DEV_NVME
) ||
5580 (super
->hba
->type
== SYS_DEV_VMD
))) {
5582 char *devpath
= diskfd_to_devpath(fd
);
5583 char controller_path
[PATH_MAX
];
5586 pr_err("failed to get devpath, aborting\n");
5593 snprintf(controller_path
, PATH_MAX
-1, "%s/device", devpath
);
5596 if (devpath_to_vendor(controller_path
) == 0x8086) {
5598 * If Intel's NVMe drive has serial ended with
5599 * "-A","-B","-1" or "-2" it means that this is "x8"
5600 * device (double drive on single PCIe card).
5601 * User should be warned about potential data loss.
5603 for (i
= MAX_RAID_SERIAL_LEN
-1; i
> 0; i
--) {
5604 /* Skip empty character at the end */
5605 if (dd
->serial
[i
] == 0)
5608 if (((dd
->serial
[i
] == 'A') ||
5609 (dd
->serial
[i
] == 'B') ||
5610 (dd
->serial
[i
] == '1') ||
5611 (dd
->serial
[i
] == '2')) &&
5612 (dd
->serial
[i
-1] == '-'))
5613 pr_err("\tThe action you are about to take may put your data at risk.\n"
5614 "\tPlease note that x8 devices may consist of two separate x4 devices "
5615 "located on a single PCIe port.\n"
5616 "\tRAID 0 is the only supported configuration for this type of x8 device.\n");
5619 } else if (super
->hba
->type
== SYS_DEV_VMD
&& super
->orom
&&
5620 !imsm_orom_has_tpv_support(super
->orom
)) {
5621 pr_err("\tPlatform configuration does not support non-Intel NVMe drives.\n"
5622 "\tPlease refer to Intel(R) RSTe user guide.\n");
5629 get_dev_size(fd
, NULL
, &size
);
5630 get_dev_sector_size(fd
, NULL
, &member_sector_size
);
5632 if (super
->sector_size
== 0) {
5633 /* this a first device, so sector_size is not set yet */
5634 super
->sector_size
= member_sector_size
;
5635 } else if (member_sector_size
!= super
->sector_size
) {
5636 pr_err("Mixing between different sector size is forbidden, aborting...\n");
5643 /* clear migr_rec when adding disk to container */
5644 memset(super
->migr_rec_buf
, 0, MIGR_REC_BUF_SECTORS
*super
->sector_size
);
5645 if (lseek64(fd
, size
- MIGR_REC_SECTOR_POSITION
*super
->sector_size
,
5647 if ((unsigned int)write(fd
, super
->migr_rec_buf
,
5648 MIGR_REC_BUF_SECTORS
*super
->sector_size
) !=
5649 MIGR_REC_BUF_SECTORS
*super
->sector_size
)
5650 perror("Write migr_rec failed");
5654 serialcpy(dd
->disk
.serial
, dd
->serial
);
5655 set_total_blocks(&dd
->disk
, size
);
5656 if (__le32_to_cpu(dd
->disk
.total_blocks_hi
) > 0) {
5657 struct imsm_super
*mpb
= super
->anchor
;
5658 mpb
->attributes
|= MPB_ATTRIB_2TB_DISK
;
5661 if (sysfs_disk_to_scsi_id(fd
, &id
) == 0)
5662 dd
->disk
.scsi_id
= __cpu_to_le32(id
);
5664 dd
->disk
.scsi_id
= __cpu_to_le32(0);
5666 if (st
->update_tail
) {
5667 dd
->next
= super
->disk_mgmt_list
;
5668 super
->disk_mgmt_list
= dd
;
5670 dd
->next
= super
->disks
;
5672 super
->updates_pending
++;
5678 static int remove_from_super_imsm(struct supertype
*st
, mdu_disk_info_t
*dk
)
5680 struct intel_super
*super
= st
->sb
;
5683 /* remove from super works only in mdmon - for communication
5684 * manager - monitor. Check if communication memory buffer
5687 if (!st
->update_tail
) {
5688 pr_err("shall be used in mdmon context only\n");
5691 dd
= xcalloc(1, sizeof(*dd
));
5692 dd
->major
= dk
->major
;
5693 dd
->minor
= dk
->minor
;
5696 dd
->action
= DISK_REMOVE
;
5698 dd
->next
= super
->disk_mgmt_list
;
5699 super
->disk_mgmt_list
= dd
;
5704 static int store_imsm_mpb(int fd
, struct imsm_super
*mpb
);
5707 char buf
[MAX_SECTOR_SIZE
];
5708 struct imsm_super anchor
;
5709 } spare_record
__attribute__ ((aligned(MAX_SECTOR_SIZE
)));
5711 /* spare records have their own family number and do not have any defined raid
5714 static int write_super_imsm_spares(struct intel_super
*super
, int doclose
)
5716 struct imsm_super
*mpb
= super
->anchor
;
5717 struct imsm_super
*spare
= &spare_record
.anchor
;
5721 spare
->mpb_size
= __cpu_to_le32(sizeof(struct imsm_super
));
5722 spare
->generation_num
= __cpu_to_le32(1UL);
5723 spare
->attributes
= MPB_ATTRIB_CHECKSUM_VERIFY
;
5724 spare
->num_disks
= 1;
5725 spare
->num_raid_devs
= 0;
5726 spare
->cache_size
= mpb
->cache_size
;
5727 spare
->pwr_cycle_count
= __cpu_to_le32(1);
5729 snprintf((char *) spare
->sig
, MAX_SIGNATURE_LENGTH
,
5730 MPB_SIGNATURE MPB_VERSION_RAID0
);
5732 for (d
= super
->disks
; d
; d
= d
->next
) {
5736 spare
->disk
[0] = d
->disk
;
5737 if (__le32_to_cpu(d
->disk
.total_blocks_hi
) > 0)
5738 spare
->attributes
|= MPB_ATTRIB_2TB_DISK
;
5740 if (super
->sector_size
== 4096)
5741 convert_to_4k_imsm_disk(&spare
->disk
[0]);
5743 sum
= __gen_imsm_checksum(spare
);
5744 spare
->family_num
= __cpu_to_le32(sum
);
5745 spare
->orig_family_num
= 0;
5746 sum
= __gen_imsm_checksum(spare
);
5747 spare
->check_sum
= __cpu_to_le32(sum
);
5749 if (store_imsm_mpb(d
->fd
, spare
)) {
5750 pr_err("failed for device %d:%d %s\n",
5751 d
->major
, d
->minor
, strerror(errno
));
5763 static int write_super_imsm(struct supertype
*st
, int doclose
)
5765 struct intel_super
*super
= st
->sb
;
5766 unsigned int sector_size
= super
->sector_size
;
5767 struct imsm_super
*mpb
= super
->anchor
;
5773 __u32 mpb_size
= sizeof(struct imsm_super
) - sizeof(struct imsm_disk
);
5775 int clear_migration_record
= 1;
5778 /* 'generation' is incremented everytime the metadata is written */
5779 generation
= __le32_to_cpu(mpb
->generation_num
);
5781 mpb
->generation_num
= __cpu_to_le32(generation
);
5783 /* fix up cases where previous mdadm releases failed to set
5786 if (mpb
->orig_family_num
== 0)
5787 mpb
->orig_family_num
= mpb
->family_num
;
5789 for (d
= super
->disks
; d
; d
= d
->next
) {
5793 mpb
->disk
[d
->index
] = d
->disk
;
5797 for (d
= super
->missing
; d
; d
= d
->next
) {
5798 mpb
->disk
[d
->index
] = d
->disk
;
5801 mpb
->num_disks
= num_disks
;
5802 mpb_size
+= sizeof(struct imsm_disk
) * mpb
->num_disks
;
5804 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
5805 struct imsm_dev
*dev
= __get_imsm_dev(mpb
, i
);
5806 struct imsm_dev
*dev2
= get_imsm_dev(super
, i
);
5808 imsm_copy_dev(dev
, dev2
);
5809 mpb_size
+= sizeof_imsm_dev(dev
, 0);
5811 if (is_gen_migration(dev2
))
5812 clear_migration_record
= 0;
5815 bbm_log_size
= get_imsm_bbm_log_size(super
->bbm_log
);
5818 memcpy((void *)mpb
+ mpb_size
, super
->bbm_log
, bbm_log_size
);
5819 mpb
->attributes
|= MPB_ATTRIB_BBM
;
5821 mpb
->attributes
&= ~MPB_ATTRIB_BBM
;
5823 super
->anchor
->bbm_log_size
= __cpu_to_le32(bbm_log_size
);
5824 mpb_size
+= bbm_log_size
;
5825 mpb
->mpb_size
= __cpu_to_le32(mpb_size
);
5828 assert(super
->len
== 0 || mpb_size
<= super
->len
);
5831 /* recalculate checksum */
5832 sum
= __gen_imsm_checksum(mpb
);
5833 mpb
->check_sum
= __cpu_to_le32(sum
);
5835 if (super
->clean_migration_record_by_mdmon
) {
5836 clear_migration_record
= 1;
5837 super
->clean_migration_record_by_mdmon
= 0;
5839 if (clear_migration_record
)
5840 memset(super
->migr_rec_buf
, 0,
5841 MIGR_REC_BUF_SECTORS
*sector_size
);
5843 if (sector_size
== 4096)
5844 convert_to_4k(super
);
5846 /* write the mpb for disks that compose raid devices */
5847 for (d
= super
->disks
; d
; d
= d
->next
) {
5848 if (d
->index
< 0 || is_failed(&d
->disk
))
5851 if (clear_migration_record
) {
5852 unsigned long long dsize
;
5854 get_dev_size(d
->fd
, NULL
, &dsize
);
5855 if (lseek64(d
->fd
, dsize
- sector_size
,
5857 if ((unsigned int)write(d
->fd
,
5858 super
->migr_rec_buf
,
5859 MIGR_REC_BUF_SECTORS
*sector_size
) !=
5860 MIGR_REC_BUF_SECTORS
*sector_size
)
5861 perror("Write migr_rec failed");
5865 if (store_imsm_mpb(d
->fd
, mpb
))
5867 "failed for device %d:%d (fd: %d)%s\n",
5869 d
->fd
, strerror(errno
));
5878 return write_super_imsm_spares(super
, doclose
);
5883 static int create_array(struct supertype
*st
, int dev_idx
)
5886 struct imsm_update_create_array
*u
;
5887 struct intel_super
*super
= st
->sb
;
5888 struct imsm_dev
*dev
= get_imsm_dev(super
, dev_idx
);
5889 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
5890 struct disk_info
*inf
;
5891 struct imsm_disk
*disk
;
5894 len
= sizeof(*u
) - sizeof(*dev
) + sizeof_imsm_dev(dev
, 0) +
5895 sizeof(*inf
) * map
->num_members
;
5897 u
->type
= update_create_array
;
5898 u
->dev_idx
= dev_idx
;
5899 imsm_copy_dev(&u
->dev
, dev
);
5900 inf
= get_disk_info(u
);
5901 for (i
= 0; i
< map
->num_members
; i
++) {
5902 int idx
= get_imsm_disk_idx(dev
, i
, MAP_X
);
5904 disk
= get_imsm_disk(super
, idx
);
5906 disk
= get_imsm_missing(super
, idx
);
5907 serialcpy(inf
[i
].serial
, disk
->serial
);
5909 append_metadata_update(st
, u
, len
);
5914 static int mgmt_disk(struct supertype
*st
)
5916 struct intel_super
*super
= st
->sb
;
5918 struct imsm_update_add_remove_disk
*u
;
5920 if (!super
->disk_mgmt_list
)
5925 u
->type
= update_add_remove_disk
;
5926 append_metadata_update(st
, u
, len
);
5931 static int write_init_super_imsm(struct supertype
*st
)
5933 struct intel_super
*super
= st
->sb
;
5934 int current_vol
= super
->current_vol
;
5936 /* we are done with current_vol reset it to point st at the container */
5937 super
->current_vol
= -1;
5939 if (st
->update_tail
) {
5940 /* queue the recently created array / added disk
5941 * as a metadata update */
5944 /* determine if we are creating a volume or adding a disk */
5945 if (current_vol
< 0) {
5946 /* in the mgmt (add/remove) disk case we are running
5947 * in mdmon context, so don't close fd's
5949 return mgmt_disk(st
);
5951 rv
= create_array(st
, current_vol
);
5956 for (d
= super
->disks
; d
; d
= d
->next
)
5957 Kill(d
->devname
, NULL
, 0, -1, 1);
5958 return write_super_imsm(st
, 1);
5963 static int store_super_imsm(struct supertype
*st
, int fd
)
5965 struct intel_super
*super
= st
->sb
;
5966 struct imsm_super
*mpb
= super
? super
->anchor
: NULL
;
5972 if (super
->sector_size
== 4096)
5973 convert_to_4k(super
);
5974 return store_imsm_mpb(fd
, mpb
);
5981 static int validate_geometry_imsm_container(struct supertype
*st
, int level
,
5982 int layout
, int raiddisks
, int chunk
,
5983 unsigned long long size
,
5984 unsigned long long data_offset
,
5986 unsigned long long *freesize
,
5990 unsigned long long ldsize
;
5991 struct intel_super
*super
;
5994 if (level
!= LEVEL_CONTAINER
)
5999 fd
= open(dev
, O_RDONLY
|O_EXCL
, 0);
6002 pr_err("imsm: Cannot open %s: %s\n",
6003 dev
, strerror(errno
));
6006 if (!get_dev_size(fd
, dev
, &ldsize
)) {
6011 /* capabilities retrieve could be possible
6012 * note that there is no fd for the disks in array.
6014 super
= alloc_super();
6019 if (!get_dev_sector_size(fd
, NULL
, &super
->sector_size
)) {
6025 rv
= find_intel_hba_capability(fd
, super
, verbose
> 0 ? dev
: NULL
);
6029 fd2devname(fd
, str
);
6030 dprintf("fd: %d %s orom: %p rv: %d raiddisk: %d\n",
6031 fd
, str
, super
->orom
, rv
, raiddisks
);
6033 /* no orom/efi or non-intel hba of the disk */
6040 if (raiddisks
> super
->orom
->tds
) {
6042 pr_err("%d exceeds maximum number of platform supported disks: %d\n",
6043 raiddisks
, super
->orom
->tds
);
6047 if ((super
->orom
->attr
& IMSM_OROM_ATTR_2TB_DISK
) == 0 &&
6048 (ldsize
>> 9) >> 32 > 0) {
6050 pr_err("%s exceeds maximum platform supported size\n", dev
);
6056 *freesize
= avail_size_imsm(st
, ldsize
>> 9, data_offset
);
6062 static unsigned long long find_size(struct extent
*e
, int *idx
, int num_extents
)
6064 const unsigned long long base_start
= e
[*idx
].start
;
6065 unsigned long long end
= base_start
+ e
[*idx
].size
;
6068 if (base_start
== end
)
6072 for (i
= *idx
; i
< num_extents
; i
++) {
6073 /* extend overlapping extents */
6074 if (e
[i
].start
>= base_start
&&
6075 e
[i
].start
<= end
) {
6078 if (e
[i
].start
+ e
[i
].size
> end
)
6079 end
= e
[i
].start
+ e
[i
].size
;
6080 } else if (e
[i
].start
> end
) {
6086 return end
- base_start
;
6089 static unsigned long long merge_extents(struct intel_super
*super
, int sum_extents
)
6091 /* build a composite disk with all known extents and generate a new
6092 * 'maxsize' given the "all disks in an array must share a common start
6093 * offset" constraint
6095 struct extent
*e
= xcalloc(sum_extents
, sizeof(*e
));
6099 unsigned long long pos
;
6100 unsigned long long start
= 0;
6101 unsigned long long maxsize
;
6102 unsigned long reserve
;
6104 /* coalesce and sort all extents. also, check to see if we need to
6105 * reserve space between member arrays
6108 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
6111 for (i
= 0; i
< dl
->extent_cnt
; i
++)
6114 qsort(e
, sum_extents
, sizeof(*e
), cmp_extent
);
6119 while (i
< sum_extents
) {
6120 e
[j
].start
= e
[i
].start
;
6121 e
[j
].size
= find_size(e
, &i
, sum_extents
);
6123 if (e
[j
-1].size
== 0)
6132 unsigned long long esize
;
6134 esize
= e
[i
].start
- pos
;
6135 if (esize
>= maxsize
) {
6140 pos
= e
[i
].start
+ e
[i
].size
;
6142 } while (e
[i
-1].size
);
6148 /* FIXME assumes volume at offset 0 is the first volume in a
6151 if (start_extent
> 0)
6152 reserve
= IMSM_RESERVED_SECTORS
; /* gap between raid regions */
6156 if (maxsize
< reserve
)
6159 super
->create_offset
= ~((unsigned long long) 0);
6160 if (start
+ reserve
> super
->create_offset
)
6161 return 0; /* start overflows create_offset */
6162 super
->create_offset
= start
+ reserve
;
6164 return maxsize
- reserve
;
6167 static int is_raid_level_supported(const struct imsm_orom
*orom
, int level
, int raiddisks
)
6169 if (level
< 0 || level
== 6 || level
== 4)
6172 /* if we have an orom prevent invalid raid levels */
6175 case 0: return imsm_orom_has_raid0(orom
);
6178 return imsm_orom_has_raid1e(orom
);
6179 return imsm_orom_has_raid1(orom
) && raiddisks
== 2;
6180 case 10: return imsm_orom_has_raid10(orom
) && raiddisks
== 4;
6181 case 5: return imsm_orom_has_raid5(orom
) && raiddisks
> 2;
6184 return 1; /* not on an Intel RAID platform so anything goes */
6190 active_arrays_by_format(char *name
, char* hba
, struct md_list
**devlist
,
6191 int dpa
, int verbose
)
6193 struct mdstat_ent
*mdstat
= mdstat_read(0, 0);
6194 struct mdstat_ent
*memb
;
6200 for (memb
= mdstat
; memb
; memb
= memb
->next
) {
6201 if (memb
->metadata_version
&&
6202 (strncmp(memb
->metadata_version
, "external:", 9) == 0) &&
6203 (strcmp(&memb
->metadata_version
[9], name
) == 0) &&
6204 !is_subarray(memb
->metadata_version
+9) &&
6206 struct dev_member
*dev
= memb
->members
;
6208 while(dev
&& (fd
< 0)) {
6209 char *path
= xmalloc(strlen(dev
->name
) + strlen("/dev/") + 1);
6210 num
= sprintf(path
, "%s%s", "/dev/", dev
->name
);
6212 fd
= open(path
, O_RDONLY
, 0);
6213 if (num
<= 0 || fd
< 0) {
6214 pr_vrb("Cannot open %s: %s\n",
6215 dev
->name
, strerror(errno
));
6221 if (fd
>= 0 && disk_attached_to_hba(fd
, hba
)) {
6222 struct mdstat_ent
*vol
;
6223 for (vol
= mdstat
; vol
; vol
= vol
->next
) {
6224 if (vol
->active
> 0 &&
6225 vol
->metadata_version
&&
6226 is_container_member(vol
, memb
->devnm
)) {
6231 if (*devlist
&& (found
< dpa
)) {
6232 dv
= xcalloc(1, sizeof(*dv
));
6233 dv
->devname
= xmalloc(strlen(memb
->devnm
) + strlen("/dev/") + 1);
6234 sprintf(dv
->devname
, "%s%s", "/dev/", memb
->devnm
);
6237 dv
->next
= *devlist
;
6245 free_mdstat(mdstat
);
6250 static struct md_list
*
6251 get_loop_devices(void)
6254 struct md_list
*devlist
= NULL
;
6257 for(i
= 0; i
< 12; i
++) {
6258 dv
= xcalloc(1, sizeof(*dv
));
6259 dv
->devname
= xmalloc(40);
6260 sprintf(dv
->devname
, "/dev/loop%d", i
);
6268 static struct md_list
*
6269 get_devices(const char *hba_path
)
6271 struct md_list
*devlist
= NULL
;
6278 devlist
= get_loop_devices();
6281 /* scroll through /sys/dev/block looking for devices attached to
6284 dir
= opendir("/sys/dev/block");
6285 for (ent
= dir
? readdir(dir
) : NULL
; ent
; ent
= readdir(dir
)) {
6290 if (sscanf(ent
->d_name
, "%d:%d", &major
, &minor
) != 2)
6292 path
= devt_to_devpath(makedev(major
, minor
));
6295 if (!path_attached_to_hba(path
, hba_path
)) {
6302 fd
= dev_open(ent
->d_name
, O_RDONLY
);
6304 fd2devname(fd
, buf
);
6307 pr_err("cannot open device: %s\n",
6312 dv
= xcalloc(1, sizeof(*dv
));
6313 dv
->devname
= xstrdup(buf
);
6320 devlist
= devlist
->next
;
6330 count_volumes_list(struct md_list
*devlist
, char *homehost
,
6331 int verbose
, int *found
)
6333 struct md_list
*tmpdev
;
6335 struct supertype
*st
;
6337 /* first walk the list of devices to find a consistent set
6338 * that match the criterea, if that is possible.
6339 * We flag the ones we like with 'used'.
6342 st
= match_metadata_desc_imsm("imsm");
6344 pr_vrb("cannot allocate memory for imsm supertype\n");
6348 for (tmpdev
= devlist
; tmpdev
; tmpdev
= tmpdev
->next
) {
6349 char *devname
= tmpdev
->devname
;
6351 struct supertype
*tst
;
6353 if (tmpdev
->used
> 1)
6355 tst
= dup_super(st
);
6357 pr_vrb("cannot allocate memory for imsm supertype\n");
6360 tmpdev
->container
= 0;
6361 dfd
= dev_open(devname
, O_RDONLY
|O_EXCL
);
6363 dprintf("cannot open device %s: %s\n",
6364 devname
, strerror(errno
));
6366 } else if (fstat(dfd
, &stb
)< 0) {
6368 dprintf("fstat failed for %s: %s\n",
6369 devname
, strerror(errno
));
6371 } else if ((stb
.st_mode
& S_IFMT
) != S_IFBLK
) {
6372 dprintf("%s is not a block device.\n",
6375 } else if (must_be_container(dfd
)) {
6376 struct supertype
*cst
;
6377 cst
= super_by_fd(dfd
, NULL
);
6379 dprintf("cannot recognize container type %s\n",
6382 } else if (tst
->ss
!= st
->ss
) {
6383 dprintf("non-imsm container - ignore it: %s\n",
6386 } else if (!tst
->ss
->load_container
||
6387 tst
->ss
->load_container(tst
, dfd
, NULL
))
6390 tmpdev
->container
= 1;
6393 cst
->ss
->free_super(cst
);
6395 tmpdev
->st_rdev
= stb
.st_rdev
;
6396 if (tst
->ss
->load_super(tst
,dfd
, NULL
)) {
6397 dprintf("no RAID superblock on %s\n",
6400 } else if (tst
->ss
->compare_super
== NULL
) {
6401 dprintf("Cannot assemble %s metadata on %s\n",
6402 tst
->ss
->name
, devname
);
6408 if (tmpdev
->used
== 2 || tmpdev
->used
== 4) {
6409 /* Ignore unrecognised devices during auto-assembly */
6414 tst
->ss
->getinfo_super(tst
, &info
, NULL
);
6416 if (st
->minor_version
== -1)
6417 st
->minor_version
= tst
->minor_version
;
6419 if (memcmp(info
.uuid
, uuid_zero
,
6420 sizeof(int[4])) == 0) {
6421 /* this is a floating spare. It cannot define
6422 * an array unless there are no more arrays of
6423 * this type to be found. It can be included
6424 * in an array of this type though.
6430 if (st
->ss
!= tst
->ss
||
6431 st
->minor_version
!= tst
->minor_version
||
6432 st
->ss
->compare_super(st
, tst
) != 0) {
6433 /* Some mismatch. If exactly one array matches this host,
6434 * we can resolve on that one.
6435 * Or, if we are auto assembling, we just ignore the second
6438 dprintf("superblock on %s doesn't match others - assembly aborted\n",
6444 dprintf("found: devname: %s\n", devname
);
6448 tst
->ss
->free_super(tst
);
6452 if ((err
= load_super_imsm_all(st
, -1, &st
->sb
, NULL
, devlist
, 0)) == 0) {
6453 struct mdinfo
*iter
, *head
= st
->ss
->container_content(st
, NULL
);
6454 for (iter
= head
; iter
; iter
= iter
->next
) {
6455 dprintf("content->text_version: %s vol\n",
6456 iter
->text_version
);
6457 if (iter
->array
.state
& (1<<MD_SB_BLOCK_VOLUME
)) {
6458 /* do not assemble arrays with unsupported
6460 dprintf("Cannot activate member %s.\n",
6461 iter
->text_version
);
6468 dprintf("No valid super block on device list: err: %d %p\n",
6472 dprintf("no more devices to examine\n");
6475 for (tmpdev
= devlist
; tmpdev
; tmpdev
= tmpdev
->next
) {
6476 if (tmpdev
->used
== 1 && tmpdev
->found
) {
6478 if (count
< tmpdev
->found
)
6481 count
-= tmpdev
->found
;
6484 if (tmpdev
->used
== 1)
6489 st
->ss
->free_super(st
);
6493 static int __count_volumes(char *hba_path
, int dpa
, int verbose
,
6496 struct sys_dev
*idev
, *intel_devices
= find_intel_devices();
6498 const struct orom_entry
*entry
;
6499 struct devid_list
*dv
, *devid_list
;
6504 for (idev
= intel_devices
; idev
; idev
= idev
->next
) {
6505 if (strstr(idev
->path
, hba_path
))
6509 if (!idev
|| !idev
->dev_id
)
6512 entry
= get_orom_entry_by_device_id(idev
->dev_id
);
6514 if (!entry
|| !entry
->devid_list
)
6517 devid_list
= entry
->devid_list
;
6518 for (dv
= devid_list
; dv
; dv
= dv
->next
) {
6519 struct md_list
*devlist
;
6520 struct sys_dev
*device
= NULL
;
6525 device
= device_by_id_and_path(dv
->devid
, hba_path
);
6527 device
= device_by_id(dv
->devid
);
6530 hpath
= device
->path
;
6534 devlist
= get_devices(hpath
);
6535 /* if no intel devices return zero volumes */
6536 if (devlist
== NULL
)
6539 count
+= active_arrays_by_format("imsm", hpath
, &devlist
, dpa
,
6541 dprintf("path: %s active arrays: %d\n", hpath
, count
);
6542 if (devlist
== NULL
)
6546 count
+= count_volumes_list(devlist
,
6550 dprintf("found %d count: %d\n", found
, count
);
6553 dprintf("path: %s total number of volumes: %d\n", hpath
, count
);
6556 struct md_list
*dv
= devlist
;
6557 devlist
= devlist
->next
;
6565 static int count_volumes(struct intel_hba
*hba
, int dpa
, int verbose
)
6569 if (hba
->type
== SYS_DEV_VMD
) {
6570 struct sys_dev
*dev
;
6573 for (dev
= find_intel_devices(); dev
; dev
= dev
->next
) {
6574 if (dev
->type
== SYS_DEV_VMD
)
6575 count
+= __count_volumes(dev
->path
, dpa
,
6580 return __count_volumes(hba
->path
, dpa
, verbose
, 0);
6583 static int imsm_default_chunk(const struct imsm_orom
*orom
)
6585 /* up to 512 if the plaform supports it, otherwise the platform max.
6586 * 128 if no platform detected
6588 int fs
= max(7, orom
? fls(orom
->sss
) : 0);
6590 return min(512, (1 << fs
));
6594 validate_geometry_imsm_orom(struct intel_super
*super
, int level
, int layout
,
6595 int raiddisks
, int *chunk
, unsigned long long size
, int verbose
)
6597 /* check/set platform and metadata limits/defaults */
6598 if (super
->orom
&& raiddisks
> super
->orom
->dpa
) {
6599 pr_vrb("platform supports a maximum of %d disks per array\n",
6604 /* capabilities of OROM tested - copied from validate_geometry_imsm_volume */
6605 if (!is_raid_level_supported(super
->orom
, level
, raiddisks
)) {
6606 pr_vrb("platform does not support raid%d with %d disk%s\n",
6607 level
, raiddisks
, raiddisks
> 1 ? "s" : "");
6611 if (*chunk
== 0 || *chunk
== UnSet
)
6612 *chunk
= imsm_default_chunk(super
->orom
);
6614 if (super
->orom
&& !imsm_orom_has_chunk(super
->orom
, *chunk
)) {
6615 pr_vrb("platform does not support a chunk size of: %d\n", *chunk
);
6619 if (layout
!= imsm_level_to_layout(level
)) {
6621 pr_vrb("imsm raid 5 only supports the left-asymmetric layout\n");
6622 else if (level
== 10)
6623 pr_vrb("imsm raid 10 only supports the n2 layout\n");
6625 pr_vrb("imsm unknown layout %#x for this raid level %d\n",
6630 if (super
->orom
&& (super
->orom
->attr
& IMSM_OROM_ATTR_2TB
) == 0 &&
6631 (calc_array_size(level
, raiddisks
, layout
, *chunk
, size
) >> 32) > 0) {
6632 pr_vrb("platform does not support a volume size over 2TB\n");
6639 /* validate_geometry_imsm_volume - lifted from validate_geometry_ddf_bvd
6640 * FIX ME add ahci details
6642 static int validate_geometry_imsm_volume(struct supertype
*st
, int level
,
6643 int layout
, int raiddisks
, int *chunk
,
6644 unsigned long long size
,
6645 unsigned long long data_offset
,
6647 unsigned long long *freesize
,
6651 struct intel_super
*super
= st
->sb
;
6652 struct imsm_super
*mpb
;
6654 unsigned long long pos
= 0;
6655 unsigned long long maxsize
;
6659 /* We must have the container info already read in. */
6663 mpb
= super
->anchor
;
6665 if (!validate_geometry_imsm_orom(super
, level
, layout
, raiddisks
, chunk
, size
, verbose
)) {
6666 pr_err("RAID gemetry validation failed. Cannot proceed with the action(s).\n");
6670 /* General test: make sure there is space for
6671 * 'raiddisks' device extents of size 'size' at a given
6674 unsigned long long minsize
= size
;
6675 unsigned long long start_offset
= MaxSector
;
6678 minsize
= MPB_SECTOR_CNT
+ IMSM_RESERVED_SECTORS
;
6679 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
6684 e
= get_extents(super
, dl
);
6687 unsigned long long esize
;
6688 esize
= e
[i
].start
- pos
;
6689 if (esize
>= minsize
)
6691 if (found
&& start_offset
== MaxSector
) {
6694 } else if (found
&& pos
!= start_offset
) {
6698 pos
= e
[i
].start
+ e
[i
].size
;
6700 } while (e
[i
-1].size
);
6705 if (dcnt
< raiddisks
) {
6707 pr_err("imsm: Not enough devices with space for this array (%d < %d)\n",
6714 /* This device must be a member of the set */
6715 if (stat(dev
, &stb
) < 0)
6717 if ((S_IFMT
& stb
.st_mode
) != S_IFBLK
)
6719 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
6720 if (dl
->major
== (int)major(stb
.st_rdev
) &&
6721 dl
->minor
== (int)minor(stb
.st_rdev
))
6726 pr_err("%s is not in the same imsm set\n", dev
);
6728 } else if (super
->orom
&& dl
->index
< 0 && mpb
->num_raid_devs
) {
6729 /* If a volume is present then the current creation attempt
6730 * cannot incorporate new spares because the orom may not
6731 * understand this configuration (all member disks must be
6732 * members of each array in the container).
6734 pr_err("%s is a spare and a volume is already defined for this container\n", dev
);
6735 pr_err("The option-rom requires all member disks to be a member of all volumes\n");
6737 } else if (super
->orom
&& mpb
->num_raid_devs
> 0 &&
6738 mpb
->num_disks
!= raiddisks
) {
6739 pr_err("The option-rom requires all member disks to be a member of all volumes\n");
6743 /* retrieve the largest free space block */
6744 e
= get_extents(super
, dl
);
6749 unsigned long long esize
;
6751 esize
= e
[i
].start
- pos
;
6752 if (esize
>= maxsize
)
6754 pos
= e
[i
].start
+ e
[i
].size
;
6756 } while (e
[i
-1].size
);
6761 pr_err("unable to determine free space for: %s\n",
6765 if (maxsize
< size
) {
6767 pr_err("%s not enough space (%llu < %llu)\n",
6768 dev
, maxsize
, size
);
6772 /* count total number of extents for merge */
6774 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
6776 i
+= dl
->extent_cnt
;
6778 maxsize
= merge_extents(super
, i
);
6780 if (!check_env("IMSM_NO_PLATFORM") &&
6781 mpb
->num_raid_devs
> 0 && size
&& size
!= maxsize
) {
6782 pr_err("attempting to create a second volume with size less then remaining space. Aborting...\n");
6786 if (maxsize
< size
|| maxsize
== 0) {
6789 pr_err("no free space left on device. Aborting...\n");
6791 pr_err("not enough space to create volume of given size (%llu < %llu). Aborting...\n",
6797 *freesize
= maxsize
;
6800 int count
= count_volumes(super
->hba
,
6801 super
->orom
->dpa
, verbose
);
6802 if (super
->orom
->vphba
<= count
) {
6803 pr_vrb("platform does not support more than %d raid volumes.\n",
6804 super
->orom
->vphba
);
6811 static int imsm_get_free_size(struct supertype
*st
, int raiddisks
,
6812 unsigned long long size
, int chunk
,
6813 unsigned long long *freesize
)
6815 struct intel_super
*super
= st
->sb
;
6816 struct imsm_super
*mpb
= super
->anchor
;
6821 unsigned long long maxsize
;
6822 unsigned long long minsize
;
6826 /* find the largest common start free region of the possible disks */
6830 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
6836 /* don't activate new spares if we are orom constrained
6837 * and there is already a volume active in the container
6839 if (super
->orom
&& dl
->index
< 0 && mpb
->num_raid_devs
)
6842 e
= get_extents(super
, dl
);
6845 for (i
= 1; e
[i
-1].size
; i
++)
6853 maxsize
= merge_extents(super
, extent_cnt
);
6857 minsize
= chunk
* 2;
6859 if (cnt
< raiddisks
||
6860 (super
->orom
&& used
&& used
!= raiddisks
) ||
6861 maxsize
< minsize
||
6863 pr_err("not enough devices with space to create array.\n");
6864 return 0; /* No enough free spaces large enough */
6875 if (!check_env("IMSM_NO_PLATFORM") &&
6876 mpb
->num_raid_devs
> 0 && size
&& size
!= maxsize
) {
6877 pr_err("attempting to create a second volume with size less then remaining space. Aborting...\n");
6881 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
6883 dl
->raiddisk
= cnt
++;
6887 dprintf("imsm: imsm_get_free_size() returns : %llu\n", size
);
6892 static int reserve_space(struct supertype
*st
, int raiddisks
,
6893 unsigned long long size
, int chunk
,
6894 unsigned long long *freesize
)
6896 struct intel_super
*super
= st
->sb
;
6901 rv
= imsm_get_free_size(st
, raiddisks
, size
, chunk
, freesize
);
6904 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
6906 dl
->raiddisk
= cnt
++;
6913 static int validate_geometry_imsm(struct supertype
*st
, int level
, int layout
,
6914 int raiddisks
, int *chunk
, unsigned long long size
,
6915 unsigned long long data_offset
,
6916 char *dev
, unsigned long long *freesize
,
6924 * if given unused devices create a container
6925 * if given given devices in a container create a member volume
6927 if (level
== LEVEL_CONTAINER
) {
6928 /* Must be a fresh device to add to a container */
6929 return validate_geometry_imsm_container(st
, level
, layout
,
6939 struct intel_super
*super
= st
->sb
;
6940 if (!validate_geometry_imsm_orom(st
->sb
, level
, layout
,
6941 raiddisks
, chunk
, size
,
6944 /* we are being asked to automatically layout a
6945 * new volume based on the current contents of
6946 * the container. If the the parameters can be
6947 * satisfied reserve_space will record the disks,
6948 * start offset, and size of the volume to be
6949 * created. add_to_super and getinfo_super
6950 * detect when autolayout is in progress.
6952 /* assuming that freesize is always given when array is
6954 if (super
->orom
&& freesize
) {
6956 count
= count_volumes(super
->hba
,
6957 super
->orom
->dpa
, verbose
);
6958 if (super
->orom
->vphba
<= count
) {
6959 pr_vrb("platform does not support more than %d raid volumes.\n",
6960 super
->orom
->vphba
);
6965 return reserve_space(st
, raiddisks
, size
,
6971 /* creating in a given container */
6972 return validate_geometry_imsm_volume(st
, level
, layout
,
6973 raiddisks
, chunk
, size
,
6975 dev
, freesize
, verbose
);
6978 /* This device needs to be a device in an 'imsm' container */
6979 fd
= open(dev
, O_RDONLY
|O_EXCL
, 0);
6982 pr_err("Cannot create this array on device %s\n",
6987 if (errno
!= EBUSY
|| (fd
= open(dev
, O_RDONLY
, 0)) < 0) {
6989 pr_err("Cannot open %s: %s\n",
6990 dev
, strerror(errno
));
6993 /* Well, it is in use by someone, maybe an 'imsm' container. */
6994 cfd
= open_container(fd
);
6998 pr_err("Cannot use %s: It is busy\n",
7002 sra
= sysfs_read(cfd
, NULL
, GET_VERSION
);
7003 if (sra
&& sra
->array
.major_version
== -1 &&
7004 strcmp(sra
->text_version
, "imsm") == 0)
7008 /* This is a member of a imsm container. Load the container
7009 * and try to create a volume
7011 struct intel_super
*super
;
7013 if (load_super_imsm_all(st
, cfd
, (void **) &super
, NULL
, NULL
, 1) == 0) {
7015 strcpy(st
->container_devnm
, fd2devnm(cfd
));
7017 return validate_geometry_imsm_volume(st
, level
, layout
,
7019 size
, data_offset
, dev
,
7026 pr_err("failed container membership check\n");
7032 static void default_geometry_imsm(struct supertype
*st
, int *level
, int *layout
, int *chunk
)
7034 struct intel_super
*super
= st
->sb
;
7036 if (level
&& *level
== UnSet
)
7037 *level
= LEVEL_CONTAINER
;
7039 if (level
&& layout
&& *layout
== UnSet
)
7040 *layout
= imsm_level_to_layout(*level
);
7042 if (chunk
&& (*chunk
== UnSet
|| *chunk
== 0))
7043 *chunk
= imsm_default_chunk(super
->orom
);
7046 static void handle_missing(struct intel_super
*super
, struct imsm_dev
*dev
);
7048 static int kill_subarray_imsm(struct supertype
*st
)
7050 /* remove the subarray currently referenced by ->current_vol */
7052 struct intel_dev
**dp
;
7053 struct intel_super
*super
= st
->sb
;
7054 __u8 current_vol
= super
->current_vol
;
7055 struct imsm_super
*mpb
= super
->anchor
;
7057 if (super
->current_vol
< 0)
7059 super
->current_vol
= -1; /* invalidate subarray cursor */
7061 /* block deletions that would change the uuid of active subarrays
7063 * FIXME when immutable ids are available, but note that we'll
7064 * also need to fixup the invalidated/active subarray indexes in
7067 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
7070 if (i
< current_vol
)
7072 sprintf(subarray
, "%u", i
);
7073 if (is_subarray_active(subarray
, st
->devnm
)) {
7074 pr_err("deleting subarray-%d would change the UUID of active subarray-%d, aborting\n",
7081 if (st
->update_tail
) {
7082 struct imsm_update_kill_array
*u
= xmalloc(sizeof(*u
));
7084 u
->type
= update_kill_array
;
7085 u
->dev_idx
= current_vol
;
7086 append_metadata_update(st
, u
, sizeof(*u
));
7091 for (dp
= &super
->devlist
; *dp
;)
7092 if ((*dp
)->index
== current_vol
) {
7095 handle_missing(super
, (*dp
)->dev
);
7096 if ((*dp
)->index
> current_vol
)
7101 /* no more raid devices, all active components are now spares,
7102 * but of course failed are still failed
7104 if (--mpb
->num_raid_devs
== 0) {
7107 for (d
= super
->disks
; d
; d
= d
->next
)
7112 super
->updates_pending
++;
7117 static int update_subarray_imsm(struct supertype
*st
, char *subarray
,
7118 char *update
, struct mddev_ident
*ident
)
7120 /* update the subarray currently referenced by ->current_vol */
7121 struct intel_super
*super
= st
->sb
;
7122 struct imsm_super
*mpb
= super
->anchor
;
7124 if (strcmp(update
, "name") == 0) {
7125 char *name
= ident
->name
;
7129 if (is_subarray_active(subarray
, st
->devnm
)) {
7130 pr_err("Unable to update name of active subarray\n");
7134 if (!check_name(super
, name
, 0))
7137 vol
= strtoul(subarray
, &ep
, 10);
7138 if (*ep
!= '\0' || vol
>= super
->anchor
->num_raid_devs
)
7141 if (st
->update_tail
) {
7142 struct imsm_update_rename_array
*u
= xmalloc(sizeof(*u
));
7144 u
->type
= update_rename_array
;
7146 strncpy((char *) u
->name
, name
, MAX_RAID_SERIAL_LEN
);
7147 u
->name
[MAX_RAID_SERIAL_LEN
-1] = '\0';
7148 append_metadata_update(st
, u
, sizeof(*u
));
7150 struct imsm_dev
*dev
;
7153 dev
= get_imsm_dev(super
, vol
);
7154 strncpy((char *) dev
->volume
, name
, MAX_RAID_SERIAL_LEN
);
7155 dev
->volume
[MAX_RAID_SERIAL_LEN
-1] = '\0';
7156 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
7157 dev
= get_imsm_dev(super
, i
);
7158 handle_missing(super
, dev
);
7160 super
->updates_pending
++;
7167 #endif /* MDASSEMBLE */
7169 static int is_gen_migration(struct imsm_dev
*dev
)
7174 if (!dev
->vol
.migr_state
)
7177 if (migr_type(dev
) == MIGR_GEN_MIGR
)
7183 static int is_rebuilding(struct imsm_dev
*dev
)
7185 struct imsm_map
*migr_map
;
7187 if (!dev
->vol
.migr_state
)
7190 if (migr_type(dev
) != MIGR_REBUILD
)
7193 migr_map
= get_imsm_map(dev
, MAP_1
);
7195 if (migr_map
->map_state
== IMSM_T_STATE_DEGRADED
)
7202 static int is_initializing(struct imsm_dev
*dev
)
7204 struct imsm_map
*migr_map
;
7206 if (!dev
->vol
.migr_state
)
7209 if (migr_type(dev
) != MIGR_INIT
)
7212 migr_map
= get_imsm_map(dev
, MAP_1
);
7214 if (migr_map
->map_state
== IMSM_T_STATE_UNINITIALIZED
)
7221 static void update_recovery_start(struct intel_super
*super
,
7222 struct imsm_dev
*dev
,
7223 struct mdinfo
*array
)
7225 struct mdinfo
*rebuild
= NULL
;
7229 if (!is_rebuilding(dev
))
7232 /* Find the rebuild target, but punt on the dual rebuild case */
7233 for (d
= array
->devs
; d
; d
= d
->next
)
7234 if (d
->recovery_start
== 0) {
7241 /* (?) none of the disks are marked with
7242 * IMSM_ORD_REBUILD, so assume they are missing and the
7243 * disk_ord_tbl was not correctly updated
7245 dprintf("failed to locate out-of-sync disk\n");
7249 units
= __le32_to_cpu(dev
->vol
.curr_migr_unit
);
7250 rebuild
->recovery_start
= units
* blocks_per_migr_unit(super
, dev
);
7254 static int recover_backup_imsm(struct supertype
*st
, struct mdinfo
*info
);
7257 static struct mdinfo
*container_content_imsm(struct supertype
*st
, char *subarray
)
7259 /* Given a container loaded by load_super_imsm_all,
7260 * extract information about all the arrays into
7262 * If 'subarray' is given, just extract info about that array.
7264 * For each imsm_dev create an mdinfo, fill it in,
7265 * then look for matching devices in super->disks
7266 * and create appropriate device mdinfo.
7268 struct intel_super
*super
= st
->sb
;
7269 struct imsm_super
*mpb
= super
->anchor
;
7270 struct mdinfo
*rest
= NULL
;
7274 int spare_disks
= 0;
7276 /* do not assemble arrays when not all attributes are supported */
7277 if (imsm_check_attributes(mpb
->attributes
) == 0) {
7279 pr_err("Unsupported attributes in IMSM metadata.Arrays activation is blocked.\n");
7282 /* count spare devices, not used in maps
7284 for (d
= super
->disks
; d
; d
= d
->next
)
7288 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
7289 struct imsm_dev
*dev
;
7290 struct imsm_map
*map
;
7291 struct imsm_map
*map2
;
7292 struct mdinfo
*this;
7300 (i
!= strtoul(subarray
, &ep
, 10) || *ep
!= '\0'))
7303 dev
= get_imsm_dev(super
, i
);
7304 map
= get_imsm_map(dev
, MAP_0
);
7305 map2
= get_imsm_map(dev
, MAP_1
);
7307 /* do not publish arrays that are in the middle of an
7308 * unsupported migration
7310 if (dev
->vol
.migr_state
&&
7311 (migr_type(dev
) == MIGR_STATE_CHANGE
)) {
7312 pr_err("cannot assemble volume '%.16s': unsupported migration in progress\n",
7316 /* do not publish arrays that are not support by controller's
7320 this = xmalloc(sizeof(*this));
7322 super
->current_vol
= i
;
7323 getinfo_super_imsm_volume(st
, this, NULL
);
7326 chunk
= __le16_to_cpu(map
->blocks_per_strip
) >> 1;
7327 /* mdadm does not support all metadata features- set the bit in all arrays state */
7328 if (!validate_geometry_imsm_orom(super
,
7329 get_imsm_raid_level(map
), /* RAID level */
7330 imsm_level_to_layout(get_imsm_raid_level(map
)),
7331 map
->num_members
, /* raid disks */
7332 &chunk
, join_u32(dev
->size_low
, dev
->size_high
),
7334 pr_err("IMSM RAID geometry validation failed. Array %s activation is blocked.\n",
7336 this->array
.state
|=
7337 (1<<MD_SB_BLOCK_CONTAINER_RESHAPE
) |
7338 (1<<MD_SB_BLOCK_VOLUME
);
7342 /* if array has bad blocks, set suitable bit in all arrays state */
7344 this->array
.state
|=
7345 (1<<MD_SB_BLOCK_CONTAINER_RESHAPE
) |
7346 (1<<MD_SB_BLOCK_VOLUME
);
7348 for (slot
= 0 ; slot
< map
->num_members
; slot
++) {
7349 unsigned long long recovery_start
;
7350 struct mdinfo
*info_d
;
7357 idx
= get_imsm_disk_idx(dev
, slot
, MAP_0
);
7358 ord
= get_imsm_ord_tbl_ent(dev
, slot
, MAP_X
);
7359 for (d
= super
->disks
; d
; d
= d
->next
)
7360 if (d
->index
== idx
)
7363 recovery_start
= MaxSector
;
7366 if (d
&& is_failed(&d
->disk
))
7368 if (ord
& IMSM_ORD_REBUILD
)
7372 * if we skip some disks the array will be assmebled degraded;
7373 * reset resync start to avoid a dirty-degraded
7374 * situation when performing the intial sync
7376 * FIXME handle dirty degraded
7378 if ((skip
|| recovery_start
== 0) && !dev
->vol
.dirty
)
7379 this->resync_start
= MaxSector
;
7383 info_d
= xcalloc(1, sizeof(*info_d
));
7384 info_d
->next
= this->devs
;
7385 this->devs
= info_d
;
7387 info_d
->disk
.number
= d
->index
;
7388 info_d
->disk
.major
= d
->major
;
7389 info_d
->disk
.minor
= d
->minor
;
7390 info_d
->disk
.raid_disk
= slot
;
7391 info_d
->recovery_start
= recovery_start
;
7393 if (slot
< map2
->num_members
)
7394 info_d
->disk
.state
= (1 << MD_DISK_ACTIVE
);
7396 this->array
.spare_disks
++;
7398 if (slot
< map
->num_members
)
7399 info_d
->disk
.state
= (1 << MD_DISK_ACTIVE
);
7401 this->array
.spare_disks
++;
7403 if (info_d
->recovery_start
== MaxSector
)
7404 this->array
.working_disks
++;
7406 info_d
->events
= __le32_to_cpu(mpb
->generation_num
);
7407 info_d
->data_offset
= pba_of_lba0(map
);
7409 if (map
->raid_level
== 5) {
7410 info_d
->component_size
=
7411 num_data_stripes(map
) *
7412 map
->blocks_per_strip
;
7414 info_d
->component_size
= blocks_per_member(map
);
7417 info_d
->bb
.supported
= 1;
7418 get_volume_badblocks(super
->bbm_log
, ord_to_idx(ord
),
7419 info_d
->data_offset
,
7420 info_d
->component_size
,
7423 /* now that the disk list is up-to-date fixup recovery_start */
7424 update_recovery_start(super
, dev
, this);
7425 this->array
.spare_disks
+= spare_disks
;
7428 /* check for reshape */
7429 if (this->reshape_active
== 1)
7430 recover_backup_imsm(st
, this);
7438 static __u8
imsm_check_degraded(struct intel_super
*super
, struct imsm_dev
*dev
,
7439 int failed
, int look_in_map
)
7441 struct imsm_map
*map
;
7443 map
= get_imsm_map(dev
, look_in_map
);
7446 return map
->map_state
== IMSM_T_STATE_UNINITIALIZED
?
7447 IMSM_T_STATE_UNINITIALIZED
: IMSM_T_STATE_NORMAL
;
7449 switch (get_imsm_raid_level(map
)) {
7451 return IMSM_T_STATE_FAILED
;
7454 if (failed
< map
->num_members
)
7455 return IMSM_T_STATE_DEGRADED
;
7457 return IMSM_T_STATE_FAILED
;
7462 * check to see if any mirrors have failed, otherwise we
7463 * are degraded. Even numbered slots are mirrored on
7467 /* gcc -Os complains that this is unused */
7468 int insync
= insync
;
7470 for (i
= 0; i
< map
->num_members
; i
++) {
7471 __u32 ord
= get_imsm_ord_tbl_ent(dev
, i
, MAP_X
);
7472 int idx
= ord_to_idx(ord
);
7473 struct imsm_disk
*disk
;
7475 /* reset the potential in-sync count on even-numbered
7476 * slots. num_copies is always 2 for imsm raid10
7481 disk
= get_imsm_disk(super
, idx
);
7482 if (!disk
|| is_failed(disk
) || ord
& IMSM_ORD_REBUILD
)
7485 /* no in-sync disks left in this mirror the
7489 return IMSM_T_STATE_FAILED
;
7492 return IMSM_T_STATE_DEGRADED
;
7496 return IMSM_T_STATE_DEGRADED
;
7498 return IMSM_T_STATE_FAILED
;
7504 return map
->map_state
;
7507 static int imsm_count_failed(struct intel_super
*super
, struct imsm_dev
*dev
,
7512 struct imsm_disk
*disk
;
7513 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
7514 struct imsm_map
*prev
= get_imsm_map(dev
, MAP_1
);
7515 struct imsm_map
*map_for_loop
;
7520 /* at the beginning of migration we set IMSM_ORD_REBUILD on
7521 * disks that are being rebuilt. New failures are recorded to
7522 * map[0]. So we look through all the disks we started with and
7523 * see if any failures are still present, or if any new ones
7527 if (prev
&& (map
->num_members
< prev
->num_members
))
7528 map_for_loop
= prev
;
7530 for (i
= 0; i
< map_for_loop
->num_members
; i
++) {
7532 /* when MAP_X is passed both maps failures are counted
7535 (look_in_map
== MAP_1
|| look_in_map
== MAP_X
) &&
7536 i
< prev
->num_members
) {
7537 ord
= __le32_to_cpu(prev
->disk_ord_tbl
[i
]);
7538 idx_1
= ord_to_idx(ord
);
7540 disk
= get_imsm_disk(super
, idx_1
);
7541 if (!disk
|| is_failed(disk
) || ord
& IMSM_ORD_REBUILD
)
7544 if ((look_in_map
== MAP_0
|| look_in_map
== MAP_X
) &&
7545 i
< map
->num_members
) {
7546 ord
= __le32_to_cpu(map
->disk_ord_tbl
[i
]);
7547 idx
= ord_to_idx(ord
);
7550 disk
= get_imsm_disk(super
, idx
);
7551 if (!disk
|| is_failed(disk
) ||
7552 ord
& IMSM_ORD_REBUILD
)
7562 static int imsm_open_new(struct supertype
*c
, struct active_array
*a
,
7565 struct intel_super
*super
= c
->sb
;
7566 struct imsm_super
*mpb
= super
->anchor
;
7567 struct imsm_update_prealloc_bb_mem u
;
7569 if (atoi(inst
) >= mpb
->num_raid_devs
) {
7570 pr_err("subarry index %d, out of range\n", atoi(inst
));
7574 dprintf("imsm: open_new %s\n", inst
);
7575 a
->info
.container_member
= atoi(inst
);
7577 u
.type
= update_prealloc_badblocks_mem
;
7578 imsm_update_metadata_locally(c
, &u
, sizeof(u
));
7583 static int is_resyncing(struct imsm_dev
*dev
)
7585 struct imsm_map
*migr_map
;
7587 if (!dev
->vol
.migr_state
)
7590 if (migr_type(dev
) == MIGR_INIT
||
7591 migr_type(dev
) == MIGR_REPAIR
)
7594 if (migr_type(dev
) == MIGR_GEN_MIGR
)
7597 migr_map
= get_imsm_map(dev
, MAP_1
);
7599 if (migr_map
->map_state
== IMSM_T_STATE_NORMAL
&&
7600 dev
->vol
.migr_type
!= MIGR_GEN_MIGR
)
7606 /* return true if we recorded new information */
7607 static int mark_failure(struct intel_super
*super
,
7608 struct imsm_dev
*dev
, struct imsm_disk
*disk
, int idx
)
7612 struct imsm_map
*map
;
7613 char buf
[MAX_RAID_SERIAL_LEN
+3];
7614 unsigned int len
, shift
= 0;
7616 /* new failures are always set in map[0] */
7617 map
= get_imsm_map(dev
, MAP_0
);
7619 slot
= get_imsm_disk_slot(map
, idx
);
7623 ord
= __le32_to_cpu(map
->disk_ord_tbl
[slot
]);
7624 if (is_failed(disk
) && (ord
& IMSM_ORD_REBUILD
))
7627 memcpy(buf
, disk
->serial
, MAX_RAID_SERIAL_LEN
);
7628 buf
[MAX_RAID_SERIAL_LEN
] = '\000';
7630 if ((len
= strlen(buf
)) >= MAX_RAID_SERIAL_LEN
)
7631 shift
= len
- MAX_RAID_SERIAL_LEN
+ 1;
7632 strncpy((char *)disk
->serial
, &buf
[shift
], MAX_RAID_SERIAL_LEN
);
7634 disk
->status
|= FAILED_DISK
;
7635 set_imsm_ord_tbl_ent(map
, slot
, idx
| IMSM_ORD_REBUILD
);
7636 /* mark failures in second map if second map exists and this disk
7638 * This is valid for migration, initialization and rebuild
7640 if (dev
->vol
.migr_state
) {
7641 struct imsm_map
*map2
= get_imsm_map(dev
, MAP_1
);
7642 int slot2
= get_imsm_disk_slot(map2
, idx
);
7644 if (slot2
< map2
->num_members
&& slot2
>= 0)
7645 set_imsm_ord_tbl_ent(map2
, slot2
,
7646 idx
| IMSM_ORD_REBUILD
);
7648 if (map
->failed_disk_num
== 0xff)
7649 map
->failed_disk_num
= slot
;
7651 clear_disk_badblocks(super
->bbm_log
, ord_to_idx(ord
));
7656 static void mark_missing(struct intel_super
*super
,
7657 struct imsm_dev
*dev
, struct imsm_disk
*disk
, int idx
)
7659 mark_failure(super
, dev
, disk
, idx
);
7661 if (disk
->scsi_id
== __cpu_to_le32(~(__u32
)0))
7664 disk
->scsi_id
= __cpu_to_le32(~(__u32
)0);
7665 memmove(&disk
->serial
[0], &disk
->serial
[1], MAX_RAID_SERIAL_LEN
- 1);
7668 static void handle_missing(struct intel_super
*super
, struct imsm_dev
*dev
)
7672 if (!super
->missing
)
7675 /* When orom adds replacement for missing disk it does
7676 * not remove entry of missing disk, but just updates map with
7677 * new added disk. So it is not enough just to test if there is
7678 * any missing disk, we have to look if there are any failed disks
7679 * in map to stop migration */
7681 dprintf("imsm: mark missing\n");
7682 /* end process for initialization and rebuild only
7684 if (is_gen_migration(dev
) == 0) {
7688 failed
= imsm_count_failed(super
, dev
, MAP_0
);
7689 map_state
= imsm_check_degraded(super
, dev
, failed
, MAP_0
);
7692 end_migration(dev
, super
, map_state
);
7694 for (dl
= super
->missing
; dl
; dl
= dl
->next
)
7695 mark_missing(super
, dev
, &dl
->disk
, dl
->index
);
7696 super
->updates_pending
++;
7699 static unsigned long long imsm_set_array_size(struct imsm_dev
*dev
,
7702 int used_disks
= imsm_num_data_members(dev
, MAP_0
);
7703 unsigned long long array_blocks
;
7704 struct imsm_map
*map
;
7706 if (used_disks
== 0) {
7707 /* when problems occures
7708 * return current array_blocks value
7710 array_blocks
= __le32_to_cpu(dev
->size_high
);
7711 array_blocks
= array_blocks
<< 32;
7712 array_blocks
+= __le32_to_cpu(dev
->size_low
);
7714 return array_blocks
;
7717 /* set array size in metadata
7719 if (new_size
<= 0) {
7720 /* OLCE size change is caused by added disks
7722 map
= get_imsm_map(dev
, MAP_0
);
7723 array_blocks
= blocks_per_member(map
) * used_disks
;
7725 /* Online Volume Size Change
7726 * Using available free space
7728 array_blocks
= new_size
;
7731 /* round array size down to closest MB
7733 array_blocks
= (array_blocks
>> SECT_PER_MB_SHIFT
) << SECT_PER_MB_SHIFT
;
7734 dev
->size_low
= __cpu_to_le32((__u32
)array_blocks
);
7735 dev
->size_high
= __cpu_to_le32((__u32
)(array_blocks
>> 32));
7737 return array_blocks
;
7740 static void imsm_set_disk(struct active_array
*a
, int n
, int state
);
7742 static void imsm_progress_container_reshape(struct intel_super
*super
)
7744 /* if no device has a migr_state, but some device has a
7745 * different number of members than the previous device, start
7746 * changing the number of devices in this device to match
7749 struct imsm_super
*mpb
= super
->anchor
;
7750 int prev_disks
= -1;
7754 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
7755 struct imsm_dev
*dev
= get_imsm_dev(super
, i
);
7756 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
7757 struct imsm_map
*map2
;
7758 int prev_num_members
;
7760 if (dev
->vol
.migr_state
)
7763 if (prev_disks
== -1)
7764 prev_disks
= map
->num_members
;
7765 if (prev_disks
== map
->num_members
)
7768 /* OK, this array needs to enter reshape mode.
7769 * i.e it needs a migr_state
7772 copy_map_size
= sizeof_imsm_map(map
);
7773 prev_num_members
= map
->num_members
;
7774 map
->num_members
= prev_disks
;
7775 dev
->vol
.migr_state
= 1;
7776 dev
->vol
.curr_migr_unit
= 0;
7777 set_migr_type(dev
, MIGR_GEN_MIGR
);
7778 for (i
= prev_num_members
;
7779 i
< map
->num_members
; i
++)
7780 set_imsm_ord_tbl_ent(map
, i
, i
);
7781 map2
= get_imsm_map(dev
, MAP_1
);
7782 /* Copy the current map */
7783 memcpy(map2
, map
, copy_map_size
);
7784 map2
->num_members
= prev_num_members
;
7786 imsm_set_array_size(dev
, -1);
7787 super
->clean_migration_record_by_mdmon
= 1;
7788 super
->updates_pending
++;
7792 /* Handle dirty -> clean transititions, resync and reshape. Degraded and rebuild
7793 * states are handled in imsm_set_disk() with one exception, when a
7794 * resync is stopped due to a new failure this routine will set the
7795 * 'degraded' state for the array.
7797 static int imsm_set_array_state(struct active_array
*a
, int consistent
)
7799 int inst
= a
->info
.container_member
;
7800 struct intel_super
*super
= a
->container
->sb
;
7801 struct imsm_dev
*dev
= get_imsm_dev(super
, inst
);
7802 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
7803 int failed
= imsm_count_failed(super
, dev
, MAP_0
);
7804 __u8 map_state
= imsm_check_degraded(super
, dev
, failed
, MAP_0
);
7805 __u32 blocks_per_unit
;
7807 if (dev
->vol
.migr_state
&&
7808 dev
->vol
.migr_type
== MIGR_GEN_MIGR
) {
7809 /* array state change is blocked due to reshape action
7811 * - abort the reshape (if last_checkpoint is 0 and action!= reshape)
7812 * - finish the reshape (if last_checkpoint is big and action != reshape)
7813 * - update curr_migr_unit
7815 if (a
->curr_action
== reshape
) {
7816 /* still reshaping, maybe update curr_migr_unit */
7817 goto mark_checkpoint
;
7819 if (a
->last_checkpoint
== 0 && a
->prev_action
== reshape
) {
7820 /* for some reason we aborted the reshape.
7822 * disable automatic metadata rollback
7823 * user action is required to recover process
7826 struct imsm_map
*map2
=
7827 get_imsm_map(dev
, MAP_1
);
7828 dev
->vol
.migr_state
= 0;
7829 set_migr_type(dev
, 0);
7830 dev
->vol
.curr_migr_unit
= 0;
7832 sizeof_imsm_map(map2
));
7833 super
->updates_pending
++;
7836 if (a
->last_checkpoint
>= a
->info
.component_size
) {
7837 unsigned long long array_blocks
;
7841 used_disks
= imsm_num_data_members(dev
, MAP_0
);
7842 if (used_disks
> 0) {
7844 blocks_per_member(map
) *
7846 /* round array size down to closest MB
7848 array_blocks
= (array_blocks
7849 >> SECT_PER_MB_SHIFT
)
7850 << SECT_PER_MB_SHIFT
;
7851 a
->info
.custom_array_size
= array_blocks
;
7852 /* encourage manager to update array
7856 a
->check_reshape
= 1;
7858 /* finalize online capacity expansion/reshape */
7859 for (mdi
= a
->info
.devs
; mdi
; mdi
= mdi
->next
)
7861 mdi
->disk
.raid_disk
,
7864 imsm_progress_container_reshape(super
);
7869 /* before we activate this array handle any missing disks */
7870 if (consistent
== 2)
7871 handle_missing(super
, dev
);
7873 if (consistent
== 2 &&
7874 (!is_resync_complete(&a
->info
) ||
7875 map_state
!= IMSM_T_STATE_NORMAL
||
7876 dev
->vol
.migr_state
))
7879 if (is_resync_complete(&a
->info
)) {
7880 /* complete intialization / resync,
7881 * recovery and interrupted recovery is completed in
7884 if (is_resyncing(dev
)) {
7885 dprintf("imsm: mark resync done\n");
7886 end_migration(dev
, super
, map_state
);
7887 super
->updates_pending
++;
7888 a
->last_checkpoint
= 0;
7890 } else if ((!is_resyncing(dev
) && !failed
) &&
7891 (imsm_reshape_blocks_arrays_changes(super
) == 0)) {
7892 /* mark the start of the init process if nothing is failed */
7893 dprintf("imsm: mark resync start\n");
7894 if (map
->map_state
== IMSM_T_STATE_UNINITIALIZED
)
7895 migrate(dev
, super
, IMSM_T_STATE_NORMAL
, MIGR_INIT
);
7897 migrate(dev
, super
, IMSM_T_STATE_NORMAL
, MIGR_REPAIR
);
7898 super
->updates_pending
++;
7902 /* skip checkpointing for general migration,
7903 * it is controlled in mdadm
7905 if (is_gen_migration(dev
))
7906 goto skip_mark_checkpoint
;
7908 /* check if we can update curr_migr_unit from resync_start, recovery_start */
7909 blocks_per_unit
= blocks_per_migr_unit(super
, dev
);
7910 if (blocks_per_unit
) {
7914 units
= a
->last_checkpoint
/ blocks_per_unit
;
7917 /* check that we did not overflow 32-bits, and that
7918 * curr_migr_unit needs updating
7920 if (units32
== units
&&
7922 __le32_to_cpu(dev
->vol
.curr_migr_unit
) != units32
) {
7923 dprintf("imsm: mark checkpoint (%u)\n", units32
);
7924 dev
->vol
.curr_migr_unit
= __cpu_to_le32(units32
);
7925 super
->updates_pending
++;
7929 skip_mark_checkpoint
:
7930 /* mark dirty / clean */
7931 if (dev
->vol
.dirty
!= !consistent
) {
7932 dprintf("imsm: mark '%s'\n", consistent
? "clean" : "dirty");
7937 super
->updates_pending
++;
7943 static int imsm_disk_slot_to_ord(struct active_array
*a
, int slot
)
7945 int inst
= a
->info
.container_member
;
7946 struct intel_super
*super
= a
->container
->sb
;
7947 struct imsm_dev
*dev
= get_imsm_dev(super
, inst
);
7948 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
7950 if (slot
> map
->num_members
) {
7951 pr_err("imsm: imsm_disk_slot_to_ord %d out of range 0..%d\n",
7952 slot
, map
->num_members
- 1);
7959 return get_imsm_ord_tbl_ent(dev
, slot
, MAP_0
);
7962 static void imsm_set_disk(struct active_array
*a
, int n
, int state
)
7964 int inst
= a
->info
.container_member
;
7965 struct intel_super
*super
= a
->container
->sb
;
7966 struct imsm_dev
*dev
= get_imsm_dev(super
, inst
);
7967 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
7968 struct imsm_disk
*disk
;
7970 int recovery_not_finished
= 0;
7975 ord
= imsm_disk_slot_to_ord(a
, n
);
7979 dprintf("imsm: set_disk %d:%x\n", n
, state
);
7980 disk
= get_imsm_disk(super
, ord_to_idx(ord
));
7982 /* check for new failures */
7983 if (state
& DS_FAULTY
) {
7984 if (mark_failure(super
, dev
, disk
, ord_to_idx(ord
)))
7985 super
->updates_pending
++;
7988 /* check if in_sync */
7989 if (state
& DS_INSYNC
&& ord
& IMSM_ORD_REBUILD
&& is_rebuilding(dev
)) {
7990 struct imsm_map
*migr_map
= get_imsm_map(dev
, MAP_1
);
7992 set_imsm_ord_tbl_ent(migr_map
, n
, ord_to_idx(ord
));
7993 super
->updates_pending
++;
7996 failed
= imsm_count_failed(super
, dev
, MAP_0
);
7997 map_state
= imsm_check_degraded(super
, dev
, failed
, MAP_0
);
7999 /* check if recovery complete, newly degraded, or failed */
8000 dprintf("imsm: Detected transition to state ");
8001 switch (map_state
) {
8002 case IMSM_T_STATE_NORMAL
: /* transition to normal state */
8003 dprintf("normal: ");
8004 if (is_rebuilding(dev
)) {
8005 dprintf_cont("while rebuilding");
8006 /* check if recovery is really finished */
8007 for (mdi
= a
->info
.devs
; mdi
; mdi
= mdi
->next
)
8008 if (mdi
->recovery_start
!= MaxSector
) {
8009 recovery_not_finished
= 1;
8012 if (recovery_not_finished
) {
8014 dprintf("Rebuild has not finished yet, state not changed");
8015 if (a
->last_checkpoint
< mdi
->recovery_start
) {
8016 a
->last_checkpoint
= mdi
->recovery_start
;
8017 super
->updates_pending
++;
8021 end_migration(dev
, super
, map_state
);
8022 map
= get_imsm_map(dev
, MAP_0
);
8023 map
->failed_disk_num
= ~0;
8024 super
->updates_pending
++;
8025 a
->last_checkpoint
= 0;
8028 if (is_gen_migration(dev
)) {
8029 dprintf_cont("while general migration");
8030 if (a
->last_checkpoint
>= a
->info
.component_size
)
8031 end_migration(dev
, super
, map_state
);
8033 map
->map_state
= map_state
;
8034 map
= get_imsm_map(dev
, MAP_0
);
8035 map
->failed_disk_num
= ~0;
8036 super
->updates_pending
++;
8040 case IMSM_T_STATE_DEGRADED
: /* transition to degraded state */
8041 dprintf_cont("degraded: ");
8042 if (map
->map_state
!= map_state
&& !dev
->vol
.migr_state
) {
8043 dprintf_cont("mark degraded");
8044 map
->map_state
= map_state
;
8045 super
->updates_pending
++;
8046 a
->last_checkpoint
= 0;
8049 if (is_rebuilding(dev
)) {
8050 dprintf_cont("while rebuilding.");
8051 if (map
->map_state
!= map_state
) {
8052 dprintf_cont(" Map state change");
8053 end_migration(dev
, super
, map_state
);
8054 super
->updates_pending
++;
8058 if (is_gen_migration(dev
)) {
8059 dprintf_cont("while general migration");
8060 if (a
->last_checkpoint
>= a
->info
.component_size
)
8061 end_migration(dev
, super
, map_state
);
8063 map
->map_state
= map_state
;
8064 manage_second_map(super
, dev
);
8066 super
->updates_pending
++;
8069 if (is_initializing(dev
)) {
8070 dprintf_cont("while initialization.");
8071 map
->map_state
= map_state
;
8072 super
->updates_pending
++;
8076 case IMSM_T_STATE_FAILED
: /* transition to failed state */
8077 dprintf_cont("failed: ");
8078 if (is_gen_migration(dev
)) {
8079 dprintf_cont("while general migration");
8080 map
->map_state
= map_state
;
8081 super
->updates_pending
++;
8084 if (map
->map_state
!= map_state
) {
8085 dprintf_cont("mark failed");
8086 end_migration(dev
, super
, map_state
);
8087 super
->updates_pending
++;
8088 a
->last_checkpoint
= 0;
8093 dprintf_cont("state %i\n", map_state
);
8098 static int store_imsm_mpb(int fd
, struct imsm_super
*mpb
)
8101 __u32 mpb_size
= __le32_to_cpu(mpb
->mpb_size
);
8102 unsigned long long dsize
;
8103 unsigned long long sectors
;
8104 unsigned int sector_size
;
8106 get_dev_sector_size(fd
, NULL
, §or_size
);
8107 get_dev_size(fd
, NULL
, &dsize
);
8109 if (mpb_size
> sector_size
) {
8110 /* -1 to account for anchor */
8111 sectors
= mpb_sectors(mpb
, sector_size
) - 1;
8113 /* write the extended mpb to the sectors preceeding the anchor */
8114 if (lseek64(fd
, dsize
- (sector_size
* (2 + sectors
)),
8118 if ((unsigned long long)write(fd
, buf
+ sector_size
,
8119 sector_size
* sectors
) != sector_size
* sectors
)
8123 /* first block is stored on second to last sector of the disk */
8124 if (lseek64(fd
, dsize
- (sector_size
* 2), SEEK_SET
) < 0)
8127 if ((unsigned int)write(fd
, buf
, sector_size
) != sector_size
)
8133 static void imsm_sync_metadata(struct supertype
*container
)
8135 struct intel_super
*super
= container
->sb
;
8137 dprintf("sync metadata: %d\n", super
->updates_pending
);
8138 if (!super
->updates_pending
)
8141 write_super_imsm(container
, 0);
8143 super
->updates_pending
= 0;
8146 static struct dl
*imsm_readd(struct intel_super
*super
, int idx
, struct active_array
*a
)
8148 struct imsm_dev
*dev
= get_imsm_dev(super
, a
->info
.container_member
);
8149 int i
= get_imsm_disk_idx(dev
, idx
, MAP_X
);
8152 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
8156 if (dl
&& is_failed(&dl
->disk
))
8160 dprintf("found %x:%x\n", dl
->major
, dl
->minor
);
8165 static struct dl
*imsm_add_spare(struct intel_super
*super
, int slot
,
8166 struct active_array
*a
, int activate_new
,
8167 struct mdinfo
*additional_test_list
)
8169 struct imsm_dev
*dev
= get_imsm_dev(super
, a
->info
.container_member
);
8170 int idx
= get_imsm_disk_idx(dev
, slot
, MAP_X
);
8171 struct imsm_super
*mpb
= super
->anchor
;
8172 struct imsm_map
*map
;
8173 unsigned long long pos
;
8178 __u32 array_start
= 0;
8179 __u32 array_end
= 0;
8181 struct mdinfo
*test_list
;
8183 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
8184 /* If in this array, skip */
8185 for (d
= a
->info
.devs
; d
; d
= d
->next
)
8186 if (d
->state_fd
>= 0 &&
8187 d
->disk
.major
== dl
->major
&&
8188 d
->disk
.minor
== dl
->minor
) {
8189 dprintf("%x:%x already in array\n",
8190 dl
->major
, dl
->minor
);
8195 test_list
= additional_test_list
;
8197 if (test_list
->disk
.major
== dl
->major
&&
8198 test_list
->disk
.minor
== dl
->minor
) {
8199 dprintf("%x:%x already in additional test list\n",
8200 dl
->major
, dl
->minor
);
8203 test_list
= test_list
->next
;
8208 /* skip in use or failed drives */
8209 if (is_failed(&dl
->disk
) || idx
== dl
->index
||
8211 dprintf("%x:%x status (failed: %d index: %d)\n",
8212 dl
->major
, dl
->minor
, is_failed(&dl
->disk
), idx
);
8216 /* skip pure spares when we are looking for partially
8217 * assimilated drives
8219 if (dl
->index
== -1 && !activate_new
)
8222 /* Does this unused device have the requisite free space?
8223 * It needs to be able to cover all member volumes
8225 ex
= get_extents(super
, dl
);
8227 dprintf("cannot get extents\n");
8230 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
8231 dev
= get_imsm_dev(super
, i
);
8232 map
= get_imsm_map(dev
, MAP_0
);
8234 /* check if this disk is already a member of
8237 if (get_imsm_disk_slot(map
, dl
->index
) >= 0)
8243 array_start
= pba_of_lba0(map
);
8244 array_end
= array_start
+
8245 blocks_per_member(map
) - 1;
8248 /* check that we can start at pba_of_lba0 with
8249 * blocks_per_member of space
8251 if (array_start
>= pos
&& array_end
< ex
[j
].start
) {
8255 pos
= ex
[j
].start
+ ex
[j
].size
;
8257 } while (ex
[j
-1].size
);
8264 if (i
< mpb
->num_raid_devs
) {
8265 dprintf("%x:%x does not have %u to %u available\n",
8266 dl
->major
, dl
->minor
, array_start
, array_end
);
8276 static int imsm_rebuild_allowed(struct supertype
*cont
, int dev_idx
, int failed
)
8278 struct imsm_dev
*dev2
;
8279 struct imsm_map
*map
;
8285 dev2
= get_imsm_dev(cont
->sb
, dev_idx
);
8287 state
= imsm_check_degraded(cont
->sb
, dev2
, failed
, MAP_0
);
8288 if (state
== IMSM_T_STATE_FAILED
) {
8289 map
= get_imsm_map(dev2
, MAP_0
);
8292 for (slot
= 0; slot
< map
->num_members
; slot
++) {
8294 * Check if failed disks are deleted from intel
8295 * disk list or are marked to be deleted
8297 idx
= get_imsm_disk_idx(dev2
, slot
, MAP_X
);
8298 idisk
= get_imsm_dl_disk(cont
->sb
, idx
);
8300 * Do not rebuild the array if failed disks
8301 * from failed sub-array are not removed from
8305 is_failed(&idisk
->disk
) &&
8306 (idisk
->action
!= DISK_REMOVE
))
8314 static struct mdinfo
*imsm_activate_spare(struct active_array
*a
,
8315 struct metadata_update
**updates
)
8318 * Find a device with unused free space and use it to replace a
8319 * failed/vacant region in an array. We replace failed regions one a
8320 * array at a time. The result is that a new spare disk will be added
8321 * to the first failed array and after the monitor has finished
8322 * propagating failures the remainder will be consumed.
8324 * FIXME add a capability for mdmon to request spares from another
8328 struct intel_super
*super
= a
->container
->sb
;
8329 int inst
= a
->info
.container_member
;
8330 struct imsm_dev
*dev
= get_imsm_dev(super
, inst
);
8331 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
8332 int failed
= a
->info
.array
.raid_disks
;
8333 struct mdinfo
*rv
= NULL
;
8336 struct metadata_update
*mu
;
8338 struct imsm_update_activate_spare
*u
;
8343 for (d
= a
->info
.devs
; d
; d
= d
->next
) {
8344 if ((d
->curr_state
& DS_FAULTY
) &&
8346 /* wait for Removal to happen */
8348 if (d
->state_fd
>= 0)
8352 dprintf("imsm: activate spare: inst=%d failed=%d (%d) level=%d\n",
8353 inst
, failed
, a
->info
.array
.raid_disks
, a
->info
.array
.level
);
8355 if (imsm_reshape_blocks_arrays_changes(super
))
8358 /* Cannot activate another spare if rebuild is in progress already
8360 if (is_rebuilding(dev
)) {
8361 dprintf("imsm: No spare activation allowed. Rebuild in progress already.\n");
8365 if (a
->info
.array
.level
== 4)
8366 /* No repair for takeovered array
8367 * imsm doesn't support raid4
8371 if (imsm_check_degraded(super
, dev
, failed
, MAP_0
) !=
8372 IMSM_T_STATE_DEGRADED
)
8375 if (get_imsm_map(dev
, MAP_0
)->map_state
== IMSM_T_STATE_UNINITIALIZED
) {
8376 dprintf("imsm: No spare activation allowed. Volume is not initialized.\n");
8381 * If there are any failed disks check state of the other volume.
8382 * Block rebuild if the another one is failed until failed disks
8383 * are removed from container.
8386 dprintf("found failed disks in %.*s, check if there anotherfailed sub-array.\n",
8387 MAX_RAID_SERIAL_LEN
, dev
->volume
);
8388 /* check if states of the other volumes allow for rebuild */
8389 for (i
= 0; i
< super
->anchor
->num_raid_devs
; i
++) {
8391 allowed
= imsm_rebuild_allowed(a
->container
,
8399 /* For each slot, if it is not working, find a spare */
8400 for (i
= 0; i
< a
->info
.array
.raid_disks
; i
++) {
8401 for (d
= a
->info
.devs
; d
; d
= d
->next
)
8402 if (d
->disk
.raid_disk
== i
)
8404 dprintf("found %d: %p %x\n", i
, d
, d
?d
->curr_state
:0);
8405 if (d
&& (d
->state_fd
>= 0))
8409 * OK, this device needs recovery. Try to re-add the
8410 * previous occupant of this slot, if this fails see if
8411 * we can continue the assimilation of a spare that was
8412 * partially assimilated, finally try to activate a new
8415 dl
= imsm_readd(super
, i
, a
);
8417 dl
= imsm_add_spare(super
, i
, a
, 0, rv
);
8419 dl
= imsm_add_spare(super
, i
, a
, 1, rv
);
8423 /* found a usable disk with enough space */
8424 di
= xcalloc(1, sizeof(*di
));
8426 /* dl->index will be -1 in the case we are activating a
8427 * pristine spare. imsm_process_update() will create a
8428 * new index in this case. Once a disk is found to be
8429 * failed in all member arrays it is kicked from the
8432 di
->disk
.number
= dl
->index
;
8434 /* (ab)use di->devs to store a pointer to the device
8437 di
->devs
= (struct mdinfo
*) dl
;
8439 di
->disk
.raid_disk
= i
;
8440 di
->disk
.major
= dl
->major
;
8441 di
->disk
.minor
= dl
->minor
;
8443 di
->recovery_start
= 0;
8444 di
->data_offset
= pba_of_lba0(map
);
8445 di
->component_size
= a
->info
.component_size
;
8446 di
->container_member
= inst
;
8447 di
->bb
.supported
= 1;
8448 super
->random
= random32();
8452 dprintf("%x:%x to be %d at %llu\n", dl
->major
, dl
->minor
,
8453 i
, di
->data_offset
);
8457 /* No spares found */
8459 /* Now 'rv' has a list of devices to return.
8460 * Create a metadata_update record to update the
8461 * disk_ord_tbl for the array
8463 mu
= xmalloc(sizeof(*mu
));
8464 mu
->buf
= xcalloc(num_spares
,
8465 sizeof(struct imsm_update_activate_spare
));
8467 mu
->space_list
= NULL
;
8468 mu
->len
= sizeof(struct imsm_update_activate_spare
) * num_spares
;
8469 mu
->next
= *updates
;
8470 u
= (struct imsm_update_activate_spare
*) mu
->buf
;
8472 for (di
= rv
; di
; di
= di
->next
) {
8473 u
->type
= update_activate_spare
;
8474 u
->dl
= (struct dl
*) di
->devs
;
8476 u
->slot
= di
->disk
.raid_disk
;
8487 static int disks_overlap(struct intel_super
*super
, int idx
, struct imsm_update_create_array
*u
)
8489 struct imsm_dev
*dev
= get_imsm_dev(super
, idx
);
8490 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
8491 struct imsm_map
*new_map
= get_imsm_map(&u
->dev
, MAP_0
);
8492 struct disk_info
*inf
= get_disk_info(u
);
8493 struct imsm_disk
*disk
;
8497 for (i
= 0; i
< map
->num_members
; i
++) {
8498 disk
= get_imsm_disk(super
, get_imsm_disk_idx(dev
, i
, MAP_X
));
8499 for (j
= 0; j
< new_map
->num_members
; j
++)
8500 if (serialcmp(disk
->serial
, inf
[j
].serial
) == 0)
8507 static struct dl
*get_disk_super(struct intel_super
*super
, int major
, int minor
)
8511 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
8512 if (dl
->major
== major
&& dl
->minor
== minor
)
8517 static int remove_disk_super(struct intel_super
*super
, int major
, int minor
)
8523 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
8524 if (dl
->major
== major
&& dl
->minor
== minor
) {
8527 prev
->next
= dl
->next
;
8529 super
->disks
= dl
->next
;
8531 __free_imsm_disk(dl
);
8532 dprintf("removed %x:%x\n", major
, minor
);
8540 static void imsm_delete(struct intel_super
*super
, struct dl
**dlp
, unsigned index
);
8542 static int add_remove_disk_update(struct intel_super
*super
)
8544 int check_degraded
= 0;
8547 /* add/remove some spares to/from the metadata/contrainer */
8548 while (super
->disk_mgmt_list
) {
8549 struct dl
*disk_cfg
;
8551 disk_cfg
= super
->disk_mgmt_list
;
8552 super
->disk_mgmt_list
= disk_cfg
->next
;
8553 disk_cfg
->next
= NULL
;
8555 if (disk_cfg
->action
== DISK_ADD
) {
8556 disk_cfg
->next
= super
->disks
;
8557 super
->disks
= disk_cfg
;
8559 dprintf("added %x:%x\n",
8560 disk_cfg
->major
, disk_cfg
->minor
);
8561 } else if (disk_cfg
->action
== DISK_REMOVE
) {
8562 dprintf("Disk remove action processed: %x.%x\n",
8563 disk_cfg
->major
, disk_cfg
->minor
);
8564 disk
= get_disk_super(super
,
8568 /* store action status */
8569 disk
->action
= DISK_REMOVE
;
8570 /* remove spare disks only */
8571 if (disk
->index
== -1) {
8572 remove_disk_super(super
,
8577 /* release allocate disk structure */
8578 __free_imsm_disk(disk_cfg
);
8581 return check_degraded
;
8584 static int apply_reshape_migration_update(struct imsm_update_reshape_migration
*u
,
8585 struct intel_super
*super
,
8588 struct intel_dev
*id
;
8589 void **tofree
= NULL
;
8592 dprintf("(enter)\n");
8593 if (u
->subdev
< 0 || u
->subdev
> 1) {
8594 dprintf("imsm: Error: Wrong subdev: %i\n", u
->subdev
);
8597 if (space_list
== NULL
|| *space_list
== NULL
) {
8598 dprintf("imsm: Error: Memory is not allocated\n");
8602 for (id
= super
->devlist
; id
; id
= id
->next
) {
8603 if (id
->index
== (unsigned)u
->subdev
) {
8604 struct imsm_dev
*dev
= get_imsm_dev(super
, u
->subdev
);
8605 struct imsm_map
*map
;
8606 struct imsm_dev
*new_dev
=
8607 (struct imsm_dev
*)*space_list
;
8608 struct imsm_map
*migr_map
= get_imsm_map(dev
, MAP_1
);
8610 struct dl
*new_disk
;
8612 if (new_dev
== NULL
)
8614 *space_list
= **space_list
;
8615 memcpy(new_dev
, dev
, sizeof_imsm_dev(dev
, 0));
8616 map
= get_imsm_map(new_dev
, MAP_0
);
8618 dprintf("imsm: Error: migration in progress");
8622 to_state
= map
->map_state
;
8623 if ((u
->new_level
== 5) && (map
->raid_level
== 0)) {
8625 /* this should not happen */
8626 if (u
->new_disks
[0] < 0) {
8627 map
->failed_disk_num
=
8628 map
->num_members
- 1;
8629 to_state
= IMSM_T_STATE_DEGRADED
;
8631 to_state
= IMSM_T_STATE_NORMAL
;
8633 migrate(new_dev
, super
, to_state
, MIGR_GEN_MIGR
);
8634 if (u
->new_level
> -1)
8635 map
->raid_level
= u
->new_level
;
8636 migr_map
= get_imsm_map(new_dev
, MAP_1
);
8637 if ((u
->new_level
== 5) &&
8638 (migr_map
->raid_level
== 0)) {
8639 int ord
= map
->num_members
- 1;
8640 migr_map
->num_members
--;
8641 if (u
->new_disks
[0] < 0)
8642 ord
|= IMSM_ORD_REBUILD
;
8643 set_imsm_ord_tbl_ent(map
,
8644 map
->num_members
- 1,
8648 tofree
= (void **)dev
;
8650 /* update chunk size
8652 if (u
->new_chunksize
> 0) {
8653 unsigned long long num_data_stripes
;
8655 imsm_num_data_members(dev
, MAP_0
);
8657 if (used_disks
== 0)
8660 map
->blocks_per_strip
=
8661 __cpu_to_le16(u
->new_chunksize
* 2);
8663 (join_u32(dev
->size_low
, dev
->size_high
)
8665 num_data_stripes
/= map
->blocks_per_strip
;
8666 num_data_stripes
/= map
->num_domains
;
8667 set_num_data_stripes(map
, num_data_stripes
);
8672 if (u
->new_level
!= 5 || migr_map
->raid_level
!= 0 ||
8673 migr_map
->raid_level
== map
->raid_level
)
8676 if (u
->new_disks
[0] >= 0) {
8679 new_disk
= get_disk_super(super
,
8680 major(u
->new_disks
[0]),
8681 minor(u
->new_disks
[0]));
8682 dprintf("imsm: new disk for reshape is: %i:%i (%p, index = %i)\n",
8683 major(u
->new_disks
[0]),
8684 minor(u
->new_disks
[0]),
8685 new_disk
, new_disk
->index
);
8686 if (new_disk
== NULL
)
8687 goto error_disk_add
;
8689 new_disk
->index
= map
->num_members
- 1;
8690 /* slot to fill in autolayout
8692 new_disk
->raiddisk
= new_disk
->index
;
8693 new_disk
->disk
.status
|= CONFIGURED_DISK
;
8694 new_disk
->disk
.status
&= ~SPARE_DISK
;
8696 goto error_disk_add
;
8699 *tofree
= *space_list
;
8700 /* calculate new size
8702 imsm_set_array_size(new_dev
, -1);
8709 *space_list
= tofree
;
8713 dprintf("Error: imsm: Cannot find disk.\n");
8717 static int apply_size_change_update(struct imsm_update_size_change
*u
,
8718 struct intel_super
*super
)
8720 struct intel_dev
*id
;
8723 dprintf("(enter)\n");
8724 if (u
->subdev
< 0 || u
->subdev
> 1) {
8725 dprintf("imsm: Error: Wrong subdev: %i\n", u
->subdev
);
8729 for (id
= super
->devlist
; id
; id
= id
->next
) {
8730 if (id
->index
== (unsigned)u
->subdev
) {
8731 struct imsm_dev
*dev
= get_imsm_dev(super
, u
->subdev
);
8732 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
8733 int used_disks
= imsm_num_data_members(dev
, MAP_0
);
8734 unsigned long long blocks_per_member
;
8735 unsigned long long num_data_stripes
;
8737 /* calculate new size
8739 blocks_per_member
= u
->new_size
/ used_disks
;
8740 num_data_stripes
= blocks_per_member
/
8741 map
->blocks_per_strip
;
8742 num_data_stripes
/= map
->num_domains
;
8743 dprintf("(size: %llu, blocks per member: %llu, num_data_stipes: %llu)\n",
8744 u
->new_size
, blocks_per_member
,
8746 set_blocks_per_member(map
, blocks_per_member
);
8747 set_num_data_stripes(map
, num_data_stripes
);
8748 imsm_set_array_size(dev
, u
->new_size
);
8758 static int apply_update_activate_spare(struct imsm_update_activate_spare
*u
,
8759 struct intel_super
*super
,
8760 struct active_array
*active_array
)
8762 struct imsm_super
*mpb
= super
->anchor
;
8763 struct imsm_dev
*dev
= get_imsm_dev(super
, u
->array
);
8764 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
8765 struct imsm_map
*migr_map
;
8766 struct active_array
*a
;
8767 struct imsm_disk
*disk
;
8774 int second_map_created
= 0;
8776 for (; u
; u
= u
->next
) {
8777 victim
= get_imsm_disk_idx(dev
, u
->slot
, MAP_X
);
8782 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
8787 pr_err("error: imsm_activate_spare passed an unknown disk (index: %d)\n",
8792 /* count failures (excluding rebuilds and the victim)
8793 * to determine map[0] state
8796 for (i
= 0; i
< map
->num_members
; i
++) {
8799 disk
= get_imsm_disk(super
,
8800 get_imsm_disk_idx(dev
, i
, MAP_X
));
8801 if (!disk
|| is_failed(disk
))
8805 /* adding a pristine spare, assign a new index */
8806 if (dl
->index
< 0) {
8807 dl
->index
= super
->anchor
->num_disks
;
8808 super
->anchor
->num_disks
++;
8811 disk
->status
|= CONFIGURED_DISK
;
8812 disk
->status
&= ~SPARE_DISK
;
8815 to_state
= imsm_check_degraded(super
, dev
, failed
, MAP_0
);
8816 if (!second_map_created
) {
8817 second_map_created
= 1;
8818 map
->map_state
= IMSM_T_STATE_DEGRADED
;
8819 migrate(dev
, super
, to_state
, MIGR_REBUILD
);
8821 map
->map_state
= to_state
;
8822 migr_map
= get_imsm_map(dev
, MAP_1
);
8823 set_imsm_ord_tbl_ent(map
, u
->slot
, dl
->index
);
8824 set_imsm_ord_tbl_ent(migr_map
, u
->slot
,
8825 dl
->index
| IMSM_ORD_REBUILD
);
8827 /* update the family_num to mark a new container
8828 * generation, being careful to record the existing
8829 * family_num in orig_family_num to clean up after
8830 * earlier mdadm versions that neglected to set it.
8832 if (mpb
->orig_family_num
== 0)
8833 mpb
->orig_family_num
= mpb
->family_num
;
8834 mpb
->family_num
+= super
->random
;
8836 /* count arrays using the victim in the metadata */
8838 for (a
= active_array
; a
; a
= a
->next
) {
8839 dev
= get_imsm_dev(super
, a
->info
.container_member
);
8840 map
= get_imsm_map(dev
, MAP_0
);
8842 if (get_imsm_disk_slot(map
, victim
) >= 0)
8846 /* delete the victim if it is no longer being
8852 /* We know that 'manager' isn't touching anything,
8853 * so it is safe to delete
8855 for (dlp
= &super
->disks
; *dlp
; dlp
= &(*dlp
)->next
)
8856 if ((*dlp
)->index
== victim
)
8859 /* victim may be on the missing list */
8861 for (dlp
= &super
->missing
; *dlp
;
8862 dlp
= &(*dlp
)->next
)
8863 if ((*dlp
)->index
== victim
)
8865 imsm_delete(super
, dlp
, victim
);
8872 static int apply_reshape_container_disks_update(struct imsm_update_reshape
*u
,
8873 struct intel_super
*super
,
8876 struct dl
*new_disk
;
8877 struct intel_dev
*id
;
8879 int delta_disks
= u
->new_raid_disks
- u
->old_raid_disks
;
8880 int disk_count
= u
->old_raid_disks
;
8881 void **tofree
= NULL
;
8882 int devices_to_reshape
= 1;
8883 struct imsm_super
*mpb
= super
->anchor
;
8885 unsigned int dev_id
;
8887 dprintf("(enter)\n");
8889 /* enable spares to use in array */
8890 for (i
= 0; i
< delta_disks
; i
++) {
8891 new_disk
= get_disk_super(super
,
8892 major(u
->new_disks
[i
]),
8893 minor(u
->new_disks
[i
]));
8894 dprintf("imsm: new disk for reshape is: %i:%i (%p, index = %i)\n",
8895 major(u
->new_disks
[i
]), minor(u
->new_disks
[i
]),
8896 new_disk
, new_disk
->index
);
8897 if (new_disk
== NULL
||
8898 (new_disk
->index
>= 0 &&
8899 new_disk
->index
< u
->old_raid_disks
))
8900 goto update_reshape_exit
;
8901 new_disk
->index
= disk_count
++;
8902 /* slot to fill in autolayout
8904 new_disk
->raiddisk
= new_disk
->index
;
8905 new_disk
->disk
.status
|=
8907 new_disk
->disk
.status
&= ~SPARE_DISK
;
8910 dprintf("imsm: volume set mpb->num_raid_devs = %i\n",
8911 mpb
->num_raid_devs
);
8912 /* manage changes in volume
8914 for (dev_id
= 0; dev_id
< mpb
->num_raid_devs
; dev_id
++) {
8915 void **sp
= *space_list
;
8916 struct imsm_dev
*newdev
;
8917 struct imsm_map
*newmap
, *oldmap
;
8919 for (id
= super
->devlist
; id
; id
= id
->next
) {
8920 if (id
->index
== dev_id
)
8929 /* Copy the dev, but not (all of) the map */
8930 memcpy(newdev
, id
->dev
, sizeof(*newdev
));
8931 oldmap
= get_imsm_map(id
->dev
, MAP_0
);
8932 newmap
= get_imsm_map(newdev
, MAP_0
);
8933 /* Copy the current map */
8934 memcpy(newmap
, oldmap
, sizeof_imsm_map(oldmap
));
8935 /* update one device only
8937 if (devices_to_reshape
) {
8938 dprintf("imsm: modifying subdev: %i\n",
8940 devices_to_reshape
--;
8941 newdev
->vol
.migr_state
= 1;
8942 newdev
->vol
.curr_migr_unit
= 0;
8943 set_migr_type(newdev
, MIGR_GEN_MIGR
);
8944 newmap
->num_members
= u
->new_raid_disks
;
8945 for (i
= 0; i
< delta_disks
; i
++) {
8946 set_imsm_ord_tbl_ent(newmap
,
8947 u
->old_raid_disks
+ i
,
8948 u
->old_raid_disks
+ i
);
8950 /* New map is correct, now need to save old map
8952 newmap
= get_imsm_map(newdev
, MAP_1
);
8953 memcpy(newmap
, oldmap
, sizeof_imsm_map(oldmap
));
8955 imsm_set_array_size(newdev
, -1);
8958 sp
= (void **)id
->dev
;
8963 /* Clear migration record */
8964 memset(super
->migr_rec
, 0, sizeof(struct migr_record
));
8967 *space_list
= tofree
;
8970 update_reshape_exit
:
8975 static int apply_takeover_update(struct imsm_update_takeover
*u
,
8976 struct intel_super
*super
,
8979 struct imsm_dev
*dev
= NULL
;
8980 struct intel_dev
*dv
;
8981 struct imsm_dev
*dev_new
;
8982 struct imsm_map
*map
;
8986 for (dv
= super
->devlist
; dv
; dv
= dv
->next
)
8987 if (dv
->index
== (unsigned int)u
->subarray
) {
8995 map
= get_imsm_map(dev
, MAP_0
);
8997 if (u
->direction
== R10_TO_R0
) {
8998 unsigned long long num_data_stripes
;
9000 map
->num_domains
= 1;
9001 num_data_stripes
= blocks_per_member(map
);
9002 num_data_stripes
/= map
->blocks_per_strip
;
9003 num_data_stripes
/= map
->num_domains
;
9004 set_num_data_stripes(map
, num_data_stripes
);
9006 /* Number of failed disks must be half of initial disk number */
9007 if (imsm_count_failed(super
, dev
, MAP_0
) !=
9008 (map
->num_members
/ 2))
9011 /* iterate through devices to mark removed disks as spare */
9012 for (dm
= super
->disks
; dm
; dm
= dm
->next
) {
9013 if (dm
->disk
.status
& FAILED_DISK
) {
9014 int idx
= dm
->index
;
9015 /* update indexes on the disk list */
9016 /* FIXME this loop-with-the-loop looks wrong, I'm not convinced
9017 the index values will end up being correct.... NB */
9018 for (du
= super
->disks
; du
; du
= du
->next
)
9019 if (du
->index
> idx
)
9021 /* mark as spare disk */
9026 map
->num_members
= map
->num_members
/ 2;
9027 map
->map_state
= IMSM_T_STATE_NORMAL
;
9028 map
->num_domains
= 1;
9029 map
->raid_level
= 0;
9030 map
->failed_disk_num
= -1;
9033 if (u
->direction
== R0_TO_R10
) {
9035 /* update slots in current disk list */
9036 for (dm
= super
->disks
; dm
; dm
= dm
->next
) {
9040 /* create new *missing* disks */
9041 for (i
= 0; i
< map
->num_members
; i
++) {
9042 space
= *space_list
;
9045 *space_list
= *space
;
9047 memcpy(du
, super
->disks
, sizeof(*du
));
9051 du
->index
= (i
* 2) + 1;
9052 sprintf((char *)du
->disk
.serial
,
9053 " MISSING_%d", du
->index
);
9054 sprintf((char *)du
->serial
,
9055 "MISSING_%d", du
->index
);
9056 du
->next
= super
->missing
;
9057 super
->missing
= du
;
9059 /* create new dev and map */
9060 space
= *space_list
;
9063 *space_list
= *space
;
9064 dev_new
= (void *)space
;
9065 memcpy(dev_new
, dev
, sizeof(*dev
));
9066 /* update new map */
9067 map
= get_imsm_map(dev_new
, MAP_0
);
9068 map
->num_members
= map
->num_members
* 2;
9069 map
->map_state
= IMSM_T_STATE_DEGRADED
;
9070 map
->num_domains
= 2;
9071 map
->raid_level
= 1;
9072 /* replace dev<->dev_new */
9075 /* update disk order table */
9076 for (du
= super
->disks
; du
; du
= du
->next
)
9078 set_imsm_ord_tbl_ent(map
, du
->index
, du
->index
);
9079 for (du
= super
->missing
; du
; du
= du
->next
)
9080 if (du
->index
>= 0) {
9081 set_imsm_ord_tbl_ent(map
, du
->index
, du
->index
);
9082 mark_missing(super
, dv
->dev
, &du
->disk
, du
->index
);
9088 static void imsm_process_update(struct supertype
*st
,
9089 struct metadata_update
*update
)
9092 * crack open the metadata_update envelope to find the update record
9093 * update can be one of:
9094 * update_reshape_container_disks - all the arrays in the container
9095 * are being reshaped to have more devices. We need to mark
9096 * the arrays for general migration and convert selected spares
9097 * into active devices.
9098 * update_activate_spare - a spare device has replaced a failed
9099 * device in an array, update the disk_ord_tbl. If this disk is
9100 * present in all member arrays then also clear the SPARE_DISK
9102 * update_create_array
9104 * update_rename_array
9105 * update_add_remove_disk
9107 struct intel_super
*super
= st
->sb
;
9108 struct imsm_super
*mpb
;
9109 enum imsm_update_type type
= *(enum imsm_update_type
*) update
->buf
;
9111 /* update requires a larger buf but the allocation failed */
9112 if (super
->next_len
&& !super
->next_buf
) {
9113 super
->next_len
= 0;
9117 if (super
->next_buf
) {
9118 memcpy(super
->next_buf
, super
->buf
, super
->len
);
9120 super
->len
= super
->next_len
;
9121 super
->buf
= super
->next_buf
;
9123 super
->next_len
= 0;
9124 super
->next_buf
= NULL
;
9127 mpb
= super
->anchor
;
9130 case update_general_migration_checkpoint
: {
9131 struct intel_dev
*id
;
9132 struct imsm_update_general_migration_checkpoint
*u
=
9133 (void *)update
->buf
;
9135 dprintf("called for update_general_migration_checkpoint\n");
9137 /* find device under general migration */
9138 for (id
= super
->devlist
; id
; id
= id
->next
) {
9139 if (is_gen_migration(id
->dev
)) {
9140 id
->dev
->vol
.curr_migr_unit
=
9141 __cpu_to_le32(u
->curr_migr_unit
);
9142 super
->updates_pending
++;
9147 case update_takeover
: {
9148 struct imsm_update_takeover
*u
= (void *)update
->buf
;
9149 if (apply_takeover_update(u
, super
, &update
->space_list
)) {
9150 imsm_update_version_info(super
);
9151 super
->updates_pending
++;
9156 case update_reshape_container_disks
: {
9157 struct imsm_update_reshape
*u
= (void *)update
->buf
;
9158 if (apply_reshape_container_disks_update(
9159 u
, super
, &update
->space_list
))
9160 super
->updates_pending
++;
9163 case update_reshape_migration
: {
9164 struct imsm_update_reshape_migration
*u
= (void *)update
->buf
;
9165 if (apply_reshape_migration_update(
9166 u
, super
, &update
->space_list
))
9167 super
->updates_pending
++;
9170 case update_size_change
: {
9171 struct imsm_update_size_change
*u
= (void *)update
->buf
;
9172 if (apply_size_change_update(u
, super
))
9173 super
->updates_pending
++;
9176 case update_activate_spare
: {
9177 struct imsm_update_activate_spare
*u
= (void *) update
->buf
;
9178 if (apply_update_activate_spare(u
, super
, st
->arrays
))
9179 super
->updates_pending
++;
9182 case update_create_array
: {
9183 /* someone wants to create a new array, we need to be aware of
9184 * a few races/collisions:
9185 * 1/ 'Create' called by two separate instances of mdadm
9186 * 2/ 'Create' versus 'activate_spare': mdadm has chosen
9187 * devices that have since been assimilated via
9189 * In the event this update can not be carried out mdadm will
9190 * (FIX ME) notice that its update did not take hold.
9192 struct imsm_update_create_array
*u
= (void *) update
->buf
;
9193 struct intel_dev
*dv
;
9194 struct imsm_dev
*dev
;
9195 struct imsm_map
*map
, *new_map
;
9196 unsigned long long start
, end
;
9197 unsigned long long new_start
, new_end
;
9199 struct disk_info
*inf
;
9202 /* handle racing creates: first come first serve */
9203 if (u
->dev_idx
< mpb
->num_raid_devs
) {
9204 dprintf("subarray %d already defined\n", u
->dev_idx
);
9208 /* check update is next in sequence */
9209 if (u
->dev_idx
!= mpb
->num_raid_devs
) {
9210 dprintf("can not create array %d expected index %d\n",
9211 u
->dev_idx
, mpb
->num_raid_devs
);
9215 new_map
= get_imsm_map(&u
->dev
, MAP_0
);
9216 new_start
= pba_of_lba0(new_map
);
9217 new_end
= new_start
+ blocks_per_member(new_map
);
9218 inf
= get_disk_info(u
);
9220 /* handle activate_spare versus create race:
9221 * check to make sure that overlapping arrays do not include
9224 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
9225 dev
= get_imsm_dev(super
, i
);
9226 map
= get_imsm_map(dev
, MAP_0
);
9227 start
= pba_of_lba0(map
);
9228 end
= start
+ blocks_per_member(map
);
9229 if ((new_start
>= start
&& new_start
<= end
) ||
9230 (start
>= new_start
&& start
<= new_end
))
9235 if (disks_overlap(super
, i
, u
)) {
9236 dprintf("arrays overlap\n");
9241 /* check that prepare update was successful */
9242 if (!update
->space
) {
9243 dprintf("prepare update failed\n");
9247 /* check that all disks are still active before committing
9248 * changes. FIXME: could we instead handle this by creating a
9249 * degraded array? That's probably not what the user expects,
9250 * so better to drop this update on the floor.
9252 for (i
= 0; i
< new_map
->num_members
; i
++) {
9253 dl
= serial_to_dl(inf
[i
].serial
, super
);
9255 dprintf("disk disappeared\n");
9260 super
->updates_pending
++;
9262 /* convert spares to members and fixup ord_tbl */
9263 for (i
= 0; i
< new_map
->num_members
; i
++) {
9264 dl
= serial_to_dl(inf
[i
].serial
, super
);
9265 if (dl
->index
== -1) {
9266 dl
->index
= mpb
->num_disks
;
9268 dl
->disk
.status
|= CONFIGURED_DISK
;
9269 dl
->disk
.status
&= ~SPARE_DISK
;
9271 set_imsm_ord_tbl_ent(new_map
, i
, dl
->index
);
9276 update
->space
= NULL
;
9277 imsm_copy_dev(dev
, &u
->dev
);
9278 dv
->index
= u
->dev_idx
;
9279 dv
->next
= super
->devlist
;
9280 super
->devlist
= dv
;
9281 mpb
->num_raid_devs
++;
9283 imsm_update_version_info(super
);
9286 /* mdmon knows how to release update->space, but not
9287 * ((struct intel_dev *) update->space)->dev
9289 if (update
->space
) {
9295 case update_kill_array
: {
9296 struct imsm_update_kill_array
*u
= (void *) update
->buf
;
9297 int victim
= u
->dev_idx
;
9298 struct active_array
*a
;
9299 struct intel_dev
**dp
;
9300 struct imsm_dev
*dev
;
9302 /* sanity check that we are not affecting the uuid of
9303 * active arrays, or deleting an active array
9305 * FIXME when immutable ids are available, but note that
9306 * we'll also need to fixup the invalidated/active
9307 * subarray indexes in mdstat
9309 for (a
= st
->arrays
; a
; a
= a
->next
)
9310 if (a
->info
.container_member
>= victim
)
9312 /* by definition if mdmon is running at least one array
9313 * is active in the container, so checking
9314 * mpb->num_raid_devs is just extra paranoia
9316 dev
= get_imsm_dev(super
, victim
);
9317 if (a
|| !dev
|| mpb
->num_raid_devs
== 1) {
9318 dprintf("failed to delete subarray-%d\n", victim
);
9322 for (dp
= &super
->devlist
; *dp
;)
9323 if ((*dp
)->index
== (unsigned)super
->current_vol
) {
9326 if ((*dp
)->index
> (unsigned)victim
)
9330 mpb
->num_raid_devs
--;
9331 super
->updates_pending
++;
9334 case update_rename_array
: {
9335 struct imsm_update_rename_array
*u
= (void *) update
->buf
;
9336 char name
[MAX_RAID_SERIAL_LEN
+1];
9337 int target
= u
->dev_idx
;
9338 struct active_array
*a
;
9339 struct imsm_dev
*dev
;
9341 /* sanity check that we are not affecting the uuid of
9344 snprintf(name
, MAX_RAID_SERIAL_LEN
, "%s", (char *) u
->name
);
9345 name
[MAX_RAID_SERIAL_LEN
] = '\0';
9346 for (a
= st
->arrays
; a
; a
= a
->next
)
9347 if (a
->info
.container_member
== target
)
9349 dev
= get_imsm_dev(super
, u
->dev_idx
);
9350 if (a
|| !dev
|| !check_name(super
, name
, 1)) {
9351 dprintf("failed to rename subarray-%d\n", target
);
9355 snprintf((char *) dev
->volume
, MAX_RAID_SERIAL_LEN
, "%s", name
);
9356 super
->updates_pending
++;
9359 case update_add_remove_disk
: {
9360 /* we may be able to repair some arrays if disks are
9361 * being added, check the status of add_remove_disk
9362 * if discs has been added.
9364 if (add_remove_disk_update(super
)) {
9365 struct active_array
*a
;
9367 super
->updates_pending
++;
9368 for (a
= st
->arrays
; a
; a
= a
->next
)
9369 a
->check_degraded
= 1;
9373 case update_prealloc_badblocks_mem
:
9376 pr_err("error: unsuported process update type:(type: %d)\n", type
);
9380 static struct mdinfo
*get_spares_for_grow(struct supertype
*st
);
9382 static int imsm_prepare_update(struct supertype
*st
,
9383 struct metadata_update
*update
)
9386 * Allocate space to hold new disk entries, raid-device entries or a new
9387 * mpb if necessary. The manager synchronously waits for updates to
9388 * complete in the monitor, so new mpb buffers allocated here can be
9389 * integrated by the monitor thread without worrying about live pointers
9390 * in the manager thread.
9392 enum imsm_update_type type
;
9393 struct intel_super
*super
= st
->sb
;
9394 unsigned int sector_size
= super
->sector_size
;
9395 struct imsm_super
*mpb
= super
->anchor
;
9399 if (update
->len
< (int)sizeof(type
))
9402 type
= *(enum imsm_update_type
*) update
->buf
;
9405 case update_general_migration_checkpoint
:
9406 if (update
->len
< (int)sizeof(struct imsm_update_general_migration_checkpoint
))
9408 dprintf("called for update_general_migration_checkpoint\n");
9410 case update_takeover
: {
9411 struct imsm_update_takeover
*u
= (void *)update
->buf
;
9412 if (update
->len
< (int)sizeof(*u
))
9414 if (u
->direction
== R0_TO_R10
) {
9415 void **tail
= (void **)&update
->space_list
;
9416 struct imsm_dev
*dev
= get_imsm_dev(super
, u
->subarray
);
9417 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
9418 int num_members
= map
->num_members
;
9421 /* allocate memory for added disks */
9422 for (i
= 0; i
< num_members
; i
++) {
9423 size
= sizeof(struct dl
);
9424 space
= xmalloc(size
);
9429 /* allocate memory for new device */
9430 size
= sizeof_imsm_dev(super
->devlist
->dev
, 0) +
9431 (num_members
* sizeof(__u32
));
9432 space
= xmalloc(size
);
9436 len
= disks_to_mpb_size(num_members
* 2);
9441 case update_reshape_container_disks
: {
9442 /* Every raid device in the container is about to
9443 * gain some more devices, and we will enter a
9445 * So each 'imsm_map' will be bigger, and the imsm_vol
9446 * will now hold 2 of them.
9447 * Thus we need new 'struct imsm_dev' allocations sized
9448 * as sizeof_imsm_dev but with more devices in both maps.
9450 struct imsm_update_reshape
*u
= (void *)update
->buf
;
9451 struct intel_dev
*dl
;
9452 void **space_tail
= (void**)&update
->space_list
;
9454 if (update
->len
< (int)sizeof(*u
))
9457 dprintf("for update_reshape\n");
9459 for (dl
= super
->devlist
; dl
; dl
= dl
->next
) {
9460 int size
= sizeof_imsm_dev(dl
->dev
, 1);
9462 if (u
->new_raid_disks
> u
->old_raid_disks
)
9463 size
+= sizeof(__u32
)*2*
9464 (u
->new_raid_disks
- u
->old_raid_disks
);
9471 len
= disks_to_mpb_size(u
->new_raid_disks
);
9472 dprintf("New anchor length is %llu\n", (unsigned long long)len
);
9475 case update_reshape_migration
: {
9476 /* for migration level 0->5 we need to add disks
9477 * so the same as for container operation we will copy
9478 * device to the bigger location.
9479 * in memory prepared device and new disk area are prepared
9480 * for usage in process update
9482 struct imsm_update_reshape_migration
*u
= (void *)update
->buf
;
9483 struct intel_dev
*id
;
9484 void **space_tail
= (void **)&update
->space_list
;
9487 int current_level
= -1;
9489 if (update
->len
< (int)sizeof(*u
))
9492 dprintf("for update_reshape\n");
9494 /* add space for bigger array in update
9496 for (id
= super
->devlist
; id
; id
= id
->next
) {
9497 if (id
->index
== (unsigned)u
->subdev
) {
9498 size
= sizeof_imsm_dev(id
->dev
, 1);
9499 if (u
->new_raid_disks
> u
->old_raid_disks
)
9500 size
+= sizeof(__u32
)*2*
9501 (u
->new_raid_disks
- u
->old_raid_disks
);
9509 if (update
->space_list
== NULL
)
9512 /* add space for disk in update
9514 size
= sizeof(struct dl
);
9520 /* add spare device to update
9522 for (id
= super
->devlist
; id
; id
= id
->next
)
9523 if (id
->index
== (unsigned)u
->subdev
) {
9524 struct imsm_dev
*dev
;
9525 struct imsm_map
*map
;
9527 dev
= get_imsm_dev(super
, u
->subdev
);
9528 map
= get_imsm_map(dev
, MAP_0
);
9529 current_level
= map
->raid_level
;
9532 if (u
->new_level
== 5 && u
->new_level
!= current_level
) {
9533 struct mdinfo
*spares
;
9535 spares
= get_spares_for_grow(st
);
9543 makedev(dev
->disk
.major
,
9545 dl
= get_disk_super(super
,
9548 dl
->index
= u
->old_raid_disks
;
9554 len
= disks_to_mpb_size(u
->new_raid_disks
);
9555 dprintf("New anchor length is %llu\n", (unsigned long long)len
);
9558 case update_size_change
: {
9559 if (update
->len
< (int)sizeof(struct imsm_update_size_change
))
9563 case update_activate_spare
: {
9564 if (update
->len
< (int)sizeof(struct imsm_update_activate_spare
))
9568 case update_create_array
: {
9569 struct imsm_update_create_array
*u
= (void *) update
->buf
;
9570 struct intel_dev
*dv
;
9571 struct imsm_dev
*dev
= &u
->dev
;
9572 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
9574 struct disk_info
*inf
;
9578 if (update
->len
< (int)sizeof(*u
))
9581 inf
= get_disk_info(u
);
9582 len
= sizeof_imsm_dev(dev
, 1);
9583 /* allocate a new super->devlist entry */
9584 dv
= xmalloc(sizeof(*dv
));
9585 dv
->dev
= xmalloc(len
);
9588 /* count how many spares will be converted to members */
9589 for (i
= 0; i
< map
->num_members
; i
++) {
9590 dl
= serial_to_dl(inf
[i
].serial
, super
);
9592 /* hmm maybe it failed?, nothing we can do about
9597 if (count_memberships(dl
, super
) == 0)
9600 len
+= activate
* sizeof(struct imsm_disk
);
9603 case update_kill_array
: {
9604 if (update
->len
< (int)sizeof(struct imsm_update_kill_array
))
9608 case update_rename_array
: {
9609 if (update
->len
< (int)sizeof(struct imsm_update_rename_array
))
9613 case update_add_remove_disk
:
9614 /* no update->len needed */
9616 case update_prealloc_badblocks_mem
:
9617 super
->extra_space
+= sizeof(struct bbm_log
) -
9618 get_imsm_bbm_log_size(super
->bbm_log
);
9624 /* check if we need a larger metadata buffer */
9625 if (super
->next_buf
)
9626 buf_len
= super
->next_len
;
9628 buf_len
= super
->len
;
9630 if (__le32_to_cpu(mpb
->mpb_size
) + super
->extra_space
+ len
> buf_len
) {
9631 /* ok we need a larger buf than what is currently allocated
9632 * if this allocation fails process_update will notice that
9633 * ->next_len is set and ->next_buf is NULL
9635 buf_len
= ROUND_UP(__le32_to_cpu(mpb
->mpb_size
) +
9636 super
->extra_space
+ len
, sector_size
);
9637 if (super
->next_buf
)
9638 free(super
->next_buf
);
9640 super
->next_len
= buf_len
;
9641 if (posix_memalign(&super
->next_buf
, sector_size
, buf_len
) == 0)
9642 memset(super
->next_buf
, 0, buf_len
);
9644 super
->next_buf
= NULL
;
9649 /* must be called while manager is quiesced */
9650 static void imsm_delete(struct intel_super
*super
, struct dl
**dlp
, unsigned index
)
9652 struct imsm_super
*mpb
= super
->anchor
;
9654 struct imsm_dev
*dev
;
9655 struct imsm_map
*map
;
9656 unsigned int i
, j
, num_members
;
9658 struct bbm_log
*log
= super
->bbm_log
;
9660 dprintf("deleting device[%d] from imsm_super\n", index
);
9662 /* shift all indexes down one */
9663 for (iter
= super
->disks
; iter
; iter
= iter
->next
)
9664 if (iter
->index
> (int)index
)
9666 for (iter
= super
->missing
; iter
; iter
= iter
->next
)
9667 if (iter
->index
> (int)index
)
9670 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
9671 dev
= get_imsm_dev(super
, i
);
9672 map
= get_imsm_map(dev
, MAP_0
);
9673 num_members
= map
->num_members
;
9674 for (j
= 0; j
< num_members
; j
++) {
9675 /* update ord entries being careful not to propagate
9676 * ord-flags to the first map
9678 ord
= get_imsm_ord_tbl_ent(dev
, j
, MAP_X
);
9680 if (ord_to_idx(ord
) <= index
)
9683 map
= get_imsm_map(dev
, MAP_0
);
9684 set_imsm_ord_tbl_ent(map
, j
, ord_to_idx(ord
- 1));
9685 map
= get_imsm_map(dev
, MAP_1
);
9687 set_imsm_ord_tbl_ent(map
, j
, ord
- 1);
9691 for (i
= 0; i
< log
->entry_count
; i
++) {
9692 struct bbm_log_entry
*entry
= &log
->marked_block_entries
[i
];
9694 if (entry
->disk_ordinal
<= index
)
9696 entry
->disk_ordinal
--;
9700 super
->updates_pending
++;
9702 struct dl
*dl
= *dlp
;
9704 *dlp
= (*dlp
)->next
;
9705 __free_imsm_disk(dl
);
9708 #endif /* MDASSEMBLE */
9710 static void close_targets(int *targets
, int new_disks
)
9717 for (i
= 0; i
< new_disks
; i
++) {
9718 if (targets
[i
] >= 0) {
9725 static int imsm_get_allowed_degradation(int level
, int raid_disks
,
9726 struct intel_super
*super
,
9727 struct imsm_dev
*dev
)
9733 struct imsm_map
*map
;
9736 ret_val
= raid_disks
/2;
9737 /* check map if all disks pairs not failed
9740 map
= get_imsm_map(dev
, MAP_0
);
9741 for (i
= 0; i
< ret_val
; i
++) {
9742 int degradation
= 0;
9743 if (get_imsm_disk(super
, i
) == NULL
)
9745 if (get_imsm_disk(super
, i
+ 1) == NULL
)
9747 if (degradation
== 2)
9750 map
= get_imsm_map(dev
, MAP_1
);
9751 /* if there is no second map
9752 * result can be returned
9756 /* check degradation in second map
9758 for (i
= 0; i
< ret_val
; i
++) {
9759 int degradation
= 0;
9760 if (get_imsm_disk(super
, i
) == NULL
)
9762 if (get_imsm_disk(super
, i
+ 1) == NULL
)
9764 if (degradation
== 2)
9778 /*******************************************************************************
9779 * Function: open_backup_targets
9780 * Description: Function opens file descriptors for all devices given in
9783 * info : general array info
9784 * raid_disks : number of disks
9785 * raid_fds : table of device's file descriptors
9786 * super : intel super for raid10 degradation check
9787 * dev : intel device for raid10 degradation check
9791 ******************************************************************************/
9792 int open_backup_targets(struct mdinfo
*info
, int raid_disks
, int *raid_fds
,
9793 struct intel_super
*super
, struct imsm_dev
*dev
)
9799 for (i
= 0; i
< raid_disks
; i
++)
9802 for (sd
= info
->devs
; sd
; sd
= sd
->next
) {
9805 if (sd
->disk
.state
& (1<<MD_DISK_FAULTY
)) {
9806 dprintf("disk is faulty!!\n");
9810 if (sd
->disk
.raid_disk
>= raid_disks
|| sd
->disk
.raid_disk
< 0)
9813 dn
= map_dev(sd
->disk
.major
,
9815 raid_fds
[sd
->disk
.raid_disk
] = dev_open(dn
, O_RDWR
);
9816 if (raid_fds
[sd
->disk
.raid_disk
] < 0) {
9817 pr_err("cannot open component\n");
9822 /* check if maximum array degradation level is not exceeded
9824 if ((raid_disks
- opened
) >
9825 imsm_get_allowed_degradation(info
->new_level
, raid_disks
,
9827 pr_err("Not enough disks can be opened.\n");
9828 close_targets(raid_fds
, raid_disks
);
9834 /*******************************************************************************
9835 * Function: validate_container_imsm
9836 * Description: This routine validates container after assemble,
9837 * eg. if devices in container are under the same controller.
9840 * info : linked list with info about devices used in array
9844 ******************************************************************************/
9845 int validate_container_imsm(struct mdinfo
*info
)
9847 if (check_env("IMSM_NO_PLATFORM"))
9850 struct sys_dev
*idev
;
9851 struct sys_dev
*hba
= NULL
;
9852 struct sys_dev
*intel_devices
= find_intel_devices();
9853 char *dev_path
= devt_to_devpath(makedev(info
->disk
.major
,
9856 for (idev
= intel_devices
; idev
; idev
= idev
->next
) {
9857 if (dev_path
&& strstr(dev_path
, idev
->path
)) {
9866 pr_err("WARNING - Cannot detect HBA for device %s!\n",
9867 devid2kname(makedev(info
->disk
.major
, info
->disk
.minor
)));
9871 const struct imsm_orom
*orom
= get_orom_by_device_id(hba
->dev_id
);
9874 for (dev
= info
->next
; dev
; dev
= dev
->next
) {
9875 dev_path
= devt_to_devpath(makedev(dev
->disk
.major
, dev
->disk
.minor
));
9877 struct sys_dev
*hba2
= NULL
;
9878 for (idev
= intel_devices
; idev
; idev
= idev
->next
) {
9879 if (dev_path
&& strstr(dev_path
, idev
->path
)) {
9887 const struct imsm_orom
*orom2
= hba2
== NULL
? NULL
:
9888 get_orom_by_device_id(hba2
->dev_id
);
9890 if (hba2
&& hba
->type
!= hba2
->type
) {
9891 pr_err("WARNING - HBAs of devices do not match %s != %s\n",
9892 get_sys_dev_type(hba
->type
), get_sys_dev_type(hba2
->type
));
9896 if (orom
!= orom2
) {
9897 pr_err("WARNING - IMSM container assembled with disks under different HBAs!\n"
9898 " This operation is not supported and can lead to data loss.\n");
9903 pr_err("WARNING - IMSM container assembled with disks under HBAs without IMSM platform support!\n"
9904 " This operation is not supported and can lead to data loss.\n");
9912 /*******************************************************************************
9913 * Function: imsm_record_badblock
9914 * Description: This routine stores new bad block record in BBM log
9917 * a : array containing a bad block
9918 * slot : disk number containing a bad block
9919 * sector : bad block sector
9920 * length : bad block sectors range
9924 ******************************************************************************/
9925 static int imsm_record_badblock(struct active_array
*a
, int slot
,
9926 unsigned long long sector
, int length
)
9928 struct intel_super
*super
= a
->container
->sb
;
9932 ord
= imsm_disk_slot_to_ord(a
, slot
);
9936 ret
= record_new_badblock(super
->bbm_log
, ord_to_idx(ord
), sector
,
9939 super
->updates_pending
++;
9943 /*******************************************************************************
9944 * Function: imsm_clear_badblock
9945 * Description: This routine clears bad block record from BBM log
9948 * a : array containing a bad block
9949 * slot : disk number containing a bad block
9950 * sector : bad block sector
9951 * length : bad block sectors range
9955 ******************************************************************************/
9956 static int imsm_clear_badblock(struct active_array
*a
, int slot
,
9957 unsigned long long sector
, int length
)
9959 struct intel_super
*super
= a
->container
->sb
;
9963 ord
= imsm_disk_slot_to_ord(a
, slot
);
9967 ret
= clear_badblock(super
->bbm_log
, ord_to_idx(ord
), sector
, length
);
9969 super
->updates_pending
++;
9973 /*******************************************************************************
9974 * Function: imsm_get_badblocks
9975 * Description: This routine get list of bad blocks for an array
9979 * slot : disk number
9981 * bb : structure containing bad blocks
9983 ******************************************************************************/
9984 static struct md_bb
*imsm_get_badblocks(struct active_array
*a
, int slot
)
9986 int inst
= a
->info
.container_member
;
9987 struct intel_super
*super
= a
->container
->sb
;
9988 struct imsm_dev
*dev
= get_imsm_dev(super
, inst
);
9989 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
9992 ord
= imsm_disk_slot_to_ord(a
, slot
);
9996 get_volume_badblocks(super
->bbm_log
, ord_to_idx(ord
), pba_of_lba0(map
),
9997 blocks_per_member(map
), &super
->bb
);
10001 /*******************************************************************************
10002 * Function: examine_badblocks_imsm
10003 * Description: Prints list of bad blocks on a disk to the standard output
10006 * st : metadata handler
10007 * fd : open file descriptor for device
10008 * devname : device name
10012 ******************************************************************************/
10013 static int examine_badblocks_imsm(struct supertype
*st
, int fd
, char *devname
)
10015 struct intel_super
*super
= st
->sb
;
10016 struct bbm_log
*log
= super
->bbm_log
;
10017 struct dl
*d
= NULL
;
10020 for (d
= super
->disks
; d
; d
= d
->next
) {
10021 if (strcmp(d
->devname
, devname
) == 0)
10025 if ((d
== NULL
) || (d
->index
< 0)) { /* serial mismatch probably */
10026 pr_err("%s doesn't appear to be part of a raid array\n",
10033 struct bbm_log_entry
*entry
= &log
->marked_block_entries
[0];
10035 for (i
= 0; i
< log
->entry_count
; i
++) {
10036 if (entry
[i
].disk_ordinal
== d
->index
) {
10037 unsigned long long sector
= __le48_to_cpu(
10038 &entry
[i
].defective_block_start
);
10039 int cnt
= entry
[i
].marked_count
+ 1;
10042 printf("Bad-blocks on %s:\n", devname
);
10046 printf("%20llu for %d sectors\n", sector
, cnt
);
10052 printf("No bad-blocks list configured on %s\n", devname
);
10056 /*******************************************************************************
10057 * Function: init_migr_record_imsm
10058 * Description: Function inits imsm migration record
10060 * super : imsm internal array info
10061 * dev : device under migration
10062 * info : general array info to find the smallest device
10065 ******************************************************************************/
10066 void init_migr_record_imsm(struct supertype
*st
, struct imsm_dev
*dev
,
10067 struct mdinfo
*info
)
10069 struct intel_super
*super
= st
->sb
;
10070 struct migr_record
*migr_rec
= super
->migr_rec
;
10071 int new_data_disks
;
10072 unsigned long long dsize
, dev_sectors
;
10073 long long unsigned min_dev_sectors
= -1LLU;
10077 struct imsm_map
*map_dest
= get_imsm_map(dev
, MAP_0
);
10078 struct imsm_map
*map_src
= get_imsm_map(dev
, MAP_1
);
10079 unsigned long long num_migr_units
;
10080 unsigned long long array_blocks
;
10082 memset(migr_rec
, 0, sizeof(struct migr_record
));
10083 migr_rec
->family_num
= __cpu_to_le32(super
->anchor
->family_num
);
10085 /* only ascending reshape supported now */
10086 migr_rec
->ascending_migr
= __cpu_to_le32(1);
10088 migr_rec
->dest_depth_per_unit
= GEN_MIGR_AREA_SIZE
/
10089 max(map_dest
->blocks_per_strip
, map_src
->blocks_per_strip
);
10090 migr_rec
->dest_depth_per_unit
*=
10091 max(map_dest
->blocks_per_strip
, map_src
->blocks_per_strip
);
10092 new_data_disks
= imsm_num_data_members(dev
, MAP_0
);
10093 migr_rec
->blocks_per_unit
=
10094 __cpu_to_le32(migr_rec
->dest_depth_per_unit
* new_data_disks
);
10095 migr_rec
->dest_depth_per_unit
=
10096 __cpu_to_le32(migr_rec
->dest_depth_per_unit
);
10097 array_blocks
= info
->component_size
* new_data_disks
;
10099 array_blocks
/ __le32_to_cpu(migr_rec
->blocks_per_unit
);
10101 if (array_blocks
% __le32_to_cpu(migr_rec
->blocks_per_unit
))
10103 migr_rec
->num_migr_units
= __cpu_to_le32(num_migr_units
);
10105 migr_rec
->post_migr_vol_cap
= dev
->size_low
;
10106 migr_rec
->post_migr_vol_cap_hi
= dev
->size_high
;
10108 /* Find the smallest dev */
10109 for (sd
= info
->devs
; sd
; sd
= sd
->next
) {
10110 sprintf(nm
, "%d:%d", sd
->disk
.major
, sd
->disk
.minor
);
10111 fd
= dev_open(nm
, O_RDONLY
);
10114 get_dev_size(fd
, NULL
, &dsize
);
10115 dev_sectors
= dsize
/ 512;
10116 if (dev_sectors
< min_dev_sectors
)
10117 min_dev_sectors
= dev_sectors
;
10120 migr_rec
->ckpt_area_pba
= __cpu_to_le32(min_dev_sectors
-
10121 RAID_DISK_RESERVED_BLOCKS_IMSM_HI
);
10123 write_imsm_migr_rec(st
);
10128 /*******************************************************************************
10129 * Function: save_backup_imsm
10130 * Description: Function saves critical data stripes to Migration Copy Area
10131 * and updates the current migration unit status.
10132 * Use restore_stripes() to form a destination stripe,
10133 * and to write it to the Copy Area.
10135 * st : supertype information
10136 * dev : imsm device that backup is saved for
10137 * info : general array info
10138 * buf : input buffer
10139 * length : length of data to backup (blocks_per_unit)
10143 ******************************************************************************/
10144 int save_backup_imsm(struct supertype
*st
,
10145 struct imsm_dev
*dev
,
10146 struct mdinfo
*info
,
10151 struct intel_super
*super
= st
->sb
;
10152 unsigned long long *target_offsets
;
10155 struct imsm_map
*map_dest
= get_imsm_map(dev
, MAP_0
);
10156 int new_disks
= map_dest
->num_members
;
10157 int dest_layout
= 0;
10159 unsigned long long start
;
10160 int data_disks
= imsm_num_data_members(dev
, MAP_0
);
10162 targets
= xmalloc(new_disks
* sizeof(int));
10164 for (i
= 0; i
< new_disks
; i
++)
10167 target_offsets
= xcalloc(new_disks
, sizeof(unsigned long long));
10169 start
= info
->reshape_progress
* 512;
10170 for (i
= 0; i
< new_disks
; i
++) {
10171 target_offsets
[i
] = (unsigned long long)
10172 __le32_to_cpu(super
->migr_rec
->ckpt_area_pba
) * 512;
10173 /* move back copy area adderss, it will be moved forward
10174 * in restore_stripes() using start input variable
10176 target_offsets
[i
] -= start
/data_disks
;
10179 if (open_backup_targets(info
, new_disks
, targets
,
10183 dest_layout
= imsm_level_to_layout(map_dest
->raid_level
);
10184 dest_chunk
= __le16_to_cpu(map_dest
->blocks_per_strip
) * 512;
10186 if (restore_stripes(targets
, /* list of dest devices */
10187 target_offsets
, /* migration record offsets */
10190 map_dest
->raid_level
,
10192 -1, /* source backup file descriptor */
10193 0, /* input buf offset
10194 * always 0 buf is already offseted */
10198 pr_err("Error restoring stripes\n");
10206 close_targets(targets
, new_disks
);
10209 free(target_offsets
);
10214 /*******************************************************************************
10215 * Function: save_checkpoint_imsm
10216 * Description: Function called for current unit status update
10217 * in the migration record. It writes it to disk.
10219 * super : imsm internal array info
10220 * info : general array info
10224 * 2: failure, means no valid migration record
10225 * / no general migration in progress /
10226 ******************************************************************************/
10227 int save_checkpoint_imsm(struct supertype
*st
, struct mdinfo
*info
, int state
)
10229 struct intel_super
*super
= st
->sb
;
10230 unsigned long long blocks_per_unit
;
10231 unsigned long long curr_migr_unit
;
10233 if (load_imsm_migr_rec(super
, info
) != 0) {
10234 dprintf("imsm: ERROR: Cannot read migration record for checkpoint save.\n");
10238 blocks_per_unit
= __le32_to_cpu(super
->migr_rec
->blocks_per_unit
);
10239 if (blocks_per_unit
== 0) {
10240 dprintf("imsm: no migration in progress.\n");
10243 curr_migr_unit
= info
->reshape_progress
/ blocks_per_unit
;
10244 /* check if array is alligned to copy area
10245 * if it is not alligned, add one to current migration unit value
10246 * this can happend on array reshape finish only
10248 if (info
->reshape_progress
% blocks_per_unit
)
10251 super
->migr_rec
->curr_migr_unit
=
10252 __cpu_to_le32(curr_migr_unit
);
10253 super
->migr_rec
->rec_status
= __cpu_to_le32(state
);
10254 super
->migr_rec
->dest_1st_member_lba
=
10255 __cpu_to_le32(curr_migr_unit
*
10256 __le32_to_cpu(super
->migr_rec
->dest_depth_per_unit
));
10257 if (write_imsm_migr_rec(st
) < 0) {
10258 dprintf("imsm: Cannot write migration record outside backup area\n");
10265 /*******************************************************************************
10266 * Function: recover_backup_imsm
10267 * Description: Function recovers critical data from the Migration Copy Area
10268 * while assembling an array.
10270 * super : imsm internal array info
10271 * info : general array info
10273 * 0 : success (or there is no data to recover)
10275 ******************************************************************************/
10276 int recover_backup_imsm(struct supertype
*st
, struct mdinfo
*info
)
10278 struct intel_super
*super
= st
->sb
;
10279 struct migr_record
*migr_rec
= super
->migr_rec
;
10280 struct imsm_map
*map_dest
;
10281 struct intel_dev
*id
= NULL
;
10282 unsigned long long read_offset
;
10283 unsigned long long write_offset
;
10285 int *targets
= NULL
;
10286 int new_disks
, i
, err
;
10289 unsigned int sector_size
= super
->sector_size
;
10290 unsigned long curr_migr_unit
= __le32_to_cpu(migr_rec
->curr_migr_unit
);
10291 unsigned long num_migr_units
= __le32_to_cpu(migr_rec
->num_migr_units
);
10293 int skipped_disks
= 0;
10295 err
= sysfs_get_str(info
, NULL
, "array_state", (char *)buffer
, 20);
10299 /* recover data only during assemblation */
10300 if (strncmp(buffer
, "inactive", 8) != 0)
10302 /* no data to recover */
10303 if (__le32_to_cpu(migr_rec
->rec_status
) == UNIT_SRC_NORMAL
)
10305 if (curr_migr_unit
>= num_migr_units
)
10308 /* find device during reshape */
10309 for (id
= super
->devlist
; id
; id
= id
->next
)
10310 if (is_gen_migration(id
->dev
))
10315 map_dest
= get_imsm_map(id
->dev
, MAP_0
);
10316 new_disks
= map_dest
->num_members
;
10318 read_offset
= (unsigned long long)
10319 __le32_to_cpu(migr_rec
->ckpt_area_pba
) * 512;
10321 write_offset
= ((unsigned long long)
10322 __le32_to_cpu(migr_rec
->dest_1st_member_lba
) +
10323 pba_of_lba0(map_dest
)) * 512;
10325 unit_len
= __le32_to_cpu(migr_rec
->dest_depth_per_unit
) * 512;
10326 if (posix_memalign((void **)&buf
, sector_size
, unit_len
) != 0)
10328 targets
= xcalloc(new_disks
, sizeof(int));
10330 if (open_backup_targets(info
, new_disks
, targets
, super
, id
->dev
)) {
10331 pr_err("Cannot open some devices belonging to array.\n");
10335 for (i
= 0; i
< new_disks
; i
++) {
10336 if (targets
[i
] < 0) {
10340 if (lseek64(targets
[i
], read_offset
, SEEK_SET
) < 0) {
10341 pr_err("Cannot seek to block: %s\n",
10346 if ((unsigned)read(targets
[i
], buf
, unit_len
) != unit_len
) {
10347 pr_err("Cannot read copy area block: %s\n",
10352 if (lseek64(targets
[i
], write_offset
, SEEK_SET
) < 0) {
10353 pr_err("Cannot seek to block: %s\n",
10358 if ((unsigned)write(targets
[i
], buf
, unit_len
) != unit_len
) {
10359 pr_err("Cannot restore block: %s\n",
10366 if (skipped_disks
> imsm_get_allowed_degradation(info
->new_level
,
10370 pr_err("Cannot restore data from backup. Too many failed disks\n");
10374 if (save_checkpoint_imsm(st
, info
, UNIT_SRC_NORMAL
)) {
10375 /* ignore error == 2, this can mean end of reshape here
10377 dprintf("imsm: Cannot write checkpoint to migration record (UNIT_SRC_NORMAL) during restart\n");
10383 for (i
= 0; i
< new_disks
; i
++)
10392 static char disk_by_path
[] = "/dev/disk/by-path/";
10394 static const char *imsm_get_disk_controller_domain(const char *path
)
10396 char disk_path
[PATH_MAX
];
10400 strcpy(disk_path
, disk_by_path
);
10401 strncat(disk_path
, path
, PATH_MAX
- strlen(disk_path
) - 1);
10402 if (stat(disk_path
, &st
) == 0) {
10403 struct sys_dev
* hba
;
10406 path
= devt_to_devpath(st
.st_rdev
);
10409 hba
= find_disk_attached_hba(-1, path
);
10410 if (hba
&& hba
->type
== SYS_DEV_SAS
)
10412 else if (hba
&& hba
->type
== SYS_DEV_SATA
)
10416 dprintf("path: %s hba: %s attached: %s\n",
10417 path
, (hba
) ? hba
->path
: "NULL", drv
);
10423 static char *imsm_find_array_devnm_by_subdev(int subdev
, char *container
)
10425 static char devnm
[32];
10426 char subdev_name
[20];
10427 struct mdstat_ent
*mdstat
;
10429 sprintf(subdev_name
, "%d", subdev
);
10430 mdstat
= mdstat_by_subdev(subdev_name
, container
);
10434 strcpy(devnm
, mdstat
->devnm
);
10435 free_mdstat(mdstat
);
10439 static int imsm_reshape_is_allowed_on_container(struct supertype
*st
,
10440 struct geo_params
*geo
,
10441 int *old_raid_disks
,
10444 /* currently we only support increasing the number of devices
10445 * for a container. This increases the number of device for each
10446 * member array. They must all be RAID0 or RAID5.
10449 struct mdinfo
*info
, *member
;
10450 int devices_that_can_grow
= 0;
10452 dprintf("imsm: imsm_reshape_is_allowed_on_container(ENTER): st->devnm = (%s)\n", st
->devnm
);
10454 if (geo
->size
> 0 ||
10455 geo
->level
!= UnSet
||
10456 geo
->layout
!= UnSet
||
10457 geo
->chunksize
!= 0 ||
10458 geo
->raid_disks
== UnSet
) {
10459 dprintf("imsm: Container operation is allowed for raid disks number change only.\n");
10463 if (direction
== ROLLBACK_METADATA_CHANGES
) {
10464 dprintf("imsm: Metadata changes rollback is not supported for container operation.\n");
10468 info
= container_content_imsm(st
, NULL
);
10469 for (member
= info
; member
; member
= member
->next
) {
10472 dprintf("imsm: checking device_num: %i\n",
10473 member
->container_member
);
10475 if (geo
->raid_disks
<= member
->array
.raid_disks
) {
10476 /* we work on container for Online Capacity Expansion
10477 * only so raid_disks has to grow
10479 dprintf("imsm: for container operation raid disks increase is required\n");
10483 if (info
->array
.level
!= 0 && info
->array
.level
!= 5) {
10484 /* we cannot use this container with other raid level
10486 dprintf("imsm: for container operation wrong raid level (%i) detected\n",
10487 info
->array
.level
);
10490 /* check for platform support
10491 * for this raid level configuration
10493 struct intel_super
*super
= st
->sb
;
10494 if (!is_raid_level_supported(super
->orom
,
10495 member
->array
.level
,
10496 geo
->raid_disks
)) {
10497 dprintf("platform does not support raid%d with %d disk%s\n",
10500 geo
->raid_disks
> 1 ? "s" : "");
10503 /* check if component size is aligned to chunk size
10505 if (info
->component_size
%
10506 (info
->array
.chunk_size
/512)) {
10507 dprintf("Component size is not aligned to chunk size\n");
10512 if (*old_raid_disks
&&
10513 info
->array
.raid_disks
!= *old_raid_disks
)
10515 *old_raid_disks
= info
->array
.raid_disks
;
10517 /* All raid5 and raid0 volumes in container
10518 * have to be ready for Online Capacity Expansion
10519 * so they need to be assembled. We have already
10520 * checked that no recovery etc is happening.
10522 result
= imsm_find_array_devnm_by_subdev(member
->container_member
,
10523 st
->container_devnm
);
10524 if (result
== NULL
) {
10525 dprintf("imsm: cannot find array\n");
10528 devices_that_can_grow
++;
10531 if (!member
&& devices_that_can_grow
)
10535 dprintf("Container operation allowed\n");
10537 dprintf("Error: %i\n", ret_val
);
10542 /* Function: get_spares_for_grow
10543 * Description: Allocates memory and creates list of spare devices
10544 * avaliable in container. Checks if spare drive size is acceptable.
10545 * Parameters: Pointer to the supertype structure
10546 * Returns: Pointer to the list of spare devices (mdinfo structure) on success,
10549 static struct mdinfo
*get_spares_for_grow(struct supertype
*st
)
10551 unsigned long long min_size
= min_acceptable_spare_size_imsm(st
);
10552 return container_choose_spares(st
, min_size
, NULL
, NULL
, NULL
, 0);
10555 /******************************************************************************
10556 * function: imsm_create_metadata_update_for_reshape
10557 * Function creates update for whole IMSM container.
10559 ******************************************************************************/
10560 static int imsm_create_metadata_update_for_reshape(
10561 struct supertype
*st
,
10562 struct geo_params
*geo
,
10563 int old_raid_disks
,
10564 struct imsm_update_reshape
**updatep
)
10566 struct intel_super
*super
= st
->sb
;
10567 struct imsm_super
*mpb
= super
->anchor
;
10568 int update_memory_size
;
10569 struct imsm_update_reshape
*u
;
10570 struct mdinfo
*spares
;
10573 struct mdinfo
*dev
;
10575 dprintf("(enter) raid_disks = %i\n", geo
->raid_disks
);
10577 delta_disks
= geo
->raid_disks
- old_raid_disks
;
10579 /* size of all update data without anchor */
10580 update_memory_size
= sizeof(struct imsm_update_reshape
);
10582 /* now add space for spare disks that we need to add. */
10583 update_memory_size
+= sizeof(u
->new_disks
[0]) * (delta_disks
- 1);
10585 u
= xcalloc(1, update_memory_size
);
10586 u
->type
= update_reshape_container_disks
;
10587 u
->old_raid_disks
= old_raid_disks
;
10588 u
->new_raid_disks
= geo
->raid_disks
;
10590 /* now get spare disks list
10592 spares
= get_spares_for_grow(st
);
10595 || delta_disks
> spares
->array
.spare_disks
) {
10596 pr_err("imsm: ERROR: Cannot get spare devices for %s.\n", geo
->dev_name
);
10601 /* we have got spares
10602 * update disk list in imsm_disk list table in anchor
10604 dprintf("imsm: %i spares are available.\n\n",
10605 spares
->array
.spare_disks
);
10607 dev
= spares
->devs
;
10608 for (i
= 0; i
< delta_disks
; i
++) {
10613 u
->new_disks
[i
] = makedev(dev
->disk
.major
,
10615 dl
= get_disk_super(super
, dev
->disk
.major
, dev
->disk
.minor
);
10616 dl
->index
= mpb
->num_disks
;
10624 sysfs_free(spares
);
10626 dprintf("imsm: reshape update preparation :");
10627 if (i
== delta_disks
) {
10628 dprintf_cont(" OK\n");
10630 return update_memory_size
;
10633 dprintf_cont(" Error\n");
10638 /******************************************************************************
10639 * function: imsm_create_metadata_update_for_size_change()
10640 * Creates update for IMSM array for array size change.
10642 ******************************************************************************/
10643 static int imsm_create_metadata_update_for_size_change(
10644 struct supertype
*st
,
10645 struct geo_params
*geo
,
10646 struct imsm_update_size_change
**updatep
)
10648 struct intel_super
*super
= st
->sb
;
10649 int update_memory_size
;
10650 struct imsm_update_size_change
*u
;
10652 dprintf("(enter) New size = %llu\n", geo
->size
);
10654 /* size of all update data without anchor */
10655 update_memory_size
= sizeof(struct imsm_update_size_change
);
10657 u
= xcalloc(1, update_memory_size
);
10658 u
->type
= update_size_change
;
10659 u
->subdev
= super
->current_vol
;
10660 u
->new_size
= geo
->size
;
10662 dprintf("imsm: reshape update preparation : OK\n");
10665 return update_memory_size
;
10668 /******************************************************************************
10669 * function: imsm_create_metadata_update_for_migration()
10670 * Creates update for IMSM array.
10672 ******************************************************************************/
10673 static int imsm_create_metadata_update_for_migration(
10674 struct supertype
*st
,
10675 struct geo_params
*geo
,
10676 struct imsm_update_reshape_migration
**updatep
)
10678 struct intel_super
*super
= st
->sb
;
10679 int update_memory_size
;
10680 struct imsm_update_reshape_migration
*u
;
10681 struct imsm_dev
*dev
;
10682 int previous_level
= -1;
10684 dprintf("(enter) New Level = %i\n", geo
->level
);
10686 /* size of all update data without anchor */
10687 update_memory_size
= sizeof(struct imsm_update_reshape_migration
);
10689 u
= xcalloc(1, update_memory_size
);
10690 u
->type
= update_reshape_migration
;
10691 u
->subdev
= super
->current_vol
;
10692 u
->new_level
= geo
->level
;
10693 u
->new_layout
= geo
->layout
;
10694 u
->new_raid_disks
= u
->old_raid_disks
= geo
->raid_disks
;
10695 u
->new_disks
[0] = -1;
10696 u
->new_chunksize
= -1;
10698 dev
= get_imsm_dev(super
, u
->subdev
);
10700 struct imsm_map
*map
;
10702 map
= get_imsm_map(dev
, MAP_0
);
10704 int current_chunk_size
=
10705 __le16_to_cpu(map
->blocks_per_strip
) / 2;
10707 if (geo
->chunksize
!= current_chunk_size
) {
10708 u
->new_chunksize
= geo
->chunksize
/ 1024;
10709 dprintf("imsm: chunk size change from %i to %i\n",
10710 current_chunk_size
, u
->new_chunksize
);
10712 previous_level
= map
->raid_level
;
10715 if (geo
->level
== 5 && previous_level
== 0) {
10716 struct mdinfo
*spares
= NULL
;
10718 u
->new_raid_disks
++;
10719 spares
= get_spares_for_grow(st
);
10720 if (spares
== NULL
|| spares
->array
.spare_disks
< 1) {
10722 sysfs_free(spares
);
10723 update_memory_size
= 0;
10724 pr_err("cannot get spare device for requested migration\n");
10727 sysfs_free(spares
);
10729 dprintf("imsm: reshape update preparation : OK\n");
10732 return update_memory_size
;
10735 static void imsm_update_metadata_locally(struct supertype
*st
,
10736 void *buf
, int len
)
10738 struct metadata_update mu
;
10743 mu
.space_list
= NULL
;
10745 if (imsm_prepare_update(st
, &mu
))
10746 imsm_process_update(st
, &mu
);
10748 while (mu
.space_list
) {
10749 void **space
= mu
.space_list
;
10750 mu
.space_list
= *space
;
10755 /***************************************************************************
10756 * Function: imsm_analyze_change
10757 * Description: Function analyze change for single volume
10758 * and validate if transition is supported
10759 * Parameters: Geometry parameters, supertype structure,
10760 * metadata change direction (apply/rollback)
10761 * Returns: Operation type code on success, -1 if fail
10762 ****************************************************************************/
10763 enum imsm_reshape_type
imsm_analyze_change(struct supertype
*st
,
10764 struct geo_params
*geo
,
10767 struct mdinfo info
;
10769 int check_devs
= 0;
10771 /* number of added/removed disks in operation result */
10772 int devNumChange
= 0;
10773 /* imsm compatible layout value for array geometry verification */
10774 int imsm_layout
= -1;
10776 struct imsm_dev
*dev
;
10777 struct intel_super
*super
;
10778 unsigned long long current_size
;
10779 unsigned long long free_size
;
10780 unsigned long long max_size
;
10783 getinfo_super_imsm_volume(st
, &info
, NULL
);
10784 if (geo
->level
!= info
.array
.level
&& geo
->level
>= 0 &&
10785 geo
->level
!= UnSet
) {
10786 switch (info
.array
.level
) {
10788 if (geo
->level
== 5) {
10789 change
= CH_MIGRATION
;
10790 if (geo
->layout
!= ALGORITHM_LEFT_ASYMMETRIC
) {
10791 pr_err("Error. Requested Layout not supported (left-asymmetric layout is supported only)!\n");
10793 goto analyse_change_exit
;
10795 imsm_layout
= geo
->layout
;
10797 devNumChange
= 1; /* parity disk added */
10798 } else if (geo
->level
== 10) {
10799 change
= CH_TAKEOVER
;
10801 devNumChange
= 2; /* two mirrors added */
10802 imsm_layout
= 0x102; /* imsm supported layout */
10807 if (geo
->level
== 0) {
10808 change
= CH_TAKEOVER
;
10810 devNumChange
= -(geo
->raid_disks
/2);
10811 imsm_layout
= 0; /* imsm raid0 layout */
10815 if (change
== -1) {
10816 pr_err("Error. Level Migration from %d to %d not supported!\n",
10817 info
.array
.level
, geo
->level
);
10818 goto analyse_change_exit
;
10821 geo
->level
= info
.array
.level
;
10823 if (geo
->layout
!= info
.array
.layout
&&
10824 (geo
->layout
!= UnSet
&& geo
->layout
!= -1)) {
10825 change
= CH_MIGRATION
;
10826 if (info
.array
.layout
== 0 && info
.array
.level
== 5 &&
10827 geo
->layout
== 5) {
10828 /* reshape 5 -> 4 */
10829 } else if (info
.array
.layout
== 5 && info
.array
.level
== 5 &&
10830 geo
->layout
== 0) {
10831 /* reshape 4 -> 5 */
10835 pr_err("Error. Layout Migration from %d to %d not supported!\n",
10836 info
.array
.layout
, geo
->layout
);
10838 goto analyse_change_exit
;
10841 geo
->layout
= info
.array
.layout
;
10842 if (imsm_layout
== -1)
10843 imsm_layout
= info
.array
.layout
;
10846 if (geo
->chunksize
> 0 && geo
->chunksize
!= UnSet
&&
10847 geo
->chunksize
!= info
.array
.chunk_size
) {
10848 if (info
.array
.level
== 10) {
10849 pr_err("Error. Chunk size change for RAID 10 is not supported.\n");
10851 goto analyse_change_exit
;
10852 } else if (info
.component_size
% (geo
->chunksize
/512)) {
10853 pr_err("New chunk size (%dK) does not evenly divide device size (%lluk). Aborting...\n",
10854 geo
->chunksize
/1024, info
.component_size
/2);
10856 goto analyse_change_exit
;
10858 change
= CH_MIGRATION
;
10860 geo
->chunksize
= info
.array
.chunk_size
;
10863 chunk
= geo
->chunksize
/ 1024;
10866 dev
= get_imsm_dev(super
, super
->current_vol
);
10867 data_disks
= imsm_num_data_members(dev
, MAP_0
);
10868 /* compute current size per disk member
10870 current_size
= info
.custom_array_size
/ data_disks
;
10872 if (geo
->size
> 0 && geo
->size
!= MAX_SIZE
) {
10873 /* align component size
10875 geo
->size
= imsm_component_size_aligment_check(
10876 get_imsm_raid_level(dev
->vol
.map
),
10877 chunk
* 1024, super
->sector_size
,
10879 if (geo
->size
== 0) {
10880 pr_err("Error. Size expansion is supported only (current size is %llu, requested size /rounded/ is 0).\n",
10882 goto analyse_change_exit
;
10886 if (current_size
!= geo
->size
&& geo
->size
> 0) {
10887 if (change
!= -1) {
10888 pr_err("Error. Size change should be the only one at a time.\n");
10890 goto analyse_change_exit
;
10892 if ((super
->current_vol
+ 1) != super
->anchor
->num_raid_devs
) {
10893 pr_err("Error. The last volume in container can be expanded only (%i/%s).\n",
10894 super
->current_vol
, st
->devnm
);
10895 goto analyse_change_exit
;
10897 /* check the maximum available size
10899 rv
= imsm_get_free_size(st
, dev
->vol
.map
->num_members
,
10900 0, chunk
, &free_size
);
10902 /* Cannot find maximum available space
10906 max_size
= free_size
+ current_size
;
10907 /* align component size
10909 max_size
= imsm_component_size_aligment_check(
10910 get_imsm_raid_level(dev
->vol
.map
),
10911 chunk
* 1024, super
->sector_size
,
10914 if (geo
->size
== MAX_SIZE
) {
10915 /* requested size change to the maximum available size
10917 if (max_size
== 0) {
10918 pr_err("Error. Cannot find maximum available space.\n");
10920 goto analyse_change_exit
;
10922 geo
->size
= max_size
;
10925 if (direction
== ROLLBACK_METADATA_CHANGES
) {
10926 /* accept size for rollback only
10929 /* round size due to metadata compatibility
10931 geo
->size
= (geo
->size
>> SECT_PER_MB_SHIFT
)
10932 << SECT_PER_MB_SHIFT
;
10933 dprintf("Prepare update for size change to %llu\n",
10935 if (current_size
>= geo
->size
) {
10936 pr_err("Error. Size expansion is supported only (current size is %llu, requested size /rounded/ is %llu).\n",
10937 current_size
, geo
->size
);
10938 goto analyse_change_exit
;
10940 if (max_size
&& geo
->size
> max_size
) {
10941 pr_err("Error. Requested size is larger than maximum available size (maximum available size is %llu, requested size /rounded/ is %llu).\n",
10942 max_size
, geo
->size
);
10943 goto analyse_change_exit
;
10946 geo
->size
*= data_disks
;
10947 geo
->raid_disks
= dev
->vol
.map
->num_members
;
10948 change
= CH_ARRAY_SIZE
;
10950 if (!validate_geometry_imsm(st
,
10953 geo
->raid_disks
+ devNumChange
,
10955 geo
->size
, INVALID_SECTORS
,
10960 struct intel_super
*super
= st
->sb
;
10961 struct imsm_super
*mpb
= super
->anchor
;
10963 if (mpb
->num_raid_devs
> 1) {
10964 pr_err("Error. Cannot perform operation on %s- for this operation it MUST be single array in container\n",
10970 analyse_change_exit
:
10971 if (direction
== ROLLBACK_METADATA_CHANGES
&&
10972 (change
== CH_MIGRATION
|| change
== CH_TAKEOVER
)) {
10973 dprintf("imsm: Metadata changes rollback is not supported for migration and takeover operations.\n");
10979 int imsm_takeover(struct supertype
*st
, struct geo_params
*geo
)
10981 struct intel_super
*super
= st
->sb
;
10982 struct imsm_update_takeover
*u
;
10984 u
= xmalloc(sizeof(struct imsm_update_takeover
));
10986 u
->type
= update_takeover
;
10987 u
->subarray
= super
->current_vol
;
10989 /* 10->0 transition */
10990 if (geo
->level
== 0)
10991 u
->direction
= R10_TO_R0
;
10993 /* 0->10 transition */
10994 if (geo
->level
== 10)
10995 u
->direction
= R0_TO_R10
;
10997 /* update metadata locally */
10998 imsm_update_metadata_locally(st
, u
,
10999 sizeof(struct imsm_update_takeover
));
11000 /* and possibly remotely */
11001 if (st
->update_tail
)
11002 append_metadata_update(st
, u
,
11003 sizeof(struct imsm_update_takeover
));
11010 static int imsm_reshape_super(struct supertype
*st
, unsigned long long size
,
11012 int layout
, int chunksize
, int raid_disks
,
11013 int delta_disks
, char *backup
, char *dev
,
11014 int direction
, int verbose
)
11017 struct geo_params geo
;
11019 dprintf("(enter)\n");
11021 memset(&geo
, 0, sizeof(struct geo_params
));
11023 geo
.dev_name
= dev
;
11024 strcpy(geo
.devnm
, st
->devnm
);
11027 geo
.layout
= layout
;
11028 geo
.chunksize
= chunksize
;
11029 geo
.raid_disks
= raid_disks
;
11030 if (delta_disks
!= UnSet
)
11031 geo
.raid_disks
+= delta_disks
;
11033 dprintf("for level : %i\n", geo
.level
);
11034 dprintf("for raid_disks : %i\n", geo
.raid_disks
);
11036 if (experimental() == 0)
11039 if (strcmp(st
->container_devnm
, st
->devnm
) == 0) {
11040 /* On container level we can only increase number of devices. */
11041 dprintf("imsm: info: Container operation\n");
11042 int old_raid_disks
= 0;
11044 if (imsm_reshape_is_allowed_on_container(
11045 st
, &geo
, &old_raid_disks
, direction
)) {
11046 struct imsm_update_reshape
*u
= NULL
;
11049 len
= imsm_create_metadata_update_for_reshape(
11050 st
, &geo
, old_raid_disks
, &u
);
11053 dprintf("imsm: Cannot prepare update\n");
11054 goto exit_imsm_reshape_super
;
11058 /* update metadata locally */
11059 imsm_update_metadata_locally(st
, u
, len
);
11060 /* and possibly remotely */
11061 if (st
->update_tail
)
11062 append_metadata_update(st
, u
, len
);
11067 pr_err("(imsm) Operation is not allowed on this container\n");
11070 /* On volume level we support following operations
11071 * - takeover: raid10 -> raid0; raid0 -> raid10
11072 * - chunk size migration
11073 * - migration: raid5 -> raid0; raid0 -> raid5
11075 struct intel_super
*super
= st
->sb
;
11076 struct intel_dev
*dev
= super
->devlist
;
11078 dprintf("imsm: info: Volume operation\n");
11079 /* find requested device */
11082 imsm_find_array_devnm_by_subdev(
11083 dev
->index
, st
->container_devnm
);
11084 if (devnm
&& strcmp(devnm
, geo
.devnm
) == 0)
11089 pr_err("Cannot find %s (%s) subarray\n",
11090 geo
.dev_name
, geo
.devnm
);
11091 goto exit_imsm_reshape_super
;
11093 super
->current_vol
= dev
->index
;
11094 change
= imsm_analyze_change(st
, &geo
, direction
);
11097 ret_val
= imsm_takeover(st
, &geo
);
11099 case CH_MIGRATION
: {
11100 struct imsm_update_reshape_migration
*u
= NULL
;
11102 imsm_create_metadata_update_for_migration(
11105 dprintf("imsm: Cannot prepare update\n");
11109 /* update metadata locally */
11110 imsm_update_metadata_locally(st
, u
, len
);
11111 /* and possibly remotely */
11112 if (st
->update_tail
)
11113 append_metadata_update(st
, u
, len
);
11118 case CH_ARRAY_SIZE
: {
11119 struct imsm_update_size_change
*u
= NULL
;
11121 imsm_create_metadata_update_for_size_change(
11124 dprintf("imsm: Cannot prepare update\n");
11128 /* update metadata locally */
11129 imsm_update_metadata_locally(st
, u
, len
);
11130 /* and possibly remotely */
11131 if (st
->update_tail
)
11132 append_metadata_update(st
, u
, len
);
11142 exit_imsm_reshape_super
:
11143 dprintf("imsm: reshape_super Exit code = %i\n", ret_val
);
11147 #define COMPLETED_OK 0
11148 #define COMPLETED_NONE 1
11149 #define COMPLETED_DELAYED 2
11151 static int read_completed(int fd
, unsigned long long *val
)
11156 ret
= sysfs_fd_get_str(fd
, buf
, 50);
11160 ret
= COMPLETED_OK
;
11161 if (strncmp(buf
, "none", 4) == 0) {
11162 ret
= COMPLETED_NONE
;
11163 } else if (strncmp(buf
, "delayed", 7) == 0) {
11164 ret
= COMPLETED_DELAYED
;
11167 *val
= strtoull(buf
, &ep
, 0);
11168 if (ep
== buf
|| (*ep
!= 0 && *ep
!= '\n' && *ep
!= ' '))
11174 /*******************************************************************************
11175 * Function: wait_for_reshape_imsm
11176 * Description: Function writes new sync_max value and waits until
11177 * reshape process reach new position
11179 * sra : general array info
11180 * ndata : number of disks in new array's layout
11183 * 1 : there is no reshape in progress,
11185 ******************************************************************************/
11186 int wait_for_reshape_imsm(struct mdinfo
*sra
, int ndata
)
11188 int fd
= sysfs_get_fd(sra
, NULL
, "sync_completed");
11190 unsigned long long completed
;
11191 /* to_complete : new sync_max position */
11192 unsigned long long to_complete
= sra
->reshape_progress
;
11193 unsigned long long position_to_set
= to_complete
/ ndata
;
11196 dprintf("cannot open reshape_position\n");
11201 if (sysfs_fd_get_ll(fd
, &completed
) < 0) {
11203 dprintf("cannot read reshape_position (no reshape in progres)\n");
11212 if (completed
> position_to_set
) {
11213 dprintf("wrong next position to set %llu (%llu)\n",
11214 to_complete
, position_to_set
);
11218 dprintf("Position set: %llu\n", position_to_set
);
11219 if (sysfs_set_num(sra
, NULL
, "sync_max",
11220 position_to_set
) != 0) {
11221 dprintf("cannot set reshape position to %llu\n",
11230 int timeout
= 3000;
11232 sysfs_wait(fd
, &timeout
);
11233 if (sysfs_get_str(sra
, NULL
, "sync_action",
11235 strncmp(action
, "reshape", 7) != 0) {
11236 if (strncmp(action
, "idle", 4) == 0)
11242 rc
= read_completed(fd
, &completed
);
11244 dprintf("cannot read reshape_position (in loop)\n");
11247 } else if (rc
== COMPLETED_NONE
)
11249 } while (completed
< position_to_set
);
11255 /*******************************************************************************
11256 * Function: check_degradation_change
11257 * Description: Check that array hasn't become failed.
11259 * info : for sysfs access
11260 * sources : source disks descriptors
11261 * degraded: previous degradation level
11263 * degradation level
11264 ******************************************************************************/
11265 int check_degradation_change(struct mdinfo
*info
,
11269 unsigned long long new_degraded
;
11272 rv
= sysfs_get_ll(info
, NULL
, "degraded", &new_degraded
);
11273 if (rv
== -1 || (new_degraded
!= (unsigned long long)degraded
)) {
11274 /* check each device to ensure it is still working */
11277 for (sd
= info
->devs
; sd
; sd
= sd
->next
) {
11278 if (sd
->disk
.state
& (1<<MD_DISK_FAULTY
))
11280 if (sd
->disk
.state
& (1<<MD_DISK_SYNC
)) {
11283 if (sysfs_get_str(info
,
11284 sd
, "state", sbuf
, sizeof(sbuf
)) < 0 ||
11285 strstr(sbuf
, "faulty") ||
11286 strstr(sbuf
, "in_sync") == NULL
) {
11287 /* this device is dead */
11288 sd
->disk
.state
= (1<<MD_DISK_FAULTY
);
11289 if (sd
->disk
.raid_disk
>= 0 &&
11290 sources
[sd
->disk
.raid_disk
] >= 0) {
11292 sd
->disk
.raid_disk
]);
11293 sources
[sd
->disk
.raid_disk
] =
11302 return new_degraded
;
11305 /*******************************************************************************
11306 * Function: imsm_manage_reshape
11307 * Description: Function finds array under reshape and it manages reshape
11308 * process. It creates stripes backups (if required) and sets
11311 * afd : Backup handle (nattive) - not used
11312 * sra : general array info
11313 * reshape : reshape parameters - not used
11314 * st : supertype structure
11315 * blocks : size of critical section [blocks]
11316 * fds : table of source device descriptor
11317 * offsets : start of array (offest per devices)
11319 * destfd : table of destination device descriptor
11320 * destoffsets : table of destination offsets (per device)
11322 * 1 : success, reshape is done
11324 ******************************************************************************/
11325 static int imsm_manage_reshape(
11326 int afd
, struct mdinfo
*sra
, struct reshape
*reshape
,
11327 struct supertype
*st
, unsigned long backup_blocks
,
11328 int *fds
, unsigned long long *offsets
,
11329 int dests
, int *destfd
, unsigned long long *destoffsets
)
11332 struct intel_super
*super
= st
->sb
;
11333 struct intel_dev
*dv
;
11334 unsigned int sector_size
= super
->sector_size
;
11335 struct imsm_dev
*dev
= NULL
;
11336 struct imsm_map
*map_src
;
11337 int migr_vol_qan
= 0;
11338 int ndata
, odata
; /* [bytes] */
11339 int chunk
; /* [bytes] */
11340 struct migr_record
*migr_rec
;
11342 unsigned int buf_size
; /* [bytes] */
11343 unsigned long long max_position
; /* array size [bytes] */
11344 unsigned long long next_step
; /* [blocks]/[bytes] */
11345 unsigned long long old_data_stripe_length
;
11346 unsigned long long start_src
; /* [bytes] */
11347 unsigned long long start
; /* [bytes] */
11348 unsigned long long start_buf_shift
; /* [bytes] */
11350 int source_layout
= 0;
11355 if (!fds
|| !offsets
)
11358 /* Find volume during the reshape */
11359 for (dv
= super
->devlist
; dv
; dv
= dv
->next
) {
11360 if (dv
->dev
->vol
.migr_type
== MIGR_GEN_MIGR
11361 && dv
->dev
->vol
.migr_state
== 1) {
11366 /* Only one volume can migrate at the same time */
11367 if (migr_vol_qan
!= 1) {
11368 pr_err("%s", migr_vol_qan
?
11369 "Number of migrating volumes greater than 1\n" :
11370 "There is no volume during migrationg\n");
11374 map_src
= get_imsm_map(dev
, MAP_1
);
11375 if (map_src
== NULL
)
11378 ndata
= imsm_num_data_members(dev
, MAP_0
);
11379 odata
= imsm_num_data_members(dev
, MAP_1
);
11381 chunk
= __le16_to_cpu(map_src
->blocks_per_strip
) * 512;
11382 old_data_stripe_length
= odata
* chunk
;
11384 migr_rec
= super
->migr_rec
;
11386 /* initialize migration record for start condition */
11387 if (sra
->reshape_progress
== 0)
11388 init_migr_record_imsm(st
, dev
, sra
);
11390 if (__le32_to_cpu(migr_rec
->rec_status
) != UNIT_SRC_NORMAL
) {
11391 dprintf("imsm: cannot restart migration when data are present in copy area.\n");
11394 /* Save checkpoint to update migration record for current
11395 * reshape position (in md). It can be farther than current
11396 * reshape position in metadata.
11398 if (save_checkpoint_imsm(st
, sra
, UNIT_SRC_NORMAL
) == 1) {
11399 /* ignore error == 2, this can mean end of reshape here
11401 dprintf("imsm: Cannot write checkpoint to migration record (UNIT_SRC_NORMAL, initial save)\n");
11406 /* size for data */
11407 buf_size
= __le32_to_cpu(migr_rec
->blocks_per_unit
) * 512;
11408 /* extend buffer size for parity disk */
11409 buf_size
+= __le32_to_cpu(migr_rec
->dest_depth_per_unit
) * 512;
11410 /* add space for stripe aligment */
11411 buf_size
+= old_data_stripe_length
;
11412 if (posix_memalign((void **)&buf
, MAX_SECTOR_SIZE
, buf_size
)) {
11413 dprintf("imsm: Cannot allocate checkpoint buffer\n");
11417 max_position
= sra
->component_size
* ndata
;
11418 source_layout
= imsm_level_to_layout(map_src
->raid_level
);
11420 while (__le32_to_cpu(migr_rec
->curr_migr_unit
) <
11421 __le32_to_cpu(migr_rec
->num_migr_units
)) {
11422 /* current reshape position [blocks] */
11423 unsigned long long current_position
=
11424 __le32_to_cpu(migr_rec
->blocks_per_unit
)
11425 * __le32_to_cpu(migr_rec
->curr_migr_unit
);
11426 unsigned long long border
;
11428 /* Check that array hasn't become failed.
11430 degraded
= check_degradation_change(sra
, fds
, degraded
);
11431 if (degraded
> 1) {
11432 dprintf("imsm: Abort reshape due to degradation level (%i)\n", degraded
);
11436 next_step
= __le32_to_cpu(migr_rec
->blocks_per_unit
);
11438 if ((current_position
+ next_step
) > max_position
)
11439 next_step
= max_position
- current_position
;
11441 start
= current_position
* 512;
11443 /* align reading start to old geometry */
11444 start_buf_shift
= start
% old_data_stripe_length
;
11445 start_src
= start
- start_buf_shift
;
11447 border
= (start_src
/ odata
) - (start
/ ndata
);
11449 if (border
<= __le32_to_cpu(migr_rec
->dest_depth_per_unit
)) {
11450 /* save critical stripes to buf
11451 * start - start address of current unit
11452 * to backup [bytes]
11453 * start_src - start address of current unit
11454 * to backup alligned to source array
11457 unsigned long long next_step_filler
;
11458 unsigned long long copy_length
= next_step
* 512;
11460 /* allign copy area length to stripe in old geometry */
11461 next_step_filler
= ((copy_length
+ start_buf_shift
)
11462 % old_data_stripe_length
);
11463 if (next_step_filler
)
11464 next_step_filler
= (old_data_stripe_length
11465 - next_step_filler
);
11466 dprintf("save_stripes() parameters: start = %llu,\tstart_src = %llu,\tnext_step*512 = %llu,\tstart_in_buf_shift = %llu,\tnext_step_filler = %llu\n",
11467 start
, start_src
, copy_length
,
11468 start_buf_shift
, next_step_filler
);
11470 if (save_stripes(fds
, offsets
, map_src
->num_members
,
11471 chunk
, map_src
->raid_level
,
11472 source_layout
, 0, NULL
, start_src
,
11474 next_step_filler
+ start_buf_shift
,
11476 dprintf("imsm: Cannot save stripes to buffer\n");
11479 /* Convert data to destination format and store it
11480 * in backup general migration area
11482 if (save_backup_imsm(st
, dev
, sra
,
11483 buf
+ start_buf_shift
, copy_length
)) {
11484 dprintf("imsm: Cannot save stripes to target devices\n");
11487 if (save_checkpoint_imsm(st
, sra
,
11488 UNIT_SRC_IN_CP_AREA
)) {
11489 dprintf("imsm: Cannot write checkpoint to migration record (UNIT_SRC_IN_CP_AREA)\n");
11493 /* set next step to use whole border area */
11494 border
/= next_step
;
11496 next_step
*= border
;
11498 /* When data backed up, checkpoint stored,
11499 * kick the kernel to reshape unit of data
11501 next_step
= next_step
+ sra
->reshape_progress
;
11502 /* limit next step to array max position */
11503 if (next_step
> max_position
)
11504 next_step
= max_position
;
11505 sysfs_set_num(sra
, NULL
, "suspend_lo", sra
->reshape_progress
);
11506 sysfs_set_num(sra
, NULL
, "suspend_hi", next_step
);
11507 sra
->reshape_progress
= next_step
;
11509 /* wait until reshape finish */
11510 if (wait_for_reshape_imsm(sra
, ndata
)) {
11511 dprintf("wait_for_reshape_imsm returned error!\n");
11517 if (save_checkpoint_imsm(st
, sra
, UNIT_SRC_NORMAL
) == 1) {
11518 /* ignore error == 2, this can mean end of reshape here
11520 dprintf("imsm: Cannot write checkpoint to migration record (UNIT_SRC_NORMAL)\n");
11526 /* clear migr_rec on disks after successful migration */
11529 memset(super
->migr_rec_buf
, 0, MIGR_REC_BUF_SECTORS
*sector_size
);
11530 for (d
= super
->disks
; d
; d
= d
->next
) {
11531 if (d
->index
< 0 || is_failed(&d
->disk
))
11533 unsigned long long dsize
;
11535 get_dev_size(d
->fd
, NULL
, &dsize
);
11536 if (lseek64(d
->fd
, dsize
- MIGR_REC_SECTOR_POSITION
*sector_size
,
11538 if ((unsigned int)write(d
->fd
, super
->migr_rec_buf
,
11539 MIGR_REC_BUF_SECTORS
*sector_size
) !=
11540 MIGR_REC_BUF_SECTORS
*sector_size
)
11541 perror("Write migr_rec failed");
11545 /* return '1' if done */
11549 /* See Grow.c: abort_reshape() for further explanation */
11550 sysfs_set_num(sra
, NULL
, "suspend_lo", 0x7FFFFFFFFFFFFFFFULL
);
11551 sysfs_set_num(sra
, NULL
, "suspend_hi", 0);
11552 sysfs_set_num(sra
, NULL
, "suspend_lo", 0);
11557 #endif /* MDASSEMBLE */
11559 struct superswitch super_imsm
= {
11561 .examine_super
= examine_super_imsm
,
11562 .brief_examine_super
= brief_examine_super_imsm
,
11563 .brief_examine_subarrays
= brief_examine_subarrays_imsm
,
11564 .export_examine_super
= export_examine_super_imsm
,
11565 .detail_super
= detail_super_imsm
,
11566 .brief_detail_super
= brief_detail_super_imsm
,
11567 .write_init_super
= write_init_super_imsm
,
11568 .validate_geometry
= validate_geometry_imsm
,
11569 .add_to_super
= add_to_super_imsm
,
11570 .remove_from_super
= remove_from_super_imsm
,
11571 .detail_platform
= detail_platform_imsm
,
11572 .export_detail_platform
= export_detail_platform_imsm
,
11573 .kill_subarray
= kill_subarray_imsm
,
11574 .update_subarray
= update_subarray_imsm
,
11575 .load_container
= load_container_imsm
,
11576 .default_geometry
= default_geometry_imsm
,
11577 .get_disk_controller_domain
= imsm_get_disk_controller_domain
,
11578 .reshape_super
= imsm_reshape_super
,
11579 .manage_reshape
= imsm_manage_reshape
,
11580 .recover_backup
= recover_backup_imsm
,
11581 .copy_metadata
= copy_metadata_imsm
,
11582 .examine_badblocks
= examine_badblocks_imsm
,
11584 .match_home
= match_home_imsm
,
11585 .uuid_from_super
= uuid_from_super_imsm
,
11586 .getinfo_super
= getinfo_super_imsm
,
11587 .getinfo_super_disks
= getinfo_super_disks_imsm
,
11588 .update_super
= update_super_imsm
,
11590 .avail_size
= avail_size_imsm
,
11591 .min_acceptable_spare_size
= min_acceptable_spare_size_imsm
,
11593 .compare_super
= compare_super_imsm
,
11595 .load_super
= load_super_imsm
,
11596 .init_super
= init_super_imsm
,
11597 .store_super
= store_super_imsm
,
11598 .free_super
= free_super_imsm
,
11599 .match_metadata_desc
= match_metadata_desc_imsm
,
11600 .container_content
= container_content_imsm
,
11601 .validate_container
= validate_container_imsm
,
11608 .open_new
= imsm_open_new
,
11609 .set_array_state
= imsm_set_array_state
,
11610 .set_disk
= imsm_set_disk
,
11611 .sync_metadata
= imsm_sync_metadata
,
11612 .activate_spare
= imsm_activate_spare
,
11613 .process_update
= imsm_process_update
,
11614 .prepare_update
= imsm_prepare_update
,
11615 .record_bad_block
= imsm_record_badblock
,
11616 .clear_bad_block
= imsm_clear_badblock
,
11617 .get_bad_blocks
= imsm_get_badblocks
,
11618 #endif /* MDASSEMBLE */